Sternberg Linearization Theorem for Skew Products


We present a special kind of normalization theorem: linearization theorem for skew products. The normal form is a skew product again, with the fiber maps linear. It appears that even in the smooth case, the conjugacy is only Hölder continuous with respect to the base. The normalization theorem mentioned above may be applied to perturbations of skew products and to the study of new persistent properties of attractors.

This is a preview of subscription content, log in to check access.


  1. 1.

    Gorodetskii AS, Ilyashenko YS. Some new robust properties of invariant sets and attractors of dynamical systems. Funct Anal Appl 1999;33:16–30 (Russian).

    MathSciNet  Article  Google Scholar 

  2. 2.

    Gorodetskii AS, Ilyashenko YS. Funct Anal Appl. 33:95–105 (English translation).

  3. 3.

    Gorodetskii AS, Ilyashenko YS. Some properties of skew products over a horseshoe and a solenoid. Tr Mat Inst Steklova 2000;231:96–118 (Russian).

    MathSciNet  Google Scholar 

  4. 4.

    Gorodetskii AS, Ilyashenko YS. Din Sist Avtom i Beskon Gruppy 2000:96–118.

  5. 5.

    Gorodetskii AS, Ilyashenko YS. Proc Steklov Inst Math 2000;231:90–112 (Engl.transl.).

  6. 6.

    Gorodetski A, Ilyashenko Y, Kleptsyn V, Nalski M. Non-removable zero Lyapunov exponents. Funct Anal Appl 2005;39(1):27–38.

    MathSciNet  Article  Google Scholar 

  7. 7.

    Gorodetskii AS. Regularity of central leaves of partially hyperbolic sets and applications. (Russian) Izv. Ross. Akad. Nauk Ser. Mat. 2006;70(6):19–44. translation in Izv. Math. 70 (2006), no. 6, 1093–1116.

    MathSciNet  Article  Google Scholar 

  8. 8.

    Guysinsky. The theory of non-stationary normal forms. Ergodic Theory Dyn Syst 2002;22(3):845–862.

    MathSciNet  Article  MATH  Google Scholar 

  9. 9.

    Guysinsky M, Katok A. Normal forms and invariant geometric structures for dynamical systems with invariant contracting foliations. Math Res Lett 1998;5(1–2): 149–163.

    MathSciNet  Article  MATH  Google Scholar 

  10. 10.

    Hirsch MW, Pugh CC, Shub M. Invariant manifolds. Lecture Notes in Mathematics. vol. 583; 1977.

  11. 11.

    Ilyashenko Y. Thick attractors of step skew products. Regular Chaotic Dyn 2010; 15:328–334.

    MathSciNet  Article  MATH  Google Scholar 

  12. 12.

    Ilyashenko Y. Thick attractors of boundary preserving diffeomorphisms. Indagationes Mathematicae 2011;22(3–4):257–314.

    MathSciNet  Article  MATH  Google Scholar 

  13. 13.

    Ilyashenko YS. Diffeomorphisms with intermingled attracting basins. Funkts Anal Prilozh 2008;42(4):60–71.

    MathSciNet  Article  MATH  Google Scholar 

  14. 14.

    Ilyashenko Y, Negut A. Invisible parts of attractors. Nonlinearity 2010;23: 1199–1219.

    MathSciNet  Article  MATH  Google Scholar 

  15. 15.

    Ilyashenko Y, Negut A. Hölder properties of perturbed skew products and Fubini regained. Nonlinearity 2012;25:2377–2399.

    MathSciNet  Article  MATH  Google Scholar 

  16. 16.

    Ilyashenko Y, Li W. Nonlocal bifurcations: American Mathematica Society; 1998.

  17. 17.

    Kleptsyn V, Volk D. Nonwandering sets of interval skew products. Nonlinearity 2014;27:1595–1601.

    MathSciNet  Article  MATH  Google Scholar 

  18. 18.

    Kudryashov YG. Bony attractors. Funkts Anal Prilozh 2010;44(3):73–76.

    MathSciNet  Article  MATH  Google Scholar 

  19. 19.

    Milnor J. Fubini foiled: Katok’s paradoxical example in measure theory, The Mathematical Intelligencer. Spring 1997;19(2):30–32.

    MathSciNet  MATH  Google Scholar 

  20. 20.

    Shub M, Wilkinson A. Pathological foliations and removable zero exponents. Inventiones Math 2000;139:495–508.

    MathSciNet  Article  MATH  Google Scholar 

  21. 21.

    Sternberg S. On the structure of local homeomorphisms of Euclidian n-space, II. Amer. J. Math 1958;80:623–631.

    MathSciNet  Article  MATH  Google Scholar 

  22. 22.

    Pugh C, Shub M, Wilkinson A. Hölder foliations, revisited. J Modern Dyn 2012;6:835–908.

    MathSciNet  Article  MATH  Google Scholar 

Download references


We are very thankful to Ilya Schurov and Stas Minkov for their attentive reading of the preliminary versions of this article and their remarks on the presentation. We are also grateful to the Referee for numerous valuable comments. Olga Romaskevich is mainly supported by UMPA ENS Lyon. (UMR 5669 CNRS) and the LABEX MILYON (ANR-10-LABX-0070) of Université de Lyon, within the program “Investissements d’Avenir” (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR) as well as by the French-Russian Poncelet laboratory (UMI 2615 of CNRS and Independent University of Moscow). Both authors are supported by RFBR project 16-01-00748.

Author information



Corresponding author

Correspondence to Yulij Ilyashenko.

Appendix: Proof of Theorem 3 and Other Calculations

Appendix: Proof of Theorem 3 and Other Calculations

In the appendix, we will prove the technical propositions stated above.

Hölder Properties of Some Auxiliary Functions

First let us prove

Proposition 2

Function π n (b) defined as a product of λ b in the first n points of the orbit of a linear diffeomorphism A, see Eq. 22 , is Hölder with the exponent α and

$$\|{\Pi}_{n}\|_{[\alpha]} \leq \frac{C_{\lambda} \theta^{n}}{(\mu^{\alpha}-1)q} $$

where C λ is Hölder constant for λ, 𝜃 is defined in Eq. 23 . Here and below α is given by Eq. 9 and μ is the largest magnitude of eigenvalues of A.

Proof of Proposition 2

$$\begin{array}{@{}rcl@{}} \left| {\Pi}_{n}(b_{1})-{\Pi}_{n}(b_{2}) \right| &=& \left| \prod\limits_{k=0}^{n-1} \lambda_{A^{k} b_{1}}-\prod\limits_{k=0}^{n-1} \lambda_{A^{k} b_{2}} \right|= \left| \lambda_{b_{1}}-\lambda_{b_{2}} \right| \times \left| \prod\limits_{k=1}^{n-1} \lambda_{A^{k} b_{1}} \right|\\&&+ \left|\lambda_{b_{2}}\right| \left| {\Pi}_{n-1}(A b_{1}) - {\Pi}_{n-1} (A b_{2}) \right| \leq \\ &\leq& q^{n-1} C_{\lambda} \sum\limits_{k=0}^{n-1} \left|\left|A^{k} b_{1}-A^{k} b_{2} \right|\right|^{\alpha} \leq q^{n-1} C_{\lambda} \frac{\mu^{n \alpha}-1}{\mu^{\alpha}-1} \|b_{1}-b_{2}\|^{\alpha} \\&\leq& \frac{C_{\lambda} \theta^{n}}{(\mu^{\alpha}-1)q} \|b_{1}-b_{2}\|^{\alpha} \end{array} $$

Proposition 3

Function P n (b) defined by \(P_{n}(b):=\frac {{\Pi }_{n}(b)}{\lambda _{A^{n} b}}\) is Hölder with exponent α and

$$\|P_{n}\|_{[\alpha]} \leq D^{2} C_{\lambda} B \theta^{n} $$

where B depends only on the initial map, the precise formula for B is given below, see Eq. 51.


$$\begin{array}{@{}rcl@{}} \left| P_{n}(b_{1})-P_{n}(b_{2}) \right| &=& \left| \frac{{\Pi}_{n}(b_{1}) \lambda_{A^{n} b_{2}}-{\Pi}_{n}(b_{2}) \lambda_{A^{n} b_{1}}}{\lambda_{A^{n} b_{1}} \lambda_{A^{n} b_{2}}} \right| \\ &\leq& D^{2} \left| \lambda_{A^{n} b_{2}}\prod\limits_{k=0}^{n-1} \lambda_{A^{k} b_{1}} \right.- \left. \lambda_{A^{n} b_{1}}\prod\limits_{k=0}^{n-1} \lambda_{A^{k} b_{2}} \right|\\ \end{array} $$
$$\begin{array}{@{}rcl@{}} &=& D^{2} \left| \left( \lambda_{A^{n} b_{2}}-\lambda_{A^{n} b_{1}}\right) {\Pi}_{n}(b_{1})+{\Pi}_{n+1}(b_{1}) \right. \\&&-\left.\left( \lambda_{A^{n} b_{1}} -\lambda_{A^{n} b_{2}}\right) {\Pi}_{n}(b_{2})-{\Pi}_{n+1}(b_{2})) \right| \\ &\leq&|{\Pi}_{n+1} (b_{1})-{\Pi}_{n+1}(b_{2})|+\left|\lambda_{A^{n} b_{1}}-\lambda_{A^{n} b_{2}}| |{\Pi}_{n}(b_{1})-{\Pi}_{n}(b_{2}) \right| \\&\leq& D^{2} \left[ \frac{C_{\lambda} \theta^{n+1}}{(\mu^{\alpha}-1)q}+2 q^{n} C_{\lambda} \mu^{n \alpha} \right] \|b_{1}-b_{2}\|^{\alpha} \leq D^{2} C_{\lambda} B \theta^{n} \|b_{1}-b_{2}\|^{\alpha} \end{array} $$

where B does not depend on anything but initial skew product:

$$ B(\theta,\mu,\alpha,q)=\frac{\theta}{(\mu^{\alpha}-1)q}+2 $$

Proposition 4

Function 𝜃 2,k (b 1 ,b 2 ) defined as 𝜃 2,k (b 1 ,b 2 )=|P k (b 2 )||Q k, 1 −Q k,2 | is Hölder with α as exponent and

$$\|\theta_{2,k}\|_{[\alpha]} \leq \theta^{k} D \left( C_{Q} + q^{k-1} \textup{Lip}_{x} Q \frac{C_{\lambda}}{\mu^{\alpha}-1}\right) $$

Here, \(Q_{k,1}=Q \circ {F_{0}^{k}}(b_{1},x)\) and \(Q_{k,2}=Q \circ {F_{0}^{k}}(b_{2},x)\) , and the definition of P k (b) was reminded in Proposition 2 above.


We use the results of Proposition 2 in the following chain of inequalities.

$$\begin{array}{@{}rcl@{}} \theta_{2,k} &\leq& q^{k} D \left| Q_{A^{k} b_{1}}({\Pi}_{k}(b_{1})x)-Q_{A^{k} b_{2}}({\Pi}_{k}(b_{2})x)\right| \\&\leq& q^{k} D \left| Q_{A^{k} b_{1}}({\Pi}_{k}(b_{1})x)-Q_{A^{k} b_{2}}({\Pi}_{k}(b_{1})x)\right|\\&&+q^{k} D \left| Q_{A^{k} b_{2}}({\Pi}_{k}(b_{2})x)-Q_{A^{k} b_{2}}({\Pi}_{k}(b_{1})x) \right| \\&\leq& q^{k} D C_{Q} \mu^{k \alpha} \|b_{1}-b_{2}\|^{\alpha} +q^{k} D \text{Lip}_{x} Q \|{\Pi}_{k}\|_{\mathcal{H}^{\alpha}}\|b_{1}-b_{2}\|^{\alpha} \\&\leq&\theta^{k} D \left( C_{Q} + q^{k-1} \text{Lip}_{x} Q \frac{C_{\lambda}}{\mu^{\alpha}-1}\right) \|b_{1}-b_{2}\|^{\alpha} \end{array} $$

Proposition 5

Operator L is bounded in the Lipschitz norm: there exists a constant Lip x L such that for any \(h \in \mathcal {M}\) holds

$$\textup{Lip}_{x} \left( Lh \right) \leq \textup{Lip}_{x} L \times \textup{Lip}_{x} h. $$

Moreover, if h(b,⋅)∈C l for any \(b \in \mathbb {T}^{d}\) , then Lh has the same smoothness as well and

$$\left\|(Lh)^{(l)}\right\|_{C} \leq C_{k}(L) \left\|h^{(l)}\right\|_{C}. $$


Using the explicit formula for the solution (24), as well as bounds (29) and (35), we have:

$$\begin{array}{@{}rcl@{}} \sup\limits_{x,y \in [0,1]} \left| \frac{L h (b,x)-Lh(b,y)}{x-y}\right|&=&\sup\limits_{x,y \in [0,1]} \left| \sum\limits_{k=0}^{\infty} P_{k}(b) \frac{h \circ {F_{0}^{k}}(b,x)-h \circ {F_{0}^{k}}(b,y)}{x-y} \right| \\ \end{array} $$
$$\begin{array}{@{}rcl@{}} &\leq& \sup\limits_{x,y \in [0,1]} \sum\limits_{k=0}^{\infty} P_{k}(b)\frac{\text{Lip}_{x} h |{\Pi}_{k}(b) x -{\Pi}_{k}(b) y|}{|x-y|} \\&=& \text{Lip}_{x} h \frac{D}{1-q^{2}}. \end{array} $$

The bounds for the derivatives are obtained analogously by differentiating term by term the series (24):

$$(Lh)^{(l)} =-\sum\limits_{k=0}^{\infty} P_{k}(b) {{\Pi}_{k}^{l}}(b) h^{(l)} \circ {F_{0}^{k}}. $$


$$\left\|(Lh)^{(l)}\right\|_{C} \leq \frac{D}{1-q^{l+1}}\left\|h^{(l)}\right\|_{C} . $$

Proposition 6

There exist polynomials T 3 (ε) and \({T_{4}^{0}}(\varepsilon )\) of degrees respectively 3 and 4 such that \({T_{4}^{0}}(0)=0\) and for any h∈N holds

$$ \textup{Lip}_{x,\varepsilon} [{\Phi} h] \leq T_{3}(\varepsilon)+{T_{4}^{0}}(\varepsilon) A_{\textnormal{Lip}} $$

Proof of Proposition 6

The proof of this proposition deals with an expression for Lip x, ε Φh which is given by

$$\sup\limits_{x,y \in [0,\varepsilon] }\frac{1}{|x-y|} \left| (1+xh_{b}(x))^{2} Q(b, x+\bar{h}_{b}(x))-(1+yh_{b}(y))^{2} Q(b, y+\bar{h}_{b}(y)) \right| $$

Since in this proposition the base coordinate b is fixed and x is changing we will permit to ourselves not to write the b index and just suppose that \(Q(x)=Q(b, x+\bar {h}_{b}(x))\) as well as h(x) = h b (x). The bound is a triangle inequality formula:

$$\begin{array}{@{}rcl@{}} \text{Lip}_{x,\varepsilon} {\Phi} h &\leq& \sup_{x,y \in [0,\varepsilon]} \left| \frac{Q(x) -Q(y)}{x-y} \right|+2 \sup_{x,y \in [0,\varepsilon]} \left| \frac{x h(x) Q(x) -y h(y) Q(y)}{x-y} \right|\\&&+ \sup_{x,y \in [0,\varepsilon]} \left| \frac{x^{2} h(x) Q(x)-y^{2} h(y) Q(y)}{x-y} \right| \leq \text{Lip}_{x} Q (1+\text{Lip}_{x,\varepsilon} \bar{h})\\ &&+ 2 \sup_{x,y \in [0,\varepsilon]} \left| \frac{x h(x) \left( Q(x)-Q(y)\right)}{x-y} \right| + 2 \sup_{x,y \in [0,\varepsilon]} \left| \frac{x Q(y) (h(x)-h(y))}{x-y} \right| \\&&+2 \sup_{y \in [0,\varepsilon]}\left| Q(y) h(y) \right|+\sup_{x,y \in [0,\varepsilon]} \left| \frac{x^{2} h(x) \left( Q(x)-Q(y)\right)}{x-y} \right|\\&&+ \sup_{x,y \in [0,\varepsilon]} \left| \frac{Q(y) \left[ x^{2}\left( h(x)-h(y)\right)+(x^{2}-y^{2})h(y)\right]}{x-y} \right| \\&\leq& \text{Lip}_{x} Q (1+\text{Lip}_{x,\varepsilon} \bar{h})+2 \varepsilon A_{C} \text{Lip}_{x} Q \text{Lip}_{x,\varepsilon} \bar{h}+ 2 \varepsilon \|Q\|_{C} A_{\text{Lip}} \\&&+ 2 \|Q\|_{C} A_{C} + \varepsilon^{2} A_{C} \text{Lip}_{x} Q \text{Lip}_{x,\varepsilon} \bar{h}+\|Q\|_{C} (\varepsilon^{2} A_{\text{Lip}}+ 2 \varepsilon A_{C}) \end{array} $$

Let us note that

$$\begin{array}{@{}rcl@{}} \text{Lip}_{x,\varepsilon} \bar{h}&=&\sup_{x,y \in [0,\varepsilon]} \left| \frac{x^{2} h(x)-y^{2} h(y)}{x-y} \right| \\&\leq& \sup_{x,y \in [0,\varepsilon]} \left|\frac{x^{2} (h(x)-h(y))}{x-y}\right|+\sup_{x,y \in [0,\varepsilon]} \left| \frac{h(y) (x^{2}-y^{2})}{x-y}\right| \\&\leq& A_{\text{Lip}} \varepsilon^{2} + A_{C} 2 \varepsilon \end{array} $$

After substitution of \(\text {Lip}_{x,\varepsilon } \bar {h}\) in Eq. 54 by Eq. 55, we have the result with

$$\begin{array}{@{}rcl@{}} T_{3}(\varepsilon)&=&2{A_{C}^{2}} \text{Lip}_{x} Q \varepsilon^{3} + 4 {A_{C}^{2}} \text{Lip}_{x} Q \varepsilon^{2}+2 A_{C} (\text{Lip}_{x} Q +\|Q\|_{C})\varepsilon+ \text{Lip}_{x} Q + 2 \|Q\|_{C} A_{C} \\ {T_{4}^{0}}(\varepsilon)&=&\text{Lip}_{x} Q A_{C} \varepsilon^{4} + 2 A_{C} \text{Lip}_{x} Q \varepsilon^{3} + \text{Lip}_{x} Q \varepsilon^{2} + 2 \|Q\|_{C} \varepsilon \end{array} $$

Now let us prove the analogous proposition for the Hölder norm of the operator Φ:

Proposition 7

If \(h \in \mathcal {H}^{\alpha }_{\varepsilon }\) then \({\Phi } h \in \mathcal {H}^{\alpha }_{\varepsilon }\) as well. And, moreover, for h∈N, there exist polynomials \(\tilde {T}_{2}(\varepsilon )\) and \(\tilde {T}_{4}^{0}(\varepsilon ), \tilde {T}_{4}^{0}(\varepsilon )(0)=0\) of degrees 2 and 4 correspondingly such that

$$\left\|{\Phi} h \right\|_{[\alpha], \varepsilon} \leq \tilde{T}_{4}^{0}(\varepsilon) A_{\alpha}+ \tilde{T}_{2}(\varepsilon) $$


To estimate Hölder norm of the shift operator, we need some more triangle inequalities.

$$\begin{array}{@{}rcl@{}} |{\Phi} h(b_{1}, x)-{\Phi} h (b_{2},x)|&=&|(1+x h_{b_{1}})^{2} Q_{b_{1}} - (1+xh_{b_{2}})^{2} Q_{b_{2}}| \\&\leq& |Q_{b_{1}}-Q_{b_{2}}|+ 2 x \left|h_{b_{1}} Q_{b_{1}}-h_{b_{2}} Q_{b_{2}}\right|\\&&+ x^{2} \left| h_{b_{1}}^{2} Q_{b_{1}} - h_{b_{2}}^{2} Q_{b_{2}} \right| \\ &\leq& \left| Q(b_{1}, x+\bar{h}_{b_{1}})-Q(b_{1}, x+\bar{h}_{b_{2}}) \right|\\&&+ \left| Q(b_{1}, x+\bar{h}_{b_{2}})-Q(b_{2}, x+\bar{h}_{b_{2}}) \right| \\&&+ 2 \varepsilon \left| h_{b_{1}} \left( Q(b_{1}, x+\bar{h}_{b_{1}})-Q(b_{1}, x+\bar{h}_{b_{2}})\right) \right|\\&&+ 2 \varepsilon \left| h_{b_{1}} \left( Q(b_{1}, x+\bar{h}_{b_{2}}) -Q(b_{2}, x+\bar{h}_{b_{2}}) \right) \right| \\&&+ 2 \varepsilon \left| Q_{b_{2}} (h_{b_{1}}-h_{b_{2}}) \right|+\varepsilon^{2} h_{b_{1}}^{2} \left| Q(b_{1}, x+\bar{h}_{b_{1}}) - Q(b_{1}, x+\bar{h}_{b_{2}}) \right|\\&&+ \varepsilon^{2} h_{b_{1}}^{2} \left| Q(b_{1}, x+\bar{h}_{b_{2}})-Q(b_{2}, x+\bar{h}_{b_{2}}) \right|\\&&+\varepsilon^{2} |Q_{b_{2}}| \left| h_{b_{1}}^{2}-h_{b_{2}}^{2} \right| \\&\leq& \|b_{1}-b_{2}\|^{\alpha} \left( \tilde{T}_{4}^{0}(\varepsilon) A_{\alpha}+\tilde{T}_{2}(\varepsilon) \right) \end{array} $$


$$\tilde{T_{4}}^{0} (\varepsilon)=\text{Lip}_{x} Q \varepsilon^{2} + 2 \varepsilon^{3} A \text{Lip}_{x} Q + 2 \varepsilon \|Q\|_{C} + \varepsilon^{4} A^{2} \text{Lip}_{Q} + 2 \|Q\|_{C} A \varepsilon^{2} $$


$$\tilde{T}_{2}(\varepsilon)=C_{Q}+ 2 \varepsilon A C_{Q}+ \varepsilon^{2} A^{2} C_{Q} $$

All the propositions stated above are proven. This completes the proof of our main result – Theorem 2.

Now we are ready to prove that the conjugacy is smooth in fiber variable, and Hölder with its derivatives in base variables.

Proof of Theorem 3

The proof of a smooth version of Theorem 2 is analogous to the proof of the latter theorem. Here, we will give a sketch of the proof: we will only show that the conjugacy H is (k−2)– smooth with respect to the fiber variable. The proof of the fact that its fiber derivatives are now Hölder on b is analogous to the proof of the Hölder property fot the function H itself and we do not give it here.

The idea is to change the space \(\mathcal {N}\) in an appropriate way. For some constants A 0,…, A l > 0 and κ 0,…, κ l > 0 let us define the space

$$\mathcal{N}_{l}:=\left\{ h(\cdot, b) \in C^{l}([0,\varepsilon]) : \|h\|_{C} \leq A_{0}, {\ldots} , \left\|h^{(l)}\right\|_{C} \leq A_{l} \right\} $$

with the norm

$$ \|h\|_{*}=\kappa_{0} \|h\|_{C} + {\ldots} + \kappa_{l} \|h^{(l)}\|_{C}. $$

We have now to prove the analogues of Lemmas 1, 2, and 3 above, and then follow the argument in Theorem 2. The homological and shift operators will stay the same although the functional spaces in which they act will be smaller, and the metric will be not continuous but a smooth one. □

Lemma 1 (Smooth case)

Operator L is bounded in the norm ( 56 ).


$$\begin{array}{@{}rcl@{}} \|L h\|_{*} &=& \sum\limits_{j=0}^{l} \kappa_{j} \| (L h)^{(j)}\|_{C} \leq \sum\limits_{j=0}^{l} \frac{D \kappa_{j}}{1-q^{j+1}}\|h^{(j)}\|_{C} \\&\leq&\frac{D}{1-q} \sum\limits_{j=0}^{\infty} \kappa_{j} \|h^{(j)}\|_{C} = \frac{D}{1-q}\|h\|_{*} \end{array} $$

For the space \(\mathcal {N}_{l}\) to map to itself by LΦ, we should choose constants A 0, A 1,…, A l appropriately. For LΦ to be contracting in the space, we should appropriately choose κ 0,…, κ l . Let us show that these two choices can be made without complications and the analogues of Lemmas 2 and 3 hold.

In what concerns the operator Φ, we will use its presentation (43) and calculate the derivatives for k = 0,…, l: by the Leibnitz rule:

$$({\Phi} h)^{(k)}=\sum\limits_{j=0}^{k} {C_{k}^{j}} ((1+x h(b,x))^{2})^{(j)}Q^{(k-j)}(b, x+\bar{h}) $$

The explicit form of the right-hand side is not as important as a fact that it can be written as a sum of polynomials in derivatives of \(h, \bar {h}\) and Q. Indeed, there exist polynomials τ 0,…, τ l and σ 0,…, σ l such that

$$\begin{array}{@{}rcl@{}} ({\Phi} h)^{(k)}&=&\sum\limits_{j=0}^{k} {C_{k}^{j}} \tau_{j}(x,h, \ldots, h^{(j)}) \sigma_{j} \\&&\times\left( x, \bar{h}, \ldots, \bar{h}^{(k-j)}, Q\left( b, x+\bar{h}), \ldots, \right.\right. \left. \left. Q^{(k-j)}(b,x+\bar{h} \right)\right) \end{array} $$

We will estimate the continuous norm of the right-hand side of Eq. 58 in \(\mathbb {T}^{d} \times [0,\varepsilon ]\). So, we will have that for some polynomials T j and S j there is a bound

$$\begin{array}{@{}rcl@{}} \|L{\Phi} h^{(k)}\|_{C, \varepsilon} &\leq& \sum\limits_{j=0}^{k} {C_{k}^{j}} T_{j}(\varepsilon, A_{0}, \ldots, A_{j}) S_{j}\\&& \times\left( \varepsilon, A_{0}, \ldots, A_{k-j}, \|Q\|_{C}, \ldots, \right. \left. \|Q^{(k-j)}\|_{C}\right) \end{array} $$

Note that the coefficient in front of A k in this expression is a polynomial that has no free term. Indeed, A k comes only from T k or S 0: in both of the cases, A k is multiplied by at least one ε. Hence, we need to find A 0,…A l such that l+1 equations hold for some polynomials \(P_{k}, {P_{k}^{0}}, {P_{k}^{0}}(0)=0\):

$$ {P_{k}^{0}}(\varepsilon) A_{k}+P_{k}(\varepsilon) C(A_{0}, \ldots, A_{k-1}) \leq A_{k}, k=0, \ldots, l $$

First, we take ε such that all polynomials \({P_{k}^{0}}(\varepsilon )<1\). Then we satisfy the Eq. 60 one by one, starting from k = 0 by choosing A k one by one, starting with A 0 and by increasing the index.

Now we have to prove that operator LΦ is contracting in the space \(\mathcal {N}\) if κ 0,…κ l are properly chosen.

One can show that

$$\begin{array}{@{}rcl@{}} \|L {\Phi} h - L {\Phi} g\|_{*} &\leq& \frac{D}{1-q} \sum\limits_{j=0}^{l} \kappa_{j} \| \sum\limits_{k=0}^{j} {C_{j}^{k}} \\&&\times\left( \tau_{k}(x,h, {\ldots} h^{(k)}) \sigma_{k} \left( x, \bar{h}, \ldots, \bar{h}^{(j-k)}, Q(b,x+h), \ldots, \right. \right. \\&&\left. \left. Q^{(j-k)}(b,x+h)\right)-\right.\left. \tau_{k}\left( x,g, {\ldots} g^{(k)}) \sigma_{k}(x, \bar{g}, \ldots, \bar{g}^{(j-k)}, Q(b,x+g), \ldots,\right.\right. \\&&\left.\left. Q^{(j-k)}(b,x+g)\right) \right) \|_{C, \varepsilon} \leq \sum\limits_{j=0}^{l} \|h^{(k)}-g^{(k)}\|_{C, \varepsilon}\\&&\times \left( \kappa_{k} {U_{k}^{0}}(\varepsilon)+\sum\limits_{j={k+1}}^{l} U_{j}(\varepsilon) \kappa_{j} \right) \end{array} $$

for some polynomials \(U_{j}, {U_{j}^{0}}\). For the right-hand side to be less than ξhg for some ξ<1, the following system should be satisfied:

$$\begin{array}{@{}rcl@{}} {U_{0}^{0}} (\varepsilon)+ \frac{\kappa_{1}}{\kappa_{0}} U_{0}(\varepsilon) + {\ldots} \frac{\kappa_{l}}{\kappa_{0}} U_{0}(\varepsilon) \leq \xi\\ {U_{1}^{0}} (\varepsilon)+ \frac{\kappa_{2}}{\kappa_{1}} U_{0}(\varepsilon) + {\ldots} \frac{\kappa_{l}}{\kappa_{1}} U_{0}(\varepsilon) \leq \xi \\ {\ldots} \\ U_{l-1}^{0} (\varepsilon) + \frac{\kappa_{l}}{\kappa_{l-1}} U_{l-1}(\varepsilon) \leq \xi\\{U_{l}^{0}}(\varepsilon) \leq \xi \end{array} $$

One can choose ε in such a way that \({U^{0}_{k}}(\varepsilon )<\xi \). Then, the last inequality in the list is true, by taking any κ l and κ l−1 big enough, we satisfy the before-last inequality, and we proceed in satisfying these inequalities from the last one till the first one.

So, we obtain a contracting operator. We have proved that in the space \(\mathcal {N}_{l}\) of functions defined in a neighborhood of the base with a metric chosen appropriately, there is a contracting operator LΦ. Its fixed point is the needed conjugacy which will be sufficiently smooth on the fiber variable x.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Ilyashenko, Y., Romaskevich, O. Sternberg Linearization Theorem for Skew Products. J Dyn Control Syst 22, 595–614 (2016).

Download citation


  • Normal forms
  • Skew products
  • Sternberg theorem

Mathematics Subject Classification (2010)

  • 37G05
  • 37D30