1: \documentclass{article}
2: \usepackage{graphicx}
3: \usepackage{url}
4:
5: \input{mac} %for SIGACT & CoRR
6: %\usepackage{latexsym} %for CoRR to get Box
7: \usepackage{amssymb}
8:
9: \setlength{\oddsidemargin}{0in}
10: \setlength{\topmargin}{-.5in}
11: \setlength{\textwidth}{6.6in}
12: \setlength{\textheight}{9in}
13:
14: \title{Computational Geometry Column 39}
15: \author{%
16: Joseph O'Rourke\thanks{
17: Dept. of Computer Science, Smith Col\-lege, North\-ampton,
18: MA 01063, USA.
19: \-orourke@cs.\-smith\-.edu.
20: Supported by NSF Grant CCR-9731804.
21: }
22: }
23: \date{}
24: \begin{document}
25: \maketitle
26: \pagestyle{empty}
27: \thispagestyle{empty}
28:
29: \begin{abstract}
30: The resolution of a decades-old open problem
31: is described: polygonal chains cannot lock in the plane.
32: \end{abstract}
33:
34: A {\em polygonal chain\/} is a connected series of line segments.
35: Chains may be open, or closed to form a polygon.
36: A {\em simple\/} chain is one that does not self-intersect:
37: only segments adjacent in the chain intersect, and then
38: only at their shared endpoint.
39: If the segments of a polygonal chain are viewed as
40: rigid bars, and the vertices as universal joints, natural
41: questions are whether every open chain can be
42: {\em straightened\/}---reconfigured to lie on a straight line---and
43: whether every closed chain can be {\em convexified\/}---reconfigured
44: to form a planar convex polygon. In both cases, the chains are
45: to remain simple throughout the motion.
46: If a chain cannot be so reconfigured, it is called {\em locked}.
47:
48: These questions were raised
49: by several researchers independently since the 1970's,%
50: \footnote{
51: G.~Bergman,
52: U.~Grenander,
53: W.~Lenhart,
54: J.~Mitchell,
55: S.~Schanuel,
56: and
57: S.~Whitesides.
58: It seems to have first appeared in print in 1995:
59: \cite{lw-rcpce-95} and~\cite[p.~270]{k-pldt-95}.
60: }
61: and were the subject of intense investigation by the late 1990's.
62: It was first established that chains can lock in three dimensions (3D):
63: both locked open chains, and locked closed but unknotted chains
64: are possible~\cite{cj-nepiu-98,bddlloorstw-lupc3d-99}.
65: In 4D, neither open nor closed chains can
66: lock~\cite{co-pccl4d-99}.
67: But aside from some special cases
68: (e.g., star-shaped polygons cannot lock~\cite{elrss-cssp-98},
69: polygonal trees can lock~\cite{bddloorsw-orfltcl-98}),
70: the problems remained unresolved for chains in 2D.
71:
72: Connelly, Demaine, and Rote have now settled the questions,
73: establishing that neither open nor closed chains can lock
74: in 2D~\cite{cdr-epcbu-00}.
75: Their result is even more general: no collection
76: of disjoint simple chains are locked
77: (although of course a chain nested inside a polygon is forever
78: confined).
79: They prove that every such collection has an
80: {\em expansive motion\/}: one during which the distance
81: between every pair of vertices increases or stays the same.
82: An expansive motion automatically maintains simplicity,
83: for the distance between some pair of vertices would have to decrease
84: to reach self-intersection.
85: Their proof first establishes that any configuration of
86: chains has an infinitesimal expansive motion,
87: and then combines these motions into a global expansive motion.
88: Although there are different ways to stitch the infinitesimal
89: motions together, one produces especially
90: natural movements.
91: This motion
92: minimizes the squared lengths
93: of the velocity vectors $v_i$ applied to each vertex $i$ at each time:
94: minimize $\sum_i \| v_i \|^2$, subject to constraints maintaining
95: segment lengths and forcing expansivity.\footnote{
96: Their proof employs a further nonquadratic
97: ``barrier'' penalty term
98: for smoothness.
99: }
100: The objective function is strictly convex and the system
101: can be solved by quadratic programming.
102: Figs.~\figref{chains.1} and~\figref{chains.2}
103: illustrate the solution so obtained for
104: a collection of three closed chains (triangles) entangled with one open
105: chain.
106: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin
107: \begin{figure}[htbp]
108: \centering
109: \includegraphics[width=0.2\linewidth]{orourke-z2-00000.ps}%
110: \hspace{2mm}%
111: \includegraphics[width=0.2\linewidth]{orourke-z2-00010.ps}%
112: \hspace{2mm}%
113: \includegraphics[width=0.2\linewidth]{orourke-z2-00020.ps}%
114: \hspace{2mm}%
115: \includegraphics[width=0.2\linewidth]{orourke-z2-00040.ps}
116: %\fbox{chains.eps}
117: \vspace{-2mm}
118: \caption{Four frames in the unlocking of four polygonal chains.
119: The initial disjoint collection is shown in the leftmost frame.
120: [Animation courtesy of Erik Demaine.]}
121: \figlab{chains.1}
122: %\end{figure}
123: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End
124: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure Begin
125: %\begin{figure}[htbp]
126: \centering
127: \includegraphics[width=0.25\linewidth]{orourke-z2e-00240.ps}%
128: \hspace{1mm}%
129: \includegraphics[width=0.25\linewidth]{orourke-z2e-00420.ps}%
130: \hspace{5mm}%
131: \includegraphics[width=0.25\linewidth]{orourke-z2e-00830.ps}%
132: %\fbox{chains.eps}
133: \vspace{-8mm}
134: \caption{Three frames toward the end of the
135: unlocking motion, continuing Fig.~\figref{chains.1}
136: (at a different scale).}
137: \figlab{chains.2}
138: \end{figure}
139: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Figure End
140:
141: Let us simplify the discussion to a single open chain
142: and offer a crude sketch of their key argument.
143: View nonadjacent vertices of the chain as connected by
144: a {\em strut}, which is permitted to increase in length
145: or stay the same, but never decrease.
146: If this bar-and-strut framework has an infinitesimal
147: motion, then the motion is necessarily expansive.
148: An {\em equilibrium stress\/} in an assignment of
149: weights ({\em stresses\/}) to each bar and strut so that (a)~the stress on
150: each strut is nonnegative, and (b)~the stress vectors
151: (vectors along the bars/struts scaled by the stresses) are
152: in equilibrium at each vertex.
153: Bar stresses may be positive (compression)
154: or negative (tension).
155: If a framework
156: is ``stuck,'' or, more formally,
157: ``first-order rigid''~\cite{cw-sorps-96,w-rsa-97},
158: then it has a nonzero equilibrium stress.
159: So if a framework only has the trivial zero
160: equilibrium stress, then it possesses an
161: infinitesimal expansive motion.
162:
163: In order to apply a theorem that holds for planar frameworks,
164: every intersection point of the framework is used to
165: divide the bars/struts, which produces a planar
166: framework
167: %after merging any multiple edges.
168: %This planar framework is
169: equivalent to the original in
170: terms of equilibrium stresses.
171: Now the century-old Maxwell-Cremona theorem is applied:
172: if the framework has a nonzero equilibrium stress,
173: then it can be ``lifted'' to a nonflat polyhedral terrain
174: that projects to the framework, with positive-stress edges
175: (always bars)
176: lifting to ``mountains,'' and negative-stress edges to ``valleys.''
177: Finally it is shown that such a lifting is impossible,
178: by concentrating on its maximum.
179: For example, if this maximum is a single point,
180: then it must lie at the junction of at least three
181: mountain-edges, which violates the fact that
182: every vertex of the chain is incident to at most two bars.
183: A case analysis on the topologically possible maxima
184: proves that only the degenerate flat lifting
185: is possible, which implies that the original framework
186: has only the zero
187: equilibrium stress, which implies that it has an
188: infinitesimal expansive motion,
189: which implies that it has a global expansive motion,
190: i.e., it is not locked.
191:
192: Building on this work, Streinu has found a way to
193: decompose the expansive global motion for chains of $n$ vertices
194: into $O(n^2)$ sections, each of which is the motion of
195: a one degree-of-freedom mechanism~\cite{s-capnc-00}.
196: This mechanism is constructed by removing one hull edge
197: from a {\em pseudotriangulation\/} of the chains,
198: a partition by diagonal bars into regions each bounded by
199: three reflex chains (where a single segment counts as a reflex chain).
200: The mechanism is opened following its only free trajectory
201: in configuration space
202: until two adjacent edges align,
203: at which point the pseudotrianguation is revised locally, and
204: the expansion process continued.
205:
206: \bibliographystyle{alpha}
207: \bibliography{39}
208: \end{document}
209:
210: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
211: