A-STABILIZATION AND THE RANGES OF COMPLEX POLYNOMIALS ON THE UNIT DISK ALEXEY SOLYANIK 7 1 To Dmitriy Dmitrishin and Alexander Stokolos 0 2 n a J 7 1 Abstract. Problemsofstabilizationoftheunstablecycleofone-dimensionalcom- plex dynamical system are briefly discussed. These questions reduced to the prob- ] A lemofdescriptionoftherangesofpolynomialsq(z)=q1z+q2z2+ +qnzndefined ··· N in the unit disk and normalized by the conditions q(1) = 1 and this is the main . subject of the present paper. h t a m 1. Introduction [ Intherecent papers[Dm 2013], [Dm 2014], [Dm 2015]authorsstudiedtheproblem 1 of local stabilization of unstable cycle of length T of the given real one-dimensional v dynamical system. A typical example is the family of logistic maps g (x) = λx(1 x) 4 λ − 8 defined on the interval [0,1] for λ [1,4] and demonstrate a chaotic dynamics for a 7 ∈ large set of λ close to 4. 4 0 Inthisworkwestudythisstabilizationprocessbutforthecomplex one-dimensional . dynamical system 1 0 7 (1.1) f : Cˆ Cˆ → 1 : where Cˆ = C is the Riemann sphere with spherical metric. This is of course a v ∪{∞} i very mild generalization, and generalization actually was not a target of the present X work. Most of the results are new even in the real case, since all of the dynamics of ar f restricted to R are contained in the dynamics on C. This notes mainly devoted to the local theory and the global behaviour will be the subject of the forthcoming paper. One remark is in order. Since all described theory is mostly devoted to the spectral analysis of the given linear operator, we can apply these results also to the problems of stabilization of unstable cycles for smooth dynamical systems defined on the given finite dimensional smooth manifold. 1.1. Complex dynamical systems. A basic question in the theory of dynamical systems is to study the asymptotic behaviour of orbits orb(f,z ) = (z ,f(z ),f2(z ),...), 0 0 0 0 Date: 17 January 2017. 1 2 ALEXEY SOLYANIK where fn(z) = f(f(...f(z))) and initial point z Cˆ. 0 ∈ A point s, such that f(s) = s is called a stable (or fixed) point of the map f. The local behaviour of the dynamical system at the stable point governed by the eigenvalues of the linear part. We call a fixed point s attracting if the multiplier µ = f (s) satisfied µ < 1. Then all orbits of f convergence uniformly to s on the ′ | | some neighbourhood of s and point s is also called asymptotically stable. A point s is called periodic if fT(s) = s for some T. The minimal T is its period and the orbit = (s,f(s),f2(s),...fT 1(s)) is called a cycle. We say that the cycle − O is attracting when the multiplier of the cycle µ( ) = (fT)(s) satisfied µ < 1. ′ O O | | For precise definitions and background theory of complex dynamics we refer to the books [Be 1991], [CG 1993] or [Mi 2006] but in this work we will use only elementary facts from the theory. As a simplest (but typical) example of the complex (or holomorphic) dynamical system (1.1) one can imagine iteration of f(z) = z2. It is clear that s = 1 is a repelling fixed point for this system, which means that f(s) = s and, since f (1) = 2, all points close enough to s repel by the map f away ′ from the point s. Our goal is to stabilize this system (at least locally) at the point s in such a way that all points close enough to s became attracted to the point s. Before we involve to the stabilization problems, we shall shortly describe the global dynamics of f(z). Since fn(z) = z2n, initial points z with z < 1 tends to 0 and 0 0 | | initial points z with z > 1 tends to . This means that this system has two 0 0 | | ∞ attracting fixed points, namely 0 and and one repelling fixed point 1. This is a ∞ reflection of general fact, that any rational function of degree d has precisely d + 1 fixed points on Cˆ counting like zeros of equation f(z) z = 0 (see [Be 1991], p. 40). − What about points on the circle ∂∆ = z : z = 1 , we have to say that their { | | } dynamics is quite complicated and we still do not know answers to a very simple questions. For instance, it seems unknown any accumulated point of the sequence sin2n = (fn(ei)). ℑ Meanwhile, of course, all initial points with z = 1 not leave the unit circle and 0 | | even more – the circle is both forward and backward invariant under f (that is each point on ∂∆ has its entire history and future lying on ∂∆). Loosely speaking, a unit circle looks like an unstable ’orbit’ (actually invariant set) for this dynamical system. Unstable, since if some iteration fn(ζ) fall from the unit circle inside or outside of the circle (for instance by the reason of approximation) it never return back. Butthemaindifferencewithperiodicorquasi-periodicmotionisthatthedynamics off(z) = z2 ontheunitcircleischaotic. Thewordchaotic hasmanymeanings, which we briefly discus here. First of all this means that the dynamics is sensitive to the initial conditions, i. e. two close to each other initial points z and w after some number of iterations 0 0 produce a later changes that grows exponentially with n (exponentially means (1.2)). It implies that long term predictions of the chaotic system is almost impossible despite the deterministic nature of the equation. For generic initial points their A-STABILIZATION AND THE RANGES OF COMPLEX POLYNOMIALS ON THE UNIT DISK 3 trajectories, which are at first very close, later diverge more and more rapidly until they no longer have anything to do with each other. Next, the dynamics is topologically transitive, which means that some point (and in fact any generic point ) has everywhere dense orbit in ∂∆ (see [Mi 2006], p. 51). Since the equation z2T = z for every T has 2T 1 roots, which are equally dis- − tributed on the circle, any such points is periodic with period at least T (but the least period could be less of course). Multipliers of these cycles µ = (fT)(z) = (z2T) satis- ′ ′ fies µ > 1, and hence they are all repelling cycles. We shall claim also, that for each | | natural number T there are periodic points with exact period T (η = exp( 2πi )) T 2T 1 and hence this dynamical system has periodic orbits of any order. − The most common definition of chaotic behaviour is to say that f has sensitively dependence on the initial conditions, f is topologically transitive and periodic orbits are dense([Dv 1989], p. 269), which are all satisfied by the system f(z) = z2 on the invariant set ∂∆ . In fact, for compact infinite metric spaces and continues transformations, the last two conditions implies the first one. We conclude our brief description of chaotic behaviour pointed out that the map f(z) = z2 on the unit circle clearly demonstrates two key features of chaos — stretch- ing and folding: the map stretched the unit circle in 2 times and 2 times folded it. Loosely speaking, stretching mechanism is responsible for sensitivity to initial condi- tions while the folding mechanism is responsible for topological mixing (topological transitivity and density of periodic orbits). Thus, in this case the so called Fatou set is Cˆ ∂∆, with the regular dynamics of f \ and the Julia set is the unit circle ∂∆, where dynamics is chaotic (and ergodic with respect to the one-dimensional Lebesgue measure on the circle). Actually this is a general case at least for all rational functions f with degree not less than 2 – we have partition of the Riemann sphere into two disjoint invariant sets, on one of which, so called Fatou set, f is well-behaved as dynamical system, on the other of which, the Julia set, f has chaotic behaviour, i.e. sensitive to the initial conditions, topologically transitive and repelling periodic points are dense. Excluding some exceptional cases, Julia set is a fractal set with Hausdorff dimension bigger than 1 and corresponding dynamics is ergodic on the Julia set with respect to the corresponding Hausdorff measure. We note that any holomorphic map from Cˆ to Cˆ is the rational function R and any rational function demonstrate interplay between expanding and contracting features, which is the main reason of complicated dynamics. Indeed, any rational function of degree d folding d times the Riemann sphere and hence, in average, expanding space in d times. On the other hand it has 2d 2 − critical points, counting multiplicity, where R(z) = 0 (or R has a pole of order two ′ and higher) and hence R highly contracting local neighbourhoods of those points (see e. g. [M1 1994]) However wehavekeep inmind, thatforsomerationalfunctions f theJuliaset J(f) equal Cˆ, the fact first discovered by Ernst Schroder in 1871 and then rediscovered in greater generality by Samuel Late in 1918 ([Mi 2006], p. 70). The simplest example is f(z) = 1 2/z2 (see [Be 1991] p.76). − 4 ALEXEY SOLYANIK If J(f) = Cˆ then it is a closed set without isolated points and with empty interior 6 and Fatou set is open set with at most countably many open components Ω, such that f(Ω) is also open component of Fatou set. Some times a given component Ω is stable, i. e. f(Ω) = Ω (like in our example f(z) = z2) and some times not (like for f(z) = z2 1). − In the latter case the corresponding Julia set is the typical fractal set called the Basilica di San Marco after John Hubbard or shortly basilica and Fatou set has infinitely many open components Ω ([Be 1991], p. 13). The component Ω contain- 0 ing point 0 is periodic with period 2, i. e. f2(Ω ) = Ω , as well as component Ω 0 0 1 − containing point 1. This is because points 0 and 1 form a 2-cycle. Other compo- − − nents Ω are eventually periodic, i. e. component became periodic with period two after some number of iterations. The component Ω , contained is stable, i. e. ∞ ∞ f(Ω ) = Ω . ∞ ∞ According to the famous theorem of Dennis Sullivan, solved the 60 year old Fatou - Julia problem on wandering domains ([Be 1991], p. 176), this is a general case — there is no wandering components in Fatou set and every component Ω is eventually periodic (or stable). Points of the Fatou set is exactly points which are stable in the Lyapunov sense, i. e. for every point z from the Fatou set every point w close enough to the point 0 0 z remains uniformly close for all iterations. Hence there is no repelling cycles in the 0 Fatou set and all repelling cycles are in the Julia set (see [Be 1991], p. 109). The number of components of the Fatou set could be 0, 1, 2 or ([CG 1993], p. ∞ 70). For the Latte maps there is 0 components, for f(z) = z2 2 one, for f(z) = z2 − two and for f(z) = z2 1 infinitely many. − After this short excursion to the complex dynamics we shall return to the main subject of the article — the local stabilization of the given dynamical system near repelling fixed point or repelling cycle. 1.2. A-stabilization. Let beaunstablecycleofthedynamical system f : Cˆ Cˆ. O → Following [Dm 2013] we use some averaging procedure to stabilize this dynamical system near . O The idea of stabilization procedure is well-known and simple. In Control Theory it calls feed-back control. We measure some previous states (trajectory) of the given dynamical system near given unstable cycle and then add a control which correct the next state. The control procedure should be of course independent of the given initial state of the system. This means that it should be the same for all initial states from the some (small enough) neighbourhood of the unstable cycle. To clarify the ideas we start with the description of stabilization of given unstable fixed point s (i. e. cycle of length one). Let z be a point close enough to s. This is an initial state. Consider the first n 0 points of the orbit, i.e. z ,f(z ),f2(z ),...fn 1(z ), which we call z ,z ,z ,...z 0 0 0 − 0 0 1 2 n 1 − and this is a part of the trajectory of z . Next point of this orbit, namely f(z ) 0 n 1 could lie far enough from the desired fixed point s. Actually, for f(z) = z2 with t−he A-STABILIZATION AND THE RANGES OF COMPLEX POLYNOMIALS ON THE UNIT DISK 5 multiplier µ = f (1) = 2 we have ′ (1.2) dist(f(z ),1) exp(nlog2)dist(z ,1) n 1 0 − ≍ at least for z close enough to 1 and numbers n log (dist(z ,1)) 1 . This means 0 ≪ 2 0 − that point s is unstable and now our aim is to stabilize it. Let a = a ,a ,...,a be an averaging set of complex numbers which will be 1 2 n h i { } chosenlater. Letz isastartingpoint, whichiscloseenoughtosand(z ,z ,...,z ) 0 0 1 n 1 − be the first n points of orbit. Define the new point z , which we still denote as z later, by the rule n∗ n z = f(z )+control = a f(z )+a f(z )+ +a f(z ) n∗ n−1 1 n−1 2 n−2 ··· n 0 in such a way, that new one z lie more close to s than the old one z = f(z ). n∗ n n 1 − Continue in the same way and define for m = n+1,n+2,... the new trajectory by the rule z = a f(z )+a f(z )+ +a f(z ) m 1 m 1 2 m 2 n m n − − ··· − where all points z are points of the new trajectory, which coincide (generally) with k the old one only at points z ,z ,...,z . We call this stabilization process corre- 0 1 n 1 − sponding to the given set a = a ,a ,...,a the a -stabilization. 1 1 n h i { } h i We would like to stress out here that z is a new point and not coincide with the n n-th point of the old orbit. It have to be clear also that the described process is not a process of averaging of the given (old) orbit of the dynamical system, rather the process of producing completely new orbit. Actually, this new system is not any more a dynamical system, it is a difference equation of n-th order. Philosophically speaking, a dynamical system has no memory and hence more easy bifurcate to the chaotic regime. When we stabilize the system by the use of memorialised coordinates it transforms to the system with memory and demonstrate more regular local behaviour. Suppose that we successfully achieved our goal and z s and let s = 0. Then m → 6 for all m big enough and k = 0,1,...,n 1 we have f(z ) f(s) = s and hence m k − − ≈ s z (a +a + +a )f(s) = (a +a + +a )s m 1 2 n 1 2 n ≈ ≈ ··· ··· It follows that necessary condition on the set a is h i (1.3) a +a + +a = 1 1 2 n ··· Define the polynomial corresponding to the averaging set a by h i (1.4) p(z) = a +a z + +a zn 1 n n 1 1 − − ··· We stress out that coefficients now are in the reverse order. Now the necessary condition (1.3) reads as p(1) = 1 and we always assume that R a = 0. Denote the set of all such polynomials by and by the set of such 1 6 Pn Pn polynomials with real coefficients. Forthecyclestabilizationonecanusethefollowinggeneralizationof a -stabilization, h i which we call a,T -stabilization (of T-cycle). h i 6 ALEXEY SOLYANIK Let the dynamical system has an unstable (repelling) cycle = (s ,s ,...,s ) of 1 2 T O the length T. Then the a,T -stabilization of T-cycle defined as follows h i z = a f(z )+a f(z )+ +a f(z ) m 1 m 1 2 m 1 T n m 1 (n 1)T − − − ··· − − − with long enough initial points of trajectory to start the process. Using time-delayed coordinates one can rise to a new (and now indeed dynamical) system F : C(n 1)T+1 C(n 1)T+1. It can be shown that after linearisation of F near − − → the point of the cycle the stabilization problem reduced to the position of roots of the polynomial (1.5) χ (z) = z(n 1)T+1 µpT(z) T − − where p(z) = a +a z + +a zn 1 n n 1 1 − − ··· is the corresponding polynomial for the a -set and h i µ = µ( ) = f (s )f (s )...f (s ) ′ 1 ′ 2 ′ T O is the multiplier of the cycle. We do not posses here an explanation of these definitions and proofs of previous statements and refer to the recent papers ([Dm 2014], [Dm 2015]), where this matter discussed in details. Thus from now the problem of a,T -stabilization has a pure algebraic context – h i for the given natural number T and complex number µ to find some averaging set a h i in such a way that all roots of polynomial χ (z) lie in the unit disk ∆ = z : z < 1 . T { | | } We fix the definition that the (holomorphical) system f : Cˆ Cˆ admits a,T - → h i stabilization of unstable cycle with multiplier µ( ) = µ if this is the case, i.e. if O O all roots of corresponding polynomial χ (z) lie in the unit disk. T Here and through the paper T is the natural number which is equal to the length of the cycle and µ is the complex number which is equal to the multiplier of the cycle. We show that for every complex number µ (which lie outside of the unit disc, i.e. the stabilized cycle should be repelling cycle) we can choose the finite set of complex numbers a in such a way, that all roots of χ (z) lie in the unit disc ∆. T h i Hence every dynamical system f : Cˆ Cˆ with the repelling cycle could be a- → O stabilized near this cycle. This means that for any initial condition z close enough to 0 s the new (a-stabilized) orbit orb (z ) = (z ,z ,z ...) tends to the cycle exponentially a 0 0 1 2 fast. For instance, as we show later, for our simple example f(z) = z2 with µ = 2 at the fixed (but repelling) point s = 1 it is impossible to find 3 numbers a ,a ,a in 1 2 3 such a way, that χ (z) has all it roots inside of the unit disk ∆, but we can choose 4 1 numbers to reach the desired aim. Namely these numbers (not unique but in some sense the best) are a = 2 √2+i√2, a = 3(√2 1) i3(√2 1) 1 2 − − − − a = 2(√2 1) i2(3 2√2), a = i(3 2√2) 3 4 − − − − − and hence dynamical system f(z) = z2 became asymptotically stable at the fixed point 1 after a -stabilization with a = (a ,a ,a ,a ). 1 2 3 4 h i h i A-STABILIZATION AND THE RANGES OF COMPLEX POLYNOMIALS ON THE UNIT DISK 7 We will show in the section 4 that we can not choose the real a in the case when k multiplier is real and bigger than 1. Let now f(z) = z2 2, which has two (finite) fixed points, namely s = 1 with − − multiplier µ = f ( 1) = 2 and s = 2 with multiplier µ = f (2) = 4. Hence both ′ ′ − − fixed points are repelling. At the point s = 1 the multiplier µ = 2 which has the same modules as − − multiplier in previous example but opposite sign. Now we can choose 2 real a for k the local a-stabilization near the point 1: − a = √3 1, a = 2 √3, 1 2 − − These examples shows that the minimal length of the set a , which we denote by h i ord a depends not only of the magnitude of multiplier, but rather of it position on h i the complex plane. 1.3. Stabilization domain and p/q-duality. For the given polynomial p we n ∈ P define it stabilization domain for T-cycle by ST(p) = µ : all roots of χ (z) lie in ∆ T { } This set describe the set of values of multiplier of the given cycle of length T for which this cycle could be asymptotically stabilized via chosen a,T -stabilization h i process. For the given family of polynomials we define it stabilization domain for P T-cycle by ST( ) = ST(p) P p [∈P We denote by ST the stabilization domain for T-cycle for the family . n Pn The set ST describes the largest domain of those µ for which we can find a,T - n h i stabilization process of the order (length) n asymptotically stabilizing a given cycle of the length T with the multiplier µ. Often we just omit the upper index T when it equal to one. Thus, for instance, we will write S instead of S1. n n In the section 2 we give purely geometrical description of the set S . It turns out n that S is an open set bounded by the sinusoidal spiral (we refer to [Ya 1952], p. n 214 for definition) with the one punctured point inside. Namely, for instance, S is 2 bounded by cardioid and S by Cayley’s sextic (with the punctured point 1 ). 3 { } Since S is a union of all S(p) for p it is useful to find some subfamily of n n ∈ P simple polynomials p , such that S(p ) also cover S . We show that for polynomials α α n of very special kind their stabilization domains S(p ) indeed cover all set S . This α n gives the practical rules to design the a-stabilization sets. In the section 4 we describe so called p/q -duality, which allow us to reformulate the problem of roots position of χ (z) in ∆ to the problem of image position of q(∆¯), T where q(z) = z(p (z))T and p (z) is the inverse polynomial of p(z). ∗ ∗ Thisleadstothesimplepracticalrule: the set a stabilize locally any dynam- h i ical system with the multiplier µ if and only if q(z) = a z+a z2+ +a zn 1 2 n omit point 1/µ in ∆¯. ··· 8 ALEXEY SOLYANIK 1.4. Stabilization of a family of maps. Usually, for the given family of dynamical systems f (z) parametrized by the parameter c we have a partition of the parameter c space into disjoint regions in such a way that dynamics of the iterates of f in this c regions display essentially the same features, while as c passes from one region to another, some significant change in the dynamics (bifurcation) take place. For instance, for the family of quadratic polynomials f (z) = z2 +c, the partition c of the parameter plane C leads to the Mandelbrot set = c : fn(0) bounded when n M { c → ∞} This extremely popular mathematical object (also in the non-mathematical world) is now well understood, except probably one, but central question of the theory — is it true that is locally connected. M We recall that a hyperbolic component of the set is an open connected com- H M ponent of the set of parameters c, such thatf has an attracting cycle . c c O It is known that for every c the attracting cycle has constant period c ∈ H O through and moves holomorphically through the parameter c moves in . c H O H Thus in hyperbolic components the behaviour of f (z) is structurally stable or robust. c The question of whether there exist non-hyperbolic components is open and essen- tially goes back to the classical work of Pierre Fatou (1920). On the other hand in hyperbolic components the corresponding repelling cycles c O has also a constant period through and lie in the corresponding Julia sets. H Previous observations rise to the next problem in a-stabilization — to find one a-stabilization process for the given domain of parameters or, in other words, for the given domain of multipliers. As a simple example consider the family of quadratic polynomials f (z) = z2 +c, c where a parameter c lie in the one of hyperbolic components of Mandelbrot set, let say in the main cardioid. Then every f has 2 fixed points, one attractive and one c repelling. At the repelling fixed point as parameter moves through the main cardioid, the corresponding multiplier describe the set M = z : z 2 < 1 which is a circle. { | − | } Thus to stabilize every dynamical system from this hyperbolic component by the one stabilization process (or in other words by the one set a ) we have to find one h i polynomial p(z), such that M S(p). ⊆ Now by the p/q-duality we reduce this question to the problem of finding a poly- nomial q(z) of the smallest degree, such that q(0) = 0, q(1) = 1 and q(z) omit the set M = w : 1/w M when z ∆¯. It is easy to see that in this particular case ∗ { ∈ } ∈ M = w : w 2/3 < 1/3 is a circle. ∗ In o{ther |wo−rds we| have }to find a polynomial q(z), such that q(∆¯) Ω , where M ⊆ Ω = Cˆ M , which is the sphere with the hole. We shall to stress out that in this M ∗ \ particular case the set Ω is closed (but not a simply connected). Hence there also M no obstructions to find the extremal polynomial in question. After we find polynomial q(z) it coefficients give us the desired set a which h i stabilise every system f (z) = z2 + c at the repelling fixed point for every value of c parameter c from the main cardioid. This how it works and now we consider the general situation. A-STABILIZATION AND THE RANGES OF COMPLEX POLYNOMIALS ON THE UNIT DISK 9 Let M bea subset ofthe complex plane and f be agiven familyof dynamical c c systems, such that every f has an unstable cy{cle} ∈X= (s ,f (s ),f2(s ),...fT 1(s )) c Oc c c c c c c − c with the multiplier of the cycle µ( ) M. c O ∈ We shall to find one polynomial p(z) of the smallest degree, such that M ST(p). ⊆ Coefficients ofthispolynomialisthecorrespondingset a which a,T -stabilizeevery h i h i unstable cycle . c O Accordingtothep/q –dualitythisproblemequivalent totheproblemofdescription of the range q(∆¯) for polynomials q(z) satisfied some conditions, namely to find polynomial q(z) = z(p (z))T of the smallest degree, such that q(1) = 1 and q(z) omit ∗ in the closed unit disc some prescribed set of values M . ∗ Our approach to the solution be as follows. First we show that the range restric- tions implies estimates of (Taylor) coefficients of polynomial q(z). This is a general principle of Geometric Function Theory. From these estimates and normalization condition q(1) = 1 we can get an estimates from below of the degree of the polyno- mial q(z). Forsimplyconnected domainΩ = C M weshallgivealsothedifferent approach M ∗ \ based on the subordination principle and growth estimates near the boundary of the Riemann function g : ∆ Ω conformally mapping ∆ onto Ω . ΩM → M M This approach lead to the degree estimate from below for polynomial q(z) for any simply connected set Ω . M In order to find concrete sequence a which stabilize the given set of unstable h i cycles, we have to find polynomial q(z) of the smallest degree which omit prescribed set of values in ∆¯, or in other words it image q(∆¯) lie in the prescribed set Ω M There is a some different approaches to solve this problem. First one is based on some classical results of Ted Suffridge ([Sf 1969] ) and can be applied for slit domains Ω . This lead to the extremal a,1 and a,2 stabilization M h i h i sets for real systems (see also [Dm 2013] and [Dm 2014]). Second one is based on the result of V. Andrievski and S. Ruscheweyh ([AR 1994] from Approximation Theory, which we call Theorem A. We use Theorem A to con- struct polynomial which subordinate to g (z) and superordinate to g ((1 c)z). ΩM ΩM − n After normalization this polynomial became the desired polynomial q(z). Third way based on the theory of maximal rangeof A. Cordova and S. Ruscheweyh ([CR 1989], [CR 1990]) and give rise to the extremal polynomials in question. But this approach has disadvantage that it works only for very special kinds of domains Ω . At least is perfectly works for slit domains and circular domains. But the ex- M tremal solutions are of a little interest in the questions of stabilization since they are very sensitive to the averaging coefficients choose — they could loose their stabiliza- tion properties after a little change in coefficients. This why we do not consider in these notes extremal solutions. We now briefly summarize some concrete results. Let f be a given family of dynamical systems, such that every f has an c c c unstab{le c}yc∈lXe = (s ,f (s ),f2(s ),...fT 1(s )) of length T with the multiplier of the cycle µ( )Oc M, wchecre cM ics acgivencs−ubsect of the complex plane C. c O ∈ 10 ALEXEY SOLYANIK Denote by µ = sup z : z M the size of the multipliers set M. It is clear M {| | ∈ } that the length of the stabilization sequence a have to depend of the size of M. But h i it also deeply depends of the shape of M and almost not depend of the length of the cycle. For the one point set M = µ , where µ 1 and µ = 1 we can always find a { } | | ≥ 6 stabilization set a of the order h i (1.6) ord a logµ M h i ≍ and this estimate is the best possible in order. For the real segment M = z : z = 0, µ z 1 , where µ > 1 we can M M { ℑ − ≤ ℜ ≤ − } always find the real stabilization set a , such that h i (1.7) ord a √µ M h i ≍ and again this estimate is the best possible in order. For the (left) horocycle M = z : z +µ /2 µ /2 we can always find the real M M { | | ≤ stabilization set a , such that h i (1.8) ord a µ M h i ≍ and again this estimate is the best possible in order. Let now M = z; z µ , argµ θ is a (wide) sector of a (big) circle. This set M { | | ≤ | | ≥ } M (in some sense) is the largest domain of admissible multipliers, since the necessary condition on the set M to satisfy the inclusion M ST(p) is that the set Cˆ M have ⊆ \ to contain some open subset connected points 1 and on the Riemann sphere Cˆ. ∞ As we show later for this set M we can find a real stabilization set a , such that h i (1.9) ord a c(θ)expµ M h i ≍ and the estimate is the best possible in order. We stress out that even for θ = π/2 we can get only exponential order of the length of the stabilization set. We shall to point out that in this notes we do not use any specific properties of complex dynamical systems and all this results are true as for a complex as well for a real dynamical systems. The only difference is that for the real systems the set of multipliers is also a real subset of the complex plane as well as a stabilized sequence a . h i In the forthcoming article we give an applications of previous results and de- scribe the global dynamics of two typical complex dynamical systems f (z) = z2 0 and f (z) = z2 2 after a-stabilization. These examples corresponds to the ’center’ 2 − − point c = 0 of the Mandelbrot set with J(f ) = ∂∆ and to the (left) extreme 0 M point c = 2 of with J(f ) = [ 2,2]. 2 − M − − For the family of real dynamical systems g (x), mentioned at the beginning, this λ two examples corresponds by the formula λ λ c = (1 ) 2 − 2 to the main trunk (c = 0 or λ = 2) of bifurcation diagram, where dynamics is completely regular and to the top of the bifurcation tree (c = 2 or λ = 4), where − dynamics is completely chaotic.