Complex Geometry Analysis Variables

CONTEMPORARY
MATHEMATICS
550
Geometric Analysis
of Several Complex Variables
and Related Topics
Marrakesh Workshop
May 10 –14, 2010
Marrakesh, Morocco
Y. Barkatou
S. Berhanu
A. Meziani
R. Meziani
N. Mir
Editors
American Mathematical Society

Geometric Analysis
and Related Topics
This page intentionally left blank
CONTEMPORARY
MATHEMATICS
550
Geometric Analysis
and Related Topics
Marrakesh Workshop
May 10 –14, 2010
Marrakesh, Morocco
Y. Barkatou
S. Berhanu
A. Meziani
R. Meziani
N. Mir
Editors

Providence, Rhode Island
Editorial Board
Dennis DeTurck, managing editor
George Andrews Abel Klein Martin J. Strauss
2010 Mathematics Subject Classification. Primary 32L05, 32Q99, 32V20, 32W05, 35A07,
35B20, 35B65, 35F05, 35F15.
Library of Congress Cataloging-in-Publication Data

Marrakesh Workshop on Geometric Analysis of Several Complex Variables and Related Topics
(2010 ; Marrakesh, Morocco)
Geometric analysis of several complex variables and related topics : Marrakesh Workshop on
Geometric Analysis of Several Complex Variables and Related Topics, May 10–14, 2010, Mar-
rakesh, Morocco / Y. Barkatou . . . [et al.], editors.
p. cm. — (Contemporary mathematics ; v. 550)
Includes bibliographical references and index.
ISBN 978-0-8218-5257-6 (alk. paper)
1. Functions of several complex variables—Congresses. 2. Complex manifolds—Congresses.
3. Analytic spaces—Congresses. I. Barkatou, Y., 1967– II. Title. III. Series.
QA331.7.M357 2010
515.94—dc23
2011014591
Copying and reprinting. Material in this book may be reproduced by any means for edu-
cational and scientific purposes without fee or permission with the exception of reproduction by
services that collect fees for delivery of documents and provided that the customary acknowledg-
ment of the source is given. This consent does not extend to other kinds of copying for general
distribution, for advertising or promotional purposes, or for resale. Requests for permission for
commercial use of material should be addressed to the Acquisitions Department, American Math-
ematical Society, 201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can
also be made by e-mail to reprint-permission@ams.org.
Excluded from these provisions is material in articles for which the author holds copyright. In
such cases, requests for permission to use or reprint should be addressed directly to the author(s).
(Copyright ownership is indicated in the notice in the lower right-hand corner of the first page of
each article.)

c 2011 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Copyright of individual articles may revert to the public domain 28 years
after publication. Contact the AMS for copyright status of individual articles.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 16 15 14 13 12 11
Contents
Preface vii
Analytic Vectors in Locally Integrable Structures
Rafael F. Barostichi, Paulo D. Cordaro,
and Gerson Petronilho 1
Subellipticity and Maximal Hypoellipticity for Two Complex Vector Fields in
(2 + 2)-Variables
Makhlouf Derridj and Bernard Helffer 15
Existence of Trace for Solutions of Locally Integrable Systems of Vector Fields
J. Hounie and E. R. da Silva 57
Chern-Moser Operators and Weighted Jet Determination Problems
Martin Kolář and Francine Meylan 75
Jet Embeddability of Local Automorphism Groups of Real-Analytic CR
Manifolds
Bernhard Lamel 89
Splitting of Holomorphic Cocycles with Estimates: Several Variables
Jürgen Leiterer 109
A Gysin Sequence for Manifolds with R-action
Gerardo A. Mendoza 139
A Potential Theoretic Characterization of Compactness of the ∂-Neumann
Problem
Sönmez Şahutoğlu 155
Duality between Harmonic and Bergman Spaces
Mei-Chi Shaw 161
On the Solvability and Hypoellipticity of Complex Vector Fields
François Treves 173
v
Preface
This volume consists of a collection of papers dealing with several complex vari-
ables, partial differential equations, and their interactions. Some of the papers are
expanded versions of the lectures given during the Workshop on Geometric Analy-
sis of Several Complex Variables and Related Topics that was held in Marrakesh,
Morocco, May 10-14, 2010.
Acknowledgment. We are very grateful and would like to thank the following
agencies for their financial support of the workshop.
• Agence Nationale de la Recherche (ANR), Projet ”Resonance”
• Ambassade de France au Maroc à Rabat
• Centre National de la Recherche Scientifique, France (CNRS)
• Centre National pour la Recherche Scientifique et Technique, Morocco (CNRST)
• Direction Générale de la Recherche Scientifique et du Développement Tech-
nologique, Algeria (DG-RSDT)
• International Mathematical Union (IMU)
• International Center for Theoretical Physics (ICTP)
• Laboratoire de Mathématiques et Applications, Université de Poitiers, France
• Laboratoire de Mathématiques Raphaël Salem, Université de Rouen, France
• National Science Foundation, USA, (NSF-OISE 1019538)
• Université Ibn Tofail, Kenitra, Morocco
Y. Barkatou
S. Berhanu
A. Meziani
R. Meziani
N. Mir
vii
Contemporary Mathematics
Volume 550, 2011
Analytic vectors in Locally Integrable Structures
Rafael F. Barostichi, Paulo D. Cordaro, and Gerson Petronilho
Abstract. In this note we introduce the sheaves of Gevrey vectors on a

smooth manifold endowed with a locally integrable structure. We discuss
the important case when the structure is hypocomplex and show that, in the
real-analytic category, if the structure is analytic hypoelliptic then every ana-
lytic vector associated to it is a real-analytic function. We also discuss some
related questions concerning the global situation as well as the regularity of
the s-Gevrey vectors associated to the structure when s > 1.
1. Introduction
Let P = P (x, D) be an analytic linear partial differential operator, of order
m ≥ 1, defined in an open subset Ω of RN , and let also s ≥ 1. A distribution
u defined in Ω is called an s-Gevrey vector for P (or an analytic vector for P
when s = 1), if P k u ∈ L2loc (Ω) for every k = 0, 1, . . . and, moreover, the estimates
P k uL2 (K) ≤ CK k! , k = 0, 1, 2, . . ., hold for each compact set K ⊂ Ω.
k+1 ms
A similar concept can be introduced for systems of operators.

During the period 1970–1990 the problem of obtaining the precise Gevrey reg-
ularity of the s-Gevrey vectors for a given operator or system has attracted the
attention of a good number of specialists in the field [BG], [BM], [BCM], [Da],
[DaH], [G], [HM], [M], [N]. See also the survey [BCR] and the references therein.
After the development of the theory of locally integrable and hypo-analytic
structures (cf. [T], [BCH]), new classes of systems of complex vector fields became
relevant, and then it is natural to analyze to what extent the known regularity
results for Gevrey vectors remain valid in this new situation. A first attempt was
done in [CCP], where the authors study such questions for the so-called locally
integrable structures of tube type.
One of the main purposes of this note is to present the notion of Gevrey vectors
in an arbitrary locally integrable structure. This, of course, requires that such
objects be defined intrinsically, a fact that is not completely obvious. Recall that a
locally integrable structure on a smooth manifold Ω is the datum of a subbundle V
of the complexified tangent bundle to Ω whose orthogonal bundle V ⊥ , which is now
a subbundle of the complexified cotangent bundle to Ω, is locally spanned by the
1991 Mathematics Subject Classification. Primary: 35B65; Secondary: 35F50.

Key words and phrases. Analytic vectors, locally integrable structures.
The second and third authors were partially supported by CNPq and Fapesp.
2011
c American
c Mathematical
0000 (copyright Society
holder)
1
2 RAFAEL F. BAROSTICHI, PAULO D. CORDARO, AND GERSON PETRONILHO
differential of smooth functions. In the bundle T(V) = C ⊗ T∗ Ω/V ⊥ we introduce

a hermitian metric and from it, inspired by the standard construction of the Fock
spaces in quantum mechanics [RS, p.53], we build a certain sequence of operators
induced by the exterior derivative on Ω, and the Gevrey vectors for V are then first
introduced as the Gevrey vectors associated to such operators. They define sheaves
over Ω, and next we show that such sheaves indeed do not depend on the choice of
the hermitian metric we started with.
All the preceding discussion is explained in sections 2 and 3. In Section 4
we turn our attention to the analytic vectors associated to V and obtain an easy
representation for them (Theorem 4.1), which is applied in Section 5 to prove their
regularity in the case when the structure V is hypocomplex (Theorem 5.1). Next,
in Section 6, we derive the regularity of the analytic vectors when Ω and V are
real-analytic: we prove that whenever V is analytic hypoelliptic then every analytic
vector for V is a real-analytic function, which gives, in this set up, the analogous
result as in the case 1 of principal type, analytic linear partial differential operators
[BM]. Finally we conclude the work by remarking that both the global version of
this result and its extension for s-Gevrey vectors when s > 1 are no longer true.
These results certainly open the doors for future investigation, to which we hope
to return.
2. Preliminaries
Let Ω be a smooth, paracompact manifold of dimension n + m endowed with
a locally integrable structure V of rank n. Thus V is a vector subbundle of CTΩ of
rank n whose orthogonal bundle V ⊥ ⊂ CT∗ Ω is locally spanned by the differential
of m smooth functions. We denote by T(V) the vector bundle CT∗ Ω/V ⊥ . Such
bundle T(V) has rank n and the exterior derivative induces a first order operator
d : C ∞ (Ω ) −→ C ∞ (Ω ; T(V)), Ω ⊂ Ω open,
through the composition
d
C ∞ (Ω ) −→ C ∞ (Ω ; CT∗ Ω) −→ C ∞ (Ω ; T(V)),
where d stands for the exterior derivative acting on scalar functions, and the last
arrow is induced by the projection map.
We recall that if E is an arbitrary vector bundle over Ω then the operator d
induces an operator
d ⊗ I : C ∞ (Ω ; E) −→ C ∞ (Ω ; T(V) ⊗ E), Ω ⊂ Ω open.
If we set, for N = 1, 2, . . .,
TN (V) = T(V) ⊗ · · · ⊗ T(V),

N-times
taking E = TN (V), gives, for each N = 1, 2, . . ., an operator
DN : C ∞ (Ω ; TN (V)) −→ C ∞ (Ω ; TN +1 (V)), Ω ⊂ Ω open.
1In general, analytic vectors for analytic hypoelliptic operators are not real-analytic functions
([G], [BCR]).
ANALYTIC VECTORS IN LOCALLY INTEGRABLE STRUCTURES 3
For completeness we shall also write D0 = d and T0 (V) = C. Finally, we shall also
set D(0) = identity and, for N ≥ 1,
D(N ) : C ∞ (Ω ) −→ C ∞ (Ω ; TN (V)), D(N ) = DN −1 ◦ · · · D1 ◦ D0 .
We now assume that T(V) is endowed with a smooth hermitian metric h. Such
hermitian metric induces a smooth hermitian metric hN on each of the bundles
TN (V) (h1 = h). From this we can define, for u ∈ C(Ω , TN (V)), and K ⊂ Ω
compact, the norms
1/2
uK,h = sup {hN (u(A), u(A))} .
A∈K
Definition 2.1. Let Ω be a smooth manifold over which a locally integrable

structure is defined. Assume that T(V) is endowed with an hermitian metric h.
Let also s ≥ 1 and Ω ⊂ Ω open. We shall denote by Gsh (Ω ; V) the space of all
u ∈ C(Ω ) such that D(N ) u ∈ C(Ω , TN (V)) for every N = 0, 1, . . . and, for each
K ⊂ Ω compact, there is a constant C = C(K) > 0 such that
D(N ) uK,h ≤ C N +1 N !s , N = 0, 1, . . .
In the next section we shall show that the sheaves Ω
→ Gsh (Ω ; V) are indeed
independent of the choice of the metric h.
3. Local expressions
We begin by recalling the standard coordinates and generators associated to a
locally integrable structure (cf. [T, I.5] and [BCH, I.10]). Each point of Ω is the
center of a coordinate system (x1 , . . . , xm , t1 , . . . , tn ), which can be assumed defined
in a product U = B × Θ, where B (respectively Θ) is an open ball centered in the
origin in Rmx (respectively Rt ), over which there is defined a smooth vector-valued
n
function Φ(x, t) = (Φ1 (x, t), . . . , Φm (x, t)) satisfying

Φ(0, 0) = 0, Φx (0, 0) = 0,
such that the differential of the functions
Zk (x, t) = xk + iΦk (x, t), k = 1, . . . , m,
⊥
span V over U .
If we define the vector fields

m
∂
Mk = μkk (x, t) , k = 1, . . . , m

∂x k
k =1
characterized by the rule
Mk Zk = δk,k , k, k = 1, . . . , m,
then the complex vector fields
∂ ∂φk
m
Lj = −i (x, t)Mk , j = 1, . . . , n,
∂tj ∂tj
k=1
span V over U .
The following properties are easily checked:
(1) L1 , . . . , Ln , M1 , . . . , Mm span CTΩ over U and are pairwise commuting.
(2) dZ1 , . . . , dZm , dt1 , . . . , dtn span CT∗ Ω over U .
Property (2) allows us to identify T(V)|U to the bundle spanned by the differ-
ential forms dt1 , . . . , dtn and the formula

n
m
du = (Lj u) dtj + (Mk u) dZk , u ∈ C 1 (U ),
j=1 k=1
allows us to express d u as

n
d u = (Lj u) dtj .
j=1
Likewise we can identify TN (V)|U to the bundle spanned by dti1 ⊗ · · · ⊗ dtiN ,

where i1 , . . . , iN ∈ {1, . . . , n}; we have
⎛ ⎞

DN ⎝ ui1 ···iN dti1 ⊗ · · · ⊗ dtiN ⎠
1≤i1 ,...,iN ≤n

n
= (Lj ui1 ···iN ) dtj ⊗ dti1 ⊗ · · · ⊗ dtiN ,
j=1 1≤i1 ,...,iN ≤n

D(N ) u = (Li1 · · · LiN u) dti1 ⊗ · · · ⊗ dtiN .
1≤i1 ,...,iN ≤n
We now assume that T(V) is endowed with a smooth hermitian metric h. In

the local coordinates just described, if I, J ∈ {1, . . . , n}N , I = (i1 , . . . , iN ), J =
(j1 , . . . , jN ), if we set
(N )
hIJ = hN (dti1 ⊗ . . . ⊗ dtiN , dtj1 ⊗ . . . ⊗ dtjN )
and if we take

n
u= uI dti1 ⊗ . . . ⊗ dtiN ∈ C(U ; TN (V))
I
we have
⎧ ⎫1/2
⎨ ⎬
(N )
uK,h = sup hIJ uI uJ .
K ⎩ ⎭
I,J
Now by definition we have, if hij = h(dti , dtj ), 1 ≤ i, j ≤ n,

(N )
hIJ = hi1 ,j1 hi2 ,j2 · · · hiN ,jN .
Lemma 3.1. For each K ⊂ U compact there are constants b > 0, B > 0 such
that

(1) bN sup |uI |2 ≤ u2K,h ≤ B N sup |uI |2 , u ∈ C(U ; TN (V)).
K K
I I
Proof. We have, by the Cauchy-Schwarz inequality,

⎧ ⎫1/2 ⎧ ⎫N/2
n ⎨ 2 ⎬ ⎨ ⎬
(N )
hIJ uI uJ ≤
(N
hIJ
)
|uI | 2
= 2
hij |uI | 2
,

I,J=1 ⎩ I,J ⎭
I
⎩
i,j
⎭
I

and hence we can take B 2 = supK i,j h2ij .
For the first inequality in (1) we first notice that the matrix that represents the
inverse of hN is given by hiN ,jN · · · hi2 ,j2 hi1 ,j1 , where hij denotes the inverse matrix
of hij .
We reason pointwise. For this we denote

|u|2 = |uI |2 , u, v = uI vI .
I I
(N )
We shall also denote by H the linear operator Hu = (wI )I , where wI = J hIJ uJ .
Then H > 0, which implies
|u|2 = |H −1/2 H 1/2 u|2 ≤ H −1/2 2 Hu, u .
Hence, in order to complete the proof, it suffices to notice that
⎧ ⎫N/2
⎨ ⎬
2
H −1/2 2 = H −1 ≤ hi,j .
⎩ ⎭
i,j

In the next statement we shall use the following notation: if α ∈ Zn+ is a
multi-index we shall write Lα = Lα αn
1 . . . Ln .
1
Corollary 3.1. Fix an hermitian metric h on T(V). Let u ∈ C(U ). Then

u ∈ Gsh (U ; V) if and only if Lα u ∈ C(U ) for every α ∈ Zn+ and, for each compact
subset K of U , there is a constant A = A(K) > 0 such that
(2) sup |Lα u| ≤ A|α|+1 α!s , α ∈ Zn+ .
K
In particular it follows that the sheaf Ω
→ Gsh (Ω ) does not depend on the choice
of the hermitian metric h.
Proof. Since the vector fields Lj are pairwise commuting we can write
N!
|Li1 . . . LiN u|2 = |Lα u|2
α1 ! · · · αn !
1≤i1 ,...,iN ≤n |α|=N
and hence
⎧ ⎫1/2
1/2
1/2 ⎨ ⎬
α! N! α 2
|Lα u| = |L u| ≤ |Li1 . . . LiN u|2 .
|α|!1/2 α! ⎩ ⎭
1≤i1 ,...,iN ≤n
Hence, if u ∈ Gsh (U ; V) and if K ⊂ U is compact, Lemma 3.1 implies the existence

of C• > 0 such that
⎧ ⎫1/2
⎨ ⎬
sup |Li1 . . . LiN u|2 ≤ C•N +1 N !s .
K ⎩ ⎭
1≤i1 ,...,iN ≤n
Then
|α|+1 |α|+1
sup |Lα u| ≤ C• |α|!s ≤ ns|α| C• α!s , α ∈ Zn+ .
K
Conversely, if estimates (2) hold then the inequalities
N!
|Li1 . . . LiN u|2 = |Lα u|2 ≤ nN |Lα u|2
α1 ! · · · αn !
1≤i1 ,...,iN ≤n |α|=N |α|=N
together with (1) imply that u ∈ Gsh (U ; V).
We shall denote by GsV the sheaf Ω

→ Gsh (Ω ) over Ω (we have the right to
drop the mention to h in this definition). If Ω is open the elements of Γ(Ω ; GsV )
will be called s-Gevrey vectors for V on Ω (resp. analytic vectors for V on Ω when
s = 1).
4. Local characterization of the analytic vectors for V

We continue to work under the local coordinates and generators as described
in the preceding section.
Theorem 4.1. Let u ∈ Γ(U ; G1V ). Then there are an open neighborhood V ⊂ U
of the origin, an open neighborhood D of the origin in Cn and a function v =
v(x, t, w), defined on V × D, satisfying
• v ∈ L∞ (V × D) ∩ C(V ; O(D));
• Lj v(·, ·, w) = 0, j = 1, . . . , n, w ∈ D;
• v(x, t, t) = u(x, t).
Here O(D) denotes the space of holomorphic functions on D.
Proof. Contracting U = B × Θ if necessary we can assume that Lα u ∈ C(U )
for every α and that supU |Lα u| ≤ C |α|+1 α! for every α and some C > 0.
We set Θ• = {t ∈ Θ : |t| < 1/(4C)} and D = {w ∈ Cn : |w| < 1/(4C)}. The
series
(Lα u)(x, t)
v(x, t, w) = (−1)|α| (t − w)α
n
α!
α∈Z+
defines an element v ∈ C(B × Θ• ; O(D)) ∩ L∞ (B × Θ• × D) which satisfies the

required properties.
Given a locally integrable structure V over a smooth manifold Ω, a distribution

u ∈ D (Ω ), Ω ⊂ Ω open, is a solution for V if Lu = 0, whatever smooth section L of
V defined in an open subset of Ω . Denote by SV the sheaf of germs of solutions for
V; denote also by SoV the subsheaf of SV formed by all solutions that are defined
by continuous functions. Finally, let O(n) denote the ring of germs of holomorphic
functions at the origin in Cn . We return to the situation described in Theorem 4.1
and consider the inverse limit
(SoV )0 ⊗
ˆ O(n) = lim SoV (U )⊗O(D),
ˆ
U×D→(0,0)
where SoV (U )⊗O(D)

ˆ stands for the completion of the tensor product between the
Fréchet space SoV (U ) and the Fréchet-nuclear space O(D). There is a homomor-
phism between stalks at the origin
ˆ O(n) −→ (G1V )0
μ0 : (SoV )0 ⊗
defined as follows: if v ∈ (SoV )0 ⊗ ˆ O(n) is represented by (x, t, ζ)
→ v(x, t, ζ)
we set μ0 (v) as being the germ of analytic vector for V at the origin defined by
(x, t)
→ v(x, t, t). The conclusion of Theorem 4.1 implies that μ0 is an isomorphism,
which provides another invariant characterization for the analytic vectors for V.
5. Analytic vectors in hypocomplex structures

We begin by recalling the following definition ([T, III.5]).
Definition 5.1. Let Ω be a smooth manifold over which a locally integrable
structure is defined. We say that V is hypocomplex at a point A ∈ Ω if there are
an open neighborhood W of A in Ω and smooth functions Zj : W → C, Zj (A) = 0,
j = 1, . . . , m, whose differentials span V ⊥ over W , and such that the following
is true: given any solution u, defined near A there is a holomorphic function H,
defined in an open neighborhood of 0 ∈ Cm , such that u = H ◦ Z near the origin.
We shall now return to the local coordinates and notation described in Section
3. We shall assume that A is the origin for the coordinate system (x1 , . . . , xm ,
t1 , . . . , tn ).
Lemma 5.1. Assume that V is hypocomplex at the origin. Let V ⊂ U be an
open neighborhood of the origin. If F is a Banach space continuously contained in
{u ∈ D (V ) : Lj u = 0, j = 1, . . . , n}, with norm denoted by · F , there exist a
complex neighborhood W of 0 ∈ Cm and a constant C > 0 satisfying the following
property:
• Given u ∈ F there is h ∈ O(W) such that u = h ◦ Z in Z −1 (W) ∩ V and
sup |h| ≤ CuF .
W
In the proof we shall use the following notation: if W is an open subset of Cm

we shall denote by O∞ (W) the Banach space of all bounded, holomorphic functions
on W endowed with the supremum norm. Also we shall denote by Bδ the open ball
in Cm centered at the origin and with radius δ > 0.
Proof. Let p be a large positive integer. Denote by Ep the Banach space of
all pairs (u, h) ∈ F × O∞ (B1/p ) satisfying u = h ◦ Z on Z −1 (B1/p ). Denote also
by Tp the continuous linear map Ep → F , Tp (u, h) = u. Since V is hypocomplex
at the origin we must have ∪p Tp (Ep ) = F . By Baire’s theorem some Tp0 (Ep0 )
must be of second category in F and the open mapping theorem then implies that
Tp0 : Ep0 → F is surjective and open. In particular there must exist > 0 such
that
{u ∈ F : uF ≤ } ⊂ Tp0 (u, h) ∈ Ep0 : sup |h| < 1 ,
B1/p0
from which the result follows after taking C = 1/ and W = B1/p0 .
Theorem 5.1. Assume that the system V is hypocomplex at the origin and let
u ∈ Γ(U ; G1V ). Then there are an open set W × W in Cm × Cn containing the
origin and G ∈ O(W × W ) satisfying u(x, t) = G(Z(x, t), t) in {(x, t) : (Z(x, t), t) ∈
W × W }.
. α
Proof. Take V , D and v as in Theorem 4.1. If we set vα (x, t) = (∂w v)(x, t, 0)
we also have Lj vα = 0, j = 1, . . . , n, α ∈ Zn+ and the Cauchy estimates give
(3) sup |vα | ≤ A|α|+1 α! .
V
We apply Lemma 5.1 with F = {u ∈ C(V ) ∩ L∞ (V ) : Lj u = 0, j = 1, . . . , n}.

We conclude the existence of an open neighborhood W of the origin in Cm , of a
constant C > 0 and of holomorphic functions gα ∈ O(W) such that vα = gα ◦ Z in

Z −1 (W) ∩ V and
(4) sup |gα | ≤ C sup |vα |, α ∈ Zn+ .
W V
Define
gα (z)
G(z, w) = wα .
n
α!
α∈Z+
Thanks to (3) and (4) there is a neighborhood W of the origin in Cn such that G
defines a holomorphic function in W × W . Finally,
gα (Z(x, t)) vα (x, t)
G(Z(x, t), t) = tα = tα
n
α! n
α!
α∈Z+ α∈Z+
(∂ α v)(x, t, 0)
w
= tα = v(x, t, t) = u(x, t),
n
α!
α∈Z+
which concludes the proof.

Remark 5.1. Fix a locally integrable structure V and consider the local coor-
dinates (x, t) and generators Lj , Mk as before. According to [T, Proposition II.4.2],
a smooth function f (x, t) can be written near the origin as f (x, t) = f˜(Z(x, t), t),
with f˜ holomorphic in Cm × Cn , if and only if f is an analytic vector for the
complete system {M1 , . . . , Mm , L1 , . . . , Ln }. Thus Theorem 5.1 just says that, in a
hypocomplex structure, the concepts of being an analytic vector for {L1 , . . . , Ln }
and for {M1 , . . . , Mm , L1 , . . . , Ln } are equivalent.
Theorem 5.1 in connection with the arguments in [CCP, Section 9] allow us to
state:
Corollary 5.1. Assume that V is hypocomplex at the origin and let u be a
C 1 function defined in an open neighborhood of the origin in U and satisfying the
system
(5) (Lj u) (x, t) = fj (Z(x, t), t, u(x, t)), j = 1, . . . , n,
where the functions fj (z, w, ζ) are holomorphic near the origin in Cm+n+1 . Then
near the origin we can write u(x, t) = G(Z(x, t), t), where G(z, w) is holomorphic
in a neighborhood of the origin in Cm+n .
6. Real-analytic locally integrable structures

In this section we assume that both the manifold Ω and the vector bundle
are real-analytic. Notice that in such situation, returning to the general set up
described in Section 3, all the vector fields Lj and the functions Zk can be assumed
real-analytic.
The proof of the next result follows after an elementary argument ([CCP, Propo-
sition 2.1], cf. also [BCR, Proposition 1.1.3]).
Proposition 6.1. Given Ω ⊂ Ω open and s ≥ 1 we have the inclusion
{u ∈ C(Ω ) : d u ∈ Gs (Ω , T(V))} ⊂ Γ(Ω ; GsV ) .
We shall now discuss the analytic regularity of the analytic vectors for V. For
this we first recall some standard concepts (cf. [T, III.5]).
Definition 6.1. We say that

• V is analytic hypoelliptic if given Ω ⊂ Ω open and u ∈ D (Ω ) then
d u ∈ C ω (Ω ; T(V)) =⇒ u ∈ C ω (Ω ) .
• V is globally analytic-hypoelliptic if given u ∈ D (Ω) then
d u ∈ C ω (Ω; T(V)) =⇒ u ∈ C ω (Ω) .
By Proposition 6.1 it follows that when V is not analytic hypoelliptic then there
are analytic vectors which are not real analytic. On the other hand, according to
[T, Proposition III.5.3], V is analytic hypoelliptic if and only if V is hypocomplex
at each point of Ω and then, as a consequence of Theorem 5.1, we can state:
Corollary 6.1. Assume that V is analytic hypoelliptic. If Ω ⊂ Ω is open
then Γ(Ω ; G1V ) ⊂ C ω (Ω ).
A similar result for global analytic hypoellipticity is no longer true. To give an
example we take as Ω the two dimensional torus Ω = S 1 × S 1 , with coordinates
written as (x, t), and let V ⊂ CT(S 1 × S 1 ) be the vector bundle spanned by the
real vector field
∂ ∂
L= −α ,
∂t ∂x
where α ∈ R. It is clear that V defines a real-analytic, locally integrable structure
over S 1 × S 1 of rank one, which is never analytic hypoelliptic; on the other hand
there are values of α ∈ R \ Q for which V is globally analytic hypoelliptic (cf. [G]).
Proposition 6.2. For any α ∈ R and any s ≥ 1 there is u ∈ Γ(S 1 × S 1 ; G1V )
such that u ∈ Gs (S 1 × S 1 ).
Proof. Since when α ∈ Q it is easy to construct global continuous solutions
to the equation Lu = 0 which are not smooth, we can restrict ourselves to the case
α ∈ R \ Q. Changing x to −x if necessary allows us to assume that α > 0 and
then there are sequences {pk }, {qk } of natural numbers such that qk → ∞ and
{pk − αqk } is bounded. We set
τk = 1/ log(log qk ) .
Then τk → 0 and consequently we can find k0 so that 0 < τk < 1/(2s) if k ≥ k0 .
We then set, for k ≥ k0 ,

.
Ak = exp −(pk + qk )−τk +1/s − |pk − αqk | .
Since
Ak ≤ exp −(pk + qk )1/(2s)
.
it follows that
u(x, t) = Ak exp{i(pk t + qk x)}
k≥k0
defines a smooth function on S × S (indeed, u ∈ G2s (S 1 × S 1 )). Moreover, for
1 1
N = 0, 1, . . .,

LN u(x, t) = iN (pk − αqk )N Ak exp{i(pk t + qk x)}
k≥k0
and thus, noticing that

|pk − αqk |N Ak ≤ N ! exp −(pk + qk )−τk +1/s ,
we obtain

|LN u(x, t)| ≤ N ! exp −(pk + qk )1/(2s) .
k≥k0
In particular u ∈ Γ(S ×S 1 1
It remains to prove that u ∈ Gs (S 1 ×S 1 ). Indeed,
; G1V ).
if this were true we would obtain constants B > 0, > 0, such that

Ak ≤ B exp −(pk + qk )1/s , k ≥ k0 .
Hence

exp (pk + qk )1/s ≤ B exp (pk + qk )−τk +1/s + |pk − αqk | , k ≥ k0 ,
or, equivalently,
(pk + qk )1/s ≤ log B + (pk + qk )−τk +1/s + |pk − αqk | , k ≥ k0 .
Dividing by (pk + qk )1/s gives
log B + |pk − αqk | 1
(6) ≤ + , k ≥ k0 .
(pk + qk ) 1/s (pk + qk )τk
If we now notice that
log B + |pk − αqk | const.
1/s
≤ −→ 0
(pk + qk ) (pk + qk )1/s
and that
1
≤ qk−τk = e−τk log qk −→ 0,
(pk + qk )τk
since τk log qk → ∞, we conclude that the right end side of (6) converges to 0 as
k → ∞, which gives the sought contradiction.
Regularity of Gevrey vectors. Still in the case when Ω and V are real-analytic, it
is a natural question to ask about the Gevrey regularity of the elements in Γ(Ω ; GsV ),
s > 1. When V is elliptic, that is when V ⊥ ∩ T∗ Ω = 0, then Γ(Ω ; GsV ) = Gs (Ω )
for every Ω ⊂ Ω open and s ≥ 1. Indeed, when V is elliptic we have m ≤ n and we
can take, in the local coordinates decribed in Section 3, Φj (x, t) = tj , j = 1, . . . , m.
The vector fields Lj now read
∂ ∂ ∂
Lj = −i , j = 1, . . . , m, Lj = , j = m + 1, . . . , n.
∂tj ∂xj ∂tj
If u ∈ Γ(U ; GsV ) then u is an s-Gevrey vector for each Lj , in the following sense:
for every K ⊂ U compact there is a constant C = C(K) > 0 such that
sup |Lpj u| ≤ C p+1 p!, p = 0, 1, . . . , j = 1, . . . , n.
K
It then follows from the results in [BCM] that the Gs -wave-front of u is contained
in σ(Lj ), the characteristic set of Lj over U , j = 1, . . . , n. Since ∩nj=1 σ(Lj ) = ∅ it
follows that u ∈ Gs (U ).
We conclude this work by presenting a partial converse of this statement, which
is inspired by a result on scalar operators due to G. Métivier [M]:
Proposition 6.3. Assume that V is not elliptic at A ∈ Ω, that is, assume that
⊥
VA ∩ T∗A Ω = 0. Then given s, s satisfying
1 < s ≤ s < 2s − 1
there is an s-Gevrey vector for V near A which is not a Gevrey function of order
s .
Proof. We can select the local coordinates (x, t) and the local generators Lj ,
dZk as in section 3 in such a way that, in these coordinates, A = (0, 0). By
hypohesis there is ζ0 = (ζ01 , . . . , ζ0m ) ∈ Cm not zero such that k ζ0k dZk (0, 0) is a
real covector.
Since dZk (0, 0) = dxk + idt Φk (0, 0) it follows that ζ0 = ξ0 ∈ Rm and
that k ξ0k dt Φk (0, 0) = 0. Consequently d(Φ · ξ0 ) = 0 at (0, 0) and hence there is
a constant C > 0 such that |Φ(x, t) · ξ0 | ≤ C(|x|2 + |t|2 ) when (x, t) ∈ U .
For convenience, we will write the local generators Lj as
∂ m
∂
Lj = + ajk (x, t) , j = 1, . . . , n,
∂tj ∂xk
k=1
where the coefficients ajk (x, t) are assumed to be real-analytic functions in a neigh-
borhood of the closure of U .
.
Let α ∈]0, 1[ be such that s < 1/α < 2s−1 and define σ = s−(1−α)/(2α) > 1.
Next we√select ζ ∈ Gc (R ) such that ζ(0)
σ n
√ = 1 and with support contained in the ball
|t| ≤ ρ/ m + 1, where ρ < min{r, 1/ C} and r is the radius of Θ. We can assume
that ]−r, r[ m ⊂ B. We also take a cut-off function ψ ∈ Gσc(R),√satisfying ψ√≡ 1 in a
neighborhood of the origin and supported in the interval −ρ/ m + 1, ρ/ m + 1 .
We then set
! ∞
eiλZ(x,t)·ξ0 −λ ζ λ(1−α)/2 t ψ λ(1−α)/2 x1 . . . ψ λ(1−α)/2 xm dλ.
α
u(x, t) =
1
If λ(1−α)/2 t ∈ supp ζ and λ(1−α)/2 x1 , . . . , λ(1−α)/2 xm ∈ supp ψ then

λ|Φ(x, t) · ξ0 | ≤ Cλ(|x|2 + |t|2 ) ≤ Cρ2 λα
and, consequently, u is well defined and smooth in U .
Since the derivatives of the function ψ vanish identically in a neighborhood of
the origin, setting ξ0 · M = m j=1 ξ0j Mj gives
! ∞ " # ! 1
1 k+1
λk e−λ dλ = Γ λk e−λ dλ.
α α
{(ξ0 · M)k u}(0, 0) = −
1 α α 0
Noticing that the last term at the right end side is bounded in k it follows from the
asymptotic behaviour of the Gamma function that u is not of Gevrey class τ near
the origin, for any τ < 1/α.
Summing up, in order to complete the argument we must show that u is an
s-Gevrey vector for V. For β ∈ Zn+ we have
Lβ u(x, t) =
"β #! ∞
λ(1−α)(|β|−|γ|)/2 eiλZ(x,t)·ξ0 −λ ζ (β−γ) (λ(1−α)/2 t)Lγ (Ψ(λ, x)) dλ,
α
(7)
γ 1
γ≤β

where Ψ(λ, x) = ψ λ(1−α)/2 x1 . . . ψ λ(1−α)/2 xm . We shall prove that, for some
constant A > 0, the estimates
(8) sup |Lγ (Ψ(λ, x))| ≤ A|γ|+1 |γ|!σ , γ ∈ Zn+

U
hold. Assuming this for a moment we obtain, from (7),

! ∞
|β|+1
λ(1−α)|β|/2 e−(1−Cρ )λ dλ
2 α
sup |L u| ≤ A1
β
|β|! σ
U 1
|β|+1 σ+(1−α)/(2α)
≤ A2 β!
|β|+1 s
= A2 β! ,
which shows that indeed u is an s-Gevrey vector for V.
The remainder of the argument will be devoted to the proof of estimates (8).
Write r = λ(1−α)/2 (≥ 1) and w(x ) = rx , = 1, . . . , m. In order to estimate

L (Ψ(λ, x)) we must first analyze the terms Lη ψ(w(x )), when η ∈ Zn+ and =
γ
1, . . . , m. For this, we first apply the Faà di Bruno’s formula

η
Lη ψ(w(x )) = Cq,θ ψ (q) (w(x ))Lθ1 w(x ) . . . Lθq w(x ).
θ1 +...+θq =η
|θj |>0
η
Here Cq,θ are universal constants. Since the vector fields Lj commute pairwise,
and since also |θj | > 0, we can choose integers 1 ≤ ij ≤ n such that Lθj w(x ) =
rLθj −eij aij (x, t) (here ek = (δkp )1≤p≤k ∈ Zn+ ). The last equality can then be
written as

Lη ψ(w(x )) = η
Cq,θ r q ψ (q) (w(x ))Lθ1 −ei1 ai1 (x, t) . . . Lθq −eiq aiq (x, t).
θ1 +...+θq =η
|θj |>0
Since the coefficients ajk are real-analytic in Ū and since ψ ∈ Gσc (R) we obtain
the existence of a constant B > 0 such that

(9) sup |Lη ψ(w(x ))| ≤ r |η| B |η|+1 η
Cq,θ B q q!σ |θ1 |! . . . |θq |!.
U
θ1 +...+θq =η
|θj |>0
For 0 < ε < 1/B and c > 0 we set

ck!
Nk = .
(k + 1)n+1 εk
Here c > 0 is chosen in such a way that
"β #
N|β | N|β−β | ≤ N|β| , β ∈ Zn+ .

β
β ≤β
Then we can write

(|θj | + 1)n+1 ε|θj | (n + 1)!e|θj |+1 ε|θj |
|θj |! = N|θj | ≤ N|θj | , ∀j = 1, . . . , q
c c
and therefore we have

" #q
(n + 1)!e
(10) |θ1 |! . . . |θq |! ≤ N|θ1 | . . . N|θq | (εe)|η|
c
" #|η|
|η| (n + 1)!e
≤ N|θ1 | . . . N|θq | (εe) 1+ .
c
Inserting (10) into (9) gives, with a new constant B1 > 0,
|η|+1 |η|

(11) sup |Lη ψ(w(x ))| ≤ B1 ε |η|!σ−1 r |η| η
Cq,θ B q q!N|θ1 | . . . N|θq | .
U
θ1 +...+θq =η
|θj |>0
If we now introduce the formal power series

∞

φ(w) = B w , w ∈ R,
=1
N|α|
τ (y) = y α , y ∈ Rn
α!
|α|>0
again the Faà di Bruno’s formula shows that (11) can be written as
|η|+1 |η|
(12) sup |Lη ψ(w(x ))| ≤ B1 ε |η|!σ−1 r |η| ∂ η (φ ◦ τ )(0).
U
Arguing as in [AM, p. 197], (see also [CCP, Section 9]), we conclude that
B
|∂ η (φ ◦ τ )(0)| ≤ N|η|
1 − Bε
and consequently, with another constant B2 > 0,
|η|+1 |η| B
sup |Lη ψ(w(x ))| ≤ B1 ε |η|!σ−1 r |η|N
1 − Bε |η|
|η|+1 |η| B c|η|!
= B1 ε |η|!σ−1 r |η|
1 − Bε (|η| + 1)n+1 ε|η|
|η|+1
(13) ≤ r |η| B2 |η|!σ .
Finally, if we observe that
Lγ (Ψ(λ, x)) =
" #" # " #

γ θ1 θm−2 γ−θ1
··· ... L ψ(w(x1 )) · · · Lθm−1 ψ(w(xm )),
θ1 θ2 θm−1
θ1 ≤γ θ2 ≤θ1 θm−1 ≤θm−2
inequality (13) allows us to estimate

|γ|+m |γ|+m
sup |Lγ (Ψ(λ, x))| ≤ 2(m−1)|γ| r |γ| B2 |γ|!σ = 2(m−1)|γ| λ(1−α)|γ|/2 B2 |γ|!σ ,
U
from which (8) follows. The proof of Proposition 6.3 is complete.

References
[AM] S. Alinhac and G. Metivier, Propagation de l’analyticité des solutions de systèmes hyper-
boliques non-linéaires. Invent. Math. 75 (1984), 189–204.
[BG] M.S. Baouendi and C. Goulaouic, Régularité analytique et itères d’operateurs elliptiques
dégénérés; applications. J. Funct. Analysis 9 (1972), 208–248.
[BM] M.S. Baouendi and G. Metivier, Analytic vectors of hypoelliptic operators of principal
type. American J. Math. 104 (1982), 287–319.
[BCH] S. Berhanu, P.D. Cordaro and J. Hounie, An introduction to involutive structures. Cam-
bridge University Press, 2008.
[BCM] P. Bolley, J. Camus, C. Mattera, Analyticité microlocale et itères d’operateurs. Seminaire
Goulaouic–Schwartz 1978–1979, Exposé XIII. École Polytechnique, France.
[BCR] P. Bolley, J. Camus and L. Rodino, Hypoellipticite analytique-Gevrey et itères d’operateurs.
Rend. Sem. Mat. Univers. Politecn. Torino 45 (1987), 1–61.
[CCP] J.E. Castellanos, P.D. Cordaro and G. Petronilho, Gevrey vectors in involutive tube struc-
tures and Gevrey regularity for the solutions to certain classes of semilinear systems,
(2010), to appear.
[Da] M. Damlakhi, Analyticité et itères d’operateus pseudo-différentiels. J. Math. Pures et Appl.
58 (1979), 63–74.
[DaH] M. Damlakhi and B. Helffer, Analyticité et itères d’un système de champs non elliptique.
Annales scient. Éc. Norm. Sup. 4e. série, 13 (1980), 397–403.
[G] C. Goulaouic, Interpolation entre des espaces localement convexes définis à l’aide de semi-
groupes. Ann. Inst. Fourier Grenoble 19 (1969), 269–278.
[HM] B. Hellfer and C. Mattera, Analyticité et itères reduits d’un système de champs de vecteurs.
Comm. PDE 5(10) (1980), 1065–1072.
[M] G. Metivier, Propriete des itères et ellipticite. Comm. PDE 3(9) (1978), 827–876.
[N] E. Nelson, Analytic vectors. Ann. Math. 70 (1959), 572–615.
[RS] M. Reed and B. Simon, Methods of modern mathematical physics I: functional analysis.
Academic Press, 1972.
[T] F. Treves, Hypo-analytic structures: local theory. Princeton University Press, 1992.
Universidade Federal de São Carlos, São Carlos, SP, Brazil

E-mail address: barostichi@dm.ufscar.br
Universidade de São Paulo, São Paulo, SP, Brazil

E-mail address: cordaro@ime.usp.br
Universidade Federal de São Carlos, São Carlos, SP, Brazil

E-mail address: gerson@dm.ufscar.br
Volume 550, 2011
Subellipticity and maximal hypoellipticity for two complex

vector fields in (2 + 2)-variables
Makhlouf Derridj and Bernard Helffer
Abstract. Our aim is to revisit some aspects of the theory of hypoelliptic

systems initiated by F. Trèves. The analysis of the links between hypoellip-
ticity and subellipticity for systems appears indeed to be far from completely
understood. Some aspects were analyzed by the authors but we will analyze
in this article new phenomena occuring in larger dimension. In particular we
will compare explicit criteria for maximal hypoellipticity to explicit criteria for
subellipticity.
1. Introduction
1.1. General context.
Our aim is to revisit some aspects of the theory of hypoelliptic systems whose
systematic analysis was started by F. Trèves [19, 20] and continued by [11, 12,
13, 14]. The analysis of the links between hypoellipticity and subellipticity for
systems appears indeed to be far from completely understood. Some aspects were
analyzed in [1, 2, 3] in connection with a very nice analysis of J-L. Journé et J.-M.
Trépreau [8] but we will analyze in this article new phenomena occuring in larger
dimension and compare, in the spirit of what was done in the book [4], with what
can be done for characterizing maximal hypoellipticity.
We will emphasize in the first part on the approach based on the theory on
nilpotent groups (collaboration of the second author with J. Nourrigat [5] and
further work of him [16, 17, 18] or of the second author with F. Nier [4] ) and in
the second part we will come back to another approach initiated by the first author
[1] and then together [2, 3] leading directly to criteria in subellipticity.
As we shall see, an interest of this analysis of the maximal hypoellipticity by
an approach based on the nilpotent Lie group techniques is that it provides global,
local or microlocal estimates.
More precisely, our aim is to analyze the maximal hypoellipticity of the system
of n first order complex vector fields
(1.1) Lj = (Xj + iYj ), where Xj = ∂tj and Yj = (∂tj B(t)) ∂x ,
1991 Mathematics Subject Classification. Primary 35F05; Secondary 35B65.

Key words and phrases. Partial Differential Equations, Complex Analysis, Subellipticity.
The authors were partially supported by the ESI (E. Schrödinger Institut).
2011
c 0000
c Mathematical
American (copyright Society
holder)
1
15
16
2 MAKHLOUF DERRIDJ AND BERNARD HELFFER
in a neighborhood V(0) × Rt of 0 ∈ Rn+1 , where B ∈ C ∞ (V(0)).

More generally, we will consider on Rn+m the system

m
Yj = (∂tj B (t)) ∂x ,
=1
and, the case m = 1 being already analyzed in [2, 3], we will mainly discuss in this
paper the case m = 2. We will show at the same time how the techniques used for
this analysis will lead to some information on the question concerning the Witten
Laplacian associated to B.
We assume that the real function B is such that the rank r Hörmander condition
is satisfied for the vector fields (Xj ), (Yj ) at (0, 0).
This is an immediate consequence of :
Definition 1.1. Hörmander’s condition of rank r

(1.2) |∂tα B(0)| > 0 .
1≤|α|≤r
By maximal hypoellipticity for the system (1.1), we mean the existence of the
inequality :
⎛ ⎞

(1.3) ||Xj u||2 + ||Yj u||2 ≤ C ⎝ ||Lj u||2 + ||u||2 ⎠ , ∀u ∈ C0∞ (V(0) × R) .
j j j
The symbol of the system is the map :

T ∗ (V(0) × R) \ {0} (x, t, ξ, τ )
(1.4)
→ σ(L)(x, t, ξ, τ ) := iτj − ξ(∂tj B)(t) j=1,...,n ∈ Cn .
The characteristic set is then by definition the set of zeroes of (the principal symbol
of) σ(L) :
(1.5) σ(L)−1 (0) = {τ = 0 , ∇B(t) = 0} .
Outside this set the system is microlocally elliptic (its (principal) symbol does
not vanish). By microlocally elliptic at (x0 , t0 , ξ0 , τ0 ) we mean the existence of a
standard pseudo-differential operator of order 0 which is elliptic at this point and
such that

(1.6) ||χ(x, t, Dx , Dt )u||21 ≤ ||Lj u||2 + ||u||2 , ∀u ∈ C0∞ ,
where || · ||s denotes the standard Sobolev’s norm.

This of course implies microlocal (maximal) hypoellipticity at this point.
So the local (maximal) hypoellipticity will result of the microlocal analysis in
the neighborhood of the characteristic set, which in this particular case has actually
two connected components defined by {±ξ > 0}.
So we are more precisely interested in the microlocal hypoellipticity in a conic
neighborhood V± of (x, t; ξ, τ ) = (0, 0; ±1, 0), that is with the microlocalized version
of the inequality (1.3).
Due to the invariance of the problem with respect to the x variable, we look for an
inequality which is local in t but global in the x variable and take the partial Fourier
transform with respect to x in order to analyze the problem (see Subsection7.1).
SUBELLIPTICITY AND MAXIMAL HYPOELLIPTICITY 17
3
In the case when we work in Rn+m with m > 1, we will work microlocally near
(0, 0; ξ0 , 0) and new phenomena appear. This is the main point of this paper.
We say that the system is subelliptic in the neighborhood of the origin is there
exists s > 0 , C > 0 and some open neighborhood of the origin V(0), such that

(1.7) ||u||2s ≤ Cs ||Lj u||2 + ||u||2 , ∀u ∈ C0∞ (V(0)) .

Due to the invariance by translation in the x-variable, it is enough to show semi-
global estimates after a partial Fourier transform in the x-variable, where the dual
variable ξ will play the role of a parameter. Due to Hörmander condition, maximal
hypoellipticity implies subellipticity with s = 1r . By microlocally subelliptic at
(0, 0, ξ, τ ) we mean the existence of a standard pseudo-differential operator of order
0 which is elliptic at (0, 0, ξ, τ ) and such that

(1.8) ||χ(x, t, Dx , Dt )u||2s ≤ Cs ||Lj u||2 + ||u||2 , ∀u ∈ C0∞ .

Similarly microlocal maximal hypoellipticity for the system (1.1) means the exis-
tence of the inequality :
(1.9) ⎛ ⎞

||χ(x, t, Dx , Dt )Xj u||2 + ||χ(x, t, Dx , Dt )Yj u||2 ≤ C ⎝ ||Lj u||2 + ||u||2 ⎠ .
j j j
Microlocal maximal hypoellipticity has been characterized in [5] for this type of
systems under a general criterion which may appear as rather implicit (see although
[16] ) and this study has been extended and developed in a series of paper by J.
Nourrigat [17, 18]. For subellipticity, the tubular case permits a simplification of
the analysis (using a partial Fourier transform). This was developed in [1, 2, 3]
and will be shortly recalled in Subsection 7.1.
1.2. What is known when m = 1.
For about twenty five years it was a kind of folk theorem that, when B analytic,
these systems were subelliptic as soon as they were hypoelliptic. This was indeed
the case when n = 1 [19] but in the case n > 1, an inaccurate reading of the proof
(based on a non standard subelliptic estimate) given by Maire [11] (see also Trèves
[20]) of the hypoellipticity of such systems, under the condition that B does not
admit any local maximum or minimum, was supporting the belief for this folk theo-
rem. This question reappears in the book of [4] in connection with the semi-classical
analysis of Witten Laplacians. Quite recently, J.L. Journé and J.M.Trépreau [8]
show by explicit examples that there are very simple systems (with polynomial B’s)
which were hypoelliptic but not subelliptic in the standard L2 -sense. But in these
examples, B is not quasihomogeneous.
In [1] and [2, 3] the homogeneous and the quasihomogeneous cases were analyzed
in dimension 2. We showed indeed that the folk theorem was correct in the quasi-
homogeneous case.
1.3. Main goals of the paper : n = 2, m = 2.
We would like to analyze the microlocal subellipticity at (0, 0, 0, 0 ; 0, 0, 1, 0) of
systems defined by
(1.10) L1 = ∂s + i∂s B(s, t, ∂x , ∂y ) , L2 = ∂t + i∂t B(s, t, ∂x , ∂y ) .
In short we denote this system by (LB ) to mention the reference to B. We will
concentrate our analysis on the homogeneous case. The new point is that we have
18
two variables x and y.

Hence we have
(1.11) B(s, t, ξ, η) = B1 (s, t)ξ + B2 (s, t)η ,
with B1 and B2 homogeneous, and we always assume that Trèves condition is
microlocally satisfied at the point (0, 0, 0, 0; 0, 0, 1, 0) , that is :
Condition 1.2 (Microlocal Trèves Condition).
For (ξ, η) close to (1, 0), (s, t) → B(s, t, ξ, η) has no local maximum in some neigh-
borhood of (0, 0).
We know from Trèves [20] that this is a necessary condition for microlocal
hypoellipticity. Moreover the sufficiency of this condition for hypoellipticity is only
known [11] for models with analytic coefficients in the form B(s, t, ξ) = B(s, t)|ξ|.
If we are not in this case, there is a sufficient condition called (RR) by Maire [11]
which is less easy to verify, except for the homogeneous case of degree 2, where one
can use Morse’s Lemma. One has indeed to verify that B is “locally trivial” (see
Definition 30 in [11]) in (ξ, η). This does not seem to be satisfied for most of the
examples we are considering in this article. Note that Journé-Trépreau [8] discuss
only the case when B = B(s, t, ξ). We emphasize that, in any case, this condition
does not give the subellipticity in the usual sense.
Finally, let us recall that H.M. Maire gave an example of a non hypoelliptic system
(1.10) satisfying the Trèves condition :
(1.12) B(s, t, ξ, η) = t4 sη + t3 η − 3tξ .
Note that this example is not homogeneous is (s, t).
In part II, we will refer to [1, 2, 3], where more general functions are considered.
Here we restrict for simplicity to homogeneous functions.
Acknowledgements
We would like to thank J.M. Trépreau for useful discussions. We thank also the
Schrödinger institute where part of this work has been done in November 2009.
5
Part 1. Maximal hypoellipticity

2. Helffer-Nourrigat’s criterion
The Helffer-Nourrigat criterion [5] in the form expressed in Nourrigat’s lecture
notes [16] (see there Th. 1, p. 118 and the remarks p. 123 showing the equivalence of
(ii) with (iii) and see also Helffer-Nier Lecture notes [4]) takes the following form :
Theorem 2.1.
We assume that B is a polynomial in (s, t) of degree ≤ k. Then the system (1.10) is
microlocally maximally hypoelliptic at (0, 0, 0, 0) , (0, 0, 1, 0) if and only if, for any
sequence (sn , tn , ξn , ηn , dn ) and any non zero polynomial function P such that
• P (0, 0) = 0 ,
• limn→+∞ |sn |2 + |tn |2 = 0 ,

• limn→+∞ |ξn |2 + |ηn |2 = +∞ , limn→+∞ ξn /( |ξn |2 + |ηn |2 = 1 ,
• dn > 0 and limn→+∞ dn = 0
•
lim d|α|
n ∂s,t B(sn , tn , ξn , ηn ) = ∂s,t P (0, 0) , if 1 ≤ |α| ≤ k ,
α α
n→+∞
P does not admit a local maximum at (0, 0).

The strategy for applying this criterion for a given B is then to analyze the set
LB of all these polynomials and to show that either it is contained in the larger set
Lk,max of the non trivial polynomials of degree k without local maximum at (0, 0)
or to exhibit in LB one polynomial having a local maximum at (0, 0). Hence we do
not have to characterize LB .
Remark 2.2.
We recall that , under Hörmander’s condition, (microlocal) maximal hypoellipticity
implies (microlocal) subellipticity.
3. Homogeneous case of degree 2

We consider first the system (1.10) with
s2 1
(3.1) B(s, t, ξ, η) = ξ + (a12 st + a22 t2 )η .
2 2
For ξ = 1, η = 0, B is independent of t and has a non degenerate local minimum
at 0 in the s-variable, hence it can not have a local maximum. Moreover the fact
that the Hessian of B has a strictly positive eigenvalue for η = 0 remains true for
η small enough. Hence it is clear that Trèves condition is microlocally satisfied at
(0, 0, 0, 0; 0, 0, 1, 0).
We would like to analyze the microlocal maximal hypoellipticity at the point
(0, 0, 0, 0 ; 0, 0, 1, 0).
Proposition 3.1.
The system (1.10) is, when B is given by (3.1), microlocally maximally hypoelliptic
at (0, 0, 0, 0 ; 0, 0, 1, 0).
Proof.
We now compute the various quantities appearing in Theorem 2.1. We get
∂s B = sξ + a12 tη , ∂t B = (a12 s + a22 t)η
2 2 2
∂ss B=ξ , ∂st B = a12 η , ∂tt B = a22 η .
20
It is immediate to see that the only term of degree 2 for a limiting polynomial P is
1 (2) 2 (2)
2 11 s , with 11 ≥ 0. Here we use that limn→+∞ ηn /|ξn | = 0.
Hence any limiting polynomial has the form :
1 (2) 2 (1) (1)
P (s, t) = s + 1 s + 2 t ,
2 11
(2)
with 11 ≥ 0.
(1) (1)
We then observe, considering successively the cases when 2 = 0, then 2 =
(1) (1) (1) (2)
0, 1 = 0, and finally 2 = 0, 2 = 0 and 11 > 0, that there is no non trivial
limiting polynomial of this form with a local maximum at 0.
4. The case of homogeneity of degree 3

We start by a general discussion.
4.1. General discussion.
We take
s3 γ3
(4.1) B(s, t, ξ, η) = ξ + γ1 ts2 + γ2 t2 s + t3 η .

3 3
We would like to analyze the microlocal maximal hypoellipticity at the point s =
0, t = 0, x = 0, y = 0 ; σ = 0, τ = 0, ξ = 1, η = 0. We have not analyzed at this
stage when the Trèves condition is satisfied (see however the remark below leading
to the condition |γ2 | + |γ3 | > 0).
We now compute the various quantities appearing in the criterion. We get

∂s B = s2 ξ + 2γ1 ts + γ2 t2 η , ∂t B = γ1 s2 + 2γ2 ts + γ3 t2 η ,
2 2 2
∂ss B = 2sξ + 2γ1 tη , ∂st B = 2 (γ1 s + γ2 t) η , ∂tt B = 2 (γ2 s + γ3 t) η ,
3 3 3 3
∂sss B = 2ξ , ∂sst B = 2γ1 η , ∂stt B = 2γ2 η , ∂ttt B = 2γ3 η .
We now analyze the properties of the non-trivial limiting polynomials. It is clear

(3)
that the term of degree 3 is of the form 13 111 s3 . Let us look at the homogeneous
term of degree 2.
If |γ2 | + |γ3 | > 0, one obtains, by considering ∂ss 2
B − γγ12 ∂st
2 2
B or ∂ss B − γγ13 ∂tt
2
B at
2
the point (sn , tn , ξn , ηn ), multiplying by dn and passing to the limit, that the limits
limn→+∞ d2n sn ξn = σ1 and limn→+∞ d2n γ1 tn ηn exist.
In addition, limn→+∞ d2n tn ηn = σ2 exists. The homogeneous term of degree 2 has
consequently the structure
P2 (s, t) = (σ1 + γ1 σ2 )s2 + 2γ2 σ2 st + γ3 σ2 t2 ,
with σ1 and σ2 ∈ R.
Remark 4.1.
In the negative example which will be analyzed later in this subsection, we have
γ1 = 0, γ2 = 0 and γ3 = 0.
Note also that if γ1 = 0, the condition |γ2 | + |γ3 | > 0 is necessary for having the
Trèves condition satisfied.
7
4.2. A maximally hypoelliptic case.

If γ3 = 0, then we have
1 2
(4.2) |γ1 s2 + 2γ2 st + γ3 t2 | ≥ t − C 0 s2 ,
C0
and using the boundedness of dn ∂t B(sn , tn , ξn , ηn ) and dn ∂s B(sn , tn , ξn , ηn ), we
obtain that dn t2n ηn and dn s2n ξn are bounded. This implies that:
limn→+∞ dn s2n ηn = 0 and limn→+∞ dn sn tn ηn = 0.
We introduce ν = limn→+∞ dn t2n ηn .
Moreover, because d3n ηn tends to 0, this implies that σ2 = 0. Then the limiting
polynomials are necessarily in the form
1 (3) 1 (3)
P2 (s, t) = 111 (s − s0 )3 + νγ2 s + νγ3 t + 111 s30 .
3 3
(3)
If 111 = 0, these (non trivial) polynomials cannot have a local maximum. This is
the same if ν = 0, because γ3 ν = 0. If ν = 0, the limiting polynomial is necessarily
1 (3)
3 111 ((s − s0 ) + s0 ) which cannot have a maximum.
3 3
Hence we have obtained

Proposition 4.2.
If B is defined by (4.1) with γ3 = 0, then the system (1.10) is microlocally maximal
hypoelliptic at (0, 0, 0, 0 ; 0, 0, 1, 0) .
4.3. A negative example, with homogeneity of degree 3.
We take γ3 = 0 and γ2 = 0 in (4.1).
Proposition 4.3.
With B defined by (4.1) with γ3 = 0 and γ2 = 0, the system (1.10) is not microlo-
cally maximally hypoelliptic at (0, 0, 0, 0 ; 0, 0, 1, 0).
Proof
By the Helffer-Nourrigat criterion in the form expressed in Nourrigat’s lecture
notes (Th. 1, p. 118 and the remarks of p. 123 showing the equivalence of (ii)
with (iii)) (see also Helffer-Nier Lecture notes), it is enough to find a sequence
(sn , tn , ξn , ηn , dn ) and a non zero polynomial function such that
• limn→+∞ (sn , tn ) = 0 ,
• limn→+∞ |ξn |2 + |ηn |2 = +∞ , limn→+∞ ξn /( |ξn |2 + |ηn |2 = 1 ,
• limn→+∞ dn = 0
•
lim d|α|
n ∂s,t B(tn , sn , ξn , ηn ) = ∂s,t P (0, 0) , if 1 ≤ |α| ≤ 3 ,
α α
n→+∞
• P admits a local maximum at (0, 0).

We now look for a sequence in the form
1
+ρ 1
+ − 5 −ρ − 5 −δ
sn = −dn2 , tn = dn2 , ξn = −γ2 dn 2 , η n = dn 2 ,
with
1 1
0<ρ< , δ < ρ , δ < , ρ + − − δ > 0 , ρ = 2 − δ .
2 2
We note first that with this condition, we get
lim d3n ξn = 0 = lim d3n ηn = 0 .
n→+∞ n→+∞
22
Hence the polynomial function we are looking at is of degree 2.

Under the above conditions on ρ, , δ, we obtain as n → +∞

lim dn ∂s B(sn , tn , ξn , ηn ) = lim dn s2n ξn + γ1 tn sn dn ηn + γ2 t2n dn ηn = 0;
2
lim dn ∂t B(sn , tn , ξn , ηn ) = lim(γ 1 s2n + γ2 sn tn )d2 n ηn = 0;
2 2
lim dn ∂ss B(sn , tn , ξn , ηn ) = lim 2dn sn ξn + γ1 dn ηn = −1 ;
lim d2n ∂st
2
B(sn , tn , ξn , ηn ) = lim d2n (γ1 sn + γ2 tn )ηn = 0;
lim dn ∂tt B(sn , tn , ξn , ηn ) = lim γ2 d2n sn ηn
2 2
= 0.
For the first line above, we have used that
lim γ1 tn sn ηn = 0 .
n→+∞
We can choose
1 1 7
δ= ,= ,ρ= .
16 4 16
Hence we obtain
P (s, t) = −s2 ,
which admits a maximum at s = 0.
Remark 4.4.
Note that all these systems satisfy Trèves’s condition microlocally. We do not know
if this model is hypoelliptic. One should for proving this property analyze if some
Maire’s condition is satisfied.
This condition was automatically satisfied when B is analytic and when we have
only the x variable (no y). As presented in Journé-Trépreau [8], this result is due
to Maire and Journé-Trépreau give a proof of an improvment of this theorem (see
their Theorem 1.2 and its proof in section 2 of [8]), which gives a minimal control
of the loss of derivatives.
4.4. Conclusion for the homogeneous case with k = 3.
When γ1 = γ2 = γ3 = 0, it is clear that it is microlocally maximal hypoelliptic
and we have already mentioned that when γ1 = 0, Trèves condition implies the
inequality |γ2 | + |γ3 | > 0. To recapitulate, we have obtained :
Theorem 4.5.
With B defined by (4.1), the system (1.10) is microlocally maximally hypoelliptic
at (0, 0, 0, 0 ; 0, 0, 1, 0) if γ3 = 0 or if γ1 = γ2 = γ3 = 0.
It is not microlocally maximally hypoelliptic at (0, 0, 0, 0 ; 0, 0, 1, 0) if γ3 = 0 and
γ2 = 0 are satisfied, or if γ2 = γ3 = 0 and γ1 = 0 are satisfied.
Remark 4.6.
It remains to analyze γ2 = γ3 = 0 and γ1 = 0. But Trèves condition is not satisfied.
5. Homogeneous examples of degree 4

We take
1 4 γ4
(5.1) B(s, t, ξ, η) = s ξ + (γ1 s3 t + γ2 s2 t2 + γ3 st3 + t4 )η .
4 4
9
5.1. γ4 = γ2 = γ1 = 0 and γ3 = 0.
We would like to analyze the microlocal maximal hypoellipticity at (0, 0, 0, 0 ; 0, 0, 1, 0).
Proposition 5.1.
If B is defined by (5.1), with γ4 = γ2 = γ1 = 0 and γ3 = 0, then the system (1.10)
is maximally hypoelliptic at (0, 0, 0, 0 ; 0, 0, 1, 0).
Proof
We compute the various quantities appearing in Helffer-Nourrigat’s criterion. We
get (we only write the non zero terms)
∂ s B = s 3 ξ + γ3 t3 η , ∂t B = 3γ3 st2 η ,
2
∂ss B = 3s2 ξ , 2
∂st B = 3γ3 t2 η , 2
∂tt B = 6γ3 stη ,
3 3 3
∂sss B = 6sξ , ∂stt B = 6γ3 tη , ∂ttt B = 6γ3 sη ,
4 4
∂ssss B = 6ξ , ∂sttt B = 6γ3 η .
The proof is a consequence of various observations for a sequence satisfying the
above properties.
(1) We have always
(5.2) lim d4n ∂sttt
4
B(sn , tn , ξn , ηn ) = 6 lim d4n ηn = 0
n→+∞ n→+∞
This implies that the only term of degree 4 in the limiting polynomial
1 (4)
function should be 24 1111 s4 .
2 4 3
(2) Observing that 6(∂st B) (∂sttt B) = (∂stt B)2 , we get
(5.3) lim d3n ∂stt
3
B(sn , tn , ξn , ηn ) = 6γ3 lim (d3n tn ηn ) = 0 .
n→+∞ n→+∞
2
(3) Observing that 3
(∂ttt B)2 = 2γ32 ηξ2 2
∂ss B 4
∂ssss B, we get
(5.4) lim d3n ∂ttt
3
B(sn , tn , ξn , ηn ) = 6 lim (d3n sn ηn ) = 0 .
n→+∞ n→+∞
The two last observations imply that the only term of degree 3 in the
(3)
limiting polynomial function should be 16 111 s3 .
2 2
(4) Observing that 2(∂tt B) = ∂t B ∂ttt B and (5.4), we obtain
(5.5) lim d2n ∂tt
2
B(sn , tn , ξn , ηn ) = 6 lim (d2n sn tn ηn ) = 0 .
n→+∞ n→+∞
This implies that the only terms of degree 2 in the limiting polynomial
(2) (2)
function should be 12 11 s2 and 12 st.
(4)
We now distinguish two cases depending on the condition that 1111 = 0 or not.
(4)
Case 1 : 1111 = 0.
Then observing that
2 4 3
(5.6) 2∂ss B ∂ssss B = (∂sss B)2 ,
we get
(5.7) lim d3n ∂sss
3
B(sn , tn , ξn , ηn ) = 24 lim (d3n sn ξn ) = 0 .
n→+∞ n→+∞
This implies that in this case our limiting polynomial function should be of degree
2:
1 (2) 2 (2)
(5.8) s + 12 st + α1 s + α2 t ,
2 11
24
2
with in addition a sign condition due to the positivity of ∂ss B:
(2)
11 ≥ 0 .
But there are no non trivial polynomial function of this form admitting a local
maximum at 0.
(4)
Case 2 : 1111 = 0.
We first observe that in this case, we have
(4)
(5.9) lim ξn d4n = 1111 > 0
n→+∞
and the limiting polynomial function takes the form

1 (4) (2) 1 (4) 4
(5.10) (s − s0 )4 + 12 st + α1 s + α2 t − s .
24 1111 24 1111 0
Here s0 is determined by using (5.9) and the identity (5.6). When α2 = 0, (0, 0) is
not a critical point of the polynomial and we are done.
We now observe that
(2)
(5.11) 12 = 0 .
We indeed observe that, using (5.9) and the properties of our sequence, we have
dn sn t2n ηn bounded and d3n sn tn ξn bounded and observing that ηξnn tends to 0, this
gives (5.11).
It remains to control that no non trivial polynomial in the form

1 (4) 1 (4)
(s − s0 )4 + α1 s − 1111 s40 ,
24 1111 24
can have a local maximum at (0, 0). This is easy having in mind (5.9).
(1)
We first see that it is clear if 2 = 0. If this term vanishes, it remains to analyze
if the polynomial
1 (4) 1 (4)
s → (s − s0 )4 + α1 s − 1111 s40 ,
24 1111 24
can have a local maximum at 0.
(4)
This is again clear, if the derivative at 0 : − 16 1111 s30 + α2 is not zero. It remains
the case when 0 is a critical point of the polynomial at 0. But in this case, the
polynomial is convex and has a unique minimum at 0.
5.2. γ4 = 0, γ1 = 0 and γ3 = 0.
Proposition 5.2.
If B is defined by (5.1) with γ4 = 0, γ1 = 0 and γ3 = 0, the system (1.10) is
maximally hypoelliptic at (0, 0, 0, 0 ; 0, 0, 1, 0).
Proof
∂s B = s3 ξ + (2γ2 st2 + γ3 t3 )η , ∂t B = (2γ2 s2 t + 3γ3 st2 )η ,
2
∂ss B = 3s2 ξ + 2γ2 t2 η , 2
∂st B = (4γ2 st + 3γ3 t2 )η , 2
∂tt B = (2γ2 s2 + 6γ3 st)η ,
3 3 3
∂sss B = 6sξ , ∂stt B = (4γ2 s + 6γ3 t)η , ∂ttt B = 6γ3 sη ,
4 4 4
∂ssss B = 6ξ , ∂sstt B = 4γ2 η , ∂sttt B = 6γ3 η .
11

above properties. We only mention the new points or new difficulties.
(1) We have always
(5.12) lim d4n ∂sttt

4
B(sn , tn , ξn , ηn ) = 6 lim d4n ηn = 0
n→+∞ n→+∞
This implies that the only term of degree 4 in the limiting polynomial
1 (4)
function should be 24 1111 s4 .
(2) We control dn tn ηn and d2n s2n ξn . We can introduce
2 2
δ1 = lim d2n s2n ξn

n→+∞
and
δ2 = lim d2n t2n ηn .
n→+∞
(3) We then deduce that the only present monomial of degree 3 is s3 .

(4) The limiting polynomials have the structure
1 (4) 1 (4)
(s − s0 )4 + δ2 γ2 s2 + δ2 γ3 st + α1 s + α2 t − 1111 s40 .
4! 1111 4!
May be there are relations between α2 and δ2 , but we will not need it.
(4)
We now distinguish two cases depending on the condition that 1111 = 0 or not.
(4)
Case 1 : 1111 = 0.
In this case, the limiting polynomials have the structure
(δ1 + δ2 γ2 )s2 + δ2 γ3 st + α1 s + α2 t .
with δ1 ≥ 0.
But there are no non trivial polynomial function of this form admitting a local
maximum at 0. We can first exclude the case when α1 or α2 is not zero. Then it
remains to consider :
s ((δ1 + δ2 γ2 )s + δ2 γ3 t)
which is easy to control (one can discuss separately δ2 = 0 and δ2 = 0).
(4)
Case 2 : 1111 = 0.
After eliminating the polynomials for whch (0, 0) is not a critical point, it
remaains to analyze the limiting polynomial function takes the form
1 (4) 4
(5.13) (s − 4s0 s3 + 6s20 s2 ) + δ2 γ2 s2 + δ2 γ3 st .
24 1111
We look at the quadratic term
(5.14) (s, t) → s ((δ1 + δ2 γ2 )s + δ2 γ3 t) ,
which is easy to control except if δ2 = δ1 = 0. If δ1 = δ2 = 0, we get simply

1 (4) 4
24 1111 s which has no local maximum at (0, 0).
26
5.3. γ4 = 0, and γ3 = 0.
Proposition 5.3.
If B is defined by (5.1) with γ4 = 0 and γ3 = 0, then the system (1.10) is maximally
hypoelliptic at (0, 0, 0, 0 ; 0, 0, 1, 0).
Proof
∂s B = s3 ξ + (3γ1 ts2 + 2γ2 st2 + γ3 t3 )η , ∂t B = (γ1 s3 + 2γ2 ts2 + 3γ3 t2 s)η ,
2
∂ss B = 3s2 ξ + (6γ1 ts + 2γ2 t2 )η , ∂st
2
B = (3γ1 s2 + 4γ2 st + 3γ3 t2 )η ,
2 2
∂tt B = (2γ2 s + 6γ3 ts)η
3 3
∂sss B = 6sξ + 6γ1 tη , ∂sst B = (6γ1 s + 4γ2 t)η ,
3 3
∂stt B = (4γ2 s + 6γ3 t)η , ∂ttt B = 6γ3 sη ,
4 4 4 4 4
∂ssss B = 6ξ , ∂ssst B = 6γ1 η , ∂sstt B = 4γ2 η , ∂sttt B = 6γ3 η , ∂tttt B = 0.
above properties. We only mention the new points or new difficulties.
(1) The only term of degree 4 in the limiting polynomial function should be
1 (4) 4
24 1111 s .
(2) We control d2n t2n ηn and d2n s2n ξn . We can introduce
δ1 = lim d2n s2n ξn
n→+∞
and
δ2 = lim d2n t2n ηn .
n→+∞
(3) We then deduce that the only present monomial of degree 3 is s3 .

(4) The limiting polynomials have the structure
1 (4) 1 (4)
(s − s0 )4 + δ2 γ2 s2 + δ2 γ3 st + α1 s + α2 t − 1111 s40 .
4! 1111 4!
Hence γ1 is not present in the limiting polynomial except in the first order terms.
Hence the proof is the same as when γ1 = 0.
5.4. The case γ4 = 0.
In this case we have:
∂s B = s3 ξ + (3γ1 s2 t + 2γ2 st2 + γ3 t3 )η , ∂t B = (γ1 s3 + 2γ2 s2 t + 3γ3 st2 + γ4 t3 )η ,
2
∂ss B = 3s2 ξ + (6γ1 st + 2γ2 t2 )η , ∂st
2
B = (3γ1 s2 + 4γ2 st + 3γ3 t2 )η ,
2 2 2
∂tt B = (2γ2 s + 6γ3 st + 3γ4 t )η
3 3
∂sss B = 6sξ + 6γ1 tη , ∂sst B = (6γ1 s + 4γ2 t)η ,
3 3
∂stt B = (4γ2 s + 6γ3 t)η , ∂ttt B = (6γ3 s + 6γ4 t)η ,
4 4 4 4 4
∂ssss B = 6ξ , ∂ssst B = 6γ1 η , ∂sstt B = 4γ2 η , ∂sttt B = 6γ3 η , ∂tttt B = 6γ4 η .
When looking at all the possible limits, we get by considering the terms relative to
2 2
∂ss B and ∂tt B that d2n t2n ηn and d2n s2n ξn are bounded. Moreover, the limits exist.
We write δ2 = limn→+∞ d2n t2n ηn and δ1 = limn→+∞ d2n s2n ξn and in the first case :
41111 = limn→+∞ d4n ξn and s0 = limn→+∞ d3n sn ξn , any limiting polynomial has the
form
1 4 1
(s − s0 )4 + γ2 δ2 s2 + 3stγ3 δ2 + γ4 t2 δ2 + α1 s + α2 t − 41111 s40 .
4! 1111 4!
13
with 41111 > 0 or

1 3
(δ1 + δ2 γ2 )s2 + stγ3 δ2 + γ4 t2 δ2 + α1 s + α2 t
2 2
with δ1 ≥ 0.
When γ4 = 0, dn ∂t B(sn , tn , ξn , ηn ) and dn ∂t B(sn , tn , ξn , ηn ) are bounded. This

implies, using that ηξnn tends to 0, that that dn s3n ξn and dn t3n ηn are bounded, which
enough for proving that δ2 = 0 by convexity.
Hence we simply obtain :
1 4 1
(s − s0 )4 + α1 s + α2 t − 41111 s40 .
4! 1111 4!
with 41111 > 0 or
1 2
δ1 s + α1 s + α2 t
2
with δ1 ≥ 0.
No one of these polynomials can have a local maximum.

Proposition 5.4.
With B given in (5.1) and γ4 = 0, then the system (1.10) is microlocally maximally
hypoelliptic at (0, 0, 0, 0; 0, 0, 1, 0).
5.5. The case γ4 = γ3 = γ2 = 0 and γ1 = 0.

Question 5.5.
If B is defined by (5.1) with γ4 = γ3 = γ2 = 0 and γ1 = 0, is the system (1.10)
maximally hypoelliptic at (0, 0, 0, 0 ; 0, 0, 1, 0) ?
Discussion
∂s B = s3 ξ + 3γ1 ts2 η , ∂t B = γ1 s3 η ,
2
∂ss B = 3s2 ξ + 6γ1 tsη , ∂st2
B = 3γ1 s2 η , ∂tt
2
B = 0,
3 3 3 3
∂sss B = 6sξ + 6γ1 tη , ∂sst B = 6γ1 sη , ∂stt B = 0 , ∂ttt B = 0,
4 4 4 4 4
∂ssss B = 6ξ , ∂ssst B = 6γ1 η , ∂sstt B = 0 , ∂sttt B = 0 , ∂tttt B = 0.
The limiting polynomial seems to have the following structure
1 (4) 4 1 (3) 1 (3)
s + 111 s3 + 11 s2 + α1 s + α2 t ,
4! 1111 3! 2
(4)
with 1111 ≥ 0.
We have no control of the other signs. Hence it is not excluded that one can
contradict maximal hypoellipticity.
Let us follow the idea which was successful for the counterexample for homogeneous
polynomials of degree 3. We consider a sequence satisfying
sn ξn + 3γ1 tn ηn = 0 .
28
and
3 1 1 −3
sn = dn4 , tn = 2 dn4 , ηn = 1 d .
γ1 n
This implies
−7
ξn = −31 2 dn 2 .
With this choice, we obtain :
d2n ∂ss
2
B(sn , tn , ξn , ηn ) = −31 2
and we have a problem, because looking at our assumption on ξn we should have
−1 2 > 0. !! So the most natural research of counterexample goes nowhere.
We will see in Proposition 9.1 that this example is actually subelliptic.
5.6. Higher homogeneity.
maximal hypoellipticity holds actually for any k ≥ 2 for the system (1.10) associated
with k
1 k 1
k−
B(s, t, ξ, η) = s ξ + γ t s η,
k! !
=1
with γk = 0. The only limiting polynomials have indeed the form
1
(s − s0 )k + α1 s + α2 t ,
k!
or
δ1 s2 + α1 s + α2 t ,
with δ1 ≥ 0.
6. More on maximal hypoellipticity

We come back to the case when m = 1 (which was considered in the previous
papers) and analyze maximal hypoellipticity in the homogeneous case of degree r.
This analysis was initiated in [4].
Let us consider a sequence such that

(6.1) lim d|α|
n ρn
r−|α| (α)
B (ωn )ξn = P (α) (0) .
n→+∞
Here we have supposed that B is homogeneous of degree r and that xn = ρn ωn ,

with ωn ∈ Sd−1 . Without loss of generality we can assume that limn→+∞ ωn = ω∞
and we recall that limn→+∞ ρn = 0.
Under Hörmander’s condition, we have
lim drn ξn = μr ,
n→+∞
and for the non degenerate case we assume that μr > 0.

Note that we have the relation
μr B (α) (ω∞ ) = P (α) (0) , for |α| = r .
Let us assume first that

|B (α) (ω∞ )| = 0 .
1≤|α|≤r−1
Then we obtain that

ρn
0 < σ∞ := lim < +∞ ,
n→+∞ dn
15
and
r−|α| (α)
μr σ ∞ B (ω∞ ) = P α (0) .
If B is an homogeneous polynomial of degree r, we obtain
−r
P (t) = μr σ∞ (B(ω∞ + σ∞ t) − B(ω∞ )) .
If B satisfies the Trèves condition globally, this is the same for P !
It remains to analyze the case when
B (α) (ω∞ ) = 0 , ∀α with 1 ≤ |α| ≤ r − 1 .
We rewrite ωn = ω∞ + wn ηn , with wn ∈ R+ and ηn ∈ Sd−1 . Without loss of
singularity we can assume that limn→+∞ ηn = η∞ and that limn→+∞ wn = 0 .
Then we can rewrite (6.1) in the form :
ρn
μr lim (dn /ρn )r (B(ωn + t) − B(ωn )) = P (t)
n→+∞ dn
We rewrite B(ωn + dρnn t) − B(ωn ) in the form

(dn /ρn )r B(ωn + ρdnn t) − B(ωn )

= (dn /ρn )r B(ω∞ + (ωn − ω∞ ) + dρnn t) − B(ω∞ + (ωn − ω∞ ))

α
α
= |α|=r B (α) (ω∞ ) − ω∞ ) + t − |α|=r B (α) (ω∞ ) dρnn (ωn − ω∞ )
dn
ρn (ωn

α
α
= |α|=r B (α) (ω∞ ) wρnndn ηn + t − |α|=r B (α) (ω∞ ) wρnndn ηn .
Taking the limit, we obtain, with σ∞ = limn→+∞ wρnndn ,

α α
P (t) = μr |α|=r B (α) (ω∞ ) (σ∞ η∞ + t) − |α|=r B (α) (ω∞ ) (σ∞ η∞ )
= μr (B(ω∞ + σ∞ η∞ + t) − B(ω∞ + σ∞ η∞ )) .
Again P satisfies Trèves condition if B satisfies Trèves condition.
Degenerate situation.
We now assume that limn→+∞ drn ξn = 0.
We assume that ω∞ satisfies for some r̃ ≤ r, B (α) (ω∞ ) = 0 for all α, such that
|α| < r̃ and that r̃ is maximal with this property.
1. r̃ = 1.
Let us treat the “elliptic” case corrresponding to

(6.2) ∇B(ω∞ ) = 0 .
In this case we have dn ρr−1
bounded and drn ξn tending to zero. It is then im-
n ξn
n ξn = 0 for 2 ≤ < r. Hence the limiting
mediate to see that limn→+∞ dn ρr−
polynomial is of degree 1 and cannot have a local maximum.
2. r̃ = 2.
In the same way, we get that the limiting polynomial is of degree 2.
We observe that by Euler formula for an homogeneous function :
(6.3) B (ω∞ ) · ω∞ = 0 .
where B denotes the Hessian. We assume that B (ω∞ ) restricted to (Rω∞ )⊥ is
non degenerate and has at least one strictly positive eigenvalue. Then no limiting
polynomial can have a local maximum.
30
The philosophy is that Trèves condition gives immediately the control of the
non-degenerate representations in the sense of C. Rockland (see [5] for this no-
tion). But maximal hypoellipticity can fail if the criterion is not satisfied for some
degenerate representation.
17
Part 2. Subellipticity
7. General discussion
7.1. Derridj’s subellipticity criterion.
We write the criterion in the case “n + m” (this will then applied for n = 2 and
m = 2). The variables are denoted by x (x ∈ Rm ) and s (s ∈ Ω ⊂ Rn ) and the
dual variables in this section by ξ and σ. We look at the system
(7.1) ∇s + i∇s B(s) · ∇x ,

associated with B(s, ξ) = j Bj (s) · ξj . Assuming that 0 ∈ Ω and that Bj (0) =
∇Bj (0) = 0 (j = 1, . . . , m) , we consider the microlocal subellipticity of the system
at the point1 (0, 0; ξ0 , 0) in Rn+m × (Rn+m \ {ξ = σ = 0}).
Assumption 7.1. (H(α, ξ0 ))
There exist a conic neighborhood V of ξ0 in Rm \ {0}, a neighborhood ω of 0 in Ω
and constants C and α such that for any ξ ∈ V , there exists ω̃ξ of full measure in
ω and a map γ
ω̃ξ × [0, 1] (s, τ ) → γ(ξ, s, τ ) ∈ Ω
s. t., writing γξ (·) = γ(ξ, ·), we have
• γξ (s, 0) = s , γξ (s, 1) ∈ / ω, ∀s ∈ ω̃ξ
• γξ is of class C 1 outside a negligeable set Eξ ⊂ ω̃ξ × [0, 1], and
(1) |∂τ γξ (s, τ )| ≤ C , ∀(s, τ ) ∈ ω̃ξ × [0, 1] \ Eξ ,
(2) |det(Ds γξ )| ≥ C1 , ∀(s, τ ) ∈ ω̃ξ × [0, 1] \ Eξ ,
(3) B(γ(ξ, s, τ )) − B(s) ≥ C1 τ α , ∀(s, τ ) ∈ ω̃ξ × [0, 1]
• γ is measurable.
Theorem 7.2.
1
If φ satisfies (H(α, ξ0 )), then the associated system (7.1) is microlocally α -subelliptic
at ((0, 0); (ξ0 , 0)).
In order to verify the assumptions of Theorem 7.2, we have to control the
variation of B in different sectors. Hence, there is a “second” localization associated
with different directions (associated with points on the circle S1 ) . We choose to
analyze the situation close to (0, 1) in S1 .
7.2. Description of the possible types of sectors.
With B defined in (1.11), and B1 and B2 homogeneous of degree r (with r ≥ 2),
we consider
(7.2) B (s, t) = ξ −1 B(s, t, ξ, η) ,
with
η
(7.3) = .
ξ
We then introduce φ by

(7.4) φ (s) = B (s, 1 − s2 ) .
The analysis in the neighborhood of the direction (0, 1) leads us to distinguish
between six types of sectors and to define the rules followed by the escaping rays
when starting from, or going through the considered (family of -dependent) sec-
tor(s). We denote these sectors by V (or V to mention their dependence on ) and
1actually the ray associated with the point
32
SV denotes the arc S ∩ V . A point on the circle is most of the time parametrized by
s. All these sectors are contained in a sufficiently small (independently of ) open
sector containing (0, 1).
Type A+ : φ is positive, monotone and attains its maximum at one end s() of SV .
Moreover, there exists c0 > 0 such that |s()| ≥ c0 .
Type A− : φ is negative, monotone and attains its minimum at one end s() of SV .
Moreover, there exists c0 > 0 such that |s()| ≥ c0 .
Type B + : φ is positive and monotone on SV .
Type B − : φ is negative, monotone on SV .
Type C + : We have
(7.5) φ ≥ c0 || and |φ | ≤ C0 ||
and the arc is contained in ((−C||

ω ), (C||ω )).
−
Type C : We have
(7.6) φ < −c0 || and |φ | ≤ C0 ||
and the arc is contained in ((−C||

ω ), (C||ω )) for some ω > 0.
Remark 7.3.
In fact, what we actually need is (immediate consequence of (7.5) and (7.6)) the
existence of C such that, in V ,
|φ | ≤ ±Cφ
with ±φ > 0 .
7.3. Description of the escaping rays.
As in our previous works there is an initial (sufficiently small) neighborhood of
(0, 0) in R2 which will be denoted by D. We will describe how points in D can
escape of D along broken straight lines. More precisely for points escaping from
D ∩ V , we will explain how they can either leave D or attain a neighboring sector.
But there is at the end a global constraint that they should escape from D. Our
escaping rays should be continuous, hence we can be can be vague for the rules
concerning points at the boundary of the sectors.
Type A+ : The rays are outgoing and parallel to the ray corresponding to the maxi-
mum of φ on SV .
Type A− : The rays are ingoing and parallel to the ray corresponding to the minimum
of φ on SV . They will necessarily touch the ray corresponding to the
maximum of φ on SV .
Type B + : The rays are outgoing and parallel to ( = 0. Note that σ
σ ) with σ is
independent of and that 0 < |
σ | < σ. Moreoever
(7.7) φ σ
≥ 0 on SV .
Type C + The rays are outgoing and parallel to ( is independent of
σ ). Note that σ
and that 0 < |
σ | < σ.
Type B − : The rays are ingoing and parallel to (
σ ). Moreover σ should satisfy
(7.8) φ σ
≤ 0 on SV .
Type C − : The rays are ingoing and parallel to (
σ ).
Remark 7.4.
In the case of Type C + or Type C − , the sign of σ is not given and determined by
global considerations : one should finally escape from D.
19
In the case of type B + , one will either leave directly D inside V or touch a neigh-
boring sector.
Once we have covered a conic neighborhood of (0, 1) by this family of sectors, we
have to verify that we can escape from any point of D (intersected with this cone).
This is not always possible and this possibility is not a consequence of Trèves con-
dition.
7.4. Control of φ .
7.4.1. Preliminaries. We refer to [3] for the general scheme of the proof and we
will restart at the level of the proof of Lemma 4.2. Hence we focus on the control
from below of B (s(τ ), t(τ )) − B (s, t), whose proof is based on the decomposition
(see formula (4.6) in [3]) :
(7.9) B (s(τ ), t(τ )) − B (s, t) = (I) + (II) ,
with
(I) := ρ(τ )m (φ (s̃(τ )) − φ (s̃(0)))
(7.10)
(II) := φ (s̃(0))(ρ(τ )m − ρm ) .
Here we recall that
s(τ )
ρ(τ )2 = s(τ )2 + t(τ )2 , s̃(τ ) = , ρ = ρ(0) ,
ρ(τ )
where (s(τ ), t(τ )) denotes the coordinates of the curve γ(s, t, τ ) starting from (s, t)
at time 0.
Our goal is to control from below, uniformly with respect to , the variation of
B along the escaping rays, i.e. the expression (I) + (II) in the different cases.
7.4.2. Type A+ . We observe that (II) is positive (ρ(τ ) is increasing). It remains
to get a lower bound of (I) by c0 τ sup(m,k) , with c0 independent of and with k the
order of the zero of φ=0 ).
For this we can apply our uniform lower bound lemma A.1 to φ(s, ) = φ (s)
together with what we did in the proof of Lemma 4.6 in [3]. We also use the lower
bound of s̃(τ ) − s̃(0) :
τ
(7.11) |s̃(τ ) − s̃(0)| ≥ C(σ) ,
ρ(τ )
with C(σ) > 0 and σ is the point of maximum of φ on the arc SV which is by
assumption uniformly away from the origin (as assumed in our definition of type
[A+ ]. This estimate is established in [2] (Formula (5.9)) and results from the fact
that our ray is not parallel to the ray (0). Hence we get for (I) the lower bound
(7.12) (I) ≥ c0 ρ(τ )m−k τ k ,
for some c0 > 0.
We now observe that for (s(τ ), t(τ )) = (s + cτ, t + dτ )
(7.13) ρ(τ )2 − ρ2 = τ (τ + 2sc + 2td) .
Hence we get that
(7.14) ρ(τ ) ≥ τ if st + cd ≥ 0 .
The conclusion is then easy by separating the two cases k ≤ m and k > m. When
k ≤ m, we use indeed the inequality (7.14) and when k > m, we use that ρ(τ ) is
uniformly bounded by the radius of D.
34
7.4.3. Type B + . The proof is the same replacing σ by σ . The important point
is independent of .
is that σ
7.4.4. Type C + . The problem is that (I) is not positive. Hence we have to find
a lower bound for (II) and (s, t, τ ) s.t. (s, t) ∈ V and γ(s, t, τ ) ∈ V ,
(II) ≥ cτ m ,
and show that |(I)| ≤ 12 (II).
We start from
m−1
ρ(τ )m − ρm = (ρ(τ ) − ρ)( j=0 ρ(τ )m−1−j ρj )
(7.15) )2 −ρ2 ) m−1
= (ρ(τ
ρ(τ )+ρ ( j=0 ρ(τ )
m−1−j j
ρ )
We now use (7.13) together with the observation that sc + td = ρ(s̃, t̃) , (c, d) and
that (s̃, t̃) , (c, d) is close to one because SV is close to (0, 1). Hence we get
(7.16) ρ(τ )2 − ρ2 ≥ τ (τ + ρ) .
Noting that by the triangular inequality
(7.17) τ + ρ ≥ ρ(τ ) ,
this leads to
ρ(τ )m − ρm ≥ τ 2ρ(τ
τ +ρ
) ρ(τ )
m−1
(7.18)
= 12 τ ρ(τ )m−1 .
Hence we obtain coming back to the definition of (II) and using the lower bound
away of 0 of |φ | :
||
(7.19) (II) ≥
|ϕ(0)|τ ρ(τ )m−1 .
4
On the other hand, using that on SV we have, using (8.4) and (8.6),
(7.20) |φ (s̃)| ≤ C|| ,
we obtain
|(I)| ≤ C1 |||s̃(τ ) − s̃(0)|ρ(τ )m .
We then observe2 that
τ
|s̃(τ ) − s̃(0)| ≤ δ(σ)
ρ(τ )
with
lim δ(σ) = 0 .
σ→0
Hence, we get :
(7.21) |(I)| ≤ C1 δ(σ)||τ ρ(τ )m−1 .
Hence we can choose σ small enough such that
1
(7.22) |(I)| ≤
(II) .
2
It remains to find a lower bound for (II). Having in mind that
(7.23) (SV ) ≤ C2 ||ω ,
2See formula (3.32) for s̃ in [2]

21
where (SV ) denotes the length of the arc SV , we observe3 that, if γ(s, t, τ ) remains
in V , we have
τ ≤ C3 ||ω ρ(τ ) ,
i.e
τ 1
|| ≥ c4 ( )ω .
ρ(τ )
From this we obtain,
τ 1 1 1
(II) ≥ c0 ||τ ρ(τ )m−1 ( ) ω = c5 τ 1+ ω ρ(τ )m−1− ω .
ρ(τ )
Having in mind (7.14) and noting that ω ≤ 1, we obtain (we recall that m ≥ 2)
(7.24) (II) ≥ c6 τ m .
So we have effectively obtained
1
(7.25) (I) + (II) ≥
c6 τ m .
2
Note that the estimate is independent of the vanishing order k.
7.4.5. Types A− , B − and C − . This is the same. We have just to replace in
the proofs φ by −φ .
7.5. Control of the Jacobian.
We refer to the discussion for the case of homogeneity 3 and to the last paragraph
of Subsection 8.3, where the general proof is sketched.
8. On subellipticity : Homogeneous case of degree 3

Our motivation is to understand the subellipticity for the system (1.10) with
γ3
(8.1) B(s, t, ξ, η) = s3 ξ + (γ1 s2 t + γ2 st2 + t3 )η .
3
When restricting, for ξ = 1, η = 0, (s, t) → B(s, t, 1, 0) to the circle, we get two
zeroes corresponding to s = 0, t = 1 and to s = 0, t = −1. For applying Theorem7.2,
we will perform a second localization attached to the decomposition√of the disk into
sectors. For the two sectors attached to s ∈] − σ, +σ[ and t = ± 1 − s2 , we will
use the techniques of the preceding papers (actually the rules (A+ ) or (A− ) will be
sufficient).
We can then concentrate the analysis on the conic neighborhoods (parametrized
by σ which should be small enough) of the points (0, 1) and (0, −1). Choosing
for example (0, 1), we are led to the analysis on the circle parametrized by s (t =
√
1 − s2 ) of the function
φ (s) = s3 + ϕ(s)
with
(8.2) ϕ(0) = 1
and
(8.3) = c̃η/ξ ,
(with c̃ = 0, corresponding to the condition γ3 = 0). The parameter runs in some
small (enough) interval containing 0.
3We have −C||ω ≤ s+cτ ≤ C||ω . If we observe that |s| ≤ C||ω ρ ≤ C||ω ρ(τ ). Hence we
ρ(τ )
get τ ≤ 2 C
c
||ω ρ(τ ) and we are done.
36
The proof below will treat a more general situation where the polynomial character
has disappeared.
8.1. Analysis of the graph of φ .
We first analyze the zeroes of φ (s) for s ∈ I := [−σ, σ], with σ > 0. σ and will
be chosen independently of small enough size (we will not mention systematically
repeat the “small enough” at each step of the proof).
We have
(8.4) φ (s) = 3s2 + ϕ (s) .
We first note that, because ϕ(0) = 0, φ (s) has a unique zero s0 () in I and this
zero satisfies :
1
(8.5) s0 () ∼ (−ϕ(0)) 3 .
It is clear that φ (s0 ()) > 0 and that φ has at most two zeroes in I. We can now
discuss different cases depending on the behavior of ϕ at the origin.
(1) ϕ (0) = 0.
If ϕ (s) ≥ 0, φ ≥ 0 and φ is monotonically increasing.
If ϕ (s) < 0, then φ has two zeroes s1 () and s2 () which corresponds
respectively to a local maximum or a local minimum. We have
s1 () ∼ −3− 2 || 2 |ϕ (0)| 2 , s2 () ∼ 3− 2 || 2 |ϕ (0)| 2
1 1 1 1 1 1
(8.6)
Note that we have, depending on the sign of ϕ(0),
(a) either
(8.7) −σ < s0 < s1 < 0 < s2 < σ
(b) or
(8.8) −σ < s1 < 0 < s2 < s0 < σ ,
and that
(8.9) φ (sj ) ∼ ϕ(0) .
Observing that
φ (s) = 6s + ϕ (s) ,
we get that
φ (sj ) ∼ 6sj ,
so s1 is a local maximum and s2 is a local minimum. This implies the
following control of φ between (s1 ) and (s2 ) :
(8.10) |φ (s)| ≤ C(|sj ()| + ||)|sj ()| ≤ C̃|| .
(2) ϕ (0) = 0, ϕ (0) = 0.
In this case, we can write ϕ = sψ1 with ψ1 (0) = 0 and we get
φ (s) = s(3s + ψ1 (s)) .
φ vanishes at 0 and at a point s1 () satisfying

(8.11) s1 () ∼ − ψ1 (0) ,
3
whose sign is determined by the sign of −ψ1 (0). Note that we have,
depending on the signs of ϕ(0) and ψ1 (0), four cases
23
(a) If ϕ(0) > 0 and ψ1 (0) < 0,

(8.12) −σ < s0 < 0 < s1 < σ .
(b) If ϕ(0) < 0 and ψ1 (0) > 0,
(8.13) −σ < s1 < 0 < s0 < σ .
(c) If ϕ(0) > 0 and ψ1 (0) > 0,
(8.14) −σ < s0 < s1 < 0 < σ .
(d) If ϕ(0) < 0 and ψ1 (0) < 0,
(8.15) −σ < 0 < s1 < s0 < σ .
In addition
(8.16) φ (s1 ) ∼ ϕ(0) .
Observing that
φ (s) = 6s + ϕ (s) ,
we get that
φ (s1 ) ∼ 3s1 () ,
so s1 is a local minimum in the first and forth cases and a local maximum
in the second and third cases.
Note also that
(8.17) |φ (s)| ≤ C(|s1 ()| + ||)|s1 ()| ≤ C̃2 .
(3) ϕ (0) = ϕ (0) = 0.
In this case, we can write ϕ (s) = s2 ψ2 (s) and we get
φ (s) = s2 (3 + ψ2 (s)) .
Hence 0 is the unique zero of φ in I and φ > 0 for s = 0. So φ is
monotonically increasing in I and crosses 0 at s0 ().
8.2. Description of the escaping rays.

8.2.1. Case (i).
Case (i)-a.
We come back to the disk and introduce the four sectors Sj (j = 1, . . . , 4) associated
with the segments on the circle ] − σ, s0 [, ]s0 , s1 [, ]s1 , s2 [ and ]s2 , σ[. We introduce a
on the circle which is independent of and satisfies : 0 < σ
point σ < σ (typically
we can choose σ = σ2 with σ small enough). For small enough we have :
−σ < s0 < s1 < 0 < s2 < σ
< σ.
We also note that −σ corresponds to a point of infimum of φ on [−σ, σ] and that
φ is increasing on ] − σ, s0 [.
For a point (s, t) in this first

√ sector we choose an ingoing ray starting from
(s, t), which is parallel to (−σ, 1 − σ 2 ), till the ray crosses the ray associated with
s0 (that is joining (0, 0) and (s0 , 1 − s20 )).
38
Notation. √
From now on, we denote by (−σ) the ray joining (0, 0) to (−σ, 1 − σ 2 ).
For a point (s, t) in the second sector (associated with the arc (s0 , s1 )) and the

third sector (associated with the arc (s1 , s2 )), we consider outgoing rays parallel to
(
σ ), till the ray meets a neighboring sector or the boundary of the disk.
Finally, for a point (s, t) in the fourth sector (associated with the arc (s 2 , σ) ),
we consider outgoing rays parallel to (σ).
Using our general considerations, we have just to verify that S1 is of type A− ,

S2 of type B + , S3 of type C + and S4 of type A+ . For short, we write that we have
the configuration A− B + C + A+ . Let us verify these assertions. The length of the
arcs appearing for S2 and S3 results from the asymptotics of s0 (), s1 () and s2 ()
established in (8.5) and (8.6). To verify that S3 is of type C + , we can use with
j = 2 (8.9) together with (8.10). The sign + of σ for S3 , which is the same as for
S2 permits to escape from D.
Case (i)(b).
In short, we have the configuration A− C − B − A+ . Coming back to the disk and
introduce the four sectors Sj (j = 1, . . . , 4) associated with the segments on the
circle ] − σ, s1 [, ]s1 , s2 [, ]s2 , s0 [ and ]s0 , σ[. We introduce a point σ on the circle
which is independent of and satisfies : −σ < σ < 0 (typically we can choose
= − σ2 with σ small enough). For small enough we have :
σ
−σ < σ
< s1 < 0 < s2 < s0 < σ .
We also note that σ corresponds to the maximum of φ on [−σ, σ] and that φ is

increasing on ] − σ, s1 [ and ]s2 , σ[.
For a point (s, t) in this first sector we choose an ingoing ray starting from
(s, t), which is parallel to(−σ), till the ray crosses the ray associated with s0 (that
is joining (0, 0) and (s0 , 1 − s20 ).
For a point (s, t) in the second sector (associated with the arc (s 1 , s2 )) and the
third sector (associated with the arc (s 2 , s0 )), we consider ingoing rays parallel to
(
σ ), till we meet another sector.
Finally, for a point (s, t) in the fourth sector (associated with the arc (s 0 , σ)),
we consider outgoing rays parallel to (σ).

S2 of type C − , S3 of type B − and S4 of type A+ . For short, we note A− B + C + A+ .
Let us verify these assertions. The length of the arcs appearing for S2 and S3
results from the asymptotics of s0 (), s1 () and s2 () established in (8.5) and (8.6).
To verify that S2 is of type C − , we can use with j = 1 (8.9) together with (8.10).
The sign − of σ for S2 , which is the same as for S3 permits to escape from D.
8.2.2. The case (ii).
Case (ii)-b.
In the case (ii)-b, we choose σ such that −σ < σ̂ < 0. Hence for small enough
we have −σ < σ < s1 () < 0 < s0 () < σ. We have to consider four arcs (−σ, s1 ),
25

(s
1 , 0), (0, s0 ) and (s0 , σ).
For a point (s, t) in the fourth sector, we consider outgoing rays parallel to (σ).
For a point (s, t) in the third sector we choose an ingoing ray starting from (s, t),
which is parallel to ( σ ), till the ray crosses the ray associated with (s0 ), we then
follow the rule of the fourth sector.
For a point (s, t) in the second sector we choose an ingoing ray starting from (s, t),
which is parallel to ( σ ), with −σ < σ̂ < 0, till the ray crosses the ray associated
with s = 0. We then follow the rules of the third and fourth sectors.
Finally, for a point (s, t) in the first sector we choose an ingoing ray starting from
(s, t), which is parallel to (−σ), till the ray crosses the ray associated with s1 and
then follow the rules of the other sectors.

S2 of type C − , S3 of type B − and S4 of type A+ . For short, we note A− B + C + A+ .
Let us verify these assertions. The length of the arcs appearing for S2 and S3 results
from the asymptotics of s0 () and s1 () established in (8.5) and (8.11). To verify
that S2 is of type C − , we can use (8.16) together with (8.17). The sign − of σ for
S2 , which is the same as for S3 permits to escape from D.
We now treat the three other cases which are actually simpler or analogous.
Case (ii)-a.
We have four sectors. In sector S1 , we consider ingoing rays parallel to (−σ). In
sectors S2 and S3 , we consider outgoing rays parallel to ( > 0. In sector
σ ) for σ
S4 , we consider outgoing rays parallel to (σ).

S2 of type B + , S3 of type C + and S4 of type A+ . For short, we note A− B + C + A+ .
Let us verify these assertions. The length of the arcs appearing for S2 and S3 results
from the asymptotics of s0 () and s1 () established in (8.5) and (8.11). To verify
that S2 is of type C − , we can use (8.16) together with (8.17). The sign + of σ for
S2 , which is the same as for S3 permits to escape from D.
Case (ii)-c.
This is the same as for case (ii) − a but we exchange the roles of (s1 ) and (0). We
get the configuration A− B + C + A+ .
Case (ii)-d.
This is the same as for case (ii) − b with exchange of s1 and 0. We get the config-
uration A− C − B − A+ .
Remark 8.1. For k = 3 the choice of σ is always determined by the rule that
we should escape by the right, if direct escaping in the sector is impossible.

8.2.3. The case (iii). We consider two arcs (−σ, s0 ()) and (s
0 (), σ). For (s, t)
in the second sector, we can take an outgoing ray parallel to (σ) and for (s, t) in
the first sector we can take an ingoing ray parallel to (−σ) till the ray crosses the
ray associated with (s0 ()) and then follow the rule of the second sector.Hence we
are simply in a configuration A− A+ .
8.3. Lower bound for the Jacobian. We illustrate the argument for the
case (i) − a, corresponding to a configuration A− B + C + A+ . Using the formulas
giving the Jacobian Dγ(s, t, τ ), it is enough to control the angles on the broken
segments when crossing the rays associated with (s0 ), (s1 ) and (s2 ). Because all
40
these angles are not zero and can be controlled by -independent lower bounds and
upper bounds. It is clear that we get a uniform positive lower bound for det(Dγ).
Here we refer to Appendix A in [2] (see Formula (A.15)). Let us illustrate what we
explained there on an example, let us assume that we start from a point (s, t) in
S1 such that the corresponding ray is broken when crossing (s0 ) and (s2 ). Then,
we have
• det(Dγ)(t, s, τ ) = 1 ,
if γ(t, s, τ ) ∈ S1 ,
• |det(Dγ)(t, s, τ )| = |Δ((−σ), (s0 ))|−1 |Δ((s0 ), ( σ ))| ,
if γ(t, s, τ ) ∈ S2 ∪ S3 ,
• |det(Dγ)(t, s, τ )|
= |Δ((−σ), (s0 ))|−1 |Δ((s0 ), ( σ ), (s2 ))|−1 |Δ((s2 )(σ))| ,
σ ))| |Δ((
if γ(t, s, τ ) ∈ S4 .
√
Here Δ((α), (β)) = α 1 − β 2 − β 1 − α2 and we observe that
lim→0 Δ((−σ), (s0 )) = Δ((−σ), (0)) , lim→0 Δ((s0 ), (

σ )) = Δ((0), (
σ )) ,
lim→0 Δ((
σ ), (s2 )) = Δ((
σ ), (0)) , lim→0 Δ((s2 ), (σ)) = Δ((0), (σ)) .
Because all these limits are not zero, we get for small enough uniform lower and
upper bounds for the Jacobian.
In all the remaining cases we have just to control the angles between the ingoing
broken line and the boundary rays of the sector V . This should be only controlled
when the line is effectively broken.Because all the crossed boundary rays tend to
the ray (0), the property results of the fact that σ and σ are not 0. Because this is
the rule which is followed when defining the rays for each type of sector, the proof
works in full generality.
8.4. Degenerate cases. We complete the discussion by considering the case

when (8.2) is not satisfied. When
φ (s) = s3 + sϕ(s) ,
with ϕ(0) = 0, we note that Trèves condition is satisfied. We recall that this class
3
contains the homogeneous model of degree 3 B(s, t, ξ, η) = s3 ξ + γ1 s2 t + γ2 st2 ,
which is not maximally hypoelliptic. The subellipticity is open. If we try to follow
our general method for ϕ(0) < 0, then ϕ has three zeroes 0, s± ()) ∼ ± −ϕ(0)
and two critical points. This determines 6 sectors, each associated arc having one
zero of φ at one end. We necessarily arrive to a configuration A− B + B + B − B − A+
for which there are no way to escape, following the rules we have given.
When
φ (s) = s3 + s2 ϕ(s) ,
with ϕ(0) = 0, we simply note that Trèves’s condition is not satisfied. Hence we do
not have hypoellipticity.
Finally when
φ (s) = s3 + s3 ϕ(s) ,
φ has a unique zero at s = 0 and we are in a configuration A− A+ .
27
8.5. Main statements.

Proposition 8.2.
If γ3 = 0, then the system (1.10) associated with B given by (8.1) is subelliptic
with coefficient 13 .
If γ3 = γ2 = 0 and γ1 = 0, then Trèves condition is not satisfied, hence the system
(1.10) is not hypoelliptic.
Remark 8.3.
The case when γ3 = 0 and γ2 = 0 is open (for hypoellipticity and subelliptic-
ity). Note that Trèves necessary condition is satisfied and that we have shown in
Subsection 4 that this case is not maximally hypoelliptic.
Remark 8.4.
Having in mind what was proven in [2, 3], we recall that we have to analyze the
zeroes of B(s, t, ξ, η) on |s|2 + |t|2 = 1. Hence, we have to look at the neighborhood
of two points on the circle : (0, 1) and (0, −1).
The proof above has treated a more general situation where the polynomial
character has disappeared. We should in this case look at the different zeros of B1
reduced to the circle. This will be explained later.
9. Subellipticity : Homogeneous case of degree 4

This time we consider homogeneous polynomials of degree 4 and after localizing
in a specific direction (we take (0, 1)), we look at the expansion of B near s = 0 on
the circle, with η close to 0. With = η/ξ, we can write
φ (s) = s4 + s ϕ(s) ,
with ϕ(0) = 0, when ≤ 3.
9.1. Analysis of φ . We now look at different cases :
A: The case = 0.
We start by the analysis of the zeroes of φ . We have
(9.1) φ (s) = 4s3 + ϕ (s) .
If ϕ (0) = 0, then we write
(9.2) ϕ (s) = sm ψ(s) ,
with ψ(0) = 0, if m ≤ 2.
Case a: m = 0.
Then φ has only one zero s0 () satisfying :
1
ψ(0) 3
(9.3) s0 () ∼ − .
4
At the minimum of φ , we have
(9.4) φ (s0 ) ∼ ϕ(0) .
Case b: m = 1.
If ϕ = s ψ, with ψ(0) = 0, we have
φ (s) = s(4s2 + ψ(s)) .
If ψ(0) > 0, φ has only one root : s = 0
If ψ(0) < 0 in this neighborhood, φ has three roots s̃1 , 0, s̃2 .
42
Computing the second derivative of φ gives :

φ (s) = 12s2 + ψ(s) + sψ (s) .
Hence 0 corresponds to a point of local maximum and the two other zeroes
of φ s̃1 and s̃2 , correspond to local minima. Moreover
1 1 1 1
(9.5) s̃1 () ∼ − |ψ(0)| 2 , s̃2 () ∼ + |ψ(0)| 2 .
2 2
It is then not difficult to see that on [s̃1 (), s̃2 ()], we have
|φ (s)| ≤ C|| 2 .

3
(9.6)
Case c: m = 2.
If ϕ (s) = s2 ψ(s), with ψ(0) = 0, we have
φ (s) = s2 (4s + ψ(s)) .
Then φ has two roots s = 0 (with multiplicity 2) and

s0 () ∼ − ψ(0) .
4
Computing the second derivative of φ gives :
φ (s) = 12s2 + 2sψ(s) + s2 ψ (s) .
At this point, we have :
1 2
φ (s0 ()) ∼ ψ(0)2 ,
4
and
φ (s0 ()) − ϕ(0) ∼ −4−4 3−1 4 ψ(0)4 .
Case d: m ≥ 3.
If ϕ = s3 ψ, we have
φ (s) = s3 (4 + ψ(s)) ,
Hence φ has a unique zero at s = 0 and φ has a unique minimum at 0.
(1) If ϕ(0) > 0, this is of course easy. φ has no zero and satisfy
(9.7) φ ≥ ϕ(0) − C2 > 0 .
Depending on m, φ has at most three zeroes.
More precisely, when m = 0, φ has a unique minimum at s0 (). We
will prove that the configuration A+ A+ works. When m = 1, we are either
in the previous situation with a unique minimum at 0 or with three zeroes.
This will correspond to the configurations A+ A+ and A+ C + C + A+ .
When m = 2, φ has an inflexion point at 0 and a positive minimum at
s0 (). We will show that a configuration A+ C + A+ works. Finally when
m ≥ 3, one has a unique positive minimum at 0. Again we will show that
a configuration A+ C + A+ works4.
4A configuration A+ A+ would also work.

29
(2) If ϕ(0) < 0, then φ has two roots s1 () and s2 () satisfying :
−σ < s1 () < 0 < s2 () < σ ,
and
1 1
(9.8) s1 () ∼ −|ϕ(0)| 4 , s2 () ∼ |ϕ(0)| 4 .
We continue the discussion in function of m.
(a) If m = 0, we have asymptotically
s1 () < s0 () < s2 () ,
We will show that configuration A+ B − B − A+ is suitable.
(b) If m = 1, and ψ(0) > 0, φ has a unique negative minima at 0 and
s1 () < 0 < s2 ()
We will show that configuration A+ B − B − A+ is suitable.
If m = 1 and ψ(0) < 0 in this neighborhood, 0 corresponds to a
point of negative local maximum
(9.9) φ (0) = ϕ(0) < 0 ,
and the two other zeroes of φ s̃1 and s̃2 , correspond to points of nega-
tive local minima. We will show that configuration A+ B − C − C − B − A+
is suitable.
(c) If m = 2, then φ has a unique negative minimum at s0 () and a
negative inflexion point at 0. Depending on the sign of s0 (), we are
either in the situation when
−σ < s1 < s0 < 0 < s2 < σ ,
or
−σ < s1 < 0 < s0 < s2 < σ .
In any case we will show that configuration A+ B − C − B − A+ is suit-

able.
(d) If m ≥ 3, φ has a unique negative minimum at 0. The suitable

configuration will be A+ B − B − A+ .
[B:] = 1.
Hence we have
φ (s) = s(s3 + ϕ(s)) .
φ has two roots : 0 and
1
s1 () ∼ −(ϕ(0)) 3 ,
whose sign is the sign of −ϕ(0).
The derivative
φ (s) = 4s3 + ϕ(s) + sϕ (s) ,
has only one zero
1 1
s0 () ∼ − (ϕ(0)) 3
4
44
Looking at the second derivative, we get

3
φ (s0 ()) ∼
2
(ϕ(0)) 3
4
and
1 4
φ (s0 ()) ∼ − 3 ϕ(0) 3 (1 − 4−3 ) .
4
4
φ has a unique minimum. We will show that a suitable configuration is A+ B − B − A+ .
[C:] = 2.
If ϕ(0) > 0, this is OK for the construction but this will not work when ϕ(0) < 0
and we have a need to treat the two signs.
This is actually a general fact. In any homogeneity we can observe that the
Trèves condition in a neighborhood of ξ = 1, η = 0) implies that cannot be even
if < 3. (see Remark 10.3)
If ϕ < 0, φ has three roots including 0 which is of multiplicity 2. But φ has
three roots (including 0). In this case the Trèves condition is not satisfied. Hence
it is not hypoelliptic.
[D:] = 3.
φ has two roots including 0 (as triple zero) and s1 () ∼ −ϕ(0) and φ has two
roots, with 0 as double zero and
3
s0 () ∼ − ϕ(0) .
4
We have
3 1
φ (s0 ()) ∼ −( )3 4 ϕ(0)4
4 4
and
9 2
φ (s0 ()) ∼
ϕ(0)2 .
16
So φ has a unique minimum at s0 () and, depending on the sign of ϕ(0), we have
−σ < 0 < s0 () < s1 () < σ ,
or
−σ < s1 () < s0 () < 0 < σ .
We will show that a suitable configuration is A+ B − B − A+ .
[E:] ≥ 4.
We have
φ (s) = s4 (1 + ϕ(s)) .
We will show that a suitable configuration is A+ A+ .
9.2. Determination of the configurations. We review all the different

cases considered in Subsection 9.1. As for the homogeneous case of degree 3, we
write, between −σ and σ, in increasing order the sequence consisting of the zeroes
of φ and φ . Except the two extreme intervals corresponding to sectors of type
A+ . We have simply to show that the other sectors are either of type B ± or of
type C ± , and that we can always choose the σ
for the C ± such that one can escape
from D.
31
Case A : = 0, ϕ(0) > 0.
When m = 0, we have two sectors corresponding to the arcs ] − σ, s0 ()[ and

]s0 (), σ[. It is clear that configuration A+ A+ is suitable.
When m = 1, and ψ(0) > 0, we can do the same thing. If ψ(0) < 0, we have
two minima and a local maximum. Hence we consider four sectors attached to the
sequence −σ, s̃1 , 0, s̃2 , σ and we have to consider the configuration A+ C + C + A+ .
The condition (7.5) is satisfied immediately using (9.7) and (9.6). Considering
some 0 < σ < σ, and when starting from S2 we take an outgoing ray parallel to
(−σ ). When starting from S3 , we take an outgoing ray parallel to (σ ).
When m = 2, we consider three sectors attached to the sequence (−σ, 0, s0 (), σ)

if ψ(0) < 0 or (−σ, s0 (), 0, σ) if ψ(0) > 0. We can take a configuration A+ B + A+
(or A+ C + A+ ).
Finally, when m ≥ 3, we can consider two sectors attached to the sequence

(−σ, 0, σ) which corresponds to the configuration A+ A+ .
Case A: = 0, ϕ(0) < 0.
When m = 0, we consider four sectors Sj (j = 1, . . . , 4) clockwise ordered and

delimited on the circle by the points : −σ < s1 < s0 < s2 < +σ. We introduce σ
< σ. For small enough we have :
on the circle satisfying : 0 < σ
−σ < − < σ.
σ < s1 < s0 < s2 < σ
We note that −σ corresponds to a point of maximum of φ on [−σ, s0 ] and that φ
is decreasing on ] − σ, s0 [, increasing on ]s0 , σ[.
The claim is that the configuration is A+ B − B − A+ . That is for points in S2 , we
consider ingoing rays parallel to (σ), and for points in S3 , we consider ingoing rays
parallel to (−σ).
When m = 1, there are two cases depending on the sign of ψ(0).

When ψ(0) > 0, we are as in the previous case in a configuration A+ B − B − A+ .
When ψ(0) < 0, we choose six adjacent sectors Sj (j = 1, . . . , 6) (ordered using
the clockwise sign) delimited by
−σ < s1 < s̃1 < 0 < s̃2 < s2 < σ .
We claim that we have the configuration A+ B − C − C − B − A+ . The only point is to
verify that S3 and S4 are of type (C − ). Hence we have to verify (7.6).
When m = 2, we treat the first case : −σ < s1 < s0 < 0 < s2 < σ . which
determine in clockwise order five sectors Sj (j = 1, . . . , 5). We can either choose
A+ B − C − B − A+ or A+ B − B − B − A+ .
For a point (s, t) in S1 , we consider outgoing rays parallel to (−σ) and for a
point in S5 an outgoing ray parallel to (σ).
For a point (s, t) in S2 ∪ S3 , we consider ingoing rays parallel to ( σ ), till the
ray enters in S1 and then use the rule of S1 .
For points (s, t) in S4 , we consider ingoing rays parallel to (−
σ ), till the ray enters
in S5 and then use the rule of this sector.
For the second case, nothing is changed except that S3 is treated like S4 .
46
When m ≥ 3, we have five points −σ < s1 < 0 < s2 < σ , which determine four
sectors. The configuration is A+ B − B − A+ .
Case B : = 1.
We treat the case when ϕ(0) > 0.
We have five characteristic points −σ < s1 < s0 < 0 < σ ,, which determine in
clockwise order four sectors Sj . The configuration is A+ B − B − A+ . For a point
(s, t) in S1 , we choose an outgoing ray starting from (s, t), which is parallel to
(−σ), till the ray leaves the disk. For a point (s, t) in S4 , we choose an outgoing
ray starting from (s, t), which is parallel to (σ), till the ray leaves the disk. For
a point (s, t) in S2 , we consider ingoing rays parallel to (σ ), till the ray enters in
S1 and then use the rule of S1 . For a point (s, t) in S3 , we consider ingoing rays
parallel to (−σ ), till the ray enters in S4 and then use the rule of S4 .
Case D : = 3.
We treat the case when ϕ(0) > 0. We have five points −σ < s1 < s0 < 0 < σ ,
which determine in clockwise order four sectors Sj (j = 1, . . . , 4). The configuration
is A+ B − B − A+ .
The other case is treated similarly by exchanging the role of 0 and s1 ().
Case E : ≥ 4.
We have three points
−σ < 0 < σ ,
which determine in clockwise order two sectors S1 and S2 . The configuration is
A+ A+ . For a point (s, t) in S1 we choose an outgoing ray starting from (s, t),
which is parallel to (−σ), till the ray leaves the disk. For a point (s, t) in S2 , we
choose an outgoing ray starting from (s, t), which is parallel to (−σ), till the ray
leaves the disk.
9.3. The results in the homogeneous case of degree 4.
In the polynomial case, we obtain
φ(s, t, ) = s4 + (γ1 s3 t + γ2 s2 t2 + γ3 st3 + γ4 t4 ) ,

√
this gives (replacing t by ± 1 − s2 ) :
Proposition 9.1.
1
We have microlocal subellipticity of order 4 in the following cases
i) γ4 = 0 or γ3 = 0 :
ii) γ2 = γ3 = γ4 = 0.
If γ4 = γ3 = 0 and γ2 = 0, the Trèves condition is not verified and hence the system
is not hypoelliptic.
Of course γ1 = γ2 = γ3 = γ4 = 0 is the trivial case. Note also that in the
cases (i) and (ii) (γ1 = 0), we have maximal hypoellipticity, and subellipticity
is also a consequence of this maximal hypoellipticity. The question of maximal
hypoellipticity in case (ii) (with γ1 = 0) is open (see Question 5.5).
10. The general case

The cases of homogeneity 3 and 4 are typical of the general discussion between
odd and even orders of the zeroes of φ=0 . In fact, as we will see below, all we used
was the variation of the functions φ and φ on sectors of type C ± .
33
10.1. Main result.

Theorem 10.1.
We continue to assume that n = 2 and m = 2. Let B homogeneous of degree r.
Let us assume that φ0 has a finite number of zeroes of finite order and let p be the
maximal order :
p = max(Ord(θ), θ zero of φ0 ) .
We also assume that B(s, t, ξ, η) satisfies Trèves condition5 at (0, 0; 1, 0) .
Let us also introduce φ̃0 = (φ − φ0 )/. If for each zero θ of φ0 , one of the following
1
conditions (i), (ii) or (iii) is satisfied, then the the system LB is sup(r,p) -microlocally
subelliptic at the point (0, 0, 0, 0; 0, 0, 1, 0) ∈ R × R \ {0} .
4 4
The three conditions are:

i) Ord(θ, φ0 ) is even;
ii) φ̃0 (θ) = 0 ;
iii) Ord(θ, φ0 ) ≤ Ord(θ, φ̃0 ).
Remark 10.2.
There is no condition on the order or the number of zeroes of φ̃0 . For example φ̃0
may vanish near a zero of φ0 , or even be identically zero on S1 .
Remark 10.3.
If we assume that
φ (s) = sp ϕ(s) + s ψ(s) ,
with ψ(0) = 0, even and 0 < < p, then φ can not satisfy Trèves condition at
(0, 0, 0, 0; 0, 0, 1, 0).
As before we will reduce the analysis to a neighborhood of a zero of φ0 and
analyze B (see (7.2)) with = ηξ . We can also assume after a linear change of
coordinates in R2 that θ = (0, 1). So we can parametrize near this point on the
circle by the variable s and consider as before the function φ of (7.4).
10.2. Proof in the case when B1 has a zero of odd order p ≥ 3 at

s = 0. We start from the decomposition
(10.1) φ (s) = sp ϕ(s) + s ψ(s) ,
with ϕ(0) = 0, and ψ(0) = 0 when < p, which is rewritten in the form
ψ(s)
φ (s) = ϕ(s)(sp + ).
ϕ(s)
We first observe that the Trèves condition implies that if < p then either = 0
or is odd. Let us assume indeed that is even with = 0. Then φ (s) =
s (sp− ϕ(s) + ψ(s)). So if ψ(0) < 0, then φ has a local maximum at 0.
So the Trèves condition permits to reduce the situation to the study of three cases
≥ p, = 0 and odd.
Subcase ≥ p.
In this case, φ has a zero of odd order at 0 and no other zero in a suitably small (but
independent) neighborhood of 0 denoted by ] − σ, σ[. Hence φ is monotone and
depending on the sign of ϕ(0), we associate two sectors to the sequence −σ < 0 < σ
and get a configuration A− A+ or A+ A− .
5See Condition 1.2.

48
Subcase = 0.
In that case, we have
ψ
φ = ϕ(sp + ) ,
ϕ
and we immediately see that φ has one zero
1
ψ(0) p
(10.2) s0 () ∼ − .
ϕ(0
Let us consider the derivative of φ .
(10.3) φ = sp−1 (pϕ + sϕ ) + ψ .
Writing
ψ (s) = sm g(s) .
We discuss between various subsubcases.
a) m ≥ p − 1.
Then φ has , for σ small enough a unique 0 at s = 0 of even order. We then obtain
four possibilities.
i) ϕ(0) > 0, s0 () < 0
ii) ϕ(0) > 0, s0 () > 0
iii) ϕ(0) < 0, s0 () < 0
iv) ϕ(0) < 0, s0 () > 0
The three sectors are determined by the sequence −σ < s0 () < 0 < σ for the
items i) and iii) and by the sequence −σ < 0 < s0 () < σ for the items ii) and iv).
It is then easy to give the corresponding configurations : A− B + A+ , A− B − A+ ,
A+ B − A− , and A+ B + A− .
b) m = 0.
In this case, assuming that g(0) = 0, we have
(10.4) φ = sp−1 (pϕ + sϕ ) + g .
Now p − 1 is even. If g(0)/ϕ(0) > 0, then φ has no zero on (−σ, σ). Hence φ is
monotone on ] − σ, +σ[ for σ > 0 small enough. The two sectors are determined
by the sequence −σ < s0 () < σ and the corresponding configuration are A− A+ or
A+ A− .
If g(0)/ϕ(0) < 0, then φ has two zeroes s1 () and s2 () of opposite signs such
that, for j = 1, 2,
1
−g(0) p−1
(10.5) |sj ()| ∼ .
pϕ(0)
Note that |sj ()| << s0 () for j = 1, 2. In this situation, we choose four sectors
corresponding to the sequence −σ < s1 () < s2 () < s0 () < σ leading to a config-
uration A− C − B − A+ (or A+ C + B + A− ) or to the sequence −σ < s0 () < s1 () <
s2 () < σ leading to a configuration A− B + C − A+ (or A+ B − C + A− ). Having in
mind what we have done for the homogeneous cases of degree 3 and 4, the only dif-
ficulty is to control the claim that the sector delimited on the circle by (s1 (), s2 ())
is indeed of type C ± . Now, because p ≥ 3 and (10.5) holds, φ (s) ∼ ψ(0) on this
interval and it is easy to see from (10.4) that |φ (s)| ≤ C||.
35
c) 1 ≤ m < p − 1.
We restart of (10.3) that we rewrite in the form

(10.6) φ = sm (pϕ + sϕ )sp−m−1 + g .
Hence, assuming again that g(0) = 0, 0 is a zero of order m and for other zeroes
we have to look at
(pϕ + sϕ )sp−m−1 + g = 0 .
The existence of solutions depends on the parity of m and of the sign of
h := −g(0)/ϕ(0)
We distinguish three subcases
i) m even and h < 0 ;
ii) m even and h > 0 ;
iii) m odd.
i)
In this case s = 0 is the only zero and is of even order. So φ is monotone and
associate with −σ < s0 () < σ, we have a configuration A+ A− or A+ A− .
ii)
In this case φ has three zeroes which are s = 0 and two other zeroes s1 () and
s2 () of opposite sign satisfying
p−m−1
1
h
(10.7) |sj ()| ∼ .
p
So we take a configuration associated to −σ < s1 () < s2 () < s0 () < σ or
−σ < s0 () < s1 () < s2 () < σ and have to show that we are in the configuration
A± C ± B ± A∓ or A± B ∓ C ∓ A∓ . As before, we have just to control |φ | from below
m
(we get |φ | ≥ C1 ||) and |φ | from above (we get |φ | ≤ C||1+ p−m−1 ) in the interval
[s1 (), s2 ()].
iii)
When m is odd, φ has another zero s1 () with asymptotics as → 0
p−m−1
1
h
(10.8) s1 () ∼ .
p
We associate to the sequence −σ < 0 < s1 () < s0 () < σ ( or to −σ <
s0 () < 0 < s1 () , or to what we get by exchanging s1 () and 0) and we get a
configuration A∓ C ∓ B ∓ A± , or A∓ B ± C ± A± . Then we have only to control |φ
from below between 0 and s1 () (we get |φ | ≥ C1 ) and |φ | from above (we get
m
|φ | ≤ C||1+ p−m−1 ).
10.3. The case when p is even, p ≥ 2. We keep the same notation and
consider three subcases.
Subcase ≥ p.
The Trèves condition implies ϕ(0) > 0. Hence we get
1 p
(10.9) ϕ (s) ≥ s ,
C
and 0 is the only zero and this zero is of even order. Rewriting
φ (s) = sp−1 (pϕ + sϕ + s−p ψ + s−p+1 ψ ) ,
50
we obtain that φ has a unique zero at 0. Associated with the sequence −σ < 0 < σ,
we get two sectors with configuration A+ A+ .
Subcase = 0.
So φ = sp ϕ + ψ with ϕ(0) > 0 and ψ(0) = 0. If ψ(0) > 0 then φ has no zero
and if ψ(0) < 0 φ has two zeroes of opposite sign satisfying
1
(10.10) s± () ∼ ±(−ψ(0)) p
We have then to analyze
φ = sp−1 [pϕ + sϕ ] + ψ ,
and to discuss in function of the behavior of ψ at 0. We write
ψ (s) = sm θ(s) ,
with θ(0) = 0 if m < p − 1. a) m ≥ p − 1.
Then φ has only one zero at 0 of odd order. Hence φ has at 0 a minimum.
When ψ(0) > 0, we can associate to the sequence −σ < 0 < σ two sectors
corresponding to the configuration A+ A+ . When ψ(0) < 0, we associate to the
sequence −σ < s− () < 0 < s+ () < σ four sectors corresponding to a configuration
A+ B − B − A+ .
b) m = 0.
Then, as θ(0) = 0, φ has a unique zero s1 (), which satisfies
1
θ(0) p−1
(10.11) s1 () ∼ − .
pϕ(0)
When ψ(0) > 0, we can associate to the sequence −σ < s1 () < σ two sectors
corresponding to the configuration A+ A+ . When ψ(0) < 0, we associate to the
sequence −σ < s− () < s1 () < s+ () < σ four sectors corresponding to a configu-
ration A+ B − B − A+ .
c) 1 ≤ m < p − 1.
We write
φ = sm sp−m−1 [pϕ + sϕ ] + θ , ,
and are let to distinguish between two possibilities :
• m even
• m odd
When m is even, then φ has two zeroes, 0 which is of even order and another one
s1 () admitting the asymptotics
1
θ(0) p−m−1
(10.12) s1 () ∼ − .
pϕ(0)
Then
– either 0 is an inflexion point and s1 () is a minimum and we consider the sequence
−σ < s− () < s1 () < s+ () < σ with a configuration A+ B − B − A+ – or 0 is the
minimum and s1 () is an inflexion point and we consider the sequence −σ < s− () <
0 < s+ () < σ with a configuration A+ B − B − A+ .
When m is odd, then φ has either one zero if θ(0) > 0 or if θ(0) < 0, three
zeroes, 0 which is of odd order and two others s± () admitting the asymptotics
1
−θ(0) p−m−1
(10.13) s± () ∼ ± .
pϕ(0)
37
We have then four cases.

i) ψ(0) > 0, θ(0) > 0
ii) ψ(0) < 0, θ(0) > 0
iii) ψ(0) > 0, θ(0) < 0
iv) ψ(0) < 0, θ(0) < 0
We meet successively the configurations A+ A+ , A+ B − B − A+ , A+ C − C − A+ and
A+ B − C − C − B − A+ . Let us verify our claim for the last case. We have six sectors
associated with the sequence −σ < s− () < s− () < 0 < s+ () < s+ () < σ. We
have to control on [ s− (), s+ ()] |φ | from below (we have |φ | ≥ |φ (0)| ∼ −ψ(0))
p−1
and |φ | from above (we have |φ | ≤ C(|| p−m−1 + ||)). So we have |φ | ≤ C|| on

s− (), s+ ()].

[
Subcase 1 ≤ ≤ p − 1.
We start from φ (s) = s (sp− ϕ(s) + ψ(s)), with ψ(0) = 0 and ϕ(0) > 0. We
know that should be odd. Otherwise Trèves Condition will not be satisfied when
ψ(0) < 0. Hence and p − are odd. Hence φ has two zeroes, 0 of odd order
and s1 () satisfying
1
ψ(0) p−
(10.14) s1 () ∼ − .
ϕ(0)
We now look at the zeroes of φ :

φ = s−1 sp− (pϕ + sϕ ) + ( ψ + sψ ) .
When = 1, φ has one zero s1 () such that
1
ψ(0) p−1
s1 () ∼ −
pϕ(0)
s1 () > 0. We meet the configuration A+ B − B − A+ associated
We observe that s1 ()
with the sequences −σ < 0 < s1 () < s1 () < σ or −σ < s1 () < s1 () < 0 < σ.
When > 1, φ has two zeroes, 0 of even order and s1 () such that
1
ψ(0) p−
s1 () ∼ − .
pϕ(0)
We meet the configuration A+ B − B − A+ associated with the sequences −σ < 0 <
s1 () < s1 () < σ or −σ < s1 () < s1 () < 0 < σ.
52
Appendix A. On a lemma of uniform lower bound of the variation of a

function on monotonicity intervals
Although looking standard (see in Maire [11] the use of Lojasevic’s inequality),
the next lemma, whose interest is the local uniformity of the constants, would be
very useful. If φ is a C k function defined in a neighborhood of some point s0 and
has a zero of order k at s0 , i.e. satisfies φ(s0 ) = φ(k−1) (s0 ) = 0 and φ(k) (s0 ) = 0, it
is easy to see that there exists α > 0 such that, for all s, s ∈]s0 − α, s0 + α[ such
that (s − s0 )(s − s0 ) > 0, we have
1
(A.1) |φ(s) − φ(s )| ≥ ck (φ)|s − s |k ,
2
with ck (φ) = k! 1
|φ(k) (s0 )|.
The multiplicative factor 12 in (A.1) can actually be rep laced by any constant in
]0, 1[. It is less clear if we can keep this estimate uniform by perturbing a little f .
The zero of order k of φ can split in at most k zeroes or disappear. We will see
below that a suitable formulation of (A.1) can be made stable by perturbation.
Lemma A.1.
Let (s, ) → φ(s, ) a C ∞ function defined in a neighborhood of (0, 0) in R × Rm .
Let us assume that, for some k ∈ N∗ , we have
(A.2) φ(0, 0) = · · · = (∂s(k−1) φ)(0, 0) = 0 ,

and
(A.3) (∂sk φ)(0, 0) = 0 .
Then there exist an interval U0 (neighborhood of 0) and a ball V0 (neighborhood of
0) such that, for any ∈ V0 , the function s → φ(s, ) has k() zeroes in U0 with
k() ≤ k. Moreover, these neighborhoods can be chosen such that, for any ∈ V0
and on any subinterval I () ⊂ U0 on which φ(·, ) is monotone, we have
1 ck (φ)
(A.4) |φ(s, ) − φ(s , )| ≥ |s − s |k , ∀s, s ∈ I () ,
2 Ck
with
1 k
(A.5) ck (φ) = |∂ φ(0, 0)| ,
k! s
and Ck a universal positive constant depending only on k.
Proof
The initial choice of U0 =] − α, +α[ is done in order that ∂s φ(., 0) have only s = 0
as zero in U0 for = 0 , · · · , k − 1.
The main ingredient is the formula
s

(A.6) φ(s , ) − φ(s, ) = ∂s φ(τ, ) dτ ,
s
that we will always use in a monotonicity interval of φ, i.e. for s and s in a -

dependent interval where ∂s φ(·, ) has a fixed sign.
39
The case k = 1.
We have simply to apply the implicit function theorem. We get for close to 0 a
unique zero s0 () depending smoothly on and such that s0 (0) = 0. We note that
∂s φ(·, ) is non zero. Say for example that ∂s φ > 0. For s and s such that s > s,
we deduce from (A.6) that (A.4) in the form
|φ(s, ) − φ(s , )| ≥ inf |∂s φ(s, )| |s − s | .
(s,)∈U0 ×V0
Choosing U0 and V0 small enough, we get (A.4)-(A.5), with C1 = 1.
At many times (but only finitely many times), we will have to take, without to
mention it explicitly, smaller U0 and V0 .
The case k = 2.
We first consider the unique zero s1 () of ∂s φ(·, ) whose existence is given by the
implicit function applied to ∂s φ(·, ). Assuming for definiteness that ∂s2 φ(0, 0) > 0,
we can choose a smaller interval U0 and a smaller neighborhood V0 such that, for
∈ V0 , s ∈ U0 such that s ≥ s1 (), we have
1
∂s φ(s, ) ≥ |∂s2 φ(0, 0)|(s − s1 ()) ≥ 0 .
2
Hence we obtain that for any s, s , such that s > s > s1 (),
s
f (s , ) − f (s, ) ≥ 12 |∂s2 φ(0, 0)| s (τ − s1 ())dτ
= 14 |∂s2 φ(0, 0)| (s − s1 ())2 − (s − s1 ())2
≥ c21 (s − s)2 .
Here we have used for the last inequality that, for ∈ N∗ ,
(A.7) (θ ) − θ ≥ (θ − θ) , ∀θ, θ s. t. θ > θ ≥ 0 .
So we have obtained the lemma for k = 2 with C2 = 1.
The case k = 3.
First we denote by s2 () the zero of ∂ss φ(·, ) and assume for definiteness that
∂sss φ(·, ) > 0. Then ∂s φ(·, ) has either no zero, one double zero, or two zeroes
depending of the sign of ∂s φ(s2 (), ).
Let us consider the two subcases.
Subcase a : ∂s φ(s2 (), ) ≥ 0
In this case, ∂s φ ≥ 0 and φ is monotone. We will prove (A.4) first with s, s ≥ s2 (),
then with s, s ≤ s2 (). The general result for s, s ∈ U0 then easily follows using a
“splitting argument” based on the following inequality, for θ0 ≤ θ1 ≤ θ2 ,
|θ2 − θ1 |3 + |θ1 − θ0 |3 ≥ 2−3 |θ2 − θ0 |3 .
More generally we could need later that, for s ≥ 1, ∈ N∗ and θ0 ≤ θ1 ≤ θ2 ≤
· · · ≤ θ ,

(A.8) |θk − θk−1 |s ≥ −s |θ − θ0 |s .
k=1
Coming back to our analysis, we have to find a lower bound of ∂s φ(τ, ) for
τ ≥ s2 (). We have

c3 (φ)
c3 (φ)
∂s φ(τ, ) ≥ ∂s φ(s2 (), ) + (τ − s2 ())2 ≥ (τ − s2 ())2 ,
2 2
54
with

c3 (φ) = inf |∂s3 φ(s, )| .
(s,)∈U0 ×V0
Implementing in the formula (A.6) for φ(s )−φ(s2 ()), we obtain, for s > s ≥ s2 () :

c3 s
c3 (φ) c3 (φ)
f (s , ) − f (s, ) ≥ (τ − s2 ())2 dτ ≥ (s − s)3 ≥ (s − s)3 .
2 s 6 2
The last inequality, based on c3 6(φ) ≥ c3 2(φ) , being achieved by taking U0 and V0
small enough.
Considering general pairs s , s and using the splitting argument, we obtain the
lemma with the condition C1 ≥ 8.
Subcase b : ∂s φ(s2 (), ) < 0
Analyzing the behavior of f (s, 0) and taking sufficiently close to 0 we get that
∂s φ has two zeroes. We denote by s− +
1 () and s2 () the two zeroes such that
s− +
1 () < s2 () < s1 () .
We then have three intervals of monotonicity for φ : ]s−∞ , s− − +

1 ()[, ]s1 (), s1 ()[ and
]s+ (), s+∞ [.
For the first interval, we write, for s < s < s− 1 (),

s
φ(s, ) − φ(s , ) = s ∂s φ(τ, )dτ
s
≥ c3 2(φ) s (s−
1 () − τ ) dτ
2

c3 (φ) 3
≥ 6 (s − s )
≥ c3 2(φ) (s − s )3 .
The third interval can be treated in the same way. For the middle one, we have to
split this interval in two subintervals ]s− +
1 (), s2 ()[ and ]s2 (), s1 ()[. One considers
− −
three cases depending on when s1 () ≤ s ≤ s ≤ s2 (), s1 () ≤ s ≤ s2 () ≤ s ≤

1 () and s2 () ≤ s ≤ s ≤ s1 () . In the first case
s+ +
s

c3 s
c3
φ(s, ) − φ(s , ) = (−∂s φ(τ, ))dτ ≥ (τ − s−
1 ()) dτ ≥
2
|s − s |3 .
s 2 s 6
This implies to choose C3 ≥ 1. In the last case,
s

c3 (φ) s
c3 (φ)
φ(s, ) − φ(s , ) = (−∂s φ(τ, ))dτ ≥ (τ − s+
1 ()) dτ ≥
2
|s − s |3 .
s 2 s 6
In the middle case, we use the “splitting” argument by writing
φ(s, ) − φ(s , ) = φ(s, ) − φ(s2 (), ) + φ(s2 (), ) − φ(s , )
and can use the previous analysis. Note here that this is why we are obliged to
increase C3 in comparison with the case k = 1 and k = 2 and get C3 ≥ 8, without
to pretend at optimality.
The recursion argument.

We observe that we have proven the lemma for k = 1 and 2 (the analysis for k = 3
was a kind of preliminary exercise to understand how to make the recursion).
We now assume that we have proved the lemma at order less than k0 − 1
for any function f satisfying (A.2)-(A.3) and will prove it for k = k0 . Hence
we assume that f satisfies (A.2)-(A.3) with k = k0 and will first apply the recursion
41
argument to ∂s φ(·, ) after observing that this function satisfies (A.2)-(A.3) with
k = k0 − 1. Hence we have, for any s, s in a monotonicity interval of ∂s φ(·, ) :
1
(A.9) |∂s φ(s, ) − ∂s φ(s , )| ≥ ck−1 (∂s φ) |s − s|k−1 .
2Ck−1
We observe that
(A.10) ck−1 (∂s φ) = k ck (φ) .
The proof goes like as for k = 3. First we determine the monotonicity νk + 1 inter-
()
vals of f (·, ξ) (at most k + 1) by considering the νk zeroes s1 () of ∂s φ(·, ).
Then we have three types of intervals :
(1) (ν ) (j) (j+1)
]s−∞ , s1 ()[, ]s1 k (), s+∞ [ and ]s1 (), s1 ()[.
2
In each of these intervals, we determine the zeroes of ∂ss φ(·, ) corresponding to an
effective change of monotonicity for ∂s φ(·, ) and get an attached decomposition of
subintervals.
In each of these subintervals, we consider the quantity φ(s, ) − φ(s , ) and use
the first order Taylor formula. We compare ∂s φ(τ, ) with ∂s φ(s̃, ) where s̃ is the
end of the subinterval where |∂s φ(·, )| is minimal. One can then use
kck kck
|∂s φ(τ, )| ≥ |∂s φ(s̃, )| + |τ − s̃|k−1 ≥ |τ − s̃|k−1 .
Ck−1 Ck−1
References
[1] M. Derridj. Subelliptic estimates for some systems of complex vector fields. In “Hyperbolic
problems and regularity questions”. Series Trends in Mathematics. Edtrs: M. Padula and L.
Zanghirati. Birkhäuser (2006), p. 101-108.
[2] M. Derridj and B. Helffer. Subelliptic estimates for some systems of complex vector fields:
quasihomogeneous case. http://www.arxiv.org/abs/math.AP/0611926 and Trans. Amer.
Math. Soc. 361 (2009), no. 5, p. 2607-2630.
[3] M. Derridj and B. Helffer. On the subellipticity of some hypoelliptic quasihomogeneous sys-
tems of complex vector fields. In ”Complex Analysis”, Series Trends in Mathematics. Edtrs:
P. Ebenfelt, N. Hungerbuehler, J.J. Kohn, N. Mok, E.J. Straube. Birkhuser (2010), p. 109-
124.
[4] B. Helffer and F. Nier.Hypoelliptic estimates and spectral theory for Fokker-Planck operators
and Witten Laplacians. Lecture Notes in Math. 1862, Springer Verlag, Berlin 2005.
[5] B. Helffer and J. Nourrigat. Hypoellipticité maximale pour des opérateurs polynômes de
champs de vecteurs. Progress in Mathematics, Birkhäuser, Vol. 58 (1985).
[6] L. Hörmander. Hypoelliptic second order differential equations.Acta Mathematica 119 (1967),
p. 147-171.
[7] L. Hörmander. Subelliptic operators. Seminar on singularities of solutions of partial differen-
tial equations.Ann. Math. Studies 91 (1978), p. 127-208.
[8] J.L. Journé and J.M. Trépreau. Hypoellipticité sans sous-ellipticité : le cas des systèmes de n
champs de vecteurs complexes en (n + 1)- variables. Séminaire EDP in Ecole Polytechnique,
April 2006.
[9] J. Kohn. Lectures on degenerate elliptic problems. Pseudodifferential operators with applica-
tions, C.I.M.E., Bressanone 1977, p. 89-151 (1978).
[10] J.J. Kohn. Hypoellipticity and loss of derivatives, with an appendix by M. Derridj and D.
Tartakoff. Ann. of Math. 162 (2), p. 943-986 (2005).
[11] H.M. Maire. Hypoelliptic overdetermined systems of partial differential equations. Comm.
Partial Differential Equations 5 (4), p. 331-380 (1980).
[12] H.M. Maire. Résolubilité et hypoellipticité de systèmes surdéterminés. Séminaire Goulaouic-
Schwartz 1979-1980, Exp. V, Ecole Polytechnique (1980).
[13] H.M. Maire. Necessary and sufficient condition for maximal hypoellipticity of ∂¯b . Unpublished
notes, Séminaire de Rennes 1981.
56
[14] H.M. Maire. Régularité optimale des solutions de systèmes différentiels et du Laplacien associé
: application au b . Math. Ann. 258, p. 55-63 (1981).
[15] F. Nier. Hypoellipticity for Fokker-Planck operators and Witten Laplacians. Course in China.
Preprint September 2006. To appear in 2010.
[16] J. Nourrigat. Subelliptic estimates for systems of pseudo-differential operators. Course in
Recife (1982). University of Recife.
[17] J. Nourrigat. Systèmes sous-elliptiques. Séminaire Equations aux Dérivées Partielles, 1986-
1987, exposé V, Ecole Polytechnique (1987).
[18] J. Nourrigat. Subelliptic systems II. Inv. Math. 104 (2) (1991), p. 377-400.
[19] F. Trèves. A new method of proof of the subelliptic estimates. Comm. Pure Appl. Math. 24
(1971), p. 71-115.
[20] F. Trèves. Study of a model in the theory of complexes of pseudo-differential operators. Ann.
of Math. (2) 104, p. 269-324 (1976). See also erratum: Ann. of Math. (2) 113, p. 423 (1981).
5 rue de la Juvinière, 78 350 Les loges en Josas, France.,

E-mail address: maklouf.derridj@wanadoo.fr
Laboratoire de Mathématiques, Univ Paris-Sud and CNRS, F 91 405 Orsay Cedex,

France.
E-mail address: Bernard.Helffer@math.u-psud.fr
Volume 550, 2011
Existence of trace for solutions of locally

integrable systems of vector fields
J. Hounie and E. R. da Silva
Abstract. We give sufficient conditions for the existence of trace of homo-

geneous solutions defined on wedges of general locally integrable structures,
extending previous results that considered locally integrable structures of a
particular nature.
Introduction
A classical result states ([H1, Thms. 3.1.14 and 3.1.15]) that a holomorphic
function in one complex variable, defined on domain with smooth boundary, that
has tempered growth at the boundary possesses a well defined distributional bound-
ary value. In the case of several complex variables, one considers the more general
situation of holomorphic functions defined on wedges and studies their boundary
values at the edges and an analogous result holds [BER, Ch. VII]. If we view holo-
morphic functions as homogeneous solutions of an overdetermined system of equa-
tions, it is natural to ask for which kind of overdetermined systems of vector fields
their continuous homogeneous solutions defined on wedges behave similarly, that
is, they have weak boundary values provided some growth restriction is assumed
at the edge. Several works have dealt with this problem in particular situations,
the case of a single vector field has been considered in [BH1], [BH2], [BH3] and
[BCH, Thm. VI.1.3] while E. Bär studied in her thesis [B] the case of solutions
defined in a wedge for a locally integrable system of vector fields of co-rank one.
Our main result applies to continuous solutions of a general overdetermined sys-
tem of first order partial differential equations that arises from a locally integrable
2010 Mathematics Subject Classification. Primary 35F45, 35B30; Secondary 35F35.

Key words and phrases. Locally integrable structures, Boundary value of homogeneous solu-
tions, Baouendi-Treves approximation formula,
Work supported in part by CNPq and FAPESP.
1
57
58
2 J. HOUNIE AND E. R. DA SILVA
involutive structure and gives a sufficient condition for the existence of boundary
values. Involutive structures arise in many geometric contexts including foliations,
complex structures, and CR structures (see for example [EG1], [EG2], [HJ] and
[Sz]). A smooth locally integrable involutive structure is a pair (M, L) where M is
a smooth manifold and L is a smooth, involutive subbundle of CT M such that L⊥ ,
the subbundle of CT ∗ M orthogonal to L, is locally generated by exact one-forms.
Similarly, a real analytic involutive structure is a pair (M, L) where M is a real
analytic manifold and L is a real analytic, involutive subbundle of CT M. It fol-
lows from the Cauchy-Kowaleska theorem that a real analytic involutive structure
is always locally integrable, in particular, our results apply to general real analytic
involutive structures.
The paper is organized as follows. In section 1 we state a sufficient growth
condition that guarantees the existence of trace for a homogeneous solution, defined
on a wedge with maximally real edge, of a locally integrable involutive structure
(Theorem 1.1) which is our main result. This condition is (in general) strictly
weaker that the usual requirement of tempered growth at the edge. However,
this condition is formulated in terms of a special first integral, so in section 2, we
address the invariance problem and prove that our growth condition is actually
independent of the choice of the first integral by attaching a local invariant to
points p ∈ (M, L, Σ), where (M, L) is a locally integrable structure and Σ ⊂ M is
a maximally real submanifold. In section 3 we prove a slightly strengthened form
of Theorem 1.1 (Theorem 3.1). In Section 4 we give an application of the invariant
defined in Section 2, showing that it can be used to characterize CR structures
among general locally integrable structures.
1. Statement of the main result
Suppose (M, L) is a smooth locally integrable structure, that is, M is a smooth

manifold of dimension N = m + n, L is a smooth subbundle of CT M of fiber
dimension n over C and its orthogonal L⊥ has fiber dimension m and can be
generated on some neighborhood of any given point by the differentials of m complex
functions Z1 , . . . , Zm . To avoid trivial cases, we will always assume that n (called
the rank of L) and m (called the co-rank of L) are ≥ 1. A system of m locally defined
functions Z1 , . . . , Zm whose differentials dZ1 , . . . , dZm span L⊥ is called a complete
set of first integrals for L or, in short form, we may say that Z = (Z1 , . . . , Zm ) is a
EXISTENCE OF TRACE AND LOCALLY INTEGRABLE STRUCTURES 59
3
first integral of the system. On the subject of locally integrable structures we refer
to [T] and [BCH]. We recall that
Definition 1.1. Let (M, L) be a smooth locally integrable structure. A sub-

manifold Σ of M is called maximally real with respect to L if
CTp M = CTp Σ ⊕ Lp , p ∈ Σ.
Definition 1.2. Let Σ be a submanifold of M, dim M = r + s, dim Σ = r,

r, s > 0. We say a subset W is a wedge in M at p ∈ Σ with edge Σ if the following
holds: there exists a diffeomorphism ϕ of a neighborhood V of 0 in Rr+s onto a
neighborhood U of p in M with ϕ(0) = p and a set B × Γ ⊆ V with B a ball
centered at 0 ∈ Rr and Γ a truncated open convex cone in Rs with vertex at 0 such
that ϕ(B × Γ) = W and ϕ(B × {0}) = Σ ∩ U .
If Σ, M, W and p ∈ Σ are as in the previous definition, the direction wedge

Γp (W) ⊆ Tp (M) is defined as the interior of
{c (0) | c : [0, 1] → M smooth, c(0) = p, c(t) ∈ W ∀t > 0} .
Equivalently, Γp (W) = {dϕ(Rr × {λv | v ∈ Γ , λ > 0})}. Thus Γp (W) is a linear

wedge in Tp M with edge dϕ(Rr × {0}) = Tp Σ. If Σ is a hypersurface in M, then
a wedge W with edge Σ is simply a side of Σ.
From now on, we will assume that Σ is a maximally real submanifold, Wis
a wedge in M at p and consider the existence problem for the trace of a con-
tinuous null solution u of L, i.e., a continuous function whose (weak) differential
du|q ∈ L⊥
q , q ∈ M. Since this is a local problem, we may choose local coordi-
nates x1 , . . . , xm , t1 , . . . , tn , such that (x(p), t(p)) = (0, 0) and assume we are in the
following situation:
(1) Σ is given by the equations tj = 0, j = 1, . . . , n, so we may set Σ = {(x, 0) :

|x| < r} after its identification with an open subset of Rm ×{0} ⊂ Rm ×Rn ;
(2) W = Brx (0) × ΓT , where Brx (0) ⊂ Rm denotes the open ball of radius
r > 0 centered at the origin, ΓT = Γ ∩ {t ∈ Rn ; |t| ≤ T }, Γ ⊂ Rn is a
convex open cone with vertex at the origin, and T > 0;
(3) the functions Z1 , . . . , Zm , whose differentials span L⊥ may be chosen to
have the form
(1.1) Zk (x, t) = xk + iϕk (x, t), k = 1, . . . , m,

60
where the functions ϕk (x, t) are real, ϕk (0, 0) = 0, k = 1, . . . , m, and

∂ϕk /∂x (0, 0) = 0, 1 ≤ k, ≤ m;
(4) L is generated by pairwise commuting vector fields L1 , . . . , Ln of the form
∂ m
∂
Lj = + λjk (x, t) , j = 1, . . . , n
∂tj ∂xk
k=1
and u satisfies in the sense of distributions the overdetermined system
(1.2) Lj u(x, t) = 0, (x, t) ∈ Brx (0) × ΓT , j = 1, . . . , n.
In view of (1.2) it is customary to write Lu = 0 rather than du|q ∈ L⊥

q , q ∈ M.
We will set Γ0 = Γ ∩ S n−1 = {t ∈ Γ : |t| = 1} so we may write ΓT = {τ t : t ∈

Γ0 , 0 < τ < T }.
We will now state our main result. Consider the map
Z(x, t) = (Z1 (x, t), . . . , Zm (x, t)) : Brx (0) × BTt (0) −→ Cm
and the function

.
d(x, τ, t ) = dist (Z(x, τ t ), Z(Σ)) : Brx (0) × (0, T ) × Γ0 −→ R+ .
Theorem 1.1. Let u be a continuous solution of (1.2) and assume that there
exists ν ∈ N such that
T
ν
(1.3) sup dist Z(x, τ t ), Z(Σ) |u(x, τ t )| dxdτ < ∞.
t ∈Γ0 0 Brx (0)
Then u(x, t) has a distributional limit as t → 0 in ΓT . More precisely, for any test
function φ(x) ∈ Cc∞ (Brx (0)), the limit

.
bu, φ = lim u(x, t)φ(x) dx
t→0
t∈ΓT
exists and defines a distribution of order ν + 1.
Remark 1.1: It is easy to check that the alternative condition
(1.4) sup d(Z(x, t), Z(Σ))ν | u(x, t)| < ∞

t∈ΓT
is stronger than (1.3). Furthermore, (1.4) is implied by
(1.5) sup |t|ν | u(x, t)| < ∞,

t∈Γ
In particular, tempered growth of u(x, t) as t → 0 guarantees the existence of bu.

Observe also that Theorem 1.1 extends all previous special results mentioned in the
introduction concerning the existence of boundary values.
5
2. Invariance of the growth condition
Although condition (1.3) was formulated in terms of a special choice of coordi-

nates, it is easy to see by changing variables in the integrals that it is coordinate-
free. On the other hand, a specific first integral Z(x, t) is present in (1.3), so it is
of interest to show that, in fact, this condition does not depend on the choice of
the first integral. This will be shown now. The basic tool is the Baouendi-Treves
approximation theorem [BT] of which several variations and extensions are known.
We now describe briefly the version we will use (see, e.g., the proof of [BCH, Thm.
II.1.1]). Assume (M, L) is a locally integrable structure, L has fiber dimension
n and M has dimension N = n + m. Then, given p ∈ M, there exist an open
neighborhood U of p and smooth complex functions Z1 , . . . , Zm , defined in U and
satisfying LZj = 0 on U , such that to every open set U2 ⊂ U that contains p, we
may find another open set U1 such that p ∈ U1 ⊂ U2 with the following property:
every function u ∈ C k (U2 ), k = 0, 1, . . . , ∞, that satisfies Lu = 0 on U2 can be
approximated uniformly in U 1 together with its derivatives up to order k, by a
sequence of functions of the form uj = Pj (Z1 , . . . , Zm ), where Pj is a polynomial in
m variables with complex coefficients. A standard consequence is that if we assume
that u ∈ C 0 (U2 ) and write Z = (Z1 , Z2 , . . . , Zm ), there exists a continuous function
: Z(U 1 ) → C, such that the factorization u = u
u ◦Z holds on U 1 . The function u
is
obtained as the limit of the polynomials Pj (ζ), ζ ∈ Cm , which converge uniformly
for ζ ∈ Z(U 1 ). We will need an improved version of this fact.
Lemma 2.1. With the previous notation, if u ∈ C 1 (U2 ), then
: Z(U 1 ) −→ Cm
u
is a Lipschitz function, i.e., there exists K > 0 such that
u(ζ1 ) − u
| (ζ0 )| ≤ K|ζ1 − ζ0 |, ζ1 , ζ0 ∈ U 1 .
Proof: By the proof of Theorem II.1.1 in [BCH] we may assume that the functions
Zk , 1 ≤ k ≤ m, are given by (1.1), and the choice of local coordinates (x, t) is such
that p = (0, 0) and U 1 is expressed by |x| ≤ a, |t| ≤ b. Let ζ0 = Z(x0 , t0 ),
ζ1 = Z(x1 , t1 ) be two arbitrary points in Z(U 1 ) and set p0 = (x0 , t0 ), p1 = (x1 , t1 ),
q = (x1 , t0 ), ζ2 = Z(q). Notice that ζ0 = x0 , ζ1 = ζ2 = x1 and consider
smooth curves γ0 and γ1 given by
γ0 = {(x, ϕ(x, t0 )) : x ∈ [x0 , x1 ] ⊂ Rm }

62
and
γ1 = [Z(p1 ), Z(q)] ⊂ {(x1 , ϕ(x1 , t)); t ∈ [t0 , t1 ] ⊂ Rn }
where [A, B] denotes the closed convex hull of the points A and B. Next consider
the approximating sequence uj = Pj ◦ Z, j = 1, 2, . . . , and write

Pj (ζ2 ) − Pj (ζ0 ) = dPj ,
γ
0
Pj (ζ1 ) − Pj (ζ2 ) = dPj .
γ1
From the fact that ∇uj converges uniformly to ∇u on U 1 we may derive in a

standard way (invoking the fact that (∂Pj /∂ζk ) ◦ Z = Mk (Pj ◦ Z), k = 1, . . . , m,
for vector fields Mk defined on U2 ) that |dPj | is bounded on Z(U 1 ) by a constant
independent of j ∈ N. We refer to [BCH, p. 24] on the definition of the Mk ’s.
Since the curves γ0 and γ1 are contained in Z(U 1 ), it follows that
(2.1) |Pj (ζ1 ) − Pj (ζ0 )| ≤ C0 (|γ0 | + |γ1 |).
We will next show that
(2.2) |γ0 | + |γ1 | ≤ C1 |ζ1 − ζ0 |
with C1 independent of ζ1 , ζ0 ∈ Z(U 1 ). Indeed, γ0 is the image by Z of the segment

[x0 , x1 ], so if C is a bound for supU 1 |Zx | we have
|γ0 | ≤ C|x1 − x0 | ≤ C|Z(x1 , t1 ) − Z(x0 , t0 )| = C|ζ1 − ζ0 |
Furthermore,
|γ1 | = |ζ1 − ζ2 | ≤ |ζ1 − ζ0 | + |ζ0 − ζ2 |
≤ |ζ1 − ζ0 | + |x1 − x0 | + |ϕ(x0 , t0 ) − ϕ(x1 , t0 )|
≤ |ζ1 − ζ0 | + |x1 − x0 | + C|x1 − x0 |
≤ C2 |ζ1 − ζ0 |,
so (2.2) holds true. Hence, (2.1) and (2.2) imply that
|Pj (ζ1 ) − Pj (ζ0 )| ≤ K|ζ1 − ζ0 |
and letting j → ∞ we obtain
|
u(ζ1 ) − u
(ζ0 )| ≤ K|ζ1 − ζ0 |
as we wished to prove.
7
Consider now a second set Z1# , . . . , Zm

#
of smooth first integrals defined on a
. .
neighborhood U of p and set U2 = U ∩ U = U2# . Then, we may find open
# #
neighborhoods of p, U1 ⊂ U2 , U1# ⊂ U2# , and continuous functions

Z# : Z(U 1 ) → Cm , Z : Z # U # → Cm ,
1
such that Z # = Z# ◦ Z and Z = Z ◦ Z # on U1 ∩ U # . Choose an open set

1
# .
p ∈ U0 ⊂ U 0 ⊂ U 1 ∩U1 . It turns out that Z maps homeomorphically W = Z U 0
#
.
onto W # = Z # U 0 and the inverse of Z# : W → W # is given by Z : W # → W .
Furthermore, by Lemma 2.1, Z and Z # are Lipschitz functions.
We now apply these considerations to the setup of Theorem 1.1 and the role of
# ◦ Z and Z = Z
Σ in condition (1.3). Using the factorizations Z # = Z ◦ Z # , we
see that
(2.3) dist (Z(p ), Z(Σ ∩ U0 )) dist (Z # (p ), Z # (Σ ∩ U0 )), p ∈ U0 .
This has the following interpretation. Let V be a neighborhood of p ∈ M and

assume that two sets of first integrals Z = (Z1 , . . . , Zm ) and Z # = (Z1# , . . . , Zm
#
)
. .
are defined on V . Consider the functions d(q) = dist (Z(q), Z(Σ ∩ V )) and d (q) =#
dist (Z # (q), Z # (Σ ∩ V )), q ∈ V . If f is a continuous function defined in a neigh-

borhood of p, denote by f its germ at p. If f and g are two such germs declare that
f ∼ g if for some representatives f ∈ f and g ∈ g and some constants c1 , c2 > 0,
and some some neighborhood V of p, the estimates
c1 |f (q)| ≤ |g(q)| ≤ c2 |f (q)|, q ∈ V .
hold. It is clear that f ∼ g is an equivalence relation and we denote by [f ] the

equivalence class of f . If f (p) = 0, g is a representative of g and f ∼ g, it follows
that the zero sets Zf and Zg of f and g coincide in a neighborhood of p and the
quotients f /g and g/f remain bounded where they are defined. Thus, the class [f ]
represents the way in which f (q) approaches 0 as q approaches the zero set Zf p.
Hence, (2.3) can be rephrased by saying that the germs at p, d and d# , of the
functions
d(q) = dist (Z(q), Z(Σ ∩ V )) and d# (q) = dist (Z # (q), Z # (Σ ∩ V ))
are equivalent and write
(2.4) d ∼ d# .
In other words, the equivalence class [d] of the germ at p of the function d(q) is
independent of the choice of the first integrals Z1 , . . . , Zm and it is a local invariant
64
at p ∈ Σ that only depends on the maximally real submanifold Σ ⊂ M and the

locally integrable structure (M, L).
We now express everything in terms of our local coordinates (x, t) (in which
Z(x, t) has the special form Z(x, t) = x + iϕ(x, t) but Z # (x, t) might not). We may
assume that U0 is of the form Brx (0) × BTt (0) if r > 0 and T > 0 are sufficiently
small (note that (1.3) still holds if we shink r and T ) and, to simplify the notation,
write Σ instead of Σ ∩ U0 . Let u(x, t) be the continuous solution in the statement
of Theorem 1.1. If follows from (2.3) that
T
ν
dist Z # (x, τ t ), Z # (Σ) |u(x, τ t )| dxdτ
0 Brx (0)

T ν
≤C dist Z(x, τ t ), Z(Σ) |u(x, τ t )| dxdτ
0 Brx (0)

T ν
≤ C sup dist Z(x, τ t ), Z(Σ) |u(x, τ t )| dxdτ
t ∈Γ0 0 Brx (0)
= C1 < ∞,
so taking the sup in t ∈ Γ0 on the left hand side we see that u satisfies a growth
restriction analogous to (1.3) with Z # in the place of Z. This argument can be
reversed to show that a growth condition in terms of Z # implies a similar a growth
condition in terms of Z, possibly after shrinking r and T .
Remark 2.1: In the special local coordinates (x, t) in which Z = (Z1 , . . . , Zm ) is

written as Z(x, t) = x + iϕ(x, t) and Σ is given by {t = 0}, it is easy to see that
(2.5) dist (Z(x, t), Z(Σ)) |ϕ(x, t) − ϕ(x, 0)| = |Z(x, t) − Z(x, 0)|,
for |x| ≤ r, |t| ≤ T.
This fact will be used in the next section.
Remark 2.2: Since the rank of the map Z : B x × B t → Cm might not be constant,
Z(B x × B t ) is, in general, neither an open set nor a submanifold and may be rather
irregular. Nevertheless, it is arc-connected by piecewise differentiable curves and
this is the main fact we exploited in the proof of Lemma 2.1. If we define a
distance between two points ζ0 , ζ1 ∈ Z(B x × B t ) as the infimum of the lengths of
the piecewise differentiable curves contained in Z(B x × B t ) that join ζ0 to ζ1 , the
arguments in the proof of Lemma 2.1 show that this distance is equivalent to the
Euclidean distance restricted to Z(B x × B t ).
9
3. Proof of the main result
Consider special coordinates (x, t) in which a set of first integrals Z1 , . . . , Zm

have the form (1.1), L is generated by vector fields of the form Lj = ∂
∂tj +
m
k=1 λjk (x, t) ∂xk , j = 1, . . . , n, and Σ = {t = 0}.
∂
In fact (cf. [BCH, Chapter I]), there exist smooth vector fields
m
∂
Mk = μk (x, t) , k = 1 · · · m,
∂x
=1
satisfying Mk Zj = δkj (Kronecker’s delta) such that
∂ ∂ϕkm
Lj = −i (x, t)Mk ,
∂tj ∂tj
k=1
∂
m
∂
= + λjk (x, t) j = 1, · · · , n.
∂tj ∂xk
k=1
The vector fields L1 , . . . , Ln , M1 . . . , Mm are pairwise commuting and span CT M
over the local patch where they are defined. Furthermore, if f is of class C 1 we
have

n
m
(3.1) df = Lj f dtj + Mk f dZk .
j=1 k=1
In view of (2.5), Theorem 1.1 will be a consequence of
Theorem 3.1. Let f (x, t) be a continuous function on the wedge Q = Brx (0) ×
ΓT with edge Σ = Brx (0) × {0} and assume that
(1) Lj f ∈ L∞ (Q), j = 1, . . . , n;
(2) for some ν ∈ N
T
(3.2) sup |ϕ(x, τ t ) − ϕ(x, 0)|ν |f (x, τ t )|dxdτ < ∞.
t ∈Γ0 0 Brx (0)
Then f (x, t) has a distributional limit as t → 0 in ΓT . More precisely, for any test
function ψ(x) ∈ Cc∞ (Brx (0)), the limit

.
bf, ψ = lim f (x, t)ψ(x) dx
t→0
t∈ΓT
exists and defines a distribution of order ν + 1.
The proof of Theorem 3.1 will be carried out in three steps. In the first step
we will assume that f is of class C 1 and will show that the limit exists as t → 0 in
ΓT along a fixed direction. In the second step we will assume that f is of class C 0
but we will still approach 0 along a fixed direction. In the final step we will deal
with the general case.
66
3.1. Step 1. Assume that f ∈ C 1 (Q), fix a point ṫ = (ṫ1 , . . . , ṫn ) ∈ Γ0 and

consider the complex vector field
L(ṫ) = ṫ1 L1 + · · · + ṫn Ln
which is tangent to the m + 1-submanifold

.
Π(ṫ) = {(x, τ ṫ) : x ∈ Brx (0), 0 < τ < T } ⊂ Rm
x × Rt
n
which is an open subset of the linear space generated by Rm

x and ṫ. We may express
the restriction of f (x, t) to Π(ṫ) as

.
f(ṫ) (x, τ ) = f (x, τ ṫ1 , . . . , τ ṫn ), 0 < τ < T,
and regard it as a function in the variables τ, x1 , . . . , xm . It is clear that

∂
m
∂
L(ṫ) = + λk (x, τ, ṫ)
∂τ ∂xk
k=1
since differentiating t = τ ṫ with respect to τ gives

∂ ∂ ∂
= ṫ1 + · · · + ṫn
∂τ ∂t1 ∂tn
∞
and that L(ṫ) f(ṫ) = k ṫk Lk f (x, τ ṫ) ∈ L (Q ∩ Π(ṫ)). We regard L(ṫ) as a single
locally integrable vector field in the m + 1 variables τ, x1 , . . . , xm with first integrals
(ṫ) .
Zj (τ, x) = Zj (τ ṫ, x), j = 1, . . . , m,
that depend on ṫ as a parameter. In particular, L(ṫ) satisfies the hypothesis of

[BCH, Thm. VI.1.3] uniformly in ṫ ∈ Γ̃0 , provided that Γ̃0 is a compact subset of
Γ0 . To see this we will briefly describe below the main steps in the proof of [BCH,
Thm. VI.1.3] that lead us to conclude that the constants involved can be taken
independently of ṫ ∈ Γ̃0 ; a more detailed proof would be a straightforward but long
and tedious repetition of the arguments in Thm.VI.1.3.
Lemma 3.1. Let ψ ∈ Cc∞ (Brx (0)) and fix a positive integer ν. There exists a
smooth function u(ξ + iη, ṫ) defined on Cm × Γ0 such that
(1)
u(Z(x, 0), ṫ) =
ψ(x), |x| < r;

∂u
(2)
(ξ + iη, ṫ)
≤ C dist (ξ + iη, Z(Σ))ν , j = 1, . . . , m, |ξ| ≤ r, |η| ≤ R,

∂ζ j
ṫ ∈ Γ̃0 ,
where R > 0 is a constant such that Z(Brx (0) × BTt (0)) ⊂ Br (0) + iBR (0), C > 0
is a constant and Γ̃0 is a compact subset of Γ0 . Furthermore, the function u is
obtained by applying to ψ(x) a linear partial differential operator P (x, t, Dx , Dt ) of
order ν with smooth coefficients.
11
Corollary 3.1. Let u(ζ, ṫ) be the function considered in the lemma above.
Setting ψ(x, τ, ṫ) = u(Z(x, τ ṫ), ṫ) we have
(1) ψ(x, 0, ṫ) = ψ(x), |x| < r;
(2) L(ṫ) ψ(x, τ, ṫ) ≤ C |ϕ(x, τ t ) − ϕ(x, 0)|ν .
m
If we set M(kṫ) = ∂
=1 μk (x, τ ṫ) ∂x , k = 1, · · · , m, it follows from (3.1) that
1
for any g(x, τ ) of class C ,

m
dg = M(kṫ) g(ṫ) dZ(kṫ) + L(ṫ) g(ṫ) dτ.
k=1
Writing dZ(ṫ) = dZ1 (x, τ ṫ) ∧ · · · ∧ dZm (x, τ ṫ), the exterior derivative of the m−form
g(x, t) dZ(ṫ) is given by
d(g dZ(ṫ) ) = L(ṫ) g(ṫ) dτ ∧ dZ(ṫ) .
We now call Corollary 3.1 and set g(x, τ ) = f(ṫ) (x, τ )ψ(x, τ, ṫ). Using the above
formulas and Stokes’ theorem we get

(3.3) f(ṫ) (x, )ψ(x, , ṫ) dx Z(ṫ) (x, ) =
Brx (0)

f(ṫ) (x, T )ψ(x, T, ṫ) dx Z(ṫ) (x, T )
Brx (0)
T
+ f(ṫ) (x, τ )L(ṫ) ψ(x, τ, ṫ) dτ ∧ dZ(ṫ)
Brx (0)
T
+ L(ṫ) f(ṫ) (x, τ )ψ(x, τ, ṫ) dτ ∧ dZ(ṫ) .
Brx (0)
By (2) of Corollary 3.1 we have
|f(ṫ) (x, τ )L(ṫ) ψ(x, τ, ṫ)| ≤ C|f(ṫ) (x, τ )||ϕ(ṫ) (x, τ ) − ϕ(ṫ) (x, 0)|ν
which shows that |f(ṫ) (x, τ )L(ṫ) ψ(x, τ )| ∈ L1 (Brx (0) × [0, T ]) in view of (3.2). Thus,
the second integral of the right hand side of (3.3) has a limit when 0 and is
bounded by a constant independent of the direction ṫ. The existence of the limit
when 0 of the other two integrals on the right hand side of (3.3) is clear. We
conclude that the limit when 0 of the left hand side of (3.3) exists and

f(ṫ) (x, )ψ(x, , ṫ) dx Z(ṫ) (x, )
≤ C,

Brx (0)
with C > 0 independent of ṫ ∈ Γ̃0 and 0 < ≤ T . We next concatenate ψ(x, τ, ṫ)
and ψ(x) ∈ Cc∞ (Brx (0)), i.e., we find a finite sequence of smooth functions ψ (x, τ, ṫ),
= 0, . . . , ν −1, whose x-support is contained in a fixed compact subset independent
of t such that
68
(1) ψν (x, τ, ṫ) = ψ(x, τ, ṫ);

(2) ψ0 (x, τ, ṫ) = ψ(x);
(3) for = 1, . . . , ν,

if lim f(ṫ) (x, τ )ψ (x, τ ) dx Z(ṫ) (x, τ ) exists, then
τ →0 Brx (0)

lim f(ṫ) (x, τ )ψ−1 (x, τ ) dx Z(ṫ) (x, τ ) exists
τ →0 Brx (0)
and both limits are equal;
(4) for = 1, . . . , ν, a bound

f(ṫ) (x, τ )ψ (x, τ, ṫ) dx Z(ṫ) (x, τ )
≤ C , 0 < τ ≤ T, ṫ ∈ Γ̃0 ,

Brx (0)
implies a bound

f(ṫ) (x, )ψ−1 (x, τ, ṫ) dx Z(ṫ) (x, τ )
≤ C−1 , 0 < τ ≤ T, ṫ ∈ Γ̃0 .

Brx (0)
The construction of the functions ψ in the concatenation is described in detail

in the proof of [BCH, Theorem VI.1.3], where the case of a single vector field is
treated (they are denoted as T g(x, t)). The only difference is that the single vector
field we are dealing with in the present case depends on the parameter ṫ and we
must check that the bounds are uniform with respect to ṫ.
Hence, by descending induction, we obtain for = 0 that

(3.4) lim f(ṫ) (x, τ )ψ(x) dx Z(ṫ) (x, τ ) exists and is equal to
τ →0 Brx (0)

f(ṫ) (x, T )ψν (x, T, ṫ) dx Z(ṫ) (x, T )
Brx (0)
T
+ f(ṫ) (x, τ )L(ṫ) ψν (x, τ, ṫ) dτ ∧ dZ(ṫ)
Brx (0) 0
T
+ L(ṫ) f(ṫ) (x, τ )ψν (x, τ, ṫ) dτ ∧ dZ(ṫ)
Brx (0) 0
(this corresponds to formula (VI.32) in [BCH, p. 281]) and

(3.5)
f(ṫ) (x, τ )ψ(x) dx Z(ṫ) (x, τ )
≤ C0 , 0 < τ ≤ T, ṫ ∈ Γ̃0 .

Brx (0)
Notice that the expression for the limit in (3.4) involves derivatives of order one
of the function ψν (x, τ, ṫ) = ψ(x, τ, ṫ) which, by Lemma 3.1 and its corollary, is a
linear combination with smooth coefficients of derivatives of ψ(x) up to order ν,
so it defines a distribution of order ν + 1. More generally, if g(x, τ ) is smooth on
13
Brx (0) × [0, T ] and the support of x → g(x, τ ) is contained in a compact subset that
does not depend on t ∈ [0, T ] we have

(3.6) lim f(ṫ) (x, τ )g(x, τ ) dxZ(ṫ) (x, τ ) exists
τ →0 Brx (0)
and

(3.7)
f(ṫ) (x, τ )g(x, τ ) dx Z(ṫ) (x, τ )
≤ C, 0 < τ ≤ T, ṫ ∈ Γ̃0 .

Brx (0)
3.2. Step 2. Assume now that f ∈ C 0 (Q). In fact, the proof of Step 1 still
holds for a continuous f but the fact that the restriction f(ṫ) of f to Π(ṫ) satisfies
weakly the equation L(ṫ) f(ṫ) ∈ L∞ and that Stokes’s formula (3.3) is valid requires
some justification (it could be proved, for instance, with the help of Baouendi-
Treves approximation formula). An alternative approach is to regularize f and
apply Step 1 to the regularizations. Let φ ∈ Cc∞ (B), with B the unit ball in Rm+1 ,
1
φ dxdτ = 1 and φδ (x, τ ) = δm+1 φ( xδ , τδ ), δ > 0. For > 0, set f(ṫ) (x, τ ) =
f(ṫ) (x, τ + ). Then, for δ < , f(ṫ) ∗ φδ (x, τ ) is smooth on {τ > 0}. Set g(,δ
ṫ)
(x, τ ) =

f(ṫ) ∗ φδ (x, τ ) ψ(x, τ + , ṫ), Z(ṫ) (x, τ ) = Z(ṫ) (x, τ + ) and
∂ m
∂
L(ṫ) = + λk (x, t + , ṫ) .
∂τ ∂xk
k=1
As in Step 1, we have
d(g(,δ
ṫ)
dZ(ṫ) ) = L(ṫ) g(,δ
ṫ)
dτ ∧ dZ(ṫ)
which we use to obtain the analogue of (3.3) for f(ṫ) ∗ φδ (x, τ ). Then, repetition
of the arguments of Step 1 lead to the analogue of (3.4), (3.5), (3.6) and (3.7) for
f(ṫ) ∗ φδ (x, τ ). If we let δ 0 and invoke Friedrichs’ lemma we derive (3.4) and
(3.5) for f(ṫ) . Finally, we let 0 to get (3.4), (3.5), (3.6) and (3.7) for f(ṫ) itself.
3.3. Step 3. Since ṫ appears on the right hand side of (3.4) the directional
limit seems to depend on the direction ṫ. To show that this is not so, consider for
ψ(x) ∈ Cc∞ (Brx (0)) the function

T (t) = f (x, t)ψ(x) dx, t ∈ ΓT .
Brx (0)
We will show that ∇T is bounded for t ∈ ΓT if t/|t| ∈ Γ̃0 . A standard computation

shows that the derivatives ∂T /∂tj , j = 1, . . . , n in the sense of distributions are
70
given by
∂T
(t) =
∂tj
∂
m
Lj f (x, t)ψ(x) dx + f (x, t) λjk (x, t)ψ(x) dx.
Brx (0) Brx (0) ∂xk
k=1
The first term on the right hand side is bounded because Lj f is bounded. To bound
the second term, write

m
∂
λjk (x, t)ψ(x) = g(x, t) = g(x, τ ṫ)
∂xk
k=1
with τ = |t|, ṫ = t/|t| and apply (3.7). Hence, T (t) is a Lipschitz function and has
a limit as t → 0 on proper subcones of ΓT . Letting t → 0 along a fixed direction ṫ
we see that the limit is given by (3.4). As we have already pointed out, this shows
that det Zx (x, 0) bf (x) is a distribution of order ν + 1 and dividing by det Zx (x, 0)
so is bf (x).
4. Another application
Let (M, L) be a smooth locally integrable structure, Σ ⊂ M a maximally real

submanifold, p ∈ Σ. If Z = (Z1 , . . . , Zm ) is a complete set of first integrals defined
in a neighborhood U of p we write
dΣ,Z (q) = dist (Z(q), Z(Σ ∩ U )), q ∈ U,
δΣ (q) = dist (q, Σ ∩ U ), q ∈ U,
and denote by dΣ,Z and δ Σ their corresponding germs at p. We have already seen
that the vanishing rate of dΣ,Z is an invariant of the pair (Σ, L). Since clearly
dΣ,Z (q) ≤ Cdist (q, Σ ∩ U ) as q → p, we always have dΣ,Z δ Σ . It is a natural
question to ask for which structures the opposite relation also holds, i.e., when
dΣ,Z ∼ δ Σ . This question has a simple answer: this property characterizes CR
structures among locally integrable structures. We recall that L is CR at p if
Lp ∩ Lp = {0} and L is CR on M if L is CR at p for every point p ∈ M. Before
stating the precise characterization result, we will need some facts about local
canonical forms for generators of L⊥ in appropriate local coordinates. As before,
N will denote the dimension of M.
Theorem 4.1. Let (M, L) be a smooth locally integrable structure of rank n

.
and co-rank m. Let p ∈ Ω and d be the real dimension of Tp0 = L⊥ ∗
p ∩ Tp M. Then
15
there is a coordinate system vanishing at p
{x1 , . . . , xν , y1 , . . . , yν , s1 , . . . , sd , t1 , . . . , tn }
and smooth, real-valued functions φ1 , . . . , φd defined in a neighborhood of the origin

and satisfying
φk (0) = 0, dφk (0) = 0, k = 1, . . . , d,
such that the differentials of the functions
.
Zj (x, y) = zj = xj + iyj , j = 1, . . . , ν;
Wk (x, y, s, t) = sk + iφk (z, s, t), k = 1, . . . , d,
span L⊥ in a neighborhood of the origin. In particular we have ν +d = m, ν +n = n

and also
Tp0 = span {ds1 |0 , . . . , dsd |0 }.
Furthermore, L is CR if and only if n = 0 (i.e., there are no t variables). In this
case, we have ν = n and m = n + d.
The theorem above summarizes well known results, see for instance [BCH,
Theorem I.10.1] and [BCH, Section I.15].
We state now the characterization theorem.
Theorem 4.2. Let (M, L) be a smooth locally integrable structure of rank n

and co-rank m. The following conditions are equivalent:
(1) L is CR on a neighborhood of p;
(2) For any maximally real submanifold Σ passing through p and any complete
set Z = (Z1 , . . . , Zm ) of local first integrals defined in a neighborhood of
p, the functions dΣ,Z and δΣ are comparable in a neighborhood of p. In
other words, dΣ,Z ∼ δ Σ ;
(3) For some maximally real submanifold Σ passing through p and some com-
plete set Z = (Z1 , . . . , Zm ) of local first integrals defined in a neighborhood
of p, dΣ,Z ∼ δ Σ .
Proof: Since it is trivial that (2) implies (3) it is enough to show that (1) implies
(2) and that (3) implies (1).
(1) =⇒ (2). Since L is CR in a neighborhood of p we may find a local coordinate
system vanishing at p,
(x1 , . . . , xn , y1 , . . . , yn , s1 , . . . , sd )
72
with 2n + d = N , such that the maximally real submanifold Σ is given by the

equations yj = 0, 1 ≤ j ≤ n and there exist smooth, real valued functions φ1 , . . . , φd
defined in a neighborhood U of p satisfying
(4.1) φk (0) = 0, dφk (0) = 0, k = 1, . . . , d,
such that the differential of the functions
Zj = xj + iyj , j = 1, . . . , n;
Wk = sk + iφk (x, y, s), k = 1, . . . , d,
form a complete set of first integrals of L in a neighborhood of the origin. Assume

that U is the cube |x| < 1, |y| < 1, |s| < 1. Given a point q = (x, y, s) ∈ U we have
that Z(q) = (x + iy, s + iφ(x, y, s)) and Z(Σ) is given by {(x + i0, s + iφ(x, 0, s))}.
Hence, δΣ (q) = |y| and
dΣ,Z (q) |y| + |φ(x, y, s) − φ(x, 0, s)| = |y| + O(|y|2 )
where we have used (4.1) in the second equality. This shows what we wanted for
this special choice of Z and the case of a general first integral Z # follows from (2.4)
(3) =⇒ (1). We will show that if (1) does not hold then (3) does not hold either.
Let Σ be a maximally real submanifold that is not CR on any neighborhood of p.
We may choose local coordinates defined on an open neighborhood U of p
{x1 , . . . , xν , y1 , . . . , yν , s1 , . . . , sd , t1 , . . . , tn }
with the properties described in Theorem 4.1 such that Σ is given by the equations
y = 0, t = 0. Notice that n ≥ 1 because Σ is not CR on U . For q = (x, y, s, t),
δΣ (q) = (|y|2 + |t|2 )1/2 while
dΣ,Z (q) |y| + |φ(x, y, s, t) − φ(x, 0, s, 0)| |y| + O(|y|2 + |t|2 ).
Taking a sequence of points qk = (0, 0, 0, tk ) ∈ U \ Σ with tk → 0, we see that

dΣ,Z (qk )/δΣ (qk ) → 0 as k → ∞ so dΣ,Z ∼ δ Σ . Invoking (2.4) we have as well that
dΣ,Z # ∼ δ Σ for any other first integral Z # .
References
[BER] M. S. Baouendi, P. Ebenfelt, and L. P. Rothschild, Real Submanifolds in Complex Space
and their Mappings, Princeton Math. Ser. 47(1999), Princeton Univ. Press.
[BT] M. S. Baouendi and F. Treves, A property of the functions and distributions annihilated by
a locally integrable system of complex vector fields, Ann. of Math. 113 (1981), 387–421.
[B] E. Bär, Um teorema de F. e M. Riesz para um sistema de campos vetoriais de co-posto
um, Tese de doutorado, UFScar, São Carlos, 2008 (in Portuguese).
[BCH] S. Berhanu, P. Cordaro and J. Hounie, An Introduction to Involutive Structures, Cam-
bridge University Press, 2008.
17
[BH1] S. Berhanu and J. Hounie, An F. and M. Riesz theorem for planar vector fields, Math.
Ann. 320 (2001), 463–485
[BH2] S. Berhanu and J. Hounie, On boundary properties of solutions of complex vector fields,
J. Funct. Anal. 192 (2002), 446–490.
[BH3] S. Berhanu and J. Hounie,Traces and F. and M. Riesz theorem for planar vector fields,
Annales de l’institut Fourier 53 (2003), 1425–1460.
[EG1] M. Eastwood and R. Graham, Some involutive structures in analysis and geometry, Trends
Math., Birkhäuser, (1997), 25–38.
[EG2] M. Eastwood and R. Graham, The involutive structure on the blow-up of Rn in Cn , Comm.
Anal. Geom. 7, (1999), 609–622.
[HJ] N. Hanges and H. Jacobowitz, Involutive structures on compact manifolds, Amer. Jour.
Math. 177, (1995),491–522.
[H1] L. Hörmander, The Analysis of linear partial differential operators I, Second edition,
Springer-Verlag, (1990).
[S] R. Szoke, Involutive structures on the tangent bundle of symmetric spaces, Math. Ann.
319, (2001), 319–348.
[T] F. Treves, Hypo-analytic structures, Princeton University Press, (1992).
Departamento de Matemática, Universidade Federal de São Carlos, São Carlos,

SP, 13565-905, Brasil
E-mail address: hounie@dm.ufscar.br
ICMC, Universidade de São Paulo, São Carlos, SP, 13560-970, Brasil

E-mail address: evandro@icmc.usp.br
Volume 550, 2011
Chern-Moser operators and weighted jet determination

problems
Martin Kolář and Francine Meylan
Abstract. We use the Chern-Moser approach for hypersurfaces of finite type

in Cn+1 in the sense of Kohn and Bloom-Graham to derive results on jet
determination for the stability group of such hypersurfaces. We then apply
these results to a “model” case to obtain a complete description of its stability
group.
1. Introduction
The Chern-Moser operator ([8]) provides a powerful tool to understand local
CR geometry of Levi nondegenerate manifolds. For instance, the celebrated Chern-
Moser normal form construction essentially reduces to the analysis of its kernel and
its image.
It has been a long-standing question whether these methods and techniques
can be adapted and applied to Levi degenerate manifolds ([10], [20]). Note that
the Levi degenerate case presents completely new challenges, which are closer to
algebraic, rather than to differential geometry.
Starting with the work of Kohn [11], the study of Levi degenerate manifolds has
lead to major advances both in analysis and geometry, in particular in microlocal
analysis, subelliptic multiplier ideal sheaves, or in the work of Baouendi, Ebenfelt,
Rothschild [1] and others. Recently, techniques developed in the degenerate case
were applied to solve major problems in algebraic geometry by Siu [16] and others.
In this paper we show how to adapt the Chern-Moser techniques in the context
of the jet determination problem to the class of finite type hypersurfaces (in the
sense of Kohn and Bloom-Graham). It can be viewed as the first step for a more
general approach, which will allow us to understand the jet determination problem
as well as the stability group itself for hypersurfaces of finite Catlin multitype ([7],
[12], [13]).
The paper is organized as follows. In Section 2, we recall the notion of Bloom-
Graham finite type and its basic properties. We also recall the notion of model
hypersurface associated to M, called MH , and then define the notion of weighted
2010 Mathematics Subject Classification. Primary 32V35, 32V40.

The first author was supported by a grant of the GA ČR no. 201/08/0397 .
The second author was supported by by Swiss NSF Grant 2100-063464.00/1 .
2011
c 0000
c Mathematical
holder)
1
75
76
2 MARTIN KOLÁŘ AND FRANCINE MEYLAN
jets. In Section 3, we show how to reduce the study of the weighted jet determina-
tion problem for Aut(M, p), the stability group of M, to the study of hol(MH , p),
the set of real-analytic infinitesimal CR automorphisms of MH at p (see Theorem
3.10). In Section 4, we introduce the notion of rigid vector fields and prove results
regarding the jet determination problem for hol(MH , p) (see Theorem 4.6). In Sec-
tion 5, we illustrate these techniques by describing completely hol(MH , p), where
MH is the model hypersurface of the form
(1.1) {(z1 , z2 , w) ∈ C3 | Im w = z1 z̄2l + z2 l z̄1 , l > 1}.
We then obtain an effective bound for the weighted jet determination problem for
Aut(M, p) for all smooth hypersurfaces with the model given by (1.1) (see Theorem
5.8).
2. Preliminaries
Let M ⊆ C n+1
be a smooth hypersurface, and p ∈ M be a point of finite type
m ≥ 2 in the sense of Kohn and Bloom-Graham ([1], [4], [11]).
We consider local holomorphic coordinates (z, w) vanishing at p, where
z = (z1 , z2 , ..., zn ) and zj = xj + iyj , w = u + iv. The hyperplane {v = 0} is
assumed to be tangent to M at p, hence M is described near p as the graph of a
uniquely determined real valued function
(2.1) v = ψ(z1 , . . . , zn , z̄1 , . . . , z̄n , u), dψ(p) = 0.
Using for instance a result of [1] (see Theorem 4.2.16), we may assume that
(2.2) ψ(z1 , . . . , zn , z̄1 , . . . , z̄n , u) = Pm (z, z̄) + o(u, |z|m ),
where Pm (z, z̄) is a homogeneous polynomial of degree m with no pluriharmonic
terms. Coordinates which provide such a description will be called standard coor-
dinates.
Following [8], we assign natural weights to the variables. The variables w and
u are given weight one, while the tangential variables z1 , . . . , zn are given weight
1
m.
Definition 2.1. Let m be the type of M at p. The weighted degree κ of a

monomial
q(z, z̄, u) = cαβl z α z̄ β ul , l ∈ N,
is defined as
1
n
κ := l + (αi + βi ).
m i=1
Definition 2.2. A polynomial Q(z, z̄, u) is weighted homogeneous of weighted

degree κ if it is a sum of monomials of weighted degree κ.
Remark 2.3. Note that according to this definition, Pm is a weighted homoge-
neous polynomial of weighted degree one, while the o(u, |z|m ) terms are of weighted
degree bigger than one.
Definition 2.4. Let M be given by (2.2). We define a model hypersurface
MH associated to M at p by
(2.3) MH = {(z, w) ∈ Cn+1 | v = Pm (z, z̄)}.
CHERN-MOSER OPERATORS 77
3
Note that standard coordinates are not unique, but it is easily shown that all
models are equivalent by a linear transformation (see the remark following formula
(3.5)).
Definition 2.5. Let (z, w) ∈ Cn+1 be standard coordinates and let F :
Cn+1
−→ C be a holomorphic function given in these coordinates. The weighted
jet of F at p of weighted order κ is given by the following set
∂ α+β F 1
(2.4) { (p), |α| + |β| ≤ κ}.
∂z α ∂wβ m
Definition 2.6. Let F1 , F2 : Cn+1 −→ C be two holomorphic functions given
in some standard coordinates. We say that F1 and F2 are weighted equivalent
modulo κ at p if
∂ α+β F1 ∂ α+β F2 1
(p) = (p), |α| + |β| ≤ κ.
∂z α ∂wβ ∂z α ∂wβ m
We have the following lemma, whose proof is immediate.
Lemma 2.7. The notion of weighted equivalence modulo κ at p is independent
of the choice of standard coordinates.
3. The basic identity

Let M ⊂ Cn+1 be a real analytic hypersurface of finite type m, given be (3),
which we rewrite as
(3.1) v = ψ(z, z̄, u) = P (z, z̄) + owt (1),

where P = Pm and owt (1) denotes terms of weighted degree bigger than one.
Recall the following definitions.
Definition 3.1. We denote by Aut(M, p) the stability group of M, that is,
those germs at p of biholomorphisms mapping M into itself and fixing p.
Definition 3.2. We denote by hol(M, p) the set of germs of real-analytic
infinitesimal CR automorphisms of M at p.
Remark 3.3. ([1]) Recall that X ∈ hol(M, p) if and only if there exists a germ
Z at p of a holomorphic vector field in Cn+1 such that ReZ is tangent to M and
X = ReZ|M . By abuse of notation, we also say that Z ∈ hol(M, p).
1
Denote Θ = m Z, i.e. Θ is the set of all integer multiples of m
1
.
We decompose the formal Taylor expansion of ψ, denoted by Ψ, into weighted
homogeneous polynomials Ψν of weighted degree ν,

Ψ= Ψν .
ν∈Θ
Notice, using (3.1), that Ψν = 0, for ν < 1, and Ψ1 = P .

Let h ∈ Aut(M, p). We write h in the form
zj = zj + f j (z, w)
(3.2)
w = w + g(z, w),
where f j and g do not contain constant terms.
78
Putting f = (f 1 , . . . , f n ), we consider the mapping given by

(3.3) T = (f, g),
and, again, decompose each power series f j and g into weighted homogeneous
polynomials f j μ and gμ of weighted degree μ,

fj = f j μ, g= gμ .
μ∈Θ μ∈Θ
Let v = ψ(z , z̄ , u ) be the defining equation of M in the coordinates (z , w ),

of the form given by (3.1), that is,

(3.4) ψ(z , z̄ , u ) = P (z , z̄ ) + owt (1).
Since h ∈ Aut(M, p), substituting (3.2) into v = ψ(z , z̄ , u ) we obtain the trans-
formation formula
ψ(z + f (z, u + iψ(z, z̄, u)), z + f (z, u + iψ(z, z̄, u)),u +
(3.5)
+ Re g(z, u + iψ(z, z̄, u)) = ψ(z, z̄, u) + Im g(z, u+iψ(z, z̄, u)).
Note that if a transformation (3.2) maps standard coordinates into standard
coordinates, then g does not contain any terms of order less or equal to one. On
the other hand, if g has this property, then by (3.5) the leading polynomial in
coordinates (z , w ) is obtained from the leading polynomial in coordinates (z, w)
by the linear part of the transformation.
Expanding (3.5) we consider terms of weight μ ≥ 1. We get

n
2Re Pzj (z, z̄)f j μ−1+ m1 (z, u+iP (z, z̄)) =
(3.6) j=1
= Im gμ (z,u + iP (z, z̄)) + . . .
∂P
where dots denote terms depending on f j ν−1+ m1 , gν , ψν , for ν < μ , and Pzj = ∂zj
(there are no dots for μ = 1).
We are now in position to give the definition of the generalized Chern-Moser oper-
ator.
Definition 3.4. The generalized Chern-Moser operator, denoted by L, is de-
fined by
(3.7) L(f, g) =
⎧ ⎫
⎨ n
∂P j ⎬
= Re ig(z, u + iP (z, z̄)) + 2 f (z, u + iP (z, z̄)) .
⎩ ∂zj ⎭
j=1
The following lemma shows the relation between the kernel of L and the infinitesi-
mal CR automorphisms of the model hypersurface given by (2.3). (See also [8] for
the analogue in the Levi non degenerate case).
Lemma 3.5. Let L be given by (3.7) and let (f, g) be given by (3.2). Then
(f, g) lies in the kernel of L if and only if the vector field
n
∂ ∂
Y = f j (z, w) + g(z, w)
j=1
∂z j ∂w
lies in hol(MH , p), where MH is given by (2.3).

5
Proof. Applying v − P to Y, we obtain

(3.8) Re Y (v − P )|MH =
⎧ ⎫
1 ⎨ n
∂P j ⎬
= − Re ig (z, u + iP (z, z̄)) + 2 f (z, u + iP (z, z̄)) =
2 ⎩ ∂zj ⎭
j=1
1
= − L(f, g).
2
The conclusion follows, using the characterisation of the set of germs of real-analytic
infinitesimal CR automorphisms.

We have the following theorem which shows how to reduce the weighted jet
determination problem from Aut(M, p) to aut(MH , p).
Proposition 3.6. Let h = (z + f, w + g) ∈ Aut(M, p) be as in (3.2). Let

(f, g) = (f, g)μ ,
where
(f, g)μ = (f 1 μ−1+ m1 , . . . f n μ−1+ m1 , gμ ).
Let μ0 be minimal such that (f, g)μ0 = 0. Then the (non trivial vector) field

n
∂ ∂
(3.9) Y = f j μ0 −1+ m1 + gμ0
j=1
∂zj ∂w
lies in hol(MH , p), where MH is given by (2.3).

Proof. Using (3.6) and the definition of μ0 , we obtain
L((f, g)μ0 ) = 0.
Therefore, using Lemma (3.5), we obtain that
n
∂ ∂
Y = fμj −1+ 1 + gμ0
0 m ∂zj ∂w
j=1
belongs to hol(MH , p). This achieves the proof of the proposition.

Definition 3.7. We say that the vector field

n
∂ ∂
Y = Fj (z, w) + G(z, w)
j=1
∂z j ∂w
has homogeneous weight μ (≥ −1) if Fj is a weighted homogeneous polynomial of
1
weighted degree μ + m , and G is a homogeneous polynomial of weighted degree
μ + 1.
Remark 3.8. The weights introduce a natural grading on hol(MH , p) in the
following sense.
Writing hol(MH , p) as
hol(MH , p) = ⊕μ+1∈Θ Gμ ,
where Gμ consists of weighted homogeneous vector fields of weight μ, we observe
that each weighted homogeneous component Xμ ∈ Gμ of X ∈ hol(MH , p) lies also
in hol(MH , p). The reason is that v − P is weighted homogeneous.
80
We recall the following definition.

Definition 3.9. A real-analytic hypersurface M ⊂ Cn+1 is holomorphically
nondegenerate at p ∈ M if there is no germ at p of a holomorphic vector field X
tangent to M.
Gathering all the previous results, we obtain the following theorem.
Theorem 3.10. Let M ⊂ Cn+1 be a smooth hypersurface of finite type m
given by (3.1), where the defining equation is given in some standard coordinates.
Let MH be the model hypersurface given by (2.3). Assume that there exist μ0 such
that
(3.10) hol(MH , p) = ⊕−1≤μ<μ0 Gμ
Then any h = (z + f, w + g) ∈ Aut(M, p) given by (3.2) such that (f, g)μ = 0 for
all μ < μ0 is the identity map.
Proof. Apply Proposition 3.6.

Remark 3.11. In the light of Theorem 3.10, we see that in order to study
the weighted jet determination problem for Aut(M, p), it is enough to study the
weighted jet determination problem for hol(MH , p).
4. Rigid vector fields

In this section, we describe an important class of vector fields X ∈ hol(MH , p),
called rigid vector fields. As we will see, they play a crucial role in the study of the
weighted jet determination problem for hol(MH , p).
As before, let M ⊂ Cn+1 be a smooth hypersurface of finite type m given by (3.1)
in some standard coordinates (z, w), that is,
(4.1) v = ψ(z, z̄, u) = P (z, z̄) + owt (1), w = u + iv.
We have the following definition.
Definition 4.1. Let X be a vector field in hol(MH , p) of the form

n
∂ ∂
X= f j (z, w) + g(z, w) .
j=1
∂zj ∂w
We say that X is rigid if f , . . . , f n , g are all independent of the variable w.

1
Remark 4.2. Note that if X ∈ hol(MH , p) is rigid homogeneous of nonnegative

weight, then g = 0, since P has no pluriharmonic terms.
Remark 4.3. Note that the rigid vector field W, homogeneous of weight −1,
given by
∂
(4.2) W =
∂w
lies in hol(MH , p).
7
Definition 4.4. The weighted homogeneous vector field of weight 0 defined

by
1
n
∂ ∂
(4.3) E= zj +w .
m j=1 ∂zj ∂w
is called the weighted Euler field.
Remark 4.5. Note that E is a non rigid vector field lying in hol(MH , p).
We have the following theorem.
Theorem 4.6. Let MH be holomorphically nondegenerate, and let X ∈ hol(MH , p)
be a nonzero rigid vector field given by
n
∂
(4.4) f j (z1 , . . . zn ) ,
j=1
∂zj
where f j are germs of holomorphic functions at p. Then the coefficients f j are sums
of monomials of weighted degree less than one.
Remark 4.7. Note that, using Remark 3.9, the weighted homogeneous com-
ponents of X are of weight less than 1 − m
1
.
The proof of the theorem is a direct application of the following lemma.
Lemma 4.8. Let MH be holomorphically nondegenerate, and let X ∈ hol(MH , p)
be a rigid vector field given by (4.4). If
∂αf j 1
α
(0) = 0, |α| < 1,
∂z m
then X = 0.
Proof. By assumption, we have
n
∂
(4.5) (Re f j (z) ) (Im w − P (z, z̄)) = 0.
j=1
∂z j
By the reality of P , we may rewrite (4.5) as

⎛ ⎞
n
∂P
(4.6) Re ⎝ fj (z, z̄)⎠ = 0.
j=1
∂zj
Write

n
∂P
(4.7) f j (z) (z, z̄) = Bαα̂ z α z̄ α̂ .
j=1
∂zj
α,α̂
Using (4.6), we obtain

(4.8) Bαα̂ = −Bα̂α .
On the other hand, since P is of weighted degree one, we have
∂P 1
(4.9) weight( )=1− < 1.
∂zj m
First we claim that Bα,α̂ are zero for all α, α̂. By contradiction, assume there is
α, α̂ with Bα,α̂ = 0. By assumption, |α| ≥ m whereas |α̂| < m, using (4.9). On
82
the other hand, by (4.8), we obtain that there exists a nonzero term with weight
in z less than one, and z̄ greater than or equal to one. That gives a contradiction,
hence all Bα,α̂ are zero. Therefore we obtain that X itself is complex tangent to
MH , and since MH is holomorphically nondegenerate, X = 0.

We have the following lemma.
Lemma 4.9. Let X ∈ hol(MH , p) be a weighted homogeneous vector field, and
let W ∈ hol(MH , p) given by (4.2). There exists an integer k ≥ 1, and a rigid vector
field Y ∈ hol(MH , p) such that [. . . [[X; W ]; W ]; . . . ]; W ] = Y, where the string of
brackets is of length k.
Proof. Observe that the effect of taking the bracket of X with W is simply
differentiation of the coefficient with respect to w. Also note that
(Re[X; W ])(v − P (z, z̄)) = [ReX, ReW ](v − P (z, z̄)).

Definition 4.10. We say that X ∈ hol(MH , p) is a k-integration of a rigid

vector field Y ∈ hol(MH , p) if the string of brackets described in the above lemma
is of length k.
Remark 4.11. By the above lemma, the general case will be reduced to the
rigid case by taking sufficiently many commutators with the vector field W . The
problem reduces then to
(i) describing rigid vector fields
(ii) analysing to what extent rigid fields can be “integrated”.
5. A model case
In this section we illustrate the results obtained in Section 4 by describing com-
pletely hol(MH , p), where MH , the model hypersurface associated to M ⊂ C3 , is
given by
(5.1) MH = {(z1 , z2 , w) ∈ C3 | Im w = z1 z̄2l + z2 l z̄1 , l > 1, }
and obtaining an efficient bound for the jet determination problem for hypersurfaces
with this model (Theorem 5.8). Note that MH is the preimage of the standard
hyperquadric under the mapping (z1 , z2 , w) → (z1 , z2l , w). The infinitesimal CR
automorphisms described below correspond precisely to the automorphisms of the
hyperquadric which can be pulled back by this mapping.
Remark 5.1. M whose model hypersurface is given by (5.1) is of finite type
m = l + 1.
We have the following lemmas.
Lemma 5.2. Let MH be given by (5.1). Then MH admits only the following
homogeneous rigid vector fields: The rigid vector fields
∂
W =a , a ∈ R,
∂w
of weight −1. The rigid vector fields
∂ ∂
(5.2) Z=a + 2iāz2 l , a ∈ C,
∂z1 ∂w
9
−1
of weight . The rigid vector fields given by
l+1
∂ ā ∂
X = az1 − z2 , a ∈ C,
∂z1 l ∂z2
of weight 0. The rigid vector fields given by
∂
Y = aiz2 l , a ∈ R,
∂z1
l−1
of weight l+1 .
Proof. A direct computation, using (3.8) and Theorem 4.6, gives all homogeneous
rigid vector fields on MH .

Lemma 5.3. Let X ∈ hol(MH , p) and Y be vector fields of nonnegative homo-

geneous weight of the form

2
∂ 2
∂
(5.3) X= aj (z) , Y = bj (z) .
j=1
∂zj j=1
∂z j
Assume that the following equation holds
(5.4) P (z, z̄)ImX(P ) + ReY (P ) = 0,

where
P (z, z̄) = z1 z̄2l + z2 l z̄1 .
Then X = 0.
Proof. Suppose X = 0. Since MH is holomorphically nondegenerate, we get
ImX(P ) = 0.
By assumption, the minimal (non weighted) order of P (z, z̄)ImX(P ) with respect
to z̄ is at least two and the term giving the order contains z̄1 . On the other hand,
the order of ReY (P ) is either one, or, if b2 = 0, the term giving the order has no
z̄1 , since b1 is of order at least l + 1. This is a contradiction. Hence, X = 0 and
ReY (P ) = 0. This achieves the proof of the lemma.

Proposition 5.4. Let X ∈ hol(MH , p) be a nonzero rigid vector field of non-

negative weight of the form

2
∂
(5.5) X= f j (z) .
j=1
∂zj
Then there exists no nonzero vector field Y ∈ hol(MH , p) such that

[Y, W ] = X.
84
Proof. Integrating the coefficients of X given by (5.5) with respect to w, we

obtain that Y would have to be of the form
2
∂
2
∂ ∂
Y =w f j (z) + φj (z) + φ(z) .
j=1
∂zj j=1 ∂zj ∂w
Suppose that there exists such a Y ∈ hol(MH , p). We then have

⎛ ⎞
2
∂P 2
∂P
Re Y (P − v) = Re ⎝2 f j (z)(u + iP (z, z̄)) + 2 φj (z) + iφ(z)⎠ =
j=1
∂z j j=1
∂zj
⎛ ⎞
2
∂P 2
∂P
= Re ⎝2 f j (z)iP (z, z̄) + 2 φj (z)⎠ − Im φ(z) = 0,
j=1
∂zj j=1
∂zj
where we have used Re X(P − v) = 0. The first summand contains only mixed
terms, while the second summand is pluriharmonic. It implies that φ(z) = 0.
Further, the first term gives
⎛ ⎞
2 2
∂P j ⎠ ∂P
(5.6) −P (z, z̄)Im ⎝2 f (z) + Re 2 φj (z) = 0.
j=1
∂zj j=1
∂zj
Using the above lemma, we conclude that X = 0. This is a contradiction. This

achieves the proof of the proposition.

Proposition 5.5. Let W ∈ hol(MH , p) be given by (4.2). There exists no

vector field lying in hol(MH , p) that is a 3-integration of W.
Proof. By integrating W we obtain a field of the form

2
∂ ∂
(5.7) φj (z) + (w + φ(z)) .
j=1
∂zj ∂w
Applying (5.7) to P − v we obtain

2
∂P
(5.8) Re (2 φj (z) ) − P (z, z̄) + Im φ(z) = 0.
j=1
∂zj
Since the first two terms are mixed, we obtain φ(z) = 0.

Therefore the 1-integration of W satisfies the following equation

2
∂P
(5.9) Re (2 φj (z) ) = P (z, z̄).
j=1
∂zj
Integrating (5.7), we obtain a field of the form

2
∂ 1 2 ∂
(5.10) (φj (z)w + ψj (z)) + w + φ(z) .
j=1
∂zj 2 ∂w
Appying (5.10) to P − v, we obtain

∂P
(5.11) −P (z, z̄)Im 2φj (z) +
∂zj
11
∂P
+Re ψj (z) + Re iφ(z) = 0.
∂zj
Since the first two summands contain only mixed terms, we obtain φ(z) = 0. Hence,
we obtain
⎛ ⎞
2 2
⎝ ∂P ⎠ ∂P
(5.12) −P (z, z̄)Im 2 φj (z) + 2Re ψj (z) = 0.
j=1
∂zj j=1
∂zj
Using the same argument as in Lemma 5.3, we obtain in particular that

⎛ ⎞
2
∂P
(5.13) Im ⎝2 φj (z)⎠ = 0, ψj = 0, φ = 0.
j=1
∂zj
Hence the 2-integration of W satisfies the following equation

2
∂P
(5.14) 2 φj (z) = P (z, z̄), ψj = 0, φ = 0.
j=1
∂zj
Integrating (5.10), and using (5.14), we obtain a field of the form

2
1 2 ∂ 1 3 ∂
(5.15) Y = w φj (z) + ψj (z) + w + φ(z) .
j=1
2 ∂zj 6 ∂w
Applying (5.15) to P − v, and using (5.14), we get (as above φ = 0),

2
1
2
∂P
(5.16) Re (u2 − P 2 + 2iuP )P + Re ψj (z)
j=1
2 j=1
∂zj
1
−
(3u2 P − P 3 ) = 0.
12
Putting u = 0 in (5.16), we obtain
5 3 n
∂P
(5.17) − P + Re (ψj (z)) = 0.
12 j=1
∂zj
By the same arguments as for Lemma 5.3, we conclude that (5.17) is impossible.
Hence, there is no 3-integration of W. This achieves the proof of the proposition.

Proposition 5.6. Let Z ∈ hol(MH , p) be given by (5.2). There exists no

vector field lying in hol(MH , p) that is a 2-integration of Z.
Proof. Integrating
∂ ∂
(5.18) Z=a + 2iāz2 l , a ∈ C,
∂z1 ∂w
we obtain a vector field of the form
∂ 2
∂ ∂
(5.19) aw + φj (z) + (w2iāz2 l + φ(z)) .
∂z1 j=1 ∂zj ∂w
86

∂P 2
∂P
(5.20) Re a(u + iP ) + Re φj (z) +
∂z1 j=1
∂zj

l
−1
+Re (u + iP )2iāz2 + φ(z) = 0.
2i
Since the first two summands contain only mixed terms, we obtain φ(z) = 0.
We may rewrite (5.20), using the hypothesis, as

2
∂P
(5.21) −P (z, z̄)Im Z(P ) + Re φj (z) = 0.
j=1
∂zj
Using (5.21), we obtain that the 1-integration of Z satisfies

(5.22) φj (z) = 0, φ(z) = 0.
Integrating (5.17), and using (5.22), we obtain a field of the form
1 2 ∂ 2
∂ 2
∂
(5.23) aw + wφj (z) + ψj (z)
2 ∂z1 j=1 ∂zj j=1 ∂zj

1 2 l ∂
+ w 2iāz2 + φ(z) .
2 ∂w
1 ∂P 2
∂P
(5.24) Re a(u2 − P 2 + 2iuP ) + Re (u + iP )φj (z)
2 ∂z1 j=1
∂zj

2
∂P 1 2 1
+Re ψj (z) + Re (u − P 2 + 2iuP )2iāz2 l + φ(z) = 0.
j=1
∂zj 2 2i
Since the first two summands contain only mixed terms, we obtain φ(z) = 0.
We may rewrite (5.24), using the hypothesis, as

2
∂P 2
∂P
(5.25) −u P (z, z̄)Im Z(P ) + Re (u + iP )φj (z) + Re ψj (z) = 0.
j=1
∂zj j=1
∂zj
Putting u = 0 in (5.25), we obtain

2
∂P 2
∂P
(5.26) −P (z, z̄)Im φj (z) + Re ψj (z) = 0.
j=1
∂zj j=1
∂zj
Using Lemma 5.3, we obtain that

2
∂P
(5.27) Re i φj (z) = 0.
j=1
∂zj
But, using Theorem 4.6, we obtain that
φj (z) = 0.
This contradicts (5.22). Hence, there is no 2-integration of Z. This achieves the
proof of the proposition.

13
Remark 5.7. Following the proofs of the above propositions, we obtain the
remaining nonrigid vector fields in hol(MH , p),
W1 = λE + X, λ ∈ R,
where E is the Euler field and X is given by Lemma 5.2,

∂ 1 ∂ 1 2 ∂
W2 = λ (z1 + z2 )w + w , λ ∈ R,
∂z1 l ∂z2 2 ∂w
∂ ∂ 1 ∂ ∂
Z1 = aw − iāz1 z2 l − iā z2 l+1 + 2iāz2 l w , a ∈ C.
∂z1 ∂z1 l ∂z2 ∂w
We have then the following theorem
Theorem 5.8. Let M ⊂ C3 be a smooth hypersurface given by (3.1), whose
model hypersurface MH is given by
(5.28) MH = {(z1 , z2 , w) ∈ C3 | Im w = z1 z̄2l + z2 l z̄1 , l > 1.}
Let h ∈ Aut(M, 0) given by (3.2), and T given by (3.3) satisfying
∂ α+β T 1
(5.29) { (0) = 0 |α| + |β| ≤ 2}.
∂z α ∂wβ l+1
Then h is the identity map.
Proof. Apply Theorem 3.10, Lemma 4.9, Lemma 5.2, Proposition 5.4, Proposition
5.5 and Proposition 5.6.

References
[1] Baouendi, M. S., Ebenfelt, P., Rothschild, L. P., Real Submanifolds in Complex Space and
Their Mappings, Princeton Mathematical Series, (1999).
[2] Baouendi, M. S., Ebenfelt, P., Rothschild, L. P., Convergence and finite determination of
formal CR mappings, J. Amer. Math. Soc. 13, (2000), 697-723.
[3] Baouendi, M. S., Ebenfelt, P., Rothschild, L. P., Local geometric properties of real submani-
folds in complex space, Bull. Amer. Math. Soc. (N.S.) 37 3 (2000), 309–336.
[4] Bloom, T. and Graham, I., On ”type” conditions for generic real submanifolds of C n , Invent.
Math. 40 (1977), no. 3, 217–243.
[5] Cartan, E., Sur la géométrie pseudo-conforme des hypersurfaces de deux variables complexes,
I , Ann. Math. Pura Appl. 11 (1932), p. 17–90.
[6] Cartan, E., Sur la géométrie pseudo-conforme des hypersurfaces de deux variables complexes,
II, Ann.Scoula Norm. Sup. Pisa 1 (1932), p. 333–354.
[7] Catlin, D., Boundary invariants of pseudoconvex domains, Ann. Math. 120 (1984), 529–586.
[8] Chern, S. S. and Moser, J., Real hypersurfaces in complex manifolds, Acta Math. 133 (1974),
219–271.
[9] D’Angelo, J., Orders od contact, real hypersurfaces and applications, Ann. Math. 115 (1982),
615–637.
[10] Ebenfelt, P., New invariant tensors in CR structures and a normal form for real hypersur-
faces at a generic Levi degeneracy, J. Differential Geom. 50 (1998), 207–247.
[11] Kohn, J. J., Boundary behaviour of ∂¯ on weakly pseudoconvex manifolds of dimension two,
J. Differential Geom. 6 (1972), 523–542.
[12] Kolář, M., The Catlin multitype and biholomorphic equivalence of models, Int. Math. Res.
Not. (2010), 3530–3548.
[13] Kolář, M., Meylan, F., Effective bounds for the weighted jet determination problem, preprint
88
[14] Kolář, M., Normal forms for hypersurfaces of finite type in C2 , Math. Res. Lett. 12 (2005),
p. 523-542.
[15] Poincaré, H., Les fonctions analytiques de deux variables et la représentation conforme
Rend. Circ. Mat. Palermo 23 (1907), p. 185-220.
[16] Siu, Y.-T., Invariance of plurigenera, Invent. Math. 134 (1998), 661–673.
[17] Vitushkin, A. G., Real analytic hypersurfaces in complex manifolds, Russ. Math. Surv. 40
(1985), p. 1-35.
[18] Webster, S. M., On the Moser normal form at a non-umbilic point, Math.Ann 233 (1978),
p. 97-102.
[19] Wells, R. O., Jr., The Cauchy-Riemann equations and differential geometry, Bull. Amer.
Math. Soc. (N.S.) 6 (1982), 187–199.
[20] Wong, P., A construction of normal forms for weakly pseudoconvex CR manifolds in C2 ,
Invent. Math. 69 (1982), p. 311-329.
Department of Mathematics and Statistics, Masaryk University, Kotlarska 2, 611 37

Brno, Czech Republic
E-mail address: mkolar@math.muni.cz
Department of Mathematics, University of Fribourg, CH 1700 Perolles, Fribourg

E-mail address: francine.meylan@unifr.ch
Volume 550, 2011
Jet embeddability of local automorphism groups

of real-analytic CR manifolds
Bernhard Lamel
Abstract. We give a survey of results establishing Lie group structures on

the local automorphism groups of real-analytic CR manifolds. These results
are obtained via “jet parametrizations”, which have been extensively stud-
ied in the course of the last 10 years, recently giving a characterization of
finite-dimensionality of the automorphism groups in the case of minimal CR
manifolds. We discuss the history and context of these results.
1. Introduction
Our goal in this paper is to provide a survey of results on the existence of Lie
group structures on the local automorphism groups of real-analytic CR manifolds,
with an emphasis of so-called “parametrizations”. This is meant to complement
the existing survey on unique determination by Zaitsev [30] and the chapter in the
book of Baouendi, Ebenfelt, and Rothschild [5] with more recent results. We will
also try to give some geometric motivation and description of the proofs of these
results (without going into details).
Given a geometric structure, understanding its automorphism group is of fun-
damental importance for its study. If one follows Klein’s approach [18], the study
a geometric structure actually is the study of its group of automorphisms. This
remains to some degree true in a setting where one considers a geometric structure
which is induced by a system of PDE’s, as in the Cauchy-Riemann (CR) setting;
however, such structures are much more rigid than traditional ones, and there-
fore carry invariants which cannot be obtained from their automorphisms (as there
might not be any).
The traditional approach to studying the structure of the automorphism group
is by considering its action on the underlying geometric structure, as for example
in Myers and Steenrod’s work [23] on the isometries of Riemannian manifolds: One
considers the action of the group G of isometries of M on M n = M × · · · × M ; after
establishing a uniqueness result, one can identify the orbit of a point (p1 , . . . , pn )
2010 Mathematics Subject Classification. 32H02,32V40.

The author was supported by the START Prize Y377 of the Austrian federal ministry of
science and research.
c 2011
XXXX
c American Mathematical Society
1
89
90
2 BERNHARD LAMEL
under this action with G. This is then used to transport the manifold structure on
M n to G.
Another example of this approach, which is somewhat closer to the subject mat-
ter of this survey, can be found in the work of Henri Cartan on the automorphism
group of a bounded domain [12]; his classical uniqueness result (if a holomorphic
automorphism of a bounded domain in Cn is tangent to the identity to first order
at some interior point of the domain, then it is the identity) plays an eminent role
in establishing the Lie group structure on these automorphism groups.
Getting back to the Riemannian setting, a comparison with Riemannian man-
ifolds also serves to point out the main differences in the CR case: Even for CR
manifolds with a very nice structure (Levi-nondegenerate ones), the automorphisms
are not comparable by linear data (which, for the Riemannian case, is obviously
true); and automorphisms of CR manifolds might be defined on arbitrarily small
open subsets (which does also not happen in the Riemannian case). Both problems
mean that one cannot apply the traditional methods to the CR case.
This survey concentrates on a particular method to construct a Lie group struc-
ture on the automorphism group of a CR manifold, namely by jet parametrizations.
Here, we use the embedding of the automorphism group into a jet group of finite
order given by a finite order Taylor expansion of the maps, and try to invert this
embedding on its image.
This also means that we do not try to describe the results known via normal-
ization methods for particular types of CR manifolds, as we think this is outside
the scope of this survey. We also only discuss results for the local automorphism
group, i.e. of automorphisms fixing a point.
This paper is organized as follows: In § 2 we define the objects which we will
discuss in this paper. We also give a complete example of the construction of a
parametrization in a special case in order to illustrate the more technical points
later. Nondegeneracy conditions are introduced in § 3, followed by a discussion of
minimality in § 4. We then discuss the known conditions implying the existence of
a jet parametrization in § 5.
2. Basics: CR manifolds, automorphism groups, and parametrizations

2.1. Real analytic CR manifolds. A real-analytic CR manifold M is a real-
analytic manifold M whose complexified tangent bundle CT M is equipped with a
subbundle V ⊂ CT M which satisfies V ∩ V̄ = {0} and [V, V] ⊂ V (i.e. any the Lie
bracket of any two sections of V is again a section of V). The CR dimension of M
is dimC V; the real dimension of M is thus 2n + d for some d ≥ 0; d is referred to
as the real codimension of M , since it turns out that for all practical purposes, one
can think about M locally as a generic submanifold of CN , where N = n + d, of
real codimension d.
Thus, M is locally given as a submanifold of U ⊂ CN by d real-valued defining
functions ρ1 , . . . , ρd which satisfy ρ1,Z ∧ · · · ∧ ρd,Z = 0 in U . We will refer to
ρ = (ρ1 , . . . , ρd ) as a defining function for M . A germ of an automorphism of M
near p is a real-analytic map h : M ⊃ V → M , defined in a neighbourhood V of p,
such that dh(V) ⊂ V. We furthermore define the group
Aut(M, p) = {h : h germ of an automorphism of M, h(p) = p}.

JET EMBEDDABILITY OF LOCAL AUTOMORPHISM GROUPS 91
3
and refer to it as the local automorphism group of (M, p). It has a natural topology
as a group of local real-analytic diffeomorphisms of M , with respect to which it is
a topological group acting continuously on the germ (M, p); in this topology, a se-
quence hn ∈ Aut(M, p) converges to h ∈ Aut(M, p) if there exists a neighbourhood
U of p on which all hn are defined, and hn → h(n → ∞) uniformly on U .
An important object attached to such an M is its complexification M. This
is the complex manifold of complex dimension 2n + d, defined locally by M =
{(Z, ζ) ∈ U × Ū : ρj (Z, ζ) = 0, j = 1, . . . , d}. M has a natural anti-holomorphic
map ι given by (Z, ζ) → (ζ̄, Z̄), and M is contained in M as the fixed point set
of ι. An automorphism H of M induces the automorphism (Z, ζ) → (H(Z), H̄(ζ))
of M, where we write H̄ for the power series whose coefficients are the complex
conjugates of the coefficients of H.
A generic submanifold M ⊂ CN also gives rise to the family of Segre varieties
Sp , defined locally–say for p ∈ U × Ū –by
(1) Sp = {Z ∈ U : ρ(Z, p̄) = 0},
where ρ is a real-analytic defining function of M in U . Sp is a complex-analytic
submanifold of U of complex dimension N − d. We see that a definition in terms
of the complexification M is given by
Sp = {Z : (Z, p̄) ∈ M}.
We disregard here and in what follows the neighbourhoods in which the Segre
varieties are defined and assure the reader that he can (and should) do the same.
The analogous concept to the CR bundle in the complexification is the (1, 0)-
tangent bundle T (1,0) M and the (0, 1)-tangent bundle and T (0,1) M, given by the
∂
linear combinations of the ∂Z j
and the ∂ζ∂ j , respectively.
It is sometimes important to have coordinates in which the Segre varieties
associated to points on a particular subvariety transverse to the complex tangent
directions are all flat; this is done by means of so-called normal coordinates. Normal
coordinates for M near p are holomorphic coordinates (z, w) ∈ Cn × Cd , where
n = dimCR M , in which p = 0, and M is given by the equation w = Q(z, z̄, w̄) for
some germ of a holomorphic funtion Q : Cn × Cn × Cd → Cd satisfying
Q(z, 0, τ ) = Q(0, χ, τ ) = τ ;
in other words, S(0,w) = {(χ, τ ) : τ = w̄}. We also note that if an equation of the
form w = Q(z, z̄, w̄) defines a real submanifold of CN , then necessarily
Q(z, z̄, Q̄(z̄, z, w)) = w.
2.2. Parametrizations of the local automorphism group. If we consider

M , locally around p ∈ CN , as a generic submanifold of CN , Aut(M, p) can be
considered as a group of biholomorphisms of CN fixing p (every h ∈ Aut(M, p) has
a unique extension to a neighbourhood U of p in CN ). We thus also have a natural
map from Aut(M, p) to the jet group of order k at p Gkp (CN ) given by truncating
the Taylor expansion of such an automorphism at order k. To be more precise, the
Gkp (CN ) is the quotient of the group of all germs of biholomorphisms of CN fixing
p by the equivalence relation which identifies two maps if they agree up to order k
at p. It can be identified with the set of all polynomial biholomorphisms of degree
at most k, equipped with the group operation which composes two such maps and
92
4 BERNHARD LAMEL
then truncates them at order k. In any case, it is easy to see that Gkp (CN ) is a
complex (algebraic) Lie group.
We are going to use the map jpk : Aut(M, p) → Gkp (CN ) and try to invert it in
order to transport the Lie group structure on the image jpk (Aut(M, p)) in Gkp (CN )
back to Aut(M, p). Such an inversion is a parametrization of Aut(M, p); it is
given by a map, defined in a neighbourhood U × V of (0, id) ∈ CN × Gkp (CN ),
complex-analytic in its first and real-analytic in its second variable, such that for
H ∈ Aut(M, p) with jpk H ∈ V ,
Ψ(Z, jpk H) = H(Z).
The observation that such a parametrization can be used to define a Lie group
structure on Aut(M, p) goes back to [2]. In order to carry out the following con-
struction, we note that we can always assume that a parametrization Ψ satisfies
jpk Ψ(Z, Λ) = Λ (this just changes the first few terms in Ψ); here we denote by Λ
coordinates in the jet space Gkp (CN ).
Let us consider a defining function ρ for M as well as a real-analytic parametriza-
tion ξ(t) of M , defined for t = (t1 , . . . , t2n+d ) in a neighbourhood of 0 ∈ R2n+d .
Then a jet Λ ∈ V is in the image of Aut(M, p) in Gkp (CN ) if and only if

ρ Ψ(ξ(t), Λ), Ψ(ξ(t), Λ) = 0
for t near 0. We develop the left hand side of this equation into a power series in t
to see that this is equivalent to

Rα (Λ)tα = 0,
α
where Rα is a real-analytic function defined in V . Thus,
jpk (Aut(M, p)) ∩ V = {Λ ∈ V : Rα (Λ) = 0},
and we recognize it as a real-analytic subvariety of V ; however, jpk (Aut(M, p)) is
also a subgroup, which implies that jpk (Aut(M, p)) is a closed real-analytic subgroup
of Gkp (CN ); clearly, the map jpk is also compatible with the natural topology on
Aut(M, p), hence, the Lie group structure on jpk (Aut(M, p)) induces a Lie group
structure on Aut(M, p) which is compatible with the topology on Aut(M, p).
A special class of parametrizations which we are also interested in are the
rational parametrizations; a parametrization Ψ is rational if we can write
Pα (Λ)
Ψ(Z, Λ) = Zα
α
Q α (Λ)
where P and Q are complex-valued real polynomials in Λ, with Qα (j0k id) = 0. If
we have a rational parametrization of Aut(M, p), then we actually get the structure
of a real-algebraic Lie group on Aut(M, p).
2.3. An illustrative example. Before diving into some technical notions and
details, we would like to give a complete discussion of a (well-known) example: We
shall construct a parametrization for the local automorphism group of
M = {(z, w) ∈ C2 : Im w = |z|2 },
the Lewy hypersurface, at the point p = (0, 0). As a byproduct, we shall also obtain
the well-known explicit formulas for the automorphisms of the Lewy hypersurface
5
in (13) from the jet parametrization technique. We use coordinates (z, w, χ, τ ) on

C4 to describe the complexification M, which is given by
w − τ = 2izχ.
A map H(z, w) = (f (z, w), g(z, w)) is an automorphism of (M, 0) if and only if
(2) g(z, w) − ḡ(χ, τ ) = 2if (z, w)f¯(χ, τ ),
when w − τ = 2izχ, f (0, 0) = g(0, 0) = 0 and if the matrix
∂f ∂f

∂z (0) ∂w (0)
∂g ∂g
∂z (0) ∂w (0)
is invertible. We now evaluate (2) at w = χ = τ = 0 to see that g(z, 0) = 0; thus,
the invertibility condition reduces to fz (0, 0)gw (0, 0) = 0.
Now substitute w = τ + 2izχ into (2) to see that
(3) g(z, τ + 2izχ) = ḡ(χ, τ ) + 2if (z, τ + 2izχ)f¯(χ, τ );
∂
an application of ∂z to (3) leads to
(4) gz + gw 2iχ = 2i(fz + fw 2iχ)f,
where we have dropped the arguments for better readability. Evaluation of (4) at
z = w = τ = 0 gives
(5) f¯(χ, 0)(fz (0) + 2iχfw (0)) = gw (0)χ.
∂
If we apply ∂τ to (4) and evaluate at again at z = w = τ = 0, we get a similar
¯
formula for fτ (χ, 0), namely
(6)
2if¯τ (χ, 0)(fz (0) + 2iχfw (0)) = gzw (0) + gw2 (0)2iχ − 2if¯(χ, 0)(fzw (0) + 2iχfw2 (0)),
∂
and applying ∂τ to (3) also a formula for ḡτ (χ, 0),
(7) ḡτ (χ, 0) = gw (0) − 2ifw (0)f¯(χ, 0).
Now we substitute set w = 0, τ = −2izχ into (2) to obtain
(8) −ḡ(χ, −2izχ) = 2if (z, 0)f¯(χ, −2izχ),
and into (4), which leads to
(9) gw (z, 0)χ = (fz (z, 0) + fw (z, 0)2iχ)f¯(χ, −2izχ).
Substituting back the complex conjugates of (5) and (7) this gives with D = f¯χ (0)−
2iz f¯τ (0)
gw (z, 0)χ
f¯(χ, −2izχ) =
fz (z, 0) + 2iχfw (z, 0)

ḡτ (0)z
χ ḡτ (0) + 2if¯τ (0)
D
= ¯
fχ (0)ḡτ (0) −ḡχτ (0) + 2izḡτ 2 (0) ḡτ (0)z(f¯χτ (0) − 2iz f¯τ 2 (0))
+ 2iχ −
D2 2iD D2
χḡτ (0)D2 + 2izχf¯τ (0)ḡτ (0)D
= ¯
fχ (0)ḡτ (0) + χD(−ḡχτ (0) + 2izḡτ 2 (0)) − 2izχḡτ (0)(f¯χτ (0) − 2iz f¯τ 2 (0))
χf¯χ (0)2 ḡτ (0) − 2izχf¯χ (0)ḡτ (0)f¯τ (0)
= ¯ .
fχ (0)ḡτ (0) − χf¯χ (0)ḡχτ (0) − 2izχ(f¯χτ (0)ḡτ (0) − f¯χ (0)ḡτ 2 (0)) + z 2 χR
94
6 BERNHARD LAMEL
Since the left hand side of this equation is a holomorphic function in χ and −2izχ,
R has to vanish. So all in all, we arrive at
fz (0)z + fw (0)w
(10) f (z, w) = .
gzw (0) gw2 (0)
1− gw (0) z + ffzw
z
(0)
(0) − g w (0) w
Using g(z, 2izχ) = 2if (z, 2izχ)f¯(χ, 0) and (5) we get a similar equation for g,
namely
gw (0)w
(11) g(z, w) =
gzw (0) gw2 (0)
1− gw (0) z + ffzw (0)
z (0)
− gw (0) w
The right hand sides of (10) and (11) are parametrizations for f and g, i.e. H =
Ψ(Z, j02 H). Of course, we would have to arrange things such that j02 Ψ(Z, Λ) = Λ;
but computationally, it is preferable to use the additional information present in
(10) and (11) in order to compute j02 (Aut(M, 0)): We thus check which map of the
form

αz + βw w
(12) (z, w) → ,
1 + γz + δw 1 + γz + δw
gives rise to an automorphism of M . Writing D(z, w) = 1 + γz + δw, this amounts

to finding all α, β, γ, δ, such that
(τ + 2izχ)D̄(χ, τ ) − ¯τ D(z, τ + 2izχ) = (αz + β(τ + 2izχ))(ᾱχ + β̄τ ).
Comparing coefficients on both sides of this equation, we get
− ¯ = 0 −2iβ β̄ + δ̄ − δ¯
=0
− αᾱ = 0 2β ᾱ + iγ̄ = 0.
This describes the image of Aut(M, p) in the closed subgroup of G20 (C2 ) defined by
mappings of the form (12). An explicit parametrization of this image is given by
(13) α = reiθ , β = reiθ a, γ = −2iā, δ = t − i|a|2 , = r2 ,
where r ∈ R \ {0}, θ, t ∈ R, a ∈ C \ {0}, with which choices the group Aut(M, 0)

can be written in the usual form as the mappings of the form

z + aw r2 w
(z, w) → re iθ
, .
1 − 2iāz + (t − i|a|2 )w 1 − 2iāz + (t − i|a|2 )w
Let us retrace our steps: We first determined H̄(χ, 0) and H̄τ (χ, 0) in (5), (6),
and (7) in terms of j02 H; this was done using a nondegeneracy condition on (M, 0),
namely, that we could solve for f¯(χ, 0) in (4) evaluated at z = w = τ = 0; then
we leveraged this knowledge to get a formula for f¯(χ, −2izχ) in terms of j02 H̄,
again using the fact that we could solve for it in (4) evaluated at w = 0 in terms of
2
j(z,0) H. This procedure led to a formula for H since the map (z, χ) → (χ, −2izχ) is
generically of full rank; we will see later that this is a consequence of the minimality
of M at 0.
7
3. Nondegeneracy conditions
3.1. From finite nondegeneracy to holomorphic nondegeneracy. Non-
degeneracy conditions tell us how “curved” a CR manifold is. The nondegeneracy
conditions we discuss here have mostly geometric significance (for example, we
do not distinguish between Levi-nondegeneracy and strict pseudoconvexity here).
They measure how the Segre varieties differ from point to point. Our main tool in
the presentation here is the following map defined on the complexification M of a
real-analytic generic submanifold M ⊂ CN of codimension d:
(14) πM : M → Hd (CN ), (Z, ζ) → (Sζ̄ , Z),
N
where Hd (C ) is the bundle of germs of complex-analytic submanifolds of codi-
mension d in CN . We can now define:
Definition 1. Let (M, p) be a germ of a generic, real-analytic submanifold of
CN . Then:
(1) (M, p) is finitely nondegenerate at p if Sp̄ ζ → πM (p, ζ) is an immersion
at ζ = p̄;
(2) (M, p) is essentially finite at p if Sp̄ ζ → πM (p, ζ) is a finite map at
ζ = p̄;
(3) (M, p) is of class C at p if Sp̄ ζ → πM (p, ζ) is generically of full rank;
(4) (M, p) is holomorphically nondegenerate if SZ̄ ζ → πM (Z, ζ) is generi-
cally of full rank for generic Z ∈ M close by p (which happens if and only
if πM is generically of full rank).
Remark 1. We remark that while the definitions given above are very con-
venient to work with, they are neither the original nor the standard definitions
(except for class C, introduced in [21]). The notion of holomorphic nondegeneracy
has been introduced by Stanton [25], and is usually defined by saying that M is
holomorphically nondegenerate if there is no nontrivial germ of a vector field
∂
X= aj (Z)
j
∂Zj
which is tangent to M . Essential finiteness has been introduced by Baouendi,
Jacobowitz, and Treves [7] in their study of analyticity of smooth CR maps; an
equivalent definition is that (M, p) is essentially finite if ∩q∈Sp Sq = {p}. Finite
nondegeneracy has been introduced by Baouendi, Huang, and Rothschild [6]; also
here, the original definition is different than the one given here.
Directly from the definition we see that if M ⊂ CN is a connected, real-analytic,
generic submanifold, then either (M, p) is holomorphically nondegenerate for all p ∈
M or for none; in the first case, we say that M is holomorphically nondegenerate,
in the second case, that M is holomorphically degenerate.
The nondegeneracy conditions in Definition 1 are increasingly weaker; i.e. a
finitely nondegenerate submanifold is essentially finite, an essentially finite one is
of class C, one of class C is holomorphically nondegenerate; there are examples that
all of these inclusions are strict, see Example 1. Furthermore, if M ⊂ CN is a
connected, real-analytic generic submanifold which is holomorphically nondegener-
ate, then there exists a nontrivial real-analytic subvariety V1 ⊂ M such that for all
p ∈ M \ V1 , (M, p) is of class C; there exists a real-analytic subvariety V2 ⊂ M ,
V2 ⊃ V1 , such that for p ∈ M \ V2 , (M, p) is essentially finite; and, there exists a
96
8 BERNHARD LAMEL
real-analytic subvariety V3 ⊂ M , V3 ⊃ V2 , such that for p ∈ M \V3 , (M, p) is finitely

nondegenerate. Actually, all of the Vj are locally of the form Wj ∩ M , where Wj
are complex-analytic subvarieties of CN . Proofs of these statements can be found
in e.g. in [21].
Example 1. The manifold in C2 given by
Im w = Re w|z|2
is holomorphically nondegenerate but not of class C. The manifold in C3 given by
Im w = |z1 z2 |2 + |z1 |2
is of class C, but not essentially finite.
The manifold in C2
Im w = |z|4
is essentially finite, but not finitely nondegenerate.
In order to connect Definition 1 with the original definitions, we shall also
summarize a characterization of the nondegeneracy conditions in coordinates.
Lemma 2. Let M ⊂ CN be a generic, real-analytic submanifold, p ∈ M , and
choose normal coordinates
(z, w) centered at p so that M is given by w = Q(z, χ, τ );
write Q(z, χ, τ ) = α Qα (χ, τ )z α . Then:
(1) (M, p) is finitely nondegenerate if and only if χ → (Qα (χ, 0))α is immer-
sive at χ = 0;
(2) (M, p) is essentially finite if and only if χ → (Qα (χ, 0))α is finite at χ = 0;
(3) (M, p) is of class C if and only if χ → (Qα (χ, 0))α is generically of full
rank;
(4) (M, p) is holomorphically nondegenerate if and only if ζ → (Qα (ζ))α is
generically of full rank, which is the case if and only if for generic τ close
by 0, the map χ → (Qα (χ, τ )) is generically of full rank.
The proof consists of working out the map πM in normal coordinates and is
left to the reader.
The equivalence of the original definition of holomorphic nondegeneracy with
the one given above in Definition 1 can now be seen in the following way: Let
us choose normal coordinates (z, w) for (M, p), so that M is locally given by the
equation

w = Q(z, z̄, w̄) = Qα (ζ)z α .
α
If X is a holomorphic tangent vector field as above, this means that X̄Qα ≡ 0 for
all α, i.e. there exist germs of holomorphic functions a1 (Z), . . . aN (Z) such that for
all α ∈ Nn ,
N
∂Qα
āj (ζ) (ζ) = 0;
j=1
∂ζj
thus we see that the generic rank of ζ → (Qα (ζ))α is full if and only if no nontrivial
holomorphic tangent vector field exists.
It is often important to get some more detailled control over numerical invari-
ants associated to the nondegeneracy conditions. We discuss these next.
9
3.2. Numerical invariants for nondegenerate manifolds. It is often im-

portant to get some information about the order of the terms in the defining equa-
tion which give rise to (any kind of ) nondegeneracy. In order to grasp this level,
we need to introduce the map
k
πM : M → Hdk (CN ), (Z, ζ) → jZk Sζ̄ ,
valued in the bundle of k-jets of complex-analytic submanifolds of codimension d
in CN .
If we look at a holomorphically nondegenerate manifold M , we know by def-
k
inition that for some k, πM is generically of full rank. However, we also need to
know some basic information about possible singularities of this map at a given
point (p, p̄). For this purpose, we will consider a numerical invariant for a holo-
morphic map which roughly measures how bad the degeneration of πM is, both in
transversal and tangential directions.
In order to do so, we will need a notion of an order which respects the splitting
in transversal and tangential directions. We start by defining this.
Definition 3. Let W be a well-ordered set, and ω : Nn → W a map. Then
the order ordω with respect to ω of a power series

ϕ(z) = ϕα z α
α∈Nn
is
ordω ϕ = min{ω(α) : ϕα = 0} ∈ W.
If ϕ ≡ 0, we set ordω ϕ = ∞.
In what follows, we will use W = N2 with the lexicographic ordering (a, b) ≺
(c, d) if b < d or, if b = d, if a < c. For a function ϕ(Z, ζ) on M, we use the
parametrization Z = (z, Q(z, χ, τ )), ζ = (χ, τ ) by some set of normal coordinates
to write ϕ(Z, ζ) = ϕ̃(z, χ, τ ), and define ω : N2n+d → N2 , (α, β, γ) → (|α| + |β|, |γ|);
we set ordω ϕ = ordω ϕ̃. The reader can check that this definition is independent
of the choice of normal coordinates.
If we are dealing with maps, we will use the order of vanishing of their Jacobians
to measure their transversal/tangential degenerations.
Definition 4. Let H : (Cn , p) → Cm be a germ of a holomorphic map, and W ,
ω as in Definition 3. We say that the determinantal order of H at p is and write
= doωp H if is the minimum order with respect to ω of the minors of the Jacobian
of H, seen as a matrix-valued holomorphic map. Similarly, for any matrix-valued
map M we define the determinantal order of M at p as the minimum of the orders
of the minors of M at p.
Given a germ (M, p) of a generic real-analytic submanifold of CN , we can now
define a sequence of lexicographically descending elements of N2 by defining
(15) n(k) = doω k
(p,p̄) πM , k ∈ N.
In terms of the n(k), we have the following characterization of the different nonde-
generacy conditions introduced earlier.
Lemma 5. Let (M, p) be a germ of a generic real-analytic submanifold of CN .
Then:
(1) (M, p) is holomorphically nondegenerate if n(k) = ∞ for some k ∈ N;
98
10 BERNHARD LAMEL
(2) (M, p) is in class C if n(k) = (n1 (k), 0) for some k ∈ N;

(3) (M, p) is finitely nondegenerate if n(k) = (0, 0) for some k ∈ N.
Remark 2. We can use the sequence n(k) to recover several numerical invari-
ants used in the literature: E.g., if (M, p) is finitely nondegenerate, then we say that
(M, p) is -nondegenerate if k() = (0, 0), but (0, 0) ≺ k(j) for j < . Also, note
that (M, p) is Levi nondegenerate if and only if n(1) = (0, 0). However, essential
finiteness and its associated numerical invariants cannot be recovered from n(k).
4. Minimality
The notion of minimality for real-analytic submanifolds in CN has its roots in
the work of Kohn [19] and Bloom-Graham [8]. In the real-analytic case, minimality
(which was introduced by Tumanov [28]) and finite type agree. Here, we will
mostly concern us with a particular method to recover the analog to CR orbits in
the complexified setting, namely through the so-called “Segre-maps”. The main
result in this setting is yet another characterization of minimality due to Baouendi,
Ebenfelt, and Rothschild [1] through these maps.
Let us recall these notions of finite type and minimality before we start. (M, p)
is of finite type if the Lie algebra g generated by the sections of V and the sections
of V̄ spans the whole complexified tangent space at p, i.e. g(p) = CTp M . (M, p) is
minimal if there is no CR manifold N M through p with dimCR N = dimCR M .
As already noted, in the case of a real-analytic submanifold these notions coincide,
see e.g. [5].
Another point of view yet is by considering the CR orbits of points in M . The
CR orbit of Oq of a point q ∈ M is a germ of a real-analytic submanifold Oq ⊂ M
which is obtained as the set of points which can be connected to q by a piecewise
smooth path γ with γ (t) ∈ Tγ(t)c
M ; with this terminology, (M, p) is of finite type
if Op = M in the sense of germs at p.
Using the Segre varieties Sq , we can define a related increasing series of sets
Spj ⊂ CN , the Segre sets, by

Sp1 = Sp , Spj = Sq , j > 1.
q∈Spj−1
The minimality criterion in [1] can then be stated by saying that (M, p) is minimal
if and only if there exists a j such that Spj contains an open set in CN .
Similar constructions can be carried out in M: If we consider all the points
which can be joined with (p, p̄) ∈ M by curves tangent to T (1,0) M, we obtain
Sp × {p̄}. Now, if we continue this process and look at the set of points which can
be joined to (p, p̄) by a segment tangent to T (1,0) M followed by a segment tangent
to T (0,1) M, we obtain in the next step the set (q, r), where q ∈ Sp and r ∈ Sq̄ .
This process can be continued inductively, and will at some point stabilize, the
projection of the set thus obtained to either the first or second component being
Sp2j+1 or Sp2j , respectively.
It is actually very useful to describe the sets obtained here as the images of
certain holomorphic maps, the Segre maps, which we define as follows. We fix
normal coordinates (z, w), so that M is given by w = Q(z, χ, τ ), or equivalently,
by τ = Q̄(χ, z, w).
11
For j ≥ 1, we define the Segre mappings S j : Cnj × Cd → Cn+d iteratively for

t ∈ Cd and xk ∈ Cn for all k as follows:
S 1 (x1 ; t) = (x1 , t),
(16)
S j+1 (x1 , . . . , xj+1 ; t) = x1 , Q(x1 , S̄ j (x2 , . . . , xj+1 ; t) , j ≥ 1.

We also agree to write S 0 (t) = (0, t), which makes the second line of (16) valid for
all j ≥ 0. Note that (S j , S̄ j−1 ) and (S j−1 , S̄ j ) map Cnj × Cd into M for all j ≥ 1.
We denote by U j : Cnj × Cd → Cd the second component of S j , i.e.,

S j (x1 , . . . , xj ; t) = x1 , U j (x1 , . . . , xj ; t) .
It will be useful to write
(17) x[j;k] = (xj , . . . , xk ), j ≤ k.
With this notation, the definition of the Segre maps (16) reads
(18) S 1 (x1 ; t) = (x1 , t); S j+1 (x[1;j+1] ; t) = (x1 , Q(x1 , S̄ j (x[2;j+1] ; t))).
This last equation ties the maps to the sequence of Segre sets S0j , since it shows
that S0j = image S j .
We are now ready to state the result from [1] already alluded to above.
Theorem 1. Let (M, 0) be a generic, real-analytic submanifold of CN , and
denote by S j the Segre mappings as defined above. Then (M, 0) is minimal if and
only if there exists a j such that the map
x[1;j] → S j (x[1;j] ; 0)
is generically of full rank.
5. Results for the local automorphism group

5.1. Protohistory: Poincaré. In his celebrated paper [24], Poincaré was
the first to observe the strong rigidity properties of real-analytic hypersurfaces in
CN , N ≥ 2; his observation also serves to connect the equivalence problem (when
does there exist a biholomorphism carrying a given real-analytic hypersurface into
another one) with the study of the automorphism group.
Here is Poincaré’s observation: Consider two germs of real-analytic hypersur-
faces in M, M ⊂ CN through 0, and let M be given by ρ(z, z̄) = 0, M given
by ρ (z, z̄) = 0. If H maps M into M , there exists a (real-analytic) function
A(z, z̄), A(0, 0) = 0, such that A(z, z̄)ρ(z, z̄) = ρ (H(z), H(z)); in other words,
ρ (H(z), H(z)) is a defining function for M . Now think of ρ as fixed. Consider the
terms in the power series

A(z, z̄)ρ(z, z̄) = ρ (H(z), H(z)) = cα,β z α z̄ β
α,β
of order k. Expanding this equation in terms which are homogeneous of degree d

in z and z̄, we can think of it as prescribing a transformation between the ρα,β and
the ρα,β where |α| + |β| = d, for each choice of Hα (where H(Z) = α Hα Z α ), and
each choice of Aα,β :

ρα,β = ϕα,β ρδ,γ , Hδ , Aδ ,γ , |δ| + |γ| ≤ d, |δ | + |γ | ≤ d − 1.

Now think of ρ as fixed, and consider ϕd = (ϕα,β )|α|+|β|=d as a map taking Hα
2n+d−1
and A , |α| = d, |δ| + |γ| = d − 1, into C( 2n−1 ) . The subspace which corresponds
δ,γ
100
12 BERNHARD LAMEL
to real valued homogeneous polynomials of degree d has (real) dimension 2n+d−1 2n−1 .
Now we count the number of independent

real variables.
From the Hα , we get 2 n+d−1 real variables; from the Aγ,δ (which also satisfy
n−1 2n+d−2
the reality condition Aγ,δ = Aδ,γ ), 2n−1 . All in all, that adds up to

n+d−1 2n + d − 2 ∼ dn−1 (d − 1)2n−1 d2n−1 ∼ 2n + d − 1
2 + =2 + < = ,
n−1 2n − 1 n − 1! (2n − 1)! (2n − 1)! 2n − 1

if n > 1. We conclude that for generic ρ, the equation ρα,β = ϕα,β ρδ,γ , Hδ , Aδ ,γ
does not have any solution in Hδ and Aδ ,γ .
The conclusion is that for a given (M , p ), the space of (M, p) such that (M, p) is
biholomorphically equivalent to (M , p ) is quite small. It also follows that there are
at least countably many independent invariants associated to this local equivalence
problem. In terms of the local automorphism group, one expects it to be rather
small for generic (M, p); indeed, the nondegeneracy conditions introduced earlier
in §3 can be thought of as conditions which ensure that the system of equations
encountered in Poincaré’s argument is nonsingular.
5.2. History: Cartan, Tanaka, and Chern-Moser. In the wake of Poincaré’s

paper, there was a considerable effort to understand the equivalence problem for
Levi-nondegenerate hypersurfaces. This problem was first taken on and solved by
Elie Cartan (using his method of equivalence) in the case of Levi-nondegenerate
hypersurfaces in C2 [11, 10]; later on, Tanaka [27] and Chern-Moser [13] were able
to solve the problem for Levi-nondegenerate hypersurfaces in CN .
In all of the approaches to this problem, the notion of a model hypersurface,
which for a Levi-nondegenerate hypersurface of signature is just the hyperquadric

n
Im w = |zj |2 − |zj |2 ,
j=1 j=+1
is essential; in fact, one compares the geometries of a Levi-nondegenerate hyper-

surface in some sense by seeing how much they differ from this (in the sense of the
pseudo-conformal geometry encountered here, flat) model hypersurface.
A particular outcome is that the local automorphism group of a Levi-nondegenerate
hypersurface (M, p) can be considered as a subgroup of the automorphism group
of its model; these are just the groups P SL( + 1, n − + 1), in coordinates (z, w)
as above given by

z + aw δ2w
(z, w) → δU , ,
1 − 2iz, a − i(a, a + it)w 1 − 2iz, a − i(a, a + it)w
where U ∈ SU (, n − ), a ∈ Cn , t ∈ R, δ ∈ R \ {0}, and

n
z, a = zj āj − zj āj
j=1 j=+1
is the standard hermitian inner product of signature on Cn .

In particular, for any Levi-nondegenerate hypersurface (M, p), Aut(M, p) turned
out to be a finite dimensional Lie group.
13
5.3. Interlude: Infinitesimal automorphisms. Before the local automor-

phism group was studied in earnest for anything besides of Levi-nondegenerate
hypersurfaces, there were a couple of results for the Lie algebra of infinitesimal au-
tomorphisms (the Lie algebra which would be the Lie algebra of the automorphism
group, if this was a Lie group).
The Lie algebra in question is
⎧ ⎫
⎨ ∂ ⎬
hol(M, p) = X = aj (Z) : Re X tangent to M near p .
⎩ ∂Zj ⎭
j
This is a Lie algebra over R; it contains as a subalgebra the algebra of tangent

holomorphic vector fields
⎧ ⎫
⎨ ∂ ⎬
ht(M, p) = hol(M, p)∩hol(M, p) = X = aj (Z) : X tangent to M near p .
⎩ ∂Zj ⎭
j
Now, ht(M, p) is also a module over C{Z − p}, and if it is not trivial, its dimension
over R is therefore necessarily infinite; by definition, ht(M, p) is trivial if and only
if (M, p) is holomorphically nondegenerate.
Stanton characterized the finite dimensionality of hol(M, p) for real-analytic hy-
persurfaces by holomorphic nondegeneracy in [26]. The question for real-analytic
submanifolds of higer codimension is more subtle, as one needs to assume in addi-
tion to holomorphic nondegeneracy that (M, p) is not everywhere nonminimal, as
the example of the codimension 2 submanifold of C3 given by the product of a hy-
perquadric in C2 and the real line shows. The defining equations of this submanifold
is
Im w1 = |z|2 , Re w2 = 0,
∂
and hol(M, p) contains every vector field of the form h(w2 ) ∂w 2
with h = h̄, which
shows that hol(M, p) cannot be finite dimensional.
Under these assumptions, i.e. that (M, p) is not everywhere nonminimal and
holomorphically nondegenerate, Baouendi, Ebenfelt and Rothschild proved that
hol(M, p) is finite dimensional in [3]. In view of the remarks above, this actually
gives a complete solution to this problem: hol(M, p) is finite dimensional if and
only if (M, p) is holomorphically nondegenerate.
What do these results tell us for the automorphism group Aut(M, p)? First off,
there is a close relation between Aut(M, p) and the subalgebras
hol0 (M, p) = {X ∈ hol(M, p) : X(p) = 0}, ht0 (M, p) = {X ∈ ht(M, p) : X(p) = 0}.
Indeed, if ht(M, p) is nontrivial, so is ht0 (M, p), and it turns out that Aut(M, p)
cannot be finite dimensional (an observation which goes back to [3]): Consider
X ∈ ht0 (M, p), and its flow ΦX t , which for small t ∈ C is in Aut(M, p). Replacing
t by a holomorphic function h(Z) of Z, we obtain a family of automorphisms
Z → ΦXh(Z) (Z) which is clearly not finite dimensional (if ord h = k, then Φh(Z)
X
agrees with the identity up to order k, but is not the identity). Thus we see that
holomorphic nondegeneracy is also necessary for Aut(M, p) to be finite dimensional.
However, that does not mean that it is also sufficient; indeed, one cannot
conclude from dim hol(M, p) < ∞ that Aut(M, p) is a Lie group. What we can
conclude is that there is a unique Lie group structure on Aut(M, p), but we do
not have any guarantee that this structure relates to the natural topology on the
102
14 BERNHARD LAMEL
topological group Aut(M, p). Indeed, in some sense, dim hol(M, p) < ∞ is a linear
problem, while the existence of a Lie group structure on Aut(M, p) is a highly
nonlinear one.
5.4. First steps: Finitely nondegenerate manifolds. The first class of

real-analytic submanifolds of CN besides of Levi-nondegenerate ones for which a
Lie group structure on Aut(M, p) was constructed is the class of finitely nondegen-
erate hypersurfaces. This was done by Baouendi, Ebenfelt, and Rothschild in [2].
Subsequently, Zaitsev [29] was able to prove this result for general finitely nonde-
generate, minimal real-analytic manifolds (M, p). Actually, the authors in these
papers prove a somewhat more general statement: Given germs of submanifolds
(M, p) and (M , p ) let us write
H(M, p; M , p ) = {H : (M, p) → (M , p ) : H biholomorphism},
and let us write Gkp,p for the set of all k-jets of germs of biholomorphisms H : (CN , p) →
(CN , p ). We then have the following main result in [2]:
Theorem 2. Let M and M be two real analytic hypersurfaces in CN which

are k0 -nondegenerate at p and p respectively and let H = H(M, p; M , p ) as above.
Then the restriction of the map η = jp2k0 : H(CN , p; CN , p ) → G2k p,p to H is one-
0
to-one; in addition, η(H) is a totally real, closed, real analytic submanifold of G2k 0
p,p
(possibly empty) and η is a homeomorphism of H onto η(H). Furthermore, global
defining equations for the submanifold η(H) can be explicitly constructed from local
defining equations for M and M near p and p .
Zaitsev had formulated his result using the notion of analytic dependency on
k-jets: If S ⊂ H(CN , p; CN ) is a set of germs of biholomorphisms, then the germs
in S depend analytically on their k-jets if
(1) Every f ∈ S is uniquely determined by jpk f ;
(2) For every jpk f0 where f0 ∈ S there exists a neighbourhood U (jpk f0 ), a neigh-
bourhood U (p) of p, and a holomorphic function F : U (jpk f0 ) × U (p) → CN
such that for all f ∈ S with jpk f ∈ U (jpk f0 ) one has f (z) = F (jpk f, z).

Given two real analytic submanifolds M ⊂ CN and M ⊂ CN , Zaitsev goes on to
call a germ f admissible if
f (M ) ⊂ M , df (Tpc M ) = Tfc(p) ;
i.e. f is a CR submersive map from M to M . We can now quote the main theorem
from [29]:
Theorem 3. Suppose that M is r-nondegenerate, M is minimal at p and

k = 2r(1 + codimCR M ). Let S be the set of all admissible germs f at p. Then the
germs in S depend analytically on their k-jets.
We note that in Zaitsevs’ result, the image point f (p) need not be fixed; we
will point out where finite nondegeneracy is used for this below.
Roughly, the proofs are obtained by the following scheme. Consider the com-
plexifications M and M . If H ∈ H(M, p; M , p ), then the following diagram
commutes:
15
(Z, ζ) → (H(Z), H̄(ζ))

M M
k k
πM πM
k
jZ H
Hdk (CN ) Hdk (CN )
This means that for (Z, ζ) ∈ M, we can say something about H̄(ζ) by knowing
something about jZk H, if we can invert the arrow on the right-hand side of this
diagram:
(Z, ζ) → (H(Z), H̄(ζ))
M M
k k k −1
πM πM (πM )
k
jZ H
Hdk (CN ) Hdk (CN )
If (M , p ) is finitely nondegenerate, this inversion is obtained by an application

of the implicit function theorem (since the implicit function theorem allows for
arbitrary parameter dependence to be passed through to the solution, one can vary
the image point as well, which is why in Zaitsevs’ result the image point need not
be fixed). It turns out one can also get prolongated versions of this, i.e. we can find
formulas of the form jζ+k H̄ = Ψ(ζ, jZ H), for (Z, ζ) ∈ M. We are thus led to an
iterative procedure: If we know j0k+ H, we can write down a formula for jζk H̄ for
(0, ζ) ∈ M, i.e. ζ ∈ S0 ; knowing jζk+ H̄ for ζ ∈ S0 leads us by the same argument
to knowing jZk H for Z ∈ S02 , and so on. Since we are interested in analytic formulas
for H(Z), we interpret this as “parametrizations along the Segre maps”, i.e. we
obtain for every k an (k) and a real-analytic map Ψk , holomorphic in its x-variable,
such that
(k)
H ◦ S k (x[1;k] , 0) = Ψk (x[1;k] , j0 H).
The end of the proof is effected by passing from these parametrizations along the
Segre varieties to parametrizations which depend on Z; here is where the minimality
of (M, p) comes into play.
It was realized later in a more general setting (CR submersive maps of a minimal
real-analytic submanifold to a finitely nondegenerate one) by Baouendi, Ebenfelt
and Rothschild [4] that one can actually obtain a rational parametrization. In that
paper they also introduced a technique which allowed for a suitable right inverse to
Segre mappings of high enough order (the “doubling trick”); this gives a possibly
singular right inverse, but whose singularity is well described, and thus can be used
to reconstruct holomorphic functions along the image of S k .
5.5. Minimal hypersurfaces in C2 . The first parametrization result valid

for finitely degenerate real-analytic hypersurfaces in C2 was obtained by Ebenfelt,
Zaitsev, and the author [15]. In that paper, we showed that a parametrization by
2-jets is possible for every minimal real-analytic hypersurface in C2 . Now, a real-
analytic hypersurface in C2 is minimal if and only if it is essentially finite; and really,
both of these properties had to be used in order to construct a parametrization.
The statement of the main parametrization theorem from [15] is as follows:
104
16 BERNHARD LAMEL
Theorem 4. Let M ⊂ C2 be a real-analytic hypersurface of finite type at a

point p ∈ M . Then there exist an open subset Ω ⊂ C2 × G2p (C2 ) and a real-analytic
map Ψ(Z, Λ) : Ω → C2 , which in addition is holomorphic in Z, such that the
following holds. For every local biholomorphism H of C2 sending (M, p) into itself,
the point (p, jp2 H) belongs to Ω and the identity
H(Z) ≡ Ψ(Z, jp2 H)
holds for all Z ∈ C2 near p.

The approach taken in [15] is somewhat different to the one described before;
in order to determine H along S0 , we used a series which transforms like a tensor
k
under biholomorphisms; in some sense, this series is the “lowest-order” term in πM
(i.e. in a particular ordering on the variables).
k
The prolongation to the second Segre set was possible since πM is, in this
setting, actually finite; an argument similar to the one in the finitely nondegenerate
setting can then be used to find a formula for a mapping H along the second Segre
map.
However, the techniques used for the proof of this result used heavily that one
was in C2 , and therefore, that the Segre varieties are one-dimensional. Finite maps
in C are easily inverted using standard methods. How one should invert a finite map
in CN remained one of the obstacles to overcome in order to find a parametrization
for automorphism groups of real-analytic submanifolds of CN .
5.6. Sufficient results in class C. This question was tackled by Mir and
the author in [21], by means of a theorem allowing for the inversion of systems
of equations of generically full rank. Let H(Cp , 0; Cq , 0) be the space of germs
of holomorphic maps (Cp , 0) → (Cq , 0). For p = q, we simply write H(Cp , 0),
and denote by B(Cp , 0) the subset of germs of biholomorphisms. These spaces
are topological vector spaces (inductive limits of Frechét spaces), and the essential
result we used is most easily stated in the following way.
Theorem 5. Let A : (Cn , 0) → (Cn , 0) be a germ of a holomorphic map which
is generically of full rank. Then there exists a holomorphic map Ψ : GLn (C) ×
H(Cn , 0) → B(Cn , 0) such that
u = Ψ(u (0), A ◦ u), u ∈ B(Cn , 0).
That is, the map Ψ inverts the map A holomorphically (depending analytically
on the parameters which make the inversion unique). Theorem 5 is a highly non-
trivial result; it turned out that it is connected via the infinite dimensional rank
theorem of Hauser and Müller [16] with several results in singularity theory, as
explained in Bruschek and Hauser [9]. The main parametrization theorem in [21]
can now be stated as follows, where κM (p) denotes the smallest integer k for which
πMk
, restricted to {p} × Sp , is generically of full rank:
Theorem 6. Let M be a real-analytic generic submanifold of CN of codimen-

sion d. Let p ∈ M and assume that (M, p) is minimal and belongs to the class C
p
and set p := (d + 1)κM (p). Then there exists an open subset Ω ⊂ CN × Gp,p (CN )
and a real-analytic map Ψ(Z, Λ) : Ω → C holomorphic in the first factor, such
N
that the following hold:

17

i) for any H ∈ Aut(M, p), the point (p, jpp H) belongs to Ω and the following iden-
tity holds:
H(Z) = Ψ(Z, jpp H) for all Z ∈ CN near p;
i) the map Ψ has the following formal Taylor expansion
Pα (Λ, Λ̄)
Ψ(Z, Λ) = (Z − p)α ,
r s
α (D(Λ1 ) )(D(Λ1 ))
α α
where for every α ∈ NN , rα and sα are nonnegative integers, Pα and D are

polynomials in their arguments, Λ1 denotes the linear part of the jet Λ and
D(jp1 H) = 0 for every H ∈ Aut(M, p).
Theorem 6 thus provides a globally defined, rational parametrization for Aut(M, p).
In its proof, Theorem 5 was used as a substitute for the implicit function theorem
in order to invert the map πM k
for M in class C. Similarly to the finitely nonde-
generate case, it is possible to prolong this in order to get formulae for the jets
of a map H. One then again obtains parametrizations along the Segre maps, and
inverts the Segre maps using the doubling trick as in [4], in order to arrive at a jet
parametrization for Aut(M, p), where (M, p) is minimal and of class C.
The parametrization thus constructed is actually also rational, and hence gives
a real-algebraic structure on Aut(M, p). This result covers a wide variety of cases
which are of importance: e.g. real-analytic boundaries of bounded domains, or more
generally, real-analytic hypersurfaces which do not contain any complex-analytic
subvarieties.
5.7. Characterization in the minimal case. The inversion theorem 5 used

in the class C case cannot be used if πM k
degenerates along {Z = 0} ∩ M. In the
general holomorphically nondegenerate setting, this will actually happen; we can
k
only guarantee that πM is generically of full rank, and it might well not be of full
rank along Z = 0.
The solution to this problem was obtained by Juhlin and the author [20]. There
are two main ingredients for this: First, we obtained an inversion theorem which
can be used in this setting. Let me state this theorem first. We use coordinates
(x, t) ∈ Cp × Cq and define for (α, β) ∈ Np × Nq ω(α, β) = (|α|, |β|) ∈ N2 in order
to define the order of a power series in x and t with respect to ω as in Definition 3
(x is going to play the role of a “tangential” and t of a “transversal” variable); N2
is ordered lexicographically as before.
k
We also need to define the notion of a transversal jet; that is, we write jt,0 h for
the collection of all derivatives
∂ |α| h
(x, 0), |α| ≤ k.
∂tα
Theorem 7. Let A ∈ H(CN , 0) be a map which is generically of full rank. If

h (x, t) is a holomorphic map valued in CN such that doω
0
(0,0) A ◦ h = n < ∞,
0
k
then there exists a sequence of integers (k), neighbourhoods of Uk of jt,0 A ◦ h0 , a
(0)
neighbourhood V of j0 h and a holomorphic map Ψ such that for all k ∈ N,
(0) (k)
k
jt,0 h = Ψ(j0 h, jt,0 A ◦ h),
(k) (0)
0 A ◦ h = n, jt,0 A ◦ h ∈ Uk , and j0
provided that doω h∈V.
106
18 BERNHARD LAMEL
We would like to use this theorem to reconstruct a transversal jet of a map

H along a Segre map from a transversal jet (of higher order) on a lower order
Segre map, and thus be able to use a similar iterative process as before. The
main problem is to show that every biholomorphism, when evaluated along a Segre
map, gives rise to the same starting term n ∈ N2 . As a second main ingredient
of the parametrization theorem for minimal holomorphically nondegenerate hyper-
surfaces, we therefore need to introduce invariants similar to the n(k) defined in
k
Definition 3 associated to πM , when considered along Segre maps. The main result
in [20] is now the following:
Theorem 8. Let (M, p) ⊂ (CN , p) be a germ of a generic real analytic subman-
ifold of CN , which is minimal and holomorphically nondegenerate at p. Then there
exists k ∈ N, a neighbourhood V ⊂ Gkp (CN ) of jpk id, and a map Ψ : CN ×Gkp (CN ) ⊃
U → CN , defined on a neighbourhood U of {p} × V , holomorphic in its first and
real-analytic in its second variable, such that
(19) H(Z) = Ψ(Z, jpk H), for all H ∈ Aut(M, p) with jpk H ∈ V.
Furthermore, there exists a polynomial s with s(jpk id) = 0 and for every α ∈ NN a
polynomial pα on Gkp (CN ) and an integer rα , such that
pα (Λ)
(20) Ψ(Z, Λ) = rα
Z α.
N
s(Λ)
α∈N
Furthermore, the integer k = k(M, p) can be bounded by local biholomorphic invari-

ants of (M, p) and can be chosen in such a way that it remains uniformely bounded
on compact subsets of M .
Note that the parametrization obtained in Theorem 8 is in contrast to the
earlier ones not globally defined. Note also that Theorem 8 actually characterizes
the existence of a jet parametrization for minimal real-analytic CR manifolds; the
necessity of holomorphic nondegeneracy follows from the fact that in the holomor-
phically degenerate case, even the space of infinitesimal automorphisms is infinite
dimensional.
5.8. Beyond minimality. In all of the results mentioned up to now, minimal-
ity played an important role; one of the main tools was the use of the Segre maps,
and this technique requires minimality. However, there is a recent result by Juhlin
and the author [17], which gives a Lie group structure on the automorphism group
Aut(M, p) for real-analytic hypersurfaces in C2 at a nonminimal point p for which a
defining function is of the form Im w = (Re w)ϕ(z, z̄, Re w) with ϕ(z, z̄, 0) = 0. This
actually turns out to be a geometric condition (i.e. it is independent of the choice
of coordinates) which is usually refered to as p being of 1-infinite type (following
Meylan [22]); we prefer to say that p is 1-nonminimal. It turns out that we can
again construct a jet parametrization in this setting. In order to get a taste of what
the statement looks in this case, let us state the main result of [17] in C2 , which
provides a jet parametrization for the space Hk (M, p; M , p ) of maps of transversal
order k:
Theorem 9. Let (M, p) and (M , p ) be germs of real-analytic hypersurfaces in
C which are 1-nonminimal at p and p , respectively, and fix some integer k. Then
2
there exist an integer m and for every Λ0 ∈ Jp,p

m 2
(C ) an integer , and for every
19
Λ̃0 ∈ jp (jpm )−1 Λ0 a neighbourhood U ⊂ Jp,p

(C ) of Λ̃0 , and a map Ψ : U → C ,
2 2
defined on a neighbourhood U ⊂ C2 × Jp,p

(C ) of {p} × U , holomorphic in its first
2
and real-analytic in its second factor, such that

(21) H(Z) = Ψ(Z, jp H), for all H ∈ Hk (M, p; M , p ) with jp H ∈ U.
In particular, for any H0 ∈ Hk (M, p; M , p ), the map H → jp H is a homeomor-
phism of a neighbourhood of H0 ∈ H(M, p; M , p ) onto a locally closed, real-analytic
2
subset of Jp,p (C ).
However, the approach taken in this case is quite distinct, and based on earlier
work by Ebenfelt [14]; we construct a certain system of singular ODEs which the
jets of maps will have to fulfill, and one then finds a parametrization theorem for
the solutions of these ODEs; in the case of 1-nonminimal hypersurfaces, it turns
out that the associated system of ODEs has a Fuchsian singularity, which makes it
possible to parametrize the solutions. Whether or not a similar procedure can be
carried out for k-nonminimal hypersurfaces with k > 1, or whether there exists a
Lie group structure for the automorphism groups of these remains an interesting
open question.
References
[1] M. S. Baouendi, P. Ebenfelt, and L. P. Rothschild. Algebraicity of holomorphic mappings
between real algebraic sets in Cn . Acta Math., 177(2):225–273, 1996.
[2] M. S. Baouendi, P. Ebenfelt, and L. P. Rothschild. Parametrization of local biholomorphisms
of real analytic hypersurfaces. Asian J. Math., 1(1):1–16, 1997.
[3] M. S. Baouendi, P. Ebenfelt, and L. P. Rothschild. CR automorphisms of real analytic man-
ifolds in complex space. Comm. Anal. Geom., 6(2):291–315, 1998.
[4] M. S. Baouendi, P. Ebenfelt, and L. P. Rothschild. Rational dependence of smooth and
analytic CR mappings on their jets. Math. Ann., 315(2):205–249, 1999.
[5] M. S. Baouendi, P. Ebenfelt, and L. P. Rothschild. Real submanifolds in complex space and
their mappings, volume 47 of Princeton Mathematical Series. Princeton University Press,
Princeton, NJ, 1999.
[6] M. S. Baouendi, X. Huang, and L. P. Rothschild. Regularity of CR mappings between alge-
braic hypersurfaces. Invent. Math., 125(1):13–36, 1996.
[7] M. S. Baouendi, H. Jacobowitz, and F. Trèves. On the analyticity of CR mappings. Ann. of
Math. (2), 122(2):365–400, 1985.
[8] T. Bloom and I. Graham. On “type” conditions for generic real submanifolds of Cn . Invent.
Math., 40(3):217–243, 1977.
[9] C. Bruschek and H. Hauser. Arcs, cords, and felts—six instances of the linearization principle.
Amer. J. Math., 132(4):941–986, 2010.
[10] E. Cartan. Sur la géométrie pseudo-conforme des hypersurfaces de l’espace de deux variables
complexes ii. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (2), 1(4):333–354, 1932.
[11] E. Cartan. Sur la géométrie pseudo-conforme des hypersurfaces de l’espace de deux variables
complexes. Ann. Mat. Pura Appl., 11(1):17–90, 1933.
[12] H. Cartan. Sur les groupes de transformations analytiques. 1935.
[13] S. S. Chern and J. K. Moser. Real hypersurfaces in complex manifolds. Acta Math., 133:219–
271, 1974.
[14] P. Ebenfelt. On the analyticity of CR mappings between nonminimal hypersurfaces. Math.
Ann., 322(3):583–602, 2002.
[15] P. Ebenfelt, B. Lamel, and D. Zaitsev. Finite jet determination of local analytic CR automor-
phisms and their parametrization by 2-jets in the finite type case. Geometric And Functional
Analysis, 13(3):546–573, Jun 2003.
[16] H. Hauser and G. Müller. A rank theorem for analytic maps between power series spaces.
Inst. Hautes Études Sci. Publ. Math., (80):95–115 (1995), 1994.
[17] R. Juhlin and B. Lamel. On maps between nonminimal hypersurfaces. pages 1–17, Nov 2010.
108
20 BERNHARD LAMEL
[18] F. Klein. Vergleichende Betrachtungen über neuere geometrische Forschungen. Math. Ann.,
43(1):63–100, 1893.
¯
[19] J. J. Kohn. Subellipticity of the ∂-neumann problem on pseudo-convex domains: sufficient
conditions. Acta Math., 142(1-2):79–122, 1979.
[20] B. Lamel and R. Juhlin. Automorphism groups of minimal real-analytic CR manifolds. 2010.
[21] B. Lamel and N. Mir. Parametrization of local CR automorphisms by finite jets and appli-
cations. J. Amer. Math. Soc., 20(2):519–572 (electronic), 2007.
[22] F. Meylan. A reflection principle in complex space for a class of hypersurfaces and mappings.
Pacific J. Math., 169(1):135–160, 1995.
[23] S. B. Myers and N. E. Steenrod. The group of isometries of a riemannian manifold. Ann. of
Math. (2), 40(2):400–416, 1939.
[24] H. Poincaré. Les fonctions analytiques de deux variables et la représentation conforme.
Palermo Rend., 23:185–220, 1907.
[25] N. K. Stanton. Infinitesimal CR automorphisms of rigid hypersurfaces. Amer. J. Math.,
117(1):141–167, 1995.
[26] N. K. Stanton. Infinitesimal CR automorphisms of real hypersurfaces. Amer. J. Math.,
118(1):209–233, 1996.
[27] N. Tanaka. On the pseudo-conformal geometry of hypersurfaces of the space of n complex
variables. J. Math. Soc. Japan, 14:397–429, 1962.
[28] A. E. Tumanov. Extension of CR-functions into a wedge from a manifold of finite type. Mat.
Sb. (N.S.), 136(178)(1):128–139, 1988.
[29] D. Zaitsev. Germs of local automorphisms of real-analytic CR structures and analytic depen-
dence on k-jets. Math. Res. Lett., 4(6):823–842, 1997.
[30] D. Zaitsev. Unique determination of local CR-maps by their jets: a survey. Atti Accad. Naz.
Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl., 13(3-4):295–305, 2002.
Universität Wien, Fakultät für Mathematik, Nordbergstrasse 15, A-1090 Wien

E-mail address: bernhard.lamel@univie.ac.at
Volume 550, 2011
SPLITTING OF HOLOMORPHIC COCYCLES WITH

ESTIMATES. SEVERAL VARIABLES
JÜRGEN LEITERER
Abstract. It is well-known from the Oka-Grauert principle that a holomor-

phic vector bundle over a domain of holomorphy is trivial if it admits a cocycle
of transition functions which is sufficiently close to the unit cocycle. We prove
this result with certain uniform estimates.
1. Introduction
Let D ⊆ C be an open set, U = {Ui }i∈I an open covering of D, L(r, C) the
n
algebra of complex r × r matrices, and GL(r, C) the group of invertible elements of

L(r, C) .
Let C 0 U, OGL(r,C) be the set of all families f = {fi }i∈I of holomorphic

functions fi : Ui → GL(r, C), and let Z 1 U, OGL(r,C) be the set of families
f = {fij }i,j∈I of holomorphic functions fij : Ui ∩ Uj → GL(r, C) satisfying the
cocycle condition
fij fjk = fik on Ui ∩ Uj , i, j ∈ I.
Set

dist(f, 1) = sup fi (ζ) − 1 for f ∈ C 0 U, OGL(r,C)
i∈I, ζ∈Ui
and

dist(f, 1) = sup fij (ζ) − 1 for f ∈ Z 1 U, OGL(r,C) ,
i,j∈I, ζ∈Ui ∩Uj
where
both the unit matrix in L(r, C) and the unit element in
1 denotes
Z 1 U, OGL(r,C) , and M denotes the operator norm of M ∈ L(r, C) as an opera-
tor acting in Cr endowed with the standard norm.
From the Oka-Grauert principle [Gr] it is well known that if D is a domain
of holo-
morphy, then there exists a constant δ > 0 such that, for each f ∈ Z 1 U, OGL(r,C)

with dist(f, 1) ≤ δ, there exists u ∈ C 0 U, OGL(r,C) which solves the Cousin prob-
lem
(1.1) fij = ui u−1
j on Ui ∩ Uj , i, j ∈ I.
In this paper, under certain additional hypotheses, we give an estimate for the
constant δ as well as for the solution u. For example, Theorem 3.5 below (the main
result) implies the following.
2000 Mathematics Subject Classification: 32L05 .

Keywords: holomorphic vector bundle, splitting of holomorphic cocycles, uniform estimates.
1 2011
c American Mathematical Society
109
110
2 JÜRGEN LEITERER
1.1. Theorem. Suppose D is bounded and strictly pseudoconvex 1. Then there

exists a constant 0 < CD < ∞, which depends only on D 2, such that:
Assume 0 < ε ≤ CD /64 and, for each a ∈ D, there exists i ∈ I such that

(1.2) D ∩ ζ ∈ C |ζ − a| < ε ⊆ Ui . 3

Then, for each f ∈ Z 1 U, OGL(r,C) such that
ε
(1.3) dist(f, 1) ≤ ,
CD

there exists u ∈ C 0 U, OGL(r,C) which solves the Cousin problem (1.1) and satisfies
CD
(1.4) dist(u, 1) ≤dist(f, 1).
2ε
For matrices close to the unit matrix, after linearization, the Cousin problem
(1.1) leads to the ∂-equation in degree one (via Dolbeault isomorphism). Since,
on bounded strictly pseudoconvex domains, this equation admits a solution with
uniform estimates [GrLi, He, Li1, Li2, Ov, RS], see also the books [R, HeLe], an
appropriate version of the implicit function theorem quickly leads to the following
result: Assume D is bounded and strictly pseudoconvex, and U satisfies condition
(1.2) for some ε > 0. Then, for each c > 0, there exists δ > 00 suchGL(r,C)
that, for
each
f ∈ Z U, O
1 GL(r,C)
with dist(f, 1) < δ, there exists u ∈ C U, O which
solves the Cousin problem (1.1) and satisfies the estimate dist(u, 1) < c.
So it is natural to analyze the proof of the implicit function theorem in order to
get a proof of Theorem 1.1. However, the author did not succeed in this way (only
a weaker estimate was obtained).
Therefore, here we go another way. First we study the equation
(1.5) U −1 ∂U = A
where A is a given continuous (0, 1)-form with values in L(r, C), and U is searched
as a continuous function from D to GL(r, C). To the knowledge of the author,
this equation appears for the first time in the work of Cornalba and Griffiths [CG],
where, by means of the Newlander-Nierenberg theorem, local solvability is proved.
Then, Gennadi Henkin found another proof 4 for the local solvability, using instead
of the Newlander-Nierenberg theorem the fact that, on balls, the ∂-equation can
be solved with uniform estimates in degrees one and two. Henkin’s proof has the
advantage that it gives local solutions with uniform estimates. Analyzing the proof
of Henkin, we obtain a global solution of (1.5) with certain uniform estimates,
provided A is sufficiently small (Theorem 5.1). Then we prove a theorem on the
solvability of Cousin problems with uniform estimates in the class of continuous
functions f with continuous ∂f (Theorem 8.2). In the last section, we deduce
Theorem 3.5 from Theorems 8.2 and 5.1.
The author has two motivations for the present paper. One motivation is to
provide the Oka-Grauert principle with certain estimates. The second motivation
1with C 2 -boundary, or with piecewise C 2 -boundary as in [RS], or of the type studied in Chapter
3 of [HeLe].
2If n = 1 and d is the diameter of D, then we can take 226 d for C (see [Le2]), but in general,
D
it is not so easy to estimate this constant.
3Below (Def. 2.2) we call such coverings ε-separated.
4Tho the knowledge of the author, this proof is published only in the form of an exercise in
the book [HeLe] (Exercise 10 at the end of Chapter 2).
SPLITTING OF HOLOMORPHIC COCYCLES 111
3
is to prove some estimates for the Weierstrass product theorems obtained in [GoR1,
GoR2, GoLe] for operator functions. The latter one is also the motivation for
another paper of the author [Le2], where the case of one variable is studied, and in
place of L(r, C) an arbitrary Banach algebra with unit is admitted.
Finally let us compare Theorem 1.1 with the following one-variable result of
B. Berndtsson and J.-P. Rosay [BR]: Let D = D be the unit disc in the complex
plane.
Assume condition (1.2) is satisfied for some ε > 0. Then, for each f ∈
Z 1 U, OGL(r,C) satisfying the condition
(1.6) f := sup fij (ζ) < ∞,
i,j∈I, ζ∈Ui ∩Uj

there exists u ∈ C 0 U, OGL(r,C) which solves the Cousin problem (1.1) and satisfies
both
(1.7) u := sup ui (ζ) < ∞ and u−1 := sup u−1
i (ζ) < ∞.
i∈I, ζ∈Ui i∈I, ζ∈Ui
Of course, our condition (1.3) is much stronger than condition (1.6). However, it
seems to the author that the method of [BR], under the stronger condition (1.3),
does not give estimate (1.4), although some weaker estimate (not explicitly stated
in [BR]) can be obtained analyzing the proof of [BR].
2. Notation
Throughout this paper the following notations are used.
• N is the set of natural numbers, zero included. N∗ = N \ {0}. Z is the set
of integers. C is the complex plane. R is the real line.
• If ξ ∈ Cn , then we denote by ξ1 , . . . , ξn the components of ξ. The standard
1/2
Euclidean norm in Cn will be denoted by |·|, i.e., |ξ| = |ξ1 |2 +. . .+|ξn |2
for ξ ∈ Cn .
• L(r, C) is the algebra of complex r × r matrices, and GL(r, C) is the group
of invertible matrices from L(r, C).
• The unit matrix in L(r, C) will be denoted by 1.
• For A ∈ L(r, C) we denote by A the norm of A as a linear endomorphism
of Cr , where Cr is endowed with the standard Euclidean norm, i.e.,

(2.1) A = max Aξ .
ξ∈Cn ,|ξ|=1
• By z1 , . . . , zn we usually mean the canonical complex coordinates on Cn .

• If X is a subset of Cn , then we denote by X the closure of X in Cn , and
by int X we denote the interior of X with respect to Cn .
2.1. In order to give our results also for holomorphic functions which admit a
continuous extension to the boundary, or to some part of the boundary of their
domain of definition, we will consider sets X ⊆ Cn with the property that
(2.2) X ⊆ int X.
By a C ∞ -function on such an X we mean a function which comes from a C ∞ -
function defined in some open (with respect to Cn ) neighborhood of X. As a
consequence of (2.2), the derivatives of such functions are well-defined on X by
their values on int X.
The following definition will be used throughout the paper.
112
4 JÜRGEN LEITERER
2.2. Definition. Let X ⊆ Cn , let U = {Ui }i∈I be a covering of X by relatively

open sets5, and let ε > 0. Then U will be called ε-separated
if for each point
a ∈ X, there exists an index i ∈ I such that X ∩ ζ ∈ Cn |ζ − a| < ε ⊆ Ui .
3. The main result

3.1. Let X be a subset of C such that X ⊆ int X.
n
We denote by BC 0 (X) the Banach space of bounded continuous functions f :

X → C endowed with the sup-norm
(3.1) f X := sup |f (ζ)|,
ζ∈X
and we denote by BC0,q

0
(X), q = 1, 2, the Banach space of all bounded continuous
6
(0, q)-forms on X endowed with the sup-norm
⎧ n
⎪
⎨ max fj X if q = 1 and f = fj dz j ,
1≤j≤n
(3.2) f X := j=1

⎪
⎩ max fjk X if q = 2 and f = fjk dz j ∧ dz k , .
1≤j<k≤n 1≤j<k≤n
We write also BC0,0

0
(X) in place of BC 0 (X).
We denote by BC0,q∂
(X), q = 0, 1, 2, the space of all f ∈ BC0,q
0
(X) such that ∂f
(defined in the sense of distributions in int X) is also bounded and continuous on
X.
Let BC0,1
Ker ∂
(X) be the subspace of all f ∈ BC0,1 ∂
(X) with ∂f = 0, and let
BC0,2
Im ∂
(X) be the subspace of all f ∈ BC0,2
0
(X) which are of the form f = ∂u with
u ∈ BC0,q (X). Notice that BC0,1 (X) is always a closed subspace of the Banach
0 Ker ∂
space BC0,1
0
(X), as ∂ is a closed operator. The space BC0,2
Im ∂
(X) need not be closed
in BC0,2 (X).
0
3.2. Definition. Let X ⊆ Cn such that X ⊆ int X. We say the X satisfies that
∂-condition if the following two conditions are satisfied:
(i) There exists a bounded (with respect to the sup-norm · X ) linear operator
T1 : BC0,1
Ker ∂
(X) → BC ∂ (X)
such that
(3.3) ∂T1 f = f for all f ∈ BC0,1
Ker ∂
(X).
(ii) There exists a bounded (with respect to the sup-norm · X ) linear operator
T2 : BC0,2
Im ∂
(X) → BC0,1
∂
(X)
such that
(3.4) ∂T2 f = f for all f ∈ BC0,2
Im ∂
(X). 7
5i.e., a covering which comes from an open covering of an open (in Cn ) neighborhood of X.
6By a continuous (0, q)-form on X we mean a continuous section over X of the bundle of
(0, q)-form over Cn .
7Notice that then BC Im ∂ (X) is closed in BC 0 (X).
0,2 0,2
5
3.3. Remark on this definition. If n = 1 and X is bounded, then X always

satisfies the ∂-condition. Indeed, then for T2 we can take the zero operator (as then
zero is the only (0, 2)-form on X) and for T1 we can take the Pompeiju operator
defined by

1 f (ζ) ∧ dζ
(T1 f )(z) = − , f ∈ BC0,1
0
(X, C).
2πi X ζ − z
Moreover, then the norm of T1 (which √ plays an important role in Theorem 3.5
below) can be easily estimated by 2 diam X, where diam X is the diameter of X.
Also for n ≥ 2, different classes of domains in Cn are well-known which satisfy
the ∂-condition [GrLi, He, HeP, Li1, Li2, ?, Ov], see also the books [R, HeLe]. But
then, in general, it is not so easy to estimate the norms of the operators T1 and T2
by the geometric properties of D.
3.4. Let X ⊆ Cn with X ⊆ int X.
Then we denote by BOL(r,C) (X) the algebra of all bounded continuous functions
f : X → L(r, C) which are holomorphic in int X, and by BOGL(r,C) (X) we denote
the group of all f ∈ BOL(r,C) (X) such that f (ζ) ∈ GL(r, C) for all ζ ∈ X and,
moreover, the function f −1 , which is then well-defined on X, is also bounded. The
unit element of this group, i.e., the constant function with value 1, will be denoted
also by 1. We define
(3.5) f X = sup f (ζ) for f ∈ BOL(r,C) (X),
ζ∈X
where f (ζ) is defined by (2.1).

Now let U = {Ui }i∈I a covering of X by relatively open subsets of X. Then we
use the
following notations:

C 0 U, BOGL(r,C) is the set of families {fi }i∈I of functions fi ∈ BOGL(r,C) (Ui ),

C 1 U, BOGL(r,C) is the set of families {fij }i,j∈I of functions fij ∈

BOGL(r,C) (Ui ∩ Uj ), and Z 1 U, BOGL(r,C) is the set of all f ∈ C 1 U, BOGL(r,C)
satisfying the multiplicative cocycle condition
(3.6) fij fjk = fik on Ui ∩ Uj ∩ Uk , i, j, k ∈ I.

The elements of Z 1 U, BOGL(r,C) are called multiplicative cocycles. Note that
the cocycle condition (3.6) implies that
−1
(3.7) fij = fji on Ui ∩ Uj , i, j ∈ I,
and
(3.8) fii = 1 on Ui , i ∈ I.

The element f ∈ C U, BOGL(r,C) defined by fi ≡ 1 on Ui , i ∈ I, will be denoted
0

by 1, and the cocycle f ∈ Z 1 U, BOGL(r,C) defined by fij ≡ 1 on Ui ∩ Uj , i, j ∈ I,
will be called the unit cocycle and also denoted by 1. Moreover, we define

(3.9) f − 1 = sup fi − 1Ui if f ∈ C 0 U, BOGL(r,C) ,
i∈I

(3.10) f − 1 = sup fij − 1Ui ∩Uj if f ∈ C 1 U, BOGL(r,C) ,
i,j∈I
where fi − 1Ui and fij − 1Ui ∩Uj are defined by (3.5).
Note that the “numbers” defined by (3.9) and (3.10) can be infinite if I is infinite.
However, in this paper, we study only those f for which f − 1 < 1.
114
6 JÜRGEN LEITERER
Now the main result of this paper can be stated as follows.

3.5. Theorem. Let X ⊆ Cn be a set with X ⊆ int X which satisfies the ∂-condition,
and let T1 , T2 be two operators as in Definition 3.2. Set Γ = 218n+11n n3n .8 Then,
for each ε > 0, the following holds:
If U = {Ui }i∈I is an
ε-separated(Definition 2.2) covering of X by relatively open
sets, and if f ∈ Z 1 U, BOGL(r,C) such that
ε 1
(3.11) f − 1 ≤ and f − 1 ≤ , 9
Γ max(T1 , T2 ) 64

then there exists h ∈ C 0 U, BOGL(r,C) such that
(3.12) fij = hi h−1
j on Ui ∩ Uj , i, j ∈ I,
and

ΓT1
(3.13) h − 1 ≤ 2+ f − 1.
2ε
4. On matrix-valued forms
4.1. Let X be a subset of Cn such that X ⊆ int X.
We denote by (BC 0 )L(r,C) (X) = (BC0,0
0 L(r,C)
) (X) the Banach space of bounded
continuous functions f : X → L(r, C) endowed with the norm
(4.1) f X := sup f (ζ),
ζ∈X
where f (ζ) is defined by (2.1). If f is an L(r, C)-valued continuous (0, q)-form

on X, q = 1, 2, then we define for each ζ ∈ X
⎧ n
⎪
⎨ max fj (ζ) if q = 1 and f = fj dz j ,
1≤j≤n
(4.2) f (ζ) = j=1

⎪
⎩ max fij (ζ) if q = 2 and f = fij dz i ∧ dz j ,
1≤i<j≤n 1≤i<j≤n
where fj (ζ) and fij (ζ) are again defined by (2.1).
We denote by (BC0,q 0 L(r,C)
) (X), q = 1, 2, the Banach space of L(r, C)-valued
bounded continuous (0, q)-forms on X, endowed with the norm
(4.3) f X := sup f (ζ), f ∈ BC0,q
0
(X).
ζ∈X
If f : X → L(r, C) is continuous matrix-valued function on X and g is an

L(r, C)-valued continuous (0, q)-form on X, q = 1, 2, then, in the canonical way, we
pointwise define the products f g and gf . Note that then
(4.4) (f g)(ζ) ≤ f (ζ)g(ζ), ζ ∈ X,
8Of course, the value 218n+11n n3n is not optimal. We present it just to show that Γ depends
only on n. The same is true for other similar constants in this paper.
9Note that the second inequality, in general, does not follow from the first one. The point
is that max(T1 , T2 ) can be very small compared to ε/Γ although the diameter of X is big
compared to ε (the case when the diameter of X is small compared to ε is not interesting, because
then the trivial covering {X} is a refinement of any ε-separated covering of X.). For example,
this happens when n = 1 and X is a rectangle one side of which is small compared to ε and the
other side of which is big compared to ε.
7
and, hence,
(4.5) f gX ≤ f X gX and gf X ≤ gX f X
if f ∈ 0 L(r,C)
(BC0,0 ) (X)and g ∈ 0 L(r,C)
(BC0,q ) (X),
q = 0, 1, 2.
For two L(r, C)-valued continuous (0, 1)-forms

n n
f= fj dz j and g = gj dz j
j=1 j=1
we pointwise define the wedge product f ∧ g by setting

(4.6) (f ∧ g)(ζ) = fi (ζ)gj (ζ) − fj (ζ)gi (ζ) dz i (ζ) ∧ dz j (ζ), ζ ∈ X.
1≤i<j≤n
Note that then

(4.7) (f ∧ g)(ζ) ≤ 2f (ζ)g(ζ), ζ ∈ X,
and, therefore,
(4.8) f ∧ gX ≤ 2f X gX
if f, g ∈ (BC0,1
0 L(r,C)
) (X).
4.2. Extension of the operators T1 and T2 from Definition 3.2 to matrix-
valued forms. Let X be a subset of Cn such that X ⊆ int X.
We denote by (BC0,q∂ L(r,C)
) (X), q = 0, 1, 2, the space of all f ∈ (BC0,q
0 L(r,C)
) (X)
such that ∂f is also bounded and continuous on X.
Ker ∂ L(r,C)
Let (BC0,1 ) (X) be the subspace of all f ∈ (BC0,1
∂ L(r,C)
) (X) with ∂f = 0,
Im ∂ L(r,C)
and let (BC0,2 ) (X) be the subspace of all f ∈ (BC0,20 L(r,C)
) (X) which are of
the form f = ∂u with u ∈ (BC0,q )
0 L(r,C)
(X).
Then each f ∈ (BC0,q0 L(r,C)
) (X), q = 0, 1, 2, can be written uniquely in the form

r
(4.9) f= erμν fμν ,
μ,ν=1
where fμν ∈ BC0,q

0
(X)
and is the matrix in L(r, C) with entry 1 at place (μ, ν)
erμν
and entry 0 at the other places.
Now suppose that X satisfies the ∂-condition and that T1 and T2 are two oper-
ators as in Definition 3.2. Then we define linear operators
T1r : (BC0,1
Ker ∂ L(r,C)
) (X) → (BC ∂ )L(r,C) (X)
and
T2r : (BC0,2
Im ∂ L(r,C)
) (X) → (BC0,1
∂ L(r,C)
) (X)
by setting

r
(4.10) Tqr f = erμν Tq fμν , q = 1, 2,
μ,ν=1
if f is written in the form (4.9). Then we see from (3.3) and (3.4) that
(4.11) ∂T1r f = f for all f ∈ (BC0,1
Ker ∂ L(r,C)
) (X)
and
(4.12) ∂T2r f = f for all f ∈ (BC0,2
Im ∂ L(rC)
) (X).
116
8 JÜRGEN LEITERER
4.3. Lemma. The operators T1r and T2r are bounded with respect to the sup-norm,
where T1r = T1 and T2r = T2 .
Proof. We only have to prove that
(4.13) Tqr ≤ Tq
for q = 1, 2. The opposite inequality is obvious.
Let L(r, C)∗ be the space of linear functionals Φ : L(r, C) → C endowed with the
norm
Φ = max Φ(A).
A∈L(r,C) , A=1
For f ∈ 0 L(r,C)
(BC0,q ) (X),q = 0, 1, 2, and Φ ∈ L(r, C)∗ , we denote by Φ ◦ f the form
0
in (BC0,q )(X) defined by

r

(4.14) Φ◦f = Φ erμν fμν
μ,ν=1
if f is written in the form (4.9).

Then it follows from the linearity of Φ that, for all ζ ∈ X,

(4.15) (Φ ◦ f )(ζ) = Φ f (ζ) if q = 0,

n
(4.16) (Φ ◦ f )(ζ) = Φ fj (ζ) dz j (ζ) if q = 1
j=1
n
and f is written in the form f = j=1 fj dz j , and

(4.17) (Φ ◦ f )(ζ) = Φ fij (ζ) dz i (ζ) ∧ dz j (ζ) if q = 2
1≤i<j≤n
n
and f is written in the form f = 1≤i<j≤n fij dz i ∧ dz j .
By the Hahn-Banach theorem it follows from (4.15) - (4.17) that, for all f ∈
0 L(r,C)
(BC0,q ) (X), q = 0, 1, 2.
(4.18) max Φ ◦ f X = f X .
Φ∈L(r,C)∗ , Φ=1
Now let f ∈ (BC0,1

Ker ∂ L(r,C)
) (X) be written in the form (4.9), and let Φ ∈ L(r, C)∗ .
Then, using the linearity of T1 , we see from (4.14) and (4.10) that
r r
T1 Φ ◦ f = T 1 Φ erμν fμν = Φ erμν T1 fμν
μ,ν=1 μ,ν=1

r
=Φ◦ erμν T1 fμν = Φ ◦ (T1r f ).
μ,ν=1
Together with (4.18) this implies that, for all f ∈ (BC0,1

Ker ∂ L(r,C)
) (X),
T1r f = max Φ ◦ (T1r f ) = max T1 (Φ ◦ f )
Φ∈L(r,C)∗ , Φ=1 Φ∈L(r,C)∗ , Φ=1
≤ T1 max Φ ◦ f = T1 f ,

Φ∈L(r,C)∗ , Φ=1
i.e., we have (4.13) for q = 1. For q = 2 the proof is the same.

9
5. An estimate for the equation U −1 ∂U = A

In this section, we use the notations introduced in the preceding section. The
aim of the present section is to prove the following theorem.
5.1. Theorem. Let X ⊆ Cn be a set with X ⊆ int X which satisfies the ∂-condition,
and let T1 , T2 be two operators as in Definition 3.2. Let A be an L(r, C)-valued
bounded continuous (0, 1)-form on X such that
(5.1) ∂A = −A ∧ A on X
and
1
(5.2) AX ≤ .
10 max(T1 , T2 )
Then there exists a bounded continuous function U : X → GL(r, C) such that
(5.3) ∂U = U A on X,
and
(5.4) U − 1X ≤ 2T1 AX .
We first prove a lemma. If A is an L(r, C)-valued bounded continuous (0, 1)-form
on X, then we define a linear differential operator
A ∂ L(r,C)
∂ : (BC0,1 ) (X) → (BC0,2
0 L(r,C)
) (X)
setting
A
∂ f = ∂f + f ∧ A.
A A
The kernel of ∂ will be denoted by Ker ∂ . It is a closed subspace of
0 L(r,C) A
(BC0,1 ) (X) (since ∂ is a closed operator) and will be also considered as
0 L(r,C)
Banach space endowed with the norm of (BC0,1 ) (X).
5.2. Lemma. Under the hypotheses of Theorem 5.1, there exists a bounded linear
operator
A
R : Ker ∂ → (BC 0 )L(r,C) (X)
such that
A
(5.5) ∂Rf = f + (Rf )A for all f ∈ Ker ∂
and
(5.6) R ≤ 2T1 .
Proof. Let
T1r : (BC0,1
Ker ∂ L(r,C)
) (X) → (BC 0 )L(r,C) (X)
and
T2r : (BC0,2
Im ∂ L(r,C)
) (X) → (BC0,1
0 L(r,C)
) (X)
be the bounded linear operators defined by (4.10). Recall that, by Lemma 4.3,
(5.7) T1r = T1 and T2r = T2 .
A
If f ∈ Ker ∂ 1 , then
(5.8) f ∧ A = −∂f.
118
10 JÜRGEN LEITERER
Therefore, then f ∧ A belongs to (BC0,2

Im ∂ L(r,C)
) (X), the domain of definition of T2r .
Moreover, then, by (4.12), (5.1), and (5.8),

∂ f + T2r (f ∧ A) = ∂f + f ∧ A = ∂f − ∂f = 0,
i.e., f + T2r (f ∧ A) belongs to the domain of definition of T1r . Hence, we can define
linear operators
S A M
(BC 0 )L(X,C) (X) ←− Ker ∂ 1 −→ (BC0,1
0 L(X,C)
) (X)
setting

Sf = T1r f + T2r (f ∧ A) and M f = T2r (f ∧ A) − (Sf )A.
It follows from (5.2), (5.7), and (4.8) that these operators are bounded, where
6
(5.9) S ≤ T1
5
and
8
(5.10) M ≤ .
25
We claim that
A
(5.11) ∂ Mf = 0
A
for all f ∈ Ker ∂ .
A A
To prove the claim, let f ∈ Ker ∂ be given. Then, by definition of ∂ and M ,
we have (in the sense of distributions)
A
∂ M f = ∂M f + (M f ) ∧ A

= ∂T2r (f ∧ A) − ∂ (Sf ) ∧ A + T2r (f ∧ A) ∧ A − (Sf )A ∧ A.
Moreover, since, by (4.11) and (5.1), both ∂Sf = f + T2r (f ∧ A) and ∂A = −A ∧ A

are continuous, we can apply the product rule and get

∂ (Sf ) ∧ A = f ∧ A + T2r (f ∧ A) ∧ A − (Sf )A ∧ A.
Together this implies that
A
∂ M f = ∂T2r (f ∧ A) − f ∧ A.
By (4.12), this further implies (5.11).
A
From (5.11)and (5.10) we see that id +M is an isomorphism of Ker ∂ , where
∞
(id +M )−1 = k=0 (−M )k and
∞
8k 25
(5.12) (id +M )−1 ≤ = .
25k 17
k=0
This implies that
R := S(id +M )−1
A
is a well-defined bounded linear operator from Ker ∂ to (BC 0 )L(X,C) (X), where,
by (5.9) and (5.12),
6 25
R ≤ S(id +M )−1 ≤ · T1 ≤ 2T1 .
5 17
11
It remains to prove (5.5).

A
Let f ∈ Ker ∂ be given. Then, by definition of R and S,

−1 −1 −1
r
∂Rf = ∂S(id +M ) f = ∂T1 (id +M ) f + T2 r
(id +M ) f ∧ A .
Moreover, by definition of M ,

T2r
(id +M ) f ∧ A = M (id +M )−1 f + S(id +M )−1 f A
−1
and hence, by definition of R,

T2r
(id +M ) f ∧ A A = M (id +M )−1 f + (Rf )A.
−1

∂Rf = ∂T1r (id +M )−1 f + M (id +M )−1 f + (Rf )A = ∂T1r f + (Rf )A .
Taking into account again (4.11), we get ∂Rf = f + (Rf )A.
5.3. Proof of Theorem 5.1. Let R be the operator from Lemma 5.2. Set U =
1 + RA. Then, by (5.6) and (5.2),
1
U − 1X = RAX ≤ 2T1 AX ≤ .
4
In particular, the values of U are invertible, and estimate (5.4) is satisfied. More-
over, from (5.5) we see that
∂U = ∂(1 + RA) = ∂RA = A + (RA)A = (1 + RA)A = U A.
6. Partitions of unity with estimates

In this section, we use the following notations. For ξ ∈ Cn and r > 0, we set

B(ξ, r) = ζ ∈ Cn |ζ − ξ| < r and B(ξ, r) = ζ ∈ Cn |ζ − ξ| ≤ r .
For ε > 0 and μ = (μ1 , . . . μ2n ) ∈ Z2n , we denote by qμε the point in Cn with
μj
(6.1) xj (qμε ) = √ ε,
2n
where x1 , . . . , x2n are the real coordinate functions on Cn with zj = xj + ixj+n . It
is easy to see that

ε ε
(6.2) B qμ , = C.
2n
2
μ∈Z
6.1. Lemma. Let ε > 0, let n ∈ N∗ , and let J ⊆ Z2n such that10 J ≥ 26n nn . Then

(6.3) B qμε , ε = ∅.
μ∈J
10J denotes the number of J.

120
Proof. For μ ∈ Z2n we denote by J(μ) the set of all indices ν ∈ Z2n such that
(6.4) |qνε − qμε | < 2ε.
By definition (6.1), the inequality (6.4) can be written in the form
2n 2
μj
√ ε − √νj ε < 4ε2 ,
2n 2n
j=1
or, equivalently,

2n
|μj − νj |2 < 8n.
j=1
In particular, for all μ ∈ Z and ν ∈ J(μ),

2n
√
(6.5) |μj − νj | ≤ 8n for j = 1, . . . , 2n,
√ √
where 8n is the entire part of 8n. Since, for fixed μ, the number of indices
√ 2n
ν ∈ Z2n satisfying (6.5) is equal to 2 8n + 1 and
√ 2n √ 2n √ 2n √ 2n
2 8n + 1 ≤ 2 8n + 1 < 6 n+1 < 8 n = 26n nn ,
we get
(6.6) J(μ) < 26n nn for all μ ∈ Z2n .
Now let J ⊆ Z2n with J ≥ 26n nn be given. Then, by (6.6), for each μ ∈ J, there
exists at least one index ν ∈ J such that ν ∈ J(μ), i.e., |qνε − qμε | ≥ 2ε. Then
B(qμε , ε) ∩ B(qνε , ε) = ∅. In particular, we have (6.3).
6.2. Lemma. Let ε > 0 and n ∈ N∗ . Then there exists a C ∞ -partition of unity
{χμ }μ∈Z2n subordinated to the open covering

(6.7) B qμε , ε 2n μ∈Z
of C (by (6.2), this is indeed a covering of Cn ) such that

n

∂χμ 26n+4 nn

(6.8) ∂z j ≤ ε
, 1 ≤ j ≤ n, μ ∈ Z2n .
Proof. Take a C ∞ -function χ : [0, ∞[→ [0, 1] such that χ ≡ 1 in a neighborhood of

[0, 1], χ ≡ 0 in a neighborhood of [2, ∞[, and

(6.9) χ ≤ 2 everywhere on [0, ∞[.
We set
|ζ − qμε |2
μ (ζ) = χ 4
χ , ζ ∈ Cn , μ ∈ Z2n .
ε2
Then, for all μ ∈ Z2n ,

ε
(6.10) μ ≡ 1 in a neighborhood of B
χ qμε , ,
2

ε
(6.11) μ ⊆ BC qμ , √ .
supp χ n
ε
2
13
Moreover, we set
φ= μ .
χ
μ∈Z2n
By Lemma 6.1, the sum in the definition of φ is locally finite. Therefore, φ is a
C ∞ -function on Cn and
∂φ ∂χ μ
(6.12) = , 1 ≤ j ≤ n.
∂z j 2n
∂z j
μ∈Z
From (6.10) and (6.2) we see that

(6.13) φ≥1 everywhere on Cn .
Therefore, setting
μ
χ
χμ = , μ ∈ Z2n ,
φ
we obtain a partition of unity {χμ }μ∈Z2n on Cn . By (6.11), this partition of unity
is subordinated to the open covering (6.7).
It remains to prove estimate (6.8). We have

∂χμ 4 |ζ − qμε |2
μ 4
(ζ) = 2 χ zj (ζ − qμε ).
∂z j ε ε2
Taking into account (6.9) and the fact that, by (6.11),

4|ζ − qμε |2
|zj (ζ − qμ )| < ε if μ
χ = 0,
ε2
this implies that

∂χ
(6.14) μ ≤ 8 on Cn .
∂z j ε
Since, by (6.11) and Lemma 6.1, the sum in (6.12) locally contains not more than
26n nn non-zero terms, this implies that

∂φ 26n+3 nn

(6.15) ∂z j ≤ ε
on Cn .
As
∂χμ 1 ∂χμ χμ ∂φ
= − 2 ,
∂z j φ ∂z j φ ∂z j
μ ≤ 1, from (6.14) and (6.15) we see that
φ ≥ 1 and χ

∂χμ ∂ χ
≤ μ + ∂φ ≤ 8 + 2
6n+3 n
n 26n+4 nn
∂z j ∂z j ∂z j ε ≤ .
ε ε

6.3. Let X ⊆ Cn such that X ⊆ int X, and let U = {Ui }i∈I be a covering of X by
relatively open subsets of X.
We say that {χi }i∈I is a C ∞ partition of unity subordinate to U if
(i) for each i ∈ I, χi is a non-negative real C ∞ -function on X (in the sense
explained in Section 2.1) such that supp χi is relatively compact in Ui ,
(ii) for each a ∈ X, there exists a relative open neighborhood U (a) ⊆ X of a
that χi ≡ 0 on U (a) for all i ∈ I except for a finite number;
such
(iii) i∈I χi ≡ 1 on X.
122
If, for some m ∈ N∗ , the number of the set J(a) in condition (ii) can be always
chosen ≤ m, then U will be called of order ≤ m.
Note that, by Lemma 6.1, each C ∞ partition of unity, which is subordinate to
the covering

(6.16) X ∩ B(qμε , ε) μ∈Z2
is of order ≤ 26n nn − 1. We now combine Lemmas 6.1 and 6.2.

6.4. Lemma. Let X ⊆ Cn , and let U = {Ui }i∈I be an ε-separated (Definition 2.2)
covering of X by relatively open sets, ε > 0. Then there exists a C ∞ -partition of
unity {χi }i∈I subordinated to U, which is of order ≤ 26n nn , such that

∂χi 212n+4 n2n

(6.17) ∂z j ≤ ε
, 1 ≤ j ≤ n, i ∈ I.
Proof. As U is ε-separated, the covering (??) is a refinement of U, i.e., there is a

map τ : Z2n → I such that

(6.18) X ∩ B qμε , ε ⊆ Uτ (μ) , μ ∈ Z2n .
∞
By Lemma 6.2, there a C partition
ε exists of unity {
χμ }μ∈Z2n subordinate to the
open covering B qμ , ε μ∈Z2n of Cn which satisfies

∂χ
μ 26n+4 nn

(6.19) ∂z j ≤ ε
, 1 ≤ j ≤ n, μ ∈ Z2n .
Now, for i ∈ I, we define on X: χi = 0 if i ∈ τ (Z2n ), and

(6.20) χi = μ
χ if i ∈ τ (Z2n ).
μ∈τ −1 (i)
As the sets τ −1 (i), i ∈ I, are pairwise disjunct and I is the union of these sets, it
is clear that
χi ≡ 1 on X,
and from (6.18) we see that supp χi ⊆ Ui . Hence, {χi }i∈I is a C ∞ partition of unity
subordinated to U.
Moreover, by Lemma 6.1, the partition { χμ }μ∈Z2n is of order ≤ 26n nn − 1. Since
−1
the sets τ (i), i ∈ I, are pairwise disjoint, this implies, by (6.20), that also the
partition {χi }i∈I is of order ≤ 26n nn − 1.
The fact that the partition of unity {χμ }μ∈Z2n is of order ≤ 26n nn − 1, moreover
implies that, in the sum (6.20), locally, not more than 26n nn − 1 terms are different
from zero. Together with (6.19) this yields the required estimate (6.17).
7. Continuous functions with continuous Cauchy-Riemann derivative.

The additive case
7.1. Let X ⊆ Cn such that X ⊆ int X.
Note that the space (BC 0 )L(r,C) (X) introduced in Section 4.1 carries also a
point-wise defined multiplicative structure and so becomes an algebra. We de-
note by (BC ∂ )L(r,C) (X) its subalgebra of all f ∈ (BC 0 )L(r,C) (X) such that ∂f ∈
0 L(r,C)
(BC0,1 ) (X) (Section 4.1).
15
Now let U = {Ui }i∈I be a covering of X by relatively open subsets of X. Then

we use the following notations:11

• C 0 U, (BC ∂ )L(r,C) is the algebra of all families {fi }i∈I of functions fi ∈
(BC ∂ )L(r,C) (Ui ).

• C 1 U, (BC ∂ )L(r,C) is the space 12 of all families {fij }i,j∈I of functions fij ∈
(BC ∂ )L(r,C) (Ui ∩ Uj ).

• Z 1 U, (BC ∂ )L(r,C) is the subspace of all f ∈ C 1 U, (BC ∂ )L(r,C) satisfying
the following (additive) cocycle condition:
(7.1) fij + fjk = fik on Ui ∩ Uj ∩ Uk , i, j, k ∈ I.
The elements of this subspace will be called additive 1-cocycles.

• C 2 U, (BC ∂ )L(r,C) is the space of all families {fijk }i,j,k∈I of functions
fijk ∈ (BC ∂ )L(r,C) (Ui ∩ Uj ∩ Uk ).

• Z 2 U, (BC ∂ )L(r,C) is the subspace of all f ∈ C 2 U, (BC ∂ )L(r,C) satisfying
the following condition (also called cocycle condition):
(7.2) − fjkl + fikl − fijl + fijk = 0 on Ui ∩ Uj ∩ Uk ∩ Ul , i, j, k, l ∈ I.
The elements
of this subspace
will be called additive 2-cocycles.
• For f ∈ C q U, (BC ∂ )L(r,C) , q = 0, 1, 2, we define (see (4.3) for the nota-
tions)
f = sup fi Ui , ∂f = sup ∂fi Ui if q = 0,
i∈I i∈I
(7.3) f = sup fij Ui ∩Uj , ∂f = sup ∂fij Ui ∩Uj if q = 1,
i,j∈I i,j∈I
f = sup fijk Ui ∩Uj ∩Uk , ∂f = sup ∂fijk Ui ∩Uj ∩Uk if q = 2,
i,j,k∈I i,j,k∈I
and
(7.4) f ∂ = f + ∂f .
Note that these “numbers” can be infinite if I is infinite. The space of all

f ∈ C q U, (BC ∂ )L(r,C) with f ∂ < ∞ is a Banach space.
• We define linear operators

δ : C q U, (BC ∂ )L(r,C) → C q+1 U, (BC ∂ )L(r,C) , q = 0, 1,

setting, for f ∈ C 0 U, (BC ∂ )L(r,C) ,
(7.5) (δf )ij = fi − fj on Ui ∩ Uj , i, j ∈ I,

and, for f ∈ C U, (BC )
1 ∂ L(r,C)
,
(7.6) (δf )ijk = −fjk + fik − fij on Ui ∩ Uj ∩ Uk , i, j, k ∈ I.

• The element f ∈ C q U, (BC ∂ )L(r,C) defined by fi ≡ 0 if q = 1, fij ≡ 0 if
q = 1, and fijk ≡ 0 if q = 2 will be denoted by 0.
11These are notations from the theory of Čech cohomology with coefficients in sheaves, but, in
this paper, we will not use this theory, except for some very simple facts, which will be explained.
Note that the map U → (BC ∂ )L(r,C) (U ) applied to the relatively open subsets of X is only a
presheaf, but not a sheaf.
12As C 0 U, (BC ∂ )L(r,C) , also C 1 U, (BC ∂ )L(r,C) is an algebra, but we will not use the mul-

tiplicative structure of C 1 U, (BC ∂ )L(r,C) .
124
It is easy to check that

(7.7) δC q U, (BC ∂ )L(r,C) ⊆ Z q+1 U, (BC ∂ )L(r,C) , q = 0, 1.

Note also that the definition of δ is chosen so that an element f ∈ C 1 U, (BC ∂ )L(r,C)
is an additive 1-cocycle if and only if δf = 0.13 Moreover, it is well-known from

the general theory of Čech cohomology that each f ∈ Z q U, (BC ∂ )L(r,C) , q = 1, 2,
is of the form
f = δu,

where u ∈ C q−1 U, (BC ∂ )L(r,C) . We need a version with estimates of the latter fact,
which is stated by the following lemma (the proof of this lemma is a modification
of the corresponding arguments from the general theory).
7.2. Lemma. Let X ⊆ Cn such that X ⊆ int X, let ε > 0, and let U be
an ε-separated (Definition 2.2) covering of X by relatively open sets. Then,

for each f ∈ Z q U, (BC ∂ )L(r,C) , q = 1, 2, such that f ∂ < ∞, there exists

u ∈ C q−1 U, (BC ∂ )L(r,C) such that
(7.8) δu = f,
(7.9) u ≤ f
and
218n+4 n3n
(7.10) ∂u ≤ ∂f + f .
ε

Proof. Let U = {Ui }i∈I , and let f = {fij }i,j∈I ∈ Z q U, (BC ∂ )L(r,C) with f ∂ <
∞ be given. Then, by Lemma 6.4, there exists a C ∞ partition of unity {χi }i∈I
subordinated to U, which is of order ≤ 26n nn , such that

∂χi 212n+4 n2n

(7.11) ∂z μ ≤
(ζ)
ε
, ζ ∈ Cn , 1 ≤ μ ≤ n, i ∈ I.

First let q = 1. Then we define a u = {ui }i∈I ∈ C 0 U, (BC ∂ )L(r,C) by

(7.12) ui = − χk fki .
k∈I

As f is an additive 1-cocycle and χi ≡ 1, then

ui − uj = χk − fki + fkj = χk fij = fij ,
k∈I k∈I

i.e., we have (7.8). Estimate (7.9) is clear, since all χk are non-negative and χk ≡
1. Further, by (7.12),
∂ui ∂fki
∂χk

(ζ) = − χ k (ζ) + f (ζ) , ζ ∈ Ui , i ∈ I.
∂z μ ∂z μ ∂z μ ki
k∈I
∂χ
Hence

∂u ≤ ∂f + f sup k (ζ).
n
∂z μ
1≤μ≤n , ζ∈C
k∈I
Since {χi } is of order ≤ 26n nn , now estimate (7.10) follows from (7.11).
13In the general theory of Čech cohomology, such an operator δ is defined also on

C2 U, (BC ∂ )L(r,C) , and its kernel is Z 2 U, (BC ∂ )L(r,C) . Here we do not need this.
17

Now let q = 2. Then we define a u ∈ C 1 U, (BC ∂ )L(r,C) setting

(7.13) uij = − χν fνij on Ui ∩ Uj , i, j ∈ I.
ν∈I
Then, for all i, j, k ∈ I,

(δu)ijk = −ujk + uik − uij = χν fνjk − fνik + fνij on Ui ∩ Uj ∩ Uk .
ν∈I
Moreover, as f is an additive 2-cocycle, for all ν, i, j, k ∈ I,

0 = (δf )νijk = −fijk + fνjk − fνik + fνij ,
i.e.,
fνjk − fνik + fνij = fijk on Uν ∩ Ui ∩ Uj ∩ Uk .
Hence
(δu)ijk = χν fijk = fijk on Ui ∩ Uj ∩ Uk , i, j, k ∈ I,
ν∈I

i.e., we have (7.8). Estimate (7.9) is clear, since all χν are non-negative and χi ≡
1. Further, by (7.13), for 1 ≤ μ ≤ n and i, j ∈ I,
∂uij ∂fνij
∂χν

(ζ) = − χν (ζ) + fνij (ζ) , ζ ∈ Ui ∩ Uj .
∂z μ ∂z μ ∂z μ
ν∈I

As all χν are non-negative and χν ≡ 1, this implies that
∂χ

∂u ≤ ∂f + f sup ν
n
∂z μ (ζ).
1≤μ≤n , ζ∈C ν∈I
Since {χν } is of order ≤ 26n nn , now estimate (7.10) follows from (7.11).
8. Continuous functions with continuous Cauchy-Riemann derivative.

The multiplicative case
8.1. Let X ⊆ Cn such that X ⊆ int X.
Then we denote by (BC ∂ )GL(r,C) (X) the group of all f ∈ (BC ∂ )L(r,C) (X) (Section
7.1) such that f (ζ) ∈ GL(r, C) for all ζ ∈ X and, moreover, f −1 also belongs to
(BC ∂ )L(r,C) (X).
Note that f belongs to this group if and only if f ∈ (BC ∂ )L(r,C) (X), all values
of f are invertible, and
sup f −1 (ζ) < ∞.
ζ∈X
This follows from the quotient rule
∂f −1 = −f −1 (∂f )f −1 ,
which holds true for each continuous f : int X → GL(r, C) such that ∂f is also
continuous on int X.
Now let U = {Ui }i∈I be a covering of X by relatively open subsets of X. Then,
additional to the notations introduced in Sections 3.4 and 7.1, here we need also
the following notations:

• C 0 U, (BC ∂ )GL(r,C) is the group of all f ∈ C 0 U, (BC ∂ )L(r,C) such that
fi ∈ (BC ∂ )GL(r,C) (Ui ) for all i ∈ I.
126

• C 1 U, (BC ∂ )GL(r,C) is the set14 of all f ∈ C 1 U, (BC ∂ )L(r,C) such that
fij ∈ (BC ∂ )GL(r,C) (Ui ∩ Uj ) for all i, j ∈ I.

• Z 1 U, (BC ∂ )GL(r,C) is the subset of all f ∈ C 1 U, (BC ∂ )GL(r,C) satisfying
the multiplicative cocycle condition
(8.1) fij fjk = fik on Ui ∩ Uj ∩ Uk , i, j, k ∈ I.
The elements of this subset will be called multiplicative cocycles.

• For g ∈ C U, (BC )
0 ∂ GL(r,C)
and f ∈ C 1 U, (BC)∂ )GL(r,C) , we define g

f ∈ C 1 U, (BC ∂ )GL(r,C) setting
(8.2) (g f )ij = gi−1 fij gj on Ui ∩ Uj , i, j ∈ I.
Note that g 1 is always a multiplicative cocycle. For later reference, note

also that, for all g, h ∈ C 0 U, (BC ∂ )GL(r,C) and f ∈ C 1 U, (BC ∂ )GL(r,C) ,
(8.3) (gh) f = h (g f ).
The aim of this section is to prove the following theorem.
8.2. Theorem. Let X ⊆ Cn such that X ⊆ int X, and let U = {Ui }i∈I
be an ε-separated (Definition 2.2) covering of X by relatively open sets. Let

f ∈ Z 1 U, (BC ∂ )GL(r,C) such that
1
(8.4) f − 1 ≤ .
64

Then there exists g ∈ C 0 U, (BC ∂ )GL(r,C) such that
(8.5) g f = 1,
(8.6) g − 1 ≤ 2f − 1,

and
218n+7 n3n
(8.7) ∂g ≤ f − 1 + 2∂f .
ε
We first prove the following lemma.
8.3. Lemma. Let X ⊆ Cn such that X ⊆ int X, and let U = {Ui }i∈I be
an ε-separated (Definition 2.2) covering of X by relatively open sets. Let f ∈

Z 1 U, (BC ∂ )GL(r,C) such that
1
(8.8) f − 1 ≤ .
64

Then there exists g ∈ C 0 U, (C ∂ )GL(r,C) such that
1
(8.9) g f − 1 ≤ f − 1,
8
218n+9 n3n
(8.10) ∂(g f ) ≤ f − 12 + 16∂f f − 1,
ε
65
(8.11) g − 1 ≤ f − 1,
64
218n+5 n3n 33
(8.12) ∂g ≤ f − 1 + ∂f .
ε 32
14This is also a group, but here we will not use the group structure.
19
Proof. We set a = f − 1. In general, a is not an additive 1-cocycle, i.e. δa = 0

(Section 7.1). But, as fik = fij fjk , we have
1 + aik = (1 + aij )(1 + ajk ) = 1 + aij + ajk + aij ajk
and therefore
(δa)ijk = −ajk + aik − aij = aij ajk on Ui ∩ Uj ∩ Uk , i, j, k ∈ I.
Hence
(8.13) δa ≤ a2
and
(8.14) ∂δa ≤ 2a∂a.
As δa is an additive 2-cocycle (see (7.7)), it follows from Lemma 7.2 that there

exists u ∈ C 1 U, (BC ∂ )L(r,C) such that
(8.15) δu = δa,
u ≤ δa,
and
218n+4 n3n
∂u ≤ ∂δa + δa.
ε
By (8.13) and (8.14), the last two estimates further imply
(8.16) u ≤ a2
and
218n+4 n3n
(8.17) ∂u ≤ 2a∂a + a2 .
ε
By (8.15), a − u is an additive 1-cocycle. Therefore, again from Lemma 7.2 we get

v ∈ C 0 U, (BC ∂ )L(r,C) such that
(8.18) δv = a − u,
(8.19) v ≤ a − u,

and
218n+4 n3n
(8.20) ∂v ≤ ∂a − ∂u + a − u.
ε
By (8.19) and (8.16), v ≤ a + a2 . By (8.8) this implies that
65
(8.21) v ≤ a
64
and
65
(8.22) v < .
212
In particular v < 1, which implies (by the arguments given at the beginning of

Section 8.1) that g := 1 + v belongs to the group C 0 U, (BC ∂ )GL(r,C) . We will
show that this g has the required properties. Estimate (8.11) is clear by (8.21).
To prove the remaining properties we set
(8.23) θ = g −1 − 1 + v.
128

As g −1 and v belong to C 0 U, (BC ∂ )L(r,C) , then also θ ∈ C 0 U, (BC ∂ )L(r,C) .
Moreover, since v < 1, for g −1 we have the representation
∞

g −1 = (−v)μ ,
μ=0
where the convergence is absolute with respect to · . (Below we shall prove

that the convergence is even absolute with respect to the Banach space norm · ∂
defined by (7.4).) Therefore, with the same kind of convergence,
∞
(8.24) θ= (−v)μ .
μ=2
More precisely, we see from (8.21) that

∞ ∞
652
(8.25) θ ≤ v2 vμ ≤ 12 a2 vμ .
μ=0
2 μ=0
Since, by (8.22),
∞
∞
μ
65 1 212
v ≤
μ
= 65 = 212 − 65 ,
μ=0 μ=0
212 1 − 212
this implies
652
(8.26) θ ≤ a2 ≤ 2a2 .
212 − 65
From (8.23) we see
(g f )ij = (1 − vi + θi )(1 + aij )(1 + vj )
= 1 − vi + aij + vj − vi aij + aij vj − vi vj − vi aij vj + θi (1 + aij )(1 + vj )
Since, by (8.18), −vi + aij + vj = aij − (δv)ij = uij , this implies that
(8.27) (g f )ij − 1 = uij − vi aij + aij vj − vi vj − vi aij vj + θi (1 + aij )(1 + vj ).
Hence
g f − 1 ≤ u + 2va + v2 + v2 a + θ(1 + a)(1 + v).
In view of (8.16), (8.21), (8.8), (8.26), and (8.22), this implies

65 652 652 65 212 + 65
(8.28) g f − 1 ≤ 1 + + 12 + 18 + 2 · · a2 ≤ 8a2 .
32 2 2 64 212
Taking again into account that a ≤ 1/64, this implies (8.9).
From (8.20), (8.16), and (8.17) we see that
218n+4 n3n
∂v ≤ ∂a + ∂u + a + u
ε
218n+4 n3n 218n+4 n3n
≤ ∂a + 2a∂a + a2 + a + a2
ε ε
218n+4 n3n
= ∂a + 2a∂a + a + 2a2 .
ε
As a ≤ 1/64, this further implies that
33 218n+5 n3n
(8.29) ∂v ≤ ∂a + a.
32 ε
21
Since ∂g = ∂v, this proves (8.12).

Next we estimate ∂θ. From the product rule it follows that
∂(−v)μ ≤ μ∂vvμ−1 if μ ≥ 1.
By (8.21) this implies that
65
∂(−v)μ ≤ μ∂v avμ−2 if μ ≥ 2,
64
and further, by (8.22),
μ−2
65 65
∂(−v) ≤ μ∂v a 12
μ
if μ ≥ 2.
64 2
Moreover,
∞ μ−2 ∞ μ−2 ∞ μ−2
65 μ 65 65
μ 12 ≤ sup μ−2 = 2
μ=2
2 μ≥2 2 μ=2
211 μ=2
211
∞ μ−2
1 32
<2 = .
μ=2
16 15
∞
65 32 7
∂(−v)μ ≤ · ∂va ≤ ∂va.
μ=2
64 15 3
Hence, by (8.24),
7
∂θ ≤ ∂va,
3
and further, by (8.29),

7 33 218n+5 n3n 218n+7 n3n 5
(8.30) ∂θ ≤ ∂a + a a ≤ a2 + ∂aa.
3 32 ε ε 2
From (8.27) and the product rule we see
∂(g f ) ≤ ∂u + 2∂va + 2∂av + 2∂vv + 2va∂v

+ v2 ∂a + ∂θ(1 + a)(1 + v) + θ ∂a + ∂v + v∂a + a∂v .
Taking into account that
65 65 1 a
v ≤ a, v < 12 , a < , and θ ≤ 2a2 ≤
64 2 64 32
(see (8.21), (8.22), (8.8), and (8.26)), this implies that
65 65 65
∂(g f ) ≤ ∂u + 2∂va + ∂aa + ∂va + 11 a∂v
32 32 2
652 a 65 1
+ 18 a∂a + 2∂θ + ∂a + ∂v + 12 ∂a + ∂v
2 32 2 64
2

65 65 1 65
≤ ∂u + + 18 + + ∂aa
32 2 32 217

65 65 1 1
+ 2+ + 11 + + 11 ∂va + 2∂θ
32 2 32 2
≤ ∂u + 3∂aa + 5∂va + 2∂θ.
130
Together with (8.17), (8.29), and (8.30) this proves (8.10):

218n+4 n3n
∂(g f ) ≤ 2a∂a + a2 + 3∂aa
ε 18n+7 3n
33 218n+5 n3n 2 n 5
+5 ∂a + a a + 2 a2 + ∂aa
32 ε ε 2
18n+4 3n 18n+5 3n 18n+7 3n

2 n 2 n 2 n
≤ +5 +2 a2
ε ε ε

33 218n+9 n3n
+ 2 + 3 + 5 + 5 ∂aa ≤ a2 + 16∂aa.
32 ε

8.4. Proof of Theorem 8.2. We write for short
218n+9 n3n
(8.31) C= .
ε
Then, from Lemma 8.3 we get sequences (gj )j∈N∗ of elements gj ∈

C 0 U, (BC ∂ )GL(r,C) and (fj )j∈N of cocycles fj ∈ Z 1 U, (BC ∂ )GL(r,C) such that
f0 = f and, for all j ∈ N,
(8.32) fj+1 = gj+1 fj ,
1
(8.33) fj+1 − 1 ≤ fj − 1,
8
(8.34) ∂fj+1 ≤ Cfj − 12 + 16∂fj fj − 1,
65
(8.35) gj+1 − 1 ≤ f − 1,
64 j
C 33
(8.36) ∂gj+1 ≤ f − 1 + ∂fj .
16 j 32
Since f0 = f , it follows from (8.33) that
1 1
(8.37) fj − 1 ≤ f − 1 = 3j f − 1, j ∈ N.
8j 2
Together with (8.35) this yields
65 1
(8.38) gj − 1 ≤ f − 1, j ∈ N∗ .
64 23j−3
Since f − 1 ≤ 1/64, this in particular implies that, for all j ∈ N∗ , gj − 1 < 1
7
and, therefore,
∞ ∞
gj − 1μ 1 7
log 1 + gj − 1 = − (−1)μ ≤ gj − 1 μ
= gj − 1.
μ=1
μ μ=0
7 6
Together with (8.38) this further implies that

7 1
(8.39) log 1 + gj − 1 ≤ f − 1, j ∈ N∗ .
6 23j−3
Hence, for each N ∈ N,
∞
∞
7 1 4 1
(8.40) log 1 + gj − 1 ≤ f − 1 = f − 1.
6 23j−3 3 23N
j=N +1 j=N +1
23
As f − 1 ≤ 1/64, this in particular implies that

∞ 1
(8.41) log 1 + gj − 1 ≤ .
j=1
48
Moreover, for N ∈ N and M ∈ N∗ , we have

N
+M
M
1 + gj − 1 = 1 + gμ1 − 1 . . . gμκ − 1.
j=N +1 κ=1 N +1≤μ1 <...<μκ ≤N +M
and, similarly,
N
+M
M

1 + (gj − 1) = 1 + gμ1 − 1 . . . gμκ − 1 .
j=N +1 κ=1 N +1≤μ1 <...<μκ ≤N +M
Hence, for N ∈ N and M ∈ N∗ ,

N N
+M +M
g − 1 = 1 + (g − 1) − 1
j j
j=N +1 j=N +1
M

=
(gμ1 − 1) . . . (gμκ − 1)

κ=1 N +1≤μ1 <...<μκ ≤N +M

M
≤ gμ1 − 1 . . . gμκ − 1
κ=1 N +1≤μ1 <...<μκ ≤N +M
N
+M N
+M

= 1 + gj − 1 − 1 = exp log 1 + gj − 1 − 1.
j=N +1 j=N +1
Taking into account (8.41) and the fact that

3 1
ex − 1 ≤ e1/48 x ≤ x for 0 ≤ x ≤ ,
2 48
this implies that
N
+M 3 N
+M
(8.42) g − 1 ≤ log 1 + g − 1 , N ∈ N, M ∈ N∗
j 2 j
j=N +1 j=N +1
and, therefore, by (8.40),

N
+M 2
gj − 1 N ∈ N, M ∈ N∗ .
(8.43) ≤ 23N f − 1,
j=N +1
In particular, as f − 1 ≤ 2−6 ,
N +M

1
(8.44) gj − 1 ≤ 5+3N , N ∈ N, M ∈ N∗ ,
2
j=N +1
and, hence,
N
+M
(8.45) g
j < 2, N ∈ N, M ∈ N∗ .

j=N +1
132
From the last two inequalities we see that, for all N, M ∈ N∗ ,

N +M
+M N
N N 1
gj −
gj ≤ gj − 1
(8.46) gj ≤ 24+3N
.
j=1 j=1 j=1 j=N +1

Denote by C 0 U, (BC 0 )L(r,C) the Banach algebra of families ϕ = {ϕi }i∈I of
continuous functions ϕi : Ui → L(r, C) such that
ϕ := sup ϕi Ui < ∞,
i∈I

and let C U, (BC )
0 0 GL(r,C)
be the group of invertible elements of this Banach

algebra, i.e., the group of all f ∈ C 0 U, (BC0 )L(r,C) such that fi (ζ) ∈ GL(r, C) for
all i ∈ I, ζ ∈ Ui and
sup ϕ−1
i Ui < ∞.
i∈I
∞
Then we see from (8.46) and (8.43) that the infinite product g := j=1 gj converges

in C 0 U, (BC 0 )GL(r,C) , where
(8.47) g − 1 ≤ 2f − 1
and, hence (as |f − 1 ≤ 1/64),
1
(8.48) g − 1 ≤ .
32
Moreover, from (8.32) and (8.3) we see that
(g1 . . . gj ) f = fj , j ∈ N∗ .
By (8.37) this implies that

(g1 . . . gj ) f − 1 ≤ f − 1 , j ∈ N∗ .
23j
Hence
g f = lim (g1 . . . gj ) f = 1.
j→∞

Therefore it remains to prove that g belongs to C 0 U, (BC ∂ )GL(r,C) and satisfies
(8.7) ((8.6) is identical (8.47)). Taking into account that g − 1 < 1 and the
arguments given at the beginning of Section 8.1, for this it is sufficient to prove

that g belongs to C 0 U, (BC ∂ )L(r,C) and satisfies (8.7).
From (8.34) and (8.37) we see that
C 16
∂fj+1 ≤ 6j
f − 12 + 3j ∂fj f − 1, j ∈ N.
2 2
Since f − 1 ≤ 1/64, this implies that
C 1
(8.49) ∂fj+1 ≤ f − 1 + 3j+2 ∂fj , j ∈ N.
26j+6 2
Setting first j = 0 and then j = 1, this gives (recall that f0 = f )
C 1
(8.50) ∂f1 ≤ f − 1 + 2 ∂f
26 2
25
and
(8.51)

C 1 C 1 C 1
∂f2 ≤ 12 f − 1 + 5 ∂f1 ≤ 12 f − 1 + 5 f − 1 + 2 ∂f
2 2 2 2 26 2

1 1 1 C 1
= C 12 + 11 f − 1 + 7 ∂f ≤ 10 f − 1 + 7 ∂f .
2 2 2 2 2
Next we prove by induction that, for all j ∈ N∗ ,
C 1
(8.52) ∂fj+1 ≤ f − 1 + j+6 ∂f .
2j+9 2
For j = 1 this holds true by (8.51). Now let j ∈ N∗ such that (8.52) is already
proved. Then, by (8.49) and (8.52),
C 1
∂f(j+1)+1 ≤ f − 1 + 3(j+1)+2 ∂fj+1
26(j+1)+6 2
C 1 C 1
≤ 6(j+1)+6 f − 1 + 3(j+1)+2 f − 1 + j+6 ∂f
2 2 2j+9 2

1 1 1
= C 6j+12 + 4j+14 f − 1 + 4j+11 ∂f
2 2 2
C 1
≤ (j+1)+9 f − 1 + (j+1)+6 ∂f .
2 2
So (8.52) is proved for all j ∈ N∗ .
As f0 = f , we see from (8.36) that
C 33
(8.53) f − 1 + ∂f .
∂g1 ≤
16 32
Moreover, from (8.36), (8.37), and (8.50) we see
C 33
∂g2 ≤ f1 − 1 + ∂f1
16 32
(8.54)
C 33 C 1 C 33
≤ 7 f − 1 + f − 1 + ∂f ≤ 5 f − 1 + 7 ∂f .
2 32 26 22 2 2
If j ≥ 3, then we see from (8.36), (8.37), and (8.52) that
C 33
∂gj ≤ f − 1 + 5 ∂fj−1
16 j−1 2
C 33 C 1
≤ 3j+1 f − 1 + 5 f − 1 + j+4 ∂f
2 2 2j+7 2
(8.55)
C 1 33 33
= j+5 + 7 f − 1 + j+9 ∂f
2 22j−4 2 2
C 33
≤ f − 1 + j+9 ∂f .
2j+5 2
For N ≥ 3, from the product rule we get the estimate
N N j−1 N
N −1 N −1

∂ g ≤ ∂g g + ∂g g + ∂g g · g
i ,
j 1 j N j j i
j=1 j=2 j=1 j=2 i=1 i=j+1
134
where, by (8.44),
N
N 1

gj ≤ 1 + gj − 1
≤ 1 + 28 ,
j=2 j=2
N −1 N
−1 1

gj ≤ 1 + gj − 1
≤ 1 + 25 ,
j=1 j=1
and
j−1 N
1 1 1

gi ·
gi ≤ 1 + 5 1 + 5+3j ≤ 1 + 6 if j ≥ 2.
2 2 2
i=1 i=j+1
Hence
N N −1
1 1 1
∂ g ≤ 1 + ∂g + 1 + ∂g + 1 + ∂gj
j
28 1
26 N
25 j=2
j=1
∞
1 1 1
≤ 1 + 8 ∂g1 + 1 + 5 ∂g2 + 1 + 5 ∂gj for N ≥ 3.
2 2 2 j=3
Together with (8.53), (8.54), and (8.55) this implies that

N 1 C 33
∂
gj ≤ 1 + 8 f − 1 + 5 ∂f
2 16 2
1

1 C 33
+ 1+ 5 f − 1 + 7 ∂f
2 25 2
∞
1 C 33
+ 1+ 5 f − 1 + ∂f for N ≥ 3,
2 j=3 2j+5 2j+9
and further, taking into account that

∞
1 C 1 C 1 C C C C C
1+ 8 + 1+ 5 + 1+ 5 = + + <
2 16 2 25 2 j=3 2j+5 8 16 64 4
and
∞
1 33 1 33 1 33
1+ 8 + 1+ 5 + 1+ 5
2 25 2 27 2 j=3
2j+9

33 1 1 1 1 1 33 1 99
= 5 1 + 2 + 6 + 7 + 8 + 11 < 5 1 + = < 2,
2 2 2 2 2 2 2 2 64
we obtain that

N C
∂ gj
(8.56) ≤ 4 f − 1 + 2∂f for N ≥ 3.
1
In particular, as f − 1 ≤ 1/64,

N C
gj
(8.57) ∂ ≤ 28 + 2∂f for N ≥ 3.
1
27
Moreover, from the product rule and (8.45) we see that

N +M N N +M −1
+M
∂
gj
≤ ∂gN +1 gj
+ ∂gN +M gj

j=N +1 j=N +2 j=N +1
−1 j−1 N
N +M
+M
+
∂gj
gi · gi

j=N +2 i=N +1 i=j+1
∞
≤4 ∂gj if N, M ∈ N and M ≥ 3.
j=N +1
By (8.55) and taking into account that f − 1 ≤ 2−6 , this implies that
N +M ∞ ∞
Cf − 1 33∂f C ∂f
∂
gj ≤ + j+7 ≤ + j+1
2j+3 2 2j+9 2
(8.58) j=N +1 j=N +1 j=N +1
C ∂f C + ∂f
= N +9 + N +1 < , N, M ∈ N, M ≥ 3.
2 2 2N +1
Again by the product rule,
N +M N

N
N +M

∂ g −∂ g = ∂ g g − 1
j j j j
j=1 j=1 j=1 j=N +1
N N +M
N N +M

≤
∂ g ·
j
gj − 1 +

g ·
j
∂ g
j
j=1 j=N +1 j=1 j=N +1
∗
for all N ∈ N and M ∈ N . If N, M ≥ 3, then, by (8.57), (8.44), (8.45), and (8.58),
this implies that
N +M
N
C 1 C + ∂f C + ∂f
∂ gj − ∂
gj ≤ + 2∂f 5+3N + ≤ .
28 2 2N 2N −1
j=1 j=1
∞
Together with (8.46) this implies that the infinite product g = j=1 gj converges

even in the Banach space C 0 U, (BC ∂ )L(r,C) endowed with the norm (7.4), where,
by (8.56),
C
∂g ≤ f − 1 + 2∂f .
4
By definition (8.31) of C, this completes the proof of Theorem 8.2.
9. Proof of Theorem 3.5

In particular, the cocycle f belongs to Z 1 U, (BC ∂ )Gl(r,C) (Section 8.1), where
even ∂f = 0. Therefore, in view of the second estimate in (3.11), it follows from

Theorem 8.2 and the definition of Γ that there exists g ∈ C 0 U, (BC ∂ )Gl(r,C) such
that
(9.1) g f = 1,
(9.2) g − 1 ≤ 2f − 1,

and
Γ
(9.3) ∂g ≤ f − 1.
16ε
136
By (9.1)
(9.4) fij = gi gj−1 on Ui ∩ Uj , i, j ∈ I.
As fij is holomorphic in Ui ∩ Uj ∩ int X, this implies that

0 = (∂gi gj−1 − gi gj−1 ∂gj gj−1 on Ui ∩ Uj , i, j ∈ I,
i.e.,
gi−1 (∂gi = gj−1 ∂gj on Ui ∩ Uj , i, j ∈ I.
Therefore, we have a well-defined form A ∈ 0 L(r,,C)
(BC0,1 ) (X) such that
(9.5) A= gi−1 ∂gi on Ui , i ∈ I.
Then, on each Ui ,
∂A = −gi−1 (∂gi )gi−1 ∧ ∂gi = −(gi−1 ∂gi ) ∧ gi−1 ∂gi = −A ∧ A.
Hence
∂A = −A ∧ A on D.
Since f − 1 ≤ 1/64, from (9.2) we see that g − 1 ≤ 1/32 and therefore
∞
1 33
(9.6) g −1 ≤ = .
32k 32
k=0
Together with (9.5) and (9.3), this implies that
33 Γ Γ
(9.7) AX ≤ f − 1 ≤ f − 1
32 16ε 10ε
and further, by the first estimate in (3.11),
1
(9.8) AX ≤ .
10 max(T1 , T2 )
Therefore, we can apply Theorem 5.1 and obtain a matrix function U ∈
(BC 0 )GL(r,C) (X) such that ∂U ∈ (BC0,1
0 L(r,C)
) (X),
(9.9) U −1 ∂U = A on X,
(9.10) U − 1X ≤ 2T1 AX .

From (9.10) and (9.8) it follows that
1
(9.11) U − 1X ≤ .
4
By the arguments given at the beginning of Section 8.1, this implies that U even
belongs to the group (BC ∂ )GL(r,C) (X). Therefore, setting
hi = gi U −1 on Ui , i ∈ I,

we obtain an element h := {hi }i∈I in the group C U, (BC ∂ )GL(r,C) . By (9.9) and
0
(9.5), for this element we have

∂hi = ∂gi U −1 − gi U −1 ∂U U −1 = ∂gi U −1 − gi AU −1

= ∂gi U −1 − gi gi−1 ∂gi U −1 = ∂gi U −1 − ∂gi U −1 = 0,

i.e., h even belongs to the group C 0 U, BOGl(r,C) . From (9.4) we see that
hi h−1
j = gi U
−1
U gj−1 = gi gj−1 = fij on Ui ∩ Uj , i, j ∈ I,
i.e., we have (3.12).
29
∞
It remains to prove (3.13). From (9.11) we see that U −1 = j=0 (1 − U )j and
∞
∞
1 4
U −1 − 1X ≤ 1 − U jX ≤ 1 − U X = 1 − U X .
j=1 j=1
4j−1 3
By (9.10) and (9.7), this further implies that

8 Γ
(9.12) U −1 − 1X ≤ T1 AX ≤ T1 f − 1.
3 3ε
Moreover, by definition of hi ,
hi − 1 = gi U −1 − 1 = (gi − 1)(U −1 − 1) + (gi − 1) + (U −1 − 1),
and, therefore
h − 1 ≤ g − 1U −1 − 1X + g − 1 + U −1 − 1X .
Together with (9.2) and (9.12) this implies that
2Γ Γ
h − 1 ≤ T1 f − 12 + 2f − 1 + T1 f − 1,
3ε 3ε
and further, as f − 1 ≤ 1/64,

Γ Γ ΓT1
h − 1 ≤ T1 + 2 + T1 f − 1 ≤ 2 + f − 1.
96ε 3ε 2ε
This completes the proof of Theorem 3.5.
References
[BR] Berndtsson, B., Rosay, J.-P., Quasi-isometric vector bundles and bounded factorization
of holomorphic matrices, Ann. Inst. Fourier, Grenoble 51, 3 (2001), 885-901.
[CG] Cornalba, M., Griffiths, Ph., Analytic cycles and vector bundles on non-compact alge-
braic varieties, Inventiones math. 28 (1975), 1-106.
[GrLi] H. Grauert, I. Lieb, Das Ramirezsche Integral und die Lösung der Gleichung ∂f = α im
Bereich der beschränkten Formen, Rice Univ. Studies 56 (1970), no. 2, 29–50.
[GoLe] Gohberg, I., Leiterer, J., Holomorphic operator funcitons on one variable and applica-
tions, Birkhüser 2009.
[GoR1] Gohberg, I., Rodman, L., Analytic operator valued functions with prescribed local data,
J. Analyse Math. 40 (1981), 90-128.
[GoR2] Gohberg, I., Rodman, L., Analytic matrix functions with prescribed local data, Acta Sci.
Math. (Szeged) 45 (1983), no. 1-4, 189-199.
[Gr] Grauert, H., Analytische Faserungen über holomorph-vollständigen Räumen, Math.
Ann. 135 (1958), 263-273.
[He] G. M. Henkin, Integral representation of functions in strongly pseudoconvex domains
and applications to the ∂-problem (Russian), Mat. Sb. 82 (1970), 300-308; Engl. transl.
in Math. USSR-Sb. 11 (1970), 273–281.
[HeLe] G. Henkin, J. Leiterer, Theorey of functions on complex manifolds, Birkhäuser 1984.
[HeP] Polyakov, P.L., Khenkin, G.M., Integral formulas for the solution of the ∂-equation and
an interpolation problem in analytic polyhedra, (English. Russian original) Trans. Mosc.
Math. Soc. 1991, 135-175; translation from Tr. Mosk. Mat. O.-va 53 (1990), 130-170.
[Le2] Leiterer, J., An estimate for the splitting of holomorphic cocycles. One Variable. To
appear in: Memorial Volume in Honour of Israel Gohberg, OT series, Birkhäuser.
[Li1] Lieb, I., Die Cauchy-Riemannschen Differentialgleichungen auf streng pseudokonvexen
Gebieten, Math. Ann. 190 (1970), 6-44.
[Li2] Lieb, I., Die Cauchy-Riemannschen Differentialgleichungen auf streng pseudokonvexen
Gebieten: Stetige Randwerte, Math. Ann. 199 (1972), 241-256.
[Ov] Øvrelid, N., Integral representation formulas and Lp -estimates for the ∂-equation, Math.
Scand. 29 (1971), 137-160.
138
[R] Range, M., Holomorphic functions and integral representations in several complex vari-
ables, Springer 1986.
[RS] M. Range, Y.-T. Siu, Uniform estimates for the ∂-equation on domains with piecewise
smooth strictly pseudoconvex boundaries, Math. Ann. 206 (1973), 325-354.
Jürgen Leiterer, Institut für Mathematik, Humboldt-Universität zu Berlin,

Rudower Chaussee 25, D-12489 Berlin , Germany
E-mail address: leiterer@mathematik.hu-berlin.de
Volume 550, 2011
A Gysin sequence for manifolds with R-action
Gerardo A. Mendoza
Abstract. We associate an exact sequence involving the cohomology groups

of a pair of differential complexes to any pair (N , T ) where N is a closed
connected smooth manifold and T a real nowhere vanishing smooth vector
field on N that admits an invariant metric. The orbits of T need not be closed.
The sequence is a natural generalization of the classical Gysin sequence (for
circle bundles) in real cohomology.
1. Introduction
In this note we shall associate an exact sequence in certain cohomology to
any pair (N , T ) where N is a smooth compact manifold without boundary and
T a real nowhere vanishing smooth vector field on N that admits an invariant
Riemannian metric. The orbits of T need not be closed. The procedure leads
to the classical Gysin sequence in real cohomology when N is the circle bundle
of a complex line bundle, somewhat more generally, to an exact sequence in real
cohomology whenever the orbits of the action are circles.
A brief discussion of, and comments about, the classical case will place the
results in context. Let B be a connected compact manifold and E → B a complex
vector bundle of rank r ≤ dim B/2 with sphere bundle ρ : SE → B. The Gysin
sequence of SE (see Gysin [2], Milnor and Stasheff [7], Bott and Tu [1], Hatcher
[3]) in integral cohomology,
e· ρ∗ ρ∗
(1.1) · · · → H q−2r (B, Z) −−→ H q (B, Z) −→ H q (SE, Z) −→ H q−2r+1 (B, Z) → · · ·
serves to define the Euler class, e ∈ H 2r (B, Z), of E. This class is also the r-th
Chern class of E. By a well known induction process this definition can be used
to define the total Chern class, c(E) = 1 + c1 (E) + · · · + cr (E), of E. The total
Chern class represents an obstruction to the vector bundle being trivial (although
the vanishing of cj (E) for all j ≥ 1 is not sufficient to ascertain the triviality of
E, see [9]). In the special case where the vector bundle is a line bundle, its Euler
class, e = c1 (E), does determine it; this property is attributable to the fact that
U(1), the unitary group in dimension 1, is abelian. Thus the first Chern class of
2010 Mathematics Subject Classification. Primary 57R19; Secondary 58J10, 55R20, 55M55,
55M65.
Key words and phrases. Gysin sequence, de Rham cohomology, R-actions.
2011
c 0000
c Mathematical
holder)
1
139
140
2 GERARDO A. MENDOZA
E plays, in the case of a line bundle, the dual role of being the element (except
for sign) which makes the sequence (1.1) exact, and of classifier of the line bundle.
Of course to discuss the Euler class B need not be a manifold, nor compact; it is
only required that E → B be of finite type. The more restricted context of this
paragraph fits the assumptions on the rest of the paper with N being SE and T
being the infinitesimal generator of the R-action t → eit p, p ∈ SE.
Denote by F the class of pairs (N , T ) as in the first paragraph:
N is a connected smooth closed manifold and T a real nowhere vanishing
smooth vector field on N that admits an invariant metric.
For a class of examples of this kind of pairs see the end of this introduction.
Let (N , T ) ∈ F . Write Op for the orbit of T through p. Defining p1 ∼ p2
if p1 ∈ Op2 gives a relation of equivalence on N because of the existence of a T -
invariant metric. We will write B for the quotient space and refer to it as the base
space of N (more precisely, of (N , T )); it is a compact Hausdorff space but may
not be a manifold.
Define the following two relations of equivalence on F (see [6]). Say that
(N , T ) is locally equivalent to (N , T ) if there are open covers {Ua }a∈A of N ,
{Ua }a∈A of N by T , resp. T invariant sets, and diffeomorphisms ha : Ua → Ua
such that dha (T ) = T and
∀a, b ∈ A ∀p ∈ Ua ∩ Ub : ha ◦ h−1
b (p) ∈ O p .
Also say that (N , T ) and (N , T ) are globally equivalent if there is a diffeomor-
phism h : N → N such that dh(T ) = T . Fix (N , T ) ∈ F , let FN ,T be the set of
elements of F which are locally equivalent to (N , T ). Global equivalence defines
a relation of equivalence within this set. Let B be the base space of N . The base
space of each element of FN ,T is homeomorphic to B.
Theorem 1.1 (Theorem 3.9 of [6]). The set of global equivalence classes of
elements of FN ,T is in one to one correspondence with the elements of H 2 (B; Zd )
for some d.
The number d is defined in (7.1). Thus the elements of H 2 (B; Zd ) are like first
Chern classes in their role as classifiers of elements of FN ,T . They are relative
Chern classes because elements within a local equivalence class are being compared
to an arbitrarily fixed element in the class, in contrast with line bundles and their
regular Chern classes where the comparison is with the trivial line bundle.
We shall define an Euler class for (N , T ) ∈ F as an element of the second
cohomology group of certain subcomplex, see (2.3), of the de Rham complex of
N . This subcomplex reduces to the de Rham complex of the base when N is a
circle bundle over a smooth manifold. The defining property of e is the following
theorem, the central result of this paper.
Theorem 1.2. Let (N , T ) ∈ F , let B be the base space of N and let ρ : N → B
be the projection map. There is an exact sequence
q−2 e∧· q ρ∗
q ρ∗
q−1
(1.2) · · · → HdR (B) −−→ HdR (B) −→ HdR (N ) −→ HdR (B) → · · · .
q
The spaces HdR (N ) are the de Rham cohomology groups of N , the groups
q
HdR (B)are the cohomology groups of the complex (2.3) defined in the next section,
the maps ρ∗ and ρ∗ are defined in Section 5, and the theorem is proved in Section 6.
A GYSIN SEQUENCE 141
3
The theorem is somewhat unsatisfactory in that it appears to be, to some

extent, a theorem in the scalar realm (however, see Theorem 7.1) that does not
reflect the fact (notwithstanding that the present context is in the realm of real
cohomology) that the classifier group established in Theorem 1.1 is cohomology
with values in Zd rather than Z. Note also that since the coefficients are real, any
nonzero multiple of e also works.
We shall in fact prove what appears to be a more general result than Theo-
rem 1.2, motivated by the following considerations in the context of the classical
case of a manifold B and a complex line bundle E → B with a given Hermitian
metric and circle bundle SE. The space SE is an m-fold covering space of the circle
bundle of the tensor product E ⊗m with covering map ℘m : SE → SE ⊗m ,

p ⊗ ... ⊗ p if m ≥ 0
℘m (p) = ∗ ∗
p ⊗ . . . ⊗ p if m < 0,
(|m| factors in either case) with p∗ ∈ Eρ(p)∗

such that p∗ , p = 1. The space
∞ ⊗m ∗ ∞ ⊗m
C (SE ) is isomorphic to ℘m C (SE ). This and the property ℘m (eit p) =
eimt ℘m (p) allows us to view the space of smooth functions on SE ⊗m as the space
{φ ∈ C ∞ (SE) : t → φ(eit p) is 2π/m-periodic for each p ∈ SE}.

q
Letting at (p) = eit p we have, more generally, that C ∞ (SE ⊗m ; SE ⊗m ) is isomor-
phic to
q
(1.3) {φ ∈ C ∞ (SE, E) : t → a∗t φ is 2π/m-periodic}.
We also note that if ψ is a section of E ⊗m , then it determines (and is determined

by) a liner map on each fiber of E −⊗m , the dual bundle, so ℘∗−m ψ is defined as a
function on SE and satisfies
(℘∗−m ψ)(eit p) = e−imt (℘∗−m ψ)(p)
equivalently, solves
T (℘∗−m ψ) − im℘∗−m ψ = 0.
More generally, smooth q-forms on B with values in E ⊗m can be viewed as elements
of the space
q
(1.4) {φ ∈ C ∞ (SE; ρ∗ B) : LT φ − imφ = 0}.
On the plus side, these observations, certainly familiar to experts in algebraic ge-
ometry, permit the analysis of objects such as de Rham cohomology to be carried
out in SE rather than SE ⊗m . We take advantage of these observations, especially
(1.3), to define the analogue of the de Rham complex for tensor products in our
more general context, see (2.2). On the minus side, in what pertains this paper
there is no real gain because, as a consequence of exactness of the Gysin sequence
(or merely by the fact that the Euler class of E m is m times that of E), the de
Rham cohomology of SE ⊗m is identical to that of SE. Also in our more general
context the cohomology groups will be the same (Theorem 2.2).
We end this introduction with an example of an element of F .

142
Example 1.3. Let S 2n+1 be the unit sphere in Cn+1 , let τ 1 , . . . , τ n+1 be
nonzero real numbers. The vector field

n+1
j ∂ j ∂
T =i j
τ z −z
j=1
∂z j ∂z j
is tangent to the sphere: T |z|2 = 0. To verify that the pair (S 2n+1 , T ) belongs to
F we only need to observe that the standard Riemannian metric of the sphere is
T -invariant.
Remark 1.4. On can show that if (N , T ) ∈ F , then there is N ∈ N and a
smooth embedding F : N → S 2N +1 ⊂ CN +1 such that
F∗ T = T
with T as in the example. The assertion, whose proof will appear elsewhere, can
be viewed as an analogue of the statement that every complex line bundle over a
compact manifold B can be realized as the pullback of the canonical line bundle of
complex projective space of some dimension.
2. Set up
Throughout the rest of the paper we work with a fixed element (N , T ) of F .
We write at for the one-parameter group of diffeomorphisms defined by T and Op
for the orbit of T through p. Given any pair (p, p ) ∈ N × N define p ∼ p if
p ∈ Op . As was already mentioned, ∼ is a relation of equivalence because of the
of the assumption on the existence of a T -invariant metric. The space B = N /∼ is
a compact Hausdorff space, not necessarily a manifold. We let
ρ:N →B
be the quotient map.
Let
q q−1
(2.1) iT : N → N
be interiormultiplication by T . If H is the kernel of iT in CT ∗ N , then the kernel
∗
q ∗
of (2.1) is H . Let LT denote Lie derivative with respect to T . For any number
τ and nonnegative integers q and k or k = ∞ define, consistently with (1.4),
q q
(2.2) C k (B; E τ ⊗ B) = {φ ∈ C k (N ; H∗ ) : LT φ − iτ φ = 0}.
The symbol E is a place holder which becomes a Hermitian line bundle with circle
bundle N if the action determined by T make N into a principal S 1 -bundle. The
symbol E τ is a reference to a virtual (not in the sense of K-theory) line bundle
which becomes the m-th tensor power of a line bundle when τ = m and E is an
actual line bundle. Thenumber τ can be any complex number but must be a real
q
number if C k (B; E τ ⊗ B) = 0; it need not be an integer, but the spaces are
nonzero only for τ in a countable subset of R.
If τ , q, and k are all zero, the space (2.2) is canonically isomorphic to
qC(B). In
general when τ = 0 we will use the notation C k (B) if q = 0 and C k (B; B) when
q > 0 for the space (2.2).
q
The space C(B; E τ ⊗ B) is a C(B)-module. It is not necessarily a finitely
generated projective module, so by a theorem of Swan [12] it need not be the space
of continuous sections of a vector bundle over B.
5
Since LT commutes with differentiation and iT d = diT −LT , there is a complex

q q+1
· · · → C ∞ (B; B) −−→ C ∞ (B;
dB
(2.3) B) → · · ·
∞
q
where dB is the restriction of d to C (B; B). We shall denote the cohomology
q
groups of (2.3) by HdR (B) even though (2.3) need not be the de Rham complex,
nor its cohomology groups isomorphic to the real cohomology groups of B. The
q
groups HdR (B) are the ones appearing in (1.2)
The following proposition links, in a special case, the cohomology groups of
(2.3) with the real cohomology groups of B.
Theorem 2.1. If the orbits of T are circles then the cohomology groups of the
complex (2.3) are canonically isomorphic to the Čech cohomology groups of B with
real coefficients.
This reduces to the de Rham Theorem [10] (see also [11]) when N is the circle
bundle of a complex line bundle (in which case B is a smooth compact manifold
and de Rham’s theorem makes sense). The proofs of this proposition, the next, and
subsequent theorem in this sections will be given later.
Let at be the one-parameter group generated by T . Following (1.3), define, for
any τ , q and k as above (but now τ = 0),
q q
C k (N τ ; N τ ) = {φ ∈ C k (N ; N ) : t → a∗t φ is 2π/τ -periodic}
The notation N τ is a reference to the circle bundle of a tensor power of a complex
line bundle.
Using that a∗t commutes with d we see that there is a complex
q q+1 τ
· · · → C ∞ (N τ ; N τ ) −→ C ∞ (N τ ;
dτ
(2.4) N ) → ···
q
where dτ is the restriction of d to C ∞ (N τ ; N τ ). We will write HdRq
(N τ ) for the
cohomology groups of (2.4). These are not a priori the real cohomology groups of
N (nor of N τ , which is a virtual object anyway). That they are, is a consequence
of the following proposition and de Rham’s theorem.
Proposition 2.2. The cohomology groups of the complex (2.4) are canonically
isomorphic to those of the de Rham complex of N .
Let θ be any smooth one-form on N with the properties

(2.5) θ, T = 1 and LT θ = 0.
q
Then dθ ∈ C ∞ (B, B): LT dθ = d(LT θ) = 0. Since dθ is obviously dB -closed, it
defines an element eτ = [τ /2π dθ] ∈ HdR 2
(B). The factor τ /2π guarantees consis-
tency with the classical case but is otherwise irrelevant.
In view of Proposition 2.2, the following theorem is equivalent to Theorem 1.2.
Theorem 2.3. Suppose that τ = 0. There is an exact sequence
q e ∧· q+2 ρ∗ q+2 ρ∗q+1
(2.6) · · · → HdR (B) −−τ−→ HdR (B) −→ HdR (N τ ) −→ HdR (B) → · · ·
The homomorphisms ρ∗ , ρ∗ are defined in Section 5. When N is the circle bun-
dle of a complex line bundle E → B this is the Gysin sequence in real cohomology
for E τ where τ is an integer. Note that the specific value of τ (which is nonzero)
does not play an essential role.
144
3. Decompositions
Let g be an arbitrarily chosen T -invariant Riemannian metric. Then g(T , T )
is smooth, positive and T g(T , T ) = 0, so we may assume g(T , T ) = 1. Assuming
this, the 1-form θ = T g satisfies (2.5).
Define q q ∗
Πθ : N → H
by
Πθ φ = φ − θ ∧ iT φ.
q ∗ q−1 q
Since iT (θ) = 1, Πθ is the projection on H with kernel θ ∧ N . If φ ∈ N,
then
φ = Π θ φ + θ ∧ iT φ
gives the decomposition
q q ∗ q−1 ∗
N = H ⊕ (θ ∧ H ).
∞
q ∗
If φ ∈ C (N ; H ), then
dφ = Πθ dφ + θ ∧ LT φ.
Define q q+1 ∗
dθ : C ∞ (N ; H∗ ) → C ∞ (N ; H ), dθ = Πθ ◦ d.
∞
q ∗ ∞
q−1 ∗
Thus, if φ ∈ C (N ; H ) and φ ∈ C (N ;
0 1
H ), then
d(φ0 + θ ∧ φ1 ) = dθ φ0 + dθ ∧ φ1 + θ ∧ (LT φ0 − dθ φ1 ).
Define Θφ = dθ ∧ φ. Since iT dθ = 0,
q−1 ∗ q+1 ∗
Θ : C ∞ (N ; H ) → C ∞ (N ; H ), d◦Θ=Θ◦d
and we may view d as the operator
q q+1 ∗
C ∞ (N ; H∗ ) C ∞ (N ; H )
dθ Θ
(3.1) d= : ⊕ → ⊕
LT −dθ q−1 ∗ q
C ∞ (N ; H ) C ∞ (N ; H∗ ).
2
The operators dθ do not form a complex unless qdθ = 0, since dθ + ΘLT = 0.
∞
However the restrictions of dθ to the spaces C (B; B) do form a complex and qare
in fact identical to the operators dB already introduced. Indeed, if φ ∈ C ∞ (B; B)
then by definition iT φ = 0 and LT φ = 0. Using φ0 = Πθ φ = φ and φ1 = iT φ in
(3.1) we get dφ = dθ φ. q
Let m be the Riemannian density on N . The spaces L2 (N ; N ) is defined
using the Riemannian metric and the density m. The Laplacian of the de Rham
complex is

Δθ − L2T + ΘΘ∗ dθ Θ − Θdθ
(3.2) Δ=
Θ dθ − dθ Θ Δθ − L2T + Θ∗ Θ.
where Δθ = dθ dθ + dθ dθ . Since Δ is elliptic, so is Δθ − L2T in each degree.
Let q
kerq Δθ = {φ ∈ L2 (N ; N ) : Δθ φ = 0}
q
This is a closed subspace of L2 (N ; N ), therefore a Hilbert space on its own right.
Let
Dom(LT ) = {φ ∈ kerq Δθ : LT φ ∈ L2 (N ; E)}.
7
Since Δθ commutes with LT , LT Dom(LT ) ⊂ kerq (δθ ) and we have that

(3.3) −iL0T : Dom(LT ) ⊂ kerq Δθ → kerq Δθ ,
the restriction of LT to Dom(LT ), is an unbounded closed operator. The ellipticity
of Δθ − L2T gives:
Theorem 3.1. The operator operator (3.3) is selfadjoint and Fredholm with
discrete spectrum.
Thus kerq Δθ decomposes as a direct sum

kerq Δθ = Eτq
τ ∈spec(−iL0T )
4. Fourier series
q
Suppose τ = 0. For arbitrary φ ∈ C ∞ (N , N ) define

2π/τ
τ
(4.1) (πτ,m φ)(p) = a∗s φ(as p) e−imτ s ds.
2π 0
q
This gives a smooth section of N . To see this, let a : R × N → N be the action
defined by T , a(t, p) = at p, let π : R × N → N be the canonical projection, and
let χ : R → R be the characteristic function of [0, 2π/τ ], regarded as a function on
R × N . Then
πτ,m φ = π∗ χa∗ φe−imτ t dt .
An analysis of the behavior of the wavefront set under pull-back and push-forward,
see Hörmander [5], gives that πτ,m φis smooth if φ is.
q
Suppose now that φ ∈ C ∞ (N τ , N τ ) and let φm = πτ,m φ. Then
(4.2) a∗t (φm (at p)) = φm (p)eiτ mt .
Indeed, using the the definition of πτ,m , the group property of a and a change of
variables we have that (2π/2π)a∗t (φm (at p)) is equal to

2π/τ
2π/τ

∗ ∗
−imτ s

at as φ(as p) (at p) e ds = a∗s+t φ(as+t p) e−imτ s ds
0 0

2π/τ
= a∗s φ(as p) e−imτ s ds eiτ mt .
0
Consequently
a∗t (φ(at p)) = φm (p)eimτ t
m∈Z
for each p ∈ N with convergence in C ∞ (R). In particular

φ= φm
m∈Z
with pointwise convergence of the series, in fact with convergence in C ∞ . It also

follows that
(4.3) LT φm = imτ φm .
Finally note that
(4.4) iT (φm ) = (iT φ)m , Θφm = (Θφ)m , Πθ φm = (Πθ φ)m , dφm = (dφ)m .
146
The first two properties are immediate, the third is a consequence of the previous
two. The last follows from (4.2), da∗t = a ∗
t d and uniqueness of the Fourier coeffi-
q ∗
cients. In particular, if φ is a section of H , then so is φm . Thus the Fourier
series representation is compatible with the decomposition 3.1 of d.
5. Homomorphisms
We will first remove τ from the picture. Let
∞
q q
C(0) (N ; N0 ) = {φ ∈ C ∞ (N ; N ) : LT φ = 0}
Since d commutes with LT , we have (yet another) complex,
∞
q d0 ∞
q+1
(5.1) · · · → C(0) (N ; N ) −→ C(0) (N ; N) → ··· ,
∞
q
where d0 is d restricted to C(0) (N ; N ). This is a subcomplex of (2.4) for every
q
τ > 0 as well as of the de Rham complex of N . We will write HdR,0 (N ) for its
cohomology groups. q
The restriction to C ∞ (N τ ; N τ ) of the map πτ,m in (4.1) satisfies
LT ◦ πτ,m = iτ m πτ,m
(see (4.3)), so with m = 0 we have
q q
πτ,0 : C ∞ (N τ ; ∞
N τ ) → C(0) (N ; N ).
This defines a chain map from the complex (2.4) to the complex (5.1).
Lemma 5.1. The cohomology groups of the complex (5.1) are isomorphic to
those of the complex (2.4).
Proof. We show that there is a homotopy from the complex (2.4) to the
subcomplex (5.1). Note that Πθ and iT preserve periodicity because they commute
∗ ∞ q
with at . Using (4.4) we see that if φ ∈ C (N ; N τ ) then in the representation
τ
φ = φ0 + θ ∧ φ1 with φ0 = Πθ φ and φ1 = iT φ, both t → a∗t φ0 and t → a∗t φ1 are

2π/τ -periodic. Let
φ0 = φ0m , φ1 = φ1m
m m
be their Fourier series expansions. Thus πτ,0 φ = φ10 + θ ∧ φ10 . Defining
1
hqτ φ = φ1
iτ m m
m=0
we get a homomorphism
q q−1 τ
hqτ : C ∞ (N τ ;
N τ ) → C ∞ (N τ ; N ).
∞
q−1 ∗
whose image is actually contained in C (N ; H ). It is easily verified that
LT hqτ φ = hqτ LT φ = φ1 − φ10 , hq+1
τ dθ = dθ hqτ .
Using (3.1) we have

dθ Θ hqτ φ d hq φ d hq φ
dhqτ φ = = θ τq = 1θ τ 1 ,
LT −dθ 0 L τ hτ φ φ − φ0
whereas 0
hq+1 (LT φ0 − dθ φ1 ) φ − φ00 − dθ hqτ φ1 )
hq+1
τ dφ = τ
= .
0 0
9
Adding these two formulas we obtain

(hq+1
τ d + dhqτ ) = I − πτ,0 .
q q
The induced homomorphism πτ,0 : HdR (N τ ) → HdR,0 (N ) has as inverse the map
induced by the inclusion
q q
ι : C ∞ (N ; N ) → C(0)
∞
(N τ ; N τ ).
This completes the proof of the lemma.
Consequently we have isomorphisms

q q
πτ,0 : HdR (N τ ) → HdR,0 (N ).
q q
Thus we may, and shall, use the groups HdR,0 (N ) in place of the groups HdR (N τ ).
q q
The space C ∞ (B; B) is a subspace of C(0)∞
(N ; N ) and the inclusion map,
which we shall denote by
q q
ρ∗ : C ∞ (B; B) → C(0) ∞
(N ; N )
satisfies d0 ρ∗ = ρ∗ dB . So there is an induced map in cohomology:
ρ∗ : HdR
q q
(B) → HdR,0 (N ).
∞
q
If φ ∈ C(0) (N ; N ), then φ1 = iT φ obviously satisfies iT φ1 = 0 and LT φ1 = 0.
q−1
So φ1 is an element of C ∞ (B; B). Since
∞
q d ∞
q+1
C(0) (N , N ) −−−0−→ C(0) (N ; N)
⏐ ⏐
⏐
iT
⏐i
T
q−1 d q
C ∞ (B; B) −−−B−→ C ∞ (B; B)
commutes, there is a map
q
q q
q−1
(5.2) ρ∗ : HdR,0 (N ; N ) → HdR (B, B)
induced by iT . q+2
∞
Finally, since LT dθ = 0 and iT dθ = 0, dθ ∧ φ ∈ C(0) (B; B) whenever
∞
q
φ ∈ C(0) (B; B). Since Θ ◦ dB = dB ◦ Θ, there is an induced map in cohomology:
q e ∧· τ q+2
HdR (B) [φ] −τ→ [ dθ ∧ φ] ∈ HdR (B).
2π
6. Exactness
Exactness of
q−2 e ∧· q q ρ∗
HdR (B) −−τ−→ HdR (B) −→ HdR,0 (N ).
q−2
Let ψ ∈ C ∞ (B; B)

q−1
represent an element of HdR (B). Then eτ [ψ] is represented
∞ q ∗
by π dθ ∧
τ
ψ ∈ C (B, B) and ρ (eτ [ψ]) is represented by the same form but in
∞ q
C(0) (N ; N ). But this form is d0 -exact:
τ τ ∞
q−1
dθ ∧ ψ = d0 (θ ∧ ψ), θ ∧ ψ ∈ C(0) (N , N ).
2π 2π
So ρ∗ (eτ [ψ]) = 0.
148
q
Now suppose φ ∈ C ∞ (B; B) represents an element in HdR q
(B) and ρ∗ [φ] = 0.
∞
q−1
Thus φ1 = iφ = 0, LT ψ = 0, dθ φ1 = 0, and there exists χ ∈ C(0) (N , N ) such
that with ψ 0 = Πθ ψ, ψ 1 = iT ψ likewise we have
0 0
φ dθ Θ ψ
φ= = dψ =
0 LT −dθ ψ 1
So φ0 = dθ ψ 0 + dθ ∧ ψ 1 , 0 = dθ ψ 1 . Since dθ ψ 0 = dψ 0 = d0 ψ 0 , the class of φ in
0
HdR,0 (N ) is equal to that of φ − dψ 0 . We may thus assume that iT φ = 0 to begin
with. So φ = dθ ∧ ψ 1 . From dφ = 0 get dψ 1 = 0, and we conclude that φ is in the
image of eτ ∧ ·.
Exactness of
q ρ∗ q ρ∗ q−1
HdR (B) −→ HdR,0 (N ) −→ HdR (B).
∞
q
Suppose ψ ∈ C (B; B) is dB -closed. Then it is d0 (or just d)-closed as an
∞ q
element of C(0) (N ; N ). Since iT ψ = 0, ρ∗ ρ∗ [ψ] = [iT ψ] = 0.
q
Now suppose [φ] ∈ HdR,0 (N ) and ρ∗ [φ] = 0, that is,
q−2
iT φ = dB χ, χ ∈ C ∞ (B; B).
The form φ is cohomologous in the complex (5.1) to ψ = φ + d0 (θ ∧ χ). But
0 0 0
φ dθ Θ 0 φ dθ ∧ χ φ + dθ ∧ χ
φ + d0 (θ ∧ χ) = 1 + = 1 + = .
φ LT −dθ χ φ dθ χ 0
q
Since φ0 + dθ ∧ χ ∈ C ∞ (B; B), [φ] = ρ∗ [φ0 + dθ ∧ φ].
Exactness of
q ρ∗ q−1 e ∧· q+1
HdR,0 (N ) −→ HdR (B) −−τ−→ HdR (B).
q ∞
q
Suppose [ψ] ∈ HdR,0 (N ). Thus ψ ∈ C (N ; N ), LT ψ = 0, and dψ = 0. We
show that dθ ∧ ψ 1 is exact in the complex (2.3). With ψ 0 , ψ 1 as usual we have

dθ ψ 1 + dθ ∧ ψ 1 0
=
−dψ 1 0
The top line gives dθ ∧ ψ 1 = −dθ ψ 0 .
q−1
Now suppose [φ] ∈ HdR (B) and dθ ∧ φ = 0. Then ψ = θ ∧ φ is a d0 -closed
∞ q
element of C(0) (N ; N ) such that iT ψ = φ.
This completes the proof of exactness of the sequence (2.6).
q
7. The groups HdR (B)
q
We shall now analyze the cohomology groups HdR (B) of the complex (2.3) from
two viewpoints. In the first we will exhibit them (for q ≥ d) as the Čech cohomology
with coefficients in a suitable sheaf. The arguments are essentially those used to
relate, for example, Čech and Rham cohomology. The sheaf in question reduces to
the sheaf of locally constant real-valued functions when the orbits of T are circles,
q
thus proving Theorem 2.1, namely that HdR (B) is isomorphic to H q (B, R). In the
second characterization we present the groups, for any q, as spaces of harmonic
forms for a Laplace-like operator
Let g be some T -invariant metric. Since LT g = 0, {at }, the one-parameter
group of diffeomorphisms generated by T consists of isometries of g. The full group
of isometries is a compact Lie group (see Myers and Steenrod [8], Helgason [4])
11
because N is compact, and {at : t ∈ R} is a subgroup. Let G be its closure, a

compact abelian Lie group. The closure of Op is the orbit of G through p, thus
an embedded submanifold of N diffeomorphic to a torus (whose dimension may
change form point to point, however). If x ∈ B, denote by Fx the set ρ−1 (x). Thus
Fx = Op for some p ∈ N . Define
(7.1) d = max{dim Fx : x ∈ B}.
∗
Let HU denote the part of H∗ over U ⊂ N . The family whose elements are
q q ∗
C ∞ (V ; V ) = {φ ∈ C ∞ (U ; HU ) : LT φ = 0}, V ⊂ B open, U = ρ−1 (V )
together with the canonical restriction maps forms a presheaf whose associated
sheaf we denote by E q . The operator dB induces a sheaf map dB : E q → E q+1 . If
d = 1 and q = 0 then Z q is the sheaf of locally constant real-valued functions on B
The family whose elements are
q
Z q (V ) = {φ ∈ C ∞ (V, V ) : dB φ = 0}, V ⊂ B open
with its restriction maps is also a presheaf. The associated sheaf, Z q , is a subsheaf
of E q containing the image of dB .
q
Theorem 7.1. If q − p ≥ d − 1 then HdR (B) is isomorphic to Ȟ p (B; Z q−p ).
q
Thus if d = 1, then HdR (B) ≈ Ȟ (B; R).
q
The proof requires an analysis of the extent to which surjectivity of

(7.2) dB : E q−1 → Z q
holds.
Fix x0 ∈ B arbitrarily. As just discussed, Fx0 is a smooth embedded subman-
ifold of N . The surjectivity of (7.2) is linked to the vanishing of the cohomology
groups in large degree of the complex
q ∗ q+1 ∗
· · · → C ∞ (Fx0 ; HF → C ∞ (Fx0 ;
d
(7.3) x0
)− HFx0 ) → . . . ,
We will write H ∗ (Fx0 ; HF ∗
x0
) for the cohomology groups of this complex.
Pick a point p0 ∈ Fx0 arbitrarily. We may then view Fx0 as an actual torus
with identity element p0 and invariant metric induced by g. Identifying the de
Rham cohomology groups of Fx0 with the spaces of harmonic forms in the various
degrees and using that the vector field T along Fx0 is translation invariant one gets
∗ q
that H q (Fx0 ; HF x0
) can be identified with the kernel of iT on HdR (Fx0 ). We thus
have
∗ q
dim H q (Fx0 ; HF x0
) = dim HdR (Fx0 ) − 1.
∗
Consequently H q (Fx0 ; HF x0
) = 0 if q ≥ dim Fx0 .
Let N Fx0 ⊂ TFx0 N be the orthogonal bundle to T Fx0 according to the metric.
Define
Bε = {v ∈ N Fx0 : g(v, v) < ε2 }
and pick ε > 0 so small that exp : Bε → N is a diffeomorphism onto U = exp(Bε ).
The latter, an open subset of N , is closed under orbits of T because as a consequence
of g being T -invariant, at ◦exp = exp ◦dat . By the definition of the quotient topology
and map, V = ρ(U ) is an open subset of B and ρ−1 (V ) = U .
150
Lemma 7.2. Let x0 ∈ B, Fx0 = ρ−1 (x0 ), and pick V ⊂ B be as just described.
The cohomology of the complex
q q+1
· · · → C ∞ (V ; V ) −−→ C ∞ (V ;
dB
(7.4) V) → ···
is isomorphic to that of the complex (7.3). In particular, these cohomology groups
are zero in all degrees q ≥ dim Fx0 .
As a corollary we obtain the following version of the Poincaré Lemma:
Proposition 7.3. Let d = max{dim Fx : x ∈ B}. The map dB : E q−1 → Z q
is surjective for q ≥ d.
q
Proof of Lemma 7.2. Write HdR (V ) for the q-th cohomology group of the
complex (7.4). Let ι : Fx0 → U be the inclusion map and ℘ : U → Fx0 the
composition
exp−1 π
U −−−−→ Bε −−−→ Fx0 .
Then ℘ commutes with iT , so there is an induced map
(7.5) ℘∗ : H q (Fx0 ; HF
∗ q
x0 ) → HdR (V )
in each degree. Since ℘ ◦ ι = I, ℘∗ is an isomorphism onto its image.

Let κ−∞ : U → U be the map ι ◦ ℘. We will exhibit a homotopy from the
identity map of the complex (7.4) to the cochain map given by
q q
(7.6) κ∗−∞ : C ∞ (V ; V ) → C ∞ (V ; V ),
The existence of this homotopy implies that (7.5) is also surjective. In constructing
the homotopy we closely follow the proof of the Poincaré Lemma in [13].
Let R be the radial vector field of N Fx0 ; its integral curve through v is s → es v.
Since at is an isometry, it sends geodesics to geodesics and so exp ◦dat = at ◦ exp.
Thus
(7.7) at exp(es v) = exp(es dat v).
This says that if X = d exp R, a well defined smooth vector field on U , then
dat X = X. Write κs for the flow of X in U . This is defined for every p ∈ U and
every s < δp where δp > 0. In particular κs maps U into U if s ≤ 0. q
that at ◦κs = κs ◦at . As a consequence, if φ ∈ C ∞ (V ; V )
It follows from (7.7)
q
then also κ∗s φ ∈ C ∞ (V ; V ) for every s ≤ 0.
The family q q
κ∗s : C ∞ (V ; V ) → C ∞ (V ; V )
converges pointwise as s → −∞ to the map (7.6). To see this, pick local coordinates
x1 . . . , xn−d0 , y 1 , . . . , y d0 (n = dim N , d0 = dim Fx0 ) near an arbitrary p ∈ Fx0
such that, near p0 , the xj are defining functions for Fx0 , the vector fields ∂xj
form an orthonormal basis of N Fx0 and d exp(∂xj |v ) = ∂xj |exp(v) if v ∈ Bε . Then

X = j xj ∂xj , κs (x, y) = (es x, y), and so if

φ= φIJ (x, y)dxI ∧ dy J
|I|+|J|=q
then
(κ∗s φ)(x, y) = es|I| φIJ (es x, y) dxI ∧ dy J .
|I|+|J|=q
13
Suppose ι∗ φ = 0. Then κ∗−∞ φ = 0 in which case one has κ∗s φ = O(es ) as

s → −∞. This estimate implies that the integral

0
αq (φ)(p) = κ∗s (φ(κs p)) ds, p ∈ U
−∞
is finite and that the resulting form αq (φ)is smooth on U . Furthermore, αq+1 dφ is
also defined because ι∗ dφ = dι∗ φ = 0. In addition,
(7.8) αq+1 dφ = dαq φ if ι∗ φ = 0
using κ∗s dφ = dκ∗s φ. q
Suppose now that φ is a general element of C ∞ (V ; V ). Obviously ι∗ LX φ
vanishes and
d ∗
κ∗s (LX φ)(p) = κ (φ(κs p)).
ds s
So αq (LX φ) is defined and

0
d ∗
αq (LX φ)(p) = κs (φ(κs p)) ds = φ(p) − (κ∗−∞ φ)(p).
−∞ ds
This gives
dαq−1 iX φ + αq iX dφ = I − κ∗−∞
∗
using LX = diX + iX d, ι iX φ = 0, and (7.8). Thus defining hq = αq−1 ◦ iX we have
d ◦ hq + hq+1 ◦ d = I − κ∗−∞ .
q q
The map κ∗−∞ vanishes on C ∞ (V ; V ) → C ∞ (Fx0 ; Fx0 ) when q ≥ dim Fx0 for
the simple reason that the target space in
q q
ι∗ : C ∞ (V ; V ) → C ∞ (Fx0 ; Fx0 )
is zero when q ≥ dim Fx0 .
Thus there is an exact sequence
ι d
0 → Z q−1 − B
→ E q−1 −−→ Zq →0
in each degree q ≥ d. Therefore for such q we have the long exact sequence
(7.9)
d
B
0 → Γ(B, Z q−1 ) → Γ(B, E q−1 ) −−→ Γ(B, Z q ) → Ȟ 1 (B, Z q−1 ) → Ȟ 1 (B, E q−1 ) →
· · · → Ȟ p (B, E q−1 ) → Ȟ p (B, Z q ) → Ȟ p+1 (B, Z q−1 ) → Ȟ p+1 (B, E q−1 ) → . . .
in Čech cohomology.
Lemma 7.4. The sheaf E q is fine.
This is a consequence of the existence of partitions of unity in C ∞ (B). The
following argument proving the existence of such partitions is taken from [6]. Define
A : G × N → N by A(h, p) = h(p). This is a smooth map. Let also π : G × N → N
be the canonical projection. Finally, let m be the Haar measure on G. Define
av : C ∞ (M) → C ∞ (M) by
avf = π∗ (A∗f m)
Again an analysis of wavefront sets gives immediately that av dos indeed send
smooth functions to smooth functions, and since avf is constant on orbits of G, av
in fact maps into C ∞ (B).
152
Now, if V = {Va }a∈A is an open cover of B and {χb }b∈B is a smooth partition
of unity subordinate to the open cover {ρ−1 (Va )} of N (so χb ∈ C ∞ (N ) for all
b), then the family {avχb }b∈B is another such partition, but now with functions in
C ∞ (B). Interpreting these functions as functions on B, they form a partition of
unity subordinate to the open cover V of B: given any open cover V = {Va }a∈A of
B there is a partition of unity {χb }b∈B with χb ∈ C ∞ (B) for all b ∈ B subordinate
to the given cover.
Since E q is a fine sheaf, Ȟ p (B, E q ) = 0 for p > 0. As in the standard case we
deduce from the long exact sequence that
Ȟ 1 (B, Z q−1 ) ≈ Γ(B, Z q )/dB Γ(B, E q−1 ), q ≥ d,
that is
q
Ȟ 1 (B, Z q−1 ) ≈ HdR (B) if q ≥ d
and
Ȟ p (B, Z q ) ≈ Ȟ p+1 (B, Z q−1 ) for p > 0 and q ≥ d.
from which we reach the thesis of the theorem.
We now turn to a purely analytic description of the groups cohomology groups
of the complex (2.3)
Proposition 7.5. The cohomology group of the qcomplex (2.3) in degree q is
isomorphic to the kernel, H q (B), of Δθ in C ∞ (B; B). The spaces H q (B) are
finite-dimensional.
Proof. The proof is similar to that of the analogous q statement in Hodge
theory. Since Δθ − L2T is elliptic, its kernel in C ∞ (N ; H∗ ), namely H q (B),
is finite dimensional.
Now, H q (B) is in fact the kernel of Δ2θ − L2T . Clearly H q (B) ⊂ ker(Δ2θ − L2T ).
On the other hand, if φ ∈ ker(Δ2θ − L2T ), then

0 = (Δθ − L2T )φ, φ = dθ φ2 + dθ φ2 + LT φ2 .
hence in particular LT φ = 0, so φ ∈ H q (B).
Let
q
ΠqB : L2 (N ; H∗ ) → H q (B)
be the orthogonal
q projection. Then LT ◦ ΠqB =
q0, while also, trivially, q
q ∗ ΠB ◦ LT = 0
∞ ∗
on C (B; B). For each q there is G : L (N ; H ) → L (N ; H ), a selfadjoint
2 2
pseudodifferential operator of order −2, such that G(Δθ − L2T ) = (Δθ − L2T )G =
I − ΠqB . Since LT commutes
q with ΠqB and with (Δθ − L2T ), G commutes with LT .
∞
Thus G maps C (B; B) to itself. The formula
dθ (Δθ − L2T ) = (Δθ − L2T )dθ .
gives
q
dθ Gφ = Gdθ φ, φ ∈ C ∞ (B; B).
q
The rest of the proof consists of using this to show that HdR (B) is isomorphic to
H (B). The details are identical to the corresponding proof in Hodge theory.
q
15
q
8. The groups HdR,0 (N )
q
We now show that HdR,0 (N ), the q-th cohomology group of the complex (5.1),
is isomorphic to H (N , R) by exhibiting a homotopy from the identity morphism
q
of the de Rham complex of N to a projection with range the subcomplex (5.1)

inducing an invertible map in cohomology. Together with Lemma 5.1 this completes
the proof of Proposition 2.2.
Let g be the Lie algebra of G. There is a basis Ŷ1 , . . . , Ŷd of g such that each
of the curves s → exp(sŶj ) is periodic of period 2π. The map (s1 , . . . , sd ) →
(s1 , . . . , sd ) = exp(s1 Ŷ1 + · · · + sd Ŷd ) is surjective, a covering map of period 2π in
each component. Setting
1 d
ω((s1 , . . . , sd )) = (eis , . . . , eis )
q
we get well defined functions ω j : G → Cd . If φ ∈ C ∞ (N ; N ) define
πα φ = π∗ (ω α A∗ φ m), α ∈ Zd .
That is, if v1 , . . . , vd ∈ Tp N and V1 , . . . , Vd denotes the canonical liftings of the
respective vj as tangent vector fields of G × N along π −1 (p), then πα φ(p) = φα (p)
is the q-covector such that

πα φ(p)(v1 , . . . , vd ) = ω α (g) A∗g (φ(Ag p))(V1 , . . . , Vd ) dm

G
Then πα φ is smooth if φ is smooth and dπα = πα d. Since

A∗exp s πα φ = ei sj αj πα φ,
j Ŷj
q
φ = α∈Zd φα with convergence in C ∞ (N ; N ).
For α = (α1 , . . . , αd ) ∈ Zd let Jα = {j : αj = 0} and let mα be the cardinality
of Jα . Let Yj be the vector field on N determined by Ŷ j :

d
Yj (p) = Aexp(sŷj ) p.
ds s=0
Then LYj πα = iαj πα . Defining
1 1
hq φ = iY π α φ
mα iαj j
α=0 j∈Jα
we have
1 1
dhq φ = (LYj πα φ − iYj dπα φ) = φα − hq+1 dφ,
mα iαj
α=0 j∈Jα α=0
that is,
dhq + hq+1 d = I − π0 .
q
It follows immediately that any class in HdR (N ) has a representative in the space
q q q−1
C(0) (N ; N ). If φ ∈ C(0) (N ; N ) and φ = dψ for some ψ ∈ C ∞ (N ;
∞ ∞
N ),
then
φ − dπ0 ψ = dhq dψ.
Applyingπ0 to both sides of this identity gives π0 dhq dψ = 0. Since φ, dπ0 ψ ∈
∞ q
C(0) (N ; N ), dhq dψ must be zero. Thus there is a well defined injective map
q q
π 0 : HdR (N ) → HdR,0 (N ).
154
q q
The same argument shows that the map ι : HdR,0 (N ) → HdR (N ) obtained from the
∞
∗ ∞
∗
inclusions C(0) (N ; N ) → C (N ; N ), which is a left inverse of π 0 , is injective.
So π 0 is an isomorphism.
References
[1] Bott, R., Tu, L. W., Differential forms in algebraic topology, Springer-Verlag Berlin, Heidel-
berg, New York 1982.
[2] Gysin, W., Zur Homologietheorie der Abbildungen und Faserungen von Mannigfaltigkeiten,
Comment. Math. Helv. 14, (1942), 61–122.
[3] Hatcher, A.,Algebraic topology, Cambridge University Press, Cambridge, 2002.
[4] Helgason, S., Differential geometry, Lie groups, and symmetric spaces. Pure and Applied
Mathematics, 80. Academic Press, Inc, New York-London, 1978.
[5] Hörmander, L., Fourier integral operators. I, Acta Math. 127 (1971), 79–183.
[6] Characteristic classes of the boundary of a complex b-manifold, in Complex Analy-
sis (Trends in Mathematics), 245–262, P. Ebenfelt, N. Hungerbühler, J. J. Kohn, N. Mok,
E. J. Straube, Eds., Birkhuser, Basel, 2010.
[7] Milnor, J., Stasheff, J., Characteristic classes, Annals of Mathematics Studies, 76, Princeton
University Press, Princeton, N. J., 1974.
[8] Myers, S. B., Steenrod, N. E., The group of isometries of a Riemannian manifold, Annals
Math. 40 (1939), 400–416.
[9] Peterson, F., Some remarks on Chern classes, Ann. Math. 69 (1959), 414–420.
[10] de Rham, G. Sur l’analysis situs des variétés à n dimensions, J. de Math., IX. Ser. 10 (1931),
115-200.
[11] , Differentiable manifolds, Grundlehren der Mathematischen Wissenschaften 266.
Springer-Verlag, Berlin, 1984.
[12] Swan, R., Vector bundles and projective modules, Trans. Amer. Math. Soc. 105 (1962) 264–
277.
[13] Warner, F. W., Foundations of differentiable manifolds and Lie groups, Graduate Texts in
Mathematics, 94, Springer-Verlag, New York-Berlin, 1983.
E-mail address: gmendoza@math.temple.edu
Department of Mathematics, Temple University, Philadelphia, PA 19122

Volume 550, 2011
A potential theoretic characterization of compactness of the

∂-Neumann problem
Sönmez Şahutoğlu
Abstract. We give a potential theoretic characterization for compactness of

the ∂-Neumann problem on smooth bounded pseudoconvex domains in Cn .
Let Ω be a domain in Cn with C ∞ -smooth boundary. The domain Ω is said

to be pseudoconvex if the Levi form of Ω, the restriction of the complex Hessian
of a defining function onto complex tangent space, is positive semi-definite on the
boundary, bΩ, of Ω. On bounded pseudoconvex domains, Hörmander [Hör65]
showed that the ∂-Neumann operator on Ω, the solution operator for is a bounded
∗ ∗ ∗
operator on L2(0,1) (Ω) (here = ∂∂ +∂ ∂ and ∂ is the Hilbert space adjoint of ∂).
We refer the reader to [CS01, Str10] for more information about the ∂-Neumann
problem.
Compactness of the ∂-Neumann operator is important to study as it is weaker
than global regularity [KN65] and it interacts with the boundary geometry. For
example, even though the general case is still open, in same cases it is know that
existence of an analytic disc in the boundary is an obstruction for compactness of
the ∂-Neumann problem (see, for example, [FS98, FS01, Şah06, ŞS06, Str10]).
Recently, Straube and Munasinghe [MS07] (see also [Mun06]) studied compact-
ness using geometric conditions involving short time flows on the boundary (this
was done in C2 earlier by Straube [Str04]). Çelik and Straube [ÇS09] (see also
[Çel08]) explored compactness in relation to the so called “compactness multipli-
ers”.
Compactness of the ∂-Neumann problem has been studied using some potential
theoretic conditions by Catlin [Cat84] using property (P ) and later by McNeal
[McN02] using property (P ). In this paper we would like to give a new potential
theoretic characterization for compactness of the ∂-Neumann problem. We refer
the reader to [Str10, Proposition 4.2] for other equivalent conditions. We would
like to note that a similar characterization has been done by Haslinger in [Has].
2010 Mathematics Subject Classification. 32W05.

Key words and phrases. Potential theory, ∂-Neumann problem.
This article is based on a part of the authors Ph.D. thesis [Şah06].
The author is supported in part by University of Toledo’s Summer Research Awards and
Fellowships Program.
2011
c 0000
c Mathematical
holder)
1
155
156
2 SÖNMEZ ŞAHUTOĞLU
Let Ω be a smooth bounded domain in Cn , K ⊂ bΩ, and U be an open

neighborhood of K. We denote the L2 norm and Sobolev −1 norm of a function
f ∈ L2 (Ω) by f and f −1 , respectively. Let I = {i1 , i2 , . . . , ip } ⊂ N such that
j1 < i2 < · · · < ip . Then we use the notation dzI = dzi1 ∧ dzi2 ∧ · · · ∧ dzip and
dz I = dz i1 ∧ dz i2 ∧ · · · ∧ dz ip . Define

∞ ∞
C0,(p,q) (U ) = fIJ dzI ∧ dz̄J : fIJ ∈ C0 (U )
|I|=p,|J|=q
for 0 ≤ q ≤ n. Define λ(p,q) (U ) as follows: for 0 ≤ p ≤ n and 1 ≤ q ≤ n − 1 let us

define
∗

∂f 2 + ∂ f 2 ∗ ∞
λ(p,q) (U ) = inf : f ∈ Dom(∂ ) ∩ C0,(p,q) (U ), f ≡ 0
f 2
and

∂f 2
λ(p,0) (U ) = inf : f ∈ (Ker∂)⊥ ∩ C0,(p,0)
∞
(U ), f ≡ 0
f 2
where (Ker∂)⊥ is the orthogonal complement of (Ker∂) in L2(p,0) (Ω) (square inte-
grable (p, 0)-forms on Ω). Notice that λ(p,q) (U ) ≤ λ(p,q) (V ) if V ⊂ U. In this paper
a finite type is meant in the sense of D’Angelo [D’A82] and infinite type point
means a point that is not finite type.
Theorem 1. Let Ω be a smooth bounded pseudoconvex domain in Cn , n ≥ 2 and

0 ≤ p ≤ n, 0 ≤ q ≤ n − 1, be given. Then the following are equivalent:
(i) the ∂-Neumann operator N(p,q) of Ω is compact on L2(p,q) (Ω),
(ii) for any compact set K ⊂ bΩ and M > 0 there exists an open neighborhood
U of K such that λ(p,q) (U ) > M,
(iii) for any M > 0 there exists an open neighborhood U of the set of infinite
type points in bΩ such that λ(p,q) (U ) > M.
Remark 1. The definition of λ(p,q) is closely connected to the so-called com-

pactness estimates (see (1) in the proof of Theorem 1) as well as Morey-Kohn-
Hörmander formula (see, for example, [CS01, Proposition 4.3.1] or [Str10, Propo-
sition 2.4]) and property (P ) of Catlin.
One can show that the Morey-Kohn-Hörmander formula implies that for a
smooth bounded pseudoconvex Ω ⊂ Cn , a non-positive function b ∈ C 2 (Ω), and
∗
u ∈ Dom(∂) ∩ Dom(∂ ) ∩ C(p,q) 1
(Ω) we have

n
∂2b ∗
eb uJ,jK uJ,kK dV ≤ ∂u2 + ∂ u2 .
Ω ∂zj ∂z k
J,K j,k=1
n
where u = J,K k=1 uJ,kK dz J ∧ dzk ∧ dzK and the prime indicates that the sum
is taken over strictly increasing (p, q − 1)-tuples (J, K). If the domain Ω satisfies
property (P ) then one can choose b to be bounded from below by −1 and with
arbitrarily large complex Hessian on the boundary of Ω. Then on a small neighbor-
hood on the boundary the Hessian is still large. Hence λ(p,q) (U ) will be arbitrarily
large for a sufficiently small neighborhood U of bΩ.
COMPACTNESS OF THE ∂-NEUMANN PROBLEM 157
3
We would like to give a simple example below to show that one can use this
characterization to show that, in some cases, compactness of the ∂-Neumann prob-
lem excludes analytic disks from the boundary. We do not claim any originality in
this example as it is a special case of Catlin’s result [FS01, Proposition 1].
Example 1. Let Ω be a smooth bounded pseudoconvex domain in C2 such that
Ω ⊂ {z ∈ C2 : Im(z2 ) < 0} and {z ∈ C2 : Im(z2 ) = 0, |z1 |2 + |z2 |2 < 1} ⊂ bΩ.
Claim: The ∂-Neumann operator on Ω is not compact.
Proof of the Claim: There exist positive numbers a1 < a2 such that
D1 × W1 ⊂ Ω ∩ {z ∈ C2 : |z1 |2 + |z2 |2 < 1} ⊂ D2 × W2
where D1 = {z ∈ C : |z| < 2/3}, D2 = {z ∈ C : |z| < 2}, and
W1 = {z = reiθ ∈ C : 0 < r < a1 , −2π/3 < θ < −π/3},
W2 = {z = reiθ ∈ C : 0 < r < a2 , −4π/3 < θ < π/3}.
Let φj (z1 , z2 ) = f (z1 )gj (z2 )dz 1 where f ∈ C0∞ (D1 ) and f ≡ 0. Later on we will
∗
choose gj ∈ C0∞ ({z ∈ C : |z| < j −2 }) so that φj ∈ Dom(∂) ∩ Dom(∂ ). There exists
a3 > 0 such that D1 × W ⊂ Ω, where W = {z ∈ C : Im(z) < 0, |z| < a3 }, and
φj (z1 , z2 ) = 0 for z ∈ Ω \ D1 × W and j −2 < a3 . Then for j −2 < a3 we have

2

2
∗
∂gj (z2 )

∂f (z )
∂φj 2 + ∂ φj 2
∂z2 f (z1 )
+
gj (z2 ) ∂z11
=
φj 2 gj (z2 )f (z1 )2

2

∂gj (z2 )

∂f (z1 )
2

∂z2
2 f L2 (D2 )
∂z1
2 gj (z2 )2L2 (W )
L (W2 ) L (D1 )
≤ +
gj (z2 )2L2 (W1 ) f L2 (D1 ) f (z1 )2L2 (D1 ) gj (z2 )2L2 (W )

∂gj (z2 )
2
∂f (z1 )
2

∂z2
2
∂z1
2
L (W2 ) L (D1 )
≤ +
gj (z2 )2L2 (W1 ) f (z1 )2L2 (D1 )
Let us choose real valued non-negative functions χj ∈ C0∞ (−j −2 , j −2 ) such that
χj (−t) = χj (t) and χ(t) = 1 for |t| ≤ 4j12 . Since z −2 is not integrable on W1 ∩B(0, ε)
for any ε > 0, we can choose a positive real number αj so that
2 2 2 2
χ |z2 | χj |z2 |
j
dV (z2 ) ≤ dV (z2 ).
W2 ∩B(0,1/j) |z2 − iαj | W1 ∩B(0,1/j) |z2 − iαj |
2 2

Now we define gj (z2 ) = χj |z2 |2 τj (z2 )(z2 − iαj )−1 where τj ∈ C ∞ (C) such that
τj (z) ≡ 1 for Im(z) ≤ 0 and τj (z) ≡ 0 for Im(z) ≥ αj /2. Then we have φj ∈
∞ ∗
C0,(0,1) (Uj ) ∩ Dom(∂ ) where Uj = {z ∈ C : |z| < 2−1 + j −1 } × {z ∈ C : |z| < j −2 }
and

∂gj

≤ gj L2 (W1 ∩B(0,1/j)) .

∂z 2
2
L (W2 ∩B(0,1/j))
∞
Hence, we constructed a sequence of (0, 1)-forms {φj } such that φj ∈ C0,(0,1) (Uj ) ∩
∗
Dom(∂ ) where
∞

K = {z ∈ C2 : |z1 | ≤ 1/2, z2 = 0} = Uj ⊂ bΩ
j=1
158
∗
∂φ 2 +∂ φ 2
and j
φj 2
j
stays bounded as j → ∞. Hence, by Theorem 1, the ∂-Neumann
operator on Ω is not compact.
Proof of Theorem 1
Proof of Theorem 1. We will show the equivalences for 0 ≤ p ≤ n and
1 ≤ q ≤ n − 1. The proof can be mimicked for the case q = 0 using the following:
compactness of N0 is equivalent to the following compactness estimate: for all ε > 0
there exists Dε > 0 such that
g2 ≤ ε∂g2 + Dε g2−1 for g ∈ (Ker∂)⊥ ∩ Dom(∂)
First let us prove that (i) implies (ii). Assume that the ∂-Neumann operator
of Ω is compact, and there exist K ⊂ bΩ and M > 0 such that λ(p,q) (U ) < M for
all open neighborhoods U of K. We may assume that there exist sequences of open
neighborhoods {Uk } of K and nonzero (p, q)-forms {fk } such that
∗
i. Uk+1 Uk , K ⊂ ∞ ∞
k=1 Uk ⊂ bΩ, fk ∈ Dom(∂ ) ∩ C0,(p,q) (Uk ),
∗
ii. fk 2 = 1, and ∂fk 2 + ∂ fk 2 < M for k = 1, 2, 3, · · ·
∞ ∞
Since K ⊂ k=1 Uk ⊂ bΩ (hence K has measure 0 in Cn ) and fk ∈ C0,(p,q) (Uk ),
by passing to a subsequence if necessary, we may assume that fk − fl ≥ 1/2.
2
Compactness of the ∂-Neumann operator is equivalent to the following so called

compactness estimate (see [Str10, Proposition 4.2] or [FS01, Lemma 1]): for all
ε > 0 there exists Dε > 0 such that
∗ ∗
(1) g2 ≤ ε(∂g2 + ∂ g2 ) + Dε g2−1 for g ∈ Dom(∂ ) ∩ Dom(∂)
∗∞ ∗
Choose ε = 16M1
. Since Dom(∂ ) ∩ C0,(p,q) (Uk ) ⊂ Dom(∂ ) ∩ Dom(∂) using (1)
and ii. above we get
1
(2) fk − fl 2−1 ≥ > 0 for k = l
4Dε
The imbedding from L2 (Ω) to W −1 (Ω) is compact and {fk } is a bounded sequence
−1
in L2(p,q) (D). Hence {fk } has a convergent subsequence in W(p,q) (Ω). This contra-
dicts with (2).
(ii) obviously implies (iii) so we will skip this part.
Next let us prove that (iii) implies (i). Let K be the set of infinite type points
∗ ∞
in bΩ and u ∈ Dom(∂ ) ∩ C(p,q) (Ω). Assume that λ(p,q) (Uk ) > k where {Uk } is a
∞
sequence of open neighborhoods of K such that Uk+1 Uk and K ⊂ k=1 Uk ⊂ bΩ.
Let ϕk ∈ C0∞ (Uk ) such that 0 ≤ ϕk ≤ 1 and ϕk ≡ 1 in a neighborhood of K. Define
ψk = 1 − ϕk . Notice that ψk is supported away from K. In following estimates, Ck
and Ck,ε are general constants meaning that the constants depend on the subscripts
only but they might change at each step. Away from K we have subelliptic estimates
as bΩ \ K is the set of finite type points (see [Cat87]). Hence, there exists s > 0
for all ε > 0 there exists Dε > 0 such that
ψk u2 ≤ εψk u2s + Dε ψk u2−1
∗
(3) ≤ εCk (∂(ψk u)2 + ∂ (ψk u)2 ) + Ck,ε u2−1
∗
≤ εCk (∂u2 + ∂ u2 + u2 ) + Ck,ε u2−1
COMPACTNESS OF THE ∂-NEUMANN PROBLEM 159
5
The first inequality follows because L2 imbedds compactly into W s for s > 0. We
used the compactness estimate for the second inequality. If we use λ(p,q) (Uk ) > k
we get:
1 ∗
ϕk u2 ≤ (∂(ϕk u)2 + ∂ (ϕk u)2 )
k
1 ∗
(4) ≤ (∂u2 + ∂ u2 ) + Dk φk u2
k
where φk ≡ 0 in a neighborhood of K, Dk > 0, and φk ≡ 1 in a neighborhood of
the support of ϕk . Calculations that are similar to ones in (3) show that
∗
(5) φk u2 ≤ ε C̃k (∂u2 + ∂ u2 + u2 ) + C̃k,ε u2−1
By choosing ε, ε > 0 small enough and combining (3) and (5) we get the following
estimate: for all k = 1, 2, 3, · · · there exists Mk > 0 such that
2 ∗ ∗ ∞
(6) u2 ≤ (∂u2 + ∂ u2 + u2 ) + Mk u2−1 for u ∈ Dom(∂ ) ∩ C(p,q) (Ω)
k
∗
∞ ∗
We note that Dom(∂ ) ∩ C(p,q) (Ω) is dense in Dom(∂ ) ∩ Dom(∂). Therefore, the
∗
above estimate (6) holds on Dom(∂ ) ∩ Dom(∂). That is, the ∂-Neumann operator
of Ω is compact on (p, q)-forms for 0 ≤ p ≤ n and 1 ≤ q ≤ n − 1.
Acknowledgement
The author would like to thank his advisor, Emil Straube, for suggesting the
problem and fruitful discussions, and Mehmet Çelik for valuable comments on a
preliminary version of this manuscript.
References
[Cat84] ¯
David W. Catlin, Global regularity of the ∂-Neumann problem, Complex analysis of
several variables (Madison, Wis., 1982), Proc. Sympos. Pure Math., vol. 41, Amer.
Math. Soc., Providence, RI, 1984, pp. 39–49.
[Cat87] David Catlin, Subelliptic estimates for the ∂-Neumann problem on pseudoconvex do-
mains, Ann. of Math. (2) 126 (1987), no. 1, 131–191.
[Çel08] Mehmet Çelik, Contributions to the compactness theory of the ∂-Neumann operator,
Ph.D. thesis, Texas A&M University, 2008.
[ÇS09] Mehmet Çelik and Emil J. Straube, Observations regarding compactness in the ∂-
Neumann problem, Complex Var. Elliptic Equ. 54 (2009), no. 3-4, 173–186.
[CS01] So-Chin Chen and Mei-Chi Shaw, Partial differential equations in several complex vari-
ables, AMS/IP Studies in Advanced Mathematics, vol. 19, American Mathematical So-
ciety, Providence, RI, 2001.
[D’A82] John P. D’Angelo, Real hypersurfaces, orders of contact, and applications, Ann. of Math.
(2) 115 (1982), no. 3, 615–637.
[FS98] Siqi Fu and Emil J. Straube, Compactness of the ∂-Neumann problem on convex do-
mains, J. Funct. Anal. 159 (1998), no. 2, 629–641.
[FS01] , Compactness in the ∂-Neumann problem, Complex analysis and geometry
(Columbus, OH, 1999), Ohio State Univ. Math. Res. Inst. Publ., vol. 9, de Gruyter,
Berlin, 2001, pp. 141–160.
[Has] ¯
Friedrich Haslinger, Compactness for the ∂-Neumann problem - a functional analysis
approach, to appear in Collect. Math., arXiv:0912.4406.
[Hör65] Lars Hörmander, L2 estimates and existence theorems for the ∂¯ operator, Acta Math.
113 (1965), 89–152.
[KN65] J. J. Kohn and L. Nirenberg, Non-coercive boundary value problems, Comm. Pure Appl.
Math. 18 (1965), 443–492.
160
[McN02] Jeffery D. McNeal, A sufficient condition for compactness of the ∂-Neumann operator,
J. Funct. Anal. 195 (2002), no. 1, 190–205.
[Mun06] Samangi Munasinghe, Geometric sufficient conditions for compactness of the ∂-
Neumann operator, Ph.D. thesis, Texas A&M University, 2006.
[MS07] Samangi Munasinghe and Emil J. Straube, Complex tangential flows and compactness
of the ∂-Neumann operator, Pacific J. Math. 232 (2007), no. 2, 343–354.
[Şah06] Sönmez Şahutoğlu, Compactness of the ∂-Neumann problem and Stein neighborhood
bases, Ph.D. thesis, Texas A&M University, 2006.
[ŞS06] Sönmez Şahutoğlu and Emil J. Straube, Analytic discs, plurisubharmonic hulls, and
non-compactness of the ∂-Neumann operator, Math. Ann. 334 (2006), no. 4, 809–820.
[Str04] Emil J. Straube, Geometric conditions which imply compactness of the ∂-Neumann
operator, Ann. Inst. Fourier (Grenoble) 54 (2004), no. 3, 699–710.
[Str10] , Lectures on the L2 -Sobolev theory of the ∂-Neumann problem, ESI Lectures in
Mathematics and Physics, vol. 7, European Mathematical Society (EMS), Zürich, 2010.
University of Toledo, Department of Mathematics, Toledo, OH 43606, USA

E-mail address: sonmez.sahutoglu@utoledo.edu
Volume 550, 2011
Volume 00, 1997
Duality between Harmonic and Bergman spaces
Mei-Chi Shaw*
Abstract. In this paper we study the duality of the harmonic spaces on the
−
annulus Ω = Ω1 \ Ω between two pseudoconvex domains with Ω− ⊂⊂ Ω1 in
C and the Bergman spaces on Ω− . We show that on the annulus Ω, the space
n
of harmonic forms for the critical case on (0, n−1)-forms is infinite dimensional
and it is dual to the the Bergman space on the pseudoconvex domain Ω− . The
duality is further identified explicitly by the Bochner-Martinelli transform,
generalizing a result of Hörmander.
Introduction
Let Ω and Ω1 be two bounded pseudoconvex domains in Cn with Ω− ⊂⊂
−
Ω1 . In this paper we study the duality of the harmonic spaces on the annulus
−
Ω = Ω1 \ Ω and the Bergman spaces on Ω− . This paper is inspired by a recent
¯
paper of Hörmander [Hö 2] where the null space of the ∂-Neumann operator on
a spherical shell as well as on an ellipsoid in Cn has been computed by explicit
formula for the critical case for (0, n − 1)-forms.
¯
The ∂-Neumann problem on the annulus has been studied in [Sh 1] on an
annulus between two pseudoconvex domains in Cn or in a hermitian Stein manifold.
When the boundary is smooth, the closed range property and boundary regularity
for ∂¯ were established in the earlier work (see [BS] or [Sh1]) for 0 < q ≤ n − 1 and
n ≥ 2. In the case when 0 < q < n − 1, the space of harmonic forms is trivial. In
this paper, we will study the critical case when q = n − 1 on the annulus Ω. In this
case the space of harmonic forms is infinite dimensional. Our goal is to establish
the duality between the harmonic forms in the critical degree with the Bergman
spaces on the domain Ω− .
In the first section, we recall the Hodge decomposition theorem on the annulus
between two pseudoconvex domains. In the second section we establish the duality
between the harmonic forms with coefficients in the Sobolev W 1 (Ω) spaces with the
Bergman spaces on Ω− . We then refine the duality to duality between L2 spaces
in Section 3.
2010 Mathematics Subject Classification. Primary 32W05, 35N15, 58J32.

Key words and phrases. Cauchy-Riemann equations, Pseudoconcave domains, Harmonic
spaces, Bergman spaces.
*partially supported by NSF grant.
c1997
2011
c American Mathematical
Society
1
161
2162 MEI-CHI SHAW*
1. L2 theory for ∂¯ on the annulus between

two weakly pseudoconvex domains in Cn
We recall the following L2 existence and estimates for ∂¯ in the annulus between
two pseudoconvex domains (see Theorems 3.2 and 3.3 in Shaw [Sh4]).
−
Theorem 1.1. Let Ω ⊂⊂ Cn , n ≥ 3, be the annulus domain Ω = Ω1 \ Ω
between two pseudoconvex domains Ω1 and Ω− . We assume that Ω− ⊂⊂ Ω1 and
Ω− has C 2 boundary. For any f ∈ L2(p,q) (Ω), where 0 ≤ p ≤ n and 0 ≤ q < n − 1,
¯ = 0 in Ω, the following hold:
such that ∂f
−1 ¯ = 0 in Ω1 in the
(1) there exists F ∈ W(p,q) (Ω1 ) such that F |Ω = f and ∂F
distribution sense.
¯ = f in Ω.
(2) there exists u ∈ L2(p,q−1) (Ω) satisfying ∂u
¯
For q = n − 1, there is an additional compatibility condition for the ∂-closed
extension of (p, n − 1)-forms.
−
Theorem 1.2. Let Ω ⊂⊂ Cn be the annulus domain Ω = Ω1 \ Ω between two
pseudoconvex domains Ω1 and Ω− . We assume that Ω− ⊂⊂ Ω1 and Ω− has C 2
¯
boundary. For any ∂-closed f ∈ L2(p,n−1) (Ω), where 0 ≤ p ≤ n, the following hold:
−1 ¯ = 0 in Ω1 in
(1) There exists F ∈ W(p,n−1) (Ω1 ) such that F |Ω = f and ∂F
the distribution sense.
(2) The restriction of f to bΩ− satisfies the compatibility condition

(1.1) f ∧ φ = 0, φ ∈ W(n−p,0)
1
(Ω− ) ∩ Ker(∂).
¯
bΩ−
¯ = f in Ω.
(3) There exists u ∈ L2(p,n−2) (Ω) satisfying ∂u
Corollary 1.3. Let Ω be the same as in Theorem 1.2. Then ∂¯ has closed range
¯
in L2(p,n−1) (Ω) and the ∂-Neumann operator N(p,n−1) exists on L2(p,n−1) (Ω).
Theorem 1.4 (Hodge Decomposition Theorem). Let Ω ⊂⊂ Cn be the

−
annulus domain Ω = Ω1 \ Ω between two pseudoconvex domains Ω1 and Ω− . We
assume that Ω− ⊂⊂ Ω1 and Ω− has C 2 boundary. Then the ∂-Neumann
¯ operator
N(p,q) exists on L(p,q) (Ω) for 0 ≤ p ≤ n and 0 ≤ q ≤ n. For any f ∈ L2(p,q) (Ω), we
2
have
f = ∂¯∗ ∂N
¯ (p,0) f + H(p,0) f, q = 0.
¯ ¯∗ ¯∗¯
f = ∂ ∂ N(p,q) f + ∂ ∂N(p,q) f, 1 ≤ q ≤ n − 2.
¯ ¯∗ ¯ ∗¯
f = ∂ ∂ N(p,n−1) f + ∂ ∂N(p,n−1) f + H(p,n−1) f, q = n − 1.
f = ∂¯∂¯∗ N(p,n) f, q = n.
We have used the notation H(p,q) to denote the projection operator from L2(p,q) (Ω)
onto the harmonic space H(p,q) (Ω) = ker((p,q) ).
For a proof of Theorem 1.4, see Theorem 3.5 in [Sh4].
Remark: All the results can be extended to annulus between pseudoconvex do-
mains in a Stein manifold with trivial modification.
DUALITY BETWEEN HARMONIC AND BERGMAN SPACES 1633
(Ω) and H (n−p,0) (Ω− )

(p,n−1)
2. The duality between HW 1
(p,q)
For k ≥ 0, we define the Dolbeault cohomology HW k (Ω) with W k (Ω)-coefficients
by
{f ∈ W(p,q)
k ¯ = 0}
(Ω) | ∂f
(p,q)
HW k (Ω) = ¯ u ∈ Wk .
{f ∈ W(p,q) (Ω) | f = ∂u,
k
(p,q−1) (Ω)}
(p,0)
For k ∈ R, we define HW k (Ω) to be the space of (p, 0)-forms with holomorphic
coefficients in W k (Ω).
If Ω is the annulus between two pseudoconvex domains as in Theorem 1.2, we
have that the space {f ∈ W(p,q)
k ¯ u ∈ Wk
(Ω) | f = ∂u, (p,q−1) (Ω)} is closed. Further-
more, we have from Theorem 1.4:
(p,n−1)
H(p,n−1) (Ω) HL2 (Ω).
(p,n−1) (0,0)
We will use the notation H (p,n−1) (Ω) for HL2 (Ω) and HW 1 (Ω) = HW 1 (Ω).
−
Theorem 2.1. Let Ω ⊂⊂ Cn be the annulus domain Ω = Ω1 \ Ω between
two pseudoconvex domains Ω1 and Ω− with smooth boundary and Ω− ⊂⊂ Ω1 ,
(p,q)
n ≥ 2. For each k ≥ 0 and 0 ≤ p ≤ n, the space HW k (Ω) = {0}, when 0 < q <
(p,n−1)
n − 1 and the space HW k (Ω) is of infinite dimension. Furthermore, we have the
isomorphism:
(Ω) (HW −k+1 (Ω− ))
(p,n−1) (n−p,0)
HW k
where the right-hand side is the space of all bounded linear functionals on the space
HW −k+1 (Ω− ).
(n−p,0)
Proof. First we assume that k = 0. Suppose that f ∈ L2(p,n−1) (Ω) and

¯ = 0. We define a pairing between H (p,n−1) (Ω) and H (n−p,0)
∂f W1 (Ω− )
(Ω− ) → C
(n−p,0)
l : H (p,n−1) (Ω) × HW 1
by

(Ω− ).
(n−p,0)
(2.1) l([f ], h) = f ∧ h, h ∈ HW 1
bΩ−
First we note that the pairing (2.1) is well-defined. It is well-known any holomorphic
function or forms with L2 (Ω) coefficients has a well-defined trace in W − 2 (bΩ) (see
1
e.g. [LM]). For any f in L2(p,n−1) (Ω) with ∂f ¯ = 0 and ∂¯∗ f = 0, we also have a
well-defined trace in W − 2 (bΩ) (see [Sh3] for details). Any function or form with
1
W 1 (Ω− ) coefficients has trace in W 2 (bΩ− ) from the Sobolev Trace Theorem. Thus
1
the pairing between f and φ in (2.1) is well-defined since

| f ∧ h| ≤
f
− 21 −
h
12 − ≤
f
L2 (Ω)
h
W 1 (Ω− ) .
W (bΩ ) W (bΩ )
bΩ−
We also note that the pairing in (2.1) is independent of the choice of the repre-
sentation function [f ]. Let f˜ be another representation of [f ], then f˜ = f + ∂u
¯ for
some element of the form ∂u¯ ∈ L2 (Ω) with u ∈ L 2
(Ω). Using Friedrichs’
(p,n−1) (p,n−2)
4164 MEI-CHI SHAW*
∞
lemma, there exists a sequence {uν } such that uν ∈ C(p,n−2) (Ω) such that uν → u
2 ¯ ν → ∂u
in L(p,n−2) (Ω) and ∂u ¯ in L 2
(Ω). It follows from Stokes’ Theorem that
(p,n−1)

¯ ∧ h = lim
∂u ¯ ν ∧h
∂u
bΩ− ν→∞ bΩ−

(Ω− ).
¯ = 0, (n−p,0)
= lim (−1)p+n uν ∧ ∂h h ∈ HW 1
ν→∞ bΩ−
Thus the pairing (2.1) is well-defined.

If we assume that f satisfies the condition

f ∧ φ = 0, φ ∈ W(n−p,0)
1
(Ω− ) ∩ Ker(∂),
¯
bΩ−
¯ −1
from Theorem 1.2, there exists a ∂-closed form F ∈ W(p,n−2) (Ω1 ) which is equal to
¯ = f . This implies
f on Ω and one can find a solution u ∈ L(p,n−2) (Ω) satisfying ∂u
2
(Ω− ) .
(n−p,0)
that [f ] = 0. Thus there is a 1-1 map from H (p,n−1) (Ω) to HW 1
(Ω− ).
(n−p,0)
On the other hand, suppose that f is a bounded linear functional on HW 1
We will show that l can be represented by some [f ] in (2.1). Since we assume that
Ω− is pseudoconvex and has smooth boundary, one has the duality for holomorphic
space H 1 (Ω− ) = L2 (Ω− ) ∩ Ker(∂)¯ and H −1 (Ω− ) (see [BB]). If the ∂-Neumann
¯
−
operator is exact regular on W (Ω ), we can use the duality between the usual L2
1
¯
spaces. Otherwise, one can use the exact regularity for the weighted ∂-Neumann
2
operator with weights t|z| for sufficiently large t > 0. The weight function can be
viewed as the bundle metric e−t|z| for the trivial line bundle C and the dual space
2
2
will be equipped with the dual metric et|z| for C. In particular the pairing (2.1)
¯
is well-defined. For simplicity, we assume that the ∂-Neumann operator is exact
regular. But all the arguments remain the same if we use weighted spaces with the
dual weighted norms.
Thus l can be represented by (n − p, 0)-form g with distribution coefficients
in H −1 (Ω− ) = ker(∂)
¯ ∩ W −1 (Ω− ). Extending g to be zero outside Ω− , then g
¯
is a (p, n)-form on Ω1 , a top degree form which is always ∂-exact. The extension
by zero of g results in a form which is in W −1 (Ω1 ). This is due to the fact that
holomorphic functions in W −1 (Ω− ) is also in the dual of W 1 (Ω− ). We remark that
for a general function or forms, this is not true. But when the functions or forms
are harmonic, then the dual space of W01 , denoted by W −1 , coincides with the dual
space of W 1 for domains with smooth boundary. For detailed explanation of this
subtle point, we refer the reader to the paper by Boas (see Appendix B in [Boa]
where the dual space of W 1 is denoted by W∗−1 .).
Thus we have that g = ∂U ¯ on Ω1 for some U ∈ L2
(p,n−1) (Ω1 ). Let f = U on
¯
Ω. It follows that ∂f = 0 on Ω and the linear functional

(Ω− ).
¯ ∧h= (n−p,0)
l(h) = g ∧ h = ∂U f ∧ h, h ∈ HW 1
Ω− Ω− bΩ−
Since f ∈ L2(p,n−1) (Ω), we have that the bounded linear functional l is represented
by [f ] ∈ H (p,n−1) (Ω). This proves the theorem for k = 0.
Suppose that k ≥ 1 and f ∈ W(p,n−1)

k ¯ = 0. We define a pairing
(Ω) and ∂f
(Ω) and HW −k+1 (Ω− ) by
(p,n−1) (n−p,0)
between HW k
(Ω) × HW −k+1 (Ω− )

(p,n−1) (n−p,0)
l : HW k

(2.2) l([f ], h) = f ∧ h.
bΩ−
It is easy to see that the pairing (2.2) is well-defined as before. If f satisfies the
condition
−k+1
f ∧ φ = 0, φ ∈ W(n−p,0) (Ω− ) ∩ Ker(∂),
¯
bΩ−
¯
there exists a ∂-closed form F ∈ W(p,n−1)
k−1
(Ω1 ) which is equal to f on Ω and one
¯ = f . For a proof, see Corollary 2
can find a solution u ∈ W(p,n−2) (Ω) satisfying ∂u
k
in the recent paper by Chakrabarti-Shaw [CS2]. This implies that [f ] = 0.

(Ω) to HW −k+1 (Ω− ) . Any
(p,n−1) (n−p,0)
Thus there is a one to one map from HW k
element in HW −k+1 (Ω− ) can be identified as a (p, n)-form g with HW k−1 (Ω− )-
(n−p,0)
coefficients. Thus repeating the same arguments as before, there exists f ∈ W k (Ω)
¯ = 0 such that any bounded linear functional can be given by f in the
with ∂f
equation (2.1).
We remark that in Theorem 2.1, the boundary is of Ω is assumed to be C ∞

smooth in order to have the duality for all k ≥ 0. For each fixed k, the duality result
holds for sufficiently smooth (depending on k) domains Ω− and Ω1 . In particular,
Theorem 2.1 holds for k = 1 for much less smooth domains Ω1 and Ω− . In the
following, we will only assume that the boundary for Ω− be Lipschitz, i.e., locally
it is the graph of a Lipschitz function.
Let ∂¯c : L2(p,n−1) (Ω− ) → L2(p,n) (Ω− ) be the minimal closure of ∂.
¯ By this we
mean that f ∈ Dom(∂¯c ) if and only if that there exists a sequence of smooth forms
∞ ¯ ν → ∂f
¯ in
fν in C(p,n−1) (Ω) compactly supported in Ω such that fν → f and ∂f
L . Let ϑ be the dual of ∂¯c . Then ϑ is equal to the maximal closure of the operator
2
ϑ : L2(p,n) (Ω− ) → L2(p,n−1) (Ω− ).
We set
c(p,n) (Ω− ) = ∂¯c ϑ : L2(p,n) (Ω− ) → L2(p,n) (Ω− ).
The kernel of c(p,n) (Ω− ) is denoted by Hc (Ω− ), the space of harmonic forms of
(p,n)
degree (p, n) with compact support.

−
bounded pseudoconvex domains Ω1 and Ω− with Ω− ⊂⊂ Ω1 , n ≥ 2. We assume
that Ω1 has C 2 boundary and Ω− has Lipschitz boundary. The space HW 1
(p,n−1)
(Ω)
is of infinite dimension and we have the isomorphism:
(Ω) Hc(p,n) (Ω− ) H (n−p,0) (Ω− ).

(p,n−1)
HW 1
6166 MEI-CHI SHAW*
Proof. From the closed range property for ∂¯ on Ω− and its L2 dual for all
degrees, it follows (see [CS2]) that range of ∂¯c is also closed for all degrees. In
particular, we have
L2(p,n) (Ω− ) = Range(∂¯c ) ⊕ Ker(ϑ).
Here we only need the boundary Ω− to be Lipschitz smooth (see [CS2] for details).
This gives that
(2.3) Hc(p,n) (Ω− ) = Ker(ϑ).
Using star operator, one has that
(2.4) Ker(ϑ) H (n−p,0) (Ω− ).

¯
From the extension of the ∂-closed forms from the annulus to Ω1 as in the
(p,n−1)
proof of Theorem 1.2, we will show that HW 1 (Ω) is isomorphic to the perp of
¯
Range(∂c ). To see this, let f ∈ W(p,n−1) (Ω) and let ∂f
1 ¯ = 0 in Ω. We extend f to
be a form f˜ ∈ W 1 (Ω1 ). The equation
(p,n−1)
(2.5) ∂¯c u = ∂¯f˜
for some u ∈ L2(p,n−1) (Ω− ) if and only if

(2.6) f ∧ φ = 0, φ ∈ L2(n−p,0) (Ω− ) ∩ Ker(∂).
¯
bΩ−
For a proof of the equivalence of (2.5) and (2.6), see the proof Proposition 5 in
[CS2].
¯
In this case, f can be extended to be ∂-closed form F where

f, z ∈ Ω,
F = ˜
f − u, z ∈ Ω− .
¯ = 0 in Ω1 and F = f on Ω. The form F is in L2
It follows that ∂F (p,n−1) (Ω1 ) but F
is in W (Ω) since F = f on Ω. Since we assume that the boundary Ω1 is C 2 , we can
1
find a solution (see [Ha]) F = ∂U ¯ for some U ∈ W 1

(p,n−2) (Ω1 ). In fact we can use
∗
the solution U = F + ∂¯t Nt F by the weighted ∂-Neumann
¯ operator Nt on Ω1 . Then
U is in W 1 near the boundary Ω1 from the boundary regularity for the weighted
¯
∂-Neumann ¯
operator. Since the weighted ∂-Neumann operator Nt is elliptic in the
interior of Ω1 , Nt F is in W 2 (Ω, loc). Thus the solution U is in W 1 (Ω, loc). Thus
(p,n−1)
1
U is in W(p,n−2) (Ω1 ). This shows that for any [f ] ∈ HW 1 (Ω), [f ] = 0 if and
only if (2.5) or (2.6) is satisfied for any representation f ∈ W(p,n−1)
1
(Ω). Repeating
the arguments as in Theorem 2.1 and using (2.3) and (2.4), we have proved the
theorem.
3. The isomorphism between H (p,n−1) (Ω) and H (n−p,0) (Ω− )

In this section we will further establish the isomorphism between the spaces
H (p,n−1) (Ω) and H (n−p,0) (Ω− ).
−
pseudoconvex domains Ω1 and Ω− and Ω− ⊂⊂ Ω1 , n ≥ 2. We assume that the
boundary of Ω− is C 2 smooth. Then we have the isomorphism:
(3.1) H (p,n−1) (Ω) H (n−p,0) (Ω− ).
Furthermore, if we assume that Ω has C 2 smooth boundary, then we have the iso-
morphism:
(p,n−1)
(3.2) HW 1 (Ω) H (p,n−1) (Ω).
Proof. It follows from Theorem 2.2 that

(Ω) H (n−p,0) (Ω− ).
(p,n−1)
HW 1
Thus it suffices to prove (3.1).
Let h ∈ H (n−p,0) (Ω− ). We will associate h with a ∂-closed
¯ form h+ in Ω as
follows:
Let ρ be a normalized C 2 defining function for Ω− . Since h has holomorphic
coefficients in L2 (Ω− ), it is well known that h has W − 2 boundary values on bΩ− .
1
Let h1 = h ∧ ∂ρ. ¯ The form h1 is a (p, n − 1)-form on Ω− and it has boundary
value in W (bΩ− ). We denote the restriction of h1 to bΩ− by
− 12
(3.3) hb = h1 |bΩ .

Let h+ = B + hb and h− = B − hb be the Bochner transform of hb defined by

(3.4) h+ = B + h b = B(ζ, z) ∧ hb , z ∈ Ω,
bΩ−

(3.5) h − = B − hb = B(ζ, z) ∧ hb , z ∈ Ω− .
bΩ−
We have the jump formula:

hb = B + hb − B − hb = h+ − h−
in terms of distributions. Also each h+ and h− are ∂-closed
¯ and L2 on Ω and Ω−
respectively.
We define a map l+ : H (n−p,0) (Ω− ) → H (p,n−1) (Ω) by
l+ h = [h+ ], h ∈ H (n−p,0) (Ω− )

¯
where h+ is defined by (3.4). Since h+ is ∂-closed on Ω and has L2 coefficients, the
+
map l is well-defined.
We next show that l+ is one to one. If l+ (h) = [h+ ] = 0 for some h ∈
H (n−p,0)
(Ω− ), we will show that h = 0. Since [h+ ] = 0, this implies that h+ can
¯
be represented by a ∂-exact form and there exists u+ ∈ L2 (Ω) such that
(p,n−2)
(3.6) ¯ +.
h+ = ∂u
Let h− be defined by by (3.5). Since Ω− is pseudoconvex, we have
(3.7) h− = ∂u
¯ −
8168 MEI-CHI SHAW*
for some u− ∈ L2(p,n−2) (Ω− ). It follows from (3.6) and (3.7) that that for each
(Ω− ),
(n−p,0)
g ∈ HW 1

− ¯ − ) ∧ g = 0.
¯ + − ∂u
hb ∧ g = (h − h ) ∧ g =
+
(∂u
bΩ− bΩ− bΩ−
(Ω− ). But from the

(n−p,0)
This implies that hb is a linear functional vanishing on HW 1
¯ 1
regularity for the weighted ∂-Neumann operator for W(n−p,1) (Ω) (since we assume
that Ω− is C 2 ), we have that the space HW 1 (Ω− ) is dense in HL2 (Ω− ).
(n−p,0) (n−p,0)
Since the functional vanishes on a dense subspace, hb must be zero. This proves
that h = 0 if l+ h = [h+ ] = 0. Thus l+ is one to one.
To show that l+ is onto, take an element F ∈ L2(p,n−1) (Ω) such that ∂F ¯ = 0.
For simplicity, we assume that p = n. We will construct a holomorphic function
h in L2 (Ω) such that l+ h = F . Note that from Theorem 1.4, any element [F ] can
be represented by a harmonic form and we may assume that F is in H(n,n−1) (Ω).
¯ = ∂¯∗ F = 0. It follows that F has boundary value with W − 12 -
This implies that ∂F √
coefficients. Choose a special orthonormal frame field basis w1 , · · · , wn = 2∂ρ¯ for
(1, 0)-forms. Then near the boundary F written in the special orthonormal frame
fields as
n
F = Fi (dV w̄i )
i=1
where dV = w1 ∧ w̄1 . . . wn ∧ w̄n is the volume element. Using
1 1
w̄n = wn = dρ = dσ
2 2
on bΩ− , we have
1
F |bΩ− = Fn dV w̄n |bΩ− = Fn dσ,
2
where dσ is the surface element on bΩ− .
We claim that F̄n is a CR distribtution on bΩ− . To see this, note that ∂¯ F = 0
since ϑF = 0. Restricted to the boundary, this implies that ∂¯F̄n ∧ w̄n = 0 on bΩ− .
Thus F̄n is a CR distribution in W − 2 (bΩ− ). Let h = F̄n be the holomorphic
1
extension of F̄n from the boundary to Ω− . Then h is an L2 holomorphic function

on Ω− . Let h+ = l+ h be the ∂-closed
¯ form in L2(n,n−1) (Ω). It remains to show that
[h+ ] = [F ]. This follows from h+ − h− = hb = (F̄n ∧ ∂ρ)¯ = Fn dV w̄n on bΩ− . If
we define
F − h+ , z ∈ Ω
G=
h − , z ∈ Ω− .
2 ¯ ¯ for some U ∈
Then G is an L ∂-closed form in Ω1 . Thus we have G = ∂U
2 ¯ +
L(n,n−2) (Ω1 ) is ∂-exact on Ω1 . Thus [F ] = [h ] in H (n,n−1)
(Ω). This proves that
l+ is onto. The theorem is proved.
Corollary 3.2. Let Ω be the same as Theorem 3.1. Each element f in the
harmonic space H(p,n−1) (Ω) can be represented by some h+ , where h is a harmonic
form in L2(n−p,0) (Ω− ). We have the following representation for the harmonic space
H(p,n−1) (Ω) = {h+ | h ∈ L2(n−p,0) (Ω− ), ∂h

¯ = 0.}
where h+ is defined by (3.4).
Proof. We have proved that for every f ∈ H(p,n−1) (Ω), we can write f =
¯ for some holomorphic h in L2 −
h∂ρ = h̄ ∂ρ (n−p,0) (Ω ). On the other hand, any
h ∈ H(n−p,0) (Ω− ), the associated B + (hb ) = h+ is in L2(p,n−1) (Ω). The form h+ is
¯
automatically ∂-closed. To see that it is in the domain of ∂¯∗ and ∂¯∗ h̄+ = 0, we
repeat the arguments before and the corollary is proved.
Remarks:
(1) All the results can be extended to any annulus between two pseudoconvex
domains in a Stein manifold with trivial modification. It can also be applied
to an annulus between two pseudoconvex domains in complex manifolds if
¯
one has the existence and the W 1 regularity of the ∂-Neumann operator on
−
the pseudoconvex domain Ω . We refer the reader to some related results
in [HI] (see also [CaS] and [CS1] and [CS2]).
(2) If we assume that the boundary is C ∞ smooth, we can also have the
(p,n−1)
isomorphism between H (p,n−1) (Ω) and HW k (Ω) for all k following the
same proof.
¯
4. The null space for the ∂-Neumann operator between balls
When the domain Ω = {z ∈ Cn | 0 < R0 < |z| < R1 } is the annulus between
two balls centered at 0, the harmonic space H(0,n−1) has been computed explicitly
in Hörmander (see equation (2.3) in [Hör2]). He proved that any (0, n − 1)-form
¯
f is in the null space of the ∂-Neumann operator if and only if

n
j z̄j z̄
(4.1) f= (−1) h dz̄1 ∧ · · · ∧ dz̄j−1 ∧ dz̄j+1 ∧ · · · ∧ dz̄n ,
1
|z|2n |z|2
where h is a holomorphic function in L2 (Ω∗ ), Ω∗ = {z ∈ Cn | |z| < R10 }. It is easy

¯ = ϑf = 0.
to check that ∂f
To see that f is in the domain of ∂¯∗ , we note that f = |z|12n h |z|z̄ 2 ∂ρ
¯ where
ρ = |z|2 . Thus f ∈ Dom(∂¯∗ ). We will show that the representation by the Bochner
transform stated the harmonic forms from Corollary 3.2 agrees with the formula
(4.1). For simplicity, we will assume that the inner ball is the unit ball, i.e., R0 = 1
and h is a holomorphic (n, 0) form in L2(n,0) (B1 ). Let ρ(z) = |z|2 − 1. Then h(z̄) is
holomorphic with L2 coefficients. The Bochner transform described in Section 3 is
given by

B + hb = ¯
B0 (ζ, z) ∧ h(ζ̄) ∧ ∂ρ(ζ)
ζ∈bB1
(4.2) ζ̄j
ζ̄
= B0 (ζ, z) ∧ h( ) ∧ [dζ̄j ]
ζ∈bB1 |ζ|2 j
|ζ|2n
where [dζ̄j ] denote the (n, n − 1)-form dV dζ̄j . It is easy to see that
⎛ ⎞
ζ̄j
∂¯ ⎝ [dζ̄j ]⎠ = 0, ζ = 0.
j
|ζ|2n
170
10 MEI-CHI SHAW*
We also have
⎛ ⎞
∂h δjk
ζ̄ ζ̄ ζ̄k ζj ζ̄j
∂¯ ⎝h ⎠
j
∧ [dζ̄ j ] = − dV
|ζ| 2
j
|ζ| 2n
j
∂wk |ζ|2 |ζ| 4 |ζ|2n
k
∂h ζ̄k ζ̄k

= − 2 dV = 0.
∂wk |ζ|2 |ζ|
k
Applying Stokes’s Theroem to (4.2), we see that

z̄j
z̄
+
B hb = h dV dz̄j .
|z|2n |z|2
¯ ∩ Dom(∂¯∗ ) ∩
We mention that in [Hör2], it is also proved that for any f ∈ Dom(∂)
H(n,n−1) (Ω)⊥ ,
(4.3) ¯
2 +
∂¯∗ f
2 .
max (n − 2, 1)
f
2 ≤ R12
∂f
Notice that the constant in (4.3) is independent of the inner diameter R0 . It is not
known if one can have such estimates on the more general annulus between two
pseudoconvex domains.
References
[BB] Bell, S. R. and Boas, H. P., Regularity of the Bergman projection and duality of holo-
morphic function spaces, Math. Ann. 267 (1984), 473-478.
[Boa] Boas, H. P., The Szegö projection: Sobolev estimates in regular domains, Trans. Amer.
Math. Soc., 300 (1987), 109–132.
[BS] Boas, H. P. and Shaw, M.-C., Sobolev Estimates for the Lewy Operator on Weakly
pseudo-convex boundaries, Math. Annalen 274 (1986), 221-231.
[BSt] Boas, H. P., and Straube, E. J., Sobolev estimates for the ∂-Neumann operator on
domains in Cn admitting a defining function that is plurisubharmonic on the boundary,
Math. Zeit., 206 (1991), 81–88.
[CaS] ¯
J. Cao and M.-C. Shaw, The ∂-Cauchy problem and nonexistence of Lipschitz Levi-flat
hypersurfaces in CP n with n ≥ 3, Math. Zeit. 256 (2007), 175-192.
[CS1] D. Chakrabarti and M.-C. Shaw, The Cauhcy-Riemann equations on product domains,
To appear in Math. Annalen (on line July 27, 2010).
[CS2] D. Chakrabarti and M.-C. Shaw, L2 Serre duality on domains in complex manifolds
with applications, To appear in Trans. Amer. Math. Society..
[CS] Chen, S.-C. and Shaw, M.-C., Partial Differential Equations in Several Complex Vari-
ables, American Math. Society-International Press, Studies in Advanced Mathematics,
Volume 19, Providence, R.I., 2001.
[Fo] Folland, G. B., The tangential Cauchy-Riemann complex on spheres, Trans. Amer. Math.
Society 171 (1972), 83-133.
[FK] Folland, G. B. and Kohn, J. J., The Neumann Problem for the Cauchy-Riemann Com-
plex, Ann. Math. Studies 75, Princeton University Press, Princeton, N.J., 1972.
[Gr] Grisvard, P., Elliptic Problems in Nonsmooth Domains, Pitman, Boston, 1985.
[Ha] Harrington, P.S., Sobolev Estimates for the Cauchy-Riemann Complex on C 1 Pseudo-
convex Domains, Math. Z. 262 (2009), 199-217.
[HI] Henkin, G. M. and Iordan, A., Regularity of ∂¯ on pseudoconcave compacts and appli-
cations, (see also Erratum: Asian J. Math., vol 7, (2003) No. 1, pp. 147-148), Asian J.
Math. 4 (2000), 855-884.
[Hör1] Hörmander, L., L2 estimates and existence theorems for the ∂¯ operator, Acta Math. 113
(1965), 89-152.
[Hör2] Hörmander, L., The null space of the ∂-Neumann operator, Ann. Inst. Fourier (Grenoble)
54 (2004), 1305-1369.
11
[Hor] ¯
Hortmann, M., Über die Lösbarketi der ∂-Giechung mit Hilfe von Lp , C k , und D -
stetigen Integraloperatoren, Math. Ann. 223 (1976), 139-156.
[Ko1] Kohn, J.J., Harmonic integrals on strongly pseudoconvex manifolds, I, Ann. of Math.
78 (1963), 112-148.
[Ko2] Kohn, J. J., Global regularity for ∂ on weakly pseudoconvex manifolds, Trans. Amer.
Math. Soc., 181 (1973), 273–292.
[Ko3] Kohn, J.J., The range of the tangential Cauchy-Riemann operator, Duke Math. Journ.
53 (1986), 525-545.
[KoR] Kohn, J. J., and Rossi, H., On the extension of holomorphic functions from the boundary
of a complex manifold, Ann. Math., 81 (1965), 451-472.
[LM] Lions, J.-L., and Magenes, E., Non-Homogeneous Boundary Value Problems and Appli-
cations, Volume I, Springer-Verlag, New York, 1972.
[MS] ¯
Michel, J., and Shaw, M.-C., Subelliptic estimates for the ∂-Neumann operator on piece-
wise smooth strictly pseudoconvex domains, Duke Math. J., 93 (1998), 115–128.
[Ra] Range, R. M., Holomorphic Functions and Integral Representations in Several Complex
Variables, Graduate Texts in Math.,Vol.108, Springer-Verlag, N.Y., 1986.
[Sh1] Shaw, M.-C., Global solvability and regularity for ∂ on an annulus between two weakly
pseudoconvex domains, Trans. Amer. Math. Soc., 291 (1985), 255-267.
[Sh2] Shaw, M.-C., L2 estimates and existence theorems for the tangential Cauchy-Riemann
complex., Invent. Math. 82 (1985), 133-150.
[Sh3] Shaw, M.-C., L2 estimates and existence theorems for ∂ b on Lipschitz boundaries, Math.
Zeit. 244 (2003), 91-123.
[Sh4] M.-C. Shaw, The closed range property for ∂ on domains with pseudoconcave boundary,
Proceedings for the Fribourg conference, Trends in Mathematics (2010), 307-320.
Department of Mathematics, University of Notre Dame, Notre Dame, IN 46556

USA
E-mail address: Shaw.1@nd.edu
Volume 550, 2011
On the solvability and hypoellipticity of complex vector

fields
François Treves
Abstract. The article is a survey of recent results about smooth complex

vector fields with critical points (i. e., that vanish at certain points) from the
viewpoint of hypoellipticity and local solvability. The results are rudimentary,
as little is known on the subject so far, outside some relatively simple special
cases. Several open problems are listed (Section 6). A stable class is isolated:
the vector fields of principal type, as well as a subclass (in two dimensions),
introduced under the name of quasi-elliptic vector fields (Section 5).
Contents
1. Generalities about Differential Operators of Principal Type 173
2. Local Solvability of Smooth Vector Fields 175
3. Complex Vector Fields in Two Dimensions 181
4. Complex Vector Fields Hypoelliptic off their Critical Points 185
5. Quasi-Elliptic Vector Fields 188
6. Problems and Possible Directions of Further Research 195
References 195
1. Generalities about Differential Operators of Principal Type

1.1. Basic concepts and notation. Throughout this article M denotes a
smooth (ie, C ∞ ) connected manifold; dim M = n ≥ 2. We use standard notation:
C ∞ (A) stands for the space of smooth functions in the open set A ⊂ M; Cc∞ (A)
for the space of test functions (ie, smooth and compactly supported) in A; D (A)
for its dual, the space of distributions in A. All functions and distributions are
taken to be complex-valued, unless specified otherwise.
As usual T M and T ∗ M stand for the tangent and cotangent bundles of M
respectively, T ∗ M\0 for the complement of the zero section in T ∗ M, π ∗ : T ∗ M −→
M for the base projection. Most of the time we reason in a local coordinate chart
2000 Mathematics Subject Classification. Primary 35A07, Secondary 35F20 .

Key words and phrases. vector fields, local solvability, hypoelliptic, foliations.
2011
c 0000
c Mathematical
holder)
1
173
174
2 FRANÇOIS TREVES
(U, x1 , ..., xn ), in which case ξ1 , ..., ξn stand for the coordinates with respect to the
basis dx1 , ..., dxn in the cotangent spaces at the points of U.
We begin by recalling some terminology and results from the general theory of
linear PDE.
Let P be a linear partial differential operator with complex coefficients of class
C ∞ in M, of order m. In the local chart (U, x1 , ..., xn ),

(1.1) P = P (x, ∂) = cα (x) ∂xα ,
|α|≤m
We are using standard multi-index notation: α = (α1 , ..., αn ), |α| = α1 + · · · + αn ,

∂xα = ∂xα11 · · · ∂xαnn , ∂xj = ∂x
∂
. Where our notation differs from the customary one is
j
√
that here the symbol of ∂xj will be ξj and not −1ξj . There are two reasons for
our choice: 1/we are not going to make use of Fourier transform; 2/ we shall deal
mainly with vector fields and it is convenient that the symbols of real vector fields be
real rather than purely imaginary. Thus the principal symbol of the differential
operator (1.1) in the domain U will be σ (P ) (x, ξ) = |α|=m cα (x) ξ α . Actually
σ (P ) is a well-defined (ie, coordinate free) function in the cotangent bundle of M,
T ∗ M. The characteristic set of P is the null-set of σ (P ) in T ∗ M\0 and will be
denoted by Char P .
If we select a smooth volume density dV in M we can define the transpose
P of P as a smooth differential operator in M by the formula

(1.2) ϕP ψdV = ψP ϕdV , ϕ,ψ ∈ Cc∞ (M) .
It is immediately verified (by reasoning in local coordinates) that

(1.3) σ P (x, ξ) = σ (P ) (x, −ξ)
for all (x, ξ) ∈ T ∗ M
The operator P is said to be elliptic if Char P = ∅; P is said to be of prin-
cipal type if the differential dσ (P ) and the “tautological” one-form τ are linearly
independent at every point of Char P . In the local coordinates x1 , ..., xn we have
τ = ξ · dx = ξ1 dx1 + · · ·+ ξn dxn . The chain rule implies that ξ · dx is invariant under
smooth (or simply C 1 ) coordinate-changes. The differential dτ is the fundamental
symplectic two-form on T ∗ M.
We shall also use the following terminology:
Definition 1. We say that P is of strong principal type in an open subset
A of M if, in every local chart (U, x1 , ..., xn ) with domain U ⊂ A, the differential
of the principal symbol of P with respect to the covariables ξj , dξ σ (P ), does not
−1
vanish at any point of Char P ∩ π ∗ (U).
Strong principal type entails principal type since τ ∧ dξ σ (P ) = 0 if dξ σ (P ) = 0
(and ξ = 0). Ellipticity entails strong principal type. According to (1.3), if P is
of principal type (resp. strong principal type, resp. elliptic) the same is true of its
transpose P .
1.2. Local solvability and hypo-ellipticity of differential operators of
strong principal type. The differential operator P in M is said to be locally
solvable at a point ℘ if there is an open set U
℘ such that to each f ∈ Cc∞ (U)
ON THE SOLVABILITY AND HYPOELLIPTICITY OF COMPLEX VECTOR FIELDS 175
3
there is u ∈ D (U) verifying the equation P u = f in U. One says that P is locally

solvable in an open subset A of M if P is locally solvable at every point of A.
The differential operator P is said to be hypoelliptic in an open set A ⊂ M
if, given any open subset U of A and any distribution u in U, P u ∈ C ∞ (U) =⇒
u ∈ C ∞ (U). Below we say that P is hypoelliptic at a point ℘ ∈ M if L is
hypoelliptic in some open neighborhood of ℘ (then P is hypoelliptic at every point
near ℘). A fundamental result of linear PDE theory is that every elliptic linear
partial differential operator in a smooth manifold M is hypoelliptic in M. Another
useful result is the following ([Treves 1967/2006], Theorem 52.2):
Theorem 1. If P is hypoelliptic in M then its transpose P is locally solvable
in M.
Given a function g ∈ C ∞ (T ∗ M) we denote by Hg the Hamiltonian vector
−1
field of g. “Over” the local chart (U, x1 , ..., xn ), which is to say, in π ∗ (U), we have
n
∂g ∂ ∂g ∂
(1.4) Hg = − .
j=1
∂ξ j ∂x j ∂x j ∂ξj
Assume that g is real-valued. A null bicharacteristic of g is an integral curve

of Hg in T ∗ M\0 on which g vanishes (g is constant along each integral curve of
Hg ). We recall two important results of the general theory of differential operator
of principal type:
Theorem 2. For a differential operator of strong principal type P in M to be
locally solvable in M it is necessary and sufficient that the following condition be
satisfied:
(P): Given an arbitrary ζ ∈ C, the function Im (ζσ(P )) does not change
sign along any null bicharacteristic of Re (ζσ(P )).
Theorem 3. For a differential operator of strong principal type P in M to be
hypoelliptic in M it is necessary and sufficient that Condition (P) be satisfied, as
well as the following condition:
(Q): Given an arbitrary ζ ∈ C, the function Im (ζσ(P )) does not vanish on
any open (nonempty) arc of a null bicharacteristic of Re (ζσ(P )).
Proofs of Theorem 2 can be found in [Hörmander, 1985] (the sufficiency of
the condition was originally proved in [Beals-Fefferman, 1973]). About Theorem
3 see [Hörmander, 1985], [Treves, 1971-b].
From (1.3) and from Theorems 1, 2, 3 we derive
Proposition 1. Let P be a differential operator of strong principal type in
M. If P is locally solvable (resp., hypoelliptic) in M then the same is true of its
transpose P . If P hypoelliptic in M then P is locally solvable in M.
2. Local Solvability of Smooth Vector Fields

2.1. The L-foliation and local solvability. Henceforth we focus on a com-
plex vector field L of class C ∞ in the manifold M. We denote by L the vector
field whose coefficients are the complex conjugates of those of L. We are espe-
cially interested in vector fields L that have critical points, ie, points at which
L = 0. We shall denote by CritL the set of critical points of L. The following
176
4 FRANÇOIS TREVES
claim is practically self-evident: A complex vector field of class C ∞ without critical

points is a (first-order) differential operator of strong principal type (Definition 1).
As a consequence of this and of Theorems 2 and 3 we see that, in any open set
A ⊂ M\CritL, the local solvability, or lack thereof, of L − χ is independent of the
zero-order term χ ∈ C ∞ (A).
By a smooth curve in M we mean the range (often denoted by c) of a smooth
·
map R
t −→ γ (t) ∈ M such that γ (t) = 0 for every t ∈ R (ie, γ is an immersion).
A complex vector field L is said to be tangent to c at a point ℘ = γ (t◦ ) if there is
a complex number ζ = 0 such that ζL|℘ is a real vector tangent to c at ℘. We say
that a smooth curve c is an orbit of L in M if the following is true:
(1) c is without self-intersections;
(2) L is tangent to c at every point of c,
(3) L = 0 at every point of c,
(4) any smooth curve with Properties #2 and #3 that intersects c must be
contained in c.
We shall make use of the Sussman foliation defined by L. Let g (L) denote
the real Lie algebra generated by the vector fields Re L and Im L for the standard
commutation bracket. Let us say that two points of M are L-connectable if they can
be joined by a continuous path consisting of finitely many arcs of orbits of vector
fields belonging to g (L); to be L-connectable is an equivalence relation among points
of M. The main theorem of [Sussman, 1973] states that every equivalence class
for this relation is an immersed submanifold of class C ∞ without self-intersections
(called an L-leaf in the sequel) having the property that the tangent space at every
one of its points contains the “freezing” of g (L) at that point and whose dimension
is minimal for these properties. (In the analytic category the tangent space is equal
to the freezing of g (L) at the point. The L-leaves form what is often called the
Nagano foliation defined by L. See [Nagano, 1966].) The orbits of L are the
one-dimensional L-leaves; the critical points of L are the zero-dimensional L-leaves.
The next statement entails the invariance of Condition (P) (see Theorem 2)
under multiplication by nonvanishing factors, at least for vector fields (for a proof
see [Treves, 1992], Ch. VIII).
Proposition 2. For L to satisfy Condition (P) in an open set A ⊂ M\CritL

it is necessary and sufficient that both the following conditions be satisfied:
(1) the dimension of every L-leaf in A is ≤ 2;
1
(2) 2i L ∧ L does not change sign on any two-dimensional L-leaf in A.
Implicit in this statement is that the two-dimensional L-leaves are orientable,

thus making sense of the property that 2i 1
L ∧ L not change sign on such an L-leaf.
According to Theorem 2 the conjunction of Properties 1 and 2 in Proposition 2 is
equivalent to the local solvability of the differential operator L in A.
2.2. Vector fields of principal type. The linear part. Let the local chart
(U, x1 , ..., xn ) be centered at ℘ ∈ CritL (meaning that xj = 0 at ℘, j = 1, ..., n).
We have, in U,

n
(2.1) L= aj (x) ∂xj
j=1
5
−1
with aj ∈ C ∞ (U) and aj (0) = 0 for all j. In π ∗ (U) the principal symbol of L is
the linear functional with respect to ξ,

n
(2.2) σ (L) (x, ξ) = aj (x) ξj
j=1
and therefore

(2.3) dσ (L) ∧ τ = σj,k (x, ξ) dxj ∧ dxk ,
1≤j<k≤n
where we have used the notation

n
∂ai ∂ai
(2.4) σj,k (x, ξ) = (x) ξk − (x) ξj ξi .
i=1
∂xj ∂xk
Saying that L is of principal type at x ∈ U ∩CritL means that, to each ξ ∈ Rn \ {0},
there is a pair of indices (j, k), 1 ≤ j < k ≤ n, such that σj,k (x, ξ) = 0.
Remark 1. The vector field L cannot be of strong principal type (Definition
1) at any one of its critical points since dξ σ (L) = 0 at such a point.
n if (2.1) is of principal type at x ∈ U ∩ CritL then

Remark 2. Obviously,
necessarily dx σ (L) = j=1 ξj daj = 0 whatever ξ ∈ Rn \ {0}. In other words, the
differentials daj at x must be linearly independent over the field R. In particular,
if the vector field L is real this means that x is an isolated critical point. If the
vector field L is real and of principal type in M then CritL is a discrete set. This
is not necessarily true of a complex vector field (see Example 2 below).
Proposition 3. If L is of principal type at ℘ ∈ CritL then there is an open
neighborhood U of ℘ in M such that L is of principal type at every point of U.
Proof. Let the local chart (U, x1 , ..., xn ) be centered at ℘ ∈ CritL. The
hypothesis is that min |σj,k (0, ξ)| > 0 for some pair of indices j, k. We will then
|ξ|=1
have min |σj,k (x, ξ)| > 0 for every x sufficiently close to 0.
|ξ|=1
It is readily checked that, at critical points of L in U,

n
(2.5) Hσ(L) ∧ ξi ∂ξi = σj,k (x, ξ) ∂ξj ∧ ∂ξk .
i=1 1≤j<k≤n
Comparing with (2.3) shows that, for L to be of principal type at ℘ ∈ CritL it

is necessary and sufficient that the Hamiltonian field Hσ(L) and the radial vector
n ∗
field i=1 ξi ∂ξi tangent to ξ-space T℘ M be linearly independent at every point of
∗
T℘ M ∩ Char L.
Remark
n 3. If all the local coordinates x1 , ..., xn vanish at ℘ theradial vector
n
field i=1 ξi ∂ξi is the Hamiltonian field of the vector field −r∂r = − i=1 xi ∂xi in
T℘ M. It follows that r∂r in R is not of principal type at the origin.
n
As before let the local chart (U, x1 , ..., xn ) be centered at ℘ ∈ CritL. We

introduce the vector field
n
(2.6) L℘ = aj,k xk ∂xj
j,k=1
178
6 FRANÇOIS TREVES
∂a
where aj,k = ∂xkj (0). We might want to view (2.6) as a vector field in U, in which
case L − L℘ vanishes to second order at ℘. From this standpoint changing the
coordinates xj might result in an expression of L℘ whose coefficients are not any
more linear. Not so if we view L℘ as a vector field in T℘ M and if we associate to
a change of coordinates in U the tangent linear transformation in T℘ M. In this
sense the following definition is coordinate-free.
Definition 2. We shall refer to the vector field L℘ in T℘ M as the linear
part of L at the critical point ℘.
We deduce immediately from (2.4) and (2.6):
Proposition 4. For L to be of principal type at a point ℘ ∈ CritL it is
necessary and sufficient that L℘ be of principal type at every point of T℘ M.
Corollary 1. For L to be of principal type in M it is necessary and sufficient
that L℘ be of principal type at every point of T℘ M whatever ℘ ∈ CritL.
Let us continue to reason within the local frame (U, x1 , ..., xn ) centered at the
critical point ℘. Let A (x) stand for the Jacobian matrix with entries ai,j (x) =
∂ai
∂xj (x); A (x) will stand for the transpose of A (x). With this notation we can
rewrite (2.4) as follows:

(2.7) σj,k (x, ξ) = A (x) ξ j ξk − A (x) ξ k ξj .
We see that the quantities σj,k are the components of the 2-covector ξ ∧ A (x) ξ.
We can state
Proposition 5. For L to be of principal type at a critical point x ∈ U it is
necessary and sufficient that the complex matrix A (x) not have any real eigenvec-
tor.
In particular, if L is of principal type at a critical point x ∈ U then the null
space of A (x) in Rn reduces to {0}. This does not mean that det A (x) = 0, as
shown by the example L = z∂z̄ in the plane.
The proofs of the claims in the exemples that follow are left to the reader. They
are direct applications of Proposition 5.
Example 1. The rotation vector field ∂θ = x1 ∂x2 − x2 ∂x1 in R2 is of principal
type.
A complex vector field L can be of principal type and still have a critical set
that is a submanifold of positive dimension, as shown in the following
Example 2. The vector field in R3 ,
(2.8) L = x1 ∂x2 − x2 ∂x1 + ix1 ∂x3
is of principal type in R (by Proposition 5).
3
A vector field L can be of principal type and still have a highly singular critical
set, as shown in the following
Example 3. Let F be an arbitrary closed subset of R and let ϕ ∈ C ∞ (R)
vanish to infinite order on F and not vanish anywhere in the complement of F .
The vector field in R3 ,
(2.9) L = x1 ∂x2 − x2 ∂x1 + (ix1 + ϕ (x3 )) ∂x3
is of principal type in R3 [The linear part of (2.9) at points of F is equal to (2.8).]
7
The set CritL may have singularities even if L is of class C ω and of principal
type:
Example 4. The vector field in R4 ,

L = x1 ∂x2 − x2 ∂x1 + x23 − x34 + ix1 ∂x3 + x23 − x34 − ix2 ∂x4 ,
is of principal type at 0. We have

CritL = x ∈ R4 ; x1 = x2 = x23 − x34 = 0 .

2.3. Local solvability of vector fields with critical points. Results and
conjectures. The next two statements show that for a complex vector field with
critical points Condition (P) is necessary but not sufficient for local solvability.
Proposition 6. Suppose L is of principal type in M. If L satisfies Condition
(P) in M\CritL then L satisfies Condition (P) in M.
Proof. Let ℘ ∈ CritL be arbitrary; then T℘∗ M ⊂ Char L and the Hamil-
tonian field of σ (L) is tangent to T℘∗ M. This implies that any bicharacteristic of
−1
Re (ζσ (L)) (0 = ζ ∈ C) that intersects T℘∗ M is entirely contained in π ∗ (℘) ⊂ T℘∗ M
and σ (L) ≡ 0 on it. The claim ensues.
Example 5. The rotation vector field ∂θ = x1 ∂x2 − x2 ∂x1 is of principal type
in R2 . It satisfies Condition (P) everywhere and yet ∂θ is not locally solvable at 0.
Proposition 7. Let ℘ ∈ CritL be arbitrary and assume that the linear part
L℘ is of principal type in the tangent space T℘ M. If L℘ does not satisfy Condition
(P) in T℘ M then, given an arbitrary open neighborhood U of ℘, L does not satisfy
Condition (P) in U\ {℘} (and therefore L is not locally solvable at ℘).
For a proof of Proposition 7 see [Treves, 2010].
In arbitrary dimension n ≥ 2 the only characterization of local solvability
known to the author concerns vector fields in Euclidean space Rn whose coefficients
are real-valued linear functions. This result is a particular case of Theorem 1,
[Müller, 1992] (for a statement and a proof closer to our present approach see
[Treves, 2009]).
Theorem 4. Consider a real vector field in Rn of the form
n
(2.10) L= aj,k xj ∂xk
j,k=1
( aj,k ∈ R). The following properties are equivalent:

(a): L is not locally solvable at the origin of Rn ;
(b): the closure of each orbit of L in Rn \CritL is compact;
(c): each point x◦ ∈ Rn \CritL lies in a torus Tx◦ ⊂ Rn \CritL such that
every orbit of L that intersects Tx◦ is a geodesic of Tx◦ ;
(d): the real matrix m (L) = (aj,k )1≤j,k≤n is semisimple and its nonzero
eigenvalues are purely imaginary;
(e): the one-parameter subgroup R
t −→ exp tm (L) has compact closure
in GL (n, R).
Corollary 2. Every real vector field (2.10) that is not of principal type is
locally solvable everywhere in Rn .
180
8 FRANÇOIS TREVES
Proof. If (2.10) is not of principal type m (L) must have a real eigenvector
and therefore also a real eigenvalue.
Theorem 4 can be generalized slightly (see [Treves, 2009]):
Theorem 5. Let the the vector field (2.10) have complex coefficients aj,k . If
L ∧ L vanishes identically then the properties (a), (b), (c) in Theorem 4 are
equivalent.
To say that L ∧ L vanishes identically is the same as saying that each L-leaf is
either a critical point or an orbit. It is not the same as saying that L = ζX where
X is a real vector field and 0 = ζ ∈ C. Example: L = (x1 + ix2 ) ∂x1 in R2 .
Note that, when L is the vector field (2.10), CritL is the vector subspace of
Rn defined by the equations nj=1 aj,k xj = 0, k = 1, ..., n. In this connection the
following observation is of interest:
Proposition 8. Suppose that the real vector field (2.10) is of principal type,
in which case CritL = {0}. For L not to be locally solvable at 0 it is necessary and
sufficient that the dimension n be an even number and that each point x◦ ∈ Rn \ {0}
lie in an n-dimensional torus Tx◦ ⊂ Rn \ {0} whose geodesics are orbits of L.
Proof. Suppose (2.10) is of principal type; then CritL = {0} by Remark 2. If
the dimension n were an odd number the real matrix A would perforce have a real
eigenvector, negating the principal type hypothesis on L according to Proposition
5. Every eigenvalue of A must be purely imaginary and thus different from zero,
implying that the torii Tx◦ have maximum dimension n.
Theorem 4 suggests a natural conjecture concerning the local solvability of
complex linear vector fields:
Conjecture 1. Let L be given by (2.10) with complex coefficients aj,k . For L
not to be locally solvable at the origin it is necessary and sufficient that one of the
following conditions be satisfied:
(1) L is not locally solvable in Rn \CritL; or
(2) to each open neighborhood U of 0 there is another open neighborhood V ⊂ U
of 0 such that every L-leaf in V\ (V ∩ CritL) has compact closure con-
tained in U\U ∩ CritL.
The conjecture is true when n = 2, as proved in [Treves, 2009], where a
different necessary and sufficient condition for local solvability is given. We explain
this in the next subsection.
Remark 4. An interesting question is whether there is a group-theoretical for-
mulation of Property 2 in Conjecture 1, of the type: the one-parameter subgroup
R
t → exp tζm (L) has compact closure in GL (n, C) where m (L) = (aj,k )1≤j,k≤n
and 0 = ζ ∈ C is suitably chosen.
In the sequel we refer to vector fields (2.10) with aj,k ∈ C as linear vector
fields. Of course, L need not be linear for the local version of Conjecture 1 to make
sense:
Conjecture 2. Let L be a C ∞ complex vector field in the C ∞ manifold M
and let ℘ ∈ CritL. For L not to be locally solvable at the origin it is necessary and
sufficient that one of the following conditions be satisfied:
9
(1) given any open neighborhood U of ℘, L is not locally solvable in U\U ∩

CritL; or
(2) to each open neighborhood U of ℘ there is another open neighborhood
V ⊂ U of ℘ such that every L-leaf in V\V ∩ CritL has compact closure
contained in U\U ∩ CritL.
Example 6. Let L◦ be a vector field in Rn with constant coefficients and sup-
pose that f ∈ C ω (Rn ) vanishes at the origin (but does not vanish identically); then
f (x) L◦ is locally solvable at 0.
The validity of Conjecture 2 is not even known in the case of linear vector fields
(2.10). It might be advisable to start tackling Conjecture 2 with these.
Contrast Conjecture 2 with Hörmander’s Theorem 26.11 in [Hörmander, 1985],
stating, as a sufficient condition for a partial differential operator of principal type
P (x, D) to be semiglobally solvable, that P (x, D) satisfy the local solvability con-
dition (P) and that the base projections of every one of its “bicharacteristics” (per
force of dimension 1 or 2) escape from every compact subset of the base, ie, not
be “trapped”. It suggest that the conclusion in Theorem 26.11, loc.cit., might be
valid even when trapped bicharaterisctic are present, provided there be at least one
that is not trapped. At any rate it raises some question as to the appropriateness
of Definition 26.11.4, ibid.
3. Complex Vector Fields in Two Dimensions

3.1. Complex vector fields of principal type in 2D. Let L be smooth
complex vector field on a surface S. Let (U, x1 , x2 ) be a local chart in S centered
at a critical point ℘ of L; in U,
(3.1) L = a1 (x) ∂x1 + a2 (x) ∂x2
∞
with aj ∈ C (U), aj (0) = 0. The linear part of L at ℘ is

2
(3.2) L℘ = aj,k xk ∂xj , aj,k ∈ C.
j,k=1

In polar coordinates r = x21 + x22 , θ = arg (x1 + ix2 ) we have
(3.3) L℘ = Q (θ) r∂r − A (θ) ∂θ ,
with the notation
(3.4) Q (θ) = a1,1 cos2 θ + a2,2 sin2 θ + (a1,2 + a2,1 ) cos θ sin θ,
(3.5) A (θ) = a1,2 sin2 θ + (a1,1 − a2,2 ) cos θ sin θ − a2,1 cos2 θ.
Note that both (3.4) and (3.5) are periodic functions of period π; they are related
to each other:
(3.6) 2Q (θ) = div L℘ + ∂θ [A (θ)] .
Going back to the definition of vector fields of principal type we see that, when
n = 2,
(3.7) dσ (L) ∧ τ = σ1,2 (x, ξ) dx1 ∧ dx2 ,
(3.8) σ1,2 (0, ξ1 , ξ2 ) = −a1,2 ξ12 + (a1,1 − a2,2 ) ξ1 ξ2 + a2,1 ξ22 .

182
10 FRANÇOIS TREVES
Comparing (3.8) to (3.5) we see that

(3.9) A (θ) = −σ1,2 (0, − sin θ, cos θ) .
Thus L℘ is of principal type at every point of T℘ S if and only if A (θ) = 0 for all
θ ∈ R. Since we have

(3.10) L = L℘ + O r2
= (Q (θ) + rF (r, θ)) r∂r − (A (θ) + rG (r, θ)) ∂θ
with F and G smooth functions in some cylindrical set [0, ε) × S1 we can state (cf.
Proposition 4):
Proposition 9. Let L be a C ∞ vector field on the surface S given by (3.1) in
the neighborhood U of ℘ ∈ CritL. For L to be of principal type at ℘ ∈ CritL it
is necessary and sufficient that A (θ) = 0 for all θ ∈ R. If L is of principal type
everywhere in S then every critical point ℘ of L is isolated and the origin in T℘ S
is an isolated critical point of L℘ .
3.2. Meziani invariant. Solvability of planar complex linear vector
fields. According to Proposition 9 the principal type property (of L at ℘) is equiva-
lent to the fact that the range of the map [0, π)
θ −→ A (θ) ∈ C is an ellipse E that
does not pass through the origin (but might be reduced to a compact straight-line
segment). This property allows us to introduce the following function of θ ∈ R:
θ
1 dt
(3.11) K (θ) = .
2iπ 0 A (t)
Theorem 6, [Treves, 2009], states that the value of K (π) is independent of the
choice of coordinates in U centered at ℘. The same is true of the quantity
π
1 A (θ) + div L℘
(3.12) μ℘ (L) = dθ,
2iπ 0 A (θ)
first introduced in [Meziani, 2001] to analyze normal forms of planar vector fields
elliptic in the complement of a circle. In [Treves, 2009] μ℘ (L) is called the Meziani
invariant of L℘ ; here we refer to it as the Meziani invariant of L at ℘. Definition
(3.12) and the periodicity of A (θ) implies
(3.13) μ℘ = (div L℘ ) K (π)
1
+ lim (arg A (π − ε) − arg A (0)) .
2π ε−→+0
The main result in [Treves, 2009] (Theorem 6 below) concerns complex linear
vector fields in S = R2 :
2
(3.14) L= aj,k xk ∂xj , aj,k ∈ C.
j,k=1
The critical set of a linear vector field L (assumed not to vanish everywhere) is
either {0}, when L is of principal type, or a straight-line through the origin, when
L is not of principal type. The class of linear vector fields of principal type can
be further subdivided into two distinct subtypes, called (IN-E) and (OUT-E)
respectively in [Treves, 2009], depending on whether the origin in the plane lies
“inside” or “outside” the convex hull of the ellipse E. Again according to Theorem 6,
[Treves, 2009], the linear vector field L is of type (IN-E) if and only if K (π) = 0.
11
It ensues from this and from (3.13) that the Meziani invariant of L is equal to ±1 if L
is of type (IN-E) and to (div L) K (π) if L is of type (OUT-E). In [Treves, 2009]
the value ∞ was assigned to the Meziani invariant of a linear vector field in the
plane that is not of principal type.
Theorem 6. Let L be a linear vector field in the plane which is locally solvable
in R2 \CritL. The following properties are equivalent:
(1): L is not locally solvable at the origin;
(2): L ∧ L vanishes identically and L is not locally solvable at the origin (cf.
Theorem 5);
(3): the origin is an isolated critical point of L and the Meziani invariant
of L at the origin vanishes;
(4): L is of principal type (OUT-E) and div L = 0.
A moment of thought shows that Theorem 6 agrees with Conjecture 1.
Remark 5. An interesting question related to Conjecture 1 in higher dimen-
sions is what should replace the Meziani invariant.
3.3. Cauchy formulas for linear vector fields of principal type in R2 .
We continue to look at a linear vector field (3.14). We assume that L is of principal
type and therefore the origin is the unique critical point of L, to which we can
attach the Meziani number of L, μ (L). We shall make use of the quadratic form
(3.15) (x1 , x2 ) = r 2 A (θ) = −a2,1 x2 + (a1,1 − a2,2 )x1 x2 + a1,2 x2
A 1 2
[cf. (3.5)]. It follows immediately from (3.14) that (L − div L) A = 0. Here we

are interested in the reciprocal A (x1 , x2 ) −1
which, thanks to the principal type
hypothersis, is a C ω function in R2 \ {0}, where it satisfies the equation:

(3.16) L A (x1 , x2 )−1 = − (L + div L) A (x1 , x2 )−1 = 0.
For each R > 0 we define

the following distribution in the open disk ΔR =

(r cos θ, r sin θ) ∈ R2 ; r < R :
2π R
−1 ϕ (r cos θ, r sin θ) − ϕ (0, 0)
(3.17) AR , ϕ = drdθ, ϕ ∈ Cc∞ (ΔR ) .
0 0 rA (θ)
Proposition 10. If L is of principal type (IN-E) then, whatever R > 0, the
−1 is equal to the distribution
distribution A R

−1
A , ϕ = lim (x1 , x2 )−1 ϕ (x1 , x2 ) dx1 dx2 , ϕ ∈ C ∞ R2 .
A c
ε→+0 x21 +x22 >ε2
2π −1
Proof. Let ϕ ∈ Cc∞ (ΔR ). Since 0 A (θ) dθ = 2iπK (π) = 0 if (and only
if) (IN-E) holds (see above) we have
2π ∞
−1
ϕ (r cos θ, r sin θ) r −1 A (θ) drdθ =
0 ε
2π R
ϕ (r cos θ, r sin θ) − ϕ (0, 0)
drdθ,
0 ε rA (θ)
whence the claim by letting r −→ 0.
Let δ denote the Dirac distribution in R2 .
184
Theorem 7. If the linear vector field L is of principal type then

1 −1
(3.18) L A = −μ (L) δ
2iπ R
whetever R > 0.

Proof. Let ϕ ∈ Cc∞ R2 be arbitrary; since (Lϕ) (0, 0) = 0 we have

A−1 , Lϕ = (x1 , x2 )−1 Lϕ (x1 , x2 ) dx1 dx2
A
R
R 2

= lim A (x1 , x2 )−1 Lϕ (x1 , x2 ) dx1 dx2 .
ε−→+0 x21 +x22 >ε2
We compute

(x1 , x2 )−1 Lϕ (x1 , x2 ) dx1 dx2 =
A
x21 +x22 >ε2
2π ∞
A (θ)−1 r −1 Lϕ (r cos θ, r sin θ) drdθ =
0 ε
2π ∞
1 −1
(div L + A (θ)) A (θ) ∂r ϕ (r cos θ, r sin θ) drdθ
2 0 ε
2π ∞
− r −1 ∂θ ϕ (r cos θ, r sin θ) drdθ =
0 ε
2π ∞
1 −1
(div L + A (θ)) A (θ) ∂r ϕ (r cos θ, r sin θ) drdθ =
2 0 ε
2π
1
− (div L + A (θ)) A (θ)−1 ϕ (ε cos θ, ε sin θ) dθ
2 0
after integration by parts. We see that, as ε > 0 converges to zero, the integral

A (x1 , x2 )−1 Lϕ (x1 , x2 ) dx1 dx2
x21 +x22 >ε2
converges to
2π
1
− ϕ (0, 0) (div L + A (θ)) A (θ)−1 dθ,
2 0
whence the claim by (3.13).
∞

To rephrase (3.18): we have, for all ϕ ∈ Cc R , 2

−1 (x1 , x2 )−1 Lϕ (x1 , x2 ) dx1 dx2 .
(3.19) ϕ (0, 0) = lim A
2iπμ (L) ε→+0 x21 +x22 >ε2
Formula (3.19) generalizes the classical inhomogeneous Cauchy formula.

Example 7. The vector field L = e2iθ (r∂r + i∂θ ) = 2z∂z̄ (with z = x1 + ix2 )
satisfies (IN-E); we have A (θ) = −ie2iθ and A (x1 , x2 ) = −i (x1 + ix2 )2 ; μ (L) =
+1. Formula (3.19) is then the standard inhomogeneous Cauchy formula:

1 −1 1
ϕ (0, 0) = lim − z ∂z̄ ϕdx1 dx2 = z −1 (∂z̄ ϕ) dz ∧ dz̄.
ε→+0 π |z|>ε 2iπ R 2
In this case div L = 0 and L = −L.

13
In the sequel we make use of the function

1 A (θ) + div L
(3.20) ω (θ) = Im ,
2 A (θ)
periodic with period 2π. In situations of interest to us it will not have a periodic
antiderivative (see Corollary 7).
Corollary 3. Suppose the linear vector field L is of principal type and let
h ∈ C 1 (Ω) satisfy Lh = 0 in Ω. If ΔR ⊂ Ω (R > 0) then
2π
1
(3.21) h (0, 0) = h (R cos θ, R sin θ) ω (θ) dθ.
2iπμ (L) 0
Proof. Let ψR denote the characteristic function of ΔR . If we put
1
L (ψR h) = (LψR ) h = − R (div L + A (θ)) h (R cos θ, R sin θ) δ (r − R)
2
into (3.19) and take (5.1) into account we get (3.21).
Formula (3.21) generalizes the classical homogeneous Cauchy formula, or, if one
prefers, the Mean Value Theorem.
Example 8. In the case of L = 2z∂z̄ (Example 7) (3.21) is the classical Cauchy
formula.
4. Complex Vector Fields Hypoelliptic off their Critical Points

4.1. Hypoellipticity of smooth vector fields. Special attention to the two-
dimensional case is justified by the following statement.
Proposition 11. Assume the smooth manifold M is connected. If the vector
field L is hypoelliptic in some open subset of M then dim M ≤ 2.
Proof. For L to be hypoelliptic in some open subset of M necessarily L must
be hypoelliptic in some open subset A of M\CritL. Then L must be locally
solvable in A (Proposition 1) and therefore the dimension of the leaves in the
foliation defined by the real vector fields Re L and Im L cannot exceed 2 (Proposition
2). This being the case, there is an open and dense subset A of M\CritL in which
the dimension of the leaves is locally constant (either 1 or 2); A can be covered with
smooth coordinates charts (U, x1 , ..., xn ) in which L = a1 (x) ∂x1 + a2 (x) ∂x2 with
aj ∈ C ∞ (U), j = 1, 2. If n = dim M > 2 any distribution u such that ∂xj u = 0,
j = 1, 2, is a solution of Lu = 0 in U.
Throughout the remainder of this article we assume that dim M = 2.
It has been recently proved by A. Meziani ([Meziani, 2010], Theorem 1) that
no analytic vector field L in a C ω surface S is hypoelliptic at an isolated critical
point. It is easily seen that L cannot be hypoelliptic at nonisolated critical points
(still in the analytic case): when the regular part of the analytic subvariety CritL is
a one-dimensional smooth submanifold R (CritL) it can be covered by coordinate
charts (U, x1 , x2 ) such that U ∩ R (CritL) = {(x1 , x2 ) ∈ U; x2 = 0} implying that,
in U,
(4.1) L = x2 (F (x) ∂x1 + G (x) ∂x2 )
with F , G ∈ C (U). But then L[H (x2 )] = 0 in U, H denoting the Heaviside
ω
function. This proves the following statement:

186
Proposition 12. A C ω analytic vector field L in a C ω surface S is not hypoel-

liptic at any of its critical points.
Corollary 4. No linear vector field (2.10) is hypoelliptic at the origin.
Actually, there is an elementary direct proof of Corollary 4:
Proof. As just seen it suffices to deal with a planar linear vector field. If L is
not of principal type, possibly after a rotation it has the form (4.1) with F and G
complex constants and this case has just been settled.
Now suppose L is of principal type. In polar coordinates L = Q (θ) r∂r −
A (θ) ∂θ , with the notation (3.4)-(3.5) and A (θ) = 0 for all θ ∈ R. When (IN-E)
holds (see end of Subsection 4.2) K (θ) is periodic of period π; we can regard K
as an L∞ function in the plane, obviously not smooth at the origin; it satisfies
LK = −1. When (OUT E) holds, the function log A (θ) can be defined since the
convex hull of the range of A (θ), ie, of the ellipse E, does not contain 0. From (3.6)
we get L(2 log r + log A (θ)) = div L.
Proposition 12 leads naturally to the following
Conjecture 3. A C ∞ complex vector field in a smooth surface S is not hy-
poelliptic at any of its critical points.
Proving Conjecture 3 might be easier than proving Conjecture 2 since local
solvability of L at ℘ does not require n = 2 nor L℘ = 0 (cf. Example 6).
The next statement is a direct consequence of Theorem 3.
Proposition 13. Let L be a C ∞ complex vector field in a smooth surface
S. For L to be hypoelliptic in an open and connected subset A of S\CritL it is
necessary and sufficient that both the following properties hold:
1
(1) 2i L ∧ L does not change sign in A;
(2) there are no orbits of L in any open subset of A.
Proof. Suppose L is given by (3.1) in a smooth coordinate chart (U, x1 , x2 ),
U ⊂ A. Then 2i 1
L ∧ L = Im (a1 ā2 ) ∂x1 ∧ ∂x2 . At least one of the two coefficients,
a1 , a2 , does not vanish at every point of A; suppose that a1 does not vanish at any
point of U. Consider a1 L = X − iY where
X = |a1 |2 ∂x1 + Re (a1 ā2 ) ∂x2 , Y = Im (a1 ā2 ) ∂x2 .
This shows that if L were tangent to a smooth curve c ⊂ U then perforce c would
be an integral curve of X in U and Im (a1 ā2 ) would vanish identically along c, in
other words c would be contained in an orbit of L in U. If L is hypoelliptic in A
then it is locally solvable there, demanding (by Proposition 1) that Im (a1 ā2 ) not
change sign along any integral curve of X in U. If this is the case and provided the
geometry of U is simple, given that the integral curves of X foliate U, for Im (a1 ā2 )
to change sign in U it would have to vanish identically on some integral curve of
X in U. This would contradict the hypoellipticity of L according to the following
easy consequence of Theorem 3: for L to be hypoelliptic in U it is necessary and
sufficient that Im (a1 ā2 ) not change sign in U and not vanish identically along any
integral curve of X in U.
Proposition 14. Let L be a C ∞ complex vector field in a smooth surface S
and let A be an open and connected subset of S\CritL. If L is locally solvable in
A then the following two properties are equivalent:
15
(1) L is not hypoelliptic in any open subset of A;

(2) A is foliated by the orbits of L.
Proof. Easy consequence of Propositions 1 and 2.
Remark 6. It is immediately checked that L is elliptic (ie, Char L = ∅) at a

point ℘ ∈ S if and only if L and L are linearly independent, ie, L ∧ L = 0, at ℘.
For L as in (3.1) and ℘ = (x1 , x2 ) this means that Im (a1 (x1 , x2 ) ā2 (x1 , x2 )) = 0.
Corollary 5. If L is hypoelliptic in the open set A ⊂ S\CritL then the set

of elliptic points of L in A is an open and dense subset of A.
Proof. If 2i 1
L ∧ L were to vanish identically in some open set A ⊂ A then A
would be foliated by the orbits of L in A and L could not be hypoelliptic in A by
Proposition 14.
We relate the hypoellipticity of L in S\CritL to hypocomplexity ([Treves, 1992],

Section III.6).
Proposition 15. Let L be a C ∞ complex vector field in a smooth surface S. If

L is hypoelliptic in S\CritL there is a covering of S\CritL by open sets U having
the following properties:
i: there is a function Z ∈ C ∞ (U) such that LZ = 0 and Z : U −→ C is a
homeomorphism of U onto an open subset of C;
ii: given an arbitrary open set V ⊂ U and an arbitrary solution h of the
homogeneous equation Lh = 0 in V there is a holomorphic function h̃ in
Z (V) such that h = h̃ ◦ Z.
Conversely, the existence of such a covering of S\CritL implies that L is
hypoelliptic in S\CritL.
For a proof see Theorem III.6.3, [Treves, 1992].
Corollary 6. If the vector field L is hypoelliptic in S\CritL then L defines

a complex structure (and therefore an orientation) on S\CritL.
When L is hypoelliptic in S\CritL the standard concepts attached to complex

structures can be introduced. Over S\CritL the complex tangent bundle CT S can
be split as a direct sum of two line bundles, VL , the line bundle spanned by L, and
VL , the line bundle spanned by the complex conjugate, L. This corresponds to the
split of the complex cotangent bundle,
(4.2) CT ∗ S|S\CritL = TL1,0 (S\CritL) ⊕ TL0,1 (S\CritL) ;
TL1,0 (S\CritL) is orthogonal to VL (for the duality between tangent vectors and
covectors) and TL0,1 (S\CritL) is orthogonal to VL . This amounts to viewing L as
(proportional to) the Cauchy-Riemann operator ∂ . The splitting (4.2) yields the
Dolbeault splitting of the cohomology:
(4.3) H 1 (S\CritL) = HL1,0 (S\CritL) ⊕ HL0,1 (S\CritL) .

188
4.2. Vector fields of principal type hypoelliptic off their critical set.

From now on we assume that the two-dimensional manifold S is orientable and
connected. Let (U, x1 , x2 ) be a local chart centered at ℘ ∈ CritL in which L has
the expression (3.1); of course, a1 (0) = a2 (0) = 0. The linear part of L at ℘ has
then the expression (3.2). We also use the polar coordinates expressions (3.3) and
(3.10).
Suppose that L is of principal type at ℘, ie, A (θ) = 0 for all θ; in this case ℘
is an isolated critical point of L (Proposition 9). We derive directly from (3.3):
1
L℘ ∧ L℘ = |A (θ)| ω (θ) ∂θ ∧ r∂r
2
(4.4)
2i
with ω given in (3.20). Below we deal with a primitive ω of ω defined in R.
Proposition 16. For L℘ to be elliptic in R2 \ {0} it is necessary and sufficient
that ω (θ) = 0 for all θ ∈ R.
Proof. Ensues directly from (4.4).

a1,1 a1,2
Remark 7. Ellipticity of L℘ in R2 \ {0} does not preclude that det =
a2,1 a2,2
0. Examples: L = (x1 ± ix2 ) (∂x1 ∓ i∂x2 ).
Proposition 17. For L℘ to be hypoelliptic in R2 \ {0} it is necessary and suf-
ficient that ω not vanish identically and not change sign.
Proof. From (4.4) we derive that if ω ≡ 0 then R2 \ {0} is foliated by the
orbits of L℘ . Proposition 14 enables us to conclude.
Corollary 7. For L℘ to be hypoelliptic in R2 \ {0} it is necessary and suffi-
cient that a primitive ω of (3.20) be strictly monotone in R.
“Strictly” because the zeros of ω are isolated.
Going back to (3.13) we can relate the function (??), or rather its primitive in
R, to the Meziani number μ℘ (L):
(4.5) π−ε
1 1 A (θ)
lim (ω (π − ε) − ω (0)) = Re (K (π) div L℘ )+ Im dθ = Re μ℘ (L) .
π ε−→+0 2π 0 A (θ)
This allows us to state:
Proposition 18. If L℘ is hypoelliptic in R2 \ {0} then Re μ℘ (L) = 0.
5. Quasi-Elliptic Vector Fields

5.1. Quasi-elliptic vector fields. Normal forms.
Proposition 19. Let L be a C ∞ vector field in the orientable surface S and
let ℘ ∈ S be an isolated critical point of L. If L℘ is elliptic in T℘ S\ {0} then there
is an open neighborhood U of ℘ such that L is elliptic in U\ {℘}.
Proof. Let (U, x, x2 ) be a smooth coordinate chart centered at ℘ in which
L has the expression (3.1) in which case L℘ has the expression (3.2). On the one
hand, to say that L℘ is elliptic in T℘ S\ {0} is the same as saying that the quadratic
form
(5.1) Q℘ (x1 , x2 ) = Im ((a1,1 x + a1,2 x2 ) (a2,1 x + a2,2 x2 ))
17
does not vanish at any point of R2 \ {0}. On the other hand, we have
1
(5.2) L ∧ L = Im a1 (x1 , x2 ) a2 (x1 , x2 ) ∂x ∧ ∂x2 .
2i
We derive from the nonvanishing of (5.2) that, for some positive constants c◦ ,
C and all points in U\ {0},
3
(5.3) |Im (a1 (x1 , x2 ) ā2 (x1 , x2 ))| ≥ c◦ x21 + x22 − C x21 + x22 2
whence the claim.
Definition 3. We say that a complex vector field L of class C ∞ in the ori-

entable surface S is quasi-elliptic if L satisfies the following conditions:
(1) L is elliptic in S\CritL;
(2) the linear part L℘ of L at an arbitrary critical point ℘ ∈ S is elliptic in
T℘ S\ {0}.
Proposition 20. A quasi-elliptic vector field in S is of principal type in S and
its critical set CritL is discrete.
Proof. Ensues directly from Proposition 9, Remark 6 and (4.4).
If L is quasi-elliptic then all smooth complex vector fields sufficiently close to

L (in the appropriate sense) are also quasi-elliptic. The vector field
B℘ (r, θ)
(5.4) L = −A (r, θ)−1 L = ∂θ − r∂r
A (r, θ)

is well defined and smooth in an open set Γε = (r, θ) ∈ R × S1 ; r < ε .

We derive from Proposition 18 that if L is quasi-elliptic in S and if ℘ ∈ S
is a critical point of L then Re μ℘ (L) = 0. Keeping this in mind Theorem 1.3
and Lemma 4.2, [Cordaro-Gong, 2004], applied to L (see also Theorem 2.1,
[Meziani, 2001]) enable us to state
Lemma 1. Let the C ∞ vector field L on S be quasi-elliptic and let ℘ ∈ S
be a critical point of L. For each positive integer N there is a C N diffeomorphism
U −→ V, with U and V open neighborhoods of {0}×S1 ⊂ Γε , that preserves {0}×S1
and transforms the vector field (5.4) into
(1) −A (r, θ) (∂θ − iμ℘ (L) r∂r ) if μ℘ (L) ∈ C\Q;

(2) −A (r, θ) ∂θ − i μ℘ (L) + r max(q,2) b (r, θ) r∂r , with r max(q,2) b (r, θ) ∈
C N −1 (V), if μ℘ (L) = pq , with 0 = p ∈ Z and 1 ≤ q ∈ Z+ coprime.
Thus, in studying the vector field L in some suitably small open neighborhood
of the critical point ℘, for many purposes we may assume that L = L◦ where
(5.5) L◦ = ∂θ − iμ℘ (L) r∂r ;

when μ℘ (L) ∈ C\Q. In the cases μ℘ (L) ∈ Q we posit

(5.6) L◦ = ∂θ − i μ℘ (L) + r max(q,2) b (r, θ) r∂r
with r max(q,2) b (r, θ) ∈ C N −1 ((−ε, ε) × S1 ), N < +∞ as large as we wish.

190
Remark 8. Note that the vector fields (5.5) and (5.6) might differ from the
vector fields L in a noteworthy respect: their divergence at the critical point ℘ is
equal to −2iμ℘ (L), possibly different from that of L. The discrepancy lies with the
proportionality factor A (r, θ)−1 . Consider, for instance, the divergence-free planar
vector field z∂z̄ = 12 e2iθ (r∂r + i∂θ ): here A (θ) = 2i
1 2iθ
e ; Property (IN-E) is valid
and μ0 (L◦ ) = 1. The corresponding form (5.6) is r∂r + i∂θ whose divergence is
equal to 2. The example z∂z̄ is also noteworthy for the property that the factor
−1
A (θ) does not even define a continuous function at the origin in R2 .
Proposition 21. Let ℘ ∈ S be a critical point of the quasi-elliptic C ∞ vector
field L. If div L = 0 at ℘ then the linear part L℘ satisfies (IN-E) and μ℘ (L) = ±1.
Proof. By Proposition 20 we know that L is of principal type. Our hypothesis
and (3.13) entail
1
μ℘ (L) = lim (arg A (π − ε) − arg A (0)) .
2π ε−→+0
It follows from Proposition 18 that μ℘ (L) = Re μ℘ (L) = 0. This is only possible if
the winding number of A (θ) at the origin is equal to ±2, whence the claim.
The real 2-vector 2i1

L ∧ L is a C ∞ section of the real line bundle Λ2 T S. Fixing
the orientation of S induces an order in the fibres of Λ2 T S and allows us to speak
of positive or negative elements of these fibres.
Example 9. Let ∞ stand for the North Pole and 0 for the South Pole of the
sphere S2 ⊂ R3 . Suppose the orientation of S2 is defined by the outer normal.
We use the stereographic projection issued from ∞ as a diffeomorphism of S2 \ {∞}
onto the tangent plane T0 S2 and the stereographic projection issued from 0 as a
diffeomorphism of S2 \ {0} onto the tangent plane T∞ S2 . We note that, if x1 and
x2 are the rectangular coordinates on T0 S2 ( resp., T∞ S2 ) inherited from ambient
space R3 then ∂x1 ∧ ∂x2 defines a negative element of Λ2 T0 S2 , whereas it defines a
positive element of Λ2 T∞ S2 . In S2 \ {∞}, we set θ = − arctan (x2 /x) and we define
the vector field L0 = − 2i1
(∂θ − ir∂r ) = z∂z if z = x1 + ix2 = re−iθ . In S2 \ {0}
define the vector field L∞ = − 2i (∂θ + ir∂r ) =
iθ 1
we set z = x1 + ix2 = re and we
−1
z∂z . The map (r, θ) −→ r , θ is an orientation preserving diffeomorphism of
S2 \ ({0} ∪ {∞}) onto itself, transforming L0 and L∞ into each other. Conclusion:
there is a unique smooth vector field L in S2 equal to L0 (resp., L∞ ) in the local
coordinates of the “southern” (resp., “northern”) affine chart. We have, in those
charts,
1
(5.7) L ∧ L = − x21 + x22 ∂x1 ∧ ∂x2 .
2i
Since the 2-vector ∂x1 ∧ ∂x2 is positive in S2 \ {0} we conclude that (5.7) is negative
in S2 \ ({0} ∪ {∞}). From the expressions of L0 and L∞ we derive that μ0 (L) = +1
whereas μ∞ (L) = −1. The two poles make up CritL.
Example 10. Let L be the vector field on the unit sphere S2 ⊂ R3 in Example
9. The vector field Lμ = (μ + 1)L + (μ − 1) L (0 = μ ∈ C) is well-defined in S2 . In
the polar coordinates (r, θ) on S2 \ {∞} we have
1 1
Lμ = − (μ + 1) (∂θ − ir∂r ) + (μ − 1) (∂θ + ir∂r ) = i (∂θ − iμr∂r ) .
2i 2i
19
Comparing with (5.5) we se that the Meziani number of L at 0 is equal to μ. In

the polar coordinates (r, θ) on S2 \ {0} we have
1 1
Lμ = − (μ + 1) (∂θ + ir∂r ) + (μ − 1) (∂θ − ir∂r ) = i (∂θ + iμr∂r ) .
2i 2i
The Meziani number of L at ∞ is equal to −μ. We observe that we can coordinatize
S2 \ {0} with the complex vaiable z = reiθ , to get the expression
Lμ = (μ + 1)z∂z + (μ − 1) z̄∂z̄ .
Consider then the smooth one-form in S2 \ ({0} ∪ {∞}),

dz dz̄
μ = (μ − 1) − (μ + 1) .
z z̄

We have dμ = 0, μ , Lμ = 0 and 2iπ
‘1
= 2μ if c is a circle |z| = r > 0.
c μ
Proposition 22. Let L be a quasi-elliptic C ∞ complex vector field on the

surface S. If the complex structure of S\CritL defined by L is the restriction of a
complex structure on S then necessarily the Meziani index of L at every one of its
critical points is equal to ±1.
Proof. Let ℘ ∈ CritL and suppose that there is an open neighborhood U of

℘ such that there is a function Z ∈ C ∞ (U) satsfying LZ = 0 and dZ ∧ dZ = 0
at every point of U. We can use r = |Z| and θ = arg Z as polar coordinates in
U; we derive that either L℘ = c + be2iθ (r∂r + i∂θ ) with b, c ∈ C, |c| = |b|. If
|c| < |b| then
(IN-E) holds and μ℘ (L) = ±1. If |c| > |b| then (OUT-E) holds and
A (θ) = −i c + be2iθ , div L = 2c. Applying (3.12) we get μ℘ (L) = 1.
Example 11. Take S = S2 , the Riemann sphere, and L = z̄∂z̄ in S2 \ {∞}; L

can be extended to the whole of S2 by setting L = −w̄∂w̄ in S2 \ {0} with w = z −1
in S2 \ ({0} ∪ {∞}). This vector field has two critical points; the complex struc-
ture it defines on S2 \ ({0} ∪ {∞}) extends trivially as the standard structure of the
Riemann sphere.
Remark 9. Concerning the converse of Proposition 22 we note that, if μ℘ (L) =

±1 the normal form in Lemma 1 implies that L is a multiple of

(5.8) L◦ = ∂θ ± i 1 + r 2 b (r, θ) r∂r
in a neighborhood of ℘. Extension of the complex structure defined by L to a full

neighborhood of ℘ would mean that there is a choice of the coordinates r, θ in which
b ≡ 0.
Energy estimates that imply the local solvability (actually, the local L2 -solvability)
of quasi-elliptic vector fields can be proved by exploiting the normal forms in Lemma
1 (see Theorem 6, [Treves, 2010]):
Theorem 8. Let L be a quasi-elliptic vector field in the surface S. Every point

℘ ∈ S has an open neighborhood U℘ with the following property:
(L2 solv): ∀f ∈ L2c (U℘ ), ∃u ∈ L2loc (U℘ ) such that Lu = f in U℘ .
192
5.2. Cauchy formulas and first integrals of a quasi-elliptic vector field

at a critical point. Throughout this subsection U will be a suitably small open
neighborhood of the critical point ℘ in which the coordinates x1 , x2 and r, θ are
well defined, ℘ corresponding to x1 = x2 = r = 0. We shall continue to use the
notation μ = μ℘ (L). We assume that L = −A (r, θ) L◦ with L◦ given by either
−1 ∞
(5.5) or (5.6). Keep in mind that both A (r, θ) and

A (r, θ) belong to L (U).
Below we use the notation Δρ = re ∈ C; r < ρ .
iθ
5.2.1. Case μ℘ (L) ∈ C\Q. In this case we apply Formula (3.19) to the linear
vector field
(5.9) L◦ = −A (r, θ)−1 L = ∂θ − iμr∂r .
[cf. (5.5)]. The function (3.5) for L◦ is A◦ (θ) = −1; we are in the case (OUT-E)
(see remarks preceding Theorem 6) and K (π) = 12 i, div L◦ = −2iμ. We get, for
arbitrary ϕ ∈ Cc∞ (U),

1 drdθ
(5.10) ϕ (0, 0) = − lim L◦ ϕ ,
2πμ ε→+0 r>ε r
equivalent to

1 drdθ
(5.11) ϕ (℘) = lim A (r, θ)−1 Lϕ .
2πiμ ε→+0 r>ε r
One can regard (5.11) as the inhomogeneous Cauchy formula for the vector field L
when μ ∈ C\Q.
Formula (??) yields here the mean value formula
2π
1
(5.12) h (℘) = h (R, θ) dθ
2π 0
if h ∈ C 1 (U) satisfies Lh = L◦ h = 0 in U (assuming ΔR ⊂⊂ U).
Keep in mind that Re μ = 0 when L is quasi-elliptic (by Proposition 18). A
first integral of L is given by
1
r μ exp (iθ) if Re μ > 0,
(5.13) Z℘ (r, θ) =
r − μ exp (−iθ) if Re μ < 0;
1
(5.13) is a Hölder continuous function in U smooth in U\ {℘} that satisfies LZ = 0

and dZ℘ = 0 at every point of U\ {℘}. We can rewrite (5.12) as

1 dZ℘
(5.14) h (℘) = ± h ,
2πi c Z℘
dZ
where ±1 = |Re Re μ ℘
μ| and c = ∂ΔR . The one-form h Z℘ being smooth and closed
in U\ {℘} we can replace c by any smooth closed curve in U\ {℘} winding once,
counterclockwise, around ℘.
5.2.2. Case μ ∈ Q\ {0}. Set μ = pq , with 0 = p ∈ Z and 0 = q ∈ Z+ coprime,
and let (5.6) be valid. The linear part of L◦ is the same as in the case μ ∈ C\Q
but Formula (5.10) is not valid unless b ≡ 0. This requires that we modify the
densities in the Cauchy formulas (5.11)-(5.12) and in the first integral (5.13). This
is achieved thanks to the following

Lemma 2. Suppose that L◦ = ∂θ − i μ + r q b (r, θ) r∂r , with

r q b (r, θ) ∈ C N −1 ((−R, R) × S1 ), N >> q = max (q, 2) .
21
If ε > 0 is sufficiently small there is a function ζ℘ ∈ C (Δε ) such that r∂r ζ℘ ∈ C (Δε )
and

(5.15) L◦ ζ℘ = ir q b, ζ℘ (0) = 0.
Proof. Suppose μ = pq > 0; if μ < 0 one must exchange z and z̄ in the

argument that follows. Since r∂r = z∂z + z̄∂z̄ and i∂θ = z̄∂z̄ − z∂z we have

q q
iL◦ = μ + 1 + |z| b (|z| , θ) z̄∂z̄ + μ − 1 + |z| b (|z| , θ) z∂z .

The equation L◦ ζ℘ = ir q b is equivalent to
(5.16)
q
μ − 1 + |z| b (|z| , θ) 2iθ −1
q
∂z̄ ζ℘ + e ∂z ζ ℘ = − 1 + μ + r b (r, θ) r q −1 eiθ b (r, θ) .
μ + 1 + |z|q b (|z| , θ)
Since q ≥ 2 the partial derivatives of order ≤ 1 of the coefficient C (r, θ) =

μ−1+|z|q b(|z|,θ) 2iθ
1+μ+r q b(r,θ)
e as well as those of the right-hand side in (5.16) are Lipschitz
continuous in U. Assuming r < ε with ε > 0 suitably small, Eq. (5.16) is elliptic.
2
This implies that ζp belongs to the Sobolev space Hloc (U) and therefore that ζ℘
and r∂r ζ℘ are continuous in U (ellipticity also implies that ζ℘ is highly regular in
Δε \ {0}).

Lemma 3. Suppose μ = pq , with p, q ∈ Z\ {0} coprime, L◦ = ∂θ −i μ + r q b (r, θ) r∂r ,

with r q b (r, θ) ∈ C N ((−R, R) × S1 ), N >> q = max (q, 2). If ε > 0 is sufficiently
small then

1 −2
(5.17) L◦ r (1 + r∂r ζ℘ ) = |μ| δ
2iπ
in Δε .
Proof. Let ϕ (r, θ) ∈ Cc∞ (Δε ) be arbitrary (with ε to be chosen later); since
(L◦ ϕ) (0, 0) = 0 we obtain
2π +∞
−2 drdθ
r (1 + r∂r ζ℘ ) , L◦ ϕ = (1 + r∂r ζ℘ ) L◦ ϕ
0 0 r
2π ∞
drdθ
= lim (1 + r∂r ζ℘ ) L◦ ϕ .
ε−→+0 0 ε r
We have
2π ∞ 2π ∞
drdθ drdθ
(1 + r∂r ζ℘ ) L◦ ϕ = (1 + r∂r ζ℘ ) ϕθ
0 ε r 0 ε r
2π ∞

−i μ + r q b (1 + r∂r ζ℘ ) ϕr drdθ =
0 ε
2π ∞ drdθ
− ϕ L◦ (r∂r ζ℘ ) − i (1 + r∂r ζ℘ ) r∂r r q b
0 ε r
2π

+i μ + εq b (ε, θ) (1 + ε∂r ζ℘ (ε, θ)) ϕ (ε, θ) dθ.
0
194
We derive from (5.15):

ir∂r r q b = r∂r (L◦ ζ℘ )

= r∂r ∂θ ζ℘ − ir∂r μ + r q b r∂r ζ℘

= ∂θ r∂r ζ℘ − i μ + r q b (r∂r )2 ζ℘ − ir∂r r q b r∂r ζ℘
whence
L◦ (r∂r ζ℘ ) = i (1 + r∂r ζ℘ ) r∂r r q b .
We obtain
2π ∞
1 drdθ
(5.18) ϕ (0) = lim (1 + r∂r ζ℘ ) L◦ ϕ
2iπμ ε→+0 0 ε r
whereby (5.17) ensues.
Recalling that L◦ = −AL we derive from (5.18) the inhomogenous Cauchy
formula for the vector field L (when 0 = μ ∈ Q):
2π ∞
1 1 + r∂r ζ℘ (r, θ) drdθ
(5.19) ϕ (℘) = lim Lϕ (r, θ)
2πμ ε−→+0 0 ε A (r, θ) r
with ϕ ∈ Cc∞ (Δε ) arbitrary.
Let χR denote the characteristic function of the disk ΔR ⊂⊂ U. If h ∈ C (U) is
a solution of the homogeneous equations Lh = L◦ h = 0 then

L◦ (χR h) = (L◦ χR ) h = iR μ + Rq b(R, θ) h (R, θ) δ (r − R) .
Replacing ϕ by χR h in (5.18) yields
2π
1
(5.20) h (℘) = h (R, θ) 1 + μ−1 Rq b(R, θ) (1 + R∂r ζ℘ (R, θ)) dθ.
2π 0
Assume, for the sake of simplicity, that μ > 0. Here
−1
(5.21) Z℘ (r, θ) = r μ exp iθ + μ−1 ζ℘ (r, θ)
is a first integral since (5.15) entails

(5.22) L◦ Z℘ = iZ℘ − iμ−1 μ + r q b Z℘ + μ−1 (L◦ ζ℘ ) Z℘ = 0.
The continuous function (5.21) is a distribution solution of the homogeneous equa-
tion Lh = 0 in a suitably small open neighborhood of ℘; let now U denote this
neighborhood. We have dZ℘ = 0 at every point of U\ {℘}. We derive once again
from (5.15) and (5.21):
∂θ Z℘

(5.23) = i + μ−1 ∂θ ζ℘ = i 1 + μ−1 r q b (1 + r∂r ζ℘ ) .
Z℘
We conclude that (5.20) is nothing else but the Cauchy formula (5.14) for the choice
(5.21) of Z℘ . If μ < 0 we must replace μ−1 by −μ−1 in (5.21).
The first integrals (5.13) and (5.21) prove that quasi-elliptic vector fields verify
Conjecture 3:
Theorem 9. A quasi-elliptic vector field on the surface S is not hypoelliptic at
any one of its critical points
23
6. Problems and Possible Directions of Further Research

We conclude with a summary of the open questions indicated in the above text
and some additional questions.
6.1. Non-hypoellipticity at critical points. Of the open problems men-
tioned in this article to prove that no C ∞ vector field is hypoelliptic at any one of
its critical points (Conjecture 3) seems the most tractable. Can one devise nor-
mal forms at a point ℘ ∈ CritL under the assumptions that L is hypoelliptic in
the punctured surface M\CritL and that its linear parts L℘ is hypo-elliptic in
T℘ M\ {0}? We know this to be true when “hypoelliptic” is replaced by “elliptic”
(Lemma 1).
6.2. Local solvability at critical points. The first question is whether Con-
jecture 1 is true for linear vector fields in dimension three and higher. If it is true
it should be related to some group theoretical properties (Remark 4). As for the
more general case it seems difficult to tackle, especially in the non-principal type
class (although, in the real linear case, non-principal type implies local solvability
- per Corollary 2.)
6.3. Meziani numbers. What should replace the Meziani number and re-
lated invariants (see Section 3) in dimension three and higher?
6.4. Global properties of quasi-elliptic vector fields. What are the global
properties of the punctured Riemann surface defined by a quasi-elliptic vector field
L?
An intriguing question (with significant implications): Can the Meziani num-
bers be regarded as residues of some (singular) differential one-form?
Can anything be said about the global solvability of a quasi-elliptic vector field?
when M is a compact Riemann surface of low genus, the Riemann sphere or the
torus T2 ? On the subject of global solvability of vector fields without critical points
on T2 see [Bergamasco-Cordaro-Petronilho, 2004] and references therein.
Is there anything special about the moduli of Riemann surfaces defined by
quasi-elliptic vector fields?
6.5. Generalization to pseudodifferential operators of principal type.
What is the correct generalization of the concept of critical point to pseudodiffer-
ential operators? That the principal symbol vanishes identically on T℘∗ M? The
straightforward microlocal generalization: the characteristic variety contains a La-
grangian, does not seem to be adequate. On the local solvability of certain second-
order PDE whose principal symbol vanishes identically on a hypersurface of the
base manifold see [Colombini-Cordaro-Pernazza, 2010].
References
[Beals-Fefferman, 1973] Beals, R. and Fefferman, C. On local solvability of lin-
ear partial differential equations, Ann. of Math. 97
(1973), 482-498.
[Bergamasco-Cordaro-Petronilho, 2004] Bergamasco, A. P., Cordaro P. D. and Petronilho, G.
Global solvability for a Class of Complex Vector Fields
on the Two Torus, Communications in P. D. E. 29
(2004), 785-819.
196
[Colombini-Cordaro-Pernazza, 2010] Colombini, F., Cordaro, P. D. and Pernazza, L. Local

solvability for a class of evolution equations, to appear.
[Cordaro-Gong, 2004] Cordaro, P. D. and Gong, X. Normalization of
complex-valued planar vector fields which degenerate
along a real curve, Advance in Math. 184 (2004), 89-
118.
[Hörmander, 1959] Hörmander, L. On the range of convolution operators,
Ann. Math. 76 (1962), 148-170.
[Hörmander, 1985] Hörmander, L. Linear Partial Differential Operators,
Springer-Verlag Berlin 1969.
[Hörmander, 1985] Hörmander, L. The Analysis of Linear Partial Differ-
ential Equations IV, Springer-Verlag Berlin 1985.
[Lojasiewiccz, 1965] Lojasiewicz, S., Notes, Institut Hautes Études, Bures-
sur-Yvette 1965.
[Meziani, 2001] Meziani, A. On planar elliptic structures with infinite
type degeneracy, J. Funct. Anal. 179 (2001), 333-373.
[Meziani, 2004] Meziani, A. Elliptic vector fields with degeneracies,
Trans. Amer. Math. Soc. 357 (2004) 4225–4248.
[Meziani, 2010] Meziani, A. On the hypoellipticity of differential forms
with isolated singularities, preprint.
[Miwa, 1973] Miwa, T. On the existence of hyperfunction solution
of Linear Differential Equations of the first order with
degenerate real principal symbol, Proc. Japan Acad. 49
(1973), 88-93.
[Müller, 1992] Müller, D. H. Local solvability of first order differential
operators near a critical point, operators with quadratic
symbols and the Heisenberg group, Comm. P. D. E. 17
(1992), 305-337.
[Nagano, 1966] Nagano, T. Linear differential systems with singular-
ities and applications to transitive Lie algebras, J.
Math. Soc. Japan 18 (1966), 398-404.
[Nirenberg-Treves, 1963] Nirenberg, L. and Treves, F. Solvability of a first-order
linear partial differential equation, Comm. Pure Appl.
Math. 16 (1963), 331-351.
[Sussman, 1973] Sussman, H. Orbits of families of vector fields and inte-
grability of distributions, Trans. Amer. Math. Soc. 180
(1973), 171-188.
[Treves 1967/2006] Treves, F. Topological Vector Spaces, Distributions and
Kernels, Academic Press New York 1967, paperback:
Dover Mineola, N. Y., 2006.
[Treves, 1971-a] Treves, F. Analytic-Hypoelliptic Partial Differential
Equations of Principal Type, Comm. Pure Applied
Math. XXIV (1971), 537-570.
[Treves, 1971-b] Treves, F. Hypoelliptic partial differential equations of
principal type. Sufficient conditions and necessary con-
ditions. Comm. Pure Applied Math. . XXIV (1971),
631-670.
[Treves, 1992] Treves, F. Hypo-Analytic Structures, Local Theory,
Princeton University Press, Princeton N. J. 1992.
[Treves, 2009] Treves, F. On the solvability of vector fields with real
linear coefficients, Proceedings A. M. S. 137 (2009),
4209-4218.
[Treves, 2009] Treves, F. On planar vector fields with complex linear
coefficients, Trans. Amer. Math. Soc. (2010), to ap-
pear.
[Treves, 2010] Treves, F. On the local solvability of vector fields with
critical points, to appear.
Titles in This Series
550 Y. Barkatou, S. Berhanu, A. Meziani, R. Meziani, and N. Mir, Editors,
Geometric analysis of several complex variables and related topics, 2011
549 David Blázquez-Sanz, Juan J. Morales-Ruiz, and Rodrı́quez Jesús Lombardero,
Editors, Symmetries and related topics in differential and difference equations, 2011
548 Habib Ammari, Josselin Garnier, Hyeonbae Kang, and Knut Sølna, Editors,
Mathematical and statistical methods for imaging, 2011
547 Krzysztof Jarosz, Editor, Function spaces in modern analysis, 2011
546 Alain Connes, Alexander Gorokhovsky, Matthias Lesch, Markus Pflaum,
and Bahram Rangipour, Editors, Noncommutative geometry and global analysis, 2011
545 Christian Houdré, Michel Ledoux, Emanuel Milman, and Mario Milman,
Editors, Concentration, functional inequalities and isoperimetry, 2011
544 Carina Boyallian, Esther Galina, and Linda Saal, Editors, New developments in Lie
Theory and its applications, 2011
543 Robert S. Doran, Paul J. Sally, Jr., and Loren Spice, Editors, Harmonic analysis
on reductive, p-adic groups, 2011
542 E. Loubeau and S. Montaldo, Editors, Harmonic maps and differential geometry, 2011
541 Abhijit Champanerkar, Oliver Dasbach, Efstratia Kalfagianni, Ilya Kofman,
Walter Neumann, and Neal Stoltzfus, Editors, Interactions between hyperbolic
geometry, quantum topology and number theory, 2011
540 Denis Bonheure, Mabel Cuesta, Enrique J. Lami Dozo, Peter Takáč, Jean Van
Schaftingen, and Michel Willem, Editors, Nonlinear elliptic partial differential
equations, 2011
539 Kurusch Ebrahimi-Fard, Matilde Marcolli, and Walter D. van Suijlekom,
Editors, Combinatorics and physics, 2011
538 José Ignacio Cogolludo-Agustı́n and Eriko Hironaka, Editors, Topology of
algebraic varieties and singularities, 2011
537 César Polcino Milies, Editor, Groups, algebras and applications, 2011
536 Kazem Mahdavi, Debbie Koslover, and Leonard L. Brown III, Editors, Cross
disciplinary advances in quantum computing, 2011
535 Maxim Braverman, Leonid Friedlander, Thomas Kappeler, Peter Kuchment,
Peter Topalov, and Jonathan Weitsman, Editors, Spectral theory and geometric
analysis, 2011
534 Pere Ara, Fernando Lledó, and Francesc Perera, Editors, Aspects of operator
algebras and applications, 2011
533 L. Babinkostova, A. E. Caicedo, S. Geschke, and M. Scheepers, Editors, Set
theory and its applications, 2011
532 Sergiy Kolyada, Yuri Manin, Martin Möller, Pieter Moree, and Thomas Ward,
Editors, Dynamical numbers: Interplay between dynamical systems and number theory,
2010
531 Richard A. Brualdi, Samad Hedayat, Hadi Kharaghani, Gholamreza B.
Khosrovshahi, and Shahriar Shahriari, Editors, Combinatorics and graphs, 2010
530 Vitaly Bergelson, Andreas Blass, Mauro Di Nasso, and Renling Jin, Editors,
Ultrafilters across Mathematics, 2010
529 Robert Sims and Daniel Ueltschi, Editors, Entropy and the Quantum, 2010
528 Alberto Farina and Enrico Valdinoci, Editors, Symmetry for Elliptic PDEs, 2010
527 Ricardo Castaño-Bernard, Yan Soibelman, and Ilia Zharkov, Editors, Mirror
symmetry and tropical geometry, 2010
526 Helge Holden and Kenneth H. Karlsen, Editors, Nonlinear partial differential
equations and hyperbolic wave phenomena, 2010
525 Manuel D. Contreras and Santiago Dı́az-Madrigal, Editors, Five lectures in
complex analysis, 2010
524 Mark L. Lewis, Gabriel Navarro, Donald S. Passman, and Thomas R. Wolf,
Editors, Character theory of finite groups, 2010
TITLES IN THIS SERIES
523 Aiden A. Bruen and David L. Wehlau, Editors, Error-correcting codes, finite
geometries and cryptography, 2010
522 Oscar Garcı́a-Prada, Peter E. Newstead, Luis Álverez-Cónsul, Indranil Biswas,
Steven B. Bradlow, and Tomás L. Gómez, Editors, Vector bundles and complex
geometry, 2010
521 David Kohel and Robert Rolland, Editors, Arithmetic, geometry, cryptography and
coding theory 2009, 2010
520 Manuel E. Lladser, Robert S. Maier, Marni Mishna, and Andrew Rechnitzer,
Editors, Algorithmic probability and combinatorics, 2010
519 Yves Félix, Gregory Lupton, and Samuel B. Smith, Editors, Homotopy theory of
function spaces and related topics, 2010
518 Gary McGuire, Gary L. Mullen, Daniel Panario, and Igor E. Shparlinski,
Editors, Finite fields: Theory and applications, 2010
517 Tewodros Amdeberhan, Luis A. Medina, and Victor H. Moll, Editors, Gems in
experimental mathematics, 2010
516 Marlos A.G. Viana and Henry P. Wynn, Editors, Algebraic methods in statistics
and probability II, 2010
515 Santiago Carrillo Menéndez and José Luis Fernández Pérez, Editors,
Mathematics in finance, 2010
514 Arie Leizarowitz, Boris S. Mordukhovich, Itai Shafrir, and Alexander J.
Zaslavski, Editors, Nonlinear analysis and optimization II, 2010
513 Arie Leizarowitz, Boris S. Mordukhovich, Itai Shafrir, and Alexander J.
Zaslavski, Editors, Nonlinear analysis and optimization I, 2010
512 Albert Fathi, Yong-Geun Oh, and Claude Viterbo, Editors, Symplectic topology
and measure preserving dynamical systems, 2010
511 Luise-Charlotte Kappe, Arturo Magidin, and Robert Fitzgerald Morse, Editors,
Computational group theory and the theory of groups, II, 2010
510 Mario Bonk, Jane Gilman, Howard Masur, Yair Minsky, and Michael Wolf,
Editors, In the Tradition of Ahlfors-Bers, V, 2010
509 Primitivo B. Acosta-Humánez and Francisco Marcellán, Editors, Differential
algebra, complex analysis and orthogonal polynomials, 2010
508 Martin Berz and Khodr Shamseddine, Editors, Advances in p-Adic and
non-archimedean analysis, 2010
507 Jorge Arvesú, Francisco Marcellán, and Andrei Martı́nez-Finkelshtein, Editors,
Recent trends in orthogonal polynomials and approximation theory, 2010
506 Yun Gao, Naihuan Jing, Michael Lau, and Kailash C. Misra, Editors, Quantum
affine algebras, extended affine Lie algebras, and their applications, 2010
505 Patricio Cifuentes, José Garcı́a-Cuerva, Gustavo Garrigós, Eugenio Hernández,
José Marı́a Martell, Javier Parcet, Alberto Ruiz, Fernándo Soria, José Luis
Torrea, and Ana Vargas, Editors, Harmonic analysis and partial differential equations,
2010
504 Christian Ausoni, Kathryn Hess, and Jérôme Scherer, Editors, Alpine
perspectives on algebraic topology, 2009
503 Marcel de Jeu, Sergei Silvestrov, Christian Skau, and Jun Tomiyama, Editors,
Operator structures and dynamical systems, 2009
502 Viviana Ene and Ezra Miller, Editors, Combinatorial Aspects of Commutative
Algebra, 2009
501 Karel Dekimpe, Paul Igodt, and Alain Valette, Editors, Discrete groups and
geometric structures, 2009
For a complete list of titles in this series, visit the

AMS Bookstore at www.ams.org/bookstore/.
This volume contains the proceedings of the Workshop on Geometric Analysis of Several
Complex Variables and Related Topics, which was held from May 10–14, 2010, in
Marrakesh, Morocco.
The articles in this volume present current research and future trends in the theory of
several complex variables and PDE. Of note are two survey articles, the first presenting
recent results on the solvability of complex vector fields with critical points while the
second concerns the Lie group structure of the automorphism groups of CR manifolds.
The other articles feature original research in major topics of analysis dealing with
analytic and Gevrey regularity, existence of distributional traces, the ∂¯-Neumann operator,
automorphisms of hypersurfaces, holomorphic vector bundles, spaces of harmonic forms,
and Gysin sequences.
CONM/550 AMS on the Web

w w w. a m s . o r g

Complex Geometry Analysis Variables

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Complex Geometry Analysis Variables

Uploaded by

Copyright:

Available Formats

CONTEMPORARY

American Mathematical Society

American Mathematical Society

Library of Congress Cataloging-in-Publication Data

Analytic vectors in Locally Integrable Structures

Rafael F. Barostichi, Paulo D. Cordaro, and Gerson Petronilho

Abstract. In this note we introduce the sheaves of Gevrey vectors on a

A similar concept can be introduced for systems of operators.

1991 Mathematics Subject Classiﬁcation. Primary: 35B65; Secondary: 35F50.

diﬀerential of smooth functions. In the bundle T(V) = C ⊗ T∗ Ω/V ⊥ we introduce

Definition 2.1. Let Ω be a smooth manifold over which a locally integrable

function Φ(x, t) = (Φ1 (x, t), . . . , Φm (x, t)) satisfying

Likewise we can identify TN (V)|U to the bundle spanned by dti1 ⊗ · · · ⊗ dtiN ,

We now assume that T(V) is endowed with a smooth hermitian metric h. In

Now by deﬁnition we have, if hij = h(dti , dtj ), 1 ≤ i, j ≤ n,

Proof. We have, by the Cauchy-Schwarz inequality,

Corollary 3.1. Fix an hermitian metric h on T(V). Let u ∈ C(U ). Then

Hence, if u ∈ Gsh (U ; V) and if K ⊂ U is compact, Lemma 3.1 implies the existence

together with (1) imply that u ∈ Gsh (U ; V).

We shall denote by GsV the sheaf Ω

4. Local characterization of the analytic vectors for V

deﬁnes an element v ∈ C(B × Θ• ; O(D)) ∩ L∞ (B × Θ• × D) which satisﬁes the

Given a locally integrable structure V over a smooth manifold Ω, a distribution

where SoV (U )⊗O(D)

5. Analytic vectors in hypocomplex structures

In the proof we shall use the following notation: if W is an open subset of Cm

We apply Lemma 5.1 with F = {u ∈ C(V ) ∩ L∞ (V ) : Lj u = 0, j = 1, . . . , n}.

constant C > 0 and of holomorphic functions gα ∈ O(W) such that vα = gα ◦ Z in

which concludes the proof.

6. Real-analytic locally integrable structures

Definition 6.1. We say that

and thus, noticing that

If λ(1−α)/2 t ∈ supp ζ and λ(1−α)/2 x1 , . . . , λ(1−α)/2 xm ∈ supp ψ then

(8) sup |Lγ (Ψ(λ, x))| ≤ A|γ|+1 |γ|!σ , γ ∈ Zn+

hold. Assuming this for a moment we obtain, from (7),

Write r = λ(1−α)/2 (≥ 1) and w(x ) = rx ,  = 1, . . . , m. In order to estimate

1, . . . , m. For this, we ﬁrst apply the Faà di Bruno’s formula

For 0 < ε < 1/B and c > 0 we set

Then we can write

and therefore we have

If we now introduce the formal power series

Finally, if we observe that

   " #" # " #

inequality (13) allows us to estimate

from which (8) follows. The proof of Proposition 6.3 is complete.

Universidade Federal de São Carlos, São Carlos, SP, Brazil

Universidade de São Paulo, São Paulo, SP, Brazil

Universidade Federal de São Carlos, São Carlos, SP, Brazil

Subellipticity and maximal hypoellipticity for two complex

Makhlouf Derridj and Bernard Helﬀer

Abstract. Our aim is to revisit some aspects of the theory of hypoelliptic

1991 Mathematics Subject Classiﬁcation. Primary 35F05; Secondary 35B65.

in a neighborhood V(0) × Rt of 0 ∈ Rn+1 , where B ∈ C ∞ (V(0)).

The symbol of the system is the map :

(1.6) ||χ(x, t, Dx , Dt )u||21 ≤ ||Lj u||2 + ||u||2 , ∀u ∈ C0∞ ,

where || · ||s denotes the standard Sobolev’s norm.

(1.7) ||u||2s ≤ Cs ||Lj u||2 + ||u||2 , ∀u ∈ C0∞ (V(0)) .

(1.8) ||χ(x, t, Dx , Dt )u||2s ≤ Cs ||Lj u||2 + ||u||2 , ∀u ∈ C0∞ .

two variables x and y.

Part 1. Maximal hypoellipticity

P does not admit a local maximum at (0, 0).

3. Homogeneous case of degree 2

4. The case of homogeneity of degree 3

(4.1) B(s, t, ξ, η) = ξ + γ1 ts2 + γ2 t2 s + t3 η .

Write r = λ(1−α)/2 (≥ 1) and w(x ) = rx , = 1, . . . , m. In order to estimate

" #" # " #

(5.14) (s, t) → s ((δ1 + δ2 γ2 )s + δ2 γ3 t) ,

and the arc is contained in ((−C||

and the arc is contained in ((−C||

(a) If ϕ(0) > 0 and ψ1 (0) < 0,

We also note that σ corresponds to the maximum of φ on [−σ, σ] and that φ is

lim→0 Δ((−σ), (s0 )) = Δ((−σ), (0)) , lim→0 Δ((s0 ), (

Computing the second derivative of φ gives :

|φ (s)| ≤ C|| 2 .

(d) If m ≥ 3, φ has a unique negative minimum at 0. The suitable

Case A : = 0, ϕ(0) > 0.

When m = 0, we have two sectors corresponding to the arcs ] − σ, s0 ()[ and

When m = 2, we consider three sectors attached to the sequence (−σ, 0, s0 (), σ)

When m = 1, there are two cases depending on the sign of ψ(0).

φ(s, t, ) = s4 + (γ1 s3 t + γ2 s2 t2 + γ3 st3 + γ4 t4 ) ,

s− (), s+ ()].

that we will always use in a monotonicity interval of φ, i.e. for s and s in a -