1: \documentclass[11pt]{article}

2: \pagestyle{plain}

3: 

4: \usepackage{times}

5: \usepackage{mathfont}

6: \usepackage{graphicx}

7: \usepackage{cite}

8: 

9: \usepackage{url}

10: \urlstyle{same}

11: \makeatletter

12: \def\UrlSpecials{\do\<{\langle}\do\>{\rangle\penalty\relpenalty}\do\_{\_%

13:  \penalty\@m}\do\|{\mid}\do\{{\lbrace}\do\}{\rbrace\penalty\relpenalty}\do

14:  \\{\mathbin{\backslash}}\do\~{\mathord{{\textstyle\sim}}}\do\ {\ }}

15: \makeatother

16: 

17: \setlength{\textwidth}{6.5in}

18: \setlength{\textheight}{9in}

19: \setlength{\topmargin}{-.5in}

20: \oddsidemargin=.25in

21: \evensidemargin=.25in

22: \setlength{\oddsidemargin}{.25in}

23: \setlength{\evensidemargin}{.25in}

24: 

25: \def\min{\mathop{\rm min}}

26: \def\max{\mathop{\rm max}}

27: \def\log{\mathop{\rm log}}

28: 

29: % magic to make big-O notation use script font

30: \mathcode`O="724F

31: 

32: \begin{document}

33: 

34: \title{Hinged Kite Mirror Dissection}

35: 

36: \author{David Eppstein\thanks{Dept. Inf. \& Comp. Sci., Univ. of

37: California, Irvine, CA 92697-3425.  Email: {\tt eppstein@ics.uci.edu}.}}

38: 

39: \date{ }

40: \maketitle   

41: 

42:  

43: \begin{abstract}

44: Any two polygons of equal area can be partitioned into congruent sets of

45: polygonal pieces, and in many cases one can connect the pieces by

46: flexible hinges while still allowing the connected set to form both

47: polygons.  However it is open whether such

48: a hinged dissection always exists.  We solve a special case of this

49: problem, by showing that any asymmetric polygon always has a hinged

50: dissection to its mirror image.  Our dissection forms a chain

51: of kite-shaped pieces, found by a circle-packing algorithm for

52: quadrilateral mesh generation. A hinged mirror dissection of a polygon

53: with $n$ sides can be formed with $O(n)$ kites in $O(n\log n)$ time.

54: \end{abstract} 

55: 

56: \section{Introduction}

57: 

58: A {\em dissection} of one polygon to another is a partition of the first

59: polygon into smaller polygonal pieces that can be rearranged to form the

60: second polygon.  Dissections are possible when (and only when) the two

61: polygons have the same area, indeed dissection was used by Hilbert as the

62: basis for an axiomatization of area~\cite{Hil-87}.  Dissection puzzles are

63: also popular in recreational geometry, where the main aim is to find

64: dissections of interesting shapes such as regular polygons that use as

65: few pieces as possible~\cite{Fre-97}.

66: 

67: \begin{figure}[wp]

68: \centering\includegraphics[width=5.5in]{trisquare.eps}

69: \caption{Hinged dissection of equilateral triangle and square.}

70: \label{fig:trisquare}

71: \end{figure}

72: 

73: \begin{figure}[wp]

74: \centering\includegraphics[width=3in]{pentomino.eps}

75: \caption{Chains of isosceles right triangles form hinged dissections of

76: any polyomino~\cite{cs.CG/9907018}.}

77: \label{fig:pentomino}

78: \end{figure}

79: 

80: \begin{figure}[wp]

81: \centering\includegraphics[width=\textwidth]{timelapse.eps}

82: \caption{A chain of kites, hinged along their axes of symmetry, can be

83: unfolded so that all axes are colinear, and refolded to form the mirror

84: image of the original polygon.}

85: \label{fig:time-lapse}

86: \end{figure}

87: 

88: A classic example is the four-piece dissection from an equilateral

89: triangle to a square, often ascribed to Dudeney (but

90: see~\cite{Fre-97,Fre-02} for speculation on its origin).

91: The four pieces in the dissection can be connected by {\em hinges},

92: points of attachment at which the two attached pieces are free to rotate

93: (Figure~\ref{fig:trisquare}), while still allowing all four pieces to fold

94: up into both the square and the triangle.  This example has sparked much

95: interest in similar hinged dissections~\cite{AkiNak-JDCG-98,Fre-02} but

96: few general results are known, and it remains open whether each equal-area

97: pair of polygons has a hinged dissection.  In one of the few theoretical

98: papers in this area, Demaine et al.~\cite{cs.CG/9907018} showed that

99: chains of isosceles right triangles form hinged dissections between any

100: pair of $n$-ominos (Figure~\ref{fig:pentomino}), and more generally that

101: hinged dissections are possibly between many pairs of {\em polyforms},

102: shapes formed by face-to-face gluing of many copies of the same basic

103: form.

104: 

105: In this paper we demonstrate the existence of another class of hinged

106: dissections: we prove that any asymmetric polygon (including polygons with

107: holes) has a hinged dissection to its mirror image.

108: Our method is to find a dissection that can be unfolded on its hinges

109: into a symmetric form: a chain of {\em kites} (quadrilaterals with

110: reflection symmetry across a diagonal) connected end-to-end, so all

111: their axes of symmetry lie on a common line.  Clearly,

112: one can then perform the mirror image of the unfolding process to fold

113: the chain back up into the mirror image of the original polygon;

114: Figure~\ref{fig:time-lapse} shows this unfolding and refolding

115: process applied to a hinged kite dissection of an asymmetric concave

116: hexagon (however this dissection is not one that would be found by the

117: algorithms we describe).  We prove that any polygon with

118: $n$ sides has a hinged kite dissection with $O(n)$ pieces, which can be

119: computed by an algorithm running in $O(n\log n)$ time.

120: 

121: Although the set of dissections we find is very limited, we discuss

122: possible implications for more general dissections between any two

123: equal-area polygons.  Our technique allows us to reduce the general

124: dissection problem into the question of whether any two equal-area

125: triangles have a hinged dissection in which some copies of a specified

126: pair of vertices of the first triangle map to a specified pair of vertices

127: of the second.

128: 

129: As in the paper of Demaine et al.~\cite{cs.CG/9907018}, we do not

130: consider the question of whether our dissections can be continuously

131: unfolded without any intermediate self-intersections, so in the

132: terminology of Frederickson~\cite{Fre-02} all our dissections are {\em

133: wobbly-hinged}.

134: 

135: \section{Dissection Process}

136: 

137: \begin{figure}[wp]

138: \centering\includegraphics[width=5.5in]{kitedis.eps}

139: \caption{Steps in our hinged dissection process.}

140: \label{fig:kitedis}

141: \end{figure}

142: 

143: \begin{figure}[t]

144: \centering\includegraphics[width=4.5in]{kitecases.eps}

145: \caption{Cases for partition of circle-packing gaps into kites,

146: from~\cite{BerEpp-IJCGA-00}.}

147: \label{fig:kite-cases}

148: \end{figure}

149: 

150: We now describe the steps by which we find a hinged dissection of an

151: arbitrary polygon into a chain of kites. These steps are also illustrated

152: in Figure~\ref{fig:kitedis}.  Our technique is based on a circle-packing

153: algorithm of Bern and the author~\cite{BerEpp-IJCGA-00} for partitioning

154: polygons into well-shaped quadrilaterals in the context of

155: finite element mesh generation; this method is based on previous

156: circle-packing nonobtuse triangulation algorithms by Bern et

157: al.~\cite{BerMitRup-DCG-95,Epp-IJCGA-97} and has also been applied to

158: problems of paper folding and cutting~\cite{BerDemEpp-Fun-98}.

159: 

160: We dissect the given polygon by the following sequence of steps.

161: Steps 1-\ref{step-cases} are taken from the kite meshing algorithm of Bern

162: and the author, while the remaining steps transform the kite mesh into a

163: hinged dissection.  In steps 1-\ref{finish-packing} we pack the polygon by

164: tangent circles, so that the polygon is

165: partitioned by the circles into regions of two types: interiors of

166: circles, and nonconvex {\em gaps} exterior to the circles.  Each gap is

167: bounded by three or more {\em sides} consisting of segments of polygon

168: boundary and arcs of tangent circles.

169: 

170: \begin{enumerate}

171: \item We begin by placing pairs of congruent circles near each reflex

172: vertex of the polygon, tangent to each other and to the polygon.

173: 

174: \item We place additional circles tangent to each boundary component of

175: the polygon, so that the circles are connected in a cycle by

176: tangencies, with a circle doubly tangent to the polygon near each

177: convex vertex.  After this step, each gap involving a segment of polygon

178: boundary (other than the four-sided gaps created in step~1) has exactly

179: three sides.  However the gaps in the interior of the polygon may still

180: have many sides.

181: 

182: \item

183: \label{finish-packing}

184: As long as there is a gap with five or more sides,

185: we place a circle to split it into two simpler gaps.

186: 

187: \item

188: The remaining gaps have three or four sides.  We draw

189: line segments between each circle center and the circle's points of

190: tangencies, partitioning the polygon into triangles and quadrilaterals

191: surrounding each gap, with distinguished points (the tangencies) towards

192: the center of each triangle or quadrilateral edge.

193: 

194: \item

195: \label{step-cases}

196: We now partition each of these triangles or quadrilaterals into kites,

197: according to a case analysis shown in Figure~\ref{fig:kite-cases}:

198: 

199: \begin{enumerate}

200: \item In a three-sided gap interior to the polygon, we place a point at

201: the circumcenter of the triangle formed by the three points of tangency,

202: and connect this center point to each tangency.

203: 

204: \item In a four-sided gap, the four points of tangency are always

205: cocircular~\cite{BerMitRup-DCG-95}.  In most such cases, as in the case

206: of three-sided gaps, we place a point at the circumcenter of these four

207: points, and connect this center point to each tangency.

208: 

209: \item There may be some four-sided gaps in which the center point is not

210: interior to the convex hull of the four tangencies, so that the previous

211: case would lead to the creation of a concave dart shape instead of a

212: kite.  Bern et al.~\cite{BerMitRup-DCG-95} call this case a {\em bad gap}

213: and show that it can always be split into two good gaps by the addition

214: of a single circle tangent to two of the four arcs of the bad gap.

215: These two good gaps can be covered by seven kites

216: (Figure~\ref{fig:kite-cases}, top right).

217: 

218: \item When two circles form a gap with a straight piece of polygon

219: boundary, we can partition this gap into two kites by a line segment

220: through the circle tangency and perpendicular to the line between the two

221: circle centers.  The same type of partition also applies to a gap

222: containing a reflex vertex, because we chose the two circles

223: forming this gap to be congruent.

224: 

225: \item The final case consists of a gap formed by a convex vertex and a

226: single circle, however this type of gap is already in the form of a kite.

227: \end{enumerate}

228: 

229: \item

230: \label{tree-step}

231: We now have a partition of the polygon into kites,

232: however we are not finished because it may not be possible to hinge the

233: kites appropriately.  We call the kites of this partition {\em large

234: kites} to distinguish them from the ones formed in step~\ref{step-bisect}

235: below. We next find a tree, with one vertex interior to each large kite,

236: where each tree edge connects points from two adjacent large kites. 

237: (I.e., this is a spanning tree of the dual graph of the large kite mesh.)

238: 

239: \item

240: \label{step-bisect}

241: We partition each large kite into four smaller pieces by

242: placing a point at the intersection of its two diagonals, and connecting

243: that point to the midpoints of the large kite edges.  This partitions the

244: large kite into four pieces, two of which are similar to the original

245: large kite (shown shaded in Figure~\ref{fig:kitedis}) and the other two of

246: which are rhombi.

247: 

248: \item We arrange the spanning tree of step~\ref{tree-step} so that its

249: vertices lie at the interior points added within each large kite, and its

250: edges lie along the connections from these interior points to the

251: large kite edge midpoints.  We add a single segment connecting this

252: spanning tree to the midpoint of an edge on the outer boundary of the

253: polygon.

254: 

255: \item Finally, we trace around the boundary of the tree, and form a

256: linear sequence of small kites and rhombs in the order in which they are

257: visited by this trace.  We hinge these kites and rhombs at the vertices

258: on the edge midpoints of the large kites.  Each small kite is hinged at

259: the two vertices of its axis of symmetry, and each rhomb is hinged at

260: two opposite vertices. 

261: \end{enumerate}

262: 

263: \begin{figure}[t]

264: \centering\includegraphics[width=5.5in]{mirrortri.eps}

265: \caption{Hinged dissections of a scalene triangle.}

266: \label{fig:mirror-tri}

267: \end{figure}

268: 

269: Bern and the

270: author~\cite{BerEpp-IJCGA-00} use a somewhat more complicated case

271: analysis in step~\ref{step-cases}, allowing four-sided gaps involving the

272: edges of the polygon, in order to show that the dissection into kites

273: used here can be performed in time

274: $O(n\log n)$ and that it need only create

275: $O(n)$ pieces.

276: 

277: Another example of a dissection created by this process, of a scalene

278: triangle, is shown in Figure~\ref{fig:mirror-tri} (center).  In this case

279: the circle packing consists of a single circle inscribed in the triangle,

280: eventually resulting in a twelve-piece dissection. However, the number of

281: pieces can be improved: as shown in the right of the figure, any scalene

282: triangle has a three-piece hinged mirror dissection into a kite and two

283: isosceles triangles, formed by cutting from the midpoints of the two

284: short sides of the triangle to a third point on the hypotenuse. The third

285: point is the reflection of the hypotenuse midpoint across the

286: perpendicular bisector of the two other midpoints; the line through it

287: and the opposite vertex is perpendicular to the hypotenuse. The three

288: pieces formed by these two cuts are then hinged at the side midpoints.

289: 

290: \section{Possible Implications}

291: 

292: We still seem to be a long way from solving the question of whether

293: hinged dissections exist between any pair of equal area

294: polygons, or even more generally between any set of equal area polygons.

295: However, our kite dissection can be used to reduce this problem to a

296: seemingly more simple form.

297: 

298: Suppose we have two equal area polygons, both dissected into chains of

299: kites hinged end-to-end.  The sequences of areas of the kites can be

300: viewed as partitions of the one-dimensional interval $[0,A]$, and we can

301: find a common refinement of these two partitions by overlaying them.

302: Geometrically, as shown in Figure~\ref{fig:kitesplit}, this corresponds

303: to introducing a sequence of cuts to the two chains of kites,

304: partitioning them into smaller kites and darts, still hinged end-to-end,

305: so that both chains are composed of polygons that form the same sequence

306: of areas. In other words, the first kite or dart from the first chain has

307: the same area as the first kite or dart from the second chain, and so on.

308: The cuts in one chain correspond to the hinges in the other chain and

309: vice versa.  If we could then further hinge-dissect each equal-area pair,

310: we could combine these parts into a hinge dissection of the original two

311: polygons. By further splitting the kites and darts along their axes of

312: symmetry, we reduce the problem to one in which we must dissect a

313: sequence of equal area triangles.

314: 

315: \begin{figure}

316: \centering\includegraphics[width=3.5in]{kitesplit.eps}

317: \caption{Refinement of two equal-area chains of kites into smaller kites

318: and darts so that each pair of pieces has the same area.}

319: \label{fig:kitesplit}

320: \end{figure}

321: 

322: Summarizing, we would be able to hinge-dissect any two equal-area

323: polygons, if only we could hinge-dissect the very simple special case of

324: two equal-area triangles, with the restriction that copies of two

325: vertices from the first triangle are mapped to two vertices of the second

326: triangle so that the dissected triangles can be connected to their

327: neighbors in the chain.  Our kite dissection method transforms any single

328: dissection problem of two polygons with a total of $n$ sides into a

329: sequence of

330: $O(n)$ triangle dissection problems.  More generally, we can use the same

331: construction to reduce any $k$-way dissection problem to one involving

332: only triangles.

333: 

334: A similar result could be obtained by using Saalfeld's decomposition of

335: equal-area polygons into combinatorially equivalent equal-area

336: triangulations~\cite{Saa-SCG-01}, however his method lacks complexity

337: bounds and seems to use a large number of pieces.

338: 

339: \section{Discussion}

340: 

341: We have shown that any polygon has a hinged dissection in

342: the form of a chain of kites, that can be unfolded and refolded to

343: form the mirror image of the original polygon.  The result also has some

344: possible consequences for the open problem of the existence of hinged

345: dissections between any pair of equal area polygons.

346: 

347: Some questions about our method remain unanswered, for instance whether

348: our dissections or modifications of them can be unfolded in a continuous

349: motion that avoids self-intersections.  Also, the number of pieces used

350: by our dissections, although asymptotically optimal, seems large, and

351: Figure~\ref{fig:mirror-tri} shows that it can be reduced by a factor of

352: four in the case of scalene triangles.  Is a similar reduction possible

353: more generally?

354: 

355: \section*{Acknowledgements}

356: 

357: My thanks go to Greg Frederickson for encouraging me to publish these

358: results, to Erik Demaine for extensive comments on a draft of this paper,

359: and to Cinderella for help with the figures.

360: This work was supported in part by NSF grant CCR-9912338.

361: 

362: %\raggedright

363: \bibliographystyle{abuser}

364: \bibliography{cpack}

365: 

366: \end{document}

367: