You are on page 1of 37

Graduate School of Mathematical Sciences of the University of Tokyo, dur-

ing the month of October 2015. They were meant to be as self-contained


as possible, taking into account time and space. The basic idea was to
introduce the concepts and some of the basic results in the theory of holo-
morphic foliations with singularities. Another goal is to guide the reader
to some of the recent questions and problems in the field, providing in this
way a motivating introduction to those who are interested in studying a
new subject.
I wish to express my gratitude to Professor Taro Asuke for his personal
effort in making this project possible and for his warm hospitality. I want
to thank Professor T. Tsuboi for his kind hospitality and support. I want
to thank all those at the Graduate School of Mathematical Sciences of the
University of Tokyo for their support and for making my stay in Tokyo
such a pleasant and fruitful period.
The first draft of this text was conceived during a visit to the Instituto
A swift introduction to holomorphic foliations de Matemática y Ciencias Afines - IMCA (Lima). It is my honor to present
these notes in the series Monografias del IMCA. I am very much indebted
with singularities with this prestigious Institute for all the support during the last couple
of decades. My special thanks to Professors Félix Escalante and Roger
Metzger for their warm hospitality.
Bruno Scárdua
Rio de Janeiro, November 2018
Bruno Scárdua

Preface Chapter 1

The theory of foliations is one of those subjects in mathematics that gathers


The classical notions of
several distinct domains such as topology, dynamical systems and geometry,
among others. Its origins go back to the works of C. Ehresmann and Shih foliations
([23], [24]) and G. Reeb ([62, 63]). It provides an interesting and valuable
approach to the qualitative study of dynamics and ordinary differential
equations on manifolds.
Although its origins are in the classical framework of real functions and This chapter is intended to introduce the classical notions of foliation in the
manifolds, the notion of foliation is also very useful in the holomorphic real framework. The reader which is already familiar with these notions
world. Indeed, it has ancient origins in the study of complex differential may skip to the next chapter. We refer to [9], [26], [32] or [54] for a more
equations. From these first problems, the introduction of singularities as complete exposition of the theory of real foliations.
an object of study is a natural step. We mention the works of P. Painlevé
([55, 56]) and Malmquist ([47]). With P. Painlevé the study of rational
complex differential equations of the form dx dy P (x,y)
= Q(x,y) has its first more 1.1 Definition of foliation
specific methods and results. After Painlevé many authors have contributed
for the initial push up of the theory, among them are E. Picard, G. Darboux, There are some ways of motivating the concept of foliation. Probably,
H. Poincaré, H. Dulac, Briot and Bouquet. the very first is given by a submersion f : M → N from a manifold M
Complex differential equations appear naturally in mathematics and in into a manifold N . If f is sufficiently differentiable (usually of class C r ,
natural sciences ([3, 4, 33]). For instance, we mention the theory of electri- r ≥ 2) then by the local form of submersions, the level sets f −1 (y), y ∈ N
cal circuits, valves and electromagnetic waves ([53]). Another motivation are embedded submanifolds of M . These fibers are locally organized as
is the search for and study of new (classes of) transcendent functions, as the fibers of a projection (x, y) → y. This local picture is not necessarily
the Liouvillian functions ([68]). global, and the fibers may be disconnected.
With the advent of the geometric theory of foliations and the modern A second important example is given by a closed non-singular 1-form
results of Cartan, Oka, Nishino, Suzuki and others, on the theory of analytic ω on a manifold M . Again, under sufficient differentiability conditions,
functions of several complex variables and some from algebraic and analytic by the integration lemma of Poincaré we can write locally ω = df for a
geometry, this field of research became quite active again. To these days it submersion map f taking values on K. Here K is the field of real numbers
is one of the active branches of modern research in mathematics. if ω is differentiable and M is a real differentiable manifold. In case M is
These are the notes of a series of lectures delivered by the author at the a complex manifold and ω is a holomorphic 1-form on M we have K = C
the field of complex numbers. In this later case f is holomorphic. Any
local function f as above, defined in an open subset in M , is called a first The map gi,j is a local diffeomorphism in its domain of definition. This
integral for ω. follows from the fact that the derivative of the transition map is given
Notice that two local first integrals f and f for ω in a same connected by D(ϕj ◦ (ϕi )−1 )(x, y) · (v, w) = (∂x fi,j (x, y) · v, Dgi,j (y) · w), (v, w) ∈
subset of M are related by f = f + constant. Therefore, they share level Rm−n × Rn . We define for all i the map gi = Π2 ◦ ϕi , where Π2 is the
sets, these local sets can therefore be globalized as immersed (locally closed) projection onto the second coordinate: Π2 : Dm−n × Dn → Dn , (x, y) → y.
submanifolds of M , again locally organized as fibers of a projection. We claim that gj = gi,j ◦ gi . Indeed, we have gi,j ◦ gi = gi,j ◦ Πi2 ◦ ϕi =
The third and last basic example we shall mention is the one provided Πj2 ◦ (ϕj ◦ ϕ−1 j
i ) ◦ ϕi = Πj ◦ ϕj = gj . Therefore, a C foliation F of
r

by a differentiable, namely C r with r ≥ 1, vector field X on a manifold M . codimension n of a manifold M is equipped with an open cover {Ui }i∈I
m

Given a non-singular point p ∈ M which is not a singular point of X, the of M and C r submersions gi : Ui → Dn such that for all i, j there is a local
flow-box theorem gives a conjugation between (X, U ), where p ∈ U ⊂ M diffeomorphism gi,j : Vi ⊂ Dn → Vj ⊂ Dn satisfying the cocycle relations
is an open neighborhood, and a constant vector field on Rm , where m =
dim M . The orbits of X in U then follow the same geometrical condition of gj = gi,j ◦ gi , gi,i = Id .
the above examples. The above examples motivate the classical definition
The gi ’s are the distinguished maps of F . 
of foliation below as follows:
Conversely, suppose that M m admits an open cover M = Ui such
Definition 1.1.1 (foliation, [9],[26],[32],[54]). Given a differentiable man- i∈I
that for each i ∈ I there is a C r submersion gi : Ui → Dn such that for
ifold M of dimension m and class C r , r ≥ 0; by a codimension 0 ≤ n ≤ m
all i, j there is a diffeomorphism gi,j : Vi ⊂ Dn → Vj ⊂ Dn satisfying the
foliation of class C r of M , we mean an atlas F = {(Uj , ϕj )}j∈J of M ,
cocycle relations above. By the local form of the submersions we can assume
where each coordinate chart ϕj : Uj ⊂ M → ϕj (Uj ) ⊂ Rm−n × Rn is of
that for each i ∈ I there is a C r diffeomorphism ϕi : Ui → Dm−n × Dn
class C r and we have the following compatibility condition:
such that
For each non-empty intersection Ui ∩ Uj = φ, the corresponding change
gi = Π2 ◦ ϕi .
of coordinates
 since
ϕj ◦ ϕ−1  : ϕi (Ui ∩ Uj ) −→ ϕj (Ui ∩ Uj )
i ϕ i (Ui ∩Uj )
Π2 ◦ (ϕj ◦ (ϕi )−1 ) = gj ◦ (ϕi )−1 = gi,j ◦ gi ◦ (ϕi )−1 = gi,j ◦ Π2 ,
preserves the natural horizontal fibration y = const. of Rm−n × Rn (x, y).
we have that the atlas
This is equivalent to say that, in coordinates (x, y) ∈ Rm−n × Rn , we F = {(Ui , ϕi )}i∈I
have  
ϕj ◦ ϕ−1 defines a foliation of class C r and codimension n of M . The above suggests
i (x, y) = hij (x, y), gij (y) ∈ R × Rn .
m−n
the following equivalent definition of foliation.
The charts ϕj : Uj → ϕj (Uj ) are called foliation charts, trivializing charts
or distinguished charts of F . The local plaques of F are the fibers of a Definition 1.2.2. A foliation of M m of class C r and of codimension n, is
foliation chart in F . Given a diffeomorphism ϕ : U → ϕ(U ) ⊂ Rm = given by the following:
Rm−n × Rn of class C r , we say that ϕ is compatible with the foliation F if
1. An open cover {Ui : i ∈ I} of M .
for any j ∈ J such that Uj ∩ U = ∅, we also have
  2. A family of C r submersions gi : Ui → Dn , ∀i ∈ I; with the following
ϕj ◦ ϕ−1 (x, y) = h(x, y), g(y) ∈ Rm−n × Rn .
compatibility property: ∀i, j ∈ I with Ui ∩ Uj = ∅, there is a local
In short, this is equivalent to say that F ∪ {(U, ϕ)} is still a foliation. Using diffeomorphism gi,j : Vi ⊂ Dn → Vj ⊂ Dn satisfying the cocycle
this and Zorn’s lemma, we may consider the foliation atlas F as maximal, relations
in the sense that it contains all the compatible charts of class C r of M . gj = gi,j ◦ gi , gi,i = Id .

In M we consider the equivalence relation induced by the connected The submersions gi ’s are the distinguished maps of the foliation F .
finite union of local plaques. This means that two point x, y ∈ M are
equivalent x ∼ y iff x and y lie in the same plaque of F or there is a This last definition leads to several interesting definitions. For instance,
finite number of plaques P1 , ..., Pr , r ≥ 2; of F such that x ∈ P1 , y ∈ Pr a foliation F of M is said to be transversely holomorphic or transversely
and Pi ∩ Pi+1 = ∅ for all i = 1, ..., r − 1. Given a point x ∈ M we call affine depending on whether, for some convenient choice, its distinguished
the corresponding equivalence class [x] ⊂ M is the leaf of F through x. maps gi,j are holomorphic or affine maps. We shall resume this subject later
Usually we denote this leaf by Fx or by Lx . The leaf Lx ⊂ M is an on. In order to distinguish foliations, we shall use the following definition.
immersed C r submanifold, but not necessarily embedded. These leaves Definition 1.2.3. Two foliations F and F  of manifolds M and M  re-
then decompose M into disjoint immersed C r submanifolds. Each leaf has spectively are C r -equivalent if there is a C r -diffeomorphism h : M → M 
dimension m − n and meets a foliation chart domain along plaques of the (h is a homeomorphism if r = 0), sending leaves of F into leaves of F  .
foliation. For instance, in the case of a submersion f : M → N , the leaves In other words, if Fx denotes the leaf of F that contains x ∈ M and Fy
of the corresponding foliation are the connected components of the level denotes the leaf of F  that contains y ∈ M  then we have:
sets f −1 (y), y ∈ N . The quotient space M/ ∼ is the leaf space of F , also

denoted by M/F . h(Fx ) = Fh(x) , ∀ x ∈ M.

The above notion can be stated for the case of holomorphic objects,
1.2 Other definitions of foliation in the obvious way. This relation defines an equivalence in the space of
foliations.
According to the literature there are essentially three ways to define folia-
tions in the real differentiable manifolds (cf. [9],[26], [32]). In addition to
the one we just have given in Definition 1.1.1 above, we have the following. 1.3 Frobenius theorem
Let M be a m-dimensional manifold, m ∈ N. Let Dk be the open unit ball
of Rk where k ∈ N. Let 0 ≤ n ≤ m be fixed. Let X, Y two vector fields on a manifold M and p ∈ M be fixed. Denote by
Definition 1.2.1. A foliation of M , of codimension n and class C r , is a Xt the local flow of X and similarly by Yt the local flow of Y assuming that
partition F of M consisting of pairwise disjoint immersed C r submanifolds X, Y ∈ C r , r ≥ 2. Given p ∈ M we define Xt∗ (Y )(p) = DX−t (Xt (p)) ·
L ⊂ M of dimension m − n, distributed as follows: for each point x ∈ M Y (Xt (p)) ∈ Tp (M ). Note that Xt∗ (X)(p) = X(p), ∀ t. All this holds for
there is a neighborhood U of x, and a C r diffeomorphism ϕ : U → Dm−n × |t| small enough.
Dn , such that for each y ∈ Dn there is L ∈ F satisfying Definition 1.3.1. The Lie bracket of X, Y is the vector field [X, Y ] on M
ϕ−1 (Dm−n × y) ⊂ L. defined at each point p ∈ M by

The elements of the partition F are the leaves of F . The element Lx of F d 


LX (Y )(p) = [X, Y ](p) = (X ∗ (Y )(p)) X, Y ∈ C r , r ≥ 2.
containing x ∈ M is the leaf of F containing x. dt t=0 t
We observe that, not every decomposition of M into immersed subman- In local coordinates (x1 , ..., xm ) ∈ M , the Lie bracket [X, Y ] has the
ifolds with the same dimension is a foliation (see [54]). following form: writing
The third definition of foliation uses the notion of distinguished maps.
Let F = {(Uj , ϕj ) , j ∈ J} be a foliation of a manifold M in the sense of 
m
∂  m

X= ai , Y = bi
Definition 1.1.1. Then ∀i, j the transition map ϕj ◦ (ϕi )−1 has the form ∂xi ∂x i
i=1 i=1
ϕj ◦ (ϕi )−1 (x, y) = (fi,j (x, y), gi,j (y)).
one has Remark 1.4.1. Given a manifold Σ of class C r and a point p ∈ Σ we
m 
∂bj ∂aj ∂ denote by Diff r (Σ, p) the group of germs of diffeomorphisms of class C r
[X, Y ] = ai − bi .
∂xi ∂xi ∂xj fixing p ∈ Σ. Each germ is induced by a C r diffeomorphism f : (U, p) →
i,j=1
(V, p), where U, V ⊂ Σ are neighborhoods of p, and f (p) = p.
When X and Y are defined in an open set of Rm , the formula above
yields The “return map” π : (Σ, p) → (Σ, p), x → π(x) above illustrated is, always
[X, Y ] = DY (p) · X(p) − DX(p) · Y (p). under suitable differentiability conditions, a germ of diffeomorphism.
Identifying the transverse section Σ with a disc in Rm−1 centered at the
A vector field X on M is tangent to a plane field P on M (denoted by origin 0 ∈ Rm−1 (m = dimension of the ambient manifold) we may consider
X ∈ P ) if X(p) ∈ P (p) for all p ∈ M . f as a (germ of a) diffeomorphism (Rm−1 , 0) → (Rm−1 , 0) (fixing the
origin).
Definition 1.3.2. A plane field P on M is involutive if X, Y ∈ P ⇒
[X, Y ] ∈ P . The same kind of idea above gives the concept of holonomy group of a
leaf of a foliation. Let us introduce it in a more formal way:
Let F be a codimension n foliation of class C r of a manifold M m . Given
Lemma 1.3.3. If F is a foliation, then its associated plane field T F is a leaf L of F we fix a point p ∈ L and, by a local trivialization of F around
involutive. p, choose a transverse section p ∈ Σp ⊂ M , diffeomorphic to a disc in Rn
Proof. Let X, Y be two vector fields tangent to T F . By using suitable and transverse to all the leaves of F . Let now γ : [0, 1] → L be any closed
local coordinates (x, y) = (x1 , ..., xm−n , y1 , ..., yn ) such that F is given in path of class C r .
these coordinates by y = (y1 , ..., yn ) = const., one can assume that X, Y Then by suitable choice of the local foliation charts, we can cover the
are of the form image γ([0, 1]) by domains of foliation charts (ϕj , Uj ), j = 0, ..., with the
following properties:
X(x, y) = (f (x, y), 0), Y (x, y) = (g(x, y), 0) (a) There is a partition 0 = t0 < t1 < · · · < tj < tj+1 < . . . < t = 1 of
    [0, 1] such that γ([tj , tj+1 ]) ⊂ Uj . In particular, we have Uj ∩ Uj+1 = ∅.
∂x g ∂y g f ∂ f ∂y f g
[X, Y ] = − x (b) For each j = 0, ..., − 1 the union Uj ∪ Uj+1 is contained in some
0 0 0 0 0 0
foliation chart domain. Moreover, the same holds for the union U ∪ U0 .
= (f · ∂x f − f · ∂x f, 0). (c) There is a neighborhood p ∈ A ⊂ Σ such that for each y ∈ A there is
a path of plaques from the plaque Py0 of (ϕ0 , U0 ) through y, to the plaque
Hence [X, Y ] ∈ T F and the proof follows. P (1)y of (U1 , ϕ1 ) that meets Py0 , then from the plaque P (1)y to the plaque
A plane field P on M of dimension k is completely integrable if there is Py2 of (U2 , ϕ2 ) that meets P (1)y , and so on. Thus we reach a plaque Py of
a foliation F of M of codimension k such that T F = P , i.e., if for each (U , ϕ ) that meets the plaque Py−1 . The intersection Py ∩ A is a single
p ∈ M we have Tp (F ) = P (p). In particular, given p ∈ M , the space P (p) point fγ (y) ∈ Σ called the γ-holonomy image of y.
is the tangent space of the leaf Lp ∈ F of F that contains p. From the This defines a map germ fγ : (Σ, p) → (Σ, p) which has the following prop-
above lemma, a completely integrable plane field is involutive. erties:
The converse of the above fact is a well-known result in theory of folia- (i) fγ has the same differentiability class as that of F .
tions: (ii) fγ does not depend on the choice of the cover {(ϕj , Uj )} of γ([0, 1])
(as a germ).
Theorem 1.3.4 (Frobenius theorem [9],[16], [26]). Involutive plane fields (iii) The map germ fγ only depends on the homotopy class [γ] ∈ π1 (L; p).
are completely integrable. (iv) The map HolL : π1 (L; p) → Diff r (Σ; p), from the fundamental group

Example 1.3.5 (integrable systems of differential forms). Let ω1 , ..., ωr be of L based at p, into the group of germs of C r diffeomorphism of Σ fixing p,
differential 1-forms of class C r on a manifold M and assume that they are and given by given [γ] → fγ , is a well-defined group homomorphism. The
linearly independent at each point p ∈ M n . We call the set S := {ω1 , ..., ωr } image is denoted by Hol(F , L, Σ; p) ⊂ Diff r (Σ; p).
a system of 1-forms on M . We may consider the corresponding distribution (v) The groups Hol(F , L, Σ; p) depend on Σ and p by natural conjugation
P (S) of (n − r)-dimensional planes defined as follows: given p ∈ M we set by diffeomorphisms.
P (S)(p) ⊂ Tp M as Thus we may speak of the holonomy group of the leaf L denoted by
Hol(F , L) or simply by Hol(L), and identify it with (a conjugacy class of)
P (S)(p) = {v ∈ Tp M, ωj (p) · v = 0, j = 1, ..., r}. a subgroup of Diff r (Rn , 0), where n is the codimension of the foliation F
of M .
The system {ω1 , ..., ωr } is called integrable if we have dωj ∧ ω1 ∧ ... ∧ ωr = 0 We close this chapter illustrating the concepts above introduced by
for all j = 1, ..., r. In particular a distribution given by a 1-form ω is means of a remarkable example.
integrable iff ω ∧ dω = 0. This occurs for instance if we have a closed
1-form ω with ω(p) = 0, ∀p ∈ M . Example 1.4.2 (foliations generated by closed 1-forms). In order to illus-
The system S is integrable if and only the distribution P (S) is involu- trate the above concepts of foliation and holonomy we consider the class of
tive. Therefore, according to Frobenius theorem above, S is integrable if codimension one foliations generated by closed differential 1-forms. Indeed,
and only if P (S) is completely integrable. we prove the following:
In the case of an integrable non-singular one-form ω we have a codi-
mension one foliation F of M which is defined by the Pffafian equation Proposition 1.4.3. A codimension one smooth foliation F tangent to a
ω = 0. closed non-singular C ∞ 1-form ω on a manifold M has trivial holonomy.
Proof. Indeed, let us fix a Riemannian metric <, > on M . Let X be the
gradient of ω, i.e., the smooth vector field on M defined by
1.4 Holonomy
ωp (vp ) = X(p), vp ,
The concept of holonomy of a foliation is motivated by the concept of return
map or Poincaré map of a periodic orbit of a vector field. for all p ∈ M and vp ∈ Tp M . Clearly X is well-defined and non-singular
since ω is non-singular. In addition X is transverse to F . Let L be a leaf
of F and γ a closed curve in L. We can assume that γ : S 1 → L is an
immersion. Set I = [−1, 1] and define the map φ : S 1 × I → S = φ(S 1 × I)
by
φ(θ, t) = Xt (γ(θ)),
where Xt (γ(θ)) stands for the integral curve of X with initial point at γ(θ).
It is clear that φ is an immersion of class C r , r ≥ 2. Then ω ∗ = φ∗ (ω) is a
well-defined 1-form on S 1 ×I. Because dω ∗ = dφ(ω ∗ ) = φ∗ (dω) = φ∗ (0) = 0
we have that ω ∗ is closed. Hence ω ∗ defines a foliation F ∗ on S 1 × I. Note
that F ∗ is conjugated to F ∩ S. It follows that the curves γ ∗ = S 1 × 0
Figure 1.1: Representation of a first return map π : (Σ, p) → (Σ, p), x → and γ have the same holonomy. Let us calculate the holonomy of γ ∗ . Fix
π(x), of a periodic orbit γ of a real vector field. (θ∗ , 0) ∈ γ ∗ and Σ∗ = {θ∗ } × I. Clearly Σ∗ is a transversal of F ∗ . Let
f ∗ : Dom(f ∗ ) ⊂ Σ∗ → Σ∗ be the holonomy of γ ∗ , p ∈ Dom(f ∗ ) and
q = f ∗ (p). Let α be an arc in Σ∗ joining p and q.
Let l be a path in a leaf of F ∗ joining p, q. Let R be the closed region Then any holomorphic (respectively, meromorphic) function f defined on
bounded by the curves γ ∗ , l and α∗ . Because the open subset U \ W admits an unique holomorphic (respectively, mero-
morphic) extension to U .
0= dω ∗ = ω∗ = ω∗ + ω∗ = 0 + ω∗
R ∂R l α α
Remark 2.1.3. The same holds for (as an immediate consequence) vector
fields and differential forms. We recall that an analytic subset of a complex
one has space is one given locally by set of common zeros of (local) holomorphic
ω ∗ = 0. functions.
α

This equality implies that α is trivial and so p = q = f ∗ (p). We conclude Remmert-Stein theorem gives conditions for the closure of an analytic
that γ ∗ has trivial holonomy. Hence γ has trivial holonomy and the proof set to be analytic.
follows. Theorem 2.1.4 (Theorem of Remmert-Stein, [30, 31]). Let M be a com-
plex manifold, W ⊂ M an irreducible analytic subset of M and V an
irreducible analytic subset of M \ W ; such that dim(V ) > dim(W ). Then
the closure V ⊂ M is an analytic subset of dimension dim(V ).
Also we need:
Theorem 2.1.5 (Theorem of Chow, [31]). A (closed) analytic subvariety
on a complex projective space is algebraic.
And the following immediate consequence:
Corollary 2.1.6. An irreducible closed analytic subset of pure codimension
one of the complex projective space CP (n) is an algebraic hypersurface.

2.2 Levi’s global extension theorem


Recall that given an open subset U ⊂ Cn , n ≥ 2 and a map f : U → R
of class C 2 we say that f is plurisubmarhomic (plush for short) or strictly
plurisubmarhomic (s-plush for short) if ∀ pi ∈ U and ∀ v ∈ Cn the restric-
tion u : z → f (p0 + zv) is subharmonic or strictly subharmonic, in the sense
that the Laplacian satisfies Δu ≥ 0 or Δu > 0 respectively.
If we denote by Hf (p0 ) the complex n × n matrix
 2
∂ f
Hf (p0 ) = (p0 )
∂zi ∂zj i,j=1,...,n

then Hf (p0 ) is hermitian Hf (p0 ) = Hf (p0 )t . It well-known that:

Lemma 2.2.1. Let f : U ⊂ Cn → R be a C 2 map in the open subset U .


Then:
1. f is plush in U ⇔ Hf∀ p∈U is definite non negative.
2. f is s-plush in U ⇔ Hf∀ p∈U is definite non positive.

Chapter 2 As a consequence:
Lemma 2.2.2. Given f ∈ C 2 (U ), U ⊂ Cn , n ≥ 2 as above, then we have:
1. f is plush in U ⇔ for each holomorphic curve γ : V ⊂ C → U the
Some results from several map f ◦ γ : V → R is subharmonic.
2. f is s-plush in U ⇔ f ◦ γ : V → R is strictly subharmonic for every
complex variables holomorphic immersed curve γ : V ⊂ C → U .
Let us see how to extend this to complex manifolds. Given f ∈ C 2 (U )
as above we define the Levi form of f as the quadratic form
n  n
∂2f
In the course of the text we shall refer to some results from the theory of Lf (p) := (p) dzi dzj , p ∈ U.
∂zi ∂zj
several complex variables. For the sake of clarity we shall now state them j=1 i=1
separately. Thus we have a quadratic form at Tp Cn defined by
∂ 2f
2.1 Some extension theorems from several com- Lf (p) · ω = (p) · wi wj = w · Hf (b) · wt , ∀ ω = (w1 , . . . , wn ) ∈ Cn .
∂zi ∂zj
plex variables Using this form we can state (cf. [30],[61]):
This section is dedicated to some useful extension theorems from several 1. f is plush in U ⇔ Lf (p) ≥ 0, ∀ p ∈ U.
complex variables. We shall start with a statement due to Riemman. We
recall that a subset X ⊂ M of a topological space is nowhere dense if the 2. f is s-plush in U ⇔ Lf (p) > 0, ∀ p ∈ U.
closure X has empty interior.
Given a holomorphic map φ : V → U , V ⊂ Cm open subset, we can
Theorem 2.1.1 (Riemann extension theorem, [29]). Let M be a connected use the Taylor expansion of order two for f in order to prove that Lf ◦φ =
complex manifold and X ⊂ M an analytic nowhere dense subset of M . φ∗ (Lf ) for the Levi-forms of f and of f ◦ φ ([30],[61]).
Then a holomorphic function f is M \ X which is bounded near X has a Given now f : M → R of class C 2 , M n a complex manifold and a chart
unique holomorphic extension to M . φ : U → Cn of M we can consider the Levi form of f at p ∈ M ⊂ U as the
quadratic form on the (complex) tangent space Tp (M ) defined by
We shall now state local and simple versions of two powerful results
from the theory of functions and analytic sets in several complex variables: Lf (p) · v := Lf ◦φ−1 (φ(p)) · (Dφ(p) · v), ∀ v ∈ Tp (M ).
Theorem 2.1.2 (Hartogs’ extension theorem, [29]). Let U ⊂ C be a n
This is well defined according to the above remarks. Finally we reach the
connected open subset and W ⊂ U be a codimension ≥ 2 analytic subset. following definition:
C2
Definition 2.2.3 ([34, 61, 69]). Given f : M −→ R, M complex manifold,
we say that f is plurisubharmonic (plush for short), respectively strictly
plurisubharmonic (s-plush for short) if Lf (p) ≥ 0, ∀ p ∈ M (respectively,
Lf (p) > 0, ∀ p ∈ M ). Given 1 ≤ k ≤ n we shall say that f is k-(strictly
plurisubharmonic) if ∀ p ∈ M, ∃ a subspace E ⊂ Tp (M ) of (complex) di-
mension k such that Lf (p)E > 0, i.e., ∀ v ∈ E \ {0} Lf (p) · v > 0.
Let M be a differentiable manifold. An exhaustion of M is a continuous
Chapter 3
function g : M → R such that:
(a) g is bounded from below g ≥ c in M
(b) g is proper: ∀ sequence {pn } ⊂ M with no accumulation point in M ,
Holomorphic foliations:
the sequence g(pn ) satisfies g(pn ) → +∞ as n → +∞.
nonsingular case
A Stein manifold is a complex manifold admitting a C ∞ s-plush ex-
haustion (Hörmander’s theorem[34]).
On the affine space Cn the function f (z) = ||z||2 is a s-plush exhaustion.
The important result below is due to Levi: 3.1 Basic concepts
Theorem 2.2.4 (Levi’s global extension theorem, [69]). Let M be a com- The purpose of this section is to introduce, in a formal way, the concept
plex manifold admitting a k-s-plush exhaustion where k ≥ 2. If K ⊂ M is of holomorphic foliation. In fact, a holomorphic foliation is, in particular,
compact subset such that M \ K is connected then any meromorphic q-form a foliation in the classical sense. Nevertheless, in this section we shall see
ω on M \ K admits an unique extension as a meromorphic q-form on M . some important examples for the development of the theory, that illustrate
the concept.
We also need the following result:
Definition 3.1.1. Let M be a complex manifold of (complex) dimension
Proposition 2.2.5. Let X ⊂ CP (n) be an algebraic subset defined by k n. A holomorphic foliation of M , of dimension k, or codimension n−k, 1 ≤
homogeneous polynomials in Cn+1 . Then the open manifold M = CP (n)\X k ≤ n−1, is a decomposition F of M in pairwise disjoint immersed complex
admits a -s-plush exhaustion where = n − k + 1. In particular if X is an submanifolds (called leaves of the foliation F ) of dimension (complex) k,
algebraic hypersurface (codimension one) the M = CP (n) \ X is a Stein and having the following properties:
manifold.
(i) ∀p ∈ M there exists a unique submanifold Lp of the decomposition that
Proof. If X is defined by the homogeneous polynomials f1 , . . . , fk in Cn+1 passes by p (called the leaf through p).
then we define f : M → R by setting in homogeneous coordinates (z1 , . . . , zn+1 ) ∈ (ii) ∀p ∈ M , there exists a holomorphic chart of M (called distinguished
Cn+1 chart of F ), (ϕ, U ), p ∈ U, ϕ : U → ϕ(U ) ⊂ Cn , such that ϕ(U ) = P × Q,
n+1
 ( |zj |2 )q where P and Q are open polydiscs em Ck and Cn−k respectively.

f (z1 , . . . , zn+1 ) := ln
j=1
(iii) If L is a leaf of F such that L∩U = ∅, then L∩U = ϕ−1 (P ×{q}),
k q∈DL,U
q
|fj j (z)|2
j=1

where if dj = deg(fj ) then q1 , . . . , qk ∈ N are such that d1 q1 = · · · = dk qk = where DL,U is a countable subset of Q.
q ∈ N. The subsets of U of the form ϕ−1 (P × {q}) are called of plaques of the
Computing the Levi form of f we can conclude by using the following: distinguished chart (ϕ, U ).
Lemma 2.2.6. ∀ x ∈ Cn the quadratic form A foliation of dimension one is also called foliation by curves. In this
case, the leaves are Riemann surfaces.

n 
Q := |dxj |2 + |xi dxj − xj dxi |2 Observe that (iii) also implies that the leaves are immersed submani-
j=1 i<j folds immersed in M . Indeed, the intersection of a leaf with a distinguished
chart is a union of plaques pairwise disjoint. Later on we shall see exam-
is definite positive. ples of foliations exhibiting leaves which are (immersed but) not embedded
submanifolds.

Remark 3.1.2. A dimension k foliation F of M , induces on M a dis-


tribution of planes of dimension k, denoted by T F , which is defined by
Tp F = Tp (Lp ), the tangent plane at p of the leaf Lp passing through p.
From (iii), this distribution is holomorphic. It defines a holomorphic vector
sub-bundle of the tangent bundle T M , which will also be denoted by T F .
The most simple example of holomorphic foliation of dimension k is the
following:
Example 3.1.3. Given the affine space Cn we may consider any decom-
position Cn = Ck × Cn−k . Such a decomposition defines a foliation F of
dimension k in Cn , whose leaves are the affine subspaces Ck ×{q}, q ∈ Cn−k .

Next we will see two ways to define foliations, equivalent to the above,
and which will be further used throughout the text.
Proposition 3.1.4. A holomorphic foliation F of dimension k on a com-
plex manifold M can also be defined in the following equivalent ways:
(I) Description by charts distinguished:
F is given by an atlas of M (also denoted F ), {(ϕα , Uα )/α ∈ A} where:
(I.1) ϕα (Uα ) = Pα × Qα , where Pα , Qα are polydiscs of dimensions k and
n − k respectively.
(I.2) If Uα ∩ Uβ = ∅ then the change of charts ϕβ ◦ ϕ−1 α is locally of the
form
ϕβ ◦ ϕ−1
α (xα , yα ) = (hαβ (xa , yα ), gαβ (yα ))
open sets, therefore it is a manifold. Finally observe that (*) shows that
In this case the plaques of F in Uα are the sets of the form ϕ−1
α (Pα × i : L → M is an injective immersion.
{q}). We leave the proof that (II) is equivalent to the definition of foliation
(II) Description by local submersions: as an exercise to reader.

F is given by an open cover M = Uα by collections {yα }α∈A and
α∈A
{gαβ }Uα ∩Uβ =∅ , that satisfy: 3.2 Examples
(II.1) ∀α ∈ A, yα : Uα → C n−k
is a submersion. Now we explore some examples of holomorphic foliations. These exam-
(II.2) If Uα ∩ Uβ = ∅ then yα = gαβ (yβ ) where gαβ : yβ (Uα ∩ Uβ ) ⊂ Ck → ples are illustrate the basic constructions and also motivate the concept of
yα (Uα ∩ Uβ ) ⊂ Ck is a local holomorphic. holomorphic foliation with singularities, to be introduced later on. Special
In this case the plaques of F in Uα are the sets of the form yα−1 (q), q ∈ attention should be paid to the dimension one and to the codimension one
Vα . cases.

Proof. First we show that (I) is equivalent to the definition of foliation. Let Example 3.2.1. Let f : M → N be a holomorphic submersion, where M
F be a dimension k foliation on M . We shall build a holomorphic atlas A of and N are holomorphic manifolds of dimensions n + k and k respectively.
M satisfying conditions (I.1) and (I.2) above. From the original definition In this case, by the local form of holomorphic submersions, the level sets
of foliation it follows that there is a holomorphic atlas A = {(ϕα , Uα ); α ∈ {F = c}, c ∈ N are holomorphic submanifolds of codimension k of M .
A} of M such that all systems of coordinates (ϕα , Uα ) of A satisfy (ii) Definition (II) of Proposition 3.1.4 then ensures that there is a foliation on
and (iii), and ϕα (Uα ) = Pα × Qα , where Pα and Qα are polydiscs of M whose leaves are the connected components level sets f . We leave the
dimension k and n − k respectively. Let us consider the change of charts proof of this fact as exercise for the reader.
ϕβ ◦ ϕ−1α : ϕα (Uα ∩ Uβ ) → ϕβ (Uα ∩ Uβ ). We can write Example 3.2.2 (Pull-back or inverse image of a foliation). Let M and N
ϕβ ◦ ϕ−1 complex manifolds, f : M → N a holomorphic map and F a foliation on N
α (xα , yα ) = (hαβ (xa , yα ), gαβ (xα , yα )) = (xβ , yβ )
of codimension k.
where (xa , yα ) ∈ Pα × Qα . We claim that gαβ does not depend on xα .
Definition 3.2.3. We say that f is transverse to F if for every point
Indeed, a point yα ∈ Qα defines the plaque ϕ−1 α (Pα × {yα }) in Uα , q ∈ N , the subspace dfq (Tq M ) and Tp F generate the tangent space Tp N ,
which is contained in a leaf L of F . On the other hand L ∩ Uβ consists of and p = f (q).
a countable union of plaques of Uβ , of the form ∪i ϕ−1
β (Pβ × {yβ }). From
i

this we obtain If this is the case then there is a foliation of M , denoted by f ∗ (F ), of


the same codimension k, whose leaves are the connected components of the
ϕβ ◦ ϕ−1
α ((Pα × {yα }) ∩ Uβ ) ⊂ ϕβ (L ∩ Uβ ) = ∪i (Pβ × {yβ }).
i inverse images by f , f −1 (L), of the leaves L of F in N . The foliation f ∗ (F)
is called pull-back or inverse image of F by F .
This last implies the following
The foliation f ∗ (F ) is obtained using (II) in Proposition 3.1.4. In-
gαβ (Pα × {yα }) ⊂ ∪i {yβi } deed, consider an open cover {Uα }α∈A of N and collections {yα }α∈A and
{gαβ }Uα ∩Uβ =∅ satisfying (II.1) and (II.2) in Proposition 3.1.4. Given α ∈ A
It is not difficult to see that the above implies that ∂gαβ /∂xα = 0. This let Vα = f −1 (Uα ) and zα = yα ◦ f : Vα → Ck . In this way we obtain an
proves the claim. open cover {Vα }α∈A of M and a collection of holomorphic maps {zα }α∈A .
Assume now that there exists a holomorphic atlas F of M satisfying Note that Vα ∩ Vβ = f −1 (Uα ∩ Uβ ), so that Vα ∩ Vβ = ∅ if, and only if,

(I.1) and (I.2). Since M is a manifold, we can assume that F is countable. Uα ∩ Uβ = ∅. Moreover, if Vα ∩ Vβ = ∅ then zα = gαβ ◦ zβ . Hence, in
Next we shall define the “leaves of F ”, taking into account (I.1)and (I.2). order to check that that f ∗ (F ) is a foliation it is enough to prove that
In M we consider an equivalence relation that identifies two points zα : Vα → Ck is a submersion for every α ∈ A. This is a consequence of the
p, q ∈ M if, and only if, there exists a finite chain of plaques (as in (I)) fact that f is transverse to F , as it can be verified from the definition.
P1 , ..., Pr of F such that Pi ∩ Pi+1 = ∅, ∀i, and p ∈ P1 , q ∈ Pr . The leaves Example 3.2.4 (foliation generated by a holomorphic vector field). Let
are the equivalence classes of M by this relation. Thus, two points p, q ∈ M M be a complex manifold of dimension n and X a holomorphic vector field
are in the same leaf if, and only if, there exists a chain of plaques as above not identically zero in M . Let S = {p ∈ M ; X(p) = 0}, the singular set of
that contains these points. Since the plaques are connected, it follows that X. Then X generates a holomorphic foliation F of dimension 1 in the open
the leaves are connected. N = M \ S. The leaves of F are the trajectories of X in N . The structure
In order to see that the leaves are immersed submanifolds in M it is of foliation arises from the Flow box theorem for holomorphic vector fields,
necessary to endow each leaf L of F with a structure of a holomorphic which can be stated as follows:
variety, in such a way that the inclusion map i : L → M is an immersion. “ For every p ∈ M such that X(p) = 0, there is a holomorphic coordi-
This structure, which is called intrinsec structure, is defined as follows: nate system (φ = (z1 , ..., zn ), U ), where P ∈ U, φ : U → φ(U ) = A × B ⊂
Fix a leaf L and consider the cover of L consisting of all the plaques C × Cn−1 and where X = ∂/∂z1 . ”
contained in L. Given a plaque Pαq = ϕ−1α (Pα × {q}) ⊂ L we define the Since the trajectories of X are the solutions of the differential equation
“coordinate system” Dz/dt = X(z) and X |U = ∂/∂z1 , we conclude that the trajectories of X
ϕqα = π1 ◦ ϕ |Pαq : Pαq → Ck in U are the form φ−1 (A × {w}) with W ∈ B. We get from this and from
Definition (I) in Proposition 3.1.4, a dimension 1 foliation, whose leaves are
where π1 : Ck × Cn−k → Ck is the first projection. Then we obtain an the trajectories of X.
“atlas”, In fact, every foliation of dimension one is locally defined by vector
fields, as we shall see in the following result, whose proof we leave to the
FL = {(ϕqα , Pαq ) ; Pαq is a plaque contained in L}.
reader as an exercise.
In order to assure that FL is a holomorphic atlas of L it is necessary Proposition 3.2.5. Let M a complex manifold of dimension n ≥ 2 and
to prove that the changes of charts are biholomorphisms between open F a dimension one foliation of M . There are collections X = {Xα }α∈A ,
subsets of Ck . This fact, left to reader as an exercise, follows from (I.2). U = {Uα }α∈A and G = {gαβ }Uα ∩Uβ =∅ such that:
We mention also without an explicit proof, that L, with the above defined (i) U is a cover of M by open sets.
structure, is a Hausdorff space. (ii) Xα is a holomorphic vector field on Uα , which is nonzero at each point.
From the definition of leaf given above, it follows that L ∩ Uα is the (iii) gαβ ∈ Ø∗ (Uα ∩ Uβ ), that is, is a holomorphic function that does not
disjoint union of plaques of Uα of the form vanish on Uα ∩ Uβ .
 (iv) In Uα ∩ Uβ = ∅ we have Xα = gαβ .Xβ .
(∗) L ∩ Uα = ϕ−1 (Pα × {q}), (v) If p ∈ Uα , then Tp F = C.Xα (p), the subspace of Tp M generated by
q∈DL,α Xα (p).

where DL,α ⊂ C n−k


. Note that each plaque of L in Uα corresponds to a Conversely, given collections X , U and G satisfying (i), (ii), (iii) and
single point in DL,α . From this, from the definition of leaf and from the (iv), there exists a foliation F that satisfies (v).
fact that F is countable, we obtain that DL,α is countable. Therefore L The following example is a complex version of results found in § 1.3 (cf.
contains only countably many plaques. Therefore L has countable basis of Example 1.3.5).
Example 3.2.6 (foliations generated by differential 1-forms). Let M a
complex manifold of dimension n and ω a holomorphic 1-form, not identi-
cally zero in M . Let S = {p ∈ M ; Ωp = 0}, the singular set of ω. In this
case, ω induces a distribution of hyperplanes Ω in the open set N = M \ S
defined by

Ωp = ker(ωp ) = {v ∈ Tp M ; ωp (v) = 0}
Chapter 4
Definition 3.2.7. We say that ω (or Ω) is integrable, if there is a holo-
morphic foliation F in N such that T F = Ω. In other words, the tangent
space at p to the leaf of F passing through P , coincides with Ωp . Holomorphic foliations
A well known fact is the following (see § 1.3 or [26]):
Lemma 3.2.8. The 1-form ω is integrable if, and only if ω ∧ dω = 0.
with singularities
The above result is known as Frobenius theorem. It is often said that
the foliation F is defined by the differential equation ω = 0 and that leaves
of F are integral submanifolds of this equation. In this chapter we introduce the concept of holomorphic foliation with
It should be noted that if η is a 1-form with η = f ω, where f is a singularities, focusing on the two cases: dimension one and codimension
holomorphic function in N that does not vanish, then the hyperplanes dis- one. Intuitively, as a first step, one may think of a holomorphic foliation
tribution induced by η coincides with Ω. In particular, η is also integrable with singularities on a complex manifold M as a a pair F := (F ∗ , S) where
and the foliations defined by η = 0 and ω = 0 coincide. S  M is an analytic subset and F ∗ is a (classical) holomorphic foliation on
Codimension one foliations are locally defined by integrable differential M ∗ := M \ S. We call S the singular set of F , and write S = sing(F ). The
1-forms, as it is stated below. leaves of F are by definition the leaves of F ∗ in M ∗ . All other concepts for
foliations (holonomy, plaques, etc) extend to this case by associating them
Proposition 3.2.9. Let M be a complex manifold of dimension n ≥ 2 and to the non-singular foliation F ∗ . Nevertheless, in the above mentioned
F a codimension one foliation of M . There exist collections W = {ωα }α∈A , cases we can say more and make a more precise definition. This is done in
U = {Uα }α∈A and G = {gαβ }Uα ∩Uβ =∅ such that: the next sections.
(i) U is a cover of M by open sets.
(ii) ωα is an integrable holomorphic 1-form in Uα , which does not vanish
at any point. 4.1 Linear vector fields on the plane
(iii) gαβ ∈ Ø∗ (Uα ∩ Uβ ).
(iv) In Uα ∩ Uβ = ∅ we have ωα = gαβ .ωβ . In this section we motivate the concept of one-dimensional holomorphic
(v) If p ∈ Uα , then Tp F = ker(ωα (p)). foliation with the study of linear plane vector fields. First of all, let us
resume a quite general remark.
Conversely, if there exist collections W, U and G satisfying (i), (ii), (iii)
Let M be a complex (always assumed to be connected) manifold of
and (iv), then there exists a foliation F that satisfies (v).
complex dimension m. Let X be a holomorphic vector field on M . We first
The proof is similar to the proof of Proposition 3.2.5 and is also left to assume that X is non-singular, i.e., X(p) = 0, ∀ p ∈ M . To X we associate
the reader as an exercise.

3.3 The Identity principle for holomorphic a holomorphic autonomous ordinary differential equation
foliations   
x = X x(t) , t ∈ C
(∗)
A well-known property of holomorphic functions and differential forms is x(0) = x0
the Identity principle (cf.[29, 30, 31]). The next proposition is the statement
of this same principle for holomorphic foliations. The solutions of (*) define the local flow of X and give a (complex) dimen-
sion one holomorphic foliation F (X) of M , obtained by gluing these local
n
Proposition 3.3.1 (identity principle for holomorphic foliations). Let M solutions
be a connected complex manifold and let F , F1 , be two holomorphic folia- The leaves of this foliation F (X) are the orbits/trajectories of X and
tions of same codimension 1 ≤ k ≤ n − 1 in M . Assume that F and F1 they are immersed Riemann surfaces on M .
coincide in some nonempty open subset U ⊂ M . Then F = F1 in M .
Example 4.1.1. We consider a linear diagonal vector field X(x, y) =
Proof. As it can be easily checked, since M n is connected, it is enough to ∂ ∂
λx + μy , μ, λ ∈ C \ {0} in the complex plane C2 . Then X is holo-
prove the case where M n is an open polydisc M = Δn in Cn . Moreover, ∂x ∂y
for the same reason, we may assume that F and F1 are given holomorphic morphic and non-singular in the punctured plane M = C2 \ {(0, 0)}. From
submersions f, f1 : Δn → Cn−k . In this case, there is some nontrivial open the above considerations, to X we can associate a dimension one holomor-
subset U ⊂ Δn where f and f1 satisfy: df ∧ df1 ≡ 0 (use the local form of phic foliation F (X) of M . The leaves
 of F (X)
 in M are parameterized by
submersions to one of these maps and then conclude it from the fact that the (global) flow of X : C t → xeλt , yeμt . The dynamical/geometrical
the other map is constant along the fibers of the first one). Thanks then behavior of these leaves very much depends on the quotient λ/μ (whether
to the classical Identity Principle for holomorphic maps, we conclude that it is real, rational positive,...).
df ∧ df1 ≡ 0 in Δn . This proves that F and F1 coincide in Δn .
Remark 4.1.2. Even though the above example seems nice, actually we
are not (usually in theory of foliations) interested in the parametrization
of the orbits; but on their dynamical/geometrical behavior on the ambient
manifold.
The point of view we want to propose is illustrated by the following
description of F (X):
• λ/μ = n/m, n/m ∈ Q+ ⇒ F (X) has a meromorphic first integral
f (x, y) = y n /xm . The leaves are contained in the curves axm + by n = 0,
a, b ∈ C and all of them accumulate only at the origin 0 ∈ C2 .
• λ/μ = −n/m, −n/m ∈ Q− ⇒ F (X) has a holomorphic first integral
f = xm y n .
The leaves are closed in C2 except for (those) the two axes; the leaves
are contained in the curves sm y n = c C.
We have a real saddle-type behavior.
• λ/μ ∈ R− \ Q: the first integral f (x, y) = y λ xmu shows that the leaves
are not closed, except for the axes. The closure of a typical leaf is a 3-
dimensional real manifold given by |f (x, y)| = |c| > 0, i.e., by |y|λ |x|−μ =
|c| (we may assume that λ, −μ ∈ R+ still with λ/μ ∈ R \ Q).
• λ/μ ∈ C \ R (hyperbolic case). In this case again the leaves are not Summarizing:
closed, except for the axes, and all leaves accumulate on both axes. In-
deed, the leaves are closed outside of the the two axes (meaning that a leaf Proposition 4.2.4. Every isolated singularity of a holomorphic foliation
accumulates only at the two axes). of dimension one is defined by a holomorphic vector field.
• λ/μ ∈ R+ : exercise (think of the radial vector field)! Thus, in a certain sense, the study of singular points of holomorphic folia-
tions of dimension one, is the study of singularities of holomorphic vector
4.2 One-dimensional foliations with isolated fields.
The next example may require some further knowledge in theory of
singularities singularities of projective varieties.

From what we have seen above, a holomorphic vector field X with singu- Example 4.2.5 (implicit complex ordinary differential equations). An al-
lar set sing(X) ⊂ M on a complex manifold M , defines a dimension one gebraic implicit ordinary differential equation in n ≥ 2 complex variables is
holomorphic foliation F (X)∗ on the open set M ∗ = M \ sing(X) (remark: given by expressions:
sing(X) ⊂ M is an analytic subset, assumed to be proper, so M \ sing(X)
is open and connected). The leaves of F (X)∗ are the orbits of X on M ∗ , (∗∗) fj (x1 , ..., xn , xj ) = 0
i.e., the non-singular orbits of X. A natural question is whether any one-
where fj (x1 , ..., xn , y) ∈ C[x1 , ..., xn , y] are polynomials j = 2, ..., n and the
dimensional holomorphic foliation is induced by a holomorphic vector field.
(x1 , ..., xn ) ∈ Cn are affine coordinates. Clearly, any polynomial vector
The answer is no, indeed there are manifolds which may be equipped with
field X on Cn defines such an equation. In general (∗∗) defines a one-
one-dimensional foliations, but that do not admit non-trivial holomorphic
dimensional singular foliation of some algebraic variety of dimension n.
vector fields. We shall see this in a while. Let us now introduce one of the
For this we begin by defining Fj (x1 , ..., xn , y2 , ..., yn ) := fj (x1 , ..., xn , yj ) ∈
main concepts in this text:
C[x1 , ..., xn , y2 , ..., yn ] polynomials in n + (n − 1) = 2n − 1 variables. Put
In what follows, as usual, M is a connected complex manifold.
also Sj := {(x, y) ∈ Cn × Cn−1 ; Fj (x, y) = o}  {(x1 , ..., xn , yj ) ∈ Cnx ×
Definition 4.2.1. A holomorphic foliation of dimension one of M with Cyj ; fj (x1 , ..., xn , yj ) = 0} × Cn−2 =: Λj × Cn−2 (y2 ,...,yˆj ,...,yn ) .
singularities is a pair F := (F ∗ , S) where S ⊂ M is a proper analytic We consider the projectivizations Sj ⊂ CP (2n − 1) and the complete
subset and F ∗ is a (classical) one-dimensional holomorphic foliation on intersection subvariety S := S2 ∩ ... ∩ Sn ⊂ CP (2n − 1). Given by the
M ∗ := M \ S. We call S the singular set of F , and write S = sing(F ). differential forms ωj := yj dx1 − dxj (j = 2, ..., n) on Cn × Cn−1 . Then
The leaves of F are by definition the leaves of F ∗ on M ∗ . {ωj = 0, j = 2, ..., n} defines an integrable system on S. We say that
the implicit differential equation (∗) is normal if S admits a normalization
Remark 4.2.2. Regarding the above concept:
(desingularization) by blow-ups σ : Ŝ → S. In particular we obtain in
(i) We may assume that S is minimal with the property “there is a non- general a singular foliation F (∗∗) of dimension one on the algebraic n-
singular foliation on M \ S that agrees with F ∗ in some open set” (cf. dimensional subvariety S ⊂ CP (2n − 1). Denote by f1 : S ∩ Cn → C1 the
Propositions 4.6.2 and 4.6.1). projection in the first coordinate f1 (x1 , ..., xn , y2 , ..., yn ) = x1 , and extend
it to a holomorphic proper mapping f1 : S → CP (1). Assume now that
(ii) Thanks to (i) above, in these notes we shall assume that F has iso- S admits a normalization σ : Ŝ → S. It is then possible to show that the
lated singularities, i.e., if dim F = 1 then sing(F ) is assumed to be a foliation F (∗∗) lifts to a foliation by curves F̂ (∗∗) on Ŝ and fˆ1 = f1 ◦ σ
discrete subset of M . ˆ over CP (1). Finally, using
defines a holomorphic proper mapping from S
The importance of the above concept also relies on the following de- fˆ α
Stein factorization theorem ([27]) we can find a splitting fˆ1 : Ŝ → B →
scription of isolated singularities.

Lemma 4.2.3. Let F be a one-dimensional holomorphic foliation with CP (1) where α : B → CP (1) is a finite ramified covering and fˆ: Ŝ → B
(isolated) singularities
 on a complex manifold M . Then there is an open is an extended holomorphic fibration over the compact Riemann surface B
cover M = Uj by connected subsets Uj ⊂ M such that: such that the following diagram therefore commutes
j∈J
σ
Ŝ −→ S
(i) sing(F ) ∩ Uj = φ or sing(F ) ∩ Uj = {pj };
fˆ ↓ ↓ f1
α
(ii) On each Uj , F is given (its plaques) by a holomorphic vector field B −→ CP (1)
Xj , with sing(Xj ) = sing(F ) ∩ Uj .
for a map fˆ1 : Ŝ → CP (1).
(iii) If Ui ∩ Uj = φ then on Ui ∩ Uj we have Xi = gij · Xj for some
non-vanishing holomorphic function gij : Ui ∩ Uk → C∗ = C \ {0}.
Conversely any such data Uj , Xj , gij defines a holomorphic, one- 4.3 Differential forms and vector fields
dimensional foliation with isolated singularities on a manifold M .
When in dimension two, the notions of codimension one and dimension
Proof. The proof is a consequence of the local case, i.e., we may assume that one, for a foliation, coincide. In this case, the vector field viewpoint though
M is a connected neighborhood of the origin 0 ∈ Cm and that sing(F ) = quite valid is not the only. Indeed the differential forms approach turns out
{0}. For sake of simplicity we assume m = 2. Given now any point to be quite useful when we look for integral of the foliation. This is what
p ∈ M \ {0} we consider the tangent space Tp (F ) := Tp (Lp ) of the leaf Lp we show through some simple examples in what follows.
of F through p. In complex dimension two, to any holomorphic vector field X(x, y) =
∂ ∂
Regarded as a line through theorigin Tp (Lp ) identifies with an element P (x, y) + Q(x, y) we can associate the corresponding dual 1-form
of the projective space C2 \ {0} C∗ = CP (1) is the Riemann sphere C = ∂x ∂y
C ∪ {∞}, so that we have defined a map f : M \ {0} → C. This map is ω(x, y) = P (x, y)dy − Q(x, y)dx.
holomorphic (as a map into C) and extends (by Hartogs’ extension theorem)
to a meromorphic map f : M → C. Indeed, the local theory of analytic The autonomous differential equation ẋ = X(x) then is equivalent to the
A(x, y) Pfaffian equation ω = 0. The solutions are the same but sometimes the
functions says that f (x, y) writes as a quotient f (x, y) = , where
B(x, y) Pfaffian viewpoint is more suitable. Let us show this in a couple of exam-
A, B are holomorphic functions on M (for M a small bidisc for instance). ples:
Now we introduce the holomorphic vector field X on M by X(x, y) =
∂ ∂   Example 4.3.1. Again we consider a linear diagonal vector field Xλ =
B(x, y) +A(x, y) · Then a local integral curve x(t), y(t) , t ∈ D ⊂ C ∂ ∂
∂x ∂y x + λy , λ ∈ C∗ , the dual 1-form is
of X satisfies ∂x ∂y
  
ẋ(t) = B x(t), y(t) ω = x dy − λy dx
 
ẏ(t) = A x(t), y(t) dy
  and Pfaffian equation is x dy − λ dx = 0. Dividing by xy we obtain −
y
y  (t) A x(t), y(t) dy A
so that  =   , that is = along the integral curves λ
dx
= 0 ⇒ d( n y − λ n x) = 0 ⇒ d n yx−λ = 0. This shows that
x (t) B x(t), y(t) dx B x
of X. This shows that such integral curves are tangent to the leaves of F f = yx−λ is a (multivalued) first integral for F .
and by same dimension we conclude that F = F (X) in M \ {(0, 0)}.
Example 4.3.2 (Poincaré-Dulac normal form). For n ∈ N \ {1}, a ∈ C∗ Example 4.4.4 (Darboux foliations). Let M be a complex manifold and
∂ ∂ let fj : M → C be holomorphic functions and λj ∈ C∗ complex numbers,
we consider the vector field X(x, y) = (nx + ay n ) +y · This is called 
r r
∂x ∂y
Poincaré-Dulac normal form as we shall see later on. It is possible to see j = 1, . . . , r. The holomorphic 1-form ω = ( fj ). λi df
fi
i
is integrable.
j=1 i=1
that the only invariant curve through the origin is the axis {y = 0}. r
1
The dual 1-form is Indeed, hω is closed and meromorphic, where h = fj . Therefore, ω is
j=1
ω = nx + ay n )dy − y dx. integrable in M \ {h = 0} and, by the Identity Principle, ω is integrable on
Dividing by y n+1 we obtain M . Thus ω defines a codimension one holomorphic foliation F (ω) in M .
Later on, we will see that we may assume that the singular set of F (ω) has
ω (nx + ay n )dy − y dx mx dy dx a dy codimension ≥ 2 (Proposition 4.6.1). We will also see that the functions
= = n+1 − n +
y n+1 y n+1 y y y fj may be assumed to be meromorphic (cf. Proposition 4.6.2). We call
 
x x 
r
λ
= −d + d( n y ) = d − n + n y .
a a F (ω) a Darboux foliation of M . The foliation F (ω) has f = fj j as a
yn y j=1
logarithmic first integral. For this reason, F (ω) is also called a logarithmic
This gives us the first integral foliation.

x n
f = exp − n + n y a = y a · e−x/y .
y 4.5 Analytic leaves
As we can easily see we may change coordinates so that a = 1 and f (x, y) =
n In general the leaves of a foliation are not embedded submanifolds, as we
y · e−x/y . Since f has a (line of) essential singularity at (y = 0) we
have already mentioned. Nevertheless when the foliation has a properly
can conclude that all leaves of, except for the one contained in (y = 0),
embedded leaf, this leaf is an analytic subset of the ambient manifold. We
accumulate at (y = 0) and are closed off (y = 0).
shall see next a criteria for a foliation, defined in a complex manifold M by
a 1-form, to have an analytic leaf. We consider the following situation:
4.4 Codimension one foliations with singu- Let F be a holomorphic foliation defined in a connected manifold M ,
by a holomorphic integrable 1-form ω. Let also f ∈ Ø(M ) be a non-
larities constant holomorphic function, that vanishes at some point of M , so that
As we have already seen (Proposition 4.2.4), any one-dimensional holomor- the analytic subset (f = 0) of M is nonempty and has codimension one. We
phic foliation F ∗ , defined in a punctured polydisc Δ\{0}, is given by a holo- shall say that the analytic set (f = 0) is invariant by F if is its connected
morphic vector field X defined on the polydisc Δ, and with sing(X) ⊂ {0}. components connected are leaves of F .
Moreover, by definition, F ∗ extends to Δ as a holomorphic foliation of di- Proposition 4.5.1. In the above situation, the analytic variety (f = 0) is
mension one, such that sing(F ) = sing(X). Similarly to this one can prove invariant by F if, and only if, there exists a holomorphic 2-form θ on M
for the case of codimension one: such that
Proposition 4.4.1. Given a polydisc Δn ⊂ Cn , centered at the origin (∗) ω ∧ df = f θ
0 ∈ Cn , denote by H ⊂ Cn any codimension two plane through the origin
and put Δ∗ = Δ \ H. Let F ∗ be a codimension-one holomorphic foliation Proof. Suppose that (f = 0) is invariant by F . In this case, since each
defined in Δ∗ . Then there exists a holomorphic 1-form ω defined in the connected component of (f = 0) is a leaf of F , these are smooth and
whole polydisc Δ with the following properties: properly embedded submanifolds of M . Thus, given a point p such that

1. ω is integrable, i.e., ω ∧ dω = 0. f (p) = 0, we can choose a trivializing chart of F , (φ = (x, y), U ), such that
p ∈ U , φ(p) = 0, x : U → Cn−1 , y : U → C and the plaques of F in U are
2. sing(ω) ⊂ H. of the form y −1 (q), q ∈ y(U ). Note that, since (f = 0) is embedded, we can
 assume that (f = 0) ∩ U = y −1 (0). We obtain then that f (x, 0) ≡ 0. From
3. The restriction ω Δ∗ defines the foliation F ∗ in the sense of Propo-
sition 3.2.9. this it follows that f (x, y) = y k .u(x, y), where k ≥ 1 and u is holomorphic
and does not vanish in U .
Proof. Indeed, by definition there is an open cover U = {Uj , j ∈ J} of Δ∗ x, On the other hand, since the plaques of F in U are of the form y =
such that on each open subset Uj the foliation is given by a holomorphic const., we can write ω |U = g.dy, where g is holomorphic and does not
integrable 1-form ωj . Moreover, on each nonempty intersection Ui ∩ Uj = ∅ vanish in U . This implies
there is a nonvanishing holomorphic function gij : Ui ∩ Uj → C \ {0}, such
that ωi = gij ωj , in this intersection. Then, the data {Uj , gij } give a Cousin df dy du du
ω∧ = g.dy ∧ (k. + ) = g.dy ∧
multiplicative problem in Δ∗ ([30],[61]). We may assume that n ≥ 3, oth- f y u u
erwise we are in dimension two, where we may use vector fields and the
dual 1-forms. Then, from Cartan’s theorem ([31, 61]) on the solution of the This proves that a 2-form θ = ω ∧ df
f is holomorphic, completing this
second Cousin problem (also called Cousin multiplicative problem), there part.
are holomorphic functions gj : Uj → C \ {0}, such that on each intersection Assume now that ω ∧ df = f θ. Let L be an irreducible component of
gij = gi /gj . Then g1i ωi = g1j ωj on each intersection Ui ∩ Uj = ∅. This (f = 0) and take p ∈ L. Computing (*) in p, we obtain,

defines a 1-form ω in Δ∗ by ω Uj = g1j ωj . This 1-form is holomorphic, inte- ωp ∧ dfp = 0 ⇒ (∗∗) dfp = λ(p).ωp , where λ(p) ∈ C
grable (each ωj is integrable) and defines F on each Uj . Thus ω defines F
on Δ∗ . By Hartogs’ extension theorem, ω admits a holomorphic extension since ωp = 0. We have two cases to consider: (a) df ≡ 0 in L, (b) df ≡ 0
to the origin. in L.
Let us consider case (a). In this case the set A = {p ∈ L; dfp = 0} is
open and dense in L (see the Identity principle in [29]). On the other hand,
The above remark also motivates the following definition: (∗∗) implies that, if p ∈ A then λ(p) = 0 and Tp L = ker(dfp ) = ker(ωp ). It
Definition 4.4.2 (codimension one holomorphic foliation with singulari- follows that A is contained in a leaf of F and therefore its L, is a leaf of F .
ties). Let M be a complex manifold. A singular holomorphicfoliation of Let us consider case (b). Here we use the fact that the set of smooth
codimension one F of M is given by an open cover M = j∈J Uj and points in L is open and dense in L (see [30, 31]). Given a smooth point
1 p of L, there exists a coordinate system (φ = (x, y), U ) such that p ∈ U ,
holomorphic integrable 1-forms ωj ∈ (Uj ) such that if Uj ∩ Uj = ∅, then
ωi = gij ωj in Ui ∩ Uj , for some gij ∈ Ø∗ (Ui ∩ Uj ). We put sing(F ) ∩ Uj = φ(p) = 0, x : U → Cn−1 , y : U → C and U ∩ L = (y = 0). Since f |L ≡ 0, we
{p ∈ Uj ; ωj (p) = 0} to obtain sing(F ) ⊂ M , a well-defined analytic subset obtain that f (x, y) = y k .u(x, y), where k ≥ 2 and u is holomorphic and does
n−1
of M , called singular set of F . M \ sing(F ) is foliated by a holomorphic not vanish in U . On the other hand, we can write ω |U = b dy + i=1 ai dxi ,
codimension one (regular) foliation F0 . By definition the leaves of F are and therefore of (*), we obtain that the 2-form below is holomorphic
the leaves of F0 .
df du 
n−1
dy
ω∧ =ω∧ +k ai dxi ∧ .
Remark 4.4.3. We may always assume that sing(F) ⊂ M has codimen- f u y
i=1
sion ≥ 2. If (fj = 0) is a local equation of a codimension one component
of sing(F ) ∩ Uj , then we get ωj = fjn ω̄j where ω̄j is a holomorphic 1-form n−1
i=1 ai dxi ∧ y is holomorphic.
dy
Since du
u is holomorphic, we obtain that
and sing(ω̄j ) does not contain (fj = 0). We conclude that y divides ai , for every i = 1, ..., n − 1, i.e., that we can
write ω |U = y.η + b.dy where b and η are holomorphic. This implies that
(y = 0) = L ∩ U is invariant by F . Since the set of smooth points of L is
open and dense in L, we can conclude that L is invariant by F , i.e., it is a
leaf of F .

4.6 Two extension lemmas for holomorphic Chapter 5


foliations
This section is dedicated to the extension of codimension one holomorphic
foliations with singularities, through codimension ≥ 2 analytic subsets. Holomorphic foliations
Moreover, we conclude that the singular set of such foliation, may always
be assumed to have codimension ≥ 2. given by closed 1-forms
Proposition 4.6.1. Let F be a codimension one holomorphic foliation
with singularities on a connected complex manifold M . Then there exists a
foliation F1 of M with the following properties:
This chapter is dedicated to the study of an important class of codimension
(a) The irreducible components of sing(F1 ) have codimension ≥ 2, and one foliations with singularities. The class of foliations given by closed 1-
sing(F1 ) ⊂ sing(F ). forms. We study their holonomy and some extension property. Starting
with the holomorphic case we are to consider the meromorphic case, making
(b) F1 coincides with F on M \ sing(F )
use of the extension results Propositions 4.6.1 and 4.6.2. This study also
(c) F1 is maximal in the following sense: if F2 is another foliation of M is related to the classification and normal forms of the isolated singularities
satisfying (a) and (b), then F2 = F1 . of foliations, as suggested for instance by § 6.2 and Example 4.3.2.

Proof. The proof is basically a consequence of Hartogs’ extension theorem


already mentioned in this text. Let {ωα }α∈A , {Uα }α∈A and {gαβ }Uα ∩Uβ =φ , 5.1 Foliations given by closed holomorphic 1-
be collections defining F . Assume that sing(F ) has some codimension one
irreducible components. Denote by W the union of these components. We
forms
develop an elimination procedure for these components. Given a point Let M be a complex manifold of dimension ≥ 2 and ω be a closed holo-
p ∈ W we choose local coordinates (x = (x1 , ..., xn ), Up ) such that p ∈ Up , morphic 1-form on M (that is dω = 0) assumed to be not identically zero.
x : Up → Cn , and x(Up ) is an open polydisc in Cn , and also W ∩Up has only Then, ω is integrable ( ω ∧ dω = 0) and therefore it defines a foliation
finitely many irreducible components, say W1p , ..., Wrp , with corresponding F of M . Classical integration lemma of Poincaré assures that, given any
irreducible equations f1 , ..., fr respectively (see [30]). We can assume that simply-connected open subset U ⊂ M , there exists a holomorphic function
Up ⊂ Uα , for some α = α(p) ∈ A. f : U → C, such that ω |U = df . Observe that if g : V → C is a func-
If g is a holomorphic function in Up vanishing in W ∩ Up , then g = tion such that dg = ω, where U ∩ V is connected and non-empty, then
f1n1 ...frnr .h, where n1 , ..., nr ∈ N and h ∈ Ø(Up ) (cf. [30, 31]). We can g − f is constant in U ∩ V . Therefore, as we have already observed before,
also write ωα |Up = nj=1 aj dxj . Since ωα |W ∩Up ≡ 0, we conclude that the the corresponding foliation F can be locally defined by holomorphic func-
coefficients aj of ωα |Up vanish in W ∩ Up and therefore ωα = f1n1 ...frnr .ωp ,

where n1 , ..., nr ∈ N and ωp is an integrable 1-form on Up , having singular tions as follows: there exist collections U = {Uα }α∈A , F = {fα }α∈A and
set of codimension ≥ 2. C = {cαβ }Uα ∩Uβ =∅ , such that: (i) U is a cover of M by simply-connected
On the other hand, if p ∈ / W then we take Up ⊂ Uα , ωp = ωα |Up , for open sets.
some α = α(p) ∈ A, in such a way that Up ∩ W = φ and that Up is the (ii) If α ∈ A, then fα is a holomorphic function in Uα such that dfα = ω |Uα .
domain of a local chart x = (x1 , ..., xn ). (iii) If Uα ∩ Uβ = ∅, then Uα ∩ Uβ is connected, cαβ ∈ C and fα = fβ + cαβ
In this way, we can define an open cover {Up }p∈M and a collection in Uα ∩ Uβ .
{ωp }p∈M , where Up ⊂ Uα(p) , ωp is a holomorphic integrable 1-form on Up Observe that if ω is free of singularities, then the functions fα are sub-
such that mersions and F is a non-singular foliation. In this case, if we denote by gαβ
codim(sing Xωp )) ≥ 2 and ωp generates F in Up \ sing(ωα(p) ) (i.e., if the translation gαβ (z) = z + cαβ , then fα = gαβ ◦ fβ , and then F can be
q ∈ Up \ sing(ωα(p) ), then Tq F = ker(ωα (q)). We shall see next that defined by local submersions as in (II) of Proposition 3.1.4, in which case
there exists a collection {gp,q }Up ∩Uq =φ , where gp,q ∈ Ø∗ (Up ∩ Uq ), such the gαβ are translations. We say then that F has an additive transverse
that ωp = gp,q .ωq in Up ∩ Uq = φ. structure. In case sing(ω) = ∅, then F has an additive transverse structure
Let p, q ∈ M such that Up ∩ Uq = φ e α = α(p) and β = α(q). Let also in M \ sing(F ).
x = (x1 , ..., xn ) : Up → Cn be a local system of coordinates. We can write Conversely, if F is a foliation with an additive transverse structure in
ωp = j=1 aj dxj and ωq |Up ∩Uq = j=1 bj dxj . Observe that ωα = gαβ .ωβ
n n
M \ sing(F ) and such that codim(sing(F )) ≥ 2, then F can be defined by
implies that a closed holomorphic 1-form ω on M . Indeed, let be given U = {Uα }α∈A ,
a1 an F = {fα }α∈A and also C = {cαβ }Uα ∩Uβ =φ collections satisfying (ii) and
= ... = = gp,q em Up ∩ Uq .
b1 bn (iii), where U is an open cover of M \ sing(F ). From (iii) we obtain that,
This means that ωp = gp,q .ωq , where gp,q is at first sight meromorphic. It if Uα ∩ Uβ = ∅, then dfα = dfβ in Uα ∩ Uβ . This implies that there exists
is then enough to prove that gp,q extends to a function in Ø∗ (Up ∩ Uq ). the 1-form holomorphic ω on M \ sing(F ) such that ω |Uα = dfα . It is not
Indeed, first we observe that the singular sets Sp and Sq , of ωp and of difficult to see that the form ω is closed and defines (induces) the foliation F
ωq , are of codimension ≥ 2. Let us put Z = (Sp ∪ Sq ) ∩ (Up ∩ Uq ). in M \ sing(F ). On the other hand, since codim(sing(F )) ≥ 2, the classical
extension theorem of Hartogs then implies that ω extends to a holomorphic
Given zo ∈ (Up ∩ Uq ) \ Z, there exists an index j ∈ {1, ..., n} such form on M , which is also closed and defines a foliation F .
that bj (z) = 0, for every z in a small neighborhood of zo . Hence gp,q = We can then state the following result:
aj (
bj ∈ Ø Up ∩ Uq \ Z). Since Z is of codimension ≥ 2, it follows from Har-
togs’ extension theorem (Theorem 2.1.2) that gp,q extends to a holomorphic Proposition 5.1.1. Let M be a holomorphic manifold and F a foliation
1
function in Up ∩ Uq (see [30, 31]). For the same reason, gp,q also extends. of M , with singular set of codimension ≥ 2. Then F can be defined by
Therefore the extension we just obtained does not vanish. The proof of (c) a closed 1-form if, and only if, F has an additive transverse structure in
is a consequence of the following proposition. M \ sing(F ).

Similarly to above we obtain: 5.1.1 Holonomy of foliations defined by closed holo-


Proposition 4.6.2. Let M be a connected complex manifold of dimension morphic 1-forms
≥ 2, and let V be an analytic subset of M of codimension ≥ 2. Given F a Let M be a complex manifold of dimension n ≥ 2 and ω a closed holomor-
codimension one holomorphic foliation of U = M \ V , there exists a unique phic 1-form not identically zero in M . Let F be the singular foliation of
foliation F  of M , whose restriction to U coincides with F . codimension one defined by ω in M . Our purpose is to prove:
Proposition 5.1.2. The foliation F is without holonomy. More precisely,
if L ⊂ M \ sing(ω) is a leaf of F , then L has trivial holonomy. Remark 5.2.3. Observe that the connected components of L are leaves
Proof. We make use of the existence of an additive transverse structure for of F  . Moreover, , given a leaf L0 ⊂ L and a coordinate system w =
the foliation. Fix a closed regular curve γ : I → L with γ(0) = γ(1) = po . (x, y) : U → Cn−1 × C such that U ∩ L = U ∩ L0 = (y = 0), we can consider
Given q ∈ γ(I) there exists a local chart (x, y) : U → Cn−1 × C such that the number
U ∩ L = (y = 0) and ω |U = dy (Poincaré integration lemma). We can then
k = min{j > 0; y j ω extends to a holomorphic form on U }
obtain a collection C = {((xj , yj ), Uj )}kj=1 of such charts and a partition
{0 = t0 < t1 < ... < tk = 1} of I, such that:
It is possible to prove that k only depends on L0 , i.e., does not depend
(i) ∪kj=1 Uj = γ(I). on the coordinate system that we consider.
(ii) γ([tj−1 , tj ]) ⊂ Uj , ∀j = 1, ..., k. We shall say then that ω has a pole of order k in L0 .
Next we shall see a particular example of the situation above.
(iii) ω |Uj = dyj , ∀j = 1, ..., k.

Since γ(0) = γ(1), we can assume that: 5.2.1 Holonomy: meromorphic case
(iv) ((x1 , y1 ), U1 ) = ((xk , yk ), Uk ) = ((x, y), U ), where x(po ) = 0 ∈ C n−1
. Let M a complex manifold of dimension ≥ 2 and ω a closed meromorphic 1-
form on M . According to Proposition 5.2.1, the foliation with singularities
Let us consider the transverse sections Σj = {(xj , yj ) ∈ Uj ; xj = = xoj defined by ω on M \ (ω)∞ ) can be extended to a foliation of M , which we
xj (γ(tj ))} ⊂ Uj , j = 1, ..., k, where we shall compute the holonomy in the shall denote by F , with the property that (ω)∞ is invariant by F . In other
transverse section Σ = Σk . For sake of simplicity, we shall denote the point words, the smooth part of of the polar set of ω is a reunion of leaves of
(xoj , yj ) ∈ Σj by yj . Moreover, for the uniformization of the notation we F . Let us now see how to compute the holonomy of a leaf of this foliation.
shall put Σ0 = Σ and y0 = y = yk . The first remark is that a leaf which is not contained in the polar set, has
Let us compute the holonomy fj : Σj−1 → Σj , j = 1, ..., k. This holon- trivial holonomy by Proposition 5.1.2. Let us then compute the holonomy
omy is of the form yj = fj (yj−1 ). It is enough to prove that fj (yj−1 ) = of the leaves contained in the polar set (ω)∞ .
yj−1 , j = 1, ..., k. This implies that the holonomy of γ, which is the com- For this sake we need some notation.
position fk ◦ ... ◦ f1 , is equal to the identity of Σ. For N k ≥ 2 and a ∈ C, we consider the following vector field:
Indeed, since ω |Uj−1 ∩Uj = dyj−1 = dyj , we obtain that d(yj − yj−1 ) = 0 yk ∂
in Uj−1 ∩ Uj . This implies that the difference yj − yj−1 is constant in the Y k,a =
1 + a.y k−1 ∂y
connected component of Uj−1 ∩ Uj that contains γ(tj−1 ), say yj = yj−1 + c.
On the other hand, because Uj ∩ L = (yj = 0) and Uj−1 ∩ L = (yj−1 = 0), defined in the open set {y ∈ C; 1+a.y k−1 = 0}. Note that Y k,a generates a
we obtain that c = 0 completing the proof. local flow in a neighborhood of 0 ∈ C, which will be denote by Yzk,a . Thus,
for a fixed z ∈ C, Yzk,a is a biholomorphic map between neighborhoods of
0 ∈ C, since Yzk,a (0) = 0. Let us denote the germ of Yzk,a at the origin by
5.2 Foliations given by closed meromorphic [Yzk,a ].
1-forms Note that, if k ≥ 3, then, [Yzk,a ] commutes with the rotation Rλ (y) =
λ.y, where λk−1 = 1. Then, for every k ≥ 2 and every a ∈ C, the set
Let M be a holomorphic manifold of dimension n ≥ 2 and ω a closed
meromorphic 1-form (not holomorphic) on M . We shall denote the polar Gk,a = {[Rλ ◦ Yzk,a ] ; z ∈ C λk−1 = 1}

divisor of ω by (ω)∞ ⊂ M (see the definition in [34]). In the current case we is an abelian group.
have (ω)∞ = ∅, since ω is assumed to be not holomorphic. Since ω is closed An important particular case is when k = 2 and a = 0. In this case
and holomorphic in the open set N = M \ (ω)∞ , it defines a codimension G2,0 is the group of homographies of the form
one foliation of N which we shall denote F . Let us see that, indeed, F
y
extends to a foliation of M . {y → ; a ∈ C}
1 + ay
Proposition 5.2.1. The foliation F extends to M as a foliation F  such
that (ω)∞ is invariant by F  .
dy
, obtained by straightforward integration of the differential equation dz =
y2.
Proof. In the proof of the extension of F we shall use the following fact Our main result in this section is the following computation of the
(see [30], [31]): holonomy:
Fact 5.2.2. The set L of smooth points of (ω)∞ is open and dense in (ω)∞ . Proposition 5.2.4. Let M a complex manifold of dimension ≥ 2 and ω a
Moreover, the set S = (ω)∞ \ L is an analytic subset of M de codimension closed meromorphic 1-form on M with polar set (ω)∞ ⊂ M . Denote by F
≥ 2. the codimension one foliation of M corresponding to ω. Let L be a leaf of
The idea is to prove first that F extends to the set M \ S and then make F . Then:
use of Proposition 4.6.2. (a) If L ⊂ M \ (ω)∞ , then Hol(L) is trivial.
(b) If L ⊂ (ω)∞ and ω has poles of order 1 in L, then Hol(L) is abelian
With this aim, let us fix a point p ∈ L. Since p is a smooth point of and analytically linearizable, that is, Hol(L) is analytically conjugate to a
(ω)∞ and this set has codimension one, there exists a system of holomorphic subgroup of linear maps (z → λz) of C, in some neighborhood of the origin.
coordinates in a neighborhood U of p, w = (x, y) : U → Cn , where w(U ) is a (c) If L ⊂ (ω)∞ and ω has poles of order k ≥ 2 in L, then Hol(L) is
polydisc, x = (x1 , ..., xn−1 ) : U → Cn−1 and such that U ∩ L = U ∩ (ω)∞ = analytically conjugate to a subgroup of Gk,a , for some a ∈ C.
{(x, y); y = 0}.
Proof. Case (a) has already been addressed. Suppose now that L ⊂ (ω)∞ .
Observe now that by definition of the polar set, there exists j > 0 such
Fix p ∈ L and a local foliation chart of F , (x, y) : U → Dn−1 ×D ⊂ Cn−1 ×C
that y j ω extends to a holomorphic form on U . Let
such that (ω)∞ ∩ U = L ∩ U = (y = 0) and the plaques of F in U are of
k = min{j > 0; y j ω extends to a holomorphic form em U } the form y = cte . We claim that:
n−1 Claim 5.2.5. We have
and let us define η=y k .ω. We can write η=an .dy + aj dxj , where some
j=1 g(y)
of the functions a1 , ..., an does not vanish identically in U ∩ L. Note that (∗) ω |U = .dy,
yk
η is integrable in U \ L and defines the same foliation that ω in this set.
Henceforth, η is integrable in U , and there it defines a foliation of U that where g is holomorphic in D, g(0) = 0 and k is the order of the polar set
extends F |U . of ω in L.
On the other hand, from ω = y −k .η we obtain Proof of Claim 5.2.5. Indeed, since ω defines F on M \ (ω)∞ and the
d(y −k ) ∧ η + y −k .dη = dω = 0 ⇒ (∗) dy ∧ η = k −1 .y.dη plaques of F em U are of the form y = cte , we have ω |U = h(x, y)dy,
where h is meromorphic in U with poles in (y = 0). On the other hand,
Then, from (*) and from Proposition 4.5.1 we get that (y = 0) = L ∩ U because ω is closed we have ∂h/∂x ≡ 0, i.e., h = f (y), only depends on y.
is invariant by the foliation defined by η. This implies then that F extends Let k be the order of the polar set of ω in L. Then f writes as in (*), as it
to M \ S in such a way that L is invariant by the extension, as desired. can be easily verified.
Now we need: the change of coordinates.
Lemma 5.2.6. Let α be a meromorphic 1-form on a neighborhood of 0 ∈ C. Case a = 0. In this case, in Uj−1 ∩ Uj we have
Suppose that 0 is a pole of order k ≥ 1 for α. Then there exists a system
dz dw f  (z) z
of coordinates y : V → C with 0 ∈ V , y(0) = 0 and such that α writes in = 2 = .dz ⇒ z 2 .f  = f 2 ⇒ f (z) =
z 2 w (f (z))2 1 + c.z
this system of coordinates as:
(i) α = a. dy
y a = 0, if k = 1. from where we conclude that f is in the group G2,0 . Similarly to above,
1+a.y k−1
(ii) α = yk
.dy, with a ∈ C, if k > 1. since the composition of elements in G2,0 is also in G2,0 , we obtain that
Hol(L) is analytically conjugate to a subgroup of G2,0 .
Proof. Let us prove the case of simple pole, k = 1. The case k > 1 is Case a = 0. The argumentation in the case k = 2 and a = 0 is quite
similar and is a simple exercise. In the case of a simple pole we can write similar to the one above: it is enough to prove that yj and yj−1 are related
α = g(z)
z dz, where g is holomorphic in neighborhood W of 0 and g(0) = by an element of G2,a . As it is easily checked, if yj = f (yj−1 ), then f
a = 0. Note that a = Res(α, 0), that is invariant by change of coordinates satisfies the differential equation
(see [2]). We have then g(z) = a + z.u(z), where u is holomorphic W , so
that (∗) z 2 .(1 + a.f (z)).f  (z) = (f (z))2 .(1 + a.z).
dz
α = a + u(z)dz.
z Thus it remains to show that in this case f is in G2,a . Let us show how
this is done.
Let ϕ(z) be a primitive (integral) of the holomorphic 1-form u(z)
a dz in a Step 1. Given b ∈ C, there exists a unique solution f of (*), defined in a
neighborhood of 0. Let us consider the function y(z) = z. exp(ϕ(z)). Since
neighborhood of 0, such that f  (0) = 1 and f  (0) = b.
y(0) = 0 and y  (0) = 0, we conclude that y is a biholomorphic map between
f (z)−z
two neighborhoods of 0. On the other hand, Indeed, if f is a solution of (*) and g(z) = z2 , then g satisfies the
following differential equation:
dy dz dz
a = a + adϕ = a + u(z)dz = α, a + azg − g
y z z (∗∗) g  = g. = F (z, g).
1 + az + azg
completing the proof.
Since F is holomorphic in a neighborhood of (0, b/2), (**) has a unique
Let us resume the proof of the proposition. solution g, defined in a neighborhood of 0 ∈ C, with g(0) = b/2. Setting
Assume first that k = 1. Fix a closed curve γ : I → L with γ(0) = f (z) = z + z 2 .g(z), we obtain the required solution.
γ(1) = po . Step 2. For each c ∈ C the function fc = Yck,a is a solution of (*) with
From Lemma 5.2.6 and with similar arguments to the holomorphic case initial condition fc (0) = 1 and fc (0) = 2c.
we have:
The proof is a straightforward computation.
Claim 5.2.7. There is a collection C = {((xj , yj ), Uj )}kj=1 of foliation Proposition 5.2.4 then follows from Steps 1 and 2 above.
charts of F , and a partition {0 = t0 < t1 < ... < tk = 1} of I, such that:
(i) ∪kj=1 Uj = γ(I).
(ii) γ([tj−1 , tj ]) ⊂ Uj , ∀j = 1, ..., k.
dy
(iii) ω |Uj = aj yjj , ∀j = 1, ..., k.

Since γ(0) = γ(1) we can assume that:


(iv) ((x1 , y1 ), U1 ) = ((xk , yk ), Uk ) = ((x, y), U ), where x(po ) = 0 ∈ Cn−1 .
Let us consider also sections Σj , j = 0, ..., k as in the holomorphic case,
with Σ0 = Σk = Σ.
Now we remark that if A is a connected component of Uj−1 ∩ Uj that
dyj−1 dy
contains γ(tj−1 ), then aj−1 . yj−1 = aj yjj in A. Comparing the residues
of these two forms in 0, we conclude that aj−1 = aj . We can then conclude Chapter 6
dyj−1 dy
that yj−1 = yjj in Uj−1 ∩ Uj , for every j = 1, ..., k. This allows us to
relate the coordinates yj and yj−1 in A. Indeed, if yj = f (yj−1 ) in A, then
we must have
dyj
=

f (yj−1 )
dyj−1 =
dyj−1

Reduction of singularities
yj f (yj−1 yj−1

z.f (z) = f (z) ⇒ f (z) = cj z
for some constant cj , as it follows from a straightforward integration of 6.1 Introduction
the differential equation z.f  = f . This last implies that the intermediate
holonomy maps fj : Σj−1 → Σj are linear. Since the composition of linear The subject of reduction of singularities is one of the most developed and
maps is also linear, we obtain that the holonomy of γ is linear in the interesting in the theory of holomorphic foliations. In the last decades var-
coordinate system that we consider. Since this system only depends on ω ious authors have given important contributions to the subject. Although
(not on the curve γ), we conclude that the holonomy de L is analytically the general problem of the reduction of singularities for a germ of a holo-
linearizable. morphic foliation singularity remains open, several are the cases where it is
Now we consider the case where k ≥ 2. Again we fix a closed curve already finished. This knowledge has been of great importance in the devel-
γ : I → L with γ(0) = γ(1) = po . Similarly to the case k = 1, from opment of the theory of holomorphic foliations with singularities as we shall
Lemma 5.2.6 and similar arguments to the ones in the holomorphic case, see in some of the results to be presented in this text (see Chapters 7 and 10
we have: for instance). We would like to mention the recent work of Camacho-Lins
Neto-Sad ([10]), Camacho-Sad ([13]), Cano([14]) and Cano-Cerveau ([15])
Claim 5.2.8. There is a collection C = {((xj , yj ), Uj )}m
j=1 of foliation as a partial list.
charts of F , and a partition {0 = t0 < t1 < ... < tm = 1} of I, satisfying Before going into the reduction of singularities, we recall some very basic
1+aj yjk−1
(i),(ii),(iv) and (iii) ω |Uj = yjk
.dyj , ∀j = 1, ..., m. facts from the theory of singularities of holomorphic vector fields. In short,
in the next section we study the analytic forms of non-degenerate isolated
Observe that aj = Res(ω, yj = 0). Thus, similarly to the above argu- singularities of holomorphic vector fields, according to Poincaré and Dulac.
mentation, we conclude that a1 = ... = am = a.
As we have seen above, it is enough to relate yj and yj−1 . Let us do 6.2 Poincaré and Poincaré-Dulac normal forms
it in the case k = 2. The case k > 2 is left to the reader. For the sake of
simplicity of notation let us put yj = w and yj−1 = z, so that w = f (z) is In this section we shall consider a holomorphic vector field X defined in a
neighborhood U of the origin 0 ∈ Cn , n ≥ 2, with a singularity (isolated) at
the origin. Since we are interested in the local analytical description of X
or F (X) in a neighborhood of 0, we shall consider U as small as necessary By using local coordinates we can introduce the notion of blow-up of
without further mention to this. complex surface M 2 at a point p ∈ M is a natural way and successively by
The eigenvalues of X at 0 are those of the linear part DX(0) (here repeating this process as many times as desired.
regarded as a linear map Cn → Cn ). Assume that DX(0) is non-singular.
Definition 6.2.1. We shall say that the singularity 0 of X is in the F
Poincaré domain if the convex hull of its eigenvalues in R2  C does not
P 2
contain the origin. Otherwise we say that the singularity 0 of X is in the 1
Siegel domain.
Remark 6.2.2. If m = 2 then the singularity is in the Siegel domain if,
and only if, the quotient of its eigenvalues is real negative.
A geometrical characterization of singularities in the Poincaré domain,
as well as a rich dynamical description of them, is found in the series of r
works of the senior Japanese mathematician, Prof. Toshikazu Ito. An
excerpt from those is the following ([37]):
Theorem 6.2.3 (T. Ito, 1992). A singularity of Cm of a holomorphic
vector field X is in the Poincaré domain if, and only if, the foliation F (X)
Figure 6.1:
is transverse to every small sphere S 2n−1 (r; 0) of center 0 ∈ Cn and radius
r > 0. This is the case if F (X) is transverse to some sphere contained in
its definition domain. The final configuration of the exceptional divisor, depends on the position
Let now m = 2. We shall say that the eigenvalues λ1 , λ2 of X are in of the blow-up centers.
resonance if λ1 = k λ2 or λ2 = k λ1 for some k ∈ N, k ≥ 2.
For dimension two the complete description of the analytical normal
forms of singularities in the Poincaré-domain is given below:
6.4 Blow-up on surfaces
Theorem 6.2.4 (Poincaré linearization theorem, [4, 6, 22]). Let X be Let us recall the blow-up of C2 at 0. Let us consider two copies of C2 ,
a holomorphic vector field with a singularity in the Poincaré domain at say U and V , with coordinates (t, x) and (s, y) respectively. We define a
the origin 0 ∈ C2 . Suppose that the eigenvalues λ1 , λ2 of X are not in complex manifold C˜2 , identifying the point (t, x) ∈ U \ (t = 0) with the
resonance. Then there is a unique holomorphic diffeomorphism ξ between point (s, y) = α(t, x) = (1/t, tx) ∈ V \ (s = 0).
neighborhoods of 0 ∈ C2 , ξ(0) = 0, such that: The divisor of C̃2 is defined as the submanifold D of C̃2 such that
U ∩ D = (x = 0) and V ∩ D = (y = 0). Note that, since y = tx, D
(i) ξ  (0) = Id
is well defined and is biholomorphic to the Riemman sphere C = CP (1).
(ii) ξ∗ X = DX(0), i.e., ξ takes X into its linear part. Moreover, we can define a submersion P : C˜2 → D by P |U (t, x) = t and
P |V (s, y) = s. Then (C˜2 , P, D) is a vector fiber space with basis D,
Remark 6.2.5. (i) The above theorem is similar to a result, by Poincaré, projection P and fiber C, having D as zero section.
for diffeomorphisms f : (Cm , 0) → (Cn , 0). The proof is based on the con-
vergence of the formal (power series) solution to the linearization problem. Let us now consider the holomorphic map π : C̃2 → C2 defined by

For the case of resonances we have: π |U (t, x) = (x, tx) and π |V (s, y) = (sy, y). Note that π is well-defined,
since in U ∩ V we have y = tx and x = sy. Moreover, , π has the fol-
Theorem 6.2.6 (Poincaré-Dulac theorem,[22]). Let X be a holomorphic lowing properties: (a) π −1 (0) = D. (b) π |C˜2 \D : C˜2 \ D → C2 \ {0} is
vector field with a singularity in the Poincaré domain at 0 ∈ C2 . Sup-
biholomorphic. (c) π is a proper map.
pose that the eigenvalues are in resonance , λ1 = k λ2 for some k ∈ N,
k ≥ 2. Then these is an unique holomorphic diffeomorphism ξ between Then, with this structure, C˜2 is the blow-up of C2 at 0, with projection
neighborhoods of 0 ∈ C2 , ξ(0) = 0, such that: (blow-down) map π.
 Let us introduce this same concept in a complex surface. We consider a
(i) ξ (0) = Id
complex manifold of dimension two M and a point q ∈ M . The blow-up of

∂ ∂ M at q is defined as follows: take a holomorphic local chart ϕ : A → B ⊂ C2
(ii) ξ∗ X = (λ1 x + ay n ) + λ2 y , with q ∈ A and ϕ(q) = 0. Let π : C˜2 → C2 be the blow-up map at 0, with
∂x ∂y
divisor D and B̃ = π −1 (B). In the disjoint union M  = (M \ {q})  B̃ we
for some a ∈ C. define an equivalence relation ∼ by setting po ∼ p1 if, and only if, po = p1
or, otherwise, po ∈ A \ {q}, p1 ∈ B̃ \ D and p1 = π −1 (ϕ(po )). The blow-up
If a = 0 then we are in the analytically linearizable case. The proof is in
of M at q is the quotient M̃ = M  / ∼.
the same spirit of the one for the non-resonant case. We refer to the book
of Ilyashenko and Yakovenko ([36]) for the proofs and a detailed study in Since B̃ is a manifold and π −1 ◦ ϕ : A \ {q} → B̃ \ D is a a biholomorphic
these normal forms. map, it follows that M̃ is a complex manifold. Roughly, M̃ is obtained from
M by replacing the point q by a projective line D  C. Indeed, the divisor
D, thanks to the above procedure, there is a natural embedding of the
6.3 Blow-up at the origin (quadratic blow- divisor D into M̃ .
up) Given a point p ∈ M̃ , we have three possibilities: (1) The equivalence
class of p is in D. (2) The equivalence class of p is in M \ A. (3) The
The (quadratic) blow-up at the origin is defined as follows: We consider two equivalence class of p contains two points po ∈ A \ {q} and p1 ∈ B̃ \ D.
copies of C2 with coordinates (a, t) and (u, y) and change of coordinates Thus, the points of M̃ are divided into two classes: those points as in
y = tx (1), that will be called points of the divisor, and the points of M̃ \ D, that
given by
x = uy will be regarded as points of M (as in (2) or (3)).
In particular ut = 1 so that we obtain a complex surface C20 that contains A map of blow-down Π : M̃ → M is defined by Π(p) = q in case (1),
a projective line P → C  2 (given by (x = 0) and by (y = 0)). Define the Π(p) = p in case (2) and Π(p) = po in case (3). It is not difficult to
0
map π : C  2 → C2 by π(x, t) = (x, tx) and π(u, y) = (uy, y); then π defines see that Π has properties analogous to those of π, i.e.,, (a’)Π−1 (q) = D.
0
a proper holomorphic projection which is a diffeomorphism between C 2 \ P (b’)Π |M̃\D : M̃ \ D → M \ {q} is a biholomorphic map. (c’)Π is a proper
0
and C2 \ {0}. map.
The manifold C  2 is the blow-up of C2 at the origin and it is a complex The above process can be iterated: start with the manifold M and a
0
(rank 1) vector bundle with basis CP (1) ≈ P, and fiber C. Roughly speak- point qo ∈ M . Blowing-up M at qo , we obtain a manifold M1 and a blow-
ing, the “explosion” of C2 at the origin “separates” the lines though 0. It up map Π1 : M1 → M with divisor D1 = Π−1 1 (po ). Next we consider any
is a remarkable fact that the self-intersection P · P of the projective line point q1 ∈ M1 , and perform the blow-up of M1 at q1 obtaining in this
P → C 2 (called exceptional divisor) is negative equal to −1. way a manifold M2 and a map blow-up map Π2 : M2 → M1 with divisor
0
D2 . Proceeding in this way, after k blow-ups, we obtain a manifold Mn
and a blow-up map Πn : Mn → Mn−1 with divisor Dn . The composition (0, 0) = D1 ∩ D2 , therefore this curve is still not resolved. With a final
Πn = Πn ◦ ... ◦ Π1 : Mn → M is a proper holomorphic map that we will call blow-up π3 at the point (u, t) = (0, 0), of the form t = vu (in one of the
the blow-up map. charts), we obtain a new divisor D3 , represented by (u = 0). The strict
The divisor Dn of Πn is defined inductively as follows: (I)D1 = D1 . transform C3 of C with equation v − 1 = 0, cuts D3 transversally at the
(II)Dn = Dn ∪ Π−1 n−1 point point (v, u) = (1, 0), which is not a corner point. Henceforth, C3 is a
n (D ).
resolution of C.
Note that Π(Dn ) is a finite subset of M : indeed, this corresponds to Note that, the original coordinates (x, y) are related to (v, u) by
centers of the explosions. Moreover, the map Πn |Mn \Dn : Mn \ Dn →
M \ Πn (Dn ) is a biholomorphism. (x, y) = π1 ◦ π2 ◦ π3 (v, u) = (v.u2 , v 2 .u3 ) = π 3 (v, u).
n
The divisor D is the union of k complex curves, each curve biholo-
morphic to C. For instance, the for the second blow-up, if q1 ∈ D1 , then Using this and the parametrization u → (1, u) of C3 , we can obtain a
D2 = D2 ∪ Π−1 −1
2 (D1 ). It follows that Π2 (D1 )  C and that D2 intersects parametrization u → (u2 , u3 ) of C.
Π−1 2
2 (D1 ) transversally at a single point, i.e., D is the union of two em-
Theorem 6.4.3 (Resolution of singularities for curves [10]). Every holo-
bedded projective lines in M2 with a single common point. For sake of
morphic curve in a complex surface admits a resolution.
simplicity we shall use the same notation for the Di and their successive
inverse images Πi , ..., Πn . Then, we can say that Dn = ∪nj=1 Dj .
Corollary 6.4.4. Let S be a holomorphic curve in a complex surface M .
In the case that for every j = 1, ..., n − 1 the j-th blow-up is centered at
Given a point q ∈ S, there exist a neighborhood U of q and holomorphic
a point of Dj , Dn will be a “graph free of cycles”, that is, for every i the
curves S1 , ..., Sm ⊂ U such that:
projective Di another projective Dj transversally at a single point, called
a corner of Dn , in such a way that if Di1 ∩ Di2 = φ,...,Dim−1 ∩ Dim = ∅, (a) q ∈ Sj for every j = 1, ..., m.
then Di1 = Dim .
Such a process will be called blow-up process centered at q. (b) S ∩ U ⊂ S1 ∪ ... ∪ Sm .
(c) Si ∩ Sj = {q}, if i = j.
6.4.1 Resolution of curves
(d) Given j = 1, ..., m, there exists a holomorphic injective map αj : Dr →
Let us now see in what consists the “resolution of a singularity of a curve”. U , where Dr = {z ∈ C ; | z |< r}, such that αj (0) = q, αj (Dr ) = Sj
Let us consider a curve C = (f (x, y) = 0) ⊂ A ⊂ C2 , where f (0, 0) = 0, and the restriction αj |Dr \{0} is an embedding.
that is, 0 ∈ C. We assume that the the Taylor expansion of f is f =
∞ ˜2 In particular, each curve Sj is homeomorphic to the disc D.
j=k fj , where fj is a homogeneous polynomial of degree j. Let π : C →
C the blow-down of C2 in 0. The expression of π in the chart ((t, x), U )
2
Definition 6.4.5. The germs at q of curves S1 , ..., Sm are called the local
of C˜2 , we obtain branches of S at q. For each j = 1, ..., m, the map αj , is called Puiseux
∞ parametrization of the branch Sj .

f ◦ π(t, x) = f (x, tx) = fj (x, tx)
j=k Proof of Corollary 6.4.4. In the case where the point q is not a singular

 point of S the result is straightforward. Indeed, in this case, the has only
= xk . xj−k .fj (1, t) = xk .fU (t, x), one branch at q.
j=k
Suppose now that q is a singularity of S. Let π : M̃ → M a resolution of
S, with divisor D = ∪nj=1 Dj , and S̃ the strict transform of S. Then S̃ cuts

so that π −1 (C) ∩ U = (x = 0) ∪ (fU (t, x) = 0). Analogously, we in the transversally D, at non corner points, forming a finite set, say {q1 , ..., qm }.
chart ((s, y), V ), we have π −1 (C) ∩ V = (y = 0) ∪ (fV (s, y) = 0), where Because the curve S̃ is regular, for each j = 1, ..., m, we can obtain an
fV (s, y) = ∞ j=k y
j−k
.fj (s, 1). Hence, we have π −1 (C) = D ∪ C̃, where embedding βj : Dr → M̃ , which is a parametrization of a neighborhood of
C̃ = (fU = 0) ∪ (fV = 0). The curve C̃ is called strict transform of C. qj in S̃ such that βj (0) = qj . Taking the restrictions of the βj to a smaller
Note that C̃ ∩D is a finite set. Indeed, C̃ ∩D∩U = {(t, 0) ; fk (1, t) = 0}, disc if necessary, we can assume that βi (Dr ) ∩ βj (Dr ) = φ if i = j. Let us
while C̃ ∩ D ∩ V = {(s, 0) ; fk (s, 1) = 0}. put αj = π ◦ βj and Sj = αj (Dr ). It is not difficult to check that S1 , ..., Sm
and α1 , ..., αm satisfy (a),(c) and (d). We leave the verification of (b) as an
In general, if we consider a blow-up process Πn : An → A with divisor
exercise.
Dn = D1 ∪ ... ∪ Dn , we obtain (Πn )−1 (C) = Dn ∪ Cn , where Cn ∩ Dn is a
finite set. The curve Cn is called the strict transform of C by Πn . As a consequence of the above we obtain:
Definition 6.4.1. Let C be a holomorphic curve in a complex surface M . Theorem 6.4.6 (cf. [28]). Let S be a holomorphic curve in a complex
We say that the blow-up process, Πn : Mn → M , with divisor Dn = ∪nj=1 Dj manifold M . Then there exist a Riemann surface S̃ and a holomorphic
is a resolution of C, if the corresponding strict transform Cn satisfies the map φ : S̃ → M with the following properties:
following properties:
(a) φ(S̃) = S.
(i) Cn is regular.
(b) There are discrete subsets A ⊂ S̃ and B ⊂ S such that φ |S̃\A : S̃\A →
(ii) Cn meets each Dj ⊂ Dn transversally. S \ B is an embedding.
(iii) Cn ∩ Dn contains no corners.
(c) φ−1 (B) = A. Moreover, , B is the singular set of S, and for every
p ∈ B, φ−1 (p) is a subfinite set of A.
Let us illustrate this with an example.
Definition 6.4.7. The curve S̃ is called the normalization of the curve S.
Example 6.4.2. We consider the singular plane curve C ⊂ C2 given by
f (x, y) = y 2 − x3 = 0. Let π1 : M1 = C˜20 → C2 the blow-up map of C2 at The theorem above implies that, given a singularity p ∈ S, we can
the origin 0 ∈ C2 . In the local coordinates ((t, x), U ) of M1 , we obtain define the branches of S at (through) p in the following way: since φ−1 (p) =
{q1 , ..., qr }, is a finite subset of S̃, we can obtain for each j = 1, ..., r a disc
f ◦ π1 (t, x) = f (x, tx) = x2 .(t2 − x). Dj ⊂ S̃ such that pj ∈ Dj and Di ∩ Dj = φ se i = j. The germs at p
of φ(D1 ), ..., φ(Dr ) are the branches of S by p. The maps φ |Dj : Dj → S,
Therefore, π1−1 (C) ∩ U consists of the divisor (x = 0) and of the strict j = 1, ..., r, are the Puiseux parametrizations of these branches.
transform C1 of C, with equation x − t2 = 0. Then clearly π1−1 (C) ⊂ U ,
so that it is not necessary to consider the other blow-up chart. The strict
transform C1 of C is regular but not transverse to the divisor D1 , since 6.5 Blow-up of a singular point of a foliation
C1 ∩ D = (0, 0) ∈ U and (x − t2 = 0) is tangent to (x = 0) at this very
point. In other words, the curve is still not resolved. Given a foliation F of M and p ∈ M , the blow-up of F at p is the pull-
back foliation F = π ∗ (F ) of F by the blow-up map π : M p → M . Both
Then a second blow-up π2 (u, t) = (t, tu) = (t, x) is made at (0, 0) ∈ U . p \π −1 (p) (recall that π −1 (p) is the
foliations are equivalent of M \{p} and M
The divisor D2 of this second blow-up is the union of two projective lines,
exceptional divisor π −1 (p) ∼
= P) so that they have the same leaves. Eventual
D1 ∪ D2 , and in the chart (u, t), D1 is represented by (u = 0) and D2 by
“new” singularities are introduced in the exceptional divisor π −1 (p). It may
(t = 0). We have then f ◦ π1 ◦ π2 (u, t) = t3 .u2 .(t − u). Thence, the strict
occur that π −1 (p) is invariant or not. If π −1 (p) is invariant by the foliation
transform C2 of C is (t − u = 0). This last curve cuts D2 at the corner
F = π ∗ (F ) we say that the blow-up is non-dicritical, otherwise it is called 1 ∈ Q+ ) and the blow-up y = tx gives:
dicritical. Let us study this procedure more closely. Let us first see what
occurs with a foliation after a single blow-up. xd(tx) − txdx = 0
Let us consider a holomorphic foliation F in a neighborhood of 0 ∈ C2 xtdx+x2 dt − txdx = 0
with an isolated singularity at the origin 0. We assume that F is represented
x2 dt = 0
by the vector field X = (P (x, y), Q(x, y)) or, equivalently, by the dual 1-
form ω = P (x, y)dy − Q(x, y)dx. We shall denote by F ∗ the foliation dt = 0.
with isolated singularities F ∗ = π ∗ (ω). Thus F ∗ is the pull-back of F via
 2 , trans-
The blow-up is therefore a non-singular foliation on the surface C
the blow-up map π : C˜20 → C2 . Let us investigate its expression in local 0

coordinates. We can write the Taylor expansion of ω at 0 as: verse to P.


 Later on we shall see what is understood as “reduction of a singularity
ω= (Pj dy − Qj dx), of a foliation”. Before we shall introduce some notions and notations.
j=k Let X be a holomorphic vector field defined in a neighborhood of 0 ∈ C2
such that 0 is an isolated singularity of X. Let λ1 and λ2 be the eigenvalues
where Pj and Qj are homogeneous polynomials of degree j, with Pk ≡ 0 of DX(0).
or Qk ≡ 0. The 1-form π ∗ (ω) writes in the chart ((t, x), U ) as:
Definition 6.5.2 (simple singularity). We say that 0 is a simple singularity

 of X, if:

π (ω) = (Pj (x, tx)d(tx) − Qj (x, tx)dx) =
j=k (a) λ1 = 0 and λ2 = 0 (or vice-versa). In this case, we shall say that the
singularity is a saddle-node.


= xk . xj−k .[(tPj (1, t) − Qj (1, t))dx − xPj (1, t)dt]. (b) λ1 , λ2 = 0 and λ2 /λ1 ∈ Q+ . The numbers λ2 /λ1 and λ1 /λ2 are then
j=k called the characteristic numbers of the singularity.

Dividing the above 1-form by xk we obtain: Note that the above conditions are invariant under holomorphic changes
of coordinates and under multiplication of X by a holomorphic function
(∗) x−k .π ∗ (ω) = (tPk (1, t) − Qk (1, t))dx + xPk (1, t)dt + x.α that does not vanish at 0. Thus, the above notion may be extended to the
∞ isolated singularities of holomorphic foliations on complex surfaces.
where α = j=k+1 xj−k−1 .[(tPj (1, t) − Qj (1, t))dx + xPj (1, t)dt].
In few words, the theorem of reduction of singularities assures that given
Set R(x, y) = yPk (x, y) − xQk (x, y), in such a way that x−k .π ∗ (ω) = a foliation F with a finite number of (isolated) singularities on a complex
R(1, t)dx + xPk (1, t)dt + x.α. Analogously, computing the expression of surface M , there exists a (finite) blow-up process, π : M̃ → M , such that
π ∗ (ω) in the chart ((s, y), V ), we obtain: the pull-back foliation F ∗ = π ∗ (F ), which is biholomorphically equivalent
to F outside of the divisor of π, has only simple singularities.
(∗∗) y −k .π ∗ (ω) = R(s, 1)dy − yQk (s, 1)ds + y.β.
Indeed, it is possible to say more.
The polynomial R(x, y) is the tangent cone of ω. We have two cases to Definition 6.5.3. We shall say that the blow-up process is a reduction of
consider: the singularity or resolution of the singularity, if:
(a) R ≡ 0. In this case, we shall say that the singularity is dicritical.
(i) All singularities of F̃ in D are simple.

(b) R ≡ 0. In this case, the singularity is non-dicritical. The tangent cone (ii) A dicritical divisor Dj contains no singularities of F̃, and no tangency
has then degree k + 1. points of F̃ with Dj .
Let us take a closer look at the above cases.
Case (a). In this case, the forms in (*) and (**) are still divisible by x and Theorem 6.5.4 (Theorem of the reduction of singularities, Seidenberg [67]
y respectively. Dividing (*) by x we obtain ). Every isolated singularity of a holomorphic foliation of a complex surface
admits a reduction by a blow-up process.
ω1 = Pk (1, t)dt + α
= Pk (1, t)dt + (tPk+1 (1, t) − Qk+1 (1, t))dx + x.α1 , 6.6 Irreducible singularities
and this form cannot be divided by x, since Pk ≡ 0. The reduction of singularities theorem above mentioned can be made more
The foliation F ∗ is then represented in this chart by ω1 and in the other accurate.
chart by the form ω2 , obtained from the division of (**) by y. Note that, at
the points of the divisor (x = 0), of the form (to , 0) such that Pk (1, to ) = 0, Definition 6.6.1. A singularity of a holomorphic vector field X in dimen-
the leaves of F ∗ are transversal to the divisor. The points (to , 0) such that sion two is called irreducible if it belongs to one of the following categories:
Pk (1, to ) = 0 will be the singular points of F ∗ , or tangency points of the (up to a change of coordinates)
leaves of F ∗ with the divisor. (i) X(x, y) = λx(1+a(x, y))

+μy(1+b(x, y))

λ/μ ∈ C\Q+ , a(x, y),
Note also that each leaf transversal to the divisor, will originate a local ∂x ∂y
b(x, y) are holomorphic with a(0, 0) = b(0, 0) = 0.
separatrix of F via blow-down. Therefore, a dicritical singularity admits
This will be called
 non-degenerate irreducible case.
infinitely many separatrices.
∂ ∂
Case (b). In this case the forms in (*) and (**) cannot be divided anymore. (ii) X(x, y) = λ xk+1 + [y(1 + μxk ) + xb(x, y)] where λ, μ ∈ C,
∂x ∂y
Therefore they already represent the foliation F ∗ in their respective charts. λ = 0, k ∈ N, b(x, y) is analytic of order ≥ k + 1 at 0 ∈ C . 2
In particular, the divisor is invariant by F ∗ . Moreover, the singularities This is the saddle-node case. A basic model for that is the formal normal
of F ∗ in the divisor, are the points, of the (x, t) chart, of the form (0, to ) form presented below:
where R(1, to ) = 0, and also the point (0, 0), of the second chart, if 0 is a
zero of R(s, 1) = 0. We also have that F ∗ has k + 1 singularities, counted Theorem 6.6.2 (Martinet-Ramis [49], Hukuara-Kimura-Matuda, [35]).
with multiplicity, in the divisor. A germ of a saddle-node foliation
 singularity is formally equivalent to an
Note that, if some of the singularities of F ∗ has some separatrix S, then ẋ = xk+1
unique model Fλ,k given by
π(S) is a separatrix of F in 0. ẏ = y(1 + λxk )
Let now us consider the blow-up process at 0 ∈ C2 , consisting of a λ = residue of the saddle-node
succession of n blow-ups, Π : M → C2 , with divisor D = ∪nj=1 Dj . 1 + k = multiplicity of the saddle-node.
The above argumentation proves that we can obtain a foliation F̃ , with Remark 6.6.3. Not all saddle-node singularities are analytically conjugate
isolated singularities, such that in M \ D  C2 \ {0} coincides with F . to the formal normal form. Indeed, the formal normal form admits two
We shall say that the divisor Dj is non-dicritical,if it is invariant by F̃ . separatrices (i.e., two invariant manifolds of dimension one,through the
Otherwise, we shall say that the divisor is dicritical. ẋ = x2
origin), this is not the case of the well-know Euler equation
Example 6.5.1. The singularity xdy − ydx = 0, is not irreducible (λ = ẏ = x + y
Theorem 6.6.4 (Seidenberg 1968, [67] ). Let F be a holomorphic folia- Proof. Let L be a leaf of F such that L \ L = {0}. By Remmert-Stein
tion with an isolated singularity at 0 ∈ C2 . There is a finite sequence of extension theorem (Theorem 2.1.4) the closure Γ := L ⊂ C2 is an analytic
quadratic blow-ups π(j) : Mj → Mj−1 , (j = 1, . . . , ) such that π(1) is the subset of dimension 1. Since Γ is clearly F -invariant and irreducible (Γ \
blow-up of C2 at 0 ∈ C2 and π(j) is the blow-up of Mj−1 at some point {0} = L is connected). We conclude that Γ = L = L ∪ {0} is a separatrix
pj−1 ∈ Mj−1 , with the following properties: of F . The converse is clear.
As a corollary of Seidenberg’s theorem (Theorem 6.6.4) we have.
(a) The pull-back foliation F := π ∗ (F ), where π = π ◦· · ·◦π1 : M → C2 ,
(i.e., is a foliation with only irreducible singularities, of one of the two Proposition 6.7.5. A foliation singularity F at 0 ∈ C2 is dicritical if,
types): and only if, it exhibits infinitely many separatrices.
(i) xdy − λy dx + hot = 0, λ ∈ C \ Q+ xk+1 dy − (y(1 + λxk ) + Proof. First we observe that after the reduction of singularities, a leaf L 
h. o. t.)dx = 0, of F projects onto a (leaf contained in a) separatrix of F if, and only if,
(ii) λ ∈ C, k ∈ N (saddle node), where h. o. t. stands for higher  is not contained in the exceptional divisor D. On the other hand, F is
L
order terms. dicritical if and only if there is a component Pj of D for which every leaf
 intersecting Pj is transverse to D except may be for those at the corners
L
(b) The exceptional divisor D = π −1 (0) is a connected union of embedded Pj ∩ Pi = φ.

Finally the invariant components of D only originate finitely many sep-
projective lines D = Pj , without triple points, transverse intersec-
j=1 aratrices of F (the only possibilities come from separatrices of singularities
tions and which are not at corners, but these singularities are irreducible and there-
fore exhibit at most two separatrices).
(c) A component P of D is either invariant by the foliation F or it is
transverse to F (without tangent points). As we have already mentioned, a separatrix Γ admits a parametrization
ϕ : (C, 0) → (Γ, 0) of type (tn , tm ), n, m = 1 so that (Γ, 0) is homeomor-
The singularity F is called non-dicritical if all components of D are phic to a disc (D, 0). In particular, the leaf L = Γ \ {0} has the topology of
invariant by F, and it is called dicritical if F exhibits some dicritical com- a punctured disc C∗ = C\{0}. Its fundamental group is cyclic isomorphism
ponent. to Z, generated by a loop γ  S 1 . The (local) holonomy group of the leaf
A characterization of dicritical singularities is now possible. This will L is then cyclic generated by a single diffeomorphism f : (C, 0) → (C, 0).
be done in the next section.
Remark 6.7.6. It is here that we notice a drastic difference between the
singular and the non-singular case for foliations: In general it is not so
6.7 Separatrices: dicricity and existence common to find leaves with fundamental group in the non-singular case.
On the other hand, these leaves are quite common in the singular foliations
We begin with a definition that comes from the theory of (Real) ordinary framework; thanks to the following theorem.
differential equations.
Theorem 6.7.7 (Separatrix theorem of Camacho-Sad, [13]). Every holo-
Definition 6.7.1 (Separatrix). Given a (germ of a) holomorphic singular- morphic foliation singularity F at 0 ∈ C2 admits some separatrix.
ity F at 0 ∈ C2 , a separatrix of F is (a germ of) an irreducible analytic
curve Γ 0 which is invariant by F (i.e., Γ \ {0} is contained in a leaf of Remark 6.7.8. The above result is typical from the holomorphic case since
F ). there are examples in the real analytic case where the foliation/vector field
admits no separatrix: take X as the Hamiltonian of f = x21 + x22 in the

Since a separatrix Γ is a germ of an irreducible analytic curve ate real plane R2 (x1 , x2 ). The orbits are concentric circles, no separatrix is
0 ∈ C2 , the Newton-Puiseux parametrization theorem gives a parametriza- allowed.
tion ϕ : (C, 0) → (Γ, 0) of type (tn , tm ), n, m = 1 so that (Γ, 0) is homeo-
morphic to a disc (D, 0). The Separatrix theorem is a by product of a suitable strategy on the re-
duction theorem of Seidenberg (organizing the reduction into linear chains)
Example 6.7.2 (Holomorphic first integral). Given a foliation with an and a residue theorem applied to an index defined in association with a sep-
isolated singularity F at 0 ∈ C2 we say that a non-constant holomorphic aratrix of a singularity. Just to give a few more words about this important
function f : (C2 , 0) → (C, 0) (defined in a neighborhood of the origin 0 ∈ theorem we have:
C2 ) is a holomorphic first integral for F if f is constant along the leaves Let F be a germ at 0 ∈ C2 of a singularity and Γ a smooth separatrix
of F . If F is given by the vector field X (with sing(X) = {0}) then this of F . In local coordinates we may assume that Γ : (y = 0). Then we choose
is equivalent to df (X) ≡ 0. If we consider the dual 1-form ω then this is ω = A(xy)dx + B(x, y)dy a holomorphic 1-form with sing ω = {0}, defining
equivalent to ω ∧ df ≡ 0. In any case, X is parallel to the Hamiltonian F . Since Γ is F -invariant we can write ω = y A1 (x, y)dx + B(x, y)dy with
vector field −A1 (x, 0)
∂ ∂ A1 (x, y) holomorphic. Then we consider the 1-form η := dx. It
Hf := −fy + fx B(x, 0)
∂x ∂y is a meromorphic 1-form with no poles off x = 0 (notice that y  B(x, y) as
and ω is of the form ω = g · df = g(fx dx + fy dy) for some meromorphic a holomorphic function, otherwise ω would have non-isolated zeros). The
function g (g is holomorphic and non-vanishing if f is chosen to be reduced). Index of F relative to Γ at 0 is defined as I(F , Γ, 0) := Res η(x = 0). The
We assume that f (0) = 0 . The separatrices of F are the branches of Index admits a geometrical interpretation as follows:
{f = 0}. It is well-know from the local theory of analytical functions ([31]) Given x, the “inclination” of the tangent space T(x,y) F of the leaf Lx,y) of

r
n F through (x, y) is given by
that f can be written (in a small bidisc centered at 0 ∈ C2 ) as f = fj j ,
j=1
with nj ∈ N, fj holomorphic and such that: dy −y A1 (x, y)
θx (y) = = ·
(a) fj = 0) is irreducible dx B(x, y)
(b) fi and fj are relatively prime
in the local ring O2 so that (fi = 0) ∩ (fj = 0) = {0}. The derivative of this function θx of y at y = 0 is then θx (0)dx =
(*) If moreover f has irreducible/connected fibers then n1 , . . . , nr  = 1. −A1 (x, 0)
dx. The Index I(γ, Γ, 0) is then the residue of this 1-form at
The separatrices are then given by (fj = 0), j = 1, . . . , r. B(x, 0)
The other leaves of F are closed in a neighborhood of 0 and do not x = 0.
accumulate at 0. The foliation is therefore with finitely many separatrices. The Camacho-Sad index theorem ([13], [41]) states that the sum of
indexes of a foliation F at all the singularities in a compact analytic smooth
Example 6.7.3 (Meromorphic first integral). A natural extension of the invariant curve Γ on a complex surface M 2 is equal to the self-intersection
above definition gives as the notion of meromorphic first integral, f : M → (first Chern class) of Γ in M , does not depend therefore on the foliation F .
g
C of a singularity F at 0 ∈ C2 . Writing f = for g, h : (C2 , 0) → (C, 0) 
h I(F , Γp , p) = Γ · Γ ∈ Z.
holomorphic functions which (in the non-trivial case) vanish at 0 ∈ C2 we
p∈sing(F )∩Γ
have that the leaves of F are contained in the curves ag + bh = 0 with
(a, b) ∈ C2 \ {0}. In particular an leaves are contained in separatrices. Exercise 6.7.9. Compute the index of each separatrix in the following
Lemma 6.7.4. Let F be a singularity at 0 ∈ C2 . A leaf L of F is contained cases: linear case, Poincaré-Dulac normal form and saddle-node case.
in a separatrix if, and only if, L \ L = {0}, i.e., L accumulates only at the
singular point.
6.8 Holonomy and analytic classification It is well-known that for a pair of regular foliations a conjugation be-
tween holonomy groups of diffeomorphic leaves induces some conjugation
6.8.1 Holonomy of irreducible singularities between the foliations in neighborhoods of the given leaves.
This is not immediate in case of foliations with singularities (how to extend
As we have already seen there is always a separatrix through a holomorphic
the equivalence/conjugation to the singularities?)
singularity in dimension two. Such a leaf has a holonomy map f : C, 0 →
In the local framework we have for irreducible singularities a precise
C, 0 and we shall study this map in some particular cases:
answer to this question thanks to the work of Martinet-Ramis and some
Example 6.8.1 (linear case). other authors.
 Theorem 6.8.5 (Martinet-Ramis, 1983 [49]). Let F1 , F2 be two germs of
ẋ = λx
saddle-node singularities at 0 ∈ C2 . We assume that (y = 0) is the strong
ẏ = μy separatrix of F1 and F2 and denote by f1 , f2 : C, 0 → C, 0 the holonomy
map if Γ with respect to F1 , F2 . Then there is a germ of a holomorphic
We fix the separatrix Γ : (y = 0). diffeomorphism Φ : C2 , 0 → C2 , 0 taking the leaves of F1 onto the leaves of
Choose a transverse section Σ : {x = 0}. F2 (preserving off course the strong separatrix Γ : (y = 0)) if, and only if,
Remark 6.8.2. Σ is a (complex) disc. there is a germ of a holomorphic diffeomorphism ϕ : C, 0 → C, 0 defining a
conjugation ϕ ◦ f1 ◦ ϕ−1 = f2 between f1 and f2 .
We consider the loop γ(t) = (x0 eu , 0) ⊂ 1; t ∈ [0, 2π]. Let C be the
product γ × Σ  S 1 × D, it is a solid torus. In this solid torus F induces a The same idea holds for non-degenerate irreducible singularities. Thanks
real flow given by the ordinary differential equation to Poincaré-Dulac theorem we only need to consider singularities in the
Siegel domain:
μy(t)
Theorem 6.8.6 (Mattei-Moussu [50], Martinet-Ramis [48]). Let F1 , F2 be
dy dy/dx λx(t) μ x (t)
= =  = y(t) two germs of non-degenerate singularities Fj : xdy − λy(1 + bj (x, y))dx = 0,
dt dx/dt x (t) λ x(t) with bj (x, y) holomorphic, bj (0, 0) = 0, λ ∈ R− . Denote by fj : C, 0 → C, 0
dy μ the holonomy map of Γ : (y = 0) with respect to Fj . Then F1 and F2 are
= y(t).i
dt λ analytically conjugate by a holomorphic diffeomorphism Φ : C2 , 0 → C2 , 0
if, and only if, the holonomy maps f1 and f2 are analytically conjugate in
The solutions are y(t) = y · etiμ/λ . Therefore the first return (holonomy)
Diff(C, 0).
map is given by f (y) = e2πiμ/λ . This is a linear map.
In general for a non-degenerate irreducible singularity F : λx[1+a(x, y)]dy− In the Siegel non-degenerate case the idea of the proof is as follows:
μy[1+b(x, y)]dx = 0 the holonomy map of the separatrix Γ : (y = 0) is given We choose vector field
by f (y) = e2πiμ/λ y + hot in particular its linear part is
∂ ∂
Xj (x, y) = x + λy(1 + bj (x, y))
f  (0) = eeπiμ/λ . ∂x ∂y

Example 6.8.3 (saddle-node normal form case). We consider a saddle- defining Fj (j = 1, 2) in a small bidisc Uj ⊂ C2 . We fix a point (x0 , 0) ∈
node in the normal form Γ\{0} close enough to origin, and consider the holonomy maps hj : (Cy , 0 →
Cy , 0) defined by Fj as the first return map of the (transversely holomor-
y k+1 dx − x(1 + λy k )dy = 0. phic) induced flow Lj := Fj  C where C is the solid torus γ × Σ ∼= S1 × D
as above.

The strong manifold Γ : (y = 0) has holonomy map h(y) given by a similar Because Fj has a saddle like dynamics we know that for 0 < R < |x0 | the
procedure. leaves of Fj intersect transversally the solid torus CR given by |z| = R,
|y| < δ where δ > 0 is small enough.
x(t) = x0 eit Remark 6.8.7. In the abstract picture of the situation above described,
dy dy/dx y k+1 /x(1 + λy k ) i y k+1 (t) the leaves of Fj are visualized as real curves but actually they are complex
= = = curves, i.e., real surfaces of real dimension two.
dt dx/dt dx/dt 1 + λy k (t)
⎧ k+1 The analytic conjugation ϕ : (Σ, (x0 , 0)) → (Σ(x0 , 0)) between h1 and
⎨y  (t) = i y (t)
1 + λy k (t) h2 satisfies
⎩ ϕ ◦ h1 = h2 ◦ ϕ
y(0) = y

f (y) = y(2π). First we extend ϕ to the solid torus C = C|x 0|
by setting

2
ẏ = y    
For instance if k = 1 and λ = 0, i.e., for the saddle-node we ϕ y1 (t, y) = y2 t, ϕ(y)
ẋ = x

then have where yj (t, y) is the solution of the flow Lj = Fj  C that starts from y ∈ Σ.

d −1 The above definition is consistent/valid because y1 (2π, y) = h1 (y) and
y  (t) = i y 2 (t) ⇒ =i
dt y(t)
y2 (2π, ϕ(y)) = h2 (ϕ(y))
−1 −1
⇒ = it + c ⇒ c =
y(t) y(0) and by hypothesis we have
−1 −1 y(0)
⇒ = it − ⇒ y(t) = ϕ ◦ h1 (y) = h2 ◦ ϕ(y).
y(t) y(0) 1 − ity(0)
y
⇒ f (y) = y(2π) = this is a homography Thus we have extended ϕ to the solid torus C. Now we show how to extend
1 − 2πiy
ϕ “radially” to the torii CR with 0 < R < |x0 |.
In general, for a general form saddle-node, the strong manifold (given by (t) = e−t x and
We consider the induced radial flow where we consider x
(y = 0) in the form y k+1 dx − [x(1 + λy k ) + (· · · )]dx = 0) the holonomy yj (t, y) the solution of
map of Γ is given by f (y) = y + ak+1 y k+1 + · · · where ak+1 = 0. It is a ⎧
map tangent to the identity. ⎨ dyj
= Xj (e−t x, yj )
dt
Exercise 6.8.4. Calculate the holonomy map of the Poincaré-Dulac nor- ⎩y (0) = y
j
mal form F : ydx − (nx + ay n )dy = 0, n ≥ 2; for the (only) separatrix
Γ : (y = 0). Then we extend ϕ : C → C  to the “interior” of C by setting
   
6.8.2 Holonomy and analytic classification of irreducible ϕ y1 (t, y) = y2 t, ϕ(y) .
singularities
Some estimative shows that
 
Let us now discuss on of the most important aspects of the concept of y1 (t, y) ≤ eλ.(1+ε)t |y|
holonomy for singularities of foliations.
for some constant 0 < ε  |λ|. (Recall that (important!) λ < 0). Similarly ⇓
we also have  
y2 (t, y) ≤ |y|.e|λ|(1+ε)t . (i) F is non-dicritical.
This shows that we have (ii) The leaves are closed off the singularity.

y2 (t, y)| ≤ A.|


| y1 (t, y)| All this for a sufficiently small neighborhood of the singular points 0 ∈
C2 .
for some constant A > 0.
Riemann extension theorem (Theorem 2.1.1) now shows that ϕ extends Remark 7.1.2. Thanks to Remmert-Stein extension theorem (Theorem 2.1.4)
to the vertical axis x = 0 for |y| ≤ δ for a certain 0 < δ. a leaf which is closed off 0 ∈ C2 and is not (contained in) a separatrix, is
contained in an analytic curve and is called analytic leaf.
We shall say that a germ F at 0 ∈ C is analytically linearizable if there is
a holomorphic diffeomorphism Φ : C2 , 0 → C2 , 0 taking the leaves of F onto The theorem of Mattei-Moussu states a direct converse to the above and
leaves of a linear foliation Fλ : xdy − λydx = 0, λ ∈ C \ {0}. In this case Fλ can be stated in dimension two as follows:
is unique and F is of the form F : xdy − λydx + hot = 0. Similarly a germ
Theorem 7.1.3 (Mattei-Moussu, [50]). Let F be a holomorphic foliation
of a holomorphic diffeomorphism f : C, 0 → C, 0 is analytically linearizable
singularity at 0 ∈ C2 . Assume that for a small neighborhood V of 0 ∈ C2
if there is a germ of a holomorphic diffeomorphism ϕ : C, 0 → C, 0 such
we have:
that that ϕ ◦ f = fλ ◦ ϕ for some linear map fλ (y) = e2πiλ .y.
Then, as a corollary of Theorem 6.8.6 above we have: (i) The leaves of F in V are closed in V \ {0} ⊂ C2 .
Theorem 6.8.8. A germ of an irreducible non-degenerate singularity F : xdy− (ii) Only a finite number of leaves of F in V accumulate at 0 ∈ C2 .
λydx + · · · = c is analytically linearizable if, and only if, its holonomy map
of a given separatrix is analytically linearizable. Then F admits a (non-constant) holomorphic first integral F : W → C in
some open subset 0 ∈ W ⊂ V .
The classical proof relies on the Reduction of Singularities, as well as
on the dynamics of holomorphic diffeomorphisms f : C, 0 → C, 0.
R. Moussu has given an alternative proof based on the classical Reeb lo-
cal stability theorem ([26]) and in the dynamics of diffeomorphisms f : C, 0 →
C, 0 ([51]).
The next example, due to M. Suzuki and Cerveau-Mattei, shows that
there is no topological criteria for the existence of a meromorphic first
integral.

Example 7.1.4 (Suzuki’s example, [72]). Consider the germ of singular


foliation F at the origin 0 ∈ C2 given by: Ω = 0 where Ω = (y 3 + y 2 −
=
xy)dx − (2xy 2 + xy − x2 )dy. The germ F has the Liouvillian first integral
=
f (x, y) = x
y exp[ y(y+1)
x ] and the following remarkable properties:

(i) F is μ − simple, that is, it is a dicritical germ which is desingularized


=
with only one blow-up and the resulting foliation has no singulari-
ties on the exceptional divisor, it is transverse to this projective line
everywhere except for (a unique) tangent point (see [40]).

Therefore it follows that:


Chapter 7 (i)’ Every leaf of F is a separatrix and therefore is given by some equation
=
(f = 0) where f ∈ O2 .
(ii) F does not admit a meromorphic first integral in any neighborhood of
Holomorphic first =
the origin 0 ∈ C2 (see [18] for a proof, or follow our argumentation).

integrals Performing a blow-up (y = tx) at the origin 0 ∈ C2 we obtain the


foliation
F̃ : t3 dx + (2xt2 + t − 1)dt = 0
given by the vector field
7.1 Mattei-Moussu theorem ẋ = 2xt2 + t − 1, ṫ = t3 .
From the structural viewpoint the singularities with a holomorphic first The initial foliation has the Liouvillian first integral f = xy exp( y(y+1) )
x
integral are the most simple singularities of a holomorphic foliation.
and therefore the foliation above has the Liouvillian first integral f (x, t) =
Definition 7.1.1. Given a singular foliation germ F at 0 ∈ C2 , a holomor- 1 t(xt+1)
t e . Restricting this function to the projective line (x = 0) we obtain
phic function F : C2 , 0 → C, 0 (i.e., a holomorphic function F defined in 1
f (0, t) = 1t e t which is a Liouvillian function on C. The map σ : (C, 1) →
some neighborhood V of 0 ∈ C2 and with F (0) = 0 ∈ C), is a holomorphic
(C, 1) defined by mapping the point p ∈ (C, 1) onto the other intersection
first integral of F if it is constant along the leaves of F .
point of the leaf Lp of F̃ though p with the projective line, is (because of
If F is given by the vector field X with an isolated singularity at 0 then the order-2 tangency) a germ of involution on (C, 1). This germ is given by
the above condition is equivalent to df (X) ≡ 0. In terms of the dual 1-form the relation f (0, t) ◦ σ = f (0, t), that is, 1t et = σ(t)
1
eσ(t) . This defines σ(t)
ω, the condition becomes ω ∧ df ≡ 0. This later, thanks to Saito’s division as a nonalgebraic Liouvillian function on C and according to [40] this is
lemma ([64]), is equivalent to ω = gdf . enough to conclude that F does not admit a nontrivial meromorphic first
As we have already seen if F admits a (non-constant) holomorphic first =
integral F : C2 , 0 → C, 0 then its leaves satisfy: integral.

(i) the leaves of F are closed outside of the origin


(ii) there are only finitely many leaves that accumulate at the origin.
7.2 Groups of germs of holomorphic diffeo-
In terms of the language of Seidenberg’s theorem (Theorem 6.6.4) we morphisms
have:
We shall start with some basic facts. We denote by Diff(C, 0) the group of
F admits a holomorphic first integral germs of holomorphic diffeomorphisms f : C, 0 → C, 0. Such a map germ

has representatives given by maps fV : V → fV (V ), where 0 ∈ V ⊂ C is • P ⊂ D is the set of periodic points of hD .
an open set, and fV : V → fV (V ) is a holomorphic diffeomorphism with
fV (0) = 0. It can be identified (the germ f ) with a power series f (z) = • F ⊂ D is the set of non-periodic points of finite D-orbit.
f  (0)z + aj z j ∈ C{z} where f  (0) = 0. • I ⊂ D is the set of points with infinite D-orbit.
j≥2

Lemma 7.2.1. Let G ⊂ Diff(C, 0) be a finite subgroup. Then G is cyclic Now we define a sequence of compact subsets An ⊂ D as follows:
 2πi 
and analytically conjugate to a group z → e ν z ν ∈ N.
A0 := D
g(z) A1 := D ∩ h−1 (D)
Proof. We define a map Φ : C, 0 → C, 0 by Φ(z) = 
· This is a well-
g∈G g (0) ..
defined holomorphic map because G is finite (in particular all elements of .
G have a common definition domain around 0 ∈ C). Moreover Φ (0) = An := D ∩ h−1 (D) ∩ · · · ∩ h−n (D)
|G| = 0 so that Φ ∈ Diff(C, 0). Given now any element g0 ∈ G we have ..
.
 g(g0 (z))  (g ◦ g0 )(z)
Φ(g0 (z)) = = g0 (0)
g  (0) g  (0)g0 (0) Also let Cn be the
 connected component of An that contains the origin
g∈G g∈G
 (g ◦ g0 )(z) 0 ∈ C and C := Cn . Then by construction a point in C is periodic or
= g0 (0) = g0 (0)Φ(z).
n≥0
(g ◦ g0 )(0) has infinite D-orbit, i.e., C ⊂ I ∪ P .
g∈G
Claim 7.2.6. If C is countable then I is not countable.
Thus Φ ∈ Diff(C, 0) is an analytic conjugation of G with a finite subgroup
of the linear group GL(1, C) = C∗ ; this ends the proof. Proof of Claim 7.2.6. By hypothesis C is countable. Therefore there is a
subdisc Dr ⊂ D of radius 0 < r < radius of D; such that C ∩ ∂Dr = φ
Given a holomorphic function (z) defined in a neighborhood of 0 ∈ C (otherwise C is not countable).
we consider the invariance group of (z) as Inv( ) = {g ∈ Diff(C, 0); ◦ g = Therefore ∃ m ∈ N such that Cm ∩ ∂Dr = φ.
} in terms of germs. Assume that is not constant, (0) = 0. Since up Let K be a compact, connected neighborhood of Cm that does not
to a change of coordinates we have (z) = z ν for some ν ∈ N we conclude intersect the other components of Am . In particular we have φ = ∂K ∩
that: Am = ∂K ∩ D ∩ h−1 (D) ∩ · · · ∩ h−m (D) so that there must exist for
Lemma 7.2.2. The invariance group Inv( ) is a finite cyclic group. every x ∈ ∂K a m ∈ N such that N p < m and such that hp (x) ∈ / D.
Consequently we get P ∩ ∂K = φ.
By the two above lemmas we have that the finite subgroups of Diff(C, 0) Now we denote the sets
are the invariance groups of holomorphic functions : C, 0 → C, 0. P := h-periodic points in K
A final simple remark concerning the finiteness of subgroups of Diff(C, 0) F := non periodic points but with finite K-orbit
is: I := points with infinite K-orbit.

We stratify P as P = Pn where Pn := { points periodic of period
Lemma 7.2.3. Let G ∈ Diff(C, 0) be a finitely generated subgroup such n≥0
that each element g ∈ G is periodic (∃ ng ∈ N such that g ng = Id). Then n}.
G is finite. We observe that the boundary ∂ Pn is finite: indeed, if x0 ∈ ∂ Pn is an
accumulation point of points ∂ Pn there exists a neighborhood V of x0 such

Proof. First we observe that a non-trivial flat element g(z) = z+ak+1 z k+1 + that hn (x) = x, ∀ x ∈ V because V ∩ Pn = φ (Identity Principle). Since
· · · (ak+1 = 0) is not periodic; indeed g n (z) = z + nak+1 z k+1 + · · · . The the orbit of x0 does not intersect ∂K we have, for V sufficiently small, that
commutator of two elements g1 , g2 ∈ G is [g1 , g2 ] = g1 g2 g1−1 g2−1 is a map hj (V ) ⊂ Int(K) ∀ j = 0, 1, . . . , n − 1. Thus V ⊂ Int(Pn ), i.e., x0 ∈ Int(Pn ).
tangent to the identity. Thus by hypothesis this is the identity and G is This actually means that points in ∂ Pn are isolated (⇒ ∂ Pn is finite).
abelian. This implies that G is finite because it is finitely generated. As a consequence the boundary of P is countable.
The compact set K can be decomposed as K = Int(P) ∪ F ∪ (I ∪ ∂ P).
Now we get to the main point in our argumentation:
Notice that Int(P) is an open subset of C since by hypothesis P ∩ ∂K = φ.
Proposition 7.2.4 (Finiteness condition). A germ of a holomorphic dif- By its turn, F is open in K and may intersect ∂K. If F = φ then ∂K ⊂
feomorphism f : C, 0 → C, 0 has finite order ( i.e., f n = Id for some n ∈ N) I∪∂ P and since ∂ P is countable (as we have just seen above) and ∂K is not
if, and only if, for some neighborhood V of 0 all the orbits of f in V are countable it follows that I is not countable. If F = φ and Int(P) = φ we
finite. conclude that for every r > 0 small enough, I∪ ∂ P intersects the boundary
The classical proof is as follows: ∂Dr so that I is not countable. If now Int(P) = φ and F = φ then, because
these are disjoint open subsets, the set K − Int(P) ∪ F is not countable.
Lemma 7.2.5. Let K ⊂ Rn be a compact connected neighborhood of 0 ∈ Rn Since ∂ P is a countable set this implies that I is not countable; proving
and h : K → h(K) ∈ Rn a diffeomorphism with h(0) = 0. Then there Claim 7.2.6.
exists a boundary point x ∈ ∂K such that the number of iterates of x by h
contained in K is infinite. Lemma 7.2.7. Suppose that C is not countable. Then I is not countable.

In general, for x ∈ K we define μK (x) := {n; h (x) ∈ K}.


n Proof. By Lemma 7.2.5 for every compact disc 0 ∈ Dr ⊂ D we have (P ∪
 I) ∩ ∂Dr = φ. Therefore P ∪ I is not countable. Suppose by contradict
Proof. Assume by contradiction that μK is bounded in ∂K, i.e., μK ∂K < that I is countable; then P is not countable. Since C = (C ∩ I) ∪ (C ∩ P )
N < ∞ for some N ∈ N. Let us consider (recall that C ⊂ I ∪ P ) we have that C ∩ P is not countable (otherwise C
would be countable, contradiction). Let us write
A = {x ∈ K; μK (x) < N } ⊃ ∂K


C ∩P = Pn
B = {x ∈ Int(K); μ ◦ (x) ≥ N }, K = Int(K)
K n≥0

Then A, B are open subset of K, A ⊃ ∂K, 0 ∈ B and A∩B = φ. Therefore, where Pn = points periodic of period n in P ∩ C. Then there exists n0
since K is connected, ∃ x0 ∈ K such that x0 ∈ / A ∪ B, i.e., μK (x0 ) ≥ N > such that Pn0 is not finite with an accumulation point in Cn0 . Since hn0 is
μ ◦ (x0 ). Thus there exists n0 such that y0 = hn0 (x0 ) ∈ ∂K. Then we have holomorphic in an open neighborhood of Cn0 we conclude from the Identity
K
μK (y0 ) = μK (x0 ) ≥ N ; contradiction. This proves Lemma 7.2.5. Principle that hn0 ≡ Id, contradicting the non-periodicity of h.
Our strategy to prove the Finiteness condition proposition is to prove: All together, Claim 7.2.6 and Lemma 7.2.7 prove that the set I of points
If h ∈ Diff(C, 0) is not periodic then exists fundamental system of neigh- x ∈ D with infinite D-orbit is not countable. This proves Proposition 7.2.4.
borhoods U of 0 ∈ C such that for each neighborhood U ∈ U the set of
points x ∈ U where the U -orbit is not finite is not countable and contains
the origin 0 ∈ C in its closure. 7.3 Irreducible singularities
We start by fixing a compact disc 0 ∈ D ⊂ C such that h : D → h(D)
is a (holomorphic) diffeomorphism. Now we consider the 3 following sets: We shall now address the irreducible case proving Mattei-Moussu in this
situation:

Lemma 7.3.1. Let F be a germ of an irreducible singularity at 0 ∈ C2 . Proof. Indeed, each map in h ∈ Inv(Fj Σ ) takes leaves of F into leaves of
Assume that F has closed leaves off the origin 0 ∈ C2 . Then F admits a F so the same holds for the maps h ∈ Inv(F, Σ). Since the leaves of F
holomorphic first integral. Indeed, F is analytically conjugate to nxdy + are analytic curves (except for those contained in separatrices) we conclude
mydx = 0, for some n, m ∈ N; in a neighborhood of 0 ∈ C2 . that each map in Inv(F,  Σ) has closed orbits. This implies that each map
Proof. Since by hypothesis F is irreducible, we divide the proof in two in Inv(F, Σ) is periodic (Finiteness condition Proposition7.2.4). Because
cases: Inv(F, Σ) is finitely generated this implies that Inv(F, Σ) is a finite group,
eπi
Case 1: F is a saddle-node singularity germ. indeed cyclic generated by a periodic rotation z → e ν z, (ν ∈ N).
In this case we have the strong manifold say F : y k+1 dx − [x(1 + λy k ) +
· · · ]dy = 0 Γ : (y = 0) The holonomy map by of the strong manifold is of
the form Thus there is a holomorphic function F on Σ such that every map in
hγ (y) = y + ak+1 y k+1 + · · · ak+1 = 0. Inv(F, Σ) leaves invariant the 
 map F ; i.e., F ◦ h = F , ∀ h ∈ Inv(F , Σ).
Because Inv(F, Σ) ⊃ Inv(Fj Σ ) this implies that F is also constant along
Therefore hγ is not periodic. This implies that the orbits of hγ are not all
the levels of Fj in Uj .
of them closed and thus F has some non-analytic leaves on small neighbor- 
r
hoods of the origin. Thus, this case cannot occur. Thus we have constructed a holomorphic first integral F for F in Uj .
j=1
Remark 7.3.2. Indeed it is possible to say much more about the dynamics 
r
Now the complementary part of Aj in P is topologically a disc, i.e.,
of hγ is no orbit is finite except for the fixed point. j=1

Case 2: (non-degenerate case). it is simply-connected.


We write F as xdy − λydx + hot = 0 λ ∈ N \ Q+ with invariant axes. This complement then has trivial holonomy and therefore F admits a
Again we consider the holonomy of Γ : (y = 0). It is a map of the form holonomy and therefore F admits a holonomy extension as a holomorphic
first integral of F to a neighborhood U  2 . This projects into a
 of P in C
hγ (y) = e2πiλ y + · · · If λ ∈
/ R then Poincaré Linearization theorem implies 0
that F is analytically linearizable and therefore hγ (y) is conjugate to y → holomorphic first integral for F in a neighborhood of 0 ∈ C2 .
e2πiλ y. This later map only has finite orbits when λ ∈ Q so we would have
λ ∈ Q+ contradiction. Therefore λ ∈ R− and indeed because hγ is periodic 7.5 The general case
we conclude that it is linearizable and the same holds for F ; moreover (still
because hγ is periodic) we must have λ ∈ Q− say, λ = −n/m, n, m ∈ N As a final word about the proof in the general case, where several blow-ups
and F is analytically linearizable as nxdy + mydx = 0. may be needed to finish the reduction of singularities, we have a similar
argumentation as above by the Induction on the number of blow-ups in the
reduction of singularities.
7.4 The case of a single blow-up
In order to illustrate the main ideas in the proof we consider the following
situation: F can be reduced with a single blow-up π : C  2 → C2 . Since F is
0
non-dicritical the exceptional divisor is F-invariant, F = π ∗ (F ). Because
the leaves of F are closed off the origin, the non-separatrices are analytic
leaves and the same holds for the leaves of F which are contained in P or
the separatrices of F transverse to P.

Remark 7.4.1. P \ sing(F ) is a leaf of F .

Write Sing(F) = {p1 , . . . , pr } ⊂ P. From the irreducible case above


we conclude that for each j ∈ {1, . . . , r} there is a neighborhood Vj of
pj in C 2 where F admits a (non-constant) holomorphic first integral say
0
Fj : Vj → C. We may assume that Vj is a product Vj = Dj × Dε of a disc

pj ∈ Dj ⊂ P and a small disc 0 ∈ Dε of radius ε > 0. Assume also that

r
Chapter 8
Di ∩Dj = φ, ∀ i = j. Fix now a point p0 ∈ P\ Dj . Since P is a 2-sphere
j=1
we may choose a simply-connected domain Aj ⊂ P such that

Aj ∩ {
p0 , p1 , . . . , pr } = {
p0 , pj }.
Dynamics of a local
Claim 7.4.2. We may extend Fj to a holomorphic first integral Fj for F diffeomorphism
in a neighborhood Uj of Dj ∪ Aj .
Proof. Indeed this is the classical holonomy extension which is possible
because Aj is simply-connected therefore it has no further holonomy that
the one already in Dj . Let us be more precise: Fix a point  aj ∈ (Dj \ We consider f a germ of holomorphic diffeomorphism at 0 ∈ C fixing 0; say
f (z) = λz + ak+1 z k+1 + . . . , k ≥ 1, λ ∈ C∗ . We shall describe its dynamics
pj }) ∩ Aj and a transverse disc Σqj to F with {
{ qj } = Σqj ∩ P.
(i.e., the dynamics of its pseudo-orbits) according to the multiplier λ =
We consider a simple path δj : [0, 1] → tj with δj (0) = p0 . δj (1) = qj .
f  (0). For a quick and effective reference in this subject we recommend [5].
We consider the holonomy map hδj : Σ → Σ qj of the path δj in the leaf
p1 , . . . , pr } of F.
P \ {
Then we define Fj (y) := Fj (hδj (y)) for y ∈ Σ close enough to p0 . 8.1 Hyperbolic case
This is well-defined because Aj is simply-connected and Fj is already invari-
ant by the local holonomy map associated to a small loop γj ⊂ Dj \ { pj }. (i) Hyperbolic Case: |λ| = 0, 1.
Thus we can construct the (holonomy) extension Fj as above. We may also By Poincaré-Königs theorem (1884) there is a unique holomorphic dif-
that Vj ⊂ Uj . feomorphism ϕ with ϕ (0) = 1, that conjugates f to the linear map z → λz.
Then we have if |λ| < 1 that f n (z) → 0, as n → ∞, for each z ≈ 0,

For each extension Fj we consider the restriction Fj Σ and the invari- following a spiralling or linear path, depending on whether λ is complex on
ance group pure real.
  
Inv Fj Σ = {h ∈ Diff(Σ, p0 ) : Fj ◦ h = Fk }.
8.2 Parabolic case
Finally we consider the global invariance group Inv(F , Σ) := subgroup of
(ii) Parabolic Case: |λ| = 1, λk = 1 for some k ∈ N.
Diff(Σ, p0 ) generated by the invariance groups Inv(Fj Σ ) j = 1, . . . , r.
Proposition 8.2.1. Let h : (C, 0) → (C, 0) be a germ of a holomorphic
Claim 7.4.3. Inv(F, Σ) is a finite group. diffeomorphism tangent to the identity h(z) = z + aj z j , a2 = 0. Then
j≥2
there exist sectors S + and S − with vertex at 0 ∈ C, angles π − θ0 (where 8.3 Elliptic case
0 < θ0 < π/2) and opposite bisectrices in such a way that:
Elliptic Case: |λ| = 1, λ = e2πiθ for some θ ∈ R \ Q.
(i) h(S + ) ⊂ S + , lim hn (z) = 0, ∀ z ∈ S + The case is pretty rich and it is a subject of deep research. The main
h→+∞
question first posed is whether f = λz + . . . is always analytically lineariz-
−1 −1
(ii) h (S ) ⊂ S−, lim h−n (z) = 0, ∀ z ∈ S − able. It was Cremer who first gave an example of an elliptic map which is
n→+∞
not analytically linearizable. Indeed Cremer introduced the following:
Proof. A linear change of coordinates allows us to write h(z) = z − z 2 +
Cremer condition, [20]: for θ ∈ R \ Q if
aj z j . For |z| < δ small enough we consider g(z) defined by
j≥3
lim sup |{nθ}|−1/n = ∞
⎧ n→∞
⎨g(0) = 1
z then there exist an elliptic germ f (z) = e2πiθ z + O(z 2 ) which is not lin-
⎩h(z) =
1 + zg(z) earizable.
In the above statement, for a positive real number x ∈ R, we have
Then g(z) is analytic (for |z| < δ small enough) and we consider the coordi-
1
nate ξ = , z = 0. Under these coordinates we have h as  h(ξ) = ξ +g(1/ξ), {x} := x − [x] [x] = the integral part of x.
z

h(∞) = ∞, a diffeomorphism defined in a neighborhood of ξ = ∞. Because A number θ satisfying the Cremer condition above is called a Cremer num-
1 ber . Cremer numbers form a dense subset of R of zero Lebesgue measure
g(0) = 1 we have g(1/ξ) = 1 +  g(ξ) with 
g bounded as ξ → ∞. Therefore
ξ [5].
 1
h(ξ) = ξ + 1 +  g(ξ) is a map which is close to the translation T (ξ) = ξ + 1 On the other hand there are arithmetical conditions by originally by
ξ Siegel and recently J.C. Yoccoz and Bryuno ([6]) for assuring that if θ ∈
for |ξ| big enough. R \ Q satisfies this arithmetical condition then f is always analytically
Hence we can find a sector S + of horizontal bisectrix and angle π − θ0 linearizable. These θ form a full Lebesgue measure subset of R.
(0 < θ0 < π/2) and a disc D+ = (ξ1 − R)2 + ξ22 ≤ (2R)2 so that if Another remarkable result is:
S+ := S+ ∩ (CD+ ) then  h(S+ ) ⊂ S+ and lim  hn (ξ) = ∞. Similarly we
n→∞
Theorem 8.3.1 (Sigel-Bryuno-Yoccoz, [58]). Let θ ∈ R \ Q and λ = e2πiθ .
h−1 (S− ) ⊂ S− and lim 
obtain S− such that  h−n (ξ) = ∞ on S− .
n→∞ If the germ pλ (z) = λz + z 2 is analytically linearizable then every germ
The sector S + and S − are then the images of S+ and S− respectively, f ∈ Diff(C, 0) with f  (0) = λ, is also analytically linearizable.
by the change of coordinates z = 1/ξ.
Regarding the dynamics we have the remarkable work of Pérez-Marco

([58, 59, 60]). He introduces the following concept:
Similarly, for the case, h : C, 0 → C, 0 h(z) = z + j
aj z , ak+1 = 0;
j≤k+1
Definition 8.3.2 (Small cycles property). A small cycle for f is a finite
we have sectors S1+ , . . . , Sk+
and sectors S1− , . . . , Sk−
so that Sj+ alternates
orbit of f (a subset {p1 , . . . , pn } ⊂ C \ {0} such that pi = pj and f (pi ) =
with Sj− which alternates with Sj+1
+
; having angles ≥ π/k, in such a way pj+1 (mod n)). We say that f has the small cycles property if for any open
that neighborhood U of 0 then exists a small cycle for f contained in U .
(a) Sj+ ∩ Sj−1 = φ In this case the small cycles accumulate at 0. In particular the germ f
is not linearizable.

(b) h(Sj+ ) ⊂ Sj+ , h−1 −


j (Sj ) ⊂ Sj

Theorem 8.3.3 (Pérez-Marco). There exist elliptic germs with the small

k cycles property. Not all non-linearizable elliptic germs have the small cycles
lim hn (z) = 0, ∀ z ∈ Sj+ property.
n→∞ j=1

k
Pérez-Marco gives an arithmetic condition on θ in order to decide whether
lim h −n
(z) = 0, ∀ z ∈ Sj−
n→∞ j=1 the non-linearizable germ has the small cycles property.
Indeed, more can be said: Actually Pérez-Marco work goes much further with the introduction of
the Hedge-Hogs. He also concludes
Theorem 8.2.2 (Camacho, [8]). Let f (z) = λz + O(z 2 ) be a holomor-
phic germ of a complex diffeomorphism, λn = 1 for some n ∈ N (with n Theorem 8.3.4 (Pérez-Marco). A non-linearizable elliptic map always has
minimal). Then: arbitrarily close to the origin some orbit which accumulates at the origin.
(i) Either f n (z) = z, Such an orbit cannot be closed.
(ii) or there exists k ∈ N such that f is topologically conjugate to Tk,λ,n : z → For an elliptic germ f : C, 0 → C, 0 we can add: Choose an open con-
λz(1 − z nk ). nected subset 0 ∈ U ⊂ C where f is univalued the stable set of f in U is


If moreover f (z) = z + ak+1 z k+1 + O(z k+2 ) ak+1 = 0, then f is topologi- K(U, f ) = f −j (U ).
j=0
cally conjugate to Tk : z → z + z k+1 .
Theorem 8.3.5 (Pérez-Marco). Let f : C, 0 → C, 0 be an elliptic map
And also:
germ with stable set K(f, U ). Then:
Theorem 8.2.3 (Fatou-Leau flower theorem, [1]). Let f (z) = z + z k+1 +
O(z k+2 ), k ∈ N. Then there exist k domains called petals, Pj , symmetric (i) K(f, U ) is compact, connected, full (i.e., U \ K(f, U ) is connected),
2πq contains the origin but is not restricted to the origin, i.e., 0 ∈ K(f, U ) =
with respect to the k directions arg(z) = , q = 0, . . . , k − 1, such that: {0}. Moreover, K(f, U ) is not locally connected at any point distinct
k
(i) Pj ∩ Pk = φ for j = k; 0 ∈ ∂Pj and each petal Pj is holomorphic to from the origin.
the right-half plane H ⊂ R2 (ii) Any point of K(f, U ) \ {0} is recurrent (that is, it is a limit point of
(ii) for each z ∈ Pj we have f m (z) → 0 as m → ∞ moreover. its orbit).

(iii) For each j the map f Pj is holomorphically conjugate to the parabolic (iii) There is an orbit in K(f, U ) that accumulates at the origin.
automorphism z → z + i on H.
(iv) No non-trivial orbit converges to the origin.
If f (z) = z + z k+1 + O(z k+2 ) then f −1 (z) = z − z k+1 + O(z k+2 ).
The stable set K(f, U ) is called a hedge-hog.
Therefore we get, from Leau-Fatou theorem, k attracting Qj for f −1
(zq + 1)...
symmetric with respect to the k directions arg(z) = , q =
k
0, . . . , k − 1. These directions are the bisectrices of the angles between
two consecutive attracting directions
 for f . The Qj ’s are repelling petals
for f , intersecting the Pj ’s and Pj ∪ Qj ∪ {0} is an open neighborhood of
j
0 ∈ C in C. We have therefore a pretty clear description of the dynamics
of f .
Thus we have a polynomial vector field X(u, v) with isolated singularities
in C2(u,v) such that in the intersection of the spaces C2(u,v) and C2(u,v) we
1 
have X =  X for some ≥ 0.
u
Remark 9.1.3. In any case we conclude that X  defines a foliation on
C2(u,v) and this foliation coincides with the one induced by X on C2(x,y) .
Chapter 9
1 x
Similarly for the coordinates r = , s = we conclude that there is
y y

Foliations on complex a polynomial vector field X(r, s), with isolated singularities on the plane
C2(r,s) , such that in the intersection C2(x,y) ∩ C2(r,s)

projective spaces X=
1
rk
X

for some 0 ≤ k ∈ N.
Therefore, just applying the definition, we conclude that X defines a folia-
tion F (X) on the complex projective plane in a natural way.
9.1 The complex projective plane and folia- Summarizing the above example we have: A polynomial vector field X on
tions C2 defines/induces a foliation F (X) on CP (2). Conversely we have:

The complex projective plane CP (2) is the quotient space C3 \ 0 by the Proposition 9.1.4. Every holomorphic foliation with singularities on CP (2)
equivalence relation p, q ∈ C3 \ 0, p ∼ q ⇔ p = λ.q for some λ ∈ C \ {0}. is the one induced by a certain polynomial vector field on C2 .
Thus CP (2) is the space of lines through the origin of C3 . By introducing Proof. Let F be a foliation on CP (2) with (finite) singular set sing(F ) ⊂
homogeneous coordinates [(x1 ; x2 ; x3 )] on CP (2) we conclude that CP (2) CP (2). Denote by π : C3 \ {0} → CP
is equipped with an atlas u consisting of three affine charts (x, y), (u, v),  (2) the canonical projection.
By definition we have CP (2) = Uj a (finite because CP (2) is compact)
(r, s) with the following changes of coordinates: j
u = 1/x, v = y/x ; r = 1/y, s = x/y ; s = 1/v r = u/v finite open cover where F has its plaques given by holomorphic vector
A well-known fact is then: fields Xj on Uj and having isolated singularities. We can assume that the
intersections Ui ∩ Uj (i = j) contain no singularities of F and that if
Proposition 9.1.1. The complex projective plane CP (2) is a compact, Ui ∩ Uj = φ then Xi = gij Xj with gij ∈ O∗ (Ui ∩ Uj ).
connected and simply-connected complex surface.
Remark 9.1.5. The idea if to “lift” F to C3 \ {0} by π : C3 \ 0 → CP (2)
Let us now investigate the structure of the space of holomorphic foli- and then prove that F (because it comes from a foliation on CP (2)) can be
ations with singularities (foliation of dimension one are the ones who are given by a polynomial system. Nevertheless, the lifted foliation π ∗ F will
interesting) on the complex projective plane. Recall that a holomorphic fo- have codimension one i.e., dimension two. So we must pass to differential
liation with singularities on CP (2) is given by an open cover CP (2) = Uj forms instead of vector fields.
j
such that on each open subset Uj the plaques of F are given by a holomor-
 Let us then choose dual 1-forms
 ωj on the Uj such that on each Ui ∩Uj =
phic vector field Xj in Uj and if Ui ∩ Uj = φ then Xi Ui ∩Uj = gij Xj Ui ∩Uj φ we have ωi = gij ωj and F Uj is given by ωj = 0. Then we lift {ωj } and

for some non-vanishing holomorphic function gij : Ui ∩ Uj → C. Finally {gij } by π : C3 \ 0 → CP (2) obtaining in this way 1-forms ω j in the open
we assume that sing(Xj ) = φ or consists of a single point pj ∈ Uj . In sets Uj = π −1 (Uj ) ⊂ C3 \ 0 and such on each intersection Ui ∩ Uj = φ we
particular we have the following example: have ω i = gij · ω
j . Notice that if Ui ∩ Uj ∩ Uk = φ then gij · gik = gik
on Ui ∩ Uj ∩ Uk . Thus we have g g g = g   
ik on Ui ∩ Uj ∩ Uk . Finally
∂  ij ij jk
Example 9.1.2 (Polynomial Vector Fields on C2 ). Let X(x, y) = P (x, y) + C3 \ {0} = U  
j so that the data Uj , g ij defines a Multiplicative Cocycle
∂x j
∂ 2
Q(x, y) be a polynomial vector field on C . (P, Q ∈ C[x, y]). For our ∨
∂y on C \ {0}. Because H 1 (C3 \ 0) = 0 and H 2 (C3 \ 0, Z) = 0 (Cartan’s
3
purposes we may assume that P , Q have no common factor on C[x, y] and theorem, [17]) the second (multiplicative) Cousin Problem has a solution
(equivalently) (P = 0) ∩ (Q = 0) is a finite subset of C2 . Thus X has iso- on C3 \ {0} so that there are holomorphic functions gj : U j → C∗ with the
lated singularities on C2 . Let us show that the foliation with singularities gi
induced by X on C2 extends to CP (2). For this we consider the change of property that gij = on each U i ∩ U j = φ ([29]). Therefore we have on
gj
coordinates u = 1/x, v = y/x . We have: each
 i ∩ U
U j = φ : ωi = gi ω 1
j ⇒ ωi =
1
j .
ω
ẋ = P (x, y) gj gi gj
ẏ = Q(x, y)
In this way we can define a holomorphic 1-form ω  on C3 \ {0} by setting
  1
−ẋ 1 v ω 
 U := j . Thanks to Hartogs’ extension theorem ω
ω  extends as a
Therefore u̇ = = −u2 .P , . j gj
x2 u u
holomorphic 1-form on C . Now we consider the pull-back foliation F =
3
Similarly v =yu ⇒ v̇ = ẏu + y u̇.  
1 v v 1 v 1 v π ∗ (F ) (induced by the pull-back of F ) on C3 \ {0}.
Thus v̇ = uQ , + · (−u2 ) · P , . Now we write P , =
u u u u u u u Claim 9.1.6. The 1-form ω  is integrable (i.e., ω
 ∧d
ω = 0) and the foliation
1 
· P (u, v) for a polynomial P (u, v) and some n ∈ N, such that u  P(u, v).  coincides with F.
induced by ω
un
Thus 1
This is quite clear since each ωj (and therefore each j ) is integrable
ω
−u2 −P (u, v) gj
u̇ = n · P (u, v) = · (in dimension two any 1-form is integrable).
u un−2
Similarly we write Now we write ω =ω ν + ω ν+1 + · · · + ω
j + · · · where ω
j is a polynomial
 homogeneous 1-form of degree j (use Taylor/power Series).
1 v 1 
Q , = m Q(u, v) Then d ω = d ων+1 + · · ·+ d
ων + d ωj + · · · and 0 = ω ∧ d
ω = ( ν+1 +
ων + ω
u u u
· · · ) ∧ (d ων+1 + · · · ) = ω
ων + d ν ∧ d
ων + · · · . Therefore ων ∧ dων ≡ 0, i.e.,
 v) ∈ C[u, v]. Then
with m ∈ N and u  Q(u, ν is integrable. The details in the proof of following claim are then lift to
ω
the reader:
 v) uv
u · Q(u,
v̇ = − n P (u, v).  on C3 .
ν defines the same foliation as ω
Claim 9.1.7. The 1-form ω
um u
⎧ ν is homogeneous so that it is radially saturated (if
Proof. Notice that ω
⎪ 
⎨u̇ = −P (u, v)
⎪ p ∈ C3 \ 0 then the leaf of ω
ν = 0 containing p, also contains the line
un−2 {λp; λ ∈ C∗ }).

⎪  v) v P(u, v)
Q(u,
⎩v̇ = −
um−1 un−1
By its turn, the 1-form ω defines F, which is the pull-back of a foliation Assume now that
1
ω ∧ df is polynomial. Then similarly to above, ω ∧ d
1
y
on CP (2). Therefore, ω is also radially saturated. Thus the claim follows: f  y
given p ∈ C3 \ {0} and v ∈ C3 we have ∀ t ∈ C is holomorphic in U and therefore yB( x, y) so that B(x, y) = y · B1 (
x, y)
for some holomorphic B1 ( x, y). In particular B( x, 0) ≡ 0 and therefore X 
 (tp) · v = ω
ω ν (tp) · v + ων+1 (tp) · v + · · ·  ∗ 
y = 0) in U . Thus C ∩ U is F -invariant and by the Identity
is tangent to (
 
= tν ων (p) · v + t
ων+1 (p) · v + · · · . Principle or by Hartogs C is an algebraic leaf of F .

Thus for v ∈ Tp (F) we have Notice that if we consider homogeneous coordinates (x1 ; x2 ; x3 ) on CP (2)
  then as we have seen before, the pull-back foliation F of F to C3 , can
0 = tν ων (p) · v + t
ων+1 (p) · v + · · · , ∀ t be defined by a homogeneous polynomial 1-form Ω(x1 , x2 , x3 ) of degree
ν; this form satisfying Ω · R  ≡ 0 (Ω is radially saturated) where R  =
ν (p)·v = 0 and hence v ∈ Tp ({
so that ω ων = 0}). By comparing dimensions ∂ ∂ ∂
we conclude the proof of the claim. x1 + x2 + x3 is the radial vector field. Given an algebraic
∂x1 ∂x2 ∂x3
curve C ⊂ CP (2) we consider an irreducible homogeneous polynomial
Since ων is homogeneous polynomial it induces a polynomial  1-form f (x1 , x2 , x3 ) such that {f = 0} is the homogeneous equation of C = Z(f ).
ω(x, y) = P (x, y)dy − Q(x, y)dx on C2(x,y) ⊂ CP (2) so that F C2 is given Let k be the degree of the coefficients of the homogeneous 1-form Ω.
(x,y)
by ω(x, y) = 0. This ends the proof of the proposition. Then the above claim rewrites as follows:
Claim 9.2.5. C is an algebraic invariant curve by F if, and only if, there
We shall adopt the following convention: exists a 2-form θ(x1 , x2 , x3 ) such that: (i) df ∧ Ω = f θ (ii) the coefficients
Given an algebraic (irreducible) (not necessarily smooth) curve C ⊂ CP (2) of θ are homogeneous polynomial of degree k − 1.
given on C2 by an affine polynomial equation f (x, y) = 0 we consider its Notice that the conclusion about the degree of (the coefficients of) θ is
lift to C3 \ {0} and then to C3 ; denoted by C  ⊂ C3 . Then C
 is an algebraic
immediate since Ω ∧ df is homogeneous of degree k + deg f − 1 while f θ is
(not necessarily smooth) hypersurface which has an homogeneous equation homogeneous of degree deg f + deg θ.
f(x1 , x2 , x3 ) = 0. We denote by Z(f) the curve C on CP (2) and by (f = 0) Let now
  Ek = θ; θ is 2-form with homogeneous coefficients of degree k−
the hypersurface C. on C3 .
Then E is a finite dimension C-vector space of finite dimension say N (k) =
dim Ek . Assume that the foliation F has N (k)+1 algebraic solutions given
9.2 The theorem of Darboux-Jouanolou by (f0 = 0), . . . , (fN (k) = 0) where fj (x1 , x2 , x3 ) is homogeneous of degree
Given a foliation F on CP (2) we may ask whether has leaf which is a closed k − 1, irreducible and fi , fj are relatively prime if i = j.
dfj
analytic subset of CP (2). In order to study this question shall use: We write ∧ Ω = θj , j = 0, . . . , N (k) as in the claim above. Since
fj
Proposition 9.2.1. Given a foliation F on CP (2) and a leaf L ∈ F the dim Ek = N (k) the set {θ0 , . . . , θN (k) } is linearly dependent. There is!
following are equivalent: N (k) N (k)
dfj
(a0 , . . . , aN (k) ) ∈ CN (k)+1 \}0} such that aj θj = 0 so that aj ∧
(i) L is contained in some algebraic curve C ⊂ CP (2). j=0 j=0 fj
!
(k)
N N (k)
dfj
(ii) L is an algebraic (invariant) curve C ⊂ CP (2). Ω = 0 and then fj α ∧ Ω = 0 for α := aj · Because
j=0 j=0 fj
(iii) L \ L ⊂ sing(F ), i.e., L only accumulate at singular points of F .

(iv) L = (L ∩ sing(F )) ∪ L. cod sing(Ω) ≥ 2 we must have f0 . . . fN · α = gΩ for some homogeneous


polynomial g(x1 , x2 , x3 ) and some N ≤ N (k) (given by the number of non-
The above proposition is a consequence of the above discussion and results. N dfj
We then define a leaf L ∈ F as an algebraic leaf if L = C is an algebraic zero coefficient aj in α). Thus since α = aj is closed, we conclude
fj
curve (which is necessarily invariant by F ). j=0
 
that F is given by a closed rational 1-form. If F admits another alge-
P braic solution (fN (k)+1 = 0) then we write dfN (k)+1 ∧ Ω = fN (k)+1 θ and
Example 9.2.2. Let R = be a rational function on C2 ; P (x, y),
Q because {θ, θ1 , . . . , θN (k) } is linearly independent a similar argumentation
Q(x, y) ∈ C[x, y] have no common factor. shows that for some
Then R defines a foliation F on CP (2) whose leaves are algebraic contained N
dfi(j)
in the algebraic curves aP + bQ = 0, (a, b) ∈ C2 = 0. In particular F has β := bj
fi(j)
infinitely many algebraic leaves. j=1

with fi(1) . . . fi(N  ) β = hΩ where h is a homogeneous polynomial, bj = 0,


Theorem 9.2.3 (Theorem of Darboux-Jouanolou, [39]). If a foliation F
i(j) = 0, j = 1, . . . , N  ≤ N (k).
on CP (2) admits infinitely many algebraic leaves then F admits a rational gfi(1) . . . fi(N  )
first integral. In particular, all leaves are algebraic. Then α = F ·β where F = · Notice that F is not constant
hf0 . . . fN
Proof. Choose a polynomial 1-form ω(x, y) = P (x, y)dy − Q(x, y)dx that because Res(f0 =0) α = α0 = 0 and Res(f0 =0) β = 0. Since α and β are closed
defines F on C2 , with isolated singularities. Given an algebraic curve C ⊂ we have 0 = dF ∧ β and therefore dF ∧ Ω = 0. Thus F is a first integral
CP (2) with C ∩ C2 having irreducible polynomial equation f (x, y) = 0 we for F. The result follows.
put C ∗ = C \ (C ∩ sing(F )).
1 As a corollary of the above argumentation we have:
Claim 9.2.4. C ∗ is an algebraic leaf of F if, and only if, ω ∧ df is a
f
polynomial 2-form on C2 . Corollary 9.2.6. For each k ∈ N there exists N (k) ∈ N such that if
a foliation on CP (2) has more than N (k) algebraic leaves then it has a
Proof of Claim 9.2.4. Assume that C ∗ is invariant by F . Choose a point rational meromorphic first integral.
p ∈ C ∗ and a local chart (x, y) ∈ U  centered at p such that f ( x, y) = y and
 : ( Remark 9.2.7. Joaunolou-Darboux theorem is an algebraic parallel to
C∗ ∩ U y = 0) and write ω( x, y) = B( x, y)d
x − A( x, y)dy where A, B are
∂ ∂ Mattei-Moussu theorem.
holomorphic near (0, 0). The vector field X(  x, y) := A( x, y) +B( x, y)
∂x ∂
y
.
then defines the foliation of U
 : ( 
9.3 Foliations given by closed 1-forms
Since C ∗ ∩ U y = 0) is X-invariant
 we conclude that B( x, 0) ≡ 0.
Therefore we may assume that yB( x, y) in local ring of holomorphic func- Since CP (2) is compact and simply-connected it does not admit a non-
1 1  . Thanks trivial closed 1-form which is holomorphic (indeed, such an 1-form ω would
tions. Therefore ω ∧ df = B( x, y)dx ∧ d y is holomorphic in U
f y be exact ω = dF for some holomorphic function F on CP (2), but then F
1 must be constant and ω ≡ 0 because CP (2) is compact).
to Hartogs’ extension theorem this shows that ω ∧ df is holomorphic in
f Nevertheless there are non-trivial closed meromorphic 1-forms. Since
all points of C ∩ C2 and therefore it is polynomial (we already have a priori any meromorphic function on CP (2) is already a rational function (Liou-
that it is rational). ville’s theorem, [31]) we consider the class of closed rational 1-forms on
CP (2) which we will denote by Ω(CP (2)).
An interesting example is the Poincaré-Dulac normal form (nx+ay n )dy− Now we consider the pull-back Ω = π ∗ (ω) which naturally extends to
ydx = 0 (n ≥ 2, a ∈ C∗ ) that defines a foliation Fa,n on CP (2) that C3 .
is also given by the closed rational 1-form Ωa,n ∈ Ω(CP (2)) defined by 
r
The polar set of Ω is (Ω)∞ = (fj = 0) ⊂ C3 and ∂Ω = 0 with
(nx + ay n )dy − ydx j=1
Ωn = · The poles of Ωa,n in C are given by (y = 0) λj = Res(fj =0) Ω, nj = order of (fj = 0) in (Ω)∞ . Now we introduce the
y n+1
which is the polar set of Ωa,n . This is a general fact as we shall below. An- 1-form
other important example is the class of linear logarithmic foliations (also  r
dfj
   r α := λj ·
r dfj  fj
called Darboux foliations) given by 1-forms as ω = fj · λj j=1
j=1 j=1 fj
where fj ∈ C[x, y], λj ∈ C∗ . In this case the foliation is given by an element This is a closed rational 1-form on C3 such that β := Ω − α is rational,
r dfj closed but with all residues equal to zero. Also (β)∞ ⊂ (Ω)∞ .
Ω= λj ∈ Ω(CP (2)). Next we describe the structure of the elements
j=1 fj Claim 9.3.4. β is exact, i.e., β = df for some meromorphic function f
in Ω(CP (2)). on C3 .
Proposition 9.3.1. Let ω be a closed rational 1-form on CP (2) and let Proof. Indeed, we start by proving given a closed path γ : S 1 → C3 \ (Ω)∞
Ω := π ∗ (ω) be its lift to C3 where π : C3 \ 0 → CP (2) is the canonical then we have β = 0. Let γ : S 1 → C3 \ (Ω)∞ be given and (by approxi-
projection. Then we have γ
 mation theory) assume that γ is of class C ∞ . Since C3 is simply-connected

r
3
 extension F : D → C (D is the closed unit disc |z| ≤ 1
dfj g there is a continuous
Ω= λj + d n1 −1
j=1
fj f1 . . . frnr −1 in C). Such that F S 1 = γ. Again we may assume that:
(i) F is of class C ∞
where
(ii) F (D) avoids the singular set of (Ω)∞ (which is a finite set).
(a) r ≥ 2 and f1 , . . . , fr  g are homogeneous polynomials in C3 ;
(iii) F is transverse to (the smooth part of) (Ω)∞ .
(b) f1 , . . . , fr are irreducible and pairwise relatively prime.
In particular F (D) ∩ (Ω)∞ = {z1 , . . . , zm } is a finite set.
(c) If nj > 1 then fj  g Then by the theorem of residues we have
r 
m
(d) deg(g) = deg(fj ) (nj − 1), i.e., deg(g) = deg(f1n1 −1 . . . frnr −1 ); β= F ∗ (β) = 2πi Reszj F ∗ (β) = 0.
j=1 γ S1 j=1
r
(e) λ1 , . . . , λr ∈ C and λj deg(fj ) = 0.
j=1
In order to conclude that β is exact on CP (2) it is enough to observe that
(f) If nj = 1 then λj = 0. we have already β = df for some meromorphic function f : C3 \ (Ω)∞ → C.
Moreover: Because (β)∞ ⊂ (Ω)∞ the function f is indeed holomorphic in C3 \ (Ω)∞ .

r Once we know that β is rational we can already conclude from df = β that
(g) The polar set of ω is given by (fj = 0) where nj = order of (fj = 0) f admits an extension to C3 as a rational function. Thus we have proved
j=1
that β is the differential of a rational function f on C3 with (f )∞ ⊂ (Ω)∞ .
as a polar curve of ω and λj = Res(fj =0) ω.

Proof. As we have seen, ω cannot be holomorphic, so that its polar set Thus we can write
(ω)∞ is not empty. Because (ω)∞ has codimension one it can be written
r 
r
dfj
as (ω)∞ = (fj = 0) where f1 , . . . , fr are irreducible polynomials in C3 , Ω= λj + df on C3
j=1 j=1
fj
pairwise relatively prime.
Let λj = Res(fj =0) ω ∈ C be the residue of ω in (fj = 0) =: Z(fj ) ⊂ CP (2) or else
and nj = order of Z(fj ) as pole of ω. 
r
dfj g
Ω= λj +d ·
fj h
Remark 9.3.2. λj can be calculated/defined as follows. Choose a point j=1
pj ∈ Z(fj ) which is not a singular point of the variety Z(fj ). Take a
transverse !
disc Zpj centered at pj , transverse to Z(fj ) and such that 9.4 Riccati foliations

r
Z(fi ) ∩ Σpj = {pj }. The classical Riccati differential equation, put in terms of complex vari-
j=1
ables, is 
Choose a small simple loop γj ⊂ Σpj \ {pj } positively oriented. ẋ = p(x)
Then " " ẏ = a(x)y 2 + b(x)y + c(x).
1 1 
λj := √ ω= √ ω Σp
2π −1 γj 2π −1 γj j We will consider the case where the coefficients are polynomials
p(x), a(x), b(x), c(x) ∈ C[x]. The appropriate space to study the geome-
By a classical result of Deligne [21] λj is well-defined (recall that Z(fj ) is
try of a Riccati foliation is perhaps the surface C × C = M . Let us see why.
irreducible). Now we claim:
First we recall the canonical coordinate changes in M .
r
Claim 9.3.3. λj deg(fj ) = 0. On M we have the natural projection π1 : M → C given by π1 (p1 , p2 ) = p1 .
j=1 The fibers π1−1 (p) = {p} × C are Riemann spheres. Now we consider a Ric-
cati foliation F of M obtained by extending from C2(x,y) to M the foliation
Proof. We consider a linear embedding E : CP (1) → CP (2) such that the

line E(CP (1)) =: L intersects (ω)∞ only at non-singular points of the induced by the polynomial vector field X(x, y) = p(x) +(a(x)y 2 +b(x)y+
∂x
variety (ω)∞ and the intersection is always transverse.  ∂
We consider the restriction, i.e., the induced 1-form ξ := ω L = E ∗ (ω). c(x)) · We claim:
∂y
Then ξ has polar at the points that correspond to the intersection points
L ∩ (ω)∞ . Moreover, since L induces a transverse disc (like the discs Σpj ) Claim 9.4.1. A vertical fiber π1−1 (p1 ) p1 = ∞ is invariant by F if, and
at each intersection points p ∈ L ∩ (ω)∞ we have that if p ∈ L ∩ Z(fj ) only if, p1 is a zero p(x).
then Resp ξ = λj . By its turn, Bezout’s theorem implies that the number 
of intersection points p ∈ L ∩ Z(fj ) is equal to deg(L) deg(fj ) = deg(fj ). 2 ẋ = p(x)
This is quite clear since F is given on C(x,y) by
Finally, because L is a Riemann Sphere the theorem of residues applied to ẏ = a(x)y 2 + b(x)y + c(x)
ξ says that and π1 (x, y) = x. Therefore F has a finite number of invariant vertical
 r
fibers, that depends on the degree if p(x). Let us change coordinates on F :
0= Res(ξ, p) = λj deg(fj ).
p∈L∩(ω)∞ p=1
As a corollary we obtain:

ẏ a(x) b(x) Proposition 9.4.6. Let F be a Riccati foliation on C×C and let F1 , . . . , Fr ⊂
Ẏ = − 2 = −Y 2 + + c(x)
y Y2 Y C × C be the invariant (vertical) fibers of F . Then F0 = F  r is
 (C×C)\ Fj
Ẏ = −(c(x)Y 2 + b(x)Y + a(x))
j=1

r
ẋ = p(x) a foliation transverse to the fibers of the fiber space ξ : (C × C) \ Fj →
j=1

r
Thus once again F is given by a Riccati ordinary differential equation in C\ Fj → C (Fj = pj × C) and in particular F0 is holomorphically
the C2(x,Y ) coordinate system. Therefore, we can claim: j=1
conjugate to the suspension of a finitely generated group of Möebius maps.
Claim 9.4.2. Given a non-invariant vertical fiber F = π1−1 (p1 ) the leaves
of F are transverse to F .
Proof. Indeed, we may assume that p1 = ∞ and therefore the non-invariance
of F is equivalent to p(x) does not vanish at x = p1 .
Since the expressions of F in the coordinate systems (x, y) and (x, Y )
are  
ẋ = p(x) ẋ = p(x)
ẏ = . . . Ẏ = . . .
we conclude that F is always transverse to F .
Actually the above transversality is a characterization of Riccati folia-
tions.
Claim 9.4.3. A foliation F on C × C = M which is transverse to some
vertical fiber π1−1 (p1 ) = F is a Riccati foliation.

Proof. We may suppose that p1 = x1 = ∞ and choose a polynomial vector


ẋ = A(x, y)
field that defines F on C2(x,y) .
ẏ = B(x, y)
By the transversality on C2(x,y) we conclude that A(x, y) = 0, ∀ y ∈ C.
Since A is a polynomial this implies that A(x, y) depends only on x , not
on y. Now, since the fiber F ∼ = C is compact, the foliation is also transverse
to the nearby fibers π1−1 ( 1 ≈ x1 . By the same reasoning above we
x1 ) for x
conclude that A( x1 , y) = A(x1 ) does not depend on y, for every x 1 ≈ x1 .
Because A(x, y) is a polynomial this  implies that A(x, y) = A(x) depends
 ẋ = A(x)
only on x. Thus F C2 is given by .
ẏ = B(x, y)

Now we change coordinates to (x, Y ) = (x, y1 ). Then Ẏ = − ẏ 2 1


y 2 = −Y (B(x, Y ))

−Y B(x, Y )
and then Ẏ =  Y ) is a polynomial, Y  B(x,
where B(x,  Y ) and
Yn ⎧
⎨ẋ = A(x)
n = degy B. Since F is given by  and by the same ar-
⎩Ẏ = − B(x, Y )
Y n−2
guments above, the transversality of F with F at the point implies that Chapter 10
n − 2 ≤ 0, i.e., n ≤ 2.
Therefore B(x, y) = b0 (x)+b1 (x)y+b2 (x)y 2 and F is a Riccati foliation.
From the structural point of view, Riccati foliations are related to sus-
pensions of groups of automorphisms of CP (1). For this let us recall the
Foliations with algebraic
classical concept of Ehresmann.
F
Definition 9.4.4. Given a fiber bundle space ξ(π : −→ B) with basis B,
limit sets
fiber F , total space E and projection π; a foliation F on E is said to be
transverse to the fibers of ξ if:
(i) dim F + dim F = dim E; 10.1 Limit sets of foliations
(ii) F is transverse to the fibers π −1 (b) ⊂ E; Let F be a holomorphic foliation with singularities on a compact complex

(iii) Given any leaf L of F the restriction π L : L → B is a (surjective) manifold M . Given a  leat L ∈ F we consider an exhaustion by compact
covering map. subsets of L, i.e., L = Kj , where each Kj ⊂ L is a compact subset and
j∈N

Recall that in this case we have a natural action of π1 (B) on Diff(F ) Kj ⊂ Int(Kj+1 ) for all j ∈ N.
given a base point b0 ∈ B and a path γ ∈ π1 (B, b0 ). We define a map Definition 10.1.1. The limit set of the leaf L is defined as lim(L) =
hγ : Fb0 → Fb0 as follows: given y ∈ Fb0 = π −1 (b) we  
 consider the lifted (L \ Kj ). The limit set of foliation F is lim(F ) = lim L.
path γy (t) ⊂ Ly obtained from the covering map π Ly : Ly → B (where j∈N L∈F
y ∈ Ly is a leaf of F ).
Remark 10.1.2. This notion is clearly motivated by the theory of real
Then we put hγ (y) := γy (1); the final point of the lifting.
Dynamical Systems and also by the dynamics of groups of rational maps
The image of this group homomorphism ϕ : π1 (B, b0 ) → Diff(F ) is called
on the Riemann sphere.
the global holonomy of F in ξ. Is is well-known that F is conjugate to
the suspension of its global holonomy. One important fact is the following The very basic properties of the limit set of a foliation are listed below:
remark by Ehresmann:
Proposition 10.1.3. Let F and M (compact) be as above, then:
F
Proposition 9.4.5 ([9, 26]). Let F be a foliation on the fiber space ξ(π : E −→
(i) lim(F ) ⊂ M is F -invariant
B). Assume that (i) dim F + dim F = dim E and (ii) F is transverse to
the fibers π −1 (b) ⊂ E. Then F is transverse to the fibers of ξ if the fiber F (ii) sing(F ) ⊂ lim(F )
is compact.
(iii) If dim M = 2 then for a leaf L ∈ F we have lim(L) ⊂ sing(F ) ⇔ L ⊂ 10.2 Groups of germs of diffeomorphisms with
M is an analytic curve.
finite limit set
(iv) If M = CP (2) then lim(F ) ⊂ sing(F ) iff F has a rational first inte-
gral. We shall study the dynamics of subgroups of Diff(C, 0) which may be as-
sociate to the holonomy groups of foliations with analytic/algebraic limit
Proof. Let us prove (iii) since (i) and (ii) are more immediate. Assume sets. We shall start with some very basic facts:
that dim M = 2 and that L ∈ F satisfies lim(L) ⊂ sing(F ). We claim
that L is an analytic curve in M . Indeed given a point p ∈ L \ L then Lemma 10.2.1 (Poincaré linearization lemma). Let f ∈ Diff(C, 0) be such
necessarily we have p ∈ sing(F ) (because lim(L) ⊂ sing(F )). Therefore by that |f  (0)| = 1. Then:
Remmert-Stein extension theorem L ⊂ M is analytic of dimension one. The (i) f is analytically linearizable: ∃ φ ∈ Diff(C, 0) such that φ ◦ f (z) =
converse of (iii) is clear. Let us now prove (iv). Assume that M = CP (2) f  (0).φ(z).
and lim(F ) ⊂ sing(F ). Then from (iii) every leaf L of F is contained in
an analytic curve which is F -invariant. By Chow’s theorem every leaf of (ii) If ψ ∈ Diff(C, 0) is any map that linearizes f (i.e., ψ ◦ f (x) =
F is algebraic. By Darboux-Joaunolou theorem F admits a rational first f  (0).ψ(z)) then φ ◦ ψ −1 is linear (i.e., φ = μψ for some μ ∈ C∗ ).
integral.
Proof. (i) is the well-known linearization theorem of Poincaré. Now we
claim:
Next we give some examples of limit sets of foliations.
Claim 10.2.2. Let g ∈ Diff(C, 0) commuting with f , i.e., f ◦ g = g ◦ f .
 We consider a linear foliation F on
Example 10.1.4 (Linear foliations). Then g is linear in any coordinate that linearizes f .
ẋ = λx Proof. Indeed, write f (z) = λz with |λ| = 1. In particular λn = 1, ∀ n = 0.
CP (2), given in an affine chart by , λ, μ ∈ C \ {0}. Then: ∞
ẏ = μy
Write g(z) = gn z n . Since g ◦ f = f ◦ g we conclude that
n=1
(i) λ/μ ∈ Q ⇒ the leaves are all algebraic and we have lim(F ) = sing(F ).
λ.gn = λn .gn , ∀ n ∈ N.
(ii) λ/μ ∈ R \ Q ⇒ lim(F ) is not algebraic, indeed for each leaf L ∈
F we have lim(L) = ML3 where ML3 ⊂ CP (2) is the singular real Since λn = λ, ∀ n = 1 we get gn = 0, ∀ n = 1 and therefore g(z) = g, z.
variety
 of dimension 3 given by |x|μ |y|−λ = c ∈ R (we assume that
λ, μ ∈ R If f1 , f2 ∈ Diff(C, 0) are such that fj−1 f fj = λz j = 1, 2 then putting
λ/μ ∈/Q
) for a certain constant c > 0. g = f1 f2−1 we conclude that g ◦ f = f ◦ g so that g(z) = μz and therefore
f1 = μf2 proving (ii).
(iii) λ/μ ∈ C \ R ⇒ lim(F ) is algebraic, it is the union of three projective
lines: the compactification of the axes (x = 0) and (y = 0), and the
The above proof actually gives:
line at infinity CP (1)∞ = CP (2) \ C2 .
Lemma 10.2.3. If f (z) = λz is linear and g ∈ Diff(C, 0) is such that
Another important property of the limit set is:
f ◦ g = g ◦ f , then we have:
Lemma 10.1.5. Let π : M  → M be a proper holomorphic map, F a fo-
(i) λn = 1 ∀ n ≤ 0 ⇒ g(z) = μ(z) (is also linear)
liation of M which is generically transverse to π (meaning that the set of
tangent points of F with π has codimension ≥ 2 in M ). Then for a leaf

L ∈ F if we denote by L  the inverse image L = π −1 (L) we have that (ii) λk = 1 for some k ∈ N ⇒ g(z) = μz(1 + u(z k )) for some holomorphic
 is a finite union of leaves of the pull-back foliation F := π ∗ (F ); say
L function u(z) with u(0) = 1.
F = L
1 ∪ · · · ∪ L
 r . Moreover we have (since π is proper)
We also recall the following (already discussed) result (see Chapter 8):

r
π −1
(lim(l)) =  j ).
lim(L Theorem 10.2.4. Let f ∈ Diff(C, 0) be of the form f (z) = z +ak+1 z k+1 +
j=1 . . . , ak+1 = 0. Then f is topologically conjugate to the diffeomorphism
z
In particular, we have f(z) = is a neighborhood of the origin.
(1 + ak+1 z k )1/k
lim F ⊂ π −1 (lim F ).
In particular we have:
Remark 10.1.6. For the proof, the essential point is that given an ex-
haustion {Kj }j∈N of L by compact subsets, since π is proper, the collection (1) For every point, close enough to the origin, its orbit is contained in
 j = π −1 (Kj )}j∈N defines a compact exhaustion of L := π −1 (L).
{K an invariant by f continuous curve that passes through the origin.
As a consequence of this and of the linear case in the above example we (2) For every point z, close enough to the origin, f n (z) or f −n (z) con-
obtain: verges to the origin as n → +∞.
Example 10.1.7. A rational pull-back π ∗ (F ) = F to CP (2) of a linear
Definition 10.2.5. Let G ⊂ Diff(C, 0) be a subgroup. Given a connected
hyperbolic foliation F : xdy − λydx = 0 (λ ∈ C \ R) (by a rational map
neighborhood V of 0 in C and z ∈ V , the (pseudo-orbit of z by G is
π : CP (2) → CP (2)) is a foliation with an algebraic limit set of dimension
defined as O(z) = {f (z); f is the representative of some element of G and
one.
z ∈ Dom(f )}. Given z ∈ V \{0} we say that the pseudo-orbit of z is discrete
Another example is given by off the originpseudo-orbit! discrete off the origin if O(z) \ O(z) ⊂ {0}. If
Example 10.1.8 (Riccati foliations). Given a Riccati foliation R on C× C this is true for all z ∈ V \ {0} then we say that G has discrete pseudo-orbits
we know that its dynamics is strongly related to the dynamics of a finitely off the origin in V .
generated group of Möebius transformations. In particular, by choosing A remarkable theorem of Isao Nakai implies the following:
suitable subgroups of SL(2, C) we can obtain Riccati foliations on C×C with
global holonomy a group G ⊂ SL(2, C) with one or two fixed points in the Theorem 10.2.6 (I. Nakai 1994, [52]). Let G ⊂ Diff(C, 0) be a non-
Riemann sphere. This foliation will have an algebraic limit set, consisting solvable finitely generated subgroup. Then there is a fundamental system of
of one or two curves. Any rational map π : CP (2) → C × C then induces neighborhoods 0 ∈ V ⊂ C such that on each V the group has no non-trivial
a following π ∗ (F ) on CP (2) with algebraic limit set. For instance, let us orbit closed or discrete off the origin.
take any finitely generated group of Möebius transformations G ⊂ SL(2, C).
Assume that the limit set of G on CP (1) is a single point, which can be A particular case of this is proved below:
assumed to be the origin 0 ∈ C. The limit point 0 is a fixed point of G. Proposition 10.2.7 (key proposition, [11]). Let G ⊂ Diff(C, 0) be a sub-
According to [42] we can find a Riccati foliation F : p(x)dy − (a(x)y 2 + group such that:
b(x)y + c(x))dx = 0 on C × C, whose holonomy group of the line (y = 0) is
conjugated to the group G. Moreover we can assume that the singularities (1) ∃ f ∈ G with |f  (0)| < 1.
of F over this horizontal line are reduced and non degenerate. The line
(y = 0) is invariant by F so that c(x) = 0, and also it is contained in the (2) ∃ neighborhood 0 ∈ V ⊂ C such that the orbits of G in V are discrete
limit set of F . This example can also be seen in CP (2) using a birational off the origin. Then G is abelian.
transformation. This will create a dicritical singularity for the foliation.
Proof. We may f (z) = λz (|λ| < 1) in some local coordinate; in some sub- (M, Λ) the reduction of singularities of F in Λ and let π ∗ F = F. Assume
neighborhood of 0 in V . Suppose that G is not abelian. Then some map that:
g ∈ G does not commute with f : otherwise G is linearizable and therefore
abelian. (1) D is invariant and sing(F) ⊂ D contains no saddle-node singularity.
Thus ∃ h ∈ G, h = [f, g] = Id. The dynamics of h is such that ∀ z ≈ 0 we (2) Each component Dj ⊂ D has abelian virtual holonomy group and
have hn (z) → 0 or h−n (z) → 0 as n → +∞. Let now A ⊂ C a fundamental contains an attractor.
domain for the attractor f , i.e., A = D \ f (D) where 0 ∈ D is a small disc
centered at the origin. Notice that for any z = 0, z ≈ 0 ∃ n ∈ Z such that (3) D has no cycles.
f n (z) ∈ A.
Then there exists a neighborhood V of D in M  where F is defined by a
Claim 10.2.8. ∃ a non-discrete orbit in A. closed meromorphic 1-form ω ω )∞ ⊃ D.
 , with simple poles and (
Proof. Choose a compact disc 0 ∈ K ⊂ D so that K ∩ A = φ. For each Proof of Proposition 10.4.1. We observe that D = D0 ∪D1 ∪· · ·∪Dr where
z ∈ A there is a minimal m1 (z) ∈ Z such that hm1 (z) ∈ K. There is .
D0 is the strict transform of Λ, i.e., d0 = π −1 (Λ \ sing(F )) ⊂ M
also a minimal positive number n1 (z) ∈ N such that f −n1 ◦ hm1 (z) ∈ A. Let us first consider the case of a component Dj ⊂ D.
Proceeding in this way we get a sequence of points {zr } ⊂ C \ {0} of the
form zr = f −nr ◦ hmr ◦ · · · ◦ f −n1 ◦ h−m1 (z) ∈ A such that hmr ◦ · · · ◦ f −n1 ◦ Lemma 10.4.2. Given a component Dj ⊂ D there exists a closed mero-
h−m1 (z) ∈ K, ∀ r ∈ N. Given two sequences of numbers m = {mj }j=1
r morphic 1-form ωj , with simple poles, defined in a neighborhood Uj of Dj ,

r
n = {nj }j=1 as above we define the set such that F is given by ωj = 0 off (ωj )∞ . The 1-form ωj is uniquely
Uj
determined by the condition: given q ∈ Dj \ sing F, Σ q transverse disc,
Vm,n := {z ∈ A; f −nr ◦ hmr ◦ · · · ◦ f −n1 ◦ h−m1 (z) = z}. Σ∩Dj = qj and a holomorphic coordinate z in Σ (z(q) = 0) that linearizes
 dz
Then Vm,n is a finite set: otherwise (since A is compact) we should have an the virtual holonomy, then ωj Σ = ·
z
accumulation point in A and then f −nr ◦ hmr ◦ · · ·◦ f −n1 ◦ h−m1 (z) = z, ∀ z.
On the other hand thederivative (f −nr ◦ hmm ◦ · · · ◦ f −n −m1  Proof of Lemma 10.4.2. Given a point p ∈ Dj \ sing F, choose a holo-
 ◦h
1
) (0) = 1,
contradiction. Thus Vm,n is countable so that A \ Vm,n = φ and we morphic chart φ = (x, u) : U → φ(U ) ⊂ C2 with p ∈ U , φ(p) = (0, 0),
m,n m,n φ(U ) = {(x, y) : |x| < 2, |y| < 2} and:
have some non-discrete orbit in A.

(1) FU is given by dy = 0;
The claim completes the proof of the proposition.
(2) Dj ∩ U ⊂ (y = 0)
As a consequence of the above results:

Proposition 10.2.9. Let G ⊂ Diff(C, 0) be a subgroup such that: (3) Σ : (x = 0) is transverse disc to F and g Σ a local chart that linearizes
Holvirt (F, Dj , Σ).
(i) G contains an attractor.
Remark 10.4.3. The existence of φ : U → φ(U ) is obtained by extending
(ii) The pseudo-orbits of G are discrete off the origin. a local transverse coordinate that linearizes the virtual holonomy from the
transverse section to a neighborhood (of bidisc type) as constant along local
Then G is abelian and linearizable.
plaques of F.

10.3 Virtual holonomy groups We obtain then an open cover

The virtual holonomy group to be introduced below is the geometric object U = {(Uα ), (xα , yα ) : Uα → C2 }α∈√α
that measures the accumulations of the leaves around a given leaf. Let us
be more precise: of Dj \ sing(F) satisfying (1), (2) and (3) above. We may also assume that
Let F be a foliation of a complex surface M , L ∈ F a leaf; q ∈ L a base
point (q ∈
/ sing(F )) and Σ a transverse disc through q ∈ Σ ∩ L. (4) If Uα ∩ Uβ = φ then Uα ∩ Uβ is connected.
We consider the holonomy group Hol(F , L, Σ, q) → Diff(Σ, q) and intro- Let us the study this situation Uα ∩ Uβ = φ.
duce the virtual holonomy group as follows ([11]):
Claim 10.4.4. We have yα = cαβ · yβ for some constant cα,β ∈ C∗ .
Holvirt (F , L, Σ, q) := {f ∈ Diff(Σ, q); Lz = Lf (z) , ∀ z ∈ Σ}
The claim is an easy consequence of (3) and the fact that Holvirt (F , Lj , Σ, q)
In the above notation Ly is the leaf of F (in M \ sing(F )) that contains contains an attractor, where Lj = Dj \ sing(F).
y ∈ M \ sing(F ). dyα yβ
We then conclude that the closed meromorphic 1-forms and coin-
Then clearly Hol(F , L, Σ, q) ⊂ Holvirt (F , L, Σ, q). Then main result of this yα yβ
chapter is the following: cide on Uα ∩ Uβ . 
This gives a closed meromorphic 1-form ωj in Vj = Uα , defined by
Theorem 10.3.1 (Linearization theorem, Camacho-Lins Neto-Sad [11]). α∈β
Let F be a holomorphic foliation on CP (2). Assume that the limit set of  dyα
lim(F ) is algebraic of codimension one and contains an irreducible compo- ωj Uα = ·

nent Λ ⊂ CP (2) of dimension one such that:
Claim 10.4.5. Given any singular point pj ∈ sing F ∩ Dj the 1-form ωj
(1) sing(F ) ∩ Λ is non-dicritical and contains no saddle-node in its re- .
extends meromorphic to a neighborhood of pj in M
duction of singularities.
Proof of Claim 10.4.5. The point pj is not a saddle-node by hypothesis.
(2) Some component in the reduction of singularities of Λ contains an
The local holonomy of the separatrix pj ∈ Γj of F contained in Dj is an-
attractor in its virtual holonomy group.
alytically linearizable because it is in the virtual holonomy. Therefore, by
Then there is a rational map φ : CP (2) → CP (2) and there is a linear Mattei-Moussu Theorem 7.1.3 we conclude that F is analytically lineariz-
foliation Lλ : xdy − λydx = 0, λ ∈ (C \ R) ∪ (R− \ Q) such that F is the able in a neighborhood of pj . Let then (x, y) : U → C2 be a local chart

pull-back F = φ∗ (Lλ ) of L
 λ by φ. such that pj ∈ Uj , x(pj ) = y(pj ) = 0, Dj ∩ U ⊂ (y = 0) and FU is given

by xdy − λydx = 0, λ ∈ C (indeed λ ∈ C \ Q+ because pj is irreducible).
The local holonomy of Γj at the transverse disc Σj : (x = 1) is given by
10.4 Construction of closed meromorphic forms dy λdx
h(y) = e2πiλ y. We set ωpj := − in U .
y x
We shall now see how to construct closed meromorphic 1-forms defining a Since ωpj and ωj define F in Vj ∩ U (Vj ∩ U contains a neighborhood of the
foliation, based on information an the virtual holonomy and on the singu- loop γ ⊂ Γj given by γ = {(x, 0); |x| = 1}) we have ωj = f · ωpj for some
larities. This is done in a neighborhood of a compact analytic invariant meromorphic function f in Vj ∩ U . Since
divisor. Given ε > 0 let Vε := {(x, y) : 1 − ε < |x| < 1 + ε; |y| < ε} and Vε ⊂ Vj ∩ U
Proposition 10.4.1. Let F be a foliation on M 2 with sing(F ) ⊂ Λ ⊂ M , for ε > 0 small enough. Since ωj and ωpj have simple poles in V we have
, D) →
where Λ is an analytic (compact) invariant curve. Denote by π : (M
that f is holomorphic in V and therefore it can be represented (in Vε and Let us proceed with the proof of Proposition 10.4.1.
therefore) in V by a Laurent Series like f (x, y) = hij xi y j . Since D has no cycles we can construct a closed 1-form ω in a neighborhood
i∈Z,j≥0 r 
Since ωj = f ωpj and dωj = 0 = dωpj we obtain df ∧ ωpj = 0. This last Uj of D by choosing ω D = cj · ωj for some suitable choice of the
j
j=0
relation can be rewritten as constants cj · ωj for some suitable choice of the constants cj ∈ C∗ , j =
0, . . . , r.
(∗) x fx + λy fy = 0
For the case of foliations on CP (2) we obtain:
and in terms of the Laurent Series of f as
Proposition 10.4.8. Let F be a foliation on CP (2) with an algebraic
(∗∗) (i + λj )fij = 0, ∀ i ∈ Z, j ≥ 0. invariant curve Λ ⊂ CP (2), having a reduction of the singularities for
sing(F ) ∩ Λ, π : (M, D) → (CP (2), Λ) as follows:
Case 1: If λ ∈ / Q then fij = 0 ∀(i, j) = (0, 0) and therefore f is constant.
Thus ωj = c.ωpj in the common domain, for some c ∈ C∗ . Comparing (1) Sing F ∩ D contains no saddle-node and is non-dicritical.
residues along (y = 0) we obtain c = 1 and therefore ωj = ωpj in Vj ∩ U .
In particular, ωj extends as ωpj to I pj . (2) Each irreducible component Dj of D has abelian virtual holonomy
m containing an attractor. Then F is given by a logarithmic (rational)
Case 2: If λ ∈ Q− say, λ = − with n, m ∈ N m, n = 1. Then (**)
n one form on CP (2).
implies that ni − mk = 0 if fij = 0. That is, fij = 0 ⇒ (i, j) = (km, kn)
for some k ≥ 0 (notice that j ≥ 0). Therefore f (x, y) = (x y  ) for some Proof. The point here is that D = D0 ∪· · ·∪Dr may have cycles. Neverthe-
holomorphic function (z) and then f admits a holomorphic extension to a less we have another argument as follows: Since F is defined on CP (2) there
neighborhood of pj . The 1-form ωj then extends to a neighborhood of pj is a rational 1-form Ω that defines F and we consider Ω  = π ∗ (Ω) which is
as f.ωpj . This ends the proof of the claim. 
a rational 1-form defining F on M . By the preceding proposition for each
j = 0, . . . , r there is a closed meromorphic 1-form ωj in a neighborhood Uj
This and the fact that two coordinates linearizing an attractor differ up to 
of Dj in M such that FUj is defined by ωj and if Uij = Ui ∩ Uj = φ then
a multiplication imply Lemma 10.4.2.
in Uij we have ωi = cij ωj for some cij ∈ C∗ . Since Ω  defines F, on each
Now we pass from (each) Dj to a neighborhood of D = D0 ∪· · ·∪Dr . Let  = hi ωj for some meromorphic function hj in Uj . Thus in
Uj we have Ω
then ωj a closed meromorphic 1-form with simple poles in a neighborhood hj hj
, such that ωj = 0 defines F in Uj \ (ωj )∞ .
Uj of Dj (j = 0, 1, . . . , r) in M Uij = φ we have hi ωi = hj ωj ⇒ ωi = ωj so that = cij and then
hi hi
Assume that Di ∩ Dj = φ say q = Di ∩ Dj . Then dhi dhj r  dhj
= · Define then η in U≡ Uj as ηUj := · There is a closed
in a neighborhood Uij ⊂ Ui ∩ Uj we have ωi = f ωj for some meromorphic hi hj j=0 hj
function f .  ) such that η = π (η).

meromorphic 1-form η in U = π(U
By Levi’s extension theorem (Theorem 2.2.4) the 1-form η extends as a
Claim 10.4.6. f is constant.
closed rational one form on CP (2).
Proof of Claim 10.4.6. Since Di and Dj are F-invariant the corner q is
 We have seen that is analytically linearizable say, dh
a singularity of F. Claim 10.4.9. η = for some rational function h on CP (2).
h
FU : xdy − λydx = 0, in some neighborhood U (x, y) of q, with Dj ⊂
(y = 0), Di ⊂ (x = 0). Proof of Claim 10.4.9. By the description of the closed rational 1-forms on
We have df ∧ (xdy − λydx) ≡ 0. CP (2) (Proposition 9.3.1) it is enough to observe that:

If λ ∈/ Q then we have seen that f is constant. Assume now that λ = (1) η has simple poles on CP (2)
−m/n ∈ Q− , m, n ∈ N, m, n = 1. The virtual holonomy of Dj is lin-
 (2) for each component H of the polar set of η we have ResH η ∈ Z.
earized by y Σj . We may choose (x, y) ∈ U such that also Holvirt (F, Di , Σi )
is linear in the coordinate x → (x, 1) ∈ Σi : (y = 1), let us see why: dhj
The proof of (1) and (2) is a consequence of the local description as
Indeed, by hypothesis the virtual holonomy of Di contains an attractor hj
say g, with g  (0) = μ, |μ| < 1. The local holonomy of the separatrix of η in a neighborhood of D = π −1 (Λ) and of Bézout’s theorem (every
n
q ∈ Γi ⊂ Di is given by h(x) = e−1πi m x. Since g and h commute we have component of (η)∞ must intersect Λ). This proves Claim 10.4.9.
g(x) = μx g(x ) for some g ∈ O1 with 
m
g(0) = 1. Let now x  = φ(x) be a
dh
change of coordinates valid at 0 ∈ Σi , such that φ ◦ g ◦ φ−1 is linear (recall Finally, let 
h = π ∗ (h) = h ◦ π. Then, η = and 
h = const. hj so that

that Holvirt (F, Di , Σi , q) is abelian linearizable). Then φ(g(x)) = μφ(x). !  h

Ω Ω
Since φ linearizes g and it also linearizes the local holonomy h of Γi we d = 0. Therefore d = 0.
conclude that φ(x) = x.φ1 (xm ) where φ1 ∈ O1 , φ1 (0) = 0. We then con- 
h h
sider  m y n ), y). Then
x, y) = ψ(x, y) = (xφ(x Ω
 the change of coordinates ( Thus ω := is a closed rational 1-form which defines F on CP (2). We
ψ Σi ≡ φ so that ψ preserves the linear foliation mydx + nxdy = 0. In- h
claim (as it is easy to see) that ω has simple poles so that ω is logarithmic
deed, ψ ∗ (mydx + nxdy) = u.(m yd
x + n xdy ) for some u holomorphic with
on CP (2). This ends the proof of Proposition 10.4.8.
u(0) = 0.
Therefore we may assume that g is linear in the coordinate (x, y).
Now given the linear singularity q : mydx + nxdy = 0 10.5 The Linearization theorem
We claim:
Claim 10.4.7. Define k : Σj → Σj by k(y) := μ1 y where μm n Now we proceed to prove Theorem 10.3.1, the main result of this chapter.
1 = μ . Then
k ∈ Holvirt (F, Dj , Σj ) and it is an attractor. Proof of Theorem 10.3.1. For the first part it is enough to prove that every
component Dj of the exceptional divisor D in the reduction of singularities
Proof of Claim 10.4.7. We consider the first integral ξ = xm y n for F in a π : (M, D) → (CP (2), Λ) of sing(F ) ∪ Λ, exhibits a hyperbolic attractor or
neighborhood of q. Then k(y) preserves the leaves of FΣj which are given
a non periodic linearizable map in its virtual holonomy group. We consider
by g n/m = constant. a component Dj0 which has a hyperbolic virtual holonomy map say hj0 .
m/n
It is enough to observe that (k(y)) = y m/n . Then if Di is an adjacent component, Di ∩Dj0 = φ we have two possibilities
for the corner singularity q = Di ∩ Dj0
The above implies that Holvirt (F, Dj , Σj ) is linear in the coordinate
y → (1, y) of Σj (because this coordinate linearizes the attractor k(y)). (1) q : xdy − λydx = 0, λ ∈ / Q, Di : (y = 0), Dj0 : (x = 0).
But from the characterization/definition/construction of ωi we have In this case the virtual holonomy (actually the holonomy) of the com-
ponent Di contains a linearizable non-resonant map hi : y → e2πiλ y.
 dx dx n dy
ωi Σi = and ωi = +
x x m y (2) q : xdy − λydx = 0, λ = −m/n ∈ Q− , m, n = 1.
In this case we have seen in the above proofs that the first integral
and analogously
dy m dx xm y n permits the “passage” of the attractor hj0 to an attractor k in
ωj = + the virtual holonomy of Di .
y n x
n
so that ωi = ωj . This proves Claim 10.4.6 in the resonant case.
m
  dfj
Thus F is given by a logarithmic 1-form ω C2 = λj where λj ∈
j=1 fj


C, fj is an irreducible polynomial, (ω)∞ = Γj where Γj = (fj = 0),
j=1
2
for a suitable choice of the coordinate system C ⊂ CP (2) with the line
CP (2) \ C2 not invariant by F .
We then know that

dj λj = 0. where dj = deg(fj ). We sketch the
Chapter 11
j=1
main steps. Let Γ1 be the component of (ω)∞ with a virtual holonomy
attractor.
Claim 10.5.1. Fix a point p1 ∈ Γ1 \ sing(F ), Σ1 p1 a transverse disc
Some modern questions
and z ∈ (Σ1 , p1 ), z(p1 ) = 0 a local coordinate such that Holvirt (F , Γ1 , Σ1 )
is linear in the coordinate z. Then for each j ≥ 2 the map hj (z) =

e2π −1
λj /λ1 .z belongs to Holvirt (F , Γ1 , Σ1 ). We introduce and comment now some modern questions in the theory of
holomorphic foliations with singularities.
We also observe that G = Holvirt (F , Γ1 , Σ1 ) is abelian, contains an
attractor, linearizable and has discrete orbits off the origin. Thus G is
2πi
generated by an attractor z → e2πiλ z and a rational rotation z → e m z. 11.1 Holomorphic flows on Stein spaces
From the above claim we then conclude that for each j ≥ 2 ∃ kj , j ∈ Z
such that We shall now discuss several aspects of the dynamics, topology and analytic
λj kj classification of the foliation with singularities defined by an action of the
= + j λ.
λ1 m groups C or C∗ on a Stein variety usually under the presence of a (singular)
Therefore m
λj
= vj − uj λ for some uj , vj ∈ Z. We define fixed point of the action. We consider an action ϕ : C × N → N , i.e., a
λ1 holomorphic map such that:

F1 = f1m f2v2 . . . fv (i) ϕ(t, ϕ(t2 , x)) = ϕ(t1 + t2 , x), ∀ t1 , t2 ∈ C, ∀ x ∈ N
F2 = f2v2 . . . fu
(ii) ϕ(0, x) = x, ∀ x ∈ C.
then we obtain
The action is periodic of period τ ∈ C∗ if ϕ(τ, x) = x, ∀ x ∈ N.
dF1 dF2 df1 
dfj
−λ =m = (vj − λuj ) The action induces a group of homomorphism
F1 F2 f1 j=2
fj
 C → Aut(N ) = {group of holomorphic diffeomorphisms of N }
df1  λj dfj m 
=m + = · ω. t → ϕt := ϕ {t} × N : N → N
f1 λ1 fj λ1
Thus the rational map φ = (F1 , F2 ) : CP (2) → CP (2) is such that If the action is periodic of period τ than ϕτ = Id and we may induce an
dx dy action ψ : C∗ × N → N by setting ψ(a, x) := ϕ 2πi
τ
n u (x).
F = φ∗ (L) for L : −λ = 0. This ends the proof of the Linearization Conversely, any action ψ : C∗ × N → N defines a periodic action ϕ : C ×
x y
Theorem. N → N by ϕ(t, x) = ψ(et , x), ∀ t ∈ C, ∀ x ∈ N ; of period 2πi.

Theorem 10.3.1 is proved in a more general setting in [11]. Indeed it Thanks to this, we will focus on actions ϕ : C × N → N . Given a point
can be stated without the assumption of absence of saddle-nodes. For this x ∈ N the orbit of x is by definition the subset O(x) = {ϕ(t, x); t ∈ C} ⊂ N .
we assume that each irreducible component of Λ contains some hyperbolic Since we are dealing with an action we know that the orbits can contain
attractor in its virtual holonomy. The precise statement is: important information provided some regularity is required: Let  Z be the
∂ϕ 
Theorem 10.5.2 ([11], page 431). Let F be a holomorphic foliation on holomorphic vector field on N defined by Z(x) := (t, x) . Then
∂t t=0
CP (2). Assume that the limit set lim(F ) is algebraic. Denote by lim1 (F ) the integral curves of Z are the orbits of ϕ and the fixed points of ϕ are
the pure codimension one component of lim(F ). Assume that lim1 (F ) = ∅ the singularities of Z, if we assume (and we shall) that Z has isolated
and: singularities on N . The vector field Z is complete and ϕ is its (globally
(1) sing(F ) ∩ lim1 (F ) and lim1 (F ) contains all separatrices of its singu- defined) flow.
larities.
Lemma 11.1.1. The orbits of the action are biholomorphic to C, C∗ = ∼ C
Z
(2) Any irreducible component of lim1 (F ) contains an attractor in its C
virtual holonomy group. or a torus ·
Z⊕Z
Then there are a rational map φ : CP (2) → CP (2) and a linear foliation The above is a straightforward consequence of the fact that the orbit
C
Lλ : xdy − λydx = 0, λ ∈ (C \ R) ∪ (R− \ Q) such that F is the pull-back O(p) is through p biholomorphic to the quotient where Gp ⊂ C is the
F = φ∗ (Lλ ) of L
 λ by φ. Gp
isotropy subgroup Gp = {t ∈ C; ϕ(t, p) = p}.
The proof is a little more elaborate as we have to prove that there are Since Gp ⊂ (C, +) is a discrete subgroup we conclude that Gp  {0}, Z or
no saddle-nodes in the reduction of singularities. Nevertheless, following Z ⊕ Z.
the same line of reasoning presented in this chapter one may be able to Assume now that N is a Stein space. Then N contains no positive
give an alternative geometrical proof of this fact. dimension compact analytic subset. Therefore we get:
Corollary 11.1.2. The orbits of ϕ on a Stein space are biholomorphic to
C
the plane C or to the cylinder C∗ ∼
= ·
Z

11.1.1 Suzuki’s theory


A fundamental contribution to the study of holomorphic flows and foliations
on Stein surfaces, was made by M. Suzuki who introduced on this subject
the use of techniques from potential theory and the theory of analytic spaces
(cf. [70] and [71]). M. Suzuki’s work is from the middle 70’s. Some of his
main results are collected below:
Theorem 11.1.3 (M. Suzuki, [70]). Given a C-action ϕ on a normal Stein
analytic space V of dimension n ≥ 2:
(i) There is a subset E ⊂ V of logarithmic capacity zero such that ϕt (E) =
E, for any t ∈ C, and all orbits of ϕ in V \E are of the same topological
type.
(ii) Any leaf of Fϕ containing an orbit of ϕ isomorphic to C∗ is closed in Example 11.1.7. Let (x, y) ∈ C2 be affine coordinates and define Z(x, y) =
V \singFϕ. ∂ ∂ n
x + λy where λ ∈ Q+ ; say λ = , n, m = 1, n, m ∈ N. Then Z is
∂x ∂y m
(iii) If n = 2 and the leaves of Fϕ are properly embedded in V \ sing(Fϕ ) complete and the flow a holomorphic action ϕ : C × C2 → C2
then there is a meromorphic first integral of Fϕ on V , not constant,  of Z ndefines

given by ϕt (x, y) = xet , ye m t . The orbits are all periodic (diffeomorphic
and one can find a Riemann surface S and a surjective holomorphic to C∗ ) and the origin is a singularity which is non-degenerate and dicritical
map p : V \singFϕ → S, such that: for the foliation F (Z).
(1) The irreducible components of the fibers {p−1 (w); w ∈ S} of p A converse of this example is as follows:
are the leaves of Fϕ .
(2) The subset E ⊂ V union of all the reducible levels p−1 (w), Theorem 11.1.8 (Global linearization theorem, [66]). Let N be a con-

w ∈ S, has zero logarithmic capacity. nected Stein surface with H 2 (N, Z) = 0 equipped with a holomorphic action
ϕ : C × N → N , with isolated singularities, having a non-degenerate dicriti-
(iv) If n = 2 and the generic leaf is isomorphic to C∗ , then any leaf of Fϕ
cal singularity p0 ∈ N . Then N is biholomorphic to C2 , indeed there is a bi-
is closed in V \singFϕ and (therefore) there is a meromorphic first
holomorphic map Φ : N → C2 that conjugates ϕ to an action C × C2 → C2 ,
integral as in (iii).
(t, (x, y)) → xeλ1 t , yeλ2 T for some λ1 , λ2 ∈ C∗ with λ1 /λ2 ∈ Q+ .
For the case of actions with isolated singular points we promptly obtain:
11.1.2 Proof of the Global linearization theorem
Theorem 11.1.4 (Suzuki, [70, 71, 72]). Given a C-action ϕ with isolated
singularities on a normal Stein analytic space N of dimension n ≥ 2 we Let us give an idea of the proof of Theorem 11.1.8. First we make a simple
have: remark:
(i) There is a subset E ⊂ N of zero logarithmic capacity such that: Claim 11.1.9. The generic leaf is diffeomorphic to C∗ .
(a) E is invariant by ϕ : ϕt (ε) ⊂ ε, ∀ t ∈ C Proof. Take a leaf L that accumulates at p0 and is contained in a separatrix
(b) all the orbits in N \ ε are diffeomorphic. Γp0 p0 . We have two possibilities for L: L  C or L  C∗ . If L  C then
in a suitable neighborhood of p0 L ⊃ Γp0 ∪{p0 } so we obtain  a holomorphic
(ii) A periodic orbit of ϕ (i.e., diffeomorphic to C∗ ) is closed in N \Fix(ϕ). map p : C → N holomorphic and non-constant such that pC (C) ⊂ L. This
where Fix(ϕ) = {p ∈ N ; ϕ ∗ t, p) = p, ∀ t ∈ C} is the set of (singular is not possible because N cannot contain a compact holomorphic curve
fixed points of ϕ. In particular, because of Remmert-Stein Theorem, Thus every leaf containing a separatrix is a periodic orbit. Since p0 ∈
such a periodic orbit is contained in an analytic curve in N . sing(Fϕ ) is dicritical its reduction of singularities exhibits some non-invariant
projective line and therefore we have an invariant sector p0 ∈ Sp0 ⊂ N with
(iii) Assume that n = dim N = 2. Then: vertex at p0 , such that every orbit in Sp) is dicritical and therefore periodic.
Because Sp0 has non-empty interior we conclude that the orbits of Fϕ in Sp0
 the orbits are property embedded in N \ Fix(ϕ) then τ =
(a) If all
are not contained in the exceptional set E ⊂ N , which has zero logarithmic
∂ϕ 
admits a meromorphic first integral f : M → C.
∂t t=0 capacity. Therefore the generic orbit of ϕ is periodic.

(b) If the generic orbit of ϕ (i.e., the one in N \ ε) is diffeomorphic Thanks to Suzuki’s theorems 11.1.3 and 11.1.4 we conclude that Fϕ
to C∗ then Z admits a meromorphic first integral. has a meromorphic first integral say f : M → C. If f is holomorphic at p0
then Fϕ has only finitely many separatrices at p0 , contradiction. Thus p0

For the case of analytic actions of C on C2 , in another remarkable work, belongs to the set of points of indetermination of f : the germ of f at p0 is
M. Suzuki proves: the quotient f = h/g, h, g ∈ O2 (p0 ) with h, g = 1 and h(p0 ) = g(p0 ) = 0.
Therefore all the orbits of ϕ that approach p0 are of type C∗ and contain
Theorem 11.1.5 (Suzuki, [71]). Any C∗ -action on C2 is analytically lin- separatrices of Fϕ at p0 .
earizable, i.e., analytically equivalent to an operation of the form s◦(x, y) = Indeed, since Z has a non-degenerate singularity with a meromorphic first
(sn x, sm y), s ∈ C∗ , (x, y) ∈ C2 , for some n, m ∈ N. integral at p0 we conclude that:
The classification of holomorphic C-actions with proper orbits on C2 is • DZ(p0 ) is in the Poincaré-domain, analytically linearizable and
∂ ∂
the following: • Z = λ1 x  + λ2 y in some local coordinates (x, y) ∈ V p0 ; for
∂x ∂
y

Theorem 11.1.6 ([71], Theorem 4). Every holomorphic action ϕ of C some λ1 , λ2 ∈ C with λ1 /λ2 ∈ Q+ .
with proper orbits on C2 is analytically equivalent to one of the following We than consider the attraction basin of p0 denoted by Bp0 (Z) = Bp0 as
operations: in the usual Dynamical Systems context.
Because of the above local linearization we have:
(i) Degenerate operations of the form (α) : t ◦ (x, y) = (x, y + a(x)t) • Bp0 is an open subset that contains a neighborhood of p0 in N .
or (β) : t ◦ (x, y) = (x, eλ(x)t (y − b(x)) + b(x)), where a(x), λ(x) are • The flow of Z is analytically conjugate to the linear flow (on Bp0 ) given
entire functions of one variable x and b(x) is a meromorphic function by
of x such that λ(x).b(x) is holomorphic on x ∈ C. ∂ ∂
Zλ1 ,λ2 = λ1 x + λ2 y
∂x ∂y
(ii) Exponential type operations of the form (γ) : t◦(x, y) = (xenλt , yemλt )
with λ ∈ C∗ , n, m ∈ N. The theorem will be proved if we prove that Bp0 = N : indeed, the attrac-
tion basin B0 (Zλ1 ,λ2 ) of the origin 0 ∈ C2 for the linear vector field Zλ1 ,λ2
(iii) Exponential type operations of the form (γ  ) : t◦(x, y) = (xenλ(u)t , ye−mλ(u)t is C2 .
with λ is an entire function of one variable, n, m ∈ N and u = xm y n . For this we shall prove:

(iv) Operations of the form α−1 ◦ρt ◦α where α(x, y) = (x, x y+P (x)), ∈ Lemma 11.1.10. The boundary ∂Bp0 is a (possibly empty) union of iso-
N, P is a polynomial of degree ≤ − 1 such that P (0) = 0 and ρ is lated singular points and if invariant analytic curves, each curve accumu-
an operation of the form (γ  ) above, where λ(z) has a zero of order lating at a unique non-dicritical singularity of Z.
≥ /m at z = 0.
In order to prove this lemma we go step by step:
Using the above M. Suzuki was already able to prove that: Claim 11.1.11. A leaf L0 ⊂ ∂Bp0 diffeomorphic to C∗ cannot be closed.
(iv) Any C∗ -action on C2 is analytically linearizable, i.e., analytically Proof. Let L0 ⊂ ∂Bp0 be a closed leaf in N which is periodic (diffeomorphic
equivalent/conjugate to s ◦ (x, y) = (sn x, sm y), s ∈ C∗ , (x, y) ∈ C2 to C∗ ). Then L0 is an analytic smooth curve in N (recall that, since
for some n, m ∈ N. Fϕ has a meromorphic first integral, all leaves have analytic closure of

Already present in some of Suzuki’s works is the viewpoint of foliation dimension one in N ). Since N is Stein and H 2 (N, Z) = 0 we can take
with singularities. Let us therefore denote by Fϕ the one-dimensional holo- a holomorphic reduced equation L0 := {h = 0}, where h : N → C is
morphic foliation with singularities
 defined by ϕ on N : Fϕ is the folia- holomorphic. Because L0  C∗ which is homeomorphic to the cylinder
∂ϕ  S 1 × R we can take γ : S 1 → L0 generator of the homology of L0 ; and a
tion F (Z) where τ = ; the leaves of Fϕ are the orbits of ϕ and
∂t t=0 holomorphic 1-form α on L0 such that α=1
sing(Fϕ ) = sing(τ ) = Fix(ϕ). γ
Since N is Stein by a theorem of Cartan there is an extension
  of α
α 11.2 Real transverse sections of holomorphic
to N . Then α is a holomorphic 1-form on N such that αL0 = α and
foliations
therefore  = 1. Choose now a transverse disc Σ centered at some point
α
γ We discuss some problems related to the interplay between geometric the-
1
q0 ∈ γ(S ) ⊂ L0 and consider the holonomy map fγ : (Σ, q0 ) → (Σ, q0 ) ory of foliations and Holomorphic foliations with singularities. We start by
induced by γ. Since Fϕ admits a meromorphic first integral this holonomy recalling that a foliation F on a manifold M is transverse to a submanifold
map fγ is periodic. This means that there n ∈ N such that the lift γ zn of N ⊂ M if for every point p ∈ N we have Tp (F ) + Tp (N ) = Tp (M ) where
γ n to the leaf Lz z for z ∈ Σ, z ≈ q0 , is such that 
γzn is closed. Let us Tp (F ) ⊂ Tp (M ) is the tangent space of F at p, defined by Tp (F ) = Tp (Lp )
assume for simplicity that n = 1. where Lp p is the leaf of F that contains p. In particular, if F is singular,
Then the lifts 
γz of γ one closed paths. Since L0 ⊂ ∂Bp0 we have then sing(F ) ∩ N = φ.
Σ ∩ Bp0 q0 , i.e., we may choose z ∈ Σ arbitrarily close to q0 such that In the real codimension one case the existence of a compact transverse
Lz ⊂ Bp0 . submanifold to the foliation is an important object in the study of the
Since Lz  C∗ and Lz = Lz ∪ {p0 } ∼= C is simply-connected we must have foliation dynamics. As example of this is given by
ω = 0 for closed 1-form ω in Lz and all closed path δ in Lz . Since Theorem 11.2.1 (Haefliger’s Theorem, [9, 26]). Let F be a real codimension
δ  one foliation of class C 2 on a manifold M . Suppose that there exists a an
 is holomorphic its restriction ω := αLz is closed. Thus we conclude
α
immersed curve γ : S 1 → M such that:
z is closed) that
(because γ  = 0, ∀ z ∈ Σ ∩ Bp0 . On the other hand
α

γz  (i) γ is homotopic to zero in M .
  1
for z ∈ Σ close enough to q0 we have   − α < so that
α = 0, (ii) γ is transverse to F .
z
γ γ 2 
γz
contradiction. Then there exists a leaf L0 ∈ F and a closed path δ : S 1 → L0 such that the
Claim 11.1.12. All leaves of Fϕ are biholomorphic to C∗ . holonomy map hδ : (R, 0) → (R, 0) corresponding to δ is a one-sided map:
∃ ε > 0 such that, up to change of orientation on γ transverse section Σ to
Proof. Indeed, in Bp0 the flow is conjugate to a periodic flow so that in L0 at δ(0) = δ(1), we have hδ  +
Σ \{0}
≡ Id and hγ  − is a contraction.
Σ \{0}
Bp0 it has a certain period say ϕτ Bp ≡ Id. The Identity Principle implies
+ −
that ϕτ ≡ Id in N .
0
Σ = Σ ∪ {δ{0}} ∪ Σ
As an important corollary we get:
Claim 11.1.13. Each leaf L0 ⊂ ∂Bp0 accumulates a unique singularity of
Zϕ and this singularity is non-dicritical. Corollary 11.2.2. If M is a compact (real) manifold of dimension m ≥ 2
with finite fundamental group, then M does not admit a (non-singular)
Proof. We already know that L0 is not closed so that it accumulates some real analytic foliation of real codimension one.
singularity say p0 ∈ L0 \ L0 , p0 ∈ sing(Fϕ ). If p0 is dicritical then the
meromorphic first integral f has an indeterminacy point at p0 and by writ- Indeed the above results rely on the following remark:
ing f = g/h g, h ∈ O2 ( p0 ) with g(p0 ) = h(
p0 ) = 0 we conclude that there Lemma 11.2.3. Let F be a real codimension one foliation of a manifold
is an open subset Vp0 (like Bp0 ) such that every leaf accumulating at p0 is M and L a leaf of F such that L \ L p. Then there is a transverse closed
contained in a separatrix of Fϕ through p0 . curve γ : S 1 → M , to F that intersects the leaf L.
In particular, because L0 ⊂ ∂Bp0 , there are leaves L ⊂ Bp0 that contain
separatrices through p0 and through p0 , such a leaf cannot exist because In particular in a compact manifold M any real codimension one has some
L would be compact (a Riemann Sphere) closed transverse curve.

Thus p0 is non-dicritical. The same type of argument shows that L0 \ L0 0


is a single point.

Claim 11.1.14. ∂Bp0 contains no isolated points. L0


Proof. This is for topological reasons: If q0 ∈ ∂Bp0 is an isolated point,
then since Bp0 is diffeomorphic to C2 (the basin of Zλ1 ,λ2 in C2 is C2 ) we
conclude that ∂Bp0 = {q0 } and therefore N is compact homeomorphic to
S 4 = R4 ∪ {∞}, contradiction because N is compact.
Thus we have proved that ∂Bp0 is a discrete union of analytic curves
and proved Lemma 11.1.10.
Let us now proceed:
Figure 11.1:
Claim 11.1.15. N = Bp0 ∪ ∂Bp0 .
Proof. Put A = N \ ∂Bp0 and B = Bp0 .
Then A and B are open subsets of N ; B is connected (diffeomorphic to Another important feature in the above results of Haefliger is the fol-
C2 ) and so is A (∂Bp0 is analytic of dimension ≤ 1). Moreover A ⊃ B and lowing:
∂A = ∂B. Then since N is connected we have A = B, i.e., N = Bp0 ∪∂Bp0 . Let X be a C 1 (real) vector field defined in a neighborhood U of the
2
closed disc D ⊂ R2 , D2 = {(x, y) ∈ R2 : x2 + y 2 ≤ 1}. Assume that:
(i) the singularities of X inside the disc are of Morse type (i.e., centers
Claim 11.1.16. ∂Bp0 = ∅.
x21 + x22 = c or saddles x21 − x22 = c) and (ii) X is transverse (pointing
2
Proof. Suppose that ∂Bp0 ⊃ L0 for some leaf L0 . Let L0 be given by the inwards) to the boundary S 1 = ∂D . Then X has some limit cycle or some
2
dh graph Γ ⊂ D with one sided (holonomy) Poincaré map.
reduced equation {h = 0}, h : N → C holomorphic. Put α := then this
h
a closed meromorphic 1-form on N with polar set (α)∞ = L0 . n

Choose a transverse disc Σ to Fϕ centered at a point q0 ∈ L0 ∩ Σ and a


loop γ : S 1 → Σ \ {q0 }
that avoids the real codimension ≥ 2 subset ∂Bp0 ; γ(S 1 ) ⊂ Bp0 . Then O
because Bp0  C2 we have α = 0. On the other hand a straightforward
γ
dh
computation shows that α= = 2πi, contradiction.
γ γ h
From Claims 11.1.15 and 11.1.16 we conclude that N = BP0 . The
proof of Theorem 11.1.8 is now complete. Figure 11.2:
These are fundamental steps in the way to the proof of the celebrated Remark 11.2.10. (1) We do not assume that Ω is integrable.
3
(2) The distribution Ker(Ω) is defined by: given p ∈ U then Ker(Ω)(p) :=
Theorem 11.2.4 (Novikov compact leaf theorem, [9, 26]). Let M be a {v ∈ Tp (Cn ); Ω(p) · v = 0}.
compact real 3-manifold with finite fundamental group. Any foliation of (3) By a classical result of Malgrange ([46]) a holomorphic foliation of
real codimension one on M 3 has a compact leaf diffeomorphic to the 2- codimension-one admits a holomorphic first integral in a small neighbor-
torus S 1 × S 1 or has all its leaves compact. hood of a point where the singular set of the foliation has codimension
Indeed, if not all leaves are compact then F has some Reeb component which ≥ 3. Therefore, in the above statement if Ω is integrable and q = sing(Ω) ∩
2 B(0; R) then the foliation F : (Ker(Ω)) : Ω = 0 admits a holomorphic f
is a region on M , invariant by F , diffeomorphic to the solid torus S 1 × D n
2
1 1 1
where the boundary ∂(S × D ) = S × S is a leaf and the interior leaves first integral: f : Vp → C, p ∈ Vp ⊂ U ; of Morse type so that f = zj in
j=1
are diffeomorphic to R2 . Resuming the study of holomorphic foliations we
z1 , . . . , zn ).
some local coordinates (
are therefore interested in the consequences of the existence of compact
transverse sections on the dynamics of the foliation. For some reasons Nevertheless, because of the Maximum Principle, F cannot be transverse
we allow the transverse manifold to have codimension different from the to any sphere in Vp . This already shows that the codimension one sit-
dimension of the foliation. We start with the simplest case: Let Z be a uation is (in view of Ito’s theorem above) pretty different from the one-
holomorphic vector field on a neighborhood 0 ∈ U ⊂ Cn , n ≥ 2. Let 0 < r dimensional case. Just to mention an (non-integrable) example we take
be such that B(0, r) = {z ∈ Cn ; |z| ≤ r} ⊂ U . n  
Ω= z2j dz2j−1 − z2j + dzzj then sing(Ω) = {0} and
Definition 11.2.5. We say that Z is transverse to S 2n−1 (0; r) = ∂B(0; r)
j=1

at a point z ∈ S 2m−1 (0; r) if the 2dimensional real foliation induced by Z Ker(Ω) ∩ S 2n−1 (0; R), ∀ R > 0.
on U \ sing(Z) is transverse to the real submanifold S n−1 (0; r) ⊂ U ⊂ R2n
at the point z. It this happens for all z ∈ S 2n−1 (0; r) then we shall say Indeed Ω is a symplectic 1-form associated to a contact structure on the
that Z is transverse to S 2n−1 (0; r) and write ZS 2n−1 (0; r). sphere S 2n−1 (0; R), ∀ R > 0.
Despite some advance in the subject, we still do not know the answer
A characterization of this situation in terms of holomorphic coordinates is:
to the following question:
n ∂ n
Lemma 11.2.6. Write Z = Aj · Then ZS 2n−1 (0; r) ⇔ z̄j Aj = Question: Is there a codimension one holomorphic foliation F defined in
j=1 ∂zj j=1
a neighborhood of a closed ball B ⊂ Cn , n ≥ 3, such that the foliation is
0,
transverse to the boundary sphere ∂B?
∀ z ∈ S 2n−1 (0; r).
→ n ∂
The geometric interpretation is as follows: Let R =
j=1
zj
∂zj
be the 11.3 Non-trivial minimal sets of holomorphic

complex radial vector field on C . Then ZS
n 2n−1
(0; r) ⇔ Z, R = 0 ∀ z ∈ foliations
S 2n−1 (0; r), where  ,  denotes the usual hermitian product on Cn . In
Given a foliation F regular, of any class of differentiability, of M , we call a
particular a vector field Z with a non-degenerate singularity at origin say
minimal set of F a closed invariant subset M ⊂ M such that M is minimal
Spec(D Z(0)) = {λ1 , . . . , λn } ⊂ R2 , is in the Poincaré-domain if and only
with this property.
if ZS 2n−1 (0; ε) for every 0 < ε small enough.
Question: What happens if Z is transverse to a “big sphere” ? Remark 11.3.1. If M is compact, then it is fairly well-known that any
The answer to this question was given by Toshikazu Ito in 1992: (regular) foliation exhibits minimal sets.

Theorem 11.2.7 (Ito,[37]). Let Z be a holomorphic vector field on a Let now F be a foliation with singular set sing(F ) on a compact mani-
neighborhood U of B(0, R) in Cn , n ≥ 2. Suppose that Z∂B(0, R) = fold M .
S 2n−1 (0, R). Then:
Definition 11.3.2. A non-trivial minimal set of F is a minimal set of
(i) Z has exactly one singular point o ∈ sing(Z) in the ball B(0.R). F 1 = F |M\sing(F ) on M \ sing(F ).

(ii) The singularity o ∈ sing(Z) is in the Poincaré domain. If F has some leaf L0 such that L̄0 ∩ sing(F ) = ∅ then m = L̄0 is a
non-trivial minimal set of F . Thus, the problem of existence of non-trivial
By a Möbius transformation we may assume that o is the origin o = 0 ∈ Cn . minimal sets in the case of singular foliations is equivalent to know whether
In this case: all leaves must accumulate singularities or not.
(iii) Z is transverse to all spheres S 2n−1 (0; r), 0 < r ≤ R. Problem 11.3.3 (Camacho-Lins Neto-Sad, [12]). Is there any non-trivial
minimal set for a foliation F on CP (2)?
(iv) There is a real analytic conjugation between the flow of Z in B(0; r) \
{0} and the product [0, ∞) × L where L is the real (transversely holo- Theorem 11.3.4 (Camacho-Lins Neto-Sad, [12]). Let F be a foliation with
morphic) flow induced by Z in S 2n−1 (0; r). a non-trivial minimal set M on CP (2). Then
In particular 1. M is unique.
(v) Each orbit of Z in B(0; R) accumulates at the singular point o. 2. If C ⊂ CP (2) is an algebraic curve then M ∩ C = ∅.
Thus the one-dimensional case is somehow well-understood. On the other 3. F has no algebraic invariant curve.
hand the codimension one case still remains open:
4. There exists an hermitian metric on CP (2) \ sing(F) such that it is
Question 11.2.8. Let n ≥ 3. Is there a codimension one holomorphic complete and induces negative curvature K ≤ −2 < 0 on the leaves
foliation F of a neighborhood B(0; r) ⊂ 0 ∈ U ⊂ Cn , with S 2n−1 (0, R)F ? of F |M . In particular each leaf L ⊂ M is covered by the disc D and
the family of uniformizations is normal.
There is a number of partial results, all of them suggesting that the answer
to the above question is NO. One of the most important is: 5. All leaves L ⊂ M have exponential growth.
Theorem 11.2.9 ([38]). Let Ω be a holomorphic 1-form on a neighborhood Given a 3-dimensional real analytic hypersurface N 3 ⊂ CP (2), we as-
0 ∈ B(0, R) ⊂ U ⊂ Nn , n ≥ 3. Suppose that the distribution Ker(Ω) is sume that N 3 ⊂ CP (2) is Levi-flat. This vanishing of its Levi-form is
transverse to S 2n−1 (0; R). Then: equivalent to the following: for any p ∈ N 3 , there exists a local chart at
p, (x, y) ∈ U for CP (2) such that N ∩ U = {Im.y = 0}. This implies the
(i) n is even (therefore n ≥ 4); following:
(ii) there exists exactly one singular point q ∈ sing(Ω) ∩ B(0; R) and this Theorem 11.3.5 (Lins Neto, [43]). There exists a holomorphic foliation F
point is simple. on CP (2) such that N 3 is F -invariant and sing(F )∩N 3 = ∅. In particular,
n N 3 gives a non-trivial minimal set of F .
If we write Ω = fj (z)dzj with fj : U → C holomorphic, then sing(Ω) =
j=1 Thus, the problem of the existence of non-trivial minimal sets for folia-
{z0 ∈ U ; fj (z0 ) = 0, ∀ j}. A singular point z0 ∈ U is simple if tions on CP (2) is also related to the existence of Levi-flat real hypersurfaces
 (or submanifolds) M 3 ⊂ CP (2). It is still not known whether a codimen-
∂fi (z0 ) sion one holomorphic foliation on CP (n) can have a non-trivial minimal
Det = 0.
∂zj i,j=1,...,n
set.
11.4 Transversely homogeneous holomorphic (i) {Ω1 , ..., Ω } is a rank integrable system which defines F .
foliations (ii) dΩk = i<j cij Ωi ∧ Ωj .
k

From the structural point of view the simplest foliations are those with Then:
an homogeneous transverse structure (cf. [26]). A holomorphic foliation
(iii) For each point p ∈ M there is a neighborhood p ∈ Up ⊆ M equipped
F on a smooth manifold M has a holomorphic homogeneous transverse with a submersion fp : Up → G which defines F in Up such that
strucutre if there are a complex Lie group G, a connected closed subgroup
fp∗ (ωj ) = Ωj in Up , for all j ∈ {1, ..., q}.
H < G such that F admits an atlas of submersions yj : Uj ⊂ M → G/H
satisfying yi = gij ◦ yj for some locally constant map gij : Ui ∩ Uj → G (iv) If Up ∩ Uq = ∅ then in the intersection we have fq = Lgpq (fp ) for
for each Ui ∩ Uj = ∅. In other words, the transversely holomorphic atlas some locally constant left translation Lgpq in G.
of submersions for F has transition maps given by left translations on G
and submersions taking values on the homogeneous space G/H. We shall (v) If M is simply-connected we can take Up = M .
say that F is transversely homogeneous of model G/H. Some important
properties of transversely homogeneous holomorphic foliations are listed 11.5 Transverselly affine foliations
below:
1. Any transversely homogeneous holomorphic foliation is a transversely Let F be a codimension one holomorphic foliation with singularities of M .
holomorphic foliation with a holomorphic homogeneous transverse We say that F is transversely additive when the maps gij in the definition
structure. of holomorphic homogeneous transverse structure are of the form gij (z) =
z + bij , bij ∈ C locally constant in Ui ∩ Uj . If gij (z) = aij z + bij , for
2. Given a foliation F of M as in (1) with model G/H then any real locally constant aij ∈ C − {0} and bij ∈ C we say that F is transversely
submanifold M ⊂ M transverse to F is equipped with a transversely aij z + bij
affine and it is transversely projective if gij (z) = with locally
holomorphic foliation F1 = F |M with holomorphic homogeneous  cij z + dij
transverse structure of model G/H. aij bij
constant ∈ SL(2, C).
cij dij
3. Let F = G/H be an homogeneous space of a complex Lie group G The problem of deciding wether there exist affine transverse structures
(H  G is a closed Lie subgroup). Any homomorphism representation for a given foliation is equivalent to a problem on differential forms as stated
ϕ : π1 (N ) → Aut(F ) gives rise to a transversely holomorphic folia- below:
tion Fϕ on (N  × F )/Φ = Mϕ which is holomorphically transversely
homogeneous of model G/H. Proposition 11.5.1 ([65]). The possible holomorphic affine transverse
structures for F in M are classified by the collections (Ωj , ηj ) of differ-
4. For the case G = P SL(2, C) and H ⊂ G is the affine group H = ential 1-forms defined in the open sets Uj ⊂ M such that:
Aff(C) (isotropy group of the point at infinity ∞ ∈ CP (1)), we have (i) Ωj and ηj are transversely holomorphic, Ωj is integrable and defines
that the quotient G/H  CP (1) is the Riemann sphere and the F in Uj , dΩj = ηj ∧ Ωj and dηj = 0 in Uj , if Ui ∩ Uj = ∅ then
foliations with this transverse model are called transversely projective. dgij
Ωi = gij Ωj and ηi = ηj + for non-vanishing transversely holomorphic
Adapting the above notion for the case of holomorphic foliation with gij
singularities we have: function gij : Ui ∩ Uj → C − {0}.
(ii) Two such collections (Ωj , ηj ) and (Ωj , ηj ) define the same affine
Definition 11.4.1 (transversely homogeneous holomorphic foliation with dgj
singularities). A holomorphic foliation F with singularities on a manifold transverse structure for F in M if and only if Ωj = gj Ωj and ηj = ηj +
gj

M . We shall say that F is transversely homogeneous of model G/H if for some transversely holomorphic non-vanishing functions gj : Uj → C −
the underlying non-singular foliation is transversely homogeneous of model {0}.
G/H on M \ sing(F ). Inparticular, F is called transversely projective if
there is an open cover Uj = M \ sing(F ) such that in each Uj the Proof. First we prove (i). Assume that F is transversely affine with trans-
j∈J versely holomorphic atlas of submersions yj : Uj → C. Given any trans-
foliation is given by a submersion fj : Uj → C and if Ui ∩ Uj = ∅ then we versely holomorphic non-singular 1-form Ωj defining F in Uj we have
have fi = fij ◦ fj in Ui ∩ Uj where fij : Ui ∩ Uj → P SL(2, C) is locally Ωj = gj dyj for some transversely holomorphic function gj : Uj → C − {0}
a f +b dgj
constant. Thus, on each intersection Ui ∩ Uj = ∅, we have fi = cijijfjj+dij ij and we define ηj = · If Ui ∩Uj = ∅ then Ωi = gij Ωj and yi = aij yj +bij
for some locally constant functions aij , bij , cij , dij with aij dij − bij cij = 1. gj
dgi dgj dgij
The data P = {Uj , fj , fij , j ∈ J} is called a projective transverse structure imply dyi = aij dyj and therefore aij gi = gj gij . Thus = + in
for F . gi gj gij
dgij
Basic references for transversely affine and transversely projective foli- Ui ∩ Uj . Clearly dηj = 0, dΩj = ηj ∧ Ωj and ηi = ηj + · This proves
gij
ations (in the nonsingular case) are found in [26]. the first part of (i). Let us prove the second part of (i), i.e., the converse
(5) Based on the Rieman-Koebe uniformization theorem we have: part. For this we assume that we have a unique 1-form Ω which is inte-
grable meromorphic on M and defines F outside the polar divisor (Ω)∞ .
Proposition 11.4.2 ([65], Theorem 6.1 page 203).). Let F be a holo- Assume then that Ω and η are as in the statement. Since η is holomorphic
morphic singular transversely homogeneous foliation of codimension and closed in M \(Ω)∞ , there exists an open cover {Ui }i∈I of M \(Ω)∞ and
one on M n . Then F is a transversely projective foliation on M n . there are holomorphic functions hi ∈ Hol(Ui ) such that η Ui = dhi . We de-

Proof. We know that G/H is a simply-connected complex manifold of di- fine gi = exp(hi ), gi ∈ Ø(Ui )∗ to obtain η Ui = dgi /gi . From dΩ = η ∧ Ω we
mension one. By the Riemann-Koebe uniformization theorem we have obtain d Ω
= 0, and therefore Ω = gi dyi for some holomorphic function
gi
a conformal equivalence G/H ≡ C, C or D the unitary disc. This im-
yi ∈ Ø(Ui ). This can be done in M \(Ω)∞ . Given a point pi ∈ (Ω)∞ we can
plies that either G ⊂ Aut(C) = PSL(2, C), G ⊂ Aut(C) = Aff(C) or
choose a local chart (x, y) ∈ Ui such that pi = (0, 0), (Ω)∞ ∩ U = {y = 0}
G ⊂ Aut(D) ∼ = PSL(2, R). The proposition follows. ∗
and η(x, y) = −n dy df
y + f where n = order of (Ω)∞ and f ∈ Ø(Ui ) . There-
−n
)
11.4.1 Transversely Lie foliations fore we have η = d(f.y dgi
f.y −n = gi , gi = f.y
−n
. The 1-form gΩi is closed and
Ω
holomorphic so that it can be written as gi = dyi for some holomorphic yi .
Let F be a codimension foliation of a manifold M . If F admits a Lie group
We have covered M \s(F ) with open sets Ui where we have the relations
transverse structure of model G, or a G-transverse structure for short, then dgj
we shall call F a G-foliation or, simply, Lie foliation. The characterization Ω = gi dyi , η = dggi . In each Ui ∩ Uj = φ we have gi = η = gj and
i dgi

of G-foliations in terms of differential forms is given below. Let {ω1 , ..., ω } gi dyi = Ω = gj dyj . The first equality implies gj = aij .gi for some locally
be a basis of the Lie algebra of G. Then we have dωk = ckij ωi ∧ ωj for a constant aij and it follows from the second equality that dyi = aij dyj and
i<j then yi = aij yj + bij with bij locally constant in Ui ∩ Uj . This shows that
family constants {ckij } called the structure constants of the Lie algebra in F is transversely affine in M .
the given basis. Now we prove (ii). For this sake it is enough to prove the following:
Theorem 11.4.3 (Darboux-Lie, [26]). Let G be a complex Lie group of Claim 11.5.2. Two pairs (Ω, η) and (Ω , η  ) define the same affine struc-
dimension . Let {ω1 , ..., ω } be a basis of the Lie algebra of G with structure ture for F in M if and only if there exists a meromorphic map g on M
constants {ckij }. Suppose that a complex manifold V m of dimension m ≥ satisfying Ω = gΩ and η  = η + dg
g .
admits a system of 1-forms Ω1 , ..., Ω on M such that:
Proof. Let (Ω, η) be given and let g : M → C be a meromorphic function. Using the notion of monodromy and its properties, F. Touzet has been
We define Ω = gΩ and η  = η + dg g . Using the same notation above we able to study the analytic classification of irreducible singularities which
   
have η   = η  + dg = dgi + dg = d(gi g) and Ω  = g.Ω = (ggi )dyi ,
Ui Ui g gi g (gi .g) Ui Ui
have a suitable projective transverse structure off its set of separatrices.
He calls such a projective structure of moderate growth, meaning that the
and this shows that: gi = aij gj and yi = yi so that aij = aij and
foliation admits a meromorphic projective triple defined in a neighborhood
bij = bij . Hence, the pairs (Ω, η) and (Ω , η  ) define the same transversal
of the singularity. He proves the following:
structure for F in M . Finally, suppose that (Ω, η) and (Ω , η  ) define the
same transversal structure for F in M . Since Ω and Ω define F , we have Theorem 11.6.3 (cf. [73], Theorem II.4.2). Let F a germ of irreducible
Ω = gΩ for some g : M → C meromorphic. Using the same notation above singularity of resonant type or of saddle-node type. Then the foliation ad-
we write (locally) Ω = gi dyi , Ω = gi dyi , η = dgi /gi and η  = dgi /gi ; but mits a meromorphic projective triple near the singularity if and only if on
gi = ggi so η  = η + dg/g completing the proof of the claim. a neighborhood of 0 ∈ C2 , F is the pull-back of a Riccati foliation on C × C
by a meromorphic map.
This ends the proof of Proposition 11.5.1.
The proof of this theorem is based on the study and classification of
Using the above and some other techniques like the index theorem of
the Martinet-Ramis cocycles of the singularity expressed in terms of some
Camacho-Sad (cf.[13]) and some linearization results like those in Chap-
classifying holonomy map of a separatrix of the singularity
ter 10, it is possible to prove that:
For the non-resonant case, without the need of the moderate growth
Theorem 11.5.3 ([65]). Let F be a codimension one foliation on CP (2) hypothesis he proves:
which is transversely affine outside an algebraic codimension one invariant
Theorem 11.6.4 ([73], Theorem II.3.1). A nondegenerate nonresonant
subset S ⊂ CP (2). Suppose that F has reduced non-degenerate singularities
singularity xdy−λydx+ω2 (x, y) = 0, λ ∈ C\Q+ , is analytically linearizable
in S. Then F is a logarithmic foliation.
if and only if the corresponding foliation F is transversely projective in
U \ sep(F , U ) for some neighborhood U of the singularity.
11.6 Transversely projective foliations Another interesting work in this direction is [7], where the authors study
Let M be a complex manifold and F be a codimension one holomorphic foli- the case of irreducible singularities with Liouvillian first integral, in the
ation with singularities of M . Recall that F is called sense of M. Singer [68].
 transversely projective
if the underlying “non-singular” foliation F0 =: F M\sing(F ) is transversely
 11.6.2 Projective structures and differential forms
projective. This means that there is an open cover Uj = M \ sing(F )
j∈J
Let F be a codimension one holomorphic foliation with singular set sing(F )
such that in each Uj the foliation is given by a submersion fj : Uj → C and
of codimension ≥ 2 on a complex manifold M . The existence of a projective
if Ui ∩ Uj = ∅ then we have fi = fij ◦ fj in Ui ∩ Uj where fij : Ui ∩ Uj →
transverse structure for F is equivalent to the existence of suitable triples
P SL(2, C) is locally constant. Thus, on each intersection Ui ∩ Uj = ∅, we
a f +b of differential forms as follows:
have fi = cijijfjj+dij
ij
for some locally constant functions aij , bij , cij , dij with
aij dij − bij cij = 1. Proposition 11.6.5 ([65], Proposition 1.1 page 190). Assume that F is
given by an integrable holomorphic 1-form Ω on M and suppose that there
exists a holomorphic 1-form η on M such that (Proj.1) dΩ = η ∧ Ω. Then
F is transversely projective of M if and only if there exists a holomorphic
1-form ξ on M such that (Proj.2) dη = Ω ∧ ξ and (Proj.3) dξ = ξ ∧ η.

11.6.1 Development of a transversely projective folia- The above proposition helps in the description of some examples of
tion - Touzet’s work transversely projective foliations:

We recall the notion of development of a transversely projective foliation, Example 11.6.6. Let α be a closed meromorphic 1-form on M and let
first mentioned in the beginning of this section, already adapting it to our f : M → C be a meromorphic function. Define (Ω, η, ξ) by: Ω = df −
current framework. Let G be a (non-singular) holomorphic foliation on a f 2 α, η = 2f α and ξ = 2α. Then (Ω, η, ξ) is a projective triple and
complex surface N . Suppose that G is transversely projective in N . There therefore Ω defines a holomorphic foliation of M , transversely projective
is a Galoisian (i.e., a transitive) covering π : P → N where π is holomorphic, in the complement of the analytic invariant codimension one set S ⊂ M ,
a homomorphism h : π1 (N ) → P SL(2, C) and a holomorphic submersion S = (α)∞ ∪ (f )∞ . The same conclusion holds for Ωλ = Ω + λα, where
Φ : P → CP 1 such that: λ ∈ C. The foliation F (Ωλ ) is also transversely affine in some smaller open
set of the form M \S  where S  ⊃ S, S  = S ∪ (f 2 − λ = 0). (In fact
(i) Φ is h-equivariant. This means that for any homotopy class [γ] ∈ Ωλ 
f 2 −λ = f 2 −λ − α is closed and holomorphic in M \S ).
df
π1 (N ), we have
 Example 11.6.7. Let h : M → C∗ be holomorphic such that dξ = − dh 2h ∧ ξ
h([γ])(Φ(x)) = Φ([γ](x)), ∀x ∈ M \ S , √
where ξ is holomorphic. (We can write this condition as d( h.ξ)
 dF = 0). Let

 : P → P we denote the covering map induced by [γ] in
where by [γ] F be any holomorphic function and write (for λ ∈ C) Ω = F · F − 12 dhh −
the Galoisian covering p : P → N . F2 1 dh
2 − λ2 h .ξ, η = 2 h +F ·ξ. The triple (Ω, η, ξ) satisfies the conditions of
  
(ii) π ∗ G N is the foliation defined by the submersion Φ. Proposition 11.6.5 and then F = F (Ω) is a transversely projective foliation
of M .
In the above construction of the development, we may take P as the
universal covering π : N → N of N . We shall refer to the submersion
Θ: N  → CP (1) as a multiform first integral of G given by the projective Proof of Proposition 11.6.5
structure in N . Given a homotopy class [γ] ∈ π1 (M \ S), the corresponding Let us now give a proof for Proposition 11.6.5. We start with a remark
monodromy map is the image h([γ]) ⊂ P SL(2, C). about its need.
Definition 11.6.1. The global monodromy of the foliation, with respect Remark 11.6.8. Proposition 11.6.5 is stated (for the real non-singular
to this development, is the image Mon(G) = h(π1 (N )) ⊂ P SL(2, C). case) with an idea of its proof, in [26] (see Prop. 3.20, pp. 262). However, it
seems that the suggested proof uses some triviality hypothesis on principal
Remark 11.6.2. Some remarks about the above construction are as fol- fiber-bundles of structural group Aff(C), over the manifold M (see [26]
lows. The construction of the development in [26] requires the foliation Prop. 3.6 pp. 249-250). In our case this is replaced by the existence of the
to be nonsingular. Assume now that F is a foliation with singular set of form η in the statement. On the other hand, since some of its elements will
codimension ≥ 2 on a complex  manifold M . Then N = M \ sing(F ) is a be useful later, we supply a proof for Proposition 11.6.5.
complex manifold and G := F N is non-singular. By definition F is trans-
We will use the two following lemmas whose proofs are straightforward
versely projective if and only if G is transversely projective. Moreover, since
computations or consequence of Darboux-Lie theorem, Theorem 11.4.3,
sing(F ) ⊂ M has real codimension ≥ 4, we conclude that there is a natural
therefore left to the reader:
isomorphism π1 (N ) ∼ = π1 (M ). In particular, we can assume in the above
construction that M = N , i.e., the notion of development above introduced Lemma 11.6.9. Let x, y, x , y : U ⊂ Cn → C be meromorphic functions sat-
can be introduced for foliations with singularities. Finally, thanks to Har- a b
isfying: (i) ydx−xdy = ydx−x y ; (ii) xy = ax+by
d , ∈ P SL(2, C).
togs’ extension theorem ([29]), any holomorphic map from M \ sing(F ) to cx+dy c d
2
CP (1) extends uniquely to a holomorphic map from M to CP (1). Then x  = ε.(ax + by) and y = ε.(cx + dy) for some ε ∈ C, ε = 1.
Lemma 11.6.10. Let x, y, x , y : U ⊂ Cn → C
 be meromorphic functions Now we only have to observe that if (Ω, η) is a pair of meromorphic
a b 1-forms and if F is transversely projective in M , then the same steps of the
 = ax + by, y = cx + dy for some
satisfying x ∈ P SL(2, C). Then
c d first part of the proof apply to construct a meromorphic 1-form ξ satisfying
xdy − ydx = x d
y − yd
x. the relations of the statement.
Proof of Proposition 11.6.5. Suppose F is transversely projective in M n , Let F be a codimension one holomorphic foliation with singular set
say, {fi : Ui → C} is a projective transverse structure for F in M \s(F ). sing(F ) of codimension ≥ 2 on a complex manifold M . As mentioned in
In each Ui we have Ω = −gi dfi for some holomorphic gi ∈ Ø(Ui )∗ . In the Introduction, the existence of a projective transverse structure for F
a f +b
each Ui ∩ Uj = φ we have: gi dfi = gj dfj and (1) fi = cijijfjj+dij
ij
as in is equivalent to the existence of suitable triples of differential forms (cf.
Definition 11.4.1. Since dΩ = d(−gi dfi ) = dg dgi Proposition 11.6.5, see [65] Section 3, page 193):
gi ∧ Ω we have η = gi − hi Ω
i

for some holomorphic hi in Ui . We define xi , yi , ui , vi : Ui → C in the This motivates the following definition:
following way: (2) yi2 = gi , xyii = fi , hi = 2v yi
i
and xi vi − yi ui = 1. Definition 11.6.15 (projective triple). Given holomorphic 1-forms (re-
Thus we have: Ω = xi dyi − yi dxi and (3) η = 2(vi dxi − ui dyi ). This spectively, meromorphic 1-forms) Ω, η and ξ on M we shall say that (Ω, η, ξ)
motivates us to define local models (see [26] Section 3.18 pp. 261): ξi = is a holomorphic projective triple (respectively, a meromorphic projective
2(vi dui − ui dvi ) in Ui . It is easy to check that we have dξi = ξi ∧ triple) if they satisfy relations (Proj.1), (Proj.2) and (Proj.3) above. The
η, dη = Ω ∧ ξi in Ui . We can assume that  dxi and dyi are independent foliation F ⊥ defined by the 1-form ξ is called transverse foliation corre-
for all i ∈ I. In fact dxi ∧ dyi = 0 ⇒ dΩUi = 2 dxi ∧ dyi = 0 ⇒ dΩ = 0 sponding to the projective triple. If η is not identically zero then F ⊥ is
in M (we can assume M to be connected) ⇒ we have 0 = dΩ = η ∧ Ω so really a foliation of M which is transverse to F outside of a proper analytic
that η = hΩ for some holomorphic function h : M → C ⇒ we can choose subset.
2
ξ = h 2Ω + hη + dh which satisfies the relations dη = Ω ∧ ξ and dξ = ξ ∧ η.
The following definition plays a fundamental role in the theory of trans-
Claim 11.6.11. ξi = ξj in each Ui ∩ Uj = φ and therefore the ξi ’s can be versely projective foliations.
glued into a holomorphic 1-form ξ in M \s(F ) satisfying the conditions of Definition 11.6.16 (moderate growth (transversely projective foliations)).
the statement. A foliation F of M will be called transversely projective of moderate growth
a x +b y if it admits a meromorphic projective triple defined in M . This means that
Proof. From (1) and (2) we obtain xyii = cijijxjj+dij j
ij yj
. Therefore according to
F is transversely projective in some the complementar of some analytic
Lemma 11.6.9 we have (4) xi = ε.(aij xj + bij xj ), yi = ε.(cij xj + dij yj ) ε2 =
subset M ⊂ M of codimension one.
1. Using (3) and (4) we obtain: (aij vi − cij ui )dxj + (bij vi − dij ui )dyj =
ε.(vj dxj − uj dyj ) and therefore: (5) vj = (aij vi − cij ui ), uj = (−bij vi + The termonilogy foliation with moderate growth has already been intro-
dij uj ). It follows form (5) and Lemma 11.6.10 that vi dui −ui dvi = vj duj − duced in [73]. With the above definitions, Proposition 11.6.5 says that F
uj dvj which proves the claim. is transversely projective on M if and only if the holomorphic pair (Ω, η)
may be completed to a holomorphic projective triple. Moreover, a foliation
Claim 11.6.12. We have ξ = ξi = h2i Ω2 + hi η + dhi in each Ui . F which is transversely projective of moderate growth exhibits a projec-
4vi2 tive transverse structure P in the complement of some codimension divisor
Proof. We have h2i Ω = (xi dyi − yi dxi ), hi η = 4v
yi (vi dxi − ui dyi ),
i
yi2 D ⊂ M (D contained in the polar set of the projective triple). One ques-
2 2
Ω
h v
dhi = 2d vyii . Hence i4 + h2i η + dh2 i = yii dxi − yv2i (xi vi − 1)dyi + dv i
yi .
tion then is whether the projective transverse structure P extends to the
i
divisor D. The other question, apparently simpler, is whether the foliation
On the other hand a straightforward calculation shows that 2 = vi dui − ξi
F is actually projective of moderate growth. According to [65] we may
v2
ui dvi = yii dxi − vyii (xi vi −1)dyi + dv i
yi . And thus Claim 11.6.12 is proved. perform modifications in a projective triple as follows:

Since codim s(F ) ≥ 2 it follows that ξ extends holomorphically to M . Proposition 11.6.17 ([65]). Let M be a connected complex manifold.
This proves the first part. Now we assume that (Ω, η, ξ) is holomorphic as
in the statement of the proposition: (i) Given a meromorphic projective triple (Ω, η, ξ) and meromorphic func-
tions g, h on M we can define a new meromorphic projective triple
Claim 11.6.13. Given any p ∈ M \s(F ) there exist holomorphic functions as follows:
x, y, u, v : U → C defined in an open neighborhood U p such that: Ω =
(Mod.1) Ω = g Ω
xdy − ydx, η = 2(vdx − udy) and ξ = 2(vdu − udv).
(Mod.2) η  = η + dg
g + hΩ
Proof. This claim is a consequence of Darboux’s Theorem 11.4.3 (see also
 1
 h2

[26] pp. 230), but we can give an alternative proof as follows: We write (Mod.3) ξ = g ξ − dh − hη − 2 Ω
locally Ω = −gdf = xdy − ydx and η = dg g − hΩ = 2(vdx − udy) as
in the proof of the first part. Using Claim 11.6.12 and the last part of (ii) Two holomorphic projective triples (Ω, η, ξ) and (Ω , η  , ξ  ) define the
2 same projective transverse structure for a given foliation F if and
Proposition 3.2.5 below we obtain locally ξ = h 2Ω + hη + dh + .Ω; for
only if we have (Mod.1), (Mod.2) and (Mod.3) for some holomorphic
some holomorphic function satisfying −2 ∧ Ω = dΩ. This last equality
d
√ )
functions g, h with g non-vanishing.
implies that d( .Ω) = 0 and then = r(f g2 for some holomorphic function
(iii) Let (Ω, η, ξ) and (Ω, η, ξ  ) be meromorphic projective triples. Then
r(z). Now we look for holomorphic functions f, g and 
h satisfying: Ω =
2 ξ  = ξ + F Ω for some meromorphic function F in M with d Ω =
−gdf, η = dg −  hΩ and ξ = h Ω +  hη + dh. We try f = U (f ) for − 12 dF
F ∧ Ω.

g 2
some holomorphic non-vanishing U (z). Using Ω = gdf = − gdf we get
g = U g(f ) . Using η = dg
   U  This last proposition implies that suitable meromorphic projective triples
g − dΩ = 
d
g − hΩ we get h = h − gU  . Using
g

 U  (f ) also define √ projective transverse structures. We can rewrite condition (iii)


ξ= h2 Ω
+ hη + dh + Ω =
2 +hη + d
h2 Ω
2 h we get d = r(f )df .
U  (f ) on F as d( F Ω) = 0. This implies that if the projective triples (Ω, η, ξ)
Therefore it is possible to write Ω, η and ξ as in the statement of the claim:
# and (Ω, η, ξ  ) are not identical then the foliation defined by Ω is transversely

define x = fy, y = g, v = hy 2 and u =
xv−1
y as in the first part of the affine outside the codimension one analytical invariant subset S = {F =
proof. This proves Claim 3. 0} ∪ {F = ∞}. ([65]).
This approach is useful because of the following result:
Using Claim 11.6.13 we prove that F is transversely projective in M \s(F ),
that is in M . The last part of Proposition 11.6.5 can be proved using Theorem 11.6.18 ([65] Theorem 4.1 page 197). Let F be a foliation of
the relation stated above between the projective structure and the local M where M is a polydisc M ⊂ Cm or a projective manifold over C of
trivializations for Ω, η and ξ. For instance we prove the following. dimension m ≥ 2. Assume that F admits a meromorphic projective triple
1 (Ω, η, ξ) defined in M . If ξ admits a meromorphic first integral in U then
Claim 11.6.14. (Ω, η, ξ) and (f Ω, η + df f , f ξ) define the same projective F is a meromorphic pull-back of a Riccati foliation.
structure for F , for any holomorphic f : M → C∗ .
√ Proof. By hypothesis, ξ defines a foliation which admits a meromorphic
Proof.
√ Using the notation of the first part we define x̂i = f . xi , ŷi =
1 1 first integral. Since we are either on a projective manifold or in a poly-
f . yi , ûi = √f . ui and v̂i = √f . vi . Then: f Ω = x̂i dŷi − ŷi dx̂i , η +
disc centered at the origin, we can write ξ = g dR for some meromorphic
df 1
f = 2(v̂i dx̂i − ûi dŷi ) and f ξ = 2(v̂i dûi − ûi dv̂i ). Furthermore we have functions g and R (these functions are rational in the case of a projec-
x̂i xi aij xj +bij yj aij x̂j +bij ŷj tive surface). Then we may replace the meromorphic triple (Ω, η, ξ) by
ŷi= =yi =
cij xj +dij yj cij x̂j +dij ŷj , and this proves the claim and finishes
1
the holomorphic part of the proof. (Ω , η  , ξ  ) where Ω = gΩ, η  = η + dg 
g and ξ = g ξ = dR. The relations
dΩ = η ∧ ξ , dη = Ω ∧ ξ , dξ = ξ ∧ η imply that η  = HdR for
       
2
some meromorphic function H. Now we define ω := H2 ξ  − Hη  + dH = After making the conversion between the notions of transversely pro-
1 2
2 H dR + dH 1-form such that dω = −HdH ∧ dR. On the other hand
jective foliation in [45] and the one we consider in this text, we can state
η  ∧ ω = HdR ∧ dH = −HdH ∧ dR. Thus dω = η  ∧ ω. We also have the main classification result of [45] as follows:
dη  = dH ∧ dR = (− 21 H 2 dR + dH) ∧ dR = ω ∧ ξ  . The meromorphic
triple (ω, η  , ξ  ) satisfies the projective relations dω = η  ∧ ω, dη  = ω ∧ ξ  , Theorem 11.6.22 (cf. [45], Theorem D). Let F be a codimension one
dξ  = ξ  ∧ η  and therefore by Proposition 11.6.17 (iii) we conclude that transversely projective foliation of moderate growth on a projective mani-
Ω = ω + F.ξ  for some meromorphic function F such that dξ  = ξ  ∧ 12 dF fold M . Then at least one of the following assertions holds true.
F ·
This implies dF ∧dR ≡ 0. By the classical Stein Factorization theorem([27])
1. There exists a generically finite Galois morphism f : Y → M such
we may assume from the beginning that R has connected fibers and there-
that f ∗ F is defined by a closed rational 1-form.
fore dF ∧ dR ≡ 0 implies F = ϕ(R) for some one-variable meromorphic
function ϕ(z) ∈ C(z). In the case where M is a projective manifold all 2. There exists a rational map f : M  S to a ruled surface S, and a
the meromorphic objects are rational and therefore ϕ(z) is also a ratio- Riccati foliation R on S such that F = f ∗ R.
nal function. We obtain therefore Ω = − 21 H 2 dR + dH + ϕ(R)dR ==
dH − ( 12 H 2 − ϕ(R))dR.If we define a meromorphic map σ : M  C × C 3. The transverse projective structure for F has at worst regular sin-
by σ(p) = R(p), H(p) then clearly Ω = σ ∗ (dy − ( 12 y 2 − ϕ(x))dx) and gularities, and the monodromy representation of F factors through
therefore F is the pull-back F = σ ∗ (R) of the Riccati foliation R given on one of the tautological representations of a polydisc Shimura modular
C × C by the meromorphic (rational if M is a projective manifold) 1-form orbifold H.
Ωϕ := dy − ( 12 y 2 − ϕ(x))dx.
There is still a number of interesting questions, on the local and on the
Definition 11.6.19. A meromorphic projective triple (Ω , η  , ξ  ) is geomet- global framework, about the classification and the description of foliations
ric if it can be written locally as in (Mod.1), (Mod.2) and (Mod.3) for some with projective transverse structure.
(locally defined) holomorphic projective triple (Ω, η, ξ) and some (locally
defined) meromorphic functions.
As an immediate consequence we obtain:
Proposition 11.6.20. A geometric projective triple (Ω , η  , ξ  ) defines a
transversely projective foliation F given by Ω on M .

Classification of projective foliations: moderate growth on pro-


jective manifolds
In [45] we find the following definition of transversely projective foliation
of a smooth projective manifold. Let M be a smooth projective manifold
over C. A (holomorphic singular) codimension one foliation F of M . The
foliation is said to be transversely projective if given a non zero rational
1-form ω defining F (and therefore satisfying the Frobenius integrability
condition ω ∧ dω = 0) we have that there are rational 1-forms α and β on
M such thatthe sl2 -connection on the rank 2 trivial vector bundle defined
α β
by Δ = d + is flat.
ω −α

Let us compare the above definition with the one we have been using
so far in this survey. Indeed, compared to Definition 11.4.1 there is a dif-
ference, quite easy to explain. In the above definition, we already assume
that the foliation admits a rational projective triple, i.e., a projective triple
meromorphic defined everywhere in the manifold M . This is not neces-
sarily the case if we just start with a foliation which is (according to our
definition Definition 11.4.1) transversely projective in M \ S for some al-
gebraic curve S ⊂ M . Nevertheless, often we cannot extend the projective Exercises
transverse structure to the curve S (for instance, in the case of Riccati foli-
ations or logarithmic foliations). Thus what is considered in [45] are what
we have called transversely projective foliations with moderate growth (cf.
Definition 11.6.15). projective structure in M \ S. ∂ ∂
1. Given the linear vector field X(x, y) = x −y on C2 describe the
The authors also introduce the following notion: ∂x ∂y
global picture of the corresponding foliation on CP (2) (the foliation
Definition 11.6.21 ([45]). A Riccati foliation over a projective manifold exhibits three singularities on CP (2), two of which are dicritical and
M consists of a pair (π : P → M ; H) = (P ; H) where π : P → M is a require one blow-up).
locally trivial P(1) fiber bundle in the Zariski topology, this means that P
is the projectivization of the total space of a rank two vector bundle E, 2. Let F1 and F2 be two holomorphic foliations with singularities on
and H is a codimension one foliation on P which is transverse to a general CP (2). Show that if F1 = F2 and they have a common leaf the this
fiber of π. In the case of a clear context, the P(1)-bundle P is omitted leaf is algebraic (hint.: Let Fj be given by the polynomial
 vector
from the notation. Then H is called a Riccati foliation. The foliation H is field Xj on C2 . Then a local parametrization
  x(z), y(z)
 , z ∈ D of

defined by the projectivization of horizontal sections of a (non unique) at the common leaf must satisfy x (z), y  (z) = λj (z) · Xj x(z), y(z) ,
meromorphic connection r on E. The connection r is uniquely determined ∀ z ∈ D for some holomorphic λj (z). Then write Xj = (Pj , Qj ) where
by H and its trace on det(E). We say that the Riccati foliation H is regular P1 P2
Pj , Qj are polynomials to conclude that = on the leaf).
if it lifts to a meromorphic connection r with at worst regular singularities Q1 Q2
(see [21]), and irregular if not. It is said that a Riccati foliation (P ; H) 3. State and prove the following local form of closed meromorphic 1-
over M factors through a projective manifold M  if there exists a Riccati forms:
foliation (π  : P  → M  , H  ) over M  , and rational maps φ : M  M  and Let ω be a closed meromorphic 1-form on a neighborhood of the origin
Φ : P  P  , such that π  ◦ Φ = φ ◦ π, and Φ has degree one when restricted 0 ∈ Cn , n ≥ 2. Then there is a neighborhood 0 ∈ U ⊂ Cn where ω is
to a general fiber of P , and H = Φ∗ H  . defined and writes
Using the notion above, alternatively, in [45] the authors state that  r $r
dfj n −1
a foliation F of M is transversely projective if there exists a triple P = ω U = λj + d(g/ fj j )
(P ; H; σ) satisfying j=1
f j j=1

1. (P ; H) is a Riccati foliation over M ; and for some holomorphic fj , g : U → C, g = 0, λj ∈ C, nj ∈ N.


2. σ : M  P is a rational section generically transverse to H such 4. Prove that an isolated singularity of a holomorphic vector field X on
that F = σ ∗ H. Cn , say 0 ∈ Cn , which is in the Poincaré domain; it is necessarily
transverse to the small spheres S 2n−1 (0; ε) (of radius ε > 0) centered
at the singular point. Where, by transverse we mean transversality
between the leaves of the foliation and the sphere, as real submani-
folds of R2n .
5. Let F be a germ of foliation singularity at 0 ∈ C2 . Assume that F
is given by a closed meromorphic one form ω with simple poles in a
neighborhood of 0 ∈ C2 . Prove that F is analytically linearizable.
Sug.: First consider the case where F is not a saddle-node. Show
that the holonomy of a separatrix of F is analytically linearizable.
Some open questions
In order to do this, show that in a neighborhood of a point p = 0
belonging to a separatrix Γ we can choose local coordinates such that
dy
Γ : (y = 0), p : (x = y = 0) and ω(x, y) = a where (a = ResΓ ω). Here are some open questions that the author thinks are relevant in the
y framework of holomorphic foliations with singularities.
Then conclude that if (x, y) are similar coordinates then we have y =
const. y. Using then the Martinet-Ramis formal normal form ([49]), Question 11.6.23. Let F be a polynomial vector field on the complex affine
get rid of the saddle-node case. space C3 . Assume that for infinitely many of its orbits they are complete
intersection of two algebraic surfaces on C3 . Is is true that the vector field
6. Let G ⊂ Diff(C, 0) be an abelian analytically linearizable subgroup
admits a strong rational first integral, i.e., a rational map R : C3  C2
containing an attractor say f ∈ G with |f  (0)| < 1. Suppose that for
such that R is constant on each orbit of X.
every point z ∈ (C, 0) we have O(z) \ O(z) ⊂ {0}. Prove that G is
2πi
generated by f and some rational rotation g(z) = e ν z, ν ∈ Z. Question 11.6.24. Let X be a germ of a holomorphic vector field at the
origin 0 ∈ C3 . Assume that X admits a formal strong first integral, i.e.,
7. Prove that a non-dicritical germ of a holomorphic foliation admitting a pair of formal functions fˆ, ĝ ∈ Ô3 such that dfˆ(X) = 0 and dĝ(X) = 0.
a meromorphic first integral, necessarily admits a holomorphic first Is is true that there is a convergent strong first integral F = (f, g) with
integral. f, g ∈ O3 ?
8. Let F be a foliation on CP (2) (holomorphic with singularities). As- Question 11.6.25. Let be given a polynomial vector field X on Cn . As-
sume that the limit set lim(F ) is algebraic, consisting of points and sume that the set of algebraic orbits of X has positive measure on Cn . Is
a finite number of (invariant) algebraic curves Λj ⊂ CP (2), j = it true that X admits some type of algebraic first integral?
1, . . . , r. Given a point p ∈ Λj \ sing(F ) and a transverse disc p ∈
Σ ∩ Λj show that the virtual holonomy group G = Holvirt (F , Λj , Σ, p) Question 11.6.26. Let G ⊂ Diff(C2 , 0) be a subgroup of germs of complex
satisfies the following property: ∀ z ∈ (Σ, p), O(z) \ O(z) ⊂ {p}. diffeomorphisms at 0 ∈ C2 . Assume that G has the origin as an stable
fixed point, in the sense of Lyapunov. Is it true that G is analytically
9. In the situation of Exercise 8 above assume that the reduction of linearizable?
singularities of a component Λj0 ⊂ lim(F ) exhibits only invariant
components (i.e., sing(F ) ∩ Λj0 is non-dicritical) and each such com- Question 11.6.27. Let X be a polynomial vector field on Cn and denote
ponent has an attractor on its virtual holonomy. Prove that there are by F the corresponding one-dimensional foliation of the complex projective
no saddle-nodes in the reduction of singularities of Λj0 . space CP (n). Assume that there is an algebraic curve Λ ⊂ CP (n) which is
irreducible, invariant by F and stable in the sense of Lyapunov. What is
10. Let F be a holomorphic foliation on CP (2), given on C2 by a closed the normal form of F ? Is it true that if the singularities in Λ are hyperbolic
meromorphic 1-form ω (not necessarily rational 1-form). Show that: then F is given by a linear vector field in some coordinate chart?

(i) If the line at infinity ∞ := CP (2) \ C2 is not F -invariant then Question 11.6.28. Let F be a germ of one-dimensional holomorphic fo-
ω is rational, i.e., ω admits an extension to CP (2). liation at 0 ∈ Cn . Assume that F is induced by a vector field with a
(ii) If ∞ is F -invariant and contains some irreducible singularity non-resonant singularity at the origin. Suppose that F is transversely ho-
of type xdy − λydx + · · · = 0 λ ∈
/ Q+ , then ω also admits an mogeneous in the complement of some invariant analytic hypersurface germ
extension to CP (2). at the origin. Is it true that the vector field germ of X at 0 is analytically
conjugate to its formal normal form?
11. Complete the details in Example 4.2.5.
Question 11.6.29. Let F be a one-dimensional holomorphic foliation of
12. Show that there is no holomorphic foliation of dimension k without the complex projective space CP (n). Assume that: (i) there is an analytic
singularities on the complex projective space CP (n) for 1 ≤ k ≤ n−1. codimension one subset Λ ⊂ CP (n) such that F is transversely homoge-
Is it also true for smooth foliations of even dimension? neous in the complement CP (n) \ Λ; (ii) the singularities of F in Λ are
generic. What is the classification of F ? Does F admit some sort of Liou-
13. Is it true that a one-dimensional holomorphic foliation with isolated villian first integral?
singularities F on the complex projective space CP (n) is given by a
rational vector field?

You might also like