Series

Series
SeriesISSN:
ISSN:
ISSN:
1938-1743
1938-1743
1938-1743

GARCÍA-RÍO • GILKEY • NIKČEVIC´ • VÁZQUEZ-LORENZO

GARCÍA-RÍO •• GILKEY
GARCÍA-RÍO
SSYNTHESIS
YNTHESISL
YNTHESIS
CCOMPUTER
OMPUTERM
OMPUTER
LECTURES
ECTURES
ECTURESON
ON
ON
MATHEMATICS
ATHEMATICS ANDSSTATISTICS
ATHEMATICSAND
AND TATISTICS
TATISTICS
MM
&C
& C Mor
Mor gan&
Morgan
gan &Cl
Cl
Claypool
aypool
aypool Publishers
Publishers
Publishers

GILKEY •• NIK
Applications
Applications of of
Series
Series
SeriesEditor:
Editor:
Editor:Steven
Steven
StevenG.
G.
G.Krantz,
Krantz,
Krantz,Washington
Washington
WashingtonUniversity,
University,
University,St.
St.
St.Louis
Louis
Louis

NIKČEVIC
Applications
Applicationsof
ofAffine
Affineand
andWeyl
WeylGeometry
Geometry

ČEVIC´´ •• VÁZQUEZ-LORENZO
Eduardo
Eduardo
EduardoGarcía-Río,
García-Río,
García-Río,University
University
UniversityofofofSantiago
Santiago
Santiagodede
deCompostela,
Compostela,
Compostela,Peter
Peter
PeterGilkey,
Gilkey,
Gilkey,University
University
UniversityofofofOregon
Oregon
Oregon
Affine
Affine and
and Weyl
Weyl Geometry
Geometry

VÁZQUEZ-LORENZO
Stana
Stana
StanaNikčević,
Nikčević,
Nikčević,Mathematical
Mathematical
MathematicalInstitute,
Institute,
Institute,Sanu,
Sanu,
Sanu,Serbia,
Serbia,
Serbia,Ramón
Ramón
RamónVázquez-Lorenzo,
Vázquez-Lorenzo,
Vázquez-Lorenzo,University
University
UniversityofofofSantiago
Santiago
Santiagodede
deCompostela
Compostela
Compostela

Pseudo-Riemannian
Pseudo-Riemannian
Pseudo-Riemanniangeometry geometry
geometryis,is, is,
toto
toa alarge
a large
largeextent,
extent,
extent,the
the
thestudy
study
studyofofofthethe
theLevi-Civita
Levi-Civita
Levi-Civitaconnection,
connection,
connection, which
which
whichisisisthe
the
theunique
unique
unique
torsion-free
torsion-free
torsion-freeconnectionconnection
connectioncompatiblecompatible
compatiblewith with
withthe the
themetric
metric
metricstructure.
structure.
structure.There There
Thereare,are,
are,however,
however,
however,other other
otheraffine
affine
affineconnections
connections
connections
which
which
whicharise
arise
ariseinin indifferent
different
differentcontexts,
contexts,
contexts,such such
suchasas asconformal
conformal
conformalgeometry,
geometry,
geometry,contact contact
contactstructures,
structures,
structures,Weyl Weyl
Weylstructures,
structures,
structures,and and
andalmost
almost
almost
Hermitian
Hermitian
Hermitiangeometry. geometry.
geometry.InIn Inthis
this
thisbook,
book,
book,we we
wereverse
reverse
reversethis
this
thispoint
point
pointofof ofview
view
viewand and
andinstead
instead
insteadassociate
associate
associateanan anauxiliary
auxiliary
auxiliarypseudo-
pseudo-
pseudo-
Riemannian
Riemannian
Riemannianstructure structure
structureofof ofneutral
neutral
neutralsignature
signature
signaturetoto tocertain
certain
certainaffine
affine
affineconnections
connections
connectionsand and
anduse use
usethis
this
thiscorrespondence
correspondence
correspondencetoto tostudy
study
study
both
both
bothgeometries.
geometries.
geometries. WeWe
Weexamine
examine
examineWalker Walker
Walkerstructures,
structures,
structures,Riemannian
Riemannian
Riemannianextensions,and
extensions,and
extensions,andKähler Kähler
KählerWeylWeyl
Weylgeometry
geometry
geometryfrom from
fromthis this
this

APPLICATIONS OF AFFINE AND WEYL GEOMETRY

APPLICATIONS OF
APPLICATIONS
viewpoint.This
viewpoint.This
viewpoint.Thisbook book
bookisisisintended
intended
intendedtoto tobebebeaccessible
accessible
accessibletoto tomathematicians
mathematicians
mathematicianswho who
whoare are
arenot
not
notexpert
expert
expertinin inthe
the
thesubject
subject
subjectand
and
andtototo
students
students
studentswithwith
witha abasicabasic
basicgrounding
grounding
groundinginin indifferential
differential
differentialgeometry.
geometry.
geometry. Consequently,
Consequently,
Consequently, the
the
thefirst
first
firstchapter
chapter
chaptercontains
contains
containsa acomprehensive
acomprehensive
comprehensive
introduction
introduction
introductiontoto tothe
the
thebasic
basic
basicresults
results
resultsandand
anddefinitions
definitions
definitionswe we
weshall
shall
shallneed
need
needproofs
proofs
proofsareare
areincluded
included
includedofof ofmany
many
manyofof ofthese
these
theseresults
results
resultstototo
make
make
makeitititasas
asself-contained
self-contained
self-containedasas aspossible.
possible.
possible.Para-complex
Para-complex
Para-complexgeometry geometry
geometryplays plays
playsanan
animportant
important
importantrole role
rolethroughout
throughout
throughoutthe the
thebook
book
book
and
and
andconsequently
consequently
consequentlyisisistreated treated
treatedcarefully
carefully
carefullyinin invarious
various
variouschapters,
chapters,
chapters,asas asisisisthe
the
therepresentation
representation
representationtheory theory
theoryunderlying
underlying
underlyingvariousvarious
various
results.
results.
results.ItItItisisisa aafeature
feature
featureofof ofthis
this
thisbook
book
bookthat, that,
that,rather
rather
ratherthan
than
thanasasasregarding
regarding
regardingpara-complex
para-complex
para-complexgeometry geometry
geometryasas asanananadjunct
adjunct
adjuncttoto to

OF AFFINE
complex
complex
complexgeometry,
geometry,
geometry,instead, instead,
instead,we we
weshall
shall
shalloften
often
oftenintroduce
introduce
introducethe the
thepara-complex
para-complex
para-complexconcepts concepts
conceptsfirst first
firstand
and
andonly
only
onlylater
later
laterpass
pass
passtoto
tothe
the
the
Eduardo
EduardoGarcía-Río
García-Río

AFFINE AND
complex
complex
complexsetting.
setting.
setting.
The
The
Thesecond
second
secondand and
andthird
third
thirdchapters
chapters
chaptersare are
aredevoted
devoted
devotedtoto tothe
the
thestudy
study
studyofofofvarious
various
variouskinds
kinds
kindsofof ofRiemannian
Riemannian
Riemannianextensions extensions
extensionsthat that
that
Peter
PeterGilkey
Gilkey

AND WEYL
associate
associate
associatetototoanan
anaffine
affine
affinestructure
structure
structureon on
ona aamanifold
manifold
manifolda aacorresponding
corresponding
correspondingmetric metric
metricofof
ofneutral
neutral
neutralsignature
signature
signatureon on
onitsits
itscotangent
cotangent
cotangent
bundle.
bundle.
bundle. These
These
Theseplayplay
playa aarole
role
roleinin
invarious
various
variousquestions
questions
questionsinvolving
involving
involvingthe the
thespectral
spectral
spectralgeometry
geometry
geometryofof ofthethe
thecurvature
curvature
curvatureoperator operator
operatorand and
and
Stana
StanaNikčević
Nikčević

WEYL GEOMETRY
homogeneous
homogeneous
homogeneousconnections
connections
connectionson on
onsurfaces.
surfaces.
surfaces. The
The
Thefourth
fourth
fourthchapter
chapter
chapterdeals
deals
dealswith
with
withKähler
Kähler
KählerWeylWeyl
Weylgeometry,
geometry,
geometry,which which
whichlies,lies,
lies,inin
ina aa
certain
certain
certainsense,
sense,
sense,midway
midway
midwaybetweenbetween
betweenaffine affine
affinegeometry
geometry
geometryand and
andKähler
Kähler
Kählergeometry.
geometry.
geometry.Another
Another
Anotherfeaturefeature
featureofof
ofthe the
thebook
book
bookisisisthat
that
thatwe we
we

Ramón
RamónVázquez-Lorenzo
Vázquez-Lorenzo

GEOMETRY
have
have
havetried
tried
triedwherever
wherever
whereverpossible
possible
possibletoto tofind
find
findthethe
theoriginal
original
originalreferences
references
referencesinin inthe
the
thesubject
subject
subjectfor
for
forpossible
possible
possiblehistorical
historical
historicalinterest.
interest.
interest. Thus,
Thus,
Thus,
we
we
wehave
have
havecited
cited
citedthe
the
theseminal
seminal
seminalpapers
papers
papersofof ofLevi-Civita,
Levi-Civita,
Levi-Civita,Ricci,
Ricci,
Ricci,Schouten,
Schouten,
Schouten,and and
andWeyl,
Weyl,
Weyl,tototoname
name
namebut but
buta aafew
few
fewexemplars.
exemplars.
exemplars.We We
We
have
have
havealso
also
alsogiven
given
givendifferent
different
differentproofs
proofs
proofsofof ofvarious
various
variousresults
results
resultsthan
than
thanthose
those
thosethat
that
thatare
are
aregiven
given
giveninin
inthe
the
theliterature,
literature,
literature,toto totake
take
takeadvantage
advantage
advantage
ofof
ofthe
the
theunified
unified
unifiedtreatment
treatment
treatmentofof ofthe
the
thearea
area
areagiven
given
givenherein.
herein.
herein.

About
About
AboutSYNTHESIs
SYNTHESIs
SYNTHESIs
This
This
Thisvolume
volume
volumeisisisa aaprinted
printed
printedversion
version
versionofof
ofa aawork
work
workthat
that
thatappears
appears
appearsinin
inthe
the
theSynthesis
Synthesis
Synthesis

MOR GAN & CL AYPOOL

MOR GAN
MOR
Digital
Digital
DigitalLibrary
Library
Libraryofof ofEngineering
Engineering
Engineeringand and
andComputer
Computer
ComputerScience.
Science.
Science.Synthesis
Synthesis
SynthesisLectures
Lectures
Lectures
provide
provide
provideconcise,
concise,
concise,original
original
originalpresentations
presentations
presentationsofof ofimportant
important
importantresearch
research
researchand
and
anddevelopment
development
development

GAN &
topics,
topics,
topics,published
published
publishedquickly,
quickly,
quickly,inin
indigital
digital
digitaland
and
andprint
print
printformats.
formats.
formats.For
For
Formore
more
moreinformation
information
information

SSYNTHESIS
YNTHESISL
LECTURES
visit
visit
visitwww.morganclaypool.com
www.morganclaypool.com
www.morganclaypool.com
YNTHESIS ECTURES
ECTURESONON
ON

& CL
MMATHEMATICS ANDSSTATISTICS

CL AYPOOL
ISBN:
ISBN:
ISBN:978-1-60845-759-5
978-1-60845-759-5
978-1-60845-759-5
Mor
Mor
Morgan
gan
gan Cl
Cl
Claypool
aypool &
&
&
aypoolPublishers
Publishers
Publishers 90000
90000
90000
ATHEMATICS
ATHEMATICSAND
AND TATISTICS
TATISTICS

AYPOOL
ww
www
www
w. .m
.m
mooor rgr gga aannnc ccl lal aay yypppooo ool l.l .c. ccooomm
m
99
9781608
781608
781608457595
457595
457595
Steven
Steven
StevenG.
G.
G.Krantz,
Krantz,
Krantz,Series
Series
SeriesEditor
Editor
Editor
Applications of
Affine and Weyl Geometry
 Synthesis Lectures on
Mathematics and Statistics
Editor
Steven G. Krantz, Washington University, St. Louis

Applications of Affine and Weyl Geometry

Eduardo García-Río, Peter Gilkey, Stana Nikčević, and Ramón Vázquez-Lorenzo
2013

Essentials of Applied Mathematics for Engineers and Scientists, Second Edition

Robert G. Watts
2012

Chaotic Maps: Dynamics, Fractals, and Rapid Fluctuations

Goong Chen and Yu Huang
2011

Matrices in Engineering Problems

Marvin J. Tobias
2011

The Integral: A Crux for Analysis

Steven G. Krantz
2011

Statistics is Easy! Second Edition

Dennis Shasha and Manda Wilson
2010

Lectures on Financial Mathematics: Discrete Asset Pricing

Greg Anderson and Alec N. Kercheval
2010

Jordan Canonical Form: Theory and Practice

Steven H. Weintraub
2009
 iii
The Geometry of Walker Manifolds
Miguel Brozos-Vázquez, Eduardo García-Río, Peter Gilkey, Stana Nikcevic, and Ramón
Vázquez-Lorenzo
2009

An Introduction to Multivariable Mathematics

Leon Simon
2008

Jordan Canonical Form: Application to Differential Equations

Steven H. Weintraub
2008

Statistics is Easy!
Dennis Shasha and Manda Wilson
2008

A Gyrovector Space Approach to Hyperbolic Geometry

Abraham Albert Ungar
2008
Copyright © 2013 by Morgan & Claypool

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in
any form or by any means—electronic, mechanical, photocopy, recording, or any other except for brief quotations in
printed reviews, without the prior permission of the publisher.

Applications of Affine and Weyl Geometry

Eduardo García-Río, Peter Gilkey, Stana Nikčević, and Ramón Vázquez-Lorenzo
www.morganclaypool.com

ISBN: 9781608457595 paperback

ISBN: 9781608457601 ebook

DOI 10.2200/S00502ED1V01Y201305MAS013

A Publication in the Morgan & Claypool Publishers series

SYNTHESIS LECTURES ON MATHEMATICS AND STATISTICS

Lecture #13
Series Editor: Steven G. Krantz, Washington University, St. Louis
Series ISSN
Synthesis Lectures on Mathematics and Statistics
Print 1938-1743 Electronic 1938-1751
Applications of
Affine and Weyl Geometry

Eduardo García-Río
University of Santiago de Compostela

Peter Gilkey
University of Oregon

Stana Nikčević
Mathematical Institute, Sanu, Serbia

Ramón Vázquez-Lorenzo
University of Santiago de Compostela

SYNTHESIS LECTURES ON MATHEMATICS AND STATISTICS #13

M
&C Morgan & cLaypool publishers
ABSTRACT
Pseudo-Riemannian geometry is, to a large extent, the study of the Levi-Civita connection, which is
the unique torsion-free connection compatible with the metric structure. There are, however, other
affine connections which arise in different contexts, such as conformal geometry, contact structures,
Weyl structures, and almost Hermitian geometry. In this book, we reverse this point of view and
instead associate an auxiliary pseudo-Riemannian structure of neutral signature to certain affine
connections and use this correspondence to study both geometries. We examine Walker structures,
Riemannian extensions, and Kähler–Weyl geometry from this viewpoint.This book is intended to be
accessible to mathematicians who are not expert in the subject and to students with a basic grounding
in differential geometry. Consequently, the first chapter contains a comprehensive introduction to
the basic results and definitions we shall need—proofs are included of many of these results to make
it as self-contained as possible. Para-complex geometry plays an important role throughout the book
and consequently is treated carefully in various chapters, as is the representation theory underlying
various results. It is a feature of this book that, rather than as regarding para-complex geometry as
an adjunct to complex geometry, instead, we shall often introduce the para-complex concepts first
and only later pass to the complex setting.
The second and third chapters are devoted to the study of various kinds of Riemannian
extensions that associate to an affine structure on a manifold a corresponding metric of neutral
signature on its cotangent bundle. These play a role in various questions involving the spectral
geometry of the curvature operator and homogeneous connections on surfaces. The fourth chapter
deals with Kähler–Weyl geometry, which lies, in a certain sense, midway between affine geometry
and Kähler geometry. Another feature of the book is that we have tried wherever possible to find
the original references in the subject for possible historical interest. Thus, we have cited the seminal
papers of Levi-Civita, Ricci, Schouten, and Weyl, to name but a few exemplars. We have also given
different proofs of various results than those that are given in the literature, to take advantage of the
unified treatment of the area given herein.

KEYWORDS
curvature decomposition, deformed Riemannian extension, Kähler–Weyl geometry,
modified Riemannian extension, Riemannian extension, spectral geometry of the cur-
vature operator
 vii

This book is dedicated to

Carmen, Emily, George, Hugo, Luis, Manuel, Montse, and Susana.
 ix

Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

1 Basic Notions and Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Basic Manifold Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Curvature Models in the Real Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Kähler Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Curvature Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.6 Walker Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.7 Metrics on the Cotangent Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.8 Self-dual Walker Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.9 Recurrent Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.10 Constant Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.11 The Spectral Geometry of the Curvature Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2 The Geometry of Deformed Riemannian Extensions . . . . . . . . . . . . . . . . . . . . . . . . 53

2.1 Basic Notational Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2 Examples of Affine Osserman Ivanov–Petrova Manifolds . . . . . . . . . . . . . . . . . . . . 57
2.3 The Spectral Geometry of the Curvature Tensor of Affine Surfaces . . . . . . . . . . . 60
2.4 Homogeneous 2-Dimensional Affine Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.5 The Spectral Geometry of the Curvature Tensor of Deformed Riemannian
Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3 The Geometry of Modified Riemannian Extensions . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.1 Four-dimensional Osserman Manifolds and Models . . . . . . . . . . . . . . . . . . . . . . . . 77
3.2 para-Kähler Manifolds of Constant para-holomorphic Sectional Curvature . . . . 86
3.3 Higher-dimensional Osserman Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.4 Osserman Metrics with Non-trivial Jordan Normal Form . . . . . . . . . . . . . . . . . . . . 95
3.5 (Semi) para-complex Osserman Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
x
4 (para)-Kähler–Weyl Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.1 Notational Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.2 (para)-Kähler–Weyl Structures if m ≥ 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.3 (para)-Kähler–Weyl Structures if m = 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.4 (para)-Kähler–Weyl Lie Groups if m = 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.5 (para)-Kähler–Weyl Tensors if m = 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.6 Realizability of (para)-Kähler–Weyl Tensors if m = 4 . . . . . . . . . . . . . . . . . . . . . . 123

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Authors’ Biographies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
 xi

Preface
The fundamental theorem of pseudo-Riemannian geometry associates to each pseudo-Riemannian
metric g a unique affine connection,
∇ = g∇ ,

called the Levi-Civita connection (we refer to Levi-Civita [151] and to Ricci and Levi-Civita [188]),
and pseudo-Riemannian geometry focuses, to a large extent, on the geometry of this connection.
There are, however, other natural connections that play a role when considering different geometric
structures, such as almost Hermitian structures, almost contact structures, and Weyl structures. Affine
connections also arise naturally in conformal geometry. In all these cases, the affine connections under
consideration are adapted to the structures under investigation. One can reverse this point of view
and associate an auxiliary pseudo-Riemannian structure to a given affine connection and then use
this correspondence to examine the geometry of both objects. As an exemplar, the Riemannian
extension is a natural, neutral signature, pseudo-Riemannian metric on the cotangent bundle of an
underlying affine manifold. This construction, which goes back to the work of Patterson and Walker
[177], has played an important role in many investigations.
This book examines a number of different areas of differential geometry which are related to
affine differential geometry—Walker structures, Riemannian extensions, and (para)-Kähler–Weyl
geometry. It is intended to be accessible to graduate students who have had a basic course in differ-
ential geometry, as well as to mathematicians who are not necessarily experts in the subject. For that
reason, Chapter 1 contains a basic introduction to the matters under consideration.The Lie derivative
and bracket, connections, Kähler geometry, and curvature are discussed. The notion of a curvature
model is treated and the basic decomposition theorems of Singer–Thorpe [194], Higa [125, 126],
and Tricerri–Vanhecke [200] are given, not only in the Riemannian, but in the pseudo-Riemannian
and in the para-complex settings as well. Almost para-Hermitian and almost Hermitian structures
are presented both in the Riemannian and pseudo-Riemannian categories.The Gray–Hervella [118]
classification of almost Hermitian structures is extended to the almost para-Hermitian and to the
almost pseudo-Hermitian contexts. The geometry of the cotangent bundle is outlined (tautological
1-form, evaluation map, complete lift) and the various natural metrics on the cotangent bundle
(Riemannian extensions, deformed Riemannian extensions, modified Riemannian extensions) are
defined – these will be examined in further detail in Chapter 2 and Chapter 3. A short introduction
to Walker geometry and recurrent curvature is included. Self-dual Walker metrics are discussed and
it is shown that any such metric is locally isometric to the metric of a Riemannian extension. The
Jacobi operator,
J (x) : y → R(y, x)x ,
xii PREFACE
is introduced and classical results concerning symmetric spaces, spaces of constant sectional curvature,
and constant holomorphic sectional curvature are presented using Jacobi vector fields. Chapter 1
concludes with a brief review of the spectral geometry of the curvature operator. Complete proofs
of a number of results are presented to keep the treatment as self-contained as possible.
Chapter 2 examines the geometry of Riemannian extensions in more detail and extends the
discussion of Chapter 1 in that regard. Riemannian extensions form a natural family of examples
and provide a first link between affine and pseudo-Riemannian geometry. The relevant facts con-
cerning the metric of the classical Riemannian extension are developed in some detail. Riemannian
extensions are then used to study the spectral geometry of the curvature tensor in the affine setting;
affine Osserman surfaces and affine Ivanov–Petrova surfaces are examined in some detail. Chap-
ter 2 concludes with a fairly lengthy treatment of homogeneous affine connections. Homogeneous
connections on surfaces that are not the Levi-Civita connection of a metric of constant curvature
form two natural classes which are not disjoint, and the intersection of the two non-metric classes
is studied using the corresponding Riemannian extensions.
Chapter 3 presents generalizations of Riemannian extensions. While Riemannian extensions
are useful in constructing self-dual Ricci flat metrics, the modified Riemannian extensions are a
source of non-Ricci flat Einstein metrics. Four-dimensional geometry is explored in some detail. A
new approach, based on the generalized Goldberg–Sachs theorem, is used to obtain some previously
known results on the classification of 4-dimensional Osserman metrics from this point of view.
The usefulness of modified Riemannian extensions is made clear when discussing 4-dimensional
Osserman metrics whose Jacobi operators have non-degenerate Jordan normal form. Para-Kähler
manifolds of constant para-holomorphic sectional curvature are treated in this fashion, as are a variety
of higher-dimensional Osserman metrics – it is shown that any such manifold is locally isometric to
the modified Riemannian extension metric of a flat affine manifold. By considering non-flat affine
Osserman connections, one obtains a family of deformed para-Kähler metrics with the same spectrum
of the Jacobi operator. This provides a useful family of Osserman metrics whose Jacobi operators
have non-trivial Jordan normal form. The related property of being (semi) para-complex Osserman
is also considered, and it is shown that any modified Riemannian extension is a semi para-complex
Osserman metric. Here, the curvature identities induced by the almost para-complex structure play
an essential role; note that the para-holomorphic sectional curvature does not necessarily determine
the curvature in the general setting.
Chapter 4 treats (para)-Kähler–Weyl geometry. Classical Weyl geometry is, in a certain sense,
midway between affine and Riemannian geometry, and (para)-Kähler–Weyl geometry is a natural
generalization of Kähler geometry. Results are presented both in the complex and para-complex
settings. Only the 4-dimensional setting is of interest in this context – any (para)-Kähler–Weyl
structure is trivial in dimension m ≥ 6. By contrast, any para-Hermitian manifold or any pseudo-
Hermitian manifold admits a unique (para)-Kähler–Weyl structure if m=4. The alternating Ricci
tensor ρa carries essential information about the structure in dimension four – the (para)-Kähler–
Weyl structure is trivial if and only if ρa = 0. All the possible algebraic possibilities for the values of ρa
 PREFACE xiii
can be realized by left-invariant structures on 4-dimensional Lie groups for Hermitian manifolds or
para-Hermitian manifolds – the case of pseudo-Hermitian manifolds of signature (2, 2) is excluded,
as quite different techniques would be needed to study that case that are tangential to the thrust
of our main discussion. This result is used to examine the space of algebraic (para)-Kähler–Weyl
curvature tensors as a module over the appropriate structure groups and to show that any algebraic
possibility can be realized geometrically.

Eduardo García-Río, Peter Gilkey, Stana Nikčević, and Ramón Vázquez-Lorenzo

April 2013
 xv

Acknowledgments
The research of all of the authors was partially supported by Project MTM2009-07756 (Spain); the
research of P. Gilkey and S. Nikčević was partially supported by Project 174012 (Srbija). The authors
are very grateful to Esteban Calviño-Louzao for assistance in proofreading.

Eduardo García-Río, Peter Gilkey, Stana Nikčević, and Ramón Vázquez-Lorenzo

April 2013
 1

CHAPTER 1

Basic Notions and Concepts

In this chapter, we establish notation and introduce the fundamental concepts that we will be dealing
with. Some of the basic theory of manifolds that is needed is presented in Section 1.1. Section 1.2
introduces the notion of a connection and the associated curvature operator on a general vector
bundle. Of particular interest is the case when the bundle in question is the tangent bundle. The
torsion tensor T is introduced; a connection is called affine if T vanishes.The Levi-Civita connection
is the unique torsion-free Riemannian connection; the notion of a Weyl structure is midway between
that of an affine structure and the Levi-Civita connection. In Section 1.3, curvature models in the real
setting are examined. Basic geometric realizability results are given in Theorem 1.12. In Section 1.4,
we pass to the complex and to the para-complex settings. A formalism is introduced that enables us to
treat both contexts in a parallel notation. Section 1.5 deals with curvature decompositions and basic
representation theory. The Singer–Thorpe [194], Higa [125, 126], and Tricerri–Vanhecke [200]
decompositions are given for the curvature tensor in the Riemannian, the Weyl, and the Hermitian
settings, respectively. Section 1.6 introduces Walker structures and Riemannian extensions. It treats
the holonomy group, parallel plane fields, and Walker coordinates. Section 1.7 deals with metrics
on the cotangent bundle: Riemannian extensions, deformed Riemannian extensions, and modified
Riemannian extensions. These notions will be analyzed in more detail subsequently. Section 1.8
discusses the Hodge  operator, self-duality, and self-dual Walker metrics. Section 1.9 is concerned
with recurrent curvature. Section 1.10 deals with constant curvature and introduces the notions of a
Jacobi vector field, of a symmetric space, of constant sectional curvature, and of constant holomorphic
sectional curvature. Section 1.11 provides an introduction to the spectral geometry of the curvature.
Osserman manifolds, higher order Osserman manifolds, and the skew-symmetric curvature operator
are introduced.

1.1 BASIC MANIFOLD THEORY

Let M be a connected smooth manifold of dimension m—for further information see Kobayashi
and Nomizu [140]. If x = (x 1 , . . . , x m ) is a system of local coordinates on M, set

∂
∂xi := ;
∂x i

{∂x1 , . . . , ∂xm } and {dx 1 , . . . , dx m } provide local coordinate frames for the tangent bundle T M and for
the cotangent bundle T ∗ M. We shall also, sometimes, use the notation ∂i for ∂xi when no confusion
is likely to result. We shall, for the most part, follow the convention of writing indices up that
2 1. BASIC NOTIONS AND CONCEPTS
refer to contravariant tensors (coordinate systems, differential forms, etc.) and indices down that
refer to covariant tensors (partial differential operators). An exception will occur to this convention
presently, when we discuss the canonical coordinates on the cotangent bundle (see Equation (1.1.e)).
Thus, for example, if I = {1 ≤ i1 < · · · < ip ≤ m} is a multi-index, let dx I := dx i1 ∧ · · · ∧ dx ip ;
the {dx I }|I |=p provide a local frame for the bundle of p-forms p M. Let S 2 (M) be the bundle of
symmetric 2-cotensors on M. If ξ and η are 1-forms, the symmetric tensor product of ξ with η is defined
by setting:
ξ ◦ η := 21 {ξ ⊗ η + η ⊗ ξ } . (1.1.a)
The {dx i ◦ dx j }1≤i≤j ≤m form a local frame for S 2 (M). Let d : C ∞ (p M) → C ∞ (p+1 M) be
the exterior derivative. Here, and henceforth, we will adopt the Einstein convention and sum over
repeated indices. Thus, for example, we may express:
 
 
∞
df = ∂xi f · dx for f ∈ C (M) and d
i
fI dx =
I
dfI ∧ dx I .
I I

Since d 2 = 0, the de Rham cohomology groups may be defined by setting:

 
p ker d : C ∞ (p M) → C ∞ (p+1 M)
HDeR (M) :=  .
Range d : C ∞ (p−1 M) → C ∞ (p M)

These groups are isomorphic to the ordinary topological cohomology groups (see de Rham [185]),
so this provides an important link between differential geometry and topology. Let

δ : C ∞ (p M) ← C ∞ (p+1 M) (1.1.b)

be the dual map, the coderivative. One forms the Laplacian

:= dδ + δd : C ∞ (p M) → C ∞ (p M) .
p
The Hodge–de Rham theorem [128] identifies ker{ } with HDeR (M) if M is closed. There are
suitable extensions of this result to the context of manifolds with boundary—see, for example, the
discussion in Gilkey [95].

SYMPLECTIC AND CONTACT STRUCTURES

If m = 2m̄ is even, then a symplectic structure on M is a 2-form such that

d = 0 and m̄
= 0 .

The pair (M, ) is then called a symplectic manifold. As we shall see presently in Equation (1.1.f),
the cotangent bundle always admits a canonical symplectic structure. Also, any (para)-Kähler metric
defines a canonical symplectic structure—see Equation (1.5.c).
 1.1. BASIC MANIFOLD THEORY 3
If (M, ) is a symplectic manifold of dimension m = 2m̄, then by Darboux’s theorem [61]
there always exist local coordinates (pi , q i ) so that

m̄
= dpi ∧ dq i .
i=1

Thus, there are no local invariants in symplectic geometry. The structure has its origin in classical
mechanics—the phase space in Hamiltonian mechanics has the structure of a symplectic manifold
where the p coordinates describe the momenta and the q coordinates describe the position of a
particle. For further details, we refer to de Gosson [116], to McDuff and Salamon [154], and to
Weinstein [205].
If m = 2m̄ + 1 is odd, then the natural analogue of symplectic geometry is contact geometry.
Let η be a smooth 1-form on M. We say that η is a contact 1-form and that the pair (M, η) is a
contact manifold if
η ∧ (dη)m̄  = 0 .
The Reeb vector field ξ is characterized by the property that
η(ξ ) = 1 and dη(ξ, X) = 0 for all X ∈ C ∞ (T M) .
We refer to Geiges [93] for further details.

THE LIE DERIVATIVE AND THE LIE BRACKET

Let X ∈ C ∞ (T M) be a smooth vector field on M. A flow curve for X starting at the point P is the
solution to the equation:
α̇(t) = X(α(t)) with α(0) = P .
The smooth 1-parameter flow associated to X is the family of local diffeomorphisms comprising the
collection of flow curves. More precisely, this is a family of local diffeomorphisms ϕtX of M so that
the curves α : t → ϕtX (P ) are flow curves:
ϕ̇tX (P ) = X(ϕtX (P )) with ϕ0X (P ) = P for all P ∈ M .
We note that
ϕtX ϕsX = ϕs+t
X
.
If is a tensor field on M (i.e., ∈ ⊗p T ∗ M ⊗q T M for some (p, q)), then we can form the tensor
(ϕtX )∗ . The Lie derivative is then defined by setting:
 
LX := ∂t (ϕtX )∗ .
t=0

Let P be a point of M. If X(P ) = 0, we can choose a system of local coordinates (x 1 , . . . , x m ) so

that X = ∂x1 and so that the flow is given locally by
ϕtX (x 1 , . . . , x m ) = (x 1 + t, x 2 , . . . , x m ) .
4 1. BASIC NOTIONS AND CONCEPTS
If we express
j ,...,j
= ai11,...,ipq · dx i1 ⊗ · · · ⊗ dx ip ⊗ ∂xj1 ⊗ · · · ⊗ ∂xjq ,
we then have
j ,...,j
LX = ∂x1 ai11,...,ipq · dx i1 ⊗ · · · ⊗ dx ip ⊗ ∂xj1 ⊗ · · · ⊗ ∂xjq .

Let [X, Y ] denote the Lie bracket of two vector fields X, Y ∈ C ∞ (T M). This is characterized
by the identity:
[X, Y ]f = X(Y (f )) − Y (X(f )) for all f ∈ C ∞ (M) .
We can express the Lie bracket in a system of local coordinates as follows. If X = a i ∂xi and if
Y = bj ∂xj are smooth vector fields on M, then:
 
[X, Y ] = a i ∂xi (bj ) − bi ∂xi (a j ) ∂xj .

One has the Jacobi identity:

[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0 .

Let  ·, ·  denote the natural bilinear pairing between a vector bundle V and the dual
bundle V∗ . It is characterized on the tangent bundle by the identity:

j j 1 if i = j
 ∂xi , dx j = δi where δi :=
0 if i  = j

is the Kronecker symbol. The Lie bracket and the exterior derivative are related by the formula:

dω(X, Y ) = X  Y, ω  −Y  X, ω  −  [X, Y ], ω  . (1.1.c)

Similarly, the Lie bracket and the Lie derivative are related by the formula:

[X, Y ] = LX Y .

The Lie bracket and the flows are also related (see Spivak [196, Chapter 5 vol. 1]). The commutator
flow
c(t; P ) := ϕ−
Y√
(ϕ X√ (ϕ√
t − t
Y
t
X
(ϕ√ t
(P )))) satisfies ċ(t; P )|t=0 = [X, Y ](P ) .

GEOMETRY OF THE COTANGENT BUNDLE

We refer to Jost [137] and to Yano and Ishihara [215] for further details concerning this material.
Let σ : T ∗ M → M be the natural projection from the cotangent bundle to the base manifold. We
may express any point P̃ of T ∗ M in the form P̃ = (P , ω) where P := σ (P̃ ) belongs to M and
where ω belongs to TP∗ M. Let m̄ = dim(M) and let

x = (x 1 , . . . , x m̄ )
 1.1. BASIC MANIFOLD THEORY 5
be a system of local coordinates on an open neighborhood U of P ∈ M. Expanding

ω = xi  dx i (1.1.d)

then defines a system of local coordinates on Ũ := σ −1 U ⊂ T ∗ M of the form

(x 1 , . . . , x m̄ ; x1 , . . . , xm̄ ) . (1.1.e)

The 1-form ω of Equation (1.1.d) is often referred to as the tautological 1-form, although, it is
also denoted as the Poincaré 1-form or the Liouville 1-form. The dual coordinates are written with
the index down, as they transform covariantly rather than contravariantly and this permits us to
retain the formalism of summing over repeated indices. Thus the canonical symplectic structure on
the cotangent bundle is given in terms of a system of local coordinates by:

:= dω = dxi  ∧ dx i . (1.1.f )

We will show presently that is invariantly defined and is independent of the particular coordinate
system chosen for its evaluation by giving an invariant characterization in Equation (1.1.h). Note
that is a closed differential form with m̄  = 0.
Let X ∈ C ∞ (T M) be a smooth vector field on M.The evaluation map ιX is a smooth function
on the cotangent bundle T ∗ M which is defined by the identity:

ιX(P , ω) = ω(XP ) .

We may express X = X i ∂xi to define the coefficients X i = X, dx i . We then have that

ιX(x i , xi  ) = xi  X i = xi   X, dx i  .

The special significance of the evaluation map comes from the fact that vector fields on T ∗ M
are characterized by their action on the evaluation maps ιX (see Yano and Ishihara [215]). More
precisely:

Lemma 1.1 Suppose that Ỹ and Z̃ are smooth vector fields on the cotangent bundle T ∗ M. Suppose that
Ỹ (ιX) = Z̃(ιX) for all smooth vector fields X on M. Then Ỹ = Z̃.

x , x )∂xi + bi  (
Proof. Let Ỹ = a i ( x , x )∂x i  be a smooth vector field on T ∗ M such that Ỹ (ιX) = 0
for all smooth vector fields on M. We must show this implies that Ỹ = 0. Let X = Xj ( x )∂xj . Since
ιX = xi X , we have:

i

x , x )(∂xi Xj )(

0 = Ỹ (ιX) = xj  a i ( x , x )Xi (
x ) + bi  ( x) .

Fix j and do not sum. If we take X = ∂xj , then Xi = δji so ιX = xj  and

x , x ) so bj  = 0 and Ỹ = a i (
0 = Ỹ (ιX) = bj  ( x , x )∂xi .
6 1. BASIC NOTIONS AND CONCEPTS
If we take X = x j ∂xj , then we have similarly

x , x ) .
0 = Ỹ (ιX) = xj  a j (

x , x ) = 0 when xj   = 0. Since the functions a j (

Thus a j ( x , x ) are smooth, this implies a j vanishes
identically and hence Ỹ = 0 as desired. 2
Let X be a smooth vector field on M. The complete lift XC is the vector field on the cotangent
bundle T ∗ M characterized via Lemma 1.1 by the identity:
XC (ιZ) = ι[X, Z] for all smooth vector fields Z on M.

Lemma 1.2 Let (P , ω) belong to T ∗ M − {0}, i.e., ω  = 0. Then the tangent space T(P ,ω) T ∗ M is
spanned by the complete lifts of all the smooth vector fields on M.

Proof. We first compute the complete lift in a system of local coordinates. Let

x , x )∂xj + bj  (
x )∂xj and XC = a j (
X = X j ( x , x )∂x j  .

Let Z = Z i (
x )∂xi . Then

x , x )(∂xj Z i )(

XC (ιZ) = XC (xi  Z i ) = xi  a j ( x , x )Z j (
x ) + bj  ( x)
= ι{(X ∂xj Z − Z ∂xj X )∂xi }
j i j i

= xi  (Xj (
x )(∂xj Z i )(x ) − Z j ( x )(∂xj Xi )(x )) .

x , x ) = Xj (
Since Z is arbitrary, we conclude a j ( x , x ) = −xi  ∂xj Xi so that
x ) and bj  (

X C = Xj (
x )∂xj − xi  (∂xj Xi )(
x )∂x j  . (1.1.g)

Taking X = ∂xj then yields XC = ∂xj . Since ω  = 0, we have xi   = 0 for some i. Taking X = x j ∂xi
then yields X C = x j ∂xi − xi  ∂x j  which completes the proof. We note that this fails on the 0-section;
if ω = 0, then
Span{XC } = Span{∂xj }. 2

Remark 1.3 A crucial point is that, since ιX and XC are invariantly characterized, we do not need
to check the local formalism transforms correctly; on the other hand, the local formalism shows that
X C in fact exists. We shall apply similar arguments subsequently.

We use Lemma 1.2 to prove the following result—see also Yano and Ishihara [215]:

Lemma 1.4 Two smooth tensor fields 1 , 2 of type (0, s) on T ∗ M coincide with each other if and only
if we have the following identity for all vector fields Xi on M:
 1.1. BASIC MANIFOLD THEORY 7
C C
1 (Xi1 , . . . , Xis ) = C C
2 (Xi1 , . . . , Xis ).

Proof. This is immediate if ω  = 0 by Lemma 1.2; we now use continuity to extend the result to the
0-section. 2
We use Lemma 1.4 to show that the symplectic 2-form = dxi  ∧ dx i of Equation (1.1.f)
is independent of the system of local coordinates. By Equation (1.1.g),

(X C , Y C ) = (Xj ∂xj − xi  (∂xj Xi )∂x j  , Y j ∂xj − xi  (∂xj Y i )∂x j  )

= xi  {−Y j ∂xj X i + X j ∂xj Y i },
ι[X, Y ] = ι((Xj ∂xj Y i − Y j ∂xj Xi )∂xi )
= xi  {−Y j ∂xj X i + X j ∂xj Y i }.

Consequently, the symplectic 2-form of Equation (1.1.f) may be characterized invariantly by the
identity:
(X C , Y C ) = ι[X, Y ] . (1.1.h)
Let T ∈ C ∞ (End{T M}) be a smooth (1, 1)-tensor field on M. Define a corresponding lifted
1-form ιT ∈ C ∞ (T ∗ (T ∗ M)) on the cotangent bundle that is characterized by the identity:

 ιT , XC = ι(T X) .

x , x )dx i + bi (
We compute in local coordinates as follows. Expand ιT = ai ( x , x )dxi  and
j
T ∂xi = Ti ∂xj . Let X = X ∂xi . We compute:
i

x , x )Xi (
(ιT )(X C ) = ai ( x , x )xi  (∂xj Xi ),
x ) − bj (
j j
ι(T X) = ι(X i Ti ∂xj ) = xj  X i Ti .
j
This shows that bj = 0 and that ai = xj  Ti , i.e., that
j j
ιT = xj  Ti dx i where T = Ti ∂xj ⊗ dx i .

We also refer to Kowalski and Sekizawa [145, 146] for additional information concerning natural lifts
in Riemannian geometry. The following result summarizes the formalisms that we have established
in this section:

Lemma 1.5

1. The evaluation map ιX for X ∈ C ∞ (T M):

(a) Invariant formalism: ιX(P , ω) = ω(XP ).

(b) Coordinate formalism: ιX = xi  X i for X = Xi ∂xi .
8 1. BASIC NOTIONS AND CONCEPTS
2. The complete lift of a vector field X ∈ C ∞ (T M).

(a) Invariant formalism: X C (ιZ) = ι[X, Z] for Z ∈ C ∞ (T M).

(b) Coordinate formalism: X C = Xj ∂xj − xi  (∂xj Xi )∂x j  for X = Xj ∂xj .

3. Symplectic form on T ∗ M.

(a) Invariant formalism: (X C , Y C ) = ι[X, Y ].

(b) Coordinate formalism: = dxi  ∧ dx i .

4. Let T ∈ C ∞ (End{T M}); ιT is the 1-form on T ∗ M:

(a) Invariant formalism:  ιT , XC = ι(T X).

j
(b) Coordinate formalism: ιT = xj  Ti dx i .

1.2 CONNECTIONS
The study of an invariantly defined directional derivative probably began with Christoffel [54].
Subsequently, additional work was done by Ricci and Levi-Civita [186, 188]. We also refer to early
work by Cartan [48]. If V is a vector bundle over M, let C ∞ (V) denote the space of smooth sections to
V. A connection ∇ on V is a first-order partial differential operator from C ∞ (V) to C ∞ (T ∗ M ⊗ V),
which satisfies the Leibnitz formula:

∇(f s) = df ⊗ s + f ∇s for s ∈ C ∞ (V) . (1.2.a)

The associated directional covariant derivative ∇X s is defined by setting

∇X s := X, ∇s  for s ∈ C ∞ (V) and X ∈ C ∞ (T M) ,

where we use  ·, ·  to denote the natural pairing from T M ⊗ T ∗ M ⊗ V to V. If {ei } is a basis

for T M and if {ei } is the associated dual basis for T ∗ M, the total covariant derivative is then given
in terms of the directional covariant derivatives by

∇s = ei ⊗ ∇ei s ,

where, as noted previously, we adopt the Einstein convention and sum over repeated indices. If V is a
complex vector bundle, we require ∇ to be complex linear. We extend ∇ to a dual connection ∇ ∗ on
V∗ by requiring that

d  s, s ∗ = ∇s, s ∗  +  s, ∇ ∗ s ∗  for all s ∈ C ∞ (V) and s ∗ ∈ C ∞ (V∗ ) .

 1.2. CONNECTIONS 9
Let s = (s1 , . . . , sk ) be a local frame for V and let x = (x 1 , . . . , x m ) be a system of local
coordinates on M. We may expand
∇∂xi sa = ia b sb .
Here, the index i ranges from 1 to m := dim(M) and the indices a, b range from 1 to k, i.e., to the
fiber dimension of V. The ia b = ∇ ia b are referred to as the Christoffel symbols of the first kind or
sometimes simply as the Christoffel symbols of the connection ∇. They are not tensorial. In view of
the Leibnitz formula given in Equation (1.2.a), ∇ is determined by the Christoffel symbols as we see
by computing:
∇α i ∂x (ν a sa ) = α i {ν a ia b sb + ∂xi (ν a )sa } .
i
∗
Let s = be the local dual frame field for the dual bundle V∗ . The dual Christoffel
(s 1 , . . . , s k )
symbols for the dual connection on V∗ are given by the identity:
∇∂∗x s b = −ia b s a .
i

If V is equipped with a non-degenerate inner product or fiber metric h, then we say that ∇ is
a Riemannian connection in the real setting or a Hermitian connection in the complex setting if the
following relation holds:
dh(s1 , s2 ) = h(∇s1 , s2 ) + h(s1 , ∇s2 ) for si ∈ C ∞ (V) .
Equivalently, ∇ is Riemannian if and only if ∇ agrees with ∇ ∗ when we use h to identify V with V∗ .
We can lower indices and define the Christoffel symbols of the second kind by setting:
iab := h(∇∂xi sa , sb ) .
If s is a local orthonormal frame for V, then ∇ is Riemannian if and only if we have the symmetry:
iab + iba = 0 for all i, a, b .

THE CURVATURE OPERATOR

We refer to Besse [12], to Kobayashi and Nomizu [140], to Riemann [189], and to Ricci and
Levi-Civita [188] for further details concerning the material of this section. The curvature operator
R = ∇ R of a connection ∇ is defined by setting:
R(X, Y )s := {∇X ∇Y − ∇Y ∇X − ∇[X,Y ] }s . (1.2.b)
This is a tensor, i.e., if X, Y ∈ C ∞ (T M), s ∈ C ∞ (V), and f ∈ C ∞ (M), then:
R(f X, Y )s = R(X, f Y )s = R(X, Y )f s = f R(X, Y )s .
Let R(∂xi , ∂xj )sa = Rij a b sb be the components of the curvature operator in a system of local coordi-
nates x = (x 1 , . . . , x m ) and relative to a local frame s = (s1 , . . . , sk ) for V. We have
Rij a b = ∂xi j a b − ∂xj ia b + ic b j a c − j c b ia c . (1.2.c)
10 1. BASIC NOTIONS AND CONCEPTS
If h is a non-degenerate inner product on V, we may lower indices to define:

R(X, Y, s, s̃) := h(R(X, Y )s, s̃) and set Rij ab := h(R(∂xi , ∂xj )sa , sb ) . (1.2.d)

If ∇ is Riemannian, and if s is a local orthonormal frame for V, then:

Rij ab + Rij ba = 0 .

AFFINE CONNECTIONS
Of particular interest is the special case in which V is the tangent bundle of M and we shall restrict
to this case henceforth unless otherwise noted. Let ∇ be a connection on T M. Define the Ricci tensor
ρ = ∇ρ, the symmetric Ricci tensor ρs = ∇ρs , and the alternating Ricci tensor ρa = ∇ρa by setting:

ρ(X, Y ) := Tr{Z → R(Z, X)Y },

ρs (X, Y ) := 21 {ρ(X, Y ) + ρ(Y, X)},
ρa (X, Y ) := 21 {ρ(X, Y ) − ρ(Y, X)}.

Let M = (M, g) be a pseudo-Riemannian manifold; here g ∈ C ∞ (S 2 (T ∗ M)) is a non-degenerate

symmetric bilinear form—see Eisenhart [77, 78] for details. We say that M is Riemannian if g is
positive definite and that M is Lorentzian if g has signature (1, m − 1). We can raise indices to
define the Ricci operator Ric = ∇ Ric. The Ricci operator is a smooth section to C ∞ (End{T M}),
which is characterized by the identity:

g(Ric X, Y ) = ρ(X, Y ) .

The Ricci tensor of the Levi-Civita connection is always a symmetric 2-cotensor; a pseudo-
Riemannian manifold is said to be Einstein if the Ricci tensor is a constant multiple of the metric
tensor; we refer to [12, 78, 140, 187] for further details.
If ∇ is a connection on T M, then we may define the torsion tensor T = ∇ T by:

T (X, Y ) = ∇X Y − ∇Y X − [X, Y ] for X, Y ∈ C ∞ (T M) .

The torsion is a tensor, i.e.,

T (f X, Y ) = T (X, f Y ) = f T (X, Y ) for X, Y ∈ C ∞ (T M) and f ∈ C ∞ (M) .

One has the following useful fact; it permits one to normalize the choice of the frame so that only
the second derivatives of the Christoffel symbols enter into the computation of the curvature:

Lemma 1.6 Let ∇ be a connection on T M and let P ∈ M. The following conditions are equivalent and
if either condition is satisfied at all points of M, then ∇ is said to be an affine connection or, equivalently,
to be a torsion-free connection.
1. T (XP , YP ) = 0 for all XP , YP ∈ TP M.
 1.2. CONNECTIONS 11
2. There exist local coordinates for M centered at P so that ∇ (P ) = 0.

Proof. It is immediate that the torsion tensor T vanishes at P if and only if we have the symmetry
ij k (P ) = j i k (P ). In particular, if there exists a coordinate system where (P ) = 0, then neces-
sarily T vanishes at P . Thus, Assertion (2) implies Assertion (1). Conversely, assume that Assertion
(1) holds. Choose any system of coordinates x = (x 1 , . . . , x m ) on M which are centered at P . Define
a new system of coordinates by setting:

zi = x i + 21 cj k i x j x k ,
j,k

where cj k i = ckj i remains to be chosen. As ∂xj = ∂zj + cj i  x i ∂z ,

∇∂xi ∂xj (P ) = ∇∂zi ∂zj (P ) + cj i  ∂z (P ) .

Set cij  := ij  (P ); the fact that cij  = cj i  is exactly the assumption that the torsion tensor of ∇
vanishes at P . Thus we conclude that the new coordinate system has vanishing Christoffel symbols
at P . 2
Although there are interesting examples where ∇ has torsion (see, for example, Cartan [47]
and Ivanov [133]), we shall henceforth assume ∇ is torsion-free. We say that (M, ∇) is an affine
manifold if ∇ is an affine connection. See, for example, the discussion in [14, 129, 168, 190, 192, 193].
In this setting, the curvature operator has the following symmetries:

R(X, Y )Z + R(Y, X)Z = 0, (1.2.e)

R(X, Y )Z + R(Y, Z)X + R(Z, X)Y = 0 . (1.2.f )

The symmetry of Equation (1.2.f) is called the first Bianchi identity or sometimes just the Bianchi
identity—see Bianchi [13]. The first covariant derivative of the curvature operator is defined by
setting:
∇ R(X1 , X2 ; X4 )X3 := ∇X4 R(X1 , X2 )X3 − R(∇X4 X1 , X2 )X3
(1.2.g)
−R(X1 , ∇X4 X2 )X3 − R(X1 , X2 )∇X4 X3 .
∇ R is a tensor which belongs to ⊗3 T ∗ M ⊗ End{T M} and which has the symmetries:

∇ R(X1 , X2 ; X4 )X3 + ∇ R(X2 , X1 ; X4 )X3 = 0, (1.2.h)

∇ R(X1 , X2 ; X4 )X3 + ∇ R(X2 , X3 ; X4 )X1 + ∇ R(X3 , X1 ; X4 )X2 = 0, (1.2.i)
∇ R(X1 , X2 ; X4 )X3 + ∇ R(X2 , X4 ; X1 )X3 + ∇ R(X4 , X1 ; X2 )X3 = 0 . (1.2.j)

Equation (1.2.i) arises as the covariant derivative of the first Bianchi identity; Equation (1.2.j) is known
as the second Bianchi identity. We can continue this process to define higher covariant derivatives
∇ k R ∈ ⊗2+k (T ∗ M) ⊗ End{T M} for any k.
12 1. BASIC NOTIONS AND CONCEPTS
We say (M, ∇) is flat if R = 0 and that (M, ∇) is locally symmetric if ∇ R = 0. The following
result of Schirokow [190] is a useful observation:

Theorem 1.7 Let ∇ be an affine connection on TM. The following assertions are equivalent. If any is
satisfied, then ∇ is said to be equiaffine or Ricci symmetric.

1. Let ω := ij j dx i . Then dωx = 0 for any system of local coordinates x on M.

2. Tr{R} = 0.

3. The connection ∇ is Ricci symmetric.

4. The connection ∇ locally admits a parallel volume form.

Proof. We use the Bianchi identity of Equation (1.2.f) to see:

0 = Tr{Z → R(X, Y )Z} + Tr{Z → R(Y, Z)X} + Tr{Z → R(Z, X)Y }

= Tr{R(X, Y )} − ρ(Y, X) + ρ(X, Y ) = 0 .

This shows that Assertion (2) and Assertion (3) are equivalent. We show that Assertion (1) and
Assertion (2) are equivalent by computing:

Tr{Rij }dx i ∧ dx j
= {∂xi j k k − ∂xj ik k + in k j k n − j n k ik n }dx i ∧ dx j
= {∂xi j k k − ∂xj ik k }dx i ∧ dx j = 2d{j k k dx j } .

Let e be a conformal factor to rescale the volume form. Since ∇∂∗x dx j = −ik j dx k , we may
i
compute:

∇∂∗x {e dx 1 ∧ · · · ∧ dx m } = {∂xi  − k ik k }{e dx 1 ∧ · · · ∧ dx m } .
i

Thus, there exists a local parallel volume form on an open subset of M if and only if ik k dx i is
exact on that open subset. As every closed 1-form is locally exact, Assertion (1) and Assertion (4)
are equivalent. 2

Geodesics
We say that a parametrized curve γ (t) in M is a geodesic if it satisfies the geodesic equation

∇γ̇ γ̇ = 0 ,

or, equivalently, in a system of coordinates where γ = (γ 1 , . . . , γ m ) we have:

γ̈ i + j k i γ̇ j γ̇ k = 0 .
 1.2. CONNECTIONS 13
In the case of the Levi-Civita connection for a positive definite metric, which will be discussed
presently, this means that γ locally minimizes distance between points on γ . An affine connection
∇ is said to be geodesically complete if every geodesic extends to the parameter range (−∞, ∞).
The exponential map expP : TP M → M is a local diffeomorphism from a neighborhood
of the origin 0 ∈ TP M to a neighborhood of P in M. It is characterized by the fact that the
curves γv (s) := expP (sv) are geodesics in M with initial velocity v ∈ TP M. One says that (M, ∇)
is complete if expP is defined on all TP M. Conjugate points arise where expP fails to be a local
diffeomorphism. Furthermore, since there can be different geodesics joining two points, expP can
fail to be globally one-to-one.

Projective Equivalence and the Projective Curvature Tensor

We refer to Nomizu [166] for further details concerning the following material.

Lemma 1.8 Let ∇ and ∇˜ be connections on T M. Let S(X, Y ) = ∇X Y − ∇˜ X Y be the difference tensor.
1. The following conditions are equivalent and if any is satisfied, then the two connections are said to
be projectively equivalent:
(a) ∇ and ∇˜ have the same unparametrized geodesics.
(b) There exists a smooth 1-form ω such that S(X, X) = 2ω(X)X for all X.
2. If ∇ and ∇˜ are torsion-free, then ∇ and ∇˜ are projectively equivalent if and only if
S(X, Y ) = ω(X)Y + ω(Y )X.
3. If ∇˜ is an arbitrary connection with torsion tensor T̃ , then setting S = − 21 T̃ yields a torsion-free
˜
connection ∇ which is projectively equivalent to ∇.

Proof. Let γ (t) be a geodesic for ∇ and let μ(s) = γ (α(s)) be a reparametrization of γ . Then
μ̇(s) = α̇(s)γ̇ (α(s)). Since ∇γ̇ γ̇ = 0,

∇˜ μ̇ μ̇ = (α̇)2 S(γ̇ , γ̇ ) + α̇ α̈ γ̇ .

Thus, it is possible to reparametrize γ to obtain a geodesic for ∇˜ if and only if S(γ̇ , γ̇ ) is a multiple of
γ̇ ; Assertion (1) now follows. If ∇ and ∇˜ are torsion-free, then the tensor S is symmetric; Assertion
(2) now follows by polarization. Set S = − 21 T̃ . Because T̃ is a skew-symmetric tensor, S(X, X) = 0,
so ∇ is projectively equivalent to ∇.˜ Assertion (3) follows as the torsion tensor for ∇ is given by
T̃ + 2S = 0. 2
If the Ricci tensor ∇ρ is a symmetric tensor, then the Weyl projective curvature operator is
defined by setting:
1
P (X, Y )Z := R(X, Y )Z − {ρ(Y, Z)X − ρ(X, Z)Y } .
m−1
14 1. BASIC NOTIONS AND CONCEPTS
We note that if ∇ and ∇˜ are projectively equivalent, then the corresponding Weyl projective curvature
operators coincide. We may show that the Weyl projective curvature operator vanishes if m = 2 by
noting that the curvature operator of any affine surface (, ∇) with symmetric Ricci tensor satisfies

R(X, Y )Z = ρ(Y, Z)X − ρ(X, Z)Y.

One says that ∇ is projectively flat if around each point there is a projective change of ∇ to a
flat affine connection. One has (see Eisenhart [77, Section 32]):

Lemma 1.9 Let (M, ∇) be an affine manifold with symmetric Ricci tensor. Then ∇ is projectively flat
if and only if one of the following holds
1. dim(M) ≥ 3 and P = 0.

2. dim(M) = 2 and (∇X ρ)(Y, X) = (∇Y ρ)(X, Z) for all X, Y , Z vector fields on M.

We note that the Ricci tensor of any projectively flat affine manifold (M, ∇) satisfies the
identity in Assertion (2) above.

THE LEVI-CIVITA CONNECTION

Let M := (M, g) be a pseudo-Riemannian manifold. The Levi-Civita connection ∇ = g ∇ is the
unique affine connection which is Riemannian; it is characterized by the identities:

Xg(Y, Z) = g(∇X Y, Z) + g(Y, ∇X Z) and ∇X Y − ∇Y X = [X, Y ] ,

for all X, Y, Z ∈ C ∞ (T M). We refer to [7, 54, 140, 151, 196] for further details.
Let x = (x 1 , . . . , x m ) be a system of local coordinates on M, and let gij := g(∂xi , ∂xj ). We
extend g dually to a metric on the cotangent bundle; g ij := g(dx i , dx j ) is the inverse matrix. Let
ij k = g ij k and ij k = g ij k be the associated Christoffel symbols of g ∇. We have the Koszul
formula:
ij k = 21 {∂xi gj k + ∂xj gik − ∂xk gij } and ij k = g k ij  . (1.2.k)
Similarly, if {ei } is a local orthonormal frame for T M, let g([ei , ej ], ek ) = Cij k give the structure
constants for the Lie bracket. We have

ij k = 21 {Cij k − Cj ki + Ckij } .

The curvature tensor R of the Levi-Civita connection has the symmetries:

R(X, Y, Z, W ) + R(Y, X, Z, W ) = 0, (1.2.l)

R(X, Y, Z, W ) + R(Y, Z, X, W ) + R(Z, X, Y, W ) = 0, (1.2.m)
R(X, Y, Z, W ) + R(X, Y, W, Z) = 0 , (1.2.n)
R(X, Y, Z, W ) − R(Z, W, X, Y ) = 0 . (1.2.o)
 1.2. CONNECTIONS 15
The symmetries of Equation (1.2.l) and Equation (1.2.m) arise from the symmetries of Equa-
tion (1.2.e) and Equation (1.2.f) since ∇ is torsion-free. The symmetry of Equation (1.2.n) arises
from the fact that ∇ is Riemannian. Note that the symmetry of Equation (1.2.n) and the symmetry
of Equation (1.2.o) are equivalent in the presence of Equation (1.2.l) and Equation (1.2.m)—see,
for example, the discussion in Blažić et al. [20]. If V is a vector bundle over M, let S 2 = S 2 (V∗ )
(resp. 2 = 2 (V∗ )) be the bundle of symmetric (resp. alternating) bilinear forms on V. Let
S 2 M = S 2 (T ∗ M) and 2 M = 2 (T ∗ M). The symmetries of Equation (1.2.l), Equation (1.2.o),
and Equation (1.2.n) permit us to regard

R ∈ C ∞ (S 2 (2 M)) . (1.2.p)

Let {x, y} be a basis for a non-degenerate 2-plane π ⊂ TP M. The sectional curvature κ(π ) is
defined by setting
R(x, y, y, x)
κ(π) := .
g(x, x)g(y, y) − g(x, y)2
The sectional curvatures determine the complete curvature tensor. One has (see, for example,
Kobayashi and Nomizu [140] and O’Neill [169]):

Lemma 1.10 Suppose that R and R̃ satisfy the identities of Equations (1.2.l)–(1.2.o) above.
If κR (π ) = κR̃ (π ) for all non-degenerate 2-planes π , then R = R̃.

We shall discuss the geometry of pseudo-Riemannian manifolds with constant sectional cur-
vature subsequently in Section 1.10—see Wolf [212] for further details. A pseudo-Riemannian
manifold is said to be locally symmetric if ∇R = 0 (see, for example, Cartan [49, 50] and Helgason
[123]); we will discuss the geometry of such manifolds as well, subsequently in Section 1.10.

WEYL GEOMETRY
Weyl geometry (see Weyl [209]) is, in a sense, midway between Riemannian geometry and affine
geometry. The literature is a vast one and we can cite only a few references [23, 24, 46, 59, 132, 179,
180, 181, 199] in the geometric setting and [108, 109] in the algebraic setting.
Let (M, g) be a pseudo-Riemannian manifold. A tensor field K is said to be recurrent if there
exists a 1-form ω such that

∇K = ω ⊗ K, i.e.,∇X K = ω(X)K for all X ∈ C ∞ (T M) .

An affine connection ∇ on T M defines a Weyl structure and (M, g, ∇) is said to be a Weyl

manifold if the metric is recurrent, i.e., if there exists a smooth 1-form φ on M so that the structures
are related by the equation:
∇g = −2φ ⊗ g .
If g ∇ is the Levi-Civita connection, then we may express ∇ = φ ∇ in the form:
φ
∇X Y = g ∇X Y + φ(X)Y + φ(Y )X − g(X, Y )φ # , (1.2.q)
16 1. BASIC NOTIONS AND CONCEPTS
where φ # is the dual vector field. Conversely, if φ is given and if we use Equation (1.2.q) to define
∇, then ∇ is a Weyl connection with associated 1-form φ. We refer to Gilkey, Nikčević, and Simon
[109], to Ovando [174], and to Özdeğer [175] for further details concerning Weyl geometry.
There is an additional curvature symmetry which pertains in Weyl geometry (see, for example,
the discussion in Gilkey, Nikčević, and Simon [109]):

R(X, Y, Z, W ) + R(X, Y, W, Z) = − m4 ρa (X, Y )g(Z, W ) , (1.2.r)

where dim(M) = m. The defining 1-form φ is related to the curvature by the equation:

dφ = − m2 ρa . (1.2.s)

Weyl geometry is a conformal theory; if g1 = e2f g is conformally equivalent to g and if

(M, g, ∇) is a Weyl manifold, then (M, g1 , ∇) is again a Weyl manifold with associated 1-form φ1
given by φ1 = φ − df . One has the following well known result characterizing trivial Weyl structures
(see, for example, Gilkey, Nikčević, and Simon [109]):

Theorem 1.11 Let (M, g, ∇) be a Weyl manifold with HDeR 1 (M) = 0. The following assertions are

equivalent and if any is satisfied, then the Weyl structure is said to be trivial.

1. dφ = 0.

2. ∇ = g1 ∇ for some conformally equivalent pseudo-Riemannian metric g1 .

3. ∇ = g1 ∇ for some pseudo-Riemannian metric g1 .

4. ρ is a symmetric tensor.

Proof. Suppose that dφ = 0. Since the first de Rham cohomology group of M vanishes by assumption,
there exists f ∈ C ∞ (M) so that φ = df . We define the conformally equivalent metric g1 := e2f g. It
is then clear that ∇g1 = e2f (2df − 2φ) = 0. Thus, ∇ is torsion-free and Riemannian with respect
to g1 ; this implies ∇ = g1 ∇ is the Levi-Civita connection defined by g1 . Thus, Assertion (1) implies
Assertion (2); Assertion (2) trivially implies Assertion (3). Since the curvature tensor of the Levi-
Civita connection has a symmetric Ricci tensor, Assertion (3) implies Assertion (4). Finally, suppose
that Assertion (4) holds so that ρa = 0. By Equation (1.2.s), dφ = − m2 ρa = 0. 2

1.3 CURVATURE MODELS IN THE REAL SETTING

Let (V , ·, ·) be an inner product space. Here V is a real vector space of dimension m and ·, · is a
non-degenerate, symmetric, bilinear form on V of signature (p, q) where m = p + q. A 4-tensor
A ∈ ⊗4 V ∗ is said to be a Riemannian algebraic curvature tensor if A satisfies the symmetries given
in Equation (1.2.l), in Equation (1.2.m), and in Equation (1.2.n); note that Equation (1.2.o) is then
 1.4. KÄHLER GEOMETRY 17
satisfied automatically. Let R(V ) be the subspace of ⊗4 V ∗ , which consists of all tensors satisfying
these relations. We say that a triple (V , ·, ·, A) is a Riemannian curvature model if A ∈ R(V ). One
says that such a triple is geometrically realizable by a pseudo-Riemannian manifold if there is a point
P of some pseudo-Riemannian manifold (M, g) and if there is an isomorphism  : V → TP M so:

∗ gP = ·, · and ∗ gRP = A .

If A ∈ End{V } ⊗ V ∗ only satisfies the identities of Equation (1.2.e) and of Equation (1.2.f),
then A will be said to be an affine curvature tensor and the pair (V , A) will be said to be an affine
curvature model; see Blažić et al. [20] and Gilkey, Nikčević, and Simon [107] for further details.
Such a pair is said to be geometrically realizable by an affine manifold if there is a point P of some
affine manifold (M, ∇) and if there is an isomorphism  : V → TP M so that ∗ ∇ RP = A. In
the presence of a metric, we can lower indices and let A(V ) ⊂ ⊗4 V ∗ be the space of all tensors
satisfying Equation (1.2.l) and Equation (1.2.m).
Finally, let W(V ) ⊂ ⊗4 V ∗ be the space of 4-tensors satisfying Equation (1.2.l), Equa-
tion (1.2.m), and Equation (1.2.r); these are the Weyl algebraic curvature tensors—see, for example,
Nomizu [167]. If A ∈ R(V ), then ρa = 0 and R(X, Y, Z, W ) + R(X, Y, W, Z) = 0. Conse-
quently,
R(V ) ⊂ W(V ) ⊂ A(V ) .
A triple (V , ·, ·, A) is said to be a Weyl curvature model if A ∈ W(V ). The notion of geometric
realizability is defined analogously in this setting. We refer to Blažić et al. [20], to Gilkey [98], and
to Gilkey, Nikčević, and Simon [109] for the proof of the following result; the first two assertions
are, of course, well known:

Theorem 1.12

1. Every Riemannian curvature model can be realized geometrically by a pseudo-Riemannian mani-

fold.

2. Every affine curvature model can be realized geometrically by an affine manifold.

3. Every Weyl curvature model can be realized geometrically by a Weyl manifold.

1.4 KÄHLER GEOMETRY

We now pass from the real to the complex setting and to the para-complex setting; we refer to
Cruceanu, Fortuny, and Gadea [60] for details on para-complex geometry. Let V be a real vector space
of even dimension m = 2m̄. A complex structure on V is an endomorphism J− of V so J−2 = − Id.
Similarly, a para-complex structure on V is an endomorphism J+ of V so J+2 = Id and Tr{J+ } = 0; this
trace-free condition is automatic in the complex setting, but must be imposed in the para-complex
18 1. BASIC NOTIONS AND CONCEPTS
setting. It is convenient to introduce the notation J± in order to have a common formulation that
permits us to treat both contexts in a similar fashion, although, we shall never be considering both
structures simultaneously. In this setting, we say that (V , J± ) is a (para)-complex vector space. In
the geometric setting, (M, J± ) is said to be an almost (para)-complex manifold if J± is a smooth
endomorphism of the tangent bundle so that (TP M, J± ) is a (para)-complex structure for every
P ∈ M. Define the Nijenhuis tensor NJ± by setting:

NJ± (X, Y ) := [X, Y ] ∓ J± [J± X, Y ] ∓ J± [X, J± Y ] ± [J± X, J± Y ] .

Theorem 1.13 Let (M, J± ) be a 2m̄-dimensional almost (para)-complex manifold. The following
conditions are equivalent and if any is satisfied, then (M, J± ) is said to be a (para)-complex manifold and
the structure J± is said to be integrable:

1. There are coordinate charts (x 1 , y 1 , . . . , x m̄ , y m̄ ) covering M so that J± ∂xi = ∂yi and

J± ∂yi = ±∂xi for 1 ≤ i ≤ m̄.

2. The Nijenhuis tensor NJ± vanishes.

3. The eigenbundles of J± are closed under the Lie bracket.

The equivalence of these two definitions in the complex setting uses the Newlander–Nirenberg
theorem [130, 157, 158, 206], which is a deep result in the theory of partial differential equations
that can be regarded as a complex version of the Frobenius theorem. By contrast, the corresponding
result in the para-complex setting only relies on the Frobenius theorem [57, 63, 81, 140, 147]—see,
for example, the discussion in Brozos-Vázquez, Gilkey, and Nikčević [37] and in Cruceanu, Fortuny,
and Gadea [60]. We refer to Frölicher and Nijenhuis [82, 83] and to Kobayashi and Nomizu [140]
for further details concerning the Nijenhuis tensor.
Let (M, J± ) be an almost (para)-complex manifold and let ∇ be a torsion-free connection on
M. We use J± to give a (para)-complex structure to the tangent bundle. The triple (M, J± , ∇) is said
to be a (para)-Kähler affine manifold if ∇ is a (para)-complex connection; this means that ∇J± = J± ∇
or, equivalently, that the covariant derivative of J± vanishes. This condition then implies that J± is
integrable and that
R(X, Y )J± = J± R(X, Y ) . (1.4.a)
Working again in an algebraic setting, let ·, · be a non-degenerate symmetric bilinear form
of signature (p, q) on V . We say that the triple (V , ·, ·, J± ) is a (para)-Hermitian vector space if
J±∗ ·, · = ∓·, ·. Thus, we permit the signature to be indefinite even when speaking of a Hermitian
vector space. If J+ defines a para-complex structure, necessarily p = q so ·, · has neutral signature.
If J− defines a complex structure, then necessarily p and q are both even.
Similarly, in the geometric setting, a triple (M, g, J± ) is said to be an almost (para)-Hermitian
manifold if J± is an almost (para)-complex structure, if (M, g) is a pseudo-Riemannian manifold, and
 1.4. KÄHLER GEOMETRY 19
ifJ±∗ g= ∓g. The triple is said to be a (para)-Hermitian manifold if J± is integrable. If g ∇J
± = 0,
we can lower indices in Equation (1.4.a) to obtain the following symmetry, which is called the
(para)-Kähler identity:
R(X, Y, Z, W ) = ∓R(X, Y, J± Z, J± W ) . (1.4.b)

A (para)-Hermitian pseudo-Riemannian manifold (M, g, J± ) will be said to be a (para)-Kähler

manifold if g ∇J± = 0 or, equivalently, if d ± = 0 where ± is the (para)-Kähler form:

(X, Y ) := g(X, J± Y ) .

Thus, in particular, a (para)-Kähler manifold inherits a natural symplectic structure. A quadruple

(M, g, J± , ∇) is said to be a (para)-Hermitian Weyl manifold if (M, g, ∇) is a Weyl manifold and if
(M, g, J± ) is a (para)-Hermitian manifold.The structure is said to be a (para)-Kähler–Weyl manifold
if ∇J± = 0. We refer to [32, 102, 103, 105] for further details.
We now pass again to the algebraic context. Define the space of (para)-Kähler tensors K± ,
the space of (para)-Kähler affine algebraic curvature tensors K±,A , the space of (para)-Kähler Rie-
mannian algebraic curvature tensors K±,R , and the space of (para)-Kähler–Weyl algebraic curvature
tensors K±,W by setting, respectively:

K± := {A ∈ ⊗4 V ∗ : A(x, y, z, w) = ∓A(x, y, J± z, J± w)},

K±,A := K± ∩ A, K±,R := K± ∩ R, K±,W := K± ∩ W .

Let (V , J± ) be a (para)-complex vector space.The triple (V , J± , A) is said to be a (para)-Kähler affine

curvature model if A ∈ K±,A . Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. The quadruple
(V , ·, ·, J± , A) is said to be a (para)-Kähler Hermitian curvature model if J±∗ ·, · = ∓·, · and
if A ∈ K±,R . The quadruple (V , ·, ·, J± , A) is said to be a (para)-Kähler–Weyl curvature model if
A ∈ K±,W . We refer to Brozos-Vázquez, Gilkey, and Nikčević [37] for the proof of Assertion (1)
and to Brozos-Vázquez, Gilkey, and Merino [34] for the proof of Assertion (2) in the following
result:

Theorem 1.14

1. Every (para)-Kähler affine curvature model is geometrically realizable by a (para)-Kähler affine

manifold.

2. Every (para)-Kähler Hermitian curvature model is geometrically realizable by a (para)-Kähler

Hermitian manifold.

The corresponding questions of geometric realizability for (para)-Kähler–Weyl curvature

models will be examined in Chapter 4.
20 1. BASIC NOTIONS AND CONCEPTS
1.5 CURVATURE DECOMPOSITIONS
In this section, we will define the structure groups O (the orthogonal group), U± (the (para)-unitary
group), and U± (the -(para)-unitary group) and present the fundamental facts from representation
theory that we shall need. We will also discuss results of Singer and Thorpe [194] giving the
decomposition of the space of Riemannian algebraic curvature tensors R as an O module, results
of Higa [125, 126] giving the decomposition of the space of Weyl algebraic curvature tensors W as
an O module, and results of Tricerri and Vanhecke [200] giving the decomposition of R as a U±
module and the decomposition of the space of (para)-Kähler algebraic curvature tensors K±,R as
a U± module. This will rise to the decomposition of W as a U± module. As we shall not need the
decomposition of K±,A as a U± module, we shall omit this decomposition and instead refer to the
discussion in Brozos-Vázquez, Gilkey, and Nikčević [36] and also in Nikčević [159, 160]. We also
refer to [22, 35, 38, 69, 79, 153, 198] for related work.

THE STRUCTURE GROUPS

If V is a vector space, let GL = GL(V ) denote the general linear group of linear transformations of
V . If (V , ·, ·) is an inner product space, then the associated orthogonal group O = O(V , ·, ·) is
given by:
O := {T ∈ GL : T ∗ ·, · = ·, ·} .
Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. The associated structure groups are:

U± := {T ∈ O : T J± = J± T },
U± := {T ∈ O : T J± = J± T or T J± = −J± T } .

The group U± is often called the (para)-unitary group. It is convenient to work with the Z2 extensions
U± as we may then interchange the roles of J± and −J± . Let χ (T ) = ±1 satisfy

J± T = χ (T )T J± for T ∈ U± . (1.5.a)

Since χ(ST ) = χ (S)χ(T ), then we may regard χ as defining a Z2 valued character of U± or,
alternatively, as being a 1-dimensional real representation of U± . We can extend ·, · to a natural
non-degenerate inner product on ⊗k V and ⊗k V ∗ . The following observation is fundamental in the
subject:

Lemma 1.15 Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. Let G ∈ {O, U− , U− , U+ } and
let ξ be a G submodule of ⊗k V ∗ . Then the restriction of the inner product on ⊗k V ∗ to ξ is non-degenerate.

Proof. As this is a crucial observation, we shall give the proof. Let {ei } be an orthonormal basis
for V and let {ei } be the associated dual basis for V ∗ . If I = (i1 , . . . , ik ) is a multi-index, set
 1.5. CURVATURE DECOMPOSITIONS 21
eI = e i1 ⊗ · · · ⊗ eik . Then:

0 if I  = J
(eI , eJ ) := ei1 , ej1  · · · eik , ejk  = . (1.5.b)
±1 if I = J

Let T ei = ei , ei  · ei define an element T ∈ O. Suppose ξ is an O invariant subspace of

⊗k V ∗ . Decompose ξ = ξ+ ⊕ ξ− and decompose ⊗k V ∗ = Y+ ⊕ Y− into the ±1-eigenspaces of T .
Since T ∈ O, these decompositions are orthogonal direct sums. By Equation (1.5.b), Y+ is spacelike
and Y− is timelike. Since ξ± ⊂ Y± , ξ+ is spacelike and ξ− is timelike; the Lemma now follows
in this special case. If G = U− or if G = U− , then we can choose the orthonormal basis so that
J− e2ν−1 = e2ν and J− e2ν = −e2ν−1 ; this additional normalization is crucial. Since J−∗ ·, · = ·, ·,
J− T = T J− . We then have T ∈ G and the same argument pertains. Finally, suppose G = U+ . We
can choose the basis so J+ e2ν−1 = e2ν , J+ e2ν = e2ν−1 , e2ν−1 is spacelike and e2ν is timelike. We
now have T ∈ U+ − U+ . 2
We note that Lemma 1.15 fails for the group G = U+ . For example, the ±1-eigenspaces
of J+ are invariant under J+ and are totally isotropic. We can combine Lemma 1.15 with same
arguments as used in the positive definite setting to establish the following result; we omit details in
the interests of brevity and instead refer to the discussion in Brozos-Vázquez, Gilkey, and Nikčević
[37]:

Lemma 1.16 Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. Let G ∈ {O, U− , U− , U+ } and
let ξ be a G submodule of ⊗k V ∗ .

1. There exist G submodules ηi of ξ so that we may decompose ξ = η1 ⊕ · · · ⊕ ηk as the orthogonal

direct sum of irreducible G modules.The multiplicity with which an irreducible representation appears
in ξ is independent of this decomposition.

2. If ξ1 and ξ2 are inequivalent irreducible submodules of ξ , then ξ1 ⊥ ξ2 . If ξ1 appears with multiplicity

1 in ξ and if η is any G submodule of ξ , then either ξ1 ⊂ η or else ξ1 ⊥ η.

We can illustrate Lemma 1.16 as follows. Let (V , ·, ·, J± ) be a (para)-Hermitian vector
space. We may decompose
⊗2 V ∗ = 2 ⊕ S 2
as the direct sum of the alternating and the symmetric bilinear forms. Let χ(T ) defined by Equa-
tion (1.5.a) and let
± (x, y) := x, J± y (1.5.c)
be the (para)-Kähler form. Let 1 be the trivial module. If T ∈ U± , then:

T∗ ± = χ (T ) ±.
22 1. BASIC NOTIONS AND CONCEPTS
Let
2± = 2± (V ∗ ) := {ω ∈ 2 : J±∗ ω = ±ω}, χ := ± · R,
2∓,0 = 2±,0 (V ∗ ) := {ω ∈ 2 : J±∗ ω = ∓ω, ω ⊥ ± },
2 = S 2 (V ∗ ) := {θ ∈ S 2 : J ∗ θ = ±θ }, (1.5.d)
S± ± ± 1 := ·, · · R,
2 = S 2 (V ∗ ) := {θ ∈ S 2 : J ∗ θ = ∓θ, θ ⊥ ·, ·}.
S∓,0 ± ±

Lemma 1.17 Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. We have the following decompo-
sition of 2 , S 2 , and ⊗2 V ∗ into inequivalent and irreducible U± modules:

2 = 2± ⊕ χ ⊕ 2∓,0 , S 2 = S±
2
⊕ 1 ⊕ S∓,0
2
,
2 ∗
⊗ V = ± ⊕ χ ⊕ ∓,0 ⊕ S± ⊕ 1 ⊕ S∓,0 .
2 2 2 2

2 are isomorphic U modules, that 2 is isomorphic to S 2 ⊗ χ

We note that 2∓,0 and S∓,0 ± ∓,0 ∓,0
as a U±
module, that χ is isomorphic to 1 as a U± module, and that 2+ is not an irreducible U+
module.

ALMOST (PARA)-HERMITIAN STRUCTURES

We now present results of Brozos-Vázquez et al. [30] giving the classification of (para)-Hermitian
structures. Let (M, g, J± ) be an almost (para)-Hermitian manifold. Let ∇ be the Levi-Civita
connection of g. The associated (para)-Kähler form ± and the covariant derivative ∇ ± are given
by:

:= g(X, J± Y ),
± (X, Y )
∇ ± (X, Y ; Z) = Zg(X, J± Y ) − g(∇Z X, J± Y ) − g(X, J± ∇Z Y ) .

One has the symmetries:

∇ ± (X, Y ; Z) = −∇ ± (Y, X; Z) and ∇ ± (X, Y ; Z) = ±∇ ± (J± X, J± Y ; Z) . (1.5.e)

It is convenient to work in an algebraic setting similar to that which we will use subsequently in
discussing curvature models. Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. Motivated by
Equation (1.5.e), we define:

H± := {H± ∈ ⊗3 V ∗ : H± (x, y; z) = −H± (y, x; z) and

H± (J± x, J± y; z) = ±H± (x, y; z), ∀ x, y, z},
U3,± := {H± ∈ H± : H± (x, y; z) = ∓H± (x, J± y; J± z), ∀ x, y, z} .

Let εij := ei , ej  give the components of the inner product relative to some basis {ei } for V . We
define:
 1.5. CURVATURE DECOMPOSITIONS 23
(τ1 H )(x) := εij H (x, ei ; ej ) for H ∈ ⊗3 V ∗ ,
ν± (φ)(x, y; z) := φ(J± x)y, z − φ(J± y)x, z
+φ(x)J± y, z − φ(y)J± x, z for φ ∈ V ∗ ,
W1,± := {H± ∈ H± : H± (x, y; z) + H± (x, z; y) = 0, ∀x, y, z},
W2,± := {H± ∈ H± : H± (x, y; z) + H± (y, z; x) + H± (z, x; y) = 0, ∀x, y, z},
W3,± := U3,± ∩ ker{τ1 },
W4,± := Range{ν± }.
In the geometric setting, if (M, g, J± ) is a (para)-Hermitian manifold (i.e., J± is integrable), then
∇ ± ∈ U3,± . The following result in Brozos-Vázquez et al. [30] is based on a decomposition of
H± , which extends the Gray–Hervella classification [118] of almost Hermitian structures in the
positive definite context to the indefinite setting and to the para-complex setting:

Theorem 1.18 Let m ≥ 6. We have a direct sum orthogonal decomposition of H± and of U3,± into
irreducible inequivalent U± modules in the form:

H± = W1,± ⊕ W2,± ⊕ W3,± ⊕ W4,± and U3,± = W3,± ⊕ W4,± .

One obtains the corresponding decompositions if m = 4, by setting W1,± = 0 and W3,± = 0.

The fact that there are only two components in dimension four has important implications in
(para)-Kähler–Weyl geometry and is one of the reasons why the 4-dimensional setting is exceptional
in that geometry.
It was shown in Brozos-Vázquez et al. [30] that every element of H± and of U3,± is geomet-
rically realizable in an appropriate context. One can, however, focus instead on the precise nature
of the classes involved. Restricting to the complex setting, we say that (M, g, J− ) is a ξ -manifold
if ∇ − belongs to ξ at every point of the manifold, where ξ is a U− submodule of H− , which is
minimal with this property.This gives rise to the celebrated 16 classes of almost Hermitian manifolds
in the positive definite setting. Many of these classes have geometrical meanings which have been
extensively investigated. For example:
1. ξ = {0} defines the class of Kähler manifolds.

2. ξ = W1,− defines the class of nearly Kähler manifolds.

3. ξ = W2,− defines the class of almost Kähler manifolds.

4. ξ = W3,− defines the class of Hermitian semi-Kähler manifolds.

5. ξ = W1,− ⊕ W2,− defines the class of quasi-Kähler manifolds.

6. ξ = W3,− ⊕ W4,− = U3,− defines the class of Hermitian manifolds.

24 1. BASIC NOTIONS AND CONCEPTS
7. ξ = W1,− ⊕ W2,− ⊕ W3,− defines the class of semi-Kähler manifolds.

8. ξ = H− defines the class of almost Hermitian manifolds.

SCALAR INVARIANTS
Let (V , ·, ·, J± ) be a (para)-Hermitian vector space and let ξ be a G submodule of ⊗k V ∗ where
G ∈ {O, U− , U− , U+ }.We say that a linear map  from ξ to R is a scalar invariant if (g · v) = (v)
for every v ∈ ξ and for every g ∈ G; let I G (ξ ) be the vector space of all such invariants. The
inner product ·, · extends naturally to an inner product ·, · on ξ , which is non-degenerate by
Lemma 1.15. Let HomG {ξ } be the set of intertwining operators, i.e., linear maps T of ξ to ξ so that
g −1 T g = T for all g ∈ G. Let θ be a bilinear form on ξ , i.e., a linear map from ξ ⊗ ξ to R. Define
a linear map Tθ from ξ to ξ , which is characterized by the property θ(v1 , v2 ) := v1 , Tθ v2 .

Lemma 1.19 The map θ → Tθ identifies I G (ξ ⊗ ξ ) with HomG {ξ }. Consequently, if

dim(I G (ξ ⊗ ξ )) = 1, then any intertwining operator T ∈ HomG {ξ } is a scalar multiple of the identity.

Proof. A linear invariant on ξ ⊗ ξ can be regarded as a bilinear form on ξ or, equivalently, as an

element of ⊗2 ξ ∗ . We have

θ(gv1 , gv2 ) = gv1 , Tθ gv2  = v1 , g ∗ Tθ gv2  = v1 , g −1 Tθ gv2  .

Thus, θ ∈ I (ξ ⊗ ξ ) is a linear invariant if and only if g −1 Tθ g = Tθ or, equivalently, if Tθ belongs to

HomG {ξ }. 2
Weyl [210, pages 53 and 66] gives a spanning set if G = O is the orthogonal group; the
corresponding result for the unitary group U− in the positive definite Hermitian setting follows
from work of Fukami [84] and of Iwahori [136]; the extension to the groups U± in general is
straightforward—see, for example, Brozos-Vázquez, Gilkey, and Nikčević [37].
We discuss this spanning set. All invariants arise by using either the inner product or the
(para)-Kähler form to contract indices in pairs; invariants of U± arise when the (para)-Kähler form
appears an even number of times. It is worth being a bit more formal about this. Let (V , ·, ·, J± )
be a (para)-Hermitian vector space. Let h = ·, · and let ±,ij be the components of the (para)-
Kähler form. If {ei } is any basis for V , let hij := ei , ej  give the components of h relative to this
basis. The inverse matrix hij = ei , ej  gives the components of the dual inner product on V ∗ . Let
 ∈ ⊗2k V ∗ . We may expand
 = i1 ...i2k ei1 ⊗ · · · ⊗ ei2k .
Let π be a permutation of the set {1, . . . , 2k}. Let κ0 := h, let κ1 := ± , and let a be a sequence of
0’s and 1’s. Define:
i iπ(2) i iπ(2k)
ψπ,a () := κaπ(1) 1 . . . κaπ(2k−1)
k i1 ...i2k .
 1.5. CURVATURE DECOMPOSITIONS 25
a ) be the number of times ai = 1. One then has:
Let n(

Lemma 1.20 If (V , ·, ·, J± ) is a (para)-Hermitian vector space and if ξ is a U± submodule of ⊗2k V ∗ ,
I U± (ξ )

then = Spann(a ) even {ψπ,a }.

Applying this formalism in the geometric setting to the curvature tensor and to the covariant
derivatives of the curvature tensor yields scalar invariants of the metric. Let In be the space of
invariants which are homogeneous of order n in the jets of the metric. Let “;" denotes covariant
differentiation. One has, see for example Gilkey [94], that:

Lemma 1.21 Let (M, g) be a Riemannian manifold. Let {ei } be a local frame orthonormal field for the
manifold. Then
I0 = Span {1}, 
I2 = Span Rijj i , 
I4 = Span Rijj i;kk , Rijj i Rkk , Rijj k Rik , Rij k Rij k ,
I6 = Span Rijj i;kk , Rijj i;k Rnn;k , Raij a;k Rbij b;k , Raj ka;n Rbj nb;k ,
Rij k;n Rij k;n , Rijj i Rkk;nn , Raj ka Rbj kb;nn , Raj ka Rbj nb;kn ,
Rij k Rij k;nn , Rijj i Rkk Rabba , Rijj i Raj ka Rbj kb , Rijj i Rabcd Rabcd ,
Raj ka Rbj nb Rcknc , Raij a Rbkb Rikj  , Raj ka Rj ni Rkni , Rij kn Rij p Rknp ,
Rij kn Rikp Rj np

These invariants are linearly independent if m ≥ 6, but in lower dimensions there are relations
amongst these invariants called universal curvature identities that arise from Weyl’s second theorem of
invariants; we refer to the discussion in Gilkey, Park, and Sekigawa [111, 112] for further details.
There is a similar formalism in the pseudo-Riemannian setting where one has to be a bit more careful
since g(ei , ei ) = ±1.
A Riemannian manifold is locally homogeneous if and only if all the scalar invariants up to
order 21 m(m − 1) are constant and the local geometry is determined by these invariants (see Prüfer,
Tricerri, and Vanhecke [183]). This fails in the pseudo-Riemannian setting as there are manifolds
all of whose scalar invariants vanish—see, for example, the discussion in Alcolado et al. [4], in Coley
et al. [58], and in Pravda et al. [182]. Such manifolds are called vanishing scalar invariants manifolds
or simply VSI manifolds. We complete our discussion of elementary representation theory with the
following result (see, for example, the discussion in Brozos-Vázquez, Gilkey, and Nikčević [37]):

Lemma 1.22 Let (V , ·, ·, J± ) be a (para)-Hermitian vector space.

1. dim(HomU− {2− }) = 1 and dim(HomU+ {2+ }) = 1.

2. Let ξ(a, b) := {(aθ, bθ )}θ∈2 ⊂ 2± ⊕ 2± for (a, b) = (0, 0). If ξ is a non-trivial proper U±
±
submodule of 2± ⊕ 2± , then there exists 0 = (a, b) ∈ R2 so ξ = ξ(a, b).
26 1. BASIC NOTIONS AND CONCEPTS
Proof. We prove Assertion (1) as follows. We have, by definition, that:

2± ⊗ 2± = {θ ∈ ⊗4 V ∗ : θ(x, y, z, w) = −θ(y, x, z, w) = −θ(x, y, w, z)

and θ (x, y, z, w) = ±θ (J± x, J± y, z, w) = ±θ(x, y, J± z, J± w) ∀x, y, z, w} .

Let {ei } be a basis for V and let εij = ei , ej . By Lemma 1.20, any U± linear invariant of 2± ⊗ 2±
is a linear combination of the invariants:

1 (θ ) = εij ε k θij k , 2 (θ ) = εik εj  θij k , 3 (θ) = εi εj k θij k ,

ij k ik j  i j k
4 (θ ) = ± ± θij k , 5 (θ ) = ± ± θij k , 6 (θ) = ± ± θij k .

Since θ is skew-symmetric in the first two indices and since ε is symmetric,

1 (θ) = 0, 2 (θ ) =− 3 (θ ), 5 (θ ) =− 6 (θ ) .

We examine cases. We can decouple the bases. We consider 4 different bases {ai }, {bi }, {ci }, and {di }
and the corresponding dual bases {a i }, {bi }, {ci }, and {d i }. We may then express

4 (θ ) = a i , J± bj ck , J± d  θ (ai , bj , ck , d ) .
i,j,k,

We set ai = J± ei , bi = ei , ci = J± ei , and di = ei . This is equivalent to interchanging the order of

summation and occasionally changing the sign. We show 4 (θ ) = 0 by computing:
4 (θ ) = J± ei , J± ej J± ek , J± e θ (J± ei , ej , J± ek , e )
= ei , ej ek , e θ (J± ei , ej , J± ek , e )
= ±ei , ej ek , e θ (J± J± ei , J± ej , J± ek , e )
= ei , ej ek , e θ (ei , J± ej , J± ek , e )
= −ei , ej ek , e θ (J± ej , ei , J± ek , e )
=− 4 (θ).

Similarly, setting ai = J± ei , bi = J± ei , and ci = di = ei , we express:

5 (θ ) = J± ei , J± ek J± ej , J± e θ (J± ei , J± ej , ek , e )
= ±ei , ek ej , e θ (ei , ej , ek , e ) = ± 2 (θ ).
U±

This shows dim(I (2± ⊗ 2± )) = 1; Assertion (1) now follows from Lemma 1.19.
Let ξ be a proper U± submodule of 2± ⊕ 2± . Let π1 (resp. π2 ) be the projection on the
first (resp. on the second) factor. Since ξ is non-trivial, we may assume, without loss of generality,
that π1 (ξ ) = {0}; since ξ is a proper submodule, ξ is necessarily irreducible, and hence π1 is an
isomorphism. If π2 (ξ ) = 0, then ξ = ξ(1, 0). Thus, we may assume that π2 = 0 and consequently
T := π2−1 π1 is a non-trivial U± equivariant map of 2± . By Assertion (1), any intertwining operator
is a scalar multiple of the identity. Consequently, T = b Id and ξ = ξ(1, b). 2
 1.5. CURVATURE DECOMPOSITIONS 27
CONFORMAL EQUIVALENCE
Two pseudo-Riemannian manifolds (M1 , g1 ) and (M2 , g2 ) are said to be (locally) conformally equiv-
alent if there is a (local) diffeomorphism ϕ : M1 → M2 such that ϕ ∗ g2 = e2f g1 for some smooth
conformal factor f , which is defined (locally) on M. The Weyl conformal curvature tensor W plays
a central role in these geometries. It forms one of the three components of the curvature tensor
decomposition of Singer and Thorpe [194] that we shall discuss presently in Theorem 1.24. It is
defined by setting:

W (X, Y, Z, T ) := R(X, Y, Z, T ) + (m−1)(m−2) τ {g(Y, Z)g(X, T ) − g(X, Z)g(Y, T )}

1

− m−2 {g(Y, Z)ρ(X, T ) − g(X, Z)ρ(Y, T )}

1
(1.5.f )
− m−2 {g(X, T )ρ(Y, Z) − g(Y, T )ρ(X, Z)} ,
1

where ρ is the Ricci tensor and τ = Tr{ρ} is the scalar curvature. Note that the Weyl conformal
curvature tensor W is not defined for m = 2 and that W vanishes identically if m = 3.
The Weyl conformal curvature tensor defined above is invariant by conformal transformations,
i.e., W = e2f g2 W (or, equivalently, the associated Weyl curvature operators satisfy g1 W = g2 W )
g 1

for any two conformally equivalent metrics. A converse implication holds true under certain con-
ditions that depend on the signature of the metric, the dimension, and the non-degeneracy of the
Weyl curvature operator—we refer to Hall [120] for further details. A pseudo-Riemannian manifold
(M, g) is said to be locally conformally flat if it is locally conformally equivalent to a flat manifold. If
(M, g) is locally conformally flat, then

W = 0 and (∇X ρ)(Y, Z) = (∇Y ρ)(X, Z) for all X, Y, Z ∈ C ∞ (T M) .

The following converse is due to Weyl [207, 208] and Schouten [191]—see also Hertrich-
Jeromin [124].

Lemma 1.23 Let (M, g) be a pseudo-Riemannian manifold. Then (M, g) is locally conformally flat if
and only if one of the following conditions holds:
1. m ≥ 4 and g W = 0.
2. m = 3 and (∇X ρ)(Y, Z) = (∇Y ρ)(X, Z) for all vector fields X, Y , Z on M.

Finally, observe that the Ricci tensor of any locally conformally flat pseudo-Riemannian
manifold (M, g) satisfies the identity in Assertion (2) above.

THE SINGER–THORPE AND HIGA DECOMPOSITIONS

We now examine the O module structure of R and W. Let (V , ·, ·) be an inner product space. Let
S 2 = S 2 (V ∗ ), and let

S02 := {θ ∈ S 2 : θ ⊥ ·, ·} and C := ker{ρ} ∩ R

28 1. BASIC NOTIONS AND CONCEPTS
be the O modules of trace-free symmetric 2-cotensors and Weyl conformal curvature tensors, respec-
tively. We refer to Singer and Thorpe [194] for the proof of Assertion (1) and to Higa [125, 126]
for the proof of Assertion (2) in the following result:

Theorem 1.24 Let m ≥ 4.

1. We may decompose R = 1 ⊕ S02 ⊕ C as the orthogonal direct sum of irreducible and inequivalent
O modules.

2. We may decompose W = 1 ⊕ S02 ⊕ C ⊕ P as the orthogonal direct sum of irreducible and inequiv-
alent O modules. Here, ρa provides an O module isomorphism from P to 2 with the inverse
≈
embedding  : 2 −→P ⊂ W being given by:

(ψ)(x, y, z, w) := 2ψ(x, y)z, w + ψ(x, z)y, w − ψ(y, z)x, w

− ψ(x, w)y, z + ψ(y, w)x, z .

The space C consists of the Weyl conformal curvature tensors defined in Equation (1.5.f). The
trivial module 1 is detected by the scalar curvature, and the module S02 is detected by the trace-free
Ricci tensor.

THE TRICERRI–VANHECKE DECOMPOSITIONS

The following decompositions of R and K±,R as U− modules were given in Mori [156], in Sitara-
mayya [195], and in Tricerri and Vanhecke [200] in the positive definite setting; they extend easily
to the more general context (see Brozos-Vázquez, Gilkey, and Merino [34] and Brozos-Vázquez,
Gilkey, and Nikčević [37]).The decomposition of W as a U± module then follows from Lemma 1.17
and Theorem 1.24.

Theorem 1.25 Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. We have the following decompo-
sitions of R(V ), K±,R (V ), and W(V ) as U± modules:
R(V ) = W±,1 ⊕ · · · ⊕ W±,10 , K±,R (V ) = W±,1 ⊕ W±,2 ⊕ W±,3 ,
W(V ) = W±,1 ⊕ · · · ⊕ W±,13 .
If m = 4, we omit the modules {W±,5 , W±,6 , W±,10 }. If m = 6, we omit the module W±,6 . The de-
compositions given above are then into irreducible U± modules. We have U± module isomorphisms:
W±,1 ≈ W±,4 ≈ 1 , W±,2 ≈ W±,5 ≈ S∓,0 2 , W
±,9 ≈ W±,13 ≈ ± . W±,8 ≈ S± , W±,11 ≈ χ, and
2 2

W±,12 ≈ ∓,0 . With the exception of these isomorphisms, the modules W±,i are inequivalent U± modules.
2

The isomorphism from 2± to W±,9 is given by setting:

 1.6. WALKER STRUCTURES 29
(ψ)(x, y, z, w) := 2x, J± yψ(z, J± w) + 2z, J± wψ(x, J± y)
+x, J± zψ(y, J± w) + y, J± wψ(x, J± z)
−x, J± wψ(y, J± z) − y, J± zψ(x, J± w).

Remark 1.26 The Bochner curvature tensor of an almost (para)-Hermitian manifold (M, g, J± )
may be defined projecting R(V ) to the subspace W±,3 ⊕ W±,6 ⊕ W±,7 ⊕ W±,10 (see Bryant [39]
or Tricerri and Vanhecke [200]). This tensor is analogous to the Weyl conformal curvature tensor
defined in Equation (1.5.f).

1.6 WALKER STRUCTURES

In this section, we will review the main results concerning Walker structures that we shall need
subsequently. Holonomy, parallel transport, and isometric decompositions are treated, as are parallel
plane fields and Walker coordinates. Subsequently, in Section 1.7, we treat a special class of Walker
metrics called Riemannian extensions. We also discuss deformed Riemannian extensions and mod-
ified Riemannian extensions. The field is a vast one and we can only give a few references on the
subject [3, 43, 45, 52, 68, 75, 89, 152, 170, 177, 203, 211]; see also Law and Matsushita [148] for a
spinor approach.

THE HOLONOMY GROUP AND ISOMETRIC DECOMPOSITIONS

Let α : [a, b] → M be a smooth curve in a connected affine manifold (M, ∇). For each tangent
vector v in the tangent space to M at the initial point of the curve α(a) = P , let v(t) be defined by
parallel transport along α; v(t) is the vector field along α given by solving the equation
∇α̇(t) v(t) = 0 with initial condition v(a) = v.
If α is a closed curve, then the map v(a) → v(b) given by the parallel transport around α defines a
linear isomorphism
Lα : TP M → TP M ,
which is called holonomy. The set of all such linear maps forms a group called the holonomy group
of the connection. Since M is assumed to be connected, the holonomy groups corresponding to
different points of M are all isomorphic; if β is any curve from P to Q, then Lβ provides an
isomorphism between the holonomy group at P and the holonomy group at Q. Consequently, the
role of the basepoint P is usually suppressed. There is a vast literature on the subject—we cite a few
representative examples [5, 11, 76, 80, 121, 150].
The holonomy group is a closed subgroup of the general linear group GL(TP M) of the tangent
bundle at P . Consequently, the holonomy group is a Lie group. When one uses the Levi-Civita
connection of a pseudo-Riemannian metric, the holonomy group is a subgroup of the orthogonal
30 1. BASIC NOTIONS AND CONCEPTS
group O(TP M, gP ), since parallel transport is realized by isometries, and we refer to it as the holonomy
group of the pseudo-Riemannian manifold.
In the Riemannian setting the holonomy group acts completely reducibly. This means that
there is an orthogonal direct sum decomposition

TP M = V1 ⊕ · · · ⊕ V ,

where each Vi is invariant under the holonomy group and where Vi contains no non-trivial invari-
ant subspaces. In the higher signature setting, however, the situation is more complicated. If Vi is
an invariant subspace, there need not exist a complementary invariant subspace. One says that the
holonomy group acts indecomposably and that the manifold is indecomposable if the metric is degen-
erate on any invariant proper subspace. In the Riemannian setting, indecomposability is equivalent
to irreducibility.
The holonomy group of the product of pseudo-Riemannian manifolds is the product of the
holonomy groups of these manifolds. Furthermore, a converse of this statement is true in the follow-
ing sense. Suppose that (M, g) is a connected pseudo-Riemannian manifold such that the tangent
space at a single point (and hence at every point) admits an orthogonal direct sum decomposition
into non-degenerate subspaces which are invariant under the holonomy representation. In this set-
ting, (M, g) is locally isometric to a product of pseudo-Riemannian manifolds corresponding to
the invariant subspaces. Moreover, the holonomy group is the product of the groups acting on the
corresponding invariant subspaces. A global version of this statement, under the assumption that
the manifold is simply connected and complete, was established by de Rham [184] for Riemannian
manifolds and by Wu [214] in the arbitrary signature case:

Theorem 1.27 Any simply connected complete pseudo-Riemannian manifold (M, g) is isometric to a
product of simply connected complete pseudo-Riemannian manifolds, one of which can be flat and the others
have an indecomposably acting holonomy group. Moreover, the holonomy group of (M, g) is the product of
these indecomposably acting holonomy groups.

PARALLEL PLANE FIELDS

Let (M, g) be a pseudo-Riemannian manifold and let D be a distribution, i.e., a smooth subbundle
of the tangent space. We say that D is a parallel plane field if

∇X Y ∈ D for all X ∈ C ∞ (T M) and for all Y ∈ C ∞ (D) .

The distribution D is said to be a null plane field if D is totally isotropic, i.e., if

g(X, Y ) = 0 for all X ∈ C ∞ (D) and Y ∈ C ∞ (D) .

 1.6. WALKER STRUCTURES 31
The existence of a null parallel plane field D in a pseudo-Riemannian manifold influences the
curvature. We refer to the discussion in Derdzinski and Roter [68] for the proof of the following
result:

Lemma 1.28 Let M := (M, g) be a pseudo-Riemannian manifold admitting a null parallel plane
field D. Then R(D, D⊥ , ·, ·) = 0, R(D, D, ·, ·) = 0, and R(D⊥ , D⊥ , D, ·) = 0.

WALKER METRICS AND WALKER COORDINATES

For indefinite metrics, there exists the possibility that some of the factors in Theorem 1.27 are
indecomposable, but not irreducible. This means that one of the factors could contain a proper null
invariant subspace. Indecomposable Lorentzian metrics were initially investigated by Brinkmann
[28], while the general pseudo-Riemannian case was first considered by Walker [204]. In all cases,
indecomposability is equivalently described by the existence of a null parallel plane field. Following
the notation in Brozos-Vázquez et al. [31], and motivated by this seminal work of Walker, a pseudo-
Riemannian manifold (M, g) that admits a non-trivial null parallel plane field D is said to be a Walker
manifold . Such metrics only arise in the pseudo-Riemannian setting. They form the underlying
structure of many pseudo-Riemannian situations with no Riemannian counterpart. We refer to
Brozos-Vázquez et al. [31] for more information on this family of metrics.
By imposing some additional curvature identities of the form given in Lemma 1.28, one can
specialize the Walker structure under consideration. This is the case, for example, for the pp-waves
which are locally characterized by the identity R(D⊥ , D⊥ , ·, ·) = 0 and by the fact that the range
of the Ricci operator is totally isotropic (see Leistner [150]). Note that TP M  = D ⊕ D⊥ since D is
a null distribution. There is a vast literature concerning pp-waves and we cite a few representative
examples here [1, 139, 149, 155].
The existence of adapted coordinates (u, v, x 1 , . . . , x m−2 ) on a Lorentzian manifold (M, g),
admitting a null parallel line field, was established by Brinkmann [28]. In those coordinates the
metric g has the form:

g = 2 du ◦ dv + f du ◦ du + ai du ◦ dx i + gij dx i ◦ dx j where ∂v gij = ∂v ai = 0 .

Moreover ∂v f = 0 if and only if the null parallel line field D = Span{∂v } is spanned by a parallel
vector field, in which case the coordinates can be chosen so that ai = 0 and so that f = 0. Walker
[204] generalized this result to arbitrary pseudo-Riemannian manifolds (M, g) that admit a null
parallel plane field D of dimension r as follows:

Theorem 1.29 If (M, g) is an m-dimensional pseudo-Riemannian manifold which admits a null

parallel plane field D of dimension r, then there exist local coordinates (x 1 , . . . , x m−r , x m−r+1 , . . . , x m )
on M such that the metric tensor g takes the form:
32 1. BASIC NOTIONS AND CONCEPTS
⎛ ⎞
B H Idr
(gij ) = ⎝ t H A 0 ⎠ ,
Idr 0 0
where Idr is the identity matrix and A, B, and H are matrices whose entries are functions of
(x 1 , . . . , x m−r , x m−r+1 , . . . , x m ) such that A and B are symmetric matrices of order m − 2r and r
respectively, H is a r × (m − 2r) matrix and t H its transpose, and such that A and H are independent
of the (x m−r+1 , . . . , x m ) coordinates. Finally, the null parallel plane field D is locally spanned by the
coordinate vector fields {∂xm−r+1 , . . . , ∂xm }.

Remark 1.30 In the special case that dim(M) = 2m̄ and dim(D) = 21 dim(M) = m̄ is maximal,
there exist Walker coordinates (x 1 , . . . , x m̄ , x1 , . . . , xm̄ ) and there exists a symmetric m̄ × m̄ matrix
B = B( x , x ) so that:
 
B Idm̄
(gij ) = .
Idm̄ 0

The metrics given in Remark 1.30 are of special interest for our purposes. Their Christoffel
symbols are given as follows (see Calviño-Louzao et al. [43]):

Lemma 1.31 Let (M, g) be a 2m̄-dimensional Walker manifold with a null parallel plane field D of
dimension m̄ and let (x 1 , . . . , x m̄ , x1 , . . . , xm̄ ) be adapted coordinates where the metric takes the form
given in Remark 1.30. The (possibly) non-zero components of the Christoffel symbols of the Levi-Civita
connection g ∇ are determined by:

ij k = − 21 ∂x k gij , i  j k = 21 ∂x i  gj k ,
   
ij k = 21 ∂xj gik − ∂xk gij + ∂xi gj k + 1≤s≤m̄ gks ∂x s  gij .
Similarly, the (possibly) non-zero components of the curvature operator are determined by:
 
Rj ik h = 21 ∂xi ∂x h gj k − ∂xj ∂x h gik
  
+ 41 1≤s≤m̄ ∂x s  gik ∂x h gj s − ∂x s  gj k ∂x h gis ,
  
Rj ik h = 21 ∂xj ∂xk gih − ∂xj ∂xh gik + ∂xi ∂xh gj k − ∂xi ∂xk gj h
   
+ 41 1≤s,t≤m̄ ∂x s  gik ∂xh gj s − ∂xs gj h − ∂xj gsh − ght ∂x t  gj s
 
− ∂x s  gj k ∂xh gis − ∂xs gih − ∂xi gsh − ght ∂x t  gis
 
− ∂x s  gj h ∂xs gik − ∂xk gis − ∂xi gks − gst ∂x t  gik
 
+ ∂x s  gih ∂xs gj k − ∂xk gj s − ∂xj gks − gst ∂x t  gj k
   
+2∂xj ghs ∂x s  gik − 2∂xi ghs ∂x s  gj k ,
Rj i  k h = 21 ∂x i  ∂x h gj k ,
 1.7. METRICS ON THE COTANGENT BUNDLE 33
h
 
Rj i  k = 1
2 ∂xh ∂x i  gj k − ∂xk ∂x i  gj h
  
+ 41 s ∂x s  gj k ∂x i  gsh + ∂x s  gj h ∂x i  gsk − 2∂x i  (ghs ∂x s  gj k ) ,
     
Rj ik  h = 21 ∂xj ∂x k gih − ∂xi ∂x k gj h + 41 s ∂x k gis ∂x s  gj h − ∂x k gj s ∂x s  gih ,

Rj i  k  h = − 21 ∂x i  ∂x k gj h .

1.7 METRICS ON THE COTANGENT BUNDLE

In this section, we continue the discussion of Section 1.6 and introduce special classes of Walker
metrics by specializing those given in Remark 1.30. We shall discuss Riemannian extensions, de-
formed Riemannian extensions, and modified Riemannian extensions. We introduce these metrics
in this section; many of their properties will be developed in subsequent chapters.

RIEMANNIAN EXTENSIONS
This class of manifolds was introduced by Patterson and Walker [177]. Their construction defines a
Walker metric on the cotangent bundle T ∗ M of any affine manifold (M, ∇). The original construc-
tion of Patterson and Walker was later extended by Afifi [3]; this gives rise to a family of metrics
on the cotangent bundle of any affine manifold. These metrics play a distinguished role in the study
of some curvature problems. We adopt the notational conventions established in Section 1.1. Let ι
be the evaluation map and let X C be the complete lift of a vector field X on M. These are defined
invariantly by the relations:

ιX(P , ω) = X(P ), ω  and XC (ιZ) = ι[X, Z] .

In a system of local coordinates (x i , xi  ) we have:

ιX = Xi xi  and XC = Xj ∂xj − xi  (∂xj Xi )∂x j  for X = Xj ∂xj .

Definition 1.32 Let (M, ∇) be an affine manifold of dimension m̄. The Riemannian extension g∇
of (M, ∇) is the pseudo-Riemannian metric of neutral signature (m̄, m̄) on the cotangent bundle
T ∗ M, which is characterized by the identity:

g∇ (X C , Y C ) = −ι(∇X Y + ∇Y X) .

In the system of induced coordinates (x i , xi  ) on T ∗ M as described in Equation (1.1.d) and in

Equation (1.1.e), the Riemannian extension takes the form:
 
−2xk  ij k (x ) Idm̄
g∇ = ,
Idm̄ 0
34 1. BASIC NOTIONS AND CONCEPTS
with respect to {∂x1 , . . . , ∂xm̄ , ∂x 1 , . . . , ∂x m̄ }; here the indices i and j range from 1, . . . , m̄,
i  = i + m̄, and ij k are the Christoffel symbols of the connection ∇ with respect to the coor-
dinates (x i ) on M. More explicitly:
j
g∇ (∂xi , ∂xj ) = −2xk  ij k (
x ), g∇ (∂xi , ∂x j  ) = δi , g∇ (∂x i  , ∂x j  ) = 0 .

Let σ be the natural projection from T ∗ M to M. The Walker distribution is the null parallel plane
field of maximal dimension m̄ given by D = ker{σ∗ }.

Let (M, g) be a pseudo-Riemannian manifold. The Riemannian extension of the Levi-Civita

connection inherits many of the properties of the base manifold. For instance, (M, g) has constant
sectional curvature if and only if (T ∗ M, gg ∇ ) is locally conformally flat (see Cendán-Verdes, García-
Río, and Vázquez-Abal [51]). However, the main applications of the Riemannian extensions appear
when considering affine connections that are not the Levi-Civita connection of any metric. We refer
to Yano and Ishihara [215] for the proof of the following result:

Lemma 1.33 Let (M, ∇) be an affine manifold. Let R̃ be the curvature operator of the Riemannian
extension (T ∗ M, g∇ ) and let R be the curvature operator of (M, ∇). Then:
  
R̃kj i h = Rkj i h , R̃kj i  h = −Rkj h i , R̃kj  i h = −Rhik j , R̃k  j i h = −Rhij k ,


R̃kj i h = xa  ∇∂xh Rkj i a − ∇∂xi Rkj h a +ht a Rkj i t + kt a Rihj t

+j t a Rhik t + it a Rkj h t ,

where Rαβγ δ (resp. R̃αβγ δ ) denote the components of R (resp. R̃).

By Lemma 1.33, (T ∗ M, g∇ ) is locally symmetric if and only if (M, ∇) is locally symmetric.

Furthermore, (T ∗ M, g∇ ) is locally conformally flat if and only if (M, ∇) is projectively flat (see Afifi
[3]). Any projectively flat pseudo-Riemannian manifold has constant sectional curvature. However,
there are many projectively flat affine connections. We refer to Weyl [208] and to Willmore [211]
for further details.

DEFORMED RIEMANNIAN EXTENSIONS

The Riemannian extensions of Definition 1.32 can be generalized.

Definition 1.34 Let  be a symmetric (0, 2)-tensor field on an affine manifold (M, ∇) and let σ
be the natural projection from T ∗ M to M. The deformed Riemannian extension g∇, is the metric of
neutral signature (m̄, m̄) on the cotangent bundle given by:

g∇, = g∇ + σ ∗  .
 1.7. METRICS ON THE COTANGENT BUNDLE 35
Let ij be the Christoffel symbols of the connection ∇ and let ij be the local components of the
k

symmetric (0, 2)-tensor field . In local coordinates the deformed Riemannian extension is given
by:
 
x ) + ij (
−2xk  ij k ( x ) Idm̄
g∇, = (1.7.a)
Idm̄ 0
or, equivalently,
j
g∇, (∂xi , ∂xj ) = −2xk  ij k (
x ) + ij (
x ), g∇, (∂xi , ∂x j  ) = δi , g∇, (∂x i  , ∂x j  ) = 0 .

Note that the crucial terms g∇, (∂xi , ∂xj ) now no longer vanish on the 0-section. As was the case
for the Riemannian extension, the Walker distribution is the kernel of the projection from T ∗ M:
D = ker{σ∗ } = Span{∂x i  }.

The tensor  plays an essential role. Even if the underlying connection is flat, the deformed
Riemannian extension need not be flat. Deformed Riemannian extensions are characterized by their
curvature as follows (see Afifi [3]):

Lemma 1.35 The Walker metric of Remark 1.30 is locally a deformed Riemannian extension if and only
if the curvature tensor satisfies R( · , D)D = 0.

Proof. We use Lemma 1.31 to see that the coefficients gij of the metric are linear in the xk  variables.
Thus, the metric locally takes the form given in Equation (1.7.a). 2

Deformed Riemannian extensions have nilpotent Ricci operator and hence, they are Einstein
if and only if they are Ricci flat. They can be used to construct non-flat Ricci flat pseudo-Riemannian
manifolds:

Lemma 1.36 A deformed Riemannian extension (T ∗ M, g∇, ) is Einstein if and only if it is Ricci flat.
This happens if and only if the Ricci tensor of the connection ∇ is skew-symmetric.

MODIFIED RIEMANNIAN EXTENSIONS

Definition 1.37 Let ∇ be a torsion-free connection and let  be a symmetric (0, 2)-tensor field
on a manifold M. Let T , S be (1, 1)-tensor fields on M. The modified Riemannian extension is the
metric on T ∗ M given by
g∇,,T ,S = ιT ◦ ιS + g∇, .
36 1. BASIC NOTIONS AND CONCEPTS
Let (T ◦ S)rs
ij := 2 (Ti Sj + Tj Si ). In local coordinates:
1 r s r s

 
xr  xs  (T ◦ S)(
x )rs − 2x k   ij
k (
x ) +  ij (
x ) Id m̄
g∇,,T ,S = ij . (1.7.b)
Idm̄ 0

In other words:

g(∂xi , ∂xj ) = xr  xs  (T ◦ S)rs x ) + ij (

ij ( x ) − 2xk  ij k (
x ),
j
g(∂xi , ∂x j  ) = δi , and g(∂x i  , ∂x j  ) = 0 .

The particular case where T = c Id and S = Id will be denoted by g∇,,c . The crucial point is that
this is now quadratic in the fiber coordinates xi  .

As well as in the previous cases, modified Riemannian extensions are Walker metrics where
the Walker distribution is given by D = ker{σ∗ }. Modified Riemannian extensions are characterized
by their curvature (see Afifi [3]). In particular, any locally symmetric Walker metric, as given in
Remark 1.30, is locally a modified Riemannian extension.

Lemma 1.38 The Walker metric of Remark 1.30 is locally a modified Riemannian extension if and only
if the covariant derivative of the curvature operator satisfies (∇D R)( · , D)D = 0.

The following is a very general observation and gives rise to a wide family of neutral signature
Einstein metrics which are not Ricci flat:

Lemma 1.39 Let (M, ∇) be an affine manifold of dimension m̄ and let c = 0.The modified Riemannian
extension (T ∗ M, g∇,,c ) is Einstein if and only if  = c(m̄−1)
4
ρs .

Proof. Let τ = 6c be the scalar curvature of the modified Riemannian extension g = g∇,,c . We
use Lemma 1.31 to see that the trace-free Ricci tensor
gρ
0 = gρ − τ
2m̄ g

of g is given by
gρ
0 = 2 σ ∗ ρs − 21 c(m̄ − 1)σ ∗ . 2

1.8 SELF-DUAL WALKER METRICS

Four-dimensional geometry is exceptional in many respects. In this section, we discuss the Hodge 
operator and self-duality. Throughout this section, let (V , ·, ·) be an oriented inner product space
of signature (p, q). Let {e1 , e2 , e3 , e4 } be an orthonormal basis so that the orientation is given by

e1 ∧ e2 ∧ e3 ∧ e4 .
 1.8. SELF-DUAL WALKER METRICS 37
THE HODGE  OPERATOR
Let  be the Hodge operator. This operator is characterized by the relation:
ω1 ∧ ω2 = ω1 , ω2 e1 ∧ e2 ∧ e3 ∧ e4 for all ωi ∈ 2 (V ∗ ) ,
where we denote by ω1 , ω2  the induced product on 2 = 2 (V ∗ ). Let εi = ei , ei  = ±1. Then:
j j
ei ∧ ej ∧ (ek ∧ e ) = (δki δ − δi δk ) εi εj e1 ∧ e2 ∧ e3 ∧ e4 .
Since 2 = Id if ·, · is definite or has neutral signature, the Hodge  operator induces a splitting
2 = + ⊕ − , where + and − denote the spaces of self-dual and anti-self-dual 2-forms
± = {α ∈ 2 :  α = ±α} .
Furthermore, the induced inner products on ± are positive definite if ·, · is Riemannian, but
they are Lorentzian if ·, · is of neutral signature. An orthonormal basis for the self-dual and
anti-self-dual space is given by
e 1 ∧ e 2 ± ε 3 ε4 e 3 ∧ e 4 ± e 1 ∧ e 3 ∓ ε 2 ε4 e 2 ∧ e 4
± = Span E1± = √ , E2 = √ ,
2 2
e 1 ∧ e 4 ± ε 2 ε3 e 2 ∧ e 3
E3± = √ . (1.8.a)
2
Henceforth, we will use these two bases when coordinates in the self-dual or anti-self-dual space are
needed, except where indicated explicitly to the contrary. The Weyl operator in the Singer–Thorpe
decomposition of Theorem 1.24 further splits in the form:
W = W + + W − where W ± := 1
2 (W ±  W ) .
The curvature operator is said to be self-dual (resp. anti-self-dual ) if we have W − = 0 (resp.
W + = 0).
We now pass to the geometric setting. We adopt the notation of Equation (1.1.b) and let δ
be the coderivative mapping C ∞ (p M) to C ∞ (p−1 M); δ is the L2 adjoint of d. We can express
δ in terms of the Hodge  operator:
δ = (−1)mp+m+1 d = (−1)p −1 d . (1.8.b)
Let m = 4. If (M, g, J+ ) is an almost para-Hermitian manifold of signature (2, 2), or if
(M, g, J− ) is a Hermitian manifold of positive definite signature (0, 4), then J± defines an ori-
entation of M so that ± is a self-dual 2-form. However, if (M, g, J− ) is a 4-dimensional almost
Hermitian manifold with a metric of neutral signature (2, 2), then − is anti-self-dual.
The special significance of the (anti)-self-dual condition is that a 4-dimensional almost (para)-
Hermitian structure is self-dual if and only if the Bochner curvature tensor described in Remark 1.26
vanishes (see, for example, Bryant [39] or Tricerri and Vanhecke [200]). In the almost pseudo-
Hermitian case of signature (2, 2), the Bochner tensor vanishes if and only if the metric is anti-self-
dual.
38 1. BASIC NOTIONS AND CONCEPTS
SELF-DUALITY
The action of the Hodge  operator on the space of self-dual Walker metrics was investigated by
Díaz-Ramos, García-Río, and Vázquez-Lorenzo [71]; a 4-dimensional Walker metric is self-dual
if and only the metric of Remark 1.30 in Walker coordinates (x 1 , x 2 , x1 , x2 ) takes the form:

g11 (x 1 , x 2 , x1 , x2 ) = x13 A +x12 B +x12 x2 C +x1 x2 D +x1 P +x2 Q+ξ,
g22 (x 1 , x 2 , x1 , x2 ) = x23 C +x22 E +x1 x22 A +x1 x2 F +x1 S +x2 T +η,
(1.8.c)
g12 (x 1 , x 2 , x1 , x2 ) = 21 x12 F + 21 x22 D +x12 x2 A +x1 x22 C + 21 x1 x2 (B + E )
+x1 U +x2 V +γ ,

where the coefficients are smooth functions of (x 1 , x 2 ). It was observed in Calviño-Louzao et al.
[43] that the metrics of Equation (1.8.c) can be realized on the cotangent bundle of an affine surface
 by means of a slight generalization of the modified Riemannian extensions as follows.

Theorem 1.40 A 4-dimensional Walker manifold is self-dual if and only if it is locally isometric to the
cotangent bundle T ∗  of an affine surface (, ∇), with a metric tensor

g = ιX(ι Id ◦ι Id) + ι Id ◦ιT + g∇ + σ ∗  ,

where X, T , ∇ and  are a vector field, a (1, 1)-tensor field, a torsion-free affine connection, and a
symmetric (0, 2)-tensor field on , respectively.

Proof. Set X = A(x 1 , x 2 )∂x1 + C (x 1 , x 2 )∂x2 to be a locally defined vector field on  so that it
defines a function ιX on T ∗  as follows:

ιX = x1 A(x 1 , x 2 ) + x2 C (x 1 , x 2 ).

Hence,

(ιX · ι Id ◦ι Id)11 = x13 A(x 1 , x 2 ) + x12 x2 C (x 1 , x 2 ),

(ιX · ι Id ◦ι Id)12 = x12 x2 A(x 1 , x 2 ) + x1 x22 C (x 1 , x 2 ),
(ιX · ι Id ◦ι Id)22 = x1 x22 A(x 1 , x 2 ) + x23 C (x 1 , x 2 ) .

Next, define locally a (1, 1)-tensor field T on  by specializing its components by

T11 = B (x 1 , x 2 ), T12 = D(x 1 , x 2 ), T21 = F (x 1 , x 2 ), T22 = E (x 1 , x 2 ).

Now the action of ι on T described in Section 1.1 gives a 1-form ιT on T ∗  with components

(ιT )1 = x1 B (x 1 , x 2 ) + x2 D(x 1 , x 2 ),

(ιT )2 = x1 F (x 1 , x 2 ) + x2 E (x 1 , x 2 ),
 1.9. RECURRENT CURVATURE 39
and therefore,

(ιT ◦ ι Id)11 = x12 B (x 1 , x 2 ) + x1 x2 D(x 1 , x 2 ),

(ιT ◦ ι Id)12 = 21 x12 F (x 1 , x 2 ) + x22 D(x 1 , x 2 )
+x1 x2 (B (x 1 , x 2 ) + E (x 1 , x 2 )) ,
(ιT ◦ ι Id)22 = x1 x2 F (x 1 , x 2 ) + x22 E (x 1 , x 2 ).

Finally, consider the affine connection ∇ locally defined by the Christoffel symbols

11 1 = − 21 P (x 1 , x 2 ), 12 1 = − 21 S(x 1 , x 2 ), 22 1 = − 21 U (x 1 , x 2 ),

11 2 = − 21 Q(x 1 , x 2 ), 12 2 = − 21 T (x 1 , x 2 ), 22 2 = − 21 V (x 1 , x 2 ),

and the symmetric (0, 2)-tensor field  locally given by

11 = ξ(x 1 , x 2 ), 12 = η(x 1 , x 2 ), 22 = γ (x 1 , x 2 ).

Now the proof follows directly from Equation (1.8.c). 2

Remark 1.41 Any 4-dimensional para-Kähler metric is necessarily a Walker metric. Consequently,
the distributions corresponding to the ±1-eigenspaces of the para-complex structure J+ form a null
parallel plane field of maximal dimension. The underlying structure of any 4-dimensional Bochner
flat para-Kähler metric is given in Theorem 1.40.

1.9 RECURRENT CURVATURE

An affine connection ∇ is said to be recurrent (or to have recurrent curvature) if there exists a 1-form
ω such that the covariant derivative of the curvature operator R satisfies ∇ R = ω ⊗ R i.e.,

∇ R(X1 , X2 ; X4 )X3 = ω(X4 )R(X1 , X2 )X3 .

As the curvature of an affine surface (, ∇) satisfies (see, for example, Nomizu [166])

R(X, Y )Z = ρ(Y, Z)X − ρ(X, Z)Y,

the curvature operator is recurrent if and only if the Ricci tensor is recurrent. Recurrent affine
surfaces have been classified by Wong [213]. The classification depends on the different possibilities
for the symmetric and the alternating Ricci tensors. Affine surfaces with recurrent curvature play an
important role in understanding the geometry of Riemannian extensions as we shall see presently.
The following result is based on work of Derdzinski [66] and of Wong [213].

Theorem 1.42 Let (, ∇) be an affine surface with recurrent curvature. There exists a coordinate system
(x 1 , x 2 ) where the (possibly) non-zero Christoffel symbols are given by:
40 1. BASIC NOTIONS AND CONCEPTS
1. If Rank{ρs } = 1 and ρa  = 0, then 11 1 = 12 2 = ∂x2 θ and 11 2 = ∂x1 θ − ∂x2 θ where
θ = θ(x 1 , x 2 ) satisfies ∂x2 ∂x2 θ = 0.
2. If Rank{ρs } = 1 and ρa = 0, then the only non-zero Christoffel symbol takes the form
11 2 = 11 2 (x 1 , x 2 ) where ∂x2 11 2  = 0.
3. If det{ρs } < 0 and ρa = 0, then either the only non-zero Christoffel symbol is 22 2 with
∂x1 22 2  = 0, or 11 1 = (1 − c)−1 ∂x1 θ and 22 2 = (1 + c)−1 ∂x2 θ where θ = θ(x 1 , x 2 ) sat-
isfies ∂x1 ∂x2 θ  = 0 and where c  = 0, ±1.
4. If det{ρs } < 0 and ρa = 0, then 11 1 = ∂x1 θ and 22 2 = ∂x2 θ where θ = θ (x 1 , x 2 ) satisfies
∂x1 ∂x2 θ  = 0.
5. If det{ρs } > 0 and ρa  = 0, then there exists θ = θ (x 1 , x 2 ) with ∂x1 ∂x1 θ + ∂x2 ∂x2 θ = 0 and
there exists 0  = c ∈ R so that
 
11 1 = 12 2 = −22 1 = (1 + c2 )−1 ∂x1 θ + c∂x2 θ  ,
22 2 = 12 1 = −11 2 = (1 + c2 )−1 ∂x2 θ − c∂x1 θ .

6. If det{ρs } > 0 and ρa = 0, then there exists θ = θ (x 1 , x 2 ) with ∂x1 ∂x1 θ + ∂x2 ∂x2 θ = 0 so that
11 1 = 12 2 = −22 1 = ∂x1 θ and 22 2 = 12 1 = −11 2 = ∂x2 θ.
7. If ρs = 0 and ρa  = 0, then there exists θ = θ (x 1 , x 2 ) with ∂x1 ∂x2 θ = 0 so that we have
11 1 = −∂x1 θ and 22 2 = ∂x2 θ .

Remark 1.43 The Ricci tensor is symmetric and non-degenerate in Cases (4) and (6). Furthermore,
in these cases, there is a function f locally defined on  such that g = ef ρ is a metric tensor on 
such that ∇ is its Levi-Civita connection (see Wong [213] for details).

1.10 CONSTANT CURVATURE

In this section, we shall discuss the Jacobi operator, symmetric spaces, manifolds of constant sectional
curvature, and Kähler manifolds of constant holomorphic sectional curvature. A vector field X along
a geodesic γ is said to be a Jacobi vector field if

Ẍ + R(X, γ̇ )γ̇ = 0 .

We use Jacobi vector fields to show that the geodesic involution is an isometry if and only if ∇ R = 0.
We also use Jacobi vector fields to show that the local geometry of a pseudo-Riemannian manifold
of signature (p, q) and of constant sectional curvature c is determined by (p, q, c). We conclude
the section with a brief introduction to the theory of manifolds of constant holomorphic sectional
curvature.
 1.10. CONSTANT CURVATURE 41
THE JACOBI OPERATOR AND JACOBI VECTOR FIELDS
The Jacobi operator J (X) is an endomorphism of the tangent bundle, which is defined for a tangent
vector X by setting
J (X) : Y → R(Y, X)X for Y ∈ C ∞ (T M) . (1.10.a)
It plays an important role in the study of geodesic sprays, which we outline as follows. Let ∇ be an
affine connection and let γ (s) be a geodesic in M. Recall that a vector field X along γ is said to be
a Jacobi vector field if it satisfies the equation

Ẍ + J (γ̇ )X = 0 i.e., ∇γ̇ ∇γ̇ X + R(X, γ̇ )γ̇ = 0 .

One says that a smooth map T : [0, ε] × [0, ε] → M is a geodesic spray if T is an embedding such
that the curves γt (s) := T (t, s) are geodesics for all t. One has (see, for example, do Carmo [74]):

Lemma 1.44 Let T (t, s) be a geodesic spray. Then the variation T∗ (∂t ) is a Jacobi vector field along the
geodesics γt (s) = T (t, s).

Proof. To simplify the notation, we identify ∂t with γ∗ ∂t and ∂s with γ∗ ∂s . Since the curves s → γt (s)
are geodesics for all t, we may compute:
0 = ∇∂t ∇∂s ∂s (The curves γt (s) are geodesics for any t)
= R(∂t , ∂s )∂s + ∇∂s ∇∂t ∂s (Definition of the curvature; [∂t , ∂s ] = 0)
= J (∂s )∂t + ∇∂s ∇∂s ∂t (∇ is a torsion-free connection)
= J (γ̇ )X + Ẍ. (Disentangling the notation)
The desired result now follows. 2

SYMMETRIC SPACES
Jacobi vector fields appear in many contexts. We present an example, arising from work of Cartan
[49, 50], to illustrate their use and to motivate the study of the spectral geometry of the Jacobi operator
in Section 1.11. Let (M, g) be a pseudo-Riemannian manifold. Let expP be the exponential map;
this is the local diffeomorphism from a neighborhood of 0 in TP M to a neighborhood of P in M,
defined in Section 1.2; it is characterized by the fact that the curves s → expP (sv) are geodesics
starting at P with initial direction v for v ∈ TP M. The geodesic symmetry at P is then defined on a
suitable neighborhood of P by setting:

SP (Q) = expP {− exp−1

P Q} .

We may use expP to identify a neighborhood of 0 in TP M with a neighborhood of P in M and

to regard TP M as a pseudo-Riemannian manifold locally isometric to M; under this identification,
42 1. BASIC NOTIONS AND CONCEPTS
SP (v) = −v is simply multiplication by −1 and the straight lines through the origin are geodesics.
One has the following result:

Lemma 1.45 Let (M, g) be a connected pseudo-Riemannian manifold.

1. The following assertions are equivalent and, if either is satisfied, then (M, g) is said to be locally
symmetric or to be a local symmetric space:
(a) The geodesic symmetry SP is an isometry for all P ∈ M.
(b) ∇ R = 0.
2. If (M, g) is locally symmetric, then (M, g) is locally homogeneous, i.e., given any two points
P , Q ∈ M, there is an isometry P ,Q from some neighborhood of P in M to some neighborhood of
Q in M.

Remark 1.46 If (M, g) is a local symmetric space that is complete and simply connected, then
(M, g) is said to be globally symmetric or to be a symmetric space. In this setting, the geodesic symmetry
extends to a global isometry of (M, g), and (M, g) is homogeneous. If G0 is the isotropy subgroup of
the group of isometries G of (M, g), then M = G/G0 with the induced metric. Global symmetric
spaces form a very special class of pseudo-Riemannian manifolds and techniques of group theory
are used to study them—the associated Lie algebras g and g0 of G and of G0 play a central role. We
refer to Helgason [123] for further details.

Proof. We use expP to identify a neighborhood of 0 in TP M with a neighborhood of P in M

henceforth. Suppose first, that SP is an isometry. Then SP commutes with ∇. Thus,

∇ RP (−X, −Y ; −Z)(−T ) = −∇ RP (X, Y ; Z)T .

This implies that ∇ RP = 0 and shows that Assertion (1a) implies Assertion (1b).
We use Jacobi vector fields to show that the reverse implication holds. Suppose Assertion (1b)
holds. Fix v ∈ TP M and consider the geodesic spray T (t, s; w) := s(v + tw). The associated Jacobi
vector field of Lemma 1.44 is given by Y (s) = sw; Y is determined by the Jacobi equation:

Ÿ + J (γ̇ )Y = 0 with Y (0) = 0 and Ẏ (0) = w .

Let {ei } be a parallel orthonormal frame along γ̇ . Expand

j
{R(ei , γ̇ )γ̇ }(s) = ci (s)ej (s) .

We compute:
j j
∇γ̇ {ci ej } = ċi · ej = ∇γ̇ {R(ei , γ̇ )γ̇ }
= (∇γ̇ R)(ei , γ̇ )γ̇ + R(∇γ̇ ei , γ̇ )γ̇ + R(ei , ∇γ̇ γ̇ )γ̇ + R(ei , γ̇ )(∇γ̇ γ̇ ) = 0.
 1.10. CONSTANT CURVATURE 43
j
Consequently, the structure functions ci are in fact constant. Let Y (s) = sw = a j (s)ej (s). Then
the Jacobi equation becomes

ä j (s) + a k (s)cki = 0 for 1 ≤ j ≤ m with a j (0) = 0 and ȧ j (0) = g0 (w, ej ) . (1.10.b)

Let bj (s) = −a j (−s). The bj (s) also satisfy Equation (1.10.b) and, hence, by the uniqueness
of the solution to an ordinary differential equation, we have bj (s) = a j (s), i.e., we have that
a j (−s) = −a j (s). We have gγ (s) (ei (s), ej (s)) = g0 (ei , ej ) is independent of s. Consequently,

gγ (s)(Y (s), Y (s)) = a i (s)a j (s)gγ (s) (ei (s), ej (s))

= a i (−s)a j (−s)gγ (−s) (ei (−s), ej (−s)) = gγ (−s) (Y (−s), Y (−s)) .

This implies

s 2 gsv (w, w) = gγ (s) (Y (s), Y (s)) = gγ (−s) (Y (−s), Y (−s)) = s 2 g−sv (w, w) .

This implies gu = g−u for u  = 0 and thus, SP is an isometry on a punctured neighborhood of P ;

continuity then implies SP is an isometry at P , as well. Thus, Assertion (1b) implies Assertion (1a);
this completes the proof of Assertion (1).
Assume that the geodesic involution is always an isometry. We establish Assertion (2) as
follows. Suppose there is a geodesic γ from P to Q. We may use the geodesic involution around
the midpoint of γ to construct a local isometry interchanging P and Q. As M is connected, we can
compose a succession of such isometries to construct a local isometry between any two points of M
and thereby show (M, g) is locally homogeneous. 2

MANIFOLDS OF CONSTANT SECTIONAL CURVATURE

Previously, we used Jacobi vector fields to show the equivalence of a geometric condition (geodesic
symmetries are isometries) with a purely algebraic condition (the curvature is parallel). In this
section, we again use Jacobi vector fields to establish the equivalence of a geometric condition with
an algebraic one. Let π be a non-degenerate 2-plane in TP M. We define the sectional curvature κ(π)
by setting:
R(x, y, y, x)
κ(π) := for any basis {x, y} for π .
g(x, x)g(y, y) − g(x, y)2
We say that a pseudo-Riemannian manifold (M, g) has constant sectional curvature c if κ(π ) = c for
any non-degenerate 2-plane π at any point P of M. The requirement that π be non-degenerate is,
of course, equivalent to the condition g(x, x)g(y, y) − g(x, y)2  = 0; this condition is independent
of the particular basis chosen.
We now give a purely algebraic formalism. Recall that a Riemannian curvature model is a triple
(V , ·, ·, A), where (V , ·, ·) is an inner product space and where A ∈ ⊗4 V ∗ satisfies the identities
of the Riemann curvature tensor given in Equations (1.2.l)-(1.2.n).

Lemma 1.47 Let M := (V , ·, ·, A) be a Riemannian curvature model.

44 1. BASIC NOTIONS AND CONCEPTS
1. The following conditions are equivalent and, if either is satisfied, then M is said to have constant
sectional curvature c.

(a) A(x, y, z, w) = c{x, wy, z − x, zy, w}.

A(e1 ,e2 ,e2 ,e1 )
(b) e1 ,e1 e2 ,e2 −e1 ,e2 2
= c for any non-degenerate 2-plane π = Span{e1 , e2 } in V .

2. If M has constant sectional curvature c and if {e1 , e3 } is an orthonormal set, then we have that
J (e1 )e3 = ce1 , e1 e3 .

Remark 1.48 The sectional curvature is a continuous function on the Grassmannian Gr 02 (V ) of

non-degenerate 2-planes in V . In the positive definite setting, this space is compact and consequently
the sectional curvature is bounded. This fact does not hold true in the indefinite setting since Gr 02 (V )
is an open subset of the full Grassmannian Gr 2 (V ) and, in fact, the sectional curvature is bounded
if and only if it is constant. Furthermore, R(x, y, y, x) = 0 whenever x, y span a degenerate plane
if and only if the sectional curvature is constant. We refer to O’Neill [169] for more information.

Proof. It is immediate that Assertion (1a) implies Assertion (1b). Assume, conversely, that Asser-
tion (1b) holds. Let {x, y, z} be an orthonormal set of vectors in V . Set:

ε = x, xy, y = ±1,

cos(t)x + sin(t)y if ε = 1
ξ(t) := .
cosh(t)x + sinh(t)y if ε = −1

We may then compute that

cos2 (t) + sin2 (t) if ε = 1

ξ(t), ξ(t) = x, x · = x, x = ±1
cosh2 (t) − sinh2 (t) if ε = −1

is independent of the parameter t. As M has constant sectional curvature c, we have that:

cx, xz, z = cξ(t), ξ(t)z, z = A(ξ(t), z, z, ξ(t))

cos2 (t) + sin2 (t) if ε = 1
= c x, xz, z
cosh (t) − sinh (t) if ε = −1
2 2

cos(t) sin(t) if ε = 1
+ 2A(x, z, z, y) .
cosh(t) sinh(t) if ε = −1

This shows that:

A(x, z, z, y) = 0 if {x, y, z} is an orthonormal set . (1.10.c)
Suppose {x, y, z, w} is an orthonormal set. We argue similarly and polarize the identity of Equa-
tion (1.10.c) to conclude
A(x, z, w, y) + A(x, w, z, y) = 0 . (1.10.d)
 1.10. CONSTANT CURVATURE 45
We now use the curvature symmetries together with Equation (1.10.d) to see:

0 = A(x, y, z, w) + A(y, z, x, w) + A(z, x, y, w)

= A(x, y, z, w) − A(y, x, z, w) − A(x, z, y, w)
= 3A(x, y, z, w) so

A(x, y, z, w) = 0 if {x, y, z, w} is an orthonormal set . (1.10.e)

We verify that Assertion (1b) implies Assertion (1a) as follows. Let {e1 , . . . , em } be an orthonormal
basis for V . By the curvature symmetries, A(ei , ej , ek , e ) = 0 unless {i, j } and {k, } are distinct.
By Equation (1.10.c) and Equation (1.10.e), A(ei , ej , ek , e ) = 0 unless either (i, j ) = (k, ) or
(i, ) = (j, k). The desired result now follows from the identity of Assertion (1b).
Let M have constant sectional curvature c. Let {e1 , e2 , e3 } be an orthonormal set of vectors
in V ; Assertion (1a) implies that A(e3 , e1 , e1 , e2 ) = 0 so J (e1 )e3 ⊥ e2 . Since A(e3 , e1 , e1 , e1 ) = 0,
we conclude that J (e1 )e3 is perpendicular to e1 and e2 . Since e2 was an arbitrary unit vector
perpendicular to e1 and e3 , we conclude J (e1 )e3 is some multiple λ of e3 . We establish Assertion (2)
by computing:
ce1 , e1 e3 , e3  = A(e3 , e1 , e1 , e3 ) = J (e1 )e3 , e3  = λe3 , e3  . 2

Let (M, g) be a pseudo-Riemannian manifold. If (TP M, gP , RP ) has constant sectional

curvature cP at each point P of M and if m ≥ 3, then a Schur type lemma shows that, in fact, cP
is actually constant. Such manifolds are often also called space forms; if c > 0, the manifold is said
to be a spherical space form while if c < 0, it is said to be a hyperbolic space form. If c = 0, then the
manifold is flat since the curvature tensor vanishes identically. We refer to Wolf [212] for further
information and content ourselves with using Lemma 1.47 to establish the following result:

Lemma 1.49 Let (M, g) and (M̃, g̃) be two connected pseudo-Riemannian manifolds of constant
sectional curvature c and signature (p, q). Let P and P̃ be points of M and M̃. Then there is a local
isometry  from M to M̃ with (P ) = P̃ .

Proof. We use expP to identify a neighborhood of 0 in TP M with a neighborhood of P in M. Let

g0 = gP . Fix v ∈ TP M so that g0 (v, v) = ε = ±1. Let w ∈ TP M be such that {v, w} forms an
orthonormal set relative to g0 . We form the geodesic γ (s) := sv. Let e(s) be a parallel vector field
along γ with e(0) = w; we regard e as a vector valued map from a neighborhood of 0 in R to TP M.
Set
⎧ ⎫
⎨ |c|−1 sin(|c|s)e(s) if cε > 0 ⎬
θ(v, s) := se(s) if c = 0 .
⎩ ⎭
|c|−1 sinh(|c|s)e(s) if cε < 0
46 1. BASIC NOTIONS AND CONCEPTS
Let {e1 , . . . , em } be an orthonormal basis for TP M where e1 = v. We wish to show that:
⎧ ⎫
⎨ 0 if i  = j ⎬
gsv (ei , ej ) = g0 (v, v) if i = j = 1 . (1.10.f )
⎩ −2 ⎭
s θ (v, s)2 g0 (ei , ei ) if i = j > 1

This determines the metric away from the origin and off the light cone of null vectors; the metric on
the light cone can then be determined by continuity. The lemma will then follow by choosing an
isometry between TP M and TP̃ M̃; this is possible as M and M̃ have the same signature.
We establish Equation (1.10.f) as follows. Examine the Jacobi vector field Y := sw on γ arising
from the geodesic spray T (t, s) := s(v + tw). We use the Levi-Civita connection to covariantly
differentiate vector fields—thus, for example, γ̈ = 0 since γ is a geodesic and Ÿ = −J (γ̇ )Y since
Y is a Jacobi vector field. We compute:

∂s ∂s gγ (s) (γ̇ (s), Y (s)) = ∂s gγ (s) (γ̈ (s), Y (s)) + ∂s gγ (s) (γ̇ (s), Ẏ (s))
= 0 + ∂s gγ (s) (γ̇ (s), Ẏ (s)) = gγ (s) (γ̈ (s), Ẏ (s)) + g(γ̇ (s), Ÿ (s))
= 0 − gγ (s) (γ̇ (s), J (γ̇ )Y (s)) = −R(Y (s), γ̇ (s), γ̇ (s), γ̇ (s)) = 0 .

Since gγ (s) (γ̇ (s), Y (s))(0) = 0 and {∂s gγ (s) (γ̇ (s), Y (s))}(0) = g0 (v, w) = 0, we may conclude that
Y (s) ⊥ γ̇ (s) for all s. By Assertion (2) of Lemma 1.47, we have:

Ÿ = −J (γ̇ )Y = −cεY with Y (0) = 0 and Ẏ (0) = w . (1.10.g)

Then Z(s) := θ (v, s)e(s) satisfies the same ordinary differential equation given in Equation (1.10.g)
that is satisfied by Y and consequently Z(s) = Y (s) = sw. Note that

gγ (s) (e(s), e(s)) = g(e(0), e(0)) = g0 (w, w) .

Thus, we have gγ (s) (sw, sw) = θ(v, s)2 g0 (w, w) if g0 (v, w) = 0. Equation (1.10.f) now follows
by polarization. 2
We have actually proved a bit more. By applying the previous lemma to M itself, we have shown
that (M, g) is a two-point homogeneous space, i.e., the local isometries of (M, g) act transitively
on the pseudo-sphere bundles

S ± (M, g) := {X ∈ T M : g(X, X) = ±1} .

Let (V , ·, ·) be an inner product space of signature (p, q). Let c > 0. The pseudo-spheres

S ± (V , ·, ·, c) := {ξ ∈ V : ξ, ξ  = ±c2 }

have constant sectional curvature ±c. Thus, they provide examples with constant sectional curvature
c in all possible signatures.
 1.10. CONSTANT CURVATURE 47
CONSTANT HOLOMORPHIC SECTIONAL CURVATURE
The fact that a (para)-Kähler manifold (M, g, J± ) of dimension m ≥ 4 and of constant sectional
curvature is necessarily flat motivates the investigation of other sectional curvatures more “adapted"
to the (para)-Kähler structure. A plane π is said to be a (para)-holomorphic line if it is left-invariant
by the almost (para)-complex structure, i.e., J± π ⊂ π . If π is non-degenerate, it has signature (0, 2)
or (2, 0) in the complex setting and signature (1, 1) in the para-complex setting. The restriction of
the sectional curvature of an almost (para)-Hermitian manifold to the set of (para)-holomorphic
lines is called the (para)-holomorphic sectional curvature.
We work first in the algebraic setting. Let M := (V , ·, ·, J± , A) be a (para)-Hermitian
curvature model. If x, x  = 0, let π(x) := Span{x, J± x} be the associated non-degenerate (para)-
holomorphic line. The (para)-holomorphic sectional curvature of π is given by:

H (π) := ∓x, x−2 A(x, J± x, J± x, x) .

Unlike the sectional curvature, the (para)-holomorphic sectional curvature does not determine the
whole curvature tensor in general—we refer to the discussion in Brozos-Vázquez, García-Río, and
Gilkey [29]. We are primarily interested in the (para)-Kähler setting—M is said to be a (para)-Kähler
curvature model if the (para)-Kähler identity of Equation (1.4.a) is satisfied:

A(x, y)J± = J± A(x, y) or equivalently A(x, y, J± z, J± w) = ∓A(x, y, z, w) .

Then the (para)-holomorphic sectional curvature determines the curvature (see Kobayashi and No-
mizu [140]). Lemma 1.47 generalizes to this setting to become (see, for example, the discussion in
Gilkey [96]):

Lemma 1.50 Let M := (V , ·, ·, J± , A) be a (para)-Kähler curvature model.The following conditions
are equivalent and, if either is satisfied, then M is said to have constant (para)-holomorphic sectional
curvature:

1. There exist constants λ0 and λ1 so

A(x, y)z = λ0 {y, zx − x, zy} +λ1 {J± y, zJ± x − J± x, zJ± y − 2J± x, yJ± z}.

2. There exists a constant c so that H (π ) = c for any (para)-holomorphic line in V .

If (V , ·, ·, J± , A) is a (para)-Kähler curvature model of constant (para)-holomorphic sec-

tional curvature and if x is a spacelike vector, then it follows from Lemma 1.50 that
⎧ ⎫
⎨ 0 if y = x ⎬
J (x)y = (λ0 ∓ 3λ1 )y if y = J± x .
⎩ ⎭
λ0 y if y ⊥ {x, J± x}
48 1. BASIC NOTIONS AND CONCEPTS
If the inner product ·, · is positive definite, the sectional curvature and hence, the holo-
morphic sectional curvature are necessarily bounded, as noted previously in Remark 1.48. How-
ever, the failure of the Grassmannian of non-degenerate (para)-holomorphic planes to be com-
pact in general causes the (para)-holomorphic sectional curvature to be unbounded. The (para)-
holomorphic sectional curvature is bounded if and only if the model is Bochner flat or, equivalently,
if A(u, J± u, J± u, u) = 0 for all null vectors u ∈ (V , ·, ·) (see, for example, Bonome et al. [25, 27]).
The condition is not that the curvature model has constant (para)-holomorphic sectional curvature.
We say that a (para)-Hermitian manifold is a (para)-complex space form if the curvature has
the form given in Lemma 1.50 (1), where λi = λi (P ). Let (M, g, J± ) be such a manifold with
dim(M) = m. The case m = 2 corresponds to that of a Riemann surface; λ1 plays no role and there
is no constraint on λ0 . The case m = 4 is exceptional. If λ1 (P ) ≡ 0, then (M, g) is a space form, so
we shall suppose this does not happen. In higher dimensions, if m ≥ 6, then, in fact, the coefficients
λi are constant and λ0 ± λ1 = 0. Furthermore, the metric in question is (para)-Kähler.
Note that there are manifolds with pointwise constant holomorphic sectional curvature where
the constant varies with the point (see, for example, Gray [117], Gray and Vanhecke [119], and
Olszak [171]). We suppose (M, g, J± ) is a (para)-Kähler manifold of constant (para)-holomorphic
sectional curvature; the local structure of such manifolds is known (see, for example, Hawley [122]
and Igusa [131]). In the Riemannian setting, they are locally modeled on complex projective space
or its non-compact dual.
We follow the discussion in Gilkey [96] to construct examples. Let (V , ·, ·, J− ) be a pseudo-
Hermitian vector space of signature (p, q), where necessarily p and q are even. Let S + (V , ·, ·) be
the pseudo-sphere of unit spacelike vectors in V . The unit circle S 1 acts on S + (V , ·, ·) by complex
multiplication and we let CPp−2,q be the quotient space; this is the set of spacelike complex lines in
V . This defines the Hopf fibration

π : S + (V , ·, ·) → CPp−2,q .

The Fubini–Study metric gF S on CPp−2,q makes π into a pseudo-Riemannian submersion; π∗ is

an isometry from the horizontal subspaces to the tangent space of the base. The construction is
 p,p
essentially the same in the para-complex setting to define the para-complex projective space CP
and the Fubini–Study metric. One has the following classification result—we refer to Barros and
Romero [10] and to Kobayashi and Nomizu [140] in the complex setting, and to Cruceanu, Fortuny,
and Gadea [60] and to Gadea and Montesinos-Amilibia [85] in the para-complex setting:

Theorem 1.51 Let (M, g) be a (para)-complex space form of dimension m ≥ 6. In the complex setting,
(M, g) is locally holomorphically isometric to (CPu,v , λgF S ) for some (λ, u, v), while in the para-complex
 n , λgF S ) for some λ.
setting, (M, g) is locally para-holomorphically isometric to (CP

The situation is again special in dimension four. There exist almost (para)-Hermitian man-
ifolds whose curvature tensor has the form of Lemma 1.50 for non-constant functions λ0 and
λ1 . Restrict first to the positive definite almost Hermitian setting and let (M, g, J− ) be such a
 1.11. THE SPECTRAL GEOMETRY OF THE CURVATURE TENSOR 49
4-dimensional almost Hermitian manifold. Any such metric is necessarily Einstein and self-dual.
Moreover, the self-dual Weyl curvature operator has exactly two distinct eigenvalues, one with mul-
tiplicity two at each point where λ1  = 0. Then, in a neighborhood of any such point, the metric
g̃ = {24W + }1/3 g is Kähler—see Derdzinski [64]. Hence (M, g, J− ) is locally conformally equiv-
alent to a self-dual Kähler surface (a Bochner flat Kähler surface). This implies, in particular, that
the almost complex structure J− is integrable. Conversely, any Bochner flat Kähler surface with
non-constant scalar curvature gives rise to a Hermitian surface with curvature tensor of the form
given in Lemma 1.50 by a conformal deformation defined in terms of W +  (see Olszak [171]).The
almost para-complex case and the almost Hermitian case of higher signature are analogous. They
depend on extensions of the generalized Goldberg–Sachs theorem (see Apostolov [8] and Ivanov
and Zamkovoy [135]) and the diagonalizability of the self-dual or anti-self-dual Weyl curvature
operator (see Cortés-Ayaso, Díaz-Ramos, and García-Río [59]).

1.11 THE SPECTRAL GEOMETRY OF THE CURVATURE

TENSOR
In this section, we discuss the spectral geometry of the Jacobi operator, of the skew-symmetric
curvature operator, and of the higher order Jacobi operator.

OSSERMAN MANIFOLDS
Osserman [173] initiated the study of the spectral geometry of the Jacobi operator and, for this
reason, his name has become attached to the subject. One says that (M, g) is timelike Osserman
(resp. spacelike Osserman) at P if the eigenvalues of the Jacobi operator J (X) : Y → R(Y, X)X
are constant on S − (TP M, gP ) (resp. S + (TP M, gP )). As these are equivalent conditions provided
that p > 0 and q > 0 (see García-Río et al. [89]), one simply says that (M, g) is Osserman at P . If
this condition holds for every point P of M, then (M, g) is said to be pointwise Osserman; if the
spectrum is independent of the point P , one either adds the words “globally" or omits the modifier
“pointwise" entirely.
Note that the Jacobi operator J (X) is self-adjoint. In the Riemannian setting, Osserman
metrics are essentially classified. Let (V , ·, ·) be a Riemannian (i.e., positive definite) inner product
space of dimension m. Let
A0 (x, y)z := y, zx − x, zy
be the curvature operator of constant sectional curvature appearing in Lemma 1.47. More generally,
if is a skew-symmetric linear operator on (V , ·, ·), then we may define an algebraic curvature
tensor by setting:

A (x, y)z :=  y, z x −  x, z y − 2 x, y z .

If 2 = − Id, then this tensor plays an important role in the analysis of Lemma 1.50. We say that
a collection { 1 , . . . , k } of skew-symmetric endomorphisms of (V , ·, ·) forms a Clif(k) module
50 1. BASIC NOTIONS AND CONCEPTS
structure if it satisfies the Clifford commutation relations:
j
i j + j i = −2δi Id .
The following result of Nikolayevsky [162, 163] uses work of Chi [53] and of Gilkey, Swann, and
Vanhecke [114] and summarizes the best results available at present; note that the exclusion of the
case of dimension 16 is essential since the Cayley plane is Osserman, but its curvature tensor is not
of this type.

Theorem 1.52 Let (V , ·, ·, A) be an m-dimensional Riemannian Osserman curvature model with
m  = 16. Then there exists a Clif(k) module structure { 1 , . . . , k } on V and there exist constants αi so

that A = α0 A0 + αi A i .

The possible Clif(k) module structures are greatly restricted by m; if m is odd, then (V , ·, ·, A)
does not admit a Clifford module structure and A is necessarily of constant sectional curvature. If
m ≡ 2 mod 4, then k = 1 and the tensor takes the form given in Lemma 1.50, where = J− . If
m ≡ 4 mod 8, then k ≤ 3. We refer to work of Adams [2] for further details.
The rank-one symmetric spaces are the space forms, the complex space forms, the quater-
nionic space forms, the Cayley plane, and their negative curvature duals. All these examples are
Osserman. The space forms correspond to k = 0, the complex space forms correspond to k = 1, and
the quaternionic space forms correspond to k = 3 in Theorem 1.52; the Cayley plane is exceptional
and does not fit into this pattern.
In the Lorentzian setting, Osserman metrics have constant sectional curvature (we refer to
Blažić, Bokan, and Gilkey [16] and to García-Río, Kupeli, and Vázquez-Abal [88]). In the higher
signature setting, however, the situation is very different and there is a vast literature on the subject—
the following are only a few of the possible references and are included to give a flavor of the subject
[26, 42, 55, 56, 70, 90, 91], much of it in the 4-dimensional setting [17, 44, 71, 88, 92]. Here, one must
emphasize that the spectrum does not determine a self-adjoint operator in the pseudo-Riemannian
setting, due to the lack of diagonalizability. Hence, one must pay attention to the possibly non-trivial
Jordan normal form of the Jacobi operators and a pseudo-Riemannian manifold (M, g) is said to be
timelike Jordan–Osserman (resp., spacelike Jordan–Osserman) if the Jordan normal form of the Jacobi
operators is constant, respectively, on the pseudo-sphere bundles. Although the timelike and the
spacelike Osserman conditions are equivalent, the corresponding Jordan–Osserman conditions are
not equivalent in the generic situation (see Gilkey [96] for more details). Moreover, the timelike
and spacelike Jordan–Osserman conditions imply that the Jacobi operators are diagonalizable if the
signature is not neutral, but the Jacobi operators can be rather arbitrary in the neutral signature case
(see the work of Gilkey and Ivanova [99, 100] for further details).
The Osserman condition is extended in Section 2.3 to the affine setting. It turns out that an
m-dimensional affine manifold (M, ∇) is affine Osserman at point P of M if J (X) is nilpotent for
all X ∈ TP M, or, equivalently, if {0} is the only eigenvalue of J (X) or, equivalently,
J (X)m = 0 for all X ∈ TP M .
 1.11. THE SPECTRAL GEOMETRY OF THE CURVATURE TENSOR 51
If this condition holds at every point P of M, then (M, ∇) is said to be affine Osserman. We refer,
for example, to the discussion in García-Río et al. [89] for further details. Note that this definition
is perfectly natural in the context of Weyl geometry.
One can also study similar conditions that are determined by the conformal curvature operator
rather than the curvature operator itself. In this setting, a pseudo-Riemannian manifold (M, g)
is said to be conformally Osserman if for each point P ∈ M the spectrum of the Jacobi operator
JW (X) corresponding to the Weyl conformal curvature tensor is constant on the corresponding unit
pseudo-spheres S ± (TP M, gP ). This condition is meaningless if M is 3-dimensional (since the Weyl
conformal curvature tensor vanishes). Furthermore, in dimension four, this condition is equivalent
to the manifold being (anti)-self-dual (see, for example, Blažić et al. [19] and Brozos-Vázquez,
García-Río, and Vázquez-Lorenzo [33]). Riemannian conformally Osserman manifolds have been
recently studied by Nikolayevsky [164, 165]. Note that one of the main difficulties in working with
the conformal Osserman property arises from the fact that the Weyl conformal curvature tensor
does not satisfy the second Bianchi identity.

THE HIGHER ORDER JACOBI OPERATOR

Following the seminal work of Stanilov and Videv [197], one can define a higher order Jacobi operator.
Let {e1 , . . . , ek } be a basis for a non-degenerate k-plane π of signature (u, v). Set:

J (π ) : X → g ij R(X, ei )ej .
i,j

This operator is independent of the particular basis chosen. If {ei } are an orthonormal basis, then

J (π ) := ε1 J (e1 ) + · · · + εk J (ek ) where εi := g(ei , ei ) . (1.11.a)

We refer to Gilkey [96] and to Gilkey, Nikčević, and Videv [110] for further details concerning this
operator. If k = 1, this is the normalized Jacobi operator while, if k = m, this is the Ricci operator.
Let Gr u,v denote the Grassmannian of non-degenerate linear subspaces of T M of signature (u, v).
We suppose 0 ≤ u ≤ p and 0 ≤ v ≤ q. We say that (M, g) is higher order Osserman of type (u, v) if
the spectrum of J (π ) is constant on Gr u,v ; if (u, v) and (ũ, ṽ) are admissible and if u + v = ũ + ṽ,
then (M, g) is higher order Osserman of type (u, v) if and only if (M, g) is higher order Osserman
of type (ũ, ṽ) and, hence, (M, g) is simply said to be k-higher order Osserman where k = u + v. This
condition is very restrictive in the Riemannian and Lorentzian settings since any k-higher order
Osserman manifold is necessarily of constant sectional curvature for 2 ≤ k ≤ m − 2; we refer to
Gilkey [97] in the Riemannian setting and to Gilkey and Stavrov [113] in the Lorentzian setting
for further details. This operator will play an important role in Section 3.5.

THE SKEW-SYMMETRIC CURVATURE OPERATOR

Let (M, g) be a pseudo-Riemannian manifold and let ∇ be an affine connection on M. We shall
usually take ∇ = g ∇ to be the Levi-Civita connection, but, again, this is not strictly necessary and
52 1. BASIC NOTIONS AND CONCEPTS
the questions involved are natural in the context of Weyl geometry. Let π be a non-degenerate
oriented 2-plane in TP M. If {x, y} is an oriented basis for π, the skew-symmetric curvature operator
is defined by setting

R(π)z, w = |g(x, x)g(y, y) − g(x, y)2 |−1/2 R(x, y, z, w) . (1.11.b)

It is independent of the particular oriented basis chosen. Following the seminal work of Ivanov
and Petrova [134], one says (M, g, ∇) is spacelike Ivanov–Petrova (resp. timelike Ivanov–Petrova or
mixed Ivanov–Petrova) at P if the eigenvalues of R(·) are constant on the Grassmannian of oriented
spacelike (resp. timelike or mixed) 2-planes in TP M. Again, these are equivalent conditions so one
simply speaks of (M, g, ∇) being Ivanov–Petrova at P ; if this condition holds for all points P ,
(M, g, ∇) is simply said to be Ivanov–Petrova or globally Ivanov–Petrova; if the eigenvalues vary
with the point in question, the adjective pointwise is often used. In the higher signature setting,
there can be non-trivial Jordan normal form and if the Jordan normal form is constant, one adds the
adjective Jordan. There is a vast literature on the subject and again, we can only cite a few references
[41, 44, 110, 115, 216].
 53

CHAPTER 2

The Geometry of Deformed

Riemannian Extensions
Let M be a smooth manifold. In this chapter, we shall consider some of the geometrical aspects
of a Riemannian extension g∇ or of a deformed Riemannian extension g∇, , which provide a link
between the affine geometry of (M, ∇) and the neutral signature pseudo-Riemannian geometry of
T ∗ M. We shall investigate the spectral geometry of the Jacobi operator and of the skew-symmetric
curvature operator both on M and on T ∗ M. Here is a brief outline to Chapter 2. Section 2.1 provides
a brief summary of some of the relevant material we shall need in the chapter. Section 2.2 gives a
construction that yields a wide variety of affine Osserman and affine Ivanov–Petrova manifolds.
Section 2.3 deals with affine surfaces; the geometric properties of affine Osserman surfaces and
of affine Ivanov–Petrova surfaces are examined in some detail. Section 2.4 examines homogeneous
affine connections in the context of 2-dimensional geometry. Finally, Section 2.5 deals with the
higher-dimensional setting.

2.1 BASIC NOTATIONAL CONVENTIONS

In this section, we review the notational conventions and basic constructions which underlie Chap-
ter 2. We refer to Chapter 1 for further details.

SPECIAL METRICS
Let (M, ∇) be an affine manifold of dimension m̄; this means that ∇ is a torsion-free connection on
M. In local coordinates x = (x 1 , . . . , x m̄ ), the Christoffel symbols are defined by the identity:
∇∂xi ∂xj = ij k ∂xk where ij k = j i k .
We adopt the notation of Equation (1.1.d) and introduce dual fiber coordinates xi  on the cotangent
bundle by decomposing a 1-form as ω = xi  dx i . The Riemannian extension of Definition 1.32 is
the neutral signature metric on T ∗ M, which is given in local coordinates relative to the frame
{∂x1 , . . . , ∂xm̄ , ∂x 1 , . . . , ∂x m̄ } by:
 
−2xk  ij k (
x ) Idm̄
g∇ = .
Idm̄ 0
More generally, if  is a symmetric (0, 2)-tensor field on M, then the deformed Riemannian extension
g∇, is the metric of neutral signature on T ∗ M given by Equation (1.7.a):
54 2. RIEMANNIAN EXTENSIONS
 
x ) + ij (
−2xk  ij k ( x ) Idm̄
g∇, = .
Idm̄ 0
These metrics are invariantly defined and are independent of the particular coordinate system
(x 1 , . . . , x m̄ ) chosen on the base. We say that a neutral signature pseudo-Riemannian manifold
(N, gN ) is a Walker metric if, relative to some system of local coordinates, we have
 
B Idm̄
gN = for 2m̄ = dim(N) .
Idm̄ 0
Thus, in particular, if B is a polynomial of order at most 1 in the xi  variables, then g is locally a
deformed Riemannian extension; a deformed Riemannian extension is locally a Riemannian exten-
sion if B vanishes on the “zero-section”. In these two instances, the linear terms in the xi  variables
give the connection 1-form of a torsion-free connection on the base manifold.

NATURAL OPERATORS DEFINED BY THE CURVATURE

Let R be the curvature operator corresponding to the Levi-Civita connection of a pseudo-
Riemannian manifold (M, g) and let R be the associated curvature tensor as defined in Equa-
tion (1.2.b) and in Equation (1.2.d), respectively:
R(X, Y )Z := (∇X ∇Y − ∇Y ∇X − ∇[X,Y ] )Z and R(X, Y, Z, W ) := g(R(X, Y )Z, W ) .
The Jacobi operator of Equation (1.10.a) is described by:
J (X) : Y → R(Y, X)X and g(J (X)Y, Z) = R(Y, X, X, Z) .
We use the symmetries of the curvature tensor given in Equations (1.2.l)-(1.2.o) to see that J (X) is
self-adjoint since R(Y, X, X, Z) = R(Z, X, X, Y ). In the positive definite setting, this implies that
J (X) is diagonalizable, but this is not the case in the indefinite signature setting and is the reason
why these two contexts are so different. Similarly, let {x, y} be an oriented basis for a non-degenerate
2-plane π := Span{x, y}. The skew-symmetric curvature operator of Equation (1.11.b) is then given
by:
R(π ) := |g(x, x)g(y, y) − g(x, y)2 |−1/2 R(x, y) so
g(R(π )z, w) = |g(x, x)g(y, y) − g(x, y)2 |−1/2 R(x, y, z, w) .

OSSERMAN GEOMETRY
If J is a linear operator, then the spectrum Spec{J } is the set of (possibly complex) eigenvalues of J .
Let pλ (J ) := det{λ Id −J } be the characteristic polynomial of J ; λ belongs to Spec{J } if and only if
pλ (J ) = 0. One has the following useful observation; we shall omit the proof as it is an elementary
exercise in linear algebra using Jordan normal form.

Lemma 2.1 Let J be a linear map of an m-dimensional vector space V . The following assertions are
equivalent and, if any is satisfied, then J is said to be nilpotent.
 2.1. BASIC NOTATIONAL CONVENTIONS 55
1. Spec{J } = {0}.

2. J m = 0.

3. pλ (J ) = λm .

4. Tr{J k } = 0 for 1 ≤ k ≤ m.

Recall that a pseudo-Riemannian manifold (M, g) is said to be pseudo-Riemannian Osserman

(if we wish to emphasize the fact that we are in the pseudo-Riemannian setting) or simply Osserman
(if the context is clear), provided that Spec{J (·)} is constant on the pseudo-sphere bundle S + (M, g)
of unit spacelike tangent vectors if q > 0 or, equivalently, on the pseudo-sphere bundle S − (M, g)
of unit timelike tangent vectors if p > 0. This implies that the eigenvalue multiplicities are constant
as well and, consequently, (M, g) is Osserman if and only if the characteristic polynomial of J (X)
is constant on S + (M, g) or on S − (M, g) (the polynomials can be different). In the positive definite
setting, Osserman [173] wondered if such a manifold must either be flat or be locally isometric
to a rank-one symmetric space. This result was proved by Chi [53] if m ≡ 0 mod 4—see Gilkey,
Swann, and Vanhecke [114] for related work and additional references. Any Osserman Lorentzian
manifold has constant sectional curvature (see Blažić, Bokan, and Gilkey [16] and also García-
Río, Kupeli, and Vázquez-Abal [88]). However, the situation is completely different for metrics of
indefinite signature. If p ≥ 2 and if q ≥ 2, then there are Osserman pseudo-Riemannian manifolds
of signature (p, q) that are not locally symmetric (see, for example, García-Río, Vázquez-Abal, and
Vázquez-Lorenzo [91]).
Let (M, ∇) be an affine manifold of dimension m̄. We say that (M, ∇) is affine Osserman if
Spec{J (X)} = {0} for all X, or, equivalently, if J (X)m̄ = 0 for any tangent vector X. Let  be an
arbitrary symmetric (0, 2)-tensor field on M. We will show in Theorem 2.15 that (M, ∇) is affine
Osserman if and only if the deformed Riemannian extension (T ∗ M, g∇, ) is pseudo-Riemannian
Osserman.

IVANOV–PETROVA GEOMETRY
A pseudo-Riemannian manifold (M, g) is said to be Ivanov–Petrova at a point P of M if Spec{R(π)}
is constant on the Grassmannian Gr + u,v (TP M) of oriented 2-planes of signature (u, v); this notion
is independent of whether (u, v) is taken to be (2, 0) (if p ≥ 2), or (1, 1) (if p ≥ 1 and q ≥ 1), or
(0, 2) (if q ≥ 2). Metrics of constant curvature are Ivanov–Petrova, but there are Ivanov–Petrova
metrics that do not have constant sectional curvature (we refer to Gilkey [96], to Gilkey, Leahy,
and Sadofsky [101] and to Gilkey and Zhang [115]). For example, let M = I × N be a product
manifold where I is a subinterval of R and where dsN 2 is a metric of constant sectional curvature K

on a pseudo-Riemannian manifold N of dimension m − 1. Give M the metric

2
dsM := dt 2 + f (t)dsN
2
where f (t) := 21 (Kt 2 + At + B)  = 0 .
56 2. RIEMANNIAN EXTENSIONS
Then M is a pseudo-Riemannian Ivanov–Petrova manifold; if A2 − 4BK  = 0, then (M, g) does
not have constant sectional curvature. Conversely, if m ≥ 9, then any Riemannian Ivanov–Petrova
metric either has constant sectional curvature or is locally isometric to this example, where f (t) > 0
and N is Riemannian.
Ivanov–Petrova metrics which are 3-dimensional have been investigated in the Rieman-
nian and Lorentzian setting (see García-Río, Haji-Badali, and Vázquez-Lorenzo [87] and Ivanov
and Petrova [134]), where a complete algebraic description is available. A 3-dimensional pseudo-
Riemannian manifold is Ivanov–Petrova if and only if at each point P in M where the sectional
curvature is non-constant, either the Ricci operator is diagonalizable of rank one, or the Ricci oper-
ator is 2-step nilpotent (see [41, 87, 134, 161]). We say an affine manifold (M, ∇) of dimension m̄
is affine Ivanov–Petrova at a point P if

Spec{R(x, y)} = {0} or equivalently if R(x, y)m̄ = 0 for all x, y ∈ TP M .

Let  be an arbitrary symmetric (0, 2)-tensor field on M. We will show in Theorem 2.15
that (M, ∇) is affine Ivanov–Petrova if and only if the deformed Riemannian extension g∇, on
the cotangent bundle is pseudo-Riemannian Ivanov–Petrova. We shall characterize those affine
connections on surfaces that are affine Ivanov–Petrova in terms of the symmetry and degeneracy of
their Ricci tensor. A local description of recurrent Ivanov–Petrova connections on surfaces will be
given in Theorem 2.7.

HOMOGENEOUS AFFINE CONNECTIONS

The study of homogeneous affine connections on surfaces which are either Osserman or Ivanov–
Petrova is more involved and relies on a classification result by Opozda [172] (see also Kowalski,
Opozda, and Vlášek [143, 144]). Any homogeneous affine connection is either the Levi-Civita
connection of a metric of constant curvature or it belongs to one of two families A and B (see
Theorem 2.8). We shall consider the equivalence problem for such connections, showing that, in
addition to the flat ones, there is an explicit family of locally homogeneous affine connections that
is of both Types A and B. This family is contained in the class of projectively flat and recurrent
homogeneous connections with symmetric and degenerate Ricci tensor (see Theorem 2.13).

PROJECTIVELY AFFINE OSSERMAN MANIFOLDS AND PROJECTIVELY

AFFINE IVANOV–PETROVA MANIFOLDS
We have J (cx) = c2 J (x) and R(cx, cy) = c2 R(x, y). In the pseudo-Riemannian setting, we can
use the metric to normalize x to belong to S ± (M, g) or to normalize {x, y} to be an orthonormal set
and eliminate the effect of this rescaling. This renormalization is not available in the affine setting
and thus, it is natural to consider the setting in that Spec{J (x)} = {0} or Spec{R(x, y)} = {0}.
There is, however, another notion available in the affine setting that has not been studied extensively
in the literature. We say (M, ∇) is projectively affine Osserman at a point P ∈ M if, given any pair of
 2.2. EXAMPLES OF AFFINE OSSERMAN IVANOV–PETROVA MANIFOLDS 57
non-zero vectors x, y ∈ TP M, there is a non-zero complex number λ(x, y) = 0 so:

Spec{J (y)} = λ(x, y) · Spec{J (x)} = {0} .

We explicitly rule out the case where the spectrum solely consists of {0}, as that is the affine Os-
serman setting discussed above. Similarly, we say that (M, ∇) is projectively affine Ivanov–Petrova if,
given a quadruple of vectors x, y, x̄, ȳ ∈ TP M with {x, y} linearly independent and {x̄, ȳ} linearly
independent, then there exists λ(x, y, x̄, ȳ)  = 0 so:

Spec{R(x̄, ȳ)} = λ(x, y, x̄, ȳ) · Spec{R(x, y)}  = {0} .

If g is a positive definite Osserman metric that is not flat, then g ∇ is projectively affine
Osserman, but is not affine Osserman. Similarly, if g is a positive definite Ivanov–Petrova metric
that is not flat, then g ∇ is projectively affine Ivanov–Petrova, but is not affine Ivanov–Petrova.
Indefinite Osserman metrics are not projectively affine Osserman since Spec{J (x)} = 0 if x is a
null vector. Similarly, indefinite Ivanov–Petrova metrics are not projectively affine Ivanov–Petrova
as Spec{R(x, y)} = {0} if the induced metric on Span{x, y} is degenerate. We refer to Gilkey and
Nikčević [106] for further details.

Example 2.2 Let (x 1 , . . . , x m̄ ) be coordinates on Rm̄ for m̄ ≥ 2. Let M = (Rm̄ , ∇) be the affine
manifold defined by setting:

11 1 = 2, i1 1 = 1i i = 1, and ii 1 = 1 for i > 1 .

The non-zero components of the curvature tensor are given by

Rijj i = 1 and Rj ij i = −1 for i  = j .

This is the curvature tensor of a metric of constant sectional curvature 1 and hence, is projectively
Osserman and projectively Ivanov–Petrova. Since the Christoffel symbols are constant, the group
of translations acts transitively on M by affine isomorphisms; thus, M is affine homogeneous.
However, if we set γ (t) = (x(t), 0, . . . , 0), then the geodesic equation becomes ẍ = −2ẋx, which
blows up in finite time with initial conditions x(0) = −1 and ẋ(0) = −1. Thus, M is geodesically
incomplete. Finally, since ∇ R = 0, these manifolds are not locally symmetric. This shows that the
affine manifold M is not affinely equivalent to a manifold of constant sectional curvature.

2.2 EXAMPLES OF AFFINE OSSERMAN IVANOV–PETROVA

MANIFOLDS
Since much of our discussion will be concerned with affine Osserman and affine Ivanov–Petrova
manifolds, it is worth giving a construction that exhibits a wide family of such examples in a uniform
context (we refer to Gilkey [98] for more details).
58 2. RIEMANNIAN EXTENSIONS
Definition 2.3 Let (x 1 , . . . , x m̄ ) be the usual system of coordinates on M = Rm̄ . We say that an
affine manifold (M, ∇) is a generalized affine plane wave manifold if

∇∂xi ∂xj = ij k (x 1 , . . . , x k−1 )∂xk .
k>max(i,j )

Here, we adopt the convention that the empty sum is zero. For example, if m̄ = 2, then the possibly
non-zero Christoffel symbols could be 11 2 (x 1 ), while, if m̄ = 3, the possibly non-zero Christoffel
symbols could be:
{11 2 (x 1 ), 11 3 (x 1 , x 2 ), 12 3 (x 1 , x 2 ) = 21 3 (x 1 , x 2 ), 22 3 (x 1 , x 2 )}.

Theorem 2.4 Let (M, ∇) be a generalized affine plane wave manifold.

1. (M, ∇) is geodesically complete.
2. There is a unique geodesic joining any two points of M.
3. (M, ∇) is Ricci flat, affine Osserman, and affine Ivanov–Petrova.

Proof. Let (M, ∇) be a generalized affine plane wave manifold. The geodesic equation takes the
form: 
γ̈ k (t) + γ̇ i (t)γ̇ j (t)ij k (γ 1 , . . . , γ k−1 )(t) = 0 . (2.2.a)
i,j <k

Let P = be the initial position and let ξ = (ξ 1 , . . . , ξ m̄ ) be the initial velocity. To

(P 1 , . . . , P m̄ )
solve Equation (2.2.a) with γ (0) = P and γ̇ (0) = ξ , we define:

γ 1 (t) := P 1 + ξ 1 t,  
t s 
γ k (t) := P k + ξ k t − γ̇ i (r)γ̇ j (r)ij k (γ 1 , . . . , γ k−1 )(r)drds for k > 1 .
0 0 i,j <k

This establishes Assertion (1). Given P , Q ∈ M, there is a unique geodesic γ = γP ,Q with γ (0) = P
and γ (1) = Q where

ξ 1 = Q1 − P 1 ,
 1 s 
ξ k = Qk − P k + γ̇ i (r)γ̇ j (r)ij k (γ 1 , . . . , γ k−1 )(r)drds for k > 1 .
0 0 i,j <k

This establishes Assertion (2). We have by Equation (1.2.c) that:

Rij k  = ∂xi j k  (x 1 , . . . , x −1 ) − ∂xj ik  (x 1 , . . . , x −1 )

+ in  (x 1 , . . . , x −1 )j k n (x 1 , . . . , x n−1 )
− j n  (x 1 , . . . , x −1 )ik n (x 1 , . . . , x n−1 ) .
 2.2. EXAMPLES OF AFFINE OSSERMAN IVANOV–PETROVA MANIFOLDS 59
Suppose  ≤ k. Then j k = ik = 0 so ∂xi j k = ∂xj ik = 0. Furthermore, for either of the
   

quadratic terms to be non-zero, there must exist an index n with k < n and n < , which is not
possible. Consequently Rij k  = 0 if  ≤ k. Thus for arbitrary {ξ1 , ξ2 } we have:

R(ξ1 , ξ2 )∂xk ∈ Spank< {∂x } .

This shows that R is nilpotent so (M, ∇) is affine Ivanov–Petrova. Suppose that  ≤ i. Then

∂xi j k  (x 1 , . . . , x −1 ) = 0 and ∂xj ik  = ∂xj 0 = 0 .

We have in  = 0 for any n. For the other quadratic term to be non-zero, there must exist an index
n so i < n and n < , which is not possible. This shows Rij k  = 0 if  ≤ i; similarly Rij k  = 0 if
 ≤ j . As a result, (M, ∇) is Ricci flat. Furthermore, for any ξ , we have:

J (ξ ) : ∂xi → Spani< {∂x } .

This shows that J (ξ ) is nilpotent as well so (M, ∇) is affine Osserman. 2

The following special case is instructive:

Theorem 2.5 Let (x 1 , x 2 , x 3 ) be the usual system of coordinates on R3 and consider the affine connection
∇ on R3 , whose only non-trivial Christoffel symbol is 12 3 = 21 3 = x 1 . Then (R3 , ∇) is Ricci flat,
locally symmetric, affine Osserman, affine Ivanov–Petrova, and not flat.

Proof. We apply Theorem 2.4. Since (R3 , ∇) is a generalized affine plane wave manifold, it is Ricci
flat, affine Osserman, and affine Ivanov–Petrova. There are no non-zero quadratic terms in the
expression
Rij a b = ∂xi j a b − ∂xj ia b + ic b j a c − j c b ia c ,

which is given in Equation (1.2.c), and the only non-zero component of the curvature is

R121 3 = −R211 3 = ∂x1 21 3 = 1 .

Consequently (R3 , ∇) is not flat. Furthermore, when considering the expression

∇ R(X1 , X2 ; X4 )X3 = ∇X4 R(X1 , X2 )X3 − R(∇X4 X1 , X2 )X3

−R(X1 , ∇X4 X2 )X3 − R(X1 , X2 )∇X4 X3 ,

which is given in Equation (1.2.g), the terms in ∇X4 Xj vanish, since ∇X4 Xj is a multiple of ∂x3 and
R(Y1 , Y2 )Y3 = 0 if any of the Yi ∈ Span{∂x3 }. Similarly since ∇∂x3 = 0, ∇X4 R(X1 , X2 )X3 = 0 as
well. Thus ∇ R = 0 as desired. 2
60 2. RIEMANNIAN EXTENSIONS
2.3 THE SPECTRAL GEOMETRY OF THE CURVATURE
TENSOR OF AFFINE SURFACES
In this section, we examine the geometry of affine surfaces (, ∇). We first study the Jacobi operator
and then we study the skew-symmetric curvature operator. If m̄ = 2, then

ρ11 = R111 1 + R211 2 = −R121 2 , ρ12 = R112 1 + R212 2 = −R122 2 ,

(2.3.a)
ρ21 = R121 1 + R221 2 = R121 1 , ρ22 = R122 1 + R222 2 = R122 1 ,
 
ρ21 ρ22
R(e1 , e2 ) = , det{λ Id −R} = λ2 + λ(ρ21 − ρ12 ) + det{ρ} .
−ρ11 −ρ12

THE JACOBI OPERATOR

Let ρa and ρs be the alternating and symmetric Ricci tensors of an affine connection ∇. Affine
connections with skew-symmetric Ricci tensor have been studied extensively in the literature [6, 15,
66, 142, 143, 213].

Theorem 2.6 Let P be a point of a connected 2-dimensional affine surface (, ∇).

1. (, ∇) is affine Osserman at P if and only if ρs = 0.

2. (, ∇) is projectively affine Osserman at P if and only ρs is definite.

3. If (, ∇) is locally symmetric and affine Osserman, then ∇ is flat.

4. Let (, ∇) be affine Osserman. Suppose that ρ = 0 at P ∈ . Then there exist local coordi-
nates (x 1 , x 2 ) defined near P and there exists a smooth function θ that is defined near P with
∂x1 ∂x2 θ(P )  = 0 so that the (possibly) non-zero covariant derivatives are ∇∂x1 ∂x1 = −(∂x1 θ )∂x1
and ∇∂x2 ∂x2 = (∂x2 θ)∂x2 .

5. Let (, ∇) be a compact affine surface with skew-symmetric Ricci tensor. Then ρ is exact and  is
diffeomorphic to the Klein bottle or to the torus.

Proof. By definition ρ(x, x) = Tr{J (x)}. Because J (x)x = R(x, x)x = 0 and because the dimen-
sion is two, one of the following two possibilities pertains:

1. Spec{J (x)} = {0} and ρ(x, x) = 0.

2. Spec{J (x)} = {0, λ(x)} for 0  = λ(x) ∈ R and ρ(x, x) = λ(x)  = 0.

Assertion (1) now follows as ρ(x, x) = 0 for all x if and only if ρ is skew-symmetric. We have
ρ is definite if and only if ρ(x, x) = 0 for x  = 0 or, equivalently, Spec{J (x)}  = {0} for x = 0.
Assertion (2) now follows.
 2.3. THE SPECTRAL GEOMETRY OF AFFINE SURFACES 61
Let (, ∇) be a connected affine surface with ∇ R = 0. Because the dimension is two, either
Rank{ρa } = 0 or Rank{ρa } = 2. Since ∇ R = 0, we have ∇ρa = 0 and thus, Rank{ρa } is constant.
Suppose that Rank{ρa } = 2 at each point. Then ρa defines a parallel volume form. By Theorem 1.7,
ρ is symmetric, which contradicts the assumption Rank{ρa } = 2. Consequently, Rank{ρa } = 0 so
ρ is symmetric. But, since (, ∇) is affine Osserman, ρ is skew-symmetric and ρ = 0. Assertion (3)
now follows from Equation (2.3.a).
We now prove Assertion (4). Let (, ∇) be an affine Osserman surface. Let ρ  = 0 near P .
Then ρ is skew-symmetric and defines a volume form in a neighborhood U of P . Therefore ρ is
recurrent. The work of Blažić and Bokan [15] shows that ∇ has recurrent curvature tensor. The work
of Wong [213] then implies that there is a system of coordinates (x 1 , x 2 ) defined near P so that
one of the following three possibilities holds:

1. There is a smooth function θ with ∂x1 ∂x2 θ = 0 so that

∇∂x1 ∂x1 = −(∂x1 θ )∂x1 and ∇∂x2 ∂x2 = (∂x2 θ )∂x2 .

2. There is a smooth function ϕ with ∂x1 ∂x2 ln ϕ  = 0 so that

∇∂x1 ∂x1 = −(∂x1 ln ϕ)∂x1 and ∇∂x2 ∂x2 = ϕ∂x1 + (∂x2 ln ϕ)∂x2 .

3. There is a smooth function ψ with ∂x1 ∂x2 ln ψ = 0 so that

x2
∇∂x1 ∂x1 = −∂x1 ln ψ + 1+x 1 x 2
∂ x1 + 1
ψ(1+x 1 x 2 )
∂x2 ,

ψ x1
∇∂x2 ∂x2 = − 1+x 1 x 2
∂x1 + ∂x2 ln ψ + 1+x 1 x 2
∂x2 .

A simplification of Wong’s result [213] was obtained by Derdzinski [66], who showed that one can
eliminate the final two cases. This completes the proof of Assertion (4); we refer to Derdzinski [66]
for the proof of the final assertion. 2
We note that Theorem 2.5 shows that Theorem 2.6 (3) fails if m̄ > 2. We also note that there
exist examples of non-flat locally symmetric Osserman pseudo-Riemannian manifolds on R4 with
metric of signature (2, 2) (we refer to García-Río and Vázquez-Lorenzo [92]).

THE SKEW-SYMMETRIC CURVATURE OPERATOR

Theorem 2.6 (1) generalizes to this setting to give an appropriate characterization of affine Ivanov–
Petrova surfaces in terms of the Ricci tensor.

Theorem 2.7 Let (, ∇) be a connected 2-dimensional affine surface.

1. (, ∇) is affine Ivanov–Petrova at P if and only if ρ is symmetric and degenerate.

2. Let (, ∇) have recurrent curvature.

62 2. RIEMANNIAN EXTENSIONS
(a) (, ∇) is affine Ivanov–Petrova if and only if there exist coordinate systems centered
at an arbitrary point P ∈  so that the only (possibly) non-zero Christoffel symbol is
11 2 = a(x 1 , x 2 ).
(b) If (, ∇) is affine Ivanov–Petrova, then (, ∇) is locally symmetric if and only if
a(x 1 , x 2 ) = α x 2 + ξ(x 1 ) for α ∈ R. In this setting, (, ∇) is projectively flat; (, ∇)
is flat if and only if α = 0.

Proof. Let {e1 , e2 } be linearly independent vectors in TP ; since the dimension is two, they form
a basis. We apply Equation (2.3.a) to see that R is nilpotent if and only if ρ12 − ρ21 = 0 (or,
equivalently, if ρ is symmetric) and det{ρ} = 0 (or, equivalently, if ρ is degenerate). Assertion (1)
follows.
We use Assertion (1) to see that (, ∇) is affine Ivanov–Petrova if and only if ρa = 0 and
det{ρs } = 0. Since the curvature is assumed to be recurrent, the work of Wong [213] shows there
exist local coordinates (x 1 , x 2 ) so that 11 2 = a(x 1 , x 2 )∂x2 and so the other Christoffel symbols
vanish; the assumption (, ∇) is not flat then yields ∂x2 a(x 1 , x 2 )  = 0. We check that the only
non-vanishing component of the Ricci tensor is ρ(∂x1 , ∂x1 ) = ∂x2 a(x 1 , x 2 ) and (, ∇) is locally
symmetric if and only if a(x 1 , x 2 ) = α x 2 + ξ(x 1 ); Assertion (2) now follows. 2

2.4 HOMOGENEOUS 2-DIMENSIONAL AFFINE SURFACES

An affine manifold (M, ∇) is said to be locally homogeneous if, for any two points P , Q ∈ M, there
exists a local diffeomorphism ϕ from a neighborhood of P to a neighborhood of Q so that ϕ ∗ ∇ = ∇.
We refer to Opozda [172] for the proof of Assertion (2) and to Arias-Marco and Kowalski [9] for
the proof of Assertions (1,3,4) in the following result:

Theorem 2.8 Let (, ∇) be an affine surface.

1. (, ∇) is locally homogeneous if and only if it admits, in a neighborhood of each point of , at least
two linearly independent affine-Killing vector fields.

2. If (, ∇) is locally homogeneous, then either ∇ is the Levi-Civita connection of a metric of constant
curvature, or there exist local coordinates (x 1 , x 2 ) and suitably chosen constants {a, b, c, d, e, f } so
that the connection has one of the following forms:

Type A: ∇∂x1 ∂x1 = a∂x1 + b∂x2 , ∇∂x1 ∂x2 = c∂x1 + d∂x2 , ∇∂x2 ∂x2 = e∂x1 + f ∂x2 .

a∂x1 + b∂x2 c∂x1 + d∂x2 e∂x1 + f ∂x2

Type B: ∇∂x1 ∂x1 = , ∇∂x1 ∂x2 = , ∇∂x2 ∂x2 = .
x1 x1 x1
3. (, ∇) is of Type A if and only if it locally admits a pair of linearly independent affine-Killing
vector fields X, Y such that [X, Y ] = 0.
 2.4. HOMOGENEOUS 2-DIMENSIONAL AFFINE SURFACES 63
4. (, ∇) is of Type B if and only if it locally admits a pair of linearly independent affine-Killing
vector fields X, Y such that [X, Y ] = X.

We emphasize that there are connections which are both Type A and Type B; we shall discuss
the intersection of these classes presently in Theorem 2.12 and in Theorem 2.13. We must first
examine some geometric properties of Type A and Type B locally homogeneous connections. Of
particular interest are those that are projectively flat and recurrent. Recall that an equiaffine surface
is projectively flat if and only if the Ricci tensor is a Codazzi tensor, or, equivalently, if we have that
(∇X ρ)(Y, Z) = (∇Y ρ)(X, Z).

TYPE A LOCALLY HOMOGENEOUS AFFINE SURFACES

Recall that we have the local coordinate representation:

∇∂x1 ∂x1 = a∂x1 + b∂x2 , ∇∂x1 ∂x2 = c∂x1 + d∂x2 , ∇∂x2 ∂x2 = e∂x1 + f ∂x2 .

We summarize the geometric properties of such a connection as follows:

Theorem 2.9 Let (, ∇) be a Type A locally homogeneous affine surface. Then the Ricci tensor is
symmetric and given by:
ρ11 = −d 2 + ad + (f − c)b, ρ12 = ρ21 = cd − eb, ρ22 = −c2 + f c + (a − d)e.
Either the Ricci tensor defines a flat metric on  or (, ∇) is affine Ivanov–Petrova. The covariant
derivatives of the Ricci tensor take the form:
1
2 (∇∂x1 ρ)(∂x1 , ∂x1 ) = −da 2 + (d 2 − bf + cb)a + (be − cd)b,
1
2 (∇∂x2 ρ)(∂x1 , ∂x1 ) = 21 (∇∂x1 ρ)(∂x1 , ∂x2 ) = −acd + (c2 − f c + de)b,
1
2 (∇∂x2 ρ)(∂x1 , ∂x2 ) = 21 (∇∂x1 ρ)(∂x2 , ∂x2 ) = bce − (ae + cf − de)d,
1
2 (∇∂x2 ρ)(∂x2 , ∂x2 ) = f c2 − (de + f 2 )c − (af − be − df )e.
Consequently, (, ∇) is projectively flat. Finally, (, ∇) is affine Ivanov–Petrova if and only if it is
recurrent.

Proof. The computation of ρ and of ∇ρ follows directly from the defining property; the fact that
either ρ defines a flat metric or (, ∇) is affine Ivanov–Petrova now follows, as does the fact that
(, ∇) is projectively flat. We argue as follows to establish the remaining assertion. Suppose first that
(, ∇) is affine Ivanov–Petrova or, equivalently, by Theorem 2.7, that the Ricci tensor is symmetric
and degenerate. Note that ρ is degenerate if and only if
 
b2 e2 − d 3 − 2ad 2 + (a 2 + 3bc − bf )d + (f − c)ab e
  (2.4.a)
+ f d 2 + a(c − f )d − b(c − f )2 c = 0.
64 2. RIEMANNIAN EXTENSIONS
We examine the possibilities seriatim to show (, ∇) is recurrent. We will show either that (, ∇)
is locally symmetric (in which case ∇ρ = 0) or we will construct ω so ∇ρ = ω ⊗ ρ.

1. If b = 0, then we obtain several sub-cases since Equation (2.4.a) reduces to

 
d · ac2 − (a − d)f c − (a − d)2 e = 0 . (2.4.b)

(a) If d = 0 then ∇ρ = ω ⊗ ρ, with ω = (−2f )dx 2 .

(b) If d  = 0 and if a = 0, then e = d −1 cf and ∇ρ = 0.
(c) If d = 0 and if a = 0, then Equation (2.4.b) shows that

ac2 − (a − d)f c − (a − d)2 e = 0, so,

1
c = (2a)−1 (a − d)(f + ε(f 2 + 4ae) 2 ),

with ε ∈ {−1, 1} (c being real). If d = a, then the connection is symmetric and

∇ρ = 0 so (, ∇) is recurrent. On the other hand, if d = a, then ∇ρ = ω ⊗ ρ, where
1
ω = (−2a)dx 1 − (f + ε(f 2 + 4ae) 2 )dx 2 .

2. If b = 0, then Equation (2.4.a) yields:

 1

e = (2b2 )−1 d 3 − 2ad 2 + (a 2 + 3bc − bf )d + (f − c)ab + ε(d 2 − ad + (c − f )b)ζ 2 ,

with ζ = (a − d)2 + 4bc and ε ∈ {−1, 1}, as e is real. Now, if c = b−1 (−d 2 + ad + bf ), then
∇ρ = 0. Otherwise, ∇ρ = ω ⊗ ρ with
1 1
ω = −(a + d + εζ 2 ) dx 1 − b−1 (d 2 − ad + 2bc + εdζ 2 ) dx 2 .

Conversely, suppose that (, ∇) is recurrent. First, observe that, if ∇ is locally symmetric, then
the vanishing of the covariant derivative of the Ricci tensor shows that Equation (2.4.a) vanishes.
Hence, the Ricci tensor is symmetric and degenerate and thus, (, ∇) is affine Ivanov–Petrova.
Next, if (, ∇) is recurrent but non-symmetric, then there exists a 1-form ω = ω1 dx 1 + ω2 dx 2
such that ∇ρ = ω ⊗ ρ, where at least one of ω1 or ω2 is non-zero. Since any Type A connection is
projectively flat, one has that

ω1 ρ22 = ω2 ρ12 and ω2 ρ11 = ω1 ρ12 .

We may assume that ω1  = 0, as the case ω2  = 0 is analogous. Identities derived previously permit
us to conclude that (, ∇) is affine Ivanov–Petrova since

det ρ = ρ11 ρ22 − ρ12

2 = ω2
ω1 ρ11 ρ12 − ρ12
2 = 0. 2
 2.4. HOMOGENEOUS 2-DIMENSIONAL AFFINE SURFACES 65
TYPE B LOCALLY HOMOGENEOUS AFFINE SURFACES
Recall that we have the local coordinate representation:

a∂x1 + b∂x2 c∂x1 + d∂x2 e∂x1 + f ∂x2

∇∂x1 ∂x1 = , ∇∂x1 ∂x2 = , ∇∂x2 ∂x2 = .
x1 x1 x1
We now summarize the relevant facts we shall need:

Theorem 2.10 Let (, ∇) be a Type B locally homogeneous equiaffine surface. Then

ρ11 = 1
(x 1 )2
{(a − d + 1)d + (f − c)b}, ρ12 = 1
(x 1 )2
{cd − be + f },
ρ21 = 1
(x 1 )2
{cd − be − c}, ρ22 = 1
(x 1 )2
{(a − d − 1)e + (f − c)c}.

Either ρs defines a metric of constant Gauss curvature on , or (, ∇) is affine Ivanov–Petrova. Moreover,

2 (x ) (∇∂x1 ρ)(∂x1 , ∂x1 ) = (a + 1)(d − a − 1)d + (2a − d + 3)bc + b2 e,

1 1 3

2 (x ) (∇∂x2 ρ)(∂x1 , ∂x1 ) = 2bc − adc + bde,

1 1 3 2

(x 1 )3 (∇∂x1 ρ)(∂x1 , ∂x2 ) = (a + 4bc − 2(a + 1)d + 2)c + (2d + 3)be,

1 1 3
2 (x ) (∇∂x2 ρ)(∂x1 , ∂x2 ) = c2 d + bec + (d − a)de,
1 1 3
2 (x ) (∇∂x1 ρ)(∂x2 , ∂x2 ) = (d + 1)(d − a + 1)e + (d + 3)c2 + bce,
1 1 3
2 (x ) (∇∂x2 ρ)(∂x2 , ∂x2 ) = −2c3 + be2 + (a − 2d)ce.
In this setting (, ∇) is projectively flat if and only if
   3c2 +2de+e −c3 −ce

e = f = c = 0 or e  = 0, f = −c, a = e ,b = e2
.

Proof. One computes the Ricci tensor directly; this shows that ρ need not be symmetric and thus,
∇ need not be equiaffine; ρ is symmetric if and only if f = −c. Suppose that ρs defines a metric on
, then the associated Gauss curvature satisfies

K = −(x 1 )−2 ρ22 det{ρ}−1

= 2c2 + e(d − a + 1)
−1
× c(4bc2 +(d 2 −2ad −1)c+2(2d −a)be)+e(((a −d)2 −1)d −b2 e)

and K is constant. One computes ∇ρ directly. It then follows that (, ∇) is projectively flat if and
only if
c(a−2d+2)+3be
0 = (∇∂x1 ρ)(∂x2 , ∂x1 ) − (∇∂x2 ρ)(∂x1 , ∂x1 ) = (x 1 )3
,
 2 
2 3c −ae+2de+e
0 = (∇∂x1 ρ)(∂x2 , ∂x2 ) − (∇∂x2 ρ)(∂x1 , ∂x2 ) = (x 1 )3
.
66 2. RIEMANNIAN EXTENSIONS
The final implication now follows. 2
We note that, in general, an affine manifold (M, ∇) is projectively flat if and only if
(T ∗ M, g∇ ) is locally conformally flat, which is equivalent to the vanishing of the associated Weyl
conformal curvature tensor (see, for example, Afifi [3]). Hence, if a Type B locally homogeneous
affine surface is projectively flat, then the associated Ricci tensor is necessarily symmetric, since
c+f
W (∂x1 , ∂x2 , ∂x1 , ∂x 1 ) = 2(x 1 )2 , which shows that f = −c, the necessary and sufficient condition for
ρ to be symmetric.
Analyzing which affine connections are recurrent is more involved, but it shows a clear
link with the Ivanov–Petrova property. First of all, note that Type A locally homogeneous affine
connections need not be recurrent in general. Set b = c = 0 in Theorem 2.8 to get ρ12 = 0, but
(∇∂x2 ρ)(∂x1 , ∂x2 ) = 2de(d − a), which shows that the connection is not recurrent in general.
We now determine which Type B locally homogeneous affine surfaces are Ivanov–Petrova.
Unlike the Type A setting, these surfaces are not necessarily recurrent.

Theorem 2.11 Let (, ∇) be a Type B locally homogeneous affine surface. Then (, ∇) is affine
Ivanov–Petrova if and only if either (, ∇) is recurrent or if the Christoffel symbols in Theorem 2.8 are
given by

∇∂x1 ∂x1 = a
∂ +b∂ ,
x 1 x1 x 1 x 2
∇∂x1 ∂x2 = c
∂ +d∂ ,
x 1 x1 x 1 x 2
∇∂x2 ∂x2 = e
∂
x 1 x1
− xc1 ∂x2 ,

for real constants a, b, c, d, and e belonging to one of the following eight cases:

1. b = 0, d  = 0, e = 0, c  = 0, a = (2d)−1 (d 2 − 1).

2. b = 0, de  = 0, c = 0, a = d ± 1.
1
3. b = 0, de = 0, c = 0, e  = −d −1 c2 , a = (de)−1 (d(c2 + de) ± ζ 2 ), with
ζ = d(c2 d + e)(c2 + de) ≥ 0.
 1
4. b  = 0, d = 0, c  = 0, e = (b2 )−1 (−abc ± b2 c2 (a 2 + 4bc − 1) 2 with a 2 + 4bc − 1 ≥ 0.

5. b  = 0, d  = 0, a  = ±(d − 1), c = (2b)−1 (a − d + 1)d, e = (2b2 )−1 (a − d + 1)(d − 1)d.

/ {0, b−1 ad, (2b)−1 (a − d + 1)d},

6. b  = 0, d  = 0, c ∈
1
e = (2b2 )−1 ((d − a)2 + 4bc − 1)d − 2abc ± ζ 2 ,
with ζ = ((d − a + 1)d + 2bc)((d − a − 1)d + 2bc)((d − a)2 + 4bc − 1) ≥ 0.

7. b  = 0, d  = 0, c = 0, a  = d − 1,
e = (2b2 )−1 (d 3 − 2ad 2 + (a 2 − 1)d ± |d||(a − d)2 − 1|)  = 0.

8. b  = 0, d  = 0, c = ad
b ,a = 1 − d,
 2.4. HOMOGENEOUS 2-DIMENSIONAL AFFINE SURFACES 67
e= (2b2 )−1 (d 3 + 2ad 2 − (a 2 + 1)d ± |d||(a + d)2 − 1|)  = −(b2 )−1 (a 2 d).

Proof. Assume that (, ∇) is affine Ivanov–Petrova. We suppose f = −c henceforth.Theorem 2.10

implies that ρ is symmetric if and only if f = −c. Thus, ρ is degenerate if and only if
   
b2 e2 − d 3 − 2ad 2 + (a 2 + 4bc − 1)d − 2abc e − d 2 − 2ad + 4bc − 1 c2 = 0. (2.4.c)

We assume that ∇ρ = ω1 dx 1 + ω2 dx 2 and examine the implications of this identity. We

proceed as in the proof Theorem 2.9 using Theorem 2.10 to examine the solutions of Equation (2.4.c).
We first observe that if b = 0, then Equation (2.4.c) becomes:
dea 2 − 2d(c2 + de)a + (d 2 − 1)(c2 + de) = 0 .
This leads to five cases which we examine seriatim:
1. Suppose d = 0. Then c = 0 so ∇ρ = ω ⊗ ρ with ω = − x21 dx 1 .

2. Suppose d = 0, e = 0, and c = 0. Then ∇ρ = ω ⊗ ρ, with ω = − 2+2a

x1
dx 1 .

3. Suppose d  = 0, e = 0, and c  = 0.Then a = (2d)−1 (d 2 − 1) and (, ∇) is not recurrent—this

leads to Case 1. If ∇ρ = ω ⊗ ρ, then
2c2 (x 1 ω2 −2c)
(∇∂x2 ρ)(∂x2 , ∂x2 ) − ω2 ρ22 = (x 1 )3
,
2c2
so ω2 = 2c
x1
and (∇∂x2 ρ)(∂x1 , ∂x2 ) − ω2 ρ12 = (x 1 )3
 = 0 and ∇ is not recurrent.
1
4. Suppose d  = 0 and 0  = e. Then a = (de)−1 (d(c2 + de) + εζ 2 ) with ε = ±1 and with
ζ = d(c2 d + e)(c2 + de) ≥ 0. We suppose ∇ρ = ω ⊗ ρ, where ω = ω1 dx 1 + ω2 dx 2 and
examine the implications. If c = 0, then ζ = d 2 e2 ≥ 0 and a = d ± 1. Consequently,
0 = (∇∂x2 ρ)(∂x1 , ∂x2 ) − ω2 ρ12 = −2ε|d||e|(x 1 )−3  = 0 ,
which gives Case 2. If c  = 0, then

0 = (x 1 )3 (∇∂x1 ρ)(∂x1 , ∂x2 ) − ω1 ρ12

1
c3 d(1−2d)−cde(2d 2 +d−2)−εc(2d−1)ζ 2
= de − c(d − 1)x 1 ω1 , (2.4.d)
1
0 = (x 1 )3 (∇∂x2 ρ)(∂x1 , ∂x2 ) − ω2 ρ12 = −2εζ − c(d − 1)x ω2 .
2 1

Assuming d = 1, the second equation above reduces to

−2ε|c2 + e|
(∇∂x2 ρ)(∂x1 , ∂x2 ) − ω2 ρ12 = ,
(x 1 )3
hence, ∇ is not recurrent if e  = −c2 , thus showing Case 3 (for d = 1). Note that, for e = −c2 ,
we have ∇ρ = 0. Next, if d  = 1, then ω1 and ω2 are determined by Equation (2.4.d) as follows
68 2. RIEMANNIAN EXTENSIONS
1
d(x 1 )3 (∇∂x1 ρ)(∂x2 , ∂x2 ) − ω1 ρ22 = c2 + de − εζ 2 ,

1
1
2 c(d − 1)e(x 1 )3 (∇∂x2 ρ)(∂x1 , ∂x1 ) − ω2 ρ11 = (c2 + de)(d(c2 + e) + εζ 2 ),

1
1
2 c(d − 1)d(x 1 )3 (∇∂x2 ρ)(∂x2 , ∂x2 ) − ω2 ρ22 = (c2 + de)(d(c2 + e) − εζ 2 ).

2
Note that the three expressions above do not vanish simultaneously for e = − cd and hence,
in such a case, the affine connection is not recurrent (Case 3 with d  = 1). If e = d −1 c2 then,
setting ω = − x21 dx 1 , it follows that ∇ρ = ω ⊗ ρ holds.

5. Suppose b  = 0. Let ε = ±1. By Equation (2.4.c), we have that:

 1

e = (2b2 )−1 ((d − a)2 + 4bc − 1)d − 2abc + εζ 2 with
ζ = ((d − a + 1)d + 2bc)((d − a − 1)d + 2bc)((d − a)2 + 4bc − 1) ≥ 0 .

Assume ∇ρ = ω ⊗ ρ. If d = 0, then

(2b)−1 (x 1 )3 (∇∂x2 ρ)(∂x1 , ∂x1 ) − ω2 ρ11 = c(2c + x 1 ω2 ).

Hence, for c  = 0, it follows that ω2 = − x2c1 and we get

(∇∂x2 ρ)(∂x1 , ∂x2 ) − ω2 ρ12 = −(x 1 )−3 2c2 ,

which does not vanish (Case 4). If d  = 0, then:

(x 1 )3 (∇∂x1 ρ)(∂x1 , ∂x1 ) − ω1 ρ11

1
= (a + d + 3)((d − a − 1)d + 2bc) + εζ 2 + ((d − a − 1)d + 2bc)x 1 ω1 ,

(x 1 )3 (∇∂x2 ρ)(∂x1 , ∂x1 ) − ω2 ρ11

 1

= b−1 ((d − a − 1)d + 2bc)((d − a + 1)d + 2bc) + dεζ 2

+((d − a − 1)d + 2bc)x 1 ω2 .

Now, if (d − a − 1)d + 2bc = 0 or, equivalently, if c = (2b)−1 (a − d + 1)d, then the expres-
sions above vanish and

(∇∂x1 ρ)(∂x1 , ∂x2 ) − ω1 ρ12 = (2b(x 1 )3 )−1 (a 2 − (d − 1)2 )d .

 2.4. HOMOGENEOUS 2-DIMENSIONAL AFFINE SURFACES 69
It follows then that, for a = ±(d − 1), the affine connection is not recurrent (Case 5), while
for a = ±(d − 1), one has ∇ρ = 0. On the other hand, if c  = (2b)−1 (a − d + 1)d, then ω1
and ω2 are determined by the expressions above and one may compute that:

(2b)−1 d((∇∂x1 ρ)(∂x1 , ∂x2 ) − ω1 ρ12 ) − ((∇∂x2 ρ)(∂x1 , ∂x2 ) − ω2 ρ12 )

= (b(x 1 )3 )−1 2c(bc − ad) .

Therefore, if c(bc − ad)  = 0 then the affine connection is not recurrent (Case 6). If we have
c = 0, then c  = (2b)−1 (a − d + 1)d.This implies that a  = d − 1. Furthermore, ∇ρ = ω ⊗ ρ
if and only if e = 0 (Case 7). If bc − ad = 0, then c = (2b)−1 (a − d + 1)d. Thus, a = 1 − d.
This shows the connection is recurrent if and only if e = −(b2 )−1 a 2 d, which yields Case 8.
Conversely, any equiaffine homogeneous connection of Type B that is recurrent, or given by the
conditions above, is affine Ivanov–Petrova (see Calviño-Louzao, García-Río, and Vázquez-Lorenzo
[45]). 2

AFFINE SURFACES WHICH ARE BOTH TYPE A AND TYPE B

Theorem 2.12 A non-flat Type B affine connection is affinely equivalent to a Type A affine connection
if and only if ∇∂x1 ∂x1 = x11 (a∂x1 + b∂x2 ), ∇∂x1 ∂x2 = x11 d∂x2 , and ∇∂x2 ∂x2 = 0.

Proof. We apply Theorem 2.9 and Theorem 2.10. First of all, recall that any Type A affine connection
is projectively flat. Moreover, the Ricci tensor is always symmetric and defines a flat metric on the
surface or, otherwise, it is degenerate. Further note that, in the later case, the Ricci tensor is always
recurrent. Observe that the projectively flat Type B locally homogeneous connections are listed in
Theorem 2.10:

e=f =c=0 (2.4.e)

3c2 +2de+e −c3 −ce
e  = 0, f = −c, a= e , b= e2
. (2.4.f )

In the situation given in Equation (2.4.f), if the Ricci tensor is non-degenerate, then it defines a
metric of non-zero curvature and thus, no such connection can be affinely equivalent to any of Type
A. Moreover, if the Ricci tensor is degenerate but non-zero, then we show that such a connection
cannot be projectively flat and recurrent simultaneously, and hence, they cannot be affinely equivalent
to any Type A connection.
First assume that Equation (2.4.f) pertains with c = 0. Then the Ricci tensor is degenerate
for d = 0 or d = −2. In the case d = 0, the connection is flat, while, if d = −2, the connection is
not recurrent. Next assume that Equation (2.4.f) holds with c  = 0. In this case, the Ricci tensor is
2 2 2
degenerate if and only if d = − ce or d = − ce − 2. If d = − ce , then the connection is flat. Hence,
2
set d = − ce − 2. If b = 0, we have the conditions a = −4, c  = 0, d = −1 and e = −c2 .This shows
that the connection is not recurrent. Thus, in any case the connection is not recurrent.
70 2. RIEMANNIAN EXTENSIONS
Next we show that any connection of Type B given by Equation (2.4.e) is affinely equivalent
to a Type A connection. In doing that, we recall the discussion in Arias-Marco and Kowalski [9]
and in Kowalski, Opozda, and Vlášek [144]: Type A locally homogeneous affine connections are
characterized by the existence of linearly independent commuting affine-Killing vector fields.
Recall that a vector field X is said to be affine-Killing if
[X, ∇Y Z] − ∇Y [X, Z] − ∇[X,Y ] Z = 0 for all Y, Z .
Thus, a vector field X = A(x 1 , x 2 )∂x1 + B(x 1 , x 2 )∂x2 on the coordinate domain U (x 1 , x 2 ) is affine-
Killing for the connection in Equation (2.4.e) if and only if
a b a a−d
A11 + 1 A1 − 1 A2 − 1 2 A = 0, A12 + A2 = 0, A22 = 0,
x x (x ) x1
2b 2d − a b b
B11 + 1 A1 + 1
B1 − 1 B2 − 1 2 A = 0,
x x x (x )
d b d 2d
B12 + 1 A1 + 1 A2 − 1 2 A = 0, B22 + 1 A2 = 0 ,
x x (x ) x
where the subscripts mean Aij = ∂xi ∂xj A(x 1 , x 2 ) (similarly for the function B). Set

x 1 ∂x 2 if a − 2d = 0
X= and
∂x2 if a − 2d = 0
x ∂x1 + (x + x )∂x2
1 1 2 if a − 2d = 0
Y = .
x 1 ∂x1 + a−2d
b
x 1 ∂x 2 if a − 2d = 0
Then {X, Y } are linearly independent commuting affine-Killing vector fields, and hence, the con-
nection is of Type A. 2
Observe the results established above show that the locally homogeneous connections given
in Theorem 2.12 are projectively flat and recurrent with symmetric and degenerate Ricci tensor.
However, not every projectively flat and recurrent locally homogeneous affine connection with sym-
metric and degenerate Ricci tensor is necessarily of Type B, as the following example shows; we refer
to Calviño-Louzao, García-Río, and Vázquez-Lorenzo [45] for further details.

Theorem 2.13 Let (, ∇) be an affine surface with symmetric and degenerate Ricci tensor, which
is recurrent and projectively flat. Then (, ∇) is locally homogeneous if and only if around each
point there exists a coordinate system (x 1 , x 2 ) in which the non-zero component of ∇ is given by
μ
∇∂x1 ∂x1 = x 2 (α+κx 1 )2 ∂x2 for some constants μ, α and κ. Moreover, any such a connection is locally

homogeneous of both Type A and Type B if and only if κ 2 − 4μ ≥ 0.

A connection of the form given in Theorem 2.13 is flat if and only if μ = 0 and locally sym-
metric if and only if κ = 0. Hence, a projectively flat locally symmetric connection with symmetric
and degenerate Ricci tensor is of Type B if and only if the only non-zero Christoffel symbol is
11 2 = Kx 2 with K ≤ 0.
 2.5. SPECTRAL GEOMETRY 71
2.5 THE SPECTRAL GEOMETRY OF THE CURVATURE
TENSOR OF DEFORMED RIEMANNIAN EXTENSIONS

Lemma 2.14 Let (M, ∇) be an affine manifold, let ∇˜ ∇, be the Levi-Civita connection of the deformed
Riemannian extension (T ∗ M, g∇, ), and let σ : T ∗ M → M be the canonical projection. Let {X̃, Ỹ } be
tangent vectors on T ∗ M and let {X := σ∗ X̃, Y := σ∗ Ỹ } be the corresponding tangent vectors on M. Then

Spec{J˜ (X̃)} = Spec{J (X)} and

Spec{R̃(X̃, Ỹ )} = Spec{R(X, Y )} ∪ − Spec{R(X, Y )} .

Proof. Let ij k be the Christoffel symbols of ∇ and let ij be the components of . We use
Lemma 1.31 to see that the (possibly) non-zero Christoffel symbols ˜ αβ γ of the Levi-Civita con-
nection of g∇, are given by:
 
˜ ij k = ij k , ˜ i  j k = −j k i , ˜ ij  k = −ik j ,
 
 
˜ ij k = xr  ∂xk ij r − ∂xi j k r − ∂xj ik r +2 r
k ij 

r 

1  
+ ∂xi j k + ∂xj ik − ∂xk ij − k ij  .
2


By Lemma 1.31, the non-zero components of the curvature tensor of (T ∗ M, g∇, ) up to the usual

symmetries are given as follows; we omit R̃kj i h , as it plays no role in our considerations:
  
R̃kj i h = Rkj i h , R̃kj i h , R̃kj i  h = −Rkj h i , R̃k  j i h = Rhij k .

Let X̃ = α i ∂xi + αi  ∂x i  and Ỹ = β i ∂xi + βi  ∂x i  be vector fields on T ∗ M. Let X = α i ∂xi and

Y = β i ∂xi be the corresponding vector fields on M. Let J (X) and R(X, Y ) be the matrices of
the Jacobi operator and of the skew-symmetric curvature operator on M relative to the basis {∂xi }.
Then the matrix of the Jacobi operator J˜ (X̃) and of the skew-symmetric curvature operator R̃(X̃, Ỹ )
with respect to the basis {∂xi , ∂x i  } have the form:
   
J (X) 0 R(X, Y ) 0
J˜ (X̃) = t J (X) and R̃(X̃, Ỹ ) = .
∗ ∗ − R(X, Y )
t

The desired result now follows. 2

The following is the main result of this section:

72 2. RIEMANNIAN EXTENSIONS
Theorem 2.15

1. Let (M, g) be a 4-dimensional self-dual Walker manifold.

(a) If (M, g) is Ricci flat, then (M, g) is locally isometric to a deformed Riemannian extension.
(b) If (M, g) is Ivanov–Petrova, then (M, g) is locally isometric to a deformed Riemannian
extension.

2. Let (M, ∇) be an affine manifold. Let ∇˜ ∇, be the Levi-Civita connection of (T ∗ M, g∇, ).

(a) The following Assertions are equivalent:

i. (M, ∇) is an affine Osserman space.
ii. (T ∗ M, ∇˜ ∇, ) is an affine Osserman space.
iii. (T ∗ M, g∇, ) is a pseudo-Riemannian Osserman space for any .
iv. (T ∗ M, g∇, ) is a pseudo-Riemannian Osserman space for some .
(b) The following Assertions are equivalent:
i. (M, ∇) is an affine Ivanov–Petrova space.
ii. (T ∗ M, ∇˜ ∇, ) is an affine Ivanova-Petrova space.
iii. (T ∗ M, g∇, ) is a pseudo-Riemannian Ivanov–Petrova space for any .
iv. (T ∗ M, g∇, ) is a pseudo-Riemannian Ivanov–Petrova space for some .

Proof. If the metric is self-dual, then we can use Equation (1.8.c) to express:

g11 (x 1 , x 2 , x1 , x2 ) = x13 A +x12 B +x12 x2 C +x1 x2 D +x1 P +x2 Q+ξ,
g22 (x 1 , x 2 , x1 , x2 ) = x23 C +x22 E +x1 x22 A +x1 x2 F +x1 S +x2 T +η,
g12 (x 1 , x 2 , x1 , x2 ) = 21 x12 F + 21 x22 D +x12 x2 A +x1 x22 C + 21 x1 x2 (B + E )
+x1 U +x2 V +γ ,

where the coefficients are smooth functions of (x 1 , x 2 ). One checks that

ρ(∂x1 , ∂x 1 ) = 41 (16x1 A + 8x2 C + 5B + E ), ρ(∂x1 , ∂x 2 ) = 2x1 C + D,

ρ(∂x2 , ∂x 2 ) = 41 (8x1 A + 16x2 C + B + 5E ), ρ(∂x2 , ∂x 1 ) = 2x2 A + F ,

and therefore, all the calligraphic letters vanish and thus, g is first order in {x1 , x2 } and has the form
of a deformed Riemannian extension. This establishes Assertion (1a).
We now prove Assertion (1b). For a general 4-dimensional pseudo-Riemannian manifold
(M, g) with metric of neutral signature, it is easy to check that the characteristic polynomial
pλ (R(π )) of R(π ) is given by

pλ (R(π )) = λ4 − 1
2 Tr{R(π )2 } λ2 + det{R(π)} .
 2.5. SPECTRAL GEOMETRY 73
Thus, (M, g) is Ivanov–Petrova if and only if det{R(π )} and Tr{R(π)2 }
do not depend on the
oriented non-degenerate spacelike (respectively, mixed or timelike) 2-plane π (see Calviño-Louzao,
García-Río, and Vázquez-Lorenzo [44]).
In the particular case of a Walker metric given in Remark 1.30, a straightforward calculation
shows that the skew-symmetric curvature operator R(π ) associated with any non-degenerate 2-
plane π , when expressed with respect to the coordinate vectors {∂xi , ∂x i  }, i = 1, 2, has the following
matrix form:  
F (π ) 0
R(π) = ,
G(π ) − tF (π )
for suitably chosen 2 × 2 matrices F (π ) and G(π ). Thus:

det{R(π )} = det{F (π )}2 and Tr{R(π )2 } = 2 Tr{F (π )2 } .

Therefore, the Walker metric of Remark 1.30 is Ivanov–Petrova if and only if det{F (π )} and
Tr{F (π )2 } do not depend on the oriented non-degenerate spacelike (respectively, mixed or timelike)
2-plane π . Set
 
f11 (π ) f12 (π )
F (π ) = .
f21 (π ) f22 (π )
Assume the Walker metric is Ivanov–Petrova and self-dual and hence, locally given by Equa-
tion (1.8.c). Let π1 = Span{∂x1 , ∂x 1 + λ∂x 2 } be a non-degenerate 2-plane. One computes:
f11 (π1 ) = x1 (λC + 3A) + x2 C + 21 (λD + 2B ),
f12 (π1 ) = x1 λA + x2 (λC + A) + 41 (λ(B + E ) + 2F ),
f21 (π1 ) = x1 C + 21 D,
f22 (π1 ) = x1 (λC + A) + x2 C + 41 (2λD + B + E ).
Then ∂x 1 ∂x 1 (det{F (π1 )}) = 2λ2 C 2 + 6λAC + 6A2 , so A = C = 0 and one gets
det{F (π1 )} = 41 λ2 D2 + 21 λBD + 41 (B 2 + BE − DF ) ,
which implies that D = 0. Hence Equation (1.8.c) simplifies in this instance to become:
g11 (x 1 , x 2 , x1 , x2 ) = x12 B + x1 P + x2 Q + ξ ,
g22 (x 1 , x 2 , x1 , x2 ) = x22 E + x1 x2 F + x1 S + x2 T + η,
g12 (x 1 , x 2 , x1 , x2 ) = 21 x12 F + 21 x1 x2 (B + E ) + x1 U + x2 V + γ .
Consider the non-degenerate 2-plane π2 = Span{∂x1 + λ∂x2 , ∂x 1 }. We compute that:
f11 (π2 ) = 21 (λF + 2B ), f12 (π2 ) = 21 F
f21 (π2 ) = 41 λ(B + E ), f22 (π2 ) = 41 (2λF + B + E ).
This implies that det{F (π2 )} = 41 λ2 F 2 + 21 λBF + 41 B (B + E ). Consequently, F = 0 and we may
express g in the form:
74 2. RIEMANNIAN EXTENSIONS
g11 (x 1 , x 2 , x1 , x2 ) = x12 B + x1 P + x2 Q + ξ ,
g22 (x 1 , x 2 , x1 , x2 ) = x22 E + x1 S + x2 T + η,
g12 (x 1 , x 2 , x1 , x2 ) = 21 x1 x2 (B + E ) + x1 U + x2 V + γ .
This permits us to compute:
det{F (π2 )} = 41 B (B + E ), Tr{F (π2 )2 } = B 2 + 16 (B
1
+ E )2 .
We also consider the non-degenerate 2-plane:
π3 = Span{∂x 1 − ∂x 2 , ∂x 2 − ∂x1 + ∂x2 } with f11 (π3 ) = 18 (5B + E ),
f12 (π3 ) = f21 (π3 ) = − 18 (B + E ), f22 (π3 ) = 18 (B + 5E ),
det{F (π3 )} = 16 (B + E 2 + 6BE ), and Tr{F (π3 )2 } = 16 (7B + 7E + 6BE ).
1 2 1 2 2

Comparing the values of det{F (π2 )} with det{F (π3 )} and of Tr{F (π2 )2 } with
Tr{F (π3 )2 }, we
conclude that E = B . Hence, det{F (π3 )} = 2 B , which implies B = κ and we may express g in the
1 2

form:
g11 (x 1 , x 2 , x1 , x2 ) = x12 κ + x1 P + x2 Q + ξ ,
g22 (x 1 , x 2 , x1 , x2 ) = x22 κ + x1 S + x2 T + η,
g12 (x 1 , x 2 , x1 , x2 ) = x1 x2 κ + x1 U + x2 V + γ .
To show that κ = 0, we consider the non-degenerate 2-plane:
π4 = Span{∂x2 − c∂x 1 − 1+b
2 ∂x 2 , ∂x1
 + 2 ∂x 1 }.
1−a 

A straightforward computation then yields:

 
f11 (π4 ) = 41 x1 x2 3κ 2 + (x1 U + x2 V )3κ + QS − U V + 4κγ − 2P2 + 2U1 ,
 
f12 (π4 ) = 41 κ + κη + S(V − P ) − T U + U 2 + 2S1 − 2U2 ,
 
f21 (π4 ) = 41 κ − κξ + Q(T − U ) + P V − V 2 − 2Q2 + 2V1 ,
 
f22 (π4 ) = 41 x1 x2 3κ 2 + (x1 U + x2 V )3κ − QS + U V + 2κγ + 2T1 − 2V2 .
We now verify that ∂x 1 ∂x 1 ∂x 2 ∂x 2 (det{F (π4 )}) = 49 κ 4 . This shows that κ = 0, and hence, g has
the form of a deformed Riemannian extension as desired; Assertion (1b) follows.
We now prove Assertion (2). The equivalence of Assertions (2-a-i) and (2-a-ii) and of As-
sertions (2-b-i) and (2-b-ii) now follows from Lemma 2.14. Also, clearly Assertion (2-a-ii) implies
Assertion (2-a-iii) and, similarly, Assertion (2-b-ii) implies Assertion (2-b-iii).
It is immediate that Assertion (2-a-iii) implies Assertion (2-a-iv) and that Assertion
(2-b-iii) implies Assertion (2-b-iv). Suppose Assertion (2-a-iv) holds so that there exists  for which
(T ∗ M, g∇, ) is pseudo-Riemannian Osserman. Let S := Spec{J˜ (X̃)} be the common spectrum
for any X̃ ∈ S + (TP̃ T ∗ M, g∇, ). For t > 0, set X̃t := t −1 ∂x1 + t∂x 1 . Then:

g∇, (X̃t , X̃t ) = 1 + O(t −1 ) so

S = (1 + O(t −1 )) · Spec{J˜ (X̃t )} = (1 + O(t −1 ))t −2 Spec{J (∂x1 )} .
 2.5. SPECTRAL GEOMETRY 75
Taking the limit as t → ∞ yields S = {0} and hence, (M, ∇˜ ∇, ) is affine Osserman. Similarly sup-
pose Assertion (2-b-iv) holds so that there exists  for which (T ∗ M, g∇, ) is pseudo-Riemannian
Ivanov–Petrova and let S := Spec{R̃(X̃, Ỹ )} be the common spectrum for any {X̃, Ỹ }, an orthonor-
mal basis for a spacelike 2-plane. Let X = ∂x1 and Y = ∂x2 . Set

X̃t := t −1 ∂x1 + t∂x 1 and Ỹt := t −1 ∂x2 + t∂x 2 .

Then

g∇, (X̃t , X̃t )g∇, (Ỹt , Ỹt ) = g∇, (X̃t , Ỹt )2 = 1 + O(t −1 ) so
S = (1 + O(t −1 )) · Spec{R̃(X̃t , Ỹt )} = (1 + O(t −1 ))t −2 Spec{R(∂x1 , ∂x2 )} .

Taking the limit as t → ∞ yields S = {0}, so (M, ∇˜ ∇, ) is affine Ivanov–Petrova. 2

In fact, we have proved a bit more. The following is a scholium to the proof and is a local
version of Theorem 2.15:

Theorem 2.16 Let (M, ∇) be an affine manifold. Let ∇˜ ∇, be the Levi-Civita connection of
(T ∗ M, g∇, ). Let P ∈ M and let ω ∈ TP∗ M.

1. The following Assertions are equivalent:

(a) (M, ∇) is an affine Osserman space at P .

(b) (T ∗ M, ∇˜ ∇, ) is an affine Osserman space at (P , ω) for any ω and any .
(c) (T ∗ M, g∇, ) is a pseudo-Riemannian Osserman space at (P , ω) for any ω and any .
(d) (T ∗ M, g∇, ) is a pseudo-Riemannian Osserman space at (P , ω) for some (ω, ).

2. The following Assertions are equivalent:

(a) (M, ∇) is an affine Ivanov–Petrova space at P .

(b) (T ∗ M, ∇˜ ∇, ) is an affine Ivanova-Petrova space at (P,ω) for any ω and for any .
(c) (T ∗ M, g∇, ) is a pseudo-Riemannian Ivanov–Petrova space at (P , ω) for any ω and for
any .
(d) (T ∗ M, g∇, ) is a pseudo-Riemannian Ivanov–Petrova space at (P , ω) for some ω and for
some .

Note that self-dual Walker metrics that have nilpotent Ricci operator are not necessarily
deformed Riemannian extensions. For instance, the self-dual Walker metric given by Remark 1.30
where
g11 = 0, g22 = x1 x2 A(x 1 , x 2 ), g12 = 21 (x1 )2 A(x 1 , x 2 ),
76 2. RIEMANNIAN EXTENSIONS
has a 2-step nilpotent Ricci operator, but does not correspond to any deformed Riemannian exten-
sion.
We may apply Theorem 2.6 (4) to the situation in Theorem 2.15 (1). Any Ricci flat self-dual
Walker manifold (M, g) is locally isometric to the cotangent bundle T ∗  of an affine surface (, ∇)
equipped with the deformed Riemannian extension g∇, . The affine connections in question are
locally parametrized by a function θ (x 1 , x 2 ); there is no restriction on the symmetric (0, 2)-tensor
field .
While any 4-dimensional Riemannian Ivanov–Petrova manifold is necessarily locally confor-
mally flat, there are examples in neutral signature of 4-dimensional Ivanov–Petrova manifolds that
are neither self-dual nor anti-self-dual, as shown in Calviño-Louzao, García-Río, and Vázquez-
Lorenzo [44]. For instance, any Walker metric of the form:
g11 (x 1 , x 2 , x1 , x2 ) = x22 P (x 1 ) + x1 S(x 1 ) + x2 T (x 1 , x 2 ) + ξ(x 1 , x 2 ),
g22 (x 1 , x 2 , x1 , x2 ) = x2 κ + η(x 1 , x 2 ), g12 (x 1 , x 2 , x1 , x2 ) = 0
is Ivanov–Petrova, but neither self-dual nor anti-self-dual. If the Walker metric is self-dual, then
it must be a deformed Riemannian extension (see, for example, Calviño-Louzao, García-Río, and
Vázquez-Lorenzo [45]).
 77

CHAPTER 3

The Geometry of Modified

Riemannian Extensions
Let  be a symmetric (0, 2)-tensor field on an affine manifold (M, ∇) of dimension m̄ and let σ be
the natural projection from T ∗ M to M. We set T = S = Id and adopt the notation of Section 1.7.
The modified Riemannian extension g∇,,c is the metric of neutral signature (m̄, m̄) on the cotangent
bundle given in a coordinate free fashion by taking:
g∇,,c = c ι Id ◦ι Id +g∇ + σ ∗ .
x , x ), this takes the form:
In local coordinates (
g∇,,c = 2 dx i ◦ dxi  + c xi  xj  + ij (
x ) − 2xk  ij k (
x ) dx i ◦ dx j .

Modified Riemannian extensions are useful in describing 4-dimensional Osserman metrics, as we

shall see in Section 3.1. In Section 3.2, we will explore the setting where the underlying affine
connection is flat since this provides local models for the para-complex space forms. This motivates
the discussion in Section 3.3 where we use non-flat affine manifolds to construct higher-dimensional
Osserman metrics on the cotangent bundle, which are not nilpotent and which have non-trivial
Jordan normal form. These examples will be explored in greater detail in Section 3.4. We conclude
Chapter 3 in Section 3.5 with a discussion of (semi) para-complex Osserman manifolds.

3.1 FOUR-DIMENSIONAL OSSERMAN MANIFOLDS AND

MODELS
The work of Blažić, Bokan, and Gilkey [16] and of García-Río, Kupeli, and Vázquez-Abal [88]
shows that any Lorentzian Osserman manifold has constant sectional curvature. Furthermore, any
Osserman metric is Einstein. Since 3-dimensional Einstein metrics have constant sectional curvature,
any Osserman manifold of dimension three has constant sectional curvature. A result of Chi [53]
(see Theorem 3.3 below) shows that any Osserman manifold of signature (0, 4) is locally isometric
to a real space form or to a complex space form. Thus, the first non-trivial case arises in signature
(2, 2) and such manifolds are not completely classified, although there are many examples and some
results in this direction.
There is a special relationship between self-duality and the Osserman condition in signature
(0, 4) and (2, 2) that will play an important role in our discussion. Let W be the Weyl conformal
curvature tensor defined in Equation (1.5.f):
78 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
W (X, Y, Z, T ) := R(X, Y, Z, T ) + (m−1)(m−2) τ {g(Y, Z)g(X, T ) − g(X, Z)g(Y, T )}
1

− m−2
1
{g(Y, Z)ρ(X, T ) − g(X, Z)ρ(Y, T ) + g(X, T )ρ(Y, Z) − g(Y, T )ρ(X, Z)}.
Let W be the associated conformal curvature operator. Fix an orientation. If m = 4, then we can de-
compose W = W + + W − into the self-dual and anti-self-dual Weyl conformal curvature operators.
The manifold (M, g) is said to be self-dual if W − = 0; this is a conformally invariant property. If
W + = 0, we will usually reverse the orientation to simplify the discussion. We refer to Gilkey, Swann,
and Vanhecke [114] in signature (0, 4) and to Brozos-Vázquez, García-Río, and Vázquez-Lorenzo
[33] in signature (2, 2) for the proof of the following result:

Lemma 3.1 Let M = (V , ·, ·, A) be a curvature model of signature (2, 2) or (0, 4). Then M is
Osserman if and only if M is Einstein and self-dual.

OSSERMAN MANIFOLDS OF SIGNATURE (0, 4)

The study of Osserman metrics is, in general, a 2-step process. First one works in the purely algebraic
setting and classifies the set of all Riemannian Osserman curvature models of a given signature. One
then passes to the geometric setting to determine which of the Riemannian Osserman curvature
models can be geometrically realized by an Osserman manifold. Although we are primarily interested
in signature (2, 2) geometry, it is worth illustrating this process first in the Riemannian setting. Let
M = (V , ·, ·, A) be an Osserman curvature model of signature (0, 4). The curvature tensor of
constant sectional curvature is given by:

A0 (x, y)z := y, zx − x, zy .

More generally, if is a skew-symmetric linear operator on (V , ·, ·), then we may define an
algebraic curvature tensor by setting:

A (x, y)z :=  y, z x −  x, z y − 2 x, y z .

We say that a collection { 1 , 2, 3} of skew-adjoint endomorphisms of V gives (V , ·, ·) a

quaternion structure if we have
j
i j + j i = −2δi Id and 3 = 2 1.

Theorem 1.52 specializes to the 4-dimensional setting to give:

Theorem 3.2 Let (V , ·, ·, A) be a 4-dimensional Riemannian Osserman curvature model. Then there
exists a quaternion structure { 1 , 2 , 3 } on (V , ·, ·) and there exist constants {α0 , α1 , α2 , α3 } so that
A = α0 A0 + α1 A 1 + α2 A 2 + α3 A 3 .
 3.1. FOUR-DIMENSIONAL OSSERMAN MANIFOLDS AND MODELS 79
We now center our attention in the geometric setting to prove a result of Chi [53] classifying
Riemannian Osserman manifolds in dimension four. Lemma 3.1 plays a crucial role in our analysis.

Theorem 3.3 If (M, g) is a 4-dimensional Riemannian Osserman manifold, then (M, g) is locally
isometric either to a real space form or to a complex space form.

Proof. Let M be an Osserman curvature model of signature (0, 4); M has the form given in The-
orem 3.2. By Lemma 3.1, M is self-dual and Einstein. Consequently, all the information we shall
need is encoded by the self-dual Weyl curvature operator W + . If W + vanishes identically, then M
is conformally flat and M has constant sectional curvature, so M is geometrically represented by a
space form. We therefore assume W + non-zero. Since W + is traceless, W + has either two or three
distinct eigenvalues.
Suppose there exists an Osserman manifold (M, g) of signature (0, 4) that is not conformally
flat and which realizes M. The eigenvalues of W + are in correspondence with the eigenvalues of
the Jacobi operator and hence, are constant on (M, g). Thus, since W + is self-adjoint, W + , W + 
is constant. Suppose that W + has exactly two distinct eigenvalues. By the generalized Goldberg–
Sachs theorem (see, for example, Apostolov [8] and Ivanov and Zamkovoy [135]), there exists a
distinguished 2-form on M with  ,  = 2 corresponding to the eigenvalue of W + which has
multiplicity 1. Furthermore, is the Kähler form of an almost Hermitian structure (g, J− ); the
work of Derdzinski [64, Proposition 5] shows that (g, J− ) is actually locally conformally Kähler
with conformal factor (24W + , W + )1/3 . Since W + , W +  is constant, (g, J− ) is in fact Kähler.
Bryant [39] has shown that 4-dimensional Bochner flat Kähler metrics are characterized by the fact
that the anti-self-dual Weyl curvature operator W − vanishes. The Einstein condition now shows
that (g, J− ) is a Kähler structure of constant holomorphic sectional curvature and thus, as desired,
(M, g) is a complex space form.
Finally, suppose that W + has 3-distinct eigenvalues. We shall use results established by
Derdzinski [64] to obtain a contradiction. Let i be the three orthogonal eigenvectors of W +
corresponding to the three different eigenvalues μi . Let  be the Hodge operator. Since ∇ = 0,
the decomposition 2 (M) = + (M) ⊕ − (M) is parallel. Consequently, there exist 1-forms a, b,
c such that
∇ 1 = c ⊗ 2 − b ⊗ 3,
∇ 2 = −c ⊗ 1 + a ⊗ 3,
∇ 3 = b ⊗ 1 − a ⊗ 2.
We use the second Bianchi identity to see that δ W + = 0, or, equivalently,
dμ1 = (μ1 − μ2 ) 3c + (μ1 − μ3 ) 2 b,
dμ2 = (μ2 − μ1 ) 3 + (μ2 − μ3 )
c 1 a,
dμ3 = (μ3 − μ1 ) 2 b + (μ3 − μ2 ) 1 a.

Recall that a hyper Kähler structure on a Riemannian manifold (M, g) is a triple of parallel orthogonal
almost complex structures J1 , J2 , J3 , such that J1 J2 = −J2 J1 = J3 ; therefore, (g, Ji ) is Kähler for
80 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
all i = 1, 2, 3. In particular, any hyper Kähler manifold is Ricci flat (see Hitchin [127] for further
details). In the setting at hand, we may now conclude either that at least two of the eigenvalues μi
are equal (which is a contradiction) or, alternatively, that the forms i are parallel for all i. In this
instance, ( 1 , 2 , 3 ) defines a hyper Kähler structure on M, which is a contradiction. 2

OSSERMAN MODELS OF SIGNATURE (2, 2)

We say that { 1 , 2 , 3 } is a Clifford module structure of type (1, 1) on an inner product space
(V , ·, ·) of signature (2, 2) if the i are skew-adjoint operators satisfying:

3 = 2 1,
2
1 = − Id, 2
2 = 2
3 = Id, and i j + j i = 0 if i = j .

We refer to Karoubi [138] for further details concerning Clifford modules. An interesting aspect of
Clifford module structures of type (1, 1) is that they give rise to skew-symmetric 2-step nilpotent
operators 1 − j for j = 2, 3 and to a skew-symmetric almost product structure (after rescaling)
2 − 3 . Theorem 3.2 generalizes to this setting; a complete description of all Osserman algebraic
curvature tensors in the neutral signature is given by the following result of Blažić et al. [21]:

Theorem 3.4 A curvature model (V , ·, ·, A) of signature (2, 2) is Osserman if and only if there exists
a Clifford module structure { 1 , 2 , 3 } of type (1, 1) on V , and there exist constants αi and βij so that
 
A = α0 A0 + αi A i + βij [A i + A j − A( i − j ) ].
i<j

Let M = (V , ·, ·, A) be an Osserman curvature model of signature (2, 2). If X is not a null
vector, then the induced metric on X⊥ has Lorentzian signature and thus, J (X) can have complex
eigenvalues. Any 3 × 3 matrix is either diagonalizable (this gives rise to Type Ia below), or has
complex roots (this gives rise to Type Ib below), or has a 2 × 2 Jordan block with real eigenvalues
(this gives rise to Type II below), or has a 3 × 3 Jordan block with a real eigenvalue (this gives rise
to Type III below). Consequently, in this setting, the Jacobi operator J (X) = R( · , X)X, viewed
as an endomorphism of X⊥ , corresponds to one of the following possibilities:
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
α 0 0 α −β 0 α 0 0 α 0 0
⎝ 0 β 0 ⎠, ⎝ β α 0 ⎠, ⎝ 0 β 0 ⎠, ⎝ 1 α 0 ⎠.
0 0 γ 0 0 γ 0 1 β 0 1 α (3.1.a)

Type Ia Type Ib Type II Type III

The possible Jordan normal forms of the Jacobi operators in Equation (3.1.a) are in a one-
to-one correspondence with the possible Jordan normal forms of the corresponding self-dual Weyl
curvature operator W + . There is a complete algebraic classification due to Blažić, Bokan, and Rakić
[17]—see also the discussion in García-Río et al. [86]. Let M be an Osserman curvature model of
signature (2, 2). In what follows, we only give the non-zero components of A and W + . We shall
 3.1. FOUR-DIMENSIONAL OSSERMAN MANIFOLDS AND MODELS 81
consider an orthonormal basis {e1 , e2 , e3 , e4 }, where {e1 , e2 } are timelike and {e3 , e4 } are spacelike.
The orthonormal basis for + given in Equation (1.8.a) has signature (2, 1):

e1 ∧ e2 + e3 ∧ e4 e1 ∧ e3 + e2 ∧ e4 e1 ∧ e4 − e2 ∧ e3
E1+ = √ , E2+ = √ , E3+ = √ ,
2 2 2
E1+ , E1+  = 1, E2+ , E2+  = −1, E3+ , E3+  = −1.

Case Ia: The Jacobi operators are diagonalizable. There is an orthonormal basis for V where {e1 , e2 }
are timelike and {e3 , e4 } are spacelike so:
A1221 = A4334 = α, A1331 = A4224 = −β, A1441 = A3223 = −γ ,
A1234 = (−2α + β + γ )/3, A1423 = (α + β − 2γ )/3, A1342 = (α − 2β + γ )/3,
⎛ ⎞
2α − β − γ 0 0
2
W+ = ⎝ 0 −α + 2β − γ 0 ⎠.
3
0 0 −α − β + 2γ
√
Case Ib: The Jacobi operators have a complex eigenvalue α ± −1β. There is an orthonormal basis
for V where {e1 , e2 } are timelike and {e3 , e4 } are spacelike so:
A1221 = A4334 = α, A1331 = A4224 = −α, A1441 = A3223 = −γ ,
A2113 = A2443 = −β, A1224 = A1334 = β, A1234 = (−α + γ )/3,
A1423 = 2(α − γ )/3, A1342 = (−α + γ )/3,
⎛ 2 ⎞
− 3 (α − γ ) −2β 0
W+ = ⎝ 2β − 23 (α − γ ) 0 ⎠.
3 (α − γ )
4
0 0
Case II: The minimal polynomial of the Jacobi operators has a double root. There is an orthonormal
basis for V where {e1 , e2 } are timelike and {e3 , e4 } are spacelike so:

A1221 = A4334 = ε α − 1
2 , A2113 = A2443 = − 2ε , A1441 = A3223 = −β,

A1331 = A4224 = −ε α + 1
2 , A1224 = A1334 = 2ε , A1423 = 23 (εα − β),

A1234 = 13 ε −α + 23 + β , A1342 = 13 ε −α − 3
2 +β ,
⎛ ⎞
− 2(α−β)
3 +1 −1 0
+ ⎜ ⎟
W =⎝ 1 − 2(α−β)
3 − 1 0 ⎠ where ε = ±1.
4(α−β)
0 0 3
Case III: The minimal polynomial of the Jacobi operators has a triple root. There is an orthonormal
basis for V where {e1 , e2 } are timelike and {e3 , e4 } are spacelike so:
82 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
A1221 = A4334 = α, A1331 = A4224 = −α, A1441 = A3223 = −α,
√ √ √
A2114 = A2334 = − 2
A3114 = −A3224 = 22 ,
2 , A1223 = A1443 = 2
2 ,
⎛ √ ⎞
√ 0 0 √2
A1332 = −A1442 = 22 , W + = ⎝ 0 0 2 ⎠.
√ √
− 2 2 0

A triple of skew-adjoint operators { 1, 2, 3} forms a para-quaternionic structure if

2
1 = − id, 2
2 = id, 2
3 = id, i j + j i = 0 for i = j .
The following result of García-Río et al. [86] arises from the classification given above:

Theorem 3.5 Let M = (V , ·, ·, A) be a curvature model of signature (2, 2).
1. The following conditions are equivalent:
(a) M is spacelike Osserman or timelike Osserman or null Osserman.
(b) M is Einstein and self-dual.
(c) M is spacelike Jordan–Osserman or timelike Jordan–Osserman.
2. M is null Jordan–Osserman if and only if M is of Type Ia and if one of the following conditions
holds for some constants κi ∈ R:
(a) A has constant sectional curvature, i.e., A = κ0 A0 .
(b) There exists a Hermitian structure J− so A = κ0 A0 + κ1 AJ− for κ1 = 0.
(c) There exists a para-Hermitian structure J+ so A = κ1 AJ+ for κ1  = 0.
(d) There exists a para-quaternionic structure { 1, 2, 3} so that
A = κ1 A 1 + κ2 A 2 + κ3 A 3

where κ2 κ3 (κ2 + κ1 )(κ3 + κ1 ) > 0 and where {3κ1 , −3κ2 , −3κ3 } are all distinct.

OSSERMAN MANIFOLDS OF SIGNATURE (2, 2)

We refer to García-Río and Vázquez-Lorenzo [92] for the proof of the following result:

Theorem 3.6 Any locally symmetric 4-dimensional Osserman manifold is either a real, a complex, or a
para-complex space form or is locally isometric to the following example, which has a 2-step nilpotent Jacobi
operator:
g = 2dx 1 ◦ dx 2 + 2dx 3 ◦ dx 4 ± (x 1 )2 dx 4 ⊗ dx 4 .

We now examine seriatim the algebraic cases in Equation (3.1.a).

 3.1. FOUR-DIMENSIONAL OSSERMAN MANIFOLDS AND MODELS 83
Type Ia: Diagonalizable Jacobi Operator
We have the following result of Blažić, Bokan, and Rakić [17]:

Theorem 3.7 Any Type Ia Osserman manifold is locally isometric to a real, a complex, or a para-complex
space form.

Proof. Let (M, g) be a Type Ia Osserman manifold of signature (2, 2). If W + has only one eigenvalue
and is diagonalizable, then the metric has constant sectional curvature. If W + has exactly two distinct
eigenvalues, let be an eigenvector corresponding to the eigenvalue of multiplicity 1. Since the
induced metric on + has Lorentzian signature, one may normalize the choice of by requiring
that  ,  = ±2 ( is not null). If is spacelike, then defines an indefinite almost Hermitian
structure (g, J− ), while, if is timelike, then defines an almost para-Hermitian structure (g, J+ ).
A generalization of the Goldberg–Sachs theorem due to Apostolov [8] and to Ivanov and Zamkovoy
[135] now shows that such structures are locally conformally (para)-Kähler. As in the Riemannian
case, which was discussed above, the conformal factor is given by the norm of the self-dual Weyl
curvature operator and hence, it is constant as (M, g) is assumed Osserman.Therefore, the structures
associated to are Kähler (with respect to the opposite orientation) or para-Kähler, depending on
the causality of . The vanishing of W − implies that the Bochner curvature tensor vanishes so the
manifold is Bochner flat (see, for example, Bryant [39]), and thus, the metric corresponds either
to an indefinite complex space form or to a para-complex space form. Finally, proceeding as in our
discussion of 4-dimensional Riemannian Osserman manifolds, one shows that the Jacobi operators
may not have 3-distinct eigenvalues. 2

Type Ib: Complex Eigenvalues

Blažić, Bokan, and Rakić [17] also showed:

Theorem 3.8 There are no Type Ib Osserman manifolds.

Proof. Although Type Ib curvature models exist, they are not geometrically realizable as Type Ib
Osserman manifolds. Suppose (M, g) is a Type Ib Osserman manifold of signature (2, 2). We argue
for a contradiction. Recall that a pseudo-Riemannian manifold is said to be curvature homogeneous
of order k if, for each pair of points P , Q ∈ M, there exists an isometry P Q from TP M to TQ M so
∗P Q ∇ j RQ = ∇ j RP for 0 ≤ j ≤ k. As the Jordan normal form of the Jacobi operator may change
from point to point, a 4-dimensional Osserman manifold is not necessarily curvature homogeneous;
the Jordan normal form of the Jacobi operator is constant, of course, on S ± (M, g) by Theorem 3.5.
However, if (M, g) is Type Ib, then the Jordan normal form must be constant on all of M and (M, g)
is 0-curvature homogeneous; a similar argument holds for the self-dual Weyl curvature operator.
Derdzinski [65] showed that, if (M, g) is a signature (2, 2) Einstein manifold with a complex
diagonalizable Weyl curvature operator, then M is either locally symmetric or is locally isometric to
84 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
a Lie group with a left-invariant metric. We describe this example as follows. Let (x 1 , x 2 , x 3 , x 4 )
be the usual coordinates on R4 . Let
√
g = ek(x ) cos k(x 1 )2 3(dx 1 ◦ dx 1 − dx 4 ◦ dx 4 ) + dx 2 ◦ dx 2 − e−2k(x ) dx 3 ◦ dx 3
1 2 1 2

1 2 √
−2ek(x ) sin k(x 1 )2 3dx 1 ◦ dx 4 .
The restriction of the curvature operator R : 2 → 2 to the spaces of (anti)-self-dual
√ 2-forms
defines maps R± : ± → ± , which have constant eigenvalues {−k 2 , 21 (k 2 ± −3k 2 )} for k  = 0.
Consequently, none of these metrics is self-dual, and hence, no 4-dimensional Osserman metric may
have a Jacobi operator of Type Ib. 2

Type II and Type III Osserman Manifolds

Theorem 3.6 exhibits a locally symmetric Osserman manifold with a 2-step nilpotent Jacobi operator.
We refer to the discussion in Bonome et al. [26], in García-Río, Kupeli, and Vázquez-Lorenzo [90],
and in García-Río, Vázquez-Abal, and Vázquez-Lorenzo [91] for other examples of 2-step and
3-step nilpotent Osserman manifolds of signature (2, 2); the classification is far from complete.
However, if the Jacobi operators are non-nilpotent, then the work of Blažić, Bokan, and Rakić
[17] provides a complete description in the Type II setting. Such a metric is necessarily of Walker
type. A systematic analysis of self-duality conditions for Walker metrics was carried out by Davidov
and Muskarov [62] and by Díaz-Ramos, García-Río, and Vázquez-Lorenzo [71], thus leading to
the following result which motivates the construction of the new examples of Osserman metrics that
we shall give in the next sections and which is closely related to Lemma 1.39:

Theorem 3.9 Let (M, g) be a Type II Osserman manifold. Then one of the following two possibilities
holds:

1. The Jacobi operators are 2-step nilpotent.

2. The scalar curvature τ is not zero. In this situation, we have:

(a) There exist local coordinates (x 1 , x 2 , x1 , x2 ) and there exist {P , Q, S, T , U, V }, which are
functions of (x 1 , x 2 ) so:

g = 2dx 1 ◦ dx1 + 2dx 2 ◦ dx2 + g11 dx 1 ◦ dx 1 + g22 dx 2 ◦ dx 2 + 2g12 dx 1 ◦ dx 2 ,

g11 = (x1 )2 τ6 + x1 P + x2 Q + 6
τ {Q(T − U ) + V (P − V ) − 2(Q2 − V1 )},
g22 = (x2 )2 τ6 + x1 S + x2 T + 6
τ {S(P − V ) + U (T − U ) − 2(S1 − U2 )},
g12 = x1 x2 τ6 + x1 U + x2 V + 6
τ {−QS + U V + T1 − U1 + P2 − V2 }.

(b) Let c = τ/6. Then (M, g) is locally isometric to a modified Riemannian extension
g∇,,c = τ6 · ι Id ◦ι Id +g∇ + 24 ∗ ∇ρ .
τ σ s
 3.1. FOUR-DIMENSIONAL OSSERMAN MANIFOLDS AND MODELS 85
If (M, g) is as in Theorem 3.9 (2), then W+
has a distinguished eigenvalue with a timelike
associated eigenspace. The normalized eigenvector defines an almost para-Hermitian structure
(g, J+ ), which is no longer integrable. Work of Cortés-Ayaso, Díaz-Ramos, and García-Río [59]
shows that any such structure is symplectic and that (g, J+ ) is an almost para-Kähler structure of
constant para-holomorphic sectional curvature.

Example 3.10 The first examples of non-Ricci flat Type II Osserman metrics were given by Díaz-
Ramos, García-Río, and Vázquez-Lorenzo [70]. For 0  = k ∈ R and f = f (x 4 ), let:

g = 2(dx 1 ◦ dx 3 + dx 2 ◦ dx 4 ) + (4k(x 1 )2 − 4k1

f (x 4 )2 )dx 3 ◦ dx 3
1  4 (3.1.b)
+4k(x 2 )2 dx 4 ◦ dx 4 + 2(4kx 1 x 2 + x 2 f (x 4 ) − 4k f (x ))dx 3 ◦ dx 4 .

Let ∇ 21 2 = ∇ 12 2 = − 21 f (x 2 ) be the Christoffel symbols of an affine connection ∇ on R2 . One

shows that the metric of Equation (3.1.b) agrees with the resulting modified Riemannian extension
by computing:
∇ρ = − 41 f (x 2 )2 dx 1 ⊗ dx 1 − 21 f  (x 2 )dx 1 ⊗ dx 2 ,
∇ρ = − 41 f (x 2 )2 dx 1 ◦ dx 1 − 21 f  (x 2 )dx 1 ◦ dx 2 ,
s
g = 4k · ι Id ◦ι Id +g∇ + k1 σ ∗ ∇ρs .

Ricci flat self-dual Walker metrics may be either Type II or Type III Osserman metrics. We
apply the results of Section 2.3 and of Section 2.5 to describe these metrics in terms of deformed
Riemannian extensions as follows:

Theorem 3.11 A 4-dimensional Ricci flat self-dual Walker manifold (M, g) is locally isometric to
a deformed Riemannian extension g∇, of an affine surface (, ∇), where ∇ has a skew-symmetric
Ricci tensor, and where  is an arbitrary symmetric (0, 2)-tensor field on . The manifold (M, g) is
Type II if and only if (, ∇) is flat. The manifold (M, g) is Type III if and only if ∇ is not flat. The
manifold (M, g) is flat if and only if (, ∇) is flat and in a system of flat coordinates on ,  satisfies
∂x 2 ∂x 2 11 − 2∂x 1 ∂x 2 12 + ∂x 1 ∂x 1 22 = 0.

García-Río, Vázquez-Abal, and Vázquez-Lorenzo [91] also constructed Osserman metrics

with 3-step nilpotent Jacobi operators. Although such metrics are neither locally symmetric nor
locally homogeneous, Theorem 3.11 provides many examples. Recently, Derdzinski [67] has con-
structed Type III Osserman metrics with non-nilpotent Jacobi operators—see also the work of
Chudecki and Przanowski [55, 56].
Blažić, Bokan, and Rakić [18] showed that Type III Osserman metrics are foliated by degen-
erate surfaces that do not have to be parallel. Derdzinski [67] exhibited Type III Osserman metrics
whose Jacobi operators are non-nilpotent. Any Type III Osserman 4-dimensional manifold (M, g)
is locally the total space of an affine plane bundle over a surface, endowed with a distinguished “non-
linear connection” in the form of a horizontal distribution H, transverse to the vertical distribution
V of the bundle, both consisting of g-null vectors.
86 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
Null Jordan–Osserman Manifolds
Complex and para-complex space forms have many similarities. For instance, both have the same
eigenvalue structure for their Jacobi operators. Moreover, all the first-, second-, and third-order scalar
curvature invariants are exactly the same for both curvature models. The null Jordan–Osserman
models of signature (2, 2) are classified in Theorem 3.5. This gives rise to the following geometrical
classification by García-Río et al. [86], which provides a geometric criterion to distinguish between
para-complex and complex geometries:

Theorem 3.12 Let (M, g) be a connected pseudo-Riemannian manifold of neutral signature (2, 2).
Then (M, g) is null Jordan–Osserman if and only if either it has constant sectional curvature or it is locally
a complex space form.

3.2 PARA-KÄHLER MANIFOLDS OF CONSTANT

PARA-HOLOMORPHIC SECTIONAL CURVATURE
Recall from Section 1.10 that a para-Kähler manifold (M, g, J+ ) is said to be of constant para-
holomorphic sectional curvature c if the sectional curvature of any para-holomorphic plane is equal
to c. The curvature tensor of any para-Kähler manifold of constant para-holomorphic sectional
curvature is completely determined by the metric and the para-complex structure. If dim(M) ≥ 6,
then c is constant and the curvature takes the form:
R(X, Y )Z = 4 {g(Y, Z)X
c
− g(X, Z)Y − g(J+ Y, Z)J+ X
+ g(J+ X, Z)J+ Y + 2g(J+ X, Y )J+ Z} .

Bonome et al. [27] established the following analogue of Lemma 1.47:

Lemma 3.13 Let (M, g, J+ ) be a para-Kähler manifold. The following conditions are equivalent:

1. (M, g, J+ ) has constant para-holomorphic sectional curvature.

2. J (X)J+ X is proportional to J+ X, for all non-null vector fields X.

3. J (U )J+ U = 0 for all null vector fields U .

If (M, ∇) is an affine manifold, let

g∇,c := cι Id ◦ι Id +g∇ on T ∗ M .

Let be the canonical symplectic structure on T ∗ M, which is characterized invariantly in Equa-

tion (1.1.h) by the identity:
(X C , Y C ) = ι[X, Y ] .
 3.2. CONSTANT PARA-HOLOMORPHIC SECTIONAL CURVATURE 87
Define an endomorphism J+ of the tangent bundle of T ∗M by the identity

(X, Y ) = g∇,c (X, J+ Y ) . (3.2.a)

We use Equation (1.1.f), Equation (1.7.b), and Equation (3.2.a) to see that in a system of canonical
coordinates for T ∗ M we have that:

= dxi  ∧ dx i ,  
g∇,c = 2 dx i ◦ dxi  + cxi  xj  − 2xk  ∇ ij k (x) dx i ◦ dx j , (3.2.b)
 
J+ (∂xi ) = ∂xi − c xi  xj  − 2xk  ∇ ij k ∂x j  and J+ (∂x i  ) = −∂x i  .

Modified Riemannian extensions of flat connections have a special significance and provide local
models for para-Kähler manifolds of constant para-holomorphic sectional curvature c.

Theorem 3.14 Adopt the notation established above.

1. J+ defines an almost para-complex structure on (T ∗ M, g∇,c ).

2. J+ is integrable if and only if (M, ∇) is flat.

3. If (M, ∇) is flat, then (T ∗ M, g∇,c , J+ ) is a para-Kähler manifold of constant para-holomorphic

sectional curvature c.

Proof. The fact that J+2 = Id and J+∗ g∇,c = −g∇,c follows directly from Equation (3.2.b); Asser-
tion (1) follows. Let P ∈ M. Choose local coordinates on M, using Lemma 1.6, so that ∇ (P ) = 0.
Let Q̃ ∈ σ −1 (P ). We use Equation (3.2.b) to establish Assertion (2) by computing at Q̃ that:
J+ [J+ ∂xi , ∂xj ] = 2xb ∂xj ∇ ia b ∂x a ,
J+ [∂xi , J+ ∂xj ] = −2xb ∂xi ∇ j a b ∂x a ,
[J+ ∂xi , J+ ∂xj ] = {2xb ∂xi ∇ j a b − 2xb ∂xj ∇ ia b }∂x a
+{xi  xa  ∂x a (xj  xc ) − xj  xa  ∂x a (xi  xc )}∂x c , so

NJ+ (∂xi , ∂xj )(Q̃) = 4xb ∇Rij a b (P )∂x a .

Let (M, ∇) be flat. As J+ is integrable and is closed, (T ∗ M, g∇,c , J+ ) is a para-Kähler
manifold. We complete the proof by showing that (T ∗ M, g∇,c , J+ ) has constant para-holomorphic
sectional curvature c. We do not sum over repeated indices in what follows. By Lemma 1.31, the non-
zero components of the curvature operator of (T ∗ M, g∇,c ) are given, up to the usual symmetries,
by:
88 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
c2   −c
g∇,cR i
ij α = 4 xj xα ,
g∇,cR  i
i ii = −c, g∇,cR  k
i ik = 2 ,
g∇,cR  j = −c , g∇,cR  i  = c2 xi2 , g∇,cR k =
2
c 2
i ji 2 i ii i  ik 2 xk  ,
g∇,cR  h = 3c2 x  x  , g∇,cR  i = 3c2 g∇,cR  h c2  
=
4 xi xk , 2 xk xh ,
i ii  
4 i h i ik i ik
 2
g∇,cR  i = c x  x  , g∇,cR  h 2 
g∇,cR  i = c2 x  x  ,
i ji 2 i j i αi = c
4 xα xh ,
  i αk 4 α k

g∇,cR  α = −c2 x  x  , i 
ij i 4 j α
g∇,cR
i  ii  = c, g∇,cR   k = c ,
i ik 2
g∇,cR   i  = c .
i jj 2

Let X = (X i ∂xi + X i ∂x i  ) and let εX = g∇,c (X, X). We now sum over repeated indices i and j to
compute:
g∇,cR(J 
+ X, X)X = −εX Xi g∇,cRi  ii i ∂xi + 2X i g∇,cRi  j i i ∂x j 
 
+εX Xi g∇,cRi  ii i ∂x i  + X i g∇,cR  i ∂ 
i ii x i

= c εX Xi ∂xi − X i c xi  xj  ∂x j  − X i ∂x i 

= c εX Xi J+ ∂xi + X i J+ ∂x i  .

Therefore, we get g∇,cJ (X)J+ X = c g∇,c (X, X) J+ X. This shows that (T ∗ M, g∇,c , J+ ) has con-
stant para-holomorphic sectional curvature c. Since para-Kähler manifolds of constant para-
holomorphic sectional curvature have a unique local-isometry type, Assertion (3) follows. 2

3.3 HIGHER-DIMENSIONAL OSSERMAN METRICS

In this section, we shall use Theorem 3.14 to construct new examples of Osserman manifolds by
perturbing the para-complex space forms. We shall be considering affine base manifolds which
are equipped with a non-flat affine Osserman connection. Since we are interested in constructing
Osserman metrics as Riemannian extensions of an affine Osserman connection, for any possible
Riemannian extension g∇,,c the symmetric (0, 2)-tensor field  must vanish identically by Lemma
1.39. We take c = 1 and consider the corresponding modified Riemannian extension g∇,1 .

Theorem 3.15 Let (M, ∇) be an affine manifold of dimension m̄.

1. If (M, ∇) is affine Osserman at P ∈ M, then the modified Riemannian extension

g∇,1 = ι Id ◦ι Id +g∇ on T ∗ M is Osserman at any Q̃ ∈ σ −1 (P ). The eigenvalues of the Jacobi
operators on the unit pseudo-sphere bundles S ± (TQ̃ T ∗ M, g∇,1 ) are ±(0, 1, 41 ) with multiplicities
(1, 1, 2m̄ − 2), respectively.

2. If (M, ∇) is globally affine Osserman, then (T ∗ M, g∇,1 ) is globally Osserman.

 3.3. HIGHER-DIMENSIONAL OSSERMAN METRICS 89
Before beginning the proof of Theorem 3.15, we first establish some preliminary results. Let
(M, ∇) be an affine manifold. Fix a point P of M. We use Lemma 1.6 to choose local coordinates
(x 1 , . . . , x m̄ ) on M so that (P ) = 0 and let 0 ∇ be the flat connection, 0 ∇∂xi ∂xj = 0. We let g0 be
the Riemannian extension defined by 0 ∇:

g0 = g0 ∇ = 2dx i ◦ dxi  + xi  xj  dx i ◦ dx j .

The auxiliary metric g0 on T ∗ M is not invariantly defined but depends on the coordinates chosen;
by Theorem 3.14, (T ∗ M, g0 , J+,g0 ) has constant para-holomorphic sectional curvature and thus, is
Osserman.

Lemma 3.16 Adopt the notation established above. Let g0R be the curvature operator of g0 , let g∇,1R be
the curvature operator of g∇,1 , and let 2 R := g∇,1R − g0R define the Jacobi operator 2 J . If ξ is a tangent
vector on T ∗ M, let a = σ∗ ξ be the corresponding tangent vector on M. Relative to the natural coordinate
frame {∂x1 , . . . , ∂xm̄ , ∂x 1 , . . . , ∂x m̄ } we have:
 ∇ 
2J (ξ ) = J (a) 0
t (∇J (a)) .


Proof. We write the curvature operator of g∇,1 as

g∇,1
R = g0R + g∇R + ER,

where ER is an additional term measuring the interactions among the metrics above. It follows from
Lemma 2.14 that the Jacobi operators corresponding to the Riemannian extension take a similar
form:    ∇ 
0 0 J (a) 0
g0
J (ξ ) = and J (ξ ) =
g∇
t (∇J (a)) .
 0 
We will complete the proof by showing that the interaction terms given by ER have Jacobi
operators of the form
 
E 0 0
J (ξ ) = .
 0
It is enough to show that the following components of ER vanish:

E
Rij k  , ERij k   , ERi  j k  , ERi  j k   , ERij  k  , E
Rij  k   ,
   

E
Ri  j  k  , ERi  j  k   , ERij  k  , ERij  k   , ERi  j  k  , ERi  j  k   .

We have ER = g∇,1R − g0R − g∇R. The metrics g∇,1 and g0 are Walker metrics. By Lemma 1.31,
the possible non-zero components of ER are:
  

E
Rij  k  , ERij  k   , ERij k  , ERij  k  .
90 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
We now show that the above four components vanish identically. By Lemma 1.31,

Rij  k  = g0Rij  k  = 21 {δi δjk + δi δk }, and g∇Rij  k  = 0, so ERij  k  = 0,

g∇,1 j

   
Rij  k   = g0Rij  k   = − 21 {δik δj + δi δk }, and g∇Rij  k   = 0, so ERij  k   = 0 .
g∇,1 j

Consider now the terms ERij k  . Once again, using Lemma 1.31, one has
g∇,1R  = −∂ ∇   + ∂ ∇  
ij k xj ik xi jk
+ 41 {(∂x s  (xj  xk  ) − 2∇ j k s )(∂x  (xi  xs  ) − 2∇ is  )
−(∂x s  (xi  xk  ) − 2∇ ik s )(∂x  (xj  xs  ) − 2∇ j s  )},
g0R
ij k
 = 41 {∂x s  (xj  xk  )∂x  (xi  xs  ) − ∂x s  (xi  xk  )∂x  (xj  xs  )},
g∇R 
ij k = −∂xj ∇ ik  + ∂xi ∇ j k  + ∇ j k s ∇ is  − ∇ ik s ∇ j s  .
We now use the fact that ∇ (P ) = 0 to see that:
g∇,1
Rij k  = g0Rij k  + g∇Rij k  .

This implies that ERij k  = 0. Finally, we consider those terms of the form ERij  k  . We use
Lemma 1.31 to see that:
g∇,1R  = −∂x ∇ ik j + ∂xk ∇ i j
ij  k
−2∇ s j ∇ ik s + ∇ ik s ∂x j  (x xs  ) + ∇ s r ∂x j  (xr  ∂x s  (xi  xk  ))
+ 41 {(∂x s  (xi  xk  ) − 2∇ ik s )(∂x j  (xs  x ) − 2∇ s j )
+(∂x s  (xi  x ) − 2∇ i s )(∂x j  (xs  xk  ) − 2∇ sk j )
−2∂x j  (x xs  ∂x s  (xi  xk  ))},
g0R   = 41 {∂x s  (xi  xk  )∂x j  (xs  x ) + ∂x s  (xi  x )∂x j  (xs  xk  )
ij k
−2∂x j  (x xs  ∂x s  (xi  xk  ))},
g∇R  = −∂x ∇ ik j + ∂xk ∇ i j − ∇ ik s ∇ s j + ∇ i s ∇ sk j .
ij  k
   
Since ∇ (P ) = 0, g∇,1Rij  k  = g0Rij  k  + g∇Rij  k  , so ERij  k  = 0 as desired. 2
We continue the discussion; we will use the underlying almost para-Kähler structure J+ of
(T ∗ M, g∇,1 ), which
is given in Equation (3.2.b). Let ξ ∈ S + (TQ̃ T ∗ M, g∇,1 ). Set

ξ1 := J+ ξ ∈ S − (TQ̃ T ∗ M, g∇,1 ) .

Let Eλ (ξ ) (resp. Eλ (ξ1 )) be the eigenspaces of g0J (ξ ) (resp. g0J (ξ1 )) for the eigenvalue λ ∈ {0, 1, 41 }
(resp. for λ ∈ {0, −1, − 41 }). Set
E0 (ξ ) = ξ · R = E−1 (ξ1 ), E1 (ξ ) = ξ1 · R = E0 (ξ1 ),
E 1 (ξ ) = {E0 (ξ ) ⊕ E1 (ξ )} = {E−1 (ξ1 ) ⊕ E0 (ξ1 )}⊥ = E− 1 (ξ1 ),
⊥
(3.3.a)
4 4
S (ξ ) := D ∩ E 1 (ξ ), U (ξ ) := E0 (ξ ) ⊕ D ,
4
 3.3. HIGHER-DIMENSIONAL OSSERMAN METRICS 91
where D = ker{σ∗ } is the null parallel vertical plane field on (T ∗ M, g∇,1 ). We then have
TQ̃ T ∗ M = E0 (ξ ) ⊕ E1 (ξ ) ⊕ E 1 (ξ ).
4

Lemma 3.17 Let (M, ∇) be an affine manifold. Let Q̃ ∈ T ∗ M. Let ξ ∈ S + (TQ̃ T ∗ M, g∇,1 ).

1. D = (ξ1 − ξ ) · R + S (ξ ).

2. 2J (ξ )D ⊂ S (ξ ).

3. g0J (ξ ) U (ξ ) ⊂ U (ξ ) and 2J (ξ ) U (ξ ) ⊂ U (ξ ).

Proof. First of all, observe that Equation (3.2.b) implies ξ1 − ξ ∈ D. Choose an orthonormal basis
for E 1 (ξ ) of the form {e1+ , . . . , em̄−1
+
, J+ e1+ , . . . , J+ em̄−1
+
}, where the ei+ are spacelike and the J+ ei+
4
are timelike. Then, Assertion (1) follows by noting that we have the following basis for D:

{ξ − ξ1 , e1+ − J+ e1+ , . . . , em̄−1

+ +
− J+ em̄−1 }.

Suppose now that Assertion (2) fails and choose η ∈ D so that 2J (ξ )η ∈ / S (ξ ). Then it follows by
Lemma 3.16 that 2J (ξ )η ∈ D. By Assertion (1), there exists c  = 0 so that
2
J (ξ )η = c(ξ − ξ1 ) + η1 for η1 ∈ S (ξ ) .

Thus, cξ ∈ E1 (ξ ) + Range{2J (ξ )} + E 1 (ξ ) ⊂ E0 (ξ )⊥ , which is false; this contradiction establishes

4
Assertion (2). To prove Assertion (3), express:

U (ξ ) = E0 (ξ ) ⊕ D = ξ · R ⊕ (ξ1 − ξ ) · R ⊕ S (ξ ) = ξ · R ⊕ ξ1 · R ⊕ S (ξ ) . (3.3.b)

Since g0J (ξ )ξ = 0, g0J (ξ )ξ1 = ξ1 , and S(ξ ) ⊂ E 1 (ξ ), g0J (ξ ) preserves U (ξ ). Analogously, since
4
2J (ξ )ξ = 0 and 2J (ξ )D ⊂ D, one has that 2 J (ξ ) preserves U (ξ ) as well. 2
We can now examine the eigenvalue structure:

Lemma 3.18 Let (M, ∇) be an affine manifold. Let Q̃ ∈ T ∗ M. Let ξ ∈ S + (TQ̃ T ∗ M, g∇,1 ). Assume
(M, ∇) is affine Osserman at P = σ (Q̃). If there is 0 = η ∈ TQ̃ T ∗ M ⊗R C with g∇,1 J (ξ )η = μη,
then:

1. If η  ∈ U (ξ ) ⊗R C, then μ = 41 .

2. If η ∈ U (ξ ) ⊗R C and if η  ∈ S (ξ ) ⊗R C, then μ = 0 or μ = 1.

3. If η ∈ S (ξ ) ⊗R C, then μ = 41 .

4. Spec{g∇,1J (ξ )} ⊂ {0, 1, 41 }.
92 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
Proof. Let (M, ∇) be affine Osserman at P . This means that ∇J (ξ ) is nilpotent at P . Thus,
by Lemma 3.16, 2J (ξ ) is also nilpotent at any point Q̃ in the fiber over P . By Assertion
(3) in Lemma 3.17, U (ξ ) is preserved by both g0J (ξ ) and 2J (ξ ). Let g0J˜ (ξ ), 2J˜ (ξ ), and
g∇,1J˜ (ξ ) = g0J˜ (ξ ) + 2J˜ (ξ ) be the induced operators on the quotient space:

V (ξ ) := {TQ̃ T ∗ M/U (ξ )} ⊗R C .

If η ∈/ U (ξ ) ⊗R C, then the projection η̃ ∈ V (ξ ) of η is different from zero and we have that

g∇,1J˜ (ξ )η̃ = μη̃. By Equation (3.3.b), one now has that

V (ξ ) = {E 1 (ξ )/S (ξ )} ⊗R C .
4

Consequently, g0J˜ (ξ ) = 41 Id. Since 2J˜ (ξ ) is nilpotent and g∇,1J˜ (ξ ) = 41 Id +2J˜ (ξ ), one obtains
that g∇,1J˜ (ξ ) has only the eigenvalue 41 . Thus, μ = 41 , which establishes Assertion (1).
In order to prove Assertion (2), suppose that there exists 0 = η ∈ U (ξ ) ⊗R C such that
η∈ / S (ξ ) ⊗R C and g∇,1J (ξ )η = μη. Note that S (ξ ) is preserved by the Jacobi operators
g0J (ξ ) and 2J (ξ ). We apply Lemma 3.17 and consider the operators g0J˜ (ξ ), 2J˜ (ξ ), and
g∇,1J˜ (ξ ) = g0J˜ (ξ ) + 2J˜ (ξ ) on the quotient space

Y (ξ ) := {U (ξ )/S (ξ )} ⊗R C .

/ S (ξ ) ⊗R C, its projection η̃  = 0 and μ is an eigenvalue of g∇,1J˜ (ξ ). Now Equation (3.3.b)

Since η ∈
shows that Y (ξ ) = ξ̃ · R ⊕ ξ̃1 · R. By Lemma 3.17,

J (ξ )ξ = 0 and
2
J (ξ )ξ1 = 2J (ξ )(ξ1 − ξ ) ∈ S (ξ ) .
2

Consequently, 2J˜ (ξ ) = 0. Since g0J˜ (ξ )ξ̃ = 0 and g0J˜ (ξ )ξ̃1 = ξ̃1 we have that g∇,1J˜ (ξ )ξ̃ = 0 and
g∇,1J˜ (ξ )ξ̃ = ξ̃ , which shows that μ ∈ {0, 1}, and Assertion (2) follows.
1 1
To prove Assertion (3), we note that g0J (ξ ) = 41 Id on S (ξ ) and that 2J (ξ ) is nilpotent and
preserves S (ξ ). Assertion (4) follows from Assertions (1)-(3). 2
Proof of Theorem 3.15: Let (M, ∇) be an affine manifold which is affine Osserman at P ∈ M.
Use Lemma 1.6 to choose local coordinates on M so that ∇ (P ) = 0 and let 0 ∇ be the flat affine
connection defined on a neighborhood of P whose Christoffel symbols vanish in these coordinates.
Define a 1-parameter family of metrics on T ∗ M interpolating between g∇,1 and g0 by considering
the modified Riemannian extensions associated to the connections
ε
∇ := ε∇ + (1 − ε)0 ∇.

Since ∇ (P ) = ∇ (P ) = 0 and since ∇ R(P ) = 0, the corresponding curvature operators satisfy
0 0

ε∇
R(P ) = ε · ∇ R(P ). Thus all the connections ε ∇ are affine Osserman at P . By Lemma 3.18,

Spec{gε ∇J (ξ )} ⊂ {0, 1, 41 }
 3.3. HIGHER-DIMENSIONAL OSSERMAN METRICS 93
for all ε and hence, the eigenvalues and their multiplicities are unchanged during perturbation defined
by ε.Taking ε = 0 yields the desired multiplicities and establishes Assertion (1) of Theorem 3.15 for ξ
spacelike. We now use results of García-Río, Kupeli, and Vázquez-Lorenzo [90] to see that spacelike
Osserman implies timelike Osserman and to relate the eigenvalues and eigenvalue multiplicities on
S + (TQ̃ T ∗ M, g∇,1 ) to the eigenvalues and eigenvalue multiplicities on S − (TQ̃ T ∗ M, g∇,1 ); alterna-
tively, of course, one could simply proceed directly as well.This proves Assertion (1) of Theorem 3.15;
Assertion (2) follows from Assertion (1). "
!
∗
The computation of the curvature on the zero section Z (T M) is qualitatively easier than in
the general setting since we may set x = 0.

Lemma 3.19 Let (M, ∇) be an affine manifold and let Q̃ ∈ Z (TP∗ M). Choose local coordinates so
∇ (P ) = 0. Then:

1. The possibly non-zero components of g∇R(Q̃) are g∇Rij k (Q̃) = ∇Rij k  (P ).

2. The non-zero components of g0R(Q̃) are: g0R

ij  i  j (Q̃) = g0Rii  j  j (Q̃) = − 21 for i  = j and
g0R   (Q̃) = −1.
ii i i

3. We have g∇,1R(Q̃) = g0R(Q̃) + g∇R(Q̃).

Proof. Let u = (u1 , . . . , un ) be coordinates on a pseudo-Riemannian manifold (U, h). Expand

h = hab dua ◦ dub . If the 1-jets of the components hab vanish at a point S of U , then differentiating
the Koszul formula yields:
h
Rabcd (S) = 21 {∂ua ∂uc hbd + ∂ub ∂ud hac − ∂ua ∂ud hbc − ∂ub ∂uc had }(S) . (3.3.c)

We apply this observation to the setting at hand. Let Q̃ be a point in the zero section of T ∗ M.
Then the coordinates x vanish at Q̃ and, since ∇ (P ) vanishes at P , the 1-jets of the coefficients
of the metrics g∇ , g0 , and g∇,1 vanish at Q̃. We establish Assertion (1) by using Equation (3.3.c) to
compute:
g∇
Rij k (Q̃) = 21 {∂xj ∂x  (−2xh ∇ ik h ) − ∂xi ∂x  (−2xh ∇ j k h )}(Q̃)
= {∂xi ∇ j k  − ∂xj ∇ ik  }(P )
= ∇Rij k  (P ) .

The proof of Assertion (2) and of Assertion (3) is similar. 2

As an immediate application of Theorem 3.15, observe that, since the product M1 × M2 of affine
Osserman manifolds (M1 , (1) ∇) and (M2 , (2) ∇) is still affine Osserman, the corresponding cotan-
gent bundle T ∗ (M1 × M2 ) provides examples of Osserman manifolds with non-zero eigenvalues.
Moreover, the Jacobi operators are not diagonalizable in general. The following result shows that
there is no scarcity of examples of affine Osserman connections and, moreover, that the Jordan
94 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
normal form of the Jacobi operators g∇,1J can be quite complicated; it also shows that (T ∗ M, g∇,1 )
need not be Jordan–Osserman:

Theorem 3.20 Let r ≥ 2 and let U be an r × r lower triangular matrix. There exists an affine
Osserman manifold (M, ∇) of dimension r + 1, there exists a point Q̃ in the zero section of the cotangent
bundle, and there exist ξi ∈ S + (TQ̃ T ∗ M, g∇,1 ) for i = 1, 2 so that g∇,1J (ξ1 ) is diagonalizable and so
that relative to a suitable basis for TQ̃ T ∗ M,
g∇,1J (ξ = 0 · Id1 ⊕1 · Id1 ⊕( 41 · Idr +U ) ⊕ ( 41 · Idr +t U ).
2)

Proof. Let r ≥ 2 and let M := Rr+1 . Let {x 0 , . . . , x r } be the usual coordinates on M and
{x0 , . . . , xr  } the dual fiber coordinates on T ∗ M. We let indices a, b, c, and d range from 1 through
r and indices i, j, and k range from 0 through r. Let Ua b be a lower triangular matrix (i.e., Ua b = 0
for b ≤ a). Let θ = θ(x 0 ) ∈ C ∞ (R). Define a torsion-free connection ∇ on T M with non-zero
Christoffel symbols:
∇
0a b = ∇ a0 b = θ (x 0 )Ua b .

This satisfies the hypotheses of Definition 2.3 and thus, by Theorem 2.4, (M, ∇) is geodesically
complete, Ricci flat, and affine Osserman. The possibly non-zero curvatures are:
∇
R0a0 b = ∂x0 ∇ a0 b + ∇ 0c b∇ a0 c = ∂x0 θ · Ua b + θ 2 · Uc b Ua c .

Set P = 0, Q̃ = (0, 0) and assume θ (0) = 0 and ∂x0 θ (0) = −1. We may then apply
Lemma 3.19 to get that the possibly non-zero components of the curvature at Q̃ are given by
g∇,1R(Q̃) = g0R(Q̃) + g∇R(Q̃) and moreover

g∇,1R
ii  i  i (Q̃) = −1, g∇,1R
ij kd  (Q̃) = ∇Rij k d (P ),
g∇,1R
ij  i  j (Q̃) = g∇,1Rii  j  j (Q̃) = − 21 (i  = j ).
Let ξ1 := √1 (∂x
1 + ∂x 1 ). Then g∇,1J (ξ1 ) = g0J (ξ1 ), so g∇,1J (ξ1 ) is diagonalizable; the curvature of
2
∇ plays no role. Take ξ2 := √1 (∂x
0 + ∂x 0 ) and J+ ξ2 = √1 (∂x
0 − ∂x 0 ). Then:
2 2
g0J (ξ = 0, g∇J (ξ = 0, g0J (ξ = 41 ∂xa ,
2 )ξ2 2 )ξ2 2 )∂xa
g∇J (ξ = Ua b ∂xb , g0J (ξ = J+ ξ2 , g∇J (ξ = 0,
2 )∂xa 2 )J+ ξ2 2 )J+ ξ2
g0J (ξ = 41 ∂x a , g∇J (ξ = Ub a ∂x b .
2 )∂x a  2 )∂x a 

This shows that the Jacobi operator g∇,1J (ξ2 ) satisfies

g∇,1J (ξ = 0 · Id1 ⊕1 · Id1 ⊕( 41 · Idr +U ) ⊕ ( 41 · Idr +t U ) . 2
2)
 3.4. OSSERMAN METRICS WITH NON-TRIVIAL JORDAN NORMAL FORM 95
3.4 OSSERMAN METRICS WITH NON-TRIVIAL JORDAN
NORMAL FORM
In Section 3.4, we will construct a family of Osserman manifolds of neutral signature whose Jacobi
operators are not nilpotent and that have non-trivial Jordan normal form. Let (, ∇)¯ be an affine
Osserman surface with non-zero curvature tensor at some point. Let

(M, ∇) := ( × Rm̄−2 , ∇¯ ⊕ 0 ∇) ,

where 0 ∇ is the flat Euclidean connection on Rm̄−2 . Since (M, ∇) is affine Osserman, Theorem 3.15
shows that (T ∗ M, g∇,1 ) is Osserman and that the spectrum of the Jacobi operator on S + (T ∗ M) is
{0, 1, 41 , . . . , 41 }. By Theorem 2.6 (4), we may choose local coordinates (x 1 , x 2 ) on  so that the
possible non-zero Christoffel symbols are:
∇¯ ∇¯
11 1 = −∂x1 θ and 22 2 = ∂x2 θ where ∂x2 ∂x1 θ (x 1 , x 2 )  = 0 . (3.4.a)

Let {i, j, k, h} be distinct indices ranging from 1 to m̄ and let {β, γ , δ} be indices ranging from 1
to m̄. By Lemma 1.31, the possibly non-zero components of curvature tensor of (T ∗ M, g∇,1 ) are
given by (do not sum):
g∇,1R
ij γ
i = 41 xj  xγ  , j = γ > 2 or, otherwise j  = γ and i > 2, or j > 2, or γ > 2,
2
211 = x2 ∂x1 ∂x1 ∂x2 θ − x1 ∂x1 ∂x2 ∂x2 θ + (x1 ∂x2 θ + x2 (x1 + ∂x1 θ))∂x1 ∂x2 θ ,
g∇,1R

1
212 = x1 ∂x1 ∂x2 ∂x2 θ − x2 ∂x1 ∂x1 ∂x2 θ − (x1 ∂x2 θ − x2 (x1 − ∂x1 θ))∂x1 ∂x2 θ ,
g∇,1R

g∇,1R  δ  = 1 x  x  , i  = β = δ > 2 or, otherwise

i βi 4 β δ
β  = δ  = i = β, and i > 2, or β > 2, or δ > 2,
g∇,1R  i  = 41 xβ  xγ  , i = β = γ > 2 or, otherwise
i βγ
β = γ  = i  = β and β  = 2 or γ = 2
g∇,1R  δ  = − 1 x  x  , j = δ > 2 or, otherwise j = δ and i > 2 or j > 2 or δ > 2,
ij i 4 j δ
1 = −1x x  − ∂ ∂ θ,
212 = 4 x1 x2 − ∂x1 ∂x2 θ,
g∇,1R g∇,1R 2 1
211 4 1 2 x1 x2
g∇,1R i = 1 x  (x  + 2∂ θ), i > 1, g∇,1R i = 1 x  (x  − 2∂ θ), i  = 2,
i11 4 1 1 x1 i22 4 2 2 x2
1 2 = x  (x  − 2∂ θ )∂ ∂ θ,
211 = x1 (x1 + 2∂x1 θ )∂x1 ∂x2 θ ,
g∇,1R g∇,1R
212 2 2 x2 x1 x 2
g∇,1R 2 = −x  x  ∂ ∂ θ, i > 2, g∇,1R 1 = x  x  ∂ ∂ θ, i > 2,
i11 1 i x1 x2 i22 2 i x1 x 2
g∇,1R 1 = x  x  ∂ ∂ θ, i > 2, g∇,1R 2 = −x  x  ∂ ∂ θ, i > 2,
i12 1 i x 1 x2 i21 2 i x1 x2
g∇,1R 1 = x  x  ∂ ∂ θ, k > 2, g∇,1R 2 = x  x  ∂ ∂ θ , k > 2,
21k 1 k x1 x 2 21k 2 k x1 x2
g∇,1R  1 = x  (x  + 2∂ θ ), g∇,1R  2 = 3 x  x  − ∂ ∂ θ,
1 11 1 1 x1 1 11 4 1 2 x1 x 2
g∇,1R  2 = x2 (x2 − 2∂x2 θ), g∇,1R  1 = 3 x  x  + ∂ ∂ θ,
2 22 2 22 4 1 2 x1 x 2
g∇,1R  1 = 43 x1 x2 + ∂x1 ∂x2 θ , g∇,1R  2 = 3 x  x  − ∂ ∂ θ,
1 12 2 21 4 1 2 x1 x 2
g∇,1R  i = 41 x1 (x1 + 2∂x1 θ ), i > 1, g∇,1R  i  = 1 x  (x  − 2∂ θ), i  = 2,
i 11 i 22 4 2 2 x2
96 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
g∇,1R i = xi2 , i > 2, g∇,1R 1 = 21 x1 (x1 + 2∂x1 θ), i > 1,
i  ii i  i1
g∇,1R 2 = 21 x2 (x2 − 2∂x2 θ ), i = 2, g∇,1R  k  = 1 x 2 , k > 2,
i  i2 i ik 2 k
g∇,1R  1 = 1 x  (x  + 2∂ θ ), i > 1, g∇,1R  2 = 1 x  (x  − 2∂ θ ), i  = 2,
i 1i 4 1 1 x1 i 2i 4 2 2 x2
g∇,1R  h = 3 x  x  , i > 2 or h > 2, g∇,1R  i  = 3 x  x  , i > 2 or k > 2,
i ii 4 i h i ik 4 i k
g∇,1R  h = 1 x  x  , g∇,1R  i  = 1 x  x  , 
212 = − 4 x1 x2 + ∂x1 ∂x2 θ,
g∇,1R 2 1
i ik 2 k h i ji 2 i j
 
211 = 4 x1 x2 + ∂x1 ∂x2 θ , 211 = 4 x2 (x2 − 2∂x2 θ),
g∇,1R 1 1 g∇,1R 2 1

g∇,1R  1 = − 1 x  (x  + 2∂ θ ), i > 1, g∇,1R  2 = − 1 x  (x  − 2∂ θ), i > 2,

i1i 4 1 1 x1 i2i 4 2 2 x2
g∇,1R  i = −1, g∇,1R  k = − 1 , g∇,1R  j = − 1 ,
i ii i ik 2 i ji 2
g∇,1R   i  = 1, g∇,1R   k  = 1 , g∇,1R   i  = 1 .
i ii i ik 2 i jj 2
We use the following technical result to simplify our subsequent computations:

Lemma 3.21 Let (M, ∇) = ( × Rm̄−2 , ∇˜ ⊕ 0 ∇), and let ξ ∈ S + (TP̃ (T ∗ M)) for P̃ ∈ T ∗ M.There
exists an isometry  of (T ∗ M, g∇,1 ) that preserves the associated almost para-complex structure J+ so
that:
1. If m̄ ≥ 3, then (P̃ ) = (a1 , a2 , 0, . . . , 0, b1 , b2 , b3 , 0, . . . , 0).
2. If m̄ ≥ 5, then ∗ ξ = (c1 , c2 , c3 , c4 , 0, . . . , 0, d1 , d2 , d3 , d4 , d5 , 0, . . . , 0).
3. Let J := g∇,1 J(P̃ ) (∗ ξ ). Relative to the canonical coordinate frame:

J ∂xi = 41 ∂xi and J ∂x i  = 41 ∂x i  for 6 ≤ i ≤ m̄,

J Span{∂xi , ∂x i  }1≤i≤5 ⊂ Span{∂xi , ∂x i  }1≤i≤5 .

Proof. Let m̄ ≥ 3. Let (x 1 , . . . , x m̄ , x1 , . . . , xm̄ ) be canonical local coordinates on T ∗ M. Let

P̃ = (a 1 , a 2 , . . . , a m̄ , ) ∈ T ∗ M, where  indicates terms not relevant for the moment. Set
x , x ) = (x 1 , x 2 , x 3 − a 3 , . . . , x m̄ − a m̄ , x1 , . . . , xm̄ ).
(
Since  arises from a transformation on M that preserves ∇,  preserves g∇,1 . Since  preserves
,  also preserves J+ . Consequently, we may assume henceforth that
P̃ = (a1 , a2 , 0, . . . , 0, b1 , b2 , . . . , bm̄ ) .
˜ =
If  = (ij ) is a linear transformation of Rm̄ , let  ˜ ij be the inverse linear transformation so
˜
that ij ki = δj . The induced transformation  =  ⊕ 
k ˜ of (
x , x ) is given by:

∗ x i = ij x j and ∗ xi  =  ˜ ki xk  so

∗ ˜
 (dx ◦ dxi  ) = ij ki dx ◦ dxk  = δjk dx j ◦ dxk  and
i j

˜ ui 
∗ (xi  xj  dx i ◦ dx j ) = ik j   ˜ vj xu xv  dx k ◦ dx 
= δk δ xu xv dx k ◦ dx  = xk  x dx k ◦ dx  .
u v  
 3.4. OSSERMAN METRICS WITH NON-TRIVIAL JORDAN NORMAL FORM 97
Let  = Id2 ⊕m̄−2 where m̄−2 ∈ GL(Rm̄−2 )
is a linear map of Rm̄−2 .
Since m̄−2 preserves
0 ∇,  preserves ∇ = ∇ ˜ ⊕ 0 ∇, so  preserves g∇,1 . Since  preserves ,  also preserves the
almost para-complex structure J+ . The maps we shall consider subsequently are of this form
and hence, are para-complex isometries. Let f := (b3 , . . . , bm̄ ). Choose m̄−2 ∈ GL(Rm̄−2 ) so
m̄−2 f = (c, 0, . . . , 0) for some c. Let  = Id2 ⊕m̄−2 and let  =  . Assertion (1) now fol-
lows since
Id2 ⊕m̄−2 P̃ = (a1 , a2 , 0, . . . , 0, b1 , b2 , c, 0, . . . , 0) .
To simplify the notation, we replace c by b3 . To prove Assertion (2), we assume m̄ ≥ 5 and that
P̃ = (a1 , a2 , 0, . . . , 0, b1 , b2 , b3 , 0, . . . , 0). Let ξ = (c1 , . . . , cm̄ , ). Let e be (c4 , c5 , . . . , cm̄ ) and
choose m̄−3 ∈ GL(Rm̄−3 ) so that m̄−3 e = (c, 0, . . . , 0) for some c. Let  = Id3 ⊕m̄−3 and let
 =  . Then

P̃ = P̃ and ∗ ξ = (c1 , c2 , c3 , c, 0, . . . , 0, d1 , d2 , d3 , d4 , . . . , dm̄ ) .

To simplify the notation, we shall replace c by c4 . Let f = (d4 , . . . , dm̄ ) = 0. We may choose
˜ m̄−4 ∈ GL(Rm̄−4 ) so m̄−4 f = (c, 0, . . . , 0) for some c . Let  = Id4 ⊕m̄−4 and let  =  .

Assertion (2) now follows since

P̃ = P̃ and ∗ ξ = (c1 , c2 , c3 , c4 , 0, . . . , 0, d1 , d2 , d3 , d4 , c, 0, . . . , 0) .

Again, we replace c by d5 to simplify the notation. To prove Assertion (3), we assume m̄ ≥ 6. Let
ei := ∂xi and fi := ∂x i  . Expand
j j j j
J ei = αi ej + βi fj and J fi = γi ej + $i fj .

Fix an index a ≥ 6 and let λ ∈ R. Let a,λ = Id5 ⊕m̄−5 (a,λ) be given by:

λei if i = a λ−1 fi if i = a
a,λ (ei ) = and (fi ) = .
ei if i  = a fi if i  = a

Then a,λ is an isometry of (T ∗ M, g∇,1 ), which fixes (P̃ , ξ ). Consequently, a,λ commutes with
J so

αia = βia = γia = $ia = αai = βai = γai = $ai = 0 for i  = a,

βaa = γaa = 0 .

If m̄ ≥ 7, fix 5 < a < b and let a,b = Id5 ⊕m̄−5 (a,b) be given by
⎧ ⎫ ⎧ ⎫
⎨ eb if i = a ⎬ ⎨ fb if i = a ⎬
a,b (ei ) = e if i = b and (fi ) = f if i = b .
⎩ a ⎭ ⎩ a ⎭
ei if i = a fi if i  = a
98 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
Again, a,b commutes with J and thus, αaa = αbb and $aa = $bb .This shows that J has the following
form relative to the basis {e1 , f1 , e2 , f2 , . . . , em̄ , fm̄ }:
⎛ ⎞
J10 0 0 0 ···
⎜ 0 α 0 0 ··· ⎟
J =⎜
⎝ 0
⎟. (3.4.b)
0 β 0 ··· ⎠
··· ··· ··· ··· ···

Since 0 and 1 are eigenvalues of multiplicity 1 and, since 1

4 has multiplicity 2m̄ − 2, we conclude
α = β = 41 . 2

By Theorem 3.20, (T ∗ M, g∇,1 ) is not Jordan–Osserman on the zero section. More generally,
we have that:

Theorem 3.22 ˜ be an affine Osserman surface with non-vanishing curvature tensor. Let
Let (, ∇)
(M, ∇) := ( × Rm̄−2 , ∇˜ ⊕ 0 ∇). Then (T ∗ M, g∇,1 ) is an Osserman manifold of signature (m̄, m̄)
whose Jacobi operators have eigenvalues {0, 1, 41 , . . . , 41 } on S + (T ∗ M). Furthermore, (T ∗ M, g∇,1 ) is
not Jordan–Osserman at any point P̃ ∈ T ∗ M.

Proof. Let (P̃ , ξ ) be normalized as in Lemma 3.21 (1,2). By Lemma 3.21 (3), (see Equation (3.4.b)),
we have that:  
J10 0
J (ξ ) := J (ξ )P̃ =
g∇,1
1 .
0 4 Id2m̄−10

This reduces the study of the Jordan normal form to the case m̄ = 5 and J = J10 . Set

A(ξ ) = J (ξ ) · (J (ξ ) − Id) · J (ξ ) − 41 Id .

We adopt the notation of Equation (3.4.a). We consider the unit spacelike vectors:

ξ̃ = ε−1/2 (0, 0, 0, ε, 0, 0, 0, 0, 21 , 0),

ξ̂ = ε−1/2 (1, 0, 0, 1, 0, 21 (ε − b12 − 2b1 ∂x1 θ (a1 , a2 )), 0, 0, 0, 0) .

Let (A(ξ̂ ))42 be the (4, 2)-entry of the (A(ξ̂ ))-matrix. One computes that

A(ξ̃ ) = 0, (A(ξ̂ ))42 = −ε−1 16

3
∂x1 ∂x2 θ (a1 , a2 )  = 0.

Thus, J (ξ̃ ) is diagonalizable while J (ξ̂ ) is not diagonalizable. This shows that (T ∗ M, g∇,1 ) is not
Jordan–Osserman at any point P̃ ∈ T ∗ M. 2
 3.4. OSSERMAN METRICS WITH NON-TRIVIAL JORDAN NORMAL FORM 99
FURTHER ANALYSIS OF THE JORDAN NORMAL FORM
We conclude this section by giving a complete analysis of the Jordan normal form for the manifold
of Theorem 3.22. There is no restriction in assuming m̄ = 5 and taking (P̃ , ξ ) normalized as in
Lemma 3.21. As in the previous theorem, set
r
Ar (ξ ) = J (ξ ) · (J (ξ ) − g∇,1 (ξ, ξ ) Id) · J (ξ ) − 41 g∇,1 (ξ, ξ ) Id , r ≥ 1.

We now examine the possible values of Ar (ξ ) for r = 1, 2, 3. From now on, we assume that all
objects defined on  (θ and its partial derivatives) are evaluated at (a1 , a2 ) and put εξ = g∇,1 (ξ, ξ ).
Then  
3 S 0
A(ξ ) = εξ ∂x1 ∂x2 θ ,
16 T tS
where the matrix S is given by
⎛ ⎞
−c1 c2 (ν1 + εξ ) c12 (ν1 + εξ ) 0 0 0
⎜ −c2 (ν1 + εξ ) c1 c2 (ν1 + εξ ) 0 0 0 ⎟
⎜ 2 ⎟
S=⎜ ⎜ −c2 c3 ν1 c 1 c 3 ν1 0 0 0 ⎟,
⎟
⎝ −c2 c4 ν1 c 1 c 4 ν1 0 0 0 ⎠
0 0 0 0 0

with ν1 = b1 b3 c1 c3 + b2 b3 c2 c3 + b32 c32 + 2c3 d3 + 2c4 d4 − εξ , and the matrix T is given by:
⎛ ⎞
−c2 t11 t12
c2 (d3 ν1 + (b3 c3 + d3 − ν2 )εξ )
2 c2 d4 ν1 c2 d5 ν1
⎜ 3εξ ∂x1 ∂x2 θ 3εξ ∂x1 ∂x2 θ ⎟
⎜ −c1 t22 ⎟
⎜ t21 ⎟
⎜ −c1 (d3 ν1 + (b3 c3 + d3 − ν2 )εξ ) −c1 d4 ν1 −c1 d5 ν1 ⎟
2
⎜ 3εξ ∂x1 ∂x2 θ 3εξ ∂x1 ∂x2 θ ⎟ ,
⎜ ⎟
⎜ c 2 ν1 ν2 −c1 ν1 ν2 0 0 0 ⎟
⎜ ⎟
⎝ c 2 d4 ν 1 −c1 d4 ν1 0 0 0 ⎠
c2 d5 ν1 −c1 d5 ν1 0 0 0

where
ν2 = b1 b3 c1 + b2 b3 c2 + b32 c3 + d3 ,
ν = ∂x1 ∂x2 θ(3b1 ∂x2 θ + 3b2 ∂x1 θ + 4∂x1 ∂x2 θ ) − 3b1 ∂x1 ∂x2 ∂x2 θ + 3b2 ∂x1 ∂x1 ∂x2 θ ,
t11 = c2 ν + 4c2 (∂x1 ∂x2 θ)2 {1 + 3ν12 + 2εξ (b32 c32 − 2c4 d4 + 2ν1 − 2c3 ν2 )}
− 3∂x1 ∂x2 θ{b1 (b1 c1 + b2 c2 + 2b3 c3 ) + εξ ν1 (b1 (b1 c1 + b2 c2 + b3 c3 ) + 2d1 )
+ 2d1 + 2b1 c1 (1 + εξ ν1 )∂x1 θ },
t22 = c1 ν + 4c1 (∂x1 ∂x2 θ )2 {1 + 3ν12 + 2εξ (b32 c32 − 2c4 d4 + 2ν1 − 2c3 ν2 )}
+ 3∂x1 ∂x2 θ{b2 (b1 c1 + b2 c2 + 2b3 c3 ) + εξ ν1 (b2 (b1 c1 + b2 c2 + b3 c3 ) + 2d2 )
+ 2d2 − 2b2 c2 (1 + εξ ν1 )∂x2 θ },
100 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
t12 = c1 c2 ν + 4c1 c2 (∂x1 ∂x2 θ )2 {1 + 3ν12 + 2εξ (b32 c32 − 2c4 d4 + 2ν1 − 2c3 ν2 )}
+ 3∂x1 ∂x2 θ{b1 b2 c1 c2 + εξ ν1 (b1 b2 c1 c2 + c1 d1 + c2 (b22 c2 + b2 b3 c3 + 3d2 ) + ν1 )
+ c1 d1 + c2 (b22 c2 + 2b2 b3 c3 + 3d2 ) + ν1 − 2b2 c22 (1 + εξ ν1 )∂x2 θ},
t21 = t12 + 3∂x1 ∂x2 θ{b32 c32 − 2(c1 d1 + c2 d2 + c4 d4 + c3 ν2 )
− εξ (ν1 (2c1 d1 + 2c2 d2 + ν1 ) − 1)}.
Hence, ⎛ ⎞
0 0
⎜ c22 −c1 c2 ⎟
⎜ 0 0 0 ⎟
⎜ ⎟
⎜ −c1 c2 c12 0 0 0 ⎟
A2 (ξ ) = 38 εξ (∂x1 ∂x2 θ )2 ψ ⎜ ⎟,
⎜ 0 0 0 0 0 0 ⎟
⎜ ⎟
⎝ 0 0 0 0 0 ⎠
0 0 0 0 0
where

ψ = b32 c32 {c1 d1 + c2 d2 − c3 d3 − c4 d4 + b1 c12 ∂x1 θ − b2 c22 ∂x2 θ }

−(2c3 d3 + 2c4 d4 − εξ ){b3 c3 (b1 c1 + b2 c2 ) + (c3 d3 + c4 d4 )} .

Thus A3 (ξ ) = 0. This shows that Rank{A(ξ )2 } ≤ 1, and Rank{A(ξ )} ≤ 2. Therefore, the Jacobi
operator corresponding to a given vector ξ is either diagonalizable, it contains a single 2 × 2 or 3 × 3
Jordan block, or it contains two 2 × 2 Jordan blocks, depending on the vector and the basepoint. All
these possibilities, except that of a single 2 × 2 Jordan block, can be realized at any point. Let

P̃ = (a1 , a2 , 0, 0, 0, b1 , b2 , b3 , 0, 0) and ξ = (c1 , c2 , c3 , c4 , 0, d1 , d2 , d3 , d4 , d5 ) .

1. If ψ  = 0, and (c1 , c2 ) = (0, 0), then the Jacobi operator has a 3 × 3 Jordan block. For example,
take: ξ = (1, 0, 0, 1, 0, 21 (ε − b12 − 2b1 ∂x1 θ − 2), 0, 0, 1, 0), g∇,1 (ξ, ξ ) = ε .

2. If c1 = c2 = 0, then the Jacobi operators are diagonalizable. For example, we could take
ξ = (0, 0, 0, ε, 0, 0, 0, 0, 21 , 0), g∇,1 (ξ, ξ ) = ε .

3. If ψ = 0, and c1 = c2 = 0 does not hold, then the Jacobi operator corresponding to the vectors
ξ = (c1 , c2 , 0, 0, 0, d1 , d2 , −b3 (b1 c1 + b2 c2 ), 0, 0) is either diagonalizable or it has a 2 × 2
Jordan block, depending on whether ν vanishes or not where

ν = ∂x1 ∂x2 θ (3b1 ∂x2 θ + 3b2 ∂x1 θ + 4∂x1 ∂x2 θ ) − 3b1 ∂x1 ∂x2 ∂x2 θ + 3b2 ∂x1 ∂x1 ∂x2 θ .

The Jacobi operator corresponding to any other unit vector has two 2 × 2 Jordan blocks. For
example, ξ = (1, 0, 0, 1, 0, 21 (ε − b12 − 2b1 ∂x1 θ ), 0, 0, 0, 0), g∇,1 (ξ, ξ ) = ε.

4. Finally, at any point where ν = 0, the Jacobi operators never have a unique 2 × 2 Jordan block.
 3.5. (SEMI) PARA-COMPLEX OSSERMAN MANIFOLDS 101
3.5 (SEMI) PARA-COMPLEX OSSERMAN MANIFOLDS
If (M, g, J+ ) is an almost para-complex manifold, let H ⊂ Gr 2 (T M) be the Grassmannian of all
para-complex 2-planes. If π ∈ Gr 2 (T M), then π ∈ H if and only if J+ π = π. If e1 is any unit
spacelike vector, then π := Span{e1 , J+ e1 } belongs to H and all elements of H arise in this way. We
adopt the notation of Equation (1.11.a) and define the para-complex Jacobi operator by setting:

J (π ) := J (e1 ) − J (J+ e1 ) .

One says that (M, g, J+ ) is semi para-complex Osserman if J has constant eigenvalues on H. If, in
addition, J (π) commutes with the almost para-complex structure J+ , then (M, g, J+ ) is said to be
para-complex Osserman. There are similar notions in the complex setting, which we shall not pursue
further; we refer to Brozos-Vázquez, García-Río, and Gilkey [29] for further details. The modified
Riemannian extensions provide examples of semi para-complex manifolds in the neutral signature
context, which are para-complex if and only if the underlying affine connection is flat.

Theorem 3.23 Let (M, ∇) be an affine manifold of dimension m̄. Give the associated modified Rie-
mannian extension (T ∗ M, g∇,1 ) the almost para-complex structure J+ of Equation (3.2.b). Let J (π) be
the associated higher order Jacobi operator for π ∈ H(T ∗ M, g∇,1 ).

1. The eigenvalues of J (π ) are (1, 21 ) with multiplicities (2, 2m̄ − 2) respectively.

2. The Jordan blocks of J (π) have size at most 2 × 2.

3. (T ∗ M, g∇,1 , J+ ) is semi para-complex Osserman.

4. J (π)J+ = J+ J (π) for all π if and only if ∇ is flat.

5. Let m̄ ≥ 3. There exists an affine Osserman manifold (M, ∇) of dimension m̄ so that

(T ∗ M, g∇,1 , J+ ) is not Jordan semi para-complex Osserman, and so that the para-complex Ja-
cobi operators are not always diagonalizable.

Proof. Let (M, ∇) be an affine manifold. Let Q̃ ∈ T ∗ M, let ξ ∈ S + (TQ̃ T ∗ M, g∇,1 ), and let
π = Span{ξ, J+ ξ }. Let a = σ∗ ξ = σ∗ J+ (ξ ). By Lemma 3.16,
 ∇   ∇ 
2J (ξ ) = J (a) 0 2J (J ξ ) = J (a) 0
t (∇J (a)) , + t (∇J (a)) ,
 1
 
2J (π ) = 2J (ξ ) − 2J (J ξ ) = 0 0
+ .
π 0
102 3. THE GEOMETRY OF MODIFIED RIEMANNIAN EXTENSIONS
Let D := ker{σ∗ } = Span{∂x 1 , . . . , ∂x m̄ }. Then Range{2J (π )} ⊂ D and 2J (π )D = 0. This shows
that 2J (π) is nilpotent. Choose a basis

{e1 , e2 , f1 , . . . , f2m̄−2 } for TQ̃ T ∗ M

so Span{e1 , e2 } is the +1 eigenspace of g0J (π ) and so Span{f1 , . . . , f2m̄−2 } is the + 21 eigenspace

of g0J (π). We compute:

(g∇,1J (π ) − Id)ei = 2J (π)ei ∈ D and (g∇,1J (π ) − 1

2 Id)fi = 2J (π )fi ∈ D .

By Equation (3.3.a),

J (π ) = Id on (ξ − J+ ξ ) · R and
g0 g0
J (π ) = 1
2 Id on S (ξ ) .

By Lemma 3.17, g0J (π)D ⊂ D. As 2J (π ) = 0 on D, g∇,1J (π )D ⊂ D and:

Range{(g∇,1J (π ) − Id) · (g∇,1J (π ) − 1

2 Id)} ⊂ D .

Since 2J (π ) = 0 on D, g∇,1J (π ) = g0J (π ) on D. Equation (3.3.a) and Lemma 3.17 yield:

(g∇,1J (π ) − Id) · (g∇,1J (π ) − 21 Id)D = {0} so

(g∇,1J (π ) − Id)2 ·(g∇,1J (π ) − 21 Id)2 = {0} .

Consequently Spec{g∇,1J (π)} ⊂ { 21 , 1} and g∇,1J (π ) has only 1 × 1 or 2 × 2 Jordan blocks. As in

the proof of Theorem 3.15, we set ε ∇ := ε∇ + (1 − ε)0 ∇ to construct a 1-parameter family of
semi para-complex Osserman metrics gε ∇ interpolating between g∇,1 and g0 . As the eigenvalues
are unchanged, the eigenvalue multiplicities are unchanged. Consequently, 21 is an eigenvalue of
multiplicity 2m̄ − 2 and 1 is an eigenvalue of multiplicity 2. Assertions (1)-(3) now follow.
Generalizing the para-Kähler identity discussed in Section 1.4, one says that an almost para-
Hermitian manifold satisfies the third Gray identity if the curvature tensor is invariant under the
almost para-complex structure:

R(J+ · , J+ · , J+ · , J+ · ) = R( · , · , · , ·) .

The third Gray identity in the complex setting is crucial (see, for example, the discussion in Brozos-
Vázquez, García-Río, and Gilkey [29], in Di Scala and Vezzoni [72], and in Di Scala, Lauret, and
Vezzoni [73]). A purely algebraic computation shows that an almost para-Hermitian manifold (in
our case (T ∗ M, g∇,1 , J+ )) satisfies the third Gray identity if and only if we have the commutation
relation J+ J (π ) = J (π)J+ for all π ∈ H(T ∗ M, J+ ).
Suppose that (T ∗ M, g∇,1 , J+ ) satisfies the third Gray identity. Fix a point Q̃ in the zero
section of T ∗ M with P = σ (Q̃). Use Lemma 1.6 to choose coordinates on M so ∇ (P ) = 0. We
apply Lemma 3.19. Since C̃P satisfies the third Gray identity, we conclude g∇R satisfies the third
 3.5. (SEMI) PARA-COMPLEX OSSERMAN MANIFOLDS 103
Gray identity at Q̃. Thus
g∇R(∂
xi , ∂xj , ∂xk , ∂x  )(Q̃) = g∇R(J+ ∂xi , J+ ∂xj , J+ ∂xk , J+ ∂x  )(Q̃)
= g∇R(∂xi , ∂xj , ∂x  , ∂xk )(Q̃)
= g∇
R(∂xi , ∂xj , ∂x  , ∂xk )(Q̃) .

Consequently g∇R(Q̃) = 0. By Lemma 3.19, this implies ∇ R(P ) = 0 and since P was arbitrary,
(M, ∇) is flat, which completes the proof; the converse implication is trivial. Assertion (4) now
follows.
We now establish Assertion (5). Let m̄ ≥ 3, let M = Rm̄ , let P = 0, and let Q̃ = (0, 0). Let
θ = θ(x 1 ) be a smooth function of a single variable. Suppose θ (0) = 0 and ∂x1 θ(0)  = 0. Define
an affine connection ∇ by requiring that the only non-zero Christoffel symbol is ∇ 22 3 = θ. Since
θ = θ(x 1 ) and the only non-zero covariant derivative is ∇∂x2 ∂x2 = θ ∂x3 , the only non-zero curvature
is ∇ R(∂x1 , ∂x2 )∂x2 = (∂x1 θ)∂x3 . Hence, the associated Jacobi operators ∇J are nilpotent and (M, ∇)
is affine Osserman. By Lemma 3.19, the non-zero components of the curvature tensor at Q̃ (which
is on the zero section of T ∗ M) are:
g∇,1
Rii  i  i (Q̃) = −1, R3 221 (Q̃) = ∂x1 θ,
g∇,1 
g∇,1
Rij  i  j (Q̃) = g ∇,1Rii  j  j (Q̃) = − 21 (i  = j ) .

We first consider

ξ := √1 (∂x
1 + ∂x 1 ), J+ ξ = √1 (∂x
1 − ∂x 1 ), πξ := Span{ξ, J+ ξ } .
2 2

As 2J (ξ ) = 2J (J+ ξ ) = 0, J (ξ ) = g0J (ξ ) and J (πξ ) = g0J (πξ ) are diagonalizable. Next consider

η := 21 (∂x1 + ∂x3 + ∂x 1 + ∂x 3 ), J+ η = 21 (∂x1 + ∂x3 − ∂x 1 − ∂x 3 ), πη := Span{ξ, J+ ξ } .

The only non-trivial components of the Jacobi operators are J (η)∂x2 = 21 ∂x1 θ ∂x 2 and
J (J+ η)∂x2 = − 21 ∂x1 θ ∂x 2 . Consequently:

J (πη )∂x2 = (∂x1 θ )∂x 2 .

Since π2 := Span{∂x2 , ∂x 2 } is contained both in the 41 -eigenspace of g0J (η) and in the 21 -eigenspace
of g0J (πη ), we see that both J (η) and J (πη ) exhibit non-trivial Jordan normal form. 2
 105

CHAPTER 4

(para)-Kähler–Weyl Manifolds
In Chapter 4 we report on work of Gilkey and Nikčević [102, 103, 104, 105] and of Brozos-Vázquez
et al. [32]. We work in the complex and in the para-complex contexts to examine (para)-Kähler–
Weyl structures. If the underlying dimension is at least six, then any (para)-Kähler–Weyl algebraic
curvature tensor is in fact Riemannian. Consequently, any (para)-Kähler–Weyl structure is trivial
in the geometric setting as well. The 4-dimensional setting is quite different. Since every (para)-
Kähler–Weyl algebraic curvature tensor is geometrically realizable, and since every 4-dimensional
(para)-Hermitian manifold admits a unique (para)-Kähler–Weyl structure, these structures can be
non-trivial. We shall always work in both the complex and the para-complex settings and shall, for
the most part, attempt to treat these two cases in parallel.

4.1 NOTATIONAL CONVENTIONS

We adopt the following notational conventions. Throughout Chapter 4, (M, g, J± ) will be a
(para)-Hermitian manifold. This means that (M, g) is a pseudo-Riemannian manifold, that J±
is an integrable (para)-complex structure on M, and that J±∗ g = ∓g. The (para)-Kähler form
± = ± (M, g, J± ) is characterized by the identity:

± (X, Y ) = g(X, J± Y ) for X, Y ∈ C ∞ (T M) ;

1 (M) = 0 henceforth, to avoid difficulties with the

it plays a central role. We shall assume HDeR
fundamental group; consequently, if φ is a smooth 1-form with dφ = 0, then we can find f so
df = φ. Let ∇ be a Weyl connection; this means that ∇ is an affine connection so that

∇g = −2φ ⊗ g where φ ∈ C ∞ (T ∗ M) is the associated 1-form .

We say that the quadruple (M, g, J± , ∇) is a (para)-Kähler–Weyl manifold if ∇J± = 0.The structure
is said to be a trivial (para)-Kähler–Weyl structure if there is a conformally equivalent metric g1 so that
∇ = g1 ∇ is the associated Levi-Civita connection; (M, g1 , J± ) is then a (para)-Kähler manifold. By
Theorem 1.11, the Weyl structure is trivial if and only if dφ = 0 or, equivalently, if the alternating
Ricci tensor ρa vanishes.
We recall the following notational conventions which were established in Section 1.3 and in
Section 1.4. Let (V , ·, ·) be an inner product space of dimension m. Let W ⊂ ⊗4 V ∗ be the space
106 4. (PARA)-KÄHLER–WEYL MANIFOLDS
of Weyl algebraic curvature tensors; these are the tensors satisfying:

A(x, y, z, w) + A(y, x, z, w) = 0,
A(x, y, z, w) + A(y, z, x, w) + A(z, x, y, w) = 0,
A(x, y, z, w) + A(x, y, w, z) = − m4 ρa (x, y)z, w .

Let R ⊂ W be the subspace of Riemannian algebraic curvature tensors satisfying additionally:

A(x, y, z, w) + A(x, y, w, z) = 0 .

Let K± ⊂ ⊗4 V ∗ denote the space of tensors satisfying the (para)-Kähler identity:

A(x, y, z, w) = ∓A(x, y, J± z, J± w) . (4.1.a)

The spaces of (para)-Kähler Riemannian tensors K±,R and of (para)-Kähler–Weyl tensors K±,W
are obtained by imposing the (para)-Kähler identity on R and on W, respectively:

K±,W = W ∩ K± and K±,R = R ∩ K± .

If the dimension is at least six, then no new phenomena arise. Section 4.2 is devoted to the
proof of the following result; Assertion (2) in the Riemannian setting is originally due to Pedersen,
Poon, and Swann [178] and to Vaisman [201, 202]; they used different approaches to the subject
than we shall employ:

Theorem 4.1

1. If (V , ·, ·, J± , A) is a (para)-Kähler–Weyl curvature model with dim(V ) ≥ 6, then A ∈ R.

Consequently K±,W = K±,R .

2. If (M, g, J± , ∇) is a (para)-Kähler Weyl manifold with dim(M) ≥ 6 and with HDeR

1 (M) = 0,

then the associated Weyl structure is trivial.

Let δ be the coderivative defined in Equation (1.1.b); we expressed δ in terms of the Hodge 
operator in Equation (1.8.b) setting δ = −  d on 2-forms. We will establish the following result
in Section 4.3; it is due to Kokarev and Kotschick [141] in the Riemannian setting; they used a quite
different approach than we shall use:

Theorem 4.2 1 (M) = 0

Every 4-dimensional (para)-Hermitian manifold (M, g, J± ) such that HDeR
1 ∗
admits a unique (para)-Kähler–Weyl structure with associated 1-form given by φ± = ± 2 J± δ ± .

By Theorem 1.24, we have a short exact sequence 0 → R → W → ρa {W} → 0. This gives

rise to the short exact sequence:

0 → K±,R → K±,W → ρa {K±,W } → 0 . (4.1.b)

 4.1. NOTATIONAL CONVENTIONS 107
Thus, attention is focused on the alternating Ricci tensor, which takes values in 2 . We consider
the structure groups which were defined in Section 1.5:
U± := {T ∈ O : T J± = J± T },
U± := {T ∈ O : T J± = J± T or T J± = −J± T } .
The group U+ is the para-unitary group and the group U− is the unitary group. The groups U± are
Z2 extensions of these groups; we let χ(T ) ∈ Z2 be defined by Equation (1.5.a):
T J± = χ (T )J± T .
It is convenient to be able to interchange the structures J± and −J± . We recall the following
definitions from Equation (1.5.d):
2∓,0 := {θ ∈ 2 : θ ⊥ ±, J±∗ θ = ∓θ }, and 2± := {θ ∈ 2 : J±∗ θ = ±θ } .
By Lemma 1.17, we have an orthogonal direct sum decomposition of
2 = ± · R ⊕ 2∓,0 ⊕ 2±
as the direct sum of inequivalent and irreducible U± modules. We shall show in Section 4.2, whilst
proving Theorem 4.2, that the module ± · R plays no role and that
ρa {K±,W } ⊂ 2± ⊕ 2∓,0 . (4.1.c)
The question of providing homogeneous examples was posed to us by Prof. Alekseevsky and we
are grateful to him for the suggestion.We note that Calvaruso [40] undertook a related 3-dimensional
problem by examining 3-dimensional homogeneous Lorentzian metrics with a prescribed Ricci
tensor. We will follow the discussion in Brozos-Vázquez et al. [32] to establish the following result
in Section 4.4:

Theorem 4.3
1. Let (V , ·, ·, J+ ) be a 4-dimensional para-Hermitian vector space of signature (2, 2). Then every
element of 2−,0 ⊕ 2+ is geometrically realizable as the alternating Ricci tensor of the para-Kähler–
Weyl structure of a suitably chosen 4-dimensional Lie group with a left-invariant para-Hermitian
structure of signature (2, 2).
2. Let (V , ·, ·, J− ) be a 4-dimensional Hermitian vector space of signature (0, 4).Then every element
of 2+,0 ⊕ 2− is geometrically realizable as the alternating Ricci tensor of the Kähler–Weyl structure
of a suitable chosen 4-dimensional Lie group with a left invariant Hermitian structure of signature
(0, 4).

We shall use Theorem 4.3 to prove the following result in Section 4.5; the proof originally
given by Gilkey and Nikčević [103, 105] was quite different and did not employ the examples of
Theorem 4.3.
108 4. (PARA)-KÄHLER–WEYL MANIFOLDS
Theorem 4.4 Let (V , ·, ·, J± ) be a 4-dimensional (para)-Hermitian vector space of dimension four.
Then K±,W = K±,R ⊕ L2± where ρa provides a U± module isomorphism from L2± to 2∓,0 ⊕ 2± .

Theorem 4.4 is one of the facts about 4-dimensional geometry that distinguishes it from the
higher-dimensional setting; the module L2± provides additional curvature possibilities in dimension
m = 4 and is the algebraic reason why Theorem 4.1 fails if m = 4.
All the algebraic possibilities of Theorem 4.4 can be realized geometrically. In Section 4.6, we
will use Theorem 4.3 and Theorem 4.4 to establish the following result; again, the proof originally
given by Gilkey and Nikčević in [105] was quite different.

Theorem 4.5 Any 4-dimensional (para)-Kähler–Weyl curvature model can be geometrically realized
by a (para)-Kähler–Weyl manifold.

4.2 (PARA)-KÄHLER–WEYL STRUCTURES IF m ≥ 6

We follow the discussion of Gilkey and Nikčević in [102, 103] to prove Theorem 4.1. By Theo-
rem 1.25, we have the following decompositions into irreducible U± modules where, if m = 4, we
omit the modules {W±,5 , W±,6 , W±,10 } and where, if m = 6, we omit the module W±,6 from these
decompositions:
R = W±,1 ⊕ · · · ⊕ W±,10 , K±,R = W±,1 ⊕ W±,2 ⊕ W±,3 ,
W = W±,1 ⊕ · · · ⊕ W±,13 .
Set:  
K1±,W := ⊕4≤i≤13 W±,i ∩ K±,W .
Since W±,1 ⊕ W±,2 ⊕ W±,3 is a submodule of K±,W , we have:
K±,W = W±,1 ⊕ W±,2 ⊕ W±,3 ⊕ K1±,W .

We prove Theorem 4.1 (1) by showing K1±,W = {0} if m ≥ 6. Suppose that 4 ≤ i ≤ 13 and i = 9, 13.
Since W±,i appears with multiplicity 1 in ⊕4≤i≤13 W±,i , Lemma 1.16 shows either W±,i ⊂ K1±,W
or W±,i ⊥ K1±,W . By Theorem 1.25, W±,i ∩ K±,R = {0} for 4 ≤ i ≤ 10. Consequently,
 
K1±,W = W±,9 ⊕ W±,11 ⊕ W±,12 ⊕ W±,13 ∩ K±,W .
We recall the notation of Theorem 1.24. Decompose W as an O module in the form:
W = 1 ⊕ S02 ⊕ C ⊕ P .
Here, ρa provides an isomorphism from P to 2 and 1 denotes the trivial module. We recall the
inverse isomorphism is provided by the map  : 2 → W described in Theorem 1.24:
(ψ)(x, y, z, w) := 2ψ(x, y)z, w + ψ(x, z)y, w − ψ(y, z)x, w
(4.2.a)
− ψ(x, w)y, z + ψ(y, w)x, z .
 4.2. (PARA)-KÄHLER–WEYL STRUCTURES IF m ≥ 6 109
We also recall the isomorphism from 2± to W±,9 given in Theorem 1.25:

(ψ)(x, y, z, w) := 2x, J± yψ(z, J± w) + 2z, J± wψ(x, J± y)

+x, J± zψ(y, J± w) + y, J± wψ(x, J± z) (4.2.b)
−x, J± wψ(y, J± z) − y, J± zψ(x, J± w) .

These maps will play a crucial role in our study. Here, is an outline to the rest of this section. We will
first examine W±,11 = ( ± · R) and show the elements of W±,11 do not satisfy the (para)-Kähler
identity of Equation (4.1.a). Our analysis here does not depend upon the assumption that m ≥ 6
and remains valid if m = 4. Then we will treat the module W±,12 = 2∓,0 . Here, the assumption
that m ≥ 6 is essential—we shall see presently that the 4-dimensional setting is very different. Note
that W±,9 ⊕ W±,13 is isomorphic to two copies of 2± ; we next examine possible submodules of this
module. Again, the assumption m ≥ 6 is essential. Finally, we will pass to the geometric setting and
use Assertion (1) to establish Assertion (2).
The calculation is purely algebraic. Let m = 2m̄. In the para-complex setting, we choose an
orthonormal basis {e1 , . . . , em } for V so that:

e2i−1 , e2i−1  = 1, e2i , e2i  = −1, J+ e2i−1 = e2i , J+ e2i = e2i−1 .

In the complex setting with an inner product of signature (2p̄, 2q̄), we choose the orthonormal basis
{e1 , . . . , em }, so:

−1 if i ≤ 2p̄
ei , ei  = , J− e2i−1 = e2i , and J− e2i = −J− e2i−1 .
+1 if i > 2p̄
j
We set hij := ei , ej  = ±δi . This vanishes if i  = j , while if i = j , it is +1 or −1.

THE MODULE W±,11

We use Equation (4.2.a) to verify Equation (4.1.c) by computing:
( ± )(e1 , e4 , e3 , e1 ) =− ± (e4 , J± e3 )e1 , e1  = −h11 h44 .
∓( ± )(e1 , e4 , J± e3 , J± e1 ) = ± ± (e1 , J± J± e1 )e4 , J± e3  = h11 h44 .
(R · ± )  ⊂ K±,W if m ≥ 4.
1

THE MODULE W±,12

Let ψ±,0 := e1 ⊗ e2 − e2 ⊗ e1 + ν± {e3 ⊗ e4 − e4 ⊗ e3 } where ν± is chosen to ensure that
ψ±,0 ⊥ ± . We have J±∗ ψ±,0 = ∓ψ±,0 and thus, ψ±,0 ∈ 2∓,0 . Equation (4.2.a) yields:
(ψ±,0 )(e5 , e1 , e2 , e5 ) = −ψ±,0 (e1 , e2 )e5 , e5  = −h55 ,
∓(ψ±,0 )(e5 , e1 , J± e2 , J± e5 ) = ±ψ±,0 (e5 , J± e5 )e1 , J± e2  = 0, and
W±,12  ⊂ K1±,W if m ≥ 6.
110 4. (PARA)-KÄHLER–WEYL MANIFOLDS
THE MODULE W±,9 ⊕ W±,13
Let ψ± := e1 ⊗ e3 − e3 ⊗ e1 ± e2 ⊗ e4 ∓ e4 ⊗ e2 . Then J±∗ ψ± = ±ψ± so ψ± ∈ 2± . We use
Equation (4.2.a) and Equation (4.2.b) to see that:
1. (ψ± )(e5 , e1 , e3 , e5 ) = −ψ± (e1 , e3 )e5 , e5  = −h55 .
2. (ψ± )(e5 , e1 , J± e3 , J± e5 ) = (ψ± )(e5 , e1 , e4 , e6 ) = 0.
3. (ψ± )(e5 , e1 , e3 , e5 ) = 0.
4. (ψ± )(e5 , e1 , J± e3 , J± e5 ) = −ψ± (e1 , J± e4 )e5 , J± e6  = −h55 .
5. (ψ± )(e5 , e6 , e1 , e4 ) = 0.
6. (ψ± )(e5 , e6 , J± e1 , J± e4 ) = 0.
7. (ψ± )(e5 , e6 , e1 , e4 ) = 2e5 , J± e6 ψ± (e1 , J± e4 ) = 2h55 .
8. (ψ± )(e5 , e6 , J± e1 , J± e4 ) = 2e5 , J± e6 ψ± (J± e1 , J± J± e4 ) = ±2h55 .
For (a, b)  = (0, 0), let ξ(a, b) := Range{a + b } ⊂ W±,9 ⊕ W±,13 . If ξ(a, b) ∩ K1±,W = {0},
then ξ(a, b) ⊂ K1±,W by Lemma 1.16. Assertions (1)-(4) then yield a = ∓b, while Assertions (5)-
(8) yield b = 0. Thus, no submodule of the form ξ(a, b) intersects K1±,W . We apply Lemma 1.22
to see that every non-trivial proper submodule of W±,9 ⊕ W±,13 is isomorphic to ξ(a, b) for some
(a, b)  = 0. Thus,
 
W±,9 ⊕ W±,13 ∩ K1±,W = {0} .

Consequently, K1±,W = {0}. This completes the proof of Theorem 4.1 (1).

THE GEOMETRIC SETTING

Suppose that the first de Rham cohomology group HDeR 1 (M) vanishes. Then, by Theorem 1.11, a

Weyl structure (M, g, ∇) is trivial if and only if the alternating Ricci tensor ρa of ∇ vanishes or,
equivalently, if the curvature R of ∇ is Riemannian, i.e., R ∈ R. Assertion (2) of Theorem 4.1 now
follows from Assertion (1), since any element of K±,W belongs to K±,R and thus, has symmetric
Ricci tensor. "
!

4.3 (PARA)-KÄHLER–WEYL STRUCTURES IF m = 4

There are a number of steps in the proof of Theorem 4.2. We first show that if a (para)-Kähler–Weyl
structure exists in dimension four, then it is necessarily unique. We then establish the result in the
para-Hermitian context. Finally, we use analytic continuation to discuss the complex setting. We
assume m = 4 henceforth in this chapter.

THE UNIQUENESS OF THE (PARA)-KÄHLER–WEYL STRUCTURE

We begin with a purely algebraic result.
 4.3. (PARA)-KÄHLER–WEYL STRUCTURES IF m = 4 111
Lemma 4.6 Let (V , ·, ·, J± ) be a 4-dimensional (para)-Hermitian vector space. If φ belongs to V ∗,
let X (Y ) := φ(X)Y + φ(Y )X − X, Y φ # , where φ # ∈ V is dual to φ. Assume that [X , J± ] = 0
for all X. Then φ = 0.

Proof. We first assume that (V , ·, ·, J+ ) is a para-Hermitian vector space. Choose a hyperbolic
basis {e1 , e2 , e3 , e4 } which diagonalizes J+ :
J+ e1 = e1 , J+ e2 = e2 , J+ e3 = −e3 , J+ e4 = −e4 , e1 , e3  = e2 , e4  = 1 .
Let φ = a1 e1 + a2 e2 + a3 e3 + a4 e4 where {e1 , e2 , e3 , e4 } is the corresponding dual basis for V ∗ .
Then
e1 e4 = a1 e4 + a4 e1 , J+ e1 e4 = −a1 e4 + a4 e1 , e1 J+ e4 = −a1 e4 − a4 e1 ,
 e 2 e3 = a2 e3 + a3 e2 , J+ e2 e3 = −a2 e3 + a3 e2 , e2 J+ e3 = −a2 e3 − a3 e2 ,
 e 4 e1 = a4 e1 + a1 e4 , J+ e4 e1 = a4 e1 − a1 e4 , e4 J+ e1 = a4 e1 + a1 e4 ,
e3 e2 = a3 e2 + a2 e3 , J+ e3 e2 = a3 e2 − a2 e3 , e3 J+ e2 = a3 e2 + a2 e3 .
Equating ei J+ ej with J+ ei ej then implies a1 = a2 = a3 = a4 = 0 so φ = 0. This establishes
the Lemma in the para-Hermitian case.
Next, assume we are in the pseudo-Hermitian setting. Complexify and extend the inner
product to be complex bilinear. Choose a local frame {Z1 , Z2 , Z̄1 , Z̄2 } for V ⊗R C so
√ √ √ √
J− Z1 = −1Z1 , J− Z2 = −1Z2 , J− Z̄1 = − −1Z̄1 , J− Z̄2 = − −1Z̄2 ,
Z1 , Z̄1  = 1, Z2 , Z̄2  = ε2 .
Take ε2 = +1 if ·, · has signature (0, 4) and take ε2 = −1 if ·, · has signature (2, 2). Set
√
J+ := − −1J− , e1 := Z1 , e2 := Z2 , e3 := Z̄1 , e4 := ε2 Z̄2
and apply the argument given above to derive the Lemma in the pseudo-Hermitian setting where
the coefficients ai are now complex. 2
In the geometric setting, let g ∇ be the Levi-Civita connection of (M, g) and let φ ∇ be a Weyl
connection. Equation (1.2.q) yields the identity:
φ
∇X Y = g ∇X Y + φ(X)Y + φ(Y )X − g(X, Y )φ # .
Let φ = φ1 − φ2 and let X (Y ) := φ(X)Y + φ(Y )X − g(X, Y )φ # . Then, if φ1 ∇ and φ2 ∇ are two
Weyl connections for (M, g), we have:
φ1
∇X − φ2 ∇X = X .
Thus, if φ1 ∇J± = 0 and φ2 ∇J± = 0, we have [X , J± ] = 0 for all tangent vector fields X. The
following result is now an immediate consequence of Lemma 4.6:
Corollary 4.7 If φ1 ∇ and φ2 ∇ are two (para)-Kähler–Weyl connections on a 4-dimensional (para)-
Hermitian manifold, then φ1 = φ2 .
112 4. (PARA)-KÄHLER–WEYL MANIFOLDS
A PARA-HERMITIAN EXAMPLE
Let {e1 , e2 , e3 , e4 } be a hyperbolic basis for an inner product space (V , ·, ·) of signature (2, 2);
e1 , e3  = e2 , e4  = 1. We take the orientation to be given by e1 ∧ e3 ∧ e2 ∧ e4 . The Hodge 
operator is characterized by the identity:

ω1 ∧ ω2 = ω1 , ω2 e1 ∧ e3 ∧ e2 ∧ e4 .

Lemma 4.8
(e1 ∧ e3 ) = −e2 ∧ e4 , (e2 ∧ e4 ) = −e1 ∧ e3 , (e1 ∧ e2 ∧ e3 ) = −e2 ,
(e1 ∧ e2 ∧ e4 ) = e1 , (e1 ∧ e3 ∧ e4 ) = −e4 , (e2 ∧ e3 ∧ e4 ) = e3 .

Proof. The calculation is straightforward, but it is important to be careful concerning the signs. To
simplify the notation, we set ei1 ...ik := ei1 ∧ · · · ∧ eik . For each ω2 = eI , there is a unique ω1 = eJ
so that eJ , eI  = ±1; eI is then composed of the complementary indices to J and the sign is
controlled by the identity eJ ∧ eI = eJ , eI e1324 . We compute:

ω1 , ω2 , ω2 , ω1 ∧ ω2 = εe1324 , ω1 , ω2  = ε,

e13 , e , −e , e ∧ (−e ) = −e
13 24 13 24 1324 , e13 , e13  = −1,
e24 , e , −e , e ∧ (−e ) = −e
24 13 24 13 1324 , e24 , e24  = −1,
e134 , e , −e , e ∧ (−e ) = +e
123 2 134 2 1324 , e134 , e123  = +1,
e234 , e124 , +e1 , e234 ∧ (+e1 ) = +e1324 , e124 , e234  = +1,
e123 , e134 , −e4 , e123 ∧ (−e4 ) = +e1324 , e123 , e134  = +1,
e124 , e234 , +e3 , e124 ∧ (+e3 ) = +e1324 , e124 , e234  = +1.

The Lemma now follows. 2

The following example will be useful in what follows; to simplify certain expressions, set
∂i := ∂xi and fi := ∂i f .

Lemma 4.9 Let (x 1 , x 2 , x 3 , x 4 ) be the usual coordinates on R4 . Let f ∈ C ∞ (R4 ). Let the metric
be determined by g(∂1 , ∂3 ) = 1 and by g(∂2 , ∂4 ) = e2f . Let J+ be the standard para-complex structure
J+ ∂1 = ∂1 , J+ ∂2 = ∂2 , J+ ∂3 = −∂3 , and J+ ∂4 = −∂4 . Let ∇ be the Weyl connection determined by
taking φ = 21 J+ δg + . Then ∇J+ = 0.

Proof. The (possibly) non-zero Christoffel symbols of g ∇ are given by:

g(g ∇∂1 ∂2 , ∂4 ) = g(g ∇∂2 ∂1 , ∂4 ) = g(g ∇∂1 ∂4 , ∂2 ) = g(g ∇∂4 ∂1 , ∂2 ) = f1 e2f ,
g(g ∇∂3 ∂2 , ∂4 ) = g(g ∇∂2 ∂3 , ∂4 ) = g(g ∇∂3 ∂4 , ∂2 ) = g(g ∇∂4 ∂3 , ∂2 ) = f3 e2f ,
g(g ∇∂4 ∂4 , ∂2 ) = 2f4 e2f ,
g(g ∇∂2 ∂2 , ∂4 ) = 2f2 e2f ,
 4.3. (PARA)-KÄHLER–WEYL STRUCTURES IF m = 4 113
g(g ∇∂2 ∂4 , ∂1 ) = g(g ∇∂4 ∂2 , ∂1 ) = −f1 e2f ,
g(g ∇∂2 ∂4 , ∂3 ) = g(g ∇∂4 ∂2 , ∂3 ) = −f3 e2f .

Consequently, the (possibly) non-zero covariant derivatives are given by:

g∇ ∂ = g ∇∂2 ∂1 = f1 ∂2 , g∇ ∂ = g ∇∂4 ∂1 = f1 ∂4 ,
∂1 2 ∂1 4
g∇ ∂ = g ∇∂2 ∂3 = f3 ∂2 , g∇ ∂ = g ∇∂4 ∂3 = f3 ∂4 ,
∂3 2 ∂3 4
g∇ ∂ = 2f4 ∂4 , g∇ = 2f2 ∂2 ,
∂4 4 ∂ 2 ∂2
g∇ ∂ = g∇ ∂ = −f1 e2f ∂ 3 − f3 e ∂1 .
2f
∂2 4 ∂4 2

Since g ∇∂1 and g ∇∂3 are diagonal, they commute with J+ so g ∇∂1 (J+ ) = g ∇∂3 (J+ ) = 0. Thus:

(g ∇∂2 J+ )∂1 = (Id −J+ )g ∇∂2 ∂1 = (Id −J+ )f1 ∂2 = 0,

(g ∇∂2 J+ )∂2 = (Id −J+ )g ∇∂2 ∂2 = (Id −J+ )2f2 ∂2 = 0,
(g ∇∂2 J+ )∂3 = (− Id −J+ )g ∇∂2 ∂3 = (− Id −J+ )f3 ∂2 = −2f3 ∂2 ,
(g ∇∂2 J+ )∂4 = (− Id −J+ )g ∇∂2 ∂4 = (− Id −J+ )e2f (−f1 ∂3 − f3 ∂1 ) = 2f3 e2f ∂1 ,
(g ∇∂4 J+ )∂1 = (Id −J+ )g ∇∂4 ∂1 = (Id −J+ )f1 ∂4 = 2f1 ∂4 ,
(g ∇∂4 J+ )∂2 = (Id −J+ )g ∇∂4 ∂2 = (Id −J+ )e2f (−f1 ∂3 − f3 ∂1 ) = −2e2f f1 ∂3 ,
(g ∇∂4 J+ )∂3 = (− Id −J+ )g ∇∂4 ∂3 = (− Id −J+ )f3 ∂4 = 0,
(g ∇∂4 J+ )∂4 = (− Id −J+ )g ∇∂4 ∂4 = (− Id −J+ )2f4 ∂4 = 0.

Set e1 = dx 1 , e2 = ef dx 2 , e3 = dx 3 , and e4 = ef dx 4 . This is a hyperbolic basis for the cotangent

space. We apply Lemma 4.8 to compute:

 + = (−e1 ∧ e3 − e2 ∧ e4 ) = dx 1 ∧ dx 3 + e2f dx 2 ∧ dx 4 ,
d  + = 2f1 e2f dx 1 ∧ dx 2 ∧ dx 4 − 2f3 e2f dx 2 ∧ dx 3 ∧ dx 4 ,
(4.3.a)
δg + = −  d  + = −2f1 dx 1 + 2f3 dx 3 ,
φ = 21 J+ δg + = −f1 dx 1 − f3 dx 3 , and φ # = −f1 ∂3 − f3 ∂1 .

Let ij := φ(∂i )∂j + φ(∂j )∂i − g(∂i , ∂j )φ # = (φ ∇ − g ∇)∂i ∂j . Then:

11 = −2f1 ∂1 , 12 = −f1 ∂2 ,

13 = (−f1 ∂3 − f3 ∂1 ) + (f1 ∂3 + f3 ∂1 ) = 0,
14 = −f1 ∂4 , 22 = 0, 23 = −f3 ∂2 ,
24 = e2f (f1 ∂3 + f3 ∂1 ), 33 = −2f3 ∂3 , 34 = −f3 ∂4 , 44 = 0 .
114 4. (PARA)-KÄHLER–WEYL MANIFOLDS
Since (∂1 ) and (∂3 ) are diagonal, [(∂1 ), J+ ] = [(∂3 ), J+ ] = 0. We compute:

[(∂2 ), J+ ]∂1 = (Id −J+ )12 = 0,

[(∂2 ), J+ ]∂2 = (Id −J+ )22 = 0,
[(∂2 ), J+ ]∂3 = (− Id −J+ )23 = 2f3 ∂2 ,
[(∂2 ), J+ ]∂4 = (− Id −J+ )24 = −2f3 e2f ∂1 ,
[(∂4 ), J+ ]∂1 = (Id −J+ )14 = −2f1 ∂4 ,
[(∂4 ), J+ ]∂2 = (Id −J+ )24 = 2f1 e2f ∂3 ,
[(∂4 ), J+ ]∂3 = (− Id −J+ )34 = 0,
[(∂4 ), J+ ]∂4 = (− Id −J +)44 = 0 .

We now observe that [g ∇, J+ ] + [, J+ ] = 0. Consequently, φ ∇J+ = 0. 2

PARA-KÄHLER–WEYL STRUCTURES FOR PARA-HERMITIAN METRICS

2 be the vector space of symmetric 2-cotensors
Adopt the notation of Equation (1.5.d) and let S−
∗
ω so that J+ ω = −ω. Let (M, g, J+ ) be the germ of a 4-dimensional para-Hermitian manifold
and let P be a point of M. We can choose local coordinates that are centered at P so J+ ∂1 = ∂1 ,
J+ ∂2 = ∂2 , J+ ∂3 = −∂3 , J+ ∂4 = −∂4 , and so

g = dx 1 ◦ dx 3 + dx 2 ◦ dx 4 + ε ,

where ε ∈ C ∞ (S−2 ) satisfies ε(0) = 0. Theorem 4.2 in the para-Hermitian setting will follow from

the following result:

Lemma 4.10 Adopt the notation established above to define the germ of a para-Hermitian manifold
Mε at the origin. If φ := 21 J+ δ + , then (φ ∇J+ )(0) = 0.

Proof. Since only the 1-jets of ε are relevant in examining (φ ∇J+ )(0), this is a linear problem and
we may take ε ∈ S−2 ⊗ V ∗ and express thereby

g = g0 + x i ε(ei ) .
2 ⊗ V ∗ → End{V } ⊗ V ∗ or, equivalently,
Then ε → (φ ∇J+ )(0) defines a linear map E : S−

E : S−
2
→ Hom{V ∗ , End{V } ⊗ V ∗ } .

Lemma 4.9 shows that E (dx 2 ◦ dx 4 ) = 0. Permuting the indices 1 ↔ 2 and 3 ↔ 4 then yields
E (dx 1 ◦ dx 3 ) = 0. The question is invariant under the action of the para-unitary group; we must
preserve J+ and we must preserve the inner product at the origin. Define a unitary transformation
T by setting:

T (e1 ) = e1 + ae2 , T (e2 ) = e2 , T (e3 ) = e3 , T (e4 ) = e4 − ae3 .

 4.3. (PARA)-KÄHLER–WEYL STRUCTURES IF m = 4 115
Then T (e1 ∧ e3 )
= e1 ◦ e3 + ae2 ◦ e3 . Consequently, E (e2 ◦ e3 ) = 0. Permuting the indices 1 ↔ 2
and 3 ↔ 4 then yields E (e1 ◦ e4 ) = 0 as well. Since

2
S− = Span{e1 ◦ e3 , e1 ◦ e4 , e2 ◦ e3 , e2 ◦ e4 } ,

E = 0. This completes the proof and, as a consequence, Theorem 4.2 in the para-Hermitian setting
follows. 2

ANALYTIC CONTINUATION
We will derive Theorem 4.2 in the complex setting from Lemma 4.10 using analytic con-
tinuation. Let V = R4 with the usual basis {e1 , e2 , e3 , e4 } and coordinates {x 1 , x 2 , x 3 , x 4 }.
Let x := x 1 e1 + x 2 e2 + x 3 e3 + x 4 e4 . We consider:
   
S := S 2 ⊗R C ⊕ (V ∗ ⊗R S 2 ) ⊗R C .

Let J+ ∈ M2 (C) be a complex 2 × 2 matrix with J+2 = Id and Tr{J+ } = 0. Let:

S (J+ ) := {(g0 , g1 ) ∈ S : det{g0 − J+∗ g0 } = 0} . (4.3.b)

For (g0 , g1 ) ∈ S (J+ ), define:

x )(X, Y ) := 21 {g0 (X, Y ) − g0 (J+ X, J+ Y )}

g(
 4
+ x i · 21 {g1 (ei , X, Y ) − g1 (ei , J+ X, J+ Y )} .
i=1

By Equation (4.3.b), this is non-degenerate at 0 and defines a complex metric on some neighborhood
of 0 so J+∗ g = −g. Let g ∇ be the complex Levi-Civita connection:

g
∇∂i ∂j = 21 g k {∂i gj  + ∂j gi − ∂x gij }∂k .

Then g ∇ is a torsion-free connection on TC M := T M ⊗R C. The para-Kähler form is defined by

setting + (x, y) = g(x, J+ y) and we have

δ + = d + and φ := 21 J+ δg +.

We then use φ to define a complex Weyl connection φ ∇ on TC M and define a holomorphic map from
S (J+ ) to V := M4 (C) ⊗R V ∗ by setting:

E (g0 , g1 ; J+ ) := φ ∇(J+ )|x=0 .

116 4. (PARA)-KÄHLER–WEYL MANIFOLDS
Lemma 4.11 Let J+ ∈ M4 (C) with J+2 = Id and Tr{J+ } = 0. Suppose (g0 , g1 ) ∈ S (J+ ).

1. If J+ is real and if (g0 , g1 ) is real, then E (g0 , g1 ; J+ ) = 0.

2. If J+ is real and if (g0 , g1 ) is complex, then E (g0 , g1 ; J+ ) = 0.
3. If J+ is complex and if (g0 , g1 ) is complex, then E (g0 , g1 ; J+ ) = 0.

Proof. Assertion (1) of the Lemma is simply a restatement of Lemma 4.10. We argue as follows
to prove Assertion (2) of the Lemma. S (J+ ) is an open dense subset of S and inherits a natural
holomorphic structure thereby. Assume that J+ is real. The map E is a holomorphic map from
S (J+ ) to V. By Assertion (1), E (g0 , g1 ; J+ ) vanishes if (g0 , g1 ) is real. Thus, by the identity theorem,
E (g0 , g1 ; J+ ) vanishes for all (g0 , g1 ) ∈ S (J+ ). Since we have removed the assumption that (g0 , g1 )
is real, Assertion (2) of Lemma 4.11 follows.
We complete the proof of Lemma 4.11 by removing the assumption that J+ is real.The complex
general linear group GL4 (C) acts on the structures involved by change of basis (i.e., conjugation). Let
(g0 , g1 ) ∈ S (J+ ) where J+ is real and Tr{J+ } = 0. We consider the real and complex orbits:

OR (g0 , g1 ; J+ ) := GL4 (R) · (g0 , g1 ; J+ ) and OC (g0 , g1 ; J+ ) := GL4 (C) · (g0 , g1 ; J+ ) .

Let F (A) := E (A · (g0 , g1 ; J+ )) define a holomorphic map from GL4 (C) to V. By Assertion (2),
F vanishes on GL4 (R). Thus, by the identity theorem, F vanishes on GL4 (C) or, equivalently, E
vanishes on the orbit space OC (g0 , g1 ; J+ ). Given any J+ ∈ M4 (C) with J+2 = Id and Tr{J+ } = 0,
we can choose A ∈ GL4 (C) so that A · J+ is real. The general case now follows from Assertion (2).
2

Let (M, g, J− ) be a 4-dimensional pseudo-Hermitian manifold of dimension four. Fix a point

P of M. Since J− is integrable, we may choose local coordinates (x 1 , x 2 , x 3 , x 4 ) so the matrix of J−
relative to the coordinate frame {∂i } is constant. Define a Weyl connection with associated 1-form
given by φ = − 21 J− δ − . Only the 0-jet and the 1-jet of the metric
√ play a role in the computation
of (φ ∇J− )(P ). So we may assume g = g(g0 , g1 ). We set J+ = −1J− . We have that
√ √ √
J+2 = −1J− −1J √ − = −J−√= Id,
2
Tr{J+ } = −1 Tr{J− } = 0,
J+∗ (g)(X, Y ) = g( −1J− X, −1J− Y ) = −g(J− X, J− Y ) = −g(X, Y ) ,
√
so J+∗ (g) = −g and (g0 , g1 ) ∈ S (J+ ). Finally, since J− = − −1J+ , we have
√
− = − −1 + , √ √
φJ− = − 21 J− δg − = − 21 (− −1J+ )δg (− −1 +) = 21 J+ δg + = φJ+ .

We apply Lemma 4.11 to complete the proof of Theorem 4.2. "

!
 4.4. (PARA)-KÄHLER–WEYL LIE GROUPS IF m = 4 117

4.4 (PARA)-KÄHLER–WEYL LIE GROUPS IF m = 4

In Section 4.4, we shall establish Theorem 4.3. We treat both the para-Hermitian setting and the
Hermitian setting; we do not deal with pseudo-Hermitian structures of signature (2, 2). We shall
describe two different Lie algebras. For generic values of the parameters, the Lie algebra in the para-
complex setting is modeled on A2,2 ⊕ A2,2 and the Lie algebra in the complex setting is modeled
on A4,12 in the classification given by Patera et al. in [176]. There is a duality between these two Lie
algebras which we describe as follows.
Definition 4.12
1. Let (V , ·, ·, J+ ) be a para-Hermitian vector space of signature (2, 2). Choose a hyperbolic
basis { 1 , 2 , 3 , 4 } for V and corresponding dual basis { 1 , 2 , 3 , 4 } for V ∗ so that
J+ is diagonalized:
 1 , 3  = 1,  2 , 4  = 1,  1 , 3  = 1,  2 , 4  = 1,
J+ 1 = 1 , J+ 2 = 2 , J+ 3 = − 3 , J+ 4 = − 4 ,
J+ 1 = 1 , J+ 2 = 2 , J+ 3 = − 3 , J+ 4 = − 4 .
Let ε1 , ε̃1 , α2 , α̃2 , α3 , α̃3 be real parameters. Define a bracket [·, ·] on V by setting:
[ 1, 2] = ε1 1 , [ 1, 3] = 0, [ 1, 4] = α3 1,
[ 2, 3] = −α̃3 3 , [ 2, 4] = α2 1 − α̃2 3, [ 3, 4] = ε̃1 3.

2. Let (V , ·, ·, J− ) be a positive definite (i.e., signature (0, 4)) Hermitian vector space. Choose
an orthogonal basis {ei } for V so that:
e1 , e1  = 2, e2 , e2  = 2, e3 , e3  = 2, e4 , e4  = 2,
e1 , e1  = 21 , e2 , e2  = 21 , e3 , e3  = 21 , e4 , e4  = 21 ,
J− e1 = e2 , J− e2 = −e1 , J− e3 = e4 , J− e4 = −e3 ,
J− e1 = −e2 , J− e2 = e1 , J− e3 = −e4 , J− e4 = e3 .
We define a complex basis {Z1 , Z2 , Z̄1 , Z̄2 } for VC := V ⊗R C and the corresponding complex
dual basis {Z 1 , Z 2 , Z̄ 1 , Z̄ 2 } for the complex dual space VC∗ by setting:
√ √ √ √
Z1 = 21 (e1 − −1e2 ), Z2 = 21 (e3 − −1e4 ), Z 1 = (e1 + −1e2 ), Z 2 = (e3 + −1e4 ),
√ √ √ √
Z̄1 = 21 (e1 + −1e2 ), Z̄2 = 21 (e3 + −1e4 ), Z̄ 1 = (e1 − −1e2 ), Z̄ 2 = (e3 − −1e4 ).
Then we have
Z1 , Z̄1  = 1, Z 1 , Z̄ 1  = 1, Z2 , Z̄2  = 1, Z 2 , Z̄ 2  = 1,
√ √ √ √
J− Z1 = −1Z1 , J− Z̄1 = − −1Z̄1 , J− Z2 = −1Z2 , J− Z̄2 = − −1Z̄2 ,
√ √ √ √
J− Z 1 = −1Z 1 , J− Z̄ 1 = − −1Z̄ 1 , J− Z 2 = −1Z 2 , J− Z̄ 2 = − −1Z̄ 2 .
Let ε1 , α2 , α3 be complex parameters. Define a complex bracket [·, ·] on VC that arises from
a corresponding real bracket on V by setting:
118 4. (PARA)-KÄHLER–WEYL MANIFOLDS
[Z1 , Z2 ] = ε1 Z1 , [Z1 , Z̄1 ] = 0, [Z1 , Z̄2 ] = α3 Z1 ,
[Z2 , Z̄1 ] = −ᾱ3 Z̄1 , [Z2 , Z̄2 ] = α2 Z1 − ᾱ2 Z̄1 , [Z̄1 , Z̄2 ] = ε̄1 Z̄1 .
3. To have a common notation, let:

para-complex setting: ξ1 = 1 , ξ2 = 2 , ξ3 = 3 , ξ4 = 4 , ε+ = +1, √

complex setting: ξ1 = Z1 , ξ2 = Z2 , ξ3 = Z̄1 , ξ4 = Z̄2 , ε− = −1 .

The {ξi } form a real (resp. complex) basis for V (resp. V ⊗R C) so that

J± ξ1 = ε± ξ1 , J± ξ2 = ε± ξ2 , J± ξ3 = −ε± ξ3 , J± ξ4 = −ε± ξ4 ,
J± ξ 1 = ε± ξ 1 , J± ξ 2 = ε± ξ 2 , J± ξ 3 = −ε± ξ 3 , J± ξ 4 = −ε± ξ 4 .

In the complex setting, we let ε̃1 = ε̄1 , α̃2 = ᾱ2 , and α̃3 = ᾱ3 .

Lemma 4.13

1. The bracket [·, ·] satisfies the Jacobi identity; let G be the associated simply connected Lie group. J±
defines an integrable left-invariant (para)-complex structure on G and ·, · defines a left-invariant
(para)-Hermitian metric on (G, J± ).

2. Use Theorem 4.2 to define a corresponding left-invariant (para)-Kähler–Weyl connection ∇. The

associated alternating Ricci tensor of this connection is given by:
ρa = α̃2 ε1 ξ 1 ∧ ξ 2 + α̃2 α3 ξ 1 ∧ ξ 4 − α2 α̃3 ξ 2 ∧ ξ 3 + α2 ε̃1 ξ 3 ∧ ξ 4 .

Proof. We verify the bracket [·, ·] satisfies the Jacobi identity:

[[ξ1 , ξ2 ], ξ3 ] + [[ξ2 , ξ3 ], ξ1 ] + [[ξ3 , ξ1 ], ξ2 ] = ε1 [ξ1 , ξ3 ] − α̃3 [ξ3 , ξ1 ] + 0 = 0,
[[ξ1 , ξ2 ], ξ4 ] + [[ξ2 , ξ4 ], ξ1 ] + [[ξ4 , ξ1 ], ξ2 ]
= ε1 [ξ1 , ξ4 ] + [α2 ξ1 − α̃2 ξ3 , ξ1 ] − α3 [ξ1 , ξ2 ] = ε1 α3 ξ1 + 0 − α3 ε1 ξ1 = 0,
[[ξ1 , ξ3 ], ξ4 ] + [[ξ3 , ξ4 ], ξ1 ] + [[ξ4 , ξ1 ], ξ3 ] = 0 + ε̃1 [ξ3 , ξ1 ] − α3 [ξ1 , ξ3 ] = 0,
[[ξ2 , ξ3 ], ξ4 ] + [[ξ3 , ξ4 ], ξ2 ] + [[ξ4 , ξ2 ], ξ3 ]
= −α̃3 [ξ3 , ξ4 ] + ε̃1 [ξ3 , ξ2 ] − [α2 ξ1 − α̃2 ξ3 , ξ3 ] = −α̃3 ε̃1 ξ3 + ε̃1 α̃3 ξ3 + 0 = 0.
√
Let V+ := Span{ξ1 , ξ2 } and V− := Span{ξ3 , ξ4 } be the ±1 (resp. ± −1) eigenspaces of J+ (resp.
J− ) . Then
[V+ , V+ ] ⊂ V+ and [V− , V− ] ⊂ V−
so J± is integrable by Theorem 1.13; this establishes Assertion (1).
 4.4. (PARA)-KÄHLER–WEYL LIE GROUPS IF m = 4 119
We work in the para-complex setting for the moment to establish Assertion (2). Express the
para-Kähler form as:
+ := −(ξ ∧ ξ + ξ ∧ ξ ).
1 3 2 4

By Equation (1.1.c), we have dξ i ( j, k) = −ξ i ([ j, k ]). Consequently:

dξ 1 = −ε1 ξ 1 ∧ ξ 2 − α3 ξ 1 ∧ ξ 4 − α2 ξ 2 ∧ ξ 4 , dξ 2 = 0,
dξ 3 = α̃3 ξ 2 ∧ ξ 3 + α̃2 ξ 2 ∧ ξ 4 − ε̃1 ξ 3 ∧ ξ 4 , dξ 4 = 0.

We apply Lemma 4.8 to compute  relative to this basis simply replacing ei by ξ i . Since δ = −  d
and  + = − + , we have:
δ + =−d  + = −  d(ξ 1 ∧ ξ 3 + ξ 2 ∧ ξ 4 )
= {(ε1 ξ 1 ∧ ξ 2 + α3 ξ 1 ∧ ξ 4 + α2 ξ 2 ∧ ξ 4 ) ∧ ξ 3 }
+  {ξ 1 ∧ (α̃3 ξ 2 ∧ ξ 3 + α̃2 ξ 2 ∧ ξ 4 − ε̃1 ξ 3 ∧ ξ 4 )}
= α̃2 ξ 1 + (−ε1 − α̃3 )ξ 2 − α2 ξ 3 + (ε̃1 + α3 )ξ 4 ,
dJ+ δ + = d{α̃2 ξ 1 + (−ε1 − α̃3 )ξ 2 + α2 ξ 3 − (ε̃1 + α3 )ξ 4 }
= −α̃2 ε1 ξ 1 ∧ ξ 2 − α̃2 α3 ξ 1 ∧ ξ 4 − α̃2 α2 ξ 2 ∧ ξ 4
+α2 α̃3 ξ 2 ∧ ξ 3 + α2 α̃2 ξ 2 ∧ ξ 4 − α2 ε̃1 ξ 3 ∧ ξ 4 .
By Theorem 4.2, φ+ = 21 J+ δ + . By Equation (1.2.s), ρa = −2dφ+ . Consequently,√ we have
ρa = −dJ
√ + δ + and the desired result follows. In the complex setting, J− = −1J+ and
− = −1 + and ρ a is unchanged as
√ √
φ− = − 21 J− δ − = − 21 −1 −1J+ δ + = φ+ . 2

Let SO(2∓ , 0) and SO(2± ) denote the special orthogonal group of these inner product
spaces; in the positive definite setting, SO(2+,0 ) ≈ S 3 and SO(2− ) ≈ S 1 .

Lemma 4.14 Let V be a real vector space of dimension four.

1. Let (·, ·, J+ ) be a para-Hermitian structure on V . Then 2−,0 has signature (2, 1) and 2+
has signature (1, 1). Every orbit of the action of U+ on 2−,0 ⊕ 2+ contains a representative
perpendicular to 1 ∧ 3 − 2 ∧ 4 .

2. Let (·, ·, J− ) be a positive definite Hermitian structure on V . Then the natural action of
the unitary group U− on 2+,0 ⊕ 2− defines a surjective group homomorphism π from U− to
SO(2+,0 ) ⊕ SO(2− ).
120 4. (PARA)-KÄHLER–WEYL MANIFOLDS
Proof. The para-Kähler form is given by + = − 1 ∧ 3 − 2 ∧ 4 . We define an orthogonal
basis {θ1 , θ2 , θ3 } for 2−,0 and an orthogonal basis {θ4 , θ5 } for 2+ by setting:

θ1 := 1 ∧ 3 − 2 ∧ 4, θ2 := 1 ∧ 4 + 2 ∧ 3, θ3 := 1 ∧ 4 − 2 ∧ 3,

θ4 := 1 ∧ 2 + 3 ∧ 4, θ5 := 1 ∧ 2 − 3 ∧ 4.

We show that 2−,0 has signature (2, 1) and that 2+ has signature (1, 1) by computing:

θ1 , θ1  = −2, θ2 , θ2  = −2, θ3 , θ3  = 2, θ4 , θ4  = 2, θ5 , θ5  = −2,

θi , θj  = 0 for i  = j .

Define an element of U+ by setting:

Tθ 1
:= cos θ 1
+ sin θ 2
, Tθ 3
:= cos θ 3
+ sin θ 4
,
Tθ 2
:= − sin θ 1
+ cos θ 2
, Tθ 4
:= − sin θ 3
+ cos θ 4
.

We then have that:

Tθ θ1 = cos(2θ )θ1 + sin(2θ)θ2 , Tθ θ2 = − sin(2θ )θ1 + cos(2θ )θ2 , Tθ θi = θi for i ≥ 3 .

We complete the proof of Assertion (1) by performing an appropriate rotation in the plane spanned
by {θ1 , θ2 } to eliminate the coefficient of θ1 .
We now establish Assertion (2). We have

2+,0 ⊗R C = SpanC {Z 1 ∧ Z̄ 1 − Z 2 ∧ Z̄ 2 , Z 1 ∧ Z̄ 2 , Z̄ 1 ∧ Z 2 },
2− ⊗R C = SpanC {Z 1 ∧ Z 2 , Z̄ 1 ∧ Z̄ 2 } .

We construct generators for the unitary group. First define Tθ ∈ U− by:

Tθ (e1 ) = cos θe1 − sin θe2 , Tθ (e2 ) = sin θ e1 + cos θe2 ,
Tθ (e3 ) = cos θe3 − sin θe4 , Tθ (e4 ) = sin θ e3 + cos θ e4 ,
√ √ √
Tθ (Z 1 ) = e −1θ Z 1 , Tθ (Z 2 ) = e −1θ Z 2 , Tθ (Z̄ 1 ) = e− −1θ Z̄ 1 ,
√
Tθ (Z̄ 2 ) = e− −1θ Z̄ 2 , Tθ (Z 1 ∧ Z̄ 1 ) = Z 1 ∧ Z̄ 1 , Tθ (Z 2 ∧ Z̄ 2 ) = Z 2 ∧ Z̄ 2 ,
Tθ (Z 1 ∧ Z̄ 2 ) = Z 1 ∧ Z̄ 2 , Tθ (Z̄ 1 ∧ Z 2 ) = Z̄ 1 ∧ Z 2 ,
√ √
Tθ (Z 1 ∧ Z 2 ) = e2 −1θ Z 1 ∧ Z 2 , Tθ (Z̄ 1 ∧ Z̄ 2 ) = e−2 −1θ Z̄ 1 ∧ Z̄ 2 .
Thus, Tθ is the identity on 2+,0 ⊗R C and acts as a rotation through an angle of 2θ on the underlying
real vector space 2− . This realizes the SO(2) on 2− action and reduces the proof of Assertion (2)
to examining the action on 2+,0 . Next, define T̃θ ∈ U− by setting:

T̃θ (e1 ) = cos θe1 − sin θe2 , T̃θ (e2 ) = sin θ e1 + cos θe2 ,
T̃θ (e3 ) = cos θe3 + sin θe4 , T̃θ (e4 ) = − sin θ e3 + cos θe4 ,
 4.4. (PARA)-KÄHLER–WEYL LIE GROUPS IF m = 4 121
√ √ √
T̃θ (Z 1 ) = e −1θ Z 1 , T̃ (Z 2 ) = e− −1θ Z 2 , T̃θ (Z̄ 1 ) = e− −1θ Z̄ 1 ,
θ
√
T̃θ (Z̄ 2 ) = e −1θ Z̄ 2 , T̃θ (Z 1 ∧ Z̄ 1 ) = Z 1 ∧ Z̄ 1 , T̃θ (Z 2 ∧ Z̄ 2 ) = Z 2 ∧ Z̄ 2 ,
√ √
T̃θ (Z 1 ∧ Z̄ 2 ) = e2 −1θ Z 1 ∧ Z̄ 2 , T̃θ (Z̄ 1 ∧ Z 2 ) = e−2 −1θ Z̄ 1 ∧ Z 2 ,
T̃θ (Z 1 ∧ Z 2 ) = Z 1 ∧ Z 2 , T̃θ (Z̄ 1 ∧ Z̄ 2 ) = Z̄ 1 ∧ Z̄ 2 .
Consequently, T̃θ is the identity on 2− ⊗R C, is the identity on {Z 1 ∧ Z̄ 1 − Z 2 ∧ Z̄ 2 } · C, and acts
as a rotation through an angle of 2θ on the real vector space
√
SpanR {Z 1 ∧ Z̄ 2 + Z 1 ∧ Z̄ 2 , −1(Z 1 ∧ Z̄ 2 − Z 1 ∧ Z̄ 2 )} .

We complete the proof of Assertion (2) by defining Ťφ ∈ U− :

Ťφ (e1 ) = cos φe1 + sin φe3 , Ťφ (e2 ) = cos φe2 + sin φe4 ,
Ťφ (e3 ) = cos φe3 − sin φe1 , Ťφ (e4 ) = cos φe4 − sin φe2 ,
Ťφ (Z 1 ) = cos φZ 1 + sin φZ 2 , Ťφ (Z 2 ) = cos φZ 2 − sin φZ 1 ,
Ťφ (Z̄ 1 ) = cos φ Z̄ 1 + sin φ Z̄ 2 , Ťφ (Z̄ 2 ) = cos φ Z̄ 2 − sin φ Z̄ 1 ,
√ √
Ťφ ( −1(Z 1 ∧ Z̄ 1 − Z 2 ∧ Z̄ 2 )) = (cos2 φ − sin2 φ) −1(Z 1 ∧ Z̄ 1 − Z 2 ∧ Z̄ 2 )
√
+2 cos φ sin φ −1(Z 1 ∧ Z̄ 2 − Z̄ 1 ∧ Z 2 ),
√ √
Ťφ ( −1(Z 1 ∧ Z̄ 2 − Z̄ 1 ∧ Z 2 )) = −2 cos φ sin φ −1(Z 1 ∧ Z̄ 1 − Z 2 ∧ Z̄ 2 )
√
+(cos2 φ − sin2 φ) −1(Z 1 ∧ Z̄ 2 − ∧Z̄ 1 ∧ Z 2 ),
Ťφ (Z 1 ∧ Z̄ 2 + Z̄ 1 ∧ Z 2 ) = Z 1 ∧ Z 2 + Z̄ 1 ∧ Z 2 ,
Ťφ (Z 1 ∧ Z 2 ) = Z 1 ∧ Z 2 , and Ťφ (Z̄ 1 ∧ Z̄ 2 ) = Z̄ 1 ∧ Z̄ 2 .
Consequently, Ťφ is the identity on 2− , is the identity on (Z 1 ∧ Z̄ 2 + Z̄ 1 ∧ Z 2 ) · C, and acts as a
rotation through an angle of 2φ on
√ √ 
SpanR −1(Z 1 ∧ Z̄ 1 − Z 2 ∧ Z̄ 2 ), −1(Z 1 ∧ Z̄ 2 − Z̄ 1 ∧ Z 2 ) .

The elements {T̃θ , Ťφ } generate SO(2+,0 ) and fix 2− . Assertion (2) follows. 2

THE PROOF OF THEOREM 4.3

The set of geometrically representable tensors is invariant under the action of the structure group
U+ . Let  ∈ 2−,0 ⊕ 2+ . By Lemma 4.14 (1), we may assume:

 = μ12 1
∧ 2
+ μ14 1
∧ 4
+ μ23 2
∧ 3
+ μ34 3
∧ 4
;

in other words, there is no 1 ∧ 3 − 2 ∧ 4 term. We consider the Lie group of Lemma 4.13
and set α2 = α̃2 = 1. The remaining parameters are then determined; we complete the proof by
taking:
122 4. (PARA)-KÄHLER–WEYL MANIFOLDS
ε1 = μ12 , α3 = μ14 , α̃3 = −μ23 , ε̃1 = μ34 .
Next we treat the complex setting. Let  = +,0 + − ∈ 2+,0 ⊕ 2− . We wish to show  is
geometrically representable. The question of representability is invariant under the action of U− or,
equivalently, by Lemma 4.14 (2), under the action of SO(2+,0 ) ⊕ SO(2− ). Thus, only the norms
|+,0 | and |− | are relevant in establishing Theorem 4.3 (2). We apply Lemma 4.13 (2) to compute
ρa and to see
|+,0 |2 = 2|α2 |2 |α3 |2 and |− |2 = 2|α2 |2 |ε1 |2 .
If we set α2 = 1, we may then adjust α3 and ε1 to obtain arbitrary norms for +,0 and − and
complete the proof. "
!

4.5 (PARA)-KÄHLER–WEYL TENSORS IF m = 4

This section is devoted to the proof of Theorem 4.4. We restate the short exact sequence of Equa-
tion (4.1.b):
0 → K±,R → K±,W → ρa {K±,W } → 0 .
If η is an irreducible U± module and if ξ is a submodule of ⊗4 V ∗ , let nη (ξ ) be the multiplicity with
which η appears in the decomposition of ξ given in Lemma 1.16. Then:

nη (K±,W ) = nη (K±,R ) + nη (ρa {K±,W }) .

We apply Theorem 1.25. If η = Wi,± for i ∈ {1, 2, 3, 6, 7, 8, 10}, then nη (2 ) = 0. Consequently,
we may estimate:

1 if i = 1, 2, 3
= nη (K±,R ) ≤ nη (K±,R ) + nη (ρa (K±,W ))
0 if i = 6, 7, 8, 10
1 if i = 1, 2, 3
= nη (K±,W ) ≤ nη (K±,R ) + nη (2 ) = +0.
0 if i = 6, 7, 8, 10
This shows that all of the inequalities in the above display must have been equalities and de-
termines nη (K±,W ) for these representations. Thus, only the multiplicities of the representations
{ ± · R, 2∓,0 , 2± } are at issue. The analysis performed previously in Section 4.2 shows that the
module ± · R plays no role. Furthermore, the analysis performed above shows that:

0 ≤ n2 (K±,W ) ≤ 1 and 0 ≤ n2 (K±,W ) ≤ 1 .

∓,0 ±

We must consider the module η := 2± or η := 2∓,0 in dimension m = 4. We have that

nη (K±,W ) ≤ 1. Thus, if we can exhibit a non-trivial element of

W±,12 ∩ K±,W or {W±,9 ⊕ W±,13 } ∩ K±,W ,

we will have nη (K±,W ) = 1. The existence of such a non-trivial element follows from Theorem 4.3
in the Hermitian and the para-Hermitian settings. To deal with the pseudo-Hermitian setting in
 4.6. REALIZABILITY OF (PARA)-KÄHLER–WEYL TENSORS IF m = 4 123
signature (2, 2), we complexify. Let (V , ·, ·, J− ) be a Hermitian vector space of signature (0, 4)
and let A ∈ η(V , ·, ·, J− ) ⊂ K−,W (V , ·, ·, J− ) for η ≈ 2+,0 or η ≈ 2− . Let VC := V ⊗R C.
Extend ·, ·, J− , and A to be complex bilinear, complex linear, and complex multilinear, respectively.
Let:
√ √
V2,2 := SpanR { −1e1 , −1e2 , e3 , e4 } .

Let &(z) and '(z) denote the real and imaginary parts, respectively, of a complex number. Then
(·, ·, J− ) restricts to a pseudo-Hermitian almost complex structure on V2,2 of signature (2, 2).
Note that

&(A|V2,2 ) ∈ η(V2,2 , ·, ·, J− ) ∩ K−,W (V2,2 ),

'(A|V2,2 ) ∈ η(V2,2 , ·, ·, J− ) ∩ K−,W (V2,2 ) .

Since A|V2,2  = 0, at least one of these tensors is non-trivial and the desired conclusion follows. This
completes the proof of Theorem 4.4. "
!

4.6 REALIZABILITY OF (PARA)-KÄHLER–WEYL TENSORS

IF m = 4
This section is devoted to the proof of Theorem 4.5. We shall first show that any element of K±,W
is geometrically realizable by the germ of a suitable structure.
Let (V , ·, ·, J± ) be a (para)-Hermitian vector space. Extend J± to an integrable (para)-
complex structure on T V by identifying the tangent bundle T V with the trivial bundle V × V over
V . Let {ei } be a basis for V and let {x i } be the dual system of coordinates on V ; we normalize the
coordinate system so that

J± ∂x1 = ∂x2 , J± ∂x2 = ±∂x1 , J± ∂x3 = ∂x4 , J± ∂x4 = ±∂x3 ,

J± dx = ±dx , J± dx = dx , J± dx = ±dx , J± dx 4 = dx 3 .
1 2 2 1 3 4

Let εij := ei , ej ; we shall also let ε denote ·, ·. Let θ ∈ S±
2 ⊗ S 2 . Set:

gθ = (εij + θij k x k x  )dx i ◦ dx j .

Since gθ (0) = ε, gθ is non-degenerate near 0 and defines a pseudo-Riemannian metric on some

neighborhood O of 0 in V . Since θ ∈ S± 2 ⊗ S 2 and since J ∗ ε = ∓ε, J ∗ g = ∓g . Thus, (g , J )
± ± θ θ θ ±
defines the germ of a (para)-Hermitian manifold at 0 in V .
By Theorem 4.2, there is a unique Weyl connection θ ∇ so that (O, J± , gθ , θ ∇) is a (para)-
Kähler–Weyl manifold. Let (θ ) := ∇R(0). Since gθ = ε + O(| x |2 ), the first derivatives of the
metric play no role at 0 and  defines an equivariant linear map:

 : S±
2
⊗ S 2 → K±,W .
124 4. (PARA)-KÄHLER–WEYL MANIFOLDS
We wish to show  is surjective. By Theorem 1.25 and Theorem 4.4, there are U± module isomor-
phisms
K±,W ≈ K±,R ⊕ 2∓,0 ⊕ 2± and K±,R ≈ 1 ⊕ S∓,0 2
⊕ W±,3 (4.6.a)
where the five modules {1 , S∓,0
2 ,W
±,3 , ∓,0 , ± } which appear in Equation (4.6.a) are irreducible
2 2

and inequivalent U± modules. We will complete the proof by showing



K±,R ⊂ Range{} and ρa : Range{} → 2∓,0 ⊕ 2± → 0 .

THE MODULE K±,R

We begin our discussion with the following observation:

Lemma 4.15 If θ ∈ S±
2 ⊗ S 2 , let g := (ε + θ
θ ij ij k x x )dx ◦ dx .
k  i j

R(θ )(x, y, z, w) := θ (x, z, y, w) + θ(y, w, x, z) − θ(x, w, y, z) − θ(y, z, x, w) and

K± (θ )(x, y, z) := 2{θ (x, J± y, z, e ) + θ (y, J± z, x, e ) + θ(z, J± x, y, e )}x  .
g
Then gθ R(0)(x, y, z, w) = R(θ)(x, y, z, w) and d ±θ (x, y, z) = 2K± (θ )(x, y, z). Furthermore,
θ ∈ ker{K± } if and only if gθ is a (para)-Kähler metric.

Proof. We apply the Koszul formula of Equation (1.2.k) to see the Christoffel symbols of the Levi-
Civita connection are given by:
gθ
ij k = 21 (∂i gθ,j k + ∂j gθ,ik − ∂k gθ,ij ) = x  (θj ki + θikj  − θij k ) .

Since gθ ink = O(|

x |) and since gθ = ε + O(|
x |), we may raise indices to see
gθ
ij n = gθnk gθ ij k = εnk gθ ij k + O(|
x |2 ) .

This permits us to determine gθR(0) by computing:

gθ
Rij k p = ∂i gθ j k p − ∂j gθ ik p + O(|x |) = εp {∂i gθ j k − ∂j gθ ik } + O(|
x |),
gθ
Rij k = ∂i j k − ∂j ik + O(|
gθ gθ
x |)
= θkj i + θj ki − θj ki − θkij − θikj + θikj + O(| x |)
= θikj  + θj ik − θj ki − θij k + O(|x |) .
2 ⊗ S 2 , θ (x, J y, z, w) = −θ (y, J x, z, w). We complete the proof by computing:
Since θ ∈ S± ± ±
gθ 
± = 2 i,j θ(ei , J± ej , ek , e )x x dx ∧ dx ,
1 k  i j

g 
d ±θ = i,j,k θ (ei , J± ej , ek , e )x  dx i ∧ dx j ∧ dx k

= 2 i<j <k {θ(ei , J± ej , ek , e ) + θ (ej , J± ek , ei , e ) + θ (ek , J± ei , ej , e )}x 
×dx i ∧ dx j ∧ dx k . 2
 4.6. REALIZABILITY OF (PARA)-KÄHLER–WEYL TENSORS IF m = 4 125
Adopt the notation of Definition 4.12. By Equation (1.2.p) and Equation (1.4.b),

R ⊂ S 2 (2 (V ∗ )) and K±,R ⊂ S 2 (2± ) .

Let ξ ij k := (ξ i ∧ ξ j ) ◦ (ξ k ∧ ξ  ). After taking into account the Bianchi identity, we have:

K+,R ⊂ Span{ξ 1313 , ξ 1414 , ξ 2323 , ξ 2424 , ξ 1323 , ξ 1424 , ξ 1314 , ξ 2324 , ξ 1324 + ξ 1423 }.
There are nine elements in this basis; this is in accordance with the dimension count (see Tricerri
and Vanhecke [200] in the complex setting and Brozos-Vázquez, Gilkey, and Nikčević [37] in the
para-complex setting) that dim(1 ) = 1, dim(S∓,0 2 ) = 3, and dim(W
±,3 ) = 5.
We consider the following example. Let

ξ1 , ξ3  = ξ2 , ξ4  = 1 and θ = 21 (ξ 1 ◦ ξ 3 ) ⊗ (ξ 1 ◦ ξ 3 ) ∈ S±

2 ⊗ S2 .

The metric gθ is a real (para)-Kähler metric that takes the form of the product metric of
M1 × M2 , where M1 is a (para)-Riemann surface and where M2 is flat; we can also verify directly
from Lemma 4.15 that the (para)-Kähler form vanishes since K± (θ ) is a 3-form that is supported
on Span{e1 , e2 }. Furthermore, an application of Lemma 4.15 permits us to compute the curvature
tensor and show
R(θ) = ξ 1313 ∈ Range{K± } . (4.6.b)
There is an algebra acting that will be central to our treatment. Let

End± = {T ∈ End{V } : T J± = J± T } .
2 ⊗ S 2 , R, and 3 (V ∗ ) ⊗ V ∗ are modules over this algebra and the maps
The vector spaces S±
θ → R(θ ) and θ → K± (θ) of Lemma 4.15 are End± module morphisms. In particular, ker{K± }
is an End± module. On ker{K± }, we have  = R since φθ = ± 21 J± δ ± = 0. Thus:

R : ker{K± } → K±,R = 1 ⊕ S∓,0

2
⊕ W±,3 .

We shall always complexify when considering J− . Let

T (e1 ) = e1 + ae2 , T (e2 ) = e2 , T (e3 ) = e3 + ãe4 , T (e4 ) = e4 .

In the para-complex setting, we let a, ã ∈ R while in the complex setting, we let ã = ā ∈ C. Then
T ∈ End± . We apply T to the element of Equation (4.6.b) to conclude

T (ξ 1313 ) = ξ 1313 + 2aξ 1323 + 2ãξ 1314 + a 2 ξ 2323 + ã 2 ξ 1414 + 2a ã(ξ 1324 + ξ 1423 )
+2a 2 ãξ 2324 + 2a ã 2 ξ 1424 + 2a 2 ã 2 ξ 2424 ∈ Range{K± } . (4.6.c)

In the para-complex setting, we take ã = ±a. Since ξ 1313 ∈ Range{K± }, we have

2aξ 2313 ± 2aξ 1314 + O(a 2 ) ∈ Range{K± } so

ξ 2313 + O(a) ∈ Range{K+ } and ξ 1314 + O(a) ∈ Range{K+ } .
126 4. (PARA)-KÄHLER–WEYL MANIFOLDS
Since Range{K+ } is a linear subspace of a finite-dimensional vector space, it is closed. Thus, we may
take the limit as a → 0 to conclude
ξ 1323 ∈ Range{K+ } and ξ 1314 ∈ Range{K+ } .
√ √
In the complex setting, we take a = t and ã = t or a = −1t and ã = − −1t to conclude
2t{ξ
√
2313
+ ξ 1314 } + O(t 2 ) ∈ Range{K± }, and
2t −1(ξ 2313 − ξ 1314 ) + O(t 2 ) ∈ Range{K± } so
ξ 2313 ∈ Range{K− } and ξ 1314 ∈ Range{K− } .
Permuting the indices 1 ↔ 2 and 3 ↔ 4 defines an element of End± and shows
ξ 2424 ∈ Range{K± }, ξ 1424 ∈ Range{K± }, ξ 2324 ∈ Range{K± } .
Consequently Equation (4.6.c) yields:
a 2 ξ 2323 + 2a ã(ξ 1323 + ξ 1423 ) ∈ Range{K± } + ã 2 ξ 1414 .
A similar argument now shows
{ξ 2323 , ξ 1323 + ξ 1423 , ξ 1414 } ⊂ Range{K± } so Range{K± } = K±,R .

THE MODULE 2∓,0 ⊕ 2±

2 , we used the symmetric Ricci tensor ρ . We use the alternating Ricci
In the study of 1 and S∓,0 s
2
tensor to study ∓,0 and 2± . We begin by studying the para-complex setting. We adopt the notation
of Lemma 4.9. Let
J+ ∂1 = ∂1 , J+ ∂2 = ∂2 , J+ ∂3 = −∂3 , J+ ∂4 = −∂4 , g(∂1 , ∂3 ) = 1, g(∂2 , ∂4 ) = e2f .
We computed in Equation (4.3.a) that φ = −f1 dx 1 − f3 dx 3 and consequently,
ρa = −2dφ = 2{−f12 dx 1 ∧ dx 2 − f14 dx 1 ∧ dx 4 + f23 dx 2 ∧ dx 3 − f34 dx 3 ∧ dx 4 },
J+ ρa = 2{−f12 dx 1 ∧ dx 2 + f14 dx 2 ∧ dx 3 − f23 dx 1 ∧ dx 4 − f34 dx 3 ∧ dx 4 } .
Since J+ ρa is not equal to ±ρa , ρa has components in both 2−,0 and 2+ ; this provides the desired
example in this instance.
To study the complex setting, we consider the example:
J− ∂1 = ∂2 , J− ∂2 = −∂1 , J− ∂3 = ∂4 , J− ∂4 = −∂3 ,
g(∂1 , ∂1 ) = g(∂2 , ∂2 ) = 1, g(∂3 , ∂3 ) = g(∂4 , ∂4 ) = εe2f .
We then set
√ √
Z1 = 21 (∂1 −√ −1∂2 ), Z2 = 21 (∂3 −√ −1∂4 ), Z3 = Z̄1 , Z4 = Z̄2 ,
Z 1 = dx 1 + −1dx 2 , Z 2 = dx 3 + −1dx 4 , Z 3 = Z̄ 1 , Z 4 = Z̄ 2 ,
and mimic the proof of Lemma 4.9. Let fi := Zi (f ). We compute:
 4.6. REALIZABILITY OF (PARA)-KÄHLER–WEYL TENSORS IF m = 4 127
Z 1 , Z 3 
= 1, Z 2 , Z 4 
= e2f ,
√
− = − −1{Z ∧ Z + εe Z ∧ Z },
1 3 2f 2 4
√
 − = −1{Z 1 ∧ Z 3 + εe2f Z 2 ∧ Z 4 },
√
d  − = 2εe2f −1{f1 Z 1 ∧ Z 2 ∧ Z 4 − f3 Z 2 ∧ Z 3 ∧ Z 4 },
√
δg − = ε −1{−2f1 Z 1 + 2f3 Z 3 },
φ− = − 21 J− δg − = ε{−f1 Z − f3 Z },
1 3

ρa = 2ε{−f12 Z 1 ∧ Z 2 − f14 Z 1 ∧ Z 4 + f23 Z 2 ∧ Z 3 − f34 Z 3 ∧ Z 4 }.

Since √ √
J− Z 1 = √
−1Z 1 , J− Z 2 = √
−1Z 2 ,
J− Z = − −1Z , J− Z = − −1Z 4 ,
3 3 4

Z 1 ∧ Z 2 ∈ 2−,0 and Z1 ∧ Z 4 ∈ 2+ . Consequently, J− ρa has components in both 2−,0 and 2+ .
This shows 2∓,0 ⊕ 2± ⊂ Range{}. This completes the proof of Theorem 4.5. "
!
 129

Bibliography
[1] R. Abounasr, A. Belhaj, J. Rasmussen, and E. H. Saidi, “Superstring theory on pp waves with
ADE geometries”, J. Phys. A 39 (2006), 2797–2841. DOI: 10.1088/0305-4470/39/11/015.
31
[2] J. Adams, “Vector fields on spheres", Ann. of Math. 75 (1962), 603–632.
DOI: 10.2307/1970213. 50
[3] Z. Afifi, “Riemann extensions of affine connected spaces", Quart. J. Math., Oxford Ser. (2) 5
(1954), 312–320. DOI: 10.1093/qmath/5.1.312. 29, 33, 34, 35, 36, 66
[4] A. Alcolado, A. MacDougall, A. Coley, and S. Hervik, “4D neutral signature VSI and CSI
spaces", J. Geom. Phys. 62 (2012), 594–603. DOI: 10.1016/j.geomphys.2011.04.012. 25
[5] D. V. Alekseevsky, V. Cortés, A. S. Galaev, and T. Leistner, “Cones over pseudo-Riemannian
manifolds and their holonomy", J. Reine Angew. Math. 635 (2009), 23–69.
DOI: 10.1515/CRELLE.2009.075. 29
[6] I. Anderson and G.Thompson, “The inverse problem of the calculus of variations for ordinary
differential equations", Mem. Amer. Math. Soc. 98 (1992), 1–110. 60
[7] G. Andreoli, “Parallelismi trasporti rigidi, riferimenti locali nelle V2 ", Ann. Scuola Norm. Sup.
Pisa Cl. Sci. (2) 1 (1932), 315–332. https://eudml.org/doc/82835 14
[8] V. Apostolov,“Generalized Golberg-Sachs theorems for pseudo-Riemannian four manifolds",
J. Geom. Phys. 27 (1998), 185–198. DOI: 10.1016/S0393-0440(97)00075-2. 49, 79, 83
[9] T. Arias-Marco and O. Kowalski, “Classification of locally homogeneous affine connec-
tions with arbitrary torsion on 2-dimensional manifolds", Monatsh. Math 153 (2008), 1–18.
DOI: 10.1007/s00605-007-0494-0. 62, 70
[10] M. Barros and A. Romero, “Indefinite Kähler manifolds", Math. Ann. 261 (1982), 55–62.
DOI: 10.1007/BF01456410. 48
[11] L. Bérard Bergery and A. Ikemakhen, “Sur L’holonomie des variétés pseudo-riemanniennes
de signature (n, n)", Bull. Soc. Math. France 125 (1997), 93–114. https://eudml.org/doc/
87759 29
[12] A. L. Besse, “Einstein manifolds", Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge
10. Springer-Verlag, Berlin, 1987. DOI: 10.1007/978-3-540-74311-8. 9, 10
130 BIBLIOGRAPHY
[13] L. Bianchi, “Vorlesungen über differentialgeometrie", Teubner (Leipzig) (1899). 11

[14] W. Blaschke, Vorlesungen über Differentialgeometrie und geometrische Grundlagen von Einsteins
Relativitätstheorie. II. Affine Differentialgeometrie, Springer, Berlin, 1923.
DOI: 10.1007/978-3-642-47392-0. 11

[15] N. Blažić and N. Bokan, “Compact Riemann surfaces with the skew–symmetric Ricci tensor",
Izv. Vyssh. Uchebn. Zaved. Mat. 9 (1994), 8–12. 60, 61

[16] N. Blažić, N. Bokan, and P. Gilkey, “A note on Osserman Lorentzian manifolds", Bull. London
Math. Soc. 29 (1997), 227–230. DOI: 10.1112/S0024609396002238. 50, 55, 77

[17] N. Blažić, N. Bokan, and Z. Rakić, “Osserman pseudo-Riemannian manifolds of signature

(2, 2)", J. Aust. Math. Soc. 71 (2001), 367–395. DOI: 10.1017/S1446788700003001. 50, 80,
83, 84

[18] N. Blažić, N. Bokan, and Z. Rakić, “Foliation of a dynamically homogeneous neutral mani-
fold", J. Math. Phys. 39 (1998), 6118–6124. DOI: 10.1063/1.532617. 85

[19] N. Blažić, P. Gilkey, S. Nikčević, and U. Simon, “The spectral geometry of the Weyl conformal
curvature tensor", Banach Center Publ. 69 (2005), 195–203. DOI: 10.4064/bc69-0-15. 51

[20] N. Blažić, P. Gilkey, S. Nikčević, and U. Simon, “Algebraic theory of affine curvature tensors",
Arch. Math. (Brno) 42 (2006), suppl. 147–168. https://eudml.org/doc/249816 15, 17

[21] N. Blažić, P. Gilkey, S. Nikčević, and I. Stavrov, “Curvature structure of self-dual 4-manifolds",
Int. J. Geom. Methods Mod. Phys. 5 (2008), 1191–1204. DOI: 10.1142/S0219887808003259.
80

[22] N. Bokan, “On the complete decomposition of curvature tensors of Riemannian manifolds
with symmetric connection", Rend. Circ. Mat. Palermo 39 (1990), 331–380.
DOI: 10.1007/BF02844767. 20

[23] N. Bokan, P. Gilkey, and U. Simon, “Geometry of differential operators on Weyl manifolds",
Proc. R. Soc. London. Ser. A 453 (1997), 2527–2536. DOI: 10.1098/rspa.1997.0134. 15

[24] N. Bokan, P. Gilkey, and U. Simon, “Asymptotic spectra for Weyl geometries", Romanian
conference on geometry, An. Stiint. Univ. Al. I. Cuza Iasi. Mat. (N.S.) 42 (1996), 59–71. 15

[25] A. Bonome, R. Castro, E. García-Río, L. Hervella, and Y. Matsushita, “Null holomorphically

flat indefinite almost Hermitian manifolds", Illinois J. Math. 39 (1995), 635–660. http://
projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&handle=
euclid.ijm/1255986270 48
 BIBLIOGRAPHY 131
[26] A. Bonome, R. Castro, E. García-Río, L. Hervella, and R. Vázquez-Lorenzo, “Nonsymmet-
ric Osserman indefinite Kähler manifolds", Proc. Amer. Math. Soc. 126 (1998), 2763–2769.
DOI: 10.1090/S0002-9939-98-04659-0. 50, 84

[27] A. Bonome, R. Castro, E. García-Río, L. Hervella, and R. Vázquez-Lorenzo, “On the para-
holomorphic sectional curvature of almost para-Hermitian manifolds", Houston J. Math. 24
(1998), 277–300. 48, 86

[28] H. W. Brinkmann, “Einstein spaces which are mapped conformally on each other", Math.
Ann. 94 (1925), 119–145. DOI: 10.1007/BF01208647. 31

[29] M. Brozos-Vázquez, E. García-Río, and P. Gilkey, “Relating the curvature tensor and the
complex Jacobi operator of an almost Hermitian manifold", Adv. Geom. 8 (2008), 353–365.
DOI: 10.1515/ADVGEOM.2008.023. 47, 101, 102

[30] M. Brozos-Vázquez, E. García-Río, P. Gilkey, and L. Hervella, “Geometric realizability of

covariant derivative Kähler tensors for almost pseudo-Hermitian and almost para-Hermitian
manifolds", Ann. Mat. Pura Appl. 191 (2012), 487–502. DOI: 10.1007/s10231-011-0192-3.
22, 23

[31] M. Brozos-Vázquez, E. García-Río, P. Gilkey, S. Nikčević, and R. Vázquez-Lorenzo, “The

geometry of Walker manifolds", Synthesis Lectures on Mathematics and Statistics 5, Morgan &
Claypool Publ., Williston, VT, 2009. DOI: 10.2200/S00197ED1V01Y200906MAS005. 31

[32] M. Brozos-Vázquez, E. García-Río, P. Gilkey, and R. Vázquez-Lorenzo, “Homogeneous 4-

dimensional Kähler Weyl Structures", accepted for publication in Results in Mathematics. 19,
105, 107

[33] M. Brozos-Vázquez, E. García-Río, and R. Vázquez-Lorenzo, “Conformally Osserman 4-

dimensional manifolds whose conformal Jacobi operators have complex eigenvalues", Proc. R.
Soc. Lond. Ser. A 462 (2006), 1425–1441. DOI: 10.1098/rspa.2005.1621. 51, 78

[34] M. Brozos-Vázquez, P. Gilkey, and E. Merino, “Geometric realizations of Kähler and

of para-Kähler curvature models", Int. J. Geom. Methods Mod. Phys. 7 (2010), 505–515.
DOI: 10.1142/S0219887810004403. 19, 28

[35] M. Brozos-Vázquez, P. Gilkey, and S. Nikčević, “Geometric realizations of affine Kähler

curvature models", Results Math. 59 (2011), 507–521. DOI: 10.1007/s00025-011-0105-1.
20

[36] M. Brozos-Vázquez, P. Gilkey, and S. Nikčević: “The structure of the space of affine Kähler
curvature tensors as a complex module", Int. J. Geom. Methods Mod. Phys. 8 (2011), 1849–1868.
DOI: 10.1142/S0219887811005981. 20
132 BIBLIOGRAPHY
[37] M. Brozos-Vázquez, P. Gilkey, and S. Nikčević, “Geometric realizations of curvature
tensors", ICP Advanced Texts in Mathematics 6, Imperial College Press, London, 2012.
DOI: 10.1142/9781848167421. 18, 19, 21, 24, 25, 28, 125

[38] M. Brozos-Vázquez, P. Gilkey, S. Nikčević, and R. Vázquez-Lorenzo, “Geometric re-

alizations of para-Hermitian curvature models", Results Math. 56 (2009), 319–333.
DOI: 10.1007/s00025-009-0403-z. 20

[39] R. L. Bryant, “Bochner-Kähler metrics", J. Amer. Math. Soc. 14 (2001), 623–715.

DOI: 10.1090/S0894-0347-01-00366-6. 29, 37, 79, 83

[40] G. Calvaruso, “Homogeneous structures on three-dimensional Lorentzian manifolds”, J.

Geom. Phys. 57 (2007), 1279-1291. DOI: 10.1016/j.geomphys.2006.10.005. 107

[41] G. Calvaruso, “Three-dimensional Ivanov-Petrova manifolds", J. Math. Phys. 50 (2009),

063509, 12 pp. DOI: 10.1063/1.3152607. 52, 56

[42] E. Calviño-Louzao, E. García-Río, P. Gilkey, and R. Vázquez-Lorenzo, “Higher dimensional

Osserman metrics with non-nilpotent Jacobi operators", Geom. Dedicata 156 (2012), 151–163.
DOI: 10.1007/s10711-011-9595-y. 50

[43] E. Calviño-Louzao, E. García–Río, P. Gilkey, and R. Vázquez-Lorenzo, “The geometry

of modified Riemannian extensions", Proc. R. Soc. Lond. Ser. A. 465 (2009), 2023–2040.
DOI: 10.1098/rspa.2009.0046. 29, 32, 38

[44] E. Calviño-Louzao, E. García-Río, and R. Vázquez-Lorenzo, “Four-dimensional Osserman

Ivanov Petrova metrics of neutral signature", Classical Quantum Gravity 24 (2007), 2343–
2355. DOI: 10.1088/0264-9381/24/9/012. 50, 52, 73, 76

[45] E. Calviño-Louzao, E. García-Río, and R. Vázquez-Lorenzo, “Riemann extensions of

torsion-free connections with degenerate Ricci tensor", Canad. J. Math. 62 (2010), 1037–
1057. DOI: 10.4153/CJM-2010-059-2. 29, 69, 70, 76

[46] E. Canfes, “On generalized recurrent Weyl spaces and Wong’s conjecture", Differ. Geom. Dyn.
Syst. 8 (2006), 34–42. 15

[47] É. Cartan,“Sur les variétés á connexion affine, et la théorie de la relativité généralisée (première
partie)", Ann. Sci. École Norm. Sup. 40 (1923), 325–412. 11

[48] É. Cartan,“Sur les varietes a connexion projective", Bull. Soc. Math. France 52 (1924), 205–241.
8

[49] É. Cartan, “Sur une classe remarquable d’espaces de Riemann, I", Bull. Soc. Math. France 54
(1926), 214–216. 15, 41
 BIBLIOGRAPHY 133
[50] É. Cartan, “Sur une classe remarquable d’espaces de Riemann, II", Bull. Soc. Math. France 55
(1927), 114–134. 15, 41

[51] J. Cendán-Verdes, E. García-Río and M. E. Vázquez-Abal, “On the semi-Riemannian struc-

ture of the tangent bundle of a two-point homogeneous space", Riv. Mat. Univ. Parma 3
(1994), 253–270. 34

[52] M. Chaichi, E. García-Río, and Y. Matsushita, “Curvature properties of four-dimensional

Walker metrics”, Classical Quantum Gravity 22 (2005), 559–577.
DOI: 10.1088/0264-9381/22/3/008. 29

[53] Q. S. Chi, “A curvature characterization of certain locally rank-one symmetric spaces", J.

Differential Geom. 28 (1988), 187–202. http://projecteuclid.org/DPubS?service=
UI&version=1.0&verb=Display&handle=euclid.jdg/1214442277 50, 55, 77, 79

[54] E. Christoffel,“Über die Transformation der homogenen Differentialausdrücke 2ten Grades.",

Borchardt J. 70 (1869), 46–70. 8, 14

[55] A. Chudecki and M. Przanowski, “From hyperheavenly spaces to Walker and Osserman
spaces I", Classical Quantum Gravity 25 (2008), no. 14, 145010, 18 pp.
DOI: 10.1088/0264-9381/25/14/145010. 50, 85

[56] A. Chudecki and M. Przanowski, “From hyperheavenly spaces to Walker and Osserman
spaces II", Classical Quantum Gravity 25 (2008), no. 23, 235019, 22 pp.
DOI: 10.1088/0264-9381/25/23/235019. 50, 85

[57] A. Clebsch, “Über die simultane integration linearer partieller differentialgleichungen", J.

Reine. Angew. Math. 65 (1866), 257–268. DOI: 10.1515/crll.1866.65.257. 18

[58] A. Coley, R. Milson, V. Pravda, and A. Pravdová, “Vanishing scalar invariant spacetimes in
higher dimensions", Classical Quantum Gravity 21 (2004), 5519–5542.
DOI: 10.1088/0264-9381/21/23/014. 25

[59] A. Cortés-Ayaso, J. C.Díaz-Ramos, and E. García-Río, “Four-dimensional manifolds with

degenerate self-dual Weyl curvature operator", Ann. Global Anal. Geom. 34 (2008), 185–193.
DOI: 10.1007/s10455-007-9101-9. 15, 49, 85

[60] V. Cruceanu, P. Fortuny, and P. M. Gadea, “A survey on paracomplex geometry", Rocky Moun-
tain J. Math. 26 (1996), 83–115. DOI: 10.1216/rmjm/1181072105. 17, 18, 48

[61] G. Darboux, “Sur le problème de Pfaff ", C. R. 94 (1882), 835–837; Darb. Bull. (2) 6 (1982),
14–36; 49–68. https://eudml.org/doc/85135 3

[62] J. Davidov and O. Muskarov, “Self-dual Walker metrics with two-step nilpotent Ricci oper-
ator", J. Geom. Phys. 57 (2006), 157–165. DOI: 10.1016/j.geomphys.2006.02.007. 84
134 BIBLIOGRAPHY
[63] F. Deahna, “Über die Bedingungen der Integrabilität lineärer Differentialgleichungen erster
Ordnung zwischen einer beliebigen Anzahl veränderlicher Grössen", J. Reine Angew. Math.
20 (1840), 340–349. DOI: 10.1515/crll.1840.20.340. 18

[64] A. Derdzinski, “Self-dual Kähler manifolds and Einstein manifolds of dimension four", Com-
pos. Math. 49 (1983), 405–433. 49, 79

[65] A. Derdzinski, “Curvature-homogeneous indefinite Einstein metrics in dimension four:

the diagonalizable case", Recent advances in Riemannian and Lorentzian geometries (Balti-
more, MD, 2003), 21–38, Contemp. Math., 337, Amer. Math. Soc., Providence, RI, 2003.
DOI: 10.1090/conm/337/06049. 83

[66] A. Derdzinski, “Connections with skew-symmetric Ricci tensor on surfaces", Results Math.
52 (2008), 223–245. DOI: 10.1007/s00025-008-0307-3. 39, 60, 61

[67] A. Derdzinski, “Non-Walker self-dual neutral Einstein four-manifolds of Petrov type III", J.
Geom. Anal. 19 (2009), 301–357. DOI: 10.1007/s12220-008-9066-3. 85

[68] A. Derdzinski and W. Roter,“Walker’s theorem without coordinates", J. Math. Phys. 47 (2006),
062504, 8 pp. DOI: 10.1063/1.2209167. 29, 31

[69] J. C. Díaz-Ramos and E. García-Río, “A note on the structure of algebraic curvature tensors",
Linear Algebra Appl. 382 (2004), 271–277. DOI: 10.1016/j.laa.2003.12.044. 20

[70] J. C. Díaz-Ramos, E. García-Río, and R. Vázquez-Lorenzo, “New examples of Osserman

metrics with nondiagonalizable Jacobi operators", Differential Geom. Appl. 24 (2006), 433–
442. DOI: 10.1016/j.difgeo.2006.02.006. 50, 85

[71] J. C. Díaz-Ramos, E. García-Río, and R. Vázquez-Lorenzo, “Four-dimensional Osser-

man metrics with nondiagonalizable Jacobi operators", J. Geom. Anal. 16 (2006), 39–52.
DOI: 10.1007/BF02930986. 38, 50, 84

[72] A. Di Scala and L. Vezzoni, “Gray identities, canonical connection, and integrability", Proc.
Edinb. Math. Soc. 53 (2010), 657–674. DOI: 10.1017/S0013091509000157. 102

[73] A. Di Scala, J. Lauret, and L. Vezzoni, “Quasi-Kähler Chern-flat manifolds and complex
2-step nilpotent Lie algebras", Ann. Sc. Norm. Super. Pisa Cl. Sci. 11 (2012), 41–60. 102

[74] M. P. do Carmo, “Geometria riemanniana", Projeto Euclides 10. Instituto de Matematica Pura
e Aplicada, Rio de Janeiro, 1979. 41

[75] V. Dryuma, “The Riemann extensions in theory of differential equations and their ap-
plications", Mat. Fiz. Anal. Geom. 10 (2003), 307–325. http://www.mathnet.ru/php/
archive.phtml?wshow=paper&jrnid=jmag&paperid=253&option_lang=eng 29
 BIBLIOGRAPHY 135
[76] M. Dunajski, “Paraconformal geometry of nth -order
ODEs, and exotic holonomy in dimen-
sion four", J. Geom. Phys. 56 (2006), 1790–1809. DOI: 10.1016/j.geomphys.2005.10.007. 29
[77] L. Eisenhart, “Non-Riemannian geometry", American Mathematical Society Colloquium
Publications 8, Amer. Math. Soc., Providence, RI, 1964. 10, 14
[78] L. Eisenhart, Riemannian Geometry, Princeton University Press, Princeton, N.J., 1949.
http://press.princeton.edu/titles/486.html 10
[79] B. Fiedler, “Determination of the structure of algebraic curvature tensors by means of Young
symmetrizers", Sém. Lothar. Combin. 48 (2002), Art. B48d, 20 pp. 20
[80] B. Fine, P. Kirk, and E. Klassen, “A local analytic splitting of the holonomy map on flat
connections", Math. Ann. 299 (1994), 171–189. DOI: 10.1007/BF01459778. 29
[81] G. Frobenius, “Über das Pfaffsche probleme", Borchardt J. 82 (1876), 230–315. http://www.
degruyter.com/view/j/crll.1877.issue-82/crll.1877.82.230/crll.1877.
82.230.xml 18
[82] A. Frölicher and A. Nijenhuis, “Theory of vector valued differential forms. Part I.", Indag.
Math. 18 (1956), 338–360. DOI: 10.1007/978-3-540-77054-1_13. 18
[83] A. Frölicher and A. Nijenhuis, “Invariance of vector form operations under mappings", Com-
ment. Math. Helv. 34 (1960), 227–248. DOI: 10.1007/BF02565938. 18
[84] T. Fukami, “Invariant tensors under the real representation of unitary groups and their appli-
cation", J. Math. Soc. Japan 10 (1958), 135–144. DOI: 10.2969/jmsj/01020135. 24
[85] P. M. Gadea and A. Montesinos-Amilibia, “Spaces of constant para-holomorphic sectional
curvature", Pacific J. Math. 136 (1989), 85–101. DOI: 10.2140/pjm.1989.136.85. 48
[86] E. García-Río, P. Gilkey, M. E. Vázquez-Abal, and R. Vázquez-Lorenzo, “Four-dimensional
Osserman metrics of neutral signature", Pacific J. Math. 244 (2010), 21–36.
DOI: 10.2140/pjm.2010.244.21. 80, 82, 86
[87] E. García-Río, A. Haji-Badali, and R. Vázquez-Lorenzo, “Lorentzian three-manifolds with
special curvature operators", Classical Quantum Gravity 25 (2008), 015003, 13pp.
DOI: 10.1088/0264-9381/25/1/015003. 56
[88] E. García-Río, D. N. Kupeli, and M. E. Vázquez-Abal, “On a problem of Osserman in
Lorentzian geometry", Differential Geom. Appl. 7 (1997), 85–100.
DOI: 10.1016/S0926-2245(96)00037-X. 50, 55, 77
[89] E. García-Río, D. N. Kupeli, M. E. Vázquez-Abal, and R. Vázquez-Lorenzo, “Affine Osser-
man connections and their Riemann extensions", Differential Geom. Appl. 11 (1999), 145–153.
DOI: 10.1016/S0926-2245(99)00029-7. 29, 49, 51
136 BIBLIOGRAPHY
[90] E. García-Río, D. N. Kupeli, and R. Vázquez-Lorenzo, “Osserman manifolds in semi-
Riemannian geometry", Lecture Notes in Math. 1777, Springer, Berlin, 2002.
DOI: 10.1007/b83213. 50, 84, 93

[91] E. García-Río, M. E. Vázquez-Abal, and R. Vázquez-Lorenzo, “Nonsymmetric Osserman

pseudo-Riemannian manifolds", Proc. Amer. Math. Soc. 126 (1998), 2771–2778.
DOI: 10.1090/S0002-9939-98-04666-8. 50, 55, 84, 85

[92] E. García-Río and R. Vázquez-Lorenzo, “Four-dimensional Osserman symmetric spaces",

Geom. Dedicata 88 (2001), 147–151. DOI: 10.1023/A:1013101719550. 50, 61, 82

[93] H. Geiges, “Contact geometry", Handbook of Differential geometry, vol. II, 315–382. Else-
vier/North Holland, Amsterdam, 2006. 3

[94] P. Gilkey, “The spectral geometry of a Riemannian manifold", J. Differential Geom.

10 (1975), 601–618. http://projecteuclid.org/DPubS?service=UI&version=1.
0&verb=Display&handle=euclid.jdg/1214433164 25

[95] P. Gilkey, “Invariance theory, the heat equation, and the Atiyah-Singer index theorem 2nd
ed.”, Studies in Advanced Mathematics, CRC Press, Boca Raton, FL., 1995. 2

[96] P. Gilkey, “Geometric properties of natural operators defined by the Riemann curvature ten-
sor", World Scientific Publishing Co., Inc., River Edge, NJ, 2001. 47, 48, 50, 51, 55

[97] P. Gilkey, “Algebraic curvature tensors which are p Osserman", Differential Geom. Appl. 14
(2001), 297–311. DOI: 10.1016/S0926-2245(01)00040-7. 51

[98] P. Gilkey, “The geometry of curvature homogeneous pseudo-Riemannian manifolds". ICP

Advanced Texts in Mathematics, 2. Imperial College Press, London, 2007. DOI: 10.1142/p503.
17, 57

[99] P. Gilkey and R. Ivanova, “Spacelike Jordan-Osserman algebraic curvature tensors in the
higher signature setting", Differential Geometry, Valencia, 2001, 179–186, World Scientific
Publishing Co., Inc., River Edge, NJ, 2002. 50

[100] P. Gilkey and R. Ivanova, “The Jordan normal form of Osserman algebraic curvature tensors",
Results Math. 40 (2001), 192–204. DOI: 10.1007/BF03322705. 50

[101] P. Gilkey, J. V. Leahy, and H. Sadofsky, “Riemannian manifolds whose skew-symmetric

curvature operator has constant eigenvalues", Indiana Univ. Math. J. 48 (1999), 615–634.
DOI: 10.1512/iumj.1999.48.1699. 55

[102] P. Gilkey and S. Nikčević, “Kaehler and para-Kaehler curvature Weyl manifolds", Publ. Math.
Debrecen 80 (2012), 369–384. 19, 105, 108
 BIBLIOGRAPHY 137
[103] P. Gilkey and S. Nikčević, “Kähler–Weyl manifolds of dimension 4", to appear in Rend. Sem.
Mat. Univ. Politec. Torino. 19, 105, 107, 108

[104] P. Gilkey and S. Nikčević, “4-dimensional (para)-Kähler–Weyl structures",

http://arxiv.org/abs/1210.6769. DOI: 10.1007/978-1-4614-4897-6_15. 105

[105] P. Gilkey and S. Nikčević, “(para)-Kähler Weyl structures", Recent trends in Lorentzian
Geometry. Springer Proceedings in Mathematics & Statistics 26, 335–353, Springer-Verlag,
New York, NY, 2013. DOI: 10.1007/978-1-4614-4897-6_15. 19, 105, 107, 108

[106] P. Gilkey and S. Nikčević, “Affine projective Osserman structures",

http://arxiv.org/abs/1304.7482. 57

[107] P. Gilkey, S. Nikčević, and U. Simon, “Geometric theory of equiaffine curvature tensors",
Results Math. 56 (2009), 275–318. DOI: 10.1007/s00025-009-0438-1. 17

[108] P. Gilkey, S. Nikčević, and U. Simon, “Curvature Properties of Weyl Geometries", Results
Math 59 (2011), 523–544. DOI: 10.1007/s00025-011-0111-3. 15

[109] P. Gilkey, S. Nikčević, and U. Simon, “Geometric realizations, curvature decompositions, and
Weyl manifolds", J. Geom. Phys. 61 (2011), 270–275. DOI: 10.1016/j.geomphys.2010.09.022.
15, 16, 17

[110] P. Gilkey, S. Nikčević, and V. Videv, “Manifolds which are Ivanov-Petrova or k-Stanilov", J.
Geom. 80 (2004), 82–94. DOI: 10.1007/s00022-003-1750-7. 51, 52

[111] P. Gilkey, J.H. Park, and K. Sekigawa, “Universal curvature identities", Differential Geom.
Appl., 29 (2011), 770-778. DOI: 10.1016/j.difgeo.2011.08.005. 25

[112] P. Gilkey, J.H. Park, and K. Sekigawa, “Universal curvature identities II", J. Geom. Phys. 62,
(2012), 814-825. DOI: 10.1016/j.geomphys.2012.01.002. 25

[113] P. Gilkey and I. Stavrov,“Curvature tensors whose Jacobi or Szabó operator is nilpotent on null
vectors", Bull. London Math. Soc. 34 (2002), 650–658. DOI: 10.1112/S0024609302001339.
51

[114] P. Gilkey, A. Swann, and L. Vanhecke, “Isoparametric geodesic spheres and a conjecture of
Osserman concerning the Jacobi operator", Quart. J. Math. Oxford Ser. (2) 46 (1995), 299–320.
DOI: 10.1093/qmath/46.3.299. 50, 55, 78

[115] P. Gilkey and T. Zhang, “Algebraic curvature tensors for indefinite metrics whose skew-
symmetric curvature operator has constant Jordan normal form", Houston J. Math. 28 (2002),
311–328. DOI: 10.1023/A:1015221317388. 52, 55
138 BIBLIOGRAPHY
[116] M. de Gosson, “Symplectic geometry and quantum mechanics", Operator Theory: Advances
and Applications, 166. Advances in Partial Differential Equations (Basel). Birkhäuser Verlag,
Basel, 2006. 3

[117] A. Gray, “Curvature identities for Hermitian and almost Hermitian manifolds", Tôhoku Math.
J. 28 (1976), 601–612. DOI: 10.2748/tmj/1178240746. 48

[118] A. Gray and L. Hervella, “The sixteen classes of almost Hermitian manifolds and their linear
invariants", Ann. Mat. Pura Appl. (4) 123 (1980), 35–58. DOI: 10.1007/BF01796539. xi, 23

[119] A. Gray and L. Vanhecke, “Almost Hermitian manifolds with constant holomorphic sectional
curvature", Časopis Pěst. Mat. 104 (1979), 170–179. 48

[120] G. S. Hall, “Some remarks on the converse of Weyl’s conformal theorem", J. Geom. Phys. 60
(2010), 1–7. DOI: 10.1016/j.geomphys.2009.08.002. 27

[121] G. S. Hall, “Covariantly constant tensors and holonomy structure in general relativity”, J.
Math. Phys. 32 (1991), 181–187. DOI: 10.1063/1.529114. 29

[122] N. Hawley, “Constant holomorphic curvature", Canad. Math. J. 5 (1953), 53–56.

DOI: 10.4153/CJM-1953-007-1. 48

[123] S. Helgason, “Differential geometry and symmetric spaces", Pure and Applied Mathematics,
vol. XII, Academic Press, New York-London, 1962. 15, 42

[124] U. Hertrich-Jeromin, “Introduction to Möbius differential geometry", London Mathematical

Society Lecture Note Series 300, Cambridge University Press, Cambridge, 2003. 27

[125] T. Higa, “Weyl manifolds and Einstein-Weyl manifolds", Comm. Math. Univ. St. Paul. 42
(1993), 143-160. xi, 1, 20, 28

[126] T. Higa, “Curvature tensors and curvature conditions in Weyl geometry", Comm. Math. Univ.
St. Paul 43 (1994), 139–153. xi, 1, 20, 28

[127] N. Hitchin, “Hyperkähler manifolds", Séminaire Bourbaki (1991/92), Ast’erisque 206 (1992),
Exp. No. 748, 3, 137–166. 80

[128] W. Hodge, The theory and applications of harmonic integrals, Cambridge University Press,
Cambridge, 1941. 2

[129] K. Honda and K. Tsukada, “Conformally flat semi-Riemannian manifolds with nilpotent
Ricci operators and affine differential geometry", Ann. Global Anal. Geom. 25 (2004), 253–
275. DOI: 10.1023/B:AGAG.0000023245.73639.93. 11

[130] L. Hörmander, “The Frobenius-Nirenberg theorem", Ark. Mat. 5 (1965), 425–432.

DOI: 10.1007/BF02591139. 18
 BIBLIOGRAPHY 139
[131] J. Igusa, “On the structure of a certain class of Kähler varieties", Amer. J. Math. 76 (1954),
669–678. DOI: 10.2307/2372709. 48
[132] M. Itoh, “Affine locally symmetric structures and finiteness theorems for Einstein-Weyl man-
ifolds", Tokyo J. Math. 23 (2000), 37–49. DOI: 10.3836/tjm/1255958806. 15
[133] S. Ivanov, “Geometry of quaternionic Kähler connections with torsion", J. Geom. Phys. 41
(2002), 235–257. DOI: 10.1016/S0393-0440(01)00058-4. 11
[134] S. Ivanov and I. Petrova, “Riemannian manifold in which the skew-symmetric curvature
operator has pointwise constant eigenvalues", Geom. Dedicata 70 (1998), 269–282.
DOI: 10.1023/A:1005014507809. 52, 56
[135] S. Ivanov and S. Zamkovoy, “Parahermitian and paraquaternionic manifolds", Differential
Geom. Appl. 23 (2005), 205–234. DOI: 10.1016/j.difgeo.2005.06.002. 49, 79, 83
[136] N. Iwahori, “Some remarks on tensor invariants of O(n), U (n), Sp(n)", J. Math. Soc. Japan
10 (1958), 146–160. DOI: 10.2969/jmsj/01020145. 24
[137] J. Jost, “Riemannian geometry and geometric analysis", Universitext, Springer-Verlag, Berlin,
2002. DOI: 10.1007/978-3-662-04745-3. 4
[138] M. Karoubi, “K-theory, an introduction", Grundlehren der mathematischen Wissenschaften 226,
Springer Verlag (Berlin) (1978). 80
[139] J. Kerimo, “AdS pp-waves”, J. High Energy Phys. (2005), 025, 18 pp.
DOI: 10.1088/1126-6708/2005/09/025. 31
[140] S. Kobayashi and K. Nomizu, “Foundations of Differential Geometry vol. I and II", Wiley
Classics Library. A Wyley-Interscience Publication, John Wiley & Sons, Inc., New York, 1996.
1, 9, 10, 14, 15, 18, 47, 48
[141] G. Kokarev and D. Kotschick, “Fibrations and fundamental groups of Kähler-Weyl mani-
folds”, Proc. Amer. Math. Soc. 138 (2010), 997–1010. DOI: 10.1090/S0002-9939-09-10110-7.
106
[142] O. Kowalski, B. Opozda, and Z. Vlášek, “Curvature Homogeneity of affine connections on
two-dimensional manifolds", Colloq. Math. 81 (1999), 123–139. 60
[143] O. Kowalski, B. Opozda, and Z. Vlášek, “A classification of locally homogeneous affine con-
nections with skew-symmetric Ricci tensor on 2-dimensional manifolds", Monatsch. Math.
130 (2000), 109–125. DOI: 10.1007/s006050070041. 56, 60
[144] O. Kowalski, B. Opozda, and Z. Vlášek, “A classification of locally homogeneous connections
on 2-dimensional manifolds via group-theoretical approach", Cent. Eur. J. Math. 2 (2004),
87–102. DOI: 10.2478/BF02475953. 56, 70
140 BIBLIOGRAPHY
[145] O. Kowalski and M. Sekizawa, “Natural lifts in Riemannian geometry", Variations, geometry
and physics, 189–207, Nova Sci. Publ., New York, 2009. 7

[146] O. Kowalski and M. Sekizawa, “On natural Riemann extensions", Publ. Math. Debrecen 78
(2011), 709–721. 7

[147] S. Lang, “Differential and Riemannian manifolds", Graduate Texts in Mathematics 160,
Springer-Verlag, Berlin, 1995. DOI: 10.1007/978-1-4612-4182-9. 18

[148] P. Law and Y. Matsushita, “A spinor approach to Walker geometry", Comm. Math. Phys. 282
(2008), 577–623. DOI: 10.1007/s00220-008-0561-y. 29

[149] T. Leistner, “Screen bundles of Lorentzian manifolds and some generalizations of pp-waves”,
J. Geom. Phys. 56 (2006), 2117–2134. DOI: 10.1016/j.geomphys.2005.11.010. 31

[150] T. Leistner, “Conformal holonomy of C-spaces, Ricci-flat, and Lorentzian manifolds", Dif-
ferential Geom. Appl. 24 (2006), 458–478. DOI: 10.1016/j.difgeo.2006.04.008. 29, 31

[151] T. Levi-Civita, “Nozione di parallelismo in una varietà qualunque e consequente specifi-

cazione geometrica della curvatura Riemanniana", Rend. Circ. Mat. Palermo 42 (1917), 73–
205. DOI: 10.1007/BF03014898. xi, 14

[152] Y. Matsushita, “Walker 4-manifolds with proper almost complex structures", J. Geom. Phys.
55 (2005), 385–398. DOI: 10.1016/j.geomphys.2004.12.014. 29

[153] P. Matzeu and S. Nikčević, “Linear algebra of curvature tensors on Hermitian manifolds",
An. Stiint. Univ. Al. I. Cuza. Iasi Sect. I. a Mat. 37 (1991), 71–86. 20

[154] D. McDuff and D. Salamon, “Introduction to symplectic topology", Oxford Mathematical

Monographs, The Clarendon Press, Oxford University Press, New York, 1998. 3

[155] J. Michelson and X. Wu, “Dynamics of antimembranes in the maximally supersymmetric

eleven-dimensional pp wave”, J. High Energy Phys. (2006), 028, 37 pp.
DOI: 10.1088/1126-6708/2006/01/028. 31

[156] H. Mori, “On the decomposition of generalized K-curvature tensor fields", Tohoku Math. J.
(2) 25 (1973), 225–235. DOI: 10.2748/tmj/1178241382. 28

[157] A. Newlander and L. Nirenberg,“Complex analytic coordinates in almost complex manifolds",

Ann. of Math. 65 (1957), 391–404. DOI: 10.2307/1970051. 18

[158] A. Nijenhuis and W. Woolf, “Some integration problems in almost-complex and complex
manifolds", Ann. of Math. 77 (1963), 424–489. DOI: 10.2307/1970126. 18
 BIBLIOGRAPHY 141
[159] S. Nikčević, “On the decomposition of curvature tensor fields on Hermitian manifolds",
Differential geometry and its applications (Eger, 1989), 555-568, Colloq. Math. Soc. Janos
Bolyai 56, North-Holland, Amsterdam 1992. 20

[160] S. Nikčević, “On the decomposition of curvature tensor", Proceedings of the Ninth Yugoslav
Conference on Geometry (Kragujevac, 1992). Zb. Rad. (Kragujevac) 16 (1994), 61–68. 20

[161] Y. Nikolayevsky, “Riemannian manifolds whose curvature operator R(X, Y ) has constant
eigenvalues", Bull. Austral. Math. Soc. 70 (2004), 301–319.
DOI: 10.1017/S0004972700034523. 56

[162] Y. Nikolayevsky, “Osserman manifolds of dimension 8", Manuscripta Math. 115 (2004), 31–
53. DOI: 10.1007/s00229-004-0480-y. 50

[163] Y. Nikolayevsky, “Osserman conjecture in dimension  = 8, 16", Math. Ann. 331 (2005), 505–
522. DOI: 10.1007/s00208-004-0580-8. 50

[164] Y. Nikolayevsky, “Conformally Osserman manifolds", Pacific J. Math. 245 (2010), 315–358.
DOI: 10.2140/pjm.2010.245.315. 51

[165] Y. Nikolayevsky, “Conformally Osserman manifolds of dimension 16 and a Weyl-Schouten

theorem for rank-one symmetric spaces", Ann. Mat. Pura Appl. 191 (2012), 677–709.
DOI: 10.1007/s10231-011-0201-6. 51

[166] K. Nomizu, “Introduction to affine differential geometry, Part I", MPI Preprint 88-37, 1988.
13, 39

[167] K. Nomizu, “On the decomposition of generalized curvature tensor fields, Codazzi, Ricci,
Bianchi and Weyl revisited", Differential geometry (in honor of K. Yano), 335-345. Kinokuniya,
Tokyo, 1972. 17

[168] K. Nomizu and T. Sasaki, “Affine differential geometry", Cambridge Tracts in Mathematics
111, Cambridge Univ. Press, Cambridge, 1993. 11

[169] B. O’Neill, “Semi-Riemannian geometry. With applications to relativity", Pure and Applied
Mathematics 103, Academic Press, Inc., New York, 1983. 15, 44

[170] Z. Olszak, “On conformally recurrent manifolds II. Riemann extensions", Tensor (N.S.) 49
(1990), 24–31. 29

[171] Z. Olszak, “On the existence of generalized complex space forms", Israel J. Math. 65 (1989),
214–218. DOI: 10.1007/BF02764861. 48, 49

[172] B. Opozda, A classification of locally homogeneous connections on 2-dimensional manifolds,

Differential Geom. Appl. 21 (2004), 173–198. DOI: 10.1016/j.difgeo.2004.03.005. 56, 62
142 BIBLIOGRAPHY
[173] R. Osserman, “Curvature in the eighties", Amer. Math. Monthly 97 (1990), 731–756.
DOI: 10.2307/2324577. 49, 55

[174] G. Ovando, “Invariant pseudo-Kähler metrics in dimension four", J. Lie Theory 16 (2006),
371–391. 16

[175] A. Özdeğer, “On sectional curvatures of a Weyl manifold", Proc. Japan Acad. Ser. A Math. Sci.
82 (2006), 123–125. DOI: 10.3792/pjaa.82.123. 16

[176] J. Patera, R. T. Sharp, P. Winternitz, and H. Zassenhaus, “Invariants of real low dimensional
Lie algebras", J. Math. Phys. 17 (1976), 986–994. DOI: 10.1063/1.522992. 117

[177] E. M. Patterson and A. G. Walker, “Riemann extensions", Quart. J. Math., Oxford Ser. (2) 3
(1952), 19–28. DOI: 10.1093/qmath/3.1.19. xi, 29, 33

[178] H. Pedersen, Y. Poon, and A. Swann, “The Einstein-Weyl equations in complex and quater-
nionic geometry", Differential Geom. Appl. 3 (1993), 309–321.
DOI: 10.1016/0926-2245(93)90009-P. 106

[179] H. Pedersen and A. Swann, “Riemannian submersions, four manifolds, and Einstein-Weyl
geometry", Proc. London Math. Soc. 66 (1991), 381–399. DOI: 10.1112/plms/s3-66.2.381. 15

[180] H. Pedersen and K. Tod, “Three-dimensional Einstein-Weyl geometry", Adv. Math. 97

(1993), 74–109. DOI: 10.1006/aima.1993.1002. 15

[181] V. Perlick, “Observer fields in Weylian space time models", Classical Quantum Gravity 8
(1991), 1369–1385. DOI: 10.1088/0264-9381/8/7/013. 15

[182] V. Pravda, A. Pravdová, A. Coley, and R. Milson, “All spacetimes with vanishing curvature
invariants”, Classical Quantum Gravity 19 (2002), 6213–6236.
DOI: 10.1088/0264-9381/19/23/318. 25

[183] F. Prüfer, F. Tricerri, and L. Vanhecke, “Curvature invariants, differential operators and local
homogeneity", Trans. Amer. Math. Soc. 348 (1996), 4643–4652.
DOI: 10.1090/S0002-9947-96-01686-8. 25

[184] G. de Rham, “Sur la reductibilité d’un espace de Riemann", Comment. Math. Helv. 26 (1952),
328–344. DOI: 10.1007/BF02564308. 30

[185] G. de Rham, “La théorie des formes différentielles extérieures et l’homologie des variétés
différentiables", Rend. Mat. e Appl. (5) 20 (1961), 105–146.
DOI: 10.1007/978-3-642-10952-2_1. 2

[186] G. Ricci, “Sulla derivazione covariante ad una forma quadratica differenziale", Rom. Acc. L.
Rend. 3 (1887), 15–18. 8
 BIBLIOGRAPHY 143
[187] G. Ricci, “Direzioni e invarianti principali in una varietà qualunque", Ven. Ist. Atti 63 (1903–
1904), 1233–1239. 10

[188] G. Ricci and T. Levi-Civita, “Méthodes de calcul différential absolu et leurs applications",
Math. Ann. 54 (1900), 125–201. DOI: 10.1007/BF01454201. xi, 8, 9

[189] B. Riemann, “On the hypotheses which lie at the foundation of geometry", Nature VIII. Nos.
183, 184 (1873), 14–17, 36, 37. (Translated by W. Clifford). 9

[190] P. A. Schirokow and A. P. Schirokow, Affine Differentialgeometrie, Deutsche Übersetzung: Olaf

Neumann; Wissenschaftliche Redaktion: Hans Reichardt B. G. Teubner Verlagsgesellschaft,
Leipzig, 1962. 11, 12

[191] J. A. Schouten, “Über die konforme Abbildung n-dimensionaler Mannigfaltigkeiten

mit quadratischer Maßbestimmung auf eine Mannigfaltigkeit mit euklidischer Maß-
bestimmung", Math. Z. 11 (1921), 58–88. DOI: 10.1007/BF01203193. 27

[192] U. Simon, “Affine differential geometry", Handbook of Differential Geometry, vol. I, 905–961,
North-Holland, Amsterdam, 2000. DOI: 10.1016/S1874-5741(00)80012-6. 11

[193] U. Simon, A. Schwenck-Schellschmidt, and H. Viesel, “Introduction to the affine differential

geometry of hypersurfaces", Lecture Notes Science University Tokyo, 1991. 11

[194] I. M. Singer and J. A. Thorpe, “The curvature of 4-dimensional Einstein spaces", Global
Analysis (Papers in honor of K. Kodaira), 355–365. Univ. Tokyo Press, Tokyo, 1969. xi, 1, 20,
27, 28

[195] M. Sitaramayya,“Curvature tensors in Kaehler manifolds",Trans. Amer. Math. Soc. 183 (1973),
341–353. DOI: 10.1090/S0002-9947-1973-0322722-1. 28

[196] M. Spivak, “A comprehensive introduction to differential geometry", Second edition. Publish

or Perish, Inc., Wilmington, Del., 1979. 4, 14

[197] G. Stanilov and V. Videv, “On a generalization of the Jacobi operator in the Riemannian
geometry", Annuaire Univ. Sofia Fac. Math. Inform. 86 (1992), 27–34. 51

[198] R. Strichartz, “Linear algebra of curvature tensors and their covariant derivatives", Canad. J.
Math. 40 (1988), 1105–1143. DOI: 10.4153/CJM-1988-046-7. 20

[199] K. Tod, “Compact 3 dimensional Einstein-Weyl structures", J. London Math. Soc. 45 (1992),
341–351. DOI: 10.1112/jlms/s2-45.2.341. 15

[200] F. Tricerri and L. Vanhecke, “Curvature tensors on almost Hermitian manifolds", Trans. Amer.
Math. Soc. 267 (1981), 365–397. DOI: 10.1090/S0002-9947-1981-0626479-0. xi, 1, 20, 28,
29, 37, 125
144 BIBLIOGRAPHY
[201] I. Vaisman, “Generalized Hopf manifolds", Geom. Dedicata 13 (1982), 231–255.
DOI: 10.1007/BF00148231. 106

[202] I. Vaisman, “A survey of generalized Hopf manifolds", Conference on differential geometry

on homogeneous spaces (Turin, 1983). Rend. Sem. Mat. Univ. Politec. Torino 1983, Special
Issue, 205–221 (1984). 106

[203] L. Vanhecke and T. J. Willmore, “Riemann extensions of D’Atri spaces", Tensor (N.S.) 38
(1982), 154–158. DOI: 10.1007/978-1-4612-2432-7_9. 29

[204] A. G. Walker, “Canonical form for a Riemannian space with a parallel field of null planes",
Quart. J. Math., Oxford Ser. (2) 1 (1950), 69–79. DOI: 10.1093/qmath/1.1.69. 31

[205] A. Weinstein, “Symplectic geometry", Bull. Amer. Math. Soc. 5 (1981), 1–13.
DOI: 10.1090/S0273-0979-1981-14911-9. 3

[206] R. Wells, “Differential analysis on complex manifolds", Graduate Texts in Mathematics 65,
Springer-Verlag, New York-Berlin, 1980. DOI: 10.1007/978-1-4757-3946-6. 18

[207] H. Weyl, “Reine Infinitesimalgeometrie", Math. Z. 2 (1918), 384–411.

DOI: 10.1007/BF01199420. 27

[208] H. Weyl, “Zur Infinitesimalgeometrie: Einordnung der projektiven und der konformen Auf-
fassung", Gött. Nachr. (1921), 99–112. 27, 34

[209] H. Weyl, Space-time matter. Fourth edition, Dover Publ., New York, 1951. 15

[210] H. Weyl, “The Classical Groups. Their Invariants and Representations", Princeton University
Press, Princeton, NJ, 1939. 24

[211] T. J.Willmore,“Riemann extensions and affine differential geometry", Results Math. 13 (1988),
403–408. DOI: 10.1007/BF03323255. 29, 34

[212] J. Wolf, “Spaces of constant curvature. Sixth edition", AMS Chelsea Publishing, Providence,
RI, 2011. 15, 45

[213] Y. C. Wong, “Two dimensional linear connexions with zero torsion and recurrent curvature",
Monatsh. Math. 68 (1964), 175–184. DOI: 10.1007/BF01307120. 39, 40, 60, 61, 62

[214] H. Wu, “On the de Rham decomposition theorem", Illinois J. Math. 8 (1964),
291–311. http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=
Display&handle=euclid.ijm/1256059674 30

[215] K. Yano and S. Ishihara, “Tangent and cotangent bundles", Pure and Applied Mathematics 16,
Marcel Dekker Inc., New York, 1973. 4, 5, 6, 34
 BIBLIOGRAPHY 145
[216] T. Zhang, “Manifolds with indefinite metrics whose skew-symmetric curvature operator has
constant eigenvalues", Steps in differential geometry (Debrecen, 2000) 401–407, Inst. Math.
Inform. Debrecen 2001. 52
 147

Authors’ Biographies

EDUARDO GARCÍA-RÍO
Eduardo García-Río1 is a Professor of Mathematics and a mem-
ber of the Institute of Mathematics of the University of Santiago
de Compostela (Spain). He received his Ph.D. degree in 1992
from the University of Santiago de Compostela and is a member
of the editorial board of the Journal of Geometric Analysis. His re-
search specialities are Differential Geometry and Mathematical
Physics.

PETER GILKEY
Peter Gilkey 2 is a Professor of Mathematics and a member of
the Institute of Theoretical Science at the University of Oregon
(USA). He is a fellow of the American Mathematical Society
and is a member of the editorial board of Results in Mathematics,
Differential Geometry and Applications, and the International Jour-
nal of Geometric Methods to Mathematical Physics. He received his
Ph.D. in 1972 from Harvard University under the direction of
L. Nirenberg. His research specialties are Differential Geometry,
Elliptic Partial Differential Equations, and Algebraic topology.
He has published more than 230 research articles and books.

1 Department of Geometry and Topology, Faculty of Mathematics, University of Santiago de Compostela,

15782 Santiago de Compostela, Spain.
email: [email protected]
2 Mathematics Department, University of Oregon, Eugene OR 97403 USA
email: [email protected]
148 AUTHORS’ BIOGRAPHIES

STANA NIKČEVIĆ
Stana Nikčević3 is a Professor of Mathematics at the University
of Belgrade (Serbia). She received her Ph.D. from the University
of Belgrade at the Mathematical Faculty and has been working at
the University since 1974. During this period, she also sporadi-
cally worked at the University of Banja Luka (Bosnia and Herce-
govina) and at the Mathematical Faculty in Kragujevac (Serbia).
Her research mainly focuses on Differential Geometry. She has
maintained international cooperation and has gone on short vis-
its to the TU Berlin, Charles University Prague, Universitate Pierre et Marie Curie (Paris VI),
University of Oregon (USA), and University of Santiago de Compostela (Spain).

RAMÓN VÁZQUEZ-LORENZO
Ramón Vázquez-Lorenzo4 is a member of the research group
in Riemannian Geometry at the Department of Geometry and
Topology of the University of Santiago de Compostela (Spain).
He is a member of the Spanish Research Network on Relativity
and Gravitation. He received his Ph.D. in 1997 from the Univer-
sity of Santiago de Compostela. His research focuses mainly on
Differential Geometry with special emphasis on the study of the
curvature and the algebraic properties of curvature operators in
the Lorentzian and in higher signature settings. He has published
more than 45 research articles and books.

3 Mathematical Institute, Sanu, Knez Mihailova 36, p.p. 367

11001 Beograd, Serbia
email: [email protected]
4 Department of Geometry and Topology, Faculty of Mathematics, University of Santiago de Compostela,
15782 Santiago de Compostela, Spain.
email: [email protected]
 149

Index

adapted coordinates, 31 complete, 13

affine, 11, 17, 53 complete lift, 6, 33
affine algebraic curvature tensor, 17 complex general linear group, 116
affine connection, 10, 66, 105 complex space form, 48, 50, 79, 82, 83, 86
affine geometry, 15 complex structure, 17
affine Ivanov–Petrova, 56–59, 63, 64, 66, 67, complex vector space, 18
69, 72, 75 complex Weyl connection, 115
affine Osserman, 50, 51, 55, 58–60, 92–95, 101 conformal curvature operator, 51, 78
affine surface, 14 conformal factor, 27
affine-Killing, 70 conformally equivalent, 16, 27, 105
almost complex, 18 conformally flat, 76, 79
almost para-complex, 18 conformally Osserman, 51
almost para-Hermitian, 18 conjugate points, 13
almost pseudo-Hermitian, 18 conjugation, 116
alternating Ricci tensor, 10, 39, 60, 105, 107 constant holomorphic sectional curvature, 47
anti-self-dual, 51, 76, 78 constant sectional curvature, 15, 34, 43, 46, 49,
anti-self-dual 2-form, 37 50, 55, 79, 86
anti-self-dual curvature operator, 37 contact 1-form, 3
Bianchi identity, 11, 12, 51, 79, 125 contact geometry, 3
bilinear pairing, 4 contact manifold, 3
Bochner curvature tensor, 29, 37, 83 contravariant tensors, 2
Bochner flat, 39, 48, 49, 79, 83 coordinate frames, 1
cotangent bundle, 1
Cayley plane, 50 covariant tensors, 2
characteristic polynomial, 54 curvature model, 16, 17, 19, 22, 43, 47, 48, 50,
Christoffel symbol, 9–11, 14, 32, 34, 35, 39, 40, 78–80, 82, 83, 86, 106, 108, 110
53, 66, 70, 71, 92, 94, 103, 112, 124 curvature operator, 9, 11, 14, 32, 34, 37, 39, 49,
Clifford module, 50, 80 54, 84, 87, 89, 92
Clifford relations, 50 curvature tensor, 14, 54
Codazzi tensor, 63
coderivative, 2, 37, 106 de Rham cohomology, 2, 16, 110
150 INDEX
deformed Riemannian extension, 34, 35, 53, 72 higher order Jacobi operator, 51, 101
degenerate Ricci tensor, 56, 61, 63, 64, 67, 69, higher order Osserman, 51
70 Hodge  operator, 37, 79, 106, 112
directional covariant derivative, 8 Hodge–de Rham theorem, 2
dual basis, 8 holomorphic, 47
dual bundle, 9 holomorphic line bundle, 47
dual Christoffel symbols, 9 holomorphic sectional curvature, 47
dual connection, 8, 9 holonomy, 29, 30
dual frame, 9 homogeneous, 25, 42, 43, 46, 53, 56, 57, 62, 63,
65, 66, 69, 70, 83, 85, 107
Einstein, 10, 35, 36, 49, 77, 79, 82, 83 homogeneous affine connection, 63
Einstein convention, 2, 8 Hopf fibration, 48
equiaffine, 12, 63, 65, 69 hyper Kähler, 79
evaluation map, 5, 33 hyperbolic space form, 45
exponential map, 13, 41
exterior derivative, 2 identity theorem, 116
indecomposable, 30
fiber metric, 9 inner product space, 16
first covariant derivative, 11 integrable complex structure, 105, 123
flat, 12, 27, 30, 35, 45, 47, 55, 57, 59, 60, 62, integrable para-complex structure, 105, 123
63, 69, 70, 87, 88, 92, 101, 103, 125 intertwining operator, 24
flow curve, 3 Ivanov–Petrova, 52, 53, 55–57, 72, 75, 76
Fubini–Study metric, 48
Jacobi equation, 43
generalized affine plane wave manifold, 58, 59 Jacobi identity, 4, 118
geodesic, 12, 13, 40–43, 45, 46, 58 Jacobi operator, 41, 49–51, 53, 54, 60, 71,
geodesic equation, 57, 58 79–86, 88, 89, 92–95, 98, 100, 103
geodesic involution, 43 Jacobi vector field, 40–43, 46
geodesic spray, 41, 42, 46
geodesic symmetry, 42, 43 Kähler affine, 18
geodesically complete, 13, 58, 94 Kähler curvature model, 19
geodesically incomplete, 57 Kähler form, 19, 21, 105
geometrically realizable, 17, 19, 23, 83, 105, Kähler identity, 19, 106
107, 123 Kähler metric, 2
germ, 114, 123 Kähler Riemannian tensors, 106
Grassmannian, 44, 48, 51, 52, 55, 101 Kähler–Weyl, 19, 105, 106, 123
Koszul formula, 14, 124
Hermitian, 18, 47 Kronecker symbol, 4
Hermitian connection, 9
Higa decomposition, 28 Laplacian, 2
 INDEX 151
Leibnitz formula, 8, 9 para-Kähler form, 19, 21, 105, 119, 120
Levi-Civita connection, 1, 10, 13–16, 22, 29, para-Kähler identity, 19
32, 34, 40, 46, 51, 54, 56, 62, 71, 72, para-Kähler metric, 2
75, 105, 111, 115, 124 para-Kähler Riemannian tensors, 106
Lie bracket, 4, 14 para-Kähler–Weyl, 19, 105, 110, 123
Lie derivative, 3 para-Kähler–Weyl tensors, 106
light cone, 46 para-quaternionic structure, 82
Liouville 1- form, 5 para-unitary group, 20, 107
locally conformally flat, 27, 34, 66 parallel distribution, 30
locally symmetric, 70 parallel plane field, 30
Lorentzian, 10, 31, 37, 50, 51, 55, 56, 77, 80, parallel volume form, 12
83, 107 Poincaré 1-form, 5
pointwise, 52
modified Riemannian extension, 36, 77, 84, 85,
pp-waves, 31
87, 88, 92, 101
projectively affine Ivanov–Petrova, 57
natural projection, 4 projectively affine Osserman, 60
Nijenhuis tensor, 18 projectively equivalent, 13
null distribution, 30 projectively flat, 14, 34, 56, 62–66, 69, 70
null Osserman, 82 pseudo-Hermitian, 19, 20, 105, 123
null parallel plane field, 31, 39 pseudo-Riemannian, 105
null parallel vertical plane field, 91 pseudo-sphere, 46, 48, 50, 88

one-parameter flow, 3 quaternion structure, 78

orbits, 116 quaternionic space form, 50
orthogonal group, 20
Osserman, 49, 55, 61, 72, 75, 77–80, 82–86, rank-one symmetric space, 50
88, 98, 103 recurrent, 15, 39, 56, 61, 63, 66–70
reducibly action, 30
para-complex, 17 Reeb vector field, 3
para-complex Jacobi operator, 101 Ricci flat, 35, 58, 59, 72, 76, 85
para-complex Osserman, 101 Ricci operator, 10, 31, 35, 56, 75, 76
para-complex space form, 48, 82, 83, 88 Ricci symmetric, 12
para-complex vector space, 18 Ricci tensor, 10, 13, 14, 27, 28, 36, 39, 40, 56,
para-Hermitian, 18–20, 47, 123 60–64, 66, 69, 70, 85, 107, 110, 118,
para-Hermitian manifold, 105, 123 126
para-holomorphic, 47, 86 Riemann surface, 125
para-Kähler, 125 Riemannian, 10
para-Kähler affine, 18 Riemannian algebraic curvature tensor, 16
para-Kähler curvature model, 19 Riemannian connection, 9
152 INDEX
Riemannian extension, 33, 39, 53, 89 tangent bundle, 1
Riemannian geometry, 15 tautological 1-form, 5
third Gray identity, 102
scalar curvature, 27
timelike Jordan–Osserman, 50
scalar invariant, 24
timelike Osserman, 82
second theorem of invariants, 25
torsion tensor, 10
sectional curvature, 15, 43, 86
torsion-free, 10
self-dual, 37, 49, 51, 76, 78, 85
totally isotropic, 30
semi para-complex Osserman, 101, 102
trace-free Ricci tensor, 36
Singer–Thorpe decomposition, 28, 37
trivial Kähler–Weyl structure, 105
skew-symmetric curvature operator, 49, 52–54,
trivial module, 108
60, 71, 73
trivial para-Kähler–Weyl structure, 105
skew-symmetric Ricci tensor, 60, 61, 85
trivial Weyl structures, 16
space form, 45, 48, 50, 79, 82, 83
spacelike Jordan–Osserman, 50 unitary group, 107
spacelike Osserman, 82 universal curvature identities, 25
special orthogonal group, 119
sphere bundle, 55 VSI manifolds, 25
spherical space form, 45
standard para-complex structure, 112 Walker, 31, 76, 85
structure constants, 14 Walker coordinates, 32
structure group, 20, 107, 121 Walker metric, 54
symmetric, 12, 13, 15, 34, 36, 42, 50, 55, 57, Walker structure, 31
59–62, 64, 70, 82–85 Weyl, 15, 17
symmetric 2-cotensor, 10 Weyl algebraic curvature tensors, 17
symmetric 2-cotensors, 2 Weyl conformal curvature tensor, 27, 28, 66
symmetric Ricci tensor, 10, 14, 16, 35, 39, 40, Weyl connection, 16, 105, 123
56, 60–67, 69, 70, 110, 126 Weyl curvature operator, 27, 49, 79, 83
symmetric space, 41, 55 Weyl geometry, 15, 51
symmetric tensor, 13, 16 Weyl operator, 37
symmetric tensor product, 2 Weyl projective curvature operator, 13
symplectic structure, 2, 5, 19 Weyl structure, 15