1: \documentclass{article}

2: \usepackage{graphicx}

3: %\usepackage{url}

4: % Blackboard R for real numbers

5: \usepackage{amsfonts}

6: \usepackage{amssymb}

7: \usepackage{latexsym}

8: \newcommand\R{\mathbb{R}}

9: 

10: 

11: %% Definitions/abbreviations

12: \newcommand{\squeezelist}{\setlength{\itemsep}{0pt}}

13: 

14: \def\e{{\epsilon}}

15: \def\b{{\beta}}

16: \def\a{{\alpha}}

17: \def\G{{\Gamma}}

18: \def\g{{\gamma}}

19: \def\t{{\theta}}

20: \def\l{{\lambda}}

21: \def\bP{{\partial P}}

22: \def\bQ{{\partial Q}}

23: \def\T{{\cal T}}

24: %%

25: %

26: \input{mac}

27: %

28: %\withcomplaints % Complaints turned ON!!!

29: %

30: \pagenumbering{roman}

31: %\title{\bf Enumerating Foldings and Unfoldings\\ 

32: %between Polygons and Polytopes}

33: \title{\bf Examples, Counterexamples, and \\

34: Enumeration Results for\\

35: Foldings and Unfoldings\\

36: between Polygons and Polytopes}

37: \author{%

38: Erik Demaine \and Martin Demaine \and Anna Lubiw\thanks{

39: Dept.\ Comput\ Sci., Univ. of Waterloo,

40: Waterloo, Ontario N2L 3G1, Canada.

41: \texttt{\{eddemaine,mdemaine,alubiw\}@\penalty \exhyphenpenalty uwaterloo.ca}.

42: }

43: \\ \and

44: Joseph~O'Rourke\thanks{

45: Dept.\ Comput.\ Sci., Smith Col\-lege, North\-ampton, 

46: MA 01063, USA.

47: \texttt{orourke@\penalty \exhyphenpenalty cs.smith.edu}.

48: Supported by NSF grant CCR-9731804.}

49: }

50: \begin{document}

51: \maketitle

52: \begin{abstract}

53: We investigate how to make the surface of a convex polyhedron 

54: (a {\em polytope\/}) by folding up a

55: polygon and gluing its perimeter shut, and the reverse process

56: of cutting open a polytope and unfolding it to a polygon.

57: We explore basic enumeration questions in both directions:

58: Given a polygon, how many foldings are there?

59: Given a polytope, how many unfoldings are there to simple polygons?

60: Throughout we give special attention to convex polygons,

61: and to regular polygons.

62: We show that every convex polygon folds to an infinite number of

63: distinct polytopes, but that

64: their number of combinatorially distinct gluings is polynomial.  

65: There are, however, simple polygons with an exponential number

66: of distinct gluings.

67: 

68: In the reverse direction, we show that there are polytopes with an

69: exponential number of distinct cuttings that lead to simple unfoldings.

70: We establish necessary conditions for a polytope to have convex unfoldings,

71: implying, for example,

72: that among the Platonic solids, only the tetrahedron has a convex

73: unfolding.

74: We provide an inventory of the polytopes that may unfold to regular polygons,

75: showing that, for $n>6$, there is essentially only one

76: class of such polytopes.

77: \end{abstract}

78: 

79: %%%%%%%%%%%%%%%%%

80: \newpage

81: \tableofcontents

82: \newpage

83: \pagenumbering{arabic}

84: \setcounter{page}{1}

85: %%%%%%%%%%%%%%%%%

86: 

87: \section{Introduction}

88: \seclab{Introduction}

89: We explore the process of folding a simple polygon 

90: by gluing its perimeter shut to form

91: a convex polyhedron, and its reverse, cutting a

92: convex polyhedron open and flattening its surface to a

93: simple polygon.

94: We restrict attention to convex polyhedra (henceforth,

95: {\em polytopes\/}), 

96: and to simple (i.e., nonself-intersecting,

97: nonoverlapping) polygons (henceforth just {\em polygons}).

98: The restriction to nonoverlapping polygons is natural,

99: as this is important to the manufacturing

100: applications~\cite{o-fucg-00}.

101: The restriction to convex polyhedra is made primarily to

102: reduce the scope of the problem.

103: See ~\cite{bddloorw-uscop-98} and~\cite{bdek-up-99}

104: for a start on unfolding nonconvex polyhedra.

105: 

106: Much recent work on unfolding revolves around an 

107: open problem that seems to have been first mentioned

108: in print in~\cite{s-cpcn-75} but is probably much older:

109: Can every polytope be cut along edges and unfolded

110: flat to a (simple) polygon?

111: Cutting along edges leads to {\em edge unfoldings\/};

112: we will not follow this restriction here.

113: Thus our work is only indirectly related to this edge-unfolding

114: question.

115: 

116: In some sense this report is a continuation

117: of the investigation started in~\cite{lo-wcpfp-96},

118: which detailed an $O(n^2)$ algorithm

119: for deciding when a polygon may be folded to a polytope,

120: with the restriction that each edge of the polyon perimeter

121: glues to another complete edge:

122: {\em edge-to-edge gluing}.

123: But here we do not following this restriction,

124: permitting arbitrary perimeter gluings.

125: Moreover, we do not consider algorithmic questions.

126: Rather we concentrate on enumerating the number of foldings

127: and unfoldings between polygons and polytopes.

128: We pay special attention to convex polygons;

129: following Shephard~\cite{s-cpcn-75}, we call an

130: unfolding of a polytope that produces a convex polygon

131: a {\em convex unfolding}.

132: Within the class of polytopes, we sometimes

133: use the five regular polytopes as examples;

134: within the class of convex polygons, we additionally

135: focus on regular polygons.

136: 

137: The basic questions we ask are:

138: \begin{enumerate}

139: \squeezelist

140: \item How many combinatorially different foldings of a polygon lead to a polytope?

141: \item How many geometrically different polytopes may be folded from one polygon?

142: \item How many combinatorially different cuttings of a polytope lead to polygon unfoldings?

143: \item How many geometrically different polygons may be unfolded from one polytope?

144: \end{enumerate}

145: Our answers to these four

146: questions are crudely summarized in Table~\tabref{results}, 

147: whose four rows correspond to

148: the four questions above, and whose columns are for general, convex, and

149: regular polygons.

150: We will not explain the entries in the table here, but only remark

151: that the increased constraints provided by convex and regular

152: polygons reduces the number of possibilities.

153: \begin{table}[htbp]

154: \begin{center}\begin{tabular}{| l | l | c | c | c |}

155:         \hline

156: 

157: \mbox{}

158: 	& \mbox{}

159: 	& General

160: 	& Convex 

161: 	& Regular 

162: 	\\

163: \mbox{}

164: 	& \mbox{}

165: 	& \mbox{}

166: 	& Polygons

167: 	& Polygons

168:         \\ \hline \hline

169: 	

170: Foldings

171:         & gluing trees

172: 	& $2^{\Omega(n)}$, $O(n^{2\l-2})$

173: 	& $O(n^3)$

174: 	& $O(1)$

175:         \\ \cline{2-5}

176: \mbox{}

177:         & polytopes

178: 	& $\infty$

179: 	& $\infty$

180: 	& $2$ classes

181: 	\\ \hline

182: 	

183: Unfoldings

184:         & cut trees

185: 	& $2^{\Omega(n)}$, $2^{O(n^2)}$

186: 	& ?

187: 	& $O(1)$

188:         \\ \cline{2-5}

189: \mbox{}

190:         & polygons

191: 	& $\infty$

192: 	& $0$, $\infty$

193: 	& $O(1)$

194: 	\\ \hline

195: \end{tabular}

196: \tablab{results}

197: \caption{Summary of Results.  $n$ is the number of polygon vertices

198: or polytope vertices; $\l$ is the number of leaves of the gluing tree;

199: the symbol $\infty$ represents nondenumerably infinite, i.e., a

200: continuum.}

201: \end{center}

202: \end{table}

203: 

204: 

205: A key tool in our work is

206: a powerful theorem of

207: Aleksandrov, which we describe and

208: immediately apply in Section~\secref{Aleksandrov}.

209: We then define the two main combinatorial objects

210: we study, cut trees and gluing trees, and make

211: clear exactly how we count them.

212: We then explore constraints on convex unfoldings

213: in Section~\secref{Convex.Unfoldings}

214: before proceeding to the general enumeration bounds

215: in Table~\tabref{results}

216: in Sections~\secref{Counting.Gluings}-\secref{Noncongruent.Polygons}.

217: A final section (\secref{Regular}) concentrates on regular polygons

218: 

219: \section{Aleksandrov's Theorem}

220: \seclab{Aleksandrov}

221: 

222: Aleksandrov proved a far-reaching generalization of

223: Cauchy's rigidity theorem

224: in~\cite{a-kp-58} that gives simple conditions for any folding

225: to a polytope.

226: Let $P$ be a polygon and $\bP$ its boundary.

227: A {\em gluing\/} 

228: maps $\bP$ to $\bP$ in a length-preserving

229: manner, as follows.

230: $\bP$ is partitioned by a finite number of distinct points

231: into a collection of open intervals whose closure covers $\bP$.

232: Each interval is mapped one-to-one (i.e., {\em glued\/})

233: to another interval of equal length.

234: Corresponding endpoints of glued intervals are glued

235: together (i.e., identified).

236: Finally, gluing is considered

237: transitive: if points $a$ and $b$ glue to point $c$, then

238: $a$ glues to $b$.%

239: \footnote{

240:         What we call {\em gluing\/} is sometimes called

241:         {\em pasting\/}~\cite[p.~13]{az-igs-67}.

242: 	In the theory of complexes, it is sometimes called

243: 	{\em topological identification}~\cite[p.~116]{h-cit-79}.

244: }

245: Aleksandrov proved that any gluing that satisfies these two conditions 

246: corresponds to a unique polytope:

247: \begin{enumerate}

248: %\item The gluing uses up the whole perimeter, i.e.,

249: %the union of the glued intervals and endpoints is $\bP$.

250: \item No more than $2 \pi$ total face angle is glued together at any point; and

251: \item The complex resulting from the gluing is homeomorphic to a sphere.

252: (This condition is satisfied if, when

253: $\bP$ is viewed as a topological circle, and the interval

254: gluings as chords of the circle, then no pair of chords cross in the

255: $\bP$-circle.)

256: \end{enumerate}

257: Aleksandrov calls any complex (not necessarily a single polygon)

258: that satisfies these properties a {\em net}~\cite{a-kp-58}.%

259: \footnote{

260: 	This may derive from the German translation, {\em Netz}.

261: 	In fact, the Russian word Aleksandrov used is

262: 	closer to ``unfolding.''

263: }

264: We call a gluing that satisfies these conditions

265: an {\em Aleksandrov gluing}.

266: 

267: Although an

268: Aleksandrov gluing of a polygon forms a unique polytope, 

269: it is an open problem to compute

270: the three-dimensional structure of the polytope~\cite{o-fucg-00}.

271: Note that there is no specification of the fold (or ``crease'') lines;

272: and yet they are uniquely determined.

273: Henceforth we will say a polygon {\em folds\/} to

274: a polytope whenever it has an Aleksandrov gluing.

275: 

276: We should mention two features of Aleksandrov's theorem.

277: First, the polytope whose existence is guaranteed may be

278: {\em flat}, that is, a doubly-covered convex polygon.

279: We use the term ``polytope'' to include flat polyhedra.

280: Second,

281: condition~(2) specifies a face angle $\le 2 \pi$.

282: The case of equality with $2 \pi$ leads to a point

283: on the polytope at which there is no curvature,

284: i.e., a nonvertex.  We make explicit 

285: what counts as a vertex below.

286: 

287: \paragraph{Polygon/Polytope Notation.}

288: We will use $P$ throughout the paper for a polygon,

289: and $Q$ for a polytope.

290: Their boundaries are $\bP$ and $\bQ$ respectively.

291: The {\em curvature\/} $\g(x)$ of a point $x \in \bQ$

292: is $2 \pi$ minus the

293: sum of the face angles incident to $x$.

294: This ``angle deficit'' corresponds to the notion of Gaussian curvature.

295: We define vertices of polygons and polytopes to be

296: {\em essential\/} in the sense

297: that the boundary is not flat there:

298: the interior angle at a polygon vertex is different from

299: $\pi$, and

300: the curvature at a polytope vertex is different from $0$.

301: Because of these definitions, there is no direct correspondence

302: between the vertices of a polytope $Q$ and the vertices of

303: a polygon $P$ unfolding of $Q$:  a vertex of $Q$ may or may

304: not unfold to a vertex of $P$; and a vertex of $P$ may or

305: may not fold to a vertex of $Q$

306: (see Section~\secref{cut.glue.comp}).

307: At the risk of confusion, we will use the terms

308: ``vertex'' and ``edge'' for both polygons and polytopes,

309: but reserve ``node'' and ``arc'' for graphs.

310: We will use $n$ for the number of vertices of $P$ or $Q$, letting

311: the context determine which.

312: 

313: We will also freely employ two types of paths on the

314: surface of a polytope:

315: {\em geodesics}, which unfold (or ``develop'') to straight lines,

316: and {\em shortest paths}, geodesics which are in addition

317: shortest paths between their endpoints.

318: See, e.g., \cite{aaos-supa-97} for details and basic properties.

319: 

320: \subsection{Perimeter Halving}

321: As a straightforward application of Aleksandrov's theorem, we

322: prove that every convex polygon folds to a polytope.

323: We will see in Section~\secref{Sharp.Vertices} that the converse does not hold.

324: 

325: For two points $x,y \in \bP$, define $(x,y)$ be the open interval

326: of $\bP$ counterclockwise from $x$ to $y$,

327: and let $|x,y|$ be its length.

328: Define a {\em perimeter-halving gluing\/}

329: as one which glues $(x,y)$ to $(y,x)$.

330: 

331: \begin{lemma}

332: Every convex polygon folds to a polytope via perimeter halving.

333: \lemlab{perim.halving}

334: \end{lemma}

335: \begin{pf}

336: Let the perimeter of a convex polygon $P$ be $L$.

337: Let $x \in \bP$ be an arbitrary point on the boundary of $P$,

338: and let $y \in \bP$ be the midpoint of perimeter around $\bP$

339: measured from $x$, i.e., $y$ is the unique point

340: satisfying $|x,y|= |y,x| = L/2$.

341: See Fig.~\figref{perim.halving} for an example.

342: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

343: \begin{figure}[htbp]

344: \centering

345: \includegraphics[scale=1.0]{perim.halving.eps}

346: \caption{A perimeter-halving fold of a pentagon. The gluing mappings

347: of vertices $v_1$ and $v_3$ are shown.}

348: \figlab{perim.halving}

349: \end{figure}

350: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

351: 

352: Now glue $(x,y($ to $(y,x)$ in the natural way, mapping each

353: point $z$ with $|x,z|=d$ to the point $z'$ the same distance

354: from $x$ in the other direction: $|z',x|=d$.

355: We claim this is an Aleksandrov gluing.  It is a gluing

356: by construction.  Because $P$ is convex, each point along

357: the gluing path has $\le 2 \pi$ angle incident to it:

358: the gluing of two nonvertex points results in exactly $2\pi$,

359: and if either point is a vertex, the total angle is strictly

360: less than $2\pi$.

361: The resulting surface is clearly homeomorphic to a sphere.

362: By Aleksandrov's theorem, this gluing corresponds to a unique

363: polytope $Q_x$.

364: \end{pf}

365: 

366: In an Aleksandrov gluing of a polygon, a point in the interior of a

367: polygon edge that glues only to itself, i.e., where a crease folds

368: the edge in two, is called a {\em fold point}.  

369: A fold point corresponds

370: to a leaf of the gluing tree, and becomes a vertex of the polytope

371: with curvature $\pi$.

372: Points $x$ and $y$ in the above proof are fold points.

373: In Theorem~\theoref{continuum} we will show that different choices

374: of $x$ result in distinct polytopes $Q_x$, leading to the

375: conclusion that every convex polygon folds to an infinite number

376: of polytopes.

377: 

378: \section{Cut Trees and Gluing Trees}

379: \seclab{cut.glue.trees}

380: 

381: The four main objects we study are polygons, polytopes,

382: cut trees, and gluing trees.

383: It will be useful in spots to distinguish between 

384: a {\em geometric\/} tree $\T$ composed of a union of line

385: segments, and the more familiar {\em combinatorial\/} tree $T$

386: of nodes and arcs.

387: A {\em geometric cut tree\/} $\T_C$ for a polytope $Q$ is a tree drawn on $\bQ$,

388: with each arc a polygonal path, 

389: which leads to a polygon unfolding when the surface is cut along $\T$,

390: i.e., flattening $Q \setminus \T$ to a plane.

391: A {\em geometric gluing tree\/} $\T_G$ specifies how $\bP$ is glued to

392: itself to fold to a polytope.

393: There is clearly a close correspondence between $\T_C$ and $\T_G$,

394: which are in some sense the same object, one viewed from

395: the perspective of unfolding, one from the perspective of folding.

396: It will nevertheless be useful to retain a distinction between

397: them, and especially their combinatorial counterparts,

398: which we define below after stating some basic properties.

399: 

400: \subsection{Cut Trees}

401: \seclab{Cut.Trees}

402: 

403: \begin{lemma}

404: If a polygon $P$ folds to a polytope $Q$, $\bP$ maps

405: to a tree $\T_C \subset \bQ$, the geometric cut tree, with

406: the following properties:

407: \begin{enumerate}

408: \item $\T_C$ is a tree.

409: \item $\T_C$ spans the vertices of $Q$.

410: \item Every leaf of $\T_C$ is at a vertex of $Q$.

411: \item 

412: A point of $\T_C$ of

413: degree $d$ (i.e., one with $d$ incident segments)

414: corresponds to exactly $d$ points of $\bP$.

415: Thus a leaf corresponds to a unique point

416: of $\bP$.

417: \item Each arc of $T_C$ is a polygonal path on $Q$.

418: \end{enumerate}

419: \lemlab{cut.tree}

420: \end{lemma}

421: \begin{pf}

422: \begin{enumerate}

423: \item If $\T_C$ contained a cycle, then it would unfold to disconnected pieces,

424: contradicting the assumption that $Q$ is folded from a single polygon $P$.

425: Thus $\T_C$ is a forest.  But because $\T_C$ is constructed by gluing

426: the connected path $\bP$ to itself, it must be connected.

427: So $\T_C$ is a tree.

428: \item If a vertex $v$ of $Q$ is not touched by $\T_C$, then,

429: because $Q$ is not flat at $v$, $P$ is not planar,

430: a contradiction to the assumption that $P$ is a polygon.

431: \item Suppose a leaf $x$ of $\T_C$ is interior to a face or edge of $Q$.

432: Then it is surrounded by $2 \pi$ face angle on $Q$, and so unfolds to a

433: point $x$ of $P$ similarly surrounded.  But by assumption, $x$ is on the

434: boundary of a simple polygon $P$, a contradiction.

435: \item Gluing exactly two distinct points of $x,y \in \bP$ together implies that

436: neighborhoods of $x$ and $y$ are glued, which leads to the interior of

437: an arc of the cut tree, i.e., a degree-$2$ point of $\T_C$.

438: Note that either or both of these points might be vertices of $P$.

439: In general, if $p \in \T_C$ has $d$ incident cut segments,

440: $p$ unfolds to $d$ distinct points of $\bP$.

441: \item If an arc of $\T_C$ is not a polygonal path,

442: then neither side unfolds to a polygonal path, contradicting

443: the assumption that $P$ is a polygon.

444: \end{enumerate}

445: \end{pf}

446: 

447: When counting cut trees, we will rely on their combinatorial

448: structure.

449: There are several natural definitions of this structure, which

450: are useful in different circumstances.  We first discuss some

451: of the options.

452: \begin{enumerate}

453: \item Make every segment of $\T_C$ an arc of $T_C$.

454: Although this is very natural, it means there are an infinite number

455: of different cut trees for any polytope, for the path

456: between any two polytope vertices could be an arbitrarily

457: complicated polygonal path, leading to different combinatorial

458: trees.

459: \item Make every point where a path of $\T_C$ crosses an edge

460: of the polytope a node of $T_C$.  This again leads to trivially

461: infinite numbers of cut trees when a path of $\T_C$ zigzags 

462: back and forth over an edge of $Q$. 

463: \item Exclude this possibility by forcing the paths between

464: polytope vertices to be

465: geodesics, and again make polytope edge crossings nodes of $T_C$.

466: This excludes many interesting cut trees---all those where a

467: polygon vertex is glued to a point with angle sum $2\pi$.

468: \item Make every maximal path of $\T_C$ consisting only of

469: degree-$2$ points a single arc of $T_C$.  This has the

470: undesirable effect of having polytope vertices in the interior

471: of such a path disappear from $T_C$.

472: \end{enumerate}

473: 

474: Threading between these possibilities, we define

475: the {\em combinatorial cut tree\/} $T_C$ corresponding

476: to a geometric cut tree $\T_C$ as

477: the labeled graph with a node (not necessarily labeled)

478: for each point of $\T_C$ with degree not equal to $2$,

479: and a labeled node for each point of $\T_C$ that corresponds

480: to a vertex of $Q$ (labeled by the vertex label);

481: arcs are determined by the polygonal paths of $\T_C$

482: connecting these nodes.

483: An example is shown in

484: Fig.~\figref{cut.tree}.  Note that not every node 

485: of the tree is labeled, but

486: every polytope vertex label is used at some node.

487: All degree-$2$ nodes are labeled.

488: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

489: \begin{figure}[htbp]

490: \centering

491: \includegraphics[width=0.6\linewidth]{cut.tree.eps}

492: \caption{(a) Geometric cut tree $\T_C$ on the surface of a cube; 

493: (b) The corresponding combinatorial cut tree $T_C$.}

494: \figlab{cut.tree}

495: \end{figure}

496: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

497: 

498: Although this definition avoids some of the listed pitfalls,

499: it does have the

500: undesirable consequence of counting different geodesics

501: on $\bQ$ between two polytope vertices as the same arc of $T_C$.

502: Thus 

503: the two unfoldings shown in Fig.~\figref{not.sp} (below)

504: have the same combinatorial cut tree under our definition,

505: even though the geodesic in (c) spirals twice around

506: compared to once around in (a).

507: 

508: \subsection{Gluing Trees}

509: Let a convex polygon $P$ have vertices $v_1,\ldots,v_n$,

510: labeled counterclockwise,

511: and edge $e_i$, $i=1,\ldots,n$ the open segment of $\bP$

512: after $v_i$.

513: There is less need to discuss the geometric gluing tree,

514: so we concentrate on the combinatorial gluing tree $T_G$.

515: $T_G$ is a tree representing the

516: identification of $\bP$ with itself.

517: Any point of $\bP$ that is identified with more or less than

518: one other distinct point of $\bP$ becomes a node

519: of $T_G$, as well as any point to which a vertex is

520: glued.

521: (Note that this means there may be nodes of degree $2$.)

522: So every vertex of $P$ maps to a node of $T_G$;

523: each node is labeled with the set of all the

524: elements (vertices or edges) that are glued together there.

525: A leaf that is a fold point is labeled by the edge label only.

526: Every nonleaf node has at least one vertex label, and at most one edge label.

527: A simple example is shown in Fig.~\figref{tri.tetra}.%

528: \footnote{

529: 	Gluing trees can be drawn by folding up the polygon toward

530: 	the viewer (as in this figure), or folding the polygon

531: 	away.  We employ both conventions but always note which

532: 	is followed.

533: }

534: Here the central node of $T_G$ is assigned

535: the label $\{v_1,v_3,e_3\}$.

536: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

537: \begin{figure}[htbp]

538: \centering

539: \includegraphics[width=10cm]{tri.tetra.eps}

540: \caption{(a) A gluing of an equilateral triangle $P$: $v_1$ and $v_3$

541: are glued to point $z$;

542: (b) the corresponding gluing tree $T_G$ [folding up].

543: Points $x$ and $y$ become fold points of the resulting tetrahedron.}

544: \figlab{tri.tetra}

545: \end{figure}

546: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

547: A more complicated example is shown in 

548: Fig.~\figref{zed.tetra2}.\footnote{

549: 	We found this example by an enumeration algorithm that

550: 	will not be discussed in this report.

551: }

552: The polygon shown folds (amazingly!) to a tetrahedron

553: by creasing as illustrated in (a).  All four tetrahedron

554: vertices are fold points.  The corresponding gluing tree

555: is shown in (b) of the figure.  The two interior nodes

556: of $T_G$ have labels $\{v_1,v_6,e_1\}$ and $\{v_2,v_5,e_5\}$.

557: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

558: \begin{figure}[htbp]

559: \centering

560: \includegraphics[width=\linewidth]{zed.tetra2.eps}

561: \caption{(a) A polygon, with fold creases shown dotted;

562: (b) A gluing tree $T_G$ [folding away] corresponding to the crease pattern.}

563: \figlab{zed.tetra2}

564: \end{figure}

565: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

566: 

567: Later (Lemma~\lemref{gluing.tree}) will show that the gluing tree

568: is determined by a relatively sparse set of gluing instructions.

569: 

570: 

571: \subsection{Comparison of Cut and Gluing Trees}

572: \seclab{cut.glue.comp}

573: 

574: \begin{lemma}

575: Let $T_C$ be a combinatorial cut tree for polytope $Q$

576: that unfolds to a polygon $P$, and let $T_G$ be the 

577: combinatorial gluing

578: tree that folds $P$ to $Q$.

579: If all degree-$2$ nodes are removed by contraction,

580: $T_C$ and $T_G$ are isomorphic as unlabeled graphs.

581: \lemlab{cut.glue}

582: \end{lemma}

583: \begin{pf}

584: %Because we have stipulated that all vertices are essential,

585: %the face angle incident to a vertex $v$ of $Q$ is strictly

586: %less than $2 \pi$. 

587: %If $v$ has face angle $\pi$ and $v$ is a leaf of $T_C$,

588: %then it unfolds to a point of $\bP$ with interior angle

589: %$\pi$, a fold point.  This establishes~(1b).

590: %If the face angle at $v$ is different from $\pi$ and $v$ is

591: %a leaf of $T_C$, then $v$ maps to a vertex of $P$ with that interior

592: %angle.

593: %If $v$ is not a leaf of $T_C$, the cuts through $v$ partition

594: %the face angle into at least two parts, at least one of which

595: %must be different from $\pi$ (because the sum is $< 2 \pi$).

596: %This establishes that at least one polygon vertex is glued

597: %there, establishing~(1a).

598: Let $(a,b,c)$ be three consecutive nodes on a path in a tree $T$,

599: with $b$ of degree $2$. Removing $b$ by contraction deletes $b$

600: and replaces it with the arc $(a,c)$.

601: Applying this to both $T_C$ and $T_G$ produces two trees

602: $T'_C$ and $T'_G$ without degree-$2$ nodes.  As the trees were

603: defined to include nodes for each point whose degree differs

604: from $2$, it must be that $T'_C$ and $T'_G$ have isomorphic

605: structures.  Of course they are labeled differently, but without

606: the labels, they are isomorphic graphs.

607: \end{pf}

608: 

609: Note that vertices in $Q$ and vertices in $P$ do not necessarily

610: map to one another:  A vertex of $Q$ can map to an 

611: interior point of $\bP$, and a vertex of $P$ can map to

612: a point interior to a face or edge of $Q$.

613: This affects the labeling of the two trees, but

614: they have essentially the same structure.

615: 

616: \section{Cut Trees for Convex Unfoldings}

617: \seclab{Convex.Unfoldings}

618: Before embarking on general enumeration results, 

619: we specialize the discussion to convex unfoldings,

620: and derive some constraints on the possible cut trees that

621: lead to convex unfoldings.

622: 

623: \subsection{Stronger Characterization}

624: We now sharpen the characterization of cut trees

625: (and via Lemma~\lemref{cut.glue}, of gluing trees)

626: under the restriction that the unfolding must be a convex 

627: polygon.

628: We first strengthen Lemma~\lemref{cut.tree}(5), which

629: only required arcs to be polygonal paths:

630: \begin{lemma}

631: Every arc of

632: a cut tree $T_C$ that leads to a convex unfolding

633: must be a {\em geodesic\/} on $Q$ (paths that unfold to straight segments),

634: but arcs might not be

635: shortest paths on $Q$.

636: \lemlab{not.sp}

637: \end{lemma}

638: \begin{pf}

639: Suppose an arc $a$ of $T$ is not a geodesic.  Then it does

640: not unfold to a straight line. Suppose a point $x \in a$ is a point

641: in the relative interior of $a$

642: at which the unfolding is locally not straight.  Then only one

643: of the two points of $\bP$ that correspond to $x$ can have an

644: interior angle $\le \pi$ in $P$, showing that $P$ has at least one

645: reflex angle.

646: This establishes that arcs of $T_C$ must be geodesics.

647: We now show that this claim cannot be strengthened to shortest

648: paths by an explicit example.

649: 

650: Let $Q$ be a doubly-covered rectangle with

651: vertices $v_i$, $i=1,2,3,4$, as shown in Fig.~\figref{not.sp}(a).

652: Let $x$ be the midpoint of edge $v_1 v_4$.

653: Let $T_C$ be the path $(v_1, v_2, x, v_3, v_4)$, where the subpath

654: $(v_2, x, v_3)$ is half on the upper rectangular face,

655: and half on the bottom face.  Clearly this subpath is not a shortest path,

656: although it is a geodesic.  The corresponding convex unfolding

657: is shown in Fig.~\figref{not.sp}(b).

658: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

659: \begin{figure}[htbp]

660: \centering

661: \includegraphics[width=0.7\linewidth]{not.sp.eps}

662: \caption{(a)Doubly covered rectangle with cut path; (b) Unfolding.

663: (c-d): Another cut path and its unfolding.}

664: \figlab{not.sp}

665: \end{figure}

666: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

667: 

668: This example can be modified to a nondegenerate 

669: ``sliver'' tetrahedron by perturbing

670: one vertex to lie slightly out of the plane of the other three.

671: \end{pf}

672: 

673: \noindent

674: Fig.~\figref{not.sp}(c-d) shows that we cannot even bound the

675: length of a geodesic arc of $T_C$.

676: 

677: One immediate corollary of Lemma~\lemref{not.sp} is that cuts need

678: not follow polytope edges (which are all shortest paths),

679: i.e., not every convex unfolding is an edge unfolding.

680: 

681: 

682: \subsection{Necessary Conditions: Sharp Vertices}

683: \seclab{Sharp.Vertices}

684: 

685: We define a vertex of a polytope to be {\em sharp\/} if it has curvature $\ge \pi$,

686: and {\em round\/} if its curvature is $< \pi$.

687: The following theorem gives a simple necessary condition for a polytope

688: to have a convex unfolding.

689: We employ this fact implied by the Gauss-Bonnet theorem:

690: 

691: \begin{fact}

692: The sum of the curvatures of all the vertices of a polytope is

693: exactly $4 \pi$.

694: \factlab{4pi}

695: \end{fact}

696: 

697: 

698: 

699: \begin{theorem}

700: If a polytope $Q$ has a convex unfolding via a cut

701: tree $T_C$, then each leaf of $T_C$ is at a sharp vertex.

702: Moreover, $Q$ must

703: have at least two sharp vertices.

704: \theolab{two.sharp}

705: \end{theorem}

706: \begin{pf}

707: Let $P$ be a convex polygon to which $Q$ unfolds via cut tree $T_C$.

708: By Lemma~\lemref{cut.tree}(3), the leaves of $T_C$ are at vertices

709: of $Q$.  Let $x$ be a leaf of $T_C$, at a vertex $v$ with curvature 

710: $\g(v) = \g$.

711: Point $x \in \bQ$ corresponds to a unique point $y \in \bP$

712: by Lemma~\lemref{cut.tree}(4).

713: The internal angle at $y$ in $P$ is $2 \pi - \g$.  Because

714: $P$ is convex, we must have 

715: $$

716: 2 \pi - \g \le \pi

717: $$

718: and so $\g(v) \ge \pi$.  Thus $v$ is sharp.

719: Because $T_C$ must have at least two distinct leaves, the lemma follows.

720: \end{pf}

721: 

722: \begin{cor}

723: Of the five Platonic solids, only the regular tetrahedron has

724: a convex unfolding.

725: \end{cor}

726: \begin{pf}

727: The curvatures at the vertices of the solids are:

728: \begin{eqnarray*}

729: 2 \pi - 3(\pi/3)  = &  \pi & \mbox{} \\

730: 2 \pi - 3(\pi/2)  = &  \pi/2 & < \pi \\

731: 2 \pi - 4(\pi/3)  = &  2\pi/3 & < \pi \\

732: 2 \pi - 3(3\pi/5) = &  \pi/5 & < \pi \\

733: 2 \pi - 5(\pi/3)  = &  \pi/3 & < \pi

734: \end{eqnarray*}

735: Only the tetrahedron has sharp vertices.

736: \end{pf}

737: 

738: We next show that two natural extensions of the previous results

739: fail.

740: 

741: \begin{lemma}

742: There is a tetrahedron with no convex unfolding.

743: \lemlab{not.tetra}

744: \end{lemma}

745: \begin{pf}

746: Let $Q_1$ be a tetrahedron whose vertices $v_1,v_2,v_3$ form an

747: equilateral triangle base in the $xy$-plane, with apex $v_4$

748: centered at a great height $z$ above. 

749: See Fig.~\figref{not.tetra}. 

750: Let $\g_i$ be the curvature of

751: vertex $v_i$.

752: If the face angle of each triangle incident to $v_4$ is $\e$,

753: then $\g_4 = 2 \pi - 3 \e$,

754: and $\g_i$ for $i=1,2,3$ is

755: $$2 \pi - [\pi/3 + 2(\pi-\e/2)] = 2 \pi /3 + \e$$

756: Choosing $z$ large makes $\e$ small,

757: and then $Q_1$ has just one sharp vertex.

758: Theorem~\theoref{two.sharp} then establishes the claim.

759: \end{pf}

760: 

761: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

762: \begin{figure}[htbp]

763: \begin{minipage}[b]{0.45\linewidth}

764: \centering

765: \includegraphics[scale=1.0]{not.tetra.eps}

766: \caption{A tetrahedron $Q_1$ without a convex unfolding.}

767: \figlab{not.tetra}

768: \end{minipage}%

769: \hspace{5mm}%

770: \begin{minipage}[b]{0.45\linewidth}

771: \centering

772: \includegraphics[scale=1.0]{not.two.sharp.eps}

773: \caption{A polytope $Q_2$ with two sharp vertices but no convex unfolding.}

774: \figlab{not.two.sharp}

775: \end{minipage}

776: \end{figure}

777: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

778: 

779: 

780: \begin{lemma}

781: There is a polytope with two sharp

782: vertices but with no convex unfolding.

783: \lemlab{not.two.sharp}

784: \end{lemma}

785: \begin{pf}

786: Our proof of this lemma is less straightforward, although the

787: example is simple.

788: Let $Q_2$ be the polytope formed by joining two copies of $Q_1$

789: from Lemma~\lemref{not.tetra} at their bases, as shown in

790: Fig.~\figref{not.two.sharp}. 

791: $Q_2$ is a $5$-vertex polytope, with vertices $v_1,\ldots,v_4$

792: as in $Q_1$, and $v_5$ the reflection of $v_4$ in the

793: central triangle $C = \triangle v_1 v_2 v_3$.

794: Again let $\e$ be the face angle incident to $v_4$ (and symmetrically $v_5$),

795: and choose $\e$ small so that only $v_4$ and $v_5$ are sharp vertices.

796: 

797: By Lemma~\lemref{cut.tree}(3), if $Q_2$ has a convex unfolding,

798: the cut tree must be a path with its two leaves at the two sharp

799: vertices.  By  Lemma~\lemref{cut.tree}(5), the path must be composed

800: of geodesics.  We now analyze the geodesics starting at $v_5$ and

801: show that there can be no piecewise simple geodesic path that

802: passes through all the vertices of $Q_2$.

803: 

804: We group the geodesics starting at $v_5$ into three classes:

805: \begin{enumerate}

806: \item The three geodesics that pass through a midpoint of an edge of

807: triangle

808: $C$.  Each of these passes through $v_4$

809: before encountering any of the other vertices, and so cannot serve

810: as the cut path.

811: \item The three geodesics that pass through a vertex of  $C$.

812: Because these vertices have low curvature ($2\e$), the geodesic must emerge

813: nearly headed toward $v_4$:  it cannot turn to hit another vertex of $C$

814: without creating a reflex angle in the unfolding.

815: If the geodesic goes directly to $v_4$, then again this cannot serve as

816: the cut path.  So it must head towards $v_4$ but miss it.  We

817: group this type of geodesic with the third class.

818: \item Geodesics that pass though an interior point of an edge

819: of $C$, but not the midpoint.

820: These geodesics all head toward $v_4$ but miss it.

821: \end{enumerate}

822: We now argue that all the geodesics in the third class (the only remaining

823: candidates) self-intersect after looping around $v_4$.

824: This will then establish the lemma.

825: 

826: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

827: \begin{figure}[htbp]

828: \centering

829: \includegraphics[width=11cm]{geodesic.crossing.eps}

830: \caption{A geodesic from $v_5$ that passes by $v_4$. The path of the 

831: geodesic on $\triangle v_1 v_2 v_4$ is shown to the left.}

832: \figlab{geodesic.crossing}

833: \end{figure}

834: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

835: 

836: An unfolding of a typical geodesic is shown in 

837: Fig.~\figref{geodesic.crossing}.

838: By choosing $\e$ small, we can arrange that every such geodesic crosses

839: several unfoldings of the three faces incident to $v_4$ before returning

840: back down to triangle $C$.  As can be seen from the copy of face

841: $\triangle v_1 v_2 v_4$ to the side, the path crosses each face several

842: times slanting one way, and then returns slanting the other way.

843: In the vicinity of the closest approach to $v_4$, the path

844: must self-cross.  We now establish this more formally.

845: 

846: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

847: \begin{figure}[htbp]

848: \centering

849: \includegraphics[width=0.7\linewidth]{X.eps}

850: \caption{$xa'$ and $yb'$ necessarily cross below the point $p$

851: of the geodesic closest to $v_4$.}

852: \figlab{X}

853: \end{figure}

854: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

855: Consider the unfolding of the three faces incident to $v_4$ (now viewed

856: as a unit) that includes the point $p$ of closest approach between the

857: geodesic and $v_4$; see Fig.~\figref{X}.

858: Let the geodesic cross the edge $v_1 v_4$ at points $a$, $x$, $y$, and $b$

859: in that order, with $xy$ including $p$.

860: Then $|v_4 b| > |v_4 x|$ and $|v_4 a| > |v_4 y|$,

861: because the distance from $v_4$ monotonically increases on either

862: side of $p$.

863: Thus the images $a'$ and $b'$ of $a$ and $b$ must fall below

864: $y$ and $x$ respectively in the figure.  Thus the geodesic must

865: cross somewhere in the stretch immediately before and after

866: the closest approach.

867: 

868: All we need for this argument to hold in general is for the geodesic to

869: cross three complete unfoldings of the three faces incident to $v_4$

870: before returning to the lower half of $Q_2$.  But this is easily

871: arranged by choosing $\e$ small.

872: 

873: We have shown that no geodesic starting from $v_5$ may serve as

874: a cut path for a convex unfolding.  Therefore $Q_2$ has no convex unfolding.

875: \end{pf}

876: 

877: \subsection{Necessary Conditions: Combinatorial Structure}

878: 

879: We now study the combinatorial structure of cut trees that lead to

880: convex unfoldings.

881: The following theorem is due to Shephard~\cite{s-cpcn-75},

882: although under different assumptions and with a different proof.%

883: \footnote{

884:         Shephard concludes that cut trees cannot have four leaves,

885:         an incorrect claim under our assumptions.

886: }

887: 

888: 

889: \begin{theorem}

890: If a polytope $Q$ of $n \neq 4$ vertices has a convex unfolding,

891: then the corresponding cut tree $T_C$ has two or three leaves:

892: it is either a path, or a `{\tt Y}' (a single degree-$3$ node).

893: If $n=4$, then additionally it may have four leaves, and

894: have the combinatorial structure of `{\tt +}' (a single degree-$4$ node),

895: or two degree-$3$ nodes connected by an edge, which we will call

896: a `{\tt I}'.

897: \theolab{cut.comb}

898: \end{theorem}

899: \begin{pf}

900: Let the cut tree $T_C$ unfold $Q$ to a convex polygon.

901: By Theorem~\theoref{two.sharp},

902: each leaf of $T_C$ must be at a sharp vertex $v$, and so have

903: curvature $\g(v) \ge \pi$.

904: If $T_C$ has more than four leaves $v$ (and therefore $n > 4$,

905: i.e., we are in the $n \neq 4$ case of the theorem claim),

906: $\sum_v \g(v) > 4 \pi$, which violates the Gauss-Bonnet theorem.

907: Therefore $T_C$ has no more than four leaves.

908: If $T_C$ has just two or three leaves, then the only possible 

909: combinatorial structures for $T_C$ are the two claimed in the theorem:

910: a path, and a `{\tt Y}'.

911: (Note that it is possible that $n=3$, when $Q$ is a doubly-covered

912: triangle.)

913: 

914: So assume that $T_C$ has exactly four leaves.

915: Because each leaf vertex is sharp,

916: $\sum_v \g(v) \ge 4 \pi$; on the other hand, we know the sum

917: over all vertices is equal $4 \pi$.  Therefore we know that

918: each leaf has curvature exactly $\pi$ and that

919: the leaves of $T_C$ are at the only vertices of $Q$.  Thus $n=4$

920: and $Q$ is a tetrahedron.

921: The only additional possible combinatorial structures for a tree with

922: four leaves are the two claimed in the theorem:

923: a `{\tt +}' and a `{\tt I}'.  Note that in both these cases, the internal

924: node(s) of $T_C$ are not at vertices of $Q$.

925: \end{pf}

926: 

927: A simple example of the `{\tt I}' possibility is shown in Fig.~\figref{rect.YY}.

928: If the rectangle is modified to become a square, the

929: `{\tt I}' becomes a `{\tt +}'.

930: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

931: \begin{figure}[htbp]

932: \centering

933: \includegraphics[width=0.5\textwidth]{rect.YY.eps}

934: \caption{A doubly-covered rectangle unfolds with a `{\tt I}' cut tree whose

935: leaves are the four rectangle corners.

936: Note all four rectangle corners are fold vertices.}

937: \figlab{rect.YY}

938: \end{figure}

939: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

940: 

941: \section{Counting Foldings: Gluing Trees}

942: \seclab{Counting.Gluings}

943: In this section we move beyond 

944: Lemma~\lemref{perim.halving}, which shows that every convex polygon

945: folds to a polytope, and explore how many different

946: ways there are to fold a given polygon, as measured by the

947: number of combinatorially distinct Aleksandrov gluing trees.

948: In Section~\secref{Noncongruent.Polytopes} we count instead the number of

949: distinct polytopes that might be produced from a given polygon.

950: In both cases, we will also examine the restriction

951: to convex polygons, which not surprisingly yields sharper results.

952: 

953: \subsection{Unfoldable Polygons}

954: We start with a natural and easily proved claim:

955: 

956: \begin{lemma}

957: Some polygons cannot be folded to any polytope.

958: \lemlab{unf}

959: \end{lemma}

960: \begin{pf}

961: Consider the polygon $P$ shown in Fig.~\figref{unf}.

962: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

963: \begin{figure}[htbp]

964: \centering

965: \includegraphics[height=3cm]{unf.eps}

966: \caption{An unfoldable polygon.}

967: \figlab{unf}

968: \end{figure}

969: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

970: $P$ has three consecutive reflex vertices $(a, b, c)$, with

971: the exterior angle $\b$ at $b$ small.  All other vertices

972: are convex, with interior angles strictly larger than $\b$.

973: 

974: Either the gluing ``zips'' at $b$, leaving $b$ a leaf of

975: $T_G$, or some other point(s) of $\bP$ glue to $b$.

976: The first possibility forces $a$ to glue to $c$, exceeding

977: $2\pi$ there; so this gluing is not Aleksandrov.

978: The second possibility cannot occur with $P$, because

979: no point of $\bP$ has small enough internal angle to fit at $b$.

980: Thus there is no Aleksandrov gluing of $P$.

981: \end{pf}

982: 

983: It is natural to wonder what the chances are that a random polygon

984: could fold to a polytope.

985: This is difficult to answer without a precise definition of

986: ``random,'' but we feel any reasonable definition would

987: lead to the same answer:

988: 

989: \begin{conj}

990: The probability that a random polygon of $n$ vertices

991: can fold to a polytope approaches $0$ as $n \rightarrow \infty$.

992: \end{conj}

993: \begin{pf}

994: ({\em Sketch\/}.)

995: %OLD PROOF -- wrong

996: %Assume that, for sufficiently large $n$, the number of

997: %reflex vertices of random polygons approaches $\frac{1}{2}$

998: %(or exceeds any constant in $(0,1)$.)

999: %Then, for large $n$, patterns of three consecutive reflex vertices

1000: %$(a, b, c)$

1001: %are common.  Note that in Fig.~\figref{unf}, it is not necessary

1002: %that $|ab|=|bc|$ to block zipping at $b$.  Thus, the need to

1003: %fit a vertex into the exterior angle at middle $b$-vertices of

1004: %$3$-reflex chains will be common.

1005: %There should be positive probability $p>0$ that some convex

1006: %vertex angle is small enough to fit into the exterior angle at $b$,

1007: %assuming a near equal distribution of internal and external angles.

1008: %Once the gluing at $b$ is made, the $\bP$ chain is pinched

1009: %to two chains, with the same problem repeated in each.

1010: %The probability that this process can proceed $k$ times

1011: %is $p^k$, which approaches $0$ as $k \rightarrow \infty$.

1012: %And $k$ is some increasing function of $n$.

1013: %NEW PROOF:

1014: Assume that random polygons on $n$ vertices satisfy two properties:

1015: \begin{enumerate}

1016: \item The distribution of the polygon angles approaches

1017: the uniform distribution on the interval $(0,2\pi)$ as $n \rightarrow \infty$.

1018: In particular, the number of reflex and convex vertices

1019: approaches balance.

1020: \item The distribution of polygon edge lengths approaches

1021: some continuous density distribution.

1022: \end{enumerate}

1023: For large $n$, we expect $P$ to have $r=n/2$ reflex vertices.

1024: Each of these reflex vertices $a$ faces one of two fates in the gluing

1025: tree:  either it becomes a leaf by ``zipping'' at $a$;

1026: or at least one convex vertex $b$ (of sufficiently small angle)

1027: is glued to $a$.

1028: The number of reflex vertices that can be zipped is limited

1029: by Fact~\factref{4pi}:

1030: if $a$ has angle $\a$, zipping there adds $2\pi-\a$ to the

1031: curvature; but the total curvature is limited to $4\pi$.

1032: Suppose we zip the largest $k$ angles out of the $r$ reflex vertices

1033: (the largest angles increment the curvature the least).

1034: Then one can compute that,

1035: under the uniform angle distribution assumption,

1036: these $k$ angles have an expected

1037: curvature sum of

1038: \begin{equation}

1039: \frac{1}{2} \frac{\pi}{r} k^2 \; .

1040: \end{equation}

1041: (For example, for $r=100$, the largest $k=10$ have an expected curvature

1042: sum of $\pi/2$.)

1043: Limiting this to $4\pi$ implies that the expected maximum number

1044: of reflex vertices that can be zipped without exceeding $4\pi$

1045: curvature is

1046: \begin{equation}

1047: k \le 2 \sqrt{2 r} = 2 \sqrt{n} \; .

1048: \end{equation}

1049: (For example, for $r=1000$ reflex vertices, the largest $k=89$

1050: lead to a curvature of $\approx 4\pi$.)

1051: Thus, at most a small portion of the reflex vertices can be zipped;

1052: the remainder (expected number: $n/2 - 2\sqrt{n}$) 

1053: must be glued to convex vertices.  We now show that this gluing is

1054: not in general possible.

1055: 

1056: Let $a$ be a reflex vertex with angle $\a$, 

1057: and $b$ a convex vertex whose angle $\b$ satisfies $\b \le 2\pi-\a$,

1058: so that $b$ can glue to $a$.

1059: It could be that this gluing forces one or more reflex vertices

1060: adjacent to $a$ or $b$ to glue to edges incident to $a$ or $b$,

1061: in which case the gluing is not possible (i.e., it is not an

1062: Aleksandrov gluing).

1063: If the adjacent vertices are convex, and/or the edge lengths are

1064: such that the gluing is Aleksandrov, then, in general, two new reflex vertices

1065: are created, as is illustrated in

1066: Fig.~\figref{random}.

1067: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1068: \begin{figure}[htbp]

1069: \centering

1070: \includegraphics[width=0.75\linewidth]{random.eps}

1071: \caption{(a) Reflex vertex $a$, convex vertex $b$; 

1072: (b) Two new reflex vertices produced by gluing $b$ to $a$.}

1073: \figlab{random}

1074: \end{figure}

1075: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1076: 

1077: To be more precise, let $A = |a a_1|$ be the length of the edge

1078: incident to $a$ which is glued to the length $B = |b b_1|$ of

1079: an edge incident to $b$.

1080: If $A > B$ and $b_1$ is reflex, the gluing is {\em not\/} Aleksandrov;

1081: but if $b_1$ is convex, a new reflex vertex is created at $b_1$.

1082: Symmetrically, if $B > A$ and $a_1$ is reflex, the gluing is

1083: not possible; but if $a_1$ is convex, a new reflex vertex is

1084: created at $a_1$.  The only circumstance in which the gluing

1085: is Aleksandrov and a new reflex vertex is not created is when

1086: $A = B$ and both $a_1$ and $b_1$ are convex with an angle sum

1087: of no more than $\pi$.

1088: 

1089: Under the assumption that the edge lengths approach some continuous

1090: distribution, the probability that two lengths match exactly

1091: approaches $0$.  Thus we conclude that gluing convex vertices

1092: to reflex vertices does not remove reflex vertices, but rather

1093: creates new ones in shorter polygonal chains, one new reflex

1094: vertex in each of the two chains produced by the gluing.

1095: Note that gluing several convex vertices to one reflex vertex

1096: does not change matters: we can view the first convex vertex

1097: as simply leaving a reflex remainder, and argue as above.

1098: 

1099: Thus, any gluing of a random polygon for large $n$

1100: will lead to shorter and and shorter chains ``pinched'' between reflex-convex

1101: gluings, each of which will contain at least one reflex vertex

1102: (actually, two reflex vertices for those pinched on both sides).

1103: Eventually these chains reach the point where either there are

1104: no convex vertices that fit into the reflex vertex gap, or

1105: there are no convex vertices at all.  In either case, the chain

1106: cannot be glued:  the reflex vertex would have to glue to a point

1107: interior to an edge, violating the Aleksandrov condition that

1108: no point have more than $2\pi$ glued angle.

1109: \end{pf}

1110: 

1111: The proof above hinges on the unlikeliness of matching

1112: edge lengths.  It is therefore natural to wonder if

1113: the same result holds for polygons all of whose edge lengths

1114: are the same.  Again we believe it does:

1115: 

1116: \begin{conj}

1117: The probability that a random polygon of $n$ vertices,

1118: all of whose edges have unit length,

1119: can fold to a polytope, approaches $0$ as $n \rightarrow \infty$.

1120: \end{conj}

1121: \begin{pf}

1122: ({\em Sketch\/}.)

1123: Assume a model of random polygons such that the angles are

1124: probabilistically independent and uniformly distributed

1125: in $(0,2\pi)$ as $n \rightarrow \infty$.

1126: The restriction to unit edge lengths means that all gluings are

1127: vertex to vertex (no vertex is ever glued to the interior of an edge).

1128: The gluing is Aleksandrov iff the angles glued together sum to at most

1129: $2\pi$

1130: everywhere.

1131: 

1132: Consider gluing two vertices to one another.  Because their angles

1133: are independent, the chance that the gluing is legal is $1/2$ (the sum of

1134: their distributions is uniform between $0$ and $4\pi$). Gluing $k$ pairs then has

1135: a $1/2^k$ chance of being Aleksandrov.

1136: 

1137: As in the above proof sketch, the gluing tree cannot have too

1138: many leaves.  Zipping just $2 \sqrt{n}$

1139: reflex vertices uses up all $4 \pi$ of

1140: curvature. 

1141: So the number of leaves is only about $2 \sqrt{n}$. As we will see

1142: in Theorem~\theoref{gluing.upper} below,

1143: specifying a ``source'' for each leaf pins down the whole

1144: tree structure.  So by selecting $4 \sqrt{n}$ vertices for the leaves and

1145: their sources, the gluing tree is determined.

1146: 

1147: Therefore we should compare the number of different gluing

1148: trees,

1149: \begin{equation}

1150: \left(

1151: 	\begin{array}{c}

1152: 		n \\ 

1153: 		4 \sqrt{n}

1154: 	\end{array}

1155: \right)

1156: \eqlab{num}

1157: \end{equation}

1158: to the probability that each one is Aleksandrov,

1159: \begin{equation}

1160: \frac{1}{2^{n-4 \sqrt{n}}}

1161: \eqlab{den}

1162: \end{equation}

1163: Note here we conservatively only concern ourselves with

1164: degree-$2$ vertex-to-vertex gluings; junctions of degree $d > 2$ have

1165: a lower probability of summing to no more than $2\pi$.

1166: We also ignore the change to the angle distribution 

1167: caused by the removal of the leaf vertices.

1168: 

1169: Using Stirling's approximation shows that the 

1170: $\log$ of Eq.~\eqref{num} grows as

1171: $2 \sqrt{n} \log n$; but the $\log$ of

1172: Eq.~\eqref{den} grows as $n$. So their ratio

1173: approaches $0$ as $n \rightarrow \infty$.

1174: \end{pf}

1175: 

1176: 

1177: We leave these results on random polygons as conjectures,

1178: as it would require a more precise definition of what

1179: constitutes a random polygon, and more careful probabilistic

1180: analyses, to establish them formally. 

1181: 

1182: \subsection{Lower Bound: Exponential Number of Gluing Trees}

1183: In contrast to the likely paucity of foldable polygons, some

1184: polygons generate many foldings.

1185: 

1186: \begin{theorem}

1187: For any even $n$, there is a polygon $P$ of $n$ vertices

1188: that has $2^{\Omega(n)}$ combinatorially distinct

1189: Aleksandrov gluings.

1190: \theolab{star}

1191: \end{theorem}

1192: \begin{pf}

1193: The polygon $P$ is illustrated in Fig.~\figref{star}(a).

1194: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1195: \begin{figure}[htbp]

1196: \centering

1197: \includegraphics[width=\linewidth]{star.eps}

1198: \caption{(a) Star polygon $P$, $m=16$, $n'=32$, $n=34$.

1199: (b) Base gluing tree.

1200: (c) A gluing tree after several contractions.}

1201: \figlab{star}

1202: \end{figure}

1203: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1204: It is a centrally symmetric star, with $m$ vertices, $m$ even,

1205: with a small convex angle $\a \approx 0$, 

1206: alternating with $m$ vertices

1207: with large reflex angle $\b < 2\pi$.

1208: All edges have the same (say, unit) length.

1209: We call this an $m$-star.

1210: We first specify the constraints on $\a$ and $\b$.

1211: 

1212: $P$ has $n'=2m$ vertices (ignoring $x$ and $y$, to be described shortly).

1213: So $m(\a+\b) = (n'-2)\pi$, which implies that 

1214: \begin{equation}

1215: \a + \b = (1 - \frac{1}{m}) 2 \pi \;. \eqlab{a+b}

1216: \end{equation}

1217: We choose $\a$ small enough so that $m$ copies of $\a$ can

1218: join with one of $\b$ and still be less than $2\pi$:

1219: 

1220: \begin{equation}

1221: m \a + \b < 2 \pi  \;. \eqlab{ma}

1222: \end{equation}

1223: Substituting this relationship into Eq.~\eqref{a+b} and solving

1224: for $\a$ yields:

1225: \begin{equation}

1226: \a < \frac{2 \pi}{ m(m-1) } \;. \eqlab{a}

1227: \end{equation}

1228: 

1229: Now we add two vertices $x$ and $y$ at the midpoints of edges,

1230: symmetrically placed so that $y$ is half the perimeter around

1231: $\bP$ from $x$.  Let $n=n'+2$ be the total number of vertices of $P$.

1232: 

1233: 

1234: The ``base'' gluing tree is illustrated in 

1235: Fig.~\figref{star}(b).

1236: $x$ and $y$ are fold vertices of the gluing.

1237: Otherwise, each $\a$ is matched with a $\b$.

1238: Because all edge lengths are the same, and because 

1239: $\a+\b < 2\pi$ by Eq.~\eqref{a+b},

1240: this path is an Aleksandrov gluing.

1241: We label it $T_{00\cdots0,00\cdots0}$,

1242: where $m/2$ zeros $00\cdots0$ represent the top chain, and another $m/2$

1243: zeros represent the bottom chain.

1244: 

1245: The other gluing trees are obtained via ``contractions''

1246: of the base tree.

1247: A {\em contraction\/} makes any particular $\b$-vertex 

1248: not adjacent to $x$ or $y$

1249: a leaf of the tree by gluing its two adjacent $\a$-vertices

1250: together.  Label a $\b$-vertex $0$ or $1$ depending

1251: on whether it is uncontracted or contracted

1252: respectively.  Then a series of contractions can be identified

1253: with a binary string.

1254: For example, Fig.~\figref{star}(c) displays the

1255: tree $T_{010100\cdots,00110\cdots0}$.

1256: Note that $k$ adjacent contractions result in $2k$ 

1257: $\a$-vertices glued together.

1258: 

1259: We now claim that if the number of contractions in the top chain

1260: is the same as the number in the bottom chain

1261: (call such a series of contractions {\em balanced\/}),

1262: the resulting

1263: tree represents an Aleksandrov gluing.

1264: Fix the position of $x$ to the left,

1265: and contract leftwards, as in  Fig.~\figref{star}(c).

1266: Then it is evident that the alternating ``parity'' pattern of

1267: $\a$'s and $\b$'s is not changed by contractions.

1268: Ignoring the arcs attached to the central path,

1269: each contraction replaces $\a \rightarrow 2\a$,

1270: and shortens the path by $2$ units.

1271: Because the contraction shortens by an even number of units,

1272: it does not affect the parity pattern.

1273: If the top and bottom chains are contracted the same number of

1274: times (twice each in (c) of the figure), then their lengths

1275: are the same.

1276: 

1277: Thus after a balanced series of contractions, we have a number

1278: of $\b$-leaves, and gluings of $2k$ $\a$-vertices to one $\b$-vertex.

1279: The $\b$-leaves are legal gluings because $\b < 2\pi$.

1280: Because there are $m/2-1$ contractible $\b$-vertices in each chain, the longest

1281: series of adjacent contractions is $m/2-1$.  So $k \le m/2 - 1$, and

1282: $2k < m$.  Eq.~\eqref{ma} then shows that each gluing produces

1283: less than $2\pi$ angle, and so is Aleksandrov.

1284: 

1285: Finally, we bound the number of gluings.

1286: There are $2^{m/2-1}$ binary numbers of $m/2-1$ bits.

1287: Thus there are this many ways to contract the top chain.

1288: The bottom chain must be contracted with the same number

1289: of $1$'s for a balanced series.  Rather than count this

1290: explicitly, we simply note that $P$ has

1291: at least  $2^{m/2-1}$ 

1292: Aleksandrov gluings, and

1293: because $P$ has $n=n'+2=2m+2$ vertices,

1294: $\Omega(2^{m/2-1}) = \Omega(2^{(n-6)/4}) = 2^{\Omega(n)}$.

1295: \end{pf}

1296: 

1297: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1298: \begin{figure}[htbp]

1299: \centering

1300: \includegraphics[width=0.75\linewidth,angle=-90]{4star_gluings.ps}

1301: \caption{Six gluing patterns for a $4$-star.}

1302: \figlab{4star_gluings}

1303: \end{figure}

1304: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1305: Fig.~\figref{4star_gluings} shows six gluings

1306: of a $4$-star.

1307: The first two in the top row correspond to the

1308: perimeter-halving construction used in the proof.

1309: By Aleksandrov's theorem, each corresponds to a

1310: unique polytope, but

1311: as mentioned in Section~\secref{Aleksandrov},

1312: we do not know how to compute the 3D structure

1313: of these polytopes.

1314: Nevertheless, our hand-exploration suggest that all

1315: fold to noncongruent polytopes, each with the combinatorial

1316: structure of the regular octahedron.

1317: Two of our conjectured crease patterns are shown in 

1318: Fig.~\figref{crease}.

1319: 

1320: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1321: \begin{figure}[htbp]

1322: \centering

1323: \centering

1324: \includegraphics[width=0.45\linewidth]{4star_zag.eps}

1325: \hspace{5mm}%

1326: \includegraphics[width=0.45\linewidth]{4star_zip.eps}

1327: \caption{Conjectured crease patterns for the first

1328: two gluing patterns in the top row of

1329: Fig.~\figref{4star_gluings}.

1330: [Constructions performed in Cinderella.]}

1331: \figlab{crease}

1332: \end{figure}

1333: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1334: 

1335: \subsection{Upper Bound: Few Leaves}

1336: Our goal is now to provide upper bounds on 

1337: the number of gluings, both for arbitary polygons and for

1338: convex polygons.  Both will rely on upper bounds for

1339: gluing trees with a small number of leaves.

1340: Let a gluing tree $T_G$ have $\l$ leaves.

1341: In this section, we prove results for $\l=2$ and $\l=3$.

1342: We then use these to obtain a general upper bound

1343: in Section~\secref{Gluing.Upper.General}, and a bound for convex polygons

1344: in Section~\secref{Gluing.Upper.Convex}.

1345: In between, we summarize the structural properties of

1346: gluing trees in Section~\secref{Gluing.char}.

1347: 

1348: It will sometimes be easier to work with ``gluing instructions''

1349: rather than with gluing trees.

1350: Toward that end, we define

1351: the combinatorial

1352: type of a gluing.

1353: Again let polgyon $P$ have vertices $v_i$ and edges $e_i$,

1354: labeled counterclockwise,

1355: The {\em combinatorial type\/} 

1356: $\G_G$ of a gluing $G$ specifies

1357: to which vertex or edge of $P$ each vertex of $P$ glues

1358: via a set of ordered pairs:

1359: $\G_G = \{ (v_i, z_j) \}$, where $z_j$ is either $v_j$ or $e_j$,

1360: the first element $j$ to which $v_i$ glues counterclockwise

1361: around $\bP$.  If $v_i$ is a leaf of the cut tree, then

1362: the pair $(v_i,v_i)$ is included; otherwise $v_i$ must glue to

1363: an element different from itself.

1364: For example, 

1365: the combinatorial type of the gluing illustrated earlier in

1366: Fig.~\figref{tri.tetra}a is

1367: $$

1368: \{

1369: (v_1,v_3),

1370: (v_2,v_2),

1371: (v_3,e_3)

1372: \}

1373: $$

1374: 

1375: We now prove that the combinatorial type of a gluing

1376: determines the gluing tree.

1377: 

1378: \begin{lemma}

1379: The combinatorial type $\G_G$ of a gluing 

1380: $G$ determines the gluing tree $T_G$.

1381: \lemlab{gluing.tree}

1382: \end{lemma}

1383: \begin{pf}

1384: A node of degree~$2$ of $T_G$ is directly labeled in $\G_G$

1385: as either $(v_i,v_j)$ or $(v_i,e_j)$.

1386: It is only nodes of degree~$\neq 2$ for which $T_G$ contains information

1387: not immediately supplied by $\G_G$.

1388: Nodes of degree~$1$ (leaves) of $T_G$ 

1389: correspond to two possible types of gluings:

1390: either $(v_i,v_i)$, which are directly labeled in $\G_G$, or

1391: fold vertices, a vertex produced by folding 

1392: at a point $x$ in the interior of an edge $e_j$.

1393: (Cf.~Fig.~\figref{perim.halving} for an example of fold vertices.)

1394: Fold vertices can be identified in $\G_G$ as gluings of $v_i$ to

1395: either $e_i$ or $e_{i-1}$: gluing to an incident edge necessarily

1396: implies a fold vertex on that edge.

1397: Or $v_i$ can be glued

1398: to the next vertex, folding the edge in half.

1399: In Fig.~\figref{tri.tetra}, the pair $(v_3,e_3)$ identifies

1400: fold vertex $x$ as labeled with $e_3$;

1401: that $v_1$ also glues to incident edge $e_3$ is known after the

1402: degree~$3$ node's labels are determined.

1403: 

1404: Nodes of degree~$d>2$ in $T_G$ have $d$ labels.

1405: Because every such node can involve at most one edge

1406: (because two edges glued to a point already gives an angle of $2\pi$

1407: there, and the other

1408: elements glued to the same point would cause the angle sum to exceed this),

1409: the labels can be gathered by following the gluings counterclockwise:

1410: $$(v_{i_1},v_{i_2}), (v_{i_2},v_{i_3}), \ldots, 

1411: (v_{i_{d{-}2}},v_{i_{d{-}1}}), (v_{i_{d{-}1}},e_j) \,.$$

1412: In Fig.~\figref{tri.tetra}, the node at point $z$ has labels

1413: $\{v_1,v_3,e_3\}$, which can be identified from

1414: the pairs

1415: $(v_1,v_3),(v_3,e_3)$ of $\G_G$.

1416: \end{pf}

1417: 

1418: \noindent

1419: This lemma permits us to count gluing trees by counting combinatorial

1420: types of gluings.

1421: 

1422: \begin{lemma}

1423: A polygon $P$ of $n$ vertices has

1424: $\Theta(n^2)$ different gluing trees of two leaves,

1425: i.e., paths.

1426: \lemlab{count.path}

1427: \end{lemma}

1428: \begin{pf}

1429: View $\bP$ as rolling continuously between the two leaves $x$ and $y$,

1430: like a conveyor belt or tank tread.

1431: Each specific position corresponds to a perimeter-halving

1432: gluing $G$ (Fig.~\figref{perim.halving}).

1433: The combinatorial type $\G_G$ changes each time a vertex $v_i$

1434: either passes another vertex $v_j$, or becomes the leaf $x$ or $y$.

1435: Each such event corresponds to two distinct types: the type at the

1436: event, and the type just beyond it:  e.g., $(v_i,v_j)$ and $(v_i,e_j)$.

1437: So counting events undercounts by half.

1438: If we count the possible pairs $(v_i,v_j)$ 

1439: for all $i \neq j$, 

1440: we will double count each type:

1441: the event $(v_i,v_j)$ leads to the same type as

1442: $(v_j,v_i)$.

1443: The undercount by half and overcount by double cancel;

1444: thus $n(n-1)$ is the number of types without a vertex at a leaf.

1445: Adding in the $n$ possible $(v_i,v_i)$ events, each of which

1446: leads to two types, yields

1447: an upper bound of $n(n-1) + 2n = O(n^2)$ on the number of

1448: combinatorial types.

1449: 

1450: A lower bound of $\Omega(n^2)$ is achieved by the

1451: example illustrated in 

1452: Fig.~\figref{lower.bounds}(a).

1453: Here $n/2$ vertices of $P$ are closely spaced within a

1454: length $L$ of $\bP$, and $n/2$ vertices are spread out by more

1455: than $L$ between each adjacent pair. Then each of the latter

1456: vertices (on the lower belt in the figure) can be placed between

1457: each pair of the former vertices (on the upper belt), yielding

1458: $n^2/4$ distinct types.  This example can be realized geometrically

1459: by making the internal angle at each vertex nearly $\pi$,

1460: i.e., by a convex polyon that approximates a circle.

1461: 

1462: Lemma~\lemref{gluing.tree} shows that the bound just

1463: obtained of $\Theta(n^2)$ on the number of combinatorial types

1464: applies as well to the number of gluing trees.

1465: \end{pf}

1466: 

1467: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1468: \begin{figure}[htbp]

1469: \centering

1470: \includegraphics[width=8cm]{lower.bounds.eps}

1471: %\fbox{figure=lower.bound.eps}

1472: \caption{(a) $\Omega(n^2)$ combinatorial types can be achieved

1473: by rolling this perimeter ``belt'';

1474: (b) There are only $O(n)$ possible disjoint vv-pairings.}

1475: \figlab{lower.bounds}

1476: \end{figure}

1477: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1478: 

1479: \begin{lemma}

1480: A polygon $P$ of $n$ vertices folds to at most

1481: $O(n^4)$ different gluing trees of three leaves,

1482: i.e., `{\tt Y}'s.

1483: \lemlab{count.Y}

1484: \end{lemma}

1485: \begin{pf}

1486: Observe that the degree-$3$ node of the `{\tt Y}' is either

1487: comprised by the gluing two vertices and an edge together

1488: (call this {\em type-vve}),

1489: or three vertices ({\em type-vvv}).

1490: It is not possible to glue two or more edges together without

1491: violating the $\le 2\pi$ angle restriction of an Aleksandrov

1492: gluing.

1493: 

1494: There are $O(n^3)$ possible type-vvv nodes for the `{\tt Y}'.

1495: Once this type of node is specified, the entire gluing tree is determined,

1496: so this bounds the number of `{\tt Y}'s with type-vvv nodes.

1497: Now consider type-vve nodes.

1498: There are $O(n^2)$ possible vv-gluings, which determine

1499: one branch of the `{\tt Y}'.  The remainder of the `Y' can be

1500: viewed as a path between its leaves; essentially this view

1501: corresponds to a conveyor belt with an appendage.

1502: Applying Lemma~\lemref{count.path} yields a bound of $O(n^4)$.

1503: 

1504: \end{pf}

1505: 

1506: We leave open the question of whether this bound is tight.

1507: We will improve it for convex polygons in Section~\secref{Gluing.Upper.Convex}.

1508: 

1509: \subsubsection{Four Fold-Point Gluing Trees}

1510: We now embark on a study of a special case that will play

1511: two roles: in the proof of our main combinatorial

1512: upper bound, Theorem~\theoref{gluing.upper},

1513: and in counting noncongruent polytopes in Section~\secref{Noncongruent.Polytopes}.

1514: Define a {\em four fold-point gluing tree\/} to be a gluing

1515: tree with (at least) four leaves, each fold points, i.e., creases in the

1516: interior of polygon edges leading to polytope vertices of

1517: curvature $\pi$.

1518: We have already encountered one such tree in Fig.~\figref{zed.tetra2}(b).

1519: We start with this straightforward lemma.

1520: 

1521: \begin{lemma}

1522: A four fold-point gluing tree must have exactly four leaves,

1523: and so have

1524: combinatorial structure

1525: `{\tt +}' or `{\tt I}'.

1526: \lemlab{four.4}

1527: \end{lemma}

1528: \begin{pf}

1529: Because each fold point leads to a

1530: vertex of the resulting polytope $Q$ which

1531: has curvature $\pi$, Fact~\factref{4pi}

1532: implies that all the curvature of the polytope is at the

1533: four fold vertices.

1534: Thus all vertices of $P$ must glue to points that have

1535: total angle $2\pi$, so that the curvature there is zero.

1536: 

1537: A leaf of a gluing tree cannot have zero curvature.

1538: This is because a leaf is either a fold point (curvature $\pi$)

1539: or a ``zipped'' polygon vertex $v$.  The only way to achieve zero

1540: curvature at a zipped vertex is to have an internal polygon

1541: angle at $v$ of $2\pi$.  But this violates simplicity of $P$:

1542: all internal angles are strictly less than $2\pi$.

1543: 

1544: Therefore, a four-fold gluing tree must have exactly four leaves.

1545: So there are only two possible combinatorial structures:

1546: `{\tt +}' and `{\tt I}' (as in Lemma~\lemref{cut.tree}).

1547: \end{pf}

1548: 

1549: Before counting the number of gluing trees, we detail one example

1550: that will be the basis for the remainder of our analysis.

1551: Start with an $L \times W$ rectangle $P$, and fold it as follows.

1552: Glue the two opposite edges of length $W$ together to form a

1553: cylinder.  Now glue the bottom rim of the cylinder to itself

1554: by creasing at two diametrically opposed points $x_1$ and $y_1$.  

1555: Similarly

1556: glue the top rim to itself by creasing at two points $x_2$ and $y_2$.

1557: The gluing tree is of structure `{\tt I}': see

1558: Fig.~\figref{I}.

1559: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1560: \begin{figure}[htbp]

1561: \centering

1562: \includegraphics[width=0.3\linewidth]{I.eps}

1563: \caption{`{\tt I}' gluing tree for an $L \times W$ rectangle.}

1564: \figlab{I}

1565: \end{figure}

1566: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1567: It is easy to see this is an Aleksandrov gluing.  Note both internal

1568: nodes of the gluing tree glue two $\pi/2$ rectangle corners

1569: to the interior of an $L$-edge; so the angle sum there is $2\pi$.

1570: The gluing is Aleksandrov

1571: even if the crease

1572: points on the top and bottom are not located at corresponding

1573: points on their rims.  In particular, identify the points $x_i$ with

1574: their distance from the rectangle corner to the left.  If $x_1 = x_2$,

1575: then the crease points correspond, and the gluing produces a

1576: flat, $L/2 \times W$ rectangle.  If $x_1 \neq x_2$, the gluing is

1577: still Aleksandrov, but the ``twist'' in the gluing results in

1578: a nondegenerate tetrahedron, with all vertices of curvature $\pi$.

1579: Let $x = |x_2 - x_1|$ characterize amount of the twist,

1580: with $x=0$ representing no twist.  

1581: 

1582: Because the $1$-skeleton

1583: of a tetrahedron is combinatorially $K_4$, each vertex is

1584: adjacent to all the others via polytope edges.

1585: This makes it trivial to decide the structure of the polytope

1586: $Q_x$ created by this rectangle gluing with twist $x$.

1587: The six distances between pairs of vertices are easily computed

1588: from the gluing,

1589: and each represents an edge length.

1590: These six lengths uniquely determine the 3D shape of the tetrahedron.

1591: It is not difficult to compute 3D vertex coordinates from the

1592: six lengths, and

1593: we have written code for this computation.

1594: An example is shown in

1595: Fig.~\figref{tetra}.

1596: Here a $2 \times 2$ rectangle is folded with a variety of

1597: different twists $x$.

1598: For both $x=0$ and $x=1$, the result is a flat $1 \times 2$

1599: rectangle, with a smooth interpolation between for $0 < x < 1$.

1600: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1601: \begin{figure}[htbp]

1602: \centering

1603: \includegraphics[width=\linewidth]{tetra.eps}

1604: \caption{Tetrahedra formed by folding a rectangle according

1605: to the gluing tree shown in Fig.~\figref{I}.}

1606: \figlab{tetra}

1607: \end{figure}

1608: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1609: 

1610: We have proven this lemma:

1611: \begin{lemma}

1612: Any rectangle may fold via a `{\tt I}' gluing tree

1613: to a uncountably infinite number of noncongruent tetrahedra.

1614: \lemlab{tetra}

1615: \end{lemma}

1616: \begin{pf}

1617: Two tetrahedra with different edges lengths are not congruent.

1618: The edge lengths of $Q_x$ for twist $x$

1619: are $L/2$ (twice), 

1620: $u(x) = \sqrt{x^2 + W^2}$ (twice),

1621: and $v(x) = \sqrt{(1-x)^2 + W^2}$ (twice).

1622: For $a \neq b$, $u(a) \neq u(b)$;

1623: and for $x < L/4$, $u(x) \neq v(x)$.

1624: Thus the number of noncongruent tetrahedra is at least the

1625: number of distinct $x \in [0, L/4)$, which is

1626: nondenumerable.

1627: \end{pf}

1628: 

1629: We return now to the task of upper-bounding the number of

1630: four fold-point gluing trees possible for a polygon of $n$ vertices.

1631: Although we do not at this point have tight bounds, they

1632: suffice for our purposes in the next section.

1633: 

1634: Define a {\em conveyor belt\/} (or just {\em belt\/})~\label{rolling.belt}

1635: in a gluing tree to be a path between

1636: two leaf fold points.

1637: Let a belt have fold points $x$ and $y$, with $x$

1638: an interior point of edge $e$.

1639: A belt can {\em roll\/} if there is a nonzero-length

1640: interval $I \subset e$ such that for every

1641: $x \in I$, the belt folded at $x$ is an Aleksandrov

1642: gluing.

1643: A belt could instead have a finite number of distinct

1644: gluings, perhaps just one.

1645: We first show that rolling belts must be vertex-free in

1646: four fold-point trees.

1647: 

1648: \begin{lemma}

1649: A rolling belt in a four fold-point tree $T$ cannot contain any

1650: vertices except those at the attachment points to other

1651: branches of $T$.

1652: \lemlab{four.belt}

1653: \end{lemma}

1654: \begin{pf}

1655: Suppose to the contrary that a rolling belt

1656: contains at least one vertex $v$ in its interior,

1657: i.e., not at an attachment point.

1658: Because under our definition, all vertices of $P$ are essential,

1659: the internal angle at $v$ is different from $\pi$.

1660: Let $x \in I$ be a particular fold point that determines the

1661: gluing of the belt.

1662: In this position, $v$ must match up with another vertex $v'$ with

1663: supplementary angle.

1664: Rolling the belt in a neighborhood of $x$ breaks the match, leaving

1665: both $v$ and $v'$ glued to points internal to an edge.

1666: At these points, the curvature is greater than zero,

1667: violating the fact that all curvature at a four fold-point gluing

1668: are concentrated at the leaves.

1669: \end{pf}

1670: 

1671: Note that the angles at the attachment points must be $\pi$.

1672: 

1673: \begin{lemma}

1674: A belt in a four-fold gluing tree $T$ has at most $O(n)$

1675: combinatorially distinct gluings.

1676: \lemlab{four.n}

1677: \end{lemma}

1678: \begin{pf}

1679: Let $B$ be a belt with attachment points $a$ and $b$.

1680: Note that because each attachment point is an internal node

1681: of $T$, the limited structural possibilities established

1682: in Lemma~\lemref{four.4} allow only one or two attachment points.

1683: Consider two cases:

1684: \begin{enumerate}

1685: \item $B$ can roll.  Then by Lemma~\lemref{four.belt},

1686: $B$ contains no internal vertices.  Thus its only vertices are

1687: at $a$ and $b$. Rolling can produce just two combinatorially distinct

1688: positions of the belt.

1689: \item $B$ cannot roll.  Then $B$ can assume a finite number of

1690: possible positions.  

1691: Define a {\em kink\/} in $B$ to be either a vertex, or an

1692: attachment point at which the angle is different from $\pi$,

1693: composed of two glued vertices.

1694: The kinks must match up in pairs.

1695: Matching one pair forces the remaining matches.  Thus

1696: This can be seen by distributing the kinks around a topological

1697: circle representing $B$.  Once one chord $v_1 v_i$ is drawn in

1698: this circle, all other chords are forced by the pairwise matching

1699: requirement.

1700: Because there are only $m-1 < n$ choices for the mate for $v_1$,

1701: $B$ has only $O(n)$ legal gluings.

1702: \end{enumerate}

1703: \end{pf}

1704: 

1705: \begin{lemma}

1706: The number of four fold-point gluing trees for a polygon

1707: of $n$ vertices is $\Omega(n^2)$ and $O(n^4)$.

1708: \lemlab{4fold.bounds}

1709: \end{lemma}

1710: \begin{pf}

1711: The lower bound is established by a variation on the foldings

1712: of a rectangle to tetrahedra (Fig.~\figref{tetra}).

1713: The idea is to make each of the two conveyor belts 

1714: in a `{\tt I}' structure (Fig.~\figref{I}) realize

1715: $\Omega(n)$ gluings independently.

1716: This can be accomplished by alternating supplementary

1717: angles along the belt at equal intervals.

1718: This is illustrated in Fig.~\figref{zigzag} with angles

1719: $\pi/2$ and $3\pi/2$.  The figure illustrates one possible

1720: folding; the fold points are midpoints of edges.

1721: The tetrahedra produced are the same as that obtained

1722: by folding a rectangle: the ``teeth'' mesh seamlessly.

1723: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1724: \begin{figure}[htbp]

1725: \centering

1726: \includegraphics[width=0.6\linewidth]{zigzag.eps}

1727: \caption{(a) Polygon $P$; (b) One four fold-point gluing.

1728: The dashed lines indicate tips of front teeth bent over

1729: and glued behind.}

1730: \figlab{zigzag}

1731: \end{figure}

1732: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1733: 

1734: 

1735: For the upper bound, Lemma~\lemref{four.4} restricts the 

1736: structures to `{\tt +}' and `{\tt I}'.

1737: \begin{enumerate}

1738: \item `{\tt +}'.

1739: Here we rely on the crude $O(n^4)$ bound determined by the

1740: four vertices, or three vertices and one edge, glued

1741: together to form the interior node of $T_G$.

1742: This fixes the combinatorial type of the gluing, which by

1743: Lemma~\lemref{gluing.tree} determines $T_G$.

1744: \item `{\tt I}'.

1745: Let $a$ and $b$ be the upper and lower nodes of the `{\tt I}'.

1746: There are two cases to consider for the upper node:

1747: \begin{enumerate}

1748: \item $a$ is of type `vv': vertices $v_i$ and $v_j$ glue to

1749: form the belt attachment point.

1750: Then the path from $a$ to $b$ is determined by the requirement that

1751: the curvature must be zero at each point:  the two sides ``zip''

1752: closed from $v_i$/$v_j$ until the first point at which the curvature

1753: is nonzero, which then must be the lower node $b$.

1754: 

1755: \item $a$ is of type `ve': vertex $v_i$ glues to the interior of edge $e_j$

1756: to form the attachment point.  The ``zipping'' down to $b$ is again

1757: determined, but this takes more argument.

1758: Let $v_{i+1}$ and $v_j$ be the two vertices closest to $a$ on the

1759: path to $b$.  Both of their angles must differ from $\pi$

1760: (because all vertices are essential).

1761: They must glue to one another with an angle sum of $2\pi$

1762: (because the curvature must be zero).  

1763: We want to show that $v_i$ cannot ``slide'' along $e_j$

1764: to another position and still result in an Aleksandrov gluing.

1765: Sliding $v_i$ up

1766: $e_j$ places  $v_{i+1}$ in the interior of $e_j$, and sliding

1767: $v_i$ down places $v_j$ in the interior of $e_i$, in both

1768: cases producing a point of nonzero curvature.  Therefore no sliding

1769: is possible.  Because any respositioning of $v_i$ on $e_j$

1770: can be viewed as such a sliding, no repositioning is possible.

1771: \end{enumerate}

1772: \end{enumerate}

1773: In both cases there are at most $O(n^2)$ choices for the 

1774: constituents glued at $a$.  Together with the $O(n^2)$ bound on

1775: the two belts from Lemma~\lemref{four.n}, this establishes

1776: the claimed $O(n^4)$ bound.

1777: \end{pf}

1778: 

1779: It seems likely that this lemma could be strengthened: 

1780: \begin{conj}

1781: The number of four fold-point gluing trees for a polygon

1782: of $n$ vertices is $\Theta(n^2)$.

1783: \end{conj}

1784: 

1785: \subsection{Upper Bound: General Case}

1786: \seclab{Gluing.Upper.General}

1787: We finally are positioned to establish an upper bound on the number

1788: of gluing trees, as a function of the number leaves.

1789: 

1790: \begin{theorem}

1791: The number of gluing trees with $\l$ leaves for a polygon $P$

1792: with $n$ vertices is $O(n^{2\l-2})$.

1793: \theolab{gluing.upper}

1794: \end{theorem}

1795: \begin{pf}

1796: Let $g(n,\l)$ be the number of gluing trees for $P$ that

1797: have $\l$ leaves.

1798: The proof is by induction on $\l$.

1799: We know from Lemma~\lemref{four.4} that at most four leaves

1800: can be fold-points.

1801: We assume for the general step of the induction that $\l > 4$,

1802: and so there is at least one non-fold-point leaf.

1803: The base cases for $\l \le 4$ will be considered later.

1804: 

1805: The bound will use one consequence of the angles or curvature of

1806: a gluing (described in this paragraph), and one consequence of the

1807: matching edge lengths of a gluing (described in the next paragraph).

1808: Because a point interior to an edge of $P$ has angle $\pi$, 

1809: a node of degree $d$

1810: of a gluing tree ($d=1, 2, \ldots$) glues together

1811: $d$ vertices of $P$ or $d-1$ vertices and one edge of $P$.

1812: Apart from this, we will use nothing else about the angles of the polygon,

1813: and in fact, our argument will hold more generally for a closed chain

1814: of $n$ vertices, with specified edge lengths.

1815: 

1816: Given a tree $T_G$ that is not a path, and a leaf $l$, 

1817: define the {\em source\/} of

1818: $l$ as the first node of degree more than 2 along the (unique) path

1819: from $l$ into $T$.  The path in $T_G$ from $l$ to its source is called

1820: the {\em branch\/} of $l$.

1821: For a tree $T_G$ and a non-fold-point leaf corresponding

1822: to polygon vertex $l$, let $s(l)$ be a vertex of $P$ closest to $l$

1823: glued at the source of the leaf.  Note that there must be such a vertex,

1824: since we cannot glue together two points interior to polygon edges at the

1825: source of the leaf.

1826: For example, in Fig.~\figref{tri.tetra}, $s(v_2)$ can be $v_3$ or $v_1$.

1827: Note---this is the single consequence of matching edge lengths referred to

1828: above---that the pair $(l, s(l))$ determines

1829: the portion of $P$'s boundary that is glued together to form the

1830: branch of $l$.

1831: We can simplify $T$ by cutting off $l$'s branch,

1832: resulting in a tree with $\l-1$ leaves.  The corresponding simplification of

1833: $\bP$ is to excise the portion of its chain of length $2d(l, s(l))$

1834: centered at $l$, resulting in a closed chain on at most $n-1$

1835: vertices. Since there are $n$ choices for $l$ and at most

1836: $n$ choices for $s(l)$ we obtain $g(n,\l) \le n^2 g(n-1, \l-1)$.

1837: For the general case there are at most 3 fold leaves, hence:

1838: $g(n,\l) \le n^{2(\l-3)} g(n-(\l-3), 3)$.

1839: 

1840: Lemmas~\lemref{count.path} and~\lemref{count.Y} established 

1841: the base cases $g(n,2) = O(n^2)$ and $g(n,3) = O(n^4)$.

1842: Substituting, this yields

1843: \begin{eqnarray}

1844: g(n,\l) \le & n^{2(\l-3)} O([n-(\l-3)]^4) \\

1845:          =  & n^{2(\l-3)} O(n^4) \\

1846:          =  & O(n^{2\l-2})

1847: \end{eqnarray}

1848: 

1849: It remains to handle the case of $\l=4$ leaves.

1850: We will separate into the cases when at least one of these leaves is not

1851: a fold-point leaf, where arguments as above yield $O(n^6)$,

1852: and the case when all 4 vertices are fold leaves.

1853: In this case, Lemma~\lemref{4fold.bounds} establishes

1854: a bound of $O(n^4)$, smaller than that claimed by the lemma.

1855: \end{pf}

1856: 

1857: Of course because $\l$ could be $\Omega(n)$, there is no contradiction

1858: between this upper bound and the exponential lower bound

1859: in Theorem~\theoref{star}.

1860: We specialize the upper bound to convex polygons in Section~\secref{Gluing.Upper.Convex},

1861: but first we summarize the structural characteristics of

1862: gluing trees we have uncovered.

1863: 

1864: \subsection{Gluing Tree Characterization}

1865: Our previous results imply that gluing trees are fundamentally

1866: discrete structures, with one or two rolling conveyor belts,

1867: and two such belts only in very special circumstances.

1868: \seclab{Gluing.char}

1869: \begin{theorem}

1870: Gluing trees satisfy these properties:

1871: \begin{enumerate}

1872: \item At any gluing tree point of degree $d \neq 2$,

1873: at most one point of $\bP$ in the interior of an edge may

1874: be glued, i.e., at most one nonvertex may be glued there.

1875: \item At most four leaves of the gluing tree can be fold points,

1876: i.e., points in the interior of an edge of $\bP$.

1877: The case of four fold-point leaves is only possible when

1878: the tree has exactly four leaves, with the combinatorial

1879: structure `{\tt +}' or `{\tt I}'.

1880: \item A gluing tree can have at most two rolling belts.

1881: \item A gluing tree with two rolling conveyor belts must

1882: have the structure `{\tt I}', and result from folding

1883: a polygon that can be viewed as a quadrilateral with two of

1884: its opposite edges replaced by complimentary polygonal paths.

1885: \end{enumerate}

1886: \theolab{gluing.char}

1887: \end{theorem}

1888: \begin{pf}

1889: \begin{enumerate}

1890: \item That $d \neq 2$ points of a gluing tree have at most

1891: one edge-interior points glued is immediate from the

1892: definition of an Aleksandrov gluing, and our insistence

1893: that all vertices are essential.

1894: \item The structure of four fold-point trees was

1895: established in Lemma~\lemref{four.4}.

1896: \item The definition of ``rolling belts'' (p.~\pageref{rolling.belt})

1897: implies four fold points, so the constraints from the previous

1898: item apply.

1899: \item Two rolling belts cannot be accommodated by the `{\tt +}'

1900: structure, which is determined by the four vertices glued

1901: at the central node.  So the tree structure must be `{\tt I}'.

1902: Lemma~\lemref{four.belt} established that the belts are

1903: vertex-free, corresponding to two opposite edges of the

1904: quadrilateral.  The central path of the `{\tt I}'

1905: must be formed by gluing vertices together whose angle sum

1906: is $2\pi$, and they are in this sense complimentary polygonal paths.

1907: \end{enumerate}

1908: \end{pf}

1909: 

1910: \noindent

1911: Thus a generic gluing tree has one rolling  belt, with

1912: trees hanging off it, 

1913: and one of those trees having a fold-point leaf.

1914: See Fig.~\figref{generic}.

1915: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1916: \begin{figure}[htbp]

1917: \centering

1918: \includegraphics[width=0.6\linewidth]{generic.eps}

1919: \caption{A generic gluing tree: three fold-point leaves

1920: (indicated by smooth arcs),

1921: two forming a rolling belt.  Vertices indicated by open circles. }

1922: \figlab{generic}

1923: \end{figure}

1924: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1925: 

1926: 

1927: 

1928: 

1929: \subsection{Upper Bound: Convex Polygons}

1930: \seclab{Gluing.Upper.Convex}

1931: For convex polygons we can prove a polynomial upper bound.

1932: We first handle the special case of $\l=4$.

1933: 

1934: \begin{lemma}

1935: A convex polygon $P$ may fold to gluing trees of four

1936: leaves only if it is a quadrilateral, a pentagon, or a hexagon;

1937: $P$ may fold to $O(1)$ such gluing trees.

1938: \lemlab{count.+I}

1939: \end{lemma}

1940: \begin{pf}

1941: As in the proof of Theorem~\theoref{cut.comb}, the two conditions

1942: $\g(v) \ge \pi$ and $\sum_v \g(v) = 4\pi$ for the four leaves

1943: $v$ of the tree imply that $\g(v) = \pi$ for each.

1944: This implies that the internal angle at $v$ in $P$ is $\pi$,

1945: which, by our assumption that all vertices are essential,

1946: implies that all four are fold vertices.

1947: 

1948: Because all available curvature is consumed by the four leaves,

1949: the internal nodes of the gluing tree must be flat.  

1950: If the $T$ has shape `{\tt +}', four vertices

1951: whose angles sum to $2\pi$ join there.

1952: Recalling that the turn angle at each vertex

1953: is $\tau_i = \pi - \a_i$ and that the total turn angle is $2\pi$,

1954: this angle sum implies that $\sum_i \tau_i = 4\pi - 2\pi = 2\pi$,

1955: for the four vertices at the `{\tt +}',

1956: and so the turn angle is completely consumed by these four vertices.

1957: Thus $P$ must be a quadrilateral, and there is just one way

1958: to form the gluing tree.

1959: 

1960: If $T$ is a `{\tt I}' shape, then 

1961: each of the two internal nodes of the `{\tt I}' are

1962: formed either by gluing together three vertices, or

1963: two vertices and an edge.

1964: For a three-vertex node, the turn angle sum

1965: is $3\pi - 2\pi = \pi$;

1966: for a two-vertex and edge node, the turn angle sum

1967: is $2\pi - \pi = \pi$.

1968: So both nodes together consume of all the $2\pi$ turn angle.

1969: Therefore $P$ has at most six vertices.

1970: The hexagon permits the most groupings of vertices, six;

1971: and so there are at most six gluing trees.

1972: \end{pf}

1973: 

1974: \noindent

1975: See Fig.~\figref{hex.YY} for an irregular hexagon that folds with

1976: a `{\tt I}' gluing tree.

1977: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1978: \begin{figure}[htbp]

1979: \centering

1980: \includegraphics[width=8cm]{hex.YY.eps}

1981: \caption{A hexagon that folds by gluing together

1982: vertices $\{v_1,v_2,v_3\}$ to form one node,

1983: and $\{v_4,v_5,v_6\}$ to form the other.  Note

1984: the angle sum at the former is $\pi/2+\pi/2$ from the

1985: right angles at $v_1$ and $v_3$,

1986: %(part of the central rectangle) 

1987: and $\pi$ from $\triangle(v_1,v_2,v_3)$,

1988: for a total of $2\pi$.

1989: The four fold vertices are marked.

1990: }

1991: \figlab{hex.YY}

1992: \end{figure}

1993: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1994: 

1995: \begin{theorem}

1996: A convex polygon $P$ of $n$ vertices folds to at most

1997: $O(n^3)$ different gluing trees.

1998: \theolab{count.gluing.trees}

1999: \end{theorem}

2000: \begin{pf}

2001: Theorem~\theoref{cut.comb} limits the combinatorial possibilities to

2002: trees with four or fewer leaves.

2003: We have settled each case for $\l \le 4$ earlier:

2004: \begin{description}

2005: \item[$\l=4$] Lemma~\lemref{count.+I}: $O(1)$.

2006: \item[$\l=3$] Lemma~\lemref{count.Y}: $O(n^4)$.

2007: \item[$\l=2$] Lemma~\lemref{count.path}: $O(n^2)$.

2008: \end{description}

2009: We now improve the $\l=3$ case to $O(n^3)$ for convex polygons.

2010: We can tighten the $O(n^4)$ bound with the following two observations:

2011: \begin{enumerate}

2012: \item The two internal angles at the two vertices glued at a type-vve

2013: node must sum to no more than $\pi$.

2014: \item A convex polygon cannot have too many vertices with small angles.

2015: \end{enumerate}

2016: To quantify the second observation, define the {\em turn angle\/}

2017: $\tau_i$ at a vertex $v_i$ with internal angle $\a_i$ to be

2018: $\tau_i = \pi - \a_i$.  For any polyon, we must have

2019: $\sum_i = 2 \pi$.  For a convex polygon, $\tau_i > 0$.

2020: Now suppose $v_i$ and $v_j$ glue at a type-vve node.

2021: Then $\a_i+\a_j \le \pi$, and so

2022: $\tau_i+\tau_j \ge \pi$.

2023: Thus two distinct vv-gluings, involving four different vertices,

2024: already consume the available $2\pi$ turn angle of the polygon.

2025: Call two pairs of vertices {\em disjoint\/} if all four vertices

2026: are different.  The turn angle bound implies

2027: that any polygon can have at

2028: most two disjoint pairs of vertices glued to a type-vve node.

2029: We now show that this implies a $O(n)$ bound on the number of

2030: type-vve nodes.

2031: 

2032: Construct a bipartite graph $H$ with $n$ nodes for the first vertex

2033: and $n$ nodes for the second vertex of vv-gluings.

2034: Let $(v_i,v_j)$ and $(v_k,v_l)$ be disjoint pairs in vv-gluings,

2035: as depicted in Fig.~\figref{lower.bounds}(b).

2036: Then because there cannot be another pair disjoint from either of

2037: these, every other pair must be incident to one of the

2038: four vertices $v_i,v_j,v_k,v_l$.  This limits $H$ to at most $4n$

2039: edges (and even this bound is loose, for this permits as many as

2040: four disjoint pairs, as is evident in the figure).

2041: 

2042: Thus there are at most $O(n)$ type-vve nodes.

2043: Repeating the argument that a vv-gluing determines one leg of the `Y'

2044: and Lemma~\lemref{count.path} bounds the remaining path to $O(n^2)$

2045: possibilities, leads to the claimed $O(n^3)$ bound.

2046: \end{pf}

2047: 

2048: \noindent

2049: We leave open the question of whether this bound is tight.

2050: 

2051: It is straightforward to list all possible gluing trees for

2052: a given convex polygon with an $O(n^3 \log n)$ time algorithm.

2053: We have implemented such an algorithm, with, however, less than

2054: maximally efficient data structures.

2055: 

2056: 

2057: 

2058: \section{Counting Foldings: Noncongruent Polytopes}

2059: \seclab{Noncongruent.Polytopes}

2060: We have so far been counting the number of different ways to fold

2061: up a given polygon, but have not addressed the question of

2062: whether all these foldings produce distinct polytopes.

2063: There are several notions of what constitutes distinctness.

2064: One natural definition relies on the combinatorial structure

2065: of the polytope, as explored by

2066: Shephard~\cite{s-cpcn-75}.  We will have little to say

2067: on this topic here.

2068: Instead, we will focus on counting noncongruent polytopes.

2069: 

2070: We have already established in Lemma~\lemref{tetra}

2071: that any rectangle can fold to an uncountably infinite 

2072: number of noncongruent tetrahedra.

2073: We extend this result

2074: in this section to the ``obvious'' fact that 

2075: any convex polygon folds (via perimeter-halving)

2076: to an uncountably infinite number of noncongruent polytopes.

2077: Despite the naturalness of this claim, our inability to determine

2078: the 3D structure of the polytope guaranteed by an Aleksandrov gluing

2079: makes our proof less than satisfactory.

2080: In the absence of any 3D information, we concentrate instead on

2081: the pattern of geodesics between vertices, for of course two

2082: congruent polytopes have the exact same set of geodesics.

2083: 

2084: \begin{lemma}

2085: A polytope $Q$ resulting from a perimeter-halving fold 

2086: of polygon $P$ has

2087: a countable number of geodesics between any pair of vertices.

2088: \lemlab{geodesics.countable}

2089: \end{lemma}

2090: \begin{pf}

2091: Let $x$ and $y$ be the fold vertices produced by the

2092: perimeter-halving (as in Fig.~\figref{perim.halving}).

2093: We will assign each geodesic a unique integer,

2094: which establishes that there are only a countable number of

2095: them.

2096: The integers are based on a ``layout'' of the

2097: surface of $Q$ in the plane.

2098: Fix $P$ in the plane, and designate it as level-$0$ of the layout.

2099: Around $\bP$ layout $2n$ copies of $P$ (where $P$ has $n$ vertices)

2100: corresponding to the perimeter gluing.

2101: These are level-$1$ $P$ copies of the layout.

2102: This level is illustrated in Fig.~\figref{pent.layout}.

2103: For example, because edge $e_4 = v_4 v_5$ of $\bP$ is glued into

2104: the edge $e_1 = v_1 v_2$ by the perimeter halving, 

2105: a level-$1$ copy of $P$ is placed 

2106: exterior to $e_4$ arranged so that the glued portions of $e_4$ and

2107: $e_1$ match.

2108: There are $2n$ level-$1$ copies of $P$ because the $n$ vertices

2109: around $\bP$ are interspersed by a reversed sequence of the same

2110: $n$ vertices.

2111: 

2112: Continuing the construction, level-$i$ of the layout is formed by

2113: surrounding each level-$(i{-}1)$ copy of $P$ with $2n$ additional

2114: copies.  Give these copies a ``sequence number'' $j=1,\dots,2n$.

2115: Now every copy of $P$ at level-$i$ in the layout may be assigned a unique

2116: integer by listing the sequence numbers for each level $0,\ldots,i$

2117: and interpreting it as a base-$2n$ number.

2118: 

2119: It is clear from the layout construction that any geodesic on $Q$

2120: ``unrolls'' to a straightline in the layout.

2121: Because we can number the copies of $P$, we can number the geodesics

2122: between any given pair of vertices.

2123: Therefore the number of geodesics is denumerable.

2124: \end{pf}

2125: 

2126: \noindent

2127: Although this proof is specialized to polytopes formed from perimeter

2128: halving, it would not be difficult to extend it to all polytopes 

2129: formed by gluings including a ``rolling'' fold-point.

2130: 

2131: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

2132: \begin{figure}[htbp]

2133: \centering

2134: \includegraphics[width=\linewidth]{pent.layout.eps}

2135: \caption{Layout of a perimeter-halving folding of a pentagon.

2136: Several geodesics are shown from $x$ to level-$1$ vertices.}

2137: \figlab{pent.layout}

2138: \end{figure}

2139: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

2140: 

2141: \begin{theorem}

2142: Any convex polygon $P$ folds, via perimeter halving,

2143: to a uncountably infinite number of

2144: noncongruent polytopes.

2145: \theolab{continuum}

2146: \end{theorem}

2147: \begin{pf}

2148: Select $x$,

2149: a fold point for perimeter halving, 

2150: interior to an edge $e_i = v_i v_{i+1}$ of $P$.

2151: The segment $x v_i \subset \bP$ in level-$0$ of the layout

2152: used in the previous lemma corresponds to a geodesic on $Q_x$.

2153: Now let $x$ vary within some neighborhood $N \subset e_i$;

2154: let $x' \neq x$ be a point in $N$.

2155: The segment $x' v_i$ corresponds to a geodesic on $Q_{x'}$

2156: of a different length.  We use this fact to establish our claim.

2157: 

2158: Let ${\cal Q} = \{Q_{x'} \;:\; x' \in N \}$ be the

2159: set of all the polytopes produced as $x$ varies over the neighborhood.

2160: Assume, for the purposes of contradiction, that the number

2161: of distinct, noncongruent polytopes in $\cal Q$ is denumerable:

2162: $Q_1, Q_2, \ldots$.

2163: By Lemma~\lemref{geodesics.countable}, each has a countable

2164: number of geodesics: a pair of numbers suffice to uniquely identify them.

2165: Thus the total number of distinct lengths of geodesics represented by

2166: all these polytopes is denumerable.

2167: But this contradicts the nondenumerable number of lengths of

2168: segments $|x' v_i|$ for $x' \in N$.

2169: Therefore the number of noncongruent polytopes in $\cal Q$ is

2170: nondenumerable.

2171: \end{pf}

2172: 

2173: Although this theorem establishes the result even for regular

2174: polygons, there is much more to say about the structure of

2175: the polytopes that can be folded from regular polygons.

2176: We explore this in Section~\secref{Regular}.

2177: 

2178: 

2179: \section{Counting Unfoldings: Cut Trees}

2180: \seclab{Unfolding.Cut.Trees}

2181: In this section we explore unfolding from the point of view of

2182: cut trees.  The general situation is that we are given one

2183: polytope $Q$ of $n$ vertices, and we would like to know

2184: how many different ways it can be cut and unfolded to a polygon.

2185: We start with some straightforward observations before

2186: proving enumeration bounds.

2187: 

2188: First, every polytope admits at least the $n$ cut trees

2189: provided by the star unfolding~\cite{ao-nsu-92}, one with each

2190: vertex as source.

2191: So in particular, every polytope unfolds to at least one polygon.

2192: (As we mentioned in the Introduction, the corresponding

2193: question for edge-unfoldings remains open.)

2194: 

2195: Second, because we permit arbitary polygonal paths between the

2196: nodes of a cut tree (Section~\secref{cut.glue.trees}),

2197: there is no upper bound on the number of polygon vertices in

2198: potential unfoldings of a given polytope.  This might lead one

2199: to wonder if \emph{any} polygon (of the appropriate area) 

2200: can be unfolded from

2201: a given polytope.  The answer is {\sc no}, as is easily

2202: established by this lemma.

2203: 

2204: \begin{lemma}

2205: Every polygon $P$ cut from $Q$ must have at least two vertices whose

2206: interior angles are of the form $2\pi - \g_i$ for some $i=1,\ldots,n$, 

2207: where $\g_i$ are the

2208: curvatures of the vertices of $Q$.

2209: \lemlab{not.all.polygons}

2210: \end{lemma}

2211: \begin{pf}

2212: Let the $n$ vertices of $Q$ have curvatures $\g_i$, $i=1,\ldots,n$.

2213: The cut tree $T_C$ must

2214: have at least two leaves,

2215: and by Lemma~\lemref{cut.tree} these leaves must be vertices of $Q$.

2216: Say they coincide with the vertices of

2217: curvatures $\g_1$ and $\g_2$.  

2218: Then any polygon $P$ that unfolds from $T_C$ must have

2219: two vertices with interior angles $2\pi - \g_1$ and $2\pi - \g_2$.

2220: \end{pf}

2221: 

2222: \noindent

2223: So let $P$ be a polygon with no interior angle equal to $2\pi - \g_i$

2224: for $i=1,\ldots,n$.  Then $P$ cannot be cut from $Q$.

2225: 

2226: \subsection{Lower Bound: Exponential Number of Unfoldings}

2227: In this section we provide an exponential lower bound.

2228: 

2229: \begin{theorem}

2230: There is a polytope $Q$ of $n$ vertices

2231: that may be cut open with exponentially many {\rm ($2^{\Omega(n)}$)}

2232: combinatorially distinct

2233: cut trees,

2234: which unfold to exponentially many geometrically distinct simple polygons.

2235: \theolab{volcano}

2236: \end{theorem}

2237: \begin{pf}

2238: $Q$ is a truncated cone, as illustrated in 

2239: Fig.~\figref{volcano.0}:

2240: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

2241: \begin{figure}[htbp]

2242: \centering

2243: \includegraphics[width=0.5\linewidth]{volcano.0.eps}

2244: \caption{

2245: Polytope $Q$.

2246: }

2247: \figlab{volcano.0}

2248: \end{figure}

2249: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

2250: the hull of

2251: two regular $n$-gons of different radii, lying in parallel planes

2252: and similarly oriented.

2253: We call this the {\em volcano\/} example.

2254: We require that $n$ be even; in the figure, $n=16$.

2255: Label the vertices on the top face $a_0,\ldots,a_{n-1}$

2256: and $b_0,\ldots,b_{n-1}$ correspondingly on the bottom face.

2257: The ``base'' cut tree, which we notate as $T_{0000000}$,

2258: unfolds $Q$ as shown in Fig.~\figref{volcano.1}.

2259: $T_{0000000}$ consists of a path on the top face

2260: $(a_0,a_1,\ldots,a_{n-1})$ supplemented by arcs

2261: $(a_i,b_i)$ for all $i=0,\ldots,n-1$.

2262: The polygon $P$ produced consists of the base face,

2263: $n$ attached trapezoids $(b_i,b_{i+1},a_{i+1},a_i)$,

2264: with the top face attached to $a_{n-1} a_0$.

2265: 

2266: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

2267: \begin{figure}[htbp]

2268: \centering

2269: \includegraphics[width=\linewidth]{volcano.1.eps}

2270: \caption{

2271: Unfolding via shaded cut tree $T_{0000000}$.

2272: }

2273: \figlab{volcano.1}

2274: %\end{figure}

2275: \vspace{5mm}

2276: %\begin{figure}[htbp]

2277: \centering

2278: \includegraphics[width=\linewidth]{volcano.2.eps}

2279: \caption{

2280: Unfolding via shaded cut tree $T_{1001101}$.

2281: }

2282: \figlab{volcano.2}

2283: \end{figure}

2284: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

2285: Define a cut tree $T_{m_{(n-1)/2} \cdots m_2 m_1 m_0}$,

2286: where $m_i$ are the digits of a binary number

2287: of $n/2 - 1$ bits,

2288: as an alteration of the base tree $T_{0\cdots0}$ as follows.

2289: If $m_i = 1$, then the arc

2290: $(a_{2i+1},b_{2i+1})$ is deleted and replaced by

2291: $(a_{2i},b_{2i+1})$.

2292: If $m_i = 0$, then the arc $(a_{2i+1},b_{2i+1})$ is used as in  $T_{0\cdots0}$.

2293: Thus the cut tree $T_{1001101}$ shown in Fig.~\figref{volcano.2}

2294: replaces 

2295: $(a_1,b_1)$ with $(a_0,b_1)$ because $m_0 = 1$,

2296: $(a_5,b_5)$ with $(a_4,b_5)$ because $m_2 = 1$,

2297: and so on.

2298: 

2299: There are $2^{n/2 - 1} = 2^{\Omega(n)}$ cut trees.

2300: 

2301: It is clear by construction that all these cut trees lead

2302: to simple polygon unfoldings.

2303: It only remains to argue that each leads to a distinct polygon,

2304: not congruent to any other.  This is not strictly true for $Q$ as

2305: defined, for any bit pattern leads to a $P$ that is congruent

2306: (by reflection) to the polygon obtained from the reverse of the bit

2307: pattern.  However, it is a simple matter

2308: to introduce some asymmetry, by, for example, lengthening edge

2309: $a_{n-1} a_0$ slightly.  Then all cut trees lead to distinct

2310: polygons.

2311: \end{pf}

2312: 

2313: \noindent

2314: A simpler example is a drum, the convex hull of two regular polygons

2315: in parallel planes.  Because some of the unfoldings used in the

2316: above proof overlap, there is a bit more argument needed to establish

2317: the exponential lower bound.

2318: 

2319: 

2320: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

2321: \begin{figure}[htbp]

2322: \centering

2323: \includegraphics[width=\linewidth]{sperm.eps}

2324: %\fbox{figure=sperm.eps}

2325: \caption{

2326: (a) Polytope $Q$,

2327: cut tree $T_{\cdots 0 0 2 2 0 2 0 1 0 0}$.

2328: The $a$-path is shown solid, the $b$-path dashed.

2329: (b) Unfolding to polygon $P$. $L$ strictly separates the head

2330: from the tail.

2331: }

2332: \figlab{sperm}

2333: \end{figure}

2334: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

2335: 

2336: Even restricting the cut tree to a path permits

2337: an exponential number of unfoldings:

2338: 

2339: \begin{theorem}

2340: There is a polytope $Q$ of $n$ vertices

2341: that may be cut open with exponentially many

2342: {\rm ($2^{\Omega(n)}$)}

2343: combinatorially distinct

2344: cut trees, all of which are paths,

2345: which unfold to exponentially many geometrically

2346: distinct simple polygons.

2347: \theolab{sperm}

2348: \end{theorem}

2349: \begin{pf}

2350: $Q$ is formed by pasting two halves of a regular $2n$-gon

2351: together to form a semicircle approximation with some 

2352: small thickness $w > 0$.

2353: Label the vertices on the front face $a_0,\ldots,a_n$

2354: and $b_0,\ldots,b_n$ correspondingly on the back face,

2355: as illustrated in

2356: Fig.~\figref{sperm}(a).

2357: Let $\a = 2\pi/n$ be the turn angle at each vertex $a_i$ (and at $b_i$),

2358: i.e., the angle $\a = \pi - \angle (a_{i-1},a_i,a_{i+1})$.

2359: (In the figure, $\a = \pi/32 \approx 11^\circ$.)

2360: We specify a series of cut trees $T_m$, where $m$ is an

2361: $n$-digit base-$3$ integer $m_n \cdots m_2 m_1$,

2362: with the following interpretation.

2363: $T_{0 \cdots 00}$ is the ``base'' cut tree on which

2364: all others are variations:

2365: \begin{equation}

2366: T_{0 \cdots 00} = (a_0,a_1,\ldots,a_n,b_n,b_{n-1},\ldots,b_1,b_0)

2367: \end{equation}

2368: Note that $T_{0 \cdots 00}$ is a path, as are all the $T_m$.

2369: We call the half of the path on the front face the $a$-path,

2370: and that on the back the $b$-path.

2371: The unfolding $P$ determined by $T_{0 \cdots 00}$ is a regular

2372: $2n$-gon, fattened by a strip $(a_0, b_0, b_n, a_n)$ of width $w$ down

2373: its middle, with a ``tail'' of $n$ rectangles attached to edge $a_0 b_0$.

2374: If $|a_i a_{i+1}| = h$, then each rectangle is $w \times h$.

2375: 

2376: In cut tree $T_{m_n \cdots m_2 m_1}$

2377: the index $m_i$ is $1$ if the $a$-path deviates to touch $b_i$

2378: on the back face

2379: via the path $(\ldots,a_{i-1},b_i,a_i,a_{i+1},\ldots)$,

2380: and the index $m_i$ is $2$ if the $b$-path similarly deviates

2381: to include $a_i$ on the front face

2382: via the path $(\ldots,b_{i-1},a_i,b_i,b_{i+1},\ldots)$.

2383: In both cases, the opposite path skips the vertex deviated to:

2384: if $m_i=1$, the $b$-path skips $b_i$ by shortcutting on the back face, and

2385: if $m_i=2$, the $a$-path skips $a_i$ by shortcutting on the front face.

2386: Fig.~\figref{sperm}(a)

2387: illustrates $T_{m_n \cdots 0 0 2 2 0 2 0 1 0 0}$,

2388: with $a$-path

2389: \begin{equation}

2390: (a_0,a_1,a_2,b_3,a_3,a_4,a_6,a_9,a_{10},\ldots,a_n)

2391: \end{equation}

2392: and $b$-path

2393: \begin{equation}

2394: (b_0,b_1,b_2,b_4,a_5,b_5,b_6,a_7,b_7,a_8,b_8,b_9,b_{10},\ldots,b_n)

2395: \end{equation}

2396: Note that when $m_i \neq 0$, the rectangle bounded between $a_{i-1} b_{i-1}$

2397: and $a_i b_i$ is crossed by an $ab$-diagonal.

2398: We insist that $m_1 = 0$, so that the cut tree starts

2399: with an uncrossed rectangle $(a_0, b_0, b_1, a_1)$.

2400: Finally, the edge $a_n b_n$ is included in $T_m$, so that

2401: it is a path from

2402: $a_0$ to $a_n$ to $b_n$ and returning to $b_0$.

2403: The digits $m_n \cdots m_2$ are each free to

2404: be any one of $\{0,1,2\}$.

2405: Thus there are an exponential number of combinatorially

2406: distinct $T_m$:  $3^{n-1}$.

2407: We return below to the issue of how many of these lead to geometrically

2408: distinct unfoldings.

2409: 

2410: It should be clear by construction that $T_m$ spans the vertices.

2411: To show that it is a tree, we need to argue that it is non-self-intersecting.

2412: This is again clear by construction, for each nonzero $m_i$ uses

2413: a diagonal in the rectangle prior to $a_i b_i$, and because

2414: $m_i$ has only one value, no such rectangle has both diagonals used.

2415: Together with the shortcutting that prevents the $a$- and $b$-paths

2416: from touching the same vertex, it follows that $T_m$ is indeed a tree;

2417: so

2418: it is a legitimate cut tree.

2419: Thus it unfolds to a single piece.

2420: It only remains to show that this unfolding is a simple

2421: polygon, i.e., it avoids overlap.

2422: 

2423: This is obvious

2424: for  $T_{0 \cdots 00}$, as mentioned previously.

2425: For general $T_m$, consider the layout of the unfolding $P$ that

2426: places $a_0 b_0$ horizontally, as in Fig.~\figref{sperm}(b).

2427: Let $L$ be the horizontal line through $a_1 b_1$;

2428: this segment is necessarily horizontal because we stipulated

2429: that  $m_1 = 0$.  We will argue that $L$ strictly separates the 

2430: {\em tail\/} of $P$ (the portion attached above $a_1 b_1$)

2431: from its {\em body\/} (the portion attached below $a_1 b_1$).

2432: 

2433: First, it is clear that the body unfolds without overlap.

2434: For it is simply truncations (due to path shortcuttings) of

2435: halves of a regular polygon glued to either side of the rectangle

2436: $(a_0, b_0, b_n, a_n)$, with attached triangle ``spikes''

2437: for each nonzero $m_i$.  None of these spikes can overlap,

2438: even when adjacent, for their length-$w$ edge juts out

2439: orthogonal to their length-$h$ edge glued to the body

2440: (see the body image of $b_7$ in Fig.~\figref{sperm}(b)).

2441: 

2442: The tail consists of $h \times w$ rectangles, or half-rectangles,

2443: glued end-to-end, with turns to the right by $\a$ for every

2444: $m_i=1$ digit, and turns to the left by $\a$ for every $m_i=2$

2445: digit.  Thus, $T_{\cdots 0 0 2 2 0 2 0 1 0 0}$ in (b)

2446: of the figure turns right once and left three times.

2447: Because there are at most $n-1$ nonzero digits, the tail

2448: can turn at most $n-1$ times.  Because $\a$ is the turn angle

2449: of a regular $2n$-gon, it takes $n$ turns of $\a$ to turn

2450: a full $\pi$.  Thus the tail turns strictly less than $\pi$,

2451: and so cannot return to line $L$.  Thus the tail remains

2452: strictly above $L$.

2453: Choosing $w < h$ guarantees that no body spike protrudes vertically

2454: as much as $h$ above $a_0 b_0$; so the body remains strictly below $L$.

2455: 

2456: It remains to argue that the tail does not self-intersect.

2457: But this follows from the same turn argument above.

2458: By construction, there are no local overlaps between two

2459: adjacent tail rectangles or half-rectangles.

2460: Thus the only overlap conceivable would result from the tail

2461: curling back to overlap itself.

2462: Choosing $w \ll h$ makes the tail essentially a series of segments

2463: of length $h$, with attached pieces of the regular polygon

2464: clipped by shortcutting.

2465: For the tail segments to curl back and overlap

2466: would require a

2467: total turn by at least $\pi$,

2468: contradicting the bound

2469: on the sum of $\a$'s.

2470: 

2471: Finally, we turn to the question of how many of the 

2472: $3^{n-1}$ combinatorially distinct $T_m$ lead to geometrically

2473: distinct (noncongruent) $P$.

2474: Let $x$ and $y$ be two base-$3$ numbers,

2475: and let $S(x)$ be the base-$3$ number obtained by changing each

2476: $1$-digit in $x$ to a $2$, and each $2$-digit in $x$ to a $1$.

2477: (For example, $S(1021) = 2012$.)

2478: Then if $S(x) = y$, $T_x$ and $T_y$ lead to congruent $P$,

2479: in that $P_y$ is the reflection of $P_x$ about a vertical line

2480: (in the layout used above).

2481: 

2482: Although we could easily ensure noncongruency for

2483: all $T_m$ by altering $Q$ to be

2484: less symmetric, we opt here for a counting argument.

2485: Let $x$ be a base-$3$ number.  Define $B(x)$ to be the binary

2486: number obtained by changing each $2$-digit in $x$ to a $1$.

2487: (For example, $B(2012) = 1011$.)

2488: Now it should be clear that for any two

2489: base-$3$ numbers $x$ and $y$, if $B(x) \neq B(y)$, then

2490: $P_x$ is noncongruent to $P_y$.  For then the pattern of spikes

2491: on the body are different in $P_x$ and $P_y$.

2492: Thus, among the $3^{n-1}$ combinatorially distinct $P$,

2493: there are at least $2^{n-1}$ geometrically distinct $P$.

2494: \end{pf}

2495: 

2496: \subsection{Lower Bound: Convex Unfoldings}

2497: It seems possible that the exponential lower bound holds

2498: even in the case of convex unfoldings,

2499: via an example similar to that used in Fig.~\figref{sperm}.

2500: \begin{conj}

2501: There is a polytope with an exponential number of

2502: convex unfoldings.

2503: \end{conj}

2504: 

2505: \noindent

2506: This represents the only `?' in Table~\tabref{results}.

2507: 

2508: 

2509: \subsection{Upper Bound}

2510: 

2511: \begin{theorem}

2512: The maximum number of edge-unfolding cut trees of a polytope of $n$

2513: vertices is $2^{O(n)}$,

2514: and the maximum number of arbitary cut trees

2515: $2^{O(n^2)}$.

2516: \end{theorem}

2517: \begin{pf}

2518: For edge unfoldings, the bound depends on the number

2519: of spanning trees of a polytope graph.

2520: We may obtain a bound here as follows.%

2521: \footnote{

2522: 	We thank B. McKay [personal communication, Jan.~2000] for guidance here.

2523: }

2524: First, 

2525: triangulating a planar graph only increases the number of spanning trees,

2526: so we may restrict attention to

2527: triangulated planar graphs.

2528: Second, it is well known that 

2529: the number of spanning trees of a connected planar graph is the same as

2530: the number of spanning trees of its dual.

2531: %     [This is a fairly well-known.  For example, Exercise 7.1.7 on page 257 of

2532: %      Introduction to Graph Theory by Douglas B. West says: "Prove that a set

2533: %      of edges in a connected plane graph $G$ forms a spanning tree of G if

2534: %      and only if the duals of the remaining edges forma spanning tree of G^*.]

2535: So we focus just on $3$-regular (cubic) planar graphs.

2536: Finally, a result of McKay~\cite{m-strg-83} proves an 

2537: upper bound of $O((16/3)^n / n)$ on the number of spanning trees

2538: for cubic graphs.

2539: This bound is $2^{O(n)}$.

2540: 

2541: For arbitrary cut trees, the underlying graph 

2542: might conceivably have a quadratic number of edges,

2543: which leads to the bound $2^{O(n^2)}$.

2544: (Note that our definition of cut tree in

2545: Section~\secref{Cut.Trees} would not count different polygonal

2546: paths between two vertices as distinct arcs of $T_C$.)

2547: \end{pf}

2548: 

2549: 

2550: \section{Counting Unfoldings: Noncongruent Polygons}

2551: \seclab{Noncongruent.Polygons}

2552: We have already seen in Theorem~\theoref{volcano}

2553: that one polytope can have an exponential number of noncongruent

2554: polygon unfoldings.

2555: In fact the possibilities range from $0$ to $\infty$,

2556: even for convex unfoldings,

2557: as this simple

2558: counterpart of Theorem~\theoref{continuum} shows:

2559: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

2560: \begin{figure}[htbp]

2561: \centering

2562: \includegraphics[width=8cm]{tri.x.eps}

2563: \caption{Unfolding a flat triangle (a) to a convex polygon (b).}

2564: \figlab{tri.x}

2565: \end{figure}

2566: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

2567: 

2568: \begin{theorem}

2569: Although some polytopes

2570: unfold to a nondenumerable number of noncongruent

2571: convex polygons, others have only a finite number of

2572: convex unfoldings.

2573: \end{theorem}

2574: \begin{pf}

2575: For the former claim, consider a doubly-covered equilateral

2576: triangle.  Choose any point $x$ interior to the top face,

2577: as shown in Fig.~\figref{tri.x}(a).  This leads to a `{\tt Y}' cut tree

2578: that unfolds to a convex polygon~(b) for every choice of $x$.

2579: All these polygons have different angles, and so are noncongruent.

2580: 

2581: The second claim of the theorem is trivially satisfied by polytopes

2582: with zero convex unfoldings.

2583: To establish it for a polytope that has at least one convex

2584: unfolding is more difficult, and 

2585: we only sketch a construction.

2586: Consider the doubly-covered trapezoid shown in  Fig.~\figref{trap}.

2587: It has just two sharp vertices, $v_1$ and $v_4$, and so,

2588: by Theorem~\theoref{cut.comb}, the cut tree must be a path

2589: connecting those vertices.  The path $(v_1,v_2,v_3,v_4)$

2590: unfolds to a convex polygon.

2591: Now consider a geodesic that starts with the segment $v_1 v_3$

2592: as illustrated.  As in the proof of Lemma~\lemref{not.two.sharp},

2593: this geodesic will either hit $v_4$ directly, in which case it is not a

2594: valid cut tree because $v_2$ is not spanned, or

2595: it spirals around the trapezoid and self-crosses.

2596: We will not prove this claim.

2597: \end{pf}

2598: 

2599: 

2600: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

2601: \begin{figure}[htbp]

2602: \centering

2603: \includegraphics[width=8cm]{trap.eps}

2604: \caption{A geodesic on a trapezoid from $v_1$ through $v_3$.}

2605: \figlab{trap}

2606: \end{figure}

2607: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

2608: 

2609: \section{Folding Regular Polygons}

2610: \seclab{Regular}

2611: In this section we study folding \emph{regular} polygons

2612: of $n$ vertices.

2613: Because all polygon vertices have the same interior angle $\t = (n-2)\pi/n$,

2614: only a limited variety of different polytope vertex curvatures

2615: may be created.  

2616: We find, not surprisingly, that this leads to a limited

2617: set of possibilities:

2618: in general,

2619: only one ``class'' of nonflat polytopes can be produced.

2620: This is established in Lemma~\lemref{ngt6}.

2621: 

2622: Let $\a_k$, $k \ge 1$, be the curvature at a polytope vertex

2623: formed by gluing $k$ $P$-angles of $\t$ together,

2624: and $\b_k$, $k \ge 0$, be the curvature at a vertex formed

2625: by gluing $k$ angles to a point interior to an edge of $\bP$.

2626: The next lemma details the possible $\a_k$ and $\b_k$ values

2627: achievable.

2628: 

2629: Throughout this section we will find that the situation is more

2630: uniform for $n > 6$ than it is for small $n$.

2631: 

2632: \begin{lemma}

2633: For $n > 6$, only four vertex curvatures can be obtained

2634: by folding a regular $n$-gon $P$; 

2635: for $n \le 6$, additional curvature values are

2636: possible.

2637: More precisely,

2638: for all $n$, these four curvature values are always achievable:

2639: \begin{itemize}

2640: \item $\a_1 = \pi(1+2/n)$.

2641: \item $\a_2 = \pi(4/n)$.

2642: \item $\b_0 = \pi$.

2643: \item $\b_1 = \pi(2/n)$.

2644: \end{itemize}

2645: The additional values possible for $n \le 6$

2646: are detailed in Tables~\tabref{alpha} and ~\tabref{beta}.

2647: \lemlab{curvature.values}

2648: \end{lemma}

2649: \begin{pf}

2650: \begin{enumerate}

2651: \item $\a_1 = 2\pi - \t = 2\pi - (n-2)\pi/n = \pi(1+2/n)$.

2652: This vertex is a leaf of the gluing/cut tree.

2653: We call this a {\em zipped\/} vertex, for $\bP$ is

2654: ``zipped shut'' at the vertex.

2655: \item $\a_2 = 2\pi - 2\t = 2\pi - 2(n-2)\pi/n =\pi(4/n)$.

2656: This vertex is a degree-$1$ node of the gluing tree.

2657: \item $\b_0 = \pi$.

2658: This is a fold vertex, when nothing is glued to an edge of $\bP$,

2659: and therefore a leaf of the gluing tree.

2660: \item $\b_1 = 2\pi - [\pi + \t] = 2\pi - [\pi + (n-2)\pi/n] = \pi(2/n)$.

2661: This is a degree-$1$ node of the gluing tree.

2662: \end{enumerate}

2663: The additional possibilities for $n \le 6$ are as follows.

2664: $\a_3$ is possible for all $n \le 6$;

2665: $\a_4$ is possible only for $n=4$;

2666: and no other $\a_k$ is possible.

2667: See Table~\tabref{alpha}.

2668: 

2669: For $n=3$, $\b_2, \b_3$, and for $n=4$, $\b_2$, are all possible.

2670: See Table~\tabref{beta}.

2671: 

2672: Explicit computation

2673: shows that all higher values of $k$ lead to nonconvex vertices, 

2674: whose total face

2675: angle exceeds $2\pi$ and so which have negative curvature.

2676: \end{pf}

2677: 

2678: \begin{table}[htbp]

2679: \begin{center}

2680: \begin{tabular}{| c | c | c | c | c |}

2681:         \hline

2682: $\a_k$ & \multicolumn{4}{c|}{$k$}

2683:         \\ \hline

2684: \mbox{} & 1 & 2 & 3 & 4

2685:         \\ \hline \hline

2686: \rule[-6pt]{0pt}{17pt}3

2687:         & $\frac{5}{3}$

2688:         & $\frac{4}{3}$

2689:         & $1$

2690:         & \mbox{}

2691:         \\ 

2692: \rule[-6pt]{0pt}{17pt}4

2693:         & $\frac{3}{2}$

2694:         & $1$

2695:         & $\frac{1}{2}$

2696:         & $0$

2697:         \\ 

2698: \rule[-6pt]{0pt}{17pt}5

2699:         & $\frac{7}{5}$

2700:         & $\frac{4}{5}$

2701:         & $\frac{1}{5}$

2702:         & \mbox{}

2703:         \\ 

2704: \rule[-6pt]{0pt}{17pt}6

2705:         & $\frac{4}{3}$

2706:         & $\frac{2}{3}$

2707:         & $0$

2708:         & \mbox{}

2709:         \\ \cline{1-3}

2710: \rule[-6pt]{0pt}{17pt}$n$

2711:         & $1+\frac{2}{n}$

2712:         & $\frac{4}{n}$

2713:         & \multicolumn{2}{c|}{\mbox{}}

2714:         \\ \hline

2715: \end{tabular}

2716: \end{center}

2717: \caption{Possible $\a_k$ curvature values, in units of $\pi$.}

2718: \tablab{alpha}

2719: \end{table}

2720: 

2721: \begin{table}[htbp]

2722: \begin{center}

2723: \begin{tabular}{| c | c | c | c | c |}

2724:         \hline

2725: $\b_k$ & \multicolumn{4}{c|}{$k$}

2726:         \\ \hline

2727: \mbox{} & 0 & 1 & 2 & 3

2728:         \\ \hline \hline

2729: \rule[-6pt]{0pt}{17pt}3

2730:         & $1$

2731:         & $\frac{2}{3}$

2732:         & $\frac{1}{3}$

2733:         & $0$

2734:         \\ 

2735: \rule[-6pt]{0pt}{17pt}4

2736:         & $1$

2737:         & $\frac{1}{2}$

2738:         & $0$

2739:         & \multicolumn{1}{c|}{\mbox{}}

2740:         \\ \cline{1-3}

2741: \rule[-6pt]{0pt}{17pt}$n$

2742:         & $1$

2743:         & $\frac{2}{n}$

2744:         & \multicolumn{2}{c|}{\mbox{}}

2745:         \\ \hline

2746: \end{tabular}

2747: \end{center}

2748: \caption{Possible $\b_k$ curvature values, in units of $\pi$.}

2749: \tablab{beta}

2750: \end{table}

2751: 

2752: Let $a_i$ and $b_i$ be the number of polytope vertices of curvature

2753: $\a_i$ and $\b_i$ respectively, formed by folding

2754: a regular $n$-gon $P$.

2755: Of course $a_i$ and $b_i$ are nonnegative integers,

2756: but there are additional significant restrictions imposed by

2757: the requirement that the total curvature be $4 \pi$:

2758: \begin{equation}

2759: \sum_{i=1} a_i \a_i + \sum_{i=0} b_i \b_i = 4 \pi \;. \eqlab{4pi}

2760: \end{equation}

2761: We now explore the implications of this constraint,

2762: separately for $n > 6$ and for $n \le 6$.

2763: Note that our notation implies that

2764: \begin{equation}

2765: \sum_{i=1} a_i i + \sum_{i=0} b_i i =  n \;, \eqlab{indices}

2766: \end{equation}

2767: because the subscripts on $\a$ and $\b$ indicate the number of

2768: vertices involved in the gluing.

2769: 

2770: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

2771: \begin{figure}[htbp]

2772: \centering

2773: \includegraphics[width=0.5\linewidth]{flat.cases.eps}

2774: \caption{(a,b) Flat foldings of an $n$-gon, $n$ even

2775: ($n=8$).

2776: (c) Flat folding of an $n$-gon, $n$ odd ($n=7$).}

2777: \figlab{flat.cases}

2778: \end{figure}

2779: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

2780: 

2781: 

2782: Now we prove that

2783: perimeter-halving is the only possible kind of folding for $n>6$.

2784: 

2785: \begin{lemma}

2786: For all $n \ge 3$,

2787: regular $n$-gons fold via

2788: perimeter-halving, using path gluing trees,

2789: to two classes of polytopes:

2790: \begin{enumerate}

2791: \item A continuum of ``pita'' polytopes of $n+2$ vertices.

2792: \item One or two flat, ``half-$n$-gons'':

2793: \begin{enumerate}

2794: \item $n$ even: Two flat polytopes, of $\frac{n}{2}+2$ and $\frac{n}{2}+1$ vertices.

2795: \item $n$ odd: One flat polytope, of $\frac{n+1}{2}+1$ vertices.

2796: \end{enumerate}

2797: \end{enumerate}

2798: For $n > 6$, these are the only foldings possible of

2799: a regular $n$-gon.

2800: \lemlab{ngt6}

2801: \end{lemma}

2802: \begin{pf}

2803: A perimeter-halving fold produces a path gluing tree.

2804: This has two leaves and all other nodes internal.

2805: From

2806: Lemma~\lemref{curvature.values},

2807: the only two curvatures can be leaves:

2808: $\{ \a_1, \b_0 \}$;

2809: and only two can be degree-$1$ nodes:

2810: $\{ \a_2, \b_1 \}$.

2811: Moreover, these are the

2812: only curvatures possible

2813: for $n > 6$.  Thus Eq.~\eqref{4pi}

2814: reduces to

2815: \begin{equation}

2816: a_1 \a_1 + a_2 \a_2 + b_0 \b_0 + b_1 \b_1 = 4 \pi \;.

2817: \end{equation}

2818: Substituting the curvature values from 

2819: Lemma~\lemref{curvature.values} and solving for $n$ yields

2820: \begin{equation}

2821: n = \frac{2(a_1 + 2 a_2 + b_1)}

2822:         {4-(a_1 + b_0)} \; \eqlab{4pi.n}

2823: \end{equation}

2824: Because

2825: only $\a_1$ and $\b_0$ are leaf vertex curvatures,

2826: we must have $a_1 + b_0 \ge 2$.

2827: The requirement that the denominator of Eq.~\eqref{4pi.n}

2828: be positive yields  $a_1 + b_0 < 4$.

2829: Therefore we know that  $a_1 + b_0 \in \{2, 3\}$.

2830: We now show that the case $a_1 + b_0 = 3$ is not possible when $n > 6$.

2831: 

2832: As both $a_1$ and $b_0$ count leaves, a tree formed

2833: with  $a_1 + b_0 = 3$ must have at least three leaves.

2834: By Theorem~\theoref{cut.comb}, because $n \neq 4$,

2835: it cannot have more than three leaves.

2836: So it has exactly

2837: three leaves, and has the combinatorial structure of a `{\tt Y}'.

2838: The interior node must be formed by gluing three distinct points of

2839: $\bP$ together (by Lemma~\lemref{cut.tree}(4)).

2840: This corresponds to curvatures

2841: $\a_k$, $k \ge 3$, or $\b_k$, $k \ge 2$.

2842: But Lemma~\lemref{curvature.values} shows that none of these

2843: are possible for $n > 6$.

2844: (Note, for later reference, that for $n \le 6$, these possibilities

2845: will need consideration.)

2846: 

2847: Therefore we must have  $a_1 + b_0 = 2$.

2848: Therefore the gluing tree must be a path for $n > 6$,

2849: and the folding must be a perimeter-halving folding.

2850: We now explore the three possible solutions to $a_1 + b_0 = 2$.

2851: \begin{description}

2852: \item[Case $a_1=0$, $b_0=2$.]

2853: The two leaves are both folds at interior points of edges

2854: of $\bP$, a perimeter-halving folding similar to that

2855: previously illustrated in 

2856: Fig.~\figref{perim.halving}.

2857: If neither fold point $x$ and $y$

2858: is the midpoint of its edge, then

2859: no pair of vertices glue together, so $a_2 = 0$

2860: and therefore $b_1 = n$.

2861: This produces a continuum of polytopes $Q_x$ of $n+2$ vertices.

2862: We call these {\em pita polytopes}

2863: (Fig.~\figref{pita}), and will study them in

2864: Section~\secref{pita} below.

2865: 

2866: Suppose one fold point $x$ is at an edge midpoint.

2867: If $n$ is even, then $y$ is also at a midpoint,

2868: and $P$'s vertices are glued in pairs.

2869: Therefore $a_2 = n/2$ and $b_1 = 0$.

2870: The polytope is a flat half-$n$-gon of $n/2+2$ vertices.

2871: See Fig.~\figref{flat.cases}(a).

2872: If $n$ is odd, then $y$ must be at a vertex.  This means that

2873: $a_1 \neq 0$, and this case does not apply.

2874: 

2875: \item[Case $a_1=2$, $b_0=0$.]

2876: Both leaves are at vertices, and so $n$ must be even.

2877: All other vertices are glued in pairs, so

2878: $a_2 = (n-2)/2$ and $b_1 = 0$.

2879: The folding produces a flat half-$n$-gon

2880: of $n/2+1$ vertices.

2881: See Fig.~\figref{flat.cases}(b).

2882: 

2883: \item[Case $a_1=1$, $b_0=1$.]

2884: One vertex is zipped to a leaf; half the perimeter around

2885: is a fold vertex.  This implies that $n$ is odd.

2886: All other vertices are glued in pairs, so

2887: $a_2 = (n-1)/2$ and $b_1 = 0$.

2888: The folding produces a flat half-$n$-gon

2889: of $(n-1)/2+2$ vertices.

2890: See Fig.~\figref{flat.cases}(c).

2891: \end{description}

2892: The details derived above are gathered into

2893: Table~\tabref{flat.cases},

2894: and the flat foldings illustrated in Fig.~\figref{flat.cases}.

2895: \end{pf}

2896: 

2897: \begin{table}[htbp]

2898: \begin{center}

2899: \begin{tabular}{| c | c | c | c | c | c | l |}

2900:         \hline

2901: $n$ & $a_1$ & $b_0$ & $a_2$ & $b_1$ & $N$ & description

2902:         \\ \hline \hline

2903: any & $0$ & $2$

2904:         & $0$

2905:         & $n$

2906:         & $n+2$

2907:         & pita polyhedra

2908:         \\ \hline

2909: \rule[-6pt]{0pt}{17pt}even & $0$ & $2$

2910:         & $\frac{n}{2}$

2911:         & $0$

2912:         & $\frac{n}{2}+2$

2913:         & flat half-$n$-gon

2914:         \\ 

2915: \rule[-6pt]{0pt}{17pt}even & $2$ & $0$

2916:         & $\frac{n}{2}-1$

2917:         & $0$

2918:         & $\frac{n}{2}+1$

2919:         & flat half-$n$-gon

2920:         \\ 

2921: \rule[-6pt]{0pt}{17pt}odd & $1$ & $1$

2922:         & $\frac{n-1}{2}$

2923:         & $0$

2924:         & $\frac{n+1}{2}+1$

2925:         & flat half-$n$-gon

2926:         \\ \hline

2927: \end{tabular}

2928: \end{center}

2929: \caption{Fold cases for regular $n$-gons, $n > 6$. 

2930: $N$ is the number of polytope vertices.}

2931: \tablab{flat.cases}

2932: \end{table}

2933: 

2934: 

2935: 

2936: \begin{lemma}

2937: For $n \le 6$, regular polygons fold to additional

2938: polytopes (beyond those listed in Lemma~\lemref{ngt6})

2939: as detailed in Table~\tabref{nle6}.

2940: \lemlab{nle6}

2941: \end{lemma}

2942: \begin{pf}

2943: Lemma~\lemref{cut.tree}

2944: limits the possible nonpath cut trees to

2945: `{\tt Y}', `{\tt +}', and `{\tt I}'.

2946: We first argue that

2947: `{\tt I}' is only possible for $n=6$.

2948: The two interior nodes of the tree must have curvatures

2949: in $\{ \a_3, \b_2 \}$.

2950: For $n=3$, there are not enough vertices to make these nodes.

2951: For $n=4$, there are enough vertices to make two $\b_2$

2952: nodes, but this then forces the `{\tt +}' structure,

2953: i.e., the interior edge of the `{\tt I}' has length zero.

2954: For $n=5$ and $n=6$, $\b_2$ is not possible.

2955: For $n=5$, there are not enough vertices to make two $\a_3$ vertices.

2956: And finally, for $n=6$, there are enough vertices, and

2957: the folding produces a flat rectangle.

2958: 

2959: Thus only `{\tt Y}' and `{\tt +}' are possible.

2960: The `{\tt +}' can only be realized in two ways:

2961: by gluing four vertices together, which is only possible for $n=4$

2962: (see $\a_4$ column in Table~\tabref{alpha}),

2963: and by gluing three vertices to an edge, which is only

2964: possible for $n=3$ (see $\b_3$ column in Table~\tabref{beta}).

2965: 

2966: There are a number of ways to realize `{\tt Y}'-trees.

2967: The constraint that the curvature add to $4\pi$, Eq.~\eqref{4pi},

2968: together with the discrete set of possible curvatures in

2969: Tables~\tabref{alpha} and~\tabref{beta},

2970: lead to the possibilities listed in Table~\tabref{nle6}.

2971: (The second line of the table

2972: was previously illustrated in Fig.~\figref{tri.tetra}.)

2973: \end{pf}

2974: 

2975: \begin{table}[htbp]

2976: \begin{center}

2977: \begin{tabular}{| c | c | l | c | c | c | c | l |}

2978:         \hline

2979: $n$ & {\em Tree} & {\em Gluing Description} & {\em Curvatures} 

2980:         & $N$ & {\em Polytope Description}

2981:         \\ \hline \hline

2982: $3$

2983:         & `{\tt Y}'

2984:         & 3 v

2985:         & $\a_3 + 3 \b_0$

2986:         & $4$

2987:         & tetrahedron

2988:         \\ \hline

2989: $3$

2990:         & `{\tt Y}'

2991:         & 2 v + inc e

2992:         & $\a_1 + 2 \b_0 + \b_2$

2993:         & $4$

2994:         & $\infty$ tetrahedra

2995:         \\ \hline

2996: $3$

2997:         & `{\tt Y}'

2998:         & 2 v + adj e

2999:         & $3 \b_0 + 3 \b_1$

3000:         & $4$

3001:         & $\infty$ $5$v polytopes

3002:         \\ \hline

3003: $3$

3004:         & `{\tt +}'

3005:         & 3 v + e

3006:         & $4 \b_0 + \b_3$

3007:         & $4$

3008:         & $\infty$ tetrahedra

3009:         \\ \hline \hline

3010: $4$

3011:         & `{\tt Y}'

3012:         & 3 v

3013:         & $\a_1 + 2 \b_0 + \a_3$

3014:         & $4$

3015:         & tetrahedron

3016:         \\ \hline

3017: $4$

3018:         & `{\tt Y}'

3019:         & 2 adj v + opp e

3020:         & $3 \b_0 +2 \b_1 + \b_2$

3021:         & $5$

3022:         & $\infty$ $5$v polytopes

3023:         \\ \hline

3024: $4$

3025:         & `{\tt Y}'

3026:         & 2 adj v + inc e

3027:         & $4 \b_0 + 2\b_2$

3028:         & $4$

3029:         & $\infty$ tetrahedra

3030:         \\ \hline

3031: $4$

3032:         & `{\tt Y}'

3033:         & 2 adj v + adj e

3034:         & $3 \b_0 +2 \b_1 + \b_2$

3035:         & $5$

3036:         & $\infty$ $5$v polytopes

3037:         \\ \hline

3038: $4$

3039:         & `{\tt Y}'

3040:         & 2 opp v

3041:         & $2 \b_0 + 2 \b_1 + \b_2$

3042:         & $4$

3043:         & $\infty$ tetrahedra

3044:         \\ \hline

3045: $4$

3046:         & `{\tt +}'

3047:         & 4 v

3048:         & $\a_4 + 4 \b_0$

3049:         & $4$

3050:         & flat square

3051:         \\ \hline \hline

3052: $5$

3053:         & `{\tt Y}'

3054:         & 2 adj v + 1 opp v

3055:         & $2 \a_1 + \a_3 + \b_0$

3056:         & $4$

3057:         & tetrahedron

3058:         \\ \hline

3059: $5$

3060:         & `{\tt Y}'

3061:         & 3 adj v

3062:         & $\a_2 + \a_3 + 3\b_0$

3063:         & $5$

3064:         & $5$v polytope

3065:         \\ \hline \hline

3066: $6$

3067:         & `{\tt Y}'

3068:         & 3 alt v

3069:         & $3 \a_1 + \a_3$

3070:         & $3$

3071:         & flat triangle

3072:         \\ \hline

3073: $6$

3074:         & `{\tt Y}'

3075:         & 3 adj v

3076:         & $\a_1 + \a_2 + \a_3 + 2 \b_0$

3077:         & $4$

3078:         & tetrahedron

3079:         \\ \hline

3080: $6$

3081:         & `{\tt Y}'

3082:         & 2 adj v + v

3083:         & $\a_1 + \a_2 + \a_3 + 2 \b_0$

3084:         & $4$

3085:         & tetrahedron

3086:         \\ \hline

3087: $6$

3088:         & `{\tt I}'

3089:         & 3 adj v, 3 adj v

3090:         & $2 \a_3 + 4 \b_0$

3091:         & $4$

3092:         & flat rectangle

3093:         \\ \hline        

3094: \end{tabular}

3095: \end{center}

3096: \caption{Additional fold possibilities for regular $n$-gons, $n \le 6$. 

3097: $N$ is the number of polytope vertices.

3098: Notation in {\em Gluing Description\/} column:

3099: v = vertex,

3100: e = edge,

3101: adj = adjacent,

3102: alt = alternate,

3103: opp = opposite,

3104: inc = included.

3105: In the {\em Polytope Description\/} column:

3106: $\infty$ = continuum of,

3107: $5$v polytope = $5$-vertex polytope.

3108: Each entry of the {\em Curvatures\/} column satisfies 

3109: Eqs.~\eqref{4pi} and~\eqref{indices}.

3110: }

3111: \tablab{nle6}

3112: \end{table}

3113: 

3114: If we treat the $n$ vertices of a regular $n$-gon as assigned

3115: the same label (as seems appropriate),

3116: Lemmas~\lemref{ngt6} and~\lemref{nle6}

3117: together show that there are only $O(1)$ ways to fold up

3118: a regular polygon, justifying the entry in Table~\tabref{results}.

3119: If we label the vertices with distinct labels, then there are $O(n)$

3120: foldings.

3121: 

3122: \subsection{Pita Polytopes}

3123: \seclab{pita}

3124: We define a {\em pita polytope\/} as one obtained by

3125: a perimeter-halving folding of a regular polygon at a

3126: point on an edge that is not a midpoint,

3127: as per the first line of Table~\tabref{flat.cases}.

3128: Let the regular $n$-gon $P$ have unit edge length, and let

3129: the fold points $x$ be distance $a$ from $v_0$ along edge $v_0 v_1$.

3130: Let $b = 1-2a$.  Call the point along $\bP$ to which $v_i$ glues

3131: $v'_i$.

3132: See

3133: Fig.~\figref{duodec} for an example with $n=12$.  We will use this

3134: example throughout the section.

3135: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

3136: \begin{figure}[htbp]

3137: \centering

3138: \includegraphics[width=0.9\textwidth]{duodec.eps}

3139: \caption{Duodecagon, $n=12$, $\a = 30^\circ$,

3140: $a \approx 0.2$, $b \approx 0.6$.

3141: $x$ and $y$ are perimeter-halving fold vertices.

3142: The dashed lines are (largely) conjectured creases.

3143: The vertices $v_i$, and the gluing points $v'_i$,

3144: are labeled with $i$ and $i'$ respectively.

3145: The left and right quadrilaterals play a role

3146: in Fig.~\figref{primes.shorter} below.

3147: }

3148: \figlab{duodec}

3149: \end{figure}

3150: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

3151: 

3152: As mentioned in Section~\secref{Introduction},

3153: we have no method for computing the 3D structure of the

3154: unique polytope determined by a particular Aleksandrov gluing.

3155: Moreover, we do not even have a general method for computing the creases,

3156: i.e., the edges of the polytope.

3157: We will therefore largely conjecture the structure of the pita

3158: polytopes in this section, although we will establish

3159: a subset of the creases.

3160: We will only explore the situation for even $n$.

3161: Let $\a = 2\pi/n$, the turn angle at each vertex of the polygon.

3162: 

3163: We view each pita polytope as composed of four parts:\footnote{

3164:         The relationship to the example in Fig.~\figref{sperm}

3165:         should be evident.

3166: }

3167: \begin{enumerate}

3168: \item A central parallelogram with short side $a$:

3169: $(x, v'_0, y, v'_{n/2})$.

3170: \item A top, nearly half-$n$-gon:

3171: $(v'_0, v'_{n-1}, v'_{n-2}, \ldots, v'_{n/2+1})$.

3172: \item A bottom, nearly half-$n$-gon, congruent by reflection to the top:

3173: $(x, v'_1, v'_2, \ldots, v'_{n/2})$.

3174: \item A ``mouth,'' a strip of triangular teeth;

3175: see Fig.~\figref{strip}.

3176: $n-2$ of the triangles in the strip are

3177: congruent; call their generic shape $T_1$.

3178: $T_1$ has sides of length

3179: $b$, $2a$, and $1$, with an angle $\a$ between $b$ and $2a$.

3180: The two extreme triangles of the mouth are smaller,

3181: of shape $T_2$:

3182: lengths $b$ and $a$ surrounding an angle $\a$.

3183: \end{enumerate}

3184: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

3185: \begin{figure}[htbp]

3186: \centering

3187: \includegraphics[width=\textwidth]{strip.eps}

3188: \caption{Mouth strip of teeth corresponding to Fig.~\figref{duodec}.

3189: (Not to same scale.)

3190: }

3191: \figlab{strip}

3192: \end{figure}

3193: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

3194: We conjecture that the central parallelogram's edges are

3195: creases, as is its central perimeter-splitting diagonal

3196: $xy$.

3197: Call the top and bottom nearly half-$n$-gons

3198: {\em pita polygons.}

3199: We have no conjectures about how the pita polygons

3200: are triangulated (except that they are triangulated the same).

3201: Finally, we prove below in

3202: Lemma~\lemref{mouth} that the mouth

3203: is creased at the edges displayed in

3204: Fig.~\figref{strip}.

3205: 

3206: The final 3D shape looks something like

3207: Fig.~\figref{pita}.

3208: As $n \rightarrow \infty$, the polytope approaches a 

3209: doubly-covered flat semicircle.

3210: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

3211: \begin{figure}[htbp]

3212: \centering

3213: \begin{minipage}{0.41\linewidth}

3214: \centering

3215: \includegraphics[width=\linewidth]{pita.1.eps}

3216: \end{minipage}%

3217: \hspace{5mm}%

3218: \begin{minipage}{0.49\linewidth}

3219: \centering

3220: \includegraphics[width=\linewidth]{pita.2.eps}

3221: \end{minipage}

3222: \caption{Two views of the approximate 3D shape of pita polytope folded as per

3223: Fig.~\figref{duodec}.}

3224: \figlab{pita}

3225: \end{figure}

3226: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

3227: 

3228: We now establish the structure of the mouth of pita polytopes.

3229: We start with this obvious claim:

3230: \begin{lemma}

3231: Pita polytopes are not flat.

3232: \lemlab{not.flat}

3233: \end{lemma}

3234: \begin{pf}

3235: A flat polytope is a pasting of two congruent polygons,

3236: oriented and aligned the same.  The vertices of the polygons

3237: are the only spots on the polytope surface with curvature.

3238: We know the location of all these $n+2$ vertices:

3239: $x$, $y$, and $v_0,\ldots,v_{n-1}$.

3240: Thus the two polygons must be 

3241: $(y,x,v_1,\ldots,v_{n/2})$

3242: $(x,y,v_{n/2+1},\ldots,v_0)$.

3243: However, because $x$ and $y$ are not at the midpoints of their

3244: edges (by definition of a pita polytope),

3245: these two polygons are not congruent.

3246: \end{pf}

3247: 

3248: Our tools will be two facts about edges of triangulated polytopes,

3249: neither of which we will prove:

3250: 

3251: \begin{fact}

3252: Every edge of a polytope is a shortest path between its

3253: endpoint vertices.

3254: \factlab{sp.edge}

3255: \end{fact}

3256: 

3257: Call two polytope edges incident to the same polytope vertex $v$

3258: {\em adjacent\/} if they are consecutive in a circular

3259: sorting around $v$.

3260: \begin{fact}

3261: The smaller surface angle between two adjacent edges incident to

3262: a polytope vertex is less than $\pi$.

3263: In other words,

3264: within every open semicircle of face angle at a polytope vertex $v$,

3265: there is at least one edge incident to $v$.

3266: \factlab{semi}

3267: \end{fact}

3268: 

3269: We use Fact~\factref{sp.edge} to eliminate certain geodesics as

3270: candidates for polytope edges.

3271: The following lemma gathers together some basic distance relationships

3272: to be used later to show that some geodesics are not shortest paths:

3273: 

3274: \begin{lemma}

3275: The following distance relationships hold 

3276: for the length of chords between points

3277: of a pita polygon:

3278: \begin{enumerate}

3279: \item $| v_i - v'_j | = | v'_i - v_j |$.

3280: \item $| v'_i - v'_j | < | v_i - v_j |$ for all $|i-j| > 1$,

3281: i.e., for all $j \neq i$ and $j \neq i \pm 1$.

3282: \item $| v'_i - x | < | v_i - x |$ for all $i \neq 0$.

3283: \item $| v'_i - y | < | v_i - y |$ for all $i \neq n/2$.

3284: \end{enumerate}

3285: \lemlab{distances}

3286: \end{lemma}

3287: \begin{pf}

3288: \begin{enumerate}

3289: \item The polygons cut off by the chords $(v_i,v'_j)$

3290: and $(v'_i,v_j)$ are congruent.

3291: For example, in Fig.~\figref{duodec},

3292: the chord $(v_4,v'_0)$ cuts off a polygon of edge

3293: lengths $(b,1,1,1)$, and

3294: the chord $(v'_4,v_0)$ cuts off a polygon of

3295: lengths $(b,1,1,1)$, both of whose outer interior angles are

3296: all $\a$.

3297: \item

3298: Distances between the $v'_i$ vertices are in general

3299: less than distances between the corresponding unprimed vertices,

3300: because the primed vertices form a regular figure inscribed in

3301: the $n$-gon.

3302: A particular instance is illustrated in

3303: Fig.~\figref{primes.shorter}.  For $j=i+1$, the distances are

3304: equal.

3305: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

3306: \begin{figure}[htbp]

3307: \centering

3308: \includegraphics[height=6cm]{primes.shorter.eps}

3309: \caption{$| v'_1 - v'_4 | < | v_1 - v_4 |$ (cf.~Fig.~\figref{duodec}).}

3310: \figlab{primes.shorter}

3311: \end{figure}

3312: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

3313: \item Here the reason is similar: the primed vertices are

3314: inscribed in the $n$-gon determined by the unprimed vertices.

3315: For example, $|v_1 - x| = a+b$, but

3316: $|v'_1 -x|$ is the length of the hypotenuse of a $T_2$ triangle,

3317: with sides $a$ and $b$, which is shorter by the triangle

3318: inequality.

3319: We will not detail the computations necessary to establish this

3320: claim for all $i$.\complaint{I'm tired}

3321: The only exception to the inequality is for $v_0$, when

3322: $|v_0 - x| = |v'_0 -x| = a$.

3323: \item Symmetric with previous case.

3324: \end{enumerate}

3325: \end{pf}

3326: 

3327: To eliminate the equal-length geodesics in

3328: Lemma~\lemref{distances}(1),

3329: we will need the following:

3330: \begin{lemma}

3331: An edge $e=vu$ of a nonflat polytope $Q$ is a uniquely shortest path,

3332: i.e., there is not another geodesic of the same length from $v$

3333: to $u$.

3334: \lemlab{uniq.sp}

3335: \end{lemma}

3336: \begin{pf}

3337: Suppose $e=vu$ is an edge of $Q$.

3338: Let $g$ be another geodesic between $v$ and $u$ of the same

3339: length as $e$.  Then because $e$ is a straight segment in 3D,

3340: and because any nonstraight path is strictly longer, it must

3341: be that $g$ is also a straight segment in 3D.  Thus it must

3342: be coincident with $e$.  If $e$ and $g$ are nevertheless distinct,

3343: then they must be on opposite sides of a flat surface.

3344: But then $Q$ must be flat (``pinched'') at $e=g$, 

3345: which by convexity implies

3346: that $Q$ is entirely flat.  This contradicts the assumption of the lemma.

3347: \end{pf}

3348: 

3349: We now have assembled enough information to pin down the structure

3350: of the mouth:

3351: \begin{lemma}

3352: The mouth of a pita polytope is

3353: triangulated as in Fig.~\figref{strip}:

3354: the edges 

3355: $$(x,v_1,\ldots,v_{n/2},y,v_{n/2+1},\ldots,v_{n-1},v_0)$$

3356: surrounding the mouth, and 

3357: the diagonals $(v_i, v_{n-i})$

3358: and $(v_{n-i+1},v_i)$ that delimit its ``teeth'' 

3359: (cf.~Fig.~\figref{strip}),

3360: are all polytope edges.

3361: \lemlab{mouth}

3362: \end{lemma}

3363: \begin{pf}

3364: Let $v_i$, $i \in \{ 1,\ldots,n/2-1 \}$ be a vertex of the

3365: mouth.  

3366: (It may help to think of $v_4$ in Fig.~\figref{duodec} as

3367: a typical $v_i$ in this proof.)

3368: By Fact~\factref{semi}, there must be a polytope

3369: edge $e$ incident to $v_i$ on the top face in the half plane

3370: bounded by the line through $v_{i-1} v_i$.

3371: By Lemma~\lemref{distances}(2), the other endpoint of $e$

3372: cannot be any $v_j$, $|i-j| > 1$, for all those are longer than

3373: $|v'_i - v'_j|$, the length of an alternate geodesic.

3374: So they are not shortest paths, and are

3375: ruled out by Fact~\factref{sp.edge}.

3376: By Lemma~\lemref{distances}(3-4), the other endpoint of $e$

3377: cannot be $x$ or $y$, for we have restricted $i$ so that

3378: $i \neq 0$ and

3379: $i \neq n/2$.

3380: This leaves $v'_j$ as a possible endpoint of $e$.

3381: But by Lemma~\lemref{distances}(1), $v_i v'_j$ is not

3382: uniquely shortest, which by Lemma~\lemref{uniq.sp} then implies

3383: that $Q$ must be flat, which we know is false by

3384: Lemma~\lemref{not.flat}.

3385: 

3386: We have excluded all candidates for the endpoint of $e$

3387: except for $j= i \pm 1$.

3388: Because we are examining the semicircle bounded by  $v_{i-1},v_i$,

3389: this leaves $v_{i+1}$ as the only possible endpoint.  Thus

3390: $v_i v_{i+1}$ is an edge of the polytope.

3391: 

3392: Repeating this argument for the bottom face,

3393: $i \in \{ n/2+1,\ldots,n-1 \}$,

3394: establishes the outer boundary of the mouth,

3395: excluding the edges incident to $x$ and $y$.

3396: Those can be argued similarly.\complaint{I hope}

3397: The teeth diagonals are now easy to see.

3398: We illustrate with $v_4$ in Fig.~\figref{duodec}.

3399: We have just proved that no edge is incident to $v_4$

3400: across the top face. But that top face must be triangulated

3401: somehow.  The only way to triangulate it without using a

3402: diagonal incident to $v_4$ is to include the diagonal

3403: $v'_9 v'_8$.  This means that $v_4 v_8$ and $v_4 v_9$

3404: are edges of the polytope. 

3405: \end{pf}

3406: 

3407: %%%%%%%%%%%%%%%%%%%BIB

3408: % Make the bibliography

3409: %\newpage

3410: \bibliographystyle{alpha}

3411: %\bibliography{cc,/home1/orourke/bib/geom/geom}

3412: \bibliography{cc}

3413: \end{document}

3414: %%%%%%%%%%%%%%%%%%%BIB

3415: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

3416: 

3417: \begin{enumerate}

3418: \item

3419: \end{enumerate}

3420: 

3421: \begin{lemma}

3422: \lemlab{X}

3423: \end{lemma}

3424: \begin{pf}

3425: \end{pf}

3426: 

3427: %9lines

3428: see Fig.~\figref{X}.

3429: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

3430: \begin{figure}[htbp]

3431: \centering

3432: \includegraphics[width=\linewidth]{X.eps}

3433: \caption{Z.}

3434: \figlab{X}

3435: \end{figure}

3436: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

3437: