1: \documentclass{article}
2: \usepackage{amssymb,amsmath,amsthm}
3: %\usepackage{epsf}
4: \usepackage{bm}% bold math
5: \usepackage{graphicx}
6: \font\cyr=wncyr8
7:
8: \input mydefs
9: \def\real{\mathbb R}
10: \def\integer{\mathbb Z}
11: \def\rational{\mathbb Q}
12: \def\bN{\mathbb N}
13: \def\emb{\iota}
14:
15: \bibliographystyle{alpha}
16:
17: \newtheorem{definition}{Definition}
18: \newtheorem{theorem}{Theorem}
19: \newtheorem{prop}{Proposition}
20: \newtheorem{remark}{Remark}
21: \newtheorem{corollary}{Corollary}
22: \newtheorem{lemma}{Lemma}
23:
24: \title{Geometry of plane sections\\ of the infinite regular skew polyhedron $\fsf$}
25:
26: \author{Roberto De Leo, Ivan A. Dynnikov \thanks{INFN Cagliari,
27: Italy $\langle$roberto.deleo@ca.infn.it$\rangle$; Moscow State University, Russia
28: $\langle$dynnikov@mech.math.msu.su$\rangle$.
29: The work of the second named author is supported by Russian Federal Science Agency (grant
30: no.~{\cyr N\char'130}-1824.2008.1)}}
31:
32: \begin{document}
33: \maketitle
34:
35: \begin{abstract}
36: The asymptotic behavior of open plane sections of triply periodic surfaces is dictated,
37: for an open dense set of plane directions, by an integer second homology
38: class of the three-torus. The dependence of this homology class
39: on the direction can have a rather rich structure, leading in special cases to a fractal.
40: In this paper we present in detail the results for the skew polyhedron $\fsf$ and
41: in particular we show that in this case a fractal arises and that such a fractal
42: can be generated through an elementary algorithm, which in turn allows us to verify
43: for this case a conjecture of S.P.Novikov that such fractals have zero measure.
44: \end{abstract}
45:
46: \pagestyle{myheadings}
47: \thispagestyle{plain}
48: \markboth{R. De Leo, I.A. Dynnikov}{Plane sections of the skew polyhedron $\fsf$}
49: %
50: \section{Introduction}
51:
52: The study of plane sections of triply periodic surfaces in $\real^3$ was
53: initiated by S.P.Novikov in~\cite{Nov82} who raised a question whether
54: open (unbounded) components of such a section have some nice asymptotic
55: behavior. This was motivated by an application to conductivity theory.
56: A number of general theoretical results has been obtained since then
57: by A.Zorich~\cite{Zor84}, I.Dynnikov \cite{Dyn97,Dyn99},
58: and R.DeLeo~\cite{DeL03b,DeL05a}. For a number of surfaces, R.DeLeo
59: performed a numerical simulation, which confirmed the general
60: conclusions of the theory~\cite{DeL03a}.
61:
62: %
63: In~\cite{DeL06} the main results were generalized to polyhedra.
64: Among the class of piecewise linear triply periodic closed
65: surfaces, the one of infinite
66: skew polyhedra~\cite{Cox37} is the most suitable for
67: numerical explorations of the geometry
68: of plane sections.
69: In~\cite{DeL06} the case of the regular skew polyhedron
70: $\cP$ of type $\sff$ was studied numerically in detail, showing that
71: %the complexity of
72: the dependence of
73: open section's asymptotics on the plane direction keeps its rich
74: fractal-like structure also in the piecewise linear case.
75:
76:
77:
78:
79: In this work we present the results for the dual of $\cP$, namely the cubic polyhedron
80: $\cC=\fsf$~\cite{GSS02}; note that
81: %. Let us point out that
82: both $\cP$ and $\cC$ are rough PL-approximations
83: of the smooth surface $\{\cos x^1+\cos x^2+\cos x^3=0\}$, which was itself studied numerically
84: in~\cite{DeL03a}.
85: It turns out that the $\cC$ case is rather noteworthy because its correspondent fractal
86: can be generated recursively through a simple algorithm, which, on one hand,
87: allowed us, for the first time,
88: to verify, in this concrete case, the conjecture~\cite{MN03} that
89: the set of exceptional directions has zero measure,
90: and, on the other hand, made
91: possible a systematic
92: comparison with the numerical data obtained through the NTC software library~\cite{NTC}.
93: % add note about closed leaves == leaves homotopic to zero
94: %\par
95: %For sake of completeness we list in next section the main results relevant for our work.
96: %
97: \section{Topological structure of plane sections of triply periodic surfaces}
98: %
99: Let $M\subset\Tt$ be an embedded closed null homologous surface
100: in the torus $\Tt$, $H=(h_1,h_2,h_3)\in(\real^3)^*$
101: a covector. We denote by $\widehat M$ the preimage of $M$
102: under the projection $\real^3\rightarrow\Tt=\real^3/\integer^3$.
103: We consider the sections of $M$ by planes
104: $\langle H,x\rangle=\const$ (we call them \emph{$H$-sections})
105: and we are interested in the asymptotical behavior of
106: their unbounded regular connected components (if any).
107: Since only the direction of covector $H$ matters,
108: sometimes we shall treat $H$ as a point of the
109: projective plane $\real{\mathrm P}^2$.
110:
111: In studying this question, the foliation $\mathcal F_H$
112: induced on $M$ by the closed one-form $\omega=(h_1dx^1+h_2dx^2+h_3dx^3)|_M$
113: plays the crucial role.
114: %All along the paper we will sometimes consider the problem on
115: %the torus and sometimes on its
116: %universal covering. To avoid cumbersome notations we will use the same
117: %letter to refer to
118: %$\omega$ whether we consider it as
119: %1-form on $\Rt$ or on $\Tt$ and we will often write $\cM$
120: %rather than $\emb(\cM)$ when there will be no ambiguity;
121: %moreover we will refer equivalently
122: %to a curve as \emph{homotopic to zero} or \emph{closed} and as \emph{homotopically non-trivial} or
123: %\emph{open}, depending on whether we consider respectively its image in $\Tt$ or in $\Rt$.
124: %%(i.e. a periodic connected surface in $\Rt$)
125: %%Let us consider now plane sections of $\cC$, namely the leaves induced on $\cC$ by a constant
126: %%1-forms $\omega=h_i dx^i$.
127: %\par
128: It is well known that, with probability $1$, in a proper sense,
129: a smooth closed one-form
130: whose critical points are all saddles induce dense leaves
131: on $\cM$. However, by restricting attention to a special
132: class of one-forms that are
133: pull-backs of a constant one-form on $\Tt$,
134: we fall exactly in the opposite situation, namely,
135: with probability $1$, open leaves are either absent or
136: confined to genus one minimal components of the foliation on $\cM$,
137: and dense leaves arise only in exceptional cases.
138:
139: There are three principally different types of foliations
140: $\mathcal F_H$ and corresponding $H$-sections that may arise,
141: which we call \emph{trivial}, \emph{integrable}, and \emph{chaotic}.
142: Most typically, trivial means that all regular
143: leaves of $\mathcal F_H$ are closed,
144: integrable means that minimal components of $\mathcal F_H$ filled
145: with open leaves are of genus one, and in the chaotic case there is
146: a minimal component of genus $>1$. More precise definitions
147: are as follows.
148:
149: Let $N\subset\torus^3$ be a piece-wise smooth embedded surface in $\torus^3$
150: such that $N\setminus M$ consists of disjoint open disks each of which lies
151: in a plane of the form $\{x\in\real^3\;;\;\langle H,x\rangle=\const\}$.
152: Such a surface is obtained by, first, cutting $M$ along some closed null homologous
153: leaves of $\mathcal F_H$ or null homologous saddle connection cycles,
154: second, removing some of the obtained connected components, and then gluing
155: up planar disks in order to obtain a closed surface.
156: For such a surface $N$, any leaf of $\mathcal F_H$ is either contained
157: in $N$ or disjoint from $N$. In the former case we say that the leaf
158: is \emph{absorbed} by $N$. When saying this we shall assume that $N$
159: is of the just specified form.
160:
161: \begin{description}
162: \item[Trivial case.]
163: Every leaf of $\mathcal F_H$ is absorbed by a
164: two-sphere or a null homologous two-torus. If covector
165: $H$ is \emph{totally irrational}, i.e., $\dim_{\rational}\langle h_1,h_2,h_3\rangle
166: =3$, then this just means that all connected components of all $H$-sections
167: of $M$ are compact. If $\dim_{\rational}\langle h_1,h_2,h_3\rangle<3$,
168: then, additionally, \emph{periodic}, i.e., invariant under a non-trivial
169: shift, unbounded component of $H$-sections may arise;
170: \item[Integrable case.]
171: Every leaf of $\mathcal F_H$ is absorbed by a sphere or a two-torus,
172: and at least one leaf is absorbed
173: by a two-torus with non-zero homology class. In this case,
174: every regular non-closed component
175: of an $H$-section is a finitely deformed straight line,
176: i.e., it has the form $\gamma(t)=t\cdot v+O(1)$ for some parametrization,
177: where $v$ is a non-zero vector.
178: If $\dim_{\rational}\langle h_1,h_2,h_3\rangle=3$,
179: then the non-zero homology class of the tori absorbing
180: leaves of $\mathcal F_H$ is uniquely defined
181: up to sign. We denote it by $L_{M,H}$ and
182: consider as an integral covector in $\real^3$.
183: The identification of $H_2(\torus^3,\real)$ and $(\real^3)^*=
184: H^1(\torus^3,\real)$ is given by the Poincare duality.
185: This covector must obviously vanish at vector $v$: $\langle L_{M,H},v\rangle=0$.
186: So, if we assume our three-space Euclidean,
187: then (unless $H$ and $L_{M,H}$ are colinear)
188: we can simply write $v=L_{M,H}\times H$.
189: If $\dim_{\rational}\langle h_1,h_2,h_3\rangle<3$ the covector
190: $L_{M,H}$ may not be uniquely defined (up to sign),
191: but there may be at most two different choices.
192: We denote the projective class $(L_1:L_2:L_3)\in\rational\mathrm{P}^2$
193: of $L_{M,H}$ by $\ell_{M,H}$ and call \emph{the soul} of the foliation
194: $\mathcal F_H$.
195: \item[Chaotic case.] None of the above.
196: If $\dim_{\rational}\langle h_1,h_2,h_3\rangle=3$,
197: this means that a minimal component of $\mathcal F_H$
198: has genus $\geqslant3$.
199: The behavior of the corresponding $H$-sections have not been studied,
200: but the known examples suggest that, typically, a chaotic $H$-section
201: contains a single unbounded curve that ``wonders all around
202: the plane'', i.e., a $d$-neighborhood of the curve is the whole
203: plane for some finite $d$.
204: \end{description}
205:
206: For a fixed surface $M$ and a rational point $\ell\in\rational\mathrm{P}^2
207: \subset\real\mathrm{P}^2$
208: we denote by $\zone_{M,\ell}$ the set
209: $$\zone_{M,\ell}=\{H\in\real\mathrm{P}^2\;;\;\ell_{M,H}=\ell\}.$$
210: If $\ell_{M,H}$ is not uniquely defined then the point $(h_1:h_2:h_3)$
211: is attributed to both corresponding subsets.
212: The set of points $(h_1:h_2:h_3)$ such that
213: the $H$-sections of $M$ are chaotic will be denoted by $\cE(M)$.
214:
215: The following three propositions are extracted from~\cite{Dyn99}.
216:
217: \begin{prop}\label{p1}
218: For a generic surface $M\subset\torus^3$
219: the sets $\zone_{M,\ell}$ are disjoint closed domains with piece-wise
220: smooth boundary. The set $\cE(M)$ is disjoint from $\rational\mathrm P^2$
221: and has zero measure. The set of
222: directions $H$ with trivial $H$-sections is open.
223: \end{prop}
224:
225: In other words, the first claim of this proposition says that $\ell$, where defined,
226: is a locally constant function of $H$. We call the non-empty
227: domains $\zone_{M,\ell}$ \emph{stability zones} and refer
228: to $\ell$ as \emph{the label} of the stability zone $\zone_{M,\ell}$.
229:
230: For studying the stability zones, it is usefull to consider
231: a 1-parametric family $M_c=\{x\in\torus^3\;;\;f(x)=c\}$ of
232: level surfaces of a fixed smooth function.
233:
234: \begin{prop}\label{p2}
235: For a generic function $f$, there are continuous
236: functions $e_1,e_2:\real{\mathrm P}^2\rightarrow\real$ such that
237: \begin{itemize}
238: \item
239: $e_1(H)\leqslant e_2(H)$ for all $H\in\real{\mathrm P}^2$;
240: \item
241: the $H$-sections of $M_c$ are trivial if and only if $c\notin[e_1(H),e_2(H)]$;
242: \item
243: if $e_1(H)<e_2(H)$, then the $H$-sections of $M_c$ are integrable
244: for all $c\in[e_1(H),e_2(H)]$, and the soul $\ell$ of the corresponding
245: foliation $\mathcal F_{c,H}$ is independent of $c$.
246: \end{itemize}
247: \end{prop}
248:
249: We define
250: \emph{generalized stability zones} $\zone_{f,\ell}$ as
251: $\mathcal D_{f,\ell}=\cup_c\mathcal D_{M_c,\ell}$,
252: and the set $\cE(f)$ as $\cE(f)=\cup_c\cE(M_c)$.
253:
254:
255: \begin{prop}\label{p3}
256: For a generic $f$,
257: generalized stability zones are closed
258: domains with piece-wise smooth boundary. If $\ell\ne\ell'$ then
259: the zones $\mathcal D_{f,\ell}$ ¨ $\mathcal D_{f,\ell'}$ can only have
260: intersections at the boundary, and, moreover, the number of their common
261: points is at most countable. If the whole $\real{\mathrm P}^2$
262: is not covered by a single generalized stability zone, then
263: the number of zones is countably infinite, and the set
264: $\cE(f)=\real{\mathrm P}^2\setminus(\cup_\ell\zone_{f,\ell})$
265: is non-empty and uncountable.
266: \end{prop}
267:
268: It may happen that there is just one generalized stability zone
269: (say, a small enough perturbation of the function $\sin(x^1)$ will work),
270: but it is also easy
271: to find a function $f$ with non-empty $\cE(f)$: any function
272: with cubical symmetry is such.
273: In all examples known to us
274: two different generalized stability zones have at most one point in common.
275:
276: It follows from Propositions~\ref{p1}--\ref{p3} that
277: the union $\cup_{\ell\in\rational\mathrm P^2}\text{int}(\mathcal D_{f,\ell})$
278: of the interiors of the zones is an open everywhere dense subset of
279: $\real\mathrm P^2$ and its complement $\overline{\mathcal E(f)}$
280: has the form of a two-dimensional cut out fractal set.
281:
282: \begin{prop}[\cite{DeL03b}]\label{p4}
283: If there is more than one generalized stability zone, then $\overline{\mathcal E(f)}$ is
284: the set of accumulation points of the set of their souls.
285: \end{prop}
286:
287: It is plausible but
288: still unknown whether $\mathcal E(f)$ has always zero measure.
289: The following stronger conjecture was proposed in~\cite{MN03}.
290:
291:
292: \begin{conjecture}
293: Whenever $\cE(f)$ is non-empty, the Hausdorff dimension of $\cE(f)$
294: is strictly between 1 and 2 for every $f$.
295: \end{conjecture}
296: %
297: So far only numerical checks of this conjecture were available
298: in the literature~\cite{DeL03a,DeL06};
299: in the next sections we shall provide, for the
300: particular case of the polyhedron $\cC$,
301: a full proof of the weaker, zero measure, conjecture and good numerical
302: evidence for the stronger one.
303: %
304: \section{Stability zones of $\cC$}
305: \label{sec:selfsim}
306: %
307: The regular skew polyhedron $\cC=\fsf$ (see Figure~\ref{fig:dc}) is, up to isometries,
308: the unique cubic polyhedron with all monkey-saddle vertices~\cite{GSS02}; the vertices of its
309: embedding in $\Tt=[0,1]^3/\sim$ are the eight points in the orbit of $P=(1/4,1/4,1/4)$ under
310: the cubic symmetry group.
311: As level surface, $\cC$ can be represented in the $[0,1]^3$ cube as $\mf^{-1}(0)$ for
312: $$\mf(x^1,x^2,x^3)=\hbox{mid}(|2x^1-1|,|2x^2-1|,|2x^3-1|)-\frac12,$$
313: where $\hbox{mid}(a,b,c)$ is the middle value among $a$, $b$, and $c$.
314: \par
315: %
316: \begin{figure}
317: \begin{center}
318: \includegraphics[width=7.5cm]{cubicsurface1.eps}
319: \end{center}
320: \caption{%
321: \small
322: The $\fsf$ polyhedron embedded in the three-torus. Shown is the fundamental
323: domain, which lies in the $[0,1]^3$ cube.
324: }
325: \label{fig:dc}
326: \end{figure}
327: %
328: This surface, as well as the surface $\cos x^1+\cos x^2+\cos x^3=0$
329: and $\cC$'s dual---the truncated octahedron, has a very strong symmetry,
330: namely, its exterior is equal
331: to its interior modulo a translation. This means that for $\mf$ the functions $e_{1,2}$ mentioned
332: in Propositoin~\ref{p2} are such that $e_1=-e_2$, and hence,
333: stability zones of the surface $\cC$ coincide
334: with generalized stability zones of the function $\mf$.
335:
336: Let us denote by $\psi_1,\psi_2,\psi_3$ the following projective transformations:
337: $$\begin{aligned}
338: \psi_1(h_1:h_2:h_3)&=(h_1:h_2+h_1:h_3+h_1),\\
339: \psi_2(h_1:h_2:h_3)&=(h_1+h_2:h_2:h_3+h_2),\\
340: \psi_3(h_1:h_2:h_3)&=(h_1+h_3:h_2+h_3:h_3).
341: \end{aligned}$$
342:
343:
344: \begin{theorem}\label{zones}
345: For the surface $\cC$ the stability zones are as follows:
346: $$\begin{aligned}
347: \mathcal D_{(1:0:0)}(\cC)&=\{(h_1:h_2:h_3)\in\real\mathrm P^2\;;\;h_1\geqslant
348: \left|h_2\right|+\left|h_3\right|\},\\
349: \mathcal D_{(1:1:1)}(\cC)&=\{(h_1:h_2:h_3)\in\real\mathrm P^2\;;\;
350: \left|h_1\right|+\left|h_2\right|+\left|h_3\right|\leqslant 4h_1,4h_2,4h_3\},\\
351: \mathcal D_{\psi_{i_1}(\psi_{i_1}(\ldots\psi_{i_k}((1:1:1))\ldots))}(\cC)&=
352: \psi_{i_1}(\psi_{i_1}(\ldots\psi_{i_k}(\mathcal D_{(1:1:1)}(\cC))\ldots)),
353: \end{aligned}$$
354: where $(i_1,\ldots,i_k)$ is an arbitrary finite sequence
355: of elements from $\{1,2,3\}$, and, in addition, all zones
356: obtained from the listed ones by cubical symmetries:
357: permutations and changing signs of the coordinates.
358: \end{theorem}
359:
360: The proof of this theorem will rest on the following two lemmas.
361:
362: \begin{lemma}\label{IDL1}
363: We have
364: $$\mathcal D_{(1:0:0)}(\cC)\supset\{(h_1:h_2:h_3)\in\real\mathrm P^2\;;\;h_1\geqslant
365: \left|h_2\right|+\left|h_3\right|\}.$$
366: \end{lemma}
367:
368: \begin{proof}%[Proof of Lemma~\ref{IDL1}]
369: If the inequality
370: $h_1\geqslant
371: \left|h_2\right|+\left|h_3\right|$ is satisfied then
372: the plane $h_1x^1+h_2x^2+h_3x^3=\mathrm{const}$
373: passing through the center of the unit cube $[0,1]^3$
374: separates the faces $x^1=0$ and $x^1=1$.
375: This means that %the
376: closed leaves of the corresponding
377: foliation will cut our triply periodic surface $\widehat\cC$
378: into parts each of which is a finitely deformed
379: plane $x^1=\mathrm{const}$ with holes (see Figure~\ref{fig:wrapped}).
380: Filling the holes by flat disks and projecting to $\torus^3$ we
381: obtain two tori whose homology class is equal up to sign to
382: $(1,0,0)\in(\real^3)^*$.
383: \end{proof}
384:
385: \begin{figure}
386: \centerline{\includegraphics[height=7cm]{wrapped.eps}}
387: \caption{A connected component of $\widehat\cC$ after cutting
388: along closed leaves}\label{fig:wrapped}
389: \end{figure}
390:
391: \begin{lemma}\label{IDL2}
392: Let $h_1,h_2,h_3\geqslant0$, $(h_1,h_2,h_3)\ne(0,0,0)$. Then we have
393: $(h_1:h_2:h_3)\in\mathcal D_\ell$ if and only if
394: $\psi_i((h_1:h_2:h_3))\in\mathcal D_{\psi_i(\ell)}$,
395: where $i=1,2,3$.
396: \end{lemma}
397:
398: \begin{proof}%[Proof of Lemma~\ref{IDL2}]
399: For convenience, we shift the coordinate system as follows:
400: $(x^1,x^2,x^3)\mapsto(x^1+1/4,x^2+1/4,x^3+1/4)$. Our surface
401: $\widehat\cC$ cuts $\real^3$ into two parts, $N_-$ and $N_+$, that
402: can now be characterized as follows: $N_-$ (resp.\ $N_+$) consists of
403: points $(x^1,x^2,x^3)\in\real^3$ such that
404: at least two (resp.\ at most one) of the three numbers $\{x^1\},\{x^2\},\{x^3\}$
405: are in the interval $[0,1/2]$ (resp.\ $(0,1/2)$), where $\{x\}$ denotes
406: the fractional part of $x$.
407:
408: We may assume $i=3$, $(h_1,h_2,h_3)=(\alpha,\beta,1)$ without loss of generality.
409: Let $\Pi$ be the plane defined by $\alpha x^1+\beta x^2+x^3=c$.
410: Denote by $Q_-$ the projection of $\Pi\cap N_-$ to the $x^1,x^2$-plane
411: along $x^3$. According to the description of $N_-$ given above
412: $Q_-$ is the set of points $(x^1,x^2)\in\real^2$
413: such that exactly two of the three numbers $\{x^1\},\{x^2\},\{c-\alpha x^1-\beta x^2\}$
414: are in the interval $[0,1/2]$.
415:
416: Denote by $\square_{a,b}$ the square $\{(x^1,x^2)\in\real^2\,;\,
417: a\leqslant x^1\leqslant a+1/2,\,
418: b\leqslant x^2\leqslant b+1/2\}$, and by $S_m$ the strip
419: defined by $m\leqslant c-\alpha x^1-\beta x^2\leqslant m+1/2$.
420: We have
421: $$Q_-=\Bigl(\bigcup_{j,k\in\integer}\square_{j,k}\Bigr)
422: \cup\Bigl(\bigcup_{j,k,m\in\integer}\square_{j+1/2,k}\cap S_m\Bigr)
423: \cup\Bigl(\bigcup_{j,k,m\in\integer}\square_{j,k+1/2}\cap S_m\Bigr).$$
424: The first part in this union, $\bigcup_{j,k\in\integer}\square_{j,k}$,
425: does not depend on $\Pi$. We call these squares \emph{mainlands}.
426:
427: Each intersection $\square_{j+1/2,k}\cap S_m$ with $j,k,m\in\integer$,
428: whenever non-empty, is a convex polygon that can be of the following three types:
429: \begin{description}
430: \item[\emph{cape}:] it has a piece of boundary in common with exactly
431: one of the mainlands
432: $\square_{j,k}$ or $\square_{j+1,k}$;
433: \item[\emph{bridge}:] it has a piece of boundary in common with both mainlands
434: $\square_{j,k}$ and $\square_{j+1,k}$;
435: \item[\emph{island}:] it is disjoint from mainlands.
436: \end{description}
437: Similarly, one defines the type of a polygon $\square_{j,k+1/2}\cap S_m$
438: regarding adjacency to the mainlands $\square_{j,k}$ and $\square_{j,k+1}$,
439: see Figure~\ref{fig:geography}.
440:
441: \begin{figure}
442: \centerline{%
443: \begin{picture}(240,230)
444: \put(0,30){\includegraphics[height=200pt]{capebridgeisland.eps}}
445: \put(130,64){\rotatebox{-23}{bridge}}
446: \put(135,92){\rotatebox{-23}{bridge}}
447: \put(135,186){\rotatebox{-23}{bridge}}
448: \put(73,80){\rotatebox{-23}{mainland}}
449: \put(183,80){\rotatebox{-23}{mainland}}
450: \put(183,190){\rotatebox{-23}{mainland}}
451: \put(73,190){\rotatebox{-23}{mainland}}
452: \put(76,117){\rotatebox{-23}{cape}}
453: \put(88,143){\rotatebox{-23}{cape}}
454: \put(152,207){\rotatebox{-23}{cape}}
455: \put(205,155){\rotatebox{-23}{cape}}
456: \put(188,130){\rotatebox{-23}{island}}
457: \multiput(30,49)(10,0)4{\line(1,0)4}
458: \put(27,47){\hbox to0pt{\hss$k$}}
459: \multiput(30,103.5)(10,0)4{\line(1,0)4}
460: \put(27,101.5){\hbox to0pt{\hss$k+\frac12$}}
461: \multiput(30,158)(10,0)5{\line(1,0)4}
462: \put(27,156){\hbox to0pt{\hss$k+1$}}
463: \multiput(30,212.5)(10,0)4{\line(1,0)4}
464: \put(27,210.5){\hbox to0pt{\hss$k+\frac32$}}
465: \multiput(65.5,13)(0,10)4{\line(0,1)4}
466: \put(65.5,3){\hbox to 0pt{\hss$j$\hss}}
467: \multiput(120,13)(0,10)4{\line(0,1)4}
468: \put(120,3){\hbox to 0pt{\hss$j+\frac12$\hss}}
469: \multiput(174.5,13)(0,10)4{\line(0,1)4}
470: \put(174.5,3){\hbox to 0pt{\hss$j+1$\hss}}
471: \multiput(229,13)(0,10)4{\line(0,1)4}
472: \put(229,3){\hbox to 0pt{\hss$j+\frac32$\hss}}
473: \end{picture}%
474: }
475: \caption{Geography of the section $\Pi\cap N_-$}\label{fig:geography}
476: \end{figure}
477:
478: Capes are not interesting for us because their removal is equivalent to
479: a finite deformation of $Q_-$. It is not hard to write
480: the necessary and sufficient condition for $\square_{j+1/2,k}\cap S_m$
481: to be a bridge:
482: \begin{equation}\label{bridge}
483: c-\alpha\Bigl(j+\frac12\Bigr)-\beta\Bigl(k+\frac12\Bigr)-\frac12
484: \leqslant m\leqslant c-\alpha(j+1)-\beta k,
485: \end{equation}
486: or an island:
487: \begin{equation}\label{island}
488: c-\alpha\Bigl(j+\frac12\Bigr)-\beta\Bigl(k+\frac12\Bigr)-\frac12
489: \geqslant m\geqslant c-\alpha(j+1)-\beta k.
490: \end{equation}
491:
492: Now we apply $\psi_3$ to $H$, which gives $H'=(\alpha',\beta',1)=(\alpha+1,\beta+1,1)$.
493: Let $Q_-'$ and $S_m'$ be defined in the same way as $Q_-$ and $S_m$
494: with $\alpha$ and $\beta$
495: replaced by $\alpha'=\alpha+1$ and $\beta'=\beta+1$, respectively.
496: Then the intersection $\square_{j+1/2,k}\cap S_m'$ is a bridge if and only if
497: $$c-(\alpha+1)\Bigl(j+\frac12\Bigr)-(\beta+1)\Bigl(k+\frac12\Bigr)-\frac12
498: \leqslant m\leqslant c-(\alpha+1)(j+1)-(\beta+1)k,$$
499: which can be rewritten as
500: $$c-\alpha\Bigl(j+\frac12\Bigr)-\beta\Bigl(k+\frac12\Bigr)-\frac12
501: \leqslant m+j+k+1\leqslant c-\alpha(j+1)-\beta k.$$
502: Thus, $\square_{j+1/2,k}\cap S_m$ is a bridge if
503: and only if so is $\square_{j+1/2,k}\cap S_{m-j-k-1}'$.
504: Similarly, the same
505: is true about islands as well as bridges and islands in
506: squares of the form $\square_{j,k+1/2}$.
507:
508: So, bridges and islands of $Q_-'$ in the square $\square_{j+1/2,k}$
509: or $\square_{j,k+1/2}$ are in a natural one to one correspondence with
510: those of $Q_-$.
511: Therefore, $Q_-'$ and $Q_-$ are obtained from each other by
512: a finite deformation. The geometrical difference between
513: $Q_-'$ and $Q_-$ can be vaguely described as follows: islands
514: and bridges of $Q_-'$ are narrower than those of $Q_-$,
515: and $Q_-'$ has more capes. See Figure~\ref{fig:morecapes} for an example.
516:
517: \begin{figure}
518: $$\begin{array}{cc}
519: \includegraphics[height=4cm]{grid1.eps}&
520: \includegraphics[height=4cm]{grid2.eps}\\
521: Q_-&Q_-'
522: \end{array}$$
523: \caption{The transition from $Q_-$ to $Q_-'$ produces more capes and makes
524: the bridges narrower}\label{fig:morecapes}
525: \end{figure}
526:
527: In the genus three case the integrability of our foliation
528: is equivalent to the existence of closed fibres of the foliation (or null homologous
529: saddle connection cycles).
530: We have just seen that $H$-sections and $H'$-sections are obtained
531: from each other by a finite deformation. Hence, they are both integrable
532: or both chaotic. If they are integrable, let $\ell$ and $\ell'$
533: be the labels of the corresponding zones. We want to show
534: that $\ell'=\psi_3(\ell)$. Since both $\ell$ and $\ell'$
535: are locally constant functions of $H$ it is enough to consider
536: the case of totally irrational $H$. Then the asymptotic direction
537: $v=H\times\ell$ is of irrationality degree two, i.e., $\dim_\rational\langle
538: v^1,v^2,v^3\rangle=2$, and $\ell$
539: is the only rational covector up to multiple that
540: vanishes at $v$. Thus, it suffices to show
541: that $\psi_3(\ell)$ vanishes at $v'$, the direction
542: of open components of $H'$-sections.
543:
544: The latter follows easily from the fact that the projections of $v$ and
545: $v'$ to the $x^1,x^2$-plane coincide, which, in turn, follows from
546: the coincidence, up to finite deformation, of $Q_-$ and $Q_-'$.
547: \end{proof}
548:
549: \begin{proof}[Proof of Theorem~\ref{zones}]
550: Due to the cubical symmetry of the surface it suffices
551: to establish the fractal structure in the region
552: $C_+=\{(h_1:h_2:h_3)\in\real\mathrm P^2\,;\,h_1,h_2,h_3\geqslant0\}$.
553: Put
554: $$\begin{aligned}
555: \mathcal R_{(1:0:0)}&=\{(h_1:h_2:h_3)\in\real\mathrm P^2\;;\;h_1\geqslant
556: \left|h_2\right|+\left|h_3\right|\},\\
557: \mathcal R_{(0:1:0)}&=\{(h_1:h_2:h_3)\in\real\mathrm P^2\;;\;h_2\geqslant
558: \left|h_1\right|+\left|h_3\right|\},\\
559: \mathcal R_{(0:0:1)}&=\{(h_1:h_2:h_3)\in\real\mathrm P^2\;;\;h_3\geqslant
560: \left|h_1\right|+\left|h_2\right|\},\\
561: \mathcal R_{(1:1:1)}&=\{(h_1:h_2:h_3)\in\real\mathrm P^2\;;\;
562: \left|h_1\right|+\left|h_2\right|+\left|h_3\right|\leqslant 4h_1,4h_2,4h_3\},\\
563: \mathcal R_{\psi_{i_1}(\psi_{i_1}(\ldots\psi_{i_k}((1:1:1))\ldots))}&=
564: \psi_{i_1}(\psi_{i_1}(\ldots\psi_{i_k}(\mathcal R_{(1:1:1)})\ldots)),
565: \end{aligned}$$
566: and $\mathcal R_\ell=\varnothing$ if $\ell\notin\{(1:0:0),(0:1:0),(0:0:1)\}$
567: and $\ell$ is not of the form
568: $\psi_{i_1}(\psi_{i_1}(\ldots\psi_{i_k}((1:1:1))\ldots))$.
569: We want to show that $\mathcal R_\ell=\mathcal D_\ell$ for all $\ell\in
570: \rational\mathrm P^2\cap C_+$. We have already shown in Lemma~\ref{IDL1}
571: that $\mathcal R_\ell\subset\mathcal D_\ell$ for $\ell=(1:0:0)$
572: By symmetry it is also true for $\ell=(0:1:0),(0:0:1)$.
573:
574: It is a straightforward check to see that
575: $$\mathcal R_{(1:1:1)}=\psi_1(\mathcal R_{(1:0:0)}\cap C_+)\cup
576: \psi_2(\mathcal R_{(0:1:0)}\cap C_+)\cup\psi_3(\mathcal R_{(0:0:1)}\cap C_+).$$
577: By Lemma~\ref{IDL2} this implies
578: $\mathcal R_{(1:1:1)}\subset\mathcal D_{(1:1:1)}$
579: as $(1:1:1)=\psi_1(1:0:0)=\psi_2(0:1:0)=\psi_3(0:0:1)$,
580: and, therefore,
581: $\mathcal R_\ell\subset\mathcal D_\ell$ for all $\ell\in C_+\cap\rational\mathrm P^2$.
582:
583: In order to establish Theorem~\ref{zones} it suffices to show that
584: the zones $\mathcal D_\ell$ are not larger that $\mathcal R_\ell$, and
585: there are no other stability zones. Both claims follow from
586: the fact that $\bigcup_\ell\mathcal R_\ell$ covers all rational points:
587: $$\bigcup_\ell\mathcal R_\ell\supset C_+\cap\rational\mathrm P^2.$$
588:
589: Indeed, let $\varphi$ be the following map from $C_+$ to itself:
590: $$\varphi(h_1:h_2:h_3)=\left\{\begin{aligned}
591: (h_1:h_2-h_1:h_3-h_1)&,\text{ if }0\leqslant h_1\leqslant h_2,h_3,\\
592: (h_2:h_3:h_1)&,\text{ otherwise}.
593: \end{aligned}
594: \right.$$
595: By construction, for every $H\in C_+$ we have
596: $H\in\bigcup_\ell\mathcal R_\ell$ if and only if
597: $\varphi(H)\in\bigcup_\ell\mathcal R_\ell$.
598: If $H$ is a rational covector from $C_+$, then after applying
599: $\varphi$ finitely many times, one of the coordinates of
600: the obtained covector becomes zero. All such points are
601: covered by $\mathcal R_{(1:0:0)}$, $\mathcal R_{(0:1:0)}$
602: and $\mathcal R_{(0:0:1)}$.
603: \end{proof}
604: %
605: \begin{figure}
606: \begin{center}
607: % \includegraphics[width=7.5cm,viewport=150 150 600 600,clip]{cubeFrDisc.eps}
608: \includegraphics[width=7cm]{cubeFrDisc.eps}
609: \caption{%
610: \small
611: Picture of the fractal in the disc model of $\RPt$. The center of the disc corresponds to the $z$ axis,
612: so the central square is the stability zone $\cD_{(0:0:1)}$, the one touching it in the left and right vertices is
613: $\cD_{(1:0:0)}$ and the third one is $\cD_{(0:1:0)}$. The green triangles are the
614: stability zones $\cD_{(1:1:1)}$,
615: $\cD_{(1:1:-1)}$, $\cD_{(1:-1:1)}$ and $\cD_{(1:-1:-1)}$.
616: }
617: \label{fig:frDisc}
618: \end{center}
619: \end{figure}
620: %
621: So, we have the following picture in $\real\mathrm P^2$: four lines
622: $h_1\pm h_2\pm h_3=0$ cut $\real\mathrm P^2$ into
623: three ``squares'', which are zones $\mathcal D_{(1:0:0)}$,
624: $\mathcal D_{(0:1:0)}$, $\mathcal D_{(0:0:1)}$, and
625: four triangles obtained from each other by cubical symmetries,
626: in which there are infinitely many stability zones.
627: We shall concentrate on the triangle that is contained in $C_+$.
628: This triangle has vertices $(1:1:0)$, $(1:0:1)$, $(0:1:1)$
629: and is defined by the inequalities
630: $$\frac{h_1+h_2+h_3}2\geqslant h_1,h_2,h_3\geqslant0.$$
631: Let us denote this triangle by $\Delta$.
632: The zone $\mathcal D_{(1:1:1)}$ is also a triangle that is contained
633: in $\Delta$ and has its vertices, $(2:1:1)$, $(1:2:1)$, $(1:1:2)$,
634: at the sides of $\Delta$.
635: The complement $\Delta\setminus\mathcal D_{(1:1:1)}$
636: consists of three triangles that are exactly $\psi_1(\Delta)$,
637: $\psi_2(\Delta)$, and $\psi_3(\Delta)$.
638: In each triangle $\Delta_1=\psi_1(\Delta)$, $\Delta_2=\psi_2(\Delta)$,
639: $\Delta_3=\psi_3(\Delta)$ the picture is obtained from
640: that in $\Delta$ by the corresponding projective transformation $\psi_i$.
641:
642: For a finite sequence $a=(a_1,a_2,\dots,a_k)$ of indices $1,2,3$
643: we denote by $\psi_a$ the mapping $\psi_{a_1}\circ\psi_{a_2}\circ\ldots
644: \circ\psi_{a_k}$. By $a'$, $a''$, and $a'''$ we denote
645: the sequences $(a_1,\dots,a_k,1)$, $(a_1,\dots,a_k,2)$, $(a_1,\dots,a_k,3)$,
646: respectively. For any such sequence $a$ we have the following.
647: The triangle $\Delta_a=\psi_a(\Delta)$
648: is bounded by the zones $\mathcal D_{\psi_a(1:0:0)}$,
649: $\mathcal D_{\psi_a(0:1:0)}$, $\mathcal D_{\psi_a(0:0:1)}$.
650: It contains the zone
651: $\mathcal D_{\psi_a(1:1:1)}$ whose vertices
652: $\psi_a(2:1:1)$, $\psi_a(1:2:1)$, $\psi_a(1:1:2)$
653: are on the sides of $\Delta_a$, and the complement
654: $\Delta_a\setminus\mathcal D_{\psi_a(1:1:1)}$ is the union
655: of the triangles $\Delta_{a'}$, $\Delta_{a''}$, $\Delta_{a'''}$.
656:
657: %
658: \begin{figure}
659: \begin{center}
660: \includegraphics[width=7.5cm,viewport=30 130 580 668,clip]{cubeFractalAnal.ps}
661: \caption{%
662: \small
663: Picture of the fractal in the square $[0,1]^2$ in the $z=1$ projective chart.
664: }
665: \label{fig:frDiscAn}
666: \end{center}
667: \end{figure}
668: %
669:
670: \begin{prop}
671: The intersection $\mathcal E(\cC)\cap C_+$
672: consists of points of the form
673: $$\lim_{k\rightarrow\infty}\psi_{i_1}(\psi_{i_2}(\ldots
674: \psi_{i_k}((1:1:1))\ldots)),$$
675: where $(i_1,i_2,\ldots)$ runs over all possible sequences
676: of elements from $\{1,2,3\}$ containing each index infinitely many times.
677: Other points in $\mathcal E(\cC)$ are obtained from these by cubical simmetries.
678: \end{prop}
679:
680: \begin{proof}
681: From the structure of stability zones established above it
682: follows that the intersection $\mathcal E\cap C_+$
683: is the union of subsets
684: $$\bigcap\limits_k\psi_{i_1}(\psi_{i_2}(\ldots
685: \psi_{i_k}(\Delta)\ldots))$$
686: over all sequences $(i_k)\in3^{\mathbb N}$ in which all three indices appear
687: infinitely many times.
688: It suffices to show that every such a subset is actually a single point,
689: which follows from Proposition~\ref{zoneasympt} below.
690: \end{proof}
691: %
692: \subsection{Measure of $\cE$}
693: %\section{Measure and fractal dimension of $\cE$}
694: %
695: In all cases studied so far no algorithm was found to generate all stability
696: zones and nothing could be
697: said about the measure of the $\cE(f)$.
698: This is therefore the first case in which it is possible to
699: check the truth of the zero measure conjecture.
700: %
701: \begin{theorem}
702: The set $\mathcal E(\cC)$ of ``chaotic'' directions has measure zero.
703: \end{theorem}
704: %
705: \begin{proof}
706: Again, due to the symmetry, it suffices to prove the claim for $\mathcal E\cap C_+$.
707: We denote this set by $\mathcal E_+$. As we have seen, it is contained in
708: the triangle $\Delta$, and the following holds:
709: $$\mathcal E_+=\psi_1(\mathcal E_+)\cup\psi_2(\mathcal E_+)\cup
710: \psi_3(\mathcal E_+).$$
711: Applying this recursively, we get
712: \begin{multline}
713: \mathcal E_+=\psi_1(\mathcal E_+)\cup\psi_2(\mathcal E_+)\cup
714: \psi_{31}(\mathcal E_+)\cup\psi_{32}(\mathcal E_+)\cup
715: \psi_{331}(\mathcal E_+)\cup\psi_{332}(\mathcal E_+)\cup\\
716: \ldots\cup
717: \psi_{\underbrace{\scriptstyle3\ldots3}_k1}(\mathcal E_+)\cup
718: \psi_{\underbrace{\scriptstyle3\ldots3}_k2}(\mathcal E_+)\cup
719: \psi_{\underbrace{\scriptstyle3\ldots3}_{k+1}}(\mathcal E_+).
720: \end{multline}
721:
722: Let $\mu$ be the measure of $\mathcal E_+$ and $\mu_k$
723: the measure of $\psi_{\underbrace{\scriptstyle3\ldots3}_k1}(\mathcal E_+)$,
724: which is equal to the measure of
725: $\psi_{\underbrace{\scriptstyle3\ldots3}_k2}(\mathcal E_+)$.
726: The measure of the triangle
727: $\psi_{\underbrace{\scriptstyle3\ldots3}_{k+1}}(\mathcal E_+)$
728: tends to zero as $k$ goes to $\infty$. Therefore, we have
729: $$\mu=2\sum_{k=0}^\infty\mu_k.$$
730: The idea now is to show that $\mu_k\leqslant c_k\mu$, where $c_0,c_1,\ldots$
731: are constants such that
732: $$\sum_{k=0}^\infty c_k<\frac12,$$
733: which, together with the previous equality, implies $\mu=0$.
734:
735: Now we provide the necessary technical details. First of all,
736: we need to parametrise the triangle $\Delta$.
737: We use the following parametrization: $(u,v)\mapsto
738: (1-v:1-u:u+v)$, $u,v\geqslant0$, $u+v\leqslant1$.
739:
740: The property of a set to be of zero measure does not
741: depend on the choice of a regular Borel measure.
742: In order to get the desired inequalities we use a somewhat artificial
743: measure on $\Delta$, namely the following one:
744: $$\frac{du\,dv}{(1+u+v)^2}.$$
745: By doing so we keep the symmetry between $\psi_1$ and $\psi_2$ (one is conjugate
746: to the other by the permutation $u\leftrightarrow v$), so the measures
747: of the sets $\psi_{\underbrace{\scriptstyle3\ldots3}_k1}(\mathcal E_+)$
748: and $\psi_{\underbrace{\scriptstyle3\ldots3}_k2}(\mathcal E_+)$ coincide.
749: We denote by $f_k$ the mapping $\psi_{\underbrace{\scriptstyle3\ldots3}_k1}\circ R$,
750: where $R(h_1:h_2:h_3)=(h_3:h_1:h_2)$, written in the $u,v$-parametrization:
751: $$f_k(u,v)=\Bigl(\frac v{u+(k+1)(v+1)},\frac 1{u+(k+1)(v+1)}\Bigr).$$
752: Since we have $R(\mathcal E)=\mathcal E$, the image $f_k(\mathcal E_+)$
753: coincides with $\psi_{\underbrace{\scriptstyle3\ldots3}_k1}(\mathcal E_+)$.
754:
755: Obviously, the measure of $f_k(\mathcal E_+)$ is bounded from above
756: by $c_k\mu$, where
757: $$c_k=\max_{u,v\geqslant0,\,u+v\leqslant1}
758: J_{f_k}(u,v)\frac{(1+u+v)^2}
759: {\Bigl(1+\frac v{u+(k+1)(v+1)}+\frac 1{u+(k+1)(v+1)}\Bigr)^2}.$$
760: By $J_f$ we denote the Jacobian of the mapping $f$. A routine check gives:
761: \begin{multline*}
762: c_k=\max_{u,v\geqslant0,\,u+v\leqslant1}
763: \frac{(1+u+v)^2}
764: {(u+(k+2)(v+1))^2(u+(k+1)(v+1))}=\\
765: \left\{\begin{array}{ll}
766: \displaystyle\frac14&\text{if }k=0,\\
767: \displaystyle\frac4{(k+2)(k+3)^2}&\text{otherwise},
768: \end{array}\right.
769: \end{multline*}
770: $$\sum_{k=0}^\infty c_k=\frac14+\sum_{k=1}^\infty\frac4{(k+2)(k+3)^2}=
771: \frac{253}{36}-\frac23\pi^2\approx0.448<\frac12,
772: $$
773: which completes the proof.
774:
775: \end{proof}
776: %
777: \subsection{Asymptotics of stability zones and fractal dimension of $\cE$}
778: \label{sec:asympt}
779: %
780: From what said at the beginning of the section it follows immediately that
781: all stability zones, except the biggest ones, which are squares, are triangles
782: and their labels satisfy a simple recursive relation.
783: %
784: \begin{prop}
785: \label{prop:alg}
786: Every stability zone of $\cC$, except for the three squares with souls
787: $(1:0:0)$, $(0:1:0)$, and $(0:0:1)$,
788: are triangles; these triangles are generated starting from $\Delta$ (and its three symmetric
789: triangles with respect to the coordinate planes) according to the the following recursive algorithm:
790: \item{1.} evaluate the vector sums $w_1=v_2+v_3$, $w_2=v_3+v_1$, $w_3=v_1+v_2$ of all pairs
791: of the three vertices $v_i$ of $\Delta$;
792: \item{2.} consider the $w_i$ as projective coordinates and add the triangle having those points
793: as vertices to the list of all stability zones -- its soul is given by the vector sum of the three vertices
794: of $\Delta$;
795: \item{3.} consider now the three triangles $\Delta_1=\langle v_1,w_2,w_3\rangle$,
796: $\Delta_2=\langle w_1,v_2,w_3\rangle$,
797: $\Delta_3=\langle w_1,w_2,v_3\rangle$ and repeat the algorithm for each of them.
798: % the label of every triangular zone is the sum of the labels of the three zones
799: % touched by its vertices or, equivalently, the sum of the (projective) coordinates of the vertices
800: % of the triangle which encloses it.
801: \end{prop}
802: %
803: This fact can be exploited to find an explicit expression for elements of $\cE$ by considering
804: ``spiralling down'' (or ``steepest descent'') sequences of stability zones (see Figure~\ref{fig:spiral}).
805: Select indeed an ordered triple
806: $(\alpha,\beta,\gamma)$ of stability zones which bind a triangle, namely such that two touch each other
807: and the third touches both, and build out of them the
808: recursive sequence of stability zones such that every
809: new stability zone is the one whose vertices touch the previous three stability zones. It is easy to see that the sequence
810: of these stability zones are arranged in a sort of spiral and by construction, the label
811: of every stability zone of this sequence is the sum of the labels of the previous three zones, so that
812: the sequence of the labels is a Tribonacci sequence in $\mathrm PH_2(\Tt,\bZ)$.
813: %
814: \begin{figure}
815: \begin{center}
816: \includegraphics[width=7cm]{spiral.eps}
817: \caption{%
818: \small
819: Detail, in the $z=1$ projective chart, of the first few zones of the Tribonacci
820: sequence of stability zones starting by $\cD_{(1:0:0)}$, $\cD_{(0:1:0)}$ and $\cD_{(0:0:1)}$.
821: The labels of the zones shown above are, in decreasing area order, $(1:2:2)$, $(2:3:4)$,
822: $(4:6:7)$, $(7:11:13)$ and $(13:20:24)$. All centers of the zones of the sequence lie on the
823: smooth ``Tribonacci projective spiral'' drawn above which is converging to
824: $(\alpha^2-\alpha-1,\alpha-1)\simeq(.543,.839)$.
825: }
826: \label{fig:spiral}
827: \end{center}
828: \end{figure}
829: %
830: %
831: \begin{prop}
832: The limit of every such sequences belongs to $\cE$.
833: In particular $(\alpha^2-\alpha-1:\alpha-1:1)\in\cE$, where $\alpha$ is the Tribonacci
834: constant, namely the real solution of the Tribonacci equation $x^3-x^2-x-1=0$.
835: \end{prop}
836: %
837: \begin{proof}
838: All vertices of stability zones of $\cC$ are 1-rational points of $\RPt$ and therefore points on the
839: boundaries have irrationality degree not higher than 2. Now, consider the sequence starting
840: from the three squares $a_1=\cD_{(1:0:0)}$, $a_2=\cD_{(0:1:0)}$, $a_3=\cD_{(0:0:1)}$, so that the
841: first few next terms of the sequence will be $a_4=\cD_{(1:1:1)}$, $a_5=\cD_{(1:2:2)}$, $a_6=\cD_{(2:3:4)}$
842: and so on: a simple calculation show that the label of the $n$th stability zone of the sequence,
843: modulo terms in $\beta^n$ and $\bar\beta^n$, will be
844: % $$
845: % ({{\beta\bar\beta}\over{(\alpha-\beta)(\alpha-\bar\beta)}}\alpha^n+B\beta^n+C\bar\beta^n
846: % :
847: % {{-\beta-\bar\beta}\over{(\alpha-\beta)(\alpha-\bar\beta)}}\alpha^n+B'\beta^n+C'\bar\beta^n
848: % :
849: % {{1}\over{(\alpha-\beta)(\alpha-\bar\beta)}}\alpha^n+B''\beta^n+C''\bar\beta^n)
850: % $$
851: $$
852: (\frac{\beta\bar\beta}{(\alpha-\beta)(\alpha-\bar\beta)}\alpha^n
853: :
854: \frac{-\beta-\bar\beta}{(\alpha-\beta)(\alpha-\bar\beta)}\alpha^n
855: :
856: \frac{1}{(\alpha-\beta)(\alpha-\bar\beta)}\alpha^n)
857: $$
858: where $\alpha=(1+\sqrt[3]{19-3\sqrt{33}}+\sqrt[3]{19+3\sqrt{33}})/3\simeq1.84$ is the Tribonacci
859: constant and $\beta,\bar\beta$ the remaining two solutions of the Tribonacci equation $x^3-x^2-x-1=0$.
860: Since $|\beta|<\alpha$ the sequence of labels converges to the triple of coefficients of
861: $\alpha^n$, namely $\ell_\infty=(\beta\bar\beta:-\beta-\bar\beta:1)=(\alpha^2-\alpha-1:\alpha-1:1)$;
862: these three coordiantes
863: % finally, since 1, $\beta$ and $\beta\bar\beta$
864: are rationally independent so that $\ell_\infty$ has rationality degree $3$ and therefore it cannot belong
865: to any boundary and for Proposition~\ref{p4} it must belong to $\cE$.
866: \end{proof}
867: %
868: Note that, since $\cE$ is invariant by the $\psi_a$, this way we automatically get a countable
869: number of explicit elements of $\cE$.
870: %every composition of all directions obtained in this way belong to $\cE$
871: %and therefore we
872: %,so that we are able to build explicitly countably many explicit examples of directions belonging to it.
873: %The asymptotics of stability zones are not yet fully understood; we start this section by presenting
874: %what is known.
875: %
876: \begin{prop}
877: Let $\{\cD_i\}_{i\in\bN}$ be the set of all stability zones
878: sorted according to any recursive algorithm, e.g.
879: %$\cD_0=D_{(1:1:1)}$,
880: $\cD_{(i_1\dots i_n)_3}=\psi_{i_n+1}\circ\dots\circ\psi_{i_1+1}(\cD_{(1:1:1)})$, where
881: $(i_1\dots i_n)_3$ is the \emph{base 3} expression for the index of the stability zone,
882: %$\cD_i=\psi_i(\cD_0)$, $i=1,2,3$, $\cD_{3+j\cdot i}=\psi_i(\cD_j)$, $i,j=1,2,3$ and so on,
883: and be $\ell_n$ the label associated to $\cD_n$.
884: Then $2\log_3^2(1+2n)+1\leqslant \|\ell_n\|^2\leqslant3(1+2n)^{2\log_3\alpha}$ where $\alpha$ is the Tribonacci constant.
885: \end{prop}
886: %
887: \begin{proof}
888: A simple induction argument shows that, at every recursion level $k$, the biggest label belongs,
889: modulo permutations of the projective coordinates, to the $k$th stability zone of the steepest descent sequence
890: having as first three elements $D_{(0,0,1)}$, $D_{(0,1,0)}$ and $D_{(1,0,0)}$. Since at the $k$th
891: recursion level there are $3^k$ stability zones it follows that $\|\ell_{\frac{3^k-1}{2}}\|\leqslant\sqrt{3}\alpha^k$
892: and therefore $\|\ell_n\|\leqslant \sqrt{3}(3n)^{\log_3\alpha}$.
893: \par
894: The slowest sequence which can be formed by picking a stability zone for every recursion level is instead the
895: one where $\cD_k$ is the spawn of, say, $D_{(0,0,1)}$, $D_{(0,1,0)}$ and $\cD_{k-1}$. In this case
896: indeed $\|\ell_{\frac{3^k-1}{2}}\|=\sqrt{2(k+1)^2+1}$ from which follows immediately the left part
897: of the inequality.
898: \end{proof}
899: %
900: %
901: \begin{prop}\label{zoneasympt}
902: Be $\cD_\ell$ a stability zone with label $\ell$, $p_\ell$ its
903: perimeter and $a_\ell$ its area in $\RPt$.
904: Then there exist four constants $A$, $B$, $C$, $D$ such that
905: $$
906: \frac{A}{\|\ell\|^{\frac{3}{2}}}\leqslant p_\ell
907: \leqslant\frac{B}{\|\ell\|}\;,\;\;\frac{C}{\|\ell\|^3}\leqslant a_\ell \leqslant\frac{D}{\|\ell\|^3}
908: $$
909: \end{prop}
910: %
911: \begin{proof}
912: The inequalities concerning $p_\ell$ are already known to be true in the general case.
913: \par
914: To prove the ones relative to the area we change coordinates in $\RPt$ so that the
915: three points $(1:0:1)$, $(0:1:1)$, $(1:1:0)$ become $(0:0:1)$, $(1:0:1)$, $(0:1:1)$.
916: This way we can use the Lebesgue measure instead of the projective one and we can use
917: the fact that $0\leqslant x,y\leqslant z$; now, be $(x_i:y_i:z_i)$, $i=0,1,2$, the projective
918: irreducible coordinates of the vertices $A_i$ of the triangle $T$ which encloses $\cD_\ell$,
919: so that $l=(\sum_i x_i:\sum_i y_i:\sum_i z_i)$; the areas of $T$ and $\cD_\ell$ are then
920: easily computed respectively as $\cA(T)=\Pi_{i=0}^2\frac{1}{z_i}$ and
921: $\cA(\cD_\ell)=\Pi_{i=0}^2\frac{1}{z_i+z_{i+1}}$ (where the indices are meant modulo 3).
922: \par
923: It is convenient for our purposes to consider the maximum norm $\|\ell\|_\infty$ since
924: % every small enough stability zone of $\cC$ can be obtained by symmetry from a zone entirely contained in the region
925: in the region under consideration $0\leqslant x,y\leqslant z$, so that we can assume without
926: loss of generality that $x_i,y_i\leqslant z_i$, $i=0,1,2$, and therefore we get trivially that
927: $$
928: \cA(\cD_\ell)=\frac{1}{(z_0+z_1)(z_1+z_2)(z_2+z_0)}\geq\frac{1}{(z_0+z_1+z_2)^3}=\frac{1}{\|\ell\|^3_\infty}
929: $$
930: The second part of the inequality comes from the fact that if $\max z_i\leqslant \hbox{mid\ } z_i + \min z_i$
931: holds for the first triangle of the algoritm in Prop.~\ref{prop:alg} then it holds for every other
932: triangle generated by the algorithm.
933: Indeed assume to fix the ideas that $z_0\leqslant z_1\leqslant z_2$; then the new $z$ coordinates of the
934: three vertices under the transform $\psi_0(z_0:z_1:z_2)=(z_0:z_1+z_0:z_2+z_0)$ satisfiy trivially
935: again the relation
936: %$\{z'_0=z_0,z'_1=z_1+z_0,z'_2=z_2+z_0\}$, so that again
937: $z'_2\leqslant z'_0 + z'_1$, and the very same happens for the tranforms under $\psi_1$. In case of
938: $\psi_2$ we have $\{z'_0=z_0+z_2,z'_1=z_1+z_2,z'_2=z_2\}$ so that the new larger $z$ is now $z'_1$
939: but again $z'_1\leqslant z'_2+z'_0$.
940: Then finally
941: $$\frac{\|\ell\|^3_\infty}{(z_0+z_1)(z_1+z_2)(z_2+z_0)}=
942: \Bigl(1+\frac{z_0}{z_1+z_2}\Bigr)\Bigl(1+\frac{z_1}{z_2+z_0}\Bigr)
943: \Bigl(1+\frac{z_2}{z_0+z_1}\Bigr)\leqslant8$$
944: so that $\cA(\cD_\ell)\leqslant8/\|\ell\|^3_\infty$. The final statement now follows from the fact that all norms
945: are equivalent in finite dimension.
946: % The second part of the inequality is not true for any triple
947: %it is enough to prove we can reduce to prove
948: %Now, be
949: % $(x_i:y_i:z_i)$, $i=0,1,2$, the projective irreducible coordinates of the vertices $A_i$
950: % of the triangle $T$ which enloses $\cD_l$, so that $l=(\sum_i x_i:\sum_i y_i:\sum_i z_i)$;
951: % the areas of $T$ and $\cD_l$ are then easily computed respectively as $\cA(T)=\Pi_{i=0}^2\frac{1}{z_i}$
952: % and $\cA(\cD_l)=\Pi_{i=0}^2\frac{1}{z_i+z_{i+1}}$ (where the indices are meant {\sl modulo 3}).
953: % indeed in our case
954: % $z_i\geq x_i,y_i$ so that $\|l\|_\infty=z_0+z_1+z_2$ and clearly $\max z_i\leq\|l\|_\infty\geq3\max z_i$.
955: % Hence
956: % $A(\cD_l)\geq A(T)=\frac{1}{z_1z_2z_3}\geq\f_1z_2z_3} \leq3\max z_i\leq3 z_1z_2z_3$
957: % (since all $z_i$ are bigger than 1)
958: % Since $\|l\|^2=(\sum_{i=0}^2x_i)^2+(\sum_{i=0}^2y_i)^2+(\sum_{i=0}^2z_i)^2$ and in our case
959: % $z_i\geq x_i,y_i$ then $\max z_i\leq\|l\|\leq3^{3/2}\max z_i$, so that
960: % $A(\cD_l)\geq A(T)=\frac{1}{z_1z_2z_3}\geq\frac{1}{(\max z_i)^3}\geq\frac{1}{\|l\|^3_\infty}\geq\frac{3^{-3/2}}{\|l\|^3}$.
961: % $A(\cD_l)\geq\frac{1}{(2\max z_i)^3}\geq\frac{3^{3/2}/2^3}{\|l\|^3}$.
962: % On the other side,
963: % $\|l\|_\infty=z_0+z_1+z_2\leq3\max z_i\leq3 z_1z_2z_3$ (since all $z_i$ are bigger than 1)
964: % and therefore, since $\|l\|\leq\sqrt{2}\|l\|_\infty$,
965: % $A(\cD_l)\leq A(T)=2/(z_1z_2z_3\leq$
966: % $A(\cD_l)\geq\leq A(T)=1/z_1z_2z_3\leq$
967: \end{proof}
968: %
969: \begin{figure}
970: \begin{center}
971: \begin{tabular}{cc}
972: \includegraphics[width=5.85cm]{minkDim.eps}&
973: \includegraphics[width=5.85cm]{bcDim.eps}\\
974: % \includegraphics[width=5.85cm]{boxDim.ps}\\
975: \end{tabular}
976: \end{center}
977: \caption{%
978: \small
979: (a) Plot of $d_n=2-\log_{1.2}V(\cE_{1.2^{-n}})$ against $n$, where $V(\cE_{1.2^{-n}})$ is the volume
980: of the neighborhood of $\cE$ of radius $1.2^{-n}$; for ``well-behaving'' fractals this sequence
981: converges to their Minkowsky dimension. The horizontal line shown in the picture has height $1.72$.
982: (b) Plot of $\log_2 N_n$ against $n$, where $N_n$ is the number of squares of side $2^{-n}$ which
983: are not completely inside one of the 797161 stability zones found at the 12th recursion level. The angular
984: coefficient of the straight line fitting the data in the picture is again $1.72$.
985: }
986: \label{fig:numfracdim}
987: \end{figure}
988: %
989: We could not find any way to evaluate exact non-trivial bounds for the fractal dimension
990: of $\cE$ but numerical calculations suggest that the dimension be smaller than 1.8.
991: \par
992: A simple way to evaluate numerically fratal dimensions is using the \emph{Minkowsky dimension},
993: namely the limit
994: $$
995: \dim_M\cE=2-\lim_{\epsilon\to0}\frac{\log V(\cE_\epsilon)}{\log \epsilon}
996: $$
997: where $V(\cE_\epsilon)$ is the surface of the $\epsilon$ neighborhood of $\cE$.
998: In order to do that we use the fact that, if $A$ is the area of $\Delta$ and $p$ its
999: perimeter, then
1000: $$
1001: V(\cE_\epsilon) = p\epsilon + \epsilon\sum_{i=1}^{k_\epsilon} p_i + A - \sum_{i=1}^{k_\epsilon} a_i +
1002: \epsilon^2(\pi-\sum_{i=1}^{k_\epsilon}\frac{p_i^2}{4a_i})
1003: $$
1004: where $k_\epsilon$ is the integer such that $\rho_{k_\epsilon+1}\leqslant\epsilon\leqslant\rho_{k_\epsilon}$
1005: and $\rho_k$ is the radius of the inscribed circle to the triangle $\cD_k$.
1006: In order to avoid infinities we make a projective change of coordinates so that the triangle
1007: $\Delta$ has vertices in $(0:0:1)$, $(1:0:1)$ and $(0:1:1)$. In Figure~\ref{fig:numfracdim}(a)
1008: we show the numerical results we got by evaluating the volume of the neighborhoods of
1009: $\cE$ of radii $r_n=1.2^{-n}$ for $n=1,\ldots,50$, which suggests a fractal dimension
1010: between $1.7$ and $1.8$.
1011: \par
1012: A second simple way is to evaluate an upper bound for the \emph{box-counting dimension},
1013: namely the limit
1014: $$
1015: \dim_B\cE=\lim_{\epsilon\to0}\frac{\log N_\epsilon(\cE)}{-\log \epsilon}
1016: $$
1017: where $N_\epsilon(\cE)$ is the smallest number of squares of side $\epsilon$ needed to
1018: cover $\cE$ (even in this case we make the same change of coordinates to avoid infinities).
1019: In Figure~\ref{fig:numfracdim}(b) we show the results relative to covering with squares
1020: of side $2^{-n}$, $n=1,\cdots,12$, the complement of the 797161 triangles obtained by
1021: applying 12 times the recursion algorithm. Again we get a clear indication of the fractal
1022: dimension to be between 1.7 and 1.8.
1023: %
1024: \section{Numerical generation of stability zones}
1025: %
1026: %
1027: \begin{figure}
1028: \begin{center}
1029: \begin{tabular}{cc}
1030: \includegraphics[width=5.85cm]{cubeFractal-n1000.eps}&
1031: \includegraphics[width=5.5cm,viewport=0 128 566 667,clip]{cubeFractalAnalNum.ps}\\
1032: % \includegraphics[width=7.5cm,viewport=150 150 600 600,clip]{cubeFrDisc.eps}&
1033: % \includegraphics[width=7.5cm]{cubeFractal.eps}\\
1034: \end{tabular}
1035: \end{center}
1036: \caption{%
1037: \small
1038: % (a) Picture of the fractal in $\RPt$; (b) detail of the fractal in the $[0,1]^2$ square in the $z=1$ chart.
1039: (a) Picture of the fractal in the square $[0,1]^2$ in the $z=1$ chart of $\RPt$ obtained
1040: by evaluating the soul of every stability zone in the lattice $(n,m,N)$, $n,m=1,\ldots,10^3$, $N=10^3$;
1041: (b) the same picture compared with the analytical boundaries of the $797161$ stability zones of the first
1042: $12$ levels of recursion. Out of the 961367 rational directions for which a label was numerically
1043: found by the algorithm, 455654 belong to those $797161$ stability zones and all of them turn out to lie
1044: within the analytical boundaries of the corresponding zone.
1045: }
1046: \label{fig:numfrac}
1047: \end{figure}
1048: %
1049: As an important byproduct of the study of stability zones
1050: of $\cC$, we were able for the first time
1051: to compare very accurately the results of the NTC code~\cite{NTC},
1052: used in~\cite{DeL03a,DeL03b,DeL06}
1053: to generate approximations of surfaces' generality stability zones against their
1054: analytical boundaries.
1055: \par
1056: Indeed no simple algorithm to generate the analytical equations of the boundaries of stability zones
1057: for a generic function is known so far but we know that all directions
1058: belonging to the same stability zone share the same soul and therefore it is possible
1059: to get an approximate picture of the set of generalized stability zones by evaluating the soul
1060: of some (big) set of rational directions. For example in all cases examined so far, thanks
1061: to the high level of symmetry, we could limit our analysis to the directions contained in the
1062: the triangle with sides $(0:0:1)$, $(1:1:1)$ and $(1:0:1)$ (concretely, to all directions
1063: $(m:n:N)$, $0<n<m<N$, for $N=10^2,10^3,10^4$).
1064: \par
1065: Note that rational directions in this setting are of paramount importance because
1066: their (non-critical) leaves are compact (in $\Tt$) and therefore can be in principle
1067: approximated with error as small as wished and therefore the corresponding soul can
1068: be, at will, evaluated \emph{exactly} through numerical calculations.
1069: Moreover, rational directions are dense in every stability zone and therefore (in principle)
1070: we do not lose any picture detail by restricting our analysis to them.
1071: \par
1072: Below we present the algorithm we use to retrieve the soul (if any) of a rational direction
1073: $(m:n:N)$ in this particular case, where we have all saddles of ``monkey'' type:
1074: %
1075: \begin{itemize}
1076: \item[{\textbf N0}] choose a representative $b_i$, $i=1,2,3$, for the cycles
1077: on $\cC$ which are respectively homologous in $\Tt$ to the $x$, $y$ and $z$ axes;
1078: \item[{\textbf N1}] retrieve the intersection between $\cC$ and the plane $$m(x-1/4)+n(y-1/4)+N(z-1/4)=0\,;$$
1079: \item[{\textbf N2}] follow the three critical loops and, if no saddle connection is detected,
1080: store them in the variables $c_{1,2,3}$, otherwise exit;
1081: \item[{\textbf N3}] evaluate the homology class of $c_{1,2,3}$ in $\Tt$;
1082: \item[{\textbf N4}] if exactly one of the three loops is non-zero in $\Tt$ then evaluate its
1083: intersection numbers with the $\{b_i\}$ and set this triple as the soul corresponding to
1084: the direction $(m,n,N)$, otherwise exit.
1085: % evaluate the intersection number
1086: % evaluate the component of the homology class in $\cC$ of either $c_1$ or $c_2$ in the
1087: % rank-3 dual of the kernel of $i_*:H_1(\cC,\bZ)\to H_1(\Tt,\bZ)$
1088: %three components of the soul $k\in H_2(\Tt,\bZ)$ are given by the
1089: % the common components of the homology classes of $c_1$ and $c_2$ in $H_1(\cC,\bZ)$.
1090: \end{itemize}
1091: \par
1092: The result of sampling the triangle $\cT$ with a $10^3 \times 10^3$ lattice are shown in
1093: Figure~\ref{fig:numfrac} and turn out to be in perfect agreement with the analytical boundaries.
1094: An evaluation of the fractal box dimension based on these numerical data leads to a value of
1095: about $1.7$,
1096: % (see fig.~\ref{fig:numbc})
1097: which is also in very good agreement with the evaluation
1098: obtained from of the analytical boundaries made in sec.~\ref{sec:asympt}.
1099: %
1100: %\section{Conclusions}
1101: %
1102: %The problem of the asymptotics of plane sections of the surface $\cC$ as function of the
1103: %direction of the planes bundle turned out to be at the same time very rich and exactly solvable
1104: %in almost all of its details; in particular, a simple recursive algorithm was found to
1105: %generate all stability zones out of the biggest three (actually out of one of them, since
1106: %the other two can be obtained from a single one by symmetry).
1107: %\par
1108: %Thanks to the high degree of symmetry we were able to prove for this particular case the
1109: %conjecture of Novikov about the negligibility of the set of exceptional directions and we
1110: %obtained strong numerical evidence for the second part of the conjecture, namely that the
1111: %fractal dimension of that set is strictly smaller than two.
1112: %Moreover we were able for the first time to compare the results of the software used so far
1113: %to investigate numerically the problem with the analytical ones, which increased our confidence
1114: %in it.
1115: %\par
1116: %The success with this particular case brings some hope about finding a general proof for
1117: %the Novikov conjecture and ultimately proves that polyhedra may lead to valuable results
1118: %in this context.
1119: %
1120: \section{Acknowledgments}
1121: %
1122: The authors gladly thanks the IPST (www.ipst.umd.edu) and the Dept. of Mathematics of the UMD (USA)
1123: (www.math.umd.edu) for their hospitality in the Spring Semester 2007 and for financial support.
1124: Numerical calculation were made on Linux PCs kindly provided by the UMD Mathematics Dept.
1125: and by the Cagliari section of INFN (www.ca.infn.it) which also provided financial support
1126: tho the first author. The authors also warmly thank S.P. Novikov and B. Hunt for several
1127: fruitful discussion during their stay at UMD.
1128: %
1129: \bibliography{dc}
1130: %
1131: \end{document}
1132: