This work is licensed under the Creative Commons Attribution 4.0 International License.
Introduction
1. This paper is the direct continuation of the work [18], where stability of the Sharkovsky's order respectively small C1-smooth perturbations of skew products of interval maps is proved. Results of [18] are announced in [19], where the part of the Author's report at the Conference “Mathematical Physics, Dynamical Systems and Infinite-Dimensional Analysis” (17–21 June 2019, Dolgoprudny, Russia) devoted to periodic orbits of C1-smooth maps defined below is presented.
We consider a map F of a closed rectangle I = I1 × I2 into itself, where I1, I2 are closed intervals of the straight line R1, Ik = [ak, bk] for k = 1, 2, and F satisfies the equality
F(x,{\kern 1pt} y) = (f(x) + \mu (x,{\kern 1pt} y),{\kern 1pt} g(x,{\kern 1pt} y)){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\rm for}\;{\rm any}{\kern 1pt} {\kern 1pt} {\kern 1pt} (x,{\kern 1pt} y) \in I.
Further we use the notation gx(y) for g(x, y), where (x, y) is an arbitrary point of the rectangle I.
In the paper [31] the integrable system of differential equations is constructed so that this system approximates the Lorenz system of differential equations [3]. The paper [31] generated the wave of interest to dynamical systems (1) (see, e.g., [5], [29] – [30]) for the case of the discontinuous Lorenz map f of the closed interval I1 into itself [8], [26].
Following [18], [19] we suppose in this paper that the map (1) is C1-smooth on I, and the map f : I1 → I1 is so that the conditions hold:
(if) f (∂ I1) ⊂ ∂ I1, where ∂ (·) is the boundary of a set;
(iif) f is the Ω-stable in the space of C1-smooth self-maps of the interval I1 with the invariant boundary.
We suppose also that the C1-smooth function μ (of variables x and y) satisfies the boundary conditions:
(iμ) the equalities μ(x, a2) = μ(x, b2) = 0 are valid for every x ∈ I1; and the equalities μ(a1, y) = μ(b1, y) = 0 are valid for every y ∈ I2.
By the properties (if) and (iμ) the set ({a1} × I2) ∪ ({b1} × I2) is F-invariant. If, in addition, the inclusion gx(∂ I2) ⊂ ∂ I2 holds for all x ∈ I1 then the union of the horizontal intervals I1× {a2} and I1× {b2} is F-invariant too.
2. Give the list of the functional spaces connected with the map (1).
Let
\widetilde C_\omega ^1({I_1})
be the set of all C1-smooth maps of the interval I1 into itself satisfying conditions (if) – (iif). The standard C1-norm || · ||1,1 of the linear normalized space of C1-smooth maps of the interval I1 into the straight line R1 generates the C1-topology in
\widetilde C_\omega ^1({I_1})
. Denote by
\widetilde B_{1,{\kern 1pt} \varepsilon }^1(f)
elements of the base of the C1-topology in
\widetilde C_\omega ^1({I_1})
for every ɛ > 0 and
f \in \widetilde C_\omega ^1({I_1})
.
By the C1- Ω-stability of the map f (see the condition (iif)) for any δ > 0 there exists an ɛ-neighborhood
\widetilde B_{1,{\kern 1pt} \varepsilon }^1(f)
of f in the space
\widetilde C_\omega ^1({I_1})
such that every map from this neighborhood is Ω-conjugate with f by means of a homeomorphism which is δ-close in the C0-topology of the uniform convergence to the identity map of the nonwandering set
A point x ∈ I1 ((x, y) ∈ I) is said to be f-nonwandering (F-nonwandering) point if for every its neighborhood U1(x) (U((x, y)) = U1(x) × U2(y)) there is a natural number n such that the inequality U1(x)∩ f n(U1(x)) ≠ ∅ (U((x, y)) ∩ Fn(U(x, y))) ≠ ∅ holds. The set of all f-nonwandering (F-nonwandering) points is said to be the nonwandering set of f (F) [22]. We use the notation Ω(f ) (Ω(F)) for this set.
of the map f.
Let
{\widetilde C^1}(I,{\kern 1pt} {I_1})
be the set of C1-smooth maps of the rectangle I into the interval I1 endowed with the standard C1-norm || · ||1,(1,1) of the linear normalized space of C1-smooth maps of the rectangle I into the straight line R1 that contains the interval I1. This norm induces the C1-topology in the space
{\widetilde C^1}(I,{\kern 1pt} {I_1})
with the base given by the set of ɛ-balls
\widetilde B_{(1,{\kern 1pt} 1),{\kern 1pt} \varepsilon }^1(\varphi )
for every
\varphi \in {\widetilde C^1}(I,{\kern 1pt} {I_1})
and ɛ > 0.
We suppose that the function μ = μ(x, y) satisfies the following “condition of smallness”:
(iiμ) ||μ||1,(1,1) < ɛ, where ɛ is found for δ > 0 by the property of the C1- Ω-stability of f.
The following inequality connects norms || · ||1,1 for every y ∈ I2 and || · ||1,(1,1):
||\mu {||_{1,{\kern 1pt} 1}} < ||\mu {||_{1,{\kern 1pt} (1,{\kern 1pt} 1)}}.
Every function f of one variable can be considered as the function of two variables of the type f ○ pr1, where pr1 : I → I1 is the natural projection of I on I1. Hence, by the condition (iiμ) and the inequality (2) we have
(f + \mu ) \in \widetilde B_{(1,{\kern 1pt} 1),{\kern 1pt} \varepsilon }^1(f \circ p{r_1})
, and the belonging
(f + \mu ) \in \widetilde B_{1,{\kern 1pt} \varepsilon }^1(f)
holds for every y ∈ I2.
Denote by
C_\omega ^1(I)
the set of C1-smooth maps (1) such that the function f satisfies conditions (if) – (iif), and μ satisfies conditions (iμ) – (iiμ). Endow
C_\omega ^1(I)
with the standard C1-norm || · ||1 of the linear normalized space of C1-smooth maps of the rectangle I into the plane R2. The base of the C1-topology generated by this norm, is given by the system of ɛ-balls
B_\varepsilon ^1(F) = \{ G \in C_\omega ^1(I):||G - F{||_1} < \varepsilon \}
with the center F for every ɛ > 0 and
F \in C_\omega ^1(I)
.
3. The map (1) from the space
C_\omega ^1(I)
is obtained by small C1-smooth perturbations (satisfying conditions (iμ) – (iiμ)) of the C1-smooth skew product of interval maps of the type
{\Phi _0}(x,{\kern 1pt} y) = (f(x),{\kern 1pt} {g_x}(y)).
In [18] it is shown also that the autonomous discrete dynamical system (1) is connected with the nonautonomous discrete dynamical system generated by skew products of interval maps. So, one can represent the value of n-th (n ≥ 1) iteration of the map F in every initial point (x0, y0) ∈ I as the composition of values of various skew products of interval maps in the corresponding points:
{F^n}({x^0},{\kern 1pt} {y^0}) = ({x^n},{\kern 1pt} {y^n}) = {\Phi _{{y^{n - 1}}}} \circ \ldots \circ {\Phi _{{y^0}}}({x^0},{\kern 1pt} {y^0}),
where a skew product Φyi : I → I for every i (0 ≤ i ≤ n − 1) is presented in the form
{\Phi _{{y^i}}}(x,{\kern 1pt} y) = ({\varphi _{{y^i}}}(x),{\kern 1pt} {g_x}(y)).
Here
{\varphi _{{y^i}}}(x) = f(x) + {\mu _{{y^i}}}(x),{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\rm and}{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\mu _{{y^i}}}(x) = \mu (x,{\kern 1pt} {y^i}).
To write the equality (5) in the coordinate form we set
{\varphi _{{y^0},{\kern 1pt} n}}({x^0}) = {\varphi _{{y^{n - 1}}}} \circ \ldots \circ {\varphi _{{y^0}}}({x^0});{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {g_{({x^0},{\kern 1pt} {y^0}),{\kern 1pt} n}}({y^0}) = {g_{{\varphi _{{y^{n - 1}}}}({x_{n - 1}})}} \circ \ldots \circ {g_{{x^0}}}({y^0}).
Then we have:
{F^n}({x^0},{\kern 1pt} {y^0}) = ({\varphi _{{y^0},{\kern 1pt} n}}({x^0}),{\kern 1pt} {\kern 1pt} {g_{({x^0},{\kern 1pt} {y^0}),{\kern 1pt} n}}({y^0})).
All previous information of the item 3 means that an important role in this paper belongs to skew products of interval maps.
Denote by
T_{\omega ,{\kern 1pt} 0}^1(I)
the space of C1-smooth skew products of interval maps with quotients satisfying conditions (if) – (iif). Endow this space with the C1-topology generated by the standard C1-norm || · ||1. The structure of the functional space
T_{\omega ,{\kern 1pt} 0}^1(I)
and dynamical properties of skew products from this space are studied in [9] – [16].
We use also the space
\widetilde T_{\omega ,{\kern 1pt} 0}^1(I)
of skew products of interval maps respectively which the boundary ∂ I of the rectangle I is invariant. Then the inclusions are valid:
\widetilde T_{\omega ,{\kern 1pt} 0}^1(I) \subset T_{\omega ,{\kern 1pt} 0}^1(I) \subset C_\omega ^1(I).
The base of the C1-topology in
\widetilde T_{\omega ,{\kern 1pt} 0}^1(I)
is given by the set of ɛ-balls
\widetilde B_\varepsilon ^1(\Phi )
with the center Φ for all ɛ > 0 and
\Phi \in \widetilde T_{\omega ,{\kern 1pt} 0}^1(I)
.
Note that by the formula (1) and the condition (iμ) every map
F \in C_\omega ^1(I)
obtained from the skew products of interval maps
{\Phi _0} \in \widetilde T_{\omega ,{\kern 1pt} 0}^1(I)
possesses the property:
F(\partial I) \subset \partial I.
4. There is a vast literature devoted to different integrability aspects of dynamical systems both with continuous time (see, e.g., [7], [23] – [24]), and with discrete time (see, e.g., [1] – [2], [35] – [36]). Originally, the concept of integrability of dynamical systems with discrete time was introduced for systems obtained by digitization of known differential equations [1] – [2], [35] – [36]). But there are discrete dynamical systems that do not belong to this class. We consider here precisely this case.
Remind the following Birkhoff's thought: “If we try to formulate the exact definition of integrability then we see that many definitions are possible, and every of them is of the specific theoretical interest” [6].
Our definition of integrability of dynamical systems with discrete time given in [4] (see also [16]) follows the paper [20] and generalizes the definition from [20] given for polynomials and rational maps, on the case of arbitrary maps. (The last set contains maps that can not be obtained by the procedure of digitization of differential equations.)
Definition 1
[4] We say that a map G of some (open or closed) domain Π ⊂ R2 to itself is integrable if there exists a self-map ψ of an interval J of the real line R1 such that G is semiconjugate with ψ by means of a continuous surjection
\widetilde H:\Pi \to J
, so that
\widetilde H \circ G = \psi \circ \widetilde H.
Remark 1
In the framework of the suggested approach in the paper [17] the definition of integrability is introduced for some multifunctions.
Remark 2
As it follows from Definition 1 skew products of interval maps are integrable maps. Here
\widetilde H = {pr_1}
. Moreover, every integrable map satisfying some natural conditions can be reduced to a skew product
The reducibility problem of integrable maps to skew products has been formulated by Grigorchuk to the Author during our verbal discussions (the formulation of the problem is not published) in the framework of the Conference devoted to the 70-th birthday of Professor V.M. Alexeev (Moscow, Russia, 2002).
.
Theorem 1
[4]. Let Π be a convex connected compact subset ofR2such that the section of Π by an arbitrary line y = const (if it is non-empty) is a non-degenerate interval, and let G : Π → Π be a continuous map. Then G is integrable in the sense of Definition 1 by means of a continuous surjection\widetilde H:\Pi \to JJ that is one-to-one with respect to x (here J is a closed interval ofR1) if and only if some homeomorphism reduces G to a skew product of interval maps defined in a compact planar rectangle.
Remark 3
Definition 1 distinguishes such feature of integrable dynamical systems satisfying conditions of Theorem 1, as the existence of an invariant foliation. This property is the key point of the proof of the integrability property of a dynamical system.
Remark 4
Point out that the existence of a continuous invariant foliation for Lorenz type maps is proven in [3], and existence of a C1-smooth invariant foliation (with C2-smooth fibers) for these maps is proven in [34].
In different problems of dynamical systems theory only existence of an invariant lamination (but not an invariant foliation!) can be proved (see, e.g., [4], [18]). Therefore, it is naturally to introduce the following concept of the partial integrability for discrete dynamical systems.
Definition 2
We say that a map G defined on some (open or closed) domain Π of the plane R2 with values in Π is partially integrable if there exist a closed invariant set A ⊂ Π (A ≠ Π), a self-map ψ of an interval J of the real line R1 and a closed invariant set B ⊂ J (B ≠ J) such that G|A is semiconjugate with ψ|B by means of a continuous surjection
\widetilde H:A \to B
, i. e. the equality holds
\widetilde H \circ {G_{|A}} = {\psi _{|B}} \circ \widetilde H.
5. In this paper we investigate stability of the integrability property of skew products of interval maps under small C1-smooth perturbations satisfying conditions (iμ) – (iiμ). We obtain here (sufficient) conditions of partial integrability for maps from the space
C_\omega ^1(I)
(§3). These conditions are formulated in the terms of properties of the unperturbed skew product
{\Phi _0} \in T_{\omega ,{\kern 1pt} 0}^1(I)
(see the formula (4)). We give also the example of the partially integrable map (1) (§3).
Preliminaries
This section contains the relevant definitions and results on dynamics both of continuous maps and C1-smooth Ω-stable maps of a closed interval.
1. We begin from the famous Sharkovsky's Theorem [32].
Theorem 2
[32] If a continuous map f : I1 → I1contains a periodic orbit of a (least) period m > 1 then it contains also periodic orbits of every (least) period n, where n precedes m (n ≺ m) in the Sharkovsky's order:\matrix{{1 \prec 2 \prec {2^2} \prec {2^3} \prec \ldots \prec \ldots \prec {2^2} \cdot 9 \prec {2^2} \cdot 7 \prec {2^2} \cdot 5 \prec {2^2} \cdot 3 \prec \ldots } \cr { \prec 2 \cdot 9 \prec 2 \cdot 7 \prec 2 \cdot 5 \prec 2 \cdot 3 \prec \ldots \prec 9 \prec 7 \prec 5 \prec 3.} \cr }
In accordance with the Sharkovsky's Theorem, the space of continuous self-maps of a closed interval can be presented as the union of three subspaces (see, e.g., [33]): the first of which consists of the maps of type ≺ 2∞, that is, the maps that have periodic points with (least) periods {1,2,22,...,2ν}, where 0 ≤ ν < +∞; the second subspace consists of the maps of type 2∞, that is, maps whose periodic points have (least) periods {1,2,22,...,2i,2i+1,...}; the third subspace consists of the maps of type ≻ 2∞, that is, maps with periodic points that possess (least) periods outside the set {2i}i ≥0.
In this paper we consider maps from the space
C_\omega ^1(I)
such that f satisfies the following additional condition: (iiif) f has type ≻ 2∞.
The above condition (iiif) means that f demonstrates a chaotic behavior (see, e.g., [33]).
2. Formulate the properties of C1-smooth Ω-stable maps of a closed interval, give the definition of Σ-stability and remind the properties of the Σ-stable maps of a closed interval.
Lemma 3
[21], [27] Letf \in \widetilde C_\omega ^1({I_1})and satisfy the condition (iiif). Then
(3.1) the nonwandering set Ω(f ) is the union of a finite number of hyperbolic periodic points (that form the rarefied set Ωr(f )) and a finite number of locally maximal (i.e., maximal quasiminimal
A quasiminimal set is the closure of an infinite recurrent trajectory (see [28]).
sets in some their neighborhood) hyperbolic perfect nowhere dense sets (that form the perfect set Ωp(f )));
(3.2) periodic points are everywhere dense in the set Ωp(f ); moreover, for every natural number m ≥ 2 periodic points with multiple m (least) periods are everywhere dense in Ωp(f );
(3.3) there are numbers α = α(f ) > 0 and c = c(f ) > 1 so that for every x ∈ Ωp(f ) and n ≥ 1 the inequality |(f n(x))′| > αcn holds (that is, Ωp(f ) is the repelling hyperbolic set);
(3.4) the subspace\widetilde C_\omega ^1({I_1})of maps satisfying condition (iiif) is open and everywhere dense in the containing it space of C1-smooth self-maps of the closed interval I1of type ≻ 2∞with the invariant boundary.
Corollary 4
[21], [27] Letf \in \widetilde C_\omega ^1({I_1})and satisfy the condition (iiif). Then the equality holds:\Omega (f) = {\Omega ^s}(f)\bigcup {\Omega ^u}(f).Here Ωs(f ) is the nonempty finite invariant set of all f-sinks. The set Ωu(f ) is invariant and equals the union of Ωp(f ) with a finite (possibly empty) set that consists of isolated sources in the set of f-periodic points.
In the set I1 \ Ω(f ) (just as in the set I1 \ Ωp(f )) the points of attraction domains of f-sinks are everywhere dense.
Point out that for the map
f \in \widetilde C_\omega ^1({I_1})
the inclusion f−1(Ω(f )) ⊂ Ω(f ) can be false (although the equality f (Ω(f )) = Ω(f ) holds). Here f−1 (·) means the first complete preimage of a set.
Following [21] we construct the set which is invariant both with respect to f, and with respect to f−1, and contains the set Ωu(f) as its subset. For this goal we need the attraction domain of all f-sinks:
\Delta (f) = \bigcup\limits_{Orb(x,{\kern 1pt} f) \subset {\Omega ^s}(f)} \bigcup\limits_{i = 0}^{ + \infty } {D^{ - i}}(Orb(x,{\kern 1pt} f)),
where Orb(x, f ) is the periodic orbit of the sink x ∈ Ωs(f ), D(Orb(x, f )) is the immediate attraction domain of the periodic orbit Orb(x, f ), D−i(Orb(x, f )) is i-th complete primage of the immediate attraction domain of the periodic orbit Orb(x, f ).
Immediate attraction domain D(Orb(x, f )) of the periodic orbit Orb(x, f ) consists of m (m is the (least) period of x) pairwise disjoint open (in the topology of the segment I1) f m-invariant intervals Df j(x) (0 ≤ j ≤ m − 1) such that every of these intervals contains the point f j(x) :
D(Orb(x,{\kern 1pt} f)) = \bigcup\limits_{j = 0}^{m - 1} {D_{{f^j}(x)}}.
Complete f-invariance of the immediate attraction domain
It means correctness of the equality f (D(Orb(x, f ))) = D(Orb(x, f )).
D(Orb(x, f )) implies correctness of the definition of preimages D−i(Orb(x, f )) (D−i(Orb(x, f )) ≠ ∅) for every i ≥ 0.
By the condition (iiif) the set Δ(f ) is a countable union (see the claim (3.1) of Lemma 3) of pairwise disjoint intervals (open in the topology of the closed interval I1); Δ(f ) is invariant both with respect to f , and with respect to f−1.
We suppose further that the set Cr(f ) of f-critical points satisfies the condition
(ivf) Cr(f ) ⊂ Δ° (f ), where Δ° (f) is the interior of the set Δ (f).
Lemma 5
[33] Letf \in \widetilde C_\omega ^1({I_1})satisfy conditions (iiif) – (ivf). Then one of the following three cases is realized for the boundary ∂ (Df j(x)) of every interval Df j(x) (0 ≤ j ≤ m − 1):
(5.1) ∂ (Df j(x)) consists of two f m-fixed points;
(5.2) points of ∂ (Df j(x)) form a periodic orbit of (least) period 2 with respect to f m;
(5.3) one of the points of ∂ (Df j(x)) is f m-fixed point source, and the other is its preimage with respect to f m.
Define the closed set that is invariant with respect to f and f−1:
\Sigma (f) = {I_1}\backslash \Delta (f).
By Lemma 5 and formula (12) we have: Ωu(f ) ⊂ Σ(f), and Ωs(f) ∩ Σ (f) = ∅.
Definition 3
[21] The map
f \in \widetilde C_\omega ^1({I_1})
is said to be Σ-stable (in the C1-topology) if for every δ > 0 there exists an ɛ-neighborhood
\widetilde B_{1,{\kern 1pt} \varepsilon }^1(f)
of the map
f \in \widetilde C_\omega ^1({I_1})
such that every map
\varphi \in \widetilde B_{1,{\kern 1pt} \varepsilon }^1(f)
is Σ-conjugate to f, that is, the equality
h \circ {f_{|\Sigma (f)}} = {\varphi _{|\Sigma (\varphi )}} \circ h
holds for some homeomorphism h : Σ(f ) → Σ(ϕ). Here h is δ-close to the identity map on Σ(f ) in the C0-topology.
The following claim is the direct corollary of Definition 3.
Lemma 6
Letf \in \widetilde C_\omega ^1({I_1})satisfy conditions (iiif) – (ivf). Then f is Σ-stable in the C1-topology.
Remark 5
By [21] the set of maps from
\widetilde C_\omega ^1({I_1})
satisfying conditions (iiif) – (ivf) contains the open everywhere dense subset of C2-smooth maps with nondegenerate critical points.
Sufficient conditions of partial integrability of the map (1)
In this section we prove the main result of the paper.
Theorem 7
Let the quotient f of the skew product of interval maps{\Phi _0} \in \widetilde T_{\omega ,{\kern 1pt} 0}^1(I)satisfy conditions (iiif) – (ivf). Then for any δ > 0 there exists an ɛ-neighborhoodB_\varepsilon ^1({\Phi _0})of the map Φ0in the spaceC_\omega ^1(I)such that every mapF \in B_\varepsilon ^1({\Phi _0})obtained from Φ0by means of the C1-perturbation μ = μ(x, y), where μ satisfies conditions (iμ) – (iiμ), is partially integrable on the closed invariant set £(F) that consists of pairwise disjoint curvelinear fibers. These fibers start from the points of the set Σ* (f) × {a2} (where Σ* (f ) = Σ(f ) ∪ Ωs(f)) and are graphs of C1-smooth functions x = x(y) defined on the interval I2. The function\widetilde H:{\rm{\pounds}}(F) \to {\Sigma ^*}(f)that realizes the partial integrability property, is C1-smooth surjection, ɛ-close in the C1-norm to the natural projection pr1 : Σ* (f ) × I2 → Σ*(f).
The following statement proved in [18], is the first step of the proof of Theorem 7.
Proposition 8
[18] Let{\Phi _0} \in \widetilde T_{\omega ,{\kern 1pt} 0}^1(I)
. Then for any δ > 0 there exists an ɛ-neighborhoodB_\varepsilon ^1({\Phi _0})of the map Φ0in the spaceC_\omega ^1(I)such that every mapF \in B_\varepsilon ^1({\Phi _0})obtained from Φ0by means of the C1-perturbation μ = μ(x, y), where μ satisfies conditions (iμ) – (iiμ), has the invariant closed set £0(F) that consists of pairwise disjoint curvelinear fibers. These fibers start from the points of the set Ω(f ) × {a2}, and are the graphs of continuous functions x = x(y) defined on the interval I2; moreover, fibers starting from the points (x, a2), where x is f-periodic point, are C1-smooth. Every curvelinear fiber is δ-close in the C0-norm to the vertical closed interval that starts from the same initial point of the set Ω(f ) × {a2} just as the curvelinear fiber.
Prove C1-smoothness of all fibers from the set £0(F).
Proposition 9
Let{\Phi _0} \in \widetilde T_{\omega ,{\kern 1pt} 0}^1(I)
. Then for any δ > 0 there exists an ɛ-neighborhoodB_\varepsilon ^1({\Phi _0})of the map Φ0in the spaceC_\omega ^1(I)such that the closed invariant set £0(F) of every mapF \in B_\varepsilon ^1({\Phi _0})obtained from Φ0by means of the C1-perturbation μ = μ(x, y), where μ satisfies conditions (iμ) – (iiμ), consists of graphs of C1-smooth functions x = x(y) defined on the interval I2. In addition, every curvelinear fiber of the set £0(F) is ɛ-close in the C1-norm to the vertical closed interval that starts from the same initial point of the set Ω(f ) × {a2} just as the curvelinear fiber.
Proof
1. Fix a number δ > 0. We find a positive number ɛ > 0 for δ using the C1- Ω-stability property of the map
f \in \widetilde C_\omega ^1({I_1})
. The neighborhood
\widetilde B_{1,{\kern 1pt} \varepsilon }^1(f)
of the map f consists of maps such that every map is Ω-conjugate with f by means of the homeomorphism that is δ-close to the identity map of the set Ω(f ).
We use also the ɛ-neighborhood
B_\varepsilon ^1({\Phi _0})
of the map Φ0 in the space
C_\omega ^1(I)
. Then by formulas (1), (4) and by the property (iiμ) the inequality
||F - {\Phi _0}{||_1} = ||\mu {||_{1,{\kern 1pt} (1,{\kern 1pt} 1)}} < \varepsilon
is valid for any map
F \in B_\varepsilon ^1({\Phi _0})
. It implies, in particular, correctness of the belonging (3) for any y ∈ I2 and means also that the map (f + μ) is Ω-conjugate with f for every y ∈ I2 by means of the homeomorphism that is δ-close to the identity map of the set Ω(f ).
We need ɛm-neighborhoods
B_{{\varepsilon _m}}^1(\Phi _0^m)
of iterations
\Phi _0^m
for any m > 1 in the space
C_\omega ^1(I)
that correspond the chosen ɛ-neighborhood
B_\varepsilon ^1({\Phi _0})
of the map Φ0. Since
F \in B_\varepsilon ^1({\Phi _0})
then
{F^m} \in B_{{\varepsilon _m}}^1(\Phi _0^m)
. Using formulas (8) – (9) we obtain the inequalities
||{\varphi _{y,{\kern 1pt} m}} - {f^m}{||_{1,{\kern 1pt} (1,{\kern 1pt} 1)}} \le ||{F^m} - \Phi _0^m{||_1} < {\varepsilon _m};
moreover, for every y ∈ I2 the inequality holds:
||{\varphi _{y,{\kern 1pt} m}} - {f^m}{||_{1,{\kern 1pt} 1}} < ||{\varphi _{y,{\kern 1pt} m}} - {f^m}{||_{1,{\kern 1pt} (1,{\kern 1pt} 1)}}.
Therefore,
{\varphi _{y,{\kern 1pt} m}} \in \widetilde B_{(1,{\kern 1pt} 1),{\kern 1pt} {\varepsilon _m}}^1({f^m} \circ p{r_1}){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\rm and},\;{\rm the}\;{\rm more}\;{\rm so},{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\varphi _{y,{\kern 1pt} m}} \in \widetilde B_{1,{\kern 1pt} {\varepsilon _m}}^1({f^m})
for every y ∈ I2. Here
\widetilde B_{(1,{\kern 1pt} 1),{\kern 1pt} {\varepsilon _m}}^1({f^m} \circ p{r_1})
is the ɛm-neighborhood of the map (fm ○ pr1) in the space
{\widetilde C^1}(I,{\kern 1pt} {I_1})
and
\widetilde B_{1,{\kern 1pt} {\varepsilon _m}}^1({f^m})
is the ɛm-neighborhood of the map f m in the space
\widetilde C_\omega ^1({I_1})
. The neighborhood
\widetilde B_{1,{\kern 1pt} {\varepsilon _m}}^1({f^m})
consists of the maps, which are Ω-conjugate to f m by means of the homeomorphisms δ-close to the identity map on the nonwandering set Ω(f m) = Ω(f ) (see Lemma 3).
2. Let f satisfy the condition (iiif). Denote by £u(F) (£u(F) ⊂ £0(F)) the set of curvelinear fibers that start from all points of Ωu(f ).
Prove that there is the universal natural number n* such that the equality
\mathop {\inf }\limits_{(x,{\kern 1pt} y) \in {{\rm{\pounds}}^u}(F)} |{\partial \over {\partial x}}{\varphi _{y,{\kern 1pt} {n_*}}}(x)| = {M_*},{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\rm where}{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {M_*} > 1,
holds for every y ∈ I2.
In fact, using the definition of the set £u(F) and Lemma 3 (see claims (3.1) – (3.3)) for every y ∈ I2 we point out the least natural number n* (y) satisfying
\mathop {\inf }\limits_{x \in ({{\rm{\pounds}}^u}(F))(y)} |{\partial \over {\partial x}}{\varphi _{y,{\kern 1pt} {n_*}(y)}}(x)| > 1.
By the claim (3.3) of Lemma 3 the inequality
\mathop {\inf }\limits_{x \in ({{\rm{\pounds}}^u}(F))(y)} |{\partial \over {\partial x}}{\varphi _{y,{\kern 1pt} n}}(x)| > 1
is valid for every n ≥ n* (y).
Since the partial derivative
{\partial \over {\partial x}}{\varphi _{y,{\kern 1pt} {n_*}(y)}}(x)
is uniformly continuous on the compact
({{\rm{\pounds}}^u}(F))(y) \times \{ y\}
then by the inequality (15) there exists a θ (y)-neighborhood Uθ(y) (£u(F))(y)×{y}) of the set (£u(F))(y)×{y} in I such that
\mathop {\inf }\limits_{(x,{\kern 1pt} y') \in {U_{\theta (y)}}(({{\rm{\pounds}}^u}(F))(y) \times \{ y\} )} |{\partial \over {\partial x}}{\varphi _{y',{\kern 1pt} {n_*}(y)}}(x)| > 1.
Moreover, by the formula (13)n* (y) is the least natural number for which the inequality (16) holds.
Let 2I1 be the topological space of all closed subsets of the closed interval I1 with the exponential topology. By compactness of the closed intervals I1, I2 the set 2I1 × I2 is the compact [24]. Then the closed set
{{\rm{\pounds}}^u}(F) = \bigcup\limits_{y \in {I_2}} (({{\rm{\pounds}}^u}(F))(y) \times \{ y\} )
is the compact in 2I1 × I2. Using compactness of the set £u(F) in 2I1× I2 we distinguish from its infinite open cover {Uθ(y) ((£u(F))(y) × {y})}y∈I2 the finite subcover. Let neighborhoods {Uθ(yj) ((£u(F))(yj) × {yj})}1≤j≤q of the sets {(£u(F))(yj) × {yj}}1≤ j≤q form this finite subcover.
The set £u(F) consists of continuous fibers (see Proposition 8). Therefore, for every j (1 ≤ j ≤ q) there exists j′ (j′ ≠ j, 1 ≤ j′ ≤ q) such that
{U_{\theta ({y_j})}}(({{\rm{\pounds}}^u}(F))({y_j}) \times \{ {y_j}\} )\bigcap {U_{\theta ({y_{j'}})}}(({{\rm{\pounds}}^u}(F))({y_{j'}}) \times \{ {y_{j'}}\} ) \ne \emptyset .
Thus, using the above considerations of this item 2 we obtain from here that the equalities hold
{n_*}({y_1}) = \ldots = {n_*}({y_j}) = \ldots = {n_*}({y_q}).
Set n* = n* (yj) (1 ≤ j ≤ q). Using continuity of the partial derivative
{\partial \over {\partial x}}{\varphi _{y,{\kern 1pt} {n_*}}}(x)
on I we verify that the equality (14) holds.
Hence, without loss of generality we will suppose further that n* = 1. In fact, if n* ≠ 1 then we get over consideration of the map Fn and use the claim (3.2) of Lemma 3.
3. Prove that every curvelinear fiber is ɛ-close in the C1-norm to the vertical closed interval that starts from the same initial point (x, a2) of the set Ω(f ) × {a2} just as the curvelinear fiber.
3.1. Begin from the curvelinear fibers γx0 that start from all points (x0, a2), where x0 ∈ Per(f ) ∩ Ωu(f ). By Proposition 8 every this fiber is the graph of the C1-smooth implicit function x = x(y) defined on the interval I2. Moreover, x = x(y) satisfies the initial conditions
x({a_2}) = x({b_2}) = {x^0}
and the equation
{\varphi _{y,{\kern 1pt} m}}(x) = x{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} [18]{\kern 1pt} ,
where m is the least period of the initial f-periodic point x0. By the theorem about the C1-smooth implicit function and the equation (17) we have
{d \over {dy}}x(y) = - {{{\partial \over {\partial y}}{\varphi _{y,{\kern 1pt} m}}(x)} \over {{\partial \over {\partial x}}{\varphi _{y,{\kern 1pt} m}}(x) - 1}}
in any point (x, y) ∈ γx0.
Note that by the item 2 the sequence {ɛm}m≥1 is increasing. Therefore, we construct the special δ-trajectory for the real trajectory {Fj(x, y)}j≥0. Denote by lx the vertical closed interval that starts from the point (x, a2).
Then (xrm+i, yrm+i) ∈ lxi. Since
{F^{rm + i}}(x,{\kern 1pt} y) = ({\varphi _{y,{\kern 1pt} rm + i}}(x),{\kern 1pt} {\kern 1pt} {g_{(x,{\kern 1pt} y),rm + i}}(y)),{\kern 1pt} {\rm and}{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {F^{rm + i}}(x,{\kern 1pt} y) \in {\gamma _{{x^i}}},
then {zrm+i}r≥0,0≤i≤m−1 is δ-trajectory for the real trajectory of the point (x, y). Moreover,
|{f^j}({x^0}) - {\varphi _{y,{\kern 1pt} j}}(x)| < \delta .
It implies, in particular, correctness of the inequality
|{\partial \over {\partial y}}{\varphi _{y,{\kern 1pt} j}}(x)| < \varepsilon
for every j ≥ 1. Therefore, using (18) we have
|{d \over {dy}}x(y)| < {\varepsilon \over {M_*^m - 1}}.
Since M* > 1 then the equality holds:
\mathop {\lim }\limits_{m \to + \infty } M_*^m = + \infty .
Then there exists m0 ≥ 1 such that the inequality
|{d \over {dy}}x(y)| < \varepsilon {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} ({\rm see}\;{\rm the}\;{\rm inequality}\, (19)){\kern 1pt}
is valid for every m ≥ m0.
3.2. As it follows from the item 3.1, the set of C1-smooth functions {x = x(y)} that start from the points {(x0, a2)}, where x0 ∈ Per(f ) ∩ Ωp(f ), m(x0) ≥ m0 (here m(x0) is the (least) period of x0), is dense in itself in the C1-topology. It implies, in particular, that for every C1-smooth function x = x(y) that starts from a point (x0, a2) for x0 ∈ Per(f ) ∩ Ωp(f ), m(x0) < m0, the inequality holds
|{d \over {dy}}x(y)| \le \varepsilon .
In addition, for every nonperiodic point x0 ∈ Ω(f ) there exists the unique C1-smooth function x = x(y) : I2 → I1 with the graph that starts from the point (x0, a2) and with the derivative satisfying the inequality (21) (see the inequality (20)). Proposition 9 is proved.
Extend the lamination £0(F) up to the lamination £(F), where £(F) consists of C1-smooth fibers that start from the points of the set Σ (f ) × {a2}.
Use conditions (iiif) – (ivf), definition of the set Δ(f ), Lemma 5 and Proposition 9. Then we obtain the following statement.
Corollary 10
Let the quotient f of the skew product of interval maps{\Phi _0} \in \widetilde T_{\omega ,{\kern 1pt} 0}^1(I)satisfy conditions (iiif) – (ivf). Then for any δ > 0 there exists an ɛ-neighborhoodB_\varepsilon ^1({\Phi _0})of the map Φ0in the spaceC_\omega ^1(I)such that every mapF \in B_\varepsilon ^1({\Phi _0})obtained from Φ0by means of the C1-perturbation μ = μ(x, y), where μ satisfies conditions (iμ) – (iiμ), has the closed invariant set £(F) that consists of pairwise disjoint curvelinear fibers. These fibers start from the points of the set Σ* (f ) × {a2}, and are graphs of C1-smooth functions x = x(y) defined on the interval I2. Every curvelinear fiber of the set £(F) is ɛ-close in the C1-norm to the vertical closed interval that starts from the same initial point of the set Σ* (f ) × {a2} just as the curvelinear fiber.
Remark 6
The set £(F) constructed in the Corollary 10 is the C1-smooth lamination, i.e. the set
{\rm{\pounds}}(F) = \{ {x^0},{\kern 1pt} x,{\kern 1pt} {\kern 1pt} y\} ,{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\rm where}{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {x^0} \in {\Sigma ^*}(f),{\kern 1pt} {\kern 1pt} x = x(y),
depends C1-smoothly on the variables x0, x, y.
Let conditions of Theorem 7 be fulfilled, and
F \in B_\varepsilon ^1({\Phi _0})
be given by the formula (1), where μ depends on x and y. Let γx0 be the curvelinear fiber from £(F) that starts from an arbitrary point (x0, a2), where x0 ∈ Σ* (f ). We set
\widetilde H(x,{\kern 1pt} y) = {x^0}
for any point (x, y) ∈ γx0, that is
\widetilde H({\gamma _{{x^0}}}) = {x^0}
, and
\widetilde H({\rm{\pounds}}(F)) = {\Sigma ^*}(f)
. It means that the curvelinear projection
\widetilde H
is the surjection of the set £(F) on the set Σ* (f).
By the equality (22) we have
\widetilde H(x,{\kern 1pt} y) = p{r_1}(x,{\kern 1pt} y) - {x_{{x^0}}}(y) + {x^0},
where x = xx0 (y) is the function with the graph γx0.
Remark 7
As it follows from the equality (23) and Corollary 10, the curvelinear projection
\widetilde H
is C1-smooth on the lamination £(F). Moreover, surjection
\widetilde H:{\rm{\pounds}}(F) \to {\Sigma ^*}(f)
is ɛ-close in the C1-norm to the natural projection pr1 : Σ (f ) × I2 → Σ* (f ).
Remark 8
F-invariance of the lamination £(F) and the formula (22) imply the equality
\widetilde H \circ {F_{|{\rm{\pounds}}(F)}} = {f_{|{\Sigma ^*}(f)}} \circ \widetilde H.
Comparison of the equalities (24) and (10) shoes that F is partially integrable map (see Definition 2). It completes the proof of Theorem 7.
In the end of the paper we give the example of the partially integrable map.
The condition “
F \in C_\omega ^1([0,{\kern 1pt} {1]^2})
” implies “
f \in \widetilde C_\omega ^1([0,{\kern 1pt} {\kern 1pt} 1])
”. We use here the model C1-smooth Ω-stable map f of type ≻ 2∞ from the paper [16]:
f(x) = \left\{ {\matrix{{\widetilde h(x),} \hfill & {{\kern 1pt} {\rm if}{\kern 1pt} } \hfill & {x \in [0,{\kern 1pt} {1 \over 4});} \hfill \cr {9({1 \over 4} - x)(x - {3 \over 4}) + {1 \over 4},} \hfill & {{\kern 1pt} {\rm if}{\kern 1pt} } \hfill & {x \in [{1 \over 4},{\kern 1pt} {3 \over 4});} \hfill \cr {\widetilde h(1 - x),} \hfill & {{\kern 1pt} {\rm if}{\kern 1pt} } \hfill & {x \in [{3 \over 4},{\kern 1pt} 1];} \hfill \cr } } \right.
where
\widetilde h
is so that f (0) = f (1) = 0, f : [0, 1/4] → [0, 1/4] is increasing bijection, f : [3/4, 1] → [0, 1/4] is decreasing bijection. Let M = sup{ f (x)} satisfy the inequality 3/4 < M < 1, and the point xM such that the equality f (xM) = M holds, be the unique. Then the equality is valid:
\Omega (f) = \{ 0\} \bigcup K(f),
where K(f ) is the unique locally maximal quasiminimal set of f,
K(f) = {\Omega _p}(f) \subset [{1 \over 4},{\kern 1pt} {3 \over 4}]
and xM ∈ Δ° (f ).
Let λ be so small that the function μ(x, y) = λ x(1 − x)y(1 − y) satisfies the condition (iiμ). We have also μ(0, y) = μ(1, y) = μ(x, 0) = μ(x, 1) = 0 for all x, y ∈ [0, 1]. Hence, the condition (iμ) is valid. It means that conditions of Theorem 7 are fulfilled, and there exists the invariant lamination £(F) over the points of the set Ω(f ) = {0} ∪ K(f ). It implies the semiconjugacy of F|£(F) and f|Ω(f ). Therefore, F is the partially integrable map (see Definition 2).