A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT

MANIFOLDS

JACKY CRESSON AND STEPHEN WIGGINS

arXiv:math.DS/0510645 v1 29 Oct 2005

Abstract. Let N be a smooth manifold and f : N → N be a C ℓ , ℓ ≥ 2 diffeomorphism.

Let M be a normally hyperbolic invariant manifold, not necessarily compact. We prove
an analogue of the λ-lemma in this case.

Contents

1. Introduction 1
2. A λ-lemma for normally hyperbolic invariant manifolds 3
2.1. Normally hyperbolic invariant manifolds 4
2.2. Normal form 4
2.3. The λ-lemma 6
3. Hamiltonian systems and normally hyperbolic invariant cylinders and annuli 7
3.1. Main result 8
3.2. An example 10
4. Proof of the λ-lemma 13
4.1. Preliminaries 13
4.2. Notation 14
4.3. Inclinations for tangent vectors in the stable manifold 14
4.4. Extending the estimates to tangent vectors not in TW s (M ) ∆n 17
4.5. Stretching along the unstable manifold 19
References 20

1. Introduction

In a recent paper, Richard Moeckel [20] developed a method for proving the existence
of drifting orbits on Cantor sets of annuli. His result is related to the study of Arnold
diffusion in Hamiltonian systems [1], and provides a way to overcome the so called gaps
problems for transition chains in Arnold’s original mechanism [17]. We refer to Lochak
[17] for a review of this problem.

1
2 JACKY CRESSON AND STEPHEN WIGGINS

The principal assumption of his work is the existence of a symbolic dynamics for a
compact normally hyperbolic invariant annulus. His assumptions can be formulated as
follows ([20],p.163):

Let Σ denote the Cantor set of all bi-infinite sequences of 0’s and 1’s, and σ : Σ → Σ
be the shift map defined for s = (si )i∈Z as σ(s)i = si+1 .
Let N be a smooth manifold and F : N → N be a C ℓ diffeomorphism. Let M ⊂ N
be a C ℓ normally hyperbolic invariant manifold for F such that the stable and unstable
manifolds W s,u(M ) intersect transversally in N . Then there exists in a neighbourhood of
M an invariant hyperbolic set Λ with the following properties:

i) Λ ∼ Σ × M .
ii) Λ is a C 0,ℓ Cantor set of manifolds, i.e. each leaf Ms = {s} × M , s ∈ Σ is C ℓ and
depends continuously on s in the C ℓ topology.
iii) some iterate F n |Σ is conjugate to a skew product over the shift

φ : Σ × M → Σ × M, φ(s, p) = (σ(s), gs (p)),

where gs : Ms → Mσ(s) .

Moeckel [20] refers to previous work of Shilnikov [24], Meyer and Sell [19] and Wiggins
[27].

Wiggins [27] proves an analogue of the Smale-Birkhoff theorem near a transversal ho-
moclinic normally hyperbolic invariant torus. However, this result can not be used to
justify Moeckel’s assumptions. Indeed, Wiggins’s result is based on:

• a particular normal form near the normally hyperbolic invariant torus obtained by
Shilnikov [24].
• an annulus is not a compact boundaryless manifold, contrary to the torus.

Moreover, in most applications the compact annulus is obtained by truncating a nor-

mally hyperbolic invariant cylinder, which is not compact. However, non-compactness can
be easily handled as it only provides technical difficulties. This is not the case when the
normally hyperbolic invariant manifold has a boundary, which leads to technical as well
as dynamical problems.
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 3

Other problems of importance deal with general compact boundaryless invariant nor-
mally hyperbolic manifolds, as for example normally hyperbolic invariant spheres. It has
recently been shown that normally hyperbolic invariant spheres are an important phase
space structure in Hamiltonian systems with three or more degrees-of-freedom. Specific
applications where they play a central role are cosmology [22], reaction dynamics [31], [25],
and celestial mechanics [26].

The proof of the Smale-Birkhoff theorem for normally hyperbolic invariant tori by Wig-
gins [27] is based on a generalized λ-lemma. This λ-lemma has been generalized by E.
Fontich and P. Martin [12] under more general assumptions and C 2 regularity for the map.

There are three settings where a new type of λ-lemma would be useful. What charac-
terizes the difference in each case is the geometrical structure of the normally hyperbolic
invariant manifold M .
• M is non-compact, and can be characterized by a global coordinate chart. This
situation arises when we consider normally hyperbolic invariant cylinders in Hamil-
tonian systems. Non-compactness is dealt with by assuming uniform bounds on
first and second derivatives of certain functions (cf. [13]).
• M is compact with a boundary. This situation arises when we truncate normally
hyperbolic invariant cylinders to form normally hyperbolic invariant annuli. The
technical difficulty is controlling the dynamics at the boundary.
• M is compact, but it cannot be described globally by a single coordinate chart.
This situation arises when we consider normally hyperbolic invariant spheres.
In this paper, we prove a λ-lemma for normally hyperbolic invariant manifolds, which
are not necessarily compact. This result allows us to prove a λ-lemma for a normally hy-
perbolic annulus, i.e. for a normally hyperbolic compact manifold with boundaries which
is a subset of a non-compact boundaryless normally hyperbolic manifold. We can also
use the same result to prove a λ-lemma for compact invariant manifolds that cannot be
described by a single coordinate chart.

The proof of the Smale-Birkhoff theorem as well as its applications for diffusion in
Hamiltonian systems will be studied in a forthcoming paper [9].

2. A λ-lemma for normally hyperbolic invariant manifolds

We first define the norms that we will use throughout this paper. Essentially, we will
only require two norms; one for vectors and one for matrices. All our vectors can be
4 JACKY CRESSON AND STEPHEN WIGGINS

viewed as elements of Rn (for some appropriate n) and our matrices will consist of real
entries. As a vector norm we will use the sup norm on Rn , denoted by | · |. Let Mm×n
denote the set of m × n matrices over R, n ≥ 1, m ≥ 1. An element of Mm×n has
the form A = (ai,j )i=1,...,m, j=1,...,n ∈ Mm×n . We define the norm of A ∈ Mm×n by
X
k A k= sup | ai,j |.
i j

2.1. Normally hyperbolic invariant manifolds. Let N be a n-dimensional smooth

manifold, n ≥ 3, and f : N → N be a C ℓ diffeomorphism, r ≥ 1. Let M be a bound-
aryless m-dimensional submanifold (compact or non compact) of class C ℓ of N , m < n,
invariant under f , such that:

i) M is normally hyperbolic,

ii) M has a m + ns -dimensional stable manifold W s (M ) and a m + nu -dimensional

unstable manifold W u (M ), with m + ns + nu = n.

Let p ∈ N , we denote by Dfp the derivative of f at p. Let TM N be the tangent

bundle of N over M . As M is normally hyperbolic, there exists a Df -invariant splitting
TM N = E s ⊕ E u ⊕ T M such that E s ⊕ T M is tangent to W s (M ) at M and E u ⊕ T M is
tangent to W u (M ) at M .

2.2. Normal form. We assume in the following that there exist a C ℓ coordinate systems
(s, u, x) ∈ Rns × Rnu × M in a neighbourhood U of M such that f takes the form:

(1) f (s, u, x) = (As (x) s, Au (x) u, g(x)) + r(s, u, x),

where r is the remainder, r = (rs (s, u, x), ru (s, u, x), rx (s, u, x)) with As , Au , g and r sat-
isfying

a) (invariance of M ) rs (0, 0, x) = ru (0, 0, x) = rx (0, 0, x) = 0 for all x ∈ M .

As a consequence, the set M is given in this coordinates system by

(2) M = {(s, u, x) ∈ U | s = u = 0},

and U can be chosen of the form

(3) U = Bρ × M,

with ρ > 0 and Bρ is the open ball defined by Bρ = {(s, u) ∈ Rns × Rnu ; | (s, u) |< ρ}.
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 5

Let ρ > 0, we denote by Bρs (resp. Bρu ) the open ball of size ρ in Rns (resp. Rnu ) around
0.

As M is normally hyperbolic, for ρ > 0 sufficiently small, the stable manifold theorem
(see [15],[28]) ensures that the stable and unstable manifolds can be represented as graphs,
i.e. there exist two C ℓ functions Gs (s, x) and Gu (u, x) such that
W s (M ) ∩ U = {(s, x) ∈ Bρs × M | u = Gs (s, x)},
(4)
W u (M ) ∩ U = {(u, x) ∈ Bρu × M | s = Gu (u, x)},
with

(5) Gs (0, x) = 0, ∂s Gs (0, x) = 0, ∂x Gs (0, x) = 0,

and

(6) Gu (0, x) = 0, ∂u Gu (0, x) = 0, ∂x Gu (0, x) = 0,

which reflect the tangency of W s (M ) and W u (M ) to E s and E u over M respectively.

Using these functions, we can find a coordinate system for which the stable and unstable
manifolds are “straightened”, i.e.

b) (straightening of the stable manifold) ru (s, 0, x) = 0 for all (s, 0, x) ∈ U ,

c) (straightening of the unstable manifold) rs (0, u, x) = 0 for all (0, u, x) ∈ U .

As a consequence, the stable and unstable manifolds of M are given by

W s (M ) = {(s, u, x) ∈ U | u = 0},
(7)
W u (M ) = {(s, u, x) ∈ U | s = 0}.
Indeed, following the classical work of Palis-deMelo [23], the change of variables
U −→ U
(8) Φ:
(s, u, x) 7−→ (s − Gu (u, x), u − Gs (s, x), x),
realizes the straightening:

For all P ∈ W s (M ), we denote by Φ(P ) = (s′ , u′ , x′ ). Then, P ∈ W s (M ) if and only if

Φ(P ) = (s′ , 0, x′ ) and P ∈ W u (M ) if and only if Φ(P ) = (0, u′ , x′ ).

d) (conjugacy on the stable and unstable manifold) We assume that rx (0, u, x) =

0 and rx (s, 0, x) = 0 for all s ∈ Bρs , u ∈ Bρu and x ∈ M .

In many examples of importance this condition is satisfied. It tells us that the dynamics
on the stable and unstable manifolds in the invariant manifold direction is given by the
6 JACKY CRESSON AND STEPHEN WIGGINS

dynamics on M . We refer to Graff [14] for such an example of rigidity in an analytic

context.

e) (hyperbolicity) k As (x) k≤ λ < 1, k Au (x)−1 k≤ λ < 1.

These results lead us to introduce the following definition of a normal form for diffeo-
morphisms near a normally hyperbolic invariant manifold:

Definition 1 (Normal form). Let N be a smooth manifold and f a C ℓ diffeomorphism

of N , ℓ ≥ 2. Let M be a compact normally hyperbolic invariant manifold of f . The
diffeomorphism is said to be in normal form if there exist a neighbourhood U of M and a
C ℓ coordinate system on U such that f takes the form (1) and satisfies conditions a)-e).

Standard results on normal form theory can be used to prove in some case that we
have a diffeomorphism in normal form. We refer to ([3],p.332) for a general normal form
theorem. In particular, we derive such a normal form in a Hamiltonian setting near a
normally hyperbolic cylinder.

Moreover, general normal form results for normally hyperbolic manifolds already imply
that our assumptions are general, at least if we restrict the regularity assumption on the
coordinates system to C 1 . Indeed, we have the following result due to M. Gidea and R.
De Llave [13]:

Theorem 1. Let N be a smooth manifold and f a C ℓ diffeomorphism of N , ℓ ≥ 2. Let M

be a normally hyperbolic invariant manifold of f (compact or non compact). There exists
a neighbourhood U of M and a C 1 coordinate system on U such that f is in normal form.

We refer to ([13],§.5.1) for a proof.

Of course, such a result is not sufficient for our purposes as we need some control on the
second order derivatives of g and r. However, it proves that our assumptions are general.
In the same paper, M. Gidea and R. De Llave [13] proves that we can take rx = 0, i.e.
that we have a decoupling between the center dynamics and the hyperbolic dynamics.

2.3. The λ-lemma. We have the following generalization of the toral λ-lemma of S.
Wiggins [27]:

Theorem 2 (λ-lemma). Let N be a smooth manifold and M be a C ℓ submanifold of N ,

normally hyperbolic, invariant under a C ℓ diffeomorphism f , ℓ ≥ 2, in normal form in a
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 7

given neighbourhood U of M and such that

i) There exists C > 0 such that

 2 ′

(9) sup k ∂σ,σ ′ ri (z) k, z ∈ U, σ ∈ {s, u, x}, σ ∈ {u, x}, i ∈ {s, x} ≤ C.

ii) There exists C̃ > 0 such that

 2
(10) sup k ∂σ,x g(z) k, z ∈ U, σ ∈ {s, u, x} ≤ C̃.

iii) There exists D > 0 such that for all x ∈ M , k ∂x As (x) k≤ D.

Let ∆ be an m + nu dimensional manifold intesecting W s (M ) transversally and let ∆k =

f k (∆)∩U be the connected component of f k (∆)∩U intersecting W s (M ). Then for ǫ > 0,
there exists a positive integer K such that for k ≥ K, ∆k is C 1 ǫ-close to W u (M ).

The proof follows essentially the same line as in ([27],p.324-329) and is given in section
4.

3. Hamiltonian systems and normally hyperbolic invariant cylinders and

annuli

Normally hyperbolic invariant annuli or cylinders are the basic pieces of all geometric
mechanisms for diffusion in Hamiltonian systems. This may seem to be an unusual state-
ment in light of the fact that the classical “transition chain” is a series of heteroclinic
connections of stable and unstable manifolds of “nearby” lower dimensional tori. How-
ever, these lower dimensional tori are contained in normally hyperbolic invariant annuli
and cylinders which have their own stable and unstable manifolds (which, in turn, contain
the stable and unstable manifolds of the lower dimensional tori used to construct Arnold’s
transition chains). The importance of normally hyperbolic invariant annuli or cylinders
can be clearly seen in the papers of Z. Xia [32], R. Moeckel [21], and A. Delshams, R. De
Llave and T. Seara [11] where normally hyperbolic annuli are a fundamental tool.

In this section, we prove a λ-lemma for Hamiltonian systems possessing, in a fixed

energy manifold, a normally hyperbolic manifold of the form T × I, where I is a given
compact interval, which belong to a non-compact boundaryless invariant manifold T × R.
We will see that the fact that a λ-lemma can be proven in this case is related to the fact
that the boundaries are partially hyperbolic invariant tori for which an analogue of the
λ-lemma for which a different type of λ-lemma has already been proven. Moreover, we
will also construct a class of three degree of freedom Hamiltonian systems which satisfy
our assumptions.
8 JACKY CRESSON AND STEPHEN WIGGINS

3.1. Main result. All of our results will be stated in terms of discrete time systems, or
maps. However, many of the applications we have in mind will be for continuous time
Hamiltonian systems. Our results will apply in this setting by considering an appropriate
Poincaré map for the continuous time system. It is important to keep this reduction from
continuous time Hamiltonian system to discrete time Poincaré map firmly in mind from
the point of view of considering the dimensions of the relevant invariant manifolds and
transversal intersection in the two systems. Finally, we note that even though the specific
applications we consider here are for Hamiltonian systems, as is true of most hyperbolic
phenomena, a Hamiltonian structure is not generally required for their validity. We now
describe the setting and our hypotheses for the applications of interest.

Let H be a C r Hamiltonian H(I, θ), (I, θ) ∈ R3 × T3 , r ≥ 3. The Hamiltonian defines

a C r−1 Hamiltonian vector field for which we make the following assumptions:

i) the Hamiltonian vector field possesses an invariant normally hyperbolic manifold Λ̄

which is diffeomorphic to T2 ×R2 , with 5-dimensional stable and unstable manifold W s (Λ̄)
and W u (Λ̄) respectively (note: these invariant manifolds are not isoenergetic).

ii) There exists a Poincaré section in a tubular neighbourhood of Λ̄ and a C 2 coordinate

system of the form (v, w, s, u) ∈ S 1 × R × R × R such that the Poincaré map is in normal
form near Λ̄, i.e. it takes the form

(11) f (s, u, v, w) = (As (z) s, Au (z) u, g(v), w) + r(s, u, v, w),

where r is the remainder, r = (rs (s, u, v, w), ru (s, u, v, w), rx (s, u, v, w)) with As , Au , g
and r satisfying assumptions a)-e) of section 2.2. We denote the intersection of Λ̄, W s (Λ̄),
and W u (Λ̄) with the Poincaré section by Λ, W s (Λ), and W u (Λ), respectively. The 4-
dimensional Poincaré section is chosen such that the intersection of these manifolds with
the Poincaré section is isoenergetic, and Λ is 2-dimensional, W s (Λ) is 3-dimensional, and
W u (Λ) is 3-dimensional.
iii) There exist two circles C0 and C1 belonging to Λ and invariant under f , possessing
2 dimensional stable and unstable manifolds.
iv) We assume that a λ-lemma is valid for the partially hyperbolic invariant circles C0
and C1 .

Assumptions iv) can be made more precise as there already exists many versions of the
λ-lemma for partially hyperbolic tori. We refer in particular to [6],[7],[18] and [12], which
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 9

is the most general.

We denote by A the invariant normally hyperbolic annulus whose boundaries are C0

and C1 . W s (A) ⊂ W s (Λ) is the stable of A and W u (A) ⊂ W u (Λ) is the unstable manifold
of A respectively. We define the boundary of W s (A) to be W s (C0 ) and W s (C1 ) and the
boundary of W u (A) to be W u (C0 ) and W u (C1 ). The important point here is that even
though A has a boundary, it is still invariant with respect to both directions of time. This
is because its boundary is an invariant manifold. Similarly, W s (A) and W u (A) are also
invariant manifolds (cf. with inflowing and outflowing invariant manifolds with boundary
described in [28]).

Our main result is the following:

Theorem 3 (λ-lemma for normally hyperbolic annuli). Let H denote the Hamiltonian
for a three degree of freedom Hamiltonian system satisfying assumptions i)-iv). Let ∆ be
a 3 dimensional manifold intersecting W s (Λ) transversally. We assume that there exists a
˜ ⊂ ∆, such that ∆
subset ∆ ˜ intersects W s (A) transversally, and such that the boundaries
˜ 0 and ∂ ∆
∂∆ ˜ 1 intersect transversally the stable manifolds of C0 and C1 respectively. Then,
for all ǫ > 0, there exists a positive integer K such that for all k ≥ K, f k (∆) is C 1 ǫ-close
to W u (A).

The proof follows from our previous Theorem 2 for the noncompact case and the λ-
˜ In considering the
lemma for partially hyperbolic tori to control the boundaries of ∆.
proof of Theorem 2 one sees that the difficulty arising for invariant manifolds is that
iterates of points may leave the manifold by crossing the boundary. This is dealt with
here by choosing the boundary of the manifold to also be (lower dimensional) invariant
manifold(s).

Proof. Since the boundary of W s (A) is invariant, we know that under iteration by f
Λ̃ ∩ W s (A) is always contained in W s (A). Hence we can use the λ-lemma proven in
Theorem 2 for non-compact normally hyperbolic invariant manifolds to conclude that for
˜ is C 1 ǫ-close to
all ǫ > 0, there exists a positive integer K such that for all k ≥ K, f k (∆)
W u (A).
˜ correctly accumulate on the boundaries
We also need to show that the boundaries of ∆
of W u (A) which are formed by W u (C0 ) and W u (C1 ). This follows by applying the λ-
lemma for partially hyperbolic tori of Fontich and Martin [12] to C0 and C1 . It follows
from this lemma that for all ǫ > 0, there exists a positive integer K ′ such that for all
˜ i ) is C 1 ǫ-close to W u (Ci ) for i = 0, 1. This concludes the proof.
k ≥ K ′ , f k (∂ ∆ 
10 JACKY CRESSON AND STEPHEN WIGGINS

We remark that if one is in a neighborhood of Λ where the manifolds are “straightened”

˜ as a graph over W u (A) such that it intersects W s (A)
it is likely that one can construct ∆
transversally, and such that the boundaries ∂ ∆˜ 0 and ∂ ∆
˜ 1 intersect transversally the stable
manifolds of C0 and C1 , respectively,

In the following section we prove that our assumptions are satisfied in a large class of
near integrable Hamiltonian systems.

3.2. An example. In order to construct a normally hyperbolic cylinder for a near inte-
grable Hamiltonian system, we follow the same geometrical set-up as in the seminal paper
of Arnold [1]. We emphasize that the results described above are not perturbative in na-
ture, but we have in mind near-integrable systems as a setting where our results provide
a key ingredient for proving the existence of symbolic dynamics an instability mechanism.

3.2.1. Variation around the Arnold example. We consider the near integrable Hamiltonian
system

1 2 1 2
(12) Hǫ,µ(p, I, J, q, θ, φ) = p + I + J + ǫ(cos q − 1) + ǫf (θ, φ) + µ(sin q)α(ν,σ) g(θ, φ),
2 2
where as usual, (I, J, θ, φ) ∈ R × R × T × T are action-angle variables, (p, q) ∈ R × T,
0 < ǫ << 1 is a small parameter and µ is such that 0 < µ << ǫ, and f and g are two
given smooth functions, and ν ∈ N is a parameter  controlling
 the order of contact be-
log ν
tween Hǫ,µ and Hǫ,0 via the function α(ν, σ) = 2 + 1 introduced in ([16], equation
4σ
(2.5)), σ > 0 and ν ≥ νσ where νσ is the smaller positive integer such that α(ν) = 2 and
α(ν, σ) ≥ 2 for ν ≥ νσ .

For ǫ = µ = 0 the system is completely integrable and the set

(13) Λ̄ = {(p, I, J, q, θ, φ) ∈ R × R × R × T × T × T | p = q = 0},

is invariant under the flow and normally hyperbolic. The dynamics on Λ̄ is given by the
completely integrable Hamiltonian system
1
(14) H(I, θ) = I 2 + J.
2
As a consequence, the set Λ̄ is foliated by invariant 2-tori.

For µ = 0 and ǫ 6= 0, the set Λ̄ persists, but the dynamics on Λ̄ is no longer integrable.
In particular, it is not foliated by invariant two tori. However, the KAM theorem applies
and we have a Cantor set of invariant two tori whose measure tends to the full measure
when ǫ goes to zero. As a consequence, we are in a situation where the “large gaps”
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 11

problem arises ([17] ), contrary to the well known example of Arnold [1] where all the
foliation by invariant two tori is preserved under the perturbation.

−1
Let h be a given real number. We denote by Hǫ,0 = Hǫ,0 (h) the energy manifold. There
exists a global cross-section to the flow denoted by S and defined by

(15) S = {(p, I, J, q, θ, φ) ∈ R × R × R × T × T × T | φ = 0}.

We denote by Λ the intersection of Λ̄ with S ∩ Hǫ,0 . We can find a symplectic analytic

coordinate system on S ∩ Hǫ,0 , denoted by (x, y, s, u) ∈ T × R × R × R such that the set
Λ is defined by

(16) Λ = {(x, y, s, u) ∈ T × R × R × R; s = u = 0 }.

Geometrically, Λ is a cylinder. This cylinder is a normally hyperbolic boundaryless man-

ifold, but not compact.

We can introduce the compact counterpart, which is a normally hyperbolic annulus, but
now with boundaries. Let T0 and T1 be two invariant 2-dimensional partially hyperbolic
tori belonging to Λ̄. We denote by C0 and C1 the intersection of T0 and T1 with S ∩ Hǫ,0 .
The invariant circles C0 and C1 are defined as

(17) Ci = {(x, y, s, u) ∈ T × R × R × R; y = yi , s = u = 0 },

for yi ∈ R well chosen, i = 0, 1. We assume in the following that y0 < y1 .

The compact counterpart of Λ is then defined as

(18) A = {(x, y, s, u) ∈ T × R × R × R; y0 ≤ y ≤ y1 , s = u = 0 }.

Let P be the Poincaré first return map associated to S ∩ Hǫ,0 . The dynamics on A is
given by an ǫ perturbation of an analytic twist map, i.e. that P |A is defined by

(19) P |A (x, y) = (x + ω(y), y) + ǫr(x, y),

where ω ′ (y) 6= 0 and r(x, yi ) = 0 for i = 0, 1.

When ǫ 6= 0 and µ 6= 0, then the set Λ persists since the perturbation vanishes on Λ.
However, the stable and unstable manifold of Λ intersect transversally for a well chosen
perturbation g.
12 JACKY CRESSON AND STEPHEN WIGGINS

We then obtain an example of a three degrees of freedom Hamiltonian system satisfying

the geometrical assumptions i), iii) and iv) of section 3.1. We prove in the following section
that the analytic assumption ii) is satisfied.

3.2.2. Poincaré section and normal form. The main problem is to prove that the diffeo-
morphism of the cross-section to the flow defined in a neighbourhood of Λ is in normal
form with respect to definition 1.

The basic theorem which we use to obtain a smooth normal form near a normally
hyperbolic manifold is a generalized version of the Sternberg linearization theorem. In
the compact case, this result has been obtain by Bronstein and Kopanskii ([3], theorem
2.3,p.334). The non compact case has been proven by P. Lochak and J-P. Marco [16]. This
theorem, which can be stated for flows or maps, ensures that we can obtain a conjugacy
as smooth as we want between φǫ,µ and φǫ,0 by choosing the two flows with a sufficiently
high order of contact (see [3],p.334). A key remark in our case is that the flows φǫ,0 and
φǫ,µ generated by Hǫ,0 and Hǫ,µ, respectively, have contact of order α(ν, σ) on Λ̄. More-
over, we can obtain an arbitrary order of contact between the two flows by choosing the
parameter ν sufficiently large. As a consequence, we will always be able to realize the
assumptions of the Sternberg linearization theorem for normally hyperbolic manifolds and
as a consequence, to obtain a normal form as smooth as we desire.

Before stating the normal form theorem, which is only a minor modification of the result
of Lochak-Marco (see [16], Theorem D), we introduce some notation:

As a general notation, for any manifold M , we denote by Mρ a tubular neighbourhood

of M of radius ρ. We let fν and f∗ denote the Poincaré maps defined in a neighborhood
of Λ obtained from the flows generated by the Hamiltonians Hǫ,µ and Hǫ,0 , respectively.

Using the Sternberg linearization theorem for normally hyperbolic manifolds proved in
[16], we obtain the following theorem:

Theorem 4 (Normal form). For ν0 large enought, there exist ρ, ρ′ with 0 < ρ′ < ρ such
that for all ν ≥ ν0 , fν (Λ2ρ′ ) ⊂ Λρ and there exists a C k diffeomorphism φν (k ≥ 1)
satisfying Λρ′ ⊂ φν (Λ2ρ′ ) ⊂ Λρ and φν ◦ f∗ = fν ◦ φν on Λρ′ . Moreover, there exists a
constant a (0 < a < 1)such that k φ±1 α(ν) , where k · k k
ν − Id kC k ≤ a C k denotes the C
norm on Λρ′ .
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 13

This theorem is a direct application of a result of Lochak-Marco [16], and we refer to

their paper for more details (in particular Theorem D).

Theorem 4 implies that the λ-lemma proven in Theorem 3 applies to the three degree-
of-freedom Hamiltonian systems defined by (12).

In this case one can, moreover, prove an analogue of the Smale-Birkhoff theorem using
the fact that symbolic dynamics is stable under small C 1 perturbations and taking for
the Poincaré map near the normally hyperbolic invariant annulus the linear mapping. A
complete study of this problem will be done in [9] in the context of homoclinic normally
hyperbolic invariant manifolds.

4. Proof of the λ-lemma

4.1. Preliminaries. In this section we develop the set-up for the proof of the λ-lemma.
First we discuss some useful consequences of the normal form assumptions for the diffeo-
morphism f .

• We denote by Dr(p) the differential of r at point p ∈ U . By invariance of M (as-

sumption a) in section 2.2) we have Dr(p) = 0 for all p ∈ M . Let 0 < k < 1, since r is a
C 1 function, we have for U sufficiently small and for all p ∈ U ,

(20) k Dr(p) k≤ k.

This implies in particular that for all p ∈ U , the partial derivatives ∂i rj (p), i, j ∈ {s, u, x}
satisfy k ∂i rj (p) k≤ k.

We take U sufficiently small in order to have the following inequalities satisfied for k:

(21) 0 < λ + k < 1,

(22) λ−1 − k > 1.

• The straightening conditions b) and c) imply that:

∀p ∈ W s (M ) ∩ U, ∂s ru (p) = ∂x ru (p) = 0,
(23)
∀p ∈ W u (M ) ∩ U, ∂u rs (p) = ∂x rs (p) = 0.
14 JACKY CRESSON AND STEPHEN WIGGINS

• The conjugacy assumption d) implies that:

∀p ∈ W s (M ) ∩ U, ∂s rx (p) = ∂x rx (p) = 0,
(24)
∀p ∈ W u (M ) ∩ U, ∂u rx (p) = ∂x rx (p) = 0.
4.2. Notation. Let v0 be a unit vector in the tangent bundle to ∆n . We denote the com-
ponents of v0 as v0 = (v0s , v0u , v0x ) where v0s ∈ E s , v0u ∈ E u and v0x ∈ T M . We denote by
vn the iterate of v0 under Df n , i.e. vn = Df n (v0 ) and vn = (vns , vnu , vnx ). Let p = (s, u, x),
we denote by pn = f n (p) the n-th iterate of p and pn = (sn , un , xn ) its components. We
identify p0 and p.

Let p ∈ U , then using (1) we have

 
As (x) + ∂s rs (p) ∂u rs (p) ∂x rs (p) + ∂x As (x) s
(25) Dfp =  ∂s ru Au (x) + ∂u ru (p) ∂x ru (p) + ∂x Au (x) u  .
∂s rx (p) ∂u rx (p) ∂x g(x) + ∂x rx (p)
We are going to prove that for all v0 ∈ T ∆n , and for n sufficiently large, we have

(26) lim sup {Inx , Ins } = 0,

n→+∞
where the inclinations are defined as:

| vnx | | vns |
(27) Inx ≡ , Ins ≡ ,
| vnu | | vnu |
and

u
| vn+1 |
(28) u
> 1.
| vn |
(26) implies that under iteration arbitrary tangent vectors align with tangent vectors to
the unstable manifold and (28) implies that these tangent vectors also grow in length.
The proof essentially involves three steps. First we prove (28) for tangent vectors in
∆n ∩ W s (M ). Next we extend this result to tangent vectors in ∆n . Finally, we prove
(28).

4.3. Inclinations for tangent vectors in the stable manifold. We first prove (26)
for v0 in the tangent bundle of ∆n ∩ W s (M ), denoted TW s (M ) ∆n . Let p ∈ W s (M ), then
by the invariance property of the stable manifold (25) simplifies to:

 
As (x) + ∂s rs (p) ∂u rs (p) ∂x rs (p) + ∂x As (x) s
(29) Dfp =  0 Au (x) + ∂u ru (p) 0 .
0 ∂u rx (p) ∂x g(x)
Acting on the tangent vector (vns , vnu , vnx ), we obtain the following relations
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 15

s
(30) vn+1 = (As (xn ) + ∂s rs (pn ))vns + ∂u rs (pn ) vnu + (∂x rs (pn ) + ∂x As (xn ) sn )vnx ,
u
(31) vn+1 = (Au (xn ) + ∂u ru (pn ))vnu ,
x
(32) vn+1 = ∂x ru (pn ) vnu + ∂x g(xn ) vnx .

Using these expressions, along with the estimates (20), (21), and (22), we then obtain

s
(33) | vn+1 | ≤ (λ + k) | vns | + k ∂u rs (pn ) k | vnu | +(k+ k ∂x As (xn ) k | sn |) | vnx |,
u
(34) | vn+1 | ≥ (λ−1 − k) | vnu |,
x
(35) | vn+1 | ≤ k ∂u rx (pn ) k | vnu | +k | vnx | .

4.3.1. Inclination in the tangential direction. Using (35), (34) and (22) gives:

x
| vn+1 | k | vnx |
(36) u ≤ + k ∂u rx (pn ) k .
| vn+1 | λ−1 − k | vnu |
Using the estimate(9) on the second derivatives with the mean value inequality gives:

x
| vn+1 | k | vnx |
(37) u ≤ + C | sn | .
| vn+1 | λ−1 − k | vnu |
Let p = (s, 0, x) ∈ W s (M ) ∩ U and pn = f n (p) = (sn , 0, xn ). By definition, we have
sn+1 = As (xn )sn + rs (sn , 0, xn ). Estimating this expression using assumption d) of section
1, as well as the mean value inequality with (20) gives:

(38) | sn+1 |≤ (λ + k) | sn |,

from which it follows that:

(39) | sn |≤ (λ + k)n | s | .

Replacing | sn | by this expression in (37), we obtain

x
| vn+1 | k | vnx |
(40) u ≤ + C | s | (λ + k)n .
| vn+1 | λ−1 − k | vnu |
As a consequence, we have

n n−1  i
| vnx | | v0x |

k X k
(41) ≤ + C | s | (λ + k)n−1−i .
| vnu | −1
λ −k | v0u | −1
λ −k
i=0
16 JACKY CRESSON AND STEPHEN WIGGINS

From (21) and (22) it follows that:

k
(42) < k + λ,
λ−1 −k
Using this we obtain

n
| vnx | | v0x |

k
(43) ≤ + C | s | n(λ + k)n−1 .
| vnu | −1
λ −k | v0u |
From (21) we have λ + k < 1, and therefore:

| vnx |
(44) lim u |
= 0.
n→+∞ | vn

4.3.2. Inclination in the stable direction. Using (33) and (34), we obtain

s
| vn+1 | | vns | | vnx |
 
λ+k
(45) u ≤ + k ∂u rs (p n ) k +(k+ k ∂x As (xn ) k | s n |) .
| vn+1 | λ−1 − k | vnu | | vnu |
Using the assumption (9), the mean value inequality, and (39) gives:

(46) k ∂u rs (sn , 0, xn ) k≤ C | sn |≤ C | s | (λ + k)n .

Moreover, recall that from assumption iii) in the statement of the λ-lemma we have:

(47) k ∂x As (xn ) k≤ D.

Using these two estimates, (45) becomes:

s
| vn+1 | | vns | x
 
λ+k n n | vn |
(48) u ≤ + C | s | (λ + k) + (k + D | s | (λ + k) ) .
| vn+1 | λ−1 − k | vnu | | vnu |
As a preliminary step to estimating (48), we first estimate the third term on the right-
hand-side of (48) using (43):

(49) n
| vx | | v0x |
 
k
(k+D | s | (λ+k) ) nu ≤ (k+D | s | (λ+k)n )
n
+ C | s | n(λ + k)n−1
| vn | λ−1 − k | v0u |
Now for n sufficiently large we have:

(50) k + D | s | (λ + k)n < 1,

 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 17

and by assumption (22) we have λ−1 − k > 1, and therefore:

| vnx | x
n | v0 |
(51) (k + D | s | (λ + k)n ) ≤ k + C | s | n(λ + k)n−1 .
| vnu | | v0u |
Substituting this expression into (48) gives:

s
| vn+1 | | vns | x
 
λ+k n n−1 n | v0 |
u ≤ + C | s | (λ + k) + C | s | n(λ + k) + k ,
| vn+1 | | vnu |
λ−1 − k | v0u |
 s
| v0x |
  
λ+k | vn | n−1
(52) ≤ + (λ + k) C | s | (1 + n) + u
λ−1 − k | vnu | | v0 |
From this expression we obtain:

n n−1 
s
| vn+1 | λ+k i
| v0s | X | v0x |
    
λ+k n−i−2
u ≤ + (λ + k) C | s | (n − i) + ,
| vn+1 | λ−1 − k | v0u |λ−1 − k | v0u |
i=0
λ + k n | v0s | | v0x |
   
n−2
(53) ≤ + (λ + k) n C|s|+ u
λ−1 − k | v0u | | v0 |
Consequently, as n goes to infinity, we have

| vns |
(54) lim u |
= 0.
n→∞ | vn

4.4. Extending the estimates to tangent vectors not in TW s (M ) ∆n . We have shown

that for n sufficiently large, U sufficiently small, for all v ∈ TW s (M ) ∆n , | v |= 1,

(55) I(v) ≤ ǫ.
˜ n ⊂ ∆n such that
By continuity of the tangent plane, there exists ∆

(56) ˜ n , I(v) ≤ 2ǫ.

∀v ∈ T ∆

For n sufficiently large ∆n is very close to W u (M ). We choose a neighbourhood Vǫs of

W u (M ) of the form

(57) Vǫs = {(s, u, x) ∈ Bǫss × Bρu × M },

for 0 < ǫs < 1. Recall the estimates

k ∂u rx (p) k≤ δ, k ∂x rx (p) k≤ δ,
(58) ∀p ∈⊂ U,
k ∂u rs (p) k≤ δ, k ∂x rs (p) k≤ δ.
18 JACKY CRESSON AND STEPHEN WIGGINS

Using the mean value theorem and (9), we choose ǫs such that:

(59) δ ≥ Cǫs .

As a consequence, we can take

(60) δ = (C + 1)ǫs .

Using (25), analogously to the estimates above we obtain the following:

s
(61) | vn+1 | ≤ (λ + k) | vns | +δ | vnu | +(Cǫs + δ) | vnx |,
u
(62) | vn+1 | ≥ (λ−1 − k) | vnu | −k | vns | −(k + ρC) | vnx |,
x
(63) | vn+1 | ≤ k | vns | +δ | vnu | +(k ∂x g(xn ) k +δ) | vnx | .

We use these expressions to obtain the following estimates of the inclinations:

x 1
(64) In+1 ≤ [kIns + δ + (k ∂x g(xn ) k +δ)Inx ] µ,
λ−1 − k
s 1
(65) In+1 ≤ −1
[(λ + k)Ins + δ + (Cǫs + δ)Inx ] µ,
λ −k
where

k k + Cρ x
(66) µ−1 = 1 − Is − I .
λ−1 − k n λ−1 − k n
2 g(z) k, z ∈ U }, we obtain the following estimates:
Since δ = (C + 1)ǫs , and C̃ ≥ sup{k ∂i,x ρ

(67) (k ∂x g(xn ) k +δ)Inx ≤ ǫs (C̃ + C + 1)Inx ,

(68) δ + (Cǫs + δ)Inx ≤ ǫs [C + 1 + Inx (2C + 1)] .

We substitute these estimates into (64) and (65), and assuming that

(69) Inx ≤ ǫ and Ins ≤ ǫ,

we obtain:

x µ∗
(70) In+1 ≤ (kǫ + (C + 1)ǫs + ǫǫs (M + C + 1)) ,
λ−1 −k
s µ∗
(71) In+1 ≤ −1
((λ + k)ǫ + ǫs (C + 1 + (2C + 1)ǫ) ,
λ −k
2k + Cρ
where µ−1
∗ =1− ǫ.
λ−1 − k
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 19

Now if we choose k small enough such that:

λ+k
(72) µ∗ < 1,
λ−1 − k
and ǫs satisfies:

    
λ−1 −k kµ∗ 1


 ǫ µ∗ 1− λ−1 −k C+1+ǫ(M +C+1) ,
(73) ǫs ≤ inf  
λ+k 1

 ǫ 1−
λ−1 −k µ∗

C+1+(2C+1)ǫ

then

x
(74) In+1 ≤ ǫ,
s
(75) In+1 ≤ ǫ.
x
Hence, we have shown that for k and ǫs sufficiently small, In+1 s
≤ ǫ and In+1 ≤ ǫ. There-
fore the estimates Ins ≤ ǫ and Ins ≤ ǫ are maintained under iteration.

As ǫ can be chosen as small as we want, the inclinations Ins and Inx is as small as we
want for n sufficiently large.

4.5. Stretching along the unstable manifold. We want to show that f n (∆) ∩ U is
stretched in the direction W u (M ). In order to see this we compare the norm of a tangent
˜ n with its image under Df :
vector in ∆

s v
s u x u
u
s

2 x

2
| vn+1 |2 + | vn+1 |2 + | vn+1 |2 | vn+1 t1
|u + In+1 + In+1
(76) = .
| vns |2 + | vnu |2 + | vnx |2 u
| vn | 1 + (Ins )2 + (Inx )2
Using (62) we obtain:

u
| vn+1 |
(77) u
≥ λ−1 − k − kǫ − (k + Cρ)ǫ.
| vn |
Since ǫ can be chosen arbitrarily small we have:

(78) λ−1 − k − kǫ − (k + Cρ)ǫ ≥ λ−1 − 2k > 1.

Since the inclinations are arbitrarily small, it follows that the norms of (nonzero) vectors
˜ n are growing by a ratio that approaches λ−1 − 2k > 1. Therefore the diameter of ∆
in ∆ ˜n
is increasing. Putting this together with the fact that the tangent spaces have uniformly
˜ n is C 1 ǫ-close to W u (M ).
small slope implies that there exists ñ such that for all n ≥ ñ ∆
20 JACKY CRESSON AND STEPHEN WIGGINS

This concludes the proof.

References
[1] Arnold V.I., Instability of dynamical systems with several degrees of freedom, Soviet Math. Dokl., 5,
1964, 581.
[2] Bourgain J., On diffusion in high-dimensional Hamiltonian systems and PDE, Journal d’Analyse
Mathématique, Vol. 80, 1-35, 2000.
[3] Bronstein I.U., Kopanskii A. Y., Smooth invariant manifolds and normal forms, World Scientific series
on Nonlinear science, Series A, Vol. 7, 1994.
[4] Beigie D., Leonard A., Wiggins S., Chaotic transport in the homoclinic and heteroclinic tangle regions
of quasiperiodically forced two-dimensional dynamical systems, Nonlinearity 4 (1991) 775-819.
[5] Cresson J., Instabilité des systèmes hamiltoniens presques intégrables, Thesis, 188.p, 1997.
[6] Cresson J., A λ-lemma for partially hyperbolic tori and the obstruction property, Letters in Mathe-
matical Physics, Vol. 42, no.4, 363-377, 1997.
[7] Cresson J., Un λ-lemme pour des tores partiellement hyperboliques, C. R. Acad. Sci. Paris, t. 331,
Série I, 65-70, 2000.
[8] Cresson J., Symbolic dynamics and Arnold diffusion, Journal of differential equations 187, 269-292
(2003).
[9] Cresson J., Wiggins S., A Smale-Birkhoff theorem for normally hyperbolic manifolds, preprint, 2005.
[10] Cresson J., Wiggins S., About the Moeckel mechanism of diffusion, in preparation, 2005.
[11] Delshams A., De La Llave R., Seara T.M., A geometric mechanism for diffusion in Hamiltonian
systems overcoming the large gap problem: heuristics and rigorous verification on a model, to appear
in Memoirs of the American Mathematical Society, 2005.
[12] Fontich E., Martin P., Differentiable invariant manifolds for partially hyperbolic tori and a lambda
lemma, Nonlinearity 13 (2000), 1561-1593.
[13] Gidea M., De La Llave R., Topological methods in the instability of Hamiltonian systems, preprint,
2005.
[14] Graff S., On the conservation of hyperbolic invariant tori for Hamiltonian systems, J. Diff. Equ. 15
(1974), 1-69.
[15] Hirsh M.W., Pugh C.C., Shub M., Invariant manifolds, Springer Lecture Notes in Mathematics, Vol.
583, Springer-Verlag, 1977.
[16] Lochak P., Marco J-P., Diffusion time and stability exponents for nearly integrable Hamiltonian
systems, Central European Science Journal, 3 (2005), 342-397.
[17] Lochak P., Arnold diffusion: a compendium of remarks and questions, in Hamiltonian systems with
three or more degrees of freedom, C.Simo ed., Kluwer Acad. Publ., (1999).
[18] Marco J-P, Transition le long de chaı̂nes de tores invariants pour les systèmes hamiltoniens analytiques,
Ann. Inst. Henri Poincaré 64, no. 2, 205-252, 1996.
[19] Meyer K., Sell G., Melnikov transforms, Bernoulli bundles, and almost periodic perturbations, Trans.
AMS, 314, 1, p. 63.
[20] Moeckel R., Generic drift on Cantor sets of Annuli, Contemporary Mathematics, Vol. 292, 163-171,
2002.
[21] Moeckel R., Transition tori in the five body problem, Jour. Diff. Eq., 129, 2, 1996, p. 290.
[22] Oliveira H.P, Ozorio de Almeida A.M., Damiao Soares I., Tonini E.V., Homoclinic chaos in the
dynamics of the general Bianchi-IX model, Physical Review D, 65, 2002.
[23] Palis J., De Melo W., A geometrical introduction to dynamical systems, Spinger, 1982.
[24] Shilnikov L.P., Structure of the neighbourhood of a homoclinic tube of an invariant torus, Sov. Math.
Dokl., 9,3, 1964, p. 624.
[25] Waalkens H., Burbanks A., Wiggins S., Phase space conduits for reaction in multidimensional systems:
HCN isomerization in three dimensions, J. Chem. Phys. 121, 6207-6225 (2004).
[26] Waalkens H., Burbanks A., Wiggins S., Escape from planetary neighborhoods, Mon. Not. R. Astron.
Soc. 361, 763-775 (2005).
[27] Wiggins S., Global bifurcation and Chaos - Analytical methods, Springer-Verlag, New-York, Heidelberg,
Berlin, 1988.
[28] Wiggins S., Normally hyperbolic manifolds in dynamical systems, Springer-Verlag, New-York, 1994.
[29] Wiggins S., Chaotic transport in dynamical systems, Interdisciplinary Applied Mathematics, Vol. 2,
1992.
 A λ-LEMMA FOR NORMALLY HYPERBOLIC INVARIANT MANIFOLDS 21

[30] Wiggins S., On the geometry of transport in phase space I. Transport in k-degree of freedom Hamil-
tonian systems, 2 ≤ k < ∞, Physica D 44 (1990), 471-501.
[31] Wiggins, S., Wiesenfeld L., Jaffé C., Uzer T., Impenetrable barriers in phase space, Phys. Rev. Lett.
86, 5478-5481 (2001).
[32] Xia Z., Arnold diffusion and oscillatory solutions in the planar three-body problem, J. Differential
Equations 110 (1994), 289-321.

Université de Franche-Comté, Équipe de Mathématiques de Besançon, CNRS-UMR 6623,

16 route de Gray, 25030 Besançon cedex, France.
E-mail address: [email protected]

School of Mathematics, University of Bristol, University Walk, Bristol, BS8 1TW,

United Kingdom.
E-mail address: [email protected]