100% found this document useful (1 vote)
228 views

The Dirac Spectrum

The Dirac spectrum
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
100% found this document useful (1 vote)
228 views

The Dirac Spectrum

The Dirac spectrum
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 168

Lecture Notes in Mathematics 1976

Editors:
J.-M. Morel, Cachan
F. Takens, Groningen
B. Teissier, Paris
Nicolas Ginoux

The Dirac Spectrum

ABC
Nicolas Ginoux
NWF I -Mathematics
University of Regensburg
Universitätsstraße 31
93040 Regensburg
Germany
nicolas.ginoux@mathematik.uni-regensburg.de

ISSN 0075-8434 e-ISSN 1617-9692


ISBN 978-3-642-01569-4 e-ISBN 978-3-642-01570-0
DOI 10.1007/978-3-642-01570-0
Springer Dordrecht Heidelberg London New York

Library of Congress Control Number: 2009926184

Mathematics Subject Classification (2000): 35P15, 53C27, 58C40, 58J32, 58J50

c Springer-Verlag Berlin Heidelberg 2009


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, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from Springer. Violations
are liable to prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant protective
laws and regulations and therefore free for general use.

Cover design: SPi Publisher Services

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)


Dedicated to my dear and loving mother
Preface

This overview is based on the talk [101] given at the mini-workshop 0648c
“Dirac operators in differential and non-commutative geometry”, Mathe-
matisches Forschungsinstitut Oberwolfach. Intended for non-specialists, it
explores the spectrum of the fundamental Dirac operator on Riemannian
spin manifolds, including recent research and open problems. No background
in spin geometry is required; nevertheless the reader is assumed to be famil-
iar with basic notions of differential geometry (manifolds, Lie groups, vector
and principal bundles, coverings, connections, and differential forms). The
surveys [41, 132], which themselves provide a very good insight into closed
manifolds, served as the starting point. We hope the content of this book
reflects the wide range of findings on and sometimes amazing applications of
the spin side of spectral theory and will attract a new audience to the topic.

vii
Acknowledgements

The author would like to thank the Mathematisches Forschungsinstitut Ober-


wolfach for its friendly hospitality and stimulating atmosphere, as well as the
organizers and all those who participated in the mini-workshop. This sur-
vey would not have been possible without the encouragement and advice of
Christian Bär and Oussama Hijazi as well as the support of the German
Research Foundation’s Sonderforschungsbereich 647 “Raum, Zeit, Materie.
Analytische und geometrische Strukturen” (Collaborative Research Center
647 / Space, Time and Matter. Analytical and Geometric Structures). We
would also like to thank Bernd Ammann and Nadine Große for their enlight-
ening discussions and useful references. The author is indebted to the referees
for their timely and nonetheless thorough and critical review, which helped
to significantly enhance the first draft. Last but not the least, we would like to
thank the whole Lecture Notes team for their effective and professional work.

ix
Contents

1 Basics of spin geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1


1.1 Spin group and spin structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Spinor bundle and Clifford multiplication . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The Dirac operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Spinors on hypersurfaces and coverings . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5 Elliptic boundary conditions for the Dirac operator . . . . . . . . . . . . . 23

2 Explicit computations of spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29


2.1 Spectrum of some non-negatively curved spaceforms . . . . . . . . . . . . 29
2.2 Spectrum of some other homogeneous spaces . . . . . . . . . . . . . . . . . . . . 35
2.3 Small eigenvalues of some symmetric spaces . . . . . . . . . . . . . . . . . . . . . 39

3 Lower eigenvalue estimates on closed manifolds . . . . . . . . . . . . . . . 41


3.1 Friedrich’s inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Improving Friedrich’s inequality in presence
of a parallel form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Improving Friedrich’s inequality in a conformal way . . . . . . . . . . . . 55
3.4 Improving Friedrich’s inequality with the energy-
momentum tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.5 Improving Friedrich’s inequality with other curvature
components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.6 Improving Friedrich’s inequality on surfaces
of positive genus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.7 Improving Friedrich’s inequality on bounding manifolds . . . . . . . . 64

4 Lower eigenvalue estimates


on compact manifolds with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.1 Case of the gAPS boundary condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Case of the CHI boundary condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3 Case of the MIT bag boundary condition . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Case of the mgAPS boundary condition. . . . . . . . . . . . . . . . . . . . . . . . . . 74

xi
xii Contents

5 Upper eigenvalue bounds on closed manifolds . . . . . . . . . . . . . . . . . 77


5.1 Intrinsic upper bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.2 Extrinsic upper bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

6 Prescription of eigenvalues on closed manifolds . . . . . . . . . . . . . . . 93


6.1 Dirac isospectrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Harmonic spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.3 Prescribing the lower part of the spectrum. . . . . . . . . . . . . . . . . . . . . . . 100

7 The Dirac spectrum on non-compact manifolds . . . . . . . . . . . . . . . 103


7.1 Essential and point spectrum. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 Explicit computations of spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3 Lower bounds on the spectrum. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.4 Absence of a spectral component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

8 Other topics related with the Dirac spectrum . . . . . . . . . . . . . . . . . 113


8.1 Other eigenvalue estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
8.2 Spectral gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.3 Pinching Dirac eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.4 Spectrum of other Dirac-type operators . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.5 Conformal spectral invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.6 Convergence of eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.7 Eta-invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.8 Positive mass theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

A The twistor and Killing spinor equations. . . . . . . . . . . . . . . . . . . . . . . . 131


A.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
A.2 Elementary properties of twistor-spinors . . . . . . . . . . . . . . . . . . . . . . . . . 134
A.3 Classification results for manifolds
with twistor-spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
A.4 Classification results for manifolds with Killing spinors . . . . . . . . . 141

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Introduction

“Find a first order linear differential operator on Rn whose square coincides


n ∂ 2
with the Laplace operator − j=1 ∂x 2 .” Give this as exercise to a group of
j
undergraduates. If they can solve it for n = 1 then they have heard of complex
numbers. If they can do it for n ≥ 2 then either they believe to have solved
it, or they claim to be students, or they know about Dirac.
For this simple-minded question and its rather involved answer lie at the
origin of the whole theory of Dirac operators. It was P. Dirac who introduced
[81] the operator now bearing his name when looking for an equation describ-
ing the probability amplitude of spin- 12 -particles (fermions, e.g. electrons)
and that would fit into the framework of both special relativity and quantum
mechanics. Mathematically formulated, his problem consisted in finding a
square root of the Klein-Gordon operator (d’Alembert plus potential) on the
4-dimensional Minkowski spacetime. It already came as a breakthrough when
Dirac showed that the problem could be solved not for the scalar operator
but for the C2 -valued one using the so-called Pauli matrices as coefficients.
Like many objects invented by physicists, the Dirac operator was soon
called upon to develop an own mathematical life. It was indeed later discov-
ered that the setup of Clifford algebras allowed it to be defined in a general
geometrical framework on “almost any” smooth semi-Riemannian manifold.
Here “almost any” means that there exists a topological restriction on the
manifold for the Dirac operator to be well-defined - the spin condition, see
Chapter 1 - which is however satisfied on most “known” manifolds. This
mathematical investigation gave birth to spin geometry. One of the first and
probably most famous achievements of spin geometry was the discovery of a
topological obstruction to positive scalar curvature as a relatively straight-
forward application of the Atiyah-Singer-Index theorem, see Chapter 3 and
references therein.
It would be very modest to claim that spin geometry has remained
lively since. Less than twenty years after Atiyah and Singer’s breakthrough,
the whole mathematical community could only gape at E. Witten’s amaz-
ingly simple proof of the positive mass theorem based on the analysis of
a Dirac-type operator of a bounding hypersurface [235]. At the same time,
noncommutative geometry made the Dirac operator one of its keystones as

xiii
xiv Introduction

it allows to reconstruct a given Riemannian spin manifold from its so-called


canonical spectral triple [78, 112]. Independently, special eigenvectors of the
Dirac operator called Killing spinors have become some of the physicists’
main tools in the investigation of supersymmetric models for string theory
in dimension 10, see e.g. [184]. In a more geometrical context, Dirac-type
operators have been successfully applied in as varied situations as finding
obstructions to minimal Lagrangian embeddings [141], rigidity issues in ex-
trinsic geometry [138] or the Willmore conjecture [14, 17], just to cite a few
of them.
Exploring the spin geometrical aspects of all the above-mentioned topics
would require a small encyclopedia, therefore we focus on a particular one.
Out of lack of up-to-date literature on the subject, we choose to deal in this
book with the spectrum of the Dirac operator on complete (mainly compact)
Riemannian spin manifolds with or without boundary. In particular we do
not intend to give any kind of extensive introduction to spin geometry, see
[63, 88] and the mother-reference [173] in this respect (the physics-oriented
reader may prefer [227]). Since it was not possible to handle all facets of
the Dirac spectrum in one volume, we had to leave some of them aside. To
keep the book as self-contained as possible, we sketch those briefly in the last
chapter.
We begin with introducing the Dirac operator and its geometrical back-
ground. Although the definition is rather involved, we try to remain as simple
as possible so as not to drown the reader in technical considerations such as
representation theory of Clifford algebras or the topological spin condition.
In Chapter 1 we define the spin group, spin structures on manifolds, spinors
(which are sections of a vector bundle canonically attached to manifolds car-
rying a spin structure) and the Dirac operator acting on spinors. We show
that the Dirac operator is an elliptic, formally self-adjoint linear differential
operator of first order and, if the underlying Riemannian manifold is further-
more complete, then it is essentially self-adjoint in L2 . In particular, if the
manifold is closed, then the spectrum of its Dirac operator is well-defined,
real, discrete and unbounded. In case the boundary of the manifold is non-
empty, elliptic boundary conditions have to be precised for the spectrum to
be well-defined and discrete.
At this point we underline that only a so-called spinc structure is needed on
the manifold in order for the Dirac operator to be well-defined. Spinc struc-
tures require weaker topological assumptions to exist than spin structures.
Since however their treatment would bring us too far, we choose to ignore
them in this book (see Section 8.4 for references).
The second chapter deals with examples of closed manifolds whose Dirac
spectrum - or at least some eigenvalues - can be explicitly computed. They
all belong to the class of homogeneous spaces, for which we recall the
representation-theoretical method allowing one to describe the Dirac operator
as a family of matrices, see Theorem 2.2.1.
Introduction xv

Since it would be illusory to aim at the explicit knowledge of the Dirac


spectrum in general, one way for studying it consists in estimating the
eigenvalues. In Chapter 3 we consider an arbitrary closed Riemannian spin
manifold and describe the main lower bounds that have been proved for its
Dirac spectrum. Almost all of them rely on the Schrödinger-Lichnerowicz for-
mula (1.15) and thus involve the scalar curvature of the manifold. Starting
from the most general estimate - Friedrich’s inequality (3.1) - we show how
it can be improved in some particular cases. The equality-case of most of
those inequalities is characterized by the existence of special sections (e.g.
Killing spinors) which give rise to interesting geometrical features. We shift
the treatment of some of them to Appendix A since they are of independent
interest.
In the situation where the manifold has a non-empty boundary, we consider
four different boundary conditions, two of which generalize those originally
introduced by Atiyah, Patodi and Singer [27]. We describe in Chapter 4 the
corresponding lower bounds à la Friedrich that have been obtained in this
context.
The techniques involved for proving lower bounds drastically differ from
those used in getting upper eigenvalue bounds. In the latter case - and if the
manifold is closed - there exist two methods available for the Dirac operator,
the first one based on index theory and the second one on the min-max
principle. Chapter 5 collects the different geometrical upper bounds that
have been proved with the help of those, separating the intrinsic - depending
on the intrinsic geometry only - from the extrinsic ones, i.e., depending on
some map from the manifold into another one.
In Chapter 6, we turn to the closely related issues of isospectrality and
prescription of eigenvalues. In a first part, we discuss isospectrality results
obtained on spaceforms of non-negative curvature and on circle bundles.
Turning to the eigenvalue 0, we detail in Section 6.2 existence as well as
non-existence results for harmonic spinors, i.e., sections lying in the kernel of
the Dirac operator. Here there is a remarkable difference between dimensions
2 and greater than 2. We end this chapter with a brief account on how the
lower part of the spectrum can always be prescribed provided it does not
contain 0.
On non-compact Riemannian spin manifolds another part of the spectrum
beside the eigenvalues must be taken into account, the so-called continuous
spectrum. For the Dirac operator it is well-defined as soon as the underlying
Riemannian manifold is complete, however the square of the Dirac operator
always has a spectrum (see Section 7.1). Only few examples are known where
the whole Dirac spectrum can be computed. In Chapter 7 we mainly discuss
the interactions between the geometry or topology of the manifold with the
Dirac spectrum, in particular we focus on whether it can be purely discrete
or continuous.
Chapter 1
Basics of spin geometry

In this chapter we define spin structures, spinors, the Dirac operator and
discuss the properties we need further on. Unless explicitly mentioned all
objects (manifolds, bundles, sections) will be assumed smooth in the whole
survey. For the thorough treatment of spin or spinc groups, spin or spinc
structures on vector bundles, representation theory of Clifford algebras and
Dirac operators on arbitrary semi-Riemannian Clifford modules we refer to
[63, 88, 173].

1.1 Spin group and spin structure

Definition 1.1.1 Let n be a positive integer. The spin group in dimension


n, denoted by Spinn , is the non-trivial 2-fold covering of the special orthogonal
group SOn .
n(n−1)
The spin group in dimension n is a compact 2 -dimensional Lie group,
ξ
connected if n ≥ 2 and simply-connected if n ≥ 3. In fact, if Spinn −→ SOn
denotes this non-trivial covering map, then ξ(z) = z 2 for any z ∈ Spin2 ∼
=
S1 = {z ∈ C, |z| = 1} and ξ is the universal covering map if n ≥ 3. In
particular the spin group provides the following short exact sequence:

ξ
0 −→ Z2 −→ Spinn −→ SOn −→ 1,

where we identify {±1} to Z2 .


From now on we denote by SO(T M ) −→ M the SOn -principal bundle of
positively oriented orthonormal frames on the tangent bundle of an oriented
Riemannian manifold (M n , g).
Definition 1.1.2 Let n ∈ N \ {0}.
i) A spin structure on an oriented Riemannian manifold (M n , g) is a Spinn -
principal bundle Spin(T M ) −→ M together with a 2-fold covering map

N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 1


DOI 10.1007/978-3-642-01570-0 1, 
c Springer-Verlag Berlin Heidelberg 2009
2 1 Basics of spin geometry
η
Spin(T M ) −→ SO(T M ) compatible with the respective group actions, i.e.,
the following diagram commutes:

Spin(T M ) × Spinn Spin(T M )

η×ξ η M

SO(T M ) × SOn SO(T M )

ii) A spin manifold is an oriented Riemannian manifold admitting a spin


structure.

A spin structure is a reduction of the bundle of oriented orthonormal frames


to the spin group. Not every oriented Riemannian manifold admits a spin
structure, the condition for its existence being of topological nature.

Proposition 1.1.3 An oriented Riemannian manifold (M n , g) is spin if and


only if the second Stiefel-Whitney class of its tangent bundle vanishes. If this
condition is fulfilled, then the set of spin structures on (M n , g) stands in
one-to-one correspondence with H 1 (M, Z2 ).
In other words, a manifold is spin if and only both its first and sec-
ond Stiefel-Whitney classes vanish (the vanishing of the first one being
equivalent to the orientability of the manifold). This explains why the spin
condition is sometimes presented as an orientability condition of second order.

Proof of Proposition 1.1.3: First recall that, for any Lie group G, there exists
a bijection between the set of equivalence classes of G-principal bundles over
some manifold N and the set H 1 (N, G) (which, if G is abelian, is the first
Čech-cohomology group with coefficients in G), see e.g. [173, App. A]. In
particular the set of two-fold coverings of - i.e., of Z2 -bundles over - N can
be identified with H 1 (N, Z2 ).
For n = 1 the result is a trivial consequence of this observation, since in
that case H 2 (M, Z2 ) = 0 and a spin structure is a 2-fold covering of the
manifold itself, hence R has exactly one and the circle 2 spin structures, see
also Example 1.4.3.1 below for a more precise description.
Assume for the rest of the proof n ≥ 2. First note that, from their
definition, the spin structures on (M n , g) coincide with the 2-fold cover-
ings of SO(T M ) which are non-trivial on each fibre of the projection map
SO(T M ) −→ M . This follows essentially from the standard lifting prop-
erty of maps through coverings. From the remark above, the set of 2-fold
coverings of SO(T M ) can be identified with H 1 (SO(T M ), Z2 ). Now the
1.1 Spin group and spin structure 3

second Stiefel-Whitney class of T M can be defined as follows: the fibration


ι π
SOn → SO(T M ) → M induces the following short exact sequence of groups

π∗ ι∗ w
0 −→ H 1 (M, Z2 ) −→ H 1 (SO(T M ), Z2 ) −→ H 1 (SOn , Z2 ) −→ H 2 (M, Z2 )

and the second Stiefel-Whitney class of T M is the image under w of the


non-trivial element of H 1 (SOn , Z2 ) ∼ = Z2 . The spin structures on (M n , g)
can therefore be reinterpreted as the elements in H 1 (SO(T M ), Z2 ) with
non-zero image under ι∗ . In particular (M n , g) is spin if and only if such an
element exists, that is, if and only if ι∗ is surjective. From w ◦ ι∗ = 0 this
is equivalent to w = 0, i.e., to the vanishing of the second Stiefel-Whitney
class of T M . This proves the first statement. If (M n , g) is spin then the
set of its spin structures identifies through the above exact sequence with
the non-zero coset in H (SO(T M ), Z2 )/π ∗ (H 1 (M, Z2 )), which has the same
1

cardinality as H 1 (M, Z2 ) itself. This concludes the proof. 

Notes 1.1.4
1. In particular the existence of a spin structure does not depend on the
metric or the orientation of a given manifold. Actually, if the manifold
M is oriented, spin structures can be defined independently of any metric
(declare them to be non-trivial 2-fold coverings of the bundle of oriented
frames of T M ). In an equivalent way, a spin structure for a given met-
ric canonically induces a spin structure for another one. For a detailed
discussion of this point we refer to [173, Chap. 2], [10] and to [62].
2. Not every orientable manifold is spin. On surfaces the spin condition is
equivalent to the vanishing of the mod 2 reduction of the Euler class, thus is
fulfilled for orientable surfaces. In dimension 3, the second Stiefel-Whitney
class is the square of the first one [173, p.86], hence any 3-dimensional
orientable manifold is spin. The simplest counter-example comes up in
dimension 4: the complex projective plane CP2 is not spin (even if it
canonically carries a so-called spinc structure as a Kähler manifold). Indeed
a complex manifold is spin if and only if the mod 2 reduction of its first
Chern class vanishes, see [173, App. D].
3. However any simply-connected manifold has a unique spin structure as
soon as it is spin, since in that case H 1 (M, Z2 ) = 0.
4. It was first noticed by J. Milnor [193] that different spin struc-
tures may provide equivalent principal Spinn -bundles. For instance, the
2-dimensional torus has 4 different spin structures (see Example 1.4.3.2),
all of which are equivalent as principal Spin2 -bundles. This is due to the
fact that, for the torus, H 1 (M, Z2 ) = H 1 (M, Z) ⊗ Z2 , see [173, p.84].
From now on, each time we assume that a manifold is spin we shall implicitly
mean that a spin structure is fixed on it.
4 1 Basics of spin geometry

1.2 Spinor bundle and Clifford multiplication

In this section we define the natural algebraic and geometric objects on a


Riemannian spin manifold, namely the Clifford multiplication and the in-
duced compatible covariant derivative on its spinor bundle. We first recall
the following central result in representation theory which can be found e.g.
in [173, Prop. 5.15 p.36].

Proposition 1.2.1 Let n ∈ N \ {0}.


i) If n is odd then there exists up to equivalence exactly one fundamental
δ
irreducible complex representation Spinn −→ n
Aut(Σn ) of Spinn that does
not come from SOn . It is called the spinor representation in dimension n
n−1
and has dimension 2 2 .
ii) If n is even then there exist up to equivalence exactly two fundamental
δ±
irreducible complex representations Spinn −→
n
Aut(Σ±n ) of Spinn that do
not come from SOn . The representations δn and δn− are called the positive
+

and negative half spinor representation in dimension n respectively and


n−2
each have dimension 2 2 .
Recall that a fundamental representation of a compact Lie group G is an irre-
ducible complex representation whose highest weight is a fundamental weight
(it belongs to a system of generators of all irreducible complex representations
(±)
of G). That δn is a representation of Spinn which does not come from SOn
(±)
means that there does not exist any representation ρ of SOn with ρ◦ξ = δn .

Proof of Proposition 1.2.1: The representation



δn+ ⊕ δn− for n even
δn :=
δn for n odd

is actually the restriction of the (or one of both if n is odd) irreducible


representation(s) of the corresponding finite-dimensional complex Clifford
algebra. Remember that, for any given (real or complex) vector space V
endowed with a non-degenerate symmetric bilinear form · , ·
, the Clifford
algebra of the pair (V, · , ·
) is the quotient of its tensor algebra through the
two-sided ideal generated by the elements of the form x ⊗ y + y ⊗ x + 2 x, y
1
where x, y ∈ V . The product law induced on the quotient, which is usually
denoted by “·” and called the Clifford product or Clifford multiplication,
satisfies the Clifford relations, namely

x · y + y · x = −2 x, y
1 (1.1)
1.2 Spinor bundle and Clifford multiplication 5

for all x, y ∈ V . Since we do not want to deal with Clifford algebras in detail
we just recall their most important properties for our purpose (see e.g. [173,
Chap. 1]):
1. The Clifford algebra is the smallest associative algebra with unit contai-
ning V and satisfying the Clifford relations.
2. The Clifford algebra of the n-dimensional Euclidean space is linearly iso-
morphic to its exterior algebra Λ∗ Rn , the Clifford product being then
given by
x· x ∧ − x (1.2)
for every x ∈ Rn .
3. If V is real then the complexification of the Clifford algebra of (V, · , ·
)
coincides with the (complex) Clifford algebra of (V ⊗ C, · , ·
C ) (where
· , ·
C is complex bilinear).
4. The Clifford algebra of C2n with its canonical complex bilinear form is
isomorphic to the algebra of all complex 2n × 2n matrices, and that of
C2n+1 to two copies of this algebra [173, Tab. I p.28].
Property 4 implies in particular the existence of exactly two and one
irreducible non-trivial representations of the Clifford algebra of Cn for n
odd and even respectively. Now Spinn can be identified with the set of
even Clifford products of unit vectors of Rn (this can actually be used
as definition of Spinn ), in particular sits in the Clifford algebra of Cn .
After restriction onto Spinn both Clifford-algebra-representations turn out
to become equivalent for n odd whereas the unique one splits into two
inequivalent equally dimensional representations of Spinn for n even. The
statement on their dimensions easily follows. 

The simplest way to distinguish δn+ from δn− consists in looking at the action
of the so-called complex volume form of Rn and which is defined for every
n by
ωnC := i[ 2 ] e1 · . . . · en
n+1
(1.3)
for any positively-oriented orthonormal basis (e1 , . . . , en ) of Rn . The complex
volume form does in general not lie in Spinn but in the complex Clifford
algebra of Cn and it can be shown that

δn± (ωnC ) = ±IdΣ±


n
(1.4)

for n even whereas δn (ωnC ) = IdΣn or −IdΣn for n odd (in the latter case both
possibilities can occur).

Definition 1.2.2 Let (M n , g) be a Riemannian spin manifold.


i) The spinor bundle of M is the complex vector bundle associated to the
principal bundle Spin(T M ) via the spinor representation, i.e.,
6 1 Basics of spin geometry

ΣM := Spin(T M ) ×δn Σn ,

− −
where, for n even, Σn := Σ+ n ⊕ Σn and δn := δn ⊕ δn .
+

i’) If M is even-dimensional, the positive (resp. negative) spinor bundle


of M is the complex vector bundle associated to the principal bundle
Spin(T M ) via the positive (resp. negative) half-spinor representation, i.e.,

Σ+ M := Spin(T M ) ×δ+ Σ+
n
n

(resp. Σ− M := Spin(T M ) ×δ− Σ−


n ).
n
ii) A Clifford multiplication is a complex linear vector-bundle-homo-
morphism
μ
T ∗ M ⊗ ΣM −→ ΣM, X  ⊗ ϕ −→ X · ϕ

such that
X ·(Y · ϕ) + Y ·(X · ϕ) = −2g(X, Y )ϕ (1.5)

for all X, Y ∈ T M and ϕ ∈ ΣM .


iii) A compatible covariant derivative on ΣM is a covariant derivative ∇Σ
on ΣM such that

∇X
Σ
(Y · ϕ) = (∇X Y )· ϕ + Y · ∇X
Σ
ϕ

for all X, Y ∈ Γ(T M ) and ϕ ∈ Γ(ΣM ), where ∇ denotes the Levi-Civita


covariant derivative of (M n , g).
In particular the spinor bundle of an even-dimensional Riemannian spin ma-
nifold always splits into ΣM = Σ+ M ⊕ Σ− M . Note moreover that, in any
n
dimension, rkC (ΣM ) = 2[ 2 ] .
From its definition any Clifford multiplication can be canonically extended
into an algebra-homomorphism from the so-called Clifford-algebra-bundle to
End(ΣM ). Since we essentially do not need the Clifford-algebra-bundle, we
just indicate how μ extends from T ∗ M ⊗ ΣM to ΛT M ⊗ ΣM : for any p-form
ω and any spinor ϕ, the product ω · ϕ is defined by

ω · ϕ := ωj1 ,...,jp ej1 · (. . . · (ejp · ϕ)),
1≤j1 <...<jp ≤n


where ω = 1≤j1 <...<jp ≤n ωj1 ,...,jp e∗j1 ∧ . . . ∧ e∗jp in a local orthonormal ba-
sis {ej }1≤j≤n of T M . Moreover the Clifford algebra being associative (see
Property 1 in the proof of Proposition 1.2.1 above), we shall in the following
forget about the parentheses and write X ·Y · ϕ instead of X ·(Y · ϕ).
The spinor bundle comes with a natural Hermitian inner product which
together with Clifford multiplication exist and are unique in some sense.
1.2 Spinor bundle and Clifford multiplication 7

Proposition 1.2.3 Let (M n , g) be a Riemannian spin manifold.


a) If n is odd then there exist up to equivalence exactly two Clifford multipli-
cations, which are opposite from each other.
b) If n is even then there exists up to equivalence exactly one Clifford multi-
plication.
c) There exists a Hermitian inner product · , ·
on ΣM , pointwise unique up
to scale, such that
X · ϕ, ψ
= − ϕ, X · ψ

for all X ∈ T M and ϕ, ψ ∈ ΣM , where “ ·” denotes any fixed Clifford


multiplication on ΣM .
d) There exists a metric compatible covariant derivative ∇Σ on ΣM .
Proof : The first two statements follow from Proposition 1.2.1 (see also its
proof), since the spinor or half-spinor representations are exactly provided
by representations of the complex Clifford algebra. The third statement is
not a direct consequence of the existence of a Hermitian inner product on Σn
(±)
preserved by δn (which would simply follow from the compactness of Spinn ),
but from the following argument: choose any Hermitian inner product · , ·

on Σn and mean it over the Clifford action of the (finite) group generated
by the canonical basis of Rn (which is a subgroup of the group of invertible
elements in the Clifford algebra). One obtains a new Hermitian inner product
· , ·
0 on Σn for which obviously every canonical basis vector acts unitarily
and hence in a skew-adjoint way from (1.1). Therefore the Clifford action
of every vector of Rn on Σn becomes also skew-adjoint w.r.t. · , ·
0 . Since
Spinn can be identified with the set of even Clifford products of unit vectors
of Rn , the inner product · , ·
0 remains invariant under the Spinn -action -
which is precisely δn . Hence · , ·
0 induces a Hermitian inner product · , ·

on ΣM which obviously has the same properties. The pointwise uniqueness


of · , ·
up to scale follows from the irreducibility of δn as a Clifford-algebra-
representation.
A covariant derivative on ΣM may be constructed by lifting locally the
connection 1-form of the Levi-Civita covariant derivative of (M n , g), which
d1 ξ
is made possible by spinn −→ son being a Lie-algebra-isomorphism, see e.g.
[134, (8) p.140] and the local formula (1.6) below. It is then straightforward
to show that the covariant derivative defined in this way is metric and
compatible in the sense of Definition 1.2.2.iii) (see e.g. [134, Prop. 4.4]).
This explains d). 

From now on we choose a Hermitian inner product · , ·


on ΣM coming from
a Hermitian inner product on Σn making Clifford multiplication of vectors
of Rn skew-Hermitian, as in Proposition 1.2.3.
Note that, if ∇Σ is a compatible covariant derivative on ΣM , then for any
real 1-form θ on M the covariant derivative ∇Σ + iθ ⊗ Id is again compatible,
hence such a covariant derivative is not unique. We choose henceforth and
8 1 Basics of spin geometry

denote by ∇ the covariant derivative on ΣM naturally induced by the Levi-


Civita covariant derivative ∇ on (M n , g) and which can be locally expressed
as [134, Prop. 4.3]

1 
n
∇ϕα = g(∇ej , ek )ej · ek · ϕα , (1.6)
4
j,k=1

where {ej }1≤j≤n denotes a local positively-oriented orthonormal basis of T M


and {ϕα }1≤α≤2[ n2 ] any corresponding local spinorial frame, that is: ϕα =
s, σα ] with η(
[ s) = (e1 , . . . , en ) and {σα }1≤α≤2[ n2 ] is a fixed orthonormal
basis of Σn .
The curvature R∇ of that covariant derivative can be explicitly expressed
through that (denoted by R) of the Levi-Civita covariant derivative on
(M n , g): for any local orthonormal basis {ej }1≤j≤n of T M ,

1 
n

RX,Y ϕ= g(RX,Y ej , ek )ej · ek · ϕ, (1.7)
4
j,k=1

for all X, Y ∈ T M and ϕ ∈ ΣM (see again [134, Prop. 4.3]). In dimension


n = 2 this identity simplifies into

∇ S
RX,Y ϕ= (X · Y − Y · X) · ϕ, (1.8)
8

where S is the scalar curvature of (M n , g).


The following very useful formula can be deduced from (1.7).
Lemma 1.2.4 Let Ric denote the Ricci-tensor of the Riemannian spin man-
ifold (M n , g), then for all X ∈ T M and ϕ ∈ ΣM one has


n
1

ej · RX,e ϕ= Ric(X) · ϕ, (1.9)
j=1
j
2

Proof : The first Bianchi identity implies that


n
(1.7) 1 
n

ej · RX,e ϕ = g(RX,ej ek , el )ej · ek · el · ϕ
j=1
j
4
j,k,l=1

1 n
= − g(Rej ,ek X, el )ej · ek · el · ϕ
4
j,k,l=1

1 n
− g(Rek ,X ej , el )ej · ek · el · ϕ
4
j,k,l=1
1.3 The Dirac operator 9

1 
n
= − g(RX,ej ek , el )(ek · el · ej − ek · ej · el ) · ϕ,
4
j,k,l=1

with

(1.5)
ek · el · ej − ek · ej · el = −ek · ej · el − 2δjl ek + ej · ek · el + 2δjk el
(1.5)
= 2ej · ek · el + 4δjk el − 2δjl ek .

We deduce that


n 
n
1 
n

3 ej · RX,e ϕ=− g(RX,ej ej , el )el · ϕ + g(RX,ej ek , ej )ek · ϕ
j=1
j
2
j,l=1 j,k=1

n
1
n
= g(Ric(X), el )el · ϕ + g(Ric(X), ek )ek · ϕ
2
l=1 k=1
3
= Ric(X) · ϕ,
2
which is the result. 

1.3 The Dirac operator

We are now ready to define the central object of this survey.

Definition 1.3.1 The fundamental Dirac operator of a Riemannian spin


manifold (M n , g) is the map D : Γ(ΣM ) −→ Γ(ΣM ) defined by


n
Dϕ := ej · ∇ej ϕ
j=1

for every ϕ ∈ Γ(ΣM ), where {ej }1≤j≤n is any local orthonormal basis of
TM.

The Dirac operator is sometimes called Atiyah-Singer operator in the lit-


erature in honour to M. Atiyah and I. Singer who brought it to mathematical
daylight through their famous index theorem [28]. To distinguish it from its
twisted and/or generalized versions on Clifford bundles, it is also called the
fundamental (or spin) Dirac operator. In this book, we only deal with D, see
Chapter 8 for results related to twisted or generalized Dirac operators. When
necessary we shall write DM or DM,g , Dg to precise the underlying manifold
M and/or the metric g.
10 1 Basics of spin geometry

The Dirac operator is obtained as the composition of the Clifford multi-


plication with the natural covariant derivative on ΣM . Alternatively one can
check that the local expression defining the Dirac operator is independent
of the local orthonormal basis chosen on T M . Beware here that, since D
depends on the choice of Clifford multiplication, it is only defined up to a
sign if n is odd. The usual convention in that case is to choose the Clifford
multiplication such that the action of the complex volume form ωnC (whose
algebraic definition (1.3) makes sense on M as an element of the so-called
Clifford-algebra-bundle) is the identity, see e.g. [173, Prop. 5.9 p.34] for the
real analog.
In case n is even the Dirac operator can be split in a canonical way.

Proposition 1.3.2 Let (M n , g) be an even-dimensional Riemannian spin


manifold, then its Dirac operator D splits into

D = D+ ⊕ D− ,

where D± : Γ(Σ± M ) −→ Γ(Σ∓ M ).

Proof : In even dimension the Clifford action of the complex volume form ωnC
is a non-trivial parallel involution of ΣM anti-commuting with the Clifford
multiplication with vectors (i.e., X · ωnC = −ωnC · X for every X ∈ T M ), so
that
D(ωnC · ϕ) = −ωnC · Dϕ (1.10)

for every ϕ ∈ Γ(ΣM ). From Σ± M = {ϕ ∈ ΣM | ωnC · ϕ = ±ϕ} we conclude. 

We come to the properties of the Dirac operator which are fundamental for
the further study of its spectrum. First we need a formula computing the
commutator of the Dirac operator with a function. For technical reasons we
include into the next lemma the computation of commutators or anticommu-
tators involving the Dirac operator and which we shall need in the following.

Lemma 1.3.3 Let ϕ be a smooth spinor field, f a smooth function and ξ a


smooth vector field on a Riemannian spin manifold (M n , g), then

i)
D(f ϕ) = grad(f ) · ϕ + f Dϕ, (1.11)

where grad(f ) denotes the gradient vector field of f on (M n , g).


ii)
D(ξ · ϕ) = −ξ ·Dϕ − 2∇ξ ϕ + (d + δ)ξ  ·ϕ, (1.12)

where d and δ denote the exterior differential and codifferential on (M n , g)


respectively.
1.3 The Dirac operator 11

iii)
D2 (f ϕ) = f D2 ϕ − 2∇grad(f ) ϕ + (Δf )ϕ, (1.13)
where Δ := δd = −tr(Hessg (·)) denotes the scalar Laplace operator on
(M n , g).
Proof : Fix a local orthonormal basis {ej }1≤j≤n of T M . We compute:


n
D(f ϕ) = ej · ∇ej (f ϕ)
j=1
n
= ej · (ej (f )ϕ + f ∇ej ϕ)
j=1
= grad(f ) · ϕ + f Dϕ,

which proves i). Moreover,


n
D(ξ · ϕ) = ej ·∇ej (ξ · ϕ)
j=1
n 
n
= ej ·(∇ej ξ) · ϕ + ej ·ξ·∇ej ϕ
j=1 j=1

(1.5) n n 
n
= ej ·(∇ej ξ) · ϕ − ξ · ej ·∇ej ϕ − 2 g(ξ, ej )∇ej ϕ
j=1 j=1 j=1

(1.2) n n
= ( ej ∧ ∇ej ξ  )·ϕ − ( ej ∇ej ξ  )·ϕ
j=1 j=1

n 
n
−ξ · ej ·∇ej ϕ − 2 g(ξ, ej )∇ej ϕ
j=1 j=1

= −ξ·Dϕ − 2∇ξ ϕ + (d + δ)ξ  ·ϕ

and

(1.11)
D2 (f ϕ) = D(df ·ϕ + f Dϕ)
(1.12)
= −df ·Dϕ − 2∇grad(f ) ϕ + (d + δ)df ·ϕ + df ·Dϕ + f D2 ϕ
= f D2 ϕ − 2∇grad(f ) ϕ + (Δf )ϕ,

where we have identified 1-forms with vector fields through the metric g.
This concludes the proof. 
12 1 Basics of spin geometry

Proposition 1.3.4 The Dirac operator of a Riemannian spin manifold is


an elliptic and formally self-adjoint linear differential operator of first order.

Proof : We deduce from Lemma 1.3.3 that D is a linear differential operator


of first order whose principal symbol is given by

T ∗ M −→ End(ΣM )
ξ −→ ξ
·,

where, as usual, g(ξ


, X) := ξ(X) for every X ∈ T M . For any ξ ∈ T ∗ M \ {0}
the map ξ
· is an automorphism of ΣM since it is injective (from (1.5) one
has ξ
·ξ
· ϕ = −g(ξ, ξ)ϕ and g is Riemannian). Therefore D is elliptic.
Let ϕ, ψ ∈ Γ(ΣM ), then using the mutual compatibility relations between
· , ·
, “ ·” and ∇ on ΣM one has


n
Dϕ, ψ
= ej · ∇ej ϕ, ψ

j=1

n
=− ∇ej ϕ, ej · ψ

j=1

n
= −ej ( ϕ, ej · ψ
) + ϕ, ∇ej ej · ψ

j=1

n
+ ϕ, ej · ∇ej ψ

j=1
= div(Vϕψ ) + ϕ, Dψ
, (1.14)

where Vϕψ ∈ Γ(T M ⊗ C) is defined by

(g ⊗ IdC )(Vϕψ , X) := ϕ, X · ψ

for all X ∈ T M . In particular, if ϕ or ψ has compact support on (and vanishes


on the boundary of) M then applying Green’s formula we obtain
 
Dϕ, ψ
vg = ϕ, Dψ
vg ,
M M

which shows that D is formally self-adjoint and concludes the proof. 

In even dimension, Proposition 1.3.4 means that the formal adjoint of D±


is D∓ . Beware however that Proposition 1.3.4 does not prove the self-
adjointness of D. It is indeed a priori not clear if D and its adjoint D∗ - which
is to be distinguished from the formal adjoint as a differential operator -
have the same domain of definition and if they can be extended to the
1.3 The Dirac operator 13

whole Hilbert space L2 (ΣM ), which is defined as the completion of the space
Γc (ΣM ) := {ϕ ∈ Γ(ΣM ) | supp(ϕ) is compact} w.r.t. (· , ·) := M · , ·
vg . The
concept needed here is that of essential self-adjointness: one has to show that
the closure of the operator in L2 (ΣM ) is self-adjoint. For D this is possible
as soon as the underlying Riemannian manifold is complete.

Proposition 1.3.5 Let (M n , g) be a complete Riemannian spin manifold,


then its Dirac operator is essentially self-adjoint.

Proof : The proof presented here is based on the talk [43]. Let dom(·) de-
note the domain of definition of an operator. By construction the operator
D is densely defined in L2 (ΣM ) and one can set dom(D) := Γc (ΣM ).
As a consequence it admits a unique closure D∗∗ , whose graph is the
closure of that of D. Proposition 1.3.4 implies that D - thus D∗∗ - is
symmetric in L2 (ΣM ). Its adjoint D∗ is defined on {ψ ∈ L2 (ΣM ) | ϕ →
(Dϕ, ψ) is bounded on (Γc (ΣM ),  · )}. Since the topological dual of L2 is
L2 itself, dom(D∗ ) = {ψ ∈ L2 (ΣM ) | ϕ → (Dϕ, ψ) ∈ L2 (ΣM )}. Considering
D at the distributional level, one deduces from the formal self-adjointness of
D (Proposition 1.3.4) that

dom(D∗ ) = {ψ ∈ L2 (ΣM ) | Dψ ∈ L2 (ΣM )}.

It remains to show that dom(D∗ ) = dom(D∗∗ ), i.e., the inclusion “⊂”. This
is equivalent to proving that Ker(D∗ − iεId) = 0 for both ε ∈ {±1}, see e.g.
[233, Cor. VII.2.9]. Let ψ ∈ dom(D∗ ) with D∗ ψ = iεψ, then (D∗ ψ, ϕ) =
iε(ψ, ϕ) for every ϕ ∈ Γc (ΣM ), i.e., (ψ, Dϕ) = iε(ψ, ϕ). Since D is formally
self-adjoint this is equivalent to Dψ = iεψ in the distributional sense. The
operator D − iεId being elliptic, general elliptic theory (see e.g. [173, Thm.
4.5 p.190]) implies that ψ is actually smooth. Let now (ρk )k>0 be a sequence
of smooth compactly supported [0, 1]-valued functions converging pointwise
to 1 and with |grad(ρk )| ≤ k1 for all k. Such a sequence exists because of the
completeness assumption on (M n , g). The preceding identity together with
Lemma 1.3.3 provide

iε(ψ, ρk ψ) = (Dψ, ρk ψ)
= (ψ, D(ρk ψ))
(1.11)
= (ψ, ρk Dψ) + (ψ, grad(ρk ) · ψ)
= (ρk ψ, Dψ) + (ψ, grad(ρk ) · ψ)
= −iε(ψ, ρk ψ) + (ψ, grad(ρk ) · ψ),

that is, 2iε(ψ, ρk ψ) = (ψ, grad(ρk ) · ψ), whose l.h.s. tends to 2iεψ2 and
whose r.h.s. tends to 0 when k goes to ∞. Therefore ψ = 0, QED. 
14 1 Basics of spin geometry

Note 1.3.6 It seems at first glance that the completeness assumption on the
metric enters the proof in a very weak manner and one could therefore think
about getting rid of it. This is unfornutately hopeless. Consider for example
M :=]0, +∞[ with standard metric and canonical spin structure. Its Dirac op-
erator is D = i dtd
(the Clifford multiplication of e1 = 1 ∈ R can be identified
with that of i on Σ1 = C). Set dom(D) := Γc (ΣM ) = Cc∞ (]0, +∞[, C), then
the Dirac operator remains symmetric in L2 (ΣM ). However a simple com-
putation shows that the kernel of D∗ − iεId in the space of distributions is
Reεt . Since et ∈/ L2 (ΣM ) one has Ker(D∗ − iId) = 0, however e−t ∈ L2 (ΣM )
so that Ker(D∗ − iId) = Re−t . Therefore D is not essentially self-adjoint in
L2 (ΣM ). Actually the fact that Ker(D∗ ∓ iId) do not have the same dimen-
sion imply the non-existence of self-adjoint extensions of D in L2 (ΣM ) (see
[233, Thm. VII.2.10]).
Another example is M :=]0, 1[ with the same metric and spin structure:
d
again D = i dt is not essentially self-adjoint, but this time it has infinitely
many self-adjoint extensions. Indeed the exponential et belongs in that case
to L2 (ΣM ) so that Ker(D∗ − iεId) = Reεt for both ε ∈ {±1}. More-
over Ker(D∗ ∓ iId) have the same dimension and it can be shown that
there exists an S1 -parametrized family of self-adjoint extensions of D, see
[233, Ex. VII.2.a)].
We summarise the spectral properties of the Dirac operator on closed
manifolds. Further basics on non-compact Riemannian spin manifolds are
given in Section 7.1.
Theorem 1.3.7 Let (M n , g) be a closed Riemannian spin manifold and de-
note by Spec(D) the spectrum of its Dirac operator D, then the following
holds:
i) The set Spec(D) is a closed subset of R consisting of an unbounded dis-
crete sequence of eigenvalues.
ii) Each eigenspace of D is finite-dimensional and consists of smooth sec-
tions.
iii) The eigenspaces of D form a complete orthonormal decomposition of
L2 (ΣM ), i.e.,

L2 (ΣM ) = Ker(D − λId).
λ∈ Spec(D)

iv) The set Spec(D) is unbounded on both sides of R and, if moreover n ≡


3 (4), then it is symmetric about the origin.

Proof : If M is compact without boundary, then the statements i)−iii) follow


from the classical spectral theory of elliptic self-adjoint operators [173, Thm.
5.8 p.196], which is applicable here to the closure of D as a consequence of
Proposition 1.3.5.
1.3 The Dirac operator 15

As for the unboundedness of Spec(D) on both sides of R, we give the proof


presented in [11, Prop. 4.30] and which consists in the following. Assume
it were wrong for n ≥ 3 (for n = 1, 2 the spectrum of D is symmetric
about 0, see Note 2.1.2.1 and end of proof respectively). Since we already
know that Spec(D) is unbounded, this would mean that either Spec(D) ⊂
[m, +∞[ or Spec(D) ⊂] − ∞, m] for some m ∈ R. Up to changing D into
−D, one can assume that Spec(D) ⊂ [m, +∞[, i.e., that (Dϕ, ϕ) ≥ mϕ2
for all ϕ ∈ Γ(ΣM ) (see e.g. Lemma 5.0.2 for a formulation of the min-max
principle). Consider a finite open covering (Ωk )k of M such that T M|Ωk is
trivial for every k and choose a partition of unity (χk )k subordinated
 to that
covering, that is, χk ∈ C ∞ (M, [0, 1]) with supp(χk ) ⊂ Ωk and k χk = 1.
Let (ej,k )1≤j≤n be an orthonormal frame trivializing T M|Ωk and set e j,k :=

χk ej,k for all j, k. Note that the e
j,k are globally defined smooth vector
fields on (M n , g) satisfying
 
e
j,k · e
j,k · = χk ej,k · ej,k ·
j,k j,k

=− χk Id
j,k
= −nId

and analogously
 
e
j,k · ∇e
j,k
= χk ej,k · ∇ej,k
j,k j,k

= χk D
k
= D.

Hence for any section ϕ of ΣM one has


 (1.12) 

ej,k · ϕ), e
(D( j,k · ϕ) = − (
ej,k · Dϕ, e
j,k · ϕ)
j,k j,k

− 2(∇e
j,k
ϕ, e
j,k · ϕ)



ej,k · ϕ, e
+ ((d + δ) j,k · ϕ)

= −n(Dϕ, ϕ) + 2(Dϕ, ϕ)
 
− ej,k ∧ d
(( ej,k ) · ϕ, ϕ)
j,k

≤ −(n − 2)(Dϕ, ϕ) + Cϕ2 ,


16 1 Basics of spin geometry
 
where C :=max |
j,k ej,k ∧ d
ej,k | is a finite nonnegative constant indepen-
M
dent of ϕ. The assumption implies that, if ϕ is a non-zero eigenvector for D
associated to the eigenvalue λ, then

(−(n − 2)λ + C)ϕ2 ≥ m ej,k · ϕ, e
( j,k · ϕ)
j,k

= nmϕ2 ,

which leads to a contradiction for λ large enough, QED.


As for the symmetry of Spec(D) about 0, it straightforward follows from
Proposition 1.3.2 (i.e., from (1.10)) in the case where n is even whereas for
n ≡ 1 (4) it can be deduced from the existence of a real or quaternionic
parallel structure on ΣM anti-commuting with the Clifford multiplication
and hence with the Dirac operator [88, Prop. p.31]. This concludes the
proof. 

From Corollary 2.1.5 below, the Dirac spectrum of the n-dimensional real
projective space RPn endowed with its round metric of sectional curvature 1
and one of its both spin structures is { n2 + n1 + 2k, − n2 − n2 − 2k, k ∈ N},
where n1 is the mod 2 reduction of n−3 n+1
4 and n2 that of 4 (here the multi-
plicities are not taken into account). Thus the symmetry property of Spec(D)
from Theorem 1.3.7.iv) breaks in dimension n ≡ 3 (4).
Eigenvectors for D associated to the eigenvalue 0 are called harmonic
spinors. Parallel spinors are harmonic but the converse is false in general.
Moreover, unlike that of harmonic forms, the number of linearly independent
harmonic spinors generally varies under a change of metric, see Section 6.2.
Turning to the square of the Dirac operator, its principal symbol is given by
ξ → −g(ξ, ξ)Id, which is exactly that of the rough Laplacian. The difference
between both must therefore be a linear differential operator of order at most
one. In fact, it turns out to be a very simple curvature expression.

Theorem 1.3.8 ([217, 175]) The Dirac operator D of a Riemannian spin


manifold (M n , g) satisfies the Schrödinger-Lichnerowicz formula

S
D 2 = ∇∗ ∇ + Id, (1.15)
4

where S is the scalar curvature of (M n , g).

Proof : Fix a local orthonormal basis {ej }1≤j≤n of T M , then using the com-
patibility relations as well as (1.5) one has, for any ϕ ∈ Γ(ΣM ),


n
D2 ϕ = ej · ∇ej (ek · ∇ek ϕ)
j,k=1
1.3 The Dirac operator 17


n
= ej · ∇ej ek · ∇ek ϕ + ej · ek · ∇ej ∇ek ϕ
j,k=1
n 
n
=− ej · ek · ∇∇ejek ϕ + ej · ek · ∇ej ∇ek ϕ
j,k=1 j,k=1

n
= (∇∇ejej − ∇ej ∇ej )ϕ
j=1

+ ej · ek · (∇ej ∇ek − ∇ek ∇ej − ∇∇ejek + ∇∇ekej )ϕ
1≤j<k≤n

= ∇∗ ∇ϕ − ej · ek · (∇[ej ,ek ] − [∇ej , ∇ek ])ϕ
1≤j<k≤n

1n
= ∇∗ ∇ϕ − ej · ek · Re∇j ,ek ϕ.
2
j,k=1

It now follows from Lemma 1.2.4 that



n
1
n
ej · ek · Re∇j ,ek ϕ = ej · Ric(ej ) · ϕ
2 j=1
j,k=1
S
= − ϕ,
2
which concludes the proof. 

In particular the square of the Dirac operator coincides with the rough Lapla-
cian acting on spinors as soon as the scalar curvature of the underlying
manifold vanishes. This provides an answer to Dirac’s original question on
Euclidean space.
Applications of the Schrödinger-Lichnerowicz formula to eigenvalue es-
timates are discussed in Chapters 3 and 4. Vanishing theorems can also
be obtained combining the Schrödinger-Lichnerowicz formula with the cel-
ebrated Atiyah-Singer index theorem [28], stating that the topological and
the analytical index of any elliptic linear differential operator coincide. In the
case of the Dirac operator, the index theorem reads as follows.

Theorem 1.3.9 (M.F. Atiyah & I.M. Singer [28]) Let (M n , g) be an


even dimensional closed Riemannian spin manifold and ind(D+ ) be the an-
alytical index of the positive part of its Dirac operator. Then
),
ind(D+ ) = A(M

)=
where A(M M ] ∈ Z is the A-genus
A[T of M .
M
18 1 Basics of spin geometry

Recall that, by definition, ind(D+ ) := dim(Ker(D+ )) − dim(Coker(D+ )).


Since D is formally self-adjoint (Proposition 1.3.4) and the formal adjoint of
D± is D∓ , one has ind(D+ ) = dim(Ker(D+ )) − dim(Ker(D− )). The A-class

of T M , denoted above by A[T M ], is the characteristic class associated to
√x
the Taylor expansion at 0 of the function x → 2
√ x . The A-genus of M
sinh( 2)
is by construction a rational number and it is already a highly non-trivial
statement that, for spin manifolds, it must be an integer. Theorem 1.3.9
can be proved using either K-theoretical methods (see [173, Chap. 3] and
references therein) or asymptotics for the heat kernel (see [58, Chap. 3] and
references therein). Beware that, if the dimension of the underlying manifold
is odd, then the index of its Dirac operator - as well as that of any elliptic
linear differential operator - vanishes, see e.g. [173, Thm. 13.12].
We end this section with the remarkable property of conformal covariance
of the Dirac operator.

Proposition 1.3.10 Let the spin structure be fixed and denote by D the
Dirac operator of M for the conformal metric g := e2u g, with u ∈ C ∞ (M, R).
Then there exists a unitary isomorphism between the spinor bundle of (M n , g)
and that of (M n , g) (denoted in the whole text by ϕ → ϕ) such that

D(e− ϕ) = e−
n−1 n+1
2 u 2 u Dϕ (1.16)

for every ϕ ∈ Γ(ΣM ), with D := Dg .

Proof : The isometry X → e−u X from (T M, g) onto (T M, g) defines a princi-


pal bundle isomorphism SOg (T M ) −→ SOg (T M ) lifting to the spin level.
More precisely, it induces a vector-bundle isomorphism Σg M −→ Σg M ,
ϕ −→ ϕ, preserving the pointwise Hermitian inner product and sending X · ϕ
onto e−u X· ϕ. As for the natural covariant derivative ∇ on the spinor bundle,
its local expression in terms of the Levi-Civita connection of g on T M (1.6)
immediately implies for all ϕ ∈ Γ(T M ) and X ∈ T M :

1 X(u)
∇X ϕ = ∇X ϕ − X · grad(u) · ϕ − ϕ. (1.17)
2 2

Since {e−u ej }1≤j≤n is a local o.n.b. of T M for g as soon as {ej }1≤j≤n is one
for g, we deduce that


n
Dϕ = e−2u ej · ∇ej ϕ
j=1

n 
(1.17) 1 ej (u)
= e−2u ej · ∇ej ϕ − ej · grad(u) · ϕ − ϕ
j=1
2 2
1.4 Spinors on hypersurfaces and coverings 19
n
 
1 ej (u)
= e−u ej · ∇ej ϕ − ej · ej · grad(u) · ϕ − ej · ϕ
j=1
2 2
n−1
= e−u (Dϕ + grad(u) · ϕ). (1.18)
2
Hence
n − 1 − n−1 u
D(e− e−u (D(e−
n−1 n−1
2 u ϕ) = 2 u ϕ) + e 2 grad(u) · ϕ)
2
(1.11) n − 1 − n−1 u −u
e 2 e grad(u) · Dϕ + e− 2 u e−u Dϕ
n−1
= −
2
n − 1 − n−1 u −u
+ e 2 e grad(u) · ϕ
2
e− 2 u Dϕ,
n+1
=

which concludes the proof. 

1.4 Spinors on hypersurfaces and coverings

In this section we discuss how spinors can be induced on submanifolds or


quotients. We restrict ourselves to the case of Riemannian hypersurfaces
and coverings, see e.g. [40, 105] and [198, 15, 119] for higher codimensional
submanifolds and submersions or foliations respectively. The case of general
homogeneous spaces, which is much more involved, is handled separately in
Section 2.2. In order to simplify the notations, we denote in this survey, for
the situation where M is a hypersurface in some spin manifold M , by “·” the
 and by “ · ” that of M .
Clifford multiplication of M
M

Proposition 1.4.1 Let ι : M −→ M  be an immersed oriented Riemannian


hypersurface in an (n + 1)-dimensional Riemannian spin manifold. Let the
unit normal ν ∈ Γ(T ⊥ M ) be chosen such that, for every local oriented basis
| is oriented. Then
{v1 , . . . , vn } of T M , the local basis {v1 , . . . , vn , ν} of T M M
the manifold M is spin and carries an induced spin structure for which a
unitary isomorphism exists

 ΣM if n is even

ΣM|M −→ 
ΣM ⊕ ΣM if n is odd
ϕ −→ ϕ

| =
where, for n odd, the two copies of ΣM correspond to the splitting ΣM M
| ⊕ Σ− M
Σ+ M | . Moreover this isomorphism can be chosen so as to satisfy
M M
20 1 Basics of spin geometry
 C
 ωn · ϕ if n is even
 M 
iν · ϕ =  0 Id (1.19)
 Id 0 , if n is odd


X · ϕ if n is even

X ·ν·ϕ= M (1.20)
 (X · ⊕ − X · )ϕ if n is odd
M M

and
 X ϕ = ∇X ϕ + A(X) · ν · ϕ
∇ (1.21)
2
for all X ∈ T M and ϕ ∈ Γ(ΣM | ), where ω C is the complex volume form
M n
(see (1.3) and A := −∇ν  denotes the Weingarten endormorphism field of ι.
In particular the fundamental Dirac operators D and D  of M and M
 res-
pectively are related through

Dϕ ν ϕ + ν · (D2 − nH )ϕ,
 =ν·∇ (1.22)
2
where 
D if n is even
D2 := 
D ⊕ −D if n is odd
1
and H := n tr(A) is the mean curvature of ι.
, it is spin. This can be seen as
Proof : Since M has trivial normal bundle in M
a consequence of a more general result [194] or, alternatively, in the follwing
way: the pull-back of Spin(T M )| to SO(T M ) over the map “completion
M
by ν”

)|
SO(T M ) −→ SO(T M M

(e1 , . . . , en ) −→ (e1 , . . . , en , ν)

provides a 2-fold covering of SO(T M ) which can easily be proved to be a


Spinn -bundle hence a spin structure on M . The identity (1.20) is just the
geometric translation of the canonical embedding of the Clifford algebra in
dimension n into that in dimension n + 1, see e.g. [134, Prop. 2.7]. The local
formula (1.6) defining the compatible covariant derivative combined with
 X Y = ∇X Y + g(A(X), Y )ν (for all
the classical Gauss-Weingarten identity ∇
X, Y ∈ Γ(T M )) lead to (1.21). The last equality (1.22) is a straightforward
consequence of both (1.20) and (1.21). 

In particular a hypersurface in a spin manifold is spin as soon as it is ori-


entable. For example the round sphere Sn is from its definition a hypersurface
in Rn+1 , which is obviously spin, therefore it is also spin and carries an in-
duced spin structure. This spin structure is unique if n ≥ 2 since Sn is then
1.4 Spinors on hypersurfaces and coverings 21

simply-connected, however there exists another spin structure on S1 , see Ex-


ample 1.4.3.1 below. Note also that, as a consequence of (1.19), (1.20) and
(1.21), the Clifford action of ν onto ΣM or ΣM ⊕ΣM respectively (according
to the parity of n) is ∇-parallel, in particular it anti-commutes with D2 : for
every ϕ ∈ Γ(ΣM ) (or in Γ(ΣM ⊕ ΣM ) if n is odd),

D2 (ν · ϕ) = −ν · D2 ϕ. (1.23)

Turning to coverings, we have the following proposition, which is also well-


known (see e.g. [200, Lemma 7.3]).
Proposition 1.4.2 Let Γ× M →M  be a properly discontinuous free isomet-
ric and orientation-preserving action of a discrete group Γ on a Riemannian
spin manifold (Mn , g). Assume that the induced principal-bundle-action of
Γ on SO(T M ) lifts to a principal-bundle-action on Spin(T M) such that the
following diagram commutes:

)
Γ × Spin(T M )
Spin(T M

Id×η η 
M

)
Γ × SO(T M )
SO(T M

Then the following holds:


i) The manifold M := Γ\M  is spin and carries an induced spin structure
 provides a unitary isomorphism
for which the Γ-action on ΣM


∼ Γ\ΣM
ΣM =

preserving the Clifford multiplication and the natural compatible covari-


ant derivative.
ii) This unitary isomorphism identifies the sections of ΣM on M with the
 on M
Γ-equivariant sections of ΣM , i.e.,

Γ(ΣM ) ∼ ) |
= {ϕ ∈ Γ(ΣM ϕ(γ · x) = γ · ϕ(x) ∀x ∈ M, ∀γ ∈ Γ},
(1.24)
. In particular, the
where we also denote by “ · ” the action of Γ on ΣM
eigenvectors of the Dirac operator on M identify with those of the Dirac
operator on M satisfying (1.24).
 is simply-connected, then the spin structures on M stand in one-
iii) If M
to-one correspondence with Hom(Γ, Z2 ).
22 1 Basics of spin geometry

) obviously defines a spin structure on M .


Proof : The left-quotient Γ\Spin(T M
Since Γ operates by principal-bundle-homomorphisms, its action commutes
with the right action of the structure groups, therefore one straightforward
obtains

) ×δ Σn ) = \ΣM
)) × Σ ∼ \(Spin(T M
ΣM = (Γ\Spin(T M .
δn n = Γ n Γ

For the same reason the Hermitian inner product of ΣM  remains preserved
by Γ - hence the above identification can be assumed to be unitary - and so
does the Clifford multiplication: for all x ∈ M , X ∈ Tx M , ϕ ∈ Σx M  and
γ ∈ Γ, γ · (X · ϕ) = (dγ(X)) · (γ · ϕ). The equivariance condition follows from
this observation. In the case where M  is simply-connected one has

H 1 (M, Z2 ) = Hom(H1 (M ), Z2 ) = Hom(π1 (M ), Z2 ) = Hom(Γ, Z2 ),

where we have used H1 (M ) = π1 (M )/[π1 (M ), π1 (M )]. This concludes the


proof. 

As noticed in Proposition 1.4.2.iii), the spin structure induced on the base


space is in general not unique: multiplying the action of Γ at the spin level
by any group homomorphism Γ −→ {±1} provides another group action
satisfying the assumptions of Proposition 1.4.2, however the spin structure
downstairs changes. Note also that the pull-back of any spin structure on
the base of a Riemannian covering is a spin structure on the total space and
sections on the base can always be lifted to equivariant sections upstairs.

Examples 1.4.3
1. Let M  := R and Γ := 2πZ acting on M  by translations. Since Spin(T R) =
R × Spin1 = R × {±1} there are only two possible lifts of the Γ-action to
the spin level which are determined by the image (−1)δ of the generator
2π of Γ in {±1}, where δ ∈ {0, 1}. We call the spin structure induced by
δ = 0 the trivial spin structure on S1 = 2πZ\R and the one induced by
δ = 1 the non-trivial one.
2. More generally, consider a lattice Γ ⊂ M  := Rn of rank n ≥ 1 and the
corresponding torus M := Γ\R with flat metric. The action of Γ by
n

translations induces the trivial group-homomorphism Γ −→ SOn , which


obviously lifts to the spin level, hence M is spin. Moreover, there ex-
ist 2n group-homomorphisms Γ −→ {±1}: fix a basis (γ1 , . . . , γn ) of Γ,
δ1 , . . . , δn ∈ {0, 1} and define the group-homomorphism εδ1 ,...,δn : Γ −→
{±1} by εδ1 ,...,δn (γj ) := (−1)δj for all 1 ≤ j ≤ n. The spin structure on M
induced by εδ1 ,...,δn is called (δ1 , . . . , δn )-spin structure. We obtain in this
way all spin structures on the torus.
3. Let n ≥ 2 and Γ be a finite subgroup of SOn+1 acting freely on Sn . For
n even there is obviously no non-trivial such subgroup, thus we assume n
1.5 Elliptic boundary conditions for the Dirac operator 23

odd. On the round sphere both bundles SO(T M ) and Spin(T M ) canoni-
cally identify to SOn+1 and Spinn+1 respectively. Therefore, the existence
of a lift of the action of Γ to the spin level is equivalent to that of a group ho-
momorphism  : Γ −→ Spinn+1 such that ξ ◦  is the inclusion Γ ⊂ SOn+1 .
If this is fulfilled then from Proposition 1.4.2.iii) there are as many spin
structures on Γ\S as there are group homomorphisms Γ −→ {±1}. For
n

example consider Γ = Z2 = {±Id}. Of course it preserves the orientation


of Sn only if n is odd. Furthermore, the pre-image of −Id under ξ can be
shown to be ±e1 · . . . · en+1 , which is involutive if and only if n ≡ 0 (4) or
n ≡ 3 (4). Hence the n-dimensional real projective space RPn is spin only
for n ≡ 3 (4) and in that case it admits two spin structures.

1.5 Elliptic boundary conditions for the Dirac operator

We end this chapter by briefly describing four types of boundary conditions


for the Dirac operator on manifolds M with non-empty boundary ∂M and
discuss their ellipticity and self-adjointness. We mostly use the notations of
[139], which is a good reference on the topic. A more general approach where
less regularity is required can be found in the seminal paper [32].
Apart from the 1-dimensional case, the Dirac operator on a compact mani-
fold with boundary is in general not Fredholm since it has infinite dimensional
kernel; for example, given any domain Ω of C, the kernel of the Dirac opera-
tor of Ω coincides with the direct sum of the space of holomorphic functions
with that of anti-holomorphic functions on Ω. The purpose of elliptic bound-
ary conditions for D is precisely to make it Fredholm. This is done in terms
of pseudo-differential operators on the boundary. To avoid technicalities, we
shortcut the original definition [139, p.380] of elliptic boundary condition
providing the following one [139, Prop. 1]:

Definition 1.5.1 Let (M n , g) be a Riemannian spin manifold with non-


empty boundary ∂M . Denote Σ := Σ∂M if n is even and Σ∂M ⊕ Σ∂M
if n is odd.
i) An elliptic boundary condition for D is a pseudo-differential operator B :
L2 (Σ) −→ L2 (V ), where V is some Hermitian vector bundle on ∂M , such
that the boundary value problem

 Dϕ = Φ on M
 (1.25)
 B(ϕ| ) = χ on ∂M
∂M

has smooth solutions up to a finite-dimensional kernel for any given smooth


data Φ ∈ Γ(ΣM ) and χ ∈ Γ(V ) belonging to a certain subspace with finite
codimension.
24 1 Basics of spin geometry

ii) An elliptic boundary condition for D is called self-adjoint if and only if


the restriction of D onto {ϕ ∈ Γ(ΣM ) | B(ϕ|∂M ) = 0} is symmetric.
As a consequence of the Fredholm alternative, one may talk about the spec-
trum of D if an elliptic boundary condition is fixed [139, Prop. 1]:

Theorem 1.5.2 Let (M n , g) be a compact Riemannian spin manifold with


non-empty boundary ∂M . Let B be an elliptic boundary condition for D.
Then the eigenvalue problem

 Dϕ = λϕ on M

 B(ϕ| ) = 0 on ∂M
∂M

has a discrete spectrum with finite dimensional eigenspaces in Γ(ΣM ), un-


less the spectrum is C itself. Moreover, if B is self-adjoint, then the Dirac
spectrum is real.
In view of the study of spectral properties of the Dirac operator four kinds
of boundary conditions have been mainly considered so far:
• generalized Atiyah-Patodi-Singer (gAPS) boundary condition (depending
on a fixed β ∈ R that we omit in the notations) [75]: define B := BgAPS to
be the L2 -orthogonal projection onto the subspace spanned by the eigen-
vectors of D2 (which is D∂M or D∂M ⊕ −D∂M according to the dimension)
to eigenvalues not smaller than β. For β = 0 this condition is called the
Atiyah-Patodi-Singer (APS) boundary condition and was originally intro-
duced by Atiyah, Patodi and Singer to prove index theorems on manifolds
with boundary, see e.g. [82, 143].
• Boundary condition associated to a chirality (CHI) operator (see e.g. [82,
143]): define the endomorphism-field BCHI := 12 (Id − ν · G) of Σ, where
ν is a unit normal on ∂M and G is a chirality operator (i.e., it is the
restriction on ∂M of an endomorphism-field G of ΣM which is involutive,
unitary, parallel and anti-commuting with the Clifford multiplication on
M ). Natural chirality operators appear in case n is even (then define G :=
ωnC ·, where ωnC is the complex volume form of M , see (1.3)) or if M is
itself a spacelike hypersurface in a Lorentzian manifold (then define G to
be the Clifford multiplication by a unit timelike normal vector field to M ).
Among others, this boundary condition has been used to prove positive
mass theorems in the presence of black holes, see references in [139].
• MIT bag boundary condition (see e.g. [144]): define the endomorphism-
field BMIT of Σ by BMIT := 12 (Id − iν·). It was first introduced in the
Lorentzian context by physicists at the MIT for the description of spin
1
2 -particles, see references in [139].
• modified generalized Atiyah-Patodi-Singer ( (mgAPS) boundary condition
(also depending on some fixed β ∈ R) [75]: define BmgAPS := BgAPS (Id +
ν·). In the particular case where β = 0 this condition is called the modified
Atiyah-Patodi-Singer (mAPS) boundary condition [139].
1.5 Elliptic boundary conditions for the Dirac operator 25

To test the ellipticity of a boundary condition, practical criteria are available


[139, Prop. 1]:

Proposition 1.5.3 Let (M n , g) be a compact Riemannian spin manifold


with non-empty boundary ∂M . A pseudo-differential operator B : L2 (Σ) −→
L2 (V ), where V is some Hermitian vector bundle on ∂M , is an elliptic
boundary condition for D if and only if its principal symbol b : T ∗ ∂M −→
Hom(Σ, V ) satisfies the following two conditions:
1
a) Ker(b(ξ)) ∩ Ker(iξ · ν · −g(ξ, ξ) 2 Id) = {0}
n
b) dim(Im(b(ξ)) = rkC (ΣM
2
)
= 2[ 2 ]−1
for every ξ ∈ T ∗ ∂M \ {0}, where ν denotes the inner unit normal.
The central result of this section is the following.

Proposition 1.5.4 Let (M n , g) be a compact Riemannian spin manifold


with non-empty boundary ∂M . Then the gAPS, CHI, MIT bag and mgAPS
boundary conditions are elliptic. Moreover, the spectrum of D is a discrete
unbounded sequence which is real for
• the gAPS and mgAPS boundary conditions with β ≤ 0,
• the CHI boundary condition,
and which is contained in the upper half of C for the MIT bag boundary
condition.

Proof : We consider the four conditions separately.


• gAPS boundary condition: It can be shown (see e.g. [59, Prop. 14.2]) that
the principal symbol of BgAPS is given on any ξ ∈ T ∗ ∂M \ {0} by bgAPS (ξ) =
1
2 (iξ · ν · +g(ξ, ξ) Id), i.e., it is the (pointwise) orthogonal projection onto the
1 2

eigenspace of the Clifford-multiplication by iξ (on ∂M , which corresponds to


1
iξ · ν·, cf. (1.20)) to the eigenvalue g(ξ, ξ) 2 . For ξ = 0 the image of bgAPS (ξ) is
[n
obviously 2 2 -dimensional. On the other hand, Ker(bgAPS (ξ)) = Ker(iξ ·
]−1
1
ν · +g(ξ, ξ) 2 Id), so that the criterium a) of Proposition 1.5.3 is also fulfilled,
hence the gAPS boundary condition is elliptic.
We now have to show that, for any ϕ, ψ ∈  Γ(ΣM ) satisfying
BgAPS (ϕ|∂M ) = BgAPS (ψ|∂M ) = 0 then M Dϕ, ψ
vg = M ϕ, Dψ
vg holds.
But from (1.14) and Green’s formula we know that, for all ϕ, ψ ∈ Γ(ΣM ),
 
Dϕ, ψ
− ϕ, Dψ
vg = ϕ, ν · ψ
vg∂M . (1.26)
M ∂M

Therefore we have to prove that BgAPS (ϕ|∂M ) = BgAPS (ψ  |∂M ) = 0 implies


the vanishing of ∂M ϕ, ν · ψ
vg∂M . Denote by (· , ·)M := M · , ·
vgM the L2 -

inner product on Γ(ΣM ), (· , ·)∂M := ∂M · , ·
vg∂M that on Γ(Σ∂M ) and for
any real numbers c < d by π≥c (resp. π>c , π<c , π≤c , π[c,d] ) the L2 -orthogonal
projection onto the subspace spanned by the eigenvectors of D2 associated to
26 1 Basics of spin geometry

the eigenvalues lying in the interval [c, +∞[ (resp. ]c, +∞[, ] − ∞, c[, ] − ∞, c],
[c, d]). Since BgAPS = π≥β , we have, for β ≤ 0 and all ϕ, ψ ∈ Γ(ΣM ) satisfying
BgAPS (ϕ|∂M ) = BgAPS (ψ|∂M ) = 0:
 
(ϕ, ν · ψ)∂M = π<β (ϕ), π<β (ν · ψ)
∂M
(1.23)
 
= (π<β (ϕ), ν · π>−β (ψ)
   ∂M
0
= 0.

• CHI boundary condition: The endomorphism-field ν · G of Σ is by definition


of G unitary, Hermitian and involutive, therefore BCHI is nothing but the
pointwise orthogonal projection onto its eigenspace to the eigenvalue −1,
n
which is 2[ 2 ]−1 -dimensional. Moreover since it is a differential operator of
zero order its principal symbol is the operator itself, therefore criterium b)
of Proposition 1.5.3 is fulfilled. On the other hand, for any ξ ∈ T ∗ ∂M \ {0},
the endomorphisms iξ · ν· and ν · G of Σ obviously anti-commute, so that, for
1
any ϕ ∈ Ker(Id − ν · G) ∩ Ker(iξ · ν · −g(ξ, ξ) 2 Id),
1
g(ξ, ξ) 2 ϕ = iξ · ν · ϕ
= iξ · ν · ν · Gϕ
= −ν · G(iξ · ν · ϕ)
1
= −g(ξ, ξ) 2 ν · Gϕ
1
= −g(ξ, ξ) 2 ϕ,

which implies ϕ = 0. Hence the criterium a) of Proposition 1.5.3 is also


fulfilled. This shows the ellipticity of the CHI boundary condition.
Let now ϕ, ψ ∈ Γ(ΣM ) satisfying BCHI (ϕ|∂M ) = BCHI (ψ|∂M ) = 0, then

(ϕ, ν · ψ)∂M = (ν · Gϕ, ν · ψ)∂M


= (ϕ, Gψ)∂M
= −(ϕ, ν · G 2 ψ)∂M
= −(ϕ, ν · ψ)∂M ,

hence (ϕ, ν · ψ)∂M = 0 and the spectrum of D must therefore be real.


• MIT bag boundary condition: The endomorphism-field iν· of Σ is unitary,
Hermitian and involutive, so that BMIT is the pointwise orthogonal projection
n
onto its eigenspace to the eigenvalue −1, which is 2[ 2 ]−1 -dimensional. More-
over, for any ξ ∈ T ∗ ∂M \ {0}, the endomorphisms iν· and iξ · ν· of Σ
anti-commute, so that the same arguments as for the CHI boundary con-
dition apply for the ellipticity.
1.5 Elliptic boundary conditions for the Dirac operator 27

Let now ϕ, ψ ∈ Γ(ΣM ) satisfying BMIT (ϕ|∂M ) = BMIT (ψ|∂M ) = 0, then


this time (ϕ, ν · ψ)∂M does not vanish in general. However (1.14) with ϕ = ψ
gives
   
2im Dψ, ψ
vg = ψ, ν · ψ
vg∂M = i |ψ|2 vg∂M ,
M ∂M ∂M

therefore any eigenvalue λ of D with associated (non-zero) eigenvector ψ


|ψ|2 v ∂M
must satisfy m(λ) = 2∂M  g
|ψ|2 vg
. If ψ|∂M = 0 then a unique continuation
M
property for the Dirac operator [59, Sec. 1.8] would imply ψ = 0 on M ,
contradiction. Therefore m(λ) > 0.
• mgAPS boundary condition: Since Id + ν· is an isomorphism-field of Σ,
the principal symbol of BmgAPS evaluated on any vector ξ ∈ T ∗ ∂M \ {0} is
n
bmgAPS (ξ) = bgAPS (ξ) ◦ (Id + ν·), hence it has rank 2[ 2 ]−1 . Moreover,
1
Ker(bmgAPS (ξ)) = (Id − ν·)Ker(bgAPS (ξ)) = (Id − ν·)Ker(iξ · ν · +g(ξ, ξ) 2 Id).
1
Let ϕ ∈ Ker(bmgAPS (ξ)) ∩ Ker(iξ · ν · −g(ξ, ξ) 2 Id), then there exists a ψ ∈
1
Ker(iξ · ν · +g(ξ, ξ) 2 Id) with ϕ = (Id − ν·)ψ so that
1 1
g(ξ, ξ) 2 (Id − ν·)ψ = g(ξ, ξ) 2 ϕ
= iξ · ν · ϕ
= (Id + ν·)iξ · ν · ψ
1
= −g(ξ, ξ) 2 (Id + ν·)ψ,

from which one deduces that ψ = 0 hence ϕ = 0. This shows the ellipticity
of the mgAPS boundary condition.
Let now ϕ, ψ ∈ Γ(ΣM ) satisfying BmgAPS (ϕ|∂M ) = BgAPS (ψ|∂M ) = 0,
then with the notations introduced above for the gAPS boundary condition
and for β ≤ 0 one has:
 
2(ϕ, ν · ψ)∂M = {Id + ν·}ϕ, {Id + ν·}ν · ψ
 ∂M

= π<β ({Id + ν·}ϕ), π<β ({Id + ν·}ν · ψ)
∂M
(1.23)
 
= π<β ({Id + ν·}ϕ), ν · π>−β ({Id + ν·}ψ )
   ∂M
0
= 0,

which shows that the spectrum of D under the mgAPS boundary condition
is real and concludes the proof. 
Chapter 2
Explicit computations of spectra

In this chapter we present the few known closed Riemannian spin mani-
folds whose Dirac spectrum - or at least some eigenvalues - can be explicitly
computed.

2.1 Spectrum of some non-negatively curved


spaceforms

We begin with the examples where no machinery is required. The simplest


ones are the flat tori. The description of the (δ1 , . . . , δn )-spin structure on
the n-torus is explained in Example 1.4.3.2.
Theorem 2.1.1 (T. Friedrich [86]) For a positive integer n let Γ ⊂ Rn be
a lattice and M = Tn := Γ\R the corresponding n-dimensional torus. Fix a
n

basis (γ1 , . . . , γn ) of Γ and δ1 , . . . , δn ∈ {0, 1}.


Then the spectrum of the Dirac operator of Tn endowed with the induced flat
metric and the (δ1 , . . . , δn )-spin structure is given by


1
n

± 2π|γ + δj γj∗ |, γ ∗ ∈ Γ∗ ,
2 j=1

where Γ∗ := {θ ∈ (Rn )∗ | θ(Γ) ⊂ Z} is the dual lattice and (γ1∗ , . . . , γn∗ ) the
basis of Γ∗ dual to (γ1 , . . . , γn ). Furthermore, if non-zero, the eigenvalue pro-
vided by γ ∗ has multiplicity 2[ 2 ]−1 . In case δ1 = . . . = δn = 0 the multiplicity
n

[n ]
of the eigenvalue 0 is 2 2 .
Beware that the multiplicities add if the corresponding eigenvalues
n are
equal. Thus, the multiplicity of the eigenvalue ±2π|γ ∗ + 12 j=1 δj γj∗ |
n
is always atleast 2[ 2 ] (even for n = 1): if it is non zero, then
γ ∗ := −γ ∗ − j=1 δj γj∗ ∈ Γ∗ provides the same eigenvalue and γ ∗ = γ ∗ .
n

N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 29


DOI 10.1007/978-3-642-01570-0 2, 
c Springer-Verlag Berlin Heidelberg 2009
30 2 Explicit computations of spectra

Proof of Theorem 2.1.1: For any f ∈ C ∞ (Rn , Σn ), the equivariance condition


(1.24) reads
f (x + γj ) = (−1)δj f (x)
for all x ∈ R
n
and 1 ≤ j ≤ n. Given γ ∗ ∈ Γ∗ , we denote by θγ the constant 1-
form γ + 2 j=1 δj γj∗ on Rn . For an arbitrary orthonormal basis (σl )1≤l≤2[ n2 ]
∗ 1 n

of Σn - which we trivially extend onto Rn as sections of Σ(Rn ) = Rn × Σn -


and 1 ≤ l ≤ n, we define the spinor field

φγ,l := e2iπθγ σl (2.1)

on Rn . It obviously satisfies the equivariance condition and, for any X ∈ Rn ,

∇X φγ,l = X(φγ,l )
= 2iπθγ (X)e2iπθγ σl
= 2iπθγ (X)φγ,l ,

so that, choosing a local orthonormal basis (ek )1≤k≤n of Rn


n
Dφγ,l = ek · ∇ek φγ,l
k=1

n
= 2iπ θγ (ek )ek · φγ,l
k=1
= 2iπθγ · φγ,l .

If θγ = 0, which only happens if γ ∗ = 0 and δ1 = . . . = δn = 0, the


spinor φγ,l = σl provides an eigenvector of D associated to the eigenvalue 0.
Moreover, as a consequence of the Schrödinger-Lichnerowicz formula (1.15),
the kernel of D consists of parallel spinors - as it does whenever the scalar
curvature of the closed manifold vanishes. We deduce in that case that 0 is
n
an eigenvalue of D with multiplicity 2[ 2 ] .
θ
If θγ = 0 and n = 1 then i |θγγ | · = Id or −Id on Σ1 = C, therefore Dφγ,l =
2π|θγ |φγ,l or −2π|θγ |φγ,l , i.e., φγ,l is a non-zero eigenvector of D associated to
the eigenvalue 2π|θγ | or −2π|θγ |. Both eigenvalues occur because, as noticed
above, θ−γ ∗ −nj=1 δj γj∗ = −θγ . Each eigenvalue has multiplicity 1.
θ
If θγ = 0 and n ≥ 2 we consider the Clifford action of i |θγγ | on Σn . It is
involutive, parallel and unitary, hence it induces the orthogonal and parallel
splitting
θγ θγ
Σn = Ker(i · −Id) ⊕ Ker(i · +Id),
|θγ | |θγ |
where both spaces on the r.h.s. have the same dimension since the Clifford
actions of two orthogonal vectors anti-commute. We replace in that case
2.1 Spectrum of some non-negatively curved spaceforms 31

(σl )1≤l≤2[ n2 ] by an orthonormal basis (σ1+ , . . . , σ +[ n2 ]−1 , σ1− , . . . , σ −[ n2 ]−1 ) of Σn ,


2 2
θ
where each (σ1 , . . . , σ [ n2 ]−1 ) is a constant orthonormal basis of Ker(i |θγγ | ·
2
−Id) ⊂ Σn for  ∈ {±1}. We redenote the φγ,l of (2.1) by φγ,l . The above
computations apply with σl instead of σl , so that

Dφγ,l = 2iπ(−i|θγ |)φγ,l


= 2π|θγ |φγ,l ,

i.e., φγ,l is a non-zero eigenvector of D associated to the eigenvalue 2π|θγ |.


Since the sections φγ,1 , . . . , φ [ n2 ]−1 are linearly independent, the multiplicity
γ,2
n
of the eigenvalue 2π|θγ | is at least 2[ 2 ]−1 .

To conclude the proof, it remains to remember that the {eiγ , γ ∗ ∈ Γ∗ }
()
form a Hilbert basis of L2 (Tn , C) and therefore so do the φγ,l in L2 (ΣTn ). 

Notes 2.1.2
1. For n = 1, Theorem 2.1.1 reads as follows: the Dirac spectrum of the
circle S1 (L) of length L > 0 (for L = 2π we just write S1 ) and the δ-spin
structure, where δ ∈ {0, 1}, is 2π L ( 2 + Z). Furthermore, each eigenvalue is
δ

simple.
2. It is remarkable that the kernel of the Dirac operator of (Tn , gflat ) is not
reduced to 0 for the (0, . . . , 0)-spin structure (called sometimes the triv-
ial spin structure) whereas it is for all other ones. Flat tori are thus the
most simple-minded examples of closed manifolds with non-zero harmonic
spinors for some spin structure. Moreover, as already noticed in the proof of
Theorem 2.1.1, the kernel of D actually consists of parallel spinors. There-
n
fore, (Tn , gflat ) admits a 2[ 2 ] -dimensional space of parallel spinors for the
trivial spin structure and no non-zero one otherwise. The reader interested
in basic results as well as the classification of complete Riemannian spin
manifolds with parallel spinors should refer to Section A.4.

Only few spectra of closed flat manifolds are known, although such manifolds
are always covered by flat tori as a consequence of Bieberbach’s theorems.
This is due to the high complexity of the groups involved, which makes
the search for equivariant eigenvectors very technical. Up to now, only di-
mension 3 (F. Pfäffle [206]) and some particular cases in higher dimensions
(R. Miatello and R. Podestá [188]) have been handled using representation-
theoretical methods, see Section 2.2.
A rather different technique leads to the Dirac spectrum of round spheres.

Theorem 2.1.3 (S. Sulanke [226], see also [38, 72, 228, 229])
Consider, for n ≥ 2, the round sphere M = Sn := {x ∈ Rn+1 , |x| = 1} with
its canonical metric g of constant sectional curvature 1 and its canonical
spin structure.
32 2 Explicit computations of spectra

Then the spectrum of the Dirac operator is {±( n2 + k), k ∈ N} and each
n n + k − 1
eigenvalue ±( n2 + k) has multiplicity 2[ 2 ] · .
k

Proof : We present here C. Bär’s proof [38, Sec. 2], which has the advantage
to get to the result in a very elementary way. It is based on the knowledge of
the spectrum of the scalar Laplacian and on the trivialization of the spinor
bundle of Sn through either − 12 - or 12 -Killing spinors, see Example A.1.3.2.
Let ϕ be a non-zero 2ε -Killing spinor on (Sn , can), where ε ∈ {±1}. Then
n ∞ n
Dϕ = 2ε j=1 ej · ej · ϕ = − nε 2 ϕ, so that, for every f ∈ C (S , R),

(1.13)
D2 (f ϕ) = f D2 ϕ − 2∇grad(f ) ϕ + (Δf )ϕ
n2
= f ϕ − εgrad(f ) · ϕ + (Δf )ϕ
4
2
(1.11) n
= f ϕ − ε(D(f ϕ) − f Dϕ) + (Δf )ϕ
4
n2 n
= ( − )f ϕ − εD(f ϕ) + (Δf )ϕ,
4 2
2
that is, ((D + 2ε Id)2 − 14 Id)(f ϕ) = ( n4 − n2 )f ϕ + (Δf )ϕ, or, equivalently,

ε n−1 2
(D + Id)2 (f ϕ) = (Δf + ( ) f )ϕ. (2.2)
2 2

The spectrum of the scalar Laplacian Δ on (Sn , can) is given by (see e.g.
[74]) {k(n + k − 1) | k ∈ N}, where the eigenvalue k(n + k − 1) appears with
n + k − 1
n+k−1 ·
multiplicity n+2k−1
k
. Since the spinor bundle ΣSn of Sn is tri-
vialized by 2ε -Killing spinors one deduces from (2.2) that, if {fk }k∈N denotes
a L2 -orthonormal basis of eigenfunctions of Δ on Sn and {ϕj }1≤j≤2[ n2 ] a tri-
vialization of ΣSn through a pointwise orthonormal basis, then {fk ϕj | k ∈ N,
n
1 ≤ j ≤ 2[ 2 ] } provides a complete L2 -orthonormal basis of L2 (ΣSn ) made out
of eigenvectors of (D + 2ε Id)2 associated to the eigenvalues ( n−1 2
2 + k) with
n  n + k − 1
k ∈ N, each of those having multiplicity 2[ 2 ] · n+2k−1
n+k−1 · . Therefore
k
the spectrum of D is contained in − 2 ± ( 2 + N), where the multiplicities
ε n−1

remain to be determined. We name the possible eigenvalues of D


n
λ+
k := +k
2
n
−k−1 := −
λ+ −k+1
2
n
λ−
k := − −k
2
2.1 Spectrum of some non-negatively curved spaceforms 33
n
λ−
−k−1 := + k − 1,
2

with k ∈ N, and denote by m(·) their corresponding multiplicity (both ε = ±1


have to be taken into account). It can already be deduced from the splitting

ε n−1 ε n−1
Ker((D + Id)2 − ( + k)2 Id) = Ker(D + ( − − k)Id)
2 2 2 2
 ε n−1
Ker(D + ( + + k)Id)
2 2
that, for every k ∈ N,

n + 2k − 1 n+k−1
m(λ±
k) + m(λ±
−k−1 )
[n
=22] · · .
n+k−1 k

n+k−1
Next we show by induction on k that m(λ±
k) =2 · [n
2] .
k
− n
For k = 0, both λ+ 0 and λ0 have multiplicity 2
[2]
because, as we have seen
ε
above, 2 -Killing spinors are eigenvectors for D associated to the eigenvalue
±
− nε
2 . Alternatively the eigenvalues λ−1 = ∓( 2 − 1) cannot appear because
n

otherwise Friedrich’s inequality (3.1) would be violated.


If the result is true for k, then the identity just above implies

n + 2k + 1 n+k
m(λ± − m(λ±
n
k+1 ) = 2
[2]
· · −k−2 )
n+k k+1

n + 2k + 1 n+k
− m(λ∓
n
= 2[ 2 ] · · k)
n+k k+1
n

n + 2k + 1 n + k  n + k − 1 
= 2[ 2 ] · · −
n+k k+1 k

n n+k
= 2[ 2 ] · ,
k+1

which was to be shown. 

On spaceforms of positive curvature the Dirac spectrum can be determined


thanks to a handy formula which reads as follows. First assume such a space-
form (M n , g) = (Γ\S , can) to be spin, which is equivalent to n odd and
n

the existence of a group homomorphism  : Γ −→ Spinn+1 such that ξ ◦ 


is the inclusion map Γ ⊂ SOn+1 , see Example 1.4.3.3. Proposition 1.4.2
states that the eigenvectors of the Dirac operator on (M n , g) are exactly the
Γ-equivariant eigenvectors of the Dirac operator on (Sn , can). In particular
the Dirac spectrum of (M n , g) is included in that of (Sn , can), so that it is
enough to find the multiplicity m(·) of each eigenvalue ±( n2 + k). To that
extent one encodes them into the two following formal power series:
34 2 Explicit computations of spectra


+∞
n
F± (z) := m(±( + k))z k .
2
k=0

Theorem 2.1.4 (C. Bär [38]) For n ≥ 3 odd, let (M n , g) := (Γ\S , can)
n

be a Riemannian spin spaceform of constant sectional curvature 1 and with


spin structure fixed by . Then the Dirac eigenvalues of (M n , g) are contained
in {±( n2 + k), k ∈ N} with multiplicities given by

1  χ∓ ((γ)) − χ± ((γ)) · z
F± (z) = ,
|Γ| det(IdRn+1 − z · γ)
γ∈Γ

± ±
where χ± := tr(δn+1 ) : Spinn+1 −→ C is the character of δn+1 .

The proof of Theorem 2.1.4 relies on a similar formula for the Laplace
eigenvalues by A. Ikeda, see [38, Sec. 3]. As an application of Theorem 2.1.4,
one obtains the Dirac spectrum of real projective spaces. We keep the nota-
tions of Example 1.4.3.3.

Corollary 2.1.5 For n ≡ 3 (4) let M n := RPn = Z2 \S be the n-


n

dimensional real projective space of constant sectional curvature 1 and with


spin structure fixed by (−Id) = (−1)δ e1 · . . . · en+1 , where δ ∈ {0, 1}. Then
the spectrum of its Dirac operator is

n n−3 n n+1
{ + k, k ∈ N ∩ (δ + + 2Z)} ∪ {− − k, k ∈ N ∩ (δ + + 2Z)},
2 4 2 4

n n + k − 1
each eigenvalue corresponding to k having multiplicity 2[ 2 ] · .
k
±
Proof : On the one hand, χ± ((Id)) = tr(δn+1
n−1
(1)) = 2 2 , on the other hand,

χ± ((−Id)) = (−1)δ tr(δn± (e1 · . . . · en+1 ))


= (−1)δ+ 4 tr(δn± (ωnC ))
n+1

(1.4) n+1
= ±(−1)δ+ 4 tr(IdΣ± n
)
n+1 n−1
= ±(−1)δ+ 4 2 2 ,

so that Theorem 2.1.4 implies that

1  χ∓ ((γ)) − χ± ((γ)) · z
F± (z) =
2 det(IdRn+1 − z · γ)
γ∈Γ

12 2 − 2 2 z ∓(−1)δ+ 4 2 2 ∓ (−1)δ+ 4 2 2 z 


n−1 n−1 n+1 n−1 n+1 n−1

= +
2 (1 − z)n+1 (1 + z)n+1
2.2 Spectrum of some other homogeneous spaces 35

n−1
 1 (−1)δ+ 4 
n+1

=2 2 ∓
2(1 − z)n 2(1 + z)n
n+1 
n−1 
+∞
1 ∓ (−1)k+δ+ 4 n+k−1
=2 2 zk ,
2 k
k=0

from which the result follows. 

Apart from the examples above, no Dirac spectrum can be computed in


such an elementary way. This is in particular the case for closed hyperbolic
manifolds (spaceforms of negative curvature), see also Theorem 2.2.3 below.
However, it remains theoretically possible to do the computations on ho-
mogeneous spaces, where there exists a representation-theoretical method to
express the Dirac operator. This is the object of the next section.

2.2 Spectrum of some other homogeneous spaces

Let us first introduce a few notations and recall basic facts. Denote by
M := G/H an n-dimensional homogeneous space and by g (resp. h) the
Lie algebra of G (resp. of H). In that case the existence as well as the
set of spin structures on M can be read off the isotropy representation of
M , which is defined as the Lie-group-homomorphism α : H → GL(g/h) in-
duced by the restriction of the adjoint map Ad of G onto H. It is indeed
well-known that M carries a homogeneous Riemannian metric (i.e., a metric
invariant under the left G-action) or an orientation if and only if α(H) is
compact or connected respectively. Assuming both conditions α becomes a
map H −→ SO(g/h). The existence of a homogeneous spin structure (i.e., a
spin structure on which the left G-action on SO(T M ) lifts) on M is then
 of α into Spin(g/h), the set of spin structures
equivalent to that of a lift α
standing then in one-to-one correspondence with that of such α ’s, that is,
with the set of group-homomorphisms H −→ {±1}, see [34, Lemma 1]. For
example, if H is connected, then there can exist only one homogeneous spin
structure on M . Moreover, if G is simply-connected, then all spin structures
are obtained in such a way.
To describe the Dirac operator on M , one has to look at the left action of
G onto the space of sections of ΣM , which can be identified with the space of
equivariant maps G −→ Σn . This left action induces a unitary representation
of G onto the space of L2 sections on M , which can be split into irreducible
and finite-dimensional components since G is compact. The Dirac operator
preserves that splitting and can be determined with the help of the following
result based on the Frobenius reciprocity principle (see reference in [34]).
Theorem 2.2.1 (see e.g. [34]) Let M := G/H be an n-dimensional Rie-
mannian homogeneous spin manifold with G compact and simply-connected.
36 2 Explicit computations of spectra

Let p be an Ad(H)-invariant supplementary subspace of h in g and fix a


p.o.n.b {X1 , . . . , Xn } of p. Choose a spin structure on M and let α
 : H −→
Spinn be the corresponding lift of the isotropy representation. Denote by
Σα M −→ M the spinor bundle of M associated with α be the set
. Let G
of equivalence classes of irreducible unitary representations of G (we identify
an element of G with one of its representants).

1. The space L2 (M, Σα M ) splits under the unitary left action of G into a
direct Hilbert sum 
Vγ ⊗ HomH (Vγ , Σn )

γ∈G

where Vγ is the space of the representation γ (i.e., γ : G −→ U(Vγ )) and


HomH (Vγ , Σn ) := f ∈ Hom(Vγ , Σn ) s.t. ∀h ∈ H,



f ◦ γ(h) = (δn ◦ α) (h) ◦ f .

2. The Dirac operator D of M preserves each summand in the splitting above;


more precisely, if {e1 , . . . , en } denotes the canonical basis of Rn , then for
the restriction of D to Vγ ⊗HomH (Vγ , Σn ) is given by Id⊗Dγ ,
every γ ∈ G,
where, for every A ∈ HomH (Vγ , Σn ),


n
Dγ (A) = − ek · A ◦ de γ(Xk )
k=1
⎛ ⎞

n 
+⎝ βi ei + αijk ei · ej · ek ⎠ · A, (2.3)
i=1 i<j<k

and

1
n
βi := [Xj , Xi ]p , Xj

2 j=1
1
αijk := ( [Xi , Xj ]p , Xk
+ [Xj , Xk ]p , Xi
+ [Xk , Xi ]p , Xj
)
4

(here Xp denotes the image of X ∈ g under the projection g −→ p with


kernel h).

Of course, if H is a discrete subgroup of the Lie group G, then Theorem 2.2.1


contains Proposition 1.4.2 (just transform the left-action into a right-one).
Theorem 2.2.1 can for instance be applied to the computation of the Dirac
spectrum of S2m+1 = U  m+1/ with Berger metric, where U  m stands for
Um
the universal covering of the unitary group Um . Recall that, if n is odd,
2.2 Spectrum of some other homogeneous spaces 37

then the round sphere Sn is the total space of an S1 -bundle called the Hopf-
n−1 n−1
fibration Sn −→ CP 2 , where CP 2 is the complex projective space of
n−1
complex dimension 2 . In particular one may decompose the round metric
on Sn along the S1 -fibres and their orthogonal complement; we denote this
decomposition by can = gS1 ⊕ g⊥ .
Theorem 2.2.2 (C. Bär [39]) For n = 2m + 1 ≥ 3, let M n := Sn with
Berger metric gt = tgS1 ⊕ g⊥ for some real t > 0. Then the Dirac operator of
(M n , gt ) has the following eigenvalues:
m + k
i) 1 m+1
t( 2 + k) + tm
2 , k ∈ N, with multiplicity .
k
1  m + k
ii) (−1)m+1 t(
m+1
2 + k) + tm
2 , k ∈ N, with multiplicity .
k

iii) (−1)j 2t ± [ 2t (m − 1 − 2j) + 1t (a1 − a2 + m−1
2 − j)] + 4(m + a1 − j)
2

(1 + a2 + j), a1 , a2 ∈ N, j ∈ {0, 1 . . . , m − 1}, with multiplicity
(m+a1 )!(m+a2 )!(m+1+a1 +a2 )!
a1 !a2 !m!j!(m−1−j)!(m+a1 −j)(1+a2 +j) .

The reader interested in the proof of Theorem 2.2.2 or in the computation


of the Dirac spectrum of 3-dimensional lens spaces S /Zk should refer to [39,
3

Sec. 3] or to [33] respectively.


In general, the formula (2.3) describing the Dirac operator as a direct sum
of endomorphisms in finite-dimensional spaces cannot be better explicited.
In case M is symmetric, (2.3) simplifies in the sense that the square of D is
given by
S
D2 = ΩG + Id, (2.4)
8
where ΩG is the Casimir operator of G and S is the scalar curvature of M ,
see e.g. [88, Prop. p.87]. However, on general homogeneous spaces, obtaining
the spectrum explicitly still remains very difficult, see e.g. [103, Sec. 3].
In case M is the quotient of a non-compact Lie group G, Theorem 2.2.1
does not apply since no Frobenius reciprocity is available. Moreover, the
irreducible representations of G are much harder to classify, so that the knowl-
edge of the spectrum is generally out of reach. We quote the only example
where this was successfully carried out and which is a quotient of the projec-
tive special linear group of R2 , that we denote PSL2 (R). Define a Fuchsian
subgroup of PSL2 (R) to be a discrete one. The signature of a Fuchsian sub-
group Γ is the tuple of non-negative integers (g; m1 , . . . , mr ) such that Γ is
presented by

Γ = < A1 , B1 , . . . Ag , Bg , X1 , . . . , Xr |
X1m1 = . . . = Xrmr = X1 . . . Xr A1 B1 A−1 −1 −1 −1
1 B1 . . . Ag Bg Ag Bg = 1 > .

Theorem 2.2.3 (J. Seade and B. Steer [218]) Let M := PSL2 (R)/Γ,
where Γ is a co-compact Fuchsian subgroup of signature (g; m1 , . . . , mr )
38 2 Explicit computations of spectra

in PSL2 (R). Consider the orientation and the 1-parameter-family


  of left-

1 0 1 0 1 1 0
invariant metrics (gt )t>0 on M for which ( t , , )
−1 0 1 0 0 −1
is a positively-oriented orthonormal basis of T1 M .
Then the Dirac eigenvalues of (M, gt ) endowed with its trivial (left-
invariant) spin structure are:
i) − 2t + 1
t with multiplicity 2.
ii) − 2t − t , k ≥ 1, with multiplicity
2k−1

mj −1
2k − 1 r
1  sin((2k − 1)mj π)
Vol(M ) − lπ
.
2π j=1
mj sin( m )
l=1 j


iii) − 2t ± (2n − 1)2 (1 + t−2 ) − (2k − 1)2 , k ≥ 1, n > k, with the same
multiplicity
 as in ii).
iv) − 2t ± (2n − 1)2 (1 + t−2 ) − s2 , n ∈ Z, s ∈ Λ, where Λ is some countable
subset of ] − 1, 1[∪iR.
The set Λ depends on representation-theoretical data and cannot be ex-
plicited in general. For t = 1, the metric gt has constant negative sectional
curvature. This is up to the knowledge of the author the only compact hy-
perbolic manifold whose Dirac spectrum has been computed.
To summarise, we reproduce - with small changes - the list in [42] of all
homogeneous spaces of which Dirac spectrum has already been computed:
space description references
G simply-connected compact Lie groups [83]
Rn/Zn flat tori [86]
R3/Γ 3-dim. (flat) Bieberbach manifolds [206]
Rn/Γ some n-dim. (flat) Bieberbach manifolds [188]
Sn round spheres [226]
Sn/Γ spherical spaceforms [38]
S2m+1 spheres with Berger metric [148, 39]
S3/Z 3-dim. lens spaces with Berger metric [34]
k
S3/Q S through the group of quaternions,
3
[103]
8
with Berger metric
H 3/Γ 3-dim. Heisenberg manifolds [15]
PSL2 (R)/Γ (Γ Fuchsian) [218]
CP2m+1 complex projective spaces [68, 69, 222]
HPm quaternionic projective spaces [66, 189]
OP2 Cayley projective plane [234]
Gr2 (R2m ) real 2-Grassmannians [224, 225]
Gr2p (R2m ) real 2p-Grassmannians [219]
Gr2 (Cm+2 ) complex 2-Grassmannians [190]
G2/SO - [219, 220]
4
2.3 Small eigenvalues of some symmetric spaces 39

2.3 Small eigenvalues of some symmetric spaces

In case of a symmetric space M := G/H where G and H have the same


rank, a formula due to R. Parthasarathy (see reference in [191]) allows to
express certain parts of the Casimir operator ΩG (see (2.4)) in terms of
representation-theoretical data:

Theorem 2.3.1 (J.-L. Milhorat [191, 192]) Let M := G/H be a spin


compact simply-connected irreducible symmetric space with G compact and
simply-connected, endowed with the metric · , ·
induced by the Killing form
of G sign-changed. Assume that G and H have the same rank and fix a spin
structure on G/H . Let βk , k = 1, . . . , p, be the H-dominant weights occurring
in the decomposition into irreducible components of the spin representation
under H.
Then the square of the first eigenvalue of D is
n n
+ 2 min βk 2 = + 2 min w · δG − δH 2 ,
8 1≤k≤p 8 w∈WG

where  ·  is the norm associated to · , ·


, WG is the Weyl group of G and
δG (resp. δH ) is the half-sum of the positive roots of G (resp. H).

Theorem 2.3.1 has been applied by J.-L. Milhorat in [192] to compute the
smallest eigenvalue λ1 (D2 ) for the following symmetric spaces (where S de-
notes the scalar curvature of M and Ep the exceptional simple Lie group of
rank p):

M = G/H dim(M ) λ1 (D2 )

m2 +6m−4 m2 +6m−4
Spinm+4/(Spin · Spin )
m 4
4m m(m+2) · m
2 = m(m+2) · S
4
(m even)

E6/(SU · SU )
6 2
40 41
6 = 41
30 · S
4

E7/(Spin
12 · SU2 ) 64 95
9 = 95
72 · S
4

E8/(E · SU )
7 2
112 269
15 = 269
210 · S
4

The quotient S4 has been each time factorized out in order to compare the
n
dimension-depending coefficient standing before with n−1 , which is the cor-
responding one in Friedrich’s inequality (3.1).
Chapter 3
Lower eigenvalue estimates
on closed manifolds

In this chapter we assume that M has empty boundary.

3.1 Friedrich’s inequality

The most general sharp lower bound for the Dirac spectrum has been proved
by T. Friedrich in [85] and is now known under the name “Friedrich’s in-
equality”. For the concept of Killing spinor we refer to Section A.1.

Theorem 3.1.1 (T. Friedrich [85]) Any eigenvalue λ of D on an n(≥ 2)-


dimensional closed Riemannian spin manifold (M n , g) satisfies
n
λ2 ≥ inf (S), (3.1)
4(n − 1) M

where S is the scalar curvature of M .


Moreover (3.1) is an equality for some eigenvalue λ if and only if there
exists a non-zero real Killing spinor on (M n , g).
Proof : It follows from the Schrödinger-Lichnerowicz formula (1.15) that, for
any ϕ ∈ Γ(ΣM ),
  
S 2
D2 ϕ, ϕ
vg = ∇∗ ∇ϕ, ϕ
vg + |ϕ| vg .
M M M 4

By definition of ∇∗ ∇ and since D is formally self-adjoint we can write


  
S 2
|Dϕ| vg =
2
|∇ϕ| vg +
2
|ϕ| vg . (3.2)
M M M 4

Decompose now w.r.t. any local orthonormal basis {ej }1≤j≤n of T M

N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 41


DOI 10.1007/978-3-642-01570-0 3, 
c Springer-Verlag Berlin Heidelberg 2009
42 3 Lower eigenvalue estimates on closed manifolds

1 ∗ 1 ∗
n n
∇ϕ = ∇ϕ + ej ⊗ ej · Dϕ − e ⊗ ej · Dϕ,
n j=1 n j=1 j
     
=P ϕ ∈ Ker(μ) ∈ Ker(μ)⊥

where P is the so-called Penrose operator of (M n , g), see also Appendix A.


We deduce that |∇ϕ|2 = |P ϕ|2 + n1 |Dϕ|2 (identity (A.11) in Appendix A).
Replacing |∇ϕ|2 in (3.2) one obtains
   
1 S 2
|Dϕ| vg =
2
|Dϕ| vg +
2
|P ϕ| vg +
2
|ϕ| vg ,
M n M M M 4

that is,
  
n n
|Dϕ| −
2 2
S|ϕ| vg = |P ϕ|2 vg . (3.3)
M 4(n − 1) n−1 M
Choose ϕ to be a non-zero eigenvector for D associated to the eigenvalue λ.
From |P ϕ|2 ≥ 0 one straightforward obtains the inequality (3.1).
If (3.1) is an equality for some eigenvalue λ then (3.3) implies P ϕ = 0
for any non-zero eigenvector ϕ for D associated to λ, hence any such ϕ is
a (necessarily real) Killing spinor on (M n , g). Conversely, if (M n , g) carries
a non-zero α-Killing spinor ϕ, then since M is compact α must be real.
Moreover, on the one hand ϕ is an eigenvector for D associated to the
eigenvalue −nα, on the other hand we know from Proposition A.4.1 that the
scalar curvature of (M n , g) must be S = 4n(n − 1)α2 , in particular it must
be non-negative. Therefore
 such 
a ϕ must be an eigenvector for D associated
to the eigenvalue 4(n−1) or − 4(n−1)
nS nS
. This shows the equivalence in the
limiting-case and concludes the proof. 

Another method for the proof of (3.1), which is actually T. Friedrich’s in [85],
relies on the modified connection

 X ψ := ∇X ψ + λ X · ψ

n
for every X ∈ T M , where Dψ = λψ: Compute |∇ψ|  2 (which plays the role
of |P ψ| above), integrate and apply the Schrödinger-Lichnerowicz formula.
2

Alternatively but still along the same idea, √ it can be directly deduced from
the Cauchy-Schwarz inequality that |Dϕ| ≤ n|∇ϕ| for every section ϕ, from
which (3.1) follows.
As a consequence of Theorem 3.1.1, if the scalar curvature S of (M n , g) is
positive then its Dirac operator has trivial kernel - whatever the spin structure
is. This had been already noticed by A. Lichnerowicz in [175] where he had
obtained as a straightforward application of (3.2) the following estimate:
1
λ2 ≥ inf (S). (3.4)
4 M
3.2 Improving Friedrich’s inequality in presence of a parallel form 43

It follows from (3.4) combined with the Atiyah-Singer index theorem [28]
(see Theorem 1.3.9) that a Riemannian manifold with positive scalar cur-
vature must have vanishing topological index. In particular, if the manifold

has non-vanishing A-genus, then it cannot carry any Riemannian manifold
with positive scalar curvature. In other words, there exists a topological ob-
struction to the existence of metrics with positive scalar curvature on closed
spin manifolds, at least in even dimensions. The reader interested in further
results in that topic - such as Gromov-Lawson’s work - should refer to [173]
or to [124]. The existence of Riemannian metrics for which the Dirac kernel
is non-zero is discussed in Section 6.2. Moreover, we mention another closely
related application of the Atiyah-Singer index theorem to geometry via the
Schrödinger-Lichnerowicz formula, namely to the so-called scalar curvature
rigidity issue which asks for the possibility of increasing the scalar curvature
without shrinking the distances of a given metric on a fixed background man-
ifold. For example this is not possible on the round sphere (M. Llarull [180,
Thm. B]) nor on any connected closed Kähler manifold with non-negative
Ricci curvature (S. Goette and U. Semmelmann [109, Thm. 0.1]), we refer to
[110] for the case of symmetric spaces and references.
Although it requires the non-negativity of S to be non-trivial, Friedrich’s
inequality (3.1) provides fine information of geometrical nature on the Dirac
spectrum. Indeed S stands for a very weak curvature invariant of a given
Riemannian manifold. This shows for example a difference of behaviour with
other differential operators such as the scalar Laplacian Δ: by a result of A.
Lichnerowicz [174], any non-zero eigenvalue λ of Δ satisfies
n
λ≥ inf (Ric),
n−1 M

where Ric is the Ricci curvature tensor of (M n , g), which is a stronger cur-
vature invariant. In case inf M (S) ≤ 0 Friedrich’s inequality (3.1) can be
improved in different ways using various techniques, see Sections 3.3 to 3.7.
Besides, (3.1) is sharp since e.g. M := Sn (n ≥ 2) admits non-zero Killing
spinors, see Example A.1.3.2. For the classification of Riemannian spin man-
ifolds carrying non-zero real Killing spinors we refer to Theorems A.4.2 and
A.4.3 in Appendix A.

3.2 Improving Friedrich’s inequality in presence


of a parallel form

O. Hijazi [128] and A. Lichnerowicz [176, 177] noticed that equality in (3.1)
cannot hold on those M admitting a non-zero parallel k-form for some k ∈
{1, . . . , n − 1}. This suggests (3.1) could be enhanced under this assumption.
44 3 Lower eigenvalue estimates on closed manifolds

The idea of proof for Theorems 3.2.1, 3.2.4 and 3.2.6 can be summarised as
follows (see [64]): given any eigenvector ϕ of D to the eigenvalue λ, decompose
|∇ϕ|2 in a sharper way than for the proof of Friedrich’s inequality, using the
splitting of ΣM induced by the Clifford action of the parallel form.

Theorem 3.2.1 (B. Alexandrov, G. Grantcharov and S. Ivanov [5])

Any eigenvalue λ of D on an n(≥ 3)-dimensional closed Riemannian spin


manifold (M n , g) admitting a non-zero parallel 1-form satisfies

n−1
λ2 ≥ inf (S), (3.5)
4(n − 2) M

where S is the scalar curvature of M .


Moreover if (3.5) is an equality for some eigenvalue λ, then the universal
cover of M is a Riemannian product of the form R × N , where N admits a
real Killing spinor.

We shall prove a more general result:


Theorem 3.2.2 (A. Moroianu and L. Ornea [202]) Inequality (3.5)
holds as soon as M n (n ≥ 3) admits a non-zero harmonic 1-form of constant
length. Furthermore, if it is an equality for some eigenvalue λ, then this
form is parallel.

Proof : Let ξ be the dual vector field to the harmonic 1-form of constant
length. We may assume that g(ξ, ξ) = 1 on M . Define the following Penrose-
like operator

1 1
TX ϕ := ∇X ϕ + (X − g(X, ξ)ξ) · Dϕ − (ng(X, ξ) + X · ξ·)∇ξ ϕ
n−1 n−1

for all X ∈ T M and ϕ ∈ Γ(ΣM ). In case ξ is parallel this operator can be


described as the sum of the orthogonal projections of ∇ϕ onto the kernels of
μ±
the Clifford multiplications T ∗ M ⊗ Σ± M −→ ΣM , where Σ± M := Ker(iξ ·
∓IdΣM ), see [5, eq. (4)] for another equivalent expression (note however that
it does not exactly coincide with the T defined in [202]). Nevertheless ξ need
not be parallel in order for T to play its role for the estimate as we shall see
in the proof.
Fix a local orthonormal frame {ej }1≤j≤n of T M . For any ϕ ∈ Γ(ΣM ), we
have


n
|T ϕ|2 = |Tej ϕ|2
j=1
1 n
= |∇ϕ|2 + |Dϕ|2 + |∇ξ ϕ|2
n−1 n−1
3.2 Improving Friedrich’s inequality in presence of a parallel form 45

2
− (|Dϕ|2 + e ( ξ · Dϕ, ∇ξ ϕ
))
n−1
2
− (n|∇ξ ϕ|2 − e ( ξ · ∇ξ ϕ, Dϕ
))
n−1
2
− e ( ξ · ∇ξ ϕ, Dϕ
))
n−1
1 n 2
= |∇ϕ|2 − |Dϕ|2 − |∇ξ ϕ|2 + e ( ξ · ∇ξ ϕ, Dϕ
)).
n−1 n−1 n−1

Now we can express the last term on the r.h.s. through the other ones, a trick
due to the authors of [202]: namely, since ξ is assumed to be harmonic, i.e.,
closed and co-closed, the identity (1.12) reads D(ξ · ϕ) = −ξ·Dϕ − 2∇ξ ϕ and
hence
1
e ( ξ · ∇ξ ϕ, Dϕ
) = |∇ξ ϕ|2 + (|Dϕ|2 − |D(ξ · ϕ)|2 ),
4
from which we obtain
1 n−2 1
|T ϕ|2 = |∇ϕ|2 − |Dϕ|2 − |∇ξ ϕ|2 + (|Dϕ|2 − |D(ξ · ϕ)|2 ).
n−1 n−1 2(n − 1)

Integrating this identity over M and applying the Schrödinger-Lichnerowicz


formula (1.15) we have
   
n−2 1 n−2
|T ϕ|2 vg = |Dϕ|2 vg − S|ϕ|2 vg − |∇ξ ϕ|2 vg
n−1 4 n − 1
M M
 M M
1
+ |Dϕ| − |D(ξ · ϕ)| vg .
2 2
2(n − 1) M

But choosing ϕ to be eigen for D for the smallest (in absolute value) eigen-
value λ, the min-max principle (see Lemma 5.0.2) implies
 
|D(ξ · ϕ)| vg ≥ λ
2 2
|ξ · ϕ|2 vg
M
 M

= λ2 |ϕ|2 vg
 M
= |Dϕ|2 vg ,
M

hence
  
n−2 2 1 n−2
( λ − inf (S)) |ϕ|2 vg ≥ |T ϕ|2 vg + |∇ξ ϕ|2 vg (3.6)
n−1 4 M M M n−1 M

and the inequality (3.5) follows.


46 3 Lower eigenvalue estimates on closed manifolds

If (3.5) is an equality for some eigenvalue λ, then (3.6) implies T ϕ = 0


and ∇ξ ϕ = 0, that is,

λ
∇X ϕ = − (X − g(X, ξ)ξ) · ϕ (3.7)
n−1

for any eigenvector ϕ associated to λ and any X ∈ T M . In particular its


length must be constant on M . As in [202] we next show that ξ must be
parallel. For this purpose we compute the curvature tensor on such a ϕ: let
X, Y ∈ T M , then

RX,Y ϕ = ∇[X,Y ] ϕ − [∇X , ∇Y ]ϕ


λ
= ((g(X, ∇Y ξ) − g(Y, ∇X ξ))ξ + g(X, ξ)∇Y ξ − g(Y, ξ)∇X ξ) · ϕ
n−1
λ2 
+ (X − g(X, ξ)ξ) · (Y − g(Y, ξ)ξ)
(n − 1)2

− (Y − g(Y, ξ)ξ) · (X − g(X, ξ)ξ) · ϕ,

hence using (1.9) and the fact that g(ξ, ξ) = 1 on M we obtain

1  n

Ric(ξ) · ϕ = ej · Rξ,e ϕ
2 j=1
j

λ 
n
= (g(ξ, ∇ej ξ) − g(ej , ∇ξ ξ))ej · ξ
n − 1 j=1

+ (g(ξ, ξ)∇ej ξ − g(ej , ξ)∇ξ ξ) · ej · ϕ

λ n
= (−2∇ξ ξ · ξ + ∇ej ξ · ej ) · ϕ.
n−1 j=1

The last sum vanishes because of (1.2) together with ξ being closed and co-
closed. On the other hand dξ = 0 means that g(∇X ξ, Y ) − g(∇Y ξ, X) = 0 for
all X, Y ∈ T M , hence for X = ξ one obtains - using once again g(ξ, ξ) = 1
- that g(∇ξ ξ, Y ) = 0 for all Y ∈ T M , i.e., ∇ξ ξ = 0. This shows Ric(ξ) = 0.
From Bochner’s formula for the Laplace operator on 1-forms (see e.g. [173,
Cor. 8.3 p.156]) one deduces that ∇ξ = 0, i.e., that ξ is parallel.
We now prove the limiting-case in Theorem 3.2.1. If ξ is parallel then the
universal cover of M must be a Riemannian product of the form R × N .
W.r.t. the pull-back spin structure the lift of ϕ to R × N also satisfies (3.7)
provided ξ is replaced by ∂t ∂
. Since each {t} × N sits totally geodesically in
3.2 Improving Friedrich’s inequality in presence of a parallel form 47

R × N the Gauss-type formula (1.21) implies that the induced spinor field
on N is a real Killing spinor for one of the constants ± n−1
λ
. 

Beware that the necessary condition for (3.5) to be an equality is not


sufficient, since e.g. R3 = R × R2 with flat metric carries non-zero paral-
lel spinors whereas (flat) T3 = Z3 \R only admits such spinors in case it
3

carries the trivial spin structure (i.e., the spin structure induced by the triv-
ial lift of the Z3 -action to the spin level, see Proposition 1.4.2). In fact (3.5)
is an equality if and only if there exists a π1 (M )-equivariant solution - in the
sense of (1.24) - to (3.7) on the universal cover R × N of M .
Although the (real) Killing-spinor-equation is completely understood (see
Theorems A.4.2 and A.4.3), the list of all local Riemannian products on
which (3.5) is sharp is not entirely known. B. Alexandrov, G. Grantcharov
and S. Ivanov have shown in [5] that, under this assumption and if n = 7 is
odd, then M is diffeomorphic - but not necessarily isometric - to S1 × Sn−1 .
It is moreover important to note that the hypothesis in Theorem 3.2.2 on
the length being constant cannot be removed: C. Bär and M. Dahl showed in
[46] that in dimension n ≥ 3 Friedrich’s inequality (3.1) cannot be improved
with the help of topological assumptions. Namely there exists on any given
compact spin manifold M n admitting a metric with positive scalar curvature
a smooth family of Riemannian metrics (gt )t>0 with Sgt ≥ n(n − 1) and

n2 n2
≤ λ1 (Dg2t ) ≤ + t,
4 4
where Dgt stands for the Dirac operator to the metric gt on M . In other words,
one can get as close as one wants to the equality in Friedrich’s inequality
(3.1) on any such manifold. Note that the set of compact spin manifolds
with positive first Betti number and admitting a metric with positive scalar
curvature is non-empty since it contains e.g. S1 × Sn−1 , n ≥ 3.
The generalization of Theorem 3.2.1 to locally reducible Riemannian man-
ifolds was achieved by B. Alexandrov, extending earlier work by E.C. Kim
[154]:

Theorem 3.2.3 (B. Alexandrov [4]) Let (M n , g) be an n(≥ 2)-dimen-


sional closed Riemannian spin manifold with positive scalar curvature S.
Assume that T M splits orthogonally into


k
TM = Tj ,
j=1

where Tj is a parallel distribution of dimension nj and n1 ≤ . . . ≤ nk . Then


any eigenvalue λ of D satisfies
nk
λ2 ≥ inf (S). (3.8)
4(nk − 1) M
48 3 Lower eigenvalue estimates on closed manifolds

Moreover, if (3.8) is an equality for some eigenvalue λ, then the universal


cover of M is isometric to M1 ×. . .×Mk , where Mj is a closed nj -dimensional
Riemannian spin manifold admitting a non-zero real non-parallel Killing
spinor for j = k, a non-zero parallel spinor if nj < nk and a non-zero real
Killing spinor if nj = nk .
Note that (3.8) contains both (3.1) and (3.5) and that nk > 1 because
of the assumption S > 0. Moreover, for any integers 1 ≤ n1 ≤ . . . ≤ np <
np+1 = . . . = nk , all Riemannian products of the form M1 × . . . × Mk , where
Mj is an nj -dimensional closed Riemannian spin manifold admitting a non-
zero parallel spinor for j ≤ p and a non-zero real Killing spinor for j ≥ p + 1
which is furthermore non-parallel for j = k, satisfy the equality in (3.8) w.r.t.
the product spin structure.
Sketch of proof of Theorem 3.2.3: The proof follows the lines of that of
Theorem 3.2.1. Define the Penrose-like operator T : for any ϕ ∈ Γ(ΣM ) and
any X ∈ T M ,
k
1
TX ϕ := ∇X ϕ + πj (X) · D[j] ϕ,
n
j=1 j
nj
where πj : T M → Tj is the orthogonal projection, D[j] ϕ := l=1 el,j · ∇el,j ϕ
and (e1,j , . . . , enj ,j ) denotes a local orthonormal frame of Tj , for every j ∈
{1, . . . , k}. A short computation gives

k
1
|T ϕ|2 = |∇ϕ|2 − |D[j] ϕ|2 .
j=1
nj

k
On the other hand, it is an exercise to show that D2 = j=1 D[j] 2
and that
D[j] is formally self-adjoint, so that, after integration and application of the
Schrödinger-Lichnerowicz formula (1.15), one obtains

nk k
nk − nj nk
Dϕ2 = T ϕ2 + D[j] ϕ2 + (Sϕ, ϕ). (3.9)
nk − 1 j=1
nj (nk − 1) 4(nk − 1)

Choosing ϕ to be an eigenvector for D associated to the eigenvalue λ leads to


the inequality. If this inequality is an equality for some λ, then (3.9) implies
that, for any non-zero eigenvector ϕ for D associated to the eigenvalue
λ, one has T ϕ = 0, D[j] ϕ = 0 as soon as nj < nk and S is constant on
{x ∈ M | ϕ(x) = 0}. In case nj < nk one deduces that ∇πj (X) ϕ = 0 for
every X. It remains to prove that, on the universal cover of M , which is a
Riemannian product of the form M1 × . . . × Mk by assumption, the lift of
ϕ induces a real Killing spinor on each Mj , which is parallel if nj < nk and
non-parallel for j = k. We refer to [4, Sec. 2] for the details. 
3.2 Improving Friedrich’s inequality in presence of a parallel form 49

For 2-forms, the canonical class of manifolds to be considered consists of that


of Kähler manifolds, i.e., of triples (M n , g, J) where (M n , g) is a Riemannian
manifold and J a parallel almost Hermitian structure on T M . Recall that
J ∈ Γ(End(T M )) is called almost Hermitian if and only if J 2 = −IdT M and
g(J(X), J(Y )) = g(X, Y ) for all X, Y ∈ T M . In this case n is even and the
Kähler-form Ω is parallel, where Ω is defined by

Ω(X, Y ) := g(J(X), Y )

for all X, Y ∈ T M . K.-D. Kirchberg was the first to enhance Friedrich’s


inequality (3.1) on Kähler manifolds:

Theorem 3.2.4 (K.-D. Kirchberg [156]) Any eigenvalue λ of D on an


n(≥ 4)-dimensional closed Kähler spin manifold (M n , g, J) satisfies
 n+2
 n
 4n inf M (S) if 2 is odd
λ2 ≥  , (3.10)
 n inf M (S) if n
is even
4(n−2) 2

where S is the scalar curvature of M . Moreover, in the case where S > 0,


(3.10) is an equality for some eigenvalue λ if and only if there exists non-zero
sections ψ, φ of ΣM satisfying

 ∇X ψ = − λ (X + iJ(X)) · φ
 n+2 (3.11)
 ∇X φ = − λ (X − iJ(X)) · ψ
n+2

for all X ∈ T M if n
2 is odd and a non-zero section ψ of ΣM satisfying
 2
D ψ = λ2 ψ

 ∇X ψ = − n1 (X + iJ(X)) · Dψ
 (3.12)
Ω · ψ = −2iψ

 Ω · Dψ =0

for all X ∈ T M if n
2 is even.
Proof : We follow the proof given in [131, 133], see also [223, Sec. 3] or [150].
We may assume that S > 0 on M (otherwise the estimate is trivial). Set, for
every X ∈ T M , p± (X) := 12 (X ∓ iJ(X)) ∈ T M ⊗ C. In the whole proof we
shall redenote m := n2 . Given a pointwise orthonormal basis (e1 , . . . , en ) of
T M such that ej+m = J(ej ) for every 1 ≤ j ≤ m, define zj := p+ (ej ) and
z j := p− (ej ) for all 1 ≤ j ≤ m. Then (z1 , . . . , zm ) and (z 1 , . . . , z m ) are bases
of T 1,0 M := p+ (T M ) and T 0,1 M := p− (T M ) respectively satisfying

zj · zk = −zk · zj , z j · z k = −z k · z j , zj · z k + z k · zj = −δjk
50 3 Lower eigenvalue estimates on closed manifolds

for all 1 ≤ j, k ≤ m. With those notations, it is elementary to show that the


ranked-2m -vector bundle (⊕m r 1,0
r=0 Λ T M ) · z 1 · . . . · z m becomes a non-trivial
Clifford submodule of the Clifford algebra bundle, which is independent of
the basis originally chosen. Therefore it can be identified with the spinor
bundle ΣM itself. Moreover, setting Σr M := (Λr T 1,0 M ) · z 1 · . . . · z m , it is an
exercise to prove that, w.r.t. the Clifford action of the Kähler-form Ω,

Σr M = Ker(Ω · −i(2r − m)Id)

and that the Clifford action by Ω is skew-Hermitian and parallel. As a con-


sequence, one obtains the orthogonal and parallel decomposition


m
ΣM = Σr M. (3.13)
r=0

By construction p± (X) · Σr M ⊂ Σr±1 M , for every X ∈ T M , where we set


Σr M := 0 as soon as r ∈ / {0, 1, . . . , m}. Inparticular the Dirac operator D
n
does not preserve (3.13): setting D± := j=1 p± (ej ) · ∇ej , we have D =
D+ + D− with D± : Γ(Σr M ) → Γ(Σr±1 M ) for every r ∈ {0, 1 . . . , m}.
Nevertheless, a more precise study of D± shows that D+ ◦D+ = D− ◦D− = 0,
so that (3.13) is preserved by D2 = D+ ◦ D− + D− ◦ D+ . Beware that the
operators D± have nothing to do with the D± of Proposition 1.3.2.
For any r ∈ {0, 1 . . . , m} and ϕ ∈ Γ(Σr M ) define

(r) 1 1
TX ϕ := ∇X ϕ + p− (X) · D+ ϕ + p+ (X) · D− ϕ
2(r + 1) 2(m − r + 1)

for all X ∈ T M . In other words, T (r) ϕ is the orthogonal projection of ∇ϕ


onto the kernel of the Clifford multiplication μ : T ∗ M ⊗ Σr M −→ Σr−1 M ⊕
Σr+1 M . Elementary computations show that


n 
n
p+ (ej )·p− (ej )· = iΩ·−mId and p− (ej )·p+ (ej )· = −iΩ·−mId,
j=1 j=1

n n
in particular j=1 p+ (ej ) · p− (ej ) · ϕ = −2rϕ and j=1 p− (ej ) · p+ (ej ) · ϕ =
−2(m − r)ϕ. We deduce for the norms that


n
|T (r) ϕ|2 = |Te(r)
j
ϕ|2
j=1

n
= |∇ej ϕ|2
j=1
3.2 Improving Friedrich’s inequality in presence of a parallel form 51


n
1 1
+ |p− (ej ) · D+ ϕ|2 + |p (e ) · D− ϕ|2
2 + j
j=1
4(r + 1)2 4(m − r + 1)
n 
1
+2 e ∇ej ϕ, p− (ej ) · D+ ϕ

j=1
2(r + 1)
n 
1
+2 e ∇ej ϕ, p+ (ej ) · D− ϕ

j=1
2(m − r + 1)
n 
1
+2 e p− (ej ) · D+ ϕ, p+ (ej ) · D− ϕ

j=1
4(r + 1)(m − r + 1)
= |∇ϕ|2
1 n
− p+ (ej ) · p− (ej ) · D+ ϕ, D+ ϕ

4(r + 1)2 j=1

1 n
− p− (ej ) · p+ (ej ) · D− ϕ, D− ϕ

4(m − r + 1)2 j=1


1 1
− |D+ ϕ|2 − |D− ϕ|2
r+1 m−r+1
= |∇ϕ|2
1 1
+ |D+ ϕ|2 + |D− ϕ|2
2(r + 1) 2(m − r + 1)
1 1
− |D+ ϕ|2 − |D− ϕ|2 ,
r+1 m−r+1

that is,

1 1
|T (r) ϕ|2 = |∇ϕ|2 − |D+ ϕ|2 − |D− ϕ|2 . (3.14)
2(r + 1) 2(m − r + 1)

Let r ∈ {0, 1 . . . , m} be the smallest integer for which Ker(D2 − λ2 Id) ∩


Γ(Σr M ) = 0. Let ψ ∈ Γ(Σr M ) be a non-zero eigenvector for D2 associated to
the eigenvalue λ2 . Since [D2 , D± ] = 0, both D+ ψr and D− ψr lie in Ker(D2 −
λ2 Id), in particular D− ψr = 0 by the choice of r. Independently, there exists
on ΣM a parallel field j of complex antilinear automorphisms commuting
with the Clifford multiplications by vectors (see e.g. [101, Lemma 1]), in
particular [Ω·, j] = [D, j] = 0, so that j(Ker(D2 − λ2 Id)) ⊂ Ker(D2 − λ2 Id)
and j(Σl M ) = Σm−l M for every l ∈ {0, . . . , m}. Thus the existence of j
imposes r ≤ m − r, hence r ≤ m−1 2 for m odd. If m is even, then r = m 2
cannot happen since otherwise D+ ψ = D− ψ = Dψ = 0 would hold, which
would contradict (3.1) together with S > 0, therefore r ≤ m−2 2 for m even.
We are now ready to prove the estimate. Integrating (3.14) and using
Schrödinger-Lichnerowicz’ formula (1.15), we obtain
52 3 Lower eigenvalue estimates on closed manifolds

S 1
T (r) ψ2 = (D2 ψ, ψ) − ( ψ, ψ) − Dψ2
4 2(r + 1)
2r + 1 2 1
=( λ − inf (S))ψ2 ,
2(r + 1) 4 M

2(r+1)
from which one deduces that λ2 ≥ 4(2r+1) inf M (S). The r.h.s. of that inequal-
ity decreases with r, so that it is bounded from below by the corresponding
expression for r = m−12 in case m is odd and for r = m−2 2 in case m is even.
Inequality (3.10) follows.
Assume now (3.10) to be an equality for some eigenvalue λ. If inf M (S) = 0
then (M n , g) has a non-zero parallel spinor, as already proved in Theorem
3.1.1. If inf M (S) > 0, then for any eigenvector ψ for D associated to the
eigenvalue λ, one has on the one hand ψ = ψ m−1 + ψ m+1 if m is odd and
2 2
ψ = ψ m−2 + ψ m2 + ψ m+2 if m is even, on the other hand T (r) ψr = 0 for
2 2
r = m±1 2 and r = m±22 for m odd and even respectively. Redenoting ψ m−1
2
by ψ and ψ m+1 by φ, we obtain (3.11) in case m is odd. If m is even then
2
redenoting ψ m−2 by ψ one obtains (3.12). Conversely, mimiking the proof
2
of Proposition A.4.1, it is elementary to show that, if (3.11) is satisfied by
some non-zero (ψ, φ), then ψ + φ is an eigenvector for D associated to the
2
eigenvalue λ and S = 4nλ n
n+2 , therefore (3.10) is an equality. Similarly, if 2 is
even and (3.12) is satisfied by some non-zero ψ, then the scalar curvature of
2
(M n , g) is equal to 4(n−2)λ
n , therefore (3.10) is an equality. 

A pair of spinors (ψ, φ) satisfying (3.11) for some non-zero real number λ is
called a real Kählerian Killing spinor. As for Killing spinors (see Proposition
A.4.1), it is not too difficult to show that, if a non-zero real Kählerian Killing
spinor exist on a given complete Kähler spin manifold and associated to
some (non-zero) real λ, then this manifold has odd complex dimension and is
Einstein with positive scalar curvature (in particular it is closed). However,
the precise classification of those Kähler spin manifolds carrying non-trivial
real Kählerian Killing spinors is more technical, even if it turns out to provide
simpler results. The idea to achieve it, due to A. Moroianu [197], can be
summarised as follows: Show the existence of a suitable S1 -bundle over such
a manifold where the pull-back of the real Kählerian Killing spinor induces a
non-zero real Killing spinor; then show that, among the possible holonomies
listed in C. Bär’s classification (Theorem A.4.3), only those associated to a
so-called regular 3-Sasaki structure can occur on that S1 -bundle. We refer to
[197] for details and mention that, before [197] was published, partial results
had been obtained by K.-D. Kirchberg [156, 157, 158] and O. Hijazi [131],
see references in [197].
The even-complex-dimensional case turns out to be more involved since
the underlying manifold is no more Einstein. In dimension n = 4, argu-
ments from complex geometry and based on Kirchberg’s work [159, Thm. 15]
3.2 Improving Friedrich’s inequality in presence of a parallel form 53

allowed T. Friedrich [87, Thm. 2] to prove that, if (M 4 , g, J) carries a non-


zero spinor ψ satisfying (3.12), then up to rescaling the metric (M 4 , g, J)
must be holomorphically isometric either to S2 × S2 or to S2 × T2 , both
with product metric and spin structure, where T2 carries a flat metric and
the trivial spin structure. In higher dimensions, if a non-zero spinor ψ ex-
ists satisfying (3.12), then A. Moroianu showed [200] that the Ricci tensor
of (M n , g) is parallel and has exactly two eigenvalues. This implies that the
universal cover of M is holomorphically isometric to the Riemannian product
N × R2 , where N is a closed Kähler spin manifold admitting a non-zero real
Kählerian Killing spinor, see [200] for details. This result had been formu-
lated by A. Lichnerowicz [179] where there remained however gaps in the
proof.
We formulate the precise statements on the characterization of the
limiting-case of (3.10).
Theorem 3.2.5 Let (M n , g, J) be a closed n(≥ 4)-dimensional Kähler spin
manifold with positive scalar curvature.
1. If n2 is odd, then (3.10) is an equality for some eigenvalue λ of D if and
n
only if (M n , g, J) is holomorphically isometric to CP 2 in case n2 ≡ 1 (4) or
to the twistor-space of a quaternionic Kähler manifold with positive scalar
curvature in case n2 ≡ 3 (4) (K.-D. Kirchberg [157] for n = 6, A. Moroianu
[197] for n ≥ 6).
2. If n2 is even, then (3.10) is an equality for some eigenvalue λ of D if and
only if (M n , g, J) is isometric to Γ\N × R , where N is a simply-connected
2

closed Kähler manifold admitting a non-zero real Kählerian Killing spinor


(ψ, φ) and Γ is generated by (γj , τj ), j = 1, 2, where the τj ’s are trans-
lations of R2 and the γj ’s are commuting holomorphic isometries of N
preserving its spin structure and (ψ, φ) (T. Friedrich [87] for n = 4, A. Mo-
roianu [200] for n ≥ 8).

Beware that, in case n2 is even, the Kähler manifold must not be holomorphi-
cally isometric to the quotient Γ\N × R endowed with the Kähler structure
2

induced from that of N . A simple criterion for holomorphicity is given in


[200, Lemma 7.6].
The last class of manifolds with a non-trivial parallel form having been
handled is that of quaternionic-Kähler manifolds, which carry a canonical
parallel 4-form. The following theorem was proved by O. Hijazi and J.-L.
Milhorat [135, 136, 137] for n = 8, 12 and by W. Kramer, U. Semmelmann
and G. Weingart in the general case [163, 164]:
Theorem 3.2.6 Any eigenvalue λ of D on an n(≥ 8)-dimensional closed
quaternionic-Kähler spin manifold (M n , g) with positive scalar curvature
satisfies
54 3 Lower eigenvalue estimates on closed manifolds

n + 12
λ2 ≥ S, (3.15)
4(n + 8)
where S is the scalar curvature of M . Moreover, this inequality is an equality
for some eigenvalue λ if and only if (M n , g) is isometric to the quaternionic
n
projective space HP 4 .
Here it should be noticed that every quaternionic Kähler manifold of
even quaternionic dimension is spin whereas only the quaternionic projective
space is spin if n4 is odd; moreover, every quaternionic Kähler manifold is
Einstein, hence has constant scalar curvature, see references in [164].

Sketch of proof of Theorem 3.2.6: We follow the proof detailed in [164], which
relies on the representation theory of Sp1 × Spk . Denote n4 by m. A quater-
nionic structure on (M n , g) is given by a triple (I, J, K) of parallel orthogonal
endomorphisms of T M with I 2 = J 2 = K 2 = −IdT M and IJ = −JI = K.
Each of those endormorphisms is a Kähler structure on T M with associated
Kähler form, so that one may define the so-called fundamental form

Ω := ΩI ∧ ΩI + ΩJ ∧ ΩJ + ΩK ∧ ΩK

on T M . The 4-form Ω is parallel and can be shown to act on ΣM so as to


split it into
m
ΣM = Σr M,
r=0

with Σr M := Ker(Ω · −(6m − 4r(r + 2))Id) ⊂ ΣM [136]. As in the Kähler


case, the Clifford multiplication sends T ∗ M ⊗Σr M into Σr−1 M ⊕Σr+1 M . De-
composing Ker(μ)|T ∗ M ⊗Σr M into irreducible components under Sp1 × Spm−1 ,
one obtains four twistor operators associated to the orthogonal projections
of ∇2 ϕ onto the irreducible components [164, p.745]. Taking ϕ ∈ Γ(Σr M )
to be an eigenvector for D2 and applying Schrödinger-Lichnerowicz’ formula
(1.15) lead to the desired inequality, see [164, Sec. 4] for the rather technical
proof where the authors determine all Weitzenböck formulas involving Dirac
and twistor operators.
The equality case in the inequality is sharp for HPm [189]. Conversely,
if it is sharp, then the spinor bundle of M carries a particular (non-zero)
section called quaternionic Killing spinor [163, p.340]. This spinor induces
a non-zero real Killing spinor on the total space of the SO3 -principal
bundle associated to the quaternionic Kähler structure and for a suitable
metric [163, p.344]. Then Bär’s classification (Theorem A.4.3) of manifolds
with real Killing spinors forces M to be isometric to HPm , we refer to
[163, Sec. 7] for the details. 
3.3 Improving Friedrich’s inequality in a conformal way 55

3.3 Improving Friedrich’s inequality in a conformal way

N. Hitchin [148] noticed in the Riemannian setting (as well as H. Baum


[49] in the pseudo-Riemannian one) that the fundamental Dirac operator is
conformally covariant, see Proposition 1.3.10. This was the starting point for
the following result.

Theorem 3.3.1 (O. Hijazi [128]) Let (M n , g) be an n(≥ 2)-dimensional


closed Riemannian spin manifold and u ∈ C ∞ (M, R), then any eigenvalue λ
of D on (M n , g) satisfies
n
λ2 ≥ inf (Se2u ), (3.16)
4(n − 1) M

where S is the scalar curvature of (M n , g := e2u g). Moreover, (3.16) is an


equality for some eigenvalue λ if and only if u is constant and (M n , g) carries
a non-zero real Killing spinor.
Proof : We use the notations of Proposition 1.3.10. For any ϕ ∈ Γ(ΣM ) one
has from (3.3) applied to ψ := e− 2 u ϕ,
n−1

  
n n
|Dψ|2 − S |ψ|2 vg = |P ψ|2 vg ≥ 0.
M 4(n − 1) n−1 M

Proposition 1.3.10 states that Dψ = e− 2 u Dϕ, so that choosing ϕ to be a


n+1

non-zero eigenvector for D associated to the eigenvalue λ one obtains D ψ =


λe−u ψ and the inequality (3.16). Furthermore, if (3.16) is an equality for
some eigenvalue λ of D on (M n , g), then for any non-zero eigenvector ϕ for
D associated to λ, the identity P ψ = 0 holds, where ψ := e− 2 u ϕ. This
n−1

implies in turn
λe−u
∇X ψ = − X ·ψ
n
for every X ∈ T M . Elementary computations as in the proof of Proposition
A.4.1 but carried out on (M n , g) (see e.g. [134, Prop. 5.12]) show that
necessarily du = 0, thus u is constant and therefore ϕ is a real Killing spinor
on (M n , g). The converse statement follows from the characterization of the
equality case in (3.1). This concludes the proof. 

Corollary 3.3.2 Any eigenvalue λ of D on an n-dimensional closed Rie-


mannian spin manifold (M n , g) satisfies:
i) (C. Bär [35]) For n = 2,

2πχ(M 2 )
λ2 ≥ , (3.17)
Area(M 2 , g)
56 3 Lower eigenvalue estimates on closed manifolds

where χ(M 2 ) is the Euler characteristic of M 2 . Moreover, (3.17) is an


equality for some eigenvalue λ of D if and only if (M 2 , g) is isometric
either to S2 with constant curvature metric or to T2 with flat metric and
trivial spin structure.
ii) (O. Hijazi [128]) For n ≥ 3,
n
λ2 ≥ μ1 , (3.18)
4(n − 1)

where μ1 denotes the first eigenvalue of the scalar conformal Laplace op-
erator 4 n−1
n−2 Δ + S. Moreover, (3.18) is an equality for some eigenvalue λ
of D if and only if (M n , g) carries a non-zero real Killing spinor.
Proof : We deduce both (3.18) and (3.17) from (3.16) and from the following
transformation formula for scalar curvature after conformal change of the
metric:
Se2u = S + 2(n − 1)Δu − (n − 1)(n − 2)|grad(u)|2 , (3.19)
for g := e2u g and u ∈ C ∞ (M, R). This is applied to a conformal metric g for
which Se2u is constant on M . 
Svg
i) Let u0 ∈ C ∞ (M, R) solve Δu0 = 2Area(M M
2 ,g) − 2 (such a solution
S

exists because the r.h.s. has vanishing integral). Since in dimension 


n = 2
2u0 2u0 Sv
the formula (3.19) reads Se = S + 2Δu0 one obtains Se = Area(M 2g,g) .
M

Applying now the Gauss-Bonnet Theorem, (3.16) becomes



1 Svg 2πχ(M 2 )
λ ≥
2 M
= ,
2 Area(M 2 , g) Area(M 2 , g)

which is (3.17). If now (3.17) is an equality for some eigenvalue λ of D, then by


construction of u0 the scalar curvature S must be constant and non-negative.
Since we explicitly know the Dirac spectra of S2 (see Theorem 2.1.3) and of
T2 (see Theorem 2.1.1), we can compare the smallest non-negative Dirac
eigenvalue with the lower bound in (3.17) and state that equality always
occurs for S > 0 and occurs on T2 only when fixing the trivial spin structure.
ii) We can assume that μ1 > 0. From Courant’s nodal domain theorem (see
e.g. [74, p.19]) every non-zero eigenfunction h0 for L := 4 n−1
n−2 Δ+S associated
to its smallest eigenvalue μ1 cannot vanish, hence may be assumed to be
positive. In particular μ1 = h−1 0 Lh0 . But in dimension n ≥ 3 the formula
(3.19) can be rewritten under the following form: for any positive smooth
function h on M n ,
Sh n−2 = h−1 Lh,
4

4
where S is the scalar curvature of (M n , g := h n−2 g). Thus, choosing
4 4
g 0 := h0 g on M n , one obtains Sh0 = μ1 , which together with Theorem
n−2 n−2

3.3.1 implies (3.18). The characterization of the equality case in (3.18)


follows from Theorem 3.3.1 as well. 
3.3 Improving Friedrich’s inequality in a conformal way 57

Another proof of (3.18) involving Kato type inequalities can be found in [71].
Inequality (3.18) improves Friedrich’s inequality (3.1) for n ≥ 3 since ob-
viously μ1 ≥inf (S). It also proves the existence in dimension n ≥ 3 of an
M
explicit conformal lower bound for the spectrum of D2 :
Corollary 3.3.3 (O. Hijazi [130]) For any Riemannian metric g on a
closed n(≥ 3)-dimensional spin manifold M n ,
2 n
2
λ1 (DM,g )Vol(M, g) n ≥ Y (M, [g]), (3.20)
4(n − 1)

2
where λ1 (DM,g ) denotes the smallest non-negative eigenvalue of D2 associ-
ated to the metric g and Y (M, [g]) is the Yamabe invariant of M w.r.t. the
conformal class of g.
Proof : Recall that the Yamabe invariant of M n w.r.t. [g] is the conformal
invariant defined by


(4 n−1
M n−2 Δg f + Sg f )f vg
Y (M, [g]) := ∞
inf  2n n−2
.
f ∈C (M,R)\{0} ( M f n−2 vg ) n
  2n n−2 2
Hölder’s inequality gives M f 2 vg ≤ ( M f n−2 vg ) n ·Vol(M n , g) n . Assuming
Y (M, [g]) > 0 (otherwise (3.20) is trivially satisfied), one obtains
2
μ1 Vol(M n , g) n ≥ Y (M, [g]),

which with (3.18) implies the result. 

Note however that (3.18) is not itself conformal. We also mention that inequa-
lity (3.18) can be combined with lower bounds of μ1 to provide an estimate
2
of λ1 (DM,g ) in terms of the total Q-curvature in dimension n = 4 and of the
first eigenvalue of the so-called Branson-Paneitz operator in dimension n ≥ 5,
see [145, Sec. 4]. The a priori existence of a qualitative conformal lower bound
for the Dirac spectrum was proved independently by J. Lott [181] using the
boundedness of particular Sobolev embeddings. More precisely, if the Dirac
operator of a given closed Riemannian spin manifold (M n , g) is invertible,
then there exists a positive constant c depending only on the conformal class
of g such that [181, Prop. 1]
2
2
λ1 (DM,g )Vol(M, g) n ≥ c (3.21)

for any metric g conformal to g on M n .


C. Bär’s estimate (3.17) gives a topologically invariant lower bound on the
Dirac spectrum. Surprisingly enough this contrasts with the situation of the
scalar Laplacian on S2 for which this invariant provides an upper bound for
58 3 Lower eigenvalue estimates on closed manifolds

the first non-zero eigenvalue in one of the corresponding estimates established


by J. Hersch (see reference in [42]) and which reads

λ1 (ΔS2 ,g )Area(S2 , g) ≤ 8π, (3.22)

where λ1 (ΔS2 ,g ) denotes the smallest positive eigenvalue of the scalar Laplace
operator Δ on (S2 , g). For lower bounds in higher genus, where (3.17) is
trivial, see Section 3.6.
As noticed in the proof of Corollary 3.3.2, one has from the Gauss-Bonnet

2πχ(M 2 ) 2 M2
Svg
Theorem in dimension 2 the following identity: Area(M 2 ,g) = 4(2−1) Area(M 2 ,g) .

Can one improve Friedrich’s


 inequality (3.1) in dimension n ≥ 3 by replacing
1
inf M (S) by Vol(M,g) M
Sv g B. Ammann and C. Bär showed [16] that this
?
is not the case at all: on any compact spin manifold of dimension n ≥ 3
and for any positive integer k there exists a sequence of Riemannian metrics
for which the k th Dirac eigenvalue remains bounded whereas the averaged
total scalar curvature tends to infinity. We refer to [42] for a detailed and
illustrated proof.

3.4 Improving Friedrich’s inequality


with the energy-momentum tensor

The main idea to prove inequality (3.1) was to split the spinorial Levi-Civita
connection in a clever way so as to make the term which is dropped off after
integration and application of the Schrödinger-Lichnerowicz formula as small
as possible. This led to the introduction of the Penrose operator. In an equiv-
alent way, this means deforming the Levi-Civita connection in the direction
of IdT M , i.e., defining TX ϕ := ∇X ϕ + f X · ϕ for some real or complex-valued
function f to be fixed later, see T. Friedrich’s method of proof in Section 3.1.
O. Hijazi’s idea for the following result is to introduce a different Penrose-
like operator, deforming the Levi-Civita connection in the direction of some
symmetric 2-tensor tensor T ψ associated to an eigenvector ψ:
Theorem 3.4.1 (O. Hijazi [132]) Let λ be an eigenvalue of the fundamen-
tal Dirac operator on a closed n(≥ 2)-dimensional closed Riemannian spin
manifold (M n , g) and ψ be a non-zero eigenvector for D to the eigenvalue λ.
Then 
S
λ ≥ inf
2
+ |T | ,
ψ 2
(3.23)
Mψ 4
 
where T ψ (X, Y ) := 12 e X · ∇Y ψ + Y · ∇X ψ, |ψ|
ψ
2
for all X, Y ∈ T M
and Mψ := {x ∈ M | ψ(x) = 0}. If furthermore (3.23) is an equality then ψ
solves
∇X ψ = −T ψ (X) · ψ (3.24)
3.4 Improving Friedrich’s inequality with the energy-momentum tensor 59

for all X ∈ T M .
on ΣM (and outside the
Proof : Define the following modified connection ∇
zero set of ψ, of which measure vanishes) by

X ψ := ∇X ψ + T ψ (X) · ψ

for every X ∈ T M . We compute in a local orthonormal frame {ej }1≤j≤n :


n
2=
|∇ψ| e ψ|2
|∇ j
j=1

n
 
= |∇ej ψ|2 + |T ψ (ej )|2 |ψ|2 + 2e ∇ej ψ, T ψ (ej ) · ψ

j=1

n
 
= |∇ψ|2 + |T ψ |2 |ψ|2 − 2 T ψ (ej , ek )e ek · ∇ej ψ, ψ

j,k=1

= |∇ψ| − |T | |ψ|
2 ψ 2 2

since from its definition the tensor T ψ is symmetric. Integrating and applying
the Schrödinger-Lichnerowicz formula (1.15) leads straightforward to the
inequality, of which limiting-case implies ∇ψ = 0. This concludes the proof. 

The tensor T ψ is sometimes called the energy-momentum tensor associated


to ψ, see e.g. [47, Sec. 6] for a justification of its name.
The lower bound in (3.23) for the eigenvalue λ has the obvious disadvan-
tage to depend on the eigenvector ψ to λ, hence Theorem 3.4.1 does not
directly provide a geometric lower bound for the Dirac spectrum. Note how-
ever that (3.23) improves Friedrich’s inequality (3.1) whatever the tensor T ψ
tr (T ψ )
could be: one can indeed write T ψ = (T ψ )0 + g n g, where (T ψ )0 de-
notes the trace-free part of T ψ . Since Dψ = λψ one has in any local o.n.b.
{ej }1≤j≤n of T M :


n 
ψ
ψ
trg (T ) = e ej · ∇ej ψ,

j=1
|ψ| 2


ψ
= e Dψ,

|ψ|2
= λ,

λ2
so that |T ψ |2 = |(T ψ )0 |2 + n and
60 3 Lower eigenvalue estimates on closed manifolds
 n n 
λ2 ≥ inf S+ |(T ψ )0 |2 ,
Mψ 4(n − 1) n−1
  
≥0

which implies (3.1).


One can also remark that the proof of Theorem 3.4.1 only needs ψ to be
eigen for D2 , which is weaker than ψ be eigen for D. However the comparison
with Friedrich’s inequality as just above is in this case not available.
Closed spin manifolds carrying a non-zero eigenvector ψ of D satisfying
(3.24) have not been completely classified yet. From the above comparison
with Friedrich’s inequality (3.1) they contain all manifolds carrying non-zero
real Killing spinors. Recent works [119, 104], where examples of manifolds
are given where (3.23) is sharp but not (3.1) (e.g. Heisenberg manifolds, see
[104, Ex. 6.4]), show that they form a strictly larger family.
Besides we mention that Theorem 3.4.1 was generalized by T. Friedrich
and E.-C. Kim [93, Lemma 5.1] and by G. Habib [119, Thm. 2.2.1] (see also
[122]). More recently, an analogous ansatz was successfully carried out by
T. Friedrich and E.C. Kim [94, Thm. 1.1] where the lower bound for the
Dirac spectrum depends on the spectrum of a Dirac-type operator associated
to a so-called nondegenerate Codazzi tensor, we refer to [94] for details.

3.5 Improving Friedrich’s inequality with other


curvature components

In case the scalar curvature of a compact spin manifold is not everywhere


positive one can try to look for lower eigenvalue bounds involving the Ricci
and Weyl components of the curvature tensor. The proof of the following
theorems relies on the application of the Schrödinger-Lichnerowicz formula
(1.15) after a suitable choice of Penrose-like operator involving those tensors,
see e.g. [161] for the highly technical details.

Theorem 3.5.1 (T. Friedrich and K.-D. Kirchberg [96]) Any eigen-
value λ of D on an n(≥ 2)-dimensional closed Riemannian spin manifold
(M n , g) with divergence-free curvature tensor, vanishing scalar curvature and
nowhere-vanishing Ricci-curvature satisfies:

1 inf M |Ric|2
λ2 >  ,
4 n−1 inf |Ric| − κ
n M 0

where κ0 denotes the smallest eigenvalue of the Ricci tensor Ric on M .


Examples of closed Riemannian manifolds satisfying the assumptions of
Theorem 3.5.1 and where the lower bound can be explicitly computed can
3.6 Improving Friedrich’s inequality on surfaces of positive genus 61

be found among the following families, see [96, Ex. 1-4] and [160, Ex. 4.1
& 4.2]: (local) Riemannian products of Einstein manifolds, warped prod-
ucts of S1 with an Einstein manifold with positive scalar curvature, warped
products on Riemannian surfaces, conformally flat manifolds. Note however
that Einstein manifolds themselves or manifolds whose Ricci tensor vanishes
somewhere cannot be handled by Theorem 3.5.1. This was the motivation
of T. Friedrich and K.-D. Kirchberg for obtaining a lower bound involving
the Weyl tensor only. The best result in this direction was obtained by K.-D.
Kirchberg, generalizing an earlier one by T. Friedrich and himself [95, Thm.
3.1]:
Theorem 3.5.2 (K.-D. Kirchberg [161]) Any eigenvalue λ of D on
an n(≥2)-dimensional closed Riemannian spin manifold (M n , g) with
divergence-free Weyl-tensor and μ > 0 satisfies:

1 n 4ν0 2
λ ≥
2
(2n − 1) inf (S) + inf (S)2 + ( ) , (3.25)
8(n − 1) M M n−1 μ

where ν0 ≥ 0 and μ are conformal invariants depending on the Weyl tensor


only.
Recall that every Einstein Riemannian manifold has divergence-free Weyl
tensor. In case inf M (S) > 0 inequality (3.25) obviously enhances Friedrich’s
inequality (3.1). In case inf M (S) ≤ 0 it is easy to see that the lower bound
in (3.25) is positive if and only if ν0 > (n−1)μ
2 | inf M (S)|. However, it is up to
now not known if (3.25) can be an equality [161, Rem. 4.2.ii)].
Theorems 3.5.1 and 3.5.2 actually follow from a whole series of estimates
[96, 161] involving curvature tensors and that can be applied to produce fine
vanishing theorems for the kernel of the Dirac operator.

3.6 Improving Friedrich’s inequality on surfaces


of positive genus

C. Bär’s inequality (3.17) does not give any information on the spectrum
of D on compact Riemannian surfaces with nonpositive Euler characteristic,
i.e., with positive genus. Estimates on such surfaces have to depend on the
choice of spin structure, as the example of the 2-torus already shows: for its
trivial spin structure (i.e., for the spin structure coming from the trivial lift of
the lattice-action to the spin level, see Proposition 1.4.2) it admits harmonic
spinors - for flat hence any metrics because of (1.16) - but not for any other
spin structure [86].
The first estimate to have been proved is a qualitative one and dates back
to J. Lott’s work [181] providing lower bounds for general conformally covari-
ant elliptic self-adjoint linear differential operators. In the case of surfaces it
62 3 Lower eigenvalue estimates on closed manifolds

states that, if the Dirac operator of a given closed oriented surface (M 2 , g) is


invertible, then there exists a positive constant c such that, for any metric g
conformal to g on M 2 (see (3.21)):
2
λ1 (DM,g )Area(M, g) ≥ c.

The constant c expresses the boundedness of particular Sobolev embeddings


hence cannot be made explicit in general.
The first successful attempt in looking for a geometric estimate is due to B.
Ammann [14]. His lower bound, which was proved for the 2-torus, involves
the so-called spinning systole spin-sys(M ) of a closed oriented surface M
with positive genus which is defined to be the minimum of the lengths of
all noncontractible loops (in our convention, loops are simply closed curves)
along which the induced spin structure is non trivial. Recall that the systole
of M is defined to be the minimum of the lengths of all noncontractible loops
in M .
Theorem 3.6.1 (B. Ammann [14]) Let g be an arbitrary Riemannian
metric on the 2-torus M := T2 carrying a non-trivial spin structure. Assume
that Kg L1 (T2 ,g) < 4π, where Kg ist the Gauss curvature of (T2 , g). Then
there exists for each p > 1 a constant Cp > 0 depending on Kg L1 (T2 ,g) ,
Kg Lp (T2 ,g) , the area and the systole of (T2 , g) such that any eigenvalue λ
of D satisfies
sup Cp
p>1
λ2 ≥ .
spin-sys(T2 )2
Moreover this inequality is an equality if and only if g is flat, the lattice is
generated by an orthogonal pair and the spin structure is the (1, 0)- or (0, 1)-
one.

Sketch of proof of Theorem 3.6.1: The proof of the inequality combines the
following steps. First one chooses a flat metric g0 := e2u g in the conformal
class of g. Using the min-max principle (see e.g. Lemma 5.0.2) it can be easily
proved that λ2 ≥ e2 max(u) λ20 , where λ0 > 0 is the smallest Dirac eigenvalue
(in absolute value) on (T2 , g0 ) and for the same spin structure. Now the Dirac
spectrum of (T2 , g0 ) for any spin structure is explicitly known (see Theorem
2.1.1), in particular the following equality holds

λ20 Area(T2 , g0 ) = 4π 2 χ2L2 (T2 ,g0 ) ,

where χ ∈ H 1 (T2 , Z2 ) is the cohomology class representing the spin


structure (it is non-zero if the spin structure is non-trivial). Obviously
Area(T2 , g) ≥ e−2 min(u) Area(T2 , g0 ) so that one obtains a lower bound of
λ in terms of the area of (T2 , g), of the L2 -norm of χ and of the so-called
oscillation osc(u) := max(u) − min(u) of u on T2 . On the other hand the
L2 -norm of χ can be proved to be only dependent of the conformal class of
3.6 Improving Friedrich’s inequality on surfaces of positive genus 63

g and can be estimated against an expression involving the spinning systole,


the area and osc(u) [14, Sec. 4]. What remains - the whole work - is to
estimate osc(u) against the desired geometric data. For an illustrated proof
of this Sobolev-type inequality we refer to [14, Sec. 6]. The limiting-case
occurs if and only if osc(u) = 0 (i.e., g is flat) and the estimate of osc(u) is
sharp, which yields strong conditions on the lattice defining the flat metric
g, see [14]. 

Another and completely different approach was developed by B. Ammann


and C. Bär in [17]. It aimed at obtaining a lower bound in terms of a geome-
tric invariant called the spin-cut diameter δ(M ). This is a positive number
which is associated to the surface M and its spin structure. The idea is
simple: apply (3.17) to the surface obtained from the genus g surface M by
cutting g suitable loops out of M . Here “suitable” means the following: on
the one hand one has to choose the loops such that the resulting surface M  is
diffeomorphic to an open subset of S2 - actually to a 2-sphere with 2g disks
removed; this is the case as soon as the Z2 -homology classes associated to
those loops form a basis of H1 (M, Z2 ). On the other hand the cut-out-process
must also respect the spin structures in the sense that the restrictions of the
original one and of the one from S2 have to coincide on M . This however
is only possible if the so-called Arf-invariant (which associates to each spin
structure on M the number 1 or −1, see [17, Def. p.430]) of the spin structure
of M is 1 [17, Cor. 3.3]. The spin-cut diameter can then be defined from the
distances between the cut-out loops, see [17, Def. p.433]. For M = T2 the
Arf-invariant of the trivial spin structure is −1 and it is 1 for the other ones.
In the latter case, extending by means of suitable cut-offs an eigenvector on
M to the S2 obtained by adding two disks to the gluing of a finite number
of copies of M (which is then a cylinder) one can prove the following result
[17, Sec. 5]:
Theorem 3.6.2 (B. Ammann and C. Bär [17]) Let M := T2 be the
2-torus with arbitrary Riemannian metric and non-trivial spin structure.
Then any eigenvalue λ of D satisfies
! " #
2 π 2
|λ| ≥ sup − 2)
+ + , (3.26)
k∈N kδ(T kArea(T2 ) k 2 δ(T2 )2
k=0

where δ(T2 ) is the spin-cut diameter of T2 associated to this spin structure.


$ √ 2
%
The supremum in the lower bound is attained for k = 4( 2+1)Area(T 2 2
)
$ √ % πδ(T )
4( 2+1)Area(T2 )
or k = πδ(T2 )2 + 1. It is positive and for the boundary T2ε of an

ε-tubular neighbourhood of a circle of radius 1ε it is asymptotic to Area(T π
2)
ε

when ε tends to 0. Therefore Theorem 3.6.2 can be viewed as a generalization


of Corollary 3.3.2.ii) for T2 with non-trivial spin structure.
64 3 Lower eigenvalue estimates on closed manifolds

In genus g ≥ 1 one can apply the same argument to the S2 obtained by


 and prove [17, Sec. 6]:
adding disks to a clever gluing of 2g + 1 copies of M

Theorem 3.6.3 (B. Ammann and C. Bär [17]) Let M be a closed Rie-
mannian surface of positive genus g with spin structure whose Arf-invariant
equals 1. Then any eigenvalue λ of D satisfies

2 π 1
|λ| ≥ · − . (3.27)
2g + 1 Area(M ) δ(M )

Although the lower bound need this time not be positive there exist examples
for which it is: as above, consider an ε-tubular neighbourhood Mε of a closed
plane curve with exactly g − 1 intersections and such that, w.r.t. any allowed
spin structure, δ(Mε ) ∼ cst (fix for instance the diameter equal to 1ε ). Then
ε→0 ε 
the lower bound is asymptotic to 2g+1 2
· Area(M
π
ε)
for ε → 0. In the case
where g = 1 the k-dependent expression in the r.h.s. of (3.26) is for k = 2
greater than the r.h.s. of (3.27), so that (3.26) is better than (3.27).
Combining Theorems 3.6.2 and 3.6.3 with the extrinsic upper bound (5.19)
for the smallest Dirac eigenvalue for surfaces embedded in R3 one obtains
a lower bound of the Willmore functional, see [17, Thm. 7.1]. Besides, we
mention that Theorems 3.6.2 and 3.6.3 can be extended to complete surfaces
with finite area [17, Thm. 8.1].

3.7 Improving Friedrich’s inequality on bounding


manifolds

In case M bounds a compact spin manifold M  the input of extrinsic geome-


trical data - such as the mean curvature of M in M - can improve Friedrich’s
inequality (3.1). The main step consists in solving a suitable boundary value
problem, see also Chapter 4. The following theorem was proved by O. Hijazi,
S. Montiel and X. Zhang in [142] for c = 0 and by O. Hijazi, S. Montiel and
A. Roldán in [140] for c < 0. Recall that D2 is the operator acting on the
sections of Σ := ΣM or Σ := ΣM ⊕ ΣM and which is defined by D2 := D
for n even or D2 := D ⊕ −D for n odd respectively, see Proposition 1.4.1.
Theorem 3.7.1 Let M n = ∂ M , where M  is a compact Riemannian spin
manifold. Assume that, for a constant c ≤ 0, the scalar curvature S of M
 and
the mean curvature  
H of M in M w.r.t. the inner normal satisfy S ≥ (n+1)nc

and H ≥ −c respectively. Then for any eigenvalue λ of D,
n 
|λ| ≥ inf ( H 2 + c). (3.28)
2 M
3.7 Improving Friedrich’s inequality on bounding manifolds 65

Moreover (3.28) is an equality √


if and

only if H is constant, the manifold
 c c
(M , g) admits a non-trivial 2 - or − 2 -Killing spinor and the eigenspace of

D2 to the eigenvalue n2 inf M ( H 2 + c) coincides with (K0 )|M for c = 0 and
with
 √ √
(Id + (−1)j i(H − H 2 + c)ν·)K(−1)j √c (M , g)| if H > −c
M
j=0,1 2
 √
, g)|
K(−1)j √c (M if H = −c
M
j=0,1 2


, g) denotes the space of ±
for c < 0, where K± √c (M c
2
2 -Killing spinors on
, g).
(M
 the spinorial Levi-Civita connection
Proof in the case c = 0: Denote by ∇
. The Schrödinger-Lichnerowicz formula for the Dirac operator D
of M  of
, g) and elementary computations as in Section 1.3 show that, for any
(M
ϕ ∈ Γ(ΣM),
 
 2
|Dϕ| =  2 ϕ, ϕ
+ div  (V )
e D M
  
(1.15)
= e ∇  ϕ
+ S |ϕ|2 + div  (V )
 ∗ ∇ϕ,
4 M


=  2 + S |ϕ|2 + div  (V + W ),
|∇ϕ| (3.29)
4 M

where 
 V and Ware the vector fields on
 M defined by the relations g(V, X) :=
 ϕ
and g(W, X) := e ∇
e X · Dϕ, X ϕ, ϕ
for all X ∈ T M
 respectively

(remember that we denote by “ · ” the Clifford multiplication of M and not



that of M ). Splitting |∇ϕ| as in (A.11) one comes to
2

S 2 n  2
|ϕ| − |Dϕ| = −|Pϕ|2 − divM
 (V + W ),
4 n+1

where P is the Penrose operator of (M


n+1 , g), see Appendix A. Let ν be the
. Integrating the last identity and
inner unit normal vector field of M in M
applying Green’s formula one obtains
 ! # 
S 2 n  2  
|ϕ| − |Dϕ| vgM = − |Pϕ|2 vgM

M 4 n+1 
M


− divM M
 (V + W )vg


M


= − |Pϕ|2 vgM − g(V + W, ν)vg

M M
66 3 Lower eigenvalue estimates on closed manifolds


= − |Pϕ|2 vgM


M
 
−  ϕ
+ ∇
e ν · Dϕ, ν ϕ, ϕ
vg
M
(1.22) 
= − |Pϕ|2 vgM

 M

nH 2
+ e D2 ϕ, ϕ
− |ϕ| vg .
M 2

Since D2 is formally self-adjoint (Proposition 1.3.4) one comes to


 ! # 
S 2 n  2  
|ϕ| − |Dϕ| vgM = − |Pϕ|2 vgM

M 4 n+1 
M
 
nH 2
+ D2 ϕ, ϕ
− |ϕ| vg . (3.30)
M 2

Let λ be an eigenvalue of D. The spectrum of D2 being the symmetrized of


that of D w.r.t. the origin (for n even it follows from (1.10) that the spectrum
of D is already symmetric w.r.t. the origin) there always exists a non-zero
eigenvector ψ for D2 associated to the eigenvalue |λ|. The crucial point is
now the existence of a smooth solution φ to the boundary value problem
with APS-boundary condition

 Dφ
 =0 
on M
 (3.31)
 π≥0 φ = ψ on M,

where π≥0 : Γ(Σ) → Γ(Σ) denotes the L2 -orthogonal projection onto the
eigenspaces of D2 to nonnegative eigenvalues, see Section 1.5 and Chapter 4.
Since S ≥ 0 and H ≥ 0 the identity (3.30) with ϕ := φ implies
  ! #
S 2 M  S 2 n  2 
0≤ |φ| vg = |φ| − |Dφ| vgM

M 4 
M 4 n + 1
 
nH 2
≤ D2 φ, φ
− |φ| vg
M 2
 
nH
≤ D2 π≥0 φ, π≥0 φ
− |π≥0 φ|2 vg
2
M
nH
= (|λ| − )|ψ|2 vg
M 2

from which the inequality follows.


In case the lower bound is attained the mean curvature H of M must be
constant, φ = π≥0 φ on M and Pφ = 0 on M , where φ is any section of ΣM

3.7 Improving Friedrich’s inequality on bounding manifolds 67

solving (3.31) for any given eigenvector ψ of D2 associated to the eigenvalue


|λ|. In particular ψ = φ|M with ∇φ  = 0 on M , i.e., every eigenvector ψ of
D2 associated to the eigenvalue |λ| must be the restriction on M of parallel
spinor on M (note that the existence of a non-zero parallel spinor on M  im-
plies S = 0, see Proposition A.4.1). Moreover (1.22) already implies that the
restriction of any parallel spinor onto a hypersurface with constant mean cur-
vature is an eigenvector of D to the eigenvalue nH 2 . Therefore the eigenspace
nH
of D2 associated to the eigenvalue 2 exactly coincides with (K0 )|M . The
other implication is trivial.
For the proof in the case c < 0, which is based on the same argument for
a Schrödinger operator associated to D, we refer to [140]. 

⊂M
In case M n+1 (c), where Mn+1 (c) is a spaceform with constant curvature
c ≤ 0, Gauss’ equations imply in particular ( n2 )2 (H 2 + c) ≥ 4(n−1)
n
S, hence

(3.28) improves (3.1) under the supplementary assumption H ≥ −c.
There exists a conformal version of (3.28) in terms of the so-called Yamabe
relative invariant, see [144].
The characterization of the equality case in (3.28) provides a short proof
of Alexandrov’s theorem (see reference in [142]) on constant mean curvature
embedded hypersurfaces in the Euclidean and hyperbolic spaces respectively
[142, Thm. 8]:
Theorem 3.7.2 (A.D. Alexandrov) Every closed embedded hypersurface
with constant mean curvature in Rn+1 or Hn+1 is a round geodesic hyper-
sphere.
Proof in the case c = 0: Let M be such a hypersurface. It is embedded so
that on the one hand it bounds a compact domain M  (in particular it is
orientable hence spin, see Proposition 1.4.1); on the other hand, it can be
shown that necessarily H ≥ 0 by a result of S. Montiel and A. Ros (see
reference in [142]). Moreover the assumption H constant implies that (3.28)
is an equality, in which case every non-zero eigenvector of D2 associated to
the eigenvalue nH 
2 must be the restriction onto M of a parallel spinor on M
(Theorem 3.7.1). But considering the spinor field

x −→ ϕx := νx · φ + Hx · φ,

, where ν is the inner unit normal and φ a parallel spinor on M


on M  ⊂ Rn+1 ,
one notices that

D2 ϕ = D2 (ν · φ) + HD2 (x · φ)
(1.19)
= −ν · D2 φ + HD2 (x · φ)
(1.22) nH  ν (x · φ) − ν · D(x
 · φ))
= −ν · D2 φ + H( x·φ−∇
2
68 3 Lower eigenvalue estimates on closed manifolds

nH nH
= − ν · φ + H( x · φ − ν · φ + (n + 1)ν · φ)
2 2
nH
= (ν · φ + Hx · φ)
2
nH
= ϕ,
2

i.e., ϕ is an eigenvector for D2 associated to the eigenvalue nH


2 . Therefore ϕ
must be either identically zero or non-zero and parallel. The first possibility
already implies that M must be a geodesic sphere. The second one means
that, for every X ∈ T M ,

 X ϕ = −A(X) · φ + HX · φ.
0=∇

Since φ has no zero on M  one deduces that A = HIdT M , i.e., that M must

be totally umbilical in M hence in Rn+1 . This concludes the proof for c = 0.
For c < 0 we refer to [140]. 

Another clever application of Theorem 3.7.1 is:

Theorem 3.7.3 (O. Hijazi and S. Montiel [138]) Let (M n+1 , g) be a


complete Riemannian spin manifold with nonnegative Ricci curvature, mean
convex boundary ∂ M  and nonnegative Einstein-tensor along the normal di-
 n+1 , g) is isometric to a Euclidean ball.
rection of ∂ M . Then (M
As a corollary, any Ricci-flat complete spin manifold with boundary isometric
to the round sphere Sn is already isometric to a Euclidean ball. Recently rigid-
ity results have been obtained by S. Raulot [213] under weaker assumptions
on the boundary. We also mention that it remains open whether analogous
estimates on the boundary of positively-curved domains can be obtained.
Chapter 4
Lower eigenvalue estimates
on compact manifolds with boundary

The study of the Dirac operator on compact spin manifolds M with bound-
ary was initiated by Atiyah, Patodi and Singer [27] in the search for index
theorems on such manifolds, see e.g. [59] which is a comprehensive reference
on the subject. In order to be able to talk about eigenvalues of the Dirac oper-
ator of M in this context, elliptic boundary conditions have to be introduced
as we have seen in Section 1.5. Following [82] a whole bunch of spectral prop-
erties of the Dirac operator have recently been proved using varied boundary
conditions, leading sometimes to very beautiful geometric results, such as
those already presented in Section 3.7. In this chapter we mainly show that,
under any of the four boundary conditions introduced in Section 1.5, some
kind of Friedrich’s inequality (3.1) holds on M although the lower bound is
not always attained according to the boundary condition chosen. For readers
interested in more details and references we suggest [139].

4.1 Case of the gAPS boundary condition

Remember that the generalized Atiyah-Patodi-Singer (gAPS) boundary con-


dition is defined by BgAPS := π≥β , i.e., it is the L2 -orthogonal projection onto
the (Hilbert) direct sum of the eigenspaces of the boundary Dirac operator
D to the eigenvalues not smaller than a fixed β ≤ 0. The following theorem
is a particular case of [75, Thm. 3.1] which is itself a generalization of [143,
Thm. 4] (see also [195, Thm. 16.5]).
Theorem 4.1.1 Let (M n , g) be an n(≥ 2)-dimensional compact Rieman-
nian spin manifold with non-empty boundary ∂M . Assume that ∂M has
non-negative mean curvature w.r.t. the inner normal. Then any eigenvalue
λ of the fundamental Dirac operator D of (M n , g) under the gAPS boundary
condition satisfies
n
λ2 > inf (S),
4(n − 1) M
where S is the scalar curvature of (M n , g).

N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 69


DOI 10.1007/978-3-642-01570-0 4, 
c Springer-Verlag Berlin Heidelberg 2009
70 4 Lower eigenvalue estimates on compact manifolds with boundary

Proof : There exists a non-zero smooth solution ϕ ∈ Γ(ΣM ) to the boundary


value problem 
 Dϕ = λϕ on M

 π≥β (ϕ| ) = 0 on ∂M.
∂M

Since π≥β (ϕ|∂M ) = 0 and ϕ|∂M = 0 (otherwise a unique continuation property


 the Dirac ∂M
for operator D [59, Sec. 1.8] would give ϕ = 0 on M ) one has
∂M
D 2 ϕ, ϕ
vg < β ∂M |ϕ|2 vg∂M ≤ 0, so that (3.30) with H ≥ 0 (beware
the different notations, in particular for the boundary Dirac operator D2 )
implies
 
S 2 n−1
|ϕ| − |Dϕ|2 vg < 0,
M 4 n
which straightforward implies the inequality. In particular it cannot be
sharp. 

4.2 Case of the CHI boundary condition

The chirality (CHI) boundary condition is defined by BCHI := 12 (Id − ν · G),


where ν is the inner unit normal and G is an endomorphism-field of ΣM
(whose restriction on ∂M is also denoted by G) which is involutive, unitary,
parallel and anti-commuting with the Clifford multiplication on M .

Theorem 4.2.1 (O. Hijazi, S. Montiel and A. Roldán [139])


Let (M n , g) be an n(≥ 2)-dimensional compact Riemannian spin manifold
with non-empty boundary ∂M . Assume that ∂M has non-negative mean cur-
vature w.r.t. the inner normal. Then any eigenvalue λ of the fundamental
Dirac operator D of (M n , g) under the CHI boundary condition satisfies
n
λ2 ≥ inf (S), (4.1)
4(n − 1) M

where S is the scalar curvature of (M n , g). Moreover, (4.1) is an equality if


n
and only if (M n , g) is isometric to the half sphere with radius 2|λ| .

Proof : There exists a non-zero smooth solution ϕ ∈ Γ(ΣM ) to the boundary


value problem 
 Dϕ = λϕ on M

 (Id − ν · G)(ϕ| ) = 0 on ∂M.
∂M

From (1.22) and the definition of G it can be easily proved that D2 G = GD2 ,
so that, if ψ := ϕ|∂M then
4.2 Case of the CHI boundary condition 71
 
D2 ψ, ψ
vg∂M = D2 (ν · Gψ), ψ
vg∂M
∂M ∂M

(1.23)
= − ν · D2 (Gψ), ψ
vg∂M
∂M

= − ν · G(D2 ψ), ψ
vg∂M
∂M

= − D2 ψ, ν · Gψ
vg∂M
∂M

= − D2 ψ, ψ
vg∂M ,
∂M

that is, ∂M D2 ψ, ψ
vg∂M = 0. Formula (3.30) together with the assumption
H ≥ 0 imply  
S 2 n−1
|ϕ| − |Dϕ| vg ≤ 0,
2
M 4 n
which leads to (4.1).
If inequality (4.1) is an equality, then (3.30) implies on the one hand
that ϕ must be a Killing spinor to the real Killing constant − nλ (it is an
eigenvector of D lying in the kernel of the Penrose operator P of (M n , g), see
Appendix A) and on the other hand H = 0, i.e., the boundary ∂M must be
minimal in M . Moreover, f := Gϕ, ϕ
defines a smooth real function on M
whose differential is given on any X ∈ M by

X(f ) = ∇X (Gϕ), ϕ
+ Gϕ, ∇X ϕ

λ λ
= G(− X · ϕ), ϕ
− Gϕ, X · ϕ

n n
λ
= − ( Gϕ, X · ϕ
+ X · ϕ, Gϕ
)
n

= − e ( Gϕ, X · ϕ
) .
n
Hence the Hessian of f evaluated on any X, Y ∈ T M is given by


Hess(f )(X, Y ) = − e ( ∇X (Gϕ), Y · ϕ
+ Gϕ, Y · ∇X ϕ
)
n
2λ2
= 2 e ( G(X · ϕ), Y · ϕ
+ Gϕ, Y · X · ϕ
)
n
2λ2
= e ( Gϕ, X · Y · ϕ
+ Gϕ, Y · X · ϕ
)
n2
4λ2
=− Gϕ, ϕ
g(X, Y ),
n2
72 4 Lower eigenvalue estimates on compact manifolds with boundary
2
i.e., Hess(f ) = − 4λ
n2 f g. On the other hand neither λ nor the function f
vanish, since (1.26) implies
 
Dϕ, Gϕ
− ϕ, D(Gϕ)
vg = ψ, ν · Gψ
vg∂M ,
M ∂M

i.e., 2λf = ∂M |ψ|2 vg∂M , which does not vanish because of the unique
continuation property mentioned above. Hence Reilly’s characterization of
the hemisphere (see reference in [139]) implies that (M n , g) is isometric to a
n
hemisphere with radius 2|λ| . Since the standard sphere for its canonical spin
structure and metric with constant sectional curvature 1 carries non-zero
− 12 and 12 -Killing spinors (see e.g. Examples A.1.3.2) the other implication
is trivial. 

In the context of manifolds with non-empty boundary the conformal covarian-


ce of the fundamental Dirac operator also improves the Friedrich-type lower
bound (4.1). For the CHI boundary condition S. Raulot proved in [211, 212]
a Hijazi-type inequality (3.18), which in dimension 2 is equivalent to a Bär-
type inequality (3.17) and where the lower bound is the smallest eigenvalue
of the Yamabe operator under a suitable boundary condition. Moreover, the
lower bound is attained exactly for the round hemispheres in Rn+1 .

4.3 Case of the MIT bag boundary condition

Remember that the “MIT bag” (denoted by MIT) boundary condition is by


definition the endomorphism-field of Σ given by BMIT := 12 (Id − iν·), where
ν is the inner unit normal of the boundary.
Theorem 4.3.1 (O. Hijazi, S. Montiel and A. Roldán [139])
Let (M n , g) be an n(≥ 2)-dimensional compact Riemannian spin manifold
with non-empty boundary ∂M . Assume that ∂M has non-negative mean cur-
vature w.r.t. the inner normal. Then any eigenvalue λ of the fundamental
Dirac operator D of (M n , g) under the MIT bag boundary condition satisfies
n
|λ|2 > inf (S), (4.2)
4(n − 1) M

where S is the scalar curvature of (M n , g).


Proof : There exists a non-zero smooth solution ϕ ∈ Γ(ΣM ) to the boundary
value problem 
 Dϕ = λϕ on M

 (Id − iν·)(ϕ| ) = 0 on ∂M.
∂M

Defining ψ := ϕ|∂M , one has


4.3 Case of the MIT bag boundary condition 73
 
D2 ψ, ψ
vg∂M = D2 (iν · ψ), ψ
vg∂M
∂M ∂M

(1.23)
= − iν · D2 (ψ), ψ
vg∂M

∂M

= − D2 (ψ), iν · ψ
vg∂M

∂M

= − D2 ψ, ψ
vg∂M ,
∂M

that is, ∂M D2 ψ, ψ
vg∂M = 0. Formula (3.30) together with the assumption
H ≥ 0 imply  
S 2 n−1
|ϕ| − |Dϕ|2 vg ≤ 0,
M 4 n
which leads to (4.2).
If (4.2) is an equality, then again ϕ must be a − nλ -Killing spinor on
(M n , g). Since in that case m(λ) > 0 (see Section 1.5), one deduces from
Proposition A.4.1 that λ ∈ iR∗+ and hence S = 4 n−1 2
n λ < 0, contradiction.
Therefore (4.2) is always a strict inequality. 

Assuming the stronger condition H > 0 on ∂M , Theorem 4.3.1 can be im-


proved. More precisely, using a suitable modified connection (or Penrose-like
operator) S. Raulot showed the following:

Theorem 4.3.2 (S. Raulot [210, 212]) Let (M n , g) be an n(≥ 2)-dimen-


sional compact Riemannian spin manifold with non-empty boundary ∂M .
Assume that ∂M has positive mean curvature w.r.t. the inner normal. Then
any eigenvalue λ of the fundamental Dirac operator D of (M n , g) under the
MIT bag boundary condition satisfies
n
|λ|2 ≥ inf (S) + nm(λ) inf (H). (4.3)
4(n − 1) M ∂M

Moreover, (4.3) is an equality if and only if the mean curvature of the bound-
ary is constant and (M n , g) admits a non-zero imaginary Killing spinor.
Besides, as for the CHI boundary condition, there exists a Hijazi-type
conformal lower bound for the Dirac spectrum under MIT bag boundary
condition, which was proved by S. Raulot in [211, 212] but which is never
sharp.
74 4 Lower eigenvalue estimates on compact manifolds with boundary

4.4 Case of the mgAPS boundary condition

The modified generalized Atiyah Patodi Singer (mgAPS) boundary condi-


tion is defined by BmgAPS := BgAPS (Id + ν·), where BgAPS = π≥β is the
generalized Atiyah Patodi Singer boundary condition and ν is the inner unit
normal to the boundary. It depends in particular on a parameter β ≤ 0.
The following theorem is a particular case of [75, Thm. 3.3] which itself
generalizes [139, Thm. 5].

Theorem 4.4.1 Let (M n , g) be an n(≥ 2)-dimensional compact Riemannian


spin manifold with non-empty boundary ∂M . Assume that ∂M has non-
negative mean curvature w.r.t. the inner normal. Then any eigenvalue λ
of the fundamental Dirac operator D of (M n , g) under the mgAPS boundary
condition to β satisfies
n
λ2 ≥ inf (S), (4.4)
4(n − 1) M

where S is the scalar curvature of (M n , g). Furthermore (4.4) is an equality


β
if and only if (M n , g) carries a non-zero α-Killing spinor for real α < n−1
and ∂M is minimal in M .
Proof : There exists a non-zero smooth solution ϕ ∈ Γ(ΣM ) to the boundary
value problem 
 Dϕ = λϕ on M
 (4.5)
 π≥β (Id + ν·)(ϕ| ) = 0 on ∂M.
∂M

Again if we prove that ∂M D2 ψ, ψ
vg∂M = 0, where ψ := ϕ|∂M , then (3.30)
 the assumption H ≥ 0 will provide the inequality. Denoting by
together with
(· , ·)∂M := ∂M · , ·
vg∂M one has
 
2(D2 ψ, ψ)∂M = {Id + ν·}D2 ψ, {Id + ν·}ψ
 ∂M

= π<β ({Id + ν·}D2 ψ), π<β ({Id + ν·}ψ)
∂M
(1.23)
 
= D2 (π<β ({Id − ν·}ψ)), π<β ({Id + ν·}ψ) ,
∂M

where we used the notations of Section 1.5. But π≥β (ψ + ν · ψ) = 0 implies

(1.23)
π<β (ψ − ν · ψ) = π<β ψ − ν · π>−β ψ
= π≤−β ψ − π[β,−β] ψ − ν · π>−β ψ
= ν · π≥β ψ − π[β,−β] ψ − ν · π>−β ψ
= ν · π[β,−β] ψ − π[β,−β] ψ,
4.4 Case of the mgAPS boundary condition 75

in particular π<β (ψ − ν · ψ) = 0, which implies (D2 ψ, ψ)∂M = 0 and (4.4).


If (4.4) is an equality, then H = 0 and ϕ must be a α-Killing spinor on (M n , g)
with α := − nλ ∈ R. Moreover (1.22) with H = 0 and for the restriction ψ on
∂M of an α-Killing spinor ϕ on M implies

D2 (ψ + ν · ψ) = (n − 1)α(ψ + ν · ψ). (4.6)

Since ψ + ν · ψ = 0 (the eigenvalues of the pointwise Clifford action by ν are


±i), the mgAPS boundary condition then requires (n − 1)α < β. Conversely,
if (M n , g) has minimal boundary and carries a non-zero α-Killing spinor ϕ
β
with real α < n−1 , then on the one hand Proposition A.4.1 implies that
(M , g) is Einstein with scalar curvature 4n(n − 1)α2 > 0, on the other hand
n

Dϕ = −nαϕ on M . Furthermore the identity (4.6) holds on ∂M so that ϕ


solves (4.5), hence −nα is an eigenvalue of D w.r.t. the mgAPS boundary
condition. This shows the characterization of the limiting-case of (4.4) and
concludes the proof. 
Chapter 5
Upper eigenvalue bounds
on closed manifolds

In this chapter we turn to the very different game of looking for upper eigen-
value bounds for the fundamental Dirac operator. We concentrate on two
methods for obtaining them. The first method, due to C. Vafa and E. Witten
[230], consists in comparing D to another Dirac-type operator D of which
kernel is non trivial for index-theoretical reasons, then in estimating the zero-
order difference D −D by geometric quantities. As a result, those bound from
above a topologically determined number of eigenvalues of D. This method
was applied to prove Theorems 5.1.1, 5.2.1 and 5.2.2. The second method
relies on the min-max principle, which is a general variational principle char-
acterizing eigenvalues of self-adjoint elliptic operators, see e.g. [74, pp.16-17]:

Lemma 5.0.2 (min-max principle) Let (M n , g) be a closed Riemannian


spin manifold. Order the eigenvalues of D2 (which are exactly the squares
of the eigenvalues of D) into a nondecreasing sequence λ1 (D2 ) ≤ . . . ≤
λk (D2 ) ≤ λk+1 (D2 ) ≤ . . .. Then, for every k ≥ 1,


2 M
D2 ϕ, ϕ
vg
λk (D ) = min max ,
Ek ϕ∈Ek
M
|ϕ|2 vg
ϕ=0

where the minimum runs over all k-dimensional vector subspaces Ek of


Γ(ΣM ). In particular one has, for every non-zero ϕ ∈ Γ(ΣM ),

D2 ϕ, ϕ
vg
M
λ1 (D ) ≤
2
, (5.1)
M
|ϕ|2 vg

with equality if and only if D2 ϕ = λ1 (D2 )ϕ.

Applying this method means choosing ϕ such that the r.h.s.



of (5.1) can
M
D 2 ϕ,ϕvg
be computed in terms of geometric quantities. The quotient |ϕ|2 vg
is
M
2
called the Rayleigh quotient of D evaluated at ϕ.

N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 77


DOI 10.1007/978-3-642-01570-0 5, 
c Springer-Verlag Berlin Heidelberg 2009
78 5 Upper eigenvalue bounds on closed manifolds

5.1 Intrinsic upper bounds

The following theorem, which can be formulated under weaker assumptions


(see [51, Prop. 1]), is an application of Vafa-Witten’s method. The upper
bound which is derived only depends on the sectional curvature of M .

Theorem 5.1.1 (H. Baum [51]) Let (M n , g) be even-dimensional, closed


with positive sectional curvature K M . Then the first eigenvalue λ1 of D
satisfies
n 
|λ1 | ≤ 2 2 −1
n
· max(K M ).
2 M

The inequality in Theorem 5.1.1 is sharp for M = S2 , however it is not


clear if it can be sharp for any other manifold. If n is odd one can apply
Theorem 5.1.1 to the product of the manifold with a circle. In that case
H. Baum proved in [51, Cor. 2] an analogous estimate, however under
the supplementary assumptions that K M is pinched and M is simply-
connected. Moreover Theorem 5.1.1 improves a qualitative result by J. Lott
[181, Prop. 4] valid on S2 .

Sketch of proof of Theorem 5.1.1: Construct a suitable map ι of degree 1


from M into a round sphere of suitable radius and such that the derivative
of this map does not deviate too far from the identity. This goes as follows:
fix a point p ∈ M and take for ι the composition of the exponential map
of the (n-dimensional) round sphere of radius √ 1M with the inverse of
K (p)
the exponential map of M at p. Of course the exponential map of M at p
is only invertible on its injectivity domain; furthermore one has to control
the behaviour of the exponential far from p by introducing a smoothing
function in the definition of ι. Actually that smoothing function may be
chosen such that ι can be extended by a constant to a smooth function on
the whole
√ of M . Then the norm of the derivative of ι can be estimated
against K M , which together with Theorem 5.2.1 below leads to the result. 

Another way to obtain upper bounds consists in comparing the Dirac spectra
for different metrics and applying the min-max principle. The remarkable
property of conformal covariance of the Dirac operator allows this method to
work. The first result is this direction is due to J. Lott [181, Prop. 3].

Theorem 5.1.2 (J. Lott [181]) Let (M n , g) be an n(≥ 2)-dimensional


closed spin manifold. Then for any conformal class of Riemannian metrics
[g] on M n , there exists a finite positive constant b([g]) such that

λ1 (Dg2 ) ≤ b([g]) sup(−Sg ) (5.2)


M

for any g ∈ [g] with scalar curvature Sg < 0.


5.1 Intrinsic upper bounds 79

Proof : Fix g0 ∈ [g] with Sg0 < 0 on M n . For some u ∈ C ∞ (M n , R), let
g = e2u g0 ∈ [g] with Sg < 0. As in Proposition 1.3.10, we denote by ϕ → ϕ
the unitary isomorphism Σg0 M −→ Σg M . Choose ψ0 to be a non-zero eigen-
vector for Dg0 associated to the eigenvalue λ1 (Dg0 ) and set ψ := e− 2 u ψ0 .
n−1

The min-max principle together with (1.16) imply



|Dg ψ|2 vg
λ1 (Dg2 ) 
≤ M
M
|ψ|2 vg

e−(n+1)u |Dg0 ψ0 |2 enu vg0
= 
M
e−(n−1)u |ψ0 |2 enu vg0
M

supM (|ψ0 |2 ) M e−u vg0
≤ λ1 (Dg0 ) ·
2
·  . (5.3)
inf M (|ψ0 |2 ) M
eu vg0

We now estimate the quotient of integrals on the r.h.s. with the help of (3.19).
Namely (3.19) reads

Sg eu = Sg0 e−u + 2(n − 1)e−u Δg0 u − (n − 1)(n − 2)e−u |gradg0 (u)|2g0 ,

so that integrating one obtains


 
Sg0 e−u vg0 = Sg eu vg0
M M

+(n − 1) −2e−u Δg0 u + (n − 2)e−u |gradg0 (u)|2g0 vg0
 M

= Sg e vg0 + n(n − 1)
u
e−u |gradg0 (u)|2g0 vg0 ,
M M

where we have used Δg0 (e−u ) = −e−u Δg0 u − e−u |gradg0 (u)|2g0 . We deduce
that
 
e−u vg0 supM (−Sg ) M −Sg0 e−u vg0
M ≤ · 
eu vg0 inf M (−Sg0 ) −Sg eu vg0
M M

supM (−Sg )  e−u |gradg0 (u)|2g0 vg0 
= 1 + n(n − 1) M 
inf M (−Sg0 ) M g
S eu vg0
supM (−Sg )
≤ ,
inf M (−Sg0 )

which together with (5.3) gives the result. 

On surfaces, Lott’s estimate (5.2) only applies if the genus is at least 2. For
genus 0 or 1 the use of (3.19) can be avoided through the fact that eigenvectors
associated to the lowest Dirac eigenvalue have constant length:
80 5 Upper eigenvalue bounds on closed manifolds

Theorem 5.1.3 (I. Agricola and T. Friedrich [2]) Let M 2 := S2 or T2


with arbitrary Riemannian metric g. Let u ∈ C ∞ (M 2 , R) be such that g0 :=
e−2u g has constant curvature.
i) The smallest eigenvalue of Dg2 satisfies

1
λ1 (Dg2 )Area(M 2 , g) ≤ λ1 (Dg20 )Area(M 2 , g0 ) + |gradg0 (u)|2g0 vg0 . (5.4)
4 M2

ii) The smallest eigenvalue of Dg2 satisfies


 −u
2 e vg
λ1 (Dg2 ) ≤ λ1 (Dg20 ) M u 0 . (5.5)
M2
e vg 0

iii) The smallest positive eigenvalue of DT22 ,g on T2 endowed with trivial spin
structure satisfies

T2
e−3u {λ2 (DT22 ,g0 ) + |gradg0 (u)|2g0 }vg0
λ2 (DT22 ,g ) ≤  . (5.6)
T2
e−u vg0

Proof : Let ψ0 be a non-zero eigenvector for DM,g0 associated to the eigenvalue


λ1 (DM,g0 ) in i), ii) and to the eigenvalue λ2 (DM,g0 ) in iii) respectively. If
M = S2 then the spinor ψ0 is a real Killing spinor, hence has constant
length on S2 . If M = T2 , the formula (2.1) implies that all eigenvectors of
D have constant length on T2 . Therefore we may assume that |ψ0 | = 1 on
M 2 . As in Proposition 1.3.10, we denote by ϕ → ϕ the unitary isomorphism
Σg0 M −→ Σg M .
i) Set ψ := ψ0 . Identity (1.16) gives

1
DM,g ψ = e−u (DM,g0 ψ0 + gradg0 (u) · ψ0 )
2
−u 1
= e (λ1 (DM,g0 )ψ + gradg0 (u) · ψ0 ),
2
where we have denoted by “·” the Clifford multiplication on Σg0 M . We deduce
that
 1
|DM,g ψ|2 = e−2u λ1 (DM,g
2
)|ψ0 |2 + |gradg0 (u) · ψ0 |2
0
4 
+ λ1 (DM,g0 )e( ψ0 , gradg0 (u) · ψ0

 1 
= e−2u 2
λ1 (DM,g ) + |gradg0 (u)|2g0 .
0
4

The min-max principle together with vg = e2u vg0 provides


5.1 Intrinsic upper bounds 81

|DM,g ψ|2 vg
M2
2
λ1 (DM,g )≤
M2
|ψ|2 vg

M2
{λ1 (DM,g
2
0
) + 14 |gradg0 (u)|2g0 }vg0
= ,
Area(M 2 , g)

which leads to (5.4).


ii) Set ψ := e− 2 ψ0 . Then identity (1.16) implies that
u

DM,g ψ = e−
3u
2 DM,g0 ψ0
= λ1 (DM,g0 )e−
3u
2 ψ0 ,

so that |DM,g ψ|2 = λ1 (DM,g


2
0
)e−3u and, by the min-max principle,
 
M2
|DM,g ψ|2 vg e−u vg0
2
λ1 (DM,g ) ≤ = λ 1 (D 2
) M2
,
M2
|ψ|2 vg M,g0
M2
eu vg0

which is (5.5).
iii) Set ψ := e− 2 ψ0 . On the one hand,
3u

e− 2
3u
(1.16)
e−u (DT2 ,g0 (e− 2 ψ0 ) +
3u
DT2 ,g ψ = gradg0 (u) · ψ0 )
2
(1.11) −u
 3 3u
− e− 2 gradg0 (u) · ψ0 + e− 2 DT2 ,g0 ψ0
3u
= e
2
e− 2
3u 
+ gradg0 (u) · ψ0
2
= e−u · e− 2 (λ2 (DT2 ,g0 )ψ0 − gradg0 (u) · ψ0 ).
3u

On the other hand, ψ is L2 -orthogonal to the kernel of DT2 ,g , which is spanned


by e− 2 σ ± , where σ ± are non-zero parallel spinors on (T2 , g0 ) with σ ± ∈
u

Γ(Σ± T2 ): indeed ψ0 is L2 -orthogonal to σ ± , so that


 
−u
ψ, e 2 σ ±
vg = e−2u ψ0 , σ ±
vg
T2 T2

= ψ0 , σ ±
vg0
T2
= 0.

We deduce that

T2
|DT2 ,g ψ|2 vg
λ2 (DT22 ,g ) ≤
T2
|ψ|2 vg
82 5 Upper eigenvalue bounds on closed manifolds

T2
e−2u · e−3u {λ2 (DT22 ,g0 ) + |gradg0 (u)|2g0 }vg
= 
T2
e−u vg0

T2
e−3u {λ2 (DT22 ,g0 ) + |gradg0 (u)|2g0 }vg0
=  ,
T2
e−u vg0

which shows (5.6) and concludes the proof. 

As a consequence of (5.4), if M 2 = S2 , then [2, Thm. 2]

λ1 (DS22 ,g )Area(S2 , g) − 4π ≤

1 
inf 2 |gradg0 (uΦ )|2g0 vg0 , Φ∗ g0 = e−2uΦ g , (5.7)
Φ∈Conf(S ,[g0 ]) 4 S2

where Conf(S2 , [g0 ]) denotes the group of conformal transformations of


(S2 , g0 ). Note that the l.h.s. of (5.7) is non-negative because of (3.17).
Beware that the r.h.s. of (5.7) is not a conformal invariant, since the product
λ1 (DS22 ,g )Area(S2 , g) is not bounded in the conformal class of g [24, Thm.
1.1]. As an application of (5.7), the smallest eigenvalue of D2 of a one-
parameter-family of ellipsoids can be asymptotically estimated: given a > 0,
x2
let Ma := {x ∈ R3 | x21 + x22 + a23 = 1} carry the induced metric from R3 , then
[2, Thm. 5]

3 1
2
lim λ1 (DM ) ∈ [2, + ln(2)] and 2
lim λ1 (DM )≤ . (5.8)
a→0 a
2 a→∞ a
4

Both estimates provide much sharper upper bounds as C. Bär’s one (5.19) in
terms of the averaged total squared mean curvature.
In the case of the 2-torus and with the notations of Theorem 2.1.1, the
smallest eigenvalue of DT22 ,g w.r.t. the (δ1 , δ2 )-spin structure and flat metric is
not greater than 4π 2 | 12 (δ1 γ1∗ + δ2 γ2∗ )|2 , so that by (5.5) it satisfies [2, Thm. 4]
 −u
2 e vg
λ1 (DT22 ,g )Area(T2 , g) ≤π 2
|δ1 γ1∗ + δ2 γ2∗ |2 T u 0 . (5.9)
T2
e vg 0

Of course (5.9) is empty if δ1 = δ2 = 0, in which case only (5.6) gives


information on the first positive eigenvalue. Besides, inequality (5.9) provides
asymptotical estimates of the smallest eigenvalue of D2 on round tori: given
0 < r < R, let T2r,R denote the tube of radius r about a circle of radius R. If
T2r,R carries the induced metric and the (1, 0)-spin structure, then [2, p.5]

lim λ1 (DT2 2 )Area(T2r,R ) = lim λ1 (DT2 2 )Area(T2r,R ) = 0


r→0 r,R R→∞ r,R
5.1 Intrinsic upper bounds 83

and in case it carries the (0, 1)-spin structure,

lim λ1 (DT2 2 )Area(T2r,R ) ≤ π 2 .


R →1
r r,R

For both the (1, 0)- and (0, 1)-spin structures (however not for the (1, 1)-one)
these estimates enhance those obtained from (5.19) below.
In higher dimensions another general upper bound in terms of the sectional
curvature can be obtained from the min-max principle. In the following the-
orem we denote by Br (p) the geodesic ball of radius r > 0 around some point
p ∈ M and, provided r is smaller than the injectivity radius radinj (M n , g) of
the Riemannian manifold (M n , g), by 0 < μ1 (Br (p)) ≤ μ2 (Br (p)) ≤ . . . the
spectrum of the scalar Laplace operator with Dirichlet boundary condition
on Br (p). For any x ≥ 0 and t > 0 we also define
⎧ √
⎪ 1−cos( xt)
⎨ √x sin(√xt) if x > 0
fx (t) :=

⎩ t
2 if x = 0.

Theorem 5.1.4 (C. Bär [36]) Let (M n , g) be an n(≥ 2)-dimensional


closed Riemannian spin manifold and assume that there exist nonnegative real
constants ρ1 , ρ2 , κ such that the sectional curvature of (M n , g) lies everywhere
n
in [κ − ρ1 , κ + ρ2 ]. Let N := 2[ 2 ] . Then the jN th eigenvalue in absolute value
of the Dirac operator D of (M n , g) satisfies


n κ
|λjN (D)| ≤ + inf inf μj (Br (p))
2 p∈M r

2 n−2
+ (n − 1)(1 + [ ])(ρ1 + ρ2 )fκ+ρ2 (r) , (5.10)
3 2

where the second infimum ranges over r ∈ ]0, min{radinj (M n , g), √κ+ρ π
2
}[.
Moreover, (5.10) is an equality for (M n , g) = (Sn , can), j = 1 and ρ1 = ρ2 = 0.
Sketch of proof of Theorem 5.1.4: First consider a geodesic ball Br (p) of
radius 0 < r < radinj (M n , g). Since Br (p) is convex its spinor bundle can
be trivialized by a pointwise orthonormal family ϕ1 , . . . , ϕN . Let {fj }j≥1
be a Hilbert basis of L2 (Br (p), C) made out of eigenfunctions for the scalar
Laplace operator Δ with Dirichlet boundary condition on Br (p), in particular
Δfj = μj (Br (p))fj with fj |∂Br (p) = 0 holds. Then {fj ϕk }1≤j,1≤k≤N is a
Hilbert basis of L2 (ΣBr (p)). Since DB
2
r (p)
is of Laplace type (see Schrödinger-
Lichnerowicz’ formula (3.2)) one may talk about its eigenvalues with respect
to the Dirichlet boundary condition and the min-max principle also applies.
84 5 Upper eigenvalue bounds on closed manifolds

Fix 1 ≤ k ≤ N and j ≥ 1, then using the formal self-adjointness of the Dirac


operator we have
2
(DB r (p)
(fj ϕk ), fj ϕk ) DBr (p) (fj ϕk )2
=
fj ϕk 2 fj 2
(1.11)
= dfj · ϕk + fj DBr (p) ϕk 2
 
= dfj 2 + 2e dfj · ϕk , fj DBr (p) ϕk
+fj DBr (p) ϕk 2
≤ μj (Br (p)) + 2dfj  · fj DBr (p) ϕk 
+ sup (|DBr (p) ϕk |2 )
Br (p)

≤ μj (Br (p)) + 2 μj (Br (p)) · sup (|DBr (p) ϕk |)
Br (p)

+ sup (|DBr (p) ϕk | )2


Br (p)
 2
= μj (Br (p)) + sup (|DBr (p) ϕk |) ,
Br (p)

therefore the min-max principle implies


 2
2
λjN (DB r (p)
)≤ μj (Br (p)) + sup (|DBr (p) ϕk |) .
Br (p)

The second step in the proof, which is the main and the most technical
one, consists in estimating the supremum on the r.h.s. by geometric data.
This can be done essentially by controlling the growth of the pointwise norm
along geodesics and applying Rauch’s comparison theorem, we refer to [36,
Lemma 1] and [36, Sec. 4]. The third and last step consists in comparing
the eigenvalues of D2 with those of DB 2
r (p)
subject to Dirichlet boundary
condition: the monotonicity principle (see e.g. [74, Cor. 1 p.18]) implies that

λj (D2 ) ≤ λj (DBr (p) )2 .

This proves the inequality (5.10).


If (M n , g) = (Sn , can), then on the one hand |λ1 (DSn )| = . . . = |λN
(DSn )| = n2 (see Theorem 2.1.3) and on the other hand the choice ρ1 = ρ2 = 0
* +
provides inf inf μ1 (Br (p)) in the r.h.s. of (5.10). Since μ1 (Br (p)) −→ 0
p∈M r r→π
(see e.g. [74, Thm. 6 p.50]), we conclude that equality holds in (5.10). 
5.2 Extrinsic upper bounds 85

5.2 Extrinsic upper bounds

In this section we assume the existence of some map from the manifold M
to another manifold and want to derive upper eigenvalue estimates in terms
of geometric invariants associated to this map. The first situation which has
been studied is the case where there exists a map of sufficiently high degree
from M into the round sphere of same dimension.

Theorem 5.2.1 (H. Baum [51]) Let (M n , g) be an even-dimensional


closed Riemannian spin manifold and assume the existence of a smooth map
ι : M n −→ Sn with degree


k−1
2 −1
n
deg(ι) ≥ 1 + 2 mj
j=1

for some positive integer k, where mj is the multiplicity of the jth eigenvalue
of D2 . Then the kth eigenvalue λk of D satisfies

n
|λk | ≤ 2 2 −1
n
· max dx ι.
2 x∈M

The inequality in Theorem 5.2.1 is sharp for M = S2 and ι = Id (but of


course only for k = 1). It is unclear if it can be sharp in higher dimensions.

Sketch of proof of Theorem 5.2.1: The proof is based on Vafa-Witten’s


method. Consider the tensor-product bundle S := ΣM ⊗ ι∗ (ΣSn ). Define
the tensor-product connection of ∇ (on ΣM ) with on the one hand the
pull-back-connection ι∗ (∇ΣS ) and on the other hand with a flat connection
n

coming from a trivialization of ΣSn through ± 12 -Killing spinors, see Example


A.1.3.2. One obtains two different covariant derivatives on S to which two
different Dirac-type operators (called twisted Dirac operators) may be
associated. The latter one (i.e., involving the flat connection on ι∗ (ΣSn )) is
n
by construction just the direct sum of 2[ 2 ] = rk(ΣSn ) copies of D. Applying
the Atiyah-Singer index theorem for twisted Dirac operators and computing
explicitly the Chern character of both positive and negative half-spinor
bundles of Sn (remember that n is assumed to be even) the dimension of the
kernel of the other twisted Dirac operator can be bounded from below by a

positive constant depending on the A-genus of T M and the degree of ι. A
further observation shows that this lower bound may be made dependent of
deg(ι) only. The assumption on deg(ι) plugged into Vafa-Witten’s method
then ensures that |λk | is not greater than the norm of the difference
of both Dirac operators, which can be itself easily estimated against the
supremum norm of dι. This concludes the sketch of proof of Theorem 5.2.1. 
86 5 Upper eigenvalue bounds on closed manifolds

Turning to the case where M is isometrically immersed into some Euclidean


space, upper eigenvalue bounds can be found in terms of the second funda-
mental form of the immersion. The first result in this direction is due to U.
Bunke.
Theorem 5.2.2 (U. Bunke [66]) Assume that (M n , g) is even-
dimensional, closed, and that there exists an isometric immersion
ι : M −→ RN for some positive integer N . Then there is a topologi-
cally determined number of eigenvalues λ of D satisfying
n
λ2 ≤ 2 2 max(IIx 2 ),
x∈M

where II is the second fundamental form of ι.


It is not clear whether the estimate in Theorem 5.2.2 can be sharp or not,
since it is a strict inequality even for the standard immersion of Sn in Rn+1 . It
has been improved by C. Bär in [40] (see (5.19)). His idea consists in choosing
the restriction of particular spinor fields (such as parallel or Killing spinors)
onto the hypersurface as test-spinors in view of the min-max principle. We
formulate a general statement, from which all known estimates à la Reilly
will follow. The notion of twistor-spinor is explained in Appendix A.
Theorem 5.2.3 Let (M n , g) be closed, n ≥ 2 and assume that there exists
an isometric immersion M −→ M
ι n+1 , where (M n+1 , g) admits a non-zero
twistor-spinor ψ. If the spin structure on M coincides with the one induced
by ι, then the smallest eigenvalue of D2 satisfies
 n2  (H 2 + R(ι))f 2 |ψ|2 v 
|df |2 |ψ|2 vg 
M  g
λ1 (D ) ≤
2
inf + M , (5.11)
f ∈C ∞ (M,R) 4 M f 2 |ψ|2 vg M
f 2 |ψ|2 vg
f =0

 is the mean curvature of ι and


where H := − n1 tr(∇ν)

1  
R(ι) := S − 2ric(ν,
 ν) .
n(n − 1)

Proof : As in Section 1.4 we denote by “·” the Clifford multiplication on


n+1 , g) and by “ · ” the one on (M n , g). In analogy with Proposition 1.4.1
(M
M
we also denote by D2 the operator

D if n is even
D2 := 
D ⊕ −D if n is odd

where D is the fundamental Dirac operator of (M n , g).


Since M is a Riemannian hypersurface of M , the operator D2 can be
2
related with the square of another Dirac operator, namely the Dirac-Witten
5.2 Extrinsic upper bounds 87

operator introduced by E. Witten in his proof of the positive mass theorem


[235] and defined by
n
:=
D e
ej · ∇ j
j=1

| ), then
in any local orthonormal basis {ej }1≤j≤n of T M : let ϕ ∈ Γ(ΣM M


n  n
D2 ϕ
(1.21)
= e ϕ − 1
ej · ν · ∇ ej · ν · A(ej ) · ν · ϕ
j
j=1
2 j=1

= + nH ϕ,
−ν · Dϕ (5.12)
2
so that
(1.23)
+ D2 ( nH
D22 ϕ = ν · D2 (Dϕ) ϕ)
2
(1.11)
+n
= ν · D2 (Dϕ) (grad(H) · ν · ϕ + HD2 ϕ)
2
(5.12)
= (D + n grad(H) · ν · ϕ + nH D2 ϕ
+ nH ν·)Dϕ
2 2 2
2 2
2ϕ + n H n
= D ϕ + grad(H) · ν · ϕ. (5.13)
4 2
n+1 , g) one has, in any local orthonormal
Since ψ is a twistor-spinor on (M
basis {ej }1≤j≤n of T M ,

n

D
= D(  e ψ)
ej · ∇ j
j=1

1 
n
= − 
D(ej · ej · Dψ)
n + 1 j=1
n 
= D(Dψ)
n+1
n  1 
n
(A.4)
= j ) · ψ + S ej · ej · ψ)
(− ej · Ric(e
n − 1 j=1 2 4n

n  S 1 S 
= ψ + ν · Ric(ν) ·ψ− ψ
n−1 2 2 4
n 1  1
 ν))ψ + ν · Ric(ν) T·ψ

= (S − 2ric(ν,
n−1 4 2
n  n(n − 1) 1 T 
= R(ι)ψ + ν · Ric(ν) · ψ
n−1 4 2
2
n n T · ψ,
= R(ι)ψ + ν · Ric(ν) (5.14)
4 2(n − 1)
88 5 Upper eigenvalue bounds on closed manifolds

T := n ric(ν,
where we denoted by Ric(ν)  ej )ej the tangential projection
j=1

of Ric(ν). Combining (1.13) (which obviously holds for D22 instead of D2 ),
(5.13) and (5.14) we obtain, for every f ∈ C ∞ (M, R),

2 2
2 ψ + n H ψ + n grad(H) · ν · ψ)
D22 (f ψ) = f (D
4 2
grad(f ) ψ − A(grad(f ))
−2(∇ · ν · ψ) + (Δf )ψ
2
n2 2 nf nf T·ψ
= (H + R(ι))f ψ + grad(H) · ν · ψ + ν · Ric(ν)
4 2 2(n − 1)
2
+ grad(f ) · DM
 ψ + A(grad(f )) · ν · ψ + (Δf )ψ. (5.15)
n+1

We deduce that
n2 2 2f
e( D22 (f ψ), f ψ
) = (H + R(ι))f 2 |ψ|2 + e( grad(f ) · DM  ψ, ψ
)
4 n+1
+f (Δf )|ψ|2
n2 2
= (H + R(ι))f 2 |ψ|2 − g(f grad(f ), grad(|ψ|2 ))
4
+f (Δf )|ψ|2 ,

so that, integrating over M and applying Green’s formula,


 
n2
D22 (f ψ), f ψ
vg = (H 2 + R(ι))f 2 |ψ|2 vg
4 M
M
 
− 2
δ(f grad(f ))|ψ| vg + f (Δf )|ψ|2 vg
M M

n2
= (H 2 + R(ι))f 2 |ψ|2 vg
4 M

+ |grad(f )|2 |ψ|2 vg . (5.16)
M

The result straightforward follows from the min-max principle. 

n+1 , g) admits an (m ≥ 2)-dimensional space of twistor-


In particular, if (M
spinors then
λ1 (D2 ) ≤ inf {r.h.s. of (5.11)}.
ψ twistor-spinor
ψ=0

The case n = 1 - the “baby case” - should not be of interest since the
spectrum of S1 for both spin structures is explicitly known (see Theorem
 = R2 or S2 -
2.1.1). However similar results turn out to hold - at least for M
and to follow from very elementary geometric properties of plane or space
curves:
5.2 Extrinsic upper bounds 89

Proposition 5.2.4 Let M be a closed regular curve in (M 2 , g) := (R2 , can)


2 2
or (S , can). Then the smallest eigenvalue λ1 (D ) of the square of the Dirac
operator on M for the induced metric and spin structure satisfies
 L
1
λ1 (D2 ) ≤ (H(t)2 + κ)dt
4L 0

where H is the curvature of M parametrized by arc-length in M, L :=


2
Length(M ) and κ denotes the sectional curvature of (M , g).
Moreover this inequality is an equality if and only if M is a simply-
.
parametrized-circle in M

Proof : First we may assume that M := c([0, L]), where c : R −→ M  is an L-


periodic arc-length-parametrized curve in M. In other words M is isometric
to S1 (L) := {z ∈ C s.t. |z| = 2π
L
}. Therefore the smallest eigenvalue of D2
w.r.t. the δ-spin structure (where δ = 0 for the trivial spin structure and 1
2
for the non-trivial one, see Example 1.4.3.1) is λ1 (D2 ) = δπ
L2 . We separate
the two cases.
• Case κ = 0: Let nc ∈ Z be the turning number of c. It is easy to show
that the induced spin structure of M in R2 is the trivial one in case nc is
even and the non-trivial one otherwise. If nc is even then the inequality is
trivial and cannot be an equality, so that we assume that nc is odd. From
the elementary formula
 L
1
nc = H(t)dt
2π 0
and the Cauchy-Schwarz inequality we deduce that

π2
λ1 (D2 ) =
L2
π2
≤ 2 n2c
L
 L
π2 1
= 2 · 2( H(t)dt)2
L 4π 0
 L
1
≤ · L · H 2 (t)dt
4L2 0
 L
1
= H 2 (t)dt,
4L 0

which proves the inequality. The equality holds if and only if H is constant
and |nc | = 1, hence we obtain the statement in that case.
• Case κ = 1: Let μc be the bridge number and H  be the curvature of c as
 2 2
space curve. From H = H + 1 and
90 5 Upper eigenvalue bounds on closed manifolds
 L
1 
μc ≤ H(t)dt
2π 0

we obtain exactly as above

π2 2
λ1 (D2 ) ≤ μ
L2 c
 L
1 
≤ ( H(t)dt) 2
4L2 0
 L
1  2 (t)dt
≤ H
4L 0
 L
1
= (H 2 (t) + 1)dt,
4L 0

which shows the inequality. As before the equality only holds if H is constant
and μc = 1, hence we obtain the statement in that case and conclude the
proof. 

Corollary 5.2.5 Under the assumptions of Theorem 5.2.3, the smallest


eigenvalue λ1 of D2 satisfies:

n2 (H 2 + R(ι))|ψ|2 vg
λ1 (D2 ) ≤ M  . (5.17)
4 M |ψ|2 vg

If furthermore ψx = 0 for every x ∈ M , then [97, Thm. 3.1 p.44]


 
n2 1
λ1 (D ) ≤
2 2
(H + R(ι))vg + |d(ln(|ψ|))|2 vg . (5.18)
4Vol(M ) M Vol(M ) M

Proof : Both results follow directly from Theorem 5.2.3: considering the
expression inside the infimum of the r.h.s. of inequality (5.11) one just has
1
to choose f to be a non-zero constant in the first case and to be |ψ| in the
second one. 

Corollary 5.2.5 provides the following estimates which were proved by C. Bär
[40, Cor. 4.2 & 4.3] for hypersurfaces of the Euclidean space Rn+1 or of the
round sphere Sn+1 and by N. Ginoux [97, 99, 98] for hypersurfaces of the
hyperbolic space Hn+1 : let Mn+1 (κ) := Rn+1 if κ = 0, Sn+1 if κ = 1 and
Hn+1 if κ = −1, then for any closed hypersurface M of M n+1 (κ) carrying
the induced metric and spin structure,
5.2 Extrinsic upper bounds 91

n2
λ1 (D2 ) ≤ (H 2 + κ)vg (5.19)
4Vol(M ) M

if κ ≥ 0,
n2
λ1 (D2 ) ≤ (max (H 2 ) − 1) (5.20)
4 M
and
  
n2 1
λ1 (D ) ≤
2
(H − 1)vg +
2
inf |d(ln(|ψ|))|2 vg
4Vol(M ) M Vol(M ) ψ∈ K ± i M
2
ψ=0

(5.21)
if κ = −1 (the space K  i refers here to the space of ± i -Killing spinors on
±2 2√
Hn+1 ). Indeed M n+1 (κ) carries at least one non-zero ± κ -Killing spinor ψ
2
(which is in particular a twistor-spinor and vanishes nowhere), see e.g. [56, 63]
and Examples A.1.3. Moreover R(ι) = κ and, if κ ≥ 0, then Proposition A.4.1
implies that |ψ| is constant on M n+1 (κ).
The inequalities (5.19)√
and (5.20) actually hold for higher eigenvalues of
D2 , since the space of ± 2κ -Killing spinors on Mn+1 (κ) is 2[ n+1
2 ] -dimensional

and the upper bound in (5.19) or (5.20) does not depend on the Killing spinor
ψ, see [40, 99].
The inequalities (5.19), (5.20) and (5.21) are equalities for all geodesic
hyperspheres in M n+1 (κ), see [40], [99] and [98] respectively. For κ ≤ 0 the
question remains open whether those are the only hypersurfaces enjoying this
property. For κ = 1, generalized Clifford tori in Sn+1 as well as minimally
embedded S /Q8 (where Q8 denotes the finite group of quaternions) in S4
3

also satisfy the limiting-case in (5.19) and it is conjectured that this actually
holds for every homogeneous hypersurface in the round sphere, see [100, 103].
For n = 2 the upper bound in (5.19) is nothing but the so-called Willmore
functional of the immersion M → M 3 (κ). Combining the estimate (5.19)
with lower bounds of the Dirac spectrum (see Section  3.6) B. Ammann [9]
and C. Bär [40] proved the Willmore conjecture (“ M H 2 vg ≥ 2π 2 for every
embedded torus M in R3 ”) for particular metrics.
Note 5.2.6 Could (5.18), (5.20) and (5.21) be enhanced in

n2
λ1 (D2 ) ≤ (H 2 + R(ι))vg ? (5.22)
4Vol(M ) M

They are already of that form as soon as there exists a twistor-spinor ψ on


n+1 , g) with constant norm on M (in (5.18) or (5.21)) or if the mean
(M
curvature H of ι is constant (in (5.20)) respectively. Both conditions are
very strong; for example the level sets of the norm of a non-zero imaginary
Killing spinor on Hn+1 are either geodesic hyperspheres or horospheres. Since
however (5.19) is already of the form (5.22) and since an analog of (5.22) for
92 5 Upper eigenvalue bounds on closed manifolds

the smallest positive Laplace-eigenvalue holds in virtue of a work by A. El


Soufi and S. Ilias (see reference in [98]), it is natural to ask if (5.22) could
hold in full generality, i.e., for every n ≥ 2, for any isometric immersion
Mn → M n+1 into any Riemannian spin manifold (M n+1 , g) admitting a
non-zero twistor-spinor. This  is unfortunately false, at least in dimension
n = 2: indeed the integral M 2 (H 2 + R(ι))vg - the Willmore functional -
is a conformal invariant, i.e., it only depends on the conformal class of g.
2
However, the product λ1 (DM,g )Area(M 2 , g) is in general not bounded on a
conformal class (take e.g. M = S2 ) [24].
In dimension n ≥ 3 one could look for a sequence {fn }n∈N of smooth real-
1
valued functions on M which would converge in L2 -norm towards f := |ψ|
L2
(provided the twistor-spinor ψ has no zero on M ) and such that dfn −→ 0 as
n → +∞. In that case {fn }n∈N would be bounded in the Sobolev space
H1,2 (M ), so that there would exist a subsequence {fϕ(n) }n∈N converging
weakly in H1,2 (M ) towards some F ∈ H1,2 (M ). In particular {fϕ(n) }n∈N
would converge weakly towards F in L2 (M ) as well as {dfϕ(n) }n∈N towards
dF . From the uniqueness of the weak limit one would conclude that F = f
and dF = 0 almost everywhere, in particular df = 0 almost everywhere, that
is, df = 0, which does not hold in general as explained just above.
Nevertheless, it is not excluded in case M is a closed hypersurface of the
hyperbolic space Hn+1 that

n2
λ1 (D ) ≤
2
(H 2 − 1)vg
4Vol(M ) M

holds.
Chapter 6
Prescription of eigenvalues
on closed manifolds

From its definition the Dirac spectrum a priori depends on the metric, the spin
structure and of course the underlying manifold. In a very formal manner,
the Dirac spectrum can be thought of as a functor from the category of
closed Riemannian spin manifolds to that of real discrete sequences with
closed image and unbounded on both sides. In this chapter, we investigate
this functor, in particular its injectivity and surjectivity. In other words, does
its Dirac spectrum determine a given Riemannian spin manifold? For a given
real sequence as above, is there a Riemannian spin manifold whose spectrum
coincides with this sequence? If the answer to the former question is definitely
negative (Section 6.1), only partial results have been found regarding the
latter in the case where the whole spectrum is replaced by a finite set of
eigenvalues. In this situation one has to distinguish between the eigenvalue
0 - whose associated eigenvectors are called harmonic spinors - and the other
ones. We shall see in Section 6.2 that, in dimension n ≥ 3 (the case of
surfaces must be handled separately), the metric can in general be modified
so as to make 0 a Dirac eigenvalue, whereas generic metrics just have as many
harmonic spinors as the Atiyah-Singer-index theorem forces them to do. If
the finite set of real numbers does not contain 0 (and is symmetric about 0 if
n ≡ 3 (4)), then it is always the lower part of the Dirac spectrum of a given
metric on a fixed spin manifold (Section 6.3).

6.1 Dirac isospectrality

The question we address in this section is: do closed Riemannian spin


manifolds which are Dirac isospectral (i.e., which show the same Dirac spec-
trum) have to be isometric, and if not are they at least diffeomorphic or
homeomorphic? Note here that we a priori have to require the isometry con-
dition to take the spin structure into account, that is, isometries are supposed
to preserve both the orientation and the spin-structure.
From Weyl’s asymptotic formula (see e.g. [18, Thm. 2.6]), the Dirac
spectrum determines the dimension and the volume of the underlying

N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 93


DOI 10.1007/978-3-642-01570-0 6, 
c Springer-Verlag Berlin Heidelberg 2009
94 6 Prescription of eigenvalues on closed manifolds

manifold. A sharper insight in the formula shows for example that, in di-
mension 4, the Euler characteristic of the manifold is determined by its Dirac
spectrum as soon as the scalar curvature is assumed to be constant [54]. How-
ever, the Dirac spectrum in general determines neither the isometry class nor
the topology of the manifold. To illustrate this, we describe different fami-
lies of examples, evolving from the “simplest” to the most sophisticated ones
where even the explicit knowledge of the Dirac spectrum is not needed.
The founding result for isospectrality issues is indisputably J. Milnor’s
famous one-page-long article (see reference in [15]), where the author de-
scribes two Laplace-isospectral but non-isometric 16-dimensional tori. The
idea is the following. The spectrum of the scalar Laplace operator on a flat
torus Γ\R is {4π 2 |γ ∗ |2 , γ ∗ ∈ Γ∗ }, where we keep the notations of Theorem
n

2.1.1. For any r ≥ 0, the multiplicity of the eigenvalue 4π 2 r2 is the number


of γ ∗ ∈ Γ∗ with |γ ∗ | = r. Now it is a surprising fact that there exist two
lattices Γ1 , Γ2 in R16 which induce non-isometric metrics on T16 but such
that, for every r ≥ 0, the sets {γ ∗ ∈ Γ∗1 , |γ ∗ | = r} and {γ ∗ ∈ Γ∗2 , |γ ∗ | = r}
have the same cardinality (see reference in Milnor’s article). Therefore the
flat tori Γ1 \R and Γ2 \R possess the same Laplace spectrum, however they
16 16

are not isometric. From Theorem 2.1.1, the Dirac spectrum of a flat torus
with trivial spin structure is {±2π|γ ∗ |, γ ∗ ∈ Γ∗ }, hence the same argument
shows that the Dirac spectra of Γ1 \R and Γ2 \R also coincide, at least for
16 16

the trivial spin structure.


On positively curved spaceforms, Theorem 2.1.4 straightforward results in
the following criterion for producing isospectrality.

Theorem 6.1.1 (C. Bär [38]) For n ≥ 3 odd let Γ1 , Γ2 ⊂ SOn+1 be finite
subgroups acting freely on Sn . For j = 1, 2 let j : Γj −→ Spinn+1 be a group
homomorphism such that ξ ◦j is the inclusion map Γj ⊂ SOn+1 and consider
the induced spin structure on Γj \S . Assume the existence of a bijective map
n

f : Γ1 −→ Γ2 such that, for every γ1 ∈ Γ1 , the elements 1 (γ1 ) and 2 (f (γ1 ))


are conjugated in Spinn+1 .
Then the spaceforms Γ1 \S and Γ2 \S are Dirac isospectral.
n n

Proof : Both the character as well as the determinant remain unchanged


under conjugation, hence Theorem 2.1.4 implies that the corresponding
series F± (z) giving the multiplicities of the Dirac eigenvalues ±( n2 + k) are
the same for Γ1 \S and Γ2 \S .
n n


For three positive integers a, b, r with a, b odd, we denote by Γ(a, b, r)


the abstract group generated by two elements A, B satisfying the relations
Aa = B b = 1 and BAB −1 = Ar . It is an exercise to prove that, if Γ(a, b, r) is
embedded in SOn+1 so as to act freely on Sn , then the corresponding space-
form has a unique spin structure [38, Lemma 7]. A more detailed study shows
that, for a good choice of a, b, r and n (as in Corollary 6.1.2 below), the group
Γ(a, b, r) can be embedded in two different ways in SOn+1 so as to satisfy the
6.1 Dirac isospectrality 95

assumptions of Theorem 6.1.1 but such that the corresponding spaceforms


are not isometric [38, Sec. 5].

Corollary 6.1.2 (C. Bär [38]) Let n ≡ 3 (8), n ≥ 19. Let a be a prime
number with a ≡ 1 ( n+1 n+1 2
4 ), let b := ( 4 ) and let r be chosen such that its
n+1 ×
mod a class is of order 4 in (Za ) . Then there exist two Dirac isospectral
non-isometric spaceforms diffeomorphic to Γ(a, b, r)\S .
n

Returning to the flat setting, Bieberbach manifolds (quotients of Rn


through Bieberbach groups, i.e., discrete co-compact and freely acting sub-
groups of the isometry group of Rn ) provide a large family of manifolds
where the Dirac spectrum can be theoretically computed - and hence the
isospectrality question be answered. As already mentioned in Section 2.1, the
complexity of Bieberbach groups in n dimensions grows up abruptly with n,
which makes the computation of the Dirac spectrum tedious. For particular
holonomies it remains possible and whole families of Dirac isospectral exam-
ples can be obtained. As an interesting fact, even pairwise non-homeomorphic
ones can be fished out. To state the result we need to introduce the following
notations, which are those of [188]. Fix an integer n ≥ 3. For non-negative
integers j, h with n − 2j − h > 0 and j + h ∈ 2N \ {0}, define in the canonical
basis {e1 , . . . , en } of Rn the orthogonal involution Bj,h by
 0 1
0 1
 
Bj,h := diag ,..., , −1, . . . , −1, 1, . . . , 1
1 0 1 0   
   h
j

and set Γj,h := Span(Bj,h · ( e2n + Id), e1 + Id, . . . , en + Id). The subgroup
Γj,h of the isometry group of Rn is orientation-preserving and co-compact,
therefore the quotient
:= Γj,h \R
n n
Mj,h
is a compact orientable flat manifold with holonomy group Z2 . Moreover it
n
can be shown that Mj,h is spin and that its first homology group with integer
coefficients is H1 (Mj,h ) = Zn−j−h ⊕ (Z2 )h [187, Prop. 4.1]. In particular, the
n

quotients corresponding to different pairs (j, h) as above are not homeomor-


phic to each other. Set, for fixed k ∈ {0, 1},

j+h
Fk+ := {Mj,h
n
| ≡ k (2)}.
2

Among the 2n−j spin structures on some fixed Mj,h n


∈ Fk+ , we shall only
consider the two so-called εk -ones, which are defined in [187, Prop. 4.2] and
[188, eq. (4.13)] and which can be roughly described as follows with the
help of Proposition 1.4.2: lift the translations e1 + Id, . . . , en−1 + Id to 1 ∈
Spinn and Bj,h ∈ SOn to one of its both pre-images through the double
96 6 Prescription of eigenvalues on closed manifolds

covering ξ. This assignment fixes exactly two spin structures since with those
assumptions the lift of en + Id must be (−1)k , see [187, Prop. 4.2] for details.
Theorem 6.1.3 (R. Miatello and R. Podestá [188]) For a fixed inte-
ger n ≥ 3 consider the Mj,h
n
’s defined above.
i) All elements of the family {(Mj,h n
, ε0 ) | Mj,h ∈ F0+ } are pairwise Dirac
isospectral closed flat Riemannian spin manifolds, which are pairwise non-
homeomorphic as soon as the pairs (j, h) are different.
ii) If n ≡ 3 (4) then all elements of the family {(Mj,hn
, ε1 ) | Mj,h
n
∈ F1+ } are
pairwise Dirac isospectral closed flat Riemannian spin manifolds, which
are pairwise non-homeomorphic as soon as the pairs (j, h) are different.

In [188] Theorem 6.1.3 comes as a corollary of a whole series of computa-


tions of spectra on the Mj,h ’s, one of which motivation is to compare Dirac
isospectrality with other isospectrality issues. The results are obtained in a
much more general setting where one twists the spinor bundle by a vector
bundle associated to some representation of the group Γj,h . We refer to [188]
for further statements and the proof of Theorem 6.1.3.
In another direction, one can try to generalize Milnor’s result, replacing
flat tori by compact quotients of nilpotent Lie groups. Initiated for Laplace
isospectrality issues (see references in [15]), this ansatz has turned out to
provide very rich families of manifolds on which, among others, Dirac isospec-
trality can be tested. In that case there also exists a general criterion for
isospectrality, see [15, Thm. 5.1] for a proof.

Theorem 6.1.4 (B. Ammann and C. Bär [15]) Let M := Γ\G and
M  := Γ \G be spin homogeneous spaces where Γ, Γ are co-compact lattices
in the simply-connected Lie group G. If the right-representations of G onto
L2 (ΣM ) and L2 (ΣM  ) are equivalent, then for any left-invariant metric on
G, the manifolds M and M  are Dirac isospectral.
The independence of the spectrum on the left-invariant metric is a very
strong statement. It allows in particular to produce continuous families of
isospectral metrics, the spin structures staying fixed. Of course the real
difficulty consists in applying Theorem 6.1.4, i.e., in picking groups G so
that the equivalence of the G-representations is satisfied. For nilpotent Lie
groups that are strictly non-singular (i.e., for every z in the center of G
and x in its complement, there exists a y with xyx−1 y −1 = z), this condition
simplifies in terms of group theoretical data [15, Thm. 5.3]. The nilpotent Lie
groups chosen by R. Gornet provide concrete examples where this criterion
is fulfilled, we refer to [15, Thm. 5.6] for details and references.

Theorem 6.1.5 (B. Ammann and C. Bär [15]) There exist in dimen-
sions 7 and 8 a continuous family of Dirac isospectral non-isometric closed
Riemannian spin manifolds. Each manifold inside a family is diffeomorphic
to the same quotient of some nilpotent Lie group by a co-compact lattice.
6.2 Harmonic spinors 97

6.2 Harmonic spinors

In analogy with the differential-form-setting, a (smooth) spinor field is called


harmonic if and only if it lies in the kernel of the Dirac operator. If the
manifold is closed then this space has always finite dimension since each
eigenspace does. For instance, if (M n , g) has positive scalar curvature, then
Theorem 3.1.1 implies that there is no non-zero harmonic spinor on M , as we
have seen in Section 3.1. The problem we investigate here is: does this dimen-
sion depend on the metric, and if so, how? Before going further on note that
it clearly depends on the choice of spin structure as the elementary example
M = Tn with flat metric already shows: the space of harmonic spinors w.r.t.
n
the trivial spin structure is 2[ 2 ] -dimensional whereas it is reduced to 0 for all
other ones, see Theorem 2.1.1.
The first fundamental remark on the dependence of the Dirac kernel in
terms of the metric was formulated by N. Hitchin in [148]: the dimension d of
the kernel of D stays constant under conformal changes. Indeed, using (1.16),
if ϕ is a harmonic spinor on (M n , g) then so is e− 2 u ϕ on (M n , g := e2u g)
n−1

for any u ∈ C ∞ (M, R). It can for example be deduced from this fact combined
with Theorem 2.1.3 that, whatever the metric chosen, the 2-sphere S2 does
not carry any non-zero harmonic spinor (there exists only one conformal class
as well as one spin structure on S2 ). Alternatively this straightforward follows
from Bär’s inequality (3.17) since the lower bound 2πχ(M 2 ) is a topological
invariant which is positive for M 2 = S2 .
On closed surfaces the presence of a quaternionic structure on the spinor
bundle, which commutes with the Dirac operator (see e.g. [101, Lemma 1]),
forces the number d to be even. On the other hand, d is bounded from above
independently of the metric and spin structure: N. Hitchin proved that d ≤
2 ] for every closed Riemann surface M of genus g and, if g ≤ 2 then
2[ g+1 2

d does not even depend on the conformal class [148, Prop. 2.3]. The case of
S2 has just been discussed. For the 2-dimensional torus T2 it can be seen as
a consequence of the conformal property above and of Theorem 2.1.1, which
implies that T2 has a 2-dimensional space of harmonic spinors for the trivial
spin structure and no non-zero one otherwise. If g = 2 the independence
on the conformal class follows from [148, Prop. 2.3] or, alternatively, from
the following argument: the Atiyah-Singer-index theorem implies that d2 ≡
α(M ) (2), where α(M ) is the α-genus of M , see (6.1) below. Either α(M ) = 1
and it follows from Hitchin’s upper bound that d = 2 as soon as g ≤ 4, or
α(M ) = 0 and, for g ≤ 2, one has d = 0. For g = 3, 4 and α(M ) = 0 both
possibilities d = 0 and d = 4 occur [20].
In higher genus the picture is more complex. As a consequence of [19, Thm.
1.1], for any given spin structure, there exists a conformal class for which d
coincides with the lower bound provided by the Atiyah-Singer-index theorem,
hence for which d ∈ {0, 2}. Next one can ask whether d can be made maximal.
To answer this question, it is more convenient to study the variations of d in
terms of the spin structure, the conformal class being fixed. Recall first that,
98 6 Prescription of eigenvalues on closed manifolds

for any closed surface M 2 of genus g, the group H 1 (M, Z2 ) is isomorphic to


(Z2 )2g , so that M 2 has exactly 22g spin structures. Assuming now the metric
on M 2 to be hyperelliptic (i.e., that there exists an isometric involution of
M 2 with exactly 2g + 2 fixed points), C. Bär and P. Schmutz Schaller [48]
have computed d explicitly for each spin structure and shown the following:
on the one hand Hitchin’s upper bound for d is attained for exactly 2g+2 spin
structures
if g is even and 1 if g is odd respectively, on the other hand there
2g + 1
exist spin structures for which d = 0. Independently H. Martens
g
showed [186] that, if d = 2[ g+1 2
2 ], then either M is hyperelliptic or g = 4 or
g = 6. For both g = 4 and g = 6 and any non-hyperelliptic conformal class,
there exists exactly one spin structure such that d = 2[ g+1 2 ], see [148, 48].
In genus g ≥ 4 it can be deduced from those results the existence of a spin
structure for which d depends on the conformal class, however it is not clear
whether this holds for every spin structure or not.
We summarise the main known results for orientable surfaces.
Theorem 6.2.1 Let M 2 be a closed oriented surface of genus g and d be
the complex dimension of the space of harmonic spinors on M 2 for some
conformal class.
i) The integer d is even and satisfies d ≤ 2[ g+1 2 ]. In particular, for any
metric, the 2-sphere S2 has no non-zero harmonic spinor.
ii) There exists at least one spin structure and one conformal class on M 2
for which d = 2[ g+1
2 ].
iii) There exists at least one spin structure and one conformal class on M 2
for which d = 0.
iv) If g ≤ 2 then d does not depend on the conformal class.
v) If g ≥ 3 then there exists a spin structure on M 2 for which d depends on
the conformal class.
vi) If g ∈ {3, 4} then M 2 has a spin structure for which d depends on the
conformal class and a spin structure for which it does not.
By contrast, in higher dimensions, if one fixes the manifold and the spin
structure, a given metric admits in general few non-zero harmonic spinors
whereas there exist particular metrics having lots of them. First note that
the Atiyah-Singer-index theorem [28] provides an a priori lower bound for d,
since the inequality dim(Ker(D)) ≥ |ind(D+ )| in even dimensions combined
with Theorem 1.3.9 implies that

⎪ )| if n ≡ 0 (4)
|A(M







⎨ |α(M )| if n ≡ 1 (8)
d = dim(Ker(D)) ≥ (6.1)



⎪ 2|α(M )| if n ≡ 2 (8)





0 otherwise,
6.2 Harmonic spinors 99

where α(M ) ∈ Z2 is the α-genus of M (see e.g. [45, Sec. 3] for a definition).
It was first conjectured by C. Bär [39] that (6.1) is an equality for generic
metrics, i.e., metrics belonging to some subset which is open in the C 1 - and
dense in the C ∞ -topology in the space of all Riemannian metrics. In dimen-
sions n = 3 and 4 perturbation methods combined with the formula linking
spinors for different metrics [62] suffice in order to prove the conjecture to
hold true (S. Maier [185]). If the underlying manifold is assumed to be simply-
connected, then C. Bär and M. Dahl [45] proved the conjecture to hold true
for all dimensions n ≥ 5. Based on bordism theory, their argument consists of
the three following steps. First, the conjecture holds true for some generators
of the spin bordism ring. Second, any closed spin manifold is spin bordant to
a spin manifold where the conjecture holds true. By a theorem of Gromov and
Lawson, any n(≥ 5)-dimensional closed simply-connected manifold which is
spin bordant to a spin manifold can be obtained from it by surgeries of codi-
mension at least 3. It remains in the last step to show that the conjecture
survives to surgeries of codimension at least 3. A set of generators is given by
spin manifolds admitting metrics of positive scalar curvature (for which we
already know that Ker(D) = 0) as well as products of some of the irreducible
manifolds of Mc.K. Wang’s classification (see Theorem A.4.2), on which it
can be relatively easily shown that the conjecture holds true. The second step
is a pure argument of bordism theory using a theorem by S. Stolz (see [45]
for references) and the explicit set of generators described above. The crucial
step is the last one, which can be deduced from the following theorem.
Theorem 6.2.2 (C. Bär and M. Dahl [45]) Let (M n , g) be a closed
Riemannian spin manifold. Let M  be obtained from M by surgery of codi-
mension at least 3. Let ε > 0 and L > 0 with ±L ∈ / Spec(D).

Then there exists a Riemannian metric g̃ on M such that the Dirac eigen-
n , g̃) in ] − L, L[ differ at most by ε.
values of (M n , g) and (M
Coming back to the conjecture, it has been proved by B. Ammann, M. Dahl
and E. Humbert [19] to hold true in full generality. Their argument relies on
a very fine generalization of the surgery theorem (Theorem 6.2.2) to codi-
mension greater than 1, we refer to [19] for a detailed proof.
At what seems to be the opposite side there have appeared since
N. Hitchin’s pioneering article [148] several results showing the existence in
certain dimensions n ≥ 3 of metrics with lots of non-zero harmonic spinors.
Already computing the Dirac spectrum on S3 with Berger metric (and canon-
ical spin structure), N. Hitchin noticed [148] that, for every N ∈ N, there
exists a metric on S3 admitting at least N linearly independent harmonic
spinors. Furthermore he constructed with the help of differential topological
methods metrics with non-zero harmonic spinors on all closed spin mani-
folds M n of dimension n ≡ 0, 1, 7 (8). Extending the computation of the
Dirac spectrum to all Berger spheres C. Bär showed [39] the existence of
such metrics on all closed spin manifolds M n with n ≡ 3, 7 (8). His proof
is based on the following simple ideas: first, for any closed odd-dimensional
100 6 Prescription of eigenvalues on closed manifolds

Riemannian spin manifolds M1 and M2 , their connected sum M1 M2 admits


a Riemannian metric for which its Dirac spectrum gets close to the union of
both Dirac spectra of M1 and M2 ; second, there exists, for any n ≡ 3 (4),
a one-parameter-family of Riemannian metrics on Sn for which at least one
eigenvalue crosses the zero line (put t = 2(m + 1), a1 = a2 = 0 and j = m−1 2
in one of both eigenvalues given in Theorem 2.2.2.iii)). We refer to [39] for
a detailed proof. As for the remaining dimensions n ≥ 3, the question of
existence of metrics with non-zero harmonic spinors is still open, although it
has been conjectured by C. Bär [39] that such metrics always exist. For the
sphere Sn , n ≥ 4, the conjecture has been proved to hold true by L. Seeger
[221] in case n = 2m ≥ 4 and by M. Dahl [80] for all n ≥ 5. The latter work,
which contains the results of C. Bär [39] and N. Hitchin [148] on the existence
of metrics with harmonic spinors on all closed spin manifolds of dimension
n ≡ 0, 1, 3, 7 (8) (n ≥ 3), is proved in the following way: one shows that,
on a given closed spin manifold, the space of Riemannian metrics for which
the Dirac operator is invertible is disconnected, if non-empty. The argument
involves special metrics with positive scalar curvature on Sn which do not
bound any metric with positive scalar curvature on the unit ball B n+1 of
Rn+1 . We refer to [80, Sec. 3] for a detailed proof.
We summarise the main known results for closed n(≥ 3)-dimensional spin
manifolds.

Theorem 6.2.3 Let M n be a closed n(≥ 2)-dimensional spin manifold and


d(g) be the complex dimension of the space of harmonic spinors on M n for
some metric g. Then the following holds.
i) There exists a subset in the space of all Riemannian metrics on M n , which
is open in the C 1 - and dense in the C ∞ -topology, such that d(g) coin-
cides with the lower bound given by (6.1) for every metric g in this subset
(B. Ammann, M. Dahl and E. Humbert [19]).
ii) If n ≡ 0, 1, 3, 7 (8) and n ≥ 3 then there exists a Riemannian metric g on
M n such that d(g) ≥ 1.
It would be interesting to know whether the following stronger conjecture by
C. Bär [39, p. 41] holds true: on every closed spin manifold of dimension ≥ 3
there exists a sequence {gm }m of Riemannian metrics for which d(gm ) ≥ m
for every m ∈ N.

6.3 Prescribing the lower part of the spectrum

In the last section we have seen that, if n ≥ 3 and n ≡ 0, 1, 3, 7 (8), then any
n-dimensional closed spin manifold admits a metric for which the kernel of the
corresponding Dirac operator is non-trivial. In other words, the eigenvalue 0
can be always prescribed in those dimensions. What about prescribing the
6.3 Prescribing the lower part of the spectrum 101

rest of the spectrum? Although this question remains open, at least the lower
part of the spectrum can be fixed. Note that, in dimension n ≡ 3 (4), Theorem
1.3.7.iv) imposes a priori this part to be symmetric about the origin.

Theorem 6.3.1 (M. Dahl [79]) For n ≥ 3 let M n be any closed spin man-
ifold. Let L > 0 (resp. m ≥ 1) be real (resp. integral) and l1 , . . . , lm be
non-zero real numbers. Then the following holds:
i) If n ≡ 3 (4) and −L < l1 < . . . < lm < L, then there exists a Riemannian
metric on M n such that Spec(D, g) ∩ (] − L, L[\{0}) = {l1 , . . . , lm }, where
each of those eigenvalues is simple.
ii) If n ≡ 3 (4) and 0 < l1 < . . . < lm < L, then there exists a Riemannian
metric on M n such that Spec(D, g) ∩ (] − L, L[\{0}) = {±l1 , . . . , ±lm },
where each of those eigenvalues is simple.

The proof of Theorem 6.3.1 relies on techniques similar to those used for
the construction of metrics with harmonic spinors. Roughly speaking, after
possibly rescaling a fixed given metric on M , one has to add sufficiently
many spheres of different sizes, each having one of the li ’s as single and
simple eigenvalue in ] − L, L[. The surgery theorem (Theorem 6.2.2) ensures
that the resulting manifold has the desired eigenvalues modulo a small error.
We refer to [79, Sec. 4] for a detailed proof.
Chapter 7
The Dirac spectrum on non-compact
manifolds

In this chapter we investigate the less familiar situation where the underlying
manifold is non-compact. A good but somewhat not up-to-date reference is
[42, Sec. 8.2]. The Dirac operator of a non-compact Riemannian spin man-
ifold has in general no well-defined spectrum, even if its square does as a
non-negative operator (Section 7.1). If the Riemannian manifold is complete
then its Dirac spectrum - which is well-defined - is composed of a discrete part
and of a disjoint union of intervals. After giving examples where the Dirac
spectrum can be explicitly computed (Section 7.2), we survey the different
situations where the geometry or particular analytical properties of the un-
derlying manifold produce either gaps in their Dirac spectrum (Section 7.3)
or the non-existence of one of both spectral components (Section 7.4).

7.1 Essential and point spectrum

We have seen in Proposition 1.3.5 that the Dirac operator is essentially self-
adjoint as soon as the underlying Riemannian spin manifold is complete. This
provides the existence of a canonical self-adjoint extension of D, which makes
its spectral theory somewhat easier, see below. In case the manifold is not
complete, there exists no canonical such extension as we have already seen in
Note 1.3.6. However, since the square of D is symmetric and non-negative on
its domain Γc (ΣM ) = {ϕ ∈ Γ(ΣM ) | supp(ϕ) compact}, it admits a canonical
non-negative self-adjoint extension, as was noticed by C. Bär in [44, Sec. 2]:

Theorem 7.1.1 (Friedrichs’ extension) Let (M n , g) be any Riemannian


spin manifold. Then the operator D2 has a unique non-negative self-adjoint
extension in HD 1
(ΣM ) ⊂ L2 (ΣM ), where HD
1
(ΣM ) is the completion of
Γc (ΣM ) w.r.t. the Hermitian inner product (ϕ, ψ) → (ϕ, ψ) + (Dϕ, Dψ).
It is called the Friedrichs’ extension of D.

For the general construction of the Friedrichs’ extension we refer to [233,


Thm. VII.2.11]. Thus one may consider the spectrum of the Friedrichs’

N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 103


DOI 10.1007/978-3-642-01570-0 7, 
c Springer-Verlag Berlin Heidelberg 2009
104 7 The Dirac spectrum on non-compact manifolds

extension of D2 , which we also denote by σ(D2 ). In general the spectrum


of an operator may be decomposed as follows:

Definition 7.1.2 Let T be a densely defined (unbounded) operator in a


Hilbert space H.
i) The point spectrum of T is the set σp (T ) := {λ ∈ C | Ker(λId−T ) = {0}}.
ii) The continuous spectrum of T is the set σc (T ) := {λ ∈ C | Ker(λId−T ) =
{0}, Im(λId − T ) = H and (λId − T )−1 unbounded}.
iii) The residual spectrum of T is the set σr (T ) := {λ ∈ C | Ker(λId − T ) =
{0} and Im(λId − T ) = H}.
iv) The essential spectrum of T is the set σe (T ) := {λ ∈ C | λId − T
not Fredholm}.
v) The discrete spectrum of T is the set σd (T ) := σp (T ) \ σe (T ).
, ,
In particular σ(T ) = σp (T ) σc (T ) σr (T ). The point spectrum is the set
of eigenvalues and the discrete spectrum the subset of eigenvalues with finite
multiplicity. Any self-adjoint operator T has real spectrum and no residual
spectrum. Furthermore, σc (T ) = σe (T ) \ σp (T ) and σe (T ) may be charac-
terized as the set of λ’s for which an orthonormal sequence (ϕk )k exists
satisfying T ϕk − λϕk  −→ 0. The essential spectrum of a self-adjoint el-
k→∞
liptic differential operator remains unchanged after modifying any compact
part of the underlying manifold [41, Prop. 1]:
Proposition 7.1.3 (decomposition principle) Let (M1 , g1 ) and (M2 , g2 )
be Riemannian spin manifolds. Assume the existence of a spin-structure-
preserving isometry M1 \K1 −→ M2 \K2 for some compact subsets Kj ⊂ Mj ,
j = 1, 2. Then DM1 ,g1 and DM2 ,g2 have the same essential spectrum.
Note that, as a consequence of Theorem 1.3.7, the Dirac operator has no
essential spectrum as soon as the underlying manifold is compact.

7.2 Explicit computations of spectra

We first determine the spectrum of the Euclidean space explicitly.

Theorem 7.2.1 The Dirac operator on (Rn , can) has no point spectrum and
its continuous spectrum is R.

Proof : We adapt [42, Sec. 8.2.2] to the case n ≥ 1. If ϕ ∈ C ∞ (Rn , Σn ) were an


eigenvector of D associated to the eigenvalue λ, then it would be in particular
an eigenvector of D2 = Δ (acting on Σn ∼
n
= CN , where N = 2[ 2 ] ) associated
2
to the eigenvalue λ . But there is no non-zero square-integrable eigenfunction
for the scalar Laplacian on Rn as a consequence of a result by F. Rellich, see
reference in [153]. Therefore σp (D) = ∅.
7.2 Explicit computations of spectra 105

Let now λ ∈ Rn be an arbitrary vector. For n = 1 and λ = 0 let ϕλ, :=


1 ∈ C = Σ1 and  := −sgn(λ). For λ = 0 let ϕλ, be any unit-length-element
in Σn and put  := 0. For n ≥ 2 and λ = 0 let ϕλ, be any unit-length-
λ
elements in Ker(i |λ| · −Id) for  = ±1 (see proof of Theorem 2.1.1). Pick

χ ∈ C (R , [0, 1]) with χ(x) = 1 whenever |x| ≤ 1 and χ(x) = 0 whenever
n

|x| ≥ 2. For j ∈ N \ {0} set


x
φλj, (x) := χ( + 3je1 ) · ei λ,x ϕλ,
j

for all x ∈ Rn , where e1 denotes the first canonical basis vector of Rn . By


construction supp(φλj, ) ∩ supp(φλj , ) = ∅ whenever j = j  , so that (φλj, )j≥1
φλ
forms an orthogonal system in L2 (Rn , Σn ). In particular setting ψj,
λ
:= j,
φλ
j, 
λ
one obtains an orthonormal system (ψj, )j≥1 in L2 (Rn , Σn ). Next we show
that (D − |λ|Id)ψj,  −→ 0, which will show |λ| ∈ σe (D) and conclude
λ
j→∞
the proof.
It follows from (1.11) that

1  x
(D − |λ|Id)ψj,
λ
= χ( + 3je1 ) · (D − |λ|)(ei λ,x ϕλ, )
φλj,  j
1 x 
+ grad(χ)( + ke1 ) · ei λ,x ϕλ, ,
j j

with


n
∂ i λ,x
D(ei λ,x ϕλ, ) = el · (e ϕλ, )
∂xl
l=1

n
= iλl el · (ei λ,x ϕλ, )
l=1

= iλ · (ei λ,x ϕλ, )


= |λ|(ei λ,x ϕλ, ),

so that
1 x
(D − |λ|Id)ψj,
λ
= grad(χ)( + ke1 ) · ei λ,x ϕλ, .
jφλj,  j
n
A simple transformation formula provides φλj,  = j 2 χL2 , from which

   12
1 x
(D − |λ|Id)ψj,
λ
= n |grad(χ)( + ke1 )|2 dx1 . . . dxn
j · j 2 χL2 Rn j
106 7 The Dirac spectrum on non-compact manifolds

1 n
= n · j 2 grad(χ)L2
j · j χL2
2

grad(χ)L2
=
jχL2

and the result follow. 

The second family where the spectrum can be explicitly described is that of
hyperbolic spaces, whose Dirac spectra are given in the following table (note
however that 0 is never a Dirac eigenvalue on RHn as claimed in [65, Cor.
4.6] for n even):

M σp (D) σc (D) references


RH n
∅ R [65], [29]
CHn , n odd ∅ R [73]
CHn , n even {0} ] − ∞, − 12 ] ∪ [ 12 , ∞[ [73]
HHn ∅ R [73]
OH2 ∅ R [73]

As a generalization, S. Goette and U. Semmelmann have shown [111] that


the point spectrum of D on a Riemannian symmetric space of non-compact
type is either empty or {0}, and in the latter case each irreducible factor
of M is of the form U(p + q)/U(p) × U(q) with p + q odd. This explains in
particular why the even complex dimensional hyperbolic space stands out as
the only one in the list above having non-empty point spectrum.

7.3 Lower bounds on the spectrum

In general not much can be said on both components of the Dirac spectrum.
However, as in the compact case, a spectral gap about 0 occurs as soon as
the scalar curvature is bounded below by a positive constant, even if the
underlying manifold is not complete:

Theorem 7.3.1 (C. Bär [44]) Let (M n , g) be any n(≥ 2)-dimensional


Riemannian spin manifold, then
n
min(σ(D2 )) ≥ inf (S), (7.1)
4(n − 1) M

where S is the scalar curvature of (M n , g).


Proof : Recall that σ(D2 ) denotes the spectrum of the Friedrichs’ extension of
D2 . As in the compact setting, min(σ(D2 )) can be characterized as follows,
see e.g. [233]:
7.3 Lower bounds on the spectrum 107

(D2 ϕ, ϕ)
min(σ(D2 )) = inf .
ϕ∈Γc (ΣM )\{0} ϕ2

For every ϕ ∈ Γc (ΣM ) the identity (D2 ϕ, ϕ) = Dϕ2 holds (Proposition


1.3.4), moreover (3.3) is valid on every Riemannian spin manifold provided
ϕ is smooth and has compact support, therefore inequality (7.1) is satisfied. 

As a consequence, if furthermore (M n , g) is complete with scalar curvature


uniformly bounded below  constant S0 , then the spectrum of D
by a positive
satisfies σ(D) ⊂ ]−∞, − 4(n−1) ]∪[ 4(n−1)
nS0 nS0
, ∞[, see [44, Cor. 3.2]. In analo-
n
gy with the compact setting, the equality λ2 = 4(n−1) inf M (S) > 0 for some
eigenvalue λ of D on the complete Riemannian manifold (M n , g) implies the
existence of a non-zero real non-parallel Killing spinor on (M n , g) [44, Thm.
3.4], in particular the manifold must be Einstein and closed (Proposition
A.4.1). Here one should beware that it is a priori not clear whether ∇ϕ is
finite or not for a non-zero eigenvector ϕ of D. Nevertheless this can be proved
with the help of standard functional analytical techniques [44, Lemma 3.3].
The finiteness of ∇ϕ implies in turn that of P ϕ (see (A.11)), therefore
(3.3) makes sense for ϕ being an eigenvector of D and the statement follows
as in the compact setting.
Combining Theorem 7.3.1 with the decomposition principle (Proposition
7.1.3) provides a positive lower bound on the essential spectrum of D as soon
as the scalar curvature is bounded below “at infinity”: if (M n , g) is again
arbitrary and contains a compact subset K for which there exists a positive
constant S0 with inf M \K (S) ≥ S0 , then min(σe (D2 )) ≥ 4(n−1)
n
S0 [44, Thm.
4.1]. In particular, the Dirac operator on (M n , g) has no essential spectrum
as soon as the scalar curvature explodes at infinity [44, Cor. 4.3]. This is
a particular situation where one of both components of the Dirac spectrum
does not appear, see Section 7.4 for further results.
A natural question arising from Theorem 7.3.1 is whether conformal lower
bounds for the Dirac spectrum can be obtained as in the compact setting.
There is no complete answer to that question. In the case of surfaces an
analog of Bär’s inequality (3.17) holds, at least when the surface has finite
area and can be embedded into S2 so as to inherit its spin structure:
Theorem 7.3.2 (C. Bär [44]) Let (M 2 , g) be a connected surface of finite
area embedded into S2 and carrying the induced spin structure. Then

min(σ(D2 )) ≥ .
Area(M 2 , g)

Theorem 7.3.2 follows from (3.17) using the above-mentioned characterization


of min(σ(D2 )) and a suitable sequence of metrics so as to make the area of
M 2 close to that of S2 .
In higher dimensions there exists no general analog of Hijazi’s inequality
(3.18), since for example the spectrum of the conformal Laplace operator on
108 7 The Dirac spectrum on non-compact manifolds

the real hyperbolic space RHn for n ≥ 3 is [ n−1


n−2 , ∞[. However the correspond-
ing inequality can be obtained under additional geometric assumptions:

Theorem 7.3.3 (N. Große [115]) Let (M n , g) be any n(≥ 3)-dimensional


complete Riemannian spin manifold with finite volume and let λ ∈ σ(D).
Assume that either λ ∈ σd (D) or λ ∈ σe (D) and n ≥ 5 as well as S ≥ S0 for
some S0 ∈ R. Then
n
λ2 ≥ min(σ(L)), (7.2)
4(n − 1)
where L := 4(n−1)
n−2 Δ + S is the conformal Laplace operator and S is the scalar
curvature of (M n , g).
The two situations according to λ ∈ σd (D) or λ ∈ σe (D) require different
approaches. In the case where λ is a eigenvalue of D, the proof consists
in adapting that of Corollary 3.3.2 using a conformal factor given by the
length of the corresponding eigenvector and cutting off near its zero-set and
at infinity. The second case, where the supplementary assumptions n ≥ 5 and
S ≥ S0 are needed for technical reasons, relies on a Kato-type inequality [115,
Lemma 2.1], we refer to [115, Sec. 4] for details. Moreover, as for inequality
(7.1), equality in (7.2) for some non-zero eigenvalue λ implies the existence of
a non-zero real non-parallel Killing spinor on (M n , g), in particular (M n , g)
must be Einstein and closed [115, Thm. 1.1].
Interestingly enough, the analog of the so-called conformal Hijazi inequal-
ity (3.20) also turns out to hold under suitable geometric assumptions. Given
an n-dimensional manifold M with Riemannian metric g and spin structure
, set

Dϕ2
λ+
1 (M, g, ) := ϕ∈Γinf ,
c (ΣM ) (Dϕ, ϕ)
(Dϕ,ϕ)>0

where D := Dg . The corresponding conformal invariant is defined by


1
λ+ 1 (M, g, ) · Vol(M, g) ,
λ+
n
min (M , [g], ) := inf n
g∈[g]
Vol(M,g)<∞

where [g] denotes a Riemannian conformal class on M .

Theorem 7.3.4 (N. Große [115]) Let M n be any n(≥ 3)-dimensional


manifold with Riemannian conformal class [g] and spin structure . Assume
the existence of a complete metric with finite volume in [g] as well as
λ+ n
min (M , [g], ) > 0. Then

n 2
min (M , [g], ) ≥
λ+ min(σ(Lg )) · Vol(M, g) n ,
n 2
inf (7.3)
4(n − 1) g∈[g]
Vol(M,g)<∞

where Lg denotes the conformal Laplace operator on (M n , g).


7.4 Absence of a spectral component 109

The assumption λ+ min (M , [g], ) > 0 actually implies σe (Dg ) ⊂] − ∞, 0] for


n

every complete g ∈ [g] with finite volume [115, Lemma 3.3]. As an example,
if (N n−1 , h) is any closed Riemannian spin manifold with positive scalar
curvature and n ≥ 5, then the Riemannian product (N × R, h ⊕ dt2 ) endowed
with the product spin structure satisfies the assumptions of Theorem 7.3.4,
therefore (7.3) holds [115, Ex. 4.1], where the r.h.s. can be shown to be
positive (see reference in [115]). However the cases n = 3, 4 remain open.

7.4 Absence of a spectral component

There are particular situations where one of both components of the Dirac
spectrum can be excluded out of geometric considerations. This kind of ques-
tion has attracted a lot of attention in the last years. We choose to present
here five different settings with sometimes non-empty mutual intersection.
Two of them deal with manifolds with cusps. A manifold with cusps can
be written as the disjoint union of a compact manifold with non-empty
boundary together with cusps, which are Riemannian manifolds of the form
(]0, ∞[×N, dt2 ⊕gt ) for some smooth 1-parameter-family of Riemannian met-
rics on the manifold N .
First consider an oriented complete n-dimensional hyperbolic manifold
(recall that a metric is called hyperbolic if it has constant sectional curvature
−1). Assume it to have finite volume. Then M can be shown to possess a
finite number of cusps of the form (]0, ∞[×N n−1 , dt2 ⊕e−2t gflat ) for some flat
metric on some closed manifold N n−1 (see reference in [41]). If M is spin,
then any spin structure on M induces a spin structure on each cusp and
hence on each N n−1 -factor. We call the spin structure trivial along a cusp
if the Dirac operator of the corresponding N has non-zero kernel and non-
trivial otherwise. Since by the decomposition principle (Proposition 7.1.3) the
essential spectrum of D is unaffected by perturbations on a compact subset,
it can be only influenced by the geometry of the cusps. As a striking fact, it
turns out to depend only on the spin structure on M , where one obtains the
following dichotomy as shown by C. Bär [41, Thm. 1]:

Theorem 7.4.1 (C. Bär [41]) Let (M n , g) be an n(≥ 2)-dimensional com-


plete hyperbolic spin manifold of finite volume.
i) If the spin structure of M is trivial along at least one cusp, then σe (D) =
R, in particular σ(D) = σe (D) = R.
ii) If the spin structure of M is non-trivial along all cusps, then σe (D) = ∅,
in particular σ(D) = σd (D).
Explicit 3-dimensional examples constructed out of complements of knots
are given in [41] where both possibilities occur according to the parity of the
linking numbers, see [41, Thm. 4].
110 7 The Dirac spectrum on non-compact manifolds

Theorem 7.4.1 has been generalized in several ways. W. Ballmann and


J. Brüning [30, Thm. E] have proved that the conclusion of Theorem 7.4.1.ii)
holds as soon as the sectional curvature is assumed to be pinched in ] − ∞, 0[
and the cusp metric is of general warped product type:

Theorem 7.4.2 (W. Ballmann and J. Brüning [30, 31]) Let (M n , g)


be any n(≥ 2)-dimensional complete Riemannian spin manifold with finitely
many cusps. Assume that, on each cusp, the metric g has sectional curvature
in [−b2 , −a2 ] with 0 < a < b < ∞ and that w.r.t. the induced spin structure
the Dirac operator of (N, gt ) has trivial kernel for large enough t.
Then σe (D) = ∅, in particular σ(D) = σd (D).
The case of a non-complete underlying manifold (M n , g) is somewhat more
delicate to handle since there are no cusps in general. A geometric situation
where Theorem 7.4.1 can be generalized has been discovered by A. Moroianu
and S. Moroianu [201, Thm. 2.1]. Their setting appears as natural when
considering the existence of Poincaré-Einstein metrics, where the metric is
required to be conformal to a product metric at infinity.

Theorem 7.4.3 (A. Moroianu and S. Moroianu [201]) Let M n be any


connected n-dimensional spin manifold with non-empty boundary ∂M . Con-
sider a Riemannian metric g on M of the form g = dx2 ⊕ g∂M in a
neighbourhood of ∂M , where x : M −→ [0, ∞[ is the distance function to
∂M and g∂M is a Riemannian metric on ∂M . For f ∈ C ∞ (M \ ∂M, ]0, ∞[)
 r only depends on x in a neighbourhood of ∂M set g := f g on M \ ∂M .
2
which
If 0 f (x)dx = ∞ for some r ∈]0, ∞[ and the Dirac operator of (∂M, g∂M ) is
essentially self-adjoint in L2 (Σ∂M ), then the Dirac operator of (M \ ∂M, g)
has no point spectrum.
r
The condition 0 f (x)dx = ∞ imposes the boundary ∂M to be at infinite
distance from its complement in M w.r.t. g. Theorem 7.4.3 enhances an
earlier result by J. Lott [182, Thm. 1]. As an application of Theorem 7.4.3,
A. Moroianu and S. Moroianu showed the following: if (M n , g) is a complete
Riemannian manifold carrying an incomplete vector field which, outside a
compact subset, vanishes nowhere, is conformal and at the same time the
gradient of a function, then σ(D) = σe (D) (see [201, Thm. 4.1]). Note that
the incompleteness assumption on the vector field is essential because of
Theorem 7.4.1.ii) (take e.g. X := e−t ∂t ∂
along a cusp) [201, Rem. 4.4]).
Further applications of Theorem 7.4.3 as well as references are discussed in
[201, Sec. 5].
Theorem 7.4.3 was also inspired by an earlier work
 r of S. Moroianu [204,
Thm. 1 & Thm. 2], where the geometric condition 0 f (x)dx = ∞ is replaced
by the invertibility of the Dirac operator on the boundary as in the spirit of
Theorem 7.4.1.ii):
7.4 Absence of a spectral component 111

Theorem 7.4.4 (S. Moroianu [204]) Let M n be any compact connected


n-dimensional spin manifold with non-empty boundary ∂M . Consider a
Riemannian metric g on M which in some local coordinates near ∂M is
of the form

dx2  
n−1 n−1
dxdyj
g = a00 (x, y) 4 + a0j (x, y) 2 + aij (x, y)dyi dyj ,
x j=1
x i,j=1

where x : M −→ [0, ∞[ is the distance function to ∂M , aαβ ∈ C ∞ (M ×M, R)


  n−1
with aij (0, y) 1≤i,j≤n−1 positive definite. Set g∂M := i,j=1 aij (0, y)dyi dyj
and, for p > 0, g := x2p g on M \ ∂M .
If a00 (0, y) = a0j (0, y) = 1 for every 1 ≤ j ≤ n and the Dirac operator of
(∂M, g∂M ) is invertible, then the Dirac operator of (M \ ∂M, g) is essentially
self-adjoint and has no essential spectrum.

The particular form of the metric near the boundary in Theorem 7.4.4
allows furthermore the existence of a nice Weyl’s asymptotic estimate for the
eigenvalue counting function [204, Thm. 3].
In the radically different situation where the curvature is non-negative, one
still may guarantee the point spectrum to be empty or almost empty. How-
ever the hypotheses needed are much stronger. Let (M n , g) be geodesically
starshaped w.r.t. some point x0 . Assume the scalar curvature of (M n , g) to
be non-negative and that particular parts of the sectional curvature remain
pointwise pinched in [0, c(n)r 2 ], where r is the distance function from x0 and
c(n) is a positive constant depending explicitly on n. Then σd (D) = ∅ or {0}
as shown by S. Kawai [153, Thm. 3].
Chapter 8
Other topics related
with the Dirac spectrum

We outline the main topics in relation with the spectrum of Dirac operators
that have been left aside in this overview.

8.1 Other eigenvalue estimates

As we have seen in Section 2.2, the Dirac operator on homogeneous spaces


can be described as a family of matrices using the decomposition of the
space of L2 -sections of ΣM into irreducible components. What happens if the
homogeneity assumption is slightly weakened? This question has first been
addressed by M. Kraus in the cases of isometric SOn -actions and warped
products over S1 respectively. Although the explicit knowledge of the Dirac
spectrum becomes out of reach, the eigenvalues can still be approximated in
a reasonable way.
Theorem 8.1.1 (M. Kraus [165, 166]) For n ≥ 2, let g be any
Riemannian metric on Sn such that SOn acts isometrically on (S n , g). Write
n−1
fmax · Vol(S n−1 , can) for the maximal volume of the orbits of the SOn -action.
Then the Dirac spectrum of (S n , g) is symmetric about the origin,

(n − 1)2
λ1 (DS2n ,g ) ≥ 2
4fmax

n n−1+k
and there are at most 2[ 2 ] · eigenvalues of DS2n ,g in the interval
k
2
( n−1
2 +k) ( n−1 +k+1)2
[ 2
fmax , 2 f2 [, for every nonnegative integer k.
max

The proof of Theorem 8.1.1 relies on the following arguments: the SOn -
action allows a dense part of (Sn , g) to be written as a warped product of
Sn−1 with an interval. On this dense part the eigenvalue problem on (Sn , g)
translates into a singular nonlinear differential equation of first order with
boundary conditions at both ends. The rest of the proof involves Sturm-
Liouville theory, we refer to [166] for details.
N. Ginoux, The Dirac Spectrum, Lecture Notes in Mathematics 1976, 113
DOI 10.1007/978-3-642-01570-0 8, 
c Springer-Verlag Berlin Heidelberg 2009
114 8 Other topics related with the Dirac spectrum

Note that the inequality in Theorem 8.1.1 is not sharp for the standard
2
metric on Sn since λ1 (DS2n ,can ) = n4 (see Theorem 2.1.3). However Theorem
8.1.1 provides sharp asymptotical eigenvalue estimates in the two follow-
ing situations. First consider the cylinder C n (L) :=]0, L[×Sn−1 with half
n-dimensional spheres glued at both ends. Obviously C n (L) admits an iso-
metric SOn -action for which fmax = 1, in particular Theorem 8.1.1 implies
2
that λ1 (DC2 n (L) ) ≥ (n−1)
4 . On the other hand, C n (L) sits in Rn+1 by con-
struction; now C. Bär’s upper bound (5.19) in terms of the averaged total
squared mean curvature is not greater than

(n − 1)2 LVol(Sn−1 , can) + n2 Vol(Sn , can)


,
4(LVol(Sn−1 , can) + Vol(Sn , can))

so that [166]
(n − 1)2
lim λ1 (DC2 n (L) ) = .
L→∞ 4
x2
For the 2-dimensional ellipsoid Ma := {x ∈ R3 | x21 + x22 + a23 = 1} (where
a > 0) the maximal length of S1 -orbits is 2π, so that by Theorem 8.1.1 the
2
inequality λ1 (DM a
) ≥ 14 holds. Combining this with the upper bound (5.8)
provides [165]
2 1
lim λ1 (DM )= .
a→∞ a
4
The technique of separation of variables used in the proof of Theorem 8.1.1
also provides a lower eigenvalue bound on warped product fibrations over
S1 in terms of the Dirac eigenvalues of the fibres, see [167, Thm. 2]. As for
the case of higher dimensional fibres over arbitrary base manifolds, the only
family which has been considered so far is that of warped products with fibre
Sk with k ≥ 2, where decomposing the Dirac operator into block operator
matrices provides similar results to those of Theorem 8.1.1, see [169].
Another natural but completely different way to study the Dirac eigen-
values consists in comparing them with those of other geometric operators.
Hijazi’s inequality (3.18) is already of that kind since μ1 is the smallest eigen-
value of the conformal Laplace operator. As for spectral comparison results
between the Dirac and the scalar Laplace operators, the first ones were proved
by M. Bordoni. They rely on a very nice general comparison principle between
two operators satisfying some kind of Kato-type inequality. The estimate
which can be deduced reads as follows.

Theorem 8.1.2 (M. Bordoni [60]) Let 0 = λ0 (Δ) < λ1 (Δ) ≤ λ2 (Δ) ≤ . . .
be the spectrum of the scalar Laplace operator Δ on a closed n(≥ 2)-
dimensional Riemannian spin manifold (M n , g). Then for any positive
integer N [60, Prop. 4.20]
8.1 Other eigenvalue estimates 115

n  λk (Δ) 
λ2N (D2 ) ≥ inf (S) + , (8.1)
4(n − 1) M
n
2(2[ 2 ] + 1)2

N
where k = [ n ].
2[ 2 ] +1

In particular Bordoni’s inequality (8.1) implies Friedrich’s inequality (3.1)


n
as well as the presence of at most 2[ 2 ] eigenvalues of D2 in the interval

n n  λ1 (Δ) 
[ inf (S), inf (S) + [,
4(n − 1) M 4(n − 1) M
n
2(2[ 2 ] + 1)2

see Section 8.2 for further results on the spectral gap.


Bordoni’s results were generalized by M. Bordoni and O. Hijazi in the
Kähler setting [61], where essentially the Friedrich-like term in the lower
bound must be replaced by the Kirchberg-type one of inequality (3.10) in
odd complex dimension.
Comparisons between Dirac and Laplace eigenvalues which go the other
way round can be obtained in particular situations. In the case of surfaces,
J.-F. Grosjean and E. Humbert proved the following.

Theorem 8.1.3 (J.-F. Grosjean and E. Humbert [113]) Let [g] be a


conformal class on a closed orientable surface M 2 with fixed spin structure,
then [113, Cor. 1.2]
 λ1 (D2 )  1
g
inf ≤ , (8.2)
g∈[g] λ1 (Δg ) 2
where here λ1 (Dg2 ) denotes the smallest positive eigenvalue of Dg2 .

Inequality (8.2) is optimal and sharp for M 2 = S2 : indeed for any Rieman-
λ1 (Δ )
nian metric g one has λ1 (DS22 ,g ) ≥ S ,g 2
2 as a straightforward consequence
of Bär’s inequality (3.17) and Hersch’s inequality (3.22). Moreover, (8.2) com-
pletes [1] where I. Agricola, B. Ammann and T. Friedrich prove the existence
of a 1-parameter family (gt )t≥0 of S1 -invariant Riemannian metrics on T2 for
which, in the same notations as just above, λ1 (ΔT2 ,gt ) < λ1 (DT2 2 ,gt ) for any
t ≥ 0, where T2 is endowed with its trivial spin structure. The inequality
λk (ΔT2 ,g ) ≥ λk (DT2 2 ,g ) for k large enough and for particular metrics g on T2
with trivial spin structure has been proved independently by M. Kraus [168].
In the case where the manifold sits as a hypersurface in some spaceform,
the best known result is the following.
Theorem 8.1.4 (C. Bär [40]) Let (M n , g) be isometrically immersed into
Rn+1 or Sn+1 and carry the induced spin structure, then [40, Thm. 5.1]

n2  
λN (D2 ) ≤ sup(H 2 ) + κ + λ[ N −1 (Δ) (8.3)
4 M 2μ ]
116 8 Other topics related with the Dirac spectrum

for every positive N ∈ N, where κ ∈ {0, 1} denotes the sectional curvature of


the ambient space, H denotes the mean curvature of M n and μ is the integer
2 ] − n mod 2.
defined by μ := [ n+1
Inequality (8.3) follows from the min-max principle and from (5.16) where
one chooses f to be an eigenfunction of Δ and ψ to be the restriction of a
non-zero Killing spinor.

8.2 Spectral gap

Another method to obtain information on the eigenvalues consists in esti-


mating their difference, which is called the spectral gap. Initiated by H.C.
Yang (see reference in [76]) for the scalar Laplacian, this approach turns out
to provide similar results for the Dirac operator. The proof of the following
theorem relies on the min-max principle and a clever input of coordinate
functions of the immersion into the Rayleigh quotient, see [76] for details.
Theorem 8.2.1 (D. Chen [76]) Let (M n , g) be any n-dimensional closed
immersed Riemannian spin submanifold of RN for some N ≥ n + 1. Denote
the spectrum of D2 by {λj (D2 )}j≥1 and set, for every j ≥ 1,

1 2 
μj := λj (D2 ) + n sup(H 2 ) − inf (S) ,
4 M M

where H and S are the mean and the scalar curvature of M respectively.
Then for any k ≥ 1


k
4
(μk+1 − μj )(μk+1 − (1 + )μj ) ≤ 0. (8.4)
j=1
n

Note that the codimension of M is arbitrary and that no compatibility


condition between the spin structure of M and that of RN is required. Ele-
mentary computations show that inequality (8.4) implies

4 
k
1
μk+1 ≤ (1 + ) μj ,
k n j=1

which itself provides


4 
k
μk+1 − μk ≤ μj ,
nk j=1

which had been shown independently by N. Anghel [26]. In particular pre-


cise estimates on the growth rate of the Dirac eigenvalues can be deduced.
8.3 Pinching Dirac eigenvalues 117

Theorem 8.2.1 has been extended by D. Chen and H. Sun to holomorphically


immersed submanifolds of the complex projective space [77, Thm. 3.2].

8.3 Pinching Dirac eigenvalues

If Friedrich’s inequality (3.1) is an equality for the smallest eigenvalue λ1 (D2 ),


then from Theorem 3.1.1 and Proposition A.4.1 the underlying Riemannian
manifold must be Einstein, which is a quite rigid geometric condition. Does
the manifold remain “near to” Einstein if λ1 (D2 ) - or at least some lower
eigenvalue - is close enough to Friedrich’s lower bound? This kind of issue
is designed under the name eigenvalue pinching. It addresses the continuous
dependence of the geometry on the spectrum, in a sense that must be precised.
We denote in the rest of this section by Ksec , diam and S the sectional
curvature, diameter and scalar curvature of a given Riemannian manifold
respectively. We also call two spin manifolds spin diffeomorphic if there exists
a spin-structure-preserving diffeomorphism between them.
The first pinching result for Dirac eigenvalues is due to B. Ammann and
C. Sprouse. It deals with the case where the scalar curvature almost vanishes.
Tori with flat metric and trivial spin structure carry a maximal number of
linearly independent parallel (hence harmonic) spinors. Theorem 8.3.1 below
states that, under boundedness assumptions for the diameter and the sec-
tional and scalar curvatures, one stays near to a flat torus in case some lower
Dirac eigenvalue is not too far away from 0. Recall that a nilmanifold is the
(left or right) quotient of a nilpotent Lie group by a cocompact lattice. If a
(left or right) invariant metric is fixed on the nilmanifold, then the trivial lift
of the lattice to the spin group provides a spin structure called the trivial
one, see Proposition 1.4.2 for spin structures on coverings.

Theorem 8.3.1 (B. Ammann and C. Sprouse [25]) Let K, d be posi-


n
tive real constants, n ≥ 2 be an integer, r := 1 if n = 2, 3 and r := 2[ 2 ]−1 +1 if
n ≥ 4. Then there exists an ε = ε(n, K, d) > 0 such that every n-dimensional
closed Riemannian spin manifold (M n , g) with

|Ksec (M n , g)| < K, diam(M n , g) < d, S(M n , g) > −ε and λr (DM


2
n ,g ) < ε

is spin diffeomorphic to a nilmanifold with trivial spin structure.

Theorem 8.3.1 implies the existence of a uniform lower eigenvalue bound for
the Dirac operator in the following family: there exists an ε = ε(n, K, d) > 0
such that on every n-dimensional closed Riemannian spin manifold (M n , g)
with |Ksec (M n , g)| < K, diam(M n , g) < d, S(M n , g) > −ε and which is
not spin diffeomorphic to a nilmanifold with trivial spin structure the rth
eigenvalue of D2 satisfies
λr (D2 ) ≥ ε.
118 8 Other topics related with the Dirac spectrum

The choice for r, which looks a priori curious, is actually optimal since the
product of a so-called K3-surface with a torus carries exactly r − 1 linearly
independent parallel spinors, see [25, Ex. (2) p.411]. The proof of Theorem
8.3.1 makes use of an approximation result by U. Abresch (see reference in
[25]) in an essential way, we refer to [25, Sec. 7] for details.
Under the supplementary assumption of a lower bound on the volume, the
metric can even be shown to stay near to some with parallel spinors.
Theorem 8.3.2 (B. Ammann and C. Sprouse [25]) Let K, d, V, δ be
positive real constants, n ≥ 2 be an integer, r := 1 if n = 2, 3 and
n
r := 2[ 2 ]−1 + 1 if n ≥ 4. Then there exists an ε = ε(n, K, d, V, δ) > 0
such that for every n-dimensional closed Riemannian spin manifold (M n , g)
with

|Ksec (M n , g)| < K, diam(M n , g) < d, S(M n , g) > −ε, Vol(M n , g) > V

2 1,α
and λr (DM n ,g ) < ε, the metric g is at C -distance at most δ to a metric
admitting a non-zero parallel spinor.
The proof of Theorem 8.3.2 relies on a similar general eigenvalue pinch-
ing valid for arbitrary rough Laplacians on arbitrary vector bundles due
to P. Petersen (see reference in [25]) and on the Schrödinger-Lichnerowicz
formula (3.2). Petersen’s method can also be applied to the rough Laplacian
associated to the deformed covariant derivative X → ∇X + ρX· and in this
case it provides the following:
Theorem 8.3.3 (B. Ammann and C. Sprouse [25]) Let K, d, V, ρ, δ be
positive real constants, n ≥ 2 be an integer, r := 1 if n = 2, 3 and r :=
n
2[ 2 ]−1 + 1 if n ≥ 4. Then there exists an ε = ε(n, K, d, V, ρ, δ) > 0 such that
for every n-dimensional closed Riemannian spin manifold (M n , g) with

|Ksec (M n , g)| < K, diam(M n , g) < d, Vol(M n , g) > V, S(M n , g) ≥ n(n−1)ρ2


2 2
2 n ρ 1,α
and λr (DM n ,g ) <
4 + ε, the metric g is at C -distance at most δ to a
2
metric with constant sectional curvature ρ .

Note that the bound on the sectional curvature is necessary because of


Bär-Dahl’s result [46] discussed in Section 3.2. However the minimal number
r necessary for the result to hold can be enhanced.

Theorem 8.3.4 (A. Vargas [231]) The conclusion of Theorem 8.3.3 holds
with 
3 if n = 6 or n ≡ 1 (4)
r := n+9
4 if n ≡ 3 (4).
8.4 Spectrum of other Dirac-type operators 119

8.4 Spectrum of other Dirac-type operators

Up to now we have concentrated onto the fundamental (or spin) Dirac opera-
tor on a spin manifold. As already mentioned at the beginning of Chapter 1,
Dirac-type operators may be defined in the more general context where a so-
called Clifford bundle [173, Sec. II.3] is at hand. Roughly speaking, a Clifford
bundle is given by a Hermitian vector bundle together with a covariant deriva-
tive and on which the tangent bundle acts by Clifford multiplication such that
all three objects (Hermitian metric, covariant derivative and Clifford multi-
plication) are compatible with each other in the sense of Definition 1.2.2 and
Proposition 1.2.3. The associated Dirac operator is defined as the Clifford
multiplication applied to the covariant derivative. One may add a zero-order
term and obtain a so-called Dirac-Schrödinger operator. In this section we dis-
cuss spectral results in relation with the spinc Dirac operator, with twisted
Dirac-Schrödinger operators, with Dirac operators associated to particular
geometrically relevant connections, with the basic Dirac operator and in the
pseudo-Riemannian setting.
First, the concept of spin structure may be weakened to that of spinc
structure, whose structure group is the spinc group Spincn := Spinn × S /Z2 .
1

Such a structure comes along with a S1 -principal bundle, or equivalently


with a complex line bundle L. We do not want to define spinc structures
more precisely but mention that all spin manifolds are spinc and that all
almost-Hermitian manifolds have a canonical spinc structure [173, App. D].
Moreover the choice of a covariant derivative on the line bundle induces
a covariant derivative and hence a Dirac operator on the associated spinor
bundle over the underlying manifold. In that case it can be expected that most
of the results valid for the spin Dirac operator remain valid for the spinc one,
except that the curvature of the line bundle must in some situations be taken
into account. For example, M. Herzlich and A. Moroianu proved the analog
of Hijazi’s inequality (3.18) in the spinc context: denote by ω the curvature
form of the line bundle L and by μ1 the smallest eigenvalue of the scalar
n 12
operator Lω := 4 n−1n−2 Δ + S − 2[ 2 ] |ω|, then any eigenvalue λ of the spin
c

Dirac operator satisfies [126, Thm. 1.2]


n
λ2 ≥ μ1 .
4(n − 1)

We note however that little has been done in the spinc context in comparison
with the spin one.
If the underlying space is again our familiar spin manifold (M n , g) and if we
choose an arbitrary Riemannian or Hermitian vector bundle E over M , then
the tensor product bundle ΣM ⊗ E carries a canonical Clifford multiplication
(extend the Clifford multiplication by the identity on the second factor). If
we endow E with a metric covariant derivative, then we obtain a structure
of Clifford bundle and an associated Dirac operator called Dirac operator of
120 8 Other topics related with the Dirac spectrum

E
M twisted with E. This operator is usually denoted by DM . For example,
the Euler operator d + δ can be seen as the Dirac operator of M twisted with
ΣM : this follows essentially from (1.2) and may actually be stated without
any spin structure on M [173, Sec. II.6]. Another prominent example is the
Dirac operator of a spin submanifold twisted with the spinor bundle of its
normal bundle (where the latter is assumed to be spin). Various studies have
been devoted to the spectrum of twisted Dirac operators, therefore we restrict
ourselves to a few ones which we hope to be representative. We include all
that concerns Dirac-Schrödinger operators, since in that case the zero order
term mainly translates the upper or lower bounds by a constant.
Let first E be as above, M be closed and f be a smooth real function
n by κ1 the smallest eigenvalue on M of the pointwise linear
on M . Denote
operator k,l=1 ek · el · ReEk ,el , where RE is the curvature tensor of the chosen
covariant derivative on E (and (ek )1≤k≤n is a local o.n.b. of TM ). If the
inequalities n(S + κ1 ) > (n − 1)f 2 > 0 hold on M , then any eigenvalue λ of
the Dirac-Schrödinger operator DM E
− f acting on Γ(ΣM ⊗ E) satisfies [105,
Prop. 4.1]
1  n 2
λ2 ≥ inf (S + κ1 ) − |f | . (8.5)
4 M n−1
Inequality (8.5), which can be deduced from a clever choice of modified covari-
ant derivative, stands for the analog of Friedrich’s inequality in this context,
see [105] for other kinds of estimates and references to earlier works on that
topic (such as [196]). In the particular case where n = 4, f = 0, E is ar-
bitrary and carries a selfdual covariant derivative, the estimate (8.5) can be
enhanced using the decomposition ΣM = Σ+ M ⊕ Σ− M and the vanishing of
one half of the auxiliary curvature term computed from RE : H. Baum proved
[52, Thm. 2] that
1
λ2 ≥ inf (S)
3 M
E
for any eigenvalue λ of DM , which is exactly Friedrich’s inequality (3.1) for
the eigenvalues of the spin Dirac operator.
Staying in dimension 4, if the spin manifold (M 4 , g) carries a Hermitian
structure J (i.e., an orthogonal complex structure on TM ) then one is led
0,∗
to the Dirac operator twisted with√ E = ΣM = Λ TM which is nothing
else than the Dolbeault operator 2(∂ + ∂). Although Kirchberg’s inequality
(3.10) does not apply, sharp lower bounds for the eigenvalues of the Dolbeault
operator are still available: B. Alexandrov, G. Grantcharov and S. Ivanov
proved [6, Thm. 2] that
1
λ2 ≥ inf (S)
6 M

for any eigenvalue λ of 2(∂ + ∂). Beware that equality cannot occur for a
non-flat Kähler metric because of (3.10). The proof of that inequality relies
on Weitzenböck formulas and the clever choice of twistor operators asso-
ciated to a canonical one-parameter-family of connections, we refer to [6]
8.4 Spectrum of other Dirac-type operators 121

for details. Besides, we mention that upper eigenvalue bounds for particular
twisted Dirac operators have been obtained in [52], [40] and [101].
From the point of view of geometers investigating the integrability of par-
ticular G-structures, there exists another interesting family of Dirac-type
1 1
operators which are usually denoted by D 3 and defined by D 3 := Dg + T4 ·,
where T is some given 3-form and Dg is the spin Dirac operator on the
Riemannian spin manifold (M n , g). For example if (M n , g) is a so-called re-
1
ductive homogeneous space then D 3 is the so-called Kostant Dirac operator
1
(see reference in [3]); if (M n , g) is a Hermitian manifold then D 3 coincides
with the Dolbeault-operator defined just above. In case T is the charac-
teristic torsion of a 5-dimensional closed spin Sasaki manifold with scalar
curvature bounded from below, the use of suitable deformations of the con-
nection by polynomials of the torsion form allowed I. Agricola, T. Friedrich
1
and M. Kassuba to prove the following estimates of any eigenvalue λ of (D 3 )2
[3, Thm. 4.1]:
 1 √
 (1 + 1 inf M (S))2 if − 4 < S ≤ 4(9 + 4 5)
 16 4
λ ≥  √
 5 inf M (S) if S ≥ 4(9 + 4 5).
16

Equality holds if (M 5 , g) is η-Einstein (see [3] for a definition). Surprisingly


enough the first lower bound depends quadratically on the scalar curvature,
which makes the estimate better for small S. We refer to [3] for the proof.
We also note that in the context of contact metric manifolds (which have a
canonical spinc structure) Weitzenböck formulas for the Dirac operator asso-
ciated to the so-called Tanaka-Webster connection have also been produced
in order to prove vanishing theorems [205], however no study of the spectrum
is still available.
Sasaki manifolds can also be viewed as particular foliated Riemannian
manifolds. Spin structures can be defined on Riemannian foliations in much
the same way as on the tangent bundle and an associated covariant derivative
and Dirac operator may be defined which are called the transversal covariant
derivative and transversal Dirac operator respectively. The transversal Dirac
operator, which acts on the space of basic spinors (spinors whose transversal
covariant derivatives vanish along all directions normal to the leaves), is in
general not formally self-adjoint, therefore one considers the symmetrized
operator called basic Dirac operator of the foliation and denoted by Db . It is
a not-so-straightforward adaptation of the proof of Friedrich’s inequality by
G. Habib and K. Richardson to show that any eigenvalue λ of Db on a closed
underlying manifold (M n , g) satisfies [123, Eq. (1.1)]
q
λ2 ≥ inf (S tr ),
4(q − 1) M

where q ≥ 2 stands for the codimension of the foliation and S tr for its
transversal scalar curvature. In case the normal bundle of the foliation carries
122 8 Other topics related with the Dirac spectrum

a Kähler or a quaternionic Kähler structure, analogs of Kirchberg’s inequality


(3.10) and of (3.15) can also be derived [151, 152, 121, 120, 119].
To close this section we mention the only result known to us about the
spectrum of the Dirac operator in the pseudo-Riemannian (non-Riemannian)
setting. First spin structures require the pseudo-Riemannian manifold to be
simultaneously space- and time-oriented in order to be well-defined, see [49]
or [53, Sec. 2]. In that case the choice of a maximal timelike subbundle
induces an L2 -Hermitian inner product on the space of spinors. Unlike its
Riemannian version the associated (spin) Dirac operator is neither formally
self-adjoint w.r.t. that inner product nor elliptic. However H. Baum could
show with the help of suitable endomorphisms of the spinor bundle commut-
ing or anti-commuting with the Dirac operator that the point spectrum, the
continuous spectrum and the residual spectrum of the Dirac operator on any
even-dimensional pseudo-Riemannian manifold are symmetric w.r.t. the real
and imaginary axes. We refer to [53] for further statements and the proof.

8.5 Conformal spectral invariants

In this section we are interested in two invariants associated to the Dirac


spectrum. A good reference for the whole section is [149]. Given a closed spin
manifold M n with fixed conformal class [g] and spin structure denoted by ,
2
let λ1 (DM,g ) be the smallest eigenvalue of the square of the Dirac operator
of (M , g). The Bär-Hijazi-Lott invariant [13, eq. (2.4.1) p.12] of (M n , [g], )
n

is the nonnegative real number λmin (M n , [g], ) defined by


 1

λmin (M n , [g], ) := inf 2 ) · Vol(M, g) n
λ1 (DM,g .
g∈[g]

Of course the expression on the r.h.s. is chosen so as to remain scaling-


invariant. By definition λmin (M n , [g], ) is a conformal invariant. The Bär-
Hijazi-Lott invariant is tightly connected to and behaves much like the
Yamabe invariant. Indeed, it already follows from Bär’s inequality (3.17)
and from Hijazi’s inequality (3.20) that
n
λmin (M 2 , [g], )2 ≥ 2πχ(M 2 ) and λmin (M n , [g], )2 ≥ Y (M, [g])
4(n − 1)
(8.6)
for every n ≥ 3, where χ(M 2 ) and Y (M, [g]) are the Euler characteristic
and the Yamabe invariant respectively. For M 2 = S2 this implies that the
Bär-Hijazi-Lott invariant is positive. More generally, as a consequence of
J. Lott’s estimate (3.21), the Bär-Hijazi-Lott invariant is positive as soon
as the Dirac operator is invertible for some - hence any - metric in the con-
formal class. In particular λmin (M n , [g], ) vanishes if and only if (M n , g)
8.5 Conformal spectral invariants 123

admits non-zero harmonic spinors. Generalizing J. Lott’s Sobolev-embedding


techniques [181] to the case where the Dirac
kernel is possibly non-trivial,
 1
B. Ammann showed the positivity of inf 2 ) · Vol(M, g) n
λ+ (DM,g to
g∈[g]
hold true in general [12, Thm. 2.3], where λ+ (DM,g
2
) denotes the smallest
2
positive eigenvalue of DM,g . As an example, the Bär-Hijazi-Lott invariant of
Sn (n ≥ 2) with standard conformal class [can] and canonical spin structure
1
is given by n2 ωnn , where ωn is the volume of Sn carrying the metric of sectional
curvature 1 (denoted by “can”): this follows from Corollaries 3.3.2 and 3.3.3
2 2
together with λ1 (DS2n ,can ) = n4 and Y (Sn , [can]) = n(n − 1)ωnn if n ≥ 3.
In a similar way as for the Yamabe invariant, the Bär-Hijazi-Lott
invariant cannot be greater than that of the sphere: if n ≥ 3 or
M 2 = S2 then B. Ammann [12, Thm. 3.1 & 3.2] proved that inf
 1
 g∈[g]

λ+ (DM,g ) · Vol(M, g) n ≤ λmin (S , [can]), in particular


2 n

λmin (M n , [g], ) ≤ λmin (Sn , [can]). (8.7)

The proof relies on a suitable cut-off argument performed on Dirac eigen-


vectors on the gluing of a sphere with large radius to the manifold, see [12,
Sec. 3] for the details.
The next step would consist in showing that (8.7) is a strict inequality if
(M n , [g]) is not conformally equivalent to (Sn , [can]). This has been done by
B. Ammann, E. Humbert and B. Morel in the conformally flat setting where
one introduces a further datum, namely the so-called mass endomorphism.
The mass endomorphism of a locally conformally flat Riemannian spin mani-
fold is a self-adjoint endomorphism field of its spinor bundle and can be locally
defined out of the difference between the Green’s operators for the Dirac op-
erators associated to the original metric and to the Euclidean one in suitable
coordinates, see [23, Def. 2.10] for a precise definition. The name comes from
the corresponding term for the Yamabe operator and which is known to pro-
vide the mass of an asymptotically flat Riemannian spin manifold. Moreover,
the mass endomorphism is “well-behaved” regarding conformal changes of
metric [23, Prop. 2.9]. In case the locally conformally flat manifold (M n , [g])
has an invertible Dirac operator (for some hence any metric in the conformal
class) and if its mass endomorphism has a non-zero eigenvalue somewhere on
M n , then [23, Thm. 1.2]

λmin (M n , [g], ) < λmin (Sn , [can]). (8.8)

At this point one should beware that the mass endomorphism of (Sn , [can])
vanishes and that this does not characterize the round sphere since flat tori
also have vanishing mass endomorphism. We refer to [23] for the details. For a
generalization of the Bär-Hijazi-Lott invariant to manifolds with non-empty
boundary we refer to [212, 214].
124 8 Other topics related with the Dirac spectrum

We also mention that the Bär-Hijazi-Lott invariant has been generalized to


the noncompact setting, where it provides an obstruction to the existence of
conformal spin compactifications of the manifold [114]. More precisely, let M n
be any n-dimensional manifold with conformal class [g] and spin structure 
and define λ+ n
min (M , [g], ) as in Section 7.3. If

min (M \ B r (p), [g], ) < λmin (S , [can]),


lim λ+ n n
r→∞

where p ∈ M is arbitrary, then (M n , [g]) is not conformal to a subdomain with


induced spin structure of a closed spin manifold [116, Thm. 3.0.1] (see also
[114, Thm. 1.4]). The vanishing of λ+ n
min (M , [g], ) also prevents the existence
of conformal spin compactifications of M , since λ+ n
min (M , [g], ) > 0 on closed
manifolds [12, Thm. 2.3] and a monotonicity principle holds for λ+ min [116,
Lemma 2.0.3], see [116, Rem. 3.0.4].
The Green’s operators for the Dirac operator have also revealed as a pow-
erful tool in general problems from geometric analysis such as the classical
Yamabe conjecture (“find a metric with constant scalar curvature in a fixed
conformal class”). As shown by R. Schoen, the Yamabe conjecture is implied
by the positive mass theorem through the fact that the constant term in the
asymptotic expansion in inverted normal coordinates of the Green’s operator
for the conformal Laplace operator is proportional to the mass of the confor-
mal blow-up. Furthermore but independently, E. Witten [235] showed that in
the spin setting the positive mass theorem is in turn implied by the existence
of spinor field which is harmonic and asymptotically constant on the con-
formal blow-up. Now it is a striking result by B. Ammann and E. Humbert
[21] that the Green’s operators for the Dirac operator provide such a spinor.
More precisely, if (M n , g) is closed Riemannian spin manifold with positive
Yamabe invariant and which is locally conformally flat if n ≥ 6, then its
conformal blow-up has positive mass. For positive mass theorems we refer to
Section 8.8.
If one lets the conformal class vary on the closed manifold M n , then one
is led to the so-called τ -invariant of M n with spin structure  and which is
defined by  
τ (M n , ) := sup λmin (M n , [g], ) .
[g]

The introduction of the spinorial invariant τ is inspired from that of


R. Schoen’s σ-invariant which is defined in dimension 2 by σ(M 2 ) :=
4πχ(M 2 ) and in dimension n ≥ 3 by
 
σ(M n ) := sup Y (M n , [g]) ,
[g]

where Y (M n , [g]) denotes the Yamabe invariant on (M n , [g]). There are at


least two motivations for the study of the τ -invariant. First, the τ -invariant
bounds the σ-invariant from above since it follows from (8.6) that, in every
dimension n ≥ 2,
8.6 Convergence of eigenvalues 125
n
τ (M n , )2 ≥ σ(M n ),
4(n − 1)
with equality for Sn . Therefore upper bounds for τ (M n , ) provide upper
bounds for σ(M n ), on which little √ is known. In an independent context,
the inequality λmin (M 2 , [g], ) < 2 π = τ (S2 ) guarantees the existence of a
metric g ∈ [g] for which any simply-connected open subset of (M 2 , g) can be
isometrically embedded with constant mean curvature into R3 [13, Sec. 5.4]. √
Hence it is of geometric interest to know when the inequality τ (M 2 , ) < 2 π
holds. In case M 2 = T2 B. Ammann and E. Humbert have shown [22, Thm.
1.1] that √
τ (T2 , ) = 2 π
for any of its non-trivial spin structures  (obviously τ (T2 , 0 ) = 0 for the
trivial spin structure 0 ). Note that this neither proves nor contradicts the
existence of immersed constant mean curvature tori in R3 . As a generaliza-
tion, the τ -invariant of Sn−1 × S1 is equal to zero if n = 2 and S1 carries the
trivial spin structure and to τ (Sn ) otherwise [22, Thm. 1.2].

8.6 Convergence of eigenvalues

Given a converging sequence of closed Riemannian spin manifolds, does their


Dirac spectrum have to converge to that of the limit? Three very different
contexts have up to now been considered where this question can be given
sense and answered. The simplest and historically the first one deals with
the behaviour of the Dirac spectrum of S1 -bundles under collapse. In that
case the behaviour depends sensitively of the spin structure as shown by
B. Ammann and C. Bär [15]. Let M denote the total space of an S1 -bundle
which is simultaneously a Riemannian submersion with totally geodesic fibres
over a base manifold B. Two kinds of spin structures can be defined on M
according to whether the S1 -action can be lifted to the spin level or not; in
the former case the spin structure is called projectable and in the latter it is
called non-projectable. Projectable spin structures on M stand in one-to-one
correspondence with spin structures on B. The main result of [15] states the
following about the convergence of the Dirac spectrum of M as the fibre-
length goes to 0: either the spin structure of M is projectable and there exist
Dirac eigenvalues of M converging to those of B or it is non-projectable and
all Dirac eigenvalues of M tend to ∞ or −∞ [15, Thm. 4.1 & 4.5]. As an
interesting application, the Dirac spectrum of all complex odd-dimensional
complex projective spaces can be deduced from that of the Berger spheres
(Theorem 2.2.2). Parts of those results have been generalized by B. Ammann
to S1 -bundles with non-geodesic fibres [7, 8].
The second natural context deals with hyperbolic degenerations, i.e., with
sequences of closed hyperbolic spin manifolds (Mj )j∈ N converging to a non-
compact complete hyperbolic spin manifold M (here a hyperbolic metric is
126 8 Other topics related with the Dirac spectrum

a metric with constant sectional curvature −1). Those sequences only exist
in dimensions 2 and 3 and, provided the convergence respects the spin struc-
tures in some sense, the limit manifold must have discrete Dirac spectrum in
dimension 3 whereas it may have continuous spectrum in dimension 2, see
references in [208] where a precise description of hyperbolic degenerations is
recalled. In case the limit manifold M is assumed to have discrete Dirac spec-
trum, F. Pfäffle proved the convergence of the Dirac spectrum of (Mj )j∈ N in
the following sense [208, Thm. 1.2] (see also [207]): For all ε > 0 and Λ ≥ 0,
there exists an N ∈ N such that for all j ≥ N the real number Λ lies neither
in the spectrum of D nor in that of DMj , both Dirac operators DMj and D
have only discrete eigenvalues and no other spectrum in [−Λ, Λ], they have
the same number m of eigenvalues in [−Λ, Λ] which can be ordered so that
(j)
|λk − λk | ≤ ε holds for all 1 ≤ k ≤ m.
The diameter of the converging sequence of degenerating hyperbolic mani-
folds cannot be controlled since the limit-manifold must have a finite number
of so-called cusps, which by definition are unbounded. The third context to
have been considered precisely deals with the situation where both the diam-
eter and the sectional curvature of the converging sequence are assumed to
remain bounded. In that case J. Lott proved the following very general result
[183]. Consider a sequence (gj )j∈ N of bundle metrics on the total space of a
spin fibre bundle M over a base spin manifold B. Assume the fibre length to
go to 0 as j tends to ∞ while both the diameter and the sectional curvature
of (M, gj ) remain bounded. Then the Dirac spectrum of (M, gj ) converges
in the sense just above to that of some differential operator of first order on
B which can be explicitly constructed. Since a precise formulation and the
discussion of the results would require too many details we recommend the
introduction of [183].

8.7 Eta-invariants

As we have seen in Theorem 1.3.7, the Dirac spectrum of any closed n-dimen-
sional Riemannian spin manifold is symmetric w.r.t. the origin in dimension
n ≡ 3 (4). To measure the asymmetry of the Dirac spectrum in case n ≡ 3 (4),
Atiyah, Patodi and Singer introduced [27] the so-called η-invariant of D which
is defined by η(D) := η(0, D), where, for every s ∈ C with e(s) > n,

 sgn(λj )
η(s, D) := .
|λj |s
λj =0

The λj ’s denote the eigenvalues of D. It is already a non-trivial statement


that s → η(s, D) can be meromorphically extended onto C and is regular
at s = 0, see [27]. Originally the η-invariant was introduced to describe
some boundary term in the Atiyah-Patodi-Singer index theorem [27]. In a
8.8 Positive mass theorems 127

simple-minded way, the η-invariant of D can be thought of as the difference


between the number of positive and that of negative Dirac eigenvalues (of
course this has no sense since both numbers are infinite). In particular the
η-invariant of D vanishes as soon as the Dirac spectrum is symmetric.
Few η-invariants are known explicitly. One of the first computations of
η-invariant goes back to Hitchin [148], where the explicit knowledge of the
Dirac spectrum on the Berger sphere S3 allows the η-invariant to be ex-
plicited. This was generalized onto all Berger spheres by D. Koh [162]. In
the flat setting, the η-invariant can also be deduced from the Dirac spec-
trum in dimension n = 3 [206] and for particular holonomies in dimension
n ≥ 4 [188]. Theorem 2.2.3 provides the η-invariant on particular closed
3-dimensional hyperbolic manifolds [218]. The most general formula allow-
ing the determination of the η-invariant has been proved by S. Goette [108,
Thm. 2.33] on homogeneous spaces, where η(D) arises as the sum of three
terms: a representation-theoretical expression, an index-theoretical one and
so-called equivariant η-invariants, which can themselves be deduced from
finer representation-theoretical data [106, 107].
Though unknown in most cases, the η-invariant behaves nicely under
connected sums: roughly speaking, if a closed Riemannian spin manifold is
separated in two pieces M1 , M2 by a closed hypersurface N , about which both
the metric and the Dirac operator split as on a Riemannian product, then
the η-invariant of D consists of the sum of the η-invariants of DM1 and DM2
plus the so-called Maslov-index of a pair of Lagrangian subspaces of Ker(DN )
making DMj self-adjoint, plus some index-theoretical integers (U. Bunke [67,
Thm 1.9]). We refer to [67] for an overview of η-invariants of general Dirac-
type operators and numerous useful references.
We also mention that some kind of η-invariant can be defined in the non-
compact setting, see [118] and references therein.

8.8 Positive mass theorems

Although this section has more to do with physics as with the Dirac spectrum,
we include it because on the one hand the proofs of the results presented
involve simple spinorial techniques as already used above, and on the other
hand positive mass theorems nowadays play a central role in many other
topics of global analysis such as the Yamabe problem. A good but not up-to-
date reference for that topic is [125].
A positive mass theorem (sometimes called positive energy theorem) is a
two-fold statement reading roughly as follows: Let (M n , g) be a Riemannian
manifold which is asymptotic to a model manifold (in a sense that must be
precised) and some of which curvature invariant satisfies a pointwise inequal-
ity, then some asymptotic geometric invariant called its mass also satisfies a
similar inequality and, if this latter inequality is an equality, then the whole
128 8 Other topics related with the Dirac spectrum

manifold is globally isometric to the original model manifold. To fix the ideas
we concentrate from now on onto the original positive mass theorem as proved
by R. Schoen and S.-T. Yau [215, 216] and independently by E. Witten [235]
in the spinorial setting, in particular we leave aside all recent developments
in what has become a whole field of research at the intersection between
mathematics and general relativity, see e.g. [236] for references.
Let (M n , g) be a Riemannian manifold of dimension n ≥ 3. Call it asymp-
totically flat of order τ ∈ R if there exists a compact subset K ⊂ M , a positive
real number R and a diffeomorphism M \ K −→ {x ∈ Rn , |x| > R} such
that the pushed-out metric fulfills: gij − δij = O(|x|−τ ), ∂xijk = O(|x|−τ −1 )
∂g

∂2g
ij
and ∂xk ∂x l
= O(|x|−τ −2 ) as |x| → ∞, for all 1 ≤ i, j, k, l ≤ n. Given such a
manifold (M n , g), set
 n
1 ∂gij ∂gii
m(g) := · lim ( − )νj dA,
16π r→∞ Sr i,j=1 ∂xi ∂xj

where Sr denotes the Euclidean sphere of radius r about 0 ∈ Rn with outside


unit normal ν and dA its canonical measure. Beware here that in general
m(g) does not make any sense: the integral need not converge, and even if
it converges it depends on the choice of asymptotic coordinates. If however
τ > n−2 2 and the scalar curvature of (M n , g) is integrable, then a highly
non-trivial theorem of R. Bartnik (see reference in [125]) ensures m(g) to
be well-defined. In that case it is called the ADM-mass of (M n , g). The
canonical example of asymptotically flat manifold (of any order) is (Rn , can),
whose ADM-mass vanishes. The positive mass theorem states that, with the
assumptions above and if the scalar curvature S of (M n , g) is non-negative,
then m(g) ≥ 0 with equality if and only if (M n , g) = (Rn , can). This is a very
deep statement since it establishes a direct relationship between the geometry
at infinity and the global geometry of M . For example, as a consequence, any
Riemannian metric on Rn with S ≥ 0 and which is flat outside a compact
subset must be flat. Surprisingly enough, the positive mass theorem follows
from relatively simple considerations involving some kind of boundary value
problem for the Dirac operator, at least in case M is spin, as shown by
E. Witten [235]. Let us sketch his idea.
The first and main step in Witten’s proof consists in choosing any non-
zero “constant” spinor field ψ0 at infinity and exhibiting a sufficiently regular
non-zero spinor field ψ lying in the kernel of D2 and being asymptotic to ψ0 .
This can be done by showing the invertibility of D2 between suitable Hölder
spaces. Applying Schrödinger-Lichnerowicz’ formula (1.15), integrating on a
Euclidean ball of (sufficiently large) radius r and using (3.29) together with
Green’s formula one obtains
  
S 2
0= D ψ, ψ
vg =
2
(|∇ψ| + |ψ| )vg −
2
∇ν ψ, ψ
dA,
Br Br 4 Sr
8.8 Positive mass theorems 129

where ν denotes here the outer unit normal to Sr = ∂Br . The miracle
in Witten’s proof happens here: it can be easily shown that the bound-
ary term Sr ∇ν ψ, ψ
dA is asymptotic to m(g) times some finite positive
constant c as r goes to ∞. After passing to the limit one is left with
m(g) = c( M (|∇ψ|2 + S4 |ψ|2 )vg , which implies m(g) ≥ 0. The equality
m(g) = 0 requires ψ to be parallel for any ψ constructed this way, in par-
ticular the spinor bundle of (M n , g) must be trivialized by parallel spinors,
from which the identity (M n , g) = (Rn , can) can be deduced. An alternative
spinorial proof but with supplementary assumptions on the dimension or the
Weyl tensor has been given by B. Ammann and E. Humbert [21] using the
Green’s operators associated to the Dirac operator, see Section 8.5.
Appendix A
The twistor and Killing spinor equations

In this section we recall basic facts as well as classification results concerning


twistor and Killing spinors on Riemannian spin manifolds. The reader is
invited to refer to [56, 70, 55, 63] for further statements and references.

A.1 Definitions and examples

Definition A.1.1 Let (M n , g) be a Riemannian spin manifold.


i) A twistor-spinor on (M n , g) is a section ψ of ΣM solving

P ψ = 0, (A.1)

where PX ψ := ∇X ψ + n1 X · Dψ for every X ∈ T M .


ii) Given a complex number α, an α-Killing spinor on (M n , g) is a section ψ
of ΣM solving
∇X ψ = αX · ψ (A.2)
for every X ∈ T M . In case α ∈ R (resp. α ∈ iR∗ ) an α-Killing spinor is
called real Killing (resp. imaginary Killing) spinor.
The operator P : Γ(ΣM ) −→ Γ(T ∗ M ⊗ ΣM ) is called the Penrose or twistor
operator. It is obtained as the orthogonal projection of the covariant deriva-
tive onto the kernel of the Clifford multiplication μ : T ∗ M ⊗ ΣM −→ ΣM .
Obviously a section of ΣM is a Killing spinor if and only if it is a twistor-
spinor which is an eigenvector of D. Beware that our definition of real Killing
spinor contains that of parallel spinor, compare [56].
Notes A.1.2
1. The name “Killing spinor” originates from the fact that, if α is real, then
the vector field V defined by g(V, X) := i ψ, X · ψ
for all X ∈ T M , is a
Killing vector field on (M n , g).
2. One could give a slightly more general definition of Killing spinor, requir-
ing (A.2) to hold for a given smooth complex-valued function α on M .

131
132 Appendix A The twistor and Killing spinor equations

If α is real-valued and n ≥ 2, then O. Hijazi has shown [127] that


it must be constant on M , see [134, Prop. 5.12]. On the other hand,
H.-B. Rademacher has proved [209] the existence of (and actually com-
pletely classified) manifolds carrying non-zero α-Killing spinors with
non-constant α ∈ C ∞ (M, iR), see Theorem A.4.5 below.
On 1-dimensional manifolds M it is a simple exercise to show the following:
every section of ΣM is a twistor-spinor, therefore this space is infinite-
dimensional. For any α ∈ C the space of α-Killing spinors on (M, g) :=
(R, can) is C · e−iαt , thus it is 1-dimensional. As for the circle S1 (L) of length
L > 0 and carrying the δ-spin structure with δ ∈ {0, 1} (see Example 1.4.3.1),
it admits a non-zero - and hence 1-dimensional - space of α-Killing spinors if
and only if α ∈ πδL + L Z (in particular α must be real).

Before we proceed to general properties of twistor-spinors in dimension


n ≥ 2, we discuss a few examples.
Examples A.1.3 We describe the twistor spinors on simply-connected
spaceforms.
1. Let (M n , g) := (Rn , can), n ≥ 2, be endowed with its canonical spin
n
structure. It obviously admits a 2[ 2 ] -dimensional space of parallel spinors,
which are the constant sections of Σ(Rn ) ∼ = Rn × Σn . There exists more-
[n ]
over a 2 -dimensional space of non-parallel twistor-spinors, which are of
2

the form ϕx := x · ψ for every x ∈ Rn , where ψ is some parallel spinor


on (Rn , can). Since this space stands in direct sum with that of parallel
spinors, we deduce that the space of twistor-spinors on (Rn , can) is at least
n
2[ 2 ]+1 -dimensional. We shall show that for n ≥ 3 the space Ker(P ) is ac-
n
tually at most (hence here exactly) 2[ 2 ]+1 -dimensional (see Proposition
A.2.1.3.b)), however in dimension n = 2 there are many more twistor-
spinors on Rn (see Proposition A.2.3).
2. Let (M n , g) := (Sn , can), n ≥ 2, be endowed with its canonical metric
(with sectional curvature 1) and spin structure. Since it is a hypersurface
of (Rn+1 , can) with Weingarten-map −IdT M w.r.t. the normal vector field
νx := x we deduce from the Gauss-type formula (1.21) that the restriction
of any parallel spinor (resp. positive half spinor for n odd, see Proposition
1.4.1) onto Sn is a 12 -Killing spinor. Therefore the space of 12 -Killing spinors
n
on (Sn , can) is at least 2[ 2 ] -dimensional. Using again Proposition 1.4.1, it is
easy to show that the restriction of any spinor of the form x → x · ψ, where
ψ is parallel on (Rn+1 , can) (and positive if n is odd), gives a − 12 -Killing
spinor. Therefore the space of − 12 -Killing spinors on (Sn , can) is also at
n
least 2[ 2 ] -dimensional. From Propositions A.2.1.4 and A.4.1.2 below we
n
deduce that both spaces have exactly dimension 2[ 2 ] and that the space
of twistor-spinors is exactly the direct sum of them.
3. Let (M n , g) := (H n , can), n ≥ 2, be endowed with its canonical metric
(with sectional curvature −1) and spin structure. It is a hypersurface of
A.1 Definitions and examples 133

the Minkowski-space (Rn+1 , · , ·

) with Weingarten-map −IdT M w.r.t.


the normal vector field νx := x. In the Lorentzian setting Proposition
1.4.1 has an analog: there exists an isomorphism ΣM | −→ ΣM (or to a
M
double copy of it if n is odd) for which quite the same Gauss-type-formula
as (1.21) holds but for which the relation (1.20) becomes X·iν·ϕ = −X · ϕ
M
| . As a
(or −(X · ⊕ − X · )ϕ if n is odd) for all X ∈ T M and ϕ ∈ ΣM M
M M
consequence the restriction of any parallel spinor (resp. positive half spinor
for n odd) onto H n is a 2i -Killing spinor. Analogously the restriction of
any spinor of the form x → x · ψ, where ψ is parallel on (Rn+1 , · , ·

) (and
positive if n is odd), gives a − 2i -Killing spinor. We deduce from Proposition
A.4.1.2 below that there are no other Killing spinors on (H n , can) than
those constructed and from Proposition A.2.1.3.b) that, if n ≥ 3, then the
space of twistor-spinors is exactly the direct sum of both spaces (for n = 2
see Proposition A.2.3).

Non simply-connected spaceforms may also admit non-zero twistor-spinors.


This is the case if and only if the (or at least some) twistor-spinor on the cor-
responding model space is preserved by the π1 -action, see Proposition 1.4.2.
This condition is not always fulfilled: for example there does not exist any
non-zero imaginary Killing spinor on closed Riemannian spin manifolds (oth-
erwise the Dirac operator would have a purely imaginary eigenvalue, which
cannot be on closed manifolds). In dimension 2 flat tori together with their
trivial spin structure are the only closed non simply-connected spaceforms
admitting twistor-spinors, which are then parallel. In dimension 3 flat tori
together with their trivial spin structure are also the only closed flat mani-
folds admitting twistor-spinors [206], however the quotient of S3 through any
of its finite subgroups carries Killing spinors [13] (for lens spaces it has been
proved independently in [84]). The real projective space RPn admits for every
n ≡ 3 (4) non-zero real Killing spinors: in the notations of Corollary 2.1.5,
n−1
if the spin structure is fixed by δ = 0, then there exists a 2 2 -dimensional
space of − 12 -Killing spinors if n ≡ 3 (8) and of 12 -ones if n ≡ 7 (8) respectively
(vice-versa for δ = 1). More generally there exists a formula for the dimen-
sion of the space of Killing spinors on every (closed) spaceform with positive
curvature [38, Thm. 3]. However, up to the knowledge of the author, there
does not exist any full classification of complete (even closed) flat manifolds
admitting parallel spinors in dimension n ≥ 4.
Non conformally flat examples are much more involved, see for instance
[170] where the authors construct on every Cn a half conformally flat (non
conformally flat) metric carrying a non-zero space of twistor-spinors. For the
reader interested in twistor-spinors on singular spaces such as orbifolds we
suggest [57].
134 Appendix A The twistor and Killing spinor equations

A.2 Elementary properties of twistor-spinors

The following fundamental results on the twistor-spinor-equation are due to


H. Baum, T. Friedrich and A. Lichnerowicz (see [56] for precise references):
Proposition A.2.1 (see [56]) Let ψ be any twistor-spinor on an n(≥ 2)-
dimensional Riemannian spin manifold (M n , g). Then the following holds:
1. For any conformal change g := e2u g of metric on M n ,

P = e 2 ◦ P ◦ e− 2 ,
u u

u
where P := Pg . In particular e 2 ψ is a twistor-spinor on (M n , g).
2. If S denotes the scalar curvature of (M n , g) then

nS
D2 ψ = ψ. (A.3)
4(n − 1)

3. If n ≥ 3 then:
a) for every X ∈ T M ,

n  1 S 
∇X (Dψ) = − Ric(X) · ψ + X ·ψ . (A.4)
n−2 2 4(n − 1)
n
b) dim(Ker(P )) ≤ 2[ 2 ]+1 .
c) The zero-set of ψ is either discrete in M n or M n itself.
d) If (M n , g) is Einstein with S = 0 then ψ is the sum of two non-parallel
Killing spinors.
e) If |ψ| is a non-zero constant then (M n , g) is Einstein. Moreover either
S = 0 and ψ is parallel or S > 0 and ψ is the sum of two real non-
parallel Killing spinors.
4. If M n is closed then Ker(P ) is finite dimensional. In the case ψ = 0 if
furthermore S is constant then either S = 0 and ψ is parallel or S > 0
and ψ is the sum of two real non-parallel Killing spinors.
Proof :
1. For any ϕ ∈ Γ(ΣM ), f ∈ C ∞ (M ) and X ∈ T M one has

1
PX (f ϕ) = ∇X (f ϕ) + X · D(f ϕ)
n
(1.11)
= X(f )ϕ + f ∇X ϕ
1 f
+ X · grad(f ) · ϕ + X · Dϕ
n n
1
= X(f )ϕ + X · grad(f ) · ϕ + f PX ϕ. (A.5)
n
A.2 Elementary properties of twistor-spinors 135

We deduce from (1.17) and (1.18) that

1
P X ϕ = ∇X ϕ + X·Dϕ
n
1 X(u)
= ∇X ϕ − X · gradg (u) · ϕ − ϕ
2 2
e−u n−1
+ X·(Dϕ + gradg (u) · ϕ)
n 2
1 X(u)
= PX ϕ − X · gradg (u) · ϕ − ϕ, (A.6)
2n 2
so that
u u
u (A.5) e 2 e2 u
P X (e 2 ϕ) = X(u)ϕ + X·gradg (u)· ϕ + e 2 P X ϕ
2 2n
u u
(A.6) e 2 e2
= X(u)ϕ + X·gradg (u)· ϕ
2 2n
u 1 X(u)
+e 2 (PX ϕ − X · gradg (u) · ϕ − ϕ)
2n 2
u
= e 2 PX ϕ,

which shows 1.
2. Let X, Y ∈ Γ(T M ), then

1 1
∇X ∇Y ψ = − ∇X Y · Dψ − Y · ∇X Dψ,
n n
from which we deduce

RX,Y ψ = ∇[X,Y ] ψ − [∇X , ∇Y ]ψ
1
= (Y · ∇X Dψ − X · ∇Y Dψ). (A.7)
n

Let {ej }1≤j≤n be a local orthonormal basis of T M . From (1.9) we obtain

1
n
1
Ric(X) · ψ = (ej · ej · ∇X Dψ − ej · X · ∇ej Dψ)
2 n j=1
1
= (−(n − 2)∇X Dψ + X · D2 ψ). (A.8)
n
Hence

1
n
1
− Sψ = ej · Ric(ej ) · ψ
2 2 j=1
136 Appendix A The twistor and Killing spinor equations

1
n
(A.8)
= (−(n − 2)ej · ∇ej Dψ + ej · ej · D2 ψ)
n j=1
2(n − 1) 2
= − D ψ
n
which shows 2.
3. Assume n ≥ 3.
a) Coming back to (A.8) using (A.3) we obtain

1 n−2 S
Ric(X) · ψ = − ∇X Dψ + X ·ψ
2 n 4(n − 1)

which is the result.


b) It follows from (A.4) that ψ is a twistor-spinor if and only if the section
ψ ⊕ Dψ of ΣM ⊕ ΣM is parallel w.r.t. the covariant derivative
 1 
∇TX (ϕ1 ⊕ ϕ2 ) := ∇X ϕ1 + X · ϕ2
n
 n 1 S 
⊕ ( Ric(X) · ϕ1 − X · ϕ1 ) + ∇X ϕ2
n−2 2 4(n − 1)
n
for all ϕ1 , ϕ2 ∈ Γ(ΣM ). From rkC (ΣM ) = 2[ 2 ] we conclude.
c) We compute the Hessian of |ψ|2 . Let X, Y ∈ Γ(T M ). From

X(|ψ|2 ) = 2e ( ∇X ψ, ψ
)
2
= − e ( X · Dψ, ψ
)
n
one has
2
Hess(|ψ|2 )(X, Y ) = − e ( Y · ∇X Dψ, ψ
+ Y · Dψ, ∇X ψ
)
n
(A.4) 1
= e( Y · Ric(X) · ψ, ψ
)
n−2
S
− e( Y · X · ψ, ψ
)
2(n − 1)(n − 2)
2
+ e( Y · Dψ, X · Dψ
)
n2
|ψ|2
= − ric(X, Y )
n−2
 S|ψ|2 2|Dψ|2 
+ + g(X, Y ). (A.9)
2(n − 1)(n − 2) n2
A.2 Elementary properties of twistor-spinors 137

2|Dψ|2
If ψp = 0 then Hess(|ψ|2 )p = n2 p gp . In the case ψ = 0 one must have
(Dψ)p = 0 (otherwise the ∇T -parallel section ψ ⊕Dψ would vanish at p
and hence identically), therefore the Hessian of |ψ|2 is positive definite
at p and the result follows.
d) If (M n , g) is Einstein then (A.4) becomes

n S S
∇X Dψ = (− + )X · ψ
n − 2 2n 4(n − 1)
S
=− X ·ψ
4(n − 1)

for every X ∈ T M . In case S = 0 the spinor ψ can be written as


ψ = ψ1 + ψ−1 where

1 1
ψ±1 := ψ ± Dψ
2 λ

nS
and with λ := 4(n−1) (if S < 0 the square root may be chosen arbi-
trarily since changing its sign just exchanges the roles of ψ1 and ψ−1 ).
We show that ψ±1 is a ∓ nλ -Killing spinor on (M n , g): for every X ∈ T M ,

1 1
∇X ψ±1 = (∇X ψ ± ∇X Dψ)
2 λ
1 1 1 λ2 
= − X · Dψ ± (− X · ψ)
2 n λ n
λ 1
= ∓ X · (ψ ± Dψ)
2n λ
λ
= ∓ X · ψ±1 ,
n

which shows d).


e) If |ψ| is a non-zero constant then (A.9) implies
 S 2(n − 2)|Dψ|2 
ric = + g,
2(n − 1) n2 |ψ|2

that is, (M n , g) is Einstein. Moreover from the latter equation the scalar
curvature of (M n , g) is then given by

4(n − 1) |Dψ|2
S= · ≥ 0. (A.10)
n |ψ|2

> 0 then using d) we deducethat ψ = ψ1 + ψ−1 where ψ±1 is


If S
a ∓ 4n(n−1)
S
-Killing spinor with 4n(n−1) ∈ R. In case S = 0 the
S
138 Appendix A The twistor and Killing spinor equations

identity (A.10) requires Dψ = 0 and hence ∇ψ = 0, i.e., ψ is parallel


on (M n , g). This proves e).
4. Assume (M n , g) to be closed and n ≥ 2. From

1
|∇ϕ|2 = |P ϕ|2 + |Dϕ|2 (A.11)
n
and

(A.11) 1 2
P ∗P = ∇∗ ∇ − D
n
(1.15) n−1 ∗ S
= ∇ ∇− Id (A.12)
n 4n
the operator P ∗ P is elliptic, hence its kernel is finite-dimensional. If fur-
thermore ψ = 0 and S is constant then integrating the Hermitian product
of (A.3) with ψ one obtains
 
nS
|ψ| vg =
2
D2 ψ, ψ
vg
4(n − 1) M
M

= |Dψ|2 vg ,
M

which shows S ≥ 0. On the other hand (A.3) already stands for the
limiting-case in T. Friedrich’s inequality (3.1), so that ψ must either be
parallel (in case S = 0) or the sum of two real Killing spinors (in case
S > 0). This shows 4. and concludes the proof. 

Note A.2.2 Actually Proposition A.2.1.4 implies that Proposition


A.2.1.3.b) and c) hold on closed M in dimension n = 2 as well: on the
one hand we deduce that the only compact orientable surfaces admitting
twistor-spinors are S2 and T2 carrying any conformal class, the latter one
being endowed with its trivial spin structure. For S2 (resp. T2 ) that space
is 4-dimensional (resp. 2-dimensional), corresponding to the direct sum of
the space of 12 -Killing spinors with that of − 12 -ones for the canonical metric
(resp. to the space of parallel spinors for any flat metric in the conformal
class). On the other hand the sum of two Killing spinors on S2 has at most
one zero.
For R2 and H2 (see Note A.2.2 for closed M 2 ) one can again make the
space of twistor-spinors explicit, however that space turns out to be infinite-
dimensional:

Proposition A.2.3 Let M be any non-empty connected open subset of R2


carrying its canonical conformal class and spin structure. Then the space of
A.2 Elementary properties of twistor-spinors 139

twistor-spinors of M for any metric in this conformal class is isomorphic


to the direct sum of the space of holomorphic with that of anti-holomorphic
functions on M . In particular the space of twistor-spinors on R2 and H2
respectively is infinite-dimensional.

Proof : Since the twistor-spinor-equation is conformally invariant (see Propo-


sition A.2.1.1) we may assume that g is the canonical flat metric on M . Let
{ϕ+ , ϕ− } be a basis of parallel spinors on M w.r.t. g such that ie1 · e2 · ϕ± =
±ϕ± where {e1 , e2 } denotes the canonical basis of R2 . Then there exist func-
tions f+ , f− : M −→ C such that ψ = f+ ϕ+ + f− ϕ− . We compute P ψ: for
every X ∈ T M ,

1
PX ψ = X(f+ )ϕ+ + X(f− )ϕ− + X · (df+ · ϕ+ + df− · ϕ− ).
2
For the Kähler structure J associated to g and the orientation of M one has
however


2
X · Y · ϕ± = g(X, ej )g(Y, ek )ej · ek · ϕ
j,k=1


2
= g(X, ej )g(Y, ej )ej · ej · ϕ
j=1
+(g(X, e1 )g(Y, e2 ) − g(X, e2 )g(Y, e1 ))e1 · e2 · ϕ±
= −g(X, Y )ϕ − g(X, J(Y ))e1 · e2 · ϕ±
= (−g(X, Y ) ± ig(X, J(Y )))ϕ±
= −2g(X, p± (Y ))ϕ± ,

where p± (X) := 12 (X ∓ iJ(X)). We deduce that

PX ψ = X(f+ )ϕ+ − g(X, p+ (df+ ))ϕ+


+X(f− )ϕ− − g(X, p− (df− ))ϕ−
= g(X, p− (df+ ))ϕ+ + g(X, p+ (df− ))ϕ− .

Therefore P ψ = 0 if and only if p± (df∓ ) = 0, that is, if and only if


f+ is anti-holomorphic and f− is holomorphic. From (H2 , canH2 ) =
({z ∈ C s.t. |z| < 1}, (1−|z|
4
2 )2 canR2 ) we conclude the proof. 

Note that Proposition A.2.3 together with Note A.2.2 imply in particular
that Proposition A.2.1.3.c) still holds in dimension n = 2, since holomorphic
and anti-holomorphic functions on a surface vanish either on a discrete subset
or identically.
140 Appendix A The twistor and Killing spinor equations

Corollary A.2.4 ([56]) Let (M n , g) be an n(≥ 3)-dimensional Rieman-


nian spin manifold carrying a non-zero twistor-spinor ψ. Then Zψ := {x ∈
n
M | ψ(x) = 0} is discrete in M and (M := M n \ Zψ , g := |ψ|1
4 g) admits a

real Killing spinor, which is parallel if Zψ = ∅.

Proof : The statement on the zero set of ψ has been proved in Proposition
ψ
A.2.1.3.c). From Proposition A.2.1.1 the spinor φ := |ψ| is a twistor-spinor
n
on (M , g). In dimension n ≥ 3 since it has constant norm it is the sum of
two real Killing spinors (Proposition A.2.1.3.e)); furthermore

(1.18) n − 1 grad(|ψ|2 )
Dφ = |ψ|2 (Dφ − · φ)
2 |ψ|2
grad(|ψ|) 1 n − 1 grad(|ψ|2 )
= |ψ|2 (− ·ψ+ Dψ − · φ)
|ψ| 2 |ψ| 2 |ψ|2
n
= |ψ|Dψ − grad(|ψ|2 ) · φ,
2

so that, for any p ∈ Zψ , |D φ|(x) −→ 0 and by (A.10) for g and φ one


x→p
obtains Sg (x) −→ 0 (both w.r.t. the topology given by g on M ). Applying
x→p
again Proposition A.2.1.3.e), since Sg is constant it must vanish identically,
n
hence φ is parallel on (M , g) as soon as Zψ = ∅. 

Note that the equivalent statement in dimension n = 2 does not hold because
of Proposition A.2.3.

A.3 Classification results for manifolds


with twistor-spinors

Corollary A.2.4 induces a dichotomy in the classification of n(≥ 3)-


dimensional Riemannian spin manifolds M carrying a non-zero twistor-spinor
ψ: either Zψ = ∅ and then up to conformal change of metric M belongs to
the class of manifolds admitting Killing spinors (which is studied in greater
detail in Section A.4), or Zψ = ∅. In the latter case and for closed M , using
the solution to the Yamabe problem about the existence of a constant scalar
curvature metric in a conformal class A. Lichnerowicz showed:
Theorem A.3.1 (A. Lichnerowicz [178]) Let (M n , g), n ≥ 2, be a closed
Riemannian spin manifold carrying a non-trivial twistor-spinor ψ with non-
empty zero-set Zψ . Then |Zψ | = 1 and (M n , g) is conformally equivalent to
(Sn , can).
A relatively simple proof of Theorem A.3.1 can be found in [172].
A.4 Classification results for manifolds with Killing spinors 141

For general M W. Kühnel and H.-B. Rademacher proved that the Ricci-flat
1
metric |ψ| 4 g on M
n
\ Zψ is either flat or locally irreducible, more precisely:

Theorem A.3.2 (W. Kühnel and H.-B. Rademacher [171]) Let


(M n , g) be a simply-connected Riemannian spin manifold carrying a non-
trivial twistor-spinor ψ with non-empty zero-set Zψ and assume that the
metric is not conformally flat. Then the following holds:
1. Every non-zero twistor-spinor on (M n , g) vanishes exactly at Zψ .
n
2. For N := dim(Ker(P )) and the reduced holonomy group Hol := Hol(M , g)
n
of the Ricci-flat metric g := |ψ|4 g on M := M n \ Zψ one has one of the
1

following:
a) n = 2m ≥ 4, Hol = SUm and N = 2.
b) n = 4m ≥ 8, Hol = Spm and N = m + 1.
c) n = 7, Hol = G2 and N = 1.
d) n = 8, Hol = Spin7 and N = 1.
Theorem A.3.2, a proof of which can be found in the beautiful paper
[172], actually requires Mc.K. Wang’s classification of manifolds with non-
zero parallel spinors, see Theorem A.4.2 below. Besides, we mention that up
to now no example with reduced holonomy of type b), c) or d) has been
described (an example with Hol = SUm is constructed in [170]).

A.4 Classification results for manifolds


with Killing spinors

We now come to the geometric properties specifically implied by the existence


of a non-zero Killing spinor (Definition A.1.1.ii)).
Proposition A.4.1 Let (M n , g) be an n(≥ 2)-dimensional Riemannian spin
manifold admitting a non-zero α-Killing spinor ψ for some α ∈ C.
1. The zero-set of ψ is empty. If furthermore α is real then |ψ| is con-
stant on M .
n
2. The space of α-Killing spinors on (M n , g) is at most 2[ 2 ] -dimensional.
3. The manifold (M n , g) is Einstein with scalar curvature S = 4n(n − 1)α2 .
In particular α must be real or purely imaginary.
Proof : By definition ψ is an α-Killing spinor if and only if it is a parallel
section of ΣM w.r.t. the covariant derivative X → ∇X − αX·; moreover that
covariant derivative is metric as soon as α is real. This shows 1. and 2.
Assuming n ≥ 3 it follows from (A.4) that, for every X ∈ T M ,

S
Ric(X) · ψ = (2(n − 2)α2 + )X · ψ
2(n − 1)
142 Appendix A The twistor and Killing spinor equations

(remember that ψ is a twistor-spinor satisfying Dψ = −nαψ). Since ψ has no


zero we obtain that (M n , g) is Einstein with scalar curvature S = 4n(n−1)α2 .
In dimension n = 2 the equation (A.7) for ψ is of the form

RX,Y ψ = α2 (X · Y − Y · X) · ψ,

hence comparing with (1.8) we obtain S = 8α2 , which concludes the


proof of 3. 

In particular Myers’ theorem implies that a complete Riemannian spin man-


ifold without boundary and carrying a non-zero real Killing spinor must be
compact.
Complete simply-connected Riemannian spin manifolds (M n , g) carrying a
non-zero space of α-Killing spinors have been classified by Mc.K. Wang [232]
for α = 0, C. Bär [37] for α ∈ R∗ and H. Baum [50] for α ∈ iR∗ respectively.
In the first case, a parallel spinor must be a fixed point of the action of the
lift of the reduced holonomy group to Spinn . Excluding the trivial example
(M n , g) = (Rn , can), which has a maximal number of linearly independent
parallel spinors, as well as local products (products of manifolds with parallel
spinors carry themselves parallel spinors), the classification can be deduced
from Berger-Simons’ list of Riemannian holonomy groups.

Theorem A.4.2 (McK. Wang[232]) Let (M n , g) be an (n ≥ 2)-


dimensional simply-connected complete irreducible Riemannian spin
manifold without boundary. Let N denote the dimension of Ker(∇). Then
the manifold (M n , g) carries a non-zero parallel spinor if and only if its
reduced holonomy group Hol := Hol(M, g) belongs to the following list:
a) Hol = SUm , n = 2m ≥ 4, and in that case N = 2.
b) Hol = Spm , n = 4m ≥ 8, and in that case N = m + 1.
c) Hol = G2 , n = 7, and in that case N = 1.
d) Hol = Spin7 , n = 8, and in that case N = 1.
There also exists a classification in the non-flat non-simply-connected case
in terms of lifts the holonomy group to the spin group, see [203] where the
proof of Theorem A.4.2 can also be found.
The classification when α ∈ R∗ relies on Mc.K. Wang’s one using the
following clever remark due to C. Bär and based on a geometric construction
by S. Gallot (see reference in [37]): a spinor field is a 12 -Killing spinor on the
manifold (M n , g) if and only if the induced spinor field on its Riemannian
cone (M × R∗+ , t2 g ⊕ dt2 ) is parallel. Hence C. Bär proved:
Theorem A.4.3 (C. Bär [37]) Let (M n , g) be an (n ≥ 2)-dimensional
simply-connected closed Riemannian spin manifold. Let p (resp. q) denote
the dimension of the space of 12 - (resp. − 12 -) Killing spinors on (M n , g).
Then the manifold (M n , g) carries up to scaling a non-zero ± 21 -Killing
A.4 Classification results for manifolds with Killing spinors 143

spinor if and only if it is either the round sphere (Sn , can) (in which case
n n
(p, q) = (2[ 2 ] , 2[ 2 ] )) or one of the following:
a) (4m + 1)-dimensional Einstein-Sasaki, m ≥ 1, and in that case (p, q) =
(1, 1).
b) (4m+3)-dimensional Einstein-Sasaki but not 3-Sasaki, m ≥ 2, and in that
case (p, q) = (0, 2).
c) (4m+3)-dimensional 3-Sasaki, m ≥ 2, and in that case (p, q) = (0, m+2).
d) 6-dimensional nearly Kähler non-Kähler, and in that case (p, q) = (1, 1).
e) 7-dimensional with a nice 3-form φ satisfying ∇φ = ∗φ but not Sasaki,
and in that case (p, q) = (0, 1).
f) 7-dimensional Sasaki but not 3-Sasaki, and in that case (p, q) = (0, 2).
g) 7-dimensional 3-Sasaki, and in that case (p, q) = (0, 3).

For the definitions of 3-Sasaki structures and nice forms as well as the proof
of Theorem A.4.3 we refer to [37]. Parts of this classification had already been
obtained in [89, 129, 90, 91, 92, 117, 87]. As an interesting fact, two higher
eigenvalues of (n = 4m + 3)-dimensional 3-Sasaki manifolds can be explicitly
computed in terms of the  scalar curvature:A. Moroianu showed [199] that on
such manifolds both − 4(n−1) nS
− 1 and nS
4(n−1) + 2 are Dirac eigenvalues
with multiplicities at least 3m and m respectively. The proof relies on a clever
combination of the Killing vector fields provided by the 3-Sasaki structure
and the Killing spinors.
In the last case (α ∈ iR∗ ) the classification turns out to rely on totally
different arguments. Studying in detail the level sets of the length function
of an imaginary Killing spinor H. Baum proved the following theorem, which
relies on Theorem A.4.2 but where the assumption π1 (M ) = 1 turns out not
to be necessary.
Theorem A.4.4 (H. Baum [50]) Let (M n , g) be an (n ≥ 2)-dimensional
connected complete Riemannian spin manifold without boundary. Then
(M n , g) admits a non-trivial α-Killing spinor with α ∈ iR∗ if and only if
it is isometric to a warped product of the form

(N × R, e4iαt h ⊕ dt2 ),

where (N n−1 , h) is a complete connected Riemannian spin manifold carrying


a non-zero parallel spinor.

Of course the n-dimensional real hyperbolic space can be obtained as a


warped product of this form (take (N, h) = (Rn−1 , can)); in the disk model,
this corresponds to the foliation by horospheres tangential to a fixed point
on the boundary at infinity.
It was noticed by O. Hijazi, S. Montiel and A. Roldán [140] that the geo-
metric part of Theorem A.4.4 - i.e., that (M n , g) must be a pseudo-hyperbolic
space - follows from a classical argument by Yoshihiro Tashiro (see reference
144 Appendix A The twistor and Killing spinor equations

in [140]), namely from the existence of a smooth non-zero real-valued function


f on M such that
Hess(f ) − f g = 0.
Here, up to rescaling g, the function f := |ψ|2 , where ψ is a non-zero
α-Killing spinor on (M n , g), satisfies that equation (use (A.9) when n ≥ 3).
Nevertheless this argument does not describe the correspondence between
spinor fields on M and those on the warped product, see [50] for a rigorous
treatment of that point.
Theorem A.4.4 generalizes to the situation where the constant α is replaced
by a smooth imaginary-valued function, in which case a similar statement on
the structure of the underlying manifold holds.

Theorem A.4.5 (H.-B. Rademacher [209]) Let (M n , g) be an n(≥ 2)-


dimensional connected complete Riemannian spin manifold without boundary.
For a given non-zero α ∈ C ∞ (M, iR) assume the existence of a non-zero
section ψ of ΣM satisfying

∇X ψ = αX · ψ

for all X ∈ T M . Then (M n , g) is isometric either to the real hyperbolic space


of constant sectional curvature 4α2 (in particular α must be constant) or to a
t
warped product of the form (N × R, e4i 0 α(s)ds h ⊕ dt2 ), where (N n−1 , h) is a
complete connected Riemannian spin manifold admitting a non-zero parallel
spinor and α ∈ C ∞ (R, iR).
Conversely, for any given 
α ∈ C ∞ (R, iR) and (N n−1 , h) as above, the
4i 0t α(s)ds
warped product (N ×R, e h⊕dt2 ) admits a non-zero section ψ of ΣM
satisfying ∇X ψ = αX · ψ for all X ∈ T M , where α is extended by a constant
onto the N -factor. Interestingly enough, there exist compact quotients ad-
mitting such spinors for some necessarily non-constant α’s, see [209, Thm. 1]
and references therein. The proof of Theorem A.4.5 relies on the classification
of complete Riemannian manifolds carrying a non-isometric conformal closed
Killing field, see [209] for details.
Bibliography

1. I. Agricola, B. Ammann, T. Friedrich, A comparison of the eigenvalues of the


Dirac and Laplace operators on a two-dimensional torus, manuscr. math. 100
(1999), no. 2, 231–258.
2. I. Agricola, T. Friedrich, Upper bounds for the first eigenvalue of the Dirac
operator on surfaces, J. Geom. Phys. 30 (1999), no. 1, 1–22.
3. I. Agricola, T. Friedrich, M. Kassuba, Eigenvalue estimates for Dirac operators
with parallel characteristic torsion, Diff. Geom. Appl. 26 (2008), no. 6, 613–624.
4. B. Alexandrov, The first eigenvalue of the Dirac operator on locally reducible
Riemannian manifolds, J. Geom. Phys. 57 (2007), no. 2, 467–472.
5. B. Alexandrov, G. Grantcharov, S. Ivanov, An estimate for the first eigenvalue
of the Dirac operator on compact Riemannian spin manifold admitting a parallel
one-form, J. Geom. Phys. 28 (1998), no. 3-4, 263–270.
6. , The Dolbeault operator on Hermitian spin surfaces, Ann. Inst. Fourier
51 (2001), no. 1, 221–235.
7. B. Ammann, Spin-Strukturen und das Spektrum des Dirac-Operators, Doktorar-
beit, Universität Freiburg, 1998.
8. , The Dirac operator on collapsing S1 -bundles, Sém. Théor. Spec. Géom.
Inst. Fourier Grenoble 16 (1998), 33–42.
9. , The Willmore conjecture for immersed tori with small curvature inte-
gral, manuscr. math. 101 (2000), no. 1, 1–22.
10. , Spectral estimates on 2-tori, arXiv:math/0101061 (2001).
11. , Analysis auf Mannigfaltigkeiten, Indextheorie und Spin-Geometrie, Lec-
ture Notes (2002/2003), available at http://www.mathematik.uni-regensburg.
de/ammann.
12. , A spin-conformal lower bound of the first positive Dirac eigenvalue,
Diff. Geom. Appl. 18 (2003), no. 1, 21–32.
13. , A Variational Problem in Conformal Spin Geometry, Habilitation the-
sis, Universität Hamburg, 2003.
14. , Dirac eigenvalue estimates on 2-tori, J. Geom. Phys. 51 (2004), no. 3,
372–386.
15. B. Ammann, C. Bär, The Dirac operator on nilmanifolds and collapsing circle
bundles, Ann. Glob. Anal. Geom. 16 (1998), no. 3, 221–253.
16. , Dirac eigenvalues and total scalar curvature, J. Geom. Phys. 33 (2000),
no. 3-4, 229–234.
17. , Dirac eigenvalue estimates on surfaces, Math. Z. 240 (2002), 423–449.
18. , The Einstein-Hilbert Action as a Spectral Action, in: F. Scheck,
W. Werner, H. Upmeier (Eds.): Noncommutative Geometry and the Standard
Model of Elementary Particle Physics, LNP 596, Springer-Verlag, 75–108 (2002).
19. B. Ammann, M. Dahl, E. Humbert, Surgery and harmonic spinors, Adv. Math.
220 (2009), no. 2, 523–539.

145
146 Bibliography

20. , La dimension du noyau de l’opérateur de Dirac, to appear in Publ. Inst.


Élie Cartan de Nancy.
21. B. Ammann, E. Humbert, Positive mass theorem for the Yamabe problem on
Spin manifolds, Geom. Funct. Anal. 15 (2005), no. 3, 567–576.
22. , The first conformal Dirac eigenvalue on 2-dimensional tori, J. Geom.
Phys. 56 (2006), no. 4, 623–642.
23. B. Ammann, E. Humbert, B. Morel, Mass endomorphism and spinorial Yamabe
type problems on conformally flat manifolds, Comm. Anal. Geom. 14 (2006),
no. 1, 163–182.
24. B. Ammann, P. Jammes, The supremum of conformally covariant eigenvalues
in a conformal class, arXiv:0708.0529v1 (2007).
25. B. Ammann, C. Sprouse, Manifolds with small Dirac eigenvalues are nilmani-
folds, Ann. Glob. Anal. Geom. 31 (2007), no. 4, 409–425.
26. N. Anghel, Extrinsic upper bounds for eigenvalues of Dirac-type operators, Proc.
Amer. Math. Soc. 117 (1993), no. 2, 501–509.
27. M.F. Atiyah, V.K. Patodi, I.M. Singer, Spectral asymmetry and Riemannian
geometry I, II and III, Math. Proc. Cambridge Philos. Soc. 77 (1975), 43–69;
78 (1975), 405–432; 79 (1975), 71–99.
28. M.F. Atiyah, I.M. Singer, The index of elliptic operators I, Ann. of Math. 87
(1968), 484–530.
29. P.D. Baier, Über den Diracoperator auf Mannigfaltigkeiten mit Zylinderenden,
Diplomarbeit, Universität Freiburg, 1997.
30. W. Ballmann, J. Brüning, On the spectral theory of manifolds with cusps,
J. Math. Pures Appl. 80 (2001), no. 6, 593–625.
31. , On the spectral theory of surfaces with cusps, in: Geometric analysis
and nonlinear partial differential equations, 13–37, Springer, Berlin, 2003.
32. W. Ballmann, J. Brüning, G. Carron, Regularity and index theory for Dirac-
Schrödinger systems with Lipschitz coefficients, J. Math. Pures Appl. (9) 89
(2008), no. 5, 429–476.
33. C. Bär, Das Spektrum von Dirac-Operatoren, Doktorarbeit, Universität Bonn,
Bonner Math. Schr. 217 (1991).
34. , The Dirac operator on homogeneous spaces and its spectrum on 3-
dimensional lens spaces, Arch. Math. 59 (1992), 65–79.
35. , Lower eigenvalue estimates for Dirac operators, Math. Ann. 293 (1992),
no. 1, 39–46.
36. , Upper eigenvalue estimates for Dirac operators, Ann. Glob. Anal.
Geom. 10 (1992), no. 2, 171–177.
37. , Real Killing spinors and holonomy, Comm. Math. Phys. 154 (1993),
no. 3, 509–521.
38. , The Dirac operator on space forms of positive curvature, J. Math. Soc.
Japan 48 (1996), 69–83.
39. , Metrics with Harmonic Spinors, Geom. Funct. Anal. 6 (1996), 899–942.
40. , Extrinsic Bounds for Eigenvalues of the Dirac Operator, Ann. Glob.
Anal. Geom. 16 (1998), 573–596.
41. , The Dirac Operator on Hyperbolic Manifolds of Finite Volume, J. Diff.
Geom. 54 (2000), 439–488.
42. , The Spectrum of the Dirac Operator, in: J.-P. Bourguignon et al (Eds.):
Dirac Operators, Yesterday and Today, International Press, 145–162 (2005).
43. , Introduction to spin geometry, Oberwolfach Report 53 (2006), 3135–
3137.
44. , Spectral Bounds for Dirac Operators on Open Manifolds,
arXiv:0810.5598v1 (2008).
45. C. Bär, M. Dahl, Surgery and the Spectrum of the Dirac Operator, J. reine
angew. Math. 552 (2002), 53–76.
Bibliography 147

46. , The First Dirac Eigenvalue on Manifolds with Positive Scalar Curva-
ture, Proc. Amer. Math. Soc. 132 (2004), 3337–3344.
47. C. Bär, P. Gauduchon, A. Moroianu, Generalized cylinders in semi-Riemannian
and Spin geometry, Math. Z. 249 (2005), no. 3, 545–580.
48. C. Bär, P. Schmutz, Harmonic spinors on Riemann surfaces, Ann. Glob. Anal.
Geom. 10 (1992), no. 3, 263–273.
49. H. Baum, Spin-Strukturen und Dirac-Operatoren über pseudo-Riemannschen
Mannigfaltigkeiten, Dissertation A, Humboldt-Universität zu Berlin, 1980,
Teubner-Texte zur Mathematik 41 (1981), Teubner Verlag Leipzig.
50. , Complete Riemannian manifolds with imaginary Killing spinors, Ann.
Glob. Anal. Geom. 7 (1989), no. 3, 205–226.
51. , An upper bound for the first eigenvalue of the Dirac operator on compact
spin manifolds, Math. Z. 206 (1991), no. 3, 409–422.
52. , Eigenvalue estimates for Dirac operators coupled to instantons, Ann.
Glob. Anal. Geom. 12 (1994), no. 2, 193–209.
53. , A remark on the spectrum of the Dirac operator on pseudo-
Riemannian spin manifolds, preprint SFB 288 136 (1994), available at
http://www.mathematik. hu-berlin.de/∼baum
54. H. Baum, T. Friedrich, Spektraleigenschaften des Dirac-Operators. Die Funda-
mentallösung seiner Wärmeleitungsgleichung und die Asymptotenentwicklung
seiner Zeta-Funktion, J. Diff. Geom. 15 (1980), 1–26.
55. , Eigenvalues of the Dirac operator, twistors and Killing spinors on Rie-
mannian manifolds, Clifford algebras and spinor structures, 243–256, Math.
Appl. 321, Kluwer Acad. Publ., Dordrecht, 1995.
56. H. Baum, T. Friedrich, R. Grunewald, I. Kath, Twistor and Killing spinors
on Riemannian manifolds, Teubner-Texte zur Mathematik, Band 124, Teubner-
Verlag Stuttgart/Leipzig 1991.
57. F.A. Belgun, N. Ginoux, H.-B. Rademacher, A singularity theorem for twistor-
spinors, Ann. Inst. Fourier 57 (2007), no. 4, 1135–1159.
58. N. Berline, E. Getzler, M. Vergne, Heat kernels and Dirac operators,
Grundlehren Text Editions, Springer-Verlag, 2003.
59. B. Booß-Bavnbek, K.P. Wojciechowski, Elliptic boundary problems for Dirac
operators, Mathematics: Theory & Applications, Birkhäuser Boston (1993).
60. M. Bordoni, Spectral estimates for Schrödinger and Dirac-type operators on
Riemannian manifolds, Math. Ann. 298 (1994), no. 4, 693–718.
61. M. Bordoni, O. Hijazi, Eigenvalues of the Kählerian Dirac operator, Lett. Math.
Phys. 58 (2001), no. 1, 7–20.
62. J.-P. Bourguignon, P. Gauduchon, Spineurs, opérateurs de Dirac et variations
de métriques, Comm. Math. Phys. 144 (1992), no. 3, 581–599.
63. J.-P. Bourguignon, O. Hijazi, J.-L. Milhorat, A. Moroianu, A Spinorial approach
to Riemannian and Conformal Geometry, in preparation.
64. T. Branson, O. Hijazi, Vanishing theorems and eigenvalue estimates in Rieman-
nian spin geometry, Int. J. Math. 8 (1997), no. 7, 921–934.
65. U. Bunke, The spectrum of the Dirac operator on the hyperbolic space, Math.
Nachr. 153 (1991), 179–190.
66. , Upper bounds of small eigenvalues of the Dirac operator and isometric
immersions, Ann. Glob. Anal. Geom. 9 (1991), no. 2, 109–116.
67. , On the gluing problem for the η-invariant, J. Diff. Geom. 41 (1995),
no. 2, 397–448.
68. M. Cahen, A. Franc, S. Gutt, Spectrum of the Dirac operator on complex pro-
jective space P2q−1 (C), Lett. Math. Phys. 18 (1989), no. 2, 165–176.
69. , Erratum to: “The spectrum of the Dirac operator on complex projective
space P2q−1 (C)”, Lett. Math. Phys. 32 (1994), no. 4, 365–368.
70. M. Cahen, S. Gutt, L. Lemaire, P. Spindel, Killing spinors, Bull. Soc. Math.
Belg. Sér. A 38 (1986), 75–102.
148 Bibliography

71. D.M.J. Calderbank, P. Gauduchon, M. Herzlich, Refined Kato inequalities and


conformal weights in Riemannian geometry, J. Funct. Anal. 173 (2000), no. 1,
214–255.
72. R. Camporesi, A. Higuchi, On the eigenfunctions of the Dirac operator on
spheres and real hyperbolic spaces, J. Geom. Phys. 20 (1996), no. 1, 1–18.
73. R. Camporesi, E. Pedon, The continuous spectrum of the Dirac operator on
noncompact Riemannian symmetric spaces of rank one, Proc. Amer. Math. Soc.
130 (2002), no. 2, 507–516.
74. I. Chavel, Eigenvalues in Riemannian geometry, Pure and Applied Mathematics
115, Academic Press, 1984.
75. D. Chen, Eigenvalue estimates for the Dirac operator with generalized APS
boundary condition, J. Geom. Phys. 57 (2007), no. 2, 379–386.
76. , Extrinsic eigenvalue estimates of the Dirac operator, arXiv:math/
0701847v1 (2007).
77. D. Chen, H. Sun, Inequalities of eigenvalues for the Dirac operator on compact
complex spin submanifolds in complex projective spaces, Chin. Ann. Math. Ser.
B 29 (2008), no. 2, 165–178.
78. A. Connes, Noncommutative geometry, Academic Press, 1994.
79. M. Dahl, Prescribing eigenvalues of the Dirac operator, manuscr. math. 118
(2005), no. 2, 191–199.
80. , On the space of metrics with invertible Dirac operator, Comment. Math.
Helv. 83 (2008), no. 2, 451–469.
81. P.A.M. Dirac, The Quantum Theory of the Electron, Proc. R. Soc. Lond. A117
(1928), 610–624.
82. S. Farinelli, G. Schwarz, On the spectrum of the Dirac operator under boundary
conditions, J. Geom. Phys. 28 (1998), no. 1-2, 67–84.
83. H.D. Fegan, The spectrum of the Dirac operator on a simply connected compact
Lie group, Simon Stevin 61 (1987), no. 2, 97–108.
84. A. Franc, Spin structures and Killing spinors on lens spaces, J. Geom. Phys. 4
(1987), no. 3, 277–287.
85. T. Friedrich, Der erste Eigenwert des Dirac-Operators einer kompakten, Rie-
mannschen Mannigfaltigkeit nichtnegativer Skalarkrümmung, Math. Nachr. 97
(1980), 117–146.
86. , Zur Abhängigkeit des Dirac-Operators von der Spin-Struktur, Colloq.
Math. 48 (1984), no. 1, 57–62.
87. , The classification of 4-dimensional Kähler manifolds with small eigen-
value of the Dirac operator, Math. Ann. 295 (1993), no. 3, 565–574.
88. , Dirac operators in Riemannian geometry, Graduate Studies in Mathe-
matics 25, American Mathematical Society, 2000.
89. T. Friedrich, R. Grunewald, On the first eigenvalue of the Dirac operator on
6-dimensional manifolds, Ann. Glob. Anal. Geom. 3 (1985), no. 3, 265–273.
90. T. Friedrich, I. Kath, Variétés riemanniennes compactes de dimension 7 admet-
tant des spineurs de Killing, C. R. Acad. Sci. Paris Sér. I Math. 307 (1988), no.
19, 967–969.
91. , Einstein manifolds of dimension five with small first eigenvalue of the
Dirac operator, J. Diff. Geom. 29 (1989), no. 2, 263–279.
92. , 7-dimensional compact Riemannian manifolds with Killing spinors,
Comm. Math. Phys. 133 (1990), no. 3, 543–561.
93. T. Friedrich, E.C. Kim, Some remarks on the Hijazi inequality and generali-
zations of the Killing equation for spinors, J. Geom. Phys. 37 (2001), no. 1-2,
1–14.
94. , Eigenvalue estimates for the Dirac operator in terms of Codazzi tensors,
Bull. Korean Math. Soc. 45 (2008), no. 2, 365–373.
95. T. Friedrich, K.-D. Kirchberg, Eigenvalue estimates for the Dirac operator de-
pending on the Weyl tensor, J. Geom. Phys. 41 (2002), no. 3, 196–207.
Bibliography 149

96. , Eigenvalue estimates of the Dirac operator depending on the Ricci ten-
sor, Math. Ann. 324 (2002), no. 4, 799–816.
97. N. Ginoux, Opérateurs de Dirac sur les sous-variétés, thèse de doctorat,
Université Henri Poincaré, Nancy (2002).
98. , Reilly-type spinorial inequalities, Math. Z. 241 (2002), no. 3, 513–525.
99. , Une nouvelle estimation extrinsèque du spectre de l’opérateur de Dirac,
C. R. Acad. Sci. Paris Sér. I 336 (2003), no. 10, 829–832.
100. , Remarques sur le spectre de l’opérateur de Dirac, C. R. Acad. Sci. Paris
Sér. I 337 (2003), no. 1, 53–56.
101. , Dirac operators on Lagrangian submanifolds, J. Geom. Phys. 52 (2004),
no. 4, 480–498.
102. , A short survey on eigenvalue estimates for the Dirac operator on com-
pact Riemannian spin manifolds, Oberwolfach Report 53 (2006), 3140–3141.
103. , The spectrum of the Dirac operator on SU2 /Q8 , manuscr. math. 125
(2008), no. 3, 383–409.
104. N. Ginoux, G. Habib, A spectral estimate for the Dirac operator on Riemannian
flows, preprint (2009).
105. N. Ginoux, B. Morel, On eigenvalue estimates for the submanifold Dirac oper-
ator, Int. J. Math. 13 (2002), no. 5, 533–548.
106. S. Goette, Äquivariante eta-Invarianten homogener Räume, Doktorarbeit,
Shaker Verlag, Aachen, 1997.
107. , Equivariant eta-Invariants on Homogeneous Spaces, Math. Z. 232
(1999), no. 1, 1–42.
108. , Eta invariants of homogeneous spaces, arXiv:math.DG/0203269 (2002),
to appear in Pure Appl. Math. Q.
109. S. Goette, U. Semmelmann, Spinc Structures and Scalar Curvature Estimates,
Ann. Glob. Anal. Geom. 20 (2001), no. 4, 301–324.
110. , Scalar curvature estimates for compact symmetric spaces, Diff. Geom.
Appl. 16 (2002), no. 1, 65–78.
111. , The point spectrum of the Dirac operator on noncompact symmetric
spaces, Proc. Amer. Math. Soc. 130 (2002), no. 3, 915–923.
112. J. Gracia-Bondı́a, J.C. Várilly, H. Figueroa, Elements of noncommutative geom-
etry, Birkhäuser, 2001.
113. J.-F. Grosjean, E. Humbert, The first eigenvalue of Dirac and Laplace operators
on surfaces, arXiv:math.DG/0609493 (2006).
114. N. Große, On a conformal invariant of the Dirac operator on noncompact man-
ifolds, Ann. Glob. Anal. Geom. 30 (2006) no. 4, 407–416.
115. , The Hijazi inequality on conformally parabolic manifolds,
arXiv:0804.3878 (2008).
116. , On a spin conformal invariant on open manifolds, Doktorarbeit, Uni-
versität Leipzig, 2008.
117. R. Grunewald, Six-dimensional Riemannian manifolds with a real Killing
spinor, Ann. Glob. Anal. Geom. 8 (1990), no. 1, 43–59.
118. C. Guillarmou, S. Moroianu, J. Park, Eta invariant and Selberg Zeta function of
odd type over convex co-compact hyperbolic manifolds, arXiv:0901.4082 (2009).
119. G. Habib, Tenseur d’impulsion-énergie et Feuilletages, thèse de doctorat, Insti-
tut Élie Cartan - Université Henri Poincaré, Nancy (2006).
120. , Eigenvalues of the basic Dirac operator on quaternion-Kähler foliations,
Ann. Glob. Anal. Geom. 30 (2006), no. 3, 289–298.
121. , Eigenvalues of the transversal Dirac operator on Kähler foliations,
J. Geom. Phys. 56 (2006), no. 2, 260–270.
122. , Energy-Momentum Tensor on Foliations, J. Geom. Phys. 57 (2007),
no. 11, 2234–2248.
123. G. Habib, K. Richardson, A brief note on the spectrum of the basic Dirac oper-
ator, arXiv:0809.2406 (2008).
150 Bibliography

124. B. Hanke, T. Schick, Enlargeability and index theory, J. Diff. Geom. 74 (2006),
no. 2, 293–320.
125. M. Herzlich, Théorèmes de masse positive, Sém. Théor. Spec. Géom. Inst.
Fourier Grenoble 16 (1998), 107–126.
126. M. Herzlich, A. Moroianu, Generalized Killing spinors and conformal eigen-
value estimates for Spinc manifolds, Ann. Glob. Anal. Geom. 17 (1999), no. 4,
341–370.
127. O. Hijazi, Opérateurs de Dirac sur les variétés riemanniennes: minoration des
valeurs propres, thèse de doctorat, Ecole Polytechnique - Paris VI (1984).
128. , A conformal lower bound for the smallest eigenvalue of the Dirac op-
erator and Killing spinors, Commun. Math. Phys. 104, (1986) 151–162.
129. , Caractérisation de la sphère par les premières valeurs propres de
l’opérateur de Dirac en dimensions 3, 4, 7 et 8, C. R. Acad. Sci. Paris Sér.
I Math. 303 (1986), no. 9, 417–419.
130. , Première valeur propre de l’opérateur de Dirac et nombre de Yamabe,
C. R. Acad. Sci. Paris Sér. I Math. 313 (1991), no. 12, 865–868.
131. , Eigenvalues of the Dirac operator on compact Kähler manifolds, Comm.
Math. Phys. 160 (1994), no. 3, 563–579.
132. , Lower bounds for the eigenvalues of the Dirac operator, J. Geom. Phys.
16 (1995), 27–38.
133. , Twistor operators and eigenvalues of the Dirac operator, Quaternionic
structures in mathematics and physics (Trieste, 1994), 151–174, Int. Sch. Adv.
Stud. (SISSA), Trieste, 1998.
134. , Spectral properties of the Dirac operator and geometrical structures,
Proceedings of the Summer School on Geometric Methods in Quantum Field
Theory, Villa de Leyva, Colombia, July 12–30, (1999), World Scientific 2001.
135. O. Hijazi, J.-L. Milhorat, Minoration des valeurs propres de l’opérateur de Dirac
sur les variétés spin Kähler-quaternioniennes, J. Math. Pures Appl. (9) 74
(1995), no. 5, 387–414.
136. , Décomposition du fibré des spineurs d’une variété spin Kähler-
quaternionienne sous l’action de la 4-forme fondamentale, J. Geom. Phys. 15
(1995), no. 4, 320–332.
137. , Twistor operators and eigenvalues of the Dirac operator on compact
quaternion-Kähler spin manifolds, Ann. Glob. Anal. Geom. 15 (1997), no. 2,
117–131.
138. O. Hijazi, S. Montiel, Extrinsic Killing spinors, Math. Z. 244 (2003), no. 2,
337–347.
139. O. Hijazi, S. Montiel, A. Roldán, Eigenvalue Boundary Problems for the Dirac
Operator, Comm. Math. Phys. 231 (2002), no. 3, 375–390.
140. , Dirac operators on hypersurfaces of manifolds with negative scalar cur-
vature, Ann. Glob. Anal. Geom. 23 (2003), 247–264.
141. O. Hijazi, S. Montiel, F. Urbano, Spinc geometry of Kähler manifolds and the
Hodge Laplacian on minimal Lagrangian submanifolds, Math. Z. 253 (2006),
no. 4, 821–853.
142. O. Hijazi, S. Montiel, X. Zhang, Dirac Operator on Embedded Hypersurfaces,
Math. Res. Letters 8 (2001), 195–208.
143. , Eigenvalues of the Dirac Operator on Manifolds with Boundary, Com-
mun. Math. Phys. 221 (2001), 255–265.
144. , Conformal Lower Bounds for the Dirac Operator of Embedded Hyper-
surfaces, Asian J. Math. 6 (2002), no. 1, 23–36.
145. O. Hijazi, S. Raulot, Branson’s Q-curvature in Riemannian and Spin Geometry,
SIGMA 3 (2007), 119, 14 p.
146. O. Hijazi, X. Zhang, Lower bounds for the eigenvalues of the Dirac operator.
I. The hypersurface Dirac operator, Ann. Glob. Anal. Geom. 19 (2001), no. 4,
355–376.
Bibliography 151

147. , Lower bounds for the eigenvalues of the Dirac operator. II. The sub-
manifold Dirac operator, Ann. Glob. Anal. Geom. 20 (2001), no. 2, 163–181.
148. N. Hitchin, Harmonic spinors, Adv. Math. 14 (1974), 1–55.
149. P. Jammes, Extrema de valeurs propres dans une classe conforme, Sém. Théor.
Spec. Géom. Inst. Fourier Grenoble 24 (2007), 23–42.
150. M. Jardim, R.F. Leão, Survey on eigenvalues of the Dirac operator and geomet-
ric structures, Int. Math. Forum 3 (2008), no. 1-4, 49–67.
151. S.D. Jung, The first eigenvalue of the transversal Dirac operator, J. Geom. Phys.
39 (2001), no. 3, 253–264.
152. S.D. Jung, T.H. Kang, Lower bounds for the eigenvalue of the transversal Dirac
operator on a Kähler foliation, J. Geom. Phys. 45 (2003), no. 1-2, 75–90.
153. S. Kawai, On the point spectrum of the Dirac operator on a non-compact man-
ifold, J. Geom. Phys. 56 (2006), no. 9, 1782–1789.
154. E.C. Kim, Lower bounds of the Dirac eigenvalues on compact Riemannian spin
manifolds with locally product structure, arXiv:math.DG/0402427 (2004).
155.
, The A-genus and symmetry of the Dirac spectrum on Riemannian prod-
uct manifolds, Diff. Geom. Appl. 25 (2007), no. 3, 309–321.
156. K.-D. Kirchberg, An estimation for the first eigenvalue of the Dirac operator on
closed Kähler manifolds of positive scalar curvature, Ann. Glob. Anal. Geom. 4
(1986), no. 3, 291–325.
157. , Compact six-dimensional Kähler Spin manifolds of positive scalar cur-
vature with the smallest possible first eigenvalue of the Dirac operator, Math.
Ann. 282 (1988), no. 1, 157–176.
158. , The first eigenvalue of the Dirac operator on Kähler manifolds,
J. Geom. Phys. 7 (1990), no. 4, 449–468.
159. , Properties of Kählerian twistor-spinors and vanishing theorems, Math.
Ann. 293 (1992), no. 2, 349–369.
160. , A relation between the Ricci tensor and the spectrum of the Dirac op-
erator, preprint SFB 288 535 (2002), available at http://www-sfb288.math.
tu-berlin.de/Publications/Preprints.html
161. , Curvature dependent lower bounds for the first eigenvalue of the Dirac
operator, J. Geom. Phys. 50 (2004), no. 1-4, 205–222.
162. D. Koh, The η-invariant of the Dirac operator on the Berger spheres, Diplomar-
beit, Universität Hamburg, 2004.
163. W. Kramer, U. Semmelmann, G. Weingart, The first eigenvalue of the Dirac
operator on quaternionic Kähler manifolds, Comm. Math. Phys. 199 (1998),
no. 2, 327–349.
164. , Eigenvalue estimates for the Dirac operator on quaternionic Kähler
manifolds, Math. Z. 230 (1999), no. 4, 727–751.
165. M. Kraus, Lower bounds for eigenvalues of the Dirac operator on surfaces of
rotation, J. Geom. Phys. 31 (1999), no. 2-3, 209–216.
166. , Lower bounds for eigenvalues of the Dirac operator on n-spheres with
SO(n)-symmetry, J. Geom. Phys. 32 (2000), no. 4, 341–348.
167. , Eigenvalues of the Dirac operator on fibrations over S1 , Ann. Glob.
Anal. Geom. 19 (2001), no. 3, 235–257.
168. , Asymptotic estimates of Dirac and Laplace eigenvalues on warped prod-
ucts over S1 , manuscr. math. 112 (2003), no. 3, 357–373.
169. M. Kraus, C. Tretter, A new method for eigenvalue estimates for Dirac operators
on certain manifolds with S k -symmetry, Diff. Geom. Appl. 19 (2003), no. 1,
1–14.
170. W. Kühnel, H.-B. Rademacher, Conformal completion of U(n)-invariant Ricci-
flat Kähler metrics at infinity, Zeitschr. Anal. Anwendungen 16 (1997), no. 1,
113–117.
171. , Asymptotically Euclidean manifolds and twistor spinors, Comm. Math.
Phys. 196 (1998), no. 1, 67–76.
152 Bibliography

172. , Asymptotically Euclidean ends of Ricci flat manifolds, and conformal


inversions, Math. Nachr. 219 (2000), 125–134.
173. H.B. Lawson, M.-L. Michelsohn, Spin geometry, Princeton University Press,
Princeton, 1989.
174. A. Lichnerowicz, Géométrie des groupes de transformations, Travaux et
Recherches Mathématiques III, Dunod, 1958.
175. , Spineurs harmoniques, C. R. Acad. Sci. Paris 257 (1963), 7–9.
176. , Killing spinors according to O. Hijazi and Applications, Spinors in
Physics and Geometry (Trieste 1986), World Scientific Publishing Singapore
(1988), 1–19.
177. , Spin manifolds, Killing spinors and the universality of the Hijazi in-
equality, Lett. Math. Phys. 3 (1987), 331–344.
178. , Killing spinors, twistor-spinors and Hijazi inequality, J. Geom. Phys.
5 (1988), 2–18.
179. , La première valeur propre de l’opérateur de Dirac pour une variété
kählérienne et son cas limite, C. R. Acad. Sci. Paris Sér. I Math. 311 (1990),
no. 11, 717–722.
180. M. Llarull, Sharp estimates and the Dirac operator, Math. Ann. 310 (1998), no.
1, 55–71.
181. J. Lott, Eigenvalue bounds for the Dirac operator, Pacific J. Math. 125 (1986),
no. 1, 117–126.
182. , The Dirac operator and conformal compactification, Internat. Math.
Res. Notices 4 (2001), 171–178.
183. , Collapsing and Dirac-type operators, Geom. Dedicata 91 (2002),
175–196.
184. J. Louis, Compactifications on generalized geometries, Fortschr. Phys. 54 (2006),
no. 2-3, 146–159.
185. S. Maier, Generic metrics and connections on Spin- and Spinc -manifolds,
Comm. Math. Phys. 188 (1997), no. 2, 407–437.
186. H.H. Martens, Varieties of special divisors on a curve, II, J. reine angew. Math.
233 (1968), 89–100.
187. R.J. Miatello, R.A. Podestá, Spin structures and spectra of (Z2 )k -manifolds,
Math. Z. 247 (2004), no. 2, 319–335.
188. , The spectrum of twisted Dirac operators on compact flat manifolds,
Trans. Amer. Math. Soc. 358 (2006), no. 10, 4569–4603.
189. J.-L. Milhorat, Spectre de l’opérateur de Dirac sur les espaces projectifs quater-
nioniens, C. R. Acad. Sci. Paris Sér. I Math. 314 (1992), no. 1, 69–72.
190. , Spectrum of the Dirac operator on Gr2 (Cm+2 ), J. Math. Phys. 39
(1998), no. 1, 594–609.
191. , The first eigenvalue of the Dirac operator on compact Spin symmetric
spaces, Comm. Math. Phys. 259 (2005), no. 1, 71–78.
192. , A formula for the first eigenvalue of the Dirac operator on compact
spin symmetric spaces, J. Math. Phys. 47 (2006), no. 4, 043503.
193. J.W. Milnor, Spin structures on manifolds, Enseignement Math. 9 (1963), no.
2, 198–203.
194. , Remarks concerning spin manifolds, Differ. and Combinat. Topology,
Sympos. Marston Morse, Princeton, 55–62 (1965).
195. S. Montiel, Dirac operators and hypersurfaces, in: J.-P. Bourguignon et al (Eds.):
Dirac Operators, Yesterday and Today, International Press, 2005, 271–281.
196. B. Morel, Eigenvalue estimates for the Dirac-Schrödinger operators, J. Geom.
Phys. 38 (2001), no. 1, 1–18.
197. A. Moroianu, La première valeur propre de l’opérateur de Dirac sur les variétés
kähleriennes compactes, Commun. Math. Phys. 169 (1995), 373–384 .
198. , Opérateur de Dirac et submersions riemanniennes, thèse de doctorat,
Ecole Polytechnique (1996).
Bibliography 153

199. , Sur les valeurs propres de l’opérateur de Dirac d’une variété spinorielle
simplement connexe admettant une 3-structure de Sasaki, Stud. Cerc. Mat. 48
(1996), no. 1-2, 85–88.
200. , Kähler manifolds with small eigenvalues of the Dirac operator and a
conjecture of Lichnerowicz, Ann. Inst. Fourier 49 (1999), no. 5, 1637–1659.
201. A. Moroianu, S. Moroianu, The Dirac spectrum on manifolds with gradient con-
formal vector fields, J. Funct. An. 253 (2007), 207–219.
202. A. Moroianu, L. Ornea, Eigenvalue estimates for the Dirac operator and har-
monic 1-forms of constant length, C. R. Math. Acad. Sci. Paris 338 (2004),
561–564.
203. A. Moroianu, U. Semmelmann, Parallel spinors and holonomy groups, J. Math.
Phys. 41 (2000), no. 4, 2395–2402.
204. S. Moroianu, Weyl laws on open manifolds, Math. Ann. 340 (2008), no. 1, 1–21.
205. R. Petit, Spinc -structures and Dirac operators on contact manifolds, Diff. Geom.
Appl. 22 (2005), no. 2, 229–252.
206. F. Pfäffle, The Dirac spectrum of Bieberbach manifolds, J. Geom. Phys. 35
(2000), no. 4, 367–385.
207. , Eigenwertkonvergenz für Dirac-Operatoren, Doktorarbeit, Universität
Hamburg, Shaker-Verlag, Aachen 2003.
208. , Eigenvalues of Dirac operators for hyperbolic degenerations, manuscr.
math. 116 (2005), no. 1, 1–29.
209. H.-B. Rademacher, Generalized Killing spinors with imaginary Killing function
and conformal Killing fields, Global differential geometry and global analysis
(Berlin, 1990), 192–198, Lecture Notes in Math. 1481, Springer, Berlin, 1991.
210. S. Raulot, Optimal eigenvalues estimate for the Dirac operator on domains with
boundary, Lett. Math. Phys. 73 (2005), no. 2, 135–145.
211. , The Hijazi inequality on manifolds with boundary, J. Geom. Phys. 56
(2006), no. 11, 2189–2202.
212. , Aspect conforme de l’opérateur de Dirac sur une variété à bord, thèse
de doctorat, Université Henri Poincaré, Nancy (2006).
213. , Rigidity of compact Riemannian spin manifolds with boundary, Lett.
Math. Phys. 86 (2008), no. 2, 177–192.
214. , On a spin conformal invariant on manifold with boundary, Math. Z.
261 (2009), no. 2, 321–349.
215. R. Schoen, S. T. Yau, On the proof of the positive mass conjecture in general
relativity, Comm. Math. Phys. 65 (1979), no. 1, 45–76.
216. , Proof of the positive mass theorem II, Comm. Math. Phys. 79 (1981),
no. 2, 231–260.
217. E. Schrödinger, Diracsches Elektron im Schwerefeld, Sitzungsber. Preuß. Akad.
Wiss., Phys.-Math. Kl. (1932), no. 11-12, 105–128.
218. J. Seade, B. Steer, A note on the eta function for quotients of PSL2 (R) by
co-compact Fuchsian groups, Topology 26 (1987), no. 1, 79–91.
219. L. Seeger, The Dirac operator on oriented Grassmann manifolds and G2 /SO(4),
Preprint, Math. Inst. Bonn, 1997.
220. , The spectrum of the Dirac operator on G2 /SO(4), Ann. Glob. Anal.
Geom. 17 (1999), no. 4, 385–396.
221. , Metriken mit harmonischen Spinoren auf geradedimensionalen
Sphären, Shaker Verlag, Aachen, 2001.
222. S. Seifarth, U. Semmelmann, The spectrum of the Dirac operator on the complex
projective space P 2m−1 (C), Preprint SFB 288, 1993.
223. U. Semmelmann, A short proof of eigenvalue estimates for the Dirac operator on
Riemannian and Kähler manifolds, Differential Geometry and its applications,
proceedings, Brno (1998), 137–140.
224. H. Strese, Über den Dirac-Operator auf Grassmann-Mannigfaltigkeiten, Math.
Nachr. 98 (1980), 53–59.
154 Bibliography

225. , Spektren symmetrischer Räume, Math. Nachr. 98 (1980), 75–82.


226. S. Sulanke, Berechnung des Spektrums des Quadrates des Dirac-Operators
auf der Sphäre und Untersuchungen zum ersten Eigenwert von D auf 5-
dimensionalen Räumen konstanter positiver Schnittkrümmung, Doktorarbeit,
Humboldt-Universität zu Berlin, 1979.
227. B. Thaller, The Dirac equation, Springer (1992).
228. A. Trautman, Spin structures on hypersurfaces and the spectrum of the Dirac
operator on spheres, Spinors, twistors, Clifford algebras and quantum deforma-
tions (Sobótka Castle, 1992), 25–29, Fund. Theories Phys. 52, Kluwer Acad.
Publ., Dordrecht, 1993.
229. , The Dirac operator on hypersurfaces, Acta Phys. Polon. B 26 (1995),
no. 7, 1283–1310.
230. C. Vafa, E. Witten, Eigenvalue inequalities for fermions in gauge theories,
Comm. Math. Phys. 95 (1984), no. 3, 257–276.
231. A. Vargas, Manifolds with Killing Spinors and Pinching of First Dirac Eigen-
values, Doktorarbeit, Universität Bonn, 2007.
232. Mc.K.Y. Wang, Parallel spinors and parallel forms, Ann. Glob. Anal. Geom. 7
(1989), no. 1, 59–68.
233. D. Werner, Funktionalanalysis, 5th ed., Springer (2005).
234. E. Wienberg, Das Spektrum des Diracoperators auf der Moufang-Ebene, Diplo-
marbeit, Universität Hamburg, 2005.
235. E. Witten, A new proof of the positive energy theorem, Comm. Math. Phys. 80
(1981), 381–402.
236. N. Xie, X. Zhang, Positive mass theorems for asymptotically AdS spacetimes
with arbitrary cosmological constant, Internat. J. Math. 19 (2008), no. 3,
285–302.
Index

Alexandrov’s theorem, 67 Friedrich’s inequality, 41


Arf invariant, 63 Friedrichs’ extension, 103
Atiyah-Patodi-Singer boundary
condition, 24
modified generalized, 24 harmonic spinors, 16
generalized, 24 Hersch’s inequality, 58
Atiyah-Singer Hijazi’s inequality, 56
index theorem, 17
operator, 9
Killing spinor, 131
real Kählerian, 52
Bär’s inequality, 55 Kirchberg’s inequality, 49
Bär-Hijazi-Lott invariant, 122
boundary condition
elliptic, 23 Laplace operator
self-adjoint, 24 conformal, 56
scalar, 11
Lichnerowicz’ inequality
chirality operator, 24 for the Dirac operator, 42
Clifford for the Laplace operator, 43
algebra, 5
bundle, 119
multiplication, 6 mass, 128
complex volume form, 5 mass endomorphism, 123
conformal covariance, 18 min-max principle, 77
cusp, 109 MIT bag boundary condition, 24

Dirac operator Penrose operator, 131


basic, 121
fundamental, 9
Kostant, 121 Rayleigh quotient, 77
spin, 9
spinc , 119
twisted, 120 Schrödinger-Lichnerowicz formula, 16
Dirac-Schrödinger operator, 119 spectrum
Dolbeault operator, 120 continuous, 104
discrete, 104
essential, 104
energy-momentum tensor, 59 point, 104
Euler operator, 120 residual, 104

155
156 Index

spin Vafa-Witten’s method, 77


group, 1
manifold, 2
structure, 1
spinning systole (of a surface), 62 Willmore conjecture, 91
spinor
bundle, 5
representation, 4
spinorial Levi-Civita connection, 8 Yamabe
invariant, 57
operator, 56
twistor-spinor, 131 problem, 124
Lecture Notes in Mathematics
For information about earlier volumes
please contact your bookseller or Springer
LNM Online archive: springerlink.com

Vol. 1791: M. Knebusch, D. Zhang, Manis Valuations Vol. 1813: L. Ambrosio, L. A. Caffarelli, Y. Brenier,
and Prüfer Extensions I: A new Chapter in Commutative G. Buttazzo, C. Villani, Optimal Transportation and its
Algebra (2002) Applications. Martina Franca, Italy 2001. Editors: L. A.
Vol. 1792: D. D. Ang, R. Gorenflo, V. K. Le, D. D. Trong, Caffarelli, S. Salsa (2003)
Moment Theory and Some Inverse Problems in Potential Vol. 1814: P. Bank, F. Baudoin, H. Föllmer, L.C.G.
Theory and Heat Conduction (2002) Rogers, M. Soner, N. Touzi, Paris-Princeton Lectures on
Vol. 1793: J. Cortés Monforte, Geometric, Control and Mathematical Finance 2002 (2003)
Numerical Aspects of Nonholonomic Systems (2002) Vol. 1815: A. M. Vershik (Ed.), Asymptotic Combi-
Vol. 1794: N. Pytheas Fogg, Substitution in Dynamics, natorics with Applications to Mathematical Physics.
Arithmetics and Combinatorics. Editors: V. Berthé, S. St. Petersburg, Russia 2001 (2003)
Ferenczi, C. Mauduit, A. Siegel (2002) Vol. 1816: S. Albeverio, W. Schachermayer, M. Tala-
Vol. 1795: H. Li, Filtered-Graded Transfer in Using Non- grand, Lectures on Probability Theory and Statistics.
commutative Gröbner Bases (2002) Ecole d’Eté de Probabilités de Saint-Flour XXX-2000.
Vol. 1796: J.M. Melenk, hp-Finite Element Methods for Editor: P. Bernard (2003)
Singular Perturbations (2002) Vol. 1817: E. Koelink, W. Van Assche (Eds.), Orthogonal
Vol. 1797: B. Schmidt, Characters and Cyclotomic Fields Polynomials and Special Functions. Leuven 2002 (2003)
in Finite Geometry (2002) Vol. 1818: M. Bildhauer, Convex Variational Problems
Vol. 1798: W.M. Oliva, Geometric Mechanics (2002) with Linear, nearly Linear and/or Anisotropic Growth
Vol. 1799: H. Pajot, Analytic Capacity, Rectifiability, Conditions (2003)
Menger Curvature and the Cauchy Integral (2002) Vol. 1819: D. Masser, Yu. V. Nesterenko, H. P. Schlick-
Vol. 1800: O. Gabber, L. Ramero, Almost Ring Theory ewei, W. M. Schmidt, M. Waldschmidt, Diophantine
(2003) Approximation. Cetraro, Italy 2000. Editors: F. Amoroso,
Vol. 1801: J. Azéma, M. Émery, M. Ledoux, M. Yor U. Zannier (2003)
(Eds.), Séminaire de Probabilités XXXVI (2003) Vol. 1820: F. Hiai, H. Kosaki, Means of Hilbert Space
Vol. 1802: V. Capasso, E. Merzbach, B. G. Ivanoff, Operators (2003)
M. Dozzi, R. Dalang, T. Mountford, Topics in Spatial Vol. 1821: S. Teufel, Adiabatic Perturbation Theory in
Stochastic Processes. Martina Franca, Italy 2001. Editor: Quantum Dynamics (2003)
E. Merzbach (2003)
Vol. 1822: S.-N. Chow, R. Conti, R. Johnson, J. Mallet-
Vol. 1803: G. Dolzmann, Variational Methods for Crys- Paret, R. Nussbaum, Dynamical Systems. Cetraro, Italy
talline Microstructure – Analysis and Computation 2000. Editors: J. W. Macki, P. Zecca (2003)
(2003)
Vol. 1823: A. M. Anile, W. Allegretto, C. Ringhofer,
Vol. 1804: I. Cherednik, Ya. Markov, R. Howe, G.
Mathematical Problems in Semiconductor Physics.
Lusztig, Iwahori-Hecke Algebras and their Representa-
Cetraro, Italy 1998. Editor: A. M. Anile (2003)
tion Theory. Martina Franca, Italy 1999. Editors: V. Bal-
doni, D. Barbasch (2003) Vol. 1824: J. A. Navarro González, J. B. Sancho de Salas,
C ∞ – Differentiable Spaces (2003)
Vol. 1805: F. Cao, Geometric Curve Evolution and Image
Processing (2003) Vol. 1825: J. H. Bramble, A. Cohen, W. Dahmen, Mul-
Vol. 1806: H. Broer, I. Hoveijn. G. Lunther, G. Vegter, tiscale Problems and Methods in Numerical Simulations,
Bifurcations in Hamiltonian Systems. Computing Singu- Martina Franca, Italy 2001. Editor: C. Canuto (2003)
larities by Gröbner Bases (2003) Vol. 1826: K. Dohmen, Improved Bonferroni Inequal-
Vol. 1807: V. D. Milman, G. Schechtman (Eds.), Geomet- ities via Abstract Tubes. Inequalities and Identities of
ric Aspects of Functional Analysis. Israel Seminar 2000- Inclusion-Exclusion Type. VIII, 113 p, 2003.
2002 (2003) Vol. 1827: K. M. Pilgrim, Combinations of Complex
Vol. 1808: W. Schindler, Measures with Symmetry Prop- Dynamical Systems. IX, 118 p, 2003.
erties (2003) Vol. 1828: D. J. Green, Gröbner Bases and the Computa-
Vol. 1809: O. Steinbach, Stability Estimates for Hybrid tion of Group Cohomology. XII, 138 p, 2003.
Coupled Domain Decomposition Methods (2003) Vol. 1829: E. Altman, B. Gaujal, A. Hordijk, Discrete-
Vol. 1810: J. Wengenroth, Derived Functors in Functional Event Control of Stochastic Networks: Multimodularity
Analysis (2003) and Regularity. XIV, 313 p, 2003.
Vol. 1811: J. Stevens, Deformations of Singularities Vol. 1830: M. I. Gil’, Operator Functions and Localiza-
(2003) tion of Spectra. XIV, 256 p, 2003.
Vol. 1812: L. Ambrosio, K. Deckelnick, G. Dziuk, Vol. 1831: A. Connes, J. Cuntz, E. Guentner, N. Hig-
M. Mimura, V. A. Solonnikov, H. M. Soner, Mathemat- son, J. E. Kaminker, Noncommutative Geometry, Mar-
ical Aspects of Evolving Interfaces. Madeira, Funchal, tina Franca, Italy 2002. Editors: S. Doplicher, L. Longo
Portugal 2000. Editors: P. Colli, J. F. Rodrigues (2003) (2004)
Vol. 1832: J. Azéma, M. Émery, M. Ledoux, M. Yor Vol. 1858: A.S. Cherny, H.-J. Engelbert, Singular
(Eds.), Séminaire de Probabilités XXXVII (2003) Stochastic Differential Equations (2005)
Vol. 1833: D.-Q. Jiang, M. Qian, M.-P. Qian, Mathemati- Vol. 1859: E. Letellier, Fourier Transforms of Invariant
cal Theory of Nonequilibrium Steady States. On the Fron- Functions on Finite Reductive Lie Algebras (2005)
tier of Probability and Dynamical Systems. IX, 280 p, Vol. 1860: A. Borisyuk, G.B. Ermentrout, A. Friedman,
2004. D. Terman, Tutorials in Mathematical Biosciences I.
Vol. 1834: Yo. Yomdin, G. Comte, Tame Geometry with Mathematical Neurosciences (2005)
Application in Smooth Analysis. VIII, 186 p, 2004. Vol. 1861: G. Benettin, J. Henrard, S. Kuksin, Hamilto-
Vol. 1835: O.T. Izhboldin, B. Kahn, N.A. Karpenko, nian Dynamics – Theory and Applications, Cetraro, Italy,
A. Vishik, Geometric Methods in the Algebraic Theory 1999. Editor: A. Giorgilli (2005)
of Quadratic Forms. Summer School, Lens, 2000. Editor: Vol. 1862: B. Helffer, F. Nier, Hypoelliptic Estimates and
J.-P. Tignol (2004) Spectral Theory for Fokker-Planck Operators and Witten
Vol. 1836: C. Nǎstǎsescu, F. Van Oystaeyen, Methods of Laplacians (2005)
Graded Rings. XIII, 304 p, 2004. Vol. 1863: H. Führ, Abstract Harmonic Analysis of Con-
Vol. 1837: S. Tavaré, O. Zeitouni, Lectures on Probabil- tinuous Wavelet Transforms (2005)
ity Theory and Statistics. Ecole d’Eté de Probabilités de
Vol. 1864: K. Efstathiou, Metamorphoses of Hamiltonian
Saint-Flour XXXI-2001. Editor: J. Picard (2004)
Systems with Symmetries (2005)
Vol. 1838: A.J. Ganesh, N.W. O’Connell, D.J. Wischik,
Big Queues. XII, 254 p, 2004. Vol. 1865: D. Applebaum, B.V. R. Bhat, J. Kustermans,
Vol. 1839: R. Gohm, Noncommutative Stationary Pro- J. M. Lindsay, Quantum Independent Increment Pro-
cesses. VIII, 170 p, 2004. cesses I. From Classical Probability to Quantum Stochas-
Vol. 1840: B. Tsirelson, W. Werner, Lectures on Probabil- tic Calculus. Editors: M. Schürmann, U. Franz (2005)
ity Theory and Statistics. Ecole d’Eté de Probabilités de Vol. 1866: O.E. Barndorff-Nielsen, U. Franz, R. Gohm,
Saint-Flour XXXII-2002. Editor: J. Picard (2004) B. Kümmerer, S. Thorbjønsen, Quantum Independent
Vol. 1841: W. Reichel, Uniqueness Theorems for Vari- Increment Processes II. Structure of Quantum Lévy
ational Problems by the Method of Transformation Processes, Classical Probability, and Physics. Editors: M.
Groups (2004) Schürmann, U. Franz, (2005)
Vol. 1842: T. Johnsen, A. L. Knutsen, K3 Projective Mod- Vol. 1867: J. Sneyd (Ed.), Tutorials in Mathematical Bio-
els in Scrolls (2004) sciences II. Mathematical Modeling of Calcium Dynam-
Vol. 1843: B. Jefferies, Spectral Properties of Noncom- ics and Signal Transduction. (2005)
muting Operators (2004) Vol. 1868: J. Jorgenson, S. Lang, Posn (R) and Eisenstein
Vol. 1844: K.F. Siburg, The Principle of Least Action in Series. (2005)
Geometry and Dynamics (2004) Vol. 1869: A. Dembo, T. Funaki, Lectures on Probabil-
Vol. 1845: Min Ho Lee, Mixed Automorphic Forms, ity Theory and Statistics. Ecole d’Eté de Probabilités de
Torus Bundles, and Jacobi Forms (2004) Saint-Flour XXXIII-2003. Editor: J. Picard (2005)
Vol. 1846: H. Ammari, H. Kang, Reconstruction of Small Vol. 1870: V.I. Gurariy, W. Lusky, Geometry of Müntz
Inhomogeneities from Boundary Measurements (2004) Spaces and Related Questions. (2005)
Vol. 1847: T.R. Bielecki, T. Björk, M. Jeanblanc, M. Vol. 1871: P. Constantin, G. Gallavotti, A.V. Kazhikhov,
Rutkowski, J.A. Scheinkman, W. Xiong, Paris-Princeton Y. Meyer, S. Ukai, Mathematical Foundation of Turbu-
Lectures on Mathematical Finance 2003 (2004) lent Viscous Flows, Martina Franca, Italy, 2003. Editors:
Vol. 1848: M. Abate, J. E. Fornaess, X. Huang, J. P. M. Cannone, T. Miyakawa (2006)
Rosay, A. Tumanov, Real Methods in Complex and CR Vol. 1872: A. Friedman (Ed.), Tutorials in Mathemati-
Geometry, Martina Franca, Italy 2002. Editors: D. Zait- cal Biosciences III. Cell Cycle, Proliferation, and Cancer
sev, G. Zampieri (2004) (2006)
Vol. 1849: Martin L. Brown, Heegner Modules and Ellip- Vol. 1873: R. Mansuy, M. Yor, Random Times and En-
tic Curves (2004) largements of Filtrations in a Brownian Setting (2006)
Vol. 1850: V. D. Milman, G. Schechtman (Eds.), Geomet- Vol. 1874: M. Yor, M. Émery (Eds.), In Memoriam Paul-
ric Aspects of Functional Analysis. Israel Seminar 2002- André Meyer - Séminaire de Probabilités XXXIX (2006)
2003 (2004) Vol. 1875: J. Pitman, Combinatorial Stochastic Processes.
Vol. 1851: O. Catoni, Statistical Learning Theory and Ecole d’Eté de Probabilités de Saint-Flour XXXII-2002.
Stochastic Optimization (2004) Editor: J. Picard (2006)
Vol. 1852: A.S. Kechris, B.D. Miller, Topics in Orbit
Vol. 1876: H. Herrlich, Axiom of Choice (2006)
Equivalence (2004)
Vol. 1853: Ch. Favre, M. Jonsson, The Valuative Tree Vol. 1877: J. Steuding, Value Distributions of L-Functions
(2004) (2007)
Vol. 1854: O. Saeki, Topology of Singular Fibers of Dif- Vol. 1878: R. Cerf, The Wulff Crystal in Ising and Percol-
ferential Maps (2004) ation Models, Ecole d’Eté de Probabilités de Saint-Flour
Vol. 1855: G. Da Prato, P.C. Kunstmann, I. Lasiecka, XXXIV-2004. Editor: Jean Picard (2006)
A. Lunardi, R. Schnaubelt, L. Weis, Functional Analytic Vol. 1879: G. Slade, The Lace Expansion and its Applica-
Methods for Evolution Equations. Editors: M. Iannelli, tions, Ecole d’Eté de Probabilités de Saint-Flour XXXIV-
R. Nagel, S. Piazzera (2004) 2004. Editor: Jean Picard (2006)
Vol. 1856: K. Back, T.R. Bielecki, C. Hipp, S. Peng, Vol. 1880: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
W. Schachermayer, Stochastic Methods in Finance, Bres- Systems I, The Hamiltonian Approach (2006)
sanone/Brixen, Italy, 2003. Editors: M. Fritelli, W. Rung- Vol. 1881: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
galdier (2004) Systems II, The Markovian Approach (2006)
Vol. 1857: M. Émery, M. Ledoux, M. Yor (Eds.), Sémi- Vol. 1882: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
naire de Probabilités XXXVIII (2005) Systems III, Recent Developments (2006)
Vol. 1883: W. Van Assche, F. Marcellàn (Eds.), Orthogo- Vol. 1910: V.D. Milman, G. Schechtman (Eds.), Geo-
nal Polynomials and Special Functions, Computation and metric Aspects of Functional Analysis. Israel Seminar
Application (2006) 2004-2005 (2007)
Vol. 1884: N. Hayashi, E.I. Kaikina, P.I. Naumkin, Vol. 1911: A. Bressan, D. Serre, M. Williams,
I.A. Shishmarev, Asymptotics for Dissipative Nonlinear K. Zumbrun, Hyperbolic Systems of Balance Laws.
Equations (2006) Cetraro, Italy 2003. Editor: P. Marcati (2007)
Vol. 1885: A. Telcs, The Art of Random Walks (2006) Vol. 1912: V. Berinde, Iterative Approximation of Fixed
Vol. 1886: S. Takamura, Splitting Deformations of Dege- Points (2007)
nerations of Complex Curves (2006) Vol. 1913: J.E. Marsden, G. Misiołek, J.-P. Ortega,
Vol. 1887: K. Habermann, L. Habermann, Introduction to M. Perlmutter, T.S. Ratiu, Hamiltonian Reduction by
Symplectic Dirac Operators (2006) Stages (2007)
Vol. 1888: J. van der Hoeven, Transseries and Real Dif- Vol. 1914: G. Kutyniok, Affine Density in Wavelet
ferential Algebra (2006) Analysis (2007)
Vol. 1889: G. Osipenko, Dynamical Systems, Graphs, and Vol. 1915: T. Bıyıkoǧlu, J. Leydold, P.F. Stadler,
Algorithms (2006) Laplacian Eigenvectors of Graphs. Perron-Frobenius and
Vol. 1890: M. Bunge, J. Funk, Singular Coverings of Faber-Krahn Type Theorems (2007)
Toposes (2006) Vol. 1916: C. Villani, F. Rezakhanlou, Entropy Methods
Vol. 1891: J.B. Friedlander, D.R. Heath-Brown, for the Boltzmann Equation. Editors: F. Golse, S. Olla
H. Iwaniec, J. Kaczorowski, Analytic Number Theory, (2008)
Cetraro, Italy, 2002. Editors: A. Perelli, C. Viola (2006) Vol. 1917: I. Veselić, Existence and Regularity Prop-
Vol. 1892: A. Baddeley, I. Bárány, R. Schneider, W. Weil, erties of the Integrated Density of States of Random
Stochastic Geometry, Martina Franca, Italy, 2004. Editor: Schrödinger (2008)
W. Weil (2007) Vol. 1918: B. Roberts, R. Schmidt, Local Newforms for
Vol. 1893: H. Hanßmann, Local and Semi-Local Bifur- GSp(4) (2007)
cations in Hamiltonian Dynamical Systems, Results and Vol. 1919: R.A. Carmona, I. Ekeland, A. Kohatsu-
Examples (2007) Higa, J.-M. Lasry, P.-L. Lions, H. Pham, E. Taflin,
Vol. 1894: C.W. Groetsch, Stable Approximate Evalua- Paris-Princeton Lectures on Mathematical Finance 2004.
tion of Unbounded Operators (2007) Editors: R.A. Carmona, E. Çinlar, I. Ekeland, E. Jouini,
Vol. 1895: L. Molnár, Selected Preserver Problems on J.A. Scheinkman, N. Touzi (2007)
Algebraic Structures of Linear Operators and on Function Vol. 1920: S.N. Evans, Probability and Real Trees. Ecole
Spaces (2007) d’Été de Probabilités de Saint-Flour XXXV-2005 (2008)
Vol. 1896: P. Massart, Concentration Inequalities and Vol. 1921: J.P. Tian, Evolution Algebras and their Appli-
Model Selection, Ecole d’Été de Probabilités de Saint- cations (2008)
Flour XXXIII-2003. Editor: J. Picard (2007) Vol. 1922: A. Friedman (Ed.), Tutorials in Mathematical
Vol. 1897: R. Doney, Fluctuation Theory for Lévy BioSciences IV. Evolution and Ecology (2008)
Processes, Ecole d’Été de Probabilités de Saint-Flour Vol. 1923: J.P.N. Bishwal, Parameter Estimation in
XXXV-2005. Editor: J. Picard (2007) Stochastic Differential Equations (2008)
Vol. 1898: H.R. Beyer, Beyond Partial Differential Equa- Vol. 1924: M. Wilson, Littlewood-Paley Theory and
tions, On linear and Quasi-Linear Abstract Hyperbolic Exponential-Square Integrability (2008)
Evolution Equations (2007) Vol. 1925: M. du Sautoy, L. Woodward, Zeta Functions
Vol. 1899: Séminaire de Probabilités XL. Editors: of Groups and Rings (2008)
C. Donati-Martin, M. Émery, A. Rouault, C. Stricker Vol. 1926: L. Barreira, V. Claudia, Stability of Nonauto-
(2007) nomous Differential Equations (2008)
Vol. 1900: E. Bolthausen, A. Bovier (Eds.), Spin Glasses Vol. 1927: L. Ambrosio, L. Caffarelli, M.G. Crandall,
(2007) L.C. Evans, N. Fusco, Calculus of Variations and Non-
Vol. 1901: O. Wittenberg, Intersections de deux Linear Partial Differential Equations. Cetraro, Italy 2005.
quadriques et pinceaux de courbes de genre 1, Intersec- Editors: B. Dacorogna, P. Marcellini (2008)
tions of Two Quadrics and Pencils of Curves of Genus 1 Vol. 1928: J. Jonsson, Simplicial Complexes of Graphs
(2007) (2008)
Vol. 1902: A. Isaev, Lectures on the Automorphism Vol. 1929: Y. Mishura, Stochastic Calculus for Fractional
Groups of Kobayashi-Hyperbolic Manifolds (2007) Brownian Motion and Related Processes (2008)
Vol. 1903: G. Kresin, V. Maz’ya, Sharp Real-Part Theo- Vol. 1930: J.M. Urbano, The Method of Intrinsic Scaling.
rems (2007) A Systematic Approach to Regularity for Degenerate and
Vol. 1904: P. Giesl, Construction of Global Lyapunov Singular PDEs (2008)
Functions Using Radial Basis Functions (2007) Vol. 1931: M. Cowling, E. Frenkel, M. Kashiwara,
Vol. 1905: C. Prévôt, M. Röckner, A Concise Course on A. Valette, D.A. Vogan, Jr., N.R. Wallach, Representation
Stochastic Partial Differential Equations (2007) Theory and Complex Analysis. Venice, Italy 2004.
Vol. 1906: T. Schuster, The Method of Approximate Editors: E.C. Tarabusi, A. D’Agnolo, M. Picardello
Inverse: Theory and Applications (2007) (2008)
Vol. 1907: M. Rasmussen, Attractivity and Bifurcation Vol. 1932: A.A. Agrachev, A.S. Morse, E.D. Sontag,
for Nonautonomous Dynamical Systems (2007) H.J. Sussmann, V.I. Utkin, Nonlinear and Optimal
Vol. 1908: T.J. Lyons, M. Caruana, T. Lévy, Differential Control Theory. Cetraro, Italy 2004. Editors: P. Nistri,
Equations Driven by Rough Paths, Ecole d’Été de Proba- G. Stefani (2008)
bilités de Saint-Flour XXXIV-2004 (2007) Vol. 1933: M. Petkovic, Point Estimation of Root Finding
Vol. 1909: H. Akiyoshi, M. Sakuma, M. Wada, Methods (2008)
Y. Yamashita, Punctured Torus Groups and 2-Bridge Vol. 1934: C. Donati-Martin, M. Émery, A. Rouault,
Knot Groups (I) (2007) C. Stricker (Eds.), Séminaire de Probabilités XLI (2008)
Vol. 1935: A. Unterberger, Alternative Pseudodifferential Vol. 1958: M.C. Olsson, Compactifying Moduli Spaces
Analysis (2008) for Abelian Varieties (2008)
Vol. 1936: P. Magal, S. Ruan (Eds.), Structured Popula- Vol. 1959: Y. Nakkajima, A. Shiho, Weight Filtrations
tion Models in Biology and Epidemiology (2008) on Log Crystalline Cohomologies of Families of Open
Vol. 1937: G. Capriz, P. Giovine, P.M. Mariano (Eds.), Smooth Varieties (2008)
Mathematical Models of Granular Matter (2008) Vol. 1960: J. Lipman, M. Hashimoto, Foundations of
Vol. 1938: D. Auroux, F. Catanese, M. Manetti, P. Seidel, Grothendieck Duality for Diagrams of Schemes (2009)
B. Siebert, I. Smith, G. Tian, Symplectic 4-Manifolds Vol. 1961: G. Buttazzo, A. Pratelli, S. Solimini,
and Algebraic Surfaces. Cetraro, Italy 2003. Editors: E. Stepanov, Optimal Urban Networks via Mass Trans-
F. Catanese, G. Tian (2008) portation (2009)
Vol. 1939: D. Boffi, F. Brezzi, L. Demkowicz, R.G. Vol. 1962: R. Dalang, D. Khoshnevisan, C. Mueller,
Durán, R.S. Falk, M. Fortin, Mixed Finite Elements, D. Nualart, Y. Xiao, A Minicourse on Stochastic Partial
Compatibility Conditions, and Applications. Cetraro, Differential Equations (2009)
Italy 2006. Editors: D. Boffi, L. Gastaldi (2008) Vol. 1963: W. Siegert, Local Lyapunov Exponents (2009)
Vol. 1940: J. Banasiak, V. Capasso, M.A.J. Chap- Vol. 1964: W. Roth, Operator-valued Measures and Inte-
lain, M. Lachowicz, J. Miȩkisz, Multiscale Problems in grals for Cone-valued Functions and Integrals for Cone-
the Life Sciences. From Microscopic to Macroscopic. valued Functions (2009)
Bȩdlewo, Poland 2006. Editors: V. Capasso, M. Lachow- Vol. 1965: C. Chidume, Geometric Properties of Banach
icz (2008) Spaces and Nonlinear Iterations (2009)
Vol. 1941: S.M.J. Haran, Arithmetical Investigations. Vol. 1966: D. Deng, Y. Han, Harmonic Analysis on
Representation Theory, Orthogonal Polynomials, and Spaces of Homogeneous Type (2009)
Quantum Interpolations (2008) Vol. 1967: B. Fresse, Modules over Operads and Functors
Vol. 1942: S. Albeverio, F. Flandoli, Y.G. Sinai, SPDE in (2009)
Hydrodynamic. Recent Progress and Prospects. Cetraro, Vol. 1968: R. Weissauer, Endoscopy for GSP(4) and the
Italy 2005. Editors: G. Da Prato, M. Röckner (2008) Cohomology of Siegel Modular Threefolds (2009)
Vol. 1969: B. Roynette, M. Yor, Penalising Brownian
Vol. 1943: L.L. Bonilla (Ed.), Inverse Problems and Imag-
Paths (2009)
ing. Martina Franca, Italy 2002 (2008)
Vol. 1970: M. Biskup, A. Bovier, F. den Hollander, D.
Vol. 1944: A. Di Bartolo, G. Falcone, P. Plaumann,
Ioffe, F. Martinelli, K. Netočný, F. Toninelli, Methods of
K. Strambach, Algebraic Groups and Lie Groups with
Contemporary Mathematical Statistical Physics. Editor:
Few Factors (2008)
R. Kotecký (2009)
Vol. 1945: F. Brauer, P. van den Driessche, J. Wu (Eds.), Vol. 1971: L. Saint-Raymond, Hydrodynamic Limits of
Mathematical Epidemiology (2008) the Boltzmann Equation (2009)
Vol. 1946: G. Allaire, A. Arnold, P. Degond, T.Y. Hou, Vol. 1972: T. Mochizuki, Donaldson Type Invariants for
Quantum Transport. Modelling, Analysis and Asymp- Algebraic Surfaces (2009)
totics. Cetraro, Italy 2006. Editors: N.B. Abdallah, Vol. 1973: M.A. Berger, L.H. Kauffmann, B. Khesin,
G. Frosali (2008) H.K. Moffatt, R.L. Ricca, De W. Sumners, Lectures on
Vol. 1947: D. Abramovich, M. Mariño, M. Thaddeus, Topological Fluid Mechanics. Editor: R.L. Ricca (2009)
R. Vakil, Enumerative Invariants in Algebraic Geo- Vol. 1974: F. den Hollander, Random Polymers: École
metry and String Theory. Cetraro, Italy 2005. Editors: d’Été de Probabilités de Saint-Flour XXXVII – 2007
K. Behrend, M. Manetti (2008) (2009)
Vol. 1948: F. Cao, J-L. Lisani, J-M. Morel, P. Musé, Vol. 1975: J.C. Rohde, Cyclic Coverings, Calabi-Yau
F. Sur, A Theory of Shape Identification (2008) Manifolds and Complex Multiplication (2009)
Vol. 1949: H.G. Feichtinger, B. Helffer, M.P. Lamoureux, Vol. 1976: N. Ginoux, The Dirac Spectrum (2009)
N. Lerner, J. Toft, Pseudo-Differential Operators. Quan-
tization and Signals. Cetraro, Italy 2006. Editors: L.
Rodino, M.W. Wong (2008) Recent Reprints and New Editions
Vol. 1950: M. Bramson, Stability of Queueing Networks, Vol. 1702: J. Ma, J. Yong, Forward-Backward Stochas-
Ecole d’Eté de Probabilités de Saint-Flour XXXVI-2006 tic Differential Equations and their Applications. 1999 –
(2008) Corr. 3rd printing (2007)
Vol. 1951: A. Moltó, J. Orihuela, S. Troyanski, Vol. 830: J.A. Green, Polynomial Representations of
M. Valdivia, A Non Linear Transfer Technique for GLn , with an Appendix on Schensted Correspondence
Renorming (2009) and Littelmann Paths by K. Erdmann, J.A. Green and
Vol. 1952: R. Mikhailov, I.B.S. Passi, Lower Central and M. Schoker 1980 – 2nd corr. and augmented edition
Dimension Series of Groups (2009) (2007)
Vol. 1953: K. Arwini, C.T.J. Dodson, Information Geo- Vol. 1693: S. Simons, From Hahn-Banach to Monotonic-
metry (2008) ity (Minimax and Monotonicity 1998) – 2nd exp. edition
Vol. 1954: P. Biane, L. Bouten, F. Cipriani, N. Konno, (2008)
N. Privault, Q. Xu, Quantum Potential Theory. Editors: Vol. 470: R.E. Bowen, Equilibrium States and the Ergodic
U. Franz, M. Schuermann (2008) Theory of Anosov Diffeomorphisms. With a preface by
Vol. 1955: M. Bernot, V. Caselles, J.-M. Morel, Optimal D. Ruelle. Edited by J.-R. Chazottes. 1975 – 2nd rev.
Transportation Networks (2008) edition (2008)
Vol. 1956: C.H. Chu, Matrix Convolution Operators on Vol. 523: S.A. Albeverio, R.J. Høegh-Krohn, S. Maz-
Groups (2008) zucchi, Mathematical Theory of Feynman Path Integral.
Vol. 1957: A. Guionnet, On Random Matrices: Macro- 1976 – 2nd corr. and enlarged edition (2008)
scopic Asymptotics, Ecole d’Eté de Probabilités de Saint- Vol. 1764: A. Cannas da Silva, Lectures on Symplectic
Flour XXXVI-2006 (2009) Geometry 2001 – Corr. 2nd printing (2008)
LECTURE NOTES IN MATHEMATICS 123
Edited by J.-M. Morel, F. Takens, B. Teissier, P.K. Maini

Editorial Policy (for the publication of monographs)

1. Lecture Notes aim to report new developments in all areas of mathematics and their
applications - quickly, informally and at a high level. Mathematical texts analysing new
developments in modelling and numerical simulation are welcome.
Monograph manuscripts should be reasonably self-contained and rounded off. Thus
they may, and often will, present not only results of the author but also related work
by other people. They may be based on specialised lecture courses. Furthermore, the
manuscripts should provide sufficient motivation, examples and applications. This clearly
distinguishes Lecture Notes from journal articles or technical reports which normally are
very concise. Articles intended for a journal but too long to be accepted by most journals,
usually do not have this “lecture notes” character. For similar reasons it is unusual for
doctoral theses to be accepted for the Lecture Notes series, though habilitation theses may
be appropriate.
2. Manuscripts should be submitted either online at www.editorialmanager.com/lnm to
Springer’s mathematics editorial in Heidelberg, or to one of the series editors. In general,
manuscripts will be sent out to 2 external referees for evaluation. If a decision cannot yet
be reached on the basis of the first 2 reports, further referees may be contacted: The author
will be informed of this. A final decision to publish can be made only on the basis of the
complete manuscript, however a refereeing process leading to a preliminary decision can
be based on a pre-final or incomplete manuscript. The strict minimum amount of material
that will be considered should include a detailed outline describing the planned contents
of each chapter, a bibliography and several sample chapters.
Authors should be aware that incomplete or insufficiently close to final manuscripts
almost always result in longer refereeing times and nevertheless unclear referees’ recom-
mendations, making further refereeing of a final draft necessary.
Authors should also be aware that parallel submission of their manuscript to another
publisher while under consideration for LNM will in general lead to immediate rejection.
3. Manuscripts should in general be submitted in English. Final manuscripts should contain
at least 100 pages of mathematical text and should always include
– a table of contents;
– an informative introduction, with adequate motivation and perhaps some historical re-
marks: it should be accessible to a reader not intimately familiar with the topic treated;
– a subject index: as a rule this is genuinely helpful for the reader.
For evaluation purposes, manuscripts may be submitted in print or electronic form (print
form is still preferred by most referees), in the latter case preferably as pdf- or zipped
ps-files. Lecture Notes volumes are, as a rule, printed digitally from the authors’ files.
To ensure best results, authors are asked to use the LaTeX2e style files available from
Springer’s web-server at:
ftp://ftp.springer.de/pub/tex/latex/svmonot1/ (for monographs) and
ftp://ftp.springer.de/pub/tex/latex/svmultt1/ (for summer schools/tutorials).
Additional technical instructions, if necessary, are available on request from:
lnm@springer.com.
4. Careful preparation of the manuscripts will help keep production time short besides en-
suring satisfactory appearance of the finished book in print and online. After acceptance
of the manuscript authors will be asked to prepare the final LaTeX source files and also
the corresponding dvi-, pdf- or zipped ps-file. The LaTeX source files are essential for
producing the full-text online version of the book (see
http://www.springerlink.com/openurl.asp?genre=journal&issn=0075-8434 for the exist-
ing online volumes of LNM).
The actual production of a Lecture Notes volume takes approximately 12 weeks.
5. Authors receive a total of 50 free copies of their volume, but no royalties. They are entitled
to a discount of 33.3% on the price of Springer books purchased for their personal use, if
ordering directly from Springer.
6. Commitment to publish is made by letter of intent rather than by signing a formal contract.
Springer-Verlag secures the copyright for each volume. Authors are free to reuse material
contained in their LNM volumes in later publications: a brief written (or e-mail) request
for formal permission is sufficient.

Addresses:
Professor J.-M. Morel, CMLA,
École Normale Supérieure de Cachan,
61 Avenue du Président Wilson, 94235 Cachan Cedex, France
E-mail: Jean-Michel.Morel@cmla.ens-cachan.fr
Professor F. Takens, Mathematisch Instituut,
Rijksuniversiteit Groningen, Postbus 800,
9700 AV Groningen, The Netherlands
E-mail: F.Takens@rug.nl
Professor B. Teissier, Institut Mathématique de Jussieu,
UMR 7586 du CNRS, Équipe “Géométrie et Dynamique”,
175 rue du Chevaleret,
75013 Paris, France
E-mail: teissier@math.jussieu.fr

For the “Mathematical Biosciences Subseries” of LNM:


Professor P.K. Maini, Center for Mathematical Biology,
Mathematical Institute, 24-29 St Giles,
Oxford OX1 3LP, UK
E-mail: maini@maths.ox.ac.uk
Springer, Mathematics Editorial, Tiergartenstr. 17,
69121 Heidelberg, Germany,
Tel.: +49 (6221) 487-259
Fax: +49 (6221) 4876-8259
E-mail: lnm@springer.com

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy