The modulus of analytic classification for the unfolding of the codimension-one flip and Hopf bifurcations

(1)

ANNALES

DE LA FACULTÉ DES SCIENCES

Mathématiques

WALDOARRIAGADA-SILVA, CHRISTIANEROUSSEAU

The modulus of analytic classification for the unfolding of the codimension-one flip and Hopf bifurcations

Tome XX, n^o3 (2011), p. 541-580.

<http://afst.cedram.org/item?id=AFST_2011_6_20_3_541_0>

L’accès aux articles de la revue « Annales de la faculté des sciences de Toulouse Mathématiques » (http://afst.cedram.org/), implique l’accord avec les conditions générales d’utilisation (http://afst.cedram.

org/legal/). Toute reproduction en tout ou partie cet article sous quelque forme que ce soit pour tout usage autre que l’utilisation à fin strictement personnelle du copiste est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.

cedram

Article mis en ligne dans le cadre du

Centre de diffusion des revues académiques de mathématiques

(2)

pp. 541–580

The modulus of analytic classiﬁcation for the unfolding of the codimension-one ﬂip

and Hopf bifurcations

Waldo Arriagada-Silva⁽¹⁾, Christiane Rousseau⁽²⁾

ABSTRACT.— In this paper we study equivalence classes of generic 1- parameter germs of real analytic familiesQε unfolding codimension 1 germs of diffeomorphismsQ0: (R^,⁰⁾^→⁽R^,0) with a fixed point at the origin and multiplier−1,under (weak) analytic conjugacy. These germs are generic unfoldings of the flip bifurcation. Two such germs are analytically conjugate if and only if their second iterates,Pε = Q^◦ε², are analytically conjugate. We give a complete modulus of analytic classification: this modulus is an unfolding of the Ecalle modulus of the resonant germQ0 with special symmetry properties reflecting the real character of the germQε.As an application, this provides a complete modulus of analytic classification under weak orbital equivalence for a germ of family of planar vector fields unfolding a weak focus of order 1 (i.e. undergoing a generic Hopf bifurcation of codimension 1) through the modulus of analytic classification of the germ of familyPε =Q^◦ε²,wherePε is the Poincaré monodromy of the family of vector fields.

R^ÉSUMÉ.— Dans cet article, nous étudions la classification sous conju- gaison analytique (faible) des germes de familles analytiques génériques

`

a un 1 paramètre,Qε,déployant des germes de difféomorphismes Q0 : (R^,⁰⁾ ^→⁽R^,0) de codimension 1,ayant un point fixe à l’origine et de multiplicateur−1.Ces germes sont des déploiements génériques de la bifurcation de doublement de période. Deux germes sont analytiquement

(∗) Re¸cu le 13/10/2010, accept´e le 28/03/2011 This work was partially supported by NSERC.

(1) Department of Mathematics and Statistics, University of Calgary, 2500 University Drive NW, AB T2N 1N4, Canada; and Instituto de matem´aticas, Universidad Austral de Chile, Casilla 567 - Valdivia, Chile.

warrigad@ucalgary.ca

(2) Département de Mathématiques et de Statistique, Université de Montréal C.P.

6128, Succ. Centre-ville Montr´eal, Qu´ebec H3C 3J7, Canada.

rousseac@dms.umontreal.ca

(3)

conjugués si et seulement si leurs itérés d’ordre 2,Pε =Q^◦ε²,sont analytiquement conjugués. On donne un module complet de classification analytique: ce module est un déploiement du module d’ Écalle du germe résonantQ0avec des propriétés de symétrie reflétant le caractère réel du germeQε.Ceci donne, comme application, un module complet de classification analytique sous équivalence orbitale faible pour un germe de famille de champs de vecteurs du plan ayant une bifurcation de Hopf générique de codimension 1 par le biais du module de classification analytique du germe de famillePε=Q^◦2ε ,oùPε est l’application de premier retour de Poincaré de la famille de champs de vecteurs.

This paper is part of a program to study the analytic classification of generic unfoldings of the simplest singularities of analytic dynamical systems. These dynamical systems can be germs of diffeomorphisms, in which case we study classification underconjugacy, orgerms of vectorfields, in which case we can study eitherclassification underorbital equivalence or underconjugacy. The analytic classification of unfoldings of singularities follows the analytic classification of the singularities themselves. The moduli of classification for the simplest 1-resonant singularities have been given by Ecalle, Voronin and Martinet-Ramis (cf. [4], [19] and [12],[13]). Except for the case of the node of a planarvectorfield, the moduli space is a huge func- tional space, while the formal invariants are in finite number. This means that there is an infinite number of analytic obstructions for the analytic equivalence of two germs, that cannot be seen at the formal level.

These obstructions can be understood when ﬁrst, one extends the under- lying space from Rⁿ to Cⁿ, n= 1,2 and then, one unfolds the singularity.

Indeed, in the simplest 1-resonance cases, the singularity of the dynamical system comes from the coallescence in a generic unfolding of the dynamical system of a ﬁnite number of hyperbolic singularities or special hyperbolic objects (like a periodic orbit or a limit cycle). Each hyperbolic object has its own geometric local model, and the modulus measures the limit of the mismatch of these local models. It is also a measure of the divergence of the normalizing series to formal normal form. Hence, if a singularity is non equivalent to its formal normal form, then we should expect a mismatch between the local models nearthe two hyperbolic objects in the unfolding.

This was the point of view suggested by Arnold and Martinet [11] and studied systematically by Glutsyuk [8] when the unfolding is considered only in certain conic regions of the parameter space considered in complex space.

The treatment had to be adapted by Lavaurs and followers when the bifur- cating objects were no more hyperbolic or when the domains of the local models did not intersect.

As far as codimension 1 singularities are concerned, the case of a germ of generic unfolding of a diffeormorphism with a double fixed point, also called parabolic diffeomorphism has been studied in [3] and [10], and the case of a germ of generic unfolding of a resonant diffeomorphism (one multiplier being a root of unity) has been studied in [15] and [16]. Germs of generic unfoldings of resonant saddle (resp. saddle-node) singularities of planar vector fields have been studied in [16] (resp. [17]). All these papers consider the unfolding of the corresponding complex singularity. The paper [15] also considers briefly the case of a saddle point of a real vector field.

(5)

The modulus of the unfolding is always constructed in the same way.

The formal normal form for the unfolding is identified and called themodel family.The germ of family is then compared to the formal normal form on special domains. The modulus is given by the comparison of the two normalizations over the intersection of the special domains. When one restricts to parameter values for which the special objects are hyperbolic and the normalization domains intersect, then the modulus is given by the comparison of the normalizations in the neighborhood of the special objects. This is called theGlutsyuk point of view and the corresponding modulus is called the Glutsyuk modulus. In the papers [3], [15], [16] and [17], another point of view was used, called the Lavaurs point of view. The Lavaurs point of view allows to compute the modulus for all values of the parameters. The domains on which we compare the germ of family to the formal normal form (model family) are no more neighborhoods of the special objects, but sectorial domains adjacent to two special objects. The correspondingLavaurs modulus depends in a ramified way on the parameter. In particular, there are two different definitions of the Lavaurs modulus for the parameter values forwhich the Glutsyuk modulus can be defined.

In this paper, we are concerned with the real character of a germQ^εof an analytic family of diﬀeomorphisms with a ﬂip bifurcation:

Q^ε(x) =−x(1−ε)±x³+o(x³).

We study how this is reflected in the modulus. So, we are especially inter- ested in the real values of the parameter. In particular, for nonzero values of the parameter, we are in the Glutsyuk point of view. Hence, in this paper, we have decided to make a profound analysis of the Glutsyuk modulus for the case of the unfolding of a periodic diffeomorphism and to determine how the real character of the diffeomorphism is reflected in the modulus.

For this reason, our study is restricted to the union of two sectors in the (complexiﬁed) parameterspace which do not covera full neighborhood of the origin. As an implication, we only obtain a modulus of classiﬁcation underweak orbital equivalence.

Ourpaperwas initially motivated by the study of the Hopf bifurcation. In [1], it is shown that two germs of analytic families of planar vector fields with a generic Hopf bifurcation of order 1 are orbitally analytically equivalent if and only if the germs of analytic families of their Poincaré monodromies are conjugate. The (unfolded) Poincaré monodromy is ex- actly a real diffeomorphismP^ε: (R,0)→(R,0),which is the second iterate of a Poincaré semi-monodromy Q^ε with a flip bifurcation. Hence, through this result, the present paper provides a complete modulus of classification

(6)

under weak orbital equivalence for germs of families undergoing a generic codimension-one Hopf bifurcation.

2. Preparation of the family

We consider a germ of codimension-one real analytic diffeomorphismQ⁰ with a fixed point at the origin and a multiplier equal to−1. Underscaling ofxand removing thex²-term by a normal form argument, such a germ of diffeomorphism has the form

Q⁰(x) =−x∓¹2x³+ax⁵+o(x⁵). (2.1) A generic 1-parameter unfolding is a germ of family of diﬀeomorphisms Qη(x) =Q(x, η) such that _∂x∂η^∂²^Q(0,0)= 0.

It was shown in [16] that two generic families Qη unfolding germs of the form (2.1) are conjugate if and only if their second iterate Pη = Q^◦_η² are conjugate. One direction is obvious. The other direction, namely prov- ing that if the second iterates are conjugate, then the diffeomorphisms are also conjugate requires more work. The proof in [16] was done for complex germs. In the case of real analytic germs a better proof of the other direction is given in [1], [2] for the case of families of real diffeomorphisms Qη. Indeed, given any two conjugate generic families of real analytic diffeomorphims of the formPη =Q^◦_η², it is proved that they are embeddable as Poincaré monodromies of analytic vector fields unfolding a weak focus, which are analytically orbitally equivalent (a weak focus of a real vector field is a singular point with two pure imaginary eigenvalues and which is not a centre). Considering afterwards their blow-up and the holonomies of a well-chosen separatrix in the blow-up, these holonomies are in turn conjugate (cf. [9]). But these holonomies are nothing else than the corresponding diffeomorphismsQη.So we will mainly discuss Pη.

Since the familiesP^η =Q^◦2η have real asymptotic expansion, then Q^η = C ◦ QC(η)◦ C,

P^η = C ◦ PC(η)◦ C, (2.2) where C is the standard complex conjugation x → x. In this paperwe consider real analytic families unfolding codimension-one diﬀeomorphisms of the form (2.1), and their second iterates. The following theorem is proved in [16] for a family of complex diﬀeomorphisms. Its proof can be adapted to respect the real character ofQ^η.Forthe sake of completeness we include the main steps here.

(7)

Theorem 2.1. — Given a germ of diﬀeomorphimsQ⁰ of the form (2.1) and a germ of generic unfolding Q^η, there exists a germ of real analytic change of coordinate and parameter (x, η) → (y, ε) conjugating the family P^η =Q^◦η² to the prepared form

P^ε(y) =y+y(ε±y²)(1 +b(ε) +a(ε)y²+y(ε±y²)h(y, ε)), (2.3) such thatPε(0) = exp(ε).In particular, the parameterεis called canonical.

It is an invariant. The formal invariant A(ε) is deﬁned implicitly through the expression

Pε(±√

−sε) = exp

− 2ε 1−sA(ε)ε

, (2.4)

whereA(ε)is real analytic, ands=±1is an invariant deﬁning two diﬀerent cases which are not equivalent by real conjugacy. Let Q^ε be the conjugate of Q^η in the (y, ε) coordinate and parameter. Conjugating P^ε either with y→ −y orQ^εyields other prepared forms with same canonical parameter.

Proof. — By the implicit function theorem we can suppose thatx= 0 is a fixed point forallη.By the Weierstrass-Malgrange preparation theorem, the othertwo fixed points of P^η (which are periodic points of period 2 of Q^η) are the roots of pη(x) = x²+β(η)x+γ(η), with γ(0) = 0 since the family is generic. Because it is a flip bifurcation, the periodic points of Q^η can only coincide when they are equal to x= 0. Hence, β(η)≡ 0. A reparametrization allows to take γ(η) = ±η1. Then, the map P^η has the form

P^η1(x) =x+x(η1±x²)(b1(η1) +c1(η1)x+a1(η1)x²+x(η1±x²)g(x, η1)), with b1(0)= 0. Since the fixed points±√η1 are periodic points of Q^η1 of order 2, then Pη1(√η1) = Pη1(−√η1). Hence, c1(η1) ≡0. Then Pη1(0) = 1 +η1b1(η1) withb1(0)= 0.An analytic change of parameterη1→εallows to suppose thatPε(0) = exp(ε).A corresponding scaling inx(replacing x by y = c(ε)x) allows to suppose that the fixed points of P^ε are given by y(ε±y²) = 0 and yields the prepared form. The analyticity ofA(ε),defined in Eq. (2.4), is well known and its real character is straightforward.

From now on, we will limit ourselves to prepared familiesQ^ε such that P^ε=Q^◦ε² has the form

P^ε(x) =x+x(ε+sx²)(1 +b(ε) +a(ε)x²+x(ε+sx²)h(x, ε)), (2.5) where s = ±1. We will mostly discuss the case s = +1. In particular, all the ﬁgures will be drawn only for this case. The case s =−1 is obtained

(8)

throughx→ix.This non-real change of coordinates exchanges the real and imaginary axes.

Strategy.The formal normal form of P^ε (also called “model family”) is the time-one map τ_ε¹of the vectorﬁeld

x(ε+sx²) 1 +A(ε)x²

∂

∂x, (2.6)

where A(ε) is defined in Eq. (2.4). Notice that the real axis is invariant forreal ε. In order to compute the modulus of analytic classification of the Poincaré monodromy, we compare it with τ_ε¹ for ε taken overspecific sectorial domains of the parameter space.

3. The modulus of analytic classiﬁcation

In order to solve the conjugacy problem for germs of families of analytic diffeomorphisms (2.5), a complete modulus of analytic classification must be identified, so that two germs of families of analytic diffeomorphisms are analytically conjugate if and only they have the same modulus (Theorem 7.2). We shall see that this modulus is an unfolding of the Ecalle modulus for the germ of diffeomorphism atε= 0,and so we recall the Ecalle modulus.

3.1. The Ecalle modulus of Q⁰

Two germs of analytic diffeomorphisms of the form (2.1) with same sign before the cubic coefficient are real analytically conjugate if and only if they have the same formal invariantA(ε) and the same orbit space. The Ecalle modulus is one way to describe the orbit space. To explain its construction we first remark that the diffeomorphism Q⁰ is topologically like the composition of x→ −xwith the time-1/2 map of the vectorfield (2.6), whose flow lines appear in Figure 1, while the diffeomorphismP⁰ is topologically like the time-1 map of Eq. (2.6).

So, we take a ﬁrstfundamental domain (forP⁰)C₁⁺ limited by a curve 1 and its image P⁰(1). If we identify x ∈ 1 with its image P⁰(x), the fundamental domain is conformally equivalent to a sphereS⁺1.The ends of the crescent C₁⁺ limited by 1 and P⁰(1) correspond to the points 0 and

∞ on the sphere. All orbits ofP⁰ (except that of 0) are represented by a most one point of the sphereS⁺1.However, there exist points in the neighborhood of 0 whose orbits have no representative on the sphereS⁺1.To coverthe

(9)

P0(x)

C₁⁻

s= +1 s=−1

C₂⁺

C₂⁺ C₂⁻

C₂⁻

C₁⁺

∞ ∞

∞

0 0

ψ_1,0^∞

ψ⁰_2,0 ψ_2,0^∞

ψ⁰_1,0

S⁺1 S⁻1

S⁺2

S⁻2

Figure 1. — The ﬂow of Eq. (2.6) in the casess=±1 and the Ecalle modulus ofQ0.

whole orbit space we need to take three other fundamental neighborhoods C₁⁻, C₂⁺, C₂⁻ limited by curves j and theirimages P⁰(j), j = 2,3,4, respectively. As before, we identifyx∈j with its imageP⁰(x) and the union of these fundamental domains is also conformally equivalent to a union of spheresS⁻1,S⁺2,S⁻2.But there are points in the neighborhood of 0 (resp.∞) which lie in different spheres but belong to the same orbit. So we need to identify a neighborhood of 0 (resp.∞) with a neighborhood of 0 (resp.∞) in two different spheres. This is done via a collection of analytic diffeomorphisms ψ⁰₁, ψ₂⁰ (resp.ψ₁^∞, ψ^∞₂ ) sending 0 to 0 (resp.∞ to ∞), so that we get a non-Hausdorff topological manifold endowed with a system of analytic charts given by the collection of spheres glued at the poles by the mapsψ⁰_j and ψ_j^∞. The size of the neighborhoods of 0 and ∞ depends on the size of the neighborhood of the origin where P⁰ is defined, but the germs of analytic diffeomorphims:

ψ₁⁰, ψ₂⁰ : (C,0)→(C,0) ψ₁^∞, ψ₂^∞ : (C,∞)→(C,∞)

are almost intrinsic as maps in the sphere w-coordinate. Indeed, the only analytic changes of coordinates on the spheresS^±j preserving 0 and∞are the

(10)

linear maps. If we choose diﬀerent coordinates onS^±j we get diﬀerent germs (ψ⁰_j,ψ_j^∞).The equivalence relation corresponding to changes of coordinates onS^±j and preserving 0 and∞is given by

(ψ₁⁰, ψ⁰₂, ψ₁^∞, ψ^∞₂ )∼(ψ₁⁰,ψ₂⁰,ψ^∞₁ ,ψ₂^∞)⇐⇒ ∃ C₁^±, C₂^± ∈C: ψ₁⁰(w) = (C₂⁻)⁻¹·ψ⁰₁(C₁⁺·w),

ψ₂⁰(w) = (C₁⁻)⁻¹·ψ⁰₂(C₂⁺·w),

ψ₁^∞(w) = (C₁⁻)⁻¹·ψ^∞₁ (C₁⁺·w), ψ₂^∞(w) = (C₂⁻)⁻¹·ψ^∞₂ (C₂⁺·w).

The identity P⁰ =Q^◦0² is reﬂected by the fact that it is possible to choose representatives of the modulus such that

ψ₁⁰(−w) = −ψ₂⁰(w), ψ₁^∞(−w) = −ψ₂^∞(w)

(see Lemma 6.2 fora proof in the unfolded case). We have a second equivalence relation induced by the action of Z² mapping x→ −x in prepared families (2.5)

(ψ⁰₁, ψ₂⁰, ψ₁^∞, ψ^∞₂ )≡(ψ₂⁰, ψ⁰₁, ψ₂^∞, ψ^∞₁ ).

Definition 3.1. — The Ecalle-modulus of the diffeomorphismP⁰is given by the tuple(ψ⁰₁, ψ₂⁰, ψ^∞₁ , ψ₂^∞),modulo the equivalence relations ∼and≡. Overa small neighborhoodD^rof the origin (whereD^ris the standard radius- r open disk of the complex plane), the dynamics of P⁰ is given along the flow curves of the field (2.6). All the study of the family P^ε will be done overthat fixed neighborhood U =Dr forsufficiently small values ofε.

3.2. Glutsyuk point of view in the spherical coordinate

Ifδ∈(0, π/2),we deﬁne the following sectorial domains in the parameter space, see Figure 2:

Vδ,l={ε∈C:|ε|< ρ,arg(ε)∈(π

2 +δ,3π 2 −δ)} Vδ,r={ε∈C:|ε|< ρ,arg(ε)∈(−π

2 +δ,3π 2 −δ)}

(3.7)

andρis a small real number depending onδ.We assume that 0∈Vδ,lr and denote

V_δ,lr^∗ =Vδ,lr\{0}.

The number ρ is chosen so that forvalues ε ∈ V_δ,lr^∗ , there exist orbits connecting the ﬁxed points inU. In this case, we say that we work in the Glutsyuk point of view (Figure 3). When s= +1,it is clearthat:

(11)

• Ifε∈V_δ,l^∗,the origin is an attractor and the two real singular points x_±=±√

−εare repellers inU.

• Ifε= 0,the origin is the only (non-hyperbolic) ﬁxed point.

• If ε ∈ V_δ,r^∗ , the origin is a repeller and two additional imaginary attracting singular points are created inU.

V_δ,l V_δ,r

Figure 2. — Sectorial domains for the parameter.

For ε ∈ V_δ,lr^∗ , the family of diffeomorphisms P^ε can be conjugated to the time-one mapτ_ε¹of the field (2.6) in the neighborhoods of each singular point, whose union is D^r. The modulus measures the obstruction to get a conjugacy on the full neighborhoodD^rin thex-space. The vectorfield (2.6) has singularpoints x0 = 0, with eigenvalue µ0(ε) = ε, and x_± =±√

−sε with eigenvalues:

µ_±(ε) =− 2ε

1−sA(ε)ε. (3.8)

Notice thatµ0andµ_±are analytic invariants of Eq. (2.6), which also depend analytically on ε. It follows that εand A(ε) are analytic invariants of the field (2.6). The multipliers of the time-one mapτ_ε¹of Eq. (2.6) areλj =e^µ^j, i.e.they are precisely the multipliers of the fixed points ofP^ε.Forε∈V_δ,lr^∗ , in order to compare P^ε with the model diffeomorphism τ_ε¹ we compare their orbit space. The orbit space of P^ε is obtained by taking 3 closed curves{0, +, ₋}around the fixed points, and their images{P^ε(#)}where

#∈ {0,+,−}.Since the ﬁxed points are hyperbolic, the closed regions{C#} between the curves and their images are isomorphic to three closed annuli.

We identify # ∼ P^ε(#). Then the quotient C#/ ∼will be shown to be a conformal torus. Hence, the orbit space is the union of the three annuli modulo the identification of points of the same orbit. Furthermore, if we identify0,±∼ P^ε(0,±) the quotient Hence, the orbit space turns out to be a non-Hausdorff space conformally equivalent to a quotient of the union of three toriT⁰_ε,T^∞_1,ε,T^∞_2,ε plus the three singular points (which represent the orbit space of the hyperbolic fixed points), such that:

(12)

•each orbit has at most one point in each torus,

•each orbit is either a ﬁxed point or is represented in a torus,

•some orbits may have representatives in two diﬀerent tori.

The Glutsyuk modulus consists in this identiﬁcation of orbits. For this, we need to introduce (almost) intrinsic coordinates on the tori. One way to introduce coordinates on a torus T is to considerthe latteras a quotient T=C^∗/L^C (whereL^C(x) =Cxis the linearmap) forsomeC∈C^∗.

S⁻₁

S⁺₁

S⁻₂ S⁺₂

ε= 0

ψ₂⁰ ψ₁^∞

ψ₂^∞ ψ₁⁰

ε <0 T⁰_ε

T^∞₁_,_ε ψ₂^G_,_ε ψ₁^G_,_ε

ψ₂^G_,_ε ψ₁^G_,_ε

ε >0 ψ₁^G_,_ε ψ₁^G_,_ε

ψ₂^G_,_ε ψ^G₂_,_ε T^∞₂_,_ε

T^∞₂_,_ε

T^∞₁_,_ε

T⁰_ε

Figure 3. — The orbit space of the Poincar´e monodromy.

Then a natural coordinate on T is the projection of a coordinate on C^∗ = CP¹\{0,∞}, i.e. a “spherical” coordinate. In toric coordinates, the identiﬁcation of orbits in two tori induce germs of families of analytic diffeomorphisms

ψ^G_j,ε:C^∗→C^∗

forj∈ {1,2},see Figure 3, such thatψj,ε◦L^C¹ =L^C²◦ψj,εifψj,εrepresents a map fromT1=C^∗/L^C¹ to T2=C^∗/L^C².

4. Lifting of the dynamics

Fatou coordinates were introduced in 1920 by former P. Fatou (cf. [5]).

They are changes of coordinates which allow to transform the prepared family P^ε into the model family τ_ε¹ overthe sectorial domains (3.7). We construct a special kind of Fatou coordinates: we show that it is possible to choose them respecting the real character of P^ε. This choice yields a symmetry property on the Glutsyuk invariant in the unfolding.

(13)

Although we want to compare the mapP^εwith its normal form, which is the time-one map of the vectorﬁeld (2.6), it has been shown (cf. [18]) that it is natural to change to the time coordinate of the simpler vector ﬁeld

˙

x=x(ε+sx²), which is a “small deformation” of Eq. (2.6) overDr. 4.1. The unwrapping coordinate

From now on, the parameter belongs to either of the Glutsyuk sectors (3.7). Consider the “unwrapping” change of coordinates pε : R^ε → C\{x0, x_±} deﬁned by:

x=pε(Z) =







sε sexp(−2εZ)−1

¹₂

, for ε= 0, − s

2Z ¹₂

, for ε= 0,

(4.1)

whereR^εis the 2-sheeted Riemann surface of the function (see Figure 4)







1−sexp(−2εZ) sε

¹₂

, for ε= 0, (sZ

2 )¹², for ε= 0,

and s = ±1 is the sign of the third order coeﬃcient of the family (2.5).

Notice that forallε∈V_δ,lr^∗ ,the mappεisα(ε)-periodic with α(ε) =−iπ

ε, (4.2)

that is,

pε(Z) =pε(Z−ikπ

ε ), k∈Z. (4.3)

(We will drop dependence onεin Eq. (4.2)). Thus, the imagep^◦−_ε ¹(U =D^r) consists in the Riemann surfaceR^εminus a countable numberof holes. The smallerthe radius ofU,the larger the radius of such holes (of order 1/2r²).

Notice that the distance between the centers of two consecutive holes, forε= 0,is equal to (4.2). Deﬁne the liftings:

Pε=p⁻_ε¹◦ P^ε◦pε,

Q_ε=p⁻¹_ε ◦ Q^ε◦pε. (4.4) By Eq. (4.3), the families Pε,Q_ε are deﬁned on R^ε minus the countable collection of holes. The dynamics of the lifting goes always from left to right

(14)

P^±

P⁰ P_ε

B_ε

Figure 4. — The surfaceRε,domain of the liftingPε.

onR^ε.We denoteP⁰andP^±the points at inﬁnity located in the direction orthogonal to the line of holes, in such a way that their images by pε be equal tox0= 0 and x_±=±√

−sε,respectively:

P⁰ = p^◦−_ε ¹(x0),

P^± = p^◦−_ε ¹(x_±). (4.5)

In a neighborhood of the points P^± (there are two such points, in corre- spondence with the leaves of R^ε) the two sheets go to diﬀerent singular points in the x-coordinate, while on the side of P⁰ both sheets go to the origin, see Figure 4.

Definition 4.1. — For any complex numberZ_∞∈Cwhose imaginary part is of order ∼ |α|in a neighborhood ofP^±,we deﬁne the translation in TZ_∞ :

TZ_∞(·) =Z_∞+·. (4.6) By Eq. (4.3), the sequence of equidistant holes can be denoted as:

{T_α^◦^k(Bε)}k∈Z, (4.7) where T_α⁰(Bε) =Bε corresponds to the integer k = 0. (Notice thatT_α^◦^k = Tkα foreveryk∈Z).It will be called the principal hole, and we will write:

Uε=p^◦−_ε ¹(U) =R^ε\

k∈ZT_α^◦^k(Bε) (4.8) the domain forthe dynamics ofPε,Q_ε.By connexity, the translation (4.6) can be analytically extended along the leaves ofR^εto allZ in a neighborhood of the pointP⁰,see Figure 5. The extension is notedTZ_∞ as well. We shall use specific values forZ_∞,the first one being α,defined in Eq. (4.2).

(15)

P^±

P⁰ Z₀

Figure 5. — Analytic extension ofTαto a neighborhood ofP⁰,whenε >0.

The familiesPε andQ_εcommute with Tα: Pε◦Tα=Tα◦Pε,

Q_ε◦Tα=Tα◦Q_ε (4.9)

along the leaves of R^ε. Indeed, they do so near P^±. Since Tα is globally deﬁned in R^ε by analytic continuation, they do so everywhere. Moreover, forsmallε,Pεis close toT1:

Proposition 4.2 [16]. — There existsK >0such that forZ∈p^◦−1_ε (D^r) andεsmall, one has

|Pε(Z)−Z−1|< Kr,

|P_ε(Z)−1|< Kr³. (4.10) Notice that the inversep^◦−_ε ¹of the change (4.1) is the multivalued function:

Z=p^◦−_ε ¹(x) =





 1

2εlog x² ε+sx²

, for ε= 0,

− s

2x², for ε= 0,

(4.11)

where log(·) is the principal branch of the logarithm.

4.2. Glutsyuk point of view and translation domains We discuss the case s= +1.We will denote

± = p^◦−_ε ¹(R±),

± = p^◦−_ε ¹(iR±). (4.12)

(16)

By Eq. (4.3), there is a countable number of such semi-infinite segments on R^ε, and by Eq. (4.11),^+,k and −,k lie on the same side of R^ε, but in different leaves. The same holds for ^+,k and −,k, see Figure 6. The half-lines (4.12) are organized differently in the cases ε 0 and ε > 0 (cf. [2]).

ε <0 ε= 0 ε >0

−

∞±

±

s−

−

− − +

∞−

∞±

∞± P⁰

P⁰ P^±

P^±

∞+

P⁰=P^±

s+

+ B_ε ± B_ε

B_ε

∞+

∞+ ±

±

s+ s+

Figure 6. — The choice of the cuts onRεfor real values of the parameter.

If the parameter is negative,the location of the ﬁxed points in thexcoordinate yields the decomposition

±=^s±∪ ^∞±

onR^ε,where ^s_± is the image byp^◦−1_ε of the straight real segment joining 0 andx_±,and^∞_± is the image byp^◦−1_ε of the straight real segment joining x_± and the boundary of the neighborhood U in the x coordinate. Again, one has a countable numberof such segments ^s,±^∞ at distance α(ε) from each otherin the Z coordinate. The cuts are located along the half-lines ±.The half-lines ^∞±,± intersecting the principal holeBε will be noted ± and±,respectively.

In the caseε= 0,there are four half-lines± and± in theZ coordinate.

They will be noted±and±,respectively. The cuts are located along±. For positive values of the parameter,on the contrary, the image of the imaginary axis by the mapp^◦−1_ε consists in the union

±=^s±∪ ^∞±

on R^ε, where ^s± is a countable collection consisting of the image of the straight imaginary segment joining 0 withx_±,where{^∞±}^k is an countable collection of the image of the straight imaginary segment joining x_± and

(17)

the boundary of the neighborhoodU in the xcoordinate. The cuts of R^ε are located along the half-lines^∞±.The half-lines±,^∞± intersecting the principal holeBε will be noted± and±,respectively.

Definition 4.3. — In the three cases above, the distinguished line ±

is called the symmetry axis in the Z coordinate.

Translation domains.Given any δ >0, there existsρ >0 such that for |ε|< ρ,there exists an orbit of the lifting Pε connecting P⁰ withP^±. In such a case, we say that we are in the “Glutsyuk point of view” of the dynamics.

C_ε( ) P^± P⁰

Q⁰_+,ε P_ε( )

Figure 7. — A translation domainQ⁰_+,εand an admissible strip on it

A slanted line⊂ R^ε,such that the imagePε() is placed on the right ofand the stripCε() betweenandPε() belongs top^◦−_ε ¹(U),is called an admissible line.Letbe an admissible line forPε.The translation domain associated to is the set

Qε() ={Z∈Uε:∃n∈Z,P^◦n_ε (Z)∈Cε(),∀i∈ {0,1, ..., n},P^◦i_ε (Z)∈Uε}. In the Glutsyuk point of view, the admissible strips are placed parallel to the line of holes, i.e. according to the α(ε) direction of the covering trans- formationTα.The induced translation domains, calledGlutsyuk translation domains,are notedQ^∞_ε andQ⁰_εaccording to whether they contain a neighborhood of P^± or P⁰,respectively, see Figure 7.

Among other properties,Qε() is a simply connected open subset ofUε; the region Cε()\{} is a fundamental domain for the restriction of Pε to Qε() : each Pε-orbit in Qε() has one and only one point in this set. For values ofεin Vδ,lr,there exist four diﬀerent Glutsyuk translation domains Q^0,_±^∞_,ε in theZ-space, which are deﬁned, depending on the sign ofε∈R,as follows, see Figure 8.

(18)

• Ifε <0, then Q^∞_±,ε is a simply connected neighborhood ofP^± containing all the segments^s±,whileQ⁰_±,ε is a simply connected neighborhood ofP⁰containing the distinguished half-line±.

• Ifε >0, then Q^∞_±_,ε is a simply connected neighborhood ofP^± containing all the segments^s±,whileQ⁰_±_,ε is a simply connected neighborhood ofP⁰containing the distinguished half-line±.

Q⁰₊_,_ε

Q⁰₊,ε Q^∞₊,ε

Q^∞₊_,_ε

s+

P⁰

P⁰ P⁺

P⁺

+

ε >0 ε <0

s + +

s+

Figure 8. — The translation domainsQ^0,∞_+,ε.

Lemma 4.4. — The translation Tα satisﬁes:

Tα(Q⁰_±_,ε) =Q⁰_∓_,ε,

Tα(Q^∞_±,ε) =Q^∞_±,ε. (4.13) Proof. — The second is clear, by definition:Tα is formerly defined in a neighborhood of the pointP^±along the leaves ofR^ε,thus leaving invariant the translation domains Q^∞_±_,ε.On the other hand, the first equality is cer- tainly true because all the possible paths defining the analytic extension of Tαto a neighborhood ofP⁰must be contained inQ^∞_±_,ε.Let us considerfor instanceQ⁰_+,εabove the principal hole. It intersectsQ^∞_+,εand because of the definition ofTα, when we applyTα (resp.T₋α) we are below the principal hole ifε >0 (resp.ε <0). In that regionQ^∞_+,ε intersects Q⁰_−,ε.Thus, each translation domainQ^∞_±,ε shares a common region with a translation domain of the kind Q⁰_±,ε.The conclusion follows.

4.3. Conjugation in the Z coordinate

Choose Z on R^ε and ﬁx any simple arc Γ joining Z with the axis of symmetry,and letγbe its image underthe mappε:γ=pε(Γ).Consider

(19)

the reﬂectionγof the pathγwith respect the real axisRin thexcoordinate.

Then deﬁne

Γ =p^◦−_ε ¹(γ).

Definition 4.5. — The pathΓis well deﬁned and is called the reﬂection of the arcΓwith respect to the axis of symmetry in the Z coordinate, see Figure 9. The starting point of Γ is called the conjugate ofZ,and is noted (Z).

The conjugationZ→(Z) is well defined: its definition is independent of the arc Γ.Indeed, if Γ+ is any simple path joiningZ with the semi-axis of symmetry⁺in theZcoordinate, then the reflection of the arc Γ+with respect to ⁺ induces a map

Z →+(Z)

along the leaves of,which is independent of the free homotopy class with endpoint on⁺.

γ₊

γ₊ x

x γ₋

γ₋

x₋ x₊

+ R

−

(Z) Z

Figure 9. — The conjugation in theZcoordinate.

Choose now any simple arc Γ₋ joining the pointZ with the semi-axis of symmetry −. The reﬂection of the arc Γ₋ with respect to− induces in turn a map Z → −(Z). Then, it is easily seen that ⁺(Z) = −(Z).

Indeed, the arc Γ+ induces a pathγ+ in thexcoordinate whose reﬂection γ+ with respect to the real axis starts at the same point as the reﬂection γ₋ of the pathγ₋ induced by the arc Γ₋ in thexcoordinate, see Figure 9.

It becomes clearby deﬁnition that:

◦=id (4.14)

(20)

for real values of the parameter. Moreover, the families Pε and Q_ε are invariant under the conjugationin the Z coordinate whenε∈R:

Pε=◦Pε◦,

Q_ε=◦Q_ε◦. (4.15)

5. Real Fatou Glutsyuk coordinates

LetW → C(W) =W be the complex conjugation inC,and let T^B(·) =B+·

be the translation inB ∈C.

Theorem 5.1. — For values of the parameter inVδ,lrwe consider translation domainsQ^0,_±^∞_,ε,lr whose admissible strips inR^εlie in a direction parallel to the line of the holesT_α(ε)^◦k (Bε).It is possible to construct four diﬀerent changes of coordinates W = Φ^0,∞_±_,ε,lr(Z)deﬁned on Q^0,∞_±_,ε,lr∩P^◦−_ε ¹(Q^0,∞_±_,ε,lr), and called Fatou coordinates with the following properties:

1. They conjugatePεwith the translation by one:

Φ^0,_±,ε,lr^∞ (Pε(Z)) = Φ^0,_±,ε,lr^∞ (Z) + 1, (5.1) for everyZ ∈Q^0,∞_±,ε,lr∩P^◦−_ε ¹(Q^0,∞_±,ε,lr).

2. They commute with complex conjugation, more precisely:

• For (real)negative values of the parameter they are related through:

Φ⁰_±,ε,l = C ◦Φ⁰_∓,ε,l◦,

Φ^∞_±_,ε,l = C ◦Φ^∞_±_,ε,l◦. (5.2) They are unique up to left composition with T^B±, B_± ∈ R, for Φ^∞_±_,ε,land up to T^D, D∈C,forΦ⁰_+,ε,l(Φ⁰₋_,ε,l being determined by Eq. (5.2)).

• For (real)positive values ofε they satisfy:

Φ⁰_±_,ε,r = C ◦Φ⁰_±_,ε,r◦,

Φ^∞_±,ε,r = C ◦Φ^∞_∓,ε,r◦. (5.3) They are unique up to left composition with T^B±, B_± ∈ R, for Φ⁰_±,ε,land up toT^D, D∈C, for Φ^∞_+,ε,l(Φ^∞_−,ε,l being determined by Eq. (5.3)).

3. They depend analytically on ε∈Vδ,lrwith the same limit atε= 0.

(21)

Proof. —

1. The construction of Fatou coordinates satisfying Eq. (5.1) exists in the literature (cf. [16]) but we wish to show additionally Eqs. (5.2) and (5.3).

2. We use the result (also well known in the literature) that a Fatou coordinate Φε on a translation domain Qε is unique up to left composition with a translation. In particular, it is uniquely determined by a base pointZ0(ε) such that Φε(Z0(ε)) = 0.

Caseε <0.Let Φ^∞_±_,ε,lbe a Fatou coordinate determined by the equa- tion Φ^∞_±_,ε,l(Z0,±(ε)) = 0.From Eqs. (4.14), (4.15), it is easy to check thatC ◦Φ^∞_±_,ε,l◦is a Fatou coordinate and, moreover, that it satis- ﬁesC ◦Φ^∞_±_,ε,l◦((Z0,±(ε))) = 0. Then,C ◦Φ^∞_±_,ε,l◦=T^D±◦Φ^∞_±_,ε,l forsome D_± ∈iR. Changing Φ^∞_±_,ε,l byT^D±

2 ◦Φ^∞_±_,ε,l yields a Fatou coordinate satisfying the second part of (5.2). As for the first part, it suffices to define Φ⁰₋_,ε,l,from the choice of a Fatou coordinate Φ⁰_+,ε,l, according to the first row of Eq. (5.2). The caseε >0 is similar.

3. It suﬃces to take the base point depending analytically on ε∈Vδ,lr

with same limit atε= 0.

Q^∞_+,ε∩Q⁰_+,ε

Q^∞_+,ε∩Q⁰_+,ε Q^∞_+,ε∩Q⁰₋_,ε

Q^∞_+,ε∩Q⁰₋_,ε

P⁰

P⁰ P^±

P^± Q⁰_+,ε∩Q^∞₋_,ε

Q⁰_+,ε∩Q^∞₋_,ε

Q⁰_+,ε∩Q^∞_+,ε Q⁰_+,ε∩Q^∞_+,ε

ε >0 Bε

Bε

ε <0

Figure 10. — The non-connected intersection of the translation domains.

Definition 5.2. — Fatou coordinates in Theorem 5.1 are called admissible Real Fatou Glutsyuk coordinates. Theorem 5.1 shows that the symmetry axis is invariant under Real Fatou coordinates when the parameter is real.

(22)

Remarks.

1. Although Real Fatou Glutsyuk changes of coordinates always exist forε∈Vδ,lr,the curve is not invariant ifε /∈R.

2. The subscriptsl, r will be dropped when the context allows no con- fusion.

3. As we will see, the modulus compares the Fatou coordinates on the left and on the right over the intersection of the left and right translations domains. Ifε= 0,the geometry ofR^εyields that the intersection of right and left translations domains is composed of a countable alternating sequence of horizontal strips, see Figure 10.

4. This yields the organization of the domains of deﬁnition for the different Real Glutsyuk coordinates. Due to periodicity, it suﬃces to describe these domains around the fundamental holeBε,see Figure 11.

Φ^∞_+,ε

Φ^∞₋_,ε

Φ⁰_+,ε Φ⁰₋_,ε Φ⁰₋_,ε Φ⁰_+,ε

Φ⁰_+,ε

Φ⁰₋_,ε

Φ^∞_+,ε Φ^∞₋_,ε +

+

Φ^∞₋_,ε Φ^∞_+,ε ε >0

B_ε B_ε

ε <0

Figure 11. — The Real Glutsyuk coordinates around the principal hole.

5.1. Pre-normalization of Real Fatou Glutsyuk coordinates The familyP^εis, by definition, the second iterate of a family of germs of diffeomorphisms Q^ε unfolding the map Q⁰, which is tangent to −id. This implies that the orbits of the family Q^ε form a 180^◦-degrees alternating sequence along the orbits of the prepared family of fields at each iteration (i.e.the pointsxandQ^ε(x) stand on opposite sides of the origin, see Figure 12). In otherwords, the liftingQ_εexchanges the two leaves.

The fact that the family of diﬀeomorphisms P^ε is a square (namely, P^ε=Q^◦ε²) is now exploited.

(23)

x

Q0(x)

Figure 12. — The “jumps” of the orbits ofQ0 in the caseε= 0.

Lemma 5.3. — For each ε∈ V_δ^G, it is possible to construct admissible Real Fatou Glutsyuk coordinates depending analytically on ε ∈ Vδ,lr, with continuous limit at ε= 0 and such that they are related through:

Φ⁰_±,ε◦Q_ε=T¹₂ ◦Φ⁰_∓,ε,

Φ^∞_±_,ε◦Q_ε=T¹₂ ◦Φ^∞_∓_,ε. (5.4) Such coordinates are unique up to left composition with translation T^C⁰ε

(resp. T^C^∞ε ) for real constant C_ε⁰ (resp. C_ε^∞) (i.e. the same constant for the two signs±in each case).

Proof. — Foreach ε, the map Q^ε commutes with P^ε. Hence Q_ε = p⁻¹_ε ◦ Q^ε◦pε commutes with Pε. Let the pairs of admissible Real Fatou Glutsyuk coordinates Φ^0,_+,ε^∞,Φ^0,₋^∞_,ε be constructed as in the proof of Theorem 5.1. Then:

Φ^0,_±_,ε^∞(Pε(Q_ε(Z))) = Φ^0,_±^∞_,ε(Q_ε(Z)) + 1 = (Φ^0,_±^∞_,ε ◦Q_ε)(Pε(Z)), (5.5) the ﬁrst equality being consequence of the fact that Φ^0,_±,ε^∞ is a solution to Eq. (5.1), and the second is true becausePεandQ_εcommute. Eq. (5.5) says that Φ^0,∞_±,ε ◦Q_ε is a Fatou Glutsyuk coordinate. By the remark above, the latteris deﬁned on the same translation domain as Φ^0,_∓^∞_,ε.Hence, as in the proof of Theorem 5.1, there existsC_±^0,_,ε^∞∈Cwith the following property:

Φ^0,_±^∞_,ε ◦Q_ε=TC_±^0,_,ε^∞◦Φ^0,_∓^∞_,ε. (5.6) We will drop the subscriptεin the constants. UsingQ^◦_ε²=Pεand iterating Eq. (5.6) yields:

Φ^0,_±^∞_,ε(Z) + 1 ≡ Φ^0,_±^∞_,ε ◦Pε(Z)

= (Φ^0,_±^∞_,ε ◦Q_ε)◦Q_ε(Z)

= TC^0,_±^∞◦(Φ^0,∞_∓,ε ◦Q_ε)(Z)

= TC^0,_±^∞◦ TC_∓^0,^∞◦Φ^0,_±^∞_,ε(Z)

= Φ^0,_±^∞_,ε(Z) +C_±^0,^∞+C_∓^0,^∞,

(5.7)

The modulus of analytic classification for the unfolding of the codimension-one flip and Hopf bifurcations

ANNALES

DE LA FACULTÉ DES SCIENCES

Mathématiques

cedram

The modulus of analytic classiﬁcation for the unfolding of the codimension-one ﬂip

and Hopf bifurcations

Contents