Content-Type: multipart/mixed; boundary="-------------0407291121184" This is a multi-part message in MIME format. ---------------0407291121184 Content-Type: text/plain; name="04-235.keywords" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="04-235.keywords" NonLinear Schrodinger Equations, Semiclassical Limit, Double-Well Potential ---------------0407291121184 Content-Type: application/x-tex; name="StabilityNLS_MP_ARC.TEX" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="StabilityNLS_MP_ARC.TEX" % % % Author: Andrea Sacchetti % Addres: Dipartimento di Matematica, % Universita' di Modena e Reggio Emilia, % Via Campi 213/B, 41100 Modena - Italy % email: Sacchet@unimo.it % % This is a LATEX file \documentclass[reqno]{amsart} \usepackage{graphicx}% Include figure files \newtheorem{lemma}{Lemma} \newtheorem{theorem}{Theorem} \newtheorem{corollary}{Corollary} \newtheorem{hypothesis}{Hypothesis} \newtheorem{remark}{Remark} \newtheorem{definition}{Definition} \newcommand{\be}{\begin{eqnarray}} \newcommand{\ee}{\end{eqnarray}} \newcommand{\bee}{\begin{eqnarray*}} \newcommand{\eee}{\end{eqnarray*}} \newcommand{\R}{\mbox {\sc R}} \newcommand{\N}{\mbox {\sc N}} \newcommand{\Z}{\mbox {\sc Z}} \newcommand{\C}{\mbox {\sc C}} \newcommand{\D}{\mbox {\sc D}} \newcommand{\I}{\mbox {\sc 1}} \newcommand{\0}{\mbox {\sc 0}} \newcommand{\Rp}{\mbox {\sc r}} \newcommand{\Np}{\mbox {\sc n}} \newcommand{\Zp}{\mbox {\sc z}} \newcommand{\Cp}{\mbox {\sc c}} \newcommand{\K}{{\it K}} \newcommand{\II}{{\it I}} \newcommand{\E}{{\it E}} \newcommand{\W}{{\it W}} \newcommand{\F}{{\it F}} \newcommand{\U}{{\it U}} \newcommand{\V}{{\it V}} \newcommand{\f}{\mbox {\sf f}} \newcommand{\g}{\mbox {\sf g}} \newcommand{\h}{\mbox {\sf h}} \newcommand{\s}{{\it S}} \newcommand{\asy}{{\it O}} \newcommand{\ind}{\hskip 0.5cm} \newcommand{\case}[2]{\textstyle{\frac{#1}{#2}}} %\parindent=0pt %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{document} \title [Semiclassical NonLinear Schr\"odinger Equations]{Nonlinear double well Schr\"odinger equations in the semiclassical limit} \author {Andrea Sacchetti} \address {Dipartimento di Matematica\\ Universit\`a di Modena e Reggio Emilia\\ Via Campi 213/B, I--41100 Modena, Italy} \date {\today} \email {Sacchetti@unimo.it} %\subjclass {Primary {34E10}; Secondary {33E30}} \thanks {This work is partially supported by the Italian MURST and INDAM-GNFM (project {\it Comportamenti Classici in Sistemi Quantistici}).} \begin {abstract} We consider time-dependent Schr\"odinger equations with a double well potential and an external nonlinear, both local and non-local, perturbation. \ In the semiclassical limit, the finite dimensional eigenspace associated to the lowest eigenvalues of the linear operator is almost invariant for times of the order of the \emph {beating period} and the dominant term of the wavefunction is given by means of the solutions of a finite dimensional dynamical system. \ In the case of local nonlinear perturbation we assume the spatial dimension $d=1$ or $d=2$ \end{abstract} \maketitle \section {Introduction} \label {Sec1} The theoretical analysis of time-dependent nonlinear Schr\"odinger (hereafter NLS) equations % \be \left \{ \begin {array}{l} i \hbar \dot \psi = H_0 \psi + \epsilon W \psi , \ \ \epsilon \in \R ,\ \dot \psi = \frac {\partial \psi }{\partial t}, \\ \psi (x,0) = \psi^0 (x) \in L^2 (\R^d ) , \ \| \psi^0 \| =1 , \label {equa1} \end {array} \right. \ee % where % \be H_0 = - \frac {\hbar^2}{2m} \Delta + V , \ \ \ \Delta = \sum_{j=1}^d \frac {\partial^2}{\partial x_j^2}, \ d \ge 1 , \ \ 2m=1 \label {equa2} \ee % is the linear Hamiltonian and where $W = W (x,|\psi |)$ is a nonlinear perturbation, has attracted an increasing interest in these last few years (see, e.g., \cite {LSY} and \cite {SS}). In this paper we consider the case of symmetric potentials $V$ with \emph {double well} shape. \ In fact, such a potential appears in several fields, from Bose-Einstein condensate states weakly coupled \cite {RSFS} to localization of symmetric molecules \cite {JPT}. \ If the nonlinear term is absent then the linear Hamiltonian $H_0$ has even--parity and odd-parity eigenstates and the state $\psi$ generically performs a \emph {beating motion}, hence the \emph {beating period} plays the role of unit of time. \ When we restore the nonlinear term a symmetry breaking phenomenon occurs: that is, if the nonlinearity is larger than a threshold value then new asymmetric stationary states appears \cite {AFGST}, \cite {GM}, \cite {Zh}. \ Furthermore, for higher nonlinearity, the beating motion is forbidden \cite {GMS}, \cite {RSFS}, \cite {VA}. \ These results can be obtained by reducing the NLS equation to a finite dimensional dynamical system exactly solvable and proving the stability of this approximation for times of the order of the beating period with a rigorous estimate of the error in the semiclassical limit \cite {GMS}, \cite {Sa}. More precisely, in \cite {GMS} has been considered the case on NLS in any dimension $d\ge 1$ where the nonlinear perturbation is \emph {non-local}, that is it is given by % \be W = \langle \psi , g \psi \rangle g(x) \label {equa3} \ee % where $g(x)$ is a given odd function. \ In \cite {Sa} has been considered the case of NLS in dimension $d=1$ and with a nonlinear \emph {local cubic} perturbation given by % \be W = | \psi |^{2\sigma} , \ \ \sigma =1 \label {equa4} \ee \ind Here, we consider NLS equation (\ref {equa1}) in the semiclassical limit in the cases of both \emph {local} (\ref {equa3}) and \emph {non-local} (\ref {equa4}), with any $\sigma >0$, nonlinearity, where we assume that $d=1$ and $d=2$ in the case of \emph {local} nonlinear perturbation. \ Under some generic assumptions on the double-well potential, we give the asymptotic behavior of the solution $\psi$ with a precise estimate of the error. \ In particular, as general results it follows that new asymmetric stationary states appear and the beating motion, between the two wells of a state initially prepared on the two lowest eigenstates, gradually disappears for increasing nonlinearity. \ind Hence, the results previously obtained by \cite {GMS} and \cite {Sa} can be seen as a particular case of the general treatment given here. Our paper is organized as follows. In Section \ref {Sec2} we introduce the assumptions on the potential. \ Moreover, we collect some semiclassical results concerning the spectrum of the linear Schr\"odinger operator. In Section \ref {Sec3} we discuss the beating motion for the unperturbed problem and the choice of the parameters. In Section \ref {Sec4} we give the existence results for equation (\ref {equa1}), the conservation laws and a priori estimate. \ The global existence of the solution is proved for both repulsive and attractive nonlinear perturbation, where, in the second case, we have to assume that the strength of the nonlinear perturbation is small enough. In Section \ref {Sec5} we introduce the two-level approximation and we discuss, in such an approximation, the appearance of new asymmetric stationary states for large nonlinearity strength. \ The two-level approximation, roughly speaking, consists in projecting equation (\ref {equa1}) onto the two-dimensional space spanned by the eigenvectors of the linear Schr\"odinger operator associated to the two lowest eigenvalues. \ For practical purposes, it is more convenient to choose, as a basis of such a two-dimensional space, the two \emph {single-well} states. \ The dynamical system we obtain is, in some cases (for instance for cubic local nonlinearity), exactly solvable. In Section \ref {Sec6} we prove the stability of the two-level approximation in the semiclassical limit. \ We make use of the comparison criterion between ordinary differential equations and of a priori estimate of the solution of the NLS equation. In Appendix we recall some useful inequalities. We close this section by introducing some notations: \begin {itemize} \item [-] Here $\| \cdot \|_p$ denotes the norm of the Banach space $L^p (\R^d )$, $p\in [1,+\infty ]$, $\| \cdot \|$ usually denotes the norm of the space $L^2 (\R^d )$ and sometimes (when this does not cause misunderstanding) it denotes the norm of a bounded operator, too; \item [-] The notation % \bee y = c_{p,d} \eee % means that there exist $\hbar^\star >0$ and a positive constant $C>0$, independent of $\hbar$, such that % \bee |y| \le C \hbar^{-d \frac {p-2}{4p}} , \ \ \forall \hbar \in (0,\hbar^\star ) \eee % \item [-] The notations $y=o(\hbar^\alpha )$, $y=\asy (\hbar^\alpha )$, $\alpha \ge 0$, and $y=\asy (e^{-\Gamma /\hbar} )$ respectively mean that $y \hbar^{-\alpha} \to 0$ as $\hbar \to 0$ and that there exist $\hbar^\star >0$ and a positive constant $C>0$, independent of $\hbar$, such that % \bee |y| \le C \hbar^\alpha \ \ \mbox { and } \ \ |y| \le C e^{-\Gamma /\hbar}, \ \ \forall \hbar \in (0,\hbar^\star ) \eee % \item [-] The notation $y=\tilde \asy (e^{-\Gamma /\hbar} )$ means that for any $\Gamma '$, $0<\Gamma ' <\Gamma$, then $y=\asy (e^{-\Gamma' /\hbar } )$; that is, there exist $\hbar^\star >0$ and a positive constant $C = C_{\Gamma'} >0$, independent of $\hbar$, such that % \bee |y| \le C e^{-\Gamma' /\hbar} , \ \ \forall \hbar \in (0,\hbar^\star ) \eee % \end {itemize} As usual, $\R$ denotes the set of real numbers, $\N$ denotes the set of positive integer numbers and $\N^\star = \N \cup \{ 0 \}$; $C$ denotes any positive constant indipendent of $\hbar$ and $t$. \section {Assumptions and preliminary results} \label {Sec2} \subsection {Linear operator} Here, we introduce the assumptions on the double-well potential $V$ and we collect some well known results on the linear operator $H_0$. \begin {hypothesis} { The potential $V (x)$ is a real valued function such that: \begin {itemize} \item [{i.}] $V$ is a symmetric potential; hence, the Hamiltonian $H_0$ is invariant under some space inversion ${\mathcal S}$: $[{\mathcal S},H_0]=0$; \item [{ii.}] $V \in C^\infty (\R^d )$; \item [{iii.}] $V (x)$ admits two minima at $x=x_\pm $, where $x_- = {\mathcal S} x_+$ and $x_+ \not= x_-$, such that % \be V (x) > V_{min} = V(x_\pm ) , \ \ \forall x \in \R^d , \ x\not= x_\pm ; \label {equa5} \ee % \item [{iv.}] finally we assume that % \be V^-_\infty = {\liminf}_{|x|\to \infty} V(x) >V_{min} \label {equa6} \ee % and % \be V^+_\infty = {\limsup}_{|x|\to \infty} V(x) <+\infty . \label {equa7} \ee % \end {itemize} } \end {hypothesis} {\it Remarks} \begin {itemize} \item [-] For the sake of definiteness we can always assume that, by means of a suitable choice of the coordinates, $V$ is symmetric with respect to the spatial coordinate $x_1$, that is % \be V(-x_1,x_2,\ldots , x_d)=V(x_1,x_2,\ldots ,x_d), \ \ \forall (x_1,\ldots , x_d) \in \R^d ; \label {equa8} \ee \item [-] For the sake of simplicity, we assume also that % \bee \nabla V (x_\pm ) =0 \ \ \mbox { and } \ \ \mbox { Hess } V (x_\pm ) >0 ; \eee % The case of degenerate minima, that is Hess $V(x_\pm )=0$, could be treated in a similar way; however, we don't dwell here on such details; \item [-] In fact, we could replace assumption {ii.} with the weaker assumption $V\in C^2$; \item [-] Assumption (\ref {equa7}) is introduced in order to obtain the local existence of the solution of the Cauchy problem (\ref {equa1}) by means of well known results \cite {CW} (see also \cite {Gi}). \ However, by making use of some ideas contained in the paper by Carles \cite {C1}, condition (\ref {equa7}) could be partially removed and anharmonic potentials like, e.g. in dimension $d=1$, $V(x) = (x^2-a^2)(x^2-b^2)$ could be considered. \end {itemize} The operator $H_0$ formally defined by (\ref {equa2}) admits a self-adjoint realization (still denoted by $H_0$) on $L^2 (\R )$ (Theorem III.1.1 in \cite {BS}). Let $\sigma (H_0 ) = \sigma_d \cup \sigma_{ess}$ be the spectrum of the self-adjoint operator $H_0$, where $\sigma_d$ denotes the discrete spectrum and $\sigma_{ess}$ denotes the essential spectrum. \ It follows that (see Theorem III.3.1 in \cite {BS}) % \bee \sigma_d \subset (V_{min},V_\infty^- ) \ \ \mbox { and } \ \ \sigma_{ess}=[V_\infty^- ,+\infty ) \eee % Furthermore, the following two Lemmas hold: \begin {lemma} \label {Lem1} Let $\sigma_d$ be the discrete spectrum of $H_0$. \ Then, for any $\hbar \in (0,\hbar^\star )$, for some $\hbar^\star >0$ fixed, it follows that: \begin {itemize} \item [{i.}] $\sigma_d$ is not empty and, in particular, it contains two eigenvalues at least; \item [{ii.}] let $\lambda_{\pm}$ be the lowest two eigenvalues of $H_0$, they are non-degenerate, in particular $\lambda_+ < \lambda_-$ and there exists $C>0$, independent of $\hbar$, such that % \be \inf_{\lambda \in \sigma (H_0) - \{ \lambda_{\pm} \} } [\lambda - \lambda_{\pm} ] \ge C \hbar . \label {equa9} \ee % \end {itemize} \end {lemma} \begin {proof} The proof is an immediate consequence of the above assumptions and standard WKB arguments. \ In fact, by assuming, for the sake of simplicity, that % \be \mbox {Hess} V (x_\pm) = 2 \mbox {diag} (\mu_1 , \mu_2 , \ldots , \mu_d ), \ \ \mu_j >0 , \ j=1,\ldots , d, \label {equa9bis} \ee % then the first two eigenvalues of $H_0$ are given by the semi-classical \emph {single-well} non-degenerate eigenvalues (Theorem 2.3.1 \cite {He}) % \be \lambda_{\pm} = V_{min} + \left [\sum_{j=1}^d \sqrt {\mu_j} \right ] \hbar + o (\hbar ) , \ \ \mbox { as } \ \hbar \to 0 ,\label {equa10} \ee % Furthermore, from conditions (\ref {equa5}) and (\ref {equa6}) the estimate (\ref {equa9}) follows (Corollary 2.3.5 \cite {He}) \end {proof} \begin {lemma} \label {Lem2} Let $\varphi_{\pm}$ be the normalized eigenvectors associated to $\lambda_{\pm}$, then: \begin {itemize} \item [{i.}] $\varphi_{\pm}$ can be chosen to be real-valued functions such that % \be \varphi_{\pm} (-x_1, x_2 , \ldots , x_d) = \pm \varphi_{\pm} (x_1, x_2 , \ldots , x_d ); \label {equa11} \ee \item [{ii.}] $\varphi_{\pm } \in H^1 (\R^d )$; \item [{iii.}] $\varphi_\pm \in L^p (\R^d )$ for any $p \in [2,+\infty ]$; \item [{iv.}] there exists a positive constant $C$, independent on $\hbar$, such that % \be \| \varphi_{\pm } \|_p = c_{p,d} \le C \hbar^{- d\frac {p-2}{4 p}} , \ \ \forall p \in [2,+\infty ], \ \ \forall \hbar \in (0, \hbar^\star ). \label {equa12} \ee % \end {itemize} \end {lemma} \begin {proof} Property {i.} immediately follows from (\ref {equa8}). \ Property {ii.} follows from Lemma III.3.1 in \cite {BS}. \ In order to prove the statement {iii.} we recall that the eigenvectors $\varphi_\pm$ satisfy to the following global estimate: for any $\delta >0$ fixed there exists a positive constant $C_{\delta , \hbar} >0$ such that (Theorem III.3.2 and Corollary III.3.1 in \cite {BS}) % \bee |\varphi_\pm (x) | \le C_{\delta , \hbar} \exp \left [- \delta |x|/\hbar \right ] \eee % Hence $\varphi_\pm \in L^\infty (\R^d )$. \ From this fact and since $\varphi_\pm \in L^2 (\R^d )$ then statement {iii.} immediately follows. \ Finally, in order to prove the statement {iv.} let % \bee \varphi_{\pm} = \case {1}{\sqrt 2} \left [ \varphi_R \pm \varphi_L \right ] \eee % where the vectors $\varphi_{R,L}$ satisfies to the following WKB estimates \cite {He} % \be \varphi_R (x)= \varphi_L (-x) \sim [2\pi \hbar]^{-d/4} e^{-[\sum_{j=1}^d (x_j-x_{+,j})^2 \sqrt {\mu_j} ] /2\hbar } ,\ \ \mbox { as } \ \hbar \to 0 , \label {equa13} \ee % in a neighborhood of the mimina $x_\pm =(x_{\pm ,1} , \ldots , x_{\pm ,d} )$ and where $\mu_j$ are defined in (\ref {equa9bis}). \ Hence, property {iv.} follows for $p=+\infty $. \ From this fact, from the normalization of the eigenvectors and from the H\"older inequality then property {iv.} follows for any $p\in [2,+\infty ]$: % \bee \| \varphi_\pm \|_p = \left [ \| \varphi_\pm^2 \varphi_\pm^{p-2} \|_1 \right ]^{1/p} \le \| \varphi_\pm \|_2^{2/p} \| \varphi_\pm \|_\infty^{(p-2)/p}=\| \varphi_\pm \|_\infty^{(p-2)/p}. \eee % \end {proof} From (\ref {equa10}) it follows that the \emph {splitting} between the two lowest eigenvalues % \be \omega = \case 12 (\lambda_{-}-\lambda_{+}) \label {equa14} \ee % vanishes as $\hbar $ goes to zero. \ In order to give a precise estimate of the splitting $\omega$ we make use of the fact that $V$ is a symmetric double-well potential with non-zero barrier between the wells. \ That is, let % \be \Gamma = \inf_\gamma \int_{\gamma} \sqrt {V (x)-V_{min}} d x >0, \label {equa15} \ee % be the Agmon distance between the two wells; where $\gamma$ is any path connecting the two wells, that is $\gamma \in AC ([0,1],\R^d)$ such that $\gamma (0)=x_-$ and $\gamma (1)=x_+$. \ From standard WKB arguments (see \cite {He}) then it follows that the splitting is \emph {exponentially small}, that is % \be \omega = \tilde \asy ( e^{- \Gamma /\hbar } ). \label {equa16} \ee % Furthermore, the vectors, usually called \emph {single well states}, % \bee \varphi_R = \case {1}{\sqrt 2} \left [ \varphi_+ + \varphi_- \right ] \ \ \mbox { and } \ \ \varphi_L = \case {1}{\sqrt 2} \left [ \varphi_+ - \varphi_- \right ] \eee % are \emph {localized on one well}, and % \be \| \varphi_{R } \varphi_{L} \|_\infty = \tilde \asy ( e^{-\Gamma /\hbar }) \label {equa17} \ee % More precisely, from (\ref {equa11}) it follows that % \bee \varphi_{R } (-x_1,x_2,\ldots ,x_d)=\varphi_{L}(x_1,x_2 , \ldots , x_d) \eee % and these functions are localized on only one of the two wells in the sense that for any $r>0$ there exists $C>0$ such that % \bee \int_{D_r(x_+)} |\varphi_{R} (x)|^2 dx = 1 + \asy (e^{-C /\hbar }) \eee % and % \bee \int_{D_r(x_-)} |\varphi_{L} (x)|^2 dx = 1 + \asy (e^{-C /\hbar }) \eee % where $D_r(x_\pm )$ is the ball with center $x_\pm$ and radius $r$. \ For such a reason we call them \emph {single-well} (normalized) states. {\it Remark:} \begin {itemize} \item [-] We underline that, by assuming some further regularity properties on the potential $V$, it is possible to obtain the precise asymptotic behavior of the splitting as $\hbar $ goes to zero \cite {HS}. \end {itemize} \subsection {Nonlinear perturbation} Here we admit both \emph {local} and \emph {non-local} nonlinear perturbations \begin {hypothesis} Let $g(x)$ be a given real-valued bounded and continuous function. \ We assume that \begin {itemize} \item [i.] {\bf Nonlinear local perturbation.} \ The perturbation $W$ has the form % \be W = W_\ell (x, |\psi |) = g(x) |\psi (x)|^{2\sigma}, \ \ \sigma >0 \label {equa18} \ee % \item [ii.] {\bf Nonlinear non-local perturbation.} The perturbation $W$ has the form % \be W = W_{n\ell} (x, |\psi |) = g(x) \langle \psi ,g \psi \rangle \label {equa19} \ee % \end {itemize} In the local perturbation case (\ref {equa18}) we assume the dimension $d=1$ or $d=2$. \ In the non-local perturbation case (\ref {equa19}) we don't introduce any assumption on the dimension $d$. \end {hypothesis} {\it Remarks:} \begin {itemize} \item [-] If $\sigma =1$ and $g\equiv 1$ then the NLS equation (\ref {equa1}) with \emph {local perturbation} (\ref {equa18}) coincides with the one previously studied by Sacchetti \cite {Sa} in dimension d=1. \ If $g(x)$ is an even function then the NLS equation (\ref {equa1}) with \emph {non-local perturbation} (\ref {equa19}) coincides with the one previously studied by Grecchi, Martinez and Sacchetti \cite {GMS}. \item [-] In the case of nonlinear local perturbation (\ref {equa18}) we have to assume that the dimension $d$ is not higher than 2. \ In fact, in the case of dimension $d>2$ then, provided that $\sigma < \frac {2}{d-2}$, the existence results and the conservation laws (see Section \ref {Sec4}) still hold, but the stability result fails (see Section \ref {Sec6}). \end {itemize} \subsection {Assumption on the initial state} Let % \bee \Pi_c = \I - \left [ \langle \varphi_+ , \cdot \rangle \varphi_+ + \langle \varphi_- , \cdot \rangle \varphi_- \right ] \eee % be the projection operator onto the eigenspace orthogonal to the bi-dimensional space associated to the doublet $\{ \lambda_\pm \}$. \ Let $\psi^0$ be the initial wavefunction, we assume that \begin {hypothesis} $\Pi_c \psi^0 =0$. \end {hypothesis} That is, we assume that % \bee \psi^0 = c_+ \varphi_+ + c_- \varphi_- = c_R \varphi_R + c_L \varphi_L \eee % for some $c_\pm$ and $c_{R,L}$. \medskip {\it Remarks:} \begin {itemize} \item [-] In fact, we could assume that the initial state $\psi^0$ belongs to a finite dimensional eigenspace of $H_0$. \ More precisely, let $\sigma_1, \sigma_2 \subset \sigma (H_0)$ such that $\sigma (H_0) = \sigma_1 \cup \sigma_2$, $\sigma_1 \cap \sigma_2 = \emptyset$, $\sigma_1 \subset \sigma_{pp} (H_0)$, where $\sigma_{pp}$ denotes the pure point spectrum of $H_0$, and $\sigma_1$ has a finite number of elements. \ Let ${\mathcal H}_1$ be the finite-dimensional spectral eigenspace associated to $\sigma_1$. \ Then we can replace the previous assumption by assuming that $\psi^0 \in {\mathcal H}_1$ and % \bee d (\sigma_1 , \sigma_2 ) = \inf_{\lambda \in \sigma_1 , \ \mu \in \sigma_2 } |\lambda - \mu | \ge C \hbar . \eee % In such a case we have to define $\omega = \case 12 \inf_{\lambda , \mu \in \sigma_1 , \ \lambda \not= \mu} |\lambda - \mu |$. \end {itemize} \section {Beating motion and choice of parameters} \label {Sec3} Let us consider, for a moment, the time-dependent linear Schr\"odinger equation % \be \left \{ \begin {array}{l} i \hbar \dot \psi = H_0 \psi , \ \ \epsilon \in \R ,\ \dot \psi = \frac {\partial \psi }{\partial t}, \\ \psi (x,0) = \psi^0 (x) \in L^2 (\R^d ) , \ \Pi_c \psi^0 =0 , \end {array} \right. \label {equa20} \ee % This equation has an explicit solution given by % \bee \psi (x,t) &=& e^{-i \lambda_+ t /\hbar} c_+ \varphi_+ + e^{-i \lambda_- t /\hbar} c_- \varphi_- \\ &=& \case {1}{\sqrt 2} \left [ c_+ e^{-i \lambda_+ t /\hbar} + c_- e^{-i \lambda_- t /\hbar} \right ] \varphi_R + \case {1}{\sqrt 2} \left [ c_+ e^{-i \lambda_+ t /\hbar} - c_- e^{-i \lambda_- t /\hbar} \right ] \varphi_L \\ &=& \case {1}{\sqrt {2}} e^{-i \Omega t /\hbar} \left [ \left ( \Delta \varphi_R + \delta \varphi_L \right ) \cos (\omega t/\hbar ) + i \left ( \delta \varphi_R + \Delta \varphi_L \right )\sin (\omega t / \hbar ) \right ] \eee % where we set % \bee \lambda_\pm = \Omega \mp \omega , \ \Delta = c_+ + c_- , \ \delta = c_+ - c_- \eee % That is $\psi (x,t)$ performs a \emph {beating motion} with \emph {beating period} % \bee T= \frac {2\pi \hbar}{\omega} \eee % Such a period will play the role of unit of time. \begin {hypothesis} Let $\omega$ be the splitting (\ref {equa14}) satisfying to the asymptotic estimate (\ref {equa16}). \ We assume that the real-valued parameter $\epsilon$ depends on $\hbar$ in such a way % \be \frac {|\epsilon |\hbar^{-d\sigma /2}}{\omega} \le C, \ \ \forall \hbar \in (0 , \hbar^\star ) \label {equa21} \ee % for some positive constant $C$, independent of $\hbar$, and for some $\hbar^\star$, where $\sigma$ is defined in (\ref {equa18}) for nonlinear local perturbations and where $\sigma =0$ for nonlinear non-local perturbations (\ref {equa19}). \end {hypothesis} {\it Remarks:} \begin {itemize} \item [-] We underline that the strength of the perturbation is, roughly speaking, given by $|\epsilon |$ is the case of nonlinear non-local perturbation (\ref {equa19}), and by $|\epsilon | \hbar^{-d\sigma /2}$ in the case of nonlinear local perturbation (\ref {equa18}). \ The ratio % \bee \eta = \frac {\epsilon \hbar^{-d\sigma /2}}{\omega} \eee % plays the role of effective nonlinearity parameter. \ The above assumption implies that $|\eta | \le C$. \item [-] Condition (\ref {equa21}) implies that % \bee && \left ( \mbox {Beating period} \right ) \times \left ( \mbox {Perturbation strength} \right ) = \\ && \ \ \ = T \times |\epsilon |\hbar^{-d \sigma /2} \approx \frac {\hbar}{\omega} \times \omega = \hbar \approx \mbox {dist} (\sigma (H_0) , \lambda_\pm ) \eee % Thus, intuitive arguments do not suggest us that the subspace $(\I - \Pi_c )L^2$ is almost invariant for times of the order of the beating period. \ In fact, we will prove that $\| \Pi_c \psi \| = \tilde \asy ( e^{-\Gamma /\hbar} )$ for any $t\in [0,T]$ for $\hbar $ small enough. \end {itemize} \section {Existence results and conservation laws} \label {Sec4} Here, making use of some results by \cite {CW}, we prove that the solution of equation (\ref {equa1}) globally exists. \ To this end we recall that $\psi^0 \in H^1 \cap L^p$ for any $p\in [2,+\infty ]$ (see Lemma \ref {Lem2}). \ The assumptions on the strength of the nonlinear perturbation, that is $\epsilon = \tilde \asy ( e^{-\Gamma /\hbar } )$ (see equations (\ref {equa16}) and (\ref {equa21})), could, in order to prove the global existence result and the conservation laws, be relaxed; in fact, here we simply require that $\epsilon = \asy (\hbar^{\alpha })$ for some $\alpha >2$. \subsection {Local existence} \begin {theorem} \label {Thm1} There exists $T^\star >0$ and an unique solution $\psi \in C([0,T^\star ),H^1) \cap C^1 ([0,T^\star), H^{-1})$ of (\ref {equa1}), where $T^\star =+\infty $ or $\| \nabla \psi \| \to +\infty$ as $t\to T^\star$. \end {theorem} \begin {proof} This result is a consequence of Theorem 2.1 and Examples 1--3 by \cite {CW}. \ In fact, $V \in L^\infty$ and $\psi^0 \in H^1$; furthermore we show that both $W_\ell$ and $W_{n\ell}$ satisfy the conditions of \cite {CW}. \ To this end let, in the case of local perturbation, % \bee f(x,z) = g(x) |z|^{2\sigma} z . \eee % If we prove that % \be \left | f(x,z_1 ) - f(x,z_2 )\right | \le M \left [ 1 + |z_1|^\gamma + |z_2|^\gamma \right ] |z_1 - z_2 | \label {equa22} \ee % for some positive constant $M$ and for some $\gamma \in [0,+\infty )$, then the local existence of the solution in the case of local nonlinear perturbation follows. \ In order to prove this inequality we assume, for the sake of definiteness, that $0<|z_1|<|z_2|$ (if $|z_1|=0$ or $|z_1|=|z_2|$ then (\ref {equa22}) immediately follows) and we obtain that % \bee \left | f(x,z_1 ) - f(x,z_2 )\right | & = & |g(x)| \cdot \left | |z_1|^{2\sigma} z_1 - |z_2|^{2\sigma } z_2 \right | \\ & \le & \tilde g \left [ |z_2|^{\gamma } |z_1 - z_2 | + |z_1 | \left | |z_1|^{\gamma} - |z_2|^{\gamma } \right | \right ] \eee % where $\gamma = 2 \sigma$ and $\tilde g = \max_x |g(x)|$. \ If $\gamma \in (0,1]$ then, recalling the inequality (\ref {equa65}), it follows that % \bee \left | f(x,z_1 ) - f(x,z_2 )\right | &\le & \tilde g \left [ |z_2|^{\gamma } |z_1-z_2| + |z_1|^\gamma |z_1|^{1-\gamma} \left | |z_1|^{\gamma} - |z_2|^{\gamma } \right | \right ] \\ &\le & \tilde g |z_1 -z_2| \left ( |z_2|^{\gamma} + |z_1|^{\gamma}\frac {1}{\gamma} \right ) \eee % from which (\ref {equa22}) follows. \ In the case % \bee 2 \sigma = \gamma >1 \eee % then the inequality (\ref {equa65}) gives that % \bee \left | f(x,z_1 ) - f(x,z_2 )\right | %&\le & \tilde g \left [ |z_2|^\gamma |z_1 - z_2 | + \gamma |z_1| \cdot |z_2|^\gamma \left ( %\frac {|z_1|}{|z_2|} \right )^{\gamma 2^{-n}} |z_1|^{-1} (|z_2|-|z_1|) \right ] \\ &\le & \tilde g \left [ |z_2|^\gamma |z_1-z_2| + \sigma |z_2|^\gamma \left | |z_2| - |z_1| \right | \right ] \\ &\le & M \left [|z_2|^\gamma (1+ \gamma )\right ] |z_2 -z_1| \eee % since % \bee |z_1 |< |z_2| \ \ \mbox { and } \ \ \left | |z_1| - |z_2| \right | \le |z_1-z_2| \eee % obtaining (\ref {equa22}). \ In the non-local case, where % \bee W = W_{n\ell} (x, \psi ) = g(x) \langle \psi , g \psi \rangle \eee % the local existence result follows by means the same arguments. \ We simply have to check that % \bee \| W_{n\ell} (x,u) u - W_{n\ell} (x,v) v \| \le C \| u-v \| \eee % Indeed, % \bee \| W_{n\ell} (x,u) u - W_{n\ell} (x,v) v \| &=& \tilde g \| \langle u , gu \rangle u - \langle v, g v \rangle v \| \\ &=& \tilde g \left \| \left ( \langle u , gu \rangle - \langle v , g v \rangle \right ) u + \langle v , g v \rangle (u-v) \right \| \\ &\le & \tilde g \| u \| \left | \langle u , gu \rangle - \langle v , g v \rangle \right | + \tilde g^2 \| v \|^2 \| u-v \| \\ &\le & \tilde g \| u \| \left | \langle u-v , gu \rangle + \langle v , g (u-v) \rangle \right | + \tilde g^2 \| v \|^2 \| u-v \| \\ &\le & \tilde g^2 \| u \| \left [ \| u \| \cdot \| u-v\| + \| v \| \cdot \| u-v \| \right ] + \tilde g^2 \| v \|^2 \| u-v \| \\ &\le & C \| u - v \| \eee % where % \bee C = \tilde g^2 \left [ \| u \|^2 + \| v \|^2 + \| u \| \cdot \| v \| \right ] \eee % \end {proof} \subsection {Conservation laws} By means of a direct computation the following first integral exists. \subsubsection {Conservation of the norm} Let % \bee {\mathcal N} (\psi ) = \| \psi \|^2 \eee % then % \bee {\mathcal N } [\psi (x,t )] = {\mathcal N} [\psi^0(x)] = 1 \eee \subsubsection {Conservation of the energy} Let us consider the case of local nonlinear perturbation (\ref {equa18}). \ Let % \be {\mathcal H} (\psi ) = {\mathcal H}_\ell (\psi ) = \langle \psi , H_0 \psi \rangle + \frac {\epsilon}{\sigma+1} \langle \psi^{\sigma+1} , g \psi^{\sigma +1} \rangle \label {equa23} \ee % defined on $H^1 (\R^d ) \cap L^{2(\sigma +1)} (\R^d )$. \ Then a direct computation gives that % \bee {\mathcal H}_\ell [\psi (x,t)] = {\mathcal H}_\ell [\psi^0 (x) ] \eee % Similarly, in the case of non-local nonlinear perturbation (\ref {equa19}) then it follows that % \bee {\mathcal H}_{n\ell} [\psi (x,t)] = {\mathcal H}_{n\ell} [\psi^0 (x) ] \eee % where the energy is defined as % \be {\mathcal H} (\psi ) = {\mathcal H}_{n\ell} (\psi ) = \langle \psi , H_0 \psi \rangle + \frac 12 {\epsilon} \langle \psi , g \psi \rangle^2 \label {equa24} \ee % on $H^1 (\R^d ) \cap L^{2 } (\R^d )$. \subsection {A priori estimates} \begin {theorem} \label {Thm2} Let % \bee \Lambda = \frac {{\mathcal H} [\psi^0 ] - V_{min}}{\hbar^2} \eee % where ${\mathcal H}$ is the energy defined above in equations (\ref {equa23}) and (\ref {equa24}) for, respectively, local and non-local perturbations. \ The solution $\psi (x,t)$ of equation (\ref {equa1}) satisfies to the following a priori estimates % \be \| \nabla \psi \| \le C \sqrt {\Lambda} \label {equa25} \ee % and % \be \| \psi \|_p \le C \Lambda^{d \frac {p-2}{4p}} \label {equa26} \ee % where % \be p \in [2,+\infty] \ \mbox { if }\ d=1, \ \ \mbox { and } \ \ p \in [2,+\infty ) \ \mbox { if }\ d=2 \label {equa27} \ee % \end {theorem} \begin {proof} We consider, at first, the case of local perturbation (\ref {equa18}). \ The conservation of the energy ${\mathcal H}_\ell (\psi )$ gives that % \bee \hbar^2 \| \nabla \psi \|^2 &=& {\mathcal H}_\ell (\psi^0 ) - \frac {\epsilon}{\sigma+1} \langle g \psi^{\sigma +1} , \psi^{\sigma +1} \rangle - \langle V \psi , \psi \rangle \\ &\le & {\mathcal H}_\ell (\psi^0 ) - V_{min} {\mathcal N} (\psi^0) + \frac {\tilde g |\epsilon |}{\sigma +1} \| \psi \|_{2(\sigma +1)}^{2(\sigma +1)} \eee % where % \bee V_{min} =\min_x V(x) =V(x_\pm ) >-\infty \ \ \mbox { and } \ \ \tilde g = \max_x |g(x)| . \eee % Hence % \bee \| \nabla \psi \|^2 &\le & \Lambda + \rho^2 \| \psi \|_{2(\sigma +1)}^{2(\sigma +1)} \eee % where % \bee \rho^2 = \frac {\tilde g |\epsilon |}{(\sigma +1) \hbar^2} \le C |\epsilon | \hbar^{-2} \ll 1 \ \ \mbox { and } \ \ \hbar |\Lambda | = C + o(1) \eee % since $\epsilon$ is a small semiclassical estimate satisfying (\ref {equa21}). \ Now, we make use of the Gagliardo-Nirenberg inequality (\ref {equa66}) obtaining % \be \| \nabla \psi \|^2 \le \Lambda + C \rho^2 \| \nabla \psi \|^{\sigma d} \| \psi \|^{2 + \sigma (2-d)} \le \Lambda + C \rho^2 \| \nabla \psi \|^{\sigma d} \label {equa28} \ee % from which and from the fixed point theorem the estimate (\ref {equa25}) follows. \ Indeed, if $\sigma d \le 2$ then the result immediately follows. \ If $\sigma d >2$ we recall that $\varphi_R$ and $\varphi_L$ satisfy to the asymptotic behavior (\ref {equa13}) then $\| \nabla \psi^0 \| \le C \hbar^{-1/2}$. \ If we set % \bee y=\hbar^{1/2} \| \nabla \psi \| , \ \alpha = \hbar \Lambda , \ \beta = C \rho^2 \hbar^{1-\sigma d /2} \eee % where, initially $y \le C$, and where % \bee C^{-1} \le \alpha \le C \ \ \mbox { and } \ \ \beta = o (1) \ \ \mbox { as } \ \hbar \to 0, \eee % then (\ref {equa28}) can be written as % \bee y^2 \le \alpha + \beta y^{\sigma d} \eee % Then fixed point arguments proved that $y \le C $ for any time, from which (\ref {equa25}) follows. \ From this fact, and making use of the Gagliardo-Nirenberg inequality again, we obtain that % \be \| \psi \|_p \le C \| \nabla \psi \|^{\delta} \| \psi \|^{1-\delta} \le C \Lambda^{\frac 12 \delta }, \ \ \delta = \frac {d(p-2)}{2p} \label {equa29} \ee % Now, we consider the case of non-local perturbation (\ref {equa19}). \ In such a case we have that the conservation of the energy ${\mathcal H}_{n\ell}$ gives inequality (\ref {equa25}) immediately. \ Indeed, % \bee \| \nabla \psi \|^2 & = & \frac {1}{k^2} \left [ {\mathcal H}_{n\ell} (\psi^0) - \langle V \psi , \psi \rangle -\frac 12 {\epsilon} \langle \psi , g \psi \rangle^2 \right ] \\ &\le & \Lambda + \rho^2 \left [ \| \psi \|^{2}_{2} \right ]^2 , \ \ \ \rho^2 = \frac 12 {\tilde g |\epsilon |} \ll 1 \\ &\le & \Lambda + C \rho^2 \| \nabla \psi \| \eee % As above, the estimates $\| \nabla \psi \| \le C \sqrt \Lambda $ and (\ref {equa26}) follow. \end {proof} {\it Remarks:} \begin {itemize} \item [-] Since $\psi^0$ is prepared on the first two states and since (\ref {equa21}) then it follows that $\Lambda \sim \hbar^{-1}$. \ Hence the above estimates take the form % \be \| \nabla \psi \| \le C \hbar^{-1/2} \ \ \mbox { and } \ \ \| \psi \|_p = c_{p,d } \le C \hbar^{-d \frac {p-2}{4p}} \label {equa30} \ee % for any $p$ satisfying (\ref {equa27}); \item [-] In fact the above a priori estimates hold for any $\epsilon >0$ (repulsive nonlinearity), and for any $\epsilon <0$ (attractive nonlinearity) such that $\epsilon = \asy (\hbar^\alpha )$ for any $\alpha >2$; \end {itemize} \subsection {Global existence} \begin {theorem} \label {Thm3} The solution $\psi$ of (\ref {equa1}) globally exists; that is $T^\star =+\infty$. \end {theorem} \begin {proof} The global solution immediately follows from Theorem \ref {Thm1} and from the estimate (\ref {equa24}). \end {proof} \section {Two-level approximation} \label {Sec5} \subsection {Two-level approximation} Since the beating period $T= \frac {2\pi \hbar}{\omega}$ plays the role of the unit of time it is more convenient to consider the \emph {slow time} % \bee \tau = \frac {\omega t}{\hbar } \eee % Therefore, if we consider the change of variable (with abuse of notation) % \bee \psi (x,t ) \to \psi (x,\tau )= e^{i \Omega t /\hbar} \psi (x,t ) ,\ \ \Omega = \frac 12 \left [ \lambda_+ + \lambda_- \right ] \eee % then equation (\ref {equa1}) takes the form (here $'$ denotes the derivative with respect to $\tau $) % \be \left \{ \begin {array}{l} i \psi' = \frac {1}{\omega} \left [ H_0-\Omega \right ] \psi + \frac {\epsilon}{\omega} W \psi \\ \psi (x,0) = \psi^0 (x) \in L^2 (\R^d ) , \ \Pi_c \psi^0 =0 , \end {array} \right. \label {equa31} \ee % since $W=W(x,|\psi |)$. \ In order to study this equation for $\tau \in [0,\tau' ]$, for any fixed $\tau' >0$, in the semiclassical limit we rewrite $\psi$ in the following form % \be \psi (x,\tau ) = \varphi (x,\tau ) + \psi_c (x,\tau ), \ \ \varphi (x,\tau ) = a_R (\tau ) \varphi_R (x)+ a_L (\tau ) \varphi_L (x) \label {equa32} \ee % where % \bee \psi_c (x,\tau ) = \Pi_c \psi (x,\tau ) \eee % and where % \bee a_R (\tau ) = \langle \varphi_R , \psi \rangle \ \ \mbox { and } \ \ a_L (\tau )= \langle \varphi_L , \psi \rangle \eee % are unknown complex-valued functions. \ Since % \bee H_0 \psi &=& a_R H_0 \varphi_R + a_L H_0 \varphi_L + H_0 \psi_c \\ &=& a_R \left [ \Omega \varphi_R - \omega \varphi_L \right ] + a_L \left [ -\omega \varphi_R + \Omega \varphi_L \right ] + H_0 \psi_c \eee % then, by substituting (\ref {equa32}) in (\ref {equa31}) and projecting the resulting equation onto the one-dimensional spaces spanned by the \emph {single-well} states $\varphi_R$ and $\varphi_L$, and on the space $\Pi_c L^2 (\R^d)$ it follows that it takes the form % \be \left \{ \begin {array}{ll} i a_R' = - a_L + r_R & r_R = r_R (a_R , a_L , \psi_c ) = \frac {\epsilon}{\omega} \langle \varphi_R , W \psi \rangle \\ i a_L' = - a_R + r_L & r_L = r_L (a_R , a_L , \psi_c ) = \frac {\epsilon}{\omega} \langle \varphi_L , W \psi \rangle \\ i \psi_c' = \frac {1}{\omega} \left [ H_0-\Omega \right ] \psi_c + r_c & r_c = r_c (a_R , a_L , \psi_c ) = \frac {\epsilon}{\omega} \Pi_c W \psi \end {array} \right. \label {equa33} \ee % with initial conditions % \bee a_{R,L} (0) = \langle \varphi_{R,L} , \psi^0 \rangle , \ \ \psi_c^0 =\psi_c (x,0) = \Pi_c \psi^0 =0 \eee % \begin {lemma}\label {Lem3} Let % \bee r_{R,L} = r_{R,L} (a_R , a_L , \psi_c ) = \frac {\epsilon}{\omega} \langle \varphi_{R,L} , W \psi \rangle \eee % Then, it follows that % \be r_{R,L} (a_R, a_L, 0)= \eta \tilde r_{R,L} (a_R, a_L) + \tilde \asy (e^{-\Gamma /\hbar } ) \label {equa34} \ee % where: % \begin {itemize} \item [i.] {\bf Local perturbation:} % \bee \tilde r_R = C_R |a_R|^{2\sigma} a_R, \ \ \tilde r_L = C_L |a_L|^{2\sigma} a_L \ \ \mbox { and } \ \ \eta= \frac {\epsilon}{\omega}\hbar^{-d\sigma /2} \eee % where % \bee C_R = \hbar^{d\sigma /2} \langle \varphi_R , g |\varphi_R |^{2\sigma } \varphi_R \rangle \ \ \mbox { and } \ \ C_L = \hbar^{d\sigma /2} \langle \varphi_L , g |\varphi_L |^{2\sigma } \varphi_L \rangle \eee % are such that $C_{R,L} = \asy (1)$ as $\hbar $ goes to zero. \item [ii.] {\bf Non-Local perturbation:} % \bee \tilde r_R = C_R a_R |a_R|^2 + \tilde C a_R |a_L|^2 , \ \ \tilde r_L = C_L a_L |a_L|^2 + \tilde C a_L |a_R|^2 \eee % and % \bee \eta =\frac {\epsilon}{\omega } \eee % where % \bee C_R = \langle \varphi_R , g \varphi_R \rangle^2 , \ \ C_L = \langle \varphi_L , g \varphi_L \rangle^2 \ \ \mbox { and } \ \ \tilde C = \sqrt {C_R C_L } \eee % are such that $C_{R,L} = \asy (1)$ as $\hbar $ goes to zero. \end {itemize} \end {lemma} \begin {proof} In order to give the explicit expression of the terms $r_{R,L}$ we consider the local and non-local perturbations separately. {\it Local perturbation.} \ In such a case we have that (where we set $\psi_c =0$ inside $W_\ell$) % \bee r_R (a_R , a_L ,0) %&=&\frac {\epsilon}{\omega}\langle \varphi_R,W_\ell (a_R \varphi_R + a_L \varphi_L ) \rangle \\ &=& \frac {\epsilon}{\omega} \langle \varphi_R , g |a_R \varphi_R + a_L \varphi_L |^{2\sigma} (a_R \varphi_R + a_L \varphi_L) \rangle \\ &=& \frac {\epsilon}{\omega} \left [ |a_R|^{2\sigma} a_R \langle \varphi_R , g |\varphi_R |^{2\sigma} \varphi_R \rangle + \tilde \asy (e^{-\Gamma /\hbar } ) \right ] \eee % since (\ref {equa17}). \ Similarly, we obtain that % \bee r_L (a_R,a_L,0) = \frac {\epsilon}{\omega} \left [ |a_L|^{2\sigma} a_L \langle \varphi_L , g |\varphi_L |^{2\sigma} \varphi_L \rangle + \tilde \asy (e^{-\Gamma /\hbar } ) \right ] \eee % If we set % \bee C_R = \hbar^{d\sigma /2} \langle \varphi_R , g |\varphi_R |^{2\sigma}\varphi_R \rangle \ \ \mbox { and } \ \ C_L = \hbar^{d\sigma /2} \langle \varphi_L , g |\varphi_L |^{2\sigma}\varphi_L \rangle \eee % then % \bee r_{R,L} = \eta C_{R,L} |a_{R,L} |^{2\sigma } a_{R,L} + \tilde \asy (e^{-\Gamma /\hbar }) \eee % for any $\tau $. \ Furthermore, from Lemma \ref {Lem2} it follows that % \bee |C_{R,L} | &\le & \tilde g \hbar^{d\sigma /2} \| \varphi_{R,L}^{\sigma +1} \|^2 = \tilde g \hbar^{d\sigma /2} \| \varphi_{R,L} \|^{2(\sigma +1)}_{2(\sigma +1)} = \tilde g \hbar^{d\sigma /2} c^{2(\sigma +1)}_{2(\sigma +1),d} \\ &\le & C \hbar^{d\sigma /2} \left [ \hbar^{-d \frac {2\sigma }{8(\sigma +1)}} \right ]^{2(\sigma +1)} \le C , \ \ \forall \hbar \in (0,\hbar^\star ) \eee % where $\tilde g = \max_x |g(x)| <+\infty$. {\it Non-Local perturbation.} \ In such a case we have that (where we set $\psi_c =0$ inside $W_{n\ell}$) % \bee && r_R (a_R,a_L,0) = %\frac {\epsilon}{\omega} \langle \varphi_R , W_{n\ell} (a_R \varphi_R + a_L \varphi_\ell ) %\rangle \\ && \ \ = \frac {\epsilon}{\omega} \langle \varphi_R , (a_R \varphi_R + a_L \varphi_L ) g \rangle \cdot \langle (a_R \varphi_R + a_L \varphi_\ell ) , g (a_R \varphi_R + a_L \varphi_\ell ) \rangle \\ && \ \ = \frac {\epsilon}{\omega} \left [ a_R \langle \varphi_R , g \varphi_R \rangle + \tilde \asy (e^{-\Gamma /\hbar }) \right ] \times \\ && \ \ \times \left [ |a_R|^2 \langle \varphi_R , g \varphi_R \rangle +|a_L|^2 \langle \varphi_L , g \varphi_L \rangle + \tilde \asy (e^{-\Gamma /\hbar }) \right ] \eee % since (\ref {equa17}). \ If we set % \bee C_R = \langle \varphi_R , g \varphi_R \rangle^2 , \ \ C_L = \langle \varphi_L , g \varphi_L \rangle^2 ,\ \ \tilde C = \sqrt {C_R C_L } \eee % then % \bee r_R &=& \eta C_R a_R |a_R|^2 + \eta \tilde C a_R |a_L|^2 + \frac {\epsilon}{\omega} \tilde \asy (e^{-\Gamma /\hbar} )\\ &=& \eta C_R a_R |a_R|^2 + \eta \tilde C a_R |a_L|^2 + \tilde \asy (e^{-\Gamma /\hbar} ) \eee % since (\ref {equa21}) and, similarly % \bee r_L &=& \eta C_L a_L |a_L|^2 + \eta \tilde C a_L |a_R|^2 + \frac {\epsilon}{\omega} \tilde \asy (e^{-\Gamma /\hbar} ) \\ &=& \eta C_L a_L |a_L|^2 + \eta \tilde C a_L |a_R|^2 + \tilde \asy (e^{-\Gamma /\hbar} ) \eee % Finally % \bee |C_{R,L} | \le \tilde g \| \varphi_{R,L} \|^4 \le C \eee % \end {proof} \begin {definition} We call \emph {two-level approximation} the system of differential equations given by % \be \left \{ \begin {array}{lcl} i b_R' &=& - b_L + \eta \tilde r_R (b_R,b_L) \\ i b_L' &=& - b_R + \eta \tilde r_L (b_R,b_L) \end {array} \right. ,\ \ b_{R,L} (0) = a_{R,L} (0) \label {equa35} \ee % \end {definition} {\it Remarks:} \begin {itemize} \item [-] The \emph {two-level approximation} is obtained, up to an exponentially small term, by substituting $\psi_c \equiv 0$ inside equation (\ref {equa33}). \item [-] The solution of equation (\ref {equa35}) globally exists. \end {itemize} \subsection {First integrals} As for the complete equation (\ref {equa1}) a direct computation proves the following conservation laws. \subsubsection {Conservation of the norm} Let % \bee \tilde {\mathcal N} (b_R,b_L ) = | b_R |^2 + |b_L |^2 \eee % then % \bee \tilde {\mathcal N } [b_R (\tau ),b_L (\tau )] = \tilde {\mathcal N} [b_R (0),b_L (0)] , \ \ \forall \tau \in \R \eee % In particular $\tilde {\mathcal N} (b_R, b_L)=1$ since % \be \tilde {\mathcal N} (b_R, b_L) \equiv |b_R (0)|^2 + |b_L (0)|^2 = |a_R (0)|^2 + |a_L (0)|^2 =\| \psi^0 \|^2 = 1 \label {equa36} \ee \subsubsection {Conservation of the energy} Let, in the case of local perturbation (\ref {equa18}), % \bee \tilde {\mathcal H} (b_R , b_L ) &=& \tilde {\mathcal H}_\ell (b_R ,b_L ) = -\omega \left (\bar b_R b_L + \bar b_L b_R \right ) + \\ && \ \ + \frac {\epsilon}{\sigma +1} \left [ |b_R|^{2(\sigma +1)} \langle \varphi_R^{\sigma +1} , g \varphi_R^{\sigma +1} \rangle + |b_L|^{2(\sigma +1)} \langle \varphi_L^{\sigma +1} , g \varphi_L^{\sigma +1} \rangle \right ] \\ &=& -\omega \left [ \left (\bar b_R b_L + \bar b_L b_R \right ) - \frac {\eta}{\sigma +1} \left ( |b_R|^{2(\sigma +1)} C_R + |b_L|^{2(\sigma +1)} C_L \right ) \right ] \eee % or, in the case of non-local perturbation (\ref {equa19}), % \bee \tilde {\mathcal H} (b_R , b_L ) &=& \tilde {\mathcal H}_{n\ell} (b_R , b_L ) \\ &=& -\omega \left (\bar b_R b_L + \bar b_L b_R \right ) + \epsilon \frac 12 \left [ |b_R|^2 \langle \varphi_R , g \varphi_R \rangle^2 + |b_L|^2 \langle \varphi_L , g \varphi_L \rangle^2 \right ] \\ &=& -\omega \left [ \left (\bar b_R b_L + \bar b_L b_R \right ) - \frac {\eta}{2} \left ( |b_R|^{2} C_R + |b_L|^{2} C_L \right ) \right ] \eee % Then a direct computation gives that % \bee \tilde {\mathcal H} [b_R (\tau ), b_L (\tau )] = \tilde {\mathcal H} [b_R (0), b_L (0) ], \ \ \forall \tau \in \R \eee % \subsection {Analysis of the two-level approximation} \ Here, we perform the qualitative analysis of the two level approximation. \ In particular, we prove that. \begin {theorem} \label {Thm4} There exists a threshold value $\eta^\star >0$ such that the two-level system (\ref {equa35}) admits just two stationary symmetric (that is $|b_R|^2 = |b_L|^2 = \frac 12$) solutions for any $\eta \in [0,\eta^\star ]$. \ At $\eta =\eta^\star$ a bifurcation phenomenon occurs and for $\eta >\eta^\star$ new stationary asymmetric (that is $|b_{R,L} |^2 \not= \frac 12$) solutions appear. \end {theorem} \begin {proof} In order to prove this result we set % \bee b_R = p e^{i \alpha} , \ b_L = q e^{i \beta}, \ z = p^2 - q^2 , \ \theta = \alpha - \beta \eee % where $p$ and $q$ are such that $p^2+q^2 =1$. \ The \emph {imbalance function} $z$ takes value in the interval $[-1,1]$; when $z=1$ then $|b_R|=1$ and $|b_L|=0$ and the wavefunction $\varphi = b_R \varphi_R + b_L \varphi_L$ is practically localized on the \emph {right-side} well, in contrast, when $z=-1$ then $|b_R|=0$ and $|b_L|=1$ and the wavefunction $\varphi $ is practically localized on the \emph {left-side} well. For the sake of definiteness we just consider the local perturbation case, the non-local case could be similarly treated. \ In such a case the \emph {two-level approximation} (\ref {equa35}) takes the form % \be \left \{ \begin {array}{lcl} z' &=& 2\sqrt {1-z^2} \sin \theta \\ \theta' &=& \frac {-2z}{\sqrt {1-z^2}} \cos \theta - \eta \left [ C_R \left ( \frac {1+z}{2} \right )^{\sigma } - C_L \left ( \frac {1-z}{2} \right )^{\sigma } \right ] \end {array} \right. \label {equa37} \ee % and the conservation of the energy takes the form % \bee \tilde {\mathcal H} = \tilde {\mathcal H} (z,\theta ) = - \omega \left \{ \sqrt {1-z^2} \cos \theta - \frac {\eta}{\sigma +1} \left [ C_R \left ( \frac {1+z}{2} \right )^{\sigma +1} + C_L \left ( \frac {1-z}{2} \right )^{\sigma +1} \right ] \right \} \eee % It is not hard to see that when the nonlinear perturbation is small enough then the above dynamical system has only two stationary solutions $(\theta_1 , z_1 )$ and $(\theta_2 , z_2)$ where $\theta_1=0$ and $\theta_2=\pi$ and where $z_{1}=z_2=0$ is the unique solution of the equation % \be \mp \frac {2z}{\sqrt {1-z^2}} - \eta \left [ C_R \left ( \frac {1+z}{2} \right )^{\sigma } - C_L \left ( \frac {1-z}{2} \right )^{\sigma } \right ] =0 \label {equa38} \ee % In contrast, when the strength of the nonlinear perturbation is larger than a threshold parameter $\eta^\star$ then new solutions $z\not= 0$ of equation (\ref {equa38}) appear. \end {proof} For instance, if $g$ is an even function then $C_R = C_L$ and the above equation take the form (where we assume, for the sake of definiteness, that $C_R=1$) % \be \frac {2z}{\sqrt {1-z^2}} - \eta \left [ \left ( \frac {1+z}{2} \right )^{\sigma } - \left ( \frac {1-z}{2} \right )^{\sigma } \right ] =0 \label {equa39} \ee % which has solutions $z_{2} =0$, corresponding to $\theta_2 =\pi$, for any $\eta$. \ In particular, if $\eta $ is larger than a threshold parameter $\eta^\star $ then new solutions appear. \ In particular: \subsubsection {Cubic ($\sigma=1$) and Quintic ($\sigma=2$) Local Nonlinearity} \ In these cases equation (\ref {equa39}) takes the form % \bee \frac {2 z}{\sqrt {1-z^2}}- \eta z =0 \eee % For $\eta$ larger than the threshold value $\eta^\star =2$ then the solution $(z_2,\theta_2 )$ bifurcates (see Figure \ref {Fig1}(a)) and two new solutions $(\theta_3 ,z_3)$ and $(\theta_4 , z_4)$ appear, where $\theta_{3,4}=\pi$ and % \bee z_{3,4} = \pm \frac {\sqrt {\eta^2 -4}}{\eta} \eee \subsubsection {Higher Local Nonlinearity: $\sigma =3$ and $\sigma=4$} \ In the case $\sigma =3$ then we have a picture similar as before. \ That is, for $\eta$ larger than the threshold value $\eta^\star =\frac 83$ then the solution $(z_2,\theta_2 )$ bifurcates (see Figure \ref {Fig1}(a)) and two new solutions $(\theta_3 ,z_3)$ and $(\theta_4 , z_4)$ appear, where $\theta_{3,4}=\pi$ and % \bee z_{3,4} = \pm \left [ \frac {2\chi}{ 3\eta} +\frac {8 \eta} { 3 \chi } -\frac 53 \right ]^{1/2} ,\ \ \eta^\star \le \eta \eee % where % \bee \chi = \left [ \left ( 8\eta^2-108+12 \sqrt { 81 -12 \eta^2 } \right ) \eta \right ]^{1/3} \eee % In the case $\sigma=4$ we have a different picture, that is at $\eta$ coinciding with the threshold value $\eta^\star =\sqrt {\frac {27}{2}}$ then four new solutions $(\theta_j ,z_j)$, $j=3,4,5,6$ appear, where $\theta_{j}=\pi$ and $z_{3,5} = \frac {1}{\sqrt 3}$ and $z_{4,6} = -\frac {1}{\sqrt 3}$. \ In particular, % \bee z_{3,4} &=& \pm \left [ \frac {2\chi }{3\eta } +\frac {2 \eta} {3 \chi } - \frac 13 \right ]^{1/2} ,\ \ \eta^\star \le \eta \\ z_{5,6} &=& \pm \left [ \left ( -\frac {\chi }{3\eta } -\frac { \eta} {3 \chi } - \frac 13 \right ) - i \frac {\sqrt 3}{2} \left ( \frac {2\chi }{3\eta } - \frac {2 \eta} {3 \chi } \right ) \right ]^{1/2} ,\ \ \eta^\star \le \eta \le \eta^+ = 4 \eee % where % \bee \chi = \left [ \left ( \eta^2-27+3 \sqrt { 81 -6 \eta^2 } \right ) \eta \right ]^{1/3} \eee % At $\eta = \eta^+ = 4$ the solutions $z_{5,6}$ collapse to the solution $z_2=0$ (see Figure \ref {Fig1}(b)). {\it Remarks:} \begin {itemize} \item [-] We emphasize that equations (\ref {equa37}) admits, when $\sigma =1$ (cubic nonlinearity) and $C_R=C_L$, an explicit solution by means of Jacobian elliptic function (see \cite {Sa} and the references therein). \item [-] The qualitative behavior of the solutions of equation (\ref {equa37}) could be easily studied by means of the conservation of the energy $\tilde H$ as done, for instance, by \cite {GMS}. \end {itemize} \begin{center} \begin{figure} \includegraphics[height=6cm,width=12cm]{figure_gpdn.eps}% \caption{\label {Fig1} In this figure we plot the graph of the solutions (full lines represent stable centers, broken lines represent unstable centers) of equation (\ref {equa39}) as function of the parameter $\eta$, for given nonlinearity. \ In figure (a) we consider the cases of $\sigma =1,2,3$, where a pitch-fork bifurcation occurs at $\eta = \eta^\star$, where $\eta^\star =2$ for $\sigma =1,2$ and $\eta^\star = \frac 83$ for $\sigma =3$. \ In figure (b) we consider the case $\sigma=4$, in such a case new solutions appear at $\eta =\eta^\star$ where $\eta^\star = \sqrt {27/2}$; at $\eta =\eta^+ =4$ two of them collapses and, then, disappear.} \end{figure} \end{center} \section {Stability of the two-level approximation} \label {Sec6} \subsection {Main result} Our main result consists in proving the stability of the two-level approximation. We prove that: \begin {theorem} \label {Thm5} Let $\psi_c = \Pi_c \psi$, $a_{R}= \langle \psi , \varphi_{R } \rangle $ and $a_{L}= \langle \psi , \varphi_{L } \rangle $, where $\psi$ is the solution of equation (\ref {equa31}), let $b_R $ and $b_L $ be the solution of the system of ordinary differential equations (\ref {equa35}). \ Then, for any fixed $\tau ' >0$ % \be \left | b_{R,L} (\tau ) - a_{R,L} (\tau ) \right | = \tilde \asy (e^{-\Gamma / \hbar}) \ \mbox { and } \ \| \psi_c (\cdot , \tau ) \| = \tilde \asy ( e^{-\Gamma / \hbar}) \label {equa40} \ee % for any $\tau \in [0, \tau' ]$, where $\Gamma >0$ is given in equation (\ref {equa15}). \end {theorem} {\it Remark:} \begin {itemize} \item [-] From this theorem it follows that the time behaviour, at least for times of the order of the beating period, of the wavefunction $\psi $, initially prepared on the lowest states, is practically described by means of the solutions of the two-level approximation given in the previous Section. \end {itemize} \subsection {Proof of Theorem \ref {Thm5}} \ For the sake of simplicity, hereafter, let us drop out the parameters where this does not cause misunderstanding. \ The proof of the theorem is organized in several Lemmas. \begin {lemma} \label {Lem4} Let % \bee \varphi (x,\tau ) = a_R (\tau )\varphi_L (x) + a_L (\tau ) \varphi_L (x) \ \ \mbox { and } \ \ \psi = \psi_c + \varphi \eee % Let $W\psi = W^I + W^{II}$ where: % \begin {itemize} \item [i.] {\bf Local perturbation:} % \be W^I = W^I_\ell = g(x) |\varphi (x,\tau )|^{2\sigma} \varphi (x,\tau ) \label {equa41} \ee % where $W^I_\ell$ does not depend on $\psi_c$, and let % \bee W^{II}_\ell = W_\ell \psi - W^I_\ell \eee % Then, it follows that % \be \| W^I_\ell \| \le C \hbar^{-\sigma d/2} \label {equa42} \ee % and % \be \| W^{II}_\ell \| \le C \hbar^{-d \sigma /2} \| \psi_c \| \label {equa43} \ee % for some positive constant $C$ independent on $\tau $ and $\hbar$. \item [ii.] {\bf Non-Local perturbation:} % \be W^I = W^I_{n\ell} = \langle \varphi (\cdot , \tau ) ,g (\cdot ) \varphi(\cdot , \tau ) \rangle g(x) \varphi (x, \tau ) \label {equa44} \ee % where $W^I_{n\ell}$ does not depend on $\psi_c$, and let % \bee W^{II}_{n\ell} = W_{n\ell}\psi - W^I_{n\ell} \eee % Then, it follows that % \be \| W^I_{n\ell} \| \le C \ \ \mbox { and } \ \ \| W^{II}_{n\ell} \| \le C \| \psi_c \| \label {equa45} \ee % for some positive constant $C$ independent on $\tau $ and $\hbar$. \end {itemize} \end {lemma} \begin {proof} The proof of this Lemma is given in several steps. \ At first we estimate the terms $W^I_{n\ell}$ and $W^{II}_{n\ell}$, then we give the proof of the estimates (\ref {equa42}) and (\ref {equa43}) for the local perturbation case; for what concerns the estimate of the term $W^{II}_\ell$ we consider different cases depending on the dimension $d$. {\it Proof of estimates (\ref {equa45}).} \ Let us consider, at first, the non-local case where $W^I_{n\ell}$ is given by (\ref {equa44}), then % \bee \| W^I_{n\ell} \| &\le & \tilde g | \langle \varphi (\cdot , \tau ) , g (\cdot ) \varphi (\cdot , \tau ) \rangle | \cdot \| \varphi \| \\ &\le & \tilde g^2 \| \varphi^2 \|_1 \cdot \| \varphi \| \\ &\le & \tilde g^2 \| \varphi \|^3 \le C \eee % since $\| \varphi \| \le \| \psi \| \le 1$. \ For what concerns the other term we have that % \bee W^{II}_{n\ell} &=& W_{n\ell} \psi -W^I_{n\ell} \\ &=& g(x) \left [ \left ( \langle \psi_c , g \psi_c \rangle + \langle \varphi , g \psi _c \rangle + \langle \psi_c , g \varphi \rangle \right ) g (\psi_c + \varphi ) + \langle \varphi , g \varphi \rangle \psi_c \right ] \eee % Hence % \bee \| W^{II}_{n\ell} \| &\le & C \left \{ \left [ \| \psi_c \|^2 + 2 \| \varphi \| \cdot \| \psi_c \| \right ] (\| \psi_c \| + \| \varphi \| ) + \| \varphi \|^2 \| \psi_c \| \right \} \\ &\le & C \| \psi_c \| \eee {\it Proof of estimate (\ref {equa42}).} \ The estimate of the term $W^I_\ell$ is immediate. \ Indeed % \bee \| W^I_\ell \| \le \tilde g \| \varphi^{2\sigma+1} \| = \tilde g \| \varphi \|_\infty^{2\sigma} \| \varphi \| \le C \hbar^{-d \sigma /2} \eee % since $\| \varphi \| \le 1$ and (see Lemma \ref {Lem2}) % \be \| \varphi \|_\infty \le | a_R (\tau )| \cdot \| \varphi_R \|_\infty + |a_L (\tau )| \cdot \| \varphi_L \|_\infty \le 2 c_{\infty ,d} = 2 C \hbar^{-d/4} \label {equa46} \ee % {\it Proof of the estimate (\ref {equa43}) --- Dimension $d=1$.} \ In such a case the proof is simpler than the case of dimension 2, because we can make use of the inequality % \be \| \psi_c \|_\infty =c_{\infty , d} \le C \hbar^{-d/4}, \ \ d=1 , \label {equa47} \ee % which immediately follows in the case $d=1$ for any $\sigma $ from the Minkowski inequality and from (\ref {equa30}): % \bee c_{p,d} = C \hbar^{- d \frac {p-2}{4 p}} \ge \| \psi \|_p \ge - \left ( |a_R(\tau )| \| \varphi_R \|_p + |a_L (\tau )| \| \varphi_L \|_p \right ) + \| \psi_c \|_p \eee % for $p=+\infty$, where $|a_{R,L} (\tau ) |\le 1$ and where $\varphi_{R,L}$ satisfy the estimate (\ref {equa12}). \ We assume, for a moment, that % \bee \sigma = n \in \N \eee % is a non negative integer number. \ Then % \be W_{\ell}^{II} &=& W_\ell \psi - W^I_\ell = g(x) \left [ |\psi_c + \varphi |^{2\sigma} (\psi_c + \varphi ) - |\varphi |^{2\sigma } \varphi \right ] \label {equa48} \\ &=& g(x) \left [ \left ( \bar \psi_c + \bar \varphi \right )^n (\psi_c + \varphi )^{n+1} - |\varphi |^{2n} \varphi \right ] \nonumber \\ &=& g (x) \sum_{j=0,\ldots , n; l = 0, \ldots , n+1; j+l >0} \left ( \begin {array}{c} n \\ j \end {array} \right ) \left ( \begin {array}{c} n+1 \\ l \end {array} \right ) \bar \psi_c^j \psi_c^l \bar \varphi^{n-j} \varphi^{n+1-l} \nonumber \ee % Hence, % \be \| W_\ell^{II} \| & \le & \tilde g \sum_{j=0,\ldots , n; l = 0, \ldots , n+1; j+l >0} \left ( \begin {array}{c} n \\ j \end {array} \right ) \left ( \begin {array}{c} n+1 \\ l \end {array} \right ) \| \bar \psi_c^{j+l} \varphi^{2n+1-j-l}\| \nonumber \\ &\le & C \sum_{k=1}^{2n+1} \| \psi_c^k \| \cdot \| \varphi \|_\infty^{2n+1-k} \label {equa49} \\ &\le & C \sum_{k=1}^{2n+1} \| \psi_c \| \cdot \| \psi_c \|_\infty^{k-1} \cdot \| \varphi \|_\infty^{2n+1-k} \nonumber \\ &\le & C \hbar^{-\frac 14 d (2n)} \| \psi_c \| = C \hbar^{- d \sigma /2} \| \psi_c \| \nonumber \ee % since (\ref {equa46}) and (\ref {equa47}). \ If $\sigma$ is not an integer number let % \bee \sigma = n + \alpha , \ \ n \in \N^\star \ \ \mbox { and } \ \ \alpha \in (0,1), \eee % and let % \bee W^{II}_{\ell} = W^{II}_{\ell ,a} + W^{II}_{\ell ,b} \eee % where % \bee W^{II}_{\ell ,a} = g(x) | \psi_c + \varphi |^{2\sigma} \psi_c \ \ \mbox { and } \ \ W^{II}_{\ell ,b} = g (x) \left [ |\psi_c + \varphi |^{2\sigma} - |\varphi |^{2\sigma } \right ] \varphi \eee % As above % \bee \| W^{II}_{\ell ,a} \| \le \tilde g \| \psi_c + \varphi \|_\infty^{2\sigma} \| \psi_c \| \le C \hbar^{-\sigma d /2} \| \psi_c \| \eee % since (\ref {equa46}) and (\ref {equa47}). \ For what concerns the other term $W^{II}_{\ell ,b}$ it follows that % \be && \| W_{\ell ,b}^{II} \| \le \tilde g \| \varphi \|_\infty \left \| |\psi_c + \varphi |^{2\sigma} - |\varphi|^{2\sigma} \right \| \nonumber \\ && \ \ \le \tilde g \| \varphi \|_\infty \left \| |\psi_c + \varphi |^{2n} |\psi_c + \varphi |^{2\alpha}- |\varphi|^{2n} |\varphi |^{2\alpha} \right \| \nonumber \\ && \ \ \le \tilde g \| \varphi \|_\infty \left \| \sum_{j, l =0,\ldots , n} \left ( \begin {array}{c} n \\ j \end {array} \right ) \left ( \begin {array}{c} n \\ l \end {array} \right ) \bar \psi_c^j \psi_c^l \bar \varphi^{n-j} \varphi^{n-l} |\psi_c + \varphi |^{2\alpha} - |\varphi|^{2n} |\varphi |^{2\alpha} \right \| \nonumber \\ & & \ \ \le \tilde g \| \varphi \|_\infty \left [ \left \| \sum_{j,l =0,\ldots , n; \ j+l>1} \left ( \begin {array}{c} n \\ j \end {array} \right ) \left ( \begin {array}{c} n \\ l \end {array} \right ) \bar \psi_c^j \psi_c^l \bar \varphi^{n-j} \varphi^{n-l} |\psi_c + \varphi |^{2\alpha} \right \| + \right. \nonumber \\ && \left. \ \ \ \ + \left \| |\varphi|^{2n} \left ( |\psi_c + \varphi |^{2\alpha} - |\varphi |^{2\alpha} \right ) \right \| \right ] \label {equa50} \\ && \ \ \le C \| \varphi \|_\infty \left [ \sum_{k=1}^{2n} \left ( \| \psi_c^k \varphi^{2n-k} \| \cdot \| \psi_c + \varphi \|^{2\alpha}_\infty \right ) + \right. \nonumber \\ && \left. \ \ \ \ + \left \| |\varphi|^{2n-(1-\alpha )} |\varphi |^{1-\alpha} \left ( |\psi_c + \varphi |^{\alpha}+ |\varphi |^{\alpha} \right ) \left ( |\psi_c + \varphi |^{\alpha} - |\varphi |^{\alpha} \right )\right \| \right ] \nonumber \\ %&& \ \ \le C \| \varphi \|_\infty %\left [ \sum_{k=1}^{2n} \| \psi_c \| \cdot \| \psi_c \|_\infty^{k-1} \cdot \| \varphi %\|_\infty^{2n-k} \left ( \| \psi_c\|_\infty + \| \varphi \|_\infty \right )^{2\alpha} + \left %\| |\varphi|^{2n} \left ( |\psi_c + \varphi |^{2\alpha} - |\varphi |^{2\alpha} \right ) \right %\| \right ] \\ && \ \ \le C \| \varphi \|_\infty \left [ \sum_{k=1}^{2n} \left ( \| \psi_c \| \cdot \| \psi_c \|_\infty^{k-1} \cdot \| \varphi \|_\infty^{2n-k} \left ( \| \psi_c\|_\infty + \| \varphi \|_\infty \right )^{2\alpha} \right ) + \right. \nonumber \\ && \ \ \ \ \left. + \| \varphi \|_\infty^{2n-(1-\alpha )} \left ( \|\psi_c + \varphi \|_\infty^{\alpha}+ \|\varphi \|_\infty^{\alpha} \right ) \left \| |\varphi |^{1-\alpha} \left ( |\psi_c + \varphi |^{\alpha} - |\varphi |^{\alpha} \right )\right \| \right ] \nonumber \\ && \ \ \le C \hbar^{-\frac 14 d} \left [ \| \psi_c \| C \hbar^{-\frac 14 d (2n+2\alpha -1)} + \alpha C \| \psi_c \| \hbar^{-\frac 14 d (2n+2\alpha -1)} \right ] \nonumber \\ &&\ \ \le C \hbar^{-d \sigma /2} \| \psi_c \| \nonumber \ee % where we have make use of the inequality (\ref {equa65}). {\it Proof of estimate (\ref {equa43}) --- Dimension $d=2$.} \ In such a case the estimate (\ref {equa47}) does not hold and we make use here of the Gagliardo-Nirenberg inequality. \ As above, we assume for a moment, that $\sigma = n \in \N$ is a non negative integer number. \ Then, from (\ref {equa49}) it follows that % \be \| W^{II}_{\ell} \| \le C \sum_{k=1}^{2n +1} \| \varphi \|^{2\sigma +1-k}_\infty \| \psi_c^k \|, \ \ \| \varphi \|^{2\sigma +1-k}_\infty \le C \hbar^{- \frac d4 (2\sigma +1-k)} \label {equa51} \ee % The Gagliardo-Nirenberg inequality (\ref {equa68}) gives that % \bee \| \psi_c^k \| &=& \| \psi_c \|^k_{2k} \le C \| \nabla \psi_c \|^{k \delta} \| \psi_c \|^{(1-\delta )k} \eee % where $\delta = \frac d2 - \frac {d}{2k}$. \ From this fact and from the estimate (\ref {equa30}) it follows that % \be \| \psi_c^k \| &\le & C \hbar^{\frac d4 (k-1)} \| \psi_c \|^{\frac d2 + (1-\frac d2)k} \label {equa52} \\ &\le & C \hbar^{-d \frac {k-1}{4}} \| \psi_c \| ,\ \ d=2 \nonumber \ee % Hence, from the above estimate and from (\ref {equa51}) it follows that % \bee \| W^{II}_\ell \| &\le & C \sum_{k=1}^{2n +1} \hbar^{- \frac 14 d (2n+1-k)} \hbar^{-d (k-1)/4} \| \psi_c \| \\ &\le & C \hbar^{-d n /2} \| \psi_c \| = C \hbar^{-d\sigma /2} \| \psi_c \| \eee % Now, if $\sigma$ is not an integer number let $\sigma = n + \alpha$ where $n\in \N^\star$ and $\alpha \in (0,1)$. \ As above, let $W_\ell^{II} = W_{\ell ,a}^{II} + W_{\ell ,b}^{II}$ where % \bee W_{\ell ,a}^{II} = g(x) |\psi_c + \varphi |^{2\sigma} \psi_c \ \ \mbox { and } \ \ W_{\ell ,b}^{II} = g(x) \left [ |\psi_c + \varphi |^{2\sigma} -|\varphi|^{2\sigma} \right ] \varphi \eee % Now, % \be \| W_{\ell ,a}^{II} \| \le \tilde g \left \| |\psi_c + \varphi |^{2\sigma} \psi_c \right \| \label {equa53} \ee % recalling that % \bee |\psi_c + \varphi |^{2\sigma} \le \left ( |\psi_c | + |\varphi |\right )^{2\sigma} \le 2^{2\sigma} \left [ \max (|\psi_c |, |\varphi | ) \right ]^{2\sigma} \le 2^{2\sigma} \left [ |\psi_c |^{2\sigma} + |\varphi |^{2\sigma } \right ] \eee % From this fact and from (\ref {equa52}) then (\ref {equa53}) takes the form % \be \| W^{II}_{\ell ,a} \| & \le & C \left [ \| \psi_c^{2\sigma +1} \| + \| \varphi \|_\infty^{2\sigma} \| \psi_c \| \right ] \label {equa54} \\ &\le & C \hbar^{-d \sigma /2} \| \psi_c \| \nonumber \ee % In order to estimate the other term $W^{II}_{\ell , b}$ from (\ref {equa50}) we have that % \be \| W_{\ell ,b}^{II} \| %&\le & \tilde g \| \varphi \|_\infty %\left [ \left \| \sum_{j=0,\ldots , n; l = 0, \ldots , n+1; j+l>1} %\left ( \begin {array}{c} n \\ j \end {array} \right ) \left ( \begin {array}{c} n \\ l \end %{array} \right ) \bar \psi_c^j \psi_c^l \bar \varphi^{n-j} \varphi^{n-l} |\psi_c + \varphi %|^{2\alpha} \right \| + \right. \\ %&& \left. \ \ + \left \| |\varphi|^{2n} \left ( |\psi_c + \varphi |^{2\alpha} - |\varphi %|^{2\alpha} \right ) \right \| \right ] \\ &\le & C \| \varphi \|_\infty \left [ \sum_{k=1}^{2n} \left ( \| \varphi \|_\infty^{2n-k} \left \| \psi_c^k |\psi_c + \varphi |^{2\alpha} \right \| \right ) + \left \| |\varphi|^{2n} \left ( |\psi_c + \varphi |^{2\alpha} - |\varphi |^{2\alpha} \right ) \right \| \right ] \nonumber \\ &\le & C \| \varphi \|_\infty \left [ \sum_{k=1}^{2n} \left ( \| \varphi \|_\infty^{2n-k} \left ( \left \| \psi_c^{k+2\alpha} \right \| + \| \psi_c^k \| \cdot \| \varphi \|_\infty^{2\alpha} \right ) \right ) + \right. \nonumber \\ && \left. \ \ + \left \| |\varphi|^{2n-(1-\alpha)} |\varphi|^{1-\alpha} \left ( |\psi_c + \varphi |^{\alpha} + |\varphi |^{\alpha} \right ) \left ( |\psi_c + \varphi |^{\alpha} - |\varphi |^{\alpha} \right )\right \| \right ] \nonumber \\ &\le & C \sum_{k=1}^{2n}\left ( \| \varphi \|_\infty^{2n-k+1} \left \| \psi_c^{k+2\alpha} \right \| \right ) + C \sum_{k=1}^{2n} \left ( \| \psi_c^k \| \cdot \| \varphi \|_\infty^{2n-k+1+2\alpha} \right ) + \nonumber \\ && \ \ + C \| \varphi \|_\infty^{2n-(1-\alpha)+1} \left \| ( |\psi_c|^\alpha + |\varphi|^\alpha ) |\psi_c | \right \| \nonumber \\ &\le & \sum_{k=1}^{2n} \left ( \| \varphi \|_\infty^{2n-k+1} \| \psi_c^{k+2\alpha} \| + \| \varphi \|_\infty^{2n-k+1+2\alpha } \| \varphi \|_\infty^{2\alpha} \| \psi_c^k \| \right ) + \nonumber \\ && \ \ + \| \varphi \|_\infty^{2n+\alpha} \left ( \| \psi_c^{1+\alpha} \| + \| \varphi \|_\infty^\alpha \| \psi_c \| \right ) \nonumber \\ &\le & C \sum_{k=1}^{2n} \left [ \hbar^{- \frac d4 (2n-k+1)} \hbar^{-\frac d4 (k+2\alpha -1)} \| \psi_c \|^{\frac d2 + \left ( 1-\frac d2 \right ) (k+2\alpha )} + \right. \nonumber \\ && \left . \ \ + \hbar^{- \frac d4 (2n+2\alpha -k+1)} \hbar^{-\frac d4 (k -1)} \| \psi_c \|^{\frac d2 + \left ( 1-\frac d2 \right ) k} \right ] +\nonumber \\ && \ \ + C \hbar^{- \frac d4 (2n+\alpha)} \left ( \hbar^{-\frac d4 \alpha } \| \psi_c \|^{\frac d2 + \left ( 1-\frac d2 \right ) (1+\alpha )} + \hbar^{-\frac d4 \alpha } \| \psi_c \| \right ) \nonumber \\ &\le & C \sum_{k=1}^{2n} \left [ \hbar^{- \frac d4 (2n+2\alpha )} \| \psi_c \|^{1+ \left ( 1-\frac d2 \right )(k+2\alpha ) } + \hbar^{- \frac d4 (2n+2\alpha)} \| \psi_c \|^{1 + \left ( 1-\frac d2 \right ) (k-1)} \right ] + \nonumber \\ && \ \ + \hbar^{- \frac d4 (2n+2\alpha)} \left ( \| \psi_c \|^{1+ \left ( 1-\frac d2 \right ) \alpha } + \| \psi_c \| \right ) \label {equa55} \\ &\le & C \hbar^{-d\sigma /2} \| \psi_c \| ,\ \ \mbox { since } \ d=2. \ \ \nonumber \ee % obtaining the wanted result by means of the inequalities (\ref {equa51}) and (\ref {equa52}). \end {proof} {\it Remarks:} \begin {itemize} \item [-] For what concerns the nonlinear local perturbation in dimension $d>2$ we emphasize that from Theorem \ref {Thm2} then $\psi = \psi_c + \varphi \in L^p$ for any $p < \frac {2d}{d-2}$ when $d>2$. \ Hence, in order to consider the $L^2$-norm of $W_\ell \psi$ we have to assume that $|\psi|^{2\sigma } \psi \in L^2$, that is $\sigma < \frac {1}{d-2}$. In such a case, then $W_\ell \psi \in L^2$ and we can apply the estimate (\ref {equa52}) to (\ref {equa54}) and (\ref {equa55}) as in the previous case obtaining that % \bee \| W^{II}_{\ell ,a} \| \le C \hbar^{-d\sigma /2} \| \psi_c \|^{\gamma} ,\ \ \gamma = 1+ ( 2-d )\sigma \eee % since $\| \psi_c \| \le 1$, and, similarly, % \bee \| W^{II}_{\ell ,b} \| \le C \hbar^{-d\sigma /2} \| \psi_c \|^{\gamma} \eee % \end {itemize} \begin {lemma} \label {Lem5} $|a_{R,L}' | \le C $ for any $\tau \ge 0$ for some positive constant $C$ independent of $\tau$ and $\hbar$. \end {lemma} \begin {proof} In the local perturbation case from (\ref {equa33}) and from the previous Lemma it follows that % \bee |a_R'| &\le & |a_L| + |r_R| \le |a_L| + \frac {|\epsilon |}{\omega} \| \varphi_R \| \cdot \| W \psi \| \\ &\le & |a_L| + \frac {|\epsilon |}{\omega} \| \varphi_R \| \left [ \| W^I_\ell \| + \| W^{II}_\ell \| \right ] \\ &\le & |a_L| + C \frac {|\epsilon |\hbar^{-\sigma d /2}}{\omega} \| \varphi_R \| \left [ 1 + \| \psi_c \| \right ] \le C \eee % since $|a_L|\le 1$ and $\| \psi_c \| \le 1$ for any $\tau$, $\| \varphi_R \| =1$ and (\ref {equa21}). \ In the same way, the estimate $|a_L'| \le C$ follows. \ The non-local perturbation case similarly follows. \end {proof} \begin {lemma} \label {Lem6} Let $W^I_\ell = W^I_\ell (x,\tau )$ and $W^{II}_{n\ell} = W^I_{n\ell} (x,\tau )$ be defined as in (\ref {equa41}) and (\ref {equa44}); then % % \bee W^I_{\ell },\ W^I_{n\ell} \in C^1 (\R , L^2 (\R^d )) \eee % and % \bee \left \| \frac {\partial W^I_{\ell }}{\partial \tau} \right \|,\ \left \| \frac {\partial W^I_{n\ell }}{\partial \tau} \right \| = c_{\infty , d}^{2 \sigma} \le C \hbar^{-\sigma d /2}, \ \ \forall \tau \ge 0 \ \mbox { and } \ \forall \hbar \in (0,\hbar^\star ) \eee % \end {lemma} \begin {proof} Let us consider, for a moment, the local case where % \bee W^I_\ell &=& W^I_\ell (x,\tau )= g(x) \left | a_R (\tau ) \varphi_R (x) + a_L (\tau ) \varphi_L (x) \right |^{2\sigma} \left ( a_R (\tau ) \varphi_R (x) + a_L (\tau ) \varphi_L (x) \right ) \\ &=& g(x) \left [ \bar a_R (\tau ) \bar \varphi_R (x) + \bar a_L (\tau ) \bar \varphi_L (x) \right ]^{\sigma} \left [ a_R (\tau ) \varphi_R (x) + a_L (\tau ) \varphi_L (x) \right ]^{\sigma +1} \eee % Then % \bee \frac {\partial W^I_\ell}{\partial \tau} &=& g(x) \left \{ \sigma \left ( \bar a_R \bar \varphi_R + \bar a_L \bar \varphi_L \right )^{\sigma -1} \left ( a_R \varphi_R + a_L \varphi_L \right )^{\sigma +1} \left ( \bar a_R' \bar \varphi_R + \bar a_L' \bar \varphi_L \right ) + \right. \\ && \left. \ \ + (\sigma +1) \left | \bar a_R \bar \varphi_R + \bar a_L \bar \varphi_L \right |^{2\sigma } \left ( a_R' \varphi_R + a_L' \varphi_L \right ) \right \} \eee % Hence, % \bee \left \| \frac {\partial W^I_\ell}{\partial \tau} \right \| \le \tilde g (2\sigma +1) \max \left [ |a_R'|,|a_L'| \right ] \max \left [ \| \varphi_R \|_\infty^{2\sigma} , \| \varphi_L \|_\infty^{2\sigma} \right ] = c^{2\sigma}_{\infty , d} \eee % where $|a_{R,L}' |\le C$ from the previous Lemma. \ A similar estimate for the non-local perturbation case follows; where we can derivate with respect to $\tau$ under the integral $\langle \varphi , g \varphi \rangle$ since the integral converges uniformly with respect to $\tau$. \end {proof} We give now an a priori estimate of the term $\psi_c$. \begin {lemma}\label {Lem7} Let $\psi_c = \Pi_c \psi$ where $\psi$ is the solution of equation (\ref {equa31}); it satisfies to the following estimate % \be e^{-C\tau }\| \psi_c \| = \tilde \asy ( e^{-\Gamma / \hbar} ),\ \ \forall \tau \ge 0 , \label {equa56} \ee % for some positive constant $C >0$ independent of $\hbar$ and $\tau$. \end {lemma} {\it Remarks:} \begin {itemize} \item [-] In particular, from (\ref {equa56}) it follows that % \bee \| \psi_c \| = \tilde \asy ( e^{-\Gamma /\hbar} ) \eee % uniformly for any $\tau \in [0,\tau' ]$ for any $\tau'$ fixed, from which the second estimate (\ref {equa40}) follows. \end {itemize} \begin {proof} Now, in order to prove the estimate (\ref {equa56}) we make use of the third equation of (\ref {equa33}) from which follows that % \be \psi_c (\cdot ,\tau ) &=& -i \int_0^\tau e^{-i (H_0-\Omega) (\tau -s)/\omega} r_c ds \nonumber \\ &=& -i \frac {\epsilon }{\omega }\int_0^\tau e^{-i (H_0-\Omega) (\tau -s)/\omega}\Pi_c W (\cdot , \psi )\psi ( \cdot ,s) ds \label {equa57} \ee % since $\psi_c^0 =\Pi_c \psi^0 =0$ from assumption Hyp.2. \ % Therefore, we can write % \bee \psi_c = -i \frac {\epsilon }{\omega }\left [ I + II \right ] \eee % where % \bee I &=& \int_0^\tau e^{-i (H_0 - \Omega )(\tau -s) /\omega } \Pi_c W^I ds , \ \ W^I = W^I (\varphi ) \\ II &=& \int_0^\tau e^{-i (H_0 - \Omega )(\tau -s) /\omega } \Pi_c W^{II} ds , \ \ W^{II} = W^{II} (\varphi , \psi_c ) \eee % For what concerns the first term we have that, by integrating by part, % \bee I &=& \left [ -i \omega e^{-i (H_0 - \Omega )(\tau -s) /\omega } [H_0 - \Omega ]^{-1} \Pi_c W^I \right ]_0^\tau + \\ && \ \ \ + i \omega \int_0^\tau e^{-i (H_0 - \Omega )(\tau -s) /\omega } [H_0 - \Omega ]^{-1} \Pi_c g \frac {\partial W^I }{\partial s} ds \eee % Let us underline that % \bee \left \| e^{-i (H_0 - \Omega )(\tau -s)/\omega } \right \| =1 \eee % and from Lemma \ref {Lem1} it follows that % \bee \| \hbar [H_0 - \Omega ]^{-1} \Pi_c \| \le C . \eee % From these facts and from Lemmas \ref {Lem4} and \ref {Lem6} then we have that % \bee \| I \| &\le & C \frac {\omega}{\hbar} \max_{s\in [0,\tau ]} \left \{ \| W^I \| + \tau \left \| \frac {\partial W^I}{\partial s} \right \| \right \} \\ &\le & C \frac {\omega}{\hbar} \left \{ \hbar^{-d\sigma /2} + \tau c_{\infty ,d}^{2\sigma} \right \} \le C \frac {\omega}{\hbar}\hbar^{-\sigma d/2} (1+\tau ) \eee % For what concerns the other term we have that % \bee \| II \| \le \int_0^\tau \| W^{II} \| ds \le C \hbar^{-\sigma d /2} \int_0^\tau \| \psi_c \| ds \eee % Collecting all these results and denoting % \bee h(\tau ) = \| \psi_c (\cdot ,\tau ) \| \eee % we have that $h(\tau )$ is a non-negative real-valued function satisfying the estimate % \be h(\tau ) &\le & \frac {\epsilon}{\omega} \left \{ C \omega \hbar^{-1-\sigma d /2} (1+\tau ) + C \hbar^{-\sigma d /2} \int_0^\tau h (s) d s \right \} \nonumber \\ &\le & a \int_0^\tau h (s) d s + b (1+\tau ) \label {equa58} \ee % where % \bee a = C \frac {\epsilon \hbar^{-\sigma d /2}}{\omega} = C \eta = O(1) \eee % since (\ref {equa21}) and % \bee b = C \epsilon \hbar^{-1-\sigma d /2} = \tilde \asy \left ( e^{-\Gamma /\hbar} \right ) \eee % since (\ref {equa21}) and (\ref {equa16}). \ Then, the Gronwall's Lemma gives that % \bee h(\tau ) \le be^{a\tau} + \frac ba \left [e^{a\tau } -1 \right ] \le C b e^{C \tau} \eee % proving so the estimate (\ref {equa56}). \end {proof} {\it Remark:} \begin {itemize} \item [-] In dimension $d>2$ for local perturbation case and where $\sigma < \frac {1}{d-2}$, then (\ref {equa58}) takes the form % \bee h(\tau ) \le a \int_0^\tau h^\gamma (s) d s + b (1+\tau ), \ \ \gamma = 1+ (2-d)\sigma \in (0,1), \eee % from which and by means of Gronwall's Lemma arguments the a priori weaker estimate follows % \bee h(\tau ) \le [b+a\tau ]^{1/(1-\gamma)} . \eee % Unfortunately, this estimate is not useful in order to extend the result of Lemma 7 to the case of dimension $d>2$. \end {itemize} Now, we are ready to complete the proof of the theorem. \ Let % \bee J = \left ( \begin {array}{cc} 0 & i \\ i & 0 \end {array} \right ) , \ A = \left ( \begin {array}{c} a_R \\ a_L \end {array} \right ) , \ R = \left ( \begin {array}{c} -i r_R \\ -i r_L \end {array} \right ) \eee % then the system (\ref {equa33}) takes the form % \be \left \{ \begin {array}{lcl} A' &=& J A + R \\ \psi_c' &=& - \frac {i}{\omega} \left [ H_0 -\Omega \right ] \psi_c -i r_c \end {array} \right. , \ \ R = R (A,\psi_c ) \label {equa59} \ee % Let \bee B = \left ( \begin {array}{c} b_R \\ b_L \end {array} \right ) , \ \tilde R = \left ( \begin {array}{c} -i\eta \tilde r_R \\ -i\eta \tilde r_L \end {array} \right ) \eee % then the \emph {two-level approximation} (\ref {equa35}) takes the form % \be B' &=& J B + \tilde R , \ \ \tilde R = \tilde R (B ) \label {equa60} \ee % We underline that % \bee A,B \in S^2 \eee % where % \bee S^2 = \left \{ A = \left ( \begin {array}{c} a_R \\ a_L \end {array} \right ), \ a_R, \ a_L \in \C \ :\ |A| = \sqrt {|a_R|^2 + |a_L|^2} \le 1 \right \} \eee % Let now % \bee F : S^2 \to C^2 \eee % defined as % \be F(A) = JA + \tilde R(A) \label {equa61} \ee % Hence, the first equation of (\ref {equa59}) and equation (\ref {equa60}) can be written as % \be B' = F(B) \ \ \mbox { and } \ \ A' = F (A) + \left [ R (A,\psi_c ) - \tilde R(A) \right ] \label {equa62} \ee \begin {lemma} \label {Lem8} The function $F$ defined in equation (\ref {equa61}) satisfies to the following Lipschitz condition % \be |F(A)-F(B)| \le C |A-B| \label {equa63} \ee % for some $C>0$. \end {lemma} \begin {proof} By definition % \bee F(A)-F(B) = J(A-B) + \left [ \tilde R(A) - \tilde R (B) \right ] \eee % where % \bee \tilde R(A) - \tilde R (B) = \left ( \begin {array}{l} \eta \left [ \tilde r_R (A) - \tilde r_R (B) \right ] \\ \eta \left [ \tilde r_L (A) - \tilde r_L (B) \right ] \end {array} \right ) \eee % In the local perturbation case it follows that (we assume, for definiteness, that $|b_R| \le |a_R|$) % \bee \tilde r_R (A)- \tilde r_R (B) &=& C_R \left [ |a_R|^{2\sigma} a_R - |b_R|^{2\sigma} b_R \right ] \\ &=& C_R \left [ |a_R|^{2\sigma} (a_R - b_R)+(|a_R|^{2\sigma} - |b_R|^{2\sigma} ) b_R \right ] \\ &=& C_R \left [ |a_R|^{2\sigma} (a_R - b_R) + \left ( |a_R|^\sigma - |b_R|^\sigma \right ) \left ( |a_R|^\sigma + |b_R|^\sigma \right ) b_R \right ] \eee % hence % \bee \left | \tilde r_R(A) - \tilde r_R (B) \right | \le C \left | a_R - b_R \right | \eee % since (\ref {equa65}) and since $|a_R|,|b_R|\le 1$. \ The estimate for the term $\tilde r_L (A) - \tilde r_L (B)$ similarly follows. \ In the non-local case it follows that % \bee \tilde r_R (A)- \tilde r_R (B) &=& C_R \left [ |a_R|^{2} a_R - |b_R|^{2} b_R \right ] + \tilde C \left [ a_R |a_L|^{2} - b_R |b_L|^{2} \right ] \\ &=& C_R \left [ |a_R|^{2} (a_R -b_R) + \left ( |a_R|^{2} - |b_R|^{2} \right ) b_R \right ] +\\ && \ \ \ + \tilde C \left [ (a_R -b_R) |a_L|^{2} + b_R \left ( |a_L|^{2}- |b_L|^{2} \right ) \right ] \eee % from which follows (\ref {equa63}) immediately, similarly for $r_L (A) - r_L (B)$. \end {proof} From (\ref {equa62}) it follows that % \be (A-B)' = F(A) - F(B) + \left [ R(A,\psi_c ) - \tilde R(A) \right ], \ \ A(0)=B(0) \label {equa64} \ee % where \begin {lemma} \label {Lem9} For any $\Gamma '$, $0<\Gamma ' < \Gamma$, there exist $\hbar^\star >0$ and $C>0$, independent of $\hbar$ and $\tau$, such that % \bee \left | R(A,\psi_c ) - \tilde R(A) \right | \le C e^{-\Gamma' /\hbar} e^{C \tau}, \ \ \forall \tau \ge 0 , \ \forall \hbar \in (0, \hbar^\star ) \eee % \end {lemma} \begin {proof} Indeed, we have that (for the sake of definiteness we consider only one term) % \bee && r_R (a_R , a_L, \psi_c ) - \eta \tilde r_R (a_R, a_L) = \\ && \ \ = \left [ r_R (a_R , a_L, \psi_c ) - r_R (a_R, a_L,0) \right ] + \left [r_R (a_R , a_L, 0 ) - \eta \tilde r_R (a_R, a_L)\right ] \\ && \ \ = \left [ r_R (a_R , a_L, \psi_c ) - r_R (a_R, a_L,0) \right ] + \tilde \asy (e^{-\Gamma /\hbar}) \eee % since (\ref {equa34}). \ Moreover % \bee r_R (a_R , a_L , \psi_c )= \frac {\epsilon}{\omega} \langle \varphi_R , W \psi \rangle \eee % where $W\psi = W^{I} + W^{II}$ and where $W^I$ does not depend on $\psi_c$. \ From this fact and from Lemmas \ref {Lem4} and \ref {Lem7} it follows that % \bee \left | r_R (a_R , a_L, \psi_c ) - r_R (a_R, a_L,0) \right | &=& \frac {|\epsilon |}{\omega} \left | \langle \varphi_R , W^{II} \rangle \right | \\ &\le & \frac {|\epsilon |}{\omega} \| \varphi_R \| \cdot \| W^{II}\| \le \frac {|\epsilon |}{\omega } C \hbar^{-d\sigma /2} \| \psi_c \| \\ &\le & C e^{-\Gamma' /\hbar} e^{C \tau } \eee % for any $\Gamma' \in (0, \Gamma )$ since (\ref {equa56}). \end {proof} Now, we are ready to complete the proof of the theorem. \ From (\ref {equa64}) it follows that % \bee A(\tau ) - B(\tau ) = \int_0^\tau \left [ F[A(s)] - F[B(s)] \right ] ds + \int_0^\tau \left [ R(A,\psi_c ) - \tilde R(A) \right ] ds \eee % If we set % \bee q(\tau ) = |A(\tau ) - B(\tau )| \eee % then from this equation and from Lemmas 8 and 9 it follows that % \bee q(\tau ) \le C \int_0^\tau q(s) ds + C e^{-\Gamma' /\hbar} \left [ e^{C\tau}-1\right ] \eee % for any $\Gamma ' \in (0, \Gamma )$, from which and from Gronwall's Lemma the desired estimate (\ref {equa40}) follows. \ The proof of the Theorem is so completed. \appendix \section {Inequalities} {\it Basic Inequalities.} \ Let $h(x)= x^\gamma$, defined for $\gamma >0$ and $x \ge 0$. \ Then, for any $y\ge x$, it follows that % \be x \left [ h(y) - h(x) \right ] \le \gamma (y-x) \left \{ \begin {array}{ll} x^{\gamma } & \ \mbox { if } \ 0<\gamma \le 1 \\ y^{\gamma } (x/y)^{\gamma 2^{-n}} & \ \mbox { if } 1 < \gamma \end {array} \right. \label {equa65} \ee % where $n$ is a positive integer such that $2^{n-1} < \gamma \le 2^n$. In order to prove this inequality we consider, at first, the case $\gamma \le 1$. \ The Taylor expansion, up to the second term, with a remainder term gives that % \bee h(y) &=& h(x) + (y-x) h'(x) + \frac 12 (y-x)^2 h'' (\bar x) \\ &=& h(x) +\gamma (y-x)x^{\gamma -1} - \gamma (1-\gamma ) \frac 12 (y-x)^2 \bar x^{\gamma -2} \eee % for some $\bar x \in (x,y)$. \ Hence, the wanted inequality follows since $y \ge x$. If $\gamma >1$ let $n$ be a positive integer such that $2^{n-1} < \gamma \le 2^n$, then % \bee y^\gamma - x^\gamma &=& \left ( y^{\frac 12 \gamma} + x^{\frac 12 \gamma} \right ) \cdot \left ( y^{\frac 12 \gamma} - x^{\frac 12 \gamma} \right ) \\ &=& \left ( y^{\frac 12 \gamma} + x^{\frac 12 \gamma} \right ) \cdot \left ( y^{\frac 14 \gamma} + x^{\frac 14 \gamma} \right ) \cdot \left ( y^{\frac 14 \gamma} - x^{\frac 14 \gamma} \right ) \\ &=& \left [ \Pi_{j=1}^{n} \left ( y^{\gamma 2^{-j}} + x^{\gamma 2^{j-1}} \right ) \right ] \cdot \left ( y^{\gamma 2^{-n}} - x^{\gamma 2^{-n}} \right ) \\ &\le & 2^{n} \left [ \Pi_{j=1}^{n} y^{\gamma 2^{-j}} \right ] x^{\gamma 2^{-n} -1 } \frac {\gamma}{2^n} (y-x) \\ &= & \gamma y^{\gamma (1- 2^{-n})} x^{\gamma 2^{-n} -1 } (y-x) \eee {\it Gagliardo-Nirenberg inequality.} \ The Gagliardo-Nirenberg inequality states that \cite {FIP} % \be \| f \|_{2\sigma +2}^{2\sigma +2} \le C_{\sigma , d} \| \nabla f \|_2^{\sigma d} \| f \|^{2+\sigma (2-d)}_2 \label {equa66} \ee % where % \be \sigma \in \left \{ \begin {array}{ll} \ [0,+\infty ] & \ \mbox { if }\ d=1 \\ \ [0,+\infty ) & \ \mbox { if }\ d=2 \\ \ [0,2/(d-2)) & \ \mbox { if }\ d>2 \end {array} \right. \label {equa67} \ee % and where $C$ is a given constant. \ Such an inequality (\ref {equa66}) can be also rewritten as % \be \| f \|_p \le C_{p,d} \| \nabla f \|^\delta \| f\|^{1-\delta}, \ \ \delta = \frac {\sigma d}{2(\sigma +1)} = \frac {(p-2)d}{2p} = \frac d2 - \frac dp \label {equa68} \ee % where % \bee p \in \left \{ \begin {array}{ll} \ [2,+\infty ] & \ \mbox { if }\ d=1 \\ \ [2,+\infty ) & \ \mbox { if }\ d=2 \\ \ [2,2d/(d-2)) & \ \mbox { if }\ d>2 \end {array} \right. \eee % \begin{thebibliography}{99} \bibitem {AFGST} W.H. Aschbacher, J. Froehlich, G.M. Graf, K. Schnee, M. Troyer {\it Symmetry breaking regime in the nonlinear Hartree equation.} J. Math. Phys. {\bf 43}, 3879-3891 (2002). \bibitem {BS} F.A. Berezin, M.A. Shubin, {\it The Schr\"odinger equation.} (Kluwer Ac. Publ., 1991). \bibitem {C1} R. Carles, {\it Nonlinear Schr\"odinger equations with repulsive harmonic potential and applications.} SIAM J. Math. Anal. {\bf 35}, 823-43 (2003). \bibitem {CW} T. Cazenave and F.B. Weissler, {\it The Cauchy problem for the nonlinear Schr\"odinger equation in $H^1$.} Manuscripta Math. {\bf 61}, 477-94 (1988). \bibitem {DGPS} F. Dalfovo, S. Giorgini, L.P. Pitavskii, S. Stringari {\it Theory of Bose-Einstein condensation in trapped gases.} Rev. Mod. Phys. {\bf 71}, 463-512 (1999). \bibitem {FIP} G. Fibich, B. Ilan, G. Papanicolaou {\it Self-focusing with fourth-order dispersion.} SIAM J. Appl. Math. {\bf 62}, 1437-1462 (2002). \bibitem {Gi} J. Ginibre {\it An introduction to nonlinear Schr\"odinger equations} in Nonlinear waves. (R. Agemi, Y. Giga, T. Ozawa, eds.) (1997). \bibitem {GM} V. Grecchi, A. Martinez {\it Non-linear Stark effect and molecular localization.} Comm. Math. Phys. {\bf 166}, 533-48 (1995). \bibitem {GMS} V. Grecchi, A. Martinez, A. Sacchetti {\it Destruction of the beating effect for a non-linear Schr\"odinger equation.} Comm. Math. Phys. {\bf 227}, 191-209 (2002). \bibitem {HS} B. Helffer, J. Sj\"ostrand, {\it Multiple wells in the semi-classical limit I.} Comm. P.D.E. {\bf 9}, 337-408 (1984). \bibitem {He} B. Helffer, {\it Semi-Classical Analysis for the Schr\"odinger Operator and Applications.} Lecture Notes in Mathematics 1336 (Springer-Verlag: Berlin Heidelberg 1988) \bibitem {JPT} G. Jona-Lasinio, C. Presilla and C. Toninelli, {\it Interaction induced localization in a gas of pyramidal molecules.} Phys. Rev. Lett. {\bf 88}, 123001 (2002). \bibitem {LSY} E.H. Lieb, R. Seiringer, J. Yngvason {\it A rigorous derivation of the Gross-Pitaevskii energy functional for a two-dimensional Bose gas.} Comm. Math. Phys. {\bf 224}, 17-31 (2001). \bibitem {RSFS} S. Raghavan, A. Smerzi, S. Fantoni, S.R. Shenoy {\it Coherent oscillations between two weakly coupled Bose-Einstein condensates: Josephson effects, $\pi$ oscillations, and macroscopic quantum self-trapping.} Phys. Rev. A {\bf 59}, 620-633 (1999). \bibitem {RS4} M. Reed, B. Simon, {\it Methods of Modern Mathematical Physics: IV Analysis of Operators.} (Academic Press: New-York 1972). \bibitem {Sa} A. Sacchetti {\it Nonlinear time-dependent one-dimesnional Schr\"odinger equation with double well potential.} SIAM: J. Math. Anal. (2004). \bibitem {SS} C. Sulem, P.L. Sulem {\it The nonlinear Schr\"odinger equation: self-focusing and wave collapse.} (Springer-Verlag, New York 1999). \bibitem {VA} A. Vardi, J.R. Anglin {\it Bose-Einstein condensates beyond mean field theory: Quantum back reaction as decoherence.} Phys. Rev. Lett. {\bf 86} (2001), 568-71. \bibitem {Zh} J. Zhang {\it Stability of attractive Bose-Einstein condensates.} J. Stat. Phys. {\bf 101}, 731-46 (2000). \end{thebibliography} \end {document} ---------------0407291121184 Content-Type: application/postscript; name="figure_gpdn.eps" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="figure_gpdn.eps" %!PS-Adobe-3.0 EPSF-3.0 %%Title: Maple plot %%Creator: MapleV %%Pages: 1 %%BoundingBox: 43 62 1067 689 %%DocumentNeededResources: font Courier %%EndComments 20 dict begin gsave /m {moveto} def /l {lineto} def /C {setrgbcolor} def /Y /setcmykcolor where { %%ifelse Use built-in operator /setcmykcolor get }{ %%ifelse Emulate setcmykcolor with setrgbcolor { %%def 1 sub 3 { %%repeat 3 index add neg dup 0 lt { pop 0 } if 3 1 roll } repeat setrgbcolor } bind } ifelse def /G {setgray} def /S {stroke} def /NP {newpath} def %%%%This draws a filled polygon and avoids bugs/features %%%% of some postscript interpreters %%%%GHOSTSCRIPT: has a bug in reversepath - removing %%%%the call to reversepath is a sufficient work around /P {gsave fill grestore reversepath stroke} def %%%%This function is needed for drawing text /stringbbox {gsave NP 0 0 m false charpath flattenpath pathbbox 4 2 roll pop pop 1.1 mul cvi exch 1.1 mul cvi exch grestore} def /thin 3 def /medium 7 def /thick 16 def /boundarythick 20 def %% thickness of bounding box %%IncludeResource: font Courier 36 36 translate 0.108 0.108 scale 1 setlinejoin 1 setlinecap 0.0 setgray /inch {72 mul} def /fheight 0.35 inch neg def 0 G medium setlinewidth [] 0 setdash NP 1836 3331 m 1855 3711 l 1874 3846 l 1893 3951 l 1912 4038 l 1957 4203 l 2001 4336 l 2040 4436 l 2080 4524 l 2166 4687 l 2248 4815 l 2334 4929 l 2413 5019 l 2498 5105 l 2586 5184 l 2662 5245 l 2745 5305 l 2831 5362 l 2915 5411 l 2996 5456 l 3086 5500 l 3167 5538 l 3253 5574 l 3331 5605 l 3417 5636 l 3498 5663 l 3582 5690 l 3664 5715 l 3750 5739 l 3833 5760 l 3918 5781 l 4002 5800 l 4079 5817 l 4168 5835 l 4247 5850 l 4332 5866 l 4413 5880 l 4500 5894 l S NP 1836 3331 m 1855 2952 l 1874 2817 l 1893 2712 l 1912 2625 l 1957 2459 l 2001 2327 l 2040 2227 l 2080 2139 l 2166 1976 l 2248 1848 l 2334 1734 l 2413 1644 l 2498 1558 l 2586 1479 l 2662 1418 l 2745 1358 l 2831 1301 l 2915 1252 l 2996 1207 l 3086 1162 l 3167 1125 l 3253 1089 l 3331 1058 l 3417 1027 l 3498 999 l 3582 973 l 3664 948 l 3750 924 l 3833 903 l 3918 882 l 4002 863 l 4079 846 l 4168 828 l 4247 812 l 4332 797 l 4413 783 l 4500 769 l S NP 500 3331 m 1836 3331 l S [50 50] 0 setdash NP 1836 3331 m 4500 3331 l S [] 0 setdash NP 1833 3331 50 0 360 arc fill medium setlinewidth NP 500 666 m 500 5999 l S thin setlinewidth NP 484 727 m 516 727 l S NP 484 846 m 516 846 l S NP 468 964 m 532 964 l S NP 484 1082 m 516 1082 l S NP 484 1201 m 516 1201 l S NP 484 1319 m 516 1319 l S NP 484 1437 m 516 1437 l S NP 468 1556 m 532 1556 l S NP 484 1674 m 516 1674 l S NP 484 1793 m 516 1793 l S NP 484 1911 m 516 1911 l S NP 484 2029 m 516 2029 l S NP 468 2148 m 532 2148 l S NP 484 2266 m 516 2266 l S NP 484 2384 m 516 2384 l S NP 484 2503 m 516 2503 l S NP 484 2621 m 516 2621 l S NP 468 2740 m 532 2740 l S NP 484 2858 m 516 2858 l S NP 484 2976 m 516 2976 l S NP 484 3095 m 516 3095 l S NP 484 3213 m 516 3213 l S NP 468 3331 m 532 3331 l S NP 484 3450 m 516 3450 l S NP 484 3568 m 516 3568 l S NP 484 3687 m 516 3687 l S NP 484 3805 m 516 3805 l S NP 468 3923 m 532 3923 l S NP 484 4042 m 516 4042 l S NP 484 4160 m 516 4160 l S NP 484 4278 m 516 4278 l S NP 484 4397 m 516 4397 l S NP 468 4515 m 532 4515 l S NP 484 4634 m 516 4634 l S NP 484 4752 m 516 4752 l S NP 484 4870 m 516 4870 l S NP 484 4989 m 516 4989 l S NP 468 5107 m 532 5107 l S NP 484 5225 m 516 5225 l S NP 484 5344 m 516 5344 l S NP 484 5462 m 516 5462 l S NP 484 5581 m 516 5581 l S NP 468 5699 m 532 5699 l S NP 484 5817 m 516 5817 l S NP 484 5936 m 516 5936 l S medium setlinewidth NP 500 666 m 4500 666 l S thin setlinewidth NP 1833 623 m 1833 709 l S medium setlinewidth /Courier findfont 150 scalefont setfont (z) dup stringbbox 2 idiv 6000 exch sub exch 600 exch sub exch m show ((a)) dup stringbbox 2 idiv 6000 exch sub exch 4000 exch sub exch m show (-0.8) dup stringbbox 2 idiv 964 exch sub exch 468 exch sub exch m show (-0.6) dup stringbbox 2 idiv 1556 exch sub exch 468 exch sub exch m show (-0.4) dup stringbbox 2 idiv 2148 exch sub exch 468 exch sub exch m show (-0.2) dup stringbbox 2 idiv 2740 exch sub exch 468 exch sub exch m show (0) dup stringbbox 2 idiv 3331 exch sub exch 468 exch sub exch m show (0.2) dup stringbbox 2 idiv 3923 exch sub exch 468 exch sub exch m show (0.4) dup stringbbox 2 idiv 4515 exch sub exch 468 exch sub exch m show (0.6) dup stringbbox 2 idiv 5107 exch sub exch 468 exch sub exch m show (0.8) dup stringbbox 2 idiv 5699 exch sub exch 468 exch sub exch m show /Courier findfont 100 scalefont setfont (x) dup stringbbox 605 exch sub exch 2 idiv 1900 exch sub exch m show /Symbol findfont 150 scalefont setfont (h) dup stringbbox 540 exch sub exch 2 idiv 1833 exch sub exch m show (h) dup stringbbox 400 exch sub exch 2 idiv 4500 exch sub exch m show 5000 0 translate 1 setlinejoin 1 setlinecap 0.0 setgray /inch {72 mul} def /fheight 0.35 inch neg def 0 G medium setlinewidth [] 0 setdash NP 500 3331 m 2500 3331 l S [50 50] 0 setdash NP 2500 3331 m 4500 3331 l S [] 0 setdash NP 1848 4983 m 1850 5029 l 1853 5049 l 1855 5063 l 1865 5104 l 1874 5134 l 1893 5177 l 1912 5211 l 1957 5272 l 2001 5318 l 2080 5382 l 2166 5437 l 2248 5480 l 2334 5519 l 2413 5549 l 2498 5579 l 2586 5607 l 2662 5628 l 2745 5650 l 2831 5671 l 2915 5689 l 2996 5706 l 3086 5723 l 3167 5738 l 3253 5752 l 3331 5765 l 3417 5777 l 3498 5789 l 3582 5800 l 3664 5811 l 3750 5821 l 3833 5831 l 3918 5840 l 4002 5849 l 4079 5857 l 4168 5865 l 4247 5873 l 4332 5880 l 4413 5887 l 4500 5894 l S NP 1848 1680 m 1850 1634 l 1853 1614 l 1855 1600 l 1865 1559 l 1874 1529 l 1893 1486 l 1912 1452 l 1957 1391 l 2001 1345 l 2080 1281 l 2166 1226 l 2248 1183 l 2334 1144 l 2413 1114 l 2498 1084 l 2586 1056 l 2662 1035 l 2745 1013 l 2831 992 l 2915 974 l 2996 957 l 3086 940 l 3167 925 l 3253 911 l 3331 898 l 3417 885 l 3498 874 l 3582 863 l 3664 852 l 3750 842 l 3833 832 l 3918 823 l 4002 814 l 4079 806 l 4168 797 l 4247 790 l 4332 783 l 4413 776 l 4500 769 l S [50 50] 0 setdash NP 1848 4983 m 1850 4934 l 1853 4913 l 1855 4896 l 1865 4847 l 1874 4810 l 1893 4751 l 1912 4701 l 1957 4603 l 2001 4517 l 2040 4447 l 2080 4379 l 2123 4307 l 2166 4235 l 2207 4166 l 2248 4095 l 2291 4019 l 2334 3937 l 2373 3855 l 2413 3761 l 2434 3703 l 2455 3636 l 2466 3596 l 2476 3550 l 2487 3492 l 2498 3395 l 2500 3331 l S NP 1848 1680 m 1850 1729 l 1853 1750 l 1855 1767 l 1865 1816 l 1874 1853 l 1893 1912 l 1912 1962 l 1957 2060 l 2001 2146 l 2040 2216 l 2080 2284 l 2123 2356 l 2166 2428 l 2207 2497 l 2248 2567 l 2291 2644 l 2334 2726 l 2373 2808 l 2413 2902 l 2434 2960 l 2455 3027 l 2466 3067 l 2476 3113 l 2487 3171 l 2498 3268 l 2500 3331 l S [] 0 setdash NP 1848 1680 50 0 360 arc fill NP 1848 4983 50 0 360 arc fill NP 2500 3331 50 0 360 arc fill medium setlinewidth NP 500 666 m 500 5999 l S thin setlinewidth NP 484 701 m 516 701 l S NP 484 815 m 516 815 l S NP 484 929 m 516 929 l S NP 468 1044 m 532 1044 l S NP 484 1158 m 516 1158 l S NP 484 1272 m 516 1272 l S NP 484 1387 m 516 1387 l S NP 484 1501 m 516 1501 l S NP 468 1616 m 532 1616 l S NP 484 1730 m 516 1730 l S NP 484 1844 m 516 1844 l S NP 484 1959 m 516 1959 l S NP 484 2073 m 516 2073 l S NP 468 2188 m 532 2188 l S NP 484 2302 m 516 2302 l S NP 484 2416 m 516 2416 l S NP 484 2531 m 516 2531 l S NP 484 2645 m 516 2645 l S NP 468 2760 m 532 2760 l S NP 484 2874 m 516 2874 l S NP 484 2988 m 516 2988 l S NP 484 3103 m 516 3103 l S NP 484 3217 m 516 3217 l S NP 468 3331 m 532 3331 l S NP 484 3446 m 516 3446 l S NP 484 3560 m 516 3560 l S NP 484 3675 m 516 3675 l S NP 484 3789 m 516 3789 l S NP 468 3903 m 532 3903 l S NP 484 4018 m 516 4018 l S NP 484 4132 m 516 4132 l S NP 484 4247 m 516 4247 l S NP 484 4361 m 516 4361 l S NP 468 4475 m 532 4475 l S NP 484 4590 m 516 4590 l S NP 484 4704 m 516 4704 l S NP 484 4819 m 516 4819 l S NP 484 4933 m 516 4933 l S NP 468 5047 m 532 5047 l S NP 484 5162 m 516 5162 l S NP 484 5276 m 516 5276 l S NP 484 5390 m 516 5390 l S NP 484 5505 m 516 5505 l S NP 468 5619 m 532 5619 l S NP 484 5734 m 516 5734 l S NP 484 5848 m 516 5848 l S NP 484 5962 m 516 5962 l S medium setlinewidth NP 500 666 m 4500 666 l S thin setlinewidth NP 1848 623 m 1848 709 l S NP 2500 623 m 2500 709 l S medium setlinewidth /Courier findfont 150 scalefont setfont (z) dup stringbbox 2 idiv 6000 exch sub exch 600 exch sub exch m show ((b)) dup stringbbox 2 idiv 6000 exch sub exch 4000 exch sub exch m show (-0.8) dup stringbbox 2 idiv 1044 exch sub exch 468 exch sub exch m show (-0.6) dup stringbbox 2 idiv 1616 exch sub exch 468 exch sub exch m show (-0.4) dup stringbbox 2 idiv 2188 exch sub exch 468 exch sub exch m show (-0.2) dup stringbbox 2 idiv 2760 exch sub exch 468 exch sub exch m show (0) dup stringbbox 2 idiv 3331 exch sub exch 468 exch sub exch m show (0.2) dup stringbbox 2 idiv 3903 exch sub exch 468 exch sub exch m show (0.4) dup stringbbox 2 idiv 4475 exch sub exch 468 exch sub exch m show (0.6) dup stringbbox 2 idiv 5047 exch sub exch 468 exch sub exch m show (0.8) dup stringbbox 2 idiv 5619 exch sub exch 468 exch sub exch m show /Courier findfont 100 scalefont setfont (x) dup stringbbox 605 exch sub exch 2 idiv 1900 exch sub exch m show (+) dup stringbbox 605 exch sub exch 2 idiv 2500 exch sub exch m show /Symbol findfont 150 scalefont setfont (h) dup stringbbox 540 exch sub exch 2 idiv 1833 exch sub exch m show (h) dup stringbbox 540 exch sub exch 2 idiv 2433 exch sub exch m show (h) dup stringbbox 400 exch sub exch 2 idiv 4500 exch sub exch m show showpage grestore end %%Trailer %%EOF ---------------0407291121184--