1: % cleaned version to send to DCG; comments removed!
2:
3: \documentclass[11pt]{article}
4: \usepackage{amsfonts,latexsym,graphicx,ifpdf,overpic,subfigure,url,wrapfig,enumerate}
5: \usepackage{amsmath}
6: \usepackage{amsthm}
7: \urlstyle{rm}
8:
9: \hyphenation{poly-gonal}
10:
11: % 1-inch margins, from fullpage.sty by H.Partl, Version 2, Dec. 15, 1988.
12: \topmargin 0pt
13: \advance \topmargin by -\headheight
14: \advance \topmargin by -\headsep
15: \textheight 8.9in
16: \oddsidemargin 0pt
17: \evensidemargin \oddsidemargin
18: \marginparwidth 0.5in
19: \textwidth 6.5in
20:
21:
22:
23:
24: % Generate hyperlinked DVI, PS, and PDF.
25: % Best used either with "latex" + "dvips -z", or with "pdflatex".
26: % If you generate PS or PDF via another mechanism, e.g., "dvipdf",
27: % you should add e.g. [dvipdf] to the \usepackage{hyperref} line.
28: % pdftitle is set automatically, but you can override and/or set pdfauthor.
29: \usepackage
30: [breaklinks,bookmarks,bookmarksnumbered,bookmarksopen,bookmarksopenlevel=2]
31: {hyperref}
32: {\makeatletter \hypersetup{pdftitle={\@title}}}
33: %\hypersetup{pdfauthor={}}
34:
35: % Automatic conversion from eps to pdf (via epstopdf).
36: % Under normal operation, replace \includegraphics{file.eps} with {file.pdf}.
37: % If you use "pdflatex -shell-escape", runs epstopdf to generate .pdf files
38: % for all existing and included .eps files. Do this whenever a file changes.
39: % \includegraphics{file.pdf} will prevent file.eps from being converted.
40: % You must \usepackage{ifpdf} so that this code knows when to take action.
41: \ifpdf
42: \usepackage{epstopdf}
43: \makeatletter
44: \edef\Gin@extensions{.eps,\Gin@extensions}
45: \DeclareGraphicsRule{.eps}{pdf}{.pdf}{`epstopdf #1}
46: \makeatother
47: \fi
48:
49: % Automatic cleaning of (e.g., Adobe Illustrator) EPS via eps2eps.
50: % Can be selectively turned off via \disablecleaneps and \enablecleaneps.
51: % If you use \disablecleaneps before \includegraphics within a {figure}
52: % environment (or any group), then just that figure will not use eps2eps.
53: % You must \usepackage{ifpdf} so that this code knows when to take action.
54: \ifpdf
55: \let\enablecleaneps=\relax
56: \let\disablecleaneps=\relax
57: \else
58: \def\enablecleaneps{\DeclareGraphicsRule{.eps}{eps}{.eps}{`eps2eps ##1 -}}
59: \def\disablecleaneps{\DeclareGraphicsRule{.eps}{eps}{.eps}{}}
60: \enablecleaneps
61: \fi
62:
63: % Avoid line breaks before citations (\cite) and references (\ref)
64: \let\latexcite=\cite
65: \def\cite{\nolinebreak\latexcite}
66: \let\latexref=\ref
67: \def\ref{\nolinebreak\latexref}
68:
69: % Put figures and text together
70: \def\textfraction{0.01}
71: \def\topfraction{0.99}
72: \def\bottomfraction{0.99}
73: \def\floatpagefraction{0.99}
74: \def\dblfloatpagefraction{0.99}
75:
76: % Theorem environments
77: \newtheorem{theorem}{Theorem}
78: \newtheorem{proposition}{Proposition}
79: \newtheorem{lemma}[theorem]{Lemma}
80: \newtheorem{corollary}[theorem]{Corollary}
81: \newtheorem{onerule}{Rule}
82:
83: % Fonts
84: \let\epsilon=\varepsilon
85: \newcommand{\eps}{\varepsilon}
86: \newcommand{\R}{\mathbb{R}}
87:
88: \begin{document}
89:
90: \title{Locked and Unlocked Chains of Planar Shapes%
91: \thanks{%
92: A preliminary extendend abstract of this paper appeared in
93: \emph{Proceedings of the 22nd Annual ACM Symposium on Computational Geometry},
94: June 2006, \protect \cite{conf-version}.
95: R. Connelly is supported in part by NSF grant DMS-0209595.
96: E.~Demaine is supported in part by NSF grant CCF-0347776
97: and DOE grant DE-FG02-04ER25647.
98: S.~Langerman is Chercheur qualifi\'e du FNRS.
99: J.~Mitchell is supported in part by
100: %the National Science Foundation
101: NSF
102: (CCF-0431030, CCF-0528209, CCF-0729019),
103: Metron Aviation, and NASA Ames.}}
104:
105: %\title{Sharp bounds on the lockedness of unsharp shapes}
106: %\title{The legend of the lockedness monster}
107:
108:
109: \author{%
110: Robert Connelly%
111: \thanks{Department of Mathematics, Cornell University,
112: Ithaca, NY 14853, USA. \protect\url{connelly@math.cornell.edu}}
113: \and
114: Erik D. Demaine%
115: \thanks{MIT Computer Science and Artificial Intelligence Laboratory,
116: 32 Vassar St., Cambridge, MA 02139, USA,
117: \protect\url{{edemaine,mdemaine}@mit.edu}}
118: \and
119: Martin L. Demaine%
120: \footnotemark[2]
121: \and
122: S\'andor P. Fekete%
123: \thanks{Department of Computer Science, Braunschweig University of Technology,
124: M\"uhlenpfordtstr. 23, D-38106 Braunschweig, Germany.
125: \protect\url{s.fekete@tu-bs.de}}
126: \and
127: Stefan Langerman%
128: \thanks{Chercheur qualifi\'e du FNRS, Universit\'e Libre de Bruxelles,
129: D\'epartement d'informatique, ULB CP212, Belgium.
130: \protect\url{Stefan.Langerman@ulb.ac.be}}
131: \and
132: Joseph S. B. Mitchell%
133: \thanks{Department of Applied Mathematics and Statistics,
134: Stony Brook University, Stony Brook, NY 11794-3600, USA.
135: \protect\url{jsbm@ams.sunysb.edu}}
136: \and
137: Ares {Rib\'o}%
138: \thanks{Institut f\"ur Informatik, Freie Universit\"at Berlin,
139: Takustra{\ss}e 9, D-14195 Berlin, Germany.
140: \protect\url{{ribo,rote}@inf.fu-berlin.de}}
141: \and
142: G\"unter Rote%
143: \footnotemark[6]
144: }
145:
146: \date{}
147: \maketitle
148:
149:
150: \begin{abstract}
151: We extend linkage unfolding results from the well-studied case of
152: polygonal linkages to the more general case of linkages of polygons.
153: More precisely, we consider chains of nonoverlapping rigid planar shapes
154: (Jordan regions) that are hinged together sequentially at rotatable joints.
155: Our goal is to characterize the familes of planar shapes that admit
156: \emph{locked chains}, where some configurations cannot be reached by
157: continuous reconfiguration without self-intersection, and which families of
158: planar shapes guarantee \emph{universal foldability}, where every chain is
159: guaranteed to have a connected configuration space.
160: Previously, only obtuse triangles were known to admit locked shapes,
161: and only line segments were known to guarantee universal foldability.
162: We show that a surprisingly general family of planar shapes,
163: called \emph{slender adornments}, guarantees universal foldability:
164: roughly, the distance from each edge along the path along the boundary of the slender adornment to each hinge should be monotone.
165: In contrast, we show that isosceles triangles with any desired apex angle
166: $< 90^\circ$ admit locked chains, which is precisely the threshold beyond
167: which the slender property no longer holds.
168: \end{abstract}
169:
170: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
171:
172: %\keywords{Linkages, folding, locked chains, hinged dissections.}
173:
174:
175:
176: \section{Introduction}
177:
178: In this paper, we explore the motion-planning problem of \emph{reconfiguring}
179: or \emph{folding} a hinged collection of rigid objects from one state to
180: another while avoiding self-intersection.
181: This general problem has been studied since the beginnings
182: of the motion-planning literature when Reif \cite{r-cmpg-79}
183: proved that deciding reconfigurability of a ``tree'' of polyhedra,
184: amidst fixed polyhedral obstacles, is PSPACE-hard.
185: This result has been strengthened in various directions over the years,
186: although the cleanest versions were obtained only very recently:
187: deciding reconfigurability of a tree of line segments in the plane, and
188: deciding reconfigurability of a chain of line segments in 3D,
189: are both PSPACE-complete \cite{Alt-Knauer-Rote-Whitesides-2004}.
190: This result is tight in the sense that deciding reconfigurability
191: of a chain of line segments in the plane is easy, in fact, trivial:
192: the answer is always yes \cite{Connelly-Demaine-Rote-2003}.
193:
194: These results illustrate a rather fine line in reconfiguration problems between
195: computationally difficult and computationally trivial. The goal of our work
196: is to characterize what families of planar shapes and hingings lead to the
197: latter outcome,
198: a \emph{universality result} that reconfiguration is always possible.
199: The only known example of such a result, however, is the family of chains
200: of line segments, and that problem was unsolved for about 25 years
201: \cite{Connelly-Demaine-Rote-2003}. (A \emph{chain} is a sequence of line segments joined end-to-end that are disjoint except except for consecutive endpoints, where they are hinged. It is called an \emph{open chain} when the first line segment is not joined to the last, and \emph{closed} when it is joined to the last segment in a closed cycle.) Even small perturbations to the problem,
202: such as allowing a single point where three line segments join, leads to
203: \emph{locked} examples where reconfiguration is impossible
204: \cite{Connelly-Demaine-Rote-2002-infinitesimally-locked}.
205:
206: What about chains of shapes other than line segments?
207: It is easy to see that a shape tucked into a ``pocket'' of a nonconvex shape
208: immediately makes trivial locked chains with two pieces.
209: Back in January~1998, the third author showed how to simulate this behavior
210: with convex shapes, indeed, just three triangles;
211: see Figure \ref{locked 3 triangles}.
212: This example has circulated throughout the years to many researchers
213: (including the authors of this paper) who have asked about chains of 2D shapes.
214: The only really unsatisfying feature of the example is that some of the angles
215: are very obtuse. But with a little more work, one can find examples with
216: acute angles, indeed, equilateral triangles, albeit of different size;
217: see Figure \ref{locked unequal equilateral}. What could be better than
218: equilateral triangles?
219:
220: \begin{figure}[htbp]
221: \centering
222: \subfigure[]
223: {\includegraphics[scale=0.7]{locked_3_triangles}
224: \label{locked 3 triangles}}\hfil\hfil
225: \subfigure[]
226: {\includegraphics[scale=0.7]{locked_unequal_equilateral}
227: \label{locked unequal equilateral}}
228: \caption{Simple examples of locked chains of triangles.
229: (a) A locked chain of three triangles.
230: (b) A locked chain of equilateral triangles of different sizes.
231: The gaps should be tighter than drawn.}
232: \label{locked few triangles}
233: \end{figure}
234:
235: It is therefore reasonable to expect, as we did for many years, that
236: there is no interesting class of shapes, other than line segments,
237: with a universality result---essentially all other shapes admit locked chains.
238: We show in this paper, however, that this guess is wrong.
239:
240: We introduce a family of shapes, called \emph{slender adornments},
241: and prove that all open
242: chains, made up of
243: arbitrarily many different shapes from this family,
244: can be universally reconfigured between any two states.
245: Indeed, we show that these chains have a natural canonical configuration,
246: analogs of the straight configuration of an open chain.
247: Our result is based on the existence of
248: ``expansive motions'', proved in
249: \cite{Connelly-Demaine-Rote-2003}.
250: Our techniques build on the theory of unfolding chains of line segments,
251: substantially generalizing and extending the results from that theory. Indeed,
252: the results in this paper essentially piggy-back onto the results of \cite{Connelly-Demaine-Rote-2003} or any of the other results and algorithms such as \cite{Streinu-2000,Streinu-2005}
253: that provide a continuous expansive motion of the base chain.
254: Our results go far beyond what \emph{we} thought was possible
255: (until recently). As part of the methods that we describe here, we also consider discrete expansive motions of the base chain, that do not necessarily come from a continuous expansive motion. (A \emph{discrete expansion} of a chain $C$ is simply another corresponding chain $C'$ such that if $x$ and $y$ are two points in $C$ the distance between corresponding points $x'$ and $y'$ is not smaller than the distance between $x$ and $y$.) In that case if all the slender adornments are \emph{symmetric} under the reflection about the line of the base chain, then any expansive discrete motion of the base chain will have the property that the attached adornments will not overlap. It turns out that the continuous case, when the adornments are not necessarily symmetric, follows from the discrete symmetric case.
256:
257: The family of slender adornments has several equivalent definitions.
258: The key idea is to distinguish the two hinge points on the boundary
259: of the shape connecting to the adjacent shapes in the chain,
260: and view the shape as an \emph{adornment} to the line segment connecting
261: those two hinges, called the \emph{base}. This view is without loss of
262: generality, but provides additional information relating the shapes and
263: how they are attached to neighbors,
264: which turns out to be crucial to obtaining a universality result.
265: An adornment is \emph{slender} if the distance from either endpoint of the base
266: to a point moving along one side of the adornment
267: changes monotonically. If the boundary curve, defining the adornment,
268: is sufficiently smooth, it is slender if and only if every inward
269: normal of the shape hits the base. Equivalently, an adornment is
270: slender if it is the union, in each half-plane having the base as a
271: boundary, of the intersection of pairs of disks centered at the two
272: endpoints of the base.
273: For the first and last edge of an open chain, there may be some freedom in defining the base.
274:
275: \begin{figure}[htbp]
276: \centering
277: \includegraphics[width=\linewidth]{slender-new}
278: \caption{Examples of slender adornments. The base is drawn bold. The examples in the top row are symmetric. Any two of these examples can be glued together along a common base so that the union also becomes a slender adornment.}
279: \label{slender}
280: \end{figure}
281:
282:
283: Slender adornments are quite general.
284: Figure \ref{slender} shows several examples of slender adornments.
285: These examples are themselves slender adornments, but also any pair can be
286: joined along their bases so that the union makes another slender adornment.
287: Our results imply that one can take any of these slender adornments,
288: link the bases together into an open chain in any way that the chain does not
289: self-intersect, and the resulting chain can be unfolded without
290: self-intersection to a straight configuration,
291: and thus the chain can be folded without self-intersection
292: into every configuration.
293:
294: We also demonstrate the tightness of the family of slender adornments by
295: giving examples of locked chains of shapes that are not slender.
296: Specifically, we show that, for any desired angle $\theta < 90^\circ$,
297: there is a locked chain of isosceles triangles with apex angle~$\theta$.
298: This is precisely the family of isosceles-triangle adornments
299: that are not slender.
300: Thus, for chains of triangles, obtuseness is really desirable,
301: contrary to our intuition from Figure~\ref{locked 3 triangles}:
302: the key is that the apex angle opposite the base (in the adornment view)
303: be nonacute, not any other angle.
304: The proof that our examples are locked uses the self-touching theory
305: developed for trees of line segments
306: in \cite{Connelly-Demaine-Rote-2002-infinitesimally-locked}.
307:
308: \begin{figure}[htbp]
309: \centering
310: \includegraphics[width=0.5\linewidth]{dudeney4}
311: \caption{Hinged dissection of square to equilateral triangle,
312: described by Dudeney~\protect\cite{Dudeney-1902-hinged}.
313: Different shades show different folded states (overlapping slightly).}
314: \label{dudeney}
315: \end{figure}
316:
317: \paragraph{Motivation.}
318: Hinged collections of rigid objects have been studied
319: previously in many contexts, particularly robotics.
320: One recent body of algorithmic work by Cheong et al.~\cite{csgor-ihp-06}
321: considers how chains of polygonal objects can be \emph{immobilized}
322: or \emph{grasped} by a robot with a limited number of actuators.
323: Grasping is a natural first step toward robotic \emph{manipulation},
324: but the latter challenge requires a better understanding of reconfigurability.
325: This paper offers the first theoretical underpinnings for reconfiguration
326: of chains of rigid objects (other than line segments).
327:
328: Another potential application is to continuous folding of
329: hinged dissections. Hinged dissections are chains or trees of polygons
330: that can be reconfigured into two or more self-touching configurations
331: with desired silhouettes. For example, Figure~\ref{dudeney} shows
332: a classic hinged dissection from 1902 of a square into an equilateral
333: triangle of the same area. Many general families of hinged dissections
334: have been established in the recent literature
335: \cite{Akiyama-Nakamura-1998,
336: Demaine-Demaine-Eppstein-Frederickson-Friedman-2005,
337: Demaine-Demaine-Lindy-Souvaine-2005,
338: Eppstein-2001-mirror-dissection,
339: Frederickson-2002}.
340: One problem not addressed in this literature, however, is whether the
341: reconfigurations can actually be executed without self-intersection,
342: as in Figure~\ref{dudeney}.
343: Our results provide potential tools, previously lacking,
344: for addressing this problem.
345: While hinged dissections have frequently been considered in recreational
346: contexts, they have recently found applications in nanomanufacturing
347: \cite{Mao-Thallidi-Wolfe-Whitesides-Whitesides-2002} and reconfigurable
348: robotics \cite{Demaine-Demaine-Lindy-Souvaine-2005}.
349:
350: \paragraph{Outline.}
351: This paper is organized as follows.
352: Section \ref{Slender Adornments} defines the model and slender adornments
353: more precisely, and proves several basic properties.
354: Section \ref{Symmetric Adornments} describes the case when each adornment is symmetric about its base and is important for proving,
355: in Section \ref{Slender Adornments Cannot Lock}, that simple chains of
356: slender adornments can always be unfolded so that the base is convex or straight. In Section \ref{generalizations} we discuss the situation when the adornments are permitted to overlap.
357: Section \ref{locked sharp} describes our examples of locked chains of
358: isosceles triangles, including the necessary background
359: from self-touching trees. The
360: conclusion (Section~\ref{sec:conclusion}) includes several closing remarks.
361:
362: The results of this paper have been reviewed and sketched in the
363: survey~\cite[Section~4]{Bobs-survey}, which cites this paper (in its
364: full form) as the source of the results and the proofs.
365:
366: \section{Slender Adornments}
367: \label{Slender Adornments}
368:
369: This section provides a formal statement of the objects we consider---adorned
370: chains consisting of slender adornments---and proves several basic results
371: about them.
372:
373: \subsection{Adorned Chains}
374:
375: Our object of study is a chain of nonoverlapping rigid planar shapes
376: (Jordan regions) that are hinged together sequentially at rotatable joints.
377: Another way to view such a chain is to consider the \emph{underlying polygonal
378: chain}, the \emph{core}, of line segments connecting successive joints.
379: (For an open chain, there is some freedom in choosing the endpoints for the
380: first and the last bar of the chain.)
381: On the one hand, these line segments can be viewed as \emph{bars} that move
382: rigidly with the shapes to which they belong.
383: On the other hand, the shapes can be viewed as ``adornments'' to the bars
384: of an underlying polygonal chain.
385: This view leads to the concept of an ``adorned polygonal chain'',
386: which we now proceed to define more precisely.
387:
388: An \emph{adornment} is a simply connected compact region in the plane,
389: called the \emph{shape},
390: together with a line segment $x y$ connecting two boundary points,
391: called the \emph{base}. There are two \emph{boundary arcs} from $x$
392: to $y$ that enclose the shape, called \emph{sides}.
393: We require the base to be contained in the shape; i.e., the base must be a chord of the shape.
394:
395: We say that two distinct adornments \emph{overlap} when some point of one adornment lies in the interior of the other, and we insist that the relative interiors of the base chains be disjoint.
396: Thus, the bases of two shapes are not allowed to touch except at common hinges of the polygonal chain. An \emph{adorned polygonal chain} is a set of nonoverlapping adornments whose
397: bases form a polygonal chain. We permit the shapes to touch on their boundary and to slide along each other.
398:
399: For our main result, Theorem \ref{thm:main}, where we assume that the motion of the base is expansive, it is not necessary to assume that the base chain is simple. It can be any finite embedded graph with straight edges whose relative interiors are pairwise disjoint; a vertex may touch an edge. When the base chain is simple the results of \cite{Connelly-Demaine-Rote-2003} or \cite{Streinu-2005} guarantee that there is such an expansive motion. On the the other hand, although an expansive motion of the base chain of a strictly simple closed polygon to a convex convex configuration can be guaranteed, it may happen that two realizations are not in the same configuration component, as shown in Figure \ref{fig:2-components}, and in the conclusion (Section~\ref{sec:conclusion}) there is a description of a case when there are infinitely many components in the configuration space.
400:
401: The viewpoint of a chain of shapes as an adorned polygonal chain is useful
402: for two reasons.
403: First, we can more easily talk about the kinds of shapes,
404: and their relation to the locations of the incident hinges,
405: in a family of chains: this information is captured by the adornments.
406: Second, the underlying polygonal chain provides a mechanism for folding
407: the chain of shapes, as well as a natural \emph{unfolding} goal:
408: straighten the underlying open chain or convexify the underlying closed chain.
409: Indeed, we show that, in some cases, unfolding motions
410: of the polygonal chain induce valid unfolding motions of the chain of shapes.
411:
412:
413:
414: \subsection{Slender Adornments}
415:
416: An adornment is defined to be \emph{slender} if, for a point moving on either side of
417: the shape, the distance to each endpoint of the base changes monotonically (possibly not strictly
418: monotonically). An adornment is called \emph{symmetric} if it is symmetric about the line through the
419: base. An adornment is called \emph{one-sided} if it lies in just one of the closed half-planes whose
420: boundary contains the base. Clearly, a general adornment is the union of two one-sided adornments, and a one-sided adornment is the intersection of a symmetric adornment with a closed half-plane whose boundary contains the base. For a base segment $[x,y]$ and a point $z$ in the plane, where $\|z-x\| \le \|y-x\|$ and $\|z-y\| \le \|x-y\|$, let $L(z)$ be the intersection of the two disks with $z$ on its boundary, centered at $x$ and
421: at $y$. We call $L(z)$ the \emph{lens determined by $z$ associated to the base $[x, y]$}. A \emph{half-lens}, denoted as $\hat{L}(z)$, is the intersection of $L(z)$ and the closed half-plane through the base containing $z$. See Figure \ref{fig:lenses} for a picture of a half-lens and lens.
422: The following are some simple, but useful, properties of lenses.
423:
424: \begin{proposition} \label{prop:lenses} For any point $z$ in a \textup(symmetric\textup) slender adornment $A$, $\hat{L}(z) \subset A$ \textup($L(z) \subset A$\textup).
425: \end{proposition}
426: \begin{proof} Let $z$ be a point on the defining boundary of $A$. Since the distance to $x$ along the boundary is monotone, no point along the path from $z$ to $y$ intersects the interior of the circle centered at $x$ through $z$. Similarly, no point along the path from $z$ to $x$ intersects the interior of the circle centered at $y$ through $z$. Thus, the intersection of the circular disks centered at $x$ and $y$, with $z$ on their boundary, and the closed half-plane containing $z$, $\hat{L}(z)$ is contained in the adornment. In the symmetric case, the intersection of the circular disks with $z$ on their boundary $L(z)$ is contained in $A$. See Figure \ref{fig:lenses}.
427: \end{proof}
428: \begin{figure} [here]
429: \begin{center}
430: \includegraphics{lenses}
431: \caption{ (a) a half-lens in non-symmetric adornment. (b) a symmetric lens with a point $z$ in the interior of the lens and the adornment.}
432: \label{fig:lenses}
433: \end{center}
434: \end{figure}
435: \begin{proposition} \label{prop:lens-interior} For any point $z$ in the interior of a \textup(symmetric\textup) slender adornment $A$, there is a half-lens $\hat{L}(z') \subset A$ \textup(lens $L(z') \subset A$\textup) that has $z$ in its interior.
436: \end{proposition}
437: \begin{proof} The half-lens (lens) through $z$ is contained in $A$ by Proposition \ref{prop:lenses}. Since $z$ is in the interior of $A$, there is another point $z'$ in $A$ on the line perpendicular to the base segment slightly further away from the base. Then $z$ is in the interior of the half-lens (lens) defined by $z'$.
438: \end{proof}
439: \begin{proposition} \label{prop:lens-union} A symmetric adornment $A$
440: of a base $[x,y]$ is slender if and only if it is the union of the
441: intersection of pairs of disks centered at $x$ and $y$.
442: \end{proposition}
443: \begin{proof}
444: Assume that $A$ is a slender adornment.
445: By Proposition \ref{prop:lenses},
446: the union of the lenses $L(z)$ for $z$ on the boundary of $A$ is
447: contained in~$A$.
448:
449: To show the reverse containment, any point $z$ in the interior of $A$
450: lies on a circle centered at $x$, and this circle must intersect the
451: boundary of $A$ in (at least) one point $z'$. Then $z$ is
452: in $L(z')$. Thus, the union of the lenses $L(z')$ for $z'$ on the
453: boundary of the slender adornment contains $A$.
454:
455: For the converse implication,
456: assume that we have a symmetric adornment $A$ which is formed as the union of lenses,
457: as stated in the proposition.
458: We first show that for two points $z$ and $z'$ on
459: the boundary of $A$,
460: we cannot have
461: $\|z-x\| < \|z'-x\|$ and
462: $\|z-y\| < \|z'-y\|$.
463: If this were the case, some lens on whose boundary $z'$ lies would contain $z$ in its interior, a contradiction.
464: Since we can exchange the role of $z$ and $z'$, we conclude
465: \begin{equation}
466: \label{two-boundary-points}
467: \|z-x\| < \|z'-x\| \implies \|z-y\| \ge \|z'-y\|
468: \text{, \ and \ }
469: \|z-x\| > \|z'-x\| \implies \|z-y\| \le \|z'-y\|
470: \end{equation}
471: Now, if we order the points $z$ on one boundary chain from $x$ to $y$ by the quantity
472: $\|z-x\|-\|z-y\|$, we conclude that the distance
473: $\|z-x\|$ cannot decrease and the distance $\|z-y\|$ cannot increase; otherwise we would derive a contradiction to~\eqref{two-boundary-points}.
474: \end{proof}
475: \begin{proposition} Finite unions and arbitrary intersections of slender adornments are slender adornments.
476: \end{proposition}
477: \begin{proof} This follows from Proposition \ref{prop:lens-union} in the symmetric case, and the non-symmetric case follows from the symmetric case by intersecting with the closed half-plane containing the line segment.
478: \end{proof}
479: \begin{proposition} Every slender adornment is contained in the symmetric lens determined by either of the points equidistant from the endpoints of the base as in Figure \ref{crescent}.
480: \end{proposition}
481: \begin{proof} Any slender adornment must be contained in the disk through the other end of the base, and thus it is in the intersection of those two disks.
482: \end{proof}
483: \begin{figure}[htbp]
484: \centering
485: \begin{overpic}[scale=0.5]{crescent}
486: \put(48,65){$z$}
487: \put(26,31.5){$x$}
488: \put(70,32){$y$}
489: \end{overpic}
490: \caption{The largest slender adornment with a given base is a lens $L(z)$, where $\|z-x\| = \|z-y\| = \|y-x\|.$}
491: \label{crescent}
492: \end{figure}
493:
494:
495:
496: \subsection{Alternate Definitions of Slender Adornments}
497:
498: In the preliminary conference version of this paper~\cite{conf-version}, we used
499: a more restricted definition of slenderness:
500: suppose that the boundary of an adornment consists of two
501: differentiable curves between the base points.
502: %
503: Then the condition of being slender is equivalent to requiring that every inward normal of the shape intersects the base before exiting the shape, as
504: illustrated in Figure~\ref{slender versus monotone}.
505: \begin{figure}[h!]
506: \centering
507: \vspace*{-2ex}
508: \includegraphics{slender_versus_monotone}
509: \caption{The normal property for slender shapes.}
510: \label{slender versus monotone}
511: \end{figure}
512: This
513: property ensures that slender adornments will not get closer together during an
514: expansive motion of the base. Our current slenderness definition by
515: the monotone distance property is more general, easier to handle, and
516: it does not raise questions of differentiability.
517:
518: \subsection{Kirszbraun's Theorem}
519: \label{Kirszbraun}
520:
521: In what follows it is very handy to have the following theorem of Kirszbraun \cite{Kirszbraun-1934}.
522:
523: \begin{theorem}\label{thm:Kirszbraun} Suppose a finite set of closed circular disks in Euclidean space are rearranged so that no pair of centers gets strictly closer together. If the original set has an empty intersection, so does the rearranged set.
524: \end{theorem}
525:
526: There is a discussion and proof of this in \cite{Alexander-1984} as well as references to other proofs. We only need this result for four disks in the Euclidean plane.
527:
528:
529: \section{Expanded Slender Symmetric Adornments Never Overlap}
530: \label{Symmetric Adornments}
531:
532: We first prove the following for the case of symmetric slender adornments. Note that the following result is for discrete expansions of the base chain. Recall that two adornments \emph{overlap} if a point in one adornment lies in the interior of the other. This allows their boundaries to touch, but not to penetrate each other. Note that the bases of a chain do not cross as well, by the expansive property of a discrete motion. We do not need the continuous expansive property for this result.
533:
534: \begin{theorem}\label{thm:symmetric} Consider two configurations $X$ and $Y$ of corresponding chains with symmetric slender adornments such that the base chain of $Y$ is an expansion of the base chain of $X$. We assume that the adornments attached to the base chain of $X$ do not overlap. Then, when the corresponding adornments are attached to the base chain of $Y$, they also do not overlap.
535: \end{theorem}
536: \begin{proof} Suppose $A_X$ and $B_X$ are two slender adornments attached to different links of the base chain of $X$, and $A_X$ and $B_X$ do not overlap. Let $A_Y$ and $B_Y$ be the corresponding adornments for $Y$. Suppose that $z$ is a point in the intersection $A_Y \cap B_Y$, where $z$ is in the interior of, say, $A_Y$. We wish to find a contradiction.
537:
538: Let $z_A$ and $z_B$ be the corresponding distinct points in $A_X$ and $B_X$, respectively, that map to $z$ under the expanding map of their bases. Thus, the lenses $L_A(z_A)$ and $L_B(z_B)$ for $A_X$ and $B_X$ have disjoint interiors, since the adornments do not overlap. Since $z$ is in the interior of $A_Y$, we can assume that $L_A(z_A)$ can be chosen so that the closed lenses $L_A(z_A)$ and $L_B(z_B)$ are disjoint also. Thus the four circular disks that correspond to the circular disks that define $L_A(z_A)$ and $L_B(z_B)$ have an empty intersection. By Kirszbraun's Theorem \ref{thm:Kirszbraun}, and the expansion property of the endpoints of the bases of $A_X$ and $B_X$, which are the centers of the four circular disks, the intersection of the corresponding lenses for $A_Y$ and $B_Y$ must also be empty, contradicting the assumption that $A_Y$ and $B_Y$ overlap. See Figure \ref{fig:symmetric-adornments}.
539: \end{proof}
540:
541: \begin{figure} [here]
542: \begin{center}
543: \includegraphics{symmetric-adornments}
544: \caption{This is the situation when two adornments overlap. The four circles that used in the application of Kirszbraun's Theorem are indicated. Note that, in this figure, the motion from $X$ to $Y$ is not an expansion, since that would contradict Theorem \ref{thm:symmetric}.} \label{fig:symmetric-adornments}
545: \end{center}
546: \end{figure}
547:
548: For discrete expansions, it is not possible to deal with non-symmetric adornments. Figure \ref{fig:non-symmetric-overlap} shows an example of two chains with corresponding slender adornments, one an expansion of the other. One starts with no overlap, and the other has such an overlap.
549: \begin{figure} [here]
550: \begin{center}
551: \includegraphics{non-symmetric-overlap}
552: \caption{This shows two chains, with slender but not symmetric adornments, where one is an expansion of the other, while there is an overlap in the expanded configuration, but not the original.}
553: \label{fig:non-symmetric-overlap}
554: \end{center}
555: \end{figure}
556:
557:
558: \section{Slender Adornments Cannot Lock}
559: \label{Slender Adornments Cannot Lock}
560:
561: We now consider the general case, assuming a continuous expansive motion.
562:
563: \begin{theorem} \label{thm:main} Suppose there is a continuous expansive motion of the base chain with slender non-overlapping, not necessarily symmetric, adornments attached. Then the adornments never overlap during the motion.
564: \end{theorem}
565: \begin{proof}Because of the expansive property, two segments of the base chain can only intersect at common endpoints of adjacent segments. Thus, suppose $z_A$, in the interior of adornment $A$, intersects $z_B$ in adornment $B$ at some time $t_1$ during the motion. We look for a contradiction. By Proposition \ref{prop:lens-interior}, there is a closed half-lens $L_A$ for $A$ that contains $z_A$ in its interior and there is a first time $t_0 < t_1$ when $L_A$ intersects another half-lens $L_B$ for $B$ that contains $z_B$. Necessarily, that intersection must be on the common boundary of $L_A$ and $L_B$. (Note that $L_B$ could be a single point on a base segment.) Then there are three cases that can occur. In each case, we will show that when the motion is continued from $t_0$ to $t_1$, $z_A$ and $z_B$ cannot intersect.
566: \begin{enumerate} [{Case} 1:]
567: \item The bases of $A$ and $B$ intersect in the interior of at least one of the bases. This cannot happen because the bases are initially disjoint and the motion is expansive. See Figure \ref{fig:overlap}(a).
568: \item The base of $A$ or $B$ intersects the half-lens of the other. The half lens can be extended to a full symmetric lens without overlaping the base of the other. Applying Theorem \ref{thm:symmetric} we see that $z_A$ and $z_B$ cannot intersect upon further expansion. See Figure \ref{fig:overlap}(b).
569: \item The half lenses of $A$ and $B$ intersect. In this case both half lenses can be extended to non-overlapping symmetric lenses. Again we apply Theorem \ref{thm:symmetric} to see that $z_A$ and $z_B$ cannot intersect upon further expansion. See Figure \ref{fig:overlap}(c).
570: \end{enumerate}
571: \begin{figure} [here]
572: \begin{center}
573: \includegraphics{overlap}
574: \caption{This presents the cases when one adornment with its base might start to overlap with the other. The dashed lines indicate where one or both of the lens of the adornment can be extended so that it does not intersect the relevant part of the other. The thick lines indicate the part of the adornment that is not to be penetrated by the other lens or base. The thin lines indicate where some of the rest of the adornment might lie, containing the point $z_A$, say, in the proof.} \label{fig:overlap}
575: \end{center}
576: \end{figure}
577: \end{proof}
578:
579: \begin{corollary}\label{cor:simple-chains} A strictly simple polygonal chain with slender adornments attached can always be straightened or convexified by a continuous motion.
580: \end{corollary}
581: \begin{proof} By \cite{Connelly-Demaine-Rote-2003}, there is a continuous expansive motion of the base chain, where the final configuration is convex in the case of a closed chain and straight in the case of an open chain. Then Theorem \ref{thm:main} implies that they can be carried along without overlap.
582: \end{proof}
583:
584:
585: \begin{corollary} \label{cor:open-chains} A strictly simple open
586: polygonal chain with slender adornments
587: can be continuously reconfigured between any
588: two states.
589: \end{corollary}
590: \begin{proof} By Corollary \ref{cor:simple-chains}, both chains can be continuously expansively reconfigured so that the base chains are straight.
591: Thus, one state can be expanded to have a straight base configuration, and then contracted to the other configuration by running its expansion backwards.
592: \end{proof}
593:
594: It is interesting to note that the conclusion of Corollary \ref{cor:open-chains} does not hold for closed chains, even though any two convex chains with no adornments can be continuously reconfigured from one to the other. Figure \ref{fig:2-components} shows an example, where the configuration space has two components, where the base chain is a quadrilateral, and where each adornment is a triangle attached to its base.
595: \begin{figure} [here]
596: \begin{center}
597: \includegraphics{2-components}
598: \caption{Two configurations (a) and (c) of a quadrilateral with two
599: slender adornments attached. It is not possible to continuously move
600: from one to the other without colliding. Figure (b) shows how the
601: two adornments collide as the quadrilateral is deformed from (a) to
602: (c). }
603: \label{fig:2-components}
604: \end{center}
605: \end{figure}
606:
607: Indeed, in the conclusion (Section~\ref{sec:conclusion}) it is shown how to create a quadrilateral with two slender adornments such that the configuration space has infinitely many components.
608:
609: It is also interesting to note that
610: when the base chain is expanded, it often happens that the motion on the
611: adorned configuration is not expansive.
612: Figure~\ref{fig:non-expansive} shows an example.
613: \begin{figure} [here]
614: \begin{center}
615: \includegraphics{non-expansive}
616: \caption{The base chain of Figure (a) expands to Figure (b). But the dark points on the corresponding slender adornments get closer together.}
617: \label{fig:non-expansive}
618: \end{center}
619: \end{figure}
620:
621: \section{Generalizations: Overlapping Adornments and Generalized Slender Symmetric Adornments}\label{generalizations}
622:
623: In the discussion so far, we have assumed, when the adornments are attached to their chains, that they do not overlap. What happens when the slender adornments do overlap? It turns out that we can apply some of the results of \cite{Bezdek-Connelly-2002} related to problems concerning areas of unions and intersections of circular disks in the plane to the case when the adornments are all symmetric.
624:
625: Proposition \ref{prop:lens-union} shows that any symmetric adornment is the infinite union of symmetric lenses $L(z)$ for all $z$ on the boundary of the adornment. To apply the theory of \cite{Bezdek-Connelly-2002} it is more convenient that there only be a finite number of sets involved in the union of lenses. But it is easy to see that each adornment can be approximated by a finite union of lenses.
626:
627: We first define a \emph{flower} as a set in the plane that can be described in terms of finite unions and intersections of circular disks, where each disk appears once and only once in the Boolean expression that describes the set. For the special case at hand we need only be concerned with flowers $F$ of the following sort:
628: \begin{equation}
629: F=(B_1 \cap B_2) \cup (B_3 \cap B_4) \cup (B_5 \cap B_6) \cup \dots \cup (B_{N-1} \cap B_{N}) ,\label{flower}
630: \end{equation}
631: where each $B_i, i=1, \dots, N$ is a circular disk in the plane. Flowers were defined by \cite{Gordon-Meyer-1995}, and a special case of Corollary 8 in \cite{Bezdek-Connelly-2002} shows the following. Let $B(x, r)$ denote a disk in the plane of radius $r$ centered at $x$.
632: \begin{lemma}\label{monotone-area}
633: Let $B(p_i, r_i)$ and $B(q_i, r_i), i= 1, \dots, N$ be two sets of planar disks, where $\|p_i-p_{i+1}\| \ge \|q_i-q_{i+1}\|$ for $i$ odd, and $\|p_i-p_{i+1}\| \le \|q_i-q_{j}\|$ for all other pairs $i<j$. Then the area of the flower $F$ in \eqref{flower} defined for the configuration of $p_i$ is less than or equal to the area of the flower $F$ defined for the configuration of $q_i$.
634: \end{lemma}
635:
636: The crucial observation is that the union of the slender adornments can be approximated by flowers. Each lens is the intersection of two disks, one of the terms in (\ref{flower}), and Proposition \ref{prop:lens-union} implies the following.
637:
638:
639: \begin{theorem} \label{thm:overlap} Suppose that one chain is a discrete expansion of the other and slender symmetric adornments are attached to each chain. Then the area of the union of the adornments does not decrease.
640: \end{theorem}
641: Figure \ref{fig:area-overlap} shows an example of overlapping
642: symmetric adornments. Figure \ref{fig:area-down} shows an example of
643: a chain with non-symmetric adornments that expands to another chain
644: and the area of the slender adornments with an expanded core
645: decreases.
646:
647: \begin{figure} [here]
648: \begin{center}
649: \includegraphics{area-overlap}
650: \caption{Three intervals with overlapping symmetric slender adornments. }
651: \label{fig:area-overlap}
652: \end{center}
653: \end{figure}
654:
655: \begin{figure} [here]
656: \begin{center}
657: \includegraphics{area-down}
658: \caption{An example of two chains with non-symmetric slender adornments, where the expanded chain with adornments has smaller area.}
659: \label{fig:area-down}
660: \end{center}
661: \end{figure}
662:
663: Another possible generalization is to attach the analog of slender
664: adornments to simplicial complexes in higher dimensions. For example,
665: a set $A$ in three-space would be called \emph{slender with respect to
666: a triangle base $B$} if for any plane $P$ perpendicular to the plane
667: of $A$, $P \cap A$ is slender with respect to $B\cap A$. Then the
668: analog of
669: Theorem~\ref{thm:main}
670: should hold using
671: the notion of symmetric slender adornments. Even the analog of
672: Theorem~\ref{thm:overlap} for the volumes of symmetric slender
673: adornments would still hold, but it could only be asserted for
674: continuous expansions of the base chain. The higher dimensional
675: version of Corollary 8 in \cite{Bezdek-Connelly-2002}, on which
676: Lemma~\ref{monotone-area} is based, is not known for discrete
677: expansions. However, in \cite{Csikos-2001}, there is a continuous
678: version that will suffice. In higher dimensions, the idea is to
679: assume simply that the base chain, to which the adornments are
680: attached, is expanded.
681:
682: \section{Locked Chains of Sharp Triangles}
683: \label{locked sharp}
684:
685: An isosceles triangle with an apex angle of $\geq 90^\circ$
686: and with the nonequal side as the base is a slender adornment.
687: By Corollary~\ref{cor:simple-chains}, any chain of such triangles
688: can be straightened. In this section we show that this result is tight:
689: for any isosceles triangle with an apex angle of $< 90^\circ$
690: and with the nonequal side as a base, there is a chain of these triangles
691: that cannot be straightened.
692:
693: \begin{figure}
694: \centering
695: \subfigure[]
696: {\includegraphics[scale=0.6]{locked_9_equilateral_loose}
697: \label{locked 9 equilateral loose}}
698: \hfil\hfil
699: \subfigure[]
700: {\includegraphics[scale=0.6]{locked_9_equilateral_tight}
701: \label{locked 9 equilateral tight}}
702: \caption{A locked chain of nine equilateral triangles.
703: (a) Drawn loosely. Separations should be smaller than they appear.
704: (b) Drawn tightly, with no separation, as a self-touching configuration.}
705: \label{locked 9 equilateral}
706: \end{figure}
707:
708: Figure~\ref{locked 9 equilateral loose} shows the construction for equilateral
709: triangles (of slightly different sizes).
710: This figure is drawn with the pieces loosely separated, but the
711: actual construction has arbitrarily small separations and arbitrarily closely
712: approximates the self-touching geometry shown in
713: Figure~\ref{locked 9 equilateral tight}.
714: Stretching the triangles in this self-touching geometry,
715: as shown in Figure~\ref{locked 9 isosceles tight},
716: defines our construction for any isosceles triangles with an opposite angle of
717: any value less than $90^\circ$.
718: In this case, however, our construction uses two different scalings
719: of the same triangle.
720:
721: \begin{figure}
722: \centering
723: \includegraphics[scale=0.6]{locked_isosceles_tight.eps}
724: \caption{Variations on the self-touching configuration from
725: Figure~\protect\ref{locked 9 equilateral tight}
726: to have any desired angle $< 90^\circ$ opposite the base of each triangle.}
727: \label{locked 9 isosceles tight}
728: \end{figure}
729:
730: \subsection{Theory of Self-Touching Configurations}
731:
732: This view of the construction as a slightly separated version of a
733: self-touching configuration allows us to apply the program developed in
734: \cite{Connelly-Demaine-Rote-2002-infinitesimally-locked} for proving a
735: configuration locked. This theory allows us to study the rigidity of
736: self-touching configurations, which is easier because vertices
737: cannot move even slightly,
738: and obtain a strong form of lockedness of non-self-touching perturbations
739: drawn with sufficiently small (but positive) separations.
740:
741: To state this relation precisely, we need some terminology from
742: \cite{Connelly-Demaine-Rote-2002-infinitesimally-locked}.
743: Call a linkage configuration \emph{rigid} if it cannot move at all.
744: Define a \emph{$\delta$-perturbation} of a linkage configuration to be
745: a repositioning of each vertex within distance $\delta$ of its original
746: position, without regard to preserving edge lengths
747: (better than $\pm 2 \delta$),
748: but consistent with the combinatorial information of which vertices are
749: on which side of which bar.
750: Call a linkage \emph{locked within~$\epsilon$} if no motion that leaves some
751: bar pinned to the plane moves any point by more than $\epsilon$.
752: Call a self-touching linkage configuration \emph{strongly locked} if,
753: for any desired $\epsilon > 0$, there is a $\delta > 0$ such that
754: all $\delta$-perturbations are locked within~$\epsilon$.
755: Thus, if a self-touching configuration is strongly locked, then
756: the smaller we draw the separations in a non-self-touching perturbation,
757: the less the configuration can move. In particular, if we choose $\epsilon$
758: small enough, the linkage must be locked in the standard sense of having
759: a disconnected configuration space locally.
760:
761: \begin{theorem}
762: {\rm \latexcite[Theorem~8.1]{Connelly-Demaine-Rote-2002-infinitesimally-locked}}
763: If a self-touching linkage configuration is rigid,
764: then it is strongly locked.
765: \label{strong lock}
766: \end{theorem}
767:
768: Therefore, if we can prove that the self-touching configuration in
769: Figure~\ref{locked 9 equilateral tight} (and its variations in
770: Figure~\ref{locked 9 isosceles tight}) are rigid, then sufficiently small
771: perturbations along the lines shown in Figure~\ref{locked 9 equilateral loose}
772: are rigid.
773:
774: The theory of \cite{Connelly-Demaine-Rote-2002-infinitesimally-locked} also
775: provides tools for proving rigidity of a self-touching configuration.
776: Specifically, we can study \emph{infinitesimal motions},
777: which just define the beginning of a motion to the first order.
778: Call a configuration \emph{infinitesimally rigid}
779: if it has no infinitesimal motions.
780:
781: \begin{lemma}
782: {\rm \latexcite[Lemma~6.1]{Connelly-Demaine-Rote-2002-infinitesimally-locked}}
783: If a self-touching linkage configuration is infinitesimally rigid,
784: then it is rigid.
785: \label{rigid rigid}
786: \end{lemma}
787:
788: \begin{figure}[htbp]
789: \centering
790: \begin{overpic}{zero_length_connection}
791: \put(10,10){\makebox(0,0)[br]{$v$}}
792: \put(10,96){\makebox(0,0)[r]{$w_1$}}
793: \put(90,46){\makebox(0,0)[t]{$w_2$}}
794: \put(23,30){\makebox(0,0)[br]{$u$}}
795: \end{overpic}
796: \caption{Two zero-length connections between vertices $u$ and~$v$.}
797: \label{zero length connection}
798: \end{figure}
799:
800: A final tool we need from
801: \cite{Connelly-Demaine-Rote-2002-infinitesimally-locked}
802: is for proving infinitesimal rigidity.
803: For each vertex $u$ wedged into a convex angle
804: between two bars $\{v,w_1\}$ and~$\{v,w_2\}$,
805: we say that there are two so-called \emph{zero-length connections} between $u$ and~$v$,
806: one perpendicular to each of the two bars $\{v,w_i\}$.
807: Such a connection between a point $u$ and a bar $\{v,w\}$ restricts $u$ to one side of the line through $v,w$,
808: %
809: \footnote{The definition of zero-length connections in
810: \cite{Connelly-Demaine-Rote-2002-infinitesimally-locked} is more
811: general, but this definition suffices for our purposes. The term
812: ``zero-length connection'' is perhaps unfortunate since it
813: mistakenly suggests a constraint on the \emph{distance} between $u$
814: and $v$ to be zero. The zero-length connection is, however, allowed
815: to slide freely on the bar $\{v,w\}$.}
816: see Figure~\ref{zero length connection}.
817: These connections must increase to the first order because $u$ must not cross
818: the two bars $\{v,w_i\}$.
819: In proving infinitesimal rigidity, we can choose to discard any zero-length
820: connections we wish, because ignoring some of the noncrossing constraints
821: only makes the configuration more flexible.
822: Together, the bars and the zero-length connections are the \emph{edges}
823: of the configuration.
824: Define a \emph{stress} to be an assignment of real numbers (\emph{stresses})
825: to edges such that, for each vertex~$v$, the vectors with directions defined by
826: the edges incident to $v$, and with magnitudes equal to the corresponding
827: stresses, sum to the zero vector.
828: We denote the stress on a bar $\{v,w\}$ by~$\omega_{v w}$,
829: and we denote the stress on a zero-length connection between vertex $u$
830: and vertex $v$ perpendicular to $\{v,w\}$ by~$\omega_{u,v w}$.
831:
832: \begin{lemma}
833: {\rm \latexcite[Lemma~7.2]{Connelly-Demaine-Rote-2002-infinitesimally-locked}}
834: If a self-touching configuration has a stress that is negative on every
835: zero-length connection, and if the configuration is infinitesimally rigid
836: when every zero-length connection is treated as a bar pinning two vertices
837: together, then the self-touching configuration is infinitesimally rigid.
838: \label{stress lemma}
839: \end{lemma}
840:
841: \subsection{Locked Chains}
842:
843: We are now in the position to state the precise senses in which the chains
844: of isosceles triangles in Figures \ref{locked 9 equilateral} and
845: \ref{locked 9 isosceles tight} are locked:
846:
847: \begin{theorem} \label{triangles rigid}
848: The self-touching chains of nine isosceles triangles shown in
849: Figures \ref{locked 9 equilateral tight} and \ref{locked 9 isosceles tight}
850: are rigid provided that the apex angle is $< 90^\circ$.
851: \end{theorem}
852:
853: Applying Theorem \ref{strong lock}, we obtain the desired result:
854:
855: \begin{corollary} \label{triangles locked}
856: The self-touching chains of nine isosceles triangles shown in
857: Figures \ref{locked 9 equilateral tight} and \ref{locked 9 isosceles tight}
858: are strongly locked provided that the apex angle is $< 90^\circ$.
859: Therefore, any sufficiently small non-self-touching perturbation,
860: similar to the one shown in Figure \ref{locked 9 equilateral loose},
861: is locked.
862: \end{corollary}
863:
864: Sections \ref{Simplifying Rules}--\ref{Stress Argument}
865: prove Theorem \ref{triangles rigid}.
866:
867: \subsection{Simplifying Rules}
868: \label{Simplifying Rules}
869:
870: We introduce two obvious rules that significantly restrict the allowable motions
871: of the self-touching configuration of isosceles triangles.
872:
873: \begin{onerule} \label{rule 1}
874: If a bar $b$ is collocated with another bar $b'$ of equal length,
875: and the bars incident to $b'$ form angles less than $90^\circ$
876: on the same side as~$b$, then any motion must keep
877: $b$ collocated with $b'$ for some positive time.
878: See Figure~\ref{rule1}.
879: \end{onerule}
880:
881: \begin{figure}
882: \centering
883: \begin{overpic}[scale=1]{rule1}
884: \put(20,31.5){\makebox(0,0)[b]{$b'$}}
885: \put(20,27.5){\makebox(0,0)[t]{$b$}}
886: \put(4,22){\makebox(0,0)[lt]{${<}90^\circ$}}
887: \put(33,21){\makebox(0,0)[rt]{$< 90^\circ$}}
888: \end{overpic}
889: \caption{Rule~\protect\ref{rule 1}
890: for simplifying self-touching configurations.}
891: \label{rule1}
892: \end{figure}
893:
894: \begin{proof}
895: The noncrossing constraints at the endpoints of $b$ and $b'$ prevent $b$
896: from moving relative to $b'$ until the angles at the endpoints of $b'$
897: open to $\geq 90^\circ$,
898: which can only happen after a positive amount of time.
899:
900: More formally, suppose without generality of loss that the endpoints
901: of $b$ and $b'$ are initially $(-1,0)$ and $(1,0)$ with $b$ below
902: $b'$ as in Figure~\ref{rule1}. Let $\alpha, \beta<90^\circ$ denote
903: the angles at the endpoints of $b'$. We attach the coordinate system
904: to $b'$ and denote denote by $(-1+x,y)$ and $(1+u,v)$ the (moving)
905: endpoints of $b$.
906: Then we have
907: \begin{equation}
908: \label{eq:a2}
909: y\le 0, \
910: v\le 0, \
911: y\ge -x\tan \alpha,\
912: v\ge u\tan \beta,
913: \end{equation}
914: and
915: \begin{equation}
916: \label{eq:a1}
917: ((x-1)-(1+u))^2+ (y-v)^2=4 .
918: \end{equation}
919: {From}~\eqref{eq:a2},
920: we conclude that $x\ge0$, $u\le 0$, and hence $x-u\ge0$,
921: and furthermore
922: \begin{equation}
923: \label{eq:a3}
924: 0\le -y-v \le d(x-u),
925: \end{equation}
926: with $d=\max\{\tan\alpha,\tan\beta\}\ge0$.
927: {From}~\eqref{eq:a1},
928: we obtain
929: \begin{align}
930: \label{a4}
931: (x-u)^2-4(x-u)& = -(y-v)^2 \ge -(y-v)^2 -4yv
932: = -(-y-v)^2 \ge -d^2(x-u)^2.
933: \end{align}
934: The last inequality is based on~\eqref{eq:a3}.
935: Now if $x-u\ne 0$, we can divide by $x-u$, knowing from~\eqref{eq:a3}
936: that $x-u> 0$, and obtain $x-u\ge 4/(1+d^2)$. Since $x$ and $u$ have to move
937: continuously from their starting values $x=u=0$, this is impossible. We conclude that $x-u=0$, and hence
938: $x=u=0$. Substituting this into~\eqref{a4}, we see that
939: $-(-y-v)^2=0$ is sandwiched between two expressions which are 0.
940: Therefore $-y-v=0$,
941: and hence
942: $y=v=0$.
943: \end{proof}
944:
945: We can apply this rule to the region shown in Figure \ref{apply rule 1},
946: resulting in a simpler linkage with the same infinitesimal behavior.
947: Although the figure shows positive separations for visual clarity,
948: we are in fact acting on the self-touching configuration
949: of Figure~\ref{locked 9 equilateral tight}.
950:
951: \begin{figure}
952: \centering
953: \begin{overpic}[scale=0.5]{apply_rule_1}
954: \put(17,58){\makebox(0,0)[b]{Rule \protect\ref{rule 1}}}
955: \end{overpic}
956: \caption{Applying Rule~\protect\ref{rule 1} to the chain of nine
957: equilateral triangles from Figure~\protect\ref{locked 9 equilateral}.}
958: \label{apply rule 1}
959: \end{figure}
960:
961: \begin{onerule} \label{rule 2}
962: If a bar $b$ is collocated with an incident bar $b'$ of the same length
963: whose other incident bar $b''$ forms a convex angle with $b'$
964: surrounding~$b$,
965: then any motion must keep $b$ collocated with $b'$ for some positive time.
966: See Figure~\ref{rule2}.
967: \end{onerule}
968:
969: \begin{figure}
970: \centering
971: \begin{overpic}[scale=1]{rule2}
972: \put(18,31.5){\makebox(0,0)[b]{$b'$}}
973: \put(18,28){\makebox(0,0)[t]{$b$}}
974: \put(33,15){\makebox(0,0)[l]{$b''$}}
975: \put(30,23){\makebox(0,0)[rt]{$< 90^\circ$}}
976: \end{overpic}
977: \caption{Rule~\protect\ref{rule 2}
978: for simplifying self-touching configurations.}
979: \label{rule2}
980: \end{figure}
981:
982: \begin{proof}
983: The noncrossing constraints at the endpoint of $b$ surrounded by
984: the convex angle formed by $b'$ and $b''$ prevent $b$ from moving relative
985: to $b'$ until the convex angle opens to $\geq 90^\circ$,
986: which can only happen after a positive amount of time.
987: The formal proof is similar (and simpler) as for Rule~\ref{rule 1}.
988: \end{proof}
989:
990: We can apply this rule twice, as shown in Figure \ref{apply rule 2},
991: to further simplify the linkage.
992:
993: \begin{figure}
994: \centering
995: \begin{overpic}[scale=0.5]{apply_rule_2}
996: \put(27,26){\makebox(0,0)[l]{Rule \protect\ref{rule 2}}}
997: \put(64,14){\makebox(0,0)[l]{Rule \protect\ref{rule 2}}}
998: \end{overpic}
999: \caption{Applying Rule~\protect\ref{rule 2} twice to the configuration
1000: from Figure~\protect\ref{apply rule 1}.}
1001: \label{apply rule 2}
1002: \end{figure}
1003:
1004: The final simplification comes from realizing that the central quadrangle gap
1005: between triangles is effectively a triangle because the right pair of edges
1006: are a rigid unit. Thus the gap forms a rigid linkage (though it is not
1007: infinitesimally rigid, because a horizontal movement of the central vertex
1008: would maintain distances
1009: to the first order), so we can treat it as part of a large rigid block.
1010: Figure~\ref{simplified} shows a simplified drawing of this self-touching
1011: configuration, which is rigid if and only if the original self-touching
1012: configuration is rigid.
1013:
1014: \begin{figure}[htbp]
1015: \centering
1016: \begin{overpic}[scale=0.5]{simplified}
1017: \put(62,38){\makebox(0,0)[l]{$A$}}
1018: \put(34,25){\makebox(0,0)[lb]{$B$}}
1019: \put(34,17){\makebox(0,0)[t]{$B'$}}
1020: \put(11,38.5){\makebox(0,0)[l]{$C$}}
1021: \put(1,37){\makebox(0,0)[r]{$C'$}}
1022: \put(37,54){\makebox(0,0)[l]{$D$}}
1023: \put(25,53.5){\makebox(0,0)[r]{$D'$}}
1024: \put(7.5,70){\makebox(0,0)[b]{$E$}}
1025: \end{overpic}
1026: \caption{The simplified configuration
1027: from Figure~\protect\ref{apply rule 2}.}
1028: \label{simplified}
1029: \end{figure}
1030:
1031: \subsection{Stress Argument}
1032: \label{Stress Argument}
1033:
1034: Finally we argue that the simplified configuration of Figure~\ref{simplified}
1035: is infinitesimally rigid using Lemma~\ref{stress lemma}.
1036: The configuration is clearly infinitesimally rigid if
1037: $B$~is pinned against~$B'$, $C$~is pinned against~$C'$, and
1038: $D$~is pinned against~$D'$.
1039: It remains to construct a stress that is negative on
1040: all length-zero connections.
1041: The stress we construct is nonzero only on the edges connecting points
1042: with labels in Figure~\ref{simplified}; we also set $\omega_{A D} = 0$.
1043:
1044: We start by assigning the stresses incident to~$A$.
1045: We choose $\omega_{A B} < 0$ arbitrarily,
1046: and set $\omega_{A B'} := -\omega_{A B} > 0$.
1047: $A$ is now in equilibrium because these stress directions are parallel.
1048:
1049: We symmetrically assign
1050: $\omega_{B C} := \omega_{A B} < 0$ and
1051: $\omega_{B' C'} = \omega_{A' B'} > 0$.
1052: The resulting forces on $B$ and $B'$ are vertical.
1053: They can be balanced by an appropriate choice of the stresses
1054: $\omega_{B, B' A}=\omega_{B, B' C'}<0$, which, taken together,
1055: also point in the vertical direction.
1056:
1057: \begin{sloppypar}
1058: Vertex $D'$ has exactly three incident stresses---$\omega_{C' D'}$,
1059: $\omega_{D', D C}$, and $\omega_{D', D E}$---which do not lie in a halfplane.
1060: Thus there is an equilibrium assignment to these stresses,
1061: unique up to scaling, and the stresses all have the same sign.
1062: Because zero-length connections must be negative,
1063: we are forced to make all three of these stresses negative.
1064: We also choose this scale factor to be substantially smaller than the
1065: stresses that have been assigned so far.
1066: \end{sloppypar}
1067:
1068: By assigning $\omega_{C D} = -\omega_{C' D'}$, we establish equilibrium at
1069: vertex $D$ as well: the forces at $D$ are the same as at $D'$, only with
1070: reversed signs.
1071:
1072: Vertex $C$ feels two stresses assigned so far---$\omega_{C D} > 0$ and
1073: $\omega_{B C} < 0$. By the choice of scale factors, the latter force
1074: dominates, leaving us with a negative force in the direction close to $CB$,
1075: and two stresses
1076: $\omega_{C, C' B'}$ and $\omega_{C, C' D'}$ which can be used to balance this
1077: force.
1078: The three directions do not lie in a halfplane.
1079: Therefore $\omega_{C, C' B'}$ and $\omega_{C, C' D'}$ can be assigned
1080: negative stresses.
1081:
1082: Finally, vertex $C'$ is also in equilibrium because
1083: $\omega_{B' C'} = -\omega_{B C}$, $\omega_{C' D'} = -\omega_{C D}$,
1084: and the stress from the zero-length connections are the same as for $C$
1085: but in the opposite direction.
1086:
1087: In summary, we have shown the existence of a stress that is positive on all
1088: zero-length connections.
1089: By Lemma~\ref{stress lemma}, the self-touching configuration is
1090: infinitesimally rigid, so by Lemma~\ref{rigid rigid}, the configuration
1091: is rigid. By the simplification arguments above,
1092: the original self-touching configuration is also rigid.
1093: By Theorem~\ref{strong lock}, the original self-touching configuration
1094: is strongly locked, so sufficiently perturbations are locked.
1095:
1096: We remark that an argument similar to the one above,
1097: using an assignment of stresses,
1098: can also be used for proving Rules~\ref{rule 1} and~\ref{rule 2},
1099: with an appropriate modification of Lemma~\ref{stress lemma};
1100: however, the direct argument that we have given is simpler.
1101:
1102: The argument relied on the isosceles triangles having an apex angle of
1103: $< 90^\circ$ (but no more) in order to guarantee that particular triples of
1104: stress directions are or are not in a halfplane. It also relies on the
1105: symmetry of the configuration through a vertical line (excluding the triangle
1106: in the upper right). Thus the argument generalizes to all isosceles triangles
1107: sharper than~$90^\circ$.
1108:
1109: \subsection{Locked Equilateral Triangles}
1110:
1111: Figure~\ref{locked 7 equilateral} shows another, simpler example of a locked
1112: chain of equilateral triangles, using just seven triangles instead of nine.
1113: However, this example cannot be stretched into a locked chain of triangles
1114: with an arbitrary apex angle of $< 90^\circ$,
1115: as in Figure \ref{locked 9 isosceles tight}.
1116:
1117: \begin{figure}
1118: \centering
1119: \subfigure[]
1120: {\includegraphics[scale=0.6]{locked_7_equilateral_loose}
1121: \label{locked 7 equilateral loose}}
1122: \hfil\hfil
1123: \subfigure[]
1124: {\includegraphics[scale=0.6]{locked_7_equilateral_tight}
1125: \label{locked 7 equilateral tight}}
1126: \caption{A locked chain of seven equilateral triangles.
1127: (a) Drawn loosely. Separations should be smaller than they appear.
1128: (b) Drawn tightly, with no separation, as a self-touching configuration.}
1129: \label{locked 7 equilateral}
1130: \end{figure}
1131:
1132: To prove that this example is locked, we first apply Rule~\ref{rule 1}
1133: and then Rule~\ref{rule 2}, as shown in Figure~\ref{apply rules}.
1134: Unlike the previous example, the resulting simplified configuration is not
1135: infinitesimally rigid (the middle vertex can move infinitesimally
1136: horizontally), so we cannot use a stress argument.
1137: In this case, however, we can use a more direct argument to prove rigidity
1138: of the simplified configuration (and thus of the original self-touching
1139: configuration).
1140:
1141: \begin{figure}
1142: \centering
1143: \begin{overpic}[scale=0.5]{apply_rules}
1144: \put(19,8){\makebox(0,0)[l]{Rule \protect\ref{rule 1}}}
1145: \put(63,16){\makebox(0,0)[l]{Rule \protect\ref{rule 2}}}
1146: \put(91,5.5){\makebox(0,0){$A$}}
1147: \put(91,33){\makebox(0,0){$B$}}
1148: \end{overpic}
1149: \caption{Applying Rules~\protect\ref{rule 1} and~\protect\ref{rule 2}
1150: to the chain of seven equilateral triangles from
1151: Figure~\protect\ref{locked 7 equilateral}.}
1152: \label{apply rules}
1153: \end{figure}
1154:
1155: Let $\ell$ denote the side length of the triangles in any of the
1156: self-touching configurations.
1157: Consider the two dashed chains connecting vertices
1158: $A$ and~$B$ in the simplified configuration.
1159: The left chain of two bars forces the distance between
1160: $A$ and $B$ to be at most $2 \ell$, with equality as in the original
1161: configuration only if the angle between the two bars remains straight.
1162: The right chain of three bars can only open its angles,
1163: because of the three triangles on the inside,
1164: so the right chain acts as a \emph{Cauchy arm}.
1165: The Cauchy-Steinitz Arm Lemma (see, e.g., \cite{Connelly-1982} or \cite{Schoenberg-Zaremba-1967})
1166: proves that the endpoints of such a chain can only get farther
1167: away from each other.
1168: Thus the distance between $A$ and $B$ is at least $2 \ell$,
1169: with equality only if the angles in the right chain do not change.
1170: These upper and lower bounds of $2 \ell$ on the distance between $A$ and $B$
1171: force the bounds to hold with equality,
1172: which prevents any angles from changing except possibly
1173: for the angles at $A$ and~$B$. However, it is impossible to change
1174: fewer than four angles of a closed chain such as the one formed by
1175: the left and right dashed chains.
1176: (This simple fact was also proved by Cauchy \cite{Cromwell-1997-Cauchy}.)
1177: Therefore, the configuration is rigid.
1178:
1179: Applying Theorem~\ref{strong lock}, we obtain that the self-touching
1180: configuration is strongly locked:
1181:
1182: \begin{theorem} \label{7 triangles rigid}
1183: The self-touching chain of seven equilateral triangles shown in
1184: Figure \ref{locked 7 equilateral tight} is rigid and thus strongly locked.
1185: Therefore, any sufficiently small non-self-touching perturbation,
1186: similar to the one shown in Figure \ref{locked 7 equilateral loose},
1187: is locked.
1188: \end{theorem}
1189:
1190:
1191:
1192: \section{Conclusion}
1193: \label{sec:conclusion}
1194:
1195: In the time since we first reported some of our results in the form of
1196: an extended abstract~\cite{conf-version}, a number of further
1197: extensions have been explored. Abbott et
1198: al.~\cite{Abbott-Abel-Charlton-Demaine-Demaine-Kominers-2008} have
1199: utilized our results in proving that hinged dissections exist.
1200: Abbott, Demaine and Gassend~\cite{adg-gcrts-09} provided a
1201: generalization for the restricted case of open chains of strictly
1202: slender adornments: even when self-touching configurations are
1203: allowed, every chain can be opened.
1204:
1205: A variety of open questions and extensions remain to be studied. It
1206: may be interesting to consider the algorithmic question of computing,
1207: for a given configuration of an adorned chain (whose adornments are
1208: not slender), whether or not there exists a motion that opens it.
1209:
1210: Our results have application to hinged dissections of polyregulars,
1211: e.g., polyominoes; this includes the (slender) case of squares
1212: connected at opposite corners. An interesting question arising in
1213: this context is whether every hinged dissection can be subdivided so
1214: that the pieces are slender.
1215:
1216: We note that there are examples of open and closed chains of
1217: self-touching squares (some of which are attached at adjacent corners)
1218: that we conjecture to be locked (Figure~\ref{fig:locked-squares});
1219: while we have been unable so far to prove that they are, the
1220: methods we employed for the locked chains of sharp triangles
1221: (Section~\ref{locked sharp}) may be applicable.
1222:
1223: \begin{figure} [here]
1224: \begin{center}
1225: \subfigure[]
1226: {\includegraphics[scale=0.8]{open_locked_squares}
1227: \label{open-locked}}
1228: \hfil\hfil
1229: \subfigure[]
1230: {\includegraphics[scale=0.8]{closed_locked_squares}
1231: \label{closed-locked}}
1232: \caption{Chains of squares that appear to be locked: (a) an open chain, and (b) a closed chain. The squares are
1233: self-touching; the drawing includes gaps only for clarity of illustration.}
1234: \label{fig:locked-squares}
1235: \end{center}
1236: \end{figure}
1237:
1238: One difficulty of exploring the space of adorned
1239: chains is highlighted by the following construction.
1240: It consists of a closed chain, a convex parallelogram,
1241: with slender adornments attached, where each adornment together with
1242: its base is convex, such that the configuration space has infinitely
1243: many components.
1244: %
1245: We attach a single obtuse triangle as a slender adornment to the top
1246: base segment, as with Figure \ref{fig:2-components}. If the bottom
1247: segment is fixed the path of the bottom vertex in the upper adornment
1248: traces out a circle, which is shown as a dashed circular arc $C$ in
1249: Figure \ref{fig:inf-comp}(a).
1250: %
1251: The second slender adornment is attached to the bottom segment and is
1252: the convex hull of infinitely many points, each slightly above $C$.
1253: The points form an infinite sequence $p_1, p_2, \dots$ converging to a
1254: point on the right $p_\infty$, and they are chosen so the straight
1255: line interval from $p_i$ to $p_{i+1}$ intersects the lower portion of
1256: $C$ (the open circular disk determined by $C$). An exaggerated
1257: picture of this construction is in Figure \ref{fig:inf-comp}(b).
1258: Thus, the upper slender adornment intersects the lower adornment and
1259: misses it alternately infinitely often.
1260:
1261: \begin{figure} [here]
1262: \begin{center}
1263: \includegraphics{inf-comp}
1264: \caption{Figure (a) shows the overall set-up of a parallelogram with two convex slender adornments attached such that the configuration space has infinitely many components. Figure (b) is an exaggerated close-up of where the two adornments are close.}
1265: \label{fig:inf-comp}
1266: \end{center}
1267: \end{figure}
1268:
1269:
1270:
1271: % \bibliography{dissect,hinge,linkage,polytopes,locked}
1272: \bibliographystyle{alpha}
1273:
1274: \newcommand{\etalchar}[1]{$^{#1}$}
1275: \begin{thebibliography}{CvdSG{\etalchar{+}}07}
1276:
1277: \bibitem[AAC{\etalchar{+}}08]{Abbott-Abel-Charlton-Demaine-Demaine-Kominers-20%
1278: 08}
1279: Timothy~G. Abbott, Zachary Abel, David Charlton, Erik~D. Demaine, Martin~L.
1280: Demaine, and Scott~D. Kominers.
1281: \newblock Hinged dissections exist.
1282: \newblock In {\em Proceedings of the 24th Annual ACM Symposium on Computational
1283: Geometry (SoCG 2008)}, pages 110--119, College Park, Maryland, June 9--11
1284: 2008.
1285:
1286: \bibitem[ADG09]{adg-gcrts-09}
1287: Timothy~G. Abbott, Erik~D. Demaine, and Blaise Gassend.
1288: \newblock A generalized {Carpenter}'s {Rule} {Theorem} for self-touching
1289: linkages, January 2009.
1290: \newblock arXiv:\href{http://arXiv.org/abs/0901.1322}{0901.1322}.
1291:
1292: \bibitem[AKRW04]{Alt-Knauer-Rote-Whitesides-2004}
1293: Helmut Alt, Christian Knauer, G{\"u}nter Rote, and Sue Whitesides.
1294: \newblock On the complexity of the linkage reconfiguration problem.
1295: \newblock In J.~Pach, editor, {\em Towards a Theory of Geometric Graphs},
1296: volume 342 of {\em Contemporary Mathematics}, pages 1--14. American
1297: Mathematical Society, 2004.
1298:
1299: \bibitem[Ale84]{Alexander-1984}
1300: Ralph Alexander.
1301: \newblock The circumdisk and its relation to a theorem of {Kirszbraun} and
1302: {Valentine}.
1303: \newblock {\em Math. Mag.}, 57(3):165--169, 1984.
1304:
1305: \bibitem[AN98]{Akiyama-Nakamura-1998}
1306: Jin Akiyama and Gisaku Nakamura.
1307: \newblock Dudeney dissection of polygons.
1308: \newblock In {\em Revised Papers from the Japan Conference on Discrete and
1309: Computational Geometry}, volume 1763 of {\em Lecture Notes in Computer
1310: Science}, pages 14--29, Tokyo, Japan, December 1998.
1311:
1312: \bibitem[BC02]{Bezdek-Connelly-2002}
1313: K\'aroly Bezdek and Robert Connelly.
1314: \newblock Pushing disks apart---the {Kneser-Poulsen} conjecture in the plane.
1315: \newblock {\em J. reine angew. Math.}, 553:221--236, 2002.
1316:
1317: \bibitem[CDD{\etalchar{+}}06]{conf-version}
1318: Robert Connelly, Erik~D. Demaine, Martin~L. Demaine, S\'andor~P. Fekete, Stefan
1319: Langerman, Joseph S.~B. Mitchell, Ares Rib\'o, and G{\"u}nter Rote.
1320: \newblock Locked and unlocked chains of planar shapes.
1321: \newblock In {\em Proceedings of the 22nd Annual Symposium on Computational
1322: Geometry, Sedona}, pages 61--70. Association for Computing Machinery, 2006.
1323: \newblock arXiv:\href{http://arXiv.org/abs/cs/0604022}{cs/0604022}.
1324:
1325: \bibitem[CDR02]{Connelly-Demaine-Rote-2002-infinitesimally-locked}
1326: Robert Connelly, Erik~D. Demaine, and G\"unter Rote.
1327: \newblock Infinitesimally locked self-touching linkages with applications to
1328: locked trees.
1329: \newblock In J.~Calvo, K.~Millett, and E.~Rawdon, editors, {\em Physical Knots:
1330: Knotting, Linking, and Folding of Geometric Objects in 3-space}, pages
1331: 287--311. American Mathematical Society, 2002.
1332:
1333: \bibitem[CDR03]{Connelly-Demaine-Rote-2003}
1334: Robert Connelly, Erik~D. Demaine, and G\"unter Rote.
1335: \newblock Straightening polygonal arcs and convexifying polygonal cycles.
1336: \newblock {\em Discrete \& Computational Geometry}, 30(2):205--239, September
1337: 2003.
1338:
1339: \bibitem[Con82]{Connelly-1982}
1340: Robert Connelly.
1341: \newblock Rigidity and energy.
1342: \newblock {\em Invent. Math.}, 66(1):11--33, 1982.
1343:
1344: \bibitem[Con08]{Bobs-survey}
1345: Robert Connelly.
1346: \newblock Expansive motions.
1347: \newblock In Jacob~E. Goodman, J\'anos Pach, and Richard Pollack, editors, {\em
1348: Surveys on Discrete and Computational Geometry---Twenty Years Later}, volume
1349: 453 of {\em Contemporary Mathematics}, pages 213--229. American Mathematical
1350: Society, 2008.
1351:
1352: \bibitem[Cro97]{Cromwell-1997-Cauchy}
1353: Peter~R. Cromwell.
1354: \newblock Equality, rigidity, and flexibility.
1355: \newblock In {\em Polyhedra}, chapter~6, pages 219--247. Cambridge University
1356: Press, 1997.
1357:
1358: \bibitem[Csi01]{Csikos-2001}
1359: Bal\'azs Csik\'os.
1360: \newblock On the volume of flowers in space forms.
1361: \newblock {\em Geom. Dedicata}, 86(1-3):59--79, 2001.
1362:
1363: \bibitem[CvdSG{\etalchar{+}}07]{csgor-ihp-06}
1364: Jae-Sook Cheong, A.~Frank van~der Stappen, Ken Goldberg, Mark~H. Overmars, and
1365: Elon Rimon.
1366: \newblock Immobilizing hinged polygons.
1367: \newblock {\em International Journal on Computational Geometry and
1368: Applications}, 17(1):45--70, 2007.
1369:
1370: \bibitem[DDE{\etalchar{+}}05]{Demaine-Demaine-Eppstein-Frederickson-Friedman-2%
1371: 005}
1372: Erik~D. Demaine, Martin~L. Demaine, David Eppstein, Greg~N. Frederickson, and
1373: Erich Friedman.
1374: \newblock Hinged dissection of polyominoes and polyforms.
1375: \newblock {\em Computational Geometry: Theory and Applications},
1376: 31(3):237--262, June 2005.
1377:
1378: \bibitem[DDLS05]{Demaine-Demaine-Lindy-Souvaine-2005}
1379: Erik~D. Demaine, Martin~L. Demaine, Jeffrey~F. Lindy, and Diane~L. Souvaine.
1380: \newblock Hinged dissection of polypolyhedra.
1381: \newblock In {\em Proceedings of the 9th Workshop on Algorithms and Data
1382: Structures}, volume 3608 of {\em Lecture Notes in Computer Science}, pages
1383: 205--217, Waterloo, Canada, August 2005.
1384:
1385: \bibitem[Dud02]{Dudeney-1902-hinged}
1386: Henry~E. Dudeney.
1387: \newblock Puzzles and prizes.
1388: \newblock {\em Weekly Dispatch}, 1902.
1389: \newblock The puzzle appeared in the April 6 issue of this column. An unusual
1390: discussion followed on April 20, and the solution appeared on May 4.
1391:
1392: \bibitem[Epp01]{Eppstein-2001-mirror-dissection}
1393: David Eppstein.
1394: \newblock Hinged kite mirror dissection, June 2001.
1395: \newblock arXiv:\href{http://arXiv.org/abs/cs.CG/0106032}{cs.CG/0106032}.
1396:
1397: \bibitem[Fre02]{Frederickson-2002}
1398: Greg~N. Frederickson.
1399: \newblock {\em Hinged Dissections: Swinging \& Twisting}.
1400: \newblock Cambridge University Press, August 2002.
1401:
1402: \bibitem[GM95]{Gordon-Meyer-1995}
1403: Y.~Gordon and M.~Meyer.
1404: \newblock On the volume of unions and intersections of balls in {Euclidean}
1405: space.
1406: \newblock In Joram Lindenstrauss and Vitali Milman, editors, {\em Geometric
1407: Aspects of Functional Analysis. Israel Seminar (GAFA) 1992--94}, volume~77 of
1408: {\em Operator Theory: Advances and Applications}, pages 91--101.
1409: Birkh\"auser, 1995.
1410:
1411: \bibitem[Kir34]{Kirszbraun-1934}
1412: M.~D. Kirszbraun.
1413: \newblock {\"Uber} die zusammenziehende und {Lipschitzsche} {Transformationen}.
1414: \newblock {\em Fundamenta Mathematicae}, 22:77--108, 1934.
1415:
1416: \bibitem[MTW{\etalchar{+}}02]{Mao-Thallidi-Wolfe-Whitesides-Whitesides-2002}
1417: Chengde Mao, Venkat~R. Thallidi, Daniel~B. Wolfe, Sue Whitesides, and George~M.
1418: Whitesides.
1419: \newblock Dissections: Self-assembled aggregates that spontaneously reconfigure
1420: their structures when their environment changes.
1421: \newblock {\em Journal of the American Chemical Society}, 124:14508--14509,
1422: 2002.
1423:
1424: \bibitem[Rei79]{r-cmpg-79}
1425: John~H. Reif.
1426: \newblock Complexity of the mover's problem and generalizations.
1427: \newblock In {\em Proceedings of the 20th Annual IEEE Symposium on Foundations
1428: of Computer Science}, pages 421--427, 1979.
1429:
1430: \bibitem[Str00]{Streinu-2000}
1431: Ileana Streinu.
1432: \newblock A combinatorial approach to planar non-colliding robot arm motion
1433: planning.
1434: \newblock In {\em Proceedings of the 41st Annual Symposium on Foundations of
1435: Computer Science}, pages 443--453, Redondo Beach, California, November 2000.
1436:
1437: \bibitem[Str05]{Streinu-2005}
1438: Ileana Streinu.
1439: \newblock Pseudo-triangulations, rigidity and motion planning.
1440: \newblock {\em Discrete \& Computational Geometry}, 34(4):587--635, November
1441: 2005.
1442:
1443: \bibitem[SZ67]{Schoenberg-Zaremba-1967}
1444: I.~J. Schoenberg and S.~K. Zaremba.
1445: \newblock On {Cauchy}'s lemma concerning convex polygons.
1446: \newblock {\em Canadian Journal of Mathematics}, 19(4):1062--1071, 1967.
1447:
1448: \end{thebibliography}
1449:
1450:
1451: \end{document}
1452: