Académique Documents
Professionnel Documents
Culture Documents
Sorbonne Université
École doctorale de sciences mathématiques de
Paris, centre 386
Equipe Analyse Algébrique, IMJ-PRG - UMR7586
Remerciements
Je remercie Pierre Schapira et Julien Grivaux d’avoir accepté de diriger les travaux
ci-exposés et pour leur support constant tout au long de ces deux dernières années. Je
croisai il y a quelques années, bien avant cette thèse, Julien et je l’écoutais évoquer
avec vigueur la force des outils de la cohomologie, tout à fait nébuleux pour moi à
l’époque et dont j’ai pu entrapercevoir le relief à l’occasion de ce travail. J’aurai un
mot tout particulier pour Pierre auprès duquel j’ai pu découvrir un bestiaire d’objects
mathématiques fascinants. Je lui suis profondément reconnaissant de m’avoir ouvert la
porte. Son sens de l’exigence et sa vivacité balistique m’auront définitivement marqué.
J’aurai un mot pour ceux dont j’ai la chance de partager le quotidien à l’Hôpital
Hôtel Dieu: Aı̈da pour ses encouragements sincères à ce que je devienne épidémiologiste,
Raphaël et Isabelle B, les chief of staff, par lesquels je comprenais rapidement le chemin
à parcourir, de quoi vite tempérer l’enthousiasme d’Aı̈da, Stéphanie et sa bienveillante
patience que je ne suis pas (encore) parvenu à éreinter, Isabelle P et son je-ne-sais-quoi
de rassurant, de constant, en fait d’équilatéral, Audrey, et la clarté si particulière de
ses éclats de rires, Elise et son flegme impeccable qui cède aussitôt l’opportunité d’une
blague se présenta, le vif François à l’enthousiasme mathématique communicatif. Tous
ont suivi les tribulations de cette thèse, avec toujours bienveillance. Je remercie aussi
Aurélien Alvarez d’avoir pris le soin de m’orienter alors que je cherchais à donner suite
à une thèse interrompue par le décès de Gennadi, Sami Mustafa et Elisha Falbel pour
leur soutien et leur confiance alors que ma situation administrative a donné lieu à de
nombreuses complications, circonvenues chaque fois à force de leur patience pour ne
pas dire endurance.
auprès desquels je goûte une aventure humaine qui n’a pas d’égal.
J’ai une pensée pour mes parents, ma mère, Sam, Mon, Wil, Karim et Hedi, neveux
et nièces, amis et amies les plus chers qui se reconnaı̂tront immédiatement, qui ont
toujours supporté mes activités, quelque fût absente ma présence.
Olivier, encore, et Frédéric pour leur infrangible amitié, Benjamin C pour sa con-
stante invitation au voyage. Et aussi Guillaume, Bertrand, Thomas, Stephan-Eloı̈se et
Benjamin S.
Résumé
Le sujet de cette thèse est une approche microlocale de la transformation de Radon.
Il s’agit d’appliquer à la dualité projective complexe et réelle les techniques initiées
dans l’article fondateur de Sato-Kashiwara-Kawai de 1972 et de retrouver, reformuler,
améliorer des travaux d’analyse plus classiques sur ce sujet, en particulier ceux de G.
Henkin ou S. Gindikin. La dualité projective vue sous l’angle microlocal et faisceautique
est apparue pour la première fois dans un travail important de J-L. Brylinski sur les
faisceaux pervers, travail repris ensuite par D’Agnolo et Schapira dans le cadre des D-
modules. Notre travail est de reprendre systématiquement cette étude avec les nouveaux
outils de la théorie microlocale des faisceaux (théorie qui n’existait pas à l’époque de
SKK72).
Ce travail se compose essentiellement de deux parties.
Dans la première, nous commençons par rappeler dans un cadre général la con-
struction des transformations canoniques quantifiées, sous l’hypothèse de l’existence
d’une section simple non-dégénérée (introduite sous un autre nom par J. Leray). Cette
construction n’avait jamais été faite dans un cadre global hors du cas projectif. Nous
montrons alors que ces transformations commutent à l’action des opérateurs microd-
ifférentiels. Il s’agit là d’ un résultat fondamental sans qu’aucune preuve consitante
n’existe dans la littérature, ce résultat étant plus ou moins sous-entendu dans SKK72.
La deuxième partie de la thèse traite des applications à la transformation de Radon
“classique”. L’idée de base est que cette transformation échange support des hyper-
fonctions (modulo analyticité) et front d’onde analytique. Nous obtenons ainsi des
théorèmes de prolongement ou d’unicité sur les ouverts linéellement concave. Nous
obtenons aussi un théoréme des résidus pour les valeurs au bord de classes de coho-
mologie définies sur les cones de signatures (1, n − 1), clarifiant substantiellement des
travaux de Cordaro-Gindikin-Trèves.
Mots Clés
Géométrie intégrale, Transformée de Radon, Analyse Algébrique, D-modules, E-modules,
Transformations de Contact, Dualité Projective, Géométrie Symplectique, Lagrangien,
Catégorie Dérivée.
8
Abstract
The subject of this thesis is a microlocal approach to the transformation of Radon. It is
a question of applying to real and complex projective duality the techniques initiated in
the founding article of Sato-Kashiwara-Kawai of 1972 and to find, reformulate, improve
more classic analytical work on this subject, in particular those of G. Henkin or S.
Gindikin. Projective duality seen from the microlocal and sheaf point of view appeared
for the first time in an important work by J-L. Brylinski on perverse sheaves, work
then taken up by D’Agnolo and Schapira in the framework of D-modules. Our work is
to systematically resume this study with the new tools of the microlocal sheaf theory
(theory which did not exist at the time of SKK72).
This work essentially consists of two parts.
In the first, we begin by recalling in a general framework the construction of quan-
tized canonical transformations, under the hypothesis of the existence of a simple non-
degenerate section (introduced under another name by J. Leray). This construction had
never been done in a global framework outside the projective case. We then show that
these transformations exchange the action of the microdifferential operators. This is a
fundamental result without any consistent proof existing in the literature, this result
being more or less implied in SKK72.
The second part of the thesis deals with the applications to the ‘ ‘classical ” Radon
transform. The basic idea is that this transform exchanges the support of hyperfunc-
tions (modulo analyticity) and the analytic wavefront set. We thus obtain theorems of
continuation or uniqueness on linearly concave domain. We also get a residue theorem
for the boundary values of finite cohomology classes defined on cones with (1, n − 1)
signature, substantially clarifying the work of Cordaro-Gindikin-Trèves.
Keywords
Integral geometry, Radon transform, Algebraic Analysis, D-modules, E-modules, Con-
tact transformations, Projective duality, Symplectic geometry, Lagrangian, Derived
Category.
Contents
1 Introduction 11
1.1 Short introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Classical results for Radon transform . . . . . . . . . . . . . . . . . . . 12
1.3 Announcement of the results . . . . . . . . . . . . . . . . . . . . . . . 16
5 Applications 57
5.1 Geometrical preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 An application of the Holmgren-Kashiwara theorem . . . . . . . . . . 60
5.3 Hyperfunctions whose Radon transfom vanishes in linearly concave do-
mains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.4 Radon transform for quadratic cones . . . . . . . . . . . . . . . . . . . 64
6 Appendix 73
6.1 Radon transform of the structure sheaf . . . . . . . . . . . . . . . . . . 73
6.2 Associativity for the composition of µhom . . . . . . . . . . . . . . . . 76
9
10 CONTENTS
1 Introduction 83
1.1 Some facts on the classical Radon transform . . . . . . . . . . . . . . . 83
1.2 The problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2 The case n = 2 95
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.2 Kernel of the Radon transform in strongly linearly concave domain of R2 . 98
2.3 Characterization of the image of the Radon transform . . . . . . . . . 104
2.4 Kernel of Radon transform through ∂¯ technic . . . . . . . . . . . . . . 107
References 141
Chapter 1
Introduction
•
(1.1.2) T ∗S (P × P∗ )
∼ ∼
• y % •
T P ∗ ∼ / T ∗ P∗
11
12 CHAPTER 1. INTRODUCTION
• u ) •
T ∗P P
∼ / T ∗ ∗ P∗
P
The above contact transformation leads to the well-known fact that the Radon trans-
form establishes an isomorphism of sheaves of microfunctions on P and P ∗ (see [KKK86]).
In fact, we will work in the more general framework of integral transforms and then
specialize our results to the case of projective duality. We will consider the following
situation. Consider two complex manifolds X and Y of the same dimension, a closed
• •
submanifold Z of X × Y , open subset U ⊂ T ∗ X and V ⊂ T ∗ Y and assume that the
•
conormal bundle T ∗Z (X × Y ) induces a contact transformation
•
(1.1.4) T ∗Z (X × Y ) ∩ (U × V a )
∼ ∼
•
u ) •
T X⊃U∗ ∼ /V ⊂ T ∗ Y.
Moreover, assume that X and Y are complexification of real analytic manifolds M and
N respectively. Then, it is known that, under suitable hypotheses, one can quantize
∗
this contact transform and get an isomorphism between microfunctions of U ∩TM X and
∗
microfunctions on V ∩ TN Y [KKK86]. One of our main results will be the commutation
of this isomorphism to the action of microdifferential operators. Although considered as
well-known, the proof of this commutation does not appear clearly in the literature (see
[SKK, p. 467]), and is far from being obvious. In fact, we will consider a more general
setting, replacing sheaves of microfunctions with sheaves of the type µhom(F, OX ).
Next, we will specialize our results to the setting of projective duality.
Finally, we will provide some applications of our results to the study of the real
Radon transform of hyperfunctions. Precisely, we will get a Holmgren-type vanishing
theorem describing the vanishing set of a hyperfunction whose wave-front set is subject
to special geometrical conditions. From there, we will recover vanishing theorems of
Kolm-Nagel [KN68] and Boman [Bom92]. Moreover, we will give an answer to a problem
formulated by Henkin [Hen04a], namely the description of the kernel of the real Radon
transform in linearly concave domains. To treat this geometrical situation, a microlocal
treatment will be peculiarly well-suited.
(1.2.1) S Se X ×Y
f g g f q1 q2
| "
X Y, Y X, X Y
14 CHAPTER 1. INTRODUCTION
Se being the image of S by the map r : X × Y → − Y × X, (x, y) 7→ (y, x). Let us fix
some notations that we will detail later.
Let G ∈ Db (CY ) and K ∈ Db (CX×Y ). The integral transform of G with respect to
the kernel K is defined to be
We will denote ΦSe(G) the integral transform of G with respect to the kernel CS [dS −dX ].
For M ∈ Db (DX ) with coherent cohomology, one defines the integral transform of
M,
ΦS (M) := g ∗ f −1 M
ΦS DF ' DG
Radon transform for sheaves and D-modules In the context of projective duality,
we have the following situation
TS∗ (P × P∗ )
p1 p2
x &
T ∗P T ∗ P∗
• •
Denoting by Λ the Lagrangian manifold T ∗S (P × P∗ ) ∩ (T ∗ P × T ∗ P∗ ), one can prove that
• •
p1 |Λ , resp. p2 |Λ , is an isomorphism on T ∗ P, resp. on T ∗ P∗ . Then a theorem of [Bry86]
asserts the equivalence of categories between perverse sheaves on P modulo constant
sheaves and perverse sheaves on P∗ modulo constant sheaves.
It is proven in [DS94] that there is an equivalence of categories between coherent
D-modules on P modulo flat connections and coherent D-modules on P∗ modulo flat
1.2. CLASSICAL RESULTS FOR RADON TRANSFORM 15
connections. At the heart of the proof, fixing an integer k such that −n − 1 < k < 0,
and denoting by OP (k) the −k-th power of the tautological line bundle, the following
isomorphism of D-modules is proven
with k ∗ = −n − 1 − k. From these results,we get the classical result, for F ∈ Db (CP ),
(1.2.5) (U × V a ) ∩ T ∗S (X × Y )
p1 p 2a
u )
∗
U ∩ TM X V ∩ TN∗ Y
Assume that p1 , p2a are isomorphisms. Then it is proven in [KKK86, Theo. 4.2.3]
∗
that there is a local isomorphism between the sheaf of microfunctions on TM X and the
∗
sheaf of microfunctions on TN Y . We will prove in this work a global version of this
isomorphism in the context of real projective duality. We will have to twist the sheaf
of microfunctions by some power of the tautological line bundle.
Following [KS90], let X, Y be two complex manifolds of same dimension n, Λ be
a closed complex Lagrangian submanifold of T ∗ (X × Y ), such that the natural mor-
phisms T ∗ X ← −Λ→ − T ∗ Y are isomorphisms. Let K ∈ Db (CX×Y ) be cohomologically
constructible, simple with shift 0 along Λ and let SS(K) be its micro-support. We
suppose that SS(K) ⊂ Λ.
Let p = (pX , paY ) ∈ Λ and consider a non-degenerate section s ∈ H 0 (µhom(K, ΩX×Y /Y )p ),
where ΩX×Y /Y := OX×Y ⊗q−1 OX q1−1 ΩX . This defines a natural morphism
1
This is, at a germ level, a microlocal version of (1.2.4) with a precision related to
the action of microdifferential operators. The main result of our work will be to give
•
a global version of this result for an open set U ⊂ T ∗ Y . From there, we will deduce
a correspondence result between solutions of systems of microdifferential equations on
the projective space and solutions of systems of microdifferential equations on its dual.
a
We will see that the composition ◦ is associative and we will see also that the morphism
a
(1.3.3) is compatible with associativity with respect to ◦.
Λ⊂U ×Va
∼ ∼
p1
w pa
2 '
• •
T X⊃U∗ ∼ / V ⊂ T ∗Y
χ
and EX the subsheaf of EXR of finite order microdifferential operators. We will prove
our main theorem:
Theorem 1.3.1. Let G ∈ Db (CY ) and assume to be given a section s of µhom(L, ΩX×Y /X ),
non-degenerate on Λ.
(i) For W ⊂ U , P ∈ EX (W ), there is a unique Q ∈ EY (χ(W )) satisfying P · s = s · Q
(P, Q considered as sections of EX×Y ). The morphism induced by s
χ−1 EY |V →
− EX |U
P 7→ Q
is a ring isomorphism.
(1.3.5) ∼
χ−1 µhom(G, OY )|V −→ µhom(ΦL[n] (G), OX )|U
(iii) The isomorphism (1.3.5) is compatible with the action of EY and EX on the left
and right side of (1.3.5) respectively.
18 CHAPTER 1. INTRODUCTION
We will see that the action of microdifferential operators in Theorem 1.3.1 (iii) is
derived from the morphism (1.3.3) given the definition (1.3.4). We refer to (1.2.2) for
the definition of ΦL[n] . The first part of this theorem is well-known, see [SKK], but
the fact that isomorphism (1.3.5) is compatible with the action of microdifferential
operators was announced for microfunctions in various papers but no detailed proof
exists to our knowledge.
AM := OX |M
BM := RH om (D0X CM , OX )
CM := µhom(D0X CM , OX )
W F (u) := supp(sp(u)) ⊂ TU X
All through this thesis, we will quantize the contact transform associated with the
•
Lagrangian submanifold T ∗S (P × P∗ ), where S is the hypersurface of P × P∗ defined by
the incidence relation hξ, xi = 0, (x, ξ) ∈ P × P∗ .
• • •
We denote by T ∗P P, resp. T ∗P ∗ P∗ , the conormal space to P in T ∗ P, resp. to P ∗ in
•
T ∗ P∗ , and we will construct and denote by χ the homogeneous symplectic isomorphism
• •
between T ∗ P and T ∗ P∗ .
For ε ∈ Z/2Z, we denote by CP (ε) the following sheaves: for ε = 0, we set
CP (0) := CP
where q is the 2 : 1 map from the universal cover Pe of P , to P and tr the integration
morphism tr : q! CPe ' q! q ! CP →
− CP .
Let an integer p ∈ Z, ε ∈ Z/2Z, we define the sheaves of real analytic functions,
hyperfunctions on P resp. P ∗ twisted by some power of the tautological line bundle,
and similarly with P ∗ instead of P . We notice that for n odd, DP0 CP ' CP (0) = CP ,
and for n even DP0 CP ' CP (1).
For X, Y either the manifold P or P∗ , for any two integers p, q, we note OX×Y (p, q)
the line bundle on X × Y with homogenity p in the X variable and q in the Y variable.
We set
and we define accordingly EX (p, q). Let us notice that EXR (−p, p) is a sheaf of rings.
Let n be the dimension of P , (of course n = dP ). For an integer k and ε ∈ Z/2Z,
we note
k ∗ := −n − 1 − k
ε∗ := −n − 1 − ε mod(2)
We have:
Theorem 1.3.2. (i) Let k be an integer such that −n − 1 < k < 0 and let s be a
•
global non-degenerate section on T ∗S (P × P∗ ) of H 1 (µhom(CS , ΩP×P∗ /P∗ (−k, k ∗ ))).
For P ∈ EP (−k, k), there is a unique Q ∈ EP∗ (−k ∗ , k ∗ ) satisfying P · s = s · Q.
The morphism induced by s
χ∗ EP (−k, k) →
− EP∗ (−k ∗ , k ∗ )
P 7→ Q
is a ring isomorphism.
In fact, we will see that the non-degenerate section of Theorem 1.3.2 is provided by
the Leray section.
Now, from classical adjunction formulas for E -modules, we get a correspondance
between solutions of systems of microdifferential equations on the projective space and
solutions of systems of microdifferential equations on its dual. We will prove our main
theorem
20 CHAPTER 1. INTRODUCTION
Theorem 1.3.3. Let k be an integer such that −n − 1 < k < 0. Let N be a coherent
EP (−k, k)-module and F ∈ Db (P). Then, we have an isomorphism in Db (CT ∗ P∗ ) •
where ΦµS it is the counterpart of ΦS for E -modules, and will be defined in Section
2.6.1.
Corollary 1.3.4. Let k be an integer such that −n − 1 < k < 0 and ε ∈ Z/2Z. The
section s of theorem 1.3.2 defines an isomorphism:
Moreover, this morphism is compatible with the respective action of χ∗ EP (−k, k) and
EP∗ (−k ∗ , k ∗ ).
Applications
Application I : the wave-front set of hyperfunctions We will obtain a descrip-
tion of the kernel of the Radon transform in linearly concave domain, generalizing some
results announed in [Hen04a]. Consider the diagram:
(1.3.7) P × P∗
q1 q2
{ $
P P∗
P ∗ \ (P
\ \ Ω)
•
T ∗S (P × P ∗ )
p1 pa
2
• •
∼
T ∗P χ T ∗P ∗
ρP PS∗ (P × P ∗ ) ρP ∗
∼ ∼
pe1 pe2
P ∗P P ∗P ∗
P P∗
(Ω × Ω) ∩ S = {(x, ξ) ∈ P × P ∗ ; x ∈ Ω, ξb ⊂ Ω}
V
We set
•
Ω := ρ−1 p1 ((Ω × Ω) ∩ S)) ⊂ T ∗ P
V
P (e
• •
We denote by Ω∗ the set χ(Ω) ⊂ T ∗ P ∗ . We will see that Ω∗ = T ∗ P∗ . V
Ω
Definition 1.3.6. Let U an open subset of P . For a section f of BP (ε, k)(U ), we
define the Radon transform Rad(f ) as the image of f by the sequence of morphisms:
• •
'
Rad : BP (ε, k)(U ) CP (ε, k)(T ∗ U ) CP ∗ (ε∗ , k ∗ )(χ(T ∗ U ))
For f ∈ BP (ε, k)(Ω), we will see that Rad(f )|Ω∗ may be viewed as an element of
BP ∗ (ε∗ , k ∗ )/AP ∗ (ε∗ , k ∗ )(Ω). We will denote this element by Rad(f )| .
V
Ω
As an application of Theorem 1.3.1, we have
Proposition 1.3.7. Let f ∈ BP (ε, k)(Ω). Then, we have
W F (f ) ∩ Ω = χ−1 (W F (Rad(f )| )) V
Ω
In the case where n = 2 and Ω is a strongly linealry concave domain, we denote
by ΩC the union of complex lines, complexification of real lines contained in Ω. Let us
consider an affine chart R2 . Then, there are 2 connected components of (ΩC \ Ω) ∩ R2 ,
which we denote by Ω± . In this condition, we get the following corollary:
Corollary 1.3.8. Assume that W F (Rad(f )| ) = ∅. Then, f is the boundary value of
V
Ω
(f+ , f− ) respectively holomorphic in Ω± .
22 CHAPTER 1. INTRODUCTION
Let V be a real finite dimensional vector space, V ∗ its dual, V its complexification,
•
and set V ∗ = V ∗ \ {0}. Let us denote by S the incidence hypersurface in V × V ∗ . For
an open set U ⊂ V , we set
B(U ) := Γ(U ; BV )
B(K) := lim
→
B(U )
U ⊃K
0 / lim H n−2 (K + iγ 0 ; OA )
Res / B(K) /
→
γ0
/ CA (K + i · Int(λ1◦ )) CA (K + i · Int(λ2◦ )) /
L
0
where γ 0 ranges over the family of open convex cones of signature (1, n − 1), with
γ ⊂ γ 0.
Residue in cones of signature (q, n − q)
Proposition 1.3.11. We have the natural morphism:
This residue was constructed in [CGT95] through several variables complex analysis
technics. Our approach provides a natural construction of this residue.
Residue diagram for Radon Transform
In the following, we see how cones of signature (1, n − 1) in A rise naturally from
the Radon transform of strip-like sets in P∗ and we explicit the diagram 1.3.8 showing
how residues and Radon transforms commutes.
Let δ be a strictly positive real number. Let us consider the open subset U of
P ∗ ,→ P∗ ,
1
X
U := {ξ ∈ A∗ , ( | ξj |2 ) 2 < δ}
2≤j≤n
Let us set
b ∩P
Z := U
K := P \ Z
We will see that U b \ P is a cone of signature (1, n − 1). Let LZ (ε∗ ) be the object of
Db (CP∗ ), defined, up to an isomorphism, by the distinguished triangle:
Theorem 1.3.12. Given a section s of µhom(CS [−1], ΩP×P∗ /P∗ (−k, k ∗ )), the following
diagram commutes:
(1.3.8) Res /
χ∗ H n−2 (µhom(CUb \A (ε), OP (k)))|T ∗ P • χ∗ H n (µhom(CZ (ε), OP (k)))|T ∗ P •
Rad Rad
H n−2 (µhom(ΦS (CUb \A (ε)), OP∗ (k ∗ )))|T ∗ P∗ • / H n (µhom(LZ (ε∗ ), OP∗ (k ∗ )))|T ∗ P∗ •
Res
•
Assuming that s is non-degenerate on T ∗S (P × P∗ ), the vertical arrows Rad are isomor-
phisms.
24 CHAPTER 1. INTRODUCTION
In this chapter, we recall classical results of Algebraic Analysis, with the exception of
section 2.7.
2.1 Sheaves
2.1.1 Notations for manifolds
(i) Let Mi (i = 1, 2, 3) be manifolds. For short, we write Mij :=Mi ×Mj (1 ≤ i, j ≤ 3),
M123 = M1 × M2 × M3 , M1223 = M1 × M2 × M2 × M3 , etc.
(ii) δMi : Mi →
− Mi × Mi denote the diagonal embedding, and ∆Mi the diagonal set of
Mi × Mi .
(iii) We will often write for short ki instead of kMi and k∆i instead of k∆Mi and
similarly with ωMi , etc., and with the index i replaced with several indices ij, etc.
(iv) We denote by πi , πij , etc. the projection T ∗ Mi →
− Mi , T ∗ Mij →
− Mij , etc.
(v) For a fiber bundle E →− M , we denote by Ė →
− M the fiber bundle with the
zero-section removed.
(vi) We denote by qi the projection Mij → − Mi or the projection M123 →− Mi and
by qij the projection M123 →
− Mij . Similarly, we denote by pi the projection
T ∗ Mij →
− T ∗ Mi or the projection T ∗ M123 →
− T ∗ Mi and by pij the projection
T ∗ M123 →
− T ∗ Mij .
(vii) We also need to introduce the maps pj a or pij a , the composition of pj or pij and
the antipodal map a on T ∗ Mj . For example,
p12a ((x1 , x2 , x3 ; ξ1 , ξ2 , ξ3 )) = (x1 , x2 ; ξ1 , −ξ2 ).
25
26CHAPTER 2. REMINDERS ON ALGEBRAIC ANALYSIS AND COMPLEMENTS
2.1.2 Sheaves
We follow the notations of [KS90].
Let X be a good topological space, i.e. separated, locally compact, countable at
infinity, of finite global cohomological dimension and let k be a commutative unital ring
of finite global dimension.
For a locally closed subset Z of X, we denote by kZ , the sheaf, constant on Z with
stalk k, and 0 elsewhere.
We denote by Db (kX ) the bounded derived category of the category of sheaves of
k-modules on X. If R is a sheaf of rings, we denote by Db (R) the bounded derived
category of the category of left R-modules.
Let Y be a good topological space and f a morphism Y → − X. We denote by
L
Rf∗ , f −1 , Rf! , f ! , RH om , ⊗ the six Grothendieck operations. We denote by the ex-
terior tensor product.
⊗−1
We denote by ωX the dualizing complex on X, by ωX the sheaf-inverse of ωX and
by ωY /X the relative dualizing complex
⊗−1
ωY /X := f ! (kX ) ' ωY ⊗ f −1 (ωX )
In the following, we assume that X is a real manifold. Recall that ωX ' orX [dim X]
where orX is the orientation sheaf and dim X is the dimension of X. We denote by
DX ( • ) resp. D0X ( • ) the duality functor
DX ( • ) = RH om ( • , ωX )
D0X ( • ) = RH om ( • , kX )
For F ∈ Db (kX ), we denote by SS(F ) its singular support, also called micro-
support. For a a subset Z ⊂ T ∗ X, we denote by Db (kX ; Z) the localization of the
category Db (kX ) by the full subcategory of objects whose micro-support is contained
in T ∗ X \ Z.
For a closed submanifold M of X, we denote by
νM : Db (kX ) →
− DbR+ (kTM X )
µM : Db (kX ) →
− DbR+ (kTM∗ X )
µM ( • ) = µhom(kM , • )
2.2. O-MODULES AND D-MODULES 27
the inverse of F. Denote by Mod(DX ) the abelian category of left DX -modules and
op
Mod(DX ) of right DX -modules. We denote by ΩX the right DX -module of holomorphic
dX forms. There is an equivalence of category between left and right DX -modules
provided by
op
Db (DX ) →
− Db (DX ), M 7→ ΩX ⊗OX M
op ⊗−1
Db (DX − Db (DX ), N 7→ N ⊗OX ΩX
)→
Let Db (DX ) be the bounded derived category of the category of left DX -modules,
Dbcoh (DX ) its full triangulated subcategory whose objects have coherent cohomology.
Let us recall the notion of good D-modules due to Kashiwara [Kas03].
Definition 2.2.1. An OX -module F is good if for any relatively compact open subset
U ⊂⊂ X, there exists a small and filtrant category I and an inductive system {Fi }i∈I
of coherent OU -modules such that lim ∼
Fi −→ F|U . A coherent DX -module is good if it
→
i
is good as an OX -module.
28CHAPTER 2. REMINDERS ON ALGEBRAIC ANALYSIS AND COMPLEMENTS
Let Dbgood (DX ) be the triangulated subcategory of Db (DX ), whose objects have all
cohomologies consisting in good DX -modules (see [Kas03] for a classical reference).
For M ∈ Dbcoh (DX ), the complex of holomorphic solutions of M is defined by
Sol(M ) := RH om DX (M , OX )
Operations on D-modules
We refer in the following to [Kas03]. Let f : Y →− X be a morphism of complex
−1
manifolds. We denote by DY →
− X the transfer bimodule, i.e. the (DY , f DX )-bimodule
−1
DY →
− X := OY ⊗f −1 OX f DX
We define the pull-back and the direct image of D-modules. Let M ∈ Db (DX ),
N ∈ Db (DY )
L L
f −1 M := DY → −1
− X ⊗f −1 DX f M, f ∗ N := Rf∗ (DX ←
−Y ⊗DY N )
The following results are proven in [KS96, Prop 2.4]. Let M ∈ Dbgood (DX ), N ∈
Db (DY ), F ∈ Db (CX ), G ∈ Db (CY ). We assume that f is non-characteristic for M.
We have
f −1 RH om DX (M, OX ) ' RH om DY (f −1 M, OY ),
2.3 E -modules
We refer in the following to [SKK] (see also [Sch85] for an exposition). For a complex
manifold X, one denotes by EX the sheaf of filtered ring of finite order holomorphic
microdifferential operators on T ∗ X. We denote by Dbcoh (EX ) the full triangulated sub-
category of Db (EX ) whose objects have coherent cohomology.
For m ∈ Z, we denote by EX (m) the abelian subgroup of EX of microdifferential
operators of order less or equal to m. For a section P of EX , we denote by σ(P ) the
principal symbol of P .
− X. Let us recall that EX is flat over
Let πX denote the natural projection T ∗ X →
π −1 (DX ). To a DX -module M , we associate an EX -module defined by
E M := EX ⊗π−1 DX πX
−1
M
X
M ' E M |TX∗ X
To a morphism of manifolds f : Y →
− X, we associate the diagram of natural mor-
phisms:
Tf fτ
(2.3.1) TY / Y ×X T X / TX
τ τX
τY
% f
/X
Y
and the dual diagram
fd fπ
(2.3.2) T ∗Y o Y ×X T ∗ X / T ∗X
π πX
πY
& f
/X
Y
where fd is the transposed of the tangent map T f : T Y →− Y ×X T X.
We denote by EY → − X , EX ←
−Y the transfer bimodules and for M, N objects of re-
spectively D (EX ) and D (EY ), one defines the functors f −1
b b
E
and f E∗ by
L
f −1
E
−1
− X ⊗fπ−1 EX fπ M)
(M) := Rfd ∗ (EY →
L
f E∗ (N ) := Rfπ ∗ (EX ← −1
−Y ⊗f −1 EY fd N ) d
E (f −1 M) ' f −1
E
(E M)
30CHAPTER 2. REMINDERS ON ALGEBRAIC ANALYSIS AND COMPLEMENTS
E (g ∗ N ) ' g E∗ (E N )
BM := RH om (D0X CM , OX )
CM := µhom(D0X CM , OX )
Let us notice that BM and CM are concentrated in degree 0. Let us denote by sp,
the isomorphism
(2.4.1) ∼
sp : BM −→ RπM ∗ CM
BZ|X := H[Z]
d
(OX )
(2.5.1) S Se
f g g f
X Y, Y X
q1 q2
For K ∈ Db (CX×Y ), and given the diagram X ←
− X ×Y −
→ Y , we define the Radon
transform of F with kernel K
ΦK (F ) := Rq2 ! (K ⊗ q1−1 F )
ΦS (M) := g ∗ f −1 M, ΦSe(N ) := f ∗ g −1 N
Let M ∈ Dbgood (DX ), assume that f is non characteristic for M and let G ∈ Db (CY ).
The functorial properties of inverse and direct image of D-modules leads to the following
adjonction formulae, proven in [DS94, Prop. 2.6.],
Let us recall that we denote by ΩX the sheaf of holomorphic n-forms and let
(n,0)
BS|X×Y := q1−1 ΩX ⊗q−1 OX BS|X×Y
1
leading to
(n,0) L
ΦS (M) ' Rq2 ! (BS|X×Y ⊗q1−1 DX q1−1 M)
2.6. MICROLOCAL INTEGRAL TRANSFORMS 33
CS|X×Y := E BS|X×Y
(n,0) L
(2.6.1) CS|X×Y := π −1 q1−1 ΩX ⊗π−1 q1−1 OX CS|X×Y
(n,0) L
(2.6.2) ΦµS (M) ' Rpa2! (CS|X×Y ⊗p−1 p−1 M)
1 EX 1
Let M ∈ Dbgood (DX ). The functors ΦµS and ΦS are linked through the following
isomorphism in Db (CT ∗ Y ), (see [SS94])
•
(2.6.4) U ×Va
p1 p2a
w '
T ∗X ⊃ U V ⊂ T ∗Y
− U and pa2 |Λ : Λ →
(i) p1 |Λ : Λ → − V are isomorphisms
(iii) (p−1 −1
1 (U ) ∪ p2 (V )) ∩ SS(K) ⊂ Λ
(2.7.1) ∼
(K ◦ L) ◦ M −→ K ◦(L ◦ M )
2 3 2 3
(K ◦(L ◦ M )) ◦ N
2 3 4
K ◦((L ◦ M ) ◦ N ) / K ◦(L ◦(M ◦ N )).
2 3 4 2 3 4
2.7. COMPLEMENTS ON THE FUNCTOR µHOM 35
M1234
3
q124 2
q134
M124 23
q14 M134
4
q12 1
q34
2
q14 3
4
q13 q14 1
q24
M123 M14 M234
3 1
q23 4
q23 2
q12 2 3
q34
q13 q24
~ ( v
M12 M13 M23 M24 M34
where the thick squares are cartesian, and where for clarity we enforced the notation:
− Mij by qijk (independently of order of appearence of the indices),
the projection Mijk →
− Mij by qijkl . We now have:
and the projection Mijkl →
3 4 −1 2 3 1 1−1 −1
3 2 34 −1
14 1 −1 −1 −1
Rq14! (q13 (Rq13! (q12 K ⊗ q23 L)) ⊗ q34 M ) ' Rq14! (Rq134! (q12 K ⊗ q23 L) ⊗ q34 M)
23 34 −1 14 −1 12 −1
' Rq14 ! (q12 K ⊗ q23 L ⊗ q34 M )
:= K1 ◦ K2 ◦ K3
2 3
which proves the isomorphism (2.7.1). And, it follows immediately that given N ∈
Db (kM45 ), the diagram (2.7.2) commutes.
∗ : Db (kM12 ) × Db (kM23 ) →
− Db (kM13 )
2
(2.7.4) L
(K1 , K2 ) 7→ K1 ∗ K2 := Rq13 ∗ q2−1 ω2 ⊗ δ2! (K1 K2 ) .
2
36CHAPTER 2. REMINDERS ON ALGEBRAIC ANALYSIS AND COMPLEMENTS
and hence
a
(2.7.6) µhom(F1 , F2 )|U1 ◦ µhom(G1 , G2 )|U2 →
− µhom(F1 ◦ G1 , F2 ◦ G2 )|U3 .
2 2 2
(a)
a
a
∼
µhom(F1 , F2 ) ◦ µhom(G1 , G2 ) ◦ µhom(H1 , H2 ) −→
2 3
a
a
µhom(F1 , F2 ) ◦ µhom(G1 , G2 ) ◦ µhom(H1 , H2 )
2 3
(b) The above isomorphism is compatible with the composition ◦ in the sense that the
following diagram commutes
a a a a
(µhom(F1 , F2 ) ◦ µhom(G1 , G2 )) ◦ µhom(H1 , H2 )
∼ / µhom(F1 , F2 ) ◦(µhom(G1 , G2 ) ◦ µhom(H1 , H2 ))
2 3 2 3
a a
µhom(F1 ◦ G1 , F2 ◦ G2 ) ◦ µhom(H1 , H2 ) µhom(F1 , F2 ) ◦ µhom(G1 ◦ H1 , G2 ◦ H2 )
2 2 3 2 3 3
µhom((F1 ◦ G1 ) ◦ H1 , (F2 ◦ G2 ) ◦ H2 )
∼ / µhom(F1 ◦(G1 ◦ H1 ), F2 ◦(G2 ◦ H2 ))
2 3 2 3 2 3 2 3
Proof. (a) This is a direct application of Lemma 2.7.1 with X, Y, Z taken to be respec-
tively T ∗ M12 , T ∗ M13 , T ∗ M34 .
and
EXR := µhom(C∆X , ΩX×X/X )[dX ]
We recall that EX is a subring of EXR .
We recall the
Lemma 3.1.1. There is a natural morphism
ΩX×Y /X ◦ OY [dY ] →
− OX .
Proof. We have
ΩX×Y /X ◦ OY [dY ] = Rq1 ! (ΩX×Y /X ⊗ q1−1 OY [dY ])
R
→ → OX ,
− Rq1 ! (ΩX×Y /X [dY ]) −
where the last arrow is the integration morphism on complex manifolds.
The following Lemma will be useful for the proof of Lemma 3.1.3. Let us first denote
by Mi (i = 1, 2, 3, 4) four complex manifolds, Li ∈ Db (CMi,i+1 ), 1 ≤ i ≤ 3. We set for
short
(0,d )
di = dimC Mi , dij = di + dj , Ωij/i = ΩMij /Mi = ΩMij j .
37
38 CHAPTER 3. COMPLEX QUANTIZED CONTACT TRANSFORMATIONS
Set for 1 ≤ i ≤ 3,
K1 ◦ K2 ◦ K3
A
w '
e 12 ◦ K3
K / e 123 o
K K1 ◦ K
e 23
B C
K12 ◦ K3 / K123 o K1 ◦ K23
Proof. Diagram labelled A commutes by the associativity of the functor µhom (see
Theorem 2.7.3). Let us prove that Diagram B and C commute. Of course, it is enough
to consider Diagram B. To make the notations easier, we assume that M1 = M4 = pt.
We are reduced to prove the commutativity of the diagram:
For F, F 0 ∈ Db (k12 ), G, G0 ∈ Db (k23 ), we saw in Theorem 2.7.3 (b) that the morphism
µhom(F, F 0 ) ◦ µhom(G, G0 ) → − µhom(F ◦ G, F 0 ◦ G0 ) is functorial in F, F 0 , G, G0 . This
fact applied to the morphism
Ω2 ◦ Ω2,3/2 [d23 ] →
− Ω3 [d3 ]
L00 := L[dY ] ◦ L0
Assume that
and that
Let us set
L := µhom(L, ΩX×Y /X )
CΛ ◦ µhom(G1 , G2 )|V →
− µhom(L ◦ G1 , ΩX×Y /X ◦ G2 )|U
µhom(L ◦ G1 , ΩX×Y /X ◦ OY ) → ∼
− µhom(L ◦ G1 , OX [−dY ]) −→ µhom(L[dY ] ◦ G1 , OX )
(CΛ ◦ CΛ0 ) ◦ H
' / CΛ ◦ (CΛ0 ◦ H )
'
CΛ ◦ CΛ0 ◦ H CΛ ◦ K 0 ◦ H
ΦCΛ (αH (s0 ))
K ◦ K0 ◦ H CΛ ◦ µhom(L0 , ΩY ×Z/Y ) ◦ H
A CΛ ◦ µhom(L0 ◦ H, ΩY ×Z/Y ◦ OZ )
R
Z
α(s·s0 ) !
CΛ ◦ µhom(L0 [dZ ] ◦ H, OY )
R
K ◦ µhom(L , ΩY ×Z/Y ) ◦ H
0 Z
/ K ◦ µhom(L0 [dZ ] ◦ H, OY )
αL0 [d (s)
Z ]◦H
B
R
µhom(L, ΩX×Y /X ) ◦ µhom(L , ΩY ×Z/Y ) ◦ H
0 Z
/ µhom(L, ΩX×Y /X ) ◦ µhom(L0 [dZ ] ◦ H, OY )
C µhom(L ◦ L0 [dZ ] ◦ H, ΩX×Y /X ◦ OY )
R
Y
R
µhom(L ◦ L0 ◦ H, ΩX×Y /X ◦ ΩY ×Z/Y ◦ OZ )
Y,Z
/ µhom(L[dY ] ◦ L0 [dZ ] ◦ H, OX )
We know from Theorem 2.7.2 that the operation ◦ is functorial, so that diagram
42 CHAPTER 3. COMPLEX QUANTIZED CONTACT TRANSFORMATIONS
CΛ ◦ K 0 ◦ H / CΛ ◦ µhom(L0 , ΩY ×Z/Y ) ◦ H ' / CΛ ◦ (µhom(L0 , ΩY ×Z/Y ) ◦ H )
R
Y
CΛ ◦ µhom(L0 [dZ ] ◦ H, OY )
R
K ◦K ◦H0 /K ◦ µhom(L , ΩY ×Z/Y ) ◦ H
0 Y / K ◦ µhom(L0 [dZ ] ◦ H, OY )
CΛ◦Λ0
' / CΛ ◦ CΛ0
α(s·s0 )
'
KU00 ×W a / µhom(L ◦ L0 [dY ], ΩX×Z/X )
Remark 3.1.4. In the following of this thesis, unless necessary, we will omit the sub-
sript for α.
Theorem 3.1.5. Let s ∈ Γ(U × V a ; L ), G ∈ Db (CY ). Then,
(i) s defines a morphism
(ii) With regards to the notation of Lemma 3.1.3, we consider the triplet of manifolds
X, X, Y , Λ = C∆X , L := µhom(C∆X [−n], ΩX×X/X ). Then, the assumption 3.1.2 is
satisfied and noticing that ΦC∆X ' IdX , we conclude by Lemma 3.1.3 that
α(P ) ◦ α(s) ' α(P · s) ' α(s · Q) ' α(s) ◦ ΦCΛ (α(Q))
α(s)
(3.2.3) CΛ ◦ µhom(G, OY )|V −−→ µhom(L[n] ◦ G, OX )|U
α(s)
(3.2.4) χ−1 µhom(G, OY )|V −−→ µhom(ΦL[n] (G), OX )|U
(3.2.5) ∼
χ−1 µhom(G, OY )|V −→ µhom(ΦL[n] (G), OX )|U
Moreover, this isomorphism is compatible with the action of EY and EX on the left and
right side of (3.2.5) respectively.
44 CHAPTER 3. COMPLEX QUANTIZED CONTACT TRANSFORMATIONS
Proof. Let us first prove the following lemma, whose proof is available at the level of
germs in [KS90, Th. 11.4.9].
Let us prove that the morphism (3.2.4) is an isomorphism. Let L∗ be the perverse
sheaf r−1 RH om (L, ωX×Y /Y ) where r is the map X × Y →− Y × X, (x, y) 7→ (y, x). Let
0 ∗
s be a section of µhom(L , ΩY ×X/Y ), non-degenerate on r(Λ), then we apply the same
precedent construction to get a natural morphism
We know from [KS90, Th. 7.2.1] that C∆X ' L∗ ◦ L, so that we get a morphism in
b
D (CV )
α(s0 )
(3.2.6) χ∗ µhom(ΦL[n] (G), OX )|U −−→ µhom(G, OY )|V
We must prove that (3.2.4) and (3.2.6) are inverse to each other. By Lemma 3.1.3(ii),
we get that the composition of these two morphisms is α(s0 · s), with s0 · s ∈ EX .
For any left EX -module M , corresponds a right EX -module ΩX ⊗OX M . Fixing a
non-degenerate form tX of ΩX |U (resp. tY of ΩY |V ), we apply now Theorem 2.3.3: s
and s0 are non-degenerate sections so that (EX×Y |U ×V a , tX ⊗s) and (EY ×X |V a ×U , s0 ⊗tY )
are simple and so isomorphic to (p−1 1 EX |U , 1) ' (p2 EY |V , 1). ΩX resp. ΩY being
a −1
invertible OX -module resp. OY -module, we get as well for the left-right (EX |U , EY |V )
bi-module, resp. left-right (EY |V , EX |U ) bi-module generated by s resp. s0 , that they
are both isomorphic to p−1 1 EX |U ' p2 EY |V .
a −1
Then, following the proof of [KS90, Th. 11.4.9], s and s0 , define ring isomorphisms
associating to each P ∈ EX (U ), P 0 ∈ EX (U ), some Q ∈ EY (V ), Q0 ∈ EY (V ), such
that P · s = s · Q, s0 · P 0 = Q0 · s0 , respectively. Hence, we get that α(s0 ) ◦ α(s) is an
automorphism µhom(G, OY )|V , defined by the left action of s0 · s ∈ EX . Hence, we can
choose s0 so that α(s0 ) ◦ α(s) is the identity.
We are now in a position to prove Theorem 3.2.1: we constructed in the proof of
the lemma, for each P ∈ p−1 1 EX |U , some Q ∈ p2a EY |V such that P · s = s · Q and we
−1
We are going to apply the results of the last chapter to the case of projective duality.
ΦK : Db (CX ) → Db (CY )
F 7→ Rq2! (K ⊗ q1−1 F )
45
46 CHAPTER 4. RADON TRANSFORM FOR SHEAVES
(
V \ {0} / P
? _H o
• •
(4.1.3) T ∗ (V \ {0}) o V \ {0} ×P T ∗ P / T ∗P
where the isomophism between H and H∗ follows from the following symplectic isomor-
phism:
• •
T ∗ (V \ {0}) ' T ∗ (V∗ \ {0})
(x, ξ) 7→ (ξ, −x)
Now, taking the quotient by the action of C∗ on both sides of the isomorphism between
• •
(V \ {0}) ×P T ∗ P and (V∗ \ {0}) ×P∗ T ∗ P∗ , we get the isomorphism:
• •
T ∗ P ' (V∗ \ {0}) ×P∗ P ∗ P∗ ' T ∗ P∗
This gives (4.1.1).
Besides, passing to the quotient by the action of C∗ × C∗ on the two central columns
of diagram (4.1.3), we get (4.1.2).
4.1. PROJECTIVE DUALITY: GEOMETRY 47
We have:
Theorem 4.2.1 ([DS96, p. 469]). Assume −n − 1 < k < 0. There exists a section s of
•
µhom(CS [−1], ΩP×P∗ /P∗ (−k, k ∗ )), non-degenerate on T ∗S (P × P∗ ).
Proof. From the exact sequence:
ω 0 (z)
(4.2.2) s(z, ζ) =
hz, ζin+1+k
Pn
where ω 0 (z) is the Leray form ω 0 (z) = k=0 (−1)
k
zk dz0 ∧ . . . ∧ dzk−1 ∧ dzk+1 ∧ . . . ∧dzn ,
Leray [Ler59].
•
Let s be a section of H 1 (µhom(CS [−1], ΩP×P∗ /P∗ (−k, k ∗ ))), non-degenerate on T ∗S (P×
P∗ ).
4.3. PROJECTIVE DUALITY FOR MICROFUNCTIONS 49
Theorem 4.2.2. Assume −n−1 < k < 0. Then, we have an isomorphism in Db (CT ∗ P ) •
Proof. Let F, G be line bundles on P, and P∗ respectively. We know from [SKK] that
• •
(n,0) ⊗−1
a global non-degenerate section s ∈ Γ(T ∗ P × T ∗ P∗ ; CS|P×P∗ ⊗p−1 EP E F ⊗p−1 EP∗ G E)
1 2
induces an isomorphism of E -modules
Now, let us set F = OP (k), G = OP∗ (k ∗ ). Then, 4.2.1 provides such a non-degenerate
• •
(n,0) ⊗−1
section in Γ(T ∗ P × T ∗ P∗ ; CS|P×P∗ ⊗p−1 EP E OP (k) ⊗p−1 EP∗ OP∗ (k ∗ ) E ). So that, we
1 2
have an isomorphism
(n,0)
On the other hand s is a non-degenerate section of CS|P×P∗ (−k, k ∗ ), hence we can apply
Theorem 2.3.3. Let us denote by
And so
In order to prove Proposition 4.3.3 below, we will use the following theorem proven
in [KS90, Th. 7.2.1], stated inthere for smooth manifolds.
Theorem 4.3.1 ([KS90, Th. 7.2.1]). Let K ∈ Db (P × P∗ ) and assume that
(i) K is cohomologically constructible
(ii) (p−1 a −1
1 (U ) ∪ (p2 ) (V )) ∩ SS(K) ⊂ Λ
50 CHAPTER 4. RADON TRANSFORM FOR SHEAVES
and
CP∗ , for j = n − 1
CP∗ \P ∗ , for j = −1 and n odd
H j (ΦK (CP (0))) '
CP ∗ (1), for j = 0 and n even
0 in any other case
where CP ×P∗ (1) denotes the exterior product CP (1) CP∗ , and C(P ×P∗ )∩S (1) :=
CP ×P∗ (1) ⊗ CS .
Let us compute the stalks of ΦK (CP ). For ξ ∈ P ∗ , we have
RΓc (q2−1 ({ξ}); C(P ×P∗ )∩S ) ' RΓc (q2−1 ({ξ}); C(P ×{ξ})∩S )
' RΓc (Pn ; CPn−1 )
' RΓc (Pn−1 ; CPn−1 )
RΓc (q2−1 ({ξ}); C(P ×P∗ )∩S ) ' RΓc (Pn ; CPn−2 )
' RΓc (Pn−2 ; CPn−2 )
so that
C[0] ⊕ C[n − 1], for ξ ∈ P ∗
ΦK (CP )ξ '
C[n − 1], for ξ ∈ P∗ \ P ∗
so that
C[n − 1], for ξ ∈ P ∗
ΦK (CP )ξ '
C[1] ⊕ C[n − 1], for ξ ∈ P∗ \ P ∗
We recall that the sheaf of CP (1) is defined by the following exact sequence
tr
0 → CP (1) → q! CPe −
→ CP → 0
Pe × P∗
qe1 qe
Pe P × P∗
q q1 q2
j
P P × P∗ P∗
q1 q2
i id
P P∗
where we used that for the closed embedding i, j, the push-forward and proper direct
image are isomorphic. Now, S e ∩ (q2 ◦ j ◦ qe)−1 ({ξ}) are the real (n − 1)-sphere and the
real (n − 2)-sphere, when ξ ∈ P ∗ and ξ ∈ P∗ \ P ∗ , respectively. From the well-known
cohomology of spheres, we get that:
0, for ξ ∈ P ∗
ΦK (CP (1))ξ '
C[1], for ξ ∈ P∗ \ P ∗
For n odd,
C[0], for ξ ∈ P ∗
ΦK (CP (1))ξ '
0, for ξ ∈ P∗ \ P ∗
So that, for n even, ΦK (CP (1)) is a locally constant sheaf, with stalks C on P∗ \ P ∗ ,
concentrated in degree 1, and for n odd, a locally constant sheaf with stalks C on P ∗ ,
concentrated in degree 0.
Now, to conclude, we refer to [DS96, Coro. 5.15], where it is shown in particular that
Hom(ΦK (CP (1)), CP (1)) is isomorphic to 0 for n even and C for n odd. And so, since
there are only 2 locally constant sheaves on P ∗ , namely CP ∗ and CP ∗ (1), we conclude
that for n odd, ΦK (CP (1)) is precisely CP (1). Besides, for n > 2 and even, P∗ \ P ∗ is
contractible, so that ΦK (CP (1)) is precisely CP∗ \P ∗ [1].
Theorem 4.3.3. Assume −n − 1 < k < 0. Recall that any section s ∈ Γ(P ×
(n,0)
P∗ ; BS (−k, k ∗ )), defines a morphism in Db (CT P ) •
•
Assume s is non-degenerate on T ∗S (P×P∗ ). Then (4.3.2) is an isomorphism. More-
over, there exists such a non-degenerate section.
Remark 4.3.4. (i) This is a refinement of a general theorem of [SKK] and is a
microlocal version of Theorem 5.17 in [DS96].
(ii) The classical Radon transform deals with the case where k = −n, k ∗ = −1.
Proof. We will deal with the case ε = 1 and n even, the complementary cases being
proven the same way. Let us apply Theorem 4.3.1 in the following particular case
• • •
- U = T ∗ P, V = T ∗ P∗ , Λ = T ∗S (P × P∗ ).
4.4. MAIN RESULTS 53
- K is CS [n − 1].
- F1 = CP (1) and F2 = OP (k)
K verifies conditions (i),(ii),(iii) of Theorem 4.3.1
(i) is fulfilled as the constant sheaf on a closed submanifold of a manifold is coho-
mologically constructible.
(ii) is fulfilled since SS(CS ) is nothing but TS∗ (P × P∗ ).
(iii) CTS∗ (P×P∗ ) −→ µhom(CS , CS ) is an isomorphism on TS∗ (P × P∗ ) (this follows from
the fact that for a closed submanifold Z of a manifold X, µZ (CZ ) − ∼
→ CTZ∗ X , see [KS90,
Prop. 4.4.3]).
By a fundamental result in [DS96, Th 5.17], we know that for −n − 1 < k < 0,
•
(n,0)
a section s ∈ Γ(P × P∗ ; BS (−k, k ∗ )), non-degenerate on T ∗S (P × P∗ ), induces an
isomorphism
(see Appendix 6.1.1 for a sketch of the proof). Formula (4.2.2) provides an example of
such a non-degenerate section. Hence, applying Lemma 4.3.2, Theorem 4.3.1 gives:
χ∗ µhom(CP (1), OP (k))|T ∗ P ' µhom(CP∗ \P ∗ [1], OP∗ (k ∗ ))|T ∗ P∗
• •
and hence,
L := CS [−1]
54 CHAPTER 4. RADON TRANSFORM FOR SHEAVES
Recall Theorem 4.2.1, and let s be a section of µhom(CS [−1], ΩP×P∗ /P∗ (−k, k ∗ )),
•
non-degenerate on T ∗S (P × P∗ ). We are in situation to apply Theorem 3.2.1.
Theorem 4.4.1. Let G ∈ Db (CP∗ ), k an integer. Assume −n − 1 < k < 0. Then, we
have an isomorphism in Db (CT ∗ P ): •
(4.4.2) ∼
χ−1 µhom(G, OP∗ (k ∗ )) −→ µhom(ΦCS [n−1] (G), OP (k))
This isomorphism is compatible with the action of EP∗ (−k ∗ , k ∗ ) and EP (−k, k) on the
left and right side of (4.4.2) respectively.
Proof. The isomorphism is directly provided by Theorem 3.2.1 in the situation where,
• • •
using the notation inthere, U = T ∗ P, V = T ∗ P∗ and Λ = T ∗S (P × P∗ ) and where we
twist by homogenous line bundles of P, P∗ as explained below.
Let us adapt (3.2.5) by taking into account the twist by homogeneous line bundles.
We follow the exact same reasoning than sections of 3.1 and 3.2.
We have the natural morphism
Indeed, we have
ΩP∗ ×P/P (−k ∗ , k) ◦ OP∗ (k ∗ ) [n] = Rq1 ! (OP∗ ×P (−k ∗ , k) ⊗q−1 OP∗ q2−1 ΩP∗ ⊗ q2−1 OP∗ (k ∗ )[n])
2
R
→ OP (k)
− Rq1 ! (OP∗ ×P (k, 0) ⊗q−1 OP∗ q2−1 ΩP∗ )[n] −
→
2
Given this morphism and considering L := µhom(CS [−1], ΩP∗ ×P/P (−k ∗ , k)), we
• •
mimic the proof of Theorem 3.1.5 so that for a section s of L on T ∗ P × T ∗ P∗a and
• •
for P ∈ Γ(T ∗ P; EP (−k, k)) and Q ∈ Γ(T ∗ P∗ ; EP∗ (−k ∗ , k ∗ )) satisfying P · s = s · Q, the
diagram below commutes:
(4.4.3)
•
From there, given a non-degenerate section of L on T ∗S (P×P∗ ), Theorem 3.2.1 gives
the compatible action of micro-differential operators on each side of the isomorphism
(4.4.2)
Proof. It suffices to prove this statement for finite free EP (−k, k)-modules, which in
turn can be reduced to the case where N = EP (−k, k). By Theorem 4.2.2, we have
χ∗ µhom(F, OP (k))|T ∗ P∗ ' RH om EP∗ (−k∗ ,k∗ ) (EP∗ (−k ∗ , k ∗ ), µhom((ΦCS [n−1] F, OP∗ (k ∗ )))|T ∗ P∗
• •
Applications
b := q2 (q1−1 (A) ∩ S)
A
b := q1 (q −1 (A) ∩ S)
B 2
A := P ∗ \ (P
V
\ \ A)
V
b = {ξ ∈ P ∗ ; {ξ}
(i) A d ∩ A 6= ∅}
V
(ii) A = {ξ ∈ P ∗ ; {ξ}
d ⊂ A}
Proof. (i) By definition, ξ ∈ Ab if and only if there exists x ∈ A, such that (x, ξ) ∈ S,
and so if and only if there exists x ∈ A, such that x ∈ {ξ}.
d
V
57
58 CHAPTER 5. APPLICATIONS
(i) If A 6= ∅, then A = P.
bb
(ii) A \
b = Conv(A).
V
b = Conv(A).
(iii) A
V
b
(iv) A
b = A.
b
(iii) The inclusion Conv(A) ⊂ A b follows immedialtely from (ii) and lemma 5.1.1
(ii). Let us prove the converse inclusion. Let x ∈/ Conv(A) and consider H a separation
∗
hyperplane of {x} and A. There exists ξ ∈ P such that {ξ} d = H. Again by Lemma
5.1.1 (ii), it suffices to prove that {ξ} ∈ d ∩ (P ∗ \ A)
b Indeed, in that case, {x}
/ A. b 6= ∅
V
Remark 5.1.3. (i) Let H∞ be a hyperplane in P . For any non empty affine set
V
A = A.
b
S
f g
X Y
We assume that f and g are smooth and proper. For A ⊂ X, B ⊂ Y , we still note
A := g(f −1 (A)), B
b b := f (g −1 (B)).
Definition 5.1.6. [DS96] Let Z be a locally closed subset of X. One says that Z is
b g −1 ({y}) has trivial cohomology in the sense that
S-trivial if for any y ∈ Z,
(5.1.1) ∼
k−→ RΓ(g −1 ({y}); kf −1 (Z) )
Refering to the notations of section 2.5, recall that for F ∈ Db (CX ), we defined
ΦS (F ) := Rg! f −1 F [dS/Y ]
Proof. We have
Rg! f −1 F ' Rg! (j −1 q1−1 F )
' Rq2 ! (Rj! j −1 )q1−1 F
' Rq2 ! ((q1−1 F )j∗ S )
' Rq2 ! (kj∗ S ⊗ q1−1 F )
Here are some conditions under which the Radon transform of locally constant
sheaves on some locally closed set A ⊂ X, is the constant sheaf on A
b with some shift
in degrees.
60 CHAPTER 5. APPLICATIONS
Lemma 5.1.8 ([DS94, Lem 2.8]). (i) Let U be an S-trivial open subset of X, then
(5.1.2) ΦS (kU ) ' kUb [c − n]
\b
ΓaK := V \ {a}
d ∩K
which is the affine cone, union of hyperplanes passing through a with empty inter-
section with K.
Let A be a subset of V . We define
Γa,b a b
K , the connected component of ΓK ∩ ΓK containing the segment ]a, b[,
•
Z = {(x, iξ) ∈ V × iV ∗; (x, ξ) ∈ S, Hx,ξ ∩ K = ∅}
Corollary 5.2.3. Let ω be an open convex set in V \ K. Let u ∈ B(V ) such that
u|ω = 0. We assume that W F (u) ∩ Z = ∅. Then u vanishes on ConvK (ω).
Proof. This is a direct application of theoreom 5.2.2 where we set Ω = Conv˚K (ω) since
we verify that for any x ∈ ConvK (ω) and (x, iξ) ∈ Z, Hx,ξ ∩ ω = ∅.
5.3.1 Notations
Let us fix some notations. Let k be an integer, ε ∈ Z/2Z. We will assume in this
chapter that 0 < k < −n − 1. For a real analytic manifold M and its complexfication
∗
X, let πM denote the projection TM X → M . Denote by P ∗ M the projectivisation of
•
the cotangent bundle of M , and by ρM the projection T ∗ M → − P ∗M .
As usual, P and P ∗ will denote the real dimension projective space and its dual,
and S the incidence submanifold of P × P ∗ . We denote by pe1 and pe2 the isomorphisms
P ∗ P ' PS∗ (P × P ∗ ) ' P ∗ P ∗ , and we will identify S to PS∗ (P × P ∗ ) (cf. remark following
62 CHAPTER 5. APPLICATIONS
• •
∼
T ∗P χ T ∗P ∗
ρP PS∗ (P × P ∗ ) ρP ∗
∼ ∼
pe1 pe2
P ∗P P ∗P ∗
P P∗
P (e
•
Ω∗ := ρ−1 p2 ((Ω × Ω) ∩ S)) ⊂ T ∗ P ∗
V
P ∗ (e
We notice that
Ω∗ = χ(Ω)
In the following, we will identify i.T ∗ P ' TP∗ P and i.T ∗ P ∗ ' TP∗ ∗ P∗ .
(i) πP (Ω) = Ω
•
(ii) Ω∗ = T ∗ P∗V
Ω
V
Proof. (i) By lemma 5.1.5, we have Ω = Ω, so that for any x ∈ Ω, there exists an
b
V
Remark 5.3.2. Let f ∈ BP (ε, k)(Ω). It follows immediately from (ii) of the above
lemma that Rad(f )|Ω∗ may be viewed as an element of BP ∗ (ε∗ , k ∗ )/AP ∗ (ε∗ , k ∗ )(Ω). We
V
Ω
We have:
Proposition 5.3.3. Let Ω ( P be a non empty linearly concave domain and let f ∈
BP (ε, k)(Ω). Then, we have
W F (f ) ∩ Ω = χ−1 (W F (Rad(f )| ))
V
Proof. For any integer −n − 1 < k < 0, Theorem 4.3.3 gives a sheaf isomorphism
between χ∗ CP (ε, k)|T ∗ P and CP ∗ (ε∗ , k ∗ )|T ∗ P∗ . Thus, we have an isomorphism of C-
• •
P P∗
modules CP (ε, k)|T ∗ P (Ω) ' CP ∗ (ε∗ , k ∗ )|T ∗ P∗ (Ω∗ ). Recalling the isomorphism sp in
• •
P P∗
2.4.1, we have
Ω
boundary value of (f+ , f− ) respectively holomorphic in Ω± .
γ ◦ := {x ∈ A; hx, ui ≥ 0, ∀u ∈ γ}
Let x = (x1 , ..., xn ) be a system of linear coordinates on A, u = (u1 , ..., un ) the dual
coordinates on A∗ . For q an integer such that 0 ≤ q < n, we denote by γq the cone of
signature (q, n − q) in A:
X X
(5.4.1) γq := {(x1 , . . . xn ) ∈ A; x2j < x2j }
1≤j≤q q+1≤j≤n
λq := A \ γq
and
X X
γq∗ := {(u1 , . . . un ) ∈ A∗ ; u2j < u2j }
q+1≤j≤n 1≤j≤q
λ∗q := A∗ \ γq∗
(5.4.2) C{0} →
− Cγq [q + 1]
CA∗ →
− Cλ∗q
Let us apply on both sides the Fourier-Sato functor using [KS97, Lem. 6.2.1], we
get
Cγq / CA / Cλ +1 /
q
we have:
C{0} →
− Cλq [q] →
− Cγq [q + 1]
Remark 5.4.2. The case of cones of signature (1, n − 1) can be handled in a more
constructive way. Indeed, the complementary set λ1 of γ1 is the union of two closed
convex sets, let us say λ11 and λ21 .
We have the two distinguished triangles:
/ / +1 /
L
(5.4.3) Cλ11 ∪λ21 Cλ11 Cλ21 C{0}
/ / CW +1 /
L
CW +i(λ11 ∪λ21 ) CW +iλ11 CW +iλ21
Applying the functor RHom (•, OA ) to this distinguished triangle, we get the long
exact sequence:
n / Hn n / Hn /
L
HW (A; OA ) W +iλ1 (A; OA ) HW +iλ2 (A; OA ) W +i(λ1 ∪λ2 ) (A; OA ) ···
From the distinguished triangle (5.4.4), and taking into account that W + iA is
convex, RH om (CW +iA , OA ) ' OW +iA is concentrated in degree 0, and we get the
isomorphism
∼
H n−2 (W + iγ; OA ) − n−1
→ HW
(5.4.8) +i(λ1 ∪λ2 ) (A; OA )
and
∼
H n−1 (W + iγ; OA ) − n
→ HW +i(λ1 ∪λ2 ) (A; OA )
Noticing that RΓW +iλi (OA ), i = 1, 2, is concentrated in degree n, and applying 5.4.8,
5.4.9 to 5.4.7, we get the exact sequence (5.4.6)
Theorem 5.4.4. Let K be a compact convex subset of A. There exists a natural exact
sequence
0 / lim H n−2 (K + iγ 0 ; OA )
Res / B(K) /
→
γ0
CA (K + i · Int(λ1◦ )) CA (K + i · Int(λ2◦ )) /
L
0
where γ 0 ranges over the open convex cones of signature (1, n − 1), with γ ⊂ γ 0 .
Proof. Let W be an open neighborhood of K in A.
Following [KS90, Th.4.3.2], we know that for an open convex cone V ⊂ T ∗W A, we
have the isomorphism
CA (V ) ' lim
→
n
HZ∩U (U ; OA )
Z,U
where Z and U range over respectively, the closed sets whose normal cone is contained
in the polar set of V , and the open subsets U of A such that π(V ) = U ∩ W , π being
the natural projection of T ∗ A onto A.
5.4. RADON TRANSFORM FOR QUADRATIC CONES 67
n n /
L
H(W +i.λ1 0 )∩U
(A; OA ) H(W +i.λ2 0 )∩U
(A; OA ) 0
CA (K + i.Int(λ1◦ )) CA (K + i.Int(λ2◦ )) /
L
0
Indeed,
CP (ε)[−n] →
− F
Rad Rad
µhom(ΦS (F ), OP∗ (k ∗ ))|T ∗ P∗ • / CP ∗ (ε∗ , k ∗ )| •
Res T ∗ P∗
•
Assume s is non-degenerate on T ∗S (P × P∗ ), then the vertical arrows Rad above are
isomorphisms.
Throughout the rest of this section, we will set s the section of 4.2.1, which provides
such a non-degenerate section.
We recall we denoted by A an affine chart of P .
Proposition 5.4.6. We have the natural morphism:
Remark 5.4.7. The residue 5.4.12 was constructed in [CGT95] through several vari-
ables complex analysis technics. Our approach provides a natural construction of this
residue.
Setting the geometric context Let δ > 0 be a real number. Let A be the affine
chart relative to {(z0 : z1 : . . . : zn ) ∈ P; z1 = 1}, and A∗ the affine chart related to
{(ξ0 : ξ1 : . . . : ξn ) ∈ P∗ ; ξ0 = 1}. We set:
H := {(z0 :Az∗1 =
: . A. .∗ :∩zP ∗
n ) ∈ P; z1 = 0}
H := {(x0 : x1 : . . . : xn ) ∈ P ; x1 = 0}
We will consider the following sets
1
X
(5.4.13) U := {ξ ∈ A∗ ; ( ξj2 ) 2 < δ} ,→ P∗
2≤j≤n
(5.4.14) b ∩P
Z := U
We have
5.4. RADON TRANSFORM FOR QUADRATIC CONES 69
(i) U is S-trivial,
Proof. (i) is related to a more general and obvious fact: convex sets, either open or
closed, are S-trivial.
b \H. There exists ξ ∈ U such that hz, ξi = 0, so that z0 +ξ1 + P zj ξj =
(ii) Let z ∈ U
2≤j≤n
0. Taking the imaginary part, and denoting y = Im(z), we have
1
X X
(5.4.15) | y0 |= | yj ξj |< δ( yj2 ) 2
2≤j≤n 2≤j≤n
We have
b \ H = A + iγ
U
The Residue diagram Let ε ∈ Z/2Z. We recall the defintion of Z by 5.4.14, and
K in 5.4.8 (iii).
CP ∗ (ε∗ ) →
− CKb [n − 1]
•
we have in Db (CP∗ ; T ∗ P∗ )
Applying Lemma 4.3.2 and Lemma 5.1.8, we get the distinguished triangle in
•
D (CP∗ ; T ∗ P∗ ):
b
(5.4.17)
χ∗ H n−2 (µhom(CUb \A (ε), OP (k)))|T ∗ P •
Res / χ∗ H n (µhom(CZ (ε), OP (k)))|T ∗ P •
Rad Rad
H n−2 (µhom(ΦS (CUb \A (ε)), OP∗ (k ∗ )))|T ∗ P∗ • / H n (µhom(LZ (ε∗ ), OP∗ (k ∗ )))|T ∗ P∗ •
Res
•
Assuming that s is non-degenerate on T ∗S (P × P∗ ), the vertical arrows Rad are isomor-
phisms.
b \ Z is a cone of signature (1, n − 1), (5.4.2) gives a morphism:
Proof. Since U
CA →
− CUb \Z [2]
and so,
(5.4.18) CZ →
− CUb \A [2]
5.4. RADON TRANSFORM FOR QUADRATIC CONES 71
(5.4.19)
' '
H n−2 (µhom(ΦS (CUb \A (ε)), OP∗ (k ∗ )))|T ∗ P∗
• / H n (µhom(LZ (ε∗ ), OP∗ (k ∗ )))|T ∗ P∗
•
Res
Proof. For any ξ ∈ ω, the set {ξ} d ∩ (U b \ A) is homeomorphic to Pn−1 \ An−1 . Hence,
the stalks of ΦS (CUb \A (ε)) are constant over ω. Since ω is simply connected, there is
L ∈ Db (mod(C)) such that ΦS (CUb \A (ε))|ω is is the constant sheaf Lω .
Hence, we have
• •
SS(CUb \A (ε)) ∩ χ−1 T ∗ ω = χ−1 (SS(ΦS (CUb \A (ε))) ∩ T ∗ ω) = ∅
As an immediate corollary,
Appendix
S̃ = r(S) ⊂ Y × X
S Λ
f g p 1 |Λ pa
2 |Λ
} !
• •
∗
X Y, T X T ∗Y
b := g(f −1 (A)), B
For A ⊂ X, B ⊂ Y , we note A b := f (g −1 (B)).
We denote by BS|X×Y the DX×Y -module H[S] d
(O(X × Y )) and we set
(n,0)
BS|X×Y := q1−1 ΩX ⊗q−1 OX BS|X×Y
1
73
74 CHAPTER 6. APPENDIX
where G D and DF are the natural DY -modules and DX -modules associated with G
and F respectively.
For F ∈ Db (CX ), G ∈ Db (CY ), we set
ΦS (F ) := Rg! f −1 F [n − c]
ΦS (M ) := g ∗ f −1 M
ΦS̃ (N ) := f ∗ g −1 N
E M := EX ⊗π−1 DX π −1 M
(n,0)
where CS|X×Y := q1−1 ΩX ⊗q−1 OX CS|X×Y and CS|X×Y := EX BS|X×Y .
1
(n,0)
Γ(X × Y ; BS|X×Y (F , G ∗ )) ' Hom DY (DG , ΦS DF )
(n,0)
So that to any s ∈ Γ(X × Y ; BS|X×Y (F , G ∗ )), one can associate a DY morphism
α(s) : DG →
− ΦS DF . We then have
(n,0)
Theorem 6.1.1 ([DS96]). For s a non-degenerate section of Γ(X ×Y ; BS|X×Y (F , G ∗ ))
over Λ, the induced morphism H 0 (α(s)) : DG →
− H 0 (ΦS DF ) is a DY -linear isomor-
phism.
6.1. RADON TRANSFORM OF THE STRUCTURE SHEAF 75
Proof. We reproduce the proof of [DS96]. It is shown in [SKK] that this morphism is an
•
isomorphism on T ∗ Y . It can be shown that α(s) induces an EY -linear morphism E G →
−
ΦµS (E F ). Let πY be the projection T ∗ Y →
− Y , we have the following distinguished
triangle for conic objects of Db (CT ∗ Y )
+1
RπY ! (.) RπY ∗ (.) Rπ̇Y ∗ (.)
So that we have
+1
RπY ! (E G ) RπY ∗ (E G ) Rπ̇Y ∗ (E G )
α̃(s) α(s) α̇(s)
(6.1.1)
+1
RπY ! (ΦµS (E F )) RπY ∗ (ΦµS (E F )) Rπ̇Y ∗ (ΦµS (E F ))
It suffices to prove that α(s) is an isomorphism at the 0 cohomology degree. From the
theory of [SKK], we know that α̇(s) is an isomorphism. It is easily seen that
(n,0)
RπY ! (ΦµS (E F )) ' Rg! (OX×Y |S ⊗f −1 OX f −1 F )[−c]
and
(n,0)
RπY ! (E G ) ' OY ×Y |Y ⊗OY G [−n]
The 0-degree cohomology of these last two terms are null since c > 0. So the long exact
sequence induced by 6.1.1 gives
+1 (n,0)
0 πY ∗ (E G ) H 0 (Rπ̇Y ∗ (E G )) H −n+1 (OY ×Y |Y ⊗OY G )
H 0 (α(s)) ' H 1 (α̃(s))
+1 (n,0)
0 πY ∗ (ΦS (E F )) H 0 (Rπ̇Y ∗ (ΦµS (E F ))) H −c+1 (Rg! (OX×Y |S ⊗f −1 OX f −1 F ))
The result is immediate for n, c > 1 and we refer to [DS96] for the cases n = 1 or
n > 1, c = 1.
Let n > 0 be the dimension of the complex projective space. Let k ∈ Z such that
−n − 1 < k < 0. We denote by k ∗ = −n − 1 − k. Specializing theorem 6.1.1 for the
projective space and its dual, we get
Corollary 6.1.2 ([DS96]). There is an isomorphism in Db (CP∗ ):
ΦS (OP (k)) ' OP∗ (k ∗ )
76 CHAPTER 6. APPENDIX
a a a a
(µhom(F1 , F2 ) ◦ µhom(G1 , G2 )) ◦ µhom(H1 , H2 )
∼ / µhom(F1 , F2 ) ◦ (µhom(G1 , G2 ) ◦ µhom(H1 , H2 ))
2 3 2 3
a a
µhom(F1 ◦ G1 , F2 ◦ G2 ) ◦ µhom(H1 , H2 ) µhom(F1 , F2 ) ◦ µhom(G1 ◦ H1 , G2 ◦ H2 )
2 2 3 2 3 3
µhom((F1 ◦ G1 ) ◦ H1 , (F2 ◦ G2 ) ◦ H2 ) ∼ / µhom(F1 ◦(G1 ◦ H1 ), F2 ◦(G2 ◦ H2 ))
2 3 2 3 2 3 2 3
a
Proof. To lighten the notation, we won’t mention in the following the subscript of ◦, ◦,
i i
and simply denote it ◦. Let us first recall the construction of the natural morphism in
Db (kT ∗ M13 )
Lemma 6.2.1 ([KS90, Prop. 4.4.11]). There is a natural morphism
a
µhom(F1 , F2 ) ◦ µhom(G1 , G2 ) →
− µhom(F1 ◦ G1 , F2 ◦ G2 )
2 2 2
tj0 p213
p̃12 p̃23
T ∗ M123 o t p0
M2 × T ∗ M13
13
p13
p13π
} u #
∗ ∗
T M12 T M13 T ∗ M23
j
→ M12 × M23 , and by p13 the
Let us denote by j the diagonal inclusion M123 ,−
∗ p13 ∗
projection T M123 −−→ T M13 . We have the natural morphisms :
? _ T ∗ M123
tj0
T ∗ M123 o T ∗ M12 ×M2 T ∗ M23 o
δ
t p0
p13π
T ∗ M123 o M2 × T ∗ M13 / T ∗ M13
13
6.2. ASSOCIATIVITY FOR THE COMPOSITION OF µHOM 77
Rt p13π ! ◦t p0−1 t 0 /
13 ◦ R j ! (µhom(F1 , F2 ) M2 µhom(G1 , G2 ))
Rt p13π ! ◦t p−1
13 (µhom(F1 M2 G1 , F2 M2 G2 ))
so that
Rt p13π ! ◦t p0−1 t 0
13 ◦ R j ! (µhom(F1 , F2 ) M2 µhom(G1 , G2 ))
∼
−→ R p13π ! ◦ Rp213! ◦ δ −1 (µhom(F1 , F2 ) M2 µhom(G1 , G2 ))
t
and on the other hand, from (6.2.2) and from [KS90, Prop. 4.4.7.(ii)]
Rt p13π ! ◦t p0−1 t 0
13 ◦ R j ! (µhom(F1 , F2 ) M2 µhom(G1 , G2 ))
− Rt p13π ! ◦t p0−1
→ 13 (µhom(F1 M2 G1 , F2 M2 G2 ))
(6.2.4) →
− µhom(Rp13 ! (F1 M2 G1 ), Rp13 ! (F2 M2 G2 ))
− ∼→ µhom(F1 ◦ G1 , F2 ◦ G2 )
Considering the following diagram, with j the embedding M1234 ,→ M12 ×M23 ×M34
and j23 the natural diagonal embedding T ∗ M1234 ,→ T ∗ M12 × T ∗ M23 × T ∗ M34
(6.2.6) T ∗ M1234 j23
/ T ∗ M12 ×M2 T ∗ M23 ×M3 T ∗ M34
p23
14
tj0
' z
T ∗ M14
and
Rp14 ! (p−1 −1 −1
12 µhom(F1 , F2 ) ⊗ p23 µhom(G1 , G2 ) ⊗ p34 µhom(H1 , H2 ))
' Rp14π ! ◦t p0−1 t 0 −1 −1 −1
14 ◦ R j ! (p̃12 µhom(F1 , F2 ) ⊗ p̃23 µhom(G1 , G2 ) ⊗ p̃34 µhom(H1 , H2 ))
We decompose j in k ◦ h, as follows:
h k
T ∗ M1234 / T ∗ M123 × T ∗ M34 / T ∗ M12 × T ∗ M23 × M34
0−1
(6.2.7) Rp14π ! ◦t p14 ◦ Rt j 0 ! ◦ jπ−1
' Rp3π ! ◦t p0−1
3 ◦ Rp134π ! ◦t p0−1 t 0 −1 t 0 −1
134 ◦ R h ! ◦ hπ ◦ R k ! ◦ kπ
Proof.
6.2. ASSOCIATIVITY FOR THE COMPOSITION OF µHOM 79
After having “separating the variables”, the associativity of the diagram of Theorem
6.2 comes essentially from the following lemma:
Lemma 6.2.5. The following diagram of canonical morsphims commutes:
(6.2.8) (νM (F ) νN (G)) νL (H) ∼ / νM (F ) (νN (G)) νL (H))
(νM ×N (F G)) νL (H) νM (F ) (νN ×L (G H))
νM ×N ×L ((F G) H) ∼ / νM ×N ×L (F (G H))
Proof. The proof relies on the following description of the speacialization sheaf given
in [KS90] th.4.2.3. Let V a conic open subet of TM X, then for an integer j
H j (V ; νM F ) ' lim
→
H j (U ; F )
U
81
Chapter 1
Introduction
The first part of this thesis clarified the geometric structure of the kernel of the Radon
transform in the geometry of strongly linearly concave domains of the euclidian space
Rn , n ≥ 2. However this did not provide explicit formulae. In the second part of this
thesis, we will develop explicit formulae for the kernel of Radon transform in strongly
concave domains. The starting point of the theory goes back to Cormack [Cor63]
who gave an inversion formula in the affine space. Relations between the behaviour
of the Radon transform of a distribution and its analytic wave-front were investigated
by Guillemin, Sternberg [GS77] and Boman [Bom92]. Then, Henkin announced in
[Hen04a] that a function whose real Radon transform vanishes in a strongly linearly
¯
concave domain can be described as the residueof a ∂-closed (0, n − 2) form.
In particular, in dimension 2, such a function can be explicitely described as bound-
ary values of holomorphic functions verifying some O( |z|12 ) asymptotic behaviour at
infinity. We also give some information when the assumption that the Radon trans-
form vanishes is relaxed. In this thesis, started under the supervision of G. Henkin, we
clarify this statement by giving explicit formulae, and describing conditions of validity.
83
84 CHAPTER 1. INTRODUCTION
This definition can be rewritten in a more invariant way. Namely, for a volume form
ω, we may consider (n − 1)-forms ϕ such that such that x ∈ Rn , dhξ, xi ∧ ϕ(x) = ω(x),
noted dhξ, xicω. These forms are uniquely determined on {hξ, xi = 0}. For instance
dµξ is such a form. Precisely, considering the volume form ω(x) = ∧ni=1 dxi , and noting
1
dµ̃ξ (x) = (n−1)! det(ξ, dx, . . . , dx), then one can see that, for ξ ∈ P ∗ \ {ξ∞ }, dµξ (x) =
| {z }
n−1 times
n 1
dµ̃ξ (x)
| ξi |2 ) 2 , (noticing of course that for ξ ∈ P ∗ \ {ξ∞ } | ξ |6= 0).
P
|ξ|2
, where | ξ |= (
i=1
So, more explicitely we have in this case
n
(−1)i−1 ξi
P V
Z j=1,...,n;j6=i dxj
i=1
(1.1.1) Rf (ξ) = f (x)
hξ,xi=0 | ξ |2
The solutions of this PDE may be written as a superposition of plane waves. Then,
for any smooth function f , noting Rf the affine Radon transform defined below
Rf (ω, t + hω, xi − p)
Z
ϕ(t, x) = dpdω
|ω|=1,p∈R p
is a solution of the wave equation. The above integrand has to be considered in the
principal value sense.
1.1. SOME FACTS ON THE CLASSICAL RADON TRANSFORM 85
Definition 1.1.2 (Gelfand, Gindikin, Graev [GGG82]). For f a section of CP∞ (−n), we
define the projective Radon transform Rf to be the section of CP∞∗ (−1) defined by
Z
Rf (ξ) = f (x)dhξ, xicω 0 (x), ξ ∈ P
hξ,xi=0
As we said before, we can choose dhξ, xicω 0 (x) to be any (n − 1)-form of the type
det(x,u,dx,...,dx)
hξ,ui
, where u is an element chosen such that hξ, ui =6 0.
Let us note that on the local chart {x0 = 1} and considering for x ∈ Rn , f˜(x) =
f (1, x), we get back on {hξ, xi = 0} ∩ {x0 = 1} the affine Radon transform. To see this,
let us take for the above element u = (ξ0 , . . . , ξn ), so that on {hξ, xi = 0} ∩ {x0 = 1}
n
det(x, ξ, dx, . . . , dx) (n − 1)! X
= (−1)i (ξ0 xi − ξi ) ∧j6=0,i dxj
| ξ |2 | ξ |2 i=0
Then noticing that
n
X n
X X
((−1)i (ξ12 +. . .+ξn2 )xi ∧k6=0,i dxk = (−1)i ((ξ12 +. . .+ξn2 )xi − (−1)j det(x, ξ)i,j )ξj ∧k6=0,i dxk
i=0 i=0 j=1,...n,j6=i
n
X
= −ξ0 (−1)i ξi ∧j6=0,i dxj
i=0
where we noted det(x, ξ)i,j := xi ξj − xj ξi . We deduce from there that dhξ, xicω 0 is
precisely the integration form of (1.1.1).
n
Integrating in the above defintion on {hξ, xi = 0} ∩ { x2i = 1}, we get the John
P
i=0
transform, which is the Radon transform on the sphere. We see the projective Radon
transform provides a common point of view for the affine Radon and the John transform.
was generalized in 1934 by John [Joh35]. Namely, for a complex valued function f ∈
C ∞ (Rn , C), and denoting by L the Laplacian, then
Z
1 n−1
(1.1.2) f (x) = n−1
(Lx ) 2 Rf (θ, hθ, xi)ω(θ), for n odd
2(2πi) θ∈S 1
Z Z
1 n dt Rf (θ, t)
(1.1.3) f (x) = (Lx ) 2 ω(θ)p.v. dt, for n even
2(2πi)n−1 θ∈S 1 t∈R t − hθ, xi
where we parametrize here hyperplanes in Rn with (θ, t) ∈ S n−1 × R and where ω(θ) =
θ1 dθ2 − θ2 dθ1 The last integral is taken in the sense of principal value.
The next question was the following: Does there exist an inversion formula when the
Radon transform is only known along hyperplanes contained in some concave domain
of Rn ? In 1963, Cormack [Cor63] was able to reconstruct a function from the data
of real Radon transform given in the complement of a compact ball. The work of
Cormack, together with the design of a specific medical computer by Hounsfield, led to
the medical X-ray computed tomography.
(n − 1)!
Z Z
f (y)
(1.1.4) f (x) = limε−→0 ω 0 (θ) ∧ dy
(2iπ)n y∈Rn θ∈S n−1 (hθ, y − xi + iε)n
and so
(−1)n−1 dn−1
Z
(1.1.5) f (x) = limε−→0 Kε ( Rf (θ, p))(hθ, xi)ω 0 (θ)
2(2iπ)n−1 θ∈S n−1 dpn−1
where KεR is the operator that takes each integrable real function ϕ to the function
ϕ(p)
Kε ϕ(t) = iπ1 p∈R p−t+iε dp.
Distinguishing on the parity of the dimension n, we get the formulae (1.1.2) and
(1.1.3).
It will be an essential fact to notice that the form
dn−1
Z
dpn−1
Rf (θ, p)
dp ∧ ω 0 (θ)
p∈R p − hθ, zi
1.1. SOME FACTS ON THE CLASSICAL RADON TRANSFORM 87
provides through Cauchy formula an extension of the integrand of 1.1.5 which depends
holomorphically on z in the domain {z ∈ Cn ; ±hθ, Im(z)i > 0}. For z ∈ Cn \ Rn ,
the above form may be not integrable on S n−1 . We will see that under cancellation
properties of the Radon transform and restrictions on the domain, this form will be
integrable.
1 dn−1 ˜
G(Rf )(ξ, ξ∞ , z) = (f (ξ, ξ∞ , λ))λ=− hξ,zi det[ξ, ξ∞ , dξ, . . . , dξ]
hξ∞ , zin λn−1 hξ∞ ,zi
Let us consider any cycle γ(ξ∞ ) such that any ray out of ξ∞ intersects γ only once,
and let us note P∞ = {z ∈ P; hξ∞ , zi = 0} then assuming that G(Rf )(ξ, ξ∞ , z) has
boundary value for x ∈ (P \ P∞ ) ∩ P , then we have the inversion formula:
Z
f (x) = k G(Rf )(ξ, ξ∞ , x)
γ(ξ∞ )
for some constant k ∈ C. This is the projective invariant inversion formula of the
projective Radon transform. We will notice that the inversion formula of the affine
Radon transform for the particular choice of ξ∞ = (1, 0, . . . , 0).
It was shown in [Cor63] that it is then possible to decompose fn with the help of Rfn
1 ∞ 2
Z
1 s
fn (r) = − (s − r2 )− 2 T|n| ( )(Rfn )0 (s)ds
π r r
where Tn (.) = cos(n arccos(.)) are Chebyshev polynomials of first kind.
88 CHAPTER 1. INTRODUCTION
So, the critical remark about this formula is that fn (r) depends only on the data
Rfn (s), s > r, so that we can deduce that f can be reconstructed with the help of Rf
given in the complement of a compact ball. It appears that the reconstructability of the
function in complementary of compact balls depend critically on the decay conditions
at infinity. For instance, one can consider in R2 \ {0}, the 2-variables function (x, y) 7→
1
(x+iy)2
. Viewing this function as holomorphic one of the complex variable, we see that
Cauchy theorem gives that its Radon trasform is null though it is not null on R2 \ {0}.
Here, the non Schwartz behaviour at infinity explains that Radon transform given in
the complement of a compact ball is not enough to reconstruct the function. In this
thesis, we will be interested in caracterizing the functions whose Radon transform is
null in strongly linearly concave domain, which are complementary in the projective
space of compact affine convex sets and will be described later.
with its adjoint operator provides an elliptic pseudo-differential operator. But given
a pseudo-differential operator P and char(P ) its characteristic variety, then for any
distribution f , the analytic wave-front set noted W FA verifies W FA (f ) ⊂ W FA (P f ) ∪
char(P ), we see here the interplay between microlocal analysis and Radon transform.
Hörmander [Hör83] showed that the vanishing of such an operator along some subset
of the function support, produces strong constraints on the regularities of this very
function, expressed in terms of wave-fronts (see also [GQ00]), which in turn may give
information on the support of the function according to a microlocal extension theorem
proved by Kawai, Kashiwara and Hörmander [Hör83]: the latter theorem tells us that,
given a distribution defined on some neighborhood of a C 2 hypersurface S in Rn , given
some x ∈ S, assuming that f is ”null on one side of S” in some neighborhood of x, and
such that for any ξ in the conormal (TS∗ Rn )x of S at x, (x, ξ) does not belong to the
analytic wave-front set of f , then f is null in some neighborhood of x.
As for the surjectivity, as said above, growth at infinity is Schwartz implies that
a function may be reconstruted from the data of its Radon transform in the comple-
mentary of a compact convex set. A natural question arises whether it is possible
to reconstruct a function from its Radon transform when the hypothesis of fast de-
cay is restricted to a limited cone of directions. We will see that the answer is that
a reconstruction formula exists but up to a kernel that we will explicit. In general,
the Radon operator is not injective when restricted to functions on linearly concave
domains, which are domains unions of hyperplanes. As an example, we can set as a
linearily concave domain {(x1 , x2 ) ∈ R2 , min(x1 + | x2 |, −x1 + | x2 |) > −1}, then take
any function ϕ ∈ Cc (R, C) of the x2 variable to define f (x1 , x2 ) = ϕ(x2 ) in the strip
{(x1 , x2 ) ∈ R2 , | x2 |< η} and f (x1 , x2 ) = −ϕ(x2 ) in {(x1 , x2 ) ∈ R2 , | x2 − 2η |< η} ,
for some η > 0, and 0 elsewhere. Then we check easily that the real Radon transform
of f is null in this domain while f is not null. Hence, we can not hope for a general
inversion formula deducing uniquely a function from its Radon transform data in a
linearily concave domain.
The exact same definition is still valid when considering strongly linearly concave
domain of the complex space. Let Ω be a proper strongly linearly concave do-
main of R2 . We suppose Ω has C 1 boundary, we denote by Ω∗ its dual consist-
ing of lines contained in Ω, parameterized by elements (θ, t) ∈ S 1 × R. We note
Ω∗x = {(θ, t) ∈ Ω∗ ; t + θ1 x1 + θ2 x2 = 0}. Also, we define ΩC consisting of the com-
plexification of each line contained in Ω. We will then note Ω± the two connected
components of ΩC \ Ω.
90 CHAPTER 1. INTRODUCTION
Let ω be the 2-form in C2 , ω(ξ) = dξ1 ∧ dξ2 . Let C l,m (R2 , C), l, m ∈ N, be the
1
space of functions f of class C l , such that | f (x) |= O( (1+|x|)m ) at infinity, for all
derivatives of f up to order m.
We will prove the existence of explicit operators BR , K± , respectively with values in
the space of continuous functions on Ω and with values in the space of holomorphic
functions in Ω± . Moreover, K± will verify in their respective definition domain
the growth condition O( |z|12 ) at infinity, allowing the decomposition f = B(Rf ) +
Im(z2 ) Im(z1 )
K+ f − K− f . We note, for z ∈ Ω± , ϑ(z) = (ϑ1 (z), ϑ2 (z)) = ( |Im(z)| , − |Im(z)| ).
The argument of the proof consists in computing ∂f¯ ± and linking it to the Radon
transform of f . We have the following proposition, interesting by itself:
i ∂ ∂
(ϑ1 (z) + ϑ2 (z) )Rf (ϑ(z), hϑ(z), <(z)i).
| Im(z) | <(z1 ) <(z2 )
We no longer assume that Radon transform vanishes on Ω∗ , then have the following
theorem, which is the first important result of this part:
Theorem 1.2.4. We assume Ω has C 3 regularity. Let f ∈ Cc3 (R2 , C), then f can
be decomposed in Ω
f = B(Rf|Ω∗ ) + K+ f − K− f.
K+ f − K− f is the difference of boundary values of K± respectively holomorphic in
Ω± , and verifying the growth condition in their respective domain K± f (z) = O( |z|12 )
when | z | tends to infinity and we have R(K+ f − K− f ) ≡ 0 in Ω∗ .
Let us recall some result of Boman that can be deepened in the perspective of a
theorem of Kolm-Nagel [KN68] expressed in dimension n > 1.
In dimension 2, this result can be clarified in view of theorem 1.2.4: we prove that a
function whose Radon transform vanishes in a strongly linearly concave domain Ω
translates into decomposition of the function into jumps of functions holomorphic
in wedges of edge Ω. However, it is known since the work of Kolm-Nagel [KN68]
that, given an open convex cone Γ, any two functions f± holomorphic in some
wedges U ± iΓ, U open set of Rn , for some integer n > 1, such that the derivatives
(m) (m)
coincide, for any m ∈ N, f+ = f− on a so called Γ-like curve γ (for any cone Γ,
a C 1 −class curve is Γ-like, if at each point of the curve, the tangent vector belongs
to the set of directions of Γ), then f+ = f− on a convex hull C(γ) of γ. Putting
together this result and the Kolm-Nagel theorem, we deduce that any function
whose Radon transform vanishes in Ω and all of which derivatives is on a segment
I whose supported line is contained in Ω , vanishes in some convex hull C(I), which
is precisely theorem of Boman.
Ωε is strongly linearly concave domain of C2 , and we note Ω∗ε the complex lines
contained in Ωε . Besides, it is known that for a strongly linearly concave domain,
there is a C 1 mapping, called Section of Leray, from ΩC to S 1 , noted ϑ , such that
hϑ(z), ξ − zi =6 0, for z ∈ Ωε , ξ ∈ ∂Ωε , we also note ϑ(z)⊥ = (−ϑ2 (z), ϑ1 (z)).
92 CHAPTER 1. INTRODUCTION
Let f a function of C 2,2 (R2 , C), and f˜ an extension of f to C2 such that, ∂¯f˜ = O(|
Im(z) |N ), for some integer N > 1, in some complex neighborhood of the real plane.
And we note F = f˜ω. We also consider some family of C 1 functions (χε )ε>0 : R → − R,
such that χε (x) = 0 for x ≤ 0, χε (x) = 1 for x ≥ ε and | χ0ε (x) |< 1ε for 0 ≤ x ≤ ε.
We consider the function defined for z ∈ C2 , by χε (hϑ⊥ (<(z)), Im(z)i), also noted
this function χε . We define the complex Radon transform of a non necessarily
¯
∂-closed (2, 1) form G in Ωε , to be
Z
1 1
(1.2.1) RC G(ξ) = G(z), ξ ∈ Ω∗ε
2πi z∈∂Ωε hξ, zi
We have the following lemma
Lemma 1.2.7. For ξ ∈ Ω∗ such that Im(ξ) = 0, we have
¯ ε )(ξ) = Rf (ξ)
limε−→0 RC (F ∧ ∂χ
Z
hRes(F|Rn−1 ), ϕi = lim F (z)ϕ(<(z)) ∧nj=1 dzj
θ,t ε−→0 n−1
z∈Rθ,t n−1
+i.(Sεn−1 ∩Rθ,0 )
Z
(1.2.2) hRes(F ), ϕi = lim F (z)ϕ(z) ∧nj=1 dzj
ε−→0 z∈U +iε.C
Theorem 1.2.11. Definition 1.2.10 does not depend on the choice of the partition
of the unity. Under these conditions, the Radon transform of f , in the sense of
distributions, vanishes along all hyperplanes of Ω.
Theorem 1.2.12. Let f ∈ C 1,2n−2 (Rn , C) such that the affine Radon transform of
¯
f vanishes on Ω∗ , then there exists a (0, n − 2)−form F , ∂−closed in ΩC \ Ω ,
1
satisfying an estimate in O( |z|2n−2 ). For z ∈ ΩC \ Ω
Z
f (y)
F (z) = n
dy ∧ cos(ϕ)dϕ ∧ ω 0 (Im(z))
y∈Rn ,ϕ∈]0,π[ h(Im(z), ϕ), y − zi
where
n−1
X
h(Im(z), ϕ), y − zi = Im(zi ) sin(ϕ)(yi − zi )+ | Im(z) | cos(ϕ)(yn − zn )
i=1
n−1
1
X
| Im(z) |= ( | Im(zj ) |2 ) 2
i=1
We then have Z
f (x) = lim F (z)
ε−→0 Im(z)∈Sεn−2
f = B(Rf ) + res(F )
where B is an operator from C ∞ (Ω∗ , C) to C ∞,n (Ω, C), where F is an (0, n−2)−form
¯
∂−closed in ΩC \ Ω and satisfying an estimate O( |z|2n−2 1
).
Chapter 2
The case n = 2
2.1 Introduction
We mean to study the kernel, the image and a possible inversion formula for the real
Radon transform in linearly concave domains in dimension 2. We know how to recon-
struct a function globally from its Radon transform when the latest is known all along
every hyperplanes of the space [Rad17]. Our purpose will be somehow to establish
a semi-global analogue of this result. In this way, we will see that, modulo a kernel
we will precise, consisting of jumps of holomorphic functions, each of which is defined
upon a wedge and submitted to an estimation in O( |z|12 ) when | z | tends to infinity, an
inversion formula is reachable as soon as the Radon transform is known in a strongly
linearly concave domain.
Let Ω be a domain of Rn , and Ω∗ the dual domain consisting in hyperplanes contained
in Ω. To get a feeling of what could be the kernel of the Radon transform in linearly
concave domain, let us recall some result related to the complex Radon transform. As
usual, let denote P , resp. P, the n-dimensional real, resp. complex, projective space.
Definition 2.1.1 ([Mar67],[GH78]). Let K be a strongly linearly convex compact in P
¯ = 0. The complex Radon transform of f is
and Ω = P\K, f ∈ Cn,n−1 (Ω), such as ∂f
defined by Z
1 ξ0 .z0
RC f (ξ) = ∂¯z ( ) ∧ f, ξ ∈ K̃.
2πi z∈Ω hξ, zi
Here K̃ defined in section ?? as elements of P such as {z ∈ P; hξ, zi = ξ0 z0 + ... + ξn zn =
0} ⊂ P \ K and Cn,n−1 (Ω) is the space of (n, n − 1)−continuous forms on Ω.
Henkin and Polyakov [HP86] proved the remarkable fact that a (n, n − 1) form
f , whose complex Radon transform vanishes on a strongly linearly concave domain is
exact. Let us notice here that this definition is consistent with (1.2.1) given in the
precedent section.
Theorem 2.1.2 ([GH78], [HP90a]). Let Ω be a strongly linearly concave domain of P.
¯
For any ∂-closed 1
form f in Cn,n−1 (Ω), the space of (n, n − 1) forms of class C 1 on Ω,
there are explicit operators Ξ and H allowing the following representation:
¯
f = Ξ(RC f ) + ∂Hf
95
96 CHAPTER 2. THE CASE N = 2
The results of Henkin and Polyakov tell us that on a strongly linearly concave
domain, the Dolbeault cohomology of (p, n − 1) degree in Ω, p 6= n − 1, is described by
the complex Radon transform.
We can deduce from this result some heuristic to link this result, involving complex
Radon transform, to one related to the real Radon transform. So heuristically, we may
get here the intuition of some function f defined in some concave domain Ω ⊂ R2
and extend it to a suitable (2, 1) integration current in some neighboorhood ΩC,ε of
the complexification ΩC of Ω, the current being supported in some hypersurface Γ of
ΩC,ε , where let say Γ = {z = (z1 , z2 ) ∈ C2 ; Im(z2 ) = 0} ∩ ΩC,ε . If there is a way to
bind the complex Radon transform of this current to the real Radon transform of f ,
then we might hope that the vanishing of real Radon transform induces the vanishing
of the complex Radon transform of the current. The above theorem tells us that the
current would then be exact so that it could lead us to the fact that f is in some sense
the residue of a holomorphic function in ΩC \ Ω, the latter consisting in 2 connected
components Ω± .
2.1.1 Notations
Let Ω̃ be a proper strongly linearly concave domain of P , with C 1 boundary. We note
for some local chart R2 , Ω = Ω̃ ∩ R2 and denote by Ω∗ ⊂ P ∗ its dual consisting of lines
contained in Ω, parameterized by a subset of S 1 × R. Precisely, (θ, t) in S 1 × R will
parametrize the line {(x1 , x2 ) ∈ R2 ; h(θ, t), xi = t + θ1 x1 + θ2 x2 = 0}. We will note π
the projection of Ω∗ on S 1 . For θ ∈ S 1 , we note θ⊥ = (−θ2 , θ1 ).
In addition, we define ΩC to be the union of the complexification of lines contained
in Ω, namely
ΩC = ∪(θ,t)∈Ω∗ {z ∈ C2 , t + θ1 z1 + θ2 z2 = 0}
m ∂k
sup (1+ | x |) | f (x) |< ∞
x∈R2 ∂xi1 ∂xj2
k = 0, 1, .., m; i + j = k.
For an open set U ⊂ S 1 × R, we will note C l,m (Ū, C), l, m ∈ N, the space of functions
f with values in C, belonging to C l (Ū, C), such that
mdi dj
sup t | i j f (θ, t) |< ∞
t∈R dθ dt
i + j = m, uniformly with regards to θ ∈ π(Ū ).
2.1. INTRODUCTION 97
Also, we note ω designates the 2-form in C2 , ω(ξ) = dξ1 ∧ dξ2 , and ϑ the continuous
Im(z2 ) Im(z1 )
mapping from ΩC to S 1 defined for z ∈ Ω± , ϑ(z) = (ϑ1 (z), ϑ2 (z)) = ( |Im(z)| , − |Im(z)| ).
⊥
We note ϑ(z) is the element (−ϑ2 (z), ϑ1 (z)).
We want to show that, for such a domain Ω, and for some integrable function f
with some regularity conditions, there exists an operator B, with value in the space
of continuous functions in Ω and operators K± with value in the space of holomorphic
functions in Ω± , satisfying in their respective definition domain the growth condition
O( |z|12 ) when | z | tends to infinity, allowing for the decomposition
f = B(Rf ) + K+ f − K− f
The purpose of this work is not just to show the existence of such operators but to
give explicit formulae.
Precisely, we first try to get a characterization of the kernel of the Radon transform
in strongly linearly concave domain. We will show that it consists of boundary values
of wedge-defined holomorphic functions, satisfying in their respective definition domain
an asympototic behaviour of O( |z|12 ) when | z | goes to infinity. This will bring us to a
statement announced by Henkin with no proof in [Hen04a], the result of which all the
beginning of this work consists in advancing a proof.
Theorem 2.1.3. Let f a function of C 1 (R2 , C) and integrable on any line contained in
Ω, let us assume that the Radon transform vanishes on Ω∗ . Then f can be written as
the boundary value of holomorphic functions f± defined respectively in the domains Ω± ,
and satisfying in their respective domain a growth is O( |z|12 ) when | z | tends to infinity.
We have for z ∈ Ω±
Z
1 Im(ξ)
f± (z) = − f (<(ξ)) det(ϑ(z), )(ϑ(z)dξ)cω(ξ))
2πi hϑ(z),ξ−zi=0,Im(ξ)=ϑ(z)⊥ hIm(ξ), <(ξ) − zi
The boundary value of f± is understood in the sense that, for x ∈ Ω and any θ ∈ S 1
such that there exists t ∈ R such that (θ, t) ∈ Ω∗x , then it is set to be limε→0 f± (x±iεθ⊥ ).
Theorem 2.1.3 can be seen as a precision of the Boman’s microlocal regularity the-
orem [Bom92] studying analysis of microlocal solutions of the equation Rf =0.
In addition, we will establish a characterization of the image of Radon transform.
Proposition 2.1.4. Let g ∈ C 3,2 (Ω̄∗ , C), such that g verifies g(θ, t) = g(−θ, −t), then
there exists f ∈ C 1 (Ω, C), integrable on any line contained in Ω, such that Rf = g
Finally, we no longer assume that Radon transform vanishes on Ω∗ . We will see
in the proof of the below theorem that given the Radon transform of a function f in
]), an extension
a strongly linearly concave domain Ω, we can construct a function R(f
1
with compact support of Rf to S × R, such that we have the following theorem
Theorem 2.1.5. We assume that Ω has C 3 boundary. Let f be a function in Cc3 (R2 , C).
Then we have the following decomposition of f in Ω
f = B(Rf ) + K+ f − K− f.
98 CHAPTER 2. THE CASE N = 2
1
Z Z ])(θ, t)
R(f
B(Rf )(x) = limε−→0 ω 0 (θ) ∧ dt
(2πi)2 t∈R θ∈S 1 (t − hθ, xi + iε)2
Im(ξ)
(f (<(ξ)) − B(R(f ))(<(ξ))) det(ϑ(z), )(ϑ(z)dξ)cω(ξ))
hIm(ξ), <(ξ) − zi
Moreover, R(K+ f − K− f ) vanishes in Ω∗ .
Z Z
1 h(y) 1 h(y)
(2.2.1) h(x) = lim( dy − dy)
ε→0 2πi y∈R y − x + iε 2πi y∈R y − x − iε
This formula can be directly derived from the one complex variable Cauchy formula.
Let f a fonction in C 1 (R2 , C) integrable on lines contained in Ω and let x ∈ Ω. By
applying Plemelj-Sohotski’s fomula on each real lines <({ξ ∈ C2 |hϑ(x), ξ − xi = 0})
embedded in complex lines {ξ ∈ C2 |hϑ(x), ξ − xi = 0}, we get:
(2.2.2) Z
1 f (<(ξ))
f (x) = lim( (ϑ(x)d<(ξ))cω(<(ξ))
ε→0 2πi hϑ(x),ξ−xi=0,Im(ξ)=0 hϑ(x)⊥ , <(ξ) − xi + iε
2.2. KERNEL OF THE RADON TRANSFORM IN STRONGLY LINEARLY CONCAVE DOMAIN OF R2 .99
Z
1 f (<(ξ))
− (ϑ(x)d<(ξ))cω(<(ξ)))
2πi hϑ(x),ξ−xi=0,Im(ξ)=0 hϑ(x)⊥ , <(ξ) − xi − iε
This is the formula (2.2.1) where we proceed to the change of variable ξ10 := hϑ(x)⊥ , ξ−xi
and ξ20 := hϑ(x), ξ − xi.
We will transform this expression to make it more convenient, as we will see in
section 2.2.2. Indeed, if one define f± on Ω± such that for z ∈ Ω± :
Z
1
f± (z) = −
2πi hϑ(z),ξ−zi=0,Im(ξ)=ϑ(z)⊥
Im(ξ)
f (<(ξ)) det(ϑ(z), )(ϑ(z)d<(ξ))cω(<(ξ)))
hIm(ξ), <(ξ) − <(z)i − ihIm(ξ), Im(z)i
then, the formula (2.2.2) gives, for x ∈ Ω :
Let us replace in the two first integrals the variable s by s − ihϑ(z)⊥ , Im(z)i +
ihϑ(z)⊥ , Im(z)i. We then obtain:
(2.2.4) + (2.2.5) =
(2.2.8)
Z
(ϑ1 (z)∂¯z1 ϑ1 (z)⊥ +ϑ2 (z)∂¯z2 ϑ1 (z)⊥ )∂1 f ()+(ϑ1 (z)∂¯z1 ϑ2 (z)⊥ +ϑ2 (z)∂¯z2 ϑ2 (z)⊥ )∂2 f ()ds
s∈R
(2.2.9)
(ϑ1 (z)(1 + i∂¯z1 ϑ1 (z)⊥ hϑ(z)⊥ , Im(z)i) + iϑ2 (z)∂¯z2 ϑ1 (z)⊥ hϑ(z)⊥ , Im(z)i)∂1 f ()
Z
+ ds
s∈R s − ihϑ(z)⊥ , Im(z)i
(2.2.10)
(iϑ1 (z)∂¯z1 ϑ2 (z)⊥ hϑ(z)⊥ , Im(z)i + ϑ2 (z)(1 + i∂¯z2 ϑ2 (z)⊥ hϑ(z)⊥ , Im(z)i))∂2 f ()
Z
+ ds
s∈R s − ihϑ(z)⊥ , Im(z)i
Besides, since hϑ(z), ϑ(z)⊥ i = 0, and Im(z1 )ϑ2 (z)⊥ = Im(z2 )ϑ1 (z)⊥ , we have by
differentiating
h∂¯z ϑ(z)⊥ , Im(z)i = h∂¯z ϑ(z)⊥ , Im(z)i = 0
1 2
We thus obtain
(2.2.6) + (2.2.7) = 0.
We also have,
ϑ1 (z)(1 + i∂¯z1 ϑ1 (z)⊥ hϑ(z)⊥ , Im(z)i) + iϑ2 (z)∂¯z2 ϑ1 (z)⊥ hϑ(z)⊥ , Im(z)i = 0
Indeed, by elmentary calculus, the term above, (ϑ1 (z)⊥ , ϑ2 (z)⊥ ) = (−ϑ2 (z), ϑ1 (z))
Im(z1 ) Im(z2 )
being ( |Im(z)| , |Im(z)| ), cancels out. Thus, in the sum (2.2.9) + (2.2.10), the number that
multiplies ∂1 f is zero, as well for the number that multiplies ∂2 f .
We can conclude that
(2.2.11)
Z
(ϑ1 (z)∂¯z1 ϑ1 (z)⊥ +ϑ2 (z)∂¯z2 ϑ1 (z)⊥ )∂1 f ()+(ϑ1 (z)∂¯z1 ϑ2 (z)⊥ +ϑ2 (z)∂¯z2 ϑ2 (z)⊥ )∂2 f ()ds
s∈R
Now, we deduce from the relations Im(z1 )ϑ2 (z)⊥ = Im(z2 )ϑ1 (z)⊥ and (ϑ1 (z)⊥ )2 +
(ϑ2 (z)⊥ )2 = 1, that
2.2. KERNEL OF THE RADON TRANSFORM IN STRONGLY LINEARLY CONCAVE DOMAIN OF R2 .10
¯ ⊥ ¯ ⊥ ϑ2 (z)⊥ ϑ1 (z)
ϑ1 (z)∂z1 ϑ1 (z) + ϑ2 (z)∂z2 ϑ1 (z) = i =i
| Im(z) | | Im(z) |
The same way,
ϑ1 (z)⊥ ϑ2 (z)
ϑ1 (z)∂¯z1 ϑ2 (z)⊥ + ϑ2 (z)∂¯z2 ϑ2 (z)⊥ = −i =i
| Im(z) | | Im(z) |
We thus have:
Z
i
(2.2.11) = ϑ1 (z)∂1 f () + ϑ2 (z)∂2 f ()ds
Im(z) s∈R
Z
i ∂ ∂
(ϑ1 (z) + ϑ2 (z) ) f (<(z1 ) + ϑ1 (z)⊥ s, <(z2 ) + ϑ2 (z)⊥ s)ds
| Im(z) | <(z1 ) <(z2 ) s∈R
Supposing that the Radon transform vanishes for any line contained in Ω, this last
term is zero and we have
This term cancels out since it is nothing more than the Radon transform of f on the
line <(hϑ(z), ξ − zi) = 0. The expansion of f± , is hence in O( |z|12 ).
The following result, known as the theorem of support, can be found in John [Joh35]
and then generalized by Cormack[Cor63] and Helgason[Hel64].
Corollary 2.2.2. Let f be a function in C 2 (R2 , C) with non empty compact support
K. Let us assume that the Radon transform vanishes in a strongly linearly concave
domain Ω. We also assume that there is a line D of R2 entirely contained in Ω such as
D ∩ K = ∅. Then f vanishes in Ω.
Remark 2.2.3. (i) This result is the one demonstrated by John [Joh35] and then
generalized by Cormack [Cor63] and Helgason [Hel64]. Indeed, just take the line D
mentioned in the above corollary, i.e. contained in Ω and with empty intersection
with the support of f , and consider it as the line at infinity in P . We are then in
the conditions of the theorem demonstrated by the above authors.
(ii) If f is a function with compact support in a strongly linearly concave domain Ω
of R2 , and if we assume that its Radon transform vanishes in Ω∗ , then f vanishes.
(iii) Let us mention some result of Boman [Bom92] sharper than corollary 2.2.2: let
S be a real analytic hypersurface S of Rn , let f be a distribution defined in some
neighborhood of S, assume that no point of the conormal bundle to S lies in the
analytic wave-front set of f , and suppose that f and all its derivative’s restriction
to S vanishes, then f vanishes on some neigborhood U of S. When f is supposed
to be a continuous function, then one gets the same vanishing result by imposing
the following flatness condition: for x ∈ U tending to S, f (x) decays to 0 faster
than any negative power of the distance of x to S.
The following result is an application of a refinement of the “edge-of-the-wedge”
theorem, which is due to Kolm and Nagel [KN68]. But it is Boman’s in [Bom92], whom
we owe the formulation of the following proposal, here demonstrated in the light of
theorem 2.1.3. It should be noted here that Boman exposes a proof which essentially
relies on a theorem of microlocal regularity of the solutions of the equation R(f ) =0.
Proposition 2.2.4. Let’s say a K a proper convex compact of R2 , and f a function of
C 2,2 (R2 ). Let us suppose that there is a I bounded segment of R2 , such as the line D that
it generates belongs to (R2 \ K)∗ . Suppose, moreover, that the Radon transform of f
vanishes on all lines of R2 \ K and that all derivatives at all orders of f on I vanishes.
Then, f vanishes in the neighborhood of I.
Proof. Let a,b be the elements of the edge of I. We consider V and V 0 , (always
parametrized by S 1 × R), subsets of (R2 \ K)∗ corresponding to lines passing through
a respectively b. Note C = π(V ) ∩ π(V 0 ), where π is the projection on S 1 . This
intersection is not empty. Let the cones C± = {y ∈ R2 |∃λ ∈ R∗ , c ∈ C, y = λc}.
We are then able to apply a refinement of the edge-of-the-wedge theorem, by Kolm
and Nagel [KN68]), stating that given an analytical curve J, parametrized by x(t), t ∈ R,
an open convex cone C, such that x0 (t) ∈ C, for any t, V(J) a neighbordhood of J, and
two functions respectively holomorphic on V(J) ± iC± such that all their derivatives, to
all orders coincide on J, then there is a analytical extension common to both functions,
holomorphic on a complex neighborhood of J. In our case, the Radon transform of f
cancelling out on (R2 \ K)∗ , f is broken down into the difference of two holomorphic
functions f = f+ − f− , and derivatives of f cancelling each other out orders on I,
derivatives of f+ and f− coincide on I. We are then in the conditions of application of
the generalized, ”edge-of-the-wedge” theorem demonstrated in [KN68].
Remark 2.2.5. This proposition may still be improved. With the same assumptions
as in proposition 2.2.4, one may conclude that f vanishes in the C-convex envelope
104 CHAPTER 2. THE CASE N = 2
Proof. Let us choose to point the frame of R2 at some point of Ω. Suppose that the
complementary Ω is made up of two connected components. The boundary of π(Ω∗ )
0
is then consisting of four elements we will note ±θ0 and ±θ0 . The set S 1 \ π(Ω∗ ) is
0
composed of two components, let say A with boundary the elements θ0 and −θ0 and
0
B with boundary the elements −θ0 and θ0 . There is no loss of generality to suppose
0 0
θ0 , θ0 are represented by angles ϕ0 , ϕ0 in [0, π[
g is known for (θ, t) ∈ S 1 × R, such that θ ∈ π(Ω∗ ) and t ∈ [aθ , bθ ], here aθ , bθ ∈ R̄
depending on θ. We will extend successively g on the whole (θ, t) ∈ π(Ω∗ )×(]−∞, aθ ]∪
[bθ , ∞[) then on the whole of (θ, t) ∈ A × [0, ∞[ and on all (θ, t) ∈ B×] − ∞, 0]. Finally,
we will complete with parity argument. We will note ge this extension.
Let us note c the real number 2 maxθ∈π(Ω∗ ) (| aθ |, | bθ |). Let Pa,b;αi ,βi the poly-
nomial interpolator of degree 8, such that, for real a, b, αi , βi , i = 0, 1, 2, 3, P (i) (a) =
αi , P (i) (b) = βi . These polynomials are chosen for that the extension built below is C 3 .
Then we set, for bθ ≤ t ≤ c,
ge(θ, t) = Pb d d2 d3
θ ,c;g(θ,bθ ), dt g(θ,bθ ), dt2 g(θ,bθ ), dt3 g(θ,bθ ),0,0,0,0
ge(θ, t) = Pa d d2 d3
θ ,−c;g(θ,aθ ), dt g(θ,aθ ), dt2 g(θ,aθ ), dt3 g(θ,aθ ),0,0,0,0
0 d d 2 d 3
where we noted G(θ0 , t), resp. G(−θ0 , t), the quadruplet (g(θ0 , t), dθ g(θ0 , t), dθ 0 0
2 g(θ , t), dθ 3 g(θ , t)),
0 d 0 d2 0 d 3 0
resp. (g(−θ0 , t), dθ g(−θ0 , t), dθ 0 0
2 g(−θ , t), dθ 3 g(−θ , t)), where we noted for θ ∈ S ,
1
d i d
and ϕ such that θ = (cos(ϕ), sin(ϕ)), dθ i g(θ, t) = dϕ g((cos(ϕ), sin(ϕ)), t), i = 1, 2, 3.
We proceed the same way for θ ∈ B. Of course, the extension above belongs to C 3,2
since it cancels for any (θ, t), with t big enough.
If the complementary Ω is made up of only one connected component, then it is
necessary to distinguish the case where this one is compact and the other case where
it is not limited. In the second case the extension is the same as above, except that
for aθ = −∞ or bθ = +∞, it is useless to extend g(θ, t) for t < aθ (resp. t > bθ ).
The extension is still C 3,2 since, for all t, the norm of the polynomial interpolator Pt is
bounded above by a polynomial combination of partial derivatives to order 3 of g(θ0 , t)
0
and g(−θ0 , t), the result will follow a behaviour in O( t12 ).
At last, in the case where the complementary of Ω is compact, we point the frame
of R2 in the latter compact. Then we have to extend g for (θ, t) ∈ S 1 × [aθ , bθ ]. Then
we note c = 12 minθ∈S 1 (| aθ |, | bθ |). And we set, for c ≤ t ≤ bθ ,
ge(θ, t) = Pb d d2 d3
θ ,c;g(θ,bθ ), dt g(θ,bθ ), dt2 g(θ,bθ ), dt3 g(θ,bθ ),0,0,0
The Radon transform of f and B(Rf ) coincide on Ω∗ , and so, R(B(Rf ))(θ, t) =
Rf (θ, t), for any (θ, t) de Ω∗ . Now,
where Z
1 zj
RC,j f (ξ) = ∂¯z ( ) ∧ f, ξ ∈ K̃.
2πi z∈Ω hξ, zi
We have the following first properties
dRC f =
P ¯
(∂ + ∂)(Rf C,j dξj )
0≤j≤n
∂
P
= ∂ξk
RfC,j dξj ∧ dξk
0≤k,j≤n
P R −zk zj f
= z∈∂Ω hξ,zi2
dξj ∧ dξk
0≤k,j≤n
P R (zj zk −zk zj )f
= z∈∂Ω hξ,zi2
dξj ∧ dξk
0≤k≤j≤n
=0
We have the follwing expression of RC f in terms of integration through complex
hyperplanes
Proposition 2.4.2. For ξ ∈ K̃, we have
X Z
RC f (ξ) = hξ, dziczj f dξj
0≤j≤n hξ,zi=0
108 CHAPTER 2. THE CASE N = 2
Z
RfC,j (ξ) = hξ, dziczj f
hξ,zi=0
Z Z
zj zj
f= f
z∈∂Ω hξ, zi |hξ,zi|=ε hξ, zi
V
Then, noting fi the dzi k6=i (dzk ∧ dz¯k ) term of f . by a change of variable, we get
Z Z X Z
zj ϕj (Z) ∗ ϕj (Z) ^
f= ϕ f (Z) = fi (ϕ(Z))dϕi (Z) (dϕk (Z) ∧ dϕk (Z
|hξ,zi|=ε hξ, zi |Z0 |=ε Z0 0≤i≤n |Z0 |=ε
Z0 0≤k≤n
k6=i
Z
Zj ^
f0 (ϕ(Z))d(hη, Zi) (dZk ∧ dZk )+
|Z0 |=ε Z0 1≤k≤n
X Z Zj ^
fi (ϕ(Z))dZi ∧ d(hη, Zi) ∧ d(hη̄, Z̄i) (dZk ∧ dZk ) =
1≤i≤n |Z0 |=ε Z0 1≤k≤n
k6=i
Z
Zj 1 X ^
(f0 (ϕ(Z)) − η̄i fi (ϕ(Z)))dZ0 (dZk ∧ dZk )
|Z0 |=ε Z0 ξ0 1≤i≤n 1≤k≤n
1 X ^
hξ, dzi ∧ zj (f0 (z) − η̄i fi (z)) (dzk ∧ dzk ) = zj f
ξ0 1≤i≤n 1≤k≤n
2.4. KERNEL OF RADON TRANSFORM THROUGH ∂¯ TECHNIC 109
We have
Ωx,ε = ∪θ∈π(Ω∗x ) {z ∈ C2 ; <(z) = x, | hθ, Im(z)i |< ε}, Ωε = ∪x∈Ω Ωx,ε , ΩC = ∩ε>0 Ωε
Ωε is strongly linearly concave domain of C2 , and we note Ω∗ε the complex lines
contained in Ωε .
Let f a function of C 2,2 (R2 , C), and f˜ an extension of f to C2 so that, there exists
|Im(z)|N
a neighborhood U of 0 so that for z ∈ R2 + iU , | ∂¯f˜(z) |< C (1+<(z)) 2 , for some C > 0,
and some integer N > 1 (cf. [HW69] for a reference, and also for section 3.2.1.3 for an
example of such an extension). And we note F = f˜ω, where ω is as always (2, 0)-form
dz1 ∧ dz2 . We also consider some family of C 1 functions (χε )ε>0 : R → − R, such that
χε (x) = 0 for x ≤ 0, χε (x) = 1 for x ≥ ε and | χ0ε (x) |< 1ε for 0 ≤ x ≤ ε. Ω being with
C 1 boundary, there exists some C 1 application from Ω to S 1 noted ϑ = (ϑ1 , ϑ2 ), such
that for x ∈ Ω, {ξ ∈ R2 , hϑ(x), ξ − xi = 0} ⊂ Ω and hϑ(x), ξ − xi 6= 0, for ξ ∈ ∂Ω. Let
ϕ1,x < ϕ2,x ∈]0, π[ be the angles parametrizing the 2 limits lines contained in Ω passing
by x. Since Ω is proper, there is Λ ∈]0, π[ such that | ϕ1,x − ϕ2,x |< Λ.
We note ϑ⊥ = (−ϑ2 , ϑ1 ), and consider the function C → − C, defined by z ∈ C2 ,
χε (hϑ⊥ (<(z)), Im(z)i), we also note this function χε . Let Γε be the domain {z ∈ C2 ; |
hϑ⊥ (<(z)), Im(z)i |< ε}, and Γ0 = {z ∈ C2 ; hϑ⊥ (<(z)), Im(z)i = 0}.
Now, we will define the complex Radon transform of a non necessarily ∂-closed ¯ (2, 1)
form G to be
110 CHAPTER 2. THE CASE N = 2
Z
1 1
RC G(ξ) = G(z), ξ ∈ Ω∗ε .
2πi z∈∂Ωε hξ, zi
To avoid confusion, we note RR f the real Radon transform. We are going to prove
the following
Lemma 2.4.3. For ξ ∈ Ω∗ , we have
Z
1 ¯ ε (z) =
(2.4.2) F (z) ∧ ∂χ
z∈∂Ωε hξ, zi
Z Z
1 ¯ ¯ ε (z) + 1 ¯ ε (z)
∂F (z) ∧ ∂χ F (z) ∧ ∂χ
z∈Ωε ,|hξ,zi|>ε hξ, zi |hξ,zi|=ε hξ, zi
(where for z ∈ Ωε and out of simplicity, we put in the integrand | hξ, zi |> ε rather
than considering some δ > 0 such that the integrand is | hξ, zi |> δε). For the first term
¯ ε (z) is 0 for | hϑ⊥ (<(z)), Im(z)i |> ε. A simple calculus
of (2.4.2), we notice that ∂χ
gives that, for z ∈ Γε , there is a k > 0, such that | Im(z) |< kε, for instance we can
take k = (1+sin(Λ/2)) ¯ ε (z) |< k0 ,
. And we can also find a constant k 0 > 0, such that | ∂χ
cos(Λ/2) ε
for z ∈ Γε
Z Z
1 ¯ ¯ ε (z) |=| 1 ¯ ¯ ε (z) |
| ∂F (z) ∧ ∂χ ∂F (z) ∧ ∂χ
z∈Ωε ,|hξ,zi|>ε hξ, zi z∈Ωε ∩Γε ,|hξ,zi|>ε hξ, zi
1 C | Im(z) |N k 0
Z
< ω(z)
<(z)∈<(Ωε );|Im(z)|<kε ε (1 + <(z))2 ε
Z Z
1 ¯ ε (z) = 1 ¯ ε (z)
F (z) ∧ ∂χ F (z) ∧ ∂χ
|hξ,zi|=ε hξ, zi {|hξ,zi|=ε}∩Γε hξ, zi
Z Z Z
1 ˜ ¯ ε (z) = 1 ˜ ∂
= f (z) ∧ ∂χ f (z) ∧ χε (z)dz1 ∧ dz2 ∧ dz¯2
|z1 |=ε,z∈Γε z1 |z1 |=ε z2 ∩Γε,z1 z1 ∂ z¯2
2.4. KERNEL OF RADON TRANSFORM THROUGH ∂¯ TECHNIC 111
Let α(z1 ) be, of the two real numbers of Im ∂(Γε,z1 ), the one on which χε equals 1,
we get by Stokes
Z Z Z Z
¯ 1 ˜ 1 ˜
dz1 ∧ ∂z2 ( f (z) ∧ χε (z)dz2 ) = dz1 ∧ ( f (z1 , x2 + iα(z1 ))dz2
|z1 |=ε z2 ∩Γε ,z1 z1 |z1 |=ε x2 z1
Z
= f˜(0, x2 )dx2 + O(ε)
x2
|∂¯f˜|
Besides, the term z2 ∩Γε ( z11 ∂¯z2 f˜(z) ∧ χε (z)dz2 ) is controlled by so is O(εN −1 ).
R
|z1 |
So that we have obtained that
Z Z
1 ¯ ε (z) = lim 1 ¯ ε (z)
lim F (z) ∧ ∂χ F (z) ∧ ∂χ
ε→0 |hξ,zi|=ε hξ, zi ε→0 {|hξ,zi|=ε}∩Γε hξ, zi
Z
= hξ, d<(z)icF (<(z))
{hξ,zi=0}∩R2
2.4.3 ¯
∂-formulae in concave domains
2.4.3.1 Notations
Pn
dξi . Let the form ω 0 (v) = k
V
As usual ω(ξ) is the (n, 0)-form k=1 (−1) vk dv1 ∧ . . . ∧
1≤i≤n
dvk−1 ∧ dvk+1 ∧ . . . ∧dvn , where v is a coordinate vector vk . We can represent this form
1
by (n−1)! det(v, dv, . . . , dv) with n − 1 columns dv. Suppose now that v is an application
dependent on complex variables ξ, z and real variable λ, we then note
q n−1
ωq0 (v) = (−1)q det(v, (∂¯ξ + dλ )v, . . . , (∂¯ξ + dλ )v , ∂¯z v, . . . , ∂¯z v )
(n − 1)! | {z } | {z }
n−q−1 q
(n − 1)! ξ¯ − z̄
Z
0
H g(z) = − g(ξ)∧ω ( )∧ω(ξ)
(2πi)n ξ∈∂Ω q−1
| ξ − z |2
∂ρ
(n − 1)! ξ¯ − z̄ (z)
Z
0 ∂ξ
(2.4.3) + g(ξ)∧ωq−1 ((1 − λ) + λ )∧ω(ξ)
(2πi)n (ξ,λ)∈∂Ω×[0,1] | ξ − z |2 h ∂ρ
∂ξ
(z), ξ − zi
¯ g(z) = g(z).
then, if q 6= n − 1, H g satisfies ∂H
In the case q = n − 1, we have for z ∈ Ω ∩ Cn
∂ρ
(z)
¯ g(z) = g(z) + (−1)n (n − 1)!
Z
0 ∂ξ
∂H g(ξ) ∧ ωn−1 ( ∂ρ ) ∧ ω(ξ)
(2πi)n ξ∈∂Ω h ∂ξ (z), ξ − zi
ξ¯ − z̄ ¯ ξ¯ − z̄ ¯
¯ξ ξ − z̄ ) ∧ ( ∂ρ (z)dξ)cω(ξ)
Z
∂ρ
g(ξ). det( (z), , ∂ξ , . . . , ∂
h ∂ρ
∂ξ
(z),ξ−zi=0 ∂ξ | ξ − z |2 | ξ − z |2 | ξ − z |2 ∂ξ
We are going to derive the same formula as (2.4.3), in the case the form is not
supposed closed anymore.
1
Proposition 2.4.6. Let an integer q > 0. Let f ∈ C0,q (Ω̄).
(2.4.4) ¯ )(z) + ∂H
f (z) = −H(∂f ¯ f (z)
¯
¯ (z) = (n − 1)! ¯ (ξ)∧ω 0 ( ξ − z̄ )∧ω(ξ)
Z
H(∂)f ∂f
(2πi)n ξ∈Ω
q
| ξ − z |2
∂ρ
(n − 1)! ξ¯ − z̄ (z)
Z
¯ 0 ∂ξ
+ ∂f (ξ)∧ωq ((1 − λ) + λ )∧ω(ξ)
(2πi)n (ξ,λ)∈∂Ω∩×[0,1] | ξ − z |2 h ∂ρ
∂ξ
(z), ξ − zi
2.4. KERNEL OF RADON TRANSFORM THROUGH ∂¯ TECHNIC 113
(n − 1)! ξ¯ − z̄
Z
0
H f (z) = f (ξ)∧ωq−1 ( )∧ω(ξ)
(2πi)n ξ∈Ω | ξ − z |2
∂ρ
(n − 1)! ξ¯ − z̄ (z)
Z
0 ∂ξ
+ f (ξ)∧ωq−1 ((1 − λ) + λ )∧ω(ξ)
(2πi)n (ξ,λ)∈∂Ω×[0,1] | ξ − z |2 h ∂ρ
∂ξ
(z), ξ − zi
¯ )(z) + ∂H
f (z) = Lf (z) − H(∂f ¯ f (z)
∂ρ
(n − 1)!
Z
0 ∂ξ
(z)
Lf (z) = f (ξ)ωn−1 ( ∂ρ )∧ω(ξ)
(2πi)n ξ∈∂Ω h ∂ξ (z), ξ − zi
Proof. In the local chart where ∂Ω ⊂ Cn , let BR the complex ball for radius R > 0. For
1
R large enough, let DR = Ω ∩ BR . We suppose that | fj (z) |= o( 1+|z|+|z j|
2 ) at infinity.
where
∂ρ
(n − 1)!
Z
0 ∂ξ
(z)
LR f (z) = f (ξ)ωq ( ∂ρ )∧ω(ξ)
(2πi)n ξ∈∂Ω∩D̄R h ∂ξ (z), ξ − zi
(n − 1)! ξ¯
Z
+ f (ξ)ωq0 ( ¯ )∧ω(ξ)
(2πi)n ξ∈∂BR ∩D̄R hξ, ξ − zi
∂ρ
(n − 1)! ξ¯ (z)
Z
∂ξ
+ f (ξ)ωq0 ((1 − λ) ¯ + λ ∂ρ )∧ω(ξ)
(2πi)n (ξ,λ)∈(∂Ω∩∂BR )×[0,1] hξ, ξ − zi h ∂ξ (z), ξ − zi
¯
¯ (z) = (n − 1)! ¯ (ξ)∧ω 0 ( ξ − z̄ )∧ω(ξ)
Z
HR (∂)f ∂f
(2πi)n ξ∈Ω∩D̄R
q
| ξ − z |2
114 CHAPTER 2. THE CASE N = 2
(n − 1)! ¯ ξ¯
¯ (ξ)∧ω 0 ((1 − λ) ξ − z̄ + λ
Z
+ ∂f ¯ ξ − zi )∧ω(ξ)
(2πi)n (ξ,λ)∈∂BR ∩D̄R ×[0,1]
q
| ξ − z |2 hξ,
∂ρ
(n − 1)! ξ¯ − z̄ (z)
Z
¯ (ξ)∧ω 0 ((1 ∂ξ
+ ∂f − λ) + λ )∧ω(ξ)
(2πi)n (ξ,λ)∈∂Ω∩D̄R ×[0,1]
q
| ξ − z |2 ∂ρ
h ∂ξ (z), ξ − zi
(n − 1)! ¯
¯ (ξ)ω 0 ((1 − λ − µ)(1 − λ) ξ − z̄ +
Z
+ ∂f
(2πi)n (ξ,λ,µ)∈(∂Ω∩∂BR )×[0,1]2 ,λ+µ=1
q
| ξ − z |2
∂ρ
ξ¯ ∂ξ
(z)
λ ¯ + µ ∂ρ )∧ω(ξ)
hξ, ξ − zi h ∂ξ (z), ξ − zi
and
(n − 1)! ξ¯ − z̄
Z
0
H R f (z) = f (ξ)∧ωq−1 ( )∧ω(ξ)
(2πi)n ξ∈Ω∩D̄R | ξ − z |2
(n − 1)! ξ¯ − z̄ ξ¯
Z
0
+ f (ξ)∧ωq−1 ((1 − λ) + λ ¯ ξ − zi )∧ω(ξ)
(2πi)n (ξ,λ)∈∂BR ∩D̄R ×[0,1] | ξ − z |2 hξ,
∂ρ
(n − 1)! ξ¯ − z̄ (z)
Z
0 ∂ξ
+ f (ξ)∧ωq−1 ((1 − λ) + λ )∧ω(ξ)
(2πi)n (ξ,λ)∈∂Ω∩D̄R ×[0,1] | ξ − z |2 h ∂ρ
∂ξ
(z), ξ − zi
(n − 1)! ξ¯ − z̄
Z
0
+ ωq−1 ((1 − λ − µ)(1 − λ) +
(2πi)n (ξ,λ,µ)∈(∂Ω∩∂BR )×[0,1]2 ,λ+µ=1 | ξ − z |2
∂ρ
ξ¯ ∂ξ
(z)
λ ¯ + µ ∂ρ )∧ω(ξ)
hξ, ξ − zi h ∂ξ (z), ξ − zi
∂ρ ∂ρ ∂ρ
∂ξ
(z) ∂ξ
(z) ∂ξ
(z)
det( , ∂¯ξ , . . . , ∂¯ξ ,
h ∂ρ
∂ξ
(z), ξ − zi h ∂ρ
∂ξ
(z), ξ − zi h ∂ρ
∂ξ
(z), ξ − zi
| {z }
n−q−1
2.4. KERNEL OF RADON TRANSFORM THROUGH ∂¯ TECHNIC 115
∂ρ ∂ρ
∂ξ
(z) ∂ξ
(z)
∂¯z , . . . , ∂¯z )
h ∂ρ
∂ξ
(z), ξ − zi h ∂ρ
∂ξ
(z), ξ − zi
| {z }
q
∂ρ
(z)
Then, since n − q − 1 > 0, ∂¯ξ ( h ∂ρ ∂ξ
(z),ξ−zi
) = 0, so that the first term of LR f is zero.
∂ξ
ξ̄
For the second term of LR f , ωq0 ( hξ̄,ξ−zi ) is essentially
ξ¯ ξ¯ ξ¯ ξ¯ ξ¯
det( ¯ , (∂¯ξ + dλ ) ¯ , . . . , (∂¯ξ + dλ ) ¯ , ∂¯z ¯ , . . . , ∂¯z ¯ )
hξ, ξ − zi hξ, ξ − zi hξ, ξ − zi hξ, ξ − zi hξ, ξ − zi
| {z } | {z }
n−q−1 q
∂ρ
ξ¯ ∂ξ
(z)
det((1 − λ) ¯ + λ ∂ρ ,
hξ, ξ − zi h ∂ξ (z), ξ − zi
∂ρ ∂ρ
¯ ξ¯ ∂ξ
(z)
¯ ξ¯ ∂ξ
(z)
(∂ξ + dλ )(1 − λ)( ¯ + λ ∂ρ ), . . . , (∂ξ + dλ )((1 − λ) ¯ + λ ∂ρ ),
hξ, ξ − zi h ∂ξ (z), ξ − zi hξ, ξ − zi h ∂ξ (z), ξ − zi
| {z }
n−q−1
∂ρ ∂ρ
¯
ξ ∂ξ
(z) ¯
ξ ∂ξ
(z)
∂¯z ((1 − λ) ¯ + λ ∂ρ ), . . . , ∂¯z ((1 − λ) ¯ + λ ∂ρ ))
hξ, ξ − zi h ∂ξ (z), ξ − zi hξ, ξ − zi h ∂ξ (z), ξ − zi
| {z }
q
which is essentially
∂ρ
ξ¯ ∂ξ
(z)
det((1 − λ) ¯ + λ ∂ρ ,
hξ, ξ − zi h ∂ξ (z), ξ − zi
∂ρ
¯ ξ¯ ¯ ξ¯ ξ¯ ∂ξ
(z)
∂ξ (1 − λ)( ¯ ), . . . , ∂ξ ((1 − λ) ¯ ), dλ ((1 − λ)( ¯ + λ ∂ρ ),
hξ, ξ − zi hξ, ξ − zi hξ, ξ − zi h ∂ξ (z), ξ − zi
| {z }
n−q−2
116 CHAPTER 2. THE CASE N = 2
∂ρ ∂ρ
∂ξ
(z) ∂ξ
(z)
∂¯z (λ ), . . . , ∂¯z (λ ))
h ∂ρ
∂ξ
(z), ξ − zi h ∂ρ
∂ξ
(z), ξ − zi
| {z }
q
∂ρ
(z)
ξ̄
max | ( hξ̄,ξ−zi , h ∂ρ ∂ξ ) |ij ¯ ∂ρ (z)) |ij
max | (ξ,
1≤i,j≤n (z),ξ−zi 1≤i,j≤n ∂ξ
∂ξ
=
¯ ξ − zi |n−q−2 | h ∂ρ (z), ξ − zi |q
| hξ, ¯ ξ − zi |n−q−1 | h ∂ρ (z), ξ − zi |q+1
| hξ,
∂ξ ∂ξ
∂ρ
ξ̄ ∂ξ
(z)
where we note, the | . |ij the (2, 2)-minor of , ∂ρ
hξ̄,ξ−zi h (z),ξ−zi
. Now, for ξ ∈ ∂Ω,
∂ξ
¯
¯ (z) = (n − 1)! ¯ (ξ)∧ω 0 ( ξ − z̄ )∧ω(ξ)
Z
¯ (z) = H(∂)f
lim HR (∂)f ∂f
R→+∞ (2πi)n ξ∈Ω
q
| ξ − z |2
∂ρ
(n − 1)! ξ¯ − z̄ (z)
Z
¯ 0 ∂ξ
+ ∂f (ξ)∧ωq ((1 − λ) + λ )∧ω(ξ)
(2πi)n (ξ,λ)∈∂Ω∩×[0,1] | ξ − z |2 h ∂ρ
∂ξ
(z), ξ − zi
(n − 1)! ξ¯ − z̄
Z
0
lim H R f (z) = H f (z) = f (ξ)∧ωq−1 ( )∧ω(ξ)
R→+∞ (2πi)n ξ∈Ω | ξ − z |2
∂ρ
(n − 1)! ξ¯ − z̄ (z)
Z
0 ∂ξ
+ f (ξ)∧ωq−1 ((1 − λ) + λ )∧ω(ξ)
(2πi)n (ξ,λ)∈∂Ω×[0,1] | ξ − z |2 h ∂ρ
∂ξ
(z), ξ − zi
(ii) For q = n − 1, the exact same steps leads to, for z ∈ Ω, (2.4.5) is valid. Indeed,
∂ρ
0 (z)
we can no longer use the fact that ωn−1 ( h ∂ρ ∂ξ
(z),ξ−zi
) is zero.
∂ξ
2.4. KERNEL OF RADON TRANSFORM THROUGH ∂¯ TECHNIC 117
¯ ε = L(F ∧ ∂χ
F ∧ ∂χ ¯ ε ) − H(∂(F
¯ ∧ ∂χ
¯ ε )) + ∂H
¯ (F ∧ ∂χ
¯ ε)
where for z ∈ Ωε ,
Z
¯ ε )(z) = 1 ¯ ε (ξ)ω 0 ( ϑε (z)
L(F ∧ ∂χ F ∧ ∂χ 1 )
(2πi)2 ξ∈∂Ωε hϑε (z), ξ − zi
¯
¯ ε )(ξ)∧ω 0 ( ξ − z̄ )
Z
¯ ∧ ∂χ
¯ ε ))(z) = 1 ¯ ∧ ∂χ
(2.4.5) H(∂(F ∂(F 1
(2πi)2 ξ∈Ωε | ξ − z |2
¯
¯ ε (ξ)∧ω 0 ( ξ − z̄ )
Z
¯ ε )(z) = 1
H (F ∧ ∂χ F ∧ ∂χ 0
(2πi)2 ξ∈Ωε | ξ − z |2
¯
¯ ε (ξ)∧ω 0 ((1 − λ) ξ − z̄ + λ
Z
1 ϑε (z)
+ F ∧ ∂χ 0 )
(2πi)2 (ξ,λ)∈∂Ωε ×[0,1] |ξ−z | 2 hϑε (z), ξ − zi
¯ ∧ χε )) = 0. Indeed, noting vol the volume form dξ1 ∧
We have that limε→0 H(∂(F
¯ ¯
dξ1 ∧ dξ2 ∧ dξ2 ,
¯ ∧ ∂χ
¯ ε )(ξ)∧ω 0 ( ξ̄−z̄ 2 ) | = | ¯ (z) ∧ ∂χ
¯ ε (ξ)∧ω 0 ( ξ̄−z̄ 2 ) |
R R
| ξ∈Ωε
∂(F 1 |ξ−z| ξ∈Ωε ∩Γε
∂F 1 |ξ−z|
0
≤ (kε)N kε 1 ξ̄−z̄ ¯ ξ̄−z̄
R
ξ∈Ωε ∩Γε 1+<(z)2
| det( |ξ−z| 2 , ∂z |ξ−z|2 ) | vol(ξ)
ξ̄−z̄ ¯ ξ̄−z̄
R
Classically, the expression for a compact K, ξ∈K | det( |ξ−z| 2 , ∂ |ξ−z|2 ) | vol(ξ), which
ξ̄−z̄,−∂¯z̄)|
is ξ∈K |det(|ξ−z| 1
R
4 vol(ξ) < ∞. Here the integrand 1+<(z) 2 garanties the integrability
on the real part, the integration on the imaginary part being bounded by some constant
times vol(Im(Ωε ∩ Γε )), so that the expression (2.4.5) tends to 0 whith ε tending to 0.
We remember we obtained in the precedent section that for v ∈ Ω∗ε such that
Im(v) = 0, we have
Z
1 1 ¯ ε (u) = RR f (v) + O(ε)
F (u) ∧ ∂χ
2πi u∈∂Ωε hv, ui
118 CHAPTER 2. THE CASE N = 2
But
Z
¯ ε )(z) = 1 ¯ ε (ξ)ω 0 ( ϑε (z)
L(F ∧ ∂χ 2 F ∧ ∂χ 1 )
(2πi) ξ∈∂Ωε hϑε (z), ξ − zi
Z
1 1 ¯ ε (ξ) det(ϑε (z), ∂¯z ϑε (z))
= F ∧ ∂χ
(2πi) ξ∈∂Ωε hϑε (z), ξ − zi2
2
(2.4.6)
(ϑε,1 (z) ∂z∂ 1 + ϑε,2 (z) ∂z∂ 2 )ϑε,1 (z)(ξ1 − z1 ) + (ϑε,1 (z) ∂z∂ 1 + ϑε,2 (z) ∂z∂ 2 )ϑε,2 (z)(ξ2 − z2 )
Z
+ + ¯ ε (ξ))]
F ∧ ∂χ
hϑ (z), ξ − zi 2
ξ∈∂Ωε ε
where we noted RR f (ϑε (z)) the Radon transform of f along the line directed the vector
ϑε (z) by passing through z (this ε-approximation is obtained reproducing step by step
the proof of 2.4.3). For z ∈ Ωε , let δε be the element Im(Ωε ∩ Im{hϑε (z), ξ − zi} ∩ Γ0 )
(cf.(2.4.2) for the definition of Γ0 ). By a simple coordinate translation, equation (2.4.1)
gives that
Z
1 1 ¯ ε (ξ) = (RF (., δε ))R (ϑε (z)) + O(ε)
F (ξ) ∧ ∂χ
2πi ξ∈∂Ωε hϑε (z), ξ − zi
where we noted (RF (., δε ))R (ϑε (z)) the Radon transform of the F (., δε ) along the line di-
rected by ϑε (z) passing <(z)+iδε . The Radon transform of f being null on Ω∗ and given
∂ ∂
the choice of ϑε , the component ∂ Im(z 1)
(RF (., δε ))R (ϑε (z)), ∂ Im(z 2)
(RF (., δε ))R (ϑε (z))
tend to 0 with ε.
1 ¯ ε (ξ).
R
The last term of the above equality is essentially ξ∈∂Ωε hϑε (z),ξ−zi F (ξ) ∧ ∂χ
Im(z2 ) Im(z1 )
Indeed, ϑε is forced to converge to ϑ0 = ( |Im(z)| , − |Im(z)| ) with ε (it parametrizes the
only complex line passing trough z and contained in ΩC ), so that
(ϑε,1 (z) ∂z∂ 1 + ϑε,2 (z) ∂z∂ 2 )ϑε , 1(z)(ξ1 − z1 ) + (ϑε,1 (z) ∂z∂ 1 + ϑε,2 (z) ∂z∂ 2 )ϑε,2 (z)(ξ2 − z2 )
lim
ε→0 hϑε (z), ξ − zi2
1
=
hϑ0 (z), ξ − zi
so that we can apply to the second term of (2.4.6) reasoning as for its first term.
2.4. KERNEL OF RADON TRANSFORM THROUGH ∂¯ TECHNIC 119
¯ ε )(z) = (n − 1)! ξ¯ − z̄
Z
H (F ∧ ∂χ ¯ ε (ξ) ∧ det(ϑε (z),
F ∧ ∂χ ) ∧ (ϑε (z)dξ)cω(ξ)
(2πi)n hϑε (z),ξ−zi=0 | ξ − z |2
¯ ε (z) = lim ∂H
lim F ∧ ∂χ ¯ ε )(z)
¯ (F ∧ ∂χ
ε→0 ε→0
¯ ε ), H(∂(F
As we may notice, the different estimation of L(F ∧ ∂χ ¯ ∧ ∂χ
¯ ε )), are uniform
with regards to z. Taking some test function, a smooth (0, 1)-form, ϕ ∈ Cc0,1 (R2 ), with
support K, and extending it to C2 , with the condition that ∂ϕ¯ is O(| Im(z) |N ),
Z Z
lim ¯ ε (ξ) ∧ ϕ(ξ) = lim
F ∧ ∂χ ¯ (F ∧ ∂χ
∂H ¯ ε )(ξ) ∧ ϕ(ξ)
ε→0 ξ∈Ωε ε→0 ξ∈Ωε
¯ ε ) is holomophic on ΩC \ Ω. By Stokes,
From there, we have that limε→0 H (F ∧ ∂χ
−
noting ∂Γε = Γ+
ε ∪ Γε , we have
Z Z Z Z
¯ ε (ξ) ∧ ϕ(ξ) =
F ∧ ∂χ F ∧ χε (ξ) ∧ ϕ(ξ) − ¯ ∧ χε (ξ) ∧ ϕ(ξ) −
∂F ¯
F ∧ χε (ξ) ∧ ∂ϕ(ξ)
−
ξ∈Γε ξ∈Γ+
ε ∪Γε ξ∈Γε ξ∈Γε
Z Z Z
¯ (F ∧ ∂χ
∂H ¯ ε )(ξ) ∧ ϕ(ξ) = ¯ ε )(ξ) ∧ ϕ(ξ) +
H (F ∧ ∂χ ¯ ε )(ξ) ∧ ∂ϕ(ξ)
H (F ∧ ∂χ ¯
−
ξ∈Γε ξ∈Γ+
ε ∪Γε ξ∈Γε
¯ ε )(z), for
The last term tends to 0 with ε. Let us note H ± (f )(z) = limε→0 H ± (F ∧ ∂χ
±
z in Ω , the two connected components of ΩC \ Ω. Putting all this together, we get
that f is the boundary value of H ± (f ).
Besides, we apply lemma (2.4.5) to get a nice expression of
ξ¯ − z̄
Z
± 1
H (f )(z) = − f (<(ξ)) det(ϑ(z), )(ϑ(z)dξ)cω(ξ)
2πi hϑ(z),ξ−zi=0∩Im(ξ)=0 | ξ − z |2
pour z ∈ Ω± .
120 CHAPTER 2. THE CASE N = 2
Chapter 3
The case n ≥ 2
3.1 Introduction
In this section, we mean to study the kernel, the image and a possible inversion formula
for the affine Radon transform in strongly linearly concave domains, in any dimension
greater than 2. The results we will prove were announced by Henkin in [Hen04a].
We will see that the kernel of the affine Radon transform in these domains, can
¯
be expressed as ∂−closed forms residues. Cordaro, Gindikin, Treves caracterized in
¯
[CGT95], at the germ level, residues of ∂−closed forms by means of their wave-front
set. We will broaden this result in a semi-global context (i.e. in a strongly linearly
concave domain context), and we will obtain an explicit expression, for any function
whose affine Radon transform vanishes in a strongly linearly concave domain, of a
¯
∂−closed form of which it is the residue.
f = Res(Ef ) + κ(Rf ),
where Ef represents a cohomology class iof H 0,n−2 (ΩC \ Ω), and where κ is an inversion
operator of the incomplete Radon transform (we mean incomplete in the sense that
Radon transform is given only in hyperplanes contained in Ω).
As usual, P, P ∗ resp. P, P∗ designates the n-dimensional real resp. complex projec-
tive space and its dual. In the complex case, the following result was established.
121
122 CHAPTER 3. THE CASE N ≥ 2
¯
f = Ξ(RC f ) + ∂Hf
3.1.2 Notations
We refer the reader to notations of section 2.1.1. We note Ω a strongly linearly concave
domain of Rn , Ω∗ ⊂ P ∗ the set of hyperplanes contained in Ω. Elements (θ, t) =
(θ1 : . . . : θn : t) of P ∗ , uniformised by 1≤i≤n θi2 = 1, will parametrise hyperplanes
P
For our purpose, we will consider the restriction of this transform to a linear concave
domain. For x ∈ Ω, we still note x b = {ξ ∈ Ω∗ , x ∈ ξ}. We define the same way as
b Φ being function C(Ω∗ , C), and x ∈ Ω, by
above in (3.1.1), RΦ,
Z
RΦ(x) =
b Φ(ξ)dµ(ξ)
ξ∈b
x
We then have
For any distribution D of compact support, we define the affine Radon transform
∗
RD on Ω as the linear form on the space of functions C ∞ (Ω , C) which associates to a
function Φ,
RD(Φ) = D(RΦ)
b
k
| F (z) |<
| Im(z) |m
1
where z = (z1 , . . . , zn ), | Im(z) |= ( nj=1 | Im(zj ) |2 ) 2 and | FP
P
(z) |= max#I=p | FI (z) |,
with I uplets of distinct integers between 1 and n, F = #I=p FI dz̄I , and dz̄I =
∧k=i1 ,...,ip dz̄ik .
¯
Proposition-definition 3. Let m a positive integer. Let F an (0, n−1)-form, ∂-closed
in Cn \ Rn with polynomial growth around Rn of index m ∈ N. The residue of F is then
the distribution defined by the linear form on Cc∞ (Rn , C), associating to each function
ϕ,
Z
hRes(F ), ϕi = lim F (z)ϕ(<(z)) ∧nj=1 dzj
ε−→0 z∈Rn +i.Sεn−1
Theorem 3.1.4. The definition 3.1.3 does not depend on the choice of the partition of
the unity. Under these conditions, the Radon transform of f , in the sense of distribu-
tions, vanishes along all hyperplanes of Ω.
Modulo linear transformation transformation, we can suppose that the hyperplane
{x ∈ Rn , xn = 0} is contained in Ω. We make a convenient change of variables, and we
set, for ϕ ∈ [0, π[, and for j = 1, . . . , n − 1, and (θ1 , . . . , θn ) ∈ S n−1 .
θj := αj sin(ϕ), θn := cos(ϕ)
where (α1 , . . . , αn−1 ) ∈ S n−2 .
Then, the following theorem asserts that a function whose Radon transform vanishes
¯
in Ω∗ , can be explicitely expressed as a residu of (0, n − 2)−form F ∂−closed in ΩC \ Ω.
Theorem 3.1.5. Let f ∈ C 1,2n−2 (Rn , C) such that the affine Radon transform of f
¯
vanishes in Ω∗ , then there exists a (0, n − 2)−form F , ∂−closed in ΩC \ Ω , satisfying
1
an estimate in O( |z|2n−2 ). For z ∈ ΩC \ Ω
Z
f (y)
F (z) = n
dy ∧ cos(ϕ)dϕ ∧ ω 0 (Im(z))
n
y∈R ,ϕ∈]0,π[ h(Im(z), ϕ), y − zi
where
n−1
X
h(Im(z), ϕ), y − zi = Im(zi ) sin(ϕ)(yi − zi )+ | Im(z) | cos(ϕ)(yn − zn )
i=1
Pn−1 1
where | Im(z) | designates ( i=1 | Im(zj ) |2 ) 2 , and where
ω 0 (Im(z)) = det(Im(z), ∂¯ Im(z), . . . , ∂¯ Im(z))
| {z }
n−2 fois
We then have Z
f (x) = lim F (z)
ε−→0 Im(z)∈Sεn−2
Z
(3.2.1) hRes(F ), ϕi = lim F (z)ϕ(<(z)) ∧nj=1 dzj
ε−→0 z∈Rn +i.Sεn−1
is well defined.
Using the notations of 3.1.3.1, let us remind a result obtained in [CGT95]: it is
shown there, using q−spheres, i.e. boundary of (q + 1)−dimensional polyhedron, that
¯
for any ∂−closed (0, q) form F , the residue
Z
(3.2.2) hRes(F ), ϕi = lim F (z)ϕ(z) ∧nj=1 dzj
ε−→0 z∈U +iε.C
Z
(3.2.3) δ(ε) = f (<(z) + i.ε Im(z))ϕ(<(z)) ∧nj=1 d<(zj )
z∈Rn +i.γ
Proof. Given the slow growth condition, consider g the (m + 1)−th order primitive of
f , then g is still holomorphic. We then have
3.2. DEFINITION OF RESIDUES IN STRONGLY LINEARLY CONCAVE DOMAINS127
∂ m+1 K
| f (x + iεγ) |=| m+1
g(x + iεγ) |<
∂ε | ε |m
for some constant K > 0. Now, let’s integrate m + 2 times this expression and we
get that g is continuous on Rn + iΓ̄.
Now, we can write for (3.2.3)
∂ m+1
Z
(3.2.4) δ(ε) = (g(<(z) + i.εγ))ϕ(<(z)) ∧nj=1 d<(zj )
z∈Rn +i.γ ∂εm+1
(3.2.5)
XZ ∂ m+1
δ(ε) = im+1 Im+1 !γ Im+1 g(<(z) + i.εγ)ϕ(<(z)) ∧nj=1 d<(zj )
Im+1 z∈Rn +i.γ ∂ Im(z)Im+1
where we noted Im+1 the set of n−uplets of integers (i1 , . . . , in ) such that nj=1 ij =
P
∂ m+1 i
m + 1, ∂X = Πnk=1 ∂ kik , here X = (X1 , . . . , Xn ) stands for Im(z), Im+1 ! = Π(m+1)!
n ik !
,
I m+1 ∂Xk k=1
∂ ∂
f = −i f
∂<(zk ) ∂ Im(zk )
So that, we have
XZ ∂ m+1
δ(ε) = (−1)m+1 Im+1 !γ Im+1 g(<(z) + i.εγ)ϕ(<(z)) ∧nj=1 d<(zj )
Im+1 z∈Rn +i.γ ∂<(z)Im+1
Z X Z
(3.2.7) lim F (z)ϕ(z) ∧nj=1 dzj = lim fi (z)ϕ(z) ∧nj=1 dzj
ε−→0 z∈K+i.Sεn−1 ε−→0 z∈K+i.εγi
i=1,...,n−1
128 CHAPTER 3. THE CASE N ≥ 2
Z
∂
(3.2.8) lim fi (z) ϕ(z) ∧nj=1 dzj = 0, 1 ≤ k ≤ n
ε−→0 z∈K+i.εγi ∂ z̄k
Z
(3.2.9) lim fi (z)(ϕ(z) − ϕ(<(z))) ∧nj=1 dzj = 0
ε−→0 z∈K+i.εγi
For the above two statements, the first one follows immediately from the fact that
one may take fi to have slow growth near K. And this is proved in the section 3.5. For
the second statement, consider the Taylor expansion of ϕ, we have ϕ(z) − ϕ(<(z)) =
Im(z)I ∂I
ϕ(<(z)) + O(Im(z)m+1 ), where we noted I a set of p−uplets, 0 ≤
P
#I≤m I! ∂ Im(zI )
∂ I ∂
p ≤ m, of integers between 1 and n, ∂ Im(z I)
= Πi∈I ∂ Im(zi)
, I! = Πi∈I multI (i)!, and
I
R
Im(z) = Πi∈I Im(zi ). Now, according to Cauchy-Riemann equations, fi (z)(ϕ(z) −
∂I
R
ϕ(<(z))) ∧nj=1 dzj , only involves Im(z)I fi (z) ∂<(z I)
ϕ(<(z)) ∧nj=1 dzj terms, which we
have precedently has a limit when ε tends to 0, knowing that Im(z)I tends to 0 with ε,
so that this limit is 0. R
Finally, putting to gether (3.2.7), (3.2.9) and (??), we get that limε−→0 z∈Rn +i.Sεn−1 F (z)ϕ(<(z))∧
dzj exists for ϕ ∈ Cc∞ (Rn , C).
¯
Suppose that (0, n−2)−form F is a ∂-closed 1
in ΩC \Ω, satisfying the estimate O( |z|2n−2 ),
for | z |−→ ∞ and with polynomial growth of index m ∈ N.
We want to prove that, for ϕ ∈ Cc∞ (Rn−1 , C)
Z
hRes(F|Rn−1 ), ϕi = lim F (z)ϕ(<(z)) ∧nj=1 dzj
θ,t ε−→0 z∈Rn−1 n−1 n−1
∩Rθ,0
θ,t +i.(Sε )
exists.
This comes from the following. Let’s consider a hyperplane H, (here parametrised
by real parameters (θ, t)) contained in ΩC , F the (n − 2)-closed form on ΩC \ Ω, and
an orthonormal coordinate system (Z1 , . . . , Zn ) where (Z1 , . . . , Zn−1 ) is an orthonormal
frame of H, then F expressed in dZ̄1 , . . . , dZ̄n coordinates is ∂¯Z -closed, since the ∂- ¯
closeness property is closed under holomorphic coordinates change transformation (here
3.2. DEFINITION OF RESIDUES IN STRONGLY LINEARLY CONCAVE DOMAINS129
(3.2.10) Z Z
lim ( F (z)ϕ(<(z)) ∧nj=1 dzj − F (z)ϕ(<(z)) ∧nj=1 dzj ) = 0
ε−→0 n−2
z∈Ωθ +i.Sε,θ n−2
z∈Ωθ0 +i.Sε,θ 0
n−2
where we note Sε,θ = Sεn−1 ∩ Rn−1
θ,0 . In [HW69], Hörmander and Wermer shows that
for fixed integer m and the above ϕ, there is an extension of ϕ, still noted ϕ, null for z
such that <(z) ∈ / U , and a positive real number K such that for z ∈ Cn ,
(3.2.11) ¯
| ∂ϕ(z) |≤ K | Im(z) |m+1
Z Z Z
F (z)ϕ(z)∧nj=1 dzj − F (z)ϕ(z)∧nj=1 dzj = ¯
F (z)∂ϕ(z)∧n
j=1 dzj
n−2 n−2
z∈U +i.Sε,θ z∈U +i.Sε,θ 0 z∈Dε
R From the polynomial growth condition controlling F and (3.2.11), it follows that
¯ n
F (z)∂ϕ(z) ∧j=1 dzj tends to 0 with ε.
z∈Dε
We decompose ϕ(<(z)) = ϕ(z) + ϕ(<(z)) − ϕ(z). (3.2.10) will follow from the fact
that
Z
lim F (z)(ϕ(z) − ϕ(<(z))) ∧nj=1 dzj = 0
ε−→0 n−2
z∈Ωθ +i.Sε,θ
I
∂I
If we develop the Taylor expansion of ϕ, we have ϕ(z)−ϕ(<(z)) = #I≤m Im(z)
P
I! ∂ Im(zI )
ϕ(<(z))+
m+1
O(Im(z) ). Now, we are going to construct explicitely the extension ϕ in order to
∂I
deal with the terms ∂ Im(z I)
ϕ(<(z)).
Let’s find a candidate ϕ̃ by decomposing into a sum of homogenous polynomials of
degree less than m in the Im(z) variable
130 CHAPTER 3. THE CASE N ≥ 2
m
X X
ϕi1 ,...,in Πnk=1 Im(zk )ik
j=0 (i1 ,...,in )∈Nn
i1 +...+in =j
m−1
∂ X X ∂
ϕ̃ = ( ϕi ,...,i + i.(il + 1)ϕi1 ,...,il +1,...,in Πnk=1 Im(zk )ik
∂ z̄l ∂<(zl ) 1 n
j=0 (i1 ,...,in )∈Nn
i1 +...+in =j
i ∂
(3.2.12) ϕi1 ,...,il +1,...,in = − ϕi ,...,i
il + 1 ∂<(zl ) 1 n
Since, ϕ̃|Rn = ϕ|Rn , we get ϕ0,...,0 (z) = ϕ(<(z)). Now, by induction all the functions
are constrained to be, for any n-uplet of integers (i1 , . . . , in ) such that i1 + . . . + in ≤ m,
I the uplet (i1 , . . . , in )
i(i1 +...+in ) ∂ I
(3.2.13) ϕi1 ,...,in (z) = ϕ(<(z))
Πnk=1 ik ! ∂<(z)I
Now, going back to the Taylor expansion of ϕ, from the construction of the extension,
∂I
we have to deal with terms of the type ∂<(z I)
ϕ(<(z)). But, we have shown that each
R ∂ I
term limε−→0 z∈Ωθ +i.S n−2 F (z) ∂<(z I)
ϕ(<(z)) ∧nj=1 dzj exists. Taking care of the Im(z)I
ε,θ
which converges to 0 with ε, we deduce that the Taylor expansion converges to 0, so
that we get (3.2.10). So, the proof of proposition-definition (3) is complete.
Then, it remains to prove that the residue (3.1.3) as we defined it in introduction
when we consider ϕ ∈ Cc∞ (Ω, C) with compact support K ⊂ Ω, is indeed well defined.
Let’s take open sets Ui , i ∈ I, #I finite, such that for each Ui , there exists θi ∈ π(Ω∗ ),
such that Ui ⊂ Ωθi , and ∪i∈I Ui ⊃ K. We defined the residue from the partition of unity
(ui ), i ∈ I, supp(ui ) ⊂ Ui , by
X
hf, ϕi = hf, ϕ.ui i
i∈I
hf, ϕ.u01 i + hf, ϕ.u02 i = hf, ϕ.(u1 + u2 ).u01 i + hf, ϕ.(u1 + u2 ).u02 i
3.2. DEFINITION OF RESIDUES IN STRONGLY LINEARLY CONCAVE DOMAINS131
Z Z
= F (z)ϕ(<(z))u1 u01 ∧nj=1 dzj + F (z)ϕ(<(z))u1 u02 ∧nj=1 dzj
n−2 n−2
z∈U1 ∩U10 +i.Sε,θ z∈U1 ∩U20 +i.Sε,θ
1 1
Z Z
+ F (z)ϕ(<(z))u2 u01 ∧nj=1 dzj + F (z)ϕ(<(z))u2 u02 ∧nj=1 dzj
n−2 n−2
z∈U2 ∩U10 +i.Sε,θ z∈U2 ∩U20 +i.Sε,θ
2 2
where we used what we shown above, i.e. for any compact support function ϕ,
whose support is in Ωθ ∩ Ωθ0 , for some θ and θ0
Z Z
n
F (z)ϕ(<(z)) ∧j=1 dzj = F (z)ϕ(<(z)) ∧nj=1 dzj
n−2 n−2
z∈Ωθ +i.Sε,θ z∈Ωθ +i.Sε,θ 0
which is finally
hf, ϕ.u1 i + hf, ϕ.u2 i
We suppose the Radon transform distribution has compact support. Using partition
of unity relatively to the compact support of f , (ui )i∈I with supp(ui ) ⊂ Ωθi , for some
θi ∈ S 1 , we need to prove that
XZ
lim F (z)ui (<(z))Φ(<(z))
b ∧nj=1 dzj = 0
ε−→0 n−2
z∈Ωθi +iSε,θ
i∈I i
Noting Ω∗θ ∗
= {ξ ∈ Ω , ξ ∩ Ωθ 6= ∅}, then this expression equals
XZ Z
lim F (z)ui (<(z))Φ(ξ) ∧nj=1 dzj ∧ dµ(ξ)
ε−→0 ξ∈Ω∗ n−2
z∈ξ∩Ωθi +iSε,θ
i∈I i
∗
Let ξ ∈ Ω and (Z1,ξ , . . . , Zn−1,ξ ) a positively oriented orthonormal basis of the
complexification of ξ, then applying Stokes formula to the unbounded domain of edge
n−2 1
(ξ + i.Sε,θ ), and taking into account that F = O( |z|2n−2 ), the following integral tends
to 0 Z
F (z) ∧n−1
j=1 dZj,ξ
n−2
z∈ξ+iSε,θ
P
We deduce, using i∈I ui = 1, that the Radon transform of f vanishes.
132 CHAPTER 3. THE CASE N ≥ 2
Let’s consider ε > 0 and Bε the ball {ξ ∈ Cn , | Im(ξ) |2 ≤ ε}. Let’s take a function
¯ (z) |= O(
f ∈ C 1 (Bε ) such that | f (z) | + | ∂f 1
) when | z | tends to infinity.
(1+|<(z)|)2n−2
With classical notations, we get some Cauchy-type Radon transform inversion for-
mula that was formulated by John [Joh35], and recovered by Henkin [Hen14] by means
of Cauchy-Leray formula, for f ∈ C 1,2n−2 (Rn ) :
(n − 1)!
Z Z
f (y)
(3.3.1) f (x) = limε−→0 ω 0 (θ) ∧ dy
(2iπ)n y∈Rn θ∈S n−1 (hθ, y − xi + iε)n
and so
(−1)n−1 dn−1
Z
f (x) = limε−→0 Kε ( n−1 Rf )(θ, s)s=hθ,xi ω 0 (θ)
2(2iπ)n−1 θ∈S n−1 ds
where Kε designates Rthe operator that takes each integrable real function ϕ to the
ϕ(s)
function Kε ϕ(t) = iπ1 s∈R s−t+iε ds.
3.3.2 ¯
Affine Radon transform’s kernel as residue of ∂−closed
(n − 2)-form
First, we are going to proceed to a change of variables. For x ∈ Ω, let Ω∗x = {(θ, t) ∈
Ω∗ , hθ, xi = t}, i.e. the set of hyperplanes of Ω passing through x, and Ωx the set of
points of these very hyperplanes. The affine Radon transform of f vanishing on Ω∗ ,
there exists for any element θ in π(Ω∗x ) (cf. notations in the introduction section) real
numbers s1,θ , s2,θ ∈ R̄ such that for any s1,θ < s < s2,θ , affine Radon transform of f
vanishes on each hyperplane {y ∈ Rn , hθ, yi = s}.
We get from (3.3.1)
(n − 1)!
Z
f (y)
(3.3.2) f (x) = limε−→0 ω 0 (θ) ∧ dy
(2iπ)n y∈Rn ,θ∈S n−1 \π(Ω∗x ) (hθ, y − xi + iε)n
(n − 1)!
Z
f (y)
(3.3.3) + limε−→0 ω 0 (θ) ∧ dy
(2iπ)n y∈Rn ,θ∈π(Ω∗x ),hθ,yi∈]s
/ 1,θ <s<s2,θ [ (hθ, y − xi + iε)n
3.3. RESIDUES AS KERNEL OF RADON TRANSFORM IN STRONGLY LINEARLY CONCAVE DOMAIN
Reminder
Case when proper // Check the labelling 000-001
θi := αi sin(ϕ), θn := cos(ϕ)
where (α1 , . . . , αn−1 ) ∈ S n−2 , ϕ ∈ [0, π[.
For α ∈ S n−2 , there exists ϕ1,α,x , ϕ2,α,x ∈ [0, π[ such that (α, ϕ1,α,x ), (α, ϕ2,α,x ) ∈
∂π(Ω∗x ). (Indeed, the image of the open π(Ω∗x ) is union of two connected open set S n−1 ,
symetric to each other, so that the intersection of the circle described by the elements
(α, ϕ), with α fixed reduces to 4 elements and therefore two by symetry when we reduce
our parameter circle to the half-circle ϕ ∈ [0, π[).
We have
(n − 1)!
Z
f (y)
(3.3.2)+(3.3.3) = limε−→0 n
( n
cos(ϕ)dϕ∧ω 0 (α)∧dy
(2iπ) α∈S n−2 n
,y∈R ,ϕ∈]ϕ1,α,x ,ϕ2α,x [ (h(α, ϕ), y − x) + iε)
Z
f (y)
+ n
cos(ϕ)dϕ∧ω 0 (α))∧dy)
α∈S n−2 ,ϕ∈]0,ϕ1,α,x [∪]ϕ2,α,x ,π[,h(α,ϕ),yi∈]s
/ 1,(α,ϕ) <s<s2,(α,ϕ) [ (h(α, ϕ), y − xi + iε)
Pn−1
where h(α, ϕ), y − xi stands for i=1 αi sin(ϕ)(yi − xi ) + cos(ϕ)(yn − xn ).
(n − 1)!
Z
f (y)
(3.3.4) F (z) = cos(ϕ)dϕ ∧ ω 0 (Im(z)) ∧ dy
(2iπ)n y∈Rn ,ϕ∈]ϕ1,z ,ϕ2,z [ h(Im(z), ϕ), y − zin
(n − 1)!
Z
f (y)
(3.3.5) + cos(ϕ)dϕ ∧ ω 0 (Im(z)) ∧ dy
(2iπ)n y∈Rn ,ϕ∈]0,ϕ1,z [∪]ϕ2,z ,π[ h(Im(z), ϕ), y − zin
where we note
n−1
X
h(Im(z), ϕ), y − zi = Im(zi ) sin(ϕ)(yi − zi )+ | Im(z) | cos(ϕ)(yn − zn )
i=1
134 CHAPTER 3. THE CASE N ≥ 2
2 21
where | Im(z) | is ( n−1 0
P
i=1 | Im(zj ) | ) , and Im(z) = (Im(z1 ), . . . , Im(zn−1 )), ω (Im(z)) =
¯ ¯
det(Im(z), ∂ Im(z), . . . , ∂ Im(z)) and where ϕ1,z and ϕ2,z are elements corresponding
respectively to ϕ , ϕ ^ designating the vector ( Im(z1 ) , . . . , Im(zn−1 ) ),
, Im(z)
^
1,Im(z),x ^
1,Im(z),x |Im z| |Im(z)|
(corresponding to the element α ∈ S n−2 used above).
^ ϕ), y − xi, we have
Let’s find out the domain of definition of F . Let s = h(Im(z),
Z
f (y)
Z ^ ϕ), s)
Rf ((Im(z),
dy = ds
y∈Rn h(Im(z), ϕ), y − zin s∈R | Im(z) |n .(s − ih(Im(z),
^ ϕ), Im(z)i)n
where ((Im(z), ϕ), s) parametrize the hyperplane {y ∈ Rn , h(Im(z), ϕ), yi = s}. For
(3.3.5), considering z ∈ ΩC,x \Ωx , ϕ ∈]0, ϕ1,z [∪]ϕ2,z , π[, then Rf ((Im(z), ϕ), h(Im(z), ϕ), xi) =
0, because the hyperplane is in Ω∗ . In the integrand of (3.3.5), h(Im(z), ϕ), y − xi is
not null for x ∈ Ω. As for (3.3.4), h(Im(z), ϕ), Im(z)i = 6 0. Indeed ΩC is the complexifi-
cation of all hyperplanes contained in Ω, so for ϕ ∈]ϕ1,z , ϕ2,z [, the complex hyperplane
{Z ∈ Cn , h(Im(z), ϕ), Z − xi = 0} is by definition not contained in ΩC (cf. definition
at the beginning of (3.3.2)) and, taking the imaginary part h(Im(z), ϕ), Im(z)i = 6 0} for
z ∈ ΩC \ Ω.
Finally, from the above expression of (3.3.2) + (3.3.3) on one hand and of (3.3.4) +
(3.3.5) on the other, we have
Z
f (x) = limε−→0 F (z)
Im(z)∈Sεn−2
3.3.3.1 ¯
Where we show F is ∂-closed
¯
The ∂-closedness comes from the following
Lemma 3.3.1. The form
1
Pn−1 det(Im(z), ∂¯ Im(z), . . . , ∂¯ Im(z))
( Im(zj )zj + | Im(z) | .zn )n | {z }
j=1
n−2 fois
¯ 2 12
where we still note | Im(z) | pour ( n−1
P
is ∂-closed, i=1 | Im(zj ) | ) , and Im(z) =
(Im(z1 ), . . . , Im(zn−1 ))
Proof. The ∂¯ of this form is
n−1
X i.zk + i. Im(z k ).zn
|Im(z)| n−1
−n (n − 1)! Im(zk ) ∧l=1 dz̄l
( n−1
P n+1
k=1 j=1 Im(z )z
j j + | Im(z) | .zn )
n−1
X 1 n−1
+ Pn−1 i.(n − 1)! ∧l=1 dz̄l
( Im(zj )zj + | Im(z) | .zn )n
k=1 j=1
3.4. INVERSION FORMULA FOR N ≥ 2 135
( n−1
P
j=1 Im(zj )zj + | Im(z) | .zn )
−n!.i Pn−1 . ∧n−1
l=1 dz̄l
( j=1 Im(zj )zj + | Im(z) | .zn ) n+1
1 n−1
+n!.i Pn−1 ∧l=1 dz̄l = 0
( j=1 Im(zj )zj + | Im(z) | .zn )n
f (y)
dy ∧ cos(ϕ)dϕ ∧ ω 0 (Im(z))
h(Im(z), ϕ), y − zin
Proof. Let us take K proper, otherwise there is no work to provide, and the extension
is trivial. Let us note K ∗ the dual of K, i.e. the set of hyperplanes with no empty
intersection with K. Let’s take a frame pointed in K. We must extend g, known for
hyperplanes in the complementary of K, to the set of hyperplanes crossing K. Let ϕ0
be the positive valued function of S n−1 such that (θ, ϕ(θ)) parametrizes the hyperplane
tangent to ∂K.
Now, for θ ∈ S n−1 , set for 0 < t < ϕ(θ), g̃(θ, ϕ(t)) = g(θ, ϕ(θ)) (since we took K
containing 0, and since it is proper, ϕ(θ) 6= 0), and g̃(θ, 0) = 0. Then, consider two
open intervals B 0 ( B, neighborhoods of 0, such that {(θ, t), θ ∈ S n−1 , t ∈ B} ⊂ K ∗
and a C ∞ test function χ of support included in S n−1 ×R\B 0 , which is 1 on S n−1 ×R\B,
such that χ(θ, t) = χ(−θ, −t) (it suffices to take an even bump function, whose support
is included in R \ B 0 , 1 on R \ B, not depending on θ). Now considering g̃.χ, we get
the sought extension because being null in {(θ, t), θ ∈ S n−1 , t ∈ B 0 }, this extension is
correctly defined. Hence we constructed an extension g̃ ∈ S(S n−1 × R, C).
Now, following [Nov02], it is proved that if g belongs to the schwartz space S(S n−1 ×
R, C), if g(θ, t) = g(−θ, −t), (θ, t) ∈ S n−1 × R, then there exists f ∈ C∞,n (Rn , C), such
that Rf = g.
136 CHAPTER 3. THE CASE N ≥ 2
Reminder
check the é typos
3.5 ¯
Estimates for solution of ∂−solution in polydisc
In the proof of proposition XXX, we will need the following statement which is inter-
esting by itself. Namely, given a (0, p)-form f in the polydisc with slow growth at the
¯ = f in the polydisc such
boundary, then one can find a solution to the d-bar problem ∂u
that u has also slow growth at the boundary. We follow the work and the notations of
Bo Berntsson XXX
3.5.1 Notations
¯
Let p ∈ N. Let f be a (0, p)-form which is ∂-closed in the n-dimensional unit polydisc
D and has polynomial growth of index m ∈ N at the boundary ∂Dn , meaning that there
n
is some real constant k and a complex neighborhood V of ∂Dn such that for z ∈ V ∩ Dn
k
| F (z) |< .
(1 − min | zi |)m
1≤i≤n
First, let us recall the construction ofVa kernel K through Henkin-type homotopy
formula. We note, for η ∈ Cn ω(η) = dηi . We define a Leray section to be an
1≤i≤n
n n n
application s : D̄ × D̄ → − C , such that hs(ξ, z), ξ − zi 6= 0. We suppose also that s is
of class C 1 , verifies, uniformly with respect to ξ ∈ D̄n and z in any compact subset of
Dn , | s(ξ, z) |≤ C1 | ξ − z |, | hs(ξ, z), ξ − zi |≥ C2 | ξ − z |2 .
We define also a map Q = (Q1 , . . . , Qn ) : D̄n × D̄n →− Cn , holomorphic in z ∈ Dn for
ξ ∈ D fixed. From there, we define the homotopy mapping ψ : D̄n × D̄n \ ∆D̄n ×]0, ∞[−
n
→
n n
C × C , ψ(ξ, z, t) = (Q(ξ, z) + ts(ξ, z), ξ − z).
¯
3.5. ESTIMATES FOR SOLUTION OF ∂−SOLUTION IN POLYDISC 137
Let A be the form exp(hξ, ηi)ω(ξ) ∧ ω(η). The kernel K will be part of the pull-
back ψ ∗ A. We decompose ψ ∗ A = (ψ ∗ A)t + (ψ ∗ A)ξ,z so that (ψ ∗ A)t is the component
involving dt, then we define
Z ∞
K= (ψ ∗ A)t
0
Then it can be proven that
X (n − 1)! s ∧ (dQ)k ∧ (ds)n−1−k
K = −C.ehQ,ξ−zi
0≤k≤n−1
k! hs, ξ − zin−k
for some positive constant C.
To produce a weighted kernel, Berndtson XXX proceeds to some Laplace transform.
Let us consider, for some positive parameter λ, the parametric Kλ obtained by replacing
Q by λQ, and g be distribution on [0, ∞[. Then the new kernel is
Z ∞
K̃ := C Kλ .e−λ g(λ)dλ
0
with C some normalisation constant. The, a simple calculation leads to
X (n − 1)! (k) s ∧ (dQ)k ∧ (ds)n−1−k
K̃ = −C G (hQ, z − ξi + 1)
0≤k≤n−1
k! hs, ξ − zin−k
where G is the laplce transform of g and G(k) is its k-th derivative and where C is some
positive constant.
Then one has the following Koppleman formula,
Proposition 3.5.1. For f a (p, q)-form, q > 0, in C 1 (D̄n ), we have
Z Z Z
f =C f ∧ K̃p,q + (−1)p+q+1
( ¯
∂f ∧ K̃p,q − ∂z ¯ f ∧ K̃p,q−1 )
∂Dn Dn Dn
where Kp,q is of degree (p, q) in z, (n − p, n − q − 1) in ξ, and C a positive constant
depending exclusively on p, q, n.
This formula is valid on any domain in Cn , with supplementary assumptions on f
¯ to be integrable. See XXX for the formula q = 0.
and ∂f
so that
∧ (dQ)k ∧ (ds)n−1−k
P
(n − 1)! −
Z λi s
˜ := −C
K̃
X
hQλ1 ,...,λn ,ξ−zi
e e 1≤i≤n
dλ1 ...dλn
0≤k≤n−1
k! (λ1 ,...,λn )∈[0,∞[n hs, ξ − zin−k
and
where (dQi1 ,...,in )k is the (j1 , . . . , jk )-minor determinant of (dQ)k , where (j1 , . . . , jk ) is
the k-uplet such that ijl = 1, 1 ≤ l ≤ k.
−ξ̄ Q 1
Now, let us choose, Q with Qj = 1−|ξjj |2 , g such that G(x1 , . . . , xn ) = Kj .
j=1,...,n xj
This way we get:
and so:
(3.5.1)
˜ := −C X (n − 1)!Ck Y 1− | ξj |2 Kj +ij s ∧ (dQi ,...,i )k ∧ (ds)n−1−k
1 n
K̃ ¯j zj
0≤k≤n−1
k! j=1,...,n
1 − ξ hs, ξ − zin−k
i1 +...+in =k
ij =0,1;1≤j≤n
We then have
Proposition 3.5.3. For f a (p, q)-form, q > 0, in C 1 (Dn ), with slow growth at the
boundary ∂Dn , the ∂¯ equation ∂u
¯ = f in Dn has a solution with slow growth at the
boundary.
¯
3.5. ESTIMATES FOR SOLUTION OF ∂−SOLUTION IN POLYDISC 139
Proof. One apply proposition (3.5.2) in Dnε , the ε-radius polydisc, with 0 < ε < 1, so
that the integrands are well-defined, and then we let ε tends to 1 to get
Z
u = (−1) p+q ˜
f ∧ K̃ p,q
Dn
141
142 REFERENCES
[GS77] V Guillemin and S Sternberg. AMS, Providence, Rhode Island, 1977. 14.
1977.
[Hel64] Sigurdur Helgason. “A duality in integral geometry; some generalizations of
the Radon transform”. In: (1964).
[Hel80] Sigurdur Helgason. The radon transform. Vol. 2. Springer, 1980.
[Hen04a] G. M. Henkin. “Abel-Radon Transform and Applications”. In: (2004). Ed.
by Olav Arnfinn Laudal and Ragni Piene, pp. 567–584.
[Hen77] Gennadi Henkin. “The Lewy equation and analysis on pseudoconvex mani-
folds”. In: Uspekhi Matematicheskikh Nauk 32.3 (1977), pp. 57–118.
[Hen14] Gennadi Henkin. “Analyse complexe plusieurs variables”. In: Cours de DEA,
Universite Pierre et Marie Curie (2014).
[HP86] Gennadi M. Henkin and Petr L. Polyakov. “Homotopy formulas for the ∂- ¯
n
operator on CP and the Radon-Penrose transform”. In: Izvestiya Rossiiskoi
Akademii Nauk. Seriya Matematicheskaya 50.3 (1986), pp. 566–597.
[Hen04b] G.M. Henkin. Analyse complexe plusieurs variables. Cours de DEA, Uni-
versit Pierre et Marie Curie. 2004.
[HP90a] GM Henkin and PL Polyakov. “Residue integral formulas and the Radon
transform for differential forms on q-linearly concave domains”. In: Mathe-
matische Annalen 286.1 (1990), pp. 225–254.
[HP90b] GM Henkin and PL Polyakov. “Residue integral formulas and the Radon
transform for differential forms onq-linearly concave domains”. In: Mathe-
matische Annalen 286.1 (1990), pp. 225–254.
[Hör83] Lars Hörmander. The Analysis of Linear Partial Differential Operators I-IV,
Grundlehren, vol. 256 (1983), 257 (1983), 274 (1985), 275 (1985). Springer-
Verlag, 1983.
[HW69] Lars Hörmander and John Wermer. “Uniform approximation on compact
sets in C n”. In: Mathematica Scandinavica 23.1 (1969), pp. 5–21.
[Joh35] F. John. “Abhngigkeiten zwischen den Flchenintegralen einer stetigen Funk-
tion.” In: Mathematische Annalen 111 (1935), pp. 541–559. url: http :
//eudml.org/doc/159791.
[Kas03] Masaki Kashiwara. “D-modules and microlocal calculus”. In: Translations
of mathematical monographs (2003).
[KKK86] Masaki Kashiwara, Takahiro Kawai, and Tatsuo Kimura. Foundations of
Algebraic Analysis (PMS-37). Princeton University Press, 1986.
[KS85] Masaki Kashiwara and Pierre Schapira. Microlocal study of sheaves. Société
mathématique de France, 1985.
[KS90] Masaki Kashiwara and Pierre Schapira. Sheaves on manifolds. Vol. 292.
Grundlehren der Mathematischen Wissenschaften. Springer-Verlag, Berlin,
1990, pp. x+512.
REFERENCES 143
[KS96] Masaki Kashiwara and Pierre Schapira. Moderate and formal cohomology
associated with constructible sheaves. Citeseer, 1996.
[KS97] Masaki Kashiwara and Pierre Schapira. “Integral transforms with exponen-
tial kernels and Laplace transform”. In: Journal of the American Mathemat-
ical Society 10.4 (1997), pp. 939–972.
[KS14] Masaki Kashiwara and Pierre Schapira. “Microlocal Euler classes and Hochschild
homology”. In: J. Inst. Math. Jussieu 13 (2014), pp. 487–516.
[KN68] Å. Kolm and B. Nagel. “A generalized edge of the wedge theorem”. In:
Communications in Mathematical Physics 8.3 (1968), pp. 185–203.
[Ler59] Jean Leray. “Le calcul différentiel et intégral sur une variété analytique com-
plexe.(Problème de Cauchy. III.)” In: Bulletin de la Société mathématique
de France 87 (1959), pp. 81–180.
[Mar67] André Martineau. “Equations differentielles d’ordre infini”. In: Bulletin de
la Société Mathématique de France 95 (1967), pp. 109–154.
[Mar68] André Martineau. “Théoremes sur le prolongement analytique du type edge
of the wedge theorem”. In: Séminaire Bourbaki 10 (1968), pp. 445–461.
[Nat01] Frank Natterer. The mathematics of computerized tomography. SIAM, 2001.
[Nov02] RG Novikov. “About asymptotic formulas for the inverse Radon transform”.
In: Bulletin des sciences mathematiques 126.8 (2002), pp. 659–673.
[NH87] Roman Novikov and Gennadi Henkin. “The-equation in the multidimen-
sional inverse scattering problem”. In: Russian Mathematical Surveys 42.3
(1987), p. 109.
[Qui01] Eric Todd Quinto. “Radon transforms, differential equations, and microlocal
Analysis”. In: 278 (2001), p. 57.
[Rad17] Johann Radon. “Uber die bestimmung von funktionen durch ihre integral-
werte langs gewisser mannigfaltigkeite”. In: Contemp. Mathematics (1917),
pp. 262–277.
[SKK] Mikio Sato, Takahiro Kawai, and Masaki Kashiwara. “Microfunctions and
pseudo-differential equations”. In: ().
[Sch85] Pierre Schapira. Microdifferential systems in the complex domain. Vol. 269.
Grundlehren der Mathematischen Wissenschaften. Springer-Verlag, Berlin,
1985.
[SS94] Pierre Schapira and Jean-Pierre Schneiders. “Index theorem for elliptic pairs”.
In: Astérisque 224 (1994), pp. 1–4.
[Trè95] François Trèves. “Boundary values of cohomology classes as hyperfunc-
tions”. In: Journées équations aux dérivées partielles (1995), pp. 1–20.
[Vla61] Vasilii Sergeevich Vladimirov. “Construction of envelopes of holomorphy for
regions of a special type and their application”. In: Trudy Matematicheskogo
Instituta imeni VA Steklova 60 (1961), pp. 101–144.