1: \documentclass{article}

2: \usepackage{amssymb,amsmath,amsthm}

3: %\usepackage{epsf}

4: \usepackage{bm}% bold math

5: 

6: \usepackage{graphicx}

7: 

8: \bibliographystyle{alpha}

9: %\bibliographystyle{plain}

10: 

11: \def\bH{{\bm H}}

12: \def\bQ{{\mathbb Q}}

13: \def\bR{{\mathbb R}}

14: \def\bS{{\mathbb S}}

15: \def\bT{{\mathbb T}}

16: \def\bZ{{\mathbb Z}}

17: \def\cA{{\cal A}}

18: \def\cC{{\cal C}}

19: \def\cD{{\cal D}}

20: \def\cM{{\cal M}}

21: \def\cN{{\cal N}}

22: \def\cP{{\cal P}}

23: \def\cS{{\cal S}}

24: \def\cZ{{\cal Z}}

25: \def\mf{{\mathfrak f}}

26: \def\mg{{\mathfrak g}}

27: \def\mh{{\mathfrak h}}

28: \def\Rt{\bR^3}

29: \def\Td{\bT^2}

30: \def\Sd{\bS^2}

31: \def\Tt{\bT^3}

32: \def\Zt{\bZ^3}

33: \def\irr{\hbox{\rm irr}}

34: \def\span{\hbox{\rm span}}

35: \def\soul{\hbox{\rm soul}}

36: \def\Star{\hbox{\rm Star}}

37: \def\RPd{\bR\hbox{\rm P}^2}

38: \def\RPt{\bR\hbox{\rm P}^2}

39: %\def\oM{\overline{M}}

40: \def\oN{\overline{\cN}}

41: \def\oM{\cM}

42: \def\M{M^2_c}

43: \def\phm{\phantom{-}}

44: \def\IdentityMatrix{1\hbox{\hskip -.12cm 1}}

45: \def\rmU{\uppercase\expandafter{\romannumeral1}}

46: \def\rmD{\uppercase\expandafter{\romannumeral2}}

47: \def\rmT{\uppercase\expandafter{\romannumeral3}}

48: 

49: \def\fig#1#2{

50:   \begin{figure}

51:     \begin{center}

52:      \includegraphics[width=15cm,bb=140 0 800 734]{#1}

53: %     \epsfxsize=15cm\epsfbox[140 0 800 734]{#1}

54:     \end{center}

55:    \caption{#2}

56:    \label{fig:#1}

57:   \end{figure}

58: }

59: 

60: \newtheorem{definition}{Definition}

61: \newtheorem{theorem}{Theorem}

62: \newtheorem{remark}{Remark}

63: \newtheorem{corollary}{Corollary}

64: \newtheorem{lemma}{Lemma}

65: 

66: \title{Topology of plane sections of periodic polyhedra with an application to the Truncated Octahedron}

67: 

68: \author{Roberto De Leo\thanks{INFN $<$roberto.deleo@ca.infn.it$>$, Dept. Physics, U. of Cagliari and Dept. of Mathematics, U. of Cagliari $<$deleo@unica.it$>$, Cagliari, Italy.}}

69: 

70: \begin{document}

71: \maketitle

72: 

73: \begin{abstract}

74: The main results of A. Zorich and I. Dynnikov about plane sections of periodic surfaces

75: are extended to the PL case. As an application, the Stereographic Map of a truncated

76: octahedron, extended to the whole $\Rt$ by periodicity, is analyzed numerically.

77: \end{abstract}

78: 

79: %\begin{keywords}

80: %low-dimension topology, Poisson geometry, multivalued functions,

81: %magnetoresistance in normal metals, polyhedra

82: %\end{keywords}

83: 

84: %\begin{AMS}

85: %57M50, 53D17, 37E35, 65D18, 82D35

86: %\end{AMS}

87: 

88: \pagestyle{myheadings}

89: \thispagestyle{plain}

90: \markboth{Roberto De Leo}{Plane sections of a periodic polyhedron}

91: 

92: \section{Introduction}

93: 

94: The problem of the asymptotics of plane sections of smooth periodic surfaces, extracted

95: from Physics literature by S.P. Novikov in 1982~\cite{Nov82}, turned out to be much richer

96: then expected, leading ultimately to the association of a fractal on $\RPt$ to every

97: element of a large class of triply periodic functions in $\Rt$ (see Section 2).

98: \par

99: In order to visualize for the first time Stereographic Maps associated to triply periodic

100: smooth functions and to get numerical confirmations of a Novikov conjecture, claiming that

101: the Hausdorff dimension of such fractals is stricly between 1 and 2, we developed a C++

102: library and used it to investigate two smooth cases \cite{DL03a}. Unfortunately, the

103: running time of our numerical explorations grows way too much as soon as we sample with

104: resolutions big enough to get a hint of the fractals, mainly because the number of

105: polygons of the meshes approximating curved smooth surfaces (e.g. to retrieve its

106: intersection with a plane) gets soon very big.

107: \par

108: This fact suggests that, from the numerical point of view, polyhedra are the best surfaces

109: to study, at the very least for the obvious reason that the number of polygons needed to

110: describe them at any resolution is constant. Moreover, such constant can be rather small

111: even in non trivial cases, e.g. like in the case of the ``extended truncated octahedron'',

112: presented in this paper, which has just eight hexagonal faces.

113: \par

114: Since the main results of the theory, due to A.V. Zorich~\cite{Zor84} and

115: I.A. Dynnikov~\cite{Dyn97,Dyn99}, refer to the case of smooth surfaces only, in Section 2

116: we will provide independent proofs for those theorems that make use of the smooth

117: structure and will mention the main properties of the system.  Then, in Section 3, we

118: present the algorithm we implemented to explore numerically the problem and the results

119: obtained in case of a the polyhedron obtained extending by periodicity the truncated

120: octahedron.

121: %

122: \section{Fundamental objects and theorems}

123: %

124: \subsection{Critical Points of height functions in polyhedra}

125: %

126: An analog of the Morse theory for height functions on polyhedra has been introduced by

127: T. Banchoff in \cite{Ban67,Ban70}. Here we recall the concepts relevant for the present

128: paper, slightly modified to cover the case of periodic polyhedra.

129: %

130: \begin{definition}

131: By ``embedded polyhedron'' $M\subset\Rt$ we mean a countable collection of cells

132: $K=\{C^r\subset\Rt\}_{r=0,1,2}$, where 0-cells are points (vertices), 1-cells are closed

133: connected segments and 2-cells are convex closed plane polygons, such that:

134: \par

135: %

136: \begin{enumerate}

137:     \item[{\textbf P1}] the boundary of any cell is union of cells of lesser degree;

138:     \item[{\textbf P2}] every cell having points in common with a higher degree cell is completely 

139: 	contained inside it;

140:     \item[{\textbf P3}] K is locally finite, i.e. every vertex has a neighborhood in $\Rt$ that intersects 

141: 	only finitely many cells;

142:     \item[{\textbf P4}] for any point $p\in M$, the union $\Star(p)$ of all cells containing that point

143: 	is omeomorphic to an open disc.

144: \end{enumerate}

145: A triply periodic polyhedron is a polyhedron that is invariant with respect to a rank-3 

146: discrete subgroup $\Gamma\simeq\Zt$ of $\Rt$. 

147: %\par

148: Finally, by polyhedron $\oM\subset\Tt$ we mean the 

149: quotient $M/\Gamma\subset\Rt/\Gamma\simeq\Tt$ of a triply periodic polyhedron.

150: \end{definition}

151: %

152: \par

153: Every triply periodic polyhedron $M$ embedded in $\Rt$ is the lift of a compact polyhedron

154: $\oM$ embedded in $\Tt$. Since we are going to study polyhedra's plane foliations, we are

155: interested in the reciprocal relation between polyhedra and height functions (or,

156: equivalently, constant 1-forms):

157: %

158: \begin{definition}

159: A height function $h(p)=h^\alpha p_\alpha$ is called ``generic'' for the polyhedron $M$

160: if no edge of $M$ is perpendicular to the direction $\bH=(h^\alpha)$. Equivalently, a constant

161: 1-form $\omega=h^\alpha dp_\alpha$ in $\Rt$ (resp. $\Tt$) is generic for $M$ (resp. $\oM$)

162: if no edge of $M$ (resp. $\oM$) is contained in a single leaf %

163: \footnote{Frobenius theorem grants that the distribution $\omega=0$ is integrable iff the 1-form

164: $\omega$ is closed.  In this case, the leaves induced by $\omega=h^\alpha dp_\alpha$ in

165: $\Tt$ are the projections of the $\Rt$ planes perpendicular to $\bH=(h^\alpha)$; those induced on a

166: polyhedron $\oM$ are the intersections of these leaves with the polyhedron.}%

167: of $\omega$.

168: %no two vertices have the same height. Equivalently, a constant 1-form $\omega=h^\alpha dp_\alpha$ 

169: %in $\Rt$ (resp. $\Tt$) is generic for $M$ (resp. $\oM$) if no two vertices lie on 

170: %the same leaf of $\omega$.

171: \end{definition}

172: %

173: \par

174: Since height functions are not single-valued on $\Tt$ (unless $\bH$ is an integer

175: direction, i.e. parallel to a lattice vector) while their differentials $\omega=dh$ are

176: always well-defined in both $\Rt$ and $\Tt$, we will refer mostly to 1-forms from now on.

177: From the definition above it is clear that, like in the smooth case, the set of

178: non-generic 1-forms has zero measure.

179: \par

180: Note that the foliation induced on $\Tt$ (and therefore on $\oM$) by $\omega$ does not

181: change by multiplying the 1-form by a non-zero scalar, so from now on we will think of

182: $\omega$ at the same time as a constant 1-form and as a point in $\RPt$.

183: %is at most countable.

184: %

185: \begin{definition}

186: The index of a point $p\in M$ (or, equivalently, $[p]\in\oM$) with respect to a

187: generic constant 1-form $\omega$ is the integer $i(p,\omega)=1-s/2$, where $s$ is the

188: number of segments, having $p$ as one of their extremes, in which the leaf of $\omega$

189: passing through $p$ cuts $\Star(p)$.

190: If $i(p,\omega)=0$, i.e. if the $\Star(p)$ is cut in exactly two components, the point is

191: said ``regular''; otherwise it is called ``critical''.

192: \end{definition}

193: %

194: \par

195: The set of critical points for any generic constant 1-form is of course a subset of the

196: set of vertices; exactly as in the smooth case, minima and maxima have index $+1$ and

197: non-degenerate saddles have index $-1$.

198: The main difference between the smooth and the piece-wise linear (PL) case, for our

199: purposes, is that saddles that are unstable in the former case, i.e. disappear for small

200: perturbations of the 1-form direction, are stable in the latter and therefore cannot be

201: disregarded; the simplest example is provided by the ``Monkey saddle'', which has index

202: $-2$ (see Figure~\ref{fig:ms}).

203: \begin{figure}

204: \begin{center}

205: \includegraphics[width=5cm]{ms.eps}

206: %\epsfxsize=5cm\epsfbox{ms.eps}

207: \end{center}

208: \caption{%

209: Piece-wise linear Monkey Saddle: in the smooth case, a however small perturbation would be

210: enough to resolve this critical point in a pair of elementary saddles; in the piece-wise

211: linear case instead it is stable and, therefore, generic.

212: }

213: \label{fig:ms}

214: \end{figure}

215: \par

216: Nevertheless, an analog of the critical point theorem for generic 1-forms holds:

217: %

218: \begin{theorem}

219: If $M$ is a triply periodic polyhedron, invariant by the action of the rank-3 group

220: $\Gamma\subset\Zt$, and $\omega$ is generic for $M$, then

221: $\sum_{[p]\in\oM}i(p,\omega)=\chi(\oM)$, where $[p]$ is the set of vertices

222: $\Gamma$-equivalent to $p$ and the sum is hence extended to any set of inequivalent

223: vertices. Equivalently, if we set to 1 the volume of a Dirichlet domain of $\Gamma$,

224: %volume spanned by any three generators of $\Gamma$,

225: then the average of the Euler characteristic of M converges to the Euler characteristic of

226: $\oM$:

227: $$\overline{\chi(M)}=\displaystyle\lim_{R\to\infty}\sum_{\|p\|<R}i(p,\omega)/Vol(B_R)=\chi(\oM)$$

228: \end{theorem}

229: %

230: \begin{proof}

231: Since Banchoff's proof \cite{Ban70} of the critical point theorem for polyhedra is only

232: based on local identities that are trivially true also in $\Tt$, that proof holds with no

233: change for our case.

234: \par

235: The second part can be proved by considering that the genus $g$ of the surface $\oM$

236: contained inside a cube of side $R$ can be evaluated by reducing by homotopy the surface

237: to a graph and then evaluating the rank of the graph's first homology group.

238: The result follows from the consideration that every component contained in a inner

239: unitary cube contributes by $g$ to the total genus of the component contained in the cube

240: of radius $R$ and their number grows with $R^3$, while the cubes on the boundary provide a

241: smaller contribute but can be disregarded in the limit for $R\to\infty$ since their number

242: grows only as $R^2$.

243: \end{proof}

244: %

245: \subsection{Structure of foliations}

246: %

247: In the most general case, a constant 1-form $\omega$ induces on a triply periodic 

248: polyhedron $M$ both open and closed leaves. 

249: \par

250: Since being homotopic to zero is an open condition, leaves close enough to closed ones are

251: also closed; maximal components of closed leaves are always enclosed between a pair of

252: critical points of $\omega$ on $M$ and form either cylinders (when both critical points

253: are saddles) or discs (when one is a saddle and the other is a center) or spheres (when

254: they are both centers). The last two cases are topologically trivial: a disc covered by

255: closed leaves around a center is exactly a homotopy to a point of that component of $M$,

256: and if $M$ is a sphere then no open orbit can ever be induced on it by a closed

257: 1-form. Hence, once the topologically trivial components are removed, what is left is a

258: collection of cylinders $\cC_i$ that separates a collection of subpolyhedra with boundary

259: $\cN_i$ filled by open leaves.

260: %

261: \begin{definition}

262: A {\sl genus-k component} of $\oM$ is a PL submanifold with boundary $\cN$ of $\oM$ such that:

263: \par

264: %

265:  \begin{enumerate}

266:    \begin{item}

267:    $\partial \cN$ is the finite disjoint union of plane parallel (topological) circles 

268: 	homotopic to zero;

269:    \end{item}

270:    \begin{item}

271:    the closed polyhedron $\oN$, obtained by filling $\cN$'s holes with plane discs, 

272: 	has genus k.

273:    \end{item}

274:  \end{enumerate}

275: \end{definition}

276: %

277: \begin{definition}

278: A {\sl Dynnikov decomposition} $\cZ$ of a polyhedron $\oM$ is a collection of subpolyhedra

279: with boundary $\{\cC_i,\cN_j\}$ of $\oM$ such that:

280: \par

281: %

282:  \begin{enumerate}

283:     \item every $\cC_i$ is homotopic either to a closed cylinder or to a closed disc;

284:     \item every $\cN_j$ is a genus-$k_j$ component of $\oM$;

285:     \item $i\neq j\implies\cN_i\cap\cN_j=\emptyset$, while every other pair of distinct subpolyhedra 

286: 	of $\cZ$ that is not disjoint shares single a boundary component;

287:     \item $\oM=\bigcup_i\cC_i \bigcup_j\cN_j$.

288:  \end{enumerate}

289: The genus and rank of a Dynnikov decomposition $\cZ$ are, respectively, the highest genus

290: and rank of the $\oN_j$s contained in $\cZ$; a genus-1 rank-2 Dynnikov decomposition is

291: called ``Zorich decomposition''.

292: \end{definition}

293: %

294: %\begin{remark}

295: From the considerations above it is clear that every constant 1-form $\omega$ in general

296: position with respect to $\oM$, i.e. such that there are no saddle connections, induces

297: naturally a Dynnikov decomposition $\cZ$ on $\oM$ where all $\cC_i$ are filled by the

298: closed leaves and all $\cN_j$ by the open ones.

299: %\end{remark}

300: %

301: \par

302: Under the same assumptions, to every focus it is associated a saddle that ``cancels'' it,

303: namely there's a homotopy of the surface that gets rid of the pair saddle-focus without

304: modifying the topology of the open orbits nearby; we call the saddle-type critical points

305: that are left ``topological saddles'', since it is them that contribute to the Euler

306: characteristic of the surface.

307: \par

308: In the smooth generic case, the number of $\Gamma$-inequivalent topological saddles,

309: i.e. the number of topological saddles in $\oM$, is of course exactly equal to $2g-2$; in

310: the PL case however, since we can have generic saddles with higher multiplicity, the

311: Euler characteristic is only an upper bound for the number of topological saddles, as

312: $\chi(\oM)=2-2g=\sum_{k\in\bZ}k\cdot\#\{[p]|i(\omega,p)=k\}$.

313: %

314: \subsection{The close-to-rational case}

315: %

316: \begin{definition}

317: By ``irrationality degree'' of a closed 1-form $\Omega$ on a piecewise smooth manifold $M$

318: we mean the number of rationally independent integrals of $\Omega$ over any base of the

319: homology integer 1-cycles of $M$: $\irr(\Omega)=\dim_\bQ\span_\bQ\{\int_\gamma\Omega\}$,

320: $\gamma\in H_1(M,\bZ)$; 1-irrational forms are also called ``rational''.

321: \end{definition}

322: %

323: The structure of foliations induced on a periodic polyhedron by rational 1-forms is much simpler 

324: than in the generic case since in this case all leaves are periodic; moreover, the appearance

325: of a topological invariant enforces all 1-forms close enough to rational to show the same

326: behaviour.

327: %Of  course $\irr(\Omega)\leq\dim H_1(M,\bR)$, so a closed 1-form $\Omega$ on $\Tt$ cannot

328: %have irrationality degree higher than three, and this inequality is inherited by

329: %any PL closed 1-form $\omega$ obtained by restricting $\Omega$ to a polyhedron $\oM$.

330: %%induced on a polyhedron $P$ by a PL embedding $i:P\to\oM\subset\Tt$.

331: \par

332: Rational 1-forms are rather special because foliations induced on $\oM$ by them are also

333: induced by well-defined circle-valued functions on the polyhedron: indeed, if

334: $\{\gamma_i\}_{i=1,2,3}$ is a base for $H_1(\Tt,\bR)$ and $N$ an integer big enough

335: such that $N\int_{\gamma_i}\omega\in\bZ$, $i=1,2,3$, and $p_0\in\oM$, then the function

336: $$

337: \vbox{

338:   \halign{

339: 	$#$\hfill&\hskip 2pt\hfill$#$\hfill&\hskip 2pt$#$\hfill&\hskip 2pt\hfill$#$\hfill\cr

340: 	f:&M&\longrightarrow&\bS^1\cr

341: 	&p&\mapsto&\exp(iN\displaystyle\int_{p_0}^p \omega)\cr

342:   }

343: }

344: $$ 

345: is well defined and differentiable, and its differential $df_p=iN\omega_p f(p)$ is

346: proportional to $\omega$; since $f$ is never zero, the set $df=0$ on $M$ coincides with

347: the restriction to $M$ of $\omega=0$.

348: \par

349: This shows that all leaves induced by rational 1-forms on a polyhedron $\oM\subset\Tt$ are

350: compact and therefore corresponding leaves on $M\subset\Rt$ will be open or closed

351: according to their homology class: leaves homotopic to zero in $\Tt$ will remain so in any

352: covering, all others will open up in $\Tt$'s universal covering.

353: %

354: \begin{definition}

355: From now on we will refer to closed leaves non homotopic to zero as ``periodic'' or, more

356: generically, ``open'' leaves, so that by ``closed leaf'' we will implicitly mean a closed

357: leaf homotopic to zero.

358: \end{definition}

359: %

360: \begin{lemma}

361: \label{lemma:4ol}

362: Be $\omega$ a rational 1-form in general position with respect to a polyhedron $\oM$: then

363: no more than two open leaves can collide at any saddle point.

364: \end{lemma}

365: %

366: \begin{proof}

367: The 1-form $\omega$ foliates $\Tt$ in a 1-parameter family of embedded 2-tori, so that all

368: open leaves at the same level of $\omega$ are parallel (i.e. they represent the same

369: 1-cycle modulo sign); moreover the number of open leaves on every level is even, since

370: they are all indivisible and their sum must be zero.

371: \par

372: It is easy to check, just by drawing pictures, that it is possible to have saddles of any

373: index with closed leaves, and adding a single pair of open leaves does not change this

374: situation; their presence though does not allow the presence of any other pair, since 

375: in any saddle point the two extremes of the same (critical) leaf must appear next to

376: each other and this is of course impossible for all open leaves, apart for the two most 

377: external ones (e.g. see Figure~\ref{fig:4ol}).

378: \end{proof}

379: %

380: \begin{figure}

381: \begin{center}

382: \includegraphics[bb=0 300 612 692,width=6cm]{4ol.ps}

383: %\epsfxsize=6cm\epsfbox[0 200 612 692]{4ol.ps}

384: \end{center}

385: \caption{%

386: A (hypothetical) saddle point where four open leaves meet simultaneously: the two branches

387: of each of the inner open leaves cannot be adjacent to each other, so this picture cannot

388: come from the section of a locally euclidean surface.

389: }

390: \label{fig:4ol}

391: \end{figure}

392: %

393: \begin{theorem}

394: \label{thm:Zor}

395: The Dynnikov decomposition induced on $\oM$ by any constant rational 1-form in general

396: position is Zorich.

397: \end{theorem}

398: %

399: \begin{proof}

400: Since all (non critical) leaves induced by $\omega$ on $\oM$ are circles, the critical

401: points of $\omega$ on $\oM$ determine a subdivision of the polyhedron in the connected sum

402: of a finite number of cobordisms. In the smooth case, the only non trivial cobordisms are

403: elementary (i.e. pants-like) and Zorich \cite{Zor84} proved that in the boundary of each

404: pant there is at least one of the three boundary loops that is homotopic to zero (in

405: $\Tt$), from which the theorem follows easily.

406: \par

407: In the polyhedra case the cobordisms are not necessarily elementary, i.e. more than two

408: leaves may collide at a saddle point, but lemma~\ref{lemma:4ol} grants us that even in

409: this case no more than two loops in the cobordism may be open.

410: This shows that the components of open leaves have genus 1 and therefore the decomposition

411: induced on $\oM$ by $\omega$ is Zorich.

412: \end{proof}

413: %

414: The presence of a genus-1 rank-2 component of open leaves is a very strong condition and leads

415: to the appearance of a topological quantity associated to the foliation~\cite{Dyn97}:

416: %

417: \begin{theorem}

418: A Dynnikov decomposition $\cZ$ of $\oM$ with rank 2 has genus 1 if and only if at least one of

419: its $\cN_j$ embedded with rank 2 has genus 1. In this case, all $\cN_j$ are genus-1

420: components of $\oM$ and those embedded with rank 2 represent, modulo sign, the same

421: indivisible non-zero 2-homology class in $\Tt$.

422: %Be $\cZ$ a Dynnikov decomposition of $\oM$. If at least one of the $\oN_j$ of $\cZ$ is a

423: %genus-1 rank-2 component of $\oM$, then $\cZ$ has genus 1 and all $\oN_j$ embedded with

424: %rank 2 represent the same indivisible homology class in $\Tt$ (modulo sign).

425: \end{theorem}

426: %

427: \begin{definition}

428: We call ``soul'' of a Zorich decomposition $\cZ$ the ``unsigned'' non-trivial

429: indivisible homology class $l\in PH_2(\Tt,\bZ)$ common to all rank-2 components of

430: $\cZ$.

431: \end{definition}

432: %

433: \begin{corollary}

434: Be $\omega$ a rational 1-form inducing on $\oM$ a rank-2 Zorich decomposition $\cZ$:

435: then any 1-form $\omega^\prime$ close enough to $\omega$ induces on $\oM$ a rank-2 Zorich

436: decomposition $\cZ^\prime$ and this decomposition is homotopic to $\cZ$.

437: In particular, all such decompositions share the same soul, i.e. the soul is a locally

438: constant function of the pair $(\oM,\omega)$. 

439: %Be $\oM$ a polyhedron embedded in $\Tt$ and $\omega$ a generic (???) constant rational 1-form. 

440: %If $\omega$ induces open leaves on $\oM$, then it also induces closed leaves and 

441: %moreover the open leaves are bound to genus-1 components of $\oM$.

442: \end{corollary}

443: %

444: \begin{proof}

445: The leaves at the boundary between $\oM$'s genus-1 components of periodic leaves $N_i$ and

446: the cilinders of closed leaves are themselves closed and therefore stable under small

447: perturbations of $\omega$'s direction; consequentely, no cilinder of closed orbits will

448: disappear in a whole neighborhood of $\omega\in\RPt$. Since no two leaves of a foliation

449: can intersect, this means that open leaves are bounded to genus-1 components of $\oM$,

450: homotopic to the $\cZ$ ones, even for all 1-forms close enough to $\omega$.

451: \par

452: Finally, since the homology class $l$ of these rank-2 genus-1 components is integer and

453: they change continously, this $l$ must be the same (modulo sign) for all them.

454: \end{proof}

455: %

456: \par

457: The soul $l$ of a Zorich decomposition is a fundamental invariant since its knowledge is

458: enough to describe the asymptotic behaviour of the open leaves. Indeed, the fact that

459: $[\cN_i]=\pm l$ means that, in $\Rt$, the lift $\hat\cN_i$ of $\cN_i$ lies between a pair

460: of parallel planes perpendicular to $l$ (seen as a direction in $\Rt$), so that the lift

461: to $\Rt$ of an open leaf, namely an open intersection of $\hat\cN_i$ with a plane

462: perpendicular to $\omega$ (seen as a vector in $\Rt$), is a curve contained in a plane

463: strip of finite width; Dynnikov~\cite{Dyn92} showed how these conditions are enough to

464: conclude that the leaf is actually a finite deformation of a straight line whose direction

465: is the axis of the strip, namely ``$\omega\times l$''.

466: \par

467: Since the foliation induced by $\omega$ is determined just by its direction $\bH$ and the

468: soul itself can be interpreted as a direction in $\Rt$, once we fix a surface $M$ we can

469: think the soul application as a locally constant application $\soul_{\oM}:\RPt\to\RPt$.

470: We will show in next section that, for a generic polyhedron $\oM$, this map is well

471: defined on the whole projective plane except for a set of meaure zero (``ergodic

472: directions'') and its image amounts to a finite number of points.

473: \begin{definition}

474: The non-empty level sets $\cD_l(\oM)=\soul_{\oM}^{-1}(l)$ corresponding to non-zero values of $l$

475: are called ``islands'' or ``stability zones'' of $\oM$. The union of all stability zones

476: $\cS(\oM)=\cup_l\cD_l(\oM)$ is called the ``Stereographic Map'' (SM) of $\cM$.

477: \end{definition}

478: Strange as it seems, this whole construction is the natural model for a very concrete

479: physical phenomenon, namely the magnetoresistance in normal metals, with the periodic

480: surface being the ``Fermi Surface'' of a metal, the 1-form a strong constant homogenous

481: external magnetic field and the Fermi Surface leaves the orbits of the momenta of the metal's

482: quasi-electrons.  The presence of open leaves is detectable experimentally, and so is each

483: stability zone (provided at least two points of it are measured \cite{NM98}). 

484: Experimental plots of the SM have been produced in Sixties and Seventies for about thirty

485: metals but only recently the first two SM, relative to the Fermi Surfaces of Au and Ag,

486: have been reproduced theoretically from first first principles~\cite{DL04,DL05a}.

487: %

488: \subsection{The generic (3-irrational) case}

489: %

490: In the previous section we showed that the structure of the foliation in a neighborhood of

491: rational directions is rather simple: either all leaves are closed or their lift is

492: strongly asymptotic to a straight line~\cite{Dyn97}, i.e. it is contained in a

493: finite-width plane strip and crosses it from one end to the other.

494: Still, this is far from being enough, since the sole density of an open set does not

495: preclude its measure from being small.

496: \par

497: We will prove in this section that the one above is nevertheless the generic situation,

498: namely:

499: %

500: \begin{theorem}

501: \label{thm:Dyn}

502: The set of directions in $\RPt$ inducing Dynnikov decompositions of genus bigger than one

503: on a generic polyhedron $\oM$ has measure zero.

504: \end{theorem}

505: %

506: To prove this theorem, Dynnikov studied the structure of the foliation induced by a

507: 3-irrational 1-form $\omega$ on 1-parameter family of surfaces $M_e=f^{-1}(e)$, where $f$ is

508: a triply periodic Morse function such that for almost all values of $f$ no more than one

509: critical point of $\omega$ lies on the same leaf. Here, we will repeat Dynnikov's steps

510: assuming $f$ to be a generic triply periodic PL function, so that almost all of its level

511: surfaces are embedded polyhedra, satisfying the same genericity condition with respect to

512: $\omega$.

513: \par

514: The following two lemmas by Dynnikov extend with no change to the polyhedra case:

515: \begin{lemma}

516: \label{thm:interval}

517: Given a triply periodic PL function, the set of values for which a constant 1-form $\omega$

518: induces open leaves filling rank-2 components is a closed connected interval

519: $[e_1(\omega),e_2(\omega)]$. The functions $e$ and $E$ are continous on the whole $\RPd$.

520: \end{lemma}

521: %

522: \begin{definition}

523: The Dynnikov index $w$ of a critical point $c$ of $\omega=h^\alpha dp_\alpha$ with respect

524: to an oriented polyhedron $\oM$ is the product of the ``Hamiltonian'' index of the critical

525: point ($+1$ for centers and $-1$ for saddles) times the sign of the scalar product between

526: $\bH=(h^\alpha)$ and any of the normals to the faces adjacent to the critical vertex.

527: \end{definition}

528: %

529: \begin{lemma}

530: The curve $\gamma_{\omega,f}(e)=\sum_i w_i c_i(e)$ is a well defined loop in $\Tt$ and the

531: quantity 

532: $$\overline{\chi}_{\omega,f}(e)=\int_{-\infty}^e i^*_{\gamma_{\omega,f}}\omega$$ 

533: is equal to the density of closed leaves on any leaf induced by $\omega$ on $\Tt$.

534: \par

535: Moreover, if $\{h^+_j\}$ (resp. $\{h^-_k\}$) are the heigths of the ``positive''

536: (resp. ''negative'') cylinders of closed leaves on $\oM$, namely those that contain points

537: with smaller (resp. bigger) values of $f$, it turns out that it is possible to choose

538: $\Rt$ representative $\hat c_i$ of the critical points so that

539: $$\overline{\chi}_{\omega,f}(e)=\sum_{i} <\bH, w_i \hat c_i(e)>=\sum h^+_j - \sum h^-_k$$

540: %where $H=(h^\alpha)$ and $<,>$ is the Euclidean scalar product.

541: \end{lemma}

542: %

543: In the weighted sum of all critical points are also contained the pairs center-saddle,

544: that can be removed by homotopy and have nothing to do with the topology of the foliation.

545: If we do not include them in the summation, we are left with a new quantity that tells us

546: the density of non-trivial (in $\oM$) closed leaves and therefore is able to spot whether

547: there are cylinders of non-zero heights. In particular, to have a full ergodic situation,

548: i.e. a leaf dense on the whole $\oM$, this function must necessarily be zero.

549: %

550: \begin{lemma}

551: Be $\{c^{top}_i\}$ the subset of the topological saddles of the pair $(\omega,f)$. Then

552: the reduced curve $\gamma^{top}_{\omega,f}(e)=\sum_i w_i c^{top}_i(e)$ is also

553: well-defined in $\Tt$ and the reduced Euler density

554: $\overline{\chi}^{top}_{\omega,f}(e)=\int_{-\infty}^e i^*_{\gamma^{top}_{\omega,f}}\omega$

555: is equal to the density of closed leaves non homotopic to zero in $\oM$.

556: \par

557: The function $\overline{\chi}^{top}_{\omega,f}$ is a strictly increasing non continous PL

558: function with respect to $e$; its discontinuity points are exactly the non-proper values

559: of $f$.

560: \end{lemma}

561: %

562: \begin{proof}

563: The original curve $\gamma_{\omega,f}(e)$ is well-defined because $\sum

564: w_i=0$~\cite{Dyn97}; since the Dynnikov indices of a pair saddle-center are opposite, the

565: restricted sum $\sum_{top} w_i$ is still zero and therefore also

566: $\gamma^{top}_{\omega,f}(e)$ is well defined.

567: \par

568: Considering only the sum of the topological saddles is equivalent to cancel from a leaf

569: all closed leaves that are homotopic to zero in $\oM$, so that the averaged Euler

570: characteristic now is only relative to the non-trivial closed leaves only.

571: \par

572: Now consider a positive cylinder: since the values inside are smaller with respect to the

573: values on the cylinder, the height of the cylinder increases with $e$; for the same

574: reason, negative cylinders decrease their height.

575: Since $\overline{\chi}^{top}_{\omega,f}(e)=\sum_{top} h^+_j - \sum_{top} h^-_k$, then the

576: function $\overline{\chi}^{top}_{\omega,f}(e)$ is strictly increasing in its continuity

577: points.  The function fails to be continous when new pairs of topological saddles are

578: created or destroyed, namely in the non-proper values of $f$, and it jumps exactly by the

579: height of the cylinder(s) created or destroyed.

580: \end{proof}

581: %

582: \begin{corollary}

583: If $\omega$ induces on $\oM_e$ a Dynnikov decomposition of genus bigger than 1, then

584: $e_1(\omega)=e_2(\omega)=e_0$, i.e. $\omega$ induces only closed leaves at any other

585: level.

586: \end{corollary}

587: %

588: \begin{corollary}

589: At almost all levels of $f$ the measure of the ``ergodic'' directions is zero.

590: \end{corollary}

591: %

592: We have now all ingredients to prove theorem~\ref{thm:Dyn}:

593: %

594: \begin{proof}

595: We will just discuss the case of ``full ergodicity'', namely the case of directions giving

596: rise to leaves dense on the whole polyhedron; the same line of arguments extends to the

597: lesser ergodic cases.

598: \par

599: If $\omega$ induces on $\oM$ fully ergodic leaves, then of course no cylinder can appear

600: and therefore, for any Morse PL function $f$ such that $\oM=f^{-1}(0)$, it must happen that

601: $\overline{\chi}^{top}_{\omega,f}(0)=0$. Let us consider now $\overline{\chi}^{top}$ as a

602: function of $\omega$ and $e$: then the surface

603: $X=(\overline{\chi}^{top}_f)^{-1}(0)\subset\RPt\times\bR$, for a generic function $f$, is

604: transversal to the sections $\RPt\times\{e\}$: indeed in a projective chart, say $h^z=1$,

605: we have that $\partial_{h^x}\overline{\chi}^{top}_{\omega,f}=\sum_i w_i (c^{top}_i)^x$ and

606: similarly for $h^y$, so that the points on $X$ where the gradient is zero are exactly the

607: points where $\gamma^{top}_{\omega,f}(e)=\sum_i w_i c^{top}_i=0$.

608: This condition is non-generic, that finally proves the claim of the theorem.

609: \end{proof}

610: %

611: \subsection{Structure of the Stereographic Map of surfaces and triply periodic functions}

612: %

613: The results of the previous two sections show that the SM $\cS(\oM)$ of a generic surface

614: is the disjoint union of a countable set of open sets $\cD_l(\oM)$ (``islands''), each

615: labeled by a $l\in PH_2(\Tt,\bZ)$, immersed in a sea of directions that give rise only to

616: closed leaves. According to our intuition of the system and the numerical experiments made

617: to date, we conjecture that generically the number of islands is finite, but no rigorous

618: proof of this fact exists. As matter of fact, this structure is exactly the one guessed,

619: from symmetry consideration, by the physicist I.M. Lifschitz and his Karkov school about

620: fifty years ago~\cite{LP59,LP60}.

621: \par

622: The boundaries of the islands are reached when the last pair of genus-1 components collide

623: because of a cylinder collapse and therefore are charaterized by the presence of (at

624: least) a pair of inequivalent critical points on the same leaf. The set of these

625: directions is the countable union of the curves $<\bH,\hat c_i-\hat c_j>=0$, $i\neq j$; such

626: curves in the polyhedra case are all straight lines, so that every island is actually

627: a (not necessarily convex) polygon. There are reasons to believe that these polygons

628: are convex for low genus, i.e. at least for genus 3 and 4, but it is easy to build examples

629: of high-genus polyhedra with islands that are either non connected or connected but non convex

630: or even connected but non simply connected.

631: \par

632: A crucial observation by Dynnikov~\cite{Dyn97} allows to associate a SM also to triply

633: periodic functions:

634: %

635: \begin{theorem}

636: Be $f$ a Morse triply periodic function: then, if $\omega$ induces on $\oM_{e0}$ a Zorich

637: decomposition $\cZ_{e0}$, it induces a Zorich decomposition $\cZ_e$ for all

638: $e\in(e_1(\omega),e_2(\omega))$ and all these decomposition share the same soul.

639: \end{theorem}

640: %

641: \begin{definition}

642: The island $\cD_l(f)$ corresponding to the label $l$ is the union of the corresponding

643: islands of $f$'s level sets: $\cD_l(f)=\cup_{e\in\bR}\cD_l(\oM_e)$.

644: The SM of $f$ is the union of all its islands: $\cS(f)=\cup_l\cD_l(f)$

645: \end{definition}

646: %

647: The SM corresponding to functions are generically drammatically different from those

648: corresponding to surfaces. 

649: \par

650: First of all, since rational directions induce necessarily Zorich decompositions, the set

651: of islands $\cS(f)$ is now always dense in $\RPt$, and of course

652: lemma~\ref{thm:interval} also tells us that in this case the sea of directions giving rise

653: to closed leaves only dried up, since every direction either belongs to an islands (or its

654: boundary) or is ergodic.

655: \par

656: Moreover, the following property shows that the islands can be sorted in a rather complex

657: way:

658: %

659: \begin{theorem}

660: Generically every two zones meet transversally and in a countable number of points.

661: \end{theorem}

662: %

663: \begin{proof}

664: No point belonging at the same time to two different zones $\cD_{l_1}$ and $\cD_{l_2}$ can

665: have irrationality degree bigger than 2, since it must contain the integer direction

666: $l_1\times l_2$. In the smooth case this would be enough, since the boundaries are smooth

667: curves and they generically contain only a countable number of 2-irrational points and no

668: rational point. In the PL case though the boundaries are actually segments of straight

669: lines and threfore they are actually contained in the set of the directions with

670: irrationality degree smaller than one.

671: \par

672: Nevertheless, the theorem holds also in this case for the following reason: since every

673: direction at the boundary between two zones is perpendicular to the direction $l_1\times

674: l_2$, then their set is the straight line (in $\RPt$) passing through $l_1$ and

675: $l_2$. Generically none of the two labels falls on the boundary and therefore two zones

676: can meet in a number of points not bigger than the number of sides of the island with the

677: smaller number of sides.

678: \end{proof}

679: %

680: \begin{corollary}

681: Either there is a single zone, i.e. it exists a label $l$ such that $\cD_l(f)=\RPt$, or there

682: are countably many zones and they are dense in the whole projective plane.

683: \end{corollary}

684: %

685: Since the islands meet trasversally, the non trivial SM will look like 2-dimensional

686: Cantor sets.  Retrieving numerically such fractals is not trivial since it involves, in

687: general, to analyze the system at several values of $f$, but there is a class of

688: interesting (non generic) cases in which it is actually enough to analyze a single level

689: surface of $f$ to get the entire fractal picture:

690: %

691: \begin{theorem}

692: \label{thm:intext}

693: Be $\oM$ a polyedron whose interior is equal to its exterior, modulo the group

694: $G\simeq\Rt\times\bZ_2$ of translations and inversion of the three axes.  Then

695: $\cS(\oM)=\cS(f)$ for any function having $\oM$ as (connected component of a) level set.

696: \end{theorem}

697: %

698: \begin{proof}

699: By symmetry, we can build a function $f$ such that $\oM=f^{-1}(0)$ and that $f^{-1}(e)$ is

700: equal to $f^{-1}(-e)$, modulo $G$. Since also the bundles of parallel planes are invariant

701: by $G$, it turns out that for such an $f$ the interval $I(\omega)$ of existence of open

702: orbits relative to any 1-form $\omega$ is of the form $I(\omega)=[-a(\omega),a(\omega)]$

703: and therefore $0\in I(\omega)$, $\forall\omega\in\RPt$, i.e. at the zero level every

704: $\omega$ induces open leaves.

705: \end{proof}

706: %

707: In particular, all triply periodic functions $f$ such that $f(c-x,c-y,c-z)=-f(x,y,z)$

708: belong to this class, and indeed the only fractals analyzed numerically to date are

709: relative to this kind of functions.

710: \par

711: Finally, we cite an important property that ties the set of all labels relative to the islands

712: of the SM of a function with the set of ergodic direction~\cite{DL03b,DL05b}:

713: %

714: \begin{theorem}

715: \label{thm:labels}

716: The closure of the set of all labels is the disjoint union of the

717: set of all zones boundaries and the set of ergodic directions.

718: \end{theorem}

719: %

720: \section{A concrete case study}

721: \label{sec:sp}

722: %

723: As pointed out in the previous section, to date a picture of the fractal has been

724: numerically produced for only two functions, an analytical one and a piece-wise quadratic

725: one~\cite{DL03b}.

726: \par

727: The analytical one is $\mf(x,y,z)=\cos(2\pi x)+\cos(2\pi y)+\cos(2\pi z)$, invariant with

728: respect to translations by integers $\Gamma=\bZ^3\subset\Rt$, that gives rise

729: to genus-3 level surfaces in the range $(-1,1)$ and spheres at every other non-critical

730: level.  This function represents the simplest non-trivial case possible from the

731: topological point of view, since any triply periodic connected surface of genus smaller

732: than 3 lies between two parallel planes and therefore the aymptotics of plane sections is

733: easily found.  Its zero level is rather special: it is known as the Schwarz primitive

734: function (or plumber's nightmare) and was studied by Schwarz in 1890 as one of the first

735: examples of triply periodic minimal surface.

736: \par

737: From the computational point of view, $\cS\cP=\mf^{-1}(0)$ has three important properties:

738: \begin{enumerate}

739: \item[{\textbf SP1}] its interior is a translate of its exterior, so that $\cS(\cS\cP)=\cS(\mf)$;

740: \item[{\textbf SP2}] it is invariant with respect to the natural action of the tetrahedral group $T_d$ 

741:       on the unitary cube, so that the whole SM can be obtained, for example, extending 

742:       by symmetry to the whole $\RPt$ the data obtained for 

743:       the triangle with vertices $[(0,0,1)]$, $[(1,0,1)]$ and $[(1,1,1)]$;

744: \item[{\textbf SP3}] the two cylinders, one negative and one positive, have the same height, so that it

745:       is enough to examine just one of the four topological critical points at the base

746:       of the two cylinders to retrieve all information about the structure of the foliation.

747: \end{enumerate}

748: %Every function with the same symmetries of $\mf$ has a zero level satisfying the same

749: %properties.

750: \par

751: The piecewise quadratic function is $\mg(x,y,z)=\sum\bar \mg(x^i)$, where $\bar\mg$ is the

752: simplest piecewise quadratic function having the same symmetries of the cosine

753: function. Its level sets have the same behaviour as the function above but the expression

754: of the critical points as function of the direction of the 1-form and the level of the

755: function is so simple that allows a comparison between analytical and numerical data also

756: at levels different from zero.

757: \par

758: Nevertheless, in both cases the number of triangles needed to describe in sufficient

759: detail the surface is so big (between $10^5$ and $10^6$) to make impossible to improve the

760: resolution of the results obtained in~\cite{DL03b}, at least until a new algorithm is found

761: or some ``ad hoc'' trick used.

762: %

763: \subsection{The polyhedron}

764: %

765: \begin{figure}

766: \begin{center}

767: \includegraphics[bb=200 512 400 712,width=3.8cm]{lincos.ps}

768: %\epsfxsize=3.8cm\epsfbox[200 512 400 712]{lincos.ps}

769: \end{center}

770: \caption{%

771: The simplest PL approximation of the cosine function.

772: }

773: \label{fig:lincos}

774: \end{figure}

775: %

776: A natural way to improve the resolution of the numerical analysis of the problem is to

777: consider PL functions, since in this case the number of triangles needed to describe their

778: level surfaces can be as low as of the order of $10^1$. Moreover, the description of the

779: surfaces is in this case exact rather than an approximation.

780: \par

781: The simplest case to study, in order to take advantage of the extremely convenient

782: properties evidenced in the $\cS\cP$ case, is the PL function

783: $\mh(x,y,z)=\sum_i\bar\mh(x^i)$, where $\bar\mh$ is the function shown in

784: Fig.~\ref{fig:lincos}.

785: %, so that the stereographic

786: %map of its $0$ level set $\cP_0$ will give the whole SM of the function.

787: %\par

788: The polyhedron $\cP_0=\mh^{-1}(0)$ is a PL embedding of a genus-3 surface in $\Tt$ having

789: 8 hexagonal faces, 20 edges and 12 vertices; the smallest triangulation for $\cP_0$ takes

790: $4\cdot8=32$ triangles, $32\cdot3/2=48$ edges and $12$ vertices, that gives the expected

791: Euler characteristic $\chi(\cP_0)=F-E+V=-4=2-2\cdot3$.

792: The basic cell of the lift $\overline\cP_0\subset\Rt$ in the unit cube is a truncated octahedron

793: (Fig.~\ref{fig:pol}); it is noteworthy to notice that, like its smooth analog, also this surface

794: is, in the discrete sense, a minimal surface~\cite{Way04}.

795: \par

796: Since exactly four edges meet at every vertex, given any 1-form $\omega$ in general

797: position with respect to $\cP_0$ only saddles with index $-1$ may arise and therefore

798: there are always four vertices that are critical for such $\omega$. If the point $p_1$ is

799: such a vertex, then the remaining three critical points are $p_{2,3}=(1/2,1/2,1/2)\pm p_1$

800: and $p_4=(1,1,1)-p_1$. In particular in our numerical study we sample the set of 1-forms

801: $\omega=(h^x,h^y,h^z)$ such that $h^x/h^z\in[0,1]$ and $h^y/h^z\in[0,1]$, for which the four

802: critical points are $(0,.5,.75)$, $(.5,1,1.25)$, $(.5,0,.25)$ and $(1,.5,.25)$.

803: %

804: \begin{figure}

805: \begin{center}

806: \includegraphics[width=5cm]{polBasic.eps}

807: %\epsfxsize=5cm\epsfbox{polBasic.eps}

808: \hskip1.cm

809: \includegraphics[width=5cm]{polUnCov.eps}

810: %\epsfxsize=5cm\epsfbox{polUnCov.eps}

811: \end{center}

812: \caption{%

813: Plot of the Truncated Octahedron inside $\Tt$ (left) and of part of its image in the

814: universal covering.

815: }

816: \label{fig:pol}

817: \end{figure}

818: %

819: \subsection{The algorithm}

820: %

821: In order to generate an approximate picture of the fractal, it is enough to produce an

822: algorithm able to evaluate the label, if any, associated to a given 1-form. Since

823: obviously no calculator can deal with irrational numbers, the numerical study will be

824: limited to rational 1-forms; luckily this is not a big restriction, since anyway rational

825: directions are dense in every stability zone.

826: \par

827: Note that the algorithm used for the numerical study of the PL case is a simplified

828: version of the more general algorithm we developed to study smooth surfaces of genus

829: three~\cite{DL03b}, since in this case we know a priori the position of all critical points

830: and moreover we know their position exactly, so that we do not have to correct ``by hand''

831: the topology of the critical section.

832: \par

833: The basic idea to retrieve the label, as suggested to me by I. Dynnikov, is that the soul

834: $l\in PH_2(\Tt,\bZ)$ associated to the Zorich decomposition $\cZ$ induced by $\omega\in\cD_l$ is

835: in 1-1 correspondance with the rank-2 sublattice of $H_1(\Tt,\bZ)$ obtained as the image,

836: through the map $i_*:H_1(\oM,\bZ)\to H_1(\Tt,\bZ)$, of the open leaves in $\oM$ that have

837: zero intersection number with the closed leaves populating the cylinders of $\cZ$.

838: Indeed, every cycle lying on the interior of a $N_j$ component of $\cZ$ has no

839: intersection with the closed leaves that form the cylinders $\cC_i$, and since all $N_j$

840: are homologous to each other (modulo sign) the image of all these cycles in $\Tt$ must

841: have rank-2; on the other side, there is an obvious 1-1 correspondance between rank-2

842: sublattices of $H_1(\Tt,\bZ)=\Gamma\simeq\Zt$ and the 2-tori embedded in $\Tt=\Rt/\Gamma$,

843: since every such 2-torus can be spanned by a pair of independent rational directions and

844: viceversa.

845: \par

846: The following algorithm {\textbf N} works for genus-3 polyhedra satisfying properties

847: {\textbf SP1-3}, in particular for $\cP_0$:

848: \par

849: {\textbf Input}: $\oM$ - the polyhedron; $\omega=(l,m,n)\in\Zt$ - the 1-form; $x$ - a critical

850: point of $\omega$ with respect to $\oM$; $\pi_{\omega,x}$ - the plane perpendicular to

851: $(l,m,n)$ and passing through $x$.

852: \par

853: {\textbf Output}: $c_{1,2}$ - the two critical loops%

854: \footnote{since numerically we can study only the 1-rational case, in $\Tt$ the saddles are

855: always wedges of circles, i.e. all critical branches close back to the critical point; according to

856: whether these loops are or not homotopic to zero in $\Tt$, their $\Rt$ lift will be open or close.}%

857: ; $h_{1,2}$ - the homology classes of $c_{1,2}$ in $\oM$; $H_{1,2}$ - the homology classes

858: of $c_{1,2}$ in $\Tt$.

859: \par

860: {\textbf Algorithm}:

861: \begin{itemize}

862: \item[{\textbf N1}] retrieve the intersection between $M$ and $\pi_{\omega,x}$;

863: \item[{\textbf N2}] check that there are exactly four critical branches and follow them

864: 	by periodicity, otherwise exit;

865: \item[{\textbf N3}] if no other critical point is met along the path, so that the four

866: 	branches are arranged in a pair of critical loops, store the two loops in the 

867: 	variables $c_{1,2}$, otherwise exit;

868: \item[{\textbf N4}] evaluate the homology class of $c_{1,2}$ in $\Tt$ and in $M$

869: 	(this is actually done while executing N2 to speed up the computations time);

870: \item[{\textbf N5}] if the saddle is half-open, i.e. if exactly one among $H_{1,2}$ is zero,

871: 	then associate to $\omega$ the complementary $h$ triple, otherwise exit.

872: %check the homology class of the two loops in $\Tt$

873: \end{itemize}

874: %

875: The main outcome of the algorithm is of course the label associated to $\omega$. The fact that

876: this label is a triple of integers is very important, since an integer evaluated numerically

877: with an error smaller than .5 becomes actually an exact measure.

878: \par

879: We implemented this algorithm in a C++ library named NTC \footnote{http://ntc.sf.net/} built

880: over an Open Source C++ library named VTK \footnote{http://www.vtk.org/}. The choice of the 

881: language comes from the fact that VTK provides the basic geometric environment and algorithms 

882: needed by the problem, mainly the capability of generating meshes for isosurfaces and

883: evaluating intersections between geometric objects. The inheritance mechanism of the 

884: C++ language allows to use transparently all functions of a library, hence we used VTK as

885: a starting point and implemented in NTC the routines to deal with periodicity and

886: evaluate the homology classes.

887: \par

888: No serious attempt to evaluate the error on such calculations has been made to date since

889: no need for it manifested. As check for the reliablity of the result were rather used indirect

890: evidences:

891: \begin{itemize}

892: \item the agreement of the biggest zones with their analytical boundary

893: (Fig.~\ref{fig:sm100}b), obtained through the independent algorithm {\textbf A};

894: \item the symmetry of the final picture with respect to the diagonal

895: (Fig.~\ref{fig:sm1000}), symmetry that was in no way used in the numerical calculations;

896: \item the agreement of the fractal picture with the labels plot (Fig.~\ref{fig:labels}).

897: \end{itemize}

898: \par	

899: The exploration of the SM was performed in the square $[0,1]^2$ of the projective chart

900: $(\omega_x/\omega_z,\omega_y/\omega_z)$ by evaluating the label associated to every

901: direction at the vertex of a uniform grid of step $r$ and it was repeated for the values

902: $r=10^2,10^3,10^4$.  Samplings with $r=10^2,10^3$ have been succesfully performed also for

903: the previous two functions \cite{DL03b} but the CPU time needed to reach $r=10^4$ in that

904: case was way too big.  It is because of the rather small number of triangles needed to

905: describe $\cP_0$ that the computation became doable.

906: %

907: %\subsection{Numerical results for $r=10^2$ and analytical boundaries}

908: \subsection{Numerical results for $r=10^2$}

909: %

910: This resolution is the lowest one that allows to have a hint of the structure of the

911: fractal.  About $r$ sections are needed to follow the critical branches for a generic

912: direction $(m,n,r)$, $m,n\in\{1,\dots,r\}$, that takes a time of $~.5s$ on a $~1GHz$ CPU

913: for the evaluaton of a single label and $10^4\times.5s\simeq1h$ for sampling the $10^4$

914: directions of the grid (Fig.~\ref{fig:sm100}).

915: \par

916: Even from this rough picture it is rather evident a further symmetry of the picture,

917: namely the one with respect to the antidiagonal of the square. This symmetry does not

918: come, like the others, from the tetrahedral group $T_d$ but it is rather of topological

919: nature. The numerical evidence is that, if a 1-form $(m,n,r)$ is labeled by $L$, then its

920: symmetric $(1-n,1-m,r)$ is labeled by $L+(1,1,0)$ but no proof of this fact is known.

921: \par

922: In order to verify the correctness of the algorithm, we found ``by hand'' the analytical

923: boundaries for the biggest zones and compared them with the numerical results

924: (Fig.~\ref{fig:sm100}b). 

925: The following algorithm, aimed at the cases similar to the cosine one, is a slight modification

926: of the original algorithm introducted and used by Dynnikov in~\cite{Dyn96}:

927: \par

928: {\textbf Algorithm A}

929: \begin{itemize}

930: \item[A1] fix a 1-form $\omega=(m,n,r)\in\bZ^3$ inside some zone 

931: 	(e.g. extracting it from the 

932: 	experimental, guessing it from symmetry arguments or simply by trial and error);

933: \item[A2] retrieve the critical section of $\omega$ passing through one of its

934: 	critical points $c$ and make sure it is half open, otherwise exit;

935: \item[A3] evaluate the homology class $l$ of the closed critical leaf $C$;

936: \item[A4] rotate $\omega$ around some direction till the cylinder of which $C$

937: 	is a base collapses and identify the critical point $c^\prime$ that is now

938: 	connected to $c$ through a saddle connection;

939: \item[A5] the equation $<\bH,c-c^\prime>=0$ contains one of the sides of the island;

940: 	follow it in one direction till four critical points fall over the critical closed

941: 	leaf: this is the point when a sides and a new one start; repeat this

942: 	step till the island boundary close up on themselves. 

943: \end{itemize}

944: %

945: \begin{figure}

946: \begin{center}

947: \includegraphics[bb=200 462 400 722,width=4.5cm]{2.4.5.ps}

948: %\epsfxsize=4.5cm\epsfbox[200 462 400 722]{2.4.5.ps}

949: \end{center}

950: \caption{%

951: A close-up of the island $\cD_{(2,4,5)}(\cP_0)=\cD_{(2,4,5)}(\mh)$.

952: Inside the island the pairs of critical points at the base of each cylinder are locally

953: constant. In correspondance to each sides, there are three different pairings sorted in

954: open subsets, labeled in the picture by roman numerals, separated by straight lines

955: segments correspoding to directions $\omega$ for which the bases of the positive and

956: negative cylinders collide, resulting in a saddle connection between two critical

957: points. These three segments meet in the single point $(.4,.8)$, that in this case happens

958: to be exactly the direction of the label (this property though is not generic).

959: In Figure~\ref{fig:sections} we show in detail the transition between two different pairs

960: within this island.

961: }

962: \label{fig:2.4.5}

963: \end{figure}

964: %

965: \par

966: Since the genus is three, only two cylinders may appear and they will be of opposite sign.

967: The pairs of critical points at the base of cylinders are locally constant; in the square

968: under investigation the pairings are $p_1,p_4$ for the positive cylinder and $p_2,p_3$ for

969: the negative one, so that boundaries are always given by an equation like

970: $<\bH,p_1-p_4+L>=0$. See Figures \ref{fig:2.4.5} and \ref{fig:sections} for a concrete

971: example worked out in detail.

972: \par

973: %

974: \begin{figure}

975: \begin{center}

976: \vbox{\halign{\hfill#\hfill&\hfill#\hfill&\hfill#\hfill\cr

977: \includegraphics[width=4cm]{polSection.x396.y81.eps}

978: &

979: \includegraphics[width=4cm]{polSection.x41.y81.eps}

980: &

981: \includegraphics[width=4cm]{polSection.x43.y82.eps}

982: \cr

983: \noalign{\vskip3pt}

984: {\textbf a.}\ \ $\omega = (396,810,1000)$

985: & 

986: {\textbf b.}\ \ $\omega = (41,81,100)$

987: & 

988: {\textbf c.}\ \ $\omega = (43,82,100)$

989: \cr 

990: \noalign{\vskip3pt}

991: \includegraphics[width=4cm]{polSection.x39.y81.eps}

992: &

993: \includegraphics[width=4cm]{polSection.x40.y80.eps}

994: &

995: \includegraphics[width=4cm]{polSection.x43.y81.eps}

996: \cr

997: \noalign{\vskip3pt}

998: {\textbf d.}\ \ $\omega = (39,81,100)$

999: & 

1000: {\textbf e.}\ \ $\omega = (40,80,100)$

1001: & 

1002: {\textbf f.}\ \ $\omega = (43,81,100)$

1003: \cr 

1004: \noalign{\vskip3pt}

1005: \includegraphics[width=4cm]{polSection.x37.y78.eps}

1006: &

1007: \includegraphics[width=4cm]{polSection.x38.y78.eps}

1008: &

1009: \includegraphics[width=4cm]{polSection.x43.y80.eps}

1010: \cr

1011: \noalign{\vskip3pt}

1012: {\textbf g.}\ \ $\omega = (37,78,100)$

1013: & 

1014: {\textbf h.}\ \ $\omega = (38,78,100)$

1015: & 

1016: {\textbf i.}\ \ $\omega = (43,80,100)$

1017: \cr 

1018: }}

1019: \end{center}

1020: \caption{%

1021: Significant examples of critical sections of $\cP_0$ for $\omega\in\cD_{(2,4,5)}(\cP_0)$.

1022: At the ``center'' of the island ({\textbf e}) there is a saddle connection between the four

1023: critical points (starting from the highest and turning clockwise) $N=p_1$,

1024: $E=p_2+(3,-3,1)$, $S=p_1+(1,-3,2)$ and $W=p_2+(1,-2,1)$. In the subzones $I$, $II$

1025: and $III$ (Fig.~\ref{fig:2.4.5}) the pairs of critical points at the base of the positive

1026: cylinder are, respectively, $N$ and $p_4+(2,-4,1)$ ({\textbf i}), $S$ and $p_3+(2,-2,0)$

1027: ({\textbf b},{\textbf c}) and $E$ and $p_3+(-1,-1,0)$ ({\textbf d}, {\textbf g}).

1028: The separating segments correspond respectively to saddle connections between the pairs of

1029: critical points $S$ and $W$ ({\textbf a}), $N$ and $S$ ({\textbf f}) and $N$ and $E$ ({\textbf h}).

1030: }

1031: \label{fig:sections}

1032: \end{figure}

1033: %

1034: Finally, a picture of the whole fractal can be obtained through the natural free action on

1035: $\RPt$ of the tetrahedral group $T_d$, whose order is 24 (Fig.~\ref{fig:sm100}(d)).

1036: %

1037: \begin{figure}

1038: \begin{center}

1039: \includegraphics[width=12cm]{polSM.100.eps}

1040: %%\epsfxsize=15cm\epsfbox{polSM.100.eps}

1041: %\epsfxsize=12cm\epsfbox{polSM.100.eps}

1042: \end{center}

1043: \caption{%

1044: Numerical plot of the square $[0,1]^2$ of the SM $\cS(\cP_0)=\cS(\mh)$ in the projective

1045: chart $h^z=1$ at a resolution $r=10^2$.  The color of the islands goes from blue to red as

1046: the norm of the label grows. In the picture are displayed the 106 islands with at least

1047: four points out of the total 1741 islands found. The missing points that is possible to

1048: see in the interior of some of the islands are due to failure of the numerical algorithm

1049: {\textbf N}, e.g. because of the presence of saddle connections.  The running time for the

1050: $10^4$ steps cycle needed to retrieve these data takes about $1h$ on a Pentium \~1GHz CPU.

1051: }

1052: \label{fig:sm100}

1053: \end{figure}

1054: %

1055: \begin{figure}

1056: \begin{center}

1057: \includegraphics[width=16cm,bb=120 280 612 732]{polSM100.BdLabels.ps}

1058: %\epsfxsize=16cm\epsfbox[120 280 612 732]{polSM100.BdLabels.ps}

1059: %%\epsfxsize=17cm\epsfbox[120 280 612 732]{polSM100.BdLabels.ps}

1060: \end{center}

1061: %\epsfxsize=14cm\epsfbox{polSM100.BdLabels.ps}

1062: \caption{%

1063: Analytical boundaries of the biggest islands found in Fig.~\ref{fig:sm100} found using the

1064: algorithm {\textbf A}. All of their boundaries are straight lines segments and the corresponding

1065: label has been reported, when possible, inside the zone itself.

1066: }

1067: \label{fig:sm100an}

1068: \end{figure}

1069: %

1070: \begin{figure}

1071: \begin{center}

1072: \includegraphics[width=12cm]{polSM.100.PtsBd.eps}

1073: %\epsfxsize=12cm\epsfbox{polSM.100.PtsBd.eps}

1074: \end{center}

1075: \caption{%

1076: Comparison between analytical (Fig~\ref{fig:sm100an}) and numerical 

1077: (Fig~\ref{fig:sm100}) boundaries.

1078: }

1079: \end{figure}

1080: %

1081: \begin{figure}

1082: \begin{center}

1083: \includegraphics[width=12cm]{polSMDisc.100.eps}

1084: %\epsfxsize=12cm\epsfbox{polSMDisc.100.eps}

1085: \end{center}

1086: \caption{%

1087: Fractal image obtained by letting the tetrahedral group $T_d$ act on the square

1088: in Fig.~\ref{fig:sm100} and then projecting on the disc through the stereographic map.

1089: }

1090: \end{figure}

1091: %

1092: \subsection{Numerical results for $r=10^3$}

1093: %

1094: This is the highest resolution reached in~\cite{DL03b}.  In this case for each generic

1095: direction are needed about $10^3$ sections to follow the critical branches, that takes a

1096: time of $~2.5s$ on a $~1GHz$ CPU for the evaluaton of a single label and therefore about

1097: $5\times10^2\times1h\simeq1month$ for sampling the $10^6$ directions of the grid

1098: (Fig.~\ref{fig:sm1000}).

1099: \par

1100: %

1101: \begin{figure}

1102: \includegraphics[width=12cm]{polSM.1000.min5.eps}

1103: %\epsfxsize=12cm\epsfbox{polSM.1000.min5.eps}

1104: \caption{%

1105: Numerical plot of the square $[0,1]^2$ of the SM $\cS(\cP_0)=\cS(\mh)$ in the projective

1106: chart $h^z=1$ at a resolution $r=10^3$. The color of the islands goes from green to red as

1107: the norm of the label grows. In the picture are displayed the 1625 islands with at least

1108: four points out of the total 10725 islands found. The missing points that is possible to

1109: see in the interior of some of the islands are due to failure of the numerical algorithm

1110: {\textbf N}, e.g. because of the presence of saddle connections.  The running time for the

1111: $10^6$ steps cycle needed to retrieve these data takes about 1 week on a Pentium 1GHz CPU.

1112: }

1113: \label{fig:sm1000}

1114: \end{figure}

1115: %

1116: A time of the order of a month is of course rather long, but thanks to the diffusion of

1117: the Linux OS, and therefore of the possibility to build big Unix clusters for cheap using

1118: PCs rather than workstations, this is still not too bad since it is easy to lower by 

1119: a factor 10 the computations time just by dividing the cycle over as many PCs.

1120: This way the running time goes down to just three days, that is a rather acceptable time.

1121: \par

1122: We point out that the situation is radically different in the smooth case: indeed in that

1123: case there is another variable to consider, that is the resolution of the mesh of the

1124: surface, that must be increased together with the grid resolution to avoid errors in the

1125: topology of the curve. The plane sections giving the complete intersection between the

1126: surface and a 2-torus with homology class $(l,m,r)$, $l,m\in[0,r]$, can be as close as

1127: $1/r$ and therefore, if the mesh is too rough, there is the risk that the program will 

1128: jump on the wrong slice. 

1129: \par

1130: Concrete tests show that a mesh resolution of $30$, meaning that the mesh is produced by

1131: dividing the unit cube in a $30^3$ uniform grid, is enough for the $r=10^2$ case but it

1132: must raised to at least $60$ for the $r=10^3$ case, increasing the time for a single label

1133: evaluation to $~15s$, an order of magnitude bigger than in the PL case. This brings back

1134: the time to $~3$ months for the execution, that is indeed the order of the time spent for the

1135: $r=10^3$ calculations made for~\cite{DL03b}.

1136: \par

1137: From the picture~\ref{fig:sm1000} it is rather evident the symmetry with respect to the

1138: anti-diagonal. Apart from this, the picture looks qualitatively very similar to the

1139: pictures found in the previous two cases at the same resolution.

1140: %

1141: \subsection{Numerical results for $r=10^4$}

1142: %

1143: \begin{figure}

1144: \begin{center}

1145: \includegraphics[width=4.5cm,bb=200 462 400 722]{boxCount.ps}

1146: %\epsfxsize=4.5cm\epsfbox[200 462 400 722]{boxCount.ps}

1147: \end{center}

1148: \caption{%

1149: Box counting evaluation of the Hausdorff dimension of $\cS(\cP_0)$ with the $r=10^4$ data.

1150: $N_n$ is the number of squares of side $2^n$ needed to cover the complement of

1151: $\cS(\cP_0)$, the angular coefficient of the linear fit provides the dimension estimate.

1152: The same evaluation made with the $r=10^3$ data gives a very close result, and also

1153: restricting the fit by canceling a few points at the extremes does not change considerably

1154: the estimate.

1155: }

1156: \label{fig:boxcount}

1157: \end{figure}

1158: %

1159: Increasing by another order of magnitude the resolution, we increase by an order of magnitude

1160: the number of sections needed to follow a generic leaf, resulting in another factor 5 in the

1161: running time for a single evaluation of a 1-form label, that is now $~10s$, so that the 

1162: total running time on a $1GHz$ CPU reaches $5\cdot10^2\cdot20d=10^4d\simeq30years$.

1163: \par

1164: Such a big running time is rather scary and suggests that there is no hope to go up by

1165: an order of magnitude in resolution without changing some significant algorithm step.

1166: Nevertheless, this big time can be once again brought down to something reasonable

1167: by running the code in 20 PCs and by restricting the numerical analysis to the upper triangle

1168: of the square $[0,.5]\times[.5,1]$ (that reduces computational time by a further factor 8).

1169: Thanks to all these expedients, the running time goes down by two orders of magnitude,

1170: reaching about 3 months, that is indeed about the time that took to us to collect the

1171: $r=10^4$ data.

1172: \par

1173: Note that for the smooth cases this would not be enough, since we must raise the mesh

1174: resolution to $10^2$ and the time for the single evaluation goes up to $50s$; this,

1175: together with the fact that there's a further factor two due to the lack of symmetry with

1176: respect to the antidiagonal, brings the total running time to $~3$ years, definitely not

1177: realisticly affordable.

1178: \par

1179: In Figures

1180: \ref{fig:polSM10000.x0-1000.y9000-10000.eps}-\ref{fig:polSM10000.x4000-5000.y5000-6000.eps} are

1181: shown the numerical results, from which it is clear beyond any doubt that the SM has a

1182: fractal-like self-repeating structure, even though no explicit construction is known for

1183: it.

1184: \par

1185: Numerical evaluations of the Hausdorff dimension $d_{\cP_0}$ of the fractal set, namely

1186: the complement of $\cS(\cP_0)$ in $\RPt$, has been performed using the box counting

1187: technique (Fig.~\ref{fig:boxcount}) and a sort of ``area distribution'' technique

1188: (Fig.~\ref{fig:areasdistr}).

1189: \par

1190: The first, and more reliable, technique involves partitioning the square $[0,1]^2$ in

1191: $2^n$ identical squares and evaluating the number of squares needed to cover the fractal.

1192: With the $r=10^3$ data, $n$ can get as big as $\log_2(10^3)\simeq10$ and a linear fit

1193: gives an evaluation of $d_{\cP_0}\simeq1.93$. The $r=10^4$ data allow $n$ to go up to

1194: $13$, representing the deeper results on the Hausdorff dimension of such fractals to date;

1195: figure~\ref{fig:boxcount} shows that the scaling law is linear to a high degree of

1196: accuracy and a linear fit gives again an estimate of $d_{\cP_0}\simeq1.93$, making us

1197: rather confident in the accuracy of this numerical result.

1198: \par

1199: \begin{figure}

1200: \begin{center}

1201: \includegraphics[width=8.5cm,bb=100 530 400 722]{areaDistr.ps}

1202: %\epsfxsize=8.5cm\epsfbox[100 530 400 722]{areaDistr.ps}

1203: \end{center}

1204: \caption{%

1205: ``Area distribution'' evaluation of the Hausdorff dimension of $\cS(\cP_0)$ with the

1206: $r=10^4$ data. $N_n$ is the number of islands whose area is between $2^{-n}$ and

1207: $2^{-n-1}$. The estimate of the Hausdorff dimension is given by the double of the angular

1208: coefficient of the linear fit and varies by $\pm0.2$ by restricting even by little the

1209: number of points on which the fit is made.

1210: }

1211: \label{fig:areasdistr}

1212: \end{figure}

1213: %

1214: \begin{figure}

1215: \begin{center}

1216: \includegraphics[width=5cm,bb=200 362 400 622]{dynConj.ps}

1217: %\epsfxsize=5cm\epsfbox[200 362 400 622]{dynConj.ps}

1218: \end{center}

1219: \caption{%

1220: Logarithmic plot of the islands' areas versus the norm of the corresponding label.

1221: There is a very good agreement between the numerical data and the Dynnikov conjecture

1222: claiming that $\cA(\cD_l)\leq C/\|l\|^3$.

1223: }

1224: \label{fig:dynConj}

1225: \end{figure}

1226: %

1227: The second one involves counting the number of zones whose size is between $b^n$ and

1228: $b^{n+1}$, where $b>1$. Computations with sevaral small values, between 2 and 1, were

1229: performed and all of them gives rough estimate of $d_{\cP_0}\simeq1.76$, further than

1230: expected from the evaluation given by the box counting. It is not clear to us the reason

1231: of this disagreement, that did not manifest for the smooth cases~\cite{DL03a}, but it is

1232: not impossible that it may be due simply to the fact that this evaluation stabilizes at

1233: higher resolutions and therefore this evaluation is at the current state of things more or

1234: less unreliable (as simple numerical tests testify).

1235: %

1236: \par

1237: Finally, the $r=10^4$ data provide enough detail to test numerically a conjecture by

1238: Dynnikov~\cite{Dyn99} and give a graphical evidence of theorem~\ref{thm:labels}.

1239: The Dynnikov conjecture claims that the area of the islands satisfy a relation

1240: $\cA(\cD_l(\cM))\leq C/\|l\|^3$ for some positive real number $C$ depending only on the

1241: polyhedron$\cM$; the numerical data suggests that the exponent 3 cannot be improved any further  

1242: (see Fig.~\ref{fig:dynConj}).

1243: \par

1244: %

1245: \begin{figure}

1246: \includegraphics[width=12cm]{polLabels.eps}

1247: %\epsfxsize=12cm\epsfbox{polLabels.eps}

1248: \caption{%

1249: Plot of the labels of the 494041 islands found in the numerical analysis, at the

1250: resolution $r=10^4$, of the portion of the SM $\cS(\cP_0)$ contained in the triangle of

1251: vertices $(0,1)$, $(.5,1)$ and $(.5,.5)$ in the projective chart $h^z=1$. According

1252: to theorem~\ref{thm:labels}, the closure of the set of labels is equal to the complement

1253: of the interiors of the islands; the striking closeness of the two pictures is one of the

1254: best indirect tests of the correctness of the implementation of the {\textbf N} algorithm in

1255: our C++ library NTC.

1256: }

1257: \label{fig:sm1000labels}

1258: \label{fig:labels}

1259: \end{figure}

1260: %

1261: According to theorem~\ref{thm:labels}, the closure of the set of labels of $\cS(\cP_0)$, 

1262: seen as points of $\RPt$, is equal to the complement of the interior of the islands.

1263: At this resolution about $5\cdot10^5$ different labels are found and their image

1264: in $\RPt$ (Fig.~\ref{fig:labels}) is one of the best indirect checks of correctness

1265: of the library NTC.

1266: %

1267: \section{Conclusions}

1268: %

1269: We proved in this paper that the structure of foliations induced on triply periodic

1270: embedded polyhedra by constant 1-forms is identical to the one induced on smooth

1271: surfaces, and we believe that the same line of arguments can also be used to further

1272: generalize the theorems to embedded piecewise smooth surfaces.

1273: This fact extends to ``Morse PL functions'' the association of a SM; in the most

1274: interesting cases such SM have a fractal nature (this condition is true for an open subset

1275: of all triply periodic PL functions, e.g for all triply periodic functions close enough to

1276: $\mh$).

1277: \par

1278: Surfaces satisfying property SP1 (see section~\ref{sec:sp}) are particularly rich, their

1279: SM being equal to the SM of any function having them as level set. We exploited this fact

1280: by studying the case of the triply periodic surface $\cP_0$ obtained extending in the

1281: three coordinate directions a truncated octahedron. The simplicity of the triangulation of

1282: the surface allowed us to improve by an order of magnitude, respect to the results

1283: obtained in~\cite{DL03b}, the resolution of the numerical analysis of the fractal

1284: and, as a consequence, to perform several numerical tests on conjectures and theorems.

1285: %

1286: \section{Acknowledgments}

1287: %

1288: We want to thank first of all S.P. Novikov for introducing the subject. We are also deeply

1289: in debt with S.P. Novikov and I.A. Dynnikov for their interest in our work and for several

1290: precious and fruitful scientific suggestions and discussions that were essential for the

1291: present work. In particular, we thank I.A. Dynnikov for suggesting the analytical algorithm

1292: to retrieve the islands boundaries.

1293: All numerical calculations were made with computers kindly provided by the INFN section

1294: of Cagliari (www.ca.infn.it) and CRS4 (www.crs4.it). We also acknowledge financial support 

1295: from the Cagliari section of INFN and from the Departments of Physics and of Mathematics

1296: of the University of Cagliari.

1297: %

1298: \bibliography{pol}

1299: %

1300: \vfill\eject

1301: %

1302: \begin{figure}

1303: \includegraphics[width=11cm,bb=200 410 400 722]{polCover.ps}

1304: %\epsfxsize=11cm\epsfbox[200 410 400 722]{polCover.ps}

1305: \caption{%

1306: The next 14 pictures show, in full detail, the fractal structure of $\cS(\cP_0)=\cS(\mh)$

1307: sampled at the resolution $r=10^4$. In order to minimize the CPU time, the numerical

1308: analysis was limited to the triangle $[(0,1,1)]$, $[(1/2,1,1)]$ $[(1/2,1/2,1)]$, divided

1309: above in smaller squares and triangles by a uniform grid and labeled by the corresponding

1310: figure number; the full picture can be retrieved by applying the ``topological'' symmetry

1311: characteristic of this surface, namely the symmetry with respect to the antidiagonal, and

1312: then the 24 symmetries coming from the action of the tetrahedral group $T_d$. To retrieve these 

1313: data we used about 20 1GHz Pentium III CPUs for about 3 months.In all

1314: pictures, the color of the islands goes from green to red as the norm of the label

1315: grows. Note that the square $[.4,.5]\times[.9,1]$ is fully contained inside the island

1316: $\cD_{(1,2,2)}$ and therefore its picture will not be shown.

1317: }

1318: \label{fig:sm10000}

1319: \end{figure}

1320: %

1321: \vfill\eject

1322: \fig{polSM10000.x0-1000.y9000-10000.eps}{Detail of $\cS(\cP_0)$ in $[0,.1]\times[.9,1]$}

1323: \vfill\eject

1324: \fig{polSM10000.x1000-2000.y9000-10000.eps}{Detail of $\cS(\cP_0)$ in $[.1,.2]\times[.9,1]$}

1325: \vfill\eject

1326: \fig{polSM10000.x1000-2000.y8000-9000.eps}{Detail of $\cS(\cP_0)$ in $[.1,.2]\times[.8,.9]$}

1327: \vfill\eject

1328: \fig{polSM10000.x2000-3000.y9000-10000.eps}{Detail of $\cS(\cP_0)$ in $[.2,.3]\times[.9,1]$}

1329: \vfill\eject

1330: \fig{polSM10000.x2000-3000.y8000-9000.eps}{Detail of $\cS(\cP_0)$ in $[.2,.3]\times[.8,.9]$}

1331: \vfill\eject

1332: \fig{polSM10000.x2000-3000.y7000-8000.eps}{Detail of $\cS(\cP_0)$ in $[.2,.3]\times[.7,.8]$}

1333: \vfill\eject

1334: \fig{polSM10000.x3000-4000.y8000-9000.eps}{Detail of $\cS(\cP_0)$ in $[.3,.4]\times[.8,.9]$}

1335: \vfill\eject

1336: \fig{polSM10000.x3000-4000.y7000-8000.eps}{Detail of $\cS(\cP_0)$ in $[.3,.4]\times[.7,.8]$}

1337: \vfill\eject

1338: \fig{polSM10000.x3000-4000.y6000-7000.eps}{Detail of $\cS(\cP_0)$ in $[.3,.4]\times[.6,.7]$}

1339: \vfill\eject

1340: \fig{polSM10000.x4000-5000.y9000-10000.eps}{Detail of $\cS(\cP_0)$ in $[.4,.5]\times[.9,1]$}

1341: \vfill\eject

1342: \fig{polSM10000.x4000-5000.y8000-9000.eps}{Detail of $\cS(\cP_0)$ in $[.4,.5]\times[.8,.9]$}

1343: \vfill\eject

1344: \fig{polSM10000.x4000-5000.y7000-8000.eps}{Detail of $\cS(\cP_0)$ in $[.4,.5]\times[.7,.8]$}

1345: \vfill\eject

1346: \fig{polSM10000.x4000-5000.y6000-7000.eps}{Detail of $\cS(\cP_0)$ in $[.4,.5]\times[.6,.7]$}

1347: \vfill\eject

1348: \fig{polSM10000.x4000-5000.y5000-6000.eps}{Detail of $\cS(\cP_0)$ in $[.4,.5]\times[.5,.6]$}

1349: %

1350: \end{document} 

1351: