1: \pdfoutput=1  %arXiv addition for pdflatex, permits no file extensions.

2: %Must go on first line!

3: \documentclass[]{article}  

4: \usepackage{url,float}

5: \usepackage{graphicx}

6: \usepackage{amsfonts}

7: \usepackage{amssymb}

8: \usepackage{latexsym}

9: 

10: 

11: %\input{/home/orourke/tex/mac}

12: %\withcomplaints

13: 

14: \newcommand{\hide}[1]{}

15: 

16: \newcommand{\ABox}{

17: \raisebox{3pt}{\framebox[6pt]{\rule{6pt}{0pt}}}

18: }

19: \newenvironment{proof}{{\bf Proof:}}{\hfill\ABox}

20: %\newenvironment{pf}{{\bf Proof:}}{\hfill\ABox}

21: 

22: \newtheorem{theorem}{{\bf Theorem}}

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

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

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

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

27: 

28: %For labels of items.

29: \newcommand{\lemlab}[1]{\label{lemma:#1}}

30: \newcommand{\thmlab}[1]{\label{thm:#1}}

31: \newcommand{\eqlab}[1]{\label{eq:#1}}

32: \newcommand{\corlab}[1]{\label{cor:#1}}

33: \newcommand{\deflab}[1]{\label{def:#1}}

34: \newcommand{\tablab}[1]{\label{tab:#1}}

35: \newcommand{\figlab}[1]{\label{fig:#1}}

36: \newcommand{\seclab}[1]{\label{sec:#1}}

37: \newcommand{\chaplab}[1]{\label{chap:#1}}

38: 

39: \newcommand{\lemref}[1]{\ref{lemma:#1}}

40: \newcommand{\thmref}[1]{\ref{thm:#1}}

41: \newcommand{\corref}[1]{\ref{cor:#1}}

42: \newcommand{\defref}[1]{\ref{def:#1}}

43: \newcommand{\chapref}[1]{\ref{chap:#1}}

44: \newcommand{\secref}[1]{\ref{sec:#1}}

45: \newcommand{\eqref}[1]{\ref{eq:#1}}

46: \newcommand{\figref}[1]{\ref{fig:#1}}

47: \newcommand{\tabref}[1]{\ref{tab:#1}}

48: 

49: %\floatstyle{ruled}

50: %\newfloat{algorithm}{htbp}{loa}

51: %\floatname{algorithm}{Algorithm}

52: 

53: 

54: %From Erik Demaine:

55: % Complex \xxx for making notes of things to do.  Use \xxx{...} for general

56: % notes, and \xxx[who]{...} if you want to blame someone in particular.

57: % Puts text in brackets and in bold font, and normally adds a marginpar

58: % with the text `xxx'' so that it is easy to find.  On the other hand, if

59: % the comment is in a minipage, figure, or caption, the xxx goes in the text,

60: % because marginpars are not possible in these situations.

61: {\makeatletter

62:  \gdef\xxxmark{%

63:    \expandafter\ifx\csname @mpargs\endcsname\relax % in minipage?

64:      \expandafter\ifx\csname @captype\endcsname\relax % in figure/caption?

65:        \marginpar{xxx}% not in a caption or minipage, can use marginpar

66:      \else

67:        xxx % notice trailing space

68:      \fi

69:    \else

70:      xxx % notice trailing space

71:    \fi}

72:  \gdef\xxx{\@ifnextchar[\xxx@lab\xxx@nolab}

73:  \long\gdef\xxx@lab[#1]#2{{\bf [\xxxmark #2 ---{\sc #1}]}}

74:  \long\gdef\xxx@nolab#1{{\bf [\xxxmark #1]}}

75:  % This turns them off:

76:  %\long\gdef\xxx@lab[#1]#2{}\long\gdef\xxx@nolab#1{}%

77:  \gdef\turnoffxxx{\long\gdef\xxx@lab[##1]##2{}\long\gdef\xxx@nolab##1{}}%

78: }

79: 

80: 

81: % Blackboard R for real numbers, S for sphere (\S taken someplace, so \Sph).

82: \def\P{{\mathcal P}}

83: \def\C{{\mathcal C}}

84: \def\B{{\mathcal B}}

85: \def\g{{\gamma}}

86: \def\l{{\lambda}}

87: \def\r{{\rho}}

88: \def\s{{\sigma}}

89: \def\d{{\delta}}

90: \def\a{{\alpha}}

91: \def\b{{\beta}}

92: \def\th{{\theta}}

93: %\def\C{{\mathcal C}}

94: %\def\hull{\mathop{\rm hull}\nolimits}

95: \def\sp{\mathop{\rm sp}\nolimits}

96: 

97: 

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

99: 

100: %\newcounter{abc}

101: 

102: 

103: \title{Unfolding Restricted Convex Caps

104: }

105: 

106: \author{%

107: Joseph O'Rourke%

108:     \thanks{Dept. Comput. Sci., Smith College, Northampton, MA

109:       01063, USA.

110:       \protect\url{orourke@cs.smith.edu}.}

111: }%author

112: 

113: \begin{document}

114: \maketitle

115: 

116: \begin{abstract}

117: This paper details an algorithm for unfolding a class of convex

118: polyhedra,

119: where each polyhedron in the class

120: consists of a convex cap over a rectangular base,

121: with several restrictions:

122: the cap's faces are quadrilaterals, with vertices over an underlying integer

123: lattice, and such that the cap convexity is ``radially monotone,''

124: a type of smoothness constraint.

125: Extensions of Cauchy's arm lemma are used in the proof of non-overlap.

126: \end{abstract}

127: 

128: \section{Introduction}

129: \seclab{intro}

130: Few classes of convex polyhedra are known to be \emph{edge-unfoldable}:

131: unfolded by cutting along edges of the polyhedron and flattening into

132: the plane to a single piece without overlap.

133: Among the classes known to be are: pyramids, prismoids, and ``domes.''

134: See~\cite[Chap.~22]{do-gfalop-07} for background on this problem.

135: 

136: The purpose of this informal note is to introduce another narrow class of polyhedra

137: for which an edge-unfolding algorithm can be provided.

138: We call this class \emph{radially monotone lattice quadrilateral convex caps}.

139: The long name reflects the several qualifications needed to guarantee

140: correctness,

141: qualifications that hopefully can be removed by subsequent research.

142: The upper surface of such a polyhedron $\P$ is a 

143: \emph{convex cap} $C$ in the sense that

144: $C$ projects parallel to the $z$-axis to its 

145: base $B$ in the $xy$-plane without overlap.

146: In other words, the intersection of the cap $C$ with a line 

147: parallel to $z$ through an interior

148: point of $B$ is a single point.

149: In other terminology, $C$ is a \emph{terrain}.

150: $C$ is composed entirely of quadrilaterals, each of which projects to a unit lattice

151: square in the $xy$-plane, and whose projections tile $B$.

152: The base $B$ is restricted to be a rectangle.

153: $\P$ then is the convex hull of $C \cup B$, which fills in the four sides

154: $S_{x^-}$,

155: $S_{x^+}$,

156: $S_{y^-}$,

157: $S_{y^+}$.

158: See Figure~\figref{unf_k16_r2}(a,b).

159: The ``radially monotone'' qualification restricts the ``sharpness'' of

160: the convexity of the cap in a way that is not easily explained until we develop

161: more notation.

162: 

163: A more general shape would be a \emph{lattice convex cap},

164: the convex hull of $B$ and a set of points $(x,y,z)$

165: over each integer lattice point $(x,y)$ in $B$.  Here the faces are in general

166: triangles rather than quadrilaterals, and $C$ is a \emph{convex height field}.

167: The restriction to quadrilaterals

168: narrows the class considerably, as we now show.

169: 

170: \paragraph{Quadrilateral Restriction.}

171: 

172: Let the base $B$ range

173: over the lattice points

174: $x=1,\ldots,n_x$ and $y=1,\ldots,n_y$.

175: If the points above the front and left of $B$ are specified,

176: $$

177: c_x(1) \; : \; \{ (x,1) \; : \; x=1,\ldots,n_x \}

178: $$

179: and

180: $$

181: c_y(1) \; : \; \{ (1,y) \; : \; y=1,\ldots,n_y \}

182: $$

183: then all the points above the remainder of $B$ are determined.

184: %The convex curves bounding $C$ above these two sides of $B$ 

185: %we call $c_x(1)$ and $c_y(1)$ respectively.

186: This can be seen as follows.

187: If three of the four corners of a lattice cell are given,

188: $$

189: (x,y,z_0),

190: (x+1,y,z_1),

191: (x,y+1,z_3)

192: $$

193: then the fourth corner, $(x+1,y+1,z_2)$, is determined by the

194: plane containing the first three corner points, whose height

195: can be computed as

196: \begin{equation}

197: z_2 = -z_0 + z_1 + z_3

198: \eqlab{z2}

199: \end{equation}

200: Thus,

201: $$

202: (1,1,z_0),

203: (2,1,z_1),

204: (1,2,z_3)

205: $$

206: determine $(2,2,z_2)$,

207: and this determination propagates similarly out over the entire $n_x \times n_y$ rectangle.

208: Thus, if $n_x=n_y=n$, the $n^2$

209: lattice points of $B$ are fixed by specifying just the $2n -1$ values along 

210: the front and left sides.

211: %

212: A consequence of

213: Lemma~\lemref{convex.cap} below is 

214: that a lattice quadrilateral cap $C$ is convex if and only if

215: the curves at its two boundaries

216: $c_x(1)$ and $c_y(1)$

217: (i.e., the upper boundaries of $S_{x^-}$ and $S_{y^-}$)

218: are convex.

219: 

220: \paragraph{Radially Monotone.}

221: The full definition of when a convex cap is radially monotone will be

222: deferred until Section~\secref{radial} below, where it is first employed.

223: Here we define when a planar convex curve is radially monotone,

224: and give two sufficient conditions for radial monotonicity of a convex cap.

225: Let $c$ be a convex curve, and let $p_0$ and $p_i$ be two vertices of the curve,

226: with $p_{i+1}$ the vertex following $p_i$.

227: Then $c$ is \emph{radially monotone} if

228: the angle $\angle p_0 p_i p_{i+1} \ge \pi/2$,

229: for every pair of vertices $p_0$ and $p_i$.

230: The condition may be interpreted as requiring that

231: the next segment $p_i p_{i+1}$ of the chain

232: does not penetrate the circle of radius $|p_0 p_i|$

233: centered on $p_0$.

234: So points on the chain increase their radial distance from $p_0$.

235: (See ahead to Figure~\figref{radially_monotone}.)

236: The two sufficient conditions are as follows:

237: \begin{enumerate}

238: \squeezelist

239: \item

240: If the two convex curves $c_x(1)$ and $c_y(1)$ are both radially monotone,

241: then the convex cap is radially monotone.

242: \item

243: If each of these convex curves fits in a semicircle connecting its two endpoints,

244: then each is radially monotone, and so~(1) applies.

245: \end{enumerate}

246: This latter condition justifies the notion that radial monotonicity

247: enforces a type of smoothness on the convexity of the cap.

248: 

249: \section{Lattice Quadrilateral Cap Properties}

250: We now derive several properties of lattice quadrilateral convex caps.

251: 

252: \begin{lemma}[Parallelograms]

253: Every quadrilateral over a lattice cell is a parallelogram.

254: \lemlab{para}

255: \end{lemma}

256: \begin{proof}

257: Let the four corners of the quadrilateral be $a$, $b$, $c$, $d$

258: in counterclockwise order,

259: let $\a$ and $\b$ be the angles at $a$ and $b$ respectively,

260: and let $\Pi$ be the plane containing the quadrilateral.

261: We argue that the two coplanar triangles

262: $\triangle abd$ and 

263: $\triangle bcd$ are congruent.

264: They share the diagonal of the quadrilateral, $bd$.

265: The edge $ab$ must be congruent to edge $dc$, because both are 

266: the intersection of $\Pi$ with a unit vertical strip,

267: parallel (say) to the $xz$-plane.

268: Therefore $|ab|=|dc|$.

269: Similarly, $bc$ must be congruent to $ad$, because

270: both are produced by $\Pi$ intersecting the a unit vertical

271: $yz$-strip.

272: Therefore $|bc|=|dc|$.

273: Therefore $\triangle abd$ and 

274: $\triangle bcd$ have all edges the same length, and

275: so are congruent, as claimed.

276: So the angles at $a$ and at $c$ are the same, $\a$.

277: Similarly, the angles at $b$ and at $d$ are the same, $\b$.

278: \end{proof}

279: 

280: Call the chain of edges of $C$ above $(y,1),(y,2),\ldots,(y,n_x)$

281: the $x$-chain $c_x(y)$, and similarly define the $y$-chain $c_y(x)$.

282: 

283: \begin{lemma}[Cap Convexity]

284: For a lattice quadrilateral cap $C$,

285: all the $x$-chains are congruent to one another, as are all the $y$-chains.

286: \lemlab{convex.cap}

287: \end{lemma}

288: \begin{proof}

289: Consider two consecutive edges of the $x$-chain $c_x(y)$,

290: over $(x,y),(x+1,y),(x+2,y)$.

291: By the Parallelogram Lemma~(\lemref{para}),

292: the corresponding edges in the next $x$-chain $c_x(y+1)$ are

293: parallel to those in $c_x(y)$.

294: Thus the convex angle at the middle point $(x+1,y)$

295: is exactly the same as that at $(x+1,y+1)$.

296: Because this is true at every lattice point of each $x$-chain,

297: all are congruent.

298: \end{proof}

299: 

300: \noindent

301: The congruency of the chains is evident in Figure~\figref{unf_k16_r2}(a,b).

302: 

303: 

304: 

305: Let $z(x,y)$ be the point of $C$ above $(x,y)$.

306: \begin{lemma}[Cap Maxima]

307: The maximum $z$-height of

308: a lattice quadrilateral cap $C$ for

309: the points at a particular $x$-value,

310: $\max_y z(x,y)$,

311: is achieved at a $y$-value $y_{max}$ independent of $x$.

312: And the same claim holds when the roles of $x$ and $y$ are interchanged.

313: \lemlab{convex.cap.max}

314: \end{lemma}

315: \begin{proof}

316: This follows immediately from the congruency of the $x$-chains

317: provided by Lemma~\lemref{convex.cap}.

318: \end{proof}

319: 

320: \noindent

321: For example, in Figure~\figref{unf_k16_r2}(a,b), $y_{max}=14$.

322: %Lemma~\lemref{convex.cap} also follows easily from the Uphill Lemma.

323: 

324: 

325: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

326: \begin{figure}[htbp]

327: \centering

328: %\includegraphics[height=0.95\textheight]{Figures/unf_k16_r2}

329: \includegraphics[width=\linewidth]{Figures/unf_k16_r2}

330: \caption{(a,b) Two views of the same convex cap $C$.

331: %The location of the side faces 

332: %$S_{x^-}$,

333: %$S_{x^+}$,

334: %$S_{y^-}$,

335: %$S_{y^+}$

336: %are marked but the faces are not shown;

337: The base $B$ is the square at $z=0$.

338: (c)~Unfolding of $\P$.}

339: \figlab{unf_k16_r2}

340: \end{figure}

341: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

342: 

343: \section{Sketch of Algorithm}

344: We first sketch the unfolding algorithm, using

345: Figure~\figref{unf_k16_r2} for illustration,

346: before justifying it in the following sections.

347: The strip of quadrilaterals between 

348: $y_{max}$ and $y_{max}+1$, call it

349: the \emph{$y_{max}$-strip} is unfolded with $(1,y_{max})$

350: at the origin and its leftmost edge

351: %(between $(1,y_{max})$ and $(1,y_{max}+1)$)

352: along the $y$-axis.  Thus it unfolds roughly horizontally.

353: See the green $y=14$ strip in Figure~\figref{unf_k16_r2}(c).

354: All the other ``parallel'' $x$-strips for different $y$-values

355: are arranged above and below this $y_{max}$-strip,

356: connected along the uncut edge between $x=1$ and $x=2$.

357: All other $x$-edges for $x \ge 2$ are cut.

358: We will show that these strips splay outward from the $y_{max}$-strip,

359: avoiding overlap.

360: The two side faces

361: $\{ S_{x^-}, S_{x^+},\}$

362: are unfolded attached to the base $B$,

363: and attached to the cap unfolding at the leftmost vertical

364: edge of the $y_{max}$-strip.

365: We will show that $S_{x^-}$ in particular does not overlap

366: with the $x=1,2$ strip (outlined in red in the figure).

367: The remaining two side faces,

368: $\{ S_{y^-},S_{y^+} \}$, are attached at the two ends of the $x=1$ strip

369: as shown in the figure.

370: 

371: \section{Quadrilateral Angles}

372: 

373: We will call the four angles of a quadrilateral

374: $(\a,\b,\a,\b)$ in counterclockwise order.

375: Note that $\a+\b=\pi$, so one angle determines all the others.

376: Much of our reasoning relies on an analysis of the behavior of

377: the quadrilateral angles.

378: 

379: Fix one quadrilateral over its lattice cell, with corner heights

380: $z_0=0,z_1,z_2,z_3$, counterclockwise

381: above $(0,0),(1,0),(1,1),(0,1)$ respectively,

382: as in Figure~\figref{alpha_z1}(a).

383: We will need detailed knowledge of how the $(0,0)$ corner

384: angle $\a$ varies as a function of $z_1$ and $z_3$:

385: 

386: \begin{lemma}[One Quad]

387: For $z_3 > 0$, $\a(z_1)$ is a strictly decreasing function

388: of $z_1$, passing through $\pi/2$ at $z_1=0$.

389: $\a(z_1,-z_3) = \pi/2 - \a(z_1,z_3)$,

390: so, for $z_3 < 0$, $\a(z_1)$ is a strictly increasing function

391: of $z_1$.

392: \lemlab{alpha.z1}

393: \end{lemma}

394: \begin{proof}

395: Explicit calculation shows that

396: $$

397: \a = \a(z_1,z_3) =

398: \cos^{-1} \left( \frac {- z_1 z_3}

399: {

400: \sqrt{ 1 + z_1^2 }

401: \sqrt{ 1 + z_3^2 }

402: }

403: \right)

404: $$

405: The claimed properties follow from the properties of the inverse cosine function.

406: See Figure~\figref{alpha_z1}(b).

407: \end{proof}

408: 

409: 

410: 

411: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

412: \begin{figure}[htbp]

413: \centering

414: \includegraphics[width=0.95\linewidth]{Figures/alpha_z1}

415: \caption{(a) one quadrilateral;

416: (b)~$\a(z_1,z_3)$ is strictly decreasing for $z_3 > 0$.}

417: \figlab{alpha_z1}

418: \end{figure}

419: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

420: 

421: 

422: We now examine angles formed by two adjacent quadrilaterals.

423: We fix one in the position detailed above, and the other

424: adjacent to its left, with height $z'_1$ above the lattice point $(-1,0)$.

425: Note that the other three corners of this second quadrilateral are

426: then determined.

427: Let $\th$ be the angle in the vertical plane containing

428: the edges above $(-1,0),(0,0),(1,0)$,

429: and let $\a$ and $\b'$ be the angles of the quadrilateral incident

430: to the origin $(0,0,0)$.

431: See Figure~\figref{alphabeta_z3}(a).

432: We need the behavior of $(\a+\b')$ as $z_3$ varies:

433: 

434: \begin{lemma}[Two Quads]

435: Let the front edges of two adjacent quadrilaterals make

436: a convex angle of $\th$ in the vertical plane containing

437: those edges.

438: Then $(\a+\b')$ is a strictly increasing function of $z_3$,

439: passing through $\pi$ at $z_3=0$, and

440: with asymptote $\th$ for $z_3 < 0$ and $2\pi-\th$ for $z_3 > 0$.

441: \lemlab{alphabeta.z3}

442: \end{lemma}

443: \begin{proof}

444: A proof by explicit calculation, using

445: the formula used in the One-Quad Lemma~(\lemref{alpha.z1})

446: twice,

447: will not be presented.

448: Alternatively, one can arrive at the claim via the

449: One-Quad Lemma %~\lemref{alpha.z1}, 

450: as follows.

451: From Figure~\figref{alpha_z1}, $\a$ decreases as $z_3$

452: increases, but, because

453: $|z'_1| > z_1$ by convexity, $\b'$ increases more than $\a$ decreases,

454: and so $\a+\b'$ increases.

455: See Figure~\figref{alphabeta_z3}(b) for sample plots.

456: Note that, as $z_3 \rightarrow +\infty$,

457: $\a$ and $\b'$ approach lying in the vertical plane above $\th$,

458: and so $\th + \a + \b' = 2 \pi$,

459: and as $z_3 \rightarrow -\infty$,

460: $\a + \b' \rightarrow \th$.

461: \end{proof}

462: 

463: 

464: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

465: \begin{figure}[htbp]

466: \centering

467: \includegraphics[width=0.95\linewidth]{Figures/alphabeta_z3}

468: \caption{(a) Two adjacent quadrilaterals;

469: here $\th=146.3^\circ$.

470: (b) $(\a+\b')$ is a strictly increasing function of $z_3$;

471: Note the asymptote $\th=146.3^\circ$ for the highlighted curve

472: corresponding to~(a).

473: }

474: \figlab{alphabeta_z3}

475: \end{figure}

476: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

477: 

478: Define the \emph{turn angle} $\tau$ at point of a curve with convex angle $\th$

479: to be $\pi - \th$.

480: 

481: \begin{lemma}[Strip Unfolding]

482: Each strip of quadrilaterals unfolds without self-overlap,

483: with convex boundary curves.

484: \lemlab{one.strip}

485: \end{lemma}

486: \begin{proof}

487: Let an $x$-strip lie between $y$ and $y+1$.

488: Let $\th(x)$ be the angle at $p=(x,y,z(x,y))$

489: in the vertical $y$-plane through $p$.

490: We know from the Convex Cap Lemma~(\lemref{convex.cap})

491: that $\th \le \pi$ because the curve is convex at $p$.

492: Let $\tau_0 = \pi - \th$ be the turn angle at $p$ in the vertical plane.

493: The turn angle of

494: the unfolded quadrilaterals incident to $p$

495: is $\tau = (\a+\b')-\pi$.

496: The Two-Quad Lemma~(\lemref{alphabeta.z3}) shows that $(\a+b') \in [ \th, 2\pi - \th ]$,

497: which implies that

498: $\tau \in [ \th - \pi, \pi - \th ]$,

499: i.e.,

500: $\tau \in [ -\tau_0, +\tau_0 ]$.

501: These are precisely the angle conditions for application of the

502: extension of Cauchy's arm lemma:

503: all turn angles of the convex curve are ``straightened''~\cite{o-ecala-01}.

504: (This is an extension of Cauchy's arm lemma because nonconvexities might be introduced.)

505: The usual conclusion is that, under such a \emph{valid reconfiguration},

506: the endpoints of the chain pull further apart.

507: In~\cite[Cor.~2]{o-dipp-03} this consequence is derived:

508: A valid reconfiguration of an open convex chain remains simple,

509: i.e., does not self-intersect.

510: 

511: Although in general a valid reconfiguration could result in a nonconvex

512: straightened curve,

513: in our case the Convex Cap Lemma~(\lemref{convex.cap}) 

514: places us  

515: to the left or right of $z_3=0$ in Figure~\figref{alphabeta_z3}(b),

516: keeping all turn angles $\tau$ the same sign, and therefore

517: resulting in a convex curve.

518: 

519: So now we have established that neither of the two unfolded boundary curves

520: of the $x$-strip,

521: call them $u_y$

522: from the chain at $y$ and $u_{y+1}$ that at $y+1$,

523: self-intersect.

524: But by the Parallelograms Lemma~(\lemref{para}),

525: $u_y$ and $u_{y+1}$ are connected by, and therefore separated by, edges that are all parallel.

526: (See Figures~\figref{unf_k16_r2}(c), \figref{16x16_unf}, and~\figref{32x32_unf}.)

527: Thus $u_y$ cannot intersect $u_{y+1}$.

528: Finally, these parallel ``ladder bars'' ensure that

529: it is not possible for the first $x$-segment to intersect

530: the last $x$-segment.

531: Therefore the entire strip unfolds without self-overlap.

532: \end{proof}

533: 

534: Finally we move to an analysis of four quadrilaterals, needed to compare

535: turn angles in the unfolding.

536: 

537: \begin{lemma}[Four Quads]

538: Let four quadrilaterals be incident to the origin $o=(0,0,0)$,

539: with those touching $y=1$ the \emph{upper quads}

540: and those touching $y=-1$ the \emph{lower quads}.

541: Let the upper quads unfold with turn angle $\tau$ at $o$,

542: and the lower quads with turn $\tau_0$ at $o$.

543: Then, if the middle edge of the lower quads, over $(0,-1),(0,0)$,

544: is uphill w.r.t. $z$ ($z'_3 < 0$ in the notation below), 

545: then $\tau_0 \ge \tau$:  the lower quads unfold to turn more sharply.

546: \lemlab{four.quads}

547: \end{lemma}

548: 

549: \begin{proof}

550: We extend the analysis in the Two-Quad Lemma~(\lemref{alphabeta.z3})

551: to four quadrilaterals incident to a central point,

552: taken to be $o=(0,0,0)$ without loss of generality.

553: We repeat all the notation developed in that lemma

554: and displayed in 

555: Figure~\figref{alphabeta_z3}(a),

556: and add two new quadrilaterals,

557: determined by one new parameter $z'_3 < 0$,

558: as shown in Figure~\figref{fourquads},

559: with angle labels as indicated.

560: Knowing from the Convex Cap Lemma~(\lemref{convex.cap}) that the

561: two convex $x$-chains are congruent, we know they determine the same

562: $\th$ in vertical planes, and so Lemma~\lemref{alphabeta.z3}

563: can be applied on an equal footing to both chains.

564: Our goal is to compare the angle $A = \a+\b'$

565: to $B=\a'_0 + \b_0$, for these angles determine the turn angles.

566: Note that $A$ is the same angle studied

567: in the Two-Quads Lemma~(\lemref{alphabeta.z3}), but $B = 2\pi -(\a_0 + \b'_0)$,

568: where $(\a_0 + \b'_0)$ is the angle to which Lemma~\lemref{alphabeta.z3} applies.

569: 

570: Now, in order to maintain convexity along the $y$-chain through the central

571: point, we must have $|z'_3| \ge z_3$

572: (recall the central point is the origin, so $z'_3 < 0$).

573: Because $|z'_3|$ is right of $z_3$ in 

574: Figure~\figref{alphabeta_z3}(b), and the angle sum is strictly increasing,

575: we know that $\a_0 + \b'_0 \ge \a+\b'$,

576: i.e., $2\pi-B \ge A$.

577: Let $\tau$ be the turn angle at $o$ of the unfolding

578: of the upper two quads, $\tau = A-\pi$,

579: and $\tau_0$ the same turn angle of the lower two quads, $\tau_0 = \pi-B$.

580: Substitution into the inequality $2\pi-B \ge A$

581: yields $\tau_0 \ge \tau$.

582: \end{proof}

583: 

584: %see Figure~\figref{fourquads}.

585: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

586: \begin{figure}[htbp]

587: \centering

588: \includegraphics[width=0.45\linewidth]{Figures/fourquads}

589: \caption{Four quadrilaterals incident to $(0,0,0)$.

590: Convexity implies that $|z'_3| \ge z_3$.}

591: \figlab{fourquads}

592: \end{figure}

593: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

594: 

595: \noindent

596: The sharper turn angles guaranteed by this lemma provide a type

597: of separation between adjacent strips, but not enough to ensure

598: non-overlap.

599: 

600: \section{Radial Monotonicity}

601: \seclab{radial}

602: Now we finally come to the point where we need to use radial monotonicty,

603: to guarantee that the unfolding of adjacent strips do not overlap.

604: The condition we need is that the unfolded edge of each quadrilateral

605: strip be radially monotone, in the sense defined in Section~\secref{intro} above.

606: The Strip Unfolding Lemma~(\lemref{one.strip}) 

607: shows that the unfolded strip boundary is

608: a straightening of the curve $c_x(y)$ in the vertical plane,

609: so that radial monotonicity of $c_x(y)$ implies radial monotonicity of

610: the strip boundary, which justifies the first sufficient condition

611: claimed in Section~\secref{intro}.

612: 

613: We now show that any opening/straightening of a radially monotone curve

614: avoids overlap:\footnote{

615:    It may be that the equivalent of this lemma is available in the literature,

616:    but no explicit reference could be found.}

617: 

618: \begin{lemma}[Radial Monotonicity]

619: Let $c'$ be a straightening of a convex curve $c$, i.e.,

620: a reconfiguration such that every convex angle either stays

621: the same or opens closer to (but not exceeding) $\pi$.

622: Let $c$ lie in the positive quadrant with origin $o$,

623: with the left endpoints of both

624: $c$ and $c'$ at $o$.

625: If $c$ is radially monotone, then $c'$ does not intersect $c$ except at $o$.

626: \lemlab{radial}

627: \end{lemma}

628: \begin{proof}

629: First consider the curve $c_2$ in 

630: Figure~\figref{radially_monotone}, which is not

631: radially monotone at $a$ because $\angle oab < \pi/2$.

632: One can see that segment $ab$ cuts into the circle of radius

633: $|oa|$, and that under rotation about $o$, $a'b'$ crosses $ab$.

634: On the other hand, curve $c_1$ in the figure is radially monotone,

635: with each succeeding segment connecting two concentric circles.

636: Now rotation about $o$ moves each segment of $c_1$ on those circles,

637: and thus keeps the segment confined to the annulus between.  Therefore

638: such rotation cannot intersect any part of $c_1$.

639: %By Cauchy's arm lemma, the rotated $c'_1$ cannot self-intersect

640: 

641: Now because the lemma assumes that $c$ lies in the positive quadrant,

642: it is radially monotone from any other vertex $p$ beyond $o$, for the

643: angle $\angle pab$ can only be larger than $\angle oab$.

644: Now we view the opening of $c$ as composed of

645: rotations about each joint playing

646: the role of $o$ successively.  As the chain is rigid beyond

647: each rotation pivot $o$, the radial monotonicity

648: beyond $o$ is not affected.  So none of the rotations can cause

649: overlap, and the lemma is established.

650: \end{proof}

651: %see Figure~\figref{radially_monotone}.

652: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

653: \begin{figure}[htbp]

654: \centering

655: \includegraphics[width=0.7\linewidth]{Figures/radially_monotone}

656: \caption{$c_1$ is radially monotone, and $c'_1 \cap c_1 = \{o\}$; 

657: $c_2$ is not, and $ab \cap a'b' \neq \emptyset$.}

658: \figlab{radially_monotone}

659: \end{figure}

660: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

661: 

662: 

663: Note that if $c$ stays inside the semicircle 

664: connecting its endpoints (drawn lightly in Figure~\figref{radially_monotone}),

665: then radial monotonicity is guaranteed (because every point on the semicircle

666: subtends $\pi/2$ from the diameter).

667: This justifies the second sufficient condition

668: claimed in Section~\secref{intro}.

669: (None of the various sufficient conditions mentioned are necessary,

670: e.g., $c_1$ in the figure does not fit in the semicircle but is nevertheless

671: radially monotone.)

672: 

673: 

674: \begin{lemma}[Adjacent Strips]

675: Adjacent $x$-strips of quadrilaterals unfold without overlap.

676: \lemlab{adj.strips}

677: \end{lemma}

678: \begin{proof}

679: This now follows immediately from 

680: the Four-Quad Lemma~(\lemref{four.quads}),

681: which establishes that adjacent strip boundaries correspond

682: to a convex curve $c$ and a straightening $c'$

683: (because $\tau_0$, corresponding to $c'$, is $\ge \tau$,

684: corresponding to $c$),

685: and 

686: the Radial Monotonicity Lemma~(\lemref{radial}), 

687: which guarantees non-overlap of the two curves.

688: \end{proof}

689: 

690: \noindent

691: \begin{sloppypar}

692: Clearly the same reasoning guarantees the non-overlap of

693: the side faces $S_{x^-},S_{y^-},S_{y^+}$ with their adjacent

694: quadrilateral strips.

695: See Figures~\figref{16x16_unf} and~\figref{32x32_unf} for further

696: examples of cap unfoldings.

697: \end{sloppypar}

698: 

699: That the radial monotonicty restriction is sometimes necessary is

700: established by the overlapping unfolding of a convex cap that has such a

701: plummet in $z$ values near one corner of $B$ that the cap curves fail to be

702: radially convex there.

703: See Figure~\figref{overlap_r2}.

704: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

705: \begin{figure}[htbp]

706: \centering

707: \includegraphics[width=0.9\linewidth]{Figures/overlap_r2}

708: \caption{Overlap caused by a sharply convex cap.

709: (Side faces are not shown.)

710: }

711: \figlab{overlap_r2}

712: \end{figure}

713: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

714: 

715: \section{Global Non-Overlap}

716: We have now established that all overlap is avoided locally, but there

717: still remains the issue of possible ``global overlap,'' overlap of widely separated

718: portions of the surface.

719: It is easy to add rays to the ends of each $x$-strip, and

720: rays on the sides of $S_{y^-}$ and $S_{y^+}$, that shoot to

721: infinity without crossing, as in

722: Figure~\figref{unf_rays}.

723: The $x$-strip rays can be thought of as edges of extensions of $c_x(y)$

724: with long ``tail'' quadrilaterals extending toward $z \rightarrow -\infty$,

725: which can be seen to preserve radial monotonicity.

726: These rays partition the plane and separate the pieces of the unfolding from

727: one another.

728: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

729: \begin{figure}[htbp]

730: \centering

731: \includegraphics[width=0.5\linewidth]{Figures/unf_rays}

732: \caption{Extension by rays establishes global non-overlap.}

733: \figlab{unf_rays}

734: \end{figure}

735: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

736: 

737: \section{Future Work}

738: The class of shapes for which the presented unfolding algorithm

739: guarantees non-overlap is narrow and of no interest except as

740: a possible stepping stone to more general shapes.

741: Next steps

742: (in order of perceived difficulty)

743: include attempting to

744: extend the algorithm to:

745: \begin{enumerate}

746: \squeezelist

747: \item remove the radially monontone assumption;

748: \item remove the assumption that $B$ is a rectangle;

749: \item include shapes with both an upper and a lower convex cap surface;

750: \item remove the assumption that all faces are quadrilaterals.

751: \end{enumerate}

752: 

753: %see Figure~\figref{16x16_unf}.

754: %see Figure~\figref{32x32_unf}.

755: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

756: \begin{figure}[htbp]

757: \begin{minipage}[b]{0.48\linewidth}

758: \centering

759: \includegraphics[width=\linewidth]{Figures/16x16_unf}

760: \caption{$16 \times 16$ cap unfolding.}

761: \figlab{16x16_unf}

762: \end{minipage}%

763: \hspace{0.04\linewidth}%

764: \begin{minipage}[b]{0.48\linewidth}

765: \centering

766: \includegraphics[width=\linewidth]{Figures/32x32_unf}

767: \caption{$32 \times 32$ cap unfolding.}

768: \figlab{32x32_unf}

769: \end{minipage}%

770: \end{figure}

771: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

772: 

773: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

774: %\small

775: %\baselineskip=0.9\baselineskip

776: % Decrease the space between bibliography items.

777: %\let\realbibitem=\bibitem

778: %\def\bibitem{\par \vspace{-0.6ex}\realbibitem}

779: 

780: \bibliographystyle{alpha}

781: %\bibliographystyle{plain}

782: \bibliography{/home/orourke/bib/geom/geom}

783: \end{document}

784: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

785: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

786: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

787: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

788: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

789: 

790: 

791: see Figure~\figref{X}.

792: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

793: \begin{figure}[htbp]

794: \centering

795: \includegraphics[width=0.75\linewidth]{Figures/X}

796: \caption{Y.}

797: \figlab{X}

798: \end{figure}

799: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

800: 

801: 

802: \begin{lemma}

803: \lemlab{x}

804: \end{lemma}

805: \begin{proof}

806: \end{proof}

807: 

808: 

809: 

810: 

811: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

812: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

813: %Old proofs, correct but plodding :-/.  Left here for nostalgia and copy&paste.

814: \hide{

815: Say a directed edge of $C$ is \emph{uphill} its

816: second endpoint is strictly above its first endpoint.

817: 

818: \begin{lemma}[Uphill]

819: For a lattice quadrilateral cap $C$,

820: the edge above

821: $(x,y)$ to $(x+1,y)$ is uphill

822: if and only iff

823: $(x,1)$ to $(x+1,1)$ is uphill.

824: And the same claim holds when the roles of $x$ and $y$ are interchanged.

825: \lemlab{uphill}

826: \end{lemma}

827: \begin{proof}

828: Consider the lattice cell at the origin,

829: $(0,0),(1,0),(1,1),(0,1)$, with

830: quadrilateral corner heights 

831: $z_0=0,z_1,z_2,z_3$ respectively

832: (as in Figure~\figref{alpha_z1}(a) below).

833: We know from Eq.~\eqref{z2} that 

834: \begin{equation}

835: z_2 = z_1 + z_3

836: \eqlab{z1}

837: \end{equation}

838: The edge $(0,0),(1,0)$ is uphill only if $z_1 > 0$.

839: The edge $(0,1),(1,1)$ is uphill only if $z_2 > z_3$,

840: which occurs exactly when $z_1 > 0$.  This establishes

841: the claim of the lemma for the edges of one particular cell.

842: But this reasoning works independent of which cell is used,

843: so it holds true of any quadrilateral of $C$.

844: Now the full claim of the lemma follows by iterating

845: the reasoning (or applying induction):

846: $(x,1)$ to $(x+1,1)$ is uphill iff

847: $(x,2)$ to $(x+1,2)$ is uphill iff

848: $(x,3)$ to $(x+1,3)$ is uphill, etc.

849: \end{proof}

850: }%hide

851: 

852: \hide{

853: We need a lemma slightly more nuanced than the Uphill Lemma:

854: 

855: \begin{lemma}[Cap Convexity]

856: For a lattice quadrilateral cap $C$,

857: the chain above the three consecutive lattice points

858: $(x,y),(x+1,y),(x+2,y)$ is convex

859: iff

860: the chain above

861: $(x,1),(x+1,1),(x+2,1)$ is convex.

862: And the same claim holds when the roles of $x$ and $y$ are interchanged.

863: \lemlab{convex.cap}

864: \end{lemma}

865: \begin{proof}

866: The proof parallels that of the uphill lemma.

867: Without loss of generality, we translate $(x+1,1,z)$ to the origin $(0,0,0)$

868: to ease the algebra.

869: Consider again the lattice cell at the origin,

870: $(0,0),(1,0),(1,1),(0,1)$, with

871: quadrilateral corner heights 

872: $0,z_1,z_2,z_3$ respectively

873: Eq.~\eqref{z1} then holds, as in Lemma~\lemref{uphill}.

874: %\begin{equation}

875: %z_2 = z_1 + z_3

876: %\eqlab{z1}

877: %\end{equation}

878: Now extend with a cell at its left,

879: $(-1,0),(0,0),(0,1),(-1,1)$, with corner heights

880: $z'_1,z_0=0,z_3,z'_2$ respectively.

881: Again Eq.~\eqref{z2} yields

882: \begin{equation}

883: z'_2 = z'_1 + z_3

884: \eqlab{z11}

885: \end{equation}

886: Now the front chain over $(-1,0),(0,0),(1,0)$ is

887: convex iff

888: \begin{equation}

889: z_1 \le -z'_1

890: \eqlab{front.convex}

891: \end{equation}

892: The back chain over $(-1,1),(0,1),(1,1)$

893: is convex iff

894: \begin{equation}

895: z_2 \le z_3 +(z_3-z'_2)

896: \eqlab{back.convex}

897: \end{equation}

898: Solving for $z_1$ and $z'_1$ in Eqs.~\eqref{z1} and~\eqref{z11}

899: and substituting into Eq.~\eqref{front.convex}

900: yields Eq.~\eqref{back.convex}.

901: 

902: This establishes

903: the claim of the lemma for one particular chain centered over $(0,0)$.

904: But this reasoning works independent of which lattice point is used.

905: so it holds anywhere on $C$.

906: Now the full claim of the lemma follows by iterating

907: the reasoning: %(or applying induction):

908: the chain centered over

909: $(x,1)$ is convex iff the chain over

910: $(x,2)$ is convex iff the chain over

911: $(x,3)$ is convex, etc.

912: \end{proof}

913: }%hide

914: