1: \documentclass[11pt]{article}

2: %\documentclass{llncs}

3: \usepackage{graphicx}

4: \usepackage{amsfonts}

5: \usepackage{amssymb}

6: \usepackage{url}

7: 

8: 

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

10: \newcommand\R{\mathbb{R}}

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

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

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

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

15: \newcommand{\ABox}{

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

17: }

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

19: 

20: \topmargin 0in                  % LaTeX adds an inch

21: \headheight 0pt                 % no headers

22: \headsep 0in                    % no headers

23: \textheight 9in                 % i.e., bottom margin = 1in

24: \textwidth 6.5in

25: \oddsidemargin 0in              % (8.5-textwidth)/2 = 1in

26:                                 % LaTeX adds an inch

27: 

28: \def\C{{\cal C}}

29: 

30: %From Erik Demaine:

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

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

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

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

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

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

37: {\makeatletter

38:  \gdef\xxxmark{%

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

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

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

42:      \else

43:        xxx % notice trailing space

44:      \fi

45:    \else

46:      xxx % notice trailing space

47:    \fi}

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

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

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

51:  % This turns them off:

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

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

54: }

55: 

56: \begin{document}

57: 

58: \title{Grid Vertex-Unfolding Orthogonal Polyhedra

59: \thanks{This is a significant revision of the preliminary version that

60: appeared in~\protect\cite{dfo-gvuop-06}.}

61: %JOR: I moved the correction notification to Sec. 6.1.

62: %Proc. of

63: %the 23rd Int. Symposium on Theoretical Aspects of Computer Science,

64: %2006

65: %The preliminary version contains an

66: %incomplete description of the unfolding tree

67: %(section~\ref{sec:spanning.tree}).}

68: }

69: 

70: %

71: %\titlerunning{Unfolding Orthogonal Polyhedra}

72: %

73: %\author{Mirela Damian\inst{1} \and Robin Flatland\inst{2} \and Joseph O'Rourke\inst{3}}

74: %

75: %\authorrunning{Mirela Damian et al.}   % abbreviated author list (for running head)

76: %

77: %%%% modified list of authors for the TOC (add the affiliations)

78: %\tocauthor{Mirela Damian (Villanova University),

79: %Robin Flatland (Siena College),

80: %Joseph O'Rourke (Smith College)}

81: %

82: %\institute{Comput. Sci., Villanova University, Villanova,

83: %    PA 19085, USA.\\

84: %%\email{mirela.damian@villanova.edu}

85: %\\ WWW home page:

86: %\texttt{http://www.csc.villanova.edu/~mdamian/}

87: %\and

88: %Comput. Sci., Siena College, Loudonville, NY 12211, USA.\\

89: %\email{flatland@siena.edu} %,\\ WWW home page:

90: %%\texttt{http://www.cs.siena.edu/~flatland/}

91: %\and

92: %Comput. Sci., Smith College, Northampton, MA 01063, USA.\\

93: %\email{orourke@cs.smith.edu}} %,\\ WWW home page:

94: %%\texttt{http://www.cs.smith.edu/~orourke/}}

95: 

96: \author{

97: Mirela Damian%

98:    \thanks{Dept. Comput. Sci., Villanova Univ., Villanova,

99:     PA 19085, USA.

100:    \protect\url{mirela.damian@villanova.edu}.}

101: \and

102: Robin Flatland%

103:    \thanks{Dept. Comput. Sci., Siena College, Loudonville, NY 12211, USA.

104:     \protect\url{flatland@siena.edu}.}

105: \and

106: Joseph O'Rourke%

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

108:       01063, USA.

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

110:        Supported by NSF

111:        %Distinguished Teaching Scholars

112:        award

113:        DUE-0123154.}

114: }

115: %\date{}

116: 

117: \maketitle              % typeset the title of the contribution

118: 

119: \begin{abstract}

120: An {\em edge-unfolding} of a polyhedron is produced by cutting along

121: edges and flattening the faces to a {\em net}, a connected planar

122: piece with no overlaps. A {\em grid unfolding} allows additional

123: cuts along grid edges induced by coordinate planes passing through

124: every vertex. A {\em vertex-unfolding} permits faces in the net to

125: be connected at single vertices, not necessarily along edges.  We

126: show that any orthogonal polyhedra of genus zero has a grid

127: vertex-unfolding. (There are orthogonal polyhedra that cannot be

128: vertex-unfolded, so some type of ``gridding'' of the faces is

129: necessary.) For any orthogonal polyhedron $P$ with $n$ vertices, we

130: describe an algorithm that vertex-unfolds $P$ in $O(n^2)$ time.

131: Enroute to explaining this algorithm, we present a simpler

132: vertex-unfolding algorithm that requires a $3 \times 1$ refinement

133: of the vertex grid.

134: \end{abstract}

135: 

136: \paragraph{Keywords:} vertex-unfolding, grid unfolding, orthogonal polyhedra,

137: genus-zero.

138: 

139: \section{Introduction}

140: Two unfolding problems have remained unsolved for many

141: years~\cite{do-sfucg-05}: (1)~Can every convex polyhedron be

142: edge-unfolded? (2)~Can every polyhedron be unfolded? An

143: \emph{unfolding} of a 3D object is an isometric mapping of its

144: surface to a single, connected planar piece, the ``net'' for the

145: object, that avoids overlap. An \emph{edge-unfolding} achieves the

146: unfolding by cutting edges of a polyhedron, whereas a \emph{general

147: unfolding} places no restriction on the cuts.

148: %

149: A net representation of a polyhedron finds use in a variety of

150: applications~\cite{o-fucg-00} --- from flattening monkey

151: brains~\cite{ssw-nsgmm-89} to manufacturing from sheet

152: metal~\cite{w-mddsmp-97} to low-distortion texture

153: mapping~\cite{thcm-pm-04}.

154: %In manufacturing parts from sheet metal

155: %for example, a 3D part is approximated as a polyhedron and its

156: %surface is mapped to a collection of 2D flat patterns. Each pattern

157: %is cut from a sheet of metal and folded by a bending

158: %machine~\cite{gbkk-appsmbo-98}, and the resulting pieces assembled

159: %to form the final part.

160: 

161: It is known that some nonconvex polyhedra cannot be unfolded without

162: overlap with cuts along edges. However, no example is known of a

163: nonconvex polyhedron that cannot be unfolded with unrestricted cuts.

164: Advances on these difficult problems have been made by specializing

165: the class of polyhedra, or easing the stringency of the unfolding

166: criteria. On one hand, it was established

167: in~\cite{bddloorw-uscop-98} that certain subclasses of

168: \emph{orthogonal polyhedra} --- those whose faces meet at angles

169: that are multiples of $90^\circ$ --- have an unfolding. In

170: particular, the class of \emph{orthostacks}, stacks of extruded

171: orthogonal polygons, was proven to have an unfolding (but not an

172: edge-unfolding). On the other hand, loosening the criteria of what

173: constitutes a net to permit connection through points/vertices, the

174: so-called \emph{vertex-unfoldings}, led to an algorithm to

175: vertex-unfold any triangulated manifold~\cite{deeho-vusm-03} (and

176: indeed, any simplicial manifold in higher dimensions).

177: A vertex unfolding maps the surface to a single, connected piece $P$

178: in the plane, but $P$ may have ``cut vertices'' whose removal

179: disconnect $P$.

180: 

181: A second loosening of the criteria is the notion of grid unfoldings,

182: which are especially natural for orthogonal polyhedra. A \emph{grid

183: unfolding} adds edges to the surface by intersecting the polyhedron

184: with planes parallel to Cartesian coordinate planes through every

185: vertex. The two approaches were recently married

186: in~\cite{dil-gvuo-04}, which established that any orthostack may be

187: grid vertex-unfolded. For orthogonal polyhedra, a grid unfolding is

188: a natural median between edge-unfoldings and unrestricted

189: unfoldings.

190: 

191: Our main result is that any orthogonal polyhedron, without shape

192: restriction except that its surface be homeomorphic to a sphere, has

193: a grid vertex-unfolding. We present an algorithm that grid

194: vertex-unfolds any orthogonal polyhedron with $n$ vertices in

195: $O(n^2)$ time. We also present, along the way, a simpler algorithm

196: for $3 \times 1$ \emph{refinement} unfolding, a weakening of grid

197: unfolding that we define below. We believe that the techniques in

198: our algorithms may help show that all orthogonal polyhedra can be

199: grid edge-unfolded.

200: 

201: \section{Definitions}

202: \label{sec:defs}

203: A $k_1 \times k_2$ \emph{refinement} of a surface~\cite{do-op04-05}

204: partitions each face into a $k_1 \times k_2$ grid of faces. We will

205: consider refinements of grid unfoldings, with the convention that a

206: $1 \times 1$ refinement is an unrefined grid unfolding.

207: 

208: We distinguish between a \emph{strict net}, in which the net

209: boundary does not self-touch, and a \emph{net} for which the

210: boundary may touch, but no interior points overlap. The latter

211: corresponds to the physical model of cutting out the net from a

212: sheet of paper, with perhaps some cuts representing \emph{edge

213: overlap}, and this is the model we use in this paper.

214: %

215: We also insist as part of the definition of a vertex-unfolding,

216: again keeping in spirit with the physical model, that the unfolding

217: ``path'' never self-crosses on the surface in the following sense.

218: If $(A,B,C,D)$ are four faces incident in that cyclic order to a

219: common vertex $v$, then the net does not include both the

220: connections $AvC$ and $BvD$.\footnote{

221:    This was not part of the original definition in~\cite{deeho-vusm-03} but was

222:    achieved by those unfoldings.

223: }

224: 

225: We use the following notation to describe the six type of faces of

226: an orthogonal polyhedron, depending on the direction in which the

227: outward normal points: {\em front}: $-y$; {\em back}: $+y$; {\em

228: left}: $-x$; {\em right}: $+x$; {\em bottom}: $-z$; {\em top}: $+z$.

229: We take the $z$-axis to define the vertical direction;

230: \emph{vertical} faces are parallel to the $xz$-plane. Directions

231: clockwise (cw), and counterclockwise (ccw) are defined from the

232: perspective of a viewer positioned at $y = -\infty$. We distinguish

233: between an original vertex of the polyhedron, which we call a

234: \emph{corner vertex} or just a \emph{vertex}, and a

235: \emph{gridpoint}, a vertex of the grid (which might be an original

236: vertex). A {\em gridedge} is an edge segment with both endpoint

237: gridpoints, and a {\em gridface} is a face of the grid bounded by

238: gridedges.

239: 

240: Let $O$ be a solid orthogonal polyhedron with the surface

241: homeomorphic to a sphere (i.e, genus zero). Let $Y_i$ be the plane

242: $y=y_i$ orthogonal to the $y$-axis. Let $Y_0, Y_1, \ldots, Y_i,

243: \ldots$ be a finite sequence of parallel planes passing through

244: every vertex of $O$, with $y_0 < y_1 < \cdots < y_i < \cdots$. We

245: define \emph{layer $i$} to be the portion of $O$ between planes

246: $Y_i$ and $Y_{i+1}$. Observe that a layer may include a collection

247: of disjoint connected components of $O$; we call each such component

248: a {\em slab}. A surface piece that surrounds a slab is called a {\em

249: band}. Referring to Fig.~\ref{fig:defs}a, layer $0$, $1$ and $2$

250: each contain one slab (with outer bands $A$, $B$ and $D$,

251: respectively). Note that each slab is bounded by an outer (surface)

252: band, but it may also contain inner bands, bounding holes. Outer

253: bands are called \emph{protrusions} and inner bands are called

254: \emph{dents} ($C$ in Fig.~\ref{fig:defs}a). In other words, band $A$

255: is a \emph{protrusion} if a traversal of the rim of $A$ in $Y_i$,

256: ccw from the viewpoint of $y=-\infty$, has the interior of $O$ to

257: the left of $A$, and a \emph{dent} if this traversal has the

258: interior of $O$ to the right.

259: 

260: For fixed $i$, define $P = \partial O \cap Y_i$ as the portion of

261: the surface of $O$ lying in plane $Y_i$. $P^+$ is the portion of $P$

262: with normal in the direction $+y$ (composed of back faces), and

263: $P^-$ the portion with normal in the direction $-y$ (composed of

264: front faces). By convention, band points in $P$ that are not

265: incident to either front or back faces (e.g., when one band aligns

266: with another), belong to both $P^+$ and $P^-$. Thus $P = P^+ \cup

267: P^-$.

268: %Note that it may be that $P^+$ and $P^-$ are not disjoint,

269: %they can share points (but not edges).

270: 

271: For a band $A$, Let $R_i(A) = A \cap Y_i$ be the polygon in $Y_i$

272: determined by the rim of band $A$, and $r_i(A)$ the closed region of

273: $Y_i$ whose boundary is $R_i(A)$. For any two bands $A$ and $B$, let

274: $r_i(AB) = r_i(A) \cap r_i (B)$ and let $R_i(AB)$ be the boundary of

275: $r_i(AB)$.

276: 

277: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

278: \begin{figure}[htbp]

279: \centering

280: \begin{tabular}{c@{\hspace{0.2\linewidth}}c}

281: \includegraphics[width=0.36\linewidth]{Figures/defs.1.eps} &

282: \raisebox{2em}{\includegraphics[width=0.08\linewidth]{Figures/defs.2.eps}}

283: \end{tabular}

284: \caption{Definitions. (a) Shaded connected pieces are bands; $A$,

285: $B$ and $D$ are protrusions; $C$ is a dent; $r_2(CB)$ coincides with

286: the back face of $C$; $R_2(DB)$ is marked in dashed lines. (b) The

287: adjacency structure of bands is a tree. \label{fig:defs} }

288: \end{figure}

289: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

290: 

291: \section{Dents vs. Protrusions}

292: We observe that dents may be treated exactly the same as protrusions

293: with respect to unfolding, because an unfolding of a $2$-manifold to

294: another surface (in our case, a plane) depends only on the intrinsic

295: geometry of the surface, and not on how it is embedded in $\R^3$.

296: Note that we are only concerned with the final unfolded ``flat

297: state''~\cite{do-sfucg-05}, and not with possible intersections

298: during a continuous sequence of partially unfolded intermediate

299: states. Our unfolding algorithm relies solely on the amount of

300: surface material surrounding each point: the cyclic ordering of the

301: faces incident to a vertex, and the pair of faces sharing an edge.

302: All these local relationships remain unchanged if we conceptually

303: ``pop-out'' dents to become protrusions, i.e., a ``Flatland''

304: creature living in the surface could not tell the difference; nor

305: can our algorithm. We note that the popping-out is conceptual only,

306: for it could produce self-intersecting objects. Also dents are

307: gridded independently of the rest of the object, so that it would

308: not matter whether they are popped out or not.

309: %The distinction between dents and protrusions

310: %only affects the case structure of the interaction between bands

311: %in Sec.~\ref{sec:Gb.connected}.

312: %\xxx[JOR]{Is this true?  We need to say something here to change the

313: %``it doesn't matter'' claim...}

314: %

315: %This shows that our algorithm works on a wider class of objects than

316: %the orthogonal polyhedra, an observation we do not pursue.

317: %======================================================================

318: 

319: Although the dent/protrusion distinction is irrelevant to the

320: unfolding, the interrelationships between dents and protrusions

321: touching a particular $Y_i$ do depend on this distinction. To cite

322: just the simplest example, there cannot be two nested protrusions to

323: the same side of $Y_i$, but a protrusion could have a dent in it to

324: the same side of $Y_i$ (e.g., protrusion $B$ encloses dent $C$ to

325: the same side of $Y_1$ in Fig.~\ref{fig:defs}). These relationships

326: are crucial to the connectivity of the band graph $G_b$, discussed

327: in Sec.~\ref{sec:Gb.connected}.

328: 

329: \section{Overview}

330: \label{sec:overview} The two algorithms we present share a common

331: central structure, with the second achieving a stronger result; both

332: are vertex-unfoldings that use orthogonal cuts only. We note that it

333: is the restriction to orthogonal cuts that makes the

334: vertex-unfolding problem difficult: if arbitrary cuts are allowed,

335: then a general vertex-unfolding can be obtained by simply

336: triangulating each face and applying the algorithm

337: from~\cite{deeho-vusm-03}.

338: 

339: The ($3 \times 1$)-algorithm unfolds any genus-0 orthogonal

340: polyhedron that has been refined in one direction 3-fold. The bands

341: themselves are never split (unlike in~\cite{bddloorw-uscop-98}). The

342: algorithm is simple. The ($1 \times 1$)-algorithm also unfolds any

343: genus-0 orthogonal polyhedron, but this time achieving a grid

344: vertex-unfolding, i.e., without refinement. This algorithm is more

345: delicate, with several cases not present in the ($3 \times

346: 1$)-algorithm that need careful detailing. Clearly this latter

347: algorithm is stronger, and we vary the detail of presentation to

348: favor it. The overall structure of the two algorithms is the same:

349: \begin{enumerate}

350: \squeezelist

351: \item A band ``unfolding tree'' $T_U$ is constructed by shooting rays vertically

352: from the top of bands. The root of $T_U$ is a {\em frontmost} band

353: (of smallest $y$-coordinate), with ties broken arbitrarily.

354: \item A forward and return \emph{connecting} path of vertical faces

355: is identified, each of which connects a parent band to a child band

356: in $T_U$.

357: \item Each band is unfolded horizontally as a unit, but interrupted

358: when a connecting path to a

359: child is encountered.  The parent band unfolding is suspended at that

360: point, and the child band is unfolded recursively.

361: \item The vertical front and back faces of each slab are partitioned

362: according to an illumination model, with variations for the more

363: complex ($1 \times 1$)-algorithm. These vertical faces are attached

364: below and above appropriate horizontal sections of the band

365: unfolding.

366: \end{enumerate}

367: The final unfolding lays out all bands horizontally, with the

368: vertical faces hanging below and above the bands.  Non-overlap is

369: guaranteed by this strict two-direction structure.

370: 

371: Although our result is a broadening of that in~\cite{dil-gvuo-04}

372: from orthostacks to all orthogonal polyhedra, we found it necessary

373: to employ techniques different from those used in that work. The

374: main reason is that, in an orthostack, the adjacency structure of

375: bands yields a path, which allows the unfolding to proceed from one

376: band to the next along this path, never needing to return. In an

377: orthogonal polyhedron, the adjacency structure of bands yields a

378: tree (cf. Fig.~\ref{fig:defs}b). Thus unfolding band-by-band leads

379: to a tree traversal, which requires traversing each arc of the tree

380: in both directions. It is this aspect which we consider our main

381: novelty, and which leads us to hope for an extension to

382: edge-unfoldings as well.

383: 

384: %======================================================================

385: \section{$(3 \times 1)$-Algorithm}

386: 

387: %Under the assumption that each front, back, top and bottom face of

388: %$O$ has been partitioned into a $3 \times 1$ grid of faces, we

389: %discuss in detail the unfolding steps listed in

390: %Sec.~\ref{sec:overview}.

391: 

392: \subsection{Computing the Unfolding Tree $T_U$}

393: \label{sec:3x1tree}

394: 

395: %We first describe finding an unfolding tree

396: %$T_U$ that spans all bands in $O$.

397: 

398: Define a {\em z-beam} to be a vertical rectangle on the surface of

399: $O$ connecting two band rims whose top and bottom edges are

400: gridedges. In the degenerate case, a $z$-beam has height zero and

401: connects two rims along a section where they coincide. We say that

402: two bands $b_1$ and $b_2$ are {\em z-visible} if there exists a

403: $z$-beam connecting a gridedge of $b_1$ to a gridedge of $b_2$.

404: There can be many $z$-beams connecting two bands, so for each pair

405: of bands we select a representative $z$-beam of minimal (vertical)

406: height. Let $G$ be the graph that contains a node for each band of

407: $O$ and an arc for each pair of $z$-visible bands. It easily follows

408: from the connectedness of the surface of $O$ that $G$ is connected.

409: Let the unfolding tree $T_U$ be any spanning tree of $G$, with the

410: root selected arbitrarily from among all bands adjacent to $Y_0$.

411: 

412: Applying the 3x1 refinement partitions each front, back, top and

413: bottom face of $O$ into a $3 \times 1$ grid of faces. This

414: partitions the top and bottom edges of each $z$-beam into three

415: refined gridedges and divides the beam itself into three vertical

416: columns of refined gridfaces. For a band $B$ in $T_U$ with parent

417: $A$, let $e$ be the gridedge on $B$'s rim where the $z$-beam from

418: $A$ attaches. We define the {\em pivot point} $x_b$ to be the

419: $\frac{1}{3}$-point of $e$ (or, in circumstances to be explained

420: below, the $\frac{2}{3}$-point), and so it coincides with a point of

421: the $3 \times 1$-refined grid. The unfolding of $O$ will follow the

422: connecting vertical ray that extends from $x_b$ on $B$ to $A$. Note

423: that if $e$ belongs to both $A$ and $B$, then the ray connecting $A$

424: and $B$ degenerates to a point. To either side of a connecting ray

425: we have two \emph{connecting paths} of vertical faces, the

426: \emph{forward} and \emph{return} path. In

427: Fig.~\ref{fig:3x1.three.boxes}a, these connecting paths are the

428: shaded strips on the front face of $A$.

429: 

430: 

431: \subsection{Unfolding Bands into a Net}

432: \label{sec:3x1band}

433: 

434: Starting at a frontmost \emph{root band}, each band is unfolded as a

435: conceptual unit, but interrupted by the connecting rays incident to

436: it from its front and back faces. In Fig.~\ref{fig:3x1.three.boxes},

437: band $A$ is unfolded as a rectangle, but interrupted at the rays

438: connecting to (front children) $B$, $C$ and (back child) $B'$. At

439: each such ray the parent band unfolding is suspended, the unfolding

440: follows the forward connecting path to the child, the child band is

441: recursively unfolded, then the unfolding returns along the return

442: connecting path back to the parent, resuming the parent band

443: unfolding from the point it left off.

444: 

445: Fig.~\ref{fig:3x1.three.boxes} illustrates this unfolding algorithm.

446: The cw unfolding of $A$, laid out horizontal to the right, is

447: interrupted to traverse the forward path down to $B$, and $B$ is

448: then unfolded as a rectangle (composed of its contiguous faces). The

449: base $x_b$ of the connecting ray is called a \emph{pivot point}

450: because the ccw unfolding of $B$ is rotated $180^\circ$ ccw about

451: $x_b$ so that the unfolding of $B$ is also horizontal to the right.

452: It is only here that we use point-connections that render the

453: unfolding a vertex-unfolding. The unfolding of $B$ proceeds ccw back

454: to $x_b$, crosses over $A$ to unfold $B'$, then a cw rotation by

455: $180^\circ$ around the second image of pivot $x_b'$ orients the

456: return path to $A$ so that the unfolding of $A$ continues horizontal

457: to the right. Note that the unfolding of $C$ is itself interrupted

458: to unfold child $D$. Also note that there is edge overlap in the

459: unfolding at each of the pivot points.

460: 

461: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

462: \begin{figure}[htbp]

463: \centering

464: \includegraphics[width=0.9\linewidth]{Figures/3x1.three.boxes.eps}\\

465: \vspace{-0.08in} \caption{ (a)~Orthogonal polyhedron. (b)~Unfolding

466: tree $T_U$. (c)~Unfolding of bands and front (hachured) face pieces.

467: Vertex connection through the pivots points $x_b$, $x_{b'}$, $x_c$,

468: $x_d$ is shown exaggerated for clarity. }

469: \label{fig:3x1.three.boxes}

470: \end{figure}

471: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

472: 

473: The reason for the $3 \times 1$ refinement is that the upper edge

474: $e'$ of the back child band $B'$ has the same $(x,z)$-coordinates as

475: the upper edge $e$ of $B$ on the front face. In this case, the faces

476: of band $A$ induced by the connecting paths to $B$ would be

477: ``overutilized'' if there were only two.  Let $a_1, a_2, a_3$ be the

478: three faces of $A$ induced by the $3 \times 1$ refinement of the

479: connecting path to $B$, as in Fig.~\ref{fig:3x1.three.boxes}. Then

480: the unfolding path winds around $A$ to $a_1$, follows the forward

481: connecting path to $B$, returns along the return connecting path to

482: $a_2$, crosses over $A$ and unfolds $B'$ on the back face, with the

483: return path now joining to $a_3$, at which point the unfolding of

484: $A$ resumes.  In this case, the pivot point $x_{b'}$ for $B'$ is the

485: $\frac{2}{3}$-point of $e'$. Other such conflicts are resolved

486: similarly. It is now easy to see that the

487: %unfolding of the bands and

488: %the connecting-path faces

489: resulted net has the general form illustrated in

490: Fig.~\ref{fig:3x1.three.boxes}b:

491: \begin{enumerate}

492: \squeezelist

493: \item The faces of each band fall within a horizontal rectangle whose height is the band

494: width.

495: \item These band rectangles are joined by vertical connecting paths on either side,

496: connecting through pivot points.

497: \item The strip of the plane above and below each band face that is not incident to a

498: connecting path, is empty.

499: \item The net is therefore an orthogonal polygon monotone with respect to

500: the horizontal.

501: \end{enumerate}

502: 

503: \subsection{Attaching Front and Back Faces to the Net}

504: \label{sec:3x1front.back}

505: 

506: Finally, we ``hang'' front and back faces from the bands as follows.

507: The front face of each band $A$ is partitioned by imagining $A$ to

508: illuminate downward lightrays from the rim in the front face. The

509: pieces that are illuminated are then hung vertically downward from

510: the horizontal unfolding of the $A$ band. The portions unilluminated

511: will be attached to the obscuring bands.

512: 

513: In the example in Fig.~\ref{fig:3x1.three.boxes}, this illumination

514: model partitions the front face of $A$ into three pieces (the

515: striped pieces in Fig.~\ref{fig:3x1.three.boxes}b). These three

516: pieces are attached under $A$; the portions of the front face

517: obscured by $B$ but illuminated downward by $B$ are hung beneath the

518: unfolding of $B$ (not shown in the figure), and so on. Because the

519: vertical illumination model produces vertical strips, and because

520: the strips above and below the band unfoldings are empty, there is

521: always room to hang the partitioned front face. Thus, any orthogonal

522: polygon may be vertex-unfolded with a $3 \times 1$ refinement of the

523: vertex grid.

524: 

525: \medskip

526: \noindent Although we believe this algorithm can be improved to $2

527: \times 1$ refinement, the complications needed to achieve this are

528: similar to what is needed to avoid refinement entirely, so we

529: instead turn directly to $1 \times 1$ refinement.

530: 

531: \section{$(1 \times 1)$-Algorithm}

532: 

533: Although the $(1 \times 1)$-algorithm follows the same general

534: outline as the $(3 \times 1)$-Algorithm, there are significant

535: complications, which we outline before detailing. First, without the

536: refinement of $z$-beams into three strips to allow avoidance of

537: conflicts on opposite sides of a slab (e.g., $B$ and $B'$ in

538: Fig.~\ref{fig:3x1.three.boxes}a), we found it necessary to replace

539: the $z$-beams by a pair of $z$-rays that are in some sense the

540: boundary edges of a $z$-beam. Selecting two rays per band permits a

541: 2-coloring algorithm (Theorem~\ref{thm:2-color}) to identify rays

542: that avoid conflicts. Generating the ray-pairs

543: (Sec.~\ref{sec:ray-pairs}) requires care to ensure that the band

544: graph $G_b$ is connected (Sec.~\ref{sec:Gb.connected}). This graph,

545: and the 2-coloring, lead to an unfolding tree $T_U$

546: (Sec.~\ref{sec:spanning.tree}). From here on, there are fewer

547: significant differences compared to the $(3 \times 1)$-Algorithm.

548: Without the luxury of refinement, there is more need to share

549: vertical paths on the vertical face of a slab

550: % and to take care with

551: % overlapping forward and return paths

552: (Fig.~\ref{fig:return}).

553: Finally, the vertical connecting paths obscure the illumination of

554: some grid faces, which must be attached to the connecting paths. We

555: now present the details, in this order:

556: 

557: %This algorithm follows a similar sequence of steps as the $3 \times

558: %1$ algorithm, but in a slightly different order and with each step

559: %involving more complex techniques discussed in detail in subsequent

560: %sections:

561: 

562: \vspace{0.1in} \hrule \hfill \vspace{-0.1in}

563: \begin{tabbing}

564: ....\=...........\=........\kill

565:        \> 1. Select Pivot Points (Sec.~\ref{sec:pivots}) via \\

566:        \>      \>a. Ray-Pair Generation (Sec.~\ref{sec:ray-pairs}) \\

567:        \>      \>b. Ray Graph (Sec.~\ref{sec:ray.graph}) \\

568:        \> 2. Construct $T_U$ (Sec.~\ref{sec:spanning.tree}) \\

569:        \> 3. Select Connecting Paths (Sec.~\ref{sec:paths}) \\

570:        \> 4. Determine Unfolding Directions (Sec.~\ref{sec:unfdir}) \\

571:        \> 5. Recurse: \\

572:        \>      \> a. Unfold Bands into a Net (Sec.~\ref{sec:unfolding.bands}) \\

573:        \>      \> b. Attach Front and Back Faces to the Net (Sec.~\ref{sec:faces})

574: \end{tabbing}

575: \hrule \hfill

576: 

577: 

578: %In the following we discuss each step in detail.

579: \subsection{Selecting Pivot Points} \label{sec:pivots}

580: The pivot $x_a$ for a band $A$ is the gridpoint of $A$ where the

581: unfolding of $A$ starts and ends. The $y$-edge of $A$ incident to

582: $x_a$ is the first edge of $A$ that is cut to unfold $A$.

583: 

584: Let $A$ be an arbitrary band delimited by planes $Y_i$ and

585: $Y_{i+1}$. Say that two gridpoints $u \in Y_i$ and $w \in Y_{i+1}$

586: are {\em in conflict} if the upward rays emerging from $u$ and $w$

587: hit the endpoints of the same $y$-edge of $A$; otherwise, $u$ and

588: $v$ are {\em conflict-free}. If $u$ lies either on a vertical edge,

589: or on a vertically extreme horizontal edge, then the ray at $u$

590: degenerates to $u$ itself.

591: 

592: Our goal is to select conflict-free pivots for all bands in $T_U$,

593: which will help us avoid later competition over the use of certain

594: faces in the unfolding, an issue that will become clear in

595: Sec.~\ref{sec:unfolding.bands}. Selecting these pivots is the most

596: delicate aspect of the $(1 \times 1)$-algorithm. Ultimately, we

597: represent pivoting conflicts in the form of a graph $G_r$

598: (Sec.~\ref{sec:ray.graph}), from which $T_U$ will be derived.

599: 

600: \subsubsection{Ray-Pair Generation}

601: \label{sec:ray-pairs} In order to avoid pivoting conflicts, for each

602: band we will need two choices for its connecting ray. Thus the

603: algorithm generates the rays in pairs. Because there is no

604: refinement, the two rays originate at grid points on the same band,

605: but they may terminate on different bands. A simple example is shown

606: in Figure~\ref{fig:face}a, where the ray pair originating on band

607: $D$ hits two different bands, $B$ and $C$. This example also

608: suggests that one cannot consider ray pairs connecting pairs of

609: bands, as in the $(3 \times 1)$-algorithm (which would connect $D$

610: to $A$ in this example), but instead we focus on shooting pairs of

611: rays upward from strategic locations on the boundary of each band,

612: and then selecting a subset of these rays so that the conflicts can

613: be resolved and $T_U$ is connected. To ensure connectedness of all

614: bands, several ray-pairs must be issued upward from each band.

615: Figure~\ref{fig:face}b shows an example: no pair of rays can emanate

616: upward from the top of $B \cap P^-$ or $C \cap P^-$; one pair of

617: rays shoots upward from the top of each component of $A \cap P^-$:

618: $(r_1, r_2)$ connects $A$ to $B$ and $(r_3, r_4)$ connects $A$ to

619: $C$; finally, one pair of rays $(r_5, r_6)$ issues from the top of

620: $A \cap P^-$, which connects $A$ to $D$. So, overall, three pairs of

621: rays are generated for band $A$. We now turn to describing in detail

622: the method for generating ray-pairs.

623: %

624: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

625: \begin{figure}[htbp]

626: \centering

627: \begin{tabular}{c@{\hspace{0.15\linewidth}}c}

628: \includegraphics[width=0.3\linewidth]{Figures/face.eps} &

629: \includegraphics[width=0.34\linewidth]{Figures/ray.shooting.eps} \\

630: (a) & (b)

631: \end{tabular}

632: \caption{(a) The ray pair $(r_1, r_2)$ connects band $D$ to two

633: different bands $B$ and $C$. (b) To ensure connectivity, three pairs

634: of rays must be issued for $A$: $(r_1, r_2)$, $(r_3, r_4)$ and

635: $(r_5, r_6)$.} \label{fig:face}

636: \end{figure}

637: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

638: 

639: 

640: Let band $A$ intersect plane

641: $Y_i$. The algorithm is a for-loop over all $A$. Let $A_1, A_2,

642: \ldots, A_m$ be the components of $A$, defined as follows. Take all

643: the maximal components of $A \cap P^+$ that contain an $x$-gridedge,

644: and union with all the maximal components of $A \cap P^-$ that

645: contain an $x$-gridedge. We define $S(A_j)$ as the set of all

646: potential rays shooting upward from $A_j$. More precisely, $S(A_j)$

647: consists of the set of all segments $s=(a,b)$, with $a \in A_j$,

648: such that

649: \begin{enumerate}

650: \squeezelist

651: \item Either $s$ is a point, with $b=a$, or

652: $s \subset P \subset Y_i$ is vertical (parallel to $z$), with $a$

653: below $b$.

654: \item $b \in B$ for some band $B \neq A$.

655: \item The open segment $s \setminus \{a,b\}$ may contain points of $A$

656: (see $r_2$ in Fig.~\ref{fig:ray.pairs}b), but no other band points.

657: %\item If $s$ is a point, it is not isolated: there is another segment in $S(A_j)$

658: %in a neighborhood of $s$.

659: \end{enumerate}

660: %\xxx{This last condition is intended to discard isolated points, to

661: %deal with Robin's example.}

662: %

663: For each band $A$, for each component $A_j \subseteq A$, if $S(A_j)$

664: is nonempty, we select one ray pair $(r_1, r_2)$, such that (i)

665: $r_1$ is the leftmost segment in $S(A_j)$ that is incident to a

666: highest $x$-gridedge in $A_j$, and (ii) $r_2$ is the segment one

667: $x$-gridedge to the right of $r_1$. Fig.~\ref{fig:ray.pairs} shows a

668: few examples.

669: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

670: \begin{figure}[htbp]

671: \centering

672: \begin{tabular}{c@{\hspace{0.1\linewidth}}c@{\hspace{0.1\linewidth}}c}

673: \includegraphics[width=0.24\linewidth]{Figures/rays.1.eps} &

674: \includegraphics[width=0.25\linewidth]{Figures/rays.2.eps} &

675: \includegraphics[width=0.21\linewidth]{Figures/rays.3.eps} \\

676: (a) & (b) & (c)

677: \end{tabular}

678: \caption{Generating ray-pairs: (a)~$(r_1, r_2)$ for $A$; $S(B) =

679: \emptyset$. (b)~$(r_1, r_2)$ for $A$ (note that $r_2$ runs along

680: source band $A$); degenerate ray-pair $(r'_1, r'_2)$ for $B$.

681: (c)~$S(A) = \emptyset$; $(r_1, r_2)$ for $B$. } \label{fig:ray.pairs}

682: \end{figure}

683: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

684: As mentioned above, several ray pairs could be generated for any one

685: band, and indeed several pairs connecting two bands.

686: 

687: Let $G_b$ be the \emph{band graph} whose nodes are bands. Two bands

688: are connected by an arc in $G_b$ if the ray-pair algorithm generates

689: a ray connecting them.  We call a collection of bands in $G_b$

690: \emph{ray-connected} if they are in the same connected component of

691: $G_b$. We establish that $G_b$ is a connected graph, i.e., all bands

692: are ray-connected to one another, even if only one ray per pair is

693: employed:

694: 

695: \begin{lemma}

696: $G_b$ is connected. Furthermore, the subgraph of $G_b$ induced by

697: exactly one ray per ray-pair (arbitrarily selected) is connected.

698: \label{lem:Gb.connected0}

699: \end{lemma}

700: 

701: \medskip

702: \noindent Whereas the connectedness of bands by $z$-beams in the $(3

703: \times 1)$-algorithm is straightforward, the complex possible

704: relationships between bands makes connectedness via rays more

705: subtle. We relegate the proof to the Appendix

706: (Sec.~\ref{sec:Gb.connected}) in order to not interrupt the main

707: flow of the algorithm.

708: 

709: The over-generation of ray-pairs noted above is designed to ensure

710: connectedness. Eventually many rays will be discarded by the time

711: $T_U$ is constructed in Sec.~\ref{sec:spanning.tree}.

712: 

713: \subsubsection{Ray Graph $G_r$} \label{sec:ray.graph}

714: %------------------------------------------------------------------

715: One pair of rays per pair of bands suffices to ensure that all bands

716: are ray-connected. If multiple pairs of rays exist for a pair of

717: bands, pick one pair arbitrarily and discard the rest. Then define a ray

718: graph $G_r$ as follows.  The nodes of $G_r$ are vertical rays in a

719: plane $Y_i$, perhaps degenerating to points, connecting gridpoints

720: between two bands that both intersect $Y_i$. The arcs of $G_r$ each

721: records a potential pivoting conflict, and are of two varieties:

722: 

723: \begin{enumerate}

724: \squeezelist

725: \item[(i)] The nodes for the two rays issuing from the top of one band $B$

726: are adjacent in $G_r$.  Call such arcs \emph{$x$-arcs};

727: geometrically they can be viewed as parallel to the $x$-axis.

728: % (although they have

729: %no particular $x$-value).

730: 

731: \item[(ii)] The nodes for two rays incident to opposite sides of

732: the rim of a band $A$, connected by a $y$-segment on the band, are

733: adjacent in $G_r$.  Call such arcs \emph{$y$-arcs}; geometrically

734: they can be viewed as parallel to the $y$-axis.

735: %; we associate

736: %the $y$-coordinate value with the arc.

737: 

738: \end{enumerate}

739: Fig.~\ref{fig:Gr.path} shows two simple examples of $G_r$ involving

740: nodes on opposite sides of one band $A$.

741: %-------------------------------------------------------------------

742: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

743: \begin{figure}[htbp]

744: \centering

745: \begin{tabular}{c@{\hspace{0.1\linewidth}}c}

746: \includegraphics[width=0.4\linewidth]{Figures/Gr.path.1.eps} &

747: \includegraphics[width=0.4\linewidth]{Figures/Gr.path.2.eps} \\

748: (a) & (b)

749: \end{tabular}

750: \caption{ Building $G_r$. (a) $G_r$ is a $4$-cycle; $\{r_1,r_2\}$

751: and $\{r_3, r_4\}$ are $x$-arcs; any other arc is a $y$-arc. (b)

752: $G_r$ is a path; $\{r_2,r_5\}$ and $\{r_3, r_6\}$ are $y$-arcs; any

753: other arc is an $x$-arc.} \label{fig:Gr.path}

754: \end{figure}

755: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

756: %

757: Before proceeding, we list the consequences of the two types of arcs

758: in $G_r$. Assuming that we can $2$-color $G_r$ $\{$red, blue$\}$,

759: and we select the base of (say) the red rays as pivots, then: (i)

760: exactly one pivot is selected for each band, and (ii) no two pivot

761: rays are in conflict across a band. So our goal now is to show that

762: $G_r$ is $2$-colorable. Because a graph is $2$-colorable if and only

763: if it is bipartite, and a graph is bipartite if and only if every

764: cycle is of even length, we aim to prove that every cycle in $G_r$

765: is of even length. We start by listing a few relevant properties of

766: $G_r$:

767: \begin{enumerate}

768: \squeezelist

769: \item

770: Every node $r \in G_r$ has exactly one incident $x$-arc. The rays

771: are generated in pairs, and the pairs are connected by an $x$-arc.

772: As no such ray is shared between two bands, at most one $x$-arc is

773: incident to any $r$.

774: 

775: \item Nodes have at most degree $3$, with the following structure:

776: degree-$1$ nodes have  an incident $x$-arc; degree-$2$ nodes have

777: both an incident $x$- and $y$-arc; and degree-$3$ nodes have an

778: incident $x$-arc and two incident $y$-arcs.

779: 

780: %\item Any path consisting entirely of $y$-arcs remains at the

781: %same $y$-coordinate.  This is because $y$-arcs follow $y$-segments

782: %crossing bands, and nodes represent $z$-vertical rays that do not

783: %alter the $y$-coordinate.

784: 

785: \item Each $x$-arc spans exactly one pair of adjacent $y$-gridlines,

786: and each $y$-arc spans exactly one band rim-to-rim. The former is by

787: the definition of ray pairs, which issue from adjacent gridpoints,

788: and the latter follows from the grid partitioning of the object into

789: bands.

790: \end{enumerate}

791: %

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

793: \begin{figure}[htbp]

794: \centering

795: \includegraphics[width=0.98\linewidth]{Figures/dent.10cycle.1.eps}

796: \caption{ (a,b) Two side views of an object;  $z$-rays and $y$-arcs

797: are marked with thick lines. (c) $G_r$ coordinatized into $xy$-plane

798: $\Pi$; $(r_5, r_6, r_{10}, r_9)$ is a $4$-cycle;

799: $(r_1,r_3,r_4,r_8,r_7,r_{11},r_{12},r_9,r_5,r_2)$ is a $10$-cycle.}

800: \label{fig:dent.10-cycle}

801: \end{figure}

802: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

803: %

804: Our next step requires embedding $G_r$ in an $xy$-plane $\Pi$.

805: Toward that end, we coordinatize the nodes and arcs of $G_r$ as

806: follows. A node $r \in G_r$ is a $z$-ray, and is assigned the

807: $(x,y)$ coordinates of the ray.  Note that this means collinear rays

808: get mapped to the same point; however we treat them as distinct. The

809: $x$-arcs are then parallel to the $x$-axis, and the $y$-arcs are

810: parallel to the $y$-axis. In essence, this coordinatization is a

811: view from $z=+\infty$.

812: 

813: Fig.~\ref{fig:dent.10-cycle} shows a more complex example

814: illustrating this viewpoint. The object is composed of $7$ bands

815: $B_i$, one of which ($B_3$) is a dent. There are $12$ ray nodes, two

816: pairs of which lie on the same $z$-vertical line, namely $(r_4,r_5)$

817: and $(r_8,r_9)$. Note that there are $y$-arcs crossing both the top

818: of and the bottom\footnote{

819:     A dent is included in this example precisely to

820:     introduce such a bottom $y$-arc into $G_r$.

821: } of $B_4$. The graph $G_r$ has a $4$-cycle and a $10$-cycle, both

822: detailed in the caption (as well as a $12$-cycle not detailed).

823: 

824: 

825: \begin{lemma}

826: Every cycle in $G_r$ is of even length.

827: \end{lemma}

828: \begin{proof}

829: Let $C$ be a cycle in $G_r$.  The coordinatization described above

830: maps $C$ to a (perhaps self-crossing) closed path in the $xy$-plane

831: $\Pi$, a path which may visit the same $(x,y)$ point more than once,

832: and/or traverse the same edge in $\Pi$  more than once. Any such

833: closed path on a grid must have even length, for the following

834: reason.

835: 

836: First, by Property~(3) above, each edge of the path in $\Pi$

837: connects adjacent grid lines: an edge never ``jumps over'' one or

838: more grid lines. Second, any such closed lattice path changes parity

839: with each step, in the following sense.

840: %Assign to $y$-gridlines in $\Pi$ $y$-bits $0, 1, 0, 1, \ldots$ from left to right,

841: %and to $x$-gridlines $x$-bits $0, 1, 0, 1, \ldots$ from bottom to top.

842: %Define the parity of a gridpoint of $\Pi$ to be the sum of its $x$- and $y$-bits

843: %mod~$2$.

844: Number the $x$- and $y$-gridlines with integers $0,1,2,\ldots$ left

845: to right and bottom to top respectively. Define the parity of a

846: gridpoint of $\Pi$ to be the sum of its $x$- and $y$-gridline

847: coordinates, mod~$2$. Then each step of the path, necessarily in one

848: of the four compass directions, changes parity, as it changes only

849: one of $x$ or $y$. Returning to the start point to close the path

850: must return to the starting coordinates, and so to the same parity.

851: Thus, there must be an even number of parity changes along any

852: closed path. Therefore, $C$ has an even number of edges.

853: \end{proof}

854: 

855: \medskip

856: \noindent We have now established this:

857: \begin{theorem}

858: $G_r$ is $2$-colorable. \label{thm:2-color}

859: \end{theorem}

860: 

861: \noindent Note that nowhere in the above proof do we assume genus

862: zero, so this theorem holds for polyhedra of arbitrary genus.

863: 

864: \paragraph{Band pivoting.}

865: By Theorem~\ref{thm:2-color}, we can $2$-color the nodes of $G_r$

866: \{red,blue\}. We choose all red ray-nodes of $G_r$ to be pivoting

867: rays, in that their base points become pivot points.  As remarked

868: before, this selection guarantees that each band is pivoted, and no

869: two pivots are in conflict.

870: 

871: \subsection{Unfolding Tree $T_U$} \label{sec:spanning.tree}The next

872: task is to define a band spanning tree $T_U$, based on the band

873: graph $G_b$. Define $G'_b$, to retain the just the arcs of $G_b$

874: corresponding to the red ray nodes (in the above $2$-coloring) in

875: $G_r$. This maintains the connectivity of

876: Lemma~\ref{lem:Gb.connected0}.

877: %bands connected by pairs of rays are now connected by one ray.

878: Then take $T_U$ to be any

879: spanning tree of $G'_b$ rooted at a frontmost band.

880: 

881: With $T_U$ finally in hand, the remainder of the $(1 \times

882: 1)$-algorithm follows the overall structure of the $3 \times 1$

883: algorithm, with variations as mentioned before, as detailed below.

884: 

885: \subsubsection{Selecting Connecting Paths}

886: \label{sec:paths} Having established a pivot point for each band, we

887: are now ready to define the {\em forward} and {\em return}

888: connecting paths for a child band in $T_U$. Let $B$ be an arbitrary

889: child of a band $A$. If $B$ intersects $A$, both forward and return

890: connection paths for $B$ reduce to the pivot point $x_b$ (e.g., $u$

891: in Fig.~\ref{fig:overhang}). If $B$ does not intersect $A$, then a

892: ray $r$ connects $x_b$ to $A$ (Figs.~\ref{fig:box-in-box1}a

893: and~\ref{fig:box-aligned-box}a). The connecting paths are the two

894: vertical paths separated by $r$ comprised of the gridfaces sharing

895: an edge with $r$ (paths $a_1$ and $a_2$ in

896: Figs.~\ref{fig:box-in-box1}a and~\ref{fig:box-aligned-box}a). The

897: path first encountered in the unfolding of $A$ is used as a forward

898: connecting path; the other path is used as a return connecting path.

899: 

900: \subsubsection{Determining Unfolding Directions}

901: \label{sec:unfdir} A top-down traversal of $T_U$ assigns an

902: unfolding direction to each band in $T_U$ as follows. The root band

903: in $T_U$ may unfold either cw or ccw, but for definiteness we set

904: the unfolding direction to cw. Let $B$ be the band in $T_U$

905: currently visited and let $A$ be the parent of $B$. If the upward

906: ray $r$ incident to $x_b$ connects $B$ to a bottom gridpoint of $A$,

907: and if $A$ unfolds cw(ccw), then $B$ unfolds cw(ccw). Otherwise, $r$

908: connects $B$ to a top or a side (for degenerate rays) gridpoint of

909: $A$; in this case, if $A$ unfolds cw(ccw), then $B$ unfolds ccw(cw).

910: In other words, $A$ and $B$ unfold in a same direction if $B$

911: ``hangs below'' $A$, and in opposite direction otherwise.

912: 

913: \subsection{Unfolding Bands into a Net}

914: \label{sec:unfolding.bands} Let $A$ be a band to unfold, initially

915: the root band. The unfolding of $A$ starts at $x_a$ and proceeds in

916: the unfolding direction (cw or ccw) of $A$. Henceforth we assume

917: w.l.o.g. that the unfolding of $A$ proceeds cw (w.r.t. a viewpoint

918: at $y = -\infty$); the ccw unfolding of $A$ is a vertical reflection

919: of the cw unfolding of $A$. In the following we describe our method

920: to unfold every child $B$ of $A$ recursively, which falls naturally

921: into several cases.

922: 

923: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

924: \begin{figure}[htbp]

925: \centering \centering

926: \begin{tabular}{c@{\hspace{0.1\linewidth}}c}

927: \includegraphics[width=0.43\linewidth]{Figures/overhang.2.eps} &

928: \includegraphics[width=0.42\linewidth]{Figures/overhang.1.eps}

929: \end{tabular}

930: \caption{Unfolding $B$ when the ray connecting $B$ to $A$

931: degenerates to $x_b$.} \label{fig:overhang}

932: \end{figure}

933: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

934: 

935: \paragraph{Case 1:} Pivot $x_b \in A \cap B$.

936: Then, whenever the unfolding of $A$ reaches $x_b$, we unfold $B$ as

937: in Fig.~\ref{fig:overhang}. The unfolding uses the two band faces of

938: $A$ incident to $x_b$ ($a_0$ and $a_1$ in Fig.~\ref{fig:overhang}).

939: The gridface $b_0$ of $B$ ccw of $x_b$ gets rotated around $x_b$ so

940: that the ccw unfolding of $B$ extends horizontally to the right. The

941: unfolding of $B$ proceeds ccw back to $x_b$, then the face $a_1$

942: incident to $x_b$ gets oriented about $x_b$ so that the unfolding of

943: $A$ continues horizontal to the right.

944: 

945: Note that, because the pivots of any two children of $A$ are

946: conflict-free, there is no competition over the use of $a_0$ and

947: $a_1$ in the unfolding. Note also that the unfolding path does not

948: self-cross. For example, the cyclic order of the faces incident to

949: $u$ in Fig.~\ref{fig:overhang}a is $(a_0,

950: A_{front},b_0,b_1,B_{back},a_1)$, and the unfolding path follows

951: $(a_0,b_0,\ldots,b_1,a_1)$.

952: 

953: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

954: \begin{figure}[htbp]

955: \centering

956: \begin{tabular}{c@{\hspace{0.2\linewidth}}c}

957: \includegraphics[width=0.28\linewidth]{Figures/box-in-box.1.eps} &

958: \includegraphics[width=0.32\linewidth]{Figures/box-in-box.2.eps} \\

959: (a) & (b)

960: \end{tabular}

961: \caption{Unfolding $B$: $u_1$ is not a corner vertex of $A$ (a)

962: $x_b$ incident to a left face of $B$ (b) $x_b$ incident to a top

963: face of $b$.} \label{fig:box-in-box1}

964: \end{figure}

965: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

966: 

967: \paragraph{Case 2:} Pivot $x_b \not\in A \cap B$ and

968: the (forward, return) connecting paths for $B$ do not overlap other

969: connecting paths (except at their boundaries);

970: %This case is more complex, because it involves conflicts over the use

971: %of the connecting paths for $B$; i.e,

972: we will later see that this may happen.

973: %We discuss these conflicts after settling some notation

974: Let us settle some notation first % before we proceed.

975: (cf.~Fig~\ref{fig:box-in-box1}a):% and~\ref{fig:box-aligned-box}a):

976: ~$r$ is the ray connecting $B$ to $A$; $a_1$ and $a_2$ are forward

977: and return connecting paths for $B$ (one to either side of $r$);

978: $u_1$ is the endpoint of $r$ that lies on $A$; and $u_2$ is the

979: other endpoint of the $y$-edge of $A$ incident to $u_1$.

980: %

981: %Assume first that the forward path $a_1$ for $B$ does not overlap

982: %the return path for another descendant of $A$.

983: We discuss three situations:

984: 

985: \paragraph{Case 2a:} $u_1$ is neither a reflex corner nor a bottom

986: corner of $A$. In this case, whenever the unfolding of $A$ reaches

987: $a_1$, the unfolding of $B$ proceeds as in

988: Fig.~\ref{fig:box-in-box1}a or Fig.~\ref{fig:box-in-box1}b,

989: depending on whether $x_b$ touches a left face of $B$ or not. In

990: either case, if $b_0$ is the face of $B$ extending ccw left of

991: $x_b$, rotate $b_0$ so that the unfolding of $B$ extends horizontal

992: to the right, recursively unfold $B$, then rotate the return path

993: $a_2$ about $x_b$ so that the unfolding of $A$ continues horizontal

994: to the right.

995: 

996: \paragraph{Case 2b:} $u_1$ is a reflex corner of $A$. In this case,

997: the unfolding of $B$ proceeds as in Fig.~\ref{fig:box-in-box2}(a,

998: b). It is the existence of the vertical strip incident to $u_1$

999: (marked $t$ in Fig.~\ref{fig:box-in-box2}) that makes handling this

1000: case different from Case 2a above. Note however that the existence

1001: of $t$ implies the existence of at least two gridfaces on either the

1002: return path or the forward path for $B$, depending on whether $t$ is

1003: a left (Fig.~\ref{fig:box-in-box2}a) or a right

1004: (Fig.~\ref{fig:box-in-box2}b) strip of faces. In the former case the

1005: unfolding starts as in Case 2a (Fig.~\ref{fig:box-in-box2}a), and

1006: once the unfolding of $B$ returns to $x_b$, it continues along the

1007: return path up to $u_1$, then unfolds $t$ and orients it about $u_1$

1008: in such a way that the unfolding of $A$ continues horizontal to the

1009: right. The portion of the return path that extends above $u_1$

1010: ($a_{20}$ in Fig.~\ref{fig:box-in-box2}a) gets attached below the

1011: adjacent top face of $A$ ($a_3$ in Fig.~\ref{fig:box-in-box2}a).

1012: %

1013: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1014: \begin{figure}[htbp]

1015: \centering

1016: \begin{tabular}{c@{\hspace{0.2\linewidth}}c}

1017: \includegraphics[width=0.33\linewidth]{Figures/box-in-box.4.eps} &

1018: \includegraphics[width=0.27\linewidth]{Figures/box-in-box.3.eps} \\

1019: (a)  & (b)

1020: \end{tabular}

1021: \caption{Unfolding $B$: $u_1$ is a corner vertex of $A$. (a) $t$ is

1022: a left strip (b) $t$ is a right strip.} \label{fig:box-in-box2}

1023: \end{figure}

1024: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1025: %

1026: 

1027: If $t$ is a strip of right faces, then $t$ gets unfolded before

1028: descending along the forward path down to $B$, as in

1029: Fig.~\ref{fig:box-in-box2}b (note the vertical symmetry with the

1030: unfolding in Fig.~\ref{fig:box-in-box2}a); the unfolding of $B$ then

1031: proceeds as in Case 2a (Fig.~\ref{fig:box-in-box1}b).

1032: 

1033: \paragraph{Case 2c:} $u_1$ is a bottom corner of $A$. In this case,

1034: the unfolding proceeds as in Fig.~\ref{fig:box-aligned-box}a or

1035: Fig.~\ref{fig:box-aligned-box}b, depending on whether $u_1$ is a

1036: right or a left bottom corner of $A$. The unfolding illustrated in

1037: Fig.~\ref{fig:box-aligned-box}a follows the familiar unfolding

1038: pattern: orient the face of $B$ ccw left of $x_b$ so that the

1039: unfolding of $B$ extends to the right; once the unfolding of $B$

1040: returns to $x_b$, follow the return path back to $A$ and unfold the

1041: face of $A$ cw to the right of $u_1$ ($a_3$ in

1042: Fig.~\ref{fig:box-aligned-box}a) so that the unfolding of $A$

1043: continues horizontal to the right.

1044: %

1045: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1046: \begin{figure}[htbp]

1047: \centering

1048: \begin{tabular}{c@{\hspace{0.06\linewidth}}c}

1049: \includegraphics[width=0.38\linewidth]{Figures/box-aligned-box.2.eps}&

1050: \includegraphics[width=0.44\linewidth]{Figures/box-aligned-box.1.eps}

1051: \\

1052: (a) & (b)

1053: \end{tabular}

1054: \caption{Unfolding $B$: $u_1$ is a bottom corner of $A$ (a)

1055: rightmost, and (b) leftmost face of $A$ vertically aligned with

1056: leftmost face of $B$.} \label{fig:box-aligned-box}

1057: \end{figure}

1058: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1059: %

1060: A similar pattern applies to the case illustrated in

1061: Fig.~\ref{fig:box-aligned-box}b, with one subtle difference meant to

1062: aid in unfolding front and back faces (discussed in

1063: Sec.~\ref{sec:faces}): in unfolding bands, we aim at maintaining the

1064: vertical position of the (forward, return) connecting paths in the

1065: unfolding, so that vertical strips hanging below these connecting

1066: paths could also hang vertically in the unfolding. More on this in

1067: Sec.~\ref{sec:faces}. Observe that $a_1$  and $a_2$ from

1068: Fig.~\ref{fig:box-aligned-box}a hang downward in the unfolding.

1069: However, if $a_2$ were to maintain its vertical position in the

1070: unfolding from Fig.~\ref{fig:box-aligned-box}b, it would not be

1071: possible to orient $a_3$ around $u_1$ so as to continue unfolding

1072: $A$ horizontal to the right of $a_2$. This is the reason for

1073: employing the face marked $t$ in the unfolding, so that vertical

1074: sides of $t$ remain vertical in the unfolding, and any face strip

1075: hanging below $t$ could be attached to $t$ vertically in the

1076: unfolding.

1077: 

1078: We note that Fig.~\ref{fig:box-aligned-box} illustrates only the

1079: situation in which $x_b$ is incident to a left face of $B$, but it

1080: should not be difficult to observe that an exact same idea applies

1081: to any top pivot of $B$; the pivot position only affects the start

1082: and end unfolding position of $B$, and everything else remains the

1083: same.

1084: 

1085: \paragraph{Case 3:} Pivot $x_b \not\in A \cap B$ and a connecting

1086: path for $B$ overlaps a connecting path for another descendant $C$

1087: of $A$. This case is slightly more complex, because it involves

1088: conflicts over the use of the connecting paths for $B$. The

1089: following three situations are possible.

1090: 

1091: \paragraph{Case 3a:} The forward path $a_1$ for $B$ overlaps the return

1092: path for another descendant $C$ of $A$. This situation is

1093: illustrated in Fig.~\ref{fig:return}a.

1094: % since $x_c$ must lie on a left face of $C$.

1095: In this case, the unfolding $B$ starts as soon as the unfolding

1096: along the return path from $C$ to $A$ meets a face of $B$ incident

1097: to $x_b$ (face $b_0$ in Fig.~\ref{fig:return}a). At this point $B$

1098: gets recursively unfolded as before (see Fig.~\ref{fig:return}b),

1099: then the unfolding continues along the return path for $C$ back to

1100: $A$.

1101: %

1102: Fig.~\ref{fig:return}b shows face $a_1$ in two positions: we let

1103: $a_1$ hang down only if the next face to unfold is a right face of a

1104: child of $A$ (see the transition from $k_7$ to $c_5$ in

1105: Fig.~\ref{fig:ex.1}); otherwise, use $a_1$ in the upward position, a

1106: freedom permitted to us by rotating about vertex $u$.

1107: 

1108: \begin{figure}[htbp]

1109: \centering

1110: \includegraphics[width=0.8\linewidth]{Figures/return.eps}

1111: \caption{(a) Return path for $C$ includes $c_{20}, c_{21}, a_1$;

1112: forward path for $B$ is $a_1$. (b) Unfolding for (a) (c) Return path

1113: for $B$ includes $a_{20}, a_{21}, c_1$; forward path for $C$ is

1114: $c_1$. (d) Return path for $B$ is $a_2$; forward path for $C$

1115: includes $a_2, c_{21}$. (e) Forward (return) paths are identical for

1116: $B$ and $C$.} \label{fig:return}

1117: \end{figure}

1118: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1119: 

1120: \paragraph{Case 3b:}

1121: The return path $a_2$ for $B$ overlaps the forward path for another

1122: descendant $C$ of $A$. This situation is illustrated in

1123: Figs.~\ref{fig:return}c and~\ref{fig:return}d. The case depicted in

1124: Fig.~\ref{fig:return}c is similar to the one in

1125: Fig.~\ref{fig:return}a and is handled in the same manner. For the

1126: case depicted in Fig.~\ref{fig:return}d, notice that $a_2$ is on

1127: both the forward path for $C$ and the return path for $B$. However,

1128: no conflict occurs here: from $a_2$ the unfolding continues downward

1129: along the forward path to $C$ and unfolds $C$ next.

1130: 

1131: \paragraph{Case 3c:}

1132: The forward path $a_1$ for $B$ overlaps the forward path for another

1133: descendant $C$ of $A$. This situation occurs when either $B$ or

1134: another band $C$ incident to $B$ is a dent, as illustrated in

1135: Figs.~\ref{fig:return}e. Again, no conflict occurs here: the

1136: recursive unfolding of $C$, which returns to $x_c = x_b$, is

1137: followed by the recursive unfolding of $B$, which returns to $x_b$,

1138: then the unfolding continues along the return path for $B$ ($C$)

1139: back to $A$.

1140: 

1141: \medskip

1142: \noindent

1143: Fig.~\ref{fig:ex.1} shows a more complex example that

1144: emphasizes these subtle unfolding issues.

1145: Note that the return path $k_1, k_8, k_9$ for $B$ overlaps the

1146: forward path $k_9$ for $C$; and the return path

1147: $k_5, k_6$ and $k_7$ for $G$ overlaps the forward path

1148: for $H$, which includes $k_7$.

1149: The unfolding produced by the method described in this section

1150: is depicted in Fig.~\ref{fig:ex.1}(b).

1151: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1152: \begin{figure}[htbp]

1153: \centering

1154: \includegraphics[width=0.98\linewidth]{Figures/ex.eps}

1155: \caption{(a) An example. (b) The vertex-unfolding.}

1156: \label{fig:ex.1}

1157: \end{figure}

1158: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1159: 

1160: \subsection{Attaching Front and Back Faces to the Net}

1161: \label{sec:faces} Front and back faces of a slab are ``hung'' from

1162: bands following the basic idea of the illumination model discussed

1163: in Sec.~\ref{sec:3x1front.back}. There are three differences,

1164: however, caused by the employment of some front and back gridfaces

1165: for the connecting paths, which can block illumination from the

1166: bands.

1167: \begin{enumerate}

1168: \squeezelist

1169: \item We illuminate both upward and downward from each band:

1170: each $x$-edge illuminates the vertical face it attaches to. This

1171: alone already suffices to handle the example in Fig.~\ref{fig:ex.1}:

1172: all vertical faces are illuminated downward from the top of $A$,

1173: upward from the bottom of $A$, and upward from the top of $B$.

1174: 

1175: \item Some gridfaces still might not be illuminated by any bands,

1176: because they are obscured both above and below by paths in

1177: connecting faces.  Therefore we incorporate the connecting faces

1178: into the band for the purposes of illumination. For example, in

1179: Fig.~\ref{fig:box-aligned-box}a, $a_2$ illuminates downward and

1180: $a_1$ illuminates upward. The reason this works is that, with one

1181: exception, each vertical connecting strip remains vertical in the

1182: unfolding, and so illuminated strips can be hung safely without

1183: overlap.

1184: 

1185: \item The one exception is the forward connecting

1186: path $a_1$ in Fig.~\ref{fig:box-aligned-box}b. This paths unfolds

1187: ``on its side,'' i.e., what is vertical in 3D becomes horizontal in

1188: 2D. Note, however, that the face $x$ below each of these paths (a

1189: face always present), is oriented vertically. We thus consider $x$

1190: to be part of the connecting path for illumination purposes,

1191: permitting the strip below to be hung under $x$.

1192: \end{enumerate}

1193: Because our cases are exhaustive, one can see now that all gridfaces

1194: of (say) the front face of $A$ are either illuminated by $A$, or by

1195: some descendant of $A$ on the front face, augmented by the

1196: connecting paths as just described. (In fact every gridface is

1197: illuminated twice, from above and below.) Hanging the strips then

1198: completes the unfolding.

1199: 

1200: %------------------

1201: %%%JOR begin 4 July

1202: \subsection{Algorithm Complexity}

1203: Because there are so few unfolding algorithms, that there is

1204: \emph{some} algorithm for a class of objects is more important than

1205: the speed of the algorithm.  Nevertheless, we offer an analysis of

1206: the complexity of our algorithm. Let $n$ be the number of corner

1207: vertices of the polyhedron, and $N=O(n^2)$ the number of gridpoints.

1208: The vertex grid can be easily constructed in $O(N)$ time, leaving a

1209: planar surface map consisting of $O(N)$ gridpoints, gridedges, and

1210: gridfaces. The computation of connecting rays

1211: (Sec.~\ref{sec:spanning.tree}) requires determining the components

1212: of $A \cap P^+$ and $A \cap P-$, for each $A$. This can be easily

1213: read of from the planar map by running through the $n$ vertices of

1214: each of the $O(n)$ bands and determining, for each vertex, whether

1215: it belongs to $P^+$ or $P^-$.

1216: %a common vertex

1217: %between bands $A$ and $B$, or, if $A \cap B = \emptyset$,

1218: %determining whether $r_i(A) \subset r_i(B)$ or  $r_i(A) \supset

1219: %r_i(B)$. The former can be easily read off in $O(n^2)$ from the

1220: %planar map by running through the $n$ vertices and noting which of

1221: %the $O(n)$ bands share a vertex. The latter can be accomplished with

1222: %$O(n)$ ray shooting queries.

1223: Each of the $O(n)$ band components shoots a vertical ray from one

1224: corner vertex, in a 2D environment (the plane $Y_i$) of $n$

1225: noncrossing orthogonal segments. Determining which band a ray hits

1226: involves a ray shooting query.

1227: %A connecting ray determines an arc of $G_b$, with the band nesting

1228: %implicit.

1229: %

1230: Although an implementation would employ an efficient data structure,

1231: perhaps BSP trees~\cite{py-obspo-92}, for complexity purposes the

1232: naive $O(n)$ query cost suffices to lead to $O(n^2)$ time to

1233: construct $G_r$. Selecting pivots (Sec.~\ref{sec:pivots}) involves

1234: 2-coloring $G_r$ in $O(n)$ time, and computing the unfolding tree

1235: $T_U$ in a breadth-first traversal of $G_r$, which takes $O(n)$

1236: time. Unfolding bands (Sec.~\ref{sec:unfolding.bands}) involves a

1237: depth-first traversal of $T_U$ in $O(n)$ time, and laying out the

1238: $O(N)$ gridfaces in $O(N)$ time. Thus, the algorithm can be

1239: implemented to run in $O(N) = O(n^2)$ time.

1240: 

1241: \section{Further Work}

1242: Extending these algorithms to arbitrary genus orthogonal polyhedra

1243: remains an interesting open problem. Holes that extend only in the

1244: $x$ and $z$ directions within a slab seem unproblematic, as they

1245: simply disconnect the slab into several components. Holes that

1246: penetrate several slabs (i.e, extend in the $y$ direction) present

1247: new challenges. One idea to handle such holes is to place a virtual

1248: $xz$-face midway through the hole, and treat each half-hole as a

1249: dent (protrusion).

1250: %These ideas require new

1251: %techniques for building the unfolding tree

1252: %and potentially new algorithmic work.

1253: 

1254: \subsection*{Acknowledgements}

1255: %The third author was supported by NSF Distinguished Teaching

1256: %Scholars DTS award DUE-0123154.

1257: %\xxx[JOR]{NSF support in another footnote.}

1258: We thank the anonymous referees

1259: on~\cite{dfo-gvuop-06} for their careful reading and insightful

1260: comments.

1261: %We also thank the participants of the DIMACS Reconnect

1262: %July 2004 Workshop held at St. Mary's College, California, for

1263: %helpful discussions.

1264: 

1265: %\small

1266: %\newpage %JOR: To separate the bibliography from the 10-page body of paper

1267: 

1268: \bibliographystyle{alpha}

1269: \bibliography{VU} %Change (eventually)

1270: 

1271: \newpage

1272: \section{Appendix: Proof of Lemma~\protect\ref{lem:Gb.connected0} (Connectedness of $G_b$)}

1273: \label{sec:Gb.connected}

1274: %We distinguish three types of connectivity, all geometric

1275: %connectivity concepts. Two subsets of $P \subset Y_i$ are

1276: %\emph{front-connected} if there are points in each that are

1277: %connected by a path that lies wholly in $P-$, and

1278: %\emph{back-connected} if connected by a path that lies wholly in

1279: %$P+$. \xxx[JOR]{We don't seem to use front- and back-connected

1280: %anymore?}

1281: 

1282: Two subsets of $P \subset Y_i$ are \emph{path-connected}, or just

1283: \emph{connected}, if there are points in each that are connected by

1284: a path that lies in $P$. We need some notation to describe the

1285: portions of $r(A)$ that are relevantly connected to each band $A$.

1286: For a protrusion $A$, let $r_c(A)$ be the subset of $r(A)$

1287: (cf.~Sec.~\ref{sec:defs}) that is path-connected to $A$ via paths

1288: that do not cross any bands. For a dent $B$, let $r_c(B)$ be the

1289: boundary of $B$ plus the subset of $r(B)$ that is both

1290: path-connected to $B$ via paths that do not cross any bands, and is

1291: not part of $r_c(A)$, for some protrusion $A$. Consider for example

1292: Figure~\ref{fig:p-d0}b. For protrusion $B'$, $r_c(B')$ consists of

1293: the boundary rim of $B'$ and the portion of the back face of $B'$

1294: that overhangs dent $B$. For dent $B$, $r_c(B)$ consists only the

1295: boundary of $B$, even though the overhanging portion of $B'$ can be

1296: reached from $B$ without crossing any bands, because that is part of

1297: $r_c(B')$. In Figure~\ref{fig:d-d}a however, the portion of the

1298: front face of $A'$ enclosed by $B$ belongs to $r_c(B)$, not to

1299: $r_c(A')$.

1300: %Dents $B$ for which $r_c(B)$ contains points interior to

1301: %$B$ can also be viewed as protrusions, a view we will not explicitly

1302: %adopt in this proof.

1303: %We define $r_c(A)$ to be the subset of $r(A)$

1304: %path-connected to $A$. This definition implies that $r_c(A) =

1305: %\partial A$ for a dent $A$, because no point in the interior of

1306: %$r(A)$ is connected to $A$ in $P$.

1307: 

1308: The genus-zero assumption implies that, for protrusion $A$ and dent

1309: $B$ on opposite sides of $Y_i$ such that $r_c(A) \cap r_c(B)$ is

1310: nonempty, it must be that $A \cap B$ is nonempty (cf.

1311: Figs.~\ref{fig:p-d}). Define

1312: 

1313: $$

1314: r_c(A,B) = \left\{

1315: \begin{tabular}{ll}

1316: $A \cap B$,           & if $A \cap B \neq \emptyset$, and at least one of $A$ and $B$ is a dent \\

1317: $r_c(A) \cap r_c(B)$  & otherwise.

1318: \end{tabular}

1319: \right.

1320: $$

1321: 

1322: This definition is intended to identify gridpoints on either $A$ or

1323: $B$ from which rays are issued by the ray-pair generation algorithm

1324: (Sec.~\ref{sec:ray-pairs}). The reason for treating intersecting

1325: dents and protrusions differently is a subtle one, and is captured

1326: by Fig.~\ref{fig:p-d0}b: $B$ is a dent behind $Y_i$ and $B'$ is a

1327: protrusion in front of $Y_i$; $r_c(B')$ is the piece of the back

1328: face of $B'$ enclosed by $B$; $u$ is a highest gridpoint in $B \cap

1329: B'$, while $w$ is a highest gridpoint in $r_c(B) \cap r_c(B')$; $u$

1330: is a potential ray basepoint, while $w$ is not. The above definition

1331: eliminates points such as $w$ from the set $r_c(A, B)$.

1332: 

1333: Our connectivity proof for $G_b$ proceeds as follows.  In general,

1334: there are a number of disconnected maximal components $P_1, P_2,

1335: \ldots$ of $P$, with $P = P_1 \cup P_2 \cup \cdots$. The bands

1336: incident to each of these are ray-connected to each other via planes

1337: other than $Y_i$. We first argue that, to prove that $G_b$ is

1338: ray-connected, it suffices to prove that each $P_j$ is

1339: ray-connected. Remove from $O$ all the slabs $S_1, S_2, \ldots$

1340: incident to $Y_0$. Establish that the bands in the resulting object

1341: $O'$ are ray-connected, via induction. Now put back the slabs.  Each

1342: $S_j$ corresponds to a component $P_j$, and we are assuming we can

1343: establish that all bands incident to $P_j$ are ray-connected to one

1344: another.  This along with the fact that $O$ itself is connected

1345: implies that all bands are ray-connected. Henceforth we concentrate

1346: on one such connected component $P_j$, call it $Q \subset Y_i$ for

1347: succinctness. Let $\C$ be the collection of all bands that intersect

1348: $Q$. Then $\displaystyle\cup_{A \in \C} r_c(A) = Q$. The idea of the

1349: connectedness proof is that the bands get connected in upward

1350: chains, and ultimately to each other through ``common ancestor''

1351: higher bands. We choose to prove it by contradiction, arguing that a

1352: highest disconnected component cannot exist.

1353: 

1354: \begin{lemma}

1355: All bands in $\C$ are ray-connected. Furthermore, if one arbitrary

1356: ray in each ray-pair is discarded, $C$ remains ray-connected.

1357: \label{lem:connected}

1358: \end{lemma}

1359: \begin{proof}

1360: For the purpose of contradiction, assume that not all bands in $\C$

1361: are ray-connected. Let $\C_1, \C_2, \ldots$ be the distinct maximal

1362: subsets of $\C$ that are ray-connected. Let $Q_j = \cup_{A \in \C_j}

1363: r_c(A)$. Then $Q = \cup_j Q_j$. Since $Q$ is connected, the subsets

1364: $Q_j$ are not disjoint, in that for every $Q_j$ there is an $Q_k$

1365: such that $Q_j \cap Q_k$ is nonempty. By the observation above, this

1366: means that

1367: %

1368: \[

1369: Q_{jk} = \cup_{A \in \C_j, B \in C_k} ~r_c(A, B)

1370: \]

1371: %

1372: is also nonempty. Let $j$ and $k$ be such that $Q_{jk}$ contains a

1373: {\em highest} $x$-gridedge (gridpoint, if $Q_{jk}$ contains only

1374: isolated points) among all $Q_{jk}$. Let $u$ be the leftmost highest

1375: gridpoint in $Q_{jk}$. Let $A \in \C_j$ and $B \in \C_k$ be such

1376: that $u \in r_c(A, B)$.

1377: 

1378: We have thus identified two bands $A$ and $B$, ray-disconnected

1379: because in different components of $Q$, which contribute this

1380: highest gridpoint $u$ in the ``highest'' intersection $Q_{jk}$. We

1381: now examine in turn the four  protrusion/dent possibilities for

1382: these two bands.

1383: 

1384: \medskip

1385: \noindent {\bf Case 1.} $A$ and $B$ are both protrusions on opposite

1386: sides of $Y_i$. Assume w.l.o.g that $A$ is behind $Y_i$, $B$ is in

1387: front of $Y_i$, and $u$ is on $B$ (as depicted in

1388: Fig.~\ref{fig:p-p-opp}).

1389: %

1390: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1391: \begin{figure}[htbp]

1392: \centering

1393: \begin{tabular}{c@{\hspace{0.06\linewidth}}c@{\hspace{0.06\linewidth}}c}

1394: \includegraphics[width=0.25\linewidth]{Figures/p-p-opp.eps} &

1395: \includegraphics[width=0.25\linewidth]{Figures/p-p-opp2.eps} &

1396: \includegraphics[width=0.25\linewidth]{Figures/p-p-opp3.eps} \\

1397: (a) & (b)

1398: \end{tabular}

1399: \caption{Case 1: $A$ and $B$ are both protrusions on opposite sides

1400: of $Y_i$ (a) $D$ is a protrusion (b) $D$ is a dent with a vertical

1401: side incident to $u$ (c) $D$ is a dent with a bottom edge incident

1402: to $u$.} \label{fig:p-p-opp}

1403: \end{figure}

1404: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1405: %

1406: We discuss two subcases:

1407: \begin{enumerate}

1408: \item[a.] $u$ is on a top edge of $B$ (Figs.~\ref{fig:p-p-opp}(a,b)).

1409: Then our ray-pair algorithm generates a ray-pair $(r, r')$, with $r$

1410: incident to $u$ and $r'$ incident to the gridpoint $u'$ cw from $u$.

1411: Consider $r$ (the analysis is similar for $r'$). If $r$ hits $A$,

1412: then in fact $A$ and $B$ are ray-connected, contradicting the fact

1413: that $A$ and $B$ belong to different ray-connected components of

1414: $\C$. So let us assume that $r$ hits another band $D \in C_{\ell}$.

1415: Fig.~\ref{fig:p-p-opp}a(b) illustrates the situation when $D$ is a

1416: protrusion (dent). If $D \in \C_j$, then $D$ and $A$ are

1417: ray-connected in $C_j$, and since $B$ and $D$ are ray-connected, it

1418: follows that $B$ and $A$ are ray-connected, a contradiction. So

1419: assume that $D \in \C_{\ell}$, with $\ell \neq j$. But then $r_c(A,

1420: D)$ (and implicitly $Q_{j\ell}$) has a gridpoint higher than $u$,

1421: contradicting our choice of $j$, $k$ and $u$.

1422: 

1423: \item[b.] $u$ is on a vertical (left, right) edge of $B$

1424: (Fig.~\ref{fig:p-p-opp}c). Then $u$ must be at the intersection

1425: between a dent $D$ and $B$, meaning that $D \cap r_c(B)$ is

1426: nonempty. Furthermore, $r_c(A, D)$ has a gridpoint higher than $u$,

1427: meaning that $D \in G_j$. Let $u_1$ be the leftmost among the

1428: highest gridpoints of $D \cap r_c(B)$. Then our ray-pair algorithm

1429: generates a ray-pair $(r, r')$ from $u_1$ and its right neighbor

1430: $u'_1$. Consider $r$ (the analysis is similar for $r'$). If $r$ hits

1431: $B$, then $B$ is ray-connected to $D$, which is ray-connected to

1432: $A$, a contradiction. If $r$ hits a band $E$ other than $D$, then it

1433: must be that $D \in C_k$, since $r_c(B,E)$ has a gridpoint higher

1434: than $u_1$, which is no lower than $u$. This means that $B$ is

1435: ray-connected to $E$, which is ray-connected to $D$, which is

1436: ray-connected to $A$, a contradiction.

1437: \end{enumerate}

1438: 

1439: \medskip

1440: \noindent {\bf Case 2.} $A$ is a protrusion and $B$ is a dent, both

1441: on a same side of $Y_i$. The case when $A$ and $B$ are both in

1442: front of $Y_i$ (illustrated in Fig.~\ref{fig:p-d0}a) is identical to

1443: Case 1 above, once one conceptually pops out $B$ into a protrusion.

1444: We now discuss the case when $A$ and $B$ are both behind $Y_i$.

1445: 

1446: Assume first that $r_c(A, B)$ contains no top edges of $B$, as

1447: depicted in Figure~\ref{fig:p-d0}b. Let $B'$ be a protrusion in

1448: front of $Y_i$ covering the top of $B$. Then $r_c(A,B')$ and

1449: $r_c(B',B)$ each contains a gridpoint higher than $u$. The following

1450: two contradictory observations settle this case:

1451: \begin{enumerate}

1452: \item[a.] It must be that $B' \not\in \C_k$; otherwise $Q_{jk}$ would contain

1453: a gridpoint in $r_c(A,B')$ higher than $u$.

1454: 

1455: \item[b.] If $B' \in C_{\ell}$, then it must be that $\ell = k$; otherwise

1456: $Q_{\ell k}$ would contain a gridpoint in $r_c(B',B)$ higher than

1457: $u$.

1458: \end{enumerate}

1459: If $r_c(A,B)$ contains at least one top gridedge of $B$, then

1460: arguments similar to the ones used for the case illustrated in

1461: Fig.~\ref{fig:p-p-opp}a (conceptually popping $B$ to become a

1462: protrusion) settle this case as well.

1463: %

1464: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1465: \vspace{-1em}

1466: \begin{figure}[htbp]

1467: \centering

1468: \begin{tabular}{c@{\hspace{0.15\linewidth}}c}

1469: \includegraphics[width=0.32\linewidth]{Figures/p-d-opp3.eps} &

1470: \includegraphics[width=0.24\linewidth]{Figures/p-d-same.eps} \\

1471: (a) & (b)

1472: \end{tabular}

1473: \caption{Case 2: $A$ is a protrusion and $B$ is a dent (a) behind

1474: $Y_i$ (b) in front of $Y_i$.} \label{fig:p-d0}

1475: \end{figure}

1476: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1477: 

1478: \medskip

1479: \noindent {\bf Case 3.} $A$ is a protrusion and $B$ is a dent on

1480: opposite sides of $Y_i$ (see Fig.~\ref{fig:p-d}). Let $B'$ be the

1481: protrusion in front of $Y_i$ enclosing $B$.

1482: %Then the ray-pair algorithm shoots a ray upwards from a

1483: %gridpoint on $A \cap r(B)$ or $A \cap r(B')$, depending on

1484: %whether $u$ is in $A \cap B$ or not.

1485: We discuss two subcases:

1486: \begin{enumerate}

1487: \item[a.] $r_c(A)$ contains a top edge of $B$

1488: (see Fig.~\ref{fig:p-d}a). This means that $r_c(A) \cap r(B)$ is

1489: nonempty, and the ray-pair algorithm shoots a ray-pair $(r, r')$

1490: upward from the endpoints of a highest gridedge $\{u_1, u'_1\}$ of

1491: $A \cap r(B)$. Consider ray $r$ (the analysis is similar for $r')$.

1492: If $r$ hits $B$, then $A$ and $B$ are in fact ray-connected, a

1493: contradiction. If $r$ hits a band $D$ other than $B$, then arguments

1494: similar to the ones for the case illustrated in

1495: Fig.~\ref{fig:p-p-opp}a (Case 1) lead to a contradiction.

1496: %

1497: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1498: \vspace{-1em}

1499: \begin{figure}[htbp]

1500: \centering

1501: \begin{tabular}{c@{\hspace{0.15\linewidth}}c}

1502: \includegraphics[width=0.25\linewidth]{Figures/p-d-opp.eps} &

1503: \includegraphics[width=0.25\linewidth]{Figures/p-d-opp2.eps} \\

1504: (a) & (b)

1505: \end{tabular}

1506: \caption{Case 3: $A$ is a protrusion behind $Y_i$; $B$ is a dent in

1507: $B'$, both in front of $Y_i$.} \label{fig:p-d}

1508: \end{figure}

1509: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1510: %

1511: 

1512: \item[b.] $r_c(A)$ contains a bottom edge of $B$.

1513: This case is symmetrical to the one above in that a ray upward from

1514: a gridpoint of $B \cap r(A)$ hits $A$, thus ray-connecting $A$ and

1515: $B$.

1516: %

1517: \item[c.] $r_c(A)$ contains neither a top nor a bottom

1518: edge of $B$ (see Fig.~\ref{fig:p-d}b). Arguments similar to the ones

1519: used in Case 1 (protrusions on opposite sides of $Y_i)$ show that

1520: $A$ and $B'$ are ray-connected. That $B$ and $B'$ are ray-connected

1521: follows immediately from the fact that $r_c(B, B')$ has a gridpoint

1522: higher than $u$ ($w$ in Fig.~\ref{fig:p-d}b). These together imply

1523: that $A$ and $B$ are ray-connected, a contradiction.

1524: \end{enumerate}

1525: %

1526: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin

1527: \begin{figure}[htbp]

1528: \centering

1529: \begin{tabular}{c@{\hspace{0.15\linewidth}}c}

1530: \includegraphics[width=0.32\linewidth]{Figures/d-d-opp.eps} &

1531: %\includegraphics[width=0.25\linewidth]{Figures/d-d-opp3.eps} &

1532: \includegraphics[width=0.25\linewidth]{Figures/d-d-opp2.eps} \\

1533: (a) & (b)

1534: \end{tabular}

1535: \caption{Case 4: $A$ is a dent behind $Y_i$, enclosed within

1536: protrusion $A'$. $B$ is a dent in front of $Y_i$, enclosed within

1537: protrusion $B'$.} \label{fig:d-d}

1538: \end{figure}

1539: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End

1540: %

1541: 

1542: \medskip

1543: \noindent {\bf Case 4.} $A$ and $B$ are both dents: $A$ is a dent

1544: behind $Y_i$ enclosed within protrusion $A'$, and $B$ is a dent in

1545: front of $Y_i$ enclosed within protrusion $B'$ (see

1546: Fig.~\ref{fig:d-d}). The genus-zero assumption implies that $r(A)

1547: \cap r(B)$ is a polygonal region of positive area. Since $u \in

1548: r_c(A) \cap r_c(B)$, we have that $u \in r(A) \cap r(B)$. Let

1549: $\beta$ be the boundary segment of $r(A) \cap r(B)$ incident to $u$.

1550: We discuss two subcases:

1551: \begin{enumerate}

1552: \item[a.] $\beta \subset P^-$, meaning that $\beta \subset A$

1553: (see Fig.~\ref{fig:d-d}a).

1554: 

1555: An analysis similar to the one for the case illustrated in

1556: Fig.~\ref{fig:p-d}a (Case 3) shows that $A$ and $B$ are

1557: ray-connected, a contradiction.

1558: 

1559: \item[b.] $\beta \subset P^+$, meaning that $\beta \subset B$

1560: (see Fig.~\ref{fig:d-d}b). We show that $A$ and $A'$ are

1561: ray-connected, $B$ and $B'$ are ray-connected, and $A'$ and $B'$ are

1562: ray-connected. This implies that $A$ and $B$ are ray-connected, a

1563: contradiction. First note that the ray-pair algorithm shoots a

1564: ray-pair $(r, r')$ upward from a highest gridedge on $\beta$. An

1565: analysis similar to the one for the case illustrated in

1566: Fig.~\ref{fig:p-p-opp}a (conceptually popping $B$ to become a

1567: protrusion) shows that $r$ and $r'$ must hit $B$, thus

1568: ray-connecting $B$ and $B'$. That $A$ and $A'$ are ray-connected

1569: follows immediately from the fact that $r_c(A, A')$ has a gridpoint

1570: higher than $u$, and similarly for $A'$ and $B'$.

1571: \end{enumerate}

1572: Having exhausted all possible cases, the connectivity claim of the

1573: lemma is established. Because the proof for each of these cases goes

1574: through by considering either the first or second ray of a ray-pair,

1575: retaining either ray suffices to preserve connectivity. Thus the

1576: second claim of the lemma is established as well.

1577: \end{proof}

1578: 

1579: \end{document}

1580: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1581: