Content-Type: multipart/mixed; boundary="-------------0407230911259" This is a multi-part message in MIME format. ---------------0407230911259 Content-Type: text/plain; name="04-227.keywords" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="04-227.keywords" non-linear wave equation, quasi-periodic solutions ---------------0407230911259 Content-Type: application/x-tex; name="prof.tex" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="prof.tex" \newcommand{\al}{\alpha} \newcommand{\bet}{\beta} \newcommand{\de}{\delta} \newcommand{\eps}{\varepsilon} \newcommand{\miti}{{-\infty}} \newcommand{\ga}{\gamma} \newcommand{\ka}{{\kappa}} \newcommand{\gam}{\gamma} \newcommand{\la}{\lambda} \newcommand{\Lam}{\Lambda} \newcommand{\fhi}{\varphi} \newcommand{\teta}{\vartheta} \newcommand{\ome}{\omega} \newcommand{\Ome}{\Omega} \newcommand{\sig}{\sigma} \newcommand{\lam}{\lambda} \newcommand{\Beq}[2]{\begin{equation}\label{#1}{#2}\end{equation}} \newcommand{\np}{\noindent} \newcommand{\1} {{\frac{1}{ 2}}} \newcommand{\6} {{\frac{1}{ 6}}} \newcommand{\sqp}{{\sqrt\eps}} \newcommand{\R}{\mathbb{R}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\C}{\mathbb{C}} \newcommand{\Q} {\mathbb {Q}} \newcommand{\T} {\mathbb {T}} \newcommand{\Ba}{{\mathcal B}} \newcommand{\B}{{\mathcal H}} \newcommand{\A}{{\cal A}} \newcommand{\N}{{\cal N}} \newcommand{\Na}{{\mathbb{N}}} \newcommand{\Ca}{{\mathcal C}} \documentclass[10pt, a4 paper, twoside, openright]{amsart} \usepackage{amsfonts} \usepackage{latexsym} \usepackage{amsthm} \usepackage{amssymb} \usepackage{amsopn} \usepackage{layout} %\usepackage[namelimits, intlimits]{amsmath} \usepackage{fancyhdr} %\usepackage[english]{babel} \usepackage{mathrsfs} \theoremstyle{plain} \newtheorem{teo}{Theorem} \newtheorem{Teo}{\em Theorem} \newtheorem{pro}{Proposition} \newtheorem{Pro}{\em Proposition} \newtheorem{lem}{Lemma}[section] \newtheorem{cor}[lem]{Corollary} \newtheorem{nota}{Notational Convention} \newtheorem{df}{Definition} \newtheorem{Df}{\em Definition} %\newtheorem{cond}[pro]{Condition} \newtheorem{oss}{Remark} \numberwithin{equation}{section} \theoremstyle{definition} \theoremstyle{remark} \newtheorem{exe}[teo]{Example} \setlength{\evensidemargin}{22pt} \setlength{\topmargin}{0pt} \setlength{\textwidth}{440pt} \setlength{\textheight}{630pt} \setlength{\hoffset}{-10pt} \setlength{\marginparwidth}{0pt} \setlength{\marginparsep}{0pt} \setlength{\oddsidemargin}{22pt} \setlength{\marginparpush}{0pt} \setlength{\baselineskip}{500pt} \title[Quasi-periodic solutions for non-linear wave equations]{Quasi-periodic solutions for completely resonant non-linear wave equations in 1D and 2D.} \author[M. Procesi]{Michela Procesi*} \email{ procesi@sissa.it} \thanks{{\bf Acknowledgments:} The author gives warm thanks to Massimiliano Berti for many helpful discussions and for his support in the writing of this article.} \begin{document}\clearpage{\pagestyle{empty}\cleardoublepage} \begin{abstract} We provide quasi-periodic solutions with two frequencies $\omega\in \mathbb{R}^2$, for a class of completely resonant non-linear wave equations in one and two spatial dimensions and with periodic boundary conditions. This is the first existence result for quasi-periodic solutions in the completely resonant case. The main idea is to work in an appropriate invariant subspace, in order to simplify the bifurcation equation. The frequencies, close to that of the linear system, belong to an uncountable Cantor set of measure zero where no small divisor problem arises. \end{abstract} \maketitle \begin{section}{Introduction} We consider the completely resonant nonlinear wave equations in $d=1$ and $d=2$ spatial dimensions with $2\pi$ periodic boundary conditions: \Beq{e-0}{\qquad\text{a})\;\left\{\begin{aligned} & v_{tt}-v_{xx}=- v^3+ f(v)\\ & v(x, t)= v(x+2\pi, t)\end{aligned}\right.\, , \qquad \text{b})\;\left\{\begin{aligned}& v_{tt}-v_{xx}-v_{yy}=- v^3+f(v)\\ & v(x, y, t)= v(x+2h\pi, y+ 2k \pi, t)\;\quad\forall k,h\in\Z\end{aligned}\right.} \np where $f(v)$ is an odd analytic function at $v=0$ of degree at least five. \vskip5pt \np In this paper we prove existence of small amplitude quasi-periodic solutions of Equations \ref{e-0}, with two frequencies $\ome\in \R^2$. %Notice that Equations \ref{e-0} linearized in $v=0$ are completely resonant, namely that all the solutions are periodic of frequency one. \vskip5pt \np Existence of periodic solutions for the one dimensional equation \ref{e-0}a) has been proved in the papers: \cite{ls}, \cite{bcp}, \cite{bb}, \cite{bb2}, \cite{gmp}, \cite{bb3} both for periodic and Dirichlet boundary conditions. \vskip5pt Up to now quasi-periodic solutions for Equations \ref{e-0} have not been found. \np One should remark that quasi-periodic solutions for non resonant (or partially resonant) nonlinear Hamiltonian PDE's have been widely studied, see for instance \cite{kp}, \cite{p}, \cite{cy}, \cite{cw}, \cite{b1}, \cite{b2}, \cite{b3} and references therein. In these cases the linearized equation at zero already has quasi-periodic solutions (which arise from a foliation by invariant tori). Indeed, given $\ome\in\R^n$, the space of quasi-periodic solutions of frequency $\ome$ is always finite dimensional (possibly empty). % which move on finite dimensional tori, the problem being their continuation. On the other hand the completely resonant equations \ref{e-0}, linearized in zero, possess infinite dimensional spaces of periodic solutions with the same period. Indeed, given $\ome\in\R^n$, the space of quasi-periodic solutions of frequency $\ome$ is always either infinite dimensional or empty. Actually for $d=1$ all the solutions of the associated linear equation are of the form: \Beq{ee}{ v_0(x, t)= r(x+t)+s(x-t), } hence $2\pi$ periodic. When looking for a quasi-periodic solution $Q(x, \ome t)$ of equation \ref{e-0}a),with $\ome\in \R^N$ and $\ome_i\sim 1$, it is not at all clear from which solution of type \ref{ee}, and from which frequency, should $Q(x, \ome t)$ branch off. For $d=2$ the picture is still more complicated as there are infinite dimensional spaces both of periodic and quasi-periodic solutions. \vskip10pt \np The main idea of this paper is to look for solutions in appropriate invariant subspaces of functions $u: \T^2\rightarrow \R$. %The main purpose of such reduction is to restrict the problem to a space of doubly periodic functions of two independent variables\footnote{notice that the variables must be independent also for $\eps=0$.}. \np In such subspaces the problem is similar to that of finding { periodic} solutions for equation \ref{e-0}a). \vskip5pt \np To motivate our choice of subspaces let us first consider the case $d=1$; the simplest way to correct \ref{ee} obtaining a quasi-periodic function is to change by small and different quantities the velocity of a forward and a backward traveling wave: $$v(x, t) = r(\ome_1 t+x)+ s( \ome_2 t-x)+ \text{ small corrections}\, , \qquad \ome_i\sim 1.$$ \np As equation \ref{e-0}a) has constant coefficients, looking for such a solution is equivalent to restricting $v(x, t)$ to the invariant subspace: \Beq{e00}{ \left\{\begin{aligned}& v(x, t)= u(x+\ome_1 t, \ome_2 t-x)\\ & u(\fhi_1+2k\pi, \fhi_2+2 h \pi)= u(\fhi_1, \fhi_2),\qquad\qquad\forall k,h\in\Z\end{aligned}\right.} \np In the case $d=2$ again we consider the simplest possible solution: $$v(x, y, t) = r( t+x)+ s( t+y), $$ and correct the velocities of the two waves; this is equivalent to looking for solutions in the invariant subspace: \Beq{e01}{ \left\{\begin{aligned}& v(x, y, t)= u(x+\Ome_1 t, y+\Ome_2 t)\, , \quad\qquad \Ome_i\sim 1\\ & u(\fhi_1+2k\pi, \fhi_2+2 h \pi)= u(\fhi_1, \fhi_2),\qquad\qquad\forall k,h\in\Z\end{aligned}\right.} We define the frequencies to be: \Beq{om}{ \ome=(1+\eps, 1+a \eps)\, , \qquad \Ome= (\sqrt{1+\eps}, \sqrt{1+a\eps});} notice that for $\eps=0$ the subspaces \ref{e00} and \ref{e01} are spaces of periodic solutions in $t$. \vskip5pt In the subspaces defined in \ref{e00}and \ref{e01}, finding quasi-periodic solutions of Equations \ref{e-0} of frequency respectively $\ome$ and $\Ome$, is equivalent to finding doubly $2\pi$ periodic solutions for the equations: \begin{equation}\label{e1}d=1:\qquad\;\;\left\{\begin{aligned}& [(\ome_1^2-1) \partial^2_{\fhi_1}+(\ome_2^2-1)\partial^2_{\fhi_2} + 2(\ome_1\ome_2+1)\partial_{\fhi_1}\partial_{\fhi_2}] u(\fhi)=- u^3(\fhi)+f(u)\\ & u(\fhi_1+2k\pi, \fhi_2+2 h \pi)= u(\fhi_1, \fhi_2),\qquad\qquad\forall k,h\in\Z\end{aligned}\right. \end{equation} \Beq{ep1}{ d=2:\qquad\;\;\left\{\begin{aligned} &[(\Ome_1^2-1)\partial^2_{\fhi_1}+ (\Ome_2^2-1)\partial^2_{\fhi_2}+ 2\Ome_1\Ome_2\partial_{\fhi_1}\partial_{\fhi_2}] u(\fhi)=- u^3(\fhi)+f(u)\\ & u(\fhi_1+2k\pi, \fhi_2+2 h \pi)= u(\fhi_1, \fhi_2),\qquad\qquad\forall k,h\in\Z.\end{aligned}\right.\qquad\;} Equations \ref{e1} and \ref{ep1} can be written as: \Beq{l2}{\left\{\begin{aligned}& L_\al[u(\fhi)]= -u^3(\fhi)+f(u)\\ &u(\fhi_1+2k\pi, \fhi_2+2 h\pi)= u(\fhi_1, \fhi_2)\end{aligned}\right.\, .} where: \Beq{l1}{ L_{ \al}[u]:= \al_0(\eps \al_1\partial_{\fhi_1}+\partial_{\fhi_2})\circ(\eps \al_2\partial_{\fhi_2}+\partial_{\fhi_1})u(\fhi), } for an appropriate choice of ${ \al}:=(\al_0, \al_1, \al_2)\in\R^3$. Having unified the notation, from now on we work on equation \ref{l2}. We rescale equation \ref{l2} in order to highlight the relationship between the amplitude and the variation in frequency: $$ u(\fhi)\rightarrow \sqp u(\fhi).$$ In the following we consider the scaled equation: \Beq{L1}{ L_\al[u]=-\eps u^3+\eps^2f(u, \eps).} \vskip5pt We are now ready to state the main results of the paper. \begin{Df} \label{d1} Given a positive $\sig\in \R$, let $\B_\sig$ be the Hilbert space of odd analytic functions $\T^2\rightarrow \R$, equipped with the norm: $$|f|^2_\sig= \sum_{j\in\Z^2} |\hat f_j|^2(|j|^2+1) e^{2|j|\sig} .$$\end{Df} \begin{Df} \label{d2} Let $r_0(\fhi_1, \al)\not\equiv 0$, $s_0(\fhi_2, \al)\not\equiv 0$ be $2\pi$ periodic solutions of: $$\left\{ \begin{aligned} - \al_0\al_1 \ddot r_0= r_0^3 +3\langle s_0^2\rangle r_0 \\ -\al_0 \al_2 \ddot s_0= s_0^3 +3\langle r_0^2\rangle s_0 \end{aligned}\right.\, \qquad \langle f\rangle \equiv \frac{1}{2\pi}\int_0^{2\pi} d\tau f(\tau).$$ \end{Df} We prove - see Lemma \ref{pro1} and Remark \ref{oss1} - the existence of such solutions $r_0,s_0\in \B_\sig$.Moreover we prove that, for appropriate values of $\al_1,\al_2$, such solutions are non degenerate. \begin{Df} \label{d3} Given $\eps_0, \gam$ such that $0<\eps_0\ll 1 $ and $ \eps_0\ll \gam< \6$, let $\Ca_\gam\subset (0, \eps_0)\times (0, \eps_0)$ be the set of badly approximable pairs: $$\Ca_\gam:=\Big\{(A_1,A_2)\in(0, \eps_0)\times (0, \eps_0):\; | n_1 + A_2 n_2|> \frac{\gam}{|n_2|}\, , \quad | n_2 + A_1 n_1|> \frac{\gam}{|n_1|},\quad n_1,n_2 \in \Z\setminus \{0\}\big\}. $$ \end{Df} \begin{Pro}There exist positive numbers $\rho, \eps_0, C_1, C_2, C_3, \gam$ such that, for any $\eps\in \R$, and $\al\in\R^3$ satisfying the assumptions: $$\text{H:}\qquad(\eps\al_1, \eps\al_2)\in \Ca_\gam, \qquad\eps\in(0, \eps_0), \qquad |\al_1/\al_2-1|\leq \rho, \qquad C_1\leq \al_i\leq C_2, $$ Equation \ref{L1} admits a doubly periodic solution $ u(\fhi, \eps, \al)\in \B_\sig$,satisfying: $$| u(\fhi, \eps, \al)- r_0(\fhi_1, \al)-s_0(\fhi_2, \al)|_\sig \gam \sim O_\eps(1), $$ for all $n_1, n_2\neq 0$. Therefore, for $(\eps \al_1, \eps\al_2) \in\Ca_\gam$, $\al_0>C_1$ the operator $ L_\al$ restricted to the P subspace has bounded inverse: $$ | \Pi_p L_\al^{-1}[p]|_\sig\leq \frac{2|p|_\sig}{C_1\gam}, $$for all $p\in P$.\end{lem} \begin{proof} By definition $x\in (0,\eps_0)$ is badly approximable if: $$ | n_1 + x n_2|> \frac{\gam}{|n_2|}\,,\qquad\forall (n_1,n_2)\in\Z^2\quad n_2\neq 0.$$ We denote this set by $\Ba_\gam(\eps_0)$ or $\Ba$ for short. $\Ba$ is known to be uncountable, zero measure and accumulating to zero see \cite{bcp} for a proof. Therefore $(\eps\al_1, \eps\al_2)\in \Ca_\gam \subset (0, \eps_0)\times(0, \eps_0) $ if both $\eps\al_1, \eps\al_2\in\Ba$ with $\gam=O_\eps(1)$. Notice that Lemma \ref{le0} is trivially satisfied if $$-n_1\neq [ \eps \al_2 n_2]\;\text{and}\; -n_2 \neq[ \eps\al_1 n_1].$$ Now, if $\eps_0$ is small enough, when $-n_1= [ \eps \al_2 n_2]$ then $|n_1|<\1|n_2|$ so that $$| n_2 + \eps\al_1 n_1|> \1 |n_2|.$$ This implies that $$| n_1 + \eps \al_2 n_2|| n_2 + \eps \al_1 n_1|> \frac{\gam|n_2|}{2|n_2|}>\1 \gam .$$ The same (exchanging $n_1$ with $n_2$) holds if $-n_2 =[ \eps \al_1 n_1]$. The eigenvalues of $L_\al$ restricted to P are $\al_0 D_n$ so $$|L_\al^{-1}[p]|^2_\sig\equiv \sum_{n\in\Z^2\atop n_1, n_2\neq 0}\frac{|p_n|^2 |n|^2e^{2 i\sig |n|}}{|\al_0 D_n|^2} \leq \frac{4}{C^2_1\gam^2 }|p|^2_\sig, $$ for all $p\in P$. \end{proof} Now we pass to item 3. and solve the p-equation keeping $q(\fhi)$ as a parameter. \begin{lem}\label{le7} Given $q(\fhi)\in Q$ such that $$ \frac{\eps |q|^2_\sig}{C_1\gam }\ll 1\, , $$ the p equations: $$ p(\fhi)=\eps L_\al^{-1}\Pi_P [(q+p)^3+\eps f(q+p)], $$ can be solved with $q$ as a parameter. The solution $p(q)$ respects the bounds: $$ |p(q)|_\sig\leq C_3 \frac{\eps|q|^3_\sig}{C_1\gam}, $$ for $C_3$ an appropriate order one constant. \end{lem} \begin{proof} By Lemma \ref{le0} $L_\al^{-1}$ is bounded on P, moreover the operator: $$ p \rightarrow \Pi_P[(q+p)^3+\eps f(q+p)]$$ is well defined and regular on $\B_\sig$; we can apply the standard contraction mapping theorem. We define the sequence: $$ p^{(h)}=\eps L_\al^{-1}\Pi_P[(q+p^{(h-1)})^3+ \eps f(q+p^{(h-1)})]\, , \qquad p^{(0)}=0$$ The sequence defined above is a contraction if $$ \frac{3\eps |q|_\sig^2}{C_1\gam}\ll 1$$ in such case we have\footnote{recall that $|q(\fhi)p(\fhi)|_\sig\leq |q|_\sig|p|_\sig$, by the Hilbert algebra property}: $$ |p^{(h)}- p^{(h-1)}|_\sig \leq C_3\frac{|q|_\sig^2\eps}{\al_0}( |p^{(h-1)}-p^{(h-2)}|_\sig)\leq (C_3\frac{|q|_\sig^2\eps}{C_1\gam})^h|q|_\sig. $$ \end{proof} We now pass to point 4. of our scheme and solve the Equations \ref{PQ1}q), with $p=p(q, \eps)$ computed in the preceding Lemma: \Beq{Q2}{\left\{\begin{aligned} \al_0\al_1 \ddot r+ r^3 +3\langle s^2\rangle r +\Pi_{\fhi_1}[(q+p(\eps, q))^3-q^3)+ \eps f(u, \eps)]=0&\quad\qquad q_1)\\ \al_0 \al_2 \ddot s+ s^3 +3\langle r^2\rangle s +\Pi_{\fhi_2}[(q+p(\eps, q))^3-q^3)+ \eps f(u, \eps)]=0. &\quad\qquad q_2)\end{aligned}\right. } For compactness of notation let us write this equations as $F(r, s, \eps)=0$, where $F$ is a well defined and regular operator. \end{subsection}\begin{subsection}{The q-equations} The q-equations \ref{Q2} are non trivial at $p(q, \eps)\equiv 0$: \Beq{PQ0}{\left\{ \begin{aligned} - \al_0\al_1 \ddot r= r^3 +3\langle s^2\rangle r \\ -\al_0 \al_2 \ddot s= s^3 +3\langle r^2\rangle s \end{aligned}\right.\, .} It is convenient to rescale the Equations \ref{PQ0} setting: $$ r(\fhi_1)= \sqrt{ \al_0\al_1} x(\fhi_1)\, , \qquad s(\fhi_2)= \sqrt{ \al_0\al_2} y(\fhi_2);\qquad\lam= \frac{\al_1}{\al_2}$$ we obtain the equations: \Beq{eqq}{\left\{\begin{aligned}-\ddot x= & x^3 +3\lam^{-1}\langle y^2\rangle x\\ -\ddot y= & y^3 +3\lam \langle x^2\rangle y, \end{aligned}\right.} \begin{lem}\label{pro1} There exists an appropriate $\rho$ such that for all $\lam$ in $$|1-\lam|\leq \rho$$ the Equation \ref{eqq} has a {\bf non-degenerate} solution: $(x(t, \lam), y(t, \lam))$ such that both $x, y$ are not identically zero. \end{lem} The proof of this Lemma is in the Appendix. Let us define $$r_0(\fhi_1, \al)= \sqrt{ \al_0\al_1} x(\fhi_1, \frac{\al_1}{\al_2})\, , \qquad s_0(\fhi_2, \al)= \sqrt{\al_0\al_2} y(\fhi_2, \frac{\al_1}{\al_2}), $$ where $x(t, \lam), y(t, \lam)$ are given by Lemma \ref{pro1}. By definition $r_0, s_0$ solve the q-equations for $p=0$. \begin{lem}\label{le8} Equations \ref{PQ1} $(q_1-q_2)$ for $p=p(q, \eps)$ have a solution $q(\fhi, \al, \eps) \in Q$, which is $\eps$ close to $$ q_0= r_0(\fhi_1)+s_0(\fhi_2).$$ \end{lem} \begin{proof} By Lemma \ref{pro1} the solutions of \ref{PQ1} for $\eps=0$ (and therefore $p(q, \eps)=0$) are non-degenerate, i.e. equation \ref{eqq}, linearized in $ (x(t, \lam), y(t, \lam))$ does not have solutions in $\B_\sig$ and the linear operator: \Beq{Q3}{O^{-1}[X, Y]:=\left\{\begin{aligned} \ddot X+ & 3 x^2(t, \lam)X+3\lam^{-1}\langle y^2(t, \lam)\rangle X+6\lam^{-1}\langle y(t, \lam) Y\rangle x(t, \lam)\\\ddot Y+ & 3y^2(t, \lam) Y +3\lam\langle x^2(t, \lam)\rangle Y+6\lam\langle x(t, \lam) X\rangle y(t, \lam)\end{aligned}\right. } is $C^1$ and invertible. In the Appendix we have studied the operator $O$ and given bounds on its norm: $$|O[f, g]|_\sig\leq |x(t, \lam=1)|_\sig^8 \max(|f|_\sig , |g|_\sig) $$ By the Implicit Function Theorem \ref{Q3} implies that for $\eps$ small enough we can solve the q-equations \ref{Q2} obtaining $q=q(\fhi, \eps)=r(\fhi_1, \eps)+s(\fhi_2, \eps)$. \end{proof} We can now prove the existence of solutions for Equation \ref{L1}, let us restate the proposition: \begin{pro}\label{te1}There exist positive numbers $\rho, \eps_0, C_1, C_2, C_3, \gam$ such that, for any $\eps\in \R$, and $\al\in\R^3$ satisfying the assumptions: $$\text{H:}\qquad(\eps\al_1, \eps\al_2)\in \Ca_\gam, \qquad\eps\in(0, \eps_0), \qquad |\al_1/\al_2-1|\leq \rho, \qquad C_1\leq \al_0\leq C_2, $$ Equation \ref{L1} admits a doubly periodic solution $ u(\fhi, \eps, \al)\in \B_\sig$, with the property: $$| u(\fhi, \eps, \al)- r_0(\fhi_1, \al)-s_0(\fhi_2, \al)|_\sig 0$ such that, for $|\lam-1|\leq \rho_1$, equation \ref{e3} has a solution\footnote{Indeed, as stated in \cite{ls}, equation\ref{e3} has solutions $(x(\lam, t),y(\lam, t))$ both non-identically zero, provided that $\lam\in(\pi/6,6/\pi)$}: \Beq{X1}{x(\lam, t)= a(m_1(\lam), C_1(\lam), t)\, , \qquad y(\lam, t)=a(m_2(\lam), C_2(\lam), t), } both not identically zero. Moreover, by construction, the solution is $2\pi$ periodic odd and therefore in $\B_\sig$ for some appropriate $\sig$. We now prove the non-degeneracy. \begin{lem}\label{le2} Given $F(t), G(t)\in \B_\sig$, the equation: $$\begin{aligned}-\ddot X= & 3x^2 X +3\lam^{-1}\langle y^2\rangle X+6\lam^{-1}\langle y Y\rangle x+ F(t)\\-\ddot Y= & 3y^2 Y +3\lam\langle x^2\rangle Y+6\lam\langle x X\rangle y+G(t) , \end{aligned}$$ where $x, y$ are defined in \ref{X1}, has a unique solution $X, Y\in \B_\sig$, for $ |\lam-1|\leq \rho$. \end{lem} \begin{proof} We first consider the operator $$X\rightarrow L[m, X]= \ddot X+ 3a^2(m) X +C(m) X $$ where $a(m)= a(m, C(m))$ is $2\pi$ periodic; notice that $$a(m_1(\lam))=x, a(m_2(\lam)=y, $$ by definition. This operator has been studied and inverted in \cite{gmp}. We have: $$X =- L^{-1}[m_1, (6\lam^{-1}\langle a(m_2) Y\rangle a(m_1)+ F(t))]\, , \quad Y=- L^{-1}[m_2, (6\lam \langle a(m_1) X\rangle a(m_2)+ G(t))], $$ which yields a linear equation for $\langle xX\rangle , \langle yY\rangle $: $$\left|\begin{array}{cc} 1& 6\lam \langle a(m_1) L^{-1}[m_1, a(m_1)]\rangle \\ 6\lam^{-1}\langle a(m_2) L^{-1}[m_2, a(m_2)]\rangle &1\end{array}\right|\left|\begin{array}{c} \langle a(m_1) X\rangle \\\langle a(m_2) Y\rangle \end{array}\right|\equiv $$ $$ M\left|\begin{array}{c} \langle x X\rangle \\\langle y Y\rangle \end{array}\right|= -\left|\begin{array}{c} \langle a(m_1) L^{-1}[m_1, F]\rangle \\\langle a(m_2) L^{-1}[m_2, G]\rangle \end{array}\right|. $$ This determines uniquely $X, Y$ provided that the matrix $M$ in the left hand side is invertible. \vskip5pt \np In \cite{gmp} the constant $6\langle a({\bf m}) L^{-1}[{\bf m}, a({\bf m})]\rangle \sim 2.5 $ is computed explicitly (computer assisted calculations); \np as $\langle a(m) L^{-1}[m, a(m)]\rangle $ is smooth and $m_1(\lam), m_2(\lam)\sim {\bf m}$ for $\lam\sim 1$ there is a $\rho$ neighborhood of $\lam=1$ where matrix is invertible. We have defined an invertible linear operator on Q : \Beq{Oo}{ O[F(\fhi_1), G(\fhi_2)]=(L^{-1}[m_1, F(\fhi_1)+ 6\lam^{-1}\langle y Y\rangle x(\fhi_1)], (L^{-1}[m_2, G(\fhi_2)+ 6\lam^{-1}\langle x X\rangle y(\fhi_2)] ), } where naturally $$M\left| \begin{array}{c} \langle x X\rangle \\\langle y Y\rangle \end{array}\right|= -M^{-1}\left|\begin{array}{c} \langle a(m_1) L^{-1}[m_1, F]\rangle \\\langle a(m_2) L^{-1}[m_2, G]\rangle \end{array}\right|. $$ Finally let us give a bound on the $\sig$ analytic norm of the operator $O$, we can use the bound on $L^{-1}$ provided in \cite{gmp} $$|L^{-1}[m]|_\sig\leq |a(m)|_\sig^2 \quad\text{which implies}\quad |O|_\sig\leq C \max(|a({\bf m})|^8), $$ for $|\lam-1|\leq \rho'$. \end{proof} \begin{oss}\label{oss1} It should be noticed that given $x(\lam, t), y(\lam, t)$ $2\pi$ solutions of \ref{xy}, the functions $$x_j(\lam, t)= j x(\lam, j t), \qquad y_h(\lam, t)= h x(\lam, h t), $$ are solutions of $$\begin{aligned}-\ddot x_j= & x_j^3 +3\bar \lam^{-1}\langle y_h^2\rangle x_j\\ -\ddot y_h= & y_h^3 +3\bar \lam \langle x_j^2\rangle y_h, \end{aligned} $$ with $\bar \lam= \frac{j^2}{h^2}\lam $, for all natural values of $j, h$ such solutions are still $2\pi$ periodic. This remark permits us to construct more solutions of \ref{xy} namely for all couple of naturals $j, h$ such that $ |\lam \frac{h^2}{j^2}-1|\leq \rho$, $$j x(\lam \frac{h^2}{j^2}, j t), h y(\lam \frac{h^2}{j^2}, ht)$$ are solutions of \ref{xy} and they are clearly different from $x(\lam, t)$, $y(\lam, t)$ studied in the proof of Lemma \ref{le2} which have minimal period $2\pi$. This implies that if $\lam$ is not close to one one can still find solutions of \ref{xy} but they will have minimal periods $2\pi/j$ and $2\pi/h$. where $j, h$ are the smallest co-prime naturals such that $$|\frac{j^2}{h^2}\lam-1|\leq \rho.$$ \end{oss} \end{section} \end{appendix} \begin{thebibliography}{30} \bibitem[BCP]{bcp}{ D. Bambusi, S. Cacciatori, S. Paleari, {\it Normal form and exponential stability for some nonlinear string equations}, Z. Angew. Math. Phys. {\bf 52} (2001), no. 6, 1033--1052. } \bibitem[BP]{bp} D. Bambusi, S. Paleari, {\it Families of periodic solutions of resonant PDE's}, J. Nonlinear Sci. {\bf 11} (2001), no. 1, 69--87 \bibitem[BB1]{bb} M. Berti, P. Bolle, {\it Periodic solutions of nonlinear wave equations with general nonlinearities}, Comm. Math. Phys. {\bf 243} (2003), no. 2, 315--328. \bibitem[BB2]{bb2} M. Berti, P. Bolle, {\it Multiplicity of periodic solutions of nonlinear wave equations}, Nonlinear Anal., no. 56 (2004), 1011-1046 \bibitem[BB3]{bb3} M. Berti, P. Bolle, {\it Cantor families of periodic solutions for completely resonant non linear wave equations}, preprint \bibitem[B1]{b1}J. Bourgain, {\it Construction of quasi-periodic solutions for Hamiltonian perturbations of linear equations and applications to nonlinear PDE}, Internat. Math. Res. Notices {\bf 1994}, no. 11, 475ff., approx. 21 pp. (electronic). \bibitem[B2]{b2}J. Bourgain, {\it Construction of periodic solutions of nonlinear wave equations in higher dimension}, Geom. Funct. Anal. {\bf 5} (1995), 629--639. \bibitem[B3]{b3}J. Bourgain, {\it Quasi-periodic solutions of Hamiltonian perturbations of 2D linear Schr\"o\-din\-ger equations}, Ann. of Math. (2) {\bf 148} (1998), no. 2, 363--439. \bibitem[CY]{cy} L.Chierchia , J. You, {\it KAM tori for 1D nonlinear wave equations with periodic boundary conditions, } Comm. Math. Phys. 211 (2) 2000 pp. 497-525 \bibitem[C]{c}W. Craig, {\it Probl\`emes de petits diviseurs dans les \'equations aux d\'eriv\'ees partielles}, Pano\-ra\-mas and Syntheses 9. Soci\'et\'e Math\'ematique de France, Paris, 2000. \bibitem[CW]{cw} W. Craig, E. Wayne, {\it Newton's method and periodic solutions of nonlinear wave equations}, Comm. Pure Appl. Math. {\bf 46} (1993), 1409--1498. \bibitem[GMP]{gmp}G. Gentile, V. Mastropietro, M. Procesi {Periodic solutions of completely resonant nonlinear wave equations} to appear in Comm. Math. Phys \bibitem[GrR]{gr} I.S. Gradshteyn, I.M. Ryzhik, {\it Table of integrals, series, and products}, Sixth edition, Academic Press, Inc., San Diego, 2000. \bibitem[K1]{k1}{ S.B. Kuksin, {\it Nearly integrable infinite-dimensional Hamiltonian systems}, Lecture Notes in Mathematics 1556, Springer, Berlin, 1994. } % \bibitem[K2]{k2}{ S.B. Kuksin, {\it Fifteen years of KAM for PDE}, Preprint, 2003. } % \bibitem[KP]{kp}{ S.B. Kuksin, J.P\"oschel, {Invariant Cantor manifolds of quasi-periodic oscillations for a nonlinear Schr\"odinger equation}, Ann. of Math. (2) {\bf 143} (1996), no. 1, 149--179. } % \bibitem[LS]{ls}{ B.V. Lidski\u\i, E.I. Shul$^{\prime}$man, {\it Periodic solutions of the equation $u_{tt}-u_{xx}+u^{3}=0$}, Funct. Anal. Appl. {\bf 22} (1988), no. 4, 332--333. } % \bibitem[P]{p}{ J. P\"oschel, {\it Quasi-periodic solutions for a nonlinear wave equation}, Comment. Math. Helv. {\bf 71} (1996), no. 2, 269--296. } % \bibitem[Pr]{pr}M. Procesi, {\it Quasi periodic solutions for a completely resonant nonlinear wave equation.} \bibitem[Wa]{32}{ C.E. Wayne, {\it Periodic and quasi-periodic solutions of nonlinear wave equations via KAM theory}, Comm. Math. Phys. {\bf 127} (1990), no. 3, 479--528. } % \bibitem[We]{33}{ A. Weinstein, {\it Normal modes for nonlinear Hamiltonian systems}, Invent. Math. {\bf 20} (1973), 47--57. } \end{thebibliography} \vskip10 pt {\small Functional Analysis sector ISAS - International School for Advanced Studies, Via Beirut 2-4, 34014 Trieste (Italy)} \end{document} ---------------0407230911259--