Kolmogorov’s theorem for degenerate Hamiltonian systems with Hölder continuous parameters

Abstract

In this paper, we investigate Kolmogorov-type theorems for small perturbations of degenerate Hamiltonian systems. These systems are index by a parameter ξ as $ H(y,x,\xi) = \langle\omega(\xi),y\rangle {+ \bar h(y,\xi)}+\varepsilon P(y,x,\xi,\varepsilon) $, where ɛ > 0. We assume that the frequency mapping $\omega(\cdot)$, $\bar h(y,\cdot)=O(|y|^2)$ and the perturbation $\varepsilon P(y,x,\cdot, \varepsilon)$ maintain Hölder continuity about ξ. We prove that the persistent invariant tori retain the same frequency as those of the unperturbed tori, under a certain topological degree condition and a weak convexity condition for the frequency mapping. Notably, this paper presents, to our understanding, pioneering results on the KAM theorem under such conditions with only assumption of Hölder continuous dependence of frequency mapping ω on the parameter.

1. Introduction

As a conservation law of energy, Hamiltonian systems are frequently considered to describe models arose in celestial mechanics or the motion of charged particles in magnetic fields, see [8, 24, 38].

The classical KAM theory, as presented by Arnold, Kolmogorov, and Moser [1, 20, 25], posits that under the Kolmogorov non-degenerate condition, most invariant tori of an integrable Hamiltonian system can withstand small perturbations. While these tori might undergo minor deformations, they transform into other invariant tori that retain the original frequency.

Numerous methods have been explored to study the persistence of invariant tori and the preservation of toral frequency within Hamiltonian systems under certain non-degenerate conditions. For instance, the KAM approach was used in [2, 4, 12, 22, 30, 35]. The direct method using Lindstedt series can be found in references [9, 13, 15], while renormalization group techniques were discussed in [5, 16]. Notably, the study presented in [10] introduced the idea of partial preservation of unperturbed frequencies and delved into the persistence problem on a specified smooth sub-manifold for real analytic Hamiltonian systems, particularly under the Rüssmann-like non-degenerate condition. For insights under analogous conditions, see also [37].

Yet, in the context of persistence, two fundamental questions emerge that warrant attention:

Q1: In the event of a failure in the Kolmogorov non-degenerate condition, can the invariant tori with the same frequency still be preserved under small perturbations?

Q2: If the regularity of the frequency mapping diminishes to mere continuity, can the aforementioned result withstand small perturbations?

To shed light on these questions, we review previous findings and offer a more comprehensive overview.

1.1. Degeneracy

Consider the real analytic nearly integrable Hamiltonian system

(1.1)\begin{align} H(y,x,\varepsilon )=h(y)+\varepsilon P(y,x,\varepsilon), \end{align}

where x is the angle variable in the standard torus $\mathbb{T}^{n}$, $n\geq2$ refers to the dimension; y is the action variable in a bounded closed region $G\subset \mathbb{R}^n$, and ɛ > 0 is a small parameter.

A fundamental assumption in historical research is the Kolmogorov non-degenerate condition. However, if we assume that there exists a $y_0\in G$ such that:

\begin{align*} \det\frac{\partial^2h(y_0)}{\partial y^2}=0, \end{align*}

then the Kolmogorov non-degenerate condition is not satisfied. The spatial solar system serves as a prominent example of this situation, as detailed in [14]. Naturally, a question arises: does the persistence result still stand under these conditions? This question has been a primary motivation for this research.

In fact, even under weaker non-degenerate conditions, KAM tori might not preserve their frequencies. As demonstrated in [6, 34, 36], under the Brjuno non-degenerate condition and Rüssmann non-degenerate condition, the presumption of an unchanged frequency may not necessarily hold true. This is because the frequency of perturbed tori can undergo slight variations. Similar observations are noted in [3, 7, 10, 11, 18, 32, 40]. Consequently, deriving conditions that assure the persistence of frequencies for KAM tori in the context of a degenerate Hamiltonian becomes rather challenging. Furthermore, the issue of the perturbed invariant tori maintaining a consistent frequency has seldom been tackled for degenerate systems.

1.2. Regularity

On the matter of regularity, it is worth noting the distinctions in the studies of various researchers. Kolmogorov [20] and Arnold [1] focused on real analytic Hamiltonian systems. In contrast, Moser [25] illustrated that Hamiltonian systems do not necessarily need to be analytic; a high, albeit finite, level of regularity for the Hamiltonian suffices. This regularity requirement was later reduced to C ⁵ in work by [33]. Further important contributions on this topic can be found in [4, 19, 21, 35]. Moreover, the scenario where the frequency mapping has Lipschitz continuous parameters has been explored in [29]. A subsequent question of interest is: what are the implications when the regularity of the frequency mapping is merely Hölder continuous with respect to its parameters?

More precisely, we consider a family of Hamiltonian systems under small perturbations:

(1.2)\begin{equation} H(y,x,\xi,\varepsilon)=\langle\omega(\xi),y\rangle{+\bar h(y,\xi)}+\varepsilon P(y,x,\xi,\varepsilon), \end{equation}

where $(y,x)\in G\times\mathbb{T}^n$ and ξ is a parameter in a bounded closed region $O\subset \mathbb{R}^n$. The function $\omega(\cdot)$, $\bar h(y,\cdot)=O(|y|^2)$ and $P(y,x,\cdot,\varepsilon)$ are Hölder continuous with respect to the parameter ξ with Hölder index β, for some $0 \lt \beta \lt 1$. The function $\bar h(\cdot,\xi)$ and $P(\cdot,\cdot,\xi,\varepsilon)$ are real analytic with respect to y and x. Additionally, ɛ > 0 is a small parameter.

It is important to note that in the conventional KAM iteration process, the regularity of the frequency mapping concerning the parameters must be at least Lipschitz continuous. This ensures that the parameter domain remains intact. However, when the regularity of the frequency mapping is less stringent than Lipschitz continuous, the traditional method of parameter excavation becomes infeasible. This necessitates the exploration of novel approaches to address the issue.

1.3. Our work

Regarding regularity, when the frequency mapping is Hölder continuous with respect to parameters, we prove that the perturbed invariant tori retain the same Diophantine frequency as their unperturbed counterparts for Hamiltonian systems as described in (1.2), see theorem 2.1. For the degeneracy problem, persistence results under the highly degenerate Hamiltonian system (1.1) are proved in theorem 2.5.

We establish sufficient conditions based on the topological degree condition $\mathrm{(A0)}$ and the weak convexity condition $\mathrm{(A1)}$ for frequency mapping. Detailed descriptions of these conditions are provided in section 2. In deriving our primary results, we employ the quasi-linear KAM iteration procedure as in [10, 17, 23, 31]. Notably, we use the technique of translation parameters to counteract frequency drift. Specifically, we adjust the action variable to maintain constant frequency for the highly degenerate Hamiltonian system (1.1). It is also noteworthy that the weak convexity condition proposed in this paper is necessary regardless of the smoothness level of the frequency mapping, as evidenced by proposition 2.3.

It should be pointed out that the KAM-type theorems associated with parameter family are due to Moser [26], Pöschel [28]. However, our results are different from theirs: a Diophantine frequency can be given in advance, but Moser’s systems need to be modified in KAM iteration and hence cannot be given beforehand; in Pöschel’s approach, the frequency set need to be dug out in KAM process. Our method is to find a parameter in the family of systems by translating parameter. Of course, it does not work generally. Recently, the persistence of invariant tori with prescribed frequencies has been developed in [39, 41]. Here we prove the persistence of invariant tori, which form a Hölder continuous family with respect to parameters. As showed by proposition 2.3, the weak convexity condition (A1) is sharp to such persistence.

The rest of this paper is organized as follows. In section 2, we state our main results (theorems 2.1, 2.5, 2.6 and 2.8). We describe the quasi-linear iterative scheme, show the detailed construction and estimates for one cycle of KAM steps in section 3. In section 4, we complete the proof of theorem 2.1 by deriving an iteration lemma and showing the convergence of KAM iterations. In section 5, we prove theorem 2.5 , which covers the analytic situation, and is also a special case of theorem 2.1. We also prove theorem 2.8 by directly computing. Finally, the proof of theorem 2.6 can be found in appendix B.

2. Main results

To state our main results, we first need to introduce a few definitions and notations.

1. (1) Given a domain $D\subset G\times \mathbb{T}^n$, we let $\bar{D}$, $\partial D$ denote the closure and the boundary of D, respectively. $D^o:=\bar{D}\setminus\partial D$ refers to the interior.

2. (2) We shall use the same symbol $\vert\cdot\vert$ to denote an equivalent vector norm and its induced matrix norm, absolute value of functions, etc, and use $\vert\cdot\vert_D$ to denote the supremum norm of functions on a domain D.

3. (3) For the perturbation function $P(y,x,\xi)$, which is analytic about y and x and Hölder continuous about ξ with Hölder index β, $0 \lt \beta \lt 1$, we define its norm as follows

   \begin{equation*}\vert\|P\|\vert_{D}={|P|_{D}}+\|P\|_{C^\beta}\end{equation*}

   where

   (2.1)\begin{equation}\|P\|_{C^\beta}=\sup_{\xi\neq\zeta,~\xi,\zeta\in {O}}\frac{\vert P(y,x,\xi)-P(y,x,\zeta)\vert}{\vert\xi-\zeta\vert^\beta},~~~\forall (y,x)\in {D}.\end{equation}

4. (4) For any two complex column vectors ξ, η in the same space, $\langle\xi,\eta\rangle$ always stands for $\xi^\top\eta$.

5. (5) i_d is the unit map, and I_d is the unit matrix.

6. (6) For a vector-value function f, Df denotes the Jacobian matrix of f, and $J_f=det Df$ its Jacobian determinant.

7. (7) All Hamiltonians in the sequel are endowed with the standard symplectic structure.

8. (8) As pointed out in [30], the real analyticity of the Hamiltonian $H(y,x)$ about y and x on $G\times{\mathbb{T}^n}$ implies that the analyticity extends to a complex neighbourhood $D(s,r)$ of $G\times{\mathbb{T}^n}$, where $D(s,r)$ is defined for some $0 \lt s,r \lt 1$, with

   \begin{equation*}D(s,r):=\{(y,x):|y| \lt s,\vert\textrm{Im}x\vert \lt r\}.\end{equation*}

9. (9) For $\forall \delta \gt 0$, $y_0\in G$, let

   \begin{align*} B_\delta(y_0)&=:\{y\in G:\vert y-y_0\vert \lt \delta\},\\ \bar B_\delta(y_0)&=:\{y\in G:\vert y-y_0\vert\leq\delta\}. \end{align*}

We are now ready to state our assumptions. Mainly we consider (1.2), i.e., for any ɛ > 0 small enough, we consider the parameterized family of perturbed Hamiltonian systems

\begin{equation*} \left\{ \begin{array}{ll} H:G\times \mathbb{T}^n\times O\rightarrow \mathbb{R}^1,\\ H(y,x,\xi)=\langle\omega(\xi),y\rangle{+\bar h(y,\xi)}+\varepsilon P(y,x,\xi,\varepsilon). \end{array} \right. \end{equation*}

First, we make the following assumptions:

1. (A0) Fix $\xi_0\in O^o$ such that

   (2.2)\begin{align} &\deg\left(\omega(\cdot), O^o, \omega(\xi_0)\right)\neq0. \end{align}

2. (A1) There exists a small neighbourhood $B_\varsigma(\xi_0)\subset O^o$ with ς > 0 and a constant σ > 0 such that

   (2.3)\begin{align} \left\vert\omega(\xi)-\omega(\xi_{*})\right\vert\geq \sigma\left\vert\xi-\xi_{*}\right\vert^\beta,~~~~\forall \xi,\xi_{*}\in B_\varsigma(\xi_0), \end{align}

   where β is defined in (2.1).

3. (A2) For the given $\xi_0\in O^o$, $\omega(\xi_0)$ satisfies the Diophantine condition

   (2.4)\begin{align} \left\vert\langle k,\omega(\xi_0)\rangle\right\vert \gt \frac{\gamma}{\vert k\vert^\tau},~~~k\in{\mathbb{Z}^n\setminus{\{0\}}}, \end{align}

   where $k=(k_1,\cdots,k_n)$, $\vert k\vert=\left\vert k_1\right\vert+\cdots+\left\vert k_n\right\vert$, γ > 0 and $\tau \gt n-1$.

Then, we have the following main results:

Theorem 2.1 Consider Hamiltonian system (1.2). Assume that $\mathrm{(A0)}$, $\mathrm{(A1)}$ and $\mathrm{(A2)}$ hold. Then there exists a sufficiently small $\varepsilon_0 \gt 0$, for any $0 \lt \varepsilon \lt \varepsilon_0$, there exist $\xi_\varepsilon\in O$ with $\xi_\varepsilon\rightarrow\xi_0$ as $\varepsilon\rightarrow0$ and a symplectic transformation $\Psi_*$ such that

\begin{equation*}H(\Psi_*(y,x,\xi_\varepsilon),\varepsilon)=e_*+\langle\omega(\xi_0),y\rangle+\bar h_*(y,\xi_{\varepsilon})+P_*(y,x,\xi_\varepsilon,\varepsilon),\end{equation*}

where $e_*$ is a constant, $\bar h_*(y,\xi_{\varepsilon})=O(\vert y\vert^2)$, $P_*=O(\vert y\vert^2)$. Thus the perturbed Hamiltonian system $H(y,x,\xi_\varepsilon,\varepsilon)$ admits an invariant torus with frequency $\omega(\xi_0)$.

Moreover, there is a Hölder continuous family of $\xi_\varepsilon(\hat\xi_0)$ with respect to $\hat\xi_0\in \hat O$, where

\begin{align*}\hat O & =\left\{\hat\xi_0\in O:|\omega(\hat \xi_0)-\omega(\xi_0)| \lt d,|\langle k,\omega(\hat\xi_0)\rangle| \gt \frac{\gamma}{|k|^\tau},k\in\mathbb{Z}^n\backslash\{0\}\right\},\nonumber\\ d & =\min_{\xi\in\partial O}\left\{\;|\omega(\xi)-\omega(\xi_0)|\;\right\},\end{align*}

such that the perturbed Hamiltonian system $H(y,x,\xi_\varepsilon(\hat\xi_0),\varepsilon)$ admits an invariant torus with frequency $\omega(\hat\xi_0)$, and $\xi_\varepsilon(\hat\xi_0)\rightarrow\hat\xi_0$ as $\varepsilon\rightarrow0$.

Remark 2.2. It should be emphasized that we deal with the degenerate Hamiltonian system in which the frequency mapping and the perturbation are Hölder continuous about parameters in this theorem.

In the following, we will give some examples to state that conditions $\mathrm{(A0)}$ and $\mathrm{(A1)}$ are indispensable, especially for condition $\mathrm{(A1)}$. See below for a counter example:

Proposition 2.3. Consider Hamiltonian system (1.2), for n = 2, with

\begin{equation*}\omega(\xi)=(\omega_1(\xi_1),\omega_2(\xi_2))^\top,~~~\bar h(y,\xi)=0~~~\varepsilon P=P_0(\varepsilon)y_2,\end{equation*}

where $\xi=(\xi_1,\xi_2)$,

\begin{align*} &\omega_1(\xi_1)=\bar{\omega}_1+\xi_1,~~\xi_1\in(-1,1),\\ &\omega_2{(\xi_2)}=\left\{\begin{array}{lll} \bar{\omega}_2+\exp\{-\frac{1}{(\xi_2+{\frac{1}{2}})^2}\},~~~&\xi_2\in(-1,-\frac{1}{2}),\\ \bar{\omega}_2,~~~&\xi_2\in[-\frac{1}{2},\frac{1}{2}],\\ \bar{\omega}_2-\exp\{-\frac{1}{(\xi_2-{\frac{1}{2}})^2}\},~~~&\xi_2\in(\frac{1}{2},1), \end{array}\right. \end{align*}

$\bar{\omega}(\xi_0)=(\bar{\omega}_1,\bar{\omega}_2)^\top$ satisfies Diophantine condition (2.4), and

\begin{equation*}P_0(\varepsilon)=\left\{\begin{array}{lll} 0,&&\varepsilon=0, \\ \varepsilon^\ell\sin\frac{1}{\varepsilon},&&\varepsilon\neq0,\,\ell\in \mathbb{Z}^+\setminus\{0\}. \end{array}\right.\end{equation*}

Then condition $\mathrm{(A1)}$ fails for any parameter $\xi\in(-\frac{1}{2},\frac{1}{2})\times(-\frac{1}{2},\frac{1}{2})$. Moreover, although the topological degree is not equal to 0, we cannot find ξ_ɛ such that $\xi_\varepsilon\rightarrow\xi_0$ as $\varepsilon\rightarrow0$.

See appendix A for the complete proof.

Remark 2.4. This counter example implies that $\mathrm{(A1)}$ is necessary to ensure $\xi_\varepsilon\rightarrow\xi_0$, as $\varepsilon\rightarrow0$.

Nevertheless, one asks what happens to the frequency mapping in the analytic situation. As a special case of our theorem 2.1, we also obtain the Kolmogorov’s theorem for analytic Hamiltonian systems with degeneracy. This is stated in the following theorem.

Theorem 2.5 Consider real analytic Hamiltonian system (1.1). Fix $\xi_0\in G$ such that $\mathrm{(A0)}$, $\mathrm{(A1)}$ and $\mathrm{(A2)}$ hold for $\omega(\xi)=\nabla h(\xi)$, O = G, and $\beta\geq 1$. Then there exists a sufficiently small positive constant $\varepsilon' \gt 0$ such that if $0 \lt \varepsilon \lt \varepsilon'$, there exists $y_\varepsilon\in G$ such that Hamiltonian system (1.1) at $y=y_\varepsilon$ admits an invariant torus with frequency $\nabla h(\xi_0)$.

This theorem is proved in section 5.1.

Next, we give an example that satisfies conditions $\mathrm{(A0)}$-$\mathrm{(A1)}$. For simplicity we use the action variable y as the parameter ξ.

Theorem 2.6 Consider Hamiltonian system (1.1) with

\begin{equation*} h(y)=\langle \omega,y\rangle+\frac{1}{2l+2}\vert y\vert^{2l+2}, \end{equation*}

where $y\in G\subset \mathbb{R}^n$, l is a positive integer, $\omega\in \mathbb{R}^n\setminus{\{0\}}$ satisfies the Diophantine condition (2.4). Then Hamiltonian system (1.1) admits an invariant torus with frequency ω for any small enough perturbation.

The proof can be found in appendix B.

Proposition 2.7. If $h(y)=\langle \omega,y\rangle+\frac{1}{2l+1}\left\vert y\right\vert^{2l+1}$ in Hamiltonian system (1.1), $\omega\in \mathbb{R}^n\setminus{\{0\}}$ satisfies the Diophantine condition (2.4), then the system may not admit invariant torus with frequency ω.

The proof can be found in appendix C.

Above results imply that condition $\mathrm{(A0)}$ is indispensable for n > 1 case. Furthermore, we also prove that for n = 1, the persistence results in theorem 2.6 hold under some weaker conditions, provided that the frequency satisfies Diophantine condition $\mathrm{(A2)}$.

Theorem 2.8 Consider Hamiltonian (1.1), for n = 1, with

\begin{align*} h(y)=\omega y+g(y),~~~~\varepsilon P(y,x,\varepsilon)=\varepsilon P(y), \end{align*}

where $y\in G=[-1,1]\subset \mathbb{R}^1$, ω satisfies Diophantine condition (2.4).

1. (1) If $g(y)\in C^{2\ell+1}$, $g'(0)=\cdots=g^{2\ell}(0)=0$, $g^{2\ell+1}(0)\neq 0$, $\ell$ is a positive integer, then the perturbed system admits at least two invariant tori with frequency ω for the small enough perturbation satisfying $\varepsilon P'(y)\, sign (g^{2\ell+1}(0)) \lt 0$; conversely, if $\varepsilon P'(y)\, sign (g^{2\ell+1}(0)) \gt 0$, the unperturbed invariant torus with frequency ω will be destroyed.

2. (2) If $g(y)\in C^{2\ell+2}$, $g'(0)=\cdots=g^{2\ell+1}(0)=0$, $g^{2\ell+2}(0)\neq 0$, $\ell$ is a positive integer, then the perturbed system admits an invariant tori with frequency ω for any small enough perturbation.

Remark 2.9. We do not know whether the results in theorem 2.8 can be extended to higher dimensions or not.

3. KAM step

In this section, we describe the quasi-linear iterative scheme, show the detailed construction and estimates for one cycle of KAM steps, which is essential to study the KAM theory, see [10, 17, 22, 23, 30]. It should be pointed out that in our KAM iteration, we translate the parameter domain; while in the usual KAM iteration, one has to dig out a decreasing series of parameter domains, see [10, 17, 23, 29–32].

3.1. Description of the 0th KAM step

Given an integer $m\geq\beta+1$, where β is defined in (2.1). Denote $\rho=\frac{1}{2(m+1)}$, and let η > 0 be an integer such that $(1+\rho)^\eta \gt 2$. We define

(3.1)\begin{equation}\gamma=\varepsilon^{\frac{1}{4(n+m+2)}}.\end{equation}

Consider Hamiltonian (1.2). We first define the following 0th KAM step parameters:

(3.2)\begin{align} &r_0=r,~~~~~\gamma_0=\gamma,~~~~~e_0=0, ~~~~~{\bar{h}_0=\bar h(y,\xi)},~~~~~\omega_0(\xi)=\omega(\xi),\notag\\ & \mu_0=\varepsilon^{\frac{1}{8\eta(\tau+1)(m+1)}},\\ &s_0=\frac{s\gamma_0}{16(M^*+2)K_1^{\tau+1}},~~~~{O_0=\left\{\xi:\xi\in B_{\varrho_0}(\xi_0),~\varrho_0= s_0^{\frac{1}{\beta}}\right\},}\notag\\ &D(s_0,r_0):=\left\{(y,x):|y| \lt s_0,\vert\textrm{Im}x\vert \lt r_0\right\},\notag \end{align}

where $0 \lt s_0,\gamma_0,\mu_0\leq 1$, $\tau \gt n-1$, $M^* \gt 0$ is a constant defined as in lemma 3.4, and

\begin{align*} K_1=\left(\left[\log\frac{1}{\mu_0}\right]+1\right)^{3\eta}. \end{align*}

Therefore, we can write

\begin{align*} H_0&=: H(y,x,\xi,\varepsilon)=N_0+P_0,\\ N_0&=: N_0(y,\xi)=e_0+\langle\omega_0(\xi),y\rangle+\bar{h}_0,\\ P_0&=:\varepsilon P(y,x,\xi,\varepsilon), \end{align*}

where $(y,x,\xi)\in D(s_0,r_0)\times O_0$.

We first prove an important estimate.

Lemma 3.1.

(3.3)\begin{equation} \vert\|P_0\|\vert_{D(s_0,r_0)}\leq\gamma_0^{n+m+2}s_0^m\mu_0. \end{equation}

Proof. Using the fact $\gamma_0^{n+m+2}=\varepsilon^{\frac{1}{4}}$ and $\left[\log\frac{1}{\mu_0}\right]+1 \lt \frac{1}{\mu_0}$, we have

\begin{align*} s_0^m=\frac{s^m\varepsilon^{\frac{m}{4(n+m+2)}}}{16^m(M^*+2)^mK_1^{m(\tau+1)}} \gt \frac{s^m\varepsilon^{\frac{m}{4(n+m+2)}}\mu_0^{3\eta m(\tau+1)}}{16^m(M^*+2)^m}\geq\frac{s^m\varepsilon^{\frac{m}{4(n+m+2)}+\frac{3}{8}}}{16^m(M^*+2)^m}. \end{align*}

Moreover, let $\varepsilon_0 \gt 0$ be small enough so that

(3.4)\begin{equation} \varepsilon_0^{\frac{1}{8}-\frac{1}{8\eta(\tau+1)(m+1)}}\vert\|P\|\vert_{D(s_0,r_0)}\frac{16^m(M^*+2)^m}{s^m}\leq 1, \end{equation}

using the fact that $\mu_0=\varepsilon^{\frac{1}{8\eta(\tau+1)(m+1)}}$, we get

(3.5)\begin{align} \gamma_0^{n+m+2}s_0^m\mu_0&\geq\frac{s^m\varepsilon^{\frac{m}{4(n+m+2)}+\frac{3}{8}+\frac{1}{4}+\frac{1}{8\eta(\tau+1)(m+1)}}}{16^m(M^*+2)^m} \geq\frac{s^m\varepsilon^{\frac{1}{4}+\frac{3}{8}+\frac{1}{4}+\frac{1}{8\eta(\tau+1)(m+1)}}}{16^m(M^*+2)^m} \notag\\ &=\varepsilon^{\frac{7}{8}}\frac{s^m\varepsilon^{\frac{1}{8\eta(\tau+1)(m+1)}}}{16^m(M^*+2)^m}, \end{align}

and by (3.4) and $0 \lt \varepsilon \lt \varepsilon_0$,

\begin{align*} \varepsilon^{\frac{1}{8}-\frac{1}{8\eta(\tau+1)(m+1)}}\vert\|P\|\vert_{D(s_0,r_0)}\frac{16^m(M^*+2)^m}{s^m}\leq 1, \end{align*}

i.e.,

(3.6)\begin{align} \varepsilon^{\frac{1}{8}}\vert\|P\|\vert_{D(s_0,r_0)}\leq \frac{s^m\varepsilon^{\frac{1}{8\eta(\tau+1)(m+1)}}}{16^m(M^*+2)^m}. \end{align}

Then by (3.5) and (3.6),

\begin{align*} \vert\|P_0\|\vert_{D(s_0,r_0)}=\varepsilon^{\frac{7}{8}} \varepsilon^{\frac{1}{8}}\vert\|P\|\vert_{D(s_0,r_0)}\leq \varepsilon^{\frac{7}{8}}\frac{s^m\varepsilon^{\frac{1}{8\eta(\tau+1)(m+1)}}}{16^m(M^*+2)^m}\leq\gamma_0^{n+m+2}s_0^m\mu_0, \end{align*}

which implies (3.3).

This completes the proof.

3.2. Induction from the νth KAM step

3.2.1. Description of the νth KAM step

We now define the νth KAM step parameters:

\begin{equation*}r_\nu=\frac{r_{\nu-1}}{2}+\frac{r_0}{4},~~~s_\nu=\frac{1}{8}\mu_{\nu-1}^{2\rho}s_{\nu-1},~~~\mu_\nu=8^m\mu_{\nu-1}^{1+\rho},\end{equation*}

where $\rho=\frac{1}{2(m+1)}$.

Now, suppose that at νth step, we have arrived at the following Hamiltonian:

(3.7)\begin{equation} \begin{aligned} H_\nu&=N_\nu+P_\nu,~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\\ N_\nu&=e_\nu+\langle\omega_\nu(\xi),y\rangle+\bar{h}_\nu(y,\xi), \end{aligned} \end{equation}

defined on $D(s_\nu,r_\nu)\times O_{\nu}$ and

(3.8)\begin{equation} \left\vert\left\|P_\nu\right\|\right\vert_{D(s_\nu,r_\nu)}\leq\gamma_0^{n+m+2}s_\nu^m\mu_\nu. \end{equation}

The equation of motion associated to H_ν is

(3.9)\begin{equation} \left\{ \begin{array}{ll} \dot{y}_\nu=-\partial_{x_\nu} H_\nu,\\ \dot{x}_\nu=~~\partial_{y_\nu} H_\nu. \end{array} \right. \end{equation}

Except for additional instructions,we will omit the index for all quantities of the present KAM step (at νth step) and use + to index all quantities (Hamiltonians, domains, normal forms, perturbations, transformations, etc.) in the next KAM step (at $(\nu+1)$th step). Using this notation, ξ_ν is abbreviated as ξ, which can easily be confused with the parameter ξ. Therefore, we will not omit the subscript of ξ_ν and continue to denote it as ξ_ν. To simplify the notations, we will not specify the dependence of P, $P_+$ etc. All the constants c ₁-c ₆ below are positive and independent of the iteration process, and we will also use c to denote any intermediate positive constant which is independent of the iteration process.

Define

\begin{align*} r_+&=\frac{r}{2}+\frac{r_0}{4},\\ s_+&=\frac{1}{8}\alpha s,~~~~~\alpha=\mu^{2\rho}=\mu^{\frac{1}{m+1}},\\ \mu_+&=8^mc_0\mu^{1+\rho},~~~~~c_0=\max\{1,c_1,c_2,\cdots,c_6\},\\ K_+&=\left(\left[\log\frac{1}{\mu}\right]+1\right)^{3\eta},\\ \hat{D}&=D\left(s,r_++\frac{7}{8}(r-r_+)\right),\\ \tilde{D}&=D\left(\frac{1}{2} s,r_++\frac{6}{8}(r-r_+)\right),\\ D(s)&=\{y\in C^n:\vert y\vert \lt s\},\\ D_{\frac{i}{8}\alpha}&=D\left(\frac{i}{8}\alpha s,r_++\frac{i-1}{8}(r-r_+)\right), ~~i=1,2,\cdots,8,\\ D_+&=D_{\frac{1}{8}\alpha}=D(s_+,r_+),\\ O_+&=\left\{\xi:\xi\in B_{\varrho}(\xi_\nu),~\varrho= s^{\frac{1}{\beta}}\right\},\\ \Gamma(r-r_+)&=\sum_{0 \lt \vert k\vert\leq K_+}\left\vert k\right\vert^{6\tau+7}e^{-\vert k\vert\frac{r-r_+}{8}}. \end{align*}

3.2.2. Construct a symplectic transformation

We will construct a symplectic coordinate transformation $\Phi_{+}$:

(3.10)\begin{align} \Phi_{+}: D(s_{+},r_{+})\times O_+\rightarrow D(s,r)\times O \end{align}

such that it transforms the Hamiltonian ($3.7$) into the Hamiltonian of the next KAM cycle (at $(\nu+1)$th step), i.e.,

(3.11)\begin{equation} H_{+}=H\circ\Phi_{+}=N_{+}+ P_{+}, \end{equation}

where $N_{+}$ and $P_{+}$ have similar properties as N and P respectively on $D(s_{+},r_{+})\times O_+$, and the equation of motion (3.9) is changed into

(3.12)\begin{equation} \left\{ \begin{array}{ll} \dot{y}_{+}=-\partial_{x_{+}} H_{+},\\ \dot{x}_{+}=~~\partial_{y_{+}} H_{+}. \end{array} \right. \end{equation}

In the following, we prove (3.12). Let $\Phi_+(y_+,x_+):=\left(\Phi_+^1(y_+,x_+),\Phi_+^2(y_+,x_+)\right)$, by (3.10), we have

\begin{align*} \left(\begin{array}{c} \dot{y}\\ \dot{x} \end{array}\right)=\left(\begin{array}{cc} (\partial_{y_+} \Phi_{+}^1)\dot{y}_{+}&(\partial_{x_+} \Phi_{+}^1)\dot{x}_{+}\\ (\partial_{y_+} \Phi_{+}^2)\dot{y}_{+}&(\partial_{x_+} \Phi_{+}^2)\dot{x}_{+} \end{array}\right)=D\Phi_{+}\left(\begin{array}{c} \dot{y}_{+}\\ \dot{x}_{+} \end{array}\right), \end{align*}

by (3.10) and (3.11), we get

\begin{align*} \left(\begin{array}{c} \partial_{y_{+}} H\\ \partial_{x_{+}} H \end{array}\right)&=\left(\begin{array}{cc} \partial_{y} H\partial_{y_+} y&\partial_{x} H\partial_{y_+} x\\ \partial_{y} H\partial_{x_+} y&\partial_{x} H\partial_{x_+} x \end{array}\right) =\left(\begin{array}{cc} \partial_{y_+} \Phi_+^1&\partial_{y_+} \Phi_+^2\\ \partial_{x_+} \Phi_+^1&\partial_{x_+} \Phi_+^2 \end{array}\right) \left(\begin{array}{cc} \partial_y H\\ \partial_x H \end{array}\right)\\ &=D\Phi_{+}^\top\left(\begin{array}{cc} \partial_y H\\ \partial_x H \end{array}\right). \end{align*}

Then this together with (3.9) yields

\begin{align*} \left(\begin{array}{c} \dot{y}_{+}\\ \dot{x}_{+} \end{array}\right)&=D\Phi_{+}^{-1}\left(\begin{array}{c} \dot{y}\\ \dot{x} \end{array}\right) =D\Phi_{+}^{-1}J\left(\begin{array}{cc} \partial_y H\\ \partial_x H \end{array}\right) =D\Phi_{+}^{-1}J(D\Phi_{+}^{-1})^\top\left(\begin{array}{c} \partial_{y_{+}} H\\ \partial_{x_{+}} H \end{array}\right)\\ & =J\left(\begin{array}{c} \partial_{y_{+}} H\\ \partial_{x_{+}} H \end{array}\right),\\ \end{align*}

where J is the standard symplectic matrix, i.e.,

\begin{align*} J=\left(\begin{array}{cc} 0&-I_n\\ I_n&0 \end{array}\right). \end{align*}

This finishes the proof of (3.12).

Next, we show the detailed construction of $\Phi_+$ and the estimates of $P_+$.

3.2.3. Truncation

Consider the Taylor-Fourier series of P:

\begin{equation*} P=\sum_{k\in \mathbb{Z}^n,~\imath\in \mathbb{Z}_+^n}p_{k\imath}y^{\imath}e^{\sqrt{-1}\langle k,x\rangle}, \end{equation*}

and let R be the truncation of P of the form

\begin{equation*} R=\sum_{\vert k\vert\leq K_+,~\vert\imath\vert\leq m}p_{k\imath}y^{\imath}e^{\sqrt{-1}\langle k,x\rangle}. \end{equation*}

Remark 3.2. As demonstrated in [30], truncating y to first order is feasible. When proving theorem 2.1 in this paper, truncating y to first order is permissible (i.e., m = 1), though truncating to higher orders is also valid. However, in proving theorem 2.5, truncating y to higher orders is necessary (i.e., $m\geq\beta+1$); see (5.12) in subsection 5.1. Because we regard the action variable as the parameter in proving theorem 2.5, and the translation of the parameter is equivalent to the translation of the action variable, we need the integer $m\geq\beta+1$ to ensure that the translated action variable y does not run out of the given domain. Here, to make the KAM steps applicable to both theorems, we truncated y to higher orders. In this paper, we denote the order of truncation as m, which provides a more general description of the KAM iteration, clarifying how estimates vary with truncation and demonstrating the role of higher-order terms in the normal form, even though it may make the proof appear somewhat more complicated.

Next, we will prove that the norm of P − R is much smaller than the norm of P by selecting truncation appropriately, see lemma below.

Lemma 3.3. Assume that

\begin{align*} {\mathbf{(H1)}}: \int_{K_+}^{\infty}t^{n}e^{-t\frac{r-r_+}{16}}dt\leq\mu. \end{align*}

Then there is a constant c ₁ such that

(3.13)\begin{align} \left\vert\left\|P-R\right\|\right\vert_{D_\alpha}&\leq c_1\gamma_0^{n+m+2}s^m\mu^2, \end{align}

(3.14)\begin{align} \left\vert\left\|R\right\|\right\vert_{D_\alpha}&\leq c_1\gamma_0^{n+m+2}s^m\mu. \end{align}

Proof. Denote

\begin{align*} I&=\sum_{\vert k\vert \gt K_+,~\imath\in \mathbb{Z}_+^n}p_{k\imath}y^{\imath} e^{\sqrt{-1}\langle k,x\rangle},\\ II&=\sum_{\vert k\vert\leq K_+,~\vert\imath\vert \gt m}p_{k\imath}y^{\imath} e^{\sqrt{-1}\langle k,x\rangle}. \end{align*}

Then

\begin{align*} P-R=I+II. \end{align*}

To estimate I, we note by (3.8) that

(3.15)\begin{align} \left\vert\sum_{\imath\in \mathbb{Z}_+^n}p_{k\imath}y^{\imath}\right\vert\leq \left\vert P\right\vert_{D(s,r)}e^{-\vert k\vert r}\leq\gamma_0^{n+m+2}s^{m}\mu e^{-\vert k\vert r}, \end{align}

where the first inequality has been frequently used in [10, 11, 17, 22, 30–32, 35] and the detailed proof see [35]. This together with ${\mathbf{(H1)}}$ yields

(3.16)\begin{align} \left\vert I\right\vert_{\hat D}&\leq\sum_{\vert k\vert \gt K_+}\left\vert\sum_{\imath\in \mathbb{Z}_+^n}p_{k\imath}y^{\imath}\right\vert e^{\vert k\vert(\frac{r_+}{8}+\frac{7r}{8})} \leq\sum_{\vert k\vert \gt K_+}\left\vert P\right\vert _{D(s,r)}e^{-\vert k\vert\frac{r-r_+}{8}}\notag\\ &\leq\gamma_0^{n+m+2}s^{m}\mu\sum_{\kappa=K_+}^{\infty}\kappa^{n}e^{-\kappa\frac{r-r_+}{8}} \leq\gamma_0^{n+m+2}s^{m}\mu\int_{K_+}^{\infty}t^{n}e^{-t\frac{r-r_+}{16}}dt\\ &\leq\gamma_0^{n+m+2}s^{m}\mu^2\notag \end{align}

It follows from (3.8) and (3.16) that

\begin{align*} \left\vert P-I\right\vert_{\hat D}\leq\vert P\vert_{D(s,r)}+\vert I\vert_{\hat D}\leq2\gamma_0^{n+m+2}s^{m}\mu. \end{align*}

For $\vert p\vert=m+1$, let $\int$ be the obvious antiderivative of $\frac{\partial^{p}}{\partial y^p}$. Then the Cauchy estimate of P − I on D_α yields

\begin{align*} \left\vert II\right\vert_{D_\alpha}&=\left\vert\int\frac{\partial^{p}}{\partial y^p}\sum_{\vert k\vert\leq K_+,~\vert\imath\vert \gt m}p_{k\imath}y^{\imath} e^{\sqrt{-1}\langle k,x\rangle}dy\right\vert_{D_{\alpha}}\\ &\leq\left\vert\int\left\vert\frac{\partial^{p}}{\partial y^p}(P-I)\right\vert dy\right\vert_{D_{\alpha}}\\ &\leq\left\vert\frac{c}{s^{m+1}}\int\left\vert P-I\right\vert_{\hat{D}}dy\right\vert_{D_{\alpha}}\\ &\leq2\frac{c}{s^{m+1}}\gamma_0^{n+m+2}s^{m}\mu(\alpha s)^{m+1}\\ &\leq c\gamma_0^{n+m+2}s^{m}\mu^2. \end{align*}

Thus,

(3.17)\begin{align} \left\vert P-R\right\vert_{D_\alpha}=\left\vert I+II\right\vert_{D_\alpha}\leq c\gamma_0^{n+m+2}s^{m}\mu^2, \end{align}

and therefore,

(3.18)\begin{align} \left\vert R\right\vert_{D_\alpha}\leq\vert P-R\vert_{D_\alpha}+\vert P\vert_{D(s,r)}\leq c\gamma_0^{n+m+2}s^{m}\mu. \end{align}

Next, we estimate $\left\|P-R\right\|_{C^\beta}$. In view of the definition of $\|\cdot\|_{C^\beta}$, for $\forall y,x\in D_\alpha$, we have

(3.19)\begin{align} \left\|P-R\right\|_{C^\beta}&=\sup_{\xi\neq\zeta}\frac{\left\vert P(x,y,\xi)-R(x,y,\xi)-(P(x,y,\zeta)-R(x,y,\zeta))\right\vert}{\left\vert\xi-\zeta\right\vert^\beta}\notag\\ &\leq\sup_{\xi\neq\zeta}\frac{\left\vert\int\right\vert\frac{\partial^p}{\partial y^p}(P(x,y,\xi)-R(x,y,\xi)-(P(x,y,\zeta)-R(x,y,\zeta)))\left\vert dy\right\vert}{\vert\xi-\zeta\vert^\beta}\notag\\ &\leq\sup_{\xi\neq\zeta}\frac{\left\vert\frac{c}{s^{m+2}}\int\left\vert P(x,y,\xi)-P(x,y,\zeta)\right\vert dy\right\vert}{\vert\xi-\zeta\vert^\beta}\notag\\ &\leq\sup_{\xi\neq\zeta}\frac{c}{s^{m+1}}\frac{\vert P(x,y,\xi)-P(x,y,\zeta)\vert}{\vert\xi-\zeta\vert^\beta}(\alpha s)^{m+1}\notag\\ &\leq c\mu\|P\|_{C^\beta}\leq c\gamma_0^{n+m+2}s^m\mu^2, \end{align}

where the third inequality follows from Cauchy estimate and the last inequality follows from (3.8).

Similarly, we get

(3.20)\begin{align} \|R\|_{C^\beta} \lt \|P-R\|_{C^\beta}+\|P\|_{C^\beta}\leq c\gamma_0^{n+m+2}s^m\mu. \end{align}

It follows from (3.17), (3.18), (3.19) and (3.20) that (3.13) and (3.14) hold.

This completes the proof.

3.2.4. Homological equation

As usual, we shall construct a symplectic transformation as the time-1 map $\phi_{F}^1$ of the flow generated by a Hamiltonian F to eliminate all resonant terms in R, i.e., all terms

\begin{align*} p_{k\imath}y^{\imath}e^{\sqrt{-1}\langle k,x\rangle},~~~~0 \lt \vert k\vert\leq K_+,\vert\imath\vert\leq m. \end{align*}

To do so, we first construct a Hamiltonian F of the form

(3.21)\begin{equation} F=\sum_{0 \lt \vert k\vert\leq K_+,\vert\imath\vert\leq m}f_{k\imath}y^{\imath}e^{\sqrt{-1}\langle k,x\rangle}, \end{equation}

satisfying the equation

(3.22)\begin{equation} \{N,F\}+R-[R]=0, \end{equation}

where $[R]=\frac{1}{(2\pi)^n}\int_{T^n}R(y,x)dx$ is the average of the truncation R.

Substituting (3.21) into (3.22) yields that

\begin{align*} &-\sum_{0 \lt \vert k\vert \lt K_+,\vert\imath\vert\leq m}\sqrt{-1}\left\langle k,\omega(\xi)+\partial_y\bar{h}\right\rangle f_{k\imath}y^\imath e^{\sqrt{-1}\langle k,x\rangle}\\ &+\sum_{0 \lt \vert k\vert \lt K_+,\vert\imath\vert\leq m} p_{k\imath}y^\imath e^{\sqrt{-1}\langle k,x\rangle}=0. \end{align*}

By comparing the coefficients above, we then obtain the following quasi-linear equations:

(3.23)\begin{equation} \sqrt{-1}\left\langle k,\omega(\xi)+\partial_y\bar{h}\right\rangle f_{k\imath}=p_{k\imath},~~~\vert\imath\vert\leq m,~~~0 \lt \vert k\vert\leq K_+. \end{equation}

We declare that the quasi-linear equations (3.23) are solvable under some suitable conditions. The details can be seen in the following lemma:

Lemma 3.4. Assume that

\begin{align*} &{\mathbf{(H2)}}: max_{\vert i\vert\leq 2}\left\vert\left\| \partial_y^i\bar{h}- \partial_y^i\bar{h}_0\right\|\right\vert_{D(s)}\leq \mu_0^{\frac{1}{2}}, \\ &{\mathbf{(H3)}}:3s \lt \frac{\gamma_0}{(M^*+2)K_+^{\tau+1}}, \end{align*}

where

\begin{align*} {M^*=\max_{\vert i\vert\leq 2, y\in D(s)}\left\vert \partial_{y}^i\bar h_0(y,\xi_0)\right\vert.} \end{align*}

Then the quasi-linear equations (3.23) can be uniquely solved on D(s) to obtain a family of functions $f_{k\imath}$ which are analytic in y, Hölder continuous about ξ with Hölder index β, and satisfy the following properties:

(3.24)\begin{equation} \left\vert\left\|\partial_y^if_{k\imath}\right\|\right\vert_{D(s)}\leq c_2|k|^{(\vert i\vert+1)\tau+\vert i\vert}\gamma_0^{n+m+1-\vert i\vert}s^{m-\vert\imath\vert}\mu e^{-\vert k\vert r}, \end{equation}

for all $\vert\imath\vert\leq m, 0 \lt \vert k\vert\leq K_+, \vert i\vert\leq 2$, where c ₂ is a constant.

Proof. For $\forall y\in D(s)$, by $\mathbf{(H2)},\mathbf{(H3)}$, we have

\begin{equation*} \left\vert\partial_y\bar h\right\vert_{D(s)}=\left\vert(\partial_y\bar h-\partial_y\bar h_0)+\partial_y\bar h_0\right\vert_{D(s)} \leq(1+M^*)\vert y\vert \lt (1+M^*)s \lt \frac{\gamma_0}{3\vert k\vert^{\tau+1}} \end{equation*}

and

\begin{equation*} \left\|\partial_y\bar h\right\|_{C^\beta}=\sup_{\xi_*\neq\xi_{**},~\xi_*,\xi_{**}\in{O}}\frac{\left\vert\partial_y\bar h(y,\xi_*)-\partial_y\bar h(y,\xi_{**})\right\vert}{\vert\xi_*-\xi_{**}\vert^\beta}\leq\mu_0^\frac{1}{2}\vert y\vert \lt \vert s\vert \lt \frac{\gamma_0}{3\vert k\vert^{\tau+1}}, \end{equation*}

which imply that

(3.25)\begin{equation} \left\vert\left\|\partial_y\bar{h}\right\|\right\vert_{D(s)} \lt \frac{\gamma_0}{3\vert k\vert^{\tau+1}}. \end{equation}

It follows from (3.25) and $\mathrm{(A2)}$ that for $\xi\in O$, $y\in D(s)$, $0 \lt |k|\leq K_+$,

(3.26)\begin{align} \left\vert \left\langle k,\omega(\xi)+\partial_y \bar h(y)\right\rangle\right\vert&\geq\left\vert\left\langle k,\omega(\xi_\nu) \right\rangle\right\vert -\left\vert\left\langle k,\omega(\xi)-\omega(\xi_\nu) \right\rangle\right\vert -\left\vert\left\langle k,\partial_y \bar h(y) \right\rangle\right\vert \notag\\ & \gt \frac{\gamma_0}{\vert k\vert^\tau}-\frac{\gamma_0}{3\vert k\vert^\tau}-\frac{\gamma_0}{3\vert k\vert^\tau}=\frac{\gamma_0}{3\vert k\vert^\tau}. \end{align}

Hence

(3.27)\begin{equation} L_k=\sqrt{-1}\left\langle k,\omega(\xi)+\partial_y\bar h(y)\right\rangle \end{equation}

is invertible, and

(3.28)\begin{equation} f_{k\imath}=L_k^{-1} p_{k\imath}, \end{equation}

for all $y\in D(s)$, $\xi\in O$, $0 \lt \vert k\vert\leq K_+$, $\vert\imath\vert\leq m$. The analyticity of $\bar h(y)$ and $p_{k\imath}$ implies that $f_{k\imath}$ is analytic with respect to y. It follows from (3.23) that the Hölder continuity of $f_{k\imath}$ with respect to the parameter ξ is determined by $\omega(\xi)$, $\bar{h}$ and $p_{k\imath}$, which are Hölder continuous about parameter ξ with Hölder index β. Let $0 \lt \vert k\vert\leq K_+$. We note by the first inequality of (3.15) and Cauchy estimate that

(3.29)\begin{align} \left\vert\left\|p_{k\imath}\right\|\right\vert_{D(s)}&\leq\left\vert\left\| \partial_y^\imath P\right\|\right\vert_{\tilde{D}}e^{-\vert k\vert r}\leq\gamma_0^{n+m+2}s^{m-\vert\imath\vert}\mu e^{-\vert k\vert r},~~~~\vert\imath\vert\leq m, \end{align}

and by (3.26) and (3.27) that

\begin{align*} \left\vert\left\|L_k^{-1}\right\|\right\vert_{D(s)}&\leq c\frac{\vert k\vert^{2\tau+1}}{\gamma_0^2}, \end{align*}

and

(3.30)\begin{align} \left\vert\left\|\partial_y^iL_k^{-1}\right\|\right\vert_{D(s)}&\leq c_2\left\vert k\right\vert^{\vert i\vert}\left\|\left\vert L_k^{-1}\right\|\right\vert_{D(s)}^{\vert i\vert+1}\leq c_2\frac{\vert k\vert^{2(\vert i\vert+1)\tau+2\vert i\vert+1}}{\gamma_0^{2(\vert i\vert+1)}},~~~~\vert i\vert\leq2. \end{align}

So, by (3.28), (3.29) and (3.30), we get

\begin{align*} \left\vert\left\| \partial_y^if_{k\imath}\right\|\right\vert_{D(s)}&\leq c_2\frac{\vert k\vert^{2(\vert i\vert+1)\tau+2\vert i\vert+1}}{\gamma_0^{2(\vert i\vert+1)}}\gamma_0^{n+m+2}s^{m-\vert\imath\vert}\mu e^{-\vert k\vert r}\\ &=c_2\vert k\vert^{2(\vert i\vert+1)\tau+2\vert i\vert+1}\gamma_0^{n+m-2\vert i\vert}s^{m-\vert\imath\vert}\mu e^{-\vert k\vert r},~~~~\vert i\vert\leq 2. \end{align*}

This completes the proof.

Next, we apply the above transformation $\phi_F^1$ to Hamiltonian H, i.e.,

\begin{align*} H\circ\phi_F^1=&(N+R)\circ\phi_F^1+(P-R)\circ\phi_F^1\\ =&(N+R)+\{N,F\}+\int_0^1\left\{(1-t)\{N,F\}+R,F\right\}\circ\phi_F^tdt\\ &+(P-R)\circ\phi_F^1\\ =&N+[R]+\int_0^1\left\{R_t,F\right\}\circ\phi_F^tdt+\left(P-R\right)\circ\phi_F^1\\ =&:N_++ P_+, \end{align*}

where

(3.31)\begin{align} & N_+=N+[R]=e_++\langle\omega_+(\xi),y\rangle+\bar{h}_+(y,\xi), \end{align}

(3.32)\begin{align} &e_+=e+p_{00}^\nu, \end{align}

(3.33)\begin{align} &\omega_+=\omega_0(\xi)+\sum_{j=0}^{\nu}p_{01}^j(\xi), \end{align}

(3.34)\begin{align} &\bar{h}_+=\bar{h}(y,\xi)+[R]-p_{00}^\nu-\left\langle p_{01}^\nu(\xi),y\right\rangle, \end{align}

(3.35)\begin{align} & P_+=\int_0^1\{R_t,F\}\circ\phi_F^tdt+(P-R)\circ\phi_F^1,\\ &R_t=(1-t)[R]+tR.\notag \end{align}

3.2.5. Translation

In this subsection, we will translate the parameter domain O to $O_+$ so that appropriate parameters can be found in $O_+$ to ensure the frequency remains unchanged in the next KAM step. To achieve this, we need to find an appropriate $\xi_+\in O$.

3.2.6. Frequency-preserving

In this subsection, we find the parameter $\xi_+$ such that the new frequency $\omega_+$ at $\xi_+$ is equal to the given frequency $\omega_0(\xi_0)$. Recall the topological degree condition $\mathrm{(A0)}$ and the weak convexity condition $\mathrm{(A1)}$. The former ensures that the parameter $\xi_+$ can be found in the parameter set to keep the frequency unchanged at this KAM step. The latter ensures that the distance between $\xi_+$ and ξ is smaller than the distance between ξ and $\xi_{\nu-1}$, i.e., the sequence of parameters is convergent after an infinite number of iteration steps. The following lemma is crucial to our arguments.

Lemma 3.5. Assume that

\begin{align*} \mathbf{(H4)}:\left\vert\left\|\sum_{j=0}^{\nu}p_{01}^j\right\|\right\vert_{D(s,r)} \lt \mu_0^{\frac{1}{2}}. \end{align*}

There exists $\xi_+\in B_{c\mu^{{1}/{\beta}}}(\xi_\nu)\subset O^o$ such that

(3.36)\begin{align} \omega_+(\xi_+)=\omega_0(\xi_0). \end{align}

Proof. The proof will be completed by an induction on ν. We start with the case ν = 0. It is obvious that $\omega_0(\xi_0)=\omega_0(\xi_0)$. Now assuming that for some ν > 0 we have

(3.37)\begin{align} \omega_i(\xi_i)=\omega_0(\xi_0),~~\xi_i\in B_{c\mu_{i-1}^{{1}/{\beta}}}(\xi_{i-1})\subset O^o,~~~i=1,\cdots,\nu, \end{align}

where $\omega_i(\xi)=\omega_0(\xi)+\sum_{j=0}^{i-1}p_{01}^j(\xi)$. We need to find $\xi_+$ near ξ such that

(3.38)\begin{align} \omega_+(\xi_+)=\omega_0(\xi_0). \end{align}

In view of the property of topological degree, ${\mathbf{(H4)}}$ and $\mathrm{(A0)}$, we have

\begin{align*} \deg\left(\omega_+(\cdot),O^o,\omega_0(\xi_0)\right)& =\deg\left(\omega_0(\cdot)+\sum_{j=0}^{\nu}p_{01}^j(\cdot),O^o,\omega_0(\xi_0)\right)\notag\\ & =\deg\left(\omega_0(\cdot),O^o,\omega_0(\xi_0)\right)\neq0, \end{align*}

i.e., there exists at least a $\xi_+\in O^o$ such that (3.36) holds.

Next, we estimate $\vert\xi_+-\xi_\nu\vert$. (3.14) in lemma 3.3 implies that

\begin{align*} \left\|p_{01}^j\right\|_{C^\beta} \lt c\mu_j,~~~j=0,1,\cdots,\nu, \end{align*}

i.e.,

(3.39)\begin{align} \left\vert p_{01}^j(\xi_+)-p_{01}^j(\xi_\nu)\right\vert \lt c\mu_j\left\vert\xi_+-\xi_\nu\right\vert^\beta,~~\forall\xi_+, \xi\in{O}. \end{align}

According to (3.37) and (3.38), we get

(3.40)\begin{align} \omega_0(\xi_+)-\omega_0(\xi_\nu)+\sum_{j=0}^{\nu-1}\left(p_{01}^{j}(\xi_+)-p_{01}^j{(\xi_\nu)}\right)=-p_{01}^{\nu}(\xi_+). \end{align}

This together with $\mathrm{(A1)}$ and (3.39) yield

(3.41)\begin{align} \left\vert p_{01}^\nu(\xi_+)\right\vert&=\left\vert\omega_0(\xi_+)-\omega_0(\xi_\nu)+\sum_{j=0}^{\nu-1}\left(p_{01}^{j}(\xi_+)-p_{01}^j{(\xi_\nu)}\right)\right\vert\notag\\ &\geq\left\vert\omega_0(\xi_+)-\omega_0(\xi_\nu)\right\vert-\sum_{j=0}^{\nu-1}\left\vert p_{01}^{j}(\xi_+)-p_{01}^j{(\xi_\nu)}\right\vert\notag\\ &\geq\sigma\left\vert\xi_+-\xi_\nu\right\vert^\beta-c\sum_{j=0}^{\nu-1}\mu_j\left\vert\xi_+-\xi_\nu\right\vert^\beta\notag\\ &\geq\left(\sigma-c\sum_{j=0}^{\nu-1}\mu_j\right)\left\vert\xi_+-\xi_\nu\right\vert^\beta. \end{align}

Then, by (3.41) and (3.14) in lemma 3.3, we have

\begin{align*} \vert\xi_+-\xi_\nu\vert^\beta&\leq\frac{p_{01}^\nu(\xi_+)}{\sigma-c\sum_{j=0}^{\nu-1}\mu_j} \lt \frac{c\mu}{\sigma-c\sum_{j=0}^{\nu-1}\mu_j} \lt \frac{2c\mu}{\sigma}, \end{align*}

which implies $\xi_+\in B_{c\mu^{{1}/{\beta}}}(\xi_\nu)$. From $\xi\in O^o$ in (3.37) and the fact that ɛ is small enough (i.e., µ is small enough), we have $B_{c\mu^{{1}/{\beta}}}(\xi_\nu)\subset O^o$.

This completes the proof.

3.2.7. Estimate on $N_+$

Now, we give the estimate on the next step $N_+$.

Lemma 3.6. There is a constant c ₃ such that the following conclusions hold:

(3.42)\begin{align} &\left\vert\xi_{+}-\xi_\nu\right\vert\leq c_3\mu^\frac{1}{\beta}, \end{align}

(3.43)\begin{align} &\left\vert e_+-e\right\vert\leq c_3s^m\mu, \end{align}

(3.44)\begin{align} &\left\vert\left\|\bar h_+-\bar h\right\|\right\vert_{D(s)}\leq c_3s^m\mu. \end{align}

Proof. It is obvious by $\xi_+\in B_{c\mu^{{1}/{\beta}}}(\xi_\nu)$ in lemma 3.5 that (3.42) holds. It follows from (3.32) and (3.33) that (3.43) and (3.44) hold.

3.2.8. Estimate on $\Phi_+$

Recall that F is as in (3.21) with the coefficients and its estimate given by lemma 3.4. Then, we have the following estimate on F.

Lemma 3.7. There is a constant c ₄ such that for all $\vert j\vert+\vert i\vert\leq 2$,

(3.45)\begin{align} &\left\vert\left\| \partial_x^j\partial_y^iF\right\|\right\vert_{\hat D}\leq c_4\gamma_0^{n+m-2|i|}s^{m-\vert i \vert}\mu\Gamma(r-r_+). \end{align}

Proof. By (3.21) and (3.24), we have

\begin{align*} \left\vert\left\| \partial_x^j\partial_y^iF\right\|\right\vert_{\hat D}&\leq \sum_{\vert\imath\vert\leq m,0 \lt \vert k\vert\leq K_+}\vert k\vert^j\left\vert\left\|\partial_y^i(f_{k\imath}y^\imath)\right\|\right\vert_{D(s)}e^{\vert k\vert(r_++\frac{7}{8}(r-r_+))}\\ &\leq c_4\sum_{0 \lt \vert k\vert\leq K_+}\vert k\vert^{2(|i|+1)\tau+2\vert i\vert+1+\vert j\vert}\gamma_0^{n+m-2\vert i\vert}s^{m-\vert i\vert}\mu e^{-\vert k\vert\frac{r-r_+}{8}}\\ &\leq c_4\gamma_0^{n+m-2\vert i\vert}s^{m-\vert i\vert}\mu\Gamma(r-r_+). \end{align*}

This completes the proof.

Lemma 3.8. Assume that

\begin{align*} &{\mathbf{(H5)}}:c_4s^{m-1}\mu\Gamma(r-r_+) \lt \frac{1}{8}(r-r_+),\\ &{\mathbf{(H6)}}:c_4s^m\mu\Gamma(r-r_+) \lt \frac{1}{8}\alpha s. \end{align*}

Then the following conclusions hold.

1. (1) For all $0\leq t\leq 1$, the mappings

   (3.46)\begin{align} \phi_F^t&:D_{\frac{1}{4}\alpha}\rightarrow D_{\frac{1}{2}\alpha} \end{align}

   are well defined.

2. (2) $\Phi_+=\phi_F^1:D_+\rightarrow D(s,r).$

3. (3) There is a constant c ₅ such that

   \begin{align*} \left\vert\left\|\phi_F^t-id\right\|\right\vert_{\tilde{D}}&\leq c_5\mu\Gamma(r-r_+),\\ \left\vert\left\|D\phi_F^t-Id\right\|\right\vert_{\tilde{D}}&\leq c_5\mu\Gamma(r-r_+),\\ \left\vert\left\|D^2\phi_F^t\right\|\right\vert_{\tilde{D}}&\leq c_5\mu\Gamma(r-r_+). \end{align*}

4. (4)

   \begin{align*} \left\vert\left\|\Phi_+-id\right\|\right\vert_{\tilde{D}}&\leq c_5\mu\Gamma(r-r_+),\\ \left\vert\left\|D\Phi_+-Id\right\|\right\vert_{\tilde{D}}&\leq c_5\mu\Gamma(r-r_+),\\ \left\vert\left\|D^2\Phi_+\right\|\right\vert_{\tilde{D}}&\leq c_5\mu\Gamma(r-r_+). \end{align*}

Proof. First, we prove (1). To verify (3.46), we denote $\phi_{F_1}^t$ and $\phi_{F_2}^t$ as the components of $\phi_{F}^t$ in the y and x planes, respectively. Let $X_F=(F_y,-F_x)^\top$ be the vector field generated by F. Then

(3.47)\begin{align} \phi_F^t=id+\int_0^tX_F\circ \phi_F^udu,~~~~0\leq t\leq 1. \end{align}

For any $(y,x)\in D_{\frac{1}{4}\alpha}$, we let $t_*=\sup\{t\in[0,1]:\phi_F^t(y,x)\in D_\alpha\}$. Then for any $0\leq t\leq t_*$, by $(y,x)\in D_{\frac{1}{4}\alpha}$, (3.45) in lemma 3.7, ${\mathbf{(H5)}}$ and ${\mathbf{(H6)}}$, we can get the following estimates:

\begin{align*} \left\vert\left\|\phi_{F_1}^t(y,x)\right\|\right\vert_{D_{\frac{1}{4}\alpha}}&\leq\left\vert y\right\vert+\int_0^t\left\vert\left\|F_x\circ\phi_F^u\right\|\right\vert_{D_{\frac{1}{4}\alpha}}du\\ &\leq \frac{1}{4}\alpha s+c_4s^{m}\mu\Gamma(r-r_+)\\ & \lt \frac{3}{8}\alpha s,\\ \left\vert\left\|\phi_{F_2}^t(y,x)\right\|\right\vert_{D_{\frac{1}{4}\alpha}}&\leq\vert x\vert+\int_0^t\left\vert\left\|F_y\circ\phi_F^u\right\|\right\vert_{D_{\frac{1}{4}\alpha}}du\\ &\leq r_++\frac{1}{8}(r-r_+)+c_4s^{m-1}\mu\Gamma(r-r_+)\\ & \lt r_++\frac{2}{8}(r-r_+). \end{align*}

Thus, $\phi_F^t\in D_{\frac{1}{2}\alpha}\subset D_\alpha$, i.e. $t_*=1$ and (1) holds.

It follows from (1) that (2) holds.

We now prove (3). Using (3.45) in lemma 3.7 and (3.47), we immediately have

\begin{align*} \left\vert\left\|\phi_F^t-id\right\|\right\vert_{\tilde{D}}\leq c_5\mu\Gamma(r-r_+). \end{align*}

By (3.45) in lemma 3.7, (3.47) and Gronwall Inequality, we get

\begin{align*} \left\vert\left\|D\phi_F^t-Id\right\|\right\vert_{\tilde{D}}&\leq\left\vert\left\|\int_0^tDX_F\circ \phi_F^\lambda D\phi_F^\lambda d\lambda\right\|\right\vert_{\tilde{D}}\\ &\leq\int_0^t\left\vert\left\|DX_F\circ \phi_F^\lambda\right\|\right\vert_{\tilde{D}}\left\vert\left\|D\phi_F^\lambda-Id\right\|\right\vert_{\tilde{D}}d\lambda+\int_0^t\left\vert\left\|DX_F\circ \phi_F^\lambda\right\|\right\vert_{\tilde{D}}d\lambda\\ &\leq c_5\mu\Gamma(r-r_+). \end{align*}

It follows from the induction and a similar argument that we have the estimate on the 2-order derivative of $\phi_F^t$, i.e.,

\begin{align*} \left\vert\left\|D^2\phi_F^t\right\|\right\vert_{\tilde{D}}\leq c_5\mu\Gamma(r-r_+). \end{align*}

(4) now follows from (3).

This completes the proof.

3.2.9. Estimate on $P_+$

In the following, we estimate the next step $P_+$.

Lemma 3.9. Assume ${\mathbf{(H1)}}$-${\mathbf{(H6)}}$. Then there is a constant c ₆ such that

(3.48)\begin{equation} \left\vert\left\|P_+\right\|\right\vert_{D_+}\leq c_6\gamma_0^{n+m+2}s^m\mu^2(\Gamma^2(r-r_+)+\Gamma(r-r_+)). \end{equation}

Moreover, if

\begin{align*} {\mathbf{(H7)}}:\mu^\rho(\Gamma^2(r-r_+)+\Gamma(r-r_+))\leq1 \end{align*}

then

(3.49)\begin{equation} \left\vert\left\|P_+\right\|\right\vert_{D_+}\leq \gamma_0^{n+m+2}s_+^m\mu_+. \end{equation}

Proof. By (3.13) and (3.14) in lemma 3.3, (3.45) in lemma 3.7 and lemma 3.8 (3), we have that, for all $0\leq t\leq 1$,

\begin{align*} \left\vert\left\| \{R_t,F\}\circ\phi_F^t\right\|\right\vert_{D_{\frac{1}{4}\alpha}}&\leq c\gamma_0^{n+m+2}s^m\mu^2\Gamma^2(r-r_+),\\ \left\vert\left\| (P-R)\circ\phi_F^1\right\|\right\vert_{D_{\frac{1}{4}\alpha}}&\leq c\gamma_0^{n+m+2}s^m\mu^2\Gamma(r-r_+). \end{align*}

So, by (3.35),

\begin{equation*} \left\vert\left\| P_+\right\|\right\vert_{D_{\frac{1}{4}\alpha}}\leq c\gamma_0^{n+m+2}s^m\mu^2\left(\Gamma^2(r-r_+)+\Gamma(r-r_+)\right). \end{equation*}

By $\mathbf{(H7)}$, we see that

\begin{align*} \left\vert\left\|P_+\right\|\right\vert_{D_+}&\leq 8^mc_0\mu^{1+\rho}s_+^m\mu^{1-2\rho-\frac{m}{m+1}}\gamma_0^{n+m+2}\left(\mu^\rho\left(\Gamma^2(r-r_+)+\Gamma(r-r_+)\right)\right)\\ &\leq\gamma_0^{n+m+2}s_+^m\mu_+, \end{align*}

which implies (3.49).

This completes the proof.

This completes one cycle of KAM steps.

4. Proof of theorem 2.1

4.1. Iteration lemma

In this subsection, we will prove an iteration lemma which guarantees the inductive construction of the transformations in all KAM steps.

Let $r_0,s_0,\alpha_0, \mu_0,H_0,N_0,e_0,\bar h_0,P_0$ be given at the beginning of section 3 and let $D_0=D(s_0,r_0)$, $K_0=0$, $\Phi_0=id$. We define the following sequences inductively for all $\nu=1,2,\cdots$:

\begin{align*} r_\nu&=r_0\left(1-\sum_{i=1}^\nu\frac{1}{2^{i+1}}\right),\\ s_\nu&=\frac{1}{8}\alpha_{\nu-1}s_{\nu-1},\\ \alpha_\nu&=\mu_\nu^{2\rho}=\mu_\nu^{\frac{1}{m+1}},\\ \mu_\nu&=8^mc_0\mu_{\nu-1}^{1+\rho},\\ K_\nu&=\left(\left[\log\left(\frac{1}{\mu_{\nu-1}}\right)\right]+1\right)^{3\eta},\\ \tilde{D}_\nu&=D\left(\frac{1}{2} s_\nu, r_\nu+\frac{6}{8}\left(r_{\nu-1}-r_\nu\right)\right). \end{align*}

Lemma 4.1. Denote

\begin{align*} \mu_*=\frac{\mu_0}{\left(M^*+2\right)^{m-1}K_1^{(\tau+1)(m-1)}}. \end{align*}

If ɛ is small enough, then the KAM step described on the above is valid for all $\nu=0,1,\cdots$, resulting the sequences

\begin{align*}e_\nu, \bar h_\nu, N_\nu, P_\nu, H_\nu, \Phi_\nu,\end{align*}

$\nu=1,2,\cdots,$ with the following properties:

1. (1)

   (4.1)\begin{align} \left\vert e_{\nu+1}-e_{\nu}\right\vert&\leq\frac{\mu_*^{\frac{1}{2}}}{2^{\nu}}, \end{align}

   (4.2)\begin{align} \left\vert e_\nu-e_{0}\right\vert&\leq2\mu_*^{\frac{1}{2}}, \end{align}

   (4.3)\begin{align} \left\vert\left\|\bar h_{\nu+1}-\bar h_{\nu}\right\|\right\vert_{D(s_{\nu})}&\leq\frac{\mu_*^{\frac{1}{2}}}{2^{\nu}}, \end{align}

   (4.4)\begin{align} \left\vert\left\|\bar h_\nu-\bar h_{0}\right\|\right\vert_{D(s_\nu)}&\leq2\mu_*^{\frac{1}{2}}, \end{align}

   (4.5)\begin{align} \left\vert\left\|P_\nu\right\|\right\vert_{D(s_\nu,r_\nu)}&\leq\frac{\mu_*^{\frac{1}{2}}}{2^{\nu}}, \end{align}

   (4.6)\begin{align} \left\vert\xi_{\nu+1}-\xi_{\nu}\right\vert&\leq\left(\frac{\mu_*^{\frac{1}{2}}}{2^{\nu}}\right)^{\frac{1}{\beta}}. \end{align}

2. (2) $\Phi_{\nu+1}:\tilde{D}_{\nu+1}\rightarrow \tilde{D}_{\nu}$ is symplectic, and

   (4.7)\begin{align} \left\vert\left\|\Phi_{\nu+1}-id\right\|\right\vert_{\tilde{D}_{\nu+1}}\leq\frac{\mu_*^{\frac{1}{2}}}{2^{\nu}}. \end{align}

   Moreover, on $D_{\nu+1}$,

   \begin{equation*} H_{\nu+1}=H_\nu\circ\Phi_{\nu+1}=N_{\nu+1}+P_{\nu+1}. \end{equation*}

Proof. The proof amounts to the verification of $\mathbf{(H1)}$-$\mathbf{(H7)}$ for all ν. For simplicity, we let $r_0=1$. It follows from ɛ small enough that µ ₀ is small. So, we see that $\mathbf{(H2)}$, $\mathbf{(H4)}$-$\mathbf{(H7)}$ hold for ν = 0. From (3.2), $\mathbf{(H3)}$ holds for ν = 0. According to the definition of µ_ν, we see that

(4.8)\begin{equation} \mu_\nu=(8^mc_0)^{\frac{(1+\rho)^\nu-1}{\rho}}\mu_0^{(1+\rho)^\nu}. \end{equation}

Let $\zeta\gg1$ be fixed and µ ₀ be small enough so that

(4.9)\begin{equation} \mu_0 \lt \left(\frac{1}{8^mc_0\zeta}\right)^{\frac{1}{\rho}} \lt 1. \end{equation}

Then

(4.10)\begin{align} \mu_1&=8^mc_0\mu_0^{1+\rho} \lt \frac{1}{\zeta}\mu_0 \lt 1,\notag\\ \mu_2&=8^mc_0\mu_1^{1+\rho} \lt \frac{1}{\zeta}\mu_1 \lt \frac{1}{\zeta^2}\mu_0,\notag\\ \vdots\notag\\ \mu_\nu&=8^mc_0\mu_{\nu-1}^{1+\rho} \lt \cdots \lt \frac{1}{\zeta^\nu}\mu_0. \end{align}

Denote

\begin{equation*} \Gamma_\nu=\Gamma(r_\nu-r_{\nu+1}). \end{equation*}

We notice that

(4.11)\begin{equation} \frac{r_\nu-r_{\nu+1}}{r_0}=\frac{1}{2^{\nu+2}}. \end{equation}

Since

\begin{align*} \Gamma_\nu&\leq\int_1^\infty t^{6\tau+7}e^{-\frac{t}{2^{\nu+5}}}dt\\ &\leq(6\tau+7)!2^{(\nu+5)(6\tau+7)}, \end{align*}

it is obvious that if ζ is large enough, then

\begin{align*} \mu_\nu^\rho\Gamma_\nu^i \lt \mu_\nu^\rho(\Gamma_\nu^i+\Gamma_\nu)\leq1,~~~i=1,2, \end{align*}

which implies that $\mathbf{(H7)}$ holds for all $\nu\geq1$, and

(4.12)\begin{equation} \mu_\nu\Gamma_\nu\leq\mu_\nu^{1-\rho}\leq\frac{\mu_0^{1-\rho}}{\zeta^{(1-\rho)\nu}}. \end{equation}

By (4.11) and (4.12), it is easy to verify that $\mathbf{(H5)}$ and $\mathbf{(H6)}$ hold for all $\nu\geq1$ as ζ is large enough and ɛ is small enough.

By (3.14) in lemma 3.3 and (4.10), we have

\begin{align*} \left\vert\left\|\sum_{j=0}^{\nu}p_{01}^j\right\|\right\vert_{D(s_\nu,r_\nu)} \lt c\sum_{j=0}^{\nu}\mu_j \lt c\sum_{j=0}^{\nu}\frac{1}{\zeta^j}\mu_0 \lt c\mu_0^{\frac{1}{2}}, \end{align*}

which implies $\mathbf{(H4)}$.

To verify $\mathbf{(H3)}$, we observe by (4.8) and (4.10) that

\begin{equation*} \frac{1}{4}\left(M^*+2\right)\mu_{\nu-1}^{2\rho}K_{\nu+1}^{\tau+1} \lt \frac{1}{2^{\nu+2}}, \end{equation*}

as ζ is large enough. Then

(4.13)\begin{align} 2\left(M^*+2\right)s_\nu K_{\nu+1}^{\tau+1}&\leq\frac{s_{\nu-1}}{4}\left(M^*+1\right)\mu_{\nu-1}^{2\rho}K_{\nu+1}^{\tau+1} \end{align}

(4.14)\begin{align} &\leq\frac{s_0}{2^{\nu+2}} \lt \frac{\gamma_0}{2^{\nu+2}} \lt \gamma_0, \end{align}

which verifies $\mathbf{(H3)}$ for all $\nu\geq1$.

Let $\zeta^{1-\rho}\geq2$ in (4.9), (4.10). We have that for all $\nu\geq1$

(4.15)\begin{align} c_0\mu_\nu&\leq\frac{\mu_0}{2^\nu}\leq\frac{\mu_*^\frac{1}{2}}{2^\nu}, \end{align}

(4.16)\begin{align} c_0\mu_\nu\Gamma_\nu&\leq\frac{\mu_0^{1-\rho}}{2^\nu}\leq\frac{\mu_*^\frac{1}{2}}{2^\nu}, \end{align}

(4.17)\begin{align} c_0s_\nu^{m-1}\mu_\nu&\leq\frac{\mu_0^{1+2\rho(m-1)}s_0^{m-1}}{2^{\nu+3}}\leq\frac{\mu_*}{2^\nu}. \end{align}

The verification of $\mathbf{(H2)}$ follows from $(4.15)$ and an induction application of $(3.44)$ in lemma 3.6 for all $\nu=0,1,\cdots.$

Since $\left(1+\rho\right)^\eta \gt 2$, we have

\begin{align*} \frac{1}{2^{\nu+6}}\left(\left[\log\frac{1}{\mu}\right]+1\right)^\eta&\geq\frac{1}{2^{\nu+6}}\left(\left(1-\left(1+\rho\right)^\nu\right)\log\left(8^mc_0\right)-\left(1+\rho\right)^\nu\log\mu_0\right)^\eta\\ &\geq-\frac{1}{2^{\nu+6}}\left(1+\rho\right)^{\eta\nu}\left(\log\mu_0\right)^\eta\geq1. \end{align*}

It follows from above that

\begin{align*} &\log\left(n+1\right)!+\left(\nu+6\right)n\log2+3n\eta\log\left(\left[\log\frac{1}{\mu}\right]+1\right)-\frac{1}{2^{\left(\nu+6\right)}}\left(\left[\log\frac{1}{\mu}\right]+1\right)^{3\eta}\\ &\leq\log\left(n+1\right)!+\left(\nu+6\right)n\log2+3n\eta\log\left(\log\frac{1}{\mu}+2\right)-\left(\log\frac{1}{\mu}\right)^{2\eta}\\ &\leq-\log\frac{1}{\mu}, \end{align*}

as µ is small, which is ensured by making ɛ small. Thus,

\begin{equation*} \int_{K_{\nu+1}}^\infty t^{n}e^{-\frac{t}{2^{\nu+6}}}dt\leq\left(n+1\right)!2^{\left(\nu+6\right)n}K_{\nu+1}^{n}e^{-\frac{K_{\nu+1}}{2^{\nu+6}}}\leq \mu, \end{equation*}

i.e. $\mathbf{(H1)}$ holds.

Above all, the KAM steps described in section 3 are valid for all ν, which give the desired sequences stated in the lemma.

Now, (4.1) and (4.3) follow from lemma 3.6, (4.15) and (4.17); by adding up (4.1) and (4.3) for all $\nu=0,1,\cdots$, we can get (4.2) and (4.4); (4.5) follows from (3.49) in lemma 3.9 and (4.15); (4.6) follows from (3.42) in lemma 3.6 and (4.15); (2) follows from lemma 3.8.

This completes the proof.

4.2. Convergence

The convergence is standard. For the sake of completeness, we briefly give the framework of proof. Let

\begin{align*} \Psi^\nu=\Phi_1\circ\Phi_2\circ\cdots\circ\Phi_\nu,~~~~\nu=1,2,\cdots. \end{align*}

By lemma 4.1, we have

\begin{align*} D_{\nu+1}&\subset D_\nu,\\ \Psi^\nu&:\tilde{D}_\nu\rightarrow \tilde{D}_0,\\ H_0\circ\Psi^\nu&=H_\nu=N_\nu+P_\nu,\\ N_\nu&=e_\nu+\left\langle\omega_\nu\left(\xi\right),y\right\rangle+\bar h_\nu(y,\xi), \end{align*}

$\nu=0,1,\cdots,$ where $\Psi^0=id$, $\omega_\nu(\xi)=\omega_0(\xi)+\sum_{j=0}^{\nu-1}p_{01}^j\left(\xi\right)$. Using (4.7) and the identity

\begin{align*} \Psi^\nu=id+\sum_{i=1}^\nu\left(\Psi^i-\Psi^{i-1}\right), \end{align*}

it is easy to verify that $\Psi^\nu$ is uniformly convergent and denote the limitation by $\Psi^\infty$.

In view of lemma 4.1, it is obvious to see that e_ν, $\bar{h}_\nu$ and ξ_ν converge uniformly about ν, and denote its limitation by $e_\infty$, $\bar{h}_\infty$ and $\xi_\infty$. By lemma 3.5, we have

(4.18)\begin{align} &\omega_1(\xi_1)=\omega_0\left(\xi_1\right)+p_{01}^0\left(\xi_1\right)=\omega_0\left(\xi_0\right),\notag\\ &\omega_2(\xi_2)=\omega_0\left(\xi_2\right)+p_{01}^0\left(\xi_2\right)+p_{01}^1\left(\xi_2\right)=\omega_0\left(\xi_0\right),\notag\\ &~~~\vdots\notag\\ &\omega_\nu(\xi_\nu)=\omega_0\left(\xi_\nu\right)+p_{01}^0\left(\xi_\nu\right)+\cdots+p_{01}^{\nu-1}\left(\xi_\nu\right)=\omega_0\left(\xi_0\right). \end{align}

Taking limits at both sides of (4.18), we get

(4.19)\begin{align} \omega_\infty(\xi_\infty)=\omega_0\left(\xi_\infty\right)+\sum_{j=0}^{\infty}p_{01}^j\left(\xi_\infty\right)=\omega_0(\xi_0). \end{align}

Then, on $D(\frac{s_0}{2})$, N_ν converge uniformly to

\begin{align*} N_\infty&=e_\infty+\left\langle\omega_\infty(\xi_\infty),y\right\rangle+\bar{h}_\infty\left(y,\xi_\infty\right).\end{align*}

Hence, on $D\left(\frac{s_0}{2},\frac{r_0}{2}\right)$,

\begin{align*} P_\nu=H_0\circ\Psi^\nu-N_\nu \end{align*}

converge uniformly to

\begin{equation*} P_\infty=H_0\circ\Psi^\infty-N_\infty. \end{equation*}

Since

\begin{align*} \left\vert\left\|P_\nu\right\|\right\vert_{D_\nu}\leq\gamma_0^{n+m+2}s_\nu^m\mu_\nu, \end{align*}

by (4.15), we have that it converges to 0 as $\nu\rightarrow\infty$. So, on $D(0,\frac{r_0}{2})$,

\begin{align*} J\nabla P_\infty=0. \end{align*}

Thus, for the given $\xi_0\in O$, the Hamiltonian

\begin{align*} H_\infty=N_\infty+ P_\infty \end{align*}

admits an analytic, quasi-periodic, invariant n-torus $\mathbb{T}^n\times\{0\}$ with the Diophantine frequency $\omega(\xi_0)$, which is the corresponding unperturbed toral frequency.

Recall

\begin{align*}\hat O& =\left\{\hat\xi_0\in O:|\omega(\hat \xi_0)-\omega(\xi_0)| \lt d,|\langle k,\omega(\hat\xi_0)\rangle|\geq\frac{\gamma}{|k|^\tau},k\in\mathbb{Z}^n\backslash\{0\}\right\},\notag\\d&=\min_{\xi\in\partial O}\left\{\;|\omega(\xi)-\omega(\xi_0)|\;\right\}.\end{align*}

For given $\hat\xi_0,\tilde\xi_0\in\hat O$, similar to (4.19), we obtain

(4.20)\begin{align} \omega(\xi_\infty(\hat\xi_0))+\sum_{j=0}^{\infty}p_{01}^j(\xi_\infty(\hat\xi_0))=\omega(\hat\xi_0), \end{align}

(4.21)\begin{align} \omega(\xi_\infty(\tilde\xi_0))+\sum_{j=0}^{\infty}p_{01}^j(\xi_\infty(\tilde\xi_0))=\omega(\tilde\xi_0). \end{align}

(4.20) minus (4.21), we get

\begin{align*} \omega(\xi_\infty(\hat\xi_0))-\omega(\xi_\infty(\tilde\xi_0))+\sum_{j=0}^{\infty}p_{01}^j(\xi_\infty(\hat\xi_0))-p_{01}^j(\xi_\infty(\tilde\xi_0))=\omega(\hat\xi_0)-\omega(\tilde\xi_0). \end{align*}

Then

\begin{align*} &\left|\omega(\hat\xi_0)-\omega(\tilde\xi_0)\right|\\ &=\left|\omega(\xi_\infty(\hat\xi_0))-\omega(\xi_\infty(\tilde\xi_0))+\sum_{j=0}^{\infty}p_{01}^j(\xi_\infty(\hat\xi_0))-p_{01}^j(\xi_\infty(\tilde\xi_0))\right|\\ &\geq \left|\omega(\xi_\infty(\hat\xi_0))-\omega(\xi_\infty(\tilde\xi_0))\right|-\left|\sum_{j=0}^{\infty}p_{01}^j(\xi_\infty(\hat\xi_0))-p_{01}^j(\xi_\infty(\tilde\xi_0))\right|\\ &\geq\sigma\left\vert\xi_\infty(\hat\xi_0)-\xi_\infty(\tilde\xi_0)\right\vert^\beta-c\sum_{j=0}^{\infty}\mu_j\left\vert\xi_\infty(\hat\xi_0)-\xi_\infty(\tilde\xi_0)\right\vert^\beta\notag\\ &\geq\left(\sigma-c\sum_{j=0}^{\infty}\mu_j\right)\left\vert\xi_\infty(\hat\xi_0)-\xi_\infty(\tilde\xi_0)\right\vert^\beta\\ &\geq\frac{\sigma}{2}\left\vert\xi_\infty(\hat\xi_0)-\xi_\infty(\tilde\xi_0)\right\vert^\beta, \end{align*}

i.e.,

(4.22)\begin{align} \left\vert\xi_\infty(\hat\xi_0)-\xi_\infty(\tilde\xi_0)\right\vert\leq\left(\frac{2}{\sigma}\right)^{\frac{1}{\beta}}\left|\omega(\hat\xi_0)-\omega(\tilde\xi_0)\right|^{\frac{1}{\beta}}. \end{align}

It follows from the Hölder continuity of ω and (4.22) that $\xi_\infty(\hat\xi_0)$ is Hölder continuous about $\hat\xi_0\in \hat O$. Using the above proof process and replacing ξ ₀ by $\hat\xi_0$, we can get an invariant torus with the Diophantine frequency $\omega(\hat\xi_0)$.

5. Proof of theorem 2.5 and theorem 2.8

First, we briefly outline the proof framework of theorem 2.5 as it follows the KAM step in section 3. Here, we mainly highlight two major differences from the proof of theorem 2.1. The first difference is that we utilize homotopy invariance and excision of the topological degree to keep the frequency unchanged during the iteration process, rather than selecting parameters. This is because we consider a Hamiltonian, not a family of Hamiltonians. The second difference is that the transformation is defined on a smaller domain because we treat the action variable as a parameter, and the translation of the parameter is equivalent to the action variable’s. We need the integer $m\geq\beta+1$ to ensure that the translated action variable y does not run out of the given domain.

5.1. Proof of theorem 2.5

In this section, we will describe the translation of action variable and state how the frequency can be preserved in the iterative process, which are different from subsection 3.2.6.

Let $\xi_0\in{G}$ be fixed as statement $\mathrm{(A0)}$. The Taylor expansion of Hamiltonian (1.1) about ξ ₀ reads

\begin{equation*} H(y,x,\xi_0)=e_0+\left\langle\omega_0(\xi_0),y-\xi_0\right\rangle+\bar{h}(y-\xi_0)+\varepsilon P(y,x,\varepsilon), \end{equation*}

where $e_0=h(\xi_0)$, $\omega_0(\xi_0)=\nabla h(\xi_0)$, $\bar{h}(y-\xi_0)=O(\left\vert y-\xi_0\right\vert^2)$. Using the transformation $(y-\xi_0)\rightarrow y$ in the above, we have

(5.1)\begin{equation} H(y,x,\xi_0)=e_0+\langle\omega_0(\xi_0),y\rangle+\bar{h}(y,\xi_0)+\varepsilon P(y,x,\xi_0,\varepsilon), \end{equation}

where (y, x) lies in a complex neighbourhood $D(s,r)$. Denote

\begin{align*} H_0&=: H(y,x,\xi_0)=N_0+P_0,\\ N_0&=:e_0+\langle\omega_0(\xi_0),y\rangle+\bar{h}_0(y,\xi_0),\\ P_0&=:\varepsilon P(y,x,\xi_0,\varepsilon). \end{align*}

For convenience, we will omit the index for all quantities of the present KAM step (at νth step) and use + to index all quantities in the next KAM step (at $(\nu+1)$th step). Now, suppose that at νth step, we have arrived at the following real analytic Hamiltonian:

(5.2)\begin{equation} \begin{aligned} H&=N+P,~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\\ N&=e+\langle\omega(\xi),y\rangle+\bar{h}(y,\xi). \end{aligned} \end{equation}

First, we shall construct a symplectic transformation as the time-1 map $\phi_{F}^1$ such that it transforms the Hamiltonian H into the following form:

\begin{align*} H\circ\phi_{F}^1=:\bar N_++\bar P_+. \end{align*}

The concrete details see subsection 3.2.4

Next, we will construct a translation so as to keep the frequency unchanged. Consider the translation

\begin{equation*}\phi:x\rightarrow x,~~~~~y\rightarrow y+\xi_+-\xi,\end{equation*}

where $\xi_+$ is to be determined. Let

\begin{equation*}\Phi_+=\phi_F^1\circ\phi.\end{equation*}

Then

(5.3)\begin{align} H\circ\Phi_+&=N_++P_+,\notag\\ N_+&=\bar N_+\circ\phi=e_++\langle\omega_+(\xi_+),y\rangle+\bar h_+(y,\xi_+), \end{align}

(5.4)\begin{align} P_+&=\bar P_+\circ\phi, \end{align}

where

(5.5)\begin{align} e_+&=e+\langle\omega(\xi),\xi_+-\xi\rangle+\bar h(\xi_+-\xi)+[R](\xi_+-\xi), \end{align}

(5.6)\begin{align} \omega_+&=\omega(\xi)+\nabla \bar h(\xi_+-\xi)+\nabla[R](\xi_+-\xi), \end{align}

(5.7)\begin{align} \bar h_+&=\bar h(y+\xi_+-\xi)-\bar h(\xi_+-\xi)-\left\langle\nabla \bar h(\xi_+-\xi),y\right\rangle +[R](y+\xi_+-\xi)\notag\\ &~~~~-[R](\xi_+-\xi)-\left\langle\nabla[R](\xi_+-\xi),y\right\rangle. \end{align}

As in subsection 3.2.6, we will show that the frequency can be preserved in the iteration process. The following lemma is crucial to our arguments.

Lemma 5.1. There exists $\xi_+\in B_{s\mu^{{1}/{\beta}}}(\xi)$ such that

(5.8)\begin{align} \omega_+(\xi_+)=\omega(\xi)=\cdots=\omega_0(\xi_0). \end{align}

Proof. The proof will be completed by an induction on ν. We begin with the case ν = 0. It is obvious that $\omega_0(\xi_0)=\omega_0(\xi_0)$. Now supposing that for some $\nu\geq0$ we have

\begin{align*} \omega_i(\xi_i)&=\omega_{i-1}(\xi_{i-1}),~~~~~\xi_i\in B_{s_{i-1}\mu_{i-1}^{{1}/{\beta}}}\left(\xi_{i-1}\right), \end{align*}

where $i=1,2,\cdots,\nu.$ Then, we need to find $\xi_+$ near ξ such that $\omega_+(\xi_+)=\omega(\xi)$. In view of (5.6), we observe that

(5.9)\begin{align} \left\vert\omega_+(y)-\omega(y)\right\vert=O(\mu). \end{align}

We split

(5.10)\begin{align} \omega_+(y)-\omega(\xi)&=(\omega(y)-\omega(\xi)) +(\omega_+(y)-\omega(y)). \end{align}

Consider homotopy $H_t(y):[0,1]\times G\rightarrow \mathbb{R}^n$,

\begin{align*} H_t(y)&=:\left(\omega(y)-\omega(\xi)\right) +t\left(\omega_+(y)-\omega(y)\right). \end{align*}

For any $y\in\partial G$, $t\in[0,1]$, by $\mathrm{(A1)}$, we have that

\begin{align*} \left\vert H_t(y)\right\vert&\geq\vert\omega(y)-\omega(\xi)\vert -\vert\omega_+(y)-\omega(y)\vert\geq\vert\omega_0(y)-\omega_{0}(\xi_0)\vert-\sum_{i=0}^\nu\left\vert\omega_{i+1}(y)-\omega_i(y)\right\vert\\ &\geq\sigma\vert y-\xi_0\vert^\beta-\sum_{i=0}^{\nu}\gamma_0^{n+m+2}s_i^{m-1}\mu_i \gt \frac{\sigma\delta^\beta}{2}, \end{align*}

where $\delta:=\min\{\vert y-\xi_0\vert, \forall y\in\partial G\}$.

So, it follows from homotopy invariance and $\mathrm{(A0)}$ that

\begin{align*} \deg\left(H_1(\cdot),G^o,0\right)=\deg\left(H_0(\cdot),G^o,0\right)\neq0. \end{align*}

We note by $\mathrm{(A1)}$, (5.9) and (5.10) that for any $y\in G\backslash B_{s\mu^{{1}/{\beta}}}(\xi)$,

(5.11)\begin{align} \left\vert H_1(y)\right\vert&=\vert\omega_+(y)-\omega(\xi)\vert\geq \vert y-\xi\vert^\beta-c_1\gamma_0^{n+m+2}s^{m-1}\mu\notag\\ &\geq s^\beta \mu-c_1\gamma_0^{n+m+2}s^{m-1}\mu\geq\frac{s^\beta\mu}{2}, \end{align}

where the last equality follows from $m\geq \beta+1$.

Hence, by excision, we have that

\begin{align*} \deg\left(H_1(\cdot),B_{s\mu^{{1}/{\beta}}}(\xi),0\right)=\deg\left(H_1(\cdot),G^o,0\right)\neq0, \end{align*}

i.e., there exists at least a $\xi_+\in B_{s\mu^{{1}/{\beta}}}(\xi)$, such that $H_1(\xi_+)=0,$ i.e.,

\begin{equation*}\omega_+(\xi_+)=\omega(\xi),\end{equation*}

which implies (5.8).

This completes the proof.

In the following, we prove

(5.12)\begin{align} \phi:D_{\frac{1}{8}\alpha}\rightarrow D_{\frac{1}{4}\alpha}. \end{align}

Recall that $m\geq\beta+1$ and $\alpha=\mu^{\frac{1}{m+1}}$, we have

(5.13)\begin{align} cs\mu^{\frac{1}{\beta}} \lt \frac{1}{8}\alpha s. \end{align}

For $\forall (y,x)\in D_{\frac{1}{8}\alpha}$, we note by $\xi_+\in B_{s\mu^{{1}/{\beta}}}(\xi)$ in lemma 5.1 and (5.13) that

\begin{equation*}\left\vert y+\xi_+-\xi\right\vert \lt \vert y\vert+\vert \xi_+-\xi\vert \lt \frac{1}{8}\alpha s+cs\mu^{\frac{1}{\beta}} \lt \frac{1}{4}\alpha s,\end{equation*}

which implies (5.12).

Next, we prove theorem 2.8 by a direct method.

5.2. Proof of theorem 2.8

(1) The unperturbed motion of (1.1) is described by the equation

\begin{equation*} \left\{ \begin{array}{ll} \dot{y}=0,\\ \dot{x}= h'(y). \end{array} \right. \end{equation*}

The flow is $x=h'(y)t+x_0,y\in G$, where x ₀ is an initial condition. Notice that

\begin{align*} h''(0)=0, \end{align*}

i.e., h(y) is degenerate at $\xi_0=0$. Obviously, by simple calculation, we get

\begin{align*} \deg\left(h'(y)-h'(0),B_\delta(0),0\right)=0, \end{align*}

i.e., $\mathrm{(A0)}$ fails, then theorem 2.6 is not applicable.

Note that the perturbed motion equation is

\begin{equation*} \left\{ \begin{array}{ll} \dot{y}=0,\\ \dot{x}=h'(y)+\varepsilon P'(y). \end{array} \right. \end{equation*}

The flow is $x=\left(h'(y)+\varepsilon P'(y)\right)t+x_1,~y\in G$, where x ₁ is an initial condition. To ensure the frequency is equal to $h'(0)$, we need to find a solution of the following equation in G:

\begin{align*} h'(y)+\varepsilon P'(y)=h'(0), \end{align*}

i.e.,

(5.14)\begin{align} g'(y)=-\varepsilon P'(y). \end{align}

Notice that the Taylor expansion of $g'(y)$ at $\xi_0=0$ is

\begin{equation*}g'(y)=g'(0)+g''(0)y+\cdots+g^{2\ell+1}(0)y^{2\ell}+o(y^{2\ell}),\end{equation*}

then the equation (5.14) is equivalent to

\begin{equation*}g^{2\ell+1}(0)y^{2\ell}+o(y^{2\ell})=-\varepsilon P'(y),\end{equation*}

which is solvable provided that $\varepsilon P'(y)\, sign \left(g^{2\ell+1}(0)\right) \lt 0$. So the perturbed system admits at least two invariant tori with frequency $\omega=h'(0)$ for the small enough perturbation satisfying $\varepsilon P'(y)\, sign \left(g^{2\ell+1}(0)\right) \lt 0$. Conversely, if $\varepsilon P'(y)\, sign \left(g^{2\ell+1}(0)\right) \gt 0$, the unperturbed invariant torus with frequency $\omega=h'(0)$ will be destroyed.

(2) Note that h(y) is degenerate in $\xi_0=0$. Obviously, by simple calculation, we get

\begin{align*} \deg\left(h'(y)-h'(0),B_\delta(0),0\right)\neq0. \end{align*}

Then, by theorem 2.6, the above persistence result hold. In addition, we can also directly prove this result. Similarly, we need to solve the following equation in G:

\begin{align*} h'(y)+\varepsilon P'(y)=h'(0), \end{align*}

i.e.,

(5.15)\begin{align} g'(y)=-\varepsilon P'(y). \end{align}

Notice that the Taylor expansion of $g'(y)$ at $\xi_0=0$ is

\begin{equation*}g'(y)=g'(0)+g''(0)y+\cdots+g^{2\ell+2}(0)y^{2\ell+1}+o\left(y^{2\ell+1}\right),\end{equation*}

then the equation (5.15) is equivalent to

\begin{equation*}g^{2\ell+2}\left(0\right)y^{2\ell+1}+o\left(y^{2\ell+1}\right)=-\varepsilon P'(y),\end{equation*}

whose solution always exists in G for any small enough perturbation. Hence, the perturbed system admits an invariant torus with frequency $\omega=h'(0)$ for any small enough perturbation.