Lecture Notes in Mathematics 1977

Editors:
J.-M. Morel, Cachan
F. Takens, Groningen
B. Teissier, Paris
C.I.M.E. means Centro Internazionale Matematico Estivo, that is, International Mathematical Summer
Center. Conceived in the early fifties, it was born in 1954 and made welcome by the world mathemat-
ical community where it remains in good health and spirit. Many mathematicians from all over the
world have been involved in a way or another in C.I.M.E.’s activities during the past years.

So they already know what the C.I.M.E. is all about. For the benefit of future potential users and co-
operators the main purposes and the functioning of the Centre may be summarized as follows: every
year, during the summer, Sessions (three or four as a rule) on different themes from pure and applied
mathematics are offered by application to mathematicians from all countries. Each session is generally
based on three or four main courses (24−30 hours over a period of 6-8 working days) held from
specialists of international renown, plus a certain number of seminars.

A C.I.M.E. Session, therefore, is neither a Symposium, nor just a School, but maybe a blend of both.
The aim is that of bringing to the attention of younger researchers the origins, later developments, and
perspectives of some branch of live mathematics.

The topics of the courses are generally of international resonance and the participation of the courses
cover the expertise of different countries and continents. Such combination, gave an excellent opportu-
nity to young participants to be acquainted with the most advance research in the topics of the courses
and the possibility of an interchange with the world famous specialists. The full immersion atmosphere
of the courses and the daily exchange among participants are a first building brick in the edifice of in-
ternational collaboration in mathematical research.

C.I.M.E. Director C.I.M.E. Secretary

Pietro ZECCA Elvira MASCOLO
Dipartimento di Energetica “S. Stecco” Dipartimento di Matematica
Università di Firenze Università di Firenze
Via S. Marta, 3 viale G.B. Morgagni 67/A
50139 Florence 50134 Florence
Italy Italy
e-mail: [email protected] e-mail: [email protected]

For more information see CIME’s homepage: http://www.cime.unifi.it

CIME activity is carried out with the collaboration and financial support of:

– INdAM (Istituto Nationale di Alta Mathematica)

Matthew J. Gursky · Ermanno Lanconelli
Andrea Malchiodi · Gabriella Tarantello
Xu-Jia Wang · Paul C. Yang

Geometric Analysis
and PDEs

Lectures given at the

C.I.M.E. Summer School
held in Cetraro, Italy
June 11–16, 2007

Editors:
Antonio Ambrosetti
Sun-Yung Alice Chang
Andrea Malchiodi

123
Editors
Alice Chang Andrea Malchiodi
Princeton University SISSA
Fine Hall Via Beirut 2-4
Washington Road 34014 Trieste
Princeton, NJ 08544-1000 Italy
USA [email protected]
[email protected]

Antonio Ambrosetti
SISSA
Via Beirut 2-4
34014 Trieste
Italy
[email protected]

Authors: see List of Contributors

ISSN 0075-8434 e-ISSN: 1617-9692

ISBN: 978-3-642-01673-8 e-ISBN: 978-3-642-01674-5
DOI: 10.1007/978-3-642-01674-5
Springer Dordrecht Heidelberg London New York
Library of Congress Control Number: 2009926187

Mathematics Subject Classification (2000): 35J60, 35A30, 35J20, 35H20, 53C17, 35Q40

c Springer-Verlag Berlin Heidelberg 2009

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

Cover design: SPi Publisher Services

Printed on acid-free paper

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

Preface

This volume contains the notes of the lectures delivered at the CIME course
Geometric Analysis and PDEs during the week of June 11–16 2007 in Cetraro
(Cosenza). The school consisted in six courses held by M. Gursky (PDEs in
Conformal Geometry), E. Lanconelli (Heat kernels in sub-Riemannian set-
tings), A. Malchiodi (Concentration of solutions for some singularly perturbed
Neumann problems), G. Tarantello (On some elliptic problems in the study of
selfdual Chern-Simons vortices), X.J. Wang (The k-Hessian Equation) and
P. Yang (Minimal Surfaces in CR Geometry).
Geometric PDEs are a field of research which is currently very active, as
it makes it possible to treat classical problems in geometry and has had a
dramatic impact on the comprehension of three- and four-dimensional mani-
folds in the last several years. On one hand the geometric structure of these
PDEs might cause general difficulties due to the presence of some invariance
(translations, dilations, choice of gauge, etc.), which results in a lack of com-
pactness of the functional embeddings for the spaces of functions associated
with the problems. On the other hand, a geometric intuition or result might
contribute enormously to the search for natural quantities to keep track of,
and to prove regularity or a priori estimates on solutions. This two-fold aspect
of the study makes it both challenging and complex, and requires the use of
several refined techniques to overcome the major difficulties encountered. The
applications of this subject are many while for the CIME course we had to
select only a few, trying however to cover some of the most relevant ones,
with interest ranging from the pure side (analysis/geometry) to the more
applied one (physics/biology). Here is a brief summary of the topics covered
in the courses of this school.
M. Gursky treated a class of elliptic equations from conformal geometry:
the general aim is to deform conformally (through a dilation which depends
on the point) the metric of a given manifold so that the new one possesses
special properties. Classical examples are the uniformization problem of two-
dimensional surfaces and the Yamabe problem in dimension greater or equal
to three, where one requires the Gauss or the scalar curvature to become

v
vi Preface

constant. After recalling some basic facts on these problems, which can be
reduced to semilinear elliptic PDEs, Gursky turned to their fully nonlin-
ear counterparts. These concern the prescription of the symmetric forms in
the eigenvalues of the Schouten tensor (a combination of the Ricci tensor
and the scalar curvature), and turn out to be elliptic under suitable con-
ditions on their domain of definition (admissible functions). The solvability
of these equations has concrete applications in geometry, since for exam-
ple they might guarantee pinching conditions on the Ricci tensor, together
with its geometric/topological consequences. After recalling some regularity
estimates by Guan and Wang, existence was shown using blow-up analy-
sis techniques. Finally, the functional determinant of conformally invariant
operators in dimension four was discussed: the latter turns out to have a uni-
versal decomposition into three terms which respectively involve the scalar
curvature, the Q-curvature and the Weyl tensor. Some conditions on the
coefficients of these three terms guarantee coercivity of the functional, and in
these cases existence of extremal metrics was obtained using a minimization
technique.
E. Lanconelli covered some topics on existence and sharp estimates on
heat kernels of subelliptic operators. Typically, in a domain or a manifold
Ω of dimension n, k vector fields X1 , . . . , Xk are given (with 2 ≤ k < n)
which satisfy the Hörmander condition, namely their Lie brackets span all
of the tangent spaces to Ω. One considers then linear operators L (or their
k
parabolic counterpart) whose principal part is given by i=1 Xi2 . During the
lectures, existence and regularity (Hörmander) theory for such operators was
recalled, and in particular the role of the Carnot-Caratheodory distance, mea-
sured through curves whose velocities belong to the linear span of the Xi ’s.
This distance is not homogeneous (at small scales), and it is very useful to
describe the degeneracy of the operators in the above form. One of the main
motivations for this study is the problem of prescribing the Levi curvature
of boundaries of domains in Cn , which for graphs amounts to solving a fully
nonlinear degenerate equation, whose linearization is of the form previously
described. Gaussian bounds for heat kernels were then given, first for constant
coefficient operators modeled on Carnot groups, and then for general oper-
ators using the method of the parametrix. Finally, applications to Harnack
type inequalities were derived in terms of the heat kernel bounds.
The course by A. Malchiodi on singularly perturbed Neumann problems
dealt with elliptic nonlinear equations where a small parameter (the singular
perturbation) is present in front of the principal term (the Laplacian). The
study is motivated by considering a class of reaction-diffusion systems (in par-
ticular the Gierer-Meinhardt model) and the (focusing) nonlinear Schrödinger
equation. First a finite-dimensional reduction technique, which incorporates
the variational structure of the problem, was presented: by means of this
method, existence of solutions concentrating at points of the boundary of
the domain was studied. Here the geometry of the boundary is significant,
as concentration occurs at critical points of the mean curvature. After this,
Preface vii

existence of solutions concentrating at the whole boundary was proved: the

phenomenon here is rather different, since the latter family has a diverg-
ing Morse index when the singular perturbation parameter tends to zero.
Initially, accurate approximate solutions were constructed (depending on the
second fundamental form of the domain), and then the invertibility of the lin-
earized equation was shown (primarily using Fourier analysis), which made
it possible to prove existence using local inversion arguments.
G. Tarantello’s course focused on self-dual vortices in Chern-Simons the-
ory. The physical phenomenon of superconductivity is described by a system
of coupled gauge-field equations whose components stand for the wave func-
tion and the electromagnetic potential. A relatively well understood model
is the abelian-Higgs (corresponding in a non-relativistic context to the
Ginzburg-Landau) variant, for which much has been accomplished even away
from the self-dual regime. A more sophisticated alternative is the Chern-
Simons model, which compared to the previous one has the advantage of
predicting the fact that gauge vortices carry electrical charge in addition to
magnetic flux, although its mathematical description is at the moment less
complete. After describing the main features of these models, Tarantello pre-
sented the approach of Taubes to the selfdual regime for the abelian-Higgs
case, which reduces the system to a semilinear elliptic problem with expo-
nential nonlinearities. This method partially extends to the Chern-Simons
case, where some natural requirements on solutions can be proved, like their
asymptotic behavior at infinity, integrability properties and the decay of their
derivatives. The structure of C-S vortices is however more rich (and, as we
remarked, far from being completely characterized) in comparison to the
abelian-Higgs ones: in addition to the topological solutions, which have a well
defined winding number at infinity, there are also non-topological solutions,
which display different asymptotic behavior.
X.J. Wang treated k-Hessian equations, a class of fully nonlinear equations
related to the problem of prescribing the Gauss curvature of a hypersur-
face, and to the Monge-Ampére equation, which is of interest in complex
geometry. First the class of admissible functions was defined, where the equa-
tions are elliptic, and then existence for Dirichlet boundary value problems
was obtained by means of a priori estimates and a continuation argument.
Next, interior gradient estimates were derived, which imply Harnack inequal-
ities, plus Sobolev-type inequalities for admissible functions which vanish on
the boundary of the domain: the embedding which follows from the latter
inequality possesses compactness properties analogous to the classical ones,
and makes it possible to derive L∞ estimates for solutions of equations with
sufficiently integrable right-hand sides. These estimates make it possible to
treat equations with nonlinear (subcritical or critical) reaction terms, where
min-max methods can be applied. After this, the notion of k-admissibility
was extended to non smooth functions using the concept of hessian measure,
and applied to the existence of weak solutions and to potential-theoretical
results. Finally, parabolic equations and several examples were treated.
viii Preface

The course by P. Yang concerned minimal surfaces in three-dimensional

CR manifolds, which possess a subriemannian structure modeled on the
Heisenberg group H1 . In this setting the volume form of M is naturally
defined in terms of the contact form θ as θ ∧ dθ: the p-area and the p-mean
curvature (p standing for pseudo) of a two-dimensional surface were defined
looking at the first and second variation of the volume form. p-minimal (regu-
lar) graphs in H1 were then considered, showing that they are ruled surfaces.
A study of the singular set (where the tangent plane to the surface coin-
cides with the contact plane, the kernel of θ) was then performed, and it was
shown that it consists either of isolated points or of smooth curves: applica-
tions were given to the classification of entire p-minimal graphs. Existence of
weak solutions (minimizers) to boundary value problems (of Plateau type)
was considered, showing the uniqueness, comparison principles and geometric
properties of the solutions. The regularity issue was then discussed, which is
a rather delicate one since the global C 2 regularity of minimizers might fail
in general.
Some of the students who attended the school were also supported by
SISSA, GNAMPA and by MiUR under the PRIN 2006 Variational methods
and nonlinear differential equations.
Finally, we would like to express our warm gratitude to the CIME Foun-
dation, to the CIME Director Prof. P. Zecca, to the CIME Board Secretary
Prof. E. Mascolo and to the CIME staff for their invaluable help and support,
and for making the environment in Cetraro so stimulating and enjoyable.

Alice S.Y. Chang

Antonio Ambrosetti
Andrea Malchiodi
List of Contributors

Matthew J. Gursky
Department of Mathematics, University of Notre Dame, Notre Dame,
IN 46556, USA, [email protected]
Ermanno Lanconelli
Dipartimento di Matematica, Universita’ degli Studi di Bologna, P.zza di
Porta S. Donato, 5 40127 - Bologna, Italy, [email protected]
Andrea Malchiodi
SISSA, Sector of Mathematical Analysis, Via Beirut 2-4, 34014 Trieste,
Italy, [email protected]
Gabriella Tarantello
Universita’ di Roma Tor Vergata-Dipartimento di Matematica, via della
ricerca scientifica, 00133 Rome, Italy, [email protected]
Xu-Jia Wang
Mathematical Sciences Institute, Australian National University, Canberra,
ACT 0200, Australia, [email protected]
Paul Yang
Department of Mathematics, Princeton University, Fine Hall, Washington
Road, Princeton, NJ 08544, USA, [email protected]

ix
Contents

PDEs in Conformal Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Matthew J. Gursky
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Some Background from Riemannian Geometry . . . . . . . . . . . . . . . . . . 2
2.1 Some Differential Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3 Some Background from Elliptic Theory . . . . . . . . . . . . . . . . . . . . . . . . 6
4 Background from Conformal Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 10
5 A Fully Nonlinear Yamabe Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.1 Ellipticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 From Lower to Higher Order Estimates . . . . . . . . . . . . . . . . . . . . 18
5.3 An Existence Result: Four Dimensions . . . . . . . . . . . . . . . . . . . . . 19
6 The Functional Determinant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.1 The Case of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.2 Four Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.3 The Euler Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.4 Existence of Extremals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.5 Sketch of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Heat Kernels in Sub-Riemannian Settings . . . . . . . . . . . . . . . . . . . . 35
Ermanno Lanconelli
1 Lecture Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2 A Motivation: Levi Curvatures and Motion by Levi Curvatures . . . . 38
3 Heat Kernels and Gaussian Bounds: Past History . . . . . . . . . . . . . . . . 40
4 Carnot-Carathèodory metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5 Heat-Type Operators with Constant Coefficients . . . . . . . . . . . . . . . . 43
6 Heat-Type Operators with Variable Coefficients.
The Levi Parametrix Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7 Hadamard-Pini Invariant Harnack Inequality . . . . . . . . . . . . . . . . . . . . 49
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

xi
xii Contents

Concentration of Solutions for Some

Singularly Perturbed Neumann Problems . . . . . . . . . . . . . . . . . . . . . 63
Andrea Malchiodi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Part I Concentration at Points

2 Perturbation in Critical Point Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 70

2.1 The Finite-Dimensional Reduction . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.2 Existence of Critical Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3 Application to the Study of P̃ε . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
 
3.1 Study of Ker[I (U )] and Ker[I + (U )] . . . . . . . . . . . . . . . . . . . . . . 76
3.2 Proof of Theorem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3 Proof of Theorem 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

Part II Higher-Dimensional Concentration

4 Some Preliminary Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5 An Approximate Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6 Eigenvalues of the Linearized Equation . . . . . . . . . . . . . . . . . . . . . . . . . 107
7 Proof of Theorem 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

On Some Elliptic Problems in the

Study of Selfdual Chern-Simons Vortices . . . . . . . . . . . . . . . . . . . . . 117
Gabriella Tarantello
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2 The Abelian-Higgs Model and The Chern-Simons-Higgs Model . . . . 119
3 Selfduality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4 On the Elliptic Problem Concerning Abelian-Higgs Vortices . . . . . . . 132
5 On the Elliptic Problem of Selfdual Chern-Simons Vortices . . . . . . . 145
5.1 Topological Chern-Simons Voritces . . . . . . . . . . . . . . . . . . . . . . . . 147
5.2 Non Topological Chern-Simons Vortices . . . . . . . . . . . . . . . . . . . . 154
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

The k-Hessian Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

Xu-Jia Wang
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
2 Admissible Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
2.1 Admissible Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
2.2 Admissible Solution is Elliptic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
2.3 Concavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
2.4 A Geometric Assumption on the Boundary . . . . . . . . . . . . . . . . . 182
2.5 Some Algebraic Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Contents xiii

3 The Dirichlet Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

3.1 A priori Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
3.2 Regularity for Fully Nonlinear, Uniformly Elliptic Equation . . . 188
3.3 Existence of Smooth Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
3.4 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4 Interior a Priori Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.1 Interior Gradient Estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.2 Harnack Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.3 Interior Second Derivative Estimate . . . . . . . . . . . . . . . . . . . . . . . 194
5 Sobolev Type Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.1 Sobolev Type Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.2 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5.3 An L∞ Estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6 Variational Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
6.1 The Sublinear Growth Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.2 The Eigenvalue Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
6.3 Superlinear Growth Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
6.4 The Critical Growth Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
7 Hessian Integral Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.1 A Basic Estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.2 Local Integral Gradient Estimates . . . . . . . . . . . . . . . . . . . . . . . . . 225
8 Hessian Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
8.1 Non-Smooth k-Admissible Functions . . . . . . . . . . . . . . . . . . . . . . . 230
8.2 Perron Lifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
8.3 Weak Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
8.4 The Dirichlet Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
9 Local Behavior of Admissible Functions . . . . . . . . . . . . . . . . . . . . . . . . 239
9.1 The Wolff Potential Estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
9.2 Hölder Continuity of Weak Solutions . . . . . . . . . . . . . . . . . . . . . . . 241
10 Parabolic Hessian Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
11 Examples of Fully Nonlinear Elliptic Equations . . . . . . . . . . . . . . . . . . 248
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250

Minimal Surfaces in CR Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 253

Paul Yang
1 Pseudo-Hermitian Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
2 Area of a Surface in (M 3 , θ, J) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
3 Structure of the Singular Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
4 Some Applications of Theorem A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
5 Weak Solutions and Condition for Minimizer . . . . . . . . . . . . . . . . . . . . 265
6 Existence and Uniqueness for Boundary Value Problems . . . . . . . . . . 268
7 Regularity of the Regular Part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
PDEs in Conformal Geometry

Matthew J. Gursky

1 Introduction

In these lectures I will discuss two kinds of problems from conformal geometry,
with the goal of showing an important connection between them in four
dimensions.
The first problem is a fully nonlinear version of the Yamabe problem,
known as the σk -Yamabe problem. This problem is, in general, not varia-
tional (or at least there is not a natural variational interpretation), and the
underlying equation is second order but possibly not elliptic. Moreover, in
contrast to the Yamabe problem, there is very little known (except for some
examples and counterexamples) when the underlying manifold is negatively
curved.
The second problem we will discuss involves the study of a fourth order
semilinear equation, and arose in the context of a natural variational prob-
lem from spectral theory. Despite their differences–higher order semilinear
versus second order fully nonlinear, variational versus non-variational–both
equations are invariant under the action of the conformal group, and we have
to address the phenomenon of “bubbling.” Therefore, in the first few sections
of the notes we will present the necessary background material, including a
careful explanation of the idea of a “standard bubble”.
After covering the introductory material, we give a description of the
σk -Yamabe problem, culminating in a sketch of the solution in the four-
dimensional case. Modulo some technical regularity estimates, the proof
is reduced to a global geometric result (Theorem 5.7) that is easy to
understand.

M.J. Gursky
Department of Mathematics, University of Notre Dame, Notre Dame, IN 46556
e-mail: [email protected]
Research supported in part by NSF Grant DMS-0500538.

S.-Y.A. Chang et al. (eds.), Geometric Analysis and PDEs, 1

Lecture Notes in Mathematics 1977, DOI: 10.1007/978-3-642-01674-5 1,

c Springer-Verlag Berlin Heidelberg 2009
2 M.J. Gursky

In the last section of the notes we discuss the functional determinant of a

four-manifold, a variational problem which is based on a beautiful calculation
of Branson-Ørsted. We end with a sketch of the existence of extremals for the
determinant functional for manifolds of positive scalar curvature. Here, the
missing technical ingredient is a sharp functional inequality due to Adams
(Theorem 65), but the proof is again reduced to Theorem 5.7. Therefore, we
see the underlying unity of the two problems in a very concrete way.
In closing, I wish to express my gratitude to the Fondazione C.I.M.E.
for their invitation and their support. The success of the meeting Geomet-
ric Analysis and PDEs was a result of the considerable efforts of the local
organizers (especially Andrea Malchiodi), the scientific contributions of the
participants, and the hospitality of our hosts in Cetraro.

2 Some Background from Riemannian Geometry

In this section we review some of the basic notions from Riemannian geom-
etry, including the basic differential operators (gradient, Hessian, etc.) and
curvatures (scalar, Ricci, etc.) This is not so much an introduction to the
subject–which would be impossible in so short a space–but rather a summary
of definitions and formulas.

2.1 Some Differential Operators

1. The Hessian

Let (M n , g) be an n-dimensional Riemannian manifold, and let ∇ denote the

Riemannian connection.

Definition 2.1. The Hessian of f : M n → R is defined by

∇2 f (X, Y ) = ∇X df (Y ). (1)

It is easy to see the Hessian is symmetric, bilinear form on the tangent

space of M n at each point. In a local coordinate system {xi }, the Christoffel
symbols are defined by
∂  ∂
∇ ∂
j
= Γijk k .
∂xi ∂x ∂x
k
PDEs in Conformal Geometry 3

Using (1), in local coordinates we have

(∇2 f )ij = ∇i ∇j f
∂2f  ∂f
= i j
− Γijk k .
∂x ∂x ∂x
k

2. The Laplacian and Gradient

Definition 2.2. The Laplacian is the trace of the Hessian: Let {e1 , . . . , en }
be an orthonormal basis of the tangent space at a point; then

Δf = ∇2 f (ei , ei ). (2)
i

In local coordinates {xi },

 
∂2f  ∂f
Δf = g ij
− Γijk k ,
∂xi ∂xj ∂x
k

where g ij = (g −1 )ij . Another useful formula is

 
1 ∂ ij √ ∂f
Δf = √ g g j ,
g ∂xi ∂x

where g = det(gij ).
The gradient vector field of f , denoted ∇f , is the vector field dual to the
1-form df ; i.e., for each vector field X,

g(∇f, X) = df (X).

In local coordinates {xi },

 ∂f
∇j f = g ij .
i
∂xi

3. The Curvature Tensor

For vector fields X, Y, Z, the Riemannian curvature tensor of (M, g) is
defined by

R(X, Y )Z = ∇[X,Y ] Z − [∇X , ∇Y ]Z,

where [·, ·] is the Lie bracket. With respect to a local coordinate system {xi },
the curvature tensor is given by
4 M.J. Gursky
  
∂ ∂ ∂ i ∂
R , = Rjkl .
∂xk ∂xl ∂xj i
∂x i

Let Π ⊂ Tp M n be a tangent plane with orthonormal basis {E1 , E2 }. The

sectional curvature of Π is the number

K(Π) = R(E1 , E2 )E1 , E2 .

(K(Π) does not depend on the choice of ON-basis.)

Example 1. For Rn with the Euclidean metric, all sectional curvatures
are zero.

Example 2. Let Sn = {x ∈ Rn+1 | x = 1} with the metric it inherits as

a subspace of Rn+1 . Then all sectional curvatures are +1.

Example 3. Let Hn = {x ∈ Rn | x < 1}, endowed with the metric

 (dxi )2
g=4 .
i
(1 − x2 )2

Then all sectional curvatures are −1.

The preceding examples are referred to as spaces of constant curvature, or
space forms. A theorem of Hopf says that any complete, simply connected
manifold of constant curvature is isometric to one of these examples (perhaps
after scaling). Thus, curvature determines the local geometry of a manifold.
Another way of thinking about curvature is that it measures the failure of
derivatives to commute:

Lemma 2.3. In local coordinates,


∇i ∇j ∇k f − ∇j ∇i ∇k f = m
Rkij ∇m f.
m

So third derivatives do not commute unless R = 0, i.e., the manifold

is flat.

4. Ricci and Scalar Curvatures

Definition 2.4. The Ricci curvature tensor is the bilinear form Ric : Tp M ×
Tp M → R defined by

Ric(X, Y ) = R(X, ei )Y, ei ,
i

where {e1 , . . . , en } is an orthonormal basis of Tp M .

PDEs in Conformal Geometry 5

In local coordinates, the components of Ricci are given by


m
Rij = Rijm .
m

For spaces of constant curvature, the Ricci tensor is just a constant multiple
of the metric:

Sn : Ric = (n − 1)g,
Rn : Ric = 0,
Hn : Ric = −(n − 1)g.

The Ricci tensor is symmetric: Ric(X, Y ) = Ric(Y, X). Therefore, at each

point p ∈ M we can diagonalize Ric with respect to an orthonormal basis
of Tp M :

 ρ1 
ρ2
Ric = ..
.
ρn

where (ρ1 , . . . , ρn ) are the eigenvalues of Ric. To say that (M n , g) has positive
(negative) Ricci curvature means that all the eigenvalues of Ric are positive
(negative).
In two dimensions, the Ricci curvature is determined by the Gauss curva-
ture K:

Ric = Kg.

Definition 2.5. The scalar curvature is the trace of the Ricci curvature:

R= Ric(ei , ei ),
i

where {e1 , . . . , en } is an orthonormal basis.

If {ρ1 , . . . , ρn } are the eigenvalues of the Ricci curvature at a point p ∈ M ,
then the scalar curvature is given by

R = ρ 1 + · · · + ρn .

For the spaces of constant curvature, the scalar curvature is a constant

function:

Sn : R = n(n − 1),
Rn : R = 0,
Hn : R = −n(n − 1).
6 M.J. Gursky

Furthermore, in two dimensions the scalar curvature is twice the Gauss

curvature:

R = 2K.

3 Some Background from Elliptic Theory

In this section we summarize some important results from functional analysis

and the theory of partial differential equations.

1. Sobolev Spaces
These are important for discussing some of the PDE topics in these lectures.
Let (M, g) be a compact Riemannian manifold. For 1 ≤ k < ∞ and 1 ≤ p ≤ ∞,
introduce the norms
 
upk,p = |∇j u|p dV,
0≤j≤k

where ∇j u denotes the iterated j th -covariant derivative.

Example. For k = 1, p = 2,

 
u21,2 = u2 dV + |du|2 dV.

The Sobolev space W k,p (M ) is the completion of C ∞ (M ) in the norm

 · k,p .

Theorem 3.1. (Sobolev Embedding Theorems; see [GT83])

(i) If

1 1 k
= − ,
r m n
then W k,m (M ) is continuously embedded in Lr (M ):

ur ≤ Cuk,m .

(ii) Suppose 0 < α < 1 and

1 k−α
≤ .
m n
Then W k,m is continuously embedded in C α .
PDEs in Conformal Geometry 7

(iii) (Rellich-Kondrakov) If

1 1 k
> − ,
r m n
then the embedding W k,m → Lr is compact: i.e., a sequence which is bounded
in W k,m has a subsequence which converges in Lr .

2. Linear Operators
Consider the linear differential operator L:

Lu = aij (x)∂i ∂j u + bk (x)∂k u + c(x)u,

where the coefficients aij , bk , c are defined in a domain Ω ⊂ Rn .

Definition 3.2. The operator L is elliptic in Ω if {aij (x)} is positive definite

at each point x ∈ Ω. If there is a constant λ > 0 such that

aij (x)ξi ξj ≥ λ|ξ|2

for all ξ ∈ Rn and x ∈ Ω, then L is strictly elliptic in Ω. If, in addition, there

is another constant Λ > 0 such that

Λ|ξ|2 ≥ aij (x)ξi ξj ≥ λ|ξ|2 ,

then we say that L is uniformly elliptic in Ω.

We can formulate a similar definition for operators defined on a Riemannian

manifold; e.g., by introducing local coordinates. Of course, the laplacian
L = Δ is an example of a linear, uniformly elliptic operator.
3. Weak Solutions
We say that u ∈ W 1,2 (M ) is a weak solution of the equation

Δu = f (x) (3)

if for each ϕ ∈ W 1,2 ,

 
−∇u, ∇ϕ dV = f ϕ dV. (4)

Of course, a smooth solution of (3) satisfies (4) by virtue of Green’s Theo-

rem. Weak solutions of elliptic equations like (3) in fact satisfy much better
estimates, as we shall see.
8 M.J. Gursky

4. Elliptic Regularity
Theorem 3.3. (See [GT83]) Suppose u ∈ W 1,2 is a weak solution of

Δu = f

on M .
(i) If f ∈ Lm , then

u2,m ≤ C f m + um . (5)

(ii) (Schauder estimates) If f ∈ C ,α then

uC +2,α ≤ C f C ,α + uC ,α . (6)

How are such estimates used?

• To prove the regularity of weak solutions.

Weak solutions are often easier to find, for example, by variational methods.

• To prove a priori estimates of solutions, that is, estimates which are

necessarily satisfied by any solution of a given equation.

Often a priori estimates can be combined with a topological argument to

establish existence.

Example. To illustrate some of these results we consider an equation that

will be an important model for much of the subsequent material.
Theorem 3.4. Suppose u ≥ 0 is a (weak) solution of

Δu + c(x)u = K(x)up , (7)

where c, K are smooth functions, and

n+2
1≤p< .
n−2
If 
2n
u (n−2) dV ≤ B, (8)

then u satisfies

sup u ≤ C(p, B).

M
PDEs in Conformal Geometry 9

In fact, we can estimate u with respect to any Hölder norm, all in terms
n+2
of p and B. The main point here is that the assumption p < n−2 is crucial.

Proof. Using the preceding elliptic regularity theorem, we know that u

satisfies

u2,m ≤ C Δum + um

≤ C up m + um (9)

≤ C ump + um .
p

Denote
2n
m0 = ,
n−2
and choose m so that
mp = m0 .
It follows from (9) that

u2,m ≤ C(p, B).

We now use the Sobolev embedding theorem, which says

ur ≤ Cu2,m

where
1 1 2 n − 2m
= − = ,
r m n mn
or,

mn ( mp0 )n
r= = .
n − 2m n − 2( mp0 )

So, we’ve passed from one Lebesgue-space estimate to another. Have things
improved?
The answer is yes, as long as

( mp0 )n
> m0 .
n − 2( mp0 )

Solving this inequality, we see that it will hold provided p satisfies

n+2
p< .
n−2
10 M.J. Gursky

In this case, we iterate this process an arbitrary number of times, and

conclude that

u2,m ≤ C(m, p, B) ∀m >> 1.

Once m is large enough, though, we can once more appeal to the Sobolev
embedding theorem, part (ii), and conclude that u is Hölder continuous–in
particular, u is bounded as claimed.

Remarks.
1. For higher order regularity we turn to the Schauder estimates, since we
actually proved that u is Hölder continuous. Iterating the Schauder estimates,
we can prove the Hölder continuity of derivatives of all orders.

2. As we mentioned above, and will soon see by explicit example, the pre-
ceding result is false if p = (n + 2)/(n − 2). However, it can be “localized”:
that is, if

2n
u (n−2) dV ≤ 0
B(x0 ,r)

for some 0 > 0 small enough, then

sup u ≤ C(r).
B(x0 ,r/2)

3. A Corollary of Theorem 3.4 is that weak solutions of (7) are regular, for
all 1 ≤ p ≤ (n + 2)/(n − 2).

4 Background from Conformal Geometry

In this, the final section of the introductory material, we present some basic
ideas from conformal geometry.
Definition 4.1. Let (M n , g) be a Riemannian manifold. A metric h is
pointwise conformal to g (or just conformal) if there is a function f such
that

h = ef g.

The function ef is referred to as the conformal factor. We used the expo-

nential function to emphasize the fact that we need to multiply by a positive
function (since h must be positive definite). However, in some cases it will be
more convenient to write the conformal factor differently.
PDEs in Conformal Geometry 11

We can introduce an equivalence relation on the set of metrics: h ∼ g iff h

is pointwise conformal to g. The equivalence class of a metric g is called its
conformal class, and will be denoted by [g]. Note that

[g] = {ef g | f ∈ C ∞ (M )}.

Definition 4.2. Let (M, g) and (N, h) be two Riemannian manifolds. A

diffeomorphism ϕ : M → N is called conformal if

ϕ∗ h = ef g.

We say that (M, g) and (N, h) are conformally equivalent. Note h and g
are pointwise conformal if and only if the identity map is conformal.
Example 1. Let δλ (x) = λ−1 x be the dilation map on Rn , where λ > 0.
Then δλ is easily seen to be conformal; in fact,

δλ∗ ds2 = λ−2 ds2 ,

where ds2 is the Euclidean metric.

Example 2. Let P = (0, . . . , 0, 1) be the north pole of Sn ⊂ Rn+1 . Let

σ : Sn \ {P } → Rn+1 denote stereographic projection, defined by

ζ1 ζn
σ(ζ 1 , . . . , ζ n , ξ) = ,..., ).
1−ξ 1−ξ

Then σ : (Sn \ {P }, g0 ) → (Rn+1 , ds2 ) is conformal, where g0 is the standard

metric on Sn .
Since the composition of conformal maps is again conformal, we can use
σ to construct conformal maps of Sn to itself: for λ > 0, let

ϕλ = σ −1 ◦ δλ ◦ σ : Sn → Sn .

Then
ϕ∗λ g0 = Ψλ2 g0 ,
where
2λ
Ψλ (ζ, ξ) = .
(1 + ξ) + λ2 (1 − ξ)

Note

(ζ, ξ) = (0, 1) ⇒ Ψλ → ∞ as λ → ∞,
(ζ, ξ) = (0, 1) ⇒ Ψλ → 0 as λ → ∞.
12 M.J. Gursky

The set of conformal maps of a given Riemannian manifold is a Lie group;

the construction above shows that the conformal group of the sphere is non-
compact. This fact distinguishes the sphere:

Theorem 4.3. (Lelong-Ferrand) A compact Riemmanian manifold with non-

compact conformal group is conformally equivalent to the sphere with its
standard metric.

This fact is the source of many of the analytic difficulties we will encounter
in the PDEs we are about to describe.

1. Curvature and Conformal Changes of Metric

Let h = e−2u g be conformal metrics, and let Ric(h), R(h) denote the Ricci
and scalar curvatures of h, and Ric(g), R(g) denote the Ricci and scalar
curvatures of g. Then

Ric(h) = Ric(g) + (n − 2)∇2 u + Δug

+(n − 2)du ⊗ du − (n − 2)|∇u|2 g,

R(h) = e 2u
R(g) + 2(n − 1)Δu

−(n − 1)(n − 2)|∇u|2 ,

where ∇2 u and Δu denote the Hessian and laplacian of u with respect to g.

2. The Uniformization Theorem and Yamabe Problem.

Let (M 2 , g) be a closed (no boundary), compact, two-dimensional Riemannian
manifold. Let K denotes its Gauss curvature.

Theorem 4.4. (The Uniformization Theorem) There is a conformal metric

h = e−2u g with constant Gauss curvature.

See ([Ber03], p. 254) for some historical background on the result. Let
Kh = const. denote the Gauss curvature of the metric h; then the sign of Kh
is determined by the Gauss-Bonnet formula:

2πχ(M 2 ) = Kh dAh

= Kh · Area(h).

Note the geometric/topological significance of the Uniformization Theo-

rem: Since h has constant curvature, by the Hopf theorem the universal cover
M̃ is isometric to either S2 , R2 , or H2 , each case being determined by the
sign of the Euler characteristic.
PDEs in Conformal Geometry 13

Now let (M n , g) be a closed, compact, Riemannian manifold of dimension

n ≥ 3. In higher dimensions there are obstructions to being even locally
conformal to a constant curvature metric. This leads to

Question: How do we generalize the Uniformization Theorem to higher

dimensions?
A major theme of these lectures is the various ways one might answer
this question (there are yet others). The first attempt we will discuss is the
Yamabe Problem: Find a conformal metric h = e−2u g whose scalar curvature
is constant.
By the formulas above, solving the Yamabe problem is equivalent to
solving the semilinear PDE

2(n − 1)Δu − (n − 1)(n − 2)|∇u|2 + R(g) = μe−2u

for some constant μ. This formula can be simplified if we write h = v 4/(n−2) g,

where v > 0. Then v should satisfy

4(n − 1) n+2
− Δv + R(g)v = λv n−2 . (10)
(n − 2)

Notice the exponent! This equation is of the form

Δv + c(x)v = K(x)v p ,

where p = (n + 2)/(n − 2). This is the critical case of the equation we

considered in Theorem 3.4.

3. The Case of the Sphere

Recall the conformal maps of the sphere described above, ϕλ : Sn → Sn .
Then h = ϕ∗λ g0 = Ψλ2 g0 has the same scalar curvature as the standard metric.
Therefore, writing
4/(n−2)
h = vλ g0 ,

where
(n−2)
2
vλ = Ψλ ,

we have a family {vλ } of solutions to

4(n − 1) n+2
− Δvλ + n(n − 1)vλ = n(n − 1)vλn−2 .
(n − 2)

As we observed above, if P is the North pole, then

14 M.J. Gursky

vλ (P ) → ∞ as λ → ∞,

whereas if x = P , then

vλ (x) → 0 as λ → ∞.

To summarize, there is good news and bad. The good news is that there
are many solutions of the Yamabe equation. The bad news is that it will
be impossible to prove a priori estimates for solutions of (10). Of course,
the non-compactness of the set of solutions arises precisely because of the
influence of the conformal group. Thus, on manifolds other than the sphere,
one would expect that the set of solutions is compact. Put another way,
ideally we would like to show that non-compactness implies the underlying
manifold is (Sn , g0 ).

2. The Yamabe Problem: A variational Approach.

There is an approach to solving the Yamabe problem by the methods of the
calculus of variations. Define the functional Y : W 1,2 → R by
4(n−1)

(n−2) |∇v| + R(g)v

2 2
dV
Y[v] =   (n−2)/n
. (11)
2n
v (n−2) dV

Using the formulas above, one can check that


Y[v] = V ol(h)−(n−2)/n R(h) dV (h),

where h = v 4/(n−2) g. The quantity on the right-hand side is called the total
scalar curvature of h.

Lemma 4.5. A function v ∈ W 1,2 is a critical point of Y iff v is a weak

solution of the Yamabe equation.
By critical point, we mean that
d 
Y(v + tϕ)t=0 = 0
dt
for all ϕ ∈ W 1,2 .
Recall that weak solutions of (10) are regular. Also, by the Sobolev
embedding theorem the number

Y (M n , [g]) = inf Y(v) (12)

v∈W 1,2

is > −∞. This number is called the Yamabe invariant of the conformal class
of g.
PDEs in Conformal Geometry 15

Some historical notes: H. Yamabe claimed to have proved the existence of

a minimizer of Y, for all manifolds (M n , g). However, N. Trudinger found a
serious gap in his proof, which he was able to fix provided Y (M n , [g]) was
sufficiently small (for example, if Y (M n , [g]) ≤ 0). Subsequently, T. Aubin
proved that for all n-dimensional manifolds

Y (M n , [g]) ≤ Y (Sn , [g0 ]), (13)

and that whenever this inequality was strict, a minimizing sequence converges
(weakly) to a (smooth) solution of the Yamabe equation. Aubin also showed
that a strict inequality holds in (13) if (M n , g) was of dimension n ≥ 6 and
not locally conformal to a flat metric.
Finally, the remaining cases were solved by Schoen: that is, he showed
that when M n has dimension 3, 4, or 5, or if M is locally conformal to a
flat metric, then the inequality (13) is strict unless (M n , g) is conformally
equivalent to (S n , g0 ). An excellent survey of the Yamabe problem can be
found in [LP87].

5 A Fully Nonlinear Yamabe Problem

In this section we begin our discussion of the σk -Yamabe problem, a more

recent attempt to generalize the Uniformization Theorem to higher dimen-
sions. To do so, we need to introduce another notion of curvature:

Definition 5.1. The Schouten tensor of (M, g) is

 
1 1
A= Ric − R·g . (14)
(n − 2) 2(n − 1)

Example. For spaces of constant curvature ±1 (e.g. the sphere or hyperbolic

space), the Schouten tensor is
 
1 1
A = diag ± , . . . , ± .
2 2

From the perspective of conformal geometry, the Schouten is actually more

natural than the Ricci tensor (but this takes some time to explain). Here’s
g = e−2u g. Then the Schouten tensor of 
one indication: Suppose  g is given by

 = A + ∇2 u + du ⊗ du − 1 |du|2 g.
A (15)
2
A complicated formula; but just think of it as saying
 = ∇2 u + · · ·
A
16 M.J. Gursky

where · · · indicates lower order terms. Contrast this with the more com-
plicated formulas for the Ricci tensor, which also involves the Laplace
operator.
The equations we will consider involve symmetric functions of the eigen-
values of A. Let λ1 , . . . , λn denote the eigenvalues of A; suppose we choose a
local basis which diagonalizes A:

 λ1 
λ2
A= .. .
.
λn

Then define

σk (A) = λi1 · · · λik , (16)
i1 <···<ik

i.e., σk is the k th elementary symmetric polynomial in n variables. Note that

R
σ1 (A) = trace(A) = ,
2(n − 1)

just a multiple of the scalar curvature. In general, the quantity σk (A) is called
the k th -scalar curvature, or σk -curvature, of the manifold.
Now, we can rephrase the Yamabe problem in the following way: Given
(M n , g) find a conformal metric g = e−2u g with constant σ1 -curvature. This
naturally leads to the σk -Yamabe problem: Given (M n , g), find a conformal
metric  g = e−2u g such that the σk -curvature is constant. By the formula
above, this is equivalent to solving the PDE
1

σk A + ∇2 u + du ⊗ du − |du|2 g = μe−2ku (17)

2
for some constant μ. Note the exponential weight on the right comes from
the fact that we are computing the eigenvalues of A  w.r.t. g.
These equations are closely related to the Hessian equations covered in
Prof. Xu-Jia Wang’s C.I.M.E. course. The differences will come from (1) The
conformal invariance, and (2) The lower order (gradient) terms.
The σk -Yamabe problem was first formulated by J. Viaclovsky in his thesis
[Via00]. Viaclovsky is also the author of a recent survey article on the subject,
[Via06].
PDEs in Conformal Geometry 17

5.1 Ellipticity

Recall from Professor Wang’s lectures that the Hessian equations

σk (∇2 u) = f (x)

are elliptic provided u is admissible, or k-convex. That is, if

σj (∇2 u) > 0, 1 ≤ j ≤ k.

In particular, a necessary condition is that f (x) > 0. We will need to impose

a similar ellipticity condition:

Definition 5.2. A metric g is admissible (or k-admissible) if the Schouten

tensor satisfies

σj (A) > 0, 1≤j≤k

at each point of M n .

What is the geometric meaning of admissibility? One can think of it as

a kind of “positivity” condition on the Schouten tensor. When k = n, it
means the Schouten tensor is positive definite; when k = 1, it means the
trace (i.e., the scalar curvature) is positive. Here is a more precise result, due
to Guan-Viaclovsky-Wang [GVW03]:

Theorem 5.3. If (M n , g) is k-admissible then

2k − n
Ric ≥ R · g.
2n(k − 1)

In particular, if k > n/2 then admissibility means positive Ricci curvature.

We can also define negative admissibility, which just means that (−A) is
k-convex.
As in the usual Yamabe problem, there is a non-compact family of solutions
to the σk -Yamabe problem on S n :

gλ = ϕ∗λ g0 = Ψλ2 g0 .

In particular, this gives an obstruction to proving a priori estimates (as it does

for the Yamabe problem). Thus, we are faced with some of the same technical
difficulties. However, there are some important technical differences between
the σk - and classical Yamabe problems. For example, equation (17) does
not have an easy variational description (though there are some important
geometric cases where it does).
A more mysterious contrast arises when studying manifolds of negative
curvature. If (M n , g) has negative scalar curvature, the Yamabe problem is
18 M.J. Gursky

very easy to solve–indeed, the solution is unique. But for negative admissible
metrics there are at this time no general existence results for the σk -Yamabe
problem. In fact, Sheng-Trudinger-Wang showed by example that the local
estimates of Guan-Wang are false for solutions in the negative cone (see
[STW05]).
Finally, we remark that the condition of admissibility can be very restric-
tive: for example, the manifold X 3 = S 2 × S 1 does not admit a k-admissible
metric for k = 2 or 3. Of course, one can consider the Yamabe problem for
any conformal class on X 3 .

5.2 From Lower to Higher Order Estimates

Our goal is to explain the main issues involved in solving the σk -Yamabe
problem, and sketch the proof of a particular case. As we shall see, the central
problem is establishing a priori estimates. Owing to a fundamental result of
Evans, Krylov ([Eva82], [Kry83]), plus the classical Schauder estimates, we
only need to worry about estimating derivatives up to order two. That is,

|u| + |∇u| + |∇2 u| ≤ C2

⇓
|u| + |∇u| + · · · + |∇k u| ≤ C(k, C2 ).

Of course, even C 2 -estimates will fail without further assumptions, again

because of the sphere. However, let’s look closer: Let ϕλ : S n → S n be the
1-parameter family of conformal maps, and write

gλ = ϕ∗λ g0 = e−2uλ g0 .

Note that as λ → ∞, the conformal factor grows like

max e−2uλ ∼ λ2 ,

while the gradient and Hessian of u grow like

max |∇uλ |2 ∼ λ2 , max |∇2 uλ | ∼ λ2 .

In particular, for this family we have

|∇u|2 + |∇2 u| ≈ max e−2uλ .

So the optimal estimate one could hope for would be

max 2nd derivatives of u ≤ C max e−2u . (18)

PDEs in Conformal Geometry 19

It turns out that such an estimate always holds:

Theorem 5.4. (See Guan-Wang, [GW03]) Assume u ∈ C 4 is an admissible

solution of the σk -Yamabe equation on B(1). Then
 
max |∇u|2 + |∇2 u| ≤ C(1 + max e−2u ).
B(1/2) B(1)

In view of this result, and the Evans and Krylov results, we see that

min u ≥ C ⇒ uC k,α(M) ≤ C(k).

M

Therefore, if we can somehow rule out “bubbling”, we obtain estimates of

all orders. Once estimates are known, there are various topological methods
to prove the existence of solutions. This shows the geometric nature of the
problem: i.e., we need to detect the global geometry of the manifold in order
to get estimates, hence existence.

5.3 An Existence Result: Four Dimensions

To finish our discussion of the σk -Yamabe problem, we want to sketch its

solution in four dimensions. This case is special because, in 4-d, the integral

σ2 (A) dV

is conformally invariant. That is, if g = e−2u g, then

 
 dV = σ2 (A) dV.
σ2 (A)

You can check this by hand using the formulas above along with the fact that

dV = e−4u dV.

Eventually, you will find that

 dV = σ2 (A) dV + (divergence terms).
σ2 (A)

We will provide some details for the case k = 2; this was first treated
by Chang-Gursky-Yang [CGY02b], and later by Gursky-Viaclovsky [GV04].
For k = 3 or 4, the scheme of the proof is essentially the same. However,
the proof presented here is a simplified version of the original one, since we
will use the local estimates of Guan-Wang (which appeared several years
20 M.J. Gursky

after [CGY02b]). As we emphasized above, existence eventually boils down

to estimates: this is what we will prove.
To begin, let us write the equation in the case k = 2:

1/2 1

σ2 A + ∇2 u + du ⊗ du − |du|2 g = f (x)e−2u . (19)

2
where f ∈ C ∞ . Using the definition of σ2 , this actually reads:

−|∇2 u|2 + (Δu)2 + c1 ∇i ∇j u∇i u∇j u

+ c2 Δu|∇u|2 + c3 |∇u|4 + · · · = f 2 (x)e−4u .

We will prove:

Theorem 5.5. Suppose (M 4 , g) is (i) admissible, and (ii) not conformally

equivalent to the round sphere. If u ∈ C 4 is a solution of (19), then there is
a constant C = C(g, f ) such that

min u ≥ −C.

Consequently,

uC k ≤ C(k).

Proof. Suppose to the contrary there is a sequence of solutions {ui } of (19)

with min ui → −∞. Let’s imagine that there is a point P with

min ui = ui (P )
M

and by introducing local coordinates we can identify P with the origin in R4

and think of ui as being defined in a neighborhood Ω of 0. (In reality, the
location of the minimum point will vary, but this doesn’t affect the argument
in a significant way).
It is time to use conformal invariance. Recall the dilations on Euclidean
space are conformal. Define
1
wi (x) = u(i x) + log ,
i
where i > 0 is chosen so that

wi (0) = 0.
1
The wi ’s are defined on i Ω, and satisfy

1/2 2 1

σ2 i A + ∇2 wi + dwi ⊗ dwi − |dwi |2 g

2
= f (i x)e−2wi .
PDEs in Conformal Geometry 21

After applying the local estimates of Guan-Wang, we can take a subsequence

k,α
{wi } which converges in Cloc to a solution of

1/2 1

σ2 ∇2 w + dw ⊗ dw − |dw|2 g = μe−2w . (20)

2
with μ > 0.
We now appeal to the following uniqueness result

Lemma 5.6. (See Chang-Gursky-Yang, [CGY02a]) Up to scaling, the unique

solution of (20) is realized by

e2w ds2 = (σ −1 )∗ g0 , (21)

where σ is the stereographic projection map, ds2 the Euclidean metric, and
g0 is the round metric on the sphere.

It is easy to check that each solution given by (21) satisfies


 dV = 4π 2 .
σ2 (A)
R4

where  g = e2w ds2 . (Remember that A = diag{1/2, . . . , 1/2}, and Vol(S 4 ) =

8π 2 /3). Also, since our solution w comes from blowing up a little piece of the
original manifold, for each gi = e−2ui g we must have

i ) dVi ≥ 4π 2 .
σ2 (A
M4

The proof now follows from the following global geometric result:

Theorem 5.7. (See Gursky, [Gur99]) If (M 4 , g) has positive scalar curva-

ture, then

σ2 (A) dV ≤ 4π 2 ,
M4

and equality holds if and only if (M 4 , g) is conformally equivalent to the

sphere.

It follows that each (M 4 , gi ) is conformally equivalent to the round sphere,

a contradiction. Therefore, assuming the manifold (M 4 , g) is not conformally
the sphere, any sequence of solutions remains bounded, as claimed.

Important Remark. The following remark is for the benefit of experts:

The proof of the preceding theorem does not use the Positive Mass Theorem!
(Or, to be precise, it uses an extremely weak form). Therefore, we are not
22 M.J. Gursky

solving the σk -Yamabe problem by somehow reducing it to the classical

Yamabe problem.

6 The Functional Determinant

In the final section we will introduce a higher order elliptic problem which has
its origins in spectral theory. Although this problem is semilinear and not fully
nonlinear, the structure of the Euler equation is related to the σ2 -Yamabe
equation in 4-d. Moreover, for 4-manifolds of positive scalar curvature, the
same result (Theorem 5.7) plays a crucial role in the existence theory.
Suppose (M n , g) is a closed Riemannian manifold, and let Δ denote the
Laplace-Beltrami operator associated to g. We can label the eigenvalues of
(−Δ) (counting multiplicities) as

0 = λ0 < λ1 ≤ λ2 ≤ λ3 ≤ . . . (22)

The spectral zeta function of (M 2 , g) is defined by


ζ(s) = λ−s
j . (23)
j≥1

By Weyl’s asymptotic law,

λj ∼ j 2/n .

Consequently, (23) defines an analytic function for Re(s) > n/2. In fact, one
can meromorphically continue ζ in such a way that ζ becomes regular at
s = 0 (see [RS71]). Note that formally–that is, if we take the definition in
(23) literally–then

ζ  (0) = − log λj
j≥1
 
= − log λj (24)
j≥1

= − log det(−Δg ).

In view of this ansatz, it is natural to define the regularized determinant of

(−Δg ) as

det(−Δg ) = e−ζ (0) . (25)
PDEs in Conformal Geometry 23

6.1 The Case of Surfaces

Clearly, the determinant is not a local quantity. Therefore, it is rather remark-

able that Polaykov ([Pol81a], [Pol81b]) was able to compute a closed formula
for the ratio of the determinants of the laplacians of two conformally related
surfaces:

Theorem 6.1. Let (Σ, g), (Σ, ĝ = e2w g) be conformal surfaces. Then

det(−Δĝ ) 1  
log =− |∇w|2 + 2Kw dA, (26)
det(−Δg ) 12π Σ

where K is the Gauss curvature and dA the surface measure associated to

(Σ 2 , g).

Remarks.
1. The formula (26) naturally defines a (relative) action on the space of
conformal metrics. That is, once we fix a metric g, we have the functional

det(−Δĝ )
ĝ ∈ [g] → log .
det(−Δg )

However, since the determinant is not scale-invariant, we should consider the

normalized functional determinant
  
 

S[w] = |∇w|2 + 2Kw dA − K dA log e2w dA , (27)

Σ Σ Σ

so that
det(−Δĝ )
S[w] = −12π log + 2πχ(Σ) log Area(ĝ),
det(−Δg )

while

S[w + c] = S[w].

2. A first variation shows that w is a critical point of S if and only if w

satisfies

Δw + ce2w = K, (28)

where c is a constant. Now, if ĝ = e2w g, then the Gauss curvature K̂ of ĝ is

related to the Gauss curvature of g via
24 M.J. Gursky

Δw + K̂e2w = K, (29)

this is called the Gauss curvature equation. Comparing (28) and (29), we
see that w is a critical point of S if and only if (Σ, ĝ) has constant Gauss
curvature. In particular, a metric extremizes the functional determinant if
and only if it uniformizes; i.e., it is a conformal metric of constant Gauss
curvature.

3. In a series of papers ([Osg88b], [Osg88a]) Osgood-Phillips-Sarnak gave a

proof of the Uniformization Theorem by showing that each conformal class on
a surface admits a metric that extermizes the determinant. Like the Yamabe
problem and its fully nonlinear version discussed earlier in the article, the
main difficulty is the invariance of the determinant under the action of the
conformal group. And like the analysis of the Yamabe problem, the solution
involves the study of sharp functional inequalities. A very nice overview of
the study of the functional determinant and related material can be found
in [Cha].

6.2 Four Dimensions

The key property of the Laplacian that Polyakov exploited in his calculation
was its conformal covariance:

Δe2w g = e−2w Δg . (30)

More generally, we say that the differential operator A = Ag : C ∞ (M n ) →

C ∞ (M n ) is conformally covariant of bi-degree (a, b) if

Ae2w g ϕ = e−bw Ag (eaw ϕ). (31)

In fact, this definition makes perfect sense for operators acting on smooth
sections of bundles (spinors, forms, etc.) as well as on functions. Two examples
of note are

Example 1. The conformal laplacian of (M n , g), where n ≥ 3, is

4(n − 1)
L=− Δ + R, (32)
(n − 2)

where R is the scalar curvature. Then L is conformally covariant with

n−2 n+2
a= , b= .
2 2
PDEs in Conformal Geometry 25

Example 2. Let (M 4 , g) be a four-dimensional Riemannian manifold. The

Paneitz operator is
2
P = (Δ)2 + div{( Rg − 2Ric) ◦ d}. (33)
3
Then P is conformally covariant with

a = 4, b = 0.

An analogue of Polyakov’s formula for conformally covariant operators

defined on four-manifolds was computed by Branson-Ørsted in [Bra91]. To
explain the Branson-Ørsted formula we need to introduce three functionals
associated to a Riemannian 4-manifold (M 4 , g). Each functional is defined on
W 2,2 , the Sobolev space of functions with derivatives up to order two in L2 .
The first functional is zeroth order in w:
  

I[w] = 4 w|W |2 dV − |W |2 dV log e4w dV, (34)

where W is the Weyl curvature tensor and dV the volume form of g. If

w ∈ W 2,2 , The Moser-Trudinger inequality ([GT83]) implies that

ew ∈ Lp , all p ≥ 1.

Therefore, I : W 2,2 → R.
The second functional is analogous to the functional S defined in (27):
   

II[w] = wP w dV + 4 Qw dV − Q dV log e4w dV, (35)

where P is the Paneitz operator and Q is the Q-curvature:

1
Q= (−ΔR + R2 − 3|Ric|2 ). (36)
12
Here we see the parallel between the Laplace-Beltrami operator/Gauss cur-
vature of a surface and the Paneitz operator/Q-curvature of a 4-manifold.
The third functional is
 
III[w] = 12 (Δw + |∇w|2 )2 dV − 4 (wΔR + R|∇w|2 ) dV. (37)

The geometric meaning of this functional is apparent if we rewrite it in terms

of the scalar curvature Rĝ and volume form dV̂ of the conformal metric
ĝ = e2w g:
 
1  
III[w] = Rĝ2 dV̂ − R2 dV . (38)
3
26 M.J. Gursky

Therefore, III is the L2 -version of the Yamambe functional in (11).

With these definitions, we can give the Branson-Ørsted formula: Suppose
A = Ag is a conformally covariant differential operator satisfying certain
“naturality” conditions (see [Bra91] for details). Then there are numbers,
γi = γi (A), 1 ≤ i ≤ 3, such that

det Ae2w g
FA [w] = log = γ1 I[w] + γ2 II[w] + γ3 III[w]. (39)
det Ag

We remark that the Branson-Ørsted formula is normalized; i.e., FA [w + c] =

FA [w].

Example 1. For the conformal laplacian, Branson-Ørsted calculated

2
γ1 (L) = 1, γ2 (L) = −4, γ3 (L) = − . (40)
3

Example 2. Later, in [Bra96], Branson calculated the coefficients for the

Paneitz operator:
1 8
γ1 (L) = − , γ2 (L) = −14, γ3 (L) = . (41)
4 3

Neglecting lower order terms, the log det functional is of the form
 
det Ae2w g
log = γ1 (Δw)2 dV + γ3 [Δw + |∇w|2 ]2 dV
det Ag
 (42)
+ κA log e4w dV + · · · ,

where κA is given by
 
κA = −γ1 |W |2 dV − γ2 Q dV, (43)

a conformal invariant. In particular, when γ2 and γ3 have the same sign (as
they do for the conformal laplacian), the main issue from the variational point
of view is the interaction of the highest order terms with the exponential term.
However, when the signs of γ2 and γ3 differ, then the highest order terms are
a non-convex combination of II and III, and the variational structure can
be quite complicated.
PDEs in Conformal Geometry 27

6.3 The Euler Equation

As we observed above, critical points of the functional determinant on a sur-

face corresponds to metrics of constant Gauss curvature. In four dimensions
the geometric meaning of the Euler equation is less straightforward: Suppose
ĝ = e2w g is a critical point of FA ; then the curvature of ĝ satisfies

γ1 |Wĝ |2 + γ2 Qĝ + γ3 Δĝ Rĝ = −κA · V ol(ĝ)−1 . (44)

The geometric significance of this condition is, at first glance, difficult to

fathom. However, this equation in some sense includes all the significant
curvature conditions studied in four-dimensional conformal geometry, as can
be seen by considering special values of the γi ’s:

• Taking γ1 = γ2 = 0 and γ1 = 1, equation (44) becomes

Δĝ Rĝ = const. = 0, (45)

which is equivalent to the Yamabe equation

Rĝ = const.

• Taking γ1 = 0 and γ2 = −12γ3 , equation (44) becomes

σ2 (Aĝ ) = const., (46)

that is, a critical point is a solution of the σ2 -Yamabe problem.

• Taking γ1 = γ3 = 0 and γ2 = 1, then

Qĝ = const. (47)

Thus, critical points are solutions of the Q-curvature problem.

Geometric properties of critical metrics were used in [Gur98] to prove var-

ious vanishing theorems, and as a regularization of the σ2 -Yamambe problem
in [CGY02b].
Turning to analytic aspects of the Euler equation, it is clear from (42) that
it is fourth order in w. A precise formula is
1  
μe4w = ( γ2 + 6γ3 )Δ2 w + 6γ3 Δ|∇w|2 − 12γ3 ∇i (Δw + |∇w|2 )∇i w (48)
2
1 1
+ γ2 Rij ∇i ∇j w + (2γ3 − γ2 )RΔw + (2γ3 + γ2 )∇R, ∇w (49)
3 6
+ (γ1 |W |2 + γ2 Q − γ3 ΔR), (50)
28 M.J. Gursky

where Rij are the components of the Ricci curvature and

κA
μ = − . (51)
e4w

To simplify this expression we can divide both sides of (49) by 6γ3 , then
rewrite the lower order terms to arrive at
 
(1 + α)Δ2 w = f (x)e4w − Δ|∇w|2 + 2∇i (Δw + |∇w|2 )∇i w
(52)
+ aij ∇i ∇j w + bk ∇k w + c(x),

where
γ2
α= . (53)
12γ3
Although writing the equation in this form clearly reveals the divergence
structure, for some purposes it is better to expand the terms on the right
and write
(1 + α)Δ2 w = f (x)e4w − 2|∇2 w|2 + 2(Δw)2 + 2∇|∇w|2 , ∇w
(54)
+ 2Δw|∇w|2 + (lower order terms).

In particular, we see that the right-hand side does not involve any third
derivatives of the solution.
The regularity of extremal solutions of (49) was proved by Chang-Gursky-
Yang in [CGY99]), and for general solutions by Uhlenbeck-Viaclovsky in
[UV00]. Similar to the harmonic map equation in two dimensions, the main
difficulty is that the right-hand side of (54) is only in L1 when w ∈ W 2,2 ,
ruling out the possibility of using a naive bootstrap argument to prove
regularity.

6.4 Existence of Extremals

The most complete existence theory for extremals of the functional determi-
nant was done by Chang-Yang in [CY95]:

Theorem 6.2. Assume

γ2 , γ3 < 0, (55)

and

κA < 8π 2 (−γ2 ). (56)

Then supW 2,2 FA is attained by some some w ∈ W 2,2 .

PDEs in Conformal Geometry 29

Remarks.
1. Recall that
 
κA = −γ1 |W | dV − γ2
2
Q dV.

If γ1 > 0, then

κA ≤ −γ2 Q dV.

Therefore, assuming γ2 < 0, then (56) holds provided


Q dV < 8π 2 . (57)

By the definition of the Q-curvature,

1
Q = 2σ2 (A) − ΔR. (58)
12
Therefore,
 
Q dV = 2 σ2 (A) dV.

In particular, for manifolds of positive scalar curvature, by Theorem 5.7 it

follows that

Q dV ≤ 8π 2 , (59)

with equality if and only if (M 4 , g) is conformal to the round sphere. Thus,

combining the existence result of Chang-Yang with the sharp inequality of
Theorem 5.7, we conclude
Corollary 6.3. If (M 4 , g) has positive scalar curvature, then an extremal for
FL exists.

2. It is easy to construct examples of 4-manifold–necessarily with negative

scalar curvature–for which

Q dV >> 8π 2 .

Thus, the existence theory for the functional determinant is quite incomplete.
This shows another parallel with the σk -Yamabe problem (and contrast with
the classical Yamabe problem): the case of negative curvature is much more
difficult than the positive case.
30 M.J. Gursky

3. Branson-Chang-Yang proved that on the sphere S 4 , the functionals II and

III are minimized by the round metric and its images under the conformal
group [Bra]. In particular, the round metric is the unique extremal (up to
conformal transformation) of FL . Later, in [Gur97], Gursky showed that the
round metric is the unique critical point.

6.5 Sketch of the Proof

In the following we give a sketch of the proof of Theorem 6.2. By Corollary

6.3, this will give the existence of extremals for FA on any 4-manifold of
positive scalar curvature.
To begin, we write the functional as

FA [w] = γ1 I[w] + γ2 II[w] + γ3 III[w]

  
= γ1 (Δw)2 + γ2 (Δw + |∇w|2 )2 + κA log e4(w−w̄ + (l.o.t.).
(60)

Next, divide by γ2 , and denote F̃ = (1/γ2 )FA :

  
κA

F̃ [w] = (Δw)2 + β (Δw + |∇w|2 )2 − log e4(w−w̄) + (l.o.t.),

−γ2
(61)

where

β = γ3 /γ2 > 0. (62)

Since γ2 < 0, we are trying to prove the existence of minimizers of F̃ .

Let us first consider the easy case, when κA ≤ 0. Then
κA

− ≥ 0.
−γ2
Also, by Jensen’s inequality,

log e4(w−w̄) ≥ 0.

Therefore,
 
F̃ [w] ≥ (Δw)2 + β (Δw + |∇w|2 )2 + (l.o.t.) (63)

Now suppose {wk } is a minimizing sequence for F̃ ; from (63) we conclude

PDEs in Conformal Geometry 31

C≥ (Δwk )2 + (l.o.t.),

which implies, for example by the Poincare inequality, that {wk } is bounded
in W 2,2 . It follows that a subsequence converges weakly to a minimizer
w ∈ W 2,2 .
For the more difficult case when κA > 0, first observe that by hypothesis,
κA < 8π 2 (−γ2 ). Therefore,
κA
= 8π 2 (1 − ) (64)
−γ2

for some  > 0. The significance of the constant 8π 2 is apparent from the
following sharp Moser-Trudinger inequality due to Adams:

Proposition 6.4. (See [Ada]) If (M 4 , g) is a smooth, closed 4-manifold, then

there is a constant C1 = C1 (g) such that
 
1
log e4(w−w̄) ≤ (Δw)2 + C1 . (65)
8π 2

Using Adams’ inequality, we will show that the positive terms in F̃ domi-
nate the logarithmic term. To see why, we argue in the following way: by the
arithmetic-geometric mean,
1
2βxy ≥ −β(1 + δ)x2 − β( )y 2 ,
1+δ
for any real numbers x, y, as long as β, δ > 0. From this inequality it follows
that
  
δ
(Δw)2 + β (Δw + |∇w|2 )2 ≥ (1 − δβ)(Δw)2 + β( )|∇w|4 . (66)
1+δ

Therefore, by (64) and (66),

  
δ
F̃ [w] ≥ (1−δβ)(Δw)2 +β( ) |∇w|4 −8π 2 (1−) log e4(w−w̄) +(l.o.t.).
1+δ
By Adams’ inequality, the logarithmic term above can be estimated by
 
−8π 2 (1 − ) log e4(w−w̄) ≥ −(1 − ) (Δw)2 − C.

Substituting this above, we get

 
δ
F̃ [w] ≥ ( − δβ)(Δw)2 + β( ) |∇w|4 + (l.o.t.).
1+δ
By choosing δ > 0 small enough, we conclude
32 M.J. Gursky

 
F̃ [w] ≥ δ  (Δw)2 + |∇w|4 − C.

Arguing as we did in the previous case, it follows that F̃ is bounded below,

and a minimizing sequence converges (weakly) to a smooth extremal.

Remarks.

1. The lower order terms that we neglected in the proof can actually dominate
the expression when γ3 = 0, e.g., when studying the Q-curvature problem. In
particular, there are known examples of manifolds for which the functional
II in not bounded below.

2. When γ2 and γ3 have different signs–for example, when A = P , the Paneitz

operator–the situation is even worse. In fact, FP is never bounded from below.
However, manifolds of constant negative curvature are always local extremals
of FP .

References

[Ada] Adams, A sharp inequality of J. Moser for higher order derivatives.

[Ber03] Marcel Berger, A panoramic view of Riemannian geometry, Springer-Verlag,
Berlin, 2003.
[Bra] Sun-Yung A.; Yang Paul C. Branson, Thomas P.; Chang.
[Bra91] Bent Branson, Thomas P.; rsted, Explicit functional determinants in four
dimensions, Proc. Amer. Math. Soc. 113 (1991), 669–682.
[Bra96] Thomas P. Branson, An anomaly associated with 4-dimensional quantum
gravity, Comm. Math. Phys. 178 (1996), 301–309.
[CGY99] Sun-Yung A. Chang, Matthew J. Gursky, and Paul C. Yang, Regularity of a
fourth order nonlinear PDE with critical exponent, Amer. J. Math. 121 (1999),
no. 2, 215–257.
[CGY02a] Sun-Yung A. Chang, Matthew J. Gursky, and Paul Yang, An a priori estimate
for a fully nonlinear equation on four-manifolds, J. Anal. Math. 87 (2002),
151–186, Dedicated to the memory of Thomas H. Wolff.
[CGY02b] Sun-Yung A. Chang, Matthew J. Gursky, and Paul C. Yang, An equation
of Monge-Ampère type in conformal geometry, and four-manifolds of positive
Ricci curvature, Ann. of Math. (2) 155 (2002), no. 3, 709–787. MR 1 923 964
[Cha] Sun-Yung Alice Chang, The moser-trudinger inequality and applications to
some problems in conformal geometry, Nonlinear partial differential equations
in differential geometry.
[CY95] Sun-Yung A. Chang and Paul C. Yang, Extremal metrics of zeta function
determinants on 4-manifolds, Ann. of Math. (2) 142 (1995), no. 1, 171–212.
[Eva82] Lawrence C. Evans, Classical solutions of fully nonlinear, convex, second-order
elliptic equations, Comm. Pure Appl. Math. 35 (1982), no. 3, 333–363. MR
83g:35038
[GT83] David Gilbarg and Neil S. Trudinger, Elliptic partial differential equations of
second order, second ed., Springer-Verlag, Berlin, 1983.
PDEs in Conformal Geometry 33

[Gur97] Matthew J. Gursky, Uniqueness of the functional determinant, Comm. Math.

Phys. 189 (1997), no. 3, 655–665.
[Gur98] , The Weyl functional, de Rham cohomology, and Kähler-Einstein
metrics, Ann. of Math. (2) 148 (1998), 315–337.
[Gur99] , The principal eigenvalue of a conformally invariant differential oper-
ator, with an application to semilinear elliptic PDE, Comm. Math. Phys. 207
(1999), no. 1, 131–143.
[GV04] Matthew Gursky and Jeff Viaclovsky, Volume comparison and the σk -Yamabe
problem, Adv. Math. 187 (2004), 447–487.
[GVW03] Pengfei Guan, Jeff Viaclovsky, and Guofang Wang, Some properties of the
Schouten tensor and applications to conformal geometry, Trans. Amer. Math.
Soc. 355 (2003), no. 3, 925–933 (electronic).
[GW03] Pengfei Guan and Guofang Wang, Local estimates for a class of fully nonlinear
equations arising from conformal geometry, Int. Math. Res. Not. (2003), no. 26,
1413–1432. MR 1 976 045
[Kry83] N. V. Krylov, Boundedly inhomogeneous elliptic and parabolic equations in
a domain, Izv. Akad. Nauk SSSR Ser. Mat. 47 (1983), no. 1, 75–108. MR
85g:35046
[LP87] John M. Lee and Thomas H. Parker, The Yamabe problem, Bull. Amer. Math.
Soc. (N.S.) 17 (1987), no. 1, 37–91.
[Osg88a] R.; Sarnak-P. Osgood, B.; Phillips, Compact isospectral sets of surfaces,
J. Funct. Anal. 80 (1988), 212–234.
[Osg88b] , Extremals of determinants of laplacians, J. Funct. Anal. 80 (1988),
148–211.
[Pol81a] A. M. Polyakov, Quantum geometry of bosonic strings, Phys. Lett. B 103
(1981), 207–210.
[Pol81b] , Quantum geometry of fermionic strings, Phys. Lett. B 103 (1981),
211–213.
[RS71] D. B. Ray and I. M. Singer, R-torsion and the Laplacian on riemannian
manifolds, Adv. in Math. 7 (1971), 145–210.
[STW05] Weimin Sheng, Neil S. Trudinger, and Xu-Jia Wang, The Yamabe problem for
higher order curvatures, preprint, 2005.
[UV00] Karen K. Uhlenbeck and Jeff A. Viaclovsky, Regularity of weak solutions to
critical exponent variational equations, Math. Res. Lett. 7 (2000), no. 5-6,
651–656.
[Via00] Jeff A. Viaclovsky, Conformal geometry, contact geometry, and the calculus of
variations, Duke Math. J. 101 (2000), no. 2, 283–316.
[Via06] , Conformal geometry and fully nonlinear equations, to appear in World
Scientific Memorial Volume for S.S. Chern., 2006.
Heat Kernels in Sub-Riemannian
Settings

Ermanno Lanconelli

1 Lecture Topics

In this lectures we present a series of results concerning a class of diffusion

second order PDE’s of heat-type. The results we show have been obtained in
collaboration with M.Bramanti, L.Brandolini and F. Uguzzoni (see [9], [10],
[24]). The exended version of the main results presented in these notes is
contained in [10].
The operators we are dealing with can be written in the general form


N 
N
qi,j (z)∂x2i ,xj + qj (z)∂xj − ∂t .
i,j=1 j=1

The coefficients qi,j = qj,i , qj are of class C ∞ in the strip

S := {z = (t, x) : x ∈ RN , T1 < t < T2 }

= ]T1 , T2 [×RN

where −∞ ≤ T1 < T2 ≤ ∞. We assume the characteristic form


N
qH (z, ξ) = qi,j (z)ξi ξj
i,j=1

is non-negative definite, and not identically zero, at any point z ∈ S. More-

over, the operator H is supposed to be hypoelliptic in S. This means that

E. Lanconelli
Dipartimento di Matematica Universita’ degli Studi di Bologna, P.zza di Porta S. Donato,
5 40127 - Bologna - Italy
e-mail: [email protected]

S.-Y.A. Chang et al. (eds.), Geometric Analysis and PDEs, 35

Lecture Notes in Mathematics 1977, DOI: 10.1007/978-3-642-01674-5 2,

c Springer-Verlag Berlin Heidelberg 2009
36 E. Lanconelli

every distributional solution to Hu = f in an open set Ω ⊂ S is of class C ∞

whenever f is C ∞ .
Together with the previous qualitative properties, we assume that H has
a fundamental solution Γ satisfying the Gaussian estimates
1
Gb (z, ζ) ≤ Γ (z, ζ) ≤ Λ Ga0 (z, ζ),
Λ 0
where, Ga (z, ζ) = Ga (t, x; τ, ξ) = 0 if t ≤ τ , and
 
1 d2 (x, ξ)
Ga (t, x; τ, ξ) = √ exp −a
|B(x, t − τ )| t

if t > τ . Λ, a0 , b0 are positive constants.

Hereafter d is a metric in RN and |B(x, r)| denotes the Lebesgue measure
of the d-ball B(x, r). We assume that the metric space (RN , d) is of sub-
Riemannian type, i.e. that it satisfies the following conditions:

• the d-topology is the Euclidean topology

• diamd (RN ) = ∞
• (RN , d) is a doubling space w.r to the Lebesgue measure, i.e. 0 <
|B(0, 2r)| ≤ cd |B(0, 2r)|, for every x ∈ RN , and r > 0
• (RN , d) has the segment property, i.e. for every x, y ∈ RN there exists
γ : [0, 1] → RN , continuous and such that d(x, y) = d(x, γ(t)) + d(γ(t), y)
for every t ∈ [0, 1]

We shall denote by |H| the constant

|H| := Λ + a0 + b0 + cd

Under the previous assumptions, we show the following scale invariant

Harnack inequality which extends to our sub-Riemannian setting the classical
Hadamard-Pini parabolic Harnack inequality
Harnack inequality. If Hu = 0 and u ≥ 0 in an open set containing
[t0 − R2 , t0 ] × B(x0 , R), where (t0 , x0 ) =: z0 ∈ S, then

max u ≤ M u(z0 ).
CR (z0 )

Here CR (z0 ) := [τ0 − γR2 , τ0 − γ2 R2 ] × Bd (ξ0 , γR)

The constants M > 0 and 0 < γ < 1 are independent of R and z0 . They
depend on the operator H only through the constant |H|.
Our basic example of diffusion operator , satisfying the assumption stated
before is the following ones


q 
q
H := L − ∂t := aij (z)Xi Xj + ak (z)Xk − ∂t
i,j=1 k=1
Heat Kernels in Sub-Riemannian Settings 37

where:
• X1 , X2 , . . . , Xq are smooth vector fields in the open set Ω ⊂ RN satisfying
the Hörmander condition

rank Lie{Xi , i = 1, 2, ..., q} = n at any point of Ω.

Then H is hypoelliptic in Ω [17].

• A(z) = (aij (z))qi,j=1 is a symmetric matrix such that

1 2  q
|ξ| ≤ aij (z)ξi ξj ≤ λ|ξ|2
λ i,j=1

for every z = (t, x) ∈ S, ξ = (ξ1 , . . . , ξq ) ∈ Rq

A natural distance for the operator H is the Carnot-Carathéodory dis-
tance d generated by the vector fields X1 , X2 , . . . , Xq . We would like to stress
that d is well defined since the system X = {X1 , X2 , . . . , Xq } satisfies the
Hörmander rank condition.
In [10]we proved that X can be extended to a system of Hörmander
vector fields, defined all over RN , in such a way that the associated Carnot-
Carathéodory distance satisfies all the assumptions stated before.
In [10] the operator H is also extended to the whole RN +1 in such a way
that outside of a compact set in the spatial variable, it becomes the classical
Heat operator.
Still denoting by H the extended operator, under the qualitative assump-
tion ai,j , aj C ∞ , we proved that H has a global fundamental solution Γ such
that
1
Gb (z, ζ) ≤ Γ (z, ζ) ≤ Λ Ga0 (z, ζ)
Λ 0
and, for every α = (αi , αj .αk ) with |α| ≤ 2,
|α|
|Dα Γ (z, ζ)| ≤ Λ|t − τ |− 2 Ga0 (z, ζ).
α
Here we have used the notation Dα := Xiαi Xj j ∂tαk and |α| := αi + αj + 2αk .
In these inequalities, Λ, a0 , b0 are positive structural constants: they only
depend on
• the doubling constant of d
• the constant λ in

1 2  q
|ξ| ≤ aij (z)ξi ξj ≤ λ|ξ|2
λ i,j=1

• the d-Hölder norms of the coefficients ai,j and aj .

38 E. Lanconelli

As a consequence: all our results extend to the operators


q 
q
H := L − ∂t := aij (z)Xi Xj + ak (z)Xk − ∂t
i,j=1 k=1

with d-Hölder continuous coefficients ai,j and aj .

We construct the heat kernel (the fundamental solution) by an adapta-
tion to our sub-riemannian setting of the classical Levi parametrix method,
as in [4]. For this adaptation we used a large amount of ideas, tech-
niques and results due to Rotschild&Stein [34], Jerison&Sanchez-Calle [19],
Fefferman&Sanchez-Calle [14] and to Kusuoka&Stroock [20], [21], [22].

2 A Motivation: Levi Curvatures and Motion by Levi

Curvatures

There are several motivations for studying heat-type operators in sub-

Riemannian settings. We want to show just one of them, arising in Complex
Geometry in studying a notion af curvature related to the Levi form.
Let M be a real hypersurface of Cn+1 which is the boundary of a domain

D := {z ∈ Cn+1 : f (z) < 0},

where f is real function of class C 2 on Cn+1 . The complex tangent space to

M = ∂D at a point p is given by

TpC (∂D) = {h ∈ Cn+1 : h, ∂¯p f = 0}.

We explicitly remark remark that TpC (∂D) is a complex vector space of dimen-
sion n, hence it can be idetified with a real vector space of real dimension 2n.
Therefore, in passing from M to TpC (∂D) we loose a real dimension
The Levi form of f at p is given by


n+1
Lp (f, ζ) = HfT (p)ζ, ζ := fj,k̄ (p)ζj ζ̄k , ζ ∈ TpC (bD).
j,k=1

2
∂f
Here fj = ∂z j
, fk̄ = ∂∂f ∂ f
z̄k , fj,k̄ := ∂zj ∂ z̄k . Let us now fix an orthonormal
basis B = {u1 , . . . , un } of TpC (bD) and consider the matrix
 
1
Lp (f, B) := H (p)uj , uk
T
.
|∂p f | f k,j=1,...,n

This is a Hermitian n × n matrix which we call Normalized Levi Matrix

of f at p. Its eigenvalues λ1 (p), . . . , λn (p) are real and independent of B and
the defining function f of D. So that they only depend on the domain D.
Heat Kernels in Sub-Riemannian Settings 39

Just proceeding as in the real case, one can define the m−th Levi curvature
of ∂D at p, 1 ≤ m ≤ n, as

σ (m) (λ1 (p), . . . , λn (p))

Kpm (∂D) = n
,
m

where σ (m) denotes the m-th elementary symmetric function. More generally,
given a generalized symmetric function s, in the sense of Caffarelli-Nirenberg-
Spruck [11], one can define the s-Levi curvature of M at p, as follows:

Sp (M ) = s (λ1 (p) , ..., λn (p)) .

When M is the graph of a function u and one imposes that its s-Levi curvature
is equal to a given function, one obtains a second order fully nonlinear partial
differential equation, which can be seen as the pseudoconvex counterpart of
the usual fully nonlinear elliptic equations of Hessian type, as studied e.g. in
[11]. In linearized form, the equations of this new class can be written as (see
[29, equation (34) p. 324])


2n

Lu ≡ aij Du, D2 u Xi Xj u = K (x, u, Du) inR2n+1 (1)

i,j=1

where:
the Xj ’s are first order differential operators, with coefficients depending
on the gradient of u, which form a real basis for the complex tangent space
to the graph of u;
the matrix {aij } depends on the function s;
K is a prescribed function.
It has to be noticed that L only involves 2n derivatives, while it lives in a
space of dimension 2n + 1. Then, L is never elliptic, on any reasonable class
of functions. However, the operator L, when restricted to the set of strictly
s-pseudoconvexk functions, becomes “elliptic” along the 2n linearly indepen-
dent directions given by the Xi ’s, while the missing one can be recovered by
a commutation. Precisely,

dim (span {Xj , [Xi , Xj ] , i, j = 1, ..., 2n}) = 2n + 1

at any point (see [29, equation (36) p. 324]). This is a Hörmander-type rank
condition of step 2.
The parabolic counterpart of (1), i.e. the equation

∂t u (t, x) = Lu (t, x) for t ∈ R, x ∈ R2n+1 (2)

arises in studying the evolution by s-Levi curvature of a real hypersurface of

Cn+1 (see [18], [28]).
40 E. Lanconelli

Before closing this section we would like to quote some remarkable isoperi-
metric inequalities involving the Levi curvatures, recently proved by Martino
and Montanari in [27]. These inequalities well enlighten the geometric content
of the Levi curvatures.
Isoperimetric inequalities Let D be a bounded smooth domain in
C n+1 . Let 1 ≤ m ≤ n and assume Kzm (∂D) ≥ 0 at any point z ∈ ∂D . Then

−1
(Kz(m) (∂D)) m dσ(z) ≥ (2n + 2)meas(D)
∂D

If Kzm (∂D) = constant w.r. to z, in the previous inequality the equality

holds if and only if D is a ball.

3 Heat Kernels and Gaussian Bounds:

Past History

Gaussian estimates for the fundamental solution of second order partial dif-
ferential operators of parabolic type, or, somehow more generally, for the
density function of heat diffusion semigroups, have a long history, starting
with Aronson’s work [1]. The relevance of two-sided Gaussian estimates to
get scaling invariant Harnack inequalities for positive solutions was firstly
pointed out by Nash in the Appendix of his celebrated paper [31]. However,
a complete implementation of the method outlined by Nash was given much
later by Fabes and Stroock in [13], also inspired by some ideas of Krylov
and Safonov. Since then, the full strength of Gaussian estimates has been
enlightened by several authors, showing their deep relationship not only with
the scaling invariant Harnack inequality, but also with the ultracontractivity
property of heat diffusion semigroups, with inequalities of Nash, Sobolev or
Poincaré type, and with the doubling property of the measure of “intrinsic”
balls. We directly refer to the recent monograph by Saloff-Coste [35] for a
beautiful exposition of this circle of ideas, and for an exhaustive list of refer-
ences on these subjects. Here we explicitly recall just the results in literature
strictly close to the core of our lectures.
For heat-type operators


q
H = ∂t − Xi2 (3)
i=1

with the Xi ’s left invariant homogeneous vector fields on a sratified Lie group
in Rn , Gaussian bounds have been proved by Varopoulos ([40], [41], see also
[42]):
1 −cy−1 ◦x2 /t c −y−1 ◦x2 /ct
e ≤ h (t, x, y) ≤ e (4)
ctQ/2 tQ/2
Heat Kernels in Sub-Riemannian Settings 41

for any x, y ∈ Rn , t > 0, where Q is the homogeneous dimension of the

group, and · any homogeneous norm of the group. Two-sided Gaussian
estimates and a scaling invariant Harnack inequality have been proved by
Saloff-Coste and Stroock for the operator


q
H = ∂t − Xi (aij Xj )
i,j=1

where {aij } is a uniformly positive matrix with measurable entries, and the
vector fields Xi are left invariant with respect to a connected unimodular Lie
group with polynomial growth [36].
In absence of a group structure, Gaussian bounds for operators (3) have
been proved, on a compact manifold and for finite time, by Jerison and
Sanchez-Calle [19], with an analytic approach (see also the previous par-
tial result in [37]), and, on the whole Rn+1 , by Kusuoka-Stroock, [21], [22],
by using Malliavin stochastic calculus.
Unlike the study of “sum of squares” Hörmander’s operators, the investi-
gation of non-divergence operators of Hörmander type has a relatively recent
history. Stationary operators like


q
L= aij (x) Xi Xj (5)
i,j=1

with X1 , ..., Xq system of Hörmander’s vector fields have been studied by Xu

[43], Bramanti, Brandolini [7], [8], Capogna, Han [12]. Evolution operators
the previuos one have been considered by Bonfiglioli, Lanconelli, Uguzzoni
[3], [4], [5]. In [7] also more general operators of kind


q
L= aij (x) Xi Xj + a0 (x) X0 (6)
i,j=1

with X0 , X1 , ..., Xq system of Hörmander’s vector fields have been studied.

In these papers, the matrix {aij } is assumed symmetric and uniformly
elliptic, and the entries aij typically belong to some function space defined in
terms of the vector fields Xi and the metric they induce. In particular, these
operators do not have smooth coefficients, so they are no longer hypoellip-
tic. Therefore the mere existence of a fundamental solution is troublesome.
For the operators (6) (without lower order terms) with Xi left invariant
homogeneous Hörmander’s vector fields on a stratified Lie group and under
assumptions (H1), (H2), (H3), it has been proved by Bonfiglioli, Lanconelli,
Uguzzoni in [3], [4], [5] that the fundamental solution h exists and satisfies
Gaussian bounds of the kind (4). Previous results about Harnack inequal-
ity for general Hörmander’s operators date back to Bony’s seminal paper
[6], where a first qualitative version of this result is proved. A first scaling
42 E. Lanconelli

invariant Harnack inequality for heat-type Hörmander’s operators was proved

later by Kusuoka-Stroock [21].

4 Carnot-Carathèodory metric

Let X1 , X2 , . . . , Xq be a system of real smooth vector fields which are defined

in some bounded connecetd domain Ω ⊆ Rn and satisfying Hörmander’s
condition of some step s in Ω. Explicitly, this means that:

n
Xi = bik (x) ∂xk
k=1

with bik ∈ C ∞ (Ω) , and the vector space spanned at every point of Ω by: the
fields Xi ; their commutators [Xi , Xj ] = Xi Xj − Xj Xi ; the commutators of
the Xk ’s with the commutators [Xi , Xj ]; . . . and so on, up to some step s, is
the whole Rn .
We say that an absolutely continuous curve γ : [0, T ] → Ω is a sub-unit
curve with respect to the system X1 , X2 , . . . , Xq if


q
γ  (t) = λj (t)Xj (γ(t))
j=1

for a.e. t ∈ [0, T ], with qj=1 λj (t)2 ≤ 1 a.e. In the following, this number T
will be denoted by l (γ).
For any x, y ∈ Ω, we define

dΩ (x, y) = inf{l (γ) | γ is X -subunit, γ(0) = x, γ(l (γ)) = y}.

It is well known (Carathèodory-Chow’s Theorem) that, if the vector

fields satisfy Hörmander’s condition, the above set in nonempty, so that
dΩ (x, y) is finite for every pair of points. Moreover, dΩ is a distance in Ω,
called the Carnot-Carathéodory distance (CC-distance) induced by the vector
fields Xi ’s.
A known result by Fefferman-Phong [15] states that

c−1 |x − y| ≤ dΩ (x, y) ≤ c |x − y|
1/s
(7)

for every x, y ∈ K ⊂ Ω, where s is the step of Hörmander condition. In

particular, this means that dΩ induces the usual topology of Rn . Moreover,
Sanchez-Calle [37] and Nagel-Stein-Weinger [30] proved that the CC-distance
is locally doubling with respect to the Lebesgue measure, i.e., denoting by
B (x, r) the dΩ -ball of center x and radius r:
Heat Kernels in Sub-Riemannian Settings 43

|B (x, 2r)| ≤ c |B (x, r)| (8)

at least for x ranging in a compact set and r bounded by some r0 .

As we have already recalled , in [10] is proved that the given system of
vector fields can be exended to the whole Rn to a new system of Hörmander
vector fields whose related Carnot-Carathèodory distance satisfies a global
doubling property, and coincide with the original one in a fixed compact
subset of Ω. Moreover, in a neighborhood of infinity, the new system is the
Euclidean one.

5 Heat-Type Operators with Constant Coefficients

Let us consider the heat-type operator in Rn+1


q
HA = ∂t − LA = ∂t − aij Xi Xj (9)
i,j=1

where:

(H1)X1 , X2 , . . . , Xq is a system of Hörmander vector fields in Rn , of step ≤

s, coinciding with the Euclidean one in a neighborhood of infinity
q
(H2)A = {aij }i,j=1 is a real symmetric positive definite matrix with
constant entries, and λ > 0 a constant such that:


q
λ−1 |ξ| ≤
2 2
aij ξi ξj ≤ λ |ξ|
i,j=1

for every ξ ∈ Rq . When condition (H2) is fulfilled, we will briefly say that

A ∈ Eλ .

Then, the following thoerem holds.

Theorem 5.1. For any T > 0 there exists c > 0 such that, for any t ∈
(0, T ) , x, y ∈ Rn the following bounds hold:
1. Upper and lower bounds on hA :
1 2 c 2
 √
 e−cd(x,y) /t ≤ hA (t, x, y) ≤  √
 e−d(x,y) /ct (10)

c B x, t   B x, t 

2. Upper bounds on the derivatives of hA of arbitrary order:

 I J i  c
X X ∂ hA (t, x, y) ≤  √
 e
−d(x,y)2 /ct
(11)
x y t |I|+|J|
ti+ 2 B x, t 
44 E. Lanconelli

3. Estimate on the difference of the fundamental solutions of two operators

(and their derivatives):
 I J i  c A − B
Xx Xy ∂t hA (t, x, y)−XxI XyJ ∂ti hB (t, x, y) ≤  √
 e
−d(x,y)2 /ct
B x, t 
|I|+|J|
ti+ 2
(12)
(here I, J are arbitrary multiindices, A, B ∈ Eλ ). The constants depend on
the matrix A only through the number λ; in (11), (12), the constant also
depends on the multiindices. The same estimates hold for hA (t, y, x).

The proof of this theorem is contained in [10], Part I. We followed an

approach which is basically inspired to the work by Jerison and Sanchez-
Calle [19], integrated with several devices to overcome the new difficulties.
The main of them are the following: first, we had to take into account the
dependence on the matrix A, getting estimates depending on A only through
the number λ; second, the globality of the estimates ; third, the estimates of
the difference of the fundamental solutions of two different operators.
The strategy we followed in order to get the uniform Gaussian bounds
for hA was as follows. First, we proved the upper bound for t ∈ (0, 1) and
ε < d (x, y) < R. In this range, the bound is equivalent to:

hA (t, x, y) ≤ ce−1/ct (13)

and is proved by means of estimates of Gevray type. This means that the
exponential decay of hA for vanishing t is deduced by a control on the supre-
mum of the time derivative of any order of a solution to HA u = 0. This
technique makes the constant c in (13) depend on:
 T 
sup dτ hA (τ, x, y) dx.
y∈Rn 0 ε<d(x,y)<R

So the next problem was to prove a uniform upper bound on this quan-
tity (i.e., depending on A only through λ). This is accomplished exploiting
suitable estimates on fractional and singular integrals on spaces of homo-
geneous type, and uniform subelliptic estimates. Next, we had to prove
the upper bound in (10) for t ∈ (0, 1) and d (x, y) < ε. For this we used
Rothschild-Stein’s technique of “lifting and approximation”. This allowed, by
a rather involved procedure, to deduce the desired uniform bound from the
analogous result proved, in the context of homogeneous groups, by Bonfiglioli-
Lanconelli-Uguzzoni [3], and to complete the proof of the upper bound in
(10) for t ∈ (0, 1) and d (x, y) < R. To prove the same upper bound for any
x, y ∈ Rn and t ∈ (0, T ), we used a comparison argument, exploiting our ad
hoc extension of the vector fields.
Heat Kernels in Sub-Riemannian Settings 45

Then, we proved the lower bound in (10), again by using the analogous
uniform lower bound which holds in the case of homogeneous vector fields.
Subsequently, we proved the Gaussian bound (11) on the derivatives of
hA . Like in [19], this bound is deduced by the upper bound on hA applying a
powerful result by Fefferman and Sanchez-Calle [14], which assures the exis-
tence of a local change of coordinates which is a good substitute of dilations
(which in our context do not generally exist).
Finally, we proved the estimate (12) on the difference of two fundamen-
tal solutions hA − hB . It relies on a suitable use of basic properties of the
fundamental solution and on the uniform bound (11) on the derivatives of hA .

6 Heat-Type Operators with Variable Coefficients.

The Levi Parametrix Method

The previuos results on the constant coefficients heat-type operators are

crucial to deal with operators of this kind

m 
m
H = ∂t − ai,j (t, x) Xi Xj − ak (t, x) Xk − a0 (t, x) . (14)
i,j=1 k=1

To exploit the results of the previous section, we make the same assumptions
on the vector fields and the structure of the matrix of the coefficients in the
principal part. Moreover, the coefficients aij , ak , a0 are supposed smooth and
m
globally defined; the matrix {aij }i,j=1 will be assumed symmetric and uni-
formly positive definite. Under these assumptions one can prove the existence
of a (global) fundamental solution for H, satisfying natural basic properties
and sharp Gaussian bounds. To state the results we need to introduce some
more definitions, notation and assumptions.

Assumptions on the Vector Fields

We assume that:
X = (X1 , X2 , ..., Xm ) (m = n + q) is a fixed system of Hörmander’s vector
fields defined in the whole Rn , and such that

X = (0, 0, ..., 0, ∂x1 , ∂x2 , ..., ∂xn ) in Rn \ Ω0 (15)

where Ω0 is a fixed bounded domain.

46 E. Lanconelli

Function Spaces

We start with the following

Definition 6.1. The intrinsic-derivative along the vector field Xj of a func-

tion v(x) at a point x0 ∈ Rn , is defined to be

d 
Xj v(x0 ) = v(γ(σ))
dσ σ=0

(if such derivative exists), where γ is the solution to

γ̇(σ) = Xj (γ(σ)), γ(0) = x0 .

We can now introduce the natural function spaces for the operator H.
We will denote by d the Carnot-Carathéodory distance induced by the
m
system {Xi }i=1 in the whole Rn and by B(x, r) the balls in the metric d.
Moreover, let dP be the corresponding “parabolic-CC-distance”:

1
dP ((t, x), (s, y)) = d4 (x, y) + (t − s)2 4

Definition 6.2. For any α ∈ (0, 1] and domain U ⊆ Rn+1 , let:

 
|u (t, x) − u (s, y)|
|u|C α (U ) = sup α : (t, x) , (s, y) ∈ U, (t, x) 
= (s, y)
dP ((t, x) , (s, y))

uC α (U ) = |u|C α (U ) + uL∞ (U )
 
C α (U ) = u : U → R : uC α (U ) < ∞ .

Also, for any positive integer k, and domain U ⊆ Rn+1 , let

 
C k,α (U ) = u : U → R : uC k,α (U ) < ∞

with   h I 
uC k,α (U ) = ∂t X u α
C (U )
|I|+2hk

where, for any multiindex I = (i1 , i2 , ..., is ) , with 1  ij  q, we say that

|I| = s and
X I u = Xi1 Xi2 ...Xis u.

Assumptions on the Coefficients

m
We assume that the matrix {aij }i,j=1 is block-diagonal, with last block the
n × n identity. Moreover , the functions aij = aji , ak , a0 , are smooth
Heat Kernels in Sub-Riemannian Settings 47

and defined on Rn+1 and satisfy, for some α ∈ (0, 1] and for some positive
constants λ, K,
q
λ−1 |w| ≤ i,j=1 aij (t, x) wi wj ≤ λ |w| ∀w ∈ Rq , (t, x) ∈ Rn+1 (16)
2 2

aij C α (Rn+1 ) + ak C α (Rn+1 ) + a0 C α (Rn+1 ) ≤ K. (17)

To state the existence result for the fundamental solution, it is also

convenient to denote the Gaussian kernel G1 (x, t; ξ, 0) as follows:
 
√ −1 d(x, ξ)2
E(x, ξ, t) = |B(x, t)| exp − , x, ξ ∈ Rn , t > 0.
t

Main Results

Theorem 6.3 (Fundamental solution for H). Let H be as in (14). Under

the above assumptions, there exists a function

h : Rn+1 × Rn+1 → R

such that:

i) h is continuous away from the diagonal of Rn+1 × Rn+1 ;

ii) h (z, ζ) is nonnegative, and vanishes for t ≤ τ ;
iii) for every fixed ζ ∈ Rn+1 , we have
2,α
h(·; ζ) ∈ Cloc (Rn+1 \ {ζ}), H (h(·; ζ)) = 0 in Rn+1 \ {ζ};

iv) the following estimates hold for every T > 0, z = (t, x) , ζ = (τ, ξ) ∈
Rn+1 , 0 < t − τ ≤ T :

c(T )−1 E(x, ξ, c−1 (t − τ )) ≤ h(z; ζ) ≤ c(T ) E(x, ξ, c(t − τ )),

|Xj (h(·; ζ)) (z)| ≤ c(T ) (t − τ )−1/2 E(x, ξ, c(t − τ )
|Xi Xj (h(·; ζ)) (z)| + |∂t (h(·; ζ)) (z)| ≤ c(T ) (t − τ )−1 E(x, ξ, c(t − τ )).

v) for any f ∈ C α Rn+1 g ∈ C (Rn ) , both satisfying suitable growth

condition at infinity , T ∈ R, the function
 
u (t, x) = h(t, x; T, ξ) g(ξ) dξ + h(t, x; τ, ξ) f (τ, ξ)dτ dξ
Rn [T,t]×Rn

2,α
is a Cloc solution to the following Cauchy problem

Hu = f in (T, ∞) × Rn , u(T, ·) = g in Rn
48 E. Lanconelli

vi) the following reproduction formula holds


h(t, x; τ, ξ) = h(t, x; s, y) h(s, y; τ, ξ)dy,
Rn

for t > s > τ and x, ξ ∈ Rn .

The Levi Paramerix Method

To construct the fundamental solution for the operator


m 
m
H = ∂t − ai,j (t, x) Xi Xj − ak (t, x) Xk − a0 (t, x) . (18)
i,j=1 k=1

we used the classical Levi paramtrix metod, adapted to our sub-Riemann set-
ting. The “Levi method” is a classical technique that allows to construct the
fundamental solution of a variable coefficient differential operator, starting
from the fundamental solution of the corresponding operator with constant
coefficients. This method was originally developed by E. E. Levi at the begin-
ning of 20th century to study uniformly elliptic equations of order 2n (see
[25], [26]) and later extended to uniformly parabolic operators (see e.g. [16]).
In the context of hypoelliptic ultraparabolic operators of Kolmogorov-
Fokker-Planck type, Polidoro in [33] managed to adapt this method, thanks
to the knowledge of an explicit expression for the fundamental solution of
the “frozen” operator, which had been constructed in [23]. For operators of
type (18), structured on homogeneous and invariant vector fields on Carnot
groups, no explicit fundamental solution is available in general. Neverthe-
less, Bonfiglioli, Lanconelli, Uguzzoni in [4] showed how to adapt the same
method, exploiting suitable sharp uniform Gaussian bounds on the funda-
mental solutions of the frozen operators. Here we follow the same line, first
giving an outline of the Levi method.
Let us consider the fundamental solution hζ0 (z, ζ) of the “frozen ” operator


m
Hζ0 := ∂t − Lζ0 := ∂t − ai,j (ζ0 )Xi Xj . (19)
i,j=1

The function z → hζ (z, ζ) is called a parametrix for H. The idea of the

Levi method is to look for a fundamental solution h (z, ζ) for H, which could
be written in the form:
 t
h (z, ζ) = hζ (z, ζ) + hη (z, η)Φ (η, ζ) dη (20)
τ Rn

for a suitable, unknown kernel Φ (z, ζ). This seems reasonable because we
expect h to be a small perturbation of theparametrix, as the integral
Heat Kernels in Sub-Riemannian Settings 49

equation (20) expresses. The following formal computation suggests how to

guess the right form of Φ (z, ζ) . If we set

Z1 (z; ζ) = −H (z → hζ (z, ζ)) (z), z = ζ ∈ Rn+1

and apply the operator H to both sides of (20) for z = ζ, we find:

 t
0 = −Z1 (z; ζ) + Φ (z, ζ) − Z1 (z, η)Φ (η, ζ) dη.
τ Rn

This means that Φ solves the integral equation

 t
Z1 (z, ζ) = Φ(z; ζ) − Z1 (z, η)Φ (η, ζ) dη
τ Rn

which, defining the integral operator T with kernel Z1 , can be rewritten as

Z1 = (I − T ) Φ

whence, formally,
∞
 ∞

Φ= T k Z1 ≡ Zk .
k=0 k=0

To make the above idea rigorous, one has to reverse the order of the
previous steps, first studying the properties of the function
∞ Z1 , then of
the functionns Zk , hence the convergence of series Φ = k=0 Zk , then the
properties of   t
J (z, ζ) = hη (z, η)Φ (η, ζ) dη
τ Rn
and finally the ones of

h (z, ζ) = hζ (z, ζ) + J (z, ζ) .

This can be accomplished by using all the results related to the constant
coefficients heat-type operators.

7 Hadamard-Pini Invariant Harnack Inequality

Here we finally come to the main goal of our lectures: the proof of the invariant
parabolic Harnack inequality, of Hadamard-Pini type, for the non negative
solution to the equation Hu = 0, where


N 
N
H= qi,j (z)∂x2i ,xj + qj (z)∂xj − ∂t
i,j=1 j=1

is a diffusion operator satisfying all the assumptions fixed in the first section.
50 E. Lanconelli

The proof is inspired to the paper by Fabes-Stroock [13], who, in turn,

exploited the original ideas by Krylov-Safanov about parabolic operators in
nondivergence form. In Fabes-Stroock’s paper, Harnack inequality is derived
by a fairly short but clever combination of estimates based only on the
Gaussian bounds (from above and below) on the Green function for a
cylinder. The radius of the cylinder incorporates the essential geometrical
information, giving dilation invariance to the Harnack estimate.
About at the same time of Fabes-Stroock paper, the same deep ideas were
applied
 by Kusuoka-Stroock [21] in the context of Hörmander’s operators
∂t − qi=1 Xi2 . Much more recently, this general strategy has been adapted
by Bonfiglioli, Uguzzoni [5] to study nonvariational operators structured on
Hörmander’s vector fields in Carnot groups.
Here we will follow the same line. The striking feature of this proof is
the “axiomatic” nature of its core: it depends only on the suitable Gaussian
estimates for the Green function, a maximum principle for H, the fact that
constants are solutions to Hu = 0 (absence of the zero order term), and some
simple geometric properties of the distance D: the doubling property of the
Lebesgue measure of the d-balls and the segment property of d. However,
a first problem arises since, for our operator H, the existence of the Green
function is not granted: the cylinders based on the metric balls could be bad
domain for the Dirichlet problem. Nevertheless, Lanconelli and Uguzzoni have
recently proved in [24] that given two metric balls B (ξ, δR), B (ξ, R), (with
δ ∈ (0, 1)), there always exists a domain A (ξ, R), regular for the (station-
ary) Dirichlet problem, and such that B (ξ, δR) ⊆ A (ξ, R) ⊆ B (ξ, R) (see
Lemma 7.2). The Green function for H on R × A (ξ, R) must be thought as
the natural substitute of the Green function for the cylinder R × B (ξ, R).

Green Function on Regular Domains

We start with the following definitions:

Definition 7.1. We shall say that a bounded cylinder

D = (T1 , T2 ) × Ω ⊆ Rn+1

is H-regular, if for every continuous function ϕ on the parabolic boundary

∂p D = ([T1 , T2 ] × ∂Ω) ∪ ({T1 } × Ω),

there exists a unique solution uϕ to

u ∈ C ∞ (D) ∩ C(D ∪ ∂p D), Hu = 0 in D, u = ϕ in ∂p D. (21)

We shall also say that a bounded open set Ω ⊆ Rn is H-regular if, for any
T1 < T2 , the cylinder (T1 , T2 ) × Ω is H-regular.
Heat Kernels in Sub-Riemannian Settings 51

By the Picone Maximum Principle [32] if D is an H-regular domain, for

any fixed z ∈ D, the linear functional

T : C(∂p D) → R,
T : ϕ −→ uϕ (z) with uϕ as in (21)

is continuous. Therefore there exists a measure μDz (supported on ∂p D) so

that 
uϕ (z) = ϕ(ζ)dμD
z (ζ).
∂p D

The measures {μD z }z∈D are called H-caloric measures.

The following lemma, proved in [24], states that it is always possible to
approximate any bounded domain Ω ⊂ Rn by H-regular domains, both from
the inside and from the outside.

Lemma 7.2. Let B be a bounded open set of Rn . Then for every δ > 0 there
exist H-regular domains Aδ , Aδ such that

{x ∈ B|d(x, ∂B) > δ} ⊆ Aδ ⊆ B ⊆ Aδ ⊆ {x ∈ Rn |d(x, B) < δ}.

At this point, one can prove the existence and basic properties of the Green
function for any regular cylinder R × Ω.

Theorem 7.3. Let Ω ⊆ Rn be an H-regular domain. Then there exists a

Green function G = GΩ on the cylinder R × Ω, with the properties listed
below.
(i) G is a continuous function defined on the set {(z, ζ) ∈ (R×Ω)×(R×Ω) :
z = ζ}. Moreover, for every fixed ζ ∈ R×Ω, G(·; ζ) ∈ C ∞ ((R×Ω)\{ζ}),
and we have

H(G(·; ζ)) = 0 in (R × Ω) \ {ζ}, G(·; ζ) = 0 in R×∂Ω.

(ii) We have 0 ≤ G ≤ h. Moreover G(t, x; τ, ξ) = 0 if t < τ .

(iii) For every ϕ ∈ C(Ω) such that ϕ = 0 in ∂Ω and for every fixed τ ∈ R,
the function

u (t, x) = G(t, x; τ, ξ) ϕ(ξ) dξ, x ∈ Ω, t > τ
Ω

belongs to the class C ∞ ((τ, ∞) × Ω) ∩ C([τ, ∞) × Ω) and solves

Hu = 0 in (τ, ∞) × Ω,
u = 0 in [τ, ∞) × ∂Ω,
u(τ, ·) = ϕ in Ω.

From this theorem, one easily obtains the following corollary.

52 E. Lanconelli

Corollary 7.4. Let Ω ⊆ Rn be an H-regular domain and let GΩ denote

the related Green function as in Theorem 7.3. The following reproduction
property of GΩ holds:

GΩ (t, x; τ, ξ) = GΩ (t, x; s, y) GΩ (s, y; τ, ξ)dy,
Ω

for every t > s > τ and x, ξ ∈ Ω.

Proof. We fix τ, ξ, s as above and we set ϕ = GΩ (·, s; τ, ξ). Then ϕ ∈ C(Ω),
ϕ = 0 in ∂Ω, by Theorem 7.3-(i). Therefore we can apply Theorem 7.3-(iii)
and obtain that the function

u (t, x) = GΩ (t, x; s, y) ϕ(y)dy, x ∈ Ω, t > s,
Ω

satisfies u ∈ C ∞ ((s, ∞) × Ω) ∩ C([s, ∞) × Ω), Hu = 0 in (s, ∞) × Ω, u = 0 in

[s, ∞) × ∂Ω, u(s, ·) = ϕ in Ω. It is now sufficient to observe that GΩ (·; τ, ξ)
has the same properties and to use the Picone maximum principle for H

We now specialize the study of Green functions to cylinders based on

regular domains which approximate metric balls.
Fix δ0 ∈ (0, 1). By Lemma 7.2, for every ξ0 ∈ Rn and R > 0, there exists
an H-regular domain A(ξ0 , R) of Rn such that

B(ξ0 , δ0 R) ⊆ A(ξ0 , R) ⊆ B(ξ0 , R). (22)

Then:
Lemma 7.5. Let R0 > 0 and δ ∈ (0, δ0 ). There exists a constant ρ =
c(δ, δ0 , R0 )−1 ∈ (0, 1), such that

GA(ξ0 ,R) (t, x; τ, ξ) ≥ c(R0 )−1 E(x, ξ, c−1 (t − τ )),

for every ξ0 ∈ Rn , R ∈ (0, R0 ], x ∈ A(ξ0 , R), ξ ∈ B(ξ0 , δR), t, τ ∈ R

satisfying d2 (x, ξ) < t − τ < ρ R2 .
Proof. Let ξ0 , R, x, ξ, t, τ be as above. Let us set

D = (τ − 1, t + 1) × A(ξ0 , R).

(t,x) ((t, t + 1] × ∂A(ξ0 , R)) = 0 we have

Observing that μD

A(ξ0 ,R) D
Ψ (t, x; τ, ξ) = Ψ(τ,ξ) (t, x) = Γ (s, y; τ, ξ) μD
(t,x) (s, y)
∂p D

≤ c(R0 ) E(ξ, y, c(s − τ )) μD
(t,x) (s, y)
[τ,t]∂A(ξ0 ,R)
 
√ (δ0 − δ)2 R2
≤ c(R0 ) sup |B(ξ, r)|−1 exp − .
0<r<t−τ cr
Heat Kernels in Sub-Riemannian Settings 53

In the last inequality we have used the fact that ξ ∈ B(ξ0 , δR) and that

(t,x) (∂p D) ≡ 1
μD

(recall the operator H is homogeneous). We now exploit the Gaussian bounds

of Γ , and obtain

GA(ξ0 ,R) (t, x; τ, ξ) = Γ (t, x; τ, ξ) − Ψ A(ξ0 ,R) (t, x; τ, ξ)

 
√ d2 (x, ξ)
≥ c1 (R0 )−1 |B(ξ, t − τ )|−1 exp −c ×
t−τ
 √  2 
|B(ξ, R ρ)| d (x, ξ) R2
× 1 − c2 (R0 ) sup √ exp c −
0<r<t−τ |B(ξ, r)| t−τ c(δ, δ0 ) r
≥ c3 (R0 )−1 E(x, ξ, c−1 (t − τ )) ×
  Q/2  
ρ R2 R2
× 1 − c4 (R0 ) sup exp − .
0<r<t−τ r c(δ, δ0 ) r

It is now sufficient to observe that the expression between square brackets is

greater than 1/2 if ρ = ρ(δ, δ0 , R0 ) is small enough, as one can easily recognize
by showing that the function

h → (ρ h)Q/2 exp(−h/c(δ, δ0 ))

is monotone decreasing on the interval [ρ−1 , ∞).

We now prove the lower Gaussian bound for the Green function.

Theorem 7.6. Let R0 > 0, T > 1 and γ ∈ (0, δ0 ). We have

GA(ξ0 ,R) (t, x; τ, ξ) ≥

 
√ d2 (x, ξ)
c(T, γ, R0 , δ0 )−1 |B(ξ, t − τ )|−1 exp −c(γ, R0 , δ0 ) ,
t−τ
for every ξ0 ∈ Rn , R ∈ (0, R0 ], x, ξ ∈ B(ξ0 , γR), and 0 < t − τ < T R2 .

Proof.
We set δ = (γ + δ0 )/2 and choose ρ = ρ(δ, δ0 , R0 ) as in Lemma 7.5. Let us
also fix ξ0 , R, x, ξ, t, τ as above. Let k be the smallest integer greater than
 
M (γ, δ0 ) max T /ρ, d2 (x, ξ)/(t − τ ) ,

where the constant M (γ, δ0 ) > 1 will be chosen later, and let us set

1
σ= (t − τ )/(k + 1).
4
54 E. Lanconelli

We claim that there exists a chain of points of Rn x = x0 , x1 , . . . , xk+1 = ξ

such that
d(x, ξ) γ+δ
d(xj , xj+1 ) ≤ c(γ, δ0 ) , d(xj , ξ0 ) ≤ R. (23)
k+1 2
Indeed, if d(x, ξ) ≤ R(δ0 − γ)/8, we can choose x1 , . . . , xk laying on a suitable
X-subunit path connecting x and ξ, so that d(xj , xj+1 ) ≤ 2d(x, ξ)/(k + 1)
and

d(xj , ξ0 ) ≤ d(x, xj ) + d(x, ξ0 ) ≤ 2d(x, ξ) + d(x, ξ0 )

≤ R(δ0 − γ)/4 + γR = R(δ + γ)/2.

On the other hand, if d(x, ξ) > R(δ0 − γ)/8, then we can choose x1 , . . . , xk
laying on suitable X-subunit paths connecting x with ξ0 and ξ0 with ξ, so that

d(xj , xj+1 ) ≤ 2γR/(k + 1) < 16γd(x, ξ)(δ0 − γ)−1 (k + 1)−1

and d(xj , ξ0 ) ≤ γR. Observing that, by the definition of k and σ, we have

σ ≤ M (γ, δ0 )−1/2 R, from (23) it follows that we can choose M (γ, δ0 ) such
that
B(xj , σ) ⊆ B(ξ0 , δR). (24)
Moreover, up to a new choice of M (γ, δ0 ), we also have

t−τ
d(yj , yj+1 ) < for every yj ∈ B(xj , σ), yj+1 ∈ B(xj+1 , σ). (25)
k+1

Indeed, from (23) and the definition of k it follows that

1 t−τ d(x, ξ)
d(yj , yj+1 ) ≤ 2σ + d(xj , xj+1 ) ≤ + c(γ, δ0 )
2 k+1 k+1
 
t−τ 1
≤ + c(γ, δ0 ) M (γ, δ0 )−1/2 .
k+1 2

Let now t = t0 , t1 , . . . , tk+1 = τ be such that tj − tj+1 = (t − τ )/(k + 1) for

j = 0, . . . , k. Using Corollary 7.4 repeatedly, we obtain (we set G = GA(ξ0 ,R) ,
y0 = x, yk+1 = ξ)

G(t, x; τ, ξ) = G(t, x; y1 , t1 ) G(y1 , t1 ; y2 , t2 ) · · · G(yk , tk ; τ, ξ) y1 · · · yk
(A(ξ0 ,R))k
 
k
≥ !k
G(yj , tj ; yj+1 , tj+1 ) y1 · · · yk ,
j=1 B(xj ,σ) j=0

by (24). Moreover, from (24), (25) and the definition of k, it follows that
yj+1 ∈ B(ξ0 , δR),
Heat Kernels in Sub-Riemannian Settings 55

d2 (yj , yj+1 ) < (t − τ )/(k + 1) = tj − tj+1 < T R2 /(k + 1) < ρR2 .

Therefore, we can apply Lemma 7.5 and obtain

 
k
−(k+1)
G(t, x; τ, ξ) ≥ c(R0 ) !k
E(yj , yj+1 , c−1 (tj −tj+1 ))dy1 · · ·dyk
j=1B(xj ,σ) j=0

1 k
|B(xj , σ)|
≥ c(R0 )−1 exp(−c(R0 ) k)
|B(x, cσ)| j=1 |B(xj , cσ)|
≥ c(R0 )−1 |B(x, cσ)|−1 exp(−c(R0 ) k)
√
≥ c(R0 )−1 |B(x, t − τ )|−1 exp(−c(R0 ) k),

by the definition of σ. Now, if d2 (x, ξ) ≥ T (t − τ )/ρ, from the definition of k

it follows that k < c(γ, δ0 ) d2 (x, ξ)/(t − τ ) and then
 
−1
√ −1 d2 (x, ξ)
G(t, x; τ, ξ) ≥ c(R0 ) |B(x, t − τ )| exp −c(γ, R0 , δ0 ) .
t−τ

On the other hand, if d2 (x, ξ) < T (t − τ )/ρ, the definition of k gives k <
c(γ, δ0 )T /ρ and then
√
G(t, x; τ, ξ) ≥ c(T, γ, R0 , δ0 )−1 |B(x, t − τ )|−1 .

This completes the proof.

Harnack Inequality

We start by giving the following oscillation lemma

Lemma 7.7. Let R0 > 0 and γ ∈ (0, δ0 ). There exists a constant μ ∈ (0, 1),
μ = c(γ, R0 , δ0 ), such that for every (τ0 , ξ0 ) ∈ Rn+1 , R ∈ (0, R0 ] and every

u ∈ C ∞ ((τ0 − R2 , τ0 ) × B(ξ0 , R)) ∩ C([τ0 − R2 , τ0 ] × B(ξ0 , R))

satisfying Hu = 0 in (τ0 − R2 , τ0 ) × B(ξ0 , R), we have

osc[τ0 −γ 2 R2 ,τ0 ]×B(ξ0 ,γR) u ≤ μosc[τ0 −R2 ,τ0 ]×B(ξ0 ,R) u. (26)

Proof. We set
D = (τ0 − R2 , τ0 ) × A(ξ0 , R)
and  
S = x ∈ B(ξ0 , γR) | u(x, τ0 − R2 ) ≥ (M + m)/2 ,
56 E. Lanconelli

where M = maxD u, m = minD u. We also define w = u − m and (for

x ∈ A(ξ0 , R), t > τ0 − R2 )

v (t, x) = GA(ξ0 ,R) (t, x; τ0 − R2 , y) w(τ0 − R2 , y) ϕ(y)d y,
A(ξ0 ,R)

where ϕ ∈ C0 (A(ξ0 , R)) is a cut-off function such that 0 ≤ ϕ ≤ 1 and ϕ ≡ 1

in B(ξ0 , γR). By means of Theorem 7.3-(iii), v is a solution to Hv = 0 in D,
v = 0 in [τ0 − R2 , τ0 ] × ∂A(ξ0 , R), v(·, τ0 − R2 ) = w(τ0 − R2 , ·) ϕ in A(ξ0 , R).
Moreover Hw = 0 since we are supposing that a0 = 0. Therefore, by the
weak maximum principle for H, v ≤ w in D. As a consequence, using the
estimate in Theorem 7.6, for every

(t, x) ∈ Dγ = (τ0 − γ 2 R2 , τ0 ) × B(ξ0 , γR)

we get
  
M +m
w (t, x) ≥ v (t, x) ≥ G (t, x; τ0 − R , y)
A(ξ0 ,R) 2
− m dy
2
 S
c 2 2 M − m
≥  √
 e−d(x,y) /c(t−τ +R ) dy

S B x, t − τ + R
2  2
M −m
≥ c(γ, R0 , δ0 )−1 |B(x, R)|−1 |S|
2
M −m
≥ c(γ, R0 , δ0 )−1 |B(ξ0 , R)|−1 |S|.
2
Now, if |S| ≥ 12 |B(ξ0 , γR)|, we infer that

min u − m ≥ c(γ, R0 , δ0 )−1 (M − m)

Dγ

and then

oscDγ u ≤ M − min u ≤ M − m − c(γ, R0 , δ0 )−1 (M − m)

Dγ

= 1 − c(γ, R0 , δ0 )−1 oscD u.

Recalling that A(ξ0 , R) ⊆ B(ξ0 , R), we have proved (26) when |S| ≥
2 |B(ξ0 , γR)|. On the other hand, if |S| < 2 |B(ξ0 , γR)|, the argument
1 1

above can be applied to u  := −u, since (with the natural notation) |S|  >
2 |B(ξ0 , γR)|. As a consequence, we get (26) for u
 and the proof is completed,
1

since osc u = osc u.

We also have the following crucial lemma.

Heat Kernels in Sub-Riemannian Settings 57

Lemma 7.8. Let R0 > 0, h ∈ (0, 1) and γ ∈ (0, δ0 ). There exists a positive
constant β = c(h, γ, R0 , δ0 ) such that

sup σ |{y ∈ B(ξ0 , γR) | u(s, y) ≥ σ}| ≤ β |B(ξ0 , R)| u(ξ0 , τ0 )

σ>0, s∈[τ0 −R2 ,τ0 −hR2 ]

for every (ξ0 , τ0 ) ∈ Rn+1 , R ∈ (0, R0 ] and every

u ∈ C ∞ (B(ξ0 , R) × (τ0 − R2 , τ0 )) ∩ C(B(ξ0 , R) × [τ0 − R2 , τ0 ])

satisfying Hu = 0, u ≥ 0 in B(ξ0 , R) × (τ0 − R2 , τ0 ).

Proof. Let us fix s ∈ [τ0 − R2 , τ0 − hR2 ]; set A = A(ξ0 , R) and


w (t, x) = GA (t, x; s, y) u(s, y) ϕ(y) dy, x ∈ A, t > s,
A

where ϕ ∈ C0 (A) is a cut-off function such that 0 ≤ ϕ ≤ 1 and ϕ ≡ 1

in B(ξ0 , γR). By means of Theorem 7.3-(iii), w is a solution to Hw = 0 in
(s, ∞) × A, w = 0 in [s, ∞) × ∂A, w(s, ·) = u(s, ·) ϕ in A. Therefore, by the
weak maximum principle for H, w ≤ u in [s, τ0 ] × A. As a consequence, using
the estimate in Theorem 7.6, we obtain

cu (s, y) ϕ (y) −d(ξ0 ,y)2 /c(τ0 −s)
u(τ0 , ξ0 ) ≥ w(τ0 , ξ0 ) ≥ √ e dy
A |B(ξ0 , τ0 − s)|

since τ0 − s ≥ hR2 and d (ξ0 , y) ≤ cR


c(h, γ, R0 , δ0 )
≥ u(s, y) dy
|B(ξ0 , R)| B(ξ0 ,γR)
c(h, γ, R0 , δ0 )
≥ σ |{y ∈ B(ξ0 , γR) | u(s, y) ≥ σ}|
|B(ξ0 , R)|

for any σ. This ends the proof.

Next theorem follows from the previous two Lemmas, with the same
technique used in [13].

Theorem 7.9. Let R0 > 0, 0 < h1 < h2 < 1 and γ ∈ (0, 1). There exists
a positive constant M = c(h1 , h2 , γ, R0 ) such that for every (ξ0 , τ0 ) ∈ Rn+1 ,
R ∈ (0, R0 ] and every

u ∈ C ∞ ((τ0 − R2 , τ0 ) × B(ξ0 , R)) ∩ C([τ0 − R2 , τ0 ] × B(ξ0 , R))

satisfying Hu = 0, u ≥ 0 in (τ0 − R2 , τ0 ) × B(ξ0 , R), we have

max u ≤ M u(ξ0 , τ0 ). (27)

[τ0 −h2 R2 ,τ0 −h1 R2 ]×B(ξ0 ,γR)
58 E. Lanconelli

Proof of Theorem 7.9. Recall that our previous estimates depend on a

number δ0 ∈ (0, 1) which can be arbitrarily chosen (see (22)). Then, for any
fixed γ ∈ (0, 1) , pick δ0 = (1 + γ) /2. We will apply all our previous results
with this particular choice of δ0 .
Let μ = μ( δ20 , R0 , δ0 ) ∈ (0, 1), β = β(h1 , γ+δ0
2 , R0 , δ0 ) > 0 be as in
Lemma 7.7 and Lemma 7.8 respectively. We define r : (0, ∞) → (0, ∞),
 1/Q
4 βc0
r(σ) = 2
σ(1 − μ)

and we set
K = (1 + μ−1 )/2
and  
M = r−1 δ0 (1 − h2 ) (δ0 − γ) (1 − K −1/Q )/4 .

We now argue by contradiction and suppose that there exist ξ0 , τ0 , R and

u satisfying the hypotheses of the theorem, for which (27) is not true (with
the above choice of M ). We first observe that u(τ0 , ξ0 ) = 0, since otherwise
(27) would follow from Lemma (7.8). Let now v = u/u(τ0 , ξ0 ). Since v is
bounded, in order to get a contradiction and thus prove the theorem, it
is sufficient to show that there exists a sequence of points {(sj , yj )}j∈N in
[τ0 − R2 , τ0 ] × B(ξ0 , R) such that

v(sj , yj ) ≥ K j M, (sj , yj ) ∈ [τ0 − R2 , τ0 ] × B(ξ0 , R).

Indeed, recalling that K > 1, this would give v(sj , yj ) → ∞. To construct

this sequence, we will prove by induction the existence of points (sj , yj ) ∈
[τ0 − R2 , τ0 ] × B(ξ0 , R) such that:

v(sj , yj ) ≥ K j M (28)
(s0 , y0 ) ∈ [τ0 − h2 R , τ0 − h1 R ] × B(ξ0 , γR)
2 2

(sj , yj ) ∈ [sj−1 − ρ2j−1 , sj−1 ] × B(yj−1 , ρj−1 ) if j ≥ 1

with ρj = 2δ0−1 r(K j M )R

The existence of (s0 , y0 ) ∈ [τ0 − h2 R2 , τ0 − h1 R2 ] × B(ξ0 , γR) such that

v(s0 , y0 ) ≥ M follows from the assumption that u does not satisfy (27). We
now suppose that, for a fixed q ∈ N, (s0 , y0 ), . . . , (sq , yq ) have been defined
and satisfy (28) for every j ∈ {0, . . . , q}. We have to prove that we can find
(sq+1 , yq+1 ) satisfying (28) for j = q + 1. We claim that

B(yq , ρq ) ⊆ B(ξ0 , (γ + δ0 )R/2). (29)

Heat Kernels in Sub-Riemannian Settings 59

Indeed, if d(y, yq ) ≤ ρq , then recalling the definition of M and using (28) for
j ∈ {0, . . . , q}, we obtain


q
d(y, ξ0 ) ≤ d(ξ0 , y0 ) + d(yj−1 , yj ) + d(yq , y)
j=1
q ∞

≤ γR + 2δ0−1 R r(K i M ) < γR + 2δ0−1 r(M ) R K −i/Q
i=0 i=0
= (γ + (1 − h2 )(δ0 − γ)/2) R < (γ + δ0 )R/2.

Moreover, with a similar computation we can prove that

[sq − ρ2q , sq ] ⊆ (τ0 − R2 , τ0 − h1 R2 ]. (30)

Indeed, sq ≤ sq−1 ≤ · · · ≤ s0 ≤ τ0 − h1 R2 and

q
sq − ρ2q = s0 + j=1 (sj − sj−1 ) − ρ2q
∞
> τ0 − h2 R2 − 4δ0−2 (r(M ))2 R2 i=0 K −2i/Q

> τ0 − h2 + (1 − h2 )(1 − K −1/Q )(1 + K −1/Q )−1 R2 > τ0 − R2 .

We now apply Lemma 7.8 (with σ = (1 − μ)K q M/2) to v and we obtain

(recalling (30) and the definition of r)

|{y ∈ B(ξ0 , (γ + δ0 )R/2) | v(y, sq ) ≥ (1 − μ)K q M/2}|

 q
Q
1 −1 r(K M)
≤ 2β|B(ξ 0 ,R)|
(1−μ)K M q = c
2 0 2 |B(ξ0 , R)|
  Q
r(K q M)
< c−1
0 2 |B(yq , 2R)| ≤ |{B(yq , r(K q M )R)}| .

In the last inequality we have used the fact that r(K q M ) ≤ 2, which follows
from the definition of M (see the proof of (29)). As a consequence, since also
(29) holds, there exists
y ∈ B(yq , r(K q M )R)
such that
v(sq , y) < (1 − μ)K q M/2.
Therefore, recalling that we are supposing that (28) holds for j = q, we have

(1 + μ)K q M/2 = K q M − (1 − μ)K q M/2 < v(sq , yq ) − v(sq , y)

≤ osc{sq }×B(yq ,r(K q M)R) v ≤ μosc[sq −ρ2 ,sq ]×B(yq ,ρq ) v
q

by means of Lemma 7.7, (29) and (30) (note that ρq ≤ R0 by the definition
of M ). Since v ≥ 0, it follows that there exists

(sq+1 , yq+1 ) ∈ [sq − ρ2q , sq ] × B(yq , ρq )

60 E. Lanconelli

such that
v(sq+1 , yq+1 ) > μ−1 (1 + μ)K q M/2 = K q+1 M.
This completes the proof of Theorem 7.9.

References

1. D. G. Aronson: Bounds for the fundamental solution of a parabolic equation. Bull.

Amer. Math. Soc. 73 1967 890–896.
2. E. Bedford, B. Gaveau: Hypersurfaces with bounded Levi form. Indiana Univ. Math.
J. 27 (1978), no. 5, 867–873.
3. A. Bonfiglioli, E. Lanconelli, F. Uguzzoni: Uniform Gaussian estimates of the funda-
mental solutions for heat operators on Carnot groups, Adv. Differential Equations, 7
(2002), 1153–1192.
4. A. Bonfiglioli, E. Lanconelli, F. Uguzzoni: Fundamental solutions for non-divergence
form operators on stratified groups. Trans. Amer. Math. Soc. 356 (2004), no. 7, 2709–
2737.
5. A. Bonfiglioli, F. Uguzzoni: Harnack inequality for non-divergence form operators on
stratified groups, to appear in Trans. Amer. Math. Soc.
6. J.-M. Bony: Principe du maximum, inégalite de Harnack et unicité du problème de
Cauchy pour les opérateurs elliptiques dégénérés. Ann. Inst. Fourier (Grenoble) 19
1969 fasc. 1, 277–304 xii.
7. M. Bramanti, L. Brandolini: Lp -estimates for uniformly hypoelliptic operators with
discontinuous coefficients on homogeneous groups. Rend. Sem. Mat. dell’Univ. e del
Politec. di Torino. Vol. 58, 4 (2000), 389–433.
8. M. Bramanti, L. Brandolini: Lp -estimates for nonvariational hypoelliptic operators
with VMO coefficients. Trans. Amer. Math. Soc. 352 (2000), no. 2, 781–822.
9. M. Bramanti, L. Brandolini, E. Lanconelli, F. Uguzzoni: Heat kernels for non-
divergence operators of Hörmander type. C.R. Acad. Sci. Paris, Mathematique, 343
(2006) 463–466.
10. M. Bramanti, L. Brandolini, E. Lanconelli, F. Uguzzoni: Non-divergence equations
structured on Hörmander vector fields: heat kernels and Harnack inequalities, preprint.
11. L. Caffarelli, L. Nirenberg, J. Spruck: The Dirichlet problem for nonlinear second-
order elliptic equations. III. Functions of the eigenvalues of the Hessian. Acta Math.
155 (1985), no. 3-4, 261–301.
12. L. Capogna, Q. Han: Pointwise Schauder estimates for second order linear equations
in Carnot groups. Harmonic analysis at Mount Holyoke (South Hadley, MA, 2001),
45–69, Contemp. Math., 320, Amer. Math. Soc., Providence, RI, 2003.
13. E. B. Fabes, D. W. Stroock, A new proof of Moser’s parabolic Harnack inequality using
the old ideas of Nash, Arch. Rational Mech. Anal. 96 (1986), 327–338.
14. C. Fefferman, A. Sánchez-Calle: Fundamental solutions for second order subelliptic
operators. Ann. of Math. (2) 124 (1986), no. 2, 247–272.
15. C. Fefferman, D. H. Phong: Subelliptic eigenvalue problems. Conference on har-
monic analysis in honor of Antoni Zygmund, Vol. I, II (Chicago, Ill., 1981), 590–606,
Wadsworth Math. Ser., Wadsworth, Belmont, CA, 1983.
16. A. Friedman: Partial differential equations of parabolic type. Prentice-Hall, Inc.,
Englewood Cliffs, N.J. 1964.
17. L. Hörmander: Hypoelliptic second order differential equations. Acta Math. 119 (1967)
147–171.
18. G. Huisken, W. Klingenberg: Flow of real hypersurfaces by the trace of the Levi form.
Math. Res. Lett. 6 (1999), no. 5-6, 645–661.
19. D. S. Jerison, A. Sánchez-Calle: Estimates for the heat kernel for a sum of squares of
vector fields. Indiana Univ. Math. J. 35 (1986), no. 4, 835–854.
Heat Kernels in Sub-Riemannian Settings 61

20. S. Kusuoka, D. Stroock: Applications of the Malliavin calculus. II. J. Fac. Sci. Univ.
Tokyo Sect. IA Math. 32 (1985), no. 1, 1–76.
21. S. Kusuoka, D. Stroock: Applications of the Malliavin calculus. III. J. Fac. Sci. Univ.
Tokyo Sect. IA Math. 34 (1987), no. 2, 391–442.
22. S. Kusuoka, D. Stroock: Long time estimates for the heat kernel associated with a
uniformly subelliptic symmetric second order operator. Ann. of Math. (2) 127 (1988),
no. 1, 165–189.
23. E. Lanconelli, S. Polidoro: On a class of hypoelliptic evolution operators. Partial dif-
ferential equations, II (Turin, 1993). Rend. Sem. Mat. Univ. Politec. Torino 52 (1994),
no. 1, 29–63.
24. E. Lanconelli, F. Uguzzoni: Potential Analysis for a Class of Diffusion Equations: a
Gaussian Bounds Approach, Preprint.
25. E. E. Levi: Sulle equazioni lineari totalmente ellittiche alle derivate parziali. Rend.
Circ. Mat. Palermo 24, 275–317 (1907) (Also with corrections in: Eugenio Elia Levi,
Opere, vol. II, 28–84. Roma, Edizioni Cremonese 1960).
26. E. E. Levi: I problemi dei valori al contorno per le equazioni lineari totalmente ellittiche
alle derivate parziali, Memorie Mat. Fis. Soc. Ital. Scienza (detta dei XL) (3) 16, 3–113
(1909). (Also with corrections in: Eugenio Elia Levi, Opere, vol. II, 207–343. Roma,
Edizioni Cremonese 1960).
27. V. Martino, A. Montanari : Integral formulas for a class of curvature PDE’s and
applications to isoperimetric inequalities and to symmetry problems, Preprint.
28. A. Montanari: Real hypersurfaces evolving by Levi curvature: smooth regularity of
solutions to the parabolic Levi equation. Comm. Partial Differential Equations 26
(2001), no. 9-10, 1633–1664.
29. A. Montanari, E. Lanconelli: Pseudoconvex fully nonlinear partial differential oper-
ators: strong comparison theorems. J. Differential Equations 202 (2004), no. 2,
306–331.
30. A. Nagel, E. M. Stein, S. Wainger: Balls and metrics defined by vector fields I: Basic
properties. Acta Mathematica, 155 (1985), 130–147.
31. J. Nash: Continuity of solutions of parabolic and elliptic equations. Amer. J. Math. 80
(1958) 931–954.
32. M. Picone: Maggiorazione degli integrali di equazioni lineari ellittico-paraboliche alle
derivate parziali del secondo ordine. Rend. Accad. Naz. Lincei 5, n.6 (1927) 138–143.
33. S. Polidoro: On a class of ultraparabolic operators of Kolmogorov-Fokker-Planck type.
Le Matematiche (Catania) 49 (1994), no. 1, 53–105.
34. L. P. Rothschild, E. M. Stein: Hypoelliptic differential operators and nilpotent groups.
Acta Math., 137 (1976), 247–320.
35. L. Saloff-Coste, Aspects of Sobolev-type inequalities. London Mathematical Society
Lecture Note Series, 289. Cambridge University Press, Cambridge, 2002.
36. L. Saloff-Coste, D. W. Stroock: Opérateurs uniformément sous-elliptiques sur les
groupes de Lie. J. Funct. Anal. 98 (1991), no. 1, 97–121.
37. A. Sanchez-Calle: Fundamental solutions and geometry of sum of squares of vector
fields. Inv. Math., 78 (1984), 143–160.
38. Z. Slodkowski, G. Tomassini: Weak solutions for the Levi equation and envelope of
holomorphy. J. Funct. Anal. 101 (1991), no. 2, 392–407.
39. G. Tomassini: Geometric properties of solutions of the Levi-equation. Ann. Mat. Pura
Appl. (4) 152 (1988), 331–344.
40. N. Th. Varopoulos: Théorie du potentiel sur les groupes nilpotents. C. R. Acad. Sci.
Paris Sér. I Math. 301 (1985), no. 5, 143–144.
41. N. Th. Varopoulos, Analysis on nilpotent groups. J. Funct. Anal. 66 (1986), no. 3,
406–431.
42. N. Th. Varopoulos, L. Saloff-Coste, T. Coulhon: Analysis and geometry on groups.
Cambridge Tracts in Mathematics, 100. Cambridge University Press, Cambridge, 1992.
43. C. J. Xu: Regularity for quasilinear second-order subelliptic equations. Comm. Pure
Appl. Math. 45 (1992), no. 1, 77–96.
Concentration of Solutions for Some
Singularly Perturbed Neumann
Problems

Andrea Malchiodi

1 Introduction

The purpose of these notes is to present some techniques for constructing

solutions to a class of singularly perturbed problems with a precise asymptotic
behavior when the perturbation parameter ε tends to zero. We first treat
the case of concentration at points, and then the case of concentration at
manifolds.
One of the main motivations for the study of these equations arises from
reaction-diffusion systems, concerning in particular the so-called Turing’s
instability. More precisely, see [18], [53], it is known that scalar diffusion
equations like "
ut = Δu + f (u) in Ω;
∂u
∂ν = 0 on ∂Ω,
for convex domains Ω admit only constant (linearly) stable steady state solu-
tions. On the other hand, as noticed in [62], reaction-diffusion systems with
different diffusivities might lead to non-homogeneous stable steady states.
A well-know example is the following one (Gierer-Meinhardt)
⎧
⎪ Up
⎨Ut = d1 ΔU − U + Vrq in Ω × (0, +∞),
U
Vt = d2 ΔV − V + V s in Ω × (0, +∞), (GM )
⎪
⎩ ∂U
∂ν = ∂V
∂ν = 0 on ∂Ω × (0, +∞),

introduced in [31] to describe some biological experiment. The functions U

and V represent the densities of some chemical substances, the numbers
p, q, r, s are non-negative and such that 0 < p−1
q
r
< s+1 , and it is assumed

A. Malchiodi
SISSA, Sector of Mathematical Analysis, Via Beirut 2-4, 34014 Trieste, Italy
e-mail: [email protected]

S.-Y.A. Chang et al. (eds.), Geometric Analysis and PDEs, 63

Lecture Notes in Mathematics 1977, DOI: 10.1007/978-3-642-01674-5 3,

c Springer-Verlag Berlin Heidelberg 2009
64 A. Malchiodi

that the diffusivities d1 and d2 satisfy d1  1  d2 . In the stationary case

of (GM ), as explained in [55], [58], when d2 → +∞ the function V is close
to a constant (being nearly harmonic and with zero normal derivative at
the boundary), and therefore the equation satisfied by U converges to the
following one ⎧
⎪−ε2 Δu + u = up in Ω,
⎨
∂u
∂ν = 0 on ∂Ω, (P̃ε )
⎪
⎩
u>0 in Ω,
with ε2 = d1 . One is interested in general in the profile of solutions, their
location and possibly in their stability properties.
The advantage of problem (P̃ε ) compared to (the stationary case of) (GM )
is that, apart from being a scalar equation, its structure is variational and
solutions can be found as critical points of the following Euler-Lagrange
functional
 
1 2
1
I˜ε (u) = ε |∇u|2 + u2 − |u|p+1 ; u ∈ H 1 (Ω).
2 Ω p+1 Ω

The typical concentration behavior of solutions

 uε to (P̃ε ) is via a scaling
of the variables in the form uε (x) ∼ u0 x−Q
ε , where Q is some point of Ω,
and where u0 is a solution of the problem

−Δu0 + u0 = up0 in Rn (or in Rn+ = {(x1 , . . . , xn ) ∈ Rn : xn > 0}),

(1)
the domain depending on whether Q lies in the interior of Ω or at the
boundary; in the latter case Neumann conditions are imposed.
n+2
When p < n−2 (and indeed only if this inequality is satisfied), problem
(1) admits positive radial solutions which decay to zero at infinity, see [14],
[15], [61]. Some details on the existence theory for (1) are reported below.
Solutions of (P̃ε ) which inherit this profile are called spike-layers, since
they are highly concentrated near some point of Ω. There is an extensive
literature regarding this type of solutions, beginning from the papers [44],
[56], [57]. Indeed their structure is very rich, and we refer for example to the
(far from complete) list of papers [21], [27], [33], [34], [35], [36], [42], [43],
[63], [64].
In these notes we are going to prove the following two theorems, to give an
idea of the methods employed in this area of research. The first concerns the
case of local extrema of the mean boundary curvature, while the second deals
with non-degenerate critical points. We will present a general perturbative
argument to deal with such concentration phenomena, and then apply it to
study (P̃ε ).

Theorem 1 Suppose Ω ⊆ Rn , n ≥ 2, is a smooth bounded domain, and

n+2
that 1 < p < n−2 (1 < p < +∞ if n = 2). Suppose X0 ∈ ∂Ω is a local
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 65

strict maximum or minimum of the mean curvature H of ∂Ω. Then for ε > 0
sufficiently small problem (P̃ε ) admits a solution concentrating at X0 .
Theorem 2 Suppose Ω ⊆ Rn , n ≥ 2, is a smooth bounded domain, and that
n−2 (1 < p < +∞ if n = 2). Suppose X0 ∈ ∂Ω is a non-degenerate
1 < p < n+2
critical point of the mean curvature H of ∂Ω. Then for ε > 0 sufficiently
small problem (P̃ε ) admits a solution concentrating at X0 .
We were intentionally vague about the meaning of concentration, but it will
become clear from the constructive proof below. A property of these solutions,
uε , is that they tend to zero uniformly on every compact set of Ω\{X0 }, while
there exists δ > 0 and some point Xε ∈ Ω such that uε (Xε ) ≥ δ as ε → 0.
We notice that the functional Iε , if p is subcritical, satisfies the assumptions
of the mountain-pass theorem (see [9]). It can be shown that the profile of
the mountain-pass solution of (P̃ε ) is a soliton of (1) in the half-space, and it
peaks at a point of ∂Ω with maximal mean curvature. Furthermore, there are
solutions with interior and multiple peaks: in particular in [35] it was shown
that for any couple of integers (k, l) there exist solutions with k boundary
peaks and l interior ones. The energy (i.e. the corresponding value of Iε )
of spike-layers (resp. multiple spikes) is of order εn , which is proportional to
the volume of their support, heuristically identified with a ball (resp. multiple
balls) of radius ε centered at the peak (resp. at each peak). Roughly speaking,
concentration at the boundary occurs at critical points of the mean curvature,
while concentration at the interior occurs at singular points of the distance
function from the boundary. Methods of construction rely mainly on min-max
arguments, finite-dimensional reductions and gluing techniques (for multiple
spikes).
In some cases, see for example [26], [28], [37], [58], it is possible to construct
stationary solutions of (GM ) starting from spike-layer solutions of (P̃ε ) and
using perturbation arguments. In these references some stability conditions
are also given: we notice that non-trivial solutions of (P̃ε ) are always unstable,
but their counterparts in (GM ) may gain stability through the coupling with
a second equation.
There are other motivations for the study of the above kind of equations:
(P̃ε ) arises also as limit of different reaction-diffusion systems (with chemo-
taxis for example, as shown in [55]). Another motivation comes from the
Nonlinear Schrödinger equation
∂ψ
i = −2 Δψ + V (x)ψ − |ψ|p−1 ψ in Rn , (2)
∂t
where ψ is a complex-valued function (the wave function), V is a potential
and p is an exponent greater than 1. Indeed, if one looks for standing waves,
namely solutions of the form ψ(x, t) = e−  u(x), for some real function u,
iωt

then the latter will satisfy

−ε2 Δu + V (x)u = up in Rn , (3)

66 A. Malchiodi

where we have set ε =  and we absorbed the constant ω into the potential V .
Therefore, the problem very similar to (P̃ε ), apart from the addition of a
potential term. Also for this problem we have highly peaked solutions, which
in general concentrate at critical points of the potential V : about this subject
we refer the reader to the (still incomplete) list of papers [1], [2], [3], [7], [8],
[10], [17], [24], [25], [30], [32], [38], [59] and to the bibliographies therein.
Apart from existence of solutions concentrating at points, one may ask
whether there exist others which scale only in some of the variables, and
not all of them, which therefore concentrate at higher dimensional sets, like
curves or manifolds. Indeed under generic assumptions, see [55], if Ω ⊆ Rn
and k = 1, . . . , n− 1, it was conjectured that there exist solutions of (P̃ε ) con-
centrating at k-dimensional sets. The phenomenon was known for particular
domains with some symmetry: for these and related issues see e.g. [5], [6],
[11], [12], [13], [22], [23], [51], [54], [60]. In these cases, some of the techniques
for studying concentration at points can be modified, but these adaptations
do not work for asymmetric situations. In [49], [50] the first general result in
this direction was obtained. It was indeed shown that there exist solutions to
(P̃ε ) which scale in one variable only and which concentrate near the whole
boundary of the domain. Their profile, as one may expect, is given by the
solution of (1) in one dimension, namely
⎧
⎪  p
⎨−w0 + w0 = w0 in R+ ;
w0 (x) → 0 as x → +∞; (4)
⎪
⎩ 
w0 (0) = 0.

The result in [49], [50] is the following.

Theorem 3 Let Ω ⊆ Rn be a smooth bounded domain, and let p > 1. Then
there exists a sequence εj → 0 and a sequence of solutions uεj of (Pεj ) with
the following properties
(i) uεj concentrates
 2
near ∂Ω as j → +∞, namely for every r > 0 one
has Ωr εj |∇uεj | + uεj → 0 as j → +∞, where Ωr = {x ∈ Ω :
2 2

dist(x, ∂Ω) ≥ r};

(ii) if x0 ∈ ∂Ω and if ν0 denotes the interior unit normal to ∂Ω at x0 , then
for every k ∈ N one has uεj (εj (x − x0 )) → w0 (x, ν0 ) in Cloc
k
(V0 ), where
V0 = {x ∈ R : x, ν0 > 0}, and where w0 is the solution of (4).
n

Below we will give an almost complete proof of this result: for the moment
we just illustrate two main features. The first is that the exponent p can be
any real number grater than 1, and no upper bound is required, differently
from Theorems 1 and 2: the reason is that the limit profile of the solutions is
one-dimensional, and for solving (4) no restriction on p is needed. The second
is that in Theorem 3 existence is proved only along a sequence εj → 0: this is
due to an underlying resonance phenomenon, described below, and is peculiar
of higher-dimensional concentration.
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 67

The latter result has been extended to the case of limit sets of dimension
k = 1, . . . , n − 2, first in [48] for n = 3 and k = 1, and then in [46] for the
general case.

Theorem 4 Let Ω ⊆ RN , N ≥ 3, be a smooth and bounded domain, and

let K ⊆ ∂Ω be a compact embedded non-degenerate
 minimal submanifold of
dimension k ∈ {1, . . . , n−2}. Then, if p ∈ 1, n−k−2
n−k+2
, there exists a sequence
εj → 0 such that (Pεj ) admits positive solutions uεj concentrating along K as
j → ∞. Precisely there exists a positive constant C, depending on Ω, K and
dist(x,K)
−
p such that for any x ∈ Ω uεj (x) ≤ Ce Cεj
; moreover for any q ∈ K,
in a system of coordinates (y, ζ), with y coordinates on K and ζ coordinates
m
Cloc (Rn−k
+ )
normal to K, for any integer m one has uεj (0, εj ·) −→ w0 (·), where
w0 : Rn−k
+ → R is the unique radial solution of
⎧
⎪
⎨−Δu + u = u
p
in Rn−k
+ ,
∂u
∂ν = 0 on ∂Rn−k
+ , (5)
⎪
⎩
u > 0, u ∈ H (R+ ).
1 n−k

By minimal submanifold we mean that K is stationary for the area

functional, namely its mean curvature in ∂Ω vanishes: this condition is
indeed quite natural, since a heuristic expansion for the energy of solu-
tions concentrated near a submanifold K̃ gives C0 area(K̃) εn−k + o(εn−k ),
with C0 a given positive constant. Therefore, since the energy of solutions
should be extremal, also the area of K has to be.  We notice that also the
bound on the exponent p is natural, since p ∈ 1, n−k−2 n−k+2
is a necessary
and sufficient condition for having solutions to (5) which decay to zero at
infinity.
For reasons of brevity, we will not give the proof of the latter theorem,
referring to [46] for complete details. Further results in this direction can be
found in [29], [47], [52], [65].
68 A. Malchiodi

Part I
Concentration at Points
In this part we investigate the case of concentration at points, and prove The-
orems 1 and 2. We first present a general perturbative technique introduced
by Ambrosetti and Badiale, and then apply it to our specific problem.
We are interested in finding solutions to (P̃ε ) with a specific asymptotic
profile, so it is convenient to make the change of variables x → εx and to
study the problem in the dilated domain
1
Ωε := Ω.
ε
After this change of variables the problem becomes
⎧
⎪
⎨−Δu + u = u
p
in Ωε ,
∂u
∂ν = 0 on ∂Ωε , (Pε )
⎪
⎩
u>0 in Ωε .

The corresponding Euler functional (for p ∈ (1, n+2

n−2 )) is then
 
1
1
Iε (u) = |∇u|2 + u2 − |u|p+1 ; u ∈ H 1 (Ωε ). (6)
2 Ωε p+1 Ωε

As already mentioned, the limit profiles for solutions to (Pε ) when ε tends
to zero is given by the function in (1). We next review the existence theory for
the latter equation: the arguments are rather standard but for the reader’s
convenience we sketch the proof below. Solutions to (1) in Rn can be found
looking for minima of the Sobolev quotient (up to a Lagrange multiplier which
can be eliminated multiplying the minimizer by a uitable positive constant)

Rn |∇u| + u
2 2
min
2 . (7)
u∈H 1 (Rn )
Rn |u|
p+1 p+1

Since a spherical decreasing rearrangement decreases the above ratio, see [40],
we can restrict ourselves to the class Hr1 (Rn ), the functions in H 1 (Rn ) which
have radial symmetry and are radially non-increasing1. In the class of radial
functions we have the following result.

1 If one wants to avoid the use of rearrangements, it is simply possible to find minima of
(7) within the radial class, but the information that they are global minima will be lost.
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 69

Lemma 5 Let n ≥ 3. Then there exists cn > 0, depending only on n, such

that for all u ∈ Hr1 (Rn )

|u(r)| ≤ cn r(1−n)/2 u, ∀ r ≥ 1. (8)

Proof. By density, we can suppose that u ∈ C0∞ (Rn ). If the prime symbol
denotes the derivative with respect to r, we have (rn−1 u2 ) = 2rn−1 uu +
(n − 1)rn−2 u2 , whence (rn−1 u2 ) ≥ 2rn−1 uu . Integrating over [r, ∞) we find
 ∞
rn−1 u2 (r) ≤ −2 rn−1 uu dr ≤ cu2,
r

where c depends on n, only, proving (8).

Corollary 6 [61] The embedding of Hr1 (Rn ) into Lq (Rn ), n ≥ 3, is compact

for all 2 < q < 2∗ . For n = 2, the embedding is compact for all q > 2.
Proof. Suppose n ≥ 3, and let (uk )k ⊆ Hr1 (Rn ) be such that uk  0 as
k → +∞. From (8) it follows that

|uk (r)| ≤ C1 r(1−n)/2 uk  ≤ C2 r(1−n)/2 .

Since q > 2 we deduce that, given ε > 0, there exists C3 > 0 and R > 0 such
that |uk (r)|q ≤ C3 ε |uk (r)|2 , for all r ≥ R. This implies
 
|uk (x)|q ≤ C3 ε |uk (x)|2 ≤ C3 ε uk 2 ≤ C4 ε. (9)
|x|≥R |x|≥R

Moreover, from the standard Sobolev embedding Theorem, we have that

uk → 0 strongly in Lq (BR ), for every 2 ≤ q < 2∗ . Thus there exists k0 > 0
such that for all k ≥ k0 one has

|uk (x)|q ≤ ε.
|x|≤R

This and (9) imply that Rn |uk (x)|q ≤ C5 ε for k ≥ k0 , proving that uk → 0,
strongly in Lq (Rn ), for every 2 < q < 2∗ . For n = 2 the proof is identical.
 
Proposition 7 If p ∈ 1, n+2
n−2 problem (1) admits a positive solution U .

Proof. We consider a radial sequence (ul )l which is minimizing for (7): by

the Sobolev embeddings the infimum is bounded from below by a positive
constant. We can assume
n the H 1 norm of each function ul is 1, so by the latter
bound we find that R |ul |p+1
≥ δ0 > 0 for every l. By the compactness of the
embedding for radial functions, ul converges strongly to a non-zero function
U0 , and also
U0  ≤ lim inf ul .
l
70 A. Malchiodi

Therefore

Rn |∇U0 |2 + U02 Rn |∇ul | + ul

2 2

2 ≤ liml inf
2 ,
Rn |U0 | Rn |ul |
p+1 p+1 p+1 p+1

so U0 realizes the minimum in (7). Finally U can be obtained from U0 simply

by scaling (to remove the Lagrange multiplier).

Remark 8 It is possible to prove that the radial solutions are unique (see
[41]) and decay to zero exponentially: more precisely satisfy the property

n−1 U  (r) U  (r)

lim er r 2 U (r) = αn,p ; lim = − lim = −1, (10)
r→+∞ r→+∞ U (r) r→+∞ U (r)

where αn,p is a positive constant depending only on n and p.

2 Perturbation in Critical Point Theory

In this and in the subsequent section we will discuss the existence of critical
points for a class of functionals which are perturbative in nature. Given a
Hilbert space H (which might as well depend on ε), we are interested in
functionals Iε : H → R of class C 2 which satisfy the following properties

i) there exists a smooth finite-dimensional manifold (compact or not) Zε ⊆

H such that Iε (z) ≤ Cε for every z ∈ Zε and for some fixed constant C
(independent of z and ε): moreover Iε (z)[q] ≤ Cεq for every z ∈ Zε
and every q ∈ Tz Zε ;
ii) there exist C, α ∈ (0, 1) and r0 > 0 (independent of ε) such that
Iε C α ≤ C in the subset {u : dist(u, Zε ) < r0 };
iii) letting Pz , z ∈ Zε , denote the projection onto the orthogonal comple-
ment of Tz Zε , there exists C > 0 (independent of z and ε) such that
Pz Iε (z), restricted to (Tz Zε )⊥ , is invertible from (Tz Zε )⊥ into itself, and
the inverse operator satisfies (Pz Iε (z))−1  ≤ C.

Example 9 One can consider functionals of the form

Iε (u) = I0 (u) + ε G(u). (11)

where I0 ∈ C 2,α (H, R), is called the unperturbed functional and G ∈

C 2,α (H, R) is a perturbation. Suppose there exists a d-dimensional smooth
manifold Z such that all z ∈ Z is a critical point of I0 . The set Z will be
called a critical manifold (of I0 ). Let Tz Z denote the tangent space to Z
at z. Since Z is a critical manifold then for every z ∈ Z one has I0 (z) = 0.
Differentiating this identity, we get

(I0 (z)[v]|ϕ) = 0, ∀ v ∈ Tz Z, ∀ ϕ ∈ H,
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 71

and this shows that all v ∈ Tz Z is a solution of the linearized equation

I0 (z)[v] = 0, namely that v ∈ Ker[I0 (z)], so Tz Z ⊆ Ker[I0 (z)]. In par-
ticular, I0 (z) has a non trivial kernel (whose dimension is at least d) and
hence all the z ∈ Z are degenerate critical points of I0 . We shall require that
this degeneracy is minimal: precisely we will suppose that
(N D) Tz Z = Ker[I0 (z)], ∀ z ∈ Z.
In addition to (N D) we will assume that
(F r) for all z ∈ Z, I0 (z) is an index 0 Fredholm map.2
The last two properties imply that I0 (z) is invertible from (Tz Z)⊥ onto itself.
Therefore, if I0 and G are bounded (near Z) in C 2,α norm and if we take
Zε = Z, properties i) − iii) above will follow.

2.1 The Finite-Dimensional Reduction

First some notation is in order. Let us set W = (Tz Zε )⊥ and let (qi )1≤i≤d
be an orthonormal set (locally smooth) such that Tz Zε = span{q1 , . . . , qd }.
In the sequel we will always assume that Zε has a (local) C 2 parametric
representation (with uniformly bounded second derivatives) z = zξ , ξ ∈ Rd .
Furthermore, we also suppose that qi = ∂ξi zξ /∂ξi zξ . This will be verified
in our applications.
We look for critical points of Iε in the form u = z + w with z ∈ Zε
and w ∈ W . If P : H → W denotes the orthogonal projection onto W , the
equation Iε (z + w) = 0 is equivalent to the following system
 
P Iε (z + w) = 0 (the auxiliary equation);
(12)
(Id − P )Iε (z + w) = 0, (the bifurcation equation).

Proposition 10 Let i)–iii) hold. Then there exists ε0 > 0 with the following
property: for all |ε| < ε0 and for all z ∈ Zε , the auxiliary equation in (12)
has a unique solution w = wε (z) such that:
(j) wε (z) ∈ W = (Tz Zε )⊥ , is of class C 1 with respect to z ∈ Zε and
wε (z) → 0 as |ε| → 0, uniformly with respect to z ∈ Zε , together with
its derivative with respect to z, wε ;
(jj) quantitatively, one has that wε (z) = O(ε) as ε → 0, for all z ∈ Zc .
Proof. Property iii) allows us to apply the contraction mapping theorem
to the auxiliary equation. In fact, by the invertibility of P Iε (z) we can rewrite
it tautologically as

w = −(P Iε (z))−1 [P Iε (z) + (P Iε (z + w) − P Iε (z) − P Iε (z)[w])] := Gε,z (w).
2 A linear map T ∈ L(H, H) is Fredholm if the kernel is finite-dimensional and if the image
is closed and has finite codimension. The index of T is dim(Ker[T ]) − codim(Im[T ]).
72 A. Malchiodi

We claim next that the latter map is a contraction on a suitable subset of H.

In fact, for the second term of Gε we can write that
  1 
 
   
P Iε (z + w) − P Iε (z) − P Iε (z)[w] = P (Iε (z + sw) − Iε (z))[w]ds
 

0
≤ Cw1+α ,

and therefore by i) and iii) we have

Gz,ε (w) ≤ C 2 ε + C 2 w1+α ; w ≤ r0 .

Similarly, one also finds

Gz,ε (w1 ) − Gz,ε (w2 ) ≤ C 2 (w1 α + w2 α )w1 − w2 .

By the last two equations, if we fix C1 > 0 sufficiently large and let Bε =

{w ∈ W : w ≤ C1 ε}, we can check that Gz,ε is a contraction in Bε , so for
every z ∈ Zε we obtain a function w satisfying (jj) (for brevity, in the sequel
the dependence on z will be assumed understood).
Let us now show that also the derivatives of w with respect to ξ are
bounded. Indeed (here we are sketchy), for the components of w tangent to
Zε we can argue as follows: since (w|∂ξ z) = 0 for every ξ, differentiating with
respect to ξ we find that

(∂ξ w|∂ξ z) = −(w|∂ξ2 z).

Since w = O(ε) and since ∂ξ2 z is bounded (we are assuming

Zε has a C 2 -controlled parameterization), the tangent components of ∂ξ w
are bounded in norm by Cε.
About the normal components, we can differentiate the relation
(Iε (z + w)|∂ξ z) = 0 with respect to ξ, to find that

(Iε (z + w)[∂ξ z + ∂ξ w]|∂ξ z) + (Iε (z + w)|∂ξ2 z) = 0,

which implies

Iε (z + w)[∂ξ w] = −(Iε (z + w)|∂ξ2 z) − (Iε (z + w)[∂ξ z]|∂ξ z).

Since Iε is locally Hölder continuous and since w = O(ε), by iii) we have
that
P ∂ξ w ≤ CIε (z + w) + Iε (z + w)[∂ξ z] ≤ Cεα .
For the latter inequality we used again w = O(ε), together with the
Lipschitzianity of Iε and i).
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 73

2.2 Existence of Critical Points

We shall now provide conditions for solving the bifurcation equation in (12).
In order to do this, let us define the reduced functional Φε : Z → R by setting

Φε (z) = Iε (z + wε (z)). (13)

From a geometric point of view the argument can be outlined as follows.

Consider the manifold Z̃ε = {z + wε (z) : z ∈ Zε }. If zε is a critical point of
Φε , it follows that uε = zε + w(zε ) ∈ Z̃ε is a critical point of Iε constrained
on Z̃ε and thus uε satisfies Iε (uε ) ⊥ Tuε Z̃ε . Moreover the definition of wε ,
see Proposition 10, implies that Iε (z + wε (z)) ∈ Tz Zε . In particular, Iε (uε ) ∈
Tzε Zε . Since, for |ε| small, Tuε Z̃ε and Tzε Zε are close, see (j) in Proposition
10, it follows that Iε (uε ) = 0. A manifold with these properties is called a
natural constraint for Iε .

Theorem 11 Suppose we are in the situation of Proposition 10, and let us

assume that Φε has, for |ε| sufficiently small, a critical point zε . Then uε =
zε + wε (zε ) is a critical point of Iε .

Proof. We use the previous notation and, to be short, we write below Di

for Dξi , etc. Let ξε be such that zε = zξε , and set qiε = ∂z/∂ξi |ξε .
From Proposition 10 we infer that there exists ε0 > 0 such that the auxil-
iary equation in (12) has a solution wε (zε ), defined for |ε| < ε0 . In particular,
from (j) and by continuity, one has that

lim (Di wε (zε ) | qjε ) = 0, i, j = 1, . . . , d.

|ε|→0

Let us consider the matrix B ε = (bεij )ij , where

bεij = Di wε (zε ) | qjε .

From the above arguments we can choose 0 < ε1 < ε0 , such that

|det(B ε )| < 1, ∀ |ε| < ε1 , (14)

Fix ε > 0 such that |ε| < min{ε0 , ε1 }. Since zε is a critical point of Φε we get

(Iε (zε + wε (zε )) | qiε + Di wε (zε )) = 0, i = 1, . . . , d.


From P Iε (z + wε (zε )) = 0, we deduce that Iε (zε + wε (zε )) = Ai,ε qiε , where

Ai,ε = (Iε (zε + wε (zε )) | qiε ).

Then we find
74 A. Malchiodi
⎛ ⎞

⎝ Aj,ε qjε | qiε + Di wε (zε )⎠ = 0, i = 1, . . . , d,
j

namely


Ai,ε + Aj,ε qjε | Di wε (zε ) = Ai,ε + Aj,ε bεij = 0, i = 1, . . . , d. (15)
j j

Equation (15) is a (d × d) linear system for which the matrix of coefficients,

IdRd + B ε , has entries of the form δij + bεij , where δij is the Kronecker symbol
and bεij are defined above and satisfy (14). Then, for |ε| < ε1 , IdRd + B ε is
invertible, thus (15) has the trivial solution only: Ai,ε = 0 for all i = 1, . . . , d.
Since the Ai,ε are the components of Φε (zε ), the conclusion follows.

We next provide a criterion for applying Theorem 11, based on expanding Iε

on Zε in powers of ε. We will assume a slow dependence of G in the parameter
ξ, since this applies to our concrete case (P̃ε ).

Theorem 12 Suppose the assumptions of Proposition 10 hold, and that for

ε small there is a local parameterization ξ ∈ 1ε U ⊆ Rd of Zε such that, as
ε → 0, Iε admits the expansion
1
Iε (zξ ) = C0 + εG(εξ) + o(ε), ξ∈ U
ε
for some function G : U → R. Then we still have the expansion

Φε (zξ ) = C0 + εG(εξ) + o(ε) as ε → 0. (16)

Moreover, if ξ ∈ U is a strict local maximum or minimum of G, then for |ε|

small the functional Iε has a critical point uε . Furthermore, if ξ is isolated,
we can take uε − zξ/ε = o(1/ε) as ε → 0.

Remark 13 The last statement asserts that, once we scale back in ε, the
solution concentrates near ξ.
Proof. To prove (16) we simply use property i) and jj) in Proposition 10,
together with a Taylor expansion of Iε , to get

Iε (zξ + wε ) = Iε (zξ ) + Iε (z)[wε ] + O(wε 2 ) = C0 + εG(ξ) + o(ε),

so the first assertion follows.

We will prove the second one when ξ is a minimum of G: the other case is
completely similar. Let γ > 0 and let Uδ be a δ-neighborhood of ξ such that

G(ξ) ≥ G(ξ) + γ, ∀ ξ ∈ ∂Uδ .

Concentration of Solutions for Some Singularly Perturbed Neumann Problems 75

Using (16) we find, for |ε| small

Φε (zξ ) − Φε (zξ ) = ε G(ξ) − G(ξ) + o(ε).

Then, there exists ε1 > 0 small such that for every ξ ∈ ∂Uδ one has

Φε (zξ ) − Φε (zξ ) > 0 if 0 < ε < ε1 ,
Φε (zξ ) − Φε (zξ ) < 0 if −ε1 < ε < 0.

In the former situation Φε (zξ ) has a local minimum for ξ in Uδ , in the latter
a local maximum. In any case, Φε has a critical point zξε with ξε ∈ Uδ and
hence, by Theorem 11, uε = zε + wε (zε ) is a critical point of Iε . If ξ is an
isolated minimum or maximum of G we can take δ arbitrarily small and hence
zε → zξ as well as uε → zξ .

The last statement in Theorem 12 can be extended to the more general

situation in which ξ̂ is a critical point of G satisfying
(G ) ∃ N ⊂ Rd open bounded such that the topological degree d(G , N , 0)
is different from zero.
For the definition of topological degree see for example [4]. Let us point out
that if (G ) holds then G has a critical point in N . Moreover, if G has either
a strict local maximum (or minimum), or any non-degenerate critical point
ξ, we can take as N the ball Br (ξ) with r  1, and (G ) holds.

Theorem 14 Suppose we are under the (first) assumptions of Theorem 12,

that Iε (z) ≤ Cε2 for all z ∈ Zε (C fixed) and that we have the expansion

∂
Iε (zξ ) = ε2 G (εξ) + o(ε2 ), ξ ∈ U as ε → 0. (17)
∂ξ

Assume (G ) holds: then for |ε| small the functional Iε has a critical point
uε and there exists ξ̂ ∈ N , G (ξ)
ˆ = 0, such that uε → z for some εj → 0.
j ξ̂
Therefore if, in addition, N contains only an isolated critical point ξ̂ of G ,
then uε − zξ̂/ε = o(1/ε) as ε → 0.

Proof. The arguments of Proposition 10 still apply, but yield the improved
estimates w ≤ Cε2 and w  ≤ Cε2α (we just perform the contraction in
a smaller set). From the definition of Φε we infer that, for all v ∈ Tz Zε

(Φε (z) |v ) = (Iε (z + wε ) | v + wε ). (18)

Therefore, we can write that

(Φε (z) |v ) = Iε (z)[v] + Iε (z)[wε , v] + o(wε )v + Iε (z)[wε ] + O(wε  wε ).

76 A. Malchiodi

Now, using (17), the fact that wε  = O(ε2 ), wε  = O(ε2α )v and
Iε (z)[v] ≤ Cε2 v (see Proposition 10 and i), which has to be modified
with an extra ε) we have that

(Φε (z) |v ) = ε2 G (εξ)[v] + o(ε) in N . (19)

Then the continuity property of the topological degree and (G ) yield, for |ε|
small,
d(Φε , N , 0) = d(G , N , 0) = 0.
This implies that, for |ε| small, the equation Φε (z) = 0 has a solution in N .
The convergence result follows from (19).

3 Application to the Study of P̃ε

In this section we apply the above methods to the Neumann problem (P̃ε ).
As we mentioned, the limit profile of the solutions we are interested in solves
equation (1): the latter has variational structure, and solutions are critical
points of the functional
 
1
1
I(u) = |∇u| + u −
2 2
|u|p+1 . (20)
2 Rn p + 1 Rn

We call I + the functional corresponding to (1) in Rn+ , namely

 
1
1
I + (u) = |∇u|2 + u2 − |u|p+1 . (21)
2 Rn+ p + 1 Rn+

n+2
Recall that we are assuming here p < n−2 . Our next goal is to characterize
 
the spectrum and some eigenfunctions of I (U ) or of I + (U ).

 
3.1 Study of Ker[I (U )] and Ker[I + (U )]

For ξ ∈ Rn we set

Uξ = U (· − ξ); Z = {Uξ : ξ ∈ Rn } .

The first goal of this section is to prove the following result.

Lemma 15 Z is non-degenerate for I, namely the following properties are
true:

(N D) TUξ Z = Ker[I (Uξ )], ∀ ξ ∈ Rn ;

(F r) I (Uξ ) is an index 0 Fredholm map, for all ξ ∈ Rn .
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 77

Proof. It is sufficient to prove the lemma for ξ = 0, hence taking Uξ = U :

the case of a general ξ will follow immediately. The proof will be carried out
in several steps.

Step 1. In order to characterize Ker[I (U )], let us introduce some notation.
We set
x
r = |x|; ϑ= ∈ S n−1
|x|
and let Δr , resp. ΔS n−1 denote the Laplace operator in radial coordinates,
resp. the Laplace-Beltrami operator on S n−1
 
∂2 n−1 ∂ 1  ∂ √ ij ∂
Δr = 2 + ; ΔS n−1 = √ gg .
∂r r ∂r g ∂yj ∂yi

In the latter formula standard notation is used: ds2 = gij dy i dy j denotes

the standard metric on S n−1 , g = det(gij ) and [g ij ] = [gij ]−1 . Consider the
spherical harmonics Yk (ϑ) satisfying

−ΔS n−1 Yk = λk Yk , (22)

and recall that this equation has a sequence of eigenvalues

λk = k(k + n − 2), k = 0, 1, 2, . . .

whose multiplicity is given by Nk − Nk−2 where

(n + k − 1)!
Nk = , (k ≥ 0); Nk = 0, ∀ k < 0,
(n − 1)! k!

see [16]. In particular, one has that

λ0 = 0 has multiplicity 1 and λ1 = n − 1 has multiplicity n.

Every v ∈ H can be written in the form

∞

v(x) = ψk (r)Yk (ϑ), where ψk (r) = S n−1 u(rϑ)Yk (ϑ)dϑ ∈ H 1 (R),
k=0

and moreover
1
Δ(ψk Yk ) = Yk (ϑ)Δr ψk (r) + ψk (r)ΔS n−1 Yk (ϑ). (23)
r2

Recall that v ∈ H belongs to Ker[I (U )] if and only if

−Δv + v = pU p−1 (x)v, v ∈ H. (24)

78 A. Malchiodi

Substituting (23) and (22) into (24) we get the following equations for ψk
 n−1  λk
Ak (ψk ) := −ψk − ψk + ψk + 2 ψk − pU p−1 ψk = 0, k = 0, 1, 2, . . . .
r r
Step 2. Let us first consider the case k = 0. Since λ0 = 0 we infer that ψ0
satisfies
 n−1 
A0 (ψ0 ) = −ψ0 − ψ0 + ψ0 − pU p−1 ψ0 = 0.
r
It has been shown in [41] that all the non trivial solutions to A0 (u) = 0 are
unbounded. Since we are looking for solutions ψ0 ∈ H 1 (R), it follows that
ψ0 = 0.

Step 3. For k = 1 one has that λ1 = n − 1 and we find

 n−1  n−1
A1 (ψ1 ) = −ψ1 − ψ1 + ψ1 + ψ1 − pU p−1 ψ1 = 0.
r r2
 (r) denote the function such that U (x) = U
Let U  (|x|). Since U (x) satisfies

−ΔU + U = U , then U solves
p

  − n − 1    p.
−U U +U =U
r
Differentiating in r, we get

  ) − n − 1   n − 1    p−1 U
 .
−(U (U ) + U + U = pU (25)
r r2

In other words, U  (r) satisfies A1 (U

  ) = 0, and U
  ∈ H 1 (R). Let us look for
a second solution of A1 (ψ1 ) = 0 in the form ψ1 (r) = c(r)U   (r). By a straight
calculation, we find that c(r) solves

−c U   ) − n − 1 c U
  − 2c (U   = 0.
r
If c(r) is not constant, it follows that

c  
U n−1
− = 2 + .
c  
U  r
This yields
1
c(r) ∼ , (r → +∞),
 2
rn−1 U
which implies c(r) → +∞ as r → +∞. Therefore, the family of solutions of
  (r), for some c ∈ R.
A1 (ψ1 ) = 0, with ψ1 ∈ H 1 (R), is given by ψ1 (r) = cU

Step 4. Let us show that the equation Ak (ψk ) = 0 has only the trivial solu-
tion in H 1 (R), provided that k ≥ 2. Actually, the equation A1 (u) = 0 has
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 79

  which does not change sign in (0, ∞). Thus, by standard

the solution U
arguments, A1 is a non-negative operator. From

λk = (n + k − 2)k = λ1 + δk ; δk = k(n + k − 2) − (n − 1),

we infer that
δk
Ak = A1 + .
r2
Since δk > 0 whenever k ≥ 2, it follows that Ak is a positive operator for any
k ≥ 2. Thus Ak (ψk ) = 0 implies that ψk = 0.

Conclusion. Collecting all the previous information, we deduce that any v ∈


Ker[I (U )] has to be a constant multiple of U  (r)Y1 (ϑ). Here Y1 satisfies

−ΔS n−1 Y1 = λ1 Y1 = (n − 1)Y1 ,

namely it belongs to the kernel of the operator −ΔS n−1 − λ1 Id. Recalling
that such a kernel is n dimensional, and letting Y1,1 , . . . , Y1,n denote a basis
of it, we finally find that
  Y1,i : 1 ≤ i ≤ n} = span{Uxi : 1 ≤ i ≤ n} = TU Z.
v ∈ span{U

This proves that (N D) holds. It is also easy to check that the operator I (U )
is a compact perturbation of the Identity, showing that (F r) holds too. This
completes the proof of Lemma 15.

Remark 16 U is a mountain pass solution of (1), so the spectrum of I (U )
has exactly one negative simple eigenvalue, p − 1, with eigenspace spanned
by U itself, and denoted by U . Moreover, we have shown in the preceding
lemma that λ = 0 is an eigenvalue with multiplicity n and eigenspace spanned
by Di U , i = 1, 2, . . . , n. Furthermore, there exists κ > 0 such that

(I (U )v|v) ≥ κv2 , ∀ v ⊥ U ⊕ TU Z, (26)

and hence the rest of the spectrum is positive.

Corollary 17 Let U be as above and consider the functional I + given in

(21). Then for every ξ ∈ Rn−1 , U (·−(ξ, 0)) is a critical point of I + . Moreover,
 ∂U
the kernel of I + (U ) is generated by ∂x 1
, . . . , ∂x∂U
n−1
. The operator has only one
negative eigenvalue, and therefore there exists δ > 0 such that

 ∂U ∂U
I + (U )[v, v] ≥ δv2 for all v ∈ H 1 (Rn+ ), v ⊥+ U, ,..., ,
∂x1 ∂xn−1

where we have used the symbol ⊥+ to denote orthogonality with respect to the
H 1 scalar product in R.+ .
80 A. Malchiodi

Proof. Given any v ∈ H 1 (Rn+ ), we define the function v ∈ H 1 (Rn ) by an

even extension across ∂Rn+ , namely we set

 v(x , xn ), for xn > 0;
v(x , xn ) =
v(x , −xn ) for xn < 0.

We prove first the following claim.


Claim. Suppose v ∈ H 1 (Rn+ ) is an eigenfunction of I + (U ) with eigenvalue λ.

Then the function v is an eigenfunction of I (U ) with eigenvalue λ.

In order to prove the claim, we notice that the function v satisfies the equation

−Δv + v − pU p−1 v = λ(−Δv + v), in Rn+ ;
∂v
∂ν = 0, on ∂Rn+ .

Similarly, by symmetry, there holds


−Δv + v − pU p−1 v = λ(−Δv + v), in Rn− ;
∂v
∂ν = 0, on ∂Rn− ,

where we have set Rn− = {(x , xn ) : x ∈ Rn−1 , xn < 0}. Then, considering
any function w ∈ H 1 (Rn ), integrating by parts and using the Neumann
boundary condition one finds
 
 1
I (U )[v, w] = (∇v, ∇w + vw) − U p−1 vw
Rn p + 1 Rn
 
=λ (−Δv + v)w + λ (−Δv + v)
Rn
+ Rn
−
 
=λ (∇v, ∇w + vw) + λ (∇v, ∇w + vw)
Rn
+ Rn
−

= λ(v|w)H 1 (Rn ) .

This proves the above claim.


We know that the functions ∂ξ1 U, . . . , ∂ξn−1 U belong to the kernel of I + (U ).
Suppose by contradiction that there exists a non zero element v in the kernel

of I + (U ), orthogonal to ∂ξ1 U, . . . , ∂ξn−1 U . Then, by the above claim, its even

extension v would belong to the kernel of I (U ). On the other hand we

know that the only element in the kernel of I (U ) which is orthogonal to
∂ξ1 U, . . . , ∂ξn−1 U is ∂ξn U . Since ∂ξn U is odd with respect to xn , while v is
even with respect to xn , we get a contradiction. This concludes the proof.
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 81

3.2 Proof of Theorem 1

Let us describe ∂Ωε near a generic point X ∈ ∂Ωε . Without loss of generality,
we can assume that X = 0 ∈ Rn , that {xn = 0} is the tangent plane of ∂Ωε
(or ∂Ω) at X, and that ν(X) = (0, . . . , 0, −1). In a neighborhood of X, let
xn = ψ(x ) be a local parametrization of ∂Ω. Then one has

xn = ψ(x ) := 12 AX x , x + CX (x ) + O(|x |4 ); |x | < μ0 , (27)

where AX is the hessian of ψ at 0 and CX is a cubic polynomial, which is

given precisely by 
CX (x ) = 16 3
∂ijk ψ|0 xi xj xk . (28)
i,j,k
1
We have clearly H(X) = n−1 trAX .
On the other hand, ∂Ωε is parameterized
 1 
by yn = ψε (x ) := ε ψ(εx ), for which the following expansion holds
ε
ψε (x ) = AX x , x + ε2 CX (x ) + ε3 O(|x |4 );
2
∂i ψε (x ) = ε(AX x )i + ε2 QiX (x ) + ε3 O(|x |3 ), (29)

where QiX are quadratic forms in x given by (see (28))

1 3
QiX (x ) = ∂ijk ψ|0 xj xk .
2
j,k

In particular, from the Schwartz’s Lemma, it follows that

(QiX )jk = (QiX )kj = (QjX )ik for every i, j, k. (30)

Concerning the outer normal ν, we have also

 
∂ψε
, . . . , ∂ψε
, −1  
∂x1 ∂xn−1   1 2  2
ν=  = ε(AX x ) + ε QX (x ), −1 + ε |Ax |
2
1 + |∇ψε |2 2
+ ε3 O(|x |3 ). (31)

Since ∂Ωε is almost flat for ε small and since the function U is radial, for
X ∈ ∂Ωε we have ∂ν ∂
U (· − X) ∼ 0. Thus U (· − X) is an approximate solution
to (Pε ). Hence, a natural choice of the manifold Zε could be the following

Zε = {U (· − X) := UX : X ∈ ∂Ωε } .

We show next that the abstract setting of the previous section can be applied
with this choice of Zε and for Iε given by (6).
82 A. Malchiodi

Proposition 18 With the above choices of Iε and Zε , the assumptions i) −

iii) in Section 2 hold true.

In order to prove this proposition, we need the following preliminary result.

Lemma 19 There exists δ > 0 such that for ε small one has
∂UX
Iε (UX )[v, v] ≥ δv2 for every v ⊥ UX , .
∂X
Proof. Let R  1 and consider a radial smooth function χR : Rn → R
such that ⎧
⎨ χR (x) = 1, in BR (0);
χR (x) = 0 in Rn \B2R (0); (32)
⎩
|∇χR | ≤ R2
in B2R (0)\BR (0),
and we set

v1 (x) = χR (x − X)v(x); v2 = (1 − χR )(x − X)v(x).

A straight computation yields


v = v1  + v2  + 2
2 2 2
[∇v1 , ∇v2 + v1 v2 ] .
Ωε

We write Ωε
[∇v1 , ∇v2 + v1 v2 ] = τ1 + τ2 , where

τ1 = χR (1 − χR )(v 2 + |∇v|2 );
Ωε

τ2 = v2 ∇v, ∇χR − v1 ∇v, ∇χR − v 2 |∇χR |2 .
Ωε

Since the integrand in τ2 is supported in {R ≤ |x| ≤ 2R}, using (32) and the
Hölder’s inequality we deduce that |τ2 | = oR (1)v2 . As a consequence we
have
v2 = v1 2 + v2 2 + 2τ1 + oR (1)v2 . (33)
After these preliminaries, let us evaluate Iε (UX )[v, v] = σ1 + σ2 + σ3 , where

σ1 = Iε (UX )[v1 , v1 ]; σ2 = Iε (UX )[v2 , v2 ]; σ3 = 2Iε (UX )[v1 , v2 ].

Since UX decays exponentially away from X, we get immediately

σ2 ≥ C −1 v2 2 + oR (1)v2 ; σ3 ≥ C −1 τ1 + oR (1)v2 , (34)

hence it is sufficient to estimate the term σ1 . By the exponential decay of UX

and the fact that (v|UX ) = (v| ∂U
∂X ) = 0 one easily finds
X
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 83

(v1 |UX ) = −(v2 |UX ) = oR (1)v;

     
∂UX ∂UX  ∂UX 
v1 | = − v2 | 
= oR (1)v  . (35)
∂X ∂X ∂X 

Therefore, using Corollary 17 we obtain


I + (U )[v1 , v1 ] ≥ δv1 2+ + oε,R (1)v2

(to be rigorous, the function v1 needs to be modified for being defined in Rn+ .
An easy way to do it is to stretch the boundary of the domain near X into a
hyperplane, but since ∂Ωε is almost flat this requires a small modification of
the H 1 norm of v1 ). Using this reasoning then one finds

σ1 ≥ I + (U )[v1 , v1 ] + oε (1)v1 2 ≥ δv1 2+ + oε,R (1)v2
≥ δv1 2 + oε,R (1)v2 . (36)

In conclusion, from (34) and (36) we deduce

δ
Iε (UX )[v, v] ≥ δv21 + v2 2 + Iv + oε,R (1)v2 ≥ v2 ,
2
provided R is taken large and ε is sufficiently small. This concludes the
proof.

Proof of Proposition 18. By the previous lemma, we need to prove only i)

and ii). To show i), we consider an arbitrary function v ∈ H 1 (Ωε ). Integrating
by parts and using (1) we find that
  
∂UX ∂UX
Iε (UX )[v] = vdσ+ (−Δg UX + UX − UX p
) vdy = vdσ.
∂Ωε ∂ν Ωε ∂Ωε ∂ν

For μ0 small and positive, we divide next ∂Ωε into its intersection with
B με0 (X) and its complement. In the latter set we use the exponential decay
of UX and its derivatives, plus the trace embedding for v to obtain
 
 
 ∂UX 
  ≤ Ce− ε0 v.
μ

 vdσ  (37)
 ∂Ωε \B μ0 (X) ∂ν 
ε

In B με0 (X) we use instead the coordinates x defined above, and the function
ψε to parameterize ∂Ωε . From (29), (31) and the fact that ∇U (x) = U  |x|x

one finds that

84 A. Malchiodi
+ ,
∂UX  (x , ε/2x AX x + O(ε2 )|x |2 )    2

= U (x) 1 , (εAX x , −1) + O(ε )|x |
2
∂ν (|x |2 + O(ε2 )|x |4 ) 2
 
1 x , AX x   2 −|x | 
= ε 
U (|x |) + O(ε e ) = O(ε|x|e−|x | ). (38)
2 |x |

This estimate, jointly with the trace embedding, yields

 
 
 ∂UX 
  ≤ Ce− ε0 v.
μ

 vdσ 
 ∂Ωε ∩B μ0 (X) ∂ν 
ε

This equation and (37) imply that Iε (UX ) ≤ Cε. Similar estimates imply
Iε (z)[q] ≤ Cεq for every z ∈ Zε and every q ∈ Tz Zε .
Property ii) is rather standard. In fact, given two functions u1 , u2 we have
that


Iε (u1 )[v, w] − Iε (u2 )[v, w] = −p |u1 |p−1 − |u2 |p−1 vw.
Ωε

Using the Hölder inequality and the Sobolev embeddings we get

  p−1
  p+1
p+1

|Iε (u1 )[v, w] − Iε (u2 )[v, w]| ≤ C |u1 |p−1 − |u2 |p−1  p−1 v w.
Oε

From the elementary inequality

  p+1

|u1 |p−1 − |u2 |p−1  p−1 ≤ C |u1 − u2 | + |u1 − u2 |p−1

p+1
≤ C|u1 − u2 | p−1 + C|u1 − u2 |p+1

and again the Sobolev embeddings one finds that

Iε (u1 ) − Iε (u2 ) ≤ C u1 − u2  + u1 − u2 p−1 ,

therefore when the ui s are bounded (e.g. in an r0 -strip around Zε ) we obtain
local Hölderianity of Iε .

We now show that also Theorem 12 applies to this case.

Lemma 20 For ε small the following expansion holds

Iε (UX ) = C0 − C1 εH(εX) + O(ε2 ),

where
 
1 1
C0 = − U p+1 ; C1 = C1 (n, p) > 0.
2 p+1 Rn
+
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 85

Proof. Integrating by parts and using (1) one finds

  
1 ∂UX 1 1
Iε (UX ) = UX + − |UX |p+1 .
2 ∂Ωε ∂ν 2 p+1 ∂Ωε

Employing (38) we also deduce

  ∞
1 ∂UX 1
UX = εσn−2 H(εX) U (r)U  (r)rn−1 dr + o(ε),
2 ∂Ωε ∂ν 4 0

where σn−2 is the volume of S n−2 .

On the other hand, from (29) we find that
  
1
|UX |p+1
= |U |p+1
− εσn−2 U p+1 x , AX x dx + o(ε)
Ωε Rn
+
2 Rn−1
  ∞
1
= |U |p+1 − εσn−2 H(εX) U p+1 (r)rn dr + o(ε).
Rn
+
2 0

In conclusion we obtain the expansion

Iε (UX ) =
  ∞   ∞ 
1 1 1 1
C0 +εσn−2 H(εX) U (r)U  (r)rn−1 dr− − U p+1 (r)rn dr .
4 0 2 2 p+1 0

Since U  (r) < 0, the coefficient of εH(εX) is negative, and we obtain the
conclusion.

Proof of Theorem 1 It is sufficient to apply Theorem 12, Proposition 18

and Lemma 20.

3.3 Proof of Theorem 2

Our goal is to apply Theorem 14, and therefore we need to find a more
accurate approximate solution. We first prove the following technical lemma.

Lemma 21 Let T = (aij ) be an (n − 1) × (n − 1) symmetric matrix, and

consider the following problem

Lw = −2T x , ∇x ∂xn U − trT ∂xn U in Rn+ ;
 (39)
∂xn w = T x , ∇x U
∂  on ∂Rn+ ,
86 A. Malchiodi

where L is the operator

Lu = −Δu + u − pU p−1 u.

Then (39) admits a solution wT , which is even in the variables x and satisfies
the following decay estimates

|w T (x)| + |∇w T (x)| + |∇2 w T (x)| ≤ C|T |∞ (1 + |x|C )e−|x| , (40)

where C is a constant depending only on n and p, and |T |∞ := maxij |aij |.

Proof. Problem (39) can be reformulated as


I + (U )[w] = vT , (41)

where vT is an element of H 1 (Rn+ ) defined by duality as

 
(vT |v)H 1 (Rn+ ) = (−2T x , ∇x ∂xn U − trT ∂xn U ) v − T x , ∇x U v.
Rn
+ ∂Rn
+

By Proposition 17, equation (41) is solvable if and only if vT is orthogonal

∂U
to ∂x 1
, . . . , ∂x∂U
n−1
, which is the case since
  
∂U ∂U
vT , =− (2T x , ∇x ∂xn U + trT ∂xn U )
∂xi H 1 (Rn Rn ∂xi
+) +

∂U
+ T x , ∇x U , i = i, . . . , n − 1.
∂Rn
+
∂xi

∂U
Indeed, all the integrals in the last formula vanish because ∂x i
is odd in x
and the other functions are even, by the symmetry of T . The decay in (40)
follows from (10) and standard elliptic estimates.
Given μ0 as in (27), we introduce a new set of coordinates on B με0 (X) ∩ Ωε .
Let
y  = x ; yn = xn − ψε (x ). (42)
The advantage of these coordinates is that ∂Ωε identifies with {yn = 0}, but
the corresponding metric coefficients gij will not be constant anymore: indeed
we have
⎛ ∂ψε ⎞
∂y1
  ⎜ .. ⎟
∂x ∂x ⎜ δij + ∂ψε ∂ψε
. ⎟
(gij ) =  , =⎜
⎜
∂yi ∂yj ⎟.
⎟
∂yi ∂yi ⎝ ∂ψε
⎠
∂yn−1
∂ψε
∂y1 ··· ∂ψε
∂yn−1 1
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 87

From the estimates in (29) it follows that

gij = Id + εA + ε2 B + O(ε3 |y  |3 ), (43)

and
∂yk (gij ) = ε∂yk A + ε2 ∂yk B + O(ε3 |y  |2 ),
where
   
0 AX y  AX y  ⊗ AX y  QX (y  ) 3
A=  t ; B=
(AX y ) 0 (QX (y  ))t 0

It is also easy to check that the inverse matrix (g ij ) is of the form g ij =

Id − εA + ε2 C + O(ε3 |y  |3 ), where
 
0 −QX (y  )
C= ,
−(QX (y  ))t |AX y  |2

and
∂yk (g ij ) = −ε∂yk A + ε2 ∂yk C + O(ε3 |y  |2 ).
Furthermore, since the transformation (42) preserves the volume, one has

det g ≡ 1.

We also recall that the Laplace operator in a general system of coordinates

is given by the expression
1   
Δg u = √ ∂j g ij det g ∂i u + g ij ∂ij
2
u,
det g
so in our situation we get

Δg u = g ij uij + ∂i (g ij )∂j u.

In particular, by (43), for any smooth function u we find

Δg u = Δu − ε (2AX y  , ∇y ∂yn u + trAX ∂yn u)

+ε2 −2QX , ∇y ∂yn u + |AX y  |2 ∂y2n yn u − divQX ∂yn u (44)

 2  3
+O(ε |y | )|∇u| + O(ε |y | )|∇ u|.
3 3 2

Here AX is the Hessian of ψ at x = 0, see the above notation. Now we choose

a cut-off function ψμ0 with the properties
⎧
⎨ ψμ0 (x) = 1 in B μ40 ;
ψμ (x) = 0 in Rn+ \B μ20 ;
⎩ 0
|∇ψμ0 | + |∇ ψμ0 | ≤ C in B μ20 \B μ40 ,
2

3 If the vector v has components (vi )i , the notation v ⊗ v denotes the square matrix with
entries (vi vj )ij .
88 A. Malchiodi

and for any X ∈ ∂Ω we define the following function, in the coordinates

(y  , yn )
zε,X (y) = ψμ0 (εy)(U (y) + εwAX (y)). (45)
where w AX is given by Lemma 21 with T = AX . We also give the expression of
the unit outer normal to ∂Ωε , ν̃, in the new coordinates
n y. Letting
n νi (resp. ν̃i )
∂ i ∂
be the components of ν (resp. ν̃), from ν = i=1 ν i ∂x i = i=1 ν̃ ∂y i , we
n i ∂y
k
have ν̃k = i=1 ν ∂xi . This implies


n−1
∂ψε
ν̃ k = ν k , k = 1, . . . , n − 1; ν̃ n = νi + νn.
i=1
∂y i

From (29) and the subsequent formulas we find

ν̃ = εAX (y  ) + ε2 QX (y  ), −1 + 32 ε2 |AX (y  )|2 + ε3 O(|y  |3 ). (46)

Finally the area-element of ∂Ωε can be expanded as

dσ = (1 + O(ε2 |y  |2 ))dy  . (47)

Next, we estimate the gradient of Iε at zε,X showing that the zε,X ’s consti-
tute, as X varies on ∂Ωε , a manifold of pseudo-critical points of Iε with a
better accuracy than UX .

Lemma 22 There exists C > 0 such that for ε small

Iε (zε,X ) ≤ Cε2 ; for all X ∈ ∂Ωε .

Proof. Let v ∈ H 1 (Ωε ). Since the function zε,X is supported in B μ2ε0 (X),
see (45), we can use the coordinates y in this set, and we obtain
   
∂zε,X
Iε (zε,X )[v] = vdσ + p
−Δg zε,X + zε,X − zε,X vdy. (48)
∂Ωε ∂ ν̃ Ωε

∂zε,X
Let us now evaluate ∂ ν̃ : one has

∂zε,X
= (U + εw AX )∇ψμ0 (εy), ν̃ + ψμ0 (εy)∇(U + εwAX ), ν̃ .
∂ ν̃
Since ∇ψμ0 (ε ·) is supported in Rn \ B μ4ε0 , and both U, wAX decay exponen-
tially to zero, we have
1
|(U + εw AX )∇ψμ0 (εy), ν̃ | ≤ C(1 + |y|C )e− Cε e−|y| .

On the other hand, from the boundary condition in (39) and from (46), the
terms of order ε in the scalar product ψμ0 (εy)∇(U + εwAX ), ν̃ cancel and
we obtain
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 89

∂zε,X μ0
= O(ε2 |y  ||∇w|) + O(ε2 |y  |2 |∇U |); |y| ≤ ;
∂ ν̃ 4ε
 
 ∂zε,X  μ0 μ0
  −|y| 1
+ Cε(1 + |y|C )e−|y| ≤ Cε−C e− Cε ;
 ∂ ν̃  ≤ Ce 4ε
≤ |y| ≤
2ε
.

The last two estimates, (47), and the trace Sobolev inequalities readily imply
 
 ∂zε,X 
 vdσ  ≤ Cε2 v. (49)

∂Ωε ∂ ν̃

On the other hand, using (40), (44) and the decay of U , the volume integrand
can be estimated as
 

 p 
−Δg zε,X + zε,X − zε,X  ≤ Cε2 |y  ||∇U | + |y  |2 |∇2 U | + |∇w| + |y  ||∇2 w|
 
+ |U + εw|p−1 (U + εw) − U p − pεU p−1 w ,
 
for |y| ≤ 4εC sup1 A  C1 , and
X X

 
 p  
−Δg zε,X + zε,X − zε,X  ≤ C(1 + |y  |C )e−|y |
1
≤ Cε−C e− Cε ,
 
for 4εC sup1 A  C1 ≤ |y| ≤ μ2ε0 . We notice that the following inequality
X X
holds true
  a
(a + b)p − ap − pap−1 b ≤ Cb2 ; a > 0, |b| ≤ .
2
In particular, by (40) we have
 
U (y) 1 1
ε|w(y)| ≤ ; for |y| ≤ ,
2 4εC supX AX  C

hence it follows that

 
 p 
−Δg zε,X + zε,X − zε,X  ≤ Cε2 (1 + |y|C )e−|y| ;
 
1 1
|y| ≤ .
4εC supX AX  C
Then, using the Hölder inequality we easily find
   
 
 −Δ z + z − z p
vdy  ≤ Cε2 v. (50)
 g ε,X ε,X ε,X 
Ωε

From (49) and (50) we obtain the conclusion.

90 A. Malchiodi

ε,X ∂z
We also need to compute ∂X in the coordinates y introduced in (42). We
notice that in the definition of zε,X , see (45), not only the analytic expression
of this function depends on X, but also the choice of the coordinates y. There-
fore, when we differentiate in X, we have to take also this dependence into
account. First we derive the variation in X of the coordinates x (introduced
before (27)) of a given point in Ω. Using the dot to denote the differentiation
with respect to X, one can prove that
∂  ∂
ẋ = x = −Ẋ; ẋn = (xn )X = −x , Hε Ẋ , (51)
∂X X ∂X
where Hε = εAX is the second fundamental form of Ωε . The second equation
in (51) is obtained by computing the variation of the distance of a fixed point
in Rn from a moving tangent plane to Ωε . Similarly, we get a dependence on
X of the coordinates y. To emphasize the dependence of zε,X on X we write

zε,X = U (yX ) + εwAX (yX ); yX = (xX , (xn )X − ψε (xX )). (52)

Since the set Ωε is a dilation of Ω, the derivatives of AX and ψε with respect

to X are of order ε (if Ẋ is of order 1). More precisely, if we set X̃ = εX,
then we have
∂AX ∂A ∂ψε ∂ψ
= ε X̃ ; =ε ,
∂X ∂ X̃ ∂X ∂ X̃
where ψ is given in (27). Differentiating (52) with respect to X and using
(51) it follows that, in the coordinates y

żε,X = −Ẋ, ∇y U + O(ε) in H 1 (Rn+ ). (53)

In this spirit, we also compute the variation of the matrix AX , see (27), with
respect to X. Differentiating the equation xn = ψε (x ) with respect to X and
using (51) we find

1 2 ∂AX̃    n−1
−x , Hε Ẋ = ε  x , x − εAX x , Ẋ − ε2 Qi Ẋi .
2 ∂ X̃ i=1

If e1 , . . . , en−1 is an orthonormal system of tangent vectors to ∂Ω with ei =

∂ X̃
∂xi , the last equation implies
 
∂A ∂AX̃
 X̃ x , x = 2QiX (x ), namely (QiX )jk = . (54)
∂ei ∂ei jk

By the symmetries in (30), we have in particular

   
∂AX̃ ∂AX̃
= for every i, j. (55)
∂ej ij ∂ei jj
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 91

The proof of Proposition 18 can be easily modified if one takes Zε =

{zε,X : X ∈ ∂Ωε }: our next goal is to expand Iε (zε,X ) up to order ε2 so
that, taking Lemma 22 into account, we can apply Theorem 14.
Lemma 23 For ε small the following expansion holds

Iε (zε,X ) = C0 − C̃1 εH(εX) + O(ε2 ),

where
   ∞ 
1 1
C0 = − U p+1 , C̃1 = rn Ur2 dr yn |y  |2 dσ.
2 p+1 Rn
+ 0 n
S+

Proof. To be short, we will often write z instead of zε,X and w instead of

w(ε, X). Since z is supported in B μ2ε0 (X), we can use the coordinates y and
integrate by parts to get
  
∂z
Iε (z) = 1
2 z + 1
2 z (−Δg z + z) − 1
p+1 |z|p+1 .
n
∂R+ ∂ ν̃ Rn
+ Rn
+

Using the definition of z given in (45), as well as the expression of the Δg in

(44) we find
   
1
2 z (−Δ g z + z) − 1
p+1 |z|p+1
= 1
2 − 1
p+1 U p+1
Rn
+ Rn
+ Rn
+

 
 ε
+ε U AX y , ∇y ∂yn U + trAX U ∂yn U + O(ε2 ).
R+
n 2 Rn
+

Moreover, from (46) we get

 
∂z
1
2 z = ε
2 U AX y  , ∇y U + O(ε2 ).
∂Rn
+
∂ ν̃ ∂Rn
+

Putting together the preceding formulas we have

  
1 1 ε
Iε (z) = − U p+1 + U AX y  , ∇y U
2 p+1 R+
n 2 n
∂R+

 
ε
+ε U AX y  , ∇y ∂yn U + trAX U ∂yn U + O(ε2 ).
Rn
+
2 Rn
+

Integrating by parts (more than once if needed), we find that the three terms
of order ε are given by
92 A. Malchiodi
  
1 1
AX y  , ∇y U 2 + U AX y  , ∇y ∂yn U + trAX ∂yn U 2
4 ∂Rn R n 4 R+
n
+
  + 
1 
= − trAX U −
2
U AX y , ∇y U − ∂yn U AX y  , ∇y U
2 n
∂R+ n
∂R+ R+
n


=− ∂yn U AX y , ∇y U .
Rn
+

Now we notice that, since U is radial, one has

yn y
∂yn U = Ur ; ∇y U = Ur ,
|y| |y|
and hence  
 yn AX (y  ), y  2
∂yn U AX (y ), ∇ U = − y Ur dy.
Rn
+ Rn
+
|y|2
Now it is sufficient to express the last integral in radial coordinates. This
concludes the proof.

Lemma 24 For ε small the following expansion holds

∂
Iε (zε,X ) = −C̃1 ε2 H  (εX) + o(ε2 ),
∂X
where C̃1 is the constant given in the previous Lemma.
Proof. We have
 
∂
Iε (z)[∂X z] = (−Δg z + z − |z| ) ∂X z +
p
∂X z zdσ.
Rn
+ ∂Rn
+
∂ ν̃

Notice that, by our construction, the terms −Δg z + z − |z|p and ∂∂ν̃ z are of
order ε2 , hence it is sufficient to take the product only with the 0-th order
term of ∂X z, see (53). In this way we obtain that Iε (z)[∂X z] = (α1 + α2 )ε2 +
o(ε2 ), where


α1 = 2Q, ∇y ∂yn U − |Ap y  |2 ∂y2n yn U + divQ∂yn U + 2AX y  , ∇y ∂yn w
R+
n

1
+ trAX ∂yn w − p(p − 1)U p−2 w2 ∂X U ;
2
 
α2 = Q, ∇y U ∂X U +
 AX y  , ∇y w ∂X U.
∂Rn
+ ∂Rn
+

Since the function w is even in y  , all the terms containing it vanish identically
and so does the term |Ap y  |2 ∂y2n yn U ∂X U , hence we get

α1 = [2Q, ∇y ∂yn U + divQ∂yn U ] ∂X U.
Rn
+
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 93

On the other hand, the boundary integral α2 is given by


α2 = Q, ∇y U ∂X U,
∂Rn
+

again by the oddness of w. In conclusion we have

 
α1 + α2 = [2Q, ∇y ∂yn U + divQ∂yn U ] ∂X U + Q, ∇y U ∂X U,
Rn
+ ∂Rn
+

which we rewrite as
  
2 Qj (x )∂j ∂yn U ∂i U + ∂j Qj (x )∂yn U ∂i U + Qj (x )∂j U ∂i U.
j Rn
+ j Rn
+ j ∂Rn
+

If we integrate by parts in the variable yj we find

 

α1 + α2 = Qj (x )∂j ∂yn U ∂i U − Qj (x )∂yn U ∂j ∂i U
R
j  +
n R n
+
 j
+ Qj (x )∂j U ∂i U.
j ∂Rn
+

Then, if we integrate by parts in the variable yn and in yi we obtain

     ∂AX  
∂yi QjX (y  ) ∂yj U ∂yn U = yj ∂yj U ∂yn U.
j Rn
+ j
∂ej ij Rn+

By the symmetry in (55) and using radial variables we finally get

  ∂AX   ∂H
α1 + α2 = yj ∂yj U ∂yn U = C1 ,
j
∂ei jj Rn+ ∂ei

which concludes the proof (recall that ∂i U = −∂Xi z).

Proof of Theorem 2. We define the new manifold of approximate solutions

Zε = {zε,X : X ∈ ∂Ωε } .

As we already noticed, since the difference between zε,X and UX is of order

O(ε), one has the counterpart of Proposition 18 for this new choice of Zε .
Then, to prove the theorem is sufficient to apply Lemmas 22, 23, 24 and
Theorem 14. In fact, using a Taylor expansion for H, one can find a small
positive number δ0 such that
det H  (X0 )
H  = 0 on ∂Bδ0 (X0 ) and deg(H  , Bδ0 (X0 ), 0) = (−1)sgn .
(56)
This implies the conclusion.
94 A. Malchiodi

Part II
Higher-Dimensional Concentration
The solutions in Theorem 3 are boundary-layers and scale qualitatively in
the following way, as ε tends to zero

uε (x , xn ) ∼ w0 (xn /ε); x ∈ Rn−1 , xn ∈ R+ ,

where now w0 is the solution of the problem (4).

In the case of spike-layers, both the energy and the Morse index stay
bounded as ε goes to zero. Viceversa, by the results in [20], if the Morse
index of a family of solutions stays bounded as ε → 0, these solutions must
concentrate at a finite number of points (at least in low dimensions).
The proof of Theorem 3 relies on a local inversion argument, via a con-
traction mapping. The main difficulty is that, since the Morse index of the
solutions is changing with ε, the linearized operator Iε (uε ) will not be invert-
ible for all the values of ε in any interval of the form (0, ε0 ). We get indeed
invertibility at least along a sequence εj , with the norm of the inverse operator
blowing-up at the rate εn−1j , see Proposition 36. This produces a resonance
phenomenon which can be described as follows.
Let γ ∈ (0, 1), and let ν denote the inner unit normal to ∂Ωε . Consider
the neighborhood Σε of ∂Ωε defined as
 
Σε = x + xn ν : x ∈ ∂Ωε , xn ∈ (0, ε−γ ) .

Let (ϕl )l denote the eigenfunctions of −Δĝ , where Δĝ is the Laplace-Beltrami
operator on ∂Ω, and let (λl )l be the corresponding (non-negative) eigenvalues.
Let also zσ be an eigenfunction of Iε (uε ) with eigenvalue σ. Then we can
decompose zσ in Fourier series (in the variable x ) in the following way
∞

zσ (x , xn ) = ϕl (εx )zσ,l (xn ); x ∈ ∂Ωε , xn ∈ (0, ε−γ ).
l=0

If the eigenvalue σ is close to zero, it turns out that all the modes ϕl zσ,l
are negligible, except those for which λl ∼ ε−2 , see also formula (57). This is
stated rigorously in Proposition 33. It follows that, qualitatively, the resonant
eigenvalues of Iε (uε ) have eigenfunctions with more and more oscillations
along ∂Ωε as ε tends to zero.
We describe below the general procedure employed here to tackle the prob-
lem. As we mentioned above, we are going to use the contraction mapping
theorem, once two preliminary steps are accomplished.

Step 1: finding an approximate solution. Given an arbitrary positive

number θ, we are able to find an approximate solution uk,ε of (Pε ) for which
Iε (uk,ε ) = O(εθ ). The function uk,ε is constructed essentially by power
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 95

series in ε with k iterations, where the number k depends on n, p and θ.

Thus uk,ε has roughly the following form

uk,ε (x , xn ) = w0 (xn ) + εw̃1 (εx , xn ) + . . . εk w̃k (εx , xn ),

x ∈ ∂Ωε , xn ∈ Jε = [0, ε−γ ].

Here w̃1 , . . . , w̃k are smooth functions on ∂Ω × Jε , which are defined induc-
tively in their index. Basically each function w̃i , which depends on w̃0 , . . . ,
w̃i−1 and the geometry of Ω, is obtained by an inversion argument. Despite
the resonance phenomena of the operator Iε , here we can perform this inver-
sion because we are assuming a smooth dependence on the variable εx . In
some sense, since resonance occurs only at high frequencies, see the com-
ments above, smooth functions on ∂Ω (scaled to Ωε ) are not affected by
this phenomenon since their Fourier modes are mainly low-frequency ones.
The rigorous derivation of uk,ε is performed in Subsection 5: in this step the
smoothness of Ω is essential to construct uk,ε .
In [49], for n = 2, we were able to satisfy the above requirement only for
θ = 32 . To deal with a general n we need a better approximation since both
the energy of the solutions and the norm of the inverse operator grow faster
when the dimension n is larger.
Step 2: inverting the linearized operator Iε (uk,ε ). This is a rather
delicate issue since, as we remarked before, the linearized operator is not
invertible for all the values of ε. Qualitatively, using the Weyl’s asymptotic
formula, one finds that the l-th eigenvalue σl of Iε (uk,ε ) is given by
2
σl ∼ −1 + ε2 l n−1 , (57)

see Subsection 6 for details. It follows that σl ∼ 0 for l ∼ ε1−n , and that
the average distance between two consecutive eigenvalues close to zero is of
order εn−1 . We show indeed that along a sequence εj the spectrum of the
linearized operator stays away from zero of order εn−1
j , and hence we get
−(n−1)
an inverse operator with norm proportional to εj . The way to prove
this fact relies on a first rough comparison of the eigenvalues with those
(essentially known, see Proposition 29) of a model problem, obtaining an
estimate of the Morse index of the solutions. Then, Kato’s Theorem allows
us to choose the values of ε appropriately and to invert the linear operator
along a sequence εj . Notice that Kato’s Theorem requires some information
not only on the eigenvalues, but also on the eigenfunctions, see Subsection 4.

Final step: the contraction argument. If the operator Iε (uk,ε ) is invert-
ible, a function uε of the form uε = uk,ε + w is a solution of (Pε ) if and only
if w is a fixed point of the operator Fε , where Fε is defined as

Fε (w) = −Iε (uk,ε )−1 [Iε (uk,ε ) + Nε (w)].

96 A. Malchiodi

min{2,p}
Here Nε (w) is a superlinear term satisfying Nε (w) ≤ O (w) . Since
along the sequence εj the norm of the operator Iε (uk,ε )−1 is of order ε−n ,
we need to choose θ to be sufficiently large (see Step 1), depending also on
p, in order to get a contraction.
A further difficulty in dealing with the case n ≥ 3 is that the exponent p
could also be supercritical. This case is tackled with a truncation argument,
proving a priori estimates in L∞ . These are based on a combination of norm
estimates, obtained using Step 2, and elliptic regularity theory.
The outline of the second part of these notes is the following. In Section 4
we collect some preliminary facts and we study a family of auxiliary one-
dimensional problems, proving some continuity and monotonicity properties.
Section 5 is devoted to the construction of the approximate solution uk,ε . In
Section 6 we give a characterization of the eigenfunctions and the eigenval-
ues of the operator TΣε , which basically coincide with those of Iε (uk,ε ), see
Proposition 37. In Section 7 we finally prove Theorem 3.

4 Some Preliminary Facts

In this section we treat the linear theory of some auxiliary one-dimensional

problems. Below, C denotes a large positive constant. For convenience, we
allow C to vary among formulas (also within the same line) and to assume
larger and larger values. The number γ will be fixed in Subsection 5.
Consider the problem
⎧
⎪ 
⎨−u + u = u in R+ ;
p

u>0 in R+ ; (58)
⎪
⎩ 
u (0) = 0,

with p > 1. This is a particular case of (1) for n = 1. Therefore, calling the
solution w0 , this satisfies the properties
"
w0 (r) < 0, for all r > 0,
w  (r) (59)
limr→∞ er w0 (r) = αp > 0, limr→∞ w00 (r) = −1,

where αp is a positive constant depending only on p. Using some ODE

analysis, one can see that all the solutions of (58) in H 1 (R+ ) coincide with w0 .
The Euler functional of (58) is
 
1  2
1
I + (u) = (u ) + u2 − |u|p+1 , u ∈ H 1 (R+ ). (60)
2 R+ p + 1 R+

Since in one dimension we have no translation invariance once we impose

Neumann conditions, Corollary 17 becomes the following.
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 97

Proposition 25 The function w0 is a non-degenerate critical point of I + .

Precisely, there exists a positive constant C such that
 
I + (w0 )[w0 , w0 ] ≤ −C −1 w0 20 ; I + (w0 )[v, v] ≥ C −1 v20 ,

for all v ∈ H 1 (R+ ), v ⊥ w0 ,

where  · 0 denotes the standard norm of H 1 (R+ ). As a consequence, we
have μ < 0 and τ > 0, where μ and τ are respectively the first and the second

eigenvalues of I + (w0 ); furthermore μ is simple.

Our next goal is to characterize the eigenvalues σ (in particular the first two)
of the following problem
"
(1 − σ)(−u + (1 + α)u) = pw0p−1 u, in R+ ;
(61)
u (0) = 0,

where α is a positive parameter. Equation (61) arises in the study of a model

for the linearization of (Pε ) near approximate solutions after performing a
Fourier decomposition, see equation (81) below.

In order to study the eigenvalues of (61), it is convenient to introduce the

following norm  · α on H 1 (R+ )

 2

u2α = (u ) + (1 + α)u2 ; u ∈ H 1 (R+ ). (62)

R+

Let Hα be the Hilbert space consisting of the functions in H 1 (R+ ), endowed

with this norm: we denote by ( , )α the corresponding scalar product. We
also define by duality the operator Tα : Hα → Hα in the following way
  
(Tα u, v)α = u v  + (1 + α)uv − pw0p−1 uv ; u, v ∈ Hα .
R+

Then equation (61) can be written in the abstract form

Tα u = σu; u ∈ Hα : (63)

note that when α = 0, Tα coincides with I + (w0 ). We have the following
result.

Proposition 26 Let μα and τα denote respectively the first and the second
eigenvalues of Tα . Then μα is simple and the following properties hold true
(i) α → μα is smooth and strictly increasing;
(ii) α → τα is non-decreasing.
98 A. Malchiodi

The eigenfunction vα of Tα corresponding to μα , normalized with vα α = 1,

can be chosen to be positive and strictly decreasing on R+ . The map α → vα
is smooth from R into H 1 (R+ ). Furthermore, μα > 0 for α sufficiently large.

Proof. The simplicity on μα and the (weak) monotonicity of α → μα ,

α → τα can be proved as follows. By the Courant-Fischer formula we have
that   2 p−1 2

R+
(u ) + (1 + α)u 2
− pw0 u
μα = inf . (64)
u∈Hα
R+
((u )2 + (1 + α)u2 )
If u is a minimizer, also the decreasing rearrangement of |u| is, and therefore
eigenfunctions corresponding to the first eigenvalue have constant sign. By
orthogonality they are then unique. The (weak) monotonicity in α is a simple
consequence of the monotonicity of the quantity
  2 p−1 2

R+
(u ) + (1 + α)u 2
− pw0 u
.
((u  )2 + (1 + α)u2 )
R+

The smoothness of α → μα and α → vα can be deduced as follows: consider

the map Ψα : H 1 (R+ ) → Hα defined as

(Ψα u)(x) = u(x); u ∈ H 1 (R+ ).

Ψα is nothing but the identity as a map between functions, and is an isomor-

phism between Sobolev spaces. We also consider the operator Tα : H 1 (R+ ) →
H 1 (R+ )
Tα = Ψα−1 ◦ Tα ◦ Ψα .
Tα depends smoothly on α and, by conjugacy to Tα , the first eigenvalue of
Tα coincides with μα and is simple. Hence, to get the smoothness of μα and
vα , it is sufficient to apply the implicit function theorem.
We now compute the derivative of μα with respect to α. The function vα
satisfies the equation
"
(1 − μα ) (−vα + (1 + α)vα ) = pw0p−1 vα in R+ ;
(65)
vα (0) = 0.

Differentiating the equation vα 2α = 1, with respect to α, we find

  
d dvα
vα α = 0,
2
which implies , vα =− vα2 . (66)
dα dα α R+
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 99

On the other hand, differentiating (65) with respect to α we obtain

⎧    
⎨−dμα (−v +(1+α)vα )+(1−μα ) − dvα  + (1+α) dvα +vα = pwp−1 dvα in R+ ;
dα α dα dα 0 dα
 
⎩ dvα
(0) = 0.
dα
(67)
Multiplying (67) by vα , integrating by parts and using (66), one gets

dμα
= (1 − μα ) vα2 > 0. (68)
dα R+

Note that, since Tα ≤ Id, every eigenvalue of Tα is strictly less than 1, and
in particular (1 − μα ) > 0. This proves the strict monotonicity of α → μα . It
remains to prove that μα > 0 for α large: in order to do this we show that
μα → 1 as α → +∞. Fixing any δ > 0, it is sufficient to notice that
   
(u )2 + (1 + α) − pw0p−1 u2 ≥ (1 − δ) (u )2 + (1 + α)u2 for all u,

provided α is sufficiently large, and to use (64). This concludes the proof.
Given α > 0, we also define the function F̃ (α) as

F̃ (α) = 2α(1 − μα ) vα2 > 0. (69)
R+

Note that, by the smoothness of α → vα , also the function α → F̃ (α) is

smooth.

We also need to consider a variant of the eigenvalue problem (61). For γ ∈

(0, 1), set
  

Jε = 0, ε−γ ; Hε1 = u ∈ H 1 (Jε ) : u ε−γ = 0 .

We let Hα,ε denote the space Hε1 endowed with the norm

 2

u2α,ε = (u ) + (1 + α)u2 ; u ∈ Hε1 ,

Jε

and ( , )α,ε the corresponding scalar product. Similarly, we define Tα,ε by

  
(Tα,ε u, v)α,ε = u v  + (1 + α)uv − pw0p−1 uv ; u, v ∈ Hα,ε .
Jε

The operator Tα,ε satisfies properties analogous to Tα . We list them in the

next proposition, which also gives a comparison between the first eigenvalues
and eigenfunctions of Tα and Tα,ε .
100 A. Malchiodi

Proposition 27 Let μα,ε and τα,ε denote respectively the first and the second
eigenvalues of Tα,ε . Then μα,ε is simple and the following properties hold true
(i) α → μα,ε is smooth and strictly increasing;
(ii) α → τα,ε is non-decreasing.
The eigenfunction vα,ε of Tα,ε corresponding to μα,ε , normalized with
vα,ε α,ε = 1, can be chosen to be positive and strictly decreasing on R+ .
Moreover there exist a large constant C and a small constant δ depending
only on p such that
 
−γ 1 −γ 1 1
|μα−μα,ε | ≤ Ce−δε ; vα−vα,ε H 1 (R+ ) ≤ Cε− 2 γ e−δε , for μα ∈ − τ, τ
4 4

and for ε small. The function vα,ε in this formula has been set identically 0
outside [0, ε−γ ].

Proof. Properties (i)–(ii) and of the monotonicity of vα,ε can be deduced

as for Proposition 26.
Also, we have

(Tα u, u)α (Tα,ε u, u)α,ε

μα = inf ; μα,ε = inf . (70)
u∈Hα u2α u∈Hα,ε u2α,ε

Since the norms  · α and  · α,ε (and the operators Tα , Tα,ε ) coincide on
Hα,ε , we have μα ≤ μα,ε . Conversely,
  μα → 1 as α → +∞ (see the
since
proof of Proposition 26), μα ∈ − 41 τ, 14 τ implies that α lies in a bounded

set of R+ . Note that, since I (w0 ) ≤ Id on H 1 (R+ ), τ is strictly less than 1,
and so is τ4 .
Using equation (61), we deduce the following decay for vα

|vα (t)| + |vα (t)| ≤ Ce−δt , t ≥ 0. (71)

Let us define a non-increasing cut-off function ϕε satisfying the following

properties
⎧  
⎪
⎨ϕε (t) = 1 for t ∈ 0, 12 ε−γ ;
 
ϕε (t) = 0 for t ∈ 34 ε−γ , ε−γ ; (72)
⎪
⎩  
|ϕε (t)| ≤ Cε ; |ϕε (t)| ≤ Cε .
γ 2γ

By (71) and some elementary computations, one finds

 
ϕε vα 2α,ε − 1 ≤ Ce−δε−γ ; |(Tα,ε (ϕε vα ), ϕε vα ) − μα | ≤ Ce−δε
−γ
.

Since ϕε vα belongs to Hα,ε , (70) and the last formulas yield μα,ε ≤ μα +
Ce−δε . This proves the first inequality in the statement.
γ
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 101

Formula (71) implies



−γ
(vα )2 + vα2 ≤ Cε−δε .
R+ \Jε

To obtain the second inequality in the statement we set

u(t) = vα,ε (t) − vα (t) + ε2γ vα (ε−γ )t2 ; t ∈ Jε .

Subtracting the differential equations satisfied by vα,ε and vα , we obtain

"
(1 − μα,ε )(−u + (1 + α)u) − pw0p−1 u = f1 + f2 in Jε ;
(73)
u (0) = 0 u ∈ Hα,ε ,

where f1 and f2 are defined by duality as


(f1 , v) = (μα,ε − μα ) (−Δvα + (1 + α)vα )vdt, ∀v ∈ Hα,ε ;
Jε
 / 0

(f2 , v) = ε2γ vα (ε−γ ) (1−μα,ε ) (1+α)t2 −2 −pw0p−1 t2 vdt, ∀v ∈ Hα,ε .

Jε

One can check that

−γ 1 −γ
f1 Hα,ε ≤ Ce−δε ; f2 Hα,ε ≤ Cε− 2 γ e−δε . (74)

We write

u = βvα,ε + u, where β∈R and (u, vα,ε )Hα,ε = 0. (75)

Since Tα,ε is invertible on (vα,ε )⊥ , (74) implies

1 −γ
uHα,ε ≤ Cε− 2 γ e−δε . (76)

Equation (75) can also be written as

(1 − β)vα,ε = u + vα − ε2γ vα (ε−γ )t2 .

Taking the norm of both sides, using (76) and some elementary estimates we
get  1 
−γ
|1 − β| = 1 + O ε− 2 γ e−δε .

This equation implies β ∼ 0 or β ∼ 2, from which we deduce (see (75))

that either vα,ε ∼ vα , or vα,ε ∼ −vα . Since we are assuming that vα,ε is
1 −γ
non-increasing, it must be β ∼ 0, and more precisely |β| ≤ Cε− 2 γ e−δε .
Then (75) and (76) yield the conclusion.
102 A. Malchiodi

Remark 28 Using the Courant-Fischer method, one can also prove that
τα,ε ≥ τα for every α and ε.

In order to study equation (Pε ) we employ Fourier analysis. For γ > 0, let us
define the set Sε and the metric g0 on Sε as
 
Sε = ∂Ωε × Jε = ∂Ωε × 0, ε−γ , g0 = gε ⊗ (dt)2 , (77)

where gε is the metric on ∂Ωε induced by Rn . We also define the functional

space  
HSε = u ∈ H 1 (Sε ) : u(x , ε−γ ) = 0 for all x ∈ ∂Ωε ,
endowed with the norm


u2HSε = |∇g0 u|2 + u2 ; u ∈ HS ε .

Sε

Let ĝ be the metric on ∂Ω induced by that of Rn , and let Δĝ be the

Laplace-Beltrami operator. Let (λl )l denote the eigenvalues of −Δĝ , counted
in non-decreasing order and with their multiplicity. Let also (ϕl )l denote the
corresponding eigenfunctions.
Given u ∈ HSε , we can decompose it in Fourier components (in the variable
x ) as follows
∞


u(x , xn ) = ϕl (εx )ul (xn ), x ∈ ∂Ωε , xn ∈ Jε . (78)
l=0

Using this decomposition, one finds


1   2
1 
uHSε = n−1
2
(ul ) + (1 + ε2 λl )u2l = n−1 ul 2ε2 μl .
ε Jε ε
l l

Writing for brevity  · l,ε instead of  · ε2 μl , the last equation becomes

1 
u2HSε = ul 2l,ε . (79)
εn−1
l

This is how the norms  · α introduced above enter in our study; in particular
we will choose α belonging to the discrete set (ε2 λl )l .
We are also interested in an eigenvalue problem of the form

TSε u = σu; u ∈ HS ε , (80)

where TSε is defined by

  
(TSε u, v) = ∇g0 u, ∇g0 v + uv − pw0p−1 uv dVg0 , u, v ∈ HSε .
Sε
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 103

Writing u as in (78), equation (80) is equivalent to

"

(1 − σ) −ul + (1 + ε2 λl )ul = pw0p−1 ul in Jε ;

for all l. (81)
ul (0) = 0,

Since TSε represents a model for the study of Iε (uk,ε ), it was essential for us
to perform a spectral analysis of the operators Tα . The spectrum of TSε is
characterized in the next proposition. We recall the definition of the (positive)
number τ in Proposition 25.

Proposition 29 Let σ < τ4 be an eigenvalue of TSε . Then σ = μl,ε for some

index l. The corresponding eigenfunctions u are of the form

u(x , xn ) = αl ϕl (εx )vl,ε (xn ), x ∈ ∂Ωε , xn ∈ Jε , (82)
{l : μl,ε =σ}

where (αl )l are arbitrary constants. Viceversa, every function of the form (82)
is an eigenfunction of TSε with eigenvalue σ. In particular the eigenvalues of
TSε which are smaller than τ coincide with the numbers (μl,ε )l which are
smaller than τ .

Proof. Let u be an eigenfunction of TSε with eigenvalue σ. Then we can

write 
u(x , xn ) = (αl vl,ε (xn ) + v̂l (xn )) ϕl (εx ),
l

where αl are real numbers, and where (v̂l , vl,ε )l,ε = 0 for all l. We have

(αl μl,ε vl,ε (xn ) + Tl,ε v̂l (xn )) ϕl (εx ) = TSε u = σu
l

=σ (αl vl,ε (xn ) + v̂l (xn )) ϕl (εx ).
l

By the uniqueness of the Fourier decomposition, the last equation implies

μl,ε = σ if αl = 0; and Tl,ε v̂l = σv̂l for all l. (83)

Proposition 27 ii) and Remark 28 yield

τα,ε ≥ τα ≥ τ, (84)

which means that Tα,ε ≥ τ Id on the subspace of Hα,ε orthogonal to vα,ε .

Therefore, the second equation in (83) and the fact that σ < τ4 imply v̂l = 0
for all l. Moreover, since λl → μl,ε is a monotone function, the first equation
in (84) shows that all the indices l for which αl = 0 correspond to the same
value λl . This concludes the proof.
104 A. Malchiodi

We finally recall the following theorem due to T. Kato, ([39], page 444) which
is fundamental to us in order to obtain invertibility, see Proposition 36.

Theorem 30 Let T (χ) denote a differentiable family of operators from an

Hilbert space X into itself, where χ belongs to an interval containing 0. Let
T (0) be a self-adjoint operator of the form Identity - compact and let σ(0) =
σ0 = 1 be an eigenvalue of T (0). Then the eigenvalue σ(χ) is differentiable
at 0 with respect to χ. The derivative of σ is given by
 
∂σ ∂T
= eigenvalues of Pσ0 ◦ (0) ◦ Pσ0 ,
∂χ ∂χ

where Pσ0 : X → Xσ0 denotes the projection onto the σ0 eigenspace Xσ0
of T (0).

Remark 31 We note that, when perturbing the operator T (0), the eigen-
value σ0 can split in several ones, so in general σ0 possesses a multivalued
derivative. Anyway, since the operator T (0) is of the form Identity - com-
pact, and since σ0 is different from 1, σ0 is an isolated eigenvalue and the
projection Pσ0 has a finite dimensional range, therefore the splitting of σ0 is
always finite.

5 An Approximate Solution

In this subsection we construct approximate solutions uk,ε of (Pε ) up to an

arbitrary order in ε. Basically, we expand (Pε ) in powers of ε, solve it term
by term, and then use suitable truncations.

Proposition 32 Consider the Euler functional Iε defined in (6) and associ-

ated to problem (Pε ). Then for any k ∈ N there exists a function uk,ε : Sε → R
with the following properties
∂uk,ε
Iε (uk,ε ) ≤ Ck εk+1−
n−1
2 ; uk,ε ≥ 0 in Ωε =0 on ∂Ωε ,
∂ν
(85)
where Ck depends only on Ω, p and k. Moreover one has
⎧ 
⎪   (m)   m −xn
⎨(∇ ) uk,ε (x , xn ) ≤Cm ε e ;

(∇ )(m) ∂xn uk,ε (x , xn ) ≤ Cm εm e−xn ; x ∈ ∂Ωε , xn ∈ Jε , , m = 0, 1, . . . ,
⎪
⎩ 
(∇ )(m) ∂x2 uk,ε (x , xn ) ≤ Cm εm e−xn ;
n
(86)
where ∇ and  ·  denote the derivative and the norm taken with respect to
the variable x (freezing xn ), and Cm is a constant depending only on Ω, p
and m.
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 105

We parameterize the set Sε , see equation (77), using coordinates x on ∂Ωε

and xn in Jε . Let ν denote the unit inner normal to ∂Ω, and define the map
Γε : ∂Ωε × Jε → Rn by

Γε (x , xn ) = x + xn ν(εx ).

We let the upper-case indices A, B, C, . . . run from 1 to n, and the lower-

case indices i, j, k, . . . run from 1 to n − 1. Using some local coordinates
(xi )i=1,...,n−1 on ∂Ωε , and letting ϕε be the corresponding immersion into
Rn , we have
⎧
⎪ ∂Γε  ∂ϕε  ∂ν
⎪
⎪ (x , xn ) = (x ) + εxn (εx )
⎪
⎨ ∂xi ∂xi ∂xi
∂ϕε 
(x ) + εxn Hij (εx )
∂ϕε  for i = 1, . . . , n − 1;
⎪
⎪ = (x )
⎪
⎪ ∂x i ∂xj
⎩ ∂Γε  
∂xn (x , xn ) = ν(εx ).

where (Hij ) are the coefficients of the mean-curvature operator on ∂Ω. Let
also gij be the coefficients of the metric on ∂Ωε in the above coordinates
(x , xn ). Then, letting g = gε denote the metric on Ωε induced by Rn , we
have 1 2  
∂Γε ∂Γε (gij ) 0
gAB = , = , (87)
∂xA ∂xB 0 1
where
1 2
∂ϕε  k  ∂ϕε  ∂ϕε  l  ∂ϕε 
gij = (x ) + εxn Hi (εx ) (x ), (x ) + εxn Hj (εx ) (x )
∂xi ∂xk ∂xj ∂xl
k

= gij + εxn Hi g kj + Hjl gil + ε2 x2n Hik Hjl g kl .

Note that also the inverse matrix (g AB ) decomposes as

 ij 
AB (g ) 0
g = .
0 1

We begin by finding a first-order approximation for uε , to show the ideas of

the general procedure. We define the following map u → ũ from functions on
∂Ωε (resp. ∂Ωε × R+ ) into functions on ∂Ω (resp. ∂Ω × R+ )

ũ(εx ) = u(x ) (resp. ũ(εx , xn ) = u(x , xn )),

(88)
x ∈ ∂Ωε (resp. (x , xn ) ∈ ∂Ωε × R+ ).

Using the above parametrization, we look for an approximate solution u1,ε

of the form
u1,ε (x , xn ) = w0 + εw̃1 (εx , xn ),
106 A. Malchiodi

where the function w̃1 is defined on ∂Ω × R+, consistently with (88). We note
that, from the above decomposition of gAB (and g AB ), one finds

1   
−Δg u = −g AB uAB − √ ∂A g AB det g uB
det g
1   1   
= −unn −g ij uij − √ ∂n det g un − √ ∂i g ij det g uj .
det g det g
We have, formally

det g = det(g −1 g) det g = (det g) 1 + εxn tr (g −1 α) + o(ε), (89)

where
αij = Hik g kj + Hjl gil .
We have

(g −1 α)is = g sj αij = g sj Hik gkj + Hjl g il ,

and hence

tr (g −1 α) = g ij Hik gkj + Hjl g il = 2gij Hik gkj = 2Hii . (90)

We recall that the quantity Hii represents the opposite of the mean curvature
of ∂Ω (by our choice of ν), and in particular it is independent of the choice
of coordinates.
Using (89), (90), writing formally (w0 + εw1 )p as w0p + pεw0p−1 w1 + o(ε),
and expanding −Δgε u1,ε + u1,ε = up1,ε up to first order in ε, we obtain the
following equation for w1
"
−w1 + w1 − pw0p−1 w1 = H11 w0 in R+ ;
(91)
w1 (0) = 0.

By Proposition 25 and the subsequent comment there, equation (91) can

be solved for any right-hand side in L2 (R2+ ), as in this case (see equation
(59)). The resulting expression for w1 will be independent of the choice of
coordinates on ∂Ωε .
The construction of a better approximate solution is performed by expand-
ing formally the equation −Δgε u + u = up in powers of ε up to order k. A
rigorous proof is in [50].
Given smooth functions w̃1 , . . . w̃k : ∂Ω × R+ , let

ûk,ε (x , xn ) = w0 (xn )+εw̃1 (εx , xn ) + . . . εk w̃k (εx , xn ); x ∈ ∂Ωε , xn ∈ R+ .

Expanding formally the equation −Δg u + u = up (imposing Neumann

boundary conditions) in powers of ε, we find that w̃i satisfies the following
recurrence formula, for i running from 1 to k
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 107
⎧
⎪  i
⎪
⎪ −w̃i + w̃i −pw0p−1 w̃i = xj−1   
⎪
⎪ n G̃i (x )w̃i−j (εx , xn )
⎪
⎪
⎪
⎪
j=1
⎪
⎪
⎪
⎪ 
i−2
⎨ − (Li−j−2 w̃j )(εx , xn )
in R+ ;
⎪
⎪ 
j=0

⎪
⎪
⎪ (εx , xn )
p− jl j1 ji−1
⎪ + Ci,j1 ,...,ji w0 w̃1 · · · w̃i−1
⎪
⎪
⎪
⎪
⎪
j1 ,...,ji−1

⎪
⎪ ljl =i
⎩
w̃k (0) = 0,
(92)
where the derivatives are taken with respect to the variable xn , the coefficients
Ci,j1 ,...,ji are constants, G̃i : ∂Ω → R are smooth functions and Li−j−2 are
linear second-order differential operators.
Proceeding by induction and using standard estimates on solutions of
ODE’s, we find that there exist polynomials Pi (t) such that
⎧ 
⎪   (m)   −xn
⎨(∇ ) w̃i (x , xn ) ≤ Ci,m Pi (xn )e ;

(∇ ) w̃ (x , xn ) ≤ Ci,m Pi (xn )e−xn ; x ∈ ∂Ωε , xn ∈ R+ , m = 0, 1, . . . ,
 (m)  
⎪
⎩
i

(∇ )(m) w̃ (x , xn ) ≤ Ci,m Pi (xn )e−xn ;
i
(93)
where ∇ denotes the derivative with respect to x , and Ci,m is a constant
depending only on Ω, p, i and m. In particular these decay estimates ensure
that the right-hand side in (92) belongs to L2 (R+ ), and (92) is solvable for
all i = 1, . . . , k.
In conclusion, uk,ε is obtained by multiplying ûk,ε by a cutoff function sup-
ported in {xn ∈ [0, ε−γ ]}: to guarantee convergence of the Taylor expansions
up to the desired order, the constant γ has to be taken sufficiently small,
depending on k, n and p (see Subsection 3.2 in [50]).

6 Eigenvalues of the Linearized Equation

Let Σε be defined as
 
Σε = Γε (Sε ) := Γε (∂Ωε × Jε ) = x ∈ Rn : dist(x, ∂Ωε ) < ε−γ .

We endow Σε with the metric gε induced by the inclusion in Rn , and introduce

the functional space


HΣε = u ∈ H 1 (Σε ) : u Γε (x , ε−γ ) = 0 for all x ∈ ∂Ωε ,
108 A. Malchiodi

with its natural norm



u2HΣε = |∇gε u|2 + u2 ; u ∈ HΣ ε ,

Σε

and the corresponding scalar product ( , )HΣε . Using the expression of the
metric gε , see equation (87) and the subsequent formulas, one finds

1 − Cε1−γ uHSε ≤ uHΣε ≤ 1 + Cε1−γ uHSε ; for all u ∈ HSε ,

(94)
where C is a constant depending only on Ω.
Let uk,ε be the function constructed in Subsection 5. Then we define TΣε :
HΣε → HΣε in the following way
  
(TΣε u, v)HΣε = ∇gε u, ∇gε v + uv − pup−1
k,ε uv dVgε ; u, v ∈ HΣε .
Σε
(95)
The goal of this subsection is to study the eigenvalues σ of the problem

TΣε u = σu, u ∈ HΣ ε . (96)

In the next proposition we characterize the eigenfunctions of TΣε when the

eigenvalue σ is close to zero. We recall the definition of the functions ϕl and
vl,ε from Subsection 4. Proposition 33 is the counterpart of Proposition 29
for a non-flat metric.
Proposition 33 Let γ > 0 be sufficiently small, and let L ∈ (0, τ /4). Sup-
pose σ ∈ [−τ /4, τ /4], and let u be an eigenfunction of TΣε with eigenvalue σ.
Then one has
 
  
  ε1−γ
u − α ϕ v  ≤C uHΣε , (97)
 l l l,ε 
L
 {l:μl,ε ∈[σ−L,σ+L]} 
HΣε

for some coefficients (αl )l and for some constant C depending only on Ω
and p.
Let uk,ε : Ω → R be the function defined by

uk,ε (x) = uk,ε (εx), x ∈ Ω.

We have next the following result.

Proposition 34 The eigenvalues of the operator TΣε are differentiable with
respect to ε, the eigenvalues being considered a possibly multivalued function
of ε. If σ(ε) is such an eigenvalue, then

∂σ
= {eigenvalues of Qσ } , (98)
∂ε
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 109

where Qσ : Hσ × Hσ → R is the quadratic form given by

 
2 ∂uk,ε
Qσ (u, v) = (1 − σ) ∇u, ∇v − p(p − 1) uvup−2
k,ε (ε ·). (99)
ε Σε Σε ∂ε

Here Hσ ⊆ HΣε denotes the eigenspace of TΣε corresponding to σ and the

function uk,ε is defined in Subsection 5.

This formula can be derived heuristically differentiating in ε (96) (scaled

to Ω). A rigorous proof, using Kato’s theorem, can be found in [49]. We
apply Proposition 34 to the eigenvalues σ which are close to 0.

Proposition 35 Let σ be an eigenvalue of TΣε belonging to the interval

[−τ /4, τ /4]. Then one has
 
 ∂σ 1  − 1+γ
 
 ∂ε − ε F̃ (σ) ≤ Cε
2 ,

where γ is sufficiently small, F̃ (·) is given in (69), and C is a constant

depending only on Ω and p.

We are now in position to prove the following proposition, which characterizes

the spectrum of TΣε .

Proposition 36 Let uk,ε and TΣε be as above. Then there exists a positive
constant C, depending on p, Ω and k with the following property. For a suit-
able sequence εj → 0, the operator TΣεj : HΣεj → HΣεj is invertible and the
inverse operator satisfies
 
 −1  C
TΣε  ≤ , for all j ∈ N.
j εn−1
j

Proof. Let (σj )j (resp. (sj )j ) denote the eigenvalues of TΣε (resp. TSε ),
counted in increasing order with their multiplicity. Let also

N (Σε ) = ! {j : σj ≤ 0} ; (resp. N (Sε ) = ! {j : sj ≤ 0}).

denote the number of non-positive eigenvalues of TΣε (resp. TSε ). Recall

that, from Proposition 29, the eigenvalues of TSε coincide with the numbers
(μj,ε )j := με2 λj ,ε , where (λj )j are the eigenvalues of the Laplace-Beltrami
operator −Δĝ , see also equations (78), (81). We also recall the Weyl’s
asymptotic formula, see e.g. [18] page 169
  n−1
2
(2π)2 j 2
λj ∼ 2 = Cn,Ω j n−1 , as j → +∞. (100)
n−1
ωn−1 V ol(∂Ω)
110 A. Malchiodi

From the strict monotonicity of α → μα := μ(α) and from the last formula
we deduce  
! j : ε2 λj ≤ μ−1 (0) ∼ C̃n,Ω,p ε1−n . (101)
From Proposition 27, we have
  / τ τ0
με2 λ − με2 λj  ≤ Ce−δε ;
−γ
provided με2 λj ∈ − , , (102)
j ,ε
4 4
which implies
 −γ
  −γ

! j : ε2 λj ≤ μ−1 (0)−Ce−δε ≤ N (Sε ) ≤ ! j : ε2 λj ≤ μ−1 (0)+Ce−δε

and hence
N (Sε ) ∼ C̃n,Ω,p ε1−n . (103)
From the Courant-Fischer method, we get
" 3
(TΣε u, u)HΣε
σj = inf sup : M subspace of HΣε , dim M = j ,
u∈M u2HΣε

and a similar characterization for sj . By (94) one finds

 
(TΣε u, u)HΣ − (TSε u, u)HS  ≤ Cε1−γ u2 ;
ε ε HSε
 
 
u2HΣε − u2HSε  ≤ Cε1−γ u2HSε ,

for all u ∈ HΣε and for some constant C depending only on Ω and p. Note
that, still by (94), in the last inequality we can also substitute  · HSε with
 · HΣε . The last two formulas yield

|sj − σj | ≤ Cε1−γ , for all j, (104)

where C depends only on Ω and p. From (102), (104) there exists a positive
constant C, depending only on Ω and p, such that
   
! j : ε2 λj ≤ μ−1 (0) − Cε1−γ ≤ N (Σε ) ≤ ! j : ε2 λj ≤ μ−1 (0) + Cε1−γ ,

which implies
    n−1
μ−1 (0) μ−1 (0) 2
N (Σε ) ∼ ! j : λj ≤ ∼ = C̃n,Ω,p ε1−n . (105)
ε2 ε2 Cn,Ω

For l ∈ N, let εl = 2−l . Then from (105) we have

 

N (Σεl+1 )−N (Σεl ) ∼ C̃n,Ω,p 2(l+1)(n−1) −2l(n−1) = C̃n,Ω,p 2n−1 −1 ε1−n

l .
(106)
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 111

Note that, by Proposition 35, the eigenvalues of TΣε close to 0 decrease when
ε decreases to zero. In other words, by the last equation, the number of
eigenvalues which cross 0, when ε decreases from εl to εl+1 , is of order ε1−n
l .
Now we define

Al = {ε ∈ (εl+1 , εl ) : ker TΣε = ∅} ; Bl = (εl+1 , εl )\Al .

By Proposition 35 and (106), it follows that card(Al ) < Cε1−n

l , and hence
there exists an interval (al , bl ) such that

meas(Bl )
(al , bl ) ⊆ Bl ; |bl − al | ≥ C −1 ≥ C −1 εnl . (107)
card(Al )

From Proposition 35 we deduce that

 
 −1 
TΣ al +bl is invertible and T ≤ C .
 Σ a +b
l l  εn−1
2 2 l

aj +bj
Now it is sufficient to set εj = 2 . This concludes the proof.

7 Proof of Theorem 3

In this subsection we prove Theorem 3. We just treat the case p ≤ n+2 n−2 : for
n+2
p > n−2 , see [50] (some comments have been given at the beginning of this
part). The following result can be proved as is [49], Proposition 5.6.

Proposition 37 Let k ∈ N be fixed, and let uk,ε be as in Subsection 5.

Let (σj )j denote the eigenvalues of TΣε , counted in increasing order and with
their multiplicity, and let (σ̂j )j denote the eigenvalues of Iε (uk,ε ). Then there
exists a positive constant C, depending only on Ω and p, such that
−
γ / τ τ0
|σj − σ̂j | ≤ Ce−δε 2 ; whenever σj ∈ − , .
4 4
The proof of this proposition consists in showing that the eigenfunctions of
Iε (uk,ε ) decay exponentially away from ∂Ωε , and and in modifying them
through suitable cut-off functions. The eigenvalues σj and σ̂j are then com-
pared using the Courant-Fischer method, as in the proof of Proposition 27.
Let εj be as in Proposition 36: then from Proposition 37 we deduce

C
Iεj (uk,εj ) is invertible and Iεj (uk,εj )−1  ≤ . (108)
εn−1
j
112 A. Malchiodi

For brevity, in the rest of the proof, we simply write ε instead of εj . We apply
the contraction mapping theorem, looking for a solution uε of the form

uε = uk,ε + w, w ∈ H 1 (Ωε ).

If Iε (uk,ε ) is invertible (which is true along the sequence εj ), we can write
−1
Iε (uk,ε + w) = 0 ⇔ w = − (Iε (uk,ε )) [Iε (uk,ε ) + Nε (w)] , (109)

where
Nε (w) = Iε (uk,ε + w) − Iε (uk,ε ) − Iε (uk,ε )[w].
Let us define the operator Fε : H 1 (Ωε ) → H 1 (Ωε )
−1
Fε (w) = − (Iε (uk,ε )) [Iε (uk,ε ) + Nε (w)] , w ∈ H 1 (Ωε ).

We are going to prove that Fε is a contraction in a suitable closed set of

H 1 (Ωε ): from Proposition 32 and (108) we get
⎧  n−1 
⎨Cε−(n−1) εk+1− 2 + wp for p ≤ 2,
j
Fεj (w) ≤  j  w ≤ 1;
⎩Cε−(n−1) ε
n−1
k+1−
j j
2
+ w 2
for p > 2;
⎧ −(n−1) (110)

⎪
⎨ j
Cε w 1  p−1
+ w 2  p−1
w1 − w2  p ≤ 2,
−(n−1)
Fεj (w1 ) − Fεj (w2 ) ≤ Cεj (w1  + w2 ) w1 − w2  p > 2;
⎪
⎩
w1 , w2  ≤ 1.
(111)
Now we choose integers d and k such that
"
n−1
for p ≤ 2, 3
d > p−1 k + 1 > d + (n − 1), (112)
n − 1 for p > 2; 2

and we set  
Bj = w ∈ H 1 (Ωε ) : w ≤ εdj .
From (110), (111) it follows that Fεj is a contraction in Bj for εj small. Since
uε is close in norm to uk,ε ≥ 0, the positivity of uε follows from standard
arguments, based on the Sobolev inequalities. The points i) and ii) in the
statement follow from the construction of uε . This concludes the proof.

Remark 38 By the above construction and some standard estimates one has
 

Iεj (uεj ) = Iεj (uk,εj ) + O εd min{1,p−1} = Iεj (uk,εj ) + o εnj .

It follows from Propositions 36, 37 that the Morse index of uεj coincides with
N (Σεj ) ∼ ε1−n
j .
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 113

References

1. Ambrosetti, A., Badiale, M., Cingolani, S., Semiclassical states of nonlinear

Schrödinger equations, Arch. Rational Mech. Anal. 140 (1997), 285–300.
2. Ambrosetti, A., Felli, V., Malchiodi, A., Ground states of nonlinear Schrödinger
equations with potentials vanishing at infinity, J. Eur. Math. Soc. 7 (2005), 117–144.
3. Ambrosetti, A., Malchiodi, A., Perturbation Methods and Semilinear Elliptic Problems
on Rn , Birkhäuser, Progr. in Math. 240, (2005).
4. Ambrosetti, A., Malchiodi, A., Nonlinear Analysis and Semilinear Elliptic Problems,
Cambridge Univ. Press, Cambridge Studies in Advanced Mathematics, No.104 (2007).
5. Ambrosetti, A., Malchiodi, A., Ni, W.M., Singularly Perturbed Elliptic Equations with
Symmetry: Existence of Solutions Concentrating on Spheres, Part I, Comm. Math.
Phys., 235 (2003), 427–466.
6. Ambrosetti, A., Malchiodi, A., Ni, W.M., Singularly Perturbed Elliptic Equations with
Symmetry: Existence of Solutions Concentrating on Spheres, Part II, Indiana Univ.
Math. J. 53 (2004), no. 2, 297–329.
7. Ambrosetti, A., Malchiodi, A., Ruiz, D., Bound states of Nonlinear Schrödinger
Equations with Potentials Vanishing at Infinity, J. d’Analyse Math., 98 (2006),
317–348.
8. Ambrosetti, A., Malchiodi, A., Secchi, S., Multiplicity results for some nonlinear sin-
gularly perturbed elliptic problems on Rn , Arch. Rat. Mech. Anal. 159 (2001) 3,
253–271.
9. Ambrosetti, A., Rabinowitz, P.H., Dual variational methods in critical point theory
and applications, J. Funct. Anal. 14 (1973), 349–381.
10. Arioli, G., Szulkin, A., A semilinear Schrödinger equation in the presence of a magnetic
field. Arch. Ration. Mech. Anal. 170 (2003), 277–295.
11. Badiale, M., D’Aprile, T., Concentration around a sphere for a singularly perturbed
Schrödinger equation. Nonlinear Anal. 49 (2002), no. 7, Ser. A: Theory Methods,
947–985.
12. Bartsch, T., Peng, S., Semiclassical symmetric Schrdinger equations: existence of solu-
tions concentrating simultaneously on several spheres, Z Angew Math Phys, 58 (5)
(2007), 778–804.
13. Benci, V., D’Aprile, T., The semiclassical limit of the nonlinear Schrödinger equation
in a radial potential. J. Differential Equations 184 (2002), no. 1, 109–138.
14. Berestycki, H., Lions, P.L. Nonlinear scalar field equations. I. Existence of a ground
state. Arch. Rational Mech. Anal. 82 (1983), no. 4, 313–345.
15. Berestycki, H., Lions, P.L. Nonlinear scalar field equations. II. Existence of infinitely
many solutions. Arch. Rational Mech. Anal. 82 (1983), no. 4, 347–375.
16. Berezin, F.A., Shubin, M.A., The Schrdinger equation, Kluwer Academic Publishers
Group, Dordrecht, 1991.
17. Byeon, J., Wang, Z.Q., Standing waves with a critical frequency for nonlinear
Schrodinger equations, Arch. Rat. Mech. Anal. 165, 295-316 (2002).
18. Casten, R.G., Holland, C.J., Instability results for reaction diffusion equations with
Neumann boundary conditions, J. Diff. Eq. 27 (1978), no. 2, 266–273.
19. Cingolani, S., Pistoia, A., Nonexistence of single blow-up solutions for a nonlinear
Schrödinger equation involving critical Sobolev exponent, Z. Angew. Math. Phys. 55
(2004), no. 2, 201–215.
20. Dancer, E.N., Stable and finite Morse index solutions on Rn or on bounded domains
with small diffusion, Trans. Amer. Math. Soc., 357 (2005), no. 3, 1225–1243.
21. Dancer, E.N., Yan, S., Multipeak solutions for a singularly perturbed Neumann
problem. Pacific J. Math. 189 (1999), no. 2, 241–262.
22. Dancer, E.N., Yan, S., A new type of concentration solutions for a singularly perturbed
elliptic problem, Trans Amer Math Soc, 359(4) (2007), 1765–1790.
114 A. Malchiodi

23. D’Aprile, T., On a class of solutions with non-vanishing angular momentum for
nonlinear Schrödinger equations. Diff. Int. Equ. 16 (2003), no. 3, 349–384.
24. Del Pino, M., Felmer, P., Local mountain passes for semilinear elliptic problems in
unbounded domains, Calc. Var. 4 (1996), 121–137.
25. Del Pino, M., Felmer, P., Semi-classcal states for nonlinear schrödinger equations,
J. Funct. Anal. 149 (1997), 245–265.
26. Del Pino, M., Felmer, P., Kowalczyk, M., Boundary spikes in the Gierer-Meinhardt
system. Commun. Pure Appl. Anal. 1 (2002), no. 4, 437–456.
27. Del Pino, M., Felmer, P., Wei, J., On the role of the mean curvature in some singularly
perturbed Neumann problems, S.I.A.M. J. Math. Anal. 31 (1999), 63–79.
28. Del Pino, M., Kowalczyk, M., Wei, J., Multi-bump ground states of the Gierer-
Meinhardt system in R2 . Ann. Inst. H. Poincaré Anal. Non Linéaire 20 (2003),
no. 1, 53–85.
29. Del Pino, M., Kowalczyk, M., Wei, J., Concentration at curves for Nonlinear
Schrödinger Equations, Comm. Pure Appl. Math., 60 (2007), no. 1, 113–146.
30. Floer, A., Weinstein, A., Nonspreading wave packets for the cubic Schrödinger equation
with a bounded potential, J. Funct. Anal. 69, (1986), 397–408.
31. Gierer, A., Meinhardt, H., A theory of biological pattern formation, Kybernetik
(Berlin), 12 (1972), 30–39.
32. Grossi, M. Some results on a class of nonlinear Schrödinger equations, Math. Zeit. 235
(2000), 687–705.
33. Grossi, M., Pistoia, A., Wei, J., Existence of multipeak solutions for a semilinear
Neumann problem via nonsmooth critical point theory. Calc. Var. Partial Differential
Equations 11 (2000), no. 2, 143–175.
34. Gui, C., Multipeak solutions for a semilinear Neumann problem, Duke Math. J., 84
(1996), 739–769.
35. Gui, C., Wei, J., On multiple mixed interior and boundary peak solutions for some
singularly perturbed Neumann problems, Canad. J. Math. 52 (2000), no. 3, 522–538.
36. Gui, C., Wei, J., Winter, M., Multiple boundary peak solutions for some singularly
perturbed Neumann problems, Ann. Inst. H. Poincaré Anal. Non Linéaire 17 (2000),
no. 1, 47–82.
37. Iron, D., Ward, M., Wei, J., The stability of spike solutions to the one-dimensional
Gierer-Meinhardt model. Phys. D 150 (2001), no. 1-2, 25–62.
38. Jeanjean, L., Tanaka, K., A positive solution for a nonlinear Schrödinger equation on
RN . Indiana Univ. Math. J. 54 (2005), no. 2, 443–464.
39. Kato, T., Perturbation theory for linear operators. Second edition. Grundlehren der
Mathematischen Wissenschaften, Band 132. Springer-Verlag, Berlin-New York, 1976.
40. Kawohl, B., Rearrangements and convexity of level sets in PDE. Lecture Notes in
Mathematics, 1150. Springer-Verlag, Berlin, 1985.
41. Kwong, M.K., Uniqueness of positive solutions of −Δu + u + up = 0 in Rn , Arch.
Rational Mech. Anal. 105, (1989), 243–266.
42. Li, Y.Y., On a singularly perturbed equation with Neumann boundary conditions,
Comm. Partial Differential Equations 23 (1998), 487–545.
43. Li, Y.Y., Nirenberg, L., The Dirichlet problem for singularly perturbed elliptic
equations, Comm. Pure Appl. Math. 51 (1998), 1445–1490.
44. Lin, C.S., Ni, W.M., Takagi, I., Large amplitude stationary solutions to a chemotaxis
systems, J. Differential Equations, 72 (1988), 1–27.
45. Lin, F.H., Ni, W.M., Wei, J., On the number of interior peak solutions for a singularly
perturbed Neumann problem, 60 (2007), no. 2, 252–281.
46. Mahmoudi, F., Malchiodi, A., Concentration on minimal submanifolds for a singularly
perturbed Neumann problem, Adv. in Math., 209-2 (2007), 460–525.
47. Mahmoudi, F., Malchiodi, A., Wei, J., Transition Layer for the Heterogeneous Allen-
Cahn Equation, Annales IHP, Analyse non Lineaire, to appear.
48. Malchiodi, A., Concentration at curves for a singularly perturbed Neumann problem
in three-dimensional domains, G.A.F.A., 15-6 (2005), 1162–1222.
Concentration of Solutions for Some Singularly Perturbed Neumann Problems 115

49. Malchiodi, A., Montenegro, Boundary concentration phenomena for a singularly

perturbed elliptic problem, Comm. Pure Appl. Math, 15 (2002), 1507–1568.
50. Malchiodi, A., Montenegro, Multidimensional Boundary-layers for a singularly per-
turbed Neumann problem, Duke Math. J. 124 (2004), no. 1, 105–143.
51. Malchiodi, A., Ni, W.M., Wei, J., Multiple clustered layer solutions for semilinear
Neumann problems on a ball, Ann. Inst. H. Poincaré Anal. Non Linéaire, 22 (2005),
143–163.
52. Malchiodi, A., Wei, J., Boundary interface for the Allen-Cahn equation, J. Fixed Point
Theory and Appl., 1 (2007), no. 2, 305–336.
53. Matano, H., Asymptotic behavior and stability of solutions of semilinear diffusion
equations, Publ. Res. Inst. Math. Sci. 15 (1979), no. 2, 401–454.
54. Molle, R., Passaseo, D., Concentration phenomena for solutions of superlinear elliptic
problems, Ann. Inst. H. Poincaré Anal. Non Linéaire 23 (2006), no. 1, 63–84.
55. Ni, W.M. Diffusion, cross-diffusion, and their spike-layer steady states. Notices Amer.
Math. Soc. 45 (1998), no. 1, 9–18.
56. Ni, W.M., Takagi, I., On the shape of least-energy solution to a semilinear Neumann
problem, Comm. Pure Appl. Math., 41 (1991), 819–851.
57. Ni, W.M., Takagi, I., Locating the peaks of least-energy solutions to a semilinear
Neumann problem, Duke Math. J. 70, (1993), 247–281.
58. Ni, W.M., Takagi, I., Yanagida, E., Stability of least energy patterns of the shadow
system for an activator-inhibitor model. Recent topics in mathematics moving toward
science and engineering. Japan J. Indust. Appl. Math. 18 (2001), no. 2, 259–272.
59. Oh, Y.G. On positive Multi-lump bound states of nonlinear Schrödinger equations
under multiple well potentials, Comm. Math. Phys. 131, (1990), 223–253.
60. Shi, J., Semilinear Neumann boundary value problems on a rectangle. Trans. Amer.
Math. Soc. 354 (2002), no. 8, 3117–3154.
61. Strauss, W.A., Existence of solitary waves in higher dimensions. Comm. Math. Phys.
55 (1977), no. 2, 149–162.
62. Turing, A.M., The chemical basis of morphogenesis, Phil. Trans. Royal Soc. London,
Series B, Biological Sciences, 237 (1952), 37–72.
63. Wang, Z.Q., On the existence of multiple, single-peaked solutions for a semilinear
Neumann problem, Arch. Rational Mech. Anal., 120 (1992), 375–399.
64. Wei, J., On the boundary spike layer solutions of a singularly perturbed semilinear
Neumann problem, J. Differential Equations, 134 (1997), 104–133.
65. Wei J., Yang J., Concentration on Lines for a Singularly Perturbed Neumann Problem
in Two-Dimensional Domains, Indiana Univ. Math. J., to appear.
On Some Elliptic Problems
in the Study of Selfdual
Chern-Simons Vortices

Gabriella Tarantello

1 Introduction

In these lectures we use an approach introduced by Taubes (cf. [JT]) in the

study of selfdual vortices for the abelian-Higgs model, in order to describe
vortex configurations for the Chern-Simons (CS in short) theory discussed
in [D1].
Notice that the abelian-Higgs model corresponds (in a non-relativistic con-
text) to the bi-dimensional Ginzburg-Landau (GL in short) model (cf. [GL],
[DGP]), for which much has been accomplished in recent years also away
from the selfdual regime. In this respect, beside the seminal work of Bethuel-
Brezis-Helein (cf. [BBH]), we mention for example: [BeR], [JS1],[JS2], [Lin1],
[Lin2], [LR1], [LR2], [PiR], [PR] and the recent monograph by Sandier-Safarty
[SS]. However, the methods and techniques introduced for the GL-model
do not seem to apply as successfully for the CS-model (see the attemps of
Kurzke-Sprin [KS1], [KS2] and Han-Kim in [HaK]). Thus, so far a rigorous
mathematical analysis of CS-vortices has been possible only at the selfdual
regime where Taubes approach applies equally well and allows one to reduce
the vortex problem to the study of elliptic problems involving exponential
nonlinearities. In this way it has been possible to treat many relevant selfd-
ual theories of interest in theoretical physics by means of nonlinear analysis,
see [Y1].
Here, we only wish to give a general idea on the type of analytical problems
that one encounters in the study of selfdual vortex configurations, and for this
purpose we shall focus our attention only on the planar Chern-Simons-Higgs
model proposed in [JW] and [HKP], and refer the interested reader to the
monographs [Y1] and [T8] for further developments.

G. Tarantello
Universita’ di Roma Tor Vergata- Dipartimento di Matematica, via della ricerca scientifica,
00133 Rome, Italy
e-mail: [email protected]

S.-Y.A. Chang et al. (eds.), Geometric Analysis and PDEs, 117

Lecture Notes in Mathematics 1977, DOI: 10.1007/978-3-642-01674-5 4,

c Springer-Verlag Berlin Heidelberg 2009
118 G. Tarantello

The physical motivation to study CS-vortices rests upon the important

property that (contrary to GL-vortices) they carry both electric and magnetic
charge, a property observed for example in superconductivity (cf. [Ab], [Park]
and [Sch]).
On this basis, Chern-Simons theory has been used more generally to resolve
important issues in condensed matter physics, as discussed in [D1], [D2], [D3],
[Fro], [FM1] and [FM2].
The mathematical reasons to treat the CS-model rests upon the fact that
it introduces a new 6th order gauge theory, whose space of moduli is rather
rich and, as we shall see, far from being completely characterized.
We are going to compare the abelian-Higgs and Chern-Simons model and
emphasize their analogies and differences. To this purpose, we first formu-
late both models within the framework of abelian gauge field theory. For
a more detailed physical background about gauge field theories we refer to
[AH], [ChNe], [Fel], [Po] and [Q]; while an introduction on the mathemati-
cal formalism of gauge theory can be found in [Tra] and [GS], (see also [JT]
and [Jo]).
Here, we shall be concerned with a gauge theory whose gauge group G
(typically a compact Lie group) is specified by the group of rotation in R2 .
Namely, recalling that the group of rotation in Rn is given by the
orthogonal group:

O(n) = {A, n × n real matrix : At A = Id}

and

SO(n) = {A ∈ O(n) : det A = 1} (special orthogonal group)

we fix G = SO(2).

Recalling that the complex counterpart of the groups above, are given
respectively by,

U (n) = {A, n × n complex matrix : At A = Id} (unitary group)

and
SU (n) = {A ∈ U (n) : det A = 1} (special unitary group)
we see that
G = SO(2) = U (1) = {z ∈ C : |z| = 1}.
Hence G is topologically equivalent to S 1 and defines an abelian group, whose
group multiplication coincides with the multiplication of complex numbers.
In particular, the group U (1) acts on C just as a multiplication by a unitary
complex number, and in this way we obtain a unitary rapresentation for it.
Since any element of our gauge group U (1) takes the form eiω , we also see
that the corresponding Lie algebra u(1) = −iR.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 119

From the physical point of view, (cf. [AH], [ChNe], [Fel], [Po] and [Q]),
a U (1)-gauge field theory describes electromagnetic particle interaction,
while a (non abelian) SU (2)-gauge field theory concerns with weak particle
interactions and an SU (3)-gauge field theory pertains with strong particle
interactions.
The celebrated Electroweak theory of Salam-Glashow-Weinberg (cf. [La])
unifies electromagnetic and weak particle interactions via a SU (2) × U (1)-
gauge field theory, and there is hope that a general unified theory of all
particle interactions should involve the gauge group SU (5).
Notice also that the Yang-Mills (cf. [YM]) and more generally the Yang-
Mills-Higgs theory are formulated as (non-abelian) SU (2)-gauge field the-
ories. It is well known (cf. [JT], [Ra]) that both theories carry selfdual
solitons-type solutions as given by the Yang-Mills instantons in 4-space
dimension, and the Yang-Mills-Higgs monopoles in 3-space dimension (see
e.g. [ADHM], [AHS1], [AHS2], [BPST], [DK], [Wit], [AtH] and [PS]). In many
respects abelian-Higgs vortices correspond to the bi-dimensional version of
such static configurations. For example, it is known that they are in one to one
correspondence with the class of O(3)-symmetric four dimensional instantons
(cf. [JT]).
Other interesting features about vortices for the abelian-Higgs model can
be found in [Bra1], [Bra2], [Hi], [Ga1], [Ga2], [Ga3] and [NO].

2 The Abelian-Higgs Model

and The Chern-Simons-Higgs Model

We denote by R1,3 the 3-dimensional Minkowski space with metric tensor

g = tr(1, −1, −1, −1), and let (x0 , x1 , x2 , x3 ) be an element of R1,3 , so that x0
denotes the time-variable and xj , j = 1, 2, 3, the space variables. Generally,
we use greek indices to denote indifferently time and space variables. As
usual we will use the metric to rise (or lower) indices in the usual way (i.e.
3
xα = g αβ xβ , with g αβ the coefficients of the inverse of g. Also, we use
α,β=0
the standard convention that repeated lower/upper indices are summed.
The U (1)-gauge field theory over R1,3 is formulated in terms of the
following variables:

(electromagnetic) gauge potential: A = −iAα dxα , Aα : R1,3 → R, α =

0, 1, 2, 3, given by a globally defined 1-form with values in the Lie algebra
u(1) = −iR, and viewed as a connection on the (trivial) principle bundle
P = R1,3 × U (1);
120 G. Tarantello

(matter) Higgs field: ϕ : R1,3 → C, which can be seen as a global section

of the associated bundle M = R1,3 × C (relative to the above mentioned
representation of U (1) over C).

The gauge potential A and the Higgs field ϕ are weakly coupled by means
of the covariant derivative DA associated to A:

DA = Dα dxα , with Dα = ∂α − iAα , α = 0, 1, 2, 3.

So that,
Dα ϕ = ∂α ϕ − iAα ϕ, α = 0, 1, 2, 3.
Also we obtain the electromagnetic gauge field FA as the curvature corre-
sponding to A, namely:
i
FA = − Fαβ dxα ∧ dxβ
2
with
Fαβ = ∂β Aβ − ∂β Aα .
The abelian Maxwell-Higgs theory is formulated in terms of the following
lagrangian density:

q2 1
L(A, ϕ) = − Fαβ F αβ + Dα ϕDα ϕ − V (1)
4 2
where V defines the scalar potential, and q is a given parameter which relates
to the electric charge.
We obtain the Ginzburg-Landau model when (in normalized units) we
specify the scalar potential:
λ
V (|ϕ|) = (1 − |ϕ|2 )2 , (2)
8
(the well known double-well potential), with λ > 0 a given parameter.
In the bi-dimensional case, this model give rise to the abelian-Higgs model.
More in general, we shall always take V = V (|ϕ|), so that (1) is invariant
under the gauge transformations:

A → A + dω; ϕ → e−iω ϕ (3)

for any function ω : R1,3 → R sufficiently smooth.

Hence, A and ϕ are defined only up to gauge transformations, and so they
are not observable quantities. On the contrary, gauge invariant quantities (i.e.
observables) are given by the gauge field FA and the magnitude of the Higgs
field |ϕ|.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 121

The Euler-Lagrange equations corresponding to (1) are given as follows,


Dα Dα ϕ = −2 ∂V
∂ϕ (a)
(4)
q ∂ν F = 2 (ϕD ϕ − ϕD ϕ) (b)
2 μν i μ μ

and obviously they inherit the invariance under the gauge transformations
(3). Furthermore, setting:

i
Jμ = (ϕDμ ϕ − ϕDμ ϕ), μ = 0, 1, 2, 3
2
we can intepret the vector field J = (J μ )μ=0,1,2,3 as the (conserved) current
generated by the internal symmetries, with ρ = J 0 the charge density, and
→
−
j = (J k )k=1,2,3 the current density.
From (4) we can recover Maxwell’s equations. Indeed, from FA we obtain
→
− →
−
the electric field: E = (F 0j )j=1,2,3, and the magnetic field: B = (B j )j=1,2,3,
with B j = 2i εjkl Fkl ; where εjkl is the totally anti-symmetric tensor fixed by
ε123 = 1.
Therefore the four equations in (4)-(b) express the following familiar set
of Maxwell’s equations:
→
− → −
− → →
−
q2 ∇ · E = ρ & q 2 ∂0 E + j = q 2 curl B .

We can complete Maxwell’s equations, simply by virtue of Bianchi identity

which is satisfied by the curvature FA as follows:

∂ μ F αβ + ∂ α F βμ + ∂ β F μα = 0. (3-equations)

The Gauss-law of the system is obtained by taking μ = 0 in (4)-(b), and thus

given by,
q 2 ∂j F 0j = ρ. (5)

So, we can obtain the total electric charge = ρ, by integration of (5), and
by neglecting boundary terms, we find ρ = 0.
In other words, the abelian Maxwell-Higgs model can only support elec-
trically neutral configurations.
Of particular interest to us are the so called vortex-configurations, namely
the stationary solutions of (4) with finite energy. Here stationary means that
it is possible to specify a gauge map so that, after the gauge transformation
(3), we obtain ϕ and the component Aα , α = 0, 1, 2, 3 indipendent of the
time-variable x0 .
In case such gauge map can be choosen to correspond to the temporal
x0
gauge A0 = 0 (i.e. ω = A0 (t, x1 , . . . , x3 )dt), then the solution is called
static.

To compute the energy density, we write the lagrangean density more

explicitly as follows,
122 G. Tarantello

q2 2 q2 2 1
L= F0,j − Fk,j + (|D0 ϕ|2 − |Dj ϕ|2 ) − V (|ϕ|).
2 4 2
Consequently,

∂L ∂L ∂L
E= ∂0 Aμ + ∂0 ϕ + ∂0 ϕ + L
∂(∂0 Aμ ) ∂(∂0 ϕ) ∂(∂0 ϕ)
q2 2 q2 2 1
= F0,j + Fk,j + (|D0 ϕ|2 + |Dj ϕ|2 ) + V + [q 2 ∂j F 0j − ρ]A0
2 4 2
+ (total spatial divergence terms). (6)

Hence, by neglecting the total spatial divergence terms (which disappear upon
integration under natural boundary conditions) and recalling (5), we obtain
the following expression for the Ginzburg-Landau energy:

q2 2 q2 2 1 λ
EGL = F0,j + Fk,j + (|D0 ϕ|2 + |Dj ϕ|2 ) + (1 − |ϕ|2 )2 . (7)
2 4 2 8
We refer to [BBH], [PR], [SS] and reference therein, for the description of
static minimizers of (7) in various situations.
However, to obtain charged vortex configurations, as they naturally occur
for example in low critical temperature superconductivity, it is necessary to
modify the conventional Maxwell’s electrodynamics of the Ginzburg-Landau
model.
The Chern-Simons theory modifies Maxwell theory in the bi-dimensional
case, by introducing the Chern-Simons lagrangian as a lower order perturba-
tion of Maxwell lagrangean.
When we deal with a bi-dimensional theory, we can consider a gauge
trasformation which specifies A3 = 0, so that the remaining variables Aα ,
α = 0, 1, 2 and ϕ are indipendent of the x3 -variable (D3 Aα = 0 = D3 ϕ,
α = 0, 1, 2). Thus, for Aα = Aα (x0 , x1 , x2 ), α = 0, 1, 2, the Chern-Simons
lagrangean density takes the form:
1
LCS = − εαβγ Aα Fβγ , (8)
4
with εαβγ the totally anti-symmetric tensor, fixed so that ε012 = 1.
Although LCS is not invariant under the gauge transformation (3), the
corresponding Euler-Lagrange equations: Fαβ = 0, are gauge invariant.
The bi-dimensional Maxwell-Chern-Simons model is described by the
Lagrangean density: LMCS = LMH +κLCS , and defines a sensible gauge field
theory, with κ > 0 the Chern-Simons coupling parameter. More precisely we
have,

q2 2 q2 2 k 1
LMCS = F0,j − Fk,j − εαβγ Aα Fβγ + (|D0 ϕ|2 − |Dj ϕ|2 ) − V (|ϕ|) (9)
2 4 4 2
with corresponding Euler-Lagrange equation:
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 123

Dα Dα ϕ = −2 ∂V
∂ϕ (a)
2 μν κ μαβ μ (10)
q ∂ν F + 2 ε Fαβ = J (b)

where again J μ = 2i (ϕDμ ϕ − ϕDμ ϕ), μ = 0, 1, 2 defines the current density

generated by the internal symmetries.
As above, letting:
i
ρ = J0 = (ϕD0 ϕ − ϕD0 ϕ) (11)
2
the charge density, then (from (10)-(b) with μ = 0) we obtain the modified
Gauss law relative to the Maxwell-Chern-Simons model as given by,

q 2 ∂j F 0j + κF12 = ρ. (12)

From (12) we see that the Maxwell-Chern-Simons theory allows for (elec-
trically and magnetically) charged vortices. In fact, upon integration of (12),
and by neglecting as usual total spatial
divergence terms, we deduce a rela-
tion between the magnetic flux Φ = F12 and the electric charge Q = ρ as
follows,
Q = κΦ.
Due to the presence of the Chern-Simons term, the study of the field equations
(10) (with κ = 0) appears much more difficult than the abelian-Higgs field
equations (4), (namely (10) with κ = 0), for which much is known especially
in the bi-dimensional case (cf. [JT], [BBH], [PR] and [SS]).
So far, it has been possible to analyze with mathematical rigor only Chern-
Simons vortex-configurations in the selfdual regime, by means of Taubes’
approach for studying (4) with V in (2) and λ = q12 , (cf. [JT]).

3 Selfduality

As observed by Bogomolnyi [Bo], the bi-dimensional abelian-Higgs field

equations admit a first order reduction. This property is ensured since the
corresponding energy density,

q2 2 1 λ
EaH = (F + F12 2
) + (|D0 ϕ|2 + |D1 ϕ|2 + |D2 ϕ|2 ) + (1 − |ϕ|2 )2
2 0j 2 8
+ (total spatial divergence terms)

can be more conveniently rewritten by completing its square terms.

To this purpose, we introduce the following gauge invariant Cauchy-
Riemann operators:
D± ϕ = D1 ϕ ± iD2 ϕ (13)
124 G. Tarantello

and observe that

|D1 ϕ|2 + |D2 ϕ|2 = |D± ϕ|2 ± F12 |ϕ|2 + (total spatial divergence terms). (14)

So we can write:
(q2 F12 ± 2q
1
(|ϕ|2 −1))2
4 56 7
q2 2 1 1 2 1
EaH = F0j + |D0 ϕ| + [q F12 ± F12 (|ϕ| − 1) + 2 (|ϕ| − 1) ]
2 2 2 2
2 2 2 4q
 
1 1 1
+ |D± ϕ|2 + λ − 2 (|ϕ|2 − 1)2 ± F12
8 q 2
+ (total spatial divergence terms). (15)

Hence, in the critical coupling:

1
λ=
q2

(which characterizes the selfdual regime) and at fixed magnetic flux, the
minimal energy is attained by solutions of the following first order selfdual
equations:

F0j = 0, D0 ϕ = 0 (16)
D± ϕ = 0 (17)
1
F12 = ± 2 (1 − |ϕ|2 ) (18)
2q

which should be satisfied together with the Gauss law equation (5).
It is not difficult to check that solutions of the selfdual equations also
satisfy the second order abelian-Higgs field equations. Moreover, equations
(16)–(18) and (5) support static vortex configurations, since in the temporal
gauge A0 = 0, equation (16) just asserts that A1 , A2 and ϕ are indipendent
of the x0 -variable. So that (5) is automatically satisfied, and we are left to
determine ϕ = ϕ(x1 , x2 ) and Aj = Aj (x1 , x2 ) j = 1, 2, satisfying (17) and
(18), with corresponding static energy density:

± 1
Eself = ± F12 . (19)
2
We shall discuss the planar case, and we shall solve (17)–(18) in terms of
two complex functions, namely: ϕ and A1 ± iA2 : C → C.
Also note that if (A, ϕ)− is a solution of (17)–(18) with “minus-sign”, then
it can be related to a solution (A, ϕ)+ of (17)–(18) with the “plus-sign”, by
the following simple change of variables:
ϕ− (x1 , x2 ) = ϕ+ (−x1 , x2 )
A−
1 (x1 , x2 ) = −A1 (−x1 , x2 ),
+

A− −
2 (x1 , x2 ) = A1 (−x1 , x2 ).
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 125

Thus, without loss of generality, we will fucus only to study (17)–(18) with
“plus-sign”.
In this case, equation (17) gives:

D+ ϕ = D1 ϕ + iD2 ϕ = (∂1 ϕ + i∂2 ϕ) − i(A1 + iA2 )ϕ = 0. (20)

Now notice that the equation:

1
∂z ψ = (A1 + iA2 ) := A (21)
2
(∂z = 12 (∂1 +i∂2 )), is locally solvable by virtue of Poincaré ∂-lemma (cf. [JT]).
Moreover, if A1 , A2 ∈ C ∞ then also the solution ψ ∈ C ∞ .
By means of (21), we can write equation (20) as follows:

∂z e−iψ ϕ = 0

and deduce that e−iψ ϕ defines an holomorphic function. So ϕ can only admit
isolated zeroes with integral multiplicity.
We point out that the zero-set of ϕ represents a relevant gauge invariant
quantity, which is going to characterize in an important way the correspond-
ing vortex configuration.
Moreover, to fullfil the finite energy condition, we see from (19) and (18),
that we must require the condition: (1 − |ϕ|2 ) ∈ L1 (R2 ). We shall see that for
solutions of (17)–(18), this property is essentially equivalent to the following
(apparently stronger) condition:

|ϕ| → 1 & |x|β |∇ϕ| → 0 as |x| → +∞, (22)

for some β > 0 (cf. [JT]). Therefore for planar vortices, the Higgs field ϕ
can only admit a finite number of zeroes, whose total number (counted with
multiplicity) we shall call the vortex number. We shall see that the vortex
number bares a topological meaning according to which we can distinguish
vortices in homotopic classes.
To this purpose, let Z = {z1 , ..., zN } be the zero-set of ϕ, and let N be the
vortex number, so that each point in Z (known as vortex point) is repeated
according to its multiplicity. Take R > 0 sufficently large so that Z ⊂⊂ BR ,
and let ψ be a solution of (21) in BR . Then (by using complex notations) we
find,
N
e−iψ ϕ = (z − zj )eh(x) , (23)
j=1

with h holomorphic in BR . From (23) we can derive some interesting

consequences.
126 G. Tarantello

Firstly,
 d −iψ   d
1 dz (e ϕ) 1 ∂ψ 1 dz ϕ
N= iψ
dz =− dz + dz.
2πi e ϕ 2π ∂z 2πi ϕ
∂BR ∂BR ∂BR

By virtue of (22), we readily deduce that,

 " 3
 d 
 dz ϕ 
Im  → 0 as R → +∞.
 ∂BR ϕ 

While,
    
∂ψ ∂2ψ ∂A
dz = 2i dxdy = 2i dxdy,
∂z BR ∂z∂z BR ∂z
∂BR

so,       
∂ψ ∂A
Re dz = −2Im dxdy =− F12 .
∂z BR ∂z BR

Hence, by letting R → +∞, we find the flux quantization property:

 
1
F12 = N = winding number of ϕ = deg(ϕ, R2 , 0). (24)
2π R 2

In other words, for each N ∈ N a selfdual solution of (17)–(18) with vortex

number N defines a minimizer of the energy within the class of configura-
tions for which the Higgs field admits fixed Brower degree N (topological
constraint). The corresponding minimal value of the energy is exactly πN . It
is possible to characterize completely the class of such (energy minimizers)
N -vortex configurations.
To this purpose, we still use (23) to observe that


N
Δ log(|e−iψ ϕ|2 ) = 4π δz j ,
j=1

where δz denotes the Dirac measure with pole at z. On the other hand, from
(21) we see that Δ{Imψ} = F12 and so we deduce:


N
−Δ log |ϕ|2 = 2Δ{Imψ} − 4π δz j .
j=1

By means of (18) we arrive at the following elliptic equation for u = log |ϕ|2 ,

1 N
−Δu = (1 − e u
) − 4π δz j (25)
q2 j=1
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 127

satisfying:
u(x) → 0, as |x| → +∞.
Similarly, if we analyze the solutions of (17), (18) with the “minus”-sign,
then by recalling that: ∂z f = ∂z f , we can apply the argument above to ϕ
(instead of ϕ) and use −ψ (instead of ψ given by (21)), to deduce similar
properties.
In particular in this case (24) holds as follows:

1
F12 = winding number of ϕ = deg(ϕ, R2 , 0) = −N ∈ Z− ,
2π
R2

where N = deg(ϕ, R2 , 0) defines the vortex number.

Furthermore, we arrive at the same elliptic problem (25) for u = log |ϕ|2 =
log |ϕ|2 . In fact in this case we have: ΔIm(−ψ) = −F12 , and at the same time
(18) is satisfied with the “minus”-sign.
In other words, we have seen that each (static) planar selfdual config-
uration (A, ϕ), solution of (17), (18) and (22) with “plus-sign” or with
“minus-sign”, identifies an homotopy class of π1 (S 1 ) = π1 (U (1)) = Z, given
by ±N = deg(ϕ, R2 , 0), where N is the vortex number and the ± sign is cho-
sen accordingly. Recall that N counts (with multiplicity) the vortex points
(i.e. the zeroes of ϕ). We also know that (A, ϕ) corresponds to a minimum
of the energy among all static configurations (satisfying (22)) with the same
homotopy. While, u = log |ϕ|2 is a solution for (25), (26), with {z1 , . . . , zN }
the corresponding vortex points.
This provides a complete analytical description of vortices for the abelian-
Higgs model in the selfdual regime, once we take into account that the
following holds:
Theorem 1 (cf. [Ta2], [JT]) Every finite energy solution (A, ϕ) of (4) in R2
with V in (2) and λ = q12 (i.e. a planar selfdual abelian-Higgs vortex) coin-
cides with a (static) solution of (17)–(18) satisfying (22), and the following
estimate holds:
∀ε > 0, ∃Cε > 0:
1 1−ε
|F12 | + |D1 ϕ| + |D2 ϕ| ≤ (1 − |ϕ|) ≤ Cε e− q |x| , in R2 . (26)
q2
We mention that similar exponential decay estimates also hold away from
the selfdual regime, i.e. when λ = q12 , see [JT].
To complete the study of (17), (18) we observe:
Theorem 2 (cf. [Ta1], [JT]) Let u be a solution (in the sense of distribution)
of (25) and suppose that (1 − eu ) ∈ L1 (R2 ). Using complex notations set,
N
ϕ± (z) = e 2 ±i , A = −iA±
u
arg(z−zj ) α
α dx (27)
j=1
128 G. Tarantello

where
A± ± ±
0 = 0, A1 ± iA2 = −i∂± (log ϕ),

with ∂± = ∂1 ± i∂2 . Then (A, ϕ)± defines a static solution for (17), (18)
(with the ± sign chosen accordingly) and with vortex points {z1 , . . . , zN }.
Furthermore, it satisfies (26) and the following holds:
1. |ϕ| < 1 in R2 ,
2. ϕ+ (z) = O((z − zj )nj ) = ϕ− (z) as z → zj , with nj the multiplicity of zj ,
for j = 1, ..., N .
3.
 
± ±
Magnetic flux Φ = F12 = ±2πN ; Total energy E ± = πN (28)
R2 R2

Proof. Note that (27) follows easily by direct computations, while (28) is a
simple consequence of our construction. We shall give indications of (26) in
the following section and we refer to [JT] for more details.

Theorem 2 reduces the search of abelian-Higgs vortices to the study of the

(singular) elliptic problem (25) subject to the integrability condition:

(1 − eu ) ∈ L1 (R2 ).

This task will be carried out in the following section.

Next we wish to show that it is possible to specify the strenght of the

Chern-Simons field and the scalar potential in order to attain a selfdual
regime also for the Chern-Simons-Higgs model (9).
To this purpose, recalling (6) and (12), we see that the Chern-Simons
energy is given as follows (where as usual we have neglected the total spatial
divergence terms):

q2 q2 2 1
EMCS = |F0j |2 + F12 + (|D0 ϕ|2 + |D1 ϕ|2 + |D2 ϕ|2 ) + V
2 2 2
In this case, a first way to attain selfduality is described by Jackiw-
Weinberg [JW] and Hong-Kim-Pac [HKP], who propose to neglect the
Maxwell term in the energy (i.e. set q = 0), and thus consider a model
whose electrodynamics is governed solely by the Chern-Simons term.
In this situation, still neglecting total spatial divergence terms and by
recalling (14), we find:

1
ECSH = (|D0 ϕ|2 + |D1 ϕ|2 + |D2 ϕ|2 ) + V =
2
1 1 1 1
= |D0 ϕ ± iϕW (|ϕ|)|2 + |D± ϕ|2 ± F12 (|ϕ|2 − 1) + V (|ϕ|) ± F12
2 2 2 2
i 1 2 2
± (ϕD0 ϕ − ϕD0 ϕ)W (|ϕ|) − |ϕ| W (|ϕ|)
2 2
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 129

with W (|ϕ|) an arbitrary real valued function to be specified later. Hence,

using (11) and the Gauss law (12) (with q = 0) we find:

1 1 1
ECSH = |D0 ϕ ± iϕW (|ϕ|)|2 + |D± ϕ|2 ± F12 (|ϕ|2 − 1) ± κF12 W (|ϕ|)
2 2 2
1 2 2 1
+V (|ϕ|) − |ϕ| W (|ϕ|) ± F12 .
2 2
Hence to obtain an energy similar to (15), we need to require that,
1 1
(|ϕ|2 − 1) + κW (|ϕ|) = 0 & V (|ϕ|) = |ϕ|2 W 2 (|ϕ|).
2 2
Consequently (see [JW] and [HKP]) we must take:
1
κW (|ϕ|) = (1 − |ϕ|2 ),
2
and the selfdual Chern-Simons scalar potential takes the form of a “triple”
well potential given as follows:
1
VCS = |ϕ|2 (1 − |ϕ|2 )2 .
8κ2
Thus, for the selfdual Chern-Simons-Higgs model:
κ 1 1
LCSH = − εαβγ Aα Fβγ + (|D0 ϕ|2 − |D1 ϕ|2 − |D2 ϕ|2 ) − 2 |ϕ|2 (1 − |ϕ|2 )2
4 2 8κ
(29)
we obtain the following set of (time-dependent) selfdual equation:

D0 ϕ = ± κi ϕ(|ϕ|2 − 1)
(30)
D± ϕ = 0

to be satisfied together with the Gauss law constraint:

i
κF12 = ρ = (ϕD0 ϕ − ϕD0 ϕ). (31)
2
In particular, by putting together the first equation in (30) and (31) we get,

1
F12 = ± |ϕ|2 (1 − |ϕ|2 ). (32)
κ2
To obtain a Chern-Simons-Higgs vortex configuration, we first solve with
ϕ = ϕ(x1 , x2 ), Aj = Aj (x1 , x2 ), j = 1, 2, the following selfdual equations:

D± ϕ = 0
(33)
F12 = ± κ12 |ϕ|2 (1 − |ϕ|2 ).
130 G. Tarantello

whose structure has much in common with the selfdual equations of the
abelian-Higgs model.
Then we deduce a stationary solution for (30) once that we check the
following:
Proposition 3 Let ϕ̂ and  = (Â1 , Â2 )± be a solution of (33) in R2 and set
A0 = ± 2κ1
(1−|ϕ̂|2 ), Aj = Âj +(∂j A0 )x0 and ϕ = e−ix0 A0 ϕ̂. Then (A, ϕ) with
A = −iAα dxα , defines a stationary solution for the selfdual equation (30). In
particular, the corresponding electric field, magnetic field and energy density
define time-independent quantities such that the following relation holds:
  
1 1
Magnetic flux = ± Total Energy = Electric charge = F̂12 dx dx 1 2
2 κ

As above, problem (33) involves as unknowns two complex functions,

namely: ϕ, A1 ± iA2 : C → C, which however are defined only up to gauge
transformations.
To solve (33) we can use the exact same argument presented for the
abelian-Higgs model, and show that via (27) we are reduced to find solutions
for the elliptic problem:

1 u N
−Δu = 2 e (1 − e ) − 4π
u
δzj in R2 , (34)
κ j=1

where u = log |ϕ|2 , N is the vortex number and z1 , . . . , zN are the corre-
sponding vortex points (repeated with multiplicity).
The finite energy condition now requires that:

eu (1 − eu ) < +∞. (35)
R2

In principle, property (35) allows for two type of admissible boundary

conditions, namely:

u → 0 as |x| → +∞ or u → −∞ as |x| → +∞.

To have physically meaningful solutions we also need that,

u < 0 (i.e. |ϕ| < 1) in R2 . (36)

So, now it makes sense to consider the following two boundary value problems:
⎧ N
⎨ −Δu = κ2 e (1 − e ) − 4π j=1 δzj
1 u
⎪ u

(I) u < 0
⎪
⎩ (1 − eu ) < +∞
R2
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 131

and ⎧ N
⎨ −Δu = κ2 e (1 − e ) − 4π j=1 δzj
1 u
⎪ u

(II) u < 0
⎪
⎩ eu < +∞
R2

We shall see that problem (I) shares many features with the elliptic prob-
lem corresponding to the abelian-Higgs model; while problem (II) admits
features common to the singular Liouville equations (cf. [Lio]) and it will be
characterized by other interesting analytical aspects.
However, contrary to the abelian-Higgs model (see Theorem 1), it is still
open the question of whether Chern-Simons vortex configurations are fully
described (via Proposition 3) by (stationary) solutions of the selfdual equa-
tion (30) with finite energy, or, on the contrary, non-selfdual solutions exist,
as it occurs for Yang-Mills fields (cf. [Ta3]).
For the moment (still using complex notations) we can easily check the
following:
Proposition 4 For given {z1 , . . . , zN } ⊂ C, let u be a solution of (34), (35)
and (36). Then

N
2 ±i
u
arg(z−zj )
ϕ(z) = e j=1
& A1 ± iA2 = −i∂± log ϕ

defines a solution for the selfdual equations (33), with ± sign chosen accord-
ingly, such that:
(a) |ϕ| < 1, and ϕ vanishes exactly at {z1 , . . . , zN };
(b) 
F12 = ±2πN.
R2

In concluding this section, we mention that after [JW] and [HKP], many
other selfdual Chern-Simons models have been introduced, including the
Maxwell term and in the non-abelian setting, see e.g. [D1], [D2] [LLM], [CaL],
[KiKi], [Va] and references therein.
For those models it has been possible to use an analogous approach to
deduce results with a similar flavor about their vortex configurations.
In this respect, we bring to the reader’s attention the following work:
[Ch3], [ChCh1], [ChK1], [ChK2], [ChNa], [DJLW2], [Ha1], [Ha2], [Ha3],
[JoLW], [JoW1], [JoW2], [LN], [NT1], [NT3], [Ol], [Ri1], [Ri2], [Ri3], [RT1],
[SY1], [T1], [T2], [T3], [Wa] and [Y3].
132 G. Tarantello

4 On the Elliptic Problem Concerning Abelian-Higgs

Vortices

In the previous section we have seen how the selfdual vortex problem for both
abelian-Higgs and Chern-Simons-Higgs model can be reduced to the study
of elliptic problems involving exponential nonlinearities and including Dirac
measures supported at the vortex points.
In this and the following section, we show how to approach such ellip-
tic problems by means of variational techniques. To this purpose, it will be
convenient to use complex notations by the usual identification: R2 " C.
We shall focus mainly on the planar case, and study vortex configurations
in R2 . See [T2], [T3], [T8] and references therein, for the study of periodic vor-
tices (as motivated by [Ab], [Ol], [’tH1] and [’tH2]) or more generally vortices
defined on compact surfaces (as motivated in [Bra1], [Ga1] [Hi] and [KiKi]).
We start in this section to analize the elliptic problem relative to abelian-
Higgs vortices.
Hence, for a given N ∈ N and {z1 , ...zN } ⊂ C a set of N -points (not
necessarily distinct), we seek solutions u ∈ C 2 (R2 \ {z1 , ...zN }) for the
problem:

1 N
−Δu = (1 − e u
) − 4π δzj in R2 (37)
q2 j=1

(1 − eu ) ∈ L1 (R2 ) (38)

where (37) is satisfied in the sense of distributions.

Clearly,
u(z) = O(2nj log(|z − zj |)) as z → zj ,
(nj is the multiplicity of zj ), j = 1, . . . , N .
So eu ∈ C 2 (R2 ), and it vanishes exactly at each zj with multiplicity 2nj ,
for every j = 1, . . . , N .
The following holds:
Proposition 5 Let u ∈ C 2 (R2 \ {z1 , ...zN }) be a solution (in the sense of
distributions) of (37), (38). Then,
i)
u < 0, (39)
ii)
|∇u(z)| + |u(z)| → 0 as |z| → +∞ (40)
iii) 
(1 − eu ) = 4πN. (41)
R2
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 133

Proof. To establish i) we start by showing that u admits positive part u+ ≡ 0.

To this purpose, we argue by contradiction and assume that Ω+ = {z :
u(z) > 0} is non-empty. In other words, Ω+ defines a non-empty open set
disjoint from {z1 , . . . , zN }. Since 0 ≤ u+ ≤ (eu − 1) ∈ L1 (R2 ), we see that
u+ ∈ L1 (R2 ) and it satisfies:

1
−Δu+ = (1 − eu+ ) ≤ 0 in Ω+ . (42)
q2

So u+ defines a (smooth) superharmonic function in Ω+ , and we can use the

mean value theorem to check that u+ → 0 as |x| → +∞. In particular, u+
is bounded from above in Ω+ , where it attains its maximum point, say at
z0 ∈ Ω+ . Hence u+ (z0 ) > 0 and Δu+ (z0 ) ≤ 0. But this is impossible since
by (42) we find:
1
0 ≥ Δu+ (z0 ) = 2 (eu+ − 1) > 0,
q
an absurd. Hence, u+ ≡ 0, and so u ≤ 0. But at this point we can use the
strong maximum principle to conclude that u < 0 as claimed.
To establish (40), we argue again by contradiction and assume there exist:
l > 0 and a sequence {yn } ⊂ C such that, |yn | → +∞ and u(yn ) ≤ −l,
∀n ∈ N. Without loss of generality we can further assume that {z1 , . . . , zN } ∩
B1 (yn ) = ∅, ∀n ∈ N and B1 (yn ) ∩ B1 (ym ) = ∅, for n = m.
We can use Harnack estimates to find two universal constants c1 > 0 and
c2 > 0 such that,

sup u ≤ c1 inf u + c2 1 − eu L2 (B2 (yn )) , ∀n ∈ N.

B1 (yn ) B1 (yn )

Since
1 − eu L2 (B2 (yn )) ≤ 1 − eu L1 (B2 (yn )) → 0, as n → ∞,
from the inequality above, we deduce that

sup u ≤ −c1 l + o(1), as n → ∞.

B1 (yn )

Consequently, we find n0 ∈ N and a suitable constant β > 0, such that

inf (1 − eu ) ≥ β, ∀n ≥ n0 .
B1 (yn )

Since the balls B1 (yn ) are mutually disjoint, this clearly violates the interabil-
ity of (1−eu ). Thus, we must have that necessarily: lim sup u = lim inf u = 0,
|z|→+∞ |z|→+∞
that is, u → 0 as |z| → +∞.
To establish ii) we only need to use the following well known gradient
estimates for Poisson’s equation (cf. [GT]),
134 G. Tarantello
 
|∇u(y)| ≤ C sup |u| + sup |Δu| (43)
B1 (y) B1 (y)

which holds with an universal constant C > 0, provided B1 (y) does not
contain any of the vortex points zj , for every j = 1, . . . , N . Thus, for R >
max (|zj | + 2), we get,
j=1,...,N
 
1
sup |∇u| ≤ C sup |u| + sup 2
(1 − e )
u
(44)
|y|≥R |y|≥R−1 |y|≥R−1 q

which allow us to complete the proof of ii). Finally, to obtain (41), we inte-
grate (37) over the ball BR . Then by letting R → +∞, we deduce the desired
conclusion by Green-Gauss theorem and by virtue of ii).

The “quantized” value of the integral (41) is consistent with the discussion
of the previous section.
We show next that the convergence in (40) actually holds exponent-
ially fast.
Proposition 6 Let u ∈ C 2 (R2 \ {z1 , . . . , zN }) satisfy (in the sense of distri-
butions) (37) and assume that (38) holds. Then, for every 0 < ε < 1, there
exists a constsant Cε > 0:
1−ε
0 < 1 − eu ≤ Cε e− q |x| , ∀x ∈ R2 . (45)

Proof. By virtue of (40) we can fix Rε > 0 sufficiently large, so that:

q|∇u(z)| + (1 − eu(z) ) < ε, ∀z : |z| ≥ Rε . (46)

For R ≥ Rε , we introduce the function:

1−ε
ψ(z) = e q (R−|z|) − (1 − eu(z) ).

It satisfies:
" 1−ε
 
Δψ = q12 e q (R−|z|) (1 − ε)2 − (1−ε)q
|z| − e u
+ 1 u
q2 e ψ + eu |∇u|2
ψ|∂BR > 0 & ψ → 0 as |z| → +∞.

We claim that,
ψ ≥ 0, ∀|z| ≥ R.
Indeed, if on the contrary we suppose inf |z|≥R ψ < 0, then ψ would assume
its minimum value at a point, say z0 , where we have:
1 − ε 1−ε z0
ψ(z0 ) < 0 ≤ Δψ(z0 ) & ∇ψ(z0 ) = − e q (R−|z0 |) +eu(z0 ) ∇u(z0 ) = 0
q |z0 |
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 135

Consequently,

0 ≤ Δψ(z0 ) =
 
1 1−ε
q (R−|z0 |)
(1 − ε)q z0
e (1 − ε)2
− − e u(z0 )
+ (1 − ε)q .∇u(z 0 )
q2 |z0 | |z0 |
1 1 1−ε
+ 2 eu(z0 ) ψ(z0 ) ≤ 2 e q (R−|z0 |) (−2ε + ε2 + q|∇u(z0 )| + (1 − eu(z0 ) )) < 0,
q q
a contradiction. Thus we have established that:
1−ε
∀ε > 0, ∃Rε > 0 : ∀R ≥ Rε , 0 < 1 − eu(z) ≤ e q (R−|z|) , (47)
1−ε
and clearly (45) holds with Cε = e q Rε .

By combining (43) and (45), we conclude:

Corollary 7 Let u ∈ C 2 (R2 \ {z1 , . . . , zN }) satisfy (37) (in the sense of
distributions) and assume that (38) holds. For every ε ∈ (0, 1) and δ > 0
there exists a constant Cε,δ > 0 such that,

8
N
− 1−ε
q |x|
|∇u| + |u| ≤ Cε,δ e , ∀x ∈ R \
2
Bδ (zj ).
j=1

By the way we have obtained the decay estimates above, it may seem that
all constants there involved depend on the given solution. As a matter of fact,
it is worth to note that the above estimates hold uniformly with respect to
the solution and the given data (namely, the vortex points and the parameter
q > 0) as follows:
Lemma 8 Given ε ∈ (0, 1), N ∈ N and ρ > 0, there exist Rε = Rε (N, ρ) > 0
such that every solution u of (37), (38) with max |zj − zk | ≤ ρ and
j,k=1,...,N
0 < q ≤ 1 satisfies:
⎛ ⎞
8
N
0 < 1 − eu(z) < ε, ∀z ∈ R2 \ ⎝ Bδ (zj )⎠
j=1

for every δ ≥ Rε .
Proof. By means of a translation, we can always be reduced to show the
following:
Claim: if |zj | ≤ ρ for every j = 1, . . . , N , then

0 < 1 − eu(z) < ε, ∀|z| ≥ Rε .

To establish the Claim, we argue by contradiction and assume there exist

ε0 ∈ (0, 1), a solution un of (37) and (38) with zj = zjn and q = qn , satisfying:
136 G. Tarantello

0 < qn ≤ 1 and |zjn | < ρ for every j = 1, . . . , N and ∀n ∈ N; such that, for
suitable Rn → +∞ and yn ∈ R2 \ B2Rn we have,

(1 − eun (yn ) ) ≥ ε0 .

Since eun (x) → 1, as |x| → +∞, we can actually claim that for every 0 < ε ≤
ε0 , there exist a sequence xn ∈ R2 \ B2Rn such that: eun (xn ) = 1 − ε. Observe
xn −z n
that qn j ∈ BRn (0), ∀j = 1, . . . , N and ∀n ∈ N sufficiently large. Thus,
setting Un (x) = un (xn + qn x), x ∈ BRn (0); we see that, ∀n ∈ N sufficiently
large, Un satisfies:
⎧
⎪ −ΔUn = 1 − eUn , in BRn (0)
⎨
− ε) U
Un (0) = log(1 (48)
⎪
⎩ U n < 0; (1 − e n ) ≤ 4πN.
BRn (0)

By Harnack estimates, we deduce that Un is uniformly bounded on compact

sets of R2 . Hence by elliptic regularity theory and a diagonalization proce-
dure, we find a subsequence (denoted the same way), such that Un → U in
2,α
Cloc (R2 ) (for some α ∈ (0, 1)) and U satisfies:
⎧
⎨ −ΔU = 1 − e in R
U 2
⎪
U (0) = log(1 − ε), U < 0 (49)
⎪
⎩ (1 − e U
) ≤ 4πN
R2

As in Proposition 5 we check that U (x) → 0 as |x| → +∞, and U ≡ 0. So

(being U regular) it must attain its minimum value at a point x0 ∈ R2 and
there holds: U (x0 ) < 0 ≤ ΔU (x0 ) = eU(x0 ) − 1 < 0, an absurd.
Remark 9 Notice that, by the analysis above, we can also treat the case
q > 1 via the transformations:
zj
u(z) → u(qz); zj → j = 1, . . . , N, (50)
q
which allows one to reduce the case q > 1 to the case q = 1.

Since Rε > 0 in Lemma 8 is indipendent of q, we can argue as in Proposi-

tion 6 and (by means of (44)) obtain the following uniform exponential decay
estimates:

Corollary 10 Given ε ∈ (0, 1), N ∈ N and ρ > 0, there exist R̂ε =

R̂ε (N, ρ) > 0 and a constant Cε = Cε (N, ρ) > 0 such that every solution u
of (37), (38) with max |zj | ≤ ρ and 0 < q ≤ 1 satisfies:
j=1,...,N
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 137
1−ε
0 < 1 − eu ≤ Cε e− q |x| , ∀|x| ≥ R̂ε ; (51)
− 1−ε
q |x|
|∇u| + |u| ≤ Cε e , ∀|x| ≥ R̂ε . (52)

Next, we direct our attention to show that problem (37), (38) is uniquely
solvable. To this purpose, by scaling:
zj
u(z) → u(qz); zj → j = 1, . . . , N, (53)
q
we can always assume that,
q = 1.
For fixed μ > 0 sufficiently large (to be specified below), we consider the
function:
N
|z − zj |2
u0 (z) = log (54)
j=1
μ + |z − zj |2

Hence,  
1
u0 < 0 in R2 , and u0 (z) = O as |z| → +∞.
|z|2
So, u0 ∈ Lp (R2 ), ∀ 1 < p < +∞; while u0 ∈ L1loc (R2 ) and it satisfies (in the
sense of distributions):


N
−Δu0 = g0 − 4π δz j (55)
j=1

with

N
1
g0 (z) = 4μ ∈ L1 (R2 ) ∩ L∞ (R2 ). (56)
j=1
(μ + |z − zj |2 )2

Since g0 L2 (R2 ) → 0 as μ → +∞, from now on we shall fix μ > 0 so that,

1
g0 L2 (R2 ) < . (57)
2
Set
u = u0 + v. (58)
We have:
Proposition 11 According to the decomposition (58), u ∈ C 2 (R2 \ {z1 , . . . ,
zN}) satisfies: ⎧
⎨ 
N
−Δu = 1 − eu − 4π δzj , in R2
(59)
⎩ j=1
(1 − e ) ∈ L (R )
u 1 2

if and only if v satisfies:

138 G. Tarantello

−Δv = 1 − eu − g0 , in R2
(60)
v ∈ H 1 (R2 ).

Proof. If u ∈ C 2 (R2 \ {z1 , . . . , zN }) satisfies (59), then v in (58) defines a

smooth solution of (60) where both functions eu0 and g0 ∈ C ∞ (R2 ). Fur-
thermore, v → 0 as |z| → +∞, and v ∈ Lp (R2 ), ∀p > 1. In particular
v ∈ L2 (R2 ).
Denote by χ ∈ C0∞ (R2 ) a standard cut off function, so that 0
≤ χ ≤ 1,
χ|B1 (0) ≡ 1 and χ ≡ 0 in R2 \ B2 (0). For R > 0, set χR (z) = χ R z
. We use
vχR as a test function in the equation (60) to obtain,
    
− (Δv)vχR = (1 − e )vχR − g0 vχR → (1 − e )v − g0 v,
u u

R2 R2 R2 R2 R2

as R → +∞, which follows by Lebesgue dominated convergence theorem.

On the other hand,
  
− (Δv)vχR = |∇v| χR + v∇v · ∇χR ,
2

R2 R2 R2

and
   
    
  1 
 v∇v · ∇χR  =  ∇v 2 · ∇χR  ≤ 1 v 2 |ΔχR | =
  2  2
2  2 
R R R≤|x|≤2R

1
= max v 2 |Δχ| → 0 as R → ∞.
2 R≤|x|≤2R
1≤|x|≤2

Consequently, |∇v| ∈ L (R ) and so v ∈ H 1 (R2 ), as claimed.

2 2

Viceversa, assume that (60) holds, namely v ∈ H 1 (R2 ) and

  
∇v · ∇ϕ = (1 − eu )ϕ − g0 ϕ, ∀ϕ ∈ H 1 (R2 ). (61)
R2 R2 R2

By well known elliptic estimates, it follows that v ∈ C ∞ (R2 ) and v → 0

as |z| → +∞. Consequently, u ∈ C ∞ (R2 \ {z1 , ..., zN }) and u(z) → 0, as
|z| → +∞. So (as in Proposition 5), we see that u < 0. Furthermore, if we
use (61) with ϕ = χR , arguing as above, we find that (1 − eu ) ∈ L1 (R2 ) as
claimed.

The advantage of introducing the equivalent problem (60) is that it admits

a variational formulation.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 139

To this purpose, recall that if v ∈ H 1 (R2 ), then (ev − 1 − v) ∈ L1 (R2 )

and (ev − 1) ∈ Lp (R2 ), ∀p ≥ 2. The proof of those integrability properties
can be found in [JT], [Y1] and [T8], and follows by an approach of Trudinger
[Tr] introduced to show similar integrability properties on bounded domains
via an inequality that subsequently has been re-derived in its sharp form by
Moser in [Mo].

Remark 12 Incidentally, let us mention that many other sharp versions of

the Moser-Trudinger inequality are now available over manifolds and in the
context of systems, (see e.g. [Ad], [Au], [Ban], [Be], [ChY3], [Che], [CSW],
[DET], [Fo], [JoW1], [JoLW], [NT2], [On], [SW1], [SW2] and [W]). All those
inequalities have proved extremely usefull in the study of related variational
problems.

Observing that 0 ≤ eu0 < 1 and that (1 − eu0 ) ∈ L2 (R2 ), we can consider
the functional:
  
1
J(v) = |∇v| + e (e − 1 − v) + (eu0 − 1 + g0 )v,
2 u0 v
(62)
2
R2 R2 R2

∀v ∈ H 1 (R2 ).
It is easy to check that J ∈ C 1 (H 1 (R2 )), and every critical point of J sat-
isfies (61). Therefore, by well known elliptic regularity theory, critical points
of J define solutions for (60).
To obtain critical points for J we prove the following:
Proposition 13 The functional J is (strictly) convex and satisfies:

J(v) ≥ av − b, ∀v ∈ H 1 (R2 ), (63)

for suitable a, b > 0.

Proof. The convexity of J easily follows once we observe that
  
1
J(v) = |∇v|2 + eu0 (ev − 1) + (g0 − 1)v.
2
R2 R2 R2

To establish (63), we use the inequality:


v2
v2L2 (R2 ) ≤ 2(1 + 2∇v2L2 (R2 ) ) , ∀v ∈ H 1 (R2 ) (64)
(1 + |v|)2
R2
140 G. Tarantello

which can be checked as follows:

⎛ ⎞ 12 ⎛ ⎞ 12
  2 
v v
v2L2 (R2 ) = v(1 + |v|) ≤ ⎝ ⎠ ⎝ v 2 (1 + |v|)2 ⎠ ≤
1 + |v| (1 + |v|)2
R2 R2 R2
⎛ ⎞ 12 ⎛ ⎞ 12
 
v2
≤⎝ ⎠ ⎝2 (v 2 + v 4 )⎠ ≤
(1 + |v|)2
R2 R2
⎛ ⎞ 12
 √
v2
≤⎝ ⎠ 1
2vL2 (R2 ) (1 + 2∇v2L2 (R2 ) ) 2 ,
(1 + |v|)2
R2

where we have used the well known Sobolev inequality:


v 4 ≤ 2v2L2 (R2 ) ∇v2L2 (R2 ) .
R2

Furtheremore, if we write v = v+ − v− (v± respectively the positive and

negative part of v) then v± ∈ H 1 (R2 ) and J(v) = J(v+ ) + J(−v− ). So it
suffices to check (63) under the assumption that v does not change sign.

Case 1. v ≤ 0 a.e. in R2 .

We have:
    
1
J(v) = |∇v|2 + (eu0 −1−u0)(ev −1)+ (ev −1−v)+ u0 (ev −1)+ g0 v.
2
R2 R2 R2 R2 R2

|t|
Observing that (eu0 − 1 − u0 ) ≥ 0, we use the inequality: 1+|t| ≤ |et − 1| to
estimate:
   
1 |v| v2
J(v) ≥ |∇v|2 − (eu0 − 1 − u0 ) + + g0 v ≥
2 1 + |v| 1 + |v|
R2 R2 R2 R2
⎛ ⎞ 12⎛ ⎞ 12
    
1 v2 v2
≥ |∇v|2 − ⎝ ⎠ ⎝ (eu0 −1−u0)2 ⎠ + + g0 v ≥
2 (1+|v|)2 (1+|v|)2
R2 R2 R2 R2 R2
  2   2 
1 1 v 1 v
≥ |∇v|2 − − (eu0 −1−u0)2 + + g0 v ≥
2 2 (1+|v|)2 2 (1+|v|)2
R 2 R 2 R 2 R 2 R2
  2 
1 1 v
≥ |∇v|2 + + g0 v − C1 ,
2 2 (1 + |v|)2
R2 R2 R2
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 141

with C1 > 0 a suitable constant. Thus, by using (64) we find:

 v2L2 (R2 ) 
1
J(v) ≥ |∇v| +
2
+ g0 v − C1 ≥
2 4(1 + 2∇v2L2 (R2 ) )
R2 R2
  
1 vH 1 (R2 )
2
∇vL2 (R2 )
2
≥ ∇v2L2 (R2 ) + − + g0 v−C1 ≥
2 2(1+2∇v2L2(R2 ) ) 4(1+2∇v2L2(R2 ) )
R 2
 
1 v 2
H 1 (R2 ) 1
≥ ∇v2L2 (R2 ) + − g0 L2 (R2 ) vL2 (R2 ) − − C1
2 2(1 + 2∇v2L2 (R2 ) ) 8
1
≥ vH 1 (R2 ) − g0 L2 (R2 ) vH 1 (R2 ) − C,
2
with C > 0 a suitable constant. Recalling (57), we conclude that (63) holds
in this case.

Case 2. v ≥ 0 a.e. in R2 .

We have:
   
1
J(v) = |∇v|2 + eu0 (ev − 1 − v) − (1 − eu0 )v + g0 v ≥
2
R2 R2 R2 R2
   
1 1
≥ |∇v|2 + eu0 v 2 − (1 − eu0 )v + g0 v.
2 2
R2 R2 R2 R2

Since u0 → 0 as |x| → ∞, we fix R ≥ 1 large enough so that eu0 ≥ 1

2 for
x ∈ BRc
= R2 \ BR . Set:

1
M := vdx ≥ 0
πR2
BR

the average of v over BR . We estimate,

   
1 1
J(v) ≥ |∇v|2 + eu0 (v−M +M )2 − (v−M )(1−eu0 )−M (1−eu0 )+
2 2
BR BR BR BR
  
1 1
+ |∇v|2 + v 2 − 1 − eu0 L2 (BRc ) vL2 (BRc ) + g0 v ≥
2 4
c
BR c
BR R2
142 G. Tarantello
   
1 1 M2
≥ |∇v| + 2
e (v − M ) + M u0
e (v − M ) + 2
eu0− u0
2 2 2
BR BR BR BR

− v − M L2 (BR ) 1 − e L2 (BR ) − M (1 − e )+
u0 u0

BR
  
1 1
+ |∇v|2 + v 2 − 21 − eu0 2L2 (B c ) + g0 v.
2 8 R
c
BR c
BR R2

By recalling the Poincaré inequality,

⎛ ⎞1 ⎛ ⎞ 12 ⎛ ⎞ 12
   2 2  
⎝ 1
v ⎠ =⎝ (v − M ) ⎠ ≤ 4R ⎝ |∇v|2 ⎠
2
v−
πR2 BR
BR BR BR
(65)
we get:
   
1 1 M2
J(v) ≥ |∇v| + 2
e (v − M ) + M
u0 2
e (v − M ) +
u0
eu0 −
2 2 2
BR BR BR BR

−4R1 − eu0 L2 (BR ) ∇vL2 (BR ) − M (1 − eu0 )+
BR
  
1 1
+ |∇v| +
2
v − 21 −
2
eu0 2L2 (B c ) + g0 v. (66)
2 8 R
c
BR c
BR R2

Minimizing the expression:

⎡ ⎤
  
⎣ u0 ⎦ t2
t e (v − M ) − (1 − e ) +
u0
eu0 ,
2
BR BR BR

with respect to t ∈ R, from (66) we obtain:

 
1 1
J(v) ≥ |∇v|2 + eu0 (v − M )2
2 2
BR BR
/ 02
1 eu0 (v − M ) − BR (1 − eu0 )
BR
− −
2 u0
BR e
 
1 1
− 4R1 − eu0 L2 (BR ) ∇vL2 (BR ) + |∇v|2 + v2 −
2 8
c
BR c
BR

− 21 − eu0 2L2 (B c ) + g0 v ≥
R

R2
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 143
 2
  eu0 (v − M )
1 1 1 BR
≥ |∇v|2 + eu0 (v − M )2 − −
4 2 2 BR
eu0
BR BR
 2
− u0
BR e (v − M ) BR (1 − e )
1 BR (1 e ) u0 u0
− + +
2 u0 u0
BR e BR e
  
1 1
+ |∇v| +
2
v + g0 v − C,
2
(67)
2 8
c
BR c
BR R2

with a suitable constant C = C(R) > 0. On the other hand, we observe that:
 2
 e u0
(v − M )
BR
eu0 (v − M )2 − ≥ 0,
u0
BR e
BR

while, via Poincaré inequality, we estimate,

 
 u0 
 BR e (v − M ) BR (1 − e ) 
u0
 ≤
 BR e
u0 
  12
BR
e2u0 BR
(1 − eu0 )   12
1

(v − M )2
≤ |∇v|2 + C,
e u0 8
BR B R
BR

with another suitable constant C = C(R) > 0. Thus we have shown that if
v ∈ H 1 (R2 ) and v ≥ 0 a.e. in R2 then
   
1 1 1
J(v) ≥ |∇v| +2
|∇v| +
2
v + g0 v − C
2
8 2 8
BR c
BR c
BR R2
  
1 1
≥ |∇v|2 + v2 + g0 v − C,
8 8
R2 c
BR R2

with a suitable constant C > 0. At this point, to verify (63), we observe that
both g0 and v are non-negative, so that
  
g0 v ≥ inf g0 v.
BR
BR BR

By the definition of g0 , we see that inf g0 ≥ c0

R4 with c0 > 0 a suitable
BR
constant indipendent of R. So, using Poincaré inequality, we get

c0

g0 v ≥ 3 vL2 (BR ) − 4R∇vL2 (BR ) .
R
BR
144 G. Tarantello

By inserting this inequality in the estimate above, we readily check that

inequality (63) holds with suitable constants a > 0 and b > 0.

As an immediate consequence of Proposition 13 we have:

Corollary 14 The functional J is bounded from below and admits a unique
critical point given by its minimum point. In particular, for every
{z1 , . . . , zN }, problem (59) admits a unique solution, which we denote by
u = u(z1 ,...,zN ) .
Proof. The strict convexity implies that J is weakly lower semicontinuos anb
admits at most one critical point. On the other hand, by (63), we also know
that J is coercive and bounded from below. Hence J attains its minimum
value at a point that gives its only critical point, and defines the unique
solution of (59).

Observe that the operator L : −Δ + eu(z1 ,...,zN ) : L2 (R2 ) → L2 (R2 ) is

strictly positive defined, as we have:

< Lϕ, ϕ >L2 (R2 ) = ∇ϕ2L2 (R2 ) + eu(z1 ,...,zN ) ϕ2 > 0, for ϕ = 0.
R2

Furthermore, due to decay rates established in Proposition 6, we see that

L defines a Fredholm operator of index zero. So L is invertible, and we can use
the Implicit Function Theorem, to see that, for every {z1 , ..., zN } ∈ R2N /ΓN
(ΓN = group of permutation of N -elements) the map:

{z1 , ..., zN } → u(z1 ,...,zN )

is smoothly defined. Moreover, we can use Corollary 10, to deduce the

following uniform decay estimates:
Corollary 15 ∀ε > 0, N ∈ N and ∀ρ > 0, there exist Rε = Rε (N, ρ) > 0
and a constant Cε = Cε (N, ρ) > 0 such that, for every {z1 , ...zN } ⊂ Bρ , we
have:
1−ε
0 < 1 − eu(z1 ,...,zN ) ≤ Cε e− q |x| in R2 \ BRε , (68)
− 1−ε
q |x|
|∇u(z1 ,...,zN ) | + |u(z1 ,...,zN ) | ≤ Cε e , ∀|x| ≥ Rε . (69)

Combining all the results above, we can conclude the following about
selfdual abelian-Higgs vortices.
Theorem 16 Every selfdual abelian-Higgs vortex (A, ϕ) (i.e. a smooth
(static) solution of (4) with V in (2) and λ = q12 , having finite energy),
admits a finite number of vortex points (i.e. zeroes of the Higgs field) with
integral multiplicity.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 145

Moreover, for any integer N ∈ N, the family of selfdual abelian-Higgs

N -vortices, having N as the total number of vortex points (counted with mul-
teplicity), is described by a family-pair of solutions {(A, ϕ)± } parametrized
by the manifold R2N /ΓN and satisfying:
(a) |ϕ| < 1 in R2 , and |ϕ| satisfies the exponential decay estimates (26);
(b)
deg(ϕ+ , R2 , 0) = N = deg(ϕ− , R2 , 0)
and (A, ϕ)+ (respectively (A, ϕ)− ) minimizes the energy among all con-
figurations for which ϕ (resp. ϕ) admits fixed Brower degree N .
(c) the magnetic flux Φ = ±2πN , the total energy E = πN .
More precisely the following holds:

for every {z1 , . . . , zN } ⊂ R2N /ΓN there exists a unique pair of N -vortices
(A, ϕ)± (up to gauge transformations) such that,

ϕ+ (z) = O((z − zj )nj ) = ϕ− (z), as z → zj ; (70)

with nj the multiplicity of zj , j = 1, ...N . In particular, ϕ± vanishes exactly

at {z1 , ...zN } (with the given multiplicity) and depends smoothly on those
points.

We shall see next that the Chern-Simons-Higgs model described above

carries a much richer vortex structure for which such a complete description
is not yet available.

5 On the Elliptic Problem of Selfdual Chern-Simons

Vortices

In this section we are concerned with the elliptic problem that describes
planar selfdual Chern-Simons vortices. Hence for given N ∈ N and {z1 , ..., zN }
⊂ R2 , (not necessarely distinct) we aim to describe solutions of the following:

1 u N
−Δu = e (1 − e u
) − 4π δz j (71)
κ2 j=1

eu (1 − eu ) ∈ L1 (R2 ) (72)
where κ > 0 and u ∈ C (R \ {z1 , . . . , zN }) satisfies (71) in the sense of
2 2

distributions.
146 G. Tarantello

To emphasize right away the difference with problem (37), (38) we have:
Proposition 17 Let u ∈ C 2 (R2 \ {z1 , . . . , zN }) be a solution (in the sense
of distributions) of (71), (72). Then,
i) u < 0 in R2 ,
ii) either u(x) → 0 as |x| → +∞, or u(x) → −∞ as |x| → +∞.
Proof. Property i) follows essentially as in Proposition 5. In fact, since: 0 ≤
u+ ≤ eu+ − 1 ≤ eu+ (eu+ − 1), we see that u+ ∈ L1 (R2 ). Consequently, we
can show that, u+ → 0, as |x| → +∞. Thus, if u+ ≡ 0, then it would admit
a maximum point, say at x0 , with u+ (x0 ) > 0, and,

1 u+ (x0 ) u+ (x0 )
0 ≥ Δu+ (x0 ) = e (e − 1) > 0,
κ2
which is clearly impossibile.
To establish ii) observe that if

lim inf u(x) = 0,

|x|→∞

then the first alternative would hold, since u < 0, and so necessarily also:
lim sup u(x) = 0. Hence, assume that,
|x|→∞

lim inf u(x) < 0.

|x|→∞

We prove that in this case we have: lim sup u(y) = lim inf u(y) = −∞. Argue
|y|→∞ |y|→∞
by contradiction and suppose there exists a sequence {yn } such that |yn | → ∞
and u(yn ) → −l, with l > 0. By taking a subsequence if necessary, we can
always assume that:
B1 (yh ) ∩ B1 (yk ) = ∅, (73)
for h = k. We use Harnack inequality to find suitable universal constants
α > 0 and β > 0 such that (for n sufficiently large) we have:

sup u ≤ α inf u + βeu (1 − eu )L2 (B2 (yh )) .

B1 (yh ) B1 (yh )

Consequently:
l
inf u ≥ − + o(1),
B1 (yh ) α

sup u ≤ −lα + o(1),

B1 (yh )

as h → ∞. Hence, for h sufficiently large, we would find a suitable constant

c0 > 0 such that:
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 147

eu (1 − eu ) ≥ c0 ,
B1 (yh )

which is impossible, since in view of (73), this would contradict the integra-
bility of eu (1 − eu ) in R2 .

Thus, it makes good sense to consider two types of solutions, namely:

(i) the topological solutions which satisfy (71)–(72) subject to the boundary
conditions:
u → 0 as |x| → +∞;
(ii) the non-topological solutions which satisfy (71)–(72) under the boundary
conditions:
u → −∞ as |x| → +∞.
The terminology topological/non-topological solution is justified by the dis-
cussion of the previous section, since only in the first case we can assign to the
vortex number N a topological meaning, that will distinguish Chern-Simons
vortex configurations into homotopic classes.
Actually, as we shall see, topological Chern-Simons vortices will have much
more in common with abelian-Higgs vortices.

5.1 Topological Chern-Simons Voritces

1
Let λ = κ2 , we devote this section to the solvability of the problem:
⎧
⎨ 
N
−Δu = λeu (1 − eu ) − 4π δz j
(74)
⎩ j=i
u → 0 as |x| → +∞

Arguing as in Section 1.3, we easily check that, any solution u ∈ C 2 (R2 \

{z1 , . . . , zN }) of (74) satisfies: u < 0 in R2 , eu (1−eu ) ∈ L1 (R2 ) and λ eu (1−
R2
eu ) = 4πN . In fact most of the properties discussed for the abelian-Higgs
vortices remain valid here. As for example the following uniform estimates
(to be compared with Lemma 8):
Lemma 18 Given ε ∈ (0, 1), N ∈ N and ρ > 0, there exist Rε =
Rε (N, ρ) > 0, such that every solution u of (74) with λ ≥ 1 and max |zj −
j,k=1,...,N
zk | ≤ ρ, satisfies:
⎛ ⎞
8
N
0 < 1 − eu(z) < ε, ∀z ∈ R2 \ ⎝ Bδ (zj )⎠ (75)
j=1

for every δ ≥ Rε .
148 G. Tarantello

Proof. Exactly as in the proof of Lemma 8, we use an argument by contradic-

tion to find ε0 ∈ (0, 1) such that, for every 0 < ε ≤ ε0 , we obtain a solution
U = Uε of the following problem:
⎧
⎨ −ΔU = e (1 − e ), in R
U U 2
⎪
U (0) = log(1 − ε), U < 0 (76)
⎪
⎩ eU (1 − eU ) ≤ 4πN
R2

Hence, we can argue as in Proposition 17, to check that either U (x) → 0 or

U (z) → −∞, as |z| → +∞. In the first case, we easily reach a contradiction,
since we find ( as in Lemma 8), that the only solution of (76) such that:
U → 0 at infinity, is given by the trivial solution U ≡ 0. Thus , U → −∞,
as |z| → +∞. According to a result of Spruck-Yang, (cf. [SY1]), U must be
radially symmetric about a point, say x0 ∈ R2 , corresponding to its unique
maximum point. Hence for V (z) = U (z + x0 ), we see that V = V (|x|) is
radially symmetric about the origin and satisfies:
⎧
⎪
⎪ −(Vrr + 1r Vr ) = eV (1 − eV ), r > 0
⎨
0 > V (0) = maxR2 V := sε ≥ log(1 − ε), V (r) → −∞ as r → +∞
(77)
⎪ V (r)
∞
⎪
⎩ e (1 − eV (r) )rdr ≤ 2N.
0

A detailed analysis of such radial problem has been carried out by Chan-
Fu-Lin in [CFL]. Thus, according to [CFL], we know that for s < 0, the
solution v(r, s) of the differential equation in (77) satisfying: v(0, s) = s and
vr (0, s) = 0, is unique and globally defined ∀r > 0. Moreover:

v(r, s) < 0 & v(r, s) → −∞ as r → +∞.

Furthermore, ev(r,s) (1 − ev(r,s) )r ∈ L1 (R+ ) and letting:

 ∞
β(s) = ev(r,s) (1 − ev(r,s) )rdr. (78)
0

then,
β(s) → +∞ as s → 0− . (79)
Consequently, V (r) = v(r, sε ) and so,
 ∞
β(sε ) = eV (r) (1 − eV (r) )rdr ≤ 2N.
0

But since: sε → 0− , as ε → 0+ , this is in contradiction with (79).

Similarly to the previous section, from Lemma 18 we can deduce the

following uniform decay estimates:
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 149

Corollary 19 For every ε ∈ (0, 1), N ∈ N and for every ρ > 0, there exist
Rε = Rε (ρ, N ) > 0 and Cε = Cε (ρ, N ) > 0, such that every solution u of
(74) with λ ≥ 1 and max |zj | ≤ ρ, satisfies:
j=1,...,N
√
0 < 1 − eu < Cε e− λ(1−ε)|x|
, ∀x : |x| ≥ Rε . (80)

√
|u(x)| + |∇u(x)| ≤ Cε e− λ(1−ε)|x|
, ∀x : |x| ≥ Rε . (81)
In particular, for fixed {z1 , . . . , zN } all corresponding solutions admits
uniform exponential decay.

At this point we turn to discuss the solvability of (74). We proceed as for

the abelian-Higgs case and let,

u = u0 + v (82)

where u0 is defined in (54), satisfies (55) with g0 given in (56).

In analogy to Proposition 11 we find:
Proposition 20 According to the decomposition (82), we have that u ∈
C 2 (R2 \ {z1, . . . , zN }) satisfies (74) (in the sense of distributions) if and only
if v satisfies (in the classical sense)

−Δv = λeu0 +v (1 − eu0 +v ) − g0
(83)
v ∈ H 1 (R2 )

and, 
λ eu0 +v (1 − eu0 +v ) = 4πN. (84)
R2

Since Proposition 20 follows exactly by the same arguments of Proposi-

tion 11, we leave the details of its proof to the interested reader.

Again, the advantage of working with problem (83) is that it admits a

variational formulation, in the sense that (weak) solutions of (83) correspond
to critical points for the functional
  
1 λ
Iλ (v) = |∇v|2 + (eu0 +v − 1)2 − g0 v, v ∈ H 1 (R2 ).
2 2
R2 R2 R2

Indeed, by recalling the integrability properties of (1 − eu0 ), and of (1 − ev )

for v ∈ H 1 (R2 ), (cf. [JT], [Y1]and [T8]), we check that Iλ ∈ C 1 (H 1 (R2 )) and
  

< Iλ (v), ϕ >= ∇v · ∇ϕ + λ (e u0 +v
− 1)e u0 +v
ϕ − g0 ϕ, ∀ϕ ∈ H 1 (R2 ).
R2 R2 R2
150 G. Tarantello

Thus, by means of elliptic regularity theory, we find that

v ∈ H 1 (R2 ) : Iλ (v) = 0 ⇐⇒ v ∈ C ∞ (R2 ) satisfies (83).

In complete analogy to the abelian-Higgs case we have:

Proposition 21 Let λ > 0. There exists a, b > 0 such that

Iλ (v) ≥ avH 1 (R2 ) − b, ∀v ∈ H 1 (R2 ). (85)

|t|
Proof. Using again the estimate 1+|t| ≤ |et − 1|, ∀t ∈ R, we find that
 
1 v2
(1 − ev+u0 )2 dx ≥ dx − 4u0 2L2 (R2 ) .
2 (1 + |v|)2
R2 R2

Consequently, by using (64), for any given ε > 0, we have:


1 λ v2
Iλ (v) ≥ ∇vL2 (R2 ) +
2
− 2λu0 2L2 (R2 ) − g0 L2 (R2 ) vL2 (R2 ) ≥
2 4 (1 + |v|)2
R2

1 λ v2
≥ ∇v2L2 (R2 ) + − 2λu0 2L2 (R2 ) −
2 4 (1 + |v|)2
R2
⎛ ⎞ 12
 2
v
− g0 L2 (R2 ) ⎝2 ⎠ (1 + 2∇v2L2 (R2 ) ) 12 ≥
(1 + |v|)2
R2

1 λ v2
≥ ∇v2L2 (R2 ) + − 2λu0 2L2 (R2 ) −
2 4 (1 + |v|)2
R2

1 v2 ε
− g0 2L2 (R2 ) − (1 + 2∇v2L2(R2 ) ) ≥
ε (1 + |v|)2 2
R2
   
1 λ 1 v2
≥ − ε ∇vL2 (R2 ) +
2
− g0 L2 (R2 )
2
2 4 ε (1 + |v|)2
R2
ε
− 2λu0 2L2 (R2 ) − .
2
Recalling that g0 L2 (R2 ) converges to zero as μ → ∞, we can choose
μ = μλ large enough to ensure that g0 2L2 (R2 ) < 16λ
. Hence, we can fix ε = 14
and obtain two suitable constants C1 > 0 and C2 > 0 such that:
⎛ ⎞
 2
v
Iλ (v) ≥ C1 ⎝∇v2L2 (R2 ) + ⎠ − C2 ≥
(1 + |v|)2
R2
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 151
 
v2L2 (R2 )
C1 ∇v2L2 (R2 ) + − C2 .
2(1 + 2∇v2L2(R2 ) )

To complete the proof, let σ ∈ [0, 1] be such that: ∇v2L2 (R2 ) = σv2H 1 (R2 )
and v2L2 (R2 ) = (1−σ)v2H 1 (R2 ) . Then the inequality above reads as follows:
 
(1 − σ)v2H 1 (R2 )
Iλ (v) ≥ C1 σv2H 1 (R2 ) + − C2 ,
2(1 + 2σv2H 1 (R2 ) )

and by minimizing the inequality above with respect to σ, we deduce the

desired estimate.
Corollary 22 For any λ > 0, Iλ attains its minumum value at v ∈ H 1 (R2 )
which defines a solution of problem (83).
Proof. From (85), we know that Iλ is coercieve and bounded from below .
Clearly, it is also weakly lower semicontinuous, and so it attains its infimun
at a critical point, that provides the desired solution.

Consequently, for every set of N -(vortex) points {z1 , . . . , zN } (not neces-

sarily distinct) and λ > 0, we have established that problem (74) admits a
solution.
However, contrary to what we have seen for the abelian-Higgs model, now
it is not clear whether ot not other topological solutions exit.
A first contribute to this question is given in case all vortex points coincide,
i.e. z1 = z2 = ... = zN . There holds,
Proposition 23 (cf. [CHMcLY] and [Ha3]) For every N ∈ N, λ > 0 and
z0 ∈ R2 the problem:

−Δu = λeu (1 − eu ) − 4πN δz0 , in R2
u → 0 as |z| → ∞

admits a unique solution uz0 which is radially symmetric about the point z0 .

As a matter of fact, the linearized operator:

L = −Δ + λeuz0 (2euz0 − 1) : L2 (R2 ) → L2 (R2 )

is strictly positive definite (hence invertible), see [CFL] and [T8] for details.
Recently this result has been extended to hold also when the vortex points
may be distinct, provided the parameter λ > 0 is taken large enough. More
precisely the following holds:
Theorem 24 (cf. [T8]) For every N ∈ N and every assigned set of points
{z1 , . . . , zN } ⊂ R2 , there exists λ∗ > 0 and μ∗ > 0, such that for every
solution u of (74) with λ > λ∗ , we have:
152 G. Tarantello

inf ∇ϕ2L2 (R2 ) + eu (2eu − 1)ϕ2 ≥ μ∗
ϕ∈H 1 (R2 ), ϕH 1 (R2 ) =1
R2

As an important consequence of Theorem 24 we have:

Theorem 25 (cf. [Cho1], [T8]) For every N ∈ N and {z1 , . . . , zN } ⊂ R2
there exists λ∗ > 0 such that, for every λ > λ∗ , problem (74) admits a unique
solution.
Proof. By virtue of Theorem 24, if u = u0 +v is a solution of (74) with λ > λ∗
then v defines a strict local mimimum for Iλ . Suppose by contradiction that
there exist two distinct solutions u1 = u2 for (74) with λ > λ∗ . Setting
uj = u0 + vj , j = 1, 2 and observing that v̂(x) = max{v1 (x), v2 (x)} defines a
subsolution for (83), we can consider the minimization problem:

inf{Iλ (v), ∀v ∈ H 1 (R2 ) : v(x) ≥ v̂(x) a.e. x ∈ R2 }

and show that the infimun is attained at a critical point for Iλ . Hence, we
obtain another solution for (83) which is pointwise greater than v1 and v2 . In
other words, we can always assume (by contradiction) to have two ordered
solutions for (83), say v1 < v2 , which define two distinct strict local minima
for Iλ . Then we can use a “mountain pass” constuction between v1 and v2 ,
(see Theorem 12.8 in [St] for details) to obtain a third solution v3 for (83),
such that v1 (x) < v3 (x) < v2 (x), ∀x ∈ R2 and v3 is not a local minimum for
Iλ . But this is in contradiction with Theorem 24.

Generalizations of this uniqueness result can be found in [T7], [ChoN],

and [MaN].

Open question: It remains as a challenging open question to know whether

uniqueness holds for problem (74), without any restriction on λ > 0.

All the results above show strong analogies between abelian-Higgs vortices
and topological Chern-Simons vortices. It should be noticed however, that in
fact their profiles around the vortex points is quite different, as we see from
the following result:
Theorem 26 For every N ∈ N, {z1 , . . . , zN } ⊂ R2 and λ > 0 sufficiently
large, denote by uλaH the unique solution for (37)–(38) (with λ = q12 ) and
by uλCS the unique (topological) solution for (74) (with λ = κ12 ). Then, as
λ → +∞:
(i)
λ 
N
1 − euaH  4π δz j
j=1
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 153

(ii) ⎛ ⎞

N 
N
1−e uλ
CS  4π ⎝ δz j + n j δz j ⎠
j=1 j=1

weakly in the sense of measure. Here nj ∈ N defines the multiplicity of zj ,

j = 1, . . . , N .
See [T8] for details.

In other words, in terms of the corresponding vortex configurations, we see

that, the function: (1 − |ϕ|2 ) concentrates around the vortex points for both
the abelian-Higgs and topological Chern-Simons-Higgs model, (respectively,
as q → 0, and κ → 0). But the “local” behaviour around those points is quite
different, as a result of the different scalar potential.
In terms of Chern-Simons vortices, the results established above can be
summarized as follows (cf. [W], [Y1] and [T8]):

Theorem 27 For any given κ > 0, N ∈ N, and any assigned set of (vortex)
points Z = {z1 , . . . , zN } ⊂ R2 (repeated according to their multiplicity) there
exists (A, ϕ)± a smooth solution to the selfdual equation (33) in R2 (with the
± sign choosen accordingly) such that:
i) |ϕ± | < 1 in R2 , ϕ± vanishes exactly at the set Z and if nj ∈ Z is the
multiplicity of zj ∈ Z j = 1, . . . , N , then:

ϕ+ (z) & ϕ− (z) = O((z − zj )nj ), as z → zj . (86)

ii) For every ε ∈ (0, 1) there exist a constant Rε > 0 such that, for every
0 < κ ≤ 1, the following estimate holds:
1
|D1 ϕ± | + |D2 ϕ± | + |F12 | ≤ cε (1 − |ϕ± |) ≤ Cε e− κ (1−ε)|z| , ∀|z| ≥ Rε , (87)

with cε > 0 and Cε > 0 suitable constants (indipendent of κ). In partic-

ular |ϕ± | → 1, as |z| → +∞.
Furthermore, if κ > 0 is sufficiently small, then (A, ϕ)± is the only
solution (up to gauge transformation) satisfying (86) and (87).
iii) The following holds respectively for the magnetic flux, electric charge and
total energy:

Magnetic flux Φ = R 2 (F12 )± = ±2πN ;
R2 (J )± = ±2πkN ;
0
Electric charge Q = (88)
Total energy E = R2 E± = πN.

In concluding this section, we mention that it is possible to deduce the

following asymptotic behaviour for the vortex solution:
154 G. Tarantello

Theorem 28 (cf. [T8]) As κ → 0+ , we have:

(A0 )± → 0, (J 0 )± → 0, inL1(R2 );

while,

N
(F12 )± → ±2π δz j , (89)
j=1


N
1 N
(A0 )2± → ±π n j δz j & (A0 )± → ±π (nj + 1)δzj ,
j=1
k j=1

weakly in the sense of measure in R2 .

We observe that the abelian-Higgs vortices of Theorem 16 also satisfy the

“concentration” property (89) as q → 0, as shown in [HJS].

5.2 Non Topological Chern-Simons Vortices

We now explore the possibility to establish the existence of non-topological

Chern-Simons vortices, by analyzing the elliptic problem:
⎧
⎨ N
−Δu = λeu (1 − eu ) − 4π δz j
(90)
⎩ j=1
u → −∞, as |x| → +∞.

To be consistent with the physical applications, we need to solve (90) together

with the properties:

u < 0 and eu (1 − eu ) ∈ L1 (R2 ). (91)

By virtue of our discussion at the biginning of this section, it is not difficult to

check that problem (90) and (91) can be equivalently formulated as follows,
⎧
⎨ 
N
−Δu = λeu (1 − eu ) − 4π δz j
(92)
⎩ u j=1
e ∈ L (R ).
1 2

The advantage of (92) is that it may be viewed as a perturbation of the (sin-

gular) Liouville problem (see (93) below) for which we have explicit solutions
available. More precisely, if u solves (92) then for every ε > 0, the scaled
function
z 1
uε (z) = u( ) + 2 log( ),
ε ε
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 155

satisfies:

N
−Δuε = λeuε − λε2 e2uε − 4π δεzj in R2 ,
j=1

which, we may regard as a perturbation of the “singular” Liouville problem:

 N
−Δu
= λeu − 4π j=1 δεzj
(93)
R2
eu < ∞.

We can exhibit an explicit solution for (93). To this purpose, let


N  z
f (z) = (N + 1) (z − zj ) and F (z) = f (ξ)dξ
j=1 0

and set,

N  z
fε (z) = (N + 1) (z − εzj ) and Fε (z) = fε (ξ) dξ.
j=1 0

By Liouville formula (cf. [Lio]) we know that,

8|fε (z)|2
u0ε,a (z) = log (94)
λ(1 + |Fε (z) + a|2 )2

satisfies (93) for any ε ∈ R and a ∈ C. Thus, it is reasonable to search for

solution of problem (92) in the form

u(z) = u0ε,a (εz) + log ε2 + ε2 w(εz) (95)

with w a suitable error term that satisfies:

 2 
0 e ε w
− 1 0 2
−Δw = λeuε,a − λe2(uε,a +ε w) . (96)
ε2

We consider the free parameters ε and a as part of our unknowns and

concentrate around the values ε = 0 and a = 0 where (96) reduces to:

1 2
Δw + ρw = ρ (97)
λ
0
with ρ = λeuε=0,a=0 , the radial function given as follows:

8(N + 1)2 r2N

ρ(r) = , r = |z|. (98)
(1 + r2(N +1) )2

We will solve (97) within the class of radial functions, and find an
explicit solution w0 = w0 (r) given in Lemma 31 below. Thus, using the
decomposition:
156 G. Tarantello

w(z) = w0 (|z|) + u1 (z) (99)

we need to solve for u1 the following equation:

P (u1 , a, ε) :=
 2

w0 +ε2 u1
u0ε,a eε −1 0 1 2
2
w0 +ε2 u1 )
Δu1 + λe − ρw0 + λe2(uε,a +ε
ρ = 0. −
ε2 λ
(100)
To this end, we aim to apply the Implicit Function Theorem (cf. [Nir]) to
the operator P acting between suitable functional spaces where it extends
smoothly at ε = 0 to satisfy P (0, 0, 0) = 0. To this purpose, Chae-Imanuvilov
in [ChI1] have introduced the following spaces:
 
Xα = u ∈ L2loc (R2 ) : (1 + |z|2+α )u2 ∈ L1 (R2 ) , α > 0
2,2 (101)
Yα = u ∈ Wloc (R2 ) : Δu ∈ Xα , u
1+ α ∈ L (R )
2
2 2
, α>0
(1+|z|)

equipped respectively with scalar product:

 
uv
(u, v)Xα = (1 + |z|2+α )uv and (u, v)Yα = (Δu, Δv)Xα +
R2 R2 (1 + |z|)2+α

and relative norms denoted by  · Xα and  · Yα respectively.

For any α > 0, the following continuous embedding properties hold:
Xα → Lq (R2 ), ∀q ∈ [1, 2);
0
Yα → Cloc (R2 ).

Furthermore, we have:
Lemma 29 Let α ∈ (0, 1) and v ∈ Yα ,
(a) if v is harmonic then v is a constant;
(b) the following estimates hold:

|v(z)| ≤ CvYα log(1 + |z|), in R2 ; (102)

∇vLp ≤ Cp vYα , for every p > 2; (103)

where C > 0 and Cp > 0 are suitable constants depending on α and (α, p)
respectively.

See [ChI1] for the proof.

Working with the spaces Xα and Yα is particularly advantageous for the
linear operator
L = Δ + ρ : Yα → Xα (104)
as we can characterize explicitly ker L ⊂ Yα and Im L ⊂ Xα . To this purpose
consider the family of functions:
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 157

Uμ,a (z) = u0ε=0,a (μ) + log μ2 , μ > 0, and a ∈ C;

which satisfy
−ΔU = λeU − 4πN δz=0 (105)
Letting u0 = Uμ=1,a=0 = log ρ and using polar coordinates we see that the
following functions:
1−r 2(N +1)
2(N +1) ∂μ Uμ,a |μ=1,a=0 = 1+r 2(N +1) ,
1 ∂
ϕ0 =
N +1
ϕ+ = − 41 ∂α
∂
Uμ,a |μ=1,a=0 = r 1+r cos((N +1)θ)
2(N +1) , (α = Re a) ; (106)
r N +1 sin((N +1)θ)
ϕ− = − 4 ∂β Uμ,a |μ=1,a=0 =
1 ∂
1+r 2(N +1)
, (β = Im a) ;

belong to ker L in Yα , ∀α > 0. More interestingly, the following holds:

Proposition 30 For α ∈ (0, 1) the operator L in (104) satisfies:
(a) ker L = span {ϕ0 , ϕ + , ϕ− } ⊂ Yα
(b) Im L = {f ∈ Xα : R2 f ϕ± = 0}
To derive Proposition 30 we start by describing the behaviour of the
operator L over radial functions. Hence denote by Lr : Yαr → Xαr the operator

d2 1 d
Lr ϕ = 2
ϕ+ ϕ + ρϕ, ϕ ∈ Yαr (107)
dr r dr
where Yαr and Xαr denote the subspaces of radial functions in Yα and Xα
respectively.
Lemma 31 Let α ∈ (0, 1) and n ∈ Z+ , then
(a) ϕ ∈ Yαr satisfies Lr ϕ = 0 if and only if ϕ ∈ span {ϕ0 }
(b) Lr : Yαr → Xαr is onto.
More precisely, for f ∈ Xα let,
  r
2 1
w(r) = ϕ0 (r) log r + ϕ0 (t)f (t)tdt +
N + 1 (1 + r2(N +1) ) 0
 r 
2 1
−ϕ0 (r) ϕ0 (t) log t + f (t)t dt (108)
0 N + 1 (1 + t2(N +1) )

then w ∈ Yαr , and satisfies: Lr w = f .

Lemma 31 follows by a variation of parameter type argument. We omit
the proof and refer to [ChI1] for details, see also [T8].
Observe that, since w(r) and ẇ(r) extend with continuity at r = 0, and
we find: w(0) = 0 = ẇ(0). Furthermore, setting:
 +∞
cf = ϕ0 (t)f (t)tdt (109)
0

(well defined) we have

158 G. Tarantello

Corollary 32 The function w in (108) admits the following asymptotic

behavior:

w(r) = −cf log r + O(1), as r → +∞, (110)

cf
w (r) = − + O(1), as r → +∞. (111)
r

In particular, by taking f (r) = λ1 ρ2 in (108) we see that:

  r
1 2
w0 (r) = (1 − r 2(N +1)
) log r + ϕ0 (t)tρ2 (t)dt +
λ(1 + r2(N +1) ) N +1 0
 r  
2
−(1 − r2(N +1) ) ϕ0 (t) log t + ρ 2
(t)tdt , (112)
0 (N + 1)(1 + t2(N +1) )

defines a solution for problem (97) in Yαr , such that,

c0
w0 (r) = − log r + O(1) as r → +∞, (113)
λ
1 c0
w0 (r) = − + O(1), as r → +∞, (114)
λ r
with

 
+∞ +∞
(1 − t2(N +1) ) 4N +1
c0 = ϕ0 (t)ρ2 (t)tdt = (8(N + 1)2 )2 t dt =
0 0 (1 + t2(N +1) )5

1 − sN +1 2N
+∞
32(N + 1)4 s ds =
0 (1 + sN +1 )5
 +∞  +∞ 
s2N sN
32(N + 1)4 − 2 s 2N +1
0 (1 + sN +1 )4 0 (1 + sN +1 )5
 +∞  +∞   
s2N 1 d 1
32(N + 1)4 + s 2N +1
ds =
0 (1+sN +1 )4 2(N + 1) 0 ds (1+sN +1)4
 +∞  +∞ 
s2N 2N + 1 s2N
32(N + 1)4 ds − =
0 (1 + sN +1 )4 2(N + 1) 0 (1 + sN +1 )4
 +∞
s2N
16(N + 1)3 ds.
0 (1 + sN +1 )4
From now on we shall use such a solution w0 into the definition of the operator
P given in (100).
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 159

Proof of Proposition 30: We start to establish (a) and so let v ∈ Yα such that
Lv = 0. We can use standard elliptic regularity theory to see that v ∈ C 2 (R2 ).
We write v according to its Fourier decomposition:

v(z) = vk (r)eikθ , z = reiθ (115)
k∈Z

with complex valued functions vk = vk (r) such that v−k = v k and whose real
and immaginary part satisfy:

k2
Lr ϕ − ϕ = 0. (116)
r2
For k = 0 the real valued function v0 (r) ∈ Yαr , and by Lemma 31 we see that
v0 (r) ∈ span {ϕ0 }. For k ∈ N, to determine a fundamental set of solutions to
(116) we use the following solutions of the (singular) Liouville equations:

8|(N + 1)z N + (k + N + 1)az N +k |2

ψa,k (z) = log ,
(1 + |z N +1 + az N +k+1 |2 )2

for a ∈ C and k ∈ N. Notice that ψa=0,k = log ρ and, according to Liouville

formula [Lio], ψa,k satisfies:


k
−Δψa,k = eψa,k − 4πN δz=0 − 4π δzja , in R2
j=1

where zja , j = 1, . . . , k, defines the k-distinct non-zero roots of the polynomial:

(N + 1)z N + (k + N + 1)az N +k . Notice in particular that, |zja | → +∞, as
a → 0, ∀j = 1, . . . , k. Therefore, for each test function ϕ ∈ C0∞ (R2 ), and |a|
sufficiently small, we have:

−Δψa,k ϕ = eψa,k ϕ − 4πN ϕ(0).

By differentiating this expression with respect to α = Re(a) and β =

Im(a), we obtain
∂ψa,k 2
ϕ1,k = |a=0 = ϕk (r) cos kθ;
∂α N +1
∂ψ 2
ϕ2,k = |a=0 = ϕk (r) sin kθ;
∂β N +1
where
k + N + 1 + (k − N − 1)r2(N +1) k
ϕk (r) = r . (117)
1 + r2(N +1)
They satisfy:
Lϕ1,k = 0 = Lϕ2,k , ∀k ∈ N,
160 G. Tarantello

so that ϕk (r) verifies (116). By replacing k with −k in (117)

we still obtain
a solution ϕ̃k for (116) and we check that ϕ̃k (r) = ϕk 1r . Thus, ϕk (r)
and ϕ̃k (r) define a fundamental set of solutions for (116). But the real and
immaginary part of vk cannot include the component ϕ̃k (r), since it admits a
1
rk
singularity at the origin, ∀k ∈ N. Furthermore, for α ∈ (0, 1) and k = N +1
the function ϕk ∈ Yαr , (this is due to its rk behavior as r → +∞) and so we
conclude that vk = 0, ∀k = N + 1. While, for k = N + 1, Re(vk=N +1 ) and
N +1
Im(vk=N +1 ) belong to span{ϕN +1 } = span { 1+rr 2(N +1) }. Thus, we conclude
that v ∈ span {ϕ0 , ϕ+ , ϕ− } as claimed.
To establish (b), we show the following:
Claim Im L is closed in Xα .
Let ϕn ∈ (ker L)⊥ ⊂ Yα be such that Lϕn = fn ∈ Xα and fn → f in Xα .
ϕ2
We claim that: R2 (1+|z|nα+2 ) ≤ C, ∀n ∈ N for suitable C > 0. Indeed, by
contradiction we assume that (along a subsequence)
  12
ϕ2n ϕn
cn = → +∞, and set Φn = .
R2 (1 + |z|α+2 ) cn

Then

Φ2n
= 1, Φn ∈ (ker L)⊥ ⊂ Yα , and LΦn → 0 in Xα . (118)
R2 (1 + |z|α+2 )

Hence, Φn is uniformly bounded in Yα and (along a subsequence) we have:

Φn → ϕ weakly in Yα ,

with ϕ ∈ Yα . We can assume also that the convergence above holds in

L2loc (R2 ).
Since

|Φn (z) − ϕ(z)| ≤ c(Φn Yα + ϕYα ) log(1 + |z|) ≤ C log(1 + |z|)

we see that,

 
(Φn − ϕ)2 log(1 + |z|)
≤ Φn − ϕ2L2 (BR ) + C =
R2 (1 + |z|1+α )2 |z|≥R (1 + |z|1+α )2

= Φn − ϕ2L2 (BR ) + o(1)

(Φn −ϕ)2
as R → +∞. Thus, R2 (1+|z|1+α )2 → 0 as n → ∞, and we conclude that

|ϕ|2
= 1. (119)
R2 (1 + |z|1+α )2
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 161

Similarly we see that R2 ρΦn → R2 ρϕ as n → ∞. Hence Lϕ = 0 and
ϕ ∈ (ker L)⊥ . So we conclude that ϕ = 0, in contradiction to (119). In
conclusion, the sequence ϕn ∈ (ker L)⊥ : Lϕn = fn is uniformly bounded
in Yα . So we can argue exactly as above to find ϕ ∈ Yα : ϕn → ϕ weakly in
Yα with Lϕ = f and the Claim is proved.
Consequently, we may decompose:

Xα = ImL ⊕ (ImL)⊥

according to the scalar product in Xα . Thus, for ξ ∈ (ImL)⊥ we have:

0 = (Lu, ξ)Xα = (Lu, (1 + |z|)2+α ξ), ∀u ∈ Yα .

The density of C0∞ in Yα implies that the function

ψ = (1 + |z|)2+α ξ ∈ Yα and Lψ = 0.

Therefore by part (a), we may write: ψ = a0 ϕ0 + a+ ϕ+ + a− ϕ− for suit-

able constants a0 , a+ and a− . Furthermore, by Lemma 31 we know that
the radial function f (r) = ϕ0 (r) (1+r1 2 )2 ∈ ImL and satisfies (f, ϕ± )L2 = 0.
Consequently
  
+∞  2
ϕ0 (r)
0 = (f, ξ)Xα = (f, ψ)L2 = a0 2π rdr ,
0 1 + r2

that is a0 = 0, and also part (b) of our claim follows.

At this point we can complete our perturbation analysis, as the operator:

P : Yα × C × R → Xα

in (100) is well defined, smooth and extends with continuity at ε = 0 with

P (0, 0, 0) = 0. Moreover, the linearized operator:

∂P
A: (0, 0, 0) : Yα × C → Xα (120)
∂(u1 , a)

takes the form:

A(ϕ, b) = Lϕ + M (b) (121)
with ϕ ∈ Yα , b = b1 + ib2 ∈ C and
   
2 2
M (b) = −4 ρw0 − ρ2 ϕ+ b1 − 4 ρw0 − ρ2 ϕ− b2 . (122)
λ λ
162 G. Tarantello

Observe that:

Lemma 33
    +∞  
2 2 r2(N +1)
ρw0 − ρ2 ϕ2± = π ρ(r)w0 (r)− ρ2 (r) rdr < 0.
R2 λ 0 λ (1+r2(N +1) )2

Proof. By easy calculation, we find:

 
1 (N + 1)2 r4N +2
Lr = .
16(1 + r2(N +1) )2 (1 + r2(N +1) )4

Furthermore we can take advantage of the decay estimates (113) and (114)
to use integration by part and obtain:
  
2
ρw0 − ρ2 ϕ2± =
R2 λ
 +∞  
8(N + 1)2 r4N +2 2 2 r2(N +1)
=π w0 (r) − ρ rdr =
0 (1 + r2(N +1) )4 λ (1 + r2(N +1) )2
 +∞    
1 r 1 2 2 r2(N +1)
=π L w0 (r) − ρ rdr =
0 2 (1 + r2(N +1) )2 λ (1 + r2(N +1) )2
 +∞  
1 r 1 2 2 r2(N +1)
= L w0 − ρ rdr =
0 2 (1 + r2(N +1) )2 λ (1 + r2(N +1) )2
  
1 +∞ 2 1 2r2(N +1)
= ρ (r) − rdr,
λ 0 2(1 + r2(N +1) )2 (1 + r2(N +1) )2

where, to derive the last identity we have used the fact that w0 satisfies (97).
The sign of the integral above can be determined by means of change of
variable t = r2 ,
 +∞  +∞
2 1 − 4r2(N +1) 4 1 − 4tN +1 2N
ρ (r) rdr = 16(N + 1) t dt =
0 2(1 + r2(N +1) )2 0 (1 + tN +1 )6
 +∞  +∞ 
t2N t3N +1
= 16(N + 1)4 − 5 dt =
0 (1 + tN +1 )5 0 (1 + tN +1 )6
 +∞  +∞   
t2N 1 2N +1 d 1
= 16(N + 1)4 + t dt =
0 (1 + tN +1 )5 N + 1 0 dt (1+tN +1 )5
 +∞  
t2N 2N + 1 +∞ t2N
16(N + 1)4 − dt =
0 (1 + tN +1 )5 N +1 0 (1 + tN +1 )5
 +∞
t2N
−16(N + 1)3 N dt < 0,
0 (1 + tN +1 )5
and the desired conclusion follows.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 163

We are now ready to show the following,

Proposition 34 For α ∈ (0, 1), the operator A : Yα → Xα in (120) and
(121) is onto and ker A = ker L = span {ϕ0 , ϕ+ , ϕ− }.
Proof. Let f ∈ Xα , we need to find ϕ ∈ Yα and b = b1 + ib2 ∈ C such that:

A(ϕ, b) = Lϕ + M (b) = f, with M (b) in (122). (123)

 ϕ+ and integrate over R to find:

2
To this purpose, multiply (123) by
f ϕ+ =
R2
      
2 2 2 2
= Lϕϕ+ − 4b1 ρw0 − ρ ϕ+ − 4b2
2
ρw0 − ρ ϕ+ ϕ− =
R2 R2 λ R2 λ
 +∞   2(N +1)
2 r
= −4πb1 ρ(r)w0 (r) − ρ2 (r) dr.
0 λ (1 + r2(N +1) )2

By Lemma 33, we may solve for b1 as follows,

  +∞   −1
1 2 r2(N +1)
b1 = − f ϕ+ ρ(r)w0 (r) − ρ2 (r) dr .
4π R2 0 λ (1 + r2(N +1) )2
(124)
Analogously we derive:
   +∞  −1
1 2 r2(N +1)
b2 = − f ϕ− ρ(r)w0 (r) − ρ2 (r) dr .
4π R2 0 λ (1 + r2(N +1) )2
(125)
Set g = f − M (b), where M (b) is given
in (122) with b 1 and b 2 specified in
(124) and (125) respectively. Since R2 gϕ± = 0, by Proposition 30 (b), we
find ϕ ∈ Xα : Lϕ = g. Thus we have checked that Im A = Xα . If we take
f = 0 in the argument above, we find that b1 = 0 = b2 , and so Aϕ = 0 if and
only if Lϕ = 0, that is ker A = ker L = span {ϕ0 , ϕ+ , ϕ− }.

Setting,
Uα = (span{ϕ0 , ϕ+ , ϕ− })⊥ ,
we obtain to the following existence result for (92):
Proposition 35 For every λ > 0 and α ∈ (0, 1), there exists ε0 > 0
sufficiently small and smooth functions:

aε : (−ε0 , ε0 ) → C, u1,ε : (−ε0 , ε0 ) → Uα

with aε=0 = 0 and u1,ε=0 = 0 such that

uε (z) = u0ε,aε (εz) + log ε2 + ε2 w0 (ε|z|) + ε2 u1,ε (εz) (126)

164 G. Tarantello

defines a solution for (92), with w0 in (112) satisfying (113)–(114). Further-

more, as ε → 0 the following estimates hold:
o(1)
|u1,ε (z)| ≤ o(1) log(1 + |z|), |∇u1,ε (z)| = , ∀z ∈ R2 ;
1 + |z|

λ euε (1 − euε ) = 8π(N + 1) + o(1) (127)
R2

Proof. By a straightforword application of the Implicit Function Theorem

(cf. [Nir]) we obtain uε in (126). To check the validity of (127) we use (102)
to estimate:

|u1,ε (z)| ≤ C1 ||u1,ε ||Yα log(1 + |z|), ∀z ∈ R2

with u1,ε Yα → 0, as ε → 0. Furthermore, we see that,

−Δu1,ε = f1,ε

with f1,ε → 0 in Xα and (1 + |z|2 )|f1,ε | ≤ cλ ε in R2 , with a suitable constant

cλ independent of ε. Since u1,ε ∈ Yα , we can write

1 y−z
∇u1,ε (z) = f1,ε (y)dy,
2π R2 |y − z|2

and for |z| ≥ 2, derive the following estimate:

 
|z||f1,ε (y)| |z||f1,ε (y)|
|z||∇u1,ε (z)| ≤ +
|z|
{|y−z|≤ 2 } |y − z| |z|
{|y−z|≥ 2 }} |y − z|
 
|y||f1,ε (y)|
≤2 +2 |f1,ε (y)|
{|y−z|≤ |z|
2 }
|y − z| {|y−z|≤ |z|
2 }
 
π|z| max |y||f1,ε (y)| + 2||f1,ε ||L1 (R2 )
|z|
{|y|≥ 2 }

≤ cλ ε + 2||f1,ε ||L1 (R2 ) → 0, as ε → 0.

and the first estimate in (127) follows. Finally, notice that

 
1
euε ( ε ) (1 − euε ( ε ) ) 2 =
z z
λ e uε (z)
(1 − e uε (z)
)=λ
2 2 ε
 R R

uε,aε (z)+ε2 (w0 +u1,ε ) 0 2
λ e − λε2 e2uε,aε (z)+2ε (w0 +u1,ε )
R2 R2

Therefore, by the estimates established above, we can pass to the limit

into the integral sign to conclude that,
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 165
   +∞
0 r2N +1
lim λ euε (z) (1−euε (z) ) = λ euε=0,a=0 = 16π(N + 1) dr
ε→0 R2 R2 0 (1+r2(N +1) )2
 +∞
dt
= 8π(N + 1) = 8π(N + 1).
0 (1 + t)2

This completes the proof.

We have established the following existence result concerning non-topo-

logical Chern-Simons vortices:
Theorem 36 For k > 0, N ∈ N and a given set of (vortex) points Z =
{z1 , . . . , zN } (repeated according to their multiplicity), there exist ε0 > 0 such
that, for every ε ∈ (0, ε0 ) we have a vortex configuration (Aε , ϕε )± solution
to the selfdual equations (33) (with the ± sign choosen accordingly) such that
i) |ϕε± | < 1 in R2 ; ϕε± vanishes exactly in the set Z, and if nj ∈ N is the
multeplicity of zj then,

ϕε+ (z) and ϕε− (z) = O((z − zj )nj ), as z → zj , j = 1, . . . , N. (128)

ii) There exist constants Cε > 0, Rε > 0 and βε → 0+ as ε → 0 such that

|F12
ε
| + |z|2 |∇|ϕε± ||2 ≤ Cε |z|−2(N +2βε ) , ∀|z| ≥ Rε . (129)

In particular ϕε± (z) → 0 as |z| → +∞.

iii)

Magnetic flux: Φε± = ε
(F12 )± = ±4π(N + 1) + o(1); (130)
R2

Electric charge: Qε± = (Jε0 )± = ±4πk(N + 1) + o(1); (131)
R2

Total energy: Eε = E± = 4π(N + 1) + o(1); (132)
R2

ε → 0.

The perturbative approach presented above has been successfully applied

to obtain non-topological type solutions in many other context, as one can
see for instance in [Ch1], [Ch2], [Ch3], [Ch4], [ChCh1], [ChI2], [ChI3], [ChT1]
and [ChT2].
On the other hand an alternative construction of non-topological vortices is
carried out in [CFL], and provides a new class of Chern-Simons vortices with
the property that they “concentrate” around the vortex points (as κ → 0).
This is a much desirable property from the point of view of the physical
applications, that as we know, it is always satisfied by topological vortices,
see (89).
166 G. Tarantello

We conclude by observing that a parallel analysis (more or less complete)

has been developed for the study of periodic vortices or vortices defined over
compact surfaces,in the framework of selfdual abelian models that include
and generalize those discussed above. In this respect, see for example: [CY],
[ChiR], [DJLPW], [DJLW1], [DJLW2], [DJLW3], [Ha1], [Ha2], [MNR], [NT1],
[NT3], [Ol], [Ri1], [Ri2], [Ri3], [RT1], [SY1], [T1], [T2], [T3] and [WY].
Furthermore, such analysis is relevant in connection to other problems
from physics and geometry where equations with similar features emerge
naturally, see e.g. [Au], [Ban], [CLMP1], [CLMP2], [CK1], [CK2], [CK3],
[ChCL], [ChY1], [ChY2], [ChY3], [H], [K], [KW1] [KW2], [Ki1], [Ki2], [Ni],
[Su2] and [Wo].
On the contrary, much less has been accomplished in the context of non-
abelian selfdual theories, where the study of non-abelian vortex configurations
give rise to systems of elliptic equations, which pose some new and delicate
analytical difficulties. In this direction, we mention the (partial) contributions
of [BT2] and [SY2] towards the understanding of W -condensates for the (non-
abelian) Electroweak theory (cf. [La]), as motivated by the work of Ambjorn-
Olesen (cf. [AO1], [AO2] and [AO3]); see also [SY3].
While, non-abelian selfdual Chern-Simons vortices are analyzed in [JoLW],
[JoW1], [JoW2] [LN], [ChOS], [NT1], [Y3] and [Y4].
But still many unresolved issues remain (see [T8]), as for instance the
understanding of the asymptotic behavior of non-abelian Chern-Simons vor-
tices. Indeed, as vortex configurations are likely to develop a “concentration”
behavior (for limiting values of the parameters), it is necessary to provide an
accurate blow-up analysis for systems, in the same spirit of what it is now
available for single Liouville-type equations, see e.g. [BP], [BCLT], [BT1],
[BLS], [BM], [CLS], [ChL1], [ChL2], [ChL3], [ChL4], [ChLW], [CD], [CL1],
[CL2], [CL3], [Che], [Chn], [CW], [DeKM], [Dj], [Dr], [Es], [EGP], [L1], [L2],
[LS], [Li1], [Li2], [LiL1], [LiL2], [LiW], [Lu], [LZ], [MW], [MNR], [MaN], [NS],
[OS1], [OS2], [PT], [Sh], [ST], [Su1], [WW1] and [WW2].
This direction of investigation is still quite open, aside from the contribu-
tions in [JoLW], [JoW1], [JoW2], [MN], [SW1], [SW2] and [W], which however
do not take into account the presence of Dirac measures, that are the cause
of even more degenerate behaviours.

References

[Ab] A.A. Abrikosov, On the magnetic properties of superconductors of second

group, Sov. Phys. JETP 5 (1957), 1174–1182.
[Ad] D.R. Adams, A sharp inequality of J. Moser for higher order derivatives, Ann.
of Math. 128 (1988), 385–398.
[AH] I. Aitchinson, A. Hey, Gauge theories in Particle physics, IoP Publisher vol.
1 (2002), vol. 2 (2003).
[AO1] J. Ambjorn, P. Olesen, On Electroweak magnetism, Nucl. Phys. B 218 (1989),
67–71.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 167

[AO2] J. Ambjorn, P. Olesen, A condensate solution of the classical electroweak

theory which interpolates between the broken and symmetric phase, Nucl.
Phys. B 330 (1990), 193–204.
[AO3] J. Ambjorn, P. Olesen A magnetic condensate solution of the classical
Electroweak theory, Phys. Lett. B 218 (1989), 67–71.
[ADHM] M.F. Atiyah, V.G. Drinfeld, N.J. Hitchin, Yu.I Mannin, Constraction of
Instantons, Phys. Lett. A 65 (1978), 185–187.
[AtH] M.F. Atiyah and N.J. Hitchin, The Geometry and Dynamics of Magnetic
Monopoles, Princeton Univ. Press, Princeton (1988).
[AHS1] M.F. Atiyah, N.J. Hitchin, I.M. Singer, Deformation of Instantons, Proc. Natl.
Acad. Sci. USA 74, (1997), 2662–2663.
[AHS2] M.F. Atiyah, N.J. Hitchin, I.M. Singer, Selfduality in four dimensional
Riemannian geometry, Proc. Roy Soc. A 362 (1978), 425–461.
[Au] T. Aubin, Some Nonlinear Problems in Riemannian Geometry, Springer-
Verlag, Berlin Heidelberg New York, (1998).
[BP] S. Baraket, F. Pacard, Construction of singular limit for a semilinear elliptic
equation in dimension 2, Calc. Var. P.D.E, 6 (1998), 1–38.
[Ban] C. Bandle, Isoperimetric Inequalities and Applications, Pitman A. Publishing
7, (1980).
[BCLT] D. Bartolucci, C.C. Chen, C.S. Lin, G. Tarantello, Profile of blow-up solutions
to mean field equations with singular data, Comm. P.D.E. 29, n. 7-8 (2004),
1241–1265.
[BT1] D. Bartolucci, G. Tarantello, The Liouville equation with singular data:
a concentration-compactness principle via a local representation formula,
J. Diff. Eq. 185 (2002), 161–180.
[BT2] D. Bartolucci, G. Tarantello, Liouville type equations with singular data and
their applications to periodic multivortices for the electroweak theory, Comm.
Math. Phys. 229 (2002), 3–47.
[Be] W. Beckner, Sharp Sobolev inequalityies on the sphere and the Moser-
Trudinger inequality Ann. of Math. 138 (1993), 213–242.
[BPST] A.A. Belavin, A.M. Polyakov, A.S. Schwartz, Yu.S. Tyupkin, Pseudoparticle
solutions of the Yang Mills Equations, Phys. Lett. B (1975), 85–87.
[BeR] J. Berger, J. Rubistein, On the zero set of the wave function in superconduc-
tivity, Comm. Math. Phys. 202, n. 3 (1999), 621–628.
[BBH] F. Bethuel, H. Brezis, F. Helein, Ginzburg Landau Vortices, Birkhauser, 1994.
[Bo] E.B. Bogomolnyi, The stability of classical solutions, Sov. J. Nucl. Phys. 24
(1976), 449–454.
[Bra1] S. Bradlow, Vortices in Holomorphic line bundles over closed Kahler Mani-
folds, Comm. Math. Phys. 135 (1990), 1–17.
[Bra2] S. Bradlow, Special metrics and solvability for holomorphic bundles with
global sections, J. Diff. Geom. 33 (1991), 169–214.
[BLS] H. Brezis, Y.Y. Li, I. Shafrir, A sup + inf inequality for some nonlinear elliptic
equations involving exponential nonlinearities, J. Funct. Anal. 115 (1993),
344–358.
[BM] H. Brezis, F. Merle, Uniform estimates and blow-up behavior for solutions of
−Δu = V (x)eu in two dimensions, Comm. P.D.E. 16 (1991), 1223–1253.
[CLS] X. Cabré, M. Lucia, M. Sanchon, On the minimizers of a Moser-Trudinger
type inequality, Comm. P.D.E. 30, n. 7-9 (2005), 1315–1330.
[CY] L. Caffarelli, Y. Yang, Vortex condensation in the Chern-Simons-Higgs model,
Comm. Math. Phys., 168 (1995), 154–182.
[CLMP1] E. Caglioti, P.L. Lions, C. Marchioro, M. Pulvirenti, A special class of sta-
tionary flows for two-dimensional Euler equations, a statistical mechanics
description, part I, Comm. Math. Phys. 143 (1992), 501–525.
168 G. Tarantello

[CLMP2] E. Caglioti, P.L. Lions, C. Marchioro C., M. Pulvirenti, A special class of

stationary flows for two-dimensional Euler equations, a statistical mechanics
description, part II, Comm. Math. Phys. 174 (1995), 229–260.
[CaL] D. Cangemi, C. Lee, Selfdual Chern-Simons solitons and (2 + 1)-dimensional
Einstein gravity, Phys. Review D 46, n. 10 (1992), 4768–4771.
[Ch1] D. Chae, Existence of multistrings solutions of the selfgravitating massive
W -boson, Lett. Math. Phys. 73, n. 2 (2005), 123–134.
[Ch2] D. Chae, On the elliptic system arising from a self-gravitating Born-Infeld
Abelian Higgs theory, Nonlinearity 18, n. 4 (2005), 1823–1833.
[Ch3] D. Chae, Existence of the semilocal Chern-Simons vortices, J. Math. Phys.
46, n. 4 (2005), 042303, 10 pp.
[Ch4] D. Chae, On the multi-string solutions of the self-dual static Einstein-Maxwell-
Higgs system, Calc. Var. PDE 20, n. 1 (2004), 47–63.
[ChCh1] D. Chae, K. Choe, Existence of selfgravitating Chern-Simons vortices,
J. Math. Phys. 44, n. 12 (2003), 5616–5636.
[ChI1] D. Chae, O. Imanuvilov, The existence of non-topological multivortex solu-
tions in the relativistic selfdual Chern-Simons theory, Comm. Math. Phys.
(2000), 119–142.
[ChI2] D. Chae, O. Imanuvilov, Non-topological solutions in the generalized self-dual
Chern-Simons-Higgs theory, Calc. Var. PDE 16, n. 1 (2003), 47–61.
[ChI3] D. Chae, O. Imanuvilov, Non-topological multivortex solutions to the self-
dual Maxwell-Chern-Simons-Higgs systems, J. Funct. Anal. 196, n. 1 (2002),
87–118.
[ChK1] D. Chae, N. Kim, Topological multivortex solutions of the selfdual Maxwell-
Chern-Simons-Higgs system, Jour. Diff. Eq. 134 (1997), 154–182.
[ChK2] D. Chae, N. Kim, Vortex condensates in the relativistic selfdual Maxwell-
Chern-Simons Higgs system, Preprint.
[ChNa] D. Chae, H.S. Nam, On the condensate multivortex solutions of the self-dual
Maxwell-Chern-Simons CP (1) model, Ann. Henri Poincaré 2, n. 5 (2001),
887–906.
[ChOS] D. Chae, H. Ohtsuka, T. Suzuki, Some existence results for solutions to SU(3)
Toda system. Calc. Var. PDE 24, n. 4 (2005), 403-429.
[ChT1] D. Chae, G. Tarantello, On Planar Electroweak vortices Ann. IHP Analyse
Non Lineaire, AN21 (2004), 187–207.
[ChT2] D. Chae, G. Tarantello, Selfgravitating Electroweak Strings, J. Diff. Eq. 213
(2005), 146–170.
[CFL] H. Chan, C.C. Fu, C.S. Lin, Non-topological multi-vortex solutions to the
selfdual Chern-Simons-Higgs equation, Comm. Math. Phys. 231 (2002),
189–221.
[ChNe] M. Chai, Chain-N.P. Nelipa, Introduction to Gauge Field Theory, Springer,
Berlin New York, (1984).
[ChCL] A. Chang, C.C. Chen, C.S. Lin, Extremal functions for a mean field equation
in two dimension, Lectures on partial differential equations, New Stud. Adv.
Math., 2, Int. Press, Somerville, MA, (2003), 61–93.
[ChY1] A. Chang, P. Yang, Conformal deformation of metric on S 2 , J. Diff. Geom.
27 (1988), 259–296.
[ChY2] A. Chang, P. Yang, Prescribing Gaussian curvature on S 2 , Acta Math., 159
(1987), 215–259.
[ChY3] A. Chang, P. Yang, The inequality of Moser-Trudinger and Applications to
Conformal Geometry, Comm. Pure Appl. Math. 56 (2003), 1135–1150.
[CK1] S. Chanillo, M. Kiessling, Rotational symmetry of solutions of some nonlinear
problems in statistical mechanics and in geometry, Comm. Math. Phys. 160
(1994), 217–238.
[CK2] S. Chanillo, M. Kiessling, Conformally invariant systems of nonlinear PDE of
Liouville type Geom. Funct. Analysis 5 (1995) 924–947.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 169

[CK3] S. Chanillo, M. Kiessling, Surfaces with prescribed scalar curvature Duke

Math. J. 105 (2002) 309–353.
[ChL1] C.C. Chen, C.S. Lin, Sharp Estimates for Solutions of Multi Bubbles
in Compact Riemann Surfaces, Comm. Pure Appl. Math. 55 (2002)
728–771.
[ChL2] C.C. Chen, C.S. Lin, Topological Degree for a Mean Field Equation on
Riemann Surfaces, Comm. Pure Appl. Math. 56 (2003), 1667–1727.
[ChL3] C.C. Chen, C.S. Lin, On the symmetry of blow up solutions to a mean field
equation, Ann. IHP Analyse Nonlineaire 18 (2001), 271–296.
[ChL4] C.C. Chen, C.S. Lin, A sharp sup + inf estimate for a nonlinear equation in
the plane Comm. Anal. Geom. 6 (1998), 1–19.
[ChLW] C.C. Chen, C.S. Lin, G. Wang, Concentration phenomena of two-vortex solu-
tions in a Chern-Simons model. Ann. Sc. Norm. Super. Pisa Cl. Sci. 5 3 n. 2
(2004), 367–397.
[CD] W. Chen, W. Ding, Scalar curvature on S 2 , Trans. AMS(1987), 365–382.
[CL1] W. Chen, C. Li, Classification of solutions of some nonlinear elliptic equations,
Duke Math. J. 63 (1991), 615–623.
[CL2] W. Chen, C. Li, Qualitative properties of solutions of some nonlinear elliptic
equations in R2 , Duke Math. J. 71 (1993), 427–439.
[CL3] W. Chen, C. Li, Prescribing Gaussian curvature on surfaces with conical
singularities, J. Geom. Anal. 1 (1991), 359–372.
[Che] X.A. Chen, A Trudinger inequality on surfaces with conical singularities, Proc.
Amer. Math. Soc. 108 (1990), 821–832.
[Chn] X.X. Chen, Remarks on the existence of branch bubbles on the blow up anal-
ysis of equation −Δu = eu in dimension two, Comm. Anal. Geom. 7 (1999),
295–302.
[CHMcLY] X. Chen, S. Hastings, J. McLeod, Y. Yang, A nonlinear elliptic equation aris-
ing from gauge field theory and cosmology, Proc Roy. Soc. Lond. A (1994),
453–478.
[ChiR] F. Chiacchio, T. Ricciardi Multiplicity for a selfdual CP (1) Maxwell-Chern-
Simons model, NoDEA 13 (2007), 563–584.
[CSW] M. Chipot, I. Shafrir, G. Wolansky, On the solutions of Liouville systems,
Jour. Diff. Eq. 140 (1997), 59–105, Erratum Jour. Diff. Eq. 178 (2002), 630.
[CM] Y.M. Cho, D. Maison, Monopole configurations in Weinberg-Salam model,
Phys. Rev. Lett. B 391 (1997), 360–365.
[Cho1] K. Choe, Uniqueness in Chern-Simons theory, J. Math. Phys. 46 (2005), n. 1,
012305.
[Cho2] K. Choe, Asymptotic behavior of condensate solutions in the Chern-Simons-
Higgs theory, J. Math. Phys. to appear.
[ChoK] K. Choe, N. Kim, Blow-up analysis of the selfdual Chern-Simons -Higgs vortex
equation, Ann. IHP Analyse Non Lineaire, to appear.
[ChoN] K. Choe, H.S. Nam, Existence and uniqueness of Topological Multivortex
solutions of the Selfdual Chern-Simons CP(1) Model, Nonlinear Anal. 66
(2007), 2794–2813.
[CW] K.S. Chou, T.Y.H. Wan, Asymptotic radial symmetry for solutions of u +
exp u = 0 in a punctured disc, Pacific J. of Math. 163 (1994), 269–276.
[DET] J. Dalbeault, M.J. Esteban, G. Tarantello, The role of Onofri type inequal-
ities in the symmetry properties of extremals for Caffarelli-Kohn-Nirenberg
inequalities in two space dimensions, Ann. Sc. Nor. Pisa vol VII, (2008),
313–341.
[DeKM] M. Del Pino, M. Kowalczyk, M. Musso Singular limits in Liouville-type
equations, Calc. Var. PDE, 24 (2005), 47–81.
[DJLPW] W. Ding, J. Jost, J. Li, X. Peng, G. Wang, Self duality equations for Ginzburg-
Landau and Seiberg-Witten type functionals with 6th order potential, Comm.
Math. Phys. 217 (2001), 383–407.
170 G. Tarantello

[DJLW1] W. Ding, J. Jost, J. Li, G. Wang, The differential equation Δu = 8π − 8πeu

on a compact Riemann surface, Asian. J. Math. 1 (1997), 230–248.
[DJLW2] W. Ding, J. Jost, J. Li, G. Wang, An analysis of the two-vortex case in the
Chern-Simons-Higgs model, Calc. Var. and P.D.E. 7 (1998), 87–97.
[DJLW3] W. Ding, J. Jost, J. Li, G. Wang, Existence results for mean field equations,
Ann. IHP Analyse Non Linéaire 16 (1999), 653–666.
[Dj] Z. Djadli, Existence results for the mean field problem in Riemann surfaces of
all genus, Comm. Contemp. Math., Vol. 10 (2008), 205–220.
[Dr] O. Druet, Multibumps analysis in Dimension 2: Quantification of blow-up
levels, Duke Math. J. 132 (2006), 217–269.
[DK] S. Donaldson, P. Kronheimer, The geometry of four-manifolds, Oxford Univ.
Press (1990).
[DGP] Q. Du, M.D. Gunzburger, J.S. Peterson, Analysis and approximation of the
Ginzburg-Landau model of superconductivity. SIAM Review, 34, 1, (1992),
54–81.
[D1] G. Dunne, Self-Dual Chern-Simons Theories, Lect. Notes in Phys., M 36 New
Series, Springer, (1995).
[D2] G. Dunne, Mass degeneracies in self-dual models, Phys. Lett. B 345 (1995),
452–457.
[D3] G. Dunne, Aspects of Chern Simons theory, Les Houches Lectures on sec-
tion LXIX: ”Topological Aspects of low dimensional systems, Eds A. Comtet,
T.Jolicoeur, S. Ouvry and F. David, EDP Sciences Springer (1998), 55–175.
[Es] P. Esposito, Blow up solutions for a Liouville equation with singular data,
SIAM J. Math. Anal. 36, n. 4 (2005), 1310–1345.
[EGP] P. Esposito, M. Grossi, A. Pistoia, On the existence of blowing-up solutions
for a mean field equation. Ann. Inst. H. Poincaré Anal. Non Linéaire 22, n. 2
(2005), 227–257.
[Fel] B. Felsager, Geometry, Particle and Fields, Springer, Berlin and New York
(1998).
[Fo] L. Fontana, Sharp borderline Sobolev inequalities on compact Riemannian
manifolds, Comment. Math. Helv. 68 (1993), 415–454.
[Fro] J. Frohlich, The fractional Quantum Hall effect, Chern-Simons theory and
integral lettice, Proc. Internat. Congr. Math. Birkhauser, Basil (1995), 75–105.
[FM1] J. Frohlich, P. Marchetti, Quantum field theory of anyons, Lett. Math. Phys.
16 (1988), 347–358.
[FM2] J. Frohlich, P. Marchetti, Quantum field theory of vortices and anyons, Comm.
Math. Phys. 121 (1989), 177–223.
[Ga1] O. Garcia-Prada, A direct existence proof for the vortex equation over a
compact Riemannian surface, Bull. London Math. Soc. 26 (1994), 88–96.
[Ga2] O. Garcia-Prada, Invariant connections and vortices, Comm. Math. Phys. 156
(1993), 527–546.
[Ga3] O. Garcia-Prada, Dimensional reduction of stable bundles vortices and
stablepairs, Intern. J. Math. 5 (1994), 1–52.
[GL] V. Ginzburg and L. Landau, On the theory of Supercondictivity, Zh.
Eksper. Theor. Fiz. 20 (1950) 1064–1082. Translated in “Collected papers
of L. Landau” ed D. Ter Haar Pergamon, New York (1965), 546–568.
[GT] D. Gilbarg, N.S. Trudinger, Elliptic partial differential equations of second
order, Springer Verlag, (1983).
[GS] M. Gockeller, T. Schucker, Differential Geometry, Gauge Theory and Gravity,
Cambridge Univ. Press (1990).
[GO] P. Goddard, D.I. Olive, Magnetic monopoles in gauge field theories, Rep. Prog.
Phys. 41 (1978) 1360–1473.
[Ha1] J. Han, Asymptotics for the vortex condensate solutions in Chern-Simons-
Higgs theory, Asymp. Anal. 28 (2001) 31–48.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 171

[Ha2] J. Han, Asymptotic limit for condensate solutions in the abelian Chern-
Simons-Higgs model, Proc. AMS 131 n. 6 (2003) 1839–1845.
[Ha3] J. Han, Existence of topological multivortex solutions in the selfdual gauge
theories, Proc. Royal soc. Edinburgh, 130A (2000), 1293–1309.
[HaK] J. Han, N. Kim, Non-selfdual Chern-Simons and Maxwell-Chern-Simons
vortices in bounded domains, J. Funct. Anal. 221 (2005), 167–204.
[H] Z.C. Han, Prescribing Gaussian curvature on S 2 , Duke Math. J. 61 (1990),
679–703.
[HJS] M.C. Hong, J. Jost, M. Struwe, Asymtotic limits of a Ginzburg-Landau
type functional, Geometric Analysis and the Calculus of Variations for
S. Hildebrandt (J. Jost ed.), International Press Boston (1996), 99–123.
[Hi] N.J. Hitchin, The selfduality equations on a Riemann surface, Proc. London
Math. Soc. 55 (1987), 59–126.
[HKP] J. Hong, Y. Kim, P.Y. Pac, Multi-vortex solutions of the Abelian Chern-
Simons theory, Phys. Rev. Lett. 64 (1990), 2230–2233.
[JW] R. Jackiw, E.J. Weinberg, Self-dual Chern-Simons vortices, Phys. Rev. Lett.
64 (1990), 2234–2237.
[JT] A. Jaffe, C. Taubes, Vortices and monopoles, Birkhauser, Boston, (1980).
[JS1] R.L. Jerrard, H.M. Soner, Dynamics of Ginzburg-Landau vortices. Arch.
Rational Mech. Anal. 142, n. 2, (1998), 99–125.
[JS2] R.L. Jerrard, H.M. Soner, The Jacobian and the Ginzburg-Landau energy.
Calc. Var. PDE 14, n. 2 (2002), 151–191.
[JoLW] J. Jost, C.S. Lin, G. Wang, Analytic aspects of the Toda system: II. Bubbling
Behavior and existence of solutions, Comm. Pure Appl. Math. 59 (2006),
526–558.
[Jo] J. Jost, Riemannian Geometry and Geometric Analysis, Spriger-Verlag Berlin,
Heidelberg (1998) (second Edition).
[JoW1] J. Jost, G. Wang, Analytic aspects of the Toda system: I. A Moser-Trudinger
inequality, Comm. Pure Appl. Math. 54 (2001), 1289–1319.
[JoW2] J. Jost, G. Wang, Classification of solutions of a Toda system in R2 , Int. Math.
Res. Not. 6 (2002), 277–290.
[K] J. Kazdan, Prescribing the curvature of a Riemannian manifold, CBMS
Lectures AMS 57, (1984).
[KW1] J. Kazdan, F. Warner, Existence and conformal deformations of metric with
prescribed Gaussian and scalar curvature, Ann. of Math. 101 (1975), 317–331.
[KW2] J. Kazdan, F. Warner, Curvature functions for compact 2-manifolds, Ann. of
Math. 99 (1974), 14–47.
[Ki1] M.K.H. Kiessling, Statistical mechanics of classical particles with logarithmic
interaction, Comm. Pure Appl. Math. 46 (1991), 27–56.
[Ki2] M.K.H. Kiessling, Statistical mechanics approach to some problems in confor-
mal geometry, Phys. A 79 (2000), 353–368.
[KiKi] S. Kim, Y. Kim, Selfdual Chern-Simons vortices on Riemann surfaces, J.
Math. Phys. 43 (2002), 2355–2362.
[KS1] M. Kurzke, D. Sprin, Gamma limit of the non-selfdual Chern-Simons-Higgs
energy, preprint (2005).
[KS2] M. Kurzke, D. Sprin, Scaling limit of the Chern-Simons-Higgs energy, preprint
(2005).
[La] C.H. Lai (ed.), Selected Papers on Gauge Theory of Weak and Electromagnetic
Interactions, World Scientific Singapore. 1981.
[LLM] C. Lee, K. Lee, H. Min, Self-dual Maxwell-Chern-Simons solitons, Phys. Lett.
B 252 (1990), 79–83.
[L1] Y.Y. Li, On Nirenberg’s problem and related topics, Top. Meth. Nonlin. Anal.
3, n. 2 (1994), 21–233.
[L2] Y.Y. Li, Harnack type inequality: the method of moving planes, Comm. Math.
Phys. 200 (1999), 421–444.
172 G. Tarantello

[LS] Y.Y. Li, I. Shafrir, Blow up analysis for solutions of −Δu = V (x)eu in
dimension two, Ind. Univ. Math. Jour. 43, n. 4 (1994), 1255–1270.
[Li1] C.S. Lin, Uniqueness of solutions to the mean field equations for the spherical
Osanger vortex, Arch. Rat. Mech. Anal. 153 (2000), 153–176.
[Li2] C.S. Lin, Topological degree for the mean field equation on S 2 , Duke Math J.
104 (2000), 501–536.
[LiL1] C.S. Lin, M. Lucia, Uniqueness of solutions for a mean field equations on
torus, J. Diff. Eq. 229 (2006), 172–185.
[LiL2] C.S. Lin, M. Lucia, One-dimensional symmetry of periodic minimizers for a
mean field equations, Ann. Sc. Norm. Pisa 5, n. 6 (2007), 269–290.
[LiW] C.S. Lin, C.L. Wang, Elliptic functions, Green functions an the mean field
equation on tori, preprint (2006).
[Lin1] F.H. Lin, Some dynamical properties of Ginzburg-Landau vortices. Comm.
Pure Appl. Math. 49, n. 4 (1996), 323–359.
[Lin2] F.H. Lin, Vortex dynamics for the nonlinear wave equation, Comm. Pure Appl.
Math. 52, n. 6, (1999), 737–761.
[LR1] F.H. Lin, T. Riviere, Quantization property for moving line vortices, Comm.
Pure App. Math. 54, (2001), 826–850.
[LR2] F.H. Lin, T. Riviere, A Quantization property for static Ginzburg-Landau
Vortices, Comm. Pure App. Math. 54, (2001), 206–228.
2
log λ
[Lio] J. Liouville, Sur l’equation aux derivées partielles ∂∂u∂v ± 2a
λ
2 = 0, J. Math.
Pure Appl. 18 (1853), 71–72.
[Lu] M. Lucia, A blowing-up branch of solutions for a mean field equation, Calc.
Var. and P.D.E. 26 (2006), 313–333.
[LN] M. Lucia, M. Nolasco, SU (N ) Chern-Simons vortex theory and Toda systems,
Jour. Diff. Eq. 184, n. 2 (2002), 443–474.
[LZ] M. Lucia, L. Zhang, A priori estimates and uniqueness for some mean field
equations, J.D.E. 217 (2005), 154–178.
[MW] L. Ma, J. Wei, Convergence for a Liouville equation, Comment. Math. Helv.
76 (2001), 506–514.
[MaN] M. Macri’, M. Nolasco, Uniqueness of topological solutions for a class of
selfdual vortex theories, Proc. Royal Soc. Edin. 137 (2007), 847–866.
[MNR] M. Macri’, M. Nolasco, T. Ricciardi, Asymptotics for selfdual vortices on the
torus and on the plane: a technique, SIAM Math. Anal. 37, n. 1 (2005), 1–16.
[MN] A. Malchiodi, C.B. Ndiaye, Some existence results for the Toda system on
closed surfaces, preprint (2005).
[Mo] J. Moser, A sharp form of an inequality by N. Trudinger, Indiana Univ. J. 20
(1971), 1077–1092.
[NS] K. Nagashi, T. Suzuki, Asymptotic analysis for a two dimensional elliptic
eigenvalue problem with exponentially dominated nonlinearity, Asymp. Anal.
3 (1990), 173–188.
[NO] H. Nielsen, P. Olesen, Vortex-Line models for dual strings, Nucl. Phys. B 61
(1973), 45–61.
[Ni] W.M. Ni, On the elliptic equation Δu + Keu = 0 and conformal metrics with
prescribed Gaussian curvature, Invent. Math. 66 (1982), 343–352.
[Nir] L. Nirenberg, Topics in nonlinear analysis, Courant Lecture Notes AMS
(2001).
[NT1] M. Nolasco, G. Tarantello, Vortex condensates for the SU (3) Chern-Simons
theory, Comm. Math. Phys. 213 (2000), 599–639.
[NT2] M. Nolasco, G. Tarantello, On a sharp Sobolev type inequality on two
dimensional compact manifolds, Arch. Rat. Mech. Anal. 145 (1998), 161–195.
[NT3] M. Nolasco, G. Tarantello, Double vortex condensates in the Chern-Simons-
Higgs theory, Calc. Var. P.D.E. 9 (1999), 31–94.
[Ob] M. Obata, The conjectures on conformal transformations of Riemannian
manifolds, J. Diff. Geom. 6 (1971), 247–258.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 173

[OS1] H. Ohtsuka, T. Suzuki, Palais-Smale sequences relative to the Trudinger Moser

inequality, Calc. Var. P.D.E. 17 (2003), 235–255.
[OS2] H. Ohtsuka, T. Suzuki, Blow-up analysis for Liouville-type equations in
selfdual-gauge field theories, Comm. Contemp. Math. 7 (2005), 117–205.
[Ol] P. Olesen, Soliton condensation in some selfdual Chern-Simons theories, Phys.
Lett. B 265 (1991), 361–365.
[On] E. Onofri, On the positivity of the effective action in a theory of random
surfaces, Comm. Math. Phys. 86 (1982), 321–326.
[PR] F. Pacard, T. Riviere, Linear and nonlinear aspects of vortices. The Ginzburg-
Landau model, Progress in Nonlinear Differential Equations and their
Applications 39, Birkhauser Boston, Inc., Boston, MA, (2000).
[Park] R.D. Parks, Superconductivity, vol. 1 and 2, Marcel Dekker publ. (1969).
[PiR] L.M. Pismen, J. Rubinstein, Motion of vortex lines in the Ginzburg-Landau
model. Phys. D. 47, n. 3 (1991), 353–360.
[Po] S. Pokorski et al. Gauge Field Theories, Cambridge Monographs on Mathe-
matical Physics, Cambridge Univ. Press (2000).
[PT] J. Prajapat, G. Tarantello, On a class of elliptic problems in R2 : Symmetry
and Uniqueness results, Proc. Roy. Soc. Edinburgh 131A (2001), 967–985.
[PS] M.K. Prasad, C.M. Sommerfield, Exact classical solutions for the ’t Hooft
monopole and the Julia-Zee dyon, Phys. Rev. Lett. 35 (1975), 760–762.
[Q] C.Quigg, Gauge theory of Strong, Weak and Electroweak interactions, West-
view Press (1997).
[Ra] R. Rajaraman, Solitons and Instantons, North Holland publ. (1982).
[Ri1] T. Ricciardi, Asymptotics for Maxwell-Chern-Simons Multivortices, NonLin-
ear Analysis TMA 50 (2002), 193–1106.
[Ri2] T. Ricciardi, Multiplicity for a nonlinear fourth order elliptic equation in
Maxwell-Chern-Simons vortex theory, Diff. Int. Eq. 17, n. 3-4 (2004), 369–390.
[Ri3] T. Ricciardi, On a nonlinear elliptic system from Maxwell-Chern-Simons
vortex theory, Asympt. Anal. 2 (2003), 113–126.
[RT1] T. Ricciardi, G. Tarantello, Self-dual vortices in the Maxwell-Chern-Simons-
Higgs theory, Comm. Pure Appl. Math. 53 (2000), 811–851.
[RT2] T. Ricciardi, G. Tarantello, On a periodic boundary value problem with
exponential nonlinearity, Diff. Int. Eqs. 11, n. 5 (1998), 745–753.
[SS] E. Sandier, S. Serfaty, Vortices in the magnetic Ginzburg-Landau model,
Progress in Nonlinear Differential Equations and their Applications, vol. 70
Birkhauser, Boston (2007).
[Sch] J.R. Schrieffer, The Theory of Superconductivity, Benjamin publ. (1964).
[Sh] I. Shafrir, Une inegalité de type sup + inf pour l’equation −Δu = V (x)eu ,
C.R. Acad. Sci. Paris 315 (1992), 159–164.
[SW1] I. Shafrir, G. Wolansky, Moser Trudinger and logarithmic HLS inequalities for
systems, J. Europ. Math. Sc. 7, n. 4 (2005), 413–448.
[SW2] I. Shafrir, G. Wolansky, The logarithmic HLS inequalities for systems on
compact manifolds, J. Funct. Anal. 227 (2005), 200–226.
[SY1] J. Spruck, Y. Yang, The existence of non-topological solutions in the self-dual
Chern-Simons theory, Comm. Math. Phys. 149 (1992), 361–376.
[SY2] J. Spruck, Y. Yang, On Multivortices in the Electroweak Theory I: Existence
of Periodic Solutions, Comm. Math. Phys. 144 (1992), 1–16.
[SY3] J. Spruck, Y. Yang, On Multivortices in the Electroweak Theory II:Existence
of Bogomol’nyi solutions in R2 , Comm. Math. Phys. 144 (1992), 215–234.
[St] M. Struwe, Variational Methods, Application to Partial Differential Equations
and Hamiltonian systems, 3rd edition Springer 34, (2000).
[ST] M. Struwe, G. Tarantello, On the multivortex solutions in the Chern-Simons
gauge theory, Boll. U.M.I. Sez. B Artic. Ric. Mat. 1 (1998), 109–121.
174 G. Tarantello

[Su1] T. Suzuki, Global analysis for a two dimensional elliptic eigenvalue problem
with exponential nonlinearity, Ann. I.H.P. Analyse Non Linéaire 9 (1992),
367–398.
[Su2] T. Suzuki, Two dimensional Emden-Fowler equations with exponential nonlin-
earities, Nonlinear Diffusion Equations and their equilibrium state 3 (1992),
493–512, Birkauser, Boston.
[’tH1] G. ’t Hooft, Computation of the quantum effects due to a four dimensional
pseudoparticle, Phys. Rev. D 14 (1976), 3432–3450.
[’tH2] G. ’t Hooft, A property of electric and magnetic flux in non abelian gauge
theories, Nucl. Phys. D 153 (1979), 141–160.
[T1] G. Tarantello, Multiple condensates solutions for the Chern-Simons-Higgs
theory, J. Math. Phys. 37, n. 8 (1996), 3769–3796.
[T2] G. Tarantello, On Chern Simons vortex theory, Nonlinear PDE’s and physical
modeling: Superfluidity, Superconductivity and Reactive flows, H. Berestycki
ed. Kluver Academic publ., (2002), 507–526.
[T3] G. Tarantello, Selfdual Maxwell-Chern-Simons vortices, Milan J. Math. 72
(2004), 29–80.
[T4] G. Tarantello, Analitycal aspects of Liouville-type equations with singular
sources, Handbook of Differential Equations. Stationary partial differential
equations, vol 1. Elsevier Sciences, M.Chipot, P.Quittner Eds.
[T5] G. Tarantello, A quantization property for blow up solutions of singular
Liouville-type equations, Journal Func. Anal. 219 (2005), 368–399.
[T6] G. Tarantello, An Harnack inequality for Liouville-type equations with
singular sources, Indiana Univ. Math J. 54, n.2 (2005), 599–615.
[T7] G. Tarantello, Uniqueness of Selfdual periodic Chern-Simons vortices of
topological-type, Calc. Var. PDE, 29 (2007), 191–217.
[T8] G. Tarantello, Selfdual gauge field vortices: an analytical approach, Progress
in Nonlinear Differential Equations and their Applications, vol. 72 Birkhauser,
Boston (2008).
[Ta1] C. Taubes, Arbitrary N-vortex solutions for the first order Ginzburg-Landau
equations, Comm. Math. Phys. 72 (1980), 277–292.
[Ta2] C.H. Taubes, On the equivalence of the first and second order equations for
gauges theories, Comm. Math. Phys. 75 (1980), 207–227.
[Ta3] C.H. Taubes, The existence of a non-minimal solution to the SU(2) Yang-
Mills-Higgs equations in R3 , Part I and II, Comm. Math. Phys. 86 (1982),
257–320.
[Tra] A. Trautmann, Differential geometry for physicists Stony Brook lectures.
Monographs and Textbooks in Physical Science, 2. Bibliopolis, Naples, 1984.
[Tr] N.D. Trudinger, On imbedding into Orlicz spaces and some applications, J.
Math. Mech. 17 (1967), 473–483.
[Va] P. Valtancoli, Classical and Chern-Simons Vortices on curved spaces, Int. J.
Mod. Phys. A, 7, n. 18 (1990), 4335–4352.
[W] G. Wang, Moser-Trudinger inequality and Liouville systems, C.R. Acad. Sci.
Paris 328 (1999), 895–900.
[WW1] G. Wang, J. Wei, On a conjecture of Wolanski, Nonlinear Analysis TMA 48
(2002), 927–937.
[WW2] G. Wang, J. Wei, Steady state solutions of a reaction-diffusion system
modelling chemotaxis, Math. Nachr. 233–234 (2002), 221–236.
[Wa] R. Wang, The existence of Chern-Simons vortices, Comm. Math. Phys. 137
(1991), 587–597.
[WY] S. Wang, Y. Yang, Abrikosov’s vortices in the critical coupling, SIAM J. Math.
Anal. 23 (1992), 1125–1140.
[Wit] E. Witten, Some exact Multipseudoparticle solutions of classical Yang-Mills
theory, Phys. Rev. Lett. 38 (1997), 121–124.
On Some Elliptic Problemsin the Study of SelfdualChern-Simons Vortices 175

[Wo] G. Wolanski, On the evolution of self-interacting clusters and applications to

semilinear equations with exponential nonlinearity, J. Anal. Math. 59 (1992),
251–272.
[YM] C.N. Yang, R. Mills, Conservation of isotopic spin and isotopic invariance,
Phys. Rev. Lett. 96 (1954), 191–195.
[Y1] Y. Yang, Solitons in Field Theory and Nonlinear Analysis, Springer Mono-
graphs in Mathematics, Springer-Verlag New York, (2001).
[Y2] Y. Yang, Topological solitons in the Weinberg-Salam theory, Physica D 101
(1997), 55–94.
[Y3] Y. Yang, The relativistic non-abelian Chern-Simons equations, Comm. Math.
Phys. 186 (1997), 199–218.
[Y4] Y. Yang, On a system of nonlinear elliptic equations arising in theoretical
physics, J. Func. Anal. (2000), 1–36.
The k-Hessian Equation

Xu-Jia Wang

Abstract The k-Hessian is the k-trace, or the kth elementary symmetric

polynomial of eigenvalues of the Hessian matrix. When k ≥ 2, the k-Hessian
equation is a fully nonlinear partial differential equations. It is elliptic when
restricted to k-admissible functions. In this paper we establish the exis-
tence and regularity of k-admissible solutions to the Dirichlet problem of
the k-Hessian equation. By a gradient flow method we prove a Sobolev
type inequality for k-admissible functions vanishing on the boundary, and
study the corresponding variational problems. We also extend the definition
of k-admissibility to non-smooth functions and prove a weak continuity of
the k-Hessian operator. The weak continuity enables us to deduce a Wolff
potential estimate. As an application we prove the Hölder continuity of weak
solutions to the k-Hessian equation. These results are mainly from the papers
[CNS2, W2, CW1, TW2, Ld] in the references of the paper.

Key Words: Hessian equation, a priori estimates, Sobolev inequality, varia-

tional problem, Hessian measure, potential estimate.

AMS subject classification: 35J60, 35J20, 35A15, 28A33.

1 Introduction

Let Ω be a bounded, smooth domain in the Euclidean space Rn . In this note

we study the k-Hessian equation

Sk [u] = f in Ω, (1.1)

X.-J. Wang
Mathematical Sciences Institute, Australian National University,
Canberra, ACT 0200, Australia
e-mail: [email protected]
This work was supported by the Australian Research Council.

S.-Y.A. Chang et al. (eds.), Geometric Analysis and PDEs, 177

Lecture Notes in Mathematics 1977, DOI: 10.1007/978-3-642-01674-5 5,

c Springer-Verlag Berlin Heidelberg 2009
178 X.-J. Wang

where 1 ≤ k ≤ n, Sk [u] = σk (λ), λ = (λ1 , · · · , λn ) are the eigenvalues of the

Hessian matrix (D2 u), and

σk (λ) = λi1 · · · λik (1.2)
i1 <···<ik

is the k-th elementary symmetric polynomial. The k-Hessian equation includes

the Poisson equation (k = 1)

−Δu = f, (1.3)

and the Monge-Ampère equation (k = n)

detD2 u = f, (1.4)

as special examples.
We say a second order partial differential equation

F (D2 u, Du, u, x) = 0 (1.5)

is fully nonlinear if F (r, p, z, x) is nonlinear in r. The k-Hessian equation is

fully nonlinear when k ≥ 2. We say F is elliptic (or degenerate elliptic) with
respect to a solution u if the matrix {Fij } is positive definite (or positive
semi-definite) at (r, p, z, x) = (D2 u(x), Du(x), u(x), x), where Fij = { ∂r ∂F
ij
}.
We say F is uniformly elliptic if there exist positive constants Λ and λ such
that
λI ≤ {F ij } ≤ ΛI, (1.6)
where I is the unit matrix. We also say F is elliptic if −F is.
The Monge-Ampère equation (1.4) is elliptic if and only if the function u is
uniformly convex or concave. For the k-Hessian equation, it is elliptic when u
is k-admissible [CNS2], namely the eigenvalues λ(D2 u) lie in the convex cone
Γk , which will be introduced in Section 2 below. Fully nonlinear equations of
mixed type are very difficult. In this note we restrict ourself to k-admissible
solutions to the k-Hessian equation.
There are many other important fully nonlinear equations, see §11 below
for examples. But the k-Hessian equation (1.1) is variational, and when
restricted to k-admissible solutions, it enjoys many nice properties which are
similar to those of the Poisson equation. In this paper we discuss the regular-
ity, variational properties, and local behaviors of solutions to the k-Hessian
equation.
We divide this note into a number of sections.
In §2 we introduce the notion of k-admissible functions, and show that the
k-Hessian equation is elliptic at k-admissible functions. We also collect some
inequalities related to the polynomial σk .
The k-Hessian Equation 179

In §3 we establish the global a priori estimates and prove the existence of

solutions to the Dirichlet problem.
In §4 we establish the interior gradient and second derivative estimates.
From the interior gradient estimate we also deduce a Harnack inequality.
In §5 we use gradient flow to prove Sobolev type inequalities for
k-admissible functions which vanish on the boundary. That is
 1/(k+1)
uLp(Ω) ≤ C (−u)Sk [u] , (1.7)
Ω

where C depends only on n, k, Ω; p = n(k+1) n−2k if k < 2 , p < ∞ if k = 2 ;

n n

and p = ∞ if k > 2 . Moreover, the corresponding embedding of k-admissible

n

functions into Lp space is compact when p is below the critical exponent. As

an application we give an L∞ estimate for solutions to the k-Hessian equation
(1.1) when f ∈ Lp (Ω) with p > 2k n
if k ≤ n2 , or p = 1 if k > n2 .
In §6 we use the Sobolev type inequality (1.7) to study variational problems
of the k-Hessian equation. We prove the existence of a min-max solution to
the Hessian equation in the sub-critical and critical growth cases.
In §7 we present some local integral estimates. In particular we show that
1,p nk
a k-admissible function belongs to Wloc (Ω) for any p < n−k .
In §8 we extend the notion of k-admissible functions to nonsmooth func-
tions; and prove that for any k-admissible function u, we can assign a measure
μk [u] to u such that if a sequence of k-admissible functions {uj } converges
to u almost everywhere, then μk [uj ] converges to μk [u] weakly as measures.
As an application we prove the existence of weak solutions to the k-Hessian
equation.
This weak continuity has many other applications as well, in particular it
enables us to establish various potential theoretical results for k-admissible
functions. In §9 we prove a Wolff potential estimate, and deduce a necessary
and sufficient condition for a weak solution to be Hölder continuous.
In §10, we include some a priori estimates for the parabolic Hessian
equations used in previous sections.
In the last Section 11, we give more examples of fully nonlinear elliptic
equations.
Main references for this note are [CNS2, W2, CW1, TW2, Ld]. There are
many other works on the k-Hessian equations. The materials in §2 and §3 are
mostly taken from [CNS2], but for the key double normal derivative estimate
we adapt the approach from [T]. See also [I] for the k-Hessian equation for
some k. The interior derivative estimates in §4 are from [CW1], but for the
Monge-Ampère equation they were first established by Pogorelov [P]. The
Sobolev type inequalities in §5 were proved by K.S. Chou for convex functions,
and in [W2] for general k-admissible functions by a gradient flow method. The
existence of min-max solutions in §6 was first obtained by K.S. Chou [Ch1]
for the Monge-Ampère equation and later in [CW1] for 2 ≤ k ≤ n2 . See also
[W1] for the Monge-Ampère equation by a degree theory method, which also
180 X.-J. Wang

applies to the case n2 < k < n by the embedding in Theorem 5.1. The local
integral estimates in §7 and weak continuity in §8 can be found in [TW2].
The Wolff potential estimate and Hölder continuity of k-admissible solutions
in §9 were proved in [Ld].
The result in §6.4 on the variational problem in the critical growth case
was not published before, it was included in the preprint [CW2]. The proof
of the weak continuity in §8, which uses ideas from [TW1,TW5], is different
from that in [TW2]. As the reader will see below, most results in the note are
generalization of the counterparts for the Poisson equation. But the study of
fully nonlinear equations requires new techniques and is usually more com-
plicated, in particular for estimates near the boundary. These results and
techniques can also be used in other problems. See e.g., [FZ, KT, STW].

2 Admissible Functions

2.1 Admissible Functions

We say a function u ∈ C 2 (Ω) ∩ C 0 (Ω) is k-admissible if

λ(D2 u) ∈ Γ k , (2.1)

where Γk is an open symmetric convex cone in Rn , with vertex at the origin,

given by

Γk = {(λ1 , · · · , λn ) ∈ Rn | σj (λ) > 0 ∀ j = 1, · · · , k}. (2.2)

Clearly σk (λ) = 0 for λ ∈ ∂Γk ,

Γn ⊂ · · · ⊂ Γk ⊂ · · · ⊂ Γ1 ,

Γn is the positive cone,

Γn = {(λ1 , · · · , λn ) ∈ Rn | λ1 > 0, · · · , λn > 0},

and Γ1 is the half space {λ ∈ Rn | Σλi > 0}. A function is 1-admissible if

and only if it is sub-harmonic, and an n-admissible function must be convex.
For any 2 ≤ k ≤ n, a k-admissible function is sub-harmonic, and the set of
all k-admissible functions is a convex cone in C 2 (Ω).
The cone Γk may also be equivalently defined as the component {λ ∈
RN | σk (λ) > 0} containing the vector (1, · · · , 1), and characterized as

Γk = {λ ∈ Rn  0 < σk (λ) ≤ σk (λ + η) for all ηi ≥ 0, ∈ R}. (2.3)
The k-Hessian Equation 181

We note that the k-Hessian operator Sk is also elliptic or degenerate elliptic

if λ(D2 u) ∈ −Γ k . But by making the change u → −u it suffices to consider
functions with eigenvalues λ ∈ Γk . In this note we consider functions with
eigenvalues in Γk only.

2.2 Admissible Solution is Elliptic

We show that if u is k-admissible, the matrix

 ij   ∂
Sk (A) = σk (λ(A))} ≥ 0 (2.4)
∂aij

is positive semi-definite at A = D2 u and so the k-Hessian operator is (degen-

erate) elliptic. To prove (2.4), note that the k-Hessian operator can also be
written in the form
Sk [u] = [D2 u]k , (2.5)
where for a matrix A = (aij ), [A]k denotes the sum of the k th principal
minors. Therefore
Sknn [u] = [D2 u]k−1 , (2.6)
where [D2 u] = {uxi xj }1≤i,j≤n−1 . Denote
 2  
[D u] , 0
D̄2 u = .
0, unn

One easily verifies that

[D̄2 u]m ≥ [D2 u]m ∀ 1 ≤ m ≤ k.

Hence by (2.2), λ(D̄2 u) ∈ Γ k . By (2.3) it follows that

∂
Sknn [u] = [D2 u]k−1 = σk (λ) ≥ 0 (λ = λ(D̄2 u)). (2.7)
∂λn
Note that (2.7) also holds after a rotation of coordinates, so the k-Hessian
equation is (degenerate) elliptic if u is k-admissible.
When u is k-admissible, Sk [u] is nonnegative. Therefore in our investiga-
tion of the k-Hessian equation, we always assume that f is nonnegative. If
f is positive and u ∈ C 2 (Ω), Sk [u] is elliptic. Note that we allow that the
eigenvalues λ(D2 u) lie on the boundary of Γk , and in such case the k-Hessian
equation may become degenerate elliptic.
182 X.-J. Wang

2.3 Concavity

When u is k-admissible,
1/k  1/k
Sk [u] = σk (λ(D2 u)) ,

is concave when regarded as a function of r = D2 u. In other words,

 1/k
aij ast ∂u2ij ust Sk [u] ≤ 0 (2.8)

for any symmetric matrix {aij }. This property follows from the concavity of
1/k
σk (λ) in Γk (see (xii) in §2.5 below). Indeed, when uij is diagonal, one can
verify (2.8) directly by the expression (2.5). When uij is not diagonal, by a
rotation of coordinates yα = cαi xi such that uαβ is diagonal, one has
 
a∗αβ a∗γδ ∂u2αβ uγδ Sk [u] ≤ 0,
1/k 1/k
aij ast ∂u2ij ust Sk [u] =

where a∗αβ = aij cαi cβj , subscripts i, j, s, t mean derivatives in x and subscripts
α, β, γ, δ mean derivatives in y. The concavity is needed in establishing the
regularity of fully nonlinear elliptic equations.

2.4 A Geometric Assumption on the Boundary

In order that there exists a smooth k-admissible function which vanishes on

ϕΩ, the boundary ϕΩ must satisfy a geometric condition, that is

σk−1 (κ) ≥ c0 > 0 on ϕΩ (2.9)

for some positive constant c0 , where κ = (κ1 , · · · , κn−1 ) denote the principal
curvatures of ϕΩ with respect to its inner normal. Indeed, let u ∈ C 2 (Ω) be
a k-admissible function which vanishes on ϕΩ. For any fixed point x0 ∈ ϕΩ,
by a translation and rotation of coordinates, we may assume that x0 is the
origin and locally ϕΩ is given by xn = ρ(x ) such that en = (0, · · · , 0, 1) is
the inner normal of ϕΩ at x0 , where x = (x1 , · · · , xn−1 ). Differentiating the
boundary condition u(x , ρ(x )) = 0, we get

uij (0) + un ρij (0) = 0. (2.10)

By our choice of coordinates, the principal curvatures of ϕΩ at x0 are the

eigenvalues of {ρij (0)}1≤i,j≤n−1 . When u is k-admissible, it is subharmonic
and so un (x0 ) < 0. We obtain

Sknn [u] = |un |k−1 σk−1 (κ). (2.11)

Hence (2.9) follows from (2.4) provided λ(D2 u) ∈ Γk .

The k-Hessian Equation 183

In this note we call a domain whose boundary satisfies (2.9) (k −1)-convex.

When k = n, it is equivalent to the usual convexity. In the following we always
assume that Ω is (k − 1)-convex.
If Ω is (k − 1)-convex, then for any smooth function ϕ on ϕΩ, there is a
function u, which is k-admissible in a neighborhood of ϕΩ and satisfies u = ϕ
on ϕΩ. Indeed, if ϕ = 0, let u(x) = −dx + td2x , where x ∈ Ω and dx is distance
from x to ϕΩ. Then u is k-admissible near ϕΩ provided t is sufficiently large.
We refer the reader to [GT] for the computation of the second derivatives of
the distance function. For a general boundary value ϕ, extend ϕ to Ω such
that it is harmonic in Ω. Then ϕ + σu is k-admissible near ϕΩ for large σ,
and Sk [ϕ + σu] can be as large as we want provided σ is sufficiently large.
Note that the function u is defined only in a neighborhood of ϕΩ. But it
suffices for the a priori estimates in §3. By the existence of solutions to the
Dirichlet problem (Theorem 3.4), there is a k-admissible function u defined
in the whole domain Ω such that u = ϕ on ϕΩ.

2.5 Some Algebraic Inequalities

We collect some inequalities related to the polynomial σk (λ), which are

needed in our investigation of the k-Hessian equation.
Denote σ0 = 1 and σk = 0 for k > n. Assume λ ∈ Γk . Arrange λ =
(λ1 , · · · , λn ) in descending order, namely λ1 ≥ · · · ≥ λn . Denote σk;i =
∂
σk (λ)|λi =0 , so that ∂λ i
σk (λ) = σk−1,i (λ). The following ones are easy to
verify

(i) σk (λ) = σk;i (λ) + λi σk−1;i (λ),

n
(ii) σk;i (λ) = (n − k)σk (λ),
i=1
(iii) σk−1,n (λ) ≥ · · · ≥ σk−1,1 (λ) > 0,
(iv) λk ≥ 0 and σk (λ) ≤ Cn,k λ1 · · · λk .

We also have

(v) σk (λ)σk−2 (λ) ≤ Cn,k [σk−1 (λ)]2 ,

(vi) σk (λ) ≤ Cn,k [σl (λ)]k/l , 1 ≤ l < k.

Furthermore we have

(vii) λ1 σk−1,1 (λ) ≥ Cn,k σk (λ).

n
(viii) σk−1;k (λ) ≥ Cn,k σk−1;i (λ),
i=1
(ix) σk−1;k (λ) ≥ Cn,k σk−1 (λ),
n
(x) σk;i (λ) ≥ Cn,k [σk (λ)]n(k−1)/k .
i=1
184 X.-J. Wang

In the above the constant Cn,k may change from line to line. There are more
inequalities useful in the study of the k-Hessian equation. For example, we
have

(xi) μi σk−1,i ≥ k[σk (μ)]1/k [σk (λ)]1−1/k ∀ λ, μ ∈ Γk ,
∂2
(xii) { σk (λ)} ≤ 0 ∀ λ ∈ Γk .
∂λi ∂λj
1/k
the last inequality means that σk (λ) is concave in Γk . We refer the reader
to [CNS2, LT, Lg] for these and more inequalities related to σk .

3 The Dirichlet Problem

In this section we study the existence and regularity of solutions to the

Dirichlet problem of the k-Hessian equation,

Sk [u] = f (x) in Ω, (3.1)

u = ϕ on ϕΩ,

where Ω is a bounded, (k − 1)-convex domain in Rn with C 3,1 boundary,

ϕ ∈ C 3,1 (ϕΩ), f ≥ 0, f ∈ C 1,1 (Ω).

3.1 A priori Estimates

First we establish the global estimate for the second derivatives.

Theorem 3.1 (CNS2, T) Let u ∈ C 3,1 (Ω) be a k-admissible solution to

the Dirichlet problem (3.1). Assume that Ω is (k − 1)-convex, ϕΩ ∈ C 3,1 ,
ϕ ∈ C 3,1 (ϕΩ), f ≥ f0 > 0, and f 1/k ∈ C 1,1 (Ω). Then we have the a priori
estimate
uC 1,1 (Ω) ≤ C, (3.2)
where C depends only on n, k, Ω, f0 , ϕC 3,1 (ϕΩ) and f C 1,1 (Ω) .

Proof. First consider the L∞ estimate. Let w = 12 a|x|2 − b, where the con-
Sk [w] > f in Ω and w ≤ ϕ on ϕΩ. Then
stants a, b are chosen large such that
w − u satisfies the elliptic equation aij (w − u)ij > 0 in Ω and w − u ≤ 0 on
1
ϕΩ, where aij = 0 Skij [u+t(w−u)]dt. It follows that w ≤ u in Ω. Extend ϕ to
Ω such that it is harmonic. By the comparison principle we have w ≤ u ≤ ϕ
in Ω.
The k-Hessian Equation 185

1/k
Next consider the gradient estimate. Denote F [u] = Sk [u], fˆ = f 1/k .
Differentiating the equation
F [u] = fˆ (3.3)
in direction xl , one obtains
L[ul ] = fˆl ,
where L = Fij ∂ij is the linearized equation of F , Fij = Fuij . So |L[ul ]| ≤ C.
Let w = 12 a|x|2 . By (ii) and (vi) above, L[w] ≥ c1 a > 0 for some positive con-
stant c1 > 0 depends only on n, k. Hence L[w ± ul ] ≥ 0, provided a is chosen
suitably large. It follows that w ± ul attains its maximum on the boundary
ϕΩ. Hence
sup |Du(x)| ≤ C(1 + sup |Du(x)|). (3.4)
x∈Ω x∈ϕΩ

Next let ŵ = ϕ + σu be the function in §2.4. Denote N = {x ∈ Ω | ŵ(x) >

w(x)}. Then when σ is sufficiently large, N is a neighborhood of ϕΩ, and
Sk [ŵ] > f in N . Therefore by the comparison principle, ŵ ≤ u ≤ ϕ in
N . Hence by the boundary condition ŵ = u = ϕ on ϕΩ, we infer that
∂γ ϕ ≤ ∂γ u ≤ ∂γ ŵ, where γ is the unit outer normal to ϕΩ. Hence Du is
bounded on ϕΩ.
Finally consider the second derivative estimate. Since u is sub-harmonic, it
suffices to prove that uξξ ≤ C for any unit vector ξ. Differentiating equation
(3.3) twice in direction ξ, we obtain, by the concavity of F ,

L[uξξ ] ≥ fˆξξ .

Hence L[Cw + uξξ ] ≥ 0 for a suitably large constant C and so

sup uξξ ≤ C + sup uξξ . (3.5)

Ω ϕΩ

Therefore we reduce the estimate to the boundary.

For any given boundary point x0 ∈ ϕΩ, by a translation and a rotation of
the coordinates we assume that x0 is the origin and locally ϕΩ is given by

xn = ρ(x ) (3.6)

such that Dρ(0) = 0, where x = (x1 , · · · , xn−1 ). Differentiating the bound-

ary condition u = ϕ on ϕΩ twice, we have, for 1 ≤ i, j ≤ n − 1,

uij (0) + un (0)ρij (0) = ϕij (0) + ϕn (0)ρij (0). (3.7)

Hence
|Dij u(0)| ≤ C i, j ≤ n − 1. (3.8)
Next we establish
|uin (0)| ≤ C i < n. (3.9)
186 X.-J. Wang

By a rotation of the x1 , · · · , xn−1 axes, we assume that x1 , · · · , xn−1 are the

principal directions of ϕΩ at the origin. Let T = ∂i + κi (0)(xi ∂n − xn ∂i ),
where κi is the principal curvature of ϕΩ in direction xi , 1 ≤ i ≤ n − 1. One
can verify that

|T (u − ϕ)| ≤ C|x |2 |∂γ (u − ϕ)| ≤ C|x |2 | on ϕΩ.

Next observing that Sk is invariant under rotation of coordinates and (xi ∂n −

xn ∂i ) is an infinitesimal generator of a rotation, we have T F [u] = L[T (u)].
Hence
|L(T (u − ϕ))| ≤ C(1 + Σi Fii ).
Let
w = ρ(x ) − xn − δ|x |2 + Kx2n , (3.10)
where K > 1 large and δ > 0 small are constants. By the assumption that Ω
is (k − 1)-convex, the function w is k-admissible in Bε (0) ∩ Ω for small ε > 0.
By the concavity of F ,

L[w] ≥ F [u + w] − F [u] ≥ F [w] − F [u]

1
≥ c1 K 1/k − C ≥ c1 K 1/k
2
for some constants c1 depending on n, k, and δ, provided K is sufficiently
large. Choose a K  large such that L[K  w ± T (u − ϕ)] ≥ 0. It follows that the
maximum of K  w ± T (u − ϕ) in Bε ∩Ω is attained on the boundary ∂(Bε ∩Ω).
But on the boundary ∂(Bε ∩ Ω), it is easy to see that
1
w ≤ − δ|x |2 on ϕΩ ∩ Bε (0),
2
w < 0 on Ω ∩ ∂Bε (0).

Hence K  w ± T (u − ϕ) ≤ 0 provided K  is chosen large enough. Hence

K  w ± T (u − ϕ) attains its maximum 0 at the origin and we obtain

|∂n (T (u − ϕ))| ≤ K  |∂n w| ≤ C,

from which (3.9) follows.

Finally we consider the double normal derivative estimate

unn (0) ≤ C. (3.11)

If ϕ = 0, by (3.7) we have uij (0) = (−un )ρij . By the geometric assumption

(2.9), we have

Sknn [u] = σk−1 [λ(D2 u) ] = |un |k−1 σk−1 (κ) > 0,
The k-Hessian Equation 187

where (D2 u) = (uij )1≤i,j≤n−1 . Note that

Sk [u] = unn σk−1 {λ[(D2 u) ]} + R = f, (3.12)

where R is the rest terms which do not involve unn , and so is bounded by
(3.8) and (3.9). Hence unn (0) must be bounded.
For general boundary function ϕ, we adapt the approach from [T]. By
(3.12) it suffices to prove σk−1 {λ[(D2 u) ]} > 0 on ϕΩ. For any boundary
point x ∈ ϕΩ, let ξ (1) , · · · , ξ (n−1) be an orthogonal vector field on ϕΩ. Denote
(i)
∇i = ξm Dm u,
(j) (j)
∇ij u = ξm
(i)
ξl Dml u, Cij = ξm
(i)
ξl Dm γl ,

and ∇2 u = {∇ij u}, C = {Cij }, where γ is the unit inner normal of ϕΩ at x.

Then we have
λ[(D2 u) ] = λ[∇2 u](x).
Similar to (3.7) we have

∇2 u = Dγ (u − ϕ)C + ∇2 ϕ. (3.13)

For any (n − 1) × (n − 1)-matrix r with eigenvalues (λ1 , · · · , λn−1 ), denote

G(r) = [σk−1 (λ)]1/(k−1) .

∂G
and Gij = ∂r ij
. Assume that inf x∈ϕΩ G(∇2 u) is attained at x0 . Then by
(3.13) and the concavity of G,

Gij ij
0 [Dγ (u − ϕ)Cij (x) + ∇ij ϕ(x)] ≥ G0 [Dγ (u − ϕ)Cij (x0 ) + ∇ij ϕ(x0 )]

for any x ∈ ϕΩ, where Gij ij 2

0 = G (∇ u(x0 )). We can also write the above
formula in the form

Gij
0 Cij (x0 )[Dγ (u − ϕ)(x) − Dγ (u − ϕ)(x0 )]
≥ Gij
0 {[Dγ (u − ϕ)(x) − Dγ (u − ϕ)(x0 )][Cij (x0 ) − Cij (x)]
+Dγ (u − ϕ)(x0 )][Cij (x0 ) − Cij (x)] − [∇ij ϕ(x) − ∇ij ϕ(x0 )]}

Assume that near x0 , ϕΩ is given by (3.6) with

1 n−1
ρ(x ) = κi x2i + O(|x |3 ).
2 i=1

Then we have Cij (x0 ) = ∂i γj = κi δij . Recall that Ω is (k − 1)-convex. The

eigenvalues of {Cij − c1 δij } (as a vector in Rn−1 ) lies in Γk−1 , provided c1 is
sufficiently small. Hence Gij 0 (Cij − c1 δij ) ≥ 0 at x0 , and so

Gij
0 Cij (x0 ) ≥ c1 0 ≥ δ0 > 0.
Gii
188 X.-J. Wang

Therefore we obtain

Dn (u − ϕ)(x) − Dn (u − ϕ)(x0 ) ≤ "(x ) + C|x |2 ,

where " is a linear function of x with "(0) = 0. Denote

v(x) = Dn (u − ϕ)(x) − Dn (u − ϕ)(x0 ) − "(x ).

We have
v(x) ≤ C|x |2 ∀ x ∈ ϕΩ. (3.14)
Differentiating equation (3.3) we have

|L(v)| ≤ C(1 + F ii ), (3.15)
 ij
where L = F ∂ij is the linearized operator of F .
Let w be the function given in (3.10). Then by (3.15) we can choose K 
sufficiently large such that L(K  w) ≥ ±L(v) in Bε ∩ Ω. By (3.14), we can
also choose K  large such that K  w + v ≤ 0 on ∂(Bε ∩ Ω). By the comparison
principle it follows that K  w + v ≤ 0 in Bε ∩ Ω. Hence K  w + v attains its
maximum at x0 . We obtain ∂n (K  w + v) ≤ 0 at x0 , namely unn (x0 ) ≤ C.
To complete the proof, one observes that in (3.12),
 ∂2
R=− u21i Sk [u] ≤ 0.
∂u11 ∂uii
Hence
f f0
σk−1 {λ[(D2 u) ]}(x0 ) ≥ (x0 ) ≥ . (3.16)
unn unn (x0 )
Recall that σk−1 {λ[(D2 u) ]} attains its minimum at x0 . Hence by (3.12) we
obtain uγγ (x) < C at any boundary point x ∈ ϕΩ.

By the a priori estimate (3.2), equation (3.1) becomes uniformly elliptic if

f is strictly positive. The uniform ellipticity follows from inequality (iii) in
§2.5. To get the higher order derivative estimates, we employ the regularity
theory of fully nonlinear, uniformly elliptic equations.

3.2 Regularity for Fully Nonlinear, Uniformly Elliptic

Equation

We say a fully nonlinear elliptic operator F is concave if F , as a function of

r = D2 u, is a concave function. From §2.3, the k-Hessian equation is concave
when u is k-admissible and the equation is written in the form (3.3).
The k-Hessian Equation 189

The regularity theory of fully nonlinear elliptic equations was established

by Evans and Krylov independently. Their proof is based on Krylov-Safonov’s
Hölder estimates for linear, uniformly elliptic equation of non-divergent form.
Theorem 3.2 Consider the fully nonlinear, uniformly elliptic equation

F (D2 u) = f (x) in Ω. (3.17)

u = ϕ on ϕΩ,

Suppose F is concave, F ∈ C 1,1 , f ∈ C 1,1 (Ω), and u ∈ W 4,n (Ω) is a solution

of (3.17). Then there exists α ∈ (0, 1) depending only on n, λ, Λ (the constants
in (1.6)) such that for any Ω ⊂⊂ Ω,

uC 2,α(Ω ) ≤ C, (3.18)

where C depends only on n, λ, Λ, α, Ω, dist(Ω , ϕΩ), f C 1,1 (Ω) , and supΩ |u|.
If furthermore ϕ ∈ C 3,1 (Ω), ϕΩ ∈ C 3,1 , and f ∈ C 1,1 (Ω), then

uC 2,α (Ω) ≤ C, (3.19)

where C depends only on n, λ, Λ, α, ϕΩ, f C 1,1 (Ω) , ϕC 3,1 (Ω) and supΩ |u|.
From (3.19) one also obtains C 3,α estimates by differentiating the equation
(3.17) and apply the Schauder theory for linear, uniformly elliptic equations.
Theorem 3.2 also extends to more general equations of the form (1.1) provided
F satisfies certain structural conditions. We refer the readers to [E, K1, GT]
for details.
As a corollary of Theorem 3.2, we obtain the higher order derivative
estimate for the k-Hessian equation.
Theorem 3.3 Let u ∈ C 3,1 (Ω) be a k-admissible solution of (3.1). Assume
that Ω is (k − 1)-convex, f ∈ C 1,1 (Ω), and f ≥ f0 > 0 in Ω. Then we have

uC 3,α (Ω) ≤ C, (3.20)

where α ∈ (0, 1), C depends only on n, k, α, f0 , Ω, ϕC 3,1 (ϕΩ) , and

f C 1,1 (Ω) .

3.3 Existence of Smooth Solutions

By Theorem 3.3 and the continuity method, we obtain the existence of smooth
solutions to the Dirichlet problem (3.1).
Theorem 3.4 Assume that Ω is (k − 1)-convex, ϕΩ ∈ C 3,1 , f ∈ C 1,1 (Ω),
and f ≥ f0 > 0. Then there is a unique k-admissible solution u ∈ C 3,α (Ω) to
the Dirichlet problem (3.1).
190 X.-J. Wang

Proof. We apply the continuity method to the Dirichlet problem

Sk [ut ] = ft in Ω,
ut = ϕt on ϕΩ,

where t ∈ [0, 1], ft = Cnk (1 − t) + tf , ϕt = 1−t

2 |x| + tϕ. Then when t = 0,
2

u0 = 2 |x| is the solution to the above Dirichlet problem at t = 0. To apply

1 2

the continuity method, we consider solution u = v + ϕt so that v ∈ C 3+α (Ω)

with v = 0 on ϕΩ. Note that the uniqueness of k-admissible solutions follows
from the comparison principle.

3.4 Remarks

(i) In the proof of Theorem 3.1, the assumption f ≥ f0 was used only once
in (3.16). Therefore this assumption can be relaxed to f ≥ 0 for the zero
boundary value problem. By approximation and Theorems 3.1 and 3.4, it
follows that there is a k-admissible solution u ∈ C 1,1 (Ω) to the k-Hessian
equation (2.1) which vanishes on ϕΩ, provided Ω is (k − 1)-convex and f 1/k ∈
C 1,1 (Ω), f ≥ 0.
The above results are also true for a general boundary function ϕ ∈
C 3,1 (ϕΩ). Indeed Krylov [K2] established the a priori estimate (3.2), not only
for solutions to the k-Hessian equation, but also for solutions to the Dirich-
let problem (3.17) for general functions ϕ ∈ C 3,1 (Ω), provided f ≥ 0 and
f ∈ C 1,1 (ϕΩ). The main difficulty is again the estimation on the boundary.
For the k-Hessian equation, Krylov’s proof was simplified in [ITW].
We also note that the geometric assumption (2.6) can be replaced by the
existence of a subsolution u to (3.1) with u = ϕ [G].
(ii) The estimate (3.2) also extends to the Hessian quotient equation [T]

Sk [u]
Sk,l [u] = = f, (3.21)
Sl [u]

where 0 ≤ l < k ≤ n and we define Sl [u] = 1 when l = 0.

(iii) For the second boundary value problem of the k-Hessian equation, and
some other boundary value problems, we refer the reader to [J,S,U3]
(iv) Much more can be said about the regularity of the Monge-Ampère equa-
tion. The interior regularity was established by Calabi and Pogorelov [GT, P].
The global regularity for the Dirichlet problem was obtained independently
by Caffarelli, Nirenberg and Spruck [CNS1], and by Krylov [K1], assuming all
data are smooth enough. Caffarelli [Ca] established the interior C 2,α and W 2,p
estimates for strictly convex solutions, assuming that f ∈ C α and f ∈ C 0 ,
respectively. The continuity of f is also necessary for the W 2,p estimate [W3].
The boundary C 2,α estimate for the Dirichlet problem was established in
[TW6], assuming that f > 0, ∈ C α (Ω), the boundary ϕΩ is uniformly convex
The k-Hessian Equation 191

and C 3 smooth, and boundary function ϕ ∈ C 3 . If either ϕΩ or ϕ is only

C 2,1 , the solution may not belong to W 2,p (Ω) for large p, even f is a positive
constant.

4 Interior a Priori Estimates

In this section we establish interior gradient and second derivative estimates

for the k-Hessian equation

Sk [u] = f (x, u). (4.1)

These estimates were previously proved in [CW1]. From the interior gradient
estimate, we also deduce a Harnack inequality. Estimates in this section will
be repeatedly used in subsequent sections.

4.1 Interior Gradient Estimate

Theorem 4.1 Let u ∈ C 3 (Br (0)) be a k-admissible solution of (4.1). Sup-

pose that f ≥ 0 and f is Lipschitz continuous. Then
M
|Du(0)| ≤ C1 + C2 , (4.2)
r
where M = 4 sup |u|, C2 is a constant depending only on n, k; C1 depends on
n, k, M, r and f C 0,1 . Moreover, if f is a constant, then C1 = 0.

Proof. Introduce an auxiliary function

G(x, ξ) = uξ (x)ϕ(u)ρ(x),
2
where ρ(x) = (1 − |x|
r 2 ) , ϕ(u) = 1/(M − u)
+ 1/2
, and M = 4 sup |u|. Suppose
G attains its maximum at x = x0 and ξ = e1 , the unit vector in the x1 axis.
Then at x0 , Gi = 0 and {Gij } ≤ 0. That is
u1
u1i = − (ui ϕ ρ + ϕρi ), (4.3)
ϕρ
(4.4)
 
0≥ Skij Gij = ϕρ∂1 f + ku1 f ϕ ρ + u1 ϕ ρSkij ui uj + u1 ϕSkij ρij
+u1 ϕ Skij (ui ρj + uj ρi ) + 2Skij u1i (uj ϕ ρ + ϕρj )
2ϕ
2
= ϕρ∂1 f + ku1 f ϕ ρ + u1 ρ(ϕ − )Skij ui uj + u1 ϕSkij ρij
ϕ
2u ϕ
−u1 ϕ Skij (ui ρj + uj ρi ) −
1
Skij ρi ρj ,
ρ
192 X.-J. Wang

where we used the relations Skij uij = kf and Skij uij1 = ∂1 f , which follows by
differentiating equation (4.1).
2
By our choice of ϕ, ϕ − 2ϕϕ ≥ 16 1
M −5/2 . Denote S = Σi Skii . Note that
the term ku1 f ϕ ρ is nonnegative. From (4.4) we obtain

M2 M 2
0 ≥ −16M 5/2 ϕρ|∂1 f | + ρSk11 u31 − CS( u1 + u ), (4.5)
ρr2 r 1

where C is independent of r, M . To prove (4.2), we assume that |Du(0)| >

CM/r, otherwise we are through. Then by G(x0 ) ≥ G(0), we have u1 ρ(x0 ) >
CM/r. Hence by (4.3) we have

ϕ 2
u11 ≤ − u at x0 . (4.6)
2ϕ 1

Hence by (ix) above, Sk11 ≥ CS.

To control ∂1 f by S, by a rotation of the coordinates, we assume that D2 u
is diagonal in the new coordinates y, and uy1 y1 ≥ · · · ≥ uyn yn . Then at the
point x0 where G reaches its maximum,

ϕ 2 1 2
u yn yn ≤ u x1 x1 ≤ − u x1 ≤ − u
2ϕ 4M x1

by (4.6). From equation (4.1),

f = uyn yn σk−1;n (λ) + σk;n (λ), λ = λ(D2 u).

By §2 (vi), we obtain

0 ≤ uyn yn σk−1;n (λ) + C[σk−1;n (λ)]k/(k−1) .

Hence
σk−1;n (λ) ≥ C|uyn yn |k−1 ≥ Cu2k−2
x1 .
We obtain
u2k−2
x1
S ≥ Cu2k−2
1 ≥C at x0 .
M k−1
Recall that in (4.6), we assumed that ux1 ≥ CM/r. Hence S ≥ CM k−1 /r2k−2
and S −1 |∂1 f | is bounded. Multiplying (4.5) by ρ2 /S, we obtain (4.2).

4.2 Harnack Inequality

From the interior gradient estimate, we obtain a Harnack inequality for the
k-Hessian equation. First we prove a lemma, which also follows from the
interpolation inequality (2.12) in [TW2].
The k-Hessian Equation 193

Lemma 4.1 Suppose u ∈ C 1 (BR (0)) is a function which satisfies for any
Br (x) ⊂ BR (0),
C1
|Du(x)| ≤ sup |u|. (4.7)
r Br (x)
Then 
C2
|u(0)| ≤ |u|, (4.8)
|BR | BR

where C2 depends only on C1 and n.

Proof. There is no loss of generality in assuming that R = 1, B1 |u| = 1, and
u is a C 1 function defined in B1+ε (0) for some small ε > 0. Let K be the
largest constant such that |u(x)| ≥ K(1 − |x|)−n for some x ∈ B1 (0), namely
K = sup(1 − |x|)n |u(x)|. Choose y ∈ B1 (0) such that |u(y)| = K(1 − |y|)−n
and |y| = sup{|x| ∈ B1 (0) | |u(x)| = K(1 − |x|)−n }. Then we have |u| ≤
2n |u(y)| = Kr−n in Br (y), where r = 12 (1 − |y|). Therefore by applying
the interior gradient estimate to u in Br (y), we get |Du(x)| ≤ CKr−n−1 .
Hence |u(x)| > 12 Kr−n whenever |x − y| ≤ r/2C. It follows that Br (y) |u| ≥

K/C. But by assumption, BR |u| ≤ 1, we obtain an upper bound for K and
Lemma 4.1 follows.

Theorem 4.2 Let u be a non-positive, k-admissible solution to

Sk [u] = c in BR (0), (4.9)

where c ≥ 0 is a constant. Then we have

sup (−u) ≤ C inf (−u), (4.10)

BR/2 (0) BR/2 (0)

where C depends only on n, k.

Proof. By Lemma 4.1,


sup (−u) ≤ C (−u).
BR/2 3R/4

Since u is subharmonic, we have [GT]


(−u) ≤ C inf (−u).
3R/4 BR/2

From the above two inequalities we obtain (4.10).

The interior gradient estimate also implies the following Liouville
Theorem.
194 X.-J. Wang

Corollary 4.1 Let u ∈ C 3 (Rn ) be an entire solution to Sk [u] = 0. If u(x) =

o(|x|) for large x, then u ≡constant.

4.3 Interior Second Derivative Estimate

Theorem 4.3 Let u ∈ C 4 (Ω) be a k-admissible solution of (4.1). Suppose

f ∈ C 1,1 (Ω × R) and f ≥ f0 > 0. Suppose there is a k-admissible function w
such that
w > u in Ω, and w = u on ϕΩ. (4.11)
Then
(w − u)4 (x)|D2 u(x)| ≤ C, (4.12)
where C depends only on n, k, f0 , supΩ (|Dw| + |Du|), and f C 1,1 (Ω) .
Proof. Writing equation (4.1) in the form

F [u] = fˆ,

where fˆ = f 1/k (x, u), and differentiating twice, we get

Fii uiiγγ + (Fij )rs uijγ ursγ = fˆγγ .

Suppose (D2 u) is diagonal. Then

⎧ 
⎨ μ σk−2;ir (λ) + μ σk−1;i σk−1;r if i = j, r = s,
(Fij )rs = −μ σk−2;ij (λ) if i = j, r = j, and s = i,
⎩
0 otherwise,

where μ(t) = t1/k . Hence


n 
n
Fii uiiγγ = fˆγγ + μ σk−2;ij u2ijγ − [μ σk−1;i σk−1;j + μ σk−2;ij ]uiiγ ujjγ .
i,j=1 i,j=1
(4.13)
Let
1
G(x) = ρβ (x)ϕ( |Du|2 )uξξ ,
2
where ρ = w − u, β = 4, ϕ(t) = (1 − Mt )−1/8 , and M = 2 supx∈Ω |Du|2 . Sup-
pose G attains its maximum at x0 and in the direction ξ = (1, 0, · · · , 0). By a
rotation of axes we assume that D2 u is diagonal at x0 with u11 ≥ · · · ≥ unn .
Then at x0 ,
ρi ϕi u11i
0 = (log G)i = β + + , (4.14)
ρ ϕ u11
The k-Hessian Equation 195

ρii ρ2i ϕii ϕ2i u11ii u211i

0 ≥ Fii (log G)ii = βFii [ − 2 ] + Fii [ − 2 ] + Fii [ − 2 ]. (4.15)
ρ ρ ϕ ϕ u11 u11

Case 1: ukk ≥ εu11 for some ε > 0.

By (4.14) we have
u11i ϕi ρi
= −( + β ). (4.16)
u11 ϕ ρ
Hence by (4.15),

ρii ρ2 ϕii ϕ2 u11ii

0 ≥ βFii [ − (1 + 2β) i2 ] + Fii [ − 3 2i ] + Fii . (4.17)
ρ ρ ϕ ϕ u11
By the concavity of F ,

Fii u11ii ≥ fˆ11 ≥ −C(1 + u11 ).

We have

ϕ ϕ ϕ ϕ
2
ϕii ϕ2
Fii [ − 3 2i ] = ( − 3 2 )Fii u2i u2ii + uγ Fii uiiγ + Fii u2ii
ϕ ϕ ϕ ϕ ϕ ϕ
 
ϕ ϕ
≥ Fii u2ii + uγ fˆγ ,
ϕ ϕ

where by §2 (ix) 
Fii u2ii > Fjj u2jj ≥ θF u211 ,
n
F= i=1 Fii , θ = θ(n, k, ε). Hence

ϕii ϕ2
Fii [ − 3 i2 ] ≥ θF u211 − C.
ϕ ϕ
Since ρ = w − u and w is k-admissible, we have

Fii ρii ≥ −Fii uii = −μ Skii uii = −kμ f.

Inserting the above estimates to (4.17) we obtain

 ρ2i kβμ f
0≥ Fii (log G)ii ≥ θF u211 − CF − − C. (4.18)
ρ2 ρ
Note that ukk ≥ εu11 , we have

F ≥ Fnn ≥ θμ u11 · · · uk−1,k−1 ≥ θ1 uk−1

11 .

Multiplying (4.18) by ρ2β ϕ2 , we obtain G(x0 ) ≤ C.

196 X.-J. Wang

Case 2: ukk ≤ εu11 .

Since (u11 , · · · , unn ) ∈ Γ k , we have ukk ≥ 0 and so |ukk | ≤ εu11 . By the
arrangement u11 ≥ · · · ≥ unn , we have ujj ≤ εu11 for all j = k, · · · , n. Noting
that
∂ k−1 
n
σk [λ] = λj ≥ 0
∂λ1 · · · ∂λk−1
j=k

we obtain
|ujj | ≤ Cεu11 for j = k, · · · , n.
By (4.14),
ρi 1 ϕi u11i
=− ( + ) i = 2, · · · , n. (4.19)
ρ β ϕ u11
Applying (4.16) for i = 1 and (4.19) for i = 2, · · · , n to (4.15), we obtain
 
n
 ρii ϕii ϕ2  ρ2
0≥ βFii + Fii ( − 3 2i ) − β(1 + 2β)F11 12
i=1
ρ ϕ ϕ ρ
 
2 
n n
u11ii u2
+ Fii − (1 + ) Fii 11i =: I1 + I2 (4.20)
i=1
u11 β i=2 u211

As in case I we have
C kβμ f
I1 ≥ θFii u2ii − F11
2
− −C
ρ ρ
1 kβμ f
≥ θF11 u211 − −C
2 ρ

provided ρ2 u211 is suitably large. By (vii) in §2 we obtain

kβμ f
I1 ≥ θ1 μ f u11 − − C.
ρ
We claim
I2 ≥ fˆ11 /u11 . (4.21)
If (4.21) is true then (4.20) reduces to

fˆ11 kβμ f
0 ≥ θ1 μ f u11 + − − C. (4.22)
u11 ρ

Multiplying the above inequality by ρβ ϕ we obtain G(x0 ) ≤ C.

To verify (4.21) we first note that by the concavity of F ,


n  ∂2
− [μ σk−1;i σk−1;j+μ σk−2;ij ]uii1 ujj1 = − μ(Sk (λ))uii1 ujj1 ≥ 0.
i,j=1
∂λi ∂λj
The k-Hessian Equation 197

Hence by (4.13),


n
2 
n
u2
u11 I2 ≥ fˆ11 + μ σk−2;ij u2ij1 − (1 +
) Fii 11i
i,j=1
β i=2 u11
 n  
2 σk−1;i 2
≥ fˆ11 + μ 2σk−2;1i − (1 + ) u11i .
i=2
β u11

Since β = 4, we need only

3 σk−1;i
σk−2;1i − ≥ 0. (4.23)
4 u11
But (4.23) follows from the following lemma.
Lemma 4.2 Suppose λ ∈ Γk and λ1 ≥ · · · ≥ λn . Then for any δ ∈ (0, 1),
there exists ε > 0 such that if

Sk (λ) ≤ ελk1 or |λi | ≤ ελ1 for i = k + 1, · · · , n

we have
λ1 Sk−1;1 ≥ (1 − δ)Sk . (4.24)
Proof. To prove (4.24) we first consider the case Sk (λ) ≤ ελk1 . We may suppose
Sk (λ) = 1. If (4.24) is not true,

Sk−1;1 < (1 − δ)λ−1

1 ≤ε
1/k
.

Hence
k/(k−1)
Sk;1 ≤ CSk−1;1 ≤ Cε1/(k−1) .
Noting that
Sk = Sk−1;1 λ1 + Sk;1 ,
we obtain (4.24).
Next we consider the case |λi | ≤ ελ1 for i = k + 1, · · · , n. Observing
that if λk << λ1 , we have Sk (λ) << λk1 and so (4.24) holds. Hence we may
suppose |λi | << λk for i = k + 1, · · · , n. In this case both sides in (4.24)
= λ1 · · · λk (1 + o(1)) with o(1) → 0 as ε → 0. Hence (4.24) holds.
In Section 6 we will investigate the existence of nonzero solutions to
equation (4.1) with zero Dirichlet boundary condition. Assume that f ∈
C 1,1 (Ω × R), f (x, u) > 0 when u < 0. Then by choosing w = −δ for small
constant δ in (4.12), we obtain a local second derivative estimate. Therefore
by the regularity theory for fully nonlinear, uniformly elliptic equations, one
also obtains local C 3,α estimate for the solution u. That is
Theorem 4.4 Let u ∈ C 4 (Ω) ∩ C 0 (Ω) be a k-admissible solution of (4.1).
Suppose u = 0 on ϕΩ, f ∈ C 1,1 and f > 0 when u < 0. Then u satisfies
3
a priori estimates in Cloc (Ω) ∩ C 0,1 (Ω), namely for any Ω ⊂⊂ Ω,
198 X.-J. Wang

uC 3(Ω ) + uC 0,1 (Ω) ≤ C, (4.25)

where C depends only on n, k, f , sup |u|, and dist(Ω , ϕΩ). If f 1/k ∈ C 1,1 (Ω),
ϕΩ ∈ C 3,1 and Ω is uniformly (k − 1)-convex, then

uC 1,1 (Ω) ≤ C.

Remark. Theorem 4.3 was established in [P] for the Monge-Ampère equation,
and in [CW1] for the k-Hessian equations. The condition (4.11) in Theorem
4.3 is necessary when k ≥ 3 [P, U1], but may be superfluous when k = 2
[WY]. Instead of (4.11), Urbas [U2] established the interior second derivative
estimate under the assumption D2 u ∈ Lp (Ω), p > 12 k(n − 1).

5 Sobolev Type Inequalities

The k-Hessian operator can also be written in the form

Sk [u] = [D2 u]k , (5.1)

see (2.5). Hence by direct computation, one has [R]


∂i Skij [u] = 0 ∀ j. (5.2)
i

It follows that the k-Hessian operator is of divergence form

1
Sk [u] = uij Skij [u]
k
1
= ∂xi (uxj Skij [u]), (5.3)
k
Denote by Φk (Ω) the set of all k-admissible functions in Ω, and by Φk0 (Ω) the
set of all k-admissible functions vanishing on ϕΩ. Let

Ik (u) = (−u)Sk [u]dx (5.4)
Ω

1
= ui uj Skij [D2 u].
k Ω

By (5.2), we can compute the first variation of Ik ,


δIk (u), h = (k + 1) (−h)Sk [u] (5.5)
Ω
The k-Hessian Equation 199

for any smooth h with compact support. Hence the Hessian equation (3.1) is
variational, namely it is the Euler equation of the functional
 
1
J(u) = ui uj Skij [u] + f u. (5.6)
k(k + 1) Ω Ω

The second variation is also easy to compute. Indeed by (5.2) we have


d2
Ik (u + tϕ) = (k + 1) ϕi ϕj Skij [u] (5.7)
dt2 Ω

for any u ∈ C 2 (Ω), ϕ ∈ C0∞ (Ω), or any u, ϕ ∈ C 2 (Ω), both vanishing on ϕΩ.
d2
In particular if u is k-admissible, then dt 2 Ik (u + tϕ) ≥ 0.

Denote
uΦk0 = [Ik (u)]1/(k+1) , u ∈ Φk0 . (5.8)

One can easily verify that  · Φk0 is a norm in Φk0 [W2]. In this section we
prove Sobolev type inequalities for the functional Ik .

5.1 Sobolev Type Inequalities

The following Theorem 5.1 was proved in [W2]. The proof below is also from
there. For convex functions, the theorem was first established in [Ch2].

Theorem 5.1 Let u ∈ Φk0 (Ω).

(i) If 1 ≤ k < n2 , we have

uLp+1(Ω) ≤ CuΦk0 ∀ p + 1 ∈ [1, k ∗ ], (5.9)

where C depends only on n, k, p, and |Ω|,

n(k + 1)
k∗ = .
n − 2k
When p+1 = k ∗ , the best constant C is attained when Ω = Rn by the function

u(x) = [1 + |x|2 ](2k−n)/2k . (5.10)

n
(ii) If k = 2,
uLp(Ω) ≤ CuΦk0 (5.11)
for any p < ∞, where C depends only on n, p, and diam(Ω).
(iii) If n2 < k ≤ n,
uL∞ (Ω) ≤ CuΦk0 , (5.12)
where C depends on n, k, and diam(Ω).
200 X.-J. Wang

Remark. Our proof of Theorem 5.1 reduces the above inequalities to rotation-
ally symmetric functions. When k = n2 , we have accordingly the embedding
(n+2)/n
of Φk0 (Ω) in the Orlicz space associated with the function e|t| .

Proof. Step 1. When u is radial and Ω = B1 (0),

 1 1/(k+1)
uΦk0 (Ω) = C rn−k |u |k+1 . (5.13)
0

One can verify Theorem 5.1 for k-admissible, radial functions vanishing on
∂B1 (0). For details see [W2].
Step 2. We prove that Theorem 5.1 holds for general k-admissible functions
when Ω = B1 (0). Indeed, let

Tp = inf{uk+1 Lp+1(Ω) | u ∈ Φ0 (Ω)},

/uk+1 k
Φk 0

Tp,r = inf{uk+1 Lp+1(Ω) | u ∈ Φ0 (Ω) is radial}.

/uk+1 k
Φk 0

Suppose to the contrary that Tp < Tp,r . Choose a constant λ ∈ (Tp , Tp,r ) and
consider the functional
    k+1
−u λ p+1
J(u) = J(u, Ω) = Sk [u] − (p + 1) F (u) , (5.14)
Ω k+1 k+1 Ω

where  |u|
F (u) = f (t)dt,
0
and f is a smooth, positive function such that
⎧ p
⎨δ |t| < 12 δ
f (t) = |t| p
δ < |t| < M,
⎩ −2
εt |t| > M + ε,

where M > 0 is any fixed constant, δ, ε > 0 are small constants. We also
assume that f is monotone increasing when 12 δ < |t| < δ, and εM −2 ≤ f (t) ≤
|t|p when M < |t| < M + ε. The introduction of ε, δ is such that f is positive
and uniformly bounded, so the global a priori estimates for parabolic Hessian
equations (Theorem 10.1) applies. Obviously F is also uniformly bounded and
J is bounded from below. The Euler equation of the functional is

Sk [u] = λβ(u)f (u), (5.15)

where
   k−p
p+1
β(u) = (p + 1) F (u) .
Ω
The k-Hessian Equation 201

Note that for a given u, β(u) is a constant. By our choice of the constant λ,
we have

inf{J(u) | u ∈ Φk0 (Ω)} < −1 (if M >> 1), (5.16)

inf{J(u) | u ∈ Φk0 (Ω), u is radial} → 0 as δ → 0.

Consider the parabolic Hessian equation

log Sk [u] − ut = log{λβ(u)f (u)} (x, t) ∈ Ω × [0, ∞), (5.17)

subject to the boundary condition

u(·, t) = 0 on ϕΩ ∀ t ≥ 0.

We say a function u(x, t) is k-admissible with respect to the parabolic equa-

tion (5.17) if for any t ∈ [0, ∞), u(·, t) is k-admissible. Equation (5.17) is a
descent gradient flow of the functional J. Indeed, let u(x, t) be a k-admissible
solution. We have

d Sk [u]
J(u(·, t)) = − (Sk [u] − ψ) log ≤ 0,
dt Ω ψ

and equality holds if and only if u is a solution to the elliptic equation (5.15),
where ψ(u) = λβ(u)f (u).
Let u0 ∈ Φk0 (Ω) be such that

J(u0 ) ≤ inf Φk0 (Ω) J(u) + ε < −1.

By a slight modification (see Remark 5.1 below), we may assume that the
compatibility condition Sk [u0 ] = λβ(u0 )f (u0 ) holds on ϕΩ × {t = 0}. In the
parabolic equation (5.17), β(u) is a function of t. By (5.16) and since F (u)
is uniformly bounded, we have

C1 ≤ β(u) ≤ C2 ,

for some positive constants C1 , C2 independent of time t. Note that C1 , C2

may depend on M but are independent of the small constants ε and δ. There-
fore by Theorem 10.1, there is a global smooth k-admissible solution u to the
parabolic Hessian equations (5.17).
By the global a priori estimates and since (5.17) is a descent gradient
flow, u(·, t) sub-converges to a solution u1 of the elliptic equation (5.15). By
the Aleksandrov’s moving plane method, see also [D] (p.327) for the Monge-
Ampère equation, we infer that u1 is a radial function. Therefore we have

inf{J(u) | u ∈ Φk0 (Ω), u is radial} ≤ −1.

We reach a contradiction when ε, δ are small.

202 X.-J. Wang

Step 3. Denote

Tp (Ω) = inf{uk+1 Lp+1(Ω) | u ∈ Φ0 (Ω)}.

/uk+1 k
Φk 0

We claim that for any (k − 1)-convex domains Ω1 , Ω2 with Ω1 ⊂ Ω2 ,

Tp (Ω1 ) ≥ Tp (Ω2 ). (5.18)

Suppose to the contrary that Tp (Ω1 ) < Tp (Ω2 ). Let λ ∈ (Tp (Ω1 ), Tp (Ω2 )) be
a constant. Let J(u, Ω) be the functional given in (5.14). Then we have

inf{J(u, Ω1 ) | u ∈ Φk0 (Ω1 )} < −1 (when M >> 1), (5.19)

inf{J(u, Ω2 ) | u ∈ Φk0 (Ω2 )} → 0 as δ → 0.

Let u1 ∈ Φk0 (Ω1 ) be the solution to (5.15) obtained in Step 2 which satisfies

J(u1 , Ω1 ) ≤ −1.

Let
1
w(x) = −M − ε − ε1/2k (R2 − |x|2 ),
2
where R is chosen large such that Ω1 ⊂ BR (0). Recall that f (t) = εt−2 when
|t| > M + ε, and C1 ≤ β(u1 ) ≤ C2 , where C1 , C2 are independent of ε.
By equation (5.15) we have Sk [u1 ] ≤ Cε. Hence Sk [w] ≥ Cε1/2 ≥ Sk [u1 ]
when u1 < −M − ε. Applying the comparison principle to u1 and w in
{u1 < −M − ε}, we obtain a lower bound for u1 ,

u1 ≥ −M − R2 ε1/2k . (5.20)

Hence when ε is sufficiently small, F (u1 ) = |u1 |p+1 + o(1) if ε, δ is small,

though f (u1 ) may violate strongly. In particular we have
  k−p
p+1
β(u1 ) = (1 + o(1)) |u1 | p+1
(5.21)
Ω1

with o(1) → 0 as ε, δ → 0.
Extending u1 to Ω2 such that u1 = 0 in Ω2 − Ω1 (so u1 is not k-admissible
in Ω2 ). Let ψ(x) = Sk [u1 ] in Ω1 and ψ(x) = 0 in Ω2 − Ω1 . Denote
   k+1
p+1
E(ϕ) = (−ϕ)ψ − λ |ϕ|p+1
.
Ω2 Ω2

Then, since u1 = 0 outside Ω1 ,

The k-Hessian Equation 203

   k+1
p+1
E(u1 ) = (−u1 )Sk [u1 ] − λ |u1 | p+1
Ω1 Ω1
    k+1
p+1
≤ (−u1 )Sk [u1 ] − λ (p + 1) F (u1 )
Ω1 Ω1
= J(u1 , Ω1 ) ≤ −1,

where we have used, by the construction of f , the fact that F (u) ≤ p+1 |u|
1 p+1
.
Let u2 = u2,m ∈ Φk0 (Ω2 ) be the solution of

Sk [u] = fm in Ω2 ,

where fm be a sequence of smooth, positive functions which converges mono-

tone decreasingly to ψ. By the maximum principle we have u2 L∞ (Ω2 ) ≤ C
for some C > 0 independent of m. By the comparison principle we have
u2 < u1 ≤ 0 in Ω1 .
By our choice of λ and by approximation and uniform boundedness of u2 ,
we have
   k+1
p+1
E(u2 ) = (−u2 )ψ − λ |u2 | p+1
Ω2 Ω2
   k+1
p+1
1
≥ (−u2 )Sk [u2 ] − λ |u2 | p+1
−
Ω2 Ω2 8
1
≥−
8
provided m is sufficiently large.
Denote ρ(t) = E[u1 + t(u2 − u1 )]. Then ρ(0) = E(u1 ) ≤ −1 and ρ(1) =
E(u2 ) ≥ − 81 . We compute
   k−p
p+1


ρ (0) = (u1 − u2 )Sk [u1 ] − (k + 1)λ |ϕ| p+1
|u1 |p (u1 − u2 ).
Ω2 Ω2 Ω2

Since u1 is a solution of (5.15), by (5.21) we have

 
(u1 − u2 )Sk [u1 ] = λβ(u1 ) |u1 |p (u1 − u2 )
Ω2 Ω1
  k−p
p+1

= λ(1 + o(1)) |u1 | p+1
|u1 |p (u1 − u2 )
Ω1 Ω2
  k−p
p+1

< (k + 1)λ |u1 | p+1
|u1 |p (u1 − u2 )
Ω1 Ω2
204 X.-J. Wang

We obtain ρ (0) < 0. Note that the functional E is linear in the first integral
and convex in the second integral, we have ρ (t) ≤ 0 for all t ∈ (0, 1).
Therefore we must have ρ(1) < ρ(0). We reach a contradiction. Hence (5.18)
holds.
Finally we remark that when k < n2 and p + 1 = k ∗ , the best constant in
(5.9) is achieved by the function in (5.10). This assertion follows from Step 2
by solving an ode. By the Hölder inequality, one also sees that when k < n2
and p + 1 < k ∗ , the constant in (5.9) depends on the volume |Ω| but not
the diameter of Ω. When k > n2 , The above proof implies the embedding
Φk0 (Ω) → L∞ (Ω). Indeed, in Step 2 we have shown that the best constant Tp
is achieved by radial functions, and so the assertion follows from Step 1.

Remark 5.1. For any initial function u0 ∈ Φk0 (Ω) satisfying J(u0 ) < 0, we can
modify u0 slightly near ϕΩ such that it satisfies the compatibility condition
Sk [u0 ] = λβ(u0 )f (u0 ) on ϕΩ × {t = 0}. Indeed, it suffices to replace u0
by the solution û0 ∈ Φk0 (Ω) of Sk [u] = g, where g(x) = (1 + a)Sk [u0 ] when
dist(x, ϕΩ) > δ1 and g(x) = λβ(u0 )f (u0 ) when dist(x, ϕΩ) < 12 δ1 . We choose
δ1 a sufficiently small constant and a also small such that β(û0 ) = β(u).

5.2 Compactness

In this section we prove the embedding Φk0 (Ω) → Lp (Ω) is compact when
k < n2 and p < k ∗ . First we quote a theorem from [TW4]

Theorem 5.2 Suppose Ω is (k − 1)-convex. Then

uΦl0(Ω) ≤ CuΦk0 (Ω) (5.22)

for any 1 ≤ l < k ≤ n, and any u ∈ Φk0 (Ω). The best constant C is achieved
by the solution u ∈ Φk0 (Ω) to the Hessian quotient equation

Sk [u]
=1 in Ω. (5.23)
Sl [u]

Proof. The proof is based on the global existence of smooth solutions to

initial boundary problem of the parabolic equation [TW4]

Sk [u]
ut − log = 0 in Ω × [0, ∞), (5.24)
Sl [u]

subject to the boundary condition u = 0 on ϕΩ× (0, ∞). As above, a solution

to (5.24) is k-admissible if for any t, u(·, t) ∈ Φk (Ω). The a priori estima-
tion for the parabolic equation (5.24) is very similar to that for the elliptic
equation (3.1).
The k-Hessian Equation 205

By constructing appropriate super- and sub-barriers, we also infer that for

any initial function u(·, 0) satisfying the compatibility condition on Sk [u] =
Sl [u] on ϕΩ, the solution u(·, t) converges to the solution u∗ of (5.23). Note
that u(·, t) ∈ Φk0 (Ω) implies the boundary condition u = 0 on ϕΩ × [0, ∞).
With the above results for the parabolic equation (5.24), Theorem 5.2
follows immediately. Indeed, let
 
1 1
J(u) = (−u)Sk [u] − (−u)Sl [u].
k+1 Ω l+1 Ω

For any u0 ∈ Φk0 (Ω), modify u slightly near ϕΩ such that Sk [u0 ] = Sl [u0 ]
on ϕΩ. Let u(·, t) ∈ Φk0 (Ω) be the solution to the parabolic equation (5.24).
Then 
d Sk [u]
J(u(·, t)) = − {Sk [u] − Sl [u]} log ≤ 0.
dt Ω Sl [u]
It follows that J(u∗ ) ≤ J(u0 ) for any u0 ∈ Φk0 (Ω). Replacing u0 by
u0 u∗ Φk0 (Ω) /u0 Φk0 (Ω) , we obtain Theorem 5.2.
n
Theorem 5.3 The embedding Φk0 (Ω) → Lp (Ω) is compact when k < 2 and
p < k∗ .
By the Hölder inequality, Theorem 5.3 follows from Theorem 5.2 and the
2n
compactness of the embedding W 1,2 (Ω) → Lp (Ω) for p < n−2 .
n
Next we show that when k > 2 , a k-admissible function is Hölder
continuous.
Theorem 5.4 Suppose u ∈ Φk (Ω) ∩ L∞ (Ω) and k > n
2. Then u ∈ Cloc
α
(Ω)
with α = 2 − nk , and for any x, y ∈ Ω ⊂⊂ Ω,

|u(x) − u(y)| ≤ C|x − y|α , (5.25)

where C depends only on n, k, Ω, dist(Ω , ϕΩ), and uL∞(Ω) .

Proof. Let
w(x) = |x|2−n/k .
By direct computation, w is k-admissible and Sk [w] = 0 when x = 0. For any
interior point x0 ∈ Ω, Applying the comparison principle to u and û(x) =
u(x0 ) + Cw(x − x0 ), where C is chosen large such that û > u on ϕΩ, we
obtain Theorem 5.4.

5.3 An L∞ Estimate

As an application of Theorem 5.1, we prove an L∞ estimate for solutions

to the k-Hessian equation. See Theorem 2.1 in [CW1]. The proof below is
essentially the same as that in [CW1].
206 X.-J. Wang

Theorem 5.5 Let u ∈ C 2 (Ω) ∩ C 0 (Ω) be a k-admissible solution of


Sk (D2 u) = f (x) in Ω,
(5.26)
u=ϕ on ϕΩ.

Suppose f ≥ 0, f ∈ Lp (Ω), where p > n/2k if k ≤ n2 , or p = 1 if k > n

2.
Then there exists C > 0 depending only on n, k, p, Ω such that
1/k
| inf u| ≤ | inf ϕ| + Cf Lp(Ω) . (5.27)
Ω Ω

Proof. By replacing the boundary function ϕ by inf ϕ and by the comparison

principle, we need only to prove (5.27) for ϕ ≡ 0. Since the k-Hessian equation
is homogeneous, we may assume that f Lp(Ω) = 1.
First we prove (5.27) for k = n2 . Multiplying (5.26) by −u and taking
integration, we obtain,

uk+1
Φk (Ω)
= | f (x)u(x)dx| ≤ f Lp uLq
0
Ω
1 1 1 1
≤ |Ω| q (1− β ) uqβ ≤ C|Ω| q (1− β ) uΦk0 (Ω) ,

1 1
where p + q = 1 and β > 1 will be chosen large. Hence
1 1
uΦk0 ≤ C|Ω| qk (1− β ) .

By the Sobolev type inequality (5.11),

1 1
uL1 (Ω) ≤ |Ω|1− β uLβ (Ω) ≤ C|Ω|1− β uΦk0 ≤ C|Ω|1+δ , (5.28)

where δ = 1
qk − β1 (1 + 1
qk ) > 0 provided β is sufficiently large. Hence

C
|{u(x) < −K}| ≤ |Ω|1+δ . (5.29)
K
From Sard’s theorem, the level set {u(x) < t} has smooth boundary for
almost all t. Therefore we may assume all the level sets involved in the proof
below have smooth boundary.
Denote u1 = u + K, Ω1 = {u1 < 0}. When K is large enough, we have
|Ω1 | ≤ 12 |Ω|. For j > 1 we define inductively uj and Ωj by uj = uj−1 + 2−δj
and Ωj = {uj < 0}. Then similarly to (5.28) we have

uj L1 (Ωj ) ≤ C|Ωj |1+δ

for some C independent of j. Therefore

|Ωj+1 | ≤ C2δ(j+1) |Ωj |1+δ ,

where Ωj+1 = {uj (x) < −2−δ(j+1) }.

The k-Hessian Equation 207

Assume by induction that |Ωi | ≤ 2 |Ωi−1 |

1
for all i = 1, 2, · · · , j, then by
(5.29),
2−δ(j−1) C 1+δ
|Ωj |δ ≤ 2−δ(j−1) |Ω1 | ≤ |Ω| .
K
When K
∞is large, we obtain |Ωj+1 | ≤ 12 |Ωj |. Therefore the set {x ∈ Ω | u(x) <
−K − j=1 2−δj } has measure zero. In other words, we have
∞
| inf u| ≤ K + 2−δj .
j=1

Therefore (5.27) is established for k = n2 .

n/2
When k < n2 , let w ∈ Φ0 be the solution to

Sn/2 [w] = f n/2k in Ω.

2k/n
By inequality (vi) in §2, Sk [w] ≥ Cn,k Sn/2 [w] ≥ Cn,k f . Hence by the
comparison principle we also obtain (5.27).
When k > n2 , multiplying (5.26) by −u and taking integration, we have

uk+1
Φk
= | f (x)u(x)dx| ≤ f L1 uΦk0
0
Ω

and (5.27) follows from (5.12). This completes the proof.

We will prove in Section 9 that the solution in Theorem 5.5 is Hölder

continuous.
Theorem 5.5 was extended by Kuo and Trudinger to more general elliptic
equations [KT]. In their paper [KT], Kuo and Trudinger considered the linear
elliptic inequality

L[u] = aij (x)uxi xj ≤ f in Ω. (5.30)
u ≤ 0 on ϕΩ

Assume that the eigenvalues λ(A) ∈ Γk∗ , where A = −{aij (x)}, Γk∗ is the
dual cone of Γk , given by

Γk∗ = {λ ∈ Rn | λ · μ ≥ 0 ∀ μ ∈ Γk }

Denote
ρ∗k (A) = inf{λ · μ | μ ∈ Γk , σk (μ) ≥ 1}.
They proved the following maximum principle
208 X.-J. Wang

Theorem 5.6 Let u ∈ C 2 (Ω) ∩ C 0 (Ω) be a solution of (5.30). Assume that

λ(A) ∈ Γk∗ and ρ∗k (A) > 0. Then we have the estimate

f
sup u ≤ C Lq (Ω) ,
Ω ρ∗k (A)

where q = k if k > n
2, and q > n
2 if k ≤ n
2, where C depends only on n, k, q,
and Ω.
Theorem 5.6 extended the well-known Aleksandrov maximum principle.

6 Variational Problems

Consider the Dirichlet problem


Sk (D2 u) = f (x, u) in Ω,
(6.1)
u=0 on ϕΩ,

where f (x, u) ∈ C 1,1 (Ω × R) is a nonnegative function in Ω × R. There

has been a huge amount of works on the existence of positive solutions to
semilinear elliptic equations, namely equation (6.1) with k = 1. In this section
we show that there are similar existence results for the k-Hessian equation.
Materials in this section are taken from in [CW1, CW2], except the eigenvalue
problem in §6.2, which was previously treated in [W2]. We note that the
published paper [CW1] is a part of the preprint [CW2]. The preprint [CW2]
also contains the existence of solutions in the critical growth case, presented
in §6.4 below.
As shown in §5, a solution of (6.1) is a critical point of the Hessian
functional  
−1
J(u) = uSk [u] − F (x, u), (6.2)
k+1 Ω Ω
0
where F (x, u) = u f (x, t)dt. The functional J is defined on the convex cone
Φk0 (Ω). We don’t know the behavior of the functional near the boundary of
Φk0 (Ω), and so we cannot use the variational theory directly. To find a critical
point of J, we employ a descent gradient flow of the functional, which was
previously used by Chou [Ch1] for the Monge-Ampère equation. That is a
parabolic Hessian equation of the form

μ(Sk [u]) − ut = μ(f (u)). (6.3)

We assume that μ is a smooth function defined on (0, ∞), satisfying μ (t) > 0,
μ (t) < 0 for all t > 0,
The k-Hessian Equation 209

μ(t) → −∞ as t → 0,
μ(t) → +∞ as t → +∞,

and such that μ(Sk [u]) is concave in D2 u. As we consider solutions in Φk0 , the
boundary condition for (6.3) is

u = 0 on ϕΩ × [0, ∞).

Let u ∈ Φk0 (Ω × R+ ) be a k-admissible solution to (6.3). Then


d
J(u(·, t)) = − (Sk (λ) − f )ut (6.4)
dt
Ω
= − (Sk (λ) − f )(μ(Sk (λ)) − μ(f )) ≤ 0.
Ω

As before, we say a solution u is k-admissible with respect to (6.3) if for

any t ≥ 0, u(·, t) is k-admissible. To simplify the notation, we will denote
u ∈ Φk0 (Ω × R+ ) if u(·, t) ∈ Φk0 (Ω) for all t ∈ R+ = [0, ∞).
A typical example of μ is μ(t) = log t [Ch1]. But for the k-Hessian equation
we have to choose a different μ in our treatment below. For the a priori
estimates for the parabolic equation (6.3), we always need to assume that f
is strictly positive. But in studying the variational problem (6.1), typically
f vanishes when u = 0. To avoid such situation, we add a small positive
constant to f , or modify f slightly near u = 0.
To study the variational problem associated with the k-Hessian equation,
similar to the Laplace equation, we divide the problem into three cases,
namely the sublinear case, the eigenvalue problem, and the superlinear case.

6.1 The Sublinear Growth Case

We say f (x, u) is sublinear with respect to the k-Hessian operator if

lim |u|−k f (x, u) → 0 (6.5)

u→−∞

uniformly for x ∈ Ω. Note that the power k is due to that the k-Hessian
operator is homogeneous of degree k.

Theorem 6.1 Let Ω be (k−1)-convex with C 3,1 -boundary. Suppose f (x, u) ∈

C 1,1 (Ω × R), f (x, u) > 0 when u < 0, f satisfies (6.5), and inf Φk0 (Ω) J(u) < 0.
Then there is a nonzero solution u ∈ C 0,1 (Ω) ∩ C 3,α (Ω) to (6.1), which is the
minimizer of the functional J.

Proof. We sketch the proof, as it was essentially included in the proof of

Theorem 5.1.
210 X.-J. Wang

Replace f by f + δ for some small δ > 0, so that f is strictly positive.

Observe that in the sublinear growth case, by the Sobolev type inequality
(Theorem 5.1), we have J(tu) → +∞ as t → ∞ for any u ∈ Φk0 (Ω), u = 0,
and J is bounded from below. As the infimum of J is negative, one can
choose an initial function u0 ∈ Φk0 (Ω) such that J(u0 ) < inf u∈Φk0 (Ω) J(u) + δ.
By Remark 5.1, we may assume the compatibility condition Sk [u0 ] = f on
ϕΩ at t = 0 is satisfied. In the sublinear growth case, one can construct
a sub-barrier u to the parabolic equation (6.3) such that u ≤ u0 . By the
comparison principle one gets a global uniform estimate, and also derivative
estimates up to the third order, for solutions to (6.3). Therefore there is a
global smooth solution to (6.3). By (6.4), the solution sub-converges to a
nonpositive solution u = uδ of (6.1).
We claim that all the solutions uδ are uniformly bounded for δ > 0 small.
Indeed, if mδ = − inf uδ → ∞, the function vδ = uδ /mδ satisfies the equation

Sk [v] = m−k
δ [f (x, mδ vδ ) + δ].

By (6.5), the right hand side converges to zero uniformly. Hence by the com-
parison principle, one infers that inf vδ → 0, which contradicts with the fact
that inf vδ = −1. Next by the assumption that inf J < 0, it is easily seen that
uδ does not converge to zero.
Sending δ → 0, by the interior a priori estimates in §4, we conclude that
uδ converges to a solution u ∈ C 0,1 (Ω)∩C 3,α (Ω) of (6.1) which is a minimizer
of J.

A particular case in Theorem 6.1 is when f ∈ C 1,1 (Ω) is a function of

x, independent of u [B]. Then for any initial u0 satisfying the compatibility
condition, the solution u ∈ Φk0 (Ω × R+ ) of (6.3) is uniformly bounded and
converges to a solution u∗ of (6.1). By the convexity of the functional J
(see (5.7)) and the uniqueness of solutions to the Dirichlet problem, u∗ is a
minimizer of the functional J in Φk0 (Ω).

6.2 The Eigenvalue Problem

Similar to the Laplace operator, the k-Hessian operator admits a positive

eigenvalue λ1 such that

Sk [u] = λ|u|k in Ω, (6.6)

u = 0 on ϕΩ,

has a nonpositive k-admissible solution when λ = λ1 . The following theorem

was proved in [W2] for k < n and in [Lp] for k = n. Here we provide a proof
which uses Theorem 6.1.
The k-Hessian Equation 211

Theorem 6.2 Let Ω be (k − 1)-convex with C 3,1 -boundary. Then there exists
λ1 > 0 depending only on n, k, Ω, such that

(i) (6.6) has a nonzero k-admissible solution ϕ1 ∈ C 1,1 (Ω) ∩ C 3,α (Ω) when
λ = λ1 .
(ii) If (λ∗ , ϕ∗ ) ∈ [0, ∞) × (C 1,1 (Ω) ∩ C 3,α (Ω)) is another solution to (6.6),
then λ∗ = λ1 and ϕ∗ = cϕ1 for some positive constant c.
(iii) If Ω1 ⊂ Ω2 , then λ1 (Ω1 ) ≥ λ1 (Ω2 ).

Proof. First consider part (i). Let p ∈ (k − 12 , k) and let c0 > 0 be a large
constant. By Theorem 6.1, there is an admissible solution up ∈ Φk0 (Ω) to the
problem
Sk [u] = c0 |u|p ,
which is a minimizer of the associated functional. Namely J(up ) = inf J(u),
where  
1 c0
J(u) = (−u)Sk [u] − |u|p+1 .
k+1 Ω p+1 Ω
Let vp = up /mp , where mp = sup |up |. Then vp satisfies

Sk [v] = c0 mp−k
p |v|p .

If mp−k
p → 0 as p → k, the right hand side converges to zero uniformly, which
contradicts with the fact that inf vp = −1. If mp−k p → ∞, then mp → 0
uniformly, which implies J(up ) = inf J(u) → 0. But if we choose c0 > 0
large, J(up ) = inf J(u) → −∞ as p → k. The contradiction implies that
mp is uniformly bounded. Hence by the a priori estimates in §4, we see that
(c0 mp−k
p , vp ) sub-converges to (λ1 , ϕ1 ), and (λ1 , ϕ1 ) is a solution of (6.6). By
Theorem 4.4, ϕ1 ∈ C 1,1 (Ω) ∩ C ∞ (Ω).
Next we consider (ii). If (λ∗ , ϕ∗ ) is also a solution of (6.6), we may assume
that λ∗ > λ1 and ϕ∗ < ϕ1 by multiplying a constant to ϕ∗ . Denote aij =
ϕ
 1/k
ϕuij Sk [u] at u = ϕ1 . Then λ1 and ϕ1 are respectively the eigenvalue
 2
and eigenfunction of the elliptic operator L = aij ∂ij . By the concavity of
Sk [u], and noting that ψ ∗ and ψ1 are negative in Ω, we deduce that
1/k

L(ϕ∗ − ϕ1 ) ≥ Sk [ϕ∗ ] − Sk [ϕ1 ] = −(λ∗ )1/k ϕ∗ + λ1 ϕ1 > λ1 (ϕ∗ − ϕ1 )

1/k 1/k 1/k 1/k

1/k
in Ω, which contradicts the fact that λ1 is the first eigenvalue of L. Hence
we have λ∗ = λ1 and ϕ1 = ϕ∗ .
Part (iii) was proved in Step 3 of the proof of Theorem 5.1.
212 X.-J. Wang

6.3 Superlinear Growth Case

We say f is superlinear with respect to the k-Hessian operator if

lim |u|−k f (x, u) → ∞ (6.7)

u→−∞

uniformly for x ∈ Ω.

Theorem 6.3 Suppose that f (x, z) > 0 for z < 0,

lim f (x, z)/|z|k < λ1 , (6.8)

z→0−
lim f (x, z)/|z|k > λ1 , (6.9)
z→−∞

where λ1 is the eigenvalue of the k-Hessian operator. Suppose there exist

constants θ > 0 and M large such that
 0
1−θ
f (x, s) ds ≤ |z|f (x, z) ∀ z < −M. (6.10)
z k +1

When k ≤ n
2, we also assume that there exists p ∈ (1, k ∗ − 1) such that

lim f (x, z)/|z|p = 0. (6.11)

z→−∞

Then (6.1) has a non-zero k-admissible solution in C 3,α (Ω) ∩ C 0,1 (Ω), α ∈
(0, 1).

When k = 1, Theorem 6.3 is a typical result in semilinear elliptic equation.

The solution in Theorem 6.3 is a min-max critical point of the functional J.
As indicated before, we cannot use the variational theory directly, but by
studying a descent gradient flow, we can use the underlying idea in the Moun-
tain Pass Lemma. The main difficulty is to prevent blowup of solutions near
the boundary for both the elliptic equation (6.1) and the parabolic equation
(6.3), in the case 2 ≤ k < n2 . It requires some new techniques.

Proof. For clarity we divide the proof into four steps.

Step 1. Let fδ,K be a smooth, positive function given by

fδ,K (x, u) = δ + ηδ1 fˆδ,K (x, u),

where ηδ1 ∈ C0∞ (Ω) is a nonnegative function satisfying ηδ1 (x) = 1 when
dist(x, ϕΩ) > 2δ1 and ηδ1 (x) = 0 when dist(x, ϕΩ) < δ1 , and
⎧
⎨δ if |u| < 12 δ,
ˆ
fδ,K (x, u) = f (x, u) if δ < |u| < K,
⎩ p
|u| if |u| > 2K.

We will choose the constants δ, δ1 > 0 small and K > 1 large.

The k-Hessian Equation 213

Remark. Before continuing, let us explain why we make these modifications

when k ≥ 2, which are not needed when considering semilinear elliptic equa-
tions (the case k = 1). The introduction of δ is such that f is positive, so
that we can apply the C 3 a priori estimates for the parabolic Hessian equa-
tion (6.3). We modify f for large |u| (namely f = |u|p when |u| > 2K) is to
use the gradient estimate for the parabolic Hessian equation. The purpose of
introducing ηδ1 is to prevent the solution to the parabolic Hessian equation
blow-up near the boundary. In the following we choose δ1 = δ.
Consider the functional
 
1
Jδ,K (x, u) = (−u)Sk [u] − Fδ,K (x, u), (6.12)
k+1 Ω Ω
0
where Fδ,K (x, u) = u fδ,K (x, t). When δ > 0 is sufficiently small, by assump-
tion (6.8), there exists a smooth, k-admissible function u0 ∈ Φk0 (Ω) with small
L∞ -norm, such that
Sk [u0 ] > fδ,K (x, u0 ). (6.13)
Consider the parabolic Hessian equation (6.3) with initial condition u(·, 0) =
su0 , where s > 0 is a parameter. We choose the function μ in (6.3) such that

μ(t) = log t if t < 1/8,

μ(t) = t1/p if t > 8, (6.14)

and
(t − s)(μ(t) − μ(s)) ≥ C(t − s)(t1/p − s1/p ) (6.15)
for all t, s > 0, where C is an absolute constant, independent of s, t. Then
equation (6.3) has a unique smooth solution us . By (6.11) and (6.14), μ(f (u))
is of linear growth in u. Hence the solution exists for all time t.
Since Sk [u0 ] > 0, u0 is a sub-barrier for the solution us for small s > 0.
That is when s > 0 is small, one has 0 > us (·, t) > u0 for all t. Hence
Jδ,K (x, us ) is uniformly bounded,

1
Jδ,K (us (·, t)) > − Fδ,K (us (x, t))dx ≥ −
Ω 2

for all t, provided δ > 0 is small.

On the other hand, when s > 1 is large, we have Jδ,K (su0 ) < −1. Hence
Jδ,K (us (·, t)) < −1 for all t > 0 as (6.3) is a descent gradient flow. Let

s∗ = inf{s | limt→∞ Jδ,K (us (·, t)) < −1 ∀ s > s}. (6.16)

Then s∗ is positive. By the continuous dependence of the solution us in s,

and the monotonicity (6.4), we see that Jδ,K (us∗ (·, t)) ≥ −1 for all time t.
We also have
214 X.-J. Wang

sup |us∗ (·, t)| ≥ C > 0 (6.17)

for some C > 0 independent of t. Indeed, if sup |us∗ (·, t)| is small at some
time t, then sup |us (·, t)| is also small at t for s > s∗ , close to s∗ . Hence by
(6.13), u0 is a sub-barrier, and so by the comparison principle, sup |us (·, t)|
is small for all t > t, which contradicts with the definition of (6.16).
Suppose for a moment that

|us∗ (·, t)| ≤ C (6.18)

uniformly for t ∈ (0, ∞). The constant C is allowed to depend on δ and K.

Then by the global regularity of the parabolic Hessian equation, we conclude
that us∗ (·, t) sub-converges to a solution u∗δ,K to the equation

Sk [u] = fδ,K (x, u). (6.19)

In Step 4 we show that u∗δ,K is uniformly bounded in δ and K, and so it

sub-converges as δ → 0, K → ∞ to a solution u ∈ C 0,1 (Ω) ∩ C 3 (Ω) in Φk0 (Ω)
of (6.1). From (6.17), u = 0.
Step 2. In the following we prove (6.18). For brevity we will write us∗ as
u, dropping the subscript s∗ . Recall that Jδ,K (u(·, t)) ≥ −1 for all time t.
Hence the set
d
K 0 = {t ∈ (0, ∞) | Jδ,K (u(·, t)) < −σ} (6.20)
dt
has finite measure, where σ > 0 is a small constant. For any t ∈ K 0 , first we
show that

(−u(·, t)Sk [u(·, t)] ≤ C, (6.21)
Ω
Fδ,K (x, u(·, t)) ≤ C, (6.22)
Ω

where C is a constant independent of t, δ and K. Indeed, if t ∈ K 0 , we have


  
Sk [u] − fδ,K (x, u) μ(Sk [u]) − μ(fδ,K (x, u))
Ω
  d
= ∂t u Sk [u] − fδ,K (x, u) = − Jδ,K (u(·, t)) ≤ σ.
Ω dt

Hence by (6.15),

 
1/p
1/p 
Sk [u] − fδ,K (x, u) Sk [u] − fδ,K (x, u) ≤ Cσ.
Ω
The k-Hessian Equation 215

1/p
1/p
Denote α = Sk [u] , β = fδ,K (x, u) . We obtain
 
|α − β|p+1 ≤ C (αp − β p )(α − β) ≤ Cσ.
Ω Ω

We have
 
 
 p 
u(α − β ) ≤ C
p
|u| |α − β| (αp−1 + β p−1 )dx
Ω Ω
  p+1
1   p(p+1)
1   p−1
p
≤C |α − β| p+1
|u| p+1
|u||α + β |
p p
Ω Ω Ω
  p−1
p
1/p
≤ Cσ 1/(p+1) uLp+1 |u||α + β |
p p
.
Ω

On the other hand,

Jδ,K (s∗ u0 ) = Jδ,K (u(·, 0)) ≥ Jδ,K (u(·, t)) (6.23)

 
1
= (−u)Sk [u] − Fδ,K (x, u) ≥ −1.
k+1 Ω Ω

By (6.10),
1−θ
Fδ,K (x, u) ≤ δ|u| + |u|fδ,K (x, u) + C.
k+1
Hence
 
1
Jδ,K (u(·, t)) = (−u)Sk [u] − Fδ,K (x, u)
k+1 Ω
 Ω
1 1−θ
≥ (−u)Sk [u] − [δ|u| + |u|fδ,K (x, u) + C]
k+1 Ω k +1
 Ω
 
1 θ
= (−u)(αp −β p )+ |u|fδ,K (x, u)− (C +δ|u|).
k+1 Ω k+1 Ω Ω

It follows that, by the Sobolev inequality (Theorem 5.1),

  
∗
|u|fδ,K (x, u) ≤ C |u| |α − β | + Jδ,K (s u0 ) + (C + δ|u|)
p p
Ω Ω Ω
   p−1
p

1/p
≤ Cσ 1/(p+1)
uLp+1 |u||α| + p
|u|β |p
+ (C +δ|u|)
Ω Ω Ω
   p−1
p

1/p
≤ Cσ 1/(p+1) uΦk (Ω) uk+1
Φk (Ω)
+ |u|β |
p
+ (C +δ|u|).
0 0
Ω Ω
216 X.-J. Wang

By the Sobolev inequality again, uL1 ≤ CuΦk0 . We obtain

 
|u|fδ,K (x, u) ≤ C1 εσ,δ |u||α|p + C2
Ω Ω

with εσ,δ → 0 as σ, δ → 0. Inserting the estimate into (6.23) we obtain (6.21)

and (6.22).
Step 3. Now we use (6.21) and (6.22) to establish (6.18). If k > n2 , in
view of (5.12), (6.18) follows readily from (6.21). We need only to consider
=t = supΩ |u(·, t)|. If Mt is not
the cases k ≤ n2 . Let Mt = supΩδ |u(·, t)|, M
uniformly bounded, there exists a sequence tj → ∞ such that Mtj → ∞ and

Mt ≤ Mtj for all t < tj . (6.24)

Let
−dx + Kd2x
w(x) = Mtj ,
−δ + Kδ 2
where dx = dist(x, ϕΩ). We choose δ small and K ∈ (1, δ −1/2 ) large such
that Sk [w] > δ in Ω − Ωδ , where Ωδ = {x ∈ Ω | dx > δ}. Then w = Mtj on
ϕΩδ . Recall that fδ,K = δ in Ω − Ωδ . By the comparison principle we have
u(x, t) ≥ w(x) for any x ∈ Ω − Ωδ , t ∈ (0, tj ). It follows that M=t ≤ Mtj for
t ∈ (0, tj ). By (6.11) and (6.14), the right hand side of the parabolic equation
(6.3) is of linear growth. Hence we have

Mt ≥ Mtj eC(tj −t) ∀ t < tj .

Hence Mt ≥ CMtj for t ∈ (tj − 2, tj ). Since the set K 0 has finite measure,
we may assume that tj ∈ K 0 for all j and Mt ≤ CMtj for all t < tj .
Suppose the maximum Mtj of |u(·, tj )| is attained at the point yj . By the
interior gradient estimate (10.10) below, we have
1
u(x, tj ) ≤ − Mtj ∀ x ∈ Br (yj ),
2

where r = c1 Mtβj and c1 > 0 is independent of j, and

p+k k−p
β =1− = .
2k 2k
By (6.21) (where the constant C is independent of δ, K) and the Sobolev
inequality (5.9) and (5.11), we have

u(·, tj )Lq (Br (yj )) ≤ u(·, tj )Lq (Ω) ≤ CuΦk0 (Ω) ≤ C,

where q = k ∗ if k < n2 and q > p + 1 is any sufficiently large constant if

k = n2 . On the other hand, we have

u(·, tj )Lq (Br (yj ) ≥ Crn Mtqj ≥ CMtq+bβ

j
.
The k-Hessian Equation 217

Since p < k ∗ − 1, we have q + bβ > 0. Hence when Mtj is sufficiently large,

we reach a contradiction. Therefore (6.18) is proved.
Step 4. We have therefore obtained a solution uδ,M to (6.19) which satisfies
(6.21) and (6.22), with the constant C in (6.21) and (6.22) independent of
δ, K. If k > n2 , by (5.12) we obtain

sup |uδ,K | ≤ C (6.25)

Ω

for a different C independent of δ, K. Sending δ → 0 and K → ∞, we obtain

a solution u ∈ C 0,1 (Ω) ∩ C 3 (Ω) in Φk0 (Ω) of (6.1).
If k < n2 , denote ψ = fδ,K (c, uδ,K ). By (6.21) and (6.11), and the Sobolev
inequality, we have ψ ∈ Lβ (Ω) for some β > 2k n
. Hence applying Theorem
5.5 to equation Sk [u] = ψ in Ω we obtain again (6.25). In all the cases, we
obtain a solution to (6.1).

Remarks
(i) In Step 4 above, if k = n2 , by the Sobolev embedding (5.11), the right
hand side of (6.19) belongs to Lβ (Ω) for any β > 1. Write equation (6.19) in
the form  1/k
aij uij = [Sk [u]]1/k = fδ,K . (6.26)
i,j

where aij = k1 [Sk [u]]1/k−1 Skij [u]. By inequality (x) in Section 2, the determi-
nant |aij | ≥ Cn,k > 0. Hence by Aleksandrov’s maximum principle,

1 n/k
sup |uδ,K | ≤ C fδ,K dx ≤ C
Ω |a ij |

for some C independent of δ, K. We also obtain (6.25).

(ii) If f is independent of x and the domain Ω is convex, by the method of
moving planes, the maximum point of uδ,K will stay away from the boundary.
Hence we can obtain (6.25) by a usual blow-up argument. We don’t need to
use Theorem 5.5.
(iii) Let u1 be a k-admissible function with small L∞ norm, u2 be a k-
admissible function such that J(u2 ) < −1, where J is the functional in (6.2).
Let Γ denote the set of paths in Φk0 (Ω) connecting u1 to u2 . Let

c0 = inf sup J(γ(s)). (6.27)

γ∈Γ s∈(0,1)

Then the assumptions (6.8)–(6.11) and the Sobolev inequality (Theorem 5.1)
implies that c0 > 0. The above proof implies that there is a solution u ∈ Φk0 (Ω)
to (6.1) such that J(u) = c0 .
218 X.-J. Wang

6.4 The Critical Growth Case

In this section we extend Theorem 6.3 to the critical growth case. Consider
the problem  ∗
Sk (D2 u) = |u|k −1 + f (x, u) in Ω,
(6.28)
u=0 on ϕΩ,
∗
−1
where 1 < k < n/2 and f is a lower order term of |u|k . For simplicity we
will consider the case
f (x, u) = λ|u|q , (6.29)
where q ∈ (k, k ∗ − 1), λ > 0. Denote
  
1 1 k∗ λ
J(u) = (−u)Sk [u] − ∗ |u| − |u|q+1 dx.
k+1 Ω k Ω q+1 Ω

c0 = inf sup J(su). (6.30)

u∈Φk
0 (Ω) s>0

By the Sobolev inequality (5.9), we have c0 > 0.

We also denote
 
1 1 ∗
J ∗ (u) = (−u)Sk [u] − ∗ |u|k .
k+1 Ω k Ω

c∗ = inf sup J ∗ (tu). (6.31)

Φk
0 (Ω) t>0

The following theorem extends the existence of positive solutions to semi-

linear elliptic equations in [BN] to the k-Hessian equation. Our proof is
completely different, due to the lack of a gradient estimate near the boundary
for equation (6.40).
Theorem 6.4 Suppose
c 0 < c∗ . (6.32)
Then (6.28) has a non-zero k-admissible solution.
Proof. For any p ∈ (q, k ∗ − 1), by Theorem 6.3, there exists a solution up ∈
C 3 (Ω) ∩ C 0,1 (Ω) of

Sk (D2 u) = ψp (x, u) =: |u|p + λ|u|q in Ω
(6.33)
u=0 on ϕΩ

with
Jp (up ) = cp ,
where  
−1
Jp (u) = uSk (D2 u)dx − Ψp (x, u)dx,
k+1 Ω Ω
0
Ψp (x, u) = u ψp (x, t)dt, and
The k-Hessian Equation 219

cp = inf sup Jp (su) > 0.

u∈Φk
0 (Ω) s>0

From equation (6.33) we have

 
2
up Sk (D up )dx = up ψp (x, up )dx,
Ω Ω

which, together with Jp (up ) = cp , implies that

up Φk0 ≤ C and up Lp+1 (Ω) ≤ C. (6.34)

We want to prove that

Mp = sup |up (x)|

x∈Ω

is uniformly bounded for p < k ∗ − 1 and close to k ∗ − 1. If this is true then

by the regularity in §4, there exists a subsequence of up (x) which converges
to a solution u0 of (6.28). Moreover, one can prove

J(u0 ) = lim Jp (up ) = lim cp > 0.

p→k∗ −1 p→k∗ −1

Hence u0 < 0 in Ω.
Suppose to the contrary that there is a subsequence pj so that Mj =:
Mpj → ∞ as pj → k ∗ − 1. Suppose the supremum Mj is attained at xj . Let

vj (y) = Mj−1 u(Rj−1 y + xj ), y ∈ Ωj ,

(p −k)/2k
where Rj = Mj j , Ωj = {y | Rj−1 y + xj ∈ Ω}. Then vj (0) = −1,
−1 ≤ vj ≤ 0 for y ∈ Ωj , and vj satisfies

Sk (Dy2 v) = ψj (y), (6.35)

where
ψj (y) = |vj |pj + λMj j |vj |q .
q−p

Moreover,  
δ
|uj |pj +1 dx = Mj j |vj |pj +1 dy, (6.36)
Ω Ωj
 
δ
|uj |Sk (D2 uj )dx = Mj j |vj |Sk (D2 vj )dy,
Ω Ωj

where δj = pj + 1 − 2k n
(pj − k) ≥ 0. Hence vj Lpj +1 and vj Φk0 (Ωj ) are
uniformly bounded.
By passing to a subsequence we assume that xj → x∞ ∈ Ω. Denote
dj = dist(xj , ϕΩ). If
dj Rj → ∞, (6.37)
220 X.-J. Wang

then for any R > 0, BR (0) ⊂ Ωj provided j is large enough. By the interior
gradient estimate (Theorem 4.1) and the interior second derivative estimate
(Theorem 4.2), we may suppose, by passing to a subsequence if necessary,

vj (y) → v∞ in Cloc
2
(Rn ),

and v∞ satisfies the equation

∗
−1
Sk (D2 v) = |v|k in Rn .

Note that to apply Theorem 4.2, we may choose the function w in (4.11) as
w = Rε2 (|x|2 − R12 ) for large R1 . Hence
1

 
∗
− v∞ Sk (D2 v∞ )dx = |v∞ |k dx,
Rn Rn

and so
J ∗ (v∞ ) = sup J ∗ (tv∞ ) ≥ c∗ . (6.38)
t>0
0
On the other hand, let Ψj (y, u) = u ψj (y, t)dt. We have

1
Ψj (y, u) → ∗ |v∞ (y)|k
∗
in L∞ n
loc (R )
k
as j → ∞. Note that by equation (6.35),

−1
vj Sk (D2 vj ) − Ψj (y, vj ) ≥ 0.
k+1
By Fatou’s lemma we obtain

 −1 
J ∗ (v∞ ) ≤ limj→∞ vj Sk (D2 vj ) − Ψj (y, vj ) dx
Ωj k + 1

−δ  −1 
= limj→∞ Mj j upj Sk2 (D2 upj ) − Ψpj (x, upj ) dx
Ω k+1
≤ limj→∞ cpj ≤ c0 < c∗ ,

which contradicts with (6.38). Hence Mp is uniformly bounded.

Next we consider the case that dj Rj is uniformly bounded. We may
suppose
dj Rj → α ≥ 0. (6.39)
By the interior gradient estimate (Theorem 4.1), vj converges locally uni-
formly to a function v∞ , and v∞ satisfies
∗
−1
Sk (D2 v∞ ) = |v∞ |k in Ω∞ (6.40)
The k-Hessian Equation 221

in a weak or the viscosity sense (see §9 for definition of the weak solution),
where by a rotation of axes, Ω∞ = {yn > −α}. Moreover we have −1 ≤
v∞ ≤ 0 in Ω∞ .
We note that the argument in Case 1 doesn’t work at the current situation,
as we don’t have uniform gradient estimate near the boundary ϕΩ∞ , we don’t
know whether v∞ = 0 on ϕΩ∞ .
Let Dj = {y ∈ Rn | vj (y) ≤ − 21 }. By (6.34) and (6.36) we have vj Lpj ≤ C
and so mes(Dj ) ≤ C. Applying Theorem 5.5 to the equation (6.35) and
noticing that inf vj = −1, we have

mes(Dj ) ≥ C1 (6.41)

for some C1 > 0 independent of j. Let vj∗ be the (usual) rearrangement of vj .

Namely vj∗ is a radially symmetric, monotone increasing function, satisfying
|{vj∗ < a}| = |{vj < a}| for any a ∈ R, where | · | denotes the Lebesgue
measure in Rn . Let v ∗ = limj→∞ vj∗ . Then v ∗ ≡ 0 because of (6.41). By
Fatou’s lemma,
 
∗
|v ∗ |k dy ≤ limj→∞ |vj∗ |pj +1 dy
Rn Rn

= limj→∞ |vj |pj +1 dy ≤ C.
Rn

Therefore v ∗ (r) = o( 1r ) as r → ∞. It follows that for any given ε > 0, there

exists δ = δε > 0, with δ → 0 at ε → 0, such that
∗
δ · mes{y ∈ Rn | |v ∗ (y)|k > δ} < ε.

Hence
∗ ∗
δ · mes{y | |vj (y)|k > δ} = δ · mes{y | |vj∗ (y)|k > δ} < ε (6.42)

for sufficiently large j.

Denote
∗
vj (y) = vj (y) + δ 1/k ,
Ωj,δ = {y ∈ Ωj |  vj (y) < 0}.

We have
  ∗ 
−tk+1 tk ∗
sup J ∗ (tvj | Ωj,δ ) = sup vj Sk (D2 vj ) − ∗ |vj |k dy
t>0 t>0 Ωj,δ k+1 k
  k+1 ∗ 
−t tk k∗
≥ sup vj Sk (D vj ) − ∗ |vj | dy.
2
t>0 Ωj,δ k+1 k
222 X.-J. Wang

By (6.42),
 
∗ ∗ ∗
|vj |k = |
vj + δ 1/k |k
Ωj,δ
Ωj,δ
∗ ∗ ∗
≤ vj |k + Cδ 1/k |
[| vj |k −1 + Cδ]
Ωj,δ
   k∗k−1
∗
∗ ∗ ∗
≤ vj |k + (δ|Ωj,δ |)1/k
[| |
vj |k + Cδ|Ωj,δ |
Ωj,δ Ωj,δ

1/k∗ ∗
≤ (1 + Cε ) |
vj |k ,
Ωj,δ

where we have used the fact that


vj Lk∗ ≥ C > 0,

which follows from (6.41). We obtain

  ∗ 
−tk+1 tk ∗ ∗
sup J ∗ (tvj | Ωj,δ ) ≥ sup vj Sk (D2 vj ) − ∗ (1 + Cε1/k )|
vj |k dy.
t>0 t>0 Ωj,δ k+1 k

Hence we obtain
∗
sup J ∗ (tvj | Ωj,δ ) ≥ (1 − Cε1/k )c∗ .
t>0

∗
On the other hand, since f (x, u) = λ|u|q is a lower order term of |u|k −1
and mes(Ωj,δ ) are uniformly bounded for fixed δ, we see that when Mj is
large enough,
∗
sup Jj (tvj , Ωj,δ ) ≥ sup J ∗ (tvj , Ωj,δ ) − εj ≥ (1 − Cε1/k )c∗ − εj
t>0 t>0

with εj → 0 as Mj → ∞, where
  
−1 
Jj (u, Ωj,δ ) = Jpj (u, Ωj,δ ) = uSk (D u) − Ψj (x, u) dx.
2
Ωj,δ k+1

For any subdomain D ⊂ Ωj , since

 
∗
vj (|vj |k −1 + λMj j |vj |q+1 )dy,
q−p
vj Sk (vj )dy =
D D

we have
Jj (vj , D) = sup Jj (tvj , D) ≥ 0.
t>0

Hence ∗
Jj (vj , Ωj ) ≥ Jj (vj , Ωj,δ ) ≥ (1 − Cε1/k )c∗ − εj .
The k-Hessian Equation 223

We reach a contradiction with (6.32) when ε and εj are sufficiently small.

This completes the proof.
The technique in the treatment of the case (6.39) is new. Moreover, it also
applies to the case (6.37). By carefully examining the argument, one sees that
the function f in (6.29) can be replaced by a more general f (x, u), provided

limt→0 f (x, t)|t|−k = 0,

∗
limt→0 f (x, t)|t|−k +1 = 0.

In the case the constant c0 in (6.30) should be replaced by

c0 = inf sup J(γ(s)),

γ∈Γ s∈[0,1]

where Γ denotes the set of all paths in Φk0 (Ω) connecting U ≡ 0 to a function
u0 satisfying J(u0 ) < 0.
The verification of (6.32) can be carried out in a similar way as [BN], and
the computation is also similar. Let
 k  1  n−2k
n − 2k ε k+1 2k
wε (x) = Cnk , (6.43)
k ε + |x|2

where Cnk is the binary coefficient. Then wε satisfies the equation

∗
−1
Sk (D2 u) = |u|k in Rn , (6.44)

and the constant c∗ in (9.3) is attained by wε when Ω = Rn .

Lemma 6.1 Suppose there exists a ball Br (x0 ) ⊂ Ω and
 
(k + 1)(nk − n + 2k)
q > max k, −1
k(n − 2k)

so that
f (x, u) ≥ λ|u|q for x ∈ Br (x0 ) (6.45)
for some λ > 0, then c0 < c∗ .
Proof. Let  
∗
B= wεk dx =− wε Sk (D2 wε )dx.
Rn Rn
B is independent of ε. By a translation we may suppose x0 is the origin. Let
ϕ(x) be a radial cut-off function so that ϕ = 1 in Br/2 (0) and ϕ = 0 outside
Br (0). We may choose ϕ so that uε =: ϕwε ∈ Φk0 for ε > 0 small. Direct
computations show that

∗
ukε dx = B + O(ε(n−2k)/2k ),
Ω
224 X.-J. Wang

(−uε )Sk (D2 uε )dx = B + O(ε(n−2k)/2k ),
Ω

If f (x, u) satisfies (6.45), we have

 
|uε |q+1 dx ≥ Cε 2 − 2(k+1) q ,
n n−2k
F (x, uε )dx ≥ C
Ω Ω
0
where F (x, u) = u f (x, t)dt. If q > (k+1)(nk−n+2k)
k(n−2k) −1, we have 2 − 2(k+1) q
n n−2k
<
n−2k ∗
2k . Hence if ε is small enough we have c 0 < c .

We refer the reader to [CGY] for more details on radially symmetric

solutions to the k-Hessian equations in the critical growth case.

7 Hessian Integral Estimates

In this section we establish some local integral estimates for k-admissible

functions [TW2]. The main estimates include (7.2) and (7.5) below. Our
estimates are based on the divergence structure of the k-Hessian operator.
As shown as the beginning of Section 5, the k-Hessian operator can also be
written as

Sk [u] = [D2 u]k (7.1)

1
= uij Skij [u]
k
1
= ∂xi (uxj Skij [u]).
k
As before we denote by Φk (Ω) the set of all smooth k-admissible functions
in Ω, and by Φk0 (Ω) the set of all smooth k-admissible functions vanishing
on ϕΩ.

7.1 A Basic Estimate

Here we establish the following basic estimate.

Theorem 7.1 Let u ∈ Φk (Ω). Suppose u ≤ 0 in Ω. Then for any subdomain

Ω ⊂⊂ Ω,
  k
Sk [u] ≤ C |u| , (7.2)
Ω Ω

where C is a constant depending on n, k, Ω and Ω .

The k-Hessian Equation 225

Proof. It suffices to consider the case Ω = BR (y), Ω = Br (y), for some

y ∈ Rn and r < R. Let η ∈ C ∞ (Ω) satisfy 0 ≤ η ≤ 1, η = 1 in Br (y), and
η = 0 when |x − y| ≥ (R + 2r)/3. Let u ∈ C ∞ (Ω) be the unique k-convex
solution of the Dirichlet problem

u] = ηSk [u] + δ n in Ω,
Sk [
=0
u on ϕΩ.

Then Sk [u + 2δ |x − y|2 ] ≥ Sk [u]. By the comparison principle, we have

2 |x − y| + u ≤ u ≤ 0 in Ω, so that
δ 2

 
|
u| ≤ |u| + Cδ.
Ω Ω

Let ζ ∈ C0∞ (Ω) be a cut-off function. Then, by integration by parts,

  
1 1
ζSk [
u] = ζSkij [ =
u]Dij u Skij [
u u]Dij ζ
k k
Ω

1

≤ max |D2 ζ| | u| S ii [
u]
k suppD2 ζ

n−k+1

= max |D2 ζ| | u| Sk−1 [

u].
k suppD2 ζ

Choose ζ = 1 in B(R+r)/2 (y), ζ = 0 for |x − y| ≥ (5R + r)/6, |D2 ζ| ≤

C(R − r)−2 . By the Harnack inequality (4.10), we have

1
|
u(x)| ≤ |
u| ∀ |x − y| ≥ (R + r).
Ω 2

Hence by sending δ → 0, we obtain

  
Sk [u] ≤ C Sk−1 [u] (−u)
Ω Ω Ω

for some constant C depending on n, k, r, R. By iteration we obtain (7.2).

7.2 Local Integral Gradient Estimates

First we prove
Theorem 7.2 Let u ∈ Φk (Ω), k = 1, · · · , n, satisfy u ≤ 0 in Ω. Then for
any sub-domain Ω ⊂⊂ Ω, we have the estimates
  q+l
|Du|q Sl [u] ≤ C |u| (7.3)
Ω Ω
226 X.-J. Wang

n(k−l)
for all l = 0, · · · , k − 1, 0 ≤ q < n−k , where C is a constant depending on
Ω, Ω , n, k, l and q.
Corollary 7.1 Let u ∈ Φk (Ω). Then ∀ Ω ⊂⊂ Ω,

DuLq (Ω ) ≤ C |u| (7.4)
Ω

for q < nk
n−k , where C depends on n, k, q, Ω, and Ω .
Inequality (7.4) follows from (7.3) by taking l = 0. When k = 1, (7.4) can
be found in [H]. Corollary 7.1 asserts that a k-admissible function is in the
1,q
local Sobolev space Wloc (Ω). When k > n/2, we have q > n, and by the
Sobolev imbedding theorem, u ∈ Cloc α
(Ω) with α ∈ (0, 2 − nk ). But recall that
in Theorem 5.4 we have shown that u ∈ Cloc α
(Ω) with α = 2 − nk .
To prove Theorem 7.2, let us denote, for a real n × n matrix A = [aij ], not
necessarily symmetric,

Sk (A) = [A]k , (7.5)

ϕ
Skij (A) = [A]k .
∂aij

Then for any vector field g = (g1 , · · · , gn ), gi ∈ C 1 (Ω), i = 1, · · · , n, it follows

that

Di Skij (Dg) = 0, j = 1, · · · , n, (7.6)

Skij (Dg)Di gj = kSk (Dg).

Now we introduce a broader class of operators, namely, the p-k-Hessian

operators, given for k = 1, · · · , n, p ≥ 2, u ∈ C 2 (Ω), by
 
Sk,p [u] = D(|Du|p−2 Du) k . (7.7)

where  
Du ⊗ Du
D(|Du|p−2 Du) = |Du|p−2 I + (p − 2) D2 u.
|Du|2
When k = 1, it is the well-known p-Laplacian operator,

S1,p [u] = div (|Du|p−2 Du). (7.8)

One can verify by direct computation that the p-k-Hessian operator is

invariant under rotation of coordinates.
Let us call a function u ∈ C 2 (Ω), p-k-admissible in Ω if Sl,p [u] ≥ 0 for all
l = 1, · · · , k. We then have the following relation between k-admissibility and
p-k-admissibility.
Lemma 7.1 If u is k-admissible, then u is p-l-admissible for l = 1, · · · , k −1
and p − 2 ≤ n(k−l)
l(n−k) .
The k-Hessian Equation 227

Proof. At a point y ∈ Ω, where Du(y) = 0, we fix a coordinate system so

that the x1 axis is directed along the vector Du(y) and the remaining axes
are chosen so that the reduced Hessian [Dij u]i,j=2,··· ,n is diagonal. It follows
then that the p-Hessian is given by
⎧
⎪
⎪ (p − 1)Di1 u if j = 1, i ≥ 1,

⎨
D1j u if i = 1, j > 1,
Di |Du| Dj u = |Du|
p−2 p−2
(7.9)
⎪
⎪ D ii u if j = i > 1,
⎩
0 otherwise.

Hence by calculation, we obtain for l = 1, · · · , k − 1 at the point y, setting

i = Dii u(y), i = 1, · · · , n,
λ


n
1 σl−1;1 (λ)
|Du|l(2−p) Sl,p [u] = (p − 1)λ  + σl;1 (λ)
 − (p − 1) 
σl−2;1i (λ)(D 2
i1 u) .
i=2
(7.10)
From the k-admissibility of u, we have

n
1 σk−1;1 (λ)
Sk [u] = λ  + σk;1 (λ)
 − 
σk−2;1i (λ)(D i1 u) ≥ 0
2

i=2

so that using Newton’s inequality, in the form

σk;1 l(n − k) σl;1

≤ , (7.11)
σk−1;1 k(n − l) σl−1;1
k(n−l)
we have, for p − 1 ≤ l(n−k) , the inequality

1 σk;1 n
|Du|l(2−p)  
Sl,p [u] ≥ λ1 σl−1;1 (λ)+ 
σl−1;1 (λ)− 
σl−2;1i (λ)(D i1 u)
2
p−1 σk−1;1 i=2

σl−1;1  
n n
≥ σk−2;1i (Di1 u)2 − σl−2;1i (Di1 u)2
σk−1;1 i=2 i=2
n  
1 
= σl−1;1 σk−2;1i − σk−1;1 σl−2;1i (Di1 u)2
σk−1;1 i=2
n  
1 
= σl−1;1i σk−2;1i − σk−1;1i σl−2;1i (Di1 u)2
σk−1;1 i=2
≥ 0.

Note that when applying Newton’s inequality to the last inequality, the
coefficient in (σl−1;1i σk−2;1i − σk−1;1i σl−2;1i ) is better than we need.
228 X.-J. Wang

Let λ = (λ1 , · · · , λn ) ∈ Γk . Suppose λ1 ≥ · · · ≥ λn . Let u = 1 2

2 λi xi . By
Lemma 7.1, we have Δp u ≥ 0 for p ≤ 2 + n(k−1) n−k . Hence

 n(k − 1)
λi + λn ≥ 0. (7.12)
i
n−k

Proof of Theorem 7.2. Setting

k(n − l)
p∗ = 1 + , k < n, l < k,
l(n − k)

we obtain from Lemma 7.1 and the formula (7.10), for 2 < p < p∗ and
u ∈ Φk (Ω),

p∗ − p p−2 ∗
|Du|l(2−p) Sl,p [u] =
∗
Sl [u] + ∗ |Du|l(2−p ) Sl,p∗ [u]
p −2 p −2
p∗ − p
≥ ∗ Sl [u],
p −2
n(k−l)
and hence, for q = (p − 2)l < n−k , we have the estimate

p∗ − 2
|Du|q Sl [u] ≤ Sl,p [u]. (7.13)
p∗ − p

Theorem 7.2 will follow by estimation of Sl,p [u] in L1loc (Ω). For any non-
negative cut-off function η ∈ C02 (Ω), we obtain
 
ηSl,p [u] = ηSl (D(|Du|p−2 Du)) (7.14)
Ω Ω

1
= ηS ij Di (|Du|p−2 Dj u)
l Ω l

1
=− |Du|p−2 Slij Di ηDj u.
l Ω

From (7.9), we have

Slij Dj u = |Du|(l−1)(p−2) Slij (D2 u)Dj u

= |Du|(l−1)(p−2) Slij [u]Dj u,

so that, by substituting in (7.14), we obtain

 
1
ηSl,p [u] = − |Du|l(p−2) Slij Di ηDj u
l Ω
Ω

1
≤ |Du|q+1 |Dη|Sl−1 [u],
l Ω
The k-Hessian Equation 229

and hence, replacing η by η l and using (7.13), we obtain

 
|Du|q η l Sl [u] ≤ C max |Dη| |Du|q+1 η l−1 Sl−1 [u].
Ω Ω

Consequently,
 

l
|Du| η Sl [u] ≤ C max |Dη|
q l
|Du|q+l , (7.15)
Ω Ω

so that the estimate (7.3) is reduced to the case l = 0. To handle this case,
we take l = 1 in (7.15) with

n(k − 1)
q = q(1) < .
n−k
If u is k-admissible for k ≥ 2, we have
1

S2 [u] = (Δu)2 − |D2 u|2 ≥ 0

2
and hence
|D2 u| ≤ Δu. (7.16)
Therefore we obtain from (7.15)
 
η|Du| |D u| ≤ C max |Dη|
q 2
|Du|1+q
Ω Ω

so that  

ηD |Du|1+q ≤ C max |Dη| |Du|1+q .

Ω Ω

Thus by an appropriate choice of η, we obtain for any subdomain Ω ⊂⊂ Ω,


−1
|Du| Ln/(n−1) (Ω ) ≤ CdΩ
1+q
|Du|1+q , (7.17)
Ω

where dΩ = dist(Ω , ϕΩ), C is a constant depending on k, q and n. The

estimate (7.3) now follows by interpolation. 
From Theorem 7.2 we may derive corresponding estimates for the k-
admissible function themselves.
Theorem 7.3 Let u be a nonpositive k-admissible function in Ω, k ≤ n/2.
Then for any subdomain Ω ⊂⊂ Ω, we have
  l+q
|u|q Sl [u] ≤ C |u| (7.18)
Ω Ω

n(k−l)
for all l = 0, · · · , k − 1, 0 ≤ q < n−2k , where C is a constant depending on
Ω, Ω , n, k, l and q.
230 X.-J. Wang

Proof. With η ≥ 0, ∈ C01 (Ω), we estimate

  
q 1
η 2 (−u)q Sl [u] = η 2 (−u)q−1 Slij Di uDj u − (−u)q Slij Di uDj η 2
l l
Ω Ω
 Ω
q(n − l + 1)
≤ η (−u) Sl−1 |Du|2
2 q−1
l Ω

2(n − l + 1)
+ η(−u)q Sl−1 |Du||Dη|
l Ω

(q + 1)(n − l + 1)
≤ η 2 (−u)q−1 Sl−1 |Du|2
l
 Ω
n−l+1
+ |Dη|2 (−u)q+1 Sl−1
l Ω

n(k−l+1)
Now, for any p < n−k , we have
  2/p  1−2/p
p(q−1)
η 2 (−u)q−1 Sl−1 |Du|2 ≤ η 2 Sl−1 |Du|p η 2 (−u) p−2 Sl−1
Ω Ω Ω

∗
so that if q < n(k−l) p(q−1)
n−2k , we may choose p so that q = p−2 <
n(k−l+1)
n−2k , and
the estimate (7.18) follows from Theorem 7.2 by induction on l.

8 Hessian Measures

In this section we extend the notion of k-admissible functions to nonsmooth

functions. We assign a measure μk [u] to a k-admissible function u and prove
the weak continuity of μk . The proof of the weak continuity of μk in [TW2]
involves delicate integral estimates and is based on the estimates in §7. Here
we provide a simpler proof, using ideas from [TW1, TW5]. As an application
we prove the existence of a weak solution to the Dirichlet problem of the
k-Hessian equation.

8.1 Non-Smooth k-Admissible Functions

Observe that a C 2 function u is k-admissible if and only if for any matrix

A = {aij } with eigenvalues in the cone

Γk∗ = {λ∗ ∈ Rn | λ∗ · λ ≤ 0 ∀ λ ∈ Γk },

there holds 
2
aij Dij u ≤ 0. (8.1)
The k-Hessian Equation 231

Note that a matrix A with eigenvalues in Γk∗ must be negative definite. From
(8.1) we can extend the notion of k-admissibility to non-smooth functions as
follows.

Definition 8.1 A function u in Ω is k-admissible if

(i) it is upper semi-continuous and the set {u = −∞} has measure zero; and
(ii) for any matrix A = {aij } with eigenvalues in Γk∗ ,

2
uaij Dij ϕ ≤ 0 ∀ ϕ ∈ C0∞ (Ω), ϕ ≥ 0. (8.2)
Ω

∗
Note that
when k = 1, Γk contains only the vector −(1, · · · , 1), and (8.2)
becomes Ω u(−Δϕ) ≤ 0 for any ϕ ≥ 0, ∈ C0∞ (Ω). The above definition
implies that an upper semi-continuous function  u is2k-admissible if it is sub-
harmonic with respect to the operator L = aij Dij for any matrix A with
eigenvalues in Γk∗ .
From (8.2) we see that if u is k-admissible, so is its mollification uε , given by
 
x−y
uε (x) = u(x − εy)ρ(y) dy = ε−n ρ( )u(y), (8.3)
Ω Ω ε

where ρ is a mollifier, namely ρ is a smooth, nonnegative function with sup-

port in the unit ball B1 (0), and B1 (0) ρ = 1. Observe that if u is k-admissible,
it is also subharmonic. Hence its mollification uε converges to u monotone
decreasingly. Therefore by Corollary 7.1, a k-admissible function is locally in
1,q nk
the Sobolev space Wloc (Ω) for any q < n−k .

Lemma 8.1 Let uj be a sequence of k-admissible functions which converges

to u almost everywhere. Suppose u is upper semi-continuous and the set
{u = −∞} has measure zero. Then u is k-admissible and uj converges to
u pointwise.

Proof. The first assertion follows readily from the definition. The second one
is due to that u is upper semi-continuous and uj is subharmonic and so it
satisfies the mean value inequality below.

Recall that a k-admissible function u is subharmonic, it satisfies the mean

value inequality

1
u(y) ≤ u ∀ Br (y) ⊂ Ω. (8.4)
|Br (y)| Br (y)

Therefore if uj and u are k-admissible and {uj } converges to u almost every-

where, uj is locally uniformly bounded in L1 (Ω). Conversely, if a sequence
of k-admissible functions {uj } is uniformly bounded in L1loc (Ω), then by
Corollary 7.1, the set {u = −∞} has measure zero.
We also have the following comparison principle for k-admissible functions.
232 X.-J. Wang

Lemma 8.2 Suppose u and v are k-admissible and v is smooth in Ω. Suppose

Sk [v] = 0 in Ω and for any point y ∈ ϕΩ, limx→y [v(x) − u(x)] ≥ 0. Then
v ≥ u in Ω.

Proof. If there is an interior point x0 ∈ Ω such that v(x0 ) < u(x0 ), by adding
a positive constant δ = 12 (u(x0 ) − v(x0 )) to v we may suppose that for any
y ∈ ϕΩ, limx→y [v(x) − u(x)] ≥ δ > 0, so that v > u in a neighborhood
of the boundary ϕΩ. Therefore for ε > 0 small, we have v > uε near ϕΩ,
where uε is the mollification of u. By the comparison principle for smooth
k-admissible functions, we conclude that v ≥ uε , which is in contradiction
with v(x0 ) < u(x0 ) ≤ uε (x0 ).

Lemma 8.3 Suppose u, v are two k-admissible functions.Then w = max(u, v)

is also k-admissible.

Proof. Let uε , vε be the mollification of u and v, respectively. Then it suffices

to show that wε = max(uε , vε ) is k-admissible. For brevity we drop the
subscript ε. Since the function w is semi-convex (i.e., w + C|x|2 is convex
for sufficiently large constant C), w is twice differentiable almost everywhere
and the eigenvalues of D2 w lies in Γk . Let wε be the mollification of w. By
integration by parts in (8.3), we have

D wε (x) ≥
2
D2 w(x − ε y)ρ(y) dy.
Ω

Hence for any matrix A with eigenvalues in Γk∗ ,


2
aij Dij wε (x) ≤ 2
aij Dij w(x − ε y)ρ(y) dy ≤ 0.
Ω

Hence wε , and so also w, is k-admissible.

8.2 Perron Lifting

Let u be a k-admissible function in Ω and ω  Ω be a subdomain of Ω. The

Perron lifting of u in ω, uω , is the upper semicontinuous regularization of the
function û,
uω (x) = lim sup û, (8.5)
t→0 B (x)
t

where

û(x) = sup{v(x) | v is k-admissible in Ω and v ≤ u in Ω − ω}.

Obviously we have uω ≥ û, and û and uω coincide in Ω except possibly

on ∂ω.
The k-Hessian Equation 233

Lemma 8.4 Assume ϕω is C 3,1 smooth. Then uω is a solution of

Sk [w] = 0 in ω, (8.6)
w = u on ϕω,

in the sense that there is a sequence of smooth k-admissible functions wε

which satisfies Sk [wε ] = 0 in ω and wε → u on ω pointwise.
Proof. Let uε be a mollification of u, as given in (8.3). Let uε,j = uε +2−j |x|2 .
Then Sk [uε,j ] ≥ C2−kj . That is, uε,j is a smooth sub-solution to the Dirichlet
problem

Sk [w] = C2−kj in ω,
w = uε,j on ϕω.

Hence from [G], there is a unique global smooth solution wε,j ∈ C 3 (ω), mono-
tone in j. By (3.5) we have supω |Dwε,j | ≤ supϕω |Dwε,j |. On the boundary
ϕω, we have uε,j ≤ wε,j ≤ uε,j , where uε,j is the harmonic extension of uε,j
in ω. Hence

sup |Dwε,j | ≤ sup |Duε,j | ≤ sup |Duε | + C2−j .

ϕω ϕω ϕω

Therefore by passing to a subsequence, wε,j converges as j → ∞ to a solution

wε of (8.6) which satisfies the boundary condition wε = uε on ϕω.
ε = wε in ω and uε = uε in Ω − ω. It is easy to see that uε is the
Let uω ω ω

Perron lifting of uε in ω. The proof of Lemma 8.3 implies that uω ε is k-

admissible. Since uε is monotone decreasing in ε, so is uω
ε . By the comparison
principle (Lemma 8.2), we have uω ε ≥ u in Ω. Hence u0 := limε→0 uε ≥ u .
ω ω ω
ω
On the other hand, let u0 be the upper semicontinuous regularization of
u0 . Then by Lemma 8.1, uω 0 is k-admissible in Ω. Obviously u0 = u in Ω − ω.
ω

Hence by definition of u , u0 ≤ u . We obtain u0 ≥ u ≥ u0 ≥ u0 . Hence

ω ω ω ω ω

lim uω ω
ε = u .
ε→0

Lemma 8.1 implies the convergence is pointwise. The interior gradient esti-
mate implies that uω is locally uniformly Lipschitz continuous in ω.
Below we will consider the Perron lifting in an annulus ωt = Br+t (x0 ) −
Br−t (x0 ). Let us fix r and let t vary. Then uωt is monotone in t, namely

lim uωt (x) ≤ uωδ (x) ≤ lim+ uωt (x) ∀ x ∈ Ω.

t→δ − t→δ

It follows that uωt L1 (Ω) , as a function of t, is monotone and bounded.

Hence, uωt L1 (Ω) is continuous at almost all t. It follows that for almost all
t>0
lim uωs (x) = uωt (x). (8.7)
s→t
234 X.-J. Wang

Lemma 8.5 Suppose uj , u are k-admissible and uj → u a.e. in Ω. Suppose

j →u
(8.7) holds at t. Then we have uω a.e. in Ω as j → ∞.
t ωt

Proof. Since uω t
j and u
ωt
are locally uniformly Lipschitz continuous in ωt ,
by passing to a subsequence, we may assume that uω t
j is convergent. Let
w = lim uj and w be the upper semicontinuous regularization of w . Then
 ωt

by Lemma 8.1, w is k-admissible and w = u in Ω−ωt . Hence by the definition

of the Perron lifting, we have uωt ≥ w.
Next we prove that for any δ > 0, w ≥ uωt−δ . Once this is proved, we have
u ≥ w ≥ uωt−δ . Sending δ → 0, we obtain uωt = w by (8.7).
ωt

To prove w ≥ uωt−δ , it suffices to prove that for any given ε > 0, uω j ≥

t

u − ε on ϕωt−δ for sufficiently large j. By the interior gradient estimate,

uω t
j is locally uniformly Lipschitz continuous in ωt . If there exists a point
x0 ∈ ϕBr−δ/2 such that u(x0 ) > uω r
j (x0 ) + ε for all large j, by (8.4), there is
a Lebesgue point x1 ∈ Bδ/4 (x0 ) of u such that u(x1 ) > uω r 1
j (x1 ) + 2 ε for all

large j. It follows that the limit function w is strictly less than u a.e. near
x1 . We reach a contradiction as w = limj→∞ uω j ≥ limj→∞ uj = u.
r

8.3 Weak Continuity

Denote μk [u] = Sk [u]dx. It is a nonnegative measure if u is a C 2 smooth,

k-admissible function. First we prove the following monotonicity formula.
Lemma 8.6 Let u, v be two smooth k-admissible function in Ω. Suppose
u = v on ϕΩ and u(x) > v(x) for x ∈ Ω, near ϕΩ. Then
 
Sk [u] ≤ Sk [v]. (8.8)
Ω Ω

Proof. We may assume that ϕΩ is smooth, otherwise it suffices to prove (8.8)

in {u − δ > v} and send δ → 0. We have
 
d
Sk [u + t(v − u)] = Skij [u + t(v − u)](v − u)ij
dt Ω
Ω

= (v − u)i γj Skij [u + t(v − u)].

ϕΩ

It is easy to see that the integrand on the right hand side is nonnegative.
Lemma 8.7 Let uj ∈ C 2 (Ω) be a sequence of k-admissible functions which
converges to a k-admissible function u in Ω almost everywhere. Then μk [uj ]
converges to a measure μ weakly, namely for any smooth function ϕ with
compact support in Ω,
 
ϕ dμk [uj ] → ϕ dμ. (8.9)
Ω Ω
The k-Hessian Equation 235

Proof. For any open set Ω  Ω, by Theorem 7.1, μk [uj ](Ω ) is uniformly
bounded. Hence there is a subsequence of μk [uj ] which converges weakly
to a measure μ. We need to prove that μ is independent of the choice of
subsequences of {uj }.
Let {uj }, {vj } be two sequences of k-admissible functions. Suppose both
sequences {uj } and {vj } converge to u almost everywhere in Ω. Suppose that

μk [uj ] → μ, μk [vj ] → ν (8.10)

weakly as measures. To prove that μ = ν, it suffices to prove that for any

ball Br (x0 )  Ω, μ(Br ) = ν(Br ), or equivalently, for any small t > 0,

μ(Br−2t ) ≤ ν(Br+2t ), (8.11)

ν(Br−2t ) ≤ μ(Br+2t ). (8.12)

Let εj be a sequence of small positive constants converging to zero. Let

ûj = 12 εj |x|2 + uj and v̂j = 12 εj |x|2 + vj . Then


k
Sk [ûj ] = Sk [uj ] + Ck,i εij Sk−i [u] ≥ Cεkj .
i=1

By Theorem 7.1, Sk−i [uj ] is locally uniformly bounded in L1 (Ω). Hence by

(8.10), μk [ûj ] → μ weakly. Therefore we may assume directly that Sk [uj ] ≥
εj > 0 and Sk [vj ] ≥ εj > 0.
We prove (8.11) and (8.12) in two steps. In the first one we assume that
uj , vj ∈ C 2 (Ω), u ∈ C 0 (Ω), and uj , vj → u locally uniformly in Ω.
Let v̂j = vj + δj [|x − x0 |2 − r2 ]. Since |uj − vj | converges to zero uniformly,
there exists δj → 0 such that v̂j < uj in Br− 12 t and v̂j > uj on ϕBr+ 12 t . Let
A = {x ∈ Ω | v̂j (x) < uj (x). Then Br− 12 t (x0 ) ⊂ A ⊂ Br+ 12 t (x0 ). Hence by
Lemma 8.6,
  
Sk [uj ] ≤ Sk [uj ] ≤ Sk [v̂j ]
Br−2t (x0 ) A A
 
≤ Sk [vj ] + O(δj ) ≤ Sk [vj ] + O(δj ).
A Br+2t

Sending j → ∞ we obtain (8.11). Similarly we can prove (8.12).

The second step essentially repeats the first step. From the first step we
see that for any continuous k-admissible function u, we can assign a measure
μk [u] such that if a sequence of smooth k-admissible functions uj converges to
u uniformly, then μk [uj ] converges to μk [u] weakly as measure. In particular
it means μk [utj ] and μk [vjt ] are well defined, where we denote by utj , vjt and ut
the Perron lifting of uj , vj and u in ωt = Br+t (x0 )−Br−t (x0 ). By Lemma 8.5,
we have
utj , vjt → ut in Ω.
236 X.-J. Wang

By the interior gradient estimate, utj , vjt are locally uniformly Lipschitz con-
tinuous in ωt . But uj and vj may not be C 2 in ωt . To avoid such situation,
we replace utj (and vjt ) in ωt by the solution of Sk [u] = εj in ωt satisfying the
boundary condition utj = uj (and vjt = vj ) on ϕωt , for sufficiently small εj .
Let v̂jt = vjt + δj [|x − x0 |2 − r2 ]. Since |utj − vjt | converges to zero uniformly
in Br+ 34 t − Br− 34 t , there exists δj → 0 such that v̂jt < utj on ∂Br− 12 t and
v̂jt > utj on ϕBr+ 12 t . Let A be the component of {utj < v̂jt } which contains
∂Br− 12 t . Let ϕ A be the boundary of A in the annulus Br+ 34 t −Br+ 14 t . Let A
be the domain enclosed by ϕ A . Then Br− 12 t ⊂ A ⊂ Br+ 12 t . Hence as above,
  
Sk [uj ] ≤ Sk [utj ] ≤
Sk [v̂jt ]
Br−2t (x0 ) A A
 
≤ Sk [vj ] + O(δj ) ≤
t
Sk [vj ] + O(δj ).
A Br+2t

Sending j → ∞ we obtain (8.11). Similarly we can prove (8.12).

Therefore for any k-admissible function u, we can assign a measure μk [u]
to u, and μk is weakly continuous in u.
Theorem 8.1 For any k-admissible function u, there exists a Radon mea-
sure μk [u] such that
(i) μk [u] = Sk [u]dx if u ∈ C 2 (Ω); and
(ii) if {uj } is a sequence of k-admissible functions which converges to u a.e.,
then μk [uj ] → μk [u] weakly as measure.
As an application, we compute the k-Hessian measure for the function
⎧
⎨ |x − y|2−n/k k > n/2,
wk (x) = log |x − y| k = n/2,
⎩
−|x − y|2−n/k k < n/2.

We have "

1/k
2 − nk nk ωn 
δy if k = n
2,
μk [wk ] =  n
1/k n
k ωn δy if k = 2,

where ωn is the area of the unit sphere, and δy is the Dirac measure at y.

8.4 The Dirichlet Problem

As another application of Theorem 8.1, we consider the Dirichlet problem

Sk [u] = ν in Ω, (8.13)
u = ϕ on ϕΩ.
The k-Hessian Equation 237

When u is not smooth, the Hessian operator Sk [u] in (8.13) is understood as

μk [u], and u is called a weak solution. The following theorem was included in
[TW2]. Here we give a different proof.
Theorem 8.2 Let Ω be a (k − 1)-convex domain with smooth boundary. Let
ϕ be a continuous function on ϕΩ and ν be a nonnegative Radon measure.
Suppose that ν can be decomposed as

ν = ν1 + ν2 (8.14)

such that ν1 is a measure with compact support in Ω, and ν2 ∈ Lp (Ω) for

some p > 2kn
if k ≤ n2 , or p = 1 if k > n2 . Then there exists a k-admissible
weak solution u to (8.13).
Proof. Let νj be a sequence of smooth, positive functions which converges to
ν weakly as measure. By the decomposition (8.14), we may assume that νj
is uniformly bounded in Lp (Nδ0 ), where Nδ = {x ∈ Ω, dist(x, ϕΩ) < δ}. Let
ϕj be a sequence of smooth functions which converges monotone increasingly
to ϕ. Let uj be the solution of

Sk [u] = νj in Ω, (8.15)
u = ϕj on ϕΩ.

If uj is uniformly bounded in L1 (Ω), by Corollary 7.1, {uj } contains a

convergent subsequence which converges to a k-admissible function u. By
Lemma 8.7, u is a weak solution to (8.13). Therefore it suffices to prove
that uj is uniformly bounded in L1 (Ω) and the limit function u satisfies the
boundary condition u = ϕ on ϕΩ.
For δ > 0, let ηδ ∈ C0∞ (Ω) be a nonnegative function satisfying ηδ (x) = 1
when dist(x, ϕΩ) > δ and ηδ (x) = 0 when dist(x, ϕΩ) < 3δ/4. Let

νj,δ = νj ηδ + δ,

νj,δ = νj (1 − ηδ ) + δ.

Then both νj,δ and νj,δ are smooth, positive functions. Let uj,δ be the solution
of the

Sk [u] = νj,δ in Ω, (8.16)

u = ϕj on ϕΩ.

Let uj,δ be the solution of the


Sk [u] = νj,δ in Ω,
u = 0 on ϕΩ.

By Theorem 5.5, we have

uj,δ L∞ (Ω) ≤ Cνj,δ

 1/k
Lp(Ω) → 0
238 X.-J. Wang

1/k
uniformly in j, as δ → 0. By the concavity of Sk [u],

Sk [uj,δ + uj,δ ] ≥ Sk [uj,δ ] + Sk [uj,δ ]

1/k 1/k 1/k

Hence uj,δ + uj,δ is a sub-barrier to the Dirichlet problem (8.15). Hence it

suffices to prove that for any given δ > 0, uj,δ is uniformly bounded in L1 (Ω)
and uδ = lim uj,δ satisfies the boundary condition uδ = ϕ on ϕΩ.
For any fixed δ > 0, we claim that uj = uj,δ is uniformly bounded in Nδ/2
(in the following we drop the subscript δ). Indeed, if this is not true, for a
fixed, sufficiently small ε ∈ (0, 14 δ), let Dε = {x ∈ Rn | dist(x, Ω) < ε} be the
ε-neighborhood of Ω. Let η = K(ρDε (x) + K  ρ2Dε (x)), where ρDε (x) is the
distance from x to ∂Dε . Then η is k-admissible when K  is large and dx < δ1
for some δ1 > 0 depending only n, k and ϕΩ. If uj is not uniformly bounded
in Nδ/2 , by the Harnack inequality (Theorem 4.2), uj (x) → −∞ uniformly
for x ∈ {x ∈ Ω | dist(x, ϕΩ) = 2ε}. Choose K large such that η < uj on ϕΩ
and uj < η on ϕΩε . Let

∂Aj = {x ∈ Dε | ρDε (x) ∈ (0, 2ε), η(x) = uj (x)}.

Then ϕAj ⊂ Nδ/2 . Let Aj be the domain enclosed by ∂Aj . By Lemma 8.6,
 
νj,δ (Ω) ≥ νj,δ (Aj ) = Sk [uj ] ≥ Sk [η] → ∞
Aj Aj

as K → ∞. But the left hand side is uniformly bounded. The contradiction

implies that uj is uniformly bounded in Nδ/2 . Since uj is sub-harmonic, by
the mean value inequality (8.4) it follows that uj is uniformly bounded in
L1loc (Ω).
To show that u = lim uj satisfies the boundary condition u = ϕ on ϕΩ,
extend ϕ to a harmonic function in Ω. Since u is sub-harmonic, by the compar-
ison principle we have u ≤ ϕ in Ω. Hence for any y ∈ ϕΩ, limx→y u(x) ≤ ϕ(y).
Next we prove that limx→y u(x) ≥ ϕ(y) ∀ y ∈ ϕΩ. Let wj be the solution to
the Dirichlet problem

Sk [w] = K in Ω
w = ϕj on ϕΩ.

Since uj is uniformly bounded in Nδ/2 , we can fix a sufficiently large K, which

may depend on δ but is independent of j, such that the solution wj < −K on
Ω ∩ ϕNδ/2 . Recall that νj,δ = δ < 1 in Nδ/2 . By the comparison principle we
have wj ≤ uj in Nδ/2 . But when K is fixed, we have limj→∞ limx→y wj (x) =
ϕ(y) uniformly.

The uniqueness is a more complicated issue. It is proved in [TW3] that if

ν ∈ L1 , the solution in Theorem 8.2 is unique.
The k-Hessian Equation 239

9 Local Behavior of Admissible Functions

In this section we prove a Wolff potential estimate and give a necessary and
sufficient condition such that a weak solution is Hölder continuous. Results
in this section are due to D. Labutin [Ld].

9.1 The Wolff Potential Estimate

Given a Radon measure μ on Ω, we denote

 r   k1
μ(Bt (x) dt
Wkμ (x, r) = . (9.1)
0 tn−2k t

Wkμ (x, r) is called Wolff potential.

Lemma 9.1 Let u ≤ 0 be k-admissible in BR (0). Then

  k1
μk [u](B9R/10 )
≤ C inf (−u). (9.2)
Rn−2k ∂BR/2

If furthermore μk [u] = 0 in (B5R/8 − B3R/8 ) ∪ (B11R/10 − B9R/10 ), then

  k1
μ[u](BR )
inf u − inf u ≤ C , (9.3)
∂BR ∂BR/2 Rn−2k

where C is independent of R and u.

Proof. First we prove (9.2). Let ψ be the solution of Sk [ψ] = 0 in BR −

B 9R/10 , ψ = 0 on ∂BR , and ψ = u on ∂B9R/10 . By replacing u by ψ in the
annulus BR − B 9R/10 , we may assume that u = 0 on ∂BR and μk [u] = 0 in
BR − B 9R/10 . By the Harnack type inequality,

sup (−u) ≤ C inf (−u). (9.4)

∂B19R/20 ∂B19R/20

Let ϕ be a radial k-admissible function satisfying Sk [ϕ] = 0 in BR − B19R/20 ,

ϕ = 0 on ∂BR , and ϕ = inf ∂B19R/20 u on ∂B19R/20 . Then by the comparison
principle we have u ≥ ϕ in BR − B 9R/10 . By Lemma 8.6, it follows that
 k
μk [u](BR ) ≤ μk [ϕ](BR ) = Cn,k Rn−2k  inf u .
∂B19R/20
240 X.-J. Wang

We obtain by the Harnack inequality (9.4)

  k1
μk [u](B9R/10 )
≤C inf |u|.
Rn−2k ∂B19R/20

Note that u is subharmonic, inf ∂BR/2 |u| ≤ inf ϕB19R/20 |u|. We obtain (9.2).
To prove (9.3), let w be the solution of Sk [w] = μk [u] in BR and w =
inf ∂BR u on ∂BR . The solution w should be obtained as the limit of the
solution wε to Sk [w] = μk [uε ] in BR and wε = uε on ∂BR . Note that by
assumption, Sk [u] = 0 near ∂BR , so u is Lipschitz continuous near ∂BR . It
follows that w ≤ u and so

inf u − inf u ≤ inf w − inf w.

∂BR ∂BR/2 ∂BR ∂BR/2

Therefore to prove (9.3), we may assume that u =constant on ϕBR . By

subtracting we may assume that u = 0 on ϕΩ.
Let ϕ be a radial k-admissible function satisfying Sk [ϕ] = 0 in BR − BR/2 ,
ϕ = 0 on ∂BR , and ϕ = sup∂BR/2 u. Then by the comparison principle we
have u ≤ ϕ in BR − BR/2 . It follows that
 k
μk [u](BR ) ≥ μk [ϕ](BR ) = Cn,k Rn−2k  sup u .
∂BR/2

By the Harnack inequality, | sup∂BR/2 u| ≥ C| inf ϕBR/2 u|. We obtain (9.3).

Theorem 9.1 Let u ≤ 0 be a k-admissible function in B2R (0). Then we
have
C −1 Wkμ (0, R) ≤ −u(0) ≤ C{Wkμ (0, 2R) + | sup u|}, (9.5)
∂BR

where μ = μ[u], and C is independent of u and R.

Proof. First we prove the left inequality, namely |u(0)| ≥ C −1 Wkμ (0, R). For
any r ∈ (0, 12 R), let ω = B9r/8 − B3r/4 , let uω be the Perron lifting of u over
 be the Perron lifting of uω over B7r/8 . By (9.2) we have
ω, and let u
 1/k
μk [
u](B9r/10 )

≤C  − sup u
sup u 
rn−2k ∂B9r/8 ∂B7r/8

≤C sup u − sup u .
∂B3r/2 ∂B3r/4

Observing that

μk [u](Br/2 ) = μk [uω ](Br/2 ) ≤ μk [uω ](B9r/10 ) = μk [

u](B9r/10 ),

we obtain,
 1/k
μk [u](Br/2 )

≤C sup u − sup u . (9.6)

rn−2k ∂B3r/2 ∂B3r/4
The k-Hessian Equation 241

For j = 0, 1, · · · , let Rj = 2−j R. We have,

∞ 
  ∞ 
 
μk (BRj ) 1/k μk (BRj ) 1/k
C −1 ≤ Wkμ (0, R) ≤ C . (9.7)
j=0
Rjn−2k j=0
Rjn−2k

Hence letting r = Rj in (9.6) and summing up we obtain the first inequality

of (9.5).
> inequality, we may suppose μkω[u] = 0 in B2R − BR .
To prove the second
Let Rj = 2−j R, ω = ∞ j=1 (B5Rj /4 − B3Rj /4 ), and let u be the Perron lifting
of u over ω. Then u = uω in B2R − ω, μk [uω ] = 0 in ω. Since μk [u] depends
on u locally, we have, for any r > 0,

μk [u](Br ) ≤ μk [uω ](B2r ) ≤ μk [u](B4r ).

Hence to prove the second inequality we may suppose directly that μk [u] = 0
in ω.
Let uj = uBRj , the Perron lifting of u over BRj . Then uj $ u pointwise.
In particular uj $ u at the origin. Hence to prove the second inequality it
suffices to show that for all s ≥ 1,
s−1 
 
μk [u](BRj ) 1/k
| inf u| ≤ C + C| sup u| (9.8)
∂BRs
j=0
Rjn−2k ∂BR

and send s → ∞.
By (9.3) we have
 1/k
μk [u](BRs−1 )
| inf u| ≤ | inf u| + C n−2k
.
∂BRs ∂BRs−1 Rs−1 )

Applying (9.3) repeatedly, we obtain, for 0 ≤ j ≤ s,

s−1 
 
μk [u](BRj ) 1/k
| inf u| ≤ | inf u| + C .
∂BRs ∂BRj
i=j
Rjn−2k

Letting j = 0, we obtain (9.8). Since μk [u] = 0 in B5R/4 − B3R/4 , we have

| inf ∂BR u| ≤ C| supϕBR u| by the Harnack inequality. This completes the
proof.

9.2 Hölder Continuity of Weak Solutions

From Theorem 9.1 we obtain a necessary and sufficient condition for a weak
solution to be Hölder continuous.
242 X.-J. Wang

Theorem 9.2 A k-admissible function u in Ω is Hölder continuous if and

only if there exists a constant ε > 0 such that for any x ∈ Ω, r ∈ (0, 1), the
measure μk [u] satisfies

μk [u](Br ∩ Ω) ≤ Crn−2k+ε . (9.9)

Proof. If u is Hölder continuous with exponent α ∈ (0, 12 ), from the first

inequality in (9.5) we obtain
 r   k1
μ(Bt (x)) dt
≤ Crα .
0 tn−2k t

Hence
μ(Br/2 (x))
≤ Crkα .
tn−2k
We obtain (9.9) with ε = kα.
Next assume that (9.9) holds. Consider the function u in BR (0). We want
to prove that |u(x) − u(0)| < Crα for |x| < r = 12 R2 . Replacing u by the
Perron lifting uω , where ω = BR − BR/2 , we may assume that μk [u] = 0 in
ω. Let w1 be the solution of Sk [w] = 0 in BR and w = u on ∂BR . Let w2 be
the solution of Sk [w] = μk [u] in BR and w = 0 on ∂BR . Then

w1 ≥ u ≥ w1 + w2 .

Hence

u(x) − u(0) ≤ w1 (x) − [w1 (0) + w2 (0)] ≤ [w1 (x) − w1 (0)] + w2 (0).

By (9.9) and the second inequality in (9.5), we have w2 (0) ≤ CRε/k . By the
interior gradient estimate, w1 is Lipschitz continuous. Hence
C
|w1 (x) − w1 (0)| ≤ |x| ≤ C|x|1/2 .
R
We obtain
u(x) − u(0) ≤ C|x|1/2 + CRε/k .
Similarly we have u(0)−u(x) ≤ C|x|1/2 +CRε/k . Hence u is Hölder continuous
at the origin with exponent ε/2k.

From Theorem 9.2, we obtain

Corollary 9.1 Let u be a k-admissible solution (k ≤ n

2) to

Sk [u] = f. (9.10)

Suppose f ∈ Lp (Ω) for some p > n

2k . Then u is Hölder continuous.
The k-Hessian Equation 243

From Theorem 9.1, one can also prove that a k-admissible function u is
continuous at x if and only if Wkμ (x, r) → 0 as r → 0. One can also introduce
the notion of capacity, and establish various potential theoretical results, such
as quasi-continuity of k-admissible functions and the Wiener criterion for the
continuity of k-admissible functions at the boundary, just as in the Newton
potential theory. We refer the reader to [TW3, Ld] for more details. More
applications of the Wolff potential estimate can be found in [PV1, PV2].

10 Parabolic Hessian Equations

This section includes the a priori estimates and existence of solutions for the
parabolic Hessian equations used before. We refer the reader to [Lg] for more
general fully nonlinear parabolic equations of parabolic type.
Consider the initial boundary value problem

F [u] − ut = f (x, t, u) in Ω × [0, ∞) (10.1)

u(·, 0) = u0 ,
u = 0 on ϕΩ × [0, ∞)

where
F [u] = μ(Sk [u]). (10.2)

We assume that μ is a smooth function defined on (0, ∞), satisfying μ (t) > 0,
μ (t) < 0 for all t > 0, and

μ(t) → −∞ as t → 0, (10.3)
μ(t) → +∞ as t → +∞. (10.4)

Furthermore we assume that μ(σk (λ)) is concave in λ, which implies that

F [u] is concave in D2 u. A natural candidate for μ is μ(t) = log t, such as in
(5.17). But we have also used different function μ, such as in (6.3) (6.14).
We say a function u(x, t) is k-admissible with respect to the parabolic
equation (10.1) if for any given t ≥ 0, u(·, t) is k-admissible. Equation (10.1) is
parabolic when u is k-admissible. Condition (10.3) is to ensure that σk (λ) > 0
so that the admissibility can be kept at all time.

Theorem 10.1 Assume u0 ∈ C 4 (Ω) is k-admissible, and satisfies the com-

patibility condition

F [u0 ] = f on ϕΩ × {t = 0}. (10.5)

2,1
Assume ϕΩ ∈ C 3,1 and Ω is (k − 1)-convex, f ∈ Cx,t (QT ), and μ satisfies
the conditions above. Let u be an k-admissible solution of (10.1). Then we
have the a priori estimate
244 X.-J. Wang

uC 2,1 (Q ≤ C, (10.6)
x,t T)

where QT = Ω × (0, T ], C depends on n, k, ϕΩ, u0 C 4 (Ω) , supQT |u|, and

f C 1,1 (QT ) .
To prove Theorem 10.1, one first establish an upper bound for
supQT |ux | and supQT |ut |, then prove supQT |uxx | is bounded. The estimates
for supQT (|ux | + |ut |) will be given in the proof of Theorem 10.2 below. The
estimate for sup |uxx | is similar to that for the elliptic equation (3.1) and is
omitted here. We refer the reader to [Ch1, W2] for details. See also [Lg].
Note that when applying Theorem 10.1 to equation (5.17), the a priori
bound for supQT |u| is guaranteed by our truncation of |u|p , namely the
function f (u) in (5.15). In equation (5.17), the right hand side involves an
integration β(u), which satisfies the estimate C1 ≤ β(u) ≤ C2 for two abso-
lute positive constants C1 , C2 . This integration does not affect the a priori
estimate for supQT |ut |. See the proof of (10.11) below. Once ut is bounded,
β(u) is positive and Lipschitz continuous in t.
Estimate (10.6) implies that equation (10.1) is uniformly parabolic. There-
fore by Krylov’s regularity theory for uniformly parabolic equation [K1], we
obtain higher order derivative estimates. By the a priori estimates, one can
then prove the local existence of smooth solutions by the contraction map-
ping theorem. In particular, if supΩ×[0,T ] |u| < ∞ for any T > 0, the smooth
solution exists at all time t > 0.
In Step 3 of the proof of Theorem 6.3, we need a special interior gradient
estimate, namely (10.10) below, for solutions to equation (10.1) with μ given
in (6.14). We provide a proof for it below. See also [CW1]. Estimate for
higher order derivatives and existence of solutions can be obtained similarly
as above.
Theorem 10.2 Let μ be the function in (6.14). Assume u0 ∈ C 4 (Ω) is
k-admissible, and satisfies the compatibility condition (10.5). Assume that
f is C 2 in x and u, C 1 in t, and satisfies,

f (x, t, u) ≤ C0 (1 + |u|) ∀ (x, t, u) ∈ Ω × R. (10.7)

4,2
Suppose u ∈ Cx,t (Ω × [0, ∞)) is a k-admissible solution of (10.1). Then we
have, for 0 < t < T ,
u(x, t) ≥ −eC1 t sup |u0 (x)|, (10.8)
Ω

1 (p+k)/2k
|∇x u(x, t)| ≤ C2 (1 + Mt ), (10.9)
r
|ut (x, t)| ≤ C3 (1 + Mt ), (10.10)
where Mt = supQt |u|, r = dist(x, ϕΩ). The constant C1 depends only on
n, k, p and C0 ; C2 and C3 depends additionally on u0 and the gradient of f .
The k-Hessian Equation 245

Proof. Estimate (10.8) is obvious as the right hand side is a lower barrier.
To prove (10.9) and (10.10) we assume for simplicity that Mt ≥ 1. First we
prove (10.10). Let
ut
G= ,
M −u
where M = 2Mt . If G attains its minimum on the parabolic boundary of Qt ,
we have ut ≥ −C for some C > 0 depending on the initial value u0 . Hence
we may suppose G attains its minimum at an interior point in Qt . At this
point we have

utt + (M − u)−1 u2t ≤ 0,

ujt + (M − u)−1 ut uj = 0, j = 1, 2, · · · , n,

and the matrix

{uijt + (M − u)−1 (uit uj + ujt ui + ut uij ) + 2(M − u)−2 ui uj ut }

= {uijt + (M − u)−1 ut uij } ≥ 0.

Differentiating the equation (10.1) we get

Fij uijt − utt = ft + fu ut ,

Fij urij − urt = fr + fu ur ,

where Fij = ∂uϕij F [u]. We may suppose ut ≤ 0 at this point. From the above
formulae we obtain

(M − u)−1 u2t ≤ −Fij uijt + ft + fu ut

≤ (M − u)−1 ut Fij uij + ft + fu ut
≤ ft + fu u t .

Hence ut ≥ −C for some C depending on supQt ft and inf Qt fu .

Similarly let
ut
G= .
M +u
If G attains its maximum on the parabolic boundary of Qt , we have ut ≤ C.
If it attains its maximum at some point in Qt . At this point we have

utt − (M + u)−1 u2t ≥ 0,

ujt − (M + u)−1 ut uj = 0, j = 1, 2, · · · , n,
{uijt − (M + u)−1 ut uij } ≤ 0.
246 X.-J. Wang

Hence as above we obtain

(M + u)−1 u2t ≤ Fij uijt − ft − fu ut

≤ (M + u)−1 ut Fij uij − ft − fu ut
= (M + u)−1 kut μ Sk [u] − ft − fu ut .

If Sk [u] ≤ 10 at the point, by the equation (10.1) we have

ut = F [u] − f ≤ C.

Otherwise we have μ(t) = t1/p and so

1 1
μ Sk [u] = μ(Sk [u]) = (ut + f ).
p p
It follows that

(M + u)−1 u2t ≤ (M + u)−1 kut (ut + f )/p − ft − fu ut .

That is
p − k u2t kf ut
≤ − ft − fu u t .
p M +u p(M + u)
We obtain ut ≥ −C for some C depending on inf Qt ft and inf Qt fu .
Next we prove (10.9). For simplicity let us take t = T . The proof below is
similar to that of the interior gradient estimate in §4. Assume that Br (0) ⊂ Ω.
Consider
G(x, t, ξ) = ρ(x)ϕ(u)uξ ,
where ρ(x) = 1 − |x|2 /r2 , ϕ(u) = (M − u)−1/4 . Suppose

sup{G(x, t, ξ) | x ∈ Br (0), t ∈ (0, T ], |ξ| = 1}

is attained at (x0 , t0 ) (with t0 > 0) and ξ0 = (1, 0, · · · , 0). Then at the point
we have
ρi ϕi u1i
0 = (log G)i = + + ,
ρ ϕ u1
0 ≥ Fij (log G)ij − (log G)t
ρij ρi ρj ρt ϕij ϕi ϕj ϕt u1ij u1i u1j u1t
= Fij ( − 2 )− +Fij ( − 2 )− +Fij ( − )−
ρ ρ ρ ϕ ϕ ϕ u1 u21 u1
ρij ρ i ρj ρt ϕij ϕi ϕj ϕt 1
≥ Fij ( −3 2 )− + Fij ( −3 2 )− + (Fij u1ij − u1t )
ρ ρ ρ ϕ ϕ ϕ u1
ϕ ϕ ϕ
2
C f1
≥− 2
F +( − 3 2 )F11 u21 + (Fij uij − ut ) + ,
ρ ϕ ϕ ϕ u1
The k-Hessian Equation 247

where F = Fii . By our choice of ϕ,

ϕ ϕ
2
1
−3 =
ϕ ϕ 8(M − u)2

Note that Fij uij ≥ 0 and ϕ ≥ 0. We obtain

F11 u21 C ϕ f1
0≥ 2
− 2 F − ut + .
32M ρ ϕ u1
Therefore we have either
CM 2
F11 u21 ≤ F, (10.11)
ρ2
or
ϕ f1
F11 u21 ≤ CM 2 ( ut − ) ≤ CM 2 . (10.12)
ϕ u1
In (10.12) we have used the estimate (10.10).
Recall that
ρ1 ϕ
u11 = −u1 ( + u1 ).
ρ ϕ
We may assume that u1 ≥ CM/r, for otherwise (10.9) is readily verified.
Hence we have
ϕ C
u11 ≤ − u21 ≤ − u21 .
2ϕ M
From the proof of Theorem 4.1 we then have

Sk11 [u] ≥ C Skii [u],
u2k−2
Sk11 [u] ≥ C 1
,
M k−1
Therefore in the case (10.11), we obtain ρu1 ≤ CM and (10.9) follows. In
the case (10.12), we observe by equation (10.1) and estimate (10.10) that
Sk [u] ≤ CM p at (x0 , t0 ). Hence μ (Sk [u]) ≥ CM −p+1 . We therefore obtain

Cu2k−2
F11 ≥ 1
.
M p+k−2
Inserting into (10.12) we obtain u1 (x0 , t0 ) ≤ CM (p+k)/2k . Hence at the center
x = 0 we have
ρϕ(x0 , t0 )
|Du(0, t)| ≤ u1 (x0 , t0 ) ≤ CM (p+k)/2k .
ρϕ(0, t)

This completes the proof.

248 X.-J. Wang

11 Examples of Fully Nonlinear Elliptic Equations

This is the notes for my lectures under the title Fully nonlinear elliptic equa-
tions, given in a workshop at C.I.M.E., Italy. So it is appropriate to give more
examples of fully nonlinear elliptic equations here.
(i) One of the most important fully nonlinear equations is the Monge-Ampère
equation
detD2 u = f (x, u, Du). (11.1)
The Monge-Ampère equation finds many applications in geometry and applied
sciences. A special case of the Monge-Ampère equation is the prescribing
Gauss curvature equation

detD2 u
= κ(x), (11.2)
(1 + |Du|2 )(n+2)/2

where κ is the Gauss curvature of the graph of u.

(ii) A related equation is the complex Monge-Ampère equation

detuzi zj = f, (11.3)

which plays an important role in complex geometry.

(iii) The k-Hessian equation

Sk [u] = f (x), (11.4)

studied in previous sections.

(iv) The k-curvature equation

Hk [u] = f (x), (11.5)

where 1 ≤ k ≤ n and Hk [u] = σk (κ), is a class of prescribing Weingarten

curvature equations, where κ = (κ1 , · · · , κn ) are the principal curvatures of
the graph of u. The k-curvature equation is just the mean curvature equation
when k = 1, and the Gauss curvature equation when k = n.
Related to the k-Hessian and k-curvature equations are the Hessian
quotient and curvature quotient equations, that is

Sk [u]
= f (x), (11.6)
Sl [u]
Hk [u]
= f (x), (11.7)
Hl [u]
The k-Hessian Equation 249

where 0 ≤ l < k ≤ n, and Sk and Hk are respectively the k-Hessian and

k-curvature operator. A special case of (11.7) is the prescribing harmonic
curvature equation, that is when k = n, l = n − 1.
(v) The special Lagrangian equation

arctanλ1 + · · · arctanλn = c (11.8)

is a fully nonlinear equation arising in geometry. If u is a solution, the graph

(x, ∇u(x)) is a minimal surface in Rn ×Rn . When n = 3 and c = kπ, equation
(11.8) can be written as
Δu = detD2 u. (11.9)

(vi) In stochastic control theory there arises the Bellman equation

F [u] = inf {Lα [u] − fα (x)}, (11.10)

α∈V

or more generally the Bellman-Isaacs equation

F [u] = sup inf {Lα,β [u] − fα,β (x)}, (11.11)

α∈U β∈V

where α, β are indexes and Lα,β are linear elliptic operators.

(vii) Another well-known fully nonlinear equation is Pucci’s equation [GT],
which is a special Bellman equation. For α ∈ (0, n1 ], let Lα denote the set of
linear uniformly elliptic operator of the form L[u] = aij (x)∂ij u with bounded
measurable coefficients aij satisfying aij ξi ξj ≥ α|ξ|2 , Σaii = 1 for all ξ ∈
Rn , x ∈ Ω. Pucci’s operators are defined by

Pα+ [u] = sup L[u], (11.12)

L∈Lα

Pα− [u] = inf L[u].

L∈Lα

By direct calculation [GT],

Pα+ [u] = αΔu + (1 − nα)λ1 (D2 u), (11.13)

Pα− [u] = αΔu + (1 − nα)λn (D2 u),

where λ1 (D2 u) and λn (D2 u) denote the maximum and minimum eigenvalues
of D2 u.
(viii) Equation (11.1) is the standard Monge-Ampère equation. In many
applications one has the Monge-Ampère equation of general form,

det{D2 u − A(x, u, Du)} = f (x, u, Du), (11.14)

250 X.-J. Wang

where A is an n × n matrix. Similarly one has an extension of the k-Hessian

equation (11.3), that is

Sk {λ(D2 u − A(x, u, Du)} = f. (11.15)

Equation (11.14) arises in applications such as reflector design, optimal

transportation, and isometric embedding. Equation (11.15) is related to the
so-called k-Yamabe problem in conformal geometry.
Some of the above equations may not be elliptic in general, such as the
Monge-Ampère equation (11.1) and the k-Hessian equation (11.4). But they
are elliptic when restricted to an appropriate class of functions.

References

[B] I.J. Bakelman, Variational problems and elliptic Monge-Ampère equations, J. Diff.
Geo. 18(1983), 669–999.
[BN] H. Brezis and L. Nirenberg, Positive solutions of nonlinear elliptic equa-
tions involving critical Sobolev exponents, Comm. Pure Appl. Math., 36(1983),
437–477.
[Ca] L.A. Caffarelli, Interior W 2,p estimates for solutions of Monge-Ampère equations,
Ann. Math., 131 (1990), 135–150.
[CNS1] L. Caffarelli, L. Nirenberg, J. Spruck, Dirichlet problem for nonlinear second
order elliptic equations I, Monge-Amp ere equations, Comm. Pure Appl. Math.,
37(1984), 369–402.
[CNS2] L. Caffarelli, L. Nirenberg, J. Spruck, Dirichlet problem for nonlinear second order
elliptic equations III, Functions of the eigenvalues of the Hessian, Acta Math.,
155(1985), 261–301.
[Ch1] K.S. Chou, On a real Monge-Ampère functional, Invent. Math., 101(1990),
425–448.
[Ch2] K.S. Chou, Remarks on the critical exponents for the Hessian operators, Ann.
Inst. H. Poincaré Anal. Non linéaire, 7(1990), 113–122.
[Ch3] K.S. Chou, On symmetrization and Hessian equations, J. d’Analyse Math.,
52(1989), 94–116.
[CGY] K.S. Chou, D. Geng, Di, and S.S.Yan, Critical dimension of a Hessian equation
involving critical exponent and a related asymptotic result. J. Diff. Eqns 129
(1996), 79–110.
[CW1] K.S. Chou and X.-J. Wang, Variational theory for Hessian equations, Comm. Pure
Appl. Math., 54(2001), 1029–1064.
[CW2] K.S. Chou and X-J. Wang, Variational solutions to Hessian equations, preprint,
1996. Available at wwwmaths.anu.edu.au/research.reports/.
[CS] A. Colesanti and P. Salani, Generalized solutions of Hessian equations, Bull.
Austral. Math. Soc. 56(1997), 459–466.
[D] P. Delanoe, Radially symmetric boundary value problems for real and complex
elliptic Monge-Ampère equations, J. Diff. Eqns 58(1985), 318–344.
[Do] H. Dong, Hessian equations with elementary symmetric functions, Comm. Partial
Diff. Eqns, 31(2006), 1005–1025.
[E] L.C. Evans, Classical solutions of fully nonlinear, convex, second order elliptic
equations, Comm. Pure Appl. Math., 25(1982), 333–362.
[FZ] D. Faraco and X. Zhong, Quasiconvex functions and Hessian equations. Arch.
Ration. Mech. Anal. 168(2003), 245–252.
The k-Hessian Equation 251

[GT] D. Gilbarg and N.S. Trudinger, Elliptic partial differential equations of second
order, Springer, 1983.
[G1] B. Guan, The Dirichlet problem for a class of fully nonlinear elliptic equations,
Comm. PDE, 19(1994), 399–416.
[G2] B. Guan, The Dirichlet problem for Hessian equations on Riemannian manifolds,
Calc. Var. PDE, 8(1999), 45–69.
[H] L. Hörmander, Notions of convexity, Birkhauser, 1994.
[I] N.M. Ivochkina, Solutions of the Dirichlet problem for certain equations of Monge-
Ampère type (in Russian), Mat. Sb., 128(1985), 403–415; English translation in
Math. USSR Sb., 56(1987).
[ITW] N.M. Ivochkina, N.S. Trudinger, X.-J. Wang, The Dirichlet problem for degenerate
Hessian equations, Comm. Partial Diff. Eqns 29(2004), 219–235.
[J] H.Y. Jian, Hessian equations with infinite Dirichlet boundary value, Indiana Univ.
Math. J. 55(2006), 1045–1062.
[K1] N.V. Krylov, Nonlinear elliptic and parabolic equations of the second order,
Moscow, Nauka 1985 (in Russian); English translation: Dordrecht, Reidel 1987.
[K2] N.V. Krylov, Lectures on fully nonlinear second order elliptic equations, Lipschitz
Lectures, Univ. Bonn, 1994.
[KT] H.J. Kuo and N.S. Trudinger, New maximum principles for linear elliptic
equations, Indiana Univ. Math. J., to appear.
[Ld] D. Labutin, Potential estimates for a class of fully nonlinear elliptic equations,
Duke Math. J., 111(2002), 1–49.
[Lg] G. Lieberman, Second order parabolic differential equations, World Scientific,
1996.
[Lp] P.L. Lions, Two remarks on Monge-Ampère equations, Ann. Mat. Pura Appl. (4)
142(1985), 263–275.
[LT] M. Lin and N.S. Trudinger, On some inequalities for elementary symmetric
functions, Bull. Austral. Math. Soc. 50(1994), 317–326.
[PV1] N.C. Phuc and I.E. Verbitsky, Local integral estimates and removable singularities
for quasilinear and Hessian equations with nonlinear source terms, Comm. Partial
Diff.Eqns 31(2006), 1779–1791.
[PV2] N.C. Phuc and I.E. Verbitsky, Quasilinear and Hessian equations of Lane-Emden
type, To appear in Ann. Math.
[P] A.V. Pogorelov, The multidimensional Minkowski problem, New York, Wiley
1978.
[R] R.C. Reilly, On the Hessian of a function and the curvatures of its graph, Michigan
Math. J. 20(1973–74), 373–383.
[S] P. Salani, Boundary blow-up problems for Hessian equations, Manuscripta Math.
96(1998), 281–294.
[STW] W.M. Sheng, N.S. Trudinger, X.-J. Wang, The Yamabe problem for higher order
curvatures, J. Diff. Geom., 77(2007), 515–553.
[T] N.S. Trudinger, On the Dirichlet problem for Hessian equations, Acta Math.,
175(1995), 151–164.
[TW1] N.S. Trudinger and X.-J. Wang, Hessian measures I, Topol. Methods Nonlin.
Anal., 10(1997), 225–239.
[TW2] N.S. Trudinger and X.-J. Wang, Hessian measures II, Ann. Math., 150(1999),
579–604.
[TW3] N.S. Trudinger and X.-J. Wang, Hessian measures III, J. Funct. Anal., 193(2002),
1–23.
[TW4] N.S. Trudinger and X.-J. Wang, A Poincaré type inequality for Hessian integrals,
Calc. Var. and PDE, 6(1998), 315–328.
[TW5] N.S. Trudinger and X.-J. Wang, The weak continuity of elliptic operators and
applications in potential theory, Amer. J. Math., 551(2002), 11–32.
[TW6] N.S. Trudinger and X.-J. Wang, Boundary regularity for the Monge-Ampère and
affine maximal surface equations, Ann. Math., to appear.
252 X.-J. Wang

[U1] J. Urbas, On the existence of nonclassical solutions for two class of fully nonlinear
elliptic equations, Indiana Univ. Math. J., 39(1990), 355–382.
[U2] J. Urbas, An interior second derivative bound for solutions of Hessian equations,
Calc. Var. PDE, 12(2001), 417–431.
[U3] J. Urbas, The second boundary value problem for a class of Hessian equations,
Comm. Partial Diff. Eqns, 26(2001), 859–882.
[W1] X.-J. Wang, Existence of multiple solutions to the equations of Monge-Ampère
type, J. Diff. Eqns 100(1992), 95–118.
[W2] X.-J. Wang, A class of fully nonlinear elliptic equations and related functionals,
Indiana Univ. Math. J., 43(1994), 25–54.
[W3] X.-J. Wang, Some counterexamples to the regularity of Monge-Ampère equations,
Proc. Amer. Math. Soc. 123(1995), 841–845.
[WY] M. Warren and Y. Yuan, Hessian estimates for the σ2 -equation in dimension three,
preprint.
Minimal Surfaces in CR Geometry

Paul Yang

In this lecture course, I plan to introduce the subject of minimal surfaces in

pseudo-hermitian geometry from an elementary point of view. The topics to
be covered are:

1. Introduction to pseudo-hermitian structure.

2. Area of surfaces in 3-D pseudo-hermitian manifold, the p-mean curvature
equation.
3. Structure of singular sets.
4. Some applications of the C 2 -structure theory.
5. Weak solution and condition for minimizer.
6. Existence and Uniqueness of Boundary value problem.
7. Regularity results.

1 Pseudo-Hermitian Structure

We are given a 3-dim contact structure (M 2 , θ) where θ is a one-form with

non-degeneracy condition θ ∧ dθ = 0. When (M 3 , θ) is given an additional
almost complex structure
J : ker θ → ker θ
satisfying J 2 = −I, then we say (M 3 , θ, J) is a 3-dimensional pseudo-
hermitian manifold.

P. Yang
Department of Mathematics, Princeton University, Fine Hall, Washington Road, Princeton,
New Jersey 08544
e-mail: [email protected]

S.-Y.A. Chang et al. (eds.), Geometric Analysis and PDEs, 253

Lecture Notes in Mathematics 1977, DOI: 10.1007/978-3-642-01674-5 6,

c Springer-Verlag Berlin Heidelberg 2009
254 P. Yang

The basic examples come from several complex variables where Ω ⊂ C2 is

a smooth strictly pseudo-convex domain given by a defining function
Ω = {p|ϕ(p) < 0} and ∂Ω = {p|ϕ(p) = 0} and ϕ is strictly pluri-
subharmonic i.e.,
 ∂2ϕ  ∂ϕ
v i v̄ i > 0 for all v satisfying v i = 0.
∂zi ∂ z̄j ∂z i

Then it is easy to verify that −∂ϕ+

¯
√ ∂ϕ is a contact form on ∂Ω, and J is given
−1
by the natural complex structure from C2 .
A particular example is the Heisenberg group H which may be represented
as the hyper-surface ϕ = (Im) w − |z|2 = 0 in the (z, w) space.
Check −∂ϕ
¯
√ + ∂ϕ = du + 2(xdy − ydx) where w = u + iv, z = x + iy.
−1
The pseudo-hermitian connection of Tanaka and Webster defines covariant
differentiation and the associated curvature and torsion.
We begin by a normalizing choice of vector field T called the Reeb vector
field satisfying ⎧
⎨ θ(T ) = 1
⎩
LT θ = d θ(T, ·) = 0.
We then choose a complex v.f Z1 to be an eigenvector of J with eigenvalue i,
and a complex 1-form θ1 s.t.

{θ, θ1 , θ1̄ } is dual to {T, Z1 , Z1̄ .}

It follows that
dθ = ih11̄ θ1 ∧ θ1̄ for real h11̄ > 0
then we can normalize further by choosing Z1 so that h11̄ = 1

dθ = iθ1 ∧ θ1̄ .

The pseudo-hermitian connection  is given by

 Z1 = ω11 ⊗ Z1 ,  Z1̄ = ω1̄1̄ ⊗ Z1̄ , T = 0.

The connection form ω11 is uniquely determined by

⎧ 1
⎨ d θ = θ1 ∧ ω11 + A11̄ θ ∧ θ1̄
⎩
ω11 + ω1̄1̄ = 0
Minimal Surfaces in CR Geometry 255

Then we have

d ω11 = W θ1̄ ∧ θ1̄ + 2i Im (A11 , 1̄ ) θ1̄ ∧ θ

A11̄ − Torsion

W − Webster scalar curvature.

√
Converting to real forms: θ1 = e1 + −1e2 , Z1 = 12 (e1 − ie2 ), ω11 = iw

dθ = 2e1 ∧ e2
e1 = ω ⊗ e2 e2 = −ω ⊗ e1

de1 = −e2 ∧ ω + θ ∧ (% A11 e1 + &A11 e2 )

de2 = e1 ∧ ω + θ ∧ (& A11 e1 − % A11 e2 )

dω(e1 , e2 ) = −2W

[e1 , e2 ] = −2T − ω(e1 )e1 − ω(e2 )e2

[e1 , T ] = (% A11 ) e1 − [(& A11 ) + ω(T )]e2

[e2 , T ] = [& A11 + ω(T )] e1 + (% A11 )e2 .

Extend J to all T M by requiring J(T ) = 0 so that

J 2 x = −x + θ(x)T ∀x ∈ T M

Then the connection is characterized by the conditions

1) x V ∈ ξ ∀ V ∈ ξ.
2) T = 0 , J = 0, (θ) = 0.
3) T or (x, y) = −dθ(x, y)T
T or (T, Jy) = −JT or (T, y)
T or (T, x) = − 12 J LT Jx for x ∈ ξ.

Remark 1.1. Vanishing torsion condition means that T is an infinitesimal CR

transformation.
Equation of Geodesics It turns out that, if x = γ̇(t) then
⎧
⎨ 'x x = αJx
⎩
α̇ = T or (T, x), x .

This can be checked considering the energy functional

 
E[γ] =  γ̇(t) 2 dt.
0
256 P. Yang

∂
Letting v = ∂u γ, then



∂
∂u E[γn ]u=0 = vx, x dt
0

=2 v x, x dt
0

=2 x v + (T or (v, x)), x dt
0

=2 xv, x − v, x x + T or (v, x), x dt
0

=2 − v, x x + T or (v, x), x dt
0


(write x x = αJx + y) = 2 −αv, Jx +v, y +θ(v)T or (T, x), x dt
0

⎛ ⎞
dθ(v, x) = −v, Jx 
⎝  ⎠ = 2 − α· (xθ(v))+v, y +θ(v)T or (T x), x dt
−xθ(v) 0


= 2 θ(v)xα(t) + T or (T, x), x + v, y dt.
0

Thus: v, y = 0 and xα(t) + T or (T, x), x = 0 which is the required Euler
equation.

Remark 1.2. Under the assumption T or (T, x) = 0 the equation of geodesics

becomes ⎧
⎨ x x = α Jx
⎩
α̇ = 0 ,
a second order system.
Example 1.3. In the Heisenberg group θ = dz + xdy − ydx
Then the geodesics γ issuing from 0 have constant curvature α. The projection
of γ to the x y plane are planar circles of radius α1
Minimal Surfaces in CR Geometry 257

2 Area of a Surface in (M 3 , θ, J )


Given a smooth surface ⊂ (M 3 , θ, J), there is at each point p a (unique
up to sign) vector e ∈ Tp ∩ ξp if the latter intersection is one-dimension,
the generic situation.

e2

e1

The line field defines a foliation consisting of integral curves of e1 , which

we shall call characteristic curves.

 
Definition 2.1. p ∈ is a singular point if Tp = ξp .
Let us define the p-area as variation of volume:

1
Vol(Ω) = θ ∧ dθ for a domain Ω ⊂ M.
2 Ω
258 P. Yang

Now suppose = ∂Ω, let us compute

δf e2 v(Ω)

by considering a variation of the surface in
the direction f e2 , where f is
a smooth function withcompact support in , we then extend f e2 locally
to a neighborhood of  . The vector field f e2 generates a 1 − p-group of
diffeomorphisms ϕt near :

d 
 
d 
δf e2 Vol (Ω) = dt t=0
Vol (ϕ t (Ω)) = dt t=0 ϕ∗t (θ ∧ dθ)
 Ω

= Lf e2 (θ ∧ dθ)
Ω

= (d ◦ if e2 + i ◦ df e2 )θ ∧ dθ
Ω

= if e2 θ ∧ dθ
∂Ω

= f θ ∧ e1
∂Ω

Hence we define p-area as



p Area ( ) = θ ∧ e1


Next we define p-mean curvature H from the first variation of the p-area:

  
δf e2 θ∧e = 1
Lf e2 (θ ∧ e ) =
1
if e2 ◦ d(θ ∧ e1 )
  


= if e2 (θ ∧ e2 ∧ ω)



= −f θ ∧ ω

Minimal Surfaces in CR Geometry 259

= −f ω(e1 ) θ ∧ e1 − f ω(e2 )θ ∧ e2

6 74 5

H


= −f ω(e1 ) + 0.

6 74 5

H

Thus we define the mean curvature to be

H = ω(e1 ).

Computation Let ψ be a defining function of then b ψ = e1 (ψ)e1 +
(e2 ψ)e2 = (e2 ψ)e2 .
b ψ
∴ | b ψ|
= e2 - called the horizontal normal

divb e2 = e1 e2 , e1 + e2 e2 , e2

= −e2 , e1 e1

= −H

Example 2.2. In the Heisenberg group

 
H = D−3 (uy + x) uxx − 2(uy + x) uxy + (ux − y)2 uyy .
2

where
/ 01/2
2
D = (ux − y) + (uy + x)2 .

Often we will also denote the horizontal normal as N .

3 Structure of the Singular Set

In this section we study the singular set of a C 2 -surface

 in the Heisenberg.
So let us consider a surface represented
 as a graph : z = u(x, y) over a
domain Ω ⊂ R2 . The singular set of is given by

S[u] = {(x, y) ∈ Ω | ux − y = 0, uy + x = 0}

We have the following dichotomy for S[u]:

260 P. Yang

Theorem A ([CHMY]) Let u ∈ C 2 (Ω) satisfy

⎧
⎨ divN (u) = H in Ω  S[u]
⎪
 
⎪
⎩ |H(p)| = O 1
|p−p0 |

where p0 is a singular point. Then either

(a) p0 is an isolated singular point of S[u] and

d(N ⊥ D)p0 = I

or
(b) S[u] ∩ Bp0 (∈) is a C 1 -curve and the characteristic curves meet S[u] ∩
Bp0 (t) in a C 1 manner transversely, so that the e1 -line field has a conti-
nuous extension over S[u].

Remark 3.1. In case (a), the condition d(N ⊥ D)p0 = I means that the e1 -field
has index one at an isolated singular point. In case (b), the assertion is that
the line field given by e1 extends continuously across the singular curve

Proof of Theorem A:
Given any unit vector (a, b) at p0 , consider the function Fab = a(ux − y) +
b(uy + x) and the set

Γa,b = {(x, y) | Fab (x, y) = 0}

= S[u] ∪ {(x, y) | Nu = ± (b, −a)}.

We claim that if p0 is not an isolated point, then Γa,b is a C 1 curve passing

through p0 for all but one choice of (a, b). Consider the matrix
 
uxx uxy − 1
U = .
uxy + 1 uyy

If rank (U (p)) = 2, then p0 is an isolated singular point. On the other hand,

if p0 is not an isolated singular point, we have

rank (U (p0 )) = 1.

Therefore there exists at most one direction (a0 , b0 ) s.t.

Fa0 b0 = ((a0 , b0 ) U ) (p0 ) = 0.

Hence for all other choices of (a, b) = ±(a0 , b0 ), The Γa,b ’s are C 1 curve
passing through p, and have a common tangent direction at p0 .
Minimal Surfaces in CR Geometry 261

If S[u] is not a C 1 -curve near p0 then for different choices of direction say
(a, b) and (a , b ), such a picture must hold:

Ωi
P0
θi

Let v be the unit normal to the ith region enclosed by Γ and Γ  . We have in
polar coordinates centered at p0 ,
   
     
   
 
N (u) · v ds  =  H dx dy 

   
∂Ωi Ωi

 i )θi
r(ē  
C
≤ dθ rdr
r
r(ei ) 0

≤ C θi | r(ēi ) − r(e1 )|.

But
LHS ≥ (|(−b, a) · ν0 − (−b , a ) · ν0 | − δi ) |r(ēi ) − r(ei )|

≥ (|c − c | − δi ) | r(ēi ) − r(ei )|

where
c = c , θi and δi −→ 0 as i −→ ∞.
This is a contradiction.
Next we consider case (a) when p0 is an isolated singular point. Let
N ⊥ D = (uy + x, −ux + y). We consider the first order Taylor expansion
near p0 :

(uy + y) (p) = (uyx + 1) (p0 ) x + uyy (p0 ) y + o(|p − p0 |)

(ux − y) (p) = uxx (p0 ) x + (uxy − 1) (p0 ) y + o(|p − p0 |).
   
uxx uxy + 1 ca
Set (p0 ) = .
uxy − 1 uyy db
262 P. Yang

Then

P = (uy + x)2 uxx − 2(uy + x) (ux − y) uxy + (ux − y)2 uyy

   
ca 'x
= (bc − ad) · ('x, 'y) + o (|p − p0 |2 )
db 'y
   3
 ab ' y 
D3 = ((ux − y)2 + (uy + x)2 )3/2 =  + o(|p − p0 |3 ).
cd 'y 
P
1

Hence the condition H = D3 = O r implies:

Choosing 'y = 0:

(bc − ad) c('x)2 (bc − ad)c

H ∼ = 2 , we find c = 0.
((a2 + c2 ) ('x)2 )3/2 (a + c2 )3/2 'x

Choosing 'x = 0:
 
1
H ∼ O =⇒ b = 0.
|p − p0 |

Hence,  
−ad (a + d) 'x 'y 1
H ∼ = O
(a2 ('x)2 + de ('y)2 )
3/2 |p − p0 |
Hence a + d = 0.  
⊥ 10
Thus d (N ') | = .
p0 01
This finishes the proof in the case (a).
In case (b), it remains to show that characteristic curves cross singular
curves transversely in C 1 -manner. Without loss of generality let us assume
Γ is a singular curve, locally represented as a graph over the y-axis. Thus
given c near y0 , there is xc s.t. (xc , c) ∈ Γ . Let us consider the ratio

(uy +x)(x,c) (uy +x) (x,c)−(uy +x) (xc ,c)

(ux −y)(x,c) = (ux −y) (x,c)−(ux −y) (xc ,c)

(x−xc ) (uxy +1) (xc ,c)

= (x−xc ) (uxx ) (xc ·c)

(uxy +1) (x1c ,c)

= uxx (xc ,c)
Minimal Surfaces in CR Geometry 263

where in the second line, the mean value theorem gives the existence of xc
and xc between x and xc . Now letting (x, c) tends to (x0 , y0 ) we find
uy + x uxy + 1
lim = (p0 ).
p>p0 ux − y uxx

Therefore the right-handed limit of N (u) must agree with the left-handed
limit of N (u) up to sign.

4 Some Applications of Theorem A

It is always important to understand the blow-up limits of minimal surfaces.

In case of graphs, the classification is known as Berstein’s theorem. In the
setting of entire p-minimal graphs, we have
Theorem 1 ([CHMY]) The entire C 2 -solutions of equation

(4.1) (uy + x)2 uxx − 2(uy + x) (ux − y) uxy + (ux − y)2 uyy = 0 are either

(a) plane or
(b) of the form
u = −abx2 + a2 − b2 xy + aby 2 + g(−bx + ay)
where g is a C 2 -function of a single variable, and a, b ∈ R are such that
a2 + b2 = 1.

Remark 4.1. 1. ([CH]) Cheng and Hwang classified entire, properly embedded
simply connected p-minimal surface. These include the following family of
surfaces:
(x − x0 ) cos θ(t) + (y − y0 ) sin θ(t) = 0
where t = z − y0 x + x0 y, and θ is a C 2 - function of t.
2. Later on we shall prove that the following surface are entire C 1,1 minimizing
solutions ⎧
⎨−xy + (cot θ) y 2 , y ≥ 0
u(x, y) =
⎩
−xy + (cot η) y 2 , y ≤ 0
where θ + η = 2π.
Proof of Theorem 1:
Geometrically H = 0 is the condition ω11 (e1 ) = 0. This is the same as saying
the integral curve of e1 -field are contact geodesics, which are lines everywhere
tangent to the contact plane. Therefore the graph of u is a ruled surfaces over
regular points. Consider the projection of the rulings to the xy plane. There
264 P. Yang

are two possibilities according to whether the rulings project onto parallel
lines. If so, we may by rotating the coordinates if necessary write

0 = e21 u = uxx

so that Nu = ∂y hence

u = xf (y) + g(y) =⇒

u = xy + g(y) for some C 2 function g.

In the other possibility, some rulings intersect at some p0 . Then p0 is a

singular point. Applying Theorem A, it follows that all the rulings must pass
through p0 . Reasoning as above, but using polar coordinates, one can check
that hence the surface is a plane.
A second application is a topological obstruction to the existence of C 2
embedded surfaces of bounded mean curvature in a general pseudo-hermitian
manifold.
Theorem
 2 Let (M 3 , θ, J) be a 3-dimensional pseudo-hermitian structure,
⊂ Mbe a closed C 2 embedded surface of bounded p-mean curvature, then
genus ( ) ≤ 1.
Proof
The index of the line field-e1 at isolated singular point is 1. The e1 -field can
be extended continuously across the singular curves. It follows from Hopf
index theorem that the Euler characteristic of such a surface is non-negative.
The classification of singular set also has application to the Isoperimetric
problem proposed by Pansu [P]. Consider the extremal problem of a domain
Ω ⊂ H1 with smooth boundary and of a given volume, and try to minimize
the p-area of the boundary ∂Ω. The optimal configuration, if smooth, will
satisfy the condition
H = λ
where λ is the optimal isoperimetric constant. Thus the optimal surface will
have rulings given by geodesics of curvature λ. The recent article [RR] uses
Theorem A to prove that such a configuration must be congruent to the
surface obtained by rotating the following geodesic γλ around the z-axis:
⎧
⎪
⎪
1
x(s) = 2λ sin(2λs)
⎪
⎪
⎨
1
y(s) = 2λ (−1 + cos(2λs))
⎪
⎪
⎪
⎪
⎩ 1
z(s) = 2λ (s − 2λ1
sin(2λs)).

This surface has precisely two singular points (0, 0, 0) and (0, 0, π).
Minimal Surfaces in CR Geometry 265

5 Weak Solutions and Condition for Minimizer

In this section we consider the functional


F [u] = |u + F| + Hu dx
Ω

for a fixed vector valued function F ∈ L1 (Ω), and a prescribed H ∈ L∞ (Ω)

for all test functions u ∈ W 1,1 (Ω). We make the following defintion:

Definition 5.1. (Weak solution): u ∈ W 1,1 (Ω) is a weak solution to the

equation
u + F
(5.1) div |u +F| = H

in Ω if for all ϕ ∈ C0∞ (Ω) we have

  
(5.2) |ϕ| + N (u) · ϕ + Hϕ ≥ 0
S[u] Ω  S[u] Ω

where
S[u] = {x ∈ Ω| u(x) + F (x) = 0}

u + F
N [u] = |u + F | .

In the general case, the singular set may be quite large. Even in the special
case Ω ⊂ R2n and F (x, y) = (−y, x) corresponding to the p-mean curvature
equation, there exists examples (see [Ba]) where the singular set has positive
measure.
We observe that when F (x, y) = (−y, x) we have

(5.3) (N (u) − N (v)) · (u − v) ≥ 0.

Under such a condition we can characterize minimizers.

Theorem B Suppose F satisfies (5.3), then u ∈ W 1,1 (Ω) is a minimizer for

F [u] = | u + F| + Hu dx
Ω

if and only if u is a weak solution.

Remark 5.2. 1. If Hn−1 (S[u]) = 0, and u is a smooth solution to (5.1) then

u is a minimizer.
2. Thus a C 2 -solution to the equation (4.1) need not be a weak solution.
In fact, an elementary calculation shows that if S[u] is a smooth singular
curve Γ , then the characteristic curves meeting Γ must do so orthogonally.
266 P. Yang

Outline of Proof of Theorem B:

Given a test function ϕ ∈ C0∞ (Ω), consider the variation u = u + ϕ.
We will verify the following:
Claims:
  
dF [u ] 
(1) = ± |ϕ| + N (uˆ) · ϕ + Hϕ.
d ˆ
S[uˆ] ΩS[uˆ] Ω
if ˆ is regular (see (3) below).
(2)  −→ F [u ] is a Lipschitz function.
(3) {| meas (S[u ] ∩ {|ϕ| = 0}) > 0} is a countable set, such  shall be
called singular.
(4) If 2 > 1 , are regular then

dF (u )  dF (u ) 
 −  ≥ 0.
d 2 d 1

(5) For any sequence of regular j $ ˆ

dF [u ]  dF [u ] 
lim  = .
j∞ d j d ˆ

It will then follow that a weak solution is a minimizer:

1
dF [u ]
F [u + ϕ] − F [u] = d ≥ 0.
d
0

Proof
  
F [u ] = |u + F | + |u + F | + Huˆ + ( − ˆ) Hϕ
S[uˆ] ΩS[uˆ] Ω
(1)   
= | − ˆ||ϕ| + |u + F | + Huˆ + ( − ˆ) Hϕ.
S[uˆ] ΩS[uˆ] Ω

Comparing with
 
F[uˆ] = |uˆ + F | + Huˆ
ΩS[uˆ] Ω

we find
  
F [uˆ] − F [u ] | − ˆ
| |u + F | − |uˆ + F |
= |ϕ| + + Hϕ.
−ˆ  −ˆ   − ˆ
S[uˆ] ΩS[uˆ] Ω
Minimal Surfaces in CR Geometry 267

The integrand in the second term is bounded by |ϕ|, hence (2).

Expanding the integrand: we find

|u + F | − |uˆ + F | 2(uˆ + F ) · ϕ + ( − ˆ)|ϕ|2

= .
 − ˆ |u + F | + |uˆ + F |

Taking limit  −→ ˆ in case ˆ is regular the first integral vanishes and we

obtain (1).
(3) Follows from the simple observation that for = the sets S[u ]∩{ϕ = 0}
and S[u ] ∩ {ϕ = 0} are disjoint.
(4) For regular values 2 > 1

dF [u ] dF [u ] 1
− = (N (u2 )−N (u1 ))· (u2−u1 ) ≥ 0
d 2 d 1 2 −1
S[u1 ]∪S[u2 ]
Ω

on account of (5.3).
(5) First observe that for each j,
  
|ϕ| = 0 hence |ϕ| = 0 and N (uj ) ·  ϕ = 0.
S[uj ] >
∞ >
∞
S[uj ] S[uj ]
1 1

 
dF [u ] 
∴ j
= N (uj ) · ϕ + Hϕ
d
ΩS[uj ] Ω

 
= N (uj ) · ϕ + Hϕ.
>
∞ Ω
Ω S[uj ]
1

>
∞
In S[uˆ]  S[uj ] we have
1

(uˆ + F ) + (j − ˆ) ϕ (j − ˆ) ϕ

N (uj ) = =
|uˆ + F + (j − ˆ)ϕ| (j − ˆ)|ϕ|
>
∞
and in Ω  (S[uˆ] ∪ S[uj ]) lim N (uj ) = N (uˆ). Hence
1 j −→∞
⎧ ⎫
⎪
⎪ ⎪
⎪
 ⎪
⎪   ⎪
⎪
⎨ ⎬
N (uj ) · ϕ = + N (uj ) ·  ϕ
⎪
⎪ ⎪
⎪
>
∞ ⎪
⎪ >
∞

>
 ⎪
⎪
Ω S[uj ] ⎩S[uˆ] S[uj ] Ω  S[uˆ]∪ S[uj ] ⎭
∞

1 1 1

 
−→ |ϕ| + N (uˆ) · ϕ.
S[uˆ] Ω  S[uˆ]
268 P. Yang

Therefore
 
dF [u ] 
j
= N (uj ) · ϕ + Hϕ
d
Ω  S[uj ] Ω

  
−→ |ϕ| + N (uˆ) · ϕ + Hϕ .
S[uˆ] Ω S[uˆ] Ω

dF [u ] 
= ˆ
d

6 Existence and Uniqueness for Boundary Value

Problems

Let Ω ⊂ R2 be a convex domain with smooth boundary. Given ϕ, we consider

the problem to find a minimizer of the p-area functional so that:
(ux − y, uy + x)
(6.1) div = 0
((ux − y)2 + (uy + x)2 )1/2
with u agreeing with ϕ on ∂Ω.
Theorem C There exists u ∈ C 0,1 (Ω̄) which is a weak solution to (6.1) that
u = ϕ on ∂Ω.
Indication of Proof
We make use of the elliptic approximation to the equation (6.1):
(ux − y, uy + x)
(6.2) div  = 0.
 + (ux − y)2 + (uy + x)2
2

and then let  −→ 0, making sure that the solution to the approximate
equation converges with uniformly bounded Lipschitz constant. This is a
somewhat routine application of known elliptic theory, we skip the details.
We discuss next the question of uniqueness. The following example was
first discovered by S. Pauls [P].

Example 6.1. 
u : u= x2 + xy

: v = xy + 1 − y 2
v
 
are C 2 solutions of the equation (4.1). u and v coincide over the unit
circle x2 + y 2 = 1, thus represent distinct solutions to the boundary value
problem over the unit disc. The singular set S[u] is given by the y-axis while
the singular set S[v] is given by {x = y}.
Minimal Surfaces in CR Geometry 269

It is easy to check that in each case the characteristic curves do not meet
the singular curve orthogonally. It follows that neither surface is a minimizer.
On the other hand, we have the following uniqueness criterion:
Theorem C ([CHMY]) (Comparison principle for C 2 solutions). Suppose
Ω ⊂ R2 is a smoothly bounded domain, and that u, v satisfy
⎧
⎨ div N (u) ≥ div N (v) in Ω  (S[u] ∪ S[v])
⎩
u ≤ v on ∂Ω

and
H 1 − meas (S[u] ∪ S[v]) = 0;
then
u ≤ v.
Claim: N (u) = N (v) on the set Ω+ = {u > v}.
We observe that it follows that Ω+ is empty: for, if not, Sard’s theorem
shows that there exists arbitrarily small  > 0 such that Ω+, = {u − v ≥ }
has smooth boundary Γ . Hence Γ may be parametrized by a C 1 curve
(x(s), y(s), u(s))
du dy dx
+x −y = 0
ds ds ds
where N ⊥ (u) is tangent to Γ : we compute

N ⊥ (u) = N ⊥ (u) · u = 1
D {(u)⊥ + (x, y)} · u

= 1
D (x, y) · u

= 1
D (x, y) · {(u) + (−y, x)}

= (x, y) · N (u).

Similarly
N + (u) · v = N − (v) · v = (x, y) · N (v).
This leads to a contradiction:
? 
0 = θ = dθ > 0.
Γ Ω,+

Proof of Comparison Principle

We begin with the formula:
D u + Dv
(6.3) (u − v) · (N (u) − N (v)) = · |N (u) − N (v)|2 .
2
270 P. Yang

For simplicity, we start by assuming there is no singular set. Let Ω+ =

{u > v}.
Proof of Claim that N (u) = N (v) on Ω+
Again it suffices to show N (u) = N (v) on Ω+,δ = {u − v > δ} where ∂Ω+,δ
is smooth. Consider the integral
?
Iδ = tan−1 (u − v) (N (u) − N (v)) v ds
∂Ω+,δ


1
= { (u − v) (N (u) − N (v))
1 + (u − v)2
Ω+,δ

+ tan−1 (u − v) div (N (u) − N (v))} dx dy


1 Du + Dv
≥ · | N (u) − N (v)|2 dx dy.
1 + (u − v)2 2
Ω+,δ

Hence if N (u) = N (v) somewhere, the RHS is bounded from below by a

positive number for all δ small. But
?  
−1 −(u − v)
I ≤ tan δ
δ
(N (u) − N (v)) · ds
|(u − v)|
∂Ω+,δ

where the integrand is non-positive. This leads to a contradiction.

>
N
When there is a singular set, we cover S[u] ∪ S[v] by small balls Bj s.t.
 j=1
|∂Bj | < 2 . Then apply the argument to the region Ω+,δ with these balls
removed. Since  can be made arbitrarily small, we still have a contradiction.

7 Regularity of the Regular Part

Our motivation in this section comes from the analytic description of sets of
bounded perimeter by Franchi, Serapioni and  Serra-Cassano ([FSS]). Their
main result is the decomposition of any set of bounded perimeter
 - which
includes in particular regions bounded by Lipschitz graphs: can be decom-
posed into a regular part which is a CH1 -hypersurface, and a singular set of
measure zero. Although such a surface needs not be C 1 in the ordinary sense,
we are interested in improving the regularity of the C 1 regular part. It will
be worthwhile to present some examples.
Minimal Surfaces in CR Geometry 271

Example 7.1. ⎧
⎨ xy, y ≥ 0
u =
⎩
0 y ≤ 0.

The characteristic lines are drawn as follows:

This is a piecewise C 1 surface with folding angles along the x-axis.

Example 7.2. This example is the minimizer of the boundary value side
considered in Example 6.1 in section VI.

x = s(sin η(t)) + α(t)

y = −s(cos η(t)) + β(t)

z = s[β(t) sin η(t) + α(t) cos η(t)] + γ(t)

where
η(t) = π
2 + θ2 (t) − δ(t)
cos (θ2 (t) − 3
2 π) = √t
2
 C D√
tan δ(t) = t 1 − t2 /2 (1 − t2 2)

α(t) = t cos 3π/8

β(t) = t sin 3π/8

γ̇(t) = cos2 3π/8 + cos 3π/8 sin 3π/8.

272 P. Yang

5π 3π
8 8

Γ 1t 1π
q1 8
Γ 2t
t
x
L

9π
8 q2

11π – 3π
8 8

This surface is C 1,1 but fails to be C 2 along the singular curve which is the
line through origin making angle 3π/8 with x-axis.

Example 7.3. This example “smoothens out” the fold of example 7.1 by intro-
ducing a buffer region. Choose α, β = [0, 1] −→ R be C ∞ smooth function
with the properties;

(7.1) α(0) = α 12 = α(1) = 0, α (0) = α (1) = 1

α(t) > 0 for 0 < t < 1/2 , α(t) < 0 for 12 < t < 1; |α(t)| < π2
β(0) = 0, β( 12 ) = 12 , β(1) = 1, β  (t) = 0 0 < t < 1
β (n) (0) = β (n) (1) = 0 for all n = 1, 2 . . .

x = s cos(α(t))

y = s sin(α(t)) + β(t)

z = s β(t) cos(α(t))
for (s, t) ∈ [0, ∞) × [0, 1]. Then for (s, t) ∈ (−∞, 0)] × [0, 1] we take

x(t) = s cos (α(t)1)

y(t) = −s sin (α(t)) + β(t)

z(t) = s β(t) cos (α(t))

Minimal Surfaces in CR Geometry 273

We then glue to the original surface

⎧
⎨ 0 for y ≤ 0
u =
⎩
x for y ≥ 1.

The diligent reader may check that this gives a minimizing surface which
is a graph of a function u ∈ C 1,α (R2 ) for every α < 1. It only misses being
C 2 at the point (0, 0), (0, 12 ) and (0, 1).

References

[B] Z. Balogh; Size of rectifiable sets and functions of prescribed gradient, J. Reine
Angew. Math. 564 (2003), 63–83.
[CHMY] J.-H. Cheng, J.-F. Hwang, A. Malchiodi and P. Yang; “Minimal surfaces in
Pseudo-hermitian geometry, ” Ann. Scuola Normal Sup., Pisa, I. Sci , (5) 2005,
129–177.
[CHY1] J.-H. Cheng, J.-F Hwang and P. Yang; “Existence and Uniqueness for P -area
minimizers in the Heisenberg group,” Math Ann., (2007), 337 = 253 − 293.
[Ch.Y] S. Chanillo and P. Yang; “Isoperimetric inequality and volume comparison on
CR manifolds (WIP).
[P] P. Pansu; “Une inegalité isoperimétrique sur le group de Heisenberg,” C.R. 295,
127–130.
[Pa] S. Pauls, “Minimal surfaces in the Heisenberg group,” Geometry Dedicata, 104
(2004), 201–231.
[T] N. Tanaka, “A differetial geometric study on strongly pseudo-convex mani-
folds,” Kinokuniya, Tokyo, (1975).
[W] S. Webster, “Pseudo-hermitian structures of a real hyper-surface,” JDG 13
(1978), 25–41.
List of Participants

1. Abdellaoui Boumediene
[email protected]
Univ. Tlemcen, Tlemcen, Algeria
2. Ambrosetti Antonio
[email protected]
SISSA/ISAS, Trieste, Italy (editor)
3. Amendola Maria Emilia
[email protected]
Univ. Salerno, Salerno, Italy
4. Ana Primo
[email protected]
Univ. Autonoma de Madrid, Madrid, Spain
5. Antonangeli Giorgio
[email protected]
Univ. La Sapienza, Roma, Italy
6. Autuori Giuseppina
[email protected]
Univ. Firenze, Firenze, Italy
7. Bonforte Matteo
[email protected]
Univ. Autonoma de Madrid, Madrid, Spain
8. Borrello Francesco
[email protected]
Univ. Catania, Catania, Italy
9. Capogna Luca
[email protected]
Univ. Arkansas, Fayetteville, AR, USA
10. Caponigro Marco
[email protected]
SISSA/ISAS, Trieste, Italy

275
276 List of Participants

11. Caravenna Laura

[email protected]
SISSA/ISAS, Trieste, Italy
12. Catino Giovanni
[email protected]
Univ. Pisa, Pisa, Italy
13. Chang Alice
[email protected]
Princeton Univ., Princeton NJ, USA (editor)
14. Charro Caballero Fernando
[email protected]
Univ. Autonoma de Madrid, Madrid, Spain
15. Coglitore Federico
[email protected]
Univ. Roma Tre, Roma, Italy
16. Fall Mouhamed
[email protected]
SISSA/ISAS, Trieste, Italy
17. Gazzini Marita
[email protected]
SISSA/ISAS, Trieste, Italy
18. Ghannam Clara
[email protected]
Univ. La Rochelle, La Rochelle, France
19. Ghergu Marius
[email protected]
Inst. of Math. Simion Stoilow, Romanian Academy, Bucharest, Romania
20. Giuffrè Sofia
[email protected]
Univ. Mediterranea di Reggio Calabria, Reggio Calabria, Italy
21. Gonzalez Maria del Mar
[email protected]
The University of Texas at Austin, Austin, TX, USA
22. Grillo Gabriele
[email protected]
Politecnico di Torino, Torino, Italy
23. Gursky Matthew J.
[email protected]
Univ. Notre Dame, Notre Dame, IN, USA (lecturer)
24. Han Xiaoli
[email protected]
ICTP, Trieste, Italy
25. Ianni Isabella
[email protected]
SISSA/ISAS, Trieste, Italy
List of Participants 277

26. Khmelynitskaya Alena

[email protected]
Kazan State Univ., Kazan, Russia
27. Kokocki Piotr
[email protected]
Nicolaus Copernicus Univ., TORUN, Poland
28. Kruglikov Boris
[email protected]
Univ. Tromso, Tromso, Norway
29. Loiudice Annunziata
[email protected]
Univ. Bari, Bari, Italy
30. Lanconelli Ermanno
[email protected]
Univ. Bologna, Bologna, Italy (lecturer)
31. Mahmoudi Fethi
[email protected]
SISSA/ISAS, Trieste, Italy
32. Malchiodi Andrea
[email protected]
SISSA/ISAS, Trieste, Italy (editor, lecturer)
33. Mantegazza Carlo
[email protected]
Scuola Normale di Pisa, Pisa, Italy
34. Mercuri Carlo
[email protected]
SISSA/ISAS, Trieste, Italy
35. Mircea Petrache
[email protected]
Scuola Normale Superiore Pisa, Pisa, Italy
36. Monticelli Dario
[email protected]
Univ. Milano, Milano, Italy
37. Montoro Luigi
[email protected]
Univ. Calabria, Cosenza Italy
38. Munoz Claudio
[email protected]
Univ. Pierre et Marie Curie, Paris 6, Paris, France
39. Ndiaye Cheikh Birahim
[email protected]
SISSA/ISAS, Trieste, Italy
40. Olech Michal
[email protected]
University Paris XI, Paris, France
278 List of Participants

41. Paniccia Irene

irene [email protected]
Univ. Roma La Sapienza, Roma, Italy
42. Rebai Yomna
[email protected]
Facultè des Sciences de Bizerte, Bizerte, Tunisia
43. Riey Giuseppe
[email protected]
Univ. Calabria, Cosenza, Italy
44. Ruiz David
[email protected]
SISSA/ISAS, Trieste, Italy
45. Scienza Matteo
[email protected]
Scuola Normale Superiore Pisa, Pisa, Italy
46. Selvitella Alessandro
[email protected]
SISSA/ISAS, Trieste, Italy
47. Siciliano Gaetano
[email protected]
Univ. Bari, Bari, Italy
48. Solferino Viviana
[email protected]
Univ. Calabria., Cosenza, Italy
49. Stephane Pia
[email protected]
Univ. Napoli, Napoli, Italy
50. Tarantello Gabriella
[email protected]
Univ. Roma Tor Vergata, Roma, Italy (lecturer)
51. Tran Vu Khanh
[email protected]
Univ. Padova, Padova, Italy
52. Vaira Giusi
[email protected]
SISSA/ISAS, Trieste, Italy
53. Wang Xu-Jia
[email protected]
Australian National Univ., Camberra, Australia (lecturer)
54. Yang Paul
[email protected]
Princeton Univ., Princeton, NJ, USA (lecturer)
55. Yashagin Eugene
[email protected]
State University, Kazan, Russia
LIST OF C.I.M.E. SEMINARS

Published by C.I.M.E
1954 1. Analisi funzionale
2. Quadratura delle superficie e questioni connesse
3. Equazioni differenziali non lineari

1955 4. Teorema di Riemann-Roch e questioni connesse

5. Teoria dei numeri
6. Topologia
7. Teorie non linearizzate in elasticità, idrodinamica, aerodinamic
8. Geometria proiettivo-differenziale

1956 9. Equazioni alle derivate parziali a caratteristiche reali

10. Propagazione delle onde elettromagnetiche automorfe
11. Teoria della funzioni di più variabili complesse e delle funzioni

1957 12. Geometria aritmetica e algebrica (2 vol.)

13. Integrali singolari e questioni connesse
14. Teoria della turbolenza (2 vol.)

1958 15. Vedute e problemi attuali in relatività generale

16. Problemi di geometria differenziale in grande
17. Il principio di minimo e le sue applicazioni alle equazioni funzionali

1959 18. Induzione e statistica

19. Teoria algebrica dei meccanismi automatici (2 vol.)
20. Gruppi, anelli di Lie e teoria della coomologia

1960 21. Sistemi dinamici e teoremi ergodici

22. Forme differenziali e loro integrali

1961 23. Geometria del calcolo delle variazioni (2 vol.)

24. Teoria delle distribuzioni
25. Onde superficiali

1962 26. Topologia differenziale

27. Autovalori e autosoluzioni
28. Magnetofluidodinamica

1963 29. Equazioni differenziali astratte

30. Funzioni e varietà complesse
31. Proprietà di media e teoremi di confronto in Fisica Matematica

1964 32. Relatività generale

33. Dinamica dei gas rarefatti
34. Alcune questioni di analisi numerica
35. Equazioni differenziali non lineari

1965 36. Non-linear continuum theories

37. Some aspects of ring theory
38. Mathematical optimization in economics
Published by Ed. Cremonese, Firenze
1966 39. Calculus of variations
40. Economia matematica
41. Classi caratteristiche e questioni connesse
42. Some aspects of diffusion theory

1967 43. Modern questions of celestial mechanics

44. Numerical analysis of partial differential equations
45. Geometry of homogeneous bounded domains

1968 46. Controllability and observability

47. Pseudo-differential operators
48. Aspects of mathematical logic

1969 49. Potential theory

50. Non-linear continuum theories in mechanics and physics and their applications
51. Questions of algebraic varieties

1970 52. Relativistic fluid dynamics

53. Theory of group representations and Fourier analysis
54. Functional equations and inequalities
55. Problems in non-linear analysis

1971 56. Stereodynamics

57. Constructive aspects of functional analysis (2 vol.)
58. Categories and commutative algebra

1972 59. Non-linear mechanics

60. Finite geometric structures and their applications
61. Geometric measure theory and minimal surfaces

1973 62. Complex analysis

63. New variational techniques in mathematical physics
64. Spectral analysis

1974 65. Stability problems

66. Singularities of analytic spaces
67. Eigenvalues of non linear problems

1975 68. Theoretical computer sciences

69. Model theory and applications
70. Differential operators and manifolds

Published by Ed. Liguori, Napoli

1976 71. Statistical Mechanics
72. Hyperbolicity
73. Differential topology

1977 74. Materials with memory

75. Pseudodifferential operators with applications
76. Algebraic surfaces

Published by Ed. Liguori, Napoli & Birkhäuser

1978 77. Stochastic differential equations
78. Dynamical systems

1979 79. Recursion theory and computational complexity

80. Mathematics of biology

1980 81. Wave propagation

82. Harmonic analysis and group representations
83. Matroid theory and its applications
Published by Springer-Verlag
1981 84. Kinetic Theories and the Boltzmann Equation (LNM 1048)
85. Algebraic Threefolds (LNM 947)
86. Nonlinear Filtering and Stochastic Control (LNM 972)

1982 87. Invariant Theory (LNM 996)

88. Thermodynamics and Constitutive Equations (LNP 228)
89. Fluid Dynamics (LNM 1047)

1983 90. Complete Intersections (LNM 1092)

91. Bifurcation Theory and Applications (LNM 1057)
92. Numerical Methods in Fluid Dynamics (LNM 1127)

1984 93. Harmonic Mappings and Minimal Immersions (LNM 1161)

94. Schrödinger Operators (LNM 1159)
95. Buildings and the Geometry of Diagrams (LNM 1181)

1985 96. Probability and Analysis (LNM 1206)

97. Some Problems in Nonlinear Diffusion (LNM 1224)
98. Theory of Moduli (LNM 1337)

1986 99. Inverse Problems (LNM 1225)

100. Mathematical Economics (LNM 1330)
101. Combinatorial Optimization (LNM 1403)

1987 102. Relativistic Fluid Dynamics (LNM 1385)

103. Topics in Calculus of Variations (LNM 1365)

1988 104. Logic and Computer Science (LNM 1429)

105. Global Geometry and Mathematical Physics (LNM 1451)

1989 106. Methods of nonconvex analysis (LNM 1446)

107. Microlocal Analysis and Applications (LNM 1495)

1990 108. Geometric Topology: Recent Developments (LNM 1504)

109. H∞ Control Theory (LNM 1496)
110. Mathematical Modelling of Industrial Processes (LNM 1521)

1991 111. Topological Methods for Ordinary Differential Equations (LNM 1537)
112. Arithmetic Algebraic Geometry (LNM 1553)
113. Transition to Chaos in Classical and Quantum Mechanics (LNM 1589)

1992 114. Dirichlet Forms (LNM 1563)

115. D-Modules, Representation Theory, and Quantum Groups (LNM 1565)
116. Nonequilibrium Problems in Many-Particle Systems (LNM 1551)

1993 117. Integrable Systems and Quantum Groups (LNM 1620)

118. Algebraic Cycles and Hodge Theory (LNM 1594)
119. Phase Transitions and Hysteresis (LNM 1584)

1994 120. Recent Mathematical Methods in Nonlinear Wave Propagation (LNM 1640)
121. Dynamical Systems (LNM 1609)
122. Transcendental Methods in Algebraic Geometry (LNM 1646)

1995 123. Probabilistic Models for Nonlinear PDE’s (LNM 1627)

124. Viscosity Solutions and Applications (LNM 1660)
125. Vector Bundles on Curves. New Directions (LNM 1649)

1996 126. Integral Geometry, Radon Transforms and Complex Analysis (LNM 1684)
127. Calculus of Variations and Geometric Evolution Problems (LNM 1713)
128. Financial Mathematics (LNM 1656)
1997 129. Mathematics Inspired by Biology (LNM 1714)
130. Advanced Numerical Approximation of Nonlinear Hyperbolic (LNM 1697)
Equations
131. Arithmetic Theory of Elliptic Curves (LNM 1716)
132. Quantum Cohomology (LNM 1776)

1998 133. Optimal Shape Design (LNM 1740)

134. Dynamical Systems and Small Divisors (LNM 1784)
135. Mathematical Problems in Semiconductor Physics (LNM 1823)
136. Stochastic PDE’s and Kolmogorov Equations in Infinite Dimension (LNM 1715)
137. Filtration in Porous Media and Industrial Applications (LNM 1734)

1999 138. Computational Mathematics driven by Industrial Applications (LNM 1739)

139. Iwahori-Hecke Algebras and Representation Theory (LNM 1804)
140. Hamiltonian Dynamics - Theory and Applications (LNM 1861)
141. Global Theory of Minimal Surfaces in Flat Spaces (LNM 1775)
142. Direct and Inverse Methods in Solving Nonlinear Evolution (LNP 632)
Equations

2000 143. Dynamical Systems (LNM 1822)

144. Diophantine Approximation (LNM 1819)
145. Mathematical Aspects of Evolving Interfaces (LNM 1812)
146. Mathematical Methods for Protein Structure (LNCS 2666)
147. Noncommutative Geometry (LNM 1831)

2001 148. Topological Fluid Mechanics (LNM 1973)

149. Spatial Stochastic Processes (LNM 1802)
150. Optimal Transportation and Applications (LNM 1813)
151. Multiscale Problems and Methods in Numerical Simulations (LNM 1825)

2002 152. Real Methods in Complex and CR Geometry (LNM 1848)

153. Analytic Number Theory (LNM 1891)
154. Inverse Problems and Imaging (LNM 1943)

2003 155. Stochastic Methods in Finance (LNM 1856)

156. Hyperbolic Systems of Balance Laws (LNM 1911)
157. Symplectic 4-Manifolds and Algebraic Surfaces (LNM 1938)
158. Mathematical Foundation of Turbulent Viscous Flows (LNM 1871)

2004 159. Representation Theory and Complex Analysis (LNM 1931)

160. Nonlinear and Optimal Control Theory (LNM 1932)
161. Stochastic Geometry (LNM 1892)

2005 162. Enumerative Invariants in Algebraic Geometry and String Theory (LNM 1947)
163. Calculus of Variations and Non-linear Partial Differential Equations (LNM 1927)
164. SPDE in Hydrodynamic. Recent Progress and Prospects (LNM 1942)

2006 165. Pseudo-Differential Operators (LNM 1949)

166. Mixed Finite Elements, Compatibility Conditions, and Applications (LNM 1939)
167. Multiscale Problems in the Life Sciences. From Microscopic (LNM 1940)
to Macroscopic
168. Quantum Transport (LNM 1946)

2007 169. Geometric Analysis and PDEs (LNM 1977)

170. Nonlinear Optimization to appear
171. Arithmetic Geometry to appear

2008 172. Nonlinear Partial Differential Equations and Applications to appear

173. Holomorphic Dynamical Systems to appear
174. Level Set and PDE based Reconstruction Methods: to appear
Applications to Inverse Problems and Image Processing
175. Mathematical models in the manufacturing of glass, polymers to appear
and textiles
Lecture Notes in Mathematics
For information about earlier volumes
please contact your bookseller or Springer
LNM Online archive: springerlink.com

Vol. 1792: D. D. Ang, R. Gorenflo, V. K. Le, D. D. Trong, Applications. Martina Franca, Italy 2001. Editors:
Moment Theory and Some Inverse Problems in Potential L. A. Caffarelli, S. Salsa (2003)
Theory and Heat Conduction (2002) Vol. 1814: P. Bank, F. Baudoin, H. Föllmer,
Vol. 1793: J. Cortés Monforte, Geometric, Control and L.C.G. Rogers, M. Soner, N. Touzi, Paris-Princeton
Numerical Aspects of Nonholonomic Systems (2002) Lectures on Mathematical Finance 2002 (2003)
Vol. 1794: N. Pytheas Fogg, Substitution in Dynam- Vol. 1815: A. M. Vershik (Ed.), Asymptotic Combi-
ics, Arithmetics and Combinatorics. Editors: V. Berthé, natorics with Applications to Mathematical Physics.
S. Ferenczi, C. Mauduit, A. Siegel (2002) St. Petersburg, Russia 2001 (2003)
Vol. 1795: H. Li, Filtered-Graded Transfer in Using Non- Vol. 1816: S. Albeverio, W. Schachermayer,
commutative Gröbner Bases (2002) M. Talagrand, Lectures on Probability Theory and
Vol. 1796: J.M. Melenk, hp-Finite Element Methods for Statistics. Ecole d’Eté de Probabilités de Saint-Flour
Singular Perturbations (2002) XXX-2000. Editor: P. Bernard (2003)
Vol. 1797: B. Schmidt, Characters and Cyclotomic Fields Vol. 1817: E. Koelink, W. Van Assche (Eds.), Orthogonal
in Finite Geometry (2002) Polynomials and Special Functions. Leuven 2002 (2003)
Vol. 1798: W.M. Oliva, Geometric Mechanics (2002) Vol. 1818: M. Bildhauer, Convex Variational Problems
Vol. 1799: H. Pajot, Analytic Capacity, Rectifiability, with Linear, nearly Linear and/or Anisotropic Growth
Menger Curvature and the Cauchy Integral (2002) Conditions (2003)
Vol. 1800: O. Gabber, L. Ramero, Almost Ring Theory Vol. 1819: D. Masser, Yu. V. Nesterenko, H. P. Schlick-
ewei, W. M. Schmidt, M. Waldschmidt, Diophantine
(2003)
Approximation. Cetraro, Italy 2000. Editors: F. Amoroso,
Vol. 1801: J. Azéma, M. Émery, M. Ledoux, M. Yor
U. Zannier (2003)
(Eds.), Séminaire de Probabilités XXXVI (2003)
Vol. 1820: F. Hiai, H. Kosaki, Means of Hilbert Space
Vol. 1802: V. Capasso, E. Merzbach, B. G. Ivanoff, Operators (2003)
M. Dozzi, R. Dalang, T. Mountford, Topics in Spatial
Vol. 1821: S. Teufel, Adiabatic Perturbation Theory in
Stochastic Processes. Martina Franca, Italy 2001. Editor:
Quantum Dynamics (2003)
E. Merzbach (2003)
Vol. 1822: S.-N. Chow, R. Conti, R. Johnson, J. Mallet-
Vol. 1803: G. Dolzmann, Variational Methods for Crys-
Paret, R. Nussbaum, Dynamical Systems. Cetraro, Italy
talline Microstructure – Analysis and Computation 2000. Editors: J. W. Macki, P. Zecca (2003)
(2003)
Vol. 1823: A. M. Anile, W. Allegretto, C. Ringhofer,
Vol. 1804: I. Cherednik, Ya. Markov, R. Howe, G. Mathematical Problems in Semiconductor Physics.
Lusztig, Iwahori-Hecke Algebras and their Representa- Cetraro, Italy 1998. Editor: A. M. Anile (2003)
tion Theory. Martina Franca, Italy 1999. Editors: V. Bal-
Vol. 1824: J. A. Navarro González, J. B. Sancho de Salas,
doni, D. Barbasch (2003)
C ∞ – Differentiable Spaces (2003)
Vol. 1805: F. Cao, Geometric Curve Evolution and Image Vol. 1825: J. H. Bramble, A. Cohen, W. Dahmen,
Processing (2003) Multiscale Problems and Methods in Numerical Simu-
Vol. 1806: H. Broer, I. Hoveijn. G. Lunther, G. Vegter, lations, Martina Franca, Italy 2001. Editor: C. Canuto
Bifurcations in Hamiltonian Systems. Computing Singu- (2003)
larities by Gröbner Bases (2003) Vol. 1826: K. Dohmen, Improved Bonferroni Inequal-
Vol. 1807: V. D. Milman, G. Schechtman (Eds.), Geomet- ities via Abstract Tubes. Inequalities and Identities of
ric Aspects of Functional Analysis. Israel Seminar 2000- Inclusion-Exclusion Type. VIII, 113 p, 2003.
2002 (2003) Vol. 1827: K. M. Pilgrim, Combinations of Complex
Vol. 1808: W. Schindler, Measures with Symmetry Prop- Dynamical Systems. IX, 118 p, 2003.
erties (2003) Vol. 1828: D. J. Green, Gröbner Bases and the Computa-
Vol. 1809: O. Steinbach, Stability Estimates for Hybrid tion of Group Cohomology. XII, 138 p, 2003.
Coupled Domain Decomposition Methods (2003) Vol. 1829: E. Altman, B. Gaujal, A. Hordijk, Discrete-
Vol. 1810: J. Wengenroth, Derived Functors in Functional Event Control of Stochastic Networks: Multimodularity
Analysis (2003) and Regularity. XIV, 313 p, 2003.
Vol. 1811: J. Stevens, Deformations of Singularities Vol. 1830: M. I. Gil’, Operator Functions and Localiza-
(2003) tion of Spectra. XIV, 256 p, 2003.
Vol. 1812: L. Ambrosio, K. Deckelnick, G. Dziuk, Vol. 1831: A. Connes, J. Cuntz, E. Guentner, N. Higson,
M. Mimura, V. A. Solonnikov, H. M. Soner, Mathemat- J. E. Kaminker, Noncommutative Geometry, Martina
ical Aspects of Evolving Interfaces. Madeira, Funchal, Franca, Italy 2002. Editors: S. Doplicher, L. Longo
Portugal 2000. Editors: P. Colli, J. F. Rodrigues (2003) (2004)
Vol. 1813: L. Ambrosio, L. A. Caffarelli, Y. Brenier, Vol. 1832: J. Azéma, M. Émery, M. Ledoux, M. Yor
G. Buttazzo, C. Villani, Optimal Transportation and its (Eds.), Séminaire de Probabilités XXXVII (2003)
Vol. 1833: D.-Q. Jiang, M. Qian, M.-P. Qian, Mathe- Vol. 1859: E. Letellier, Fourier Transforms of Invariant
matical Theory of Nonequilibrium Steady States. On the Functions on Finite Reductive Lie Algebras (2005)
Frontier of Probability and Dynamical Systems. IX, 280 Vol. 1860: A. Borisyuk, G.B. Ermentrout, A. Friedman,
p, 2004. D. Terman, Tutorials in Mathematical Biosciences I.
Vol. 1834: Yo. Yomdin, G. Comte, Tame Geometry with Mathematical Neurosciences (2005)
Application in Smooth Analysis. VIII, 186 p, 2004. Vol. 1861: G. Benettin, J. Henrard, S. Kuksin, Hamilto-
Vol. 1835: O.T. Izhboldin, B. Kahn, N.A. Karpenko, nian Dynamics – Theory and Applications, Cetraro, Italy,
A. Vishik, Geometric Methods in the Algebraic Theory 1999. Editor: A. Giorgilli (2005)
of Quadratic Forms. Summer School, Lens, 2000. Editor: Vol. 1862: B. Helffer, F. Nier, Hypoelliptic Estimates and
J.-P. Tignol (2004) Spectral Theory for Fokker-Planck Operators and Witten
Vol. 1836: C. Nǎstǎsescu, F. Van Oystaeyen, Methods of Laplacians (2005)
Graded Rings. XIII, 304 p, 2004. Vol. 1863: H. Führ, Abstract Harmonic Analysis of Con-
Vol. 1837: S. Tavaré, O. Zeitouni, Lectures on Probabil- tinuous Wavelet Transforms (2005)
ity Theory and Statistics. Ecole d’Eté de Probabilités de Vol. 1864: K. Efstathiou, Metamorphoses of Hamiltonian
Saint-Flour XXXI-2001. Editor: J. Picard (2004) Systems with Symmetries (2005)
Vol. 1838: A.J. Ganesh, N.W. O’Connell, D.J. Wischik, Vol. 1865: D. Applebaum, B.V. R. Bhat, J. Kustermans,
Big Queues. XII, 254 p, 2004. J. M. Lindsay, Quantum Independent Increment Pro-
Vol. 1839: R. Gohm, Noncommutative Stationary Pro- cesses I. From Classical Probability to Quantum Stochas-
cesses. VIII, 170 p, 2004. tic Calculus. Editors: M. Schürmann, U. Franz (2005)
Vol. 1840: B. Tsirelson, W. Werner, Lectures on Probabil- Vol. 1866: O.E. Barndorff-Nielsen, U. Franz, R. Gohm,
ity Theory and Statistics. Ecole d’Eté de Probabilités de B. Kümmerer, S. Thorbjønsen, Quantum Independent
Saint-Flour XXXII-2002. Editor: J. Picard (2004) Increment Processes II. Structure of Quantum Lévy
Vol. 1841: W. Reichel, Uniqueness Theorems for Vari- Processes, Classical Probability, and Physics. Editors:
ational Problems by the Method of Transformation M. Schürmann, U. Franz, (2005)
Groups (2004) Vol. 1867: J. Sneyd (Ed.), Tutorials in Mathematical Bio-
Vol. 1842: T. Johnsen, A. L. Knutsen, K3 Projective Mod- sciences II. Mathematical Modeling of Calcium Dynam-
els in Scrolls (2004) ics and Signal Transduction. (2005)
Vol. 1843: B. Jefferies, Spectral Properties of Noncom- Vol. 1868: J. Jorgenson, S. Lang, Posn (R) and Eisenstein
muting Operators (2004) Series. (2005)
Vol. 1844: K.F. Siburg, The Principle of Least Action in Vol. 1869: A. Dembo, T. Funaki, Lectures on Probabil-
Geometry and Dynamics (2004) ity Theory and Statistics. Ecole d’Eté de Probabilités de
Vol. 1845: Min Ho Lee, Mixed Automorphic Forms, Saint-Flour XXXIII-2003. Editor: J. Picard (2005)
Torus Bundles, and Jacobi Forms (2004)
Vol. 1870: V.I. Gurariy, W. Lusky, Geometry of Müntz
Vol. 1846: H. Ammari, H. Kang, Reconstruction of Small
Spaces and Related Questions. (2005)
Inhomogeneities from Boundary Measurements (2004)
Vol. 1871: P. Constantin, G. Gallavotti, A.V. Kazhikhov,
Vol. 1847: T.R. Bielecki, T. Björk, M. Jeanblanc,
Y. Meyer, S. Ukai, Mathematical Foundation of Turbu-
M. Rutkowski, J.A. Scheinkman, W. Xiong, Paris-
lent Viscous Flows, Martina Franca, Italy, 2003. Editors:
Princeton Lectures on Mathematical Finance 2003 (2004)
M. Cannone, T. Miyakawa (2006)
Vol. 1848: M. Abate, J. E. Fornaess, X. Huang,
Vol. 1872: A. Friedman (Ed.), Tutorials in Mathemati-
J. P. Rosay, A. Tumanov, Real Methods in Complex
cal Biosciences III. Cell Cycle, Proliferation, and Cancer
and CR Geometry, Martina Franca, Italy 2002. Editors:
(2006)
D. Zaitsev, G. Zampieri (2004)
Vol. 1873: R. Mansuy, M. Yor, Random Times and En-
Vol. 1849: Martin L. Brown, Heegner Modules and Ellip-
tic Curves (2004) largements of Filtrations in a Brownian Setting (2006)
Vol. 1850: V. D. Milman, G. Schechtman (Eds.), Geomet- Vol. 1874: M. Yor, M. Émery (Eds.), In Memoriam Paul-
ric Aspects of Functional Analysis. Israel Seminar 2002- André Meyer - Séminaire de Probabilités XXXIX (2006)
2003 (2004) Vol. 1875: J. Pitman, Combinatorial Stochastic Processes.
Vol. 1851: O. Catoni, Statistical Learning Theory and Ecole d’Eté de Probabilités de Saint-Flour XXXII-2002.
Stochastic Optimization (2004) Editor: J. Picard (2006)
Vol. 1852: A.S. Kechris, B.D. Miller, Topics in Orbit Vol. 1876: H. Herrlich, Axiom of Choice (2006)
Equivalence (2004) Vol. 1877: J. Steuding, Value Distributions of L-Functions
Vol. 1853: Ch. Favre, M. Jonsson, The Valuative Tree (2007)
(2004) Vol. 1878: R. Cerf, The Wulff Crystal in Ising and Percol-
Vol. 1854: O. Saeki, Topology of Singular Fibers of Dif- ation Models, Ecole d’Eté de Probabilités de Saint-Flour
ferential Maps (2004) XXXIV-2004. Editor: Jean Picard (2006)
Vol. 1855: G. Da Prato, P.C. Kunstmann, I. Lasiecka, Vol. 1879: G. Slade, The Lace Expansion and its Applica-
A. Lunardi, R. Schnaubelt, L. Weis, Functional Analytic tions, Ecole d’Eté de Probabilités de Saint-Flour XXXIV-
Methods for Evolution Equations. Editors: M. Iannelli, 2004. Editor: Jean Picard (2006)
R. Nagel, S. Piazzera (2004) Vol. 1880: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
Vol. 1856: K. Back, T.R. Bielecki, C. Hipp, S. Peng, Systems I, The Hamiltonian Approach (2006)
W. Schachermayer, Stochastic Methods in Finance, Vol. 1881: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
Bressanone/Brixen, Italy, 2003. Editors: M. Fritelli, Systems II, The Markovian Approach (2006)
W. Runggaldier (2004) Vol. 1882: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
Vol. 1857: M. Émery, M. Ledoux, M. Yor (Eds.), Sémi- Systems III, Recent Developments (2006)
naire de Probabilités XXXVIII (2005) Vol. 1883: W. Van Assche, F. Marcellàn (Eds.), Orthogo-
Vol. 1858: A.S. Cherny, H.-J. Engelbert, Singular nal Polynomials and Special Functions, Computation and
Stochastic Differential Equations (2005) Application (2006)
Vol. 1884: N. Hayashi, E.I. Kaikina, P.I. Naumkin, Vol. 1911: A. Bressan, D. Serre, M. Williams,
I.A. Shishmarev, Asymptotics for Dissipative Nonlinear K. Zumbrun, Hyperbolic Systems of Balance Laws.
Equations (2006) Cetraro, Italy 2003. Editor: P. Marcati (2007)
Vol. 1885: A. Telcs, The Art of Random Walks (2006) Vol. 1912: V. Berinde, Iterative Approximation of Fixed
Vol. 1886: S. Takamura, Splitting Deformations of Dege- Points (2007)
nerations of Complex Curves (2006) Vol. 1913: J.E. Marsden, G. Misiołek, J.-P. Ortega, M.
Vol. 1887: K. Habermann, L. Habermann, Introduction to Perlmutter, T.S. Ratiu, Hamiltonian Reduction by Stages
Symplectic Dirac Operators (2006) (2007)
Vol. 1888: J. van der Hoeven, Transseries and Real Dif- Vol. 1914: G. Kutyniok, Affine Density in Wavelet
ferential Algebra (2006) Analysis (2007)
Vol. 1889: G. Osipenko, Dynamical Systems, Graphs, and Vol. 1915: T. Bıyıkoǧlu, J. Leydold, P.F. Stadler,
Algorithms (2006) Laplacian Eigenvectors of Graphs. Perron-Frobenius and
Vol. 1890: M. Bunge, J. Funk, Singular Coverings of Faber-Krahn Type Theorems (2007)
Toposes (2006) Vol. 1916: C. Villani, F. Rezakhanlou, Entropy Methods
Vol. 1891: J.B. Friedlander, D.R. Heath-Brown, for the Boltzmann Equation. Editors: F. Golse, S. Olla
H. Iwaniec, J. Kaczorowski, Analytic Number Theory, (2008)
Cetraro, Italy, 2002. Editors: A. Perelli, C. Viola (2006) Vol. 1917: I. Veselić, Existence and Regularity Prop-
Vol. 1892: A. Baddeley, I. Bárány, R. Schneider, W. Weil, erties of the Integrated Density of States of Random
Stochastic Geometry, Martina Franca, Italy, 2004. Editor: Schrödinger (2008)
W. Weil (2007) Vol. 1918: B. Roberts, R. Schmidt, Local Newforms for
Vol. 1893: H. Hanßmann, Local and Semi-Local Bifur- GSp(4) (2007)
cations in Hamiltonian Dynamical Systems, Results and Vol. 1919: R.A. Carmona, I. Ekeland, A. Kohatsu-
Examples (2007) Higa, J.-M. Lasry, P.-L. Lions, H. Pham, E. Taflin,
Vol. 1894: C.W. Groetsch, Stable Approximate Evalua- Paris-Princeton Lectures on Mathematical Finance 2004.
tion of Unbounded Operators (2007) Editors: R.A. Carmona, E. Çinlar, I. Ekeland, E. Jouini,
Vol. 1895: L. Molnár, Selected Preserver Problems on J.A. Scheinkman, N. Touzi (2007)
Algebraic Structures of Linear Operators and on Function Vol. 1920: S.N. Evans, Probability and Real Trees. Ecole
Spaces (2007) d’Été de Probabilités de Saint-Flour XXXV-2005 (2008)
Vol. 1896: P. Massart, Concentration Inequalities and Vol. 1921: J.P. Tian, Evolution Algebras and their Appli-
Model Selection, Ecole d’Été de Probabilités de Saint- cations (2008)
Flour XXXIII-2003. Editor: J. Picard (2007) Vol. 1922: A. Friedman (Ed.), Tutorials in Mathematical
Vol. 1897: R. Doney, Fluctuation Theory for Lévy BioSciences IV. Evolution and Ecology (2008)
Processes, Ecole d’Été de Probabilités de Saint-Flour Vol. 1923: J.P.N. Bishwal, Parameter Estimation in
XXXV-2005. Editor: J. Picard (2007) Stochastic Differential Equations (2008)
Vol. 1898: H.R. Beyer, Beyond Partial Differential Equa- Vol. 1924: M. Wilson, Littlewood-Paley Theory and
tions, On linear and Quasi-Linear Abstract Hyperbolic Exponential-Square Integrability (2008)
Evolution Equations (2007) Vol. 1925: M. du Sautoy, L. Woodward, Zeta Functions
Vol. 1899: Séminaire de Probabilités XL. Editors: of Groups and Rings (2008)
C. Donati-Martin, M. Émery, A. Rouault, C. Stricker Vol. 1926: L. Barreira, V. Claudia, Stability of Nonauto-
(2007) nomous Differential Equations (2008)
Vol. 1900: E. Bolthausen, A. Bovier (Eds.), Spin Glasses Vol. 1927: L. Ambrosio, L. Caffarelli, M.G. Crandall,
(2007) L.C. Evans, N. Fusco, Calculus of Variations and Non-
Vol. 1901: O. Wittenberg, Intersections de deux Linear Partial Differential Equations. Cetraro, Italy 2005.
quadriques et pinceaux de courbes de genre 1, Intersec- Editors: B. Dacorogna, P. Marcellini (2008)
tions of Two Quadrics and Pencils of Curves of Genus 1 Vol. 1928: J. Jonsson, Simplicial Complexes of Graphs
(2007) (2008)
Vol. 1902: A. Isaev, Lectures on the Automorphism Vol. 1929: Y. Mishura, Stochastic Calculus for Fractional
Groups of Kobayashi-Hyperbolic Manifolds (2007) Brownian Motion and Related Processes (2008)
Vol. 1903: G. Kresin, V. Maz’ya, Sharp Real-Part Theo- Vol. 1930: J.M. Urbano, The Method of Intrinsic Scaling.
rems (2007) A Systematic Approach to Regularity for Degenerate and
Vol. 1904: P. Giesl, Construction of Global Lyapunov Singular PDEs (2008)
Functions Using Radial Basis Functions (2007) Vol. 1931: M. Cowling, E. Frenkel, M. Kashiwara,
Vol. 1905: C. Prévôt, M. Röckner, A Concise Course on A. Valette, D.A. Vogan, Jr., N.R. Wallach, Representation
Stochastic Partial Differential Equations (2007) Theory and Complex Analysis. Venice, Italy 2004. Edi-
Vol. 1906: T. Schuster, The Method of Approximate tors: E.C. Tarabusi, A. D’Agnolo, M. Picardello (2008)
Inverse: Theory and Applications (2007) Vol. 1932: A.A. Agrachev, A.S. Morse, E.D. Sontag,
Vol. 1907: M. Rasmussen, Attractivity and Bifurcation H.J. Sussmann, V.I. Utkin, Nonlinear and Optimal
for Nonautonomous Dynamical Systems (2007) Control Theory. Cetraro, Italy 2004. Editors: P. Nistri,
Vol. 1908: T.J. Lyons, M. Caruana, T. Lévy, Differential G. Stefani (2008)
Equations Driven by Rough Paths, Ecole d’Été de Proba- Vol. 1933: M. Petkovic, Point Estimation of Root Finding
bilités de Saint-Flour XXXIV-2004 (2007) Methods (2008)
Vol. 1909: H. Akiyoshi, M. Sakuma, M. Wada, Vol. 1934: C. Donati-Martin, M. Émery, A. Rouault,
Y. Yamashita, Punctured Torus Groups and 2-Bridge C. Stricker (Eds.), Séminaire de Probabilités XLI (2008)
Knot Groups (I) (2007) Vol. 1935: A. Unterberger, Alternative Pseudodifferential
Vol. 1910: V.D. Milman, G. Schechtman (Eds.), Geo- Analysis (2008)
metric Aspects of Functional Analysis. Israel Seminar Vol. 1936: P. Magal, S. Ruan (Eds.), Structured Popula-
2004-2005 (2007) tion Models in Biology and Epidemiology (2008)
Vol. 1937: G. Capriz, P. Giovine, P.M. Mariano (Eds.), Vol. 1960: J. Lipman, M. Hashimoto, Foundations of
Mathematical Models of Granular Matter (2008) Grothendieck Duality for Diagrams of Schemes (2009)
Vol. 1938: D. Auroux, F. Catanese, M. Manetti, P. Seidel, Vol. 1961: G. Buttazzo, A. Pratelli, S. Solimini,
B. Siebert, I. Smith, G. Tian, Symplectic 4-Manifolds E. Stepanov, Optimal Urban Networks via Mass Trans-
and Algebraic Surfaces. Cetraro, Italy 2003. Editors: portation (2009)
F. Catanese, G. Tian (2008) Vol. 1962: R. Dalang, D. Khoshnevisan, C. Mueller,
Vol. 1939: D. Boffi, F. Brezzi, L. Demkowicz, R.G. D. Nualart, Y. Xiao, A Minicourse on Stochastic Partial
Durán, R.S. Falk, M. Fortin, Mixed Finite Elements, Differential Equations (2009)
Compatibility Conditions, and Applications. Cetraro, Vol. 1963: W. Siegert, Local Lyapunov Exponents (2009)
Italy 2006. Editors: D. Boffi, L. Gastaldi (2008) Vol. 1964: W. Roth, Operator-valued Measures and
Vol. 1940: J. Banasiak, V. Capasso, M.A.J. Chaplain, Integrals for Cone-valued Functions and Integrals for
M. Lachowicz, J. Miȩkisz, Multiscale Problems in Conevalued Functions (2009)
the Life Sciences. From Microscopic to Macro- Vol. 1965: C. Chidume, Geometric Properties of Banach
scopic. Bȩdlewo, Poland 2006. Editors: V. Capasso, Spaces and Nonlinear Iterations (2009)
M. Lachowicz (2008) Vol. 1966: D. Deng, Y. Han, Harmonic Analysis on
Vol. 1941: S.M.J. Haran, Arithmetical Investigations. Spaces of Homogeneous Type (2009)
Representation Theory, Orthogonal Polynomials, and Vol. 1967: B. Fresse,Modules over Operads and Functors
Quantum Interpolations (2008) (2009)
Vol. 1942: S. Albeverio, F. Flandoli, Y.G. Sinai, SPDE in Vol. 1968: R. Weissauern, Endoscopy for GSP(4) and the
Hydrodynamic. Recent Progress and Prospects. Cetraro, Cohomology of Siegel Modular Threefolds (2009)
Italy 2005. Editors: G. Da Prato, M. Röckner (2008) Vol. 1969: B. Roynette, M. Yor, Penalising Brownian
Vol. 1943: L.L. Bonilla (Ed.), Inverse Problems and Imag- Paths (2009)
ing. Martina Franca, Italy 2002 (2008) Vol. 1970: M. Biskup, A. Bovier, F. den Hollander,
Vol. 1944: A. Di Bartolo, G. Falcone, P. Plaumann, D. Ioffe, F. Martinelli, K. Netočný, F. Toninelli, Methods
K. Strambach, Algebraic Groups and Lie Groups with of Contemporary Mathematical Statistical Physics.
Few Factors (2008) Editor: R. Kotecký (2009)
Vol. 1945: F. Brauer, P. van den Driessche, J. Wu (Eds.), Vol. 1971: L. Saint-Raymond, Hydrodynamic Limits of
Mathematical Epidemiology (2008) the Boltzmann Equation (2009)
Vol. 1946: G. Allaire, A. Arnold, P. Degond, T.Y. Hou, Vol. 1972: T. Mochizuki, Donaldson Type Invariants for
Quantum Transport. Modelling, Analysis and Asymp- Algebraic Surfaces (2009)
totics. Cetraro, Italy 2006. Editors: N.B. Abdallah, Vol. 1973: M.A. Berger, L.H. Kauffmann, B. Khesin,
G. Frosali (2008) H.K. Moffatt, R.L. Ricca, De W. Sumners, Lectures
Vol. 1947: D. Abramovich, M. Mariño, M. Thaddeus, on Topological Fluid Mechanics. Centaro, Italy 2001.
R. Vakil, Enumerative Invariants in Algebraic Geo- Editor: R.L. Ricca (2009)
metry and String Theory. Cetraro, Italy 2005. Editors: Vol. 1974: F. den Hollander, Random Polymers: École
K. Behrend, M. Manetti (2008) d’Été de Probabilités de Saint-Flour XXXVII – 2007
Vol. 1948: F. Cao, J-L. Lisani, J-M. Morel, P. Musé, (2009)
F. Sur, A Theory of Shape Identification (2008) Vol. 1975: J.C. Rohde, Cyclic Coverings, Calabi-Yau
Vol. 1949: H.G. Feichtinger, B. Helffer, M.P. Lamoureux, Manifolds and Complex Multiplication (2009)
N. Lerner, J. Toft, Pseudo-Differential Operators. Quan- Vol. 1976: N. Ginoux, The Dirac Spectrum (2009)
tization and Signals. Cetraro, Italy 2006. Editors: Vol. 1977: M.J. Gursky, E. Lanconelli, A. Malchiodi,
L. Rodino, M.W. Wong (2008) G. Tarantello, X.-J. Wang, P.C. Yang, Geometric Analysis
Vol. 1950: M. Bramson, Stability of Queueing Networks, and PDEs. Centaro, Italy 2007. Editors: A. Ambrosetti,
Ecole d’ Eté de Probabilités de Saint-Flour XXXVI-2006 S.-Y.A. Chang, A. Malchiodi (2009)
(2008)
Vol. 1951: A. Moltó, J. Orihuela, S. Troyanski, Recent Reprints and New Editions
M. Valdivia, A Non Linear Transfer Technique for
Renorming (2009) Vol. 1702: J. Ma, J. Yong, Forward-Backward Stochas-
Vol. 1952: R. Mikhailov, I.B.S. Passi, Lower Central and tic Differential Equations and their Applications. 1999 –
Dimension Series of Groups (2009) Corr. 3rd printing (2007)
Vol. 1953: K. Arwini, C.T.J. Dodson, Information Geom- Vol. 830: J.A. Green, Polynomial Representations of
etry (2008) GLn , with an Appendix on Schensted Correspondence
Vol. 1954: P. Biane, L. Bouten, F. Cipriani, N. Konno, and Littelmann Paths by K. Erdmann, J.A. Green and
N. Privault, Q. Xu, Quantum Potential Theory. Editors: M. Schoker 1980 – 2nd corr. and augmented edition
U. Franz, M. Schuermann (2008) (2007)
Vol. 1955: M. Bernot, V. Caselles, J.-M. Morel, Optimal Vol. 1693: S. Simons, From Hahn-Banach to Monotonic-
Transportation Networks (2008) ity (Minimax and Monotonicity 1998) – 2nd exp. edition
Vol. 1956: C.H. Chu, Matrix Convolution Operators on (2008)
Groups (2008) Vol. 470: R.E. Bowen, Equilibrium States and the Ergodic
Vol. 1957: A. Guionnet, On Random Matrices: Macro- Theory of Anosov Diffeomorphisms. With a preface by
scopic Asymptotics, Ecole dŠEté de Probabilités de D. Ruelle. Edited by J.-R. Chazottes. 1975 – 2nd rev.
Saint-Flour XXXVI-2006 (2009) edition (2008)
Vol. 1958: M.C. Olsson, Compactifying Moduli Spaces Vol. 523: S.A. Albeverio, R.J. Høegh-Krohn, S. Maz-
for Abelian Varieties (2008) zucchi, Mathematical Theory of Feynman Path Integral.
Vol. 1959: Y. Nakkajima, A. Shiho, Weight Filtrations 1976 – 2nd corr. and enlarged edition (2008)
on Log Crystalline Cohomologies of Families of Open Vol. 1764: A. Cannas da Silva, Lectures on Symplectic
Smooth Varieties (2008) Geometry 2001 – Corr. 2nd printing (2008)
LECTURE NOTES IN MATHEMATICS 123
Edited by J.-M. Morel, F. Takens, B. Teissier, P.K. Maini

Editorial Policy (for Multi-Author Publications: Summer Schools/Intensive Courses)

1. Lecture Notes aim to report new developments in all areas of mathematics and their
applications - quickly, informally and at a high level. Mathematical texts analysing new
developments in modelling and numerical simulation are welcome. Manuscripts should be
reasonably self-contained and rounded off. Thus they may, and often will, present not only
results of the author but also related work by other people. They should provide sufficient
motivation, examples and applications. There should also be an introduction making the
text comprehensible to a wider audience. This clearly distinguishes Lecture Notes from
journal articles or technical reports which normally are very concise. Articles intended for
a journal but too long to be accepted by most journals, usually do not have this “lecture
notes” character.
2. In general SUMMER SCHOOLS and other similar INTENSIVE COURSES are held to
present mathematical topics that are close to the frontiers of recent research to an audience
at the beginning or intermediate graduate level, who may want to continue with this area of
work, for a thesis or later. This makes demands on the didactic aspects of the presentation.
Because the subjects of such schools are advanced, there often exists no textbook, and so
ideally, the publication resulting from such a school could be a first approximation to such
a textbook. Usually several authors are involved in the writing, so it is not always simple
to obtain a unified approach to the presentation.
For prospective publication in LNM, the resulting manuscript should not be just a col-
lection of course notes, each of which has been developed by an individual author with
little or no co-ordination with the others, and with little or no common concept. The sub-
ject matter should dictate the structure of the book, and the authorship of each part or
chapter should take secondary importance. Of course the choice of authors is crucial to the
quality of the material at the school and in the book, and the intention here is not to belittle
their impact, but simply to say that the book should be planned to be written by these
authors jointly, and not just assembled as a result of what these authors happen to submit.
This represents considerable preparatory work (as it is imperative to ensure that the
authors know these criteria before they invest work on a manuscript), and also consider-
able editing work afterwards, to get the book into final shape. Still it is the form that holds
the most promise of a successful book that will be used by its intended audience, rather
than yet another volume of proceedings for the library shelf.
3. Manuscripts should be submitted either online at www.editorialmanager.com/lnm/ to
Springer’s mathematics editorial, or to one of the series editors. Volume editors are
expected to arrange for the refereeing, to the usual scientific standards, of the individual
contributions. If the resulting reports can be forwarded to us (series editors or Springer)
this is very helpful. If no reports are forwarded or if other questions remain unclear in
respect of homogeneity etc, the series editors may wish to consult external referees for an
overall evaluation of the volume. A final decision to publish can be made only on the basis
of the complete manuscript; however a preliminary decision can be based on a pre-final
or incomplete manuscript. The strict minimum amount of material that will be considered
should include a detailed outline describing the planned contents of each chapter.
Volume editors and authors should be aware that incomplete or insufficiently close to
final manuscripts almost always result in longer evaluation times. They should also be
aware that parallel submission of their manuscript to another publisher while under con-
sideration for LNM will in general lead to immediate rejection.
4. Manuscripts should in general be submitted in English. Final manuscripts should contain
at least 100 pages of mathematical text and should always include
– a general table of contents;
– an informative introduction, with adequate motivation and perhaps some historical
remarks: it should be accessible to a reader not intimately familiar with the topic
treated;
– a global subject index: as a rule this is genuinely helpful for the reader.
Lecture Notes volumes are, as a rule, printed digitally from the authors’ files. We strongly
recommend that all contributions in a volume be written in the same LaTeX version,
preferably LaTeX2e. To ensure best results, authors are asked to use the LaTeX2e style
files available from Springer’s web-server at
ftp://ftp.springer.de/pub/tex/latex/svmonot1/ (for monographs) and
ftp://ftp.springer.de/pub/tex/latex/svmultt1/ (for summer schools/tutorials).
Additional technical instructions are available on request from: [email protected].
5. Careful preparation of the manuscripts will help keep production time short besides ensur-
ing satisfactory appearance of the finished book in print and online. After acceptance of
the manuscript authors will be asked to prepare the final LaTeX source files and also the
corresponding dvi-, pdf- or zipped ps-file. The LaTeX source files are essential for pro-
ducing the full-text online version of the book. For the existing online volumes of LNM
see: http://www.springerlink.com/openurl.asp?genre=journal&issn=0075-8434.
The actual production of a Lecture Notes volume takes approximately 12 weeks.
6. Volume editors receive a total of 50 free copies of their volume to be shared with the au-
thors, but no royalties. They and the authors are entitled to a discount of 33.3% on the price
of Springer books purchased for their personal use, if ordering directly from Springer.
7. Commitment to publish is made by letter of intent rather than by signing a formal contract.
Springer-Verlag secures the copyright for each volume. Authors are free to reuse material
contained in their LNM volumes in later publications: a brief written (or e-mail) request
for formal permission is sufficient.
Addresses:
Professor J.-M. Morel, CMLA, Professor B. Teissier,
École Normale Supérieure de Cachan, Institut Mathématique de Jussieu,
61 Avenue du Président Wilson, UMR 7586 du CNRS,
94235 Cachan Cedex, France Équipe “Géométrie et Dynamique”,
E-mail: [email protected] 175 rue du Chevaleret,
Professor F. Takens, Mathematisch Instituut, 75013 Paris, France
Rijksuniversiteit Groningen, Postbus 800, E-mail: [email protected]
9700 AV Groningen, The Netherlands
E-mail: [email protected]
For the “Mathematical Biosciences Subseries” of LNM:
Professor P.K. Maini, Center for Mathematical Biology,
Mathematical Institute, 24-29 St Giles,
Oxford OX1 3LP, UK
E-mail: [email protected]
Springer, Mathematics Editorial I, Tiergartenstr. 17,
69121 Heidelberg, Germany,
Tel.: +49 (6221) 487-8259
Fax: +49 (6221) 4876-8259
E-mail: [email protected]