1: \documentclass{conm-p-l}

2: 

3: 

4: \copyrightinfo{2006}{International Press}

5: 

6: \setcounter{page}{235}

7: 

8: 

9: 

10: \usepackage{graphics}

11: 

12: \newtheorem{theorem}{Theorem}[section]

13: \newtheorem{lemma}[theorem]{Lemma}

14: \newtheorem{corollary}[theorem]{Corollary}

15: 

16: \theoremstyle{definition}

17: \newtheorem{definition}[theorem]{Definition}

18: \newtheorem{example}[theorem]{Example}

19: \newtheorem{xca}[theorem]{Exercise}

20: 

21: \theoremstyle{remark}

22: \newtheorem{remark}[theorem]{Remark}

23: 

24: \numberwithin{equation}{section}

25: 

26: %    Absolute value notation

27: \newcommand{\abs}[1]{\lvert#1\rvert}

28: 

29: %    Blank box placeholder for figures (to avoid requiring any

30: %    particular graphics capabilities for printing this document).

31: \newcommand{\blankbox}[2]{%

32:   \parbox{\columnwidth}{\centering

33: %    Set fboxsep to 0 so that the actual size of the box will match the

34: %    given measurements more closely.

35:     \setlength{\fboxsep}{0pt}%

36:     \fbox{\raisebox{0pt}[#2]{\hspace{#1}}}%

37:   }%

38: }

39: 

40: \newcommand{\LL}{{\mathcal L}}

41: \newcommand{\tla}{\tilde{\lambda}}

42: \newcommand{\im}{\mathop{\rm Im}\nolimits}

43: \newcommand{\hk}{\hat{k}}

44: \newcommand{\Z}{\ZZ^2/\{0\}}

45: \newcommand{\E}{{\mathcal E}}

46: \newcommand{\HH}{{\mathcal H}}

47: \newcommand{\cS}{{\mathcal S}}

48: \newcommand{\q}{\vec{q}}

49: \newcommand{\vq}{\vec{q}}

50: \newcommand{\p}{\vec{p}}

51: \newcommand{\CC}{{\mathbb C}}

52: \newcommand{\RR}{{\mathbb R}}

53: \newcommand{\ZZ}{{\mathbb Z}}

54: \newcommand{\PP}{{\mathbb P}}

55: \newcommand{\f}{\vec{f}}

56: \newcommand{\F}{{\mathcal F}}

57: \newcommand{\e}{\epsilon}

58: \newcommand{\vth}{\vartheta}

59: \newcommand{\vphi}{\varphi}

60: \newcommand{\U}{{\mathcal U}}

61: \newcommand{\N}{{\mathcal N}}

62: \newcommand{\cq}{\tilde{q}}

63: \newcommand{\tDl}{\tilde{\Delta}}

64: \newcommand{\tQ}{\tilde{Q}}

65: \newcommand{\vQ}{\vec{Q}}

66: \newcommand{\A}{{\mathcal A}}

67: \newcommand{\B}{{\mathcal B}}

68: \newcommand{\C}{{\mathcal C}}

69: \renewcommand{\k}{\kappa}

70: \newcommand{\ga}{\gamma}

71: \newcommand{\hga}{\hat{\gamma}}

72: \newcommand{\Ga}{\Gamma}

73: \newcommand{\ve}{\bf{e}}

74: \newcommand{\we}{{\hat{{\bf{e}}}}}

75: \newcommand{\he}{\hat{e}}

76: \newcommand{\vv}{\vec{v}}

77: \newcommand{\hcS}{\hat{\cS}}

78: \newcommand{\tcS}{\tilde{\cS}}

79: \newcommand{\hS}{\hat{S}}

80: \newcommand{\tS}{\tilde{S}}

81: \newcommand{\hD}{\hat{D}}

82: \newcommand{\hH}{\hat{H}}

83: \newcommand{\hJ}{\hat{J}}

84: \newcommand{\tD}{\tilde{D}}

85: \newcommand{\tC}{\tilde{C}}

86: \newcommand{\dl}{\delta}

87: \newcommand{\Dl}{\Delta}

88: \renewcommand{\th}{\theta}

89: \newcommand{\Th}{\Theta}

90: \newcommand{\ra}{\rightarrow}

91: \newcommand{\al}{\alpha}

92: \newcommand{\be}{\beta}

93: \newcommand{\sg}{\sigma}

94: \newcommand{\Sg}{\Sigma}

95: \newcommand{\bM}{\bar{M}}

96: \newcommand{\pa}{\partial}

97: \newcommand{\z}{\zeta}

98: \newcommand{\hQ}{\hat{Q}}

99: \newcommand{\hq}{\hat{q}}

100: \newcommand{\hv}{\hat{v}}

101: \newcommand{\hw}{\hat{w}}

102: \newcommand{\hx}{\hat{x}}

103: \newcommand{\bv}{\bar{v}}

104: \newcommand{\bw}{\bar{w}}

105: \newcommand{\La}{\Lambda}

106: \newcommand{\tLa}{\tilde{\Lambda}}

107: \newcommand{\la}{\lambda}

108: \newcommand{\bq}{\bar{q}}

109: \newcommand{\bp}{\bar{p}}

110: \newcommand{\bQ}{\bar{Q}}

111: \newcommand{\bE}{\bar{E}}

112: \newcommand{\nid}{\noindent}

113: \newcommand{\rc}{S_\omega}

114: \newcommand{\hrc}{\hat{S}_\omega}

115: \newcommand{\bW}{\bar{W}}

116: \newcommand{\hN}{\hat{N}}

117: \newcommand{\hF}{\hat{F}}

118: \newcommand{\bB}{\bar{B}}

119: \newcommand{\hG}{\hat{G}}

120: \newcommand{\tF}{\tilde{F}}

121: \newcommand{\om}{\omega}

122: \newcommand{\Om}{\Omega}

123: \newcommand{\na}{\nabla}

124: \newcommand{\lag}{\langle}

125: \newcommand{\rag}{\rangle}

126: \newcommand{\tx}{\tilde{x}}

127: \newcommand{\tq}{\tilde{q}}

128: \newcommand{\tE}{\tilde{E}}

129: \newcommand{\tv}{\tilde{v}}

130: \newcommand{\tth}{\tilde{\theta}}

131: \newcommand{\cH}{{\mathcal H}}

132: \newcommand{\W}{{\mathcal W}}

133: \renewcommand{\O}{{\mathcal O}}

134: \newcommand{\hz}{\hat{z}}

135: 

136: 

137: 

138: %Commuting Diagram Need

139: \def\maprightu#1{\smash{

140:     \mathop{\longrightarrow}\limits^{#1}}}

141: \def\maprightd#1{\smash{

142:     \mathop{\longrightarrow}\limits_{#1}}}

143: \def\mapdownl#1{

144:     \llap{$\vcenter{\hbox{$\scriptstyle#1$}}$}\Big\downarrow}

145: \def\mapdownr#1{\Big\downarrow

146:     \rlap{$\vcenter{\hbox{$\scriptstyle#1$}}$}}

147: \def\mapupl#1{

148:     \llap{$\vcenter{\hbox{$\scriptstyle#1$}}$}\Big\uparrow}

149: \def\mapupr#1{\Big\uparrow

150:     \rlap{$\vcenter{\hbox{$\scriptstyle#1$}}$}}

151: 

152: 

153: 

154: \begin{document}

155: 

156: \title[Arnold Diffusion]{Arnold Diffusion of the Discrete Nonlinear Schr\"odinger Equation}

157: 

158: 

159: %    Information for first author

160: \author{Y. Charles Li}

161: %    Address of record for the research reported here

162: \address{Department of Mathematics, University of Missouri, 

163: Columbia, MO 65211}

164: %    Current address

165: \curraddr{}

166: \email{cli@math.missouri.edu}

167: %    \thanks will become a 1st page footnote.

168: \thanks{}

169: 

170: %    Information for second author

171: %\author{Author Two}

172: %\address{Mathematical Research Section, School of Mathematical Sciences,

173: %Australian National University, Canberra ACT 2601, Australia}

174: %\email{two@maths.univ.edu.au}

175: %\thanks{Support information for the second author.}

176: 

177: %    General info

178: \subjclass{37, 34, 35, 78, 76}

179: \date{}

180: 

181: 

182: \keywords{Arnold diffusion, Darboux transformation, isospectral theory, 

183: Melnikov-Arnold integrals, $\lambda$-lemma, transition chain.}

184: 

185: \dedicatory{Communicated by Y. Charles Li, received April 3, 2006 \\

186: and, in revised form, June 25, 2006.}

187: 

188: 

189: 

190: \begin{abstract}

191: In this article, we prove the existence of Arnold diffusion for an interesting 

192: specific system -- discrete nonlinear Schr\"odinger equation. The proof is for 

193: the 5-dimensional case with or without resonance. In higher dimensions, the 

194: problem is open. Progresses are made by establishing a complete set of 

195: Melnikov-Arnold integrals in higher and infinite dimensions. The openness lies 

196: at the concrete computation of these Melnikov-Arnold integrals. New machineries 

197: introduced here into the topic of Arnold diffusion are the Darboux transformation 

198: and isospectral theory of integrable systems. 

199: \end{abstract}

200: 

201: \maketitle

202: 

203: %\section*{}

204: %This is an example of an unnumbered first-level heading.

205: 

206: %\specialsection*{This is a Special Section Head}

207: %This is an example of a special section head%

208: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

209: %\footnote{Here is an example of a footnote. Notice that this footnote

210: %text is running on so that it can stand as an example of how a footnote

211: %with separate paragraphs should be written.

212: %\par

213: %And here is the beginning of the second paragraph.}%

214: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

215: 

216: %\section{This is a numbered first-level section head}

217: %This is an example of a numbered first-level heading.

218: 

219: %\subsection{This is a numbered second-level section head}

220: %This is an example of a numbered second-level heading.

221: 

222: %\subsection*{This is an unnumbered second-level section head}

223: %This is an example of an unnumbered second-level heading.

224: 

225: %\subsubsection{This is a numbered third-level section head}

226: %This is an example of a numbered third-level heading.

227: 

228: %\subsubsection*{This is an unnumbered third-level section head}

229: %This is an example of an unnumbered third-level heading.

230: 

231: \tableofcontents

232: 

233: \section{Introduction}

234: 

235: For a simple example posed by V. I. Arnold \cite{Arn64}, the existence of 

236: the so-called Arnold diffusion has been proved (see e.g. \cite{Arn64} 

237: \cite{AA68} \cite{FM00}). The argument involves two parts: A calculation 

238: of the Melnikov-Arnold integrals \cite{Arn64} and a transversal intersection 

239: argument \cite{AA68} supported by a $\la$-lemma \cite{FM00}. Other arguments 

240: of variational type were also developed \cite{Mat04} \cite{Xia98} \cite{BBB03}

241: \cite{CY04}. 

242: 

243: Nevertheless, the theory of Arnold diffusion is far from complete \cite{Dou88}

244: \cite{CG94} \cite{Mar96} \cite{Cre97} \cite{Per98} \cite{BT99} \cite{CV00} 

245: \cite{DLS00} \cite{FM03} \cite{LW04} \cite{DLS06}. The main challenge is dealing with high 

246: dimensional specific systems of interest in applications. When the dimensions 

247: of the KAM tori are large, more Melnikov-Arnold integrals are needed to establish 

248: Arnold diffusion. Calculating these integrals is a dauting task if not impossible, 

249: even with computers. In an infinite dimensional phase space, even the dimensions of 

250: the KAM tori become a challenging issue. Are there infinite dimensional KAM tori ?

251: in what form ? In the classical setting of a Banach space with angles-momenta 

252: coordinates, the challenge is how to deal with the perturbation v.s. the decay of 

253: the sequence of momenta. It is a very interesting problem. 

254: 

255: The aim of the current article is to draw attention to two canonical systems of 

256: mathematical physics: The discrete nonlinear Schr\"odinger equation (DNLS) and 

257: its continuous version -- the nonlinear Schr\"odinger equation (NLS). DNLS and 

258: NLS are integrable systems that describe many different phenomena in physics 

259: \cite{APT04}. DNLS is an integrable finite difference discretization of NLS. An 

260: interesting fact about DNLS is that one can choose and change the dimensions of 

261: the phase space by selecting the number of particles in the discretization.

262: For a two particle case, under periodic Hamiltonian perturbations, the resulting 

263: system is 5-dimensional, for which the existence of Arnold diffusion will be 

264: proved here. For a three particle case, the system is 7-dimensional, and will 

265: be a good testing ground for a higher dimensional theory.

266: 

267: The integrable theory offers the missing link from low dimension to high dimension

268: via two powerful and beautiful machineries: Darboux transformation and isospectral 

269: theory. Darboux transformation generates explicit expressions of separatrices, 

270: while isospectral theory produces all the Melnikov vectors. Together they provide a 

271: complete set of Melnikov-Arnold integrals with elegant universal formulae. This is 

272: the case for both DNLS and NLS \cite{Li92} \cite{Li03} \cite{Li04} \cite{Li99} 

273: \cite{Li04a}. On the other hand, specific calculation of these integrals is the 

274: challenge. For the purpose of proving the existence of chaos, often one Melnikov 

275: integral is enough and easily computable \cite{Li04} \cite{Li04a} \cite{Li04b}

276: \cite{Li04c} \cite{Li06c}. The reason is that one can utilize locally invariant center 

277: manifolds instead of KAM tori. For the purpose of Arnold diffusion, local invariance

278: (i.e. orbits can only enter or leave the submanifold through its boundaries) is 

279: not enough. This is due to the second part of the argument for Arnold diffusion 

280: mentioned above. To establish a $\la$-lemma, one needs the tori to be invariant 

281: (not locally invariant). For a different application of $\la$-lemma in (1). 

282: establishing shadowing lemma in infinite dimensional autonomous systems, (2).

283: proving the existence of homoclinic tubes and heteroclinically tubular cycles, 

284: (3). proving the existence of tubular chaos, (4). proving the existence of chaos 

285: cascade; we refer the readers to \cite{Li03a} \cite{Li03b} \cite{Li04c} \cite{Li06c}.

286: 

287: The article is organized as follows: Section 2 deals with isospectral theory and 

288: Darboux transformation for both DNLS and NLS. Section 3 deals with Arnold diffusion 

289: for a 5-dimensional perturbed DNLS.

290: 

291: \section{Isospectral Theory and Darboux Transformation}

292: 

293: In this section, we are going to present the isospectral theory and Darboux 

294: transformation for both the discrete nonlinear Schr\"odinger equation (DNLS) and 

295: the nonlinear Schr\"odinger equation (NLS).

296: 

297: \subsection{Discrete Nonlinear Schr\"odinger Equation}

298: 

299: Consider the discrete nonlinear Schr{\"{o}}dinger equation (DNLS),

300: \begin{equation}

301: i \dot{q}_n = {1 \over h^2}[q_{n+1}-2 q_n +q_{n-1}] + |q_n|^2(q_{n+1}+q_{n-1})

302: -2 \om^2 q_n \ ,

303: \label{DNLS}

304: \end{equation}

305: where $q_n$'s are complex-valued, $i = \sqrt{-1}$ is the imaginary unit, $\om$ is 

306: a positive parameter, and $q_n$ satisfies the periodic boundary condition and even 

307: constraint,

308: \begin{equation}

309: q_{n+N}=q_n,\ \ \ \ q_{-n}=q_n \ ,  \label{bc}

310: \end{equation}

311: where $N$ is a positive integer $N \geq 3$ and $h=1/N$. The DNLS (\ref{DNLS}) is a 

312: $2(M+1)$-dimensional system, where $M=N/2$ ($N$ even) and $M=(N-1)/2$ ($N$ odd).

313: The DNLS can be rewritten in the Hamiltonian form

314: \begin{equation}

315: i \dot{q}_n = \rho_n \pa H_0/ \pa \overline{q_n} \ , \label{HDNLS}

316: \end{equation}

317: where $\rho_n = 1 +h^2|q_n|^2$ and 

318: \[

319: H_0={1 \over h^2}\sum^{N-1}_{n=0}

320: \bigg \{ \overline{q_n}(q_{n+1}+q_{n-1})-{2 \over h^2}(1+\om^2 h^2)\ln \rho_n

321: \bigg \} \ .

322: \]

323: The phase space is defined as 

324: \[

325: \cS = \bigg\{ \vq = ({\bf q} , \bar{{\bf q}}) \bigg | \ {\bf q}=(q_0,q_1,...,q_{N-1}), 

326: \ q_{N-n}=q_n  \ (1 \leq n \leq N-1)  \bigg\} \ . 

327: \]

328: In $\cS$ (viewed as a vector space over the real numbers), we 

329: define the inner product, for any two points $\vq^{\ +}$ and $\vq^{\ -}$,

330: as follows:

331: \[

332: \lag \vq^{\ +}, \vq^{\ -} \rag = \sum_{n=0}^{N-1} ( q_n^{+} 

333: \overline{q_n^{-}} + \overline{q_n^{+}} q_n^{-}) \ . 

334: \]

335: \nid

336: And the norm of $\vq$ is defined as $\|\vq \ \|^2 = \lag \vq,\vq \rag$. 

337: \begin{remark}

338: In the expression of $H_0$, both $\sum^{N-1}_{n=0}[ \overline{q_n}(q_{n+1}+q_{n-1})]$ 

339: and $I= \frac{1}{h^2}\sum^{N-1}_{n=0}\ln \rho_n$ are constants of motion too. $I$ will be 

340: used later to establish Arnold diffusion in the non-resonant case. Also the constant 

341: of motion $D$ given by $D^2=\prod^{N-1}_{n=0}\rho_n$ will play an important role in 

342: the isospectral theory. In the continuum limit (i.e. $h\ra 0$), the Hamiltonian $H_0$ 

343: has a limit in the manner: $h H_0 \ra H_c$, where $H_c$ is the Hamiltonian for 

344: NLS, $H_c=-\int^1_0 [ |q_x|^2+2\om^2|q|^2-|q|^4 ]dx$. Also as $h\ra 0$, $D\ra 1$. 

345: \end{remark}

346: 

347: \subsection{Isospectral Theory of DNLS}

348: 

349: For more details on the topic of this subsection, see \cite{Li03}. DNLS has the Lax 

350: pair \cite{AL76}

351: \begin{eqnarray}

352: \varphi_{n+1}&=&L_n\varphi_n, \label{DLax1} \\

353: \dot{\varphi}_n&=&B_n\varphi_n, \label{DLax2}

354: \end{eqnarray}

355: \noindent

356: where

357: \[

358: L_n=\left( \begin{array}{cc} z& ihq_n \cr ih\overline{q_n} & 1/z \cr \end{array} \right),

359: \]

360: \[

361: B_n={i\over h^2}\left( \begin{array}{ll}

362: 1-z^2+2i\la h-h^2q_n\overline{q_{n-1}}+\om^2 h^2 & -izhq_n+(1/z)ihq_{n-1}  \cr

363: -izh\overline{q_{n-1}}+(1/z)ih\overline{q_n} & \frac{1}{z^2}-1+2i\la h+h^2\overline{q_n}

364: q_{n-1}-\om^2 h^2 \cr \end{array} \right),

365: \]

366: and where $z=\exp(i\la h)$. Compatibility of the over determined system 

367: (\ref{DLax1},\ref{DLax2}) gives the ``Lax representation'' 

368: \[

369: \dot{L}_n=B_{n+1}L_n-L_nB_n

370: \]

371: of the DNLS (\ref{DNLS}). Focusing our attention upon the discrete spatial part 

372: (\ref{DLax1}) of the Lax pair, let $M_n = M_n(z,\vq )$ be the $2\times 2$ fundamental 

373: matrix solution such that $M_0$ is the $2\times 2$ identity matrix. The {\em Floquet 

374: discriminant} $\Dl : \mathbb{C} \times \cS \ra \mathbb{C}$ is defined by 

375: \[

376: \Dl = \frac{1}{D}\ \text{trace } M_N(z,\vq )\ ,

377: \]

378: where $D^2=\prod^{N-1}_{n=0}\rho_n$. The isospectral theory starts from the fact 

379: that for any $z \in \mathbb{C}$, 

380: $\Dl (z, \vq )$ is a constant of motion of DNLS. An easy way to understand this is that 

381: as $q_n$ evolves in time according to DNLS, the parameter $z$ in the Lax pair does not 

382: change. $\Dl (z, \vq )$ is a meromorphic function in $z$ of degree ($+N,-N$), and 

383: provides $(M+1)$ functionally independent constants of motion, where 

384: $M = N/2$ ($N$ even), $M=(N-1)/2$ ($N$ odd). There are many ways to generate 

385: $(M+1)$ functionally independent constants of motion from $\Dl$. The approach which proved to be 

386: most convenient is by employing all the critical points $z^c$ of $\Dl (z, \vq )$:

387: \[

388: \frac{\pa \Dl}{\pa z} (z^c, \q )=0\ .

389: \]

390: \begin{definition}

391: We define a sequence of ($M+1$) constants of motion $F_j$ as follows

392: \begin{equation}

393: F_j(\vq ) = \Dl (z^c_j(\vq ), \vq )\ , \quad (j=1, \cdots , M+1)\ .

394: \label{deff}

395: \end{equation}

396: \end{definition}

397: There is a good description on the locations of these critical points $z^c_j$ in 

398: the NLS setting \cite{Li04}. These $F_j$'s can be used to build a complete set of 

399: Melnikov-Arnold integrals for the Arnold diffusion purpose. The Melnikov vectors 

400: are given by the gradients of these $F_j$'s.

401: \begin{theorem}

402: Let $z^c_j(\vq)$ be a simple critical point, then

403: \begin{equation}

404: \left (\begin{array}{c} \frac{\pa F_j}{\pa q_n} \cr \cr

405: \frac{\pa F_j}{\pa \overline{q_n}} \cr \end{array} \right ) = 

406: {i h (\z -\z^{-1}) \over 

407: 2 W_{n+1}} \left (\begin{array}{c} \psi^{(+,2)}_{n+1}\psi^{(-,2)}_{n}+ 

408: \psi^{(+,2)}_{n}\psi^{(-,2)}_{n+1} \cr \cr -\psi^{(+,1)}_{n+1}\psi^{(-,1)}_{n}-

409: \psi^{(+,1)}_{n}\psi^{(-,1)}_{n+1} \cr \end{array} \right )\ , 

410: \label{fidr}

411: \end{equation}

412: where $\psi_n^\pm = (\psi_n^{(\pm,1)}, \psi_n^{(\pm,2)})^T$, ($T$: transpose), are two 

413: Bloch solutions of the Lax pair (\ref{DLax1},\ref{DLax2}) at ($z^c_j,\vq$) such that 

414: \[

415: \psi^\pm_n = D^{n/N} \z^{\pm n/N} \tilde{\psi}^\pm_n\ ,

416: \]

417: where $\tilde{\psi}^\pm_n$ are periodic in $n$ with period $N$, $\z$ is a complex constant, 

418: and $W_n = \ \mbox{det}\ ( \psi^+_n,\psi^-_n )$ is the Wronskian.

419: \end{theorem}

420: For a perturbation $H_0+\e H_1$ of the Hamiltonian $H_0$, a complete set of 

421: Melnikov-Arnold integrals is then given by 

422: \[

423: I_j = \int_{-\infty}^{\infty} \sum_{n=0}^{N-1} 2 \ \text{Im} \left [ \frac{\pa F_j}{\pa q_n} \rho_n 

424: \frac{\pa H_1}{\pa \overline{q_n}} \right ]dt \ , \quad (j=1, \cdots , M+1)\ ,

425: \]

426: which are evaluated on the unstable manifolds of tori that persist into KAM tori, 

427: where $\frac{\pa F_j}{\pa q_n}$ is given by (\ref{fidr}). Expression (\ref{fidr}) is a 

428: universal expression for all the Melnikov vectors. The challenge is of course how 

429: to compute these $I_j$'s. The difficulty lies at how to obtain expressions of the orbit $q_n$

430: and the corresponding $\psi_n^\pm$. By utilizing Darboux transformations, these 

431: expressions can be obtained even by hand in some cases.  

432: 

433: \subsection{Darboux Transformation of DNLS}

434: 

435: For more details on the topic of this subsection, see \cite{Li03}.

436: Expressions of unstable manifolds of tori can be generated via Darboux transformations.

437: For DNLS, such a Darboux transformation was established in \cite{Li92}. 

438: 

439: Let $q_n$ be any solution of DNLS. Pick $z = z^d$ ($|z^d| \neq 1$) at which  

440: the linear operator $L_n$ in the Lax pair (\ref{DLax1})-(\ref{DLax2}) has two 

441: linearly independent periodic solutions (or anti-periodic solutions) $\phi^{\pm}_n$ 

442: in the sense:

443: \[

444: \phi^{\pm}_{n+N} = D \phi^{\pm}_n \ , \quad (\text{or } \phi^{\pm}_{n+N} = -D \phi^{\pm}_n) \ ,

445: \]

446: where $D$ is defined above. Let 

447: \[

448: \phi_n=(\phi_{n1}, \phi_{n2})^T = c^+ \phi_n^+ + c^- \phi_n^-\ ,

449: \]

450: where $c^+$ and $c^-$ are complex parameters. Define the matrix $\Ga_n$ by

451: \[

452: \Ga_n=\left(\begin{array}{cc} z+(1/z)a_n & b_n \cr c_n &-1/z+z d_n \cr

453:             \end{array} \right),

454: \]

455: \nid

456: where

457: \begin{eqnarray*}

458: a_n &=& {z^d \over \overline{z^d}^2\Dl_n}\bigg [|\phi_{n2}|^2+|z^d|^2|\phi_{n1}|^2

459:     \bigg ],\\

460: d_n &=& -{1 \over z^d\Dl_n}\bigg [|\phi_{n2}|^2+|z^d|^2|\phi_{n1}|^2

461:     \bigg ],\\

462: b_n &=& {|z^d|^4-1 \over \overline{z^d}^2\Dl_n}\phi_{n1} \overline{\phi_{n2}}, \\

463: c_n &=& {|z^d|^4-1 \over |z^d|^2\Dl_n}\overline{\phi_{n1}}\phi_{n2}, \\

464: \Dl_n &=& -{1 \over \overline{z^d}}\bigg [|\phi_{n1}|^2+|z^d|^2|\phi_{n2}|^2

465:     \bigg ].

466: \end{eqnarray*}

467: \nid

468: From these formulae, we see that

469: \[

470: \overline{a_n}=-d_n,\ \ \overline{b_n}=c_n.

471: \]

472: \begin{theorem}[Darboux Transformation \cite{Li92}]

473: Define $Q_n$ and $\Psi_n$ by

474: \[

475: Q_n = {i\over h}b_{n+1}-a_{n+1}q_n\ , \quad \Psi_n = \Ga_n \psi_n \ ,

476: \]

477: where $\psi_n$ solves the Lax pair (\ref{DLax1})-(\ref{DLax2}) 

478: at $(q_n,z)$. Then $Q_n$ is also a solution of DNLS, and $\Psi_n$ 

479: solves the Lax pair (\ref{DLax1})-(\ref{DLax2}) at $(Q_n,z)$.

480: \end{theorem}

481: In principle, by choosing $q_n$ to be orbits on the tori, one can 

482: generate $Q_n$ to be orbits on the unstable manifolds of the tori.

483: In special cases, $Q_n$ can be calculated by hands.

484: 

485: Next we present an example. Define the 2-dimensional invariant plane

486: \begin{equation}

487: \Pi = \left \{ \vq \in \cS \ | \ q_n =q\ , \quad \forall n \right \}\ .

488: \label{invp}

489: \end{equation}

490: On $\Pi$, the solutions of DNLS are given by the periodic orbits (1-tori)

491: \begin{equation}

492: q_n=q_c \ ,\ \forall n; \quad q_c=a\exp \left \{-i[2(a^2-\om^2)t - \ga]

493: \right \}\ , 

494: \label{us}

495: \end{equation}

496: where $a$ and $\ga$ are real constants. We choose the amplitude $a$ in 

497: the following range so that the unstable direction of $q_c$ is 1-dimensional

498: \begin{eqnarray}

499: & & N\tan{\pi \over N}< a <N\tan{2\pi \over N}\ ,\ \ \ \mbox{when}\ N>3 \ ,

500: \nonumber \\ \label{constr} \\

501: & & 3\tan{\pi \over 3}< a < \infty\ ,\ \ \ \mbox{when}\ N=3 \ .\nonumber

502: \end{eqnarray}

503: Increasing the unstable dimensions of $q_c$ amounts to iterations of 

504: the Darboux transformation which are still doable by hands, and does not 

505: add difficulty substantially in the Arnold diffusion problem. To apply the 

506: Darboux transformation we choose 

507: \[

508: z^d = \sqrt{\rho}\cos \frac{\pi}{N} +\sqrt{\rho \cos^2 \frac{\pi}{N} -1}\ ,

509: \]

510: where $\rho = 1 +h^2a^2$. This $z^d$ is also a critical point of $\Dl$. We 

511: label it by $z^c_1 = z^d$. Direct calculation leads to the following 

512: formulae

513: \begin{eqnarray}

514: Q_n &=& q_c \bigg [ \Ga / \La_n -1 \bigg ]\ , \label{horbit} \\ \nonumber \\

515: \left (\begin{array}{c} \frac{\pa F_1}{\pa q_n} \cr \cr

516: \frac{\pa F_1}{\pa \overline{q_n}} \cr \end{array} \right )\bigg |_{Q_n}

517:     &=& K [K_n]^{-1} \ \mbox{sech}[2\mu t + 2p] \left ( \begin{array}{c} 

518: X^1_n \cr X^2_n \cr \end{array} \right ) \ , \label{melv}

519: \end{eqnarray}

520: where

521: \begin{eqnarray*}

522: \Ga &=& 1-\cos 2 \varphi - i \sin 2 \varphi \tanh [ 2 \mu t + 2p]\ , \\

523: \La_n &=& 1 \pm \cos \varphi [\cos \be ]^{-1}\ \mbox{sech}[2 \mu t + 2p] 

524: \cos [2n\be]\ , \\

525: K &=& -2N (1-\hz^4) [8a\rho^{3/2} \hz^2]^{-1} \sqrt{\rho \cos^2\be - 1}\ , \\

526: K_n &=& \bigg [ \cos \be \pm \cos \varphi \ \mbox{sech}[2\mu t +2p] 

527: \cos[2(n-1)\be]\bigg ] \times \\

528: & &\bigg [ \cos \be \pm \cos \varphi \ \mbox{sech}[2\mu t +2p] 

529: \cos[2(n+1)\be]\bigg ]\ , \\

530: X^1_n &=& \bigg [ \cos \be \ \mbox{sech}\ [ 2\mu t +2p] 

531: \pm (\cos \varphi \\

532: & & -i \sin \varphi \tanh [ 2\mu t +2p]) \cos [2n\be]\bigg ] e^{i2\th(t)}\ , \\

533: X^2_n &=& \bigg [ \cos \be \ \mbox{sech}\ [ 2\mu t +2p] 

534: \pm (\cos \varphi \\

535: & & +i \sin \varphi \tanh [ 2\mu t +2p]) \cos [2n\be]\bigg ] e^{-i2\th(t)}\ , \\

536: & & \be = \pi / N\ , \ \ \mu = 2h^{-2} \sqrt{\rho}\sin \be \sqrt{\rho \cos^2 \be -1}\ ,\\

537: & & \rho = 1+h^2 a^2\ , \ \ h=1/N, \ \ \hz =\sqrt{\rho}\cos \be +\sqrt{\rho \cos^2 \be -1}\ , \\

538: & & \th(t)=(a^2-\om^2)t - \ga/2\ , \ \ h a e^{i \varphi} =  

539: \sqrt{\rho \cos^2 \be -1} + i \sqrt{\rho} \sin \be \ . 

540: \end{eqnarray*}

541: and $p$ is a real parameter. One can easily see that $Q_n$ represents homoclinic 

542: orbits asymptotic to the periodic orbits $q_c$: As $t \ra \pm \infty$,

543: \[

544: Q_n \ra q_c e^{i(\pi \pm 2\varphi )} \ .

545: \]

546: The union 

547: \[

548: \bigcup_{\ga \in [0, 2\pi ]}Q_n 

549: \]

550: represents the 2-dimensional unstable (=stable) manifold of the 1-torus (\ref{us}). 

551: When $N>3$, the 1-torus also has a center manifold of codimension 2. 

552: 

553: 

554: \subsection{Nonlinear Schr\"odinger Equation}

555: 

556: Consider the nonlinear Schr{\"{o}}dinger equation (NLS),

557: \begin{equation}

558: iq_t = q_{xx} + 2[|q|^2 - \om^2] q \ ,

559: \label{NLS}

560: \end{equation}

561: where $q = q(t,x)$ is a complex-valued function of the two real 

562: variables $t$ and $x$, $t$ represents time, and $x$ represents

563: space. $q(t,x)$ is subject to periodic boundary condition of period 

564: $2 \pi$, and even constraint, i.e. 

565: \[

566: q(t,x + 2 \pi) = q(t,x)\ , \ \ q(t,-x) = q(t,x)\ .

567: \]

568: $\om$ is a positive parameter. The DNLS (\ref{DNLS}) is an integrable finite 

569: difference discretization of the NLS. The NLS can be rewritten in the Hamiltonian form

570: \begin{equation}

571: i \dot{q} = \pa H_0/ \pa \bq \ , \label{HNLS}

572: \end{equation}

573: where 

574: \[

575: H_0=-\int^{2\pi}_0 \left [ |q_x|^2+2\om^2|q|^2-|q|^4 \right ] dx \ .

576: \]

577: The phase space is defined as 

578: \[

579: \cS = \bigg\{ \vq = (q,\bq ) \bigg | \ q \in H^k \ , \quad (k \geq 1) \bigg\} \ ,

580: \]

581: where $H^k$ is the Sobolev space of periodic and even functions.

582: 

583: \subsection{Isospectral Theory of NLS}

584: 

585: For more details on the topic of this subsection, see \cite{Li04}. NLS has the Lax 

586: pair

587: \begin{eqnarray}

588: \psi_x &=& U \psi \ , \label{ZS1} \\

589: \psi_t &=& V \psi \ , \label{ZS2}

590: \end{eqnarray}

591: where

592: \[

593: U = i \left ( \begin{array}{lr} \la & q \cr \bq & -\la \cr 

594: \end{array} \right ) \ , 

595: \]

596: \[

597: V = i \left ( \begin{array}{lr} 2\la^2 -|q|^2 +\om^2 & 

598: 2\la q -i q_x \cr 2 \la \bq + i \overline{q_x} & -2 \la^2 +|q|^2 -\om^2 \cr 

599: \end{array} \right ) \ . 

600: \]

601: Focusing our attention on the spatial part (\ref{ZS1}) of the Lax 

602: pair (\ref{ZS1},\ref{ZS2}), we can define the fundamental matrix 

603: solution $M(x)$ such that $M(0)$ is the $2\times 2$ identity matrix. Then 

604: the {\em Floquet discriminant} $\Dl$ is defined as

605: \[

606: \Dl = \ \mbox{trace}\ M(2\pi)\ .

607: \]

608: The isospectral theory starts from the fact that for any $\la \in \mathbb{C}$, 

609: $\Dl (\la , \vq )$ is a constant of motion of NLS. $\Dl = \Dl (\la, q)$ is 

610: an entire function in both $\la$ and $\vq$. $\Dl$ provides enough functionally 

611: independent constants of motion to make the NLS (\ref{NLS}) integrable in the 

612: classical Liouville sense. There are many ways to generate these

613: functionally independent constants of motion from $\Dl$. The approach which proved to be 

614: most convenient is by employing all the critical points $\la^c$ of $\Dl (\la ,\vq )$:

615: \[

616: \frac{\pa \Dl}{\pa \la} (\la^c, \vq )=0\ .

617: \]

618: For each $\vq$, there is a sequence of critical points $\{ \la^c_j \}$ whose 

619: locations can be estimated \cite{Li04}. 

620: \begin{definition}

621: We define a sequence of constants of motion $F_j$ as follows

622: \[

623: F_j(\vq ) = \Dl (\la^c_j(\vq ), \vq )\ .

624: \]

625: \end{definition}

626: These $F_j$'s can be used to build a complete set of Melnikov-Arnold integrals for 

627: the Arnold diffusion purpose. The Melnikov vectors are given by the gradients of 

628: these $F_j$'s.

629: \begin{theorem}

630: Let $\la^c_j(\vq)$ be a simple critical point, then

631: \begin{equation}

632: \left (\begin{array}{c} \frac{\pa F_j}{\pa q} \cr \cr

633: \frac{\pa F_j}{\pa \overline{q}} \cr \end{array} \right ) = 

634: i \frac {\sqrt{\Dl^2 -4}} {W(\psi^+, \psi^-)}

635: \left ( \begin{array}{c} \psi_2^+ \psi_2^- \cr - \psi_1^+ \psi_1^-

636: \cr \end{array} \right ) \ ,

637: \label{nlsv}

638: \end{equation}

639: where $\psi^{\pm} = (\psi^{\pm}_1, \psi^{\pm}_2)^T$, ($T$: transpose), are two 

640: Bloch solutions of the Lax pair (\ref{ZS1})-(\ref{ZS2}) at ($\la^c_j, \vq$) such that 

641: \[

642: \psi^{\pm}(x) = e^{\pm \sg x} \tilde{\psi}^{\pm}(x)\ ,

643: \]

644: where $\sg$ is a complex constant and $\tilde{\psi}^{\pm}(x)$ are periodic in $x$ 

645: with period $2\pi$, 

646: \[

647: W(\psi^+,\psi^-)=\psi^+_1 \psi^-_2 - \psi^+_2 \psi^-_1

648: \]

649: is the Wronskian, and $\Dl$ is evaluated at $\la = \la^c_j$. 

650: \end{theorem}

651: For a perturbation $H_0+\e H_1$ of the Hamiltonian $H_0$, a complete set of 

652: Melnikov-Arnold integrals is then given by 

653: \[

654: I_j = \int_{-\infty}^{\infty} \int_0^{2\pi} 2 \ \text{Im} \left [ \frac{\pa F_j}{\pa q}  

655: \frac{\pa H_1}{\pa \overline{q}} \right ]dxdt \ , 

656: \]

657: which are evaluated on the unstable manifolds of tori that persist into KAM tori, 

658: where $\frac{\pa F_j}{\pa q}$ is given by (\ref{nlsv}). Expression (\ref{nlsv}) is a 

659: universal expression for all the Melnikov vectors. The challenge is of course how 

660: to compute these $I_j$'s. The difficulty lies at how to obtain expressions of the orbit $q$

661: and the corresponding $\psi^\pm$. By utilizing Darboux transformations, these 

662: expressions can be obtained even by hand in some cases. Also as mentioned in the 

663: introduction, there is the issue of elusiveness of infinite dimensional KAM tori. 

664: As $j \ra \infty$, The magnitude of $I_j$ v.s. the size of the perturbation $\e$ 

665: is another tricky problem.  

666: 

667: \subsection{Darboux Transformation of NLS}

668: 

669: For more details on the topic of this subsection, see \cite{Li04}.

670: Expressions of unstable manifolds of tori can be generated via Darboux transformations.

671: For NLS, such a Darboux transformation was also known. 

672: 

673: Let $q$ be any solution of NLS. Pick $\la = \nu$ (complex constant) at which  

674: (\ref{ZS1}) has two linearly independent periodic solutions (or anti-periodic solutions) 

675: $\phi^{\pm}$ in the sense:

676: \[

677: \phi^{\pm}(x+2\pi ) = \phi^{\pm}(x)\ , \quad (\text{or } \phi^{\pm} (x+2\pi )= - 

678: \phi^{\pm}(x)) \ .

679: \]

680: Let 

681: \[

682: \phi =c_+ \phi^+ + c_- \phi^-\ ,

683: \]

684: where $c_+$ and $c_-$ are complex parameters. Define the matrix $G$ by

685: \[

686: G = \Ga \left ( \begin{array}{lr} \la -\nu & 0 \cr 

687: 0 & \la - \bar{\nu} \cr \end{array} \right )  \Ga^{-1} \ ,

688: \]

689: where 

690: \[

691: \Ga = \left ( \begin{array}{lr} \phi_1 & - \overline{\phi_2} \cr 

692: \phi_2 & \overline{\phi_1} \cr  \end{array} \right ) \ .

693: \]

694: \begin{theorem}[Darboux Transformation]

695: Define $Q$ and $\Psi$ by

696: \[

697: Q = q + 2 (\nu - \bar{\nu}) \frac {\phi_1 \overline{\phi_2}} {|\phi_1|^2 

698: + |\phi_2|^2}\ , \quad \Psi = G \psi \ ,

699: \]

700: where $\psi$ solves the Lax pair (\ref{ZS1})-(\ref{ZS2}) 

701: at $(q,\la )$. Then $Q$ is also a solution of NLS, and $\Psi$ 

702: solves the Lax pair (\ref{ZS1})-(\ref{ZS2}) at $(Q,\la )$.

703: \end{theorem}

704: In principle, by choosing $q$ to be orbits on the tori, one can 

705: generate $Q$ to be orbits on the unstable manifolds of the tori.

706: In special cases, $Q$ can be calculated by hands.

707: 

708: Next we present an example. Define the 2-dimensional invariant plane

709: \[

710: \Pi = \left \{ \vq \in \cS \ | \ \pa_x q = 0 \right \}\ .

711: \]

712: On $\Pi$, the solutions of NLS are given by the same periodic orbits (1-tori)

713: as in (\ref{us})

714: \begin{equation}

715: q_c=a\exp \left \{-i[2(a^2-\om^2)t - \ga]\right \}\ , 

716: \label{usc}

717: \end{equation}

718: where $a$ and $\ga$ are real constants. We choose the amplitude $a$ in 

719: the range $a \in (1/2, 1)$ so that the unstable direction of $q_c$ is 1-dimensional.

720: Increasing the unstable dimensions of $q_c$ amounts to iterations of 

721: the Darboux transformation which are still doable by hands \cite{Li04a}, and does not 

722: add difficulty substantially in the Arnold diffusion problem. To apply the 

723: Darboux transformation we choose 

724: \[

725: \nu = i \sg \ , \quad \sg = \sqrt{a^2 - 1/4}\ .

726: \]

727: This $\nu$ is also a critical point of $\Dl$. We label it by $z^c_1 = \nu$.

728: Direct calculation leads to

729: \begin{eqnarray*}

730: Q &=& q_c \bigg [ 1 \pm \sin \vth_0 \ \mbox{sech} \tau  

731: \cos x \bigg ]^{-1} \\

732: & & \cdot \bigg [ \cos 2\vth_0 - i \sin 2\vth_0 \tanh \tau \mp 

733: \sin \vth_0 \ \mbox{sech} \tau \cos x \bigg ] \ ,  \\

734: \left (\begin{array}{c} \frac{\pa F_1}{\pa q} \cr \cr

735: \frac{\pa F_1}{\pa \overline{q}} \cr \end{array} \right )\bigg |_{Q} 

736: &=& \frac{1}{4} a^{-2} i (\nu - \bar{\nu})

737: \sqrt{\Dl(\nu)\Dl''(\nu)} \frac {1}{(|u_1|^2 + |u_2|^2)^2}

738: \left ( \begin{array}{c} \overline{q_c}\ \overline{u_1}^{\ 2} \cr 

739: - q_c\ \overline{u_2}^{\ 2} \cr \end{array} \right ) \ ,

740: \end{eqnarray*}

741: where 

742: \begin{eqnarray*}

743: u_1 &=&  \cosh \frac{\tau}{2} \cos z -i \sinh \frac{\tau}{2} \sin z \ , \\

744: u_2 &=&  -\sinh \frac{\tau}{2} \cos (z - \vth_0) + i \cosh  \frac{\tau}{2}

745: \sin (z - \vth_0) \ , \\

746: & & a e^{i\vth_0} =\frac{1}{2} + \nu \ , \quad \tau = 2 \sg t - \rho \ , 

747: \quad z = x/2 + \vth_0/2 \mp \pi /4 \ ,

748: \end{eqnarray*}

749: and $\rho$ is a real parameter. As $t \ra \pm \infty$,

750: \[

751: Q \ra q_c e^{\mp i2\vth_0} \ .

752: \]

753: The union 

754: \[

755: \bigcup_{\ga \in [0, 2\pi ]}Q 

756: \]

757: represents the 2-dimensional unstable (=stable) manifold of the 1-torus (\ref{usc}). 

758: The 1-torus also has a center manifold of codimension 2. 

759: 

760: \section{Arnold Diffusion}

761: 

762: To establish the existence of Arnold diffusion, one needs three ingredients:

763: (1). Melnikov-Arnold integrals, (2). A $\la$-lemma, (3). A transversal intersection 

764: argument. Melnikov-Arnold integrals have been studied above. Next we discuss the 

765: other two ingredients.

766: \begin{lemma}[The $\la$-lemma of Fontich-Martin \cite{FM00}]

767: Let $F$ be a $C^2$ diffeomorphism in $\mathbb{R}^n$, $\mathbb{T}$ be a $C^2$ 

768: invariant torus in $\mathbb{R}^n$. The dynamics on $\mathbb{T}$ is quasi-periodic.

769: $\mathbb{T}$ has $C^2$ unstable, stable and center manifolds ($W^u,W^s,W^c$). Let 

770: $\Ga$ be a $C^1$ manifold intersecting transversally $W^s$ at a point. Then

771: \[

772: W^u \subset \overline{\bigcup_{m \geq 0}F^m(\Ga )}\ .

773: \]

774: \label{lal}

775: \end{lemma}

776: \begin{remark}

777: Like every other $\la$-lemma, the claim is very intuitive, but the proof is always 

778: delicate. The proof in \cite{FM00} takes about eight pages. For a quick glance of the 

779: basic idea, see e.g. \cite{Li03a}. It turns out to be crucial to use Fenichel's 

780: fiber coordinates. With respect to base points, Fenichel fibers drop one degree of 

781: smoothness. That is why $C^2$ smoothness is required in the lemma to obtain $C^1$ 

782: families of $C^2$ unstable and stable Fenichel fibers. Using the Fenichel's 

783: fiber coordinates, $W^u$ and $W^s$ are rectified, i.e. they coincide with their 

784: tangent bundles. This makes the estimate a lot easier. Fenichel fibers also make it 

785: easier to track orbits inside $W^u$ and $W^s$ via the fiber base points in $\mathbb{T}$.

786: The main argument is to track the tangent space of a submanifold $S$ of $\Ga$ 

787: starting from the intersection point of $\Ga$ and $W^s$. After enough iterations of $F$, 

788: one can obtain some estimate of closeness to $W^u$. The novelty of \cite{FM00} is that 

789: they also track the tangent space at every point in $S$ and off $W^s$. This is 

790: necessary in order to obtain the claim of the lemma. The claim is proved by showing 

791: that any neighborhood of any point on $W^u$ has a nonempty intersection with $F^m(S)$ 

792: for some $m$. The claim of the lemma should also be true in proper infinite 

793: dimensional settings. 

794: \end{remark}

795: \begin{definition}[Transition Chain]

796: A finite or infinite sequence of tori $\{ \mathbb{T}_j \}$ forms a transition chain if 

797: $W^s(\mathbb{T}_j)$ intersects transversally $W^u(\mathbb{T}_{j+1})$ at some point, for 

798: all $j$, and dynamics on $\mathbb{T}_j$ is quasi-periodic.

799: \end{definition}

800: \begin{lemma}[Arnold \cite{AA68}]

801: Let $\{ \mathbb{T}_j \}$ ($1 \leq j \leq N$) be a finite transition chain. Then an 

802: arbitrary neighborhood of an arbitrary point in $W^u(\mathbb{T}_1)$ is connected to an 

803: arbitrary neighborhood of an arbitrary point in $W^s(\mathbb{T}_N)$ by an orbit.

804: \label{al}

805: \end{lemma}

806: \begin{proof}

807: Let $\Om_1$ be an arbitrary neighborhood of an arbitrary point $u_1 \in W^u(\mathbb{T}_1)$.

808: Let $\bB_{u_1}(r_1) \subset \Om_1$ be a closed ball of radius $r_1>0$ centered at $u_1$. 

809: Then using the $\la$-lemma \ref{lal}, one can find 

810: \begin{equation}

811: \bB_{u_N}(r_N) \subset \bB_{u_{N-1}}(r_{N-1}) \cdots \subset \bB_{u_2}(r_2)

812: \subset \bB_{u_1}(r_1)

813: \label{bnt}

814: \end{equation}

815: such that

816: \[

817: u_j \in W^u(\mathbb{T}_j)\ , \quad r_j >0\ , \quad (1 \leq j \leq N)\ .

818: \]

819: Indeed, by the $\la$-lemma \ref{lal},

820: \[

821: B_{u_1}(r_1) \cap W^u(\mathbb{T}_2) \neq \emptyset

822: \]

823: where $B_{u_1}(r_1)$ is the open ball. Then one can find 

824: \[

825: u_2 \in B_{u_1}(r_1) \cap W^u(\mathbb{T}_2) \ , \text{ and }

826: \bB_{u_2}(r_2) \subset B_{u_1}(r_1)\ .

827: \]

828: Again by the $\la$-lemma \ref{lal},

829: \[

830: B_{u_2}(r_2) \cap W^u(\mathbb{T}_3) \neq \emptyset \ ,

831: \]

832: and one can find 

833: \[

834: u_3 \in B_{u_2}(r_2) \cap W^u(\mathbb{T}_3) \ , \text{ and }

835: \bB_{u_3}(r_3) \subset B_{u_2}(r_2)\ ,

836: \]

837: and so on. Let $\Om_N$ be an arbitrary neighborhood of an arbitrary point in 

838: $W^s(\mathbb{T}_N)$. Inside $\Om_N$, one can find a $C^1$ submanifold intersecting 

839: transversally $W^s(\mathbb{T}_N)$ at the point. Again by the $\la$-lemma \ref{lal},

840: \[

841: B_{u_N}(r_N) \cap F^m(\Om_N) \neq \emptyset \text{  for some } m \ .

842: \]

843: Thus $\Om_1$ and $\Om_N$ are connected by an orbit.

844: \end{proof}

845: \begin{remark}

846: It is easy to see that around the connecting orbit, there is in fact a connecting flow 

847: tube \cite{Li06} \cite{Li03b} \cite{Li04c} \cite{Li06c}

848: \[

849: \bigcup_{0\leq m \leq M}F^m(D) \quad \text{for some } M

850: \]

851: where $D$ is a neighborhood. When $N=\infty$, relation (\ref{bnt}) leads to a 

852: point in the intersection

853: \[

854: u \in \bigcap_{j=0}^{\infty} \bB_{u_j}(r_j) \ .

855: \]

856: Starting from $u$, one obtains a connecting orbit. In a Banach space setting, if 

857: $N=\infty$, then one can choose $r_{j+1} \leq \frac{1}{2} r_j$ for any $j$ in 

858: (\ref{bnt}). Choosing an arbitrary point $v_j$ in $\bB_{u_j}(r_j)$, one gets a 

859: Cauchy sequence $\{ v_j \}$. Thus

860: \[

861: \lim_{j\ra \infty}v_j = v \in \bigcap_{j=0}^{\infty} \bB_{u_j}(r_j) \ .

862: \]

863: Starting from $v$, one still obtains a connecting orbit.

864: \end{remark}

865: 

866: \subsection{Arnold Diffusion of DNLS ($N=3$, Non-resonant Case)}

867: 

868: In this subsection, we prove the existence of Arnold diffusion for a perturbed 

869: DNLS when $N=3$, which is a 5-dimensional system. For arbitrary $N$, one can 

870: find large enough annular region inside the invariant plane $\Pi$ (\ref{invp}), which 

871: is normally hyperbolic, for which the current proof can be easily applied. 

872: The point is that increasing unstable and stable dimensions does not pose 

873: substantial computational difficulty to establishing Arnold diffusion, while 

874: increasing the dimensions of tori does. We will study here the case 

875: that there is no resonance ($\om = 0$) inside the invariant plane $\Pi$ (\ref{invp}).

876: The resonant case ($\om \neq 0$) will be studied in next subsection.

877: 

878: Consider the following perturbation of the DNLS (\ref{DNLS})

879: \[

880: H = H_0 +\e H_1 \ , 

881: \]

882: where

883: \[

884: H_1 = \al \sin t \sum_{n=0}^{N-1} 

885: \left | \frac{q_n-q_{n-1}}{h}\right |^2 + \sum_{n=0}^{N-1} 

886: \left [ \left ( \frac{q_n-q_{n-1}}{h}\right )^2 + 

887: \left ( \frac{\overline{q_n}-\overline{q_{n-1}}}{h}\right )^2\right ] \ ,

888: \]

889: where $\al$ is a real parameter. Under this perturbation, dynamics inside $\Pi$ is 

890: unchanged. $\Pi$ consists of periodic orbits forming concentric circles (\ref{us}) 

891: [Figure \ref{dpin}]. 

892: We are interested in the following normally hyperbolic annular region inside $\Pi$

893: \[

894: \A = \left \{ \vq \in \Pi \ | \quad q_n =q, \ \forall n, \quad 3\tan{\pi \over 3}< |q| < B \right \} 

895: \]

896: where $B$ is an arbitrary large constant. Denote by $\{ \F^{u,s}(\vq): \ \vq \in \A \}$ the 

897: $C^1$ families of $C^2$ one dimensional unstable and stable Fenichel fibers with base points in $\A$ 

898: \cite{Li04} such that for any $\vq_* \in \F^u(\vq )$ or $\vq_* \in \F^s(\vq )$,

899: ( $\vq \in \A$),

900: \[

901: \| F^t(\vq_* ) - \vq \| \leq C e^{\k t} \| \vq_* - \vq \| \ , \quad \forall t \in (-\infty , 0] \ ,

902: \]

903: or 

904: \[

905: \| F^t(\vq_* ) - \vq \| \leq C e^{-\k t} \| \vq_* - \vq \| \ , \quad \forall t \in [0, \infty ) \ ,

906: \]

907: where $F^t$ is the evolution operator of the perturbed DNLS, $\k$ and $C$ are some 

908: positive constants. The Fenichel fibers are $C^1$ in $\e \in [0, \e_0)$ for some $\e_0 >0$. 

909: It turns out that the constant of motion of DNLS (\ref{DNLS}) 

910: \[

911: I = \frac{1}{h^2} \sum_{n=0}^{N-1} \ln \rho_n 

912: \]

913: and $F_1$ [cf: (\ref{deff}) and (\ref{melv})] are 

914: the best choices to build the two Melnikov-Arnold intergals. Restricted to $\Pi$,

915: \[

916: I = \frac{1}{h^3} \ln \rho \ , \quad  \rho = 1+ h^2 |q|^2 \ .

917: \]

918: The level sets of $I$ lead to all the periodic orbits (1-tori) in $\Pi$. The unstable 

919: and stable manifolds of an 1-torus given by $I = A$ (a constant) in $\A$ are 

920: \[

921: W^{u,s}(A) = \bigcup_{\vq \in \A ,\ I(\vq ) = A} \F^{u,s}(\vq) \ ,

922: \]

923: which are three dimensional (taking into account the time dimension). 

924: \begin{theorem}[Arnold Diffusion]

925: For any $A_1$ and $A_2$ such that 

926: \[

927: \frac{1}{h^3} \ln \rho_0 < A_1 < A_2 < +\infty \ ,

928: \]

929: where $\rho_0 = 1+ h^2 \left ( 3\tan \frac{\pi}{3}\right )^2$, 

930: there exists a $\al_0 >0$ such that when $|\al | > \al_0$, 

931: $W^u(A_1)$ and $W^s(A_2)$ are connected by an orbit.

932: \label{nrthm}

933: \end{theorem}

934: \begin{proof}

935: One can check directly that for any $\vq \in \A$, $F_1 (\vq ) = -2$ and 

936: $\pa F_1 (\vq ) / \pa \vq = 0$ [cf: (\ref{deff}) and (\ref{melv})].

937: Now consider $W^s(a_1)$ and $W^u(a_2)$. Along any 

938: orbit $\vq^{\ s}(t)$ in $W^s(a_1)$, we have 

939: \begin{eqnarray*}

940: & & \lim_{t \ra +\infty} F_1 (\vq^{\ s} (t)) - F_1 (\vq^{\ s} (t)) = -2 - F_1 (\vq^{\ s} (t)) \\

941: & & = \int_{t}^{+\infty} \frac{dF_1}{dt} dt = -i\e \int_{t}^{+\infty}\{ F_1, H_1 \} dt\ ,  \\

942: & & \lim_{t \ra +\infty} I (\vq^{\ s} (t)) - I (\vq^{\ s} (t)) = a_1 - I (\vq^{\ s} (t)) \\

943: & & = \int_{t}^{+\infty} \frac{dI}{dt} dt = -i\e \int_{t}^{+\infty}\{ I, H_1 \} dt\ , 

944: \end{eqnarray*}

945: where 

946: \[

947: \{ f,g \} = \sum_{n=0}^{N-1} \rho_n \left [ \frac{\pa f}{\pa q_n} \frac{\pa g}{\pa \overline{q_n}}

948: - \frac{\pa f}{\pa \overline{q_n}} \frac{\pa g}{\pa q_n} \right ]

949: \]

950: is the Poisson bracket. Notice that $\{ F_1, H_0 \} = \{ I, H_0 \} = 0$ at any $\vq \in \cS$. 

951: Since $\pa H_1 /\pa \vq \ra 0$ exponentially as $t \ra +\infty$, the corresponding integrals converge. 

952: Similarly along any orbit $\vq^{\ u}(t)$ in $W^u(a_2)$, we have

953: \begin{eqnarray*}

954: & & F_1 (\vq^{\ u} (t)) - \lim_{t \ra -\infty} F_1 (\vq^{\ u} (t)) = F_1 (\vq^{\ u}(t))+2 \\

955: & & = \int^{t}_{-\infty} \frac{dF_1}{dt} dt = -i\e \int^{t}_{-\infty}\{ F_1, H_1 \} dt\ ,  \\

956: & & I (\vq^{\ u} (t)) - \lim_{t \ra -\infty} I (\vq^{\ u} (t)) = I (\vq^{\ u} (t)) - a_2 \\

957: & & = \int^{t}_{-\infty} \frac{dI}{dt} dt = -i\e \int^{t}_{-\infty}\{ I, H_1 \} dt\ . 

958: \end{eqnarray*}

959: Thus a neighborhood of $W^s(a_1)$ in $\cS$ can be parameterized by ($\ga , t_0 , t , F_1 , I$)

960: where $\ga$ is defined in (\ref{us}), $t_0$ is the initial time, and 

961: \[

962: F_1 = F_1 (\vq^{\ s} (t)) + v_1^s \ , \quad  I = I (\vq^{\ s} (t)) + v_2^s \ .

963: \]

964: When $v_1^s = v_2^s = 0$, we get $W^s(a_1)$. Thus $W^s(a_1)\cap W^u(a_2) \neq \emptyset$ 

965: if and only if 

966: \[

967: F_1 (\vq^{\ s} (t))=F_1 (\vq^{\ u} (t))\ , \quad I (\vq^{\ s} (t)) = I (\vq^{\ u} (t)) \ ,

968: \]

969: for some $\ga$ and $t_0$. In such a case, there is an orbit 

970: $\vq (t,\e ) \subset W^s(a_1)\cap W^u(a_2)$ along which 

971: \[

972: \int^{+\infty}_{-\infty}\{ F_1, H_1 \} |_ {\vq (t,\e )}dt = 0 \ , 

973: \quad a_1 -a_2 = -i\e \int_{-\infty}^{+\infty}\{ I, H_1 \} |_ {\vq (t,\e )}dt \ .

974: \]

975: Let $\vq (t,0 )$ be an orbit of DNLS such that $\vq (0,0 )$ and  $\vq (0,\e )$ have the 

976: same stable fiber base point. Then 

977: \[

978: \|  \vq (0,0 ) - \vq (0,\e ) \| \sim \O (\e ) \ .

979: \]

980: For any small $\dl >0$, there is a $T>0$ such that 

981: \[

982: \left | \int^{\pm T}_{\pm \infty}\{ F_1, H_1 \} |_ {\vq (t,\e )}dt \right | < \dl \ , 

983: \quad 

984: \left | \int^{\pm T}_{\pm \infty}\{ I, H_1 \} |_ {\vq (t,\e )}dt  \right | < \dl \ , 

985: \quad \forall \e \in [0,\e_0]\ ,

986: \]

987: for some $\e_0 >0$. For this $T$, when $\e$ is sufficiently small,

988: \[

989: \|  \vq (t,0 ) - \vq (t,\e ) \| \sim \O (\e ) \ , \quad \forall t \in [-T,T]\ .

990: \]

991: Thus 

992: \begin{eqnarray*}

993: & & \int^{+ T}_{-T}\{ F_1, H_1 \} |_ {\vq (t,\e )}dt =

994: \int^{+ T}_{-T}\{ F_1, H_1 \} |_ {\vq (t,0 )}dt + \O (\e ) \ , \\

995: & & \int^{+ T}_{-T}\{ I, H_1 \} |_ {\vq (t,\e )}dt =

996: \int^{+ T}_{-T}\{ I, H_1 \} |_ {\vq (t,0 )}dt + \O (\e ) \ .

997: \end{eqnarray*}

998: Finally we have 

999: \begin{eqnarray}

1000: & & \int^{+\infty}_{-\infty}\{ F_1, H_1 \} |_ {\vq (t,\e )}dt = 

1001: \int^{+\infty}_{-\infty}\{ F_1, H_1 \} |_ {\vq (t,0 )}dt + \O (\dl ) = 0 \ ,  \label{meq1}\\

1002: & & a_1 -a_2 = -i\e \int_{-\infty}^{+\infty}\{ I, H_1 \} |_ {\vq (t,\e )}dt

1003: = -i\e \int_{-\infty}^{+\infty}\{ I, H_1 \} |_ {\vq (t,0 )}dt + \O (\dl \e ) \ . \label{meq2}

1004: \end{eqnarray}

1005: Next we solve the above equations at the leading order in $\dl$ and $\e$. Rewrite the derivative 

1006: given in (\ref{melv}) as follows

1007: \[

1008: \pa F_1 / \pa q_n = V_n e^{-i\hga}\ ,

1009: \]

1010: where $\hga = \ga + 2(a^2-\om^2)t_0$, $t_0 = p /\mu$, and $V_n$ represents the rest 

1011: which does not depend on $\hga$. We also rewrite $q_c$ (\ref{us}) and $q_n$ (\ref{horbit}) 

1012: as 

1013: \[

1014: q_c = \hq_c e^{i\hga}\ , \quad q_n = \hq_n e^{i\hga}\ ,

1015: \]

1016: where $\hq_c$ and $\hq_n$ do not depend on $\hga$. Then substitute all these into the leading 

1017: order terms in (\ref{meq1})-(\ref{meq2}), we obtain the following equations

1018: \begin{eqnarray}

1019: & & \al \sqrt{M_1^2+M_2^2} \sin (t_0 +\th_1) + \sqrt{M_3^2+M_4^2} \sin (2\hga +\th_2) = 0 \ ,

1020: \label{mele1} \\

1021: & & a_1-a_2 = 2\e \sqrt{M_5^2+M_6^2} \sin (2\hga +\th_3) \ , \label{mele2}

1022: \end{eqnarray}

1023: where 

1024: \begin{eqnarray*}

1025: & & \cos \th_1 = \frac{M_1}{\sqrt{M_1^2+M_2^2}} \ , \quad 

1026: \sin \th_1 = \frac{M_2}{\sqrt{M_1^2+M_2^2}} \ , \quad 

1027: \cos \th_2 = \frac{M_3}{\sqrt{M_3^2+M_4^2}} \ ,  \\

1028: & & \sin \th_2 = \frac{M_4}{\sqrt{M_3^2+M_4^2}} \ , \quad 

1029: \cos \th_3 = \frac{M_5}{\sqrt{M_5^2+M_6^2}} \ , \quad 

1030: \sin \th_3 = \frac{M_6}{\sqrt{M_5^2+M_6^2}} \ ,  \\

1031: & & M_1 = \int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \cos \tau \ \rho_n \text{ Im } 

1032: [V_n G_n^1 ] \ d \tau \ , \\

1033: & & M_2 = -\int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \sin \tau \ \rho_n \text{ Im } 

1034: [V_n G_n^1 ] \ d \tau \ , \\

1035: & & M_3 = \int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \rho_n \text{ Re } 

1036: [V_n G_n^2 ]\ d \tau \ , \quad 

1037: M_4 = -\int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \rho_n \text{ Im } 

1038: [V_n G_n^2 ] \ d \tau \ , \\

1039: & & M_5 = -\int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \text{ Re } G_n^3 \ d \tau \ , \quad 

1040: M_6 = \int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \text{ Im } G_n^3 \ d \tau \ , \\

1041: & & G_n^1 = \frac{\hq_{n+1} - 2\hq_n + \hq_{n-1}}{h^2} \ , \quad 

1042: G_n^2 = 2\frac{\overline{\hq_{n+1}} - 2\overline{\hq_n} + \overline{\hq_{n-1}}}{h^2} \ , \\

1043: & & G_n^3 = -2\frac{\overline{\hq_n}(\overline{\hq_{n+1}} - 2\overline{\hq_n} + 

1044: \overline{\hq_{n-1}})}{h^2} \ ,

1045: \end{eqnarray*}

1046: and $\tau = t +p/\mu$. Equations (\ref{mele1})-(\ref{mele2}) are easily solvable as long as

1047: neither $\sqrt{M_1^2+M_2^2}$ nor $\sqrt{M_5^2+M_6^2}$ vanishes. In Figures \ref{nr1}-\ref{nr3}, we 

1048: plot the graphs of them as functions of $a$. We solve equation (\ref{mele2}) for $\hga$, 

1049: then solve equation (\ref{mele1}) for $t_0$. Thus when $|\al |$ is large enough, we have 

1050: solutions. It is also clear from equations (\ref{mele1})-(\ref{mele2}) that $W^s(a_1)$ and 

1051: $W^u(a_2)$ intersect transversally. Then we can choose a sequence

1052: \[

1053: A_1=a_1 < a_2 < \cdots < a_N=A_2 \ ,

1054: \]

1055: such that $W^s(a_j)$ and $W^u(a_{j+1})$ ($ 1\leq j \leq N-1$) intersect transversally.

1056: The period of the 1-tori (\ref{us}) is $\pi / a^2$. Thus we can always choose the $a_j$'s 

1057: such that the frequencies $\frac{1}{2a^2}$ of the corresponding 1-tori are irrational. 

1058: Therefore we obtain a transition chain. Apply Lemma \ref{al} to the period-$2\pi$ map 

1059: of the DNLS, we obtain the claim of the theorem.

1060: \end{proof}

1061: \begin{remark}

1062: The constant of motion $I$ is equivalent to $F_2$ for $z_2^c =1$ (\ref{deff}).

1063: The continuum limit of $H_1$ has the form 

1064: \[

1065: H_1 = \al \sin t \int_0^1 |q_x|^2 dx + \int_0^1 (q_x^2 + \overline{q_x}^{\ 2})dx \ ,

1066: \]

1067: which is suitable for the NLS setting. One can regularize the perturbation by replacing the 

1068: partial derivative $\pa_x$ in $H_1$ by a Fourier multiplier $\hat{\pa}_x$, e.g. a 

1069: Galerkin truncation. One can use the constant of motion 

1070: \[

1071: I = \int_0^1 |q|^2 dx 

1072: \]

1073: to build the second Melnikov-Arnold integral. 

1074: \end{remark}

1075: 

1076: \begin{figure}

1077: \includegraphics{fig01.eps}

1078: \caption{Dynamics inside $\Pi$ (non-resonant case).}

1079: \label{dpin}

1080: \end{figure}

1081: 

1082: \begin{figure}

1083: \includegraphics{fig1.eps}

1084: \caption{The graph of $\sqrt{M_1^2+M_2^2}$ as a function of $a$ in the non-resonant

1085: case $\om =0$.}

1086: \label{nr1}

1087: \end{figure}

1088: \begin{figure}

1089: \includegraphics{fig2.eps}

1090: \caption{The graph of $\sqrt{M_3^2+M_4^2}$ as a function of $a$ in the non-resonant

1091: case $\om =0$.}

1092: \label{nr2}

1093: \end{figure}

1094: \begin{figure}

1095: \includegraphics{fig3.eps}

1096: \caption{The graph of $\sqrt{M_5^2+M_6^2}$ as a function of $a$ in the non-resonant

1097: case $\om =0$.}

1098: \label{nr3}

1099: \end{figure}

1100: 

1101: \begin{figure}

1102: \includegraphics{fig02.eps}

1103: \caption{Dynamics inside $\Pi$ (resonant case).}

1104: \label{dpir}

1105: \end{figure}

1106: 

1107: \begin{figure}

1108: \includegraphics{fig4.eps}

1109: \caption{The graph of $\sqrt{M_1^2+M_2^2}$ as a function of $a$ in the resonant

1110: case $\om =10$.}

1111: \label{r1}

1112: \end{figure}

1113: \begin{figure}

1114: \includegraphics{fig5.eps}

1115: \caption{The graph of $\sqrt{M_3^2+M_4^2}$ as a function of $a$ in the resonant

1116: case $\om =10$.}

1117: \label{r2}

1118: \end{figure}

1119: \begin{figure}

1120: \includegraphics{fig6.eps}

1121: \caption{The graph of $\sqrt{M_5^2+M_6^2}$ as a function of $a$ in the resonant

1122: case $\om =10$.}

1123: \label{r3}

1124: \end{figure}

1125: 

1126: \subsection{Arnold Diffusion of DNLS ($N=3$, Resonant Case)}

1127: 

1128: In this subsection, we prove the existence of Arnold diffusion for a perturbed 

1129: DNLS when $N=3$, which is a 5-dimensional system. We will study here the case 

1130: that there is a resonance ($\om > 3\tan{\pi \over 3}$) inside the invariant 

1131: plane $\Pi$ (\ref{invp}).

1132: 

1133: Consider the following perturbation of the DNLS (\ref{DNLS})

1134: \[

1135: H = H_0 +\e (H_1 +H_2) \ , 

1136: \]

1137: where

1138: \[

1139: H_1 = \al \sum_{n=0}^{N-1}(q_n +\overline{q_n}) \ , \quad 

1140: H_2 = \sin t \sum_{n=0}^{N-1}

1141: \left | \frac{q_n-q_{n-1}}{h}\right |^2 \ ,

1142: \]

1143: where $\al$ is a real parameter. Under this perturbation, dynamics inside $\Pi$ is 

1144: changed. Due to the resonance $a =\om$ in (\ref{us}), some tori do not persist into 

1145: KAM tori. A secondary separatrix is generated. Inside this separatrix are the 

1146: secondary tori [Figure \ref{dpir}]. As can be seen below, resonance does not add 

1147: difficulty to the Arnold diffusion problem. Instead of $I$ in last subsection, we use 

1148: \[

1149: \hH = H = H_0 +\e H_1 \ ,

1150: \]

1151: to build one of the two Melnikov-Arnold integrals. 

1152: Restricted to $\Pi$, the level sets of $\hH$ produces Figure \ref{dpir}. The unstable 

1153: and stable manifolds of an 1-torus given by $\hH = A$ (a constant) in $\A$ are 

1154: \[

1155: W^{u,s}(A) = \bigcup_{\vq \in \A ,\ \hH(\vq ) = A} \F^{u,s}(\vq) \ .

1156: \]

1157: Let 

1158: \[

1159: A_* = \frac{1}{h^3}\left [ 2\left ( 3\tan \frac{\pi}{3}\right )^2

1160: -\frac{2}{h^2}(1+\om^2h^2) \ln \rho_0 \right ]\ ,

1161: \]

1162: where $\rho_0 = 1+ h^2 \left ( 3\tan \frac{\pi}{3}\right )^2$.

1163: \begin{theorem}[Arnold Diffusion]

1164: For any $A_1$ and $A_2$ such that 

1165: \[

1166: A_* < A_1 < A_2 < +\infty \ ,

1167: \]

1168: there exists a $\al_0 >0$ such that when $|\al | > \al_0$, 

1169: $W^u(A_1)$ and $W^s(A_2)$ are connected by an orbit.

1170: \end{theorem}

1171: \begin{proof}

1172: Again one can check directly that for any $\vq \in \A$, $F_1 (\vq ) = -2$ and 

1173: $\pa F_1 (\vq ) / \pa \vq = 0$ [cf: (\ref{deff}) and (\ref{melv})].

1174: Similar to the proof of Theorem \ref{nrthm}, consider $W^s(a_1)$ and $W^u(a_2)$. 

1175: Along any orbit $\vq^{\ s}(t)$ in $W^s(a_1)$, we have 

1176: \begin{eqnarray*}

1177: & & \lim_{t \ra +\infty} F_1 (\vq^{\ s} (t)) - F_1 (\vq^{\ s} (t)) = -2 - F_1 (\vq^{\ s} (t)) \\

1178: & & = \int_{t}^{+\infty} \frac{dF_1}{dt} dt = -i\e \int_{t}^{+\infty}\{ F_1, H_1+H_2 \} dt\ ,  \\

1179: & & \lim_{t \ra +\infty} \hH (\vq^{\ s} (t)) - \hH (\vq^{\ s} (t)) = a_1 - \hH (\vq^{\ s} (t)) \\

1180: & & = \int_{t}^{+\infty} \frac{d\hH}{dt} dt = -i\e \int_{t}^{+\infty}\{ \hH, H_2 \} dt\ , 

1181: \end{eqnarray*}

1182: where 

1183: \[

1184: \{ f,g \} = \sum_{n=0}^{N-1} \rho_n \left [ \frac{\pa f}{\pa q_n} \frac{\pa g}{\pa \overline{q_n}}

1185: - \frac{\pa f}{\pa \overline{q_n}} \frac{\pa g}{\pa q_n} \right ]

1186: \]

1187: is the Poisson bracket. Notice that $\{ F_1, H_0 \} = \{ \hH , \hH\} = 0$ at any $\vq \in \cS$. 

1188: Since $\frac{\pa F_1}{\pa \vq}, \frac{\pa H_2}{\pa \vq} \ra 0$ exponentially as $t \ra +\infty$, 

1189: the corresponding integrals converge. 

1190: Similarly along any orbit $\vq^{\ u}(t)$ in $W^u(a_2)$, we have

1191: \begin{eqnarray*}

1192: & & F_1 (\vq^{\ u} (t)) - \lim_{t \ra -\infty} F_1 (\vq^{\ u} (t)) = F_1 (\vq^{\ u}(t))+2 \\

1193: & & = \int^{t}_{-\infty} \frac{dF_1}{dt} dt = -i\e \int^{t}_{-\infty}\{ F_1, H_1+H_2 \} dt\ ,  \\

1194: & & \hH (\vq^{\ u} (t)) - \lim_{t \ra -\infty} \hH (\vq^{\ u} (t)) = \hH (\vq^{\ u} (t)) - a_2 \\

1195: & & = \int^{t}_{-\infty} \frac{d\hH}{dt} dt = -i\e \int^{t}_{-\infty}\{ \hH, H_2 \} dt\ . 

1196: \end{eqnarray*}

1197: Thus a neighborhood of $W^s(a_1)$ in $\cS$ can be parameterized by ($\vth , t_0 , t , F_1 , \hH$)

1198: where $\vth$ is the angle of the 1-torus $\hH = a_1$ in $\Pi$, $t_0$ is the initial time, and 

1199: \[

1200: F_1 = F_1 (\vq^{\ s} (t)) + v_1^s \ , \quad  \hH = \hH (\vq^{\ s} (t)) + v_2^s \ .

1201: \]

1202: When $v_1^s = v_2^s = 0$, we get $W^s(a_1)$. Thus $W^s(a_1)\cap W^u(a_2) \neq \emptyset$ 

1203: if and only if 

1204: \[

1205: F_1 (\vq^{\ s} (t))=F_1 (\vq^{\ u} (t))\ , \quad  \hH (\vq^{\ s} (t)) = \hH (\vq^{\ u} (t)) \ ,

1206: \]

1207: for some $\vth$ and $t_0$. In such a case, there is an orbit 

1208: $\vq (t,\e ) \subset W^s(a_1)\cap W^u(a_2)$ along which 

1209: \[

1210: \int^{+\infty}_{-\infty}\{ F_1, H_1+H_2 \} |_ {\vq (t,\e )}dt = 0 \ , 

1211: \quad a_1 -a_2 = -i\e \int_{-\infty}^{+\infty}\{ \hH , H_2 \} |_ {\vq (t,\e )}dt \ .

1212: \]

1213: Let $\vq (t,0 )$ be an orbit of DNLS such that $\vq (0,0 )$ and  $\vq (0,\e )$ have the 

1214: same stable fiber base point. Then 

1215: \[

1216: \|  \vq (0,0 ) - \vq (0,\e ) \| \sim \O (\e ) \ .

1217: \]

1218: For any small $\dl >0$, there is a $T>0$ such that 

1219: \[

1220: \left | \int^{\pm T}_{\pm \infty}\{ F_1, H_1+H_2 \} |_ {\vq (t,\e )}dt \right | < \dl \ , 

1221: \quad 

1222: \left | \int^{\pm T}_{\pm \infty}\{ \hH , H_2 \} |_ {\vq (t,\e )}dt  \right | < \dl \ , 

1223: \quad \forall \e \in [0,\e_0]\ ,

1224: \]

1225: for some $\e_0 >0$. For this $T$, when $\e$ is sufficiently small,

1226: \[

1227: \|  \vq (t,0 ) - \vq (t,\e ) \| \sim \O (\e ) \ , \quad \forall t \in [-T,T]\ .

1228: \]

1229: Thus 

1230: \begin{eqnarray*}

1231: & & \int^{+ T}_{-T}\{ F_1, H_1+H_2 \} |_ {\vq (t,\e )}dt =

1232: \int^{+ T}_{-T}\{ F_1, H_1+H_2 \} |_ {\vq (t,0 )}dt + \O (\e ) \ , \\

1233: & & \int^{+ T}_{-T}\{ \hH , H_2 \} |_ {\vq (t,\e )}dt =

1234: \int^{+ T}_{-T}\{ \hH , H_2 \} |_ {\vq (t,0 )}dt + \O (\e ) \ .

1235: \end{eqnarray*}

1236: Finally we have 

1237: \begin{eqnarray}

1238: & & \int^{+\infty}_{-\infty}\{ F_1, H_1+H_2 \} |_ {\vq (t,\e )}dt = 

1239: \int^{+\infty}_{-\infty}\{ F_1, H_1+H_2 \} |_ {\vq (t,0 )}dt + \O (\dl ) = 0 \ ,  \label{mer1}\\

1240: & & a_1 -a_2 = -i\e \int_{-\infty}^{+\infty}\{ \hH , H_2 \} |_ {\vq (t,\e )}dt

1241: = -i\e \int_{-\infty}^{+\infty}\{ H_0, H_2 \} |_ {\vq (t,0 )}dt + \O (\dl \e ) \ . \label{mer2}

1242: \end{eqnarray}

1243: To the leading order terms in (\ref{mer1})-(\ref{mer2}), we obtain the following equations

1244: \begin{eqnarray}

1245: & & \sqrt{M_1^2+M_2^2} \sin (t_0 +\th_1) + \al \sqrt{M_3^2+M_4^2} \sin (\hga +\th_2) = 0 \ ,

1246: \label{melr1} \\

1247: & & a_1-a_2 = 2\e \sqrt{M_5^2+M_6^2} \sin (t_0 +\th_3) \ , \label{melr2}

1248: \end{eqnarray}

1249: where 

1250: \begin{eqnarray*}

1251: & & \cos \th_1 = \frac{M_1}{\sqrt{M_1^2+M_2^2}} \ , \quad 

1252: \sin \th_1 = \frac{M_2}{\sqrt{M_1^2+M_2^2}} \ , \quad 

1253: \cos \th_2 = \frac{M_3}{\sqrt{M_3^2+M_4^2}} \ ,  \\

1254: & & \sin \th_2 = \frac{M_4}{\sqrt{M_3^2+M_4^2}} \ , \quad 

1255: \cos \th_3 = \frac{M_5}{\sqrt{M_5^2+M_6^2}} \ , \quad 

1256: \sin \th_3 = \frac{M_6}{\sqrt{M_5^2+M_6^2}} \ ,  \\

1257: & & M_1 = \int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \cos \tau \ \rho_n \text{ Im } 

1258: [V_n G_n^1 ] \ d \tau \ , \\

1259: & & M_2 = -\int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \sin \tau \ \rho_n \text{ Im } 

1260: [V_n G_n^1 ] \ d \tau \ , \\

1261: & & M_3 = -\int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \rho_n \text{ Re } 

1262: [V_n]\ d \tau \ , \quad 

1263: M_4 = \int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \rho_n \text{ Im } 

1264: [V_n] \ d \tau \ , \\

1265: & & M_5 = \int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \cos \tau \text{ Im } [G_n^1G_n^2] \ d \tau \ , \\

1266: & & M_6 = -\int_{-\infty}^{+\infty} \sum_{n=0}^{N-1} \sin \tau \text{ Im } [G_n^1G_n^2] \ d \tau \ , \\

1267: & & G_n^1 = \frac{\hq_{n+1} - 2\hq_n + \hq_{n-1}}{h^2} \ , \\

1268: & & G_n^2 = \frac{1}{h^2}\left [\overline{\hq_{n+1}} - 2\overline{\hq_n} + \overline{\hq_{n-1}}

1269: \right ] 

1270: + |\hq_n|^2 \left [\overline{\hq_{n+1}} + \overline{\hq_{n-1}}\right ] -2\om^2 \overline{\hq_n}\ ,

1271: \end{eqnarray*}

1272: and $\tau = t +p/\mu$. Equations (\ref{melr1})-(\ref{melr2}) are easily solvable as long as

1273: neither $\sqrt{M_3^2+M_4^2}$ nor $\sqrt{M_5^2+M_6^2}$ vanishes. In Figures \ref{r1}-\ref{r3}, we 

1274: plot the graphs of them as functions of $a$. We solve equation (\ref{melr2}) for $t_0$, 

1275: then solve equation (\ref{melr1}) for $\hga$. Thus when $|\al |$ is large enough, we have 

1276: solutions. It is also clear from equations (\ref{melr1})-(\ref{melr2}) that $W^s(a_1)$ and 

1277: $W^u(a_2)$ intersect transversally. Then we can choose a sequence

1278: \[

1279: A_1=a_1 < a_2 < \cdots < a_N=A_2 \ ,

1280: \]

1281: such that $W^s(a_j)$ and $W^u(a_{j+1})$ ($ 1\leq j \leq N-1$) intersect transversally.

1282: The period of the 1-tori ($\hH = a_j$) depends on $a_j$ no matter they are KAM tori or 

1283: secondary tori. We can always choose the $a_j$'s 

1284: such that the frequencies of the corresponding 1-tori are irrational. 

1285: We can use one secondary torus inside and close enough to the separatrix (Figure \ref{dpir})

1286: to bridge across the resonant region of width $\O(\sqrt{\e})$.

1287: Therefore we obtain a transition chain. Apply Lemma \ref{al} to the period-$2\pi$ map 

1288: of the DNLS, we obtain the claim of the theorem.

1289: \end{proof}

1290: \begin{remark}

1291: The continuum limit of $H_1$ and $H_2$ have the form 

1292: \[

1293: H_1 = \al \int_0^1 (q +\bq ) dx \ , \quad 

1294: H_2 = \sin t \int_0^1 |q_x|^2 dx \ ,

1295: \]

1296: which is suitable for the NLS setting. One can regularize the perturbation by replacing the 

1297: partial derivative $\pa_x$ in $H_2$ by a Fourier multiplier $\hat{\pa}_x$, e.g. a 

1298: Galerkin truncation. 

1299: \end{remark}

1300: 

1301: 

1302: 

1303: 

1304: 

1305: 

1306: 

1307: \begin{thebibliography}{99}

1308: 

1309: \bibitem{AL76} M. Ablowitz, J. Ladik, A nonlinear difference scheme and inverse 

1310: scattering, {\it Studies in Appl. Math.} {\bf 55} (1976), 213-229.

1311: 

1312: \bibitem{APT04} M. Ablowitz, B. Prinari, A. Trubatch, {\it Discrete and Continuous

1313: Nonlinear Schr{\"{o}}dinger Systems}, Cambridge University Press, 2004.

1314: 

1315: \bibitem{Arn64} V. Arnold, Instability of dynamical systems with several degrees of 

1316: freedom, {\it Soviet Math. Doklady} {\bf 5} (1964), 581-585.

1317: 

1318: \bibitem{AA68} V. Arnold, A. Avez, {\it Ergodic Problems of Classical Mechanics}, 

1319: W. A. Benjamin, Inc., New York, 1968. [Page 112, Lemma 23.8]

1320: 

1321: \bibitem{BBB03} M. Berti, L. Biasco, P. Bolle, Drift in phase space: a new

1322: variational mechanism with optimal diffusion time, {\it J. Math. Pures Appl.}

1323: {\bf 82, no.6} (2003), 613-664.

1324: 

1325: \bibitem{BT99} S. Bolotin, D. Treschev, Unbounded growth of energy in nonautonomous 

1326: Hamiltonian systems, {\it Nonlinearity} {\bf 12} (1999), 365-388.

1327: 

1328: \bibitem{CY04} C. Cheng, J. Yan, Existence of diffusion orbits in a priori unstable 

1329: Hamiltonian systems, {\it J. Diff. Geom.} {\bf 67, no.3} (2004), 457-517.

1330: 

1331: \bibitem{CG94} L. Chierchia, G. Gallavotti, Drift and diffusion in phase space, 

1332: {\it Ann. Inst. H. Poincar{\'{e}} Phys. Theor.} {\bf 60} (1994), 1-144.

1333: 

1334: \bibitem{CV00} L. Chierchia, E. Valdinoci, A note on the construction of Hamiltonian 

1335: trajectories along heteroclinic chains, {\it Forum Math.} {\bf 12} (2000), 247-255.

1336: 

1337: \bibitem{Cre97} J. Cresson, A $\lambda$-lemma for partially hyperbolic tori and the 

1338: obstruction property, {\it Lett. Math. Phys.} {\bf 42} (1997), 363-377.

1339: 

1340: \bibitem{DLS06} A. Delshams, R. de la Llave, T. Seara, A geometric mechanism for 

1341: diffusion in Hamiltonian systems overcoming the large gap problem: Heuristics and 

1342: rigorous verification on a model, {\it Memoirs of AMS} {\bf 179, no.844} (2006).

1343: 

1344: \bibitem{DLS00} A. Delshams, R. de la Llave, T. Seara, A geometric approach to the 

1345: existence of orbits with unbounded energy in generic periodic perturbations by potential

1346: of generic geodesic flows on $\mathbb{T}^2$, {\it Comm. Math. Phys.} {\bf 209} (2000),

1347: 353-392.

1348: 

1349: \bibitem{Dou88} R. Douady, Stabilit{\'{e}} ou instabilit{\'{e}} des points fixes 

1350: elliptiques, {\it Ann Sci. de l'ENS} {\bf 21} (1988), 1-46.

1351: 

1352: \bibitem{FM00} E. Fontich, P. Martin, Differentiable invariant manifolds for 

1353: partially hyperbolic tori and a lambda lemma, {\it Nonlinearity} {\bf 13} (2000), 

1354: 1561-1593. [Page 1585]

1355: 

1356: \bibitem{FM03} E. Fontich, P. Martin, Hamiltonian systems with orbits covering densely 

1357: submanifolds of small codimension, {\it Nonlinear Analysis} {\bf 52} (2003), 

1358: 315-327. 

1359: 

1360: \bibitem{LW04} R. de la Llave, C.E. Wayne, Whiskered and low dimensional tori in

1361: nearly integrable Hamiltonian systems, {\it Math. Phys. Electron. J.} {\bf 10} (2004), 

1362: paper 5, 45pp.

1363: 

1364: \bibitem{Li92} Y. Li, B{\"{a}}cklund transformations and homoclinic structures for 

1365: the discrete NLS equation, {\it Phys. Lett. A} {\bf 163} (1992), 181-187.

1366: 

1367: \bibitem{Li03} Y. Li, Homoclinic tubes in discrete nonlinear 

1368: Schr{\"{o}}dinger equation under Hamiltonian perturbations, {\it 

1369: Nonlinear Dynamics} {\bf vol.31, no.4} (2003), 393-434.

1370: 

1371: \bibitem{Li04} Y. Li, {\it Chaos in Partial Differential Equations}, 

1372: International Press, 2004.

1373: 

1374: \bibitem{Li99} Y. Li, Homoclinic tubes in nonlinear Schr{\"{o}}dinger 

1375: equation under Hamiltonian perturbations, {\it Progr. Theoret. Phys.} 

1376: {\bf 101, no.3} (1999), 559-577.

1377: 

1378: \bibitem{Li04a} Y. Li, Persistent homoclinic orbits of nonlinear Schr{\"{o}}dinger 

1379: equation under singular perturbations, {\it Dynamics of PDE} {\bf 1, no.1} (2004),

1380: 87-123.

1381: 

1382: \bibitem{Li04b} Y. Li, Existence of chaos for nonlinear Schr{\"{o}}dinger 

1383: equation under singular perturbations, {\it Dynamics of PDE} {\bf 1, no.2} (2004),

1384: 225-237.

1385: 

1386: \bibitem{Li03a} Y. Li, Chaos and shadowing lemma for autonomous 

1387: systems of infinite dimensions, {\it J. Dyn. Diff. Eq.} {\bf vol.15, no.4}

1388: (2003), 699-730. [Page 705]

1389: 

1390: \bibitem{Li03b} Y. Li, Chaos and shadowing around a homoclinic tube,

1391: {\it Abstract and Applied Analysis} {\bf vol.2003, no.16} (2003), 923-931.

1392: 

1393: \bibitem{Li04c} Y. Li, Homoclinic tubes and chaos in perturbed Sine-Gordon 

1394: equation, {\it Chaos, Solitons and Fractals} {\bf vol.20, no.4} (2004), 791-798.

1395: 

1396: \bibitem{Li06c} Y. Li, Chaos and shadowing around a heteroclinically 

1397: tubular cycle with an application to sine-Gordon equation, {\it Studies

1398: in Appl. Math.} {\bf 116} (2006), 145-171.

1399: 

1400: \bibitem{Li06} Y. Li, Tubes in dynamical systems, {\it http://www.math.missouri.edu/{\~{ }}cli} 

1401: (2006), Submitted.

1402: 

1403: \bibitem{Mar96} J. Marco, Transition le long des cha{\^{i}}nes de tores invariants 

1404: pour les syst{\`{e}}mes Hamiltoniens analytiques, {\it Ann. Inst. H. Poincar{\'{e}}} 

1405: {\bf 64} (1996), 205-252.

1406: 

1407: \bibitem{Mat04} J. Mather, Arnold diffusion. I: Announcement of results, 

1408: {\it J. Math. Sci.} {\bf 124, no.5} (2004), 5275-5289.

1409: 

1410: \bibitem{Per98} P. Perfetti, Fixed point theorems in the Arnold model about 

1411: instability of Hamiltonian dynamical systems, {\it Discrete Contin. Dynam. Syst.} 

1412: {\bf 4} (1998), 379-391.

1413: 

1414: \bibitem{Xia98} Z. Xia, Arnold diffusion: a variational construction, 

1415: {\it Proc. ICM} {\bf vol.II} (1998), 867-877.

1416: 

1417: 

1418: \end{thebibliography}

1419: 

1420: 

1421: \end{document}

1422: 

1423: %-----------------------------------------------------------------------

1424: % End of article.tex

1425: %-----------------------------------------------------------------------

1426: