\documentclass[reqno]{amsart} \usepackage{hyperref} \usepackage{graphicx} \AtBeginDocument{{\noindent\small \emph{Electronic Journal of Differential Equations}, Vol. 2015 (2015), No. 275, pp. 1--10.\newline ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu \newline ftp ejde.math.txstate.edu} \thanks{\copyright 2015 Texas State University.} \vspace{9mm}} \begin{document} \title[\hfilneg EJDE-2015/275\hfil Oscillations with discontinuous energy] {Oscillations with one degree of freedom and discontinuous energy} \author[M. Frasson, M. Gadotti, S. Nicola, P. T\'aboas \hfil EJDE-2015/275\hfilneg] {Miguel V. S. Frasson, Marta C. Gadotti, \\ Selma H. J. Nicola, Pl\'acido Z. T\'aboas} \address{Miguel V. S. Frasson \newline Departamento de Matem\'atica Aplicada e Estat\'istica, ICMC-Universidade de S\~ao Paulo, Avenida Trabalhador S\~ao-carlense 400, 13566-590 S\~ao Carlos SP, Brazil} \email{frasson@icmc.usp.br} \address{Marta C. Gadotti \newline Departamento de Matem\'atica, IGCE -- Universidade Estadual Paulista, Avenida 24A 1515, 13506-700 Rio Claro SP, Brazil} \email{martacg@rc.unesp.br} \address{Selma H. J. Nicola \newline Departamento de Matem\'atica,\quad Universidade Federal de S\~ao Carlos, Rodovia Washington Luis, km 235 Norte, 13565-905 S\~ao Carlos SP, Brazil} \email{selmaj@dm.ufscar.br} \address{Pl\'acido Z. T\'aboas \newline Departamento de Matem\'atica Aplicada e Estat\'istica, ICMC-Universidade de S\~ao Paulo, Avenida Trabalhador S\~ao-carlense 400, 13566-590 S\~ao Carlos SP, Brazil} \email{pztaboas@icmc.usp.br} \thanks{Submitted September 30, 2015. Published October 23, 2015.} \subjclass[2010]{34C25, 34D20, 37G15} \keywords{Periodic solutions; discontinuous energy; orbital stability; bifurcation} \begin{abstract} In 1995 for a linear oscillator, Myshkis imposed a constant impulse to the velocity, each moment the energy reaches a certain level. The main feature of the resulting system is that it defines a nonlinear discontinuous semigroup. In this note we study the orbital stability of a one-parameter family of periodic solutions and state the existence of a period-doubling bifurcation of such solutions. \end{abstract} \maketitle \numberwithin{equation}{section} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{remark}[theorem]{Remark} \newtheorem{definition}[theorem]{Definition} \allowdisplaybreaks \section{Introduction} The solutions of the damped linear oscillator \begin{equation}\label{eq:oscill-plain} \ddot x+2\alpha\dot x+\omega^2x=0,\quad \omega>\alpha>0, \end{equation} are supposed to undergo a fixed instantaneous increase of velocity whenever they reach a certain level $E_0>0$ of energy. More precisely, the following condition is imposed \[ \frac12(\dot x^2(t)+\omega^2x^2(t))=E_0\Rightarrow \lim_{s\to{t+}}\dot x(s)=\dot x(t)+\sigma,\quad\sigma>0. \] This note concerns the resulting discontinuous dynamical system in the plane $x\dot x$. Motivated by a pioneering work by Myshkis \cite{myshkis-1}, we obtain the existence of orbitally asymptotically stable \emph{simple} periodic solutions, i.e., solutions which have exactly one impulse in the period. We accomplish a period-doubling bifurcation for such solutions. The main feature of the problem is to be autonomous; that is, besides the involved equation being autonomous, the moments of impulses are not previously known. Therefore the solution operator of the whole system defines a discontinuous semigroup. Specific references to the subject are Myshkis \cite{myshkis-3} and Samoilenko-Perestyuk \cite{samoilenko-perestyuk}. For a wider class of related poblems see \cite{gadotti-taboas, gadotti-taboas-2, gyori-3, halanay-wexler, ladeira-nicola-taboas, laks-bainov-simeonov, myshkis, myshkis-2, myshkis-3, pandit-deo} and references therein. Section 2 aims to build a context for the problem. In Section 3 we state elementary properties of positive simple periodic solutions. In Section 4 we prove the existence of orbitally unstable positive simple periodic solutions with small amplitude and of orbitally asymptotically stable with large amplitudes. Finally, in Section 5 we give a sufficient condition for a period-doubling bifurcation of such solutions. \section{Object of study and basic facts} By the time scaling $\tau=\omega t$ and the change of variables $\xi(\tau)=(\omega/\sqrt{2E_0})x(\tau/\omega)$ Equation \eqref{eq:oscill-plain} is written as $\xi''+2a\xi'+\xi=0$, where $'=d/d\tau$, $a=\alpha/\omega\in(0,1)$ and the locus of level $E_0$ of energy is taken to the circle $S:\xi^2+{\xi'}^2=1$ in the plane $\xi\xi'$. Retrieving the original notation and formulating the problem in the $x\dot x$ plane we obtain \begin{equation}\label{eq:system} \begin{gathered} \dot x=y,\\ \dot y=-x-2ay \end{gathered} \end{equation} with the impulsive condition \begin{equation} (x(t),y(t))\in S\,\Rightarrow\, (x(t+),y(t+))=(x(t),y(t)+v).\label{eq:jump} \end{equation} Solutions of \eqref{eq:system} will be denoted by $z$ and $z(\cdot;t_0,z_0)$, if $z(t_0;t_0,z_0)=z_0$, or briefly $z(\cdot;z_0)=z(\cdot;0,z_0)$. As the eigenvalues of \eqref{eq:system} are $-a\pm\delta i$, with $\delta=\sqrt{1-a^2}>0$, the origin is a stable focus and the energy decreases strictly along nontrivial solutions, since \begin{equation}\label{eq:dissipacao} \dot E(z(t))= -2a(y(t))^2, \quad t\in\mathbb{R}. \end{equation} Let $a=\sin b$, $b\in(0,\pi/2)$, so that $\delta=\cos b$. If $\bar z(\cdot)=z(\cdot;(0,-1))$, \begin{equation}\label{eq:spannersolution} \bar z(t) = -\delta^{-1} e^{-at} \bigl(\sin \delta t, \cos (\delta t +b)\bigr), \quad t\in\mathbb{R}. \end{equation} As $\bar z(\cdot)$ crosses the $y$ axis at $(0,-\sigma)=(0,-e^{-2a\pi/\delta})$, completing a lap around the origin, if $\gamma= \bar z(\mathbb{R})$, the family $\{\mu\gamma\}_{\mu\in(\sigma,1]}$ describes all nontrivial orbits of \eqref{eq:system}. That is, the general nontrivial solution is \begin{equation*}\label{eq:generalsolution} z(\cdot)=\mu\bar z(\cdot+\tau),\quad \tau\in\mathbb{R},\quad\sigma<\mu\leq 1. \end{equation*} \begin{definition}\label{solsia} \rm A solution of \eqref{eq:system}, \eqref{eq:jump} through $b_0\in\mathbb{R}^2$ at $t=t_0$ is a function $\phi:[t_0,\infty)\to\mathbb{R}^2$ such that $\phi(t_0)=b_0$ and \begin{enumerate} \item $\phi(t-)=\phi(t)$, for all $t\in(t_0,\infty)$; \item $\phi\in C^1$ and satisfies \eqref{eq:system} in $(t,t+\epsilon_t)$, for all $t\in[t_0,\infty)$ and some $\epsilon_t>0$. \item $\phi$ is continuous in $t$ if $\phi(t)\in\mathbb{R}^2\setminus S$ and $\phi(t+)=\phi(t)+(0,v)$ if $\phi(t)\in S$. \end{enumerate} \end{definition} \begin{remark}\rm \begin{enumerate} \item $\phi$ is denoted by $\phi(\cdot;t_0,b_0)$ or $\phi(\cdot;b_0)$ if $t_0=0$. \item A function $\psi:(\tau,\infty)\to\mathbb{R}^2$ is solution of \eqref{eq:system}, \eqref{eq:jump} in $(\tau,\infty)$ if $\psi\bigl|_{[\,t_0,\infty)}= \phi(\cdot;t_0,\psi(t_0))\bigr.$, for any $t_0\in(\tau,\infty)$. \item The solution $\phi(\cdot;t_0,b_0)$ is unique, but in general there is no uniqueness for backward continuations. If $|b_0|\geq 1$, $\phi(\cdot;t_0,b_0)$ has a continuation to $(-\infty,\infty)$. If $|b_0|<1$, in general a maximal interval of existence to the left is bounded below. \end{enumerate} \end{remark} \section{Positive simple solutions} \label{sec: pss} For the dynamics of \eqref{eq:system}, \eqref{eq:jump} the only relevant solutions are $\phi(\cdot;b)$ with $|b|\geq 1$, as they are the only that eventually undergo impulses. There is no loss of generality in taking $|b|=1$ and we do so. We denote by $\mathfrak C$ the class of such solutions. \begin{definition} \rm Let $\phi(\cdot;b)$, $|b|=1$, be a periodic solution of \eqref{eq:system}, \eqref{eq:jump} with minimal period $\omega>0$. The point $\phi(0;b)$ is called \emph{vertex} of $\gamma=\phi(\cdot;b)$. We say that $\phi(0;b)$ is simple if it has a unique impulse in $[0,\omega)$. If $\phi(\cdot;b)=(x(\cdot),y(\cdot))$, it is positive when $x(t)>0$ for all $t$. \end{definition} We close this section by setting some standing notations. A number $\beta$, identified to any $\beta'\equiv\beta\mod2\pi$, indicates a point $(\cos\beta,\sin\beta)\in S$ or its arc length coordinate in $S$. The context will clarify the meaning in each case. For $\beta\in S$ we denote $\phi_\beta=\phi(\cdot;\beta)$ and, if $|\beta+(0,v)|>1$, we set $t_1=t_1(\beta)>0$ such that $\phi_\beta(t_1)\in S$ and $\phi_\beta(t)\notin S$ for $01\}$, we define the return map $\Phi_v:D\to S$ by $\Phi_v(\beta)=\phi_\beta(t_1(\beta))$ for all $\beta\in D$. \end{definition} Clearly, if $\beta^*\in D$ is a fixed point of $\Phi_v$, $\phi_{\beta^*}$ is a simple periodic solution whose period is $t_1(\beta^*)$ and $\beta^*$ is the vertex of the simple cycle $\phi_{\beta^*}(\mathbb{R})$. If $\beta^*$ is an attractor fixed point, $\phi_{\beta^*}$ is orbitally asymptotically stable and, if it is repelling, $\phi_{\beta^*}$ is orbitally unstable. Here the orbital stability must be in the sense of conditional stability relative to the class $\mathfrak C$, see \cite{lefschetz}, since if $\phi=\phi(\cdot;b)$, $|b|=1$, there are points $b'$ inside $S$ arbitrarily close to $b$ and therefore $\phi(t;b')\to(0,0)$, as $t\to\infty$. If $\beta\in S$, let $s_\beta$ be the vertical line $s_\beta\!: x=\cos\beta$ and $t_\beta>0$ such that $z(-t_\beta;\beta)= (\cos\beta,y_\beta)\in s_\beta$ and $z(t;\beta)\notin s_\beta$ for $-t_\beta0$. We denote by $\alpha=\alpha_\beta$ the polar angle of $z(-t_\beta;\beta)$, according to Figure \ref{fig:positsimpcycle}. \begin{figure}[htb] \begin{center} \includegraphics[width=0.5\textwidth]{fig1} % art-fgnt-figure0 \end{center} \caption{Positive simple cycle.} \label{fig:positsimpcycle} \end{figure} \begin{remark} \rm For any $v\in(0,e^{a\pi/\delta}+1)$, there exists exactly one positive simple cycle of \eqref{eq:system}, \eqref{eq:jump} since $\beta\in(-\pi/2,0)\mapsto v_\beta\in (0,\,e^{a\pi/\delta}+1)$ is a continuous bijection. \end{remark} \section{Orbital stability} Now we show that, for some $\zeta>0$, the solution $\phi_\beta$ of \eqref{eq:system}, \eqref{eq:jump} is orbitally unstable if $\beta\in(-\zeta,0)$ and orbitally asymptotically stable if $\beta\in(-\pi/2,-\pi/2+\zeta)$. \begin{lemma}\label{lem:technicallemma} $v_\beta = -2\beta + o(\beta)$ as $\beta\to0-$. \end{lemma} \begin{proof} Let $\beta\in(-\pi/2,0)$. System \eqref{eq:system} in polar coordinates, \begin{gather*} \dot r=-(2a\sin^2\theta)r,\\ \dot\theta=-(1+a\sin2\theta), \end{gather*} yields \begin{equation}\label{eq:cp} r'=\bigl(2a\sin^2\theta/(1+a\sin2\theta)\bigr)r,\quad( '=d/d\theta). \end{equation} and a parametrization of $\phi_\beta$ is \begin{equation}\label{eq:A(theta)} r_\beta(\theta)=e^{A_\beta(\theta)} =\exp\Big[2a\int_\beta^\theta\frac{\sin^2s} {1+a\sin2s}\,ds\Big],\quad \theta\in\mathbb{R}. \end{equation} As the integrand in \eqref{eq:A(theta)} will be a regular participant, we introduce the notation \[ q_a(s)=\frac{\sin^2s} {1+a\sin2s}. \] For any small $\epsilon>0$ such that $\alpha=-(1+\epsilon)\beta<\pi/2$, the inequality \[ A_\beta(\theta)\leq -\frac{2a(2+\epsilon)(1+\epsilon)^2}{1-a}\beta^3,\quad \theta\in[\beta,-(1+\epsilon)\beta], \] yields \[ r_\beta(-(1+\epsilon)\beta)=e^{A_\beta(-(1+\epsilon)\beta)}=1+O(\beta^3)\quad\text{as } \beta\to{0-}. \] \begin{figure}[htb] \begin{center} \includegraphics[width=0.7\textwidth]{fig2} % art-fgnt-figure1 \end{center} \caption{$v_\beta=-2\beta+o(\beta)$ as $\beta\to0-$.}\label{fig:orderofv_beta} \end{figure} If $r^\epsilon=|p_\epsilon|$, $p_\epsilon$ being the intersection of the half lines $s_1:\theta=-(1+\epsilon)\beta$ and $s_2:r(\theta)\cos\theta=\cos\beta$, $\theta\in(0,\pi/2)$, the similarity of the triangles $mnO$ and $p_\epsilon qO$ seen in Figure \ref{fig:orderofv_beta} yields \[ r^\epsilon=\frac{\cos\beta}{\cos(1+\epsilon)\beta} =1+ \frac{(2+\epsilon)\epsilon}{2!}\beta^2+O(\beta^4) \quad\text{as }\beta\to{0-}. \] For $|\beta|$ small enough, the estimates above imply $r_\beta(-(1+\epsilon)\beta)0$ such that if $\beta\in(-\zeta,0)$, the simple periodic solution $\phi_\beta$ of \eqref{eq:system}, \eqref{eq:jump} is orbitally unstable and if $\beta\in(-\pi/2,-\pi/2+\zeta)$, $\phi_\beta$ is orbitally asymptotically stable. \end{theorem} \begin{proof} Let $\beta\in(-\pi/2,0)$ and $\epsilon_1\ne0$ so that $\beta+\epsilon_1=\beta_1\in(-\pi/2,0)$. We take $|\epsilon_1|$ smaller if necessary to assure the existence of $\Phi_{v_\beta}(\beta_1)=\beta+\epsilon_2\in(-\pi/2,0)$, as it is seen in Figure \ref{fig:epsilon xi} for the case $\epsilon_1<0$. \begin{figure}[htb] \begin{center} \includegraphics[width=0.6\textwidth]{fig3} % art-fgnt-figure2 \end{center} \caption{$\beta+\epsilon_2= \Phi_{v_\beta}(\beta+\epsilon_1)$.} \label{fig:epsilon xi} \end{figure} Firstly we notice that $\epsilon_1$ and $\sigma$ are related by the equation \[ \frac{v_\beta+\sin(\beta+\epsilon_1)}{\cos(\beta+\epsilon_1)}=\tan(\alpha+\sigma), \] therefore, the implicit function theorem about $(\epsilon_1,\sigma)=(0,0)$ yields \begin{equation}\label{eq:dsigmadepsilon} \sigma=\frac{v_\beta\sin\beta+1}{|b_\beta|^2}\,\epsilon_1+o(\epsilon_1), \end{equation} as $\epsilon_1\to0$. By \eqref{eq:A(theta)}, if $b_1=\beta_1+(0,v_\beta)$, $\epsilon_2$ must satisfy \[ |b_1|\exp\Big[2a\int_{\alpha+\sigma}^{\beta+\epsilon_2} q_a(s)\,ds\Big]=1. \] As $|b_1|=\sqrt{(v_\beta+\sin(\beta+\epsilon_1))^2+\cos^2(\beta+\epsilon_1)}$, we have \[ \bigl(v_\beta^2+2v_\beta\sin(\beta+\epsilon_1)+1\bigr) \exp\Big[4a\int_{\alpha+\sigma(\epsilon_1)}^{\beta+\epsilon_2} q_a(s)\,ds\Big]=1 \] and the implicit function theorem leads to \begin{equation}\label{eq:key} \epsilon_2=\frac1{q_a(\beta)|b_\beta|^2} \big[q_a(\alpha)(1+v_\beta\sin\beta)- \frac{v_\beta\cos\beta}{2a}\big] \epsilon_1 +o(\epsilon_1), \end{equation} as $\epsilon_1\to0$. Let \begin{equation}\label{eq:F(beta)} F(\beta)=\frac1{q_a(\beta)|b_\beta|^2} \big[q_a(\alpha)(1+v_\beta\sin\beta)- \frac{v_\beta\cos\beta}{2a}\big], \end{equation} so that $F(\beta)<0$ and \eqref{eq:key} is $\epsilon_2=F(\beta)\epsilon_1+o(\epsilon_1)$, as $\epsilon_1\to0$, for short. Since $\lim_{\beta\to-\pi/2}|b_\beta|= \lim_{\beta\to-\pi/2}-(1+v_\beta\sin\beta) =e^{a\pi/\delta}$, \begin{equation}\label{eq:Fbeta<1} |F(\beta)|\to e^{-a\pi/\delta}<1,\quad\text{ as }\beta\to-\pi/2. \end{equation} On the other hand, we have $|\sin\beta|<|\sin\alpha|0$, Eqs. \eqref{eq:Fbeta<1} and \eqref{eq:Fbetatoinf} imply that $|F(\beta)|<1$ if $\beta\in(-\pi/2,-\pi/2+\zeta)$ and $|F(\beta)|>1$ if $\beta\in(-\zeta,0)$. In other words, any $\beta\in (-\pi/2,-\pi/2+\zeta)$ is an attractor fixed point of the return map $\Phi_{v_\beta}$ and any $\beta\in(-\zeta,0)$ is a repelling fixed point of $\Phi_{v_\beta}$. \end{proof} \section{Period doubling bifurcation} \label{sec: per-doub} Solutions $\phi_\beta$ of \eqref{eq:system}, \eqref{eq:jump} change from stable to unstable when $\beta$ varies over $(-\pi/2,0)$ from left to the right. Therefore it is natural to expect a bifurcation in between. In this section we apply the theorem below \cite[Theorem 12.7]{devaney} to confirm that this indeed occurs at least for small dampings. \begin{theorem}[Period doubling bifurcation]\label{theo:devaney} Let $\{f_\lambda\}$ a one-parameter family of real functions and suppose that \begin{enumerate} \item $f_\lambda(0)=0$ for all $\lambda$ in an interval about $\lambda_0$; \item $f_{\lambda_0}'(0)=-1$; \item $\displaystyle \frac{\partial(f_\lambda^2)'}{\partial\lambda}\Big|_{\lambda=\lambda_0}(0)\neq 0$. \end{enumerate} Then there is an interval $I$ about $0$ and a function $p:I\to\mathbb{R}$ such that \[ f_{p(x)}(x)\neq x \quad\text{and}\quad f^2_{p(x)}(x)=x. \] \end{theorem} By the proof of Theorem \ref{theo:unst-st} there is a $\beta^*_a \in (-\pi/2,0)$, $00$, $\beta^*_a\notin(-\pi/4-\eta,-\pi/4+\eta)$. That is, $F'(\beta^*_a)\ne0$ for $a\in(0,1)$ sufficiently small. \end{proof} Figure \ref{fig:doubleperiod} shows a typical positive periodic orbit emanating from $\beta^*_a$. \subsection*{Final remarks} Smallness of $a$ is a request of our proof of Theorem \ref{theo:perdoub}, possibly this hypothesis can be weakened or even discarded. The larger is the coefficient $a \in (0,1)$, the larger is the region of stability in $(-\pi/2,0)$. In fact, by \eqref{eq:A(theta)}, $r_{-\pi/2}(\pi)=e^{a\pi/\delta} \to \infty$ as $a \to 1$. Therefore, for any fixed $\beta \in (-\pi/2,0)$, one has $|b_{\beta}| \to \infty$ as $a \to 1$, so that the number $\epsilon_{2}$ in \eqref{eq:key} satisfies $\epsilon_{2} \to 0$, as $a \to 1$. \begin{thebibliography}{10} \bibitem{devaney} Robert~L. Devaney; \emph{An introduction to chaotic dynamical systems}, second ed., Addison-Wesley Studies in Nonlinearity, Addison-Wesley Publishing Company, Advanced Book Program, Redwood City, CA, 1989. \MR{1046376 (91a:58114)} \bibitem{gadotti-taboas} M. C. Gadotti, P.Z. T\'aboas; \emph{Oscillations of planar impulsive delay differential equations}, Funkcialaj Ekvacioj \textbf{48} (2005), 33--47. \bibitem{gadotti-taboas-2} M. C. Gadotti, P.Z. T\'aboas; \emph{Periodic and backset solutions of differential delay systems with self-supporting condition}, J. Differential Equations \textbf{229} (2006), 138--153. \bibitem{gyori-3} I.~Gy{\"o}ri; \emph{Asymptotic periodicity in impulsive differential equations of retarded type with applications to compartmental models}, Proceedings of the First World Congress of Nonlinear Analysis - Berlin \textbf{IV} (1996), 1403--1413. \bibitem{halanay-wexler} A.~Halanay, D.~Wexler; \emph{Teoria calitativa a sistemelor cu impulsiori}, Acad. Rep. Soc. Romania, Bucaresti, 1968, In Romanian. \bibitem{ladeira-nicola-taboas} L. A. C. Ladeira, S. H. J. Nicola, P.Z. T{\'a}boas; \emph{Periodic solutions of an impulsive differential system with delay: an {$L^p$} approach}, Fields Inst. Comm. \textbf{31} (2002), 201--215. \bibitem{laks-bainov-simeonov} V.~Lakshmikantham, D. D. Bainov, P.S. Simeonov; \emph{Theory of impulsive differential equations}, World Scientific, Singapore, 1989. \bibitem{lefschetz} S.~Lefschetz; \emph{Differential equations; geometric theory}, second ed., Interscience, New York, 1959. \bibitem{myshkis} A. D. Myshkis; \emph{Autonomous differential equations with impulsive self-support and infinite delay}, Funct. Differential Equations \textbf{3} (1995), no.~2, 145--154. \bibitem{myshkis-1} A. D. Myshkis; \emph{On autonomous self-supporting impulsive-continuous system}, World Sc. Series in Applied An.; Dynam. Syst. Appl. \textbf{4} (1995), 541--548. \bibitem{myshkis-2} A. D. Myshkis; \emph{Vibrations of the string with energy dissipation and impulsive feedback support}, Nonlinear Anal., T. M. A. \textbf{26} (1996), no.~7, 1271--1278. \bibitem{myshkis-3} A. D. Myshkis; \emph{Auto-oscillations in continuous systems with impulsive self-support}, Resenhas IME-USP \textbf{3} (1997), no.~1, 93--106. \bibitem{pandit-deo} S. G. Pandit, S. G. Deo; \emph{Differential systems involving impulses}, L. N. in Math. 954, Springer, Berlin, 1982. \bibitem{samoilenko-perestyuk} A. M. Samoilenko, N. A. Perestyuk; \emph{Differential equations with impulsive action}, Vishcha Shkola, Kiev, 1987, In Russian. \end{thebibliography} \end{document}