1: \documentclass[12pt,reqno,twoside]{amsart}

2: \usepackage{amsmath}

3: \usepackage{amssymb}

4: \usepackage[all]{xy}

5: \usepackage{epsfig}

6: \usepackage{color}

7: \title{Characterizations of lattice surfaces}

8: \author{John Smillie%\thanks{Partially supported by  NSF grant DMS-0302357

9: %  } 

10: }

11: \address{Cornell University, Ithaca, NY {\tt smillie@math.cornell.edu}}

12: \author{Barak Weiss}

13: \address{Ben Gurion University, Be'er Sheva, Israel 84105

14: {\tt barakw@math.bgu.ac.il}}

15: \font\sb = cmbx8 scaled \magstep0

16: \font\sn = cmssi8 scaled \magstep0

17: \font\si = cmti8 scaled \magstep0

18: \long\def\comjohn#1{\ifdraft{\si #1 }\else\ignorespaces\fi}

19: \long\def\combarak#1{\ifdraft{\sb #1 }\else\ignorespaces\fi}

20: \newcommand\qu{quasiunipotent}

21: \newcommand\ba{badly approximable}

22: \newcommand\hol{\mathrm{hol}}

23: \newcommand\wa{well approximable}

24: \newcommand\vwa{very well approximable}

25: \newcommand\da{Diophantine approximation}

26: \newcommand\di{Diophantine}

27: \newcommand\de{Diophantine exponent}

28: \newcommand\hs{homogeneous space}

29: \newcommand\tfae{the following are equivalent}

30: \newcommand\stl{strongly tree-like}

31: \newif\ifdraft\drafttrue

32: %for nondraft mode when typesetting this file by itself

33: %uncomment the following line:

34: \draftfalse

35: %\long\def\comdima#1{\ifdraft{\tt #1 }\else\ignorespaces\fi}

36: %\long\def\com#1{\ifdraft{\sb #1 }\else\ignorespaces\fi}

37: \newcommand\name[1]{\label{#1}{\ifdraft{\sn [#1]}\else\ignorespaces\fi}}

38: \newcommand\bname[1]{{\ifdraft{\sn [#1]}\else\ignorespaces\fi}}

39: \newcommand\eq[2]{{\ifdraft{\ \tt [#1]}\else\ignorespaces\fi}\begin{equation}\label{#1}{#2}\end{equation}}

40: \newcommand {\equ}[1]{\eqref{#1}}

41: \newcommand{\under}[2]{\underset{\text{#1}}{#2}}

42:     \newcommand{\absolute}[1] {\left|{#1}\right|}

43: 

44:     \newcommand{\grad}{\nabla}

45: \newcommand{\MM}{{\mathcal{M}}}

46: \newcommand{\HH}{{\mathcal{H}}}

47: \newcommand{\HHH}{{\mathbb{H}}}

48: \newcommand{\goth}[1]{{\mathfrak{#1}}}

49: \newcommand{\Q}{{\mathbb {Q}}}

50: \newcommand{\qsubgp}{\stackrel{\Q}{<}}

51: \newcommand{\G}{{\bf{G}}}

52: \newcommand{\compose}{{\circ}}

53: \newcommand{\U}{{\bf{U}}}

54: \newcommand{\R}{{\mathbb{R}}}

55: \newcommand{\TT}{{\mathcal{T}}}

56: \newcommand{\ii}{{\mathbf{i}}}

57: \newcommand{\Z}{{\mathbb{Z}}}

58: 

59: \newcommand{\C}{{\mathbb{C}}}

60: 

61: \newcommand{\h}{{\bf{H}}}

62: 

63: \newcommand{\epi}{<_{e}}

64: 

65: \newcommand{\obs}{<_{o}}

66: 

67: \newcommand{\E}{{\mathbf{e}}}

68: 

69: \newcommand{\se}{{\bf{S}}}

70: 

71: \newcommand{\K}{{\bf{K}}}

72: 

73: \newcommand{\N}{{\mathbb{N}}}

74: 

75: \newcommand{\F}{{\bf{F}}}

76: 

77: \newcommand{\EL}{{\bf{L}}}

78: 

79: \newcommand{\cl}{\overline}

80: 

81: \newcommand{\ww}{{\bf{w}}}

82: 

83: \newcommand{\A}{{\bf{a}}}

84: 

85: \newcommand{\vv}{{\bf{v}}}

86: 

87: \newcommand{\homeo}{{\operatorname{Homeo}}}

88: 

89: \newcommand{\Ad}{{\operatorname{Ad}}}

90: 

91: \newcommand{\GL}{\operatorname{GL}}

92: 

93: \newcommand{\Mod}{\operatorname{Mod}}

94: 

95: \newcommand{\SL}{\operatorname{SL}}

96: \newcommand{\PSL}{\operatorname{PSL}}

97: 

98: \newcommand{\ggm}{G/\Gamma}

99: 

100: \newcommand{\sln}{{{\mathcal sl}}}

101: 

102: \newcommand{\Ss}{{{\mathcal{S}}}}

103: 

104: \newcommand{\PGL}{\operatorname{PGL}}

105: 

106: \newcommand{\Lie}{\operatorname{Lie}}

107: 

108: \newcommand{\SO}{\operatorname{SO}}

109: 

110: \newcommand{\PSO}{\operatorname{PSO}}

111: 

112: %\newcommand{\ij}{1 \leq i \neq j \leq n}

113: 

114: \newcommand{\diag}{{\rm diag}}

115: 

116: \newcommand{\Aut}{{\rm AUT}}

117: 

118: \newcommand{\End}{{\rm End}}

119: 

120: \newcommand{\downto}{{\downarrow}}

121: 

122: \newcommand{\Gal}{{\rm Gal}}

123: \newcommand {\ignore}[1]  {}

124: 

125: \newcommand{\spa}{{\rm span}}

126: 

127: \newcommand{\bigplus}{+}

128: 

129: \newcommand{\interior}{{\rm int}}

130: 

131: \newcommand{\conv}{{\rm conv}}

132: 

133: \newcommand{\EE}{{\goth{E}}}

134: 

135: \newcommand{\dist}{{\rm dist}}

136: 

137: \newcommand{\diam}{{\rm diam}}

138: 

139: \newcommand{\bq}{{\mathbf{q}}}

140: 

141: \newcommand{\sameas}{\simeq}

142: 

143: \newcommand{\QQ}{{\mathcal Q}}

144: 

145: \newcommand{\QQQ}{{\til{\mathcal Q}}}

146: \newcommand{\MMM}{{\til{\mathcal{M}}}}

147: \newcommand{\PP}{{\mathcal P}_1}

148: 

149: \newcommand{\LL}{{\mathcal L}}

150: 

151: \newcommand{\avg}{{\rm Avg}}

152: 

153: \newcommand{\II}{{\mathcal E}}

154: 

155: \newcommand{\df}{{\, \stackrel{\mathrm{def}}{=}\, }}

156: 

157: \newcommand{\FF}{{\mathcal{F}}}

158: 

159: \newcommand{\s}{{\bf y}}

160: 

161: \newcommand{\x}{{\bf x}}

162: 

163: \newcommand{\vu}{{\bf u}}

164: 

165: \newcommand{\p}{{\bf p}}

166: 

167: \newcommand{\til}{\widetilde}

168: 

169: \newcommand{\supp}{{\rm supp}}

170: 

171: \newcommand{\ff}{{\mathbf{f}}}

172: 

173: \newcommand{\hh}{{\mathbf{h}}}

174: 

175: \newcommand{\T}{{\mathbf{t}}}

176: \newcommand{\Mat}{{\operatorname{Mat}}}

177: \newcommand{\NVWA}{{\rm NVWA}}

178: 

179: \newcommand{\NMVWA}{{\rm NVWMA}}

180: 

181: \newcommand{\esssup}{{\mu \mathrm{.ess \ sup}}}

182: 

183: \newcommand{\sm}{\smallsetminus}

184: 

185: \newcommand{\vre}{\varepsilon}

186: 

187: \newcommand {\Heads}[1]   {\smallskip\pagebreak[1]\noindent{\bf

188: #1{\hskip 0.2cm}}}

189: 

190: \newcommand {\Head}[1]    {\Heads{#1:}}

191: 

192: \font\comment = cmbx10 scaled \magstep0

193: 

194: \newcommand\hd{Hausdorff dimension}

195: \newcommand\nz{\smallsetminus \{0\}}

196: \newcommand{\NST}{{\mathrm{NST}}}

197: \newcommand{\NSVT}{{\mathrm{NSVT}}}

198: \newcommand{\aNSVT}{{\mathrm{aNSVT}}}

199: \newcommand{\NSMP}{{\mathrm{NSMP}}}

200: \newcommand{\LCM}{{\mathrm{LCM}}}

201: \newcommand{\LT}{{\mathrm{LT}}}

202: \newcommand{\NLC}{{\mathrm{NLC}}}

203: \newcommand{\Aff}{{\mathrm{Aff}}}

204: \newcommand{\SC}{{\mathrm{SC}}}

205: \newtheorem*{main}{Theorem}

206: 

207: \newtheorem{thm}{Theorem}[section]

208: 

209: \newtheorem{lem}[thm]{Lemma}

210: 

211: \newtheorem{prop}[thm]{Proposition}

212: 

213: \newtheorem{cor}[thm]{Corollary}

214: 

215: \newtheorem{step}{Step}

216: 

217: \newtheorem{claim}{Claim}

218: 

219: \newtheorem{remark}[thm]{Remark}

220: 

221: \newtheorem{example}[thm]{Example}

222: 

223: \newtheorem{conj}{Conjecture}

224: 

225: \newtheorem{question}{Question}

226: 

227: \newtheorem{dfn}{Definition}

228: 

229: 

230: \begin{document}

231: \maketitle

232: 

233: 

234: \begin{abstract}

235: We answer a question of Vorobets by showing that the lattice property

236: for flat surfaces is equivalent to the existence of a positive lower

237: bound for the areas of affine triangles.

238: We show that the set of affine equivalence classes of lattice surfaces

239: with a fixed positive lower bound for the areas of triangles is finite

240: and we obtain explicit bounds on its cardinality. We deduce 

241: several other characterizations of the lattice property. 

242: 

243: \end{abstract}

244: 

245: \section{Introduction}

246: 

247: 

248: Our objects of study are translation and half-translation surfaces and

249: their affine automorphism groups. These structures arise in the study 

250: of rational polygonal billiards. They also arise in Thurston's classification of

251: surface diffeomorphisms in connection with measured foliations.

252:  Isomorphic structures arise in

253: complex analysis where they are called respectively abelian differentials

254: and quadratic differentials. We will use the term flat surface for both

255: translation and half translation surfaces when it is not important to distinguish

256: between the two types. For more details on flat surfaces see

257: \cite{Vorobets, MT, zorich survey}. 

258: 

259: Let $\Aff(M)$ denote the affine automorphism group of a flat surface $M$. 

260: For a typical flat surface this group is trivial; however surfaces

261: with non-trivial automorphism groups are quite interesting. 

262:  Taking the differential of the automorphism yields a

263: homomorphism $D: \Aff(M) \to G$ with finite kernel, 

264: where $G$ is either $\SL(2, \R)$ or $\PSL(2,\R)$ depending on whether

265: $M$ is a translation or half-translation surface. The image 

266: $\Gamma_M$ of this homomorphism is called the {\em Veech group} of $M$.  

267: We say that $M$ is a {\em lattice surface} if $\Gamma_M$ is a lattice,

268: i.e. has finite

269: covolume in $G$.

270: %The structure of flat surfaces with arithmetic

271: %lattice groups was given by Gutkin  \cite{Gutkin}. 

272: In a celebrated

273: paper \cite{Veech - alternative}, Veech

274: constructed a family of lattice surfaces, and 

275: showed that lattice surfaces have striking

276: dynamical properties, in particular they satisfy the `Veech dichotomy'

277: which will be discussed below.

278: Other examples have been constructed by several authors 

279: and classifications

280: of lattice surfaces are known in special cases (see \cite{GJ, KS, Puchta,

281: Calta, McMullen}),

282: but an overall classification does not yet exist.

283: 

284: 

285: Vorobets \cite{Vorobets} found a connection between the lattice

286: property and the collection of areas of triangles in the

287: surface. A flat surface is equipped with a finite set of distinguished

288: points $\Sigma = \Sigma_M$ which contains all cone points (those with

289: cone angle not equal to $2\pi$) and may contain other points. We

290: always assume $\Sigma \neq \varnothing$, and that elements of

291: $\Aff(M)$ preserve $\Sigma$. A {\em triangle in $M$} is the image

292: of an affine map from a triangle in the plane to $M$, which takes the

293: vertices of the triangle to points in $\Sigma,$ is injective on

294: the interior of the triangle, and such that interior points do not map

295: to $\Sigma$. In particular the vertices of the triangle need not be

296: distinct. The collection of areas of triangles will be denoted by

297: $\TT(M)$. Vorobets proved that $M$ is a lattice surface if and only if

298: $\TT(M)$ is finite. Say that $M$ {\em has no small triangles} if $\inf

299: \, \TT(M)>0$. Vorobets also showed that if $M$ has no small triangles

300: then it satisfies the Veech dichotomy, and raised the question of

301: whether the no small triangles property is equivalent to the lattice

302: property. 

303: 

304: 

305: \begin{thm}

306: \name{thm: characterization}

307: A flat surface has the lattice property if and only if it has no small

308: triangles. 

309: 

310: \end{thm}

311: 

312: 

313: Let us consider only surfaces $M$ which have total area 1. For

314: $\alpha>0$, let 

315: $$\NST(\alpha) = \left\{M: \inf \, \TT(M) \geq \alpha \right \}.$$

316: Say that $M$ and $M'$ are {\em affinely equivalent} if there

317: is an affine homeomorphism between them; note that for surfaces of

318: area 1, $\mathcal{T}(M)$

319: depends only on the affine equivalence class of $M$. 

320: %Denote the set of affine equivalence classes in $\NST(\alpha)$ by $\til{

321: %\NST}(\alpha)$. 

322: 

323: 

324: \begin{thm}

325: \name{thm: NST finite}

326: For any $\alpha>0$, 

327: $\NST(\alpha)$ contains a finite number of affine equivalence classes.

328: \end{thm}

329: 

330: See  Proposition

331: \ref{prop: explicit} for an explicit bound on this number, and see

332: Proposition \ref{prop: coarea} for a bound on the sum, over all $M \in

333: \NST(\alpha),$ of the co-areas of $\Gamma_M$.

334: 

335: Besides Theorem \ref{thm: characterization}, our results yield several

336: different characterizations of lattice 

337: surfaces, which we collect in Theorem \ref{thm: TFAE} below. To state

338: them we introduce some terminology.

339: A {\em saddle

340: connection} on $M$ is a straight segment on $M$ connecting points of

341: $\Sigma$ (which need not be distinct), with no points of $\Sigma$ in

342: its interior. Let $\LL= \LL_M$ denote the set of all

343: saddle connections on $M$, and let $\LL(\theta) = \LL_M(\theta)$ denote the

344: set of saddle connections in direction $\theta$.

345: We say that $\theta$

346: is a {\em periodic direction} if

347: % $\LL(\theta)$ consists of saddle

348: %connections and 

349: each connected 

350: component of $M \sm \LL(\theta)$ is a cylinder with a waist curve in

351: direction $\theta$.

352: %We say that a singular foliation

353: %$\FF$ is {\em uniquely ergodic} if there is a unique (up to scaling)

354: %collection of measures defined on all arcs transverse to $\FF$ and

355: %invariant under holonomy along leaves. 

356: Veech \cite{Veech - alternative} showed that a lattice surface

357: satisfies the following dichotomy:

358: \begin{itemize}

359: \item[I.] If $\LL(\theta) \neq \varnothing$ then $\theta$ is

360: a periodic direction. 

361: \item[II.] If $\LL(\theta) = \varnothing$ then $\FF_\theta$ is 

362: uniquely ergodic. 

363: \end{itemize}

364: This dichotomy does not characterize the lattice property (see

365: \cite{SW -- Veech}), but a modification of property I introduced by

366: Vorobets does. 

367: Namely, say

368: that $M$ is {\em uniformly completely periodic} if there is $s>0$ such

369: that 

370: each $\theta$ for which $\LL(\theta) \neq \varnothing$ is a periodic

371: direction, and the ratio of 

372: lengths of any two segments in $\LL(\theta)$ does not exceed

373: $s$. We will show that this property characterizes the

374: lattice property. Moreover, for a lattice surface in $\NST(\alpha)$ we

375: will provide an effective estimate for $s$ in terms of $\alpha$, see

376: Proposition \ref{prop: s bound}. 

377: %, and say that $M$ is 

378: %{\em uniformly completely periodic} if it is $s$-uniformly completely

379: %parabolic for some $s$.

380: 

381: A periodic direction

382: $\theta$ is called {\em parabolic} if the moduli of

383: all the cylinders are commensurable, and $M$ is called {\em uniformly

384: completely 

385: parabolic} if it is uniformly completely periodic, with all periodic

386: directions parabolic. 

387: %Say that 

388: %that $M$ is $s$-{\em completely parabolic} if 

389: %each $\theta$ for which $\LL(\theta) \neq \varnothing$ is a parabolic

390: %direction. 

391: 

392: Another characterization involves the set of holonomy vectors of

393: saddle connections. Associated with $\delta \in \LL$ is a vector

394: hol($\delta$) in the plane 

395: of the same length and direction as $\delta$. We set 

396: $$\mathrm{hol}(M) = \{ \mathrm{hol}(\delta) : \delta \in \LL \}.

397: $$

398: This is a discrete $\Gamma_M$-invariant subset of the plane which has attracted

399: considerable attention. 

400: For $\beta>0$, we let 

401: $$\NSVT(\beta) = \left\{ M : \inf \{ |v_1 \wedge v_2| : v_i \in \mathrm{hol}(M), v_1 \wedge v_2

402: \neq 0 \} \geq \beta \right \},

403: $$

404: and say that $M$ has {\em no small virtual triangles} if it belongs

405: to some $\NSVT(\beta)$. 

406: \ignore{The {\em virtual triangle

407: spectrum} of $M$ is 

408: $$

409: \mathcal{VT} (M) =\{ |v_1 \wedge v_2| : v_i \in \mathrm{hol}(M) \}.

410: $$

411: We let 

412: $$

413: \NSVT(\alpha) = \{M : \inf \, \mathcal{VT}(M) \geq \alpha \}

414: $$

415: (NSVT stands for `no small virtual triangles'). }

416: We will show that having no

417: small virtual 

418: triangles is also equivalent to the lattice property. 

419: %Moreover for

420: %computing explicit bounds it is more convenient to work with

421: %$\til \NSVT(\alpha).$ 

422: 

423: 

424: Two other characterizations involve the dynamics of the $G$-action on

425: the space of flat surfaces. 

426: Associated with $M$ is the topological data consisting of the

427: underlying surface, the number of points in $\Sigma$ and the

428: associated cone angles, and whether or not the corresponding

429: foliations are orientable. The set of all $M$ sharing this data is

430: a noncompact orbifold called a {\em stratum}. The restriction of the

431: action of $G$ to the subgroup 

432: $\{g_t\}$, where 

433: $$

434: g_t = \left(\begin{array}{cc}

435: e^{t/2} & 0 \\ 0 & e^{-t/2} \end{array} \right)

436: $$ 

437: is called the {\em geodesic flow}. 

438: 

439: %\end{itemize}

440: 

441: \begin{thm}

442: \name{thm: TFAE}

443: The following are equivalent for a flat surface $M$:

444: \begin{itemize}

445: \item[(i)]

446: $M$ is a lattice surface.

447: \item[(ii)]

448: $M$ is uniformly completely periodic. 

449: \item[(iii)]

450: $M$ is uniformly completely parabolic.

451: \item[(iv)]

452: $\left|\mathcal{T}(M) \right| < \infty$.

453: \item[(v)]

454: The set of triangles for $M$ consists of finitely many $\Aff(M)$-orbits.

455: \item[(vi)]

456: $M$ has no small triangles. 

457: \item[(vii)]

458: $\{u \wedge v : u, v \in \hol(M)\}$ is a discrete set of numbers. 

459: \item[(viii)]

460: For any $T>0$, the set 

461: $$\{(\xi,\eta) \in \LL_M \times \LL_M :

462: |\hol(\xi) \wedge

463: \hol(\eta)| < T\}$$

464: contains finitely many $\Aff(M)$-orbits. 

465: \item[(ix)]

466: $M$ has no small virtual triangles. 

467: \item[(x)]

468: The $G$-orbit of $M$ is closed.

469: \item[(xi)]

470: There is a compact subset $K$ of the stratum containing $M$ such

471: that for any $g \in G$, the geodesic orbit of $gM$ intersects $K$. 

472: \end{itemize}

473: \end{thm}

474: 

475: Many of these implications are due to Vorobets \cite[\S

476: 6]{Vorobets}. The main new implication is (vi) $\implies$ (i). Our

477: first 

478: proof of this implication is sketched in \cite{toronto}. 

479: 

480: The proofs of Theorems \ref{thm: characterization} and

481: \ref{thm: NST finite} are similar. Vorobets showed that if $M$

482: has no small triangles any saddle connection direction is parabolic,

483: so associated with two non-parallel saddle connections are

484: a pair of cylinder decompositions of $M$, each with cylinders of

485: commensurable moduli. 

486: Moreover a bound on triangle areas leads to an upper bound on certain 

487: combinatorial data associated with two cylinder

488: decompositions. Arguments of Thurston and Veech show that such 

489: combinatorial data determine a flat surface up to affine

490: equivalence. This immediately 

491: yields Theorem \ref{thm: NST finite}. 

492: 

493: To derive Theorem \ref{thm:

494: characterization} we introduce the {\em spine} $\Pi$ of a translation

495: surface $M$; 

496: this is a $\Gamma_M$-invariant tree in $\HHH$, whose edges are

497: labeled by two saddle connections on $M$, and correspond to the surfaces

498: affinely equivalent to $M$ for which these saddle

499: connections are simultaneously shortest. There is a retraction $\rho:

500: \HHH \to \Pi$ taking a surface in which all shortest saddle

501: connections are parallel, to an affinely equivalent surface which has

502: shortest saddle connections in two or more directions. For each edge $e$ of

503: $\Pi$, $\rho^{-1}(e)$ is a finite-area domain in $\HHH$, and $e$

504: is associated with a pair of cylinder decompositions. A bound on

505: triangle areas bounds the combinatorial data and thus bounds the

506: number of  edges in $\Pi/\Gamma_M$, giving a

507: bound on the area of $\HHH/\Gamma_M$. 

508: 

509: Theorems \ref{thm: characterization} and \ref{thm: NST finite} suggest a

510: natural ordering on the set of lattice surfaces. For a lattice

511: surface $M$, let $\alpha(M)$ denote the largest $\alpha$ for which $M

512: \in \NST(\alpha)$. 

513: Then all (affine equivalence classes of) lattice surfaces may be

514: written as 

515: %$$

516: \eq{eq: alphaM}{

517: M_1, M_2, \ldots, \ \ \mathrm{with \ \ }\alpha(M_1) \geq \alpha(M_2)

518: \geq \cdots,

519: }

520: %SS

521: and a similar ordering exists with the NSVT condition instead of the NST

522: condition. 

523: 

524: It would be desirable to have an algorithm which,

525: given $\alpha >0$, lists all surfaces in $\NST(\alpha)$ or

526: $\NSVT(\alpha)$. 

527: Our analysis yields an explicit finite set of

528: surfaces, presented in terms of Thurston-Veech combinatorial

529: data, which contains all the above lattice surfaces, and implicitly, an

530: algorithm for distinguishing the lattice from the non-lattice

531: surfaces. This is our motivation for providing effective proofs 

532: and explicit estimates for the cardinality of certain finite sets. 

533: We suspect our explicit estimates are far from optimal and

534: believe additional theoretical work would be required in order to

535: create an algorithm which is practically feasible. 

536: 

537: 

538: 

539: 

540: In the interest of presenting the

541: simplest formulae, our estimates may be presented in terms of either

542: triangle or virtual triangle areas. We do not present 

543: estimates involving the genus of $M$ or the stratum

544: containing $M$, both because these are harder to obtain, and because

545: the same effective search will produce the lattice surfaces in all

546: genera. 

547: 

548:  The relation between

549: the no small triangles and no small virtual triangles properties is

550: described in the following:

551: \begin{thm}

552: \name{thm: nst vs nsvt}

553: $\NSVT(\alpha/2) \subset \NST(\alpha) \subset \NSVT(2\alpha

554: e^{-1/(2\alpha e)})$. 

555: 

556: \end{thm}

557: 

558: 

559: Without conducting computer searches, we are able to

560: determine $M_1, \ldots, M_7.$ They are all

561: arithmetic, and comprise $\NST\left(\frac16\right).$ 

562: 

563: 

564: \ignore{

565: 

566: 

567: An interesting class of flat surfaces arises from

568: rational polygonal billiards. To conclude this introduction, we

569: reinterpret the no small triangle property for billiards in a

570: polygon. This makes it possible to recognize a billiard table which

571: corresponds to a lattice surface, without making reference to the

572: associated $G$-action. 

573: 

574: Given a polygon $\mathcal{P}$ in

575: the plane, one can study the {\em billiard flow} on $\mathcal{P}$; see

576: \cite{MT} for a precise definition. 

577: If all angles of $\mathcal{P}$ are rational multiples of $2\pi$ there is a naturally

578: associated flat surface $M_{\mathcal{P}}$ and a finite-to-one map

579: $M_{\mathcal{P}} \to \mathcal{P}$, and we say that

580: $\mathcal{P}$ is a {\em lattice 

581: polygon} if $M_{\mathcal{P}}$ is a lattice surface. We make the

582: assumption that the {\em pre-images of vertices of $\mathcal{P}$ are

583: contained in $\Sigma_{M_{\mathcal{P}}}$}; see \S 8 for a discussion.  

584: Veech showed for example that the regular polygons are

585: lattice polygons, which implies via the Veech dichotomy that any billiard orbit

586: is either closed or uniformly distributed. 

587: %Our work gives a 

588: %characterization of lattice polygons which does not make

589: %explicit reference to the associated flat surface.

590: 

591: 

592: A {\em vertex connection} on a polygon $\mathcal{P}$ is a billiard

593: trajectory $\sigma$ 

594: which begins and ends at a vertex. We denote its length by

595: $L=L(\sigma)$. 

596: Let $\phi:[0,L]\to \mathcal{P}$ be a parametrization of $\sigma$, and let

597: $\delta(t)$ be the minimum of the distances from the point $\phi(t)$ 

598: to the vertices of the table. Denote the local minima of $\delta$ by

599: $t_0=0,t_1, \ldots, t_k=L$. We say that $\sigma$ {\em has close

600: approaches} if $k>1$, and define 

601: $$D(\sigma) = \min \{\delta(t_i): i=1, \ldots, k-1\}.$$

602: We say that $\mathcal{P}$ has a {\em linear bound on close 

603: approaches} if there is a constant $C >0$ such that for all vertex

604: connections $\sigma$ which have close approaches, $D(\sigma)\ge C/ L(\sigma).$

605: 

606: 

607: 

608: 

609: \begin{cor}

610: \name{cor: billiard reformulation}

611: A rational polygon $\mathcal{P}$ is a lattice polygon if and only if it has

612: a linear bound on close approaches.

613: \end{cor}

614: 

615: As another application, we show that the no small virtual triangles

616: property implies a quantitative closing lemma for geodesic segments on the

617: flat surface. A geodesic segment on a flat surface is a straight line segment

618: in each chart, or a finite concatenation of such segments and saddle

619: connections parallel to the segments.

620: A closing lemma asserts that given a geodesic trajectory

621: which almost closes up, there is a nearby closed geodesic. 

622: We have:

623: 

624: \begin{cor}\name{cor: closing lemma}

625: Suppose $M$ is a flat surface with a bound of $\beta$ for the areas of virtual 

626: triangles.

627: If $\sigma$ is a geodesic trajectory of length $L$, whose endpoints are a

628: distance $\vre$ apart, with %$L > \diam(M)$ and 

629: $(L+\vre)\vre < \beta/8$, then there is a closed

630: geodesic on $M$ contained in a $2\vre$-neighborhood of $\sigma.$ 

631: \end{cor} 

632: 

633: }

634: 

635: \medskip

636: {\bf Acknowledgements.} We thank Alex Eskin for drawing our attention

637: to \cite{Vorobets} at an early stage of this work, and we thank Yaroslav Vorobets for 

638: useful discussions. The authors gratefully acknowledge support from NSF grant

639: DMS-0302357, BSF grant 2004149 and ISF grant 584/04. 

640: 

641: 

642: \section{Basics}

643: %In this section we set some

644: %notation, and collect standard results. 

645: %

646: %\subsection{Notation}

647: We review some definitions here. For more details we refer the reader to 

648: \cite{MT, Vorobets, zorich survey}.

649: Throughout this paper, $M$ denotes a connected oriented surface with a flat

650: structure. When there is no danger of confusion, we will also denote

651: by $M$ the underlying topological surface. We denote the genus of $M$

652: by $g$. 

653: A half-translation structure may be

654: thought of as an equivalence class of atlases 

655: of charts $(U_{\alpha}, \varphi_{\alpha})$ covering all but a

656: non-empty finite

657: set $\Sigma = \Sigma_M$, such that

658: the transition functions $\R^2 \to \R^2$ are of the form $\vec{x}

659: \mapsto  \pm \vec{x} + \vec{c}$, and such that around each $\sigma \in

660: \Sigma$ the charts glue together to form a cone point with cone angle

661: $2\pi(r+1)$, where $r =r_{\sigma}\in \{ -\frac12,   \frac12, 1,

662: \frac32, 

663: \ldots\}$. 

664: A translation

665: structure is similarly defined, with the  

666: requirement that the transition functions are of the form $\vec{x}

667: \mapsto \vec{x} + \vec{c}$. 

668: Points in $\Sigma$ are called {\em singularities}. 

669: Several authors permit singularities $\sigma$ 

670: for which $r_\sigma =0$; such points are called {\em removable

671: singularities} or {\em marked points}. Except in \S\ref{section:

672: simplest}, we will always 

673: assume that our singularities are not removable. 

674: 

675: 

676: An orientation-preserving homeomorphism $\varphi: M_1 \to M_2$ which is

677: affine in each chart is 

678: called an {\em affine isomorphism}, or an {\em affine automorphism} if

679: $M_1=M_2$. We denote the set of affine automorphisms of $M$ by

680: $\Aff(M)$. An affine isomorphism whose linear part is $\pm 

681: \mathrm{Id}$ is called an {\em translation equivalence}. 

682: The map $D :\Aff(M) \to G$ which assigns to

683: $\varphi$ its linear part has a finite kernel, consisting of 

684: translation equivalences of $M$. 

685: Two atlases are considered equivalent if there is a translation

686: equivalence betweeen them. 

687: 

688: 

689: There is a standard {\em orientation

690: double cover} construction which associates to each half-translation

691: surface $M$ a translation surface $M'$ with a branched degree two

692: translation cover $M' \to M$; thus many statements about flat surfaces

693: can be reduced to statements about translation surfaces. 

694: 

695:  A standard analogue of the Gauss-Bonnet formula (see \cite{Vorobets})

696: says that 

697: \eq{eq: Gauss Bonnet}{\sum r_{\sigma} = 2g-2.}

698: Summing the total angle at all the singularities we obtain the number 

699: \eq{eq: defn tau}{

700: \tau=\tau(M) = \sum_\sigma 2 \pi(r_\sigma+1) = 2\pi\left(2g-2+ |\Sigma|

701: \right). 

702: }

703: Since each triangle in a triangulation of $M$ with vertices in

704: $\Sigma$ contributes a total 

705: angle of $\pi$, the number of triangles in such a triangulation is

706: exactly $\tau/\pi$. 

707: 

708: 

709: A flat surface inherits a

710: Euclidean area form from 

711: the plane and we normalize our surfaces by assuming that each surface has unit area.

712: An affine automorphism of $M$ is a  

713: self-homeomorphism which is affine in each chart.

714: 

715: 

716: 

717: 

718: For a flat surface $M$ let $\vec{r}_M = \left(r_{\sigma} \right)_{\sigma \in

719: \Sigma_M}$. The set of all translation equivalence classes of 

720: (half-) translation surfaces of a given combinatorial type, namely those for

721: which the data $\vec{r}_M$ is fixed, 

722: is called a {\em stratum}. 

723: Each stratum is equipped with a structure of

724: an affine orbifold, which is locally modeled on $H^1(M, \Sigma; \R^2)$

725: (in the case of translation surfaces), or an appropriate subspace of

726: the $H^1(\til M, \til \Sigma; \R^2)$ (in the case of half-translation

727: surfaces, where $\til M, \til \Sigma$ are the orientation double cover

728: of $M, \Sigma$). 

729: %%One property of this topology is that for

730: %a flat surface $M$ and a saddle connection $\delta$ on $M$, any $M'$

731: %sufficiently close to $M$ has a saddle connection $\delta'$ with

732: %%$\mathrm{hol}(\delta')$ close to $\mathrm{hol}(\delta)$. See \cite{MS}

733: %for more details.

734: 

735: There is an action of $G$ on $\HH$ by post-composition on each

736: chart in an atlas. It follows from the above that $\Gamma_M = \{g \in

737: G: gM=M\}.$ 

738: We define 

739: \[

740: %\begin{split}

741: %g_t & = \left(\begin{array}{cc}

742: %e^{t/2} & 0 \\ 0 & e^{-t/2} \end{array} \right), \ \ \ 

743: r_{\theta}=

744: \left(\begin{array}{cc}

745: \cos \theta & -\sin \theta \\

746: \sin \theta & \cos \theta

747: \end{array}

748: \right),\ \ \ \ \ 

749: h_s 

750: %& 

751: = \left(\begin{array}{cc} 1 & s \\ 0 & 1 

752: \end{array}

753: \right),

754:  \ \ \ \ \, \ \til h_s = \left(\begin{array}{cc} 1 & 0 \\ s & 1 

755: \end{array}

756: \right),

757: %\end{split}

758: \]

759: which we view as one-parameter subgroups of $G$.

760: \ignore{

761: Let 

762: $$\mathcal{F}_{\theta}(q) = \mathcal{F}(r_{-\theta}q).$$

763: A singular foliation $\mathcal{F}$ is called {\em minimal} if every

764: nonsingular leaf is dense. It is called {\em uniquely ergodic} if

765: Lebesgue measure is the unique measure on segments transverse to

766: $\mathcal{F}$, invariant under holonomy along leaves.

767: }

768: 

769: 

770: \ignore{

771: A Fuchsian group is a discrete subgroup of $G$. Fixing a haar measure

772: $\mu$ on $G$, we define the covolume of $\Gamma$ in $G$ as

773: $\mu(\Omega)$, where $\Omega$ is any fundamental 

774: domain for the action of $\Gamma$ on $G$, and denote the covolume by

775: $\bar{\mu}(\Gamma)$. It is easily checked that this is

776: well-defined (independent of the choice of $\Omega$), and that if $q$

777: and $q'$ are affinely 

778: equivalent then $\bar{\mu}(\Gamma_q) = \bar{\mu}(\Gamma_{q'}).$ If

779: $\bar{\mu}(\Gamma) < \infty$ 

780: then $\Gamma$ is called a lattice.  

781: 

782: 

783: For a Fuchsian group $\Gamma$ we now define a set of cusp

784: areas. 

785: An element of a Fuchsian group is called {\em parabolic} if it is

786: conjugate to $h_1$. An infinite cyclic subgroup of $G$

787: generated by a parabolic element is called {\em parabolic}.

788: Suppose that $\Gamma$ contains a parabolic subgroup $P$ and is {\em 

789: non-elementary}, that is,  

790: not a finite extension of an abelian group. Suppose also that $P$ is

791: {\em maximal}, i.e.\ not properly 

792: contained in a parabolic subgroup of $\Gamma$. Choose an element $g

793: \in G$ such that 

794: \begin{equation}

795: \name{eq: prop of g}

796: gPg^{-1} = \left \langle h_1  \right \rangle,

797: % \ \ \ \mathrm{where} \ 

798: %h_1= \pm \left(\begin{matrix} 1 & 1 \\ 0 & 

799: %1\end{matrix} \right),

800: \end{equation}

801:  and relabeling, replace $P$ and $\Gamma$ by

802: $gPg^{-1}$ and $g\Gamma g^{-1}$ respectively. Let $\HH$ be the complex upper half

803: plane, let $\ii = \sqrt{-1}$, let $\mathcal{C}_t$ be the image of the disk $\left\{z \in \HH:

804: \left|z-\frac{t}{2}\ii \right | < \frac{t}{2} \right\}$ in $\HH/P$, let

805: $\varphi: \HH/P \to 

806: \HH/\Gamma$ be the natural map, and let 

807: \begin{equation}

808: \name{eq: defn t0}

809: t_0=t_0(\Gamma, P) = \sup \{t>0 : \varphi|_{\mathcal{C}_t} \mathrm{\ is \ injective }

810: \}.

811: \end{equation}

812: It is easily seen (see Propositions

813: \ref{prop: cusp}, \ref{prop: conjugation}) that the set in the right

814: hand side of \equ{eq: defn 

815: t0} is nonempty and bounded above, so that $t_0$ is well-defined, and that $t_0$ does not

816: depend on the choice of $g$ in \equ{eq: prop of g} and satisfies $t_0(\Gamma,

817: P) = t_0(x\Gamma x^{-1}, xPx^{-1})$ for $x \in G$. We call $t_0$ the

818: {\em cusp area} of $P$ in $\Gamma$; a simple computation shows that it

819: is equal to the hyperbolic area of $\mathcal{C}_{t_0}$.

820: }

821: 

822: By a {\em direction} we mean an element of $P(\R^2)$ (resp. $S^1$) in

823: the case of half-translation (resp. translation) surfaces. There is a

824: natural action of $G$ on the set of directions. 

825: 

826: A {\em cylinder} for $M$ is a topological annulus on the surface which is

827: isometric to $\R /c \Z \times (0,h)$, where $c$ is the {\em

828: circumference} of the cylinder and $h$ is its {\em height}.

829: We say that the cylinder is {\em horizontal (vertical)} if the angle

830: between the horizontal 

831: direction and the direction of the waist curve is $0$

832: (resp. $\pi/2$). Note that if the cylinder is vertical then the height

833: actually measures its horizontal width. 

834: A cylinder is {\em maximal} if it is not contained in a

835: larger cylinder, and this implies that both of its boundary components

836:  contain singularities. 

837: We define $\LL(\theta)$ to be the collection of closed saddle connections

838: in direction $\theta$. Recall that if $\theta$ is a periodic direction

839: then $M$ has a {\em cylinder decomposition in direction

840: $\theta$} i.e., the complement of $\LL(\theta)$ is a union of

841: cylinders. 

842: An automorphism $\varphi \in \Aff(M)$ is called {\em parabolic} if

843: $D\varphi \in \Gamma_M$ is a parabolic matrix. 

844: The 

845: {\em inverse modulus} of a cylinder as above is $\mu=w/h$. 

846: We say that real numbers $\mu_1, \ldots, \mu_k \in \R$ are {\em commensurable} if

847: $\mu_i/\mu_j \in \Q$ for all $i,j \in \{1, \ldots, k\}$. If this

848: holds, we denote by $\mu=\LCM(\mu_1, \ldots, \mu_k)$ the smallest

849: positive number which is an integer multiple of all the $\mu_i$, and

850: we call $(n_1, \ldots, n_k)$, where $n_i = \mu /\mu_i$, the

851: {\em Dehn twist vector} corresponding to the 

852: cylinder decomposition. Note that by definition $\gcd(n_1, \ldots,

853: n_k)=1$. This terminology is motivated by the fact if $\varphi \in

854: \Aff(M)$ is such that $D\varphi$ is parabolic and fixes the direction

855: $\theta$, then there is a cylinder

856: decomposition in direction $\theta$ with 

857: the inverse moduli of the cylinders commensurable.

858: Conversely, given a decomposition of $M$ into cylinders for which

859: the $\mu_i$ are commensurable, one can construct an associated parabolic 

860: affine automorphism. We say that $\varphi$ is {\em simple}

861: if $\varphi$ fixes the saddle connections $\LL(\theta)$

862: pointwise. This implies that each cylinder is taken to itself and that

863: the map on each cylinder is topologically a certain number of Dehn

864: twists.  Let $n_1, \ldots, n_k$ be a collection of natural numbers and

865: let $\mu \neq 0$ satisfy the equation $n_j\mu_j=\mu$. Then there is a simple parabolic automorphism

866: $\varphi$ with derivative $D\varphi=r_\theta h_\mu r_{-\theta}$ which induces $n_j$ Dehn

867: twists in the cylinder $C_j$. %We will say that $(m_1, \ldots, m_k)$ is

868: %the {\em Dehn twist vector} corresponding to the cylinder

869: %decomposition $C_1, \ldots, C_k$ if $m_i \mu_i = \LCM(\mu_1, \ldots,

870: %\mu_k)$ for each $i$, so that $\gcd(m_1, \ldots, m_k)=1$. 

871: %This particular automorphism generates

872: %the group of simple parabolic automorphisms with axis in direction

873: %$\theta$. 

874: %Let us call the collection of these Dehn twist numbers,

875: %$(n_1,\ldots,n_r)$, the {\em Dehn twist vector} corresponding to

876: %$\varphi$. 

877: %We say that $\varphi$ is {\em primitive} if it is simple

878: %and is not of the form $\psi^k$ for some simple parabolic affine

879: %automorphism $\psi$ and $k \geq 2$. Assuming that $\varphi$ we see

880: %that the Dehn twist vector is 

881: %completely determined by the affine structure of $M$. Moreover, it may

882: %be reconstructed completely from the ratios $n_i/n_j$ \combarak{give

883: %an argument using the fact that $\varphi$ is primitive.}

884: 

885: \section{Cylinder decompositions}

886: Vorobets showed that cylinder decompositions arise in connection with

887: the no small triangles condition: 

888: 

889: \begin{prop}[Vorobets]

890: \name{prop: Vorobets parabolic}

891:  If $M$ has no small triangles then any

892: direction $\theta$ for which $\LL(\theta) \neq \varnothing$ is

893: parabolic.

894: \end{prop}

895: 

896: In addition Vorobets obtained bounds on the ratios of lengths of saddle connections in 

897: $\LL(\theta)$ and the Dehn twist numbers $n_j$. We will review

898: Vorobets' arguments here and use them to relate the no small triangles

899: and the no small virtual triangle conditions. Where possible we will

900: improve the constants that arise. 

901: 

902: 

903: \begin{prop}

904: \name{prop: s bound}

905: If $M \in \NST(\alpha)$ then $M$ is $s$-uniformly periodic for $s =

906: \min \left\{e^{1/(2\alpha e)},  (2(r-1)\alpha)^{1-r}\right\}$, where

907: $r$ is the maximal number of cylinders in a cylinder decomposition on

908: $M$.  

909: \end{prop}

910: \begin{proof}

911: Suppose $\theta$ is a saddle connection direction for $M \in

912: \NST(\alpha)$. We know from Proposition \ref{prop: Vorobets parabolic} that

913: $\theta$ is a periodic direction on $M$. Let $\sigma, \sigma' \in

914: \LL_M(\theta)$. Suppose first that $\sigma, \sigma'$ are on the

915: boundary of the same cylinder $C$, of area $A$. Let $h$ be the height

916: of $C$, and let $x, x'$ be the lengths of $\sigma,

917: \sigma'$. Then $x h \leq A$ since the circumference of $C$ is at

918: least $x$, and $x' h \geq 2\alpha$ since $C$ contains a triangle of

919: area $x'h/2$. This implies that

920: \eq{eq:3}{

921: %$$ 

922: \frac{x}{x'} = \frac{xh}{x'h} \leq \frac{A}{2 \alpha}.

923: }

924: %$$

925: Now suppose $\sigma, \sigma'$ are not on the boundary of the same

926: cylinder. Since $M$ is connected, there is a chain of cylinders 

927: $C_1,\ldots C_{j}$ such that $\sigma = \sigma_1$ is on the boundary of

928: $C_1$ and $\sigma'=\sigma_{j+1}$ is on the boundary of $C_j$, and $C_i$ and $C_{i+1}$

929: meet along a segment $\sigma_i$ of length $x_i$ for $i=1, \ldots,

930: j$. Applying \equ{eq:3} $j$ times we have  

931: \begin{equation}

932: \name{eq:product}

933: \frac{x}{x'}=\prod_{i=1}^{j}\frac{x_i}{x_{i+1}}\le\prod_{i=1}^{j}\frac{A_i}{2

934: \alpha}.

935: \end{equation}

936: Since $M$ has area 1 and the cylinders $C_i$ are disjoint,

937: \begin{equation}

938: \name{eq:area}

939: \sum A_i\le 1

940: \end{equation}

941: The maximum of the right hand side of \equ{eq:product} subject to the

942: constraint \equ{eq:area}  

943: occurs when when $A_i=1/j$, so that 

944: $$\frac{x}{x'}\le (2 j \alpha)^{-j}. $$

945: 

946: The maximum of  $(2 j\alpha)^{-j}$ as a function of $j$ occurs when

947: $j=1/(2 \alpha e)$.

948: If $1/(2\alpha e)\ge r-1$ then the maximum value of ${2j\alpha}^{-j}$ occurs when

949: $j=r-1$. In this case the maximum value of ${2 j\alpha}^{-j}$ is $(2(r-1)\alpha)^{1-r}$.

950: If $1/(2\alpha e)\le r-1$ then the maximum value of $(2j\alpha)^{-j}$ occurs when $j=r-1$.

951: In this case the maximum value of $(2j\alpha)^{-j}$ is $e^{1/(2\alpha e)}$.

952: \end{proof}

953: 

954: \ignore{

955: \combarak{seems to me we can prove Prop. 3.3 using 3.4 and can do

956: without 3.2 entirely.}

957: 

958: 

959: 

960: \begin{prop}

961: \name{prop:height bounds}

962: Let $M\in \NST(\alpha)$. Consider a cylinder decomposition of $M$ with

963: $r$ cylinders. Then the ratio of heights of cylinders is bounded above

964: by 

965: $$\min \left\{ ((r-1)\alpha)^{1-r}, e^{1/(\alpha e)} \right\}.$$

966: \end{prop}

967: 

968: \begin{proof}

969: We begin by considering the

970: case of neighboring cylinders. Let $C$ and $C'$ be cylinders of height $h$ and $h'$ which meet along a

971: segment $\sigma$ of length $L$. 

972: Since $C$ is a maximal cylinder the boundary component of $C$ opposite

973: $\sigma$ 

974: contains a point $p\in\Sigma$. Thus we can construct a triangle

975: $\Delta$ in $C$ which has $\sigma$ as a base and $p$ as a vertex. The

976: area $A$ of $\Delta$ is $hL/2$. Similarly we can construct a triangle

977: $\Delta'\subset C'$ with area $A'=h'L/2$.  By our assumption both $A$

978: and $A'$ are at least $\alpha.$ This implies that 

979: \eq{eq:3}{

980: \frac{h}{h'} = \frac{A}{A'} \leq \frac{A}{\alpha}.

981: }

982: 

983: %The lower bound on areas of triangles gives:

984: %

985: %\begin{equation}

986: %\name{eq:1}

987: %\alpha\le A, A'.

988: %\end{equation}

989: %

990: %and

991: %

992: %\begin{equation}

993: %\name{eq:2}

994: %\alpha\le \frac{h'L}{2}\le A'

995: %\end{equation}

996: %

997: %Inverting the terms in equation \ref{eq:2} gives

998: %\begin{equation}

999: %\name{eq:3}

1000: %\frac{1}{A'}\le \frac{2}{h'L}\le \frac{1}{\alpha}

1001: %\end{equation}

1002: %

1003: %Taking the product of equations \ref{eq:1} and \ref{eq:3}  gives:

1004: %

1005: %\begin{equation}

1006: %\name{eq:3}

1007: %\frac{\alpha}{A'}\le h/h' \le A/\alpha

1008: %\end{equation}

1009: 

1010: Now we want to bound the ratio of heights of two different cylinders

1011: $C$ and $C'$. Since $M$ is connected there is a chain of cylinders

1012: $C_1,\ldots C_{j+1}$ with $C=C_1$ and $C_{j+1}=C'$ so that $C_i$ and $C_{i+1}$

1013: meet along a segment. Applying \equ{eq:3} $j$ times we have 

1014: \begin{equation}

1015: \name{eq:product}

1016: \frac{h}{h'}=\prod_{i=1}^{j}\frac{h_i}{h_{i+1}}\le\prod_{i=1}^{j}\frac{A_i}{\alpha}.

1017: \end{equation}

1018: Since $M$ has area 1 and the cylinders $C_i$ are disjoint,

1019: \begin{equation}

1020: \name{eq:area}

1021: \sum A_i\le 1

1022: \end{equation}

1023: The maximum of the right hand side of \equ{eq:product} subject to the

1024: constraint \equ{eq:area}  

1025: occurs when when $A_i=1/j$, so that 

1026: $$\frac{h}{h'}\le (j \alpha)^{-j}$$

1027: 

1028: The maximum of  $(j\alpha)^{-j}$ as a function of $j$ occurs when $j=1/(e\alpha)$.

1029: If $1/(e\alpha)\ge r-1$ then the maximum value of ${j\alpha}^{-j}$ occurs when

1030: $j=r-1$. In this case the maximum value of ${j\alpha}^{-j}$ is $((r-1)\alpha)^{1-r}$.

1031: If $1/(e\alpha)\le r-1$ then the maximum value of ${j\alpha}^{-j}$ occurs when $j=r-1$.

1032: In this case the maximum value of ${j\alpha}^{-j}$ is $e^{1/(\alpha e)}$.

1033: \end{proof}

1034: }

1035: We deduce the following statement, which implies Theorem 

1036: \ref{thm: nst vs nsvt}. 

1037: \begin{prop}

1038: \name{prop: nst vs nsvt}

1039: If $M \in \NST(\alpha)$ and the number of cylinders in a decomposition of

1040: $M$ is at most $r$, then $M \in \NSVT(\beta)$ where 

1041: $$\beta = 2\alpha \max \left\{e^{-1/(2\alpha e)}, \left(2(r-1)\alpha

1042: \right)^{r-1} \right\}.$$ 

1043: 

1044: \end{prop}

1045: \begin{proof}Let $M\in

1046: \NST(\alpha)$. Let $\xi$ and $\eta$ be nonparallel saddle

1047: connections on $M$. By applying an element of $G$ we may assume that $\xi$

1048: is horizontal and $\eta$ vertical. Since $M \in \NST(\alpha)$ there is a

1049: decomposition of $M$ into horizontal cylinders, so there is a saddle

1050: connection $\xi'$ parallel to $\xi$ making an angle of $\pi/2$ with

1051: $\eta$ at one of the endpoints of $\eta$. By Proposition \ref{prop: s

1052: bound} we have $x' \leq sx,$ where $x, x'$ are the lengths of $\xi,

1053: \xi'$ and $s = \min \left\{e^{1/(2\alpha e)},  (2(r-1)\alpha)^{1-r}\right\}$. 

1054: 

1055: 

1056: Consider the right triangle $\Delta$ in the plane with horizontal and

1057: vertical sides the same length as $\xi', \eta$. If there is an

1058: isometric copy of $\Delta$ embedded in $M$ under an affine mapping

1059: sending its edges to $\xi', \eta$, we would have $|\xi' \wedge \eta|

1060: /2  = \mathrm{area} \, \Delta \geq \alpha$. If there is no such isometric

1061: copy of $\Delta$ then $M$ contains an isometric copy of a triangle

1062: contained in $\Delta$, and again $|\xi' \wedge \eta| \geq 2 \alpha$.

1063: Therefore 

1064: $$\left|\xi \wedge \eta \right| \geq \frac1s |\xi' \wedge \eta| \geq

1065: \frac{2\alpha}{s} = \beta.

1066: $$

1067: \end{proof}

1068: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1069: 

1070: 

1071: \begin{remark}

1072: When applying Propositions %\ref{prop:height bounds}, 

1073: \ref{prop: s bound}, \ref{prop: nst vs

1074: nsvt} it is useful to 

1075: have a bound for the maximal number of cylinders in a cylinder

1076: decomposition on $M$. 

1077: A standard bound of $2g + |\Sigma| -2 $ can be found in e.g. \cite[proof of Lemma,

1078: p. 302]{KMS}. A better bound is 

1079: $g+i+\left \lfloor \frac{j}{2} \right \rfloor -1,$

1080: where $g$ is the genus of $M$, and $i$ (resp. $j$) is the number of

1081: even (resp. odd) -angled singularities of $M$. A proof of this bound,

1082: due to the first-named author, is given 

1083: in \cite{Naveh} for the case of translation surfaces; the same proof

1084: works for half-translation surfaces.

1085: \end{remark}

1086: 

1087: \ignore{

1088: 

1089: An automorphism $\varphi \in \Aff(M)$ is called {\em parabolic} if

1090: $D\varphi \in \Gamma_M$ is a parabolic matrix. 

1091: One way in which cylinder decompositions arise is in connection with parabolic automorphisms. 

1092: 

1093: \begin{prop}[Vorobets] If $M$ has no small triangles then the cylinder

1094: decomposition that arises in each saddle connection direction is

1095: parabolic. 

1096: \end{prop}

1097: }

1098: For the cylinder decompositions that arise via Proposition \ref{prop: Vorobets

1099: parabolic}, Vorobets

1100: bounded the number and size of the Dehn twist numbers in terms of the

1101: lower bound for areas of triangles. We give our version here. 

1102: 

1103: 

1104: \begin{prop}

1105: \name{prop:diagonal count}

1106: The number of  Dehn twist vectors for a 

1107: cylinder decomposition of a surface in $\NST(\alpha)$ into  $r$ cylinders

1108: is bounded above by:

1109: $$(8\alpha)^{2(1-r)}(1-2\log 8 \alpha)^{r-1}

1110: %(2\alpha)^{2(1-r)}(1-2\log 2\alpha)^{2(r-1)}

1111: .$$

1112: \end{prop}

1113: 

1114: 

1115: \begin{proof}

1116: Let $M=C_1 \cup \cdots \cup C_r$ be a cylinder decomposition, where the cylinders have

1117: inverse moduli $\mu_1,\ldots, \mu_r$. We can recover the Dehn twist

1118: vector $\vec{n}$ from the numbers $\mu_i/\mu_j$. 

1119: 

1120: 

1121: We first recall Vorobets' argument which gives a restriction on

1122: the ratio of moduli of neighboring cylinders. Say that we have a pair of cylinders

1123: $C_i$ and $C_j$ with an edge in common. Let $h$ and $c$ be the height and

1124: circumference of a cylinder and let $A$ be its area. %from here

1125: As shown in \cite[proof of Prop. 6.2]{Vorobets}, applying the $G$-action and

1126: Dehn twists in the cylinders one can find a triangle 

1127: in $C_i \cup C_j$ with base $h_i$ and height in $\displaystyle{w_0 + \Z \mu_i h_j +

1128: \Z w_j = w_j\left(\frac{\mu_i}{\mu_j} \Z + \Z \right)}$ for some $w_0$. 

1129: Since for $\mu_i/\mu_j  = p/q$ we have that $\displaystyle{\Z +

1130: \frac{\mu_i}{\mu_j}\Z}$ is $1/2q$-dense in $\R$, we have by our hypothesis

1131: \begin{equation*}

1132: q\le\frac{h_iw_j}{4\alpha}.

1133: \end{equation*}

1134: If we interchange the roles of the cylinders we get 

1135: $

1136: p\le h_jw_i / (4\alpha),

1137: $

1138: and multiplying the equations gives

1139: \eq{eq: *}{

1140: pq\le \frac{h_iw_ih_jw_j}{16\alpha^2}=\frac{A_iA_j}{16\alpha^2}.

1141: }

1142: The maximum of the right hand side of \equ{eq: *} under the constraint

1143: $A_i+A_j \leq 1$ is attained when $A_i = A_j =1/2$ so we get 

1144: $$pq \leq \frac1{64\alpha^2}.$$

1145: 

1146: 

1147: 

1148: To count the possible values we note that the number of pairs of natural numbers $(x,y)$ that

1149: satisfy $xy\le K$ is bounded by the area under the graph of $y=\min(K,\log x)$ from $x=0$ to $x=K$

1150: and this area is $K(1+\log K)$.

1151: Thus the number of possible ratios of moduli of neighboring cylinders is bounded by

1152: $\displaystyle{

1153: %\begin{equation*}

1154: \frac{1}{64\alpha^2}\left(1+\log \frac{1}{64\alpha^2}\right).

1155: %\end{equation*}

1156: }$

1157: 

1158: The connections between cylinders can be recorded in a graph whose

1159: vertices are cylinders and whose edges correspond to cylinders

1160: which share a segment. Since $M$ is connected this graph is connected. 

1161: In order to recover $\vec{n}$ it suffices to

1162: know the ratios of moduli for a collection of edges that span a maximal tree

1163: in this graph. The number of edges in a maximal tree is $r-1$, and we

1164: obtain our bound of 

1165: \[

1166: %\begin{equation*}

1167: \prod_{edges}\frac{1}{64\alpha^2}\left(1+\log

1168: \frac{1}{64\alpha^2}\right )= (8\alpha)^{2(1-r)}(1-2\log 8\alpha)^{r-1}.

1169: %\end{equation*}

1170: \]

1171: \end{proof}

1172: 

1173: We will also need a similar bound in terms of $\beta,$ when $M \in

1174: \NSVT(\beta)$. Note that this bound is independent of $r$. 

1175: \begin{prop}

1176: \name{prop: twist bounds beta}

1177: If $M \in \NSVT(\beta)$ and $\vec{n} = (n_1, \ldots, n_r)$ is a Dehn

1178: twist vector for a parabolic direction on $M$ with a decomposition

1179: into $r$ cylinders, 

1180: then for any $i, j$:

1181: \eq{eq: twist bounds beta}{

1182: pq \leq \frac{1}{64\beta^2}, \ \ \ \ \  \mathrm{where} \ \ \frac{n_i}{n_j} =

1183: \frac{p}{q} \ \ \mathrm{and} \, \ \gcd(p,q)=1.

1184: }

1185: \end{prop}

1186: 

1187: \begin{proof}

1188: By renumbering assume $i=1,\ j=2$. Let $C_1, C_2$ be the corresponding

1189: cylinders and let $\delta_1, \delta_2$ be saddle connections passing

1190: across the cylinders, so that $\hol(\delta_i) = (x_i, h_i)$, where

1191: $|h_i|$ is the height of $C_i$ and $|x_i|$ is no greater than the

1192: circumference $c_i$ of $C_i$. By

1193: applying Dehn twists in $C_1, C_2$ and the $G$-action, we find an

1194: affinely equivalent surface for which $\hol(\delta_1) = (0, h_1)$ and

1195: $\hol(\delta_2) = (x, h_2)$ with $|x| \leq c_1/2q$. This implies

1196: $$\beta \leq |\hol(\delta_1) \wedge \hol(\delta_2)| = |xh_2| \leq 

1197: \frac{c_1h_2}{2q}.$$

1198: Interchanging the roles of $C_1, C_2$ we obtain $\beta \leq

1199: \frac{c_2h_1}{p}$ so that 

1200: \eq{eq: Lagrange}{

1201: pq \leq \frac{A_1A_2}{4\beta^2},

1202: }

1203: where $A_i$ is the area of $C_i$. 

1204:  The

1205: maximum of the right hand side of \equ{eq: Lagrange} subject to the

1206: requirement $A_1+A_2 \leq 1$ is obtained when $A_1 = A_2 = 1/2$, and

1207: we obtain 

1208: \equ{eq: twist bounds beta}.

1209: \end{proof}

1210: 

1211: 

1212: \section{The spine and a criterion for finite covolume}

1213: Given a flat surface $M$ of area 1 we can consider the collection of all

1214: area 1 surfaces $M'$ affinely equivalent to $M$. An affine equivalence

1215: $f: M \to M'$ determines an element $g = Df \in G$, whose coset

1216: $g\Gamma_M$ in $G/\Gamma_M$ depends only on $M'$. This establishes an

1217: identification of $G/\Gamma_M$ with its affine equivalence class.  

1218: The Veech group $\Gamma = \Gamma_M$ is the isotropy

1219: group of $M$ so we can identify this family with

1220: $G/\Gamma$. Identifying two flat surfaces which are isometric, we find

1221: that the collection of isometry classes of surfaces affinely equivalent to

1222: $M$ can be identified with $\PSO(2, \R)\backslash G/\Gamma$ or

1223: $\HHH/\Gamma$ where $\HHH$ is the hyperbolic plane. The image of this

1224: quotient in moduli space is also called the Teichm\"uller 

1225: disk associated to $M$. 

1226: 

1227: We define $\lambda(M)$ to be the infimum of lengths of saddle

1228: connections in $M$. Since the collection of lengths of saddle

1229: connections is discrete this infimum is achieved and is positive. The

1230: function $\lambda$ depends only on the isometry class of $M$ so it is

1231: well defined on $\HHH/\Gamma$. It is also useful to view it as a

1232: $\Gamma$-invariant function on $\HHH$. 

1233: 

1234: The function $\lambda$ is bounded above.

1235: 

1236: \begin{prop} \name{prop: go bears}

1237: If $\tau = \tau(M)$ is as in \equ{eq: defn tau} then 

1238: $\lambda(M)\le\sqrt{2/\tau}.$

1239: \end{prop}

1240: 

1241: \begin{proof} Write $\lambda$ for $\lambda(M)$. Let $M_0\subset M$

1242: consist of points $p\in M$ for which the distance from $p$ to a point

1243: of $\Sigma$ is less than $\lambda/2$, so that $p$ is closest to a

1244: unique $q\in \Sigma$. If we fix a $q\in\Sigma$  

1245: then the set of points in $M_0$ closest to $q$ has area $c_q\lambda^2/2$ where

1246: $c_q$ is the cone angle at $q$. By \equ{eq: defn tau}, 

1247: the area of $M_0$ is $\tau\lambda^2/2$. Since the total area 

1248: of $M$ is 1 we have $\lambda\le\sqrt{2/\tau}$.

1249: \end{proof}

1250: 

1251: 

1252: We say a saddle connection $\sigma$ {\em has minimal length} if

1253: $\ell(\sigma)=\lambda(M)$. The number of minimal length saddle

1254: connections in $M$ is finite and is generically one. 

1255: Let $\Pi$ be the subset of $\HHH$ corresponding to surfaces affinely

1256: isomorphic to $M$ which

1257: have at least two non-parallel minimal length saddle connections. This

1258: is a locally finite geodesic subcomplex of

1259: $\HHH$ which we call the {\em spine} of $M$. 

1260: A related but distinct construction is the tesselation studied in Veech \cite{Veech

1261: - spine} and Bowman \cite{Bowman}.  

1262: 

1263: The edges of $\Pi$ correspond to surfaces with minimal length saddle

1264: connections in exactly two directions. These edges are

1265: geodesic segments in $\HHH$. Indeed, if $\xi$ and $\eta$ are minimal length saddle

1266: connections in distinct directions on $M$ then the vectors $v=\hol(\xi)$

1267: and $w=\hol(\eta)$ have the same length. We can rotate the translation

1268: structure so that the vector $v+w$ is horizontal and the vector $v-w$

1269: is vertical. In these coordinates $v=(x,y)$ and $w=(x,-y)$ for some

1270: $x, y \in \R$. If we apply the geodesic flow to $M$ the holonomy vectors

1271: of these saddle connections with respect to $g_t(M)$ are

1272: $(e^{t/2}x,e^{-t/2}y)$ and $(e^{t/2}x,-e^{-t/2}y)$, so the

1273: vectors continue to have the same length. In fact this geodesic in

1274: $\HHH$ is exactly the set of translation structures for which $\xi$

1275: and $\eta$ have the same length. The edge in $\Pi$ that contains $M$

1276: is a segment contained in this geodesic. 

1277: 

1278: 

1279: Clearly $\Pi$ is $\Gamma_M$-invariant. 

1280: The vertices of $\Pi$ correspond to surfaces with minimal length

1281: saddle connections in three or more directions. 

1282: We will use $\Pi$ to get a condition for the lattice property.

1283: 

1284: \begin{prop}

1285: \label{prop: area est} The area of $\HHH/\Gamma_M$ is bounded above by

1286: $2\pi N$ where $N$ is the number of edges in $\Pi/\Gamma_M$. 

1287: \end{prop}

1288: 

1289: \begin{proof}

1290: Write $\Gamma = \Gamma_M$. We define a retraction $\rho: \HHH/\Gamma \to

1291: \Pi/\Gamma$. We will describe $\rho$ as a $\Gamma$-invariant map

1292: from $\HHH$ to $\Pi$. Let $z \in \HHH$. It corresponds to a set of

1293: isometric translation surfaces, i.e. to $\SO(2, \R) \cdot M'$ for some

1294: $M'$ in the $G$-orbit of $M$. If $M'$ has minimal length saddle

1295: connections in two or more non-parallel directions then $M'$

1296: represents a point in $\Pi$ 

1297: and we define $\rho(z)=z$. If all minimal length saddle connections on

1298: $M$ are in the same direction, by rotating the coordinate system on

1299: the surface we can assume 

1300: that this direction is horizontal. Now apply the geodesic flow $g_t$

1301: to the rotated surface. The length of the shortest saddle connection

1302: $\sigma$ in the 

1303: surface $g_t(M')$ is $e^{t/2}$ times the original length of

1304: $\sigma$. As long as $g_t(M')$ is in the complement of $\Pi$, $\sigma$

1305: remains a minimal length segment. Since the length of the shortest

1306: curve in $g_t(M')$ is bounded above, for some $t_0$ we have $g_{t_0}(M')

1307: \in \Pi$, and we define $\rho(z)$ to be the isometry class of $g_{t_0}(M')$. 

1308: 

1309: We now show that the set of all points which retract to a given

1310: point corresponding to the surface $M_0$ in $\Pi$ is a union of geodesic

1311: rays, one for each direction of a minimal length saddle connection

1312: in $M_0$. Pick a 

1313: saddle connection $\sigma$ of minimal length in $M_0$.  

1314: By rotating the coordinate system assume that $\sigma$ is

1315: horizontal. Consider the collection of surfaces that we get by flowing

1316: by $g_{-t}$ for $t>0$. The matrix for $g_{-t}$ contracts the

1317: horizontal direction more than any other direction so in each of these

1318: surfaces $g_{-t}M_0$ the saddle connection $\sigma$ still has minimal length

1319: and is shorter than all saddle connections in other directions. It

1320: follows that for each of these surfaces the retraction takes $g_{-t}M_0$

1321: to $M_0$. 

1322: 

1323: Pick an edge $e$ of $\Pi$. There are minimal length saddle

1324: connections in two directions. Pick one direction and let $\sigma$ be

1325: a saddle connection in this direction.  

1326: The collection of surfaces which retract to $e$ and 

1327: correspond to the direction of $\sigma$ in $\Pi$ is a union of

1328: geodesic rays all of which are asymptotic to the same point on the

1329: boundary of $\HHH$ (which corresponds to the direction of $\sigma$). It

1330: follows that the set of points that retract to $e$ consists of two

1331: geodesic triangles each with one vertex at infinity. 

1332: Since the area of a geodesic triangle is bounded above by $\pi$ the

1333: total area of $\HHH/\Gamma$ is bounded above by $2\pi N$.  

1334: \end{proof}

1335: 

1336: \begin{remark}

1337: The existence of the retraction $\rho: \HHH \to \Pi$ implies that

1338: $\Pi$ is homotopy-equivalent to $\HHH$, i.e. a tree. 

1339: \end{remark}

1340: 

1341: As we explained each edge $e$ in $\Pi$ is a geodesic segment

1342: corresponding to the surfaces for which two saddle 

1343: connections, say $\xi$ and $\eta$ have the same length. There is a

1344: unique 

1345: point on the geodesic containing $e$ corresponding to a surface $M_0$

1346: where the 

1347: saddle connections $\xi$ and $\eta$ have equal length and are perpendicular. It is

1348: conceivable that $M_0$ is not in $e$, but nevertheless we have: 

1349: 

1350: \begin{lem}\name{lem: conceivable}

1351: In the surface $M_0$ the common length of $\xi$ and $\eta$ is no more than

1352: $\sqrt{2/\tau},$ where $\tau$ is as in \equ{eq: defn tau}. 

1353: \end{lem}

1354: 

1355: \begin{proof} Let $M'$ correspond to a point in

1356: $e$. By Proposition \ref{prop: go bears}, in this surface the lengths

1357: of $\xi$ and $\eta$ are bounded above by 

1358: $\sqrt{2/\tau}$. Since the $g_t$-action preserves 

1359: $|\xi \wedge \eta| = \ell(\xi) \ell(\eta) \sin \theta$, where $\theta$

1360: is the angle between $\xi$ and $\eta$,  

1361: the minimum of $\ell(\xi)$  is

1362: attained when $\xi$ and $\eta$ are perpendicular. Thus the lengths at

1363: $M_0$ are no greater than

1364: the lengths at $M'$, and we can use Proposition \ref{prop: go bears}. 

1365: \end{proof}

1366: 

1367: We say that a translation surface $M$ is in {\em standard form} if it has

1368: vertical and horizontal saddle connections, and if the shortest

1369: horizontal and vertical saddle connections have the same length and

1370: this length is bounded by $\sqrt{2/\tau(M)}$. 

1371: 

1372: \begin{cor}\name{cor: standard}

1373: The area of $\HHH/\Gamma_M$ is at most $2\pi E$ where $E$ is the

1374: number of standard form surfaces affinely equivalent to $M$. 

1375: \end{cor}

1376: ~\qed

1377: %\begin{proof}

1378: %\end{proof}

1379: \section{The complexity of pairs of cylinder decompositions}

1380: 

1381: A pair of cylinder decompositions in distinct directions determine a

1382: decomposition of the surface into parallelograms. We measure the

1383: complexity of the pair by counting the number of parallelograms in

1384: this decomposition. We will show that a lower bound on the area of

1385: virtual triangles allows us to find a pair of cylinder decompositions

1386: for which we can bound the complexity. 

1387: 

1388: 

1389: 

1390: 

1391: 

1392: \begin{prop}

1393: \name{prop: rectangle number bound}

1394: Suppose  $M \in \NSVT(\beta_0)$ and $\xi$ and $\eta$ are saddle

1395: connections on $M$ with $|\hol(\xi)\wedge\hol(\eta)|=\beta \ge \beta_0$. 

1396: Then the number of rectangles in the pair of cylinder

1397: decompositions determined by $\xi$ and $\eta$ is no more than

1398: $\beta/\beta_0^2$. 

1399: \end{prop}

1400: 

1401: \begin{proof} By changing coordinates using an element of $G$ we may assume

1402: that $\xi$ is horizontal and $\eta$ is vertical, and they have the

1403: same length. The product of the

1404: lengths is $\beta$ so both lengths are 

1405: $\sqrt{\beta}$. Let $h$ be the height of a horizontal cylinder. There

1406: are singularities on the top and bottom of this cylinder and a saddle

1407: connection $\sigma$ between them. The quantity

1408: $|\hol(\sigma)\wedge\hol(\xi)|$ is just $h\sqrt{\beta}$. By

1409: assumption $|\hol(\sigma)\wedge\hol(\xi)|\ge \beta_0$ 

1410: so $h\ge\beta_0/\sqrt{\beta}$. A similar

1411: calculation for the width of the vertical cylinders gives

1412: $w\ge\beta_0/\sqrt{\beta}$. It follows that the area of any rectangle

1413: is at least $\beta_0^2/\beta$. Since the total area of $M$ is 1, the

1414: number of cylinders is at most  $\beta/\beta_0^2$.  

1415: \end{proof}

1416: 

1417: 

1418: \begin{cor} 

1419: \name{cor: stand form equiv class} A surface in $\NSVT(\beta)$ is

1420: affinely equivalent to a

1421: standard form surface with at most $1/\beta$ rectangles. 

1422: \end{cor}

1423: 

1424: \begin{proof} 

1425: This follows from Proposition \ref{prop: rectangle number bound} with

1426: $\beta = \beta_0$.  

1427: \end{proof}

1428: 

1429: \begin{cor} 

1430: \name{cor: stand form area est}

1431: A standard form surface in $\NSVT(\beta)$ has a decomposition into at

1432: most  $2/(\tau \beta^2)$ rectangles. 

1433: \end{cor}

1434: 

1435: \begin{proof} Let $\xi$ and $\eta$ be horizontal and vertical saddle

1436: connections of common length at most $\sqrt{2/\tau}. $ Then 

1437: $|\hol(\xi)\wedge\hol(\eta)| \le 2/\tau$. By Proposition \ref{prop:

1438: rectangle number bound}, the 

1439: number of rectangles in the corresponding pair of cylinders

1440: decomposition is bounded above by $2/(\tau\beta^2)$. 

1441: \end{proof}

1442: 

1443: %Our next objective is to show that each combinatorial type corresponds

1444: %to a finite number of affine structures. We do this by controlling the

1445: %moduli of the horizontal and vertical cylinders. 

1446: 

1447: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1448: 

1449: 

1450: \section{Pairs of cylinder decompositions}

1451: \name{section: pairs of parabolics}

1452: In this section we review ideas of Thurston \cite{Thurston} and Veech

1453: \cite{Veech - alternative} that make it possible to bound the number

1454: of standard form surfaces in $\NSVT(\beta).$ To this end assume $M$ is

1455: a flat surface for which the horizontal and vertical directions

1456: are periodic. 

1457: %is

1458: %%in standard form, with saddle connections of equal 

1459: %length no more than $\sqrt{2/\tau(M)}$ in the horizontal and vertical

1460: %directions, and moreover that the horizontal and vertical directions

1461: %are parabolic. Note that if $M \in \NSVT(\beta)$ the latter condition is

1462: %satisfied by Proposition \ref{prop: Vorobets parabolic}. 

1463: The intersections of 

1464: the horizontal and vertical cylinders divide the surface into

1465: rectangles. Let us label the rectangles with numbers $1 , \ldots,

1466: \ell$ and call the resulting surface a {\em labeled surface}. Let

1467: $S_\ell$ denote the permutation group on $\ell$ 

1468: symbols. Assume first that $M$ is a translation surface, so that each

1469: rectangle has a well defined top, bottom, left and right 

1470: edge. For $k \in \{1, \ldots, \ell\}$ let $\sigma_1(k)=k'$ if the

1471: $k'$th rectangle is attached to the right side of the $k$th

1472: rectangle. Let $\sigma_2(k)=k'$ if the $k'$th rectangle is attached to

1473: the top side of the $k$th rectangle. As in \cite{EO1} we observe that

1474: connectedness of $M$ is equivalent to the condition that group generated by

1475: $\sigma_1$ and $\sigma_2$ acts transitively on $\{1, \ldots, \ell\}$.

1476: The collection of horizontal cylinders corresponds to the set of

1477: cycles of $\sigma_1$ and the collection of vertical cylinders

1478: corresponds to the set of cycles of $\sigma_2$. The orders of the

1479: singular points of $M$ is determined by the cycle structure of the

1480: commutator of $\sigma_1$ and $\sigma_2$. %We refer to the permutations

1481: %$\sigma_1, \sigma_2$ as the {\em gluing pattern} for the two 

1482: %cylinder decompositions. 

1483: If $M$ is a half-translation surface, a gluing pattern can be defined

1484: in a similar way. For example this can be defined using the

1485: orientation double cover. We will omit the details. 

1486: 

1487: We can also reverse this process. Say that $\sigma_1$ and

1488: $\sigma_2$ are elements of $S_\ell$, for which 

1489: the group

1490: generated by $\sigma_1$ and $\sigma_2$ acts transitively on $\{1,

1491: \ldots, \ell\}$. Let $\vec{a}^{(1)},

1492: \vec{a}^{(2)}$ be vectors with positive real entries, indexed by the

1493: cycles of $\sigma_1, \sigma_2$. 

1494: Then there is a labeled translation surface $M$ where the heights of

1495: the horizontal (vertical) cylinders are recorded in $\vec{a}^{(1)}$

1496: (resp. $\vec{a}^{(2)}$); indeed, let $R_1, \ldots, R_\ell$ be

1497: (labeled) rectangles where, if $k$ belongs to the $i$th (resp. $j$th)

1498: cycle of $\sigma_1$ ($\sigma_2$) then the height (width) of

1499: $R_k$ is $a^{(1)}_i$ (resp $a^{(2)}_j$). We see that this

1500: surface is determined uniquely 

1501: as a labeled surface. 

1502: 

1503: We say that a labeled surface is {\em normalized} if the horizontal

1504: and vertical directions are parabolic, and the shortest saddle

1505: connections in the horizontal and vertical direction have equal

1506: length. Associated with such a surface is a pair of Dehn twist 

1507: vectors corresponding to the horizontal and vertical cylinder

1508: decompositions. We have:

1509: 

1510: \begin{prop}[Veech] \name{prop: parameters determine surface}

1511: The pair of permutations $(\sigma_1, \sigma_2)$ and the Dehn twist

1512: vectors uniquely determine $M$ as a normalized labeled translation surface. 

1513: \end{prop}

1514: 

1515: \begin{proof}

1516: By the preceding discussion it suffices to show that the permutations

1517: and Dehn twist vectors uniquely determine the vectors

1518: $\vec{a}^{(d)}, \ d=1,2.$ The Dehn twist vectors prescribe the ratios

1519: of inverse moduli of cylinders in a labeled surface; we will show that

1520: determining these ratios uniquely determines $\vec{a}^{(d)}$ up to two

1521: free variables. The assumptions that $M$ is normalized and has area 1

1522: will be used to get a unique solution. 

1523: 

1524: Let $B = (b_{ij})$ be an integer matrix where $i$ ranges over the cycles

1525: of $\sigma_1$ and $j$ over the cycles of $\sigma_2$, and $b_{ij}$ is

1526: the number of elements in the intersection of the corresponding cycles.

1527: Define 

1528: $$\vec{c}^{(1)} = B \vec{a}^{(2)} \ \ \ \mathrm{and \ }  \ \vec{c}^{(2)} =B^t

1529: \vec{a}^{(1)}.$$

1530: Note that on a labeled surface, the circumference of the $j$th horizontal

1531: cylinder is the sum of 

1532: the widths of the rectangles contained in it. This is just $\sum

1533: b_{ij}a^{(2)}_j$. Thus, if $\vec{a}^{(1)}$ contains the heights of horizontal cylinders

1534: on a labeled surface, then $\vec{c}^{(2)}$ contains the circumferences of

1535: vertical cylinders and vice versa. 

1536: 

1537: Now let $A_d = 

1538: \diag(\vec{n}^{(d)})$, i.e. an integer diagonal matrix, indexed by

1539: cycles of $\sigma_d$, containing the Dehn twist data. 

1540: In order for the surface determined by $\vec{a}^{(1)}, \vec{a}^{(2)}$

1541: to be parabolic in the horizontal and vertical directions, with the

1542: prescribed Dehn twist vectors, the inverse

1543: modulus $\mu^{(d)}_i$ of the $i$th cylinder must satisfy 

1544: $n^{(d)}_i\mu^{(d)}_i=\mu^{(d)}$. Here 

1545: $\mu^{(1)}$ and $\mu^{(2)}$ are variables.

1546: The relationship between inverse moduli and Dehn twist

1547: numbers gives:

1548: $$\mu^{(d)} \vec{a}^{(d)}=  A_d \vec{c}^{(d)}.$$

1549: Putting these matrix equations together gives 

1550: $$\mu^{(1)} \mu^{(2)} \vec{a}^{(1)} = A_1B A_2 B^t \vec{a}^{(1)},

1551: $$

1552: so $\vec{a}^{(1)}$ is an eigenvector of $E_1= A_1BA_2B^t$ corresponding

1553: to the eigenvalue $\mu^{(1)} \mu^{(2)}$. The

1554: assumption that $\langle \sigma_1, \sigma_2 \rangle$ acts

1555: transitively on $\{1,\ldots,\ell\}$ implies that $E_1$ is 

1556: an irreducible non-negative matrix. By the uniqueness of a positive

1557: eigenvector for an irreducible non-negative matrix \cite{Gantmacher},

1558: $\vec{a}^{(1)}$ is determined, up to a positive scalar, by the matrix

1559: $E_1$. Similarly $\vec{a}^{(2)}$ is uniquely 

1560: determined up to scaling by $E_2 = A_2B^tA_1B$.  

1561: 

1562:  The shortest horizontal

1563: (vertical) 

1564: saddle connection is a sum of some of the entries in $\vec{a}^{(2)}$

1565: (resp. $\vec{a}^{(1)}$). Requiring these sums to be equal implies that

1566: the scaling of $\vec{a}^{(1)}$ determines the scaling of

1567: $\vec{a}^{(2)}$. Rescaling both vectors by the same factor

1568: $\gamma$ changes the area of the surface

1569: by $\gamma^2$. There is a unique positive value of $\gamma$ which produces a

1570: surface of area 1. 

1571: \end{proof} 

1572: 

1573: %\begin{proof}[Proof of Proposition \ref{prop: parameters determine surface}]

1574: %The permutations $\sigma_1$ and $\sigma_2$ describe how the

1575: %rectangles are glued together to create the surface $M$. The cycle

1576: %structure of the commutator $[\sigma_1, \sigma_2]$ also determines the

1577: %number and cone angle of the  singularities. Thus the

1578: %permutations and the vectors $\vec a_d$ determine the translation

1579: %structure% up to label preserving isomorphism

1580: %. By Proposition

1581: %\ref{prop: twists determine heights},

1582: %the vectors $\vec{a}_d$ are determined by the Dehn twist vectors. 

1583: %\end{proof}

1584: 

1585: \begin{cor}\name{cor: viii}

1586: If $M \in \NSVT(\beta)$ then condition (viii) of Theorem \ref{thm:

1587: characterization} holds. 

1588: 

1589: \end{cor}

1590: 

1591: \begin{proof}

1592: Consider two saddle connections $\xi$ and $\eta$ on $M$ with $|\hol(\xi)

1593: \wedge \hol(\eta)| = \beta_0 <T$. By Propositions \ref{prop: Vorobets

1594: parabolic} the directions of $\xi$ and $\eta$ are parabolic directions

1595: on $M$, by

1596: Proposition \ref{prop:diagonal count} the Dehn twist vectors are

1597: bounded, and by Proposition \ref{prop: rectangle number bound} the

1598: number of rectangles in the corresponding pair of cylinder

1599: decompositions is bounded, hence so are the possibilities for the

1600: corresponding permutations. For any two such pairs $(\xi_i, \eta_i), \

1601: i=1,2$, there is $g \in G$ such that $g\xi_1= \xi_2$ and $g\eta_1 =

1602: \eta_2$. If the corresponding Dehn twist vectors and permutations

1603: are the same, then $gM$ is affinely equivalent to $M$ by Proposition

1604: \ref{prop: parameters determine surface}, i.e. $g \in \Gamma_M$. Thus

1605: the numbers of $\Gamma_M$-orbits of such pairs $(\xi, \eta)$ is finite. 

1606: \end{proof}

1607: 

1608: 

1609: %We now give the proof of Theorem \ref{thm: characterization} and Theorem \ref{thm: NST finite}.

1610: %%%%%%%%%%%%%%%%%%%%%%%%%% Insert closed orbit argument here.

1611: 

1612: 

1613: \section{Counting cylinder intersection patterns}

1614: In this section we will obtain a combinatorial formula which will

1615: bound the number of affine equivalence classes of

1616: translation surfaces in

1617: $\NSVT(\beta)$ and the sums of the co-areas of the Veech groups of

1618: surfaces in $\NSVT(\beta)$. The first bound will imply Theorem

1619: \ref{thm: NST finite}

1620: and the second will imply Theorem \ref{thm: characterization}, since

1621: the case of half-translation surfaces follows via the orientation

1622: double cover. 

1623: %One may obtain better bounds for

1624: %half-translation surfaces by applying a similar analysis; we

1625: %omit the details. 

1626: 

1627: Let $X_\ell$ be the set of pairs of

1628: permutations $(\sigma_1,\sigma_2)$ in $S_\ell$ for which the group

1629: $\langle \sigma_1,\sigma_2\rangle$ acts transitively on the set

1630: $\{1,\ldots,\ell\}$. As we have seen such a pair together with the

1631: Dehn twist vectors determine a normalized labeled surface 

1632: decomposed into $\ell$ rectangles. Since the

1633: horizontal and vertical cylinders of the surface correspond to cycles of

1634: $\sigma_1$ and 

1635: $\sigma_2$, the Dehn twist vector $\vec{n}_j$ may be viewed as a function

1636: from the cycles of $\sigma_j$ to the natural numbers. Let $\mathcal

1637: Y(\ell,\beta)$ denote the set of quadruples

1638: $(\sigma_1,\sigma_2,\vec{n}_1,\vec{n}_2)$ where $(\sigma_1,\sigma_2)\in

1639: X_\ell$ and $\vec{n}_j$ are Dehn twist vectors satisfying the bound in

1640: \equ{eq: twist bounds beta}, in particular, compatible with $M$ being

1641: in $\NSVT(\beta)$.

1642: 

1643: A permutation $\lambda \in S_\ell$ acts on the set of labels $\{1,

1644: \ldots, \ell\}$, and this induces an action on $(\sigma_1, \sigma_2,

1645: \vec{n}_1, \vec{n}_2)$ by simultaneously conjugating

1646: $\sigma_j$ 

1647: and changing the domains of $\vec{n}_j$. If $\lambda$ fixes the 

1648: quadruple $(\sigma_1, \sigma_2,

1649: \vec{n}_1, \vec{n}_2)$ then re-arranging the rectangles of the

1650: corresponding cylinder decompositions gives rise to a translation

1651: equivalence of $M$. Moreover the relabeling operation gives the entire

1652: translation equivalence class, since any translation equivalence of

1653: a surface preserves the horizontal and vertical

1654: directions, thus preserves the cylinder decompositions and takes

1655: rectangles to rectangles. 

1656: 

1657: This gives:

1658: \begin{prop} 

1659: \name{prop: order bound}

1660: The cardinality of any $S_\ell$-orbit in $\mathcal{Y}(\ell, \beta)$ 

1661: is at least $(\ell -1)!$ 

1662: \end{prop}

1663: 

1664: \begin{proof}

1665: For $y = (\sigma_1, \sigma_2, \vec{n}_1, \vec{n}_2) \in

1666: \mathcal{Y}(\ell, \beta),$ let $M$ be the corresponding normalized labeled surface

1667: as in Proposition

1668: \ref{prop: parameters determine surface}. By the above discussion, the

1669: stabilizer in $S_\ell$ of $y$ is isomorphic to the

1670: kernel $K_M$ of the map $D: \Aff(M) \to G$. Since any translation

1671: equivalence permutes the rectangles on $M$, and any equivalence fixing

1672: a rectangle must fix all rectangles, the order of $K_M$ is at most

1673: $\ell$. 

1674: \end{proof} 

1675: 

1676: \begin{prop}\name{prop: explicit}

1677: Let $\til \NSVT(\beta)$ be the set of affine equivalence classes in

1678: $\NSVT(\beta)$. Then 

1679: $$\left|\til \NSVT(\beta) \right| \le \sum_{\ell

1680: \leq 1/\beta} \frac{1}{(\ell-1)!} \sum_{(\sigma_1,\sigma_2)\in X_\ell} 

1681: \eta(|\sigma_1|,\beta)\eta(|\sigma_2|,\beta),$$

1682: where 

1683: $$\eta(r, \beta)=(8\beta)^{2(1-r)}(1-2\log 

1684: 8\beta)^{r-1}$$ 

1685: and 

1686:   $|\sigma|$ is the number of cycles for

1687: $\sigma$.

1688: \end{prop}

1689: 

1690: \begin{proof} 

1691: Every class in $\til \NSVT(\beta)$ contains a

1692: standard form surface $M$, with a decomposition into $\ell \leq

1693: 1/\beta$ rectangles, by Corollary \ref{cor: stand form equiv class}. 

1694: The corresponding combinatorial data gives rise to a quadruple in

1695: $\mathcal{Y}(\ell, \beta)$, and distinct classes in $\til

1696: \NSVT(\beta)$ will give rise to quadruples in distinct

1697: $S_{\ell}$-orbits. 

1698: According

1699: to Proposition \ref{prop:diagonal count}, for each $(\sigma_1,

1700: \sigma_2) \in X_\ell$, 

1701: the number of Dehn twist vectors $\vec{n}_j$ for which $(\sigma_1,

1702: \sigma_2, \vec{n}_1, \vec{n}_2) \in \mathcal{Y}(\ell, \beta)$ is at most 

1703: $\eta(|\sigma_j|, \beta).$

1704: Thus the cardinality of $\mathcal Y(\ell,\beta)$

1705: is bounded by

1706: $$\Phi(\ell, \beta) = \sum_{(\sigma_1,\sigma_2)\in X_\ell}

1707: \eta(|\sigma_1|,\beta)\eta(|\sigma_2|,\beta).$$ 

1708: The size of an $S_\ell$-orbit in $y$ is at least $(\ell - 1)!$ by

1709: Proposition \ref{prop: order bound}. So the number of orbits in

1710: $\mathcal{Y}(\ell, \beta)$ is at most 

1711: $\Phi(\ell, \beta)/(\ell -1)!$ 

1712: The claim follows by summing over $\ell$.

1713: \end{proof}

1714: 

1715: This explicit estimate, along with a similar bound for

1716: half-translation surfaces, proves Theorem \ref{thm: NST finite}.

1717: We define ${\rm area}\NSVT(\beta)$ to be the sum over the affine

1718: equivalence classes in $\til \NSVT(\beta)$ of the areas of the

1719: Teichm\"uller curves  $\HHH/\Gamma_M$. 

1720: 

1721: \begin{prop}\name{prop: coarea}

1722: We have 

1723: $${\rm area}\NSVT(\beta)\le 2\pi \sum_{\ell \leq 2/\beta^2}

1724: {\Phi(\ell,\beta)\over (\ell-1)!}.$$

1725: 

1726: \end{prop}

1727: 

1728: \begin{proof} Every $M \in \NSVT(\beta)$ which is in standard form

1729: contains at most $2/\beta^2$ rectangles by Corollary 

1730: \ref{cor: stand form area est}. Thus $\sum_{\ell \leq 2/\beta^2} \Phi(\ell,\beta) /

1731: (\ell-1)!$ is an upper bound for the number of standard form

1732: surfaces which are affinely isomorphic to some $M \in

1733: \NSVT(\beta)$. Now the bound follows from Corollary \ref{cor:

1734: standard}. 

1735: \end{proof}

1736: 

1737: Since the sum of the areas is finite each individual area is finite

1738: and we get a proof of Theorem \ref{thm: characterization}. 

1739: 

1740: \begin{remark}

1741: Our formulae do not distinguish surfaces on different strata, nor do

1742: they distinguish connected surfaces from disconnected ones. 

1743: %Eskin and Okounkov \cite{EO1} have developed very sophisticated methods of

1744: %analyzing related quantities and their asymptotic behavior which they

1745: %apply to questions of computing volumes of strata. 

1746: 

1747: \end{remark}

1748: 

1749: 

1750: %\section{Counting connected surfaces}

1751: 

1752: \section{Proof of Theorem \ref{thm: TFAE}}

1753: 

1754: The following characterization of lattice surfaces is part of

1755: Theorem \ref{thm: TFAE}. 

1756: 

1757: \begin{thm}

1758: \name{prop: closed implies Veech}

1759: Let $\HH$ be the stratum containing the flat surface

1760: $M$. Then $M$ is a lattice surface if and only if the $G$-orbit of $M$

1761: is closed in $\HH$. 

1762: \end{thm}

1763: 

1764: \begin{proof} For two proofs of this result, see the sketch in Veech

1765: \cite{Veech - closed} or the proof in \cite[\S 5]{toronto}. 

1766: %we can proceed as follows. 

1767: %Recall that for a fixed

1768: %genus $g$ and a fixed $n \in \N$, there are finitely many 

1769: %strata of flat surface structures on a surface of genus $g$, with at most $n$ singularities of

1770: %cone angle $\pi$. The union of these strata, which we denote by $\QQ$,

1771: %is known as the

1772: %quadratic differential space of a surface of genus $g$ with $n$

1773: %punctures. Let $M$ and $\HH$ be as in the statement of the proposition. 

1774: %It was shown in \cite[Prop. 8]{SW} that $M$ is a lattice surface in case the

1775: %$G$-orbit of $M$ is closed in $\QQ$. However the same proof works if

1776: %one assumes only that the orbit is closed in $\HH$, since

1777: %the proof of \cite[Cor 2.6]{MW} works for locally finite measures on $\HH$

1778: %as well as on $\QQ$. This proves the `if' direction. The converse is

1779: %well-known, see \cite{Veech - alternative} for details. 

1780: \end{proof}

1781: 

1782: 

1783: \begin{proof}[Proof of Theorem \ref{thm: TFAE}]

1784:  Vorobets \cite{Vorobets} proved the implications

1785: (i) $\Longleftrightarrow$ (v) $\Longrightarrow$ (iv) 

1786: $\Longrightarrow$ (vi) $\Longleftrightarrow$ (ii)

1787: $\Longleftrightarrow$ (iii). Theorem \ref{thm: characterization} shows

1788: that (i) and (vi) are equivalent, Theorem \ref{thm: nst vs nsvt} shows

1789: that (vi) and (ix) are equivalent, and Theorem \ref{prop: closed implies

1790: Veech} gives the equivalence of (i) and (x). Clearly 

1791: that (viii) $\Longleftrightarrow$ (vii) $\Longleftrightarrow$ (ix),

1792: and 

1793: Corollary \ref{cor: viii} implies 

1794: (ix) $\Longleftarrow$ (viii). 

1795: Putting all these together one sees that 

1796: (i)--(x) are equivalent. To conclude the proof we will prove that 

1797:  (i) $\Longrightarrow$ (xi)

1798: $\Longrightarrow$ (vii). 

1799: 

1800: Assume (i). 

1801: It is well-known that if $\Gamma \subset G$ is a lattice then there is

1802: a compact $K_1 \subset G/\Gamma$ such that 

1803: any geodesic orbit intersects $K$; that is, denoting by $\pi: G \to

1804: G/\Gamma$ the projection map, for every $g \in G$ one has

1805: $$K_1

1806: \cap \{g_t\pi(g) : t \in \R \}\neq \varnothing.$$

1807: Now taking $\Gamma = \Gamma_M$, and letting $\varphi: G/\Gamma \to

1808: \HH$ be the orbit map $\varphi(\pi(g)) = gM,$ we see that

1809: (xi) holds with $K = \varphi(K_1).$

1810: 

1811: Now assume (xi). 

1812: Let $\eta>0$ be small enough so that for any $M_0 \in K$, the length

1813: of any saddle connection for $M_0$ is at least $\eta.$ 

1814: Let $v_1, v_2 \in \mathrm{hol}(M)$ such that $v_1\wedge v_2 \neq 0,$

1815: and suppose $v_i = \mathrm{hol}(\delta_i)$ where $\delta_i \in \LL.$ 

1816: For $h \in G$, let $l_i(h)$ be the length of $\delta_i$ with respect

1817: to the Euclidean metric on $hM$. 

1818: 

1819: Let $g \in G$ be

1820: a linear map such that $gv_1$ is horizontal, $gv_2$ is vertical and

1821: both have the same length, which we denote by $c$. By assumption,

1822: there is $t=t(g) \in \R$ with $g_tgM \in K$.  

1823: If $t\geq 0$ then 

1824: $$\eta \leq l_2(g_tg) = e^{-t/2} l_2(g) = e^{-t/2}c,$$

1825: so $c \geq \eta.$ If $t<0$ we apply the same argument with $l_1$

1826: instead of $l_2$ to see that $c \geq \eta.$ 

1827: Since $g$ preserves the two dimensional volume element, 

1828: $$|v_1 \wedge

1829: v_2| = |gv_1 \wedge gv_2| = c^2 \geq \eta^2,$$

1830: which is a positive constant independent of $v_1, v_2$. 

1831: \end{proof}

1832: 

1833: 

1834: 

1835: \begin{remark}

1836: Let $C$ be the number of cusps in

1837: $\HHH/\Gamma_M$. Then the arguments of \cite{Vorobets} show that the 

1838: number of $\Aff(M)$-orbits of triangles as in alternative (iv) of Theorem

1839: \ref{thm: TFAE} is between $C$ and $C (2g+ \left|\Sigma \right| -2)$. 

1840: \end{remark}

1841: 

1842: \ignore{

1843: 

1844: \section{Billiard tables and safe starting points}

1845: In this section we interpret our results for polygonal billiard

1846: tables. Given a polygon $\mathcal{P} \subset \R^2,$ let

1847: $\Gamma=\Gamma_{\mathcal{P}}$ be

1848: the subgroup of $O(2,\R)$ generated by the linear parts of the

1849: reflections in the sides of 

1850: $\mathcal{P}$, and assume $\Gamma$ is finite. The associated

1851: flat surface $M_{\mathcal{P}}$ is constructed as follows: consider the

1852: disjoint union 

1853: $\bigcup_{\gamma \in \Gamma} \gamma \mathcal{P}$, with the edges $e_i

1854: \in \gamma_i \mathcal{P}$ glued to each other if there is an edge $e$

1855: of $\mathcal{P}$ such that $e_i = \gamma_i e$ and $\gamma_1^{-1}

1856: \gamma_2$ is the reflection in $e$. This defines a translation

1857: surface $M_{\mathcal{P}}$ whose singularities are the

1858: vertices of the $\gamma_i \mathcal{P}$. We assume that the 

1859: set of singularities of $M_{\mathcal{P}}$ contains all 

1860: vertices of $\mathcal{P}$ and their $\Gamma$-orbits. Note that

1861: different conventions also appear in the literature, cf. \cite{HS, MT, Vorobets}. 

1862: 

1863: \begin{example}

1864: Let $\mathcal{P}$ be a triangle with 

1865: angles $\frac{\pi}{2}, \frac{\pi}{5}, \frac{3\pi}{10}$. Then

1866: $\left|\Gamma_{\mathcal{P}_1} \right| =

1867: 20$ and $M_{\mathcal{P}}$ can be obtained by gluing opposite edges in

1868: two regular pentagons. The acute vertex of $\mathcal{P}$ gives rise to

1869: two points $p_1, p_2$ on $M_{\mathcal{P}}$ which are the centers of these

1870: pentagons. The total angle around each $p_i$ is $2 \pi$. By our

1871: convention these points belong to $\Sigma_{M_{\mathcal{P}}}$ (as

1872: marked points). Hubert and Schmidt \cite{HS} showed that although 

1873: $M_{\mathcal{P}}$ is not a lattice 

1874: surface, changing $M_{\mathcal{P}}$ by removing $p_1, p_2$ from

1875: $\Sigma_{M_{\mathcal{P}}}$ (and retaining the

1876: underlying flat structure) gives rise to a

1877: lattice surface. 

1878: \end{example}

1879: 

1880: 

1881: 

1882: In order to relate

1883: close approaches and triangles we will need the following:

1884: 

1885: \begin{prop}

1886: \label{prop: triangles and approaches}

1887: For any vertex connection $\sigma$ on $\mathcal{P}$ which has close approaches, there is a

1888: triangle $\Delta$ in $M=M_{\mathcal{P}}$ satisfying $D(\sigma)

1889: L(\sigma)/2 =\mathrm{area}(\Delta).$

1890: For any triangle $\Delta$ in $M$, with the angle at two of its vertices less than $\pi/4$,

1891: there is a vertex 

1892: connection $\sigma$ on $\mathcal{P}$ which has close approaches, for

1893: which $D(\sigma) L(\sigma)/2

1894:  \leq \mathrm{area}(\Delta).$

1895: 

1896: \end{prop}

1897: 

1898: \begin{proof}

1899: Suppose $\sigma$ is a vertex connection, between vertices $p_1, p_2$

1900: on $\mathcal{P}$, which has close approaches. Let $L=L(\sigma)$, let $\phi: [0, L] \to 

1901: \mathcal{P}$ be a parametrization of $\sigma$, and let $t \in (0,

1902: L)$ be such that $\delta(t) =D= D(\sigma).$ Let $\sigma'$ be a path of

1903: length $D$ from $\phi(t)$ to a nearest vertex $p_3$, and let

1904: $\delta$ be the path from $p_1$ to $p_3$ which follows $\sigma$ until

1905: $\phi(t)$ and continues along $\sigma'$. We can lift $\sigma$ and

1906: $\delta$ to segments $\til \sigma$ and $\til \delta$ on $M$, and we have

1907: $$|\hol(\til \sigma) \wedge

1908: \hol(\til \delta)| = DL.$$

1909: Let $\Delta$ be a triangle in the plane with sides $\hol(\til \sigma),

1910: \hol(\til \delta)$, and let $\til p_1, \til p_2, \til p_3$ be the

1911: singularities on the ends of 

1912: $\til \sigma, \til \delta$ covering $p_1, p_2, p_3.$ We map $\hol(\til

1913: \sigma)$ to $\til \sigma$ and by linear continuation in each chart,

1914: isometrically map $\Delta$ to obtain a triangle in $M$, of area

1915: $DL/2$. To see that this map can be continued to all of $\Delta$, note

1916: that the only possible obstruction is a singularity of $M$ as the

1917: image of an interior point of $\Delta$, which would contradict the

1918: definition of $D$. 

1919: 

1920: Now suppose $\Delta$ is a triangle for $M$ with the angle at the two

1921: vertices $p_1, p_2$ less than $\pi/4$, and let $\sigma$ be the

1922: side of $\Delta$ joining $p_1$ and $p_2$. The condition on angles

1923: ensures that 

1924: the height $h$ of $\Delta$ is achieved by a segment from the vertex $p$ opposite to

1925: $\sigma$ to  $p' \in \sigma$, and $h$ is less than the distance from

1926: $p'$ to either $p_1$ or $p_2$. Thus $\sigma$ is a vertex

1927: connection which has close approaches, and 

1928: $$\mathrm{area}(\Delta) = \frac{hL(\sigma)}{2} \geq \frac{D(\sigma) L(\sigma)}{2}.$$

1929: \end{proof}

1930: 

1931: \begin{proof}[Proof of Corollary \ref{cor: billiard reformulation}]

1932: In view of Theorem \ref{thm: characterization} it suffices to show

1933: that $\mathcal{P}$ has a linear bound of small

1934: approaches if and only if $M_{\mathcal{P}}$ has no small 

1935: triangles. The `if' direction is immediate from the first

1936: assertion in Proposition \ref{prop: triangles and approaches}. Moreover,

1937: for any $M$, the set of triangles for $M$ with two angles greater than

1938: or equal to a fixed positive number is finite due to the discreteness

1939: of $\hol(M)$ and the fact that the diameter of a triangle is bounded

1940: above by the diameter of $M$. Now the converse follows from the second

1941: statement in Proposition \ref{prop: triangles and approaches}. 

1942: \end{proof}

1943: 

1944: \begin{question}

1945: Are there irrational polygons having a linear bound on close

1946: approaches?

1947: \end{question}

1948: 

1949: 

1950: By definition, a lattice surface $M$ is called {\em arithmetic} if it

1951: has a conjugate commensurable to $\PSL(2,\Z

1952: )$, and a point on a flat surface $M$ is called {\em periodic} if its 

1953: stabilizer under $\Aff(M)$ is of finite index in $\Aff(M).$ Periodic

1954: points were studied by Gutkin, Hubert and Schmidt, who showed:

1955: 

1956: \begin{thm}[\cite{GHS}]

1957: \label{GHS - main}

1958: Suppose $M$ is a lattice surface. Then the number of periodic points

1959: on $M$ is infinite if and only if $M$ is arithmetic. 

1960: \end{thm}

1961: 

1962: \begin{proof}[Proof of Corollary \ref{cor: safe start}]

1963: Note that for a polygon $\mathcal{P}$, all the vertices are safe

1964: starting points if and only if $\mathcal{P}$ has a linear bound on

1965: close approaches. In particular, if $\mathcal{P}$ is a lattice polygon then any

1966: of its singularities are safe starting points by 

1967: Corollary \ref{cor:

1968: billiard reformulation}. 

1969: Thus to prove (i) it is enough to show that if $\mathcal{P}$ has a

1970: safe starting point $p$ then $M_{\mathcal{P}}$ has no small

1971: triangles. So let $p$ be a safe starting point, 

1972: \combarak{How to continue from here?}

1973: 

1974: Now suppose $\mathcal{P}$ is a lattice polygon and $p$ is a safe

1975: starting point. We claim that any pre-image $\til p \in M_{\mathcal{P}}$

1976: of $p$ is a periodic point. Indeed, consider the surface

1977: $M'$ obtained from $M_{\mathcal{P}}$ 

1978: by marking $\til p$, that is $M'$ has the same underlying flat

1979: structure, with $\Sigma_{M'} = \Sigma_M \cup \{\til p\}.$ Arguing as

1980: in the proof of Proposition \ref{prop: triangles and approaches} we

1981: see that $M'$ has no small 

1982: triangles, and hence is a lattice surface. \combarak{Explain why $M'$

1983: can't have a small triangle with all vertices at $\til p$.} This implies that

1984: $\Gamma_{M'}$ is a lattice in $G$, hence of finite index in

1985: $\Gamma_M$. Therefore $\Aff(M')$ is of finite index in $\Aff(M)$. The subgroup

1986: of $\Aff(M')$ fixing $\til p$ is of finite index in $\Aff(M')$, so we

1987: find that $\til p$ is a periodic point for $M'$. Using Theorem

1988: \ref{GHS - main} we obtain (ii) and

1989: (iii). 

1990: \end{proof}

1991: 

1992: \begin{proof}[Proof of Corollary \ref{cor: closing lemma}]

1993: Let $\sigma, L, \vre, \alpha$ be as in the statement of the

1994: Corollary. By rotating $M$ we can assume that $\sigma$ is horizontal. 

1995: We fix an orientation on $\sigma$ and denote the initial point of

1996: $\sigma$ by $p$ and the terminal point by $q$. Let $\sigma'$ be a

1997: segment of length $\vre$ from $q$ to $p$, and let $\gamma$ be the path

1998: on $M$ obtained by concatenating $\sigma$ and $\sigma'$, so that

1999: $\gamma$ begins and ends at $p$. Let $\delta$ be a path of shortest

2000: length which is homotopic to $\gamma$ rel endpoints. There is a

2001: list  $x_1, \ldots, x_s$ of

2002: singularities (possibly with repetition), such that $\delta$ is a

2003: concatenation of an initial segment from $p$ to $x_1$, saddle

2004: connections from $x_i$ to $x_{i+1}$, and a terminal segment

2005: from $x_s$ to $p$. Any of the 

2006: segments $\lambda$ comprising $\delta$ satisfies 

2007: \eq{eq: segments satisfy}{

2008: |\xi| \leq L+\vre, \ |\eta| \leq \vre \ \ \mathrm{where \ } \hol(\lambda) =

2009: (\xi, \eta).

2010: }

2011: It is possible that $s=0$, in which case

2012: $\delta$ is just a straight segment from $p$ to $p$, and we are done. 

2013: So assume that $s \geq 1$, and repeat the previous construction with 

2014: $\gamma$ replaced with the

2015: concatenation of $\sigma'$ and $\sigma$, which begins and ends at

2016: $q$. We obtain a polygonal path $\delta'$ from $q$ to $q$ passing through

2017: singularities $y_1, \ldots, y_t$, where $t \geq 1$. Any of the

2018: segments $\lambda$ comprising $\delta'$ also satisfies \equ{eq: segments

2019: satisfy}. Moreover there is a saddle connection $\lambda_1$

2020: (respectively $\lambda_2$) joining one of the $x_i$ to one of the

2021: $y_j$ and passing across

2022: $\sigma$ (resp. $\sigma'$). Both $\lambda_i$ satisfy 

2023: \eq{eq: segments2}{

2024: |\xi| \leq 2(L+\vre), \ |\eta| \leq 2\vre \ \ \mathrm{where \ } \hol(\lambda_i) =

2025: (\xi, \eta).

2026: }

2027: 

2028: We now claim that all the saddle connections above joining

2029: singularities in $\{x_1, \ldots, x_s, y_1, \ldots, y_t\}$ are

2030: parallel. Indeed, let $\lambda, \lambda'$ be two such saddle

2031: connections with $\hol(\lambda) = (\xi, \eta)$ and $\hol(\lambda') =

2032: (\xi', \eta')$. If they are not parallel, then using \equ{eq: segments

2033: satisfy}, \equ{eq: segments2} we find

2034: \[\alpha \leq |\hol(\lambda) \wedge \hol(\lambda') | = |\xi \eta' -

2035: \xi' \eta| \leq |\xi \eta'| + | \xi' \eta| \leq 8 (L+\vre)\vre < \alpha,

2036: \]

2037: a contradiction. 

2038: 

2039: So the saddle connections are parallel and their concatenation gives

2040: the required loop.

2041: \end{proof}

2042: 

2043: 

2044: }

2045: 

2046: \section{The simplest lattice surfaces}

2047: \name{section:

2048: simplest}

2049: In this section we will list the first few lattice surfaces, ordered

2050: as in \equ{eq: alphaM}. We will also list surfaces with removable

2051: singularities. 

2052: 

2053: Recall that any triangulation of $M$ has $\tau(M)/\pi$ triangles,

2054: implying

2055: \eq{eq: easy upper bound}{\alpha(M) \leq \frac{\pi}{\tau(M)}. }

2056: 

2057: 

2058: \begin{prop}

2059: \name{prop: bottom of spectrum}

2060: If

2061: equality holds in \equ{eq: easy upper bound} then $M$ is an arithmetic

2062: lattice surface, all 

2063: triangles on $M$ have the same area, all cylinders in a cylinder

2064: decomposition of $M$ have the same height and all parallel 

2065: saddle connections on $M$ have the same length.

2066: \end{prop}

2067: \begin{proof}

2068: Since 

2069: $\alpha(M)$ is both a lower bound and the 

2070: average of the triangle areas in any triangulation, the area of

2071: each triangle for $M$ must be equal 

2072: to $\alpha(M).$ We will now show that for any $\theta$, the length

2073: $|\sigma|$ of any

2074: $\sigma \in \LL_{M}(\theta)$ is the same. By Theorem \ref{thm: TFAE}, $\theta$

2075: is a parabolic direction 

2076: for $M$, so $\sigma$ is contained in the boundary 

2077: of a cylinder $C$, and there is a singularity $x$ in the

2078: boundary component of $C$ opposite to $\sigma$. If $\sigma' \in

2079: \LL_M(\theta)$ is in the same boundary component of $C$ as $\sigma$,

2080: consider the triangles $\Delta$ and $\Delta'$ with apex $x$ and base $\sigma$ and

2081: $\sigma'$ respectively. Since these triangles have the same height and

2082: area we find that $|\sigma| = |\sigma'|$. Similarly, if $\sigma''$ is a segment on the boundary

2083: component of $C$ opposite to $\sigma$ then a triangle $\Delta''$ with base

2084: $\sigma''$ and apex an endpoint of $\sigma$ has the same area and

2085: height as $\Delta$ so we find that $|\sigma| = |\sigma''|.$ 

2086: 

2087: Now consider another cylinder $C'$ whose boundary contains

2088: $\sigma$. Since it contains a triangle with base $\sigma$, its height

2089: is the same as that of $C$, and any saddle connection on either of its

2090: boundary components has length $|\sigma|$. Continuing in this fashion and using the

2091: connectedness of $M$ we find that all segments in $\LL_M(\theta)$ have

2092: the same length and all cylinders in the corresponding

2093: cylinder decomposition have the same height.

2094: 

2095: Now consider $C, \sigma$ as before and let $\sigma'$ be a saddle

2096: connection passing from one 

2097: boundary component of $C$ to another. Write $v = \hol(\sigma), v' =

2098: \hol(\sigma')$, and consider any saddle connection $\lambda$ on

2099: $M$. We can write $\hol(\lambda) = nv + n'v'$, where $n$ (respectively

2100: $n'$) is the number

2101: of times $\lambda$ passes through a cylinder in the direction of

2102: $\sigma'$ (resp. $\sigma$). In particular $\hol(M) \subset \Z v \oplus

2103: \Z v'.$ It follows by \cite[\S 5]{GJ} 

2104: that $M$ is arithmetic. 

2105: \end{proof}

2106: 

2107: 

2108: We now list some examples of arithmetic lattice surfaces $M$ and calculate

2109: $\alpha(M)$. Let $\gamma = \tau/\pi$ denote the number of triangles in

2110: a triangulation of $M$, so that by \equ{eq: Gauss Bonnet} we have 

2111: \eq{eq: GB}{

2112: \gamma= 2 \left(\sum r_{\sigma} + \left|\Sigma \right| \right) =

2113: 2\left(2g-2+\left|\Sigma \right| \right).

2114: }

2115: 

2116: \begin{enumerate}

2117: \item

2118: Let $M_1$ be the standard flat torus $[0,1]^2$ with opposite sides

2119: identified, with one marked point at the origin. By \equ{eq: GB} $

2120: \gamma  =2$

2121: so $\alpha(M_1) \leq 1/2.$ On the other hand $\hol(M_1) \subset \Z^2$

2122: so that $|v_1 \wedge v_2| \geq 1$ for any two linearly independent

2123: $v_1, v_2 \in \hol(M_1)$. Since the area of a triangle with sides

2124: $v_1, v_2$ is $|v_1 \wedge v_2|/2$ we see that $\alpha(M_1)=1/2.$

2125: \item

2126: Let $M_2$ be the standard pillowcase with 4 singularities of total

2127: angle $\pi$. Then $\gamma = 4$ so that 

2128: $\alpha(M) \leq 1/4.$ On the other hand $\hol(M_3) \subset

2129: \left(\frac1{\sqrt{2}} \Z \right)^2,$ so that $\alpha(M_3) =1/4.$ 

2130: \item

2131: Let $M_3$ be the standard torus with one marked

2132: point at the origin

2133: and another at $\left(\frac12,0\right)$. Then

2134: $\gamma

2135: = 4$ so that $\alpha(M_2)\leq 1/4$. On the other hand $\hol(M_2)

2136: \subset \Z\left[\frac12\right] \oplus \Z$ so that $\alpha(M_2) = 1/4.$

2137: 

2138: \ignore{

2139: \item

2140: Let $M_4$ be the standard torus with one marked point at the origin,

2141: one marked point at $\left(\frac13, 0\right)$ and a third marked point at

2142: $\left(\frac23, 0\right)$. Then $\gamma =3$ so that $\alpha(M) \leq

2143: 1/6$ and $\hol(M_4) \subset \Z\left[\frac13 \right] \oplus \Z$ so that

2144: $\alpha(M_4) = 1/6.$ 

2145: \item

2146: Let $M_5$ be the half-translation surface obtained by gluing 3 squares

2147: to make a flat torus attached to a pillowcase along a slit (see Figure 2). Then there are two

2148: singularities with total angle $\pi$ and one with total angle $4 \pi$

2149: so that \equ{eq: GB} gives $\gamma =3.$ Also $\hol(M_5) \subset

2150: \Z\left[\frac1{\sqrt{3}} \right] \oplus \Z\left[\frac1{\sqrt{3}}

2151: \right]$ so that 

2152: $\alpha(M_5) = 1/6.$

2153: 

2154: 

2155: \item

2156: Let $M_6$ be the translation surface obtained by gluing opposite sides

2157: of an $L$-shaped

2158: polygon made of three equal squares of sidelength

2159: $\frac1{\sqrt{3}}$ (see Figure 2). Then there is

2160: one singularity of cone angle $6 \pi$ and \equ{eq: GB} gives $\gamma

2161: =3,$ so that $\alpha(M_6) \leq 1/6.$ Also $\hol(M_6) \subset

2162: \Z\left[\frac1{\sqrt{3}} \right] \oplus \Z\left[\frac1{\sqrt{3}}

2163: \right] $ so that 

2164: $\alpha(M_5) = 1/6.$

2165: }

2166: \end{enumerate}

2167: 

2168: We now show:

2169: \begin{prop}

2170: $\NST\left(\frac14 \right)$ consists of the affine equivalence classes

2171: of $M_1, M_2, M_3$. 

2172: 

2173: \end{prop}

2174: 

2175: \begin{proof}

2176: The above discussion shows $M_1, M_2, M_3 \in \NST\left(\frac14

2177: \right)$ and it remains to show that if $M$ is a flat surface with

2178: $\alpha = \alpha(M) \geq 1/4$ then $M$ is affinely equivalent to one of the

2179: $M_i$. Let $\gamma$ be as above, so by \equ{eq: easy upper bound} we

2180: have $\gamma \in \{2,4\}.$ If $\gamma =2$ then either $g=1$ and $\left|

2181: \Sigma\right| =1$ or $g=0$ and $\left|\Sigma\right| =3.$ In the first case, since

2182: the moduli space of tori with one marked point is a 

2183: single $G$-orbit, we find that $M$ is affinely equivalent to

2184: $M_1$. The second case does not occur as there is no solution to

2185: \equ{eq: Gauss Bonnet} with three singularities. 

2186: 

2187: Now suppose $\gamma =4.$ The only solutions to \equ{eq: easy upper

2188: bound} are 

2189: $\left(g=0, \left|\Sigma\right| =4\right)$ and $\left(g=1, \left|

2190: \Sigma \right|=2\right)$. In the

2191: first case it follows from \equ{eq: Gauss Bonnet} that the four

2192: singularities have $r_{\sigma} = -1/2$ so $M$ is a pillowcase. Since

2193: the moduli space of the pillowcase consists of a single $G$-orbit, we

2194: have that $M$ is affinely equivalent to $M_2$. In the second case by

2195: \equ{eq: GB} the two singularities $\sigma_1, \sigma_2$ satisfy either

2196: \begin{itemize}

2197: \item[(i)]

2198: $r_{\sigma_1} = -1/2, r_{\sigma_2}

2199: = 1/2.$

2200: \item[(ii)]

2201: $r_{\sigma_1} = r_{\sigma_2}=0$;

2202: \end{itemize}

2203: In case (i) we obtain a half-translation structure on a torus, but

2204: such a flat surface does not exist (see \cite{MS2}).  

2205: In case (ii) $M$ is a torus with two marked points. Applying an element

2206: of $G$ we may identify $M$ with the unit square, and there is no

2207: loss of generality in assuming that one of the marked points is at the

2208: origin. By Proposition \ref{prop: bottom of spectrum}, if $\sigma$ is

2209: the saddle connection connecting the two marked points inside the unit

2210: square, then there is a parallel segment from the second marked point

2211: to a singularity, of the same length. This implies that the second

2212: marked point is at either of the points $\left(\frac12, 0 \right),

2213: \left(0, \frac12, \right), \left(\frac12, \frac12\right).$ All of

2214: these cases are affinely equivalent to $M_3$. 

2215: \end{proof}

2216: 

2217: If $\alpha$ is not too small one can continue applying such arguments

2218: to identify $\NST\left(\alpha 

2219: \right)$. For example, in addition to $M_1, M_2,

2220: M_3$, $\NST\left(\frac16\right)$ consists of a torus with two marked

2221: point, a torus with three marked points, a genus 1 half-translation

2222: surface made by gluing a torus and a pillowcase along a slit, and a

2223: genus 2 surface (see Figure 1). All these examples are arithmetic. 

2224: 

2225: 

2226: \begin{figure}[htp] 

2227: \input{m5m6.pstex_t}

2228: \caption{Two surfaces in $\NST\left( \frac16 \right)$}

2229: \end{figure}

2230: 

2231: 

2232: \ignore{

2233: Now suppose $\gamma =3$. The possible solutions to \equ{eq: easy upper

2234: bound} are:

2235: \begin{itemize}

2236: \item[(a)]

2237: $g=0$, $\left|\Sigma \right| = 5$. 

2238: \item[(b)]

2239: $g=1$, $\left|\Sigma\right| = 3.$

2240: \item[(c)]

2241: $g=2$, $\left|\Sigma \right| =1.$

2242: 

2243: \end{itemize}

2244: First suppose (a) holds. The only solution to \equ{eq: Gauss Bonnet}

2245: is that there are four singularities with $r_{\sigma} = -1/2$ and one

2246: marked point. Thus $M$ is a pillowcase, and by applying the $G$-action

2247: we may assume it is a standard pillowcase with two squares of

2248: side-length $1/\sqrt{2}$ glued to each other. The marked point is on

2249: one of these squares. Arguing as in case (ii) above we find that the

2250: marked point is on the midpoint of a segment connecting two of the

2251: corners of the squares. A direct computation then shows that $M$

2252: contains a triangle of area $1/8$ contradicting $\alpha \geq 1/6.$ 

2253: 

2254: Now suppose (b) holds. Then one of the following holds:

2255: \begin{itemize}

2256: \item[(b1)]

2257: $M$ is a torus with singularities of orders $1/2, 0, -1/2.$

2258: \item[(b2)]

2259: $M$ is a torus with 3 marked points. 

2260: \item[(b3)]

2261: $M$ is a torus with singularities of orders $1, -1/2, -1/2$.

2262: \end{itemize}

2263: 

2264: Case (b1) gives a half-translation structure which cannot occur by

2265: \cite{MS2}. In case (b2) there is no loss of generality in assuming

2266: that $M$ is identified with the standard square with one marked point

2267: at the origin. Then one uses Proposition \ref{prop: bottom of

2268: spectrum} to analyze the possible locations of the marked points. One

2269: finds that up to affine equivalence the only possibility is

2270: $M_4$. Similar arguments show that in case (b3) the only possibility

2271: is $M_5$.

2272: 

2273: Finally in case $c$ the only possibility is $M_6$.  

2274: \end{proof}

2275: }

2276: \begin{thebibliography}{99}

2277: %\bibitem[Bo]{Boshernitzan} M.\ Boshernitzan, {\em Rank two

2278: %interval exchange transformations}, Erg. Th. Dyn. Sys. {\bf 8} (1988)

2279: %379--394.

2280: 

2281: \bibitem[Bo]{Bowman} J. Bowman, {\em Doctoral thesis, Cornell

2282: University}, in preparation. 

2283: 

2284: \bibitem[Cal]{Calta} K. Calta, {\em Veech surfaces and complete

2285: periodicity in genus 2}, J. Amer. Math. Soc. {\bf 17} (2004) 871--908.

2286: 

2287: %\bibitem[CoFoSi]{CFS} I. P. Cornfeld, S. V. Fomin and Ya. G. Sinai,

2288: %{\bf Ergodic theory} (1982) Springer.

2289: 

2290: %\bibitem[DaSm]{DS} S. G. Dani and J. Smillie, {\em Uniform

2291: %distribution of horocycle orbits for Fuchsian groups}, Duke

2292: %Math. J. {\bf 51} (1984) 185--194.

2293: 

2294: %\bibitem[DoHu]{DH} A. Douady and J. Hubbard, {\em On the density of

2295: %Strebel differentials}, Inv. Math. {\bf 30} (1975) 175--179.

2296: 

2297: %\bibitem[EsMaZo]{emz} A. Eskin, H. Masur and A. Zorich, {\em Moduli

2298: %spaces of abelian differentials: the principal boundary, counting

2299: %problems, and the Siegel-Veech constants } Publ. I. H. E. S. {\bf 97}

2300: %(2003) 631--678.

2301: 

2302: \bibitem[EsOk]{EO1} A. Eskin and A. Okounkov, {\em Asymptotics of

2303: numbers of branched coverings of a torus and volumes of moduli spaces

2304: of holomorphic differentials,} Inv. Math.

2305: {\bf 145} (2001), no. 1, 59--103.

2306: 

2307: 

2308: %\bibitem[EsOk2]{EO2} A. Eskin and A. Okounkov, {\em Pillowcases and

2309: %quasi-modular forms} (2005) preprint.

2310:  

2311: \bibitem[Ga]{Gantmacher} F. R. Gantmacher, {\bf Theory of matrices,

2312: Vol. 2,} Chelsea publishing, 1959.   

2313: 

2314: 

2315: 

2316: 

2317: \bibitem[Gu]{Gutkin} E. Gutkin, {\em Billiards on almost integrable

2318: polyhedral surfaces,} Erg. Th. Dyn. Sys. {\bf 4} (1984) 569--584.

2319: 

2320: %\bibitem[GuHuSc]{GHS} E. Gutkin, P. Hubert, T. Scmidt, {\em Affine 

2321: %diffeomporphisms of translation surfaces: periodic points, fuchsian

2322: %groups, and arithmeticity,} Ann. Sci. Ec. Norm. Sup. {\bf 36} (2003) 847--866. 

2323: 

2324: \bibitem[GuJu]{GJ} E. Gutkin and C. Judge, {\em Affine maps

2325: of translation surfaces: geometry and arithmetic}, Duke Math. J. {\bf

2326: 103}, no.2 (2000) 191--213.

2327: 

2328: 

2329: %\bibitem[HaKa]{HK} B. Hasselblatt and A. Katok, {\bf An introduction

2330: %to the modern theory of dynamical systems??} ?.

2331: 

2332: %\bibitem[Hoo]{Hooper} P. Hooper, {\em Yet another Veech triangle},

2333: %preprint (2005). 

2334: 

2335: %\bibitem[HuSc]{HS} P. Hubert and T. Schmidt, {\em Veech groups and

2336: %polygonal coverings}, J. Geom. Phys. {\bf 35}  (2000), 

2337: %79--91.

2338: 

2339: %\bibitem[HuSc]{HS} P. Hubert and T. Schmidt, {\em Infinitely

2340: %generated Veech groups}, Duke Math. J.  {\bf 123}  (2004),  no. 1,

2341: %49--69. 

2342: 

2343: 

2344: \bibitem[KeSm]{KS} R. Kenyon and J. Smillie, {\em Billiards on

2345: rational-angled triangles}, Comm. Math. Helv. {\bf 75} (2000) 65--108.

2346: 

2347: \bibitem[KeMaSm]{KMS} S. Kerckhoff, H. Masur and J. Smillie, {\em

2348: Ergodicity of billiard flows and quadratic differentials},

2349: Ann. Math. {\bf 124} (1986) 293--311.

2350: 

2351: %\bibitem[KoZo]{KZ} M. Kontsevich and A. Zorich, {\em Connected

2352: %components of the moduli spaces of Abelian differentials with

2353: %prescribed singularities,} Invent. Math. 153 (2003), no. 3, 631--678.

2354: 

2355: %\bibitem[Kul]{Kulikov} M. Kulikov, {\em Infinitely generated Fuchsian

2356: %groups and minimal sets of the geodesic and horocycle flow}, to appear

2357: %in Comptes. Rend. Acad. Sci.

2358: 

2359: %\bibitem[La]{lanneau} E. Lanneau, {\em Hyperelliptic components of the

2360: %moduli spaces of quadratic differentials with prescribed

2361: %singularities}, Comm. Math. Helvetici, to appear (2004). 

2362: 

2363: 

2364: %\bibitem[Le]{Leininger} C. Leininger, {\em On groups generated by two

2365: %positive multi-twists: Teichm\"uller curves and Lehmer's number} (2003) preprint. 

2366: 

2367: %\bibitem[MaSm]{MS} H. Masur and J. Smillie, {\em Hausdorff dimension

2368: %of sets of nonergodic measured foliations}, Ann. Math. {\bf 134}

2369: %(1991) 455--543.

2370: 

2371: \bibitem[MaSm]{MS2} H. Masur and J. Smillie, {\em Quadratic

2372: differentials with prescribed singularities and pseudo-Anosov

2373: diffeomorphisms}, Comment. Math. Helvitici {\bf 68} (1993) 289--307.

2374: 

2375: \bibitem[MaTa]{MT} H. Masur and S. Tabachnikov, {\em Rational

2376: billiards and flat structures}, in {\bf Handbook of dynamical systems,

2377: Enc. Math. Sci. Ser.} (2001).

2378: 

2379: \bibitem[McM]{McMullen} C. McMullen, {\em Billiards and Teichm\"uller

2380: curves on Hilbert modular surfaces} J. Amer. Math. Soc. {\bf 16}

2381: (2003) 857--885.

2382: 

2383: %\bibitem[MiWe]{MW} Y. Minsky and B. Weiss, {\em Nondivergence of

2384: %horocyclic flows on moduli spaces}, J. Reine Angew. Math. {\bf 552}

2385: %(2002) 131-177.

2386: 

2387: 

2388: \bibitem[Na]{Naveh} Y. Naveh, {\em Tight upper bounds on the number of

2389: invariant components}, Isr. J. Math. (to

2390: appear).

2391: 

2392: 

2393: %\bibitem[PaPe]{PP}A. Papadopoulos and R. C. Penner, {\em Enumerating

2394: %pseudo-Anosov foliations}, Pac. J. Math {\bf 142} (1990) 159--173.

2395: 

2396: \bibitem[Pu]{Puchta} J.-C. Puchta, {\em On triangular billiards,} 

2397: Comment. Math. Helv. {\bf 76} (2001), no. 3, 501--505.

2398: 

2399: %\bibitem[Ra]{Raghunathan} M. S. Raghunathan, {\bf Discrete subgroups

2400: %of Lie groups}, Springer 1972. 

2401: 

2402: %\bibitem[SmWe1]{SW} J. Smillie and B. Weiss, {\em Minimal sets for

2403: %flows on moduli space}, Isr. J. Math, {\bf 142} (2004) 249--260.

2404: 

2405: \bibitem[SmWe1]{SW -- Veech} J. Smillie and B. Weiss, {\em Veech's

2406: dichotomy and the lattice property},

2407: preprint (2005).

2408: 

2409: \bibitem[SmWe2]{toronto} J. Smillie and B. Weiss, {\em Finiteness

2410: results for flat surfaces: a survey and problem list},  (2006) to

2411: appear in {\bf  Partially hyperbolic dynamics, laminations, and 

2412:      Teichm\"uller flow} (Proceedings of a conference, Fields

2413: Institute, Toronto  Jan 2006), G. Forni (ed.)   

2414:  

2415:  

2416: %\bibitem[St]{S} K. Strebel, {\bf Quadratic Differentials},  Springer, Berlin, 1984.

2417: 

2418: %\bibitem[Ta]{Tabachnikov} S. Tabachnikov, {\bf Billiards,} Panoramas

2419: %et synth\`eses 1, Soci\'et\'e Math\'ematique de France (1995). 

2420: 

2421: \bibitem[Th]{Thurston} W. Thurston, {\em On the geometry and dynamics

2422: of diffeomorphisms of surfaces}, Bull. AMS (new series) {\bf 19} no. 2

2423: (1988) 417--431.

2424: 

2425: %\bibitem[Ve1]{Veechstrata}

2426: %W. A. Veech, {\em Moduli spaces of

2427: %quadratic differentials}, J. Analyse Math. {\bf 55} (1990) 117--171.

2428: 

2429: 

2430: 

2431: %\bibitem[Ve3]{Veech}  W. A. Veech, {\em Measures Supported on the

2432: %Set of Uniquely Ergodic Directions of an Arbitrary Holomorphic

2433: %1-form}, Erg. Th. Dyn. Sys. {\bf 19} (1999) 1093--1109.

2434: 

2435: \bibitem[Ve1]{Veech - alternative} W. A. Veech, {\em

2436: Teichm\"uller curves in moduli space, Eisenstein series and an

2437: application to triangular billiards}, 

2438: Invent. Math. {\bf 97} (1989), no. 3, 553--583.

2439: 

2440: \bibitem[Ve2]{Veech - closed} W. A. Veech,

2441: {\em Geometric realizations of hyperelliptic curves}, in {\bf

2442: Algorithms, fractals, and dynamics (Okayama/Kyoto, 1992)} , 217--226,

2443: Plenum, New York, 1995.

2444: 

2445: \bibitem[Ve3]{Veech - spine} W. A. Veech, {\em Bicuspid F-structures

2446: and Hecke groups,}

2447: preprint (2006). 

2448: 

2449: \bibitem[Vo]{Vorobets} Ya. B. Vorobets, {\em Planar structures and

2450: billiards in rational polygons: the Veech alternative}, (Russian)

2451: Uspekhi Mat. Nauk  {\bf 51}  (1996),  no. 5(311), 3--42;  translation in

2452: Russian Math. Surveys  {\bf 51}  (1996),  no. 5, 779--817.

2453: 

2454: %\bibitem[Wa]{Ward} C. Ward, {\em Calculation of fuchsian groups

2455: %associated to billiards in a rational triangle,}

2456: %Erg. Th. Dyn. Sys. {\bf 18} (1998) 1019--1042.

2457: 

2458: \bibitem[Zo]{zorich survey} A. Zorich, {\em Flat surfaces}, in {\bf

2459: Frontiers in number theory, physics and geometry,} P. Cartier,

2460: B. Julia, P. Moussa and P. Vanhove (eds), Springer (2006). 

2461: 

2462: %\bibitem[Zo]{Zorich cambridge} A. Zorich, {\em Square tiled surfaces

2463: %and Teichm\"uller volumes of the moduli spaces of abelian

2464: %differentials}, in {\bf Rigidity in dynamics and geometry,} M. Burger

2465: %and A. Iozzi, eds, Springer (2000).

2466: \end{thebibliography}

2467: 

2468: \end{document}

2469: