Persistent homoclinic tangencies and infinitely many sinks for residual sets of automorphisms of
low degree in C 3
Contents
1 Introduction 1
1.1 Background . . . . 1 1.2 Results and outline . . . . 3
2 Preliminaries 4
2.1 One complex dimension . . . . 4 2.2 Two complex dimensions . . . . 6 2.3 Three complex dimensions . . . . 7 3 Structure of the basic set : unperturbed case 9 3.1 Generalities and main result in the unperturbed case . . . . 9 3.2 Proof of Proposition 3.1.7 . . . . 12 4 Structure of the basic set : perturbed case 14 4.1 Generalities and main result in the perturbed case . . . . 14 4.2 Proof of Proposition 4.1.8 . . . . 16 4.3 Central curves . . . . 19
5 First heteroclinic tangency 20
5.1 Perturbation . . . . 20 5.2 Motion of the point of tangency . . . . 22
6 Persistent homoclinic tangencies 23
6.1 Persistent heteroclinic tangencies . . . . 23 6.2 Persistent homoclinic tangencies . . . . 25
7 Proof of the main result 26
A From homoclinic tangencies to sinks 27
References 29
1 Introduction
1.1 Background
Hyperbolic systems such as the horseshoe introduced by Smale were orig-
inally supposed to be dense in the set of parameters in the 1960’s. This was
quickly discovered to be false in general for diffeomorphisms of manifolds of dimension greater than 2 (see [1]). The discovery in the seventies of the so- called Newhouse phenomenon, i.e. the existence of residual sets of C
2diffemor- phisms of compact surfaces with infinitely many sinks (periodic attractors) in [15] showed it was false in dimension 2 too. In a subsequent work, Newhouse showed (see [16]) that such sets appear in fact close to any diffeomorphism with an homoclinic tangency. The technical core of the proof is the reduction to a line of tangency between the stable and unstable foliations where two Cantor sets must have persistent intersection. This gives persistent homoclinic tangencies between the stable and unstable foliations, ultimately leading to infinitely many sinks. Indeed, it is a well known fact that a sink is created in the unfolding of a generic homoclinic tangency.
Palis and Viana showed in [17] an analogous result for real diffeomorphisms in higher dimensions. More precisely, they proved that near any smooth dif- feomorphism of R
3exhibiting a homoclinic tangency associated to a sectionally dissipative saddle, there is a residual subset of an open set of diffeomorphisms such that each of its elements displays infinitely many coexisting sinks.
In the complex setting, this reduction is not possible anymore and to get persistent homoclinic tangencies and then residual sets of diffeomorphisms dis- playing infinitely many sinks, we have to intersect two Cantor sets in the plane.
This was done by Buzzard who proved in [7] that there exists d > 0 such that there exists an automorphism G ∈ Aut
d( C
2) and a neighborhood N ⊂ Aut
d( C
2) of G such that N has persistent homoclinic tangencies. Then, there is a resid- ual subset of Aut
d( C
2) of automorphisms with infinitely many sinks. In fact, after extending the stable and unstable foliations of a basic set, there is still a complex disk of tangency where intersections with the two foliations are two Cantor sets in two (real) dimensions. Buzzard gives an elegant criterion (see [6]) which generates the intersection of two such Cantor sets, hence leading to persistent homoclinic tangencies. More precisely, he gets persistent intersections between a "spiralic" Cantor set and a second Cantor set with high topological dimension. In his article, Buzzard uses a Runge approximation argument to get a polynomial automorphism, which implies that the degree d remains unknown and is supposedly very high.
In the article [4], Bonatti and Diaz introduce a type of horseshoe they called
blender horseshoe. The important property of such hyperbolic sets lies in the
fractal configuration of one of their stable/unstable manifold which implies per-
sistent intersection between any well oriented graph and this foliation. In some
sense, the foliation behaves just as it had greater Hausdorff dimension than ev-
ery individual manifold of the foliation. They find how to get robust homoclinic
tangencies for some C
r-diffeomorphism of R
3with an homoclinic tangency by
some geometric intersection procedure using the properties which define the
blender in [5] . In the article [9] , one can find real polynomial maps of degree
2 with a blender. Other studies using blenders include [8], [2] and [3].
1.2 Results and outline
In this article, we adapt a complex blender. More precisely, we solve here the degree problem in dimension 3 by introducing a kind of 3-dimensional central- stable complex blender. Here is our main result :
Main Theorem. There exists f ∈ Aut
5( C
3) such that for every g ∈ Aut( C
3) sufficiently close to f , g admits a tangency between the stable and unstable laminations of some hyperbolic set.
Notice that in the previous result, g is not assumed to be polynomial.
Corollary 1. For each d ≥ 5, there exists an open set V ⊂ Aut
d( C
3) in which the automorphisms having a homoclinic tangency are dense.
Corollary 2. For each d ≥ 5, there exists an open set V ⊂ Aut
d( C
3) in which the automorphisms having infinitely many sinks are dense.
Let us remark that there are classes of interesting polynomial automorphisms of C
3called regular and semi-regular automorphisms which have received much attention due to their interesting dynamical properties (e.g [13],[18]). It is pos- sible to choose the automorphisms to be regular or semi-regular in the above results because the condition of being regular or semi-regular is dense for the Zariski topology.
Let us present quickly our method to prove this result. We consider an au- tomorphism of C
3which is a perturbation of the following map f
1:
f
1: (z
1, z
2, z
3) → (p(z
1) + b.z
2, z
1, λ.z
1+ µ.z
3) (1)
where p is a polynomial and the coefficients b, λ, µ are complex numbers. Then, we prove that f
1has a hyperbolic set with the property that the stable lamina- tion of dimension 2 can be fixed with enough freedom while the 1-dimensional unstable lamination presents one direction where the lamination is very thick.
Moreover, this hyperbolic set has a fixed point which is sectionally dissipa- tive. Perturbating the automorphism f
1creates an initial heteroclinic tangency.
Then, we show that the two laminations are such that there are robust hetero- clinc tangencies. Then we get robust homoclinic tangencies. By a classical Baire category argument, this gives a subset of the set of automorphisms of limited degree 5 in which automorphisms with infinitely many sinks are dense.
In section 2, we precise the automorphism. We give first a complex polyno- mial p, then a 2-dimensional Henon automorphism h using p and finally the 3-dimensional automorphism f
1by adding to h a third linear component.
In section 3 and 4, we describe the basic set : we prove propositions which precise the transversal shape of the unstable foliation.
In section 5, then we perturb f
1to get a heteroclinic tangency with good ori- entation of the two foliations. This is done by increasing the degree of p and adding a small term in the first component of f
1.
In section 6, we prove that these heteroclinic tangencies are robust and we con- clude the proof of the main result in section 7.
For the convenience of the reader we recall how to create sinks from homoclinic
tangencies in the sectionally dissipative case.
Acknowledgments : This research was partially supported by the ANR project LAMBDA, ANR-13-BS01-0002.
2 Preliminaries
2.1 One complex dimension
We denote by D ⊂ C the unit disk.
Lemma 2.1.1. For r =
101and r
0=
1001, there exists a polynomial p of degree 4 such that :
(i) p
−1( D ) ⊂⊂ D and p
−1( D ) admits 3 components D
1, D
2, D
3(ii) For D
1and D
2we have : D
1⊂ B(1 − r, r
0), D
2⊂ B(r − 1, r
0), p
|Diis a biholomorphism and |p
0| > 100 on D
ifor i ∈ {1, 2}
(iii) p admits a fixed point α
p∈ D
3such 1 < |p
0(α
p)| < 1.1
Proof. Let us start with a real polynomial z
2+ c where c = 0.249. We modify it by adding a term of degree 4 : p(z) = z
2+ 0.249 −
Rd2.z
4on a disk of radius R. Then, it suffices to rescale to get the result on D . We want to choose R such that the following inequalities hold :
p((1 − r + 2r
0)R) < −2R (2)
p((1 − r − 2r
0)R) > 2R (3)
that is :
((1 − r + 2r
0)R)
2−
Rd2(1 − r + 2r
0)
4)R
4+ 0.249 < −2R ((1 − r − 2r
0)R)
2−
Rd2(1 − r − 2r
0)
4)R
4+ 0.249 > 2R is equivalent to :
((1 − r + 2r
0)
2− d(1 − r + 2r
0)
4)R
2+ c < −2R ((1 − r − 2r
0)
2− d(1 − r − 2r
0)
4)R
2+ c > 2R
Once for all, we take d =
(1−r)1 2. Then, if we take R sufficiently large, the two inequalities are verified.
The critical points of p are {0, c
p= +
√R2d
, −
√R2d
}. Since p(0) = c ∈ B (0, R) and :
p( R
√
2d ) = R
22d + c − d. R
44d
2R
2= R
24d + c ' R
24d > R
by the Riemann-Hürwitz formula we see that p
−1(B(0, R)) admits 3 compo- nents, two of them are univalent, we call them D
1and D
2, and one is not, we call it D
3. In R , p
−1((−R, R)) admits 3 components I
1, I
2, I
3with I
1near −R, I
2near +R and 0 ∈ I
3(see Figure 1). Moreover, the 3 components D
1, D
2, D
3of p
−1(D(0, R)) in C are such that D
i∩ R = I
i. Indeed, p is real, the sets D
iare symmetric w.r.t. to R and simply connected by the maximum principle so the intersections D
i∩ R must be intervals. p
|Diis a biholomorphism for i ∈ {1, 2}
and we have : p
−1( D ) ⊂⊂ D .
Then, let us consider the univalent map q = p
−1|D0 1: B(0, 2R) → D
01where D
0iis the component of p
−1(D(0, 2R)) such that : D
i⊂ D
i0. Since p((1−r+2r
0)R) <
−2R and p((1 − r − 2r
0)R) > 2R, there is a point z
q∈ B(0, R) such that :
|p
0(q(z
q)| >
2R4r0=
2rR0, that is |q
0(z
q)| <
2rR0. Now, by the Koebe Theorem we know that for an univalent map g : D → C we have the following inequalities :
∀z ∈ D , 1 − |z|
(1 + |z|)
3< |g
0(z)|
|g
0(0)| < 1 + |z|
(1 − |z|)
3Figure 1 : graph of p on the real line
Applying this result to q and the ball B(0, R) ⊂ B(0, 2R) we get that
∀z ∈ B(0, R), 12.|q
0(0)| > |q
0(z)| >
274.|q
0(0)| then : ∀z ∈ B(0, R), |q
0(z)| <
144.|q
0(z
q)| <
144.2rR 0then : ∀z ∈ D
1, |p
0(z)| >
288rR 0>
r10= 100 and ∀z ∈ B(0, R), |q
0(z)| < r
0if R > 288. This inequality implies that D
1⊂ B(1 − r, r
0).
Then, |p
0| > 100 on D
1. The corresponding result holds for D
2.
The multiplier of the repelling fixed point z
2+ 0.249 is smaller than 1.1 in modulus, so increasing the value of R if necessary we get the same estimate for p and we are done.
Remark 2.1.2. Let us further introduce two sets D
001and D
002such that : D
1⊂ D
001⊂ D
01and D
2⊂ D
200⊂ D
02, D
001and D
200are 2 components of p
−1(D(0,
3R2)).
The Koebe Theorem gives us that |p
0| > 100 on D
i00.
2.2 Two complex dimensions
We now perturb the polynomial p into a complex Henon map with small Jacobian −b.
h : (z
1, z
2) 7→ (p(z
1) + bz
2, z
1)
In this subsection, we will denote by C
uthe two-dimensional cone centered at e
1of opening
101and C
ssthe two-dimensional cone centered at e
2of opening
1
10000
. In the next subsection, we will introduce analogous three dimensional cones. To simplify the notations, we will denote them too by C
uand C
ss. Definition 2.2.1. Given an automorphism F : C
2→ C
2, we say that T
∞−∞
F
n( D
2) is a horseshoe for F if :
- F ( D
2) ∩ D
2is an union of two bidisks D
1and D
2such that D
i∩ ∂( D
2) is included in ∂( D ) × D for i = 1, 2
- on F
−1( D
2) ∩ F ( D
2), the cone field C
uis F-invariant and the cone field C
ssis F
−1-invariant
- there exists C
F> 1 such that on F
−1( D
2) ∩ F( D
2) :
∀u ∈ C
u, ||DF (u)|| > C
F||u|| and ∀v ∈ C
ss, ||D(F
−1)v|| > C
F||v||
In the following, we will have to take the constant b such that 0 < |b| < b
iwhere b
iis a bound which will be reduced a finite number of times so that it will ensure some properties on f
1. The following result is classical. For instance, it follows from the work of Hubbard-Oberste-Vorth, see [14] or [11], we just give a justification for the constants of the cones.
Proposition 2.2.2. There exists a positive number b
0such that if |b| < b
0, then T
−∞<n<+∞
h
n( D
2) is a horseshoe for h
|D2. Besides, the fixed point α
hcontinuation of the fixed point α
pof p is a saddle point of expanding eigenvalue 1 < |ξ
h| < 1.1.
Proof. We have that |p
0| > 100 on D
100∪ D
200. This implies that there exists a positive number b
0such that if |b| < b
0, then C
uis h-invariant and C
ssis h
−1-invariant with C
h= 50.
Definition 2.2.3. We denote h[1] by the restriction of h to D
100× D and by h[2]
the restriction of h to D
200× D (where D
001and D
200were defined in the precedent Remark). Then we inductively define h[I] for an arbitrary sequence of digits by h[Ij] = h[j] ◦ h[I] on (h[I])
−1(D
00j× D ). We define : H
1= (h[1])(D
001× D ) ∩ D
2, H
2= (h[2])(D
200× D ) ∩ D
2and H
Ij= (h[j])(U
I) ∩ D
2.
The following Proposition is a consequence of the general study on complex horseshoes.
Proposition 2.2.4.
(i) For each finite sequence of indices I = (i
1, ..., i
p), i
j∈ {1, 2}, the set H
Iis of the form : H
I= S
z1∈D
{z
1} × I
Iz1where I
Iz1is an open topological disk included in D .
(ii) If I is an infinite sequence I ∈ {1, 2}
N, the intersection H
I= T
p≥0
H
I≤pis a piece of the intersection of the unstable manifold of one point of the set T
∞−∞
h
n( D × D ) with D
2× Q (where I ≤ p denotes the finite subsequence of the
p first digits of I).
Let us call t
z1,I= δ
1(I
Iz1) = δ
2({z
1} × I
Iz1). We call t
n= max
z1,|I|=nt
z1,I. We first prove that the rate of decay of t
ncan be made arbitrary small.
Lemma 2.2.5. For all k ∈]0, 1[, there exists 0 < b
1< b
0such that for all
|b| < b
1, n ∈ N we have : t
n+1< kt
n. Consequently : P
+∞n=0
t
n<
1−k1Proof. Let us prove the lemma by induction. The property is obvious for I = ∅.
Let us suppose the lemma is proven for all the I until a certain rank. We bound here t
z1,I∪{ip+1}. The intersection of the set H
I= h[I]( D
2) with p(z
1) + bz
2= C
stis a curve Γ
Isuch that δ
1(Γ
I) < k.t
nif we take 0 < b
1< b
0sufficiently small. Mapping by h : (z
1, z
2) 7→ (p(z
1) + b.z
2, z
1), we get : t
n+1< k.t
n.
2.3 Three complex dimensions
We consider now the 3 dimensional map f
1introduced in (1) : f
1: (z
1, z
2, z
3) 7→ (p(z
1) + b.z
2, z
1, λ.z
1+ µ.z
3)
In the following, we will see that the first direction is expanded by f
1and corresponds to the direction of the unstable manifolds of a hyperbolic set we are going to describe. The second and third directions are contracted by f
1and correspond to the direction of the stable manifolds of this hyperbolic set. We fix 3 small angular cones C
u, C
ss, C
cscentered in the three axis of coordinates with thin opening.
Notation. C
uis the cone centered at e
1of opening
101, C
ssis the cone centered at e
2of opening
100001and C
csis the cone centered at e
3of opening
100001. Definition 2.3.1. Given an automorphism F : C
3→ C
3, we say that T
∞−∞
F
n( D
2× Q) is a horseshoe for F if :
- F ( D
2× Q) ∩ ( D
2× Q) is an union of two tridisks D
1and D
2such that D
i∩ ∂( D
2× Q) is included in (∂( D ) × D × Q) for i = 1, 2
- on F
−1( D
2× Q) ∩ F( D
2× Q), the cone field C
uis F-invariant and the cone fields C
ss, C
csare F
−1-invariant
- there exists C
F> 1 such that on F
−1( D
2× Q) ∩ F ( D
2× Q) :
∀u ∈ C
u, ||DF (u)|| > C
F||u|| and ∀v ∈ C
ss∪ C
cs, ||D(F
−1)v|| > C
F||v||
Definition 2.3.2. We say that a saddle periodic point q of multipliers |λ
1| ≤
|λ
2| < 1 < |λ
3| is sectionally dissipative if the product of any two of its eigen- values is less than 1 in modulus, that is, |λ
1λ
3| < 1 and |λ
2λ
3| < 1 .
Once for all, we fix now : µ = 9
10 i, λ = 1 10
√ 2
2 e
iπ4= 1
10 (1 + i) (4)
We denote by pr
ithe projection on the i
thcoordinate in C
3.
Proposition 2.3.3. Let f
1: (z
1, z
2, z
3) → (p(z
1) + b.z
2, z
1, λ.z
1+ µ.z
3) with
|b| < b
1. Then, T
−∞<n<+∞
f
1n( D
2× Q) is a horseshoe. Moreover, f
1has a
fixed point α
f1that is sectionally dissipative.
Proof. The fact that C
uis f
1-invariant and C
ssis f
1−1-invariant comes from the analogous result on h. The fact that C
csis f
1−1-invariant comes from the fact that e
3is an eigenvector at each point for Df
1−1of associated eigenvalue
1
µ
=
109i. The existence of α
f1is obvious. Looking at the differential, we see that α
f1has three eigenvalues ξ
h, µ,
−bξh
so the result follows since 1 < |ξ
h| < 1.1,
|µ| =
109and
109.
1110< 1.
Definition 2.3.4. We denote f
1[1] the restriction of f
1on D
001× D ×Q and f
1[2]
the restriction of f
1on D
200× D × Q (where D
100and D
200were defined in Remark 2.1.2). Then, we inductively define f
1[I] for an arbitrary sequence of digits by f
1[Ij] = f
1[j] ◦ f
1[I] on (f
1[I])
−1(D
00j× D × Q). We define U
1= (f
1[1])(D
100× D × Q) ∩ D
3, U
2= (f
1[2])(D
002× D × Q) ∩ D
3and U
Ij= (f
1[j])(U
I) ∩ ( D
2× Q).
In the following, we denote by D
.,z2,z3the disk parallel to the z
1axis. In this subsection we gather a few properties of complex horseshoes which will be useful afterwards. In particular, horseshoes are compact, hyperbolic, transitive and locally invariant sets.
Lemma 2.3.5. For |b| < b
1and for every sequence of digits I, f
1[I]( D
.,z2,z3) is a graph over the first coordinate z
1of the form
f
1[I]( D
.,z2,z3) = {(z
1, y
I2(z
1), y
3I(z
1)) : z
1∈ D } with derivatives |(y
2I)
0| <
501, |(y
3I)
0| <
501.
Proof. We show the result by induction. We suppose that the conditions are ver- ified for each step until I and we consider the following index I
0= I∪{j}. In C
I: z
2= y
I2(z
1) with the derivative of y
I2in modulus bounded by
15: |(y
2I)
0(z
1)| <
501. Every curve {(z
1, y
2I(z
1), y
I3(z
1)) : z
1∈ D } intersects, if |b| < b
1, each hyper- surface p(z
1) + bz
2= C
stonly one and one time. Then the curve f
1[I
0]( D
.,z2,z3) is a graph {(z
1, y
Ij2(z
1), y
Ij3(z
1)) : z
1∈ D }. The implicit function Theorem says that the derivate of this graph is
p0(z1)+b(y1 2I)0(z1)inside this domain. Then, {z
1: p(z
1) + by
2I(z
1) ∈ D } = D
1∪ D
2with D
i⊂ D
00i, we have the bound
|p
0| ≥ 100. Moreover, |(y
I2)
0| <
501and b < 1, so this gives : |(y
Ij2)
0| <
501So, f
1[I]( D
.,z2,z3) is a graph {(z
1, y
I2(z
1), y
3I(z
1)) : z
1∈ D }. It remains to prove the bound on the derivative of y
3I(z
1). We do this by induction. For |I| = ∅, this is obvious. Then for I
0= I ∪ {j}, we see that f
1[I
0]( D
.,z2,z3) is the image of a piece of f
1[I]( D
.,z2,z3) so we can write :
y
I30(z
1) = µ.y
I3(ζ(z
1)) + λ.ζ (z
1) (y
3I0)
0(z
1) = µ.(y
I3)
0(ζ(z
1)).ζ
0(z
1)) + λ.ζ
0(z
1)
where ζ(z
1) denotes the first coordinate of the inverse image of the point of C
I0whose first coordinate is z
1. The derivative of ζ is bounded by
701because
|p
0| > 100 on D
001∪ D
002and for small values of b, ζ
0is near the derivate of the
local inverse branch of p. We infer that |(y
I3)
0(z
1)| ≤ B
nwhere B
nsatisfies
B
0= 0 and B
n+1<
501|µ|B
n+
701where n = |I|. It follows that for every n ≥ 0,
B
n<
501and we are done.
We finish this section by a lemma saying that horizontal graphs are not very perturbated by adding a term of degree 5, p
5z
5, to p.
Lemma 2.3.6. For every sequence of digits I, f [I]( D
.,z2,z3) = {(z
1, y
2f,I(z
1), y
3f,I(z
1)) : z
1∈ D } is a horizontal graph such that for all z
1∈ D , p
57→ [(y
f,I2)(z
1)](p
5) and p
57→ [(y
3f,I)(z
1)](p
5) are 2-lipschitz continuous.
Proof. The lemma is a consequence of the Schwarz-Pick Lemma applied to the two holomorphic maps D 7→ D , p
57→ [(y
1f,I)(z
1)](p
5) and D 7→ D , p
57→
[(y
f,I2)(z
1)](p
5).
A consequence of this lemma is that the horizontal graphs T
p≥0
U
I≤p= W ∩ ( D
2× Q) = {(z
1, y
fW(z
1), y
Wf,2(z
1)) : z
1∈ D } verify the same 2-Lipschitz continuous property.
3 Structure of the basic set : unperturbed case
3.1 Generalities and main result in the unperturbed case
In this section, we describe the geometry of a horseshoe in C
3which is induced by f
1. We describe a subset of the set K
f1= T
∞−∞
f
1n( D × D × Q), more precisely T
∞−∞
f
1n((D
1∪ D
2)× D ×Q). Recall that f
1(z
1, z
2, z
3) = (p(z
1) + b.z
2, z
1, λ.z
1+ µ.z
3). The choice of Q instead of D in the last coordinate is a matter of convenience only, since the projection on the 3
rdcoordinate is easier to analyse in terms of subsquares (see Figure 3).
Definition-Proposition 3.1.1. Given a set E ⊂ C
kand i ∈ {1, 2, 3}, we call δ
i(E) = diam(pr
i(E)). Given z
1, z
2∈ C , we let : L
z1,z2= {(z
1, z
2)} × C . Definition 3.1.2. Given a set E ⊂ C
2and δ ∈ (0, 1), we say that E is of δ-product type w.r.t the 2
ndcoordinate if there is a square S = c
S+ `
S.Q ⊂ C such that
pr
1(E) × (c
S+ (1 − δ).`
S.Q) ⊂ E ⊂ pr
1(E) × (c
S+ (1 + δ).`
S.Q) We call admissible square for E a square S with this property.
Proposition 3.1.3.
(i) For each finite sequence of indices I = (i
1, ..., i
p), i
j∈ {1, 2}, the set U
Iis a fibration by squares over the set H
I(see Figures 2,3) :
U
I= [
(z1,z2)∈HI
(z
1, z
2) × Q
z1,z2,IMore precisely, U
I= S
z1∈D
{z
1} × R
z1,Iwhere R
z1,Iis a non empty open con- nected subset in D × Q of the form :
R
z1,I= [
z2∈pr1(Rz1,I)
{z
2} × {β
z1,z2,I+ l
I.Q} (5)
where Q
z1,z2,I= {β
z1,z2,I+ l
I.Q} is a square whose length side l
Iis independent
of z
1and depends only of the length |I| of I : l
I= |µ|
|I|.
(ii) If I is an infinite sequence I ∈ {1, 2}
N, the intersection U
I= T
p≥0
U
I≤pis a piece of the intersection of the unstable manifold of one point of the set T
∞−∞
f
1n( D × D × Q) with D
2× Q (where I ≤ p denotes the finite subsequence of the p first digits of I).
The result is essentially a consequence of Proposition 2.2.4 and Lemma 3.1.5 below.
Let us first note the following obvious consequence of the definition of f
1. Lemma 3.1.4. For all z
1, z
2, f
1(L
z1,z2) = L
h(z1,z2).
The following lemma describes the geometry of non empty intersections of the form L
z1,z2∩ ({z
1} × R
z1,I) :
Lemma 3.1.5. For every z
1, z
2, if L
z1,z2intersects {z
1}×R
z1,I, then pr
3(L
z1,z2∩ ({z
1} × R
z1,I)) is a square inside the line L
z1,z2. We denote `
z1,z2,Ithe length side of the square L
z1,z2∩ U
I= L
z1,z2∩ ({z
1} × R
z1,I) when this intersection is non empty. Then, for all z
1, I, `
z1,z2,Iis constant in z
1, z
2, we denote it by l
Iand its value is l
I= l
z1,z2,I= |µ|
|I|.
Proof. The third coordinate of f , for a fixed value of z
1, is affine in z
3of mul- tiplier µ =
109i. This implies : ∀I, l
I∪{ip+1}=
109l
Iand the result follows.
In the following, the projections of the sets R
z1,Iwill have a special config- uration. This is why we introduce the next definition.
Definition 3.1.6. Let Q the unit square centered at the origin. For any square S whose axes are parallel to those of Q, we denote by N E, SE, SW, N W its four corners. Given such a square S of length size l
Swe denote by S
c,s,s0where c ∈ {N E, SE, SW, N W }, s ∈ [
12, 1[, s
0∈]0, 1 − s[ the subsquare of side length s.l
Spositioned near the corner c at a distance s
0l
Sof the two sides.
A configuration NE-SW is a triple (S, S
c,s,s0, S
c0,s,s0) where S
c,s,s0and S
c0,s,s0are two subsquares of S such that {c, c
0} = {N E, SW }, a configuration NW-SE is a triple (S, S
c,s,s0, S
c0,s,s0) such that S
c,s,s0and S
c0,s,s0are two subsquares of S and {c, c
0} = {N W, SE} with s
0< max(
501, 1 − s).
Proposition 3.1.7. There is a constant 0 < b
2< b
1such that for all |b| < b
2, we get the following properties :
1. For every z
1, I , R
z1,Iis of
10001-product type.
2. For each admissible square S
z1,Ifor R
z1,I, its length side satisfies
`
z1,I∈ 899
1000 l
I, 901 1000 l
I!
3. Let S
z1,I, S
z1,I∪{1}and S
z1,I∪{2}be admissible squares for R
z1,I, R
z1,I∪{1}and R
z1,I∪{2}respectively. There are c, c
0∈ {N E, SE, SW, N W}, e, e
0∈ (
1000899,
1000901) and 0 < s
01, s
02<
1001such that S
z1,I∪{1}= (S
z1,I)
c,e,s01
and S
z1,I∪{2}= (S
z1,I)
c0,e0,s02
and (S
z1,I, S
z1,I∪{1},S
z1,I∪{2}) forms a configuration NE-SW if |I| is even and
a configuration NW-SE if |I| is odd.
4. The β
z1,z2,Idepend holomorphically on z
1, z
2and |
∂βz∂z1,z2,I1
| <
501.
Figure 2 : U , U
1and U
2Figure 3 : R
z1,Iand its two subsets in projection in the complex case
This proposition simply means that pr
3({z
1}×R
z1,I) contains a square which
contains two subsquares included in, pr
3({z
1}×R
z1,I∩{1}), pr
3({z
1}×R
z1,I∩{2}),
whose lengths are at least
1000899times the length of the larger one, and positioned
in opposite corners.
3.2 Proof of Proposition 3.1.7
Proof. First, let us remark that item 4 is a direct consequence of Lemma 2.3.5.
Let us prove items 1 and 2. It is a consequence of the following lemma.
Lemma 3.2.1. For all I and z
1, there exists a constant r
Iwhich only depends on |I| such that if |b| < b
1, for all z
1and I, R
z1,Iis of r
I-product type and for all admissible squares S
z1,I= c
z1,I+ `
z1,I.Q ∈ C for R
z1,Iwe have :
pr
1(R
z1,I)×(c
z1,I+(1−r
I).`
z1,I.Q) ⊂ R
z1,I⊂ pr
1(R
z1,I)×(c
z1,I+(1+r
I).`
z1,I.Q) Besides, we have :
∀z
1∈ D , ∀I, r
I≤ 1 1000
Proof. We have that : t
z1,I= δ
2(R
z1,I). We show Lemma 3.2.1 step by step, the result is the consequence of the convergence of an infinite series. The property is clear for the set U since U = S
z1∈D
({z
1} × D × Q). Let us suppose it is true for a finite sequence I and consider a value z
1∈ D , then z
2∈ I
Iz1and two indices I, I
0= I ∪ {i
p+1} such that p(z
1) + bz
2∈ D . We study the set R
Z1,I0with the value Z
1= p(z
1) + bz
2. We suppose inductively that for all z
1such that f
1−1(R
Z1,I0) ∩ R
z1,Iis not empty, R
z1,Iis of r
I-product type and for all admissible square S
z1,I= c
z1,I+ `
z1,I.Q ⊂ C for R
z1,Iwe have : pr
1(R
z1,I) × (c
z1,I+ (1 − r
I).`
z1,I.Q) ⊂ R
z1,I⊂ pr
1(R
z1,I) × (c
z1,I+ (1 + r
I).`
z1,I.Q) for a constant r
I. We denote : R = f
1−1(R
Z1,I0) and δ
R,Iz1= δ
1(R). Let us show the following lemma.
Lemma 3.2.2. If R
z1,Iis of r
I-product type, then R
Z1,I0is of r
I0-product type with : r
I0= r
I+ (
501+
2|µ||λ|)δ
R,Iz1Proof. If |b| < b
1, by Property 4, we get that there is a square S
Z1,I0of length
|µ|.(1 − r
I−
501.δ
R,Iz1).`
z1,Isuch that : pr
1(R
Z1,I0) ×S
Z1,I0⊂ R
Z1,I. We consider now the biggest complex disk S
0included in R
Z1,I0\ (pr
1(R
Z1,I0) × S
Z1,I0) and
∆ the line which contains the preimage of this complex disk by f
1. This line is a line parallel to the z
3axis, we bound here the diameter δ
3(S) of the preimage complex disk S = f
1−1(S
0). Let us call :
M = {(z
1, z
2, z
3) ∈ R, pr
2(M ∩ {z
3= C
st}) = pr
2(R)}
Then, δ
3(S ∩ (R/M)) < (r
I+
501.δ
R,Iz1).`
z1,I(this is still due to Property 4) so δ
3(f
1(S ∩ (R/M))) ≤ |µ|.(r
I+
501.δ
zR,I1).`
z1,I. Besides, δ
1(M ) ≤ δ
zR,I1so δ
3(f
1(S ∩ M )) ≤ |λ|.δ
zR,I1.`
z1,I. Then the length of S
0which is the length of the image of S by f
1is bounded by : (|µ|.(r
I+
501.δ
zR,I1) + |λ|.δ
zR,I1).`
z1,I. Then, there is a square S
Z+1,I0
of length (|µ|.(1 + r
I+
501.δ
zR,I1) + |λ|.δ
zR,I1).`
z1,Isuch that : R
Z1,I⊂ pr
1(R
Z1,I0) × S
+Z1,I0
. It implies that R
Z1,I0is of r
I0-product type with : pr
1(R
Z1,I0) × (c
Z1,I0+ (1 − r
I0).`
Z1,I0.Q) ⊂ R
z1,I0⊂ pr
1(R
Z1,I0) × (c
Z1,I0+ (1 + r
I0).`
Z1,I0.Q) with `
Z1,I0=
109`
z1,Iand : 2r
I0= 2r
I+
502δ
R,Iz1+
|λ||µ|δ
zR,I1so : r
I0= r
I+ (
501+
2|µ||λ|)δ
zR,I1.
The same arguments as in the proof of Lemma 2.2.5 imply that for every
k ∈]0, 1[, there exists 0 < b
2< b
1such that for all |b| < b
2, I, i
p+1, we have
the following : δ
R,I∪{iz1p+1}
< kδ
zR,I1. It thus follows from Lemma 3.2.2 that r
Iconverges as |I| → +∞ and
r
I0≤ ( 1 50 + |λ|
2|µ| )( X
n≥0
k
n)δ
0so this can be made smaller than
10001if δ
0is small enough.
Property 1 of Proposition 3.1.7 is a direct consequence of the previous lemma.
Property 2 of Proposition 3.1.7 is a consequence of Property 1 and Lemma 3.1.5.
Let us now prove Property 3.
Lemma 3.2.3. For every z
1, if a complex line ∆ parallel to the third axis of coordinates intersects both {z
1} × R
z1,Iand {z
1} × R
z1,I∪{ip+1}, then the sets S = pr
3(∆ ∩ ({z
1} × R
z1,I)) and S
0= pr
3(∆ ∩ ({z
1} × R
z1,I∪{ip+1})) are squares such that S
0= S
c,910,s0
for c ∈ {N E, SE, SW, N W } and s
0≤
1001. Proof. Let us first study the case where |I| = 1.
Lemma 3.2.4. If a complex line ∆ parallel to the third axis of coordinates intersects U
1, then pr
3(U
1∩ ∆) is the square Q
N E,910,s
, and if ∆ intersects U
2, then pr
3(U
2∩ ∆) is the square Q
SW,910,s
with s <
1001.
Proof. This follows from the choice of constants in Section 2. Indeed, the term λz
1=
101√2
2
e
iπ4z
1in the expression of pr
3◦ f
1is just intended to push the two subsquares in the direction of each of the two NE and SW corners, the two subsquare are of length of side multiplied by the modulus of µ this is
109.
The demonstration of Lemma 3.2.3 is made by induction on |I|. The in- duction step is simply the fact that the image by z
37→ λ.z
1+ µ.z
3of a NW (resp. SW,SE,NE) configuration, this is a subsquare positioned in the NW (resp.
SW,SE,NE) corner of a greater square, is a SW (resp. SE,NE,NW) configuration with length sides multiplied by
109.
Now, for fixed z
1and I, it is enough to take ∆ and ∆
0such that ∆ intersects R
z1,I∪{1}(resp. ∆
0intersects R
z1,I∪{2}). We then get two squares which are respectively the square intersection of ∆ and R
z1,I∪{1}in ∆, and the square in- tersection of ∆
0and R
z1,I∪{2}in ∆
0. These two squares are admissible squares for R
z1,I∪{1}and R
z1,I∪{2}. Lemma 3.2.3 and Property 1 give us that every admissible square for R
z1,Iwith these two subsquares form the configuration given by Item 3.
The third Property has been shown, the proof of Proposition 3.1.7 is now com-
plete.
4 Structure of the basic set : perturbed case
4.1 Generalities and main result in the perturbed case
In this section, we analyze the structure of unstable manifolds of the horse- shoe induced by an arbitrary perturbation of f
1. Some arguments in Section 3 used the special form of f
1, so we need to adapt the arguments. We can take perturbations of f
1, not necessarily polynomial, in the space of automorphisms of C
3.
Notation. So far we have considered an automorphism f
1= f
1(p, b) defined in (1). In the following, we will reduce constant b a finite number of times to ensure a certain number of properties. Furthermore, in Section 4, we will introduce decreasing neighborhoods ν
i(f
1) for i ∈ {0, 1, 2, 3}. In Section 5, we perturb f
1to f
2= f
2(p, b, σ) with parameters p = p +p
5z
5and σ such that f
2∈ ν
3(f
1) and ν
4(f
2) will denote a non empty neighborhood of f
2with ν
4(f
2) ⊂ ν
3(f
1). We will subsequently make a number of further reductions until the final result.
Because of the structural stability of horseshoes, it is well known that Defi- nition 2.3.4 can be applied in a neighborhood ν
0(f
1) of f
1= f
1(p, b) in the space of automorphisms of C
3. For such a map f ∈ ν
0(f
1), we can define f [1], f [2]
and a sequence of sets U
I= f [I]( D
2× Q) as before. We have the following properties for the sets U
I:
Proposition 4.1.1. For every |b| < b
2, for every f ∈ ν
0(f
1), for every finite sequence of indices I = (i
1, ..., i
p), i
j∈ {1, 2}, the set U
Isatisfies the following properties :
1. U
I= S
z1∈D
{z
1} × R
z1,Iwhere R
z1,Iis an open connected set with R
z1,I⊂⊂
D × Q.
2. If I is an infinite sequence I ∈ {1, 2}
N, the intersection T
p≥0
U
I≤pis a component of the intersection of the unstable manifold of some point in the set T
n∈Z
f
n( D
2× Q) with D
2× Q.
Definition 4.1.2. A u-curve C is a curve whose tangent vectors are all in C
u(recall the cones C
u, C
ss, C
cswere defined in Subsection 2.3).
A s-plane is a plane which admits a basis of two vectors, one of them belonging to C
ssand the other one to C
cs.
A s-surface is a hypersurface whose tangent planes are all s-planes.
We have the counterpart of Lemma 2.3.5 in this context.
Lemma 4.1.3. For |b| < b
2and f ∈ ν
0(f
1), for every sequence of digits I, f [I]( D
.,z2,z3) is a u-curve.
Proof. Lemma 2.3.5 defines a cone field which is invariant by f which thus persists under small perturbations.
The previous Lemma 4.1.3 implies in particular that every piece of unstable manifold T
p≥0
U
I≤pis a u-curve. The following lemma deals with the horseshoe
for f ∈ ν
0(f
1).
Lemma 4.1.4. For each I, the intersection of U
Iwith a s-surface S is home- omorphic to D × Q. Moreover, there is a canonical homeomorphism Ψ
Sgiven by the map S ∩ U
I7→ D × Q, z 7→ ((pr
2, pr
3) ◦ (f [I])
−1)(z).
Proof. The property can be proven by induction. For U
∅= D
2×Q, the property is obvious. Let us suppose it is true until the rank I. Let S be a s-surface intersecting U
I0= U
I∪{i0}with i
0∈ {1, 2}. Then (f [i
0])
−1(S∩( D
2×Q))∩( D
2×Q) is a s-surface too. Then, its intersection with U
Iis by the induction hypothesis homeomorphic to D × Q by the map z 7→ ((pr
2, pr
3) ◦ (f [I])
−1)(z). Now, taking its image by f , we have that : S ∩ U
I0is homeomorphic to D × Q by the map z 7→ ((pr
2, pr
3) ◦ (f [I
0])
−1)(z).
In particular, we have :
Lemma 4.1.5. For all z
1, I, R
z1,Iis homeomorphic to D × Q.
Definition 4.1.6. Given a set E ⊂ C and δ ∈ (0, 1), we say that E is δ-square shaped if there is a square S = c
S+ `
S.Q ⊂ C (`
Scan be complex non real) such that
(c
S+ (1 − δ).`
S.Q) ⊂ E ⊂ (c
S+ (1 + δ).`
S.Q) We call admissible square for E a square S with this property.
Definition 4.1.7. Let S, S
11, S
12, S
21, S
22be squares in C . We say that S is well subdivided by S
11, S
12, S
21, S
22if there exist squares S
1, S
2such that S
1, S
2⊂ S, S
11, S
12⊂ S
1and S
21, S
22⊂ S
2and a linear isomorphism ψ of C such that (ψ(S), ψ(S
1), ψ(S
2)) is a configuration NE-SW, and (ψ(S
1), ψ(S
11), ψ(S
12)), (ψ(S
2), ψ(S
21), ψ(S
22)) are configurations NW-SE (see Definition 3.1.6).
Let δ ∈ (0,
1001). Given five δ-square shaped sets (E
i)
1≤i≤5, we say that E
1is well subdivided by the other sets if there are 5 admissible squares (S
i)
1≤i≤5respectively for E
isuch that S
1is well subdivided by the other squares.
The following Proposition is an analogous of Proposition 3.1.7.
Proposition 4.1.8. For every |b| < b
2, there exists a neighborhood ν
3(f
1) ⊂ ν
0(f
1) of f
1= f
1(p, b) such that for every f ∈ ν
3(f
1), we have :
1. For every z
1, I , pr
3({z
1} × R
z1,I) is
5001-square shaped (recall that the sets R
z1,Iwere defined in Proposition 4.1.1).
2. If S
z1,Iand S
z1,I∪{ip+1}are admissible squares for R
z1,Iand R
z1,I∪{ip+1}respectively, their length sides satisfy :
`
z1,I∪{ip+1}∈ 895
1000 `
z1,I, 905 1000 `
z1,I!
3. Let S
z1,I, S
z1,I∪{11}, S
z1,I∪{12}, S
z1,I∪{21}and S
z1,I∪{22}be admissible squares for R
z1,I, R
z1,I∪{11}, R
z1,I∪{12}, R
z1,I∪{21}and R
z1,I∪{22}respectively.
Then, S
z1,Iis well subdivided by these 4 subsquares.
So pr
3(R
z1,I) contains a square which contains two subsquares included in
pr
3(R
z1,I∩{1}), pr
3(R
z1,I∩{2}) whose lengths are the length of the great square
times >
1000895and the two squares are near two opposite corners of the great
square, with a change of direction at each step.
4.2 Proof of Proposition 4.1.8
If we take ν
1(f
1) ⊂ ν
0(f
1) sufficiently small, the properties are automatically verified for (U, U
1, U
2). Our strategy is to consider the sets U
I0as images of pieces of the sets U
Iby f which looks like locally more and more precisely to a linear function when we map the sets U
Iat a smaller scale. Then, by an infinite series argument we will get a result which looks like a bounded distorsion property.
Notation. We will call pr
w3,w2
the orthogonal projection on a vector w
3∈ C
csparallel to w
2∈ C
ss.
Definition 4.2.1. We say that a family of sets {W
I0, I
0∈ {I, I ∪{1}, I ∪{2}, I ∪ {11}, I ∪ {12}, I ∪ {21}, I ∪ {22}} is a family of u-oriented-subdivided tridisks if we have that :
- each W
I0is an union of u-curves
- W
I∪{1}, W
I∪{2}⊂ W
I, W
I∪{11}, W
I∪{12}⊂ W
I∪{1}and W
I∪{21}, W
I∪{22}⊂ W
I∪{2}In particular, for each finite sequence I, we easily get this new property for the sets U
I: (U
I, U
I∪{1}, U
I∪{2}, U
I∪{11}, U
I∪{12}, U
I∪{21}, U
I∪{22}) is a family of u-oriented-subdivided tridisks
Definition 4.2.2. Let I be a finite sequence and η
1, η
2<
101. Let {W
I0, I
0∈ {I, I ∪ {1}, I ∪ {2}, I ∪ {11}, I ∪ {12}, I ∪ {21}, I ∪ {22}} be a family of u- oriented-subdivided tridisks. We say that {W
I0, I
0∈ {I, I ∪ {1}, I ∪ {2}, I ∪ {11}, I ∪ {12}, I ∪ {21}, I ∪ {22}} is of (η
1, η
2)-type if for every s-plane P and for every w
3∈ P ~ ∩ C
cs, w
2∈ P ~ ∩ C
ss, denoting by Π
I0= pr
w3,w2