1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2: %

3: %  K. Grosse-Brauckmann, R. B. Kusner, J.M. Sullivan

4: %                 Coplanar constant mean curvature surfaces

5: %

6: % AMS2000: 53A10, 58D10

7: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

8: %

9: \documentclass[11pt,leqno,twoside]{amsart}

10: %\usepackage[totalwidth=17cm,totalheight=9in]{geometry}

11: \usepackage{amsmath,amsthm,amsfonts}

12: \usepackage{amssymb,verbatim,url,graphicx}

13: \usepackage[percent]{overpic}

14: 

15: %% -----------------------------  Layout  -----------------------------------

16: %

17: % Layout for A4, two-sided output

18: %\textwidth16cm \textheight23.5cm

19: %\hoffset-4mm \voffset-10mm

20: %\setlength{\oddsidemargin}{2mm} \setlength{\evensidemargin}{2mm}

21: \textwidth=5in

22: \textheight=7.5in

23: 

24: %\parskip0.6ex plus0.4ex minus 0.2ex         % Absatzabstand

25: %\setlength{\parsep}{0.2ex plus 0.1ex minus 0.1ex}

26: 

27: %\headheight=6.15pt % to avoid Overfull vbox messages with 12pt amsart

28: 

29: %\renewcommand{\baselinestretch}{1.15} %Zeilenabstandsvergroesserung

30: 

31: %% ---------------------------- Figures ----------------------

32: \graphicspath{{figs/}}

33: \newcommand{\figr}[1]{Figure~\ref{fig:#1}}

34: %

35: \newcommand{\fig}[5]{%

36: \begin{figure}

37: \begin{overpic}[width=#2]{#1}

38: #4

39: \end{overpic}

40: \caption[#3]{#5}

41: \label{fig:#1}

42: \end{figure}%

43: }

44: %

45: \newcommand{\figtwo}[7]{%

46: \begin{figure}

47: \begin{overpic}[width=#3]{#1} #5 \end{overpic}

48: \hspace{1cm}

49: \begin{overpic}[width=#3]{#2} #6 \end{overpic}

50: \caption[#4]{#7} \label{fig:#1-#2}

51: \end{figure}%

52: }

53: 

54: %% ---------------------------  Theorems   -----------------------------

55: %

56: \newtheorem{theorem}{Theorem}[section]

57: \newtheorem*{maintheorem}{Main Theorem}

58: \newtheorem{lemma}[theorem]{Lemma}

59: \newtheorem{proposition}[theorem]{Proposition}

60: \newtheorem{corollary}[theorem]{Corollary}

61: \newtheorem*{maincorollary}{Corollary}

62: 

63: \newcommand{\thm}[1]{Theorem~\ref{th:#1}}

64: \newcommand{\prop}[1]{Proposition~\ref{pr:#1}}

65: \newcommand{\cor}[1]{Corollary~\ref{co:#1}}

66: 

67: \theoremstyle{definition}

68:   \newtheorem{definition}[theorem]{Definition}

69:   \newtheorem{conjecture}[theorem]{Conjecture}

70: \theoremstyle{remark}

71:   \newtheorem*{remark}{Remark}

72:   \newtheorem*{remarks}{Remarks}

73:   \newtheorem*{examples}{Examples}

74:   \newtheorem*{example}{Example}

75: 

76: %% ----------------------------  Lists   -----------------------------

77: %

78: \renewcommand{\labelenumi}{({\it \roman{enumi}\/})}

79: 

80: \renewcommand{\labelitemii}{$*$}

81: 

82: \newcommand{\ia}{{\rm({\it i\/}\rm)}}

83: \newcommand{\ii}{{\rm({\it ii\/}\rm)}}

84: \newcommand{\iii}{{\rm({\it iii\/}\rm)}}

85: \newcommand{\iv}{{\rm({\it iv\/}\rm)}}

86: \newcommand{\va}{{\rm({\it v\/}\rm)}}

87: 

88: %% ---------- Mathematical symbols in alphabetical ordering  -----------

89: %

90: \newcommand{\area}{\operatorname{area}}

91: \def\alex-emb/{Alexandrov-embedded}

92: \newcommand{\after}{\circ}

93: \newcommand{\C}{\mathbb{C}}

94: \newcommand{\cmc}{{\footnotesize{CMC}}~}

95: \def\CMC/{{\footnotesize{CMC}}}

96: \newcommand{\const}{\operatorname{const}}

97: \def\d{\partial}

98: \newcommand{\dist}{\operatorname{dist}}

99: \newcommand{\D}{\mathcal{D}}

100: \newcommand{\Dk}{\D_k}

101: \newcommand{\Dbox}{{D^{{}_\Box}_{\vphantom{j}}}}  % cheating, but looks nice

102: \newcommand{\Dboxj}{{D^{{}_\Box}_j}}

103: \newcommand{\E}{\mathcal{E}}

104: \newcommand{\F}{\mathcal{F}}

105: \renewcommand{\epsilon}{\varepsilon}

106: \newcommand{\grad}{\mathrm{grad}}

107: \renewcommand{\H}{\mathbb{H}}

108: \newcommand{\id}{\operatorname{Id}}

109: \newcommand{\ident}{\equiv}

110: \newcommand{\isom}{\cong}

111: \renewcommand{\Im}{\operatorname{Im}}

112: \newcommand{\intr}{\operatorname{int}}

113: \let\dotlessi\i

114: \renewcommand{\i}{\mathbf{i}}  % bold face type for quaternion i

115: \renewcommand{\j}{\mathbf{j}}  % bold face type for quaternion j

116: \renewcommand{\k}{\mathbf{k}}  % bold face type for quaternion k

117: \newcommand{\K}{ K} %{{\mathcal K}} 

118: \newcommand{\length}{\operatorname{length}}

119: \newcommand{\M}{\mathcal{M}}

120: \newcommand{\Mk}{\mathcal{M}_k}

121: \newcommand{\N}{\mathbb{N}}

122: \let\paragraph=\S  % must come before redefinition of \S

123: \newcommand{\Phik}{\Phi_k}

124: \renewcommand{\P}{{\mathcal P}}

125: \newcommand{\Pk}{\P_k}

126: %\newcommand{\Pl}{{\P_\ell}

127: \newcommand{\Psik}{\Psi_k}

128: \renewcommand{\phi}{\varphi}

129: \newcommand{\Pik}{\Pi_{\k}}               % for Hopf projection

130: % conjugates: with tildes %%

131: % a not-so-wide tilde, thanks to Silvio Levy (note: extra {} are needed!)

132: \def\wtilde#1{{\mskip4mu\widetilde{\mskip-4mu #1\mskip-1mu}\mskip1mu}}

133: \def\mtilde#1{{\mskip7mu\widetilde{\mskip-7mu #1\mskip-7mu}\mskip7mu}}

134: \def\plustilde#1{\wtilde{#1}{}^+}

135: \newcommand{\R}{\mathbb{R}}

136: \renewcommand{\Re}{\operatorname{Re}}

137: \renewcommand{\S}{\mathbb{S}}

138: \newcommand{\sfrac}[2]{#1/#2} % for easy change from \frac

139: \newcommand{\SO}{\mathsf{SO}}

140: \newcommand{\setm}{\smallsetminus}

141: \newcommand{\T}{\mathcal{T}}

142: \newcommand{\Tl}{{\mathcal{T}}_\ell}

143: \newcommand{\Tk}{{\mathcal{T}}_k}

144: \renewcommand{\theta}{\vartheta}

145: \def\takes{\colon}

146: \def\uhalf/{upper half}

147: \newcommand{\vol}{\operatorname{vol}}

148: \newcommand{\vertical}{\operatorname{vert}}

149: \newcommand{\X}{\mathcal{X}}

150: \newcommand{\Y}{\mathcal{Y}}

151: \newcommand{\Z}{\mathbb{Z}}

152: \newcommand{\Zodd}{\mathbb{Z}_\mathrm{odd}} %why doesn't textrm work?

153: 

154: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

155: % Temporary commands

156: 

157: \let\mgp=\marginpar \marginparwidth18mm \marginparsep1mm

158: \def\marginpar#1{\mgp{\raggedright\tiny #1}}

159: %\def\marginpar#1{}   %Uncomment this to hide all marginpars

160: 

161: %\let\lbl=\label

162: %\def\label#1{\lbl{#1}\ifinner\else\marginpar{\ref{#1} #1}\ignorespaces\fi}

163: 

164: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

165: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

166: \begin{document}

167: 

168: \title{Coplanar constant mean curvature surfaces}

169: \date{2007 December 5}

170: 

171: \author[Grosse-Brauckmann]{Karsten~Gro{\ss}e-Brauckmann}

172: \address{Technische Universit\"at Darmstadt, Fb.~Mathematik (AG~3),

173:     Schlossgartenstr.~7, 64289~Darmstadt, Germany}

174: \email{kgb@mathematik.tu-darmstadt.de}

175: 

176: \author[Kusner]{\\Robert~B.~Kusner}

177: \address{Mathematics Dept., University of Massachusetts,

178:       Amherst~MA~01003, USA}

179: \email{kusner@math.umass.edu}

180: 

181: \author[Sullivan]{John~M.~Sullivan}

182: \address{Technische Universit\"at Berlin, MA 3--2,

183: Str.~des 17.~Juni~136, 10623~Berlin, Germany}

184: \email{sullivan@math.tu-berlin.de}

185: 

186: \begin{abstract}

187:   We consider constant mean curvature surfaces with finite topology,

188:   properly embedded in three-space in the sense of Alexandrov.

189:   Such surfaces with three ends and genus zero were constructed

190:   and completely classified by the authors~\cite{gkspnas, gks}.  

191:   Here we extend the arguments to the case of an arbitrary number of ends,

192:   under the assumption that the asymptotic axes of the

193:   ends lie in a common plane:

194:   we construct and classify the entire family of these genus-zero,

195:   coplanar constant mean curvature surfaces.

196: \end{abstract}

197: \maketitle

198: 

199: \begin{center}

200: \small \emph{Dedicated to Hermann Karcher\\

201: on the occasion of his sixty-fifth birthday.}

202: \end{center}

203: \bigskip

204: 

205: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

206: \section*{Introduction}

207: 

208: The search for examples of surfaces with constant mean curvature

209: dates back to the nineteenth century.

210: Unlike minimal surfaces, which have a Weierstrass representation,

211: the case of nonzero constant mean curvature (\CMC/) does not readily yield 

212: explicit examples; 

213: it thus appeared that constant mean curvature surfaces were scarce.  

214: In recent decades, however, several construction methods have evolved.

215: Wente's discovery of an immersed torus of constant mean curvature

216: initiated the integrable-systems approach explored 

217: by Abresch, Bobenko, Pinkall, Sterling and others; 

218: this in turn led to a generalized Weierstrass representation 

219: by Dorfmeister, Pedit and Wu.

220: The ground-breaking analytic perturbation (or gluing) techniques of Kapouleas 

221: have been extended by Mazzeo, Pacard, Pollack, Ratzkin and others.  

222: Conjugate surface techniques, introduced by Lawson, 

223: were developed further by Karcher 

224: and---combined with new analytic methods---by the present authors.  

225: An abundance of known \CMC/ surfaces has emerged, 

226: inviting their systematic study and classification.

227: 

228: With the conjugate surface method, we can study \CMC/

229: surfaces with sufficient symmetries.  This method is powerful in

230: that it enables us to study entire families of surfaces, that is,

231: components of suitably defined moduli spaces.

232: In contrast, the gluing techniques mentioned above, while more widely

233: applicable, tend to produce examples only near a boundary of the

234: relevant moduli space.

235: The generalized Weierstrass representation also has the potential to deal

236: with moduli spaces, even without any symmetry assumptions,

237: but it has difficulty detecting key geometric features

238: such as embeddedness.

239: %% also period problem is a problem

240: 

241: The \emph{Delaunay unduloids}~\cite{del} are

242: the noncompact embedded \CMC/ surfaces

243: of revolution.  These singly periodic surfaces decompose along

244: their \emph{necks} (minimizing closed geodesics) into \emph{bubbles}.  

245: All known embedded \CMC/ surfaces display similar bubbles.

246: In light of experience with

247: doubly and triply periodic surfaces \cite{kar,law,kgb},

248: the discovery that bubbles can be created and deleted 

249: continuously on surfaces of finite topology~\cite{gk} was surprising.

250: This insight is what led us to expect our moduli spaces to be connected,

251: and thus to look for the existence proof presented here, 

252: based on the continuity method of~\cite{gks}.

253: 

254: Let us now make the setting of our paper precise.

255: A surface has \emph{finite topology} when it is homeomorphic to

256: a compact surface~$\Sigma$ of genus~$g$ from which a finite number~$k$ of 

257: points has been removed;  

258: a neighborhood of any of these punctures is called an \emph{end}.

259: If the surface has nonzero constant mean curvature, 

260: then by a scaling we may assume the mean curvature is identically~$1$; 

261: from now on \cmc will denote this condition.

262: As in \cite{gks} we consider a class of surfaces more general

263: than the embedded ones, namely those bounding an immersed solid.

264: To be precise, we define a finite topology \CMC/ surface~$M$

265: to be \emph{\alex-emb/} if~$M$ is properly immersed, if each end of~$M$ is

266: embedded, and if there exists a compact 3-manifold~$W$ with boundary

267: $\d W=\Sigma$, and a proper immersion

268: $F\takes W\setm\{q_1,\ldots,q_k\}\to\R^3$ whose boundary restriction

269: $f\takes \Sigma\setm\{q_1,\ldots,q_k\} \to\R^3$ parametrizes~$M$.  

270: Moreover, we assume that the (inward) mean-curvature normal~$\nu$ of~$M$ 

271: points into~$W$.  

272: 

273: A Delaunay unduloid is an embedded \CMC/ surface with two ends and genus zero;

274: it is determined by its \emph{necksize} $n\in(0,\pi]$, the circumference of any neck.

275: More generally, 

276: we define a \emph{$k$-unduloid} to be an \alex-emb/ \CMC/ surface with

277: finite genus and a finite number $k\in\N$ of ends.  Our terminology

278: is justified by the asymptotics result of Korevaar, Kusner and Solomon:

279: \begin{theorem}[\cite{kks}]  \label{th:kksasym}

280:   Each end of a $k$-unduloid is exponentially asymptotic to 

281:   a Delaunay unduloid, and thus has an asymptotic axis and 

282:   necksize $n\in(0,\pi]$.

283: \end{theorem}

284: For $k\le2$, all $k$-unduloids have genus zero, and their classification

285: is well-known: Alexandrov~\cite{al1,al2} showed the round sphere

286: is the only $0$-unduloid; Meeks~\cite{mee} showed there are

287: no $1$-unduloids; and the Delaunay unduloids themselves

288: are the only $2$-undu\-loids~\cite{kks}. 

289: 

290: A \cmc surface is called \emph{coplanar} 

291: if it is contained in a half-space of~$\R^3$. 

292: For $k\le3$, all $k$-unduloids are coplanar~\cite{mee}.

293: The Alexandrov reflection method extends to coplanar $k$-unduloids~\cite{kks},

294: showing that any such surface~$M$ is \emph{Alexandrov-symmetric} 

295: in the following sense:

296: it has a reflection plane~$P$    %, which we take to be the $xy$-plane,

297: such that $M\setm P$ has two connected

298: components $M^{\pm}$ whose Gauss images~$\nu(M^{\pm})$ are

299: contained in disjoint open hemispheres; %~$(\S^2)^{\mp}$;

300: moreover $\nu(M\cap P)$ is contained in the equator.

301: %(Note that our convention of using inward normals means that the

302: %Gauss map of $\S^2$ itself is the antipodal map, explaining the

303: %switched signs above.)

304: %

305: \begin{theorem}[\cite{kks}]  \label{th:kkscopl}

306:   A coplanar $k$-unduloid $M$ is contained in a slab 

307:   and is Alexandrov-symmetric across a plane~$P$ within that slab.  

308:   The plane~$P$ contains the asymptotic axes of the ends of~$M$.

309: % Each of its ends is exponentially asymptotic to a Delaunay unduloid,

310: % and thus has an asymptotic axis and necksize $n\in(0,\pi]$.

311:   % see pp.472, 476, 477

312: \end{theorem}

313: 

314: We want to study moduli spaces of coplanar $k$-unduloids with genus zero.

315: Thus we identify two such $k$-unduloids 

316: when they differ by a rigid motion in space and we 

317: label the ends to avoid orbifold points at surfaces with extra symmetry.

318: It is convenient to represent any coplanar $k$-unduloid

319: with the horizontal $xy$-plane as its fixed mirror plane~$P$,

320: such that the labels of the ends occur in cyclic order:

321: \begin{definition}

322:   Let us fix $k\ge0$ points $q_1,\ldots,q_k$

323:   in cyclic order along the equator of~$\S^2$.

324:   We consider proper \alex-emb/ \CMC/ immersions of 

325:   $\S^2\setm \{q_1,\ldots ,q_k\}$ into~$\R^3$

326:   which are equivariant under reflection across the $xy$-plane.

327:   The \emph{moduli space}~$\Mk$ of {coplanar $k$-unduloids} 

328:   of genus zero consists of all such immersions, modulo 

329:   diffeomorphisms of the domain fixing the points~$q_i$,

330:   and modulo isometries of~$\R^3$.

331:   Uniform convergence in Hausdorff distance

332:   on compact subsets of~$\R^3$ defines the topology on~$\Mk$.

333: \end{definition}

334: 

335: %(Note that our equivariance and orientation assumptions mean that

336: %it would be equivalent, in the above definition, to identify only

337: %those immersions which differ by orientation-preserving diffeomorphisms

338: %of the domain and/or orientation-preserving isometries of~$\R^3$.)

339: 

340: Our main theorem classifies coplanar $k$-unduloids of genus zero in~$\Mk$.

341: This generalizes our classification of 

342: \emph{triunduloids} (with $k=3$) 

343: by triples of points in~$\S^2$ modulo rotation~\cite{gks},

344: which shows $\M_3$ is homeomorphic to the open three-ball.

345: % This triple uniquely determines a spherical metric triangle.

346: For $k\ge4$, there is still a natural map from

347: coplanar $k$-unduloids of genus zero to $k$-tuples on~$\S^2$,

348: but this no longer suffices to classify the surfaces.

349: Instead, we need to employ a space~$\Dk$ of spherical metrics,

350: whose precise definition will be given below.

351: Our classification says that $\Mk$ is homeomorphic to~$\Dk$:

352: \begin{maintheorem}

353:   For each $k\ge3$, there is a homeomorphism $\Phi$

354:   from the moduli space~$\Mk$ of coplanar $k$-unduloids

355:   of genus~$0$

356:   to the connected \hbox{$(2k-3)$}--manifold~$\Dk$ of $k$-point metrics.  

357:   The necksizes of a $k$-unduloid $M\in\Mk$ are 

358:   the spherical distances of the $k$~consecutive pairs of

359:   completion points in~$\Phi(M)$.

360: \end{maintheorem}

361: As a consequence we have the following constraints 

362: on the necksizes of a coplanar $k$-unduloid of genus~$0$.

363: (These do not hold for higher genus, as shown by

364: the coplanar $k$-unduloids of genus~$1$ 

365: with all ends cylindrical~\cite{cylends}.)

366: \begin{maincorollary}

367:   Let $k\ge 3$.

368:   The necksizes $n_1,\ldots,n_k\in(0,\pi]$

369:   of any coplanar $k$-unduloid of genus~$0$ satisfy

370:   \begin{align}

371:     \label{necksizesumodd}

372:     && n_1+ \cdots+ n_k &\le (k-1)\pi &\text{for }& k \text{ odd}, &&\\

373:     \label{necksizesumeven}

374:     && n_1+\cdots + n_k & < k\pi      &\text{for }& k\ge4 \text{ even}, &&

375:   \end{align}

376:   and these inequalities are sharp.

377:   At most $k-2$ ends of the surface can be cylindrical,

378:   and examples of coplanar $k$-unduloids with $k-2$ cylindrical ends

379:   exist for each $k\ge 3$.

380: \end{maincorollary}

381: 

382: %---------- rhombic 4-unduloids with cyl. segment  ------------------------

383: \begin{figure}

384: \hspace*{-7mm}

385: \begin{overpic}[width=.23\textwidth]{rhcyl1} \end{overpic}

386: \hspace{-2mm}

387: \begin{overpic}[width=.23\textwidth]{rhcyl2} \end{overpic}

388: \hspace{5mm}

389: \begin{overpic}[width=.23\textwidth]{rhcyl3} \end{overpic}

390: \hspace{2mm}

391: \begin{overpic}[width=.23\textwidth]{rhcyl4} \end{overpic}

392: \hspace*{-5mm}

393: \caption[rhombic with cyl segment]{

394:   Our main theorem characterizes which sets of~$k$ asymptotic

395:   necksizes a genus-zero coplanar $k$-unduloid can attain.

396:   In addition, for $k=3$ the necksizes uniquely 

397:   determine the angles between the ends, as follows from force balancing.

398:   For $k\ge4$, however, there is no such dependence:

399:   the $4$-unduloids shown above all have all necksizes equal to~$\pi/2$,

400:   but the angles between the ends vary.  

401: % (These surfaces were computed in Grape.)

402: }\label{fig:rhombic with cyl segment}

403: \end{figure}

404: %---------------------------------------------------------------------------

405:  

406: Our theorem should be compared with the classification of

407: coplanar minimal surfaces by Cos{\'\dotlessi}n and Ros~\cite{cr}.

408: The classifying space for their minimal $k$-noids consists of planar polygons,

409: whose combinatorial description is well-known.

410: % it's the generalization of the "art gallery theorem", contained in 

411: % the Aigner-Ziegler book

412: Another simplification arises because they are able

413: to prove directly that all their surfaces are nondegenerate.

414: We are not aware of any other classification results

415: for moduli spaces of minimal or \CMC/ surfaces with more than three ends.

416: 

417: Very recently, in joint work~\cite{GKKRS} with Korevaar and Ratzkin

418: (following up on~\cite{kkr}),

419: we have augmented the results of this paper as follows:

420: All coplanar $k$-unduloids of genus zero are nondegenerate,

421: so that $\Mk$ is a real-analytic manifold, and a tubular neighborhood

422: of it in the space of all $k$-unduloids is a $(3k-6)$--manifold.

423: Furthermore, our classifying map~$\Phi$ is a (real-analytic) diffeomorphism.

424: \medskip

425: 

426: We conclude this introduction with an overview of our paper.

427: In Section~1, we define the classifying map $\Phi$ as a composition

428: of the following maps:  By coplanarity and Theorem~\ref{th:kkscopl}, we first 

429: associate to each $M\in\Mk$ its \uhalf/~$M^+$. 

430: Since $M$ has genus zero, $M^+$ is an open disk.

431: It thus has a conjugate cousin, an isometric minimal

432: immersion to the three-sphere.

433: Hopf projection of this cousin disk to the $2$-sphere

434: produces a metric on $M^+$ locally isometric to $\S^2$, 

435: which we call the spherical $k$-point metric $\Phi(M)$.

436: 

437: Central for our paper is the analysis (in Sections~2 and~3)

438: of the space~$\Dk$ of spherical $k$-point metrics.

439: In Section~2,

440: we start with a combinatorial decomposition of each metric $D\in\Dk$.

441: In view of the classifying map, analyzing 

442: metrics amounts to analyzing coplanar $k$-unduloids,

443: yielding a decomposition of the surface into bubbles.

444: On cylindrical parts of a $k$-unduloid, such a decomposition 

445: is necessarily ambiguous.  For this reason, 

446: our decomposition cannot be unique, making the analysis somewhat involved.

447: 

448: In Section~3, we apply our combinatorial decomposition to

449: show that $\Dk$ is a connected manifold.

450: % with a known combinatorial description.

451: We analyze the example~$\D_4$ completely to find that 

452: that~$\D_4\isom \M_4$ is a five-ball, generalizing the

453: result of~\cite{gks} that $\D_3\isom\M_3$ is a three-ball.

454: Our combinatorial description is not explicit enough 

455: to decide whether~$\Dk$ is always a $(2k-3)$--ball.

456: (In ongoing joint work with Korevaar, using a M{\"o}bius-invariant

457: decomposition of $k$-point metrics, we expect to be

458: able to show that indeed $\Dk\isom\R^{2k-3}$ for all $k\ge3$.)

459: 

460: In Section~4, we prove injectivity and properness of our classifying map.

461: Injectivity follows by applying

462: a maximum principle to the cousin surfaces

463: (just as in the case $k=3$ of~\cite{gks}).

464: Properness involves \emph{a priori} estimates, 

465: corresponding to the closedness part of a traditional

466: continuity method.  The proof here is somewhat different than

467: in the case $k=3$ because of the possibility of small necks occurring

468: away from the ends.  Still, our characterization of compact subsets

469: of~$\Dk$ allows us to prove the needed area and curvature estimates.

470: 

471: In Section~5, we establish the surjectivity of our classifying map.

472: As in \cite{gks}, the crucial point is to apply a structure theorem

473: for real-analytic varieties.  Thus we must show

474: that our spaces~$\Mk$, consisting of coplanar surfaces

475: with a mirror symmetry, are real-analytic.

476: For this, we adapt the methods of~\cite{kmp},

477: which dealt with the space of all $k$-unduloids.

478: More generally, we show that for any fixed finite symmetry group,

479: the moduli space of symmetric $k$-unduloids is real-analytic,

480: a result which has applications beyond this paper~\cite{groisman}.

481: Then we can apply our version of the continuity method, as introduced

482: in~\cite{gks}, to prove the main theorem.

483: 

484: {\small

485: We gratefully acknowledge partial support from 

486: the Deutsche Forschungsgemeinschaft and the National Science Foundation.

487: Our figures of \cmc surfaces were produced with the 

488: Graphics Programming Environment \emph{Grape}; our

489: sketches of spherical polygons, with \emph{Spherical Easel}.

490: }

491: 

492: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

493: \section{The classifying map}       \label{se:class}

494: 

495: In this section we define the space~$\Dk$ of $k$-point spherical metrics

496: and the classifying map $\Phi\colon\Mk\to\Dk$.

497: We assume $k\ge 2$ to avoid trivial cases.

498: 

499: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

500: \subsection{Conjugate cousins and necksizes}

501: 

502: Consider a coplanar $k$-unduloid $M\in\Mk$;

503: it is Alexandrov symmetric across the $xy$-plane~$P$.

504: % and representing a class in~$\Mk$.

505: Take a component $M^+$ of $M\setminus P$

506: and orient $M^+$ such that the boundary labels occur in

507: increasing cyclic order:  

508: this means that $M^+$ lies to the left of~$\d M^+$.  

509: By a rotation of $M$ in $\R^3$, we may assume that

510: $M^+ := M\cap\{z>0\}$; we call $M^+$ the \emph{\uhalf/}.

511: 

512: When $M$ is embedded the ends occur in counterclockwise order in~$P$.

513: In other words, if at each end of~$M$ 

514: we truncate $P\cap M$ and join in a locally embedded way,

515: then we obtain an embedded loop in~$P$ with turning number one.

516: 

517: However, in general ends of~$M$ may cross 

518: (as in~\figr{crossed ends}, right)

519: and thus no longer appear in the same order in~$P$.

520: We note that for large~$k$, most coplanar $k$-unduloids

521: have such crossed ends and thus fail to be embedded.

522: By Alexandrov symmetry, 

523: the projection of~$M^+$ to~$P$ gives an immersed disk,

524: so that the turning number of the corresponding boundary

525: loop is still one.

526: 

527: %---------- 4-unduloids with crossed ends ------------------------

528: \begin{figure}

529: %\hspace*{7mm}

530: \begin{overpic}[width=.47\textwidth]{cr4und} \end{overpic}

531: \hspace*{\fill}

532: \begin{overpic}[width=.47\textwidth]{cr4und_bdy} \end{overpic}

533: \caption{

534:   A coplanar 4-unduloid and its

535:   intersection with the symmetry plane~$P$.

536:   Typical coplanar $k$-unduloids with $k>3$ 

537:   have such crossed ends and are not embedded.

538:   % unlike most triunduloids.

539: } 

540: \label{fig:crossed ends}

541: \end{figure}

542: %---------------------------------------------------------------------------

543: 

544: We now review the conjugate cousin relationship~\cite{law},

545: in the first-order form presented in \cite[Thm.~1.1]{gks}.

546: Given any simply connected \CMC/ surface, such as~$M^+$, there

547: is a cousin minimal surface $\plustilde M$ in the three-sphere~$\S^3$.

548: We regard~$\S^3$ as the Lie group of unit quaternions

549: and identify~$\R^3$ with its Lie algebra $T_1\S^3=\Im\H$, as in~\cite{gks}.

550: The cousin is determined uniquely up to left translation in~$\S^3$. 

551: 

552: The upper half~$M^+\subset\R^3$ is a disk isometric to its

553: minimal cousin $\plustilde M\subset\S^3$.

554: This isometry has the following effect on normals and tangent vectors.

555: 

556: If $\nu\in\S^2$ is the normal at $p\in M^+$,

557: then the normal $\tilde\nu$

558: at the corresponding point $\tilde p\in\plustilde M$

559: is simply the left translation of~$\nu$, that is, $\tilde \nu=\tilde p\nu$.

560: Left translation preserves the inner product, and so

561: for the unit quaternion $\k\in\S^2\subset\Im\H$ we find

562: $$

563:   \langle \tilde\nu, \tilde p\k\rangle_{\S^3} = \langle\nu, \k\rangle_{\R^3}.

564: $$

565: Since $\k$ is vertical,

566: Theorem~\ref{th:kkscopl} gives $\langle\nu,\k\rangle<0$ 

567: on the \uhalf/ $M^+$,  % transverse to vertical lines, with 

568: and thus the cousin $\plustilde M$ is transverse to 

569: the \emph{$\k$-Hopf circles} $t\mapsto\tilde p(\cos t+\k\sin t)$. 

570: %at each $\tilde p\in\plustilde M$.

571: 

572: If $X$ is any tangent vector to $M^+$ at~$p$,

573: then its isometric image $\tilde X = \tilde p\,JX$ at~$\tilde p$

574: is obtained through a rotation~$J$ by~$\sfrac\pi2$ within~$T_pM$,

575: followed by left translation to~$\tilde p$.

576: When $X$ is the (horizontal) unit tangent to~$\d M^+$ at~$p$,

577: the tangent plane at~$p$ is vertical and so 

578: $JX$ agrees with the constant conormal~$\k$ to~$\d M^+$. 

579: Thus $\tilde X=\tilde p\k$,

580: implying that the component of~$\d\plustilde M$ through~$\tilde p$

581: covers precisely the $\k$-Hopf circle $t\mapsto\tilde p(\cos t+\k\sin t)$.

582: 

583: The $\k$-Hopf circles are the fibers of the

584: \emph{$\k$-Hopf projection} $\Pik\colon\S^3\to\S^2$,

585: defined by $\Pik(q):=q\k\overline q$ in terms of quaternion multiplication. 

586: Thus each component of $\d\plustilde M$,

587: corresponding to the arc from some end~$i$ 

588: to end $i+1$, projects to a single point~$p_i$.

589: Collecting these points in~$\S^2$, we obtain an ordered $k$-tuple 

590: $$

591:   \Psi(M):=\Pik(\d\plustilde{M}) = (p_1,\ldots,p_k).

592: $$

593: As in \cite[Sect.~2]{gks} it follows that $\Psi(M)$ is

594: well-defined up to rotation in~$\SO(3)$.

595: % In detail: The cousin is defined up to left translation,

596: % but Hopf projecting of a left translated set yields

597: % a rotation of the Hopf image.

598: 

599: Theorem~\ref{th:kksasym} says that the ends of a $k$-unduloid $M$ are

600: asymptotic to certain unduloids, defining 

601: \emph{asymptotic necksizes} $n_1,\ldots,n_k$ of~$M$.

602: These are determined by $\Psi(M)$:

603: \begin{theorem}\label{th:necksizes}

604:   Let $M$ be a coplanar $k$-unduloid of genus zero. 

605:   Its asymptotic necksizes $n_1,\ldots,n_k$

606:   are the distances between consecutive points

607:   of the spherical $k$-tuple $\Psi(M)=(p_1,\ldots,p_k)$:

608:   $$\dist_{\S^2}(p_{i-1},p_{i})=n_i.$$

609: \end{theorem}

610: \begin{proof}

611:   As in~\cite{gks} the result is immediate for $k=2$:

612:   the cousin of the \uhalf/ of an unduloid of necksize~$n$ is a spherical

613:   helicoid bounded by two $k$-Hopf circles;

614:   these circles Hopf-project to two points in~$\S^2$ at distance~$n$.  

615:   Then the proof given in \cite[Theorem 2.1]{gks}

616:   for the case $k=3$ carries over to the general case:

617:   the asymptotics of Theorem~\ref{th:kksasym} imply for any~$k$ that

618:   each end of the cousin~$\plustilde M$ has the same bounding great

619:   circle rays as its asymptotic unduloid cousin.

620: \end{proof}

621: 

622: \begin{corollary} \label{co:oddk}

623:   The set of $k$~necksizes of a coplanar $k$-unduloid obeys

624:   the inequalities satisfied by the edgelengths of a 

625:   spherical $k$-gon:

626:   for each subset $P\subset\{1,\ldots,k\}$ 

627:   with an odd number~$|P|$ of elements

628:   \begin{equation} \label{biswas}

629:     \sum_{i\in P} n_i \le \sum_{i\not\in P} n_i + \big(|P|-1\big)\pi .

630:   \end{equation}

631: \end{corollary}

632: \begin{proof}

633:   Note that the case of $|P|=k$ odd in~\eqref{biswas}

634:   is~\eqref{necksizesumodd}.  We prove this case only; 

635:   for the general case, see~\cite{bs}, where it is also shown

636:   that these inequalities are sharp.

637: % characterize the possible edgelengths of a spherical $k$-gon.

638: 

639:   We proceed by induction,

640:   noting that for $k=3$ we recognize \eqref{necksizesumodd}

641:   as the perimeter inequality for spherical triangles~\cite{gks}.

642: 

643:   For $k\ge 5$, we draw minimizing geodesics of length~$a\ge 0$ 

644:   and $b\ge 0$ from~$p_k$ to~$p_{k-3}$ and~$p_{k-2}$, as in~\figr{kgon}.

645: %

646: %------------------------------------------------------------------------- 

647: \fig{kgon}{7cm}{Perimeter inequality for spherical polygons}

648: {\put(83,59){$p_k$} \put(26,15){$p_{k-3}$} \put(62,15){$p_{k-2}$}

649:  \put(63,46){$a$} \put(77,39){$b$} \put(55,82){$p_1$}

650:  \put(44,14){$n_{k-2}$}

651:  \put(63,64){$n_k$} \put(43,66){$p_{k-1}$} \put(43,39){$n_{k-1}$} }

652: {Edges (dashed) of length~$a$ and~$b$ from~$p_k$

653: to~$p_{k-3}$ and~$p_{k-2}$ divide a $k$-gon into two triangles

654: and a $(k-2)$--gon; the pieces overlap in this example.

655: %The perimeter inequality of Corollary~\ref{co:sphpolybound} follows by induction.

656: %(This figure, among others, was drawn with Spherical Easel.)

657: }

658: %------------------------------------------------------------------------- 

659: %

660:   These divide the $k$-gon into a $(k-2)$--gon and two triangles,

661:   which degenerate when $a$ or~$b$ vanishes.

662:   By induction the $(k-2)$-gon satisfies

663:   $$

664:     n_1+\cdots+n_{k-3}+a \le (k-3)\pi.

665:   $$

666:   The perimeter and triangle inequalities for the two triangles give

667:   $$

668:     n_{k-1}+n_k +b \le 2\pi

669:     \qquad\text{and}\qquad 

670:     n_{k-2}\le a+b.

671:   $$

672:   Adding these three inequalities gives~\eqref{necksizesumodd} as desired.

673:   Since all necksizes are positive at most $k-2$ necksizes can be~$\pi$.

674: \end{proof}

675: 

676: \begin{remark}

677:   This corollary extends to a broader class of 

678:   immersed (and not Alexandrov embedded) \cmc surfaces,

679:   including the surfaces constructed in~\cite{kss}.

680:   If $M$ has genus~$0$ with $k$~ends,

681:   each of which is asymptotically unduloidal, and $M$ has a mirror plane,

682:   then it can be shown that the disk~$M^+$ is still bounded by $k$~curves

683:   of planar reflection.  

684:   Thus the map~$\Psi$ is still defined, so that~\eqref{necksizesumodd} 

685:   and~\eqref{biswas} hold.

686:   However, inequality~\eqref{necksizesumeven} 

687:   % stated in the Corollary to the Main Theorem

688:   depends on~\prop{dk2tk}, and need not extend to the broader class.

689: % Thus an interesting question is whether a \cmc surface of genus 0 

690: % with 4 asymptotically cylindrical ends exists.  

691: % Such a surface might perhaps be obtained

692: % by fusing together two cylindrical bubbeltons at the bubbles.

693: \end{remark}

694: 

695: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

696: \subsection{Spherical metrics and the classifying map}

697: 

698: For $k\le3$, the spherical $k$-tuple $\Psi(M)$ determines

699: the coplanar $k$-unduloid~$M$ uniquely. 

700: Since for $k>3$ this is no longer the case, we are led to introduce

701: a family of metrics to serve as our classifying space.

702: 

703: A metric is \emph{spherical} if it is locally isometric to~$\S^2$.

704: We will consider spherical metrics~$D$ on the oriented open two-disk.

705: We denote the completion of such a metric by~$\hat D$,

706: and refer to $\d D:=\hat D \setm D$

707: as the \emph{completion boundary} of $D$.

708: To the spherical metric $D$ we can associate 

709: an oriented \emph{developing map} $\phi\colon D\to\S^2$, 

710: unique up to rotation, which extends to~$\hat D$.

711: %unique up to isometry of~$\S^2$, which extends to~$\hat D$.

712: \begin{definition}

713:   Consider a spherical metric on the disk~$D$ 

714:   whose completion boundary consists of %$k\ge1$ 

715:   fixed points $\{q_1,\ldots,q_k\}$, called \emph{vertices},

716:   in increasing cyclic order.

717:   Two such metrics are equivalent if they are related by an isometry

718:   extending to fix each vertex $q_1,\ldots,q_k$.

719:   A \emph{$k$-point metric} is an equivalence class of such metrics,

720:   and we let~$\Dk$ denote the space of all $k$-point metrics.

721:   The developing map defines a topology on~$\Dk$:

722:   a sequence of metrics in~$\Dk$ is said to converge if there are 

723:   developing maps (from the disk to~$\S^2$) which converge

724:   in the compact-open topology.

725: \end{definition}

726: 

727: As we noted above, if $M$ is a coplanar $k$-unduloid,

728: $\plustilde M$~is transverse to the $\k$-Hopf circles.

729: Hence $\Pik\colon \plustilde M\to\S^2$ is an immersion 

730: well-defined up to rotation. 

731: We use $\Pik(\plustilde M)$ to denote the spherical metric

732: on $M^+$ pulled back from~$\S^2$.

733: As in \cite[Lem.~3.4]{gks}, 

734: the exponential asymptotics of the ends of~$M$ to unduloids shows that

735: this spherical metric is a $k$-point metric:

736: The $k$~curves in~$\d M^+$ correspond to the vertices,

737: which develop to the $k$~labelled points~$(p_1,\ldots,p_k)=\Psi(M)$.

738: Thus we have shown:

739: \begin{theorem} \label{th:PhimapsMtoD}

740:   For each $k\ge2$, there is a map $\Phi\colon\Mk\to\Dk$ defined by

741:   $$

742:     \Phi(M):=\Pik(\plustilde{M}).

743:   $$ 

744: % in the sense of immersed surfaces.

745: \end{theorem}

746: \noindent

747: We call this map $\Phi$ the \emph{classifying map}.

748: Our main theorem asserts that $\Phi$ is a homeomorphism.

749: 

750: 

751: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

752: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

753: \section{Decompositions of a $k$-point metric}

754: 

755: We analyze a $k$-point metric $D\in\Dk$ by decomposing it

756: into standard pieces.  

757: We will use this combinatorial analysis 

758: later to prove that the space~$\Dk$ is connected.

759: We assume $k\ge 2$ throughout this section.

760: 

761: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

762: \subsection{Polygonal metrics} \label{se:polmet}

763: 

764: It will simplify the exposition to consider a further

765: class of spherical metrics on the disk, whose completions are compact.  

766: %

767: \begin{definition}

768:   A \emph{polygonal metric} or \emph{(spherical) $k$-gon} 

769:   is an isometry class of spherical metrics on the open disk

770:   whose completion boundary is topologically a circle

771:   consisting of $k$~\emph{vertices} and $k$~\emph{edges} each

772:   developing to a great-circle arc in~$\S^2$ of length at most~$\pi$.

773:   We denote the space of all $k$-gons by~$\Pk$.

774: \end{definition}

775: \noindent

776: At each vertex~$v$ of a polygonal metric, there is an 

777: interior \emph{angle}~$\alpha(v)$, well-defined as a positive real number.

778: We introduce some further terminology:

779: 

780: A \emph{diagonal} in a spherical metric~$D$ is a geodesic of length

781: at most~$\pi$ in~$D$, whose completion in~$\hat D$ joins two vertices

782: (which may be consecutive).

783: A diagonal develops onto a minimizing great circle arc.

784: 

785: Two kinds of $2$-gons (see \figr{slit-lune}) will be important throughout:

786: A \emph{slit sphere} is 

787: the complement of a minimizing great circle arc in~$\S^2$.

788: Each of its vertices has angle $\alpha=2\pi$;

789: its \emph{slit length} is the length of each edge.

790: A \emph{lune} of angle~$\alpha$ is a $2$-gon whose vertices

791: develop to antipodal points of~$\S^2$; 

792: it has a continuous family of diagonals.

793: 

794: %--------------------------------------------------------------------------

795: \figtwo{slit}{lune}{5cm}{Slit sphere and lune}{}{}

796: {Left: The complement of the great circle arc is a slit sphere.

797: Right: Two geodesics of length~$\pi$,

798: connecting a pair of antipodal points, cut the sphere into

799: two complementary lunes, one convex and one concave.}

800: %--------------------------------------------------------------------------

801: 

802: %A \emph{lune} of angle~$\alpha$ is a $2$-gon whose vertices

803: %develop to antipodal points of~$\S^2$,

804: %as in \figr{slit-lune} (right).

805: %A lune has a continuous family of diagonals.

806: %The complement of a minimizing great circle arc in $\S^2$,

807: %called a \emph{slit sphere}, is the other example of a $2$-gon.

808: %Each of its vertices has angle $\alpha=2\pi$;

809: %its \emph{slit length} is the length of each edge.

810: 

811: A \emph{ray of spheres} is a spherical metric~$D$ whose completion boundary

812: is a single arc, developing to a minimizing geodesic~$\gamma$.

813: The preimage of~$\gamma$ decomposes~$D$ into 

814: an infinite union of slit spheres. %~\cite[Lemma 3.2]{gks}.  

815: \begin{comment}

816: This is the content of~\cite[Lemma 3.2]{gks}.  

817:  Since the proof there omits some details, we fill them in here:

818:  By joining two copies of~$D$ across the boundary arc, we obtain a

819:  $2$-point metric~$D'$ whose completion boundary develops to

820:  the endpoints $p,q\in\S^2$ of~$\gamma$.  Consider the preimage

821:  $Y\subset D'$ of $X:=\S^2\setm\{p,q\}$.  Since any path in~$X$ lifts,

822:  $Y$ is a covering space of~$X$, hence either an annulus (if the

823:  degree is finite) or a disk (the universal cover).  But the first

824:  case cannot happen, for to extend~$Y$ to create the disk~$D'$

825:  we would have to insert a point developing to~$p$ or~$q$, and the

826:  developing map would branch there.  Thus $D'=Y$ is the universal

827:  cover of~$X$, a bi-infinite join of slit spheres.

828:  Any $2$-point metric can be decomposed along a diagonal into two such rays.

829: \end{comment}

830: Let us indicate an argument for this fact which is more

831: direct than \cite[Lemma~3.2]{gks}.  

832: Note that each point of a slit sphere is the endpoint

833: of a geodesic emanating from one of the endpoints of the slit;

834: each of these geodesics is minimizing and disjoint from the bounding slit.   

835: Thus the completeness of $D\cup\gamma$ gives

836: that $D$ contains a slit sphere %of slit length~$|\gamma|$ 

837: attached across the completion boundary~$\gamma$.  

838: Continuing, by induction, we see 

839: $D$ contains a ray $R$ of slit spheres attached across~$\gamma$.

840: But $R\cup\gamma$ is complete, and thus $R$ is relatively open and closed 

841: in~$D$.  This shows $D=R$.

842: 

843: \begin{comment}%-----------------------------------------------------

844: If $\alpha=2\pi$ we call $D$ a 

845: Let us first consider the case $k=2$.  

846: The two angles of any $2$-gon~$D$ agree.

847: % and $D$ is embedded if and only if this angle~$\alpha$ is at most~$2\pi$.

848: If $\alpha=2\pi$ we call $D$ a \emph{slit sphere},

849: as in \figr{slit-lune} (left),

850: because it develops to the complement in~$\S^2$ of a geodesic arc,

851: whose length we call the \emph{slit length}.

852: If the vertices of a $2$-gon~$D$ develop to antipodal points

853: of~$\S^2$, as in \figr{slit-lune} (right),

854: then we call~$D$ a \emph{lune} of angle~$\alpha$;

855: two complementary lunes, one convex and one concave.

856: 

857: otherwise, the edges of~$D$ necessarily develop to the same

858: unique minimizing geodesic in~$\S^2$, so $\alpha$ is a multiple of~$2\pi$.

859: 

860: If the vertices of a $2$-gon~$D$ develop to antipodal points

861: of~$\S^2$, as in \figr{slit-lune} (right),

862: then we call~$D$ a \emph{lune} of angle~$\alpha$;

863: two complementary lunes, one convex and one concave.

864: 

865: Note that a lune has a continuous family of diagonals,

866: while a slit sphere of slit length less than~$\pi$ has no diagonal.

867: It is easy to see that any $2$-gon can be decomposed by

868: diagonals into a union of slit spheres of equal slit length,

869: plus possibly (if the slit length is $\pi$) an embedded lune of

870: angle less than~$2\pi$.

871: The only $2$-gons without diagonals are slit spheres (with

872: slit length less than~$\pi$).

873: 

874: \begin{lemma}\label{lem:hasray}

875:   Let~$D$ be a spherical metric such that $\d D$ contains

876:   an arc~$\alpha$ developing to a great-circle arc $\gamma\subset\S^2$.

877:   Suppose there are infinitely many diagonals $\alpha_i\subset D$

878:   whose endpoints in $\partial D$ agree with $\alpha$

879:   and which develop to~$\gamma$. 

880:   Then $D$ is a ray of slit spheres.

881: \end{lemma}

882: \begin{proof}

883:   For each~$i$, the loop $\alpha\cup\alpha_i$ bounds a $2$-gon~$D_i$.

884:   The union $\bigcup D_i =: R$ has no boundary other than~$\alpha$,

885:   so it is, by definition, a ray of slit spheres.

886:   Clearly $R$ is an open subset of~$D$.

887:   On the other hand, the metric space $\hat R=\alpha\cup R$ is complete,

888:   hence closed in~$\hat D$.

889:   Thus $R=\hat R\cap D$ is closed in~$D$. 

890:   Since the disk~$D$ is connected, $D=R$ is a ray.

891: \end{proof}

892: \end{comment}

893: %-----------------------------------------------------

894: 

895: The analysis of $k$-point metrics can be reduced to $k$-gon metrics:

896: \begin{proposition} \label{pr:metrics}

897:   Given any $k$-gon, we % for $k\ge 2$, we

898:   can attach rays of spheres along its edges to give a $k$-point metric.

899:   Conversely, any $k$-point metric $D$ can be decomposed

900:   into a $k$-gon~$D'$ together with $k$ rays of spheres;

901:   we call $D'$ a \emph{truncation} of~$D$.

902: \end{proposition}

903: \begin{proof}

904:   Given a $k$-gon~$D$, we can attach to each edge~$e$

905:   a ray of spheres with slit length equal to the length of~$e$.

906:   The resulting metric has no completion boundary except for

907:   the~$k$ vertices of~$D$, so it is a $k$-point metric.

908: 

909:   Conversely, given a $k$-point metric~$D$, note that its

910:   vertices are cyclicly ordered.  If we fix one vertex~$v$, the others

911:   inherit a linear ordering; let~$w$ be the last vertex in this ordering.

912:   We claim there is a diagonal from~$v$ to~$w$.

913: 

914:   First consider some reference geodesic arc~$c_0$ from~$v$ into~$D$.

915:   By completeness, for each angle~$\theta\in\R$ there is a

916:   geodesic arc~$c_\theta$ making a signed angle~$\theta$ with~$c_0$ at~$v$;

917:   we continue each arc~$c_\theta$ only until either

918:   it hits $\d D$ (at another vertex)

919:   or it reaches length~$\pi$ staying inside~$D$.

920:   Whenever the first case happens, we have found a diagonal from~$v$.

921: 

922:   Suppose for all $\theta$ in some interval $(\alpha,\beta)$ the second case

923:   occurs.  All arcs~$c_\theta$ then end inside~$D$ at the same point~$v'$

924:   (developing antipodally to~$v$).  Since the angle around~$v'$ is only~$2\pi$, 

925:   we must have $\beta-\alpha\le2\pi$,

926:   meaning there is never an angular gap of more than~$2\pi$ between diagonals.

927:   Thus we find a sequence $(d_i)_{i\in\Z}$ of diagonals from~$v$,

928:   indexed in increasing order of~$\theta$.  

929:   (If some diagonal has length~$\pi$, it is part of a continuous family

930:   of diagonals for some range of~$\theta$; in this case we include only

931:   a discrete subset of these in our sequence.)

932: 

933:   Let $v_i$ be the endpoint of~$d_i$.

934:   Since each $d_i$ cuts~$D$ into two subdiscs,

935:   the sequence $(v_i)$ of vertices is increasing with respect to their

936:   linear order.

937:   Thus for~$i\ge I$, the point $v_i$ is a constant vertex.

938:   For each~$i>I$, the loop $d_i\cup d_I$ bounds a $2$-gon~$D_i$.

939:   The union $\bigcup D_i =: R$ has no boundary other than~$d_I$,

940:   and so is a ray of slit spheres.

941:   Since such a ray has no further

942:   completion points, $v_I$ must be~$w$, the neighbor of~$v$.

943:   This proves the claim: each vertex has a diagonal to the consecutive vertex.

944: 

945: % Clearly $R$ is an open subset of~$D$.

946: % On the other hand, the completion~$\hat R$ is closed in~$\hat D$.

947: % Thus $R=\hat R\cap D$ is closed in~$D$. 

948: % Since the disk~$D$ is connected, $D=R$ is a ray.

949: % 

950: % Applying Lemma~\ref{lem:hasray}, the metric beyond~$d_i$ is then

951: % a ray~$R$ of spheres. 

952: % Since such a ray has no further

953: % completion points, $v_i$ must be~$w$, the neighbor of~$v$.

954: % This proves the claim: each vertex has a diagonal to the consecutive vertex.

955: 

956:   Consider the set of $k$ diagonals in~$D$, connecting the pairs of

957:   consecutive vertices.  Since these are minimizing geodesics

958:   between consecutive vertices, they cannot intersect one another.

959:   Together, they bound a $k$-gon

960:   whose complement consists of $k$ rays of spheres.% like~$R$.

961: \end{proof}

962: 

963: \begin{example}

964:   We exhibit polygonal metrics with

965:   $k-2$ edges of length~$\pi$, for each $k\ge 3$. 

966:   By Proposition~\ref{pr:metrics} these examples can be 

967:   extended with $k$~rays of spheres to form $k$-point metrics

968:   with $k-2$ edgelengths~$\pi$.

969:   By the Main Theorem, these metrics establish

970:   the existence of $k$-unduloids with $k-2$ cylindrical ends.

971: 

972: %--------------------------------------------------------------------------

973: \figtwo{4gon}{4und2cyl}{5cm}{A 4-unduloid with two cylindrical ends}{

974: \put(26,54){$\pi$}

975: \put(33,31){$n$}

976: \put(73,46){$\pi-n$}

977: }{}

978: {Left: The solid arcs have lengths $\pi$ and $n$, while the dashed arc

979: has length $\pi-n$.  The $3$-gon $D_n$ develops to the upper hemisphere,

980: while the $4$-gon described in the text (with edgelengths $n$, $\pi$, $\pi$, $n$)

981: develops to the complement of the solid arcs.

982: Right: The classifying map takes this $4$-unduloid (with

983: necksizes $n$, $\pi$, $\pi$, $n$) to the $4$-point metric one

984: of whose truncations is the $4$-gon shown on the left. 

985: }

986: %--------------------------------------------------------------------------

987:   

988:   Let $D_n$ be a $3$-gon developing to a hemisphere, 

989:   with edgelengths~$\pi$, $n$, $\pi-n$ where $0<n<\pi$.

990:   Reflecting across the edge of length $\pi-n$, we obtain

991:   a $4$-gon with edgelengths~$n$, $\pi$, $\pi$ and~$n$, as in \figr{4gon-4und2cyl}.

992: 

993:   We can reflect a copy of~$D_n$ across an edge of length~$n$

994:   of the $4$-gon to obtain a $5$-gon with

995:   edgelengths~$n$, $\pi$, $\pi$, $\pi$, and~$\pi-n$.

996:   Continuing the same way, we obtain

997:   $k$-gons with $k-2$ edgelengths~$\pi$; the remaining edgelengths

998:   are $n$, $n$ when $k$ is even, and $n$, $\pi-n$ when $k$ is odd.

999:   Note that the $k$~vertices of these gluings always just develop

1000:   to the three original vertices of~$D_n$.

1001: \end{example}

1002: 

1003: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1004: \subsection{Polygonal metrics without diagonal}

1005: 

1006: We will use diagonals to decompose a $k$-point metric in~$\Dk$

1007: into embedded pieces.  

1008: Let us introduce the particular $k$-gons which arise in the decomposition,

1009: beginning with triangles ($3$-gons).

1010: \begin{definition}

1011:   A triangle is \emph{small} if its completion develops

1012:   into an open hemisphere.  

1013:   Equivalently, its developing image has area less than~$2\pi$, 

1014:   or it has angles less than~$\pi$.  

1015:   Its edge lengths are also less than~$\pi$.

1016: \end{definition}

1017: 

1018: In the case of a flat metric with piecewise geodesic boundary there is a

1019: set of diagonals which decomposes the metric into flat triangles.

1020: This fact was used by Cos{\'\dotlessi}n and Ros 

1021: to analyze coplanar minimal $k$-noids~\cite{cr}.

1022: For spherical metrics, the situation is more involved.

1023: Besides slit spheres and small triangles, our decomposition

1024: will include pieces of the following types:

1025: %

1026: \begin{definition}

1027:   % Let $k\ge 2$.

1028:   A $k$-gon in~$\Pk$ is \emph{concave} if all its

1029:   angles are at least~$\pi$ and no edge has length~$\pi$.

1030:   It is \emph{embedded} if its developing

1031:   map is an embedding into~$\S^2$ (its completion need not be embedded).

1032:   A \emph{marked lune} is a lune of angle at most~$2\pi$

1033:   with $k-2$ additional vertices marked on one of its edges, 

1034:   as in \figr{mlune-longdiag};

1035:   when the angle is less than $\pi$ we call it \emph{convex}.

1036: \end{definition}

1037: %--------------------------------------------------------

1038: \figtwo{mlune}{longdiag}{5cm}{Slit sphere and lune}{}

1039: { \put(29,46){$p$} \put(72,46){$q$} \put(85,60){$D$} }

1040: %\fig{mlune}{6cm}{Marked lune}{}

1041: %\fig{longdiag}{6cm}{Embedded concave polygons have no diagonals}

1042: {Left: A lune with $k-2$ marked points on one of its edges

1043: forms an embedded $k$-gon called a marked lune.

1044: Right: The complement of the small convex pentagon shown on the front of the 

1045: sphere is an embedded concave $5$-gon~$D$.  It has no diagonals.  

1046: The great circle arc $pq$ shown, for instance, is longer than~$\pi$

1047: and so by definition it is not a diagonal.  (A minimum-length path

1048: from~$p$ to~$q$ exists in~$\hat D$, but it stays within $\d D$.)}

1049: %--------------------------------------------------------

1050: 

1051: Slit spheres of slit length less than~$\pi$ are concave $2$-gons.

1052: A triangle without diagonals is embedded, 

1053: and is either concave or a marked lune or small.

1054: %Triangle metrics were considered in~\cite{gks}.  

1055: For larger~$k$, \figr{mlune-longdiag} shows an example of 

1056: an embedded concave $k$-gon, with no diagonals.

1057: Another concave example is a metric isometric to a hemisphere of~$\S^2$

1058: with $k$~vertices on its bounding equator;

1059: it admits a diagonal if and only if some

1060: pair of vertices is antipodal.

1061: For slit length less than~$\pi$, slit spheres 

1062: or chains of slit spheres are always concave.

1063: 

1064: Our goal (in \prop{conemb} below)

1065: is to show that a concave $k$-gon with no diagonals is embedded;

1066: for the special case of triangles this was shown in~\cite{gks}.

1067: % In fact, Lemma 3.5 shows this for 3-point metrics; for a triangle

1068: % it follows by joining rays to its edges.

1069: To formulate a lemma, let us fix a vertex~$v$ of~$D$

1070: and consider geodesics in~$D$ emanating from~$v$,

1071: as in the proof of Proposition~\ref{pr:metrics}.

1072: For each angle~$\theta\in\big(0,\alpha(v)\big)$,

1073: there is a unique geodesic arc~$c_\theta$ making an angle~$\theta$ 

1074: with the edge after~$v$; again we continue~$c_\theta$ only until

1075: either it hits $\d D$, or it reaches length~$\pi$ staying inside~$D$.

1076: Thus the only points of~$\d D$ contained 

1077: in an arc $c_\theta$ are its initial point~$v$ and perhaps its endpoint.

1078: 

1079: \begin{lemma}\label{lem:twocase}

1080:   Let $D$ be a $k$-gon which is not isometric to a hemisphere, and

1081:   suppose that $D$ has no diagonals. 

1082:   Then all edges of $D$ have length less than~$\pi$.

1083:   Moreover, at each vertex $v$ the following alternative holds.  Either

1084: %\centerline{ \begin{minipage}{.92\textwidth}

1085: \begin{itemize}

1086: \item[\ia] all arcs~$c_\theta$ can be continued to length~$\pi$, 

1087:   where they converge at a point $v'\in D$

1088:   which develops antipodally to~$v$, or 

1089: \item[\ii] all arcs~$c_\theta$ have length at most~$\pi$ 

1090:   and end on the interior of the same edge~$e$ (not incident to~$v$),

1091:   where they meet~$e$ transversely.

1092: \end{itemize}

1093:   The second case can only arise when $\alpha(v)<\pi$. 

1094: \end{lemma}

1095: 

1096: \begin{proof}

1097:   If some edge $e$ has length~$\pi$, 

1098:   then there is a sufficiently thin lune contained in~$D$ 

1099:   whose boundary includes~$e$.

1100:   But such a lune contains a continuous family of diagonals, 

1101:   joining the endpoints of~$e$.  

1102:   This contradiction shows the edges have length less than~$\pi$.

1103:   

1104:   Let us now prove the alternative.

1105:   Since~$D$ has no diagonals, no arc~$c_\theta$ can hit~$\d D$ at a vertex.

1106:   For any fixed~$\theta$,

1107:   either~$c_\theta$ reaches length~$\pi$ without encountering~$\d D$,

1108:   or $c_\theta$ first hits~$\d D$ in an edge $e$ transversely.

1109:   Since the edges have length less than~$\pi$, the edge~$e$

1110:   in the second case is one of the $k-2$ edges not incident to~$v$.

1111: %(so that $k\ge 3$ in this case).

1112:   We claim that each of these $k-1$ possibilities occurs for

1113:   an open set of angles~$\theta$.  

1114:   Since $(0,\alpha(v))$ is connected, it then follows that we must have

1115:   the same endpoint behavior for all angles~$\theta$.  

1116: 

1117:   To prove the claim, consider~$c_\theta$.  Let~$X$ be~$\d D$ with~$v$

1118:   and its two incident edges removed; if $c_\theta$ ends on

1119:   an edge~$e$ we also remove~$e$.  Then 

1120:   the two disjoint closed subsets~$c_\theta$ and~$X$ of~$\hat D$

1121:   have a positive distance~$\delta$.

1122:   Within the $\delta$-neighborhood of~$c_\theta$,

1123:   we can construct the required nearby geodesics starting from~$v$

1124:   and with the same endpoint behavior.

1125: 

1126:   It remains to show that for $\alpha(v)\ge\pi$, case \ii\ cannot occur.

1127:   If $\alpha>\pi$ then we have two opposite arcs from~$v$ hitting

1128:   the same boundary edge~$e$ transversely.  The two intersection points

1129:   of this great circle with~$e$ must be antipodal, contradicting

1130:   the fact that $e$ has length less than~$\pi$.

1131:   To rule out $\alpha=\pi$, we similarly use two limit arcs in~$\hat D$

1132:   which hit the same edge~$e$.  If~$e$ is transverse to the limit

1133:   arcs, we get the same contradiction.  Otherwise $e$ lies along the

1134:   limit arcs; however, by assumption $D$ is not a hemisphere, so in

1135:   this case one of the limit arcs must include a diagonal of~$D$,

1136:   a contradiction.

1137: \end{proof}

1138: 

1139: \begin{proposition}\label{pr:concave}

1140:   If a $k$-gon $D\in\Pk$ has no diagonal,

1141:   then either $D$ is concave or it is a small triangle. 

1142: \end{proposition}

1143: \begin{proof}

1144:   For $k=2$, the $2$-gons without diagonals are slit spheres,

1145:   and hence concave.  So we assume $k\ge 3$.

1146:   We first consider the case that all angles of~$D$ equal~$\pi$.

1147:   The developing image of the completion boundary is then contained

1148:   in a single great circle and, by the Gauss--Bonnet formula, 

1149:   the metric must have area~$2\pi$.  

1150:   Therefore $D$ is isometric to a hemi\-sphere.

1151:   It cannot have an edge of length~$\pi$ for then

1152:   it would have diagonals near this edge;

1153:   thus it is concave and we are done.

1154: 

1155:   We may thus assume that $D$ is not a hemisphere,

1156:   meaning the hypotheses of Lemma~\ref{lem:twocase} hold.

1157:   We suppose $D$ is not concave and claim $D$ must be a small triangle. 

1158:   In the following we use $\overline{vw}$ to denote

1159:   the edge between consecutive vertices~$v$ and~$w$, 

1160:   not including its endpoints.

1161: 

1162:   There is at least one vertex~$w$ with $\alpha(w)<\pi$.

1163:   Consider the consecutive vertices $t,u,v,w,x$ 

1164:   (not necessarily distinct for small~$k$). 

1165:   We apply the lemma to the arcs~$c_\theta$ at~$v$. 

1166:   Because the length of~$\overline{vw}$ and the angle at~$w$ are

1167:   both less than~$\pi$, as in \figr{nodiag},

1168:   arcs~$c_\theta$ near~$\overline{vw}$ end on edge~$\overline{wx}$.

1169:   The lemma then says all $c_\theta$ end on~$\overline{wx}$;

1170:   this implies that we also have $\alpha(v)<\pi$.

1171: %--------------------------------------------------------

1172: \fig{nodiag}{6cm}{Polygonal metrics with no diagonals}

1173: { \put(21,40){$t$} \put(46,17){$u$}

1174:   \put(56,50){$v$} \put(81,38){$w$} \put(54,90){$x$} }

1175: {The rays $c_\theta$ from $v$ initially hit~$\overline{wx}$

1176: so by Lemma~\ref{lem:twocase} they all do, and we deduce

1177: that, unlike in this picture, $\alpha(v)<\pi$.}

1178: %--------------------------------------------------------

1179:   

1180:   Repeating this argument from~$v$, we find next that $\alpha(u)<\pi$.

1181:   Now consider again the arcs $c_\theta$ at~$v$.

1182:   They have length less than~$\pi$ and

1183:   end on the edge~$\overline{wx}$ when~$\theta$ is small, while they 

1184:   end on the edge~$\overline{tu}$ for~$\theta$ close to $\alpha(v)$.

1185:   By the lemma, we must have $\overline{wx}=\overline{tu}$, 

1186:   implying that $k=3$; moreover all edges and angles are less than~$\pi$.  

1187:   That is, $D$ is a small triangle, as claimed.

1188: \end{proof}

1189: 

1190: We can characterize concave $k$-gons without diagonals further:

1191: \begin{proposition} \label{pr:conemb}

1192:   A concave $k$-gon with no diagonals is embedded.  

1193: \end{proposition}

1194: \begin{proof}

1195:   It follows as in the previous proof that 

1196:   if all vertex angles equal~$\pi$ then

1197:   equivalently the metric is a hemisphere and hence embedded.

1198:   Otherwise Lemma~\ref{lem:twocase} is applicable. 

1199: 

1200:   We first claim that $\alpha(w)\in[\pi,2\pi]$ for each vertex~$w$.

1201:   Since $D$ is concave, $\alpha(w)\ge\pi$,

1202:   so if we apply Lemma~\ref{lem:twocase} we are in case~\ia.

1203:   That is, all arcs~$c_\theta$ from $w$ continue to length~$\pi$,

1204:   where they end at a common point $w'\in D$.  We call their

1205:   union the \emph{vertex lune} of~$D$ at~$w$.  Since $w'$ is an

1206:   interior point of~$D$, and~$D$ immerses into~$\S^2$ under the developing map, 

1207:   as in the proof of Proposition~\ref{pr:metrics},

1208:   the angle $\alpha(w)$ of this lune cannot exceed~$2\pi$.

1209:   This proves the claim.

1210: 

1211:   We then use induction on~$k$.

1212:   For $k=2$, we know a $2$-gon without diagonals is an embedded slit sphere.  

1213:   For the induction step, we will show that given a $k$-gon~$D$,

1214:   we can join a triangle to it to obtain a $(k-1)$--gon~$D'$ that still

1215:   has no diagonals.  By the induction hypothesis, $D'$ is embedded, 

1216:   and so the same must hold for its subset~$D$.

1217: 

1218:   Since the $k$-gon is not a hemisphere

1219:   we can choose a vertex~$v$ with $\alpha(v)\in(\pi,2\pi]$.

1220:   Let us assume for now that $\alpha(v)<2\pi$.

1221:   Then the three consecutive vertices $u,v,w$ of~$D$

1222:   develop to the three vertices of a small

1223:   triangle~$\Delta$, which is locally exterior to the image of~$D$.

1224:   We define a polygonal metric~$D'\in\D_{k-1}$, as in \figr{concind},

1225:   isometric to the union of~$D$ with~$\Delta$, glued across

1226:   the two edges incident to~$v$.

1227:   More precisely, we glue

1228:   the closure of~$\Delta$ to $D$ and then remove~$\overline{uw}$.

1229:   The completion boundary of~$D'$ agrees with that of~$D$, 

1230:   except that $\overline{uv}$ and~$\overline{vw}$ are

1231:   replaced by~$\overline{uw}$.

1232: %

1233: \fig{concind}{6cm}{Concave metrics with no diagonal are embedded.}

1234: { \put(57,23){$u$} \put(81,50){$v$} \put(64,77){$w$}

1235:   \put(77,28){$D$} \put(67,50){$\Delta$} }

1236: {By induction, the $(k-1)$--gon~$D'$, obtained by gluing

1237: the triangle~$\Delta$ to the concave $k$-gon $D$, is embedded.}

1238: 

1239:   The angles at~$u$ and~$w$ have only increased under this gluing,

1240:   so $D'$ is still concave.

1241:   We claim that $D'$ does not have a diagonal.

1242:   To see this, consider the open hemisphere $H\subset D'$,

1243:   bounded by $\overline{uw}$, and contained 

1244:   in the union of~$\Delta$ with the vertex lune at $v\in D$.

1245:   No vertices of $D'$ are in~$H$, so a

1246:   diagonal of~$D'$, being a minimizing geodesic with endpoints outside~$H$,

1247:   stays outside~$H$.  Hence any diagonal of~$D'$ stays outside~$\Delta$

1248:   and is thus also a diagonal of~$D$.

1249:   However, by assumption there are no such diagonals.

1250:   This completes the inductive proof under the assumption that $\alpha(v)<2\pi$.

1251: 

1252:   When $\alpha(v)=2\pi$, essentially the same ideas apply, except that

1253:   $\Delta$ degenerates to a segment.  When $u$ and $w$ develop to

1254:   distinct points, we obtain exactly as above a metric~$D'\in\P_{k-1}$

1255:   with no diagonals.  When~$u$ and~$w$ develop to the same point,

1256:   note that we must have $k\ge4$.  Here we get $D'\in\P_{k-2}$

1257:   by gluing the edge $\overline{vw}$ (being the degenerate triangle~$\Delta$) 

1258:   to~$D$, and removing the vertex $u=w$.  Again by the same argument,

1259:   the metric~$D'$ cannot have any diagonals.

1260: \end{proof}

1261: 

1262: Note that an embedded concave polygon has diagonals only if it has

1263: two antipodal vertices; in this case it is a lune with angle in $[\pi,2\pi]$,

1264: possibly with extra vertices marked along its edges.

1265: 

1266: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1267: \subsection{Decomposition of polygonal and $k$-point metrics}

1268: 

1269: We will now use the previous proposition to decompose any polygonal

1270: metric into a finite union of embedded pieces.

1271: Each piece is either a small triangle,

1272: or an embedded concave polygon with no diagonals,

1273: or a marked lune of angle less than~$\pi$.

1274: It will follow immediately that any

1275: $k$-point metric can be decomposed into the same pieces.

1276: 

1277: \begin{theorem}  \label{thm:decomp}

1278:   %For $k\ge2$, a

1279:   Any polygonal metric $D\in\Pk$ can be decomposed

1280:   by diagonals into a finite union of slit spheres,

1281:   small triangles, embedded concave polygons without diagonals,

1282:   and marked lunes.

1283: \end{theorem}

1284: \begin{proof}

1285:   By Propositions~\ref{pr:concave} and \ref{pr:conemb}, 

1286:   if $D$ has no diagonals, it is already one of the allowed pieces.  

1287:   Otherwise, we work by induction on~$k$, using diagonals to simplify $D$.  

1288:   Suppose that $D$ has a diagonal~$d$ connecting two nonconsecutive vertices. 

1289:   In this case the induction step is immediate

1290:   since $d$ splits $D$ into two pieces each with fewer edges. 

1291: 

1292:   The remaining case is when $D$ has a diagonal connecting consecutive

1293:   vertices, but no diagonals connecting nonconsecutive ones.

1294:   This includes the base case ($k=2,3$) of the induction on~$k$.

1295:   Here we will establish the decomposition directly, without using 

1296:   the induction hypothesis.

1297:   Instead we work by induction on the area of~$D$ (in multiples of~$4\pi$).

1298: 

1299:   So we have a metric~$D$, of area less than $4\pi n$ for some~$n\in\N$,

1300:   where all diagonals connect consecutive vertices.

1301:   Choose a shortest diagonal~$d$, and call its endpoints $v$ and~$w$.

1302:   Suppose first~$d$ has length less than~$\pi$.

1303:   Then the $2$-gon bounded by~$d$ and the edge $\overline{vw}$ of~$D$

1304:   is a union of $m$ slit spheres, where $1\le m\le n$.  

1305:     % m \ge 1 since diagonal is interior, m\le n by area comparison.

1306:   The $k$-gon on the other side of~$d$ has area $4\pi m$ less than 

1307:   that of~$D$.  Thus for the base case $n=1$ this is impossible, 

1308:   meaning that no such diagonal can exist so that $D$ itself is embedded.

1309:   On the other hand, for $n>1$ the metric decomposes into the allowed 

1310:   embedded pieces by induction.

1311:   

1312:   The more difficult case is when~$d$ has length~$\pi$.

1313:   In this case we consider the lune~$L$ of maximal angle which is

1314:   contained in~$D$ and incident to the edge~$\overline{vw}$.

1315:   Then the other edge~$d'$ bounding~$L$ must include some point of~$\d D$.

1316:   Since there are no diagonals shorter than~$\pi$,

1317:   the arc~$d'$ must lie entirely in~$\d D$.

1318:   Thus $D$ is a marked lune, though with angle~$\alpha$

1319:   possibly greater than~$2\pi$.

1320:   It has a continuous family of diagonals,

1321:   so we can use these to decompose $D$ into $n$~slit spheres

1322:   and one marked lune of angle at most~$2\pi$ (adjacent to~$d'$).

1323:   Thus also in this case we have arrived at a decomposition into

1324:   the desired embedded pieces.

1325: \end{proof}

1326: 

1327: \begin{comment} %-----------------------------------------------------

1328: % John, Darmstadt 2/02  

1329: Claim: can decompose into three kinds of pieces:\\

1330: 1. small triangles\\

1331: 2. embedded concave l-gons $l\ge 2$ without diagonals\\

1332: 3. lunes (maybe not marked) of angle $< pi$

1333: 

1334: Claim: Each basic piece can be decomposed into small triangles.

1335: 1. is obvious.

1336: 3. On each diagonal of length~$\pi$ let us

1337: introduce the midpoint as a marker point.

1338: 2. if area < 4pi, find p in complement of interior, cone to boundary

1339: from antipodal -p.  If area=4pi, disk is complement of some slit.

1340: Introduce three new vertices, antipodal to midpoint of slit and at poles.

1341: 

1342: For the record only (currently not needed):

1343: 

1344: Note: changes in combinatorics are as follows:

1345: A concave l-gon with one angle going to pi becomes a concave

1346: (l-1)-gon union a convex triangle (area ~ 0).  (includes l=3,4)

1347: A convex triangle shrinking to 0 is either adjacent to a concave

1348: piece (inverse of above), or is adjacent to another convex

1349: triangle; in the latter case we could have flipped the diagonal

1350: ahead of time, and then see no change.

1351: If convex triangle shrinks to 0 area, 

1352: Claim: can decompose into three kinds of pieces:

1353: 1. small triangles

1354: (strictly convex <=> contained open hemisphere <=> area < 2 pi)

1355: 2. embedded concave l-gons (l>=2) without diagonals

1356: 3. lunes (maybe not marked) of angle < pi

1357: \end{comment}

1358: %-----------------------------------------------------

1359: 

1360: Given a $k$-point metric $D$ we can 

1361: remove its rays as in Proposition~\ref{pr:metrics} 

1362: and then apply Theorem~\ref{thm:decomp} to obtain the following 

1363: decomposition:

1364: %

1365: \begin{corollary}\label{co:kptdec}

1366:   Any $k$-point metric $D\in\Dk$ can be decomposed into $k$ rays

1367:   of slit spheres together with a finite union of 

1368:   slit spheres, small triangles, embedded concave polygons

1369:   with no diagonals, and marked lunes. 

1370: \end{corollary}

1371: %

1372: When applied to the Hopf projection $D=\Phi(M)$ of

1373: a coplanar $k$-unduloid $M\in\Mk$ this decomposition 

1374: can be regarded as a decomposition of~$M$ into bubbles.

1375: %

1376: 

1377: There is an equivalent reformulation of the decomposition in terms of

1378: graphs dual to the tessellating pieces.  This graph is a tree,

1379: topologically equivalent to the skeletal graphs considered earlier by

1380: Schoen~\cite{sch}, Kusner~\cite{kmsri}, and Kapouleas~\cite{kap}.

1381: % More precisely, we can associate to $P$ a 

1382: % connected graph $G(P)$ with no terminal (one-valent) nodes,  

1383: % infinitely many two-valent nodes, and at most 

1384: % $k-3$~multivalent nodes, of total valence $k-2$;

1385: % each twovalent node represents a slit sphere, and

1386: % each node of valence $\kappa$ represents an embeddable $\kappa$-gon.

1387: 

1388: Note that the decomposition of a $k$-point metric in general is not unique. 

1389: Clearly, the location of diagonals of length $\pi$ is not determined;

1390: in addition, many metrics admit combinatorially different choices of

1391: diagonals, with the simplest example being a convex square.  

1392: 

1393: In Proposition~\ref{pr:metrics} we removed rays (of slit spheres)

1394: from a $k$-point metric.

1395: In view of the decomposition established by the corollary above,

1396: we can make this more precise:

1397: \begin{definition}

1398:   Given a $k$-point metric $D\in\Dk$, a \emph{small truncation}

1399:   is a metric $\Dbox\in\Pk$ obtained from~$D$

1400:   by removing (closed) rays so that no slit 

1401:   sphere remains incident to any bounding edge.

1402: \end{definition}

1403: When $D$ has no pair of consecutive vertices at distance~$\pi$, 

1404: the small truncation is uniquely defined.  In the general case,

1405: the ambiguity is by less than one slit sphere at each of the $k$~ends.  

1406: 

1407: \figr{ex6pt} illustrates the results of this section.

1408: \fig{ex6pt}{7cm}{A six-point metric on the open disk}

1409: {}{This $6$-point metric~$D$ on the open disk is decomposed

1410:     by the six thick arcs into its small truncation $\Dbox$

1411:     and six rays of slit spheres.  In turn, the polygon

1412:     metric $\Dbox$ is, as in \cor{kptdec}, decomposed into two

1413:     triangles, three slit spheres, and one $4$-gon (necessarily

1414:     embedded, concave, and without diagonals).  A $6$-point

1415:     metric with such a decomposition will correspond (under

1416:     our classifying map~$\Phi$) to a $6$-unduloid with two

1417:     midsegments, one having two unduloid bubbles and the

1418:     other having just one.}

1419: 

1420: 

1421: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1422: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1423: \section{The space of $k$-point metrics}

1424: 

1425: We now show that $\Dk$ is a connected manifold. 

1426: As for $k=3$ we consider the projection~$\Pi$ from $\Dk$ to $k$-tuples 

1427: of points on $\S^2$, given by restricting the developing map 

1428: of a $k$-point metric to its vertices.

1429: For $k=3$ the mapping $\Pi$ is a diffeomorphism from~$\D_3$ 

1430: to the open 3-ball of triples in~$\S^2$, as in~\cite{gks}.

1431: For larger~$k$ this is only true locally.  

1432: From now on we assume $k\ge 3$.

1433: 

1434: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1435: \subsection{$\Dk$ is a manifold}

1436: 

1437: Let us define ${\mathcal S}_k$ to be the set of $k$-tuples

1438: of sequentially distinct points in~$\S^2$:

1439: $$

1440:   {\mathcal S}_k:=\{(p_{1},p_{2},\ldots,p_{k})\in (\S^2)^k, 

1441:   p_{i}\not=p_{i+1} \text{ for }i=1,\ldots,k\}.

1442: $$

1443: Define ${\mathcal A}_k\subset {\mathcal S}_k$ to be the set 

1444: $\{(p,q,\ldots,p,q): p\not=q\in\S^2\}$ for $k$~even and the empty set

1445: for $k$~odd.

1446: Note that ${\mathcal A}_k$ contains all $k$-tuples in ${\mathcal S}_k$ 

1447: consisting only of a pair of antipodal points (namely, when $q=-p$).

1448: Thus the action of $\SO(3)$ is free on 

1449: the manifold ${\mathcal S}_k\setm {\mathcal A}_k$, 

1450: % S_k is a manifold and A_k is closed

1451: and the quotient

1452: $$

1453:   \T_k:=\big({\mathcal S}_k\setm {\mathcal A}_k\big) \big/ \SO(3),

1454: $$

1455: is a manifold.

1456: 

1457: \begin{proposition}\label{pr:dk2tk}

1458:   For $k\ge3$, the completion boundary of any $k$-point metric develops

1459:   to a $k$-tuple in $\Tk$.  

1460:   This gives a surjective map $\Pi\colon \Dk\to\Tk$,

1461: \end{proposition}

1462: Note, however, that for $k=2$ the image of $\D_2$ includes

1463: $(p,-p)\in {\mathcal A}_2={\mathcal S}_2$.

1464: This fact explains why $\D_2\isom(0,\pi]$ fails to be a manifold

1465: but instead has a boundary point~$\pi$.  Of course,

1466: $\pi$~corresponds to the cylinder in $\M_2\isom\D_2$,

1467: where rotations also fail to act freely.

1468: %

1469: \begin{proof}

1470:   For $D\in\Dk$, Proposition~\ref{pr:metrics} gives

1471:   a $k$-gon $\Dbox\in\Pk$, obtained by removing rays from~$D$.

1472:   The spheres of each ray have nonzero slit length, 

1473:   and so consecutive vertices must develop to distinct points. 

1474:   Thus any developing image of the completion boundary $\d D$

1475:   lies in ${\mathcal S}_k$.

1476: 

1477:   To show the image is not in ${\mathcal A}_k$, suppose

1478:   to the contrary that the completion points of $D\in\Dk$ develop 

1479:   onto just one pair $p\not=q$ of points in~$\S^2$.

1480:   Then the decomposition of Corollary~\ref{co:kptdec}

1481:   can only contain slit spheres of slit length $\dist(p,q)$.

1482:   Such slit spheres

1483:   cannot be joined together to produce a $k$-gon for $k>2$.

1484: 

1485:   Finally we show~$\Pi$ is surjective.  

1486:   For $k=3$ this is obvious, while for $k>3$ this follows by induction:

1487:   There exists a nonconsecutive pair $p_i,p_j$ of vertices

1488:   with distinct images.  By joining two metrics along a

1489:   diagonal developing to~$\overline{p_ip_j}$

1490:   we obtain the desired $k$-point metric.

1491: \end{proof}

1492: \begin{corollary} \label{co:evenk}

1493:   For $k$ even, at least two ends of a $k$-unduloid

1494:   are non-cylindrical, and so $n_1+\cdots+n_k < k \pi$ holds.

1495: \end{corollary}

1496: \begin{proof}

1497:   If $k-1$ ends were cylindrical,

1498:   then the $k$-tuple~$\Psi(M)$ 

1499:   would alternate between a pair of antipodal points

1500:   (making also the $k^{\text{th}}$ necksize cylindrical).

1501:   But this is impossible by~\prop{dk2tk}.

1502: \end{proof}

1503: 

1504: The map~$\Pi$ relates the $k$-tuple map~$\Psi$ and classifying

1505: map~$\Phi$ introduced in Section~1:  For any $k$-unduloid $M\in\Mk$,

1506: we have $\Phi(M)\in\Dk$ and $\Psi(M)=\Pi(\Phi(M))\in\Tk$.

1507: 

1508: Let $\Tk^*\subset\Tk$ denote the open set of 

1509: all pairwise distinct $k$-tuples in $\S^2$ (mod $\SO(3)$),

1510: and let $\Dk^*:=\Pi^{-1}(\Tk^*)\subset\Dk$.

1511: For $k=3$ we have $\T_3^*=\T_3$ and so $\D_3^*=\D_3$.

1512: 

1513: \begin{theorem} \label{th:mfld}

1514:   For any~$k\ge3$, the map $\Pi\takes\Dk\to\Tk$ is a covering map over $\Tk^*$.

1515:   It is a local diffeomorphism everywhere, implying that~$\Dk$ is a

1516:   real-analytic manifold of dimension $2k-3$.

1517: \end{theorem}

1518: \begin{proof}

1519:   Consider a $k$-point metric $D\in\Dk$, which maps to some $k$-tuple 

1520:   $T\in\Tk$.   Pick any representative $(v_1,\ldots,v_k)\in(\S^2)^k$

1521:   for $T$; the $v_i$ are the images of the completion boundary points $q_i$

1522:   of~$D$ under some developing map~$\phi$.

1523: 

1524:   By introducing a finite set of additional vertices

1525:   we can find a triangulation~$\Delta$ of~$\S^2$ by small triangles

1526:   whose vertices include the~$v_i$.  Passing to a

1527:   subdivision if necessary, we may assume that the~$v_i$ are not

1528:   adjacent in~$\Delta$, in other words that their 

1529:   star neighborhoods~$S_i\subset\S^2$ are disjoint when $v_i\ne v_j$.  

1530:   (Note that we may have $v_i=v_j$ for $i\ne j$ if $T\not\in\Tk^*$.)

1531:   We claim that $\Pi$ is a local diffeomorphism from a neighborhood of~$D$ 

1532:   onto $U:=(S_1\times\cdots\times S_k)/\SO(3)$.  

1533:   It follows that $\Dk$ is a manifold of dimension $2k-3$; 

1534:   we endow it with the real-analytic structure coming from these charts.

1535: 

1536:   To prove the claim, note that the pullback $\phi^{-1}(\Delta)$

1537:   gives a triangulation of the metric space~$\hat D$.  For any

1538:   $(w_1,\ldots,w_k)\in S_1\times\cdots\times S_k$, we construct

1539:   a completed $k$-point metric with that developing image as follows.

1540:   Each triangle of $\phi^{-1}(\Delta)$ incident to the completion

1541:   boundary~$\d D$ has exactly one vertex $q_i\in\d D$.  

1542:   We continuously change the metric

1543:   on any such triangle according to the motion of $v_i$ to~$w_i$, but we

1544:   fix the metric on the remaining triangles (those not incident to $\d D$).

1545:   Under this new $k$-point metric, $q_i$ develops to~$w_i$; this confirms

1546:   that we have constructed a continuous local inverse for~$\Pi$.

1547: 

1548:   To show that $\Pi$ is a covering over $\Tk^*$, note that $\Tk^*$ is

1549:   exactly the set where the $v_i$ are distinct.  In this case, the

1550:   neighborhood $U=(S_1\times\cdots\times S_k)/\SO(3)$ is trivially

1551:   covered by $\Pi^{-1}(U)$ because we can use the same triangulation~$\Delta$

1552:   for any $D\in\Pi^{-1}(U)$.

1553: \end{proof}

1554: 

1555: \begin{comment}

1556:   We note that the covering map is not a regular covering of $\Tk^*$.

1557:   Also, $\Tk\setm\Tk^*$ has codimension two.  Some of the loops in $\Tk^*$

1558:   that are not unwrapped in the covering space get filled in in certain

1559:   sheets of $\Dk$.

1560: \end{comment}

1561: 

1562: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1563: \subsection{$\Dk$ is connected}

1564: 

1565: What initially led us to believe that~$\Mk$ is connected

1566: was that we can imagine a path from any given $k$-unduloid $M\in\Mk$

1567: to a standard one, say a dihedrally symmetric one.

1568: In particular, we could deform~$M$ such that one of its midsegments becomes

1569: cylindrical (as in \figr{curvature blow up} viewed from right to left),

1570: and then (as in \figr{rhombic with cyl segment})

1571: telescope inward along that midsegment. 

1572: Repeating this process until all midsegments have been retracted, 

1573: we could finally adjust the necksizes of the ends until they are equal. 

1574: 

1575: This intuitive idea will guide our proof of connectedness of~$\Dk$.

1576: Our first lemma singles out a class of $k$-point metrics 

1577: for which the decomposition of Corollary~\ref{co:kptdec} is unique.

1578: Let us introduce the subset $\E_k\subset\Tk$ 

1579: of equatorial $k$-tuples subject to the condition

1580: that all points occur in counterclockwise order on the equator,

1581: and consecutive pairs have distances in $(0,\pi)$.

1582: %

1583: \begin{lemma} \label{le:equatorial}

1584:   Suppose that $D\in\Dk$ is a metric such that $\Pi(D)\in\E_k$ 

1585:   and $D$ has no diagonal of length~$\pi$. 

1586:   Then the decomposition of~$D$ asserted in Corollary~\ref{co:kptdec} 

1587:   is unique.  

1588:   More specifically, the small truncation $\Dbox$ is a union of

1589:   some number $m(D)\ge 0$ of pieces isometric to slit spheres

1590:   and exactly one piece isometric to a hemisphere. 

1591: \end{lemma}

1592: \begin{proof}

1593:   The decomposition pieces %from Corollary~\ref{co:kptdec} 

1594:   can only be slit spheres or embedded concave $k$-gons without diagonals.

1595:   Let us consider the developing image of~$D$.

1596:   The diagonals of~$D$ can only develop to the equator, 

1597:   and so the concave $k$-gons are isometric either to hemispheres 

1598:   or to slit spheres.   Moreover, all rays of slit spheres

1599:   are joined from the same 

1600:   side to the equator, meaning that 

1601:   the small truncation $\Dbox$ has area an odd multiple of~$2\pi$.

1602:   % This follows from Gauss-Bonnet: Exterior angles are

1603:   % 0 or pi, and there is an even number of the pi angles

1604:   % (a loop in the equator S^1 has an even number of turnings)

1605:  

1606:   Thus there is at least one

1607:   decomposition piece isometric to a hemisphere.

1608:   The orientation of~$D$ means that any such hemisphere~$H$

1609:   has its bounding vertices in increasing cyclic order;  

1610:   however, the developing image of~$\d H$ covers the equator once,

1611:   so~$H$ must develop to the upper hemisphere. 

1612:   If there were two or more hemispheres in the decomposition,

1613:   we could find a pair 

1614:   connected by a chain of pieces isometric to slit spheres;

1615:   this pair would develop to complementary hemispheres, contradicting

1616:   the observation above. 

1617: 

1618: % This leads to the following combinatorial picture for a metric in~$F$:

1619: % It corresponds to an assignment of $0$ to~$k-3$ nonintersecting diagonals 

1620: % in~$D$, each labelled with a number in~$\N_0$. Indeed, this number is the 

1621: % number of slit spheres enclosed by the edges of two concave $l$-gons.

1622: % Our metric $D_0$ has all labels equal to~$0$, and it is the unique metric

1623: % in $F$ with this property.

1624:   

1625:   Let us now show the decomposition is unique.

1626:   We use preimages of the equator to subdivide~$D$ into hemispherical tiles,

1627:   developing alternately to the upper and lower hemisphere.

1628:   This canonical decomposition is the same as the one above, except that

1629:   each slit sphere has been divided (along an internal arc of length

1630:   greater than~$\pi$) into two hemispheres.

1631:   Thus the hemisphere piece in the original decomposition is distinguished

1632:   as the unique tile which has all edge lengths less than~$\pi$.

1633: \end{proof}

1634: 

1635: We can now prove the following:

1636: \begin{theorem}\label{th:connected}

1637:   The manifold $\Dk$ is connected.

1638: \end{theorem}

1639: \begin{proof}

1640:   By Theorem~\ref{th:mfld} the complement of~$\Dk^*$

1641:   has codimension~2 in~$\Dk$; for $k=3$ it is empty. 

1642:   Thus proving $\Dk$ is connected is equivalent 

1643:   to showing that $\Dk^*$ is connected.

1644:   %

1645:   % For $i\not=j\pm1$, fixing $p_i$ to $p_j$ is two equations ==> codim = 2

1646:   % Then $\Tk-\Tk^*$ is the union of these sets 

1647:   % over all $i,j$ with i\not=j\pm 1$.

1648:   % For $k=3$ the space of such $i,j$'s is empty!

1649:   % By the local diffeomorphism property, dimensions in $\Dk$ agree.

1650:   %

1651:   To prove the latter, it is sufficient to pick some $T\in\Tk^*$ and

1652:   show that the deck transformation group of the covering

1653:   $\Pi\colon\Dk^*\to\Tk^*$

1654:   acts transitively on the particular fiber~$F:=\Pi^{-1}(T)\subset\Dk^*$.

1655: 

1656:   To define~$F$, we pick a $k$-tuple $T\in\E_k$ 

1657:   with no pairs of points antipodal.

1658:   For instance, if $k$ is odd, 

1659:   then $T$ could be $k$~points with maximal dihedral symmetry,

1660:   while for even~$k$, 

1661:   a slight perturbation of this configuration within~$\E_k$ would work.  

1662:   Since a metric $D\in F$ cannot have diagonals of length~$\pi$,

1663:   Lemma~\ref{le:equatorial} is applicable to~$D$ and gives a well-defined

1664:   number~$m(D)$ of pieces isometric to slit spheres

1665:   in the unique small truncation~$\Dbox$.

1666:   Note there is a unique metric in~$D_0\in F$

1667:   with decomposition number~$m(D_0)=0$, isometric to 

1668:   a hemisphere with $k$~rays of spheres attached.

1669: 

1670:   For $k=3$, any $D\in F$ has $m(D)=0$ since the

1671:   small truncation~$\Dbox$ is the unique triangle determined by $\Pi(D)\in\T_3$.

1672:   This shows $\D_3^*=\T_3^*$; but $\T_3^*=\T_3$ is an open 3-ball 

1673:   (see~\cite{gks}). 

1674:   So let us from now on assume $k\ge 4$.

1675: 

1676:   For $D\in F$ we will construct a loop in~$\Tk^*$ based at~$T$

1677:   whose lift in~$\Dk^*$ runs from $D$ to~$D_0$.  

1678:   We can assume $D$ has decomposition number $m(D)>0$;

1679:   by induction it is sufficient to construct a loop~$\gamma$

1680:   whose lift~$\tilde\gamma$ connects~$D$ to some $D'\in F$

1681:   for which~$m$ has decreased by one.

1682:   We will choose~$\gamma$ as the concatenation $\alpha * \lambda * \alpha^{-1}$

1683:   of three subpaths, where $\lambda$ is a loop in~$\Tk^*$

1684:   and $\alpha$ is a path in~$\E_k$.

1685:   On the level of \CMC/ surfaces, the lift of~$\lambda$ telescopes

1686:   a midsegment, while $\alpha$ adjusts necksizes.

1687: 

1688:   Since $m(D)>0$, the metric $D$ has at least one

1689:   diagonal $d$ between two nonconsecutive vertices $v_i$, $v_j$.

1690:   There is a path $\alpha$ in $\E_k$ from $T=\Pi(D)$ 

1691:   to some $k$-tuple $S\in\E_k$ in which~$v_i$ and~$v_j$ are antipodal.

1692:   Among pairs of vertices joined by a diagonal,

1693:   we can assume that no pair becomes antipodal along our path $\alpha$, 

1694:   since otherwise we instead take the first such diagonal to be~$d$,

1695:   and replace $\alpha$ by the shortened path.

1696: 

1697:   The path~$\alpha$ has a unique lift $\mtilde\alpha$ in~$\Dk^*$ through~$D$.

1698:   We let $E\in\Pi^{-1}(S)$ be its terminal metric.

1699:   By Lemma~\ref{le:equatorial} the decomposition of each metric

1700:   along~$\mtilde\alpha$ is unique.

1701:   Hence up until the terminal time, the diagonal~$d$ as well as the

1702:   entire decomposition continues uniquely along the lift~$\mtilde\alpha$. 

1703:   At $E$ the diagonal and decomposition are no longer unique, but

1704:   each can be defined by taking the limit along~$\mtilde\alpha$.

1705:   The limiting diagonal $e$ in~$E$ then develops to a semicircle 

1706:   contained in the equator, 

1707:   bounded by the antipodal vertices $v_i$ and~$v_j$ of~$S$.

1708: 

1709:   We now define $\lambda$ so that it lifts to a path in $\Dk^*$ 

1710:   which removes a slit sphere whose boundary contains the diagonal~$e$.

1711:   Of the pieces adjacent to~$e$ at least

1712:   one---call it~$P$---is isometric to a slit sphere,

1713:   because there is only one piece isometric to a hemisphere 

1714:   in the decomposition.

1715:   Now consider the vertices developing into the image of~$e$.  

1716:   Note that by the definition of~${\mathcal E}_k$, these vertices are either

1717:   all the vertices with index between $i$ and~$j$ or all the complementary ones.

1718:   But~$e$ decomposes the metric into two connected halves, and these are

1719:   exactly the vertices that lie to one side of~$e$.

1720:   In order for the lift of~$\lambda$ to remove~$P$,

1721:   we revolve the vertices which develop into~$e$ 

1722:   by an angle of~$2\pi$ about the fixed endpoints $v_i,v_j$;

1723:   we must revolve in the direction of~$P$.

1724:   By a slight perturbation of $\alpha$ and~$S$, if necessary, 

1725:   we can assume that the orbits of the revolving vertices

1726:   stay disjoint from the fixed vertices.  

1727:   This ensures that $\lambda$ lies in~$\Tk^*$ and so has a unique lift 

1728:   $\tilde\lambda$ in~$\Dk^*$.  

1729:   

1730:   The complement in~$E$ of the slit sphere~$P$ has two components.

1731:   We can explicitly define the path~$\tilde\lambda$ by attaching these

1732:   components to a path of lunes whose angles decrease from $2\pi$ to~$0$.

1733:   By leaving the rest of the decomposition unchanged,

1734:   we obtain a natural decomposition of the metrics along~$\tilde\lambda$.

1735:   The decomposition of the terminal metric~$E'$ of~$\tilde\lambda$

1736:   then agrees with that of~$E$, except that the one slit sphere

1737:   is removed.  

1738:   

1739:   There is a unique lift $\mtilde\alpha^{-1}$ starting from $E'$;

1740:   it terminates at some $D'\in F$.

1741:   As with~$\mtilde\alpha$, the metrics along $\mtilde\alpha^{-1}$

1742:   (aside from~$E'$) have a unique decomposition.

1743:   This path of decompositions agrees with that along the reversed

1744:   path~$\mtilde\alpha$ except one slit sphere has been removed;

1745:   the limiting decomposition of this path therefore 

1746:   agrees with the decomposition at~$E'$.

1747:   The small truncation of~$D'$ has one less slit sphere than the 

1748:   small truncation of~$D$, so $m(D')=m(D)-1$.

1749: \end{proof}

1750: 

1751: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1752: \subsection{The space $\D_4$}

1753: 

1754: We now proceed to analyze in detail the structure of the

1755: space~$\D_4$ of $4$-point metrics, proving that it is

1756: diffeomorphic to a $5$-ball.

1757: Recall that $\T_4^*$ is the $5$-manifold of all pairwise distinct 

1758: spherical $4$-tuples $(p_1,p_2,p_3,p_4)$ modulo rotation;

1759: the $5$-manifold $\T_4$ additionally includes the $4$-tuples

1760: where either $p_1=p_3$ or $p_2=p_4$.

1761: 

1762: \begin{lemma}

1763:   The space $\T_4^*$ is a $\C$-bundle

1764:   whose base is the open $3$-ball with two proper arcs removed;

1765:   it thus has the homotopy type of $\S^1\vee\S^1$, a figure-8.

1766:   The space $\T_4$ is also a $\C$-bundle; the base is an open $3$-ball 

1767:   with a proper arc and a point removed, so

1768:   the homotopy type is $\S^1\vee\S^2$.

1769: \end{lemma}

1770: 

1771: \begin{proof}

1772:   Let us identify $\S^2$ with $\hat\C:=\C\cup\infty$.  

1773:   We use a rotation to move $p_4$ to~$\infty\in\hat\C$.

1774:   Then

1775:   $$ 

1776:     \T_4 \isom \Big(\{(p_1,p_2,p_3):

1777:       p_1,p_3\in\C, p_2\in\hat\C, p_1\ne p_2\ne p_3\}

1778:       \setm \{(p,\infty,p)\} \Big) \,\Big/\; \S^1, 

1779:   $$

1780:   where $\S^1$ acts diagonally by multiplication by~$e^{i\theta}$.

1781: 

1782:   We define $\X_4\subset\T_4$ to be the subset where $p_3=0$.

1783:   Then

1784:   $$

1785:     \pi\colon \T_4\to\X_4,\qquad

1786:     (p_1,p_2,p_3)\mapsto(p_1-p_3,p_2-p_3,0)

1787:   $$

1788:   is well-defined and indeed a deformation retract.

1789:   % from~$\T_4$ to~$\X_4$.  

1790:   In fact, $\pi$ is the desired bundle with fiber~$\C$:

1791:   the $\C$-valued expression $p_3/(p_2-p_3)$ is $\S^1$-invariant,

1792:   hence well-defined on~$\T_4$, and it gives a trivialization

1793:   away from $p_2=\infty$;

1794:   by symmetry a similar trivialization exists away from $p_1=p_3$.

1795: 

1796:   In our picture, $\T_4^*\subset\T_4$ is the subspace

1797:   where $p_1\ne p_3$ and $p_2\ne\infty$.  So if we define

1798:   $$

1799:     \X_4^*:=

1800:       \big\{(p_1,p_2,0): p_1, p_2\in\C\setm\{0\}, p_1\ne p_2\big\} / \S^1,

1801:   $$

1802:   the bundle $\T_4\to\X_4$ restricts to a $\C$-bundle $\T_4^*\to\X_4^*$.

1803: 

1804:   Now consider the smooth map $\phi\colon\X_4\to\C\times\R$ defined by

1805:   $\phi(p_1,p_2,0) := (z,t)$, where

1806:   $$

1807:     z := \frac{p_1}{p_2}, \qquad

1808:     t := \log\big(1+|p_1|\big)-\log\Big(1+\frac1{|p_2|}\Big)

1809: %     = \log\Big(|p_1|\frac{|p_1|+1}{|p_1|+|p_1/p_2|}\Big). 

1810: %     = \log\bigg(\frac{|p_1|+1}{\tfrac{|z|}{|p_1|}+1}\bigg). 

1811:       = \log\frac{1+|p_1|}{1+\sfrac{|z|}{|p_1|}}. 

1812:   $$

1813:   The map $\phi$ is well-defined since it is unchanged when multiplying

1814:   $p_1$ and~$p_2$ by the same factor~$e^{i\theta}$.  

1815:   Note that $z$ is a cross-ratio of the $4$-tuple

1816:   $$(p_1,p_2,p_3,p_4)=(p_1,p_2,0,\infty).$$

1817:   

1818:   We now examine the behavior of $\phi$ on~$\X_4^*$; 

1819:   here $z$ takes every value except~$0$ and~$1$.  

1820:   For any fixed $z=\sfrac{p_1}{p_2}$, the function

1821:   $$

1822:     r\mapsto \log \frac{1+r}{1+\sfrac{|z|}r}

1823:   $$

1824:   is strictly increasing from $\R^+$ onto~$\R$.

1825:   Thus $\X_4^*$ is diffeomorphic to its image

1826:   $\phi(\X_4^*)=\big(\C\setm\{0,1\}\big)\times\R$,

1827:   % z=0 <=> p_1=0 <=> p_1=p_3

1828:   % z=1 <=> p_1=p_2

1829:   and so has the homotopy type of $\S^1\vee\S^1$.

1830:   This completes the proof for~$\T_4^*$.

1831: 

1832:   The space $\X_4$ includes two open arcs in addition to~$\X_4^*$,

1833:   namely the arcs~$\gamma_{13}$, where $p_1=p_3$,

1834:   and~$\gamma_{24}$, where $p_2=p_4$.  Under~$\phi$, both

1835:   arcs map to the line $z=0$; the coordinate $t$

1836:   maps $(0,p_2,0)\in\gamma_{13}$ to $-\log\big(1+\tfrac1{|p_2|}\big)$, 

1837:   and $(p_1,\infty,0)\in\gamma_{24}$ to $\log(1+|p_1|)$.

1838:   Hence this coordinate

1839:   maps $\R^+$ diffeomorphically onto~$\R^-$ and~$\R^+$, respectively.

1840:   Thus $\phi(\X_4)$ is the union of $\phi(\X_4^*)$ with $0\times(\R\setminus 0)$.

1841:   We conclude $\phi$ is a diffeomorphism from~$\X_4$ to its image,

1842:   which is all of~$\C\times\R$ except for the origin 

1843:   % origin: (z,t)=0 <==> p_1=0 and p_2=\infty 

1844:   % but this is (0,\infty,0)\not\in\X_4: excluded point p_1=p_3 and p_2=p_4

1845:   and the line $\{1\}\times\R$.

1846:   This is homotopy-equivalent to $\S^2\vee\S^1$,

1847:   completing the proof for~$\T_4$.

1848: \end{proof}

1849: 

1850: Remembering that $\D_4^*$ is a covering space of $\T_4^*$,

1851: we know it will be a $\C$-bundle over the corresponding

1852: cover $\Y_4^*$ of $\X_4^*$.  Letting $\Zodd := 2\Z+1$ denote the

1853: odd integers, we have:

1854: 

1855: \begin{lemma}

1856:   The space $\D_4^*$ is a $\C$-bundle over $\Y_4^*\isom(\C\setm\Zodd)\times\R$.

1857: \end{lemma}

1858: \begin{proof}

1859:   The fundamental group of $\T_4^*$ is $\Z *\Z$.

1860:   The first generator~$\beta$ is a loop represented by $p_1$ circling

1861:   counterclockwise around~$p_2$;

1862:   in terms of $\phi(\X_4^*)=\big(\C\setm\{0,1\}\big)\times\R$,

1863:   it corresponds to a loop around the deleted line at~$1$.

1864:   The second generator~$\gamma$ is represented

1865:   by $p_1$ circling clockwise around~$p_3$

1866:   (or equivalently, counterclockwise around both $p_2$ and $p_4$);

1867:   in $\phi(\X_4^*)$, it is a loop around the deleted line at~$0$.

1868: 

1869:   We claim that every lift of $\gamma$ to $\D_4^*$ is a loop,

1870:   while the lifts of $\beta$ are not.  

1871:   This means that the covering $\D_4^*\to\T_4^*$ is normal,

1872:   corresponding to the normal subgroup generated by~$\gamma$.  

1873:   The covering map $\Y_4^*\to\X_4^*$ can be then explicitly given as

1874:   $(w,t)\mapsto(1+e^{\pi i w},t)$ with $\Y_4^*\isom(\C\setm\Zodd)\times\R$,

1875:   and we can represent $\D_4^*$ as the pullback $\C$-bundle over~$\Y_4^*$.

1876: 

1877:   To prove the claim, we will work over the deformation retract~$\X_4^*$.

1878:   Because $p_3=0$ and $p_4=\infty$ are antipodal, 

1879:   truncations of metrics projecting to~$\X_4^*$ are not unique.

1880:   To overcome this difficulty we adopt the convention that we always truncate 

1881:   that end along an arc developing to the positive reals.  

1882:   This corresponds to viewing the point $(p_1,p_2,0)$

1883:   as a limit of $(p_1,p_2,\epsilon)$ for $\epsilon$ positive real.

1884: 

1885:   We fix a basepoint $P:=(1,2,0)\in\X_4^*\subset\T_4^*$.

1886:   We can use our decomposition results to analyze any metric

1887:   in the fiber over~$P$.  Because of the order of the four points in

1888:   the circle, and our convention for drawing diagonals~$p_3p_4$,

1889:   each decomposition piece is isometric to a slit sphere.

1890:   Furthermore exactly two pieces are $3$-gons, with vertices

1891:   either $p_1p_2p_3$ and $p_1p_3p_4$, or $p_1p_2p_4$ and $p_2p_3p_4$.

1892:   Metrics in the fiber over~$P$ are completely characterized by

1893:   the choice of one of these two types,

1894:   together with the nonnegative number of slit spheres between the two

1895:   triangles.

1896: 

1897:   More explicitly, suppose we truncate a $4$-point metric~$D$ in the

1898:   fiber over~$P$, using again a diagonal $p_3p_4$

1899:   which projects to the positive reals. 

1900:   The $4$-gon $D'$ obtained has four interior angles $\alpha_i:=\alpha(p_i)$; 

1901:   each of these is an integer multiple of~$2\pi$.

1902:   The alternating sum

1903:   $$

1904:     \alpha_1-\alpha_2+\alpha_3-\alpha_4=2\pi n

1905:   $$

1906:   is unchanged if we add $2\pi$ to two consecutive angles,

1907:   and hence is independent of the truncation $D'$ we have chosen.

1908:   Note that the area of~$D'$ and the number $2(\alpha_2+\alpha_4)$ 

1909:   are each multiples of~$4\pi$.

1910:   Thus the Gauss--Bonnet Theorem shows that $n$ must be odd;

1911:   the odd integer~$n$ uniquely characterizes a metric $D=D_n$ 

1912:   in the fiber over~$P$.

1913:   This also follows from our combinatorial picture:

1914:   indeed, $D_n$ has a diagonal $p_1p_3$ when $n>0$ and 

1915:   a diagonal $p_2p_4$ when $n<0$; the number of slit spheres

1916:   along this diagonal is $\sfrac{(|n|-1)}2$.

1917:   %% Gauss-Bonnet:

1918:   %% 2\pi-(a_1+a_2+a_3+a_4-4*\pi) = \int_D' K dA = area(D') = 4\pi m

1919:   %% <==>   2\pi (1+2-2m) = a_1+a_2+a_3+a_4 

1920:   %% claim ok since  2(a_2+a_4) = 4\pi l  

1921: 

1922:   Now choose a truncation $D'$ of~$D_n$ with at

1923:   least one slit sphere added at each end.

1924:   The loops $\beta$ and~$\gamma$ in~$\X_4^*$

1925:   can be lifted to paths in~$\D_4$ with initial point~$D_n$.

1926:   These paths also have a unique lift starting at~$D'$ in

1927:   the space $\P_4$ of $4$-gons, provided no angle~$\alpha_i$ becomes zero.  

1928:   Indeed, this is easy to see if the paths are chosen to avoid antipodal 

1929:   pairs in the $4$-tuple: then the unique geodesics between them

1930:   lead to a unique continuation of the truncation.

1931: 

1932:   Let us now observe the effect of these lifts of $\beta$ and~$\gamma$

1933:   on the angles~$\alpha_i$.

1934:   Since only $p_1$ moves and the edge $p_3p_4$ stays 

1935:   fixed (by our convention), the angle~$\alpha_3$ remains constant.

1936:   For $\gamma$, as $p_1$ moves around $p_3$ (which is antipodal to~$p_4$)

1937:   the edge $\overline{p_1p_4}$ rotates once around the sphere.

1938:   This means that $\alpha_1$ and~$\alpha_4$ each decrease by~$2\pi$.

1939:   Since the angle~$\alpha_2$ stays nearly constant all along~$\gamma$,

1940:   it must return to its initial value.  We conclude that $n$ is fixed,

1941:   and so $\gamma$ lifts to a loop in~$\D_4^*$.

1942:   For~$\beta$, on the other hand, $\alpha_4$ is nearly constant,

1943:   $\alpha_2$ increases by $2\pi$ and~$\alpha_1$ decreases by~$2\pi$;

1944:   thus $n$ decreases by~$2$.  

1945:   Thus $\beta$ changes the metric $D_n$ to~$D_{n-2}$.  

1946:   This completes the proof of the claim.

1947: \end{proof}

1948: 

1949: \begin{proposition}

1950:   The space~$\D_4$ is diffeomorphic to an open 5-ball.

1951: \end{proposition}

1952: \begin{proof}

1953:   Remember that $\X_4\setm\X_4^*$ consists of two arcs $\gamma_{13}$

1954:   and~$\gamma_{24}$.  The decision of where to fill in copies of these

1955:   arcs into the covering space $\Y_4^*$ is purely combinatorial, based

1956:   on the decompositions of the metrics in $\D_4\setm\D_4^*$.

1957: 

1958:   We can connect our basepoint $P=(1,2,0)$

1959:   to the fibers over these arcs by simple paths:

1960:   the first path moves only~$p_1$ (along the reals to~$0$); 

1961:   the second moves only~$p_2$ (along the reals to~$\infty$).  

1962:   We use these paths to extend the odd labels~$n$ from the fiber over~$P$.   

1963:   Now we note that $p_i=p_j$ is allowed in $\D_4$ exactly when there

1964:   is no diagonal $p_ip_j$.

1965:   Thus the arc $\gamma_{13}$ gets filled in exactly at the levels

1966:   with negative label $n<0$.  

1967:   Similarly the arc $\gamma_{24}$ gets filled in exactly 

1968:   at the levels with $n>0$.

1969: 

1970:   This means that at each level, exactly one of the two arcs gets filled in.

1971:   Again, the decision is purely combinatorial: whenever we fill in a single

1972:   point, we fill in an entire arc into~$\Y_4^*$,

1973:   and $\D_4$ contains the entire $\C$-bundle over this arc.

1974:   In the picture $\Y_4^*\isom(\C\setm\Zodd)\times\R$,

1975:   we fill in the lower half ($t<0$) of the missing lines where $z=n<0$,

1976:   and the upper half ($t>0$) of the others where $z=n>0$; 

1977:   this leaves us with a space $\Y_4$ diffeomorphic to a three-ball.  

1978:   Finally, $\D_4$ is a $\C$-bundle over this ball, 

1979:   which must be trivial, meaning that $\D_4$ is a five-ball.

1980: \end{proof}

1981: 

1982: 

1983: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1984: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1985: \section{Injectivity and Properness}

1986: 

1987: To show the classifying map $\Phi$ is continuous, injective, and proper,

1988: we follow the approach we used for the case $k=3$ of triunduloids.

1989: The proof of properness is where substantial modifications are needed.

1990: In particular, since the classifying space is~$\Dk$, not~$\Tk$,

1991: we need to characterize the compact subsets of~$\Dk$ for general~$k$.

1992: 

1993: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1994: \subsection{Injectivity}

1995: 

1996: As for triunduloids we have for coplanar $k$-unduloids of genus zero:

1997: \begin{theorem}\label{th:inj}

1998:   The classifying map $\Phi$ is injective.

1999: \end{theorem}

2000: \begin{proof}

2001:   The proof is essentially the same as for triunduloids.

2002:   We sketch the main steps and refer to \cite{gks} for details.

2003:  

2004:   Suppose two $k$-unduloids $M,M'\in\Mk$ map to the same 

2005:   $k$-point metric $D:=\Phi(M)=\Phi(M')\in\Dk$.  We need to show $M=M'$. 

2006:   We pull back the $\k$-Hopf bundle $\S^3\to\S^2$ to 

2007:   the disk $D$ by the developing map.  Taking the universal cover of

2008:   the fibers, we obtain a line bundle $E\to D$.

2009:   It carries a Riemannian structure, locally isometric to~$\S^3$. 

2010:   The $\k$-Hopf flow acts isometrically on~$E$, moving points vertically

2011:   along the fibers.

2012: 

2013:   The cousin disks $\plustilde M$ and $\plustilde{M'}$ can be considered 

2014:   sections of~$E$.   Over each of the $k$ rays of the decomposition of~$D$

2015:   given by Theorem~\ref{thm:decomp}, the $\k$-Hopf flow foliates the 

2016:   bundle~$E$ with unduloid cousins.  

2017:   Over each ray, the cousin disks are asymptotic to one of the leaves.

2018:   But across each end of~$M$ and~$M'$ the \emph{periods}

2019:   (see \cite[Sect.~3.3]{gks}) vanish,

2020:   which by \cite[Lem.~3.7]{gks} implies for the cousins that 

2021:   the $\k$-Hopf flow can move $\plustilde M$ to become

2022:   asymptotic to~$\plustilde{M'}$ over all rays simultanuously.  

2023:   This allows us to deal with

2024:   $\plustilde M$ and $\plustilde {M'}$ as if they were compact;

2025:   as in \cite[Thm.~3.8]{gks} the 

2026:   maximum principle then shows 

2027:   $\plustilde M=\plustilde {M'}$.

2028:   Therefore $M=M'$, up to translation.

2029: \end{proof}

2030: 

2031: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2032: \subsection{Properness}

2033: 

2034: To show properness, we want to characterize compact subsets of~$\Dk$.

2035: We need the following notion.

2036: The \emph{vertex distance} $d(D)$ of $D\in\Dk$ is the minimum  

2037: distance between any pair of vertices in the completion $\hat D$.

2038: 

2039: \begin{proposition} \label{pr:cptK}

2040:   For any compact $\K\subset\Dk$, there exist $m,\epsilon>0$ 

2041:   such that for all $D\in\K$\\

2042:   \ia\ the vertex distance $d(D)$ is at least~$\epsilon$, and \\

2043:   \ii\ there is a small truncation $\Dbox$ with at 

2044:      most~$m$ decomposition pieces. 

2045: % perhaps place lemma in Sect. 2?

2046: \end{proposition}

2047: \begin{proof}

2048:   \ia\  

2049:   The distance between any two given vertices

2050:   depends continuously on $D\in\Dk$ and is always positive;

2051:   thus it assumes a positive minimum over the compact set~$K$.

2052:   Minimizing over the finitely many pairs of distinct vertices

2053:   still yields a positive minimum for~$d(D)$.

2054:   (It is not hard to see that, in fact, $d(D)$ is realized

2055:   by the spherical distance of some pair of distinct

2056:   points of~$\Pi(D)\in\Tk$.)

2057: 

2058:   \smallskip\noindent\ii\

2059:   Consider a metric $D\in\Dk$ which admits a small truncation

2060:   with $n$ decomposition pieces.  

2061:   If no piece is a lune or a concave polygon with angle~$\pi$ then 

2062:   $D$ has a neighborhood of metrics with the same number of pieces.

2063:   Otherwise, at most $k$~additional pieces are needed.

2064: %KGB+RBK 9.7.05: John, please check.

2065: 

2066:   Now let $V_j\subset\Dk$ be the set of metrics

2067:   which admit a small truncation with at most $j-k$ decomposition pieces.  

2068:   The above discussion shows $V_j$ has an 

2069:   open neighborhood $U_j\subset\Dk$, such that any metric in~$U_j$ 

2070:   admits a truncation with no more than $j$~decomposition pieces. 

2071:   The $\{U_j\}$ define an open exhaustion of~$\Dk$, so that the

2072:   compact set $K$ is contained in some~$U_m$.

2073: %

2074: % Suppose there is a sequence $D_j\in\Dk$ such that $\Dboxj$ has

2075: % area at least~$j$.  No subsequence can have a limit~$D$, for

2076: % then the area of~$\Dbox$ would have to be infinite.

2077: \end{proof}

2078: 

2079: \begin{lemma}\label{le:curv}

2080:   Suppose $\K$ is a subset of $\Dk$ such that the vertex

2081:   distance of each $D\in\K$ is at least~$\epsilon>0$.

2082:   Then there is $C=C(\epsilon)>0$ such that the curvature estimate

2083:   $\sup_M |A|<C(\epsilon)$ holds for each $M\in\Phi^{-1}(\K)$.

2084: \end{lemma}

2085: %---------- Curvature blow up of rhombic 4-unduloids ------------------------

2086: \begin{figure}

2087: \hspace*{-9mm}

2088: \begin{overpic}[width=.32\textwidth]{rhompi4_3} \end{overpic}

2089: %\hspace{2.5mm}

2090: \begin{overpic}[width=.32\textwidth]{rhompi4_4} \end{overpic}

2091: \hspace{3mm}

2092: \begin{overpic}[width=.32\textwidth]{rhompi4_5} \end{overpic}

2093: \caption[Curvature blow up]{

2094:     This family of $4$-unduloids with fixed necksizes $\pi/2$

2095:     exhibits a curvature blow-up on the three necks attached to the

2096:     two central bubbles.  Its singular limit is a union of two unduloids 

2097:     and two spheres.  Under~$\Psi$, the surfaces Hopf project

2098:     to the two poles of~$\S^2$ as well as two points on the equator 

2099:     which approach  one another; thus the vertex distance has limit~$0$ 

2100:     and Lemma~\ref{le:curv} does not apply.

2101: }\label{fig:curvature blow up}

2102: \end{figure}

2103: %---------------------------------------------------------------------------

2104: % 4.10.02

2105: \begin{proof}

2106:   We modify the argument given in \cite[Sect.~4]{gks}.

2107:   Suppose on the contrary that $\sup_{M_j}|A|\to\infty$ with $\Phi(M_j)\in \K$.

2108:   Then by \cite[Lem.~4.2]{gks} we find a sequence of rescaled surfaces~$M'_j$,

2109:   converging to some minimal surface~$N$ which is Alexandrov-symmetric

2110:   across the plane~$P$.

2111:   Note that, as in \cite[Prop.~4.4]{gks}, the $M_j$ have genus zero,

2112:   meaning that $M_j\cap P$ has no closed loops.  Since the convergence

2113:   is with multiplicity one, each component of $N\cap P$ is likewise

2114:   an unbounded curve.

2115: 

2116:   By \cite[Lem.~4.3]{gks} the minimal surface~$N$ contains a simple

2117:   closed curve~$\gamma$ with nonzero force, symmetric with respect to~$P$

2118:   and not contained in~$P$.  So the surfaces~$M'_j$ contain closed 

2119:   curves~$\gamma'_j$, whose lengths and forces converge to those of~$\gamma$.  

2120:   Now consider the closed curves~$\gamma_j$ on the unrescaled surfaces~$M_j$.

2121:   Note that the length of~$\gamma_j$ converges to~$0$.

2122:   For sufficiently large~$j$, however, the force of $\gamma'_j$ 

2123:   is nonzero, and so $\gamma'_j$ as well as~$\gamma_j$ 

2124:   must be homologically nontrivial.  

2125:   But since the~$M_j$ have genus zero this means that 

2126:   the curve~$\gamma_j$ must connect two different components of $M_j\cap P$.

2127:   Therefore, the conjugate curve $\tilde\gamma^+_j$

2128:   to its \uhalf/ $\gamma^+_j$ Hopf projects to a curve~$\eta_j$ 

2129:   in $\Phi(M_j)\in\Dk$ which connects two different vertices.  

2130: 

2131:   By definition, the length of any such curve is at least the vertex 

2132:   distance $d(D)\ge\epsilon$.

2133:   Since the Hopf projection $\Pik$ has Lipschitz constant~$2$,

2134:   it follows that the length of $\tilde\gamma^+_j$, and thus also

2135:   the length of $\gamma_j$, is bounded below.

2136:   This contradicts the fact that $\gamma_j$ has length converging to~$0$.

2137: \end{proof}

2138: 

2139: \begin{lemma}\label{le:ball}

2140:   Given a compact subset $\K\subset\Dk$ there is $R>0$ with the following

2141:   property.  For each $M\in \Phi^{-1}(\K)$ there exists a ball $B_R\in\R^3$

2142:   of radius~$R$ such that

2143:   the noncompact components of $M\setm B_R$ are exactly $k$ annular ends, 

2144:   each of which is contained in a solid cylinder of radius~$3$.

2145: \end{lemma}

2146: \begin{proof}

2147:   The enclosure theorem, proved in~\cite{kk} for embedded surfaces,

2148:   carries over without change to say that any \alex-emb/ \CMC/

2149:   surface $M$ is contained

2150:   in the union of $k$~solid cylindrical rays as well as a finite

2151:   number of balls of radius $11$ and solid cylindrical segments of radius~$3$; 

2152:   the numbers of these depend only on~$k$ and the genus.

2153:   Collecting the balls and finite cylindrical segments into one larger ball,

2154:   we see that for each $M\in\Mk$ there is an $R=R(M)$ with the property

2155:   desired.  Note that $\Mk$ contains sequences of $k$-unduloids 

2156:   where the length of some midsegment, and hence~$R$, diverges.

2157:   However we claim that, over~$\K$, there is a uniform choice of~$R$.

2158: 

2159:   Suppose, by contradiction, that for some

2160:   sequence $M_j\in \Phi^{-1}(\K)$ we have $R(M_j)\to\infty$.

2161:   Consequently, the enclosing sets for the~$M_j$ 

2162:   have solid cylindrical segments with

2163:   length~$l(j)$ tending to infinity.

2164:   These segments contain annuli~$A_j\subset M_j$.

2165:   By the uniform bound on the vertex distance from 

2166:   Proposition~\ref{pr:cptK}\ia,

2167:   the force carried by these annuli are uniformly bounded away from $0$.

2168:   Thus, as $l(j)\to\infty$, it follows as in~\cite[Thm.~5.1]{kks} 

2169:   that after %an appropriate 

2170:   rigid motion a subsequence of the~$A_j$ 

2171:   converge to some (complete) unduloid.

2172: 

2173:   Since this unduloid has a cousin which Hopf-projects to

2174:   a line of slit spheres, we find that the spherical metrics

2175:   for~$A_j$ must contain increasing numbers of slit spheres.

2176:   This contradicts Proposition~\ref{pr:cptK}\ii.

2177: \end{proof}

2178: 

2179: \begin{theorem}                      \label{th:proper}

2180:   The classifying map $\Phi\colon\Mk\to\Dk$ is continuous and proper.

2181: \end{theorem}

2182: \begin{proof}

2183:   Given a compact $\K\subset\Dk$, Lemma~\ref{le:curv} (combined with

2184:   Proposition~\ref{pr:cptK}\ia) gives a uniform

2185:   curvature estimate for $k$-unduloids in $\Phi^{-1}(\K)$.  The continuity 

2186:   of~$\Phi$ on this set then follows just as in~\cite[Thm.~4.5]{gks}.

2187: 

2188:   To show that $\Phi^{-1}(\K)$ is compact,

2189:   we now prove that any sequence $M_j\in \Phi^{-1}(\K)$

2190:   subconverges with respect to the topology on~$\Mk$.

2191:   The~$M_j$ are given only up to horizontal translation; to

2192:   get convergence, we choose representatives such that

2193:   the enclosing balls $B_R$ from Lemma~\ref{le:ball}

2194:   are centered at the origin.

2195:   Note also that by~\cite{kk}, there is a uniform area bound, 

2196:   for any coplanar genus-zero $k$-unduloid $M$,

2197:   of the form $\area(M_j\cap B_r)\le Cr^2$.

2198:   This area bound, together with the curvature estimate

2199:   of Lemma~\ref{le:curv}, now gives a subsequence of the~$M_j$

2200:   that converges (on each compact subset) to some \CMC/ surface~$M$.

2201: 

2202:   By smooth convergence, it is clear that $M$ is \alex-emb/.

2203:   Because a sequence of contractible loops has a contractible limit,

2204:   $M$ has genus zero and at most $k$ ends.

2205:   On the other hand, by Lemma~\ref{le:ball}, on each $M_j$

2206:   there are $k$ closed curves in $B_{R+1}(0) \setm B_{R}(0)$

2207:   which bound the $k$ ends.  These $k$ sequences of

2208:   curves subconverge to $k$ closed curves which are disjoint in

2209:   the limit surface $M$, and thus bound $k$ different ends.

2210:   Thus $M$ is a coplanar $k$-unduloid, as desired.

2211: \end{proof}

2212: 

2213: 

2214: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2215: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2216: \section{Surjectivity of the classifying map}

2217: 

2218: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2219: \subsection{Moduli spaces of coplanar $k$-unduloids}

2220: 

2221: \begin{comment} %-----------------------------

2222: % We will use below that any nondegenerate surface has an open neighborhood 

2223: % of nondegenerate surfaces in the moduli space

2224: % $\M_{g,k}$ of all $k$-unduloids of genus $g$.

2225: We are interested in surfaces with

2226: Alexandrov symmetry across the $xy$-plane~$P$.

2227: Let $\tau$ be the reflection across~$P$.

2228: Vertical translation induces a positive bounded Jacobi field 

2229: $v:=-\langle \nu,\k\rangle$ on $M$ which is odd with respect to~$\tau$, 

2230: that is, $v\circ\tau=-v$.

2231: The following proposition was suggested by Nick Korevaar; it also holds when

2232: the genus is nonzero, as well as for coplanar minimal $k$-noids.

2233: \begin{proposition} \label{pr:oddjacobi}

2234:   On a coplanar $k$-unduloid~$M\in\Mk$, any 

2235:   odd $L^2$ Jacobi field must vanish.

2236: % odd bounded Jacobi field is a multiple of~$v$.

2237: % -- is still true (with more involved proof), but we don't need it, 

2238: \end{proposition}

2239: \begin{proof}

2240:   Let $u\not\ident 0$ be an odd Jacobi field in $L^2(M)$.

2241:   By Alexandrov symmetry, the solution $v$ is positive on $M^+$.

2242:   We claim that $\sfrac uv$ is bounded on $M^+$ and decays to zero

2243:   on each end of~$M$.

2244: 

2245:   Both~$u$ and~$v$ vanish on the boundary $\d M^+$,

2246:   so to estimate $\sfrac uv$ near $\d M^+$ it suffices to

2247:   estimate $\sfrac{|\grad u|}{|\grad v|}$.

2248:   Since $u\in L^2$, it and its gradient decay exponentially~\cite{kks} along

2249:   the ends of~$M$.

2250:   By the Hopf boundary maximum principle, $|\grad v|$ is

2251:   bounded away from~$0$ along compact subsets of the boundary.

2252:   Each end of~$M$ is asymptotic to an unduloid, 

2253:   and on each unduloid the vertical component of the normal also

2254:   has gradient bounded away from~$0$.  It follows that 

2255:   $|\grad v|$ is bounded away from~$0$ on all of~$\d M^+$.

2256:   Hence $\sfrac{|\grad u|}{|\grad v|}$ is bounded on~$\d M^+$

2257:   and decays to zero on its ends.

2258:   

2259:   Because $u,v$ are bounded Jacobi fields, their second derivatives are bounded.

2260:   Hence the estimate on $\sfrac{|\grad u|}{|\grad v|}$ extends

2261:   to a neighborhood of~$\d M^+$ as well.

2262:   

2263:   Finally, since $u\in L^2$ while $v>0$ is asymptotically periodic,

2264:   away from~$\d M^+$ the quotient $\sfrac uv$ is bounded and decays

2265:   to zero on the ends of~$M$.

2266:   This proves our claim.

2267: 

2268:   By changing the sign of $u$ if necessary, $u$ is positive at some point.

2269:   Let $\mu$ be the (positive) supremum of $\sfrac{u}v$ over $M^+$.

2270:   Then the function $u-\mu v\le 0$ attains its maximum~$0$ 

2271:   at some point $p\in \overline{M^+}$ (in particular all along the boundary).

2272:   Then the Hopf boundary maximum principle shows that $u\ident \mu v$. 

2273:   But $v\not\in L^2$, which gives $\mu=0$ and $u\ident 0$.

2274: \end{proof}

2275: 

2276: We can decompose $L^2(M)$ into odd and even parts with respect to $\tau$.

2277: %The Jacobi field $v$ is not in $L^2$, and so the proposition 

2278: %implies that any $L^2$ Jacobi field lies in $L^2_{\text{even}}(M)$.  

2279: Thus, a coplanar surface~$M$ is nondegenerate if and only if

2280: all even $L^2$ Jacobi fields $u=u\circ\tau$ vanish.

2281: \end{comment}

2282: %---------------------------------------------------------------

2283: 

2284: Given a minimal or \CMC/ surface~$M$, a solution~$u$ to the

2285: linearized mean-curvature equation

2286: $L(u):=\Delta_M u+|A|^2 u=0$ is called a \emph{Jacobi field}.

2287: The surface~$M$ is called 

2288: \emph{nondegenerate} if all square-integrable Jacobi fields

2289: $u\in L^2(M)$ vanish.

2290: 

2291: Let $\M_{g,k}$ denote the moduli space of all $k$-unduloids of genus~$g$.

2292: Kusner, Mazzeo and Pollack~\cite{kmp} showed that~$\M_{g,k}$

2293: is a real-analytic variety (for $k\ge3$), and that near any

2294: nondegenerate surface it is a $(3k-6)$--dimensional manifold.

2295: We now prove a similar result for coplanar $k$-unduloids:

2296: 

2297: \begin{theorem} \label{th:kmp}

2298:   The moduli space $\Mk$ of coplanar $k$-unduloids of genus zero

2299:   is locally a real-analytic variety of finite dimension.

2300:   In a neighborhood of a nondegenerate coplanar $k$-unduloid,

2301:   it is a manifold of dimension~$2k-3$.

2302: \end{theorem}

2303: In their proof, \cite{kmp} consider the \emph{premoduli space} 

2304: $\wtilde\M_{g,k}$, which is the space of $k$-unduloids of genus~$g$

2305: \emph{before} dividing by euclidean motions.

2306: At a nondegenerate surface~$M$, they show the premoduli space $\wtilde\M_{g,k}$

2307: is locally a $3k$-dimensional Lagrangian submanifold of a symplectic vector 

2308: space~$W$.  Here, each end of~$M$ has a

2309: $6$-dimensional space of geometric Jacobi fields, 

2310: induced by translations, rotations, and change of asymptotic necksize.

2311: The direct sum of these, over the $k$~ends of~$M$, forms the 

2312: $6k$-dimensional space~$W$.

2313: The tangent space to $\wtilde\M_{g,k}$ at~$M$ is the Lagrangian subspace 

2314: consisting of all globally defined Jacobi fields on~$M$.  

2315: \begin{proof}

2316:   In the premoduli space $\wtilde\M_{0,k}$ of all genus-zero $k$-unduloids,

2317:   let $\wtilde\M_k$ denote the subset consisting 

2318:   of coplanar $k$-unduloids with Alexandrov symmetry

2319:   across the $xy$-plane~$P$.

2320:   The reflection $\tau$ across~$P$ acts on~$\wtilde\M_{0,k}$,

2321:   fixing~$\wtilde\M_k$.

2322: 

2323:   Near any surface~$M$, the variety~$\wtilde\M_{0,k}$

2324:   is defined by real-analytic 

2325:   equations on a finite-dimensional vector space $W'\supset W$.

2326:   Our premoduli space~$\wtilde\M_k$ is obtained locally by adding one more set

2327:   of linear equations, $u\after\tau=u$, and is thus still real-analytic.

2328:   Dividing by euclidean motions of the $xy$-plane~$P$, we find the same

2329:   is true for~$\Mk$.

2330:  

2331:   To compute the dimension of~$\Mk$ we show that 

2332:   $\wtilde\M_k$ has dimension~$2k$ in the neighborhood of a

2333:   nondegenerate surface~$M$.  

2334:   We use the following transversality argument. 

2335:   The reflection~$\tau$ induces an automorphism $\tau^*$

2336:   on the $6k$-dimensional vector space~$W$ which encodes the asymptotic

2337:   behavior of Jacobi fields on~$M$. 

2338:   This automorphism preserves the natural symplectic form on~$W$.  

2339:   The global Jacobi fields on~$M$ form a Lagrangian subspace $L\subset W$,

2340:   which corresponds to the tangent space to~$\wtilde\M_{0,k}$ at~$M$.

2341:   Since $\tau$ takes one global Jacobi field on~$M$ to another,

2342:   $\tau^*$ carries $L$ into itself.  

2343:   The fixed point set of~$\tau^*$ is a $4k$-dimensional vector space~$V$, 

2344:   consisting of those asymptotic changes that keep the ends in the plane~$P$.

2345:   The following lemma then says that $L\cap V$, 

2346:   corresponding to the tangent space of $\wtilde\M_k$ at~$M$,

2347:   is $2k$-dimensional.

2348: \end{proof}

2349: 

2350: \begin{lemma}

2351:   Suppose $L$ is a Lagrangian (maximal isotropic) subspace of

2352:   a symplectic vector space $W$, and suppose $G$ is a compact group

2353:   of symplectic automorphisms of~$W$ preserving~$L$.

2354:   Then the fixed-point set $W^G$ of~$G$ is a symplectic subspace 

2355:   $W^G\subset W$, and $L\cap W^G$ is Lagrangian in $W^G$.

2356: \end{lemma}

2357: \begin{proof}

2358:   By averaging over the group~$G$, we can choose a compatible almost complex

2359:   structure~$J$ on~$W$ that is also invariant under~$G$.

2360: % (or equivalently, an inner product which tames the symplectic structure).

2361:   That is, $g\after J=J\after g$ for all $g\in G$.

2362:   Let $W^g := \ker(g-I)$, which is preserved by $J$.

2363:   Then the fixed set $W^G$ is the intersection $W^G=\bigcap_{g\in G} W^g$.

2364:   In particular, $J(W^G)=W^G$, showing this is symplectic.

2365: 

2366:   Because $L$ is Lagrangian, we have $W=L\oplus JL$, so

2367:   any vector in $W$ can be represented as $v+Jw$, where $v,w\in L$.

2368:   If, moreover, $v+Jw$ is in $W^g$ 

2369:   then $g(v+Jw)=v+Jw$ and so $-(g-I)v = (g-I)Jw = J(g-I)w$.

2370:   But since $g$ (and therefore $g-I$) preserves~$L$, 

2371:   this equates a vector in~$L$ with one in~$JL$.  

2372:   Thus $(g-I)v=0=(g-I)w$, so that in fact $v,w\in L\cap W^g$.

2373:   Since the preceding argument holds for all $g\in G$ we obtain 

2374:   a decomposition $W^G=(L\cap W^G)\oplus J(L\cap W^G)$, 

2375:   meaning that $L\cap W^G$ is Lagrangian in~$W^G$.  

2376: \end{proof}

2377: 

2378: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2379: \subsection{Existence of nondegenerate coplanar $k$-unduloids}

2380: 

2381: A \emph{minimal $k$-noid} is a finite-total-curvature minimal surface

2382: of finite topology with genus zero which has only catenoidal ends.

2383: Jorge and Meeks determined the $k$-noids which are coplanar and dihedrally 

2384: symmetric.  They have only two umbilic points, so that the result of

2385: Montiel and Ros~\cite{mr} applies (in the form mentioned in~\cite{gks}) to give:

2386: \begin{proposition} \label{pr:knoids}

2387:   The Jorge--Meeks $k$-noids are nondegenerate.

2388: \qed

2389: \end{proposition}

2390: %As we noted above, Proposition~\ref{pr:oddjacobi} also applies to coplanar

2391: %minimal $k$-noids.

2392: %Moreover, translations perpendicular to the axis

2393: %of a catenoid end induce a Jacobi field which fails to be in $L^2$.  

2394: %Thus, as for our coplanar \CMC/ surfaces, a coplanar minimal surface is

2395: %nondegenerate if and only if there are no even $L^2$ Jacobi fields.

2396: 

2397: As in the case of triunduloids, we want to use these minimal surfaces

2398: in the gluing technique of Mazzeo and Pacard~\cite{mp} to produce nondegenerate

2399: coplanar \CMC/ surfaces.  Their theorem generalizes to an

2400: equivariant setting as follows:

2401: 

2402: \begin{theorem}\label{th:mp}

2403:   Let $M_0\subset\R^3$ be a nondegenerate minimal $k$-noid that

2404:   can be oriented so that each end has normal pointing inwards towards its

2405:   asymptotic axis.  Assume $M_0$ has symmetry group~$G$ in~$\R^3$,

2406:   consisting of isometries preserving the normal.

2407:   Then for some $\epsilon_0$,

2408:   there is a family $M_{\epsilon}$, $0<\epsilon<\epsilon_0$,

2409:   of nondegenerate \CMC/ surfaces with $k$~embedded ends,

2410:   also with symmetry group~$G$, such that on any compact set in~$\R^3$,

2411:   the dilated surfaces

2412:   $\tfrac 1{\epsilon} M_{\epsilon}$ converge smoothly and uniformly to~$M_0$.

2413: \end{theorem}

2414: %Here, \emph{$G$-nondegenerate} means that there are no

2415: %nonzero $G$-invariant $L^2$ Jacobi fields on~$M$.

2416: 

2417: \begin{proof}[Sketch of proof]

2418: The proof of this theorem follows exactly the argument

2419: of Mazzeo and Pacard~\cite{mp}, so we merely outline

2420: how to adapt their argument to our $G$-invariant setting.

2421: They first glue exactly unduloidal ends to~$M_0$, 

2422: truncated along $k$~circles, and then consider normal perturbations.

2423: They slightly perturb the bounding circles, and

2424: derive solution operators for the respective Dirchlet problems

2425: for the mean curvature equation, both over~$M_0$ and over the ends.  

2426: Unique solutions to these Dirichlet problems are established

2427: as fixed points of a contraction mapping \cite[Cor.~5,~8]{mp};

2428: thus $G$-invariant Dirichlet data lead to $G$-invariant solutions. 

2429: % as claimed in Cor. 5 (p.210) and Cor. 8 (p.231)

2430: The surface $M_\epsilon$ is then obtained by a Cauchy matching

2431: of a pair of Dirichlet solutions: 

2432: This matching is accomplished \cite[Prop.~32]{mp}

2433: % (on p.233)

2434: by applying the Leray-Schauder fixed-point theorem

2435: to a nonlinear operator~$F_\epsilon$ on the Dirichlet data.

2436: The operator~$F_\epsilon$ is designed to have a fixed point exactly

2437: when the first-order Cauchy data match.

2438: 

2439: Given a $G$-invariant initial surface~$M_0$ %for nontrivial~$G$,

2440: we note that $G$ acts by composition on all relevant function spaces.

2441: Also, the operators used in the various fixed-point arguments, being

2442: geometrically defined, all commute with~$G$.

2443: Therefore we can simply replace the function spaces by their

2444: closed linear subspaces consisting of $G$-invariant functions.

2445: Clearly, all estimates in~\cite{mp} continue to hold, 

2446: and so the contraction-mapping and Leray--Schauder fixed-point theorems

2447: still apply, giving us a $G$-invariant \CMC/ surface~$M_\epsilon$.

2448: 

2449: The arguments of \cite[Sect.~12]{mp} show that

2450: for sufficiently small~$\epsilon$ any function in~$L^2(M)$

2451: (whether $G$-invariant or not)

2452: satisfying the linearized mean curvature equation vanishes.

2453: Thus~$M_\epsilon$ is nondegenerate.

2454: \end{proof}

2455: 

2456: Now let~$M_0$ be the Jorge--Meeks $k$-noid of Proposition~\ref{pr:knoids},

2457: and $G$ be its symmetry group.

2458: Applying Theorem~\ref{th:mp} gives:

2459: \begin{corollary}\label{co:nondegkund}

2460:   For each $k\ge 3$ there exists a nondegenerate

2461:   %dihedrally symmetric,

2462:   coplanar genus-zero $k$-unduloid.

2463: \qed

2464: \end{corollary}

2465: 

2466: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2467: \subsection{Proof of Main Theorem and Corollary}

2468: 

2469: In \cite[Sect.~5.3]{gks} we show:

2470: \begin{theorem} \label{th:homeo}

2471:   If $f\takes X\to Y$ is a continuous, proper, injective map from

2472:   a $d$-dimensional real-analytic variety $X$ to a connected $d$-manifold $Y$,

2473:   then $f$ is surjective and thus a homeomorphism. \qed

2474: \end{theorem}

2475: 

2476: We now have all the ingredients needed to prove:

2477: \begin{theorem}

2478:   The classifying map $\Phi\colon \Mk\to\Dk$ is a homeomorphism

2479:   between connected $(2k-3)$--manifolds.

2480: \end{theorem}

2481: \begin{proof}

2482:   To apply Theorem~\ref{th:homeo} to $\Phi$ we need to verify its assumptions.

2483:   The space $\Dk$ is a connected $(2k-3)$--manifold by Theorems~\ref{th:mfld}

2484:   and~\ref{th:connected}.  The map $\Phi$ is proper and injective by

2485:   Theorems~\ref{th:proper} and~\ref{th:inj}.

2486: 

2487:   From Theorem~\ref{th:kmp}, the moduli space~$\Mk$ of coplanar $k$-unduloids

2488:   is locally a real-analytic variety; injectivity of $\Phi$ shows it has

2489:   dimension at most $2k-3$.

2490:   But by Corollary~\ref{co:nondegkund} there is a nondegenerate coplanar

2491:   $k$-unduloid, so then by Theorem~\ref{th:kmp}, $\Mk$ is

2492:   a $(2k-3)$--manifold in some neighborhood of this $k$-unduloid.

2493: 

2494:   Thus $\Mk$ has dimension exactly $2k-3$, so

2495:   Theorem~\ref{th:homeo} applies to conclude that $\Phi$ is a homeomorphism.

2496: \end{proof}

2497: 

2498: Together with Theorem~\ref{th:necksizes}, this proves our Main Theorem.

2499: 

2500: The corollary on necksizes stated in the introduction collects the

2501: statements of Corollary~\ref{co:oddk} and Corollary~\ref{co:evenk}.

2502: The existence statement follows 

2503: from the example of Section~\ref{se:polmet}; in particular

2504: the example shows that \eqref{necksizesumodd} and \eqref{necksizesumeven}

2505: are sharp.

2506: 

2507: 

2508: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2509: \begin{thebibliography}{GK${}^2$RS}

2510: \frenchspacing

2511: \renewcommand{\baselinestretch}{1}

2512: 

2513: \bibitem[Al1]{al1} Alexandrov A. D.,

2514:   \emph{Uniqueness theorems for surfaces in the large V},

2515:   Vestnik Leningrad Univ.~19:13 (1958) 5--8;

2516:   Amer.~Math.~Soc. Transl. (Ser.~2) 21 (1962) 412--416.

2517: 

2518: \bibitem[Al2]{al2} Alexandrov A. D.,

2519:   \emph{A characteristic property of spheres},

2520:   Ann. Mat. Pura Appl.~58 (1962) 303--315.

2521: 

2522: %\bibitem[AD]{AD} Austin D., Dickinson W.,

2523: %  Spherical Easel software, 2004,

2524: %  \url{merganser.math.gvsu.edu/easel/}

2525: 

2526: \bibitem[BS]{bs} Buckman R., Schmitt N.,

2527:   \emph{Spherical polygons and unitarization},

2528:   preprint, 2002; \url{www.gang.umass.edu/reu/2002/polygon.html}

2529: 

2530: \bibitem[CR]{cr} Cos{\'\dotlessi}n C., Ros A.,

2531:   \emph{A Plateau problem at infinity for properly immersed minimal

2532:   surfaces with finite total curvature},

2533:   Indiana Univ. Math. J. 50 (2001) 847--879.

2534: 

2535: \bibitem[Del]{del} Delaunay C.,

2536:   \emph{Sur la surface de r\'evolution, dont la courbure moyenne est constante},

2537:   Journal de math\'ematiques~6 (1841) 309--320.

2538: 

2539: \bibitem[Gro]{groisman} Groisman, P.,

2540:   \emph{A new family of symmetric constant-mean-curvature surfaces},

2541:   Ph.D. thesis, University of Illinois, 2004.

2542: 

2543: \bibitem[GB1]{kgb} Gro{\ss}e-Brauckmann K.,

2544:   \emph{New surfaces of constant mean curvature},

2545:   Math.~Zeit.~214 (1993) 527--565.

2546: 

2547: \bibitem[GB2]{kgb2} Gro{\ss}e-Brauckmann K.,

2548:   \emph{Cousins of constant mean curvature surfaces},

2549:  in: Global Theory of Minimal Surfaces,

2550:  Clay Math. Proc.~{2}, AMS (2005) 747--767.

2551: 

2552: \bibitem[GK${}^2$RS]{GKKRS}

2553:   Gro{\ss}e-Brauckmann K., Korevaar N., Kusner R., Ratzkin J., Sullivan J. M.,

2554:   \emph{Coplanar $k$-unduloids are nondegenerate},

2555:   preprint, 2007.

2556: 

2557: \bibitem[GK]{gk} Gro{\ss}e-Brauckmann K., Kusner R.,

2558:   \emph{Embedded constant mean curvature surfaces with special symmetry},

2559:   Man. Math.~99 (1999) 135--150.

2560: 

2561: \bibitem[GKS1]{gks} Gro{\ss}e-Brauckmann K., Kusner R., Sullivan J. M.,

2562:   \emph{Triunduloids: Embedded constant mean curvature surfaces

2563:   with three ends and genus zero},

2564:   J. Reine Angew. Math.~564 (2003) 35--61.

2565:   \url{arXiv:math.DG/0102183}

2566: 

2567: \bibitem[GKS2]{gkspnas} Gro{\ss}e-Brauckmann K., Kusner R., Sullivan J. M.,

2568:   \emph{Constant mean curvature surfaces with three ends},

2569:   Proc. Natl. Acad. Sci. USA~97 (2000) 14067--14068.

2570:   \url{arXiv:math.DG/9903101}

2571: 

2572: \bibitem[GKS3]{cylends} Gro{\ss}e-Brauckmann K., Kusner R., Sullivan J. M.,

2573:   \emph{Constant mean curvature surfaces with cylindrical ends}

2574:   in: Mathematical Visualization, %(H.-C. Hege, K. Polthier, eds.),

2575:   Springer (1998) 106--117.

2576: 

2577: \bibitem[Kap]{kap} Kapouleas N.,

2578:   \emph{Complete constant mean curvature surfaces in Euclidean three-space},

2579:   Ann. of Math.~131 (1990) 239--330.

2580: 

2581: \bibitem[Kar]{kar} Karcher H.,

2582:   \emph{The triply periodic minimal surfaces of

2583:   A.~Schoen and their constant mean curvature companions},

2584:   Man. Math.~64 (1989) 291--357.

2585: 

2586: \bibitem[KSS]{kss} Kilian M., Schmitt N., Sterling I.,

2587:   \emph{Dressing \CMC/ $n$-noids},

2588:   Math. Z.~246 (2004) 501--519.

2589: 

2590: \bibitem[KK]{kk} Korevaar N., Kusner R.,

2591:   \emph{The global structure of constant mean curvature surfaces},

2592:   Invent. Math.~114 (1993) 311--332.

2593: 

2594: \bibitem[KKR]{kkr} Korevaar N., Kusner R., Ratzkin J.,

2595:   \emph{On the nondegeneracy of constant mean curvature surfaces},

2596:   Geom. Funct. Anal.~16 (2006) 891--923; \url{arXiv:math.DG/0407153}

2597: 

2598: \bibitem[KKS]{kks} Korevaar N., Kusner R., Solomon B.,

2599:   \emph{The structure of complete embedded surfaces with constant mean curvature},

2600:   J. Diff.\ Geom.~30 (1989) 465--503.

2601: 

2602: \bibitem[Kus]{kmsri} Kusner R.,

2603:   \emph{Bubbles, conservation laws and balanced diagrams},

2604:   Geometric Analysis and computer graphics (Berkeley, CA, 1988) 103--108,

2605:   Math. Sci. Res. Inst. Pub.~17, Springer, New York, 1991. 

2606: 

2607: %???

2608: %\bibitem[K2]{kclay} Kusner R.,

2609: %  Clay/MSRI conference on minimal surfaces (Berkeley, CA, 2001).

2610: 

2611: \bibitem[KMP]{kmp} Kusner R., Mazzeo R., Pollack D.,

2612:   \emph{The moduli space of complete embedded constant mean curvature surfaces},

2613:   Geom. Funct. Anal.~6 (1996) 120--137.

2614: 

2615: \bibitem[Law]{law} Lawson H. B.,

2616:   \emph{Complete minimal surfaces in $\S^3$},

2617:   Annals of Math.~92 (1970) 335--374.

2618: 

2619: \bibitem[MP]{mp} Mazzeo R., Pacard F.,

2620:   \emph{Constant mean curvature surfaces with Delaunay ends},

2621:   Comm. Anal. Geom. 9 (2001) 169--237.

2622: 

2623: \bibitem[Mee]{mee} Meeks W. H.,

2624:   \emph{The topology and geometry of embedded surfaces of constant mean curvature},

2625:   J. Diff.\ Geom.~27 (1988) 539--552.

2626: 

2627: \bibitem[MR]{mr} Montiel S., Ros A.,

2628:   \emph{Schr\"odinger operators associated to a holomorphic map},

2629:   in: Global differential geometry and global analysis,

2630:   Lect. Notes Math.~1481, Springer (1991) 148--174.

2631: 

2632: %\bibitem[O]{obe} Oberknapp B.,

2633: %  Eine Konjugiertenkonstruktion zur Berechnung diskreter $H$-Fl\"achen,

2634: %  Diplomarbeit Bonn 1998.

2635: 

2636: \bibitem[Sch]{sch}

2637:   Schoen A.~H.,

2638:   \emph{Infinite periodic minimal surfaces without selfintersections},

2639:   NASA Technical Note TN D-5541 (1970).

2640: 

2641: \end{thebibliography}

2642: 

2643: \end{document}

2644: