1:
2: \documentclass{amsart}
3: \usepackage{amssymb}
4: %\usepackage{stmaryrd}
5: \usepackage{color}
6:
7: \linespread{1}
8:
9: \usepackage{epsfig}
10: \usepackage{graphicx}
11: \usepackage{psfrag}
12:
13: %\usepackage[latin1]{inputenc}
14: %\usepackage[english]{babel}
15: %\usepackage{babel}
16: %\usepackage[T1]{fontenc}
17:
18: \usepackage{amsfonts}
19: \usepackage{amsmath}
20: \usepackage{url}
21:
22: %\usepackage{graphicx}
23: %\usepackage{epsfig}
24: %\psfigdriver{dvips}
25: %\usepackage{rotating}
26: %\rotdriver{dvips}
27:
28: \newtheorem{theorem}{Theorem}
29: \newtheorem{sublemma}[theorem]{Sublemma}
30: \newtheorem{proposition}[theorem]{Proposition}
31: \newtheorem{property}[theorem]{Property}
32: \newtheorem{corollary}[theorem]{Corollary}
33: %\newtheorem{slogan}[theorem]{Slogan}
34: \newtheorem{lemma}[theorem]{Lemma}
35:
36: \theoremstyle{remark}
37: \newtheorem{remark}[theorem]{Remark}
38: \newtheorem{observation}[theorem]{Observation}
39: \newtheorem{trivial}[theorem]{Easy Fact}
40: \newtheorem{definition}[theorem]{Definition}
41:
42: \newcommand{\matris}[4]{\left ( \!\! \begin{array}{cc}
43: #1 & #2 \\ #3 & #4 \end{array} \!\! \right )}
44: \newcommand{\1}{1\hspace{-2.5pt}{\rm l}}
45: \newcommand{\binomial}[2]{\left
46: (\!\!\begin{array}{c}#1\\#2\end{array}\!\!\right)}
47: \newcommand{\llbracket}{[\![}
48: \newcommand{\rrbracket}{]\!]}
49:
50: \newcommand{\llabel}[1]{\label{#1}
51: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
52: %%% D I S P L A Y L A B E L S %%%
53: %%% %%%
54: %\boxed{\sf #1} %%%
55: %%% %%%
56: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
57: }
58:
59: \renewcommand{\labelenumi}{{\sf \roman{enumi} --}}
60:
61: %\newcommand{\color}[6]{}
62: % (A mettre s'il n'y a pas le package {color}).
63: %
64: % Explication de la ligne precedente : les fichiers .pstex_t (inseres ici par
65: % \include et qui contiennent les legendes $x$ des figures) contiennent des
66: % lignes de la forme : \color[rgb]{0,0,0}$x$ . Le compilateur ne reconnait
67: % pas \color. On definit donc une commande \color qui prend 6 arguments et
68: % ne fait rien : les 5 premiers correspondent aux 5 caracteres [rgb] et le
69: % dernier au bloc {0,0,0}...
70:
71: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
72: %% D I S A B L E P I C T U R E S %%
73: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
74: %
75: \newcommand{\ledessin}[1]{\includegraphics{#1.eps}}
76: %\newcommand{\ledessin}[1]{\includegraphics[scale=1]{#1.gif}}
77: %\newcommand{\ledessin}[1]{\texttt{(Dessin : #1...)}}
78: %
79: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
80:
81: \setcounter{section}{-1}
82:
83: \begin{document}
84:
85: \title[Punctured-torus groups]
86: {Triangulated cores of punctured-torus groups}
87: \author[F. Gu\'eritaud]{Fran\c{c}ois Gu\'eritaud}
88: %\date{February 2005}
89: \date{May 2006}
90:
91: \begin{abstract}
92: We show that the interior of the convex core of a quasifuchsian punctured-torus group admits an ideal decomposition (usually an infinite triangulation) which is canonical in two different senses: in a combinatorial sense \emph{via} the pleating invariants, and in a geometric sense \emph{via} an Epstein-Penner convex hull construction in Minkowski space. The result extends to certain non-quasifuchsian punctured-torus groups, and in fact to all of them if a strong version of the Pleating Lamination Conjecture is true.
93: \end{abstract}
94:
95: \maketitle
96: %\tableofcontents
97:
98: \section{Introduction}
99:
100: Among Kleinian groups with infinite covolume, quasifuchsian groups, which are deformations of Fuchsian surface groups, are fundamental examples. It has been noted \cite{jorgensen} that such groups can be analyzed much more explicitly when the base surface is just a once-punctured torus --- the curve complex is then dual to a (locally finite) tree, which simplifies many issues. Punctured-torus groups, which still retain many features of general quasifuchsian groups, have thus become a favorite ``training ground'': Minsky's work \cite{minsky} on end invariants is an example. Similarly, Caroline Series \cite{series} was able to prove the Pleating Lamination Conjecture for quasifuchsian punctured-torus groups (general pleating laminations, in contrast, seem to pose tremendous technical challenges). In this paper, we prove that the convex core $V$ of any quasifuchsian punctured-torus group admits an ideal triangulation (or slightly more general decomposition) which relates ``as nicely as one could hope'' both to the intrinsic geometry of $V$, and to the combinatorics of the boundary pleatings of $V$. This answers several conjectures made in \cite{aswy1}. As a byproduct, we get enough control to re-prove the result of \cite{series}.
101:
102: \subsection{Objects of study} \llabel{definelambda}
103: Let $S$ be the once-punctured torus endowed with its differential
104: structure. Let $\mathcal{C}$ be the set of homotopy classes of
105: simple closed curves in $S$. Choose in $\mathcal{C}$ a meridian $m$
106: and longitude $l$ whose intersection number is $1$. Fix an
107: identification $s$, called the \emph{slope}, between $\mathcal{C}$
108: and $\mathbb{P}^1{\mathbb Q}$, so that $s(m)=\infty$ and $s(l)=0$.
109: Let $(\alpha^+,\beta^+)$ and $(\alpha^-,\beta^-)$ be elements of
110: ${\mathbb{R}^*}^2$ such that $\beta^+/\alpha^+$ and
111: $\beta^-/\alpha^-$ are distinct irrationals. Define the \emph{pleatings}
112: $\lambda^{\pm}:\mathcal{C} \rightarrow {\mathbb R}^{\pm}$ by
113: $$\lambda^+(\eta/\xi)=\left|\left|\begin{array}{cc} \beta^+& \eta \\
114: \alpha^+&\xi \end{array}\right|\right| ~~\text{ and }~~
115: \lambda^-(\eta/ \xi)=-\left|\left|\begin{array}{cc} \beta^-& \eta \\
116: \alpha^-& \xi \end{array}\right|\right|,$$ where the double bars denote
117: the absolute value of the determinant (and $\xi$, $\eta$ are coprime integers). Replacing a pair $(\alpha^{\pm}, \beta^{\pm})$ by its
118: negative does not change $\lambda^{\pm}$. Notice that
119: $\lambda^-<0<\lambda^+$ (as functions on $\mathcal{C}$).
120:
121: \begin{theorem} \llabel{pureexistence}
122: There exists a cusped, non-complete hyperbolic
123: $3$-manifold $V$ homeomorphic to $S\times \mathbb{R}$ whose metric
124: completion $\overline{V}$ is homeomorphic to $S\times [0,1]=V\sqcup
125: S_{-\infty} \sqcup S_{+\infty}$, where $S_{-\infty}, S_{+\infty}$ are pleated surfaces whose pleating measures restrict to $\lambda^-$ and $\lambda^+$ (respectively) on $\mathcal{C}$. \end{theorem}
126: The bulk of this paper is devoted to producing such a $V$ (with some adjustments, the method also applies to rational pleatings $\lambda^{\pm}$). The puncture of $S$ is required to correspond to a cusp of $V$, so $\overline{V}$ will be the convex core of a quasifuchsian punctured-torus group $\Gamma \subset \text{Isom}^+(\mathbb{H}^3)$, i.e. the convex core of the manifold $\mathbb{H}^3/\Gamma$ (a punctured-torus group is a group freely generated by two elements with parabolic commutator). The interior $V$ of $\overline{V}$ is called the \emph{open convex core}.
127:
128: Specifically, $V$ is constructed as an infinite union of
129: ideal hyperbolic tetrahedra $(\Delta_i)_{i\in \mathbb{Z}}$ glued
130: along their faces, and the $\lambda^{\pm}$ encode the gluing rule between $\Delta_i$ and $\Delta_{i+1}$ (see Section \ref{sectionstrategy}): so this ideal decomposition $\mathcal{D}$ of $V$ is canonical in a combinatorial
131: sense, with respect to the data $\lambda^{\pm}$. By \cite{series}, the group $\Gamma$ is determined uniquely up to conjugacy in $\text{Isom}(\mathbb{H}^3)$ by the $\lambda^{\pm}$. Our construction therefore provides a decomposition $\mathcal{D}$ of the open convex core $V$ of an \emph{arbitrary} quasifuchsian, non-Fuchsian punctured-torus group $\Gamma$. A good drawing of $V$ is Figure 3 of \cite{thurstonlimitedouble}.
132:
133: \subsection{Geometric canonicity}
134: Akiyoshi and Sakuma \cite{comparing} generalized the Epstein-Penner convex
135: hull construction in Minkowski space $\mathbb{R}^{3+1}$ (see \cite{epsteinpenner}) to show that $V$ also admits a decomposition $\mathcal{D}^G$, canonical in a purely geometric sense, and related to the
136: Ford-Voronoi domain of $\mathbb{H}^3/\Gamma$. Roughly speaking, $\mathcal{D}^G$ is defined by considering the $\Gamma$-orbit $\mathcal{O}\subset \mathbb{R}^{3+1}$ of an isotropic vector representing the cusp, and projecting the cell decomposition of the boundary of the convex hull of $\mathcal{O}$ back to $\mathbb{H}^3/\Gamma$ (see Section \ref{sectioneph} for more detail). Our main theorem is
137:
138: \begin{theorem} \llabel{ephgeneral}
139: If $V$ is the open convex core of a quasifuchsian once-punctured torus group,
140: the decompositions $\mathcal{D}$ and $\mathcal{D}^G$ of $V$ are the same.
141: \end{theorem}
142:
143: %Excision ("strong pleating lamination conjecture")
144:
145: It is reasonable to understand ending laminations of geometrically infinite surface groups as \emph{infinitely strong pleatings}, and to conjecture that the group is determined by its ending and/or pleating laminations.
146: Indeed, our method allows to construct (conjecturally unique) punctured-torus groups with arbitrary admissible ending and/or pleating laminations: the precise statement, with a full description of $\mathcal{D}^G$ (especially in the case of rational laminations), is Theorem \ref{ephhyper} in Section \ref{sectionextensions}.
147:
148: \subsection{Context} The identity $\mathcal{D}=\mathcal{D}^G$ of Theorem \ref{ephgeneral} (and the existence of $\mathcal{D}$, as realized by positively oriented cells) was Conjecture 8.2 in \cite{aswy1}, also called the \emph{Elliptic-Parabolic-Hyperbolic (EPH) Conjecture}.
149: Akiyoshi subsequently established the identity in the case of two
150: infinite ends, in \cite{akiyoshi}. Near a finite end however, the ideal tetrahedra $\{\Delta_i\}_{i\in \mathbb{Z}}$ of $\mathcal{D}$ flatten at a very quick rate: the smallest angle of $\Delta_i$ typically goes to $0$ faster than any geometric sequence, as $i$ goes to $\pm \infty$. In a sense, the difficulty is to show that these angles nevertheless stay positive for the hyperbolic metric.
151:
152: Under the hypotheses of Theorem \ref{ephgeneral}, for finitely many indices $i\in\mathbb{Z}$, the tetrahedron $\Delta_i$ of $\mathcal{D}^G$ comes from a \emph{spacelike} face in Minkowski space, and is therefore dual to a singular point (a vertex) of the Ford-Voronoi domain of $\mathbb{H}^3/\Gamma$. The latter domain is described in great detail in \cite{jorgensen, aswy1, aswy2} and \cite{aswy3}, relying on a \emph{geometric continuity} argument in the space of quasifuchsian groups (which is known to be connected). To study \emph{all} tetrahedra $\Delta_i$ at once, including those not seen in the Ford-Voronoi domain, the present paper takes a somewhat opposite approach: first describe geometric shapes for the tetrahedra of the candidate triangulation $\mathcal{D}$, then establish that the gluing of these tetrahedra defines (the open convex core of) a quasifuchsian group.
153:
154: The construction of $\mathcal{D}$ (namely, of angles for the tetrahedra) will be fairly explicit: the solution will arise as the maximum of an explicit concave ``volume'' functional $\mathcal{V}$ over an explicit convex domain. The domain has infinite dimension, but there are explicit bounds on the contributions to $\mathcal{V}$ of the ``tail'' coordinates. This should allow for numerically efficient implementations.
155:
156: The geometrically canonical decomposition $\mathcal{D}^G$ can be defined for arbitrary cusped manifolds, but is quite mysterious and hard to study in general. It seems to be unknown, for instance, whether $\mathcal{D}^G$ is always locally finite (see \cite{comparing}). For quasifuchsian punctured-torus groups however, Theorem \ref{ephgeneral} can be said to completely describe the combinatorics of $\mathcal{D}^G$. In fact, aside from coarsenings of $\mathcal{D}$, the author does not know of any ideal cell decomposition that is invariant under the hyperelliptic involution (a property which $\mathcal{D}^G$ must a priori enjoy).
157:
158: An identity of the form $\mathcal{D}=\mathcal{D}^G$ can be established by the same methods in several closely related contexts: punctured-torus bundles (where the result is due to Lackenby \cite{lackenby}); complements of two-bridge links (see the announcement in \cite{aswy3}) or of certain arborescent links --- see \cite{these} for a synthesis, but the present paper contains the key ideas. We will use several results (and notation) from \cite{mapomme}.
159:
160: \subsection*{Acknowledgements}
161: My thanks go to Makoto Sakuma for having drawn my attention to the EPH conjecture. I am also deeply indebted to Francis Bonahon, Fr\'ed\'eric Paulin, Makoto Sakuma and David Futer for many discussions and insights.
162:
163: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
164:
165: \section{Strategy} \llabel{sectionstrategy}
166: \subsection{Setup} \llabel{subsectionsetup}
167: We return to the irrational pleating data $\lambda^{\pm}$.
168: In the hyperbolic plane ${\mathbb H}^2$ with boundary $\partial
169: {\mathbb H}^2={\mathbb P}^1{\mathbb R}$, consider the Farey
170: triangulation (the ideal triangle $01\infty$ iteratedly reflected in
171: its sides, see e.g. Section 2 of \cite{minsky}). The irrationals $\beta^+/\alpha^+$ and $\beta^-/\alpha^-$ belong to $\partial \mathbb{H}^2$, and the oriented line $\Lambda$ from $\beta^-/\alpha^-$ to $\beta^+/\alpha^+$ crosses infinitely many Farey edges $(e_i)_{i\in{\mathbb Z}}$ (the choice of $e_0$ is
172: arbitrary). To every pair of consecutive integers $i,i+1$ is
173: associated a letter, $R$ or $L$, according to whether $e_i$ and
174: $e_{i+1}$ share their Right or Left end, with respect to the
175: orientation of $\Lambda$ (we say that $\Lambda$ \emph{makes a Right} or
176: \emph{makes a Left} across the Farey triangle between $e_i$ and $e_{i+1}$).
177: We thus get a bi-infinite word $...RLLLRR...$ with infinitely many
178: $R$'s (resp. $L$'s) near either end.
179:
180: Two rationals of $\mathbb{P}^1\mathbb{Q}$ are Farey neighbors exactly when
181: the corresponding elements (simple closed curves) of $\mathcal{C}$ have number of intersection $1$. Therefore, each Farey triangle $\tau$ defines an
182: \emph{ideal triangulation}
183: %(see \cite{epstein} for a general definition)
184: of the punctured torus $S$ in the following way. The
185: vertices of $\tau$ are the slopes of three curves of $\mathcal{C}$,
186: each parallel to a properly embedded line running from the puncture
187: to itself (so we may speak of the \emph{slopes} of such properly
188: embededd lines). These three lines are embedded disjointly and
189: separate $S$ into two \emph{ideal triangles}. Moreover, two
190: triangulations corresponding to Farey triangles which share an edge
191: differ by a \emph{diagonal move} (see Figure \ref{diagonalmove}).
192: Such a diagonal move must be seen as a (topological) ideal
193: tetrahedron in $S\times \mathbb{R}$ filling the space between two
194: (topological) pleated surfaces, pleated along the two ideal
195: triangulations.
196: \begin{figure}[h!] \centering
197: \ledessin{diagonalmove}
198: \caption{\llabel{diagonalmove} Left: a diagonal move in $S$ (the puncture is in
199: the corners). Right: how to see it as an ideal tetrahedron, here with
200: truncated vertices (grey).} \end{figure}
201:
202: Our strategy will be to consider triangulated surfaces (pleated
203: punctured tori) corresponding to the Farey triangles living between
204: $e_i$ and $e_{i+1}$ for $i$ ranging over ${\mathbb Z}$, interpolate
205: these surfaces with tetrahedra $\Delta_i$ corresponding to the
206: diagonal moves, and provide geometric parameters (dihedral angles)
207: for all these objects, using Rivin's Maximum Volume Principle (see
208: \cite{rivin}).
209:
210: Throughout the paper, we will deal with an infinite family of
211: tetrahedra $(\Delta_i)_{i\in \mathbb{Z}}$, separated by pleated
212: once-punctured tori $S_i$. By an arbitrary choice, we resolve that
213: $S_i$ is the surface between $\Delta_i$ and $\Delta_{i-1}$ --- or
214: equivalently, that $\Delta_i$ is bounded by the surfaces $S_i$ and
215: $S_{i+1}$. However, the numbering of the tetrahedra should be seen
216: as the more essential one (see especially Definition
217: \ref{definitionhinge} below).
218:
219: \subsection{Plan of the paper}
220: In Section \ref{sectionangles}, we describe the space of possible
221: dihedral angle assignments $x_i, y_i, z_i$ for the $\Delta_i$. In
222: Section \ref{sectionboundingthebending}, we encode $\lambda^{\pm}$
223: into constraints on the $x_i, y_i, z_i$. In Section
224: \ref{sectionhyperbolicvolume}, we carry out (constrained) volume
225: maximization. Important asymptotic features of the solution are analyzed in
226: Section \ref{sectionbehaviorofwi}. In Section
227: \ref{sectionthecusplink}, we describe the Euclidean triangulation of
228: the cusp. In Sections \ref{sectionintrinsicconvergence} and
229: \ref{sectionextrinsicconvergence}, we show that the pleated surfaces
230: $S_i$ converge in a strong enough sense, so that their limit as $i$
231: goes to $\pm \infty$ describes the (pleated) boundary of the metric
232: completion of $V=\bigcup_{i\in \mathbb{Z}} \Delta_i$. At that point,
233: we have constructed (the convex core of) a quasifuchsian group. The corresponding instance of Theorem \ref{ephgeneral} then follows from a computation, carried out in Section \ref{sectioneph}. In Section \ref{sectionextensions}, we provide a similar construction of punctured-torus groups with \emph{rational} pleating slopes $\beta^{\pm}/\alpha^{\pm}$ and/or with infinite ends, and re-prove that $(\lambda^+, \lambda^-)$ are continuous coordinates for the space of quasifuchsian groups (see \cite{series}).
234:
235: \section{Dihedral angles} \llabel{sectionangles}
236:
237: In this section we find positive dihedral angles for the ideal tetrahedra
238: $\Delta_i$, following Section 5 of \cite{mapomme}. More precisely, we describe the convex space $\Sigma$ of positive
239: angle configurations for the $\Delta_i$ such that:
240: \begin{itemize}
241: \item the three dihedral angles near each ideal vertex of $\Delta_i$ add up to $\pi$
242: (this is true in any ideal tetrahedron of ${\mathbb H}^3$);
243: \item the dihedral angles around any edge of $V=\bigcup_{i\in \mathbb{Z}}\Delta_i$
244: add up to $2\pi$ (this is necessary, though not sufficient, for a
245: hyperbolic structure at the edge);
246: \item the three pleating angles of each pleated punctured torus $S_i$ add up to $0$
247: (this is necessary, though not sufficient, to make the puncture $p$
248: of $S$ a cusp of $V$, i.e. make the loop around $p$ lift to a
249: parabolic isometry of $\mathbb{H}^3$).
250: \end{itemize}
251: (The first condition implies that opposite edges in $\Delta_i$ have
252: the same dihedral angle.) Later on we shall apply Rivin's Maximum Volume
253: Principle on a certain convex subset of (the closure of) $\Sigma$.
254:
255: If the tetrahedron $\Delta_i$ realizes a diagonal exchange that
256: kills an edge $\varepsilon'$ and replaces it with $\varepsilon$,
257: denote by $\pi-w_i$ the interior dihedral angle of $\Delta_i$ at
258: $\varepsilon$ and $\varepsilon'$. Observe that the slope of
259: $\varepsilon$ (resp. $\varepsilon'$) is the rational located
260: opposite the Farey edge $e_i$ in the Farey diagram, on the side of
261: $\beta^+/\alpha^+$ (resp $\beta^-/\alpha^-$).
262:
263: Thus, the pleating angles of the surface $S_i$ living between
264: $\Delta_{i-1}$ and $\Delta_i$ are \begin{equation}
265: \llabel{pleatingangles} w_{i-1}~~,~~-w_i~~\text{ and }~~
266: w_i-w_{i-1}~.\end{equation} Observe the sign convention: the pleated
267: punctured torus embedded in $S\times\mathbb{R}$ receives an
268: \emph{upward} transverse orientation from $\mathbb{R}$, and the
269: angles we consider are the dihedral angles \emph{above} the surface,
270: minus $\pi$. Thus, the ``new'' edge of $\Delta_{i-1}$, pointing
271: upward, accounts for a positive pleating $w_{i-1}$, while the
272: ``old'' edge of $\Delta_i$, pointing downward, accounts for a
273: negative pleating $-w_i$. This is in accordance with the convention
274: $\lambda^-<0<\lambda^+$ of the Section \ref{definelambda}. One may write the three numbers (\ref{pleatingangles}) in the corners of the corresponding
275: Farey triangles (Figure \ref{rl2}, top).
276:
277: In the tetrahedron $\Delta_i$, let $x_i$ (resp. $y_i$) be the
278: interior dihedral angle at the edge whose slope is given by the
279: right (resp. left) end of the Farey edge $e_i$. Let $z_i=\pi-w_i$ be
280: the third dihedral angle of $\Delta_i$. For instance, $2x_i$ (resp.
281: $2y_i$) is the difference between the numbers written just below and
282: just above the right (resp. left) end of $e_i$ in Figure \ref{rl2}
283: (the factor $2$ comes from the fact that the two edges of $\Delta_i$
284: with angle $x_i$ [resp. $y_i$] are identified). For notational
285: convenience, write $(w_{i-1},w_i,w_{i+1})=(a,b,c)$. By computing
286: differences between the pleating angles given in Figure \ref{rl2}
287: (bottom), we find the following formulae for $x_i, y_i, z_i$
288: (depending on the letters, $R$ or $L$, living just before and just
289: after the index $i$):
290:
291: \begin{figure}[h!] \centering
292: \psfrag{w}{$-w_i$}
293: \psfrag{w1}{$w_{i-1}$}
294: \psfrag{ww}{$w_i-w_{i-1}$}
295: \psfrag{e}{$e_i$}
296: \psfrag{e1}{$e_{i-1}$}
297: \psfrag{L}{$L$}
298: \psfrag{R}{$R$}
299: \psfrag{a}{$a$}
300: \psfrag{bb}{$-b$}
301: \psfrag{b}{$b$}
302: \psfrag{cc}{$-c$}
303: \psfrag{cb}{$c-b$}
304: \psfrag{ba}{$b-a$}
305: \psfrag{p}{$p$}
306: \psfrag{p1}{$p'$}
307: \psfrag{p2}{$p''$}
308: \ledessin{LLRRLRRL}
309: \caption{Bottom: $e_i$ is the horizontal edge and
310: $(a,b,c)=(w_{i-1},w_i,w_{i+1})$.\llabel{rl2}}
311: \end{figure}
312:
313: \begin{equation} \llabel{xiyi}
314: \begin{array}{c|c|c|c|c}
315: & L~~~~~L & R~~~~~R & L~~~~~R & R~~~~~L\\ \hline
316: x_i&\frac12 (a+c)&\frac12 (-a+2b-c)&\frac12 (a+b-c)&\frac12 (-a+b+c)\\
317: y_i&\frac12 (-a+2b-c)&\frac12 (a+c)&\frac12 (-a+b+c)&\frac12 (a+b-c)\\
318: z_i&\pi-b&\pi-b&\pi-b&\pi-b\\
319: \end{array}\end{equation}
320:
321: %\begin{equation} \begin{array}{c|c|c|c|c}
322: %& L~~~~~L & R~~~~~R & L~~~~~R & R~~~~~L\\ \hline
323: %x_i&\frac{a+c}{2}&\frac{-a+2b-c}{2}&\frac{a+b-c}{2}&\frac{-a+b+c}{2}\\
324: %y_i&\frac{-a+2b-c}{2}&\frac{a+c}{2}&\frac{-a+b+c}{2}&\frac{a+b-c}{2}\\
325: %z_i&\pi-b&\pi-b&\pi-b&\pi-b\\ \end{array}\end{equation}
326:
327:
328: The first of the three conditions defining $\Sigma$ can be checked immediately;
329: the other two are true by construction. From (\ref{xiyi}), the condition for
330: all angles to be positive is that:
331:
332: \begin{equation} \llabel{positivity} \left \{ \begin{array}{l}
333: \text{ For all $i$ one has $0<w_i<\pi$.} \\
334: \text{ If $i$ separates identical letters (first two cases),
335: $2w_i>w_{i+1}+w_{i-1}$.} \\
336: \text{ If $i$ separates different letters (last two cases),
337: $|w_{i+1}-w_{i-1}|<w_i$.} \end{array} \right . \end{equation}
338: Denote by $\Sigma$ the non-empty, convex solution set of (\ref{positivity}).
339:
340: \section{Bounding the Bending} \llabel{sectionboundingthebending}
341:
342: \subsection{A natural constraint on the pleating of $S_i$} \llabel{subsectionnaturalconstraint}
343: Next, we describe a certain convex subset of $\Sigma$. It is
344: obtained in the following way. Consider the pleated punctured torus,
345: $S_i$, lying between the tetrahedra $\Delta_i$ and $\Delta_{i-1}$.
346: Let $\epsilon_1, \epsilon_2,\epsilon_3$ be the edges of $S_i$ and
347: $\delta_1,\delta_2,\delta_3$ the corresponding pleating angles
348: (exterior dihedral angles, counted positively for salient edges as
349: in (\ref{pleatingangles}) above). Then, define the pleating measure
350: $\lambda_i:\mathcal{C}\rightarrow \mathbb{R}^+$ of $S_i$ by
351: \begin{equation} \lambda_i(\gamma)=\nu_1^{\gamma}\delta_1+\nu_2^{\gamma}\delta_2 +\nu_3^{\gamma}\delta_3, \llabel{definelambdai} \end{equation} where
352: $\nu_s^{\gamma}\in\mathbb{N}$ is the intersection number of $\gamma$ with the
353: simple closed curve $\varepsilon_s$ parallel to $\epsilon_s$. We
354: shall require that $$\lambda^-(\varepsilon_s)<
355: \lambda_i(\varepsilon_s)< \lambda^+(\varepsilon_s)$$ for each
356: $s\in\{1,2,3\}$, and the same thing for every pleated punctured
357: torus $S_i$ in the $\mathbb{Z}$-family (\footnote{In fact a similar inequality will follow for all simple closed curves $\gamma$ of $\mathcal{C}$ (see Lemma \ref{bendingbounded} below), namely, $\lambda^-<\lambda_i<\lambda^+$ as functions on $\mathcal{C}$. Forcing this ``natural'' inequality
358: is the whole point of our constraint.}).
359:
360: In other words, denote by $\eta/\xi$ the (rational) slope of
361: $\varepsilon_s$. Observe that
362: $\lambda_i(\varepsilon_s)=\delta_{s'}+\delta_{s''}=-\delta_s$ where
363: $\{s,s',s''\}=\{1,2,3\}$, because the slopes of $\varepsilon_1,
364: \varepsilon_2, \varepsilon_3$ are Farey neighbors. Therefore the
365: requirement is that, for every $\varepsilon_s$ as above,
366: \begin{equation} \llabel{encadrement} -\left|\left|
367: \begin{array}{cc} \beta^+&\eta\\
368: \alpha^+&\xi\end{array} \right|\right|< \delta_s
369: <\left|\left|\begin{array}{cc} \beta^-&\eta\\
370: \alpha^-&\xi\end{array} \right|\right|.\end{equation}
371:
372: To express (\ref{encadrement}) in terms of the $w_i$, we need some
373: notation (the $\delta_s$ are determined by the $w_i$ via (\ref{pleatingangles}) above). For each Farey edge $e_i$, let $q_i^+$ (resp. $q_i^-$) be the rational
374: located opposite $e_i$, on the same side of $e_i$ as
375: $\beta^+/\alpha^+$ (resp. $\beta^-/\alpha^-$). For an arbitrary
376: rational $p=\eta/\xi$ (reduced form), introduce the slightly
377: abusive (\footnote{Abusive in that it depends on the ordered pair
378: $(\alpha,\beta)$ rather than just on the real
379: $\frac{\beta}{\alpha}$.}) notation \begin{equation} \llabel{wedge}
380: \frac{\beta}{\alpha} \wedge p := \left | \left | \begin{array}{cc} \beta&\eta\\
381: \alpha &\xi \end{array} \right | \right | \text{ (absolute value of
382: the determinant)}. \end{equation} Then, if $l,r$ are the rationals
383: living at the left and right ends of the Farey edge $e_i$, one has
384: \begin{equation} \llabel{signes} \textstyle{(\frac{\beta^+}{\alpha^+}
385: \wedge l ) + (\frac{\beta^+}{\alpha^+} \wedge r) =
386: \frac{\beta^+}{\alpha^+} \wedge q_i^-~~;~~
387: (\frac{\beta^-}{\alpha^-}\wedge l ) + (\frac{\beta^-}{\alpha^-}
388: \wedge r) = \frac{\beta^-}{\alpha^-} \wedge q_i^+}~.\end{equation}
389: Indeed, the $||\cdot||$-notation is invariant under $PSL_2({\mathbb
390: Z})$, acting on $({\mathbb H}^2, \partial {\mathbb H}^2)$ by
391: isometries and on ordered pairs $\pm(\alpha,\beta)$ as a matrix
392: group. Since $PSL_2({\mathbb Z})$ acts transitively on oriented
393: Farey edges, we are reduced to the case
394: $(l,r,q_i^+,q_i^-)=(\infty,0,1,-1)$ where $\beta^-/\alpha^-<0<
395: \beta^+/\alpha^+$, which is straightforward.
396:
397: Let us now translate Equation (\ref{encadrement}) in terms of the
398: $w_i$. Let $e_{i-1}, e_i$ be two consecutive Farey edges; $p$ and
399: $p'$ are the ends of $e_{i-1}$; $p$ and $p''$ are the ends of $e_i$.
400: One has $q_{i-1}^+=p''$ and $q_i^-=p'$. In view of
401: (\ref{pleatingangles}) and Figure \ref{rl2} (top), Equation
402: (\ref{encadrement}) translates to $$\begin{array}{rcccl}
403: -(\frac{\beta^+}{\alpha^+} \wedge p'') &<&w_{i-1}&<&
404: \frac{\beta^-}{\alpha^-} \wedge p'' \\ -( \frac{\beta^+}{\alpha^+}
405: \wedge p' ) &<&-w_i&<& \frac{\beta^-}{\alpha^-} \wedge p' \\ -(
406: \frac{\beta^+}{\alpha^+} \wedge p ) &<&w_i-w_{i-1}&<&
407: \frac{\beta^-}{\alpha^-} \wedge p \\ \end{array}$$ Using the fact
408: that the $w_i$ are positive, this simplifies to (respectively)
409: $$\begin{array}{rcccl} &&w_{i-1}&<&\frac{\beta^-}{\alpha^-} \wedge q_{i-1}^+ \\
410: -(\frac{\beta^+}{\alpha^+} \wedge q_i^-) &<&-w_i&& \\
411: -(\frac{\beta^+}{\alpha^+} \wedge p) &<&w_i-w_{i-1}&<&
412: \frac{\beta^-}{\alpha^-} \wedge p \\ \end{array}$$
413:
414: Finally, observe that $\frac{\beta^+}{\alpha^+} \wedge
415: p=(\frac{\beta^+}{\alpha^+} \wedge q_{i-1}^-) -
416: (\frac{\beta^+}{\alpha^+} \wedge q_i^-)$ while
417: $\frac{\beta^-}{\alpha^-} \wedge p=(\frac{\beta^-}{\alpha^-} \wedge
418: q_i^+) - (\frac{\beta^-}{\alpha^-} \wedge q_{i-1}^+)$, by Equation
419: (\ref{signes}). Therefore, if we introduce
420:
421: \begin{equation}\llabel{phiphi} \phi_i^+:=\frac{\beta^+}{\alpha^+}
422: \wedge q_i^- ~~~\text{and }~~~
423: \phi_i^-:=\frac{\beta^-}{\alpha^-}\wedge q_i^+
424: \end{equation} then Equation (\ref{encadrement}) reduces to \begin{equation}
425: \left . \begin{array} {rcccl} w_i&<&\min\{\phi_i^+,\phi_i^-\} &&\\
426: \phi^-_{i-1}-\phi^-_i&<& w_{i-1}-w_i &<&\phi^+_{i-1}-\phi^+_i
427: \end{array} \right \}\forall i\in
428: \mathbb{Z}.\llabel{bounding}\end{equation}
429:
430: \subsection{Study of $\phi^+$ and $\phi^-$} \llabel{subsectionstudyphi}
431:
432: \begin{definition} \llabel{definitionhinge}
433: In \ref{subsectionsetup} we associated to each $i\in \mathbb{Z}$ a
434: Farey edge $e_i$ living between two letters of $\{R,L\}$. We call
435: $i$ a \emph{hinge index}, and $\Delta_i$ a \emph{hinge tetrahedron},
436: if the two letters are distinct ($RL$ or $LR$). \emph{Non-hinges}
437: correspond to $RR$ or $LL$.
438: \end{definition}
439:
440: \begin{lemma} \llabel{studyphi} The following holds concerning the sequences $\phi^+, \phi^-:{\mathbb Z} \rightarrow {\mathbb R}_+^*$.
441: \begin{enumerate}
442: \item First, $\phi^-$ is strictly increasing and $\phi^+$
443: is strictly decreasing.
444:
445: \item For all $i$ one has $1<\phi_{i+1}^-/\phi_i^-<2$ and
446: $1<\phi_{i-1}^+/\phi_i^+<2$.
447:
448: \item If $i\in \mathbb{Z}$ is non-hinge then
449: $\phi_{i-1}^++\phi_{i+1}^+=2\phi_i^+$ and
450: $\phi_{i-1}^-+\phi_{i+1}^-=2\phi_i^-$.
451:
452: \item If $i\in \mathbb{Z}$ is hinge then
453: $\phi_{i+1}^-=\phi_i^-+\phi_{i-1}^-$ and
454: $\phi_{i-1}^+=\phi_i^++\phi_{i+1}^+$.
455:
456: \item The sequences $\phi^{\pm}$ are (weakly) convex, i.e. $2\phi^{\pm}_i\leq
457: \phi^{\pm}_{i-1}+\phi^{\pm}_{i+1}$.
458:
459: \item If $i<j$ are consecutive hinge indices, then $1+\frac{j-i}{2} \leq
460: \{~\frac{\phi_i^+}{\phi_j^+}~,~ \frac{\phi_j^-}{\phi_i^-}~\} \leq
461: 1+j-i$.
462:
463: \item One has
464: $\underset{+\infty}{\lim}~\phi^+=\underset{-\infty}{\lim}~\phi^-=0$ and
465: $\underset{-\infty}{\lim}~\phi^+=\underset{+\infty}{\lim}~\phi^-=+\infty$.
466:
467: \item One has $\underset{-\infty}{\lim}~(\phi^+_{i-1}-\phi^+_i) =
468: \underset{+\infty}{\lim}~(\phi^-_{i+1}-\phi^-_i)=+\infty$.
469: \end{enumerate}
470: \end{lemma}
471:
472: \begin{remark} Points {\sf iii} and {\sf iv} are just equality cases of the
473: inequalities (\ref{positivity}), which encode positivity of the
474: angles $x_i, y_i, z_i$. \llabel{remarkphi}
475: \end{remark}
476:
477: \begin{proof} We will deal only with $\phi^+$: the arguments for $\phi^-$ are
478: analogous. Let $e_{i-1}=pr_{i-1}$ and $e_i=pr_i$ be consecutive
479: Farey edges. We have $\phi_i^+=\frac{\beta^+}{\alpha^+} \wedge
480: r_{i-1}$ while $\phi_{i-1}^+=\frac{\beta^+}{\alpha^+}\wedge r_{i-1}
481: + \frac{\beta^+}{\alpha^+}\wedge p$ by Equation (\ref{signes}),
482: hence {\sf i}.
483:
484: For {\sf ii}, we need care only about the upper bound. Just observe
485: that $\phi_{i-1}^+=(\frac{\beta^+}{\alpha^+}\wedge r_i +
486: \frac{\beta^+}{\alpha^+}\wedge p)+\frac{\beta^+}{\alpha^+}\wedge p$
487: while $\phi_i^+=\frac{\beta^+}{\alpha^+}\wedge r_i +
488: \frac{\beta^+}{\alpha^+}\wedge p$.
489:
490: For {\sf iii}, assume $e_{i+1}=pr_{i+1}$ so that
491: $\phi^+_k=\frac{\beta^+}{\alpha^+} \wedge p+\frac{\beta^+}{\alpha^+}
492: \wedge r_k$ for $|i-k|\leq 1$. For $k\in\{i,i+1\}$ the right hand
493: side is $\frac{\beta^+}{\alpha^+} \wedge r_{k-1}$ , so
494: $(\phi^+_{k-1}-\phi^+_k)=\frac{\beta^+}{\alpha^+} \wedge p$, which
495: is sufficient.
496:
497: For {\sf iv}, assume $e_{i+1}=p'r_i$. In the notations of Formula
498: (\ref{phiphi}), we have $q_{i+1}^-=p$ and $q_i^-=r_{i-1}$, the ends
499: of $e_{i-1}$. This together with Equation (\ref{signes}) yields the
500: result. Point {\sf v} follows from {\sf iv} and {\sf ii} at hinges
501: indices, and from {\sf iii} at other indices. For Point {\sf vi},
502: observe that $\phi_i^+=\phi_j^++(j-i)\phi_{j+1}^+$ by {\sf iii-iv},
503: and conclude using {\sf ii}. Point {\sf vii} follows from {\sf vi},
504: and the presence of infinitely many hinge indices near either end.
505: Point {\sf viii} follows from {\sf vii}, {\sf v}, and {\sf iv}.
506: \end{proof}
507:
508: \subsection{Behavior of the pleatings
509: $\lambda_i:\mathcal{C}\rightarrow\mathbb{R}$} \llabel{behavior}
510: For any real sequence $u\in \mathbb{R}^{\mathbb{Z}}$, define the
511: real sequence $\nabla u$ by $(\nabla u)_i=u_{i-1}-u_i$. Let us
512: summarize the conditions imposed on $w:{\mathbb Z}\rightarrow
513: {\mathbb R}_+^*$ (Eq. \ref{positivity} and \ref{bounding} above):
514: \begin{equation} \llabel{doubleve} \left \{ \begin{array}{rcccll}
515: 0&<&w_i&<&\min\{\phi_i^+,\phi_i^-,\pi\}& \\
516: \nabla \phi^-_i&<&\nabla w_i&<&\nabla \phi^+_i
517: &\text{for all $i\in \mathbb{Z}$;} \\
518: &&|w_{i+1}-w_{i-1}|&<&w_i&\text{if $i$ is a hinge;} \\
519: &&w_{i+1}+w_{i-1}&<&2w_i&\text{otherwise.}
520: \end{array} \right .\end{equation}
521: It is a simple exercise to check that $w_i= \tanh\phi_i^+
522: \tanh\phi_i^-$, for instance, satisfies this system ($\tanh$ may be
523: replaced by any strictly concave monotonous function from ${\mathbb
524: R}_+$ to $[0,1)$ with the same $1$-jet at $0$).
525: \begin{definition} If (\ref{doubleve}') denotes the system (\ref{doubleve}) in which all strong inequalities have been turned into weak ones, let $W\subset\mathbb{R^Z}$ be the solution space of (\ref{doubleve}'). \llabel{defineW}\end{definition}
526: Suppose $(w_i)_{i\in\mathbb{Z}}\in W$
527: and consider the corresponding pleating measures $\lambda_i:\mathcal{C}\rightarrow \mathbb{R}$ of $S_i$, the pleated punctured torus lying between the tetrahedra
528: $\Delta_i$ and $\Delta_{i-1}$.
529:
530: \begin{lemma} \llabel{bendingbounded}
531: For any curve $\gamma \in \mathcal{C}$, the sequence
532: $(\lambda_i(\gamma))_{i\in\mathbb{Z}}$ is nondecreasing, with
533: $$\lambda^-(\gamma)~\leq~\underset{i\rightarrow -\infty}{\lim}\lambda_i(\gamma)
534: ~\leq~\underset{i\rightarrow +\infty}{\lim}\lambda_i(\gamma)~\leq~\lambda^+(\gamma).$$
535: \end{lemma}
536: \begin{proof} First, observe that the pleating angles of $S_{i-1},S_i$ are
537: always of the form given in Figure \ref{monotonpleat}, where $x$
538: (resp. $y$) is positive, equal to the interior dihedral angle of the
539: tetrahedron at the horizontal (resp. vertical) edges of the square.
540: The closed curve $\gamma$ in the surface traverses the given
541: square a number of times, either vertically, or horizontally, or
542: diagonally (cutting off one of the four corners). The pleating along
543: $\gamma$ increases by $2y$ per vertical passage, $2x$ per horizontal
544: passage, and $0$ per diagonal passage, hence the monotonicity
545: statement. (This argument, or a variant of it, is also valid for
546: higher-genus surfaces and non-simple closed curves, as long as the
547: tetrahedron has positive angles.)
548: \begin{figure}[h!]
549: \centering
550: \psfrag{xmy}{$-x-y$}
551: \psfrag{xpy}{$x+y$}
552: \psfrag{A}{$A$}
553: \psfrag{B}{$B$}
554: \psfrag{A2x}{$A-2x$}
555: \psfrag{B2y}{$B-2y$}
556: \psfrag{ga}{$\gamma$}
557: \ledessin{monotonpleat}
558: \caption{The pleatings of
559: $S_{i-1}$ and $S_i$.\llabel{monotonpleat}}
560: \end{figure}
561:
562: For the bounding, we will focus only on the positive side.
563: Consider the slope $s=\frac{\eta}{\xi}\in
564: \mathbb{P}^1\mathbb{Q}$ of $\gamma$ (a reduced fraction); recall the definition
565: $\lambda^+(s)=||^{\beta^+}_{\alpha^+} ~^{\eta}_{\xi}||$ from the
566: Introduction. Consider a large enough index $i$, such that the Farey edge $e_{i-1}$ separates $s$ from $\beta^+/\alpha^+$. Consider the points $p,p',p'' \in \mathbb{P}^1\mathbb{Q}$ such that $(e_{i-1},e_i)=(pp',pp'')$: the points $(s,p,\frac{\beta^+}{\alpha^+},p'',p',s)$ are cyclically arranged in $\mathbb{P}^1\mathbb{R}$.
567: Observe that the $\wedge$-notation
568: (\ref{wedge}) applied at rationals is just the (geometric)
569: intersection number of the corresponding curves. Therefore, using
570: the angle information given in Figure \ref{rl2} (top), one has
571: \begin{equation}\llabel{determinant} \begin{array}{rcl} \lambda_i(\gamma)&=&(p\wedge s)(w_i-w_{i-1})-(p'\wedge s)w_i+(p''\wedge s)w_{i-1} \\ &=& w_i (p\wedge s) + \nabla w_i (p' \wedge s) \hspace{25pt} \text{since $p''\wedge s=p\wedge s\,+\,p'\wedge s$} \\ &\leq &\phi^+_i(p\wedge s)+\nabla \phi^+_i(p'\wedge s)
572: \\ &=&\textstyle
573: {\left(\frac{\beta^+}{\alpha^+}\wedge p'\right)(p\wedge s)
574: +\left(\frac{\beta^+}{\alpha^+}\wedge p\right)(p'\wedge s)} \hspace{20pt}(*)
575: \end{array} \end{equation}
576: by definition (\ref{phiphi}) of $\phi^+$. The last quantity is $\frac{\beta^+}{\alpha^+}\wedge s$ (hence the upper bound): by $SL_2$-invariance of the $\wedge$-notation, it is enough to check this when $p=\infty$ and $p'=0$ --- in that case, $\frac{\beta^+}{\alpha^+}$ and $s=\frac{\eta}{\xi}$ have opposite signs, and $(*)$ does indeed
577: become $|\beta^+ \xi|+|\alpha^+ \eta|=\left | \left |^{\beta^+}_{\alpha^+} ~^{\eta}_{\xi} \right | \right |$.
578: \end{proof}
579:
580: \section{Hyperbolic volume} \llabel{sectionhyperbolicvolume}
581:
582: %EXCISION...
583: The product topology on $\mathbb{R^Z}$ induces a natural topology
584: on the space $W$ of Definition \ref{defineW}: clearly, $W$ is nonempty, convex, and compact.
585:
586: If $(x,y,z)$ is a nonnegative triple such that $x+y+z=\pi$, let ${\mathcal
587: V}(x,y,z)$ be the hyperbolic volume of an ideal tetrahedron whose interior
588: dihedral angles are $x,y,z$. We wish to compute the total hyperbolic volume of
589: all tetrahedra when $w\in W$, i.e. $$\mathcal{V}(w):=\sum_{i\in\mathbb{Z}}
590: \mathcal{V}(x_i,y_i,z_i)$$ where $x_i,y_i,z_i$ are defined from the $w_i$
591: \emph{via} Table (\ref{xiyi}). This poses the problem of well-definedness --- the sum of the volumes might diverge. Let us estimate $\mathcal{V}$: a well-known explicit formula \cite{milnor} gives
592:
593: \begin{equation} \begin{array}{rcl}
594: {\mathcal V}(x,y,z)&=&\displaystyle{-\int_0^x \log 2\sin - \int_0^y\log 2\sin - \int_0^z\log 2\sin} \\
595: &=&{\displaystyle \int_0^x \log \frac{\sin(\tau+y)}{\sin \tau} d\tau}
596: \hspace{15pt}\text{(as $\int_0^{\pi}\log 2 \sin = 0$)}\\
597: &\leq&\displaystyle{\int_0^x \log \frac{\tau+y}{\tau} d\tau} \hspace{20pt}
598: \text{(by concavity of $\sin$)}\\&=&\displaystyle{x\log \frac{x+y}{x}
599: +y\log \frac{x+y}{y} } \\ &\leq& (x+y)\log 2 \hspace{35pt} \text{(concavity of
600: $\log$).} \end{array} \llabel{volumeformula} \end{equation}
601:
602: \begin{lemma} \llabel{volumelemma} There exists a universal constant $K>0$ such
603: that the sum of the volumes of the tetrahedra $\Delta_j$ for $j\geq i$ (resp.
604: $j\leq i$) is at most $K\phi^+_i$ (resp. $K\phi^-_i$). \end{lemma}
605: \begin{proof} We will focus only on the $\phi^+$-statement. First, by the
606: computation above, the volume of the tetrahedron $\Delta_i$ is at
607: most $w_i \log 2 \leq \phi^+_i \log 2$ (see Table \ref{xiyi}). In
608: view of Lemma \ref{studyphi}-{\sf vi}, this implies that the total
609: volume of all \emph{hinge} tetrahedra beyond the index $i$ is at
610: most $3\phi^+_i\log 2$. For the same reason, it is sufficient to
611: prove
612: \begin{sublemma} \llabel{boundhops} There exists a
613: universal constant $L>0$ such that if $0$ and $N\in{\mathbb N}$ are two
614: consecutive hinge indices, then the sum of the volumes of the tetrahedra
615: $\Delta_1, \Delta_2, \dots, \Delta_{N-1}$ is at most $L \phi_0^+$.
616: \end{sublemma}
617:
618: \begin{proof} In view of homogeneity in the
619: estimation (\ref{volumeformula}), it is sufficient to
620: assume $\phi_0^+=1$ and replace the volume with its estimate.
621: Therefore, let $(w_i)_{0\leq i \leq N}$ be a concave sequence in
622: $[0,1]$: following Table (\ref{xiyi}), we want to find a universal
623: upper bound $L$ (not depending on $N$) for $$
624: \sum_{i=1}^{N-1}\frac{w_{i+1}+w_{i-1}}{2}\log\frac{2w_i}{w_{i+1}+w_{i-1}}+
625: \frac{2w_i-w_{i+1}-w_{i-1}}{2}\log\frac{2w_i}{2w_i-w_{i+1}-w_{i-1}}.$$
626: If $A<B$ are positive integers, denote by $\Sigma_A^B$ the
627: restriction of the above sum to indices $A\leq i < B$. Observe that
628: the general term of $\Sigma_A^B$ is bounded by $2e^{-1}$, because
629: $\frac{\tau}{2}\log\frac{2}{\tau}\leq e^{-1}$ for all positive
630: $\tau$. To be more efficient, we bound the first half of the general
631: term by
632: $(\frac{w_{i+1}+w_{i-1}}{2})(\frac{2w_i}{w_{i+1}+w_{i-1}}-1)$, and
633: the second half, by concavity of $\log$. This produces
634: \begin{eqnarray*} \Sigma_A^B&\leq&
635: \sum_{i=A}^{B-1}\left (\frac{2w_i-w_{i+1}-w_{i-1}}{2} \times \frac{w_{i+1}+w_{i-1}}{w_{i+1}+w_{i-1}} \right ) \\
636: && + \left ( \sum_{i=A}^{B-1} \frac{2w_i-w_{i+1}-w_{i-1}}{2}\right )
637: \log\frac{\sum_{i=A}^{B-1}w_i}{\sum_{i=A}^{B-1}
638: \frac{2w_i-w_{i+1}-w_{i-1}}{2}} \\ &=& \sigma \log \frac{e
639: \sum_{i=A}^{B-1}w_i}{\sigma}~,~~~~~\text{ where
640: }\sigma=\frac{w_A-w_{A-1} +w_{B-1}-w_B}{2}. \end{eqnarray*} Denote
641: by $M\in \llbracket 0, N \rrbracket$ a value of the index $i$ for
642: which $w_i$ is maximal. If $A<B\leq M$, we have $$0 \leq \sigma \leq
643: \frac{w_A-w_{A-1}}{2} \leq \frac{1}{2A} \leq 1$$ (the third
644: inequality follows from concavity of $w$ between $0$ and $A$). Also
645: notice that $f: \tau \mapsto \tau \log \frac{e}{\tau}$ is
646: nondecreasing on $[0,1]$. We shall apply these facts for $A=2^{k-1}$
647: and $B=\min\{2^k,M\}$: the previous bound on $\Sigma_A^B$ can be rewritten
648: \begin{eqnarray*} \Sigma_{A}^{B}&\leq&
649: \textstyle{ f(\sigma)+ \sigma \log\left ( \sum_{i=A}^{B-1} w_i \right )}\\
650: &\leq&\textstyle{f(2^{-k})+2^{-k}\log 2^{k-1} = 2^{-k}[1+(2k-1)\log
651: 2].} \end{eqnarray*} The latter numbers (for $k$ ranging over
652: $\mathbb{N}^*$) add up to some universal $L'<+\infty$. After a
653: similar argument for the indices $M<i<N$, we can take
654: $L=2e^{-1}+2L'$.
655: \end{proof} Finally, we can take $K=3L+3\log 2$. Lemma \ref{volumelemma} is
656: proved. \end{proof}
657:
658: \begin{corollary} The volume functional ${\mathcal V}:W\rightarrow {\mathbb
659: R}^+$ is well-defined, continuous, and concave.
660: \end{corollary}
661:
662: \begin{proof} Well-definedness is the point of Lemma \ref{volumelemma}. Given
663: $\varepsilon>0$, only finitely many indices $i$ satisfy
664: $\min\{\phi_i^+,\phi_i^-\}>\varepsilon/K$, and the others contribute at most
665: $2\varepsilon$ to the volume: hence continuity in the product topology.
666: Concavity follows from the concavity of the volume of \emph{one} tetrahedron
667: (parametrized by its angles): see e.g.
668: %\cite{concavityref}
669: Proposition 8 of \cite{mapomme}. \end{proof}
670:
671: Therefore, by compactness, there exists a sequence
672: $(w_i)_{i\in\mathbb{Z}}\in W$ which maximizes the hyperbolic volume
673: $\mathcal{V}$. From this point on, $w$ will denote that
674: maximizer.
675:
676: \begin{proposition}
677: For each $j\in\mathbb{Z}$, if $x_j y_j z_j=0$ then $\max\{x_j,y_j,z_j\}=\pi$.
678: \llabel{tricot} \end{proposition}
679: \begin{proof}
680: Assume the tetrahedron $\Delta_j$ has exactly one vanishing angle, and aim for a contradiction. If $\mathcal{V}(\Delta^t)$ is the volume of a tetrahedron $\Delta^t$ having angles
681: $x^t,y^t,(\pi-x^t-y^t)$ with $(x^t, y^t)_{t\geq 0}$ smooth, $x^0=0<y^0<\pi$ and $dx^t/dt_{|t=0}>0$, then
682: $d\mathcal{V}(\Delta^t)/dt_{|t=0}=+\infty$ (by Formula (\ref{volumeformula}) above).
683:
684: Let $(w'_i)_{i\in\mathbb{Z}}$ be a sequence satisfying all (strict) inequalities of (\ref{doubleve}) and define $w^t:=w+t(w'-w)$ for $0\leq t \leq 1$.
685: Denote by $\Delta^t_i$ the $i$-th tetrahedron determined via (\ref{xiyi}) by $(w^t_i)_{i\in\mathbb{Z}}$: then the angles of $\Delta^t_j$ satisfy the hypotheses above, so
686: $$\mathcal{V}(w^t)=\mathcal{V}(\Delta^t_j)+\left ( \sum_{i\neq j}\mathcal{V}(\Delta^t_i)\right )$$
687: has right derivative $+\infty$ at $t=0$ (the second summand is concave and continuous, so it has a well-defined right derivative in $\mathbb{R}\cup \{+\infty\}$ at $0$). Therefore, $\mathcal{V}$ was not maximal at $w$. \end{proof}
688:
689: %%%%%%%%%%%%%%%%%%%%%%
690: %%%%%%%%%%%%%%%%%%%%%%
691: %%% %%%
692: %%% TRICOT %%%
693: %%% %%%
694: %%% --> JUNK %%%
695: %%% %%%
696: %%%%%%%%%%%%%%%%%%%%%%
697: %%%%%%%%%%%%%%%%%%%%%%
698:
699: In Section \ref{sectionextensions} we will need the following consequence of Proposition \ref{tricot}:
700:
701: \begin{proposition} \llabel{hingefragile}
702: If $j\in \mathbb{Z}$ and $x_j y_j z_j=0$, then $j$ is a hinge index and $w_j=0$. \qed \end{proposition}
703: This is Proposition 13 in \cite{mapomme}. But in fact much more is true:
704: \begin{proposition} \llabel{winterior}
705: All the (strict) inequalities of (\ref{doubleve}) are true at $w$.
706: \end{proposition}
707: \begin{proof}
708: If some inequality of (\ref{doubleve}) involving $\phi^+$
709: fails to be strict, it is easy to see
710: by induction that $w=\phi^+$ near $+\infty$, so all tetrahedra $\Delta_i$ (for $i$ large enough) have exactly one vanishing angle: $w$ was not maximal, by Proposition \ref{tricot}. Therefore all inequalities of (\ref{doubleve}) involving $\phi^{\pm}$ are strict.
711: The arguments in \cite{mapomme} (especially
712: Lemma 16 and the argument of Section 9 there) can then be used
713: to show that \emph{all} inequalities (\ref{doubleve}) are strict at
714: $w$, so $w$ is a \emph{critical} point of the volume $\mathcal{V}$.
715: \end{proof}
716:
717: Proposition \ref{winterior} implies that the holonomy representation
718: is trivial, i.e. the gluing of any finite number of consecutive
719: tetrahedra $\Delta_i$ defines a complete hyperbolic metric with
720: (polyhedral) boundary (the shapes of the $\Delta_i$ ``fit together
721: correctly'' around the edges): see \cite{rivin}, \cite{chanhodgson} or \cite{mapomme}. Therefore, the links of the vertices of the ideal tetrahedra
722: (Euclidean triangles) form a triangulation of the link of the
723: puncture: the latter is naturally endowed with a Euclidean structure
724: and its universal cover can be drawn in the plane (Figure
725: \ref{frise} --- more on the combinatorics of this triangulation in
726: Section \ref{sectionthecusplink}; see also \cite{mapomme}). Denote by $\Gamma$ the image of the induced holonomy representation $\pi_1(S)\rightarrow \text{Isom}^+(\mathbb{H}^3)$.
727: \begin{figure}[h!]
728: \centering
729: \psfrag{L}{$\mathcal{L}$}
730: \ledessin{FriseLive}
731: \caption{The cusp triangulation is shown, in anticipation, against the limit set $\mathcal{L}\subset \mathbb{P}^1\mathbb{C}$ of the quasifuchsian group (a complicated Jordan curve). The picture extends periodically to the right and left. Each broken line is the puncture link of a pleated surface $S_i$; the vertices (whose design artificially sets the $S_i$ apart from each other) are all parabolic fixed points at which the limit set $\mathcal{L}$ becomes pinched. Infinitely many very flat triangles accumulate along the top and bottom horizontal lines. This picture was generated with Masaaki Wada's computer program Opti \cite{opti}. \llabel{frise}}
732: \end{figure}
733:
734: \section{Behavior of $(w_i)$} \llabel{sectionbehaviorofwi}
735:
736: \subsection{Properties of $(w_i)$}
737: We want to prove that the pleatings $\lambda_i$ (see Section
738: \ref{sectionboundingthebending}) of the pleated punctured tori $S_i$
739: converge (in the weak-* topology, i.e. on any test curve $\gamma \in
740: \mathcal{C}$) to the pleatings $\lambda^{\pm}$ near infinity. For
741: this we must study $(w_i)$, and especially show that the bounds
742: specified by $\phi^{\pm}$ in Equations (\ref{doubleve}) are almost
743: (but not quite) achieved. We use the $\nabla$-notation as in Section \ref{behavior}.
744:
745: \begin{lemma} \llabel{hiatus} Recall that
746: $\nabla \phi^-<0<\nabla \phi^+$ (Lemma \ref{studyphi}). One has $$\max \left\{
747: \underset{\mathbb{Z}}{\overline{\lim}}\frac{w}{\phi^+}~,~
748: \underset{\mathbb{Z}}{\overline{\lim}}\frac{\nabla
749: w}{\nabla\phi^+}\right\} = \max
750: \left\{\underset{\mathbb{Z}}{\overline{\lim}}\frac{w}{\phi^-}~,~
751: \underset{\mathbb{Z}}{\overline{\lim}}\frac{-\nabla w}{-\nabla
752: \phi^-}\right\} = 1.$$ \end{lemma}
753:
754: \begin{proof} We focus on the $\phi^+$-statement; the $\phi^-$-part is
755: analogous. Since $\frac{w}{\phi^+}<1$, assuming $\overline{\lim}
756: \frac{w}{\phi^+}<1$ implies $\sup \frac{w}{\phi^+}<1$, and the same
757: holds true for $\frac{\nabla w}{\nabla \phi^+}$ (see Equation
758: \ref{doubleve}). Therefore, suppose $\sup \frac{w}{\phi^+} \leq
759: 1-\varepsilon$ and $\sup \frac{\nabla w}{\nabla \phi^+}\leq
760: 1-\varepsilon$ for some $\varepsilon >0$, and aim at a
761: contradiction.
762:
763: Recall the ordered pair $(\alpha^+, \beta^+)$ that helped define
764: $\phi^+$. For each $\mu>0$, define
765: $(\alpha^{\mu},\beta^{\mu}):=(\mu \alpha^+, \mu \beta^+)$. This
766: defines a new pleating function $\lambda^{\mu}=\mu \lambda^+ :
767: \mathcal{C} \rightarrow \mathbb{R}^+$, a new $\phi^{\mu}=\mu
768: \phi^+$ and a new domain $W^{\mu}$ by (\ref{doubleve}) (the numbers
769: $\alpha^-$ and $\beta^-$ are left unchanged). By definition, $W=W^1$
770: and $W^{\mu}\subset W^{\mu'}$ if and only if $\mu \leq \mu'$. Let
771: $(w^{\mu}_i)_{i\in \mathbb{Z}}$ be the maximizer of the volume
772: functional $\mathcal{V}$ on $W^{\mu}$. By assumption, we have
773: $w^1\in W^{1-\varepsilon}\subset W^1$, so $w^1=w^{1-\varepsilon}$.
774:
775: Write $V(\mu)=\mathcal{V}(w^{\mu})$, so that
776: $V(1)=V(1-\varepsilon)$. It is straightforward to check that for any
777: $t\in [0,1]$, one has $(t\cdot w^{\mu}+(1-t)\cdot w^{\mu'}) \in
778: W^{t\mu+(1-t)\mu'}$. Since the volume of any tetrahedron is a
779: concave function of its angles, this is enough to imply that
780: $V:\mathbb{R}^+\rightarrow \mathbb{R}^+$ is (weakly) concave. By
781: inclusion, $V$ is also nondecreasing. In fact, $V$ is strictly
782: increasing (which will finish the proof by contradiction): to prove
783: this, since $V$ is concave, we just need to produce arbitrarily
784: large values of the volume $\mathcal{V}$ at points $v^{\mu} \in
785: W^\mu$, for large enough $\mu$. We may assume (up to a translation
786: of indices, see Lemma \ref{studyphi}, {\sf vii} -- {\sf viii}) that
787: $\phi^-_0>\pi$ and $-\nabla\phi^-_0>\pi$. Then, start by defining
788: $v^{\mu}_i=\min\{\phi^{\mu}_i,\phi^-_i,\pi\}$, so that
789: $v^{\mu}=\pi$ on $\llbracket 0,N+1\rrbracket$ for arbitrarily large
790: $N$. Without loss of generality, by just taking $\mu$ large enough,
791: we may further assume $|\nabla\phi^{\mu}|>\pi$ on $\llbracket
792: 0,N+1\rrbracket$. Then, each time $i\in\llbracket 1,N\rrbracket$ is
793: a \emph{hinge} index, replace $v^{\mu}_i$ by $2\pi/3$: this is
794: allowed in $W^{\mu}$, by our assumptions on $\nabla\phi^-,
795: \nabla\phi^{\mu}$. The angles of the tetrahedron $\Delta_i$ are
796: then all in $\{\pi/2,\pi/3,\pi/6\}$ (see Table \ref{xiyi}). Since
797: there are infinitely many hinge indices near $+\infty$, the volume
798: can become arbitrarily large: QED. \end{proof}
799:
800: \begin{corollary} \llabel{hiatus2} In fact,
801: $\displaystyle{ \underset{i\rightarrow +\infty}{\overline{\lim}}
802: \min \left \{ \frac{w_i}{\phi^+_i} , \frac{\nabla w_i}{\nabla\phi^+_i} \right\}
803: = \underset{i\rightarrow -\infty}{\overline{\lim}}
804: \min \left \{ \frac{w_{i-1}}{\phi^-_{i-1}} , \frac{-\nabla
805: w_i}{-\nabla\phi^-_i} \right \} = 1 }$. \end{corollary}
806:
807: \begin{proof} Again, we focus only on $\phi^+$. By Lemma \ref{hiatus}, there
808: exists a subsequence $(w_{\nu(i)})_{i\in\mathbb{N}}$ such that
809: $w_{\nu(i)}\sim\phi^+_{\nu(i)}$ or $\nabla w_{\nu(i)}\sim
810: \nabla\phi^+_{\nu(i)}$. Suppose the latter is the case. For an
811: arbitrary integer $i$, let $n$ be the smallest hinge index larger
812: than or equal to $\nu(i)$: observe that $\phi^+_{n+1}=\nabla\phi^+_n
813: =\nabla\phi^+_{\nu(i)}$ by Lemma \ref{studyphi}{\sf -iii-iv}, while
814: $w_{n+1}\geq \nabla w_n\geq \nabla w_{\nu(i)}$ by the positivity
815: conditions (\ref{doubleve}). Therefore, $\frac{w_{n+1}}
816: {\phi^+_{n+1}}\geq \frac{\nabla w_{\nu(i)}} {\nabla\phi^+_{\nu(i)}}$
817: so up to redefining $\nu$ we may assume simply
818: $w_{\nu(i)}\sim\phi^+_{\nu(i)}$.
819:
820: Pick $\varepsilon>0$. Take $i$ such that \begin{equation}
821: \llabel{press}w_{\nu(i)}\geq(1-\varepsilon)\phi^+_{\nu(i)}.\end{equation}
822: Let $n$ be the smallest hinge index strictly larger than $\nu(i)$.
823: If $n=\nu(i)+1$ then $$w_n\geq
824: w_{n-1}-\nabla\phi^+_n\geq(1-\varepsilon)
825: \phi^+_{n-1}-\phi^+_{n+1}=\phi^+_n-\varepsilon \phi^+_{n-1}\geq
826: (1-2\varepsilon)\phi^+_n;$$ $$\nabla w_n=w_{n-1}-w_n \geq (1-\varepsilon)\phi^+_{n-1}-\phi^+_n =\phi^+_{n+1}-\varepsilon\phi^+_{n-1}\geq
827: (1-3\varepsilon) \phi^+_{n+1}= (1-3\varepsilon)\nabla\phi^+_n$$
828: where Lemma \ref{studyphi} has been used several times. Therefore,
829: $\min\left \{\frac{w_n}{\phi^+_n}, \frac{\nabla w_n}{\nabla\phi^+_n}
830: \right\} \geq 1-3\varepsilon$.
831:
832: If $n\geq \nu(i)+2$, we can find an index $k$ such that
833: $\frac{k-\nu(i)}{n-\nu(i)}\in[\frac12,\frac23]$. We will show that
834: $\min\left \{\frac{w_k}{\phi^+_k}, \frac{\nabla w_k}{\nabla\phi^+_k}
835: \right\} \geq 1-8\varepsilon$, which will finish the proof.
836:
837: \noindent $\bullet$ By positivity of $w$ and concavity of $w$ between the points
838: $(\nu(i),k,n)$ one has $w_k\geq \frac13 w_{\nu(i)}$. Therefore,
839: $$\phi^+_k-w_k\leq \phi^+_{\nu(i)}-w_{\nu(i)}\leq
840: \left (\frac{1}{1-\varepsilon}-1\right )w_{\nu(i)} \leq
841: \frac{3\varepsilon}{1-\varepsilon}w_k$$ (here the first inequality
842: holds because $(\phi^+-w)$ is decreasing by Condition
843: (\ref{doubleve}), and the second follows from the assumption
844: (\ref{press}) above). Hence,
845: $$\frac{w_k}{\phi_k^+}\geq\frac{1-\varepsilon}{1+2\varepsilon}\geq
846: 1-3\varepsilon.$$
847:
848: \noindent $\bullet$ Observe that $$(1-\varepsilon)\phi^+_{\nu(i)}\leq w_{\nu(i)}\leq w_k+(k-\nu(i))\nabla w_k \leq \phi^+_k+(k-\nu(i))\nabla w_k$$ where
849: the second inequality follows from concavity of $w$ between the
850: points $(\nu(i),k-1,k)$. It follows that $$(k-\nu(i))\nabla w_k\geq
851: \phi^+_{\nu(i)}-\phi^+_k-\varepsilon \phi^+_{\nu(i)}
852: =(k-\nu(i))\nabla\phi^+_k-\varepsilon \phi^+_{\nu(i)}$$ hence $$
853: \nabla w_k \geq
854: \nabla\phi^+_k-\frac{\varepsilon}{k-\nu(i)}\phi^+_{\nu(i)} \geq
855: \nabla\phi^+_k-\frac{2\varepsilon}{n-\nu(i)}\phi^+_{\nu(i)}.$$ Since
856: Lemma \ref{studyphi}-{\sf vi} gives us $\phi^+_{\nu(i)}\leq
857: 2(n-\nu(i))\phi^+_n$, this yields $$\nabla w_k \geq
858: \nabla\phi^+_k-4\varepsilon \phi^+_n \geq
859: \nabla\phi^+_k-8\varepsilon \phi^+_{n+1}=(1-8\varepsilon)\nabla
860: \phi^+_k.$$
861: \end{proof}
862:
863: \begin{corollary} Recall the pleating $\lambda_i$ of the pleated surface $S_i$.
864: For any simple closed curve $\gamma\in \mathcal{C}$ we have $\underset{i\rightarrow +\infty}{\lim}\lambda_i(\gamma)=\lambda^+(\gamma)$ and
865: $\underset{i\rightarrow -\infty}{\lim}\lambda_i(\gamma)=\lambda^-(\gamma)$.
866: \end{corollary}
867: \begin{proof}
868: By Corollary \ref{hiatus2}, the member ratio in the inequality (\ref{determinant}) can be made arbitrarily close to $1$.
869: \end{proof}
870:
871: \section{The cusp link} \llabel{sectionthecusplink}
872:
873: We now aim to investigate the behavior of the pleated surfaces $S_i$
874: as $i$ goes to $\pm \infty$ --- or, more precisely, to find two
875: limiting pleated surfaces $S_{\pm\infty}$ with pleatings
876: $\lambda^{\pm}$ such that $\overline{V}=\bigcup_{i\in\mathbb{Z}}
877: \Delta_i \sqcup S_{+\infty} \sqcup S_{-\infty}$ is metrically
878: complete with locally convex boundary. The difficult part is to
879: prove that the intrinsic moduli of the $S_i$ converge in
880: Teichm\"uller space. This question will be addressed in the next
881: section. In this section, we just describe the cusp link, introduce
882: notation and prove a few inequalities.
883:
884: As always, we will mainly work near $+\infty$. Define
885: $V=\bigcup_{i\in\mathbb{Z}}\Delta_i$. We begin by orienting all the
886: edges of $V$ in a way that will be consistent with the pleating data
887: $(\alpha^{\pm},\beta^{\pm})$. Namely, denote by $\mathcal{E}
888: \subset {\mathbb P}^1{\mathbb Q}$ the collection of all the
889: endpoints of all the Farey edges $(e_i)_{i\in {\mathbb Z}}$. For
890: $s\in \mathcal{E}$, let $Q_s$ be the edge of $V$ of slope $s$.
891: Recall that the punctured torus is defined as $({\mathbb
892: R}^2\smallsetminus {\mathbb Z}^2)/{\mathbb Z}^2$. Orienting $Q_s$ is
893: therefore equivalent to orienting the line $F_s$ of slope $s$ in
894: $\mathbb{R}^2$. We decide that the positive half of $F_s$ should be
895: on the same side of the line ${\mathbb R} (\alpha^-, \beta^-)$ as
896: $(\alpha^+, \beta^+)$, and orient $Q_s$ accordingly.
897:
898: Let us now describe the link of the puncture, or cusp triangulation.
899: Each tetrahedron $\Delta_i$ contributes four similar Euclidean
900: triangles at infinity to the link of the puncture, corresponding to
901: the four ideal vertices of $\Delta_i$. The bases of these four
902: triangles form a closed curve which is a broken line of four
903: segments, and the triangles point alternatively up and down from
904: this broken line (see Figure \ref{fourtriangles}). The two upward
905: (resp. downward) pointing triangles have the same Euclidean size,
906: an effect of the hyperelliptic involution (rotation of $180^{\circ}$
907: around the puncture) which acts isometrically on $V$ (reversing all
908: edge orientations) and as a horizontal translation on the cusp link.
909:
910: \begin{figure}[h!] \centering
911: \psfrag{Q0}{$Q_0$}
912: \psfrag{Q1}{$Q_1$}
913: \psfrag{Qinf}{$Q_{\infty}$}
914: \psfrag{inf}{$(\!\bigstar\!)$}
915: \psfrag{tof}{: toward $(\!\bigstar\!)$}
916: \psfrag{frf}{: from $(\!\bigstar\!)$}
917: \psfrag{x}{$x$}
918: \psfrag{y}{$y$}
919: \psfrag{z}{$z$}
920: \psfrag{bap}{$\displaystyle{\frac{\beta^+}{\alpha^+}}$}
921: \ledessin{fourtriangles}
922: \caption{In the left panel, $(\!\bigstar\!)$ marks the cusp. In the right panel, the cusp is at infinity. \llabel{fourtriangles}}
923: \end{figure}
924:
925: \begin{definition} \llabel{definebases}
926: If the loop around the cusp has Euclidean length $4$, let $b_i$
927: (resp. $b'_i$) be the length of the base of a downward-pointing
928: (resp. upward-pointing) triangle contributed by the tetrahedron
929: $\Delta_i$, and define $\sigma_i=b'_i/b_i$ (see e.g. Figure \ref{sigmalimit}).
930: \end{definition}
931:
932: Figure \ref{fourtriangles} also shows some additional information,
933: assuming that $e_i=0\infty$ and $\alpha^+, \beta^+>0$ (hence
934: $\beta^-/\alpha^-<0$). Namely, the pleated surfaces $S_{i+1}$ and
935: $S_i$ (above and below $\Delta_i$) are pleated along $Q_0,
936: Q_{\infty}, Q_1$ and $Q_0, Q_{\infty}, Q_{-1}$ respectively, and the
937: orientations of the $Q_s$ are as shown in the left panel of Figure
938: \ref{fourtriangles} (edge $Q_{-1}$ has no determined orientation).
939: The orientations of the lines to/from the puncture are also shown in
940: the right panel, at the vertices, with the help of a color code.
941: Moreover, each segment $\epsilon$ of the upper broken line in the
942: right panel corresponds to an arc about a vertex $v$ of a face $f$
943: of $S_{i+1}$ in the left panel, so $\epsilon$ receives the
944: orientation of the edge of $f$ opposite $v$. If $\tau$ is one of the
945: upward-pointing triangles drawn in the plane $\mathbb{C}$ (right
946: panel), consider the tetrahedron $\Delta$ whose vertices are
947: $\infty$ and those of $\tau$: all edges of $\Delta$, except one,
948: receive orientations from the construction above, and $\Delta$ is
949: isometric (respecting these orientations) to $\Delta_i$. Finally,
950: notice the labels in the $3$ corners of each triangle in the right
951: panel: the corner of the free vertex is labeled $z$ (the angle there
952: being $z_i$); the other two corners are labeled $x$ and $y$
953: accordingly. The labels $x$-$y$-$z$ appear clockwise in each triangle.
954:
955: The contribution of the tetrahedron $\Delta_{i-1}$ to the link at
956: infinity is also a union of four triangles bounded by two broken
957: lines. Moreover, the upper broken line from $\Delta_{i-1}$ is the
958: lower broken line from $\Delta_i$, and the orientations of the lines
959: to/from infinity must agree. Inspection shows that there are only
960: two possibilities, corresponding to whether the letter living on the
961: surface $S_i$ is $R$ or $L$: for $R$, the $x$-corners of the two
962: $4$-triangle families live near a common vertex $C$; for $L$, the same
963: is true of the $y$-corners (Figure \ref{eighttriangles}).
964: \begin{figure}[h!] \centering
965: \psfrag{tof}{: to $\infty$}
966: \psfrag{frf}{: from $\infty$}
967: \psfrag{A}{$A$}
968: \psfrag{B}{$B$}
969: \psfrag{C}{$C$}
970: \psfrag{b}{$b$}
971: \psfrag{c}{$c$}
972: \psfrag{R}{$(R)$}
973: \psfrag{L}{$(L)$}
974: \ledessin{eighttriangles}
975: \caption{Left and Right transitions. Compare with the right panel of Figure \ref{fourtriangles}. \llabel{eighttriangles}} \end{figure}
976:
977: \begin{definition}
978: In a downward-pointing triangle defined by $\Delta_i$, the edge
979: lengths are $b_i$ (the basis from Def. \ref{definebases}), $b_{i-1}$, and a third number which we call $c_i$.
980: \end{definition}
981:
982: \begin{property} For $i$ large enough, $(b_i)$ is increasing and $(b'_i)$ is
983: decreasing. \end{property}
984: \begin{proof} Let $\tau_i$ (resp. $\tau'_i$) be a downward-pointing
985: (resp. upward-pointing) triangle defined by the tetrahedron
986: $\Delta_i$. For large enough $i$ we have $z_i=\pi-w_i>\pi-\phi_i^+
987: >\pi/2$, so $b_i$ is the longest edge of $\tau_i$, and $b'_i$ the longest edge of $\tau'_i$. Since $b_{i-1}$
988: is an edge of $\tau_i$ and $b'_{i+1}$ is an edge of $\tau'_i$
989: (Figure \ref{eighttriangles} or \ref{sigmalimit}), the conclusion follows.
990: \end{proof}
991: \begin{property}We have $\lim_{i\rightarrow +\infty}\sigma_i=0$.
992: \end{property}
993: \begin{proof}
994: We already know that $(\sigma_i)=(b'_i/b_i)$ is ultimately
995: decreasing. It is therefore enough to show that
996: $\sigma_{i+1}/\sigma_{i-1}\leq 1/2$ for large enough hinge indices
997: $i$. Consider Figure \ref{sigmalimit},
998: where angles labeled $z$ are obtuse (as a rule, we shade a cusp triangle contributed by $\Delta_i$ whenever $i$ is a hinge index). Check that
999: $$\frac{\sigma_{i+1}}{\sigma_{i-1}}= \frac{b'_{i+1}}{b'_{i-1}}
1000: \frac{b_{i-1}}{b_{i+1}} < \frac{b'_{i+1}}{b'_i} \frac{b_{i-1}}{b_i}=
1001: \frac{\sin x_i \sin y_i}{\sin^2 z_i}\leq \frac{\sin^2(w_i/2)}{\sin^2
1002: w_i}<\frac12$$ (the last two inequalities follow from an easy study
1003: of $\sin$, using $x_i+y_i=\pi-z_i=w_i<\pi/2$). As an immediate
1004: consequence, we find $\lim_{+\infty}b'_i=0$ and
1005: $\lim_{+\infty}b_i=2$. Similarly, $\lim_{-\infty}b'_i=2$ and
1006: $\lim_{-\infty}b_i=0$. (Compare with Figure \ref{frise}).
1007: \begin{figure}[h!] \centering
1008: \psfrag{xi}{$x_i$}
1009: \psfrag{yi}{$y_i$}
1010: \psfrag{zi}{$z_i$}
1011: \psfrag{zip}{$z_{i+1}$}
1012: \psfrag{zim}{$z_{i-1}$}
1013: \psfrag{bi}{$b_i$}
1014: \psfrag{bip}{$b_{i+1}$}
1015: \psfrag{bim}{$b_{i-1}$}
1016: \psfrag{Bi}{$b'_i$}
1017: \psfrag{Bip}{$b'_{i+1}$}
1018: \psfrag{Bim}{$b'_{i-1}$}
1019: \ledessin{sigmalimit}
1020: \caption{The index $i$ is a hinge; the nature of $i\pm 1$ is undetermined. \llabel{sigmalimit}} \end{figure}
1021: \end{proof}
1022: \begin{definition}
1023: Let $J\subset {\mathbb Z}$ be the set of all integers $j$ such that
1024: $j-1$ is a hinge index. \llabel{definitionJ}
1025: \end{definition}
1026: \begin{proposition} \llabel{ecrase}
1027: If $j<l$ are two large enough consecutive elements of $J$, and $k$
1028: is not in $J$, then $$\frac{c_k}{c_{k-1}}= \sigma_{k-1}~~~
1029: \text{and} ~~~ \frac{c_l}{c_j} \leq \sigma_{j-2}.$$
1030: \end{proposition}
1031: \begin{proof}
1032: Since $k-1$ is not a hinge index, $b_{k-1}$ shares the same end with
1033: $b_k$ and with $b_{k-2}$
1034: (see Figure \ref{ecrasefigure}, left) so
1035: $c_k/c_{k-1}=b'_{k-1}/b_{k-1}=\sigma_{k-1}$.
1036:
1037: Since $c_l$ is always an edge of $\tau'_{l-1}$, we have $c_l\leq
1038: b'_{l-1}$, hence
1039: \begin{equation} \frac{c_l}{c_j}\leq\frac{b'_{l-1}}{c_j}
1040: \leq\frac{b'_j}{c_j}
1041: =\frac{c_{j-1}}{b_{j-2}}\leq\frac{b'_{j-2}}{b_{j-2}}
1042: =\sigma_{j-2}\end{equation} where the equality in the middle just
1043: translates the similarity property of the ``hinge'' triangles
1044: $\tau_{j-1},\tau'_{j-1}$
1045: (shaded in Figure \ref{ecrasefigure}, right).
1046: \end{proof}
1047: \begin{figure}[h!] \centering
1048: \psfrag{cj}{$c_j$}
1049: \psfrag{bj}{$b_j$}
1050: \psfrag{Bj}{$b'_j$}
1051: \psfrag{cjm}{$c_{j-1}$}
1052: \psfrag{bjm}{$b_{j-1}$}
1053: \psfrag{Bjm}{$b'_{j-1}$}
1054: \psfrag{bjM}{$b_{j-2}$}
1055: \psfrag{BjM}{$b'_{j-2}$}
1056: \psfrag{ck}{$c_k$}
1057: \psfrag{ckm}{$c_{k-1}$}
1058: \psfrag{bkm}{$b_{k-1}$}
1059: \psfrag{Bkm}{$b'_{k-1}$}
1060: \ledessin{ecrasefigure}
1061: \caption{Left: $k-1$ is not a hinge. Right: $j-1$ is a hinge. \llabel{ecrasefigure}}
1062: \end{figure}
1063:
1064: \section{Intrinsic convergence of the surfaces $S_i$} \llabel{sectionintrinsicconvergence}
1065:
1066: \subsection{Thickness of the tetrahedra}
1067:
1068: Consider a tetrahedron $\Delta_i$ bounded by the pleated surfaces
1069: $S_i$ and $S_{i+1}$. Let $Q$ (resp. $Q'$) be the pleating edge of
1070: $S_i$ (resp. $S_{i+1}$) not lying in $S_i \cap S_{i+1}$. Let $s_i$
1071: be the shortest segment between $Q$ and $Q'$, across $\Delta_i$.
1072:
1073: \begin{definition}
1074: Recall the orientations on the edges of $V=\bigcup_{i\in
1075: \mathbb{Z}}\Delta_i$. Let $\ell_i$ be the complex length of the
1076: hyperbolic loxodromy along $s_i$ sending $Q$ to $Q'$, respecting the
1077: orientations of $Q, Q'$ (with $-\pi< {\rm Im} \, \ell_i \leq
1078: \pi$).
1079: \end{definition}
1080:
1081: \begin{proposition} The series $(\ell_i)_{i\in {\mathbb Z}}$ is
1082: absolutely convergent. \llabel{helicoptere} \end{proposition}
1083:
1084: \begin{proof} Consider a downward-pointing triangle $\tau$ contributed by
1085: $\Delta_i$. Label the vertices of $\tau$ by $A,B,C$ in such a way
1086: that $AC=b_i$, $BC=b_{i-1}$ and $AB=c_i$ (see Figure
1087: \ref{eighttriangles}). Let $\gamma_i$ be the hyperbolic loxodromy of
1088: complex length $\ell_i$ along the common perpendicular to $B\infty$
1089: and $AC$, sending $B,\infty$ to $A,C$ (in that order).
1090:
1091: \begin{sublemma} Let $\ell=\ell_i=\rho+\theta\sqrt{-1}$ be the complex length of
1092: $\gamma_i$, with $\theta\in]-\pi,\pi]$. Then
1093: $\max\{\rho,|\theta|\}\leq\pi\sqrt{c_i/b_i}$.
1094: \llabel{thickness}\end{sublemma} \begin{proof} Up to a plane
1095: similarity, we may assume $A=1$ and $C=-1$. Also, for convenience, relabel the edges of $ABC$ by $a,b,c$. Let $L$ be the fixed
1096: line of $\gamma_i$. The hyperbolic isometry defined by $z\mapsto
1097: f(z)= \frac{2B+1-z}{z+1}$ exchanges the oriented lines $AC$ and
1098: $B\infty$, so it reverses the orientation of $L$ around the center
1099: of the tetrahedron $ABC\infty$. Therefore, $\gamma_i$ is given by
1100: $\gamma_i(z)= 2B-f(z)$. If $M$ is a matrix of $\gamma_i$, one
1101: has $\displaystyle{\frac{ \text{tr}^2 M}{4\det
1102: M}=\frac{\cosh \ell +1}{2}}$. Use
1103: $M=\matris{2B+1}{-1}{1}{1}$ to find $B=\cosh \ell$. Compute
1104: \begin{eqnarray*} \frac{a\pm c}{b}&=&\frac{|B+1|\pm
1105: |B-1|}{2}=\left|\cosh^2 \frac{\ell}{2}\right|\pm \left|\sinh^2
1106: \frac{\ell}{2}\right|
1107: \\&=&\textstyle{ \cosh \frac{\ell}{2}\cosh
1108: \frac{\overline{\ell}}{2}\pm\sinh \frac{\ell}{2} \sinh
1109: \frac{\overline{\ell}}{2}= \cosh\frac{\ell\pm\overline{\ell}}{2}}.
1110: \end{eqnarray*} Use $\cosh(i\theta)=\cos \theta$ to get $\cosh(\rho)=
1111: \frac{a+c}{b}$ and $\cos(\theta)=\frac{a-c}{b}$. The estimates
1112: $\text{Argcosh}(y)\leq 2 \sqrt{\frac{y-1}{2}}$ and $\text{Arccos}
1113: (y) \leq \pi \sqrt{\frac{1-y}{2}}$, since $\frac12 \left |\frac{a\pm
1114: c}{b}-1 \right |\leq
1115: \frac{c}{b}$, finally yield $\rho\leq 2\sqrt{c/b}$ and $|\theta|\leq
1116: \pi\sqrt{c/b}$, hence Sublemma \ref{thickness}.\end{proof}
1117:
1118: Since $(b_i)$ goes to $2$, Proposition
1119: \ref{helicoptere} will follow if the $c_i$ go to $0$ fast enough
1120: near $+\infty$ (with a similar argument near $-\infty$). Such fast
1121: decay is given by Proposition \ref{ecrase}: using the fact that
1122: $(\sigma_i)$ goes to $0$, we can bound the $c_i$ by decreasing
1123: geometric sequences on intervals of the form $\llbracket j, l-1
1124: \rrbracket$ where $j<l$ are consecutive elements of the set $J$.
1125: \end{proof}
1126:
1127: \subsection{Teichm\"uller charts}
1128:
1129: Let $\mathcal{T}_S$ be the Teichm\"uller space of the punctured
1130: torus: for each $i\in {\mathbb Z}$ we denote by $\mu(S_i)\in
1131: \mathcal{T}_S$ the intrinsic modulus of the marked surface $S_i$. In
1132: order to prove that the $\mu(S_i)$ converge in $\mathcal{T}_S$, let
1133: us first introduce appropriate charts for $\mathcal{T}_S$.
1134:
1135: Consider a topological ideal triangulation $E$ of the punctured torus $S$,
1136: with labeled edges $\epsilon_1,\epsilon_2,\epsilon_3$. Then $E$ defines an isomorphism
1137: $$h_E: {\mathbb P}^2 {\mathbb R}_+^* =
1138: (\mathbb{R}_+^*)^{\{\epsilon_1,\epsilon_2,\epsilon_3\}}/\mathbb{R}_+^*~\widetilde{\longrightarrow}~
1139: \mathcal{T}_S.$$ Namely, given a hyperbolic metric $g$ on $S$, in
1140: order to compute $h_E^{-1}(g)$, straighten $E$ to an ideal
1141: triangulation for $g$ and return the positive projective triple of
1142: Euclidean lengths defined (in the link of the puncture) by the
1143: sectors opposite $\epsilon_1,\epsilon_2,\epsilon_3$. We consider the $h_E$ as charts of $\mathcal{T}_S$. We endow $\mathbb{P}^2\mathbb{R}_+^*$ with the
1144: distance $d$ given by
1145: $$d([a:b:c],[a':b':c']):=\min_{\lambda>0}\max\left \{ \left | \log
1146: \frac{\lambda a}{a'}\right |, \left | \log \frac{\lambda
1147: b}{b'}\right |, \left | \log \frac{\lambda c}{c'}\right | \right
1148: \}.$$ Also define $\overline{h}_E:\mathbb{P}^2{\mathbb
1149: C}^*\rightarrow \mathcal{T}_S$ by
1150: $\overline{h}_E([a:b:c]):=h_E([\,|a|:|b|:|c|\,]).$
1151:
1152: %The map $(\mathbb{R}_+^*)^3\rightarrow \mathcal{T}_S$ induced by $h_E$
1153: %will be noted $H_E$.
1154:
1155: In particular, if the pleated punctured torus $S_{i+1}$, pleated
1156: along the ideal triangulation $E_{i+1}$, gives rise in the cusp link
1157: to a broken (oriented) line whose segments have complex coordinates
1158: $(a,b,c)$, then $$\mu(S_{i+1})=\overline{h}_{E_{i+1}}([a:b:c]).$$
1159:
1160: \begin{figure}[h!] \centering
1161: \psfrag{x}{$a$}
1162: \psfrag{y}{$b$}
1163: \psfrag{z}{$c$}
1164: \psfrag{xy}{$a+b$}
1165: \psfrag{Y}{$\displaystyle{\frac{b}{a+b}c}$}
1166: \psfrag{X}{$\displaystyle{\frac{a}{a+b}c}$}
1167: \psfrag{Si}{$S_i$}
1168: \psfrag{Sp}{$S_{i+1}$}
1169: \ledessin{chartmap}
1170: \caption{The complex projective triples associated to $S_{i+1}$ and $S_i$.\llabel{chartmap}}
1171: \end{figure}
1172:
1173: Moreover, Figure \ref{chartmap} shows \emph{two} broken lines corresponding
1174: respectively to $S_{i+1}$ and the previous pleated surface $S_i$.
1175: Since the triangles (links of vertices of $\Delta_i$) are similar,
1176: the complex coordinates of the segments forming the lower broken
1177: line are functions of $(a,b,c)$, as shown. Therefore, if we define
1178: the substitution formula
1179: $$\Psi([a:b:c]):=\left (\left [
1180: a+b~:~\frac{b}{a+b}c~:~\frac{a}{a+b}c \right ] \right )$$ (which is
1181: a birational isomorphism from $\mathbb{P}^2\mathbb{C}$ to itself),
1182: then $$\mu(S_{i})=\overline{h}_{E_{i}} (\Psi([a:b:c])).$$ Finally,
1183: by viewing the diagonal exchange between $E_i$ and $E_{i+1}$ as a
1184: \emph{flat} tetrahedron (as in Figure \ref{diagonalmove}), we see
1185: that
1186: $$\mu(S_{i+1})=h_{E_i}(\Psi([\,|a|:|b|:|c|\,])).$$ In other words, the
1187: restriction $\psi$ of $\Psi$ to $\mathbb{P}^2\mathbb{R}_+^*$ is the
1188: chart map $h_{E_{i+1}}\rightarrow h_{E_i}$, i.e.
1189: $$\begin{array}{rcl}
1190: \mathbb{P}^2\mathbb{R}_+^* & \overset{\psi}{\longleftarrow} & \mathbb{P}^2\mathbb{R}_+^* \\
1191: h_{E_i} \searrow && \swarrow h_{E_{i+1}} \\
1192: &\mathcal{T}_S& \end{array}$$ commutes.
1193:
1194: \begin{property} \llabel{skid} We have $\displaystyle{d\left ( h_{E_i}^{-1}\mu(S_i), h_{E_i}^{-1}\mu(S_{i+1}) \right ) = \log \frac{|a|+|b|}{|a+b|}}$.
1195: \end{property}
1196: \begin{proof} The right member is
1197: $$d\left(\left[ |a+b|:\left |\frac{b}{a+b}c \right |: \left |\frac{a}{a+b}c \right |\, \right],
1198: \left [|a|+|b|:\frac{|b|}{|a|+|b|}|c|:\frac{|a|}{|a|+|b|}|c|\right
1199: ]\right ).$$ \end{proof} It is easy to see that
1200: $\psi:\mathbb{P}^2\mathbb{R}_+^*\rightarrow
1201: \mathbb{P}^2\mathbb{R}_+^*$ is $3$-Lipschitz for $d$. In fact,
1202:
1203: \begin{proposition} There exists $K>0$ such that the $n$-th
1204: iterate $\psi^n$ is $Kn$-bilipschitz for all $n>0$. \llabel{magique}
1205: \end{proposition} \begin{proof} Since
1206: $\psi([a:b:c])=\left [\frac{(a+b)^2}{c}:b:a\right ]$ and
1207: $\psi^{-1}([a:b:c])=\left [c:b:\frac{(c+b)^2}{a}\right ]$, it is
1208: enough to check the Lipschitz statement. Set
1209: $A:=\sqrt{a}~;~C:=\sqrt{c}~$;
1210: $$P_n:=\frac{A^{n+1}}{C^n}+\sum_{i,j\in\mathbb{Z}}\binomial{j}{i-1}
1211: \binomial{n-i}{j-i} A^{2i-n-1} C^{n-2j}~~\text{ for all $n\geq
1212: -1$}$$ (the sum is really on $0<i\leq j \leq n$), so that
1213: $P_0=A~,~P_{-1}=C~,~P_1= \frac{A^2+1}{C}$. We claim that
1214: $\psi^n[a:1:c]=[P_n^2:1:P_{n-1}^2]$. The claim is seen by induction
1215: on $n$: the only difficult thing is the induction step
1216: $P_{n+1}^2=\frac{(P_n^2+1)^2}{P_{n-1}^2}$. First, it is
1217: straightforward to check that
1218: $P_{n+1}+P_{n-1}=(\frac{A}{C}+\frac{C}{A}+\frac{1}{AC})P_n$ (using
1219: the Pascal relation twice). Hence, for all $n\geq 1$, one has
1220: \begin{eqnarray*} &&
1221: (P_{n+1}P_{n-1}-P_n^2)-(P_n P_{n-2}-P_{n-1}^2)\\&=&
1222: P_{n-1}(P_{n+1}+P_{n-1})-P_n(P_n+P_{n-2})\\&=&(P_{n-1}P_n-P_n P_{n-1})
1223: \textstyle{ \left (\frac{A}{C}+\frac{C}{A}+\frac{1}{AC}\right )}=0.
1224: \end{eqnarray*} Therefore $P_{n+1}P_{n-1}-P_{n}^2=P_1 P_{-1}-P_0^2=1$, which
1225: proves the induction step. Since $P_n$ is a Laurent polynomial in
1226: $A,C$ with partial degrees of order $n$ and positive coefficients,
1227: we see that $\log P_n$ is $Ln$-bilipschitz in $\log a, \log c$ for
1228: some universal $L$. The Proposition follows.
1229: \end{proof}
1230: The proof of Proposition \ref{magique} may seem extremely \emph{ad
1231: hoc} and unsatisfactory. However, Proposition \ref{magique} is a
1232: special case of a more general phenomenon for \emph{Markoff maps}
1233: (in the sense of \cite{bowditch}), which we describe in
1234: \cite{abeilles}.
1235:
1236: \subsection{Convergence of the moduli}
1237:
1238: \begin{proposition} The moduli $(\mu(S_i))_{i\rightarrow \pm \infty}$ converge in Teichm\"uller space $\mathcal{T}_S$.
1239: \llabel{surfaceconvergence} \end{proposition}
1240:
1241: \begin{proof} To prove the $+\infty$-statement, we will fix a large
1242: enough index $i$, and prove that the series $$\eta_j:= d\left
1243: (h_{E_i}^{-1}\mu(S_j), h_{E_i}^{-1}\mu(S_{j+1}) \right ),$$ defined
1244: for $j> i$, has finite sum.
1245:
1246: For $j>i$, consider a downward-pointing triangle $\tau$ contributed
1247: by $\Delta_j$, with its edge lengths $b_j$, $b_{j-1}$ and $c_j$. The
1248: angles of $\tau$ at the ends of $b_j$ are $x_j$ and $y_j$. By
1249: Property \ref{skid}, we have $d\left
1250: (h_{E_j}^{-1}\mu(S_j),h_{E_j}^{-1}\mu(S_{j+1})\right )= \log
1251: \frac{c_j+b_{j-1}}{b_j}$. Compute \begin{eqnarray*}\log
1252: \frac{c_j+b_{j-1}}{b_j}&\leq& \frac{c_j+b_{j-1}-b_j}{b_j}=\frac{\sin
1253: x_j+\sin y_j-\sin z_j}{\sin z_j}=\frac{2\sin \frac{x_j}{2} \sin
1254: \frac{y_j}{2}}{\cos\frac{x_j+y_j}{2}}\\
1255: &\leq& \frac{\sin x_j \sin y_j}{\sin^2 z_j} \sin^2 z_j = \frac{c_j
1256: b_{j-1}}{b_j^2} \sin^2 w_j \leq \frac{c_j}{b_j} {\phi^+_j}^2
1257: \end{eqnarray*} (the inequality at the start of the second line
1258: holds for large enough $j$ because $x_j+y_j=2w_j\rightarrow 0$).
1259: Define $\delta_j=\frac{c_j}{b_j}{\phi_j^+}^2$, and let $M_j$ denote
1260: the best bilipschitz constant for the chart map $h_{E_j}\rightarrow
1261: h_{E_i}$. Observe that $\eta_j\leq M_j \delta_j$.
1262:
1263: Let $j<l$ be two consecutive elements of $J$ (see Definition
1264: \ref{definitionJ}), and $k\notin J$ an integer. We shall bound the
1265: $M_n\delta_n$ by geometric sequences on intervals of the form
1266: $\llbracket j,l-1\rrbracket$, using Proposition \ref{ecrase} as in
1267: the proof of Proposition \ref{helicoptere}. Since $\psi$ is
1268: $3$-bilipschitz, one clearly has $M_k\leq 3 M_{k-1}$. By Proposition
1269: \ref{ecrase},
1270: $$\frac{M_k \delta_k}{M_{k-1}\delta_{k-1}} \leq 3
1271: \frac{\delta_k}{\delta_{k-1}} \leq 3 \frac{c_k}{c_{k-1}} =
1272: 3\sigma_{k-1}$$ because $(\phi^+_k)$ and $(1/b_k)$ are decreasing.
1273: The right member goes to $0$ for large $k$.
1274:
1275: By Sublemma \ref{magique}, $M_{l}\leq (l-j)L\cdot M_j$ for some
1276: universal $L$, and by Lemma \ref{studyphi} ({\sf i-iii-iv}),
1277: $\phi^+_{j}=\phi^+_{l-1}+(l-j-1)\phi^+_l\geq(l-j)\phi^+_{l}$. Using
1278: Proposition \ref{ecrase}, it follows that
1279: $$\frac{M_l\delta_l}{M_j\delta_j}\leq (l-j)L \cdot \frac{c_l}{c_j}
1280: {\left ( \frac{\phi^+_l}{\phi^+_j} \right )}^2 \leq
1281: \frac{L\sigma_{j-2}}{l-j}\leq L\sigma_{j-2}.$$ The right member goes
1282: to $0$ as $j$ goes to infinity, hence Proposition
1283: \ref{surfaceconvergence}.
1284: \end{proof}
1285:
1286: \section{Extrinsic convergence of the surfaces $S_i$}
1287: \llabel{sectionextrinsicconvergence}
1288:
1289: \subsection{Pleated surfaces} \llabel{definepleating} Propositions
1290: \ref{helicoptere} and \ref{surfaceconvergence}, together with Lemma
1291: \ref{bendingbounded}, are the key ingredients to prove that the
1292: metric completion of $V=\bigcup_{i\in \mathbb{Z}}\Delta_i$ has two
1293: boundary components which are pleated punctured tori with pleating
1294: measure $\lambda^{\pm}$. A pleated surface is by definition (up to taking a universal cover) a map
1295: $\varphi:\mathbb{H}^2\rightarrow \mathbb{H}^3$ which sends
1296: rectifiable arcs to rectifiable arcs of the same length, such that
1297: through each point $p$ of $\mathbb{H}^2$ runs an open segment $s_p$
1298: on which $\varphi$ is totally geodesic. It is known (see \cite{lms},
1299: 5.1.4) that the direction of $s_p$ is unique if and only if $p$
1300: belongs to a certain \emph{geodesic lamination} $\Lambda$ (closed
1301: union of disjoint geodesics), and that $\varphi$ is totally geodesic
1302: away from $\Lambda$.
1303:
1304: To wrestle with pleated surfaces, we will use the fact that if
1305: $\varphi$ is a locally convex immersion, and $\Lambda$ has zero
1306: Lebesgue measure, then $\Lambda$ comes with a transverse (pleating)
1307: measure $\nu_{\Lambda}$. More precisely, $\nu_{\Lambda}$ can be
1308: defined on any segment $s$ transverse to $\Lambda$ in the following
1309: way (see Sections 7 to 9 of \cite{bonahon}). Immerse $\varphi(\mathbb{H}^2)$ into the Poincar\'e upper half-space model. Each component of $\mathbb{H}^2\smallsetminus \Lambda$ crossed by $s$ can be extended to a subset $A$ of
1310: $\mathbb{H}^2$ bounded by only one or two lines of $\Lambda$ crossed by
1311: $s$. Endow $s$ with a transverse orientation. The boundary component
1312: of $A$ in $\partial \mathbb{H}^3=\mathbb{C}\cup\{\infty\}$ on the
1313: positive side of the transverse orientation defines a circle arc
1314: $c_A$, of angle $\theta_A \in (-2\pi,2\pi)$ (we may assume
1315: $\infty\notin c_A$). The closure of the union of all the $c_A$ forms
1316: a rectifiable curve $c$, of length $\sum_A\text{\sf length}(c_A)$.
1317: Then $c$ has a well-defined \emph{regular curvature} $RC(c)$,
1318: defined as the absolutely convergent sum $\sum_A\theta_A$. But $c$
1319: also has a \emph{total curvature} $TC(c)$, defined (up to an
1320: appropriate multiple of $2\pi$) as the difference between the
1321: arguments of the initial and final tangent vectors to $c$. (The appropriate multiple of $2\pi$ can be determined by closing off the \emph{embedded} arc $c$ with a broken line, and requiring that the resulting Jordan curve have total curvature $2\pi$). Then, $\nu_{\Lambda}(s)$ is defined as the \emph{singular curvature} $SC(c)=TC(c)-RC(c)$.
1322:
1323: Conversely, if $\varphi:\mathbb{H}^2\rightarrow \mathbb{H}^3$ is a
1324: pleated immersion and $SC(c)$ is well-defined and non-negative for
1325: all transverse segments $s$, then $\varphi$ is locally convex with
1326: pleating measure $\nu_{\Lambda}$ as above. We refer to \cite{bonahon} for greater detail.
1327: % Bonahon, aussi ! (Annales de Toulouse)
1328:
1329: \subsection{Setup} Consider the marked once-punctured torus
1330: $S_{+\infty}$, endowed with the hyperbolic metric $\lim_{i\rightarrow
1331: +\infty}\mu(S_i)$. There exists a unique compact geodesic lamination
1332: $\Lambda_c^+$ of slope $\beta^+/\alpha^+$ on $S_{+\infty}$. Exactly
1333: two lines $\ell,\ell'$ issued from the puncture of $S_{+\infty}$ fail
1334: to meet $\Lambda_c^+$: define $\Lambda^+=\Lambda_c^+\sqcup \ell
1335: \sqcup \ell'$. Then $S_{+\infty}$ is the disjoint union of
1336: $\Lambda^+$ and the interiors of two ideal triangles $A,A'$. The
1337: union $A\cup A'\cup \ell \cup \ell'$ is a punctured ideal bigon.
1338:
1339: Recall the ideal tetrahedra $\Delta_i$, the space $V=\bigcup_{i\in \mathbb{Z}} \Delta_i$ and the
1340: hyperelliptic involution $h:V\rightarrow V$ which reverses all edge
1341: orientations. The slopes of the (oriented) pleating lines of the
1342: surface $S_i$ (between $\Delta_{i-1}$ and $\Delta_i$) are
1343: elements of ${\mathbb P}^1{\mathbb Q}$ projecting to $0,1,\infty$ in
1344: ${\mathbb P}^1({\mathbb Z}/2)$: accordingly, we call these pleating lines $l^0_i, l^1_i, l^{\infty}_i$. For $*\in\{0,1,\infty\}$, denote by
1345: $\omega^*_i$ the unique point of $l^*_i$ fixed under the
1346: hyperelliptic involution (the $\omega^*_i$ are called Weierstrass
1347: points). Let $s^*_i$ be the segment from $\omega^*_i$ to
1348: $\omega^*_{i+1}$ (across the tetrahedraon $\Delta_i$ if
1349: $\omega^*_i\neq\omega^*_{i+1}$): each $s^*_i$ is contained in a
1350: (pointwise) fixed line $\Omega^*$ of the hyperelliptic involution.
1351:
1352: Fix the value of the superscript $*$ (soon we shall omit it). Fix a
1353: point $\omega$ in $\mathbb{H}^2$ and an oriented line $l$ through
1354: $\omega$. For each (oriented) surface $S_i$, consider the oriented, marked
1355: universal covering $\pi_i:(\mathbb{H}^2,\omega,l)\rightarrow
1356: (S_i,\omega^*_i,l^*_i)=(S_i,\omega_i,l_i)$. Endow a universal
1357: covering $\widetilde{V}$ of $V$ with lifts $\widetilde{\omega_i}$ of
1358: the $\omega_i$ connected by lifts of the segments $s_i$, and fix a
1359: developing map $\Phi:\widetilde{V}\rightarrow \mathbb{H}^3$. There
1360: is a unique map $h_i$ such that
1361: \begin{equation} \llabel{comdiagram} \begin{array}{ccccc}
1362: (\mathbb{H}^2,\omega,l)&\overset{h_i}{\longrightarrow}&(\widetilde{V},\widetilde{\omega_i})
1363: & \overset{\Phi}{\longrightarrow}&\mathbb{H}^3 \\ \pi_i \downarrow
1364: ~~ && \downarrow &&\\ S_i & \longrightarrow & V && \end{array}
1365: \end{equation}
1366: commutes. We will prove that the (developing) pleated immersions
1367: \begin{equation}\varphi_i=\Phi \circ h_i : \mathbb{H}^2 \rightarrow
1368: \mathbb{H}^3 \llabel{immersion}\end{equation} converge as pleated
1369: maps.
1370:
1371: \subsection{Convergence of the $\varphi_i$} \llabel{hausdorff}
1372: By Proposition \ref{helicoptere}, it is already clear that the
1373: restriction of $\varphi_i$ to $l$ converges to a totally geodesic
1374: embedding of the line $l$ into $\mathbb{H}^3$ (the convergence is
1375: uniform on all compacts of $l$). By Ascoli's theorem, since the
1376: $\varphi_i$ are $1$-Lipschitz, there exists an increasing sequence
1377: $\nu$ such that the $\varphi_{\nu(i)}$ converge to a certain map
1378: $\varphi_{+\infty}$, uniformly on all compact sets of $\mathbb{H}^2$.
1379: In Section \ref{pleatingphiinf} below, $\varphi_{+\infty}$ is shown to be independent of the subsequence $\nu$: in anticipation, we now abusively write $\varphi_i$ instead of $\varphi_{\nu(i)}$.
1380:
1381: We shall work in the projective tangent bundle
1382: $\mathcal{E}=\mathbb{P}T\mathbb{H}^2$, a circle bundle over
1383: $\mathbb{H}^2$ in which geodesic laminations naturally live as
1384: closed sets. Length and angle measurements define a (canonical)
1385: complete Riemannian metric on $\mathcal{E}$. For $K\subset
1386: \mathbb{H}^2$ compact and $A,B\subset \mathcal{E}$ closed, let
1387: $K^{\mathcal E}\subset \mathcal{E}$ be the preimage of $K$ under the
1388: natural projection $\mathcal{E}\rightarrow \mathbb{H}^2$, and define
1389: $$d_K(A,B):=\inf \left \{ \delta>0 ~|~
1390: A\cap K^{\mathcal E} \subset B+\delta ~,~ B\cap K^{\mathcal E}
1391: \subset A+\delta \right \}$$ where $X+\delta$ denotes the set of
1392: points within $\delta$ of $X$. Then $\inf(1,d_K)$ is a pseudometric,
1393: and the set of closed subsets of $\mathcal{E}$ is compact for the
1394: Hausdorff metric
1395: $$d_H=\sum_{n>0}2^{-n}\inf(1,d_{K_n})$$ where the $K_n$ are concentric balls
1396: of radius $n$.
1397:
1398: \medskip
1399: Observe that $S_{+\infty}$ also has Weierstrass points
1400: $\omega^*_{+\infty}$, belonging to leaves $l^*_{+\infty}$ of the
1401: lamination $\Lambda^+$. Fixing the value of $*$ as before, denote by
1402: $\pi_{+\infty}:(\mathbb{H}^2,\omega,l) \rightarrow (S_{+\infty},
1403: \omega^*_{+\infty}, l^*_{+\infty})$ an oriented universal cover.
1404:
1405: A consequence of Proposition \ref{surfaceconvergence} is that the
1406: lifts to $\mathcal{E}$ of the $\pi_i^{-1}(l^0_i\cup l^1_i \cup
1407: l^{\infty}_i)$ converge for $d_H$ to the lift of
1408: $\pi_{+\infty}^{-1}(\Lambda^+)$. We know that
1409: $U=\pi_{+\infty}^{-1}(S_{+\infty}\smallsetminus \Lambda^+)\subset
1410: \mathbb{H}^2$ is a disjoint union of (open) ideal triangles, of full
1411: Lebesgue measure in $\mathbb{H}^2$. For any connected compact set
1412: $K\subset U$, we have $K\cap \pi_i^{-1}(l^0_i\cup l^1_i \cup
1413: l^{\infty}_i)=\emptyset$ for $i$ large enough, so $\varphi_{+\infty}$
1414: is totally geodesic on $K$. Therefore, $\varphi_{+\infty}$ is totally
1415: geodesic on each component of $U$. Since $\varphi_{+\infty}$ is
1416: clearly $1$-Lipschitz, we can approximate any segment in
1417: $\mathbb{H}^2\smallsetminus U$ by segments in $U$ to show that
1418: $\varphi_{+\infty}$ is totally geodesic on each leaf of
1419: $\mathbb{H}^2\smallsetminus U$. By Lemma 5.2.8 in \cite{lms},
1420: $\varphi_{+\infty}$ sends rectifiable segments to rectifiable
1421: segments of the same length, and is a pleated map.
1422:
1423: \subsection{The map $\varphi_{+\infty}$ is a topological immersion} \llabel{patches}
1424: Define $\mathcal{P}:=\pi_{+\infty}^{-1}(\Lambda^+)$, which contains
1425: the pleating locus of $\varphi_{+\infty}$. To prove
1426: $\varphi_{+\infty}$ is an immersion, it is enough to find a short
1427: geodesic segment $m$ of $\mathbb{H}^2$, through the base point $\omega$, transverse to $\mathcal{P}$, and prove that $\varphi_{+\infty}$ is an immersion
1428: on the union $\Upsilon$ of all strata (lines and complementary ideal triangles) of $\mathcal{P}$ crossed by $m$ (indeed,
1429: $\pi_{+\infty}(\Upsilon)=S_{+\infty}$). Clearly, $\varphi_{+\infty}$ is already an immersion near any point of $\mathbb{H}^2\smallsetminus \mathcal{P}$. At other points, the key fact will be an ``equidistribution'' property of the $3$ pleating lines of the surface $S_i$, as $i$ goes to $+\infty$.
1430:
1431: Choose a small $\mu_1>0$, and pick $k\in \mathbb{Z}$ large enough so
1432: that $\phi_k^+\leq\mu_1$. Let $q$ be the rational opposite the Farey
1433: edge $e_k$, on the same side as $\beta^-/\alpha^-$, so that
1434: $\lambda^+(q)=\phi^+_k$. Let $m^{\mathcal{C}}_i$ be the simple
1435: closed geodesic of slope $q$ in $S_i$ (equipped with the intrinsic
1436: metric): for some superscript $*$ independent of $i$, the
1437: Weierstrass point $\omega_i=\omega^*_i$ belongs to
1438: $m^{\mathcal{C}}_i$. By Proposition \ref{surfaceconvergence} and
1439: Hausdorff convergence, there exists $\mu_2>0$ such that for all
1440: $i>k$, the angle between $m^{\mathcal{C}}_i$ and the pleating line
1441: $l_i$ of $S_i$ at $\omega_i$ is at least $\mu_2$, and there exists
1442: $\mu_3>0$ such that the segment $m_i$ of $m^{\mathcal{C}}_i$ of
1443: length $2\mu_3$, centered at $\omega_i$, is embedded in $S_i$. The
1444: ends of $m_i$ are at distance at least $\frac12 \mu_2\mu_3$ from the
1445: pleating line $l_i$ in $S_i$. Observe that the simple closed curve
1446: $m^{\mathcal{C}}_i$ meets the pleating edges of $S_i$ in a perfectly
1447: equidistributed (Sturmian) order (as would a straight line on a Euclidean grid): therefore, the algebraic sum of the pleating angles crossed by any given subsegment of $m_i$ does not exceed $2\mu_1$. Finally, let $\kappa_i$ be a subsegment of the pleating line $l_i$, centered at $\omega_i$, of length $2$: if
1448: $\mu_3$ is small enough, any pleating line $L$ met by $m_i$ makes an
1449: angle at least $\frac12 \mu_2$ with $m_i$, and comes within $3\mu_3$
1450: of both ends of $\kappa_i$ for the intrinsic metric of $S_i$.
1451:
1452: \medskip
1453: Arrange the developing map $\Phi:\widetilde{V}\rightarrow
1454: \mathbb{H}^3$ in the upper half-space model so that the
1455: $\varphi_i(\omega)$ lie on the line $0\infty$ at heights less than
1456: $1$, and $\varphi_{+\infty}(l)$ is the oriented line from $-1$ to
1457: $1$. Consider lifts $\widetilde{m}_i$ of the arcs $m_i$ through the
1458: $\varphi_i(\omega)$. By the above (considering lifts of the
1459: $\kappa_i$), if the $\mu$'s are small enough, any pleating line of
1460: $\varphi_i(\mathbb{H}^2)$ met by $\widetilde{m}_i$ has its endpoints
1461: within distance $1/2$ from $1$ and $-1$ in $\mathbb{C}$ (recall
1462: $\varphi_i$ is $1$-Lipschitz). Following Subsection
1463: \ref{definepleating}, let $c^{+1}_i$ (resp. $c^{-1}_i$) denote the
1464: piecewise smooth curve defined by the transverse segment
1465: $\widetilde{m}_i$ of $\varphi_i(\mathbb{H}^2)$ near $1$ (resp.
1466: $-1$).
1467:
1468: Let $\tau$ be a subsegment of $\widetilde{m}_i$ across an ideal
1469: triangle of $\varphi_i(\mathbb{H}^2)$. Let $\tau'$ be the circle arc
1470: contributed by $\tau$ to $c^{\pm 1}_i$. By the above, for some
1471: universal $K_1>0$,
1472: $$ \frac{\mu_2}{K_1} \leq \frac{\text{euclidean length
1473: of }\tau'}{\text{hyperbolic length of }\tau} \leq K_1.$$ In
1474: particular, the $c^{\pm 1}_i$ have length at most $2K_1\mu_3$. But
1475: the regular curvature radii of $c^{\pm 1}_i$ are at least $\frac12$
1476: (the corresponding circles come near $1$ and $-1$), so the total
1477: regular curvature of $c^{\pm 1}_i$ is at most $4K_1\mu_3$. By the
1478: above, the total singular curvature on any subinterval of $c^{\pm
1479: 1}_i$ is at most $2\mu_1$. If the $\mu$'s are small enough, it
1480: follows that all tangent vectors of $c^{\pm 1}_i$ have complex
1481: arguments within $[\pi/4, 3\pi/4]$. As a consequence, if $\tau$
1482: ranges over the ideal triangles of $\varphi_i(\mathbb{H}^2)$ crossed
1483: by $\widetilde{m}_i$, and $p:\mathbb{H}^3\rightarrow \mathbb{C}$ is
1484: the vertical projection, the different $p(\tau)$ intersect only
1485: along their edges: so $\varphi_i(\bigcup_{\tau}\tau)$ is an embedded
1486: surface (which we can see, say, as the graph of a function from an
1487: open set of $\mathbb{C}$ to $\mathbb{R}^+$). Moreover, define the
1488: \emph{breadth} of $p(\tau)$ as the length of the segment
1489: $p(\tau)\cap \sqrt{-1}\mathbb{R}$ (of the imaginary axis). Then, for some
1490: universal $K_2>0$,
1491: $$\frac{\text{breadth of }p(\tau)}{\text{hyperbolic length of }
1492: \tau\cap \widetilde{m}_i} \geq \frac{\mu_2}{K_2}.$$
1493:
1494: To conclude concerning $\varphi_{+\infty}$, define a lift
1495: $m^{\mathcal{C}}$ through $\omega \in \mathbb{H}^2$ of the simple
1496: closed geodesic of slope $q$ in $S_{+\infty}$, and a subsegment $m$
1497: of $m^{\mathcal{C}}$, of length $2\mu_3$, centered at $\omega$. The
1498: angle between $m$ and any pleating line of $\Lambda^+$ it encounters
1499: is at least $\frac12 \mu_2$. Let $I$ be the (infinite) collection of
1500: ideal triangles of $\pi_{+\infty}^{-1}(S_{+\infty}\smallsetminus
1501: \Lambda^+)$ crossed by $m$. For $\tau\in I$, denote by $|\tau|$ the
1502: length of $m\cap \tau$. By convergence in the Hausdorff metric,
1503: $\varphi_{+\infty}(\tau)$ is approached by triangles of the
1504: $\varphi_i(\mathbb{H}^2)$, of breadth at least
1505: $\frac{|\tau|\mu_2}{2K_2}$: so $p(\varphi_{+\infty}(\tau))$ has nonzero
1506: breadth. Injectivity follows: if $x,x' \in \overline{\bigcup_{\tau\in
1507: I}\tau}$ do not belong to the same stratum of the lamination $\Lambda^+$, find $\tau$ separating
1508: $x$ from $x'$ to prove that $\varphi_{+\infty}(x)\neq
1509: \varphi_{+\infty}(x')$. By vertical projection to $\mathbb{C}$, we see
1510: that $\varphi_{+\infty}(\mathbb{H}^2)$ is topologically immersed in
1511: $\mathbb{H}^3$.
1512:
1513: \subsection{Pleating measure of $\varphi_{+\infty}$} \llabel{pleatingphiinf}These arguments can be extended to prove that the pleating measure of
1514: $\varphi_{+\infty}$, as defined in \ref{definepleating}, is the limit
1515: of the pleating measures of the $\varphi_i$: the rectifiable
1516: curves $c^{\pm 1}_{+\infty}$ defined near $\pm 1$ by
1517: $\varphi_{+\infty}$ have lengths $\ell^{\pm 1}$, and for any
1518: $\varepsilon >0$, there exists a finite disjoint union of circle
1519: arcs $\gamma_s$ in $c^{\pm 1}_{+\infty}$ whose lengths add up to at
1520: least $\ell^{\pm 1}-\varepsilon$ (moreover the direction of $c^{\pm 1}_{+\infty}$, like that of $c^{\pm 1}_i$, is everywhere within $\pi/4$ of the vertical axis). The $\gamma_s$ can be approached by (unions of) arcs of the $c^{\pm 1}_i$, and the regular curvature \emph{not} contributed by the $\gamma_s$ is bounded by $3\varepsilon$. It follows that the pleating of $\varphi_{+\infty}$ is the limit of the pleatings of the $\varphi_i$ (on any transverse arc, and therefore, on any simple closed curve): that pleating is simply $\lambda^+$. In particular, $\varphi_{+\infty}=\lim \varphi_{\nu(i)}$ is independent of the original subsequence $\nu$, and the $\varphi_i$ converge to a pleated map whose pleating is given by $\lambda^{\pm}$, as $i$ goes to $\pm \infty$.
1521:
1522: \subsection{Completeness} The construction of \ref{patches} further
1523: allows us to embed the universal cover $\widetilde{V}$ of
1524: $V=\bigcup_{i\in \mathbb{Z}}\Delta_i$ into a (topological) manifold
1525: with boundary $\widetilde{V}_{\partial}$, as follows. For each $x\in
1526: \mathbb{H}^2$, consider a neighborhood $U_x$ of $x$ such that
1527: $\varphi_{+\infty}$ is an embedding on $U_x$. Then
1528: $\varphi_{+\infty}(U_x)$, which has a well-defined transverse
1529: (``outward'') orientation, splits a small ball $B_x$ centered at
1530: $\varphi_{+\infty}(x)$ into two (topological) hemispheres, which we
1531: can call ``inner'' and ``outer'', referring to the transverse
1532: orientation. The inner hemispheres $H_x$, for $x$ ranging over
1533: $\mathbb{H}^2$, can be patched together to obtain a manifold with
1534: boundary $H$. Without loss of generality, the balls $B_x$ can be
1535: chosen small enough so that, by the construction of Section
1536: \ref{patches}, each $H_x\smallsetminus \varphi_{+\infty}(U_x)$ is
1537: identified with a subset of $\widetilde{V}$, embedded in
1538: $\mathbb{H}^3$. Then, $H$ can further be patched to $\widetilde{V}$.
1539: Since each $H_x$ is homeomorphic to
1540: $\mathbb{R}^2\times\mathbb{R}^+$, the space
1541: $\widetilde{V}_{\partial}=\widetilde{V}\cup H$ is a (possibly
1542: non-complete) topological manifold with boundary.
1543:
1544: \begin{proposition} \llabel{properlydiscontinuous}
1545: The action of the fundamental group $\Gamma$ of the punctured torus $S$ on
1546: $\widetilde{V}$ extends to a properly discontinuous action on
1547: $\widetilde{V}_{\partial}$.
1548: \end{proposition}
1549: \begin{proof}
1550: Consider the representation $\rho:\Gamma\rightarrow
1551: \text{Isom}^+(\mathbb{H}^3)$ given by the developing map $\Phi$ from
1552: (\ref{comdiagram}), and the representations
1553: $\rho_n:\Gamma\rightarrow \text{Isom}^+(\mathbb{H}^2)$ which satisfy
1554: $\varphi_n \circ \rho_n(g)=\rho(g) \circ \varphi_n$ for all $g\in
1555: \Gamma$. By Lemma \ref{surfaceconvergence}, the $\rho_n$ converge to
1556: some $\rho_{+\infty}$. Convergence of the $\varphi_n$ immediately
1557: implies $\varphi_{+\infty} \circ \rho_{+\infty}(g)=\rho(g) \circ
1558: \varphi_{+\infty}$. Therefore, the hemispheres $H_x$ can be chosen in
1559: an equivariant fashion, and the action of $\Gamma$ on
1560: $\widetilde{V}_{\partial}$ is well-defined. The action is already
1561: properly discontinuous at every point $x$ of $\widetilde{V}$ (namely, as $g$ ranges over $\Gamma$, the $gx$ do not accumulate at $x$). But if
1562: $\Gamma$ acts without fixed points and by isometries on a locally compact metric space $X$, the set of $x\in X$ such that the action is properly
1563: discontinuous at $x$ is open (obviously) and closed: if $g_n x
1564: \rightarrow x$ for some sequence $(g_n)$ of $\Gamma \smallsetminus
1565: \{1\}$ and $U$ is a compact neighborhood of $x$, then $U$ contains a
1566: ball of radius $\varepsilon$ centered at $x$, and whenever
1567: $d(x,x')\leq \varepsilon/2$, the $g_n x'$ accumulate at some point
1568: of $U$, so the action is not totally discontinuous at $x'$. The
1569: Proposition follows by connectedness of $\widetilde{V}_{\partial}$.
1570: \end{proof}
1571: As a consequence, the space
1572: $V_{\partial}:=\widetilde{V}_{\partial}/\Gamma$ is a (topological)
1573: manifold with boundary, containing $V=\bigcup_{i\in
1574: \mathbb{Z}}\Delta_i$; and $\partial V_{\partial}$ consists of two
1575: pleated punctured tori (intrinsically isometric to $S_{+\infty}$ and
1576: another surface $S_{-\infty}$), with pleatings $\lambda^+$ and
1577: $\lambda^-$.
1578:
1579: \begin{proposition}
1580: The manifold with boundary $V_{\partial}$ is complete.
1581: \end{proposition}
1582: \begin{proof}
1583: Consider the metric completion $\overline{V} \supset V_{\partial}$
1584: and assume the inclusion is strict. Define a continuous function
1585: $f:V_{\partial}\rightarrow \mathbb{R}^{>0}$ by
1586: $f(x)=d(x,\overline{V} \smallsetminus V_{\partial})$. By assumption,
1587: $\inf (f)=0$.
1588:
1589: Consider the immersion of $\widetilde{V}_{\partial}$ into the upper
1590: half-space model of $\mathbb{H}^3$ obtained by sending a lift of the
1591: cusp to infinity (Figures \ref{frise} and
1592: \ref{fourtriangles}-\ref{eighttriangles}). The image of $\partial
1593: \widetilde{V}_{\partial}$ contains, in particular, vertical
1594: half-planes (interrupted at some height above $\mathbb{C}$).
1595: Therefore, any geodesic of $\widetilde{V}_{\partial}$ starting high
1596: enough above $\mathbb{C}$ is defined for all times $t\leq 1$ (unless
1597: it hits $\partial\widetilde{V}_{\partial}$). As a result, if
1598: $H\subset V_{\partial}$ denotes a small enough open horoball
1599: neighborhood of the cusp, we have $f\geq 1$ on $H$.
1600:
1601: For $i\in \mathbb{Z}$, consider the compact set
1602: $K_i:=S_i\smallsetminus H$ in $V_{\partial}$. There exists a ball
1603: $B$ of $\mathbb{H}^2$ centered at the base point $\omega$, with radius independent
1604: of $i$, such that $K_i\subset\pi_i(B)$: by convergence of $(\varphi_i)$,
1605: the $\pi_i(B)$ converge metrically to a compact subset $K'$ of
1606: $\partial V_{\partial}$, on which $f$ is positive. Therefore $f$ is bounded away from $0$ one some neighborhood $U$ of $K'$ in $V_{\partial}$, and $K_i\subset U$ for large enough $i$. So $f$ is bounded away from $0$ on $\bigcup_{i\in\mathbb{Z}} K_i$, and therefore on
1607: $\bigcup_{i\in\mathbb{Z}} S_i$.
1608:
1609: However, assume $\gamma(t)$ is a rectifiable $1$-Lipschitz arc of
1610: $V_{\partial}$, defined for $t<M$, with no limit at $M$. For any
1611: $\varepsilon>0$, the restriction $\gamma_{|[M-\varepsilon, M)}$
1612: meets $V_{\partial}\smallsetminus
1613: \partial V_{\partial}$ (because $\partial
1614: V_{\partial}=S_{+\infty}\sqcup S_{-\infty}$ is complete), but then
1615: $\gamma_{|[M-\varepsilon, M)}$ must meet $\Delta_i$ for an unbounded
1616: set of indices $i$ (any finite union of tetrahedra is complete).
1617: Therefore, we can find a sequence $t_n\rightarrow M$ such that
1618: $\gamma(t_n)\in \bigcup_{i\in \mathbb{Z}} S_i$. Clearly,
1619: $f(\gamma(t_n))\leq M-t_n$ which goes to $0$: a contradiction. So
1620: $V_{\partial}$ is complete.
1621: \end{proof}
1622:
1623: \subsection{A quasifuchsian punctured-torus group}
1624: The end of the argument is now quite standard: recall the complete
1625: manifold with locally convex boundary $\widetilde{V}_{\partial}$,
1626: which is a universal cover of $\overline{V}=V_{\partial}$. Given
1627: distinct points $x,x'\in \widetilde{V}_{\partial}$, consider a
1628: shortest possible path $\gamma$ from $x$ to $x'$. If $\gamma$ has an
1629: interior point in $\partial\widetilde{V}_{\partial}$, by local
1630: convexity, we must have $\gamma\subset \partial
1631: \widetilde{V}_{\partial}$, and $\gamma$ is a geodesic segment of
1632: $\partial\widetilde{V}_{\partial}$. If not, $\gamma$ is (the closure
1633: of) a geodesic segment of $\widetilde{V}$. At any rate, the extended
1634: developing map $\Phi:\widetilde{V}_{\partial}\rightarrow
1635: \mathbb{H}^3$ is an embedding (it sends $\gamma$ to a segment with
1636: distinct endpoints) and has a closed convex image $C$, endowed with
1637: a properly discontinuous action of the fundamental group $\Gamma$ of
1638: $S$ (Proposition \ref{properlydiscontinuous}). The action extends
1639: properly discontinuously to $\mathbb{H}^3$ (this can be seen by
1640: projecting any point of $\mathbb{H}^3$ to $C$). The manifold
1641: $\mathbb{H}^3/\Gamma$ contains $\overline{V}\simeq C/\Gamma$, which has
1642: the desired boundary pleatings $\lambda^{\pm}$. Clearly, $C$ is the
1643: smallest closed, convex set containing all parabolic fixed points of
1644: $\Gamma$; therefore $\overline{V}$ is the convex core, and $\Gamma$
1645: is quasifuchsian, with pleating data $\lambda^{\pm}$. Theorem \ref{pureexistence} is proved.
1646:
1647:
1648: \section{Application: the EPH theorem
1649: %conjecture of Akiyoshi, Sakuma, Wada and Yamashita
1650: }\llabel{sectioneph}
1651:
1652: Let us quickly recall the correspondence between the horoballs of
1653: ${\mathbb H}^3$ and the vectors in the positive light cone of
1654: Minkowski space. Endow ${\mathbb R}^4$ with the Lorentzian product
1655: $\left \langle (x,y,z,t) | (x',y',z',t') \right \rangle :=
1656: xx'+yy'+zz'-tt'$. Define $$X:=\{v=(x,y,z,t)\in{\mathbb R}^4 ~|~ t>0
1657: \text{ and } \langle v | v \rangle = -1 \}.$$ Then
1658: $\langle.|.\rangle$ restricts to a Riemannian metric on $X$ and
1659: there is an isometry $X \simeq {\mathbb H}^3$, with
1660: $\text{Isom}^+(X)$ a component of $SO_{3,1}({\mathbb R})$. We will
1661: identify the point $(x,y,z,t)$ of $X$ with the point at Euclidean
1662: height $\frac{1}{t+z}$ above the complex number $\frac{x+iy}{t+z}$
1663: in the Poincar\'e upper half-space model. Under this convention, the
1664: closed horoball $H_{d,\zeta}$ of Euclidean diameter $d$ centered at
1665: $\zeta=\xi+i\eta\in {\mathbb C}$ in the half-space model corresponds
1666: to $\{v\in X ~|~ \langle v | v_{d,\zeta}\rangle\geq -1\}$, where
1667: $v_{d,\zeta}= \frac{1}{d}(2\xi,2\eta,1-|\zeta|^2,1+|\zeta|^2)$. We
1668: therefore identify $H_{d,\zeta}$ with the point $v_{d,\zeta}$ of the
1669: isotropic cone. Similarly, the closed horoball $H_{h,\infty}$ of
1670: points at Euclidean height no less than $h$ in the half-space model
1671: corresponds to $\{v\in X ~|~ \langle v | v_{h,\infty} \rangle \geq
1672: -1\}$ where $v_{h,\infty}=(0,0,-h,h)$, so we identify $H_{h,\infty}$
1673: with $v_{h,\infty}$.
1674:
1675: Consider the following objects: a complete oriented hyperbolic $3$-manifold $M$ with one cusp, a horoball neighborhood $H$ of the cusp, the universal
1676: covering $\pi:{\mathbb H}^3\rightarrow M$, and the group $\Gamma
1677: \subset \text{Isom}^+({\mathbb H}^3)$ of deck transformations of
1678: $\pi$. Then $H$ lifts to a family of horoballs $(H_i)_{i\in I}$ in
1679: ${\mathbb H}^3$, corresponding to a family of isotropic vectors
1680: $(v_i)_{i\in I}$ in Minkowski space. The closed convex hull $C$ of
1681: $\{v_i\}_{i\in I}$ is $\Gamma$-invariant, and its boundary $\partial
1682: C$ comes with a natural decomposition
1683: $\widetilde{K}$ into polyhedral \emph{facets}. Let $U$ be the convex core of $M$, minus the
1684: pleating locus. In \cite{comparing}, Akiyoshi and Sakuma extended
1685: %\marginpar{Quote the actual result?}
1686: the Epstein-Penner convex hull construction to prove that
1687: $\widetilde{K}$ defines an $H$-independent decomposition $K$ of $U$ into ideal hyperbolic polyhedra, typically tetrahedra, allowing for a few clearly defined types of degeneracies. In \cite{aswy1}, with Wada and Yamashita, they also conjectured
1688:
1689: \begin{theorem} \llabel{eph}
1690: If $M$ is quasifuchsian, homeorphic to the product of the punctured
1691: torus with the real line, and has irrational pleating laminations of
1692: slopes $\beta^+/\alpha^+$ and $\beta^-/\alpha^-$, then the
1693: restriction of $K$ to the interior of the convex core of $M$ is
1694: combinatorially the triangulation $(\Delta_i)_{i\in \mathbb{Z}}$
1695: defined in Section \ref{sectionstrategy}.
1696: \end{theorem}
1697:
1698: \begin{proof}
1699: It is known (see \cite{series} or Corollary \ref{plth} below) that
1700: quasifuchsian punctured-torus groups are fully determined by their measured pleating laminations. It follows that the manifold constructed in Sections
1701: \ref{sectionstrategy}-\ref{sectionextrinsicconvergence}, with the
1702: same pleating data as $M$, is isometric to $M$. Recall the
1703: triangulated space $\widetilde{V}\subset{\mathbb H}^3$ of
1704: (\ref{comdiagram}), which is a universal cover of the union
1705: $\bigcup_{i\in{\mathbb Z}}\Delta_i$ of all tetrahedra $\Delta_i$.
1706: Given a horoball neighborhood $H$ of the cusp of $M$, consider its
1707: lifts $(H_i)_{i\in I}$ in ${\mathbb H}^3$ and the corresponding
1708: isotropic vectors $(v_i)_{i\in I}$ in Minkowski space ${\mathbb
1709: R}^4$. Then each tetrahedron $\Delta$ of $\widetilde{V}$ has its
1710: vertices at the centers of four horoballs $H_1,\dots,H_4$: we
1711: introduce the convex hull $\tau_{\Delta}$ of $v_1,\dots,v_4$, and
1712: define $D:=\bigcup_{\Delta} \tau_{\Delta}$. The central projection
1713: to the hyperboloid $X$ with respect to the origin sends
1714: $\tau_{\Delta}$ homeomorphically to the lift of $\Delta$ in $X$, so
1715: the interiors of the $\tau_{\Delta}$ are pairwise disjoint, and each
1716: $\tau_{\Delta}$ comes with a transverse orientation
1717: $\overrightarrow{u}$ (given by any ray through $\Delta$ issued from
1718: the origin). The theorem claims exactly that $D\subset\partial C$
1719: (the inclusion is expected to be strict, since $\partial C$ also
1720: contains faces projecting, say, to the boundary of the convex core
1721: --- they are analyzed in detail in \cite{comparing}).
1722:
1723: \medskip
1724:
1725: Suppose that $D$ is locally convex, with $\overrightarrow{u}$
1726: pointing inward. Define $I:=[1,+\infty)$ and, for any subset $Y$ of
1727: a vector space, $IY:=\{ty~|t\in I, y\in Y\}$. Since the interior of
1728: the convex core is convex, $ID$ is a convex set. Moreover, $ID$
1729: contains all the $v_i$, so its closure $\overline{ID}$ contains
1730: their convex hull $C$. Conversely, $D$ is clearly included in $C$
1731: and it is easy to check that $IC\subset C$ (because $I\{v_i\}\subset
1732: C$ for all $i$, see \cite{comparing}). So $ID \subset C$ and, by
1733: closedness, $\overline{ID}=C$. Clearly, $D\subset \partial
1734: \overline{ID}=\partial C$, as wished. So we only need to prove
1735:
1736: \begin{lemma} \llabel{minkonwex}
1737: The codimension-1 simplicial complex $D\subset {\mathbb R}^4$ is
1738: locally convex ($\overrightarrow{u}$ pointing inward).
1739: \end{lemma}
1740:
1741: \emph{Proof ---} Consider adjacent ideal tetrahedra $\Delta,
1742: \Delta'$ in ${\mathbb H}^3$ which are lifts from tetrahedra
1743: $\Delta_{i-1}, \Delta_i$ of the manifold. We must prove that the
1744: dihedral angle in $\mathbb{R}^4$ between $\tau_{\Delta}$ and $\tau_{\Delta'}$ points ``downward''. We will assume that the letter between $i-1$ and $i$
1745: is an $L$ belonging to a subword $RL^nR$ of $\Omega$. In the link of the cusp, the pleated surface $S_i$ between $\Delta_{i-1}$ and $\Delta_i$ contributes
1746: % appears as
1747: a broken line $(-1,\zeta,\zeta',1)$ in ${\mathbb C}$ together
1748: with its iterated images under $u\mapsto u\pm 2$, as in Figure
1749: \ref{minkowski} (we assume that the vertices $-1,1$ both belong to the
1750: base segments of the Euclidean triangles just below and just above
1751: the broken line, in the sense of Figure \ref{fourtriangles}). We use
1752: the notation
1753: \begin{eqnarray*}\zeta+1&=&\overrightarrow{a}=a\,e^{iA} \\ \zeta'-\zeta &
1754: =&\overrightarrow{b}=b\,e^{iB} \\1-\zeta'&=&\overrightarrow{c}=c\,e^{iC}
1755: \end{eqnarray*} (so far $A, B, C$
1756: are only defined modulo $2\pi$). Above this broken line lives a lift of $S_i$ which admits as a deck transformation
1757: $$f: u\mapsto 1+\frac{(\zeta+1)(\zeta'-1)}{u+1}~,$$ because $f(-1)=\infty$ ; $f(\infty)=1$ ; $f(\zeta)=\zeta'$. Therefore, $f(H_{1,\infty})=H_{|\zeta+1||\zeta'-1|,1}=H_{ac,1}$. Similarly, the following horoballs all belong to the same orbit: $$H_{1,\infty}~;~
1758: H_{ac,-1}~;~ H_{ab,z}~;~ H_{bc,z'}~;~ H_{ac,1}~.$$ If $\zeta=\xi+\eta \sqrt{-1}$ and $\zeta'=\xi'+\eta'\sqrt{-1}$, the corresponding isotropic vectors in Minkowski space are respectively
1759: \begin{equation} \begin{array}{lcccccccl}
1760: v_{\infty}&=&&(&0,&0,&-1,&1&) \\ v_{-1}&=&\frac{1}{ac}&(&-2,&0,&0,&2&)\\
1761: v_{\zeta}&=&\frac{1}{ab}&(&2\xi,&2\eta,&1-|\zeta|^2,&1+|\zeta|^2&)\\
1762: v_{\zeta'}&=&\frac{1}{bc}&(&2\xi',&2\eta',&1-|\zeta'|^2,&1+|\zeta'|^2&) \\
1763: v_{1}&=&\frac{1}{ac}&(&2,&0,&0,&2&). \end{array}\end{equation}
1764:
1765: \begin{figure} [h!]\centering
1766: \psfrag{m1}{$(-1)$}
1767: \psfrag{p1}{$(1)$}
1768: \psfrag{z}{$(\zeta)$}
1769: \psfrag{Z}{$(\zeta')$}
1770: \psfrag{a}{$\overrightarrow{a}$}
1771: \psfrag{b}{$\overrightarrow{b}$}
1772: \psfrag{c}{$\overrightarrow{c}$}
1773: \psfrag{ds}{$\overrightarrow{d_s}$}
1774: \psfrag{dS}{$\overrightarrow{d_{s'}}$}
1775: \psfrag{xi}{$x_i$}
1776: \psfrag{yi}{$y_i$}
1777: \psfrag{zi}{$\pi-w_i$}
1778: \psfrag{xim}{$x_{i-1}$}
1779: \psfrag{yim}{$y_{i-1}$}
1780: \psfrag{zim}{$\pi-w_{i-1}$}
1781: \ledessin{minkowski}
1782: \caption{Adjacent tetrahedra $\Delta_{i-1}, \Delta_i$ (cusp view).\llabel{minkowski}} \end{figure}
1783:
1784: To prove that the dihedral angle at the codimension-2 face projecting to $(
1785: \zeta \zeta' \infty)$ is convex, it is enough to show that if
1786: $Pv_{\zeta}+Qv_{\zeta'}+Rv_{\infty}=\lambda v_1+(1-\lambda)v_{-1}$ then
1787: $P+Q+R>1$ (moreover, this will in fact take care of \emph{all}
1788: codimension-2 faces of the simplicial complex $D$). One easily finds the unique solution
1789: $$P=\frac{-b\eta'}{c(\eta-\eta')}~;~Q=\frac{b\eta}{a(\eta-\eta')}~;~R= \frac{\eta(1-|\zeta'|^2)-\eta'(1-|\zeta|^2)}{ac(\eta-\eta')}$$
1790: hence $$P+Q+R=1+\frac{Z}{ac(\eta-\eta')}~~\text{ where }
1791: Z=bc\eta-ab\eta'+\eta(1-|\zeta'|^2)-\eta'(1-|\zeta|^2)+ac(\eta'-\eta).$$ Observe that $\eta>\eta'$
1792: because the triangles $-1\zeta'\zeta$ and $1\zeta\zeta'$ are counterclockwise
1793: oriented. So it is enough to prove that $Z>0$. Endow ${\mathbb
1794: C}\simeq {\mathbb R}^2$ with the usual scalar product $\bullet$ and
1795: observe that $1-|\zeta|^2=\overrightarrow{a}\bullet
1796: (\overrightarrow{b}+\overrightarrow{c})$ and
1797: $1-|\zeta'|^2=(\overrightarrow{a}+\overrightarrow{b})\bullet\overrightarrow{c}$.
1798: Hence \begin{eqnarray*} Z&=&
1799: \eta(bc+\overrightarrow{b}\bullet\overrightarrow{c})-\eta'(ab+\overrightarrow{a}\bullet\overrightarrow{b})+
1800: (\eta'-\eta)(ac-\overrightarrow{a}\bullet \overrightarrow{c}) \\
1801: &=& abc\left [\frac{\eta}{a}(1+\cos(B-C))-\frac{\eta'}{c}(1+\cos(A-B))+\frac{\eta'-\eta}{b}(1-\cos(A-C))\right ] \\ &=& abc[\sin A(1+\cos(B\!-\!C)) + \sin C(1+\cos(A\!-\!B)) + \sin B(1-\cos(A\!-\!C))] \\ &=&
1802: 4abc\,\sin\frac{A+C}{2}\cos\frac{A-B}{2}\cos\frac{B-C}{2}.
1803: \end{eqnarray*} by standard trigonometric formulae. Observe that the
1804: last expression is a well-defined function of $A,B,C \in \mathbb{R}/2\pi\mathbb{Z}$ (although each factor is defined only up to sign). Next, however, we shall
1805: carefully pick representatives of $A,B,C$ in ${\mathbb R}$.
1806:
1807: \begin{observation} \llabel{argumentnormal}
1808: There exists a unique triple of representatives $(A,B,C) \in \mathbb{R}^3$ such that $\{A,B,C\}\subset J$ where $J$ is an open interval of length less than $\pi$, containing $0$. (It is easy to see that the broken line $(\dots,-1,\zeta,\zeta',1,\dots)$ has no self-intersection if and only if there exists an open half-plane $H$ containing the vectors $\overrightarrow{a}, \overrightarrow{b}, \overrightarrow{c}$: the observation follows since $\overrightarrow{a}+\overrightarrow{b}+\overrightarrow{c}=2$ belongs to $H$). \end{observation}
1809: We pick the representatives $A,B,C \in \mathbb{R}$ given by Observation \ref{argumentnormal}, and do the same for the broken line contributed by each surface $S_i$ for $i\in\mathbb{Z}$ (each broken line is oriented from $-\infty$ to $+\infty$, so each edge $s$ of the cusp link inherits an orientation, as in Figure \ref{minkowski}: we write its complex coordinate
1810: $\overrightarrow{d_s}=d_s\, e^{D_s}$). By Observation \ref{argumentnormal}, all the complex arguments $D_s$ are in $(-\pi,\pi)$. But since $0\leq w \leq \pi$, the existence of the interval $J$ also implies (in the notation of Figure \ref{minkowski}):
1811: $$A=B+w_{i-1}~~~\text{ and }~~~C=B+w_i$$
1812: hence $B=\inf J$ and $B\in (-\pi,0)$. It follows that $B+x_i \in (-\pi,\pi)$, hence $$D_s=B+x_i~~~\text{ and similarly }~~~ D_{s'}=B+x_{i-1}.$$
1813: In other words (by transitivity), for any two edges $s_1,s_2$ of the cusp link, the difference of arguments $D_{s_1}-D_{s_2}\in\mathbb{R}$ can be read off ``naively'' as a linear combination of the $\{w_i\}_{i\in\mathbb{Z}}$, with no multiple of $2\pi$ added.
1814:
1815: %Therefore $A,B,C$ are contained in an interval $J$ of length
1816: %strictly less than $\pi$ (because $0<w<\pi$) and, since
1817: %$\overrightarrow{a}+\overrightarrow{b}+\overrightarrow{c}=2$, we may
1818: %assume that $J$ contains $0$. Since $B=\min J$, it follows that
1819: %$$B\in (-\pi,0) ~~\text{ and }~~ A,C\in(-\pi,\pi).$$ If these values
1820: %of $A,B,C$ in $\mathbb R$ are used to define all other $D_s$, then
1821: %$-\pi<D_s<\pi$ holds for any edge $s$: to see this by induction, it
1822: %is sufficient to prove that at least one edge of the next broken
1823: %line above (resp. below) inherits an argument in $(-\pi,\pi)$. And
1824: %indeed, Figure \ref{minkowski} shows that $D_s=B+x_i \in(-\pi,\pi)$
1825: %and $D_{s'}=B+x_{i-1} \in (-\pi,\pi)$.
1826:
1827: Now that all arguments are fixed in the real interval $(-\pi, \pi)$,
1828: we can see that $\cos\frac{A-B}{2}=\cos (w_{i-1}/2)$ and
1829: $\cos\frac{B-C}{2}=\cos (w_i/2)$ are positive. So to prove $Z>0$ it
1830: remains to see that $0<A+C<2\pi$. This in turn follows from a small
1831: variant of Lemma 16 in \cite{mapomme}, which we prove now (it implies an empirical observation of J{\o}rgensen which was Conjecture 8.6 in \cite{aswy1}).
1832: \begin{proposition} \llabel{tiltlemma}
1833: With the above notation, one has $0<A+C<2\pi$.
1834: \end{proposition}
1835: \begin{figure}[h!] \centering
1836: \psfrag{a}{$\overrightarrow{a}$}
1837: \psfrag{c}{$\overrightarrow{c}$}
1838: \psfrag{ceq}{$\left (\!\! \begin{array}{l} \overrightarrow{a}=a\,e^{iA} \\ \overrightarrow{c}=c\,e^{iC} \end{array} \!\! \right )$}
1839: \psfrag{m1}{$-1$}
1840: \psfrag{p1}{$1$}
1841: \psfrag{zm1}{$\zeta_{-1}$}
1842: \psfrag{z0}{$\zeta_0$}
1843: \psfrag{z1}{$\zeta_1$}
1844: \psfrag{znm}{$\zeta_{n-1}$}
1845: \psfrag{zn}{$\zeta_n$}
1846: \psfrag{znp}{$\zeta_{n+1}$}
1847: \psfrag{ze0}{$\xi_0$}
1848: \psfrag{zen}{$\xi_n$}
1849: \psfrag{zep}{$\xi_{n+1}$}
1850: \ledessin{manytriangles}
1851: \caption{\llabel{manytriangles} Cusp view: a subword $RL^nR$, bounded by hinge indices $0$ and $n$.} \end{figure}
1852:
1853: Consider the maximal subword $RL^nR$, with the broken line
1854: $(-1,\zeta,\zeta',1)$ corresponding to some $L$. There is in fact a sequence
1855: of points $(\zeta_0, \zeta_1, \dots,\zeta_n)$ in ${\mathbb C}$ such that the
1856: broken line corresponding to the $j$-th letter $L$ is
1857: $(-1,\zeta_j,\zeta_{j-1},1)$ for all $1\leq j \leq n$. The broken line
1858: corresponding to the initial (resp. final) $R$ is $(-1,\zeta_0,\zeta_{-1},1)$
1859: for some $\zeta_{-1}$ (resp. $(-1,\zeta_{n+1},\zeta_n,1)$ for some $\zeta_{n+1}$). There exists $1\leq k \leq n$ such that $(\zeta,\zeta')=(\zeta_k,\zeta_{k-1})$.
1860:
1861: Set $\xi_j:=\zeta_j+2$ for all $j$. By construction, the rays issued from $1$ through $\zeta_{-1},\zeta_0,\dots,\zeta_{n-1},\zeta_n,\xi_{n+1},\xi_n,\dots,\xi_1,\xi_0,\zeta_{-1}$ (in that cyclic order) divide ${\mathbb C}$ clockwise into salient angular sectors of sum $2\pi$ (these rays realize the link of the
1862: vertex $1$ in the cusp triangulation, see Figure \ref{manytriangles}).
1863: Comparing the angles at $1$ and at $-1$, we see immediately that
1864: \begin{equation} \llabel{convert} A+C=D_{-1 \zeta_k}+D_{\zeta_{k-1} 1}=D_{-1
1865: \zeta_{k+1}}+D_{\zeta_k 1}=\dots=D_{-1 \zeta_j}+D_{\zeta_{j-1} 1} \end{equation}
1866: (for all $0\leq j \leq n+1$). Since the triangles $\zeta_0 \zeta_{-1}1$
1867: and $\xi_0 1\zeta_{-1}$ are counterclockwise oriented, we have
1868: ${\rm Im}(\zeta_{-1})<{\rm Im}(1)$ i.e. ${\rm Im}(\zeta_{-1})<0$.
1869: Similarly, $\zeta_n 1 \xi_{n+1}$ and $\xi_n \xi_{n+1}1$ are
1870: counterclockwise oriented, so ${\rm Im}(\xi_{n+1})>0$. So there exists
1871: $0\leq j \leq n+1$ such that ${\rm Im}(\zeta_{j-1})< 0$ and ${\rm
1872: Im}(\zeta_j)\geq 0$. This implies $D_{-1 \zeta_j} \in (0,\pi)$ and
1873: $D_{\zeta_{j-1} 1}\in[0,\pi)$. By (\ref{convert}), this implies $A+C\in
1874: (0,2\pi)$. Theorem \ref{eph} is proved. \end{proof}
1875:
1876: In \cite{mapomme}, we studied punctured-torus bundles over the circle by indexing the tetrahedra $\Delta_i$ in $\mathbb{Z}/m\mathbb{Z}$ (instead of $\mathbb{Z}$): note that the proof of Theorem \ref{ephgeneral} applies without alteration to that context.
1877:
1878: \section{Generalizations} \llabel{sectionextensions}
1879: In this section, we extend all the previous results to punctured-torus groups with rational pleatings and/or infinite ends. The general theorem is as follows.
1880: \begin{theorem} \llabel{ephhyper}
1881: Let $\lambda^+\neq \lambda^-$ be nonzero, possibly projective (if irrational), measured laminations on the punctured torus $S$, and let $s(\lambda^{\pm})\in \mathbb{P}^1\mathbb{R}$ be the slope of $\lambda^{\pm}$. If $\lambda^{\pm}$ is a closed leaf, denote its weight by $|\lambda^{\pm}|$ and assume $|\lambda^{\pm}|\leq\pi$. There exists a punctured-torus group $\Gamma$ with ending and/or pleating laminations $\lambda^{\pm}$, and the open convex core $V$ of $\mathbb{H}^3/\Gamma$ has an ideal decomposition $\mathcal{D}$ into polyhedral cells (of positive volume) whose combinatorics are given by $\lambda^{\pm}$ in the following sense: if $\Lambda$ is the line from $s(\lambda^-)$ to $s(\lambda^+)$ across the Farey diagram in $\mathbb{H}^2$, then
1882: \begin{enumerate}
1883: \item If $s(\lambda^+)$ and $s(\lambda^-)$ are irrational, $\mathcal{D}$ consists of ideal tetrahedra $(\Delta_i)_{i\in\mathbb{Z}}$ in natural bijection with the Farey edges crossed by $\Lambda$, as in Section \ref{sectionstrategy}.
1884: \item If only $s(\lambda^+)$ is rational and $|\lambda^+|<\pi$, then $\mathcal{D}$ has one ideal tetrahedron per Farey edge crossed by $D$, and one cell $T$ homeomorphic to a solid torus: $\partial_1=\overline{\partial T \cap \partial V}$ is a punctured torus pleated along a simple closed curve of slope $s(\lambda^+)$, and $\partial_2=\overline{\partial T \smallsetminus \partial V}$ is a punctured torus pleated along the ideal triangulation associated to the Farey triangle with vertex $s(\lambda^+)$ crossed by $\Lambda$. Finally, $\partial_1\cap\partial_2$ is a line from the puncture to itself of slope $s(\lambda^+)$. See the left panel of Figure \ref{toricpieces}.
1885: \item If only $s(\lambda^+)$ is rational and $|\lambda^+|=\pi$, all the statements of the previous case apply, except that $\partial_1=\overline{\partial T \cap \partial V}$ becomes a thrice-punctured sphere (the simple closed curve of slope $s(\lambda^+)$ has been ``pinched'' to become a cusp): see the right panel of Figure \ref{toricpieces}.
1886: \item If only $s(\lambda^-)$ is rational, the situation is similar to the two previous cases, exchanging $\lambda^-$ and $\lambda^+$.
1887: \item If $s(\lambda^+),s(\lambda^-)$ are rationals but not Farey neighbors, the situation is again similar, with two solid torus cells $T^+$ and $T^-$.
1888: \item If $s(\lambda^+), s(\lambda^-)$ are Farey neighbors, $\mathcal{D}$ only consists of two solid tori $T^+$ and $T^-$ as above, glued along a punctured torus $S$ pleated along only two lines, of slopes $s(\lambda^+)$ and $s(\lambda^-)$.
1889: \end{enumerate}
1890: Moreover, $\mathcal{D}$ agrees with the geometrically canonical decomposition $\mathcal{D}^G$ of the open convex core given by the Epstein-Penner convex hull construction.
1891: \end{theorem}
1892: Note that the combinatorics of $\mathcal{D}^G$ do not depend on the nature (finite or infinite) of the ends of $\mathbb{H}^3/\Gamma$. At this point, we have treated the case of two irrational pleatings (finite ends). We proceed to prove the remaining cases of Theorem \ref{ephhyper}.
1893:
1894: \begin{figure}[h!] \centering
1895: \psfrag{tlp}{$\theta<\pi$}
1896: \psfrag{tp}{$\theta=\pi$}
1897: \ledessin{toricpieces}
1898: \caption{Toric cells: the exterior dihedral angle $\theta$ is the weight $|\lambda^{\pm}|$. Shaded faces are identified. \llabel{toricpieces}} \end{figure}
1899:
1900:
1901: \subsection{One rational pleating} \llabel{onerationalpleating}
1902: We focus on the case of two finite ends, with only $\beta^+/\alpha^+$ rational.
1903: We can choose to end the word $\Omega\in \{R,L\}^{\mathbb{Z}}$ with an infinite suffix $LRR...R...$ (or of $RLL...L...$: that is an arbitrary choice) and proceed from
1904: Section \ref{sectionstrategy} onward. We shall assume that $i=0$ is
1905: the greatest hinge index. Sections
1906: \ref{sectionangles} through \ref{subsectionnaturalconstraint} are unchanged:
1907: the sequence $(w_i)$ is just concave (thus convergent and non-decreasing) on
1908: $\mathbb{N}$. In Subsection \ref{subsectionstudyphi}, we find that
1909: the sequence $(\phi^+_i)$ is constant on $\mathbb{N}$, equal to some
1910: $\theta>0$. By (\ref{phiphi}), the number $\theta$ is the weight of the rational lamination $\lambda^+$, so we assume $\theta<\pi$.
1911: %We will need to distinguish the case $\theta<\pi$ from
1912: %the case $\theta=\pi$ (the latter, as well as $\theta>\pi$, means
1913: %that the constraints on $w$ involving $\phi^+$ in (\ref{doubleve})
1914: %are vacuous).
1915: Section \ref{sectionhyperbolicvolume} goes through
1916: essentially unchanged: by the computation of Sublemma \ref{boundhops} (and with the same notation), the sum of the volumes of all tetrahedra $\Delta_i$ for $i\geq 2^n$ is at most
1917: $$\sum_{k>n} \Sigma_{2^{k-1}}^{2^k}\leq \sum_{k>n}
1918: 2^{-k}[1+(2k-1)\log 2]=O(2^{-n/2}).$$ Therefore the volume
1919: functional $\mathcal{V}$ is bounded, continuous for the product
1920: topology, and concave.
1921: We can find a maximizer $w$ of $\mathcal{V}$, and it still satisfies Propositions \ref{tricot} and \ref{hingefragile}: in particular, all tetrahedra $\Delta_i$ for $i>0$ have positive angles.
1922:
1923: Something new is required in Section \ref{sectionbehaviorofwi}: we
1924: must prove that $\lim_{i\rightarrow +\infty}w_i=\theta$.
1925:
1926: The $\{w_i\}_{i\geq 2}$ contribute only to the angles of the $\{\Delta_i\}_{i\geq 1}$, which are positive: so the volume $\mathcal{V}$ is critical with respect to each $w_i$ for $i\geq 2$. By Sublemma 6 of \cite{mapomme}, it follows that the cusp triangles of the $\{\Delta_i\}_{i\geq 1}$ fit together correctly and can be drawn in the Euclidean plane $\mathbb{C}$. More precisely, there exists a sequence of complex numbers $(\zeta_i)_{i\geq 0}$ such that the triangles contributed by $\Delta_i$ have vertices at $(-1,\zeta_i,\zeta_{i-1})$ and $(1,\zeta_i,\zeta_{i+1})$ (Figure \ref{nautilus}, left). These triangles being similar, we have $(\zeta_i+1)(\zeta_{i+1}-1)=(\zeta_{i-1}+1)(\zeta_i-1)=...=(\zeta_0+1)(\zeta_1-1)$, hence
1927: $$\zeta_{i+1}=\frac{\zeta_i+\kappa}{\zeta_i+1}=:\varphi(\zeta_i)$$ for some complex number $\kappa\neq 1$ independent of $i$. Observe that the complex length $\ell$ of the hyperbolic isometry (extending) $\varphi$ satisfies $\cosh \ell = \frac{1+\kappa}{1-\kappa}$. In the end, $\varphi$ will be a lift of the loop along the rational pleating line of the convex core.
1928:
1929: \begin{proposition}
1930: The number $\kappa$ lies in the real interval $(0,1)$.
1931: \end{proposition}
1932: \begin{proof} Denote by $Z_i$ the periodic broken line $(\dots,-1,\zeta_{i-1},\zeta_i,1,2+\zeta_{i-1},2+\zeta_i,\dots)$ contributed by the pleated surface $S_i$. First, $\kappa$ is real: if not, the $\zeta_i$ have a limit in $\mathbb{C} \smallsetminus \mathbb{R}$ (a square root of $\kappa$), so $w_i-w_{i-1}$ (the angle of $Z_i$ at $1$) cannot go to $0$. If $\kappa<0$, then $\varphi$ is a pure rotation: the $\zeta_i$ all belong to a circle of the lower half-plane, which also contradicts $w_i-w_{i-1}\rightarrow 0$. If $\kappa>1$, then $\varphi$ is a gliding axial symmetry: the $\zeta_i$ go to $\pm\sqrt{\kappa}$ and belong alternatively to the upper and lower half-plane, so $Z_i$ has self-intersection for large $i$. If $\kappa=0$, then $\varphi$ is a parabolic transformation fixing $0$: the $\zeta_i$ go to $0$ along a circle tangent to $\mathbb{R}$, and one can see that $w_i=\widehat{\zeta_i \zeta_{i+1}1}$ goes to $\pi>\lim_{+\infty}\phi^+=\theta$. The only remaining possibility is $\kappa \in (0,1)$, where $\varphi$ is a pure translation.
1933: \end{proof}
1934:
1935: Thus, the $\zeta_i$ lie on a circle arc $C$ of the lower half-plane which meets the real line at $\pm \sqrt{\kappa}$, and $\lim_{+\infty} \zeta_i=\sqrt{\kappa}$. We denote by $\theta^*$ the angle between $C$ and the segment $[-\sqrt{\kappa},\sqrt{\kappa}]$ (more precisely, the angle between their half-tangents at $\sqrt{\kappa}$). It is easy to see that $$\theta^*=\underset{i\rightarrow +\infty}{\lim} w_i$$ (see Figure \ref{nautilus}).
1936: % --- if $\kappa=0$, then $C$ is a full circle tangent to $\mathbb{R}$ at $0$).
1937: Hence, $\theta^*\leq \theta$.
1938:
1939: \begin{figure}[h!] \centering
1940: \psfrag{m1}{$-1$}
1941: \psfrag{p1}{$1$}
1942: \psfrag{msqk}{$-\sqrt{\kappa}$}
1943: \psfrag{sqk}{$\sqrt{\kappa}$}
1944: \psfrag{C}{$C$}
1945: \psfrag{z0}{$\zeta_0$}
1946: \psfrag{z1}{$\zeta_1$}
1947: \psfrag{z2}{$\zeta_2$}
1948: \psfrag{fz0}{$f(\zeta_0)$}
1949: \psfrag{fz1}{$f(\zeta_1)$}
1950: \psfrag{fz2}{$f(\zeta_2)$}
1951: \psfrag{0}{$0$}
1952: \psfrag{r}{$\rho$}
1953: %\psfrag{}{$$}
1954: \ledessin{nautilus}
1955: \caption{All marked angles (grey) are $\pi-\theta^*$, and $\zeta_{j+1}=\varphi(\zeta_j)=\frac{\zeta_j+\kappa}{\zeta_j+1}$. \llabel{nautilus}} \end{figure}
1956:
1957: \begin{proposition} One has $\theta^*=\theta$. \llabel{behaviorofwirational} \end{proposition}
1958: \begin{proof}
1959: First, it is easy to check that any data $0\leq w_0<w_1<\theta^*< \pi$ smoothly determines a \emph{unique} pair of complex numbers $\zeta_0,\zeta_1$ such that:
1960: \begin{enumerate}
1961: \item the broken line $(-1,\zeta_0,\zeta_1,1,2+\zeta_0,2+\zeta_1,\dots)$ has angles $(w_0,-w_1,w_1-w_0)$ as in (\ref{pleatingangles}) above;
1962: \item the number $\kappa$ such that $\zeta_1=\frac{\zeta_0+\kappa}{\zeta_0+1}$ lies in $(0,1)$;
1963: \item the circle through $\zeta_0$ and $\zeta_1$ centered on the imaginary axis intersects the real axis at an angle $\theta^*$. \end{enumerate}
1964:
1965: These $\zeta_0,\zeta_1$ in turn define all $\{\zeta_j\}_{j\geq 2}$ \emph{via} $\zeta_{j+1}=\varphi(\zeta_j)$, and we can read off the angle $w_j=\widehat{1 \zeta_{j+1} \zeta_j}\leq \theta^*$ and construct the associated ideal tetrahedron $\Delta_j$. In what follows, we investigate the shape of the space $U:=\bigcup_{j\geq 1} \Delta_j$, whose boundary (the punctured torus $S_1$) has pleating angles $(w_0,-w_1,w_1-w_0)$ (see (\ref{pleatingangles}) above).
1966:
1967: Define $f(\zeta):=\frac{\zeta+\sqrt{\kappa}}{\zeta-\sqrt{\kappa}}$, so that $f(\varphi(\zeta))=\rho f(\zeta)$ where $\rho:=\frac{1+\sqrt{\kappa}}{1-\sqrt{\kappa}}$. The convex hull of $(\infty,1,\zeta_j,\zeta_{j+1})$ is isometric to the tetrahedron $\Delta_j$: pushing forward by $f$, we obtain a tetrahedron $\Delta'_j$, isometric to $\Delta_j$, with vertices $(1,\rho, \rho^j f(\zeta_0),\rho^{j+1} f(\zeta_0))$ (Figure \ref{nautilus}, right). Moreover, all the $f(\zeta_j)=\rho^j f(\zeta_0)$ lie on the half-line $e^{i(\pi-\theta^*)}\mathbb{R}^+$. By reasoning in a fundamental domain of the loxodromy $\Phi: u\mapsto \rho u$, it is then easy to see that $U$ has the same volume as $D:= D_1 \cup D_2$, where $D_1,D_2$ are ideal tetrahedra of vertices $(\infty, f(\zeta_1),1,\rho)$ and $(\infty,f(\zeta_1),1,f(\zeta_0))$ respectively. Moreover, $\Phi$ identifies the faces $(\infty,1,f(\zeta_0))$ and $(\infty,\rho,f(\zeta_1))$ of $D$, so that $D/\Phi$ is a manifold with polyhedral boundary, homeomorphic to a solid torus, with interior dihedral angles $$(\pi-w_1,\pi+w_0,\frac{w_1-w_0}{2},\frac{w_1-w_0}{2},\pi-\theta^*).$$ The edge of $D/\Phi$ with dihedral angle $\pi-\theta^*$ is a simple closed curve of length $\log \rho$, toward which $D_1,D_2$ spiral. A picture of $D/\Phi$ is obtained by replacing $\theta$ with $\theta^*$ in the left panel of Figure \ref{toricpieces}.
1968:
1969: Using the smooth dependence on $\theta^*$, the Schl\"afli volume formula then gives $d\mathcal{V}(D/\Phi)/d\theta^*=\frac{1}{2}\log \rho>0$: so $D/\Phi$ (and therefore $U$) has largest volume when $\theta^*$ is largest. Regard the $\{w_j\}_{j\leq 1}$ as fixed, and the $\{w_j\}_{j\geq 2}$ as unknowns: then $(w_2, w_3,\dots)$ is clearly the solution to the maximization problem for the volume of $U$ (with fixed pleating angles on $S_1$). Therefore, the $w_i$ will choose the largest possible limit $\theta^*$ at $+\infty$, namely $\theta^*=\theta$.
1970: \end{proof}
1971:
1972: The above proof does more than determining $\lim_{+\infty} w_j$: as in Sections \ref{sectionbehaviorofwi}-\ref{sectionextrinsicconvergence}, it gives a full description of $U=\bigcup_{j\geq 1} \Delta_j$ and of its boundary (whose pleating turns out to be $\lambda^+$). Here is, however, an important observation:
1973: \begin{observation} \llabel{depli}
1974: Proposition \ref{tiltlemma} above, ``$A+C>0$'', does not hold for the family of pleated surfaces $(S_i)_{i\geq 0}$. Instead, we have $A+C=0$.
1975: This simply means that $\overrightarrow{-1,\zeta_i}$ and $\overrightarrow{\zeta_{i+1},1}$ make opposite angles with the real line. Indeed, $(\zeta_i+1)(1- \zeta_{i+1})=1-\kappa^2$ is a positive real.\end{observation}
1976:
1977: We can now establish
1978: \begin{proposition} All the (strict) inequalities of (\ref{doubleve}) are true.\llabel{winteriorrational} \end{proposition}
1979: \begin{proof}
1980: This is Proposition \ref{winterior} (in the new context where $\beta^+/\alpha^+$ is rational). The proof is the same, with the following caveat: in ruling out $w_j=0$ for $j$ hinge, we call upon \cite{mapomme} (especially Lemma 16 and the argument of Section 9 there). The strategy is to assume $w_j=0$, and then perturb $w$ to a well-chosen $w^{\varepsilon}$ so as to make the volume increase: $\partial\mathcal{V}/\partial\varepsilon>0$. The latter inequality holds essentially because the inequality of Proposition \ref{tiltlemma} (``$A+C>0$'') is true, both in the $R^n$-word preceding $j$ and in the $L^m$-word following $j$ (\footnote{In Section 9 of \cite{mapomme}, the inequality ``$A+C>0$'' is formulated in terms of lengths and takes the guise ``$Q<P+T$''.}). More exactly, $\partial\mathcal{V}/\partial\varepsilon$ will be positive when at least one of the two instances of ``$A+C\geq 0$'' is strict. But it is always strict, except in a single case (the infinite suffix $LRR...R...$): so we can conclude.
1981: % O U H L A L A
1982: \end{proof}
1983:
1984: As a result, all the tetrahedra $\Delta_i$ have positive angles and fit together correctly: Sections \ref{sectionbehaviorofwi}-\ref{sectionextrinsicconvergence} carry through for the $\lambda^-$-end, and $\bigcup_{i\in\mathbb{Z}}\Delta_i$ is the open convex core of a quasifuchsian punctured-torus group, with the prescribed pleatings $\lambda^{\pm}$.
1985:
1986: The results of Section \ref{sectioneph} extend readily: the only modification is that tetrahedra $(\Delta_i)_{i\geq 0}$ lift to a family of \emph{coplanar} cells in Minkowski space, because the key inequality of Proposition \ref{tiltlemma} has become an equality. Therefore the geometrically canonical decomposition of the convex core contains the non-contractible cell $D/\Phi$ (Figure \ref{toricpieces}, left).
1987:
1988: \subsection{Two rational pleatings} \llabel{tworationalpleatings}
1989: When both $\beta^+/\alpha^+$ and $\beta^-/\alpha^-$ are rational, the encoding word $\Omega\in\{R,L\}^{\mathbb{Z}}$ can be chosen with an infinite prefix $...R...RRL$ and an infinite suffix $LRR...R...$ We apply the same argument as above to both ends simultaneously. Again, Proposition \ref{winteriorrational} holds because no hinge index $j$ belongs both to the prefix and to the suffix.
1990:
1991: If the rational pleating slopes $s(\lambda^{\pm})$ are not Farey neighbors, we can convert the prefix to $...LLR$ and/or the suffix to $RLL...$, obtaining different triangulations of the convex core of the same quasifuchsian group.
1992:
1993: If $s(\lambda^+),s(\lambda^-)$ are Farey neighbors, then $\Omega=...RRLRR...$ (observe that prefix and suffix overlap, so the two word conversions do \emph{not} commute). If the indices before and after the $L$ are $0$ and $1$, we obtain $w_0=w_1$ by applying Observation \ref{depli} to prefix and suffix. In other words, the points $a,b,c$ in Figure \ref{fareyneighbors} (left) are collinear. It is therefore possible to triangulate the same convex core according to a word $\Omega=...LLLRRR...$, provided that we allow the hinge tetrahedron $\Delta_0$ to become flat (black in Figure \ref{fareyneighbors}, right). In any case, the sequence $(w_i)_{i\in\mathbb{Z}}$ maximizes the total volume.
1994:
1995: \begin{figure}[h!] \centering
1996: \psfrag{rlr}{$...RRLRR...$}
1997: \psfrag{lr}{$...LLLRRR...$}
1998: \psfrag{a}{$a$}
1999: \psfrag{b}{$b$}
2000: \psfrag{c}{$c$}
2001: \ledessin{fareyneighbors}
2002: \caption{Two triangulations seen against the same limit set. \llabel{fareyneighbors}} \end{figure}
2003:
2004: \subsection{Pinching}
2005: The case where one (or both) of the pleatings $\lambda^{\pm}$ has weight $\pi$ is a straightforward limit case of Subsections \ref{onerationalpleating} and \ref{tworationalpleatings} (the term ``pinching'' refers to the fact that the pleating curve becomes shorter and shorter, and eventually turns into a cusp as the pleating angle reaches $\pi$).
2006:
2007: Suppose $|\lambda^+|=\pi$ (note that the conditions of (\ref{doubleve}) involving $\phi^+$ become vacuous, because $\phi^+\geq\pi$). Subsection \ref{onerationalpleating} carries through with $\theta=\pi$ and $\kappa=0$. The circle arc $C$ of Figure \ref{nautilus} becomes a full circle, tangent to $\mathbb{R}$ at $0$. The analysis of Proposition \ref{behaviorofwirational} (existence and uniqueness of $\zeta_0, \zeta_1\in\mathbb{C}$) extends smoothly to $\theta^*=\pi$. One then finds that the tetrahedron with vertices $(\infty,1,\zeta_j,\zeta_{j+1})$ is sent by $f:u\mapsto 1/u$ to a tetrahedron $\Delta'_j$ of $\mathbb{H}^3$, isometric to $\Delta_j$, with vertices $(0,1,\tau+j,\tau+j+1)$ for a certain $\tau \in \mathbb{C}$ independent of $j$ (in the right panel of Figure \ref{nautilus}, the grey bounding rays are replaced by parallel lines). It is straightforward to check that $\bigcup_{j\geq 1}\Delta_j$ is the solid torus pictured in Figure \ref{toricpieces} (right), and the Schl\"afli formula again implies that $\theta^*=\pi$ realizes the maximum volume.
2008:
2009: After Proposition \ref{behaviorofwirational}, the argument is unchanged.
2010:
2011: \subsection{Infinite ends}
2012: By \cite{minsky}, quasifuchsian groups are dense in the set of all discrete, faithful, type-preserving representations of $\pi_1(S)\rightarrow \text{Isom}^+(\mathbb{H}^3)$. In fact, a geometrically infinite end of such a representation comes with an \emph{ending lamination}, namely an irrational projective measured lamination which should be thought of as an ``infinitely strong pleating''. In this section we will assume $\beta^+/\alpha^+ \notin \mathbb{P}^1 \mathbb{Q}$ and consider $w^T$, the solution of the volume maximization problem for $(\phi^-,T\phi^+)$, where $T>0$ (namely, $w$ is subject to conditions (\ref{doubleve}), where $\phi^+$ is replaced by $T \phi^+$; and Table (\ref{xiyi}) still expresses $x_i, y_i, z_i$).
2013:
2014: In a discrete, type-preserving representation of a surface group, each measured lamination $\lambda$ on the surface receives a \emph{length}, which can be computed by measuring the (weighted) lengths of weighted curves converging to $\lambda$.
2015: It is known \cite{bridgeman} that the lengths of the pleating laminations of a quasifuchsian group are bounded by a constant depending only on the underlying surface (here the punctured torus). By Thurston's double limit theorem (see Theorems 4.1 and 6.3 of \cite{thurstonlimitedouble}), the space of discrete, type-preserving representations in which two fixed measured laminations have length bounded by a given constant is \emph{compact}. Therefore, up to taking a subsequence, the groups $\Gamma^T$ corresponding to $w^T$ converge algebraically to a certain $\Gamma$. By \cite{minsky}, $\mathbb{H}^3/\Gamma$ is homeomorphic to $S\times \mathbb{R}$ and must have an infinite end (otherwise, $\Gamma$ would be quasifuchsian and the volumes would stay bounded).
2016:
2017: \begin{proposition}
2018: The $\lambda^-$-end of $\mathbb{H}^3/\Gamma$ is finite, with pleating lamination $\lambda^-$. The $\lambda^+$-end is infinite, with projective ending lamination $[\lambda^+]$. \end{proposition}
2019: \begin{proof}
2020: By \cite{bonahon} (Theorem D), the space of type-preserving representations of the abstract group $\Gamma$ is smoothly (in fact, holomorphically) parametrized by the data $(\tau, \omega)$ of a point $\tau$ of Teichm\"uller space $\mathcal{T}$, and a \emph{transverse $\mathbb{R}/2\pi \mathbb{Z}$-valued cocycle} relative to a fixed topological lamination $\mu$ (such cocycles include pleating measures as special cases). Taking for $\mu$ the support of $\lambda^-$, we see that the moduli of the $\lambda^-$-boundaries of the convex cores of the $\Gamma^T$ must converge in $\mathcal{T}$. Therefore, $\mathbb{H}^3/\Gamma$ contains a locally convex pleated surface $H$ with pleating $\lambda^-$, which must be a boundary of the convex core ($\partial H$ contains all parabolic fixed points).
2021:
2022: The $\lambda^+$-end of $\mathbb{H}^3/\Gamma$ must therefore be infinite.
2023: The parabolic fixed points of the limit group $\Gamma$ determine a version of Figure \ref{frise} (a Euclidean cusp link), and therefore a family of \emph{non-negative} angle assignments for the tetrahedra $\Delta_i$. By algebraic convergence, the $w^T$ converge to $w$ in the product topology. For any $i\leq j$, the total volume of $\Delta_{i-1}, \Delta_i, \dots, \Delta_j, \Delta_{j+1}$ is maximal with respect to $w_i, \dots,w_j$: in particular, Propositions \ref{tricot} and \ref{hingefragile} are still true. The techniques of \cite{mapomme} (Lemma 16 and Section 9 there) show that all $\Delta_i$ have positive angles, and Proposition \ref{tiltlemma}, hence also Lemma \ref{minkonwex}, still hold: $\{\Delta_i\}_{i\in\mathbb{Z}}$ is the geometrically canonical decomposition of $\mathbb{H}^3/\Gamma$. In particular, the family of all edges of all tetrahedra $\{\Delta_i\}_{i\geq 0}$ forms a sequence of laminations which exits $\mathbb{H}^3/\Gamma$: therefore $[\lambda^+]$ is the end invariant.
2024: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2025: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2026: %%% %%%
2027: %%% B A N C A L . %%%
2028: %%% %%%
2029: %%% T E R M I N E R ? %%%
2030: %%% %%%
2031: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2032: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2033: \end{proof}
2034: The case of two infinite ends is already treated in \cite{akiyoshi}. Theorem \ref{ephhyper} is proved.
2035:
2036: \subsection{The Pleating Lamination Conjecture for punctured-torus groups}
2037:
2038: \begin{proposition} The group $\Gamma$ constructed at the end of Section \ref{sectionhyperbolicvolume} is \emph{continuously} parametrized by $(\lambda^+,\lambda^-)$. \llabel{groupcontinuous} \end{proposition}
2039: \begin{proof}
2040: Our first observation is that if $\beta^+/\alpha^+$ is rational, the initial choice of infinite prefix/suffix in Subsection \ref{onerationalpleating} does not change the resulting group $\Gamma$: it just induces different triangulations of the toric piece of Figure \ref{toricpieces}, whose deformation space is still the same.
2041:
2042: Define the open set $$\mathcal{U}:=\mathbb{R}^2\smallsetminus \bigcup_{m,n\in \mathbb{Z}}[\pi,+\infty)\cdot\{(m,n)\}$$ (note that $0\notin \mathcal{U}$). An admissible pleating lamination $\lambda$ can be identified with an element $\pm(\alpha,\beta)$ of $\mathcal{U}/\pm$. Suppose $(\alpha^+_n, \beta^+_n) \rightarrow (\alpha^+, \beta^+)$ and $(\alpha^-_n, \beta^-_n) \rightarrow (\alpha^-, \beta^-)$ in $\mathcal{U}$, and define the oriented line $\Lambda_n$ from $\beta^-_n/\alpha^-_n$ to $\beta^+_n/\alpha^+_n$ across the Farey diagram. Also define the associated functions $\phi^{\pm,n}$ as in (\ref{phiphi}), domains $W^n$ as in Definition \ref{defineW}, and solutions $(w^n_i)_{i\in\mathbb{Z}}$ to the volume maximization problem over $W^n$. If $\beta^+/\alpha^+$ is rational, we may assume (up to restricting to two subsequences) that the $(\alpha^+_n, \beta^+_n)$ converge to $(\alpha^+,\beta^+)$ in the clockwise direction for the natural orientation of $\mathbb{P}^1\mathbb{R}$. We also make a similar assumption for $\beta^-/\alpha^-$.
2043:
2044: A priori, the sequences $(w^n_i)_{i\in \mathbb{Z}}$ are defined only up to a shift of the index $i$. However, we can choose these shifts in a consistent way:
2045: there exists a Farey edge $e$ which is crossed by all the lines $\Lambda_n$ for $n$ large enough, so we decide that $w^n_0$ always lives on $e$ (namely, $e^n_0=e$). By compactness of $[0,\pi]^{\mathbb{Z}}$, some subsequence of $(w^n)_{n\in \mathbb{N}}$ converges to some $w^*$. It is enough to show that $w^*=w$: indeed, the group $\Gamma$ is completely determined by the shapes of a finite number of tetrahedra $\Delta_i$.
2046:
2047: The main observation is that the words $\Omega^n \in \{R,L\}^{\mathbb{Z}}$ converge pointwise to $\Omega$, and $\phi^{\pm,n}\rightarrow \phi^{\pm}$ (pointwise in $\mathbb{R}^{\mathbb{Z}}$), by definition (\ref{phiphi}). Therefore, $w^*$ belongs to the space $W$, hence the volume inequality $\mathcal{V}(w^*)\leq \mathcal{V}(w)$. Since $\max_W \mathcal{V}$ is achieved at a unique point (the volume is a strictly concave function), it is enough to prove the reverse inequality. We proceed by contradiction.
2048:
2049: Suppose $\mathcal{V}(w^*) < \mathcal{V}(w)$. Pick $\varepsilon>0$: there exist integers $m<0<M$ such that the tetrahedra $\{\Delta_i\}_{m<i<M}$ defined by $w$ have total volume at least $\mathcal{V}(w)-\varepsilon$. If we can extend $(w_i)_{m\leq i\leq M}$ to a sequence $(v_i)_{i\in\mathbb{Z}}$ of $W^n$ for some large $n$, we will obtain a contradiction (assuming $\varepsilon$ small enough).
2050:
2051: By Corollary \ref{hiatus2}, we can assume $\frac{w_m}{\phi^-_m}>\frac{w_{m+1}}{\phi^-_{m+1}}$ and $\frac{w_M}{\phi^-_M}>\frac{w_{M-1}}{\phi^-_{M-1}}$. Since $w$ satisfies (\ref{doubleve}) (strong inequalities), for $n$ large enough, the restricted sequence $(w_m,\dots,w_M)$ satisfies the corresponding inequalities defining $W^n$, by convergence of the $\phi^{\pm, n}$. Pick such a large $n$ and define
2052: $$v_i:=\left \{ \begin{array}{ll} \phi^{-,n}_i \frac{w_m}{\phi^{-,n}_m} & \text{ if } i\leq m~; \\ w_i & \text{ if } m\leq i \leq M ~; \\ \phi^{+,n}_i \frac{w_M}{\phi^{+,n}_M} & \text{ if } M\leq i~. \end{array} \right .$$ It is a straightforward exercise to check that $(v_i)_{i\in \mathbb{Z}}$ belongs to $W^n$. \end{proof}
2053:
2054: \begin{corollary} \emph{(C. Series)}\llabel{plth}
2055: A quasifuchsian punctured-torus group $\Gamma$ is determined up to conjugacy in $\text{Isom}(\mathbb{H}^3)$ by its pleating measures $\lambda^{\pm}$.
2056: \end{corollary}
2057: \begin{proof}
2058: We will use the well-known fact that the space $\mathcal{QF}$ of quasifuchsian, non-fuchsian (punctured-torus) groups is a connected real manifold of dimension $4$. Recall the open set $\mathcal{U}$ from the proof of Proposition \ref{groupcontinuous}, and consider the map
2059: $$f:\mathcal{U} \times \mathcal{U} \longrightarrow \mathcal{QF}$$ defined by the construction of the group $\Gamma$ (end of Section \ref{sectionhyperbolicvolume}). We know that $f$ is well-defined and injective (Theorem \ref{pureexistence}), and continuous (Proposition \ref{groupcontinuous}). Since $\mathcal{U}^2$ has dimension $4$, the theorem of domain invariance states that the image $\text{Im}(f)$ is open. It remains to show that $\text{Im}(f)$ is closed.
2060:
2061: Consider pairs $(\lambda^+_n, \lambda^-_n)$ such that the corresponding groups $\Gamma_n = f(\lambda^+_n, \lambda^-_n)$ converge to some $\Gamma$ in $\mathcal{QF}$. The function which to a group associates its pleatings is continuous (see \cite{keenseries}), so the $\lambda^{\pm}_n$ converge to some $\lambda^{\pm}$ in $\mathcal{U}$. Proposition \ref{groupcontinuous} then implies that $\Gamma=f(\lambda^+,\lambda^-)$. \end{proof}
2062: Theorem \ref{ephgeneral} now follows from Theorem \ref{ephhyper}.
2063:
2064: \begin{thebibliography}{MaR}
2065:
2066: \bibitem[Ak]{akiyoshi} Hirotaka Akiyoshi, \emph{On the Ford domains of once-punctured torus groups}, in \emph{Hyperbolic spaces and related topics}, RIMS, Kyoto, Kokyuroku {\bf 1104} (1999), 109-121.
2067:
2068: \bibitem[AS]{comparing} Hirotaka Akiyoshi, Makoto Sakuma,
2069: \emph{Comparing two convex hull constructions of cusped hyperbolic
2070: manifolds}, Proceedings of the Workshop ``Kleinian groups and
2071: hyperbolic 3-manifolds'' (Warwick 2002), Lond. Math. Soc. Lecture
2072: Notes {\bf 299} (2003), 209--246.
2073:
2074: \bibitem[ASWY1]{aswy1} H. Akiyoshi, M. Sakuma, M. Wada, Y. Yamashita, \emph{Jorgensen's picture of quasifuchsian punctured torus groups}, Proceedings of the Workshop ''Kleinian groups and hyperbolic 3-manifolds'' (Warwick 2002), Lond. Math. Soc. Lecture Notes {\bf 299} (2003), 247--273.
2075:
2076: \bibitem[ASWY2]{aswy2} H. Akiyoshi, M. Sakuma, M. Wada, Y. Yamashita, \emph{Ford domains of punctured torus groups and two-bridge knot groups}, in \emph{Knot Theory}, Proceedings of the workshop held in Toronto dedicated to 70th birthday of Prof. K. Murasugi, 1999.
2077:
2078: \bibitem[ASWY3]{aswy3} Hirotaka Akiyoshi, Makoto Sakuma, Masaaki Wada,
2079: Yasushi Yamashita, \emph{Punctured torus groups and 2-bridge knot groups} , Preprint.
2080:
2081: %\bibitem[BH]{haefliger} Martin Bridson, Andr\'e Haefliger, \emph{Metric spaces
2082: %of non-positive curvature}, Grundlehren der mathematischen
2083: %Wissenschaften 319, Springer (1999).
2084:
2085: \bibitem[Bon]{bonahon} Francis Bonahon, \emph{Shearing hyperbolic surfaces, bending pleated surfaces and Thurston's symplectic form}, Ann. Fac. Sci. Toulouse Math. {\bf 5} (1996), 233--297.
2086:
2087: \bibitem[Bow]{bowditch} Brian H. Bowditch, \emph{Markoff triples and quasifuchsian groups}, Proc. Lond. Math. Soc. Vol. {\bf 77} (1998) 697--736.
2088:
2089: \bibitem[Br]{bridgeman} Martin Bridgeman, \emph{Average bending of convex pleated planes in hyperbolic three-space}, Invent. Math. {\bf 132}, no. 3 (1998), 381--391.
2090:
2091: \bibitem[CEG]{lms} R. D. Canary, D.B.A. Epstein, R. Green, \emph{Notes on notes of Thurston}, in
2092: \emph{Analytical and Geometric Aspects of Hyperbolic Space} (Epstein
2093: ed.), Lond. Math. Soc. Lecture notes {\bf 111}, Cambridge University
2094: Press, 1987.
2095:
2096: \bibitem[CH]{chanhodgson} Ken Chan, \emph{Constructing hyperbolic 3-manifolds}, Undergraduate thesis with Craig Hodgson, University of Melbourne, 2002.
2097:
2098: \bibitem[EP]{epsteinpenner} David B.A. Epstein, Robert C. Penner, \emph{Euclidean decompositions of noncompact hyperbolic manifolds}, J. Diff. Geom. {\bf 27} (1988), 67--80.
2099:
2100: \bibitem[G1]{abeilles} Fran\c{c}ois Gu\'eritaud, \emph{Formal Markoff maps are positive}, in preparation.
2101: %{\sf les nids d'abeilles......}
2102:
2103: \bibitem[G2]{these} Franois Gu\'{e}ritaud, PhD thesis, in preparation.
2104:
2105: \bibitem[GF]{mapomme} Fran\c{c}ois Gu\'eritaud, with an Appendix by David Futer,
2106: \emph{On canonical triangulations of once-punctured torus bundles
2107: and two-bridge link complements}, arXiv:math.GT/0406242, 2004.
2108:
2109: \bibitem[J{\o}]{jorgensen} Troels J{\o}rgensen, \emph{On pairs of punctured tori}, (unfinished manuscript), in \emph{Kleinian groups and hyperbolic 3-manifolds} (Y. Komori, V. Markovic, C. Series, Editors), Lond. Math. Soc. Lecture notes {\bf 299} (2003), 183--207.
2110:
2111: \bibitem[KS]{keenseries} Linda Keen and Caroline Series, \emph{Continuity of convex hull boundaries}, Pacific J. Math. {\bf 168}, no. 1 (1995), 183--206.
2112:
2113: \bibitem[La]{lackenby} Marc Lackenby, \emph{The canonical decomposition of once-punctured torus bundles}, Comment. Math. Helv. {\bf 78} (2003), 363--384.
2114:
2115: \bibitem[Mil]{milnor} John Milnor, \emph{Hyperbolic geometry: the first 150
2116: years}, Bull. Amer. Math. Soc. {\bf 6} (1982), no. 1, 9--24.
2117:
2118: \bibitem[Min]{minsky} Yair Minsky, \emph{The classification of punctured-torus groups}, Annals of Math. {\bf 149} (1999), 559--626.
2119:
2120: %\bibitem[Ot]{otallimitedouble} Jean-Pierre Otal, \emph{Le th\'eor\`eme
2121: %d'hyperbolisation pour les vari\'et\'es fibr\'ees de dimension 3}, Ast\'erisque %{\bf 235}, SMF (1996)
2122:
2123: \bibitem[Ri]{rivin} Igor Rivin,
2124: \emph{Euclidean structures on simplicial surfaces and hyperbolic volume},
2125: Ann. of Math. {\bf 139} (1994) 553--580.
2126:
2127: \bibitem[Se]{series} Caroline Series, \emph{Thurston's bending measure
2128: conjecture for once punctured torus groups}, arXiv:math.GT/0406056,
2129: 2004.
2130:
2131: \bibitem[Th]{thurstonlimitedouble} William P. Thurston, \emph{Hyperbolic Structures on 3-manifolds, II: Surface groups and 3-manifolds which fiber over the circle}, arXiv:math.GT/9801045.
2132:
2133: \bibitem[Wa]{opti} Masaaki Wada, \emph{Opti}, a computer program for the study of punctured-torus groups: \texttt{http://vivaldi.ics.nara-wu.ac.jp/\~{}wada/OPTi/}
2134:
2135: \end{thebibliography}
2136:
2137: \begin{flushright}
2138: USC Mathematics (KAP) \\ 3620 South Vermont Avenue \\ Los Angeles, CA 90089 (USA) \\
2139: \texttt{Francois.Gueritaud@normalesup.org}
2140: \end{flushright}
2141: \end{document}
2142:
2143:
2144: 888888888888888888888888888888888888888888888888888888888
2145: 888888888888888888888888888888888888888888888888888888888
2146: 888888888888888888888888888888888888888888888888888888888
2147: 888888 %%%%%%
2148: 888888 %%%%%%
2149: 888888 JJJJ UU UU NNN NN KK KK %%%%%%
2150: 888888 JJ UU UU NNNN NN KK KK %%%%%%
2151: 888888 JJ UU UU NN NN NN KKKK %%%%%%
2152: 888888 JJ JJ UU UU NN NNNN KK KK %%%%%%
2153: 888888 JJJJJ UUUUU NN NN KK KK %%%%%%
2154: 888888 %%%%%%
2155: 888888 %%%%%%
2156: 888888888888888888888888888888888888888888888888888888888
2157: 888888888888888888888888888888888888888888888888888888888
2158: 888888888888888888888888888888888888888888888888888888888
2159:
2160:
2161: if $\frac{d}{dt}(x_t,y_t,z_t)=(\xi,\eta,\zeta)$
2162: (where by symmetry we may assume $x_t, y_t\leq \pi/2$), then
2163: \begin{eqnarray*} \textstyle{-\frac{d^2}{dt^2}\mathcal{V}(x_t,y_t,z_t)}
2164: &=&\textstyle{\frac{d^2}{dt^2} \left (\int_0^{x_t}\log 2\sin+\int_0^{y_t}\log
2165: 2\sin+\int_0^{z_t}\log 2\sin \right )}\\ &=&\xi^2\cot x_t+\eta^2\cot
2166: y_t+\zeta^2\cot z_t\\ &=&\xi^2\cot x_t+\eta^2\cot y_t +(\xi+\eta)^2\frac{1-\cot
2167: x_t\cot y_t}{\cot x_t+\cot y_t}\\ &=&\frac{(\xi+\eta)^2+(\xi\cot x_t-\eta\cot
2168: y_t)^2}{\cot x_t+\cot y_t} \geq 0. \end{eqnarray*} \end{proof}
2169:
2170: %=(p_1+p_0)(q_1-q_0)-(q_1+q_0)(p_1-p_0)
2171:
2172: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2173: Therefore the result will follow if we find a sequence $w'\in W$ such that $w'-w$ has finite support, and the tetrahedron $\Delta'_i$ defined by $w'$ has positive angles ($\mathcal{V}$ will have infinite positive derivative at $w$ in the direction of $w'$).
2174:
2175: \begin{definition}
2176: Recall the rationals $q^{\pm}_i$ opposite the Farey edge $e_i$. For each $k\in \mathbb{Z}$, consider the sequences $\phi^{k+},\phi^{k-}$ defined by
2177: $$\phi^{k+}_i:=\left \{ \begin{array}{ll} w_i & \text{if $i\geq k-1$} \\ \lambda_k(q^-_i) & \text{if $i\leq k$} \end{array} \right . ~~;~~ \phi^{k-}_i:=\left \{ \begin{array}{ll} -\lambda_k(q^+_i) & \text{if $i\geq k-1$} \\ w_i & \text{if $i\leq k$}\end{array}\right..$$
2178: \end{definition}
2179: (It is easy to check that the two definitions of $\phi^{k\pm}$ are equally valid on $\{k-1,k\}$). Intuitively speaking, $\phi^{k+}$ (resp. $\phi^{k-}$) will be the largest Fibonacci-concave function that agrees with $w$ on $[k-1,+\infty)$ (resp. $(-\infty,k]$).
2180:
2181: \begin{property}
2182: The sequences $\phi^{k+}$, $\phi^{k-}$ are Fibonacci-concave.
2183: \end{property}
2184: \begin{proof}
2185: We focus on $\phi^{k+}$. At indices $i\geq k$ it follows from Fibonacci-concavity of $w$. For $i\leq k-1$, consider the Farey triangle $\tau$ between $e_{k-1}$ and $e_k$, corresponding to the pleated surface $S_k$ (whose pleating is $\lambda_k$). The sequence $\phi^{k+}$ actually satisfies the properties $\sf{iii-iv}$ of Lemma \ref{studyphi} relevant to $\phi^+$: the proof is the same, replacing $\beta^+/\alpha^+$ by each vertex of $\tau$, and then summing with coefficients $w_{k-1}, -w_k, w_k-w_{k-1}$ (by definition (\ref{definelambdai}) of $\lambda_k$).
2186: \end{proof}
2187:
2188: \begin{property} \llabel{phiN}
2189: For every $k\in \mathbb{Z}$ we have $\nabla \phi^- \leq \nabla \phi^{k-} \leq
2190: \nabla w \leq \nabla \phi^{k+} \leq \nabla \phi^+$ and $w\leq
2191: \phi^{k\pm}\leq \phi^\pm$.
2192: \end{property}
2193: \begin{proof}
2194: Let us focus on $\phi^+$ only. For indices $i\geq k$ we have $\nabla
2195: \phi^{k+}_i=\nabla w_i\leq \nabla \phi^+_i$ and $\phi^{k+}_i =
2196: w_i\leq \phi^+_i$. For $i<k$, the downward induction hypothesis
2197: $$ \begin{array}{rcrcr} w_{i+1}&\leq & \phi^{k+}_{i+1} & \leq &
2198: \phi^+_{i+1} \\ \nabla w_{i+1}& \leq& \nabla \phi^{k+}_{i+1} & \leq & \nabla \phi^+_{i+1} \\
2199: \end{array}$$ gives $w_i \leq \phi^{k+}_i \leq \phi^+_i$ right away. We know $$\nabla \phi^{k+}_i=\left \{
2200: \begin{array}{rl} \phi^{k+}_{i+1} & \text{if $i$ is a hinge} \\
2201: \nabla\phi^{k+}_{i+1}& \text{otherwise}
2202: \end{array} \right . ~~~~~\nabla \phi^+_i=\left \{
2203: \begin{array}{rl} \phi^+_{i+1}& \text{if $i$ is a hinge} \\
2204: \nabla \phi^+_{i+1} & \text{otherwise}
2205: \end{array} \right .$$ and, by (\ref{doubleve}): $$\begin{array}{lrcrcrcrc}\text{if $i$ is a hinge,}&
2206: \nabla w_i &\leq& w_{i+1} &\leq& \phi^{k+}_{i+1} &\leq& \phi^+_{i+1}&; \\
2207: \text{otherwise,}&\nabla w_i &\leq& \nabla w_{i+1} &\leq& \nabla
2208: \phi^{k+}_{i+1}&\leq& \nabla \phi^+_{i+1}&. \end{array}$$ The last
2209: two members in each line are none but $\nabla \phi^{k+}_i$ and
2210: $\nabla \phi^+_i$. \end{proof}
2211:
2212: \begin{property} \llabel{findgoodw}
2213: Fix $k<K$ in $\mathbb{Z}$.
2214: The real sequence $\widehat{w}=\inf\{\phi^{k-},\phi^{K+}\}$ is
2215: Fibonacci-concave and satisfies $\nabla \phi^-\leq \nabla \widehat{w} \leq
2216: \nabla \phi^+$. \end{property}
2217: \begin{proof}
2218: Since $\nabla \phi^{K+}\geq \nabla w$, the sequence $\phi^{K+}-w$ is
2219: non-increasing. Therefore, if $\widehat{w}_j=\phi^{K+}_j$ (which holds for
2220: $j=K-1$), then $\widehat{w}_i=\phi^{K+}_i$ for all $i\geq j$. A similar
2221: statement holds for $\phi^{k-}$ near $-\infty$, so there exists $j\in\llbracket
2222: k, K-1 \rrbracket$ such that $\widehat{w}_i=\phi^{K+}_i$ for $i\geq j$, and
2223: $\widehat{w}_i=\phi^{k-}_i$ for $i<j$. Fibonacci-concavity of $\widehat{w}$ thus
2224: follows from the Fibonacci-concavity of $\phi^{K+}$ or $\phi^{k-}$ at every
2225: index, except $j$ and $j-1$. If $j$ is a hinge, we have
2226: $\widehat{w}_{j-1}=\phi^{k-}_{j-1}\leq \phi^{K+}_{j-1}=
2227: \phi^{K+}_j+\phi^{K+}_{j+1}=\widehat{w}_j+\widehat{w}_{j+1}$ and $\nabla
2228: \widehat{w}_{j+1}=\nabla\phi^{K+}_{j+1}\geq
2229: \nabla\phi^{k-}_{j+1}=-\phi^{k-}_{j-1}=-\widehat{w}_{j-1}$, so $\widehat{w}$ is
2230: Fibonacci-concave at $j$. If $j$ is non-hinge, we have
2231: $\widehat{w}_{j-1}=\phi^{k-}_{j-1}\leq
2232: \phi^{K+}_{j-1}=2\phi^{K+}_j-\phi^{K+}_{j+1}=2\widehat{w}_j-\widehat{w}_{j+1}$, so
2233: $\widehat{w}$ is Fibonacci-concave at $j$. The argument at $j-1$ is the
2234: same, interchanging $\phi^{K+}$ and $\phi^{k-}$: so $\widehat{w}$ is
2235: Fibonacci-concave. The property $\nabla \phi^{k-} \leq \nabla \widehat{w}
2236: \leq \nabla \phi_{K+}$ is clear from Proposition \ref{phiN} at
2237: indices $i\neq j$; and we have $\nabla
2238: \phi^{k-}_j=\phi^{k-}_{j-1}-\phi^{k-}_j\leq
2239: \phi^{k-}_{j-1}-\phi^{K+}_j\leq \phi^{K+}_{j-1}-\phi^{K+}_j=\nabla
2240: \phi^{K+_j}$: the middle term is $\phi^{k-}_{j-1}-\phi^{K+}_j=\nabla
2241: \widehat{w}_j$, hence Property \ref{findgoodw}.
2242: \end{proof}
2243:
2244: Recall our assumption that $\Delta_i$ has exactly one vanishing angle. By Proposition \ref{bendingbounded}, we can find $k<i$ small enough and $K>i$ large enough, so that $\widehat{w}_i >0$. The sequence $w^{\varepsilon}:=\inf \{\widehat{w},\varepsilon \}$ can therefore be made equal to $\varepsilon$ on an arbitrarily large \emph{finite} neighborhood of the index $i$, and $w^{\varepsilon}$ belongs to $W$ (if we assume $\varepsilon<\pi$). If $i$ is a hinge index, we can thus assume $w^{\varepsilon}_{i-1}=w^{\varepsilon}_i=w^{\varepsilon}_{i-1} = \varepsilon$, and the sequence $w':=w^{\varepsilon}$ does define a tetrahedron $\Delta'_i$ with positive angles $(\pi-\varepsilon,\varepsilon,\varepsilon)$, as wished. If $i$ is not a hinge, find two hinge indices $h<i<h'$ and make sure that $w^{\varepsilon}=\varepsilon$ on $\llbracket h-1,h'+1 \rrbracket$: it is then easy to deform $(w^{\varepsilon}_{h+1},\dots,w^{\varepsilon}_i,\dots, w^{\varepsilon}_{h'-1})$ into a strictly concave sequence $w'$ without breaking the conditions (\ref{doubleve}) defining $W$: the tetrahedron $\Delta'_i$ defined by $w'$ has positive angles, hence Proposition \ref{tricot}.
2245:
2246: %%%%%%%%%%%%%%%%%%%%
2247: The last inequality of (\ref{positivity}) can be written : $w_{i+1}<w_i+w_{i-1}$ and $w_{i-1}<w_i+w_{i+1}$. For that reason, a positive sequence $(w_i)_{i\in\mathbb{Z}}$ satisfying the last two inequalities of (\ref{positivity}) will be called \emph{strongly Fibonacci-concave} (it may exceed $\pi$). A non-negative sequence $w$ satisfying the corresponding weak inequalities is called (weakly) Fibonacci-concave.
2248:
2249: Assume $w$ is Fibonacci-concave. If $\varepsilon\geq 0$, the sequence $\inf\{w,\varepsilon\}$ is also Fibonacci-concave. If $w_k=w_{k+1}=0$ for some $k$, then $w=0$.
2250: %%%%%%
2251:
2252: For the lower bound, consider the further identities
2253: \begin{eqnarray*} \lambda_i(\gamma)&=& w_{i-1} (a\wedge s) + w_i (b \wedge s \,-\, a \wedge s) \\ &=& w_{i-1}(a\wedge s) + w_i (z\wedge s). \end{eqnarray*}
2254: Since $w_{i-1},w_i$ are not both $0$ (we assumed $w\neq 0$), this quantity is positive.
2255:
2256: %%%%%%%%%%%%%%%%%%%
2257: To an infinite end of an arbitrary Kleinian group, by the Tameness Theorem \cite{tameness}, one can associate an ending lamination, which is a projective class of measured laminations. It is natural to conjecture that the group is fully determined by the topology of the quotient manifold, plus the ending and/or pleating laminations: if all ends are infinite, this is a case of the Ending Lamination Theorem \cite{endinglamination}; if all ends are finite, it is the Pleating Lamination Conjecture, known only for quasifuchsian once-punctured torus groups \cite{series}. The mixed case is an open problem.
2258:
2259:
2260: