PAIRS OF FOLIATIONS AND MATTEI-MOUSSU’S THEOREM.

ADJARATOU ARAME DIAW AND FRANK LORAY

À J.-F. Mattei, J. Martinet, J.-P. Ramis et R. Moussu.

Abstract. We prove a reduction of singularities for pairs of foliations by blowing-up,

and then investigate the analytic classification of the reduced models. Those reduced
pairs of regular foliations are well understood. The case of a regular and a singular
foliation is dealt with Mattei-Moussu’s Theorem for which we provide a new proof,
avoiding Gronwall’s inequality. We end-up announcing results recently obtained by
the first author in the case of a pair of reduced foliations sharing the same separatrices.

1. Introduction
A singular holomorphic foliation by curves on a surface is locally given by a holo-
morphic vector field with isolated zeroes. Outside of the zero set, the complex integral
curves of the vector field v define a regular holomorphic foliation by curves F. Another
vector field will define the same foliation if, and only if, it takes the form f · v for a non
vanishing holomorphic function f .
In [11], Mattei and Moussu provide a topological characterization of singular points
of holomorphic foliations that admit a non constant holomorphic first integral, i.e. such
that the leaves are locally defined as the level curves of a holomorphic function h. For
this, they use the reduction of singularities by blow-up, and they provide a careful study
of the reduced singular points. In particular, they prove that the saddle singular points
are determined by their eigenvalues and holonomy map (see Theorem 3.1, and also [4]).
The proof of this famous result is in two parts. They consider two saddles with the same
eigenvalues {λ1 , λ2 } and assume that the holomomies of the separatrices associated to
λ1 say are conjugated. First they use this conjugacy to construct a conjugacy of the
foliations between neighborhoods of annuli contained into the separatrices; this is done
by a simple and standard geometric argument. In a second step, they prove that the
conjugacy extends on a neighborhood of the singular point minus the second separatrix,
by controlling the boundedness by means of Gronwall’s Inequality, and extends holomor-
phically along the separatrix by Riemann Extension Theorem. This second part of the
proof is delicate and, in general, not written in full details. Here, we provide in Section
3 a complete and elementary proof avoiding Gronwall’s Inequality. In fact, we compare
the analytic conjugacy on the annulus with the formal conjugacy which is transversely
Date: July 31, 2020.
2010 Mathematics Subject Classification. 32M25, 32S65, 34C20, 34M35.
Key words and phrases. Foliations, Web, Holonomy, Reduction of singularities.
We warmly thank Jorge Vitório Pereira and Frédéric Touzet for useful discussions. This work is
supported by ANR-16-CE40-0008 “Foliage” grant, CAPES-COFECUB Ma 932/19 project, Université
de Rennes 1 and Centre Henri Lebesgue CHL.
1
2 A. A. DIAW AND F. LORAY

formal on the disc: they differ by a transversely formal symmetry of the foliation on
the annulus. It suffices to prove that all such symmetry extend as a transversely formal
symmetry on the disc, and this allow to conclude the analytic extension of the conjugacy.
Our proof remains valid in the saddle-node case when the central manifold (invariant
curve tangent to the zero-eigendirection) is analytic (not only formal) although Gron-
wall’s Inequality cannot be used to conclude in that case. We hope that this proof can
be useful in other situations, like for higher dimensional saddles, for instance weakening
assumptions of Reis’ result in [12]. At least, this approach has been recently used by the
first author to classify some natural pairs of singular foliations. Let us explain.
Mattei-Moussu’s Theorem provide fibered conjugacy, that is to say, it can be thought
as classification of pairs (F1 , F2 ) of a saddle F2 with a regular foliation F1 such that one
of the separatrices of F2 is a leaf of F1 . We start a study of pairs of singular foliations
by proving a reduction of singularities. In Theorem 4.2, we prove that, given a pair of
singular foliations on a complex surface M , we can construct a proper map π : M̃ → M
obtained by a finite sequence of blowing-up such that the singular points of the lifted
pair π ∗ (F1 , F2 ) fits with a list of simple models. Reduced pairs of regular foliations
can be easily handled and are well-known, let us mention [15] for a recent contribution.
Mattei-Moussu’s Theorem and some results of Martinet and Ramis allows us to deal
with the analytic classification of those models where only one of the two foliations is
regular. In Section 5, we state a recent contribution of the first author to the case
of two singular foliations: reduced pairs consist of pairs of reduced singular foliations
sharing the same invariant curves. In [2, 3], the first author provides a complete analytic
classification in the case the two foliations have a reduced tangency divisor (i.e. without
multiplicity) along the invariant curves (see Theorem 5.1). All details will be published
in the forthcoming paper [3]. In the study of reduced pairs, only remains the case of two
reduced singular foliations with higher tangency along the invariant curves. This seems
feasible, but more technical and left to the future.

2. Singular holomorphic foliations in dimension 2

A singular holomorphic foliation by curves F on a complex surface M is defined by
coherent analytic subsheaf TF ⊂ T M of rank 1 such that the quotient sheaf T M/TF is
torsion free. The locus where T M/TF is not locally free is the singular locus, a discrete
set. Locally, sections of TF take the form f · v where v is a holomorphic vector field with
isolated zeroes, and f is any holomorphic function. Any other generating vector field
takes the form f · v with f non vanishing. Equivalently, one can locally define TF as the
kernel of a holomorphic 1-form ω with isolated zeroes, which is also defined up to a non
vanishing factor, and globally by the conormal bundle NF∗ , a rank 1 coherent subsheaf
of TF∗ with torsion-free cokernel (see [1, Chapter 2]).
Let v be a germ of holomorphic vector field at (C2 , 0):

v = a(x, y)∂x + b(x, y)∂y , a, b ∈ C{x, y}

and we assume 0 is an isolated singular point of v, i.e. the common zero of a and b. The
linear part of v is defined as the linear part of the map V : (x, y) 7→ (a(x, y), b(x, y)),
i.e. its differential D0 V at 0; we denote by lin(v) this linear map. If Φ ∈ Diff(C, 0) is a
 MATTEI-MOUSSU’S THEOREM 3

coordinate change, then we have

lin (Φ∗ v)) = D0 Φ−1 · lin(v) · D0 Φ.
Therefore, the eigenvalues {λ1 , λ2 } of lin(v) are invariant under change of coordinates.
If we are interested in the foliation F defined by v, then we note that
lin(f · v) = f (0) · lin(v)
so that only the ratio λ = λ2 /λ1 (when λ1 6= 0 say) is well defined by the foliation.
When λ 6∈ R≤0 , then Poincaré and Dulac proved that we can reduce the foliation to the
unique normal form
(1) x∂x + λy∂y , or x∂x + (ny + xn )∂y
by an analytic change of coordinate; the latter case occur when λ or λ1 = n ∈ Z>0 . When
λ < 0, we are in the saddle case. Briot and Bouquet proved that there are two invariant
curves, one in each eigendirection, so that after straigthening them on coordinates axis,
we get a preliminary normalization
(2) x∂x + (λ + f (x, y))y∂y
where f is a function vanishing at 0. Poincaré and Dulac proved that we can reduce the
foliation to the unique normal form by a formal change of coordinate:


p
 p, q, k ∈ Z>0
k 2k
(3) x∂x + λy∂y , or x∂x + − + u + αu y∂y with u = xp y q ,
q 
α∈C
The latter case, called resonant saddle, occurs for generic f when λ ∈ Q<0 . We will
provide below a proof of this formal reduction. It is known however that this formal
normalization is divergent in general, and the analytic classification fails to be finite
dimensional in this case. In fact, the classification of saddles up to analytic conjugacy
is equivalent to the classification of germs of diffeomorphisms Diff(C, 0) up to analytic
conjugacy. Before explaining this, let us just end our review of non degenerate singular
points by the saddle-node case λ = 0 which admits a normal form similar to (3) with
(p, q) = 0 and u = y:
  
k+1 2k+1 k ∈ Z>0
(4) x∂x + y + αy ∂y with
α∈C
In order to explain this, let us recall the construction of the holonomy.
2.1. Holonomy. Fix a disc ∆ = {(x, 0) ; |x| < r} ⊂ C2 such that the foliation is of the
form (2), with f analytic at the neighborhood of ∆, λ + f non vanishing on ∆. Denote
by ∆∗ the punctured disc, where we have deleted the singular point of F. Fix a point
p ∈ ∆∗ and choose a minimal first H(x, y) ∈ O(C2 , p): locally at p, the 1-form dH is
non vanishing and its kernel defines the foliation, i.e. dH(v) = v · H = 0. Consider a
loop γ : [0, 1] → ∆∗ based at p = γ(0) = γ(1), and consider the analytic continuation of
H along γ.
We claim that such a continuation exists because γ is contained in a regular leaf ∆∗
of the foliation. Indeed, one can cover γ by small balls Ui equipped with a local minimal
first integral Hi and choose finitely many of them by compacity: we enumerate such
that γ successively intersect U0 , U1 , . . . , Un . Denote Vi = Ui ∩ ∆∗ . Near overlappings
4 A. A. DIAW AND F. LORAY

Vi ∩ Vi+1 , we have Hi = ϕi,i+1 ◦ Hi+1 for some ϕi,i+1 ∈ Diff(C, 0) because they both
define minimal first integrals; therefore, ϕi,i+1 ◦ Hi+1 defines an analytic extension of Hi
near Vi+1 . Starting from (U0 , H0 ) = (U, H), the analytic continuation H γ of H along γ
is given by:
H = H0 = ϕ0,1 ◦ H1 = ϕ0,1 ◦ ϕ1,2 ◦ H2 = · · · = ϕ0,1 ◦ ϕ1,2 ◦ · · · ◦ ϕn,0 ◦ H0 =: H γ .
|{z} | {z } | {z } | {z }
V0 V1 V2 back to V0

By construction, H γ is a minimal first integral of F near V = U ∩ ∆∗ , and we have

H γ = ϕγ ◦ H
with ϕγ ∈ Diff(C, 0). It is easy to check that ϕγ only depend on the homotopy type of
γ; it is called the monodromy of H along γ. Moreover, we get a group morphism
Mon(H) : π1 (∆∗ , p) → Diff(C, 0) ; γ 7→ ϕγ .
Finally, if we start with another minimal first integral H̃ = ϕ ◦ H, then the monodromy
is changed by
H̃ γ = ϕ ◦ H γ = ϕ ◦ ϕγ ◦ H = ϕ ◦ ϕγ ◦ ϕ−1 ◦H̃.
| {z }
ϕ̃γ
We call holonomy of F along γ the class of ϕγ
up to conjugacy in Diff(C, 0) (or a
representative by abuse of notation); in the sequel, we will call holonomy of F along the
leaf ∆∗ the holonomy of F along a loop of index one, i.e. of the form γ(t) = (x0 e2iπt , 0).
One can check that the holonomy of F of the form (2) along ∆∗ is of the form
(5) ϕ(z) = e−2iπλ z + o(z).
Example 2.1. The holonomy of the linear model in 3 is the monodromy of the local first
integral H(x, y) = x−λ y (after choosing a local determination of x−λ = exp(−λ log(x))),
namely ϕ(z) = e−2iπλ z. Equivalently, one can integrate the vector field v = x∂x + λy∂y
and deduce a global symmetry φ(x, y) = exp(−2iπv) = (x, e−2iπλ y) of the foliation; and
then restrict to a transversal x = x0 .
2iπ p
The holonomy of the non-linear model in 3 is ϕ(z) = e q exp(v) where v is the
holomorphic vector field
 
v = −2iπ z kq+1 + αz 2kq+1 ∂z .

Indeed, after introducing variables x = x̃q and z = x̃p y, we get u = z q and the foliation
is defined by
F̃ : x̃∂x̃ + q(z kq+1 + αz 2kq+1 )∂z
which is the formal normal form for a saddle-node singular point. The holonomy of F̃
˜ ∗ = {(x̃, 0) ; 0 < |x̃| < r̃} is the q th iterate of the holonomy of F along ∆∗ ,
along ∆
namely
ϕ̃ = ϕ◦q = exp(q · v).
We conclude by noticing that any q th root takes the form ϕ = a · exp(v), and we must
2iπ p
have the linear part a = e−2iπλ = e q (compare [10] and [9]). This latter computation
is also valid for the saddle-node case (4) by setting (p, q) = (0, 1) and u = y.
 MATTEI-MOUSSU’S THEOREM 5

Mattei-Moussu’s Theorem tells us that any two saddles of the form (2) are conjugated
by an analytic diffeomorphism preserving the two separatrices (i.e. not permuting the
coordinate axis) if, and only if, they share the same invariant λ, and the corresponding
holonomies along ∆∗ are analytically conjugated. On the other hand, Pérez-Marco
and Yoccoz proved in [17] that any diffeomorphism germ ϕ(z) = e−2iπλ z + o(z) is the
holonomy of a saddle of the form (2). Therefore, for a fixed λ < 0, analytic classification
of saddles and one-dimensional diffeomorphisms are equivalent. A result of Siegel shows
that for generic (with respect to Lebesgue measure) λ, any diffeomorphism, and therefore
any saddle is analytically linearizable. But for special “diophantine” λ, the moduli space
is infinite dimensional as shown by the works of Yoccoz in the irrational case, and Écalle-
Malgrange-Voronin in the rational case. Martinet and Ramis gave a complete analytic
classification of resonant saddles (i.e. non linear case of (3)) in [10], and of saddle-nodes
λ = 0 in [9].
2.2. Normalization in the transversely formal setting. Let D ⊂ C be a domain.
A transversely formal function on D × (C, 0) ∋ (x, y) is a power series
X
f= an (x)y n ∈ O(D)[[y]]
n≥0

i.e. where all an are analytic on D, and nothing is asked about convergence in the y-
variable. A transversely formal diffeomorphism is a “map” Φ(x, y) = (x+yf (x, y), yg(x, y))
where f, g ∈ O(D)[[y]] and g(x, 0) does not vanish on D. We denote by Diff(D × (C,\ 0))
the group of transversely formal diffeomorphisms.
Proposition 2.2. Let ∆ = {x ; |x| < r} be a disc and let F be the foliation defined
by (2) with f transversely formal on ∆ × (C, 0) ∋ (x, y) (for instance, analytic on
the neighborhood of ∆ × {0}). Then there exists a transversely formal diffeomorphism
\
Φ ∈ Diff(D × (C, 0)) of the form
Φ(x, y) = (x, yg(x, y))
such that Φ∗ F is defined by (3).
Remark 2.3. The resonant case of (1) does not occur since we have imposed in expres-
sion (2) that F has two invariant curves. Also, in the saddle-node case λ = 0, the central
manifold, tangent to the 0-eigendirection, is convergent in expression (2).
Proof. It is more convenient to define F as the kernel of the 1-form
dy dx
ω = xdy − (λ + f )ydx, or better − (λ + f )
y x
as coordinate change will be easier to handle on 1-forms. Our strategy is to make
successive
P changes of y-coordinate to kill (or reduce) step-by-step the coefficients an (x)
in f = n an (x)y n . Let us first consider a linear change y 7→ ϕ0 (x)y. Then Φ∗ ω writes
   
dy xϕ′0 (x) dx
− λ + a0 (x) − y + o(y) ;
y ϕ0 (x) x
R
as a0 vanishes at x = 0, we can integrate and find ϕ0 = exp( ax0 ) ∈ O∗ (∆). Now
we can assume a0 ≡ 0, i.e. f (x, 0) ≡ 0, and all further changes of coordinate will
6 A. A. DIAW AND F. LORAY

be tangent to the identity, i.e. with g(x, 0) ≡ 1. Consider now a change of the form
y 7→ y(1 + ϕn (x)y n ), n > 0. Then Φ∗ ω writes after normalization (i.e. multiplication by
a non vanishing analytic function)
dy

dx
− λ + f − nλϕn + xϕ′n y n + o(y n ) .
y x
Since linear operator
O(∆) → O(∆) ; ϕ 7→ nλϕ + xϕ′
writes X X
bm xm 7→ (nλ + m)bm xm
m≥0 m≥0
we deduce, when λ 6∈ Q≤0 , that the operator is onto, so that we can find at each step a
ϕn killing the coefficient an of f . The composition
Φ = · · · ◦ Φn ◦ · · · ◦ Φ2 ◦ Φ1 ◦ Φ0
converges in Diff(D × (C, \ 0)) providing a linearization of F, i.e. Φ∗ ω is colinear to
xdy − λydx. Indeed, for each n > 0, the jet of order n of Φ in y-variable is determined
by Φn ◦ · · · ◦ Φ2 ◦ Φ1 ◦ Φ0 since all other terms in the composition are tangent to the
identity up to order n. Assume now that λ ∈ Q<0 , i.e. λ = − pq with p, q ∈ Z>0 relatively
prime. Then we see that we can kill successively all the terms expect powers of u = xp y q
in f . Therefore, we are led by a transversely formal diffeomorphism to the preliminary
normal form  
dx dy du dx
(p − f (u)) +q = f (u) −
x y uf (u) x
where f is now a formal power series, with f (0) = 0. One easily check that the freedom
in this normalization is up to a change of the form Φ(x, y) = (x, yψ(u)), ψ(0) 6= 0;
moreover, this induces a change
   
q ∗ du dx ∗ du dx
u 7→ ϕ(u) = u(ψ(x)) , and therefore Φ − =ϕ −
uf (u) x uf (u) x
and we are led to a normalization of a (formal) meromorphic 1-form in one variable.
It is well known (see proof of [10, Proposition 2.1] or [7, Proposition 1.1.3]) that there
exists a unique change of coordinate ϕ tangent to the identity such that
 
du du du
ϕ∗ = k+1 + α
uf (u) u u
which is analytic (resp. formal) if f is analytic (resp. formal). The only invariants are
the pole order k + 1 (here, k > 0 is the vanishing order of f ), and the residue α. The last
 1/q
normalization is therefore given by the choice of a determination of ψ(u) = ϕ(u) u .
du
In the normal form (3), we have choosen another normalization, namely quk+1 (1+αu k)

where the residue is − αq . Finally, in the saddle-node case λ = 0, we note that everything
works the same with u = y (i.e. p = 0 and q = 1). 
We deduce a formal version of Mattei-Moussu’s Theorem:
 MATTEI-MOUSSU’S THEOREM 7

Corollary 2.4. Let F1 and F2 be two foliations of the form

Fi = ker(ωi ), ωi = xdy − (λ + o(y))ydx, λ∈C
and assume they are both analytic at the neighborhood of the disc ∆. Then, there exists
a transversally formal diffeomorphism Φ̂(x, y) = (x, y + o(y)) along ∆ conjugating the
foliations Φ̂∗ F2 ∧ F1 , i.e. Φ̂∗ ω2 ∧ ω1 = 0, if and only if, the respective holonomies along
the punctured disc ∆∗ are formally conjugated.
Proof. If we have a transversally formal conjugacy between foliations along even the
punctured disc is enough to conclude that the holonomies are conjugated. For the
converse, we can assume that F1 and F2 are into normal form (3), with the same λ
(resp. (4) when λ = 0. Only the resonant saddle case λ = − pq and saddle-node case
λ = 0 need some arguments, as we have to distinguish between the linear case, and the
several possible non-linear cases of (3), i.e how to retrieve k and α from the holonomy.
But, as shown in Example 2.1, the holonomy of the non-linear saddle (3) or saddle-node
(4) takes the form
      
kq+1 α 2kq+1 kq+1 kq + 1 α
◦q
ϕ (z) = exp z − z ∂z = z+z + − z 2kq+1 +o z 2kq+1 .
2iπq 2 2iπq
It is well-known (see [10, p. 581] or [7, Section1.3]) that the formal conjugacy class of
the tangent-to-identity diffeomorphism ϕ◦q is characterized by two invariants, namely
the maximal contact kq to the identity, and the coefficient of the monomial z 2kq+1 once
we have killed intermediate coefficients; the first one gives us k, and we deduce α from
the latter one. 

3. Mattei-Moussu’s Theorem
Theorem 3.1 (Mattei-Moussu). Let F1 and F2 be two foliations of the form
Fi = ker(ωi ), ωi = xdy − (λ + o(y))ydx, λ ∈ R<0 .
Assume that they are well defined near ∆∗ = {(x, 0) ; 0 < |x| < r} and have same
holonomy along this leaf. Then, there exists an analytic diffeomorphism Φ(x, y) = (x, y +
o(y)) conjugating the foliations: Φ∗ F2 ∧ F1 , i.e. Φ∗ ω2 ∧ ω1 = 0.
Proof. The first part of the proof follows the paper [11]. Let p0 ∈ ∆∗ and H1 , H2 be
local first integrals of F1 , F2 respectively at p0 . Assume that H1 and H2 have same
monodromy ϕ = ϕγ ∈ Diff(C, 0) along a generating loop γ of π1 (∆∗ ). Notice that the
local diffeomorphism Φi (x, y) = (x, Hi (x, y)) at p0 is conjugating Fi to the horizontal
foliation {y = constant}, i.e. Φ∗i dy ∧ ωi = 0. Therefore, the local diffeomorphism
Φ := (Φ2 )−1 ◦ Φ1 is conjugating F1 to F2 as in the statement, near p0 . By analytic
continuation of H1 , H2 along paths in ∆∗ (see Section 2.1) one can deduce the analytic
continuation of Φ by the same formula. Since H1 and H2 have the same monodromy ϕ,
we see that Φ is uniform: after analytic continuation along the generating loop γ, we get
Φγ = (Φγ2 )−1 ◦ Φγ1 = (φ ◦ Φ2 )−1 ◦ (φ ◦ Φ1 ) = (Φ2 )−1 ◦ φ−1 ◦ φ ◦ Φ1 = (Φ2 )−1 ◦ Φ1 = Φ
where φ(x, y) = (x, ϕ). We therefore obtain an analytic diffeomorphism Φ : U1 → U2
between neighborhoods Ui of ∆∗ conjugating the restrictions F1 to F2 .
8 A. A. DIAW AND F. LORAY

In the second part of the proof, we want to extend Φ analytically on the neighborhood
of 0. Our idea is to compare Φ with the transversely formal diffeomorphism Φ̂ along ∆
provided by Corollary 2.4. As they are both conjugating F1 to F2 ,
φ̂ := Φ̂−1 ◦ Φ
defines a transversely formal diffeomorphism along ∆∗ that preserves the foliation F1 .
We are going to show, in Lemma 3.2, that this forces φ̂ to extends as a transversely
formal diffeomorphism on the whole disc ∆. Therefore, the diffeomorphism Φ̂ ◦ φ̂ = Φ is
both analytic near ∆∗ and transversely formal on ∆. Equivalently, the Laurent power
series decomposition of its coefficients are convergent and with only positive exponents:
they define holomorphic germs at 0, therefore extending Φ. 
Lemma 3.2. Let F be a foliation of the form
F = ker(ω), ω = xdy − (λ + o(y)) ydx, λ ∈ C \ Q>0
analytic near the disc ∆ = {(x, 0) ; |x| < r}. Then, any transversely formal diffeo-
morphism Φ̂(x, y) = (x, φ̂(x, y)) along the punctured disc ∆∗ commuting with F, i.e.
Φ̂∗ ω ∧ ω = 0, extends as a transversely formal diffeomorphism along the whole disc ∆.
Proof. By Proposition 2.2, we can assume that F is in formal normal form (3) and
consider first the linear case: we can equivalently define F with ω = dy dx
y − λ x . If we
write Φ̂(x, y) = (x, y · g(x, y)), then the condition Φ̂∗ ω ∧ ω = 0 writes
 
dy dx
dg ∧ −λ = 0 (or (x∂x + λy∂y ) · g = 0)
y x
i.e. g is a first integral for F. If we decompose g in Laurent power series, then we get
X X
g= am,n xm y n (x∂x + λy∂y ) · g = (m + λn)am,n xm y n = 0.
m∈Z, n≥0 m∈Z, n≥0

The obstruction to extend at 0 comes from non zero coefficients am,n with negative m;
but this can only occur when λ ∈ Q>0 , what we have excluded.
Consider now the resonant case, and assume F is defined by
 
du du dx 1 du dx
ω = k+1 + α − = k
+α −
u u x u u x
(see proof of Proposition 2.2). Then we have
  
∗ 1 du dg dx
Φ̂ ω = k k
+α + − .
u g u g x

The ratio of the closed one-forms ω and Φ̂∗ ω must be a first integral of F and therefore
be constant. Moreover, since they have the same residu along ∆∗ (mind that g is not
vanishing), then they should be equal, and therefore dg ∧ du = 0, i.e. g = g(u). But
non zero coefficients am,n of g occur only for n ≥ 0, and therefore only for m ≥ 0 since
u = xp y q . So g extends on the whole of ∆. 
 MATTEI-MOUSSU’S THEOREM 9

Remark 3.3. Our proof of Mattei-Moussu’s Theorem works not only for saddles λ < 0,
but also for saddle-node λ = 0 provided that the central manifold is convergent (i.e. the
formal invariant curve along the 0-eigendirection). It was proved in [9] that the holonomy
of the strong manifold characterize the foliation, but the proof was completely indirect,
by comparing the two moduli spaces. The original approach of Mattei and Moussu
fails to conclude in that case. Here, we get a direct and simple proof when the central
manifold is convergent. We will explain later how to adapt to the case of a divergent
central manifold (the generic case).
Remark 3.4. In the saddle-node case λ = 0, it is well-known that the holonomy of
the central manifold, when it is convergent, fails to characterize the saddle-node, even
formally. In fact, if we want to normalize a saddle-node of the form x2 dy − (1 + o(y))ydx
by a transversly formal change of coordinates Φ(x, y) = (x, φ(x, y)) along the disc ∆ or
even along the punctured disc ∆∗ , we find obstructions at each step. Indeed, if it takes
the form
dy dx
− (1 + f (x)y n + o(y n )
y x
and we apply a change of the form y 7→ y(1 + g(x)y n ), then we find
dy dx
− (1 + f˜(x)y n + o(y n ) with f˜ = f − x2 g′ (x) − ng(x).
y x
If we want to make f˜ = 0 as it should be for n >> 0, we see by integration that we
R n/x
must set g = e−n/x e x2f (x) which is multiform on ∆∗ in general. Moreover, there are
many transversely formal symmetries along ∆∗ that do not extend: we can take any
transformation of the form y 7→ yg(ye1/x ) with g(0) 6= 0.
Remark 3.5. If we want to compare our proof to the original one of Mattei and Moussu,
the present one has the disadvantage that we have to discuss between the different formal
types, and deal with existence of first integrals, closed one-forms; only in the non resonant
part λ 6∈ Q it is very short and easy. On the other hand, this approach can be used to
classify pairs of singular foliations as we shall see in the next section.
Remark 3.6. Mattei-Moussu’s Theorem can be considered as a classification of pairs
of foliations. Indeed, consider a pair (F1 , F2 ) of foliations germs such that F1 is regular
and F2 is a saddle, one invariant curve of which is a leaf of F1 . If we have two such
pairs, (F1 , F2 ) and (G1 , G2 ), one can easily find coordinates in which F1 , G1 are vertical,
defined by ker(dx), and the two invariant curves of F2 , G2 are on coordinate axis. Then,
the two pairs are conjugated if and only if they share the same λ and have conjugated
holonomies. Indeed, the conjugacy between the foliations F2 and G2 provided by Theo-
rem 3.1 preserves the variable x and therefore F1 = G1 . This was one of the motivation
to us for looking at classifications of pairs in a more general setting.

4. Reduction of singularities for a pair of foliations

Consider a pair (F1 , F2 ) of holomorphic singular foliations on a complex surface M .
Denote by Tang(F1 , F2 ) the tangency divisor between the two foliations: locally, if Fi =
ker(ωi ) where ωi is a holomorphic one-form with isolated zeroes, then Tang(F1 , F2 ) is
10 A. A. DIAW AND F. LORAY

defined by the ideal (f ) given by ω1 ∧ ω2 = f dx ∧ dy. We note that Tang(F1 , F2 ) is

passing through all singular points of each Fi . Recall Seidenberg’s Theorem:
Theorem 4.1 (Seidenberg [14]). Let F be a holomorphic singular foliations on a complex
surface M . Then there is a proper map π : M̃ → M obtained by a finite sequence of
punctual blowing-up such that the lifted foliation F̃ = π ∗ F has only reduced singular
points, i.e. of the type
F = ker(xdy − (λ + o(1))ydx) with λ ∈ C \ Q>0 .
Moreover, this property is stable under additional blowing-up.
This allows to reduce the study of singular points to non degenerate models. We note
that after reduction of singularities, all singular points have exactly two invariant curves
(one of them might be formal divergent in the saddle-node case λ = 0) which are smooth
and transversal. We want to prove now a similar result for pairs of foliations.
Theorem 4.2. Let (F1 , F2 ) be a pair of holomorphic singular foliations on a compact
complex surface M , and T be the support of Tang(F1 , F2 ). Then there is a proper map
π : M̃ → M obtained by a finite sequence of punctual blowing-up such that the pair of
lifted foliations (F̃1 , F̃2 ) = π ∗ (F1 F2 ) is, at each point of M̃ , one of the following types:
(1) T = ∅, both Fi are regular and transversal to each other;
(2) T is smooth, both Fi are regular, transversal to T ;
(3) T is smooth, both Fi are regular, tangent to T ;
(4) T has a normal crossing, both Fi are regular, tangent to one component of T ,
and transversal to the other one;
(5) T is smooth, F1 is regular, F2 has a reduced singular point, and T is a common
leaf/invariant curve of Fi ;
(6) T has a normal crossing, both Fi have a reduced singular point, and T is the
common set of invariant curves of Fi ; moreover, in case both Fi are saddle-node,
they share the non zero invariant curve.
Before proving the theorem, let us comment on the possible types of reduced singular
points. First of all, types (1) - (4) admit simple local normal form:
(1) (F1 , F2 ) ∼ (dy, dx) (the generic regular case);
(2) (F1 , F2 ) ∼ (dy, d(y + xk+1 )) and T = (xk ), k ∈ Z>0 (see [5, Lemma 5.1]);
(3) (F1 , F2 ) ∼ (dy, d(y + xy k )) and T = (y k ), k ∈ Z>0 (see [6, Lemma 5]);
(4) (F1 , F2 ) ∼ (dy, d(y + xk+1 y l )) and T = (xk y l ), k, l ∈ Z>0 .
Generic points of each branch of T are either of type (2), or of type (3).
Type (5) splits into 3 cases:
(5.1) (F1 , F2 ) ∼ (dx, xdy − λydx), λ ∈ C \ (R≤0 ∪ Q>0 ), and T = (x), or
(5.2) (F1 , F2 ) ∼ (dx, xdy − (λ + o(1))ydx), λ ∈ R≤0 , and T = (x), or
(5.3) (F1 , F2 ) ∼ (dy, xdy − (1 + o(1))y k+1 dx) and T = (y k+1 ), k ∈ Z>0 ;
Moreover, the classification of the pair up to analytic diffeomorphism is equivalent to
the analytic classification of F2 in each case. Poincaré Linearization Theorem provides
linearization in (5.1). The saddle case (5.2) with λ < 0 corresponds to Mattei-Moussu’s
Theorem (see Remark 3.6). The saddle-node case splits into (5.2) with λ = 0 (see
Remark 3.3) and (5.3) (see [9, Proposition 2.2.3]). Moreover, for fixed λ, the analytic
 MATTEI-MOUSSU’S THEOREM 11

class of F2 is characterized by λ and the analytic class of its holonomy along the invariant
curve y = 0 in (5.2) (Mattei-Moussu’s Theorem), or x = 0 in (5.3) (see [9, Corollaire
3.3]).
So far, only the classification of type (6) was not known. In this direction, the first
author recently obtained (see [2, 3]) the complete classification in the case Tang(F1 , F2 )
is reduced (i.e. T = (xy), without multiplicity). We will state the results in Section 5.
We note that in case (6), any saddle occuring among Fi must have analytic invariant
curves, i.e. not formal divergent component, since they are contained in T which is
analytic.
The list of Theorem 4.2 is not stable by arbitrary blowing-up: if we blow-up items
(2) or (3) for instance, then we have to blow-up more in order to recover a pair with
reduced singular points as in the statement.

Proof. Let M , (F1 , F2 ) and T as in the statement. First of all, there exists a proper
map π : M̃ → M obtained by a finite sequence of punctual blowing-up such that the
pull-back π ∗ Fi have both only reduced singular points. This is done by applying twice
Seidenberg’s Theorem [14], successively for F1 and F2 . These properties are stable by
additional blowing-up and the final map π : M̃ → M of the statement will be obtained
by additional blowing-ups. Therefore, we can assume without lost of generality that
both Fi have reduced singular points from the beginning.
Recall that T is passing through all singular points of F1 or F2 . Outside the support
of T , the pair is regular, of type (1). Along each irreducible component of T , the
pair is generically of type (2) or (3), depending whether that component is generically
transversal to the Fi ’s, or is Fi -invariant; moreover, this generic feature is for a Zariski
open set of T . We therefore conclude that, apart from a finite set of points in M , the
situation is as in (1), (2) and (3) of the list. So the problem is local: let us consider
a point p where (F1 , F2 , T ) is not as (1), (2) or (3), and prove that after finitely many
blowing-up infinitesimally close to p, we get only points in the list (1)-(6); as we shall
see, we will get only local models of type (3)-(6) along the exceptional divisor.
In order to show this, let us consider the local formal invariant curve Γi of Fi at p:
it consists of one or two smooth and transversal branches depending on whether the
foliation is regular or singular. Then consider the germ Γ = Γ1 ∪ Γ2 ∪ T at p; mind
that they can share common branches. Then we can blow-up until Γ̃ = π ∗ Γ has only
normal crossing singular points. We note that all local invariant curves for F̃i = π ∗ Fi
are contained in Γ̃; otherwise, any extra invariant curve would descend as an additional
invariant curve for Fi outside Γ, providing a contradiction. We also note that Γ̃ contains
the support of Tang(F̃1 , F̃2 ) for a similar reason. We claim that all points along Γ̃ are
of type (3)-(6). Let us check this.
Let us consider first the case where Γ̃ is smooth. Then the two foliations are smooth
(only one invariant curve) with Γ̃ as a common leaf and no other tangency: we are in type
(3). Now, in order to consider the case Γ̃ has two components, note that any common
component between two among the three curves Γ̃i = π ∗ Γi , i = 1, 2, and T̃ = π ∗ T must
be also a component of the third curve. Indeed, if Γ̃1 and Γ̃2 have a common component,
then F̃1 and F̃2 must be tangent along this branch and it must be a component of T̃ ; in
a similar way, if T and Γ̃1 , say, have a common component, then that branch of Γ̃1 must
12 A. A. DIAW AND F. LORAY

be invariant by F̃2 as well, and be a component of Γ̃2 . We conclude that, each branch
of Γ̃ consists of
• either a branch of T̃ only,
• or a branch of Γ̃i only,
• or a common branch of Γ̃1 , Γ̃2 and T̃ .
In the first case above, then F̃1 and F̃2 are regular (only one invariant curve contined
in the other branch of Γ̃) and transversal to that branch of T̃ only: we are in case (4).
In the second case above, say i = 1, then F̃2 is regular (only one invariant curve) and
transversal to that branch of Γ̃1 , and tangent to the other branch: we are in case (5).
Finally, if all branches are shared by T̃ , Γ̃1 and Γ̃2 , then we are in case (6). Finally, in
case (6), if ever F̃1 and F̃2 are saddle-nodes oriented in opposite way, i.e. each branch of
T is the strong manifold (non zero eigendirection) of one and the central manifold (zero
eigendirection) of the other, then one check that after one additional blowing-up, we get
two singular pairs along the exceptional divisor with only one saddle-node in each pair.
To conclude, applying the above strategy of resolution by blowing-up at each point p
in M where the pair is not locally of type (1)-(3) yields a global proper map π : M̃ → M
such that the lifted pair is locally of type (1)-(6) everywhere. 

5. Reduced pairs of foliations: a partial classification

This section is devoted to announcement of results of the first author in her thesis [2].
All details will appear in a forthcoming paper [3].
We consider a local pair (F1 , F2 ) of type (6) in the list of Theorem 4.2 and assume
that Tang(F1 , F2 ) = (xy) is reduced, i.e. without multiplicity. One easily check that
this implies that we can write
Fi = ker( xdy − (λi + o(1))ydx)
with λi ∈ C \ Q≥0 , and with λ1 6= λ2 (otherwise Tang(F1 , F2 ) cannot be reduced).
Conversely, any pair (F1 , F2 ) as above with λ1 6= λ2 is of type (6) with reduced tangency
along axis.
Denote
Diff × (C2 , 0) = {Φ(x, y) = (xa(x, y), yb(x, y)) ; a, b ∈ O× (C2 , 0)} ⊂ Diff(C2 , 0)
the subgroup of those diffeomorphisms preserving the axis (without permutation) and
F∼G or (F1 , F2 ) ∼ (G1 , G2 )
when two foliations or two pairs are conjugated by a diffeomorphism Φ ∈ Diff × (C2 , 0).
Theorem 5.1 (A. A. Diaw [2, 3]). Consider two pairs (F1 , F2 ) and (G1 , G2 ) of the form
Fi = ker(xdy − (λi + fi (x, y))ydx) and Gi = ker(xdy − (λi + gi (x, y))ydx)
with λ1 6= λ2 , λi ∈ C \ Q≥0 , and fi , gi ∈ O(C, 0) vanishing at 0. Then we have
(F1 , G1 ) ∼ (F2 , G2 ) ⇔ F1 ∼ F2 and G1 ∼ G2 .
In other words, if there are Φi ∈ Diff (C2 , 0) such that Fi = Φ∗i Gi for i = 1, 2, then
×

there exists a single Φ ∈ Diff × (C2 , 0) such that Fi = Φ∗ Gi . An equivalent statement is

given by
 MATTEI-MOUSSU’S THEOREM 13

Theorem 5.2 (A. A. Diaw [2, 3]). Consider a pair (F1 , F2 ) as in Theorem 5.1. Then,
for any Ψ ∈ Diff × (C2 , 0), there exist Ψ1 , Ψ2 ∈ Diff × (C2 , 0) such that
Ψ = Ψ1 ◦ Ψ2 and Ψ∗1 Fi = Fi , i = 1, 2.
Proof of the equivalence between the two statements. Assume we are two pairs (F1 , F2 )
and (G1 , G2 ) as in Theorem 5.1 and Φi ∈ Diff × (C2 , 0) such that Fi = Φ∗i Gi . Then
we can apply Theorem 5.2 to decompose Ψ := Φ−1 1 ◦ Φ2 = Ψ1 ◦ Ψ2 and check that
Φ := Φ1 ◦ Ψ1 = Φ2 ◦ Ψ−1 2 is conjugating the pairs (F 1 , F2 ) to (G1 , G2 ). Conversely, given
Ψ ∈ Diff × (C2 , 0), we can apply Theorem 5.1 to produce conjugacy of pairs
Φ
(F1 , Ψ∗ F2 ) −→
1
(F1 , F2 )
so that we have a conjugacy
Ψ◦Φ−1
(F1 , F2 ) −→1 (Ψ∗ F1 , F2 ).
Then we have decomposition Ψ = Ψ1 ◦Ψ2 where Ψ1 = Φ1 preserves F1 , and Ψ2 = Ψ◦Φ−1
1
preserves F2 . 
The technics involved to prove Theorem 5.1 use local study of a pair of regular folia-
tions along a common leaf, as was already done in [6, 15, 8, 16]; this allows to construct
conjugacy between of pairs of foliations at the neighborhood of a punctured disc ∆∗
contained in the invariant curve. On the other hand, one can prove a version of the
decomposition in Theorem 5.2 in the transversely formal setting along the complete disc
∆, implying a transversely formal conjugacy between the pairs of foliations. Then a care-
ful study of transversely formal symmetries of a pair along ∆∗ allow the first author to
conclude to the extension of the conjugacy at 0, likely as in our proof of Mattei-Moussu’s
Theorem.
We expect that it will be possible to obtain the complete classification even for non
reduced tangency divisors, but there will be additional invariants in that case, and more
complicated statement. The reduced case is motivated by the study of Hilbert Modular
Surfaces which admit a pair of foliations with reduced tangency divisor (see [13]).

References
[1] M. Brunella, Birational Theory of Foliations, IMPA Monographs, No. 1, Springer, 2015.
[2] A. A. Diaw, Géométrie de certains tissus holomorphes singuliers en dimension 2. Thèse de
l’Université de Rennes 1 (2019). http://www.theses.fr/2019REN1S063
[3] A. A. Diaw, Pairs of foliations: reduction of singularities and local classification. (in preparation)
[4] P. M. Elizarov and Yu. S. Ilyashenko, Remarks on orbital analytic classification of germs of vector
fields. Mat. Sb. (N.S.) 121 (1983) 111-126.
[5] F. Loray, Sur les Théorèmes I et II de Painlevé. Geometry and dynamics, 165-190, Contemp. Math.,
389, Aportaciones Mat., Amer. Math. Soc., Providence, RI, 2005.
[6] F. Loray, Versal deformation of the analytic saddle-node. Analyse complexe, systèmes dynamiques,
sommabilité des séries divergentes et théories galoisiennes. II. Astérisque No. 297 (2004) 167-187.
[7] F. Loray, Pseudo-groupe d’une singularité de feuilletage holomorphe en dimension 2. Preprint (2005).
hal-00016434 https://hal.archives-ouvertes.fr/hal-00016434
[8] F. Loray, O. Thom and F. Touzet, Two-dimensional neighborhoods of elliptic curves: formal classi-
fication and foliations. Mosc. Math. J. 19 (2019) 357-392.
[9] J. Martinet and J.-P. Ramis, Problèmes de modules pour des équations différentielles non linéaires
du premier ordre. Inst. Hautes Études Sci. Publ. Math. 55 (1982) 63-164.
14 A. A. DIAW AND F. LORAY

[10] J. Martinet and J.-P. Ramis, Classification analytique des équations différentielles non linéaires
résonnantes du premier ordre. Ann. Sci. École Norm. Sup. 16 (1983) 571-621.
[11] J.-F. Mattei and R. Moussu, Holonomie et intégrales premieres. Ann. Sci. Éc. Norm. Supér. 13
(1980) 469-523.
[12] H. Reis, Equivalence and semi-completude of foliations. Nonlinear Anal. 64 (2006) 1654-1665.
[13] E. Rousseau, F. Touzet, Curves in Hilbert modular varieties. Asian J. Math. 22 (2018) 673-689.
[14] A. Seidenberg, Reduction of singularities of the differential equation Ady=Bdx. Amer. J. Math. 90
(1968) 248-269.
[15] O. Thom, Classification locale de bifeuilletages holomorphes sur les surfaces complexes. Bull. Braz.
Math. Soc. (N.S.) 47 (2016) 989-1005.
[16] O. Thom, Formal classication of two-dimensional neighborhoods of genus g ≥ 2 curves with trivial
normal bundle. Preprint (2018). arXiv:1807.01046 https://arxiv.org/abs/1807.01046
[17] R. Pérez Marco and J.-C. Yoccoz, Germes de feuilletages holomorphes à holonomie prescrite. Com-
plex analytic methods in dynamical systems (Rio de Janeiro, 1992). Astérisque 222 (1994) 345-371.

Univ Rennes, CNRS, IRMAR - UMR 6625, F-35000 Rennes, France

E-mail address: [email protected]

Univ Rennes, CNRS, IRMAR - UMR 6625, F-35000 Rennes, France

E-mail address: [email protected]