EMS Textbooks in Mathematics

EMS Textbooks in Mathematics is a series of books aimed at students or professional mathemati-

cians seeking an introduction into a particular field. The individual volumes are intended not only to
provide relevant techniques, results, and applications, but also to afford insight into the motivations
and ideas behind the theory. Suitably designed exercises help to master the subject and prepare
the reader for the study of more advanced and specialized literature.

Jørn Justesen and Tom Høholdt, A Course In Error-Correcting Codes

Markus Stroppel, Locally Compact Groups
Peter Kunkel and Volker Mehrmann, Differential-Algebraic Equations
Dorothee D. Haroske and Hans Triebel, Distributions, Sobolev Spaces, Elliptic Equations
Thomas Timmermann, An Invitation to Quantum Groups and Duality
Oleg Bogopolski, Introduction to Group Theory
Marek Jarnicki and Peter Pflug, First Steps in Several Complex Variables: Reinhardt Domains
Tammo tom Dieck, Algebraic Topology
Mauro C. Beltrametti et al., Lectures on Curves, Surfaces and Projective Varieties
Wolfgang Woess, Denumerable Markov Chains
Eduard Zehnder, Lectures on Dynamical Systems. Hamiltonian Vector Fields and Symplectic
Capacities
Andrzej Skowroński and Kunio Yamagata, Frobenius Algebras I. Basic Representation Theory
Piotr W. Nowak and Guoliang Yu, Large Scale Geometry
Joaquim Bruna and Juliá Cufí, Complex Analysis
Eduardo Casas-Alvero, Analytic Projective Geometry
Fabrice Baudoin, Diffusion Processes and Stochastic Calculus
 Olivier Lablée

Spectral
Theory in
Riemannian
Geometry
Author:

Olivier Lablée
Université Joseph Fourier Grenoble 1
Laboratoire de Mathématiques
BP 74
38402 Saint Martin d’Hères
France
E-mail: [email protected]

2010 Mathematics Subject Classification: 58J35, 58J37, 58J50, 58J53, 35P05, 35P15, 35P20,
47A05, 47A10, 47A12, 47A60, 47A75, 49R05, 53C21

Key words: Spectral theory, linear operators, spectrum of operators, spectral geometry, eigenvalues,
Laplacian, inverse problems, Riemannian geometry, analysis on manifolds

ISBN 978-3-03719-151-4

The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography,
and the detailed bibliographic data are available on the Internet at http://www.helveticat.ch.

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broad-
casting, reproduction on microfilms or in other ways, and storage in data banks. For any kind of use
permission of the copyright owner must be obtained.

© 2015 European Mathematical Society

Contact address:

European Mathematical Society Publishing House

Seminar for Applied Mathematics
ETH-Zentrum SEW A27
CH-8092 Zürich
Switzerland

Phone: +41 (0)44 632 34 36

Email: [email protected]
Homepage: www.ems-ph.org

Typeset using the author’s TEX files: le-tex publishing services GmbH, Leipzig, Germany
Printing and binding: Beltz Bad Langensalza GmbH, Bad Langensalza, Germany
∞ Printed on acid free paper
987654321
Acknowledgements

I wish to express my gratitude to Laurent Bessières (Université Bordeaux 1),

Pierre Bérard (Université Grenoble 1), Gilles Courtois (CNRS & Université Pierre
et Marie Curie Paris 6), the referees, and Thomas Hintermann (European Math-
ematical Society Publishing House) for their comments and suggestions. It is
a pleasure for me to thank each participant of the “Séminaire de Théorie Spectrale
et Géométrie de Grenoble” and members of Institut Fourier, especially Gérard
Besson, Yves Colin de Verdière, Sylvain Gallot, Frédéric Faure, Frédéric Mou-
ton and San Vu Ngoc (Université Rennes 1), from whom I have learned much on
geometry. My thanks go also to Géraldine Rahal (CNRS & Université Greno-
ble 1) for having suggested many improvements to the presentation. Finally, I am
grateful for the support of my family members and friends.

Saint Martin d’Hères, September 2014 Olivier Lablée

Contents

1 Introduction to Spectral Geometry 1

2 Detailed introduction to abstract spectral theory 7

2.1 Linear operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Bounded operators . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Lax–Milgram theorem . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Unbounded operators . . . . . . . . . . . . . . . . . . . . 10
2.2 Closed operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Adjoint operators . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.2 Symmetry and self-adjointness . . . . . . . . . . . . . . . 16
2.3.3 An important result concerning self-adjoint operators . . . 17
2.4 Spectrums and resolvent set . . . . . . . . . . . . . . . . . . . . . 19
2.4.1 Spectrum versus point spectrum . . . . . . . . . . . . . . . 19
2.4.2 The self-adjoint case . . . . . . . . . . . . . . . . . . . . . 25
2.5 Spectral theory of compact operators . . . . . . . . . . . . . . . . 27
2.5.1 The notion of compact operators . . . . . . . . . . . . . . 27
2.5.2 Algebraic properties of compact operators . . . . . . . . . 28
2.5.3 Main spectral theorem for compact self-adjoint operators . 30
2.5.4 A proof using the Riesz–Schauder theory . . . . . . . . . . 30
2.6 The spectral theorem multiplication operator form for unbounded
operators on a Hilbert space . . . . . . . . . . . . . . . . . . . . . 35
2.6.1 Spectral theorem multiplication operator form . . . . . . . 35
2.6.2 Functional calculus . . . . . . . . . . . . . . . . . . . . . 38
2.6.3 Functional calculus in Quantum Mechanics . . . . . . . . . 39
2.7 Some complements on operators theory . . . . . . . . . . . . . . 40
2.7.1 Closable operators . . . . . . . . . . . . . . . . . . . . . . 40
2.7.2 Unbounded operators with compact resolvents . . . . . . . 41
2.7.3 Numerical range and applications . . . . . . . . . . . . . . 44
2.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3 The Laplacian on a compact Riemannian manifold 51

3.1 Basic Riemannian Geometry . . . . . . . . . . . . . . . . . . . . 51
3.1.1 Differential Geometry: conventions and notations . . . . . 51
3.1.2 Riemannian manifolds and examples . . . . . . . . . . . . 57
3.1.3 Metric structure and geodesics on a Riemannian manifold . 59
3.1.4 Curvatures on a Riemannian manifold . . . . . . . . . . . 61
3.1.5 Integration on a Riemannian manifold . . . . . . . . . . . 64
viii Contents

3.2 Analysis on manifolds . . . . . . . . . . . . . . . . . . . . . . . . 66

3.2.1 Distributions on a Riemannian manifold . . . . . . . . . . 66
3.2.2 Sobolev spaces on a Riemannian manifold . . . . . . . . . 67
3.2.3 The Laplacian operator and the Green formula . . . . . . . 68
3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4 Spectrum of the Laplacian on a compact manifold 71

4.1 Physical examples . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1.1 The wave equation on a string . . . . . . . . . . . . . . . . 71
4.1.2 The heat equation . . . . . . . . . . . . . . . . . . . . . . 72
4.1.3 The Schrödinger equation . . . . . . . . . . . . . . . . . . 73
4.2 A class of spectral problems . . . . . . . . . . . . . . . . . . . . 75
4.2.1 The closed eigenvalue problem . . . . . . . . . . . . . . . 75
4.2.2 The Dirichlet eigenvalue problem . . . . . . . . . . . . . . 75
4.2.3 The Neumann eigenvalue problem . . . . . . . . . . . . . 76
4.2.4 Other problems . . . . . . . . . . . . . . . . . . . . . . . 76
4.3 Spectral theorem for the Laplacian . . . . . . . . . . . . . . . . . 76
4.4 A detailed proof by a variational approach . . . . . . . . . . . . . 78
4.4.1 Variational generic abstract eigenvalue problem . . . . . . 78
4.4.2 The closed eigenvalue problem . . . . . . . . . . . . . . . 81
4.4.3 The Dirichlet eigenvalue problem . . . . . . . . . . . . . 84
4.4.4 The Neumann eigenvalue problem . . . . . . . . . . . . . 87
4.4.5 A remark on the variational formulation . . . . . . . . . . 89
4.5 The minimax principle and applications . . . . . . . . . . . . . . 90
4.5.1 A physical example . . . . . . . . . . . . . . . . . . . . . 90
4.5.2 The minimax theorem . . . . . . . . . . . . . . . . . . . . 91
4.5.3 Properties of the first eigenvalue . . . . . . . . . . . . . . 95
4.5.4 Monotonicity domain principle . . . . . . . . . . . . . . . 97
4.5.5 A perturbation of metric result . . . . . . . . . . . . . . . 98
4.6 Complements: The Schrödinger operator and the Hodge–de Rham
Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.6.1 The Schrödinger operator . . . . . . . . . . . . . . . . . . 100
4.6.2 The Hodge–de Rham Laplacian . . . . . . . . . . . . . . . 102

5 Direct problems in spectral geometry 107

5.1 Explicit calculation of the spectrum . . . . . . . . . . . . . . . . . 107
5.1.1 Flat tori . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.1.2 Rectangular domains with boundary conditions . . . . . . 109
5.1.3 Spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.1.4 Harmonic oscillator . . . . . . . . . . . . . . . . . . . . . 110
5.2 Qualitative properties of the spectrum . . . . . . . . . . . . . . . 113
5.2.1 From the bottom of the spectrum . . . . . . . . . . . . . . . 113
5.2.2 . . . to the large eigenvalues: the Weyl formula . . . . . . . 117
 Contents ix

5.2.3 Weyl formula for a flat torus . . . . . . . . . . . . . . . . . 117

5.2.4 The semi-classical paradigm . . . . . . . . . . . . . . . . 118
5.3 The spectral partition function ZM . . . . . . . . . . . . . . . . . 119
5.4 Eigenvalues and eigenfunctions of surfaces . . . . . . . . . . . . . 121
5.4.1 Spectrum of surfaces . . . . . . . . . . . . . . . . . . . . 121
5.4.2 Eigenfunctions of surfaces and Courant’s nodal theorem . . 122
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

6 Intermezzo: “Can one hear the holes of a drum?” 125

6.1 The main result . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.2 Some useful spaces . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.3 Electrostatic capacity and the Poincaré inequality . . . . . . . . . 129
6.4 A detailed proof of the main Theorem 6.1.1 . . . . . . . . . . . . 131

7 Inverse problems in spectral geometry 139

7.1 Can one hear the shape of a drum? . . . . . . . . . . . . . . . . . 139
7.2 Length spectrum and trace formulas . . . . . . . . . . . . . . . . 139
7.3 Milnor’s counterexample . . . . . . . . . . . . . . . . . . . . . . 142
7.4 Prescribing the spectrum on a manifold . . . . . . . . . . . . . . . 146
7.5 Heat kernel and spectral geometry . . . . . . . . . . . . . . . . . 147
7.5.1 The heat equation . . . . . . . . . . . . . . . . . . . . . . 147
7.5.2 The heat kernel . . . . . . . . . . . . . . . . . . . . . . . 148
7.5.3 The spectral partition function ZM as a trace . . . . . . . . 158
7.6 The Minakshisundaram–Pleijel expansion and the Weyl formula . 159
7.7 Two planar isospectral nonisometric domains . . . . . . . . . . . 161
7.7.1 Construction of the domains D1 and D2 . . . . . . . . . . 161
7.7.2 Isospectrality of the domains D1 and D2 . . . . . . . . . . 163
7.8 Few words about Laplacian and conformal geometry . . . . . . . 166
7.8.1 Conformal geometry on surfaces . . . . . . . . . . . . . . 166
7.8.2 Spectral zeta function and uniformization theorem
on surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.8.3 Ricci flow on surfaces . . . . . . . . . . . . . . . . . . . . 172
7.8.4 What about three manifolds? . . . . . . . . . . . . . . . . 173

Bibliography 177

Index 185
Chapter 1
Introduction to Spectral Geometry

From P.-S. Laplace to E. Beltrami

The Laplace operator was first introduced by P.-S. Laplace (1749–1827) for de-
scribing celestial mechanics (the notation  is due to G. Lamé). For example, in
our three-dimensional (Euclidean) space the Laplace operator (or just Laplacian)
is the linear differential operator:
8
< C 2 .U / ! C 0 .U /
W
: @2 f @2 f @2 f
f 7! f WD @x 2
C @y 2
C @z 2
;

where U is an open set of R3 . This operator can be generalized to a Riemannian

manifold .M; g/: this generalization is called the Laplace–Beltrami operator and
is denoted by the symbol g . The study of this operator and in particular the
study of its spectrum is called spectral geometry. The Laplace–Beltrami operator
is very useful in many fields of physics, in particular in all diffusion processes:
 Fluid mechanics: the Navier–Stokes equations
8
< @u
@t
C .u:r/u  u D rp
:
div.u/ D 0I

 potential theory and gravity theory (with Newton potential): the Laplace
equation and the Poisson equation

u D f I

 heat diffusion process: the heat equation

@u
.x; t/  u.x; t/ D f .x; t/I
@t

 wave physics: the wave equation

@2 u
.x; t/  u.x; t/ D 0I
@t 2
2 Chapter 1. Introduction to Spectral Geometry

 quantum physics: the Schrödinger equation

@
i„ .x; t/ D  .x; t/ C V .x/ .x; t/I
@t

 etc. . . .
Let us mention that this operator appears also in:
 computer science: in particular, in computer vision (blob detection)
 economy: financial models, Black–Scholes equations
 etc. . . .
Moreover, spectral geometry is an inter-disciplinary field of mathematics; it in-
volves
 analysis of ODE and PDE
 dynamical systems: classical and quantum completely integrable systems,
quantum chaos, geodesic flows and Anosov flows on (negatively curved)
manifolds (for example, Ruelle resonances are related to the spectrum of
the Laplacian, see the recent articles [Fa-Ts1], [Fa-Ts2]).
 geometry and topology (the main purpose of these notes is to explain this)
 geometric flow: the parabolic behaviour of scalar curvature
@Rg.t / ˇ ˇ2
D Rg.t / C 2 ˇRicg.t / ˇ I
@t
the Ricci flow (see also Sections 7.8.3 and 7.8.4)
@g
D 2Ricg.t /
@t

 probability: the Brownian motion on a Riemannian manifold .M; g/ is de-

fined to be a diffusion on M (generator operator is given by 12 g ).
 etc. . . .

Meaning of the Laplacian operator

To get a better feeling about the Laplacian operator, consider for example a one-
dimensional C 3 function uW R ! R. The mean value of u on the compact set
Œh; h is given by
Z
1 h
u WD u.x/ dx:
2h h
 3

Now, using the Taylor expansion of u around the origin we get: for all x 2 Œh; h,

x2 x3
u.x/ D u.0/ C u0 .0/x C u00 .0/ C u000 .0/ C o.x 4 /:
2 6
Therefore,
Z !
h 2 3
1 x x
uD u.0/ C u0 .0/x C u00 .0/ C u000 .0/ dx C o.h4 /
2h h 2 6

u00 .0/ 2 u00 .0/ 2

i.e., u D u.0/ C 12
h C o.h4 /, hence u  u.0/ D 12
h C o.h4 /. In other
words,
12
u.0/ D u00 .0/ D .u  u.0// C o.h2 /;
h2
thus the Laplacian of u measures the difference between the function u at 0 and
the mean value of u on the neighbourhood Œh; h.
Another way to understand this interpretation is to use the finite difference
method: for example, on the domain I D Œ0; 1 and for an integer N > 0 consider
the discretization grid of I given by fti WD ih; i 2 f0 : : : N gg with h WD N1 . Thus
using the Taylor expansion of u around the point ti we have

h2 h3
u.ti C1 / D u.ti C h/ D u.ti / C u0 .ti /h C u00 .ti / C u000 .ti / C o.h4 /
2 6
and
h2 h3
u.ti 1/ D u.ti  h/ D u.ti /  u0 .ti /h C u00 .ti /  u000 .ti / C o.h4 /;
2 6
whence
ui 1  2ui C ui C1
u.ti / D u00 .ti / D C o.h2 /
h2
where uk WD u.tk /. The principal term ui 1 2ui Cui C1 represents the difference
between the value of the function u at the point ti and the values of u on the grid-
neighborhood of ti . In particular, the discretization of the problem u00 D f on
I with the boundary conditions u.0/ D u.1/ D 0 is the linear equation AX D B
with 0 1
2 1 0
B :: C
B 1 2 : C
B C
ADB B 1
:: ::
: : C
C
B C
@ : : : 2 1 A
0 1 2
4 Chapter 1. Introduction to Spectral Geometry

and 0 1 0 1
u1 f1
B C B C
X D @ ::: A ; B D @ ::: A :
uN fN
More generally: the Laplace–Beltrami operator at a point x measures the differ-
ence of the mean value of f on a neighbourhood of x and the value of the function
at the point x.

Main Topics in Spectral Geometry

Given a (compact) Riemannian manifold .M; g/, we can associate to it the (lin-
ear) unbounded operator g . This operator is self-adjoint and its spectrum is
discrete (see Chapter 4): namely, the spectrum consists of an increasing sequence
.k .M //k of real eigenvalues of finite multiplicity such that k .M / as k ! C1.
We denote this spectrum by
Spec.M; g/ WD .k .M //k :
In others words, for any integer k  0 there exists a non trivial function uk on M
(eigenfunction) such that
g uk D k .M /uk :
The spectral theory of the Laplacian on a compact Riemannian manifold .M; g/
is in particular interested in the connection between the spectrum Spec.M; g/ and
the geometry of the manifold .M; g/. Indeed there are many deep connections
between spectrum and geometry. Therefore spectral geometry studies such con-
nections. The main topics in spectral geometry can be split into two types.

Direct problems
The main questions in direct problems are (see Chapter 5):

Question. Can we compute .exactly or not/ the spectrum Spec.M; g/? And .or/:
can we find properties on the spectrum Spec.M; g/?

Then the principle of direct problems is to compute or find some properties

on the spectrum of a compact Riemannian manifold .M; g/. Obviously the ques-
tion to compute the spectrum of Laplacian (or for other operators) arises in a lot
of problems: analysis of PDE, dynamical systems, mathematical physics, differ-
ential geometry, probability, etc. See, for example, Section 4.1.3 for a concrete
application in quantum dynamics.
 5

For example, the first non-null eigenvalue ? .M / plays a very important role
in Riemannian geometry, and one of the main questions is to find a lower bound
for ? .M / depending on the geometric properties of the manifold, e.g. the di-
mension n, the volume Vol.M; g/, the curvature R, etc. (see Section 5.2.1):

? .M /  a.n; Vol.M; g/; R;    /:

Another example of (asymptotic) computation (based on the Weyl formula, see

Section 7.6) is
  n2
.2/n 2
k .M /  knI
k!C1 Bn Vol.M; g/
n
here n is the dimension of the compact manifold .M; g/ and Bn WD 2
. n
.
2 C1/

Inverse problems
The data of a metric on a Riemannian manifold determine completely the Lapla-
cian g and therefore its spectrum Spec.M; g/. Hence the spectrum is a Rieman-
nian invariant: if two Riemannian manifolds .M; g/ and .M 0 ; g 0 / are isometric,
then they are isospectral, i.e., Spec.M; g/ D Spec.M 0 ; g 0 /. Conversely the main
question is

Question. Does the data of the spectrum Spec.M; g/ determine the “shape” of
the manifold .M; g/?

In other words: which geometric information can we deduce from the spec-
trum? For example, we have the classical heat invariants; indeed the spectrum of
the manifold .M; g/ determines
 the dimension of .M; g/
 the volume of .M; g/
 the integral of the scalar curvature Scalg over .M; g/.
Another main topic in inverse problems is

Question. What sequences of real numbers can be spectra of a compact mani-

fold?

A simpler version of this question is: Let M be a fixed manifold, given a finite
increasing sequence of real numbers 0 < a1  a2      aN does, there exists a
metric g such that the k t h first eigenvalues of .M; g/ are equal to 0 < a1  a2 
    aN ? Y. Colin de Verdière proved in 1987 that the answer is positive (see
Section 7.4).
The next important example of inverse problem concerns the length spectrum
(see Section 7.2). The length spectrum of a compact Riemannian manifold .M; g/
6 Chapter 1. Introduction to Spectral Geometry

is the set of lengths of closed geodesics on .M; g/ counted with multiplicities. As

it turns out, the spectrum of the manifold determines the length spectrum. Spectral
theory is also an important tool for understanding the relationships between the
formalism of classical mechanics and that of quantum mechanics:
 The formalism of classical mechanics on a Riemannian manifold is ex-
pressed in terms of geodesics.
 The formalism of quantum mechanics on a Riemannian manifold is ex-
pressed in terms of the Laplace–Beltrami operator.
The question of isospectrality in Riemannian geometry may be traced back to H.
Weyl in 1911–1912 and became popularized thanks to M. Kac’s article of 1966
[Kac1]. The famous sentence of Kac “Can one hear the shape of a drum?” refers
to this type of isospectral problem. The exact formulation of the isospectrality
question is

Question. If two Riemannian manifolds .M; g/ and .M 0 ; g 0 / are isospectral, are

they isometric?

The answer is negative and was given first by J. Milnor in 1964 (see Section
7.3). In 1984 and 1985, respectively C. Gordon, E.N. Wilson [Go-Wi] and T.
Sunada [Sun] gave a systematic construction of counter-examples. In 1992, C.
Gordon, D. Webb, and S. Wolpert [GWW1] gave the first planar counter-example.
The main classical references on spectral geometry are the book of M. Berger,
P. Gauduchon and E. Mazet [BGM], the book of P. Bérard [Bér8][Bér], of I.
Chavel [Cha1] and the book of S. Rosenberg [ROS].

Organization of the book

The present book is a basic introduction to spectral geometry. The reader is as-
sumed to have a good grounding in functional analysis and differential calcu-
lus. Chapter 2 discusses the fundamental notions of spectral theory for compact
and unbounded operators. Chapter 3 is a review of differentiable manifolds and
Riemannian geometry, including the definition of Laplace–Beltrami operator. In
Chapter 4 we define the spectrum of the Laplace–Beltrami operator on a Rieman-
nian manifold and then we present the minimax principle and some geometrical
consequences. Chapter 5 discusses principles underlying the treatment of direct
problems of spectral geometry, including some exact computations of spectrum.
In Chapter 6 we present a topological perturbation result for eigenvalues of a man-
ifold. Chapter 7 is devoted to inverse problems in spectral geometry; in particular,
at the end of the chapter we introduce briefly some results of conformal geometry
in dimension 2 and 3.
Chapter 2
Detailed introduction to abstract spectral
theory

In order to study the Laplace–Beltrami operator on a compact Riemannian man-

ifold, we need to introduce some basic material of abstract spectral theory. In
this chapter we give some classical results concerning spectral theory in Hilbert
spaces for bounded and unbounded operators. We focus on the case of compact
self-adjoint operators.
For a complete course on Hilbert space and compact operators see for example
the book of H. Brézis [Bré]. For the general case of unbounded operators see
for example the bible of T. Kato [Kat] and the books of M. Reed and B. Simon
[Re-Si], see also [Dav], [Yos], [Le-Br]. For a pure algebraic approach see N.
Bourbaki [Bou] (in French) or W. Arverson [Arv1], [Arv2].
Throughout this section H is a separable Hilbert space over the field C and for
the inner product we use the conventions
˝ ˛ ˝ ˛
x C x 0 ; y H D  hx; yiH C x 0 ; y H
˝ ˛ ˝ ˛
x; y C y 0 H D  hx; yiH C x; y 0 H

2.1 Linear operators

2.1.1 Bounded operators
A bounded operator on H is a linear map T W H ! H with the property that there
exists M  0, such that for any vector x 2 H we have
kT xkH  M kxkH :
In other words, a bounded operator is a continuous linear map. In this book we
denote by B .H; kkH //, or simply B.H /, the set of bounded operators of H . It
is easy to see that B.H / is an algebra. The space B.H / can be equipped with the
operator norm
8
ˆ
ˆ B.H / ! RC
<
jjjjjj W
ˆ T 7! jjjT jjj WD sup kT xkH :
:̂
x2H;x¤0 kxkH
8 Chapter 2. Detailed introduction to abstract spectral theory

This norm is also an algebra norm, i.e., for all T; S 2 B.H / we have

jjjT S jjj  jjjT jjj jjjS jjj :

Since H is complete, the space .B.H /; jjjjjj / is a Banach space. Observe that we
also have

jjjT jjj D sup kT xkH D sup jhT x; yiH j: (2.1)

kxkH 1 kxkH ;kykH 1

Similarly, a linear form `W H ! C is bounded if that there exists M  0, such

that for any vector x 2 H
j`xj  M kxkH :
In this case we put
j`xj
jjj`jjj WD sup :
x2H;x¤0 kxkH

Remark 2.1.1. Note that every operators on a finite-dimensional space is bounded.

2.1.2 Lax–Milgram theorem

We start this section by a very useful theorem in functional analysis and PDE
analysis.

Theorem 2.1.2 (Lax–Milgram theorem). Let V be a Hilbert space and consider

a bilinear symmetric form a.; / on V 2 which is continuous and elliptic (namely,
there exists a constant C  0 such that for all .u; v/ 2 V 2 we have ja.u; v/j 
C kukV kvkV , and there exists a constant ˛ > 0 such that for all u 2 V we have
ja.u; u/j  ˛ kuk2V ). Also, let `./ be a bounded linear form on V . Then there
exists a unique vector u 2 V such that for all v 2 V we have a.u; v/ D `.v/.
Moreover, kukV  jjj`jjj
˛
:

Proof. Since ` is a bounded linear form on the Hilbert space V , the Riesz–Fréchet
representation theorem provides a unique ` 2 V such that for all v 2 V we
have `.v/ D h` ; viV . Similarly, for a fixed vector u 2 V the map v 2 V 7!
a.u; v/ 2 R is a bounded linear form, so there exists a unique fu 2 V such that
for all v 2 V we have a.u; v/ D hfu ; viV . So we can consider the map
8
< V ! V
fW
: u 7! f :
u
 2.1 Linear operators 9

Using the linearity of a it is clear that this map is linear. Moreover,

kfu k2V D hfu ; fu iV D a .u; fu /  C kukV kfu kV ;

therefore
kf .u/kV D kfu kV  C kukV :
Thus the map f W .V; kkV / ! .V; kkV / is bounded and jjjf jjj  C (here C D
jjjajjj).
Now let us show that f is bijective. First for u 2 ker.f / we have

hfu ; uiV D a .u; u/ D 0;

and since a .u; fu /  ˛ kuk2V we deduce that u D 0. Hence f is injective.

Now to show that f is surjective observe first that im.f /? D f0g and im.f /
is a closed subspace of V . The first claim is clear: for v 2 im.f /? we have
in particular hv; fv iV D a .v; v/ D 0, and using also that a .v; v/  ˛ kvk2V
we deduce that v D 0. Consequently im.f / is dense in V . Next to see that
im.f / D V it suffices to prove that im.f / is a closed subspace of V . Consider
a Cauchy sequence .xn /n WD .f .vn //n in the space im.f /. Since V is closed,
the sequence .xn /n converge to some vector x 2 V . Using the Cauchy–Schwarz
inequality we have
 2   ˝   ˛
˛ vp  vq V  a vp  vq ; vp  vq D f vp  vq ; vp  vq V
˝     ˛
D f vp  f vq ; vp  vq V
˝ ˛    
D xp  xq ; vp  vq  C xp  xq  vp  vq  :
V V V

Since .xn /n is a Cauchy sequence in V , .vn /n is also a Cauchy sequence in the

Hilbert space V , therefore it converges to some vector v 2 V . Further, since the
map f is continuous,

kxn  f .v/kV D kf .vn /  f .v/kV  jjjf jjj kvn  vkV ;

so the sequence .xn /n converges to the vector f .v/ 2 im.f /, which shows that
the subspace im.f / is closed. Hence, im.f / D V .
Since the map f is bijective, there exists a unique u 2 V such that

f .u/ D ` :

Therefore, for all v 2 V

hf .u/; viV D h` ; viV ;
i.e.,
a.u; v/ D `.v/:
10 Chapter 2. Detailed introduction to abstract spectral theory

To finish the proof remark that

˛ kuk2V  a.u; u/ D `.u/  jjj`jjj kukV ;

and so
jjj`jjj
kukV  kukV : 
˛

2.1.3 Unbounded operators

A generalization of bounded operators is provided by the notion of unbounded
operators (or operators with domains). In fact, it is a necessity to deal with such
operators if one wishes to study Quantum Mechanics since they appear as soon as
one wishes to consider, say, a free quantum particle in R.
We said A is an unbounded linear operator with domain D.A/ if D.A/ is
a subspace of H and the map
8
< D.A/ ! H
AW
: x 7! Ax;

is linear.
Remark 2.1.3. Be careful: unbounded operators are not continuous.

Example 2.1.4. Consider the operator of multiplication by x on the space L2 .R/,

8
< D .Mx / ! L2 .R/
Mx W
: f 7! xf .x/;
˚ 
with the domain D .Mx / WD f 2 L2 .R/I xf .x/ 2 L2 .R/ . Obviously, D .Mx /
is a subspace of L2 .R/ and Mx is linear. Let us remark that here D .Mx / ¤
L2 .R). Note that Mx is the position operator of Quantum Mechanics.

Definition 2.1.5. We said that an unbounded linear operator .A; D.A// is in-
cluded in another operator .B; D.B// if D.A/  D.B/ and for all x 2 D.A/,
Bx D Ax. If A is included in B we write A  B.

In others words, A  B if and only if the graph of A is included into the graph
of B. Now, recall the following usual definition:

Definition 2.1.6. Let .A; D.A// be an unbounded linear operator. We define:

(i) ker.A/, the kernel of A, by ker.A/ WD fx 2 D.A/I Ax D 0g :
(ii) im.A/, the image of A, by im.A/ WD fAx; x 2 D.A/g.
 2.1 Linear operators 11

A bounded operator A is said to be invertible if there exists B 2 B.H / such

that AB D I and BA D I . In this case we denote by A1 D B the inverse of A.
Now, let us recall a classical result on bounded operators.

Lemma 2.1.7 (von Neumann lemma). Let T be a bounded operator on H such

that jjjT jjj < 1. Then the operator I  T is invertible and
C1
X
.I  T /1 D T k 2 B.H /:
kD0
Pn
Proof. Consider the sequence Sn WD kD0 T , n  0. It is clear that for all
k

n  0 the operator Sn is bounded. Next, for all integers .n; m/ with m > n we
have
C1
X 1
jjjSm  Sn jjj  jjjT jjjk D jjjT jjjnC1 ;
1  jjjT jjj
kDnC1
so jjjSm  Sn jjj converges to 0 as n ! 1, hence .Sn /n is a Cauchy sequence in
the Banach algebra of bounded operators. Consequently, there exists a bounded
operator S such that limn!1 Sn D S in the norm jjjjjj. It immediately fol-
lows that limn!1 .I  T /Sn D .I  T /S and limn!1 Sn .I  T / D S.I  T /
for the norm jjjjjj. And, observe that for all n  0
X
n X
n X
nC1
.I  T /Sn D Sn  T kC1 D Tk  Tk
kD0 kD0 kD1
D I  T nC1 :
Since jjjT jjj < 1 we have limn!1 .I  T /Sn D Id in the norm jjjjjj. Similarly,
limn!1 .I  T /Sn D I . By the unicity of the limits, .I  T /S D I and S.I 
T / D I . In other words, I  T is invertible and
C1
X
1
.I  T / D S WD T k 2 B.H /: 
kD0

A very useful consequence of Lemma 2.1.7 is

Corollary 2.1.8. Let T be a bounded operator on H . For all z 2 C such that

jzj > jjjT jjj the operator zI  T is invertible with bounded inverse, and
C1
X 1
.zI  T /1 D T k 2 B.H /:
z kC1
kD0
 
Proof. It suffices to remark that .zI  T / D z I  1z T and apply the von Neu-
mann lemma. 
12 Chapter 2. Detailed introduction to abstract spectral theory

2.2 Closed operators

We have seen that an unbounded linear operator is not continuous. The analogue
of boundedness for unbounded linear operators is the notion of closed operators:

Definition 2.2.1. Let .A; D.A// be an unbounded linear operator. We say that
.A; D.A// is closed if the graph Gr.A/ WD f.x; Ax/ I x 2 D.A/g is closed in H 2
with respect to the graph norm k.x; Ax/k2 WD kxk2H C kAxk2H .

For the practice we use:

Proposition 2.2.2. An unbounded linear operator .A; D.A// is closed if and only
if for any sequence .xn /n 2 D.A/N such that
8
ˆ
< n!1 lim xn D x in the norm kkH
and
:̂ lim Axn D y in the norm kkH
n!1

we have x 2 D.A/ and y D Ax.

Proof. Suppose .A; D.A// is closed and let .xn /n 2 H N be a sequence such
that limn!1 xn D x and limn!1 Axn D y. Then the sequence .xn ; Axn /n 2
Gr.A/N converges in the graph norm as n ! 1 to the couple .x; y/. Since the
graph is closed we deduce that x 2 D.A/ and y D Ax.
Conversely, suppose that for any sequence .xn /n 2 D.A/N such that
limn!1 xn D x and limn!1 Axn D y we have x 2 D.A/ and y D Ax. Let
.xn ; Axn / 2 Gr.A/N be a sequence such that limn!1 .xn ; Axn / D .x; y/ in H 2
in the graph norm. In particular, .xn /n converge as n ! 1 to x and .Axn /n
converge as n ! 1 to y, hence x 2 D.A/ and y D Ax, that is

lim .xn ; Axn / D .x; Ax/ 2 Gr.A/:

n!1

Therefore, the graph Gr.A/ WD f.x; Ax/ ; x 2 D.A/g is closed (in the graph
norm) in H 2 . 

We have also the following criterion.

Proposition 2.2.3. An unbounded linear operator .A; D.A// is closed if and only
if the set D.A/ is complete for the norm kkA , where kxk2A WD kxk2H C kAxk2H .

Proof. Suppose first that .A; D.A// is closed. Let .xn /n be a Cauchy sequence in
.D.A/; kkA /. Then .xn /n and .Axn /n are Cauchy sequences in the Hilbert space
H , so there exists .x; y/ 2 H 2 such that limn!1 xn D x and limn!1 Axn D y
 2.2 Closed operators 13

in the norm kkH . Since .A; D.A// is closed we deduce that x 2 D.A/ and
y D Ax. Consequently,

kxn  xk2A D kxn  xk2H C kAxn  Axk2H

converge as n ! 1 to 0, i.e., limn!1 xn D x in the norm kkA . So D.A/ is

complete in the norm kkA .
Conversely, suppose that D.A/ is complete in the norm kkA . Let .xn /n be
a sequence in D.A/ such that limn!1 xn D x and limn!1 Axn D y in kkH . In
particular .xn /n and .Axn /n are Cauchy sequences, so .xn /n is a Cauchy sequence
in .D.A/; kkA /. Since D.A/ is supposed to be complete, there exists a x0 2
D.A/ such that limn!1 xn D x0 in the norm kkA and we deduce that x0 D x 2
D.A/ and y D Ax. 

Example 2.2.4. The operator

8
< D.A/ WD C 1 .Œ0; 1/ ! C 0 .Œ0; 1/;
AW
: ' 7! ' 0
 
is closed. Indeed, A is well defined and linear. C 0 .Œ0; 1/; kk1 is a Banach
space. Moreover RŒX jŒ0; 1  C 1 .Œ0; 1/  C 0 .Œ0; 1/ and since the space of real
polynomials RŒX  restricted to Œ0; 1 is dense in C 0 .Œ0; 1/ in the norm kk1 , we
have
C 1 .Œ0; 1/ D C 0 .Œ0; 1/;

where the closure is taken in the norm kk1 . Next, let .'n /n be a sequence in
 1 N
C .Œ0; 1/ such that limn!1 'n D ' and limn!1 'n0 D in the norm kk1 .
0
In particular, the sequence .'n /n converges pointwise to ' and the sequence 'n n
converges uniformly to , so since every 'n is C 1 , it follows that ' is C 1 and
D ' 0 . Therefore, the operator .A; D.A// is closed.

Now consider an operator .A; D.A// with D.A/ D H . Using the closed
graph theorem we have immediately that .A; D.A// is closed if and only if A is
bounded. Hence, every bounded operator is closed.

Example 2.2.5. Consider the differentiation operators

8
< D.Ak / ! L2 .0; 1Œ/
Ak W
: u 7! i u0 ;
14 Chapter 2. Detailed introduction to abstract spectral theory

where

D.A0 / WD H 1 .0; 1Œ/ ;

˚ 
D.A1 / WD u 2 H 1 .0; 1Œ/ I u.0/ D 0 ;
˚ 
D.A2 / WD u 2 H 1 .0; 1Œ/ I u.1/ D 0 ;
˚ 
D.A3 / WD u 2 H 1 .0; 1Œ/ I u.0/ D u.1/ and
˚ 
D.A4 / WD H01 .0; 1Œ/ D u 2 H 1 .0; 1Œ/ I u.0/ D u.1/ D 0 :

We claim that all these operators are closed with dense domains.
For the first operator: since H 1 .0; 1Œ/ is dense in L2 .0; 1Œ/ for the L2 -norm
and H 1 .0; 1Œ/ is a Banach space for the H 1 -norm, therefore the space D.A0/
equipped with the norm kkA0 is complete, consequently the operator A0 ; D.A0/
is closed. For the second operator: recall that the injection of H 1 .0; 1Œ/ into
C 0 .Œ0; 1/ is continuous (see [Bré]) hence D.A1 / is a closed subset of D.A0 /,
consequently the space D.A1 / equipped with the norm kkA1 is complete, it fol-
lows that A1 ; D.A1 / is closed. For the cases D.A2 /, D.A3 / and D.A4 / proceed
in the same manner.

2.3 Adjoint operators

2.3.1 Introduction
Let .A; D.A// be a closed unbounded linear operator with domain dense in H .
The adjoint of .A; D.A// is the closed unbounded linear operator A? with domain
D.A? / defined by:

D.A? / WD fy 2 H I 9 2 H; such that 8x 2 D.A/; hAx; yiH D hx; iH g :

In fact, here the vector is unique (see below). So the adjoint operator A? is
defined by 8
< D.A? / ! H
A W
?
: y 7! A? y WD :

Let us explain the uniqueness of . Suppose that there exists 1; 2 2 H 2 such

that for all x 2 D.A/,

hy; AxiH D h 1 ; xiH D h 2 ; xiH :

Then h 1  2 ; xiH D 0 for all x 2 D.A/, therefore 1  2 2 D.A/? and since

D.A/ is dense in E we have D.A/? D f0g, consequently 1 D 2.
 2.3 Adjoint operators 15

Note that the operator .A? ; D.A? // is always closed (even A is not closed) and
A?? D A. If we have two unbounded linear operators A; B such that A  B, then
B ?  A? . Indeed, let y 2 D.B ? /. Then hy; Bxi D hB ? y; xi for all x 2 D.B/.
Since the domain D.A/ is included in D.B/ and B coincides with A on D.A/,
˝ ˛
hy; Axi D B ? y; x

for all x 2 D.A/, therefore y 2 D.A? / and A? y D B ? y, i.e.: B ?  A? .

Example 2.3.1. Consider the differentiation operators

8
< D.Ak / ! L2 .0; 1Œ/
Ak W
: u 7! i u0 ;

where

D.A0/ WD H 1 .0; 1Œ/ ;

˚ 
D.A1/ WD u 2 H 1 .0; 1Œ/ I u.0/ D 0 ;
˚ 
D.A2/ WD u 2 H 1 .0; 1Œ/ I u.1/ D 0 ;
˚ 
D.A3/ WD u 2 H 1 .0; 1Œ/ I u.0/ D u.1/ and
˚ 
D.A4/ WD H01 .0; 1Œ/ D u 2 H 1 .0; 1Œ/ I u.0/ D u.1/ D 0 :

We have seen in a previous example that all these operators are closed with dense
domains. Here we determine their adjoints: we claim that A?0 D A4 , A?1 D A2 ,
A?2 D A1 , A?3 D A3 , and A?4 D A.
Let us show only the first equality. For all v 2 D.A?4 / and all u 2 Cc1 .0; 1Œ/ 
D.A4/, we have ˝ ? ˛
A4 v; u D hv; A4 ui ;
i.e., ˝ ˛ ˝ ˛
A?4 v; u D i v; u0 :
Therefore, for all u 2 Cc1 .0; 1Œ/  D.A4 /,
˝ ˛ ˝ ˛
v; u0 D i A?4 v; u

which means that v 2 H 1 .0; 1Œ/ and v 0 D iA?4 v. In others words A?4 v D iv 0 ,
and so A?4  A0 .
Now let g 2 D.A4 / D H01 .0; 1Œ/. Then for all u 2 D.A0 / we have
˝ ˛ ˝ ˛
hA0 u; gi D i u0 ; g D i u; g 0 ;

which shows that g 2 D.A?0 / and A?0 g D ig 0 . So A4  A?0 . It follows that

A0 D A??
0  A4 and since A4  A0 we finally have A0 D A4 .
? ? ?
16 Chapter 2. Detailed introduction to abstract spectral theory

Proposition 2.3.2 (von Neumann). For every unbounded linear operator A with
domain D.A/ dense in H we have
 
ker A? D .im.A//? : (2.2)

Proof. Let x 2 ker .A? /, i.e., A? x D 0. Then for all y 2 D.A/ we have
hx; Ayi D hA? x; yi D 0 and so x 2 .im.A//? .
Conversely, let x 2 .im.A//? , i.e., hx; Ayi D 0 for all y 2 D.A/. Then
x 2 D.A? / and A? x D 0. 
What happens if A is bounded? Using the Riesz–Fréchet representation theo-
rem we have D.A? / D H and the operator A? is also bounded, with jjjA? jjj D
jjjAjjj.

2.3.2 Symmetry and self-adjointness

Definition 2.3.3. An operator .A; D.A// is symmetric if A  A? . The operator
.A; D.A// is said to be self-adjoint if A D A? .

Thus an operator .A; D.A// is symmetric if and only if for any couple .x; y/ 2
D.A/2 we have
hAx; yiH D hx; AyiH :
In particular, for all x 2 D.A/

hx; AxiH D hAx; xiH D hx; AyiH

and so hAx; xiH 2 R.

Every self-adjoint operator is symmetric. If an operator .A; D.A// is symmet-
ric and D.A? /  D.A/ then .A; D.A// is self-adjoint.

Definition 2.3.4. A unbounded symmetric operator .A; D.A// is said to be posi-

tive if hAx; xiH  0 for all x 2 D.A/.

Remark 2.3.5. In the case of bounded operators symmetry and self-adjointness

coincide.

Proposition 2.3.6. For every bounded symmetric .self-adjoint/ operator T on H

we have
jjjT jjj D sup jhT x; xiH j :
kxkH 1

Proof. By (2.1), jjjT jjj D supkxkH ;kykH 1 jhT x; yiH j, whence

sup jhT x; xiH j  jjjT jjj :

kxkH 1
 2.3 Adjoint operators 17

Using the polarization identity, we see that for all vectors x; y 2 H such that
kxkH ; kykH  1
  
1 2 xCy xCy
jhT x; yiH j  kx C ykH T ;
4 kx C ykH kx C ykH H
  
1 x  y xy
C kx  yk2H T ;
4 kx  ykH kx  ykH H
1 1
 kx C yk2H c C kx  yk2H c
4 4
c
D 2 kxk2H C 2 kyk2H  c;
4
where cWD sup jhT x; xiH j. 
kxkH 1

We have also
Proposition 2.3.7. For every bounded positive symmetric .self-adjoint/ operator
T on H ,
kT xk2H  jjjT jjj jhT x; xiH j
for all x 2 H .
Proof. Consider the bilinear form a.x; y/ WD hT x; yiH on H 2 . Since T is posi-
tive and symmetric we have, by the Cauchy–Schwarz inequality,
ja.x; y/j2  a.x; x/ a.y; y/
for all x; y 2 H , whence
jhT x; yiH j2  hT x; xiH hT y; yiH
 hT x; xiH jjjT jjj kyk2H
for all x; y 2 H . Hence, for the normed vector y WD Tx
kT xkH
we get

kT xk2H  hT x; xiH jjjT jjj : 

2.3.3 An important result concerning self-adjoint operators

Observe first that for every symmetric operator .A; D.A// of H , for all x 2 D.A/
and for all  D  C i 2 C such that ¤ 0 we have
k.I  A/xk2H D h.I  A/ x C i x; .I  A/ x C i xiH
D k.I  A/ xk2H  i h.I  A/ x; xiH
C i hx; .I  A/ xiH C 2
kxk2H
D k.I  A/ xk2H C 2
kxk2H  2
kxk2H ;
18 Chapter 2. Detailed introduction to abstract spectral theory

because A is symmetric and  is real. Therefore,

k.I  A/ xk2H  2
kxk2H : (2.3)

Lemma 2.3.8. Let .A; D.A// be a symmetric operator of H . For all  D  C

i 2 C such that ¤ 0, ker.I  A/ D f0g and im.I  A/ is closed.

Proof. Using (2.3) we have

k.I  A/ uk2H  2
kuk2H

for all u 2 D.A/. Thus it is clear that ker .I  A/ D f0g. Next, consider
a Cauchy sequence yn D .I  A/ xn in im .I  A/ (obviously the sequence xn
belongs to D.A/). Using (2.3) we see that .xn /n is a Cauchy sequence of D.A/.
Since H is complete, there exists a couple .x; y/ 2 H 2 such that limn!1 xn D x
and limn!1 yn D y in the norm kkH . Therefore,

lim Axn D lim xn  yn D x  y

n!1 n!1

and since A is closed we have x 2 D.A/ and Ax D x  y. Consequently,

y D .I  A/x 2 im.I  A/. This means that the sequence yn D .I  A/xn
converges in im.I  A/, thus im.I  A/ is closed. 
So we deduce

Theorem 2.3.9. Let .A; D.A// be a symmetric operator of H . The following

properties are equivalent:
(i) A is self-adjoint.
(ii) A is closed and ker .A? ˙ iI / D f0g.
(iii) im.A ˙ iI / D H .

Proof.
 First we show .i/ ) .ii/. Suppose A self-adjoint, then A is closed, be-
cause A D A? . Next if w 2 ker .A?  iI / then A? w D iw, and we get
hA? w; wiH D i kwk2H . On the other hand, since A D A? , we also have
˝ ? ˛ ˝ ˛
A w; w H D hAw; wiH D w; A? w H D i kwk2H ;

whence w D 0. The case ker .A? C iI / is similar.

 Now let us prove .ii/ ) .iii/. Assume (ii) holds. Then .im.˙iI  A//? D
f0g because of Proposition 2.3.2, so we have im.˙iI  A/ D H . By
Lemma 2.3.8 (with  D i / the subspace im.˙iI  A/ is closed.
 2.4 Spectrums and resolvent set 19

 Finally, we show .iii/ ) .i/. Since im .A ˙ iI / D H we have (by Propo-

sition 2.3.2)  
ker A? ˙ iI D f0g:
Let v 2 D.A? /. Since .A?  iI / v 2 H , there exists u 2 D.A/ such that
.A?  iI / v D .A  iI / u. Since A is symmetric, .A?  iI / .v  u/ D 0
and it follows that v D u (because ker .A? ˙ iI / D f0g). This shows that
v 2 D.A/, hence A is self-adjoint. 

2.4 Spectrums and resolvent set

2.4.1 Spectrum versus point spectrum
Definition 2.4.1. Let AW D.A/  H ! H be a closed unbounded linear operator.
(i) A scalar  is an eigenvalue of A if ker.I  A/ ¤ f0g. The point spectrum
of A is the set of eigenvalues of A, denoted by Spec.A/ or p .A/.
(ii) A scalar  is said to belong to the resolvent set of A if the operator AI is
an isomorphism from D.A/ onto H and the inverse .I A/1 is a bounded
operator in H . The resolvent set of A is denoted by .A/.
(iii) The spectrum of A is the complement of the set .A/ in C. The spectrum
of A is denoted by .A/.

Let  be a fixed eigenvalue of A. An eigenvector associated to  is a vector

' 2 D.A/  H such that A' D '. The set of eigenvectors associated to  is
a vector space called the eigenspace associated to .
In the case of bounded operators the definitions of the spectra are similar
(for (ii) just replace D.A/ by H ). We have the inclusion
Spec.A/  .A/:
Indeed, if  … .A/ i.e.,  2 .A/, then in particular the operator .I  A/ is
injective; in other words, ker .I  A/ D f0g, i.e.,  … Spec.A/.
The spectrum of an operator is a closed set of C (see the following theorem).
Moreover, if the operator A is bounded, we have also the inclusion
.A/  BC .0; jjjAjjj/
where BC .0; r/ WD fz 2 CI jzj  rg. Indeed, for each z 2 C such that jj >
jjjAjjj the operator zI  A is invertible with bounded inverse (Corollary 2.1.8 of
von Neumann’s lemma), i.e., z 2 .A/. Therefore if A is bounded the spectrum
.A/ is a compact subset of C.
Remark 2.4.2. In the case of finite-dimensional spaces the spectrum and the point
spectrum coincide.
20 Chapter 2. Detailed introduction to abstract spectral theory

For any  2 .A/ we denote by RA ./ the bounded operator .I  A/1 .
This operator is called the resolvent operator of A at the point .
Example 2.4.3. Consider the family of derivation operators
8
< D.Ak / ! L2 .0; 1Œ/
Ak W
: u 7! i u0 ;
where
D.A0 / WD H 1 .0; 1Œ/ ;
˚ 
D.A1 / WD u 2 H 1 .0; 1Œ/ I u.0/ D 0 ;
˚ 
D.A2 / WD u 2 H 1 .0; 1Œ/ I u.1/ D 0 ;
˚ 
D.A3 / WD u 2 H 1 .0; 1Œ/ I u.0/ D u.1/ and
˚ 
D.A4 / WD H01 .0; 1Œ/ D u 2 H 1 .0; 1Œ/ I u.0/ D u.1/ D 0 :
We have: .A0 / D C, .A1 / D .A2 / D ;, .A3 / D f2i  n; n 2 Zg, .A4/ D
C.
 Start with A0 and note that for any constant  2 C the function x 7!
u.x/ D eix with x 2 Œ0; 1 is a solution of the ODE
i u0  u D 0:
This means that for all  2 C we have
ker .A0  I / ¤ f0g;
hence the point spectrum of A0 is “full”, i.e., Spec.A0 / D C, therefore
.A0/ D C.
 For A1 (and similarly for A2 ), for any function f and any  2 C, using
classical ODE techniques one can verify that the equation
i u0  u D f
with u.0/ D 0 has a unique solution, namely
Z x
ix ix
u.x/ D u.0/e e i ei t f .t/ dt
0
Z x
ix
D e i ei t f .t/ dt:
0

Hence, for any  2 C we have  2 .A1 /, and so .A1/ D ; and for all
2C Z x
ix
.R A1 / u.x/ D e i ei t f .t/ dt:
0
 2.4 Spectrums and resolvent set 21

 For A3 : for any function f and for any  2 C, using classical ODE tech-
niques one can verify that the solutions of the equation

i u0  u D 0

are given by u.x/ D Ceix with C 2 C. The condition

u.0/ D u.1/ ”  D 2 n; n 2 Z;

hence Spec.A3 / D f2i  n; n 2 Zg, and f2i  n; n 2 Zg  .A3/. Next for

 2 C such that  ¤ 2i  n for any integer n, the equation

i u0  u D f

has a unique solution, namely

Z x
ix ix
u.x/ D u.0/e e i ei t f .t/ dt:
0

Imposing the boundary conditions (u.0/ D u.1// we get

Z x
1
u.0/ D ix i ei.t 1/ f .t/ dt;
e 1 0
so we deduce (arguing as for the operator A1 ) the expression of the solution
x 7! u.x/ and the expression of R A3 . We see that im.A3  I / D
L2 .0; 1Œ/. Finally, if  2 C, such that  ¤ 2i  n for any integer n, then
 2 .A3 /; therefore .A3 /  f2i  n; n 2 Zg.
 To finish we look at the operator A4 . Since for all  2 C
˚ 
im .A0  I / D i u0  uI u 2 H01 .0; 1Œ/

and since for all u 2 H01 .0; 1Œ/ we have, using integrating by parts, that
Z Z Z
1  0
 1
ix 0
1
e ix
i u .x/  u.x/ dx D i e u .x/ dx   eix u.x/ dx
0 0 0
Z 1 Z 1
D eix u.x/ dx   eix u.x/ dx
0 0
D 0;

it follows that for any  2 C we have .im .A0  I //? ¤ f0g, so  2

.A4 /. We conclude that .A4 / D C.
22 Chapter 2. Detailed introduction to abstract spectral theory

Theorem 2.4.4. Let AW D.A/  H ! H be a closed unbounded linear operator.

(i) For all .; / 2 .A/2 we have the resolvent equation

RA ./  RA ./ D .  /RA ./RA ./: (2.4)

(ii) For all .; / 2 .A/2 the resolvents of A in  and in  commutes

RA ./RA ./ D RA ./RA ./:

(iii) The resolvent set .A/ is an open subset of C.

(iv) The mapping 8
< .A/ ! B.H /
RA W
:  7! R ./;
A

0
is holomorphic on .A/ and for all  2 .A/ we have RA ./ D  .RA .//2 .
(v) For all  2 .A/ we have

ARA ./ D RA ./  I 2 B.H /: (2.5)

Proof.
 Start with (i). For all .; / 2 .A/2 ,

RA ./  RA ./ D .I  A/1  .I  A/1

D .I  A/1 I  .I  A/ .I  A/1
D .I  A/1 ..I  A/  .I  A// .I  A/1
D .  / .I  A/1 .I  A/1 :

 For item (ii): the case  D  is clear. So, suppose  ¤ . Using (i) we
have
RA ./  RA ./
RA ./RA ./ D

RA ./  RA ./
D D RA ./RA ./:


 The argument for (iii) is the following: for each fixed 0 2 .A/ and for all
 2 C,

.0 C / I  A D 0 I  A C I D .0 I  A/ .I C RA .0 /I / :

 2.4 Spectrums and resolvent set 23

Now for all  2 C such that jj < jjjRA1.0 /jjj (i.e., jjjRA .0 /jjj < 1),
the von Neumann’s lemma shows that .0 C / I  A is invertible with
a bounded inverse (so 0 C  2 .A// and
C1
X
RA .0 C / D ./k .RA .0 //k :
kD0

In particular, for each 0 2 .A/ there exists a radius r WD jjjRA1.0 /jjj such
that the open disc centered at 0 and of radius r is a subset of .A/; in other
words, .A/ is an open set.
 Next, item (iv): for any 0 2 .A/ and for any complex number z such that
jz  0 j < jjjRA1.0 /jjj we have seen that .0 C / I  A is invertible with
a bounded inverse and
C1
X
RA .z/ D .zI  A/1 D .0  z/k .RA .z//kC1 :
kD0

This formula show that  2 .A/ 7! RA ./ is analytic, hence holomor-

phic. Moreover, using the proof of (ii) we get
0 RA .z/  RA ./
RA ./ D lim D  .RA .//2 :
z! z
 Finally, let us prove (v). For any  2 .A/ we have .I  A/ RA ./ D I ,
whence
RA ./  I D ARA ./: 
Next we give an important theorem concerning the spectrum of resolvent op-
erators:

Theorem 2.4.5. Let AW D.A/  H ! H be a closed unbounded linear operator

such that .A/ ¤ ;. For any 0n 2 .A/ we have o
(i) Spec .RA .0 //  f0g D 01 I  2 Spec.A/
n o
(ii) .RA .0 //  f0g D 01 I  2 .A/ .

Proof.
 Start with item (i): if  2 Spec .RA .0 // with  ¤ 0, then there exists
' 2 H; ' ¤ 0, such that RA .0 /' D ', so ' 2 D.A/ and
 
1 1
A' D A .RA .0 /'/ D .0 RA .0 /  I / '
 
 
0 1 1
D '  ' D 0  ';
  
24 Chapter 2. Detailed introduction to abstract spectral theory

so with  WD 0  1

we have  2 Spec.A/ and  D 1
0 
.
Conversely, arguing in the same way we have also
1
I  2 Spec.A/  Spec .RA .0 //  f0g:
0  

 Let us prove item (ii): for all  2 C? and for all 0 2 .A/ we have
  
1
.  RA .0 // ' D  0  I  A RA .0 /'

for any vector ' 2 H (because ..0 1/ I A/ RA .0 / D 0 RA .0 /
RA .0 /  0 RA .0 / C I . Moreover, for any vector 2 H ,
   
1
.  RA .0 // D  RA .0 / 0  I A '

  
hence if x 2 ker .I  RA .0 // then x 2 D.A/ and 0  1 I  A x 2
ker .RA .0 // D f0g. Consequently,
  
1
x 2 ker 0  I A :

  
We also have ker 0  1 I  A  ker .I  RA .0 //.
Next, let y 2 im .I  RA .0 //. Then  there exists
 x  2 H such that
y D .I  RA .0 // x hence y 2 im 0  1 I  A . Now, if y 2
     
im 0  1 I A , there exists x 2 D.A/ such that y D 0  1 I A x
and there exists a unique w 2 H such that x D RA .0 /w, therefore y 2
im .I  RA .0 //. So we have proved that
  
1
ker 0  I  A D ker .I  RA .0 //

and   
1
im 0  I  A D im .I  RA .0 // :

It follows that
 
1
 2 .RA .0 // ” 0  2 .A/ with  ¤ 0;

i.e.,  
1
 2 .RA .0 // ” 0  2 .A/ with  ¤ 0;

n o
so we get .RA .0 //  f0g D 01 ;  2 .A/ . 
 2.4 Spectrums and resolvent set 25

2.4.2 The self-adjoint case

Theorem 2.4.6. If an unbounded operator A is self-adjoint, then Spec.A/ 
.A/  R. Moreover, the eigenspaces of A are orthogonal to one another.

Proof. Let  2 Spec.A/, i.e., there exists x ¤ 0 in D.A/ such that Ax D x. We
have
hAx; xiH D  kxk2H :
On the other hand,

hAx; xiH D hx; AxiH D  kxk2H

so we deduce that  2 R, hence Spec.A/  R. Now for  2 C such that

im./ ¤ 0, Lemma 2.3.8 shows that

ker.I  A/ D f0g;

hence
.im .I  A//? D f0g
because  2 R and A D A? . So we have im .I  A/ D H and by Lemma 2.3.8
the subspace im.I  A/ is closed. So we have im.I  A/ D H and the operator
I  A is surjective. Consequently, if  2 .A/, then I  A is not surjective, it
follows that Im./ D 0. Hence .A/  R.
To finish, let 1 and 2 be two distinct (real) eigenvalues of A and consider
two associated eigenvectors x1 and x2 . We have

hAx1 ; x2 iH D 1 hx1 ; x2 iH :

On the other hand,

hAx1 ; x2 iH D hx1 ; Ax2 iH D 2 hx1 ; x2 iH :

Therefore, hx1 ; x2 iH D 0: 

We have also

Proposition 2.4.7. For every bounded self-adjoint operator T on H let us denote

m WD infkxkH 1 jhT x; xiH j and M WD supkxkH 1 jhT x; xiH j. Then .T / 
Œm; M  and the numbers m; M belong to .T /.

Proof. Let us prove the statement for M (for m the proof is similar). Consider
a fixed f 2 H and  2 R with  > M . Then for all x 2 H we have

hT x; xiH  M kxk2H ;
26 Chapter 2. Detailed introduction to abstract spectral theory

hence

h.I  T / x; xiH   kxk2H  M kxk2H D .  M / kxk2H :

Now consider the symmetric (because T is self-adjoint) bilinear form a.x; y/ WD

h.I  T / x; yiH . It is clear that this form is bounded (i.e., continuous) and for
all x 2 H we have a.x; x/  .  M / kxk2H , so this form is elliptic (coercive)
on H . Next, consider the bounded linear form `.y/ WD hf; yiH . Using the Lax–
Milgram theorem we see that for all f 2 H there exists a unique vector x ? 2 H
(which depends on f ) such that for all y 2 H

a.x ? ; y/ D `.y/

and kx ? kH  K kf kH , with a constant K  0. In other words the problem: for

a fixed f 2 H , find x 2 H such that

h.I  T / x; yiH D hf; yiH

for all y 2 H , admits a unique solution x ? 2 H . This problem is equivalent to

the problem: for a fixed f 2 H , find x 2 H such that

.I  T / x D f:

In fact, we have shown that for every f 2 H there exists an unique solution x ? 2
H , consequently the operator I T is bijective and x ? D .I  T /1 f . More-
over, kx ? kH D k.I  T /1 f kH  K kf kH , so we deduce that .I  T /1 is
bounded, thus the operator .I  T / is invertible. Summing up, if  > M , then
 2 .T /, i.e., if  2 .T / then   M .
Next let us verify that M 2 .T /: if M 2 .T /, then the operator .MI  T /
is invertible and consequently .MI  T /1 is bounded. Then by the definition of
M there exists a sequence .xn /n in H with kxn kH D 1 such that

lim h.MI  T / xn ; xn iH D 0:
n!1

On the other hand since for all kxkH  1 we have

h.MI  T / x; xiH D M hx; xiH  hT x; xiH  0;

the operator MI  T is self-adjoint and positive, consequently for all kxkH  1

(Proposition 2.3.7)

k.MI  T / xk2H  jjjMI  T jjj jh.MI  T / x; xiH j :

Applying this to the sequence .xn /n we have

lim k.MI  T / xn kH D 0;
n!1
 2.5 Spectral theory of compact operators 27

hence since xn D .MI  T /1 .MI  T / xn , we get

ˇˇˇ ˇˇˇ
ˇˇˇ ˇˇˇ
kxn kH  ˇˇˇ.MI  T /1 ˇˇˇ k.MI  T / xn kH :

Finally we deduce that limn!1 kxn kH D 0, which is absurd (because kxn kH D

1). It follows that M 2 .T /. 

Corollary 2.4.8. Let T be a bounded self-adjoint operator on H . If .T / D f0g,

then T D 0.

Proof. In particular, M D 0, so using Proposition 2.3.6 we get

0D sup jhT x; xiH j D jjjT jjj : 

kxkH 1

2.5 Spectral theory of compact operators

2.5.1 The notion of compact operators
Definition 2.5.1. Let A be a bounded operator on H and denote by BH .0; 1/ WD
fx 2 H I kxkH < 1g the open ball of radius 1 in H . We say that A is a compact
operator if the closed set A .BH .0; 1// is compact.

Proposition 2.5.2. A bounded operator A is compact if and only if for any boun-
ded subset X of H , the closed set A.X / is a compact subset of H .

Proof. Suppose A is compact and let X be a bounded subset of H . There exists

r > 0 such that X  BH .0; r/ WD fx 2 H I kxkH < rg. Let .yn /n be a sequence
 a sequence .xn /n in X such that for all n, yn D A.xn /.
in A.X /, so there exists
For any n we have A xrn D 1r A .xn / and xrn 2 BH .0; 1/, thus the sequence
  xn 
A r n belong to A.BH .0; 1//. Since A .BH .0; 1// is compact one can sup-
  
pose, by extracting a subsequence if necessary, that the sequence A xrn n con-
verge in A .BH .0; 1//. It follows that the sequence yn D .A .xn //n converge in
A.X /.
To prove the converse, it suffices to take X D BH .0; 1/. 

The Riesz lemma asserts that the closed unit ball BH .0; 1/ WD
fx 2 H I kxkH  1g is compact if and only if dim.H / < C1. Thus every fi-
nite rank operator (i.e., a bounded operator A such that dim .im.A// < C1) is
compact.
28 Chapter 2. Detailed introduction to abstract spectral theory

2.5.2 Algebraic properties of compact operators

The algebraic structure of the set of compact operators is described as follows.

Theorem 2.5.3. The set of compact operators of H is a closed subspace of the

space of bounded operators. Moreover, this set is also a (two-sided) ideal of the
algebra of bounded operators of H .

Proof. Let T; S be two compact operators and ˛ 2 C. Obviously,

.˛T C S / .BH .0; 1//  ˛T .BH .0; 1// C S .BH .0; 1// ;
so
.˛T C S / .BH .0; 1//  ˛T .BH .0; 1// C S .BH .0; 1//:
Hence, since T; S are compact operators the set ˛T .BH .0; 1// C S .BH .0; 1//
is compact; indeed, if K1 ; K2 are two compact sets, the set ˛K1 C K2 is also
compact, because K1 K2 is compact and the map
8
< K1 K2 ! ˛K1 C K2
: .x; y/ 7! ˛x C y;

is continuous. Therefore, the set .˛T C S / .BH .0; 1// is compact, i.e., the opera-
tor ˛T C S is compact.
Now let us consider a sequence of compact operators .Tn /n such that
limn!1 Tn D T in the normSjjjjjj. Let " > 0, then there exists N 2 N such
that jjjTN  T jjj < 2" . Since y2H BH y; 2" H we have, in particular,
[ "
TN .BH .0; 1//  BH y; :
2
y2H

By the compactness of TN .BH .0; 1//, there is a finite set .y1 ; y2 ; : : : ; yk / in H

such that
[k
"
TN .BH .0; 1//  BH yi ; :
2
i D1

Let x 2 BH .0; 1/, then k.TN  T / xkF < "

2
and

TN x 2 TN .BH .0; 1//  TN .BH .0; 1//;

 
so there exists ` 2 f1; : : : ; kg such that TN x 2 BH y` ; 2" . Therefore, T x 2
BH .y` ; "/, hence
[k
T .BH .0; 1//  BH .yi ; "/:
i D1
 2.5 Spectral theory of compact operators 29

Consequently, the set T .BH .0; 1// is compact, i.e., the operator T is compact. It
follows that set of compact operators is closed.
Let A be a bounded operator on H and T a compact operator on H . First,

A .T .BH .0; 1///  A T .BH .0; 1// ;

thus
A .T .BH .0; 1///  A T .BH .0; 1//

and since T .BH .0; 1// is compact and A is continuous (because is bounded), the
set A T .BH .0; 1// is compact. Hence the closed set A .T .BH .0; 1/// is com-
pact, i.e., AT is compact. To finish, (since A is bounded) we have the inclusion
A .BH .0; 1//  A .BH .0; jjjAjjj//, therefore

T .A .BH .0; 1///  T .A .BH .0; jjjAjjj///

thus
T .A .BH .0; 1///  T .A .BH .0; jjjAjjj///:
Here the set T .A .BH .0; jjjAjjj/// is compact (see the previous proposition), so
the closed set T .A .BH .0; 1/// is compact, consequently TA is compact. 
Another useful result is the following proposition.

Proposition 2.5.4. Let T be a compact operator in an infinite-dimensional Hilbert

space H . Then for every orthonormal sequence .en /n in H we have T .en / ! 0
in the norm kkH as n ! 1.

Proof. Suppose T .en / does not converge to 0, i.e., there exists m > 0 such that
for any integer n we have kT en k2H  m. Since the sequence .en /n is bounded
(it is orthonormal) and since T is compact, there exists a compact subspace X of
H such that for any n we have T en 2 X . Thus one can suppose, by extracting
a subsequence if necessary, that the sequence .T en /n converges to a vector y 2
X  H . In particular, kyk2H  m. Moreover,
˝ ˛
0 < m  hy; yiH D lim hT en ; yiH D lim en ; T ? y H
n!C1 n!C1
PC1
and hen ; T ? yiH ! 0 as n ! C1 because hT ? y; en iH D nD0 jhT ? y; en iH j2 <
C1, and this is absurd. 
For finish recall (without proof) the:

Theorem 2.5.5 (Schauder theorem). Let T be a compact operator on H . Then

the adjoint T ? is a compact operator on H .
30 Chapter 2. Detailed introduction to abstract spectral theory

2.5.3 Main spectral theorem for compact self-adjoint operators

The main result concerning compact self-adjoint operators is the following

Theorem 2.5.6. Let T be a compact self-adjoint operator on a infinite-dimen-

sional separable Hilbert space H such that 0 … Spec.T /. Then the spectrum of T
is a real discrete set: it consists of an infinite sequence .k /k of real eigenvalues
with finite multiplicity such that k ! 0 as k ! C1. Moreover, the associated
eigenfunctions .ek /k0 form a Hilbert basis of the space H .

In particular, if the operator T is positive the eigenvalues .k /k are positive.

In this case we can suppose that
0  1      k1  k ! 0:

2.5.4 A proof using the Riesz–Schauder theory

The key ingredient of our proof for Theorem 2.5.6 is the following result:

Lemma 2.5.7. Let T be a compact operator on a separable Hilbert space H .

Then
(i) ker.I  T / is a finite-dimensional space.
(ii) The bounded operator I  T restricted to the subspace .ker.I  T //? is
invertible .with continuous inverse/ from .ker.I  T //? onto im.I  T /.
(iii) The subspace im.I  T / is closed.
(iv) ker.I  T / D f0g , im.I  T / D H .

Proof.
 Start with (i). Denote B WD BH .0; 1/ \ ker.I  T /. Then T .B/ D B.
Since T is compact and B is bounded, T .B/ is a compact subset of H , so
B is also compact. Thus the Riesz lemma1 applied to space ker.I  T /
implies that the dimension of ker.I  T / is finite.
 Item (ii): It is clear that the operator I  T restricted to the subspace
.ker.I  T //? is injective and surjective onto image im.Id  T /. Thus
the operator:

S WD I  T W .ker.I  T //? ! im.I  T /

is bijective. Now let us verify that its inverse

S 1 D .I  T /1 W im.I  T / ! .ker.I  T //?

1 Riesz lemma: in a Hilbert space H the closed unit ball is compact if and only if the dimension of H

is finite.
 2.5 Spectral theory of compact operators 31

is continuous. Suppose S 1 is not bounded: there existsa sequence

 .yn /n 2
N 
.im.I  T // with kyn kH D 1 for all n and such that S yn H ! C1
1

as n ! C1. So for any n we have

yn D .I  T / S 1 .yn / D S 1 yn  T S 1 yn ;
yn
and ! 0 as n ! C1. Let
kS 1 yn kH
 
1 yn S 1 yn
xn WD S 1
D :
kS yn kH kS 1 yn kH

Then for any n, kxn k D 1 and xn 2 im.S 1/ D .ker.I  T //? . Moreover,
xn  T xn ! 0 as n ! C1. Since the sequence .T xn /n belongs to T .B/
and since T .B/ is compact one can assume, by extracting a subsequence
if necessary, that the sequence .T xn /n converge to a vector x 2 T .B/.
Thus the sequence .xn /n converge also to x and since .xn /n belongs to the
closed subspace .ker.I  T //? we have x 2 .ker.I  T //? . But we also
have x  T x D 0; in other words, x 2 ker.I  T /, therefore x D 0 and
this is absurd (because for any n, kxn k D 1, so kxk D 1).
 Item (iii): The subspace

im.I  T / D S .ker.I  T //?

is closed because by (ii) the operator S is invertible (in particular bijective

and bi-continuous) and .ker.I  T //? is closed.
 Item (iv): First, suppose ker.I  T / D f0g and im.I  T / ¤ H . Let
consider
H1 WD S.H / D im.S /  H:
By hypothesis, H1 ¤ H , in fact since H1 D im.I  T / is closed by (iii),
this space admits an orthogonal complement in H . Therefore, there exists
a vector e1 2 H with ke1 k D 1 such that e1 2 H1? . Now consider the
space
H2 WD S.H1 / D S.S.H //  H1 :
We have H2 ¤ H1 : otherwise, S.S.H // D S.H / whence S.H / D H be-
cause S is bijective by (ii), and then H1 D H , which is absurd. Moreover,
the space H2 is closed: indeed, H2 D S.H1 / with H1 closed and S 1
is continuous. Consequently since H2 is closed this space admits an or-
thogonal complement in H1 . Therefore, there exists a vector e2 2 H with
ke2 k D 1 such that e2 2 H1 \ H2? . By induction we construct a sequence
of closed subspaces
Hn WD S.Hn1 /
32 Chapter 2. Detailed introduction to abstract spectral theory

such that Hn  Hn1 with Hn ¤ Hn1 , as well as a sequence of vec-

?
tors en 2 H with ken k D 1 such that en 2 Hn \ HnC1 . By construc-
tion, Sen 2 HnC1 . It follows that Sen is orthogonal to en and by the
Pythagorean theorem
kT en k2H D k.I  S / en k2H D ken k2H C kSen k2H  ken k2H D 1:
But this is absurd, because the operator T is compact (see Proposition
2.5.4). Finally, im.I  T / D H .
Conversely, suppose im.I  T / D H . Since im.I  T / D ker .I  T ? /?
(see Proposition 2.3.2) we get ker .I  T ? / D f0g and therefore (with the
implication ) of (iv) applied to the compact operator T ? ) we deduce that
im.I  T ? / D H , so ker .I  T / D f0g. 

Corollary 2.5.8 (Riesz–Schauder). Let T be a compact operator on a separable

Hilbert space H . Then .T /f0g D Spec.T /f0g and every non-null eigenvalue
has a finite multiplicity.

Proof. Obviously, Spec.T /  f0g  .T /  f0g. Conversely, let  2 C? . Without

loss generality one can suppose, upon replacing T by the compact operator 1 T
if necessary, that  D 1. If 1 … Spec.T /, then using item (iv) of the previous
theorem we have im.I  T / ¤ H and thus 1 … .T /. The assertion concerning
the finite multiplicity is a also a direct consequence of the previous theorem. 

Proposition 2.5.9. Let T be a compact operator on a separable infinite-dimen-

sional Hilbert space H . Then 0 2 .T /.

Proof. If 0 … .T /, then 0 is in the resolvent set of T , the operator T is invertible,

and I D T T 1 . It follows that I is compact (because T is compact and T 1
bounded). Consequently, I .BH .0; 1// D .BH .0; 1// is closed, but this is absurd
because dim.H / D C1 (using the Riesz lemma). 
The last part in our proof of Theorem 2.5.6 is

Lemma 2.5.10. Let T be a compact operator on a separable infinite-dimensional

Hilbert space H . Then every element of .T /  f0g is isolated.

Proof. Suppose  2 .T /  f0g is not isolated in .T /  f0g. Then there exists

a sequence .n /n 2 . .T /  f0g/N such that n !  as n ! C1. By Corollary
2.5.8, for all n the scalar n is a non-null eigenvalue of T , so let us denote by en
an associated eigenvector (with ken kH D 1) and consider the sets En ; E defined
by En WD span fek ; 1  k  ng and
C1
[
E WD En D span fen ; n 2 Ng :
nD0
 2.5 Spectral theory of compact operators 33

For any n  1 we have En  EnC1 and En ¤ EnC1 . Indeed, assuming that

En D EnC1 , there exists .˛1 ; : : : ; ˛n / 2 Cn such that
X
n
enC1 D ˛i ei ;
i D1

whence
X
n
nC1 enC1 D ˛i nC1 ei ;
i D1
and also
X
n
T enC1 D ˛i T ei ;
i D1
i.e.,
X
n
nC1 enC1 D ˛i i ei :
i D1
It follows that
X
n
˛i .nC1  i / ei D 0
i D1
and consequently ˛i D 0 for all i 2 f1; : : : ; ng. Therefore, enC1 D 0, which is
absurd. Now since for any integer n  1 we have En  EnC1 and En ¤ EnC1 ,
we deduce that dim.E/ D C1.
Next, using the Gram–Schmidt orthogonalization process we construct a Hil-
?
bert basis .fn /n of E: f1 WD e1 and for all n  2 we have fn 2 En \ EnC1 . Note
Xn
that Tfn  n fn 2 En1 : indeed if fn D ˛i ei , then
i D1

X
n
Tfn D ˛i i ei
i D1
so
X
n1
Tfn  n fn D ˛i .i  n / ei 2 En1 :
i D1
It follows that the vector Tfn  n fn is
 orthogonal
 to the vector fn .
To finish we consider the sequence 1n fn n . This sequence belong the closed
ball BH .0; min1 / . Since Tfn  n fn 2 En1 and Tfn  n fn is orthogonal
i
to fn , for all m < n:
 
Tfn Tfm T  n Id T  m Id ?
 D fn  fm  fm Cfn 2 En1 ˚span ffn g :
n m n m
34 Chapter 2. Detailed introduction to abstract spectral theory

Therefore, by the Pythagorean theorem,

 
 Tfn Tfm 
 
      kfn kH D 1:
n m H
  
Consequently, the sequence T 1 n fn n does not have Cauchy subsequences,
so the operator T is not compact, which is absurd. Finally every  2 .T /  f0g
is isolated. 

Corollary 2.5.11. Let T be a compact operator on a separable infinite-dimen-

sional Hilbert space H . Then
 .T / D f0g, or
 .T /  f0g is a sequence converging to 0.

Proof. If " > 0 is given, using the previous lemma and the fact that the spec-
trum .T / is compact, we conclude that the intersection of .T / with the set
f 2 CI jj  "g is empty or finite. 
All this leads to the following result.

Theorem 2.5.12. Let T be a compact self-adjoint operator on a infinite-dimen-

sional separable Hilbert space H such that 0 … Spec.T /. Then the spectrum of
T is a real discrete set, namely, it consists of an infinite sequence .k /k of real
eigenvalues with finite multiplicity such that k ! 0 as k ! C1. Moreover, the
associated eigenfunctions .ek /k0 form a Hilbert basis of H .

Proof. If .T / D f0g, then by Corollary 2.4.8, T D 0, and the result is trivial, so

we suppose that .T / ¤ f0g. First, since .T /  f0g D Spec.T /  f0g (Corollary
2.5.8), we have .T /f0g D Spec.T /. Moreover, by Lemma 2.5.10, the bounded
set .T /  f0g is discrete. In other words,
.T /  f0g D Spec.T / D fn ; n  1g
and every eigenvalue k has a finite multiplicity (Corollary 2.5.8). Next, the
eigenspaces Ek WD ker .T  k I / are all finite-dimensional and orthogonal to
one another (Theorem 2.4.6). For every eigenspace Ek WD ker .T  k Id / we
can choose a finite orthogonal basis (here the eigenvalues are repeated according
to their multiplicities). We thus get a family of vectors .en /n1 such that i ¤
j ) ei ? ej . Denote F WD span fen ; n  1g. We claim that F is dense in H ,
i.e., F ? D f0g. By construction, the family .en /n1 is orthonormal. Moreover,
Xn Xn
T .F /  F : if x D ˛i ei 2 F , then T x D ˛i i ei 2 F . And since T is
i D1 i D1
self-adjoint, we also have T .F ? /  F ? : if x 2 F ? , then for all y 2 F
0 D hx; T yiH D hT x; yiH
 2.6 The spectral theorem multiplication operator form 35

hence T x 2 F ? . Therefore, we can consider the restriction S of the operator T

to F ? 8
< F ? ! F ?
SW
:
x 7! S x WD T x:
The operator S is compact and self-adjoint (because T is compact and self-adjoint).
Then by Corollary 2.5.11 we have two cases: if .S / D f0g, then S D 0 (by
Corollary 2.4.8), hence F D H . Else, if .S / ¤ f0g, then Spec.S / ¤ f0g, thus
there exists an eigenvector e 2 F ? , e ¤ 0, of S , Se D e. But then

T e D e;

i.e., the vector e is an eigenvector of T , thus e 2 F . Therefore e 2 F \ F ? ,

which implies that e D 0. This is absurd, so F ? D f0g.
Finally, the sequence .en /n1 is a Hilbert basis of H , and since the operator T
is compact, T .en / ! 0 as n ! 1 (Proposition 2.5.4), i.e., n en ! 0 as n ! 1
and we deduce that n ! 0 as n ! 1. 

2.6 The spectral theorem multiplication operator form

for unbounded operators on a Hilbert space
2.6.1 Spectral theorem multiplication operator form
Start by the following lemma:

Lemma 2.6.1. Let .X; F ; / be a measure space and F a ˚real-valued function

finite -almost
 everywhere on X . Then the set D .MF / WD ' 2 L2 .X /I F ' 2
L .X / is dense in L2 .X / and the multiplication operator defined by:
2

˚ 
MF W D .MF / WD ' 2 L2 .X /I F ' 2 L2 .X / ! L2 .X /
' 7! F '

is self-adjoint. Moreover, if F is bounded, then the operator MF is bounded on

L2 .X / with jjjMF jjj D kF k1 .

Proof. First, let us verify that the domain D.MF / is dense in L2 .X /. Let ' 2
L2 .X / and consider the sequence .'n /n of L2 .X / defined by 'n WD ' .jF jn/
where denotes the indicator function. Since for any n 2 N, jF 'n j  nj'j, the
sequence .'n /n belongs to D.MF /. Next, since F is finite almost everywhere,
the sequence .'n /n converge almost everywhere on X to '. Since for all n 2 N,
j'n j  j'j 2 L2 .X /, using the Lebesgue dominated convergence theorem shows
36 Chapter 2. Detailed introduction to abstract spectral theory

that the sequence .'n /n converges in the L2 .X /-norm on X to '. Hence the
domain D.MF / is dense in L2 .X /.
Now, let us verify that the operator MF is closed. Consider a sequence .'n /n
of D.MF / such that 'n converge in the L2 .X /-norm on X to some function '.
And suppose also that the sequence .MF 'n /n converge in the L2 .X /-norm on X
to a function 2 L2 .X /. Using the Lebesgue dominated convergence theorem
one can suppose, by extracting a subsequence if necessary, that .'n /n converge al-
most everywhere on X to '. Consequently, the sequence .F 'n /n converge almost
everywhere on X to F ', hence D F ' almost everywhere on X . Therefore,
' 2 D.MF / and MF .'/ D .
To complete the proof, let us show that the operator MF is self-adjoint. Since
F is a real-valued function, the operator MF is symmetric  on its domain
 D.MF /.

To see that MF is self-adjoint
  we use the criterion ker MF ˙ iI D f0g (Theo-

rem 2.3.9): let ' 2 ker MF  iI ; then MF .'/ D i'. On the other hand, since
for all 2 D.MF / we have
˝  ˛
MF '; L2 D h'; MF iL2
we get for all 2 D.MF /
Z Z
i' d D 'F d
X X
R
i.e., for all 2 D.MF /, X '.i  F / d D 0: Since D.MF / is dense 2
  in L .X /
and F is a real-valued function, we  ' D 0, hence ker MF  iI D
 obtain that
f0g. By the same argument, ker MF C iI D f0g. The assertion for the case
F 2 L1 is trivial. 
In the spectral theory of C algebras [Arv1], [Arv2] the following result plays
an important role: (Recall that a bounded operator A is normal if AA? D A? A
and a unitary map is an isometric and surjective map).

Theorem 2.6.2 (Spectral theorem for bounded normal operators). Let N be a

bounded normal operator on a separable Hilbert space H . There exist a finite
measure space .X ,/ and F a bounded function on X such that N is unitarily
equivalent to the operator of multiplication by F in L2 .X / in the sense of the
previous lemma.

Notation 2.6.3. Let A be a self-adjoint operator with domain D.A/ on a separable

Hilbert space H . Since Spec.A/  R; the complex numbers ˙i are not in the
spectrum of A. We denote R˙i D .˙iI C A/1 .
Theorem 2.6.2 has the following

Corollary 2.6.4. With the previous notations, there exist a finite measure space
.X ,/, a unitary operator U W H ! L2 .X /, and a bounded function F with
 2.6 The spectral theorem multiplication operator form 37

F ¤ 0 almost everywhere on X , such that

URi U 1 D MF :

Proof. First let us verify that the operators R˙i are normal. Since Im .˙iI CA/ D
H , for any couple of vectors .u; v/ 2 H 2 there exists an unique couple .z; w/ 2
D.A/2 such that v D .iI C A/ z and u D .iI C A/ w. Hence we get
˝ ˛
hv; Ri uiH D h.iI C A/ z; wiH D z; .iI C A/ w H
D hz; .iI C A/ wiH D hRi .v/; uiH :

Consequently, R˙i D Ri . On the other hand, by the resolvent equation (2.4)
the operator Ri and Ri commute, therefore Ri and Ri are bounded normal
operators.
Next we use the Theorem 2.6.2: we must verify that the function F ¤ 0 almost
everywhere on X . By its definition the operator Ri is injective, hence by unitary
conjugation MF is also injective, i.e., F ¤ 0 almost everywhere on X . 
From this result we get the usual result:

Corollary 2.6.5 (Spectral theorem multiplication operator form). In the previous

notations, there exist a finite measure space .X; /, a unitary operator U W H !
L2 .X /; and f a real-valued function finite -almost everywhere on X such that:
v 2 D.A/ , U.v/ 2 D.Mf / and on the set U.D.A//

UAU 1 D Mf :

Proof. Let F be the function provided by Corollary 2.6.4. Consider the function
f WD F1  i: Then f is finite -almost everywhere on X (because F ¤ 0 almost
everywhere on X ). By construction, F .f C i / D 1 and URi D MF U . First
we show that v 2 D.A/ , U.v/ 2 D.Mf /. Let v 2 D.A/; since the operator
Ri is invertible, there exists a unique u 2 H such that v D Ri u. Consequently,
U v D MF U u: Multiplying this equality by f we obtain f U v D .1  iF /U u.
Since U u 2 L2 .X / and F is bounded, we get f U v 2 L2 .X /, i.e., U v 2 D.Mf /.
Conversely, let U v 2 D.Mf / be a vector, so we have .f Ci /U v 2 L2 .X /. On
the other hand, since the operator U between the spaces H and L2 .X / is unitary,
there exists a unique vector u 2 H such that .f C i /U v D U u. Multiplying this
equality by F we obtain U v D F U u, whence
v D U 1 F U u D Ri u;
which in particular shows that v 2 D.A/.
Now we must verify that UAU 1 D Mf on U.D.A//. Note that for all u 2 H
we have F U u D URi u, therefore
1
Uu D URi u:
F
38 Chapter 2. Detailed introduction to abstract spectral theory

Let v 2 D.A/. Then there exists a unique u 2 H such that v D Ri u. By the

resolvent equation, Av D u  iv, hence we deduce that UAv D . F1  i /U v. In
other words, UAv D f U v and this establishes the needed equality.
To finish we must verify that f is a real-valued function. Suppose that the
imaginary part of f is strictly non-negative on a subset  of X such that ./ >
0. Since the measure space .X; / is finite,  2 D.Mf /. By Lemma 2.6.1,
the operator Mf is self-adjoint, so h  ; f  iL2 2 R. This is absurd by the
definition of the set . Hence, f is real-valued almost everywhere. 

2.6.2 Functional calculus

Using Corollary 2.6.5 we construct the standard functional bounded calculus

Definition 2.6.6. With the previous notations, for h 2 L1 .R/ we define the
bounded operator h.A/ on H by

h.A/ WD U 1 Mhıf U:

An obvious consequence of this definition is the following result.

Theorem 2.6.7 (Bounded functional calculus). Let .A; D.A// be a bounded self-
adjoint operator on H . The map  defined by:
8
< L1 .R/ ! B.H /
W
: h 7! h.A/

has the following properties.

(i)  is a bounded ?-normed algebra homomorphism, and for any function
h 2 L1 .R/
jjjh.A/jjj  khk1 :

(ii) If .hn /n 2 .L1 .R//N is such that .x 7! hn .x//n converges to x 7! x

and such that for any .x; n/ 2 R N we have jhn .x/j  jxj, then for any
u 2 D.A/
hn .A/u ! Au i n H:

(iii) If .; u/ 2 Spec.A/ .H  f0g/ is such that Au D u and if h 2 L1 .R/,

then h./ 2 Spec.h.A// and

h.A/u D h./u:
 2.6 The spectral theorem multiplication operator form 39

2.6.3 Functional calculus in Quantum Mechanics

The functional calculus of operators has numerous applications in the analysis of
PDE and physics. One of them concerns the mathematical formalism of Quantum
Mechanics (see for example [Car], [Gu-Si]). Quantum mechanics is ubiquitous,
in the study of elementary particles, semi-conductors, optical physics, etc. Quan-
tum mechanics is one of the greatest intellectual revolution and one of the most
fantastic theory in history. The mathematical formalism of Hamiltonian classical
mechanics is grounded in symplectic geometry. In Hamilton’s formalism physical
particles are describes by positions and velocities: for example, in the Euclidean
space R3 every point is characterized by a vector

.x1 ; x2 ; x3 ; 1 ; 2 ; 3 / 2 R6 :

For a Hamiltonian f 2 C 1 .R6 ; R/, the classical dynamics is governed by Hamil-

ton’s equations 8
ˆ P @f
< j D  @x j

:̂ xP D @f
:
j @j

In quantum physics the description of a particle M 2 Rn at time t  0 is given

by a wave function i.e., a vector:

.x; t/ 2 L2 .Rn /

with the following interpretation: for any subset  of Rn the real number
Z
j .x; t/j2 dx1 : : : dxn


is the probability of finding the particle M in  at time t. In quantum mechanics

the space of all possible states of the particle at a given time is called the state
space. This set is the Hilbert space L2 .Rn /:
Another principle of quantum mechanics governs the dynamics (the analogue
of Hamilton’s equations): if the particle is placed in a force field that derives from
a potential V , the time evolution of the wave function .x; t/ is governed by the
famous Schrödinger equation

@ „2
i„ .x; t/ D  g .x; t/ C V .x/ .x; t/:
@t 2
More generally, for a self-adjoint operator .A; D.A// on a Hilbert space H , using
Theorem 2.6.7 we can define the one-parameter family
˚ 
U.t/ D ei tA t 2R :
40 Chapter 2. Detailed introduction to abstract spectral theory

This one-parameter family is a continuous unitary group of bounded operators

with generator .iA; D.A//. In quantum mechanics the behavior of a physical
system S is described by a unitary group with a generator .iA; D.A//. More
precisely, if at time t D 0 the system is represented by the initial state 0 2 D.A/;
then at time t  0 it is represented by the state .t/ D U.t/ 0 . In other words,
the quantum dynamics is given by the formula

.t/ D U.t/ 0 2 B.H /:

The operator P„ WD „A is called the quantum Hamiltonian of the system S and
it represents the total energy of S . Thus for all 0 2 D.A/ we have
t
.t/ D ei h P„ 0:

Taking the derivative of this equation we obtain

@ .t/ i
D  P„ .t/;
@t „
2
so if P„ D  „2  C V we get

@
i„ .x; t/ D g .x; t/ C V .x/ .x; t/:
@t

2.7 Some complements on operators theory

2.7.1 Closable operators
Definition 2.7.1. An operator .A; D.A// is said to be closable if .A; D.A// has
a closed extension. In this case we denote by A the minimal (in the sense of
inclusion of domains) extension. This operator is called the closure of .A; D.A//.

We have:

Proposition 2.7.2. The unbounded linear operator .A; D.A// is closable if and
only if for any sequence .xn /n 2 D.A/N the conditions
8
ˆ
< n!1lim xn D 0 in the norm kkE
and
:̂ lim Axn D y in the norm kkF
n!1

implies that y D 0.
 2.7 Some complements on operators theory 41

Proof. Suppose .A; D.A// is closable, i.e., there exists a closed operator
.B; D.B// such that A  B. Let .xn /n 2 D.A/N satisfy the two conditions
in the proposition. Since the sequence .xn /n 2 D.A/N we have
8
ˆ
< n!1lim xn D 0 in the norm kkE
and
:̂ lim Bxn D y in the norm kkF ;
n!1

thus y D B0 D 0 (because B is closed).

Conversely: suppose there exists a sequence .xn /n 2 D.A/N satisfying the
hypothesis. Consider the operator e
A with domain

D.e
A/ D x 2 H I 9 ..xn /n ; y/ 2 D.A/N H I lim xn D x and lim Axn D y
n!1 n!1

and for all x 2 D.e A/, we define the operator e A by e Ax WD limn!1 Axn .
Now if we have two sequences .xn /n ; .xn /n 2 D.A/N such that limn!1 xn D x,
0

limn!1 xn0 D x and limn!1 Axn D y, limn!1 Axn0 D y, then

limn!1 .xn  xn0 / D 0 and limn!1 .Axn  Axn0 / D y  y 0 ; so using the hy-
pothesis we get y  y 0 D 0. Thus e Ax D limn!1 Axn does not depend on the
sequence .xn /n , therefore the operator e
A is well defined and linear. Moreover we
have A  eA and e A is closed. 

Proposition 2.7.3. Every symmetric operator is closable.

Definition 2.7.4. An operator A is said to be essentially self-adjoint if its closure

A is self-adjoint.

2.7.2 Unbounded operators with compact resolvents

Definition 2.7.5. Let A; D.A/ be a closed unbounded linear operator of H such
that .A/ ¤ ;. We said that A has a compact resolvent if for all  2 .A/ the
bounded operator .I  A/1 is compact.

In fact, we have

Proposition 2.7.6. If there exists  2 .A/ such that .I  A/1 is compact,
then A has a compact resolvent.

Proof. Using the resolvent equation (2.4) we have

RA ./ D .  /RA ./RA ./ C RA ./
for all  2 .A/, and since the set of compact operators is a vector space and an
ideal in the algebra bounded operator, it is clear that RA ./ D .I  A/1 is also
compact. 
42 Chapter 2. Detailed introduction to abstract spectral theory

Recall that the following result

Proposition 2.7.7. If dim.H / D C1 and if the operator A has compact resol-

vent, then A is not bounded.

Proof. We have for all  2 .A/ the equality .I  A/ .I  A/1 D I thus if
the operator A is bounded, then I D .I  A/ .I  A/1 is compact, which is
absurd because dim.H / D C1. 

A main result concerning operators with compact resolvents is

Theorem 2.7.8. Let .A; D.A// be a closed unbounded linear operator with com-
pact resolvent in a Banach space E. Then the spectrum of A is a discrete set and
coincides with the point spectrum of A, i.e.,

.A/ D Spec.A/:

Moreover, every eigenvalue has a finite multiplicity.

Proof. For a fixed z 2 .A/, the operator B WD .zI  A/1 is compact, and for
all  2 C we have seen in Theorem 2.4.5 that
1
.B/  f0g D ;  2 .A/
z 

and
1
Spec.B/  f0g D ;  2 Spec.A/ :
z
So we have
1
.A/ D z  ;  2 .B/  f0g

and
1
Spec.A/ D z  ;  2 Spec.B/  f0g :

Since the operator B is compact, Corollary 2.5.8 shows that .B/  f0g D
Spec.B/  f0g: Consequently,

1
.A/ D z  ;  2 Spec.B/  f0g ;


thus .A/ D Spec.A/ and this proves the theorem because every eigenvalue in
Spec.B/  f0g is isolated and has a finite multiplicity (Corollary 2.5.8 and Lemma
2.5.10). 
 2.7 Some complements on operators theory 43

Finally, the analogue of Theorem 2.5.6 for unbounded self-adjoint operators

with compact resolvent reads:

Theorem 2.7.9. Let H be a separable Hilbert space and .A; D.A// be a self-
adjoint operator with compact resolvent. Then the spectrum and the point spec-
trum of A coincide. This spectrum is a real discrete set, namely it consists of
a sequence .k /k of infinite real eigenvalues with finite multiplicity such that
k ! C1 as k ! C1. Moreover, the associated eigenfunctions .ek /k0 form
a Hilbert basis of the space H .

Proof. Since A is self-adjoint, Spec.A/  .A/  R. Let  2 .A/. Then the

bounded operator R WD .I  A/1 is compact, thus .R/  f0g D Spec.R/  f0g
and this is a discrete subset of R. Consequently, the sets Spec.A/ and .A/ are
equal and discrete (see Lemma 2.5.10). Hence there exists  2 R such that  2
.A/, and so we can assume now that  2 R. It follows that the operator R is
compact, self-adjoint and 0 … Spec.R/ (else there exists x 2 H with x ¤ 0 such
that Rx D 0, whence, .I  A/1 x D 0, i.e., x D 0 which is absurd). By the
spectral theory of compact self-adjoint operators (Theorem 2.5.6), the spectrum
of R is a real discrete set, more precisely, the spectrum is an infinite sequence
.k /k of real non-null eigenvalues with finite multiplicity such that k ! 0 as
k ! C1. Moreover, the associated eigenfunctions .ek /k0 form a Hilbert basis
of the space H . In other words, for any integer n

Ren D n en ;

i.e.,
.I  A/1 en D n en ;
whence
en D n en  n Aen :
Therefore,  
1
Aen D   en
n
so, with n WD   1n we have n ! ˙1 as n ! C1, .A/ D Spec.A/ D
fn ; n  0g, and for any integer n

Aen D n en : 

In particular, if the operator A is positive the eigenvalues .n /n are positive.

In this case we can arrange the eigenvalues so that

0  1      k ! C1:
44 Chapter 2. Detailed introduction to abstract spectral theory

2.7.3 Numerical range and applications

Definition 2.7.10. Let .A; D.A// be an unbounded linear operator on H . The
numerical range ‚.A/ of A is the subset of C defined by

‚.A/ WD fhAu; uiH I u 2 D.A/ with kukH D 1g :

A theorem from Hausdorff asserts that this set is convex. If the operator
.A; D.A// is closed,  WD ‚.A/ is a closed convex subset of C. It is easy to
see that the set  WD C   is
 connected (here  is compact), or
 non-connected if  is a band bounded by two parallel lines (here  is not
compact).

Lemma 2.7.11. For any  2 :

(i) The inequality k.I  A/ukH  d.; / kukH holds for any u 2 D.A/
.here d.; / denotes the distance between  and /.
(ii) ker.I  A/ D f0g and im.I  A/ is closed.
(iii) For any  2 .A/,
ˇˇˇ ˇˇˇ 1
ˇˇˇ.I  A/1 ˇˇˇ  :
d.; /

Proof. Here d.; / > 0 because  2 .

 Item (i): for all u 2 D.A/ with kukH D 1, using the Cauchy–Schwarz
inequality we get

d.; /  j  hAu; uiH j D jh.I  A/ u; uiH j

 k.I  A/ukH :

Therefore we obtain (i).

 Item (ii): using (i) it is clear that ker .I  A/ D f0g. Next consider
a Cauchy sequence yn D .I  A/ xn in im .I  A/ (obviously the se-
quence .xn /n belongs to D.A/). Using (i) we see that .xn /n is a Cauchy se-
quence in D.A/, and since H is complete there exists a couple .x; y/ 2 H 2
such that limn!1 xn D x and limn!1 yn D y in the norm kkH . There-
fore,
lim Axn D lim xn  yn D x  y:
n!1 n!1

So, since A is closed, x 2 D.A/ and Ax D x  y, whence y D

.I A/ x 2 im .I A/. This means that the sequence yn D .I A/ xn
converges in im .I  A/, thus im .I  A/ is a complete subspace of H .
It follows that im .I  A/ is closed.
 2.7 Some complements on operators theory 45

 Finally, let us prove (iii). If  2 .A/, then im .I  A/ D H , so using (i)

we have for all u 2 D.A/ with u ¤ 0

k.I  A/ ukH
d.; /  :
kukH

For all v 2 H with v ¤ 0,

kvkH
d.; /    ;
 
.I  A/1 v 
H

so for all v 2 H with v ¤ 0,

 
 
.I  A/1 v  1
 : 
kvkH d.; /

Now consider the map

8
ˆ
< C ! N [ fC1g
dW
:̂  7! dim ker I  A? :

Using (2.2), we have also

d./ D dim .im .I  A//? :

If  2 .A/, then we immediately get im .I  A/ D H , hence d./ D 0. In

fact we can be more precise

Proposition 2.7.12. For any  2  D C  ‚.A/,

d./ D 0 ”  2 .A/:

Proof. Let  2  D C  ‚.A/. If d./ D 0, then ker I  A? D f0g, and

using (2.2) we have
im .I  A/? D f0g:
Applying here ?, we get
im .I  A/ D H
and since ker .I  A/ D f0g and im .I  A/ is closed (Lemma 2.7.11), we
obtain im .I  A/ D H , hence  2 .A/. 
46 Chapter 2. Detailed introduction to abstract spectral theory

Lemma 2.7.13 (Perturbation lemma). Let .A; D.A// be a closed unbounded lin-
ear operator of H with the property that there exists > 0 such that kAxkH 
kxkH for all x 2 D.A/. Then for every complex number  such that jj < 2
we have:   
dim ker A?  I D dim ker A? :

Proof. Let  be a complex such that jj < and consider the sets M WD ker .A? /,
N WD ker A?  I . Our proof is carried out in three steps:
 Step 1. We claim that M ? \ N D f0g. Indeed, if u 2 M ? \ N , since
M ? D im.A/ D im.A/ (the subspace im.A/ is closed because for all x 2
D.A/ we have kAxkH  kxkH , see the proof of Lemma 2.3.8), there
exists v 2 D.A/ such that u D Av, so by hypothesis, we have kukH 
kvkH . Now, since u belongs to N ,
D E
0 D A?  I u; v D hu; .A  I / viH
H
D hu; u  viH D kuk2H   hu; viH
 
2 2 jj
 kukH  jj kukH kvkH  kukH 1 

and since we have supposed that jj < , we have 1  jj 

> 0. It follows
that u D 0.
 Step 2. We claim that dim.N /  dim.M /. Indeed, if dim.M / D C1, this
is clear. If now dim.M / D m < C1, let .e1 ; : : : ; em ; emC1 / be a family of
mC1 vectors in N and denote by P the orthogonal projector onto M . Then
.P e1 ; : : : ; P em ; P emC1 / is a family of mC1 vectors in M , so these vectors
are not linearly independent: there exists .1 ; : : : ; m ; mC1 / 2 CmC1 with
.1 ; : : : ; m ; mC1 / ¤ .0; 0; : : : ; 0/ such that
!
X
mC1 X
mC1
0D i P ei D P i ei :
i D1 i D1
PmC1
Thus, i D1 i ei 2 N \ M ? (because ker.P / D M ? ), consequently us-
ing Step 1, we get
X
mC1
i ei D 0
i D1
with .1 ; : : : ; m ; mC1 / ¤ .0; 0; : : : ; 0/, i.e., that the vectors .e1 ; : : : ; em ;
emC1 / are not linearly independent, and since the family was arbitrary, we
conclude that dim.N /  m D dim.M /. In other words,
 
dim ker A?  I  dim ker.A? / :
 2.7 Some complements on operators theory 47

In the same way, if we replace  by , we have also

 
dim ker A? C I  dim.ker A? /:

 Step 3. We claim that for all  such that jj < 2 and for all x 2 D.A/ we
have k.A  I / xkH  2 kxkH . Indeed, for all u 2 D.A/
k.A  I / xkH  jkAxkH  jj kxkH j
 j kxkH  jj kxkH j  kxkH ;
2
so if we consider the operator B WD A  I we have for all x 2 D.A/,
kBxkH  2 kxkH . By step 2, applied to the operator B, for any complex
number  such that jj < 2 it holds that
  
dim ker B ? C I  dim ker B ? ;

i.e.,
  
dim ker A?  dim ker A?  I :

Finally, we get dim ker A?  I D dim .ker .A? //. 

Corollary 2.7.14. The function  7! d./ is constant on every connected com-

ponents of  D C  ‚.A/.

Proof. We have seen (Lemma 2.7.11) that for all  2  and for all u 2 D.A/,
k.I  A/ ukH  kukH ;
where WD d.; / > 0. Here for any complex number  we have
     
d./ D dim ker A?  I D dim ker I  A?
D dim ker I  A? C .  /I :

Consider the operator B WD I A. Then B ? D I A? , and we have kBukH 

kukH . Therefore, by Lemma 2.7.13: if j  j < 2 ,
  
dim ker B ?  .  /I D dim ker B ? ;

i.e., for any complex number  such that j  j < 2
we have
d./ D d./:
Hence, the dimension of I  A? is locally constant, so the dimension of I  A?
is constant on every connected components of  D C  ‚.A/. 
48 Chapter 2. Detailed introduction to abstract spectral theory

Proposition 2.7.15. Let A be a bounded operator of H . We have .A/  ‚.A/.

Proof. Here

‚.A/ D fhAu; uiH I u 2 H with kukH D 1g

thus for all u 2 H with kukH  1 we have jhAu; uiH j  jjjAjjj. Therefore
‚.A/  B, where B is the closed disc of C with center 0 and radius jjjAjjj, so

‚.A/  B:

Thus  WD C  ‚.A/ is an open and connected subset of C (because ‚.A/ is

compact) and f 2 CI jj > jjjAjjjg   (because ‚.A/  B). By the classical
theory of bounded operators in Hilbert spaces, we have also .A/  B, hence if
jj > jjjAjjj then  2 .A/. It follows that d./ D 0 (because im .I  A/ D H ,
i.e., ker A?  I D f0g). Now, since the map  7! d./ is constant on every
connected component of , we get for all  2  that d./ D 0. Finally, using
Proposition 2.7.12 we have   .A/, i.e., .A/  ‚.A/. 

Consider now a symmetric unbounded operator .A; D.A//. First, it is clear

that ‚.A/  R, Hence for a symmetric operator we always have ˙i 2 . Denote

mC WD dim ker I  A? for im./ > 0:

In fact the number mC does not depend on the choice of  2 C such that im./ >
0. Indeed, since ‚.A/  R, the set PC WD f 2 CI im./ > 0g is included in one
of the connected components of , and we conclude that the map  7! d./ is
constant on PC . Similarly, define

m WD dim ker I  A? for im./ < 0

m does not depend on the choice of  2 C such that im./ < 0. Note that if the
set  D C  ‚.A/ is connected then mC D m . Hence for a symmetric operator
˙i 2  and mC D d.i /; m D d.i /.

Theorem 2.7.16. Let .A; D.A// be a closed symmetric operator of H . A is self-

adjoint if and only if mC D m D 0.

Proof. Since A is closed and symmetric, using Theorem 2.3.9 we get

 
mC D m D 0 ” d.˙i / D 0 ” ker A? ˙ iI D f0g ” A D A? : 
 2.8 Exercises 49

2.8 Exercises
Exercise 2.8.1. Prove that the set B.H / of bounded operators on a Hilbert space
H is a Banach space for the norm jjjjjj. Show that for all T; S 2 B.H / we have

jjjT S jjj  jjjT jjj jjjS jjj :

Exercise 2.8.2. Prove that for a bounded operator the spectrum and point spec-
trum coincide.

Exercise 2.8.3. Prove Theorem 2.5.5.

Exercise 2.8.4. Show that the operator

8
< `2 ! `2
TW
: u 7!  u
n n n

is bounded if and only if the sequence .n /n is bounded. In this case show that T
is invertible with a not-bounded inverse.

Exercise 2.8.5. Show that the operator

8
< `2 ! `2
TW
:
un 7! n1 un

is invertible with a not-bounded inverse.

Exercise 2.8.6. Show that the operator

8
< `2 ! `2
SW
: u 7! u
n nC1

is a bounded.

Exercise 2.8.7. Let us denote by Binv .H / the set of bounded invertible operators
with bounded inverse. Show that the mapping
8
< Binv .H / ! Binv .H /
: T 7! T 1

is continuous .for the norm jjjjjj/.

50 Chapter 2. Detailed introduction to abstract spectral theory

Exercise 2.8.8. Show that if an operator is symmetric then its eigenvalues are
real.

Exercise 2.8.9. Prove that the point spectrum of the operator T from the Exercise
2.8.4 is Spec.T / D fn ; n  1g and that the spectrum of T is .T / D Spec.T /.

Exercise 2.8.10. Consider the operator

8
< L2 Œ0; 1 ! L2 Œ0; 1
Mx W
: f .x/ 7! xf .x/:

Show that Mx is bounded and jjjMx jjj  1. Next prove that Spec.Mx / D ; and
.T / D Œ0; 1.

Exercise 2.8.11. Show that the operator of Exercise 2.8.10 is self-adjoint.

Exercise 2.8.12. Show that the operator of Exercise 2.8.4 is compact if and only
if n ! 0.

Exercise 2.8.13. Show that operator of Exercise 2.8.10 is not compact.

Exercise 2.8.14 (Cayley transform). Let .T; D.T // be a closed symmetric opera-
tor on a Hilbert space H .

1. Show that T C iI is injective with a closed image.

2. Show that the operator UT WD .T  iI / .T C iI /1 with domain D.UT / D
Im .T C iI / is closed with a closed domain.
3. Prove that UT is isometric with a closed image.
4. Prove that ker .I  UT / D f0g.
5. Show that T D i .I C UT / .I  UT /1 .
Chapter 3
The Laplacian on a compact Riemannian
manifold

The goal of this chapter is to introduce some fundamental notions of Riemannian

Geometry and analysis on manifold. At the end of the chapter we define the main
object of this book: the Laplace–Beltrami operator on a compact Riemannian
manifold.

Assumption. All manifolds we work with are supposed to be connected, smooth,

and countable at infinity.

3.1 Basic Riemannian Geometry

One of the principal characteristics of modern physics is the geometry: general
relativity, string theory, quantum field theory . . . Riemannian geometry is a gener-
alisation of Euclidean geometry, Riemannian geometry is in particular associated
to general relativity and gauge theory.
For an introduction on differentiable manifolds see for example [War] or
[DRh] and for a complete course on Riemannian Geometry see for example the
books of I. Chavel [Cha2], Gallot–Hullin–Lafontaine [GHL], J. Jost [Jos], M. Do
Carmo [DoC], T. Sakai [Sak]. See also the bible of M. Berger [Ber].

3.1.1 Differential Geometry: conventions and notations

First we need to introduce some notations: start with some very basics facts (with-
out proofs) on differential geometry.
From now on, all manifolds are supposed to be connected.

Definition 3.1.1. A Hausdorff (i.e., possessing a countable basis of topology)

separable topological space M is a n-dimensional topological manifold if for each
point x 2 M there exists an open neighbourhood U of x and a function 'W U !
Rn such that ' is an homeomorphism from U onto on open set of Rn .

The pair .U; '/ is called a chart of M and for y 2 U the components of the
vector
'.y/ D .y1 ; y2 ; : : : ; yn / 2 Rn
52 Chapter 3. The Laplacian on a compact Riemannian manifold

are called the coordinates of the point y in theSchart .U; '/. An atlas of M is
a collection of charts .Ui ; 'i / such that M D i Ui : We say that a topological
manifold M is a smooth manifold if for every pair of charts .U; '/ and .V; /
such that U \ V ¤ ;, the map

ı ' 1 W '.U \ V /  Rn ! .U \ V /  Rn

is smooth in the classical sense of differential calculus in Rn . This type of maps

ı ' 1 are called transition functions.

Example 3.1.2.
 The first simple example is the Euclidean space Rn : here the atlas is re-
duced to the open set U D Rn and the unique chart is the identity map.
˚ 
 The second usual example is the n-sphere Sn WD x 2 RnC1 I kxk D 1 :
here one can consider an atlas with two charts: U D Sn  fN g and V D
Sn  fS g, where N; S are the north and the south poles of the sphere. The
homeomorphisms are the stereographic projections
8
ˆ
< U  RnC1 ! Rn
'N W
:̂ .x1 ; x2 ; : : : ; xnC1 / 7! x1
; x2 ; : : : ; 1xxnnC1 ;
1xnC1 1xnC1

and
8
ˆ
< V  RnC1 ! Rn
'S W
:̂ .x1 ; x2 ; : : : ; xnC1 / 7! x1
; x2 ; : : : ; 1CxxnnC1 :
1CxnC1 1CxnC1

Definition 3.1.3. A map f W M ! N between two smooth manifolds is said to be

smooth (or C 1 ) if for any two charts .U; '/ of M and .V; / of N the map
 
ı f ı ' 1 W ' U \ f 1 .V /  Rn ! .f .U / \ V /  Rn

is smooth in the classical sense.

In the case N D R we denote by C 1 .M / the set of smooth functions between

M and R.

Definition 3.1.4. A tangent vector to a smooth manifold M at a point x 2 M

is a derivation on C 1 .M /, that is, a linear form X W C 1 .M / ! R such that the
Leibniz rule
X.fg/ D X.f /g.x/ C f .x/X.g/
holds for all .f; g/ 2 C 1 .M /2 .
 3.1 Basic Riemannian Geometry 53

For a fixed point x on M the set of derivations can be equipped with the
following operations: if X; Y are two derivations and  a scalar, for all f 2
C 1 .M / we put
.X C Y /.f / D X.f / C Y .f /:
Thus the set of derivations is a n-dimensional vector space, called the tangent
space of M at the point x. This vector space is denoted by Tx M . The rela-
tion between this definition and the geometric intuitive notion of tangent vector is
the following: for a curve W r; rŒ! M such that .0/ D x; the corresponding
derivation acts on f 2 C 1 .M / as
d .f ı /
X.f / D .0/:
dt
@
Given a chart .U; '/ containing x, we define the tangent vector @xi
by
x
   
@ @ f ı ' 1
.f / D .'.x//
@xi x @xi
n o
for all f 2 C 1 .M /. In fact, the n-tuple @x@ is a basis of Tx M , so for
i x 1i n
every vector X 2 Tx M there exists n-scalars X1 ; X2 ; : : : ; Xn such that
Xn  
@
XD Xi ;
@xi x
i D1
 
and for a local chart U; ' D .' 1 ; ' 2 ; : : : ; ' n / we have

Xi D X.' i /:

Definition 3.1.5. Let f W M ! R be a smooth function on the manifold M . For

any point x 2 M we define the differential df .x/ by
8
< Tx M ! R
df .x/W
: X 7! .df .x// X WD X.f /:

Remark 3.1.6. There exists another equivalent ways to define the tangent space
(see [GHL]).
The (disjoint) union of the tangent spaces
a
TM WD Tx M
x2M

admit a structure of smooth 2n-dimensional manifold, called the tangent bundle

of M .
54 Chapter 3. The Laplacian on a compact Riemannian manifold

Definition 3.1.7. A vector field of a manifold M is a section of the tangent bundle

TM , i.e., a smooth map
8
< M ! TM
XW
: x 7! X 2 T M:
x x

We denote by .M / the set of vector fields on M . This set can be equipped
with the Lie bracket: for .X; Y / 2 .M /2 , for all f 2 C 1 .M / and for all x 2 M ,

ŒX; Y x .f / WD Xx .Y .f //  Yx .X.f //

where Y .f /W y 2 M 7! Yy .f / 2 R and X.f /W y 2 M 7! Xy .f / 2 R are

smooth functions.
Let X be a smooth vector field of M and x a fixed point on M . By classical
results on dynamical systems and ODE theory there exists an unique smooth curve
W r; rŒ! M such that: 8
< 0 .t/ D X. .t//
: .0/ D x:
Hence the local flow of the vector field X with origin x at time t is defined by

't .x/ D .t/:

For a fixed point x 2 M , the cotangent space of M at x is as defined

n as theo (alge-
braic) dual space Tx M of Tx M , and the associated dual basis of @x@
?
i x 1i n
is denoted f.dxi /x g1i n : namely,
  
@
.dxi /x D ıij ;
@xi x

where 8
< 1; if i D j
ıij D
: 0; otherwise:

The (disjoint) union of the cotangent spaces

a
T ? M WD Tx? M
x2M

also admits a structure of smooth 2n-dimensional manifold, called the cotangent

bundle of M .
 3.1 Basic Riemannian Geometry 55

Let f W M ! N be a smooth function between two smooth manifolds. For

a fixed point x 2 M we define the map Tx f W Tx M ! Tf .x/ N by the rule: for all
u 2 Tx M and g 2 C 1 .N /,

.Tx f .u// .g/ WD u.g ı f /:

It is easy to see that this map in linear between the spaces Tx M and Tf .x/ N .
Consequently we can also define the (linear) tangent map of f by
8
< TM ! T N
Tf W
: .x; u/ 7! .f .x/; .T f .u/// :
x

Definition 3.1.8. A connection on M is a map

DW TM .M / ! TM

such that:
(i) for all x 2 M , if X 2 Tx M and Y 2 .M / then D.X; Y / 2 Tx M ,
(ii) for all x 2 M , the map DW Tx M .M / ! Tx M is bilinear,
(iii) for all x 2 M , all X 2 Tx M , and all Y 2 .M /, if f W M ! R is
differentiable, then

D.X; f Y / D X.f /Yx C f .x/D.X; Y /:

If D is a connection on M , we denote

DX Y WD D.X; Y /:

The operator D is a local operator in the sense that for any point x 2 M and for
any X; Y; Z 2 .M /, if Y D Z in a neighbourhood of x, then  DXY D DX Z.
In this book, for x 2 M and k 2 N we denote by ƒk Tx? M the set of k-
alternating forms on the vector space Tx? M . The (disjoint) union of these spaces
is denoted by a  
ƒk .M / WD ƒk T ? M
x2M

and admits a structure of fiber bundle. If k D 1, ƒ1 .M / D T ? M , and by

convention for k D 0 we put ƒ0 .M / D R.

Definition 3.1.9. A differential form of degree k (or a k-form) on M is a smooth

section of the bundle ƒk .M /. We denote by k .M / the set of smooth sections
of ƒk .M /, with the convention 0 .M / D C 1 .M /.
56 Chapter 3. The Laplacian on a compact Riemannian manifold

The set k .M / can be equipped with the exterior derivative operator d ,

8
< k .M / ! kC1 .M /
dW
:
˛ 7! d˛:

Namely, if X
˛D ˛i dxi ;
i

then we put
X  @˛i 
d˛ D dxj ^ dxi ;
@xj
i;j

where ^ is the wedge (exterior) product on alternating forms: for an n1 -form !1

and for an n2 -form!2 , the element !1 ^  !2 is the .n1 C n2 /-form defined in local
coordinates for all V1 ; V2 ; : : : ; Vn1 Cn2 2 Tx? M by the formula
  X .1/  
.!1 ^ !2 / V1 ; V2 ; : : : ; Vn1 Cn2 D !1 V .1/ ; V .2/ ; : : : ; V .n1 /
n1 Šn2 Š
2Sn1 Cn2
 
 !2 V .n1 C1/ ; V .n1 C2/ ; : : : ; V .n1 Cn2 /

where Sn1 Cn2 denotes the symmetric group with n1 C n2 elements

The operator d has the fundamental property

d ı d D 0:

If k D 0, the operator d coincides with the usual differential operator on functions.

For ˛ 2 k .M /, ˇ 2 ` .M / we have the relation:

d .˛ ^ ˇ/ D d˛ ^ ˇ C .1/k ˛ ^ dˇ:

A differential form ˛ 2 k .M / is said to be closed if d˛ D 0, and ˛ is said to be

exact if there exists a differential form ˇ 2 k1 .M / such that ˛ D dˇ. Since
d 2 D 0, every exact form is closed. By the Poincaré lemma, on a contractible1
domain every closed form is exact. The codifferential operator ı is the formal
L2 -adjoint of d on k .M /.
On the set of differential forms we have for any vector field X the interior
product operator iX , defined by
8
< kC1 .M / ! k .M /
iX W
:
˛ 7! iX .˛/;
1 The fundamental group is trivial.
 3.1 Basic Riemannian Geometry 57

where for all x 2 M and for all .V1 ; V2 ; : : : ; Vk / 2 Tx? M ,

iX .˛/.x/ .V1 ; V2 ; : : : ; Vk / D ˛.x/.X.x/; V1; V2 ; : : : ; Vk /:

For any smooth map 'W M ! N between two manifolds M; N , the pull-back of
' is defined by: for any x 2 M , ˇ 2 k .M /, and .V1 ; V2 ; : : : ; Vk / 2 Tx? M ,
 ? 
' ˇ .x/.V1 ; V2 ; : : : ; Vk / D ˇ.'.x// .Tx '.V1 /; Tx '.V2 /; : : : ; Tx '.Vk // :

Finally, the Lie derivative associated to a smooth vector field X is defined by the
formula: for all ˇ 2 k .M /,
't? ˇ  ˇ
LX .ˇ/ WD lim ;
t !0 t
where 't is the local flow generated by X . We have the Cartan formula

LX .ˇ/ D iX dˇ C d .iX ˇ/ I
in particular, for a function f (i.e., f 2 0 .M /), we get

LX .f / D iX df D df .X /:

3.1.2 Riemannian manifolds and examples

Definition 3.1.10. A Riemannian metric g on a manifold M is a map g which
associates to each point x of M a scalar product g.x/.; / on the vector space
Tx M . We also require that g is smooth: if X and Y are C 1 vector fields, then
the map x 7! g./.X; Y / is smooth on M . A Riemannian manifold is a manifold
equipped with a Riemannian metric.

Given a manifold M , using a partition of unity, the existence of a Riemannian

metric on M is given by the following classical result:

Theorem 3.1.11. On every manifold M there exists at least one Riemannian met-
ric.

Example 3.1.12.
 The simplest example of Riemannian manifold is the Euclidean space Rn
with the canonical metric “can”, which is the usual scalar product on Rn :
for every x 2 Rn , and all X; Y 2 Tx Rn with X D .x1 ; x2 ; : : : ; xn / and
Y D .y1 ; y2 ; : : : ; yn /
X
n
can.x/.X; Y / WD hX; Y iRn D xi y i :
i D1
58 Chapter 3. The Laplacian on a compact Riemannian manifold

 Another example of Riemannian metric on the Euclidean space Rn is ob-

tained as follows: for a fixed positive definite n n matrix A we define the
metric gA by: for every x 2 Rn , and all X; Y 2 Tx Rn ,
gA .x/.X; Y / WD hAX; Y iRn :

Let us give an example on the sphere.

˚ 
Example 3.1.13. The Euclidean sphere: the sphere Sn WD x 2 RnC1 I kxk D 1
equipped with the metric it inherits as a subspace of RnC1 is a Riemannian mani-
fold.

We finish with a hyperbolic example.

Example 3.1.14. The hyperbolic space: let Hn WD fx 2 Rn I kxk < 1g, the open
unit ball in Rn . The hyperbolic space Hn can be equipped with the hyperbolic
metric g defined for every point x 2 Hn and for all X; Y 2 Tx Hn by
4 hX; Y iRn
g.x/.X; Y / WD 2
:
1 kxk2Rn

In the case of dimension 2, H2 is also called the Poincaré disk.

Consider a Riemannian manifold .M; g/ of dimension n. Using the metric g

we can do differential calculus on M via the Levi-Civita connection.

Definition 3.1.15. Let .M; g/ be a Riemannian manifold. The Levi-Civita con-

nection is the unique connection D on TM such that
(i) D is torsion free: for all smooth vector fields X; Y of M ,
DX Y  DX Y D ŒX; Y  :

(ii) D is compatible with the metric g: for all smooth vector fields X; Y; Z of
M,
X.g.Y; Z// D g .DX Y; Z/ C g .Y; DX Z/ :

For a Riemannian manifold .M; g/ of dimension

 n and for a local chart W U 
M ! Rn which coordinates are denoted by x 1 ; x 2 ; : : : ; x n , we denote by
@
; @ ;
@x 1 @x 2
; @x@n the associated vector fields. Then we have the following
local expression of the Levi-Civita connection:

@ X
n
@
D @ D ijk ;
@xi @xj @xk
kD1

where ijk denote the Christoffel symbols.

 3.1 Basic Riemannian Geometry 59

3.1.3 Metric structure and geodesics on a Riemannian manifold

On a Riemannian manifold we have a natural metric structure, indeed, for a curve
W Œ0; 1 ! M , the length of is defined by
Z 1p
L. / WD g. .t// . P .t/; P .t// dt;
0

where P .t/ 2 T.t / M is the velocity vector of the curve. We can now define
a distance (called the Riemannian distance) on the manifold .M; g/: for any pair
of points .x; y/ in M , the Riemannian distance between x and y is

d.x; y/ WD inf L. /;
 2C .x!y/

where the infimum is taken over the set C .x ! y/ of smooth curves on M going
from the point x to the point y. The topology associated to this metric coincides
with the natural topology of the manifold M: If the metric space .M; d / is com-
plete (i.e., every Cauchy sequence is convergent), then every pair of points .x; y/
on M can be joined by a curve and

d.x; y/ D L. /:

This result is a consequence of the Hopf–Rinow theorem (see Theorem 3.1.19).

The curve going from the point x to the point y is a geodesic. Geodesics are
curves which locally minimize length and realize the distance between two points
on the manifold, i.e., for " small enough the curve is the shortest path between
the points .t/ and .t C "/. Geodesic curves are curves which satisfy the non-
linear second-order differential equation

P / P .t/ D 0;
D.t

in local coordinates,
X
n
Rk .t/ C ijk Pi .t/ Pj .t/ D 0:
i;j

By the standard existence and uniqueness results for ordinary differential equa-
tions, for every point x 2 M and for every vector U 2 Tx M there exist > 0
and a unique geodesic U W  ; Π! M such that U .0/ D x and PU .0/ D U .
For example
 on the Euclidean plane R2 geodesics are straight lines,
 on the round sphere S2 geodesics are great circles,
 on the hyperbolic Poincaré disk geodesics are arcs of circles meeting the
unit circle orthogonally, or segments through the origin.
60 Chapter 3. The Laplacian on a compact Riemannian manifold

B
A

B
A

The Hopf–Rinow theorem states that the mapping t 7! U .t/ is well defined
for all t 2 R if and only if the metric space .M; d / is complete.

Definition 3.1.16. We say a Riemannian manifold .M; d / is complete if the metric

space .M; d / is complete.

We have also

Definition 3.1.17. Let .M; d / be a complete Riemannian manifold. For every

fixed point x 2 M we define the exponential map at the point x by
8
< Tx M ! M
expx W
: U 7! exp U WD .1/;
x U

where t 7! U .t/ is the unique geodesic such that U .0/ D x and PU .0/ D U .

Geometrically, the point expx U on M is obtained by drawing the geodesic

starting from the point x and going in the direction of the vector U . The Rie-
mannian distance between the point x and expx U is exactly equal to kU k. If we
denote by B .x; r/ the open ball of center x and radius r of M for the Riemannian
distance, and by BTx M .0x ; r/ the open ball of center x and radius r of Tx M for
the Euclidean norm, then

expx .BTx M .0x ; r// D BM .x; r/

for all r > 0. The exponential map is well defined on the whole tangent space
(since the manifold is complete), but is not a global diffeomorphism from Tx M
 3.1 Basic Riemannian Geometry 61

on M . By the local inversion theorem, there exists a radius r > 0 such that
exponential map at the point x is a local diffeomorphism from BTx M .0x ; r/ to
expx .BTx M .0x ; r// D BM .x; r/.

Definition 3.1.18. The injectivity radius of the manifold .M; g/ at the point x 2
M is the largest radius r > 0 such that the exponential map at the point x is
a local diffeomorphism from the ball of radius r and center 0x 2 Tx M . The
injectivity radius of the manifold .M; g/ is the infimum of the injectivity radius
over all points x 2 M .

In other words, the injectivity radius of M is the largest number r > 0 such
that for every pair of points .x; y/ 2 M 2 with dM .x; y/ < r there exists an
unique geodesic curve which joins x and y. If the manifold .M; g/ is compact,
the injectivity radius is strictly positive. We have (see [GHL] for a proof):

Theorem 3.1.19 (Hopf–Rinow Theorem). Let .M; g/ be a Riemannian connected

manifold. The following conditions are equivalent:
(i) The manifold .M; g/ is complete.
(ii) There exists x 2 M such that the map expx is defined on the whole tangent
space Tx M .
(iii) The map expx is defined on the whole tangent space Tx M for all x 2 M .
(iv) Compact sets of M are exactly closed and bounded sets of M .
Moreover, any of these conditions implies that every pair of points .x; y/ 2 M 2
can be joined by a geodesic curve.

From now on, all Riemannian manifolds are supposed to be complete.

3.1.4 Curvatures on a Riemannian manifold

Now, let us recall the main notions of curvature in Riemannian geometry. We start
with the tensorial point of view. Let D be the Levi-Civita connection associated
to the metric g. Let X and Y be smooth vector fields on M . The curvature tensor
is defined by
R.X; Y / WD ŒDX ; DY   DŒX;Y :
So, for two smooth vector fields X and Y; R.X; Y / is a map from the set of smooth
vector fields on M into itself; and for any smooth vector field U on M we have

R.X; Y /U D DX .DY U /  DY .DX U /  DŒX;Y U:

In fact, we can use R to build a tensor of type .0; 4/. Namely, for four smooth
vector fields X; Y; Z; T , we define the tensor R.X; Y; Z; T / by

R.X; Y; Z; T / WD g .R .X; Y / T; Z/ :
62 Chapter 3. The Laplacian on a compact Riemannian manifold

For a fixed point x on the manifold M , the sectional curvature of a 2-plane P 

Tx M spanned by a basis X1 ; X2 is the number
g .R .X1 ; X2 / X2 ; X1 / R.X1 ; X2 ; X1 ; X2 /
Kx .P / WD D ;
g .X1 ^ X2 ; X1 ^ X2 / kX1 ^ X2 k2
where ^ is the wedge product.
The Ricci curvature tensor of .M; g/ is defined as follows: for any x 2 M and
for any vector v 2 Tx M ,

X
n
Ricx .v; v/ WD R .v; ei ; v; ei /
i D1

where .ei /1i n is an orthonormal basis of the vector space Tx M .

Next, the scalar curvature of .M; g/ is defined by: for all x 2 M

X
n
Scal.x/ WD Ricx .ei ; ei / 2 R;
i D1

where .ei /1i n is an orthonormal basis of Tx M . In Riemannian geometry, the

scalar curvature at the point x is also denoted by R.x/.
With the tensorial point of view the above notions of curvatures are not really
intuitive. So let us give here their geometrical interpretation. Start by the simplest,
the sectional curvature. Let x be a fixed point on the manifold M , and P  Tx M
a 2-plane in the vector space Tx M . For r > 0, denote by C .0; r/ the circle in P
of center 0 and radius r. Consider the image of C .0; r/ under the exponential map
expx at the point x. This image is a closed curve in the manifold M whose length
has the expansion (see for example [BGM]):
 
r2
L .expx .C .0; r/// D 2 r 1  Kx .P / C O.r / 3
6
as the radius r goes to zero. In particular,

L .expx .C .0; r///  2 r:

r!0C

Therefore, the sectional curvature Kx .P / can be expressed as an infinitesimal

deviation of circles tangent to P from having Euclidean length.
Remark 3.1.20. In the case of surfaces embedded in R3 , the notions of sectional
curvature and the Gaussian curvature coincides. In fact, the sectional curvature is
the generalization of the Gaussian curvature.
 3.1 Basic Riemannian Geometry 63

Example 3.1.21.
 In the case M D Rn , Tx M ' Rn for all x 2 M . For any 2-plane P 
Tx M and any radius r > 0 we have

L .expx .C .0; r/// D 2 r;

hence Kx .P / D 0. Therefore, all sectional curvatures on Rn are 0.

 The second usual example is the 2-sphere M D S2 : in this case for all
x 2 M , Tx M ' R2 , so for any radius r > 0 we have

L .expx .C .0; r/// D 2 sin.r/:

3
Since sin.r/ D r  r6 C o.r 4 /, we have Kx .P / D 1. More generally, all
sectional curvatures on Sn are C1.
 We finish with the hyperbolic space (and Poincaré disk), for which Kx .P / D
1 (see [GHL] for details); hence, all sectional curvatures on Hn are 1.

By the definition of Ricci curvature, for x in M and for v a vector of Tx M

X
n1
 
Ricx .v; v/ D Kx Pej ;v ;
j D1
 
where ej 1j n1 is an orthonormal basis of the .n  1/-dimensional vector v ? ,
and Pej ;v denotes the 2-plane spanned by ej and v.

Example 3.1.22. In the case of spaces of constant curvature, the Ricci curvature
is just a multiple of the metric:
 on Rn , Ric D 0;
 on Sn , Ric D .n  1/g;
 on Hn , Ric D .n  1/g.

For finish observe that the scalar curvature is just the trace of the Ricci curva-
ture on Tx M
X
n X
n X
 
Scal.x/ D Ricx .ei ; ei / D Kx Pej ;ei ;
i D1 i D1 j ¤i

where .ei /1i n is an orthonormal basis of Tx M . In other words, the scalar cur-
vature is the sum of the eigenvalues of the Ricci curvature tensor.
64 Chapter 3. The Laplacian on a compact Riemannian manifold

Example 3.1.23. In the case of spaces of constant curvature, the scalar curvature
is just a constant function:
 on Rn , Scal D 0;
 on Sn , Scal D n.n  1/;
 on Hn , Scal D n.n  1/.

Thus, the sectional curvatures determine the Ricci curvature, and the Ricci
curvature determines the scalar curvature. Conversely, in the case of dimension 3
the Ricci curvature determines also all sectional curvatures, because we have the
relation
  Ricx .e1 ; e1 / C Ricx .e2 ; e2 /  Ricx .e3 ; e3 /
Kx Pe1 ;e2 D :
2

3.1.5 Integration on a Riemannian manifold

Let us construct the Hilbert space L2 .M; g/ associated to a Riemannian manifold
.M; g/. For this, we first need to define the canonical measure on a Riemannian
manifold.
Consider a Riemannian manifold .M; g/ of dimension
 n anda local chart
W U  M ! Rn with coordinates denoted by x 1 ; x 2 ;    ; x n . Set gij D
g @x@ i ; @x@j , g WD det.gij / and g j k WD .gj k /1 . The canonical measure of
.M; g/ is
p
d Vg WD g dx 1 dx 2 : : : dx n
where dx 1 dx 2 : : : dx n is the standard Lebesgue measure of Rn . The volume form
of .M; g/ (here we suppose M is oriented) is
p
!g WD g dx 1 ^ dx 2 : : : ^ dx n
R
Hence we have Vol.M / D M !g .
For a local chart W U  M ! Rn of the manifold M and for a measurable
function f W U ! R we define the integral of f on U  M (in the chart ) by
Z Z
 p
f d Vg WD f ı  1 g ı  1 dx 1 dx 2 : : : dx n :
U .U /

Using the change of variables formula we see that this integral does not depend
on the chart W U  M ! Rn . Next, to define integration over the whole of M ,
we use a partition of unity on M , i.e., a finite family .U˛ ; ˛ /˛ where .U˛ /˛ are
open sets of M and .˛ /˛ are smooth positives functions on M which satisfy

supp .˛ /  U˛
 3.1 Basic Riemannian Geometry 65

and X
˛ D 1:
˛
So we can define the integral of f on M by the following formula (which does
not depend on the choice of the partition):
Z XZ
f d Vg WD f ˛ d Vg :
M ˛ U˛

Using the Riemannian metric g we get a canonical isomorphism [ between the

tangent bundle TM and the cotangent bundle T ? M :
8
< TM ! T ? M
[W
: .x; V / 7! [.V /

where [.V /.U / WD g.x/.V; U / with V; U 2 Tx M . With this isomorphism we

can define the notion of gradient of a function: for a function f 2 C 1 .M / we
define the gradient by
rf WD [1 .df /I
in other words, for all V 2 TM we have g.rf; V / D .df /.V /. For a local chart
W U  M ! Rn of M , we have the local formula
X
n
@.f ı  1 / j k @
rf D g :
@x j @x k
j;kD1

Now, let us introduce the divergence operator. Let X be a C 1 vector field on M .

The divergence of X is given by
div.X / WD tr .Y 7! DY X / :
For a local chart W U  M ! Rn of M and for a vector field
X
n
@
XD Xj ;
@x j
j D1

we have p 
1 X @ gX j
n
div.X / D p :
g @x j
j D1

A classical result about this operator is the following: if X is a vector field of M

with compact support, then divergence formula
Z
div.X / d Vg D 0:
M
66 Chapter 3. The Laplacian on a compact Riemannian manifold

3.2 Analysis on manifolds

3.2.1 Distributions on a Riemannian manifold
For a complete introduction to the theory of L. Schwartz distributions on mani-
folds and more generally to the theory of currents (generalization for differential
forms) see for example the book of G. de Rham [DRh].
Let us denote by D .M / the set of smooth functions with compact support on
M . In the language of distributions, D .M / is called the set of test functions.

Definition 3.2.1. A distribution on M is a linear form

8
< D .M / ! R
TW
: ' 7! T ';

which is continuous in the following sense: if .'n /n is a sequence in D .M / such

that for any n the support of 'n is contained in a single compact  set which lies
in the interior of a local coordinate system x 1 ; x 2 ;    ; x n and such that all the
derivatives of 'n tend uniformly to zero as n ! C1, then T 'n tends in R to zero
as n ! C1. The set of distributions on M is denoted D 0 .M /.

Remark 3.2.2. In fact, the space D .M / can be equipped with a topology and
D 0 .M / is the topological dual space of D .M /.
In this book we also use the following notation: for T 2 D 0 .M / and for
' 2 D .M /,
hT; 'iD0 .M /D.M / WD T ':

Example 3.2.3. Let a be a point M . The Dirac distribution ıa is defined by: for
all ' 2 D .M /,
hıa ; 'iD0 .M /D.M / WD '.a/:

Example 3.2.4. A locally integrable function f on M defines a distribution,

called the regular distribution associated to f , by: for all ' 2 D .M /
Z
˝ ˛
Tf ; ' D0 .M /D.M / WD f ' d Vg :
M

Remark 3.2.5. From now we do not distinguish between locally integrable func-
tions and the associated regular distributions.
For a distribution T 2 D 0 .M / and for a local coordinate system .x 1; x 2 ;    ; x n /
i of T by: for all ' 2 D .M /
@T
we define the partial derivative @x
 
@T @'
; ' WD  T; i :
@x i D 0 .M /D .M / @x D 0 .M /D .M /
 3.2 Analysis on manifolds 67

3.2.2 Sobolev spaces on a Riemannian manifold

First we define the Lebesgue space L2 .M; g/ (or simply L2 .M /) on the manifold
.M; g/ by
Z
L2 .M; g/ WD f W M ! R measurable such that jf j2 d Vg < C1 :
M

This space is a Hilbert space for the scalar product

Z
hu; viL2 .M / WD uv d Vg :
M

Next we define the Sobolev space H 1 .M; g/ (or simply H 1 .M /) by

H 1 .M; g/ WD C 1 .M /;

where the closure is with respect to the norm kkH 1 defined by

q
kukH 1 WD kuk2L2 C kruk2L2 :

Another way to define the space H 1 .M / is

˚ 
H 1 .M / D u 2 L2 .M /I ru 2 L2 .M /

where the derivation r is in the sense of distributions.

H 1 .M / is a Hilbert space for the scalar product

hu; viH 1 WD hu; viL2 C hru; rviL2 :

Finally, we define the Sobolev space H01 .M / (or simply H01 .M /) by

H01 .M; g/ WD D .M /;

where the closure is taken with the respect to the norm kkH 1 .
So we have:

D .M /  H01 .M /  H 1 .M /  L2 .M /:

Recall that for the norm kkL2 we have

D .M / D L2 .M /:

To finish let us state two important theorems in Sobolev space theory. The first is
the Poincaré inequality (see, for example, [Heb]):
68 Chapter 3. The Laplacian on a compact Riemannian manifold

Theorem 3.2.6 (Poincaré theorem). Suppose the manifold .M; g/ is compact.

There exists a positive constant CM such that for any function u 2 H01 .M /
kukL2 .M /  CM krukL2 .M / :

Consequently, the bilinear form

8
< H01 .M / H01 .M / ! R
h; iH 1 W
0 : .u; v/ 7! hu; vi 1 WD hru; rvi 2
H 0
L

is a scalar product on H01 .M /. Moreover, the associated norm kkH 1 is equivalent

0

to the norm kkH 1 . Therefore H01 .M /; h; iH 1 is a Hilbert space.

0
An obvious consequence of the Poincaré inequality is:

Corollary 3.2.7. Suppose the manifold .M; g/ is compact. The canonical embed-
 
ding of H01 .M /; kkH 1 into L2 .M /; kkL2 is bounded.
0

Now, let us recall the Rellich theorem (see for example [Heb]).

Theorem 3.2.8 (Rellich theorem). Suppose the manifold .M; g/ is compact. The
canonical embeddings
 1 
H .M /; k:kH 1 ! L2 .M /; k:kL2
and
H01 .M /; k:kH 1 ! L2 .M /; k:kL2
0

are compacts.

3.2.3 The Laplacian operator and the Green formula

We define the Laplace–Beltrami operator (or simply Laplacian) on the Rieman-
nian manifold .M; g/ by
8
< C 2 .M / ! C 0 .M /
g W
: f 7!  .f / WD div.rf /:
g

In local coordinates, for a C 2 real valued function f on M and for a local chart
W U  M ! R of M , the Laplace–Beltrami operator is given by the local
expression
 
1 X @ p j k @.f ı  1 /
n
g f D p gg ;
g @xj @xk
j;kD1
 3.3 Exercises 69

where g D det.gij / and .g j k / D .gj k /1 . The Laplacian operator is an un-

bounded linear operator which satisfies for any C 2 pair of functions .'; /
g .' / D g ' C 2g .r'; r / C 'g :
This formula in turn yields the famous Green’s formula:

Theorem 3.2.9 (Green’s formula aka. Integration by parts formula). Let .M; g/
be an oriented Riemannian manifold and  a subset of M with a smooth boundary
@M . Denote by  the unit normal vector field to the boundary @M . For any
' 2 C 2 .M / and 2 C 1 .M /, at least one of which has a compact support, we
have:
Z Z Z
g ' d Vg D .g .; r'/  g .; r / '/ d Ag C 'g d Vg :
M @M M

3.3 Exercises
Exercise 3.3.1. If f is a constant function on a manifold show that df D 0
everywhere.

Exercise 3.3.2. Prove that flat tori are Riemannian manifolds.

Exercise 3.3.3. Verify that the hyperbolic space Hn WD fx 2 Rn ; kxk < 1g

equipped with the hyperbolic metric g is a Riemannian manifold.

Exercise 3.3.4. Let .M; g/ be a Riemannian manifold and f W M ! R a function

such that krf k D 1. Show that the integral curves of rf are geodesics.

Exercise 3.3.5. Let . t /t be a one-parameter family of closed geodesics on a Rie-

mannian manifold .M; g/. Prove that for any t; t 0 the length of t and t 0 are
equals.

Exercise 3.3.6. Prove that a Riemannian manifold M equipped with the Rieman-
nian distance .see Section 3.1.3/ is a metric space and the topology associated to
this metric coincides with the natural topology of the manifold M .

Exercise 3.3.7. Verify that the volume of the sphere with the metric can is
.4/n .n  1/Š
Vol.S2n ; can/ D
.2n  1/Š
and
 nC1
Vol.S2nC1 ; can/ D 2 :
nŠ
70 Chapter 3. The Laplacian on a compact Riemannian manifold

Exercise 3.3.8. Show that the Ricci tensor is symmetric.

Exercise 3.3.9. Show that:

1. on the Euclidean plane R2 geodesics are straight lines;
2. on the round sphere S2 geodesics are great circles;
3. on the hyperbolic Poincaré disk geodesics are arcs of circles meeting the
unit circle orthogonally, or segments through the origin.

Exercise 3.3.10. Prove that H 1 .M / is a Hilbert space.

Chapter 4
Spectrum of the Laplacian on a compact
manifold

In this chapter our first main goal is to prove that the Laplace–Beltrami operator
on a compact Riemannian manifold is self-adjoint with a discrete spectrum. Next,
we will show that the spectrum is a nice geometric invariant. We also present the
minimax principle for the Laplacian. At the end we also define the Schrödinger
operator on a compact manifold.
From now on, all manifolds are supposed to be compact and connected.

4.1 Physical examples

4.1.1 The wave equation on a string
Consider an one-dimensional string of length L > 0. The vibrations of this string
are governed by the wave equation

@2 u @2 u
2
.x; t/  2 .x; t/ D 0
@t @x
with boundary conditions

u.0; t/ D u.L; t/ D 0

and initial data

u.x; 0/ D u0 .x/:
Here u.x; t/ is the amplitude of the point x on the string at time t.
Using the theory of Fourier series (see Section 5.1.1 for more details), the
family
nx
en .x/ WD sin
L n1
˚ 
form a Hilbert basis of H WD f 2 L .Œ0; T / I f .0/ D f .T / D 0 , and for any
2

integer n  1 we have the relation

@2 en n2  2
 .x/ D en .x/:
@x 2 L2
72 Chapter 4. Spectrum of the Laplacian on a compact manifold

2
In fact, the spectrum of the operator  @x
@
2 with the boundary conditions u.0; t/ D
u.L; t/ D 0 is equal to  2 2
n 
:
L2 n1
So if u.x; t/ is a solution of the wave equation, there exists a sequence of real
number .an .t//n1 (which depends on the time t/ such that
C1
X
u.x; t/ D an .t/en .x/:
nD1

Plugging this expression into the wave equation we get

X
C1
n2  2

an00 .t/ C a n .t/ en .x/ D 0;
nD1
L2

and since .en /n is a basis, for any integer n  1 we obtain the following ODE in
the time variable
n2  2
an00 .t/ C an .t/ D 0:
L2
Thus we get for any integer n  1 and for all t  0
n n
an .t/ D An cos t C Bn sin t ;
L L
where An ; Bn are real constants.
Finally, we obtain the expression
C1
X n n nx
u.x; t/ D An cos t C Bn sin t sin ;
nD1
L L L

the coefficients An ; Bn of which are calculated from the initial data u.x; 0/ D
u0 .x/.

4.1.2 The heat equation

Consider here a domain   R3 with a boundary  WD @. The heat equation
on  describes the heat diffusion process on this domain. It is the linear equation
@u
.x; t/  u.x; t/ D f .x; t/
@t
with boundary conditions
u.x; t/ D u` .x; t/ for x 2 
 4.1 Physical examples 73

and initial data

u.x; 0/ D u0 .x/:
Here u.x; t/ is the temperature at the point x 2  at time t. For simplify, we
suppose that f D 0 and u D 0. Using the separation variables technique, we
look for a solution in the form u.x; t/ D '.t/w.x/. Plugging this into the heat
equation we get
' 0 .t/w.x/  ' 0 .t/w.x/ D 0:
Formally, we have for all x 2 , t  0

' 0 .t/ w.x/

D ;
'.t/ w.x/

and since the variables x and t are independent, there exists a real constant  such
that for all x; t we have

' 0 .t/ w.x/

D D :
'.t/ w.x/

In particular, ' 0 .t/ D '.t/, hence

'.t/ D '.0/et :

Moreover,
w.x/ D w.x/
with w.x/ D 0 for x 2 . In other words, the scalar  is an eigenvalue of 
with Dirichlet conditions, therefore if we have a basis .ek /k0 of eigenvectors,

ek D k ek ;

and we get for all x; t

C1
X
u.x; t/ D ck ek t ek .x/:
kD1

4.1.3 The Schrödinger equation

In Quantum Mechanics, a physical particle on a manifold M in a time interval I
is describe by a wave function, that is, a function
8
< M I ! L2 .M /
W
: .x; t/ 7! .x; t/;
74 Chapter 4. Spectrum of the Laplacian on a compact manifold
Z
where the quantity .x; t/ dx represents the probability to find the particle in

the domain   M at time t. In particular, this requires that
Z
2
k kL2 .M / D .x; t/ dx D 1:


The Schrödinger operator on a Riemannian manifold .M; g/ is the linear operator

h2
P„ WD  g C V
2
with V a function of L1 loc .M / such that limjxj!1 V .x/ D C1. This operator
is self-adjoint and its spectrum is discrete. For an initial state 0 2 L2 .M / the
quantum dynamics associated to Ph is given by the formula (using the bounded
functional calculus)
.t/ D U.t/ 0 2 L2 .M /;
where t
U.t/ WD ei h Ph :
The derivative of this equation is
@ .t/ i
D  Ph .t/;
@t h
so we arrive at the famous Schrödinger equation
@
ih .x; t/ D g .x; t/ C V .x/ .x; t/:
@t
Now if we have a basis .ek /k0 of eigenvectors of , using also the bounded
functional calculus, we get for all t and for any integer n
t t
ei h Ph en D ei h n en I
PC1
thus, if we write the initial state as 0 D nD0 an en with an D h 0 ; en iL2 , we
get for all t
C1
! C1
X X
i ht Ph t
.t/ D U.t/ 0 D e an en D an ei h n en :
nD0 nD0

Note that a good way to understand (in the semi-classical limit h ! 0: see Sec-
tion 5.2.4) this quantum dynamics is to study the quantum auto-correlation func-
tion
a.t/ WD jr.t/j D jh .t/; 0 iH j :
 4.2 A class of spectral problems 75

This function is a quantum analog of the Poincaré’s return function: if the wave
packet .t/ is localised on the initial state 0 , then a.t/ is close to 1 (because
k 0 kL2 D 1/. The auto-correlation function is a fundamental tool in the dy-
namical study of quantum integrable systems (see for example [Co-Ro], [Robi1],
[Robi2], [Rob2], [Rob3], [Lab1], [Lab2]). Indeed, in the case of semi-classical
completely integrable systems and for a localized initial state, at the beginning
the quantum dynamics is periodic with a period which corresponds to the clas-
sical Hamiltonian flow period. Next for a large time scale, a new period Trev of
the quantum dynamics appears: this is the revival phenomenon. For t D Trev the
initial wave packet reforms again at t D Trev . We have also the phenomenon of
fractional revival of initial wave packets for time t D pq Trev , with pq 2 Q: there
is a formation of a finite number of clones of the original wave packet 0 with
a constant amplitude and differing in the phase plane from the initial wave packet
by fractions pq Trev .

4.2 A class of spectral problems

4.2.1 The closed eigenvalue problem
Let .M; g/ be a closed manifold, i.e., compact without boundary, for example the
sphere, the torus. . . The closed spectral problem is: find all real numbers  such
that there exists a function u 2 C 1 .M / with u ¤ 0 for which

g u D u:

From a spectral point of view we want to find the (point) spectrum of g on the
domain D D C 1 .M /.

4.2.2 The Dirichlet eigenvalue problem

Let .M; g/ be a compact manifold with boundary. The Dirichlet spectral problem
is: find all real numbers  such that there exists a function u 2 C 1 .M / with
u ¤ 0 for which
8
< g u D u
: u D 0 on the boundary of M:

From a spectral point of view we want to find the (point) spectrum of g on the
domain D D D .M /.
76 Chapter 4. Spectrum of the Laplacian on a compact manifold

4.2.3 The Neumann eigenvalue problem

Let .M; g/ be a compact manifold with boundary. The Neumann spectral problem
is: find all real numbers  such that there exists a function u 2 C 1 .M / with u ¤ 0
for which 8
< g u D u
: ru  D 0 on the boundary of M;
where is the unit inward normal vector field on .
From a spectral point of view we want to find the (point) spectrum of g on
the domain D D ff 2 C 1 .M /I rf  D 0 on the boundary of M g.

4.2.4 Other problems

Obviously there exist many other problems, for example
 Mixed
` problem: suppose the boundary  of M can be written as  D
1 2 the problem is: find all real numbers  such that there exists
a function u 2 C 1 .M / with u ¤ 0 for which
8
< g u D u
u D 0 on 1
: ru  D 0 on  :
2

 Steklov problem: for a fixed function 2 C 0 ./, find all real numbers  for
which there exists a function u 2 C 1 .M / with u ¤ 0 such that
8
< g u D 0 in M
: ru  D  u on the boundary on M:

4.3 Spectral theorem for the Laplacian

For each of the three above cases (closed, Dirichlet, or Neumann problem), using
Green’s formula conclude that for all functions '; in the domain
 D D C 1 .M /
 D D D .M /
 D D ff 2 C 1 .M /I rf  D 0 on  the boundary of M g
we always have the relation
Z Z
g ' d Vg D 'g d Vg :
 
 4.3 Spectral theorem for the Laplacian 77

Hence, the operator g is symmetric. Moreover, for any functions '; in the
domain D we have also
Z Z
 'g ' d Vg D jr'j2 d Vg  0;
 
therefore the operator g is positive.
Consequently, let ;  be two eigenvalues such that  ¤  and let u; v be
respective eigenfunctions (i.e., g u D u and g v D v/. Since g is
symmetric, we have ˝ ˛ ˝ ˛
g u; v L2 D u; g v L2 ;
i.e.,
.  / hu; viL2 D 0:
Therefore, if we denote by E ./ the eigenspace corresponding to the eigenvalue
, the spaces E ./ and E ./ are L2 -orthogonal.
Moreover, every eigenvalue is non-negative. Indeed, let  be an eigenvalue
and let u be an eigenfunction of . Then
˝ ˛
g u; u L2 D  kukL2  0;
so   0.
In fact, we have the more general theorem concerning the spectral theory of
the Laplacian (see also [Bér2]):

Theorem 4.3.1. Let .M; g/ be a compact Riemannian manifold with boundary 

.possibly empty/. Then the following assertions hold true for each of the above
spectral problems.
(i) The spectrum and the point spectrum of g coincide and consist of a real
infinite sequence
0  1 .M /  2 .M /      k .M /    
such that k .M / ! C1 as k ! C1.
(ii) Each eigenvalue has finite multiplicity and the eigenspaces corresponding
to distinct eigenvalues are L2 -orthogonal, and if we denote by E .k / the
eigenspace corresponding to the eigenvalue k , then
M
E .k / D L2 .M /
k1

where the closure is taken for the norm kkL2 .

Moreover, for the closed problem each eigenfunction belongs to the space
H 1 .M /, for the Dirichlet problem each eigenfunction belongs to the space
H01 .M /, and for the Neumann problem the gradient of every eigenfunction
vanishes on the boundary of M .
(iii) Each eigenfunction is analytic.
78 Chapter 4. Spectrum of the Laplacian on a compact manifold

Hence, in conclusion, for each Riemannian manifold there exists a unique

sequence of real numbers .k .M //k0 such that

Spec.g ; D/ D fk .M /; k  0g :

Recall that the multiplicity

 of an eigenvalue k .M / is the dimension of the vector
space ker I C g .

Definition 4.3.2. The sequence above is called the spectrum of the manifold
.M; g/ and is denoted by Spec.M; g/.

Spectral geometry studies the relationships between the geometry of .M; g/

and the spectrum Spec.M; g/. A fundamental notion in spectral geometry is the
following:

Definition 4.3.3. Two (compact) Riemannian manifolds .M; g/ and .M 0 ; g 0 / are

isospectral if Spec.M; g/ D Spec.M 0 ; g 0 /. A spectral invariant on .M; g/ is
a quantity which is completely determined by Spec.M; g/.

Since the spectrum of a manifold depends only on the divergence and the gra-
dient operators, and those operators depend only on the Riemannian structure on
.M; g/, the spectrum depends only on the Riemannian structure. Consequently,
we have

Proposition 4.3.4. Any two isometric Riemannian manifold are isospectral.

4.4 A detailed proof by a variational approach

In this subsection we give a detailed proof of the Theorem 4.3.1. The proof uses
only the spectral theory of bounded operators. In the book [Bér8] by P. Bérard we
can find another type of proof. Here we use the Lax–Milgram theorem.

4.4.1 Variational generic abstract eigenvalue problem

Consider two Hilbert spaces H and V such that V  H and V is dense in H
(in the norm of H ). Let us denote by I10 the canonical injection from V into
H and suppose that I10 W .V; k  kV / ! .H; k  kH / is bounded and compact. In
particular, this means that there exists a constant K  0 such that for all u 2 V
we have kukH  K kukV . We also consider a bilinear symmetric form a on
V 2 which is continuous and elliptic (or coercive) (that is, there exists a constant
C  0 such that for all .u; v/ 2 V 2 we have ja.u; v/j  C kukV kvkV , and there
exists a constant ˛ > 0 such that for all u 2 V we have ja.u; u/j  ˛ kuk2V ).
 4.4 A detailed proof by a variational approach 79

Now, for f a fixed vector in H we shall be interested in the following varia-

tional problem: 8
< find u 2 V; such that for all v 2 V
: a.u; v/ D hf; viH :
If we denote by ` the map
8
< V ! R
`W
: v 7! hf; vi ;
H

it is clear that ` is linear and by the Cauchy–Schwarz inequality, for all v 2 V

j`.v/j D jhf; viH j  kf kH kvkH
 K kf kH kvkV ;
and so the operator `W .V; k  kV / ! .R; j  j/ is bounded. Therefore, by the Lax–
Milgram theorem, there exists a unique vector u 2 V such that
a.u; v/ D `.v/
for all v 2 V , and moreover
K
kukV  p kf kH :
˛
Since for every vector f 2 H the vector solution u 2 V of the variational problem
is unique, we can define the following map
8
< H ! V
TW
: f 7! u;

where u is the unique solution of the problem

8
< find u 2 V; such that for all v 2 V
: a.u; v/ D `.v/:
This mean that for all v 2 V ,
a .Tf; v/ D hf; viH :
It is clear (use the unicity of solution) that T is a linear map between the Hilbert
spaces H and V . Moreover,
K
kTf kV D kukV  p kf kH
˛
for all f 2 H , and so the operator T W .H; kkH / ! .V; kkV / is bounded.
80 Chapter 4. Spectrum of the Laplacian on a compact manifold

Let us introduce the restriction Tr of the operator T to the space Hilbert space
V H 8
< V ! V
Tr W
: f 7! Tf:

Since
K K2
kTf kV  p kf kH  p kf kV
˛ ˛
for all f 2 V , we deduce that the operator Tr W .V; kkV / ! .V; kkV / is bounded.
Clearly,
Tr D T ı I10
and since the I10 is bounded and compact, the operator Tr is also compact (the
set of compact operator is an ideal of the bounded operators, see Theorem 2.5.3).
Now let us verify that zero is not an eigenvalue of the operator Tr . Suppose on the
contrary, that there exists u 2 V with u ¤ 0 such that Tr u D T u D 0. Since
a .Tr u; u/ D hu; uiH , we get 0 D kukH , which is a contradiction.
Next for all u; v 2 V we have

a .Tr u; v/ D a .T u; v/
D hu; viH D hv; uiH
D a .Tr v; u/ D a .u; Tr v/ ;

so the bounded operator Tr is self-adjoint with respect to the scalar product a .; /
on the space Hilbert space V .
Thus, the operator Tr is bounded, compact, self-adjoint and 0 is not an eigen-
value, so using the fundamental theorem of the spectral theory of compact op-
erators (see Theorem 2.5.6) the spectrum .Tr / and the point spectrum Spec.Tr /
coincide and this set is discrete: it consists of a sequence .k /k of real eigenvalues
such that k > 0 and k ! 0 as k ! C1. For any integer k  0 the dimen-
sion of ker .k I  Tr / is finite, i.e., each eigenvalue of Tr has finite multiplicity.
Moreover, the corresponding eigenvectors .vk /k of .k /k form a Hilbert basis of
the space V . In particular for every integer k; `,

T vk D Tr vk D k vk and a .vk ; v` / D ık;` :

To finish, observe that for every integer k and for all v 2 V ,

a .Tr vk ; v/ D hvk ; viH

and on the other hand we have also

a .Tr vk ; v/ D k a .vk ; v/ :
 4.4 A detailed proof by a variational approach 81

Consequently,
1
a .vk ; v/ D hvk ; viH :
k
So, if we denote for each integer n  0
vk
ek WD p
k

we have for all .k; `/ 2 N2


vk vk 1
hek ; ek iH D p ; p D hvk ; vk iH
k k H
k
D a .vk ; vk / D 1;

and if k ¤ `

vk v` 1 1
hek ; e` iH D p ; p D p p hvk ; v` iH
k ` H k `
p
k
D p a .vk ; v` / D 0:
`

Hence, the sequence .ek /k is an H -orthonormal family in V , therefore by the

density of V in H , the family .ek /k form a Hilbert basis of H such that for every
integer k and for all v 2 V we have
1 1 1 1
a .ek ; v/ D p a .vk ; v/ D p hvk ; viH D hek ; viH :
k k k k

Now let us apply this theory to our spectral problems (closed, Dirichlet or Neu-
mann problem).

4.4.2 The closed eigenvalue problem

For a fixed f 2 L2 .M / consider the following problem
8
< find u 2 L2 .M /; such that
:  u C u D f in D 0 .M /:
g

The corresponding variational formulation reads

8
< find u 2 H 1 .M /; such that for all v 2 H 1 .M /
: a.u; v/ D `.v/
82 Chapter 4. Spectrum of the Laplacian on a compact manifold

where for all u; v 2 H 1 .M /

Z Z
a.u; v/ WD hu; viH 1 D rurv d Vg C uv d Vg
M M

and for all v 2 H 1 .M / Z

`.v/ WD f v d Vg :
M

The two previous formulations are equivalent. Indeed, suppose u 2 L2 .M / is

a solution of g u C u D f in D 0 .M /. This means that for all v 2 D .M / we
have ˝ ˛
g u C u; v D0 .M /D.M / D hf; viD0 .M /D.M / :
Hence, for all v 2 D .M /,

hru; rviD0 .M /D.M / C hu; viD0 .M /D.M / D hf; viD0 .M /D.M / ;

so by definition of regular L2 -distributions, for all v 2 D .M /

Z Z Z
rurv d Vg C uv d Vg D f v d Vg :
M M M

Therefore, since D .M / is dense in H 1 .M / in the L2 norm,

Z Z Z
rurv d Vg C uv d Vg D f v d Vg ;
M M M

for all v 2 H 1 .M /, i.e., a.u; v/ D `.v/. Conversely, suppose u 2 H 1 .M /

satisfies a.u; v/ D `.v/ for all v 2 H 1 .M /, hence for all v 2 D .M / we have
Z Z Z
rurv d Vg C uv d Vg D f v d Vg :
M M M

Integrating by parts we get

Z Z Z
 g uv d Vg C uv d Vg D f v d Vg
M M M

for all v 2 D .M /. Consequently,

˝ ˛
g u C u; v D0 .M /D.M / D hf; viD0 .M /D.M /

for all v 2 D .M /, hence g u C u D f in D 0 .M /.

Now let us focus on the variational formulation. Here both V WD H 1 .M /
and H WD L2 .M / are Hilbert spaces and obviously H 1 .M /  L2 .M / and
H 1 .M / is dense in L2 .M / (in the norm of L2 .M /). Since for all u 2 H 1 .M /,
 4.4 A detailed proof by a variational approach 83

kukL2  kuk  H 1 Rellich’s theorem

  (Theorem 3.2.8)
 shows that the canonical in-
jection I10 W H 1 .M /; kkH 1 ! L2 .M /; kkL2 is bounded and compact. More-
over by the Cauchy–Schwarz inequality, we obtain for any u; v 2 H 1 .M /:

ja.u; v/j D jhu; viH 1 j  kukH 1 kvkH 1

and for any u 2 H 1 .M / we have a.u; u/ WD kuk2H 1 . In consequence

 the bilinear

form a .; / is bounded and elliptic on the Hilbert space H 1 .M /; h; iH 1 . On
the other hand, for any v 2 H 1 .M /
ˇZ ˇ
ˇ ˇ
ˇ
j`.v/j D ˇ f v d Vg ˇˇ  kf kL2 kvkL2
M
 kf kL2 kvkH 1 :

So the linear form ` is bounded on H 1 .M /; h; iH 1 . It follows from the previous
theory with V WD H 1 .M / and H WD L2 .M / (in this case the operator T is equal
 1
to T D g C I , which means that 1 is in the resolvent set of g ) that
there exists a infinite sequence .k /k of real numbers such that every k > 0 and
k ! 0 as k ! C1 (in fact Spec.Tr / D f.k /k g) and there exists a Hilbert
basis .ek /k of H such that for every integer k and for all v 2 V we have

1
a .ek ; v/ D hek ; viL2 :
k

Hence, for every integer k and for all v 2 H 1 .M /,

Z Z Z
1
rek rv d Vg C ek v d Vg D ek v d Vg :
M M k M

Integrating by parts (here M is closed) we get for all v 2 H 1 .M /

Z Z Z
1
 g ek v d Vg C ek v d Vg D ek v d Vg
M M k M

i.e., for all v 2 H 1 .M /

Z    
1
g C 1  I .ek / v d Vg D 0:
M k

Using the density of H 1 .M / into L2 .M / we obtain that for any integer k

   
1
g C 1  I ek D 0;
k
84 Chapter 4. Spectrum of the Laplacian on a compact manifold

hence, for all integers k,

 
1
g ek D  1 ek :
k

So, if we denote k WD 1
k
 1 for any integer k, we have

g ek D k ek :

Consequently, .k /k  Spec.g / and .ek /k is a Hilbert eigenbasis of g on

the space L2 .M /.
Conversely let us verify the important fact that there are no other eigenvalues
in Spec.g /. Indeed, let  2 Spec.g /, so there exists f ¤ 0 in H 1 .M /
such that
g f D f:
Then
g f C f D . C 1/f;
i.e.,  
g C I f D . C 1/f:
 1
Since 1 is in the resolvent set of g we have (here T D g C I )

f D . C 1/Tf:

Next, since f 2 V D H 1 .M / we deduce that

f D . C 1/Tr f

and since  ¤ 1 (because 1 is in the resolvent set of g ) we get

1
Tr f D f:
C1
1
Therefore C1 2 Spec.Tr / hence there exists an integer k such that C1
1
D k . It
follows that  D k  1, i.e., and so we have .k /k D Spec.g /.
1

4.4.3 The Dirichlet eigenvalue problem

For a fixed f 2 L2 .M / consider the problem
8
< find u 2 L2 .M /; such that
g u D f in D 0 .M /
: uj D 0:
 4.4 A detailed proof by a variational approach 85

The variational formulation of this problem reads

8
< find u 2 H01 .M /; such that for all v 2 H01 .M /
: a.u; v/ D `.v/

where for all u; v 2 H01 .M /,

Z
a.u; v/ WD hu; viH 1 D rurv d Vg ;
0
M

and for all v 2 H01 .M /, Z

`.v/ WD f v d Vg :
M

The two previous formulations are equivalent. Indeed, suppose u 2 L2 .M / is

solution of g u D f in D 0 .M /, this means that for all v 2 D .M / we have
˝ ˛
g u; v D0 .M /D.M / D hf; viD0 .M /D.M / ;

hence for all v 2 D .M /

hru; rviD0 .M /D.M / D hf; viD0 .M /D.M / :

By the definition of regular L2 -distributions, we get for all v 2 D .M /

Z Z
rurv d Vg D f v d Vg ;
M M

whence, since D .M / is dense in H01 .M / in the L2 norm, we have for all v 2

H01 .M / Z Z
rurv d Vg D f v d Vg ;
M M

i.e., a.u; v/ D `.v/. Now suppose u 2 H01 .M / satisfies a.u; v/ D `.v/ for all
v 2 H01 .M /, hence for all v 2 D .M / we get
Z Z
rurv d Vg D f v d Vg :
M M

Integrating by parts (here v vanishes on the boundary of M ) for all v 2 D .M / we

have Z Z
 g uv d Vg D f v d Vg :
M M
86 Chapter 4. Spectrum of the Laplacian on a compact manifold

Consequently,
˝ ˛
g u; v D0 .M /D.M / D hf; viD0 .M /D.M /

for all v 2 D .M /, hence g u D f in D 0 .M /.

Now let us focus on the variational formulation. Here both V D H01 .M / and
H D L2 .M / are Hilbert spaces, H01 .M /  L2 .M /, and H01 .M / is dense in
L2 .M / (in the norm of L2 .M /). From the Poincaré inequality (Theorem 3.2.6)
and from the Rellich theorem (Theorem 3.2.8) we deduce that the canonical injec-
 
tion I10 W H01 .M /; k  kH 1 ! L2 .M /; k  kL2 is bounded and compact. More-
0
over, by the Cauchy–Schwarz inequality, for any u; v 2 H01 .M /
ˇ ˇ
ˇ ˇ
ja.u; v/j D ˇhu; viH 1 ˇ  kukH 1 kvkH 1 ;
0 0 0

and for any u 2 H01 .M /, a.u; u/ WD kuk2H 1 . Therefore, the bilinear form a .; / is
0

bounded and elliptic on the Hilbert space H01 .M /; h; iH 1 . On the other hand,
0
the Poincaré inequality implies that for any v 2 H01 .M / we get
ˇZ ˇ
ˇ ˇ
j`.v/j D ˇˇ f v d Vg ˇˇ  kf kL2 kvkL2
M
 CM kf kL2 krvkL2 D CM kf kL2 kvkH 1 ;
0

and so the linear form ` is bounded on the Hilbert space H01 .M /; h; iH 1 . It
0
follows from the previous theory with V WD H01 .M / and H WD L2 .M / (in this
case the operator T is equal to T D 1 g , this meaning that 0 is in the resolvent
set of g ) that there exists a infinite sequence .k /k of real numbers such that
every k > 0 and k ! 0 as k ! C1 (in fact Spec.Tr / D f.k /k g); and there
exists a Hilbert basis .ek /k of H , such that for every integer k and for all v 2 V
1
a .ek ; v/ D hek ; viL2 :
k

Therefore, for every integer k and for all v 2 H01 .M /,

Z Z
1
rek rv d Vg D ek v d Vg :
M k M
Integrating by parts (here v vanishes on the boundary of M ) we obtain for all
v 2 H01 .M / Z Z
1
 g ek v d Vg D ek v d Vg ;
M k M
 4.4 A detailed proof by a variational approach 87

i.e., Z  
1
g  I .ek / v d Vg D 0
M k
for all v 2 H01 .M /, so by the density of H01 .M / in L2 .M / we have for any k
 
1
g  I ek D 0;
k

i.e.,
1
g ek D ek :
k
Thus, if we denote k WD 1
k for any k; we get

g ek D k ek :

Hence, .k /k  Spec.g / and .ek /k form a Hilbert eigenbasis of g on the
space L2 .M /.
Conversely, let  2 Spec.g /, so there exists f ¤ 0 in H01 .M / such that

g f D f:
 1
Since 0 is in the resolvent set of g we have (here T D g )

f D Tf

and since v 2 V D H01 .M /

f D Tr f:
To conclude, since  ¤ 0 (because 0 is in the resolvent set of g ) we get

1
Tr f D f;


thus 1 2 Spec.Tr /: there exists a k such that 1


D k , hence  D 1
k
and  D k .
We have shown that .k /k D Spec.g /.

4.4.4 The Neumann eigenvalue problem

To finish, we consider the Neumann eigenvalue problem: for a fixed f 2 L2 .M /,
8
< find u 2 L2 .M /; such that
g u C u D f in D 0 .M /
: ru  D 0 on :
88 Chapter 4. Spectrum of the Laplacian on a compact manifold

The variational formulation of this problem reads

8
< find u 2 H 1 .M /; such that for all v 2 H 1 .M /
: a.u; v/ D `.v/

where for all u; v 2 H 1 .M /

Z Z
a.u; v/ WD hu; viH 1 D rurv d Vg C uv d Vg
M M

and for all v 2 H 1 .M / Z

`.v/ WD f v d Vg :
M

The two previous formulations are equivalent. Indeed, suppose u 2 L2 .M / is

a solution of g u C u D f in D 0 .M / with ru  D 0 on ; this means that
for all v 2 D .M / we have
˝ ˛
g u C u; v D0 .M /D.M / D hf; viD0 .M /D.M / :

Hence, for all v 2 D .M /

hru; rviD0 .M /D.M / C hu; viD0 .M /D.M / D hf; viD0 .M /D.M / ;

so by the definition of regular L2 -distributions, we get for all v 2 D .M /

Z Z Z
rurv d Vg C uv d Vg D f v d Vg :
M M M

Since D .M / is dense in H 1 .M / in the L2 norm, for all v 2 H 1 .M / we get

Z Z Z
rurv d Vg C uv d Vg D f v d Vg
M M M

i.e., a.u; v/ D `.v/. Now, suppose u 2 H 1 .M / satisfies a.u; v/ D `.v/ for all
v 2 H 1 .M /, hence for all v 2 D .M /
Z Z Z
rurv d Vg C uv d Vg D f v d Vg ;
M M M

so integrating by parts we get for all v 2 D .M /

Z Z Z
 g uv d Vg C uv d Vg D f v d Vg :
M M M
 4.4 A detailed proof by a variational approach 89

Consequently, for all v 2 D .M /

˝ ˛
g u C u; v D0 .M /D.M / D hf; viD0 .M /D.M /

hence g u C u D f in D 0 .M /. On the other hand, for all v 2 H 1 .M / inte-

grating by parts we get
Z Z Z
g uv d Vg D ru  v d  rurv d Vg :
M  M

Therefore,
Z Z Z Z
rurv d Vg  ru  v d C uv d Vg D f v d Vg ;
M  M M

Whence Z
ru  v d D 0

for all v 2 H 1 .M /. By density, we conclude that ru  D 0 on .
From this point on, the study of the variational formulation is carried out in
exactly the same way as in the case of closed problem.

4.4.5 A remark on the variational formulation

The variational formulation is also very important in numerical analysis, indeed,
the finite element method is based on this formulation. To outline this method,
consider the generic variational problem
8
< find u 2 V; such that for all v 2 V
: a.u; v/ D `.v/

where V is a Hilbert space, a.; / is a bilinear symmetric form on V 2 continuous

and elliptic and `./ is a bounded linear form on V . The approach of the finite
element method is to find a “nice” finite-dimensional subspace Vh of V and to
solve the associated variational problem on Vh :
8
< find uh 2 Vh ; such that for all vh 2 Vh
: a.uh ; vh / D `.vh /:

If we have a basis .ei /1i n of Vh , the previous problem is equivalent to

8
< find uh 2 Vh ; such that for all i 2 f1; : : : ; ng
: a.uh ; ei / D `.ei /:
90 Chapter 4. Spectrum of the Laplacian on a compact manifold

j
Since every vector uh 2 Vh is characterized by its coordinates uh in the
  1j n
basis ej 1j n , namely
Xn
uh D ujh ej ;
j D1

the previous problem is equivalent to

8
ˆ
ˆ find the scalars ujh 2 R; such that for all i 2 f1; : : : ; ng
ˆ
ˆ 1j n
<
ˆ
ˆ X
n
ˆ j
uh a.ej ; ei / D `.ei /:
:̂
j D1

 t
In the matrix language we have to find a vector U D u1h u2h : : : unh such that

AU D L
 
where A WD a.ej ; ei / 1i;j n 2 Mn .R/ and L WD .`.e1 / `.e2 / : : : `.en //.

4.5 The minimax principle and applications

4.5.1 A physical example
Let us consider a one-dimensional elastic beam of length L > 0. The energy of
the deformation (under vibrations) for a homogeneous elastic beam is given by
Z L ˇ 00 ˇ2
E0 D ˇu .x/ˇ dx:
0

Therefore, the main frequency of the vibrations is equal to

Z L ˇ 00 ˇ2
ˇu .x/ˇ dx
F0 D min Z0 L :
u¤0 2
ju.x/j dx
0

In the case of a nonhomogeneous elastic beam the energy is

Z L ˇ ˇ
ED ˇa.x/u00 .x/ˇ2 dx
0
 4.5 The minimax principle and applications 91

where a is a non-negative function, hence in this case the main frequency of the
vibrations for the nonhomogeneous elastic beam is equal to
Z L ˇ ˇ
ˇa.x/u00 .x/ˇ2 dx
F WD min 0
Z L
:
u¤0 2
ju.x/j dx
0

4.5.2 The minimax theorem

Let .M; g/ be a compact Riemannian manifold (closed or with a boundary where
Dirichlet or Neumann boundary conditions are imposed).

Definition 4.5.1. The Rayleigh quotient R.'/ of a function '; ' ¤ 0 on .M; g/
is defined by
Z
jr'j2 d Vg
R.'/ WD M Z :
' 2 d Vg
M

Remark 4.5.2. In the case where M is closed, for any function ' 2 C 1 .M / (or
more generally for any function ' 2 H 1 .M /) we have the equality
Z Z
2
jr'j d Vg g ' ' d Vg
R.'/ D M Z D MZ :
' 2 d Vg ' 2 d Vg
M M

This also holds in the case of the Dirichlet problem for any function ' 2 D .M /
(or more generally for any function ' 2 H01 .M /), and the same is true for the
Neumann problem for any function ' 2 C 1 .M / such that the gradient vanishes
on the boundary of M .
First, let us show

Proposition 4.5.3. For each spectral problem, let  be an eigenvalue of .M; g/

and u 2 C 1 .M /, u ¤ 0 an associated eigenfunction. Then
Z
jruj2 d Vg
 D R.u/ D M Z :
u2 d V g
M
92 Chapter 4. Spectrum of the Laplacian on a compact manifold

Proof. Let u 2 C 1 .M /; u ¤ 0 be an eigenfunction:

g u D u:

Then, by multiplying by u and integrating over the manifold M we obtain

Z Z
g uu d Vg D  u2 d V g :
M M

In integrating by parts1 we have

Z Z Z
g uu d Vg D ru  ru d Vg D jruj2 d Vg ;
M M M

so we finally get Z
jruj2 d Vg
D Z
M
D R.u/: 
u2 d V g
M

Proposition 4.5.4. For each spectral problem and for any k  1, the following
characterization holds true:

k .M / D max R.'/;
'2Vk
'¤0

where Vk WD span .e1 ; e2 ; : : : ; ek /. An alternative characterization is

1 .M / D min R.'/
'2V
'¤0

and for all k  2

k .M / D min R.'/
?
'2Vk1
'¤0

?
where Vk1 D span .ek ; ekC1 ; : : :/.

Proof. For any integer n we compute the Rayleigh quotient for the Hilbert eigen-
basis .en /n and we get

R .en / D n .M / D R .vn / :
1 In the closed case this is clear, in the Dirichlet case the eigenfunction u vanishes on the boundary of

M , and in the Neumann case the gradient of the eigenfunction u vanishes on the boundary of M .
 4.5 The minimax principle and applications 93

Since R.˛'/ D R.'/ for all scalar ˛ and for any function ', it is clear that
1 .M / D max R.'/;
'2V1
'¤0
where V1 WD span .e1 /.
Next for any function ' such that
C1
X
'D h'; ei iL2 ei ;
i D0
we have
Z C1
X
g ' ' d Vg i .M / h'; ei i2L2
i D0
R.'/ D MZ D C1
:
' 2 d Vg X
M
h'; ei i2L2
i D0
Proceeding by induction, consider the set Vk WD span .e1 ; e2 ; : : : ; ek /. Then, for
all ' 2 Vk ,
Xk X k
i .M / h'; ei i2L2 k .M / h'; ei i2L2
i D0 i D0
R.'/ D  D k .M /:
X
k X
k
h'; ei i2L2 h'; ei i2L2
i D0 i D0

Hence, max'2Vk ; '¤0 R.'/  k .M / and since R .ek / D k .M /, we get

k .M / D max R.'/:
'2Vk
'¤0

For the other characterization, we have

1 .M / D min R.'/
'2V
'¤0
and for all k  2
k .M / D min R.'/;
?
'2Vk1
'¤0
? ?
with Vk1 D span .ek ; ekC1 ; : : :/. Indeed, for all ' 2 Vk1
C1
X C1
X
i .M / h'; ei i2L2 k .M / h'; ei i2L2
i Dk i Dk
R.'/ D C1
 C1
D k .M /:
X X
h'; ei i2L2 h'; ei i2L2
i Dk i Dk
94 Chapter 4. Spectrum of the Laplacian on a compact manifold

Hence, min'2V ? '¤0 R.'/  k .M /, and since R .ek / D k .M / we get

k1 ;

k .M / D min R.'/: 
?
'2Vk1
'¤0

Finally we get the minimax variational characterization of eigenvalues:

Theorem 4.5.5 (Minimax principle). For each of the above spectral problems and
for any k  1, we have Z
jr'j2 d Vg
k .M / D min max Z
M
E H ? .M / '2E
dim.E /Dk '¤0 ' 2 d Vg
M

where H ? .M / WD H01 .M / in the case of Dirichlet problem and H ? .M / D

H 1 .M / in the case of Neumann or closed problem.

Proof. For all E  H ? .M / such that dim.E/ D k, we have

k .M /  max R.'/:
'2E
'¤0

?
Indeed, for all nontrivial functions ' 2 E such that ' 2 Vk1 (here we have
?
E \ Vk1 ¤ ; because dim.E/ D k), since k .M / D min'2V ? ; '¤0 R.'/ we
k1
obtain
k .M /  R.'/:
Consequently, for all E  H ? .M / such that dim.E/ D k we have k .M / 
max'2E R.'/, hence

k .M /  min max R.'/:

E H ? .M / '2E
dim.E /Dk

And if E D Vk we have (Proposition 4.5.4)

k .M / D max R.'/ D R .ek /

'2Vk
'¤0

so we deduce that:
k .M /  min max R.'/: 
E H ? .M / '2E
dim.E /Dk
 4.5 The minimax principle and applications 95

4.5.3 Properties of the first eigenvalue

An obvious consequence of the minimax principle is the following result.

Proposition 4.5.6. Let .M; g/ be a compact Riemannian manifold.

(i) In the cases of a closed manifold or of Neumann boundary conditions the
first eigenvalue is
1 .M / D 0
and the corresponding eigenfunction is the constant function M 3 x 7! 1.
(ii) In the case of the Dirichlet boundary condition the first eigenvalue satisfies

1 .M / > 0:

Proof.
(i) By the spectral theorem, 1 .M /  0. With the minimax principle we also
get Z
jr'j2 d Vg
1 .M / D min MZ :
'2C 1 .M /
'¤0 ' 2
d V g
M

Since the constant function M 3 x 7! 1 belongs to the space H 1 .M / and

since Z
jr1j2 d Vg
MZ D0
1 d Vg
2
M
we deduce that 1 .M /  0, therefore 1 .M / D 0. Moreover, g .1/ D
0.
(ii) We consider the Dirichlet problem and by contradiction assume that
1 .M / D 0, there exists a non trivial function u such that g u D 0
and u vanishes on the boundary of M . Therefore,
Z
g u u d Vg D 0
M

hence integrating by parts (here u vanishes on the boundary of M / we get

Z
jruj2 d Vg D 0:
M

Thus u is constant on M; and from the boundary conditions it follows that

u D 0 on M , which is a contradiction. 
96 Chapter 4. Spectrum of the Laplacian on a compact manifold

Remark 4.5.7. In fact we have already seen in the proof of the spectral theorem
4.3.1 (Section 4.4.3: The Dirichlet eigenvalue problem) that the eigenvalues of the
Laplacian for the Dirichlet problem satisfy k .M / WD 1k , where k > 0, thus
1 .M / > 0.
More generally, we have also:

Proposition 4.5.8. For each spectral problem above we have:

(i) The first eigenfunction e1 of the manifold .M; g) satisfy e1 > 0 or e1 < 0
in the interior of M .
(ii) The first eigenvalue 1 .M / is always simple.

Proof.
(i) Suppose that the function e1 changes sign in M . Since e1 2 H 1 .M /, the
function f WD je1 j belongs to H 1 .M / and jdf j D jde1 j (see for example
[Gi-Tr]), hence
R.f / D R .e1 / D 1 .M /:
Therefore the function f is also a first eigenfunction of .M; g/ which sat-
isfies f  0 on M . Moreover, f vanishes in the interior of M and

g f D 1 .M /f  0

on M . By the maximum principle [Pr-We], the function f cannot attain its

minimum in an interior point of the manifold M , hence f does not vanish
on M , which is a contradiction.
(ii) Suppose that the first eigenvalue 1 .M / is not simple: there exists two
orthogonal non-trivial eigenfunctions f1 ; '1 associated to 1 .M /. Using
the assertion (i) we may assume that f1 ; '1 > 0 in the interior of M , thus
Z
hf1 ; '1 iL2 D f1 '1 d Vg > 0I
M

but this is absurd because f1 ; '1 are orthogonal. 

Note that the assertions (i) and (ii) hold for a large class of operators.

Proposition 4.5.9. In the cases of a closed manifold or the Neumann problems

we have for any integer k  2
Z
ek d Vg D 0
M

where ek is an eigenvector of the k-th eigenvalue k .M /.

 4.5 The minimax principle and applications 97

Proof. In the cases of a closed manifold or the Neumann problem the first eigen-
value 1 .M / is equal to zero and simple. Thus for all k  2 we have

k .M /  2 .M / > 0

and by definition g ek D k .M /ek . Therefore,

Z Z
g ek d Vg D k .M / ek d Vg
M M

and integrating by parts2 we get

Z Z Z
g ek d Vg D g ek :1 d Vg D rek :r1 d Vg D 0
M M M

so for all k  2 Z
0 D k .M / ek d Vg :
M

Next, since for all k  2 we have k .M / > 0 we deduce that for all k  2
Z
ek d Vg D 0. 
M

4.5.4 Monotonicity domain principle

Another obvious classical application of minimax property is the monotonicity
principle for Dirichlet problem:

Proposition 4.5.10. Let .M; g/ be a Riemannian manifold and A; B two subman-

ifolds of .M; g/ of same dimension n such that A  B and A; B are compact with
boundary. Then for every integer k  1 we have:

k .B/  k .A/:

Proof. For every integer k  1 and for any i 2 f1; : : : ; kg, let fk 2 H01 .A/ be
an eigenfunction corresponding to the eigenvalue i .A/. This eigenfunction may
be extended by 0 on B, and this extension belong to the space H01 .B/. Hence we
have Z Z
jr'j d Vg
2
jr'j2 d Vg
min max M Z  min max M Z : 
E H01 .B/ '2E E H01 .A/ '2E
dim.E /Dk '¤0
' 2
d V g dim.E /Dk '¤0
' 2
d V g
M M
2 In the closed manifold case this is clear and in the Neumann case one uses the fact that the gradient of
the eigenfunction u vanishes on the boundary of M .
98 Chapter 4. Spectrum of the Laplacian on a compact manifold

Let us finish this section by introducing a notation. For each spectral problem
(closed, Neumann and Dirichlet) the spectrum of g on .M; g/ can be rewritten
as
Spec.M; g/ D f0  1 .M / < 2 .M /      k .M /     g
or simply
Spec.M; g/ D f0  1 < 2      k     g
In the specific cases of closed manifold or Neumann problems we have

Spec.M; g/ D f0 D 1 < 2      k     g ;

whereas for the Dirichlet problem

Spec.M; g/ D f0 < 1 < 2      k     g

The notation 0  1 < 2      k     does not indicate the multiplicity of

eigenvalues. In certain situations it is convenient to use the following alternative
Notation 4.5.11. Let us denote the distinct eigenvalues of .M; g/ by

0  e1 ; m1 D 1 < e2 ; m2 <    < 

fk ; mk <    ;

where the integer mi denote the multiplicity of the eigenvalue ei .

4.5.5 A perturbation of metric result

We present here a very simple result concerning a metric perturbation on the man-
ifold .M; g/:

Theorem 4.5.12. Let M be a compact manifold equipped with a family of Rie-

mannian metrics .gn /n . If the sequence of metrics .gn /n converge as n ! C1
(in the Whitney topology) to a Riemannian metric g0 , then for every problem
.Dirichlet, Neumann, or closed manifold/, and for any fixed integer k, the eigen-
values k .M; gn / converge as n ! C1 to k .M; g0 /.

The proof is a direct consequence of the following lemma.

Lemma 4.5.13. Let g and g0 be two metrics on a compact Riemannian manifold

such that ˛g0  g  ˇg0 . Then for every problem .Dirichlet, Neumann, or closed
manifold/ and for any integer k  1 we have
n n
˛2 ˇ2
n k .M; g0 /  k .M; g/  n k .M; g0 /:
ˇ 2 C1 ˛ 2 C1
 4.5 The minimax principle and applications 99

Proof. Using the definition of the volume form Vg , we have

n n
˛ 2 V g0  V g  ˇ 2 V g0 :
On the other hand, using the definition of the gradient we also have
1 ˇˇ ˇ2 ˇ ˇ2 1ˇ ˇ2
rg0 ' ˇ  ˇrg ' ˇ  ˇrg0 ' ˇ :
ˇ ˛
Therefore for any function ' 2 H 1 .M / we get
Z Z Z
ˇ ˇ2 n ˇ ˇ
jr'j d Vg
2 1 ˇ ˇ
rg0 ' d Vg ˇ2 ˇrg ' ˇ2 d Vg
˛ ˛ 0 0
MZ  MZ  MZ
n
' 2 d Vg ' 2 d Vg ˛2 ' 2 d V g0
M M
Z M
ˇ ˇ2
n ˇr ' ˇ d Vg
ˇ 2 M g0 0
D n C1 Z :
˛2 ' 2 d V g0
M

Consequently for any integer k  1 we obtain:

Z Z
ˇ ˇ
2
jr'j d Vg n ˇrg ' ˇ2 d Vg
0 0
ˇ2
min max MZ  n min max MZ
E H ? .M / '2E
' 2 d Vg ˛ 2 C1 E H ? .M / '2E ' 2 d V g0
dim.E /Dk '¤0 dim.E /Dk '¤0
M M

where H .M / WD
?
in the case of the Dirichlet problem and H ? .M / D
H01 .M /
1
H .M / in the case of the Neumann or closed manifold problems. Hence we get
for any integer k  1
n
ˇ2
k .M; g/  n k .M; g0 /:
˛ 2 C1
Obviously we also have
Z Z Z
ˇ ˇ n ˇ ˇ
jr'j2 d Vg 1 ˇrg ' ˇ2 d Vg ˛2 ˇrg ' ˇ2 d Vg
ˇ 0 ˇ 0 0
Z
M
 Z
M
 MZ
n
' 2 d Vg ' 2 d Vg ˇ2 ' 2 d V g0
M
Z
M M
ˇ ˇ
n ˇrg ' ˇ2 d Vg
0 0
˛2 MZ
D n :
ˇ 2 C1 ' 2 d V g0
M
100 Chapter 4. Spectrum of the Laplacian on a compact manifold

It follows that for any integer k  1

Z Z
ˇ ˇ
n jr'j 2
d V g
ˇrg ' ˇ2 d Vg
0
˛2
n min max M Z  min max MZ ;
ˇ 2 C1 E H ? .M / '2E ' 2
d V
E H ? .M / '2E
' 2
d V
dim.E /Dk '¤0 g0 dim.E /Dk '¤0 g
M M

i.e., for any integer k  1

n
˛2
n k .M; g0 /  k .M; g/: 
ˇ 2 C1

4.6 Complements: The Schrödinger operator and the

Hodge–de Rham Laplacian
4.6.1 The Schrödinger operator
In this section .M; g/ is a complete connected Riemannian manifold of dimension
n  1, h denotes the Planck constant, and V is a potential on M : a function
V W M ! R.

Definition and basic properties

Definition 4.6.1. The Schrödinger operator on .M; g/ with potential V is the
linear unbounded operator

h2
H WD g C V
2
h2
with domain D .M /, in the sense that H ' D 2
g C V '.

The Schrödinger operator plays a central role in Quantum Mechanics; in par-

ticular, we use it to determine the quantum dynamics of a particle via the famous
Schrödinger equation: see Sections 2.6.3 and 4.1.3. Let us give here some exam-
ples of potentials:
 Free motion potential V D 0.
 The double slit experiment: V ! C1 on the plate, and V D 0 elsewhere.
 The hydrogen atom potential: V D  kxk
k
.
kxk2
 The harmonic oscillator: V D 2
.
 4.6 Complements: The Schrödinger operator and the Hodge–de Rham Laplacian 101

If V D 0 the operator H is just the Laplace–Beltrami operator, and we have seen

in this case that the spectrum is discrete. In the general case the main questions
about spectrum for Schrödinger operators are:
1. Is the operator H self-adjoint (or essentially self-adjoint)?
2. Is the spectrum of H discrete?

Self-adjointness
Starting with the first question, recall that an unbounded linear operator H is said
to be essentially self-adjoint if its closure H is selfadjoint. In the case of M D Rn
with standard metric, Carleman [Car] showed that if the function V is locally
bounded and if there exists C such that V  C on M , then the Schrödinger op-
erator H is essentially self-adjoint. Later T. Kato [Kat1] proved that it is possible
to replace the hypothesis V 2 L1 loc .M / by V 2 Lloc .M /. Next, in the works of
2

I.M. Oleinik [Ole1], [Ole2], [Ole3] we can find a general theorem with complex
hypotheses on V . A very useful consequence of Oleinik’s result is:

Theorem 4.6.2. Let .M; g/ be a complete connected Riemannian manifold of

dimension n  1 and let V 2 L1 loc .M / be a potential such that for all x 2
M; V .x/  C , where C is a real constant. Then the operator H D g C V is
essentially self-adjoint.

Discreteness of the spectrum

As for the Laplace–Beltrami operator, if the manifold .M; g/ is compact, then
the spectrum of the Schrödinger operator H D g C V is discrete. In the
non-compact case, the spectrum of g is not discrete. Nevertheless, if V is a
confining potential, then the spectrum of H D g C V is discrete.

Compact setting
Here we suppose that the manifold .M; g/ is compact. The spectral problem as-
sociated to the operator H is: find all pairs .; u/ with  2 R and u 2 L2 .M /
such that
g u C V u D u:
In the case of manifolds with boundary, we need boundary conditions on the func-
tions u (Dirichlet, Neumann, or closed manifold conditions).
Similarly to the Laplace–Beltrami operator case (Theorem 4.3.1), we have

Theorem 4.6.3. Consider a compact connected Riemannian manifold .M; g/ of

dimension n  1, and a potential V 2 L1
loc .M /. For the Dirichlet, Neumann, or
102 Chapter 4. Spectrum of the Laplacian on a compact manifold

closed manifold problems, the spectrum of H D g CV is discrete. Specifically,

it consists of an infinite increasing sequence of eigenvalues with finite multiplicity,

inf V .x/  1  2      k     C 1:
x2M

Moreover, the associated eigenfunctions .ek /k0 form a Hilbert basis of the space
L2 .M /.

Non-compact setting
Now suppose that the manifold .M; g/ is not compact.

Definition 4.6.4. Let .M; g/ be a smooth manifold and V W M ! R. We said that

the function V is confining if limjxj!1 V .x/ D C1, in the sense that

8A > 0; 9K  M; 8x 2 M n K; jf .x/j  A:

A general theorem concerning the spectrum of the Schrödinger operator on

a manifold (with bounded geometry) is due to Kondrat’ev and Shubin [Ko-Sh1],
[Ko-Sh2]:

Theorem 4.6.5. Let .M; g/ be a complete connected Riemannian manifold of

dimension n  1, and let V 2 L1 loc .M / be a potential such that limjxj!1 V .x/ D
C1. Then the spectrum of H D g C V is discrete. Specifically, it consists of
an infinite increasing sequence of eigenvalues with finite multiplicity,

inf V .x/  1  2      k     C 1:
x2M

Moreover, the associated eigenfunctions .ek /k0 form a Hilbert basis of the space
L2 .M /.

For an example of spectrum computation see Section 5.1.4.

4.6.2 The Hodge–de Rham Laplacian

Hodge theory links harmonic differential forms to cohomology, thus providing
a bridge between algebraic topology and differential geometry. For details on the
theory of Hodge–de-Rham Laplacian see the famous article of J. Dodziuk [Dod2],
see also for example [Tay] or [War].
In this subsection let us consider .M; g/ a closed Riemannian manifold with
dimension n  1.
 4.6 Complements: The Schrödinger operator and the Hodge–de Rham Laplacian 103

The de Rham chain complex

Using the exterior derivative operator d (see Section 3.1.1) we construct the so-
called de Rham chain complex:
0 .M / ! 1 .M / !    ! n .M / ! 0
which is locally exact (by the Poincaré lemma).
A differential form 2 p .M / is called co-closed if ı D 0, and 2 p .M /
is called co-exact if there exists ˇ 2 pC1 .M / such that D ıˇ. Note that
˚ 
d p1 .M / D im d p1 WD 2 p .M /I 9˛ 2 p1 .M /; D d˛
is the vector space of exact differential forms on p .M /,
˚ 
ıpC1 .M / D im ıpC1 WD 2 p .M /I 9 2 pC1 .M /; Dı
is the vector space of co-exact differential forms on p .M /,
ker d p WD f 2 p .M /I d D 0g
is the vector space of closed differential forms on p .M /.

Some definitions
Definition 4.6.6. The Hodge–de Rham Laplacian is the second-order elliptic op-
C1
[
erator acting on the set of differential forms  .M / WD k .M / by the for-
kD0
mula
 WD d ı C ıd;
where d is the exterior derivative operator and ı is the codifferential operator, i.e.,
the formal L2 -adjoint of d on  .M /.

For a fixed integer p  0, the pth -Hodge–de Rham Laplacian

p WD jp .M /
is the restriction of  to p .M /. Note that the operator ı sends functions to zero,
therefore the 0th -Hodge–de Rham operator is the classical Laplacian on functions.
A form 2 p .M / is called harmonic if p D 0; using an integration by parts
formula for differential forms one can show that a differential form 2 p .M /
is harmonic if and only if is closed and co-closed.
Since d 2 D d ı d D 0, the set im d p1 is a subspace of ker d p . One of
the main principal objects in Hodge theory is the pth -de Rham cohomology group,
defined as the quotient of vector spaces
p
HDR .M / WD ker d p =im d p1 :
104 Chapter 4. Spectrum of the Laplacian on a compact manifold

Spectral theory of the Hodge–de Rham Laplacian

For any p  0, the pth -Hodge–de Rham Laplacian has the same spectral proper-
ties as the classical Laplacian on functions: this operator is self-adjoint, positive,
its spectrum and point spectrum coincide and consist of an infinite sequence of
real eigenvalues (with finite multiplicity)

p;1 .M /  p;2 .M /      p;k .M /    

such that p;k .M / ! C1 as k ! C1.

Hodge decomposition and applications

In Hodge theory we have the following main result

Theorem 4.6.7 (Hodge decomposition). For any fixed integer 0  p  n,

p .M / D im d p1 ˚ ker p ˚ im ıpC1 ;

i.e., for any 2 p .M / there exists a unique triplet .˛; ˇ; / 2 p .M /

p1 .M / pC1 .M / such that

D ˛ C dˇ C ı

and p ˛ D 0.

Remark 4.6.8. In the case p D 0, for any function 2 C 1 .M / there exists

a unique couple .˛; / with a function ˛ 2 C 1 .M / such that g ˛ D 0 and
˛ 2 1 .M /, such that D ˛ C ı .
Since  commutes with the operators d and ı, the previous decomposition is
invariant under the action of p . Moreover, since d W p .M / ! pC1 .M / and
ıW pC1 .M / ! p .M /, if we denote by

p;1 .M /  p;2 .M /      p;k .M /    

the spectrum of the Hodge–de Rham Laplacian restricted to ıpC1 .M /, then we

can write the spectrum of the Hodge–de Rham Laplacian restricted to d p1 .M /:

p1;1 .M /  p1;2 .M /      p1;k .M /     ;

and we have a natural vector space isomorphism between the associated eigen-
spaces.
One of the most important result of Hodge theory is:

Theorem 4.6.9. The kernel of p is isomorphic to the pth -de Rham cohomology
p
group HDR .M /.
 4.6 Complements: The Schrödinger operator and the Hodge–de Rham Laplacian 105

Proof. Let be a closed form on p .M /: d D 0. Thanks to the Hodge decom-

position, there exists a unique triplet .a; b; c/ 2 .p .M //3 with p a D 0, and
there exists .ˇ; / 2 p1 .M / pC1 .M / such that:

b D dˇ; c D ı

and
D a C b C c:
Then we have
d D da C db C dc;
i.e.,
0 D dc
(da D 0 because a is harmonic, and hence closed; and db D d 2 ˇ D 0). On the
other hand, we have also ıc D ı2 D 0. Consequently the differential form c is
harmonic. Using the unicity of the decomposition, we get c D 0 and finally we
have D a C b. In other words, we obtained the equality

ker d p D im d p1 ˚ ker p :

Therefore
p    
HDR .M / D ker d p =im d p1 D im d p1 ˚ ker p = im d p1 ˚ 0
' ker p : 

In algebraic topology the pth Betti number is defined as

p
bp .M / WD dim HDR .M /

(these numbers are topological invariants). bp .M / is also the dimension of the

kernel of p . An example of application is the following result (see Theorem 4.7
in [GHL]):

Theorem 4.6.10. Let .M; g/ be a compact Riemannian manifold of dimension

n  1. Then:
p
(i) If the Ricci curvature of M is strictly positive, HDR .M / D f0g.
p
(ii) If the Ricci curvature is non-negative, 0  dim HDR .M /  n.
p
(iii) If the Ricci curvature is non-negative and dim HDR .M / D n, the manifold
.M; g/ is isometric to a flat torus .of dimension n/.
Chapter 5
Direct problems in spectral geometry

As we have seen in the Introduction, there are two main types of problems in
spectral geometry: direct problems and inverse problems. This chapter is devoted
to direct problems. The general statement of a direct problem is: given a Rie-
mannian manifold .M; g/, can we compute or find some properties of the spec-
trum .k .M //k0 of .M; g/? We start this chapter with some examples of exact
computation for compact manifolds, including an example of computation with
a Schrödinger operator. We then present some classical examples of inequalities
and asymptotic results concerning eigenvalues.

5.1 Explicit calculation of the spectrum

Unfortunately, the calculation of the spectrum is very difficult and in fact we have
only a few examples where it can be done explicitly. Let us give some standard
examples.

5.1.1 Flat tori

We begin with the one-dimensional case with the canonical metric: using the
theory of Fourier series on the one-dimensional torus TL of length L > 0, the
d2
spectrum of the Laplacian  dx 2 with closed manifold conditions is

4 2 n2
Spec.TL ; can/ D ;n2Z
L2
with associated eigenfunctions:
2i  nx
x 7! en .x/ D e L ; n 2 Z:
2
Indeed, if the scalar number  is in the spectrum of  dx
d
2 , then there exists a non-

trivial function u 2 L .TL / such that

2

u00 D u:
It is a well known fact that (by the theory of L2 Fourier series) the sequence of
functions
2i  nx
x 7! en .x/ D e L
n2Z
108 Chapter 5. Direct problems in spectral geometry

form a Hilbert basis of L2 .TL /, hence there exists an unique sequence .an /n2Z 2
CZ such that X
uD an en :
n2Z

Hence, we can rewrite the differential equation u00  u D 0 as

X  4 2 n2 
an 2
  en D 0:
n2Z
L

Since u ¤ 0, there exists n0 2 Z such that an0 ¤ 0, hence because .en /n2Z is
a basis we get
4 2 n20
u00  u D 0 H)  D :
L2
n 2 2 o
Thus we have the inclusion Spec.TL ; can/  4L2n ; n 2 Z .
2 4 2 n2
Conversely, it is clear that for any integer n 2 Z we have  dx
d
2 en D L2
en ,
so indeed
4 2 n2
Spec.TL ; can/ D ;n2Z :
L2
More generally, let a WD .a1 ; a2 ; : : : ; an / 2 Rn be such that for all i , ai ¤ 0 and
consider the lattice  WD a1 Z C a2 Z C    C an Z in Rn of rank n  1. The
associated flat torus is
Tna WD Rn = :
Let us also denote by  ? the dual lattice of :

 ? WD fx 2 Rn I 8y 2 ; hx; yi 2 Zg

Since .a1 ; a2 ; : : : ; an / is a Z-basis of , the dual lattice  ? is given by

 T
 ? D A1 Zn ;

where 0 1
a1 0  0
B 0 a2    0 C
A WD B
@ ::: :: : : :: C
: : : A
0 0  an

thus a11 ; a12 ; : : : ; a1n is Z-basis of  ? , and it follows that the volume of  ? is
given by
1
Vol. ? / D :
Vol./
 5.1 Explicit calculation of the spectrum 109

Moreover, it is easy to show that

 ?? D 

but unfortunately we do not have  ? D .

Now, using the theory of the multidimensional Fourier series, the sequence of
functions
 2i k1 x1

C C 2i kn xn
x 7! e 2i h!;xi
?
D .x1 ; : : : ; xn / 7! e a1 an
!2 .k1 ;:::;kn /2Zn
 
form a Hilbert basis of L2 Tna , so the same argument as in dimension one shows
that
 2 
2 k1 kn2
Spec.Ta ; can/ D 4
n
C    C 2 ; .k1 ; : : : ; kn / 2 Zn :
a12 an

In other words, the spectrum of the Laplacian (in the closed situation) on a flat
torus  n Rn is given by:
˚ 
Spec. n Rn ; can/ D 4 2 k  kRn ;  2   :

5.1.2 Rectangular domains with boundary conditions

Consider the Dirichlet problem on a rectangular domain Œ0; a Œ0; b in R2 with
a; b > 0. Using the same argument as for the flat torus and the fact that the family
n1 x n2 y
.x; y/ 7! en1 ;n2 .x; y/ WD sin sin ? 2
a b .n1 ;n2 /2.N /
˚ 
form a Hilbert basis of H WD f 2 L2 .Œ0; a Œ0; b/ I f .a/ D f .b/ D 0 , it is
easy to see that
 2 
n1 n22 ? 2
SpecDirichlet .Œ0; a Œ0; b; can/ D  2 2
C 2
; .n1 ; n2 / 2 N ;
a b

? 2
with associated eigenfunctions en1 ;n2 ; .n1 ; n2 / 2 N .

5.1.3 Spheres
The one-dimensional case is trivial: since S1 ' T11 , we have
˚ 
Spec.S1 ; can/ D 4 2 n2 ; n 2 Z :
110 Chapter 5. Direct problems in spectral geometry

Consider the n-dimensional sphere Sn in RnC1 . Using spherical coordinates

.r;  / 2 R?C Sn , the Euclidean Laplacian may be expressed via the spherical
Laplacian as
@2 n @ 1
RnC1 D 2 C C 2 :
@r r @r r
In fact, the homogeneous harmonic polynomials restricted to Sn are eigenfunc-
tions of Sn . Indeed, let P be a degree k homogeneous harmonic polynomial on
RnC1 , i.e., P D r k P jSn and RnC1 P D 0. Using the previous expression for the
Euclidean Laplacian we get

0 D RnC1 P
D k.k  1/r k2 P jSn C nkr k2 P jSn  r k2  .P jSn / ;

whence
 .P jSn / D k.n C k  1/P jSn :
Now, since the space of homogeneous harmonic polynomials restricted to Sn is
dense in L2 .Sn / (see [GHL]), the homogeneous harmonic polynomials restricted
to Sn are the only eigenfunctions of Sn . Consequently,

Spec.Sn ; can/ D fk.k C n  1/; k 2 Ng

   
nCk nCk1
and the multiplicity of k.k C n  1/ is equal to  .
k k1

5.1.4 Harmonic oscillator

This is a very famous example of explicit computation in Quantum Mechanics.
The one-dimensional harmonic oscillator is the Schrödinger operator

d2
 C x2
dx 2

on the manifold R. Consider the set Y WD ff 2 H 1 .R/; xf 2 L2 .R/g. This set

a subspace of H 1 .R/, and equipped with the scalar product
˝ ˛
hf; giY WD hf; giL2 C f 0 ; g 0 L2 C hxf; xgiL2

Y is a Hilbert space and is dense in L2 .R/. Moreover, by a classical argument of

functional analysis, the canonical inclusion of Y in L2 .R/ is compact.
Let us introduce the annihilation operator AW D.A/ ! L2 .R/ with domain
D.A/ D Y , defined by
Af WD f 0 C xf:
 5.1 Explicit calculation of the spectrum 111

Using the integration by parts formula (and the fact that if f 2 Y then
limjxj!C1 xjf .x/j2 D 0) we get for all f 2 Y
 2
kAf k2L2 D  kf k2L2 C f 0 L2 C kxf k2L2 ;

for all f 2 Y , thus we obtain

2 kf k2L2 C kAf k2L2 D kf k2Y :

Consequently, the norms kkY and kkA are equivalent on Y . Therefore the set
Y D D.A/ is complete for the norm kkA . It follows that the operator AW D.A/ !
L2 .R/ is closed. The adjoint of A is the operator A? W D.A?/ ! L2 .R/ with
domain D.A? / D Y and defined by

A? f D f 0 C xf:

This operator is called the creation operator. Now we consider the operator

H WD A? A C I

with domain D.H / WD ff 2 Y; Af 2 Y g. In fact we have

˚ 
D.H / D f 2 H 2 .R/; x 2 f 2 L2 .R/

and for all f 2 D.H /

A? Af D  .Af /0 C xA.f / D f 00 C x 2 f  f:

Hence,
Hf D f 00 C x 2 f
for all f 2 D.H /. In particular, for all f 2 D.H / such that kf k2L2 D 1 we have
˝ ˛
hHf; f iL2 D A? Af; f L2 C kf k2L2 D kAf k2L2 C 1  1;

therefore the closure of the numerical range ‚.H / is a subset of Œ1; C1Œ. We see
that the set C  ‚.H / has just one connected component and the map
8
ˆ
< C ! N [ fC1g
dW
:̂  7! dim ker H ?  I

is constant on C  ‚.H /, i.e., for all  2 C  ‚.H /,

d./ D d.i / D d.i /

112 Chapter 5. Direct problems in spectral geometry

(because i; i 2 C  ‚.H /). But since H is self-adjoint

ker H ?  I D f0g

thus d.i / D d.i / D 0, and so d./ D 0 for all  2 C  ‚.H /. In other words,
C  ‚.H /  .H / so we get Spec.H /  .H /  ‚.H / because the spectrum
of H is discrete. We conclude that

Spec.H /  Œ1; C1Œ:

By a simple computation
 
A; A? f D AA? f  A? Af D 2f

for all f 2 D.H /. Next, if a vector ' 2 D.H / satisfies H ' D ' (where  is
a scalar), then A' 2 D.H /; A? ' 2 D.H /, and we have

H.A'/ D A? AA' C A' D AA? A'  2A' C A'

D A .H '  '/  A' D A .'  '/  A' D .  2/A':

On the other hand,

˝ ˛
kA'k2L2 D hA'; A'iL2 D A? A'; ' L2
D hH '  '; 'iL2 D  k'k2L2  k'k2L2 D .  1/ k'k2L2 :

Similarly,
 ? 2
H.A? '/ D . C 2/A? '; A '  2 D . C 1/ k'k2 2 :
L L

Now we want to show that Spec.H /  f.2n C 1/; n 2 Ng. For the moment we
only know that Spec.H /  Œ1; C1Œ. If  is an eigenvalue of H in the interval
1; 3Œ, there exist ' ¤ 0 in D.H / such that H ' D ', thus we have

H.A'/ D .  2/A'

and
kA'k2L2 D .  1/ k'k2L2 > 0
hence A' ¤ 0. Consequently,   2 2 Spec.H / therefore   2 2  1; 1Œ, which
is a contradiction. So
Spec.H /  f1g [ Œ3; C1Œ:
Using the same argument by induction we get easily that

Spec.H /  f.2n C 1/; n 2 Ng ;

 5.2 Qualitative properties of the spectrum 113

as claimed. To finish, using the Hermite family fen gn2N

p 1 x2 2 d
n
2
en .x/ D .2n nŠ / 2 e 2 Hn .x/ with Hn .x/ D .1/nex n
.ex /
dx
(which form a Hilbert basis of the space L2 .R/: for a proof of this fact see, for ex-
ample [Tay] or the notes [Ev-Zw], and for a different method using the Bargmann
transform see the original article of V. Bargmann [Bar] or also in [Tay]).
We have for any n p
A? en D 2n C 2enC1
and p
Aen D 2nen1 ;
so for any n
Hen D .2n C 1/en :
Consequently the spectrum of the operator H is

Spec.H / D f.2n C 1/; n 2 Ng

and the associated eigenvectors are given the Hermite’s family.

5.2 Qualitative properties of the spectrum

When it is impossible to compute the spectrum, we try to describe some of its
properties, and two kinds of questions arise:
 Determine or estimate the bottom of the spectrum, i.e., look at the first
eigenvalues 1 .M / and/or 2 .M /.
 Determine the asymptotics of large eigenvalues (the semi-classical para-
digm).

5.2.1 From the bottom of the spectrum . . .

We give here some examples concerning the bottom of the spectrum. For example,
the first nonzero eigenvalue of a compact manifold is very important. We begin
with the Lichnerowicz–Obata theorem:

Theorem 5.2.1 (Lichnerowicz–Obata Theorem). Let .M; g/ be a closed Rieman-

nian manifold of dimension n such that

Ric  .n  1/ > 0

114 Chapter 5. Direct problems in spectral geometry

(i.e., for any  2 TM we have Ric.; /  .n  1/ kk2 > 0). Then the first
nonzero eigenvalue ? of g satisfies

?  n:

Moreover, we have equality if and only if M is isometric to an Euclidean sphere

of dimension n.

Proof. The equality case is due to Obata. Here we just give a proof for the in-
equality. The proof is based on the Bochner–Lichnerowicz formula (see for exam-
ple [BGM]): for any function ' 2 C 1 .M / it holds that

1  
g jrf j2 D jjjHess f jjj2 C rf:rg f C Ric .rf; rf / ;
2
where Hess f is the Hessian matrix of the function f and jjjjjj is the matrix–
operator norm.
Now, let ? be an nonzero eigenvalue of the operator g : there exists a non-
trivial function u such that g u D ? u; and we have (see Proposition 4.5.3)
Z
jruj2 d Vg
? D Z
> 0:
u d Vg
2


Under the hypotheses of the theorem, the Bochner–Lichnerowicz formula yields

1  
g jruj2 D jjjHess ujjj2 C ru  rg u C Ric .ru; ru/
2
D jjjHess ujjj2  ? ru  ru C Ric .ru; ru/
 jjjHess ujjj2  ? jruj2 C .n  1/jruj2 :

By a classical result of the theory of finite-dimensional vector space (here Hess u

is a n n-matrix),
tr2 .Hess u/
 jjjHess ujjj2 ;
n
so in our context we get
 2
1   g u
g jruj 
2
 ? jruj2 C .n  1/jruj2
2 n
2 u2
D ? C ..n  1/  ? / jruj2 :
n
 5.2 Qualitative properties of the spectrum 115

Upon integrating over the manifold .M; g/ we obtain

Z Z
1   2? u2
g jruj2 d Vg  C ..n  1/  ? / jruj2 d Vg :
2 M M n
Recall that the manifold .M; g/ has no boundary, so using the integration by parts
formula we obtain Z
 
g jruj2 d Vg D 0;
M
whence Z
1 2
0 ? kuk2L2 .M / C ..n  1/  ? / jruj2 d Vg :
n M
Since u is not trivial we further have
Z
jruj2 d Vg
1 2
0  ? C ..n  1/  ? / Z M
;
n
u d Vg
2
M

hence
1 2
0  C ..n  1/  ? / ?
n ?
1
D 2? C  n?  ?  2? :
n
To finish observe that
? .n  1/ 1
. n  ? / D 2? C  n?  ?  2? ;
n n
and so
? .n  1/
. n  ? /  0;
n
hence  n  ?  0. 
Remark 5.2.2. There exist analogous versions of this theorem for the case of com-
pact manifolds with boundary (Dirichlet and Neumann problems).
We present here the Cheeger inequality. Let .M; g/ be a compact connected
Riemannian manifold. For every domain D (D is a bounded regular subset) of
M , consider the quantity
Vol.@D; g/
h.D; g/ D ;
Vol.D; g/
where @D denotes the boundary of D (thus Vol.@D; g/ is the .n  1/-dimensional
volume).
116 Chapter 5. Direct problems in spectral geometry

Definition 5.2.3. The Cheeger’s constant on .M; g/ is the number

h.M; g/ D inf h.D; g/;
D2X

where X is the set of domains D  M such that Vol.D/  Vol.M;g/

2
.
Using the co-area formula we obtain:
Theorem 5.2.4 (Cheeger’s inequality). For each of the spectral problem .Dirichlet,
Neumann or closed/ the first nonzero eigenvalue ? of g satisfies
h.M; g/2
?  :
4
Another remarkable result is the Li–Yau Theorem: for a closed Riemannian
manifold the first nonzero eigenvalue of g is bounded (up to a constant) from
below by the inverse of the diameter of the manifold
Theorem 5.2.5 (Li–Yau Theorem). Let .M; g/ be a closed Riemannian manifold
of dimension n with nonnegative Ricci curvature. Then the first nonzero eigen-
value ? of g satisfies
2
? 
4diam2 .M; g/
˚ 
where diam.M; g/ WD sup d.x; y/; .x; y/ 2 M 2 .
Let us also gives examples of results about comparison between the first eigen-
values of two different domains.
Theorem 5.2.6 (Faber–Krahn Theorem, 1953). Let M be a bounded domain of
Rn , and let us denote by 1 .M / the first non-null eigenvalue of  with Dirichlet
conditions. We have the inequality
1 .M /  1 .BM /;
where BM is the Euclidean ball with volume equal to Vol.M /. Moreover, equality
holds if and only if M is isometric to BM .
In the same spirit, we have also
Theorem 5.2.7 (Szegö–Weinberger Theorem, 1954). Let M a bounded domain of
Rn and let us denote by 1 .M / the first non-null eigenvalue of  with Neumann
boundary conditions. We have the inequality
1 .M /  1 .BM /;
where BM is the Euclidean ball with volume equal to Vol.M /. Moreover, equality
holds if and only if M is isometric to BM .
 5.2 Qualitative properties of the spectrum 117

5.2.2 . . . to the large eigenvalues: the Weyl formula

The most popular result concerning large eigenvalues is the Weyl asymptotic for-
mula. For the Laplacian in a rectangular domain  of R2 with Dirichlet boundary
conditions, the physicist P. Debye arrived at the formula
Vol./
N ./  ;
!C1 4
where N ./ WD Card .fk 2 N; k  g/ and Vol./ is the area of the domain
. In 1911, H. Weyl gave a proof of this conjecture. In the next subsection we
explain the proof in the case of flat tori.

5.2.3 Weyl formula for a flat torus

Consider the flat torus of Section 5.1.1, i.e., Tna D Rn =  with  WD a1 Z C a2 Z C
   C an Z and a WD .a1 ; a2 ; : : : ; an / 2 Rn . We have seen therein that
 2 
2 k1 k12 kn2
Spec.Ta ; can/ D 4
n
C 2 C    C 2 ; .k1 ; k2 ; : : : ; kn / 2 Zn :
a12 a1 an
Here the eigenvalue counting function is
 
k12 k22 kn2 
N ./ D Card k D .k1 ; k2 ; : : : ; kn / I 2 C 2 C    C 2 
a1 a2 an 4 2
D  ? \ B;
p
where B WD BRn 0; 
2 denotes the closed ball of Rn centered in 0 and of
p
radius 
2
.For all ! WD .!1 ; !2 ; : : : ; !n / 2  ? let us denote by I! the following
compact set of Rn :
Yn  
1 1
I! WD !i  ; !i C
2a1i 2ai
i D1

and consider the union [

P WD I! :
!2 ? \B

Clearly the volume Vol.P / of P is equal to

1 1 1
Vol.P / D N ./  :
a1 a2 an
Now we claim that the inclusions
p ! p !
 
BRn 0;  R  P  BRn 0; CR
2 2
118 Chapter 5. Direct problems in spectral geometry

hold with  
 1 1 1 

R WD  ; ;:::;  :
2a1 2a2 2an Rn
Indeed, using the triangle inequality for all .x1 ; x2 ; : : : ; xn / 2 B, we have
 
 1 1 1 

k.x1 ; x2 ; : : : ; xn /kRn  R   x1 ˙ ; x2 ˙ ; : : : ; xn ˙ 
2a1 2a2 2an Rn
 k.x1 ; x2 ; : : : ; xn /kRn C R:

Consequently,
p !! p !!
 
Vol BRn 0; R  Vol.P /  Vol BRn 0; CR ;
2 2

with p !! p !n
  ˙ R2
Vol BRn 0; ˙R D Bn
2 2
n
where Bn WD 2
. n
is the volume of the closed unit ball in Rn .
2 C1/
Moreover, since
1 1 1 N ./
Vol.P / D N ./  D ;
a1 a2 an Vol.Tna /

we obtain
Vol.Tna /Bn n
N ./  2:
!C1 .2/n

5.2.4 The semi-classical paradigm

Passer de la mécanique de Newton à celle d’Einstein doit être un peu, pour le
mathématicien, commme de passer du bon vieux dialecte provençal à l’argot
parisien dernier cri. Par contre, passer à la mécanique quantique, j’imagine
c’est passer au chinois.
Alexandre Grothendieck [Grot], Récoltes et Semailles (1986).
The semi-classical analysis is a nice way to understand the relationship be-
tween classical Hamiltonian mechanics (symplectic1 geometry) and quantum me-
chanics (operator algebras). The principle of semi-classical analysis is to extend
the classical Hamiltonian formalism by a deformation, therefore we can see clas-
sical mechanics as a limit of quantum mechanics. For a complete introduction
1 A symplectic manifold .M; !/ is a smooth differentiable manifold of dimension m equipped with a

non-degenerate closed 2-form !.

 5.3 The spectral partition function ZM 119

to semi-classical analysis see for example [Ba-We], [Di-Sj], [Ev-Zw], [Gr-Sj],

[Mar], [Rob]. Here, we just explain the philosophy of semi-classical analysis on
a concrete example.
Let .M; g/ be a compact Riemaniann manifold, and consider a (semi-classical)
parameter2 h > 0 and an another real number E > 0. The equation

h2
 g ' D E'
2
admits an eigenvector 'k as solution if and only if we have the scalar relation

h2
 k D E:
2

Hence, if h ! 0C then k ! C1. So there exists a correspondence between

the semi-classical limit (h ! 0C ) and large eigenvalues of the Laplacian. The
asymptotics of large eigenvalues for the Laplace–Beltrami operator g on a Rie-
maniann manifold .M; g/, or more generally for a pseudo-differential operator
Ph , is linked to symplectic geometry, the phase space geometry. This is the same
correspondence between quantum mechanics (spectrum, operator algebra) and
classical mechanics (length of periodic geodesics, symplectic geometry). More
precisely, for a pseudo-differential operator Ph on L2 .M / with a principal sym-
1
 1p 2 C .T M /, there exist a link between the geometry of the foliation
?
bol
p ./ 2R and the spectrum of the operator Ph . Indeed, we have the famous
result
.Ph  I / uh D O.h1 /
then
MS.uh /  p1 ./;
where MS.uh /  T ? M denote the microsupport of the function uh 2 L2 .M /.

5.3 The spectral partition function ZM

Definition 5.3.1. Let .M; g/ be a compact Riemannian manifold. The series of
functions
C1
X
ZM .t/ WD et k
kD1

is called the spectral partition function of the manifold .M; g/.

2 For example, the Planck constant.
120 Chapter 5. Direct problems in spectral geometry

Using the Weyl asymptotics, the series converge uniformly on R?C (see Section
7.5.3). Using this uniform convergence it is clear that the function t 7! ZM .t/ is
continuous and non-increasing on R?C . We have also that
lim ZM .t/ D `;
t !C1

with ` D 0 if 0 2 Spec.M; g/ and ` D 1 otherwise.

With the notation (see 4.5.11)

0  e1 ; m1 D 1 < e2 ; m2 <    < 

fk ; mk <   

we have also the expression

C1
X
ZM .t/ D mk et ek :
kD1

The interest in the spectral partition function is justified by the following result

Proposition 5.3.2. The function ZM determine the spectrum of the manifold

.M; g/.

Proof. Suppose for example, that 0 2 Spec.M; g/, hence e1 D 0. For  > 0
consider the function
C1
X C1
X
et ZM .t/  et D mk e.ek /t  et D mk e.ek /t
kD1 kD2

If  < e2 , then

lim et ZM .t/  et D 0I
t !C1

if  D e2 , then
lim et ZM .t/  et D m2 I
t !C1

finally, if  > e2 , then

lim et ZM .t/  et D C1:
t !C1

Thus e2 is the unique real number  > 0 such that the function t 7! et ZM .t/ 
et has a finite limit as t tends to infinity. Consequently, the function ZM de-
termine the first non-null eigenvalue e2 . By induction, for any integer i  2
considering the expression
i 1
X C1
X
et ZM .t/  mj e.ej /t D mj e.ej /t
j D1 j Di
 5.4 Eigenvalues and eigenfunctions of surfaces 121

we get: if  < ei , then

i 1
X
lim et ZM .t/  mj e.ej /t D 0I
t !C1
j D1

if  D ei , then

i 1
X
lim et ZM .t/  mj e.ej /t D mi I
t !C1
j D1

finally, if  > ei , then

i 1
X
lim et ZM .t/  mj e.ej /t D C1:
t !C1
j D1

Hence, for any integer i the function ZM determines the eigenvalue ei .
The case 0 … Spec.M; g/ is similar. 

5.4 Eigenvalues and eigenfunctions of surfaces

5.4.1 Spectrum of surfaces
In the case of dimension two (surfaces) there exists important topological restric-
tions on the multiplicities of the eigenvalues (see also Section 7.4). Here let us
present quickly (without proofs) the main result concerning the relationship be-
tween the topology of a surface and the multiplicities of its eigenvalues. The
following theorem is due to S. Y. Cheng [Che], G. Besson [Bes], Y. Colin de
Verdière [Col4], N. Nadirashvili [Nad], and B. Sévennec [Sev].

Theorem 5.4.1. For a connected complete surface .M; g/ we have

(i) If X D S2 or R2 , then for any k  3 we have mk  2k  3.
(ii) If X D P2 .R/ .the projective plane/ or K2 .the Klein Bottle/, then for any
k  1 we have mk  2k C 1.
(iii) If X D T2 .the two-torus/, then for any k  1 we have mk  2k C 2.
(iv) If .M / < 0 .where .M / denotes the Euler characteristic of the surface/,
then for any k  1 we have mk  2k  2 .M /.
122 Chapter 5. Direct problems in spectral geometry

5.4.2 Eigenfunctions of surfaces and Courant’s nodal theorem

After studying eigenvalues, an important further question is to understand the be-
haviour of the eigenfunctions. Let .M; g/ be a compact Riemannian surface with
boundary. For every eigenfunction ' of g (with Dirichlet conditions) the nodal
set (or set of nodal lines) associated to ' is
N.'/ WD fx 2 M I '.x/ D 0g:
Nodal sets appear explicitly in experimental physics: for example on a vibrating
drum with dust, the dust accumulates along the nodal lines of the drum (this is the
Chladni’s acoustic patterns phenomenon, see for example in the classical [Co-Hi]
and in [Bé-He] for recent pictures and discussion on this topic).
The connected components of the set M  N.'/ are called the nodal domains
of the eigenfunction '. Here let us denote by .'/ the number of nodal domains
of '. We have the celebrated Courant’s nodal theorem (see the original article
[Coura] and also [Co-Hi], [Cha]):
Theorem 5.4.2 (Courant’s nodal Theorem). Let k  1. Then for every non-trivial
eigenfunction ' associated to the k-eigenvalue k .M / we have .'/  k.
Consequently, the first eigenfunction never vanishes on M and the multiplicity
of the first eigenvalue is exactly equal to one (see also Proposition 4.5.8).
In 1956 A. Pleijel [Ple] showed that the case of equality in Courant’s nodal
theorem holds only for finitely many values of the integer k. For a recent dis-
cussion on the geometry of nodal domains for the Dirichlet problem on a square
menbrane; see the recent article of P. Bérard and B. Helffer [Bé-He].
Remark 5.4.3. Courant’s nodal theorem and Pleijel’s theorem extends for compact
manifolds (see [Cha] for a proof).

5.5 Exercises
Exercise 5.5.1. Compute the spectrum of the Laplacian on a rectangular domain
Œ0; a Œ0; b in R2 with Neumann boundary conditions.
Exercise 5.5.2. Consider the operator:
h2
H D   e2 r 1
2m
p
with domain H 2 .R3 /, where m; e > 0 and r WD x 2 C y 2 C z 2 . Show that H
is self-adjoint and
me4
Spec.H / D ;n1 :
2h2 n2
 5.5 Exercises 123

Exercise 5.5.3. Prove that the Hermite functions

p 1 x2 2 d
n
2
en .x/ D .2n nŠ / 2 e 2 Hn .x/ with Hn .x/ D .1/nex .ex /
dx n
form a Hilbert basis of L2 .R/.

Exercise 5.5.4. Compute the spectrum of a flat Klein bottle .see in [BGM] page
151/.

Exercise 5.5.5. Prove the Weyl formula for the Laplacian on a rectangular do-
main.

Exercise 5.5.6. Let .M1 ; g1 / and .M2 ; g2 / be two closed Riemannian manifolds.
Find the spectrum and eigenvectors of the product .M1 M2 ; g1 g2 /.
Chapter 6
Intermezzo: “Can one hear the holes
of a drum?”

In this chapter we describe in details an application of the minimax principle.

Here, we consider a closed Riemannian manifold .M; g/ and a subset A of M . We
are interested in comparing the eigenvalues .k .M //k1 of the manifold .M; g/
and the eigenvalues .k .M  A//k1 of the manifold .M  A; g/ with Dirichlet
boundary conditions.
The behaviour of the spectrum of a Riemannian manifold .M; g/ under topo-
logical perturbation has been the subject of intensive research. The most famous
example is the crushed ice problem [Kac2], see also [Ann]. This problem is to
understand the behaviour of eigenvalues of the Laplacian with Dirichlet boundary
conditions on a domain with small holes. This subject was first studied by M.
Kac [Kac2] in 1974. Then, J. Rauch and M. Taylor [Ra-Ta] studied the case of
Euclidean Laplacian in a compact set M of Rn ; they showed that the spectrum of
Rn is invariant under topological excision of a compact subset A with a Newto-
nian capacity zero. Later, S. Osawa, I. Chavel and E. Feldman [Ch-Fe1, Ch-Fe2]
treated Riemannian manifold case. They used complex probabilistic techniques
based on Brownian motion. In [Ge-Zh], F. Gesztesy and Z. Zhao investigated the
case of a Schrödinger operator in Rn with Dirichlet boundary conditions, also us-
ing probabilistic tools. In 1995, in a nice article [Cou] G. Courtois studied the
case of the Laplace–Beltrami operator on a closed Riemannian manifold. He used
very simple techniques of analysis based on the minimax principle. In [Be-Co]
J. Bertrand and B. Colbois treated also the case of the Laplace–Beltrami operator
on a compact Riemannian manifold with boundary conditions, see also [Lab3]
for the case of a Schrödinger operator g C V treated on a closed Riemannian
manifold.
In this section we present a part of the paper of G. Courtois [Cou].

6.1 The main result

Let .M; g/ be a closed Riemannian manifold. Following G. Courtois [Cou], we
show that if we remove topologically a “small” part A from the manifold, the
spectrum of M  A is close of the spectrum of M . More precisely, the “good”
126 Chapter 6. Intermezzo: “Can one hear the holes of a drum?”

parameter for measuring the smallness of A is a type of electrostatic capacity,

defined by
nZ
cap.A/ WD inf jruj2 d Vg I u 2 H 1 .M /;
ZM
o
u d Vg D 0; u  1 2 H01 .M  A/
M

where H01 .M  A/ is the Sobolev space

H01 .M  A/ WD fg 2 H 1 .M /I g D 0 on a open neighborhood of A g;

the closure is taken with respect to the norm kkH 1 .M / and H 1 .M / is the usual
Sobolev space on M (see Section 3.2.2).
Indeed, the smaller cap.A/ is, the closer the spectrum of g on M  A to
the spectrum on M in the following sense:

Theorem 6.1.1 (Courtois, 1995). Let .M; g/ be a closed Riemannian manifold.

For any integer k  1, there exists a constant Ck depending on the manifold
.M; g/ such that for any subset A of M we have
p
0  k .M  A/  k .M /  Ck cap.A/:

6.2 Some useful spaces

The space Cc1 .M  A/ is the set of smooth functions with compact support on
M  A. For a compact subset A of the manifold M the usual Sobolev space
H01 .M  A/ is defined as the closure of Cc1 .M  A/ with respect to the norm
kkH 1 .M / (see Section 3.2.2):

H01 .M  A/ WD Cc1 .M  A/:

But what happens when the set A is not compact? For example, if A is a dense
and countable subset of M , the space of test functions Cc1 .M  A/ is reduced to
f0g. Therefore we cannot define the space H01 .M  A/. This leads us to a more
general definition of H01 .M  A/ which is valid for any subset A of M .

Definition 6.2.1. We define the Sobolev spaces H01 .M  A/ and H01 .M  A/ by

˚ 
H01 .M  A/ WD g 2 H 1 .M /I g D 0 on a open neighborhood of A
and
H01 .M  A/ WD H01 .M  A/;
where the closure is taken with respect to the norm kkH 1 .M / .
 6.2 Some useful spaces 127

We have the

Proposition 6.2.2. If the set A is compact, the previous definition of the space
H01 .M  A/ coincides with the usual ones .see Section 3.2.2/.

Proof. Let f 2 H01 .M  A/ WD H01 .M  A/. Then, by definition, for all "  0
there exists g 2 H01 .M  A/ such that kf  gkH 1 .M /  ". Let us show that
we can write g as a limit of a sequence from the space Cc1 .M  A/. Since
g 2 H01 .M  A/, there exists an open set U A such that gjU D 0. Consider
two open sets U1 and U2 of the manifold M such that
A  U1 ; M  U  U2 ; U1 \ U2 D ;I
and consider also a function ' 2 D .M / such that
'jU1 D 0; 'jU2 D 1:
Obviously, ' belongs to the space Cc1 .M  A/. Next, since g 2 H01 .M  A/ 
H 1 .M / and since the set of smooth functions C 1 .M / is dense in H 1 .M /, there
exists a sequence .gn /n in C 1 .M / such that limn!C1 gn D g in the norm
kkH 1 .M / . We claim that limn!C1 'gn D g in the norm kkH 1 .M / . Indeed
for any n we have
k'gn  gk2H 1 .M /  kgn  gk2H 1 .M U / C k'gn  gk2H 1 .U /
 kgn  gk2H 1 .M / C k'gn  gk2H 1 .U / :
Next, note that for any n
k'gn  gk2H 1 .U / D k'gn k2H 1 .U /
Z Z
D j'gn j d Vg C
2
jr'gn C 'rgn j2 d Vg
ZU
ZU
Z
 j'gn j d Vg C
2
jr'gn j d Vg C
2
j'rgn j2 d Vg
U
Z U U

C2 jr'gn 'rgn j d Vg
U
 k'k1 kgn k2L2 .U /
2
C kr'k2L1 .M / kgn k2L2 .U /
Z
C k'k21 krgn k2L2 .U / C 2 kr'k1 k'k1 jgn rgn j d Vg
U
 k'k21 kgn k2L2 .U / C kr'k21 kgn k2L2 .U /
C k'k21 krgn k2L2 .U /
C 2 kr'k1 k'kL1 .M / kgn kL2 .U / krgn kL2 .U /
thanks to the Cauchy–Schwarz inequality.
128 Chapter 6. Intermezzo: “Can one hear the holes of a drum?”

Therefore,

k'gn  gk2H 1 .U /  kgn k2H 1 .U / 2 k'k21 C kr'k21 C 2 kr'k1 k'k1 :

As a consequence, for any n

k'gn  gk2H 1 .M /  kgn  gk2H 1 .M U /

C kgn k2H 1 .U / 2 k'k21 C kr'k21 C 2 kr'k1 k'k1 :

Now it suffices to note that

kgn k2H 1 .U / D kgn  gk2H 1 .U /  kgn  gk2H 1 .M /

(since g D 0 on the open set U ) and we finally conclude that

 
k'gn  gk2H 1 .M /  kgn  gk2H 1 .M / 1C2 k'k21 Ckr'k21 C2 kr'k1 k'k1 :

The sequence .'gn /n belongs to Cc1 .M  A/N and since limn!C1 gn D g in

the norm kkH 1 .M / , the last inequality implies limn!C1 'gn D g in the norm
kkH 1 .M / .
So we have show that every function f 2 H01 .M  A/ WD H01 .M  A/ is the
limit (in the norm kkH 1 .M / ) of a sequence from Cc1 .M  A/.
Conversely, since Cc1 .M  A/  H01 .M  A/, we get

H01 .M  A/ WD Cc1 .M  A/  H01 .M  A/ WD H01 .M  A/: 

Let us also define the spaces H?1 .M / and SA .M / by

Z
H?1 .M / WD f 2 H 1 .M /I f d Vg D 0
M

and ˚ 
SA .M / WD u 2 H?1 .M /I u  1 2 H01 .M  A/ ;
R
respectively. In the definition of the space H?1 .M / the condition M f d Vg D 0
is analogous to a boundary condition.
Note that H?1 .M / is a Hilbert space for the norm
Z
2 2
kuk? WD kukH 1 D jruj2 d Vg
0 M

and SA .M / is an affine closed subset of H?1 .M /.

 6.3 Electrostatic capacity and the Poincaré inequality 129

6.3 Electrostatic capacity and the Poincaré inequality

Next, we introduce the notion of the electrostatic capacity of a set A.

Definition 6.3.1. The electrostatic capacity cap.A/ of the set A is defined by

Z
cap.A/ WD inf jruj2 d Vg I u 2 SA .M / : (6.1)
M

Let us remark that there exists a unique function uA 2 SA .M / such that

Z
cap.A/ D jruA j2 d Vg :
M

Indeed, here the capacity cap.A/ is just the distance between the function 0 and the
closed space SA .M /. This distance is equal to kuA k2? , where uA is the orthogonal
projection of 0 on SA .M /:
n o
cap.A/ D d? .0; SA .M // WD inf kuk2? I u 2 SA .M / D kuA k2? :

The following lemma establishes the relationships between the capacity cap.A/,
the function uA , and the Sobolev spaces H01 .M  A/; H 1 .M /.

Lemma 6.3.2. For subset A of the manifold M , the following properties are
equivalent
(i) cap.A/ D 0;
(ii) uA D 0;
(iii) 1 2 H01 .M  A/;
(iv) H01 .M  A/ D H 1 .M /.

Proof. It is clear from the formula (6.1) that .i/ , .ii/ , .iii/.
Next, suppose that (iii) holds. Hence, there exists a sequence .vn /n  H01 .M 
N
A/ such that limn!C1 vn D 1 in the norm kkH 1 .M / . So, for any smooth func-
tion ' 2 C 1 .M / we have limn!C1 'vn D ' in the norm kkH 1 .M / ; indeed, for
any integer n
Z Z
k'vn  'k2H 1 .M / D j'vn  'j2 d Vg C jr .'vn /  r'j2 d Vg :
M M

First, for any n, we have

Z Z
j'vn  'j2 d Vg D j' .vn  1/j2 d Vg  k'k21 kvn  1k2L2 .M / :
M M
130 Chapter 6. Intermezzo: “Can one hear the holes of a drum?”

Since limn!C1 vn D 1 in the norm kkH 1 .M / , we get

Z
lim j'vn  'j2 d Vg D 0:
n!C1 M

On the other hand, for any n

Z
jr .'vn /  r'j2 d Vg D kr'vn  r' C 'rvn k2L2 .M /
M
2
 kr'vn  r'kL2 .M / C k'rvn kL2 .M /
2
 kr'k1 kvn  1kL2 .M / C k'k1 krvn kL2 .M / :

Since limn!C1 vn D 1 in the norm kkH 1 .M / , we deduce that

Z
lim jr .'vn /  r'j2 d Vg D 0:
n!C1 M

Therefore, for any ' 2 C 1 .M / we have limn!C1 'vn D ' in the norm kkH 1 .M / .
Next, since C 1 .M / is dense in H 1 .M /, for any function f 2 H 1 .M / we
have limn!C1 f vn D f . Since the sequence .'vn /n  H01 .M  A/N we con-
clude that f belongs to space H01 .M  A/.
Finally, it is easy to see that .iv/ ) .iii/. 
An obvious consequence of this lemma is the following result:

Proposition 6.3.3. The spectra of .M; g/ and .M  A; g/ coincide if and only if

cap.A/ D 0.

Now, let us state the Poincaré inequality.

Theorem 6.3.4 (Poincaré inequality). If 2 .M / denotes the first non-null eigen-

value of the operator g on the manifold .M; g/, then
cap.A/
kuA k2L2 .M / 
2 .M /
for any subset A of M .

Proof. The case cap.A/ D 0 is an obvious consequence of the previous lemma.

Suppose now that cap.A/ > 0. Then kuA kL2 .M / > 0. By the minimax principle,
the second eigenvalue 2 .M / of .M; g/ is given by:
Z
jr'j2 d Vg
2 .M / D min max M Z :
E H 1 .M / '2E
dim.E /D2 '¤0 ' 2
d V g
M
 6.4 A detailed proof of the main Theorem 6.1.1 131

Since uA belongs to the space H 1 .M /, we get:

Z
jr'j2 d Vg
2 .M /  max MZ
'2spanfuA ;1g
'¤0 ' 2 d Vg
M
Z
˛2 jruA j2 d Vg
 max Z M Z
:
.˛;ˇ /2R2
.˛;ˇ /¤.0;0/ ˛ 2 u 2
A d V g C 2˛ˇ u A d V g C ˇ 2
vol.M /
M M
Z
Since uA 2 SA .M /, we get uA d Vg D 0, whence
M Z
˛2 jruA j2 d Vg
2 .M /  max Z M
.˛;ˇ /2R2
.˛;ˇ /¤.0;0/ ˛
2 uA d Vg C ˇ 2 vol.M /
2

Z M
2
jruA j d Vg
MZ
 :
u2A d Vg
M

And the statement of the theorem is established. 

6.4 A detailed proof of the main Theorem 6.1.1

Recall our main result of Chapter 6 (Theorem 6.1.1):

Theorem (Courtois, 1995). Let .M; g/ be a closed Riemannian manifold. For any
integer k  1, there exists a constant Ck depending on the manifold of .M; g/,
such that for any subset A of M we have
p
0  k .M  A/  k .M /  Ck cap.A/:

Proof. Let us denote by .ek /k1 an orthonormal eigenbasis (for the operator g
on the closed manifold .M; g/) of the space L2 .M /. For any integer k  1; we
consider the sets

Fk WD span fe1 ; e2 ; : : : ; ek g and Ek WD ff .1  uA / ; f 2 Fk g :

132 Chapter 6. Intermezzo: “Can one hear the holes of a drum?”

First, observe that Ek  H01 .M  A/. For all j 2 f1; : : : ; kg we introduce also
the functions
j WD ej .1  uA / 2 Ek :
˝ ˛
 Step 1: we compute the L2 -inner product i ; j L2 .M / for any pair .i; j / 2
f1; : : : ; kg2 : Z
˝ ˛
i ; j L2 .M / D ei ej .1  uA /2 d Vg
M
Z Z
D ıi;j  2 ei ej uA d Vg C ei ej u2A d Vg :
M M

Thus, for any pair .i; j / 2 f1; : : : ; kg2 ,

ˇ˝ ˇ Z Z
ˇ ˛ ˇ ˇ ˇ ˇ ˇ
ˇ i ; j L2 .M /  ıi;j ˇ  2 ˇ ˇ
ei ej uA d Vg C ˇei ej u2 ˇ d Vg :
A
M M

Hence, by the Cauchy–Schwarz inequality,

ˇ˝ ˛ ˇ  
ˇ ˇ ei ej  kuA k 2
ˇ i j L2 .M /
 ;   ıi;j ˇ  2 max 1 L .M /
1i;j k
 
C max ei ej 1 kuA k2L2 .M /
1i;j k
  p
 2 max ei ej 1 Vol.M / kuA kL2 .M /
1i;j k
 
C max ei ej 1 kuA k2L2 .M / ;
1i;j k

and with the Poincaré inequality we have

ˇ˝ ˛ ˇ p
ˇ ˇ
ˇ i j L2 .M /
 ;   ıi;j ˇ  Bk;M cap.A/ C cap.A/

where Bk D Bk .e1 ; e2 ; : : : ; ek ; 2 .M /; M /  0. Since the eigenfunctions

e1 ; e2 ; : : : ; ek and the eigenvalue 2 .M / depend only on .M; g/, for any integer
k the constant Bk depends only on .M; g/, i.e., Bk D Bk .M /.
Therefore, there exists "k 2 0; 1Π(which depends on the constant Bk ) such
that for all A  M we have

cap.A/  "k ) dim.Ek / D k and

ˇ  ˇ p
ˇ ˇ
8j 2 f1; : : : ; kg; ˇj L2 .M /  1ˇ  Dk cap.A/
2

where (and for the same reasons as for Bk ) for all integer k, the constant Dk
depends only on M , i.e., Dk D Dk .M /.
 Step 2: Take  D f .1  uA / 2 Ek , with f 2 Fk . Without loss generality
we can assume that kf kL2 .M / D 1, indeed, we have R./ D R kf k 2 and
L .M /
 6.4 A detailed proof of the main Theorem 6.1.1 133

in our context we are interested in the Rayleigh quotient of  (see the end of the
final step of the proof).
We have
Z Z
2
jrj d Vg D jrf  r .f uA /j2 d Vg
M
ZM
Z
2
D jrf j d Vg C jrf uA C f ruA j2 d Vg
M
Z M

2 rf r .f uA / d Vg
Z M
Z Z
D jrf j2 d Vg C jrf uA j2 d Vg C jf ruA j2 d Vg
M
Z M
Z M

C2 rf ruA f uA d Vg  2 jrf j uA d Vg
2

Z M M

2 rf ruA f d Vg
Z M
Z Z
D jrf j2 d Vg C jrf uA j2 d Vg C jf duA j2 d Vg
M
Z M
Z M

2 jrf j uA d Vg  2
2
rf ruA f .1  uA / d Vg :
M M
Hence
Z Z Z Z
2 2 2
jrj d Vg D jrf j d Vg C jrf uA j d Vg C jf ruA j2 d Vg
M
„M
ƒ‚ … „ M
ƒ‚ … „ M
ƒ‚ …
WDA.f / WDB.f / WDC.f /
Z Z
2 jrf j2 uA d Vg  2 rf ruA f .1  uA / d Vg :
„M
ƒ‚ … „M
ƒ‚ …
WDD.f / WDE.f /
R
 Estimate of A.f / WD M jrf j2 d Vg  0. Since f 2 Fk we can write f D
Pk Pk
i D1 ˛i ei where .˛i /1i k 2 R with i D1 ˛i D 1 (since kf kL2 .M / D 1),
k 2

and so
* k +
X Xk X ˝ ˛
A.f / D ˛j rej ; ˛i rei D ˛i ˛j rej ; rei L2 .M /
j D1 i D1 L2 .M / i;j
X ˝ ˛ X ˝ ˛
D ˛i ˛j  ej ; g ei L2 .M /
D ˛i ˛j ej ; g ei L2 .M /
i;j i;j

X ˝ ˛ X
k
D ˛i ˛j i .M / ej ; ei L2 .M / D ˛i2 i .M /  k .M /:
i;j i D1
134 Chapter 6. Intermezzo: “Can one hear the holes of a drum?”

Hence, for any integer k, and for any function f 2 Fk such that kf kL2 .M / D 1
we have
0  A.f /  k .M /:
R
 Estimate of B.f / WD M jrf uA j2 d Vg . We get B.f /  krf k21 kuA k2L2 .M /
and, with the Poincaré inequality

cap.A/
kuA k2L2 .M / 
2 .M /

we deduce that for any k and any function f 2 Fk such that kf kL2 .M / D 1 we
have
0  B.f /  Ek cap.A/
where Ek D Ek .e2 ; 2 .M // > 0. Moreover, since the eigenfunction e2 and
the eigenvalue 2 .M / depend only on .M; g/, for any integer k the constant Ek
depend only on .M; g/, i.e., Ek D Ek .M /. R
 Estimate of C.f /. The term C.f / is equal to M jf ruA j2 d Vg . We have

C.f /  kf k21 kruA k2L2 .M /

so, since kruA k2L2 .M /  cap.A/, for any k and any function f 2 Fk such that
kf kL2 .M / D 1, it holds that

0  C.f /  Fk cap.A/;

where Fk D Fk .f; e2 ; 2 .M // > 0. Here, for k fixed, the constant Fk depends

also on f , and f depends on the functions f1 ; f2 ; : : : ; fk (which depend only on
P
M ) and on the scalars ˛ 1 ; ˛2 ; : : : ; ˛k ; since kiD1 ˛i2 D 1, all the .˛i /1i k are
bounded in R, so finally, for any k the constant Fk can be bounded by a constant
(we denote it also by Fk D Fk .M /) which depends only on M .
 Estimate of jD.f /j: we have
ˇZ ˇ Z
ˇ ˇ
ˇ
jD.f /j D ˇ jrf j uA d Vg ˇˇ  krf k21
2
juA j d Vg
M M
p
 krf k21 Vol.M / kuA kL2 .M /
s
p cap.A/
 krf k21 Vol.M / :
2 .M /

Hence, for any k and any f 2 Fk such that kf kL2 .M / D 1

p
jD.f /j  Gk cap.A/
 6.4 A detailed proof of the main Theorem 6.1.1 135

where (and for the same reasons as in the estimation of C.f /, see the constant
Fk ) for any k, the constant Gk depends onlyRon M , i.e., Gk D Gk .M /.
 Estimate of jE.f /j: recall that E.f / D M rf ruA f .1  uA / d Vg , hence
Z Z
jE.f /j  jrf ruA j jf j d Vg C jrf ruA j jf uA j d Vg :
M M

For the first term we have

Z p
jrf ruA j jf j d Vg  kf k1 krf k1 Vol.M / kruA kL2 .M / I
M
p
so if we denote Hk WD kf k1 krf k1 Vol.M /, we get
Z q p
jrf ruA j jf j d Vg  K kruA k2L2 .M /  Hk cap.A/
M

where (for the same reasons as above), for any integer k, the constant Hk depends
only on M , i.e., Hk D Hk .M /.
Next, for the second term we have
Z
jrf ruA j jf uA j d Vg  krf k1 kf k1 kruA kL2 .M / kuA kL2 .M /
M
s
cap.A/ p
 krf k1 kf k1 Hk cap.A/
2 .M /
0
 Hk;M cap.A/;

where (for the same reasons as above), for any integer k, the constant Hk depends
only on M , i.e., Hk0 D Hk .M /.
So, for any integer k we get

00
p
jE.f /j  Hk;M cap.A/ C cap.A/ ;

where Hk00 WD Hk00 .M /.

Combining the estimates for A.f /; B.f /; C.f /; jD.f /j and jE.f /j, for any
k and any function  D f .1  uA / 2 Ek , with f 2 Fk such that kf kL2 .M / D 1
we obtain
Z p
jrj2 d Vg  k .M / C Ik cap.A/ C cap.A/
M

where, for any k, the constant Ik depend only on M , i.e., Ik D Ik .M; V /.

136 Chapter 6. Intermezzo: “Can one hear the holes of a drum?”

 Step 3: Now we claim that for any A  M such that cap.A/  "k and any
function  2 Ek , p
0
kk2L2 .M /  1  Jk;M cap.A/;
0 0 0
where, for any k, the constant Jk;M depend only on M , i.e., Jk;M D Jk;M .M /.
Indeed, let  2 Ek . We have seen above in Step 1 that

cap.A/  "k ) dim.Ek / D k and

ˇ  ˇ p
ˇ ˇ
8j 2 f1; : : : ; kg; ˇj L2 .M /  1ˇ  Dk cap.A/:
2

P
Hence, since  2 Ek , we can write  D .1  uA /f with f D kiD1 ˛i ei , where
.˛i /1i k 2 Rk . As in the Step 2, we can assume that kf kL2 .M / D 1, and so
Pk 2
i D1 ˛i D 1. Next, we compute kkL2 .M / :
2

 k 2  k 2
X  X 
   
kk2L2 .M / D .1  uA / ˛i ei  D ˛i i 
   
i D1 L2 .M / i D1 L2 .M /

X
k X
k
˝ ˛
D ˛i2 ki k2L2 .M / C ˛i ˛j i ; j L2 .M / :
i D1 i;j D1; i ¤j

Since

X
k X
k  Z Z 
˛i2 ki k2L2 .M / D ˛i2 1  2 ei2 uA d Vg C ei2 u2A d Vg
i D1 i D1 M M

X
k  Z Z 
D1 2
˛i 2 ei2 uA d Vg  ei2 u2A d Vg
i D1 M M

X
k Z
 
D1 ˛i2 ei2 2uA  u2A d Vg ;
i D1 M

we obtain

X
k Z X
k
  ˝ ˛
kk2L2 .M / D 1 ˛i2 ei2 2uA  u2A d Vg C ˛i ˛j i ; j L2 .M / :
i D1 M
i;j D1; i ¤j

We have seen in Step 1 that, for cap.A/ small enough,

ˇ˝ ˛ ˇ p
ˇ ˇ
ˇ i ; j L2 .M /  ıi;j ˇ  Bk cap.A/ C cap.A/ :
 6.4 A detailed proof of the main Theorem 6.1.1 137

hence, since all the .˛i /1i k are bounded in R, for cap.A/ small enough we can
0
find a constant Bk;M which depends only on M , i.e., Bk0 D Bk0 .M / such that, for
cap.A/ small enough
ˇ ˇ
ˇ ˇ
ˇ X k
˝ ˛ ˇ p
ˇ ˛i ˛j i ; j L2 .M / ˇˇ  Bk0 cap.A/:
ˇ
ˇi;j D1; i ¤j ˇ
Finally, in the same spirit as in the estimations in Step 2, there exists a constant
00
Bk;M which depend only on M , i.e., Bk00 D Bk00 .M / such that, for cap.A/ small
enough ˇ k ˇ
ˇX Z   ˇ p
ˇ ˇ
ˇ 2
˛i ei 2uA  uA d Vg ˇ  Bk00 cap.A/:
2 2
ˇ M ˇ
i D1
It follows that, p
kk2L2 .M /  1  Bk000 cap.A/;
where the constant Bk000 depend only on M , i.e., Bk000 WD Bk000 .M /.
 Final step: By Steps 2 and 3, for any function  2 Ek we get
Z
p
jrj2 d Vg k .M / C Ik cap.A/ C cap.A/
MZ  p :
000
 d Vg
2 1  B k
cap.A/
M
Hence, for cap.A/ small enough (i.e., cap.A/  "k ),
Z
jrj2 d Vg p
MZ  k .M / C Lk cap.A/;
 2 d Vg
M
where Lk WD Lk .M /. Next, since for any k  1
Z
jr'j2 d Vg
k .M  A/ D min max Z
M
E H01 .M A/ '2E
dim.E /Dk '¤0 ' 2 d Vg
M

and since  2 H01 .M  A/, we get for all k  1

Z
jrj2 d Vg p
k .M  A/  M Z  k .M / C Ck cap.A/
 2 d Vg
M
This completes the proof of the theorem. 
Chapter 7
Inverse problems in spectral geometry

This chapter is devoted to inverse problems. First we present a brief historical re-
view of inverse problems, next we give a proof of Milnor’s counter-example, and
then we explain the notion of heat kernel and its applications to spectral inverse
problems. We also provide a detailed example of isospectral but non-isometric do-
mains in the plane R2 . We conclude this chapter with some elements of conformal
geometry, including some recent results in Riemannian geometry.

7.1 Can one hear the shape of a drum?

In 1966, M. Kac [Kac1] in his famous article “Can one hear the shape of a drum?”
investigated the following question: given a Riemannian manifold .M; g/ (the
membrane of a drumhead), does the spectrum of g (the harmonics of the drum-
head) determine geometrically, up to an isometry, the manifold .M; g/? So, one
of the most fundamental questions in spectral geometry is: if two manifolds are
isospectral, are they isometric?
Before 1964, this problem remained quite mysterious. In 1964 J. Milnor
[Mil] gave the first counter-example: a pair of 16-dimensional flat tori which are
isospectral and nonisometric (see Section 7.3 for details). Subsequently, many
other counter-examples have been constructed: see the papers of T. Sunada in
1985 [Sun], the papers [Go-Wi] of C. Gordon and E.N. Wilson in 1984, and also
P. Bérard’s method of isospectral construction by transplantation [Bér2], [Bér3].
In 1992, C. Gordon, D. Webb, and S. Wolpert [GWW1], produced the first pla-
nar counter-example. For more details see [GWW1], [GWW2] and the survey of
P. Bérard [Bér5], [Bér6], [Bér7]. Let us also mention the article of S. Zelditch
[Zel] in 2000, in which he showed that for a simply connected domain in R2
with analytic boundary and with two orthogonal axes of symmetry, the spectrum
determines completely the geometry of the domain.

7.2 Length spectrum and trace formulas

The spectrum of the Laplacian determines some important geometric invariants:
the dimension, the volume of the manifold, and the integral of the scalar curvature
Scalg over the manifold, (for the details see the next section). In fact, the spectrum
of the Laplacian determines some other invariants, as for example the length spec-
trum of the manifold. The length spectrum of a Riemannian manifold .M; g/ is
140 Chapter 7. Inverse problems in spectral geometry

the set of lengths of closed geodesics1 on .M; g/ counted with multiplicities (the
multiplicity of a length is the number of free homotopy classes of closed curves
containing a geodesic of the given length). In 1973 Y. Colin de Verdière [Col1],
[Col2] showed that, in the compact case, up to a generic hypothesis (which it
always satisfied in the case of negative sectional curvature), the spectrum of the
Laplacian determines the length spectrum. Colin de Verdière’s proof is based on
trace formulas. More generally, trace formulas are useful tools in the analysis of
PDE, in spectral theory, mathematical physics, etc. . . .
The formal principle of trace formulas is the following. Consider an unboun-
ded linear operator H on a Hilbert space and suppose the spectrum of H is dis-
crete:
Spec.H / D fk ; k  1g :
Let f be a “nice” function. The principle is to compute the trace of the operator
f .H / in two ways:
 the first: using the eigenvalues of the linear operator f .H /:
X
trace.f .H // D f .k /;
k1

 the second: using the integral kernel of f .H /: if

Z
f .H /'.x/ D Kf .x; y/'.y/ dy
M

for all x 2 M , then

Z
trace.f .H // D Kf .x; x/ dx:
M

We then get the following equality:

X Z
f .k / D Kf .x; x/ dx:
M
k1

However, the major difficulty with this method is to find the “good” choice for the
function f . Some of the usual choices are
 f .x/ D ext , where t  0 (heat function);
 f .x/ D 1
xs
, where s 2 C such that Re.s/ > 1 (Riemann Zêta function);
i tx
 f .x/ D e h , where t  0 (Schrödinger function).
1 Note that geodesics are trajectories associated to the dynamics of classical Hamiltonian flow of H D
 g (geodesic flow) in the phase plane, identified with the cotangent bundle T ? M .
 7.2 Length spectrum and trace formulas 141

The simplest example of exact trace formula is the Poisson summation formula
for a lattice  of Rn .
Recall that for any f 2 S .Rn / the Poisson summation formula reads
X 1 X
f .k/ D b.`/;
f
Vol ./ ?
k2 `2

where Z
b.y/ WD
f f .t/e2i hy;t i dt
Rn
is the Fourier transform of the function f . A classical calculation shows that the
Fourier transform of the function
2
f .x/ D e˛kxk
for x 2 Rn , where ˛ > 0, is given for all y 2 Rn by
  n2
b   2 kyk2
f .y/ D e ˛ :
˛
2 2
Now, applying Poisson’s formula to f .x/ D e4 t kxk , where t > 0 we get
X 2 2 1 X kyk2
e4 t kxk D n e 4t :
x2
Vol./.4 t/ 2 y2 ?

Replacing the lattice  by  ? we obtain

X 2 t kxk2 Vol./ X kyk2
e4 D n e 4t ;
x2 ?
.4 t/ 2
y2

because  ?? D  and Vol. ? / D Vol./1

.
Since the spectrum of the Laplacian (with closed manifold conditions) on the
flat torus  is given by (see Section 5.1.1):
˚ 
Spec.; can/ D 4 2 kxkRn ; x 2  
we obtain for all t  0
X Vol./ X kyk2
Z .t/ D et D n e 4t : (7.1)
2Spec.;can/
.4 t/ 2
y2

Further, since the distance between the origin and a point of the lattice  is exactly
the length of a periodic geodesic on the flat torus, we get the equality
X Vol./ X  `2
et D n e 4t ;
2Spec.;can/
.4 t/ 2
`2†
142 Chapter 7. Inverse problems in spectral geometry

where † is the length spectrum. In this last equality on the right-hand side we
have the geometric information (dimension, volume, . . . ) and on the left-hand
side we have spectral information. For a recent reference on trace formulas, see
the paper of Y. Colin de Verdière [Col3].

7.3 Milnor’s counterexample

In this section we present a counterexample due to J. Milnor [Mil] in 1964.

Theorem 7.3.1 (Milnor, 1964). There exist two lattices  and  0 of R16 such that
the associated tori T16 ./ and T16 . 0 / are isospectral, but not isometric.

To prove this theorem we consider for any integer n 2 8N the following sub-
lattice of Zn of index 2:
8 9
< X
n =
2 WD .x1 ; x2 ; : : : ; xn / 2 Zn I xj 2 2Z
: ;
j D1

and consider also the vector

 
1 1 1
!n WD ; ;:::; 2 Rn :
2 2 2

Denote by .n/ the lattice generated by 2 and !n .

Since 2!n D .1; 1; : : : ; 1/ 2 2 the lattice 2 is of index 2 in .n/, thus we have
1
Vol..n// D Vol.2 /
2
D Vol.Zn / D 1:

The following two lemmas are obvious.

Lemma. For any n 2 8Z, if y 2 .n/, then kyk2 2 2Z:

Pn
Proof. If x D j D1 xj 2 2 , then
0 12
X
n X
n X
kxk2 D xj2 D @ xj A  2 xi xj ;
j D1 j D1 i <j

thus kxk2 2 2Z.

 7.3 Milnor’s counterexample 143
1 
If x 2 !n Z, then there exists ˛ 2 Z such that x D ˛ ; 1 ; : : : ; 12
2 2
, whence
X
n
1 ˛2 n
kxk2 D ˛ 2 D 2 2Z
4 4
j D1

because n 2 8Z.
Finally, if y D x C ˛!n with ˛ 2 Z; x D .x1 ; x2 ; : : : ; xn / 2 2

1X
n
hx; !n i D xi 2 Z
2
j D1

thus
kyk2 D kxk2 C ˛ 2 k!n k2 C 2˛ hx; !n i 2 2Z: 
The second lemma is

Lemma 7.3.2. For any n 2 8Z, we have  ? .n/ D .n/.

Proof. First consider .y; y 0 / 2 .n/2 such that y D x C k!n and y 0 D 0 0

 x C k !n
0 0 0 0 0
with .k; k / 2 Z ; x D .x1 ; x2 ; : : : ; xn / 2 2 ; x D x1 ; x2 ; : : : ; xn 2 2 . We
2

claim that ˝ ˛
y; y 0 2 Z;
and consequently y 0 2  ? .n/. Indeed,
˝ ˛ ˝ ˛ ˝ ˛
y; y 0 D x; x 0 C kk 0 h!n ; !n i C k 0 hx; !n i C k x 0 ; !n :
Here it is clear that hx; x 0 i 2 Z, and
X
n
1 n
h!n ; !n i D k!n k2 D D 2Z
4 4
j D1

(because n 2 8N). Moreover,

X
n
xj k0 X
n
k 0 hx; !n i D k 0 D xj 2 Z
2 2
j D1 j D1

and obviously for the same reason we have k hx 0 ; !n i 2 Z. We obtain that

hy; y 0 i 2 Z and y 0 2  ? .n/. Therefore, we have
.n/   ? .n/:
Since
1
Vol. ? .n// D D 1;
Vol..n//
the lattices .n/ and  ? .n/ have the same volume, and so  ? .n/ D .n/: 
144 Chapter 7. Inverse problems in spectral geometry

From the two previous lemmas we get

Proposition 7.3.3. The lattices .16/ and .8/ ˚ .8/ are non-isometric, there-
fore the associated tori are also non-isometric.

Proof. Let us denote by e1 ; e2 ; : : : ; e8 the canonical basis of R8 . It is clear that

the family
G WD fe1  e8 ; e2  e8 ; : : : ; e7  e8 ; 2e8 g
generates the lattice 2 . Moreover, since

2e8 D 2!8  .e1 C e2 /  .e3 C e4 /  .e5 C e6 /  .e7  e8 /

the family

G WD fe1  e8 ; e2  e8 ; : : : ; e7  e8 ; e1 C e2 ; e3 C e4 ; e5 C e6 ; !8 g
p
generates the lattice .8/. Here the norm of every vectors from G is equal to 2.
pother hand, the lattice .16/ cannot be generated by vectors
On the
P16
with norm
equal to 2; indeed, every element of .16/ can be expressed as j D1 ai ei or
P16  
j D1 ai C 2 ei , with .a1 ; a2 ; : : : ; an / 2 2 . So a generator system of .16/
1
P  
admits necessarily elements of second type, and for x WD 16 j D1 ai C 2 ei such
1

a vector we have

1X
16
2 16
kxk D .2ai C 1/2  D 4;
4 4
j D1

p D R ˚R the lattice .8/˚.8/ is generated by elements

hence kxk  2. In R 16 8 8

with norm equal to 2 and the lattice .16/ is not, consequently .8/ ˚ .8/ and
.16/ cannot be isometric. 
To finish, prove

Proposition 7.3.4. The tori associated to the lattices .16/ and .8/ ˚ .8/ are
isospectral.

Proof. The proof is based on the theta function

X 2
 .t/ WD et kxk
x2

defined for t > 0. Note that for all t > 0, the formula (7.1) yields
  X X
t t 2
Z D e 4 D et kxk D  .t/
4 ?
2Spec.;can/ x2
 7.3 Milnor’s counterexample 145

thus the theta function determines the spectrum of  (because the spectral partition
function Z determine the spectrum of , see Proposition 5.3.2).
Next, the trick here is to consider the function
2
f .y/ D ekyk :
Then, for all x we get
b.x/ D ekxk2 :
f
p
Consider the lattice t WD t, where t > 0 and  is a lattice of Rn such that
 ? D  (hence Vol./ D 1/ and such that for all x 2  we have kxk2 2 2Z.
Applying the Poisson summation formula2 to f and the lattice t , we get for all
t >0
X 2 1 X 2
ekxk D ek`k :
Vol .t / ?
x2t `2t

Since t? D 1
p
t
 ?, we further have
X 2 1 X k`k2
et kxk D n e t :
x2
t Vol./ `2 ?
2

Therefore,
X 2 1X k`k2
et kxk D n e t ;
x2
t 2
`2
i.e.,  
1 1
 .t/ D n  ;
t2 t
where the function  is defined for all z 2 C such that Re.z/ > 0 by the same
formula X 2
 .z/ WD ezkxk
x2
 admits a holomorphic extension to H WD fz 2 CI Re.z/ > 0g, the half-plane
H is invariant under the mapping z 7! 1z and the function
 
1 1
'W z 7!  .z/  n 
z 2 z
is holomorphic on H (here n 2 8N). Next, since ' vanishes on the real positive
axis R?C , the function ' vanishes on the whole H, so for all z 2 H we have
 
1 1
 .z/ D n  :
z2 z
2 Based on the theory of Fourier series; for a proof see for example [BGM].
146 Chapter 7. Inverse problems in spectral geometry

Moreover, for all z 2 H

X 2 2 X 2
 .z C i / D e t zkxk ei kxk D e t zkxk D  .z/;
x2 x2

because kxk2 2 2Z.

Thus the theta function  is a modular function, in fact even a modular form,
because if z D u C iv we have  .z/ ! 1 for v ! C1 uniformly in u. For
n D 16 the weight of the modular form  is n4 D 4 and the space of modular
forms with weight equal to 4 is a space of dimension one (see for example [Ser]).
Consequently, since  .1/ D  0 .1/ D 1, we get

 D  0

therefore the tori .16/ and .8/˚.8/ are isospectral (because the theta function
determines the spectrum, see Proposition 5.3.2). 
In dimension two and three the spectrum of the Laplacian characterizes a flat
torus. This is no longer true for dimension n  4: there exists isospectral but
non-isometric tori. In the general case, for a given flat torus the number of flat tori
with the same spectrum is finite and an upper bound of this number is known.

7.4 Prescribing the spectrum on a manifold

As discussed into the Introduction, for a fixed topological manifold M and a finite
increasing sequence of real numbers 0 < a1  a2      aN there exists
a metric g such that for any integer 1  k  N we have k .M; g/ D ak . Indeed,
in 1987, Y. Colin de Verdière [Col4] proved the following amazing theorem:

Theorem 7.4.1 (Colin de Verdière, 1987). Let M be a compact manifold of di-

mension n  3. For every finite increasing sequence 0 < a1  a2      aN
there exists a metric g on M such that for any k  N we have k .M; g/ D ak :

This theorem implies that for a compact manifold of dimension n  3, there

are no restrictions on the multiplicities of the (non-null) eigenvalues. The hy-
pothesis on the dimension is fundamental, since we have seen in Section 5.4 that
in dimension two there exists topological restrictions on the multiplicities of the
eigenvalues.
For the proof of this theorem Y. Colin de Verdière considered a complete graph
with N C 1 vertices and constructed a Laplacian operator on this graph with the
prescribing spectrum (see the book of Y. Colin de Verdière for a complete intro-
duction about Laplacians on graphs). Next the trick of the proof is to embed the
 7.5 Heat kernel and spectral geometry 147

graph into the manifold and tending the N first eigenvalues of the manifold to
the spectrum of the graph; the final step if based on the Arnold’s transversality
argument.

7.5 Heat kernel and spectral geometry

In this section .M; g/ is a closed Riemannian manifold and .k /k0 denotes the
spectrum of .M; g/.

7.5.1 The heat equation

Definition 7.5.1. The Cauchy problem for the heat equation on a closed Rieman-
nian manifold .M; g/ is the following: find a function uW M RC ! R such
that: 8
ˆ @u
< .x; t/ D g u.x; t/
@t
:̂
u.x; 0/ D u0 .x/ 8x 2 M:
where u0 is a given smooth function on M .

Physically, for every point x 2 M the number u0 .x/ is an initial temperature

on the manifold M and the solution .x; t/ 7! u.x; t/ describes the evolution of
the temperature at point x 2 M with the time t.
The first way to solve this equation is to use the spectral theory of g and
the bounded functional calculus (see Section 2.6.2). Indeed, we can consider the
bounded one-parameter semi-group fU.t/gt 0 defined for all t  0 by

U.t/ D et g
:
The generator of this semi-group is the operator g . Consider the function
u.x; t/ WD U.t/u0 .x/:
Then
du
D g u.t/
dt
and
u.x; 0/ D U.0/u0 .x/ D u0 .x/:
Finally, the function u.x; t/ D et g u0 .x/ is a solution of the heat equation. By
the bounded functional calculus, for any t  0 and for any integer k we also have
 t g
e ek D et k ek ;
148 Chapter 7. Inverse problems in spectral geometry

where k D k .M /. So for any initial condition u0 2 L2 .M /, if we denote

by .an /n2N the sequence of `2 .N/ defined by .an /n2N WD . 0 /, where  is the
following unitary projection operator:
8
< L2 .M / ! `2 .N/
W
: ' 7!
h'; ek iL2 ;

then for all t  0

C1
!
  X
u.x; t/ D U.t/u0 .x/ D e t g
ak ek .x/
kD0
C1
X
D ak et k ek .x/:
kD0

In conclusion, for all x 2 M and for all t  0,

C1
X
u.x; t/ D hu0 ; ek iL2 et k ek .x/:
kD0

7.5.2 The heat kernel

Another way to solve the heat equation is to use the heat kernel of the manifold.
The theory of heat kernels on Riemannian manifolds is deep and fundamental,
providing powerful tools for understanding the relationship between geometry and
diffusion processes. The diffusion operator related to the Laplacian on a compact
Riemannian manifold is the heat operator. This operator acts on smooth functions
and admits a fundamental solution which is called the heat kernel. Indeed, the
heat kernel is the fundamental solution of the heat process diffusion equation (see
also Sections 4.1.2 and 7.5.1)
@
u D u:
@t
The heat kernel (see Definition 7.5.2 below) is a function

E W .x; y; t/ 2 M M RC 7! E.x; y; t/

such that for all y 2 M the function .x; t/ 7! E.x; y; t/ is a solution of the heat
equation and for all x 2 M and every test function ' 2 D .M / we have the initial
data condition Z
lim E.x; y; t/'.y/ d Vg .y/ D '.x/:
t !0C M
 7.5 Heat kernel and spectral geometry 149

The theory of the heat kernel has many applications, for example
 in physics: from a physical point of view, the heat kernel is very impor-
tant for understanding the space–time behaviour of the heat conduction in
a material. Indeed, for an initial heat data at time t D 0 given by a Dirac
distribution at the point y, the real number E.x; y; t/ is the temperature
after time t > 0 at the point x.
 In analysis: for example, the heat kernel is an analytic tool which is very
useful in the theory of function spaces (Hölder spaces, BMO spaces . . . ).
In particular, the heat kernel is used for approximation of functions (see for
example [Bu-Be], [Tri]).
 In spectral theory: the heat kernel can be used to build approximations of
spectral projectors (see [Dav], [Dav2], [Kat], [Re-Si]).
 In geometry: one example of interaction between the heat kernel and the
geometry of a compact manifold .M; g/ is the famous Varadhan’s formula
[Var]: for x; y close enough,
2
dM .x; y/
lim t log E.x; y; t/ D  ;
t !0C 4
where E.x; y; t/ is the heat kernel of .M; g/ and dM .x; y/ is the Rieman-
nian distance between x and y. Let us give another example in geometry.
In the article [BBG] P. Bérard, G. Besson and S. Gallot use the heat ker-
nel and a finite number of eigenfunctions of g in order to embed3 every
compact Riemannian manifold in the usual following Hilbert space
( C1
)
X
`2 .R/ WD .an /n 2 RN I jan j2 < C1 :
nD0

In the next section we present in detail another major application of heat

kernel in geometry and topology: the Minakshisundaram–Pleijel expansion
(see Section 7.6).
 In applied mathematics and computer science: in 2006, R. Coifman and S.
Lafon [Co-La] gave an algorithm for embedding a data set (graph or subset
of Rn ) into a Euclidean space via the “diffusion map”. In this space the
data points are reorganized in such a way that the Euclidean distance cor-
responds to the “diffusion metric distance”; the “diffusion metric distance”
describes the connectivity rate between two points of the set.
 In probability: the heat kernel is the distribution of a Brownian motion
generated by the Laplacian on the manifold (see for example [SaC]).
 etc. . . .
3 In fact this embedding is asymptotically isometric (see [BBG]).
150 Chapter 7. Inverse problems in spectral geometry

For a more complete introduction to heat kernel theory, see for example [Dav],
[Ya-Sc], [BGV], [Gri] and for some application see also the article [SaC].

Definition 7.5.2. The heat kernel (or fundamental solution) of the heat equation
on a closed Riemannian manifold .M; g/ is a function:
8
< M M RC ! R
E W
: .x; y; t/ 7! E.x; y; t/

such that:
(i) E is C 0 in the three variables .x; y; t/; E is C 2 in the second variable y and
C 1 in the third variable t.
(ii) For all .x; y; t/ 2 M 2 RC ,
@E
.x; y; t/ D g;y E.x; y; t/;
@t
where g;y is the Laplacian for the second variable y; in other words,
for all y 2 M the function .x; t/ 7! E.x; y; t/ is a solution of the heat
equation.
(iii) For all x 2 M ,
E.x; ; t/ * ıx
in D 0 .M / when t ! 0C .

Remark 7.5.3. Item (iii) of this definition means that for all x 2 M and for all
' 2 D .M / we have
lim hE.x; ; t/; 'iD0 .M /D.M / D hıx ; 'iD0 .M /D.M / ;
t !0C

i.e., for all x 2 M and for all ' 2 D .M /

Z
lim E.x; y; t/'.y/ d Vg .y/ D '.x/:
t !0C M

Since the manifold is compact, for every fixed y 2 M we have existence and
uniqueness of a such function E (for the construction see [BGM] or [Cha2]). We
call this solution the heat kernel of M .

Example 7.5.4. In the Euclidean case M D Rn with g D can, the heat kernel is
given by the expression
1 d 2 .x;y/
E.x; y; t/ D n e 4t ;
.4 t/ 2

where d is the usual Euclidean distance on Rn . This example is very important

because it can be used to it for construct the heat kernel in the general case.
 7.5 Heat kernel and spectral geometry 151

Now, let us focus on the link between the heat kernel and the spectrum of the
Laplacian. First, formally4 , consider the function E W M 2 RC ! R defined by
C1
X
E .x; y; t/ WD et k ek .x/ek .y/:
kD0

We have
C1
X C1
X
g;y E .x; y; t/ D et k ek .x/g;y ek .y/ D k et k ek .x/ek .y/:
kD0 kD0

On the other hand,

C1
X
@E
.x; y; t/ D k et k ek .x/ek .y/
@t
kD0

hence for all .x; y; t/ 2 M 2

RC we obtain the equality
@E
.x; y; t/ D g;y E .x; y; t/:
@t
Moreover, for all ' 2 D .M / we have
Z C1
X Z
t k
E .x; y; t/'.y/ d Vg .y/ D e ek .x/ ek .y/'.y/ d Vg .y/
M M
kD0
C1
X
D et k ek .x/ h'; ek iL2 ;
kD0

so we get
Z C1
X
lim E .x; y; t/'.y/ d Vg .y/ D ek .x/ h'; ek iL2 D '.x/:
t !0C M
kD0

Hence, formally the function E is a fundamental solution of heat equation which

can be expressed as
C1
X
E .x; y; t/ WD et k ek .x/ek .y/:
kD0

4
Here, we suppose that all objects are well defined.
152 Chapter 7. Inverse problems in spectral geometry

The rigorous argument for the previous calculation is provided by

Theorem 7.5.5. If we denote by E the heat kernel of .M; g/, then the series of
functions
X
.x; y; t/ 7! et k ek .x/ek .y/
k0

converges uniformly and absolutely on M M Œ"; C1Œ and the limit is equal
to E.x; y; t/.

Proof. Following J. Dodziuk [Dod1], let us introduce the one-parameter family

of integral operators
8
< L2 .M / ! L2 .M /
Pt W
: ' 7! R E.x; y; t/'.y/ d V .y/;
M g

where t > 0 is the time parameter. In other words, for all x 2 M ,

Z
Pt .'/.x/ D E.x; y; t/'.y/ d Vg .y/:
M

For all t > 0, the associated integral kernel is .x; y/ 7! E.x; y; t/. Since for all
t > 0 the function .x; y/ 7! E.x; y; t/ is continuous on the compact manifold
.M; g/, it belongs to the space L2 .M M /, so for all t > 0 the map Pt is well
defined and is a compact operator.
An important fact is that for all ' 2 L2 .M /, the function Pt .'/ is a solution
of the heat equation.
Next, observe that for all .'; / 2 L2 .M / and for all t > 0 we have

jE.x; y; t/'.y/ .x/j  sup jE.x; y; t/j j'.y/ .x/j 2 L2 .M M /;

.x;y/2M M

so we can apply Fubini’s theorem and get that for all t > 0
Z
hPt '; iL2 D Pt '.x/ .x/ d Vg .x/
M
Z Z 
D E.x; y; t/'.y/ d Vg .y/ .x/ d Vg .x/
M M
Z Z
D E.x; y; t/'.y/ .x/ Vg .y/d Vg .x/
M M
 7.5 Heat kernel and spectral geometry 153
Z Z 
D E.x; y; t/ .x/ d Vg .x/ '.y/ d Vg .y/
ZM M

D Pt .y/'.y/ d Vg .y/ D h'; Pt iL2 :

M

Therefore, for all t > 0 the bounded operator Pt is self-adjoint.

To finish, note that by Duhamel’s principle (see [Cha2])
Z
E.x; z; t C s/ D E.x; y; t/E.y; z; s/ d Vg .y/;
M

and so
Z
.Pt ı Ps / ' D E.x; y; t/ .Ps '/ .y/ d Vg .y/
M
Z Z 
D E.x; y; t/ E.y; z; s/'.z/ d Vg .z/ d Vg .y/
ZM Z M

D E.x; y; t/E.y; z; s/'.z/ d Vg .z/ d Vg .y/

ZM ZM
D E.x; y; t/E.y; z; s/'.z/ d Vg .y/ d Vg .z/
ZM M
Z 
D E.x; y; t/E.y; z; s/ d Vg .y/ '.z/ d Vg .z/
ZM M

D E.x; z; t C s/'.z/ d Vg .z/

M
D Pt Cs .'/.x/:
Therefore, the family fPt gt 0 is a semi-group of bounded operator.
An obvious consequence of the semi-group property is that the operator Pt is
positive; indeed, for all t > 0
˝  ˛
hPt '; 'iL2 D Pt =2 ı Pt =2 '; ' L2
˝ ˛  
D Pt =2 '; Pt =2 ' 2 D Pt =2 '  2  0:
L L

Thus, for all t > 0, Pt is bounded, compact, self-adjoint and positive operator.
Hence there exists a Hilbert basis .k .t//k0 of L2 .M / consisting of eigenvectors
of Pt with corresponding eigenvalues k .t/ (see Theorem 2.5.6), i.e.,
Pt k .t/ D k .t/k .t/:
Moreover, for all t > 0 we have
0 .t/  1 .t/      k .t/  0
and k .t/ ! 0 as k ! C1.
154 Chapter 7. Inverse problems in spectral geometry

On the other hand, for all ' 2 D .M / it holds that

d d
kPt 'k2L2 D hPt '; Pt 'iL2
dt dt
d ˝ ˛
D2 Pt '; Pt ' D 2 g Pt '; Pt ' L2
dt L2
Z
D2 g .Pt '/.x/.Pt '/.x/ d Vg .x/
Z
M

D 2 r.Pt '/.x/r.Pt '/.x/ d Vg .x/

M
D 2 kr .Pt '/kL2  0:

Hence, the real valued function t 7! kPt 'k2L2 is non-increasing on 0; C1Œ.
Since for all t > 0 and for all ' 2 D .M /
Z
Pt .'/.x/ D E.x; y; t/'.y/ d Vg .y/;
M

we have that
Z
lim Pt .'/.x/ D lim E.x; y; t/'.y/ d Vg .y/ D '.x/
t !0C t !0C M

for all x 2 M and for all ' 2 D .M /, and it follows that

kPt 'kL2 < k'kL2

for all ' 2 D .M / and for all t > 0. Consequently, by the density of D .M / in
L2 .M /, for all ' 2 L2 .M /, we have

lim Pt ' D ' and kPt 'kL2 < k'kL2 :

t !0C

Now, for any integer k  0, let us consider the functions k WD k .1/ and k WD
k .1/, so we have
P1 k D k k :
Consequently, for any integer n we have

Pn k D P1C1C C1 k D .P1 /n k D nk k :

This implies that Pr k D rk k for any rational number. Since Q is dense in R
and since the functions t 7! Pt k and t 7! tk k are continuous on 0; C1Πwe
get by density that for all t > 0

Pt k D tk k D et ln k k :
 7.5 Heat kernel and spectral geometry 155

Here, for any k  0 we have k > 0. Indeed, suppose k D 0. Since the function
k is non-trivial and since
lim Pt k D k
t !0C
and
lim Pt k D lim tk k
t !0C t !0C
we get a contradiction, so k > 0.
If we denote ˛k WD  ln k , i.e., k WD e˛k , then

Pt k D tk k D e˛k t k ;
 
so we get kPt k kL2 D e˛k t k L2 D e˛k t kk kL2 : Since kPt k kL2 <
kk kL2 we deduce that ˛k > 0 for any k.
Now, for any fixed " > 0 we can apply Mercer’s theorem (see, for example,
[Hoc]) to the operator P1 to conclude that the series of functions
X
.x; y; t/ 7! et ˛k k .x/k .y/
k0

converges uniformly and absolutely on M M Œ"; C1Œ and the limit is equal
to E.
To complete the proof, we show that for any k, ˛k D k and k D ek . Indeed,
for any k
Pt k D e˛k t k
and since the function Pt k is a solution of the heat equation we get
@  ˛k t   
e k .x/ D g e˛k t k .x/;
@t
i.e.,
˛k e˛k t k .x/ D e˛k t g .k / .x/;
whence
g k D ˛k k : 
Remark 7.5.6. In fact the family fPt gt 0 is a non-negative semi-group of bounded
operators and coincides with U.t/ WD et g :
To finish, we sketch the construction of the heat kernel (for technical details
see [BGM], [Cha2]). It is based on the heat kernel G of Rn with the canonical
metric can:
1  kxyk
2
G.x; y; t/ D n e
4t :
.4 t/ 2
In fact, we regard G as the first-order approximation for the heat kernel of a Rie-
mannian manifold M . Now, we want to construct the second-order approximation
156 Chapter 7. Inverse problems in spectral geometry

of E. We achieve this by multiplying G by a finite series of differentiable func-

tions on a certain domain. More precisely, if we denote by r the injectivity radius
of M , this domain is

Ur WD f.x; y/ 2 M M I d.x; y/ < rg :

For any integer N > 0, let us introduce the function HN defined by

X
N
HN .x; y; t/ WD G.x; y; t/ t k uk .x; y/;
kD0

where for any integer 0  k  N , uk 2 C 1 .Ur / and is such that for all x 2 M

u0 .x; x/ D 1

and
Ly HN .x; y; t/ D G.x; y; t/t N g;y uN .x; y/;
where Ly WD g;y  @t@ . This function is an approximation of the heat kernel, but
is defined only on the open set Ur . We want to extend this solution to M M . To
this end we introduce a smooth radial function W M M ! R such that
ˇ
ˇ 1 if d.x; y/ < r
ˇ 4
.x; y/ D ˇˇ
ˇ 0 if d.x; y/ > r :
2

Thus for N large enough (namely, N > n2 C 1), the function SN WD HN is

a parametrix5 of the heat kernel. Next, using geodesic spherical coordinates, one
can compute the uk for any k such that 0  k  N

1 1
uk .x; y/ D p
.x; y/ d.x; y/k
Z 1
   
 t k1  x; expx .t exp1 1
x y/ g;y uk1 x; expx .t expx y/ dt;
0

where p
det g
.x; y/ WD :
d.x; y/n1
@
5A parametrix of a function Ly D g;y  @t is a function H such that (see [BGM]):
(i) H is continuous on M  M  0; C1Œ;
(ii) Ly H admit a continuous continuation to M  M  Œ0; C1Œ;
(iii) limt!0C H.x; ; t / D ıx for all x 2 M .
 7.5 Heat kernel and spectral geometry 157

In particular for k D 0 we get

1
u0 .x; y/ D p ;
.x; y/
and for k D 1 we have
1
u1 .x; y/ D Scal.x/:
6
To finish, using the parametrix, we construct the heat kernel of .M; g/ in the form
E.x; y; t/ D SN C SN ? H;
where H is continuous on M M Œ0; C1Œ. Here ? is the space–time convolution
product6, namely, for two continuous functions H and J on M M Œ0; C1Œ,
Z tZ
.H ? J /.x; y; t/ WD H.x; z; s/J.z; y; t  s/ d Vg .z/ ds:
0 M

Since (by a computation)

Ly .SN ? H / D Ly SN ? H  H;
we get  
0 D Ly K D Ly SN C Ly SN ? H  H:
We can solve this equation and find (note the analogy with the real equation x C
P
xy  y D 0, with solution y D 1xx
D C1 k
kD1 x )

C1
X  ?k
H D Ly SN :
kD1
Therefore,
C1
X  ?k
E.x; y; t/ D SN C SN ? Ly SN :
kD1
P  ?k
Moreover, the function C1
kD1 Ly SN is m C ` times differentiable on M
M 0; C1Œ, where 2` C m < N  n2 . More precisely, the series of functions
C1
X @` m  ?k
`
@y Ly SN
@t
kD1

(where @my D @y ı @y ı    ı @y is the derivative with respect to the variable y

in a local system of coordinates) converges uniformly and absolutely on every
compact subset of M M 0; C1Œ.
6 This convolution product is associative.
158 Chapter 7. Inverse problems in spectral geometry

Finally,
C1 C1
@` m X  ?k X @`  ?k
`
@y Ly SN D `
@m
y Ly SN D O t N 2n2`mC1 ;
@t @t
kD1 kD1

PC1  ?k
if we denote QN .x; y; t/ WD SN ? kD1 Ly SN we get

X
N
E.x; y; t/ D G.x; y; t/ t k .x; y/uk .x; y/ C QN .x; y; t/;
kD0

with
n
QN .x; y; t/ D O t N  2 1

uniformly in .x; y/ 2 M M.

7.5.3 The spectral partition function ZM as a trace

Next, let us examine what happens with the heat kernel on the diagonal y D x.
For a fixed " > 0, for all t  " and for all x 2 M , the series
X
.x; t/ 7! et k ek2 .x/
k1

is convergent and its limit is equal to E.x; x; t/. Consequently, for all t 2 Œ"; C1Œ
Z Z C1
X
E.x; x; t/ d Vg .x/ D et k ek2 .x/ d Vg .x/
M M
kD1
C1
X Z
D et k ek2 .x/ d Vg .x/;
M
kD1
C1
X C1
X
et k kek k2L2 D et k :
kD1 kD1

So, we can define the trace of the operator et g by

Z C1
X
 
ZM .t/ WD trace et g
D trace .Pt / D E.x; x; t/ d Vg .x/ D et k :
M
kD1
PC1
We conclude that the series of functions t 7! kD1 et k is uniformly and abso-
lutely convergent on Œ"; C1Œ.
 7.6 The Minakshisundaram–Pleijel expansion and the Weyl formula 159

7.6 The Minakshisundaram–Pleijel expansion and the

Weyl formula
A very important corollary of the heat kernel construction is the Minakshisundaram–
Pleijel expansion:

Theorem 7.6.1 (Minakshisundaram–Pleijel expansion). Let .M; g/ be a compact

Riemannian manifold of dimension n, and let us denote by E the heat kernel of
.M; g/. We have the expansion
1
E.x; x; t/  n u0 .x; x/ C u1 .x; x/t C    C uk .x; x/t k C : : : ;
t !0C .4 t/ 2

where the functions x 7! uk .x; x/ are smooth on M and depend only on the
curvature tensor and its covariant derivatives.

Hence, if for any integer k we denote

Z
ak WD uk .x; x/ d Vg .x/;
M

then
C1
X 1
ZM .t/ D et k  n a0 C a1 t C    C ak t k C    ;
t !0C .4 t/ 2
kD1

i.e., for any integer N > 0 and for all t > 0,

1 X
N
n
ZM .t/ D n ak t k C O t N  2 C1 :
.4 t/ 2
kD0

The computation of the coefficients ak is very difficult. Nevertheless, we have

(see [BGM]):
a0 D Vol.M; g/
and Z
1
a1 D Scalg d Vg :
6 M
In particular, if dim.M / D 2, the Gauss–Bonnet formula yields

a1 D .M /;
3
where .M / is the Euler–Poincaré characteristic of the surface M .
160 Chapter 7. Inverse problems in spectral geometry

The expression of a2 is also known:

Z
1
a2 D 2 jRj2  2 jRicj2 C 5Scal2g d Vg ;
360 M
where R is the Riemann curvature tensor.
Remark 7.6.2. In particular, for a closed surface M , we have:
Z
Vol.M; g/ .M / t
ZM .t/ D C C K 2 d Vg CP .t/t 2 ;
4 t 6 60 M
where t 7! P .t/ is a bounded function on M .
A fundamental corollary of the Minakshisundaram–Pleijel expansion is the
following fact: if we know the spectrum of .M; g/, then in particular we know
 the dimension of the manifold;
 the volume of the manifold;
 the integral of the scalar curvature Scalg over the manifold.
Hence we deduce

Corollary 7.6.3. If two Riemannian manifolds .M; g/ and .M 0 ; g 0 / are isospec-

tral, then:
(i) the dimensions of M and M 0 are equal.
(ii) The volumes of .M; g/ and .M 0 ; g 0 / are equal.
(iii) The integrals of the scalar curvature over .M; g/ and .M 0 ; g 0 / are equal.
In particular, if M and M 0 are surfaces, then .M / D .M 0 /.

In the case of surfaces, the dimension and the Euler–Poincaré characteristic

are topological invariants.
The coefficients ak for k  2 are very complicated; more details can be found
in the books [BGM] and [Bér8].
To finish, we note that the Karamata Tauberian theorem allows us to deduce
from Minakshisundaram–Pleijel expansion the famous Weyl asymptotic formula:

Theorem 7.6.4 (Weyl asymptotic formula). Let .M; g/ be a compact Riemannian

manifold of dimension n, and let k denote the eigenvalues of g on .M; g/.
Then
Bn Vol.M; g/ n
Card .fk 2 N; k  g/  2
!C1 .2/n
n
where Bn WD 2
. n
is the volume of the closed unit ball in .Rn ; can/.
2 C1/
 7.7 Two planar isospectral nonisometric domains 161

An obvious consequence of this asymptotics is the equivalent one

  n2
.2/n 2
k  kn:
k!C1 Bn Vol.M; g/

7.7 Two planar isospectral nonisometric domains

In this subsection we present a very simple counter-example for the isospectrality
problem. This example is based on P. Bérard’s transplantation method [Bér2],
[Bér3]. We largely follow [Bér1].

7.7.1 Construction of the domains D1 and D2

Let us consider the following elementary domain, called a brick:
A

@
@
@ B
@
@
Here A; B, and  are labels for pieces of the boundary. Consider also the following
graphs:
 
 2 A 1
A B  A  B A 
         
1 7 6 5 4@ 3 4 2 7 6@
B@@ B@@
 
3 5

where the integers f1; 2; : : : ; 7g on the vertices are labels for bricks and A; B; 
are labels for edges. Using seven bricks Pi , i 2 f1; 2; : : : ; 7g and following com-
binatorial the first graph, we glue the bricks along the labeled edges. For example:
we glue bricks P1 and P7 along the boundary A, we glue bricks P7 and P6 along
the boundary B, etc. . . . Using this we obtain the following planar domain D1 :
162 Chapter 7. Inverse problems in spectral geometry

 @
@
2  4 BB 3 @
@
@

A
A

 6


B
B

1 7
@
@
@ AA
@
@

Carrying out the same type of construction with the second graph, we get
another planar domain D2 .
 7.7 Two planar isospectral nonisometric domains 163

@
AA 7  6 AA 1 @@
@@
2

B B
B B

5
@
4  3 @
@
@@

Obviously, for the metric can on R2 we have

Proposition 7.7.1. The domains D1 and D2 are non-isometric.

7.7.2 Isospectrality of the domains D1 and D2

Now let us consider the Laplacian  on the domains D1 and D2 with, for example,
Neumann conditions (the Dirichlet case is very similar) and show that D1 and D2
are isospectral. For this let us introduce some notations and facts. First we define
the transplantation matrix
0 1
a a a a b b b
B a b a b a a b C
B C
B a a b b b a a C
B C
T WD B a b b a a b a C;
B C
B b a b a a a b C
@ b a a b a b a A
b b a a b a a
164 Chapter 7. Inverse problems in spectral geometry

where a; b 2 R are such that

8
< 4a2 C 3b 2 D 1
2a C 4ab C b 2 D 0
2
:
4a C 3b D 1:
An elementary computation establishes that the matrix T is orthogonal (i.e.,
T ? T D I7 ).
Next consider the sets Xj , j 2 f1; 2g, defined by
80 1 9
ˆ
< x1 >
=
B :: C
Xj WD @ : A I 8i 2 f1; 2; : : : ; 7g xi 2 L .Pi / and .?/j ;
2
:̂ x >
;
7

where
8 8
ˆ x 1 jA D x 7 jA ˆ x 4 j D x 3 j
ˆ
ˆ ˆ
ˆ
ˆ
ˆ x j D x 6 jB ˆ
ˆ x j D x 2 jB
< 7B < 4B
x 6 j D x 5 j x 2 jA D x 7 jA
.?/1 W and .?/2 W
ˆ
ˆ x 5 jA D x 4 jA ˆ
ˆ x 7 j D x 6 j
ˆ
ˆ x 4 j D x 2 j ˆ
ˆ x 6 jB D x 5 jB
:̂ :̂
x 4 jB D x 3 jB x 6 jA D x 1 jA
and xi jZ denotes the restriction of the function xi to the domain Z 2 fA; B;  g.
Now consider the maps j ; j 2 f1; 2g:
8  
ˆ
ˆ L2 Dj ! Xj
ˆ
ˆ
ˆ
< 0 1 0 1
j W '1 'jP1
ˆ
ˆ B : C B : C
ˆ ' 7! j .'/ WD @ :: A WD @ :: A :
ˆ
:̂ '7 'jP7

It is clear that 1 and 2 are isomorphisms.

Definition 7.7.2. The transplantation map between domains D1 and D2 is de-

fined by 8
< L2 .D1 / ! L2 .D2 /
T W
:
' 7! 21 .T:1 .'// :
Let us verify that the map T is well defined. Given any function ' 2 L2 .D1 /,
consider the vector 0 1
1
B C
D @ ::: A WD T  .'/;
7
 7.7 Two planar isospectral nonisometric domains 165

so we have D a'1 C a'2 C a'3 C a'4 C b'5 C b'6 C b'07 , etc. 1

0a 1 1
'1 1
B :: C B :: C
Since @ : A belongs to X1 one can easily verify that D @ : A belongs
'7 7
to X2 .

Proposition 7.7.3. The transplantation map is unitary .i.e., a surjective isometry/

from L2 .D1 / onto L2 .D2 /. In particular, H 1 .D1 / is isometric to H 1 .D2 /.

Proof. Since 2 ; T; 1 are isomorphisms, the map T is an isomorphism from

L2 .D1 / onto L2 .D2 /. Next for all ' 2 L2 .D1 / using the fact that the matrix T
is orthogonal we get

X
7
  X
7
kT'k2L2 .D2 / D T'jP 2 2 D k 2
k kL2 .Pk /
k L .P k/
kD1 kD1
Z Z
D kT 1 .'/k2R7 d Vg D k1 .'/k2R7 d Vg D k'k2L2 .D1 / :
D1 D1

Similarly,
krT'k2L2 .D2 / D kr'k2L2 .D1 /

for all ' 2 H 1 .D1 /; in particular,

kT'k2H 1 .D2 / D k'k2H 1 .D1 /

for all ' 2 H 1 .D1 /. To finish one can easily verify the equivalence

' 2 H 1 .D1 / ” T' 2 H 1 .D2 / :

Therefore, the map T induces an isometry from H 1 .D1 / onto H 1 .D2 /. 

Corollary 7.7.4. The domains D1 and D2 are isospectral.

Proof. Since the map T is bijective and

kT'k2L2 .D2 / D k'k2L2 .D1 /

and
krT'k2L2 .D2 / D kr'k2L2 .D1 /
166 Chapter 7. Inverse problems in spectral geometry

for all ' 2 H 1 .D1 / (see the previous proof), using the minimax principle we get
for each integer k

kr k2L2 .D2 /
k .D2 / D min max
E H 1 .D2 / 2E k k2L2 .D2 /
dim.E /Dk ¤0

krT'k2L2 .D2 /
D min max
E H 1 .D1 / '2E kT'k2L2 .D2 /
dim.E /Dk '¤0

kr'k2L2 .D1 /
D min max D k .D1 /: 
E H 1 .D1 / '2E k'k2L2 .D1 /
dim.E /Dk '¤0

7.8 Few words about Laplacian and conformal geome-

try
In this last Section we explain the relationship between the eigenvalues of a closed
surface and its associated conformal geometry. For more details, see, for example
[Cha] or [Gur].

7.8.1 Conformal geometry on surfaces

We begin with some basic facts on the conformal geometry on surfaces.

Definition 7.8.1. Let .M; g0 / be a surface. A metric h on M is said conformal to

g0 if there exists a smooth function u on M such that h D e2u g0 .
We define the conformal class Œg0  of the metric g0 on M by
˚ 
g D e2u g0 I u 2 C 1 .M / :
Œg0  WD e
For all vectors X; Y 2 Tx M (]g denotes the angle for the metric g) we have
  g0 .X; Y /
cos ]g0 X; Y D p p
g0 .X; X / g0 .Y; Y /
e2ue g .X; Y /
Dp p
e2ueg .X; X / e2ue g .Y; Y /
e
g .X; Y /
Dp p
e
g .X; X / eg .Y; Y /
 
D cos ]eg X; Y :
 7.8 Few words about Laplacian and conformal geometry 167

Therefore, angles are conformally invariant: if we change the metric g0 on a sur-

face to a conformal metric e
g D e2u g0 , the length of vectors and curves changes,
but the angle between two vectors is unchanged:

]g0 X; Y D]e
g X; Y

for all x 2 M and for all vectors X; Y 2 Tx M . In fact the geometries of .M; g0 /
and .M;e g / are the sames up to a change of scale.
When we change the initial metric g0 to another conformal metric g the cur-
vature changes according to the Gaussian curvature equation:

Proposition 7.8.2. For a conformal metric e

g D e2u g0 on M we have the local
relation
g0 u C Kg0 D Ke 2u
ge :

One of the main questions in conformal geometry is: given a closed surface M
with a fixed metric g0 and for a smooth function KW M ! R, does there exist a
conformal metric e g D K? In other words, does there exist a smooth
g such that Ke
function u such that
g0 u C Kg0 D Ke2u ‹
The answer is yes: using complex analysis, A. Weil solved this problem for Rie-
mann surfaces.

7.8.2 Spectral zeta function and uniformization theorem

on surfaces
Now, let us define the spectral zeta function associate to a surface .M; g0 / by the
formula
C1
X 1
g0 .s/ WD s
;
nD1
n

where s 2 C is such that Re.s/ > 1. Then we have

C1
X C1
X
g0 .s/ D s
n D es ln.n / :
nD1 nD1

Using the Weyl formula, i.e.,

n  C n;
n!C1

where
4 2
C WD > 0;
B2 Vol.M; g/
168 Chapter 7. Inverse problems in spectral geometry
P PC1 1
we see that the series C1 1
nD1 sn and nD1 ns have the same nature. The series
PC1 1
nD1 ns is the usual Riemann function which is convergent for all s 2 C such
P
that Re.s/ > 1. So the series g0 .s/ WD C1 1
nD1 sn is well defined and convergent
for all s 2 C such that Re.s/ > 1.
Now we shall prove that the function g0 admits a meromorphic continuation
to C with a simple pole at s D 1. First recall that  the Euler Gamma function
Z C1
.s/ WD et t s1 dt
0

converges for all s 2 C such that Re.s/ > 0. Moreover, for all s 2 C in the
indicated range,
1 nŠns
D lim :
.s/ n!C1 s.s C 1/.s C 2/    .s C n/
The function  admits a meromorphic continuation to C n f0; 1; 2; : : :g with
simple poles at s D 0; 1; 2; : : : . By changing variables we have for all x > 0
Z C1
1 1
D ext t s1 dt:
xs .s/ 0
Therefore, for all s 2 C such that Re.s/ > 1 we get
C1
X Z C1
1
g0 .s/ D en t t s1 dt
nD1
.s/ 0
Z C1 C1 !
1 X
n t
D e t s1 dt
.s/ 0 nD1
Z C1
1
D .ZM .t/  1/ t s1 dt
.s/ 0
because 0 D 0. Thus,
Z 1 Z C1
1 1
g0 .s/ D .ZM .t/  1/ t s1 dt C .ZM .t/  1/ t s1 dt:
.s/ 0 .s/ 1
We start by studying the first integral. Using Remark 7.6.2, we have for all s 2 C
such that Re.s/ > 1
Z 1
1
.ZM .t/  1/ t s1 dt
.s/ 0
Z 1 Z 
1 Vol.M; g/ .M / t
D C C K 2 d Vg CP .t/t 2  1 t s1 dt
.s/ 0 4 t 6 60 M
 7.8 Few words about Laplacian and conformal geometry 169
Z 1 Z 
1 Vol.M; g/ .M / t
D C C K d Vg  1 t s1 dt
2
.s/ 0 4 t 6 60 M
Z 1
1
C P .t/t sC1 dt:
.s/ 0
Z 1
The function s 7! .s/
1
P .t/t sC1 dt is holomorphic for Re.s/ > 2 and
0
Z 1 Z 
1 Vol.M; g/ .M / t
C C K d Vg  1 t s1 dt
2
.s/ 0 4 t 6 60 M
 s1 Z t D1
1 t Vol.M; g/ .M /t s  t sC1 ts
D C C K d Vg 
2
:
.s/ s  1 4 6s 60.s C 1/ M s t D0
This quantity converges for all s 2 C such that Re.s/ > 1 and has a meromorphic
continuation to C with a simple pole in s D 1 (this follows from the relation
.s/ D limn!C1 s.sC1/.sC2/ nŠns
.sCn/
).
To finish, let us study the second integral. For all s 2 C such that Re.s/ > 1,
Z C1 Z C1 C1 !
1 1 X
n t
.ZM .t/  1/ t s1
dt D e t s1 dt;
.s/ 1 .s/ 1 nD1

and for all t  1, en t D e1 t e.1 n /t , so since the eigenvalues n grows like
n as n tends to infinity (Weyl asymptotic formula) and since 0 < 1  2    
n , it follows that for t large enough
C1
X
en t  C e1 t :
nD1

Hence, using the holomorphic theorem

! under integral sign we get easily that the
Z C1 C1
X
function s 7! .s/
1
en t t s1 dt is holomorphic on C. Finally, the
1 nD1
function g0 has a meromorphic extension to C with a simple pole in s D 1. In
particular, g0 is analytic in 0, hence the value g0 0 .0/ exists and

g0 .s/  g0 .0/

g0 0 .0/ D lim :
s!0 s0
Next, the derivative of g0 is
C1
X 1
g0 0 .s/ D  ln.i / ;
si
i D1
170 Chapter 7. Inverse problems in spectral geometry

so for s D 0 we get

C1
!
X Y
n
g0 0 .0/ D ln.i / D  ln i :
i D1 i D1

Therefore,
C1
Y
0  
eg0 .0/ D i D det g0 :
i D1

Thus we can define the determinant of g0 by the formula :

  0
det g0 D eg0 .0/ :

Now, let g 2 Œg0 , so there exists u 2 C 1 .M / such that g D e2u g0 . The Polyakov
formula (see for example [Cha]) says that
 ! Z
det g 1
ln   D kruk2 C Kg0 u d Vg0 :
det g0 12 M

We can consider the functional

8
ˆ
ˆ C 1 .M / ! R
<
F W  
ˆ u 7! ln det. g / :
:̂ det. g / 0

Since the determinant is not a scale invariant we consider the normalized func-
tional determinant
8
ˆ
ˆ C 1 .M / ! R
<
S W  
ˆ u 7! 12 ln det. g / C 2 .M / ln .Vol.M; g// ;
:̂ det. g / 0

where .M / is the Euler characteristic of M . The critical functions of S are

functions u such that
g0 u C Kg0 D ce2u ;
where c is a real constant. Using the Gaussian curvature equation we see that u
is a critical function of S if and only if the surface M with the conformal met-
 7.8 Few words about Laplacian and conformal geometry 171

ric g WD e2u g0 has a constant Gaussian curvature Kg . In 1998, B. Osgood, R.

Phillips, and P. Sarnak [OPS1], [OPS2] proved the following theorem:

Theorem 7.8.3 (Osgood, Philipps, Sarnak, 1998). The supremum supu2C 1.M/ S.u/
of the above functional exists and it is a maximum: there exists u0 2 C 1 .M / such
that Gaussian curvature Kg (where g WD e2u g0 ) is constant on M .

It is turn out that each surface can be endowed with a metric conformally
equivalent to a metric with constant Gaussian curvature. A main consequence is
the famous Poincaré–Klein–Koebe uniformization theorem.

Theorem 7.8.4 (Uniformization Theorem). Let .M; g0 / be a closed surface with

a Riemannian metric g0 . Then there exists an unique metric7 g conformal to g0
on M with a constant Gaussian curvature Kg 2 fC1; 0; 1g.

The values of the constant Kg 2 fC1; 0; 1g depends on the topology of the
surface M . Recall that the Gauss–Bonnet formula for a closed surface reads
Z
K d Vg D 2 .M /;
M

and for a surface with boundary

Z Z
K d Vg C Kg dSg D 2 .M /;
M M

where .M / is the Euler–Poincaré characteristic of M and Kg is the geodesic

curvature on the boundary. According to the Gauss–Bonnet formula, the sign of
the curvature is determined by the Euler–Poincaré characteristic8 .M / of M .
Therefore, the universal covering of every closed surface can be isometrically
embedded in
1. the round sphere S2 for genus zero with .M / D 2 > 0 (in this case
Kg D C1).
2. the Euclidean plane R2 for genus one with .M / D 0 (in this case Kg D
0).
3. the hyperbolic space H2 for genus g  2 with .M / D 2 < 0 (in this
case Kg D 1).
7 Called a uniformization metric.
8 For M a closed orientable surface .M / D 2  2g, where g is the genus of the surface.
172 Chapter 7. Inverse problems in spectral geometry

K=1

K=0

K=-1

7.8.3 Ricci flow on surfaces

At the beginning of the 1980’s R. Hamilton [Ham] introduced a nice tool for
studying topological spaces in low dimensions. This geometric tool is called the
Ricci flow. In the case of dimension 2 (surface) the Ricci flow on a fixed surface
.M; g0 / is given by the non-linear PDE

@g
D 2Kg.t / g.t/;
@t
with g.0/ D g0 .
R. Hamilton also introduced a variant of this equation, that for the so-called
normalized Ricci flow:
 
@g 4 .M /
D  2Kg.t / g.t/
@t Vol.M; g0 /

with g.0/ D g0 .
This version of Hamilton’s surface Ricci flow preserves the total area, more-
over, the Ricci flow converges to a metric with constant Gaussian curvature and
for all t the metric g.t/ is conformal to the initial metric. Indeed, we have
 7.8 Few words about Laplacian and conformal geometry 173

Theorem 7.8.5 (Hamilton, 1988). For a closed two-dimensional Riemannian man-

ifold .M; g0 / such that .M /  0 there exists a unique solution g.t/ of the nor-
malized Ricci flow and this solution converge (in the Whitney topology) to a metric
of constant Gaussian curvature and conformal to g0 .

In the case of positive Euler–Poincaré characteristic we have:

Theorem 7.8.6 (Chow, 1991). For a closed two-dimensional Riemannian mani-

fold .M; g0 / such that .M /  0 there exists a unique solution g.t/ of the nor-
malized Ricci flow and this solution converge (in the Whitney topology) to a metric
of constant Gaussian curvature and conformal to g0 .

These theorems are fundamental, but they are not a proof of the uniformization
theorem using Ricci flow. In fact the arguments for convergence in the proofs of
R. Hamilton and B. Chow use the uniformization theorem. However in 2006,
X. Chen, P. Lu and G. Tian gave a proof of the uniformization theorem based on
Hamilton’s surface Ricci flow.
Remark 7.8.7. The Ricci flow on surfaces has many applications in computer
vision (shape matching, medical imaging, visualization and design in computer
games). For an introduction to the Ricci flow on surfaces see Chapter 5 in the
book of B. Chow [Ch-Kn]. Nice applications of Ricci flow on surfaces can be
also found in the book of X. D Gu and S. T Yau [Gu-Ya].

7.8.4 What about three manifolds?

In dimension n D 2 the situation is very simple: indeed, by the Uniformiza-
tion Theorem 7.8.4, every manifold admits a metric of constant sectional curva-
ture. Thus in dimension n D 2 every geometry9 is isotropic10. The situation
for dimension 3 (and beyond) is more complex. The geometric classification of
three-dimensional manifolds was conjectured by W. Thurston in the end of 70’s.
Thurston’s Geometrization Conjecture asserts that every three-dimensional man-
ifold admits a canonical topological decomposition into elementary “geometric”
pieces11. “Geometric” pieces mean here that every pieces can be endowed with
one of the eight homogeneous Thurston geometries:
9 A geometry X is a homogeneous unimodular complete and simply-connected Riemannian manifold.

Homogeneous means that the isometry group Isom.X/ acts transitively on X. Homogeneous means also
that the local geometry of the manifold is the same at all points. Unimodular means that the quotient
X=Isom.X/ admits a finite volume.
10 A manifold X is isotropic if the isometry group Isom.X/ acts transitively on the unit tangent bundle.

Isotropic means that the local geometry of the manifold is the same in all directions. A geometry X is
isotropic if and only if its sectional curvature is constant.
11 The existence of this decomposition is based on the Kneser sphere decompostion and on the Jaco–

Shalen–Johannson torus decomposition.

174 Chapter 7. Inverse problems in spectral geometry

1. Spherical isotropic geometry, e.g., S3 , PR3 ; Poincaré’s sphere, lens space

...
2. Euclidean isotropic geometry, e.g., Euclidean space E3 , three-torus T3 . . .
3. Hyperbolic isotropic geometry, e.g., Seifert–Weber dodecahedral space . . .
4. Trivial product S2 E1 geometry, e.g., S2 S1 ; S2 I (with I an interval)
...
5. Trivial product H2 E1 geometry, e.g., H2 S1 . . .
6. Nil geometry.
B
7. SL2 .R/ geometry.
8. Sol geometry.

Conjecture 7.8.8 (Thurston’s Geometrization Conjecture12). The interior of every

oriented compact 3-manifold can be split (along 2-spheres and 2-tori) into a finite
number of pieces, such that each piece is geometric.

An important corollary of the Geometrization Conjecture is the so-called Ellipti-

sation Conjecture, which asserts that every compact orientable 3-manifold with
a finite fundamental group has a Riemannian metric of constant positive sectional
curvature:

Conjecture 7.8.9 (Elliptisation Conjecture). Every oriented closed 3-manifold

with a finite fundamental group is spherical13.

In particular the elliptisation conjecture is true for any compact and simply
connected three-manifold. A consequence is the famous Poincaré Conjecture
(1904) [Poi]:

Conjecture 7.8.10 (Poincaré Conjecture, 1904). If M is a compact and simply

connected three-manifold, then M is homeomorphic to the round sphere S3 .

In fact, W. Thurston proved the Geometrization Conjecture for a specific class

of manifolds14. For more details on 3-manifolds and geometrization see for ex-
ample the notes of W. Thurston [Thu], the book of J. Weeks [Wee] and the book
of M. Boileau, S. Maillot and J. Porti [BMP].
12 We present here a simplified version of the conjecture; for details see [BBBMP].
13 A manifold M is called spherical if M admits a Sn - geometric model structure, i.e., M is diffeomor-
phic to the quotient of Sn by a discrete sub-group acting freely on Sn .
14 Haken 3-manifolds.
 7.8 Few words about Laplacian and conformal geometry 175

Hamilton’s Ricci flow on a fixed 3-manifold .M; g0 / is the one-parameter fam-

ily of smooth Riemannian metrics on .M; g0 / satisfying the non-linear PDE
@g
D 2Ricg.t /
@t
with g.0/ D g0 . Note that if we use harmonic local coordinates the Laplace–
Beltrami operator appears in the Ricci flow equation
@gij
D g.t / gij C Qij .g; @g/;
@t
where Q is a quadratic expression.
In fact the Ricci flow is very similar to the heat equation, but here we are not
interested in the diffusion of heat, but in the diffusion of curvature. The Ricci
flow process is a sort of uniformization process of curvature on the manifold.
R. Hamilton proved that every compact 3-manifold with strictly positive Ricci
curvature has a Riemannian metric of constant strictly positive sectional curvature.
Hence, if the manifold is simply connected, then M is diffeomorphic to the round
sphere S3 . Without the Ricci strict positivity hypothesis, the behaviour of the Ricci
flow is more complex and singularities appear during the Ricci flow evolution, so
the idea of R. Hamilton is to stop the flow, make surgery around the singularities,
and restart the flow after the surgery (this is called the Ricci flow with surgery).
But unfortunately this method requires to control the number of surgeries and
understand precisely the nature of these singularities.
In 2002–2003, G. Perelman [Per1], [Per2], [Per3] completely solved this prob-
lem. More precisely, he proved that it is possible to make a finite number of surg-
eries in a finite time interval, and also that if the fundamental group 1 .M / is
finite (as is the case for the Poincaré conjecture), then the Ricci flow with surgery
extincts after a finite time (i.e., the manifold shrinks to a point). In the case where
1 .M / is infinite, the Ricci flow is defined on the interval Œ0; C1Œ and G. Perel-
man analyzed the behaviour of the Ricci flow with surgery for t ! C1 (he
used a thin–thick decomposition). He was then able to prove the Geometrization
Conjecture and the Poincaré Conjecture.
For more details on this amazing story using Perelman’s arguments see for ex-
ample the surveys [Bes1], [Bes2], the article by B. Kleiner and J. Lott [Kl-Lo1],
the book by J. Morgan and G. Tian [Mo-Ti], the article by H.D. Cao and X.P. Zhu
[Ca-Zh] and the book by B. Chow and D. Knopf [Ch-Kn]. There exist alterna-
tives approaches to some of Perelman’s arguments, see for example the articles
by T. Colding and W. Minicozzi [Co-Mi], T. Shioya and T. Yamaguchi [Sh-Ya],
J. Cao and J. Ge [Ca-Ge], B. Kleiner and J. Lott [Kl-Lo2]. See also the recent book
by L. Bessières, G. Besson, M. Boileau, S. Maillot and J. Porti [BBBMP]. In that
monograph the authors give a complete proof of the Geometrisation Conjecture
with an alternative approach to Perelman: they replace Perelman’s Ricci flow by
a variant called Ricci flow with bubbling-off . Perelman’s Ricci flow with surgery
176 Chapter 7. Inverse problems in spectral geometry

and the Ricci flow with bubbling-off are similar, but in the case of Ricci flow
with bubbling-off the metric evolves on a fixed manifold rather than an evolving
manifold. One other major difference is that in the Ricci flow with bubbling-off
process the surgery occurs before the flow becomes singular, rather than exactly
at the time the singularity is formed (see [BBBMP]).
Bibliography

[Ann] C. A NNÉ , Problème de la glace pilée, Séminaire de théorie spectrale et géométrie

de Grenoble 3: 1–13, 1984–1985.
[Arv1] W. A RVERSON, A Short Course on Spectral Theory, Graduate Texts in Mathemat-
ics, Springer, 2001.
[Arv2] W. A RVERSON, An Invitation to C*-Algebras, Graduate Texts in Mathematics,
Springer, 1976.
[Bar] V. BARGMANN, On a Hilbert space of analytic functions and an associated integral
transform, Comm. Pure. Appl. Math. 19: 187–214, 1961.
[Ba-We] S. BATES & A. W EINSTEIN, Lectures on the Geometry of Quantization, Berkeley
Mathematics Lecture Notes, 1997.
[Bér1] P. B ÉRARD, Domaines plans isospectraux à la Gordon-Webb-Wolpert: une preuve
terre à terre, Séminaire de Théorie Spectrale et Géométrie, 131–142, 1991–1992.
[Bér2] P. B ÉRARD, Transplantation et isospectralité I, Math. Ann. 292: 547–559, 1992.
[Bér3] P. B ÉRARD, Transplantation et isospectralité II, J. London Math. Soc. 48: 565–576,
1993.
[Bér4] P. B ÉRARD, The isospectral problem for Riemannian manifolds, Cours de DEA,
Grenoble 1993. http://www-fourier.ujf-grenoble.fr/~pberard/D/isos-dea93.pdf
[Bér5] P. B ÉRARD, On ne peut pas entendre la forme d’un tambour: Introduction, Auxerre
2001. http://www-fourier.ujf-grenoble.fr/~pberard/D/tambours.pdf
[Bér6] P. B ÉRARD, On ne peut pas entendre la forme d’un tambour: Exposé 1, Auxerre
2001. http://www-fourier.ujf-grenoble.fr/~pberard/D/auxerre1.pdf
[Bér7] P. B ÉRARD, On ne peut pas entendre la forme d’un tambour: Exposé 2, Auxerre
2001. http://www-fourier.ujf-grenoble.fr/~pberard/D/auxerre2.pdf
[Bé-Be] P. B ÉRARD & M. B ERGER, Le spectre d’une variété riemannienne en 1982, Kaigai
(Tokyo), 1983.
[Bér8] P. B ÉRARD, Spectral Geometry: Direct and Inverse Problems, Lecture Notes in
Mathematics, Vol. 1207, Springer-Verlag, Berlin, 1987.
[BBG] P. B ÉRARD, G. B ESSON & S. G ALLOT , Embedding Riemannian manifolds by their
heat kernels, Geom. Funct. Anal. 4, 232-1, 373–398, 1994.
[Bé-He] P. B ÉRARD & B. H ELFFER, Dirichlet eigenfunctions of the square membrane:
Courant’s property, and A. Stren’s and A. Pleijel analyses, available online at
arXiv:1402.6054, 2014.
[BGM] M. B ERGER, P. G AUDUCHON & E. M AZET , Le Spectre d’une Variété Riemanni-
enne, Lectures Notes in Mathematics 194, Springer-Verlag 1971.
[BGV] N. B ERLINE , E. G ETZLER & M. V ERGNE , Heat Kernels and Dirac Operators,
Springer-Verlag, 1992.
178 Bibliography

[Be-Co] J. B ERTRAND & B. C OLBOIS, Capacités et inégalités de Faber-Krahn dans

l’espace euclidien, J. of Functional Anal., 232-1, 1–28, 2006.
[Ber] M. B ERGER, A panoramic view of Riemannian Geometry, Springer, 2003.
[Bes1] L. B ESSIÈRES, Conjecture de Poincaré: la preuve de R. Hamilton et G. Perelman.
Gaz. Math. 106: 7–35, 2005.
[Bes2] L. B ESSIÈRES, Poincaré conjecture and Ricci flow. An outline of the work of R.
Hamilton and G. Perelman. EMS Newsletter 59: 11–15, 2006.
[Bes] G. B ESSON, Sur la multiplicité de la première valeur propre des surfaces riemanni-
ennes, Ann. Inst. Fourier. 30: 109–128, 1980.
[BBBMP] L. B ESSIÈRES, G. B ESSON, M. B OILEAU, S. M AILLOT and J. P ORTI, Geometrisa-
tion of 3-manifolds, Tracts in Mathematics, European Mathematical Society, 2010.
[BMP] M. B OILEAU, S. M AILLOT & J. P ORTI, Three-dimensional orbifolds and their Ge-
ometric Structures. Panor. Synthèses 15, Soc. Math. France, Paris 2003.
[Bou] N. B OURBAKI, Théorie Spectrale, Hermann, Paris, 1970.
[Bré] H. B RÉZIS, Functional Analysis, Sobolev Spaces and Partial Differential Equa-
tions, Universitext, Springer, 2010.
[Bu-Be] P. B UTZER & H. B ERENS, Semi-Groups of Operators and Approximation, Die
Grundlehren der mathematischen Wissenschaften, 145, Springer-Verlag, 1967.
[Ca-Zh] H. D. C AO and X. P. Z HU, A complete proof of the Poincaré and geometrization
conjectures – application of the Hamilton-Perelman theory of the Ricci flow. Asian
J. Math. 10: 165–492, 2006.
[Ca-Ge] J. C AO & J. G E , A proof of Perelman’s collapsing theorem for 3-manifolds,
arXiv:0908.3229, Preprint 2009.
[Car] P. C ARTIER, Introduction aux Problèmes Mathématiques de la Mécanique Quan-
tique, Strasbourg: Département de Mathématique, 1971.
[Cha] S. Y. A. C HANG, Non-linear Elliptic Equations in Conformal Geometry, Zürich
Lectures in Advanced Mathematics, European Mathematical Society, 2004.
[Chap] S. J. C HAPMAN, Drums that sound the same, Amer. Math. Monthly, 124–138, 1995.
[Cha1] I. C HAVEL , Eigenvalues in Riemannian Geometry, Academic Press, Orlando, 1984.
[Cha2] I. C HAVEL , Riemannian Geometry: a Modern Introduction, Cambridge Tracts in
Mathematics, Cambridge University Press, Cambridge, 1993.
[Ch-Fe1] I. C HAVEL & E. F ELDMAN, Spectra of domains in compact manifolds, J. Funct.
Anal. 30: 198–222, 1978.
[Ch-Fe2] I. C HAVEL & E. F ELDMAN, Spectra of manifolds less a small domain, J. Duke
Math J. 56: 399–414, 1988.
[Che] S. Y. C HENG, Eigenfunctions and nodal sets, Comment. Math. Helv. 51: 43–55,
1979.
[Ch-Kn] B. C HOW & D. K NOPF, The Ricci Flow: An Introduction, Amer. Mathematical
Soc., Providence R.I., 2004.
[Co-La] R. R. C OIFMAN & D. L AFON, Diffusion maps, Appl. Comput. Harmo. Appli. 21:
5–30, 2006.
 Bibliography 179

[Co-Mi] T. H. C OLDING & W. P. M INICOZZI, Width and finite extinction time of Ricci flow.
Geom. Topol. 12: 2537–258, 2008.
[Col1] Y. C OLIN DE V ERDIÈRE , Spectre du Laplacien et longueurs des géodésiques péri-
odiques I, Compositio Mathematica, 27: 80–106, 1973.
[Col2] Y. C OLIN DE V ERDIÈRE , Spectre du Laplacien et longueurs des géodésiques péri-
odiques II, Compositio Mathematica, 27: 159–184, 1973.
[Col3] Y. C OLIN DE V ERDIÈRE , Spectrum of the Laplace operator and periodic geodesics:
thirty years after, Ann. Inst. Fourier. 57(7): 2429–2463, 2008.
[Col4] Y. C OLIN DE V ERDIÈRE , Construction de laplaciens dont une partie finie du spectre
est donnée, Ann. Sci. Ec. Norm. Sup. Paris 20: 599–615, 1987.
[Co-Ro] M. C OMBESCURE & D. ROBERT , A phase study of the quantum Loschmidt Echo in
the semi-classical limit, [arXiv : quant-ph0510151], 2005.
[Coura] R. C OURANT , Eim allgemeiner Satz zur Theorie der Eigenfunktionen selbstad-
jungierter Differentialausdrücke, Nachr. Ges. Gottingen 81–84, 1923.
[Co-Hi] R. C OURANT & D. H ILBERT , Methods of Mathematical Physics, Intersciences Pub-
lishers, New York, 1953.
[Cou] G. C OURTOIS, Spectrum of manifolds with holes, J. Funct. Anal. 134: 194–221,
1995.
[Dav] E. B. DAVIES, Heat Kernels and Spectral Theory, Cambridge University Press,
1990.
[Dav2] E. B. DAVIES, Spectral Theory and Differential operators, Cambridge University
Press, 1995.
[Di-Sj] M. D IMASSI & J. S JÖSTRAND, Spectral Asymptotics in the Semi-classical Limit,
London Math Society Lectures Note Series 268, 1999.
[DoC] M. DO C ARMO, Riemannian Geometry, Birkhäuser, Basel, 1992.
[Dod1] J. D ODZIUK, Eigenvalues of the Laplacian and the heat equation, Amer. Math.
Monthly 88(9): 686–695, 1981.
[Dod2] J. D ODZIUK, Eigenvalues of the Laplacian on forms, Proc. Amer. Math. Soc 85(9):
438–443, 1982.
[DRh] G. DE R HAM, Variétés Différentiables: Formes, Courants, Formes Harmoniques,
Paris, Hermann, 1960.
[Ev-Zw] L. E VANS & M. Z WORSKI, Lectures on Semiclassical Analysis. http://math.
berkeley.edu/~evans/semiclassical.pdf.
[Fa-Ts1] F. FAURE & M. T SUJII, Band structure of the Ruelle spectrum of contact Anosov
flows, Comptes rendus – Mathématique 351, 385–391, 2013.
[Fa-Ts2] F. FAURE & M. T SUJII, The semiclassical zeta function for geodesic flows on neg-
atively curved manifolds, preprint 2013.
[Ge-Zh] F. G ESZTESY & Z. Z HAO, Domain perturbations, Browian motion, capacities, and
ground states of Dirichlet Schrödinger operator, Math. Z. 215: 143–250, 1994.
[GHL] S. G ALLOT , D. H ULLIN & J. L AFONTAINE , Riemannian Geometry, Universitext
Springer-Verlag, Berlin, 2004.
180 Bibliography

[Gor] C. G ORDON, When you can’t hear the shape of a manifold, The Mathematical In-
telligencer 11-3, 39–47, 1989.
[Go-Wi] C. G ORDON & E. N. W ILSON, Isospectral deformations on compact manifolds, J.
Diff. Geom. 19: 241–256, 1984.
[GWW1] C. G ORDON, D. W EBB & S. W OLPERT , Isospectral plane domains and surfaces via
Riemannian orbifolds, Inventiones Mathematicae 110: 1–22, 1992.
[GWW2] C. G ORDON, D. W EBB & S. W OLPERT , One cannot hear the shape of a drum,
Bulletin AMS 27: 134–138, 1992.
[Gi-Tr] S. G ILBARD & N.S. T RUDINGER, Elliptic Partial Differential Equations of Second
Order, Springer-Verlag Berlin, 1983.
[Gr-Sj] A. G RIGIS &. J. S JÖSTRAND, Microlocal Analysis and Differential Operators, an
Introduction, London Math Society Lectures Note Series 196, Cambridge Univ.
Press, 1994.
[Gri] A. G RIGOR ’ YAN, Heat Kernels on Riemannian Manifolds in Spectral Theory and
Geometry, London Math Society Lectures Note Series 273, Cambridge Univ. Press,
1999.
[Grot] A. G ROTHENDIECK, Récoltes et Semailles, 1986.
[Gu-Si] S. J. G USTAFSON & I. M. S IGAL , Mathematical Concepts of Quantum Mechanics,
Springer, 2003.
[Gu-Ya] X. D. G U & S. T. YAU, Computational Conformal Geometry, Advanced Lectures
in Mathematics Vol. 3, International Press, 2008.
[Gur] M. G URSKY, PDEs in Conformal Geometry, Geometric analysis and PDEs, Lecture
Notes in Mathematics, Springer-Verlag, Berlin, 2009.
[Ham] R. H AMILTON, Three-manifolds with positive Ricci curvature, J. Differential
Geom. 17(2): 255–306, 1982.
[Heb] E. H EBEY, Nonlinear Analysis on Manifolds: Sobolev Spaces and Inequalities,
Courant Institute Lectures notes, 5, 1999.
[Hoc] H. H OCHSTAD, Integral Equations, Wiley, New York, 1973.
[Jos] J. J OST , Riemannian Geometry and Geometric Analysis, Universitext, Springer,
Berlin-Heidelberg, 1995.
[Kac1] M. K AC, Can one hear the shape of a drum?, Amer. Math. Monthly 73(4): 1–23,
1966.
[Kac2] M. K AC, Probabilistic methods in some problems of scattering theory, Rocky
Mountain, J. Math 4: 511–538, 1974.
[Kat] T. K ATO, Pertubation Theory for Linear Operators, Grundlehren der mathematis-
chen Wissenschaften 142, Springer, 1980.
[Kat1] T. K ATO, Schrodinger operators with singular potentials, Israel J. Math. 13: 135–
148, 1972.
[Kl-Lo1] B. K LEINER and J. L OTT , Notes on Perelman’s papers. Geom. Topol. 12: 2587–
2855, 2008.
 Bibliography 181

[Kl-Lo2] B. K LEINER and J. L OTT , Locally collapsed 3-manifolds, arXiv:1005.5106,

Preprint 2010.
[Ko-Sh1] V. A. KONDRAT ’ EV & M.A. S HUBIN, Discreteness of the spectrum for Schrödinger
operators on manifolds of bounded, geometry, in: Proceedings of the conference
“Functional Analysis, Partial Differential Equations and Applications” dedicated to
the V. G. Maz’ya 60th birthday, Rostock, 1998.
[Ko-Sh2] V. A. KONDRAT ’ EV & M.A. S HUBIN, Conditions for the discreteness of the spec-
trum for Schrödinger operators on manifolds, Funct. Anal. and Appl. 33, 1999.
[Lab1] O. L ABLÉE , Semi-classical behaviour of Schrödinger’s dynamics: Revivals of wave
packets on hyperbolic trajectory, Asymptotic Analysis, vol. 71, 59–99, 2011.
[Lab2] O. L ABLÉE , Quantum revivals in two degrees of freedom integrable systems: the
torus case. Asymptotic Analysis, vol. 77, 1–41, 2012.
[Lab3] O. L ABLÉE , Eigenvalues of Schrödinger operator on a closed Riemannian manifold
with holes, Functional Analysis, Approximation and Computation, vol. 5.2, 43–58,
2013.
[Le-Br] P. L EVY-B RUHL , Introduction à la Théorie Spectrale, Dunod, 2003.
[Mar] A. M ARTINEZ , An Introduction to Semiclassical and Microlocal Analysis, Springer,
2001.
[Mo-Ti] J. M ORGAN & G. T IAN, Ricci Flow and Poincaré Conjecture, Clay Mathematics
Institute, 2006.
[Mil] J. M ILNOR, Eigenvalues of the Laplace operator on certain manifolds, Proceedings
of the National Academy of Sciences of the United States of America 51: 542,
1964.
[Nad] N. NADIRASHVILI, Multiple eigenvalues of the Laplace operator, Math. USSR
Sbornik, 61: 225–238, 1988.
[Ole1] I. M. O LEINIK, On the essential self-adjointness of the Schrödinger operators on a
complete Riemannian manifold, Math. Notes. 54: 934–939, 1993.
[Ole2] I. M. O LEINIK, On the connection of the classical and quantum mechanical com-
pleteness of a potential at infinity on complete Riemannian manifolds, Math. Notes.
55: 380–386, 1994.
[Ole3] I. M. O LEINIK, On the essential self-adjointness of the Schrödinger-type operators
on a complete Riemannian manifold, PhD thesis, Northeastern University, 1997.
[OPS1] B. O SGOOD, R. P HILLIPS & P. S ARNAK, Compact isospectral sets of surfaces, J.
Funct. Anal. 80: 212–234, 1988.
[OPS2] B. O SGOOD, R. P HILLIPS & P. S ARNAK, Extremals of determinants of laplacians,
J. Funct. Anal. 80: 148–211, 1988.
[Per1] G. P ERELMAN, The entropy formula for the Ricci flow and its geometric applica-
tions. ArXiv: math.DG/0211159.
[Per2] G. P ERELMAN, Ricci flow with surgery on three-manifolds. ArXiv:
math.DG/0303109.
[Per3] G. P ERELMAN, Finite extinction time for the solutions to the Ricci flow on certain
threemanifolds. ArXiv: math.DG/0307245.
182 Bibliography

[Ple] A. P LEIJEL , Remarks on Courant’s nodal theorem. Comm. Pure. Appl. Math. 9:
543–550, 1956.
[Poi] H. P OINCARÉ , Analysis Situs, Cinquième complément à l’analysis Situs, Rend.
Circ. mat. Palerma 18, 45–110, 1904.
[Pr-We] M. P ROTTER & H. W EINBERGER, Maximum Principles in Differential Equations,
Springer-Verlag, 1984.
[Ra-Ta] J. R AUCH & M. TAYLOR, Potential and scattering on widly perturbed domains, J.
Funct. Anal. 18: 27–59, 1975.
[Re-Si] M. R EED & B. S IMON, Methods of Modern Mathematical Physics, Academic
Press, 1975.
[Rob] D. ROBERT , Autour de l’Approximation Semi-classique, volume 68 of Progress in
Mathematics, Birkhäuser, 1987.
[Rob2] D. ROBERT , Revivals of wave packets and Bohr-Sommerfeld quantization rules, in:
Adventures in Mathematical Physics, 219–235, Contemp. Math., 447, Amer. Math.
Soc., Providence, RI, 2007.
[Rob3] D. ROBERT , Propagation of coherent states in quantum mechanics and applica-
tions, proc. CIMPA, 2005.
[Robi1] R. W. ROBINNET , Quantum wave packet revivals, Physics Reports. 392: 1–119,
2005.
[Robi2] R. W. ROBINNET , Wave packets revivals and quasirevivals in one-dimesionnal
power law potentials, Journal of Mathematical Physics, vol. 41, 2000.
[ROS] S. ROSENBERG, The Laplacian on a Riemannian Manifold, London Mathematical
Society, Student Texts 31, Cambridge Univ. Press, 1997.
[SaC] L. S ALOFF -C OSTE , The heat kernel and its estimates, in: Advanced Studies in Pure
Mathematics, 8: 13, pp. 1–32, 2009.
[Sak] T. S AKAI, Riemannian Geometry, vol. 149 of Translations of Mathematical Mono-
graphs. American Mathematical Society, Providence, RI, 1996.
[Ser] J. P. S ERRE , Cours d’Arithmetique, Presses Universitaires de France, 1970.
[Sev] B. S EVENNEC, Majoration topologique de la multiplicité du spectre des surfaces,
Séminaire de Théorie Spectrale et Géométrie de Grenoble, 12: 29–35, 1993–1994.
[Sh-Ya] T. S HIOYA & T. YAMAGUCHI, Volume collapsed three-manifolds with a lower cur-
vature bound. Math. Ann. 333: 131–155, 2005.
[Tay] M. E. TAYLOR, Noncommutative Harmonic Analysis, Mathematical Surveys and
Monographs, AMS, 1986.
[Thu] W. T HURSTON, The Geometry and Topology of Three- Manifolds, preprint, 1978,
Princeton University, available online at http://www.msri.org/publications/books/
gt3m; and Three-Dimensional Geometry and Topology, Vol. 1, Princeton University
Press, 1997.
[Tri] H. T RIEBEL , Theory of Function Spaces, Monographs in Mathematics 84,
Birkauser, Basel, 1992.
[Sun] T. S UNADA, Riemannian covering and isospectral manifolds, Ann. of Math. 121:
169–186, 1985.
 Bibliography 183

[Var] S. R. S. VARADHAN, On the behavior of the fundamental solution of the heat equa-
tion with variables coefficients, Comm. Pure. Appl. Math. 20: 431–455, 1967.
[War] F. W. WARNER, Foundations of Differentiable Manifolds and Lie Groups, Springer,
New York, 1983.
[Wee] J. R. W EEKS, The Shape of Space, Marcel Dekker Inc, New York, 2002.
[Ya-Sc] S. T. YAU & R. S CHOEN, Differential Geometry, Academic Press, 2006.
[Yos] K. YOSIDA, Functional Analysis, CIM, Springer, 1980.
[Zel] S. Z ELDITCH, Spectral determination of analytic bi-axisymmetric plain domains,
Geom. Func. Anal. 10: 628–677, 2000.
Index

adjoint operator, 14 divergence operator, 65

annihilation operator, 109
eigenspace, 19
Betti number, 105 eigenvalue, 19
Bochner–Lichnerowicz formula, 113 eigenvector, 19
bounded operator, 7 eight Thurston geometries, 170
brick, 158 electrostatic capacity, 124, 127
elliptic, 8, 78
canonical measure, 64 elliptisation conjecture, 171
canonical metric, 57 essentially self-adjoint, 41
chart, 51 exponential map, 60
Cheeger inequality, 114 exterior derivative operator, 56
Cheeger’s constant, 115
Christoffel symbols, 58 Faber–Krahn Theorem, 115
closable operators, 40 finite difference method, 3
closed eigenvalue problem, 75 finite element method, 89
closed operators, 12 finite rank operator, 27
co-closed, 103 flat torus, 106
co-exact, 103 Fourier transform, 138
compact operator, 27 functional calculus, 38
compact resolvent, 41
confining potential, 101 Gaussian curvature equation, 164
conformal, 163 geodesics, 59
conformal class, 163 geometry, 170
connection, 55 gradient, 65
cotangent bundle, 54 Green’s formula, 69
cotangent space, 54
Courant’s Nodal Theorem, 121 harmonic, 103
creation operator, 110 harmonic oscillator, 109
curvature tensor, 61 heat equation, 72, 144
heat kernel, 145
de Rham chain complex, 103 Hermite family, 112
de Rham cohomology group, 103 Hodge decomposition, 104
derivation, 52 Hodge–de Rham Laplacian, 103
differential, 53 homogeneous, 170
differential form, 55 Hopf–Rinow Theorem, 61
Dirac distribution, 66 hyperbolic space, 58
Dirichlet eigenvalue problem, 75
distribution, 66 image, 10
divergence formula, 65 injectivity radius, 61
186 Index

integration by parts formula, 69 positive, 16

interior product operator, 56 pseudo-differential operator, 118
invertible, 11
isospectral, 78 quantum auto-correlation function, 74
isotropic, 170 quantum Hamiltonian, 40

kernel, 10 Rayleigh quotient, 91

rectangular domain, 108
Laplace–Beltrami operator, 68 regular distribution, 66
Lax–Milgram theorem, 8 Rellich Theorem, 68
Lebesgue space, 67 resolvent equality, 22
length spectrum, 136 resolvent operator, 20
Levi-Civita connection, 58 resolvent set, 19
Li–Yau Theorem, 115 Ricci curvature, 62
Lichnerowicz–Obata Theorem, 112 Ricci flow, 169
Lie bracket, 54 Ricci flow with bubbling-off, 172
linear tangent map, 55 Riemannian distance, 59
local flow, 54 Riemannian manifold, 57
Riemannian metric, 57
manifold, 51
Riesz lemma, 27
microsupport, 118
Riesz–Schauder theory, 30
Milnor’s counterexample, 139
Minakshisundaram–Pleijel expansion, scalar curvature, 62
156 Schauder theorem, 29
minimax principle, 90 Schrödinger equation, 39, 73
mixed problem, 76 Schrödinger operator, 100
monotonicity principle, 97 sectional curvature, 62
multiplicity, 78 self-adjoint, 16
semi-classical paradigm, 117
Neumann eigenvalue problem, 76
Sobolev space, 67
nodal domains, 121
spectral invariant, 78
nodal set, 121
normalized functional determinant, 167 spectral partition function, 118
numerical range, 44 Spectral Theorem for bounded normal
operators, 36
operators with domains, 10 Spectral Theorem multiplication opera-
tor form, 37
partition of unity, 64 spectral zeta function, 164
Poincaré conjecture, 171 spectrum, 19
Poincaré disk, 58 spectrum of the manifold .M; g/, 78
Poincaré inequality, 67, 128 sphere, 108
point spectrum, 19 spherical Laplacian, 109
Poisson summation formula, 138 Steklov problem, 76
Polyakov formula, 167 symmetric, 16
position operator, 10 Szegö–Weinberger Theorem, 115
 Index 187

tangent bundle, 53 Uniformization Theorem, 168

tangent space, 53 unimodular, 170
tangent vector, 52
test functions, 66 vector field, 54
trace formulas, 137 volume form, 64
transition functions, 52 von Neumann lemma, 11
transplantation map, 161
transplantation matrix, 160
wave equation, 71
transplantation method, 158
wave function, 39
unbounded operators, 10 Weyl formula, 116