1: \documentclass{elsart}
2: \usepackage[T1]{fontenc}
3: \usepackage[latin1]{inputenc}
4: \usepackage{graphicx}
5:
6: \providecommand{\tabularnewline}{\\}
7:
8: \usepackage{amssymb}
9:
10: \begin{document}
11:
12: \begin{frontmatter}
13: \title{Universal Scaling in Saddle-Node Bifurcation Cascades (I) }
14:
15: \author[addr1,addr2]{Jesús San-Martín}
16: \address[addr1]{Departamento de Matemática Aplicada, E.U.I.T.I, Universidad Politécnica de Madrid. Ronda de Valencia 3, 28012 Madrid Spain}
17: \address[addr2]{Departamento de Física Matemática y Fluidos, U.N.E.D. Senda
18: del Rey 9, 28040 Madrid Spain}
19: \ead{jsm@dfmf.uned.es}
20:
21: \begin{abstract}
22: A saddle-node bifurcation cascade is studied in the logistic equation,
23: whose bifurcation points follow an expression formally identical to
24: the one given by Feigenbaum for period doubling cascade. The Feigenbaum
25: equation is generalized because it rules several objects, which do
26: not have to be orbits. The outcome is that an attractor of attractors
27: appears, and information about the birth, death and scaling of windows
28: is obtained.
29: \end{abstract}
30:
31: \begin{keyword}
32: Saddle-Node bifurcation cascade. Attractor of attractors. Generalized
33: Feigenbaum equation. Scaling windows. Scaling Myrberg-Feigenbaum points.
34: Bifurcation rigidity.
35: \end{keyword}
36:
37: \end{frontmatter}
38:
39:
40: \section{Introduction}
41:
42: The study of the logistic equation
43:
44: \begin{equation}
45: x_{n+1}=f(x_{n})=rx_{n}(1-x_{n})\,\,\,\,\,\,0\leq r\leq4\label{eq:recurrencia}\end{equation}
46: bears relevance because it is a simple expression with a numerical
47: and theoretical easy use that at the same time shows universal and
48: varied behaviors. It is enough to mention the patterns of periodic
49: orbit, found by Metropolis, Stein and Stein \cite{Metropolis73},
50: or the Feigenbaum cascade \cite{Feigenbaum78,Feigenbaum79}, the accumulating
51: point of which was discovered by Myrberg \cite{Myrberg63}.
52:
53: As all quadratic maps are topologically conjugate \cite{Milnor88}
54: we can just focus the study in one of them. On the other hand, the
55: iterated one-dimensional maps, under rather general conditions, are
56: nearly quadratic if they are renormalized \cite{Gukenheimer87}. As
57: a consequence, the work done with the logistic equation can be generalized
58: to other maps. Therefore, any progress that makes clear the hierarchies
59: of bifurcations and the structure of this equation helps to the extension
60: of our knowledge of dynamical systems and the understanding of other
61: associated phenomena.
62:
63: Within the study of the logistic equation the identification and organization
64: of orbits is outstanding, that is, its organization and hierarchies
65: in more detailed structures which are self-similar. It is also relevant
66: to establish in which order they are created, and from which kind
67: of bifurcation, as well as, for which parameter values the orbits
68: prevail \cite{Sharkovsky64,Sharkovsky93,Li75,Zehg84}. In short, it
69: is a question of establishing the structure of periodic windows, a
70: question to which we want to contribute.
71:
72: The study of the logistic equation is also important in Physics, both
73: as a model \cite{Beck99} and because it helps to calculate magnitudes
74: which describe certain physics processes: Lyapunov exponents and topological
75: entropy. From an experimental point of view the spotting stable orbit
76: would allow the experimenter to locate himself in a stable region,
77: which is close to phenomena that he wants to observe: bifurcations
78: and several kinds of chaos. On the other hand, the periodic orbits
79: allow us a semiclassical approach of quantum mechanics \cite{Gutzwiller71}.
80: Therefore, the progress in the understanding, in its mathematical
81: aspect, of periodic orbits helps to the theoretical and experimental
82: development of Physics.
83:
84: Lately, an area of the study of application of logistic equation has
85: focused on the distribution of periodic windows, their widths, their
86: locations and its bifurcation diagram \cite{Hunt97,Hunt99}, clearly
87: linked to central question, above mentioned, of establishing which
88: is the complex structure of those windows. Similarly, we want to contribute
89: with our work, in which we will show a saddle-node bifurcation cascade.
90: This is a sequence of saddle-node bifurcations (tangent bifurcation),
91: with a recurrence law identical to the one Feigenbaum found for period
92: doubling cascade, which we will allow us to a) calculate the parameter
93: values where the windows are born and die, and therefore their length,
94: b) the identification and organization of orbits in windows, c) spot
95: the accumulating point of a saddle-node bifurcation cascade, in which
96: there will be chaos.
97:
98: The works by Feigenbaum \cite{Feigenbaum78,Feigenbaum79,Feigenbaum80}
99: and Hunt et al. \cite{Hunt99} are the cornerstones to get the above
100: mentioned points. The first one will be fundamental in the numerical,
101: geometrical and theoretical interpretation of the results, meanwhile
102: the second one will be so for the theoretical reconstruction of the
103: behavior of the logistic equation as shown in its numerical analysis.
104:
105: This paper is organized as follows:
106:
107: In section II numerical results are shown. These results suggest that
108: in an arbitrary period-$j$ window, the birth of successive period-$q*2^{n}$
109: saddle-node orbits scale in the same way as in a Feigenbaum cascade.
110: In general, inside the same period-$j$ window, there will be several
111: saddle-node bifurcation cascades with the same basic period $q$.
112:
113: In section III we get the theoretical explanation of these properties.
114: In the process we will find that not only the period doubling cascade
115: and saddle-node bifurcation cascade scale in the same way but also
116: that it is typical of other sequences which appear in the logistic
117: equation, such as accumulating points of Myrberg-Feigenbaum points.
118:
119: In section IV, and following the work by Feigenbaum for the period
120: doubling cascade, we develop a similar work for saddle-node cascade
121: which ends up in Feigenbaum-Cvitanoviç equation. It allows to explain
122: the self-similarity of iterated functions and to generalize the results
123: to polynomials with a maximum other than quadratic.
124:
125: The results of section IV allow to generalize the bifurcation rigidity
126: Principle \cite{Hunt99} which is done in section V.
127:
128: Section VI shows a rule to get the symbolic sequences of the saddle-node
129: bifurcation cascade orbits, as well as a rule to order some of them.
130:
131: In the conclusion section, some connections with a possible physical
132: model are shown, which will most likely make the phenomenon fit in
133: the framework described in this work.
134:
135:
136: \section{Numerical results}
137:
138: In the logistic map the period-$2^{n}$ orbits stem from pitchfork
139: bifurcation cascade: Feigenbaum cascade. However, the period-$q=3,5,7,...,2n+1,..$
140: orbits stem from saddle-node bifurcations (it is also possible to
141: find saddle-orbits in which $q$ is even, but not power of two).
142:
143: Let $f^{q}$ be the $q$-th iterated of $f$. When $f^{q}$, $f$
144: obtained from Eq. (1), has a saddle-node bifurcation $q$ saddle-node
145: fixed points are generated at the same time. The node and saddle points
146: pull away from each other when the control parameter is varied. Every
147: $q$ node point stands for an orbit, and each orbit will be, in turn,
148: subjected to a Feigenbaum cascade. As result of that period-$q*2^{n}$
149: orbits are born. However, there are period-$q*2^{n}$ orbits which
150: do not stem from a period-$q$ orbit displaying pitchfork bifurcation,
151: but from $f^{q*2^{n}}$ displaying saddle-node bifurcations.
152:
153: Whereas period-$q*2^{n}$ orbits duplicate their period because of
154: Feigenbaum cascade as the parameter $r$ is increased in Eq. (1),
155: contrary to what happens with saddle-node orbits. So, there is a period-$3$
156: saddle-node bifurcation at $r_{3}=1+\sqrt{8}$ \cite{Myrberg58} (see
157: Fig. \ref{cap:fig1}). If $r$ is increased, to get the point of pitchfork
158: bifurcation, a period-$3*2$ orbit is generated. A further increase
159: will generate a new pitchfork bifurcation and a period-$3*2^{2}$
160: orbit. A Feigenbaum cascade is taking place inside a period-$3$ window
161:
162: On the contrary if we decrease $r$ from $r_{3}=1+\sqrt{8}$ the period-$3$
163: orbit vanishes thus generating a saddle-node-$3*2$ period at $r_{3*2}\cong3.6265$
164: (see Fig. \ref{cap:fig2}). If $r$ is decreased further the saddle-node
165: period-$3*2$ orbit vanishes, and so a saddle-node period-$3*2^{2}$orbit
166: is generated at $r_{3*2^{2}}\cong3.5820$ (see Fig. \ref{cap:fig3}).
167: A period-$3,3*2,3*2^{2},..,3*2^{n},..$. saddle-node bifurcation cascade
168: takes place. Refer to Table \ref{cap:table1} for several birth saddle-node
169: orbit values. We can see that this saddle-node bifurcation cascade
170: is taking place at a canonical window, whereas pitchfork bifurcations
171: are taking place at a period-$3$window.
172:
173: Following Feigenbaum \cite{Feigenbaum78} let's define
174:
175: \[
176: \delta_{3*2^{n}}=\frac{r_{3*2^{n+1}}-r_{3*2^{n}}}{r_{3*2^{n+2}}-r_{3*2^{n+1}}}\]
177: and
178:
179: \[
180: \alpha_{3*2^{n}}=-\frac{d_{3*2^{n}}}{d_{3*2^{n+1}}}\]
181: where $r_{3*2^{n}}$ is the value of $r$ at which a period-$3*2^{n}$
182: orbit is created. And for this same saddle-node orbit we define $d_{3*2^{n}}$
183: as the distance from $x=\frac{1}{2}$ to the nearest saddle-node point.
184:
185: In Table \ref{cap:table1} estimates of $\alpha_{n}$ and $\delta_{n}$
186: for several values of $n$ are shown, which apparently approach the
187: $\delta=4.66920160..$ and $\alpha=2.50290787..$. Feigenbaum constants.
188: Similar results are reached if we repeat the process for the saddle-node
189: period-$5,5*2,5*2^{2},..,5*2^{n},..$ orbit, as shown in Table \ref{cap:table2}.
190:
191: The former two cases relate to saddle-node bifurcation cascade at
192: a canonical window, but the same kind of results are obtained working
193: with other windows. For instance, inside a period-$3$ window a saddle-node
194: bifurcation cascade can be generated, that is, period-$3,3*2,3*2^{2},..,3*2^{n},..$
195: orbits are created at this window. Nevertheless, it relates to period
196: $3*3,3*3*2,3*3*2^{2},..,3*3*2^{n},..$ orbits for the original map
197: Eq. (1). If this same cascade was generated at period-$5$window period-$5*3,5*3*2,5*3*2^{2},..,5*3*2^{n},..$
198: orbits would be created at a canonical window. In both cases, as shown
199: in Tables \ref{cap:table3} and \ref{cap:table4}, again we have the
200: same convergence to the $\delta$ Feigenbaum constant.
201:
202: Let's emphasize that the saddle-node period-$5*3$ orbit is a saddle-node
203: period-$3$ orbit which displays at a period-$5$window; different
204: from a saddle-node period-$15$ displayed at a canonical window, and
205: different from a saddle-node period-$3*5$ orbit, as well, the latter
206: being born as a saddle-node period-$5$ orbit at a period-$3$ window
207: .
208:
209: As well as numerical similarity with Feigenbaum's work, there is another
210: geometrical one. As shown in Figs. \ref{cap:fig1}, \ref{cap:fig2}
211: and \ref{cap:fig3} the graph of $f^{3}$(Fig. \ref{cap:fig1}) is
212: replicated in the neighborhood of $(\frac{1}{2},\frac{1}{2})$, for
213: any orbit of the period-$3,3*2,3*2^{2},..,3*2^{n},..$. saddle-node
214: bifurcation cascade. The same phenomenon can be noticed in other critical
215: points of $f^{2^{n}}$close to the line $y=x$. These geometrical
216: results apply not only to a $3,3*2,3*2^{2},..,3*2^{n},..$ cascade,
217: but also to $q,q*2,q*2^{2},..,q*2^{n},..$ $q\neq2^{m}$ cascades.
218:
219: The geometrical, as well as the numerical, phenomenon apply not only
220: to a canonical window, as shown in Fig. \ref{cap:fig4}, where the
221: $3,3*2,3*2^{2},..,3*2^{n},..$ cascade is at the period-$5$window.
222:
223: Summarizing, what the numerical results suggest is that for a given
224: saddle-node bifurcation, where a period-$q\neq2^{m}$ orbit is created,
225: there is a saddle-node bifurcation cascade where period-$q*2^{n}$
226: orbits are created. Let $r_{q*2^{n}}$ $n=0,1,2,3,...$ be the value
227: of $r$ at which a period-$q*2^{n}$ saddle-node bifurcation is created.
228: As $r$ is decreased from $r_{q}$ the different bifurcation values
229: appear as
230:
231: \[
232: ...<r_{q*2^{n}}<...<r_{q*2}<r_{q}\]
233: resulting in
234:
235: \begin{equation}
236: \delta=\lim_{n\rightarrow\infty}\delta_{q*2^{n}}=\lim_{n\rightarrow\infty}\frac{r_{q*2^{n+1}}-r_{q*2^{n}}}{r_{q*2^{n+2}}-r_{q*2^{n+1}}}\label{nueva 2}\end{equation}
237: and
238:
239: \begin{equation}
240: \alpha=\lim_{n\rightarrow\infty}\alpha_{q*2^{n}}=\lim_{n\rightarrow\infty}\left(-\frac{d_{q*2^{n}}}{d_{q*2^{n+1}}}\right)\label{nueva 3}\end{equation}
241: where $\alpha,\delta$ are Feigenbaum's constants, and $d_{q*2^{n}}$
242: the distance from $x=\frac{1}{2}$ to the nearest saddle-point at
243: $r=r_{q*2^{n}}$.
244:
245: As well as in the Feigenbaum cascade, when three consecutive values
246: $r_{q*2^{bin-3}},r_{q*2^{n-2}},r_{q*2^{n-1}}$are given the following
247: bifurcation value $r_{q*2^{n}}$ can be approximately predicted. This
248: result is important because the values where saddle-node bifurcation
249: appear are the values where period-$q*2^{n}$ windows are born. Therefore,
250: we can spot the birth of period-$q*2^{n}$ windows, once we have found
251: three consecutive bifurcation values. In addition, it is possible
252: to know how the lower endpoints of such periodic windows scale.
253:
254: The Eq. (\ref{nueva 2}) and (\ref{nueva 3}) allow us to calculate
255: what it is happening in the non-primary windows, because each period-$q*2^{n}$
256: window, created in a saddle-node bifurcation, necessarily mimics the
257: canonical window.
258:
259:
260: \section{Generalization of Feigenbaum's formulas}
261:
262:
263: \subsection{Saddle-Node bifurcation cascade}
264:
265: Following Feigenbaum work, let be the sequence
266:
267: \begin{equation}
268: r_{1},r_{2},r_{3},.....,r_{n}.....\label{2 nueva 4}\end{equation}
269: made up by values of $r$, for which Eq. (\ref{eq:recurrencia}) shows
270: a pitchfork bifurcation. If the sequence is ordered in such a way
271: that a period-$j*2^{n}$ orbit is created at $r_{n}$, being $j$
272: the original orbit period, according to Feigenbaum \cite{Feigenbaum78}
273: there exist the limits
274:
275: \begin{equation}
276: \delta=\lim_{n\rightarrow\infty}\frac{r_{n+1}-r_{n}}{r_{n+2}-r_{n+1}}=\lim_{n\rightarrow\infty}\delta_{n}\label{delta-F}\end{equation}
277:
278:
279: \[
280: r_{\infty}=\lim_{n\rightarrow\infty}r_{n}\]
281: From the former limits, the expression
282:
283: \begin{equation}
284: r_{n+1}=\frac{1}{\delta}r_{n}+(1-\frac{1}{\delta})r_{\infty}\label{3 nueva 5}\end{equation}
285: follows, valid for $n>N$, being $N$ big enough.
286:
287: For further development it is necessary to write Eq. (\ref{3 nueva 5})
288: in terms of the bounds of an interval and we can do so by using a
289: period-$j$ window originating from a saddle-node bifurcation. $r=r_{j,ini}$
290: is the lower bound of the window and $r=r_{j,end}$ the upper bound.
291: The window has a Myrberg-Feigenbaum point $r_{\infty,j}$, where a
292: period-$j*2^{n}$cascade finishes. Then, as Eq. (\ref{3 nueva 5})
293: there is an $r_{N,j}$ so that for every $r_{n,j}$ accomplishing
294: $r_{N,j}\leq r_{n,j}\leq r_{\infty,j}$ the equation
295:
296: \begin{equation}
297: r_{n+1,j}=\frac{1}{\delta}r_{n,j}+(1-\frac{1}{\delta})r_{\infty,j}\label{4 nueva 6}\end{equation}
298: is accomplished.
299:
300: With this posing, the lineal Eq. (\ref{3 nueva 5}) turns to lineal
301: Eq. (\ref{4 nueva 6}), which is accomplished in the subset $\left[r_{n,j};r_{\infty,j}\right]\subset\left[r_{j,ini};r_{j,end}\right]$.
302:
303: We know that bifurcation points $r_{n,j}$ of doubling-period cascade
304: scale as Eq. (\ref{4 nueva 6}). We are going to prove that bifurcation
305: points of saddle-node bifurcation cascade, which have been encountered
306: numerically, escalate as Eq. (\ref{4 nueva 6}) too, that is
307:
308: \[
309: r_{q*2^{n+1},SN,j}=\frac{1}{\delta}r_{q*2^{n},SN,j}+(1-\frac{1}{\delta})r_{\infty,j}\]
310: at $r_{q*2^{n},SN,j}$ the function $f^{j}$ bifurcate to a saddle-node
311: period-$q*2^{n}$ orbit.
312:
313: To get this result the Eq. (\ref{4 nueva 6}) must extend to interval
314: $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$. Accordingly let's take the
315: interval $\left[r_{n,j};r_{q*2^{n},SN,j}\right]\subset\left[r_{j,ini};r_{j,end}\right]$,
316: which holds the interval $\left[r_{n,j};r_{\infty,j}\right]$ where
317: Eq. (\ref{4 nueva 6}) is accomplished.
318:
319: The key of proof is to chose two intervals $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$
320: and$\left[r_{n+1,j};r_{q*2^{n+1},SN,j}\right]$, whose bounds are
321: bifurcation points of $f^{j}$. Both intervals will be related linearly.
322: The Eq. (\ref{4 nueva 6}) shows that the lower bound of an interval
323: is linearly transformed in the lower bound of the other interval.
324: By the uniqueness of linear transformation between both intervals,
325: the upper bounds are linearly transformed by Eq. (\ref{4 nueva 6}).
326:
327: Starting from \cite{Yorke85}, Hunt \emph{et al.} \cite{Hunt99} demonstrated
328: the following principle:.
329:
330: \emph{Principle 1: Let $X$ and $\Lambda$ be compact intervals. For
331: a typical $C^{3}$ family maps $f:X\times\Lambda\rightarrow X$ and
332: a typical superstable period $n$ orbit with critical point ($x_{0}$,$\lambda_{0}$),
333: there is a linear change of coordinates that conjugates $f^{n}$near
334: ($x_{0}$,$\lambda_{0}$) to the quadratic family $y\longmapsto y^{2}-c$
335: in the square $[-2.5,\,2.5]\times[-2.5,\,2.5]$ to within an error
336: $\varepsilon$ in the $C^{2}$ norm, where $\varepsilon\rightarrow0$
337: as $n$ increases.}
338:
339: Principle 1 shows that close to critical point the bifurcation diagram
340: for a window can be linearly transformed into a canonical bifurcation
341: diagram.
342:
343: Let's see Fig. \ref{cap:fig4}. The first element of the $3*2^{m}$
344: saddle-node cascade is shown. In the neighborhood of $x=\frac{1}{2}$
345: of Fig. \ref{cap:fig4} the graph of $f^{3}$ is seen (Fig. \ref{cap:fig1}).
346: The graph of $f^{5}$ has been superimposed, which has a parabolic
347: shape in the neighborhood of $x=\frac{1}{2}$. If we expanded $f^{5}$
348: in the neighborhood of critical point $x=\frac{1}{2}$ it would turn
349: into a quadratic polynomial, which if iterated three times it would
350: look like a graph of $f^{3}$ as shown in Fig. \ref{cap:fig4}. The
351: same happens for all five critical points of $f^{5}$, each one of
352: which would have a graph of $f^{3}$ tangent to line $y=x,$and the
353: result would be a period-$3*5=15$ orbit in the canonical window.
354:
355: Principle 1 suggests that the bifurcation diagram (of quadratic approximation
356: of $f^{5}$ close to critical point) is linearly transformed into
357: canonical bifurcation diagram. This linear transformation turns value
358: $r$ , which accounts for graph of $f^{3}$in Fig. \ref{cap:fig4},
359: into value $r$ which accounts for graph of $f^{3}$ in Fig. \ref{cap:fig1}.
360:
361: As said before we formulate the following principle:
362:
363: \textbf{Principle}.- In a period-$j$ window, saddle-node bifurcation
364: cascades scale by
365:
366: \[
367: r_{q*2^{n+1},SN,j}=\frac{1}{\delta}r_{q*2^{n},SN,j}+(1-\frac{1}{\delta})r_{\infty,j}\]
368:
369:
370: Proof.
371:
372: If the bifurcation diagram of $f^{j}$in $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$
373: and the $f^{2*j}$in $\left[r_{n,2*j};r_{q*2^{n},SN,2*j}\right]$
374: are considered, according to Principle 1 there will be
375:
376: i) A linear transformation that maps bifurcation diagram of $f^{j}$
377: in $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$ onto canonical bifurcation
378: diagram of $f$ in $\left[r_{n,1};r_{q*2^{n},SN,1}\right]$, let this
379: be $h_{j}$. At $r_{n,1}$ the $n$th pitchfork bifurcation of $f$
380: occurs. At $r_{q*2^{n},SN,1}$a period-$q*2^{n}$ saddle-node bifurcation
381: of $f$ occurs.
382:
383: ii) a linear transformation that maps bifurcation diagram of $f^{2*j}$in
384: $\left[r_{n,2*j};r_{q*2^{n},SN,2*j}\right]$ onto bifurcation diagram
385: of $f$ in $\left[r_{n,1};r_{q*2^{n},SN,1}\right]$, let this be $h_{2j}$.
386:
387: Therefore, in the parameter space, there will be a linear transformation
388: $h_{2j}^{-1}\circ h_{j}$ that maps the bifurcation diagram of $f^{j}$
389: in $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$ onto bifurcation diagram
390: of $f^{2*j}$ in $\left[r_{n,2*j};r_{q*2^{n},SN,2*j}\right]$. As
391: the bifurcation diagram of $f^{2*j}$in $\left[r_{n,2*j};r_{q*2^{n},SN,2*j}\right]$
392: coincides with bifurcation diagram of $f^{j}$ $\left[r_{n+1,j};r_{q*2^{n+1},SN,j}\right]$,
393: the result is that the linear transformation $h_{2j}^{-1}\circ h_{j}$
394: maps the bifurcation diagram of $f^{j}$ of $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$
395: onto the bifurcation diagram of $f^{j}$in $\left[r_{n+1,j};r_{q*2^{n+1},SN,j}\right]$.
396:
397: Let's notice how Eq. (\ref{4 nueva 6}) is a linear transformation
398: that maps the interval $\left[r_{n,j};r_{\infty,j}\right]$ onto $\left[r_{n+1,j};r_{\infty,j}\right]$.
399: As the linear transformation is unique, the linear transformation
400: $h_{2j}^{-1}\circ h_{j}$ within the limits of $\left[r_{n,j};r_{\infty,j}\right]\subset\left[r_{n,j};r_{q*2^{n},SN,j}\right]$
401: must coincide with Eq. (\ref{4 nueva 6}), i.e., $r_{n+1,j}=h_{2j}^{-1}\circ h_{j}(r_{n,j})=\frac{1}{\delta}r_{n,j}+(1-\frac{1}{\delta})r_{\infty,j}$.
402: However, the linear transformation is accomplished on the whole interval
403: $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$and not only on the $\left[r_{n,j};r_{\infty,j}\right]\subset\left[r_{n,j};r_{q*2^{n},SN,j}\right]$
404: where a pitchfork bifurcation cascade occurs. By applying the same
405: lineal transformation to upper bound of interval the expression
406:
407: \begin{equation}
408: r_{q*2^{n+1},SN,j}=\frac{1}{\delta}r_{q*2^{n},SN,j}+(1-\frac{1}{\delta})r_{\infty,j}\label{5 nueva 7}\end{equation}
409: results.
410:
411: For that reason it is demonstrated that bifurcation parameter values
412: in saddle-node bifurcation cascade are governed by the same relation
413: that governs Feigenbaum cascade as shown in Eq. (\ref{3 nueva 5}).
414: Accordingly, the convergence on $\delta$of Feigenbaum
415:
416: \[
417: \delta=\lim_{n\rightarrow\infty}\delta_{q*2^{n}}=\lim_{n\rightarrow\infty}\frac{r_{q*2^{n+1}}-r_{q*2^{n}}}{r_{q*2^{n+2}}-r_{q*2^{n+1}}}\]
418: suggested by numerical results is satisfied.
419:
420: Obviously a saddle-node bifurcation cascade in a canonical window
421: escalates in the same way as in a period-$j$ window, as a result
422: of the linear relation between the bifurcation diagram of one window
423: and the other according to Principle 1.
424:
425:
426: \subsection{Non-uniqueness of Saddle-Node bifurcation cascade}
427:
428: Most of the times, $f^{j}$ will have saddle-node bifurcations not
429: only for a single value of $r$ but for several (although $f^{3}$
430: has only one saddle-node bifurcation). Therefore, the interval $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$
431: used to obtain the Eq. (\ref{5 nueva 7}) will not be unique, as there
432: will be different values $r_{q*2^{n},SN,j}$, with fixed $q,n,j$,
433: for which $f^{j}$ has saddle-node bifurcation. That means, there
434: will be different values $r_{q*2^{n},SN,j}$, with fixed $q,n,j$,
435: accomplishing the same Eq. (\ref{5 nueva 7}). Therefore, the saddle-node
436: bifurcation cascade that escalates as Eq. (\ref{5 nueva 7}) will
437: not be unique either. Although saddle-node orbits will have the same
438: period $q*2^{n}$.
439:
440: For instance, $f^{6}$ shows three different $6*2^{n}$ saddle-node
441: bifurcation cascades, and none of them is related with the $3*2^{n+1}$
442: saddle-node cascade, although all of them have orbits with the same
443: periods. As shown in Table \ref{cap:table1}, it is obvious that $f^{6}$
444: has a saddle-node bifurcation at $r\cong3.6265$, but it must not
445: be understood that a $6*2^{n}$ saddle-node bifurcation cascade starts
446: at this point, but it must be considered as the second element of
447: the $3*2^{n}$ cascade, since the graph of $f^{3}$ appears in the
448: neighborhood of the critical points of $f^{2}$nearest to line $y=x$
449: (see Fig. \ref{cap:fig2}). The following element of the $3*2^{n}$cascade
450: occurs at $r\cong3.5820$ (see Fig. \ref{cap:fig3}), where the same
451: figure is repeated $2^{2}$ times. However, $f^{6}$ has saddle-node
452: bifurcation at $r_{1}\cong3.9375$, $r_{2}\cong3.9779$ and $r_{3}\cong3.9976$,
453: and it is impossible for $f^{3}$ to have saddle-node bifurcation
454: at these points. In each of all three former points a $6*2^{n}$ saddle-node
455: cascade starts. The following element of the respective cascades must
456: reproduce the graph of $f^{6}$ and duplicate its period, which occur
457: at $r_{a}\cong3.6552$, $r_{b}\cong3.6684$ and $r_{c}\cong3.6767$
458: respectively. Obviously, for these three new values $f^{12}$ has
459: a saddle-node bifurcation, but as said before, it must not be understood
460: as the beginning of a $12*2^{n}$ cascade, because the graph of $f^{6}$
461: is repeated twice. In Fig. \ref{cap:fig6} and \ref{cap:fig7} two
462: elements of $6*2^{n}$ cascade are shown, corresponding to $r_{2}$
463: and $r_{b}$.
464:
465:
466: \subsection{Other objects scaling with Feigenbaum's relation}
467:
468: The reasoning followed to show the saddle-node bifurcation cascade
469: scaling can be used with other {}``objects'' . To do so, such {}``objects''
470: must belong to a sequence, and later build intervals with the elements
471: of the sequence in a similar way to the proof of a saddle-node bifurcation
472: cascade.
473:
474: We can probe it with Myrberg-Feigenbaum points. To do so, let us consider
475: the period-$q*2^{n}$window $n=0,1,2,..$ , born of a primary saddle-node
476: orbit. Each window will have a Myrberg-Feigenbaum point $r_{\infty,q*2^{n}}$,
477: where the Feigenbaum cascade finishes.
478: Let us build the sequence
479:
480: \begin{equation}
481: r_{\infty,q,j},r_{\infty,q*2,j},...,r_{\infty,q*2^{n},j}\label{nueva 8}\end{equation}
482: that represents the parameter values of Eq. (\ref{eq:recurrencia}),
483: in which every one is the end of a Feigenbaum cascade in period-$q,q*2,q*2^{2},..,q*2^{n},..$
484: windows respectively, that is , Myrberg-Feigenbaum points. Period-$q,q*2,q*2^{2},..,q*2^{n},..$
485: windows are inside period-$j$ window, because of this the subindex
486: $j$ in Eq. (\ref{nueva 8}). This sequence is vital for the proof.
487:
488: In proof of the saddle-node cascade convergence let $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$
489: and $\left[r_{n,2*j};r_{q*2^{n},SN,2*j}\right]$ be substituted by$\left[r_{n,j};r_{\infty,q*2^{n},j}\right]$
490: and $\left[r_{n,2*j};r_{\infty,q*2^{n},2*j}\right]$. Initially $r_{q*2^{n},SN,j}$
491: used to indicate the value of parameter for which $f^{j}$ had a period-$q*2^{n}$
492: saddle-node orbit. Such orbit will have a Myrberg-Feigenbaum at some
493: value of $r$, which we represent by $r_{\infty,q*2^{n},j}$. This
494: is the value that appears in the new intervals, and also in the sequence
495: (\ref{nueva 8}).
496:
497: In order to demonstrate that the Myrberg-Feigenbaum points are governed
498: by
499:
500: \begin{equation}
501: r_{\infty,q*2^{n+1},j}=\frac{1}{\delta}r_{\infty,q*2^{n},j}+(1-\frac{1}{\delta})r_{\infty,j}\label{6 nueva 9}\end{equation}
502: $n>N$ , being $N$ large enough, we simply need to rebuild the saddle-node
503: cascade demonstration in the initial way, but with new intervals.
504:
505: Let's expound what Eq. (\ref{6 nueva 9}) means. There will be saddle-node
506: period-$q*2^{n}$orbits inside a period-$j$ window born of a period-$j$
507: saddle-node orbit. After the birth of these orbits at $r_{q*2^{n},SN,j}$a
508: Feigenbaum cascade occurs, which will finish at Myrberg-Feigenbaum
509: point $r_{\infty,q*2^{n},j}$. Then, the Eq. (\ref{5 nueva 7}), which
510: interrelates birth-point of period-$q*2^{n}$ orbits, is identical
511: to the equation which interrelates the end-points where doubling-period
512: cascades of those same orbits finish.
513:
514: The Eq. (\ref{6 nueva 9}) brings more information about Myrberg-Feigenbaum
515: points. It is well known that the Myrberg-Feigenbaum point $r_{\infty,j}$
516: is the accumulation point of values $r$ where pitchfork bifurcations
517: occur, because a doubling-period cascade finishes at $r_{\infty,j}$.
518: However, the Myrberg-Feigenbaum point $r_{\infty,j}$is not only the
519: accumulation point of Feigenbaum cascade because the Eq. (\ref{6 nueva 9})
520: shows that $r_{\infty,j}$ is the accumulation point of Myrberg-Feigenbaum
521: points $r_{\infty,q*2^{n},j}$,that is, Myrberg-Feigenbaum point $r_{\infty,j}$
522: is the accumulation point of Myrberg-Feigenbaum points of the period-$q*2^{n}$
523: $q\neq2^{m}$ primary windows, which have been born inside period-$j$
524: windows. As Myrberg-Feigenbaum points are attractors what we are in
525: front of is an attractor of attractors of attractors of ...
526:
527: The problem of the convergence has not yet been completed. Let's notice
528: that at $r_{\infty,j}$ not only a single sequence of Myrberg-Feigenbaum
529: points convergences, but infinite ones, because for each $q\neq2^{m}$
530: there is an associated sequence $q*2^{n}$ $n=0,1,2,3,...$, and for
531: each sequence the same happens. The whole process is repeated for
532: each value of $q\neq2^{m}$.
533:
534: The reasoning followed with Myrberg-Feigenbaum points can be applied
535: to other {}``objects''. The reader cannot miss that if instead of
536: choosing Myrberg-Feigenbaum points $r_{\infty,q*2^{n},j}$, where
537: a Feigenbaum cascade finishes, we choose the point $r_{q*2^{n},end,j}$
538: where the window started at $r_{q*2^{n},SN,j}$finishes then equation
539:
540: \begin{equation}
541: r_{q*2^{n+1}end,j}=\frac{1}{\delta}r_{q*2^{n},end,j}+(1-\frac{1}{\delta})r_{\infty,j}\label{7 nueva 10}\end{equation}
542: $n>N$ , being $N$ large enough, will be the same.
543:
544: Notice that three demonstration have been carried out for three different
545: points of a same window respectively, that is, $r_{q*2^{n},SN,j}$,
546: $r_{\infty,q*2^{n},j}$ and $r_{q*2^{n},end,j}$, initial point, intermediate
547: point and final point.
548:
549: From the equations found we can answer part of the questions initially
550: posed: {}``where windows begin and end'', the answers to which are
551: given by Eq. (\ref{5 nueva 7}) and (\ref{7 nueva 10}). From these
552: results the relation between the length $L_{n}$of the period-$q*2^{n}$
553: windows is obtained, because by subtracting Eq. (\ref{5 nueva 7})
554: and (\ref{7 nueva 10})
555:
556: \[
557: L_{n+1=}r_{q*2^{n+1},end,,j}-r_{q*2^{n+1},SN,j}=\frac{1}{\delta}r_{q*2^{n},end,,j}-\frac{1}{\delta}r_{q*2^{n},SN,j}=\frac{1}{\delta}L_{n}\]
558: is obtained, that is,
559:
560: \begin{equation}
561: \delta=\frac{L_{n}}{L_{n+1}}\label{8 nueva 11}\end{equation}
562:
563:
564: This equation simply means that the quotient of the length of two
565: intervals which are linearly transformed one onto other must be constant.
566: The constant $\delta$ must be the proportional factor in the lineal
567: map, or else, how the lengths of the two consecutive windows born
568: of saddle-node bifurcation cascade are contracted. For the same reason,
569: the equation
570:
571: \begin{equation}
572: \delta=\frac{r_{q*2^{n},end,j}-r_{q*2^{n},M,j}}{r_{q*2^{n+1},end,j}-r_{q*2^{n+1},M,j}}\label{9 nueva 12}\end{equation}
573: applies.
574:
575: Let's bear in mind that we are talking about period-$q*2^{n}$ windows,
576: which scales in the same period-$j$ window, and not about how a period-$j$
577: window scales with respect to a canonical window. Therefore there
578: is no conflict with other theoretical results \cite{Hunt97,Yorke85}.
579:
580:
581: \subsection{Generalized Feigenbaum's relation}
582:
583: Eqs. (\ref{5 nueva 7}), (\ref{6 nueva 9}) and (\ref{7 nueva 10})
584: are identical to the Eq. (\ref{3 nueva 5}), and they are valid for
585: $r>r_{\infty}$, meanwhile Eq. (\ref{3 nueva 5}) is valid for $r<r_{\infty}$.
586: We can express all these results in only one equation and generalize
587: the Feigenbaum equation (Eq. (\ref{3 nueva 5})) , by writing
588:
589: \begin{equation}
590: r_{n+1,j}=\frac{1}{\delta}r_{n,j}+(1-\frac{1}{\delta})r_{\infty,j}\label{10 nueva 13}\end{equation}
591: where points $r_{n,j}$ show the elements of a sequence (pitchfork
592: bifurcation, saddle bifurcation, Myrberg-Feigenbaum point, supercycle
593: or any other element to which the proof of saddle-node bifurcation
594: applies).
595:
596: The Eq. (\ref{10 nueva 13}), apart from being a generalization, informs
597: us about the order in which the different elements of the sequence
598: appear. The order will depend on the parameter value $r_{n,j}$ being
599: smaller or bigger than $r_{\infty,j}$.
600:
601: From the Eq. (\ref{10 nueva 13}) we can deduce that if $r_{n,j}\prec r_{\infty,j}$
602: then $r_{n,j}\prec r_{n+1,j}\prec r_{\infty,j}$, on the contrary
603: if $r_{n,j}\succ r_{\infty,j}$ then $r_{\infty,j}\prec r_{n+1,j}\prec r_{n,j}$.
604: From which we deduce that if a bifurcation occurs for a value such
605: that $r_{n,j}\prec r_{\infty,j}$ then the next value of bifurcation
606: increases, in the same way as for pitchfork bifurcation. On the contrary,
607: when the bifurcation occurs for $r_{n,j}\succ r_{\infty,j}$ then
608: the next values of bifurcation decreases, in the same way as for saddle-node
609: bifurcation cascades.
610:
611: In this ordering, part of these results are new. If we consider first-occurrence
612: orbits the former results turn out to be a particular case of a theorem
613: (Theorem 2.10 in \cite{Sharkovsky93}). However, the ordering shown
614: in this section deals with period-$j$ window, which does not have
615: to originate from first-occurrence period-$j$ windows. For this reason,
616: the ordering is not restricted to first-occurrence orbits and therefore
617: it is not regarded in the Sharkovsky theorem.
618:
619: On the other hand, the ordering that we have just shown is not restricted
620: only to orbits, as the Sharkovsky theorem, but it also applies to
621: Myrberg-Feigenbaum points and to other objects that are not orbits.
622: An ordering which was hidden as far as we know.
623:
624: Therefore the Eq. (\ref{10 nueva 13}) enlarges the ordering to both
625: new orbits and other objects which are not orbits.
626:
627: Let's summarize the conclusions obtained in this section for a period-$j$window:.
628:
629: i) It is possible to spot the birth of a saddle-node period-$q*2^{n}$
630: orbit and therefore to fix the birth of period-$q*2^{n}$ windows
631: from Eq. (\ref{5 nueva 7}). This equation justifies the numerical
632: results shown in Tables \ref{cap:table1}, \ref{cap:table2}, \ref{cap:table3}
633: and \ref{cap:table4}, where the convergence to Feigenbaum $\delta$constant
634: is shown.
635:
636: ii) Equally it spots where a period-$q*2^{n}$ window finishes from
637: Eq. (\ref{7 nueva 10})
638:
639: iii) According to the former two points the scaling of windows related
640: to saddle-node bifurcation cascades is proved Eq. (\ref{8 nueva 11}).
641:
642: iv) The Feigenbaum equation (Eq. (\ref{3 nueva 5})) has been generalized,
643: both in the range of the parameter validity and in the objects to
644: which is applied
645:
646: v) In particular, the generalization of Eq. (\ref{3 nueva 5}) is
647: applied to Myrberg-Feigenbaum points, showing that these points behave
648: as an attractor of attractors.
649:
650: vi) The generalization, applied to Myrberg-Feigenbaum points, allows
651: us to spot where the Feigenbaum cascade of period-$q*2^{n}$ window
652: finishes by means of Eq. (\ref{6 nueva 9}), or else, it shows the
653: relative position of the attractors in which doubling-period cascades
654: finish.
655:
656: vii) The ordering of the objects governed by Eq. (\ref{10 nueva 13})
657: is shown, and if they happen to be orbits they do not have to be first-occurrence
658: orbits.
659:
660: In this way, we account for the question initially posed, about the
661: birth, end, scaling and structure of the orbits.
662:
663:
664: \section{Self-similarity and convergence to Feigenbaum's $\alpha$ constant}
665:
666:
667: \subsection{Self-similarity}
668:
669: In the numerical results we anticipated that the graph of $f^{3}$(Fig.
670: \ref{cap:fig1}), when it has a saddle-node bifurcation, is repeated
671: in the neighborhood of point $(\frac{1}{2},\frac{1}{2})$ both $f^{3\cdot2}$
672: and $f^{3\cdot2^{2}}$, when $f^{3\cdot2}$ and $f^{3\cdot2^{2}}$have
673: also saddle-node bifurcation. Furthermore, the graph of $f^{3}$ in
674: the neighborhood of point $(\frac{1}{2},\frac{1}{2})$ appears contracted
675: by a factor $\frac{1}{\alpha}$, each time $f^{3}$is iterated with
676: itself, where $\alpha$is a Feigenbaum constant.
677:
678: Let's justify the numerical results, as well as why similar figures
679: are shown at the $2^{n}$ critical points of $f^{2^{n}}$near line
680: $y=x$ (see Fig. \ref{cap:fig3}) These similar figures are like saddle-node
681: orbits trapped in the critical points, and we will use them for the
682: mathematical analysis.
683:
684: A saddle-node period-$q*2^{n}$ orbit is obtained when the derivative
685: of $f^{q*2^{n}}$ takes value 1. This orbit is forced to a period
686: doubling process, a process that will occur when the derivative takes
687: value -1. Accordingly, the derivative vanishes in some middle point,
688: which corresponds to a supercycle. Therefore, we can associate each
689: saddle-node orbit with a supercycle.
690:
691: Feigenbaum \cite{Feigenbaum78} introduced the quotient
692:
693: \begin{equation}
694: \delta_{n}=\frac{R_{n+1}-R_{n}}{R_{n+2}-R_{n+1}}\label{nueva 14}\end{equation}
695: where $R_{n}$is the value for which Eq. (\ref{eq:recurrencia}) has
696: a period-$2^{n+1}$ supercycle in a period doubling cascade. The quotient
697: (Eq. (\ref{nueva 14})) approaches a$\delta$ Feigenbaum constant.
698:
699: Let's follow Feigenbaum and define
700:
701: \begin{equation}
702: \delta_{n}=\frac{r_{n+1}-r_{n}}{r_{n+2}-r_{n+1}}\label{nueva 15}\end{equation}
703: being $r_{n}$the value where the period-$q*2^{n}$ supercycle occurs.
704: Such supercycle is associated to a period-$q*2^{n}$ saddle-node orbit.
705: Then we will get the same results as Feigenbaum: the convergence of
706: the quotient (Eq. (\ref{nueva 15})) on $\delta$ constant (see Table
707: \ref{cap:table1}). This result leads us to the definition of a family
708: of functions similar to those created by Feigenbaum in his work
709:
710: As for Eqs. (\ref{nueva 14}) and (\ref{nueva 15}), we have to emphasize
711: that we will work with supercycles associated to orbits born of saddle-node
712: bifurcations instead of supercycles associated to orbits born of pitchfork
713: bifurcation as Feigenbaum does. There is an important difference with
714: regard to the behavior of both kinds of orbits in the parameter space.
715: The orbits of period doubling cascade are adjacent in a connected
716: set: an interval of the parameter space. However, the saddle-node
717: period-$q*2^{n}$ orbits only exist in discrete points of the parameter
718: space. Besides, saddle-node orbits that do not belong to the same
719: cascade can be found between saddle-node period-$q*2^{n}$and period-
720: $q*2^{n+1}$ orbits, which proves that the latter orbits are not adjacent.
721: Let us mention other essential difference between both bifurcations:
722: the pitchfork bifurcation has a local character, whereas the Saddle-Node
723: bifurcation leads to a global qualitative change in the behavior of
724: the logistic equation.
725:
726: Bearing in mind the previous explanation and following Feigenbaum,
727: we define ( see Eq. (35) of \cite{Feigenbaum80})
728:
729: \[
730: g_{1,q}=\lim_{n\rightarrow\infty}\left(-\alpha\right)^{n}f_{R_{n+1,q}}^{2^{n}}\left[\frac{x}{\left(-\alpha\right)^{n}}\right]\]
731: where the supercycle associated to saddle-node period-$q*2^{n}$ orbit
732: is obtained at $R_{n,q}$, and $q$ is fixed.
733:
734: From previous definition we define a family of functions (see Eq.
735: (39) of \cite{Feigenbaum80})
736:
737: \begin{equation}
738: g_{i,q}=\lim_{n\rightarrow\infty}\left(-\alpha\right)^{n}f_{R_{n+i,q}}^{2^{n}}\left[\frac{x}{\left(-\alpha\right)^{n}}\right]\label{11 nueva 16}\end{equation}
739: where $q$ is fixed, which fulfills the next equation (see Eq. (42)
740: of \cite{Feigenbaum80})
741:
742: \begin{equation}
743: g_{i-1,q}=\left(-\alpha\right)g_{i,q}\left[g_{i,q}\left(-\frac{x}{\alpha}\right)\right]\equiv Tg_{i,q}(x)\label{12 nueva 17}\end{equation}
744: Taking the limit in Eq. (\ref{11 nueva 16}) (see Eq. (40) of \cite{Feigenbaum80})
745: the function converges to a limiting function
746:
747: \begin{equation}
748: g(x)=\lim_{i\rightarrow\infty}g_{i,q}\label{13 nueva 18}\end{equation}
749:
750:
751: This limit exists at a fixed value of $r$, the Myrberg-Feigenbaum
752: $r_{\infty}$, and it means that there is an accumulation point at
753: $r_{\infty}$ for a saddle-node $q*2^{n}$bifurcation cascade.
754:
755: It follows from Eqs. (\ref{12 nueva 17}) and (\ref{13 nueva 18})
756: that
757:
758: \begin{equation}
759: g(x)=-\alpha g\left[g\left(-\frac{x}{\alpha}\right)\right]\label{14 nueva 19}\end{equation}
760: (see Eq. (43) of \cite{Feigenbaum80})
761:
762: It follows from Eq. (\ref{14 nueva 19}) that if $g(x)$ is the solution
763: of this equation then $g^{q}(x)$ is the solution of
764:
765: \begin{equation}
766: g^{q}(x)=-\alpha g^{q}\left[g^{q}(-\frac{x}{\alpha})\right]\label{15 nueva 20}\end{equation}
767:
768:
769: This equation means that the iterated of $g^{q}$ is self-similarly
770: scaled by $\alpha$. Precisely, this is shown in Fig. \ref{cap:fig1},
771: \ref{cap:fig2} and \ref{cap:fig3}, in the neighborhood of $\left(\frac{1}{2},\frac{1}{2}\right)$:
772: the graph of $f^{3}$, close to a saddle-node bifurcation, is repeated
773: scaled by a constant. This constant is the nearer to $\alpha$ the
774: bigger $n$ is in $f^{3*2^{n}}$, if $f^{3*2^{n}}$ has a saddle-node
775: bifurcation (see Table \ref{cap:table1}). With this last reasoning
776: all numerical results shown in section 2 have been verified.
777:
778: The Eq. (\ref{14 nueva 19}), that we have just found, is the Feigenbaum-Cvitanoviç
779: equation, which Feigenbaum utilized to show why function $f$ is repeated
780: in the neighborhood of $(\frac{1}{2},\frac{1}{2})$ in a period doubling
781: cascade. If so, we wonder, why the original work of Feigenbaum is
782: not utilized to justify the self-similarity of $f^{3}$in a saddle-node
783: bifurcation cascade. The reason lies in the fact that Feigenbaum reached
784: Eq. (\ref{14 nueva 19}) from a limit of functions associated with
785: pitchfork bifurcations, and accordingly its validity is not justified
786: when we work with functions associated with other kinds of bifurcations,
787: as is the case with saddle-node bifurcations.
788:
789: This is not the only difference. It has been said that Eq. (\ref{14 nueva 19})
790: is the limit of Eq. (\ref{12 nueva 17}) only at $r=r_{\infty}$,
791: and Feigenbaum got this value on the left, i.e., with $r<r_{\infty}$
792: which is how a period doubling cascade occurs. However we have got
793: $r_{\infty}$ on the right, as saddle-node bifurcations occur for
794: $r>r_{\infty}$. That is, the Eq. (\ref{14 nueva 19}) can be reached
795: by limits of functions defined for both $r<r_{\infty}$and $r>r_{\infty}$,
796: and that explains why Eq. (\ref{10 nueva 13}) generalizes Eq. (\ref{3 nueva 5})
797: of Feigenbaum, as we will see later.
798:
799:
800: \subsection{Derivation of the generalized Feigenbaum formula.}
801:
802: Feigenbaum \cite{Feigenbaum79} got the Eq. (\ref{3 nueva 5}) from
803: the equation
804:
805: \begin{equation}
806: g_{i+1}(x)-g_{i}(x)=\delta^{-i}h(x)\,\,\,\, i\gg1\label{rel-F}\end{equation}
807: where $g_{i}(x)$ is the family of functions that we have utilized
808: to define the family $g_{i,q}$ given by Eq. (\ref{11 nueva 16}).
809: $g_{i}(x)$ and $g_{i,q}$ are formally identical, but $g_{i}(x)$
810: refers to pitchfork bifurcations and $g_{i,q}$ refers to saddle-node
811: bifurcations. In order to define well the Eq. (\ref{rel-F}) it is
812: necessary to point out what $h(x)$ and $\delta$ are, which are calculated
813: from
814:
815: \begin{equation}
816: \delta h(x)=L_{g}h(x)\label{F-C-lineal}\end{equation}
817: where\begin{equation}
818: L_{g}h(x)=-\alpha\left\{ g'\left[g(\frac{-x}{\alpha})\right]h(\frac{-x}{\alpha})+h\left[g(\frac{-x}{\alpha})\right]\right\} \label{oper-linea}\end{equation}
819: where $\alpha$ and $g(x)$ are given by the Eq. (\ref{14 nueva 19}).
820:
821: If we follow Feigenbaum, working with $g_{i,q}$ instead of $g_{i}$,
822: we will calculate an equation identical to Eq. (\ref{rel-F}), i.e.,
823:
824: \begin{equation}
825: g_{i+1,q}(x)-g_{i,q}(x)=\delta^{-i}h(x)\,\,\,\, i\gg1\label{rel-Jesus}\end{equation}
826: where $g_{i,q}$ is the family of functions given by Eq. (\ref{11 nueva 16}),
827: whereas both $h(x)$ and $\delta$ are given by Eqs. (\ref{F-C-lineal})
828: and (\ref{oper-linea}). As both $g_{i}(x)$ and $g_{i,q}$ are identical
829: and both Eqs. (\ref{rel-F}) and (\ref{rel-Jesus}), which govern
830: the former functions, are also identical then the result has to be
831: the same: the Eq. (\ref{3 nueva 5}). Although, this time the equation
832: refers to a saddle-node bifurcations for $r>r_{\infty}$instead of
833: a pitchfork bifurcation cascade for $r<r_{\infty}$, as Feigenbaum
834: reached, that is, the Eq. (\ref{3 nueva 5}) is valid for both $r>r_{\infty}$
835: and $r<r_{\infty}$. The thing is that depending on the fact that
836: wether $r$ is smaller or bigger than $r_{\infty}$, the equation
837: is assigned to one bifurcation or another.
838:
839: The result of the former explanation is that Eq. (\ref{10 nueva 13}),
840: which generalizes Eq. (\ref{3 nueva 5}), is obtained without the
841: bifurcation rigidity Principle of Hunt et al. \cite{Hunt99}, as it
842: was done in section 3.1, and it will be the starting point to generalize
843: the bifurcation rigidity Principle in section 5.
844:
845: Now we have obtained the equation which governs bifurcation points
846: in saddle-node bifurcation cascades. We must distinguish two cases:
847: saddle-node bifurcation cascades with the same or different basic
848: periods. In a $q,q*2,q*2^{2},..,q*2^{n},..$ $q\neq2^{m}$ saddle-node
849: bifurcation cascade $q$ will be named a basic period.
850:
851:
852: \subsubsection{Saddle-Node bifurcation cascades with the same basic period}
853:
854: As shown in section 3, there are different saddle-node bifurcation
855: cascades with the same basic period in the canonical window, which
856: are generated because the same period basic $q$ is produced for different
857: values of parameter $r$. Accordingly, there are different values
858: $R_{n,q}$ , with fixed $q$ and $n$, and therefore the family of
859: functions given by Eq. (\ref{11 nueva 16}) are not the only ones,
860: but there are as many families as there are values of $R_{n,q}$ (see
861: sections 3.2 and 4.1). Given that there are different $g_{i,q}$families
862: then there are different equations like Eq. (\ref{rel-Jesus}). As
863: Eq. (\ref{3 nueva 5}) is obtained from Eq. (\ref{rel-Jesus}) it
864: turns out that there will be one Eq. (\ref{3 nueva 5}) for each saddle-node
865: bifurcation cascade. All equations will be identical and all of them
866: will approach to $r_{\infty}$, because the limit given by Eq. (\ref{14 nueva 19})
867: exists only for $r_{\infty}$ which does not depend on value of the
868: $R_{n,q}$. What we have just found is something already known: for
869: each fixed period $q$ there are different saddle-node bifurcation
870: cascades, each one of which is governed by the Eq. (\ref{3 nueva 5}),
871: and all of them approach to $r_{\infty}$(see section 3.2).
872:
873:
874: \subsubsection{Saddle-Node bifurcation cascades with different basic period}
875:
876: For each basic period there will be one saddle-node bifurcation cascade,
877: and therefore a family of functions $g_{i,q}$. Following the explanation
878: of section 4.2.1 there will be an equation like Eq. (\ref{3 nueva 5})
879: for each bifurcation cascade. All equations will be identical and
880: all of them will approach to $r_{\infty}$.
881:
882: We have already found this result in section 3 (see Eq. (\ref{5 nueva 7})).
883:
884:
885: \subsection{Extension to non-canonical windows}
886:
887: Until now we have only been working in the canonical window, but it
888: is not difficult to extend the results to any window. For instance,
889: if we had worked in a period-$p$ window then a $q\cdot2^{n}$ saddle-node
890: bifurcation cascade would have occurred inside it. It would be enough
891: to take into account that the $R_{n,q}$ of Eq. (\ref{11 nueva 16})
892: is the value for which the supercycle associated to a saddle-node
893: period-$q*2^{n}$ orbit occurs, but this time inside a period-$p$
894: window, so that the whole process is identical and the same universal
895: results are obtained. The difference would lie in the fact that the
896: approaching to the Eq. (\ref{14 nueva 19}) would be at a Myrberg-Feigenbaum
897: point $r_{\infty,p}$ of a period-$p$ window instead of at the Myrberg-Feigenbaum
898: point $r_{\infty}$ of the canonical window. Therefore, if $r_{\infty}$
899: were substituted by $r_{\infty,p}$ all equations valid in the canonical
900: window would remain valid in a period-$p$ window.
901:
902:
903: \subsection{Extension to functions with a non-quadratic maximum}
904:
905: As we have generalized the results from a canonical window to an arbitrary
906: periodic window, one more generalization remains to be done: the results
907: are generalized to maps other than logistic ones.
908:
909: In this paper, the work has been developed with Eq. (\ref{eq:recurrencia}),
910: which is topologically conjugate to
911:
912: \[
913: x_{n+1}=x_{n}^{2}-c\]
914:
915:
916: The whole work could have been developed with the function
917:
918: \[
919: f(x,c)=x^{2}-c\]
920: and it could have been repeated in an identical way with the family
921: of functions
922:
923: \begin{equation}
924: f_{n}(x,c)=x^{2n}-c\,\,\,\,\, n=1,2,3,...\label{f-conj}\end{equation}
925:
926:
927: The sections 4.1 and 4.2 can be rewritten, step by step, defining
928: a new family of functions $g_{i,q}$, which are identical to the ones
929: given by Eq. (\ref{11 nueva 16}), but this time $f$ is replaced
930: by Eq. (\ref{f-conj}). This new family of functions applies to an
931: equation identical to Eq. (\ref{12 nueva 17}), and it results in
932: Eq. (\ref{14 nueva 19}), which has a solution \cite{Epstein89},
933: once the maximum of $g(x)$has been fixed as $x^{2n}$ , $n=1,2,3,...$,
934: although $\alpha$ and $\delta$ will be different, depending on the
935: kind of maximum.
936:
937: As the universal behavior comes from Eq. (\ref{14 nueva 19}), and
938: this remains identical, it turns out that $g^{q}$is self-similar,
939: although we will have a different $g(x)$ depending on the kind of
940: maximum we choose. We will see how the graph of functions $x^{2n}$
941: is repeated in the saddle-node bifurcation cascade.
942:
943: The results relative to saddle-node bifurcation cascades remain the
944: same, because these results came from Eqs. (\ref{rel-Jesus}), (\ref{F-C-lineal})
945: and (\ref{oper-linea}), which did not change after $f$ was replaced
946: by Eq. (\ref{f-conj}) in $g_{i,q}$. As everything is identical it
947: turns out that equations identical to Eq. (\ref{3 nueva 5}) will
948: be obtained, but the value of $\delta$ will depend on the kind of
949: maximum with which we are dealing with.
950:
951: In short, in this section we have expound the convergence to Feigenbaum
952: constant $\alpha$ and the self-similarity of the function $f^{q}$
953: in the neighborhood of $\left(\frac{1}{2},\frac{1}{2}\right)$ as
954: it was shown in section 2. We have also laid the foundations for a
955: generalization of the Feigenbaum Eq. (\ref{3 nueva 5}) and, at last,
956: we have generalized these results to functions with maximum like $x^{2n}$
957: , $n=1,2,3,...$.
958:
959:
960: \section{Generalization of Bifurcation Rigidity Principle}
961:
962: We have reached Eq. (\ref{3 nueva 5}) (see section 3.1) coming from
963: the bifurcation rigidity Principle \cite{Hunt99} for functions with
964: a quadratic maximum. The proof can be followed the other way round,
965: starting from Eq. (\ref{3 nueva 5}) and reaching the bifurcation
966: rigidity Principle. If we start from Eq. (\ref{3 nueva 5}), valid
967: for functions like Eq. (\ref{f-conj}), we will arrive to the bifurcation
968: rigidity Principle. This time it will be valid for functions like
969: Eq. (\ref{f-conj}) and not only for functions with quadratic maximum
970: as it was initialed expounded.
971:
972: Let a function be like Eq. (\ref{f-conj}). According to section 4.4,
973: we know that both successive pitchfork and successive saddle-node
974: bifurcations are governed by an equation like
975:
976: \begin{equation}
977: r_{n+1}=\frac{1}{\delta}r_{n}+(1-\frac{1}{\delta})r_{\infty}\label{20}\end{equation}
978:
979:
980: In period-$j$ window successive saddle-node bifurcations will be
981: governed by
982:
983: \begin{equation}
984: r_{q*2^{n+1},SN,j}=\frac{1}{\delta}r_{q*2^{n},SN,j}+(1-\frac{1}{\delta})r_{\infty,j}\label{21}\end{equation}
985: meanwhile in a canonical window the successive saddle-node bifurcations
986: will be governed by
987:
988: \begin{equation}
989: r_{q*2^{n+1},SN}=\frac{1}{\delta}r_{q*2^{n},SN}+(1-\frac{1}{\delta})r_{\infty}\label{22}\end{equation}
990:
991:
992: Accordingly there is a linear transformation between both saddle-node
993: bifurcation cascades given by
994:
995: \begin{equation}
996: \frac{r_{q*2^{n+1},SN,j}-\frac{1}{\delta}r_{q*2^{n},SN,j}}{r_{q*2^{n+1},SN}-\frac{1}{\delta}r_{q*2^{n},SN}}=\frac{r_{\infty,j}}{r_{\infty}}\label{23}\end{equation}
997:
998:
999: For pitchfork bifurcations there is an identical equation
1000:
1001: \begin{equation}
1002: \frac{r_{n+1,j}-\frac{1}{\delta}r_{n,j}}{r_{n+1}-\frac{1}{\delta}r_{n}}=\frac{r_{\infty,j}}{r_{\infty}}\label{24}\end{equation}
1003: where $n$ is related to $n$-th pitchfork bifurcation.
1004:
1005: Let's take the intervals
1006:
1007: $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$ and $\left[r_{n+1,j};r_{q*2^{n+1},SN,j}\right]$\\
1008: in the period-$j$ window and the intervals
1009:
1010: $\left[r_{n};r_{q*2^{n},SN}\right]$ and $\left[r_{n+1};r_{q*2^{n+1},SN}\right]$\\
1011: in the canonical window. If a linear transformation maps the interval
1012: $\left[r_{n,j};r_{q*2^{n},SN,j}\right]$ onto the interval $\left[r_{n};r_{q*2^{n},SN}\right]$then
1013: Eqs. (\ref{23}) and (\ref{24}) will force the interval $\left[r_{n+1,j};r_{q*2^{n+1},SN,j}\right]$
1014: to be linearly transformed into the $\left[r_{n+1};r_{q*2^{n+1},SN}\right]$.
1015: Accordingly the bifurcation diagrams of both window are linearly changed
1016: as set by the bifurcation rigidity Principle, but this time the Principle
1017: works with $x^{2n}$-maps and not only with quadratic maps. As the
1018: Eq. (\ref{20}) is applied to $n\gg1$, the smaller the windows are,
1019: the better the linear approximation to the transformation will be.
1020:
1021:
1022: \section{Symbolic sequences and orbit ordering}
1023:
1024: \textbf{Sequence principle}.- Let be a period-$p*(q*2^{n})$ saddle-node
1025: orbit. The points of this orbit lie inside the $p*2^{n}$ neighborhoods
1026: of the $p*2^{n}$ supercycle points of $f^{p*2^{n}}$, each one of
1027: these neighborhoods having $q$ saddle-node points. Then the sequence
1028: of the saddle-node orbit is obtained with the following process
1029:
1030: i) Write the sequence of the orbit of the supercycle of $f^{p*2^{n}}$,
1031: which will be like $CI_{1}.....I_{p*2^{n}-1}$, where $C$ indicates
1032: the center and represents the point $x=\frac{1}{2}$ and $I_{i}$
1033: , $i=1,...,p*2^{n}-1$ can take the values L (left) or R (right) with
1034: respect to $C$.
1035:
1036: ii) Write consecutively $q$ times the sequence obtained in the former
1037: point i), getting a sequence like
1038:
1039: $C_{1}I_{1}.....I_{p*2^{n}-1}C_{2}I_{1}.....I_{p*2^{n}-1}.........C_{q}I_{1}.....I_{p*2^{n}-1}$.
1040:
1041: iii) Write the sequence of period-$q$ saddle-node orbit, that is,
1042: the sequence of the saddle-node orbit of $f^{q}$.
1043:
1044: iv) Calculate $\left(-1\right)^{n}$. If the result is negative then
1045: conjugate the letters obtained in point iii) by means of L$\leftrightarrow$R.
1046: Bear in mind that $n\in Z^{+}$.
1047:
1048: v) Replace consecutively each letter $C_{1}C_{2}.....C_{q}$ of the
1049: sequence obtained in ii) by the complete sequence obtained in iv),
1050: keeping the order.
1051:
1052: Proof
1053:
1054: i) Let be period-$p*2^{n}$ supercycle at $r=r_{0}$, in the period-$p$
1055: window, whose points are $\left\{ x_{1},...x_{p*2^{n}}\right\} $.
1056: The order in which these points are visited determines the sequence
1057: of the supercycle orbit of $f^{p*2^{n}}$, that is, $CI_{1}.....I_{p*2^{n}-1}$.
1058:
1059: ii) The function $f^{p*2^{n}}$can be approximated by a quadratic
1060: polynomial $g(x)$ in the neighborhood of each point of the supercycle
1061: $\left\{ x_{1},...x_{p*2^{n}}\right\} $ \cite{Hunt99}.
1062:
1063: The period-$p*(q*2^{n})$ saddle-node orbit of $f$ occurs when $g(x)$
1064: has a period-$q$ saddle-node orbit. This period-$q$ saddle-node
1065: orbit is like $\left\{ x_{0},g(x_{0}),.....,g^{q-1}(x_{0})\right\} $,
1066: with $g^{q}(x_{0})=x_{0}$. As $g(x)$ is an approximation of $f^{p*2^{n}}$to
1067: pass from point $g^{i}(x_{0})$ to point $g^{i+1}(x_{0})$ we will
1068: have to iterate $p*2^{n}$ times the function $f$, that is, the iterated
1069: of $f$ visit the neighborhood of the points of the supercycle $\left\{ x_{1},...x_{p*2^{n}}\right\} $.
1070: With each $p*2^{n}$iterated, the same neighborhood is visited again,
1071: not to the same point but the following one of the period-$q$ saddle-node
1072: orbit trapped in this neighborhood.
1073:
1074: With each $p*2^{n}$ iterated of $f$ we go back to the same neighborhood
1075: of a given point of the supercycle. This neighborhood will be to the
1076: left or to the right of the center $C$, or it will be the point $C$
1077: itself. For the time being, let's leave the neighborhood of the point
1078: $C$ and let's consider the period-$q$ saddle-node orbits trapped
1079: in the other neighborhoods. If this orbit is trapped in a neighborhood
1080: to the left (right) of the center $C$ then all its points will be
1081: placed to the left (right) of $C$ and they will generate an L (R)
1082: in the sequence of the period-$p*(q*2^{n})$ saddle-node orbit. Therefore,
1083: writing consecutively $q$ times the sequence of the orbit of the
1084: supercycle we would obtain the sequence of the period-$p*(q*2^{n})$
1085: orbit.
1086:
1087: iii) A drawback of the previous procedure takes place when the neighborhood
1088: of the point $C$ is visited. For in this neighborhood not all the
1089: points of the period-$q$ saddle-node orbit are to the left or to
1090: the right of the point $C$. Apparently, to solve the problem, it
1091: would be enough to calculate the sequence of the period-$q$ saddle-node
1092: orbit, and with this sequence to replace the $C_{1}C_{2}.....C_{q}$
1093: of the sequence $C_{1}I_{1}.....I_{p*2^{n}-1}C_{2}I_{1}.....I_{p*2^{n}-1}.........C_{q}I_{1}.....I_{p*2^{n}-1}$.
1094: This is so because each time the neighborhood of $C$ is visited a
1095: point of the period-$q$ saddle-node orbit trapped in this neighborhood
1096: is visited.
1097:
1098: iv) However there is a drawback as the letters $C_{1}C_{2}.....C_{q}$are
1099: replaced by the the letters of period-$q$ saddle-node orbit, because
1100: the period-$q$ saddle-node orbit is conjugated in the neighborhood
1101: of $C$, as the value of $n$ is in $f^{p*2^{n}}$. This is so because
1102: of the negative sign in Eq. (\ref{15 nueva 20}). If $n$ is even
1103: the sequence holds and if $n$ is odd the sequence is conjugated.
1104: Therefore to solve the problem $\left(-1\right)^{n}$ is calculated
1105: and if this value is negative the sequence of the orbit is conjugated.
1106:
1107: v) Finally, it is enough to replace the letters $C_{1}C_{2}.....C_{q}$
1108: by the letters of the period-$q$ saddle-node orbit sequence correctly
1109: conjugated to obtain the right result.
1110:
1111: Notice that it is enough to know the sequences of the supercycles
1112: of $f^{2^{n}}$and the primary saddle-node orbits $f^{p}$ to obtain
1113: the sequence of each orbit in the logistic map. According to the rule
1114: that we show here we get the sequence of the supercycle of $f^{p*2^{n}}$in
1115: the period-$p$ window, and from this supercycle we calculate the
1116: sequence of the period-$p*(q*2^{n})$ saddle-node orbit that will
1117: occur in a period-$q$ window inside a period-$p$ window.
1118:
1119: Let's see how the principle is used to obtain the period-$3*2$ saddle-node
1120: orbit sequence of Fig. \ref{cap:fig2}, in which $p=1,n=1,q=3$.
1121:
1122: i) This time there is a supercycle of $f^{2}$, whose orbit is formed
1123: by the points $\left\{ \frac{1}{2},f(\frac{1}{2})\right\} $and its
1124: own sequence is $CR$.
1125:
1126: ii) There is a period-$3$ saddle-node orbit which is trapped in the
1127: neighborhoods of $\frac{1}{2}$ and $f(\frac{1}{2})$, accordingly
1128: we write $CRCRCR$
1129:
1130: iii) The sequence of the period-$3$ saddle-node is $RRL$, as shown
1131: in Fig. \ref{cap:fig1}
1132:
1133: iv) As $\left(-1\right)^{1}=-1$ the sequence $RRL$ is conjugated
1134: to give $LLR$
1135:
1136: v) Finally, the letters $C$ of the sequence $CRCRCR$ are replaced
1137: by the letters of the orbit $LLR$, turning into $LRLRRR$. The first
1138: letter indicates a place with respect to $C$, the point $x=\frac{1}{2}$.
1139: In Fig. \ref{cap:fig2} it can be observed that the saddle-node orbit
1140: goes through as the sequence obtained .
1141:
1142: The principle of ordering shown above, for saddle-node orbits, is
1143: very similar to the one described by Derrida \emph{et al.} in \cite{Derrida79}
1144: for supercycle orbits.
1145:
1146: \textbf{Principle of saddle-node orbits} \textbf{ordering}.- Let be
1147: $f$ the logistic map, if $q>s$ then the period-$qp2^{n}$saddle-node
1148: orbit occurs at a value $r$ smaller than that for which the period-$sp2^{n}$saddle-node
1149: orbit occurs, that is, the period $qp2^{n}$ precedes period $sp2^{n}$
1150: in the parameter space. We note it down like $qp2^{n}\triangleright sp2^{n}$
1151: if $q>s$.
1152:
1153: Proof
1154:
1155: As the function $f^{p*2^{n}}$can be approximated by a quadratic polynomial
1156: $g(x)$ in the neighborhood of each point of the supercycle $\left\{ x_{1},...x_{p*2^{n}}\right\} $
1157: \cite{Hunt99}, it will turn out that as $g(x)$ has a period-$s$
1158: saddle-node orbit then $f$ will have a period-$sp2^{n}$ orbit ,
1159: and as $g(x)$ has a period-$q$ saddle-node orbit then $f$ will
1160: have a period-$qp2^{n}$ orbit. Since for the saddle-node $q\triangleright s$
1161: if $q>s$ so $qp2^{n}\triangleright sp2^{n}$ if $q>s$.
1162:
1163: Comments:
1164:
1165: i) Although we have a Sharkovsky ordering the principle is not limited
1166: to first-occurrence orbits to which the Sharkovsky theorem applies.
1167:
1168: ii) We cannot discard halfway orbits in this ordering, which come
1169: from other values of $n$, $p$ or $q$, that is, we have ordered
1170: non-first-occurrence orbits althugh all of them.
1171:
1172:
1173: \section{Conclusion}
1174:
1175: The Feigenbaum equation
1176:
1177: \[
1178: r_{n+1}=\frac{1}{\delta}r_{n}+(1-\frac{1}{\delta})r_{\infty}\]
1179: which allows us to locate the successive pitchfork bifurcations in
1180: a period doubling cascade, turns out to be an expression of a general
1181: nature. This is applied to cascades of other objects present in the
1182: logistic equation. For instance, this expression rules, among other
1183: things, saddle-node bifurcations, Myrberg-Feigenbaum points, and the
1184: birth and death of some windows. That is, not only the equation gives
1185: the ordering but also the parameter values for which such objects
1186: appear.
1187:
1188: An immediate consequence of these results is the posibility to determine:
1189: the scaling of windows, the parameters of birth and death, the relative
1190: position of the Myrberg-Feigenbaum points, the relative position of
1191: the attractors in which the period doubling cascade finishes.
1192:
1193: We would like to highlight another consequence of these results, a
1194: new concept: an attractor of attractors, which coincide with the Myrberg-Feigenbaum
1195: points.
1196:
1197: The former results have been first expounded with linear transformations
1198: benefiting from the properties of the windows (bifurcation rigidity
1199: Principle), and later, part of theses results have been derived using
1200: the linearization of the Feigenbaum-Cvitanoviç equation. Later the
1201: bifurcation rigidity Principle is generalized to functions with a
1202: maximum other than quadratic.
1203:
1204: The concept upon which the whole work relies is the saddle-node bifurcation
1205: cascade, which is a sequence of successive saddle-node bifurcations,
1206: which in turn double the number of saddle-node fixed points. The bifurcation
1207: points are ruled by the same equation obtained by Feigenbaum for the
1208: period doubling cascade. By using the bifurcation parameter given
1209: by saddle-node bifurcation cascade, we follow the Feigenbaum work
1210: to end up in a Feigenbaum-Cvitanoviç equation, which proves the universal
1211: behavior of our results, and it allows us the generalization to maps
1212: other than the logistic one.
1213:
1214: The saddle-node bifurcation cascade shows a further behavior similar
1215: to a period doubling cascade: it is self-similar after successive
1216: iterations. Nonetheless, for a fixed period that is duplicated in
1217: each bifurcation, there are different saddle-node bifurcations, however
1218: this they have the same accumulating point and the same scaling law.
1219:
1220: The ordering of orbits missing from Sharkovsky theorem have been given,
1221: as well as the ordering of Myrberg-Feigenbaum points associated to
1222: the windows born of saddle-node bifurcation cascades.
1223:
1224: Lastly symbolic sequences of the orbits found in the saddle-node bifurcation
1225: cascade have been described.
1226:
1227: We have summarized above the mathematical contributions of our work.
1228: We started the introduction by pointing out that any progress, in
1229: the logistic equation, would also yield progress to Physics . We would
1230: like to quote {}``\textbf{the self-similar cascade of bifurcations}
1231: (....) \textbf{it is characterized by an infinite series of saddle-node
1232: bifurcation of cycles, accumulating at a finite parameter value}''
1233: reported by Yeung and Strogatz \cite{Yeung00}. This mechanism coincides
1234: with the one described for a saddle-node bifurcation cascade, where
1235: the accumulating parameter value is the Myrberg-Feigenbaum point,
1236: which will most likely make the phenomenon fit in the framework described
1237: in this work.
1238:
1239:
1240: \section*{Acknowledgments}
1241:
1242: The author wishes to thank Daniel Rodríguez-Pérez for helpful discussions
1243: and help in the preparation of the manuscript.
1244:
1245:
1246: \newpage
1247:
1248: %
1249: \begin{table}[h]
1250:
1251: \caption{\label{cap:table1}Period $3\cdot2^{n}$ Saddle-Node bifurcation
1252: cascade inside the canonical window. The convergence to $\alpha$
1253: and $\delta$ Feigenbaum constants is shown.}
1254:
1255: \begin{center}\begin{tabular}{|c|c|c|c|c|}
1256: \hline
1257: Period&
1258: $r_{3*2^{n}}$&
1259: $\delta_{3*2^{n}}$&
1260: $\alpha_{3*2^{n}}$&
1261: $n$\tabularnewline
1262: \hline
1263: \hline
1264: $3$&
1265: 3.82842712&
1266: &
1267: &
1268: 0\tabularnewline
1269: \hline
1270: $3*2$&
1271: 3.62655316&
1272: &
1273: 2.30066966&
1274: 1\tabularnewline
1275: \hline
1276: $3*2^{2}$&
1277: 3.58202300&
1278: 4.5334294&
1279: 2.52112183&
1280: 2\tabularnewline
1281: \hline
1282: $3*2^{3}$&
1283: 3.57253281&
1284: 4.69223061&
1285: 2.49753642&
1286: 3\tabularnewline
1287: \hline
1288: $3*2^{4}$&
1289: 3.57049971&
1290: 4.66784221&
1291: 2.50264481&
1292: 4\tabularnewline
1293: \hline
1294: $3*2^{5}$&
1295: 3.57006433&
1296: 4.66971381&
1297: 2.50263906&
1298: 5\tabularnewline
1299: \hline
1300: \end{tabular}\end{center}
1301: \end{table}
1302:
1303:
1304: %
1305: \begin{table}[h]
1306:
1307: \caption{\label{cap:table2}Period $5\cdot2^{n}$ Saddle-Node bifurcation
1308: cascade inside the canonical window. The convergence to Feigenbaum
1309: $\delta$ constant is shown.}
1310:
1311: \begin{center}\begin{tabular}{|c|c|c|c|}
1312: \hline
1313: Period&
1314: $r_{5*2^{n}}$&
1315: $\delta_{5*2^{n}}$&
1316: n\tabularnewline
1317: \hline
1318: \hline
1319: $5$&
1320: 3.73817238&
1321: &
1322: 0\tabularnewline
1323: \hline
1324: $5*2$&
1325: 3.60520807&
1326: &
1327: 1\tabularnewline
1328: \hline
1329: $5*2^{2}$&
1330: 3.57751225&
1331: 4.80088006&
1332: 2\tabularnewline
1333: \hline
1334: $5*2^{3}$&
1335: 3.57156559&
1336: 4.65737406&
1337: 3\tabularnewline
1338: \hline
1339: $5*2^{4}$&
1340: 3.57029262&
1341: 4.67144810&
1342: 4\tabularnewline
1343: \hline
1344: $5*2^{5}$&
1345: 3.57001998&
1346: 4.66925870&
1347: 5\tabularnewline
1348: \hline
1349: \end{tabular}\end{center}
1350: \end{table}
1351:
1352:
1353: %
1354: \begin{table}[h]
1355:
1356: \caption{\label{cap:table3}Period $3\cdot2^{n}$ Saddle-Node bifurcation
1357: cascade inside the period-3 window. The convergence to Feigenbaum
1358: $\delta$ constants is shown.}
1359:
1360: \begin{center}\begin{tabular}{|c|c|c|c|}
1361: \hline
1362: Period&
1363: $r_{3*2^{n}}$&
1364: $\delta_{53*2^{n}}$&
1365: n\tabularnewline
1366: \hline
1367: \hline
1368: $3$&
1369: 3.85361311&
1370: &
1371: 0\tabularnewline
1372: \hline
1373: $3*2$&
1374: 3.85031470&
1375: &
1376: 1\tabularnewline
1377: \hline
1378: $3*2^{2}$&
1379: 3.84962047&
1380: 4.75117756&
1381: 2\tabularnewline
1382: \hline
1383: $3*2^{3}$&
1384: 3.84947362&
1385: 4.72747702&
1386: 3\tabularnewline
1387: \hline
1388: $3*2^{4}$&
1389: 3.84944223&
1390: 4.67824148&
1391: 4\tabularnewline
1392: \hline
1393: $3*2^{5}$&
1394: 3.84943551&
1395: 4.67113095&
1396: 5\tabularnewline
1397: \hline
1398: \end{tabular}\end{center}
1399: \end{table}
1400:
1401:
1402: %
1403: \begin{table}[h]
1404:
1405: \caption{\label{cap:table4}Period $3\cdot2^{n}$ Saddle-Node bifurcation
1406: cascade inside the period-5 window. The convergence to Feigenbaum
1407: $\delta$ constants is shown.}
1408:
1409: \begin{center}\begin{tabular}{|c|c|c|c|}
1410: \hline
1411: Period&
1412: $r_{3*2^{n}}$&
1413: $\delta_{3*2^{n}}$&
1414: n\tabularnewline
1415: \hline
1416: \hline
1417: $3$&
1418: 3.744003&
1419: &
1420: 0\tabularnewline
1421: \hline
1422: $3*2$&
1423: 3.74321655&
1424: &
1425: 1\tabularnewline
1426: \hline
1427: $3*2^{2}$&
1428: 3.743050198&
1429: 4.727522&
1430: 2\tabularnewline
1431: \hline
1432: $3*2^{3}$&
1433: 3.7430150831&
1434: 4.737633&
1435: 3\tabularnewline
1436: \hline
1437: $3*2^{4}$&
1438: 3.7430075758&
1439: 4.677634&
1440: 4\tabularnewline
1441: \hline
1442: $3*2^{5}$&
1443: 3.7430059688&
1444: 4.671437&
1445: 5\tabularnewline
1446: \hline
1447: \end{tabular}\end{center}
1448: \end{table}
1449:
1450: \newpage
1451:
1452: \begin{figure}[h]
1453:
1454: \caption{\label{cap:fig1}Period-$3$ Saddle-Node orbit in the canonical window,
1455: for $r\simeq3.82842712$.}
1456:
1457: \includegraphics[%
1458: width=0.70\textwidth,
1459: keepaspectratio,
1460: angle=-90]{figs/I/fig1.eps}
1461: \end{figure}
1462: \newpage
1463:
1464:
1465: %
1466: \begin{figure}[h]
1467:
1468: \caption{\label{cap:fig2}Period-$3\cdot2$ Saddle-Node orbit in the canonical
1469: window, for $r\simeq3.62655316$.}
1470:
1471: \includegraphics[%
1472: width=0.70\textwidth,
1473: keepaspectratio,
1474: angle=-90]{figs/I/fig2.eps}
1475: \end{figure}
1476:
1477: \newpage
1478:
1479: %
1480: \begin{figure}[h]
1481:
1482: \caption{\label{cap:fig3}Period-$3\cdot2^{2}$ Saddle-Node orbit in the canonical
1483: window, for $r\simeq3.58202300$.}
1484:
1485: \includegraphics[%
1486: width=0.70\textwidth,
1487: keepaspectratio,
1488: angle=-90]{figs/I/fig3.eps}
1489: \end{figure}
1490:
1491: \newpage
1492:
1493: %
1494: \begin{figure}[h]
1495:
1496: \caption{\label{cap:fig4}Period-$3$ Saddle-Node orbit in the period-$5$
1497: window, for $r\simeq3.74400300$. The shape of Fig. \ref{cap:fig1}
1498: is reproduced five times; here is shown a magnification of the replica
1499: near $(1/2,\,1/2)$.}
1500:
1501: \includegraphics[%
1502: width=0.70\textwidth,
1503: keepaspectratio,
1504: angle=-90]{figs/I/fig4.eps}
1505: \end{figure}
1506:
1507: \newpage
1508:
1509: %
1510: \begin{figure}[h]
1511:
1512: \caption{\label{cap:fig5}Period-$3\cdot2$ Saddle-Node orbit in the period-$5$
1513: window, for $r\simeq3.74321655$. The shape of Fig. \ref{cap:fig2}
1514: is reproduced five times; here is shown a magnification of the replica
1515: near $(1/2,\,1/2)$.}
1516:
1517: \includegraphics[%
1518: width=0.70\textwidth,
1519: keepaspectratio,
1520: angle=-90]{figs/I/fig5.eps}
1521: \end{figure}
1522:
1523:
1524: \newpage
1525:
1526: \begin{figure}[h]
1527:
1528: \caption{\label{cap:fig6}One of the three possible period-$6$ Saddle-Node
1529: orbits close to $r=3.97790000$.}
1530:
1531: \includegraphics[%
1532: width=0.70\textwidth,
1533: keepaspectratio,
1534: angle=-90]{figs/I/fig6.eps}
1535: \end{figure}
1536:
1537: \newpage
1538: %
1539: \begin{figure}[h]
1540:
1541: \caption{\label{cap:fig7}Period-$6\cdot2$ Saddle-Node orbit in the canonical
1542: window, close to $r\simeq3.66840000$. The shape of Fig. \ref{cap:fig6}
1543: is reproduced twice.}
1544:
1545: \includegraphics[%
1546: width=0.70\textwidth,
1547: keepaspectratio,
1548: angle=-90]{figs/I/fig7.eps}
1549: \end{figure}
1550:
1551: \newpage
1552:
1553:
1554: \begin{thebibliography}{10}
1555: \bibitem{Metropolis73}N. Metropolis, M. L. Stein, and P. R. Stein, On Finite Limit Sets
1556: for Transformations on the Unit Interval. Journal of Combinatorial
1557: Theory 15 (1973) 25-44
1558: \bibitem{Feigenbaum78}M. J. Feigenbaum, Quantitative Universal for a Class of Nonlinear
1559: Transformations. Journal of Statistical Physics, 19 (1978) 25-52
1560: \bibitem{Feigenbaum79}M. J. Feigenbaum, The universal Metric Properties of Nonlinear Transformations.
1561: Journal of Statistical Physics, 21 (1979) 669-706
1562: \bibitem{Myrberg63}P. J. Myrberg, Iteration der Reellen Polynome Zweiten Grades III.
1563: Ann. Acad. Sci. Fenn. Ser. AI 336(3) (1963) 1-18
1564: \bibitem{Milnor88}J. Milnor and W. Thurston, On iterated Maps of the interval. 465-563
1565: Lect. Notes in Math. 1342 Springer, Berlin, (1988)
1566: \bibitem{Gukenheimer87}J. Gukenheimer, Renormalization of one dimension mappings. Contemp.
1567: Math 58, pt. III (1987) 143-160
1568: \bibitem{Sharkovsky64}A. N. Sharkovsky, Coexistence of cycles of a continuous map of a line
1569: into itself. Ukrain. Math. Zhurn, 16 (1964) 61-71
1570: \bibitem{Sharkovsky93}A. N. Sharkovsky, Yu L. Maistrenko, E. Yu. Romanenko, \emph{Difference
1571: Equations and Their Applications}. Kluwer Academic Publishers. London,
1572: 1993
1573: \bibitem{Li75}T. Y. Li and J. A. Yorke, Period three implies chaos. Am. Math. Mon.
1574: 82 (1975), 985-992
1575: \bibitem{Zehg84}W-H Zehg, B-L Hao, G-R Wang, S-G Gheng, Scaling property of period-n-tupling
1576: sequences in one-dimensional mappings. Commun. Theor. Phys. Vol.3
1577: Nº. 3 (1984) 283-295
1578: \bibitem{Beck99}C. Beck, Physical meaning for Mandelbrot and Julia Sets. Physica D
1579: 125 (1999) 171-182
1580: \bibitem{Gutzwiller71}M. C. Gutzwiller, Periodic Orbits and Classical Quantization Conditions.
1581: J. Math. Phys. 12 (1971) 343-358
1582: \bibitem{Hunt97}B. R. Hunt and E. Ott, Structure in the Parameter Dependence of Order
1583: and Chaos for the Quadratic Map. J. Phys. A 30 (1997) 7067-7076
1584: \bibitem{Hunt99}B. R. Hunt, J. A. C. Gallas, C. Grebogi, J. A. Yorke, H. Koçak, Bifurcation
1585: rigidity. Physica D 129 (1999) 35-56
1586: \bibitem{Feigenbaum80}M. J. Feigenbaum, Universal Behavior in Nonlinear Systems. Los Alamos
1587: Science 1 4-27(1980)
1588: \bibitem{Myrberg58}P. J. Myrberg , Iteration von Quadratwurzeloperationen. Annals Acad.
1589: Sci. Fenicae A I Math. 259,1
1590: \bibitem{Yorke85}J. A. Yorke, C. Grebogi, E. Ott and L. Tedeschini-Lalli , Scaling
1591: behavior of windows in dissipative dynamical systems. Phys. Rev. Lett.
1592: 54 (1985), 1095-1098
1593: \bibitem{Epstein89}H. Epstein, Fixed points of composition operators II. Nonlinearity
1594: 2 (1989) 305-310.
1595: \bibitem{Derrida79}B. Derrida, A. Gervois, Y. Pomeau, Universal metric properties of
1596: bifurcations of endomorphisms. J. Phys. A-Math. Gen. 12 (1979) 269-196.
1597: \bibitem{Yeung00}M. K. S. Yeung and S. H. Strogatz , Nonlinear dynamics of a solid-state
1598: laser with injection. Phys. Rev. E 58(4) (1998) 4421-4435; Erratum,
1599: Phys. Rev. E 61(2) (2000) 2154
1600:
1601: \end{thebibliography}
1602:
1603: \end{document}
1604: