ARTICLE IN PRESS

J. Differential Equations 196 (2004) 289–300

Hyperbolicity, transversality and analytic

first integrals
Jacky Cresson
Equipe de Mathématiques de Besan@on CNRS-UMR 6623, Université de Franche-Comté, 16 route de Gray,
25030 Besan@on, Cedex, France
Received October 25, 2000; revised August 11, 2003

Abstract
Let A be a (normally) hyperbolic compact invariant manifold of an analytic diffeomorphism
f of an analytic manifold M: We assume that the stable and unstable manifold of A intersect
transversally (in an admissible way), the dynamics on A is ergodic and the modulus of the
eigenvalues associated to the stable and unstable manifold, respectively, satisfy a non-
resonance condition. In the case where A is a point or a torus, we prove that the discrete
dynamical system associated to f does not admit an analytic first integral. The proof is based
on a triviality lemma, which is of combinatorial nature, and a geometrical lemma. The same
techniques, allow us to prove analytic non-integrability of Hamiltonian systems having Arnold
diffusion. In particular, using results of Xia, we prove analytic non-integrability of the elliptic
restricted three-body problem, as well as the planar three-body problem.
r 2003 Elsevier Inc. All rights reserved.

Keywords: Normally hyperbolic manifold; Analytic first integral; Partially hyperbolic tori; Three-body
problem

1. Introduction

The aim of this paper is to discuss the following conjecture [2]

Conjecture 1.1. Let f be an analytic diffeomorphism of an analytic manifold M, and A

be a compact hyperbolic invariant set for f. We assume that
(i) the stable and unstable manifold of A intersect transversally,
(ii) f is ergodic on A,
(iii) the eigenvalues of f associated to the stable (resp. unstable) manifold satisfy a non-
resonance condition.

E-mail address: [email protected].

0022-0396/$ - see front matter r 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.jde.2003.10.002
 ARTICLE IN PRESS
290 J. Cresson / J. Differential Equations 196 (2004) 289–300

Then the discrete dynamical system associated to f does not admit an analytic first
integral, except constant.
In this paper, we prove this conjecture for a point or a torus.
The proof is based on two results. The first one, called triviality lemma, states that
under assumption (ii) and (iii), an analytic function, which is zero on a generic orbit
of the stable (or unstable) manifold, is identically zero. Then, conditions (ii) and (iii)
are sufficient conditions under which the stable and unstable manifold of a point, or
a torus, are a key-set for analytic functions.
The second result, called geometrical lemma, states that a C 1 function, constant on
the stable and unstable manifold has a differential which is zero at all point of
transverse intersection.
The same techniques allow us to prove that a Hamiltonian system H possessing a
partially hyperbolic torus satisfying assumption (i)–(iii) does not admit an analytic
first integral independent of H: This result implies, via Xia study of Arnold diffusion
in the three-body problem, non-existence of analytic first integrals for the elliptic
restricted three-body problem, as well as the planar three-body problem, extending a
well-known result of Poincaré.

2. Hyperbolic fixed point

2.1. On a theorem of Moser

Let f be an analytic diffeomorphism of Rn : We say that f possesses a transverse

hyperbolic homoclinic structure if f admits an invariant hyperbolic fixed point p;
whose stable and unstable manifolds, denoted W
ðpÞ and W þ ðpÞ; intersect
transversally.
In 1973, Moser [5] proves, for n ¼ 2; the following theorem.

Theorem 2.1. Let f be an analytic diffeomorphism of R2 ; possessing a transverse

hyperbolic homoclinic structure, then the dynamical system associated to f does not
admit an analytic first integral.

His proof is based on the Birkhoff–Smale theorem. Precisely, he uses the existence
of a hyperbolic invariant set in the neighbourhood of the homoclinic orbit, on which
the dynamics is complicated. In particular, there exists a dense orbit. This set is then
a key-set for analytic functions.
The generalization of this result in higher dimension is difficult (see [3]) if one
wants to follow Moser’s scheme of proof. This is due in particular, to the fact that
key’s sets of analytic functions with several variables are complicated to characterize.

2.2. Main result

Let f be a diffeomorphism of Rn ; possessing a transverse hyperbolic homoclinic

structure.
 ARTICLE IN PRESS
J. Cresson / J. Differential Equations 196 (2004) 289–300 291

We call local analytic first integral for f ; a C 1 first integral, such that its restriction
to an open neighbourhood U of W
ðpÞ,W þ ðpÞ is analytic.

Remark 2.1. This definition has been suggested by R. Roussarie in order to cover
some problems concerning first integrals of polynomial vector fields.

We say that the local analytic first integral is C o -trivial, if its restriction to U is
constant.

Theorem 2.2. Let f be an analytic diffeomorphism of Rn such that p is a hyperbolic

fixed point for f. We assume that

(i) W
ðpÞ and W þ ðpÞ intersect transversally in an admissible homoclinic point h,

(ii) the eigenvalues of Df ðpÞ associated to W
ðpÞ (resp. W þ ðpÞ), denoted l
i ;

þ þ
i ¼ 1; y; n and li ; i ¼ 1; y; n ; respectively, satisfy the following non-
resonance condition:

jðls Þn ja1 ð1Þ

s
for s ¼ 7; where nAZn \ f0g; n ¼ ðn1 ; y; nns Þ; ls ¼ ðls1 ; y; lsns Þ; ðls Þn ¼
ðls1 Þn1 yðlsns Þnn :
s

Then, the dynamical system defined by f does not possess an analytic first integral
which is not C o -trivial.

The notion of admissible homoclinic point will be precised during the proof of the
theorem (see Definition 2.1).
For diffeomorphisms of R2 ; the non-resonance condition is empty, as well as the
condition on the homoclinic point to be admissible. Then, if we look for an analytic
first integral defined on the whole space, the theorem implies that it is trivial. As a
consequence, the theorem of Moser is a corollary of our result.

2.3. Proof of Theorem 2.2

2.3.1. Preliminary
The proof of Theorem 2.2 is based on two key results. The first one is of
combinatorial nature, and is related to the dynamics on the stable or unstable
manifold (the triviality lemma). The second one, if of geometrical nature, and is
related to the transverse structure in each iterates of the homoclinic point.
Let x0 ARn ; we denote gðx0 Þ the orbit of x0 under f :

Lemma 2.1 (Triviality lemma). Let f be an analytic diffeomorphism of Rn satisfying

assumption (ii) of Theorem 2.2. Let A be an analytic function on W
ðpÞ (resp. W þ ðpÞ)
such that AðxÞ ¼ 0 for all xAgðx
Þ (resp. xAgðxþ Þ), where x
(resp. xþ ) is an
admissible point of W
ðpÞ (resp. W þ ðpÞ), then A ¼ 0 on W
ðpÞ (resp. W þ ðpÞ).
 ARTICLE IN PRESS
292 J. Cresson / J. Differential Equations 196 (2004) 289–300

The proof is given in the next section.

Lemma 2.2 (Geometrical lemma). Let f be an analytic diffeomorphism of Rn

satisfying the assumptions of Theorem 2.2. Let A be a function of class C 1 which is
constant on W
ðpÞ,W þ ðpÞ; then DAðxÞ ¼ 0 for all xAgðhÞ:

The proof is given in appendix.

2.3.2. Proof
Let P be an analytic first integral for f : The idea is to prove by induction, the
cancellation of the successive derivatives of P; denoted DPi ðxÞ; for all xAgðhÞ where
h is an admissible point. As P is analytic on U and W
ðpÞ,W þ ðpÞCU which is a
connected set, we deduce that P ¼ const on U:
The induction is based on the following property.
ðhn Þ We have DPi ðxÞ ¼ 0 for all xAgðhÞ; and 1pipn:
This property is satisfied for n ¼ 1: Indeed, we have PðxÞ ¼ const on
W
ðpÞ,W þ ðpÞ by definition. The geometrical lemma implies DPðxÞ ¼ 0 for all
xAgðhÞ:
We now prove that ðhn Þ implies ðhnþ1 Þ: By ðhn Þ; we have DPn ðxÞ ¼ 0 for all
xAgðhÞ: By the triviality lemma, we deduce that DPn jW
ðpÞ ¼ 0 and DPn jW
ðpÞ ¼ 0:
Then, by the geometrical lemma, we obtain DPnþ1 ðxÞ ¼ 0 for all xAgðhÞ:
By induction, we then have DPi ðxÞ ¼ 0 for all xAgðhÞ and iX1; which concludes
the proof of the theorem. &

2.4. Proof of the triviality lemma

2.4.1. Reduction to a linear diffeomorphism

Let f
ðxÞ be the restriction of f to W
ðpÞ: The linear map Df
ðpÞ admits
eigenvalues l




i ; i ¼ 1; y; n ; such that 0oj Reðli Þ jo1 for i ¼ 1; y; n : Moreover,
by the non-resonance condition (ii) of Theorem 2.2, we have the Poincaré theorem
[1, p. 186], for analytic linearization of f
: There exists an analytic coordinates
system y ¼ zðxÞ; defined on an open neighbourhood U of p in W
ðpÞ; such that
z3f
3z
1 ¼ flin



; where flin ðxÞ ¼ Df
ðpÞ:x: We denote by x
the image of h in this
coordinates system.
As Aððf
Þk ðhÞÞ ¼ const by assumption, we have A3z
1 3ðflin Þk 3zðx
Þ ¼ const: We
denote à ¼ P3z
1 and y
¼ zðx
Þ; then

k

Ãððflin Þ ðy ÞÞ ¼ const:

The function à is still analytic on U: If à 0 on U; then A 0 on z
1 ðUÞ: As

z
1 ðUÞ an open neighbourhood of W
ðpÞ; and W
ðpÞ is a connected set, we have
A 0 on W
ðpÞ:
P
We can always find an open set V CU; containing p; such that ÃðxÞ ¼ n an xn ; for
all xAV :
 ARTICLE IN PRESS
J. Cresson / J. Differential Equations 196 (2004) 289–300 293

Moreover, we can find an holomorphic coordinates system which diagonalizes flin :
n
n

We denote by Flin : C -C the linear mapping defined by Flin ðxÞ ¼ L :x; where

L
is a diagonal matrix, constituted of the eigenvalues l

i ; i ¼ 1; y; n : We denote
by h the image of h in this coordinates system. A similar reasoning for W þ ðpÞ

produces a point hþ :

Definition 2.1. A point hAW
ðTÞ (resp. W þ ðTÞ) is called admissible if h
(resp. hþ )
þ
belongs to ðC Þn
(resp. ðC Þn ). A homoclinic point h is called admissible if hþ and



h
belongs to ðC Þ and ðC Þn , respectively.
n

Remark 2.2. We do not know if admissible points are generic in the analytic
category.

By the previous remarks, the triviality lemma follows from the following lemma.

Lemma 2.3. Let F : Cn -Cn be a linear mapping defined by F ðxÞ ¼ L:x; where L is a
diagonal matrix whose eigenvalues li ; i ¼ 1; y; n; satisfy the non-resonance condition
and are such that jli jo1 for i ¼ 1; y; n (resp. jli j41 for i ¼ 1; y; n). Let A be an
holomorphic function and h a point in ðC Þn : If AðF k ðhÞÞ ¼ 0 for all kAN then A 0:

The proof is detailed in the next paragraph.

2.4.2. Proof of Lemma 2.3

We give the proof for a contracting mapping. The case of an expansive mapping is
similar.
Let F : Cn -Cn be of the form

F ðxÞ ¼ ðl1 x1 ; y; ln xn Þ;

where 0ojli jo1 for i ¼ 1; y; n and satisfy the non-resonance condition. Let h ¼
n
ðh1 ; y; hn ÞAðC Þ ; and
X
AðxÞ ¼ an xn ;
nANn

n ¼ ðn1 ; y; nn Þ; xn ¼ xn11 yxnnn ; an holomorphic function. We denote

jnj ¼ n1 þ ? þ nn :
We have AðF k ðhÞÞ ¼ 0 for all kAN by assumption, then
X
a0 þ an lkn hn ¼ 0; 8kAN: ð2Þ
n

As the eigenvalues ðl1 ; y; ln Þ satisfy a non-resonance condition, the quantities

jln j; nANn can be totally ordered, i.e.

jln0 j4jln1 j4 ? 4jlnk j4 ? :

 ARTICLE IN PRESS
294 J. Cresson / J. Differential Equations 196 (2004) 289–300

As a consequence, Eq. (2) is equivalent to

X
ani lkni hni ¼ 0; 8kAN: ð3Þ
iX0

The cancellation of A is done by induction. We first factorize Eq. (3) by ln0 :

We obtain

X  k
n0 ln i
an 0 h þ ani hni ¼ 0; 8kAN: ð4Þ
iX1
ln 0

As jlni =ln0 jo1 for all iX1; taking the limit of (4) when k-N; we obtain

an0 hn0 ¼ 0: ð5Þ

As h is an admissible point, we have hn0 a0 and an0 ¼ 0:

A simple induction on i allows us to prove that A 0: This concludes the proof of
the lemma. &

3. Normally hyperbolic tori

Let f be an analytic diffeomorphism of an analytic manifold M; and T an

invariant, n-dimensional normally hyperbolic torus for f : Following [7, p. 322], there
exists an analytic coordinates system, defined in a neighbourhood U of T; such that
f takes the form

f ðy; s; uÞ ¼ ðy þ 2po; Lþ ðyÞs; L
ðyÞuÞ þ rðy; s; uÞ; ð6Þ

þ
where ðy; s; uÞATn Rl Rl ; r is of order 2 in s and u; and rðy; 0; uÞ ¼ 0;
rðy; s; 0Þ ¼ 0:
In this coordinates system, the invariant torus T is given by

þ
T ¼ fðy; s; uÞATn Rl Rl js ¼ u ¼ 0g

and its stable and unstable manifolds are given by

þ
W
ðTÞ ¼ fðy; s; uÞATn Rl Rl ju ¼ 0g;

þ
W þ ðTÞ ¼ fðy; s; uÞATn Rl Rl js ¼ 0g:

respectively.
The torus T is said to be reducible if the matrices Ls ðyÞ; s ¼ 7; are
independent of y:
 ARTICLE IN PRESS
J. Cresson / J. Differential Equations 196 (2004) 289–300 295

Definition 3.1. An homoclinic point h to T is called admissible if for some iterates,

þ

we have f n ðhÞ ¼ ðy
; s
; 0ÞAW
ðTÞ and f n ðhÞ ¼ ðyþ ; 0; uþ Þ with s
AðR Þl and
þ
uþ AðR Þl :

We have the following theorem.

Theorem 3.1. Let T be a reducible invariant normally hyperbolic torus of an analytic

diffeomorphism f. We assume that

(i) the stable and unstable manifold intersect transversally in an admissible

homoclinic point h,
(ii) the dynamics on the torus is minimal,
(iii) the eigenvalues associated to the stable and unstable manifold, denoted by lsi ;
i ¼ 1; y; l s satisfy the non-resonance condition

ðjlsi jÞn a1;

for all nAZls \f0g; s ¼ 7:

Then, the discrete dynamical system defined by f does not admit an analytic first
integral.

The scheme of proof is similar to that of Theorem 2.2. The geometrical lemma can
be applied. We only need to prove the following version of the triviality lemma.

Lemma 3.1 (Triviality lemma for normally hyperbolic tori). Let T be a reducible
normally hyperbolic torus. Let hþ (resp. h
) be an admissible point of W þ ðTÞ (resp.
W
ðTÞ), and A an analytic function, which vanishes on the orbit gðhþ Þ (resp. gðh
Þ) of
hþ (resp. h
). If the modulus of the eigenvalues of Lþ (resp. L
) satisfy the
non-resonance condition and the flow on T is minimal, then A 0 on W þ ðTÞ
(resp. W
ðTÞ).

Proof. We detail the proof for W
ðTÞ: The proof is similar for W þ ðTÞ:
Let A be an analytic function on W
ðTÞ: In a sufficiently small neighbourhood V
of T; A takes the form
X
Aðy; sÞ ¼ ak ðyÞsk ; ð7Þ
l

kAN

where ak ðyÞ is a 2p periodic function of y:

Let h
¼ ðy
; s
ÞAU-W
ðTÞ such that s
a0: The restriction of f to W
ðTÞ is
defined in U-W
ðTÞ by

f
ðy; sÞ ¼ ðy þ o; L
sÞ: ð8Þ

 ARTICLE IN PRESS
296 J. Cresson / J. Differential Equations 196 (2004) 289–300

By assumption, we have Aððf
Þm ðh
ÞÞ ¼ 0 for all mAN; hence

X
ak ðy
þ moÞðl
Þmk ðs
Þk ¼ 0; mAN: ð9Þ
l

kAN

As the eigenvalues l
satisfy a non-resonance condition, the quantities ðl
Þk ;

kANl ; are totally ordered, i.e.

ðl
Þk0 4ðl
Þk1 4 ? 4ðl
Þki 4 ? : ð10Þ

As a consequence, Eq. (9) is equivalent to

X
aki ðy
þ moÞðl
Þmki ðs
Þki ¼ 0; 8mAN: ð11Þ
iX0

The cancellation of A is done by induction on i: We factorize ðl
Þmk0 in Eq. (11).

We obtain
!m
X ðl
Þki

k0
ak0 ðy
þ moÞðs Þ þ aki ðy
þ moÞ
k0
ðs
Þki ¼ 0; 8mAN: ð12Þ
iX1 ðl Þ

As jðl
Þki =ðl
Þk0 jo1 for all iX1; and s
AðC Þl ; we have, taking the limit of (12)
when m-N;

lim ak0 ðy
þ moÞ ¼ 0: ð13Þ

m-N

As o is non-resonant, we deduce, by a density argument, that ak0 ðyÞ ¼ 0 for all

yATn :
A simple induction on i concludes the proof. &

4. Partially hyperbolic tori

Let M be an analytic symplectic manifold of dimension 2n þ 2l; and H an analytic

Hamiltonian system defined on M: We call partially hyperbolic torus, an invariant n-
dimensional torus, for which there exists a neighbourhood such that the
Hamiltonian takes the form

* þ 12 I:Gf þ x:Py þ gðf; I; x; yÞ;

Hðf; I; s; yÞ ¼ o:I

where ðf; I; x; yÞATn Rn Rl Rl ; with the usual scalar product, G and P two
symmetricals matrices, and g is of order 3 in ðI; x; yÞ:
 ARTICLE IN PRESS
J. Cresson / J. Differential Equations 196 (2004) 289–300 297

* satisfies a diophantine condition

We assume that o
g
*
jo:kjX ð14Þ
jkjt

for all kAZn \f0g; g40 and t41:

The partially hyperbolic torus T is defined by

T ¼ fðf; I; x; yÞATn Rn Rl Rl jI ¼ x ¼ y ¼ 0g:

Its stable manifold (resp. unstable manifold), denoted by W
ðTÞ (resp. W þ ðTÞ), is
defined by

W
ðTÞ ¼ fðf; I; x; yÞATn Rn Rl Rl jI ¼ y ¼ 0g;

ðresp:W þ ðTÞ ¼ fðf; I; x; yÞATn Rn Rl Rl jI ¼ x ¼ 0gÞ:

As o* is non-resonant, there exists a 2n-dimensional Poincaré section S; defined in a

neighbourhood of T; such that the first return map takes the form

f ðy; r; s; uÞ ¼ ðy þ 2po þ nr; r; Ls; L
1 uÞ þ rðy; r; s; uÞ; ð15Þ

where ðy; r; s; uÞATn
1 Rn
1 R R; r is of order 2 in r; s and u; o is non-

resonant, and L is a diagonal matrix with real eigenvalues li ; i ¼ 1; y; l:

Remark 4.1. Invariant partially hyperbolic tori of near integrable Hamiltonian

systems, obtained by bifurcation of resonant tori of integrable Hamiltonian systems
along simple resonance ðl ¼ 1Þ; possess a first return map of form (15). For l41;
this form is valid only under particular conditions of reductibility of the flow on the
torus [6].

Theorem 4.1. Let T be an invariant l-partially hyperbolic torus of an analytic

Hamiltonian system H; lX1: Let H be the energy level containing T. We assume that
(i) the stable an unstable manifold of T intersect transversally in H;
(ii) the eigenvalues li ; i ¼ 1; y; l; satisfy the non-resonance condition

ln a1; ð16Þ

where nAZl \f0g; n ¼ ðn1 ; y; nl Þ; ln ¼ ln11 ylnl l :

Then, the Hamiltonian system does not admit an analytic first integral
independent of H.

The proof is similar to the normally hyperbolic case. We are then reduce to prove the
triviality lemma for partially hyperbolic tori.
 ARTICLE IN PRESS
298 J. Cresson / J. Differential Equations 196 (2004) 289–300

Lemma 4.1 (Triviality lemma for partially hyperbolic tori). Let T be a partially
hyperbolic torus. Let hþ (resp. h
) be an admissible point of W þ ðTÞ (resp. W
ðTÞ),
and A an analytic function, which vanishes on the orbit gðhþ Þ (resp. gðh
Þ) of hþ (resp.
h
). If the modulus of the eigenvalues of L satisfy the non-resonance condition and the
flow on T is minimal, then A 0 on W
ðTÞ (resp. W þ ðTÞ).

Proof. We prove the lemma for the stable manifold, the case of the unstable
manifold being similar. The stable manifold is defined by W
ðTÞ ¼
fðy; I; s; uÞATn Rn Rl Rl jI ¼ u ¼ 0g: An analytic function on W
ðTÞ is then
P
of the form Aðy; sÞ ¼ kANn ak ðyÞsk : As the dynamics on W
ðTÞ is of the form
f
ðy; sÞ ¼ ðy þ 2po; LsÞ; we must solve an equation similar to (9). &

5. The three-body problem and Arnold diffusion

The elliptic restricted three-body problem is the study of the behaviour of a

particle A; of mass zero, in Newtonian interaction with two points J and S; of mass
mA 0; 1 and 1
m; respectively, such that the vector SJ describes an ellipse, with
eccentricity e and focus at the centre of mass.
The Hamiltonian of this system is given by

 
jjpjj2 m 1
m
He;m ðt; q; pÞ ¼
þ ; ð17Þ
2 dðt; q; eÞ sðt; q; eÞ

where ðt; q; pÞAR R2 R2 ; jj  jj is the Euclidean norm, dðt; q; eÞ ¼ jjq
Jt jj;

sðt; q; eÞ ¼ jjq
St jj with St ¼ ðð1
mÞr cos u; ð1
mÞr sin uÞ; Jt ¼ ð
mr cos u;
2
mr sin uÞ; r ¼ 1þe1
e pffiffiffiffiffiffiffi
t ffi
cos u and u ¼ e sin u þ 1
e2 :
In [8], Xia proves, in his study of Arnold diffusion in the three-body problem, the
following theorem.

Theorem 5.1. For 0oe51 and 0om5e; there exists invariant 1-partially hyperbolic
tori for He;m : Let T be such a torus. We denote by He;m the energy level containing T.
The stable and unstable manifold of T intersect transversally in He;m :

The first return map defined in a neighbourhood of Xia tori is of form (15).
Moreover, the dynamics on each of these tori is minimal. As they are 1-hyperbolic,
the non-resonance condition, as well as the genericity of the homoclinic point are
always satisfied. Then, Theorem 4.1 applies, and we have:

Theorem 5.2. The elliptic restricted three-body problem does not admit an analytic first
integral independent of He;m ; for 0oe51 and 0om5e:
 ARTICLE IN PRESS
J. Cresson / J. Differential Equations 196 (2004) 289–300 299

This theorem is announced by Xia in [8] without proof. This result extend to the
planar three-body problem using Xia work [9].

6. Conclusion

The proof of Theorem 3.1 extends immediately to the following case with minor
modifications.

Theorem 6.1. Let A be a normally hyperbolic manifold of an analytic diffeomorphism f

defined on an analytic manifold M. We assume that there exists an analytic coordinates
systems ðx; s; uÞAA Rl Rn ; defined on an open neighbourhood of A such that f is of
the form
f ðx; s; uÞ ¼ ðgðxÞ; L
s; Lþ uÞ þ rðx; s; uÞ; ð18Þ

where Ls ; s ¼ 7 are diagonal matrices, r is of order 2 in s and u, and g : A-A is a

diffeomorphism.
We assume that
(i) the stable and unstable manifold of A intersect transversally in M. There exists an
admissible homoclinic point h in W
ðAÞ-W þ ðAÞ;
(ii) g is ergodic on A,
(iii) the eigenvalues ls of Ls ; s ¼ 7 satisfy a non-resonance condition.

Then, f does not admit an analytic first integral except constant.

We follow the proof of Theorem 3.1. The analogue of Eq. (11) is

X
aki ðgm ðx
ÞÞðl
Þki ðs
Þki ¼ 0; 8mAN; ð19Þ
iX0

where ðx
; s
; 0Þ are the coordinates of some iterates of the homoclinic point h: We
deduce
lim ak0 ðgm ðx
ÞÞ ¼ 0: ð20Þ
m-N

As g is ergodic, a density argument implies ak0 ðxÞ ¼ 0 for all xAA: A simple
induction on i concludes the proof.
Theorem 6.1 is then a first step toward the conjecture.
However, in order to cover a more general situation, we must deal with
non-reducible normally hyperbolic manifolds, i.e. the normal form (18) is replaced by

f ðx; s; uÞ ¼ ðgðxÞ; L
ðxÞs; Lþ ðxÞuÞ þ rðx; s; uÞ: ð21Þ

In this case, the analogue of Eq. (11) is very complicated. Even in the (non-generic)
case of diagonal matrices Ls ðxÞ; s ¼ 7; we must use Oseledec multiplicativ ergodic
theorem (see [4, p. 665]) in order to conclude.
 ARTICLE IN PRESS
300 J. Cresson / J. Differential Equations 196 (2004) 289–300

Appendix. The geometrical lemma

The intersection between W
ðpÞ and W þ ðpÞ is transversal in a space of dimension

n; with W
ðpÞ ¼ n
and dimW þ ðpÞ ¼ nþ ; n ¼ n
þ nþ : The geometrical lemma is a
consequence of the following lemma.

Lemma A.1 (General geometrical lemma). Let M be an n-dimensional manifold of

class C k ; kX1; and V
(resp. V þ ) a n
-submanifold (resp. nþ -submanifold) of class
C k ; such that V
and V þ intersect transversally in M. Let P a function on M of class
C 1 ; constant on V
,V þ ; then DPðxÞ ¼ 0 for xAV þ -V
:

Proof. As V
and V þ intersect transversally, the tangent bundle in xAV þ -V
is

Tx M ¼ Tx V
þ Tx V þ : As P is constant on V
(resp. V þ ), DPðxÞ is zero on Tx V

(resp. Tx V þ ), then identically zero. &

References

[1] V.I. Arnold, Chapitres supplémentaires de la théorie des équations différentielles ordinaires, Ed. Mir-
Librairie du globe, 1996.
[2] J. Cresson, Hyperbolicité et non-intégrabilité analytique II. Tores normalement et partieliement
hyperboliques, C. R. Acad. Sci. Paris, Serie I 333 (2001) 229–232.
[3] S. Dovbysh, Transversal intersection of separatrices and branching of solutions as obstructions to the
existence of an analytic integral in many-dimensional systems I, Basic result, sepraratrices of
hyperbolic points, Collectanea Math. 50 (2) (1999) 119–197.
[4] A. Katok, B. Hassenblatt, Introduction to the Modern Theory of Dynamical Systems, Cambridge
University Press, Cambridge, 1995.
[5] J. Moser, Stable and random motions in dynamical systems (with a special emphasis in celestial
mechanics), in: Ann. Math. Stud., Vol. 77, Princeton University Press, Princeton, NJ, 1973.
[6] D. Treshchev, The mechanism of destruction of resonance tori for Hamiltonian systems, Math. USSR
Sbornik 68 (1991) 181–203.
[7] S. Wiggins, Global Bifurcation and Chaos, Springer, Berlin, 1988.
[8] J. Xia, Arnold diffusion in the elliptic restricted three-body problem, J. Dynamics Differential
Equations 5 (2) (1993).
[9] J. Xia, Arnold diffusion and oscillatory solutions in the planar three-body problem, J. Differential
Equations 110 (1994) 289–321.