Ann. Scient. Éc. Norm. Sup.

,
4e série, t. 35, 2002, p. 231 à 266.

SOME EXAMPLES FOR THE POINCARÉ

AND PAINLEVÉ PROBLEMS

B Y A LCIDES LINS NETO 1

A BSTRACT . – In 1891, Poincaré started a series of three papers in which he tried to answer the following
question (cf. [21–23]): “Is it possible to decide if an algebraic differential equation in two variables is
algebraically integrable?” (in the sense that it has a rational first integral). More or less at the same time
P. Painlevé asked the following question: “Is it possible to recognize the genus of the general solution of an
algebraic differential equation in two variables which has a rational first integral?”. In this paper we give
examples of one-parameter families which show that both problems have a negative answer. With some of
the families we can also answer a question posed by M. Brunella in [5].
 2002 Éditions scientifiques et médicales Elsevier SAS

R ÉSUMÉ . – En 1891, Poincaré commençait une série de trois articles dans lesquels il tentait de répondre
à la question suivante (cf. [21–23]) : « Est-il possible de décider si une équation différentielle algébrique en
deux variables est algébriquement intégrable ? » (au sens où il admet une intégrale première rationnelle).
À peu près à la même époque, P. Painlevé posait la question suivante : « Est-il possible de déterminer
le genre de la solution générale d’une équation différentielle algébrique en deux variables qui admet une
intégrale première rationnelle ? ». Dans cet article, nous donnons des exemples de familles à un paramètre
qui montrent que les deux problèmes ont une réponse négative. À l’aide de certaines de ces familles, nous
pouvons également répondre à une question de M. Brunella dans [5].
 2002 Éditions scientifiques et médicales Elsevier SAS

1. Introduction

The main purpose of this paper is to describe some examples of one-parameter families
of algebraic differential equations on two variables which show that the so-called “Poincaré
problem” and “Painlevé problem” have a negative answer. With some of the families we give an
answer for a problem stated by M. Brunella in [5].
H. Poincaré in three papers [21,22] and [23], studied the following problem:

“Is it possible to decide if an algebraic differential equation in two variables has a rational first
integral?”

In [22] he starts, by observing that it is sufficient to bound the degree of a possible algebraic
solution. In fact, in [22] and [23] he tries to bound this degree, by supposing that the equation has
a fixed degree (he calls “dimension” of the equation, its degree) and non-degenerate singularities
of fixed local analytic type, that is, with local first integral of the type up .v q = cte, where p

1 This research was partially supported by Pronex.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE
0012-9593/02/02/ 2002 Éditions scientifiques et médicales Elsevier SAS. All rights reserved
232 A. LINS NETO

and q are non zero relatively primes integers which depend of the singularity. When p > 0 and
q < 0, the equation has a local meromorphic first integral and he calls the singularity “dicritical”
or “node” (“nœud”). When p, q > 0 the equation has a local holomorphic first integral and he
calls the singularity a “saddle” (“col”). In [21] he announces the results that he proves in [22]. In
particular, he solves the problem in some particular cases like, for instance, in the case where in
all the saddles we have p/q = −1. He also observes that this problem was studied twenty years
before by Darboux (cf. [11]) and also by Painlevé and Autonne more or less at the same time
than himself. Another problem, posed by P. Painlevé, is the following (cf. [20] and [21]):
“Is it possible to recognize the genus of the general solution of an algebraic differential
equation in two variables which has a rational first integral?”
As we will see, in the families constructed in the begining, all differential equations with
rational first integral have their general integral curve of genus one. However, by pulling-back
these examples by rational maps, it is possible to construct families which contain differential
equations with general integral curves of arbitrarily large genus.
More recently the problem of bounding the degree of a solution, was studied by several authors
in a more general context, that is, without the hypothesis that the equation has a rational first
integral. In particular, it was solved in some particular cases, where conditions are imposed either
on the solution or on the singular points of the equation (cf. [9,8] and [7]).
Before stating our results, we will recall some basic notions which will be used along
the text. We will use freely some basic facts and the terminology of the theory of foliations
∂ ∂
(leaf, holonomy, etc. . .) (cf. [13] and [6]). Given a vector field X = P (x, y) ∂x + Q(x, y) ∂y ,
where P and Q are polynomials on C , the singular set of X is, by definition, the algebraic set
2

sing(X) = {(x, y) ∈ C2 ; P (x, y) = Q(x, y) = 0}. In this paper we will suppose that sing(X)
is finite (i.e., P and Q are relatively prime polynomials). A singular point po = (xo , yo ) of X
is non-degenerate if the jacobian matrix DX(po ) is non singular. In this case, if λ1 and λ2 are
the eigenvalues of DX(po ), then the quotients λ1 /λ2 and λ2 /λ1 are called the characteristic
numbers of the singularity. These characteristic numbers are analytic invariants of the singularity
(cf. [2] or [18]).
Outside sing(X), the system of complex differential equations associated to X:

dx dy
(1) = P (x, y) = Q(x, y),
dt dt
generates a foliation of C2 , whose leaves are the images of the complete solutions of (1), that is,
the immersed Riemann surfaces on C2 locally parametrized by the solutions of (1). Since P
and Q are polynomials, this foliation can be extended to a unique singular foliation on the
projective space CP (2), say F . We will say that the vector field X represents F in the affine
chart C2 ⊂ CP (2). The degree of F is, by definition, the number of tangencies of F with a
generic line, linearly embeded in CP (2). If F is of degree d then the vector field X is of the
form

 ∂
 ∂
(2) X = p(x, y) + x.g(x, y) + q(x, y) + y.g(x, y) ,
∂x ∂y
where p, q are polynomials of degree
d and g is a homogeneous polynomial of degree d, such
that, if g ≡ 0 then max{dg(p), dg(q)} = d and the homogeneous part of p and q of degree d
are not of the form x.h and y.h, respectively. When all singularities of F are non-degenerate,
then F has exactly d2 + d + 1 singularities (where in this number we count also the singularities
which appear at line of infinite after the compactification). This formula is also true when all
singularities of F are isolated, if we count the singularities with multiplicity (cf. [17]).

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 233

We say that an algebraic curve C is invariant by F , or a solution of F , if it is the union of

leaves and singularities of F . For instance, the line at infinity of C2 , L = CP (2) \ C2 , is an
algebraic solution of F if and only if g ≡ 0 in (2). More generally, if C = L is given in the
affine coordinate system C2 by (f = 0), where f is a reduced polynomial, then C is a solution
of F if, and only if, X(f ) := P ∂f ∂f
∂x + Q ∂y = f.h, where h is some polynomial (cf. [17]). The
foliation F has a rational first integral (or is algebraically integrable) if there exists a rational
function F : CP (2)− → C such that all level curves of F are solutions of F .
Another way of representing the foliation F defined by (1) in C2 , is by the differential equation
ω = 0, where ω = P (x, y) dy − Q(x, y) dx. When we represent F in this way, then the rational
function F is a first integral of F if, and only if, dF = H.ω, where H is a rational function.
Since F is a quotient of polynomials, say F = g/h, then this condition is equivalent to

(3) h.dg − g.dh = r.ω,

where r is some polynomial. One of the main difficulties of the problem to recognize if a
differential equation is algebraically integrable or not, is that, although the degree of ω is fixed,
the degrees of g, h and r are, in principle, arbitrary. If we could, in some way, bound the degree
of a possible algebraic solution of (1), then the existence of g, h and r like in (3) would be
reduced to an algebraic system of equations involving the (unknown) coefficients of these three
polynomials and the (known) coefficients of P and Q.
Example. – We say that a foliation F in CP(2) has a Darboux first integral if it can be defined
in an affine chart by a differential equation ω = f.η = 0, where f is a rational function and η is
a closed meromorphic form. It can be shown that η must be of the following type:


r
dfj
η= λj + dg
j=1
fj

where f1 , . . . , fr are polynomials and g is a rational function (cf. [10]). The Darboux integral is
f1λ1 · · · frλr .eg , in this case.
A simple example in which a family of differential equations of some fixed degree contains
members with first integrals of arbitrarily large degrees is given by a family like above, when
g = 0. Consider the family of foliations in CP (2) given in an affine chart by the differential
equations

r
dfj
(4) ωλ := f1 · · · fr λj = 0,
j=1
fj

where r  2, λ = (λ1 , . . . , λr ) ∈ (C∗ )r and f1 , . . . , fr are polynomials in C2 . The foliation

F (ωλ ), defined by (4) in CP(2), has a meromorphic first integral if, and only if, λ = c.m where
m = (m1 , . . . , mr ) ∈ Zr and c ∈ C∗ . However, in this family the singularities of F (ωλ ) are not
 r  2 and that f1 , . . . , fr are chosen in such a way
of fixed analytic type. In fact, suppose that
that there is a point p ∈ (f1 = f2 = 0) \ j
3 (fj = 0), which is not a singularity of the curves
(f1 = 0) and (f2 = 0), and that these two curves meet transversely at p. In this case, a simple
calculation shows that ωλ (p) = 0, that this singularity is non-degenerate and their characteristic
numbers are −λ1 /λ2 and −λ2 /λ1 .
The above example shows that, in order to bound the degree of a possible algebraic solution,
we must fix the analytic type of the singularities of the foliation.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

234 A. LINS NETO

D EFINITION 1. – Let (Fs )s∈S be a family of algebraic foliations on CP (2), where S is some
complex manifold of parameters and the coefficients of the differential equations which define
the family in some fixed affine chart depend holomorphically of s ∈ S. We say that the family
has singularities of fixed analytic type, if
(i) The set of singularities of Fs , s ∈ S, can be written as sing(Fs ) = {p1 (s), . . . , pk (s)},
where the functions s ∈ S → pj (s) are holomorphic.
(ii) For each j ∈ {1, . . . , k} and s1 , s2 ∈ S, there exists a biholomorphism g from a
neighborhood U1 of pj (s1 ) to a neighborhood U2 of pj (s2 ) which sends pj (s1 ) to pj (s2 )
and leaves of Fs1 in U1 onto leaves of Fs2 in U2 . In particular, if the singularity pj (s1 ) is
non-degenerate for Fs1 then pj (s2 ) is also non-degenerate for Fs2 and they have the same
characteristic numbers. When all singularities of Fs are non-degenerate (for all s ∈ S), we
will say that the family has non-degenerate singularities.
Our main result is the following:
M AIN T HEOREM. – For d = 2, 3, 4, there are families of foliations in CP(2), say (Fαd )α∈C , of
degree d, with the following properties:
(a) There is a finite set of parameters Ad ⊂ C such that the restricted family (Fαd )α∈C\Ad has
non-degenerate singularities of fixed analytic type.
(b) There exists a countable and dense set of parameters E ⊂ C, such that for any α ∈ E
the foliation Fad has a rational first integral, say Fα = Pα /Qα , of degree dα (that is the
general level curve Fα = cte is irreducible of degree dα ), satisfying the property that for
any k > 0 the set {α ∈ E; dα
k} is finite. In particular, the degrees of the rational first
integrals can be chosen arbitrarily large and these families provide counterexample for
the Poincaré problem.
/ E, then Fαd has no rational first integral. In particular, almost all leaves of Fαd
(c) If α ∈
are transcendent (non-algebraic). Furthermore, for any of these transcendent leaves,
say L, there exists a holomorphic non-constant map f : C → CP (2), such that f (C) ⊂
L ∪ sing(Fαd ).
Remark 1. – For the foliations like in (b) of the theorem, almost all levels Fα = cte are elliptic
curves, that is are curves of genus one (see Proposition 6 of §2.2). Therefore these families are
not couterexamples for the Painlevé problem. However, by pulling-back such families by rational
maps Φ : CP (2) → CP (2) of topological degree > 1, we can obtain such couterexamples, as in
the following:
C OROLLARY. – For any d  5, there exists a family of foliations of degree d in CP(2), say
(Fαd )α∈C , satisfying (a) of the main Theorem and
(b.1) There exists a countable and dense set of parameters E ⊂ C, such that for any α ∈ E the
foliation Fad has a rational first integral, say Fα , of degree dα , satisfying the property that for
any k > 0 the sets {α ∈ E d ; dα
k} and

{α ∈ E; such that the genus of the general level curve of Fα is
k}

are finite. In particular, the degrees of the rational first integrals and the genus of their general
level curves can be chosen arbitrarily large and these families provide couterexamples for
Painlevé problem.
Remark 2. – As we will see, the singularities of the foliations in Fαd , α ∈
/ Ad , are singularities
with local first integrals of the following analytic types:
(a.1) In the degree four situation: v/u = cte for the dicritical singularities and v 3 .u = cte for
the saddles.

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 235

(a.2) In the degree three situation: v/u = cte or v 2 /u = cte for the dicritical singularities and
v 3 .u = cte or v 6 .u = cte for the saddles.
(a.3) In the degree two situation: v 2 /u = cte or v 3 /u2 = cte for the dicritical singularities
and v 3 .u = cte or v 6 .u = cte for the saddles.
Remark 3. – At the end of §2.2 we will prove that, for the foliations Fα4 in the main
Theorem, α ∈ / E, after the resolution of the dicritical singularities of Fa4 , almost all leaves of
the resolved foliation are biholomorphic to C. We observe that these foliations answer a question
of M. Brunella in [5]. In [5], Brunella proved that, if F is a foliation in CP(2) of degree  5,
whose singularities are non-degenerate, and if f : C → CP(2) is a non-constant holomorphic map
tangent to F , then f (C) must be contained in an algebraic solution of F . At the end of [5] he
asks if the same result could be proved for foliations of degree  2.
Remark 4. – In the main Theorem and in the Corollary, when α ∈ / E the foliation Fαd is not
Darboux integrable. On the other hand, it is Liouvillian integrable, that is, the foliation Fαd can
be defined by a meromorphic 1-form ωα in CP(2) in such a way that there exists a meromorphic
closed 1-form η such that dωα = η ∧ ωα . This fact will be proved for the family of degree four
in Remark 6 of §2.2. We would like to observe also that, since the families are not Darboux
integrable, the set E cannot be described explicitly (in principle). It would be interesting to know
what kinds of properties this set has.
Remark 5. – Another interesting fact about any of the families is that, if α ∈
/ E then:
(i) For any p ∈ CP(2), there exists a neighborhood U of p such that Fαd |U has a first integral
(holomorphic or meromorphic).
(ii) Fαd has no meromorphic global first integral.
As far as we know, these are the first known examples of algebraic foliations in CP(2) with
such properties.
We would like to state some problems which arise naturally from the examples and some
well-known results.
Problem 1. – An useful result, due to Darboux and later generalized by Jouanolou, is the
following (cf. [14] (p. 29), [11] and [15]):
T HEOREM. – For all d  1 there exists N (d) ∈ N such that if a foliation of degree d has more
than N (d) algebraic solutions, then it has a rational first integral. In particular, a foliation with
infinitely many algebraic solutions has a rational first integral.
A natural question should be the following: given d  2, is there M (d) ∈ N such that if a
foliation of degree d has an algebraic solution of degree greater than or equal to M (d), then it has
a rational first integral? However, I recently have received a preprint by J. Moulin Ollagnier [19]
in which he claims that this question has a negative answer, at least for foliations of degree two. In
this paper he exhibits a countable family of Lotka–Volterra polynomial differential equations, say
LV (,), , ∈ N, without rational first integrals, with associated foliations of degree two, and such
that LV (,) has an irreducible algebraic solution of degree 2,. We observe that the singularities
of the foliations in Ollagnier’s family are not of fixed analytic type. On the other hand, all such
foliations have Liouvillian first integrals, so that a more realistic question would be the following:
Question. – Given d  2, is there M (d) ∈ N such that if a foliation of degree d has an algebraic
solution of degree greater than or equal to M (d), then it has a Liouvillian first integral?
Problem 2. – Classify all families of foliations in CP(2) having properties (a), (b) and (c) of
the families of the main Theorem.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

236 A. LINS NETO

Problem 3. – Classify all families of foliations in CP(2) having properties (a) and (b.1) of
the families of the Corollary.
I would like to thank the referee for helpful ideas and suggestions which enabled me to
simplify and improve some of the steps of the proofs.

2. Proof of the main Theorem

We begin by describing a configuration of lines and points in CP(2) that will be used in the
construction of the family of degree four.

2.1. A particular configuration of lines and points

Let us consider the lines in CP(2) defined in homogeneous coordinates by the equation
 3   
(5) Y − X 3 Z 3 − Y 3 X 3 − Z 3 = 0.

It is not difficult to see that the nine lines defined by (5), say ,1 , . . . , ,9 , intersect in twelve points
of CP(2), which we will denote by p1 , . . . , p12 . We will use the notation j for the primitive cubic
root of the unity e2πi/3 .
In Fig. 1 we sketch the configuration in the affine plane (Z = 1). The lines ,1 , . . . , ,9 in
this case are represented by (x = 1), (x = j), (x = j 2 ), (y = 1), (y = j), (y = j 2 ), (y = x),
(y = jx) and (y = j 2 x). The points of the configuration in the finite plane are (0, 0) and (j k , j  ),
k, , ∈ {0, 1, 2}, and the two points [1 : 0 : 0] and [0 : 1 : 0] at the line (Z = 0). We will denote this
configuration by C = (L, P), where L = {,1 , . . . , ,9 } and P = {p1 , . . . , p12 }.
We observe that the configuration C satisfies the following properties:
(I) Each line contains four points of the configuration.
(II) Each point belongs to three lines of the configuration.
(III) If three points of P are not in a line , ∈ L, then the points are not aligned.
In the next proposition we obtain some symmetries of the configuration C and prove that it is
essentially the unique configuration satisfying (I), (II) and (III).

Fig. 1.

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 237

P ROPOSITION 1. – Let C  = (L , P  ) be a configuration of nine lines and twelve points, where
L = {,1 , . . . , ,9 } and P  = {p1 , . . . , p12 } are the sets of lines and points of C  . Suppose that C 


satisfies properties (I), (II) and (III) above. Then P  can be divided into four subsets P1 , . . . , P4 ,
each one containing three points, with the following properties:
(a) Pi ∩ Pj = ∅ if i = j.
(b) If Pj = {p1 , p2 , p3 }, then there exists an automorphism T of CP(2) which sends C 
into C (that is T (L ) = L and T (P  ) = P) in such a way that T (p1 ) = [0 : 0 : 1],
T (p2 ) = [0 : 1 : 0] and T (p3 ) = [1 : 0 : 0].
(c) Given i, j ∈ {1, . . . , 4}, i = j, there exists an automorphism S of CP(2) such that
S(C  ) = C  and S(Pi ) = Pj .
(d) Given pi , pj ∈ P, there exists an automorphism S of CP(2) such that S(C  ) = C  and
S(pi ) = pj .

Proof. – Choose a point p1 ∈ P  . It follows from (II) that p1 belongs to three lines, say ,1 , ,2
and ,3 . Since each of these lines contain four points of P  , it is clear that ,1 ∪ ,2 ∪ ,3 contains
ten points of P  , so that there are two other points in P  , which we can suppose to be p2 and p3 .
Let P1 = {p1 , p2 , p3 }. If we consider now a point pi ∈ P  \ P1 , then we can define in a similar
way a subset P2 , with three points, such that pi ∈ P2 , P1 ∩ P2 = ∅ and the other two points of
P2 are not contained in the three lines of L which contain pi .
In the same way, we can define subsets P3 and P4 (inductively), such that P1 , . . . , P4 satisfy (a)
of the proposition and for each j ∈ {1, . . . , 4} the three points of Pj are not aligned and two of
them do not belong to the same line of L . We leave the details for the reader.
Now, let us consider P1 = {p1 , p2 , p3 } and prove that it satisfies (b). Since p1 , p2 and p3
are not aligned, there exists an automorphism T1 of CP(2) such that T1 (p1 ) = [0 : 0 : 1],
T1 (p2 ) = [0 : 1 : 0] and T1 (p3 ) = [1 : 0 : 0], and so we can suppose that p1 = [0 : 0 : 1],
p2 = [0 : 1 : 0] and p3 = [1 : 0 : 0]. Let us consider the configuration in the affine plane (Z = 1),
with affine coordinates (x, y) ∈ C2 . In this coordinate system, the three lines of L through p2
are of the form (x = a1 ),(x = a2 ) and (x = a3 ), and the three lines through p3 are of the form
(y = b1 ),(y = b2 ) and (y = b3 ) (see Fig. 2).
Now, each one of the other three lines of the configuration contains p1 = (0, 0), is neither
horizontal nor vertical and contains exactly three points in the set {(ai , bj ); 1
i, j
3}. After
changing the sub-indexes in the aj  s , if necessary, we can suppose that one of these lines contains
the points (aj , bj ), j = 1, 2, 3. In this way, one of the other lines contains the points (a1 , b2 ),
(a2 , b3 ) and (a3 , b1 ) and the third one the points (a1 , b3 ), (a2 , b1 ) and (a3 , b2 ). This implies that

b1 b2 b3 b2 b3 b1 b3 b1 b2
(6) = = = α, = = =β and = = = γ,
a1 a2 a3 a1 a2 a3 a1 a2 a3

Fig. 2.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

238 A. LINS NETO

where α, β, γ = 0, are distinct complex numbers. It follows from (6) that α3 = β 3 = γ 3 , and
so β/α and γ/α are distinct primitive cubic roots of the unity. Let us suppose, for instance, that
β/α = j and γ/α = j 2 . This implies that
   
(a1 , a2 , a3 ) = a1 1, j, j 2 and (b1 , b2 , b3 ) = αa1 1, j, j 2

as the reader can check. Now, it is not difficult to see that the automorphism T of CP(2) defined
by


T [X : Y : Z] = a−1 1 X : (αa1 )
−1
Y :Z

satisfies (b). Assertions (c) and (d) follow easily from (b). We leave the details for the reader. ✷

2.2. The family of degree four

Let F be a foliation of degree four in CP(2) for which all the lines of the configuration C
are invariant. In the affine chart (x, y) ∈ C2 such that the lines of L are the components of
(y 3 − 1).(x3 − 1).(y 3 − x3 ) = 0, such a foliation F must be given by a differential equation of
the form:
 dx  

dt = x − 1 . a0 + a10 x + a01 y + by ,
3 2
(7)  3 

dt = y − 1 . b0 + b10 x + b01 y + bx
dy 2

because the components of (x3 − 1).(y 3 − 1) = 0 are invariants by F . We recall that a foliation
of degree four in the affine chart z = (x, y), is defined by a differential equation of the form
dt = F (x, y) + g(x, y)R(x, y), where F = (p, q), p and q are polynomials of degree
4, g
dz

is a homogeneous polynomial of degree 4 and R(x, y) = (x, y) (see (2) in §1). Therefore, the
homogeneous part of fifth degree in the right side of (7) must be of the form x3 .by 2 in the first
component and y 3 .bx2 in the second. Since the components of (y 3 − x3 ) = 0 are also invariants,
(0, 0) must be a singularity of F , so that a0 = b0 = 0. On the other hand, the fact that the lines
(y = x), (y = jx) and (y = j 2 x) are invariant, implies that
 

 b10 + b01 = a10 + a01 

 
b10 + b01 j = a10 j + a01 j 2 ⇒ a01 = b10 = 0 and a10 = b01

 

 2 2 
b10 + b01 j = a10 j + a01 j

so that, if we multiply both components of (7) by a convenient constant, we can write (7) in one
of the following forms:
 dx  3  
dt = Pα = x − 1 . x − αy ,
2
(8)  3   or
dt = Qα = y − 1 . y − αx ,
dy 2
 dx  3  2
dt = P∞ = x − 1 y ,
(9)  3  2
dt = Q∞ = y − 1 x .
dy

We call Fα4 the foliation given by (8) and F∞

4
the foliation given by (9). This defines the family
4
(Fa )α∈C , of degree four.

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 239

Remark 6. – The foliation Fα4 is Liouvillian integrable as asserted in Remark 4 of §1. Let us
verify this fact. Observe first that Fα4 can be defined by the meromorphic 1-form

x − αy 2 y − αx2
ωα = dy − dx.
y3 − 1 x3 − 1

On the other hand, dωα = η ∧ ωα where

H +1 y3 − 1
η= dH, and H = .
3H(H − 1) x3 − 1

Since dη = 0, Fα4 is Liouvillian integrable.

Remark 7. – Let S be an automorphism of CP(2) such that S(C) = C (see (c) and (d) of
Proposition 1). It follows from the above arguments, that S sends the family (Fa4 )α∈C in itself,
that is, for every α ∈ C, we have that S ∗ (Fa4 ) = Fβ4 for some β ∈ C.
P ROPOSITION 2. – Let (Fa4 )α∈C be as above. Then:
(a) All points of the configuration C are singularities of Fα4 , for all α ∈ C.
(b) If α ∈/ {1, j, j 2 , ∞}, then all singularities of Fα4 are non-degenerate. Moreover, these
singularities are the following:
(b.1) Twelve singularities of radial type at the points of P = {p1 , . . . , p12 }. The foliation Fα4
has a local meromorphic first integral of the type v/u = cte, in a neighborhood of each
singularity of this set.
(b.2) Nine singularities of saddle type, say {q1 (α), . . . , q9 (α)}, where, for each j ∈ {1, . . . , 9},
qj (α) ∈ ,j , a line of L. The foliation Fα4 has a local holomorphic first integral of the
type v 3 .u = cte, in a neighborhood of each qj (α).
(c) If α ∈ {1, j, j 2, ∞}, then sing(Fα4 ) = P. Among the points in P, nine are of radial type
(like in (b.1)) and the three others are contained in one of the subsets Pj , j = 1, 2, 3, 4,
as in Proposition 1. Moreover, Fα4 has a rational first integral of the type Hα = Pα /Qα ,
where Pα and Qα are the product of three lines in L. These six lines can be chosen in the
following way: Suppose that the three non radial singularities of Fα4 are p1 , p2 , p3 ∈ Pj .
Let ,1 , ,2 , ,3 and ,4 , ,5 , ,6 be the lines of L which pass through two of the points of Pj .
Then Hα = 14 . .5 .6 is a rational first integral of Fα . In particular, for α = ∞, the first
2 .3 4

integral can be chosen to be:

y3 − 1
(10) H(x, y) := H∞ (x, y) = .
x3 − 1
Proof. – Assertion (a) follows from the fact that the points of P are the singularities of the
algebraic set (Y 3 − Z 3 ).(X 3 − Z 3 ).(Y 3 − X 3 ) = 0, which is invariant for all Fα4 s . Let us prove
assertion (b).
The fact that the singularities of Fα4 are non-degenerate for α ∈ / {1, j, j 2 , ∞} can be proved
by calculating directly its singularities and the Jacobian matrix of the vector field given by (8) in
these points. However, another way to prove it, is just observing that sing(Fα4 ) contains exactly
21 = 42 + 4 + 1 points, if α ∈ / {1, j, j 2 , ∞} (see §1). Let us prove this fact. Consider, for instance
the singularities of Fα in the line (x = 1). Since Fα4 has degree four, this line must contain five
4

singularities (counted with multiplicities). Four of these singularities are the points (1, 1), (1, j),
(1, j 2 ) (in the finite plane) and [0 : 1 : 0] ∈ (Z = 0). The other one is, of course, the solution of
x − 1 = y − αx2 = 0, which is (1, α). If α ∈ / {1, j, j 2, ∞}, then this singularity does not coincide
with any of the others, so that, in this case (x = 1) contains exactly five singularities of Fα4 . With

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

240 A. LINS NETO

a similar computation, it can be proved that if α ∈ / {1, j, j 2 , ∞}, then each line of L contains
exactly five singularities of Fα . This implies that Fα4 has nine singularities outside the set P, so
4

that it has exactly 21 singularities. Let us prove (b.1).

Let Y be the polynomial vector field which represents Fα4 in an affine chart which contains
p ∈ P. Since p is a non-degenerate singularity of Fα4 , the Jacobian matrix DY (p) has two
non zero eigenvalues, say λ1 , λ2 = 0. On the other hand, since Fα4 has three invariant straight
lines through p, it is clear that λ1 = λ2 = λ and DY (p) = λ.I, where I is the 2 × 2 identity
matrix. Now, it follows from the Poincaré’s linearization theorem, that Y is linearizable in a
neighborhood of p (cf. [2]). Therefore, there exists a (local) holomorphic coordinate system
(U, (u, v)) such that p ∈ U , u(p) = v(p) = 0 and Y (u, v) = λ(u ∂u ∂ ∂
+ v ∂v ), so that, in these
coordinates v/u is a meromorphic first integral of Fα in U .
4

In order to prove (b.2), let us fix a singularity qj (α) ∈ ,j ∈ L, say (1, α) ∈ ,1 = (x = 1).
Consider the vector field X which represents Fα4 in the affine chart (x, y) (see (8)):

 3   ∂
   ∂
X(x, y) = x − 1 . x − αy 2 + y 3 − 1 . y − αx2 .
∂x ∂y

A direct computation shows that

   
3 1 − α3 0
DX(1, α) =  
0 α3 − 1

and so the eigenvalues of DX(1, α) are λ1 = 3(1 − α3 ) and λ2 = α3 − 1 and satisfy λ1 = −3.λ2 ,
where λ1 is the eigenvalue in the normal direction to ,1 and λ2 in the tangent direction to ,1 .
With a similar calculation (or using Proposition 1), it can be proved that this property is true for
all singularities qj (α), j = 2, . . . , 9. It remains to prove that Fα4 has a local holomorphic first
integral in a neighborhood of qj (α) of the form v 3 .u = cte. In order to prove this fact, we recall
that the other four singularities of Fα4 on ,j are radial and we will consider the resolution of these
singularities by blowing-up. From now on, we shall use the following notations:
(i) M will be the manifold obtained from CP(2) by blowing-up in the twelve points of P.
The blowing-up map will be denoted by Π : M → CP(2).
(ii) F̃α will be the strict transform of Fα4 by Π.
9
(iii) ,̃j , j = 1, . . . , 9, will be the strict transform of the line ,j ∈ L by Π and L = j=1 ,̃j .
(iv) Di will be the divisor Π−1 (pi ).
Observe that F̃α has just one singularity on ,̃j , say q̃j (α), where Π(q̃j (α)) = qj (α). This
implies that ,̃j \ {q̃j (α)} is a leaf of F̃α biholomorphic to C. Therefore, the holonomy of this
leaf is trivial, and this implies that the holonomy of the local separatrix of q̃j (α) contained
in ,̃j is the identity. In particular, this holonomy is linearizable. It follows from a Theorem of
Mattei and Moussu (cf. [18]), that F̃α is linearizable in a neighborhood of q̃j (α). Since Π is a
biholomorphism in a neighborhood of q̃j (α), the same is true for Fα4 in a neighborhood of qj (α).
Therefore, we can suppose that in a suitable holomorphic coordinate system (V, (u, v)) around
qj (α), we have
∂ ∂
(11) f (u, v).X(u, v) = 3u −v ,
∂u ∂v
where f ∈ O∗ (V ) and V ∩ ,j = (u = 0). This implies that v 3 .u is a holomorphic first integral of
Fα4 in V , which proves (b.2).

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 241

y 3 −1 P
Let us consider now the rational map H(x, y) = x3 −1 = Q. Then

dH dP dQ 3y 2 dy 3x2 dx 3
    
= − = 3 − 3 = 3 . x3 − 1 y 2 dy − y 3 − 1 x2 dx .
H P Q y − 1 x − 1 (x − 1)(y − 1)
3

This implies that H = H∞ is a rational first integral for F∞ 4

(see (9)). Observe that H − 1 =
y 3 −x3 x3 −y 3
x3 −1 and H
1
− 1 = y 3 −1 are also rational first integrals for F 4
∞ . Therefore, if we choose six
lines of L, say ,1 , . . . , ,6 , such that ,1 , . . . , ,3 pass through one of the points of P1 = {[1 : 0 : 0],
[0 : 1 : 0], [0 : 0 : 1]} and ,4 , . . . , ,6 pass through another one, then 14..25.63 is a rational first
integral for F∞ 4
. Now let Pj , j ∈ {2, 3, 4}, be another subset of P as in Proposition 1. It follows
from (c) of Proposition 1 that there exists an automorphism S of CP(2) such that S(C) = C
and S(Pj ) = P1 . Since S(C) = C we must have S ∗ (F∞ 4
) = FT4 for some T ∈ C \ {∞}. In
fact, T ∈ {1, j, j }, because for α ∈
2
/ {1, j, j , ∞}, Fα has 21 singularities, whereas F∞
2 4 4
has 12
singularities. It follows that H ◦ S is a rational first integral of FT of the type required in (c).
This finishes the proof of Proposition 2. ✷
One of the main facts about the family (Fα4 )α∈C , is the following:
P ROPOSITION 3. – If α = β then F̃α and F̃β are transversal outside L = ,̃1 ∪ · · · ∪ ,̃9 .
Proof. – We will prove the proposition for α, β = ∞ and leave for the reader the case
where either α = ∞ or β = ∞. Let us prove first that F̃α and F̃β are transversal outside
L ∪ D1 ∪ · · · ∪ D12 . Since Π is a biholomorphism outside D1 ∪ · · · ∪ D12 , it is sufficient to
prove that Fα4 and Fβ4 are transversal outside ,1 ∪ · · · ∪ ,9 . By considering the determinant of the
components of vector fields which define Fα4 and Fβ4 (see (8)), we get:
   
Pα .Qβ − Pβ .Qα = (β − α). x3 − 1 . y 3 − 1 . y 3 − x3 .

This proves that Fα4 and Fβ4 are transversal outside ,1 ∪ · · · ∪ ,9 ∪ (Z = 0). In order to verify that
they are tranversal in (Z = 0) \ (,1 ∪ · · · ∪ ,9 ) = (Z = 0) \ {[1 : 0 : 0], [0 : 1 : 0]} we can use (c) of
Proposition 1. It follows from (c) of Proposition 1, that it is possible to find an automorphism S
of CP(2) such that S(C) = C and S((Z = 0) \ {[1 : 0 : 0], [0 : 1 : 0]}) ⊂ C2 . It follows from
Remark 7 and the above computation that Fα4 and Fβ4 are transversal in (Z = 0) \ {[1 : 0 : 0],
[0 : 1 : 0]}. This proves that F̃α and F̃β are transversal outside L ∪ D1 ∪ · · · ∪ D12 .
It remains to prove that F̃α and F̃β are transversal in (D1 ∪ · · · ∪ D12 ) \ L. Observe that,
(d) of Proposition 1 and Remark 7, imply that it is sufficient to prove that they are tranversal in
Dj \ L, for some j ∈ {1, . . . , 12}. Since p1 = (0, 0) ∈ P, we will prove that they are tranversal
in D1 \ L, by doing a blowing-up π at 0 ∈ C2 . Consider, for instance this blowing-up in the
chart (u, x) ∈ C2 , where π(u, x) = (x, u.x). A straightforward computation shows that the strict
transform of F̃α can be defined in this chart by:
  3 
dt = α u − 1 + x.h1 (u, x),
du
(12) dx
dt = 1 + x.h2 (u, x),

where h1 and h2 are polynomials. In this chart, the equation of the divisor D1 is (x = 0), so
that the slopes of F̃α and F̃β along D1 are given by du dx = α(u − 1) and dx = β(u − 1),
3 du 3

respectively. This implies that they are tranversal along D1 , outside (u − 1 = 0), which is the
3

equation of L in this chart. With a similar computation, it is possible to prove that they are also
tranversal outside L in the other chart (v, y), where π(v, y) = (v.y, y). We leave this computation
for the reader. This finishes the proof of Proposition 3. ✷

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

242 A. LINS NETO

Fig. 3.

Now, consider the first integral H = xy 3 −1

3

−1 of F∞ and set h = H ◦ Π. Since the base points

4

of H are contained in P and these base points are radial points for F∞ , it is not difficult to see
that h : M → C is holomorphic and the singular fibers of h are h−1 (0), h−1 (1) and h−1 (∞).
Moreover, if V = M \ h−1 {0, 1, ∞} and W = C \ {0, 1, ∞}, then h|V : V → W is an elliptic
regular fibration. The idea of the proof of the main Theorem is to use Proposition 3, this fibration
and the theory of foliations transversal to the fibers of a fibration (see for instance [12] or [6]).
According to this theory, if we fix one of the foliations F̃α , α = ∞ and a fiber of h|V , say
Ta = h−1 (a), then there exists a group representation

Hα : Π1 (W, a) → diff (Ta ),

called global holonomy representation, with the following properties:

(I) If Lo is a leaf of F̃α |V , then h|Lo : Lo → W is a covering map.
(II) Given a closed path γ : [0, 1] → W such that γ(0) = γ(1) = a, with class [γ] ∈ Π1 (W, a),
and a point q ∈ Lo ∩ Ta , then Hα ([γ])(q) is the end point of the unique path γ̃ : [0, 1] → Lo such
that γ̃(0) = q and h ◦ γ̃ = γ. In particular, the orbit of q by Hα ([γ]) is contained in Lo ∩ Ta .
(III) If we denote by Gα the subgroup Hα (Π1 (W, a)) of diff (Ta ), then
 
(13) Lo ∩ Ta = f (q): f ∈ Gα ,

that is Lo ∩ Ta is the orbit of some point q ∈ Lo ∩ Ta by the group of diffeomorphims Gα .

Remark 8. – Since the foliation F̃α and the fibration h|V are holomorphic, it follows that all
elements f ∈ Gα are automorphims of the elliptic curve Ta . Moreover, if we fix γ ∈ Π1 (W, a)
then we can define a map Fγ : C × Ta → Ta by

Fγ (α, q) = Hα (γ)(q),

where Hα is the global holonomy of F̃α . It follows from the theorem of holomorphic dependence
of the solutions of a holomorphic differential equation with respect to parameters and initial
conditions, that the map Fγ is holomorphic in both variables.
On the other hand, W = C \ {0, 1, ∞} = C \ {0, 1}, so that Π1 (W, a) is generated by the two
closed paths γ1 and γ2 illustrated in Fig. 3. We will denote by fα and gα the automorphisms of
Ta defined by
fα (q) = Fγ1 (α, q) and gα (q) = Fγ2 (α, q),
respectively. In particular, Gα is generated by fα and gα . In the next result, we will see how fα
and gα look like.
P ROPOSITION 4. – Let h, Ta = h−1 (a), F̃α , fα and gα be as before. Then:
(a) For all b ∈ W , the fiber Tb = h−1 (b) is biholomorphic to the torus C/Γ, where Γ = 1, j
is lattice generated by 1 and j = e2πi/3 .

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 243

(b) There exists a holomorphic map A : C → C/Γ, such that for every α ∈ C, the global
holonomy group Gα of F̃α in the fiber Ta , written in a suitable holomorphic universal
covering Pα : C → Ta , is generated by fα (z) = j.z and gα (z) = j.z + A(α).
(c) The map A : C → C/Γ is not constant.
Proof. – Since the fibers of h|V are tranverse to F̃α , α = ∞, it follows that all these fibers are
biholomorphic. The fact that they are biholomorphic to C/Γ is a consequence of the following:
A SSERTION. – fα3 = id (resp. gα3 = id ), fa (resp. gα ) has a fixed point, but fα = id (resp.
gα = id ).
It is well-known that an elliptic curve which admits an automorphism satisfying the properties
of fα in the assertion is biholomorphic to C/Γ (cf. [1]).
Proof of the assertion. – We will use (b.2) of Proposition 2. Consider the singular level
h−1 (0) = D1 ∪ ,̃1 ∪ ,̃2 ∪ ,̃3 , where D1 is the divisor Π−1 (0) and ,̃1 , ,̃2 and ,̃3 are the strict
tranforms of (y = x), (y = jx) and (y = j 2 x), respectively. Let q̃k (α) ∈ ,̃k be the singularity
of saddle type of F̃α , such that Π(q̃k (α)) = qk (α) = (j k−1 /α, j 2(k−1) /α) ∈ (y = j k−1 x),
 
k = 1, 2, 3. Since Π is a biholomorphism outside 12 i=1 Di , there are neighborhoods Uk of q̃k (α)

and Uk of qk (α) such that Π|U : Uk → Uk is a biholomorphism. In particular, the foliations F̃α
k

and Fα4 are analytically equivalent in U k and Uk , respectively. For simplicity, let us consider the
case k = 1. In this case, there exists an analytic coordinate system (u, v) : U 1 → C2 such that:
(i) u(q̃1 (α)) = v(q̃1 (α)) = 0 and ,̃1 ∩ U 1 = (u = 0).

(ii) F̃α |U is defined in U1 by the differential equation 3u dv + v du = 0, so that v 3 .u is a
1

holomorphic first integral of F̃α |U .

1
(iii) The maps h|U = H ◦ Π|U and H|U1 are equivalent.
1 1
It follows from (ii) that the holonomy of the local separatrix (v = 0) of q˜1 (α) in a transversal
(u = uo ) (uo = 0), is fo (v) = e−2πi/3 .v = j 2 .v. In particular, we have fo = id and fo3 = id .
This implies that, if Σ ⊂ U 1 is any transversal to (v = 0) through a point qo ∈ (v = 0) and
f : (Σ, qo ) → (Σ, qo ) is the holonomy of (v = 0) in this transversal, then
(iv) f (qo ) = qo , f = id and f 3 = id .
Now, observe that H|U1 = 1 + yx3−x
3 3

−1 , and so h has multiplicity one along the curve ,̃1 , that

is, there exists an analytic coordinate system (W1 , (u1 , v1 )) around (0, 0) such that
(v) W1 ⊂ U 1 , W1 = {(u1 , v1 ); |u1 |, |v1 | < δ}, (u = 0) ∩ W1 = (u1 = 0), v1 = v and
h(u1 , v1 ) = 1 + u1 .
Let us suppose for a moment that |a| < δ and that the curve γ1 is contained in Dδ =
{z; |z − 1| < δ}. In this case, the fiber Ta = h−1 (a) cuts W1 in the transversal to (v1 = 0),
Σ = {(u1 , v1 ); u1 = 1 + a}. Moreover, the global holonomy fα of γ1 , is such that fα |Σ = f −1 ,
where f is the holonomy of the local separatrix (v1 = 0) as in (iv) (observe that γ1 in Fig. 3
is oriented in the negative sense of C). Since fα is holomorphic, (iv) implies that fα = id and
fα3 = id . The proof of the assertion in the general case, that is γ1 ⊂ Dδ , follows from this case.
In fact, it is enough to observe that fα = k −1 ◦ f1 ◦ k, where f1 is the global holonomy of simple
curve γ(s) = 1 + δ2 exp(2πis), s ∈ [0, 1], contained in Dδ and k : h−1 (a) → h−1 ( 1+δ 2 ) is the
global holonomy of a simple curve joining a to 1 + 2 (see Fig. 3). This proves the assertion. ✷
δ

Remark 9. – It follows from Remark 8 that the assertion is also true for α ∈ {1, j, j 2 }.
Let us fix a holomorphic universal covering P : C → Ta , with associated lattice Γ. It follows
from the assertion, that the liftings of fα and gα , which we will denote by the same symbols, can

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

244 A. LINS NETO

be written as
fα (z) = j.z + a(α) and gα (z) = j.z + b(α),
where a : C → C/Γ and b : C → C/Γ are holomorphic (observe that the senses of γ1 and γ2 were
chosen in Fig. 3 in such a way that fα = gα = j). On the other hand, the point zo (α) = a(α) 1−j
is a fixed point of fα , so that kα−1 ◦ fα ◦ kα (z) = j.z, where kα (z) = z + zo (α). Since
kα−1 ◦ gα ◦ kα (z) = j.z + A(α), where A(α) = b(α) − a(α), we obtain that in the covering
Pα = P ◦ kα : C → Ta the generators fα and gα of the global holonomy, are written as in (b).
This proves (b).
In order to prove (c), it is enough to exhibit two foliations F̃α1 and F̃α2 , such that the global
holonomy groups, Gα1 and Gα2 are finite and have a different number of elements. In fact, we
will prove that, in the representation of (b), G1 has three elements (in this case G1 = j.z and
f1 (z) = g1 (z) = j.z) and G0 has at least nine elements, so that A(1) = 0 (mod Γ) and A(0) = 0
(mod Γ).
(c.1) We assert that G1 has three elements.
In fact, as we have seen in (c) of Proposition 2, F14 has a rational first integral, say H1 = pq11 ,
where p1 and q1 are the product of three lines which pass through two of the points of
P2 = {(1, 1), (j, j 2 ), (j 2 , j)}, so that we can take

(x − j 2 )(y − j)(y − j 2 x)
H1 (x, y) = .
(x − j)(y − j 2 )(y − jx)

It follows from Bézout’s Theorem, that the general fiber H1−1 (c) of H1 , cuts H∞ −1
(a) = Π(Ta )
in 3 × 3 = 9 points, counted with multiplicities. Now, observe that, between the twelve points of
P, H1−1 (c) does not pass through the points of P2 and H∞ −1
(a) does not pass through the points
of P1 = {(0, 0), [1 : 0 : 0], [0 : 1 : 0]}. It follows that P ∩ H1−1 (c) ∩ H∞−1
(a) contains six points
−1 −1
of P. As the reader can check, the intersection number of H1 (c) and H∞ (a) in each of these
−1
points is one. This implies that, the general leaf of F̃1 cuts the fiber h (a) = Ta in 3 = 9 − 6
points, so that #G1 = 3. Therefore f1 (z) = g1 (z) = j.z and G1 = j.z.
(c.2) We assert that G0 has at least nine elements, so that G0 = G1 and A(0) = 0 (mod Γ).
First of all observe that
p0 x3 (y 3 − 1)
H0 (x, y) = = 3 3
q0 y (x − 1)
is a rational first integral of F04 . In fact, we have

    
q02 .dH0 = q0 dp0 − p0 dq0 = 3x2 y 2 x x3 − 1 dy − y y 3 − 1 dx ,

which implies that H0 is a first integral of F04 (see (8)).

Now, the line , = (x = 0), which is contained in H0−1 (0), cuts the general level H −1 (a) in
three points, (0, yk ), where y1 , y2 , y3 are cubic roots of 1 − a. Since H0 has multiplicity three
along (x = 0), it follows that, if |b| is small, then H0−1 (b) cuts H −1 (a) in at least nine points,
which are near the points (0, yk ), k = 1, 2, 3. But (0, yk ) ∈
/ P, k = 1, 2, 3, so that if |b| is small,
none of these nine points is in P. This implies that h−1 (a) contains at least nine points of the
generic leaf of F̃0 . Therefore, #(G0 )  9, which proves (c.2) and the proposition. ✷
In the next result, we will see how a group G of the form j.z, j.z + A looks like. We will
also characterize the finite groups of this form.

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 245

P ROPOSITION 5. – Let G be a subgroup of the group of automorphisms of C/Γ, generated by

two elements f (z) = j.z and g(z) = j.z + A. Then
 
(14) G = k; k(z) = c.z + d.A where c ∈ {1, j, j 2} and d ∈ Z ⊕ j.Z .

In particular, the following assertions are equivalent:

(a) G is finite.
(b) There exists n ∈ Z such that nA ∈ Γ.
(c) G has a finite orbit in C/Γ.

Proof. – We will prove (14) and leave the proof of the equivalence between (a), (b) and (c) for
the reader. Set
 
G1 = k; k(z) = c.z + d.A where c ∈ {1, j, j 2 } and d ∈ Z ⊕ j.Z .

It is not difficult to see that G1 is a subgroup of Aut(C/Γ). Since f, g ∈ G1 , it is clear that

G ⊂ G1 .
On the other hand, since f and g have order three and G = f, g, any k ∈ G can be written as
a word

k = f α1 ◦ g β1 ◦ · · · ◦ f αr ◦ g βr ,
where r  1, 0
α1 , βr
2, 1
αm , βn
2, if 2
m
r and 1
n
r − 1. Let us define
the length of the word k as |k| := α1 + · · · + αr + β1 + · · · + βr and prove that any k ∈ G1
is in G, by induction on |k|. If |k| = 0 then k is the identity map, so that k ∈ G1 . Suppose
that for any element k ∈ G with 1
|k| < n we have k ∈ G1 . Let h ∈ G be of length n. Then
h = f ◦ k or h = g ◦ k, where |k| < n. By the induction hypothesis, we have k(z) = c.z + d.A,
where c ∈ {1, j, j 2 } and d ∈ Z ⊕ j.Z. On the other hand, f ◦ k(z) = j.c.z + j.d.A and g ◦ k(z) =
j.c.z + (j.d + 1).A, so that h ∈ G1 , as the reader can check. This finishes the proof of the
proposition. ✷

As a consequence, we obtain the following:

C OROLLARY. – The following assertions are equivalent:

(a) Fα4 has a rational first integral.
(b) There exists n ∈ Z such that nA(α) ∈ Γ, where A(α) is like in Proposition 4.
(c) Fα4 has an algebraic leaf which is not contained in the lines of L.

Proof. – Clearly (a) ⇒ (c). Let us prove that (c) ⇒ (b) ⇒ (a).
(c) ⇒ (b). Let L be an algebraic leaf of Fα4 which is not contained in the lines of L and L̃ be
its strict transform by Π. Since L is algebraic, it follows that L̃ ∩ Ta is finite. But, L̃ ∩ Ta is an
orbit of Gα , so that Gα has a finite orbit. Therefore (c) ⇒ (b).
(b) ⇒ (a). Observe that (b) implies that Gα is finite, say #Gα = m. Therefore, if L ⊂ L
is a leaf of F˜α , then #L ∩ Tb
m, for any fiber Tb = h−1 (b), b ∈ W . On the other hand,
if L does not contain the local separatrices of the singularities qj (α), then L cuts the singular
levels h−1 (c), c = 0, 1, ∞, transversely in the divisors contained in these levels (Proposition 3),
say D1 , D2 and D3 . It is not difficult to see that this implies that #L ∩ Dj
m and that L can
be covered by a finite number of relatively compact disks, so that L is compact. In particular,
Fα4 has infinitely many algebraic solutions and so a meromorphic first integral, by Darboux’s
Theorem. ✷

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

246 A. LINS NETO

We have already proved assertion (a) of the main Theorem for the family (Fα4 )α∈C . In order
to prove (b) we have to define the set E of the statement of the main Theorem. Let
 
E = α ∈ C; there exists n ∈ Z such that n.A(α) ∈ Γ ∪ {∞}.

Since A : C → C(mod Γ) is holomorphic and not constant, it follows that E is countable and
dense. The Corollary of Proposition 5 implies that Fα4 has a rational first integral Fα = Q
Pα
α
if, and
only if, α ∈ E. Let d(α) be the degree of a generic level curve of Fα , say Lc := (Pα − c.Qα = 0),
and L̃c be the strict transform of Lc by Π. By Bézout’s Theorem, we have
 −1
  
3.d(α) = # Lc ∩ H∞ (a)  # L̃c ∩ h−1 (a) = #Gα ,

and so d(α)  13 #Gα . On the other hand, given A ∈ C∗ such that there exists n ∈ Z with
n.A ∈ Γ, let
(15) k = k(A) = min{n ∈ Z; n > 0 and n.A ∈ Γ}.
Then, it is not difficult to see from (14) that #(j.z, j.z + A)  3.k 2 . Therefore, if α ∈ E then
d(α)  k(A(α))2 . Given ko ∈ N, let
   
Eko = α ∈ E; k A(α)
ko ∪ {∞}.

It is not difficult to see that Eko is finite and that, if α ∈

/ Eko , then d(α)  ko2 . This implies (b) of
the main Theorem. It remains to prove (c) for the family (Fα4 )α∈C . We need a definition.
D EFINITION 2. – Let G = j.z, j.z + A. We say that b ∈ C/Γ is generic for G, if for any
f ∈ G \ {id } the point b is not a fixed point of f . We observe that the set of points b ∈ C/Γ
which are not generic for G is countable. In particular, the set of generic points is dense and non
countable in a fundamental domain of Γ.
Let α ∈ C and L be a leaf of Fα4 . We say that L is generic, if
(i) L ⊂ L.
(ii) Let L̃ be a leaf of F̃α such that Π(L̃) = L. Then Ta ∩ L̃ contains a point b which is generic
for Gα .
We will say also that L̃ is generic. We observe that condition (ii) does not depend on the point
b ∈ Ta ∩ L̃. Moreover, for all α ∈ C, the set of generic leaves of F̃α contains a non countable
and dense set of leaves. We would like to observe also that, the leaves of F̃α which contain the
local separatrices of the saddle points of this foliation are not generic (see the proof of (b) of
Proposition 4). Assertion (c) of the main Theorem will follow from the next result.
P ROPOSITION 6. – Let L̃ be a generic leaf of F̃α . Then there exists a holomorphic covering
map π : L̃ → C/Γ. In particular, if α ∈ E then the generic leaf of Fα4 has genus one.
Proof. – We will use (a) of Proposition 4 and the first integral H1 of F14 in the proof of
this proposition. Let L̃1 be a generic leaf of F̃1 . According to (a) of Proposition 4, L̃1 is
biholomorphic to C/Γ. Let V = M \ h−1 {0, 1, ∞} be as before. Denote by L1 and L the
intersections of L̃1 and L̃ with V . We know that h1 := h|L1 : L1 → W and ho := h|L : L → W
are covering maps, where W = C \ {0, 1, ∞}. Moreover, the fiber h−1 1 (z) of a point z ∈ W has
three points in L1 (see (c.1) in the proof of Proposition 4). Let us construct π|L : L → L1 in such
a way that ho = h1 ◦ π.
Fix a point b1 ∈ h−1 −1
1 (a) ⊂ Ta ∩ L1 and a point bo ∈ ho (a) ⊂ Ta ∩ L and set π(bo ) = b1 .
Given m ∈ L, fix a path γ̃m : [0, 1] → L such that γ̃m (0) = bo and γ̃m (1) = m. Let

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 247

γm = ho ◦ γ̃m : [0, 1] → W and γ̂m : [0, 1] → L1 be the lifting of γm by h1 in L1 , such that

γ̂m (0) = b1 . In order to define π(m) it is enough to prove the following:
A SSERTION 1. – The final point of γ̂m , γ̂m (1), does not depend on the path γ̃m . Therefore,
we can define π(m) = γ̂m (1).
Proof. – It is enough to prove the assertion in the case where the path γ̃m is closed, that
is γ̃m (0) = γ̃m (1) = bo . In this case, since L̃ is generic for F̃α , the global holonomy of the
path γm = ho ◦ γ̃m in Gα is the identity. In fact, if we denote this holonomy by k ∈ Gα , then
k(bo ) = γ̃m (1) = bo . Therefore, the definition implies that k = id . Now, write γm as a word in γ1
and γ2 , γm = γ1m1 ◦ γ2n1 ◦ · · · ◦ γ1mr ◦ γ2nr . We have

k(z) = fαm1 ◦ gαn1 ◦ · · · ◦ fαmr ◦ gαnr (z) = c.z + e,

where c = j m1 +n1 +···+mr +nr . Since c = 1, it follows that 3 divides m1 + n1 + · · · + mr + nr .

Now, let K be the global holonomy of γm with respect to the foliation F̃1 . As we have seen
in (c.1) of the proof of Proposition 4, G1 = j.z, j.z, so that K(z) = j m1 +n1 +···+mr +nr .z = z.
This implies that γ̂m is closed and the assertion. ✷
Now, we have to extend π to L̃ ∩ h−1 {0, 1, ∞}. To do this, we first observe that L̃1 ∩ h−1 (c)
contains exactly one point, for all c ∈ {0, 1, ∞}, say p(c). Moreover, L̃1 and L̃ intersect
transversely h−1 (c) (Proposition 3).
In fact, let us consider, for instance, the case c = 1. Since h−1 (1) = D1 ∪ ,̃1 ∪ ,̃2 ∪ ,̃3 and
,̃1 ∪ ,̃2 ∪ ,̃3 is invariant for F̃1 and F̃α , we must have, L̃1 ∩ h−1 (1) ⊂ D1 and L̃ ∩ h−1 (1) ⊂ D1 .
On the other hand, L̃1 = h̃−1 1 (e), where h̃1 = H1 ◦ Π and e ∈ C. In the chart (u, x) ∈ M , such
that Π(u, x) = (x, u.x), we have

(x − j 2 )(ux − j)(u − j 2 ) u − j2
h̃1 (u, x) = H1 ◦ Π(u, x) = ⇒ h̃1 (u, 0) = ,
(x − j)(ux − j )(u − j)
2 u−j
2
−1
and so L̃1 ∩ D1 = ( je−j
1−e , 0) = p(0). Hence, we can define π in L̃ ∩ h (1), by π(q) = p(0) for
all q ∈ L̃ ∩ h (1). In the same way, we can extend π to L̃ ∩ h (0) and L̃ ∩ h−1 (∞), obtaining
−1 −1

a map
π : L̃ → L̃1 ,
−1
which is holomorphic in L = L̃ \ h {0, 1, ∞}.
A SSERTION 2. – π is holomorphic in L̃. Moreover, it is a local homeomorphism in a
neighborhood of each point q ∈ L̃ \ L.
Proof. – Let q ∈ L̃ ∩ h−1 {0, 1, ∞}. We can suppose without lost of generality that q ∈
L̃ ∩ h−1 (1), so that q ∈ L̃ ∩ D1 . Let us consider the case where q is in the domain of the chart
(u, x) in which the blowing up at 0 ∈ C2 is written as Π(u, x) = (x, ux). In this chart, we have
2
q = (uo , 0), p(0) = (ao , 0) = ( je−j
1−e , 0) and

x3 (1 − u3 )
(16) h(u, x) − 1 = .
1 − x3

Since L̃ and L̃1 intersect transversely D1 we can parametrize these leaves in a neighborhood
of q and p(0) as x → (u(x), x) and x → (u1 (x), x) respectively, where u, u1 : D(0, r) → C are
holomorphic and uo = u(0), ao = u1 (0). It follows from (16) that there are local charts z and w

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

248 A. LINS NETO

in L̃ and L̃1 , around q and p(0), respectively, such that z(q) = w(p(0)) = 0 and ho (z) = 1 + z 3
and h1 (w) = 1 + w3 . Since ho = h1 ◦ π outside D1 ∩ L̃, we get
  3  
(17) z 3 = w π(z) ∀z = 0 ⇒ w π(z) = δ.z where δ 3 = 1.

Clearly (17) implies the assertion. ✷

Now, in order to prove that π : L̃ → L̃1 is a covering map, we use the so-called “property of
the unique lifting of paths” (cf. [16]):
“Let f : X → Y be a local homeomorphism, where X and Y are connected manifolds. Then,
f is a covering map if, and only if, it has the property of the unique lifting of paths, that is, given
a path γ : [0, 1] → Y such that γ(0) = yo ∈ Y and xo ∈ X such that f (xo ) = yo , then there exists
a unique path γ̃ : [0, 1] → X such that γ̃(0) = xo and f ◦ γ̃ = γ.”
It is not very difficult to verify that π : L̃ → L̃1 satisfies the property of the unique lifting of
paths. We leave the proof of this fact for the reader. This proves Proposition 6. ✷
Let us prove assertion (c) of the main Theorem for the family (Fα4 )α∈C . Let α ∈ / E and L̃ be
a generic leaf of F̃α . Let π : L̃ → L̃1  C/Γ be the covering of Proposition 6. Since L̃ is not
compact (see (c) of the Corollary of Proposition 5), we must have L̃  C or L̃  C∗ . Since the
holomorphic universal covering of C∗ is C, in both cases there exists a holomorphic non-constant
map f˜: C → L̃. Hence f = Π ◦ f˜ : C → L = Π(L̃) is a non-constant holomorphic map such that
f (C) ⊂ L ∪ sing(Fα4 ), which proves (c) of the main Theorem.
It remains to prove that a generic leaf of F̃α is biholomorphic to C, as asserted in Remark 3.
Let us consider the automorphisms of the covering π; Aut(π) = {k ∈ Aut(L̃); π ◦ k = π}. We
have two possibilities: either (i) L̃  C∗ and Aut(π)  |Z, or (ii) L̃  C and Aut(π)  Z2 . We
will prove that Aut(π)  Z2 .
First of all, fix a point b1 ∈ L̃1 and a point b ∈ L̃, such that h(b1 ) = h(b) = a ∈ W =
C \ {0, 1, ∞}. Recall that for any class of closed curve [δ] ∈ Π1 (L̃1 , b1 ) there exists kδ ∈ Aut(π)
such that, if δ̃ is the lifting of d by π such that δ̃(0) = b, then δ̃(1) = kδ (b). Now, consider the
curves γ2 ◦ γ12 and γ1 ◦ γ2 ◦ γ1 in W . The lifting of each one of these curves by h1 in L̃1 is a
closed curve, because g1 ◦ f12 = f1 ◦ g1 ◦ f1 = idTa . Let us denote by δ1 and δ2 , respectively,
these liftings (where δ1 (0) = δ2 (0) = b1 ), by δ̃1 and δ̃2 the liftings of δ1 and δ2 by π (where
δ̃1 (0) = δ̃2 (0) = b) and by k1 and k2 the associate elements of Aut(π). Since ho = h1 ◦ π, δ̃1
and δ̃2 coincide with the liftings of γ2 ◦ γ12 and γ1 ◦ γ2 ◦ γ1 , respectively, by h1 . It follows that
 
k1 (b) = δ̃1 (1) = gα ◦ fα2 (b) = b + A mod(Γ) and
 
k2 (b) = δ̃2 (1) = fα ◦ gα ◦ fα (b) = b + j.A mod(Γ) ,
where A = A(α). Now, suppose by contradiction that Aut(π)  Z. If k is a generator of Aut(π),
then k1 = k m and k2 = k n for m, n ∈ Z, so that k1n = k2m . But, this implies that
   n  
b + n.A mod(Γ) = gα ◦ fα2 (b) = (fα ◦ gα ◦ fα )m (b) = b + mj.A mod(Γ)

so that
 
(n − mj).A ∈ Γ ⇒ n2 + m2 + mn A ∈ Γ
as the reader can check. This contradiction finishes the proof of Remark 3.

2.3. The family of degree three

In order to construct the family of degree three, we observe that the equations in (8) and (9),
which define the family (Fα4 )α∈C , are symmetric with respect to the change of variables

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 249

S(x, y) = (y, x). It is possible to prove that any foliaton with this property is the pull-back of
another foliation by the map T (x, y) = (x + y, x.y). In our case, the vector field which defines
α = X
the foliation Fα4 is Z  − αY , where
      ∂   ∂
 = x x3 − 1 ∂ + y y 3 − 1 ∂
X and Y = y 2 x3 − 1 + x2 y 3 − 1 .
∂x ∂y ∂x ∂y
 = T ∗ (X) and Y = T ∗ (Y ), where
As the reader can check, we have X

  ∂   ∂
X(u, v) = −u + 2v 2 − 4u2 v + u4 + v −2 − 3uv + u3 ,
∂u ∂v

and
  ∂   ∂
Y (u, v) = 2v − u2 + uv 2 + 3uv − u3 + 2v 3 ,
∂u ∂v
so that Zα = T ∗ (Zα ), where Zα = X − αY .
Observe that, although the vector field Zα has degree four, its homogeneous part of degree
∂ ∂ ∂ ∂
four is u4 ∂u + v u3 ∂v = u3 (u ∂u + v ∂v ), so that the foliation defined by Zα has degree three.
We will denote this foliation by Fα . The foliation F∞
3 3
will be the foliation defined by the vector
field Y , which is also of degree three.
In order to study the foliation Fα3 , let us see first how the rational map T transforms the lines
of the configuration C. Observe that:
(I) The lines (x = j k ) and (y = j k ) are sent by T into the line (v = j k u − j 2k ), k = 0, 1, 2.
This implies that the foliation Fα3 has three invariant lines; ,k := (v = j k u − j 2k ),
k = 0, 1, 2.
(II) The line (y = x) is sent by T into the conic C1 := (v = 14 u2 ).
(III) The lines (y = j x) and (y = j 2 x) are sent by T into the conic C2 := (v = u2 ).
It follows from (I), (II) and (III) that, for all α ∈ C, Fα3 leaves invariant an algebraic curve of
degree 7, which is the union of five rational curves: three lines, ,k , k = 0, 1, 2, and two conics
C1 and C2 . The conics are tangent in the points q1 = [0 : 0 : 1] and q2 = [0 : 1 : 0]. The lines ,k ,
k = 0, 1, 2, intersect in the points p01 = [−j 2 : j : 1] ∈ ,0 ∩ ,1 , p02 = [−j : j 2 : 1] ∈ ,0 ∩ ,2
and p12 = [−1 : 1 : 1] ∈ ,1 ∩ ,2 . Observe that p01 , p02 , p12 ∈ C2 . Finally, the lines are tangent
to the conic C1 in the points p0 = [2 : 1 : 1] ∈ ,0 ∩ C1 , p1 = [2j : j 2 : 1] ∈ ,1 ∩ C1 and
p2 = [2j 2 : j : 1] ∈ ,2 ∩ C1 . In Fig. 4, we sketch the five curves and the points.
Since the five curves are invariant for Fα3 , for all α ∈ C, it follows that all points in the
intersections of the curves are singularities for Fα3 . As we will see, if α ∈ / {0, 1, j, j 2, ∞}, then
the foliation Fα has five more singularities, one in each curve of the configuration sketched in
3

Fig. 4. In the next result we precise how the family of foliations (Fα3 )α∈C looks like.
P ROPOSITION 7. – Assertions (a), (b) and (c) of the main Theorem are true for the family
(Fα3 )α∈C . Moreover, the set A3 of (a) is {0, 1, j, j 2, ∞} and if α ∈
/ A3 then the singularities of
Fα are of the following local analytic types:
3

(a.1) The eight points in the intersection of the five curves are dicritical; p01 , p02 and p12 are
of radial type and in a neighborhood of each of the points q1 , q2 , p0 , p1 and p2 , Fα3 has
a meromorphic first integral of the type z 2 /w = cte.
(a.2) Each of the five curves contains another singularity, say P1 (α) ∈ C1 , P2 (α) ∈ C2
and Qk (α) ∈ ,k , k = 0, 1, 2. Moreover, Fα3 has a local holomorphic first integral in
a neighborhood of each of these singularities, of the following types: z 6 .w = cte in a
neighborhood of P1 (α) ∈ C1 and z 3 .w = cte in a neighborhood of the others.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

250 A. LINS NETO

Fig. 4.

Proof. – We observe that most of the above assertions follow from the analysis that we have
done for the family Fα4 . So, for instance, the fact that if α ∈ E then Fα4 has a first integral,
implies that all leaves of Fα3 are algebraic, and so it has a rational first integral, by Darboux’s
Theorem.
Let us prove that A3 = {0, 1, j, j 2, ∞} and assertions (a.1) and (a.2). Observe first that
the eight intersection points of the five curves are singularities of Fα3 . Moreover, in each of
the five invariant curves, Fα3 has a unique singularity contained in T {q1 (α), . . . , q9 (α)} (see
Proposition 2). These singularities are:
   
1 1 2 1
T , = , := P1 (α) ∈ C1 ,
α α α α2
     
j j2 j2 j −1 1
T , =T , = , := P2 (α) ∈ C2 ,
α α α α α α2

T (1, α) = T (α, 1) = (1 + α, α) := Q0 (α) ∈ ,0 ,

     
T j, j 2 α = T j 2 α, j = j + j 2 α, α := Q1 (α) ∈ ,1
and
     
T j 2 , jα = T jα, j 2 = j 2 + jα, α := Q2 (α) ∈ ,2 .
If α ∈/ {0, 1, j, j 2, ∞}, then these singularities do not coincide with the points in the intersections
of the curves. Therefore, in this case, Fα3 has 8 + 5 = 13 = 32 + 3 + 1 distinct singularities, which
implies that all singularities are non-degenerate.
In order to prove (a.1) and (a.2), we will study the map T near the points of sing(Fα4 ). In
homogeneous coordinates, we can write T (X, Y, Z) = (Z(X + Y ), X.Y, Z 2 ) = (U, V, W ). It
follows that T is not defined in the points [1 : 0 : 0] and [0 : 1 : 0]. Moreover, the singular set
of T is S = (Y = X) ∪ (Z = 0), where (Y = X) is the ramification set of T and T (Z = 0) =
[0 : 1 : 0] ∈ C1 ∩ C2 . Remark also that:
(i) T (Y = X) = C1 .
(ii) If m ∈ CP(2) \ C1 , then T −1 (m) contains two points, say m1 and m2 . Moreover T is a
local biholomorphism in a neighborhood of mk , k = 1, 2.

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 251

(iii) If m ∈ C1 \ {[0 : 1 : 0]} = {(u, v) ∈ C2 ; v = 14 u2 }, say mo = (uo , 14 u2o ), then

T −1 (mo ) = {( 21 uo , 12 uo )} = {m}. Moreover, there exists a holomorphic coordinate
system, (W1 , (z1 , w1 )) in a neighborhood of m and (W2 , (z2 , w2 )) in a neighborhood of
mo , such that z1 (m) = w1 (m) = 0, z2 (mo ) = w2 (mo ) = 0, W1 ∩ (Y = X) = (w1 = 0),
W2 ∩ C1 = (w2 = 0) and T (z1 , w1 ) = (z1 , w12 ).
It follows from (ii) that for any singularity p of Fα3 , which is not in C1 , there is a singularity q
of Fα4 such that T (q) = p. Moreover, Fα3 and Fα4 are locally equivalent at p and q. Using this
fact, it is not difficult to see that:
(iv) p01 , p02 and p12 are radial singularities for Fα3 .
(v) Fα3 has a local holomorphic first integral of the type z 3 .w = cte, in a neighborhood of
each one of the singularities P2 (α), Q0 (α), Q1 (α) and Q2 (α).
The next result will allow us to study the dicritical singularities of Fα3 which are contained
in C1 .
∂ ∂
L EMMA 1. – Let Z(x, y) = p(x, y) ∂x + q(x, y) ∂y be a holomorphic vector field defined in a
neighborhood of 0 ∈ C . Suppose that:
2

(a) 0 is a non-degenerate singularity of Z and the characteristic numbers of Z at 0 are

rational and positive, say p/q and q/p, where p, q ∈ N are relatively primes.
(b) Either p, q  2 or Z has at least two distinct local analytic separatrices through 0.
Then there exists a holomorphic coordinate system (W, (u, v)) with 0 ∈ W , u(0) = v(0) = 0,
in which Z can be written as
 
∂ ∂
Z(u, v) = k q.u + p.v ,
∂u ∂v

up
where k ∈ C∗ . In particular, vq is a meromorphic first integral of Z in a neighborhood of 0.
Proof. – If p, q  2 then Z has no resonance at 0 and the Poincaré’s linearization Theorem
implies that Z is linearizable (cf. [2]). Therefore, we can suppose that p = 1 and q  1. In this
case, Dulac’s normal form (cf. [2]) implies that in a suitable coordinate system (W, (u, v)) such
that 0 ∈ W and u(0) = v(0) = 0, the vector field Z can be written as
 
∂   ∂
Z(u, v) = k u + qv + b.uq ,
∂u ∂v

where k ∈ C∗ . Let us prove that hypothesis (b) implies that b = 0. Observe that, if this is the
case, then v/uq is a local meromorphic first integral of Z.
Suppose by contradiction that b = 0. In this case, the function F : C2 \ (u = 0) → C given by
 
v
F (u, v) = u. exp − q ,
b.u

is a first integral of Z, as the reader can check. Observe that Z has a holomorphic local analytic
separatrix through 0: (u = 0). If it had another local separatrix through 0, then it would be
contained in a level curve of F , say (F = c), where c = 0. But (u = 0) is an essential singularity
of F , and so these level curves are not analytic at 0 ∈ C2 . ✷
∂ ∂
C OROLLARY. – Let Z(x, y) = p(x, y) ∂x + q(x, y) ∂y be a holomorphic vector field defined
in a neighborhood of 0 ∈ C . Suppose that 0 is a non-degenerate singularity of Z and that Z
2

has two smooth invariant separatrices through 0, with a tangency of order two at 0. Then the

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

252 A. LINS NETO

characteristic numbers of Z at 0 are 2 and 1/2 and Z has a local meromorphic first integral of
the form z 2 .w = cte.
Proof. – It is sufficient to prove that the characteristic numbers of Z at 0 are 2 and 1/2. Let
γ1 , γ2 be the smooth invariant separatrices through 0. Since they have a tangency of order two
at 0, there is a holomorphic coordinate system (W, (u, v)) such that 0 ∈ W , u(0) = v(0) = 0,
γ1 ∩ W = (v = 0) and γ2 ∩ W = (v − u2 = 0). Since (v = 0) is invariant for Z, in this coordinate
system Z can be written as

  ∂   ∂
Z(u, v) = au + bv + r1 (u, v) + v c + r2 (u, v) ,
∂u ∂v
∂ ∂
where the linear part of Z at 0 is (au + bv) ∂u + c.v ∂v and the characteristic numbers are c/a,
a/c. Now, since γ2 is invariant for Z, we must have
       
Z v − u2 = −2u au + bv + r1 (u, v) + v c + r2 (u, v) = f (u, v). v − u2 ,

where f is holomorphic. By comparing the two jets of −2u(au + bv + r1 (u, v)) and f (u, v).
(v − u2 ) at 0, it is not difficult to see that c = f (0, 0) and that f (0, 0) = 2a. Therefore
c/a = 2. ✷
2
The above result implies that Fα3 has a meromorphic first integral of the form zw in a
neighborhood of each one of the singularities p0 , p1 , p2 , q1 and q2 , because each of those
singularities belongs to C1 and to another one of the four other invariant curves, and C1 has
a tangency of order two with the curve at the point (see Fig. 4).
We have already proved (a.1) of Proposition 7. In order to finish the proof of (a.2), it remains
to prove that Fα3 has a holomorphic first integral of the form z 6 .w = cte in a neighborhood
of P1 (α) ∈ C1 . We will use (iii). Since T −1 (P1 (α)) = q1 (α) ∈ (Y = X), there exists a holo-
morphic coordinate system, (W1 , (z1 , w1 )) in a neighborhood of m = q1 (α) and (W2 , (z2 , w2 ))
in a neighborhood of mo = P1 (α), such that z1 (m) = w1 (m) = 0, z2 (mo ) = w2 (mo ) = 0,
W1 ∩ (Y = X) = (w1 = 0), W2 ∩ C1 = (w2 = 0) and T (z1 , w1 ) = (z1 , w12 ). We assert that
the characteristic numbers of Fα3 at P1 (α) are −6 and −1/6.
Observe first that Fα3 can be represented in the coordinate system (z2 , w2 ) by a differential
equation of the form
 
(18) az2 + bw2 + r(z2 , w2 ) dw2 + 6w2 dz2 = 0,

where the characteristic numbers are −6/a and −a/6, because (w2 = 0) is invariant for Fα3 .
Since T (z1 , w1 ) = (z1 , w12 ), we get from (18) that Fα4 is represented in a neighborhood of q1 (α)
by
  
az1 + bw12 + r z1 , w12 2w1 .dw1 + 6w12 dz1 = 0
  
=⇒ az1 + bw12 + r z1 , w12 dw1 + 3w1 dz1 = 0.
By using the local form of Fα4 in a neighborhood of ( α1 , α1 ) = q1 (α) (see (b.2) of Proposition 2),
it follows that a = 1, which proves that the characteristic numbers of Fα3 at P1 (α) are −6
and −1/6. Moreover, after a linear change of coordinates, we can suppose that the linear
part in (18) is z2 dw2 + 6w2 dz2 = 0. We want to prove that the foliation Fα3 is linearizable
in a neighborhood of P1 (α), which means that Fα3 can be represented by z dw + 6w dz = 0,
after a holomorphic local change of coordinates. In order to prove this fact we use the
following:

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 253

(vi) Fα4 has a holomorphic first integral in a neighborhood of q1 (α) (Proposition 2).
(vii) If there is a neighborhood V of P1 (α) such that the leaves of the restricted foliation
Fα3 |V are closed in V , then Fα3 has a holomorphic first integral in (a possibly smaller)
neighborhood U of P1 (α) (cf. [18]). In particular, Fα3 is linearizable in U .
Since Fα4 = T ∗ (Fα3 ), it is clear that (vi) ⇒ (vii). This finishes the proof of (a.2).
Let us prove that the family (Fα3 )α∈C satisfies (b) of the main Theorem. First of all, observe
that the rational map
(u3 − 3uv)2 − 4v 3
R(u, v) =
v 3 − u3 + 3uv + 1
is a first integral of F∞
3
, because

 y) = R ◦ T (x, y) = (y 3 − x3 )2 H2
R(x, = ,
(y 3 − 1)(x3 − 1) H + 1

where H = H(x, y) = yx3−x

3 3

−1 is a first integral of F∞ . Let α ∈ E. As we have seen in the

4

beginning of the proof, Fα3 has a rational first integral, say Rα , which we suppose primitive (that
is, with irreducible generic level curve) and of degree dα . It follows from Bézout’s Theorem,
that the generic level (Rα = c1 ) cuts the generic level (R = c) in 6dα points, counted with
multiplicities. On the other hand, the foliations F∞ 3
and Fα3 are transversal outside the curve
3
C = k=0 ,k ∪ C1 ∪ C2 . This implies that the intersection number of (R = c) and (Rα = c1 ) in
each point of (R = c) ∩ (Rα = c1 ) \ C is one. Let kα = #((R = c) ∩ (Rα = c1 ) \ C), so that
6dα > kα . Now, observe that, T −1 ((R = c) ∩ (Rα = c1 ) \ C) is contained in the intersection
of two leaves of F∞ 4
with at most two algebraic leaves of Fα4 , but outside the set of lines L. It
follows from (ii) above, that 2kα = #(T −1 ((R = c)∩(Rα = c1 )\C))  2#(Gα )  2k(A(α))2 .
Therefore, 6dα  kα  k(A(α))2 , where k(A) is like in (15). This proves that the family
(Fα3 )α∈C satisfies (b) of the main Theorem.
Now, let us see how the resolution of the elements of the family looks like. We have seen that
Fα3 has eight dicritical singularities, three radial and five with local meromorphic first integrals of
2
the type zw = cte. The resolution of each of these five singularities is done with two blowing-ups,
as sketched in Fig. 5.
In Fig. 5, π1 and π2 are the two blowing-ups, D1 and D2 the two divisors, and m is a
singularity of π ∗ (Fα3 ), in which this foliation has a local holomorphic first integral of the type
z 2 .w = cte, where π = π1 ◦ π2 .
We observe that D1 is invariant for π ∗ (Fα3 ), whereas D2 is not. It follows that the resolution of
Fα involves a total of 13 blowing-ups, one at each radial singularity and two at each singularity
3

as in Fig. 5. We call M1 the manifold obtained from CP(2) after these blowing-ups, and
Π1 : M1 → CP(2) the blowing-up map. We denote by F̂α the strict transform of Fα3 by Π1 .

Fig. 5.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

254 A. LINS NETO

Let T1 : M − → M1 be the map which makes the diagram below commute:

T1
M M1
Π Π1

CP(2) CP(2)
T

where Π : M → CP(2) is the resolution of the foliation Fα4 .

Now, let us prove that the family satisfies (c) of the main Theorem and Remark 3 of §1. Let
α∈ / E. If L is a leaf of F̃α such that L ⊂ L and L1 = T1 (L), then L1 is a leaf of F̂α and
the restriction T1 |L : L → L1 extends to a holomorphic map TL : L → L1 , which is a ramified
covering of degree at most two (see (ii) in the proof of Proposition 7). Moreover, since L is
a transcendent leaf of F̃α , then L1 is also transcendent. When L is a generic leaf of F̃α , then
L  C and we have a ramified covering TL : C  L → L1 of degree at most two. This implies
that L1 is biholomorphic to C and finishes the proof of Proposition 7. ✷

2.4. The family of degree two

The family of degree two is obtained from Fα3 by a Cremona transformation, as illustrated in
Fig. 6.
In this figure, we denote by π1 : N → CP(2) the blowing-up at the three points p01 , p02
and p12 . After this blowing-up process, we obtain three divisors, Dij = π1−1 (pij ), not invariant
for the strict transform π1∗ (Fα3 ), because p01 , p02 and p12 are radial singularities (α ∈ /
{0, 1, j, j , ∞}). Moreover, the strict transforms of ,0 , ,1 and ,2 , say ,̂0 ,̂1 and ,̂2 , have
2

self-intersection −1 in N , so that we can blow down these three curves, obtaining another
surface, say N1 , with three points, J, K, L ∈ N1 and blowing-up map π2 : N → N1 , such that
π2−1 {J, K, L} =  ,̂0 ∪ ,̂1 ∪ ,̂2 . The map π = π2 ◦ (π1 )−1 is well defined, holomorphic and
injective outside j ,j . This type of blowing-up-blowing-down process is known in the literature
as a “Cremona transformation”. It is well-known that the surface N1 , obtained after a Cremona
transformation in CP(2), is again CP(2).
The curve indicated by C 1 in Fig. 6.2 is the strict transform of the curve C1 . This curve is
sent by π2 to the curve Q of Fig. 6.3. This curve has degree four. In fact, the conics through the
base points pij of π are sent by π to straight lines. Since a generic conic passing through these
points cuts C1 in four points, its image by π has intersection number four with Q, so that Q has
degree four. Observe that the curve C2 is conic passing through the base points of π, therefore it
is transformed by π into a line, indicated by R in Fig. 7. The quadric Q has three cuspidal points,

Fig. 6.

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 255

Fig. 7.

indicated by J , K and L in Fig. 6.3 and in Fig. 7. The line R and the curve Q are tangent in two
points indicated by M and N . Observe that M = π(q1 ) and N = π(q2 ) (or vice-versa), where q1
and q2 are the tangent points between C1 and C2 .
Let Fα2 be the strict transform of Fα3 by π −1 . We will see that the foliation Fα2 has degree two
for all α ∈ C, but first let us study the singularities of Fα2 for α ∈
/ {0, 1, j, j 2, ∞}.
We begin by observing that the foliation Fα has four non-degenerate singularities outside
3
2
k=0 ,k namely q1 , q2 ∈ C1 ∩ C2 , P1 (α) ∈ C1 and P2 (α) ∈ C2 (see Fig. 4). These singularities
are sent by π into singularities of the same analytic type, because π is a biholomorphism
2
outside k=0 ,k . After the three blowing-ups π1 , each of the invariant curves ,̂k contains
two non-degenerate singularities, pk and qk (α). This implies that after the blowing-down π2 ,
these curves are contracted in the three points J , K and L, which are singularities of Fα2 .
Therefore, the singularities of Fα2 are J , K, L and the four non-degenerate singularities
π{q1 , q2 , P1 (α), P2 (α)}. We will use the same symbols to denote Pj (α) and π(Pj (α)), j = 1, 2.
Since π(q1 ) = M , π(q2 ) = N and π(P2 (α)) = P2 (α) are non-degenerate, belong to line R
and this line contains no other singularity of Fα2 , it follows that this foliation has degree two
(see [17]). This implies that J , K and L are also non-degenerate because if a foliation of degree
two has 7 = 22 + 2 + 1 singularities,
2 then these singularities are non-degenerate. Since π is a
biholomorphism outside k=0 ,k , we have already proved that Fα2 has the following local first
integrals (α ∈ / {0, 1, j, j 2, ∞}):
2
(i) Meromorphic of the type zw = cte in neighborhoods of M and N .
6
(ii) Holomorphic of the type z .w = cte in a neighborhood of P1 (α).
(iii) Holomorphic of the type z 3 .w = cte in a neighborhood of P2 (α).
In order to prove that the family has singularities of fixed analytic type, we will prove that:
3
(iv) If α ∈ / {0, 1, j, j 2, ∞}, then Fα2 has a local meromorphic of the type wz 2 = cte in
neighborhoods of J , K and L.
This follows from the fact that the characteristic numbers of Fα2 at J , K and L are 2/3 and
3/2. In fact, the curve Q has cuspidal singularities at these points, that is in a convenient local
chart (u, v) near one of them, Q has an equation of the form v 2 − u3 = 0. As the reader can
check, since Fα2 is non-degenerate and Q is invariant, this implies that the characteristic numbers
are 2/3 and 3/2. Now, the Poincaré’s linearization Theorem implies that Fα2 is linearizable near
J , K and L, so that it has a local meromorphic first integral like in (iv).

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

256 A. LINS NETO

We have proved that the family (Fα2 )α∈C\{0,1,j,j 2 ,∞} is a family of degree two with
singularities of fixed analytic type. Since this family is obtained from (Fα3 )α∈C by a Cremona
transformation, it is not difficult to see that it satisfies also (b) and (c) of the main Theorem. We
leave the details for the reader.
Remark 10. – We would like to observe that in a suitable affine coordinate system, the
foliation Fα2 is defined by the following differential equations:
 dx  dx
dt = 4x − 9x2 + y 2 + α(2y − 4xy), dt = 2y − 4xy,
Fα2   or F∞
2
 
dy
dt = 6y − 12xy + 3α x2 − y 2 , dy
dt = 3 x2 − y 2 .

In this coordinate system, the line R is the line at infinity and the quartic Q is given by
 2
4y 2 (1 − 3x) − 4x3 + 3x2 + y 2 = 0.
2
The finite singularities are J = (1/2, −1/2), K = (0, 0), L = (1/2, 1/2) and P1 (α) = ( 4(α +1)
(α2 +3)2 ,
−8α
√ √
(α2 +3)2 ) ∈ Q and the singularities at the line R are M = [1 : 3 i : 0], N = [1 : − 3 i : 0] and
P2 (α) = [1 : α : 0].

3. Proof of the Corollary

The families of degree greater than four are obtained by pulling back one of the families of §2.
The idea is that, if f : S → T is a non-constant holomorphic map between two compact Riemann
surfaces S and T , where T is an elliptic curve and f has k points of ramification, then

(22) X (S)
−k,

where X (S) is the Euler characteristic of S. We observe that (22) follows from Riemann–
Hurwitz formula. In fact, if the ramification points are z1 , . . . , zk and the ramification number
of f at zj is nj  2, j = 1, . . . , k, then the Riemann–Hurwitz formula says that,


k 
k
X (S) = dg(f ).X (T ) − (nj − 1) = − (nj − 1)
−k,
j=1 j=1

where dg(f ) is the topological degree of f .

3.1. The families of degree 3r − 1, r  2

Let us sketch how to obtain a family of degree 3r − 1, r  2, from the family of degree
two. Consider homogeneous coordinates (X, Y, Z) on C3 such that the invariant line R for all
Fα2 is the line (Z = 0). Let F : CP(2) → CP(2) be a rational map which in this homogeneous
coordinates in the target, is written as
 
(23) F (U, V, W ) = P (U, V, W ), Q(U, V, W ), W r ,

where P and Q are homogeneous polynomials of degree r. We make the following generic
assumptions on P and Q:

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 257

(i) {(U, V ); P (U, V, 0) = Q(U, V, 0) = 0} = {(0, 0)}. This condition implies that F is well
defined.
(ii) Let ∆˜ = ∂P ∂Q − ∂P ∂Q and ∆ = Π(∆ ˜ = 0), where Π : C3 \ {0} → CP(2) is the
∂U ∂V ∂V ∂U
canonical projection. Then ∆ is a smooth curve of degree 2r − 2, transversal to the
line L∞ = Π(W = 0).
(iii) If po ∈ ∆ \ L∞ , then dimC (ker(dF (po )) = 1.
(iv) The set P = {po ∈ ∆ \ L∞ ; Tpo ∆ = ker(dF (po ))} is finite.
(v) If po ∈ ∆ \ (P ∪ L∞ ) and γ : (C, 0) → (CP(2), po ) is a curve such that
0 = γ  (0) ∈ ker(dF (po )), then (F ◦ γ) (0) = 0.
In this case, there are holomorphic coordinate systems (W1 , (u, v)) and (W2 , (x, y)) around
po and F (po ), respectively, such that u(po ) = v(po ) = 0, x(F (po )) = y(F (po )) = 0 and
F (u, v) = (u, v 2 ). In these coordinates, we have ∆ ∩ W1 = (v = 0) and F (∆) ∩ W2 = (y = 0).
(vi) If po ∈ P then there are holomorphic coordinate systems (W1 , (u, v)) and (W2 , (x, y))
around po and F (po ), respectively, such that u(po ) = v(po ) = 0, x(F (po )) = y(F (po ))
= 0 and F (u, v) = (u, 13 v 3 − uv). In these coordinates, we have ∆ ∩ W1 = (u = v 2 )
and F (∆) ∩ W2 = (9y 2 = 4x3 ).
(vii) Let C = F (∆). Then C has two types of singularities: the points in F (P ), which are
singularities of cuspidal type; and singularities of nodal type. If qo ∈ C is a singularity
of nodal type, then F −1 (qo ) ∩ ∆ contains two points, which are not on P .
Assumptions (ii)–(vii) are called Whitney’s conditions. It is well-known that they are generic
in the set of maps as in (23). We impose another condition in the points of ∆ ∩ L∞ . Observe
that the critical points of the map F |L∞ : L∞ → L∞ , F [u : v : 0] = [P (u, v, 0) : Q(u, v, 0) : 0],
coincide with the critical points of the map [P : Q].
(viii) The critical points of [P : Q] : C → C are of order two.
Let us denote by Cr the set of critical points of F |L∞ . We observe that Cr = L∞ ∩ ∆. This
follows from the fact that,
 
∂Q ∂P
po = [1 : vo : 0] ∈ Cr ⇔ 0 = r P. − Q. (po )
∂v ∂v
∂P ∂Q ∂P ∂Q
= (po ) (po ) − (po ) (po ),
∂u ∂v ∂v ∂u
where in the second equality, we have used Euler’s identity. It is not difficult to see that, if
po ∈ Cr then there are holomorphic coordinate systems (W1 , (v, w)) and (W2 , (x, y)) around
po and F (po ), respectively, such that, v(po ) = w(po ) = 0, x(F (po )) = y(F (po )) = 0 and
F (v, w) = (v 2 + f (w), wr ), where f is holomorphic.
Now, we take the quartic Q, which is invariant for all Fα2 , in such a way that:
(ix) The points M and N of tangency between L∞ = R and Q are not in C (see Fig. 7).
(x) C cuts Q transversely, so that the cuspidal points of Q (J , K and L) are not in C ∪ L∞
and the singular points of C are not in Q ∪ L∞ .
Let us consider the family of foliations (Gα )α∈C , where Gα = F ∗ (Fα2 ).

L EMMA 2. – There exists a finite subset F0 ⊂ C, such that the family (Gα )α∈C\F0 is non-
degenerate with singularities of fixed local analytic type. Moreover, the degree of any Gα is
3r − 1.

Proof. – Let P1 (α) ∈ Q and P2 (α) ∈ R be as in (ii) and (iii) of §2.4. Observe that the maps
α → P1 (α) and α → P2 (α) are 1–1 parametrizations of Q and R, respectively. We have seen
that, if α ∈
/ F1 := {0, 1, j, j 2, ∞}, then all singularities of Fα2 are non-degenerate. This means

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

258 A. LINS NETO

that P1 (F1 ) = {J, K, L, M, N } and P2 (F1 ) ⊃ {M, N }. Define

 
F2 = F1 ∪ α ∈ C; P1 (α) ∈ C or P2 (α) ∈ C .

It is not difficult to see that F2 is finite. Now, observe that the topological degree of F is r2 and
the set of critical values of F is C ∪ R. This implies that, if p ∈ CP(2) \ (C ∪ R), then F −1 (p)
contains r2 points. Moreover, for any q ∈ F −1 (p), F is a biholomorphism in a neighborhood
of q. It follows that, if α ∈/ F2 , then Gα has:
(a) 3r2 non-degenerate singularities with a local meromorphic first integral of the type
z3 −1
w 2 = cte, namely, the points in F (J) ∪ F −1 (K) ∪ F −1 (L).
2
(b) r non-degenerate singularities with a local holomorphic first integral of the type
z 6 .w = cte, namely, the points in F −1 (P1 (α)).
Now, the topological degree of F |L∞ : L∞ → R is r. This implies that if p ∈ R \ C,
then F −1 (p) contains r points, all of them in L∞ \ ∆. It follows that if α ∈ / F2 , then
F −1 (M ) ∪ F −1 (N ) contains 2r points, and F −1 (P2 (α)) r points. Let us study the singularities
of Gα in F −1 {M, N, P2 (α)}.
Observe that, if q ∈ L∞ \ ∆, then there are holomorphic coordinate systems (W1 , (v, w)) and
(W2 , (x, y)) around q and p = F (q), respectively, such that v(q) = w(q) = 0, x(p) = y(p) = 0,
W1 ∩ L∞ = (w = 0), F (v, w) = (v, wr ) and R ∩ W2 = L∞ ∩ W2 = (y = 0). Let us see how the
foliation Gα looks like in a neighborhood of a point q ∈ L∞ \ ∆. Since R = (y = 0) is invariant
for Fα2 , this foliation can be defined in W2 by a differential equation of the form

(24) η = y dx − (a + bx + cy + h.o.t) dy = 0,

where a = 0 if, and only if p is a singular point of Fα2 . If α ∈

/ F2 and p is a singular point of Fα2 ,
then a = 0 and b = 0 (because p is non-degenerate). It follows from (24) that Gα is defined in
W1 by
F ∗ (η) = wr dv − (a + bv + h.o.t.)rwr−1 dw = 0
⇒ w dv − r(a + bv + h.o.t.) dw = 0.
This implies that, if α ∈/ F2 and p is not a singularity of Fα2 , then q is not a singularity of Gα .
Moreover, if p is a singularity of Fα2 , then q is a non-degenerate singularity of Gα and their
characteristic numbers are rb and rb 1
. It follows that, if α ∈
/ F2 , then Gα has:
(c) 2rr non-degenerate singularities on L∞ with a local meromorphic first integral of the type
−1
w
z 2 = cte. These singularities are the points of F (M ) ∪ F −1 (N ).
(d) r non-degenerate singularities on L∞ with a local holomorphic first integral of the type
wr .z 3 = cte. These singularities are the points of F −1 (P2 (α)).
The proofs of (c) and (d) are based on Lemma 1 and are left for the reader. Let us consider
now the points in Cr = ∆ ∩ L∞ . If α ∈ / F2 and q ∈ Cr, then p = F (q) is not a singularity of Fα2 ,
so that (24) and the local form of F in (viii) imply that Gα can be defined in a neighborhood of q
by a differential equation of the type:
 
  
wr d v 2 + f (w) − a + b v 2 + f (w) + cwr + · · · rwr−1 dw = 0
 
⇒ wd v 2 + f (w) − r(a + h.o.t.) dw = 0.
Therefore, q is not a singularity of Gα .
It remains to study the singularities of Gα on ∆1 := ∆ \ L∞ . Let us define

F3 = {α ∈ C; Gα has some degenerate singularity on ∆1 }

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 259

and F0 = F2 ∪ F3 . In order to finish the proof of the lemma it is enough to prove that F3 is
finite.
Observe first that condition (x) implies that C is not invariant for Fα2 . In fact, Q is invariant
for Fα2 and C ∩ Q contains more than one point. If C was an invariant for Fα2 for some α, then
all points in C ∩ Q would be singular points of Fα2 . On the other hand, the unique singularity
of Fα2 in Q \ {J, K, L} ∪ L∞ is P1 (α), which implies that one of the points of C ∩ Q is not a
singularity of Fα2 .
Another fact that we will use is that F (∆1 ) = C1 := C \R. Let us consider an affine coordinate
system (x, y) ∈ C2 = CP(2) \ R, two polynomial vector fields X and Y such that Xα = X + αY
defines Fα2 in this system (see Remark 10) and a reduced polynomial f such that C1 = (f = 0).
Define gα = Xα (f ) and hα = Xα (gα ). The fact that F3 is finite follows easily from the following
assertions:

A SSERTION 1. – Let α ∈ C be such that Gα has some degenerate singularity q ∈ ∆1 . Then

p = F (q) satisfies f (p) = gα (p) = hα (p) = 0.

A SSERTION 2. – The subset A of C × C2 defined by

 
A = (α, x, y) ∈ C × C2 ; f (x, y) = gα (x, y) = hα (x, y) = 0 ⊂ C × C1

is algebraic of dimension zero.

Proof of Assertion 1. – We will prove that if q ∈ ∆1 is such that F (q) ∈ C1 but (α, F (q)) ∈
/A
then, either q is not a singularity of Gα , or q is a non-degenerate singularity of Gα .
Let us observe first, that the conditions f (p) = gα (p) = hα (p) = 0 depend only on the divisor
f and on the foliation Fα2 . In other words, if U is an open set of C2 , Zα is a holomorphic vector
field on U which represents Fα2 |U , f 1 ∈ O(U ) is reduced and such that C1 ∩ U = (f 1 = 0) and
gα1 = Zα (f 1 ), h1α = Zα (gα1 ), then
   
p ∈ U ; f 1 (p) = gα1 (p) = h1α (p) = 0 = p ∈ U ; f (p) = gα (p) = hα (p) = 0 .

This follows from the fact that Zα = u.Xα and f1 = v.f where u, v ∈ O∗ (U ), as the reader can
check.
Let q ∈ ∆1 and p = F (q) ∈ C1 , so that f (p) = 0. Suppose that gα (p) = 0. Since gα =
Xα (f ) = df (Xα ), it follows that p is neither a singularity of Xα nor a singularity of f , so
that p ∈
/ P . Let us consider local coordinate systems (W1 , (u, v)) and (W2 , (z, w)) around q
and p = F (q), respectively, such that u(q) = v(q) = 0, x(p) = y(p) = 0 and F (u, v) = (u, v 2 ),
like in (v). In these coordinates, we have ∆1 ∩ W1 = (v = 0) and C1 ∩ W2 = (w = 0). The
∂
vector field X can be written in the coordinate system (W2 , (z, w)) as Xα (z, w) = P (z, w) ∂z +
Q(z, w) ∂w and Fα can be represented in W2 by the 1-form ωα = Q(z, w) dz − P (z, w) dw.
∂ 2

Since C1 ∩ W2 = (w = 0) and gα (p) = 0, we must have Xα (w)(p) = Q(0, 0) = 0. Now, Gα is

represented in W1 by
   
(25) ηα := F ∗ (ωα ) = Q u, v 2 du − P u, v 2 2v dv,

so that ηα (0) = 0 and q is not a singular point of Gα .

Suppose now that gα (p) = 0, but hα (p) = 0. Since hα = Xα (gα ), p is not a singularity of Xα ,
but it can be a singularity of f . We can have one of the following cases: (I) p is not a singularity
of f ; (II) p is a singularity of nodal type of f and (III) p is a cuspidal singularity of f .

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

260 A. LINS NETO

Case (I). In this case we can consider local coordinate systems (W1 , (u, v)) and (W2 , (z, w))
as before. Now, consider
gα1 = Xα (w) = Q(z, w)
and
 
h1α = Xα gα1 = P (z, w)Qz (z, w) + Q(z, w)Qw (z, w).
Since gα (p) = 0 and hα (p) = 0 we must have Q(0, 0) = 0 and P (0, 0)Qz (0, 0) = 0, so that
P (z, w) = a + h.o.t and Q(z, w) = bz + cw + h.o.t., where a, b = 0. It follows from (25) that
Gα can be represented in a neighborhood of q = (0, 0) by a vector field of the form
  ∂   ∂ ∂ ∂
2v.P u, v 2 + Q u, v 2 = 2av + bu + h.o.t.,
∂u ∂v ∂u ∂v
so that q is a non-degenerate singularity of Gα .
Case (II). In this case, we have local charts (W1 , (u, v)) and (W2 , (z, w)) in neighborhoods
of q and p, as before. Moreover, we have that f = w.f1 , where f1 (z, w) = az + bw + h.o.t. where
a = 0. On the other hand, a straightforward computation shows that gα = Q.f1 + w(P.(f1 )z +
Q.(f1 )w ) and hα (0) = Q(0, 0)[2aQ(0, 0) + 2bP (0, 0)] = 0, so that Q(0, 0) = 0. Therefore (25)
implies that q is not a singularity for Gα .
Case (III). In this case we take holomorphic coordinate systems (W1 , (u, v)) and (W2 , (z, w))
around q and p, respectively, such that
 
1 3
u(q) = v(q) = 0, z(p) = w(p) = 0 and F (u, v) = u, v − uv .
3

In these coordinates, we have

   
∆1 ∩ W1 = u = v 2 and C1 ∩ W2 = 9w2 − 4z 3 = 0 .

If we set f 1 = 9w2 − 4z 3 , gα1 = Xα (f 1 ) and h1α = Xα (gα1 ), where Xα is as before, then

 2
h1α (0, 0) = 18 Q(0, 0) =⇒ Q(0, 0) = 0,

as the reader can check. Therefore, if ωα = Q(z, w) dz − P (z, w) dw, then


ηα = F ∗ (ωα ) = [Q ◦ F + vP ◦ F ] du + P ◦ F. u − 3v 2 dv =⇒ ηα (q) = Q(0, 0) du = 0.

Therefore q is not a singularity of Gα .

Proof of Assertion 2. – Since f , gα and hα are polynomials in the three variables (α, x, y), it
is clear that A is an algebraic subset of C × C2 . In order to prove that it has dimension zero,
it is sufficient to show that their points are isolated. Fix a point (α0 , p0 ) = (α0 , x0 , y0 ) ∈ A.
Suppose first that p0 is not a singularity of C1 . Observe that there exists α1 = αo such that Fα21 is
transversal to C1 at p0 . Let X and Y be two polynomial vector fields on C2 which represent Fα20
and Fα21 , respectively, and such that Fα2 is represented by Xα = (α − α1 )X + (α − α0 )Y . Since
p0 is a smooth point of C1 and Y is transversal to C1 at p0 , there exists a local coordinate system
(W, (u, v)) around p0 such that u(p0 ) = v(p0 ) = 0, W ∩ C1 = (f 1 = v = 0) and Y (u, v) = ∂v ∂
.
In this coordinate system X can be written as

∂   ∂
X(u, v) = A(u, v) + k(u) + v,(u, v) .
∂u ∂v
4e SÉRIE – TOME 35 – 2002 – N◦ 2
 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 261

Since C1 is not invariant for Fα0 , we must have k(u) ≡ 0. Moreover, A(u, 0) ≡ 0, because Y
is transversal to X in almost all points of C1 ∩ W = (v = 0). Let gα1 = Xα (f 1 ) = Xα (v) and
h1α = Xα (gα1 ), so that A ∩ (C × W ) = (v = gα1 = h1α = 0). Now, a straighforward computation
shows that
 
gα1 (u, v) = α − α0 + (α − α1 ) k(u) + v,(u, v)
and

  
h1α (u, 0) = (α − α1 ) (α − α1 ) A(u, 0)k  (u) + k(u),(u, 0) + (α − α0 ),(u, 0) .

Therefore, the set A can be defined in C × W by gα1 (u, 0) = h1α (u, 0) = 0 =⇒

(26) α − α0 + (α − α1 )k(u) = (α − α1 )2 .A(u, 0).k  (u) = 0,

as the reader can check. Since (α0 − α1 )k(0) = gα1 0 (0, 0) = 0, the function k is non-constant, so
that k  (u) ≡ 0. This fact together with (26) imply that (α0 , p0 ) is an isolated point of A.
The above argument implies that, if S is the set of singularities of C1 , then all irreducible
components of A in C × (C1 \ S) have dimension zero. Since S is finite, if A contains some
irreducible component, say B, of dimension greater than one, then B = C × {p}, where p ∈ S.
However, as the reader can verify in the proofs of cases (II) and (III) in Assertion 1, we have the
following:
(II.1) If p is a nodal point of C1 then A ∩ (C × {p}) contains exactly two points: the points
(α, p) such that Xα (p) is tangent to one of the branchs of C1 at p.
(III.1) If p is a cuspidal point of C1 then A ∩ (C × {p}) contains exactly one point: the point
(α, p) such that Xα (p) is in the direction of the tangent cone of C1 at p.
We leave the details of the proof of these facts for the reader. This finishes the proof of
Assertion 2.
Now, let α ∈ / F0 . It follows from the above arguments that all singularities of Ga are non-
degenerate as in (a), (b), (c), (d) or
(e) Singularities of Ga on ∆1 . The foliation Gα has a local first integral of the type z.w = cte
in a neighborhood of each singularity of this type.
In fact, if q ∈ ∆1 is a singularity of Gα , as we have seen in the proof of case (I) in Assertion 1,
the linear of a vector field which represents Gα in a neighborhood of q, can be written in a
∂ ∂
suitable coordinate system as 2av ∂u + bu ∂v , where a, b = 0. This implies that the characteristic
numbers of Gα at q are both −1. On the other hand, since p = F (q) is not a singularity for Fα2 ,
this foliation has a local regular holomorphic first integral, say h, in a neighborhood of p. It
follows that h ◦ F is a local holomorphic first integral for Gα . Since the characteristic numbers
of Gα at q are both −1, this first integral is of Morse type, so that it can be written in a suitable
coordinate system as z.w = cte.
It remains to prove that Gα has degree 3r − 1. We will use Baum–Bott formula in the proof
(cf. [4] and [3]). The Baum–Bott index of a foliation F in a non-degenerate singularity p is
defined as follows. Let Z be a holomorphic vector field which represents F in a neighborhood
tr(A)2
of p and A = DZ(p). The number det(A) does not depend on the vector field Z, representing F
near p. Define the Baum–Bott index of F at p by

tr(A)2
BB (F , p) = .
det(A)

In particular, if the characteristic numbers of F at p are a and a−1 , then BB (F , p) = a + a−1 + 2.

We would like to observe that the Baum–Bott index can be defined for any isolated singularity

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

262 A. LINS NETO

of F (cf [4]), but we will use it only in the non-degenerate case. The Baum–Bott formula in
CP(2) can be stated as follows (cf. [17]): let F be a foliation of degree d in CP(2). Then

BB (F , p) = (d + 2)2 .
p∈sing (F )

/ F0 , then the singularities of Gα are the following:

In our case, if α ∈
2
Case (a). 3r singularities of index (2+3)
2
2×3 = 6 .
25

(1−6)2
−6×1 = − 6 .
25
Case (b). r2 singularities of index
2
(r+2)2
Case (c). 2r singularities of index (2+r)
2×r = 2r .
(3−r)2 (r−3)2
Case (d). r singularities of index −r×3 = − 3r .
Case (e). The singularities like in (e) have Baum–Bott index zero.
This implies that:
 25 25 (r + 2)2 (r − 3)2
BB (F , p) = × 3r2 − × r2 + × 2r − ×r
6 6 2r 3r
p∈sing (Gα )

= (3r + 1)2 = (3r − 1 + 2)2 .

Therefore dg(Gα ) = 3r − 1, which proves the lemma. ✷

We now finish the proof of the Corollary for this type of family.
L EMMA 3. – Let E be as in the main Theorem. If α ∈ E, then the foliation Ga has a rational
first integral, say of degree dα . Moreover, for any k ∈ N, the sets {α ∈ E; dα
k} and

{α ∈ E; the genus of the general integral curve of Fα is
k}

are finite.
Proof. – If α ∈ E then Fα2 has a rational first integral, say R α , where the general level surface
Rα = cte is irreducible of degree dˆα . Observe that the rational map Rα = R α ◦ F is a first
integral of Gα , whose general level Rα = cte is irreducible. This implies that the degree of the
general level surface Rα = cte is dα = r2 .dˆα . Therefore the set {α ∈ E; dα < k} is finite for all
k > 0.
Now consider an irreducible level surface Lc = (Rα = c) such that Lc does not contain
any non-dicritical singularity of Gα and F (Lc ) does not contain any singular point of C. Let
L̂c = F (Lc ). Since Lc is irreducible, we must have Lc = F −1 (L̂c ), so that the topological degree
of F |Lc : Lc → L̂c is r2  4. Observe that L̂c is transversal to C. In fact, if it was not transversal
to C it would be tangent to C in a point p ∈ L̂c ∩ C. This point is not a singular point of C,
by assumption. Therefore p must be a tangency in a regular point of C, so that F −1 (p) contains
some non-dicritical singularity of Gα (see case (I) in the proof of Assertion 2 in Lemma 2), which
contradicts the choice of Lc . This implies that L̂c ∩ C contains δ.dˆα points, δ = dg(C). Now, for
each point p ∈ L̂c ∩ C, let q(p) ∈ ∆ be such that p = F (q(p)). It is clear that the point q(p) ∈ Lc
is a ramification point of F |Lc , so that F |Lc has at least δ.dˆα ramification points. This implies
that, if L̃c is the normalization of Lc , then X (L̃c )
−δ.dˆα , so that g(Lc )  1 + δ.2d̂α . It follows
that {α ∈ E; the genus of the general integral curve of Fα is
k} is finite for any k > 0. ✷
This finishes the proof of the Corollary for families of degrees of the form 3r − 1, r  2.
In order to obtain families of degrees of the form 3r and 3r + 1 we can use the same type of

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 263

construction, but now using the families of degrees 3 and 4, respectively. In what follows, we
will sketch how these families can be constructed. Since the proof of the Corollary for them is
analogous to the previous case, we will only say how is the map F : CP(2) → CP(2) and how
are the singularities of the families.

3.2. The families of degree 3r, r  2

Consider the family (Fα3 )α∈C . Let ,1 and ,2 be two of the straight lines invariant for all folia-
tions of the family. Consider homogeneous coordinates (X, Y, Z) ∈ C3 such that ,1 = Π(Y = 0)
and ,2 = Π(Z = 0). Take F : CP(2) → CP(2), which in homogeneous coordinates is of the
form (X, Y, Z) = F (U, V, W ) = (P (U, V, W ), V r , W r ), where P is a homogeneous polyno-
mial of degree r. Let ∆ = Π({(U, V, W ); ∂P ∂U (U, V, W ) = 0}), L1 = (V = 0), L2 = (W = 0),
and C = F (∆), so that the set of critical points of F is ∆∪L1 ∪L2 , F (L1 ) = ,1 and F (L2 ) = ,2 .
We suppose that P satisfies the following generic properties:
(i.1) {U ; P (U, 0, 0) = 0} = {0}. This implies that the map F is well defined and that
[1 : 0 : 0] = L1 ∩ L2 ∈
/ ∆.
(ii.1) The curve ∆ is smooth of degree r − 1 and transversal to L1 ∪ L2 .
(iii.1) F satisfies Whitney’s conditions in CP(2) \ (L1 ∪ L2 ) (see conditions (iii)–(vii)
in §3.1).
(iv.1) The critical points of the polynomials U → P (U, 1, 0) and V → P (U, 0, 1) are of order
two (see condition (viii) in §3.1).
(v.1) The curve C is transversal to the curve C1 ∪ C2 ∪ ,0 ∪ ,1 ∪ ,2 . In particular the dicritical
singularities of the generic foliation Fα3 , p01 , p02 , p12 , p0 , p1 , p2 , q1 and q2 are not in
the curve C.
We define Hα = F ∗ (Fα3 ). It can be proved that there exists a finite set H0 such that, if α ∈ / H0
then all singularities of Hα are non-degenerate of the following types:
(a.1) One radial singularity at the point p = [1 : 0 : 0] = F −1 (p12 ). In this case Hα has a local
meromorphic first integral of the type wz = cte in a neighborhood of p and Baum–Bott
index 4.
(b.1) 3r2 singularities at F −1 {q1 , q2 , p0 } with Baum–Bott indexes 92 . In this case Hα has a
2
local meromorphic first integral of the type wz = cte in a neighborhood of each of these
points.
2
(c.1) 2r singularities at F −1 ({p01 , p02 }) with Baum–Bott indexes (r+1) r . In this case Hα
wr
has a local meromorphic first integral of the type z = cte in a neighborhood of each of
these points.
2
(d.1) 2r singularities at F −1 ({p1 , p2 }) with Baum–Bott indexes (r+2) 2r . In this case Hα has a
wr
local meromorphic first integral of the type z2 = cte in a neighborhood of each of these
points.
2
(e.1) 2r singularities at F −1 ({Q1 (α), Q2 (α)}) with Baum–Bott indexes − (r−3) 3r . In this
case Hα has a local holomorphic first integral of the type wr .z 3 = cte in a neighborhood
of each of these points.
(f.1) 2r2 singularities at F −1 ({P2 (α), Q0 (α)}) with Baum–Bott indexes − 43 . In this case
Hα has a local holomorphic first integral of the type w3 .z = cte in a neighborhood of
each of these points.
(g.1) r2 singularities at F −1 (P1 (α)) with Baum–Bott indexes − 25 6 . In this case Hα has a
6
local holomorphic first integral of the type w .z = cte in a neighborhood of each of
these points.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

264 A. LINS NETO

(h.1) The remaining singularities correspond to the points of tangency between Fα3 and C.
They have Baum–Bott index 0 and Hα has a local holomorphic first integral of the type
w.z = cte in a neighborhood of each of these points.
The foliation Hα has degree 3r because:
9 (r + 1)2 (r + 2)2 (r − 3)2
4 + 3r2 × + 2r × + 2r × − 2r ×
2 r 2r 3r
4 25
− 2r2 × − r2 × = (3r + 2)2 .
3 6
3.3. The families of degree 3r + 1, r  2

Consider the family (Fα4 )α∈C . Let ,1 , ,2 and ,3 be three of the straight lines invari-
ant for all foliations of the family, such that ,1 ∩ ,2 = ,1 ∩ ,3 . Consider homogeneous
coordinates (X, Y, Z) ∈ C3 such that ,1 = Π(X = 0), ,2 = Π(Y = 0) and ,3 = Π(Z = 0). Take
F : CP(2) → CP(2), which in homogeneous coordinates is of the form
 
(X, Y, Z) = F (U, V, W ) = U r , V r , W r .

Let Kα = F ∗ (Fα4 ). It can be proved that if α ∈ / {1, j, j 2 , ∞} then all singularities of Kα are
non-degenerate of the following types:
(a.2) 3r2 + 3 radial singularities. The foliation Kα has Baum–Bott index 4 and a local
meromorphic first integral of the type wz = cte in a neighborhood of each of these points.
2
(b.2) 6r singularities with Baum–Bott indexes (r+1) r and a local meromorphic first integral
wr
of the type z = cte in a neighborhood of each of these points. This set of points is
3
F −1 ({p1 , . . . , p6 }), where p1 , . . . , p6 are the six radial singularities contained in i=1 ,i
which are not in the intersection of any two of these lines.
2
(c.2) 3r singularities with Baum–Bott indexes − (r−3) 3r and a local holomorphic first integral
of the type wr .z 3 = cte in a neighborhood of each of these points.
(d.2) 6r2 singularities with Baum–Bott indexes −4/3 and a local holomorphic first integral
of the type w3 .z = cte in a neighborhood of each of these points.
Therefore,

BB (Kα , p)
p∈sing(Kα )

  (r + 1)2 (r − 3)2 4
= 3r2 + 3 × 4 + 6r × − 3r × − 6r2 × = (3r + 3)2 ,
r 3r 3
which implies that dg(Kα ) = 3r + 1.
We observe that the set of singular points of F is (U = 0) ∪ (V = 0) ∪ (W = 0) and its image
by F is ,1 ∪ ,2 ∪ ,3 , which is invariant for all foliations Fα4 . In particular, there are no singularities
which come from tangencies, like in (e) or (h.1).
We remark also that, if L is a leaf of Kα such that the leaf of Fα4 , F (L) := L̂, is not contained
in the set L of straight lines invariant for Fα4 , then the “ramification points” of F |L : L → L̂
are “contained” in the dicritical points of type (b.2). In order to precise this sentence, we must
solve by blowing-up these dicritical points. Let us give an idea. Fix a point q ∈ F −1 (p1 ), for
instance, where p1 ∈ ,1 and q ∈ L1 . Then there are holomorphic coordinate systems (W1 , (u, v))
and (W2 , (x, y)) around q and p1 respectively, such that u(q) = v(q) = 0, x(p1 ) = y(p1 ) = 0,
W1 ∩ L1 = (v = 0), W2 ∩ ,1 = (y = 0) and (x, y) = F (u, v) = (u, v r ). Since p1 is radial and

4e SÉRIE – TOME 35 – 2002 – N◦ 2

 SOME EXAMPLES FOR THE POINCARÉ AND PAINLEVÉ PROBLEMS 265

(y = 0) is invariant for Fα4 , this foliation can be defined in a neighborhood of p1 by a differential

equation of the form
 
η = y dx − x + A(x, y) dy = 0,
∗
where the order of A at (0, 0) is at least two. This implies that Kα is defined by ω = 1
v r−1 F (η),
which can be written as
      
ω = v du − ru + u2 A1 u, v r + uv r A2 u, v r + v 2r A3 u, v r dv = 0,

as the reader can check. Therefore, the resolution of Kα at q involves r blowing-ups in which
a dicritical divisor appears at the last blowing-up. Let us denote by π the composition of these
blowing-ups and by D the last divisor. One of the charts of D can be written as (t, v), where
π(t, v) = (tv r , v). Observe that π ∗ (Kα ) is defined in this chart by vr+11
π ∗ (ω), which can be
written as
      
dt − v r−1 t2 A1 tv r , v + tA2 tv r , v r + A3 tv r , v r dv = 0,
so that the foliation is transversal to the divisor D = (v = 0). Let us consider now the chart
(s, y) of the blowing-up π1 at p1 such that π1 (s, y) = (sy, y). If F1 is the rational map such that
π1 ◦ F1 = F ◦ π, then it is not difficult to see that F1 (t, v) = (t, v r ), so that F1 ramifies in order r
along D. This procedure provides us the “points of ramification” of F |L needed to use (22) and
to prove the analogous of Lemma 3 for the family Kα .

REFERENCES

[1] A HLFORS L.V., S ARIO L., Riemann Surfaces, Princeton, 1960.

[2] A RNOLD V., Chapitres supplémentaires de la théorie des équations différentielles ordinaires, Éditions
MIR, 1980.
[3] BAUM P., B OTT R., On the zeroes of meromorphic vector fields, in: Essais en l’honneur de De Rham,
1970, pp. 29–47.
[4] B RUNELLA M., Feuilletages holomorphes sur les surfaces complexes compactes, Ann. Scient. Éc.
Norm. Sup., 4e série 30 (1997) 569–594.
[5] B RUNELLA M., Courbes entières et feuilletages holomorphes, Enseig. Math. 45 (1–2) (1999) 196–
216.
[6] C AMACHO C., L INS N ETO A., Geometric Theory of Foliations, Birkhäuser, 1985.
[7] C AMPILLO A., C ARNICER M., Proximity inequalities and bounds for the degree of invariant curves
by foliations in PC2 , Trans. Amer. Math. Soc. 349 (9) (1997) 2211–2228.
[8] C ARNICER M., The Poincaré problem in the non-dicritical case, Ann. de Math. 140 (1994) 289–294.
[9] C ERVEAU D., L INS N ETO A., Holomorphic foliations in CP (2) having an invariant algebraic curve,
Ann. Inst. Fourier 41 (4) (1991) 883–903.
[10] C ERVEAU D., M ATTEI J.F., Formes intégrables holomorphes singulières, Astérisque 97 (1982).
[11] DARBOUX G., Mémoire sur les équations différentielles algébriques du premier ordre et du premier
degré (Mélanges), Bull. Sci. Math. (2) 2 (1878) 60–96, 123–144, 151–200.
[12] E HRESMANN C., R EEB G., Les connexions infinitésimales dans un espace fibré différentiable, in:
Colloque de topologie, Bruxelles, 1950, pp. 29–55.
[13] G ODBILLON C., Études géométriques I, Université Louis Pasteur, 1985.
[14] I NSE E.L., Ordinary Differential Equations, Dover Books on Advanced Mathematics.
[15] J OUANOLOU J.-P., Équations de Pfaff algébriques, Lecture Notes in Math., Vol. 708, Springer-Verlag,
Berlin, 1979.
[16] L IMA E.L., Grupo Fundamental e espaços de recobrimento, Projeto Euclides, IMPA, 1993.
[17] L INS N ETO A., Algebraic solutions of polynomial differential equations and foliations in dimension
two, in: Lect. Notes Math., Vol. 1345, pp. 192–231.

ANNALES SCIENTIFIQUES DE L’ÉCOLE NORMALE SUPÉRIEURE

266 A. LINS NETO

[18] M ATTEI J.F., M OUSSU R., Holonomie et intégrales premières, Ann. Sc. Ec. Norm. Sup. 13 (1980)
469–523.
[19] M OULIN O LLAGNIER J., About a conjecture on quadratic vector fields, Preprint École Polytechnique,
Paris, 2000.
[20] PAINLEVÉ P., “Sur les intégrales algébriques des équations différentielles du premier ordre” and
“Mémoire sur les équations différentielles du premier ordre”, Oeuvres de Paul Painlevé; Tome II,
Éditions du Centre National de la Recherche Scientifique, 15, quai Anatole-France, 75700, Paris,
1974.
[21] P OINCARÉ H., Sur l’intégration algébrique des équations différentielles, C. R. Acad. Sci. 112 (1891)
761–764.
[22] P OINCARÉ H., Sur l’intégration algébrique des équations différentielles du premier ordre et du premier
degré, Rendiconti del Circolo Matemático di Palermo 5 (1891) 161–191.
[23] P OINCARÉ H., Sur l’intégration algébrique des équations différentielles du premier ordre et du premier
degré, Rendiconti del Circolo Matemático di Palermo 11 (1897) 193–239.

(Manuscript received September 6, 2000;

accepted, after revision, June 13, 2001.)

Alcides L INS N ETO

Instituto de Matemática Pura e Aplicada,
Estrada Dona Castorina, 110, Horto,
Rio de Janeiro, Brazil
E-mail: [email protected]

4e SÉRIE – TOME 35 – 2002 – N◦ 2