1: \documentclass[fleqn,11pt]{article}
2: \usepackage{amsmath,amssymb}
3: \usepackage{graphicx}
4: % \usepackage{hyperref}
5:
6: \topmargin -2cm \oddsidemargin=0.0cm \rightmargin=3cm
7: \textwidth=15cm \textheight=23cm
8:
9: \newcommand{\vfi}{\varphi}
10: \newcommand{\eps}{\varepsilon}
11: \newcommand{\R}{\mathbb{R}}
12: \newcommand{\C}{\mathbb{C}}
13: \newcommand{\dee}{{\mathrm d}}
14: \newcommand{\imag}{{\mathrm i}}
15: \newcommand{\rem}[1]{}
16: \newcommand{\sign}{{\rm sign}}
17:
18: \newtheorem{teo}{Theorem}
19:
20: \begin{document}
21:
\title{(Vanishing) Twist in the Saddle-Centre and Period-Doubling Bifurcation}
\author{H.~R.~Dullin$^{1,2}$, A.~V.~Ivanov$^{1}$
\thanks{This research is funded by the EPSRC under contract GR/R44911/01.
22: Partial support by the European Research Training Network
23: {\it Mechanics and Symmetry in Europe\/} (MASIE), HPRN-CT-2000-00113,
24: is also gratefully acknowledged. AVI was also supported in part by INTAS grant 00-221,
25: RFBI grant 01-01-00335 and RFME grant E00-1-120.}\\
26: \\ $^{1}$ Department of Mathematical Sciences, % Loughborough University,
27: \\ Loughborough University, LE11 3TU, UK
28: \\ $^{2}$ Fachbereich 1, Physik, Universit\"at Bremen
29: \\ 28334 Bremen, Germany
30: \\ {\small H.R.Dullin@lboro.ac.uk}, {\small A.V.Ivanov@lboro.ac.uk}
}
\date{\today}
%\date{}
\maketitle
\begin{abstract}
31: The lowest order resonant bifurcations of a periodic orbit of a Hamiltonian system
32: with two degrees of freedom have frequency ratio $1:1$ (saddle-centre)
33: and $1:2$ (period-doubling). The twist, which is the derivative of the
rotation number with respect to the action, is studied near these bifurcations.
34: When the twist vanishes the nondegeneracy condition
35: of the (isoenergetic) KAM theorem is not satisfied,
36: with interesting consequences for the dynamics.
37: We show that near the saddle-centre bifurcation the twist always vanishes.
38: At this bifurcation a ``twistless'' torus is created, when the resonance is passed.
The twistless torus replaces the colliding periodic orbits in phase space.
39: We explicitly derive the position of the twistless torus depending on the resonance
40: parameter, and show that the shape of this curve is universal.
41: For the period doubling bifurcation the situation is different.
42: Here we show that the twist does not vanish in a neighborhood of the bifurcation.
43:
44: % \noindent
% PACS: 03.20.+i; 05.45.+b
\vspace*{1ex}
\noindent
Keywords: Twist Maps; Hamiltonian Systems; Saddle-Centre Bifurcation;
Period-doubling Bifurcation; KAM; Normal Forms; Elliptic Integrals
\end{abstract}
% \newpage
45:
46: \section{Introduction}
47:
48: The dynamics near a periodic orbit of a Hamiltonian system can be
49: studied in terms of a local Poincar\'e section transversal to the orbit.
50: In two degrees of freedom the first return map restricted to the
51: surface of constant energy is an area preserving map with a fixed (or periodic) point.
52: %
53: If the multipliers $\mu_i$, $i=1,2$ of the fixed point have modulus 1 (but are not equal to $\pm 1$)
54: the fixed point is called elliptic. Then $\mu_1 = \overline{\mu_2} \in \C$
55: and we can write $\mu = \exp(2\pi i\omega)$ with the rotation number
56: $0<\omega<1/2$ of the periodic orbit.
57: %
58: If the periodic orbit is elliptic and the rotation number is irrational
59: the map can (formally) be transformed to Birkhoff normal form
60: which in action-angle variables $(\vfi,I)$ reads
61: \begin{equation} \label{eqn:BNF}
62: (\vfi,I)\to (\vfi+2\pi\Omega(I),I)\,.
63: \end{equation}
64: The action $I$ is like the radial coordinate in polar coordinates,
65: hence the map in normal form maps circles to circles by rotating
66: them $\Omega(I)$ times.
The {\em rotation number} (or winding number) $\Omega(I)$
67: near the periodic orbit can be expanded as
68: \[
69: \Omega(I) = \omega + \tau_{0}I + \frac{1}{2}\tau_1 I^2+...
70: \]
The {\em twist} (or torsion) $\tau(I)$ is the derivative of the rotation number with
71: respect to the action,
\[
72: \tau(I) = \frac{{\rm d}\Omega}{{\rm d} I}(I)=\tau_{0}+\tau_{1}I+...
73: \]
When the rotation number is a strictly monotone function of the action
74: in some interval the map (\ref{eqn:BNF}) restricted to the corresponding
75: annulus is called a {\em monotone twist map}.
76: Moser's KAM theorem \cite{SM71} states that the invariant torus $I=I_0$
77: of (\ref{eqn:BNF}) persist under perturbation when its frequency $\Omega(I_0)$
78: is diophantine and its twist $\tau(I_0)$ does not vanish.
79: Arnold's KAM theorem \cite{Arnold78} is the same statement for flows where
80: the nonvanishing of the twist corresponds to the isoeneregetic nondegeneracy
81: conditions.
82: A well known corollary of the KAM theorem is the stability of an elliptic fixed point
83: in two degrees of freedom when $\Omega(0) = \omega \not = 1/3, 1/4$ and the twist at the
84: origin is non-vanishing, $\tau(0) = \tau_0 \not = 0$.
85:
86: When the twist vanishes the perturbed dynamics can be more complicated.
87: The stability of an elliptic point can be lost when its twist vanishes,
88: see \cite{DMS98b} for an example of an unstable elliptic point with $\omega = 1/5$.
89: The effects of vanishing twist away from the origin was first described by
90: Howard \cite{HowHoh84}, and the resulting effects have been observed in many
91: examples \cite{WVCP88,Sadovskii96}.
92: The probably most spectacular effect is the appearance of so-called meandering
93: curves \cite{HowHoh84,HowHum95,Simo98}.
94: The properties of non-twist maps also show interesting behaviour under
95: renormalisation \cite{CGM96} and recently it has been shown \cite{DL98} that an
96: extension of the KAM theorem can also be proved in this context.
97: In \cite{DMS98b,Moeckel90} it was finally shown that the vanishing of twist at the
98: fixed point generically occurs in a one parameter family when the rotation number
99: of a fixed point
100: passes through the interval $[1/4, 1/3]$. When the twist vanishes at the fixed
101: point a twistless torus is created in a {\em twistless bifurcation} \cite{DMS98b}.
102: After creation the twistless torus passes through resonances and in this way
103: non-twist maps generically appear in one parameter families of area preserving maps.
104: The truncated resonant Birkhoff normal form shows that this twistless torus
105: eventually collides with a saddle-centre bifurcation that gives rise to the
106: period 3 orbits that collide with the fixed point when $\omega = 1/3$ \cite{DMS98b}.
107: Such a connection between resonance and vanishing twist can also be found
108: in 4 dimensional symplectic maps \cite{DM02}.
109:
110: The techniques of \cite{DMS98b} can also be applied to the higher order resonances,
111: in particular for $\omega = 1/4$.
112: In this paper we study the two remaining generic bifurcations
113: at even stronger resonance $\omega = 0, 1/2$. By definition the corresponding
114: fixed point is not elliptic. The two cases will be denoted as the $1:1$ and
115: $1:2$ resonance, or as the saddle-centre and period doubling bifurcation, respectively.
116: The main result is that near the saddle-centre bifurcation the twist always vanishes,
117: while it does not vanish near the period doubling bifurcation.
118:
119: The method is based on the analysis of the resonant normal form, in which the
120: Poincar\'e map is approximated by the time $1$ map of a one
121: degree of freedom system, see e.g.~\cite{MH92}.
122: This normal form is an approximation, that is local near the bifurcation
123: in parameter space and in phase space. At first we will completely ignore this, and
124: just analyse the normal forms in the following two sections.
125: In section~\ref{sec:Universality} we will
126: address the problem of non-locality in phase space, and also comment on the
127: effect of higher order perturbations on the twistless tori.
128: We will talk of invariant tori even though the invariant curves $H(u,v)=h$ of
129: the normal form may not be compact.
130: This will also be justified in Sec.~\ref{sec:Universality}.
131: Finally we treat a saddle-centre bifurcation in the H\'enon map as an example.
132:
133: \section{Saddle-Centre Bifurcation}
134:
135: The normal form of a Hamiltonian system with two degrees of freedom near the $1:1$ resonance has the form
\begin{equation} \label{Hsn}
136: H(u,v)=\frac{v^2}{2}+\frac{u^3}{3}+\eps u\,.
137: \end{equation}
138: The coefficient of $u^3$ has been scaled so that it equals $1/3$.
139: The variables $u$ and $v$ are canonically conjugate variables
140: on a local transversal Poincar\'e section and $\eps$ is a parameter, typically
141: the energy of the original system.
142: The Poincar\'e map is given by the time $1$ map of the flow of $H(u,v)$.
143: If the period $T$ is large, the time $1$ map advances little.
144: The rotation number of the full system is the period $1$ divided by
145: the period of the reduced one degree of freedom flow, $\Omega = 1/T$.
146: The more familiar $\Omega = 2\pi/T$ is obtained when the time $2\pi$ map
147: is taken instead of the time $1$ map, but the time $1$ map is more natural
148: at least for the example of the H\'enon map we are going to discuss.
149: Since $\Omega$ is determined by $T$, we now study in detail the period $T$
150: of the one degree of freedom system given by $H$.
151:
152: The critical points and critical values of the energy map $H : \R^2 \to \R$
153: and their dependence on $\eps$ give the main structure to the bifurcation.
154: Instead of a one degree of freedom system $H(u,v;\eps)$ depending on the
155: parameter $\eps$ one may consider $H(u,v,\eps,\theta)$ as a Hamiltonian in
156: $\R^3\times S^1$, with action $\eps$ and conjugate angle $\theta$. The set of
157: critical values of the energy-momentum map $(H,\eps):\R^3 \times S^1 \to \R^2$
158: is called the bifurcation diagram. A simple way to compute it is to find
159: the critical values of the energy map of $H(u,v)$ and consider their
160: parameter dependence on $\eps$.
161: %
162: The Hamiltonian has critical points $(u,v) = (\pm\sqrt{-\eps}, 0)$
163: and corresponding critical values $h=\mp2(-\eps)^{3/2}/3$. They exist when
164: $\eps < 0$ and the upper sign corresponds to a local minimum of $H$,
165: while the lower sign gives a saddle.
166: The corresponding phase portraits are shown in Fig.~\ref{fig:SNphase}.
167:
168: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=5cm]{scpnegM2}
\includegraphics[width=5cm]{scpzeroM1}
\includegraphics[width=5cm]{scpposM1}}
169: \caption{Lines of constant energy on the phase space $(u,v)$ for the saddle-centre
170: bifurcation ($1:1$ resonance, $\omega = 0$).
171: left: $\eps = -0.1$, middle: $\eps = 0$, right: $\eps = 0.1$.} \label{fig:SNphase}
\end{figure}
172: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
173:
174:
175: The dynamics is given by Hamiltons equation
176: $\dot u = v$ and eliminating $v$ using the Hamiltonian gives a first
177: order equation for $u$. After separation of variables the period
178: of motion with energy $h$ is given by the elliptic integral
179: \begin{equation} \label{Teq3}
180: T(h,\eps)= \oint\frac{{\rm d}u}{\sqrt{2h-\frac23 u^3-2\eps u}}\,,
181: \end{equation}
182: where the integration path is encircling the interval on the real
183: axis where the argument of the square root is positive. If there are
184: two positive intervals either one can be taken, the result is the same.
185: %
186: By scaling $u=z (\sigma \eps)^{1/2}$, where $\sigma = {\rm sign}(\eps)$
187: and introducing the one essential parameter
188: \begin{equation} \label{gammadef}
189: \gamma = \frac{3 h}{2(\sigma\eps)^{3/2}}\,,
190: \end{equation}
191: the period $T$ is an elliptic integral on the curve
192: \begin{equation} \label{ElliCurve3}
193: {\cal E}: w^2 = P_3(z) = 2\gamma - z^3 - 3 \sigma z\,.
194: \end{equation}
195: The case $\eps = 0$ has to be excluded in this scaling, but
196: it is simple to treat it separately. We are mostly interested
197: in the case where $\sigma = 1$.
198: The essential integral now reads
199: \begin{equation} \label{Seq3}
200: S(\gamma) = \oint \frac1w \, {\rm d}z
201: \end{equation}
202: and it is related to the period by
203: \begin{equation} \label{TSeq3}
204: T(h,\eps) = \frac{\sqrt{3/2}}{ (\sigma\eps)^{1/4}}S(\gamma) \,.
205: \end{equation}
206: The polynomial $P_3$ has one or three real roots.
207: The collision of two real roots corresponds to the
208: unstable equilibrium and its separatrix. It occurs when
209: the discriminant
210: \[
211: \Delta = -108(\sigma + \gamma^2)
212: \]
213: vanishes. $\Delta = 0$ is only possible for $\sigma = -1$
214: and hence the critical parameters for which a double root
215: occurs are given by $\gamma = \pm 1$, hence
216: \begin{equation} \label{bifcurve}
217: 9 h^2 = -4\eps^3\,,
218: \end{equation}
219: which has a cusp at the origin.
220: At the origin $h=\eps=0$ all three roots collide
221: in the saddle-centre bifurcation.
222: The discriminant (\ref{bifcurve}) is shown in Fig.~\ref{fig:SNbif}.
223: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=5cm]{SNbif}}
\caption{Schematic sketch of the bifurcation diagram for the saddle-centre bifurcation.
224: Graphs of $P_3$ are shown together with a horizontal line indicating the value of $h$.
225: The bold lines are the critical values of the saddle. } \label{fig:SNbif}
\end{figure}
226: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
227: In the case of the saddle-centre bifurcation the bifurcation diagram is given by
228: the discriminant of $P_3$.
229: The bifurcation diagram divides the parameter plane $(\eps,h)$ into
230: two regions: one with 3 real roots and one with 1 real and two
231: complex roots. The latter has positive $\eps$ everywhere, while
232: the former is the wedge shaped region contained in the negative
233: half-plane. For $\eps < 0$ the phase portraits
234: contain a pair of stable/unstable fixed points. The critical
235: value of the energy of the stable fixed point is a local minimum
236: given by the lower branch of the bifurcation diagram, while that
237: of the unstable fixed point is a saddle given by the branch with
238: positive $h$. For this range of energies there are 3 real roots.
239: It will turn out that the upper branch of the bifurcation diagram
240: corresponding to the unstable fixed point is crucial for the
241: existence of vanishing twist.
242: For $\eps > 0$ the phase portrait is without fixed points.
243: %
244: Even though the topology is trivial in this case
245: we will now show that the rotation number has a maximum
246: on a certain invariant torus containing points near the origin
247: in the phase space. At this maximum of the rotation number
248: the twist vanishes. The vanishing twist occurs at the vertical
249: tangents of the contours of the rotation number shown in
250: Fig.~\ref{fig:SNWconst}. The fact that the invariant curves
251: are all unbounded for $\eps > 0$ will be dealt with in section~\ref{sec:Universality}.
252: For now observe that the integral (\ref{Teq3}) is finite,
253: even though the invariant curves are unbounded in $v$.
254: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=9cm]{saddlecentreM1}}
\caption{Lines of constant period $T=1/\Omega$ equidistant with $\Delta T = 0.3$ on the
255: parameter plane $(\eps, h)$ for the saddle-centre bifurcation ($1:1$ resonance, $\omega = 0$).
256: The vanishing twist is indicated by a curve of vertical tangents $\partial \Omega / \partial h = 0$.
257: }
258: \label{fig:SNWconst}
\end{figure}
259: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
260: The main feature of the level lines of the rotation number as shown in Fig.~\ref{fig:SNWconst}
261: is that it
262: diverges when the unstable periodic orbit is approached. This occurs
263: for the positive critical value of $h$ when $\eps < 0$. Everywhere else
264: the rotation number is a well defined, smooth and bounded function of
265: $h$ and $\eps$. Accordingly the level lines ``hug'' the curve of critical
266: values that correspond to the unstable orbit. Already from this property one
267: can deduce the existence of a curve with vanishing twist using topological arguments.
268: Here we proceed along the analytical route, because it will give us more
269: detailed information.
270: Note that the curve of critical values with negative $h$, see Fig.~\ref{fig:SNbif},
271: does not appear in Fig.~\ref{fig:SNWconst}.
272: One reason for this is that we chose to plot the rotation
273: number for the (non-compact) invariant tori with motion between
274: $-\infty$ and the smallest real zero of $P_3$. These invariant tori do
275: not contain critical points of the energy map, even though the corresponding
276: energy might be a critical value.
277: Accordingly the rotation number is smooth across this line of critical values.
278: The critical point corresponding to the critical values is the stable fixed point
279: at the local minimum of $H$.
280: But even if we would plot the rotation number of
281: the bounded invariant tori near the local minimum of the potential the
282: picture is unchanged. The reason is that for a cubic elliptic curve the
283: integrals of first kind over either one of the two real intervals (if they exist)
284: are equal.
285:
286:
287: Since $\Omega = 1/T$ and $2\pi T = \partial J/\partial h$ the derivative of
288: the rotation function is
289: \[
290: \frac{\partial \Omega}{\partial J} = \frac{\partial \Omega/\partial h}{\partial J/\partial h}
291: = -\frac{\partial T/\partial h}{2\pi T^3} \,.
292: \]
293: Hence the twist vanishes when
294: \[
295: \frac{\partial T}{\partial h} = \frac{(3/2)^{3/2}}{\eps^{7/4}}
296: \frac{\partial S}{\partial \gamma} = 0
297: \]
298: and this is only possible for finite $\eps$ when
299: \[
300: \frac{\partial S}{\partial \gamma} = - \oint \frac{1}{w^3} \,{\rm d}z = 0\,.
301: \]
302: This complete elliptic integral can be written as a linear combination of
303: Legendre's standard integrals. In this way a condition
304: for the vanishing of the twist is now obtained.
305: The relevant case for this purpose is that of
306: one real root, for which the phase portrait has no fixed point.
307: The integrand $w$ is positive for $u \in (-\infty, z_0)$, where
308: $z_0$ is the single root of $P_3(z)$.
309: Let the factorized polynomial be given by
310: \begin{equation} \label{P3fac}
311: P_3(z) = -(z-z_0)P_2(z), \qquad
312: P_2(z) = (z-\zeta_1)^2 + \zeta_2^2\,,
313: \end{equation}
314: so that the complex roots are $\zeta_1 \pm i \zeta_2$.
315: Denote the distance between the real and complex roots by $r$,
316: hence $r^2 = P_2(z_0)$, so that the discriminant is given by
317: $\Delta = -4r^4\zeta_2^2$.
318: Legendre's standard integral of the first kind $K(k)$ has differential
319: \begin{equation} \label{eq:omega1}
320: \omega_1 = \frac{1}{\sqrt{P_3(z)}} {\rm d}z
321: \end{equation}
322: up to a constant factor, where the modulus $k$ is given by
323: \begin{equation} \label{keq}
324: k^2 = \frac12 \left( 1+\frac{z_0 - \zeta_1}{r} \right)
325: = \frac12\left( 1 + \frac{\sign(\gamma)}{\sqrt{1+\alpha^2}}\right) \,,
326: \end{equation}
327: and in the last equality the parameter $\alpha = \zeta_2/(z_0-\zeta_1)$
328: has been introduced.
329: In the second equality in addition $\sign( z_0 - \zeta_1) = \sign( \gamma)$ is used,
330: which is true because $z_0 = \zeta_1$ in (\ref{P3fac}) together with
331: the vanishing of the quadratic coeffcient in (\ref{ElliCurve3}) implies $z_0 = \zeta_1 = 0$,
332: and therefoe the polynomial has no constant term and $h = \gamma=0$.
333:
334: A non-standard form of the differential of Legendre's standard integral
335: of second kind $E(k)$ is
336: \[
337: \omega_2 = \frac{P_2(z){\rm d}z}{(z-(z_0 \pm r))^2} \omega_1 \,,
338: \]
339: up to the same constant factor as in (\ref{eq:omega1}).
340: The differential $\dee z/w^3 $ we are interested in is of the second kind,
341: and can therefore be written as a linear combination of $\omega_1$
342: and $\omega_2$ with constant coefficients, up to a total differential:
343: \[
344: \frac{{\rm d}z}{w^3} = A \omega_1 + B \omega_2 + {\rm d} F
345: \]
346: where $F = Q_2(z)/((z-(z_0\pm r))w)$. Together with the undetermined
347: coefficients of the quadratic polynomial $Q_2$ this gives a system of 5
348: linear equations for the 5 unknown coefficients. Solving these equations gives
349: \[
350: A = \frac{2r}{\Delta} ((z_0 - \zeta_1)r - r^2 + 4\zeta_2^2), \quad
351: B = \frac{4r}{\Delta} (r^2 - 4\zeta_2^2)
352: \]
353:
354: Since the quadratic coefficient of $P_3$ is zero, the roots of $P_3$
355: add up to zero. Therefore the real parts satisfy $z_0 + 2 \zeta_1 = 0$,
356: hence
357: \[
358: 2k^2 = 1 - 3\zeta_1/r \quad \mbox{and} \quad r^2 = 9 \zeta_1^2 + \zeta_2^2\,.
359: \]
360: With these equations the coefficients $A$ and $B$ can be expressed in
361: terms of $k$ alone, up to the factor $(r^2 \Delta)^{-1}$.
362: The condition of vanishing twist, $\partial S/\partial \gamma = 0$,
363: finally reads
364: \[
365: (8k^4 - 9k^2 + 1)K(k)=(16k^4 - 16k^2 + 1)E(k)\, ,
366: \]
367: where $K(k)$ and $E(k)$ stand for elliptic integrals of the first and the second kind, respectively.
368:
369: For $k=0$ we have equality since both elliptic integrals equal $\pi/2$.
370: The first derivatives of either side vanishes, but the second derivatives
371: are $-35\pi/4$ and $-65\pi/4$, respectively,
372: so that the left hand side is larger for small $k$.
373: For $k=1$ the prefactor of $K$ vanishes, while that of $E$ gives 1
374: and $E(1) = 1$. Hence for $k \to 1$ the right hand side dominates.
375: This proves that there exists a solution of this equation
376: for $k \in (0,1)$. Numerically we find $k_0^2 \approx 0.7097215$.
377: In order to calculate the corresponding $\gamma_0$ we observe that
378: $\alpha$ as introduced in (\ref{keq}) is related to $\gamma$ by
379: \begin{equation} \label{alphaeq}
380: \alpha = \frac{\zeta_2}{z_0 - \zeta_1} = -\frac{\zeta_2}{3\zeta_1}
%= \frac{1}{\sqrt{3}}\left(\frac{(\gamma+\sqrt{1+\gamma^2})^{2/3}+1}
381: % {(\gamma+\sqrt{1+\gamma^2})^{2/3}-1}\right)\,
382: = \frac{1}{\sqrt{3}}\frac{\Gamma^2 + \sigma}{\Gamma^2 - \sigma} \,,
383: \end{equation}
384: where
385: \begin{equation} \label{Gdef}
386: \Gamma = ( \gamma + \sqrt{\sigma+\gamma^2})^{1/3} \,.
387: \end{equation}
388: Using (\ref{keq}) and $k_0$ the corresponding value of $\alpha$ is
389: $\alpha_0 \approx 2.164255$,
390: and from $\alpha_0$ using (\ref{alphaeq}) we find $\gamma_0 \approx 0.9152203 $.
391: Therefore the curve of vanishing twist in the parameter plane occurs for positive $\eps$
392: when
393: \begin{equation} \label{VTg0curve}
394: 3 h = 2 \gamma_0 \eps^{3/2}\,.
395: \end{equation}
396: Since $\gamma_0 < 1$ the curve of vanishing twist is bent downward
397: as compared to the bifurcation curve (\ref{bifcurve}) for
398: $\eps < 0$ and $h>0$. See Fig.~\ref{fig:SNWconst} for a graph of this
399: curve together with the numerically computed lines of constant rotation number.
400: The lines of constant rotation number have vertical slope at their intersection
401: with the critical curve, as must be the case.
402:
403: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=7cm]{SNubifdi}}
\caption{Bifurcation diagram of the position in phase space $u$ versus the bifurcation
404: parameter $\eps$. For $\eps < 0$ the location of the fixed points at $\pm \sqrt{-\eps}$ are shown,
405: while for $\eps > 0$ the maximal $u$ of the twistless torus at $z_0(\gamma_0) \sqrt{\eps}$ is shown.}
406: \label{fig:Snubd}
\end{figure}
407: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
408:
409: The scaling of $u$ reduces the number of parameters to one.
410: The essential parameter $\gamma$, in its dependence on $h$
411: and $\eps$, organizes the bifurcation. It allows to compute explicitly
412: all the important characteristics of the twistless torus.
413: Combining (\ref{keq}) and (\ref{alphaeq}) shows that $k$ is a function
414: of $\gamma$. The curves in the parameter plane that have the same
415: value of $\gamma$ are given by (\ref{gammadef}). The most prominent
416: ones are the curve of critical values $\gamma = \pm1$, as shown in Fig.~\ref{fig:SNbif},
417: and $\gamma = \gamma_0$, the curve of twistless tori, see Fig.~\ref{fig:SNWconst}.
418: They all have the same shape
419: of a semicubical parabola, except when $\gamma = 0$, hence $h=0$,
420: or $\gamma = \pm\infty$, hence $\eps = 0$ with $\pm h > 0$.
421: \[
422: \begin{array}{cccc}
423: k^2 & \gamma & \alpha & \text{curve} \\ \hline
424: 1 & +1 & 0 & 3h = -2(-\eps)^{3/2} \\
425: \frac14(2+\sqrt{3}) & +\infty & {1}/{\sqrt{3}} & \eps = 0, h>0 \\
426: 0.709721497 & 0.91522 & 2.164255 & 3h = \gamma_0(-\eps)^{3/2} \\
427: \frac12 & 0 & \infty & \eps = 0, h > 0 \\
428: \frac14(2-\sqrt{3}) & -\infty & {1}/{\sqrt{3}} & \eps = 0, h<0 \\
429: 0 & -1 & 0 & 3h = 2(-\eps)^{3/2}
430: \end{array}
431: \]
432: Approaching the bifurcation point along the curve (\ref{gammadef}) gives
433: $k(\gamma)$ in the limit. The function $k(h,\eps)$ is therefore
434: not continuous at the origin.
435: Moving on curves (\ref{gammadef}) in the parameter plane for any $\gamma \neq 1$
436: the change in the period $T(h,\eps)$ given by (\ref{Teq3}) is elementary. From
437: (\ref{TSeq3}) it follows that $T$ is proportional to $|\eps|^{-1/4}$, and the constant
438: of proportionality is determined from (\ref{TSeq3}) and (\ref{Seq3}).
439: The divergence of the period upon approaching the bifurcation is therefore
440: not caused by the elliptic integral, but merely by the algebraic dependence
441: $|\eps|^{-1/4}$.
442: %
443: In this way the rotation number of the stable periodic orbit is given by
444: \[
445: \Omega(\gamma=-1,\eps) = \frac{1}{\sqrt{2}\pi} (-\eps)^{1/4} \approx 0.225079 \, |\eps|^{1/4}\,.
446: \]
447: The integral $S(-1)$ contained in this expression can be easily calculated using
448: residue calculus because for $\gamma = -1$ the curve has a double root, $P_3(z) = -(z+2)(z-1)^2$.
449: For general values of $\gamma$ and in particular for the twistless torus the
450: elliptic integral $S(\gamma)$ (\ref{Seq3}) needs to be calculated.
451: The single real root is given by
452: \begin{equation} \label{z00}
453: z_0 = \Gamma - \frac{\sigma}{\Gamma}, \quad \text{hence} \quad
454: z_0(\gamma_0) \approx 0.5535942\,.
455: \end{equation}
456: Finally $S(\gamma)$ is obtained as
457: \[
458: S(\gamma) = \frac{4K(k) }{\sqrt{3 z_0 \sqrt{1+\alpha^2}/2 } } \,.
459: \]
460: In particular when $\gamma = \gamma_0$ the rotation number
461: of the twistless torus is
462: \begin{equation} \label{Omlim}
463: \Omega(\gamma=\gamma_0, \eps) =
464: \frac{ \sqrt{ z_0(\gamma_0) \sqrt{1+\alpha^2_0}}} {4K(k_0) } \eps^{1/4}
465: \approx 0.1374244 \, \eps^{1/4}\,.
466: \end{equation}
467: The constant of proportionality is close to $\sqrt{3}/4\pi$.
468: Using the above value of $z_0$ the position of the rightmost point
469: of the twistless torus in phase space is located at $u_0 = z_0 \sqrt{\eps}$.
470: This means that for $\eps = \eps_0$ this point is on the same side as
471: the stable periodic orbit was for $\eps = -\eps_{0}$ before the bifurcation,
472: but by a factor of $1/z_0(\gamma_0) \approx 2$ closer to the origin.
473: See Fig.~\ref{fig:Snubd} for an illustration in the form
474: of a standard bifurcation diagram showing position in phase space $u$
475: versus bifurcation parameter $\eps$.
476:
477: % \clearpage
478:
479: \section{Period-Doubling Bifurcation}
480:
481: The normal form of a Hamiltonian system with two degrees of freedom near a
482: periodic orbit in $1:2$ resonance is
\begin{equation} \label{Hpd}
483: H(u,v)=\frac{v^2}{2}+D\frac{u^4}{4}+\eps u^2\,,
\end{equation}
484: where $D = \pm 1$.
485: As in the case of the saddle-centre bifurcation the variables $u$ and $v$ are canonically conjugate variables
486: on a local transversal Poincar\'e section and $\eps$ is a parameter, typically corresponding to
487: the energy of the original system. Contourplots of this Hamiltonian
488: show the intersection of invariant tori with the Poincar\'e section transversal
489: to the bifurcating orbit.
490: For $D=1$ they are shown in Fig.~\ref{fig:PDphaseP},
491: for $D=-1$ in Fig.~\ref{fig:PDphaseN},
492: for $\eps < 0$, $\eps = 0$, and $\eps > 0$, respectively.
493: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=5cm]{pdApnegM2}
\includegraphics[width=5cm]{pdApzeroM1}
\includegraphics[width=5cm]{pdApposM1}}
494: \caption{Lines of the constant energy on the phase space $(u,v)$
495: for the period-doubling bifurcation $(D = 1)$ for $\eps = -0.1$, $\eps = 0$, $\eps = 0.1$.}
496: \label{fig:PDphaseP}
\end{figure}
497: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
498:
499: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{%
500: \includegraphics[width=5cm]{pdBpposM1}
\includegraphics[width=5cm]{pdBpzeroM1}
\includegraphics[width=5cm]{pdBpnegM2}
501: }
\caption{Lines of the constant energy on the phase space $(u,v)$
502: for the period-doubling bifurcation $(D = -1)$, for $\eps = -0.1$, $\eps = 0$, $\eps = 0.1$}
503: \label{fig:PDphaseN}
\end{figure}
504: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
505:
506:
507: The critical points and critical values of the energy map $H : \R^2 \to \R$ and their
508: dependence on the parameter $\eps$ describe the structure of the bifurcation.
509: The Hamiltonian has critical points at $(0,0)$ and at $(u,v) = (\pm\sqrt{-2\eps D},0)$
510: with critical values $h=0$ and $h=-D\eps^2$. The origin is a local minimum
511: for $\eps>0$, a saddle otherwise. The second critical point
512: exists when $\eps D < 0$, and is a local minimum for $\eps<0$, a saddle otherwise.
513: The set of critical values, see Fig.~\ref{fig:PDbif}, therefore is the union of the line $h=0$ with
514: the half of the parabola $h = -D\eps^2$ for which $D\eps < 0$.
515: For $D>0$ the only unstable branch in the bifurcation diagram is
516: $h=0$ for $\eps < 0$. It divides the two regions of real motion.
517: A third region $\{h < -\eps^2\} \cup \{\eps > 0, -\eps^2 < h < 0\}$ is not
518: in the image of $H$, hence there is no real motion corresponding
519: to $(h,\eps)$ from this region.
520:
521: For $D<0$ the line of critical values $h=0$ again has a saddle
522: as critical point when $\eps < 0$. In addition the half parabola
523: $h=\eps^2$, $\eps > 0$ also corresponds to a saddle of $H$ that
524: is not at the origin. When $D< 0$ the whole plane $(h,\eps)$ is
525: in the image of $H$ considering all $\eps$. The critical values divide
526: the plane into three regions with 0, 2, and 4 real roots.
527:
528: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=12cm]{PDbif}}
\caption{Schematic sketch of the bifurcation diagram for the two cases of
529: the period doubling bifurcation.
530: Graphs of $P_4$ are shown together with a horizontal line indicating the value of $h$.
531: The bold line denote critical values of unstable orbits.
532: The dashed line is not critical; it indicates a vanishing of the discrimiant.
533: } \label{fig:PDbif}
\end{figure}
534: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
535:
536: The dynamics of the reduced one degree of freedom system is given by
537: $\dot u = v$ and eliminating $v$ using the Hamiltonian gives a first
538: order equation for $u$. Separation of variables then gives the period
539: of motion in the reduced one degree of freedom system as the elliptic integral
540: \[
541: T(h,\eps)=\oint\frac{{\rm d}u}{\sqrt{2h - \frac12 D u^4-2\eps u^2}}\,.
\]
542: The number of parameters could be reduced by introducing the ratio $h/\eps^2$,
543: but for clarity we do not introduce this scaling.
544: The period $T$ is an elliptic integral on the curve
545: \begin{equation} \label{ElliCurve4}
546: {\cal E}: w^2 = P_4(z) = P_2(z^2) = 2h - \frac12 D z^4 - 2\eps z^2 \,.
547: \end{equation}
548: The discriminant of $P_4$ is
549: \[
550: \Delta = -256 D h (D h + \eps^2)^2\,.
551: \]
552: It vanishes for $h = 0$ and $Dh = -\eps^2$, corresponding to the critical
553: values $h=0$ and $h= -D \eps^2$ already found above.
554: As usual the set of critical values is contained in the discriminant, however,
555: the discriminant vanishes on a larger set. Namely the branch of the parabola
556: with $D\eps > 0$ is not part of the critical values.
557: The three regions already found correspond to regions with
558: 4, 2, and 0 real roots of $P_4$, respectively, see Fig.~\ref{fig:PDbif}.
559: The part of the discriminant that is not part of the critical values is dashed.
560: The polynomial can be factored as
561: \[
562: P_4(z) = -\frac12 D(\xi_- - z^2 )(\xi_+ - z^2 ), \quad
563: \frac12 D\xi_{\pm} = - \eps \pm \sqrt{\eps^2 + Dh}\,.
564: \]
565: Comparing coefficients gives $\xi_-\xi_+ = -4hD$ and $\xi_- + \xi_+ = -4\eps D$.
566: % and the roots are $D\xi_{\pm}/2 = - \eps \pm \sqrt{\eps^2 + Dh}$.
567: The factorization of $P_4$ has real factors, i.e.\ $\xi_\pm \in \R$, with one exception.
568: It occurs when the quadratic equation $P_2(\xi)= 0$ has complex roots.
569: The position of the roots in the regions of the parameter plane is
570: as follows.
571: The two real cases for $D = 1$ are
572: \begin{align*}
573: +2: && h > 0 && \xi_- < 0 < \xi_+ && 2 [-\sqrt{\xi_+},\sqrt{\xi_+}] \\
574: %, \eps>0: |+| < |-| \\
575: +4: && \eps < 0, -\eps^2 < h < 0 && 0 < \xi_- < \xi_+ && 2 [\sqrt{\xi_-},\sqrt{\xi_+}]
576: \end{align*}
577: In the first column a label is given containing the number of real roots.
578: All 4 cases appear when $D=-1$ for
579: \begin{align*}
580: -2: && h < 0 && \xi_+ < 0 < \xi_- && 2[-\infty, -\sqrt{\xi_-} ] \\
581: -4: && \eps > 0, 0 < h < \eps^2 && 0 < \xi_+ < \xi_- && 2[-\infty, -\sqrt{\xi_-} ] \\
582: -0i: && \eps < 0, 0 < h < \eps^2 && \xi_+ < \xi_- < 0 && [-\infty,\infty] \\
583: -0c: && h > \eps^2 && \xi_\pm \text{ complex} && [-\infty,\infty]
584: \end{align*}
585: The last column gives the interval of real motion along which the period integral is taken.
586: The region without real roots contains two parts separated by the branch
587: of the parabola $h = -D\eps^2$ with $D\eps > 0$.
588: On this branch there occurs a collision of complex roots at $z^2 = -2\eps D$,
589: and they move from the imaginary axis into the complex plane.
590:
591: The intervals of integration and multiplication factors as given in the last
592: column of the previous table can be read off from
593: the phase diagrams, see Fig.~\ref{fig:PDphaseP}, \ref{fig:PDphaseN}.
594: In the case $D=-1$ with 4 real roots
595: there are different orbits for the same $(h,\eps)$. One is the compact orbit
596: near the stable fixed point with extent $2[-\sqrt{\xi_+}, \sqrt{\xi_+} ]$. The interval
597: given above is for the non-compact orbit, which has half of this period.
598:
599:
600: In order to calculate the derivative of $T$ with respect to $h$ the integral
601: is first written in standard form and then differentiated. This more traditional
602: approach (as compared to the previous section) is preferable in this case
603: because the roots of the quartic are easily written down.
604:
605: Denote the ratio of the roots $\xi_\pm$ by
606: \[
607: r = \frac{\xi_-}{\xi_+} = \frac{-\eps - \sqrt{\eps^2 + Dh}}{-\eps + \sqrt{\eps^2 + Dh}} \,.
608: \]
609: Then the period $T(h,\eps)$ in the 6 cases is given by
610: \begin{align}
611: +2,-2: && T = & \left(\frac{8(1-2k^{2})}{\eps}\right)^{1/2} K(k),
612: && k^2 = \frac{1}{1 - r}, \\
613: +4,-0i: && T = & \left(\frac{8(k^{2}-2)}{\eps}\right)^{1/2} K(k),
614: && k^2 = 1 - r, \\
615: -4: && T = & \left(\frac{8(1+k^{2})}{\eps}\right)^{1/2} K(k),
616: && k^2 = \frac{1}{r}, \\
617: -0c: && T = & \left(\frac{8(2k^{2}-1)}{\eps}\right)^{1/2} K(k),
618: && k^2 = \frac12 + \frac{\eps}{2\sqrt{h}}\,.
619: \end{align}
620: The level lines of the period $T$ (and hence the rotation number $\Omega = 1/T$) are shown in
621: Fig.~\ref{fig:PDWconstP} for $D=1$ and in
622: Fig.~\ref{fig:PDWconstN} for $D=-1$.
623: These numerically computed pictures show that there are no
624: vertical tangents, hence the twist does not vanish.
625: This is now proved by differentiating the period in Legendre
626: normal form.
627:
628: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=9cm]{perioddoublingAM2}}
\caption{Lines of constant period $T=1/\Omega$ equidistant with $\Delta T = 0.3$
629: on the parameter plane $(\eps, h)$ for the period-doubling bifurcation ($1:2$ resonance,
630: $\omega = 1/2$) with $D = 1$}
631: \label{fig:PDWconstP}
\end{figure}
632: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
633:
634: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{figure}
\centerline{\includegraphics[width=9cm]{perioddoublingBM2}}
\caption{Lines of constant period $T=1/\Omega$ equidistant with $\Delta T = 0.3$ on the
635: parameter plane $(\eps, h)$ for the period-doubling bifurcation
636: ($1:2$ resonance, $\omega = 1/2$) with $D = -1$}
637: \label{fig:PDWconstN}
\end{figure}
638: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
639:
640: The twist vanishes when
641: \[
642: 0 = \frac{\partial T}{\partial h} = \frac{\partial T}{\partial k} \frac{\partial k}{\partial h} \,.
643: \]
644: The last factor $\partial k/\partial h$ does not vanish for $k\in (0,1)$.
645: In the cases $+2$, $-2$, and $-0c$ it seems to vanish when $k=1/2$.
646: But this implies that $\eps = 0$ and the singularity cancels.
647: Therefore it is enough to consider ${\partial T}/{\partial k} = 0$.
648: After removing common non-vanishing factors the conditions for
649: vanishing twist are $V_i(k) = 0$, $i=1,2,3$ where
650: \begin{align}
651: +2, -2, -0c: && V_1(k) =& (1-2k^2) E(k) - (1-k^2) K(k), \\
652: +4, -0i: && V_2(k) =& \left(1-\frac{k^2}{2}\right) E(k) - (1-k^2) K(k), \\
653: -4: && V_3(k) =& (1+k^2) E(k) - (1-k^2) K(k) \,.
654: \end{align}
655: The three equalities $V_i(k) = 0$, $i=1,2,3$ are never satisfied on the range $k \in (0,1)$.
656: Obviously $V_i(0) = 0$, while $V_i(1) = -1, 1/2, 2$ for $i=1,2,3$, respectively. So
657: we need to show that $V_1$ is negative and $V_2$, $V_3$ are positive for
658: $k \in (0,1)$.
659: Differentiating $V_1$ and $V_2$ gives the simple results
660: \begin{align}
661: % V'_1(k) = & -3k( K(k) - 2E(k)) \\
662: V'_2(k) = & -\frac32k (K(k) - E(k)) \\
663: V'_3(k) = & -3k E(k)
664: \end{align}
665: In the last case $V'_3(k)$ is obviously non-positive, so that the twist in the case
666: $-4$ is a monotone funtion rising from 0, hence it is nonzero.
667: Similarly also $V'_2(k)$ is non-positive, which follows from
668: the well known inequality $K(k) > E(k)$.
669: %
670: The first function $V_1(k)$ is negative on $(0,1)$, but not monotone.
671: Rewriting it as
672: \[
673: V_1(k) = (1-k^2)(E(k)-K(k)) - k^2 E(k) < 0
674: \]
675: the inequality is clear because both terms are negative for $k\in (0,1)$.
676: Therefore the twist never vanishes in a neighborhood of
677: the period doubling bifurcation.
678:
679:
680: The relation between the cases $D= \pm1$ is interesting. The main observation is
681: that changing the sign of $D$ and $\eps$ inverts the overall sign of the potential
682: $V(x;D,\eps) = Dx^4/4 + 2\eps x^2$. Therefore changing the signs of $D$, $\eps$,
683: and $h$ leaves the roots of $P_4$ invariant.
684: By this mapping the regions in parameter space with the same numbers of real
685: roots are mapped into each other. The integration path does change in a
686: less trivial way. The integration needs to be taken over the positive intervals
687: of $P_4$ on the real axis. Changing the sings of $D$, $\eps$, and $h$ does change
688: the sign of $P_4$.
689: As a result the periods for
690: $D=1$ are the $\alpha$ cycles of the elliptic curve, while those for $D=-1$ are
691: the $\beta$ cycles. Hence the period for the case $-4$ can be obtained
692: from that of $+4$ by replacing $K(k)$ by $K'(k) = K(\sqrt{1-k^2})$.
693: Similarly, the cases $+2$ and $-2$ are mapped into each other.
694:
695: The obvious symmetry in Figures~\ref{fig:PDWconstP} and \ref{fig:PDWconstN}
696: with respect to changing the sign of $h$
697: is related to the fact that changing the sign of $D$ and $h$ leaves the
698: ratio $r$, and therefore also the corresponding modulus $k^2$, invariant.
699: This means that the level lines in region $+2$ and $-2$ can be obtained from
700: each other by reflection on the $h=0$ axis. In a similar way the regions $+4$ and
701: $-0i$ have the same rotation number.
702:
703: The period doubling could have been treated in a scaled version with only
704: one essential parameter $\delta = h/\eps^2$. However, the presentation seems
705: more transparent in the unscaled version. Similar to the case of the saddle-centre
706: bifurcation the modulus $k^2$ of the elliptic integral for the period is constant
707: on the parabolas $h = \delta \eps^2$. Again dependence on the parameters on
708: these curves is simply algebraic, as before through $|\eps|^{1/4}$.
709: The value of the modulus is not defined at the origin, but depends on the
710: parabola on which it is approached. But in any case, there are no twistless tori
711: near the origin in this bifurcation.
712:
713: \section{Universality} \label{sec:Universality}
714:
715: A major problem in our approach seems to be that the invariant twistless tori
716: in the normal form are not compact. The normal form is obtained from
717: an expansion near the bifurcation point, and is therefore local in phase space
718: and local in the parameter. How can the rotation number of a non-local
719: invariant torus be determined from this local normal form?
720: To answer this question higher order terms need to be considered in
721: the Hamiltonian. The Poincar\'e map near the bifurcation can be described by
722: the time $1$ map of the Hamiltonian
723: \[
724: \tilde H(u,v,t;\eps) = H(u,v;\eps) + G(u;\eps) + R(u,v,t;\eps)\,.
725: \]
726: The remainder terms $R$ containing the periodic time dependence can
727: be made arbitrarily small, but they cannot in general be removed while
728: retaining a non-zero radius of convergence of the normal form.
729: The first term $H$ is the normal form analysed in the previous chapters.
730: We will concentrate on the saddle-centre case (\ref{Hsn}),
731: but similar remarks apply to (\ref{Hpd}).
732: The invariant tori near the origin for $\eps>0$ of $H$ are not compact.
733: The higher order terms in $G$ can compactify them. The results about
734: vanishing twist can be applied when $G$ does compactify these curves.
735: However, the precise form of $G$ does not matter.
736: Under the compactness assumption KAM theory can be applied to $H+G$,
737: where $R$ is the perturbation. Many of the invariant tori of $H+G$ will persist.
738: In particular a twistless invariant torus of $H+G$ will persist if it is
739: sufficiently irrational. The curve (\ref{VTg0curve}) in the parameter plane
740: therefore does not have invariant twistless tori in its perimage for
741: every point, instead just for a cantorset of points. This is well understood.
742: The main issue in this section is to understand the effect of adding $G$ to $H$.
743: %
744: For small $\eps$ the essential contribution to the
745: diverging period comes from the dynamics near the origin, while
746: the dynamics on the invariant torus away from the origin has finite period.
747: This is why the local normal form can give a statement
748: about the dynamics on a non-local invariant torus near the bifurcation point.
749: In particular we will now show that the curve of vanishing twist that was found
750: to be emanating from the cusp of the saddle-centre bifurcation has
751: a universal shape sufficiently close to the cusp singularity.
752: In particular this means that the constant $\gamma_0$ that determines the
753: shape of the curve of twistless tori in relation to the curve of critical values of the
754: unstable orbits has the universal value $\gamma_0 \approx 0.91522$.
755: Quantities derived from $\gamma_0$, like $z_0$ and the coefficients in (\ref{Omlim}),
756: are accordingly also universal.
757: %
758: Our calculation will show that the value of $\gamma_0$ is not influenced by the higher order
759: terms $G$ in the Hamiltonian. Moreover, the following argument also shows that
760: integrating the non-compact invariant torus up to $v = \pm \infty$ does not introduce
761: and additional error.
762:
763: Let the high-order truncated normal form Hamiltonian be
764: \[
765: H(u,v;\eps) = \frac12 v^2 + \frac13 u^3 + \eps u + G(u,\eps)
766: \]
767: where $G(0,\eps) = 0$ and $G(u,0) = 0$ is an analytic function.
768: It is not necessary to assume that the higher order terms in $G$ depend
769: on $v$ also, see e.g.~\cite{MH92}, but even with such a dependence a slightly
770: modified argument would work. As already explained, it is now assumed that
771: $G(u,\eps)$ is such that the invariant curves near the origin for
772: $\eps > 0$ are compact.
773:
774: Hamiltons equation for $u$ reads $\dot u = v$ as before.
775: The period is obtained by solving $H(u,v;\eps) = h$ for $v = v(u;h,\eps)$
776: and then by integrating
777: \[
778: T(h,\eps) = \oint \frac{{\rm d} u}{v(u;h,\eps)} \,.
779: \]
780: The idea is to split the integral into two parts; one part near the origin, where the
781: main contribution originates, and the rest, which is called $T_3$.
782: In addition the singular integral near
783: the origin is split again into two parts, $T_1$ which has the same integrand
784: as in the previous calculation with $G=0$, and a correction $T_2$ which contains
785: the contribution from $G$. The integral $T_1$ will be the most singular,
786: $T_2$ is mildly singular, and $T_3$ is regular. For sufficiently small $\eps$
787: then $T_1$ dominates and the previous result is recovered.
788:
789: To achieve the splitting into $T_1$ and $T_2$ the multiplicative structure of
790: the original inegrand has to be recovered. Solving $H=h$ for $v$ and
791: inserting into $\dot u$ gives
792: \[
793: \dot u^2 = v(u;h,\eps)^2 = Q_0(u;h,\eps) \hat Q(u;h,\eps)
794: \]
795: where $\hat Q = 1 + O(u)$ when $|u| \le C \le 1$ for some fixed constant $C$,
796: and $Q_0$ is a polynomial of degree 3 in $u$ whose zeroes approach those
797: of the original case with $G=0$ when $\eps \to 0$. Denote by $\gamma_C$ the part of
798: $H(u,v;\eps) = h$ for which $|u| < C$, and $\bar \gamma_C$ the rest of the
799: invariant torus. Then the period integral can be split as
800: \begin{align}
801: T(h,\eps) = & \int_{\gamma_C} \frac{\dee u}{\sqrt{Q(u,\eps,h)}} +
802: \int_{\bar \gamma_C} \frac{\dee u}{\sqrt{Q(u,\eps,h)}} \\
803: = & \int_{\gamma_C} \frac{\dee u}{\sqrt{Q_0(u,\eps,h)}} +
804: \int_{\gamma_C} \frac{{\hat Q(u,h,\eps)}^{-1/2} - 1}{\sqrt{Q_0(u,\eps,h)}} \, \dee u +
805: \int_{\bar \gamma_C} \frac{\dee u}{\sqrt{Q(u,\eps,h)} } \\
806: = & T_1 + T_2 + T_3 \,.
807: \end{align}
808: Now $T_3$ is regular, and gives a bounded contribution, so we can ignore it.
809: The integral $T_1$ is singular in the limit $h,\eps \to 0$. $T_2$ is less
810: singular, and in particular bounded, because the numerator goes to zero in this limit.
811: So we only need to show that $T_1$ approaches the complete integral when
812: $\eps \to 0$. The integral is
813: \[
814: T_1(h,\eps) = \frac{4\sqrt{3/2}}{r^{1/2}} F(\phi,k), \quad
815: \phi = 2 \arctan \sqrt{\frac{u_0+C}{r}} \,,
816: \]
817: where $u_0$ is the real root, $\zeta$ is the complex roots of $Q_0$,
818: and $r$ and $k$ are as before,
819: \[
820: r = \sqrt{2u_0^2 + |\zeta|^2}, \quad
821: 2k^2 = 1 + \frac{3u_0}{2r} \,.
822: \]
823:
824: The integral $T_1$ does not diverge because of the modulus $k^2 \to 1$,
825: but because in the prefactor $r \to 0$.
826: In fact, we already observed that the modulus $k^2 \le (\sqrt{3} + 2)/4 < 1$
827: inside the first quadrant $\eps \ge 0, h > 0$.
828: The behaviour of $r$ for small $\eps$ is obtained from
829: \[
830: r^2 = \frac32 u_0 \sqrt{1 + \alpha^2} = D(\gamma) \eps^{1/4} \,.
831: \]
832: This shows that $\phi \to \pi$, and the integral approaches the
833: complete integral with the same modulus $k(\gamma)$ as before.
834: Therefore even though $T_1$ is an incomplete integral, in the limit
835: of small $\eps$ its value approaches that of the complete integral
836: $T(h,\eps)$, which diverges. Since the other integrals $T_2$ and $T_3$
837: are finite the analysis obtained from the complete integral over the
838: non-compact curve of $H(u,v;\eps)$ with $G=0$ therefore correctly
839: describes the behaviour of the rotation number of the compact invariant
840: curve obtained when $G \not = 0$.
841:
842:
843: \goodbreak
844:
845: \section{Example: H\'enon Map}
846:
847: \begin{table}
848: \centerline{
849: \begin{tabular}{lll}
850: $\eps$ & $u \eps^{-1/2}$ & $\Omega \eps^{-1/4}$ \\ \hline
851: 0.1 & 0.027 & 0.28314 \\
0.01 & 0.328 & 0.17278 \\
0.001 & 0.478 & 0.14656 \\
852: 0.0001 & 0.528 & 0.14008 \\
853: 0.00001 & 0.545 & 0.13823 \\
854: 0.000001 & 0.550 & 0.13767
855: \end{tabular}
856: }
857: \caption{Numerically measured position $u$ and rotation number $\Omega$ of the
858: twistless curve for
859: the three times iterated H\'enon map with parameter $k = 1 - \eps$.} \label{tabHen}
860: \end{table}
861:
862:
863: The H\'enon map in the area preserving case,
864: \[
865: (x', y') = (y - k + x^2, -x) \,,
866: \]
867: illustrates the above.
868: There are saddle-centre bifurcations in the H\'enon map for which the invariant tori
869: for $\eps > 0$ are not compact, and hence the vanishing twist cannot be observed.
870: This applies to the initial bifurcation at $k=-1$ that creates the pair of
871: fixed points, and also to many saddle-centre bifurcations that occur for $k > 4$.
872: %
873: However for $k=1$ a pair of period 3 orbits is created at
874: the origin in a saddle-centre bifurcation of the third iterate of the map,
875: and the corresponding invariant tori are compact.
876: One of the period three points is located on the symmetry line $x = -y$.
877: In new coordinates $(u,v) = (x, y+x)$ the third iterate of the map expanded
878: near the origin with parameter $k = 1- \eps$ is
879: \[
880: (u', v') = \left( (u-v)(1 + 4v) , v + \eps + (u^2 - 2uv + 3v^2)\right) + O(u^3,v^3,\sqrt{|\eps|}^3)\,.
881: \]
882: The (exact) location of the fixed points is $(u,v) = ( \pm \sqrt{-\eps}, 0)$,
883: with trace of the Jacobian $2 \mp 2 \sqrt{-\eps} + 8\eps \mp 8 (-\eps)^{3/2}$
884: so that the (approximate) multiplier is $\mu \approx 1 + \imag \sqrt{2} (-\eps)^{1/4} $
885: and $\bar\mu$ for the fixed point at $u = \sqrt{-\eps}$. The corresponding
886: rotation number $\omega$ of the elliptic fixed point
887: is obtained from $\mu = \exp( 2\pi \imag\omega)$, so
888: that $\omega = (-\eps)^{1/4} / (\sqrt{2}\pi)$.
889: For small positive $\eps$ the H\'enon map possesses compact invariant curves
890: near the origin. A higher order normal form would give a $G$ that describes
891: these invariant curves. The H\'enon map is non-integrable, so that only
892: sufficiently irrational invariant curves of the (high order) normal form will
893: exist in the H\'enon map. For the situation under consideration numerical
894: experiments show that many of these invariant curves do exist.
895:
896: Considering the third iterate of the H\'enon map turns the pair of period three orbits
897: into three pairs of fixed points with heteroclinic connections. In the normal form there
898: is only one pair of fixed points, and the unstable fixed point has a homoclinic
899: connection. To match the prediction in the case of more than one unstable fixed
900: point the period and hence rotation number must be calculated for the heteroclinic
901: connection. For $\eps > 0$ there are invariant tori on which the dynamics becomes
902: slow near the three points that are close to the three bifurcation points of the third
903: iterate of the map.
904: Hence the rotation number for the third iterate
905: of the map between two successive such points gives the correct rotation number.
906: The results together with the position of the twistless curve are shown in
907: table~\ref{tabHen}. The values shown converge to the predicted values
908: given in (\ref{z00}) and (\ref{Omlim}),
909: however, fairly small $\eps$ are needed to see this.
910: The convergence to the true value $0.1374244...$ occurs
911: approximately as $\tfrac12 \eps^{5/4}$.
912:
913: \bibliographystyle{plain}
\bibliography{notwist}
914:
915: \end{document}
916:
917:
918: \rem{
919: % for arxiv.org:
920: % include the bbl file, and all figures, make one archive from all of them, then web upload
921: \\
922: Title: (Vanishing) Twist in the Saddle-Centre and Period-Doubling Bifurcation
923: Authors: Holger R. Dullin, Alexey V. Ivanov (Loughborough University)
924: Comments: 18 pages, 9 figures
925: Report-no:
926: \\
927: The lowest order resonant bifurcations of a periodic orbit of a Hamiltonian system
928: with two degrees of freedom have frequency ratio 1:1 (saddle-centre)
929: and 1:2 (period-doubling). The twist, which is the derivative of the
rotation number with respect to the action, is studied near these bifurcations.
930: When the twist vanishes the nondegeneracy condition
931: of the (isoenergetic) KAM theorem is not satisfied,
932: with interesting consequences for the dynamics.
933: We show that near the saddle-centre bifurcation the twist always vanishes.
934: At this bifurcation a ``twistless'' torus is created, when the resonance is passed.
The twistless torus replaces the colliding periodic orbits in phase space.
935: We explicitly derive the position of the twistless torus depending on the resonance
936: parameter, and show that the shape of this curve is universal.
937: For the period doubling bifurcation the situation is different.
938: Here we show that the twist does not vanish in a neighborhood of the bifurcation.
939: \\
940:
941: }