(Michael Farber) Invitation To Topological Robotic
(Michael Farber) Invitation To Topological Robotic
(Michael Farber) Invitation To Topological Robotic
Edited by
Erwin Bolthausen (Managing Editor), Freddy Delbaen, Thomas Kappeler (Managing Editor), Christoph Schwab,
Michael Struwe, Gisbert Wstholz
Mathematics in Zurich has a long and distinguished tradition, in which the writing of lecture notes volumes
and research monographs play a prominent part. The Zurich Lectures in Advanced Mathematics series
aims to make some of these publications better known to a wider audience. The series has three main constituents: lecture notes on advanced topics given by internationally renowned experts, graduate text books
designed for the joint graduate program in Mathematics of the ETH and the University of Zurich, as well
as contributions from researchers in residence at the mathematics research institute, FIM-ETH. Moderately
priced, concise and lively in style, the volumes of this series will appeal to researchers and students alike,
who seek an informed introduction to important areas of current research.
Previously published in this series:
Yakov B. Pesin, Lectures on partial hyperbolicity and stable ergodicity
Sun-Yung Alice Chang, Non-linear Elliptic Equations in Conformal Geometry
Sergei B. Kuksin, Randomly forced nonlinear PDEs and statistical hydrodynamics in 2 space dimensions
Pavel Etingof, Calogero-Moser systems and representation theory
Guus Balkema and Paul Embrechts, High Risk Scenarios and Extremes A geometric approach
Demetrios Christodoulou, Mathematical Problems of General Relativity I
Camillo De Lellis, Rectifiable Sets, Densities and Tangent Measures
Published with the support of the Huber-Kudlich-Stiftung, Zrich
Michael Farber
Invitation to
Topological Robotics
Author:
Prof. Michael Farber
Department of Mathematical Sciences
Durham University
DH1 3LE Durham
United Kingdom
E-mail: [email protected]
ISBN 978-3-03719-054-8
The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography,
and the detailed bibliographic data are available on the Internet at http://www.helveticat.ch.
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in other ways, and storage in data banks. For any kind of use permission
of the copyright owner must be obtained.
To Madeleine
Preface
Topological robotics is a new mathematical discipline studying topological problems inspired by robotics and engineering as well as problems of practical robotics requiring topological tools. It is a part of a
broader newly created research area called computational topology.
The latter studies topological problems appearing in computer science
and algorithmic problems in topology.
This book is based on a one-semester lecture course Topics of
Topological Robotics which I gave at the ETH Z
urich during April
June 2006. I describe here four selected mathematical stories which
have interesting connections to other sciences.
Chapter 1 studies conguration spaces of mechanical linkages, a
remarkable class of manifolds which appear in several elds of mathematics as well as in molecular biology and in statistical shape theory.
Methods of Morse theory, enriched with new techniques based on properties of involutions, allow eective computation of their Betti numbers.
We describe here a recent solution of the conjecture raised by Kevin
Walker in 1985. This conjecture asserts that the relative sizes of bars of
a linkage are determined, up to certain equivalence, by the cohomology
algebra of the linkage conguration space.
In Chapter 1 we also discuss topology of random linkages, a probabilistic approach to topological spaces depending on a large number of
random parameters.
In Chapter 2 we describe a beautiful theorem of Swiatoslaw R. Gal
[38] which gives a general formula for Euler characteristics of conguration spaces F (X, n) of of n distinct particles moving in a polyhedron X,
for all n. The Euler Gal power series is a rational function encoding
all numbers (F (X, n)) and Gals theorem gives an explicit expression
for it in terms of local topological properties of the space.
Chapter 3 deals with the knot theory of the robot arm, a variation of
the traditional knot theory question motivated by robotics. The main
result (which in my view is one of the remarkable jewels of modern
mathematics) is an unknotting theorem for planar robot arms, proven
recently by R. Connelly, E.Demaine and G. Rote [10].
viii
PREFACE
Chapter 4 discusses the notion of topological complexity of the robot motion planning problem TC(X) and mentions several new results
and techniques. The number TC(X) measures the complexity of the
problem of navigation in a topological space X viewed as the conguration space of a system. In this chapter we explain how one may use
stable cohomology operations to improve lower bounds on the topological complexity based on products of zero-divisors. These results were
obtained jointly with Mark Grant.
I would like to thank Jean-Claude Hausmann, Thomas Kappeler
and Dirk Schuetz for useful discussions of various parts of the book.
My warmest thanks go also to Mark Grant who helped me in many ways
to make this text readable and in particular for his advice concerning
the material of Chapter 4.
Problems of topological robotics can roughly be split into two main
categories: (A) studying special topological spaces, conguration spaces
of important mechanical systems; (B) studying new topological invariants of general topological spaces, invariants which are motivated and
inspired by applications in robotics and engineering. Class (A) includes
describing the topology of varieties of linkages, conguration spaces of
graphs, knot theory of the robot arm topics which are partly covered
below. The story about TC(X) represents a theory of class (B).
The book is intended as an appetizer and will introduce the reader
to many fascinating topological problems motivated by engineering.
Michael Farber
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Chapter
1.1.
1.2.
1.3.
1.4.
1.5.
1.6.
1.7.
1.8.
1.9.
1.10.
1.11.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
2
5
6
9
11
16
19
24
30
31
36
Chapter
2.1.
2.2.
2.3.
2.4.
2.5.
2.6.
2.7.
2.8.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
41
41
45
46
49
54
55
57
59
Chapter
3.1.
3.2.
3.3.
3.4.
3.5.
3.6.
3.7.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
61
61
64
67
68
72
75
77
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Chapter
4.1.
4.2.
4.3.
4.4.
4.5.
4.6.
4.7.
4.8.
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
87
87
89
99
102
106
114
117
121
CHAPTER 1
(following [28]) how one may compute Betti numbers of planar polygon
spaces. We also discuss classication results for these spaces in terms
of combinatorics of chambers and strata of their length vectors (the
Walker conjecture); here we briey describe results of [29]. Finally
we present a probabilistic approach to polygon spaces, which is very
eective in situations when one does not know the bar lengths and the
number of links n is large, n ; these results are described in [31]
in more detail.
The style of the exposition is not uniform and varies while we advance into the chapter. In the beginning it is very elementary, with
many simple examples, pictures and explanations. Sections 1.51.8
are written in the style of a research article containing theorems and
full proofs. In several concluding sections we adopt the survey style,
simply stating major results and referring to original articles for proofs
and further details.
A special role is played by section 1.7 on Morse theory of manifolds
with involution which can be read independently of the other material
of the chapter and may be applied in various contexts.
1.1. Conguration space of a linkage
Consider a simple planar mechanism consisting of n bars of xed lengths
l1 , . . . , ln connected by revolving joints forming a closed polygonal chain,
see Figure 1.1. The positions of two adjacent vertices are xed but the
other vertices are free to move so that angles between the bars change
but the lengths of the bars remain xed and the links are not disconnected from each other. Our task is to understand the topology of
l2
l1
l4
l3
l5
Z = (l1 , l2 , . . . , ln ) Rn+ ,
l1 > 0, . . . , ln > 0,
is the collection of the bar lengths, called the length vector of the linkage. Clearly, a conguration of the linkage is fully determined by angles
the bars make with the horizontal direction. Hence, the conguration
space MW can be identied with
1
(1.2) MW = {(u1 , . . . , un ) S S ;
n
=
li ui = 0, un = e1 }.
i=1
Here ui is +
the unit vector in the direction of the bar number i and the
condition ni=1 li ui = 0 expresses the property of the polygonal chain
to be closed. The equation un = e1 means that the last bar always
points in the direction opposite to the x-axis.
The topological space MW can also be understood as the moduli
space of planar n-gons with sides of length l1 , . . . , ln , viewed up to the
action of orientation-preserving isometries of the plane. It is important
to emphasize that our n-gons have cyclically oriented sides which are
labeled by integers 1, 2, . . . , n as shown on Figure 1.2. For any such planar n-gon there is a unique rotation of the plane bringing the polygon
in the position with a given side pointing in a xed direction. Hence
there is a one-to-one correspondence between the conguration space
of the mechanism shown on Figure 1.1 and the variety of all dierent
polygonal shapes of planar n-gons with sides of lengths l1 , . . . , ln . This
l5
l2
l3
l4
l1
(1.3) MW = {(u1 , . . . , un ) S S ;
n
=
li ui = 0}/SO(2).
i=1
Spaces MW appear also in the statistical shape theory, see [63], dealing with shapes of nite clouds of points viewed up to orientationpreserving Euclidean isometries. A nite set z1 , . . . , zn R2 has the
center c = (z1 + + zn )/n and is partly characterized by the distances
li = |zi c|. Let us assume that all li > 0, i.e., zi == c. Then
ui = (zi c)/li S 1 , i = 1, . . . , n
+
is a collection of unit vectors satisfying ni=1 li ui = 0. If z1( , . . . , zn( R2
is obtained from z1 , . . . , zn R2 by applying an orientation-preserving
planar isometry, then the corresponding set u(1 , . . . , u(n S 1 is obtained
from u1 , . . . , un S 1 by a global orientation-preserving rotation. This
explains why MW parameterizes all planar shapes with given distances
li from the central point.
Any permutation : {1, . . . , n} {1, . . . , n} denes a dieomorphism
: S 1 S 1 S 1 S 1 ,
given by
(u1 , . . . , un ) = (u(1) , . . . , u(n) ).
Clearly, maps the polygon space MW S 1 S 1 dieomorphically
onto the polygon space MW! S 1 S 1 where Z( = (l(1) , . . . , l(n) ).
Hence, we conclude that the order in which the numbers l1 , l2 , . . . , ln
appear in the length vector (1.1) is irrelevant for the dieomorphism
type of MW .
The case n = 3 is trivial. By the remark
above without loss of generality we may assume that l1 l2 l3 . If the triangle inequalP3
P2
ity l1 < l2 + l3 is satised, then MW consists of
two points two triangles with sides l1 , l2 , l3
P1
which are congruent to each other via a reection of the plane. If l1 = l2 + l3 , then MW
consists of a single point the degenerate triangle.
In the remaining case l1 > l2 + l3 the
P2
P3
variety MW is empty, MW = .
The following lemma describes all cases
when the variety MW is empty, n 3.
Lemma 1.1. For any n 3, the variety MW is empty if and only if one
of the links li is longer than the sum of all other links, i.e.,
(1.4)
i = 1, . . . , n
l2
l1
C=T
:CW
associates with each conguration of the arm the corresponding position of the gripper, see Figure 1.4. Analytically is given by
(1.7)
(u1 , u2 ) = l1 u1 + l2 u2 ,
u1 , u2 S 1 .
Lemma 1.2. (a) The preimage 1 (w) of any internal point w of the
annulus W consists of exactly two congurations which are symmetric
to each other with respect to the line passing through the origin and w.
(b) If |w| = l1 + l2 , the preimage 1 (w) contains a single conguration with u1 = u2 .
(c) If |w| = l1 l2 > 0, the preimage 1 (w) contains a single
conguration with u1 = u2 .
(d) In the case l1 = l2 , the preimage 1 (0) equals the anti-diagonal
= {(u, u); u S 1 }.
(e) Let T 2 denote the diagonal = {(u, u); u S 1 }. The complement T 2 ( ) is a union of two connected components (which
we denote C+ and C ) and maps each of them dieomorphically onto
the domain l1 l2 < |w| < l1 + l2 .
Proof. To prove (e) we note that T 2 ( ) is the set of linearly
independent pairs of unit vectors u1 , u2 S 1 and C+ is dened as the
set of pairs u1 , u2 with det(u1 , u2 ) > 0 (the positive frames). The other
set C is dened as the set of negative frames.
Remaining statements of the lemma are obvious.
!
1.3. Varieties of quadrangles: n = 4
Next we use the method of J. Milgram and J. Trinkle [75] to understand
the variety MW in the case n = 4. This elementary discussion will give
l1 l2 l3 l4 .
A
(5,2,1,1)
B
(4,2,1,1)
C
(3,2,1,1)
D
(3,2,2,1)
E
(4,3,3,1)
single point,
S 1,
S 1 S 1,
S1 5 S1
F
(3,3,2,1)
G
(2,2,1,1)
H
(1,1,1,1)
Case G,
Case H,
n
=
li ii == 0 for ii = 1.
i=1
10
iJ
/
A subset of a short set is short. As we shall see later, the set of all
short subsets with respect to Z fully determines the dieomorphism
type of MW . Hence one expects to be able to express all topological
characteristics of MW in terms of the family of all short subsets with
respect to Z.
A subset J {1, . . . , n} is called long with respect to Z if
=
=
li >
(1.11)
li ,
iJ
iJ
/
iJ
/
B
J " {2, 3, 4}
single point
C
J = {1} or J " {2, 3, 4}
S1
D
|J| 1 or J = {2, 4} or J = {3, 4}
S1 S1
E & F |J| 2 and J == {1, 2}, {1, 3}, {2, 3}
S1 5 S1
G
|J| 1 or J = {3, 4}
Figure 1.7 left
H
|J| 1
Figure 1.7 right
To explain our notations, let us mention that in Case B the short
subsets are all subsets of {2, 3, 4} except J = {2, 3, 4}.
11
l3
l2
l4
l1
W = {(u1 , . . . , un ) S 1 S 1 }/SO(2).
Clearly, W is dieomorphic to torus T n1 of dimension n 1. A dieomorphism can be specied, for example, by assigning to a conguration
1
1
n1
(u1 , . . . , un ) the point (1, u2 u1
(measur1 , u3 u1 , . . . , un1 u1 ) T
ing angles between the directions of the rst and the other links).
The moduli space of polygons MW (where Z = (l1 , . . . , ln )) is naturally embedded into W .
MP
0
fP
12
fW : W R,
$2
$
n
$
$=
$
$
l i ui $ .
fW (u1 , . . . , un ) = $
$
$
i=1
n
=
li cos i )2 (
i=1
n
=
i=1
li2 2
=
i<j
n
=
li sin i )2
i=1
li lj cos(i j ).
13
We nd
=
fW
lj sin(j i ).
= 2li
i
j=1
n
(1.15)
The equation
(1.16)
f,
i
= 0 gives
sin i
n
=
lj cos j = cos i
n
=
lj sin j .
j=1
j=1
lj cos j = 0 =
j=1
n
=
lj sin j ,
j=1
LJ =
n
=
li ui > 0.
i=1
li k1=i lk cos(i k ), if i = j,
1 2 fW
=
(1.18)
2 i j
li lj cos(j i ),
if i =
= j.
Since cos(i j ) = ui uj , we may rewrite (1.18) in the form
li2 (di 1) if i = j,
2
1 fW
(1.19)
(pJ ) =
l l u u if i == j,
2 i j
i j i j
where
di =
ui L J
.
li
14
(1.20)
r
9
dj (1
j=1
r
=
d1
).
=1
Let us show that the number in the brackets in formula (1.20) is always
non-negative and it vanishes only for r = n. Indeed, if r < k then
dj < 0 for 1 j r, and the statement is obvious. For r k we have
(1.21)
r
=
d1
L1
J (LJ
r
=
l i ui ) =
n
=
li ui 0,
i=r+1
i=1
=1
L1
J
()r
for 1 r < k,
A = D E,
15
2
MW = fW1
! ((ln ) ).
16
ak + an3k + bk ,
for any k = 0, 1, . . . , n 3.
By Theorem 1.7 the Poincare polynomial
p(t) =
n3
=
dim Hk (MW ; Q) tk
k=0
where
(1.26)
q(t) =
n3
=
k=0
ak t ,
r(t) =
n3
=
bk t k ;
k=0
17
18
then {1, r} and {r( , s( } would be two disjoint long subsets which is
impossible.
Example 1.13. As another example consider the equilateral case Z =
(1, 1, . . . , 1) with n = 2r + 1 odd. It is clearly generic and hence bk = 0.
We may x the rst link l1 as being the longest. The short subsets in
this case are subsets of {1, . . . , n} of cardinality r. Hence we nd
that the number ak equals
.n1(
for k r 1,
k
(1.27)
ak =
0,
for k r.
By Theorem 1.7 the Betti numbers of MW are given by
.n1(
for k < r 1,
. (
for k = r 1,
2 n1
bk (MW ) =
(1.28)
r1
.n1(
for k > r 1.
k+2
This coincides with the result of Theorem 1.1 from [58], see also Theorem C in [57].
Example 1.14. Consider now the equilateral case Z = (1, 1, . . . , 1) with
n = 2r + 2 even. We x l1 = 1 as the longest link. The short subsets
are all subsets of cardinality r and the median subsets are all subsets
of cardinality r + 1. Hence. we (nd that ak is given by formula (1.27),
bk = 0 for k =
= r and br = 2r+1
. Applying Theorem 1.7 we obtain
r
.n1(
for k r 1,
. (
n
for k = r,
bk (MW ) =
(1.29)
r
.n1(
for r + 1 k n 3.
k+2
This is consistent with the result of Theorem 1.1 from [58].
We will use later the observation (following from the last two examples) that the total Betti number
n3
=
i=0
bi (M(1,1,...,1) )
19
r
where r = [ (n1)
].
2
1.7. Morse theory on manifolds with involutions
Our main tool in computing the Betti numbers of the moduli space of
planar polygons MW is Morse theory of manifolds with involution.
Theorem 1.15. Let M be a smooth compact manifold with boundary.
Assume that M is equipped with a Morse function f : M [0, 1]
and with a smooth involution : M M satisfying the following
properties:
(1) f is -invariant, i.e., f ( x) = f (x) for any x M ;
(2) The critical points of f coincide with the xed points of the
involution;
(3) f 1 (1) = M and 1 [0, 1] is a regular value of f .
Then each homology group Hi (M ; Z) is free abelian of rank equal to
the number of critical points of f having Morse index i. Moreover, the
induced map
: Hi (M ; Z) Hi (M ; Z)
coincides with multiplication by (1)i for any i.
As an illustration for Theorem 1.15 consider a surface in R3 (see Figure 1.11) which is symmetric with respect to the z-axis. The function f
is the orthogonal projection onto the z-axis, the involution : M M
is given by (x, y, z) = (x, y, z). The critical points of f are exactly
z
20
dp (v) = v,
for any v Tp M.
H(f )p : Tp (M ) Tp (M ) R
v (x) = dx (vx ),
x M.
where in the summation is taken over the critical points q with Morse
index ind(q) = ind(p) 1. The incidence numbers [p : q] Z are
dened as follows
=
(1.35)
[p : q] =
i(), i() = 1,
21
i() + i( ) = 0,
22
W (p)
(t)
W (q)
23
for any i.
Proof of Theorem 1.16. First we note that each submanifold Xp
is orientable. Indeed, Theorem 1.15 applied to the restriction f |Xp
implies that f |Xp has a unique maximum and unique minimum3 and
the top homology group Hi (Xp ; Z) = Z is innite cyclic where i =
dim Xp = ind(p).
For a regular value a R of f we denote by M a M the preimage
f 1 (, a]. It is a compact manifold with boundary. It follows from
Theorem 1.15 that f has a unique local minimum and therefore M a
is either empty or connected. For a slightly above the minimum value
f (p0 ) = min f (M ) the manifold M a is a disc and the homology of M a
is obviously realized by the submanifold Xp0 = {p0 } M a .
We proceed by induction on a. Our inductive statement is that
the homology of M a is freely generated by the homology classes of the
submanifolds Xp where p runs over all critical points of f satisfying
f (p) a.
Suppose that the statement is true for a and the interval [a, b] contains a single critical value c. Let p1 , . . . , pr be the critical points of f
lying in f 1 (c). Set
r
/
X=
Xp i
i=1
0 C (X a ) C (X) C (M a ) C (M b ) 0
we assume that M and Xp are connected.
24
which (by the arguments indicated in the proof of Theorem 1.15) have
trivial dierentials and hence the sequence
(1.40)
0 Hi (X a ) Hi (X) Hi (M a ) Hi (M b ) 0
Proof. One simply checks that the coordinate changes in the standard
proof of the Morse lemma (compare [76], 2) can be chosen so that the
subspace Rk {0} is mapped to itself.
!
1.8. Proof of Theorem 1.7
Consider the moduli space W of the robot arm (dened by (1.13)) with
the function fW : W R (dened by (1.14)). There is an involution
(1.43)
:W W
25
given by
(1.44)
(u1 , . . . , un ) = (
u1 , . . . , un ).
Here the bar denotes complex conjugation, i.e., the reection with respect to the real axis. It is obvious that formula (1.44) maps SO(2)orbits into SO(2)-orbits and hence denes an involution on W . The
xed points of are the collinear congurations of the robot arm, i.e.,
the critical points of fW in W MW , see Lemma 1.4. Our plan is to
apply Theorems 1.15 and 1.16 to the sublevel sets
(1.45)
W a = fW1 (, a]
fW (pJ ) = (LJ )2 .
(1.47)
+
Here LJ = ni=1 li ui with pJ = (u1 , . . . , un ).
The number a which appears in (1.45) will be chosen so that
for any long subset J such that the manifold W a contains all the critical
points pJ . The situation is shown schematically on Figure 1.14.
MP
a
fP
Wa
26
: Hi (W a ; Z) Hi (W ; Z)
27
[WJ ] [WJ ! ] =
1, if |J J ( | = 1,
0,
if |J J ( | > 1.
t [0, 1].
Here satises 0 < < . Then WJ( ! = g1 (WJ ! ) is clearly disjoint from
WJ ; indeed, the links l1 and l2 are parallel in WJ and make an angle
in WJ( ! .
28
(1.51)
1, if J J ( = {1},
0,
if |J J ( | > 1.
(1.52)
Similarly, one has
(1.53)
Hi (W ; Z) = Ai Ci Di ,
where:
A is as above;
C H (W ; Z) is the subgroup generated by the homology
classes [WJ ] with J {1, . . . , n} short and 1 J;
D is the subgroup generated by the classes [WJ ] H (W ; Z)
where J is median and contains 1.
It is clear that (see (1.48)) is identical when restricted to Ai ,
compare (1.52) and (1.53). We claim that the image (Bi ) is contained
in Ai . This would follow once we show that
(1.54)
[WJ ] [WK ] = 0
J is long and 1
/ J,
J ( is short or median and 1 J ( ,
|J| = |J ( |,
K = CJ ( {1}.
29
rk(cokern1j ) = aj + bj ,
30
n3
=
bi (MW ).
i=0
(1.60)
.
r
This estimate is sharp: Bn equals the total Betti number of the moduli
space of planar equilateral n-gons, see Examples 1.13 and 1.14.
Note that for n even the equilateral linkage with n sides is not
generic and hence Theorem 1.21 does not answer the question about
the maximum of the total Betti number on the set of all generic length
vectors with n even.
Theorem 1.22. Assume that n is even and Z = (l1 , . . . , ln ) is a generic
length vector. Then the total Betti number of MW does not exceed
(1.61)
Bn( = 2 Bn1 ,
31
Cr =
,
r
r+1
r3/2
see [98]. One obtains the asymptotic formula
2
# 5
2
(1.62)
.
Bn 2n1 1
n
1.10. On the conjecture of Kevin Walker
In 1985 Kevin Walker in his study of topology of polygon spaces [100]
raised an interesting conjecture in the spirit of the well-known question
Can you hear the shape of a drum? of Marc Kac. Roughly, Walkers
conjecture asserts that one can recover relative lengths of the bars of
a linkage from intrinsic algebraic properties of the cohomology algebra
of its conguration space. In this section we survey results of a recent
paper [29] answering the conjecture for polygon spaces in R3 . In [29]
it is also proven that for planar polygon spaces the conjecture holds
in several modied forms: (a) if one takes into account the action of a
natural involution on cohomology, (b) if the cohomology algebra of the
involutions orbit space is known, or (c) if the length vector is normal.
Some results mentioned below allow the length vector to be non-generic
(the corresponding polygon space has singularities).
Walkers conjecture [100] states that for a generic length vector
Z the cohomology ring of MW determines the length vector Z up to a
natural equivalence (described below). To state the Walker conjecture
in full detail we recall the dependence of the conguration space MW on
the length vector
(1.63)
Z = (l1 , . . . , ln ) Rn+ .
Here Rn+ denotes the set of vectors in Rn having non-negative coordinates. Clearly, MW = MtW for any t > 0. Also, MW is dieomorphic to
MW! if Z( is obtained from Z by permuting coordinates.
5This
32
iJ
/
Here the symbol A(i) denotes the set of points Z A lying in at least
n 1 i linearly independent hyperplanes HJ for various subsets J. A
stratum of dimension k is a connected component of the complement
A(k) A(k1) . By Theorem 1.1 of [49], manifolds with singularities MW
and MW! are dieomorphic if the vectors Z and Z( belong to the same
stratum.
Lemma 1.23. Two length vectors Z, Z( An1 lie in the same stratum
of An1 if and only if the family of all subsets J {1, 2, . . . , n} which
are short with respect to Z coincides with the family of all subsets J
{1, 2, . . . , n} which are short with respect to Z( .
Proof. J is long i the complement J is short and J is median i
neither J nor J is short. Hence, vectors Z, Z( satisfying conditions of
the lemma have identical families of short, long and median subsets.
This clearly implies that Z and Z( lie in the same stratum of A.
!
Strata of dimension n 1 are called chambers. Vectors Z lying in
chambers are generic. Non-generic length vectors lie in walls separating
chambers and hence satisfy linear equations (1.64) for some J.
Walkers conjecture: Let Z, Z( A be two generic length vectors; if the corresponding polygon spaces MW and MW! have isomorphic graded integral cohomology rings, then for some permutation :
{1, . . . , n} {1, . . . , n} the length vectors Z and (Z( ) lie in the same
chamber of A.
It is important to recall that polygon spaces (1.3) come with a
natural involution
(1.66)
: MW MW ,
(u1 , . . . , un ) = (
u1 , . . . , un )
33
induced by complex conjugation; a similar involution played an important role earlier in section 1.8. Geometrically, this involution asso!2
!3
!1
!4
!4
!1
!3
!2
Theorem 1.24. [29] Suppose that two length vectors Z, Z( An1 are
ordered, i.e., Z = (l1 , l2 , . . . , ln ) with l1 l2 ln > 0 and
similarly for Z( . If there exists a graded ring isomorphism of the integral
cohomology algebras
f : H (MW ; Z) H (MW! ; Z)
commuting with the action of the involution (1.67), then Z and Z( lie in
the same stratum of A. In particular, under the above assumptions the
moduli spaces MW and MW! are -equivariantly dieomorphic.
Let Z, Z( An1 be length vectors (possibly non-generic) lying in
the same stratum. Then there exists a dieomorphism : MW! MW
which is equivariant with respect to the involution , see page 36 and
Remark 3.3 in [49].
W denote the factor-space of MW with respect to the involution
Let M
W is given by
(1.66). An alternative denition of M
(1.68)
W = {(u1 , . . . , un ) S 1 S 1 ;
M
n
=
i=1
li ui = 0}/O(2).
34
Theorem 1.25. [29] Suppose that two generic ordered length vectors
Z, Z( An1 are such that there exists a graded algebra isomorphism
W ; Z2 ) H ( M
W! ; Z2 )
f : H (M
of cohomology algebras with Z2 coecients. If n == 4, then Z and Z( lie
in the same chamber of A.
Theorem 1.25 is false for n = 4. Indeed, for the length vectors
W and M
W! are circles.
Z = (2, 1, 1, 1) and Z( = (2, 2, 2, 1) the manifolds M
However MW and MW! are not dieomorphic (the rst is S 1 and the
second is S 1 5 S 1 ) and thus Z and Z( do not lie in the same chamber.
In [29] we also prove a result in the spirit of Walkers conjecture
for the spatial polygon spaces. These spaces are dened by
n
=
2
2
(1.69) NW = {(u1 , . . . , un ) S S ;
li ui = 0}/SO(3).
i=1
35
Indeed, if this set is not short, then neither of the sets {n 3, n 2, n},
{n 3, n 1, n}, {n 2, n 1, n} is short and the intersection of
these four sets of cardinality three is empty. On the other hand, if the
set {n 3, n 2, n 1} is short, then any median or long subset of
cardinality three J {1, . . . , n} contains n and therefore J, where
|J| = 3, also contains n.
Examples of non-normal length vectors are (1, 1, 1, 1, 1) (for n = 5)
and (3, 2, 2, 2, 1, 1) for n = 6. Only 7 chambers out 21 are normal for
n = 6. However, for large n it is very likely that a randomly selected
length vector is normal. For n = 9, where there are 175428 chambers
up to permutation, 86% of them are normal. It is shown in [31] that
the (n 1)-dimensional volume of the union Nn An1 of all normal
strata satises
vol(An1 Nn )
n6
<
,
vol(An1 )
2n
i.e., for large n the relative volume of the union of non-normal strata
is exponentially small.
Theorem 1.28. [29] Suppose that Z, Z( An1 are two ordered length
vectors such that there exists a graded algebra isomorphism between the
integral cohomology algebras H (MW ; Z) H (MW! ; Z). Assume that
one of the vectors Z, Z( is normal. Then the other vector is normal as
well and Z and Z( lie in the same stratum of the simplex A.
Consider the action of the symmetric group n on the simplex An1
induced by permutations of vertices. This action denes an action of
n on the set of strata and we denote by cn and by cn the number
of distinct n -orbits of chambers (or chambers consisting of normal
length vectors, respectively).
Theorems 1.241.28 imply:
Theorem 1.29. [29] (a) For n == 4 the number of distinct dieomorphism types of manifolds NW , where Z runs over all generic vectors of
An1 , equals cn ;
(b) for n == 4 the number of distinct dieomorphism types of mani W , where Z runs over all generic vectors of An1 , equals cn ;
folds M
(c) the number xn of distinct dieomorphism types of manifolds MW ,
where Z runs over all generic vectors of An1 , satises cn xn cn ;
(d) the number of distinct dieomorphism types of manifolds with
singularities MW , where Z varies in An1 , is bounded above by the number of distinct n -orbits of strata of An1 and is bounded below by the
number of distinct n -orbits of normal strata of An1 .
36
Statements (a), (b), (c), (d) remain true if one replaces the words
dieomorphism types by homeomorphism types or by homotopy
types.
It is an interesting combinatorial problem to nd explicit formulae
for the numbers cn and cn and to understand their behavior for large
n . For n 9, the numbers cn have been determined in [49], by giving
an explicit list of the chambers. The following table gives the values cn
and cn for n 9:
n 3 4 5
65
1700
151317
We refer the reader to the original paper [29] for proofs of theorems
mentioned in this section and more details.
1.11. Topology of random linkages
In this section we study polygon spaces NW and MW assuming that the
number of links n is large, n . This approach is motivated by
applications in topological robotics, statistical shape theory and molecular biology. We view the lengths of the bars of the linkage as random
variables and study asymptotic values of the average Betti numbers
when n tends to innity. We describe a surprising fact (established in
[30], [31]) that for a reasonably ample class of sequences of probability measures the asymptotic values of the average Betti numbers are
independent of the choice of the measure.
The following picture summarizes our description of the eld of
topological spaces Z 7 NW viewed as a single object. The open simplex
n1 is divided into a huge number of tiny chambers, each representing
a dieomorphism type of manifolds NW . The symmetric group n acts
on the simplex n1 mapping chambers to chambers and manifolds NW
and NW! are dieomorphic if and only if the vectors Z, Z( lie in chambers
belonging the same n -orbit.
The main idea of this section is to use methods of probability theory and statistics in dealing with the variety of dieomorphism types
of conguration spaces NW and MW for n large. In applications dierent
manifold types appear with dierent probabilities and our intention is
to study the most frequently emerging manifolds NW and the mathematical expectations of their topological invariants. Formally, we view
the length vector Z n1 as a random variable whose statistical
37
n1
<
n1
.
bp (MW )dn
p
n1
It might appear surprising that the asymptotic values of average Betti
numbers do not depend on the sequence of probability measures n
which are allowed to vary in an ample class of admissible probability
measures described below. First we have to dene what is meant by
an admissible sequence of measures.
For a vector Z = (l1 , . . . , ln ) we denote by
;
vol(A)
vol(n1 )
Clearly, p q for p q.
6It
is well known that all odd-dimensional Betti numbers of N% vanish, see [65].
38
(1.72)
fn (Z) = kn |Z|n ,
Z n1 .
as the
If Z pn1 then fn (Z) kn (2p)n . We can represent n1
p
union A1 An where
Ai = {(l1 , . . . , ln ) n1 ; li (2p)1 },
!
>n1
2p1
Clearly, n (Ai ) = 2p
and hence
i = 1, . . . , n.
<n1
2p 1
.
n (
p ) 1 n
2p
;
!
>n1 <
2p1
1
n
Using (1.74) we nd that kn (2p) 1 n 2p
. This shows
;
n1
39
The following two theorems are the main results of this section.
Theorem 1.32. [31] Fix an admissible sequence of probability measures
n and an integer p 0, and consider the average 2p-dimensional Betti
number (1.70) of polygon spaces NW in R3 for large n . Then there
exist constants C > 0 and 0 < a < 1 (depending on the sequence of
measures n and on the number p but independent of n) such that the
average Betti numbers (1.70) satisfy
$
$
$ *
<
p ;
= n 1 $$
$
$ < C an
$
(1.75)
b2p (NW )dn
$
$
i
$
$ n1
i=0
for all n.
Theorem 1.33. [31] Fix an admissible sequence of probability measures
n and an integer p 0, and consider the average p-dimensional Betti
number (1.70) of planar polygon spaces for large n . Then there
exist constants C > 0 and 0 < a < 1 (depending on the sequence of
measures n and on the number p but independent of n) such that
$
$
$ *
;
<$
$
n 1 $$
n
$
bp (MW )dn
(1.76)
$
$<C a
p
$ n1
$
for all n.
Proofs of Theorems 1.32 and 1.33 can be found in [31].
Matthew Hunt [52] applied Monte Carlo simulation to compute
numerically the average Betti numbers of planar polygon spaces MW for
various n 11. His numerical results conrm Theorem 1.33.
CHAPTER 2
(B(X, n)) =
(F (X, n))
n!
where n = 1, 2, . . . . One formally denes F (X, 0) and B(X, 0) as singletons (i.e., spaces consisting of a single point) so that
(B(X, 0)) = (F (X, 0)) = 1.
42
=
=
tn
n
(B(X, n)) t =
(F (X, n)) .
(2.2) euX (t) =
n!
n=0
n=0
The latter is called the Euler Gal power series of X. The constant
term of euX (t) is 1. We shall see that euX (t) has a fairly simple expression while the individual numbers (2.1) are much more involved.
Theorem 2.1. For any nite polyhedron X the Euler Gal power
series euX (t) represents a rational function
(2.3)
euX (t) =
p(t)
,
q(t)
where p(t) and q(t) are polynomials with integral coecients satisfying
p(0) = 1 = q(0),
n = a1 n1 + a2 n2 + + ar nr .
43
C(L)
(X)
the reduced Euler characteristic. Here C(X) denotes the cone over X.
The reduced Euler characteristic behaves well with respect to the join
operation:
(2.6)
(X
Y ) = (X)
(Y
).
44
(S
k ) = (1)k+1 .
The statement given below is not used in the sequel and therefore it
is left as an exercise. It is a special case of Theorem 2.3 the equation
for the rst order terms in t.
Exercise: Show that for a nite polyhedral cell complex X one has
=
(2.8)
(L
),
(X) =
and
(2.10)
q(t) =
[1 t(L
)] .
dim =odd
In (2.9) and in (2.10) runs over all cells of X having even or odd
dimension, correspondingly.
Corollary 2.4. The zeros of the rational function euX (t) are of the
form
(2.11)
t = (L
)1 ,
t = (L
)1 ,
45
( 1) . . . ( k + 1) if dim X is even,
(F (X, k)) =
( + 1) . . . ( + k 1) if dim X is odd.
Here X is a compact manifold, possibly with boundary, and = (X).
Theorem 2.5 may also be obtained by examining the towers of
Fadell Neuwirth brations [21]: if X is a manifold without boundary,
then projecting onto the rst coordinate gives a locally trivial bration
F (X, n) X. Its bre above a point p X equals F (X {p}, n 1),
the conguration space of n 1 distinct points in X {p}. Using the
multiplicative property of the Euler characteristic1 we nd
(2.14)
Iterating we obtain
(F (X, n)) = (X) (X1 ) (Xn1 ),
where each Xi is obtained from X by removing i distinct points. This
gives the formulae mentioned above since (Xi ) = (X) (1)dim X i.
Next we give a proof of Theorem 2.5 based on Theorem 2.3. For
every cell lying in the interior of X one has L = S nd 1 and
1It
states that for a bration E B with bre F one has (E) = (B)(F ).
46
(L
) = (1)nd where n = dim X and d = dim . If is a cell
lying in the boundary, then (L
) = 0. Hence Theorem 2.3 gives
(2.15)
.
((X)(X)
.
euX (t) = 1 + (1)dim X t
This implies (2.13) since for dim X even one has (X) = 0 and for
dim X odd, (X) = 2(X).
2.3. Conguration spaces of graphs
Next we examine the special case of Theorem 2.3 when X = is a nite
graph, i.e., a one-dimensional nite simplicial complex. For any vertex
v the link Lv is the discrete set of vertices which are connected to
v by an edge in . Hence
(L
v ) = 1 (v)
where (v) denotes the valence of v. For any edge e the link Le is
empty and therefore
(L
e ) = 1.
Applying Theorem 2.3 we nd
Theorem 2.6. The Euler Gal power series of a graph is given by
the formula
(2.16)
eu (t) = (1 t)E
0
v
[1 + t(1 (v))]
? 0
. (2
4
. (
t + [1 + t(1 (v))] .
= 1 + E1 t + E+1
2
v
Here E denotes the total number of edges in and v runs over all
vertices of .
Observe that in the product appearing in (2.16) the univalent vertices (v) = 1 give no contribution. As another observation note that
subdividing an edge by introducing a new vertex of valence 2 makes no
change to the Euler Gal series (2.16) as two new terms cancel each
other.
As an illustration we use formula (2.16) to compute explicitly the
Euler characteristic (F (, 2)), which equals twice the coecient of t2
in the above series.
47
=
1=
(v)
(w),
2 v
w
(V 2 2V E + E 2 ) + (V E) 2V + 6E F
which is equivalent to (2.17).
48
1 + t(1 )
.
(1 t)
Hence,
(F ( , n)) =
( + n 2)!
[(n 1) 2n + 1] .
( 1)!
(B(, n)) c nE 1 .
(v)1=2
(E ( 1)!
where
(v) = 1 (v) and
9
aV ! =
(2 (v)) == 0.
(v)1=2
<
;
=
k+n1
n=0
tn
49
aV !
For n large the rst term in square brackets dominates and we may
ignore the other terms. We obtain that asymptotically one has
(2.21)
(F (, n))
9
(v)1=2
(2 (v))
(n + E ( 1)!
.
(E ( 1)!
where runs over all simplices of X. Formula (2.22) can be equivalently rewritten as
=
d
(2.23)
euX (t) =
euX&3 (t) (L
).
dt
50
!"
!"
(2.24)
i=1
as follows from invariance of the Euler characteristic. Under the subdivision the term
(2.25)
(F (X +4, n 1) (L
)
k
=
(F (X +i 4, n 1) (L
i ).
i=1
Note that
(2.27)
F (X +i 4, n 1) = F (X +4, n 1)
(L
i ) = (1)ddi (L
).
51
Ak Bk Ck Ak+1 .
d(x1 , X (l) ) il ,
for l < k.
52
(Ak , Ak+1 )
k=0
k=0
(F (X +4, n 1) (L
).
N2
N3
N 1
Figure 2.4. Neighborhoods N .
omorphic to the product int(C(L )) and the boundary of N in
(Bk ) is homeomorphic to L . Over N the map : F (X, n) X
is a locally trivial bration with bre F (X +4, n 1). Hence we
obtain
+
(Bk , Ck ) =
(F (X +4, n 1) () (CL , L )
dim =k
+
dim =k
(F (X +4, n 1) (L
).
53
R ik
2i2k R2
t
+
t2
k
k
R(2i R)
R(2i R)
y = (|y|, arg(y))
54
F, (y)
x1
pk(x1)
= (x1 , . . . , xn ),
[0, 1]
55
for any n.
Proof. We prove (2.32) by induction on n. It is trivial for n = 0 and
is obvious for n = 1. Suppose that it is true for n 1. By Theorem 2.9
we have
=
(F (X +4), n 1) (L
X
(2.33) (F (X, n)) =
),
and
(2.34)
(F (Y, n)) =
(F (Y +4), n 1) (L
Y ).
56
[X] [Y ] = [X Y ].
(F (X 5 Y, n))
n=0
tn
.
n!
n
=
(F (X 5 Y, n))
(F (X, k)) (F (X, n k))
=
,
n!
k!
(n
k)!
k=0
A
S
57
To see this start with the disjoint union X 5 (S [1, 1]), cut along
S 5 S (one copy S X and another S = S 0 S [1, 1]) and
glue one of these copies to another one and vice versa. The result is
homeomorphic to A 5 B which proves (2.36). Combining this result
with Proposition 2.11 we obtain:
Corollary 2.12. Under the conditions described above one has
(2.37)
euX (t) =
euCX (t)
= 1 + t(X).
euXI (t)
58
euCX&! 3 (t) =
euCX (t)
euXI&! 3 (t).
euXI (t)
eu(CX (t) =
euCX&3 (L
CX
)
euCX (t) =
euXI&3 (L
XI
) + euXI (t) (X).
euXI (t)
In the rst sum in (2.41) the symbol runs over all cells of X I of
the form I where is a simplex of X. However, links of the cells
of X I which intersect the top and the bottom faces of the cylinder
X I are contractible and therefore for such cells (L
) = 0. Hence,
we may apply Theorem 2.9 again to obtain:
(2.42)
eu(CX =
euCX
eu(XI + euXI (X),
euXI
euCX
euXI
<(
= (X).
This proves Proposition 2.13 since the constant term of Euler Gal
power series is always 1.
!
59
(2.44)
(Y
) = (1)dim (L
).
1
1 + t(Y
)
1
1+
(1)dim
(L
) t
(L
)
.
dim
1 + (1) (L
) t
(log euX ) =
log
[1 + (1)dim (L
)t](1)
dt
dt
or, equivalently,
(2.45)
euX = C
p(t)
q(t)
CHAPTER 3
62
li
pi+1
pi pi1
,
|pi pi1 |
i = 2, . . . , n
63
64
+vi+1 vi , pi+1 pi 4 = 0.
This is a consequence of the fact that the derivative of |pi+1 (t) pi (t)|2
equals twice the scalar product 2 +p i+1 (t) p i (t), pi+1 (t) pi (t)4. The
velocity vector w of a point c lying on the bar pi pi+1 is given by
(3.2)
w=
|c pi |
|c pi+1 |
vi +
vi+1 .
|pi pi+1 |
|pi pi+1 |
65
p1 , p2 , p3 ) satisfying
(3.3)
(3.4)
(3.5)
+v2 v1 , p2 p1 4 = 0,
+v3 v2 , p3 p2 4 0,
+v3 v1 , p3 p1 4 0.
|c p2 |
|c p1 |
v1 +
v2 .
|p1 p2 |
|p2 p1 |
+v3 w, p3 c4 0.
v3
p3
p1
v1
c
v2
p2
66
0, 0.
|p1 c|
|p2 c|
(p3 p2 ) +
(p3 p1 ).
|p1 p2 |
|p2 p1 |
Note in the last formula the coecients of both and are positive.
Finally, if one of the inequalities (3.4) or (3.5) is strict, then either
> 0 or > 0 and hence +v3 , p3 c4 > 0.
Now consider the remaining case when p3 p1 and p3 p2 are linearly dependent, i.e., the points p1 , p2 , p3 lie on a straight line. There
are two subcases: (a) p3 lies between p1 and p2 and (b) p3
/ [p1 , p2 ].
Assuming (a), we see that (3.6) is obvious, however none of (3.4), (3.5),
(3.6) can be satised with a strong inequality.
In the case (b) the statement of Lemma 3.3 is obvious.
!
Corollary 3.4. For any expansive motion, the distance between any
pair of points of the robot arm is nondecreasing. In particular, no selfintersection may occur in an expansive motion p1 (t), . . . , pn (t) starting
with a conguration p1 (0), . . . , pn (0) without self-intersections.
Proof. Consider a point c lying on the bar p1 p2 and its distance to
another vertex p3 (see Figure 3.5, left). We know that p3 moves such
that the distances from p3 to p1 and p2 are nondecreasing. By Lemma
3.3, the distance from p3 to c is nondecreasing as well.
p3
p1
p4
p2 p1
67
p5
p2
+vi+1 vi , pi+1 pi 4 = 0,
where pj = pj (0). On the other hand, the distance |pi (t) pj (t)| must
be nondecreasing in t for |i j| > 1 and hence
(3.8)
+vi vj , pi pj 4 0,
for |i j| > 1.
+vj vi , pj pi 4 > 0.
One of the key steps in the proof of Theorem 3.2 is the following
statement [10].
68
Figure 3.6. The initial conguration (left) and the planar framework obtained by adding struts (right).
Let us denote by B the set of bars and by S the set of struts. We
will write [i, j] S or [i, j] B to indicate that the segment connecting
pi and pj is a strut or a bar, correspondingly. Here [i, j] denotes an
unordered pair of indices. The conditions describing expansive innitesimal motions are:
+vi vj , pi pj 4 = 0,
if [i, j] B,
(3.10)
+vi vj , pi pj 4 > 0,
if [i, j] S.
69
for
i = 1, . . . , k,
+ai , x4 = 0
for i = k + 1, . . . , m.
i ai = 0,
i=1
+aij , x4 > 0,
+aij , x4 = 0,
if [i, j] S,
if [i, j] B.
By Theorem 3.7, this system has a solution if and only if there does
not exist a function assigning a real number
ij = ji ,
i == j
70
vanishes. In the last sum the summation is over all indices j such that
the pair [i, j] belongs to B S.
There is a useful mechanical interpretation of the weights ij . One
can view ij as the stresses induced in the bar or strut [i, j] B S.
A negative stress means that the edge is pushing on its endpoints by
an equal amount, a positive stress means that that the edge is pulling
on its endpoints by an equal amount, and zero means that the edge
induces no force. The major condition (3.13) can now be interpreted
by saying that the total force applied to any vertex equates to zero.
For that reason a stress
(3.14)
[i, j] 7 ij ,
[i, j] B S
--1
--1
1
--1
--1
71
(
kl
= ij
|pi pj |
.
|p(k p(l |
72
If the initial stress was proper, i.e., the struts had non-negative
stresses, then the obtained stress ( is proper as well.
Suppose that the initial graph had a strut with a positive stress.
After performing the operations of subdivision and merging edges, a
part of this edge remains to be a strut with a positive stress.
3.5. Maxwell Cremona Theorem
As the result of planarization described in the previous section we obtain a convex polygonal domain Y R2 subdivided into nitely many
polyhedral cells, see Figure 3.8. The exterior of Y in R2 will be denoted
by F0 . We consider continuous functions f : R2 R which are ane
when restricted onto every cell F of Y or onto the exterior F0 . In other
words,
(3.16)
f (x) = +aF , x4 + bF ,
xF
aF ! aF = e e ,
e R,
F
e
73
assigning real valued weights2 to the edges of Y . Note that the order of
F and F ( is irrelevant: if the roles of F and F ( are interchanged, then
the vector e reverses and the value e remains unchanged.
Clearly, e is the jump of the slope of f when crossing the edge e
along the line perpendicular to the edge e. The graph of f is a piecewise
linear surface with edges lying above the edges of . One may write
e = tan(e )
where e is the dihedral external angle between the faces of the graph
of f lying above F and F ( , see Figure 3.10. Edges of having positive
stresses are called valleys and edges having negative stresses are called
mountains.
e >0
F
Valley
e <0
Mountain
Figure 3.10. Geometric interpretation of the stress determined by a piecewise ane function.
The following statement is known as the Maxwell Cremona theorem; it was known to James Clerk Maxwell and Luigi Cremona in the
nineteenth century.
Theorem 3.9. (i) The stress function e
7 e dened by the equation
(3.17) is an equilibrium stress on the planar framework . (ii) Any
equilibrium stress on can be realized by a continuous piecewise ane
function f : R2 R which is ane when restricted to any cell of Y
and on the exterior of Y . Such f is unique up to addition of an ane
function.
Proof. Consider a vertex p and the cells F1 , . . . , Fk incident to
it; one of the Fi s can be the exterior F0 of Y . We will assume that
F1 , . . . , Fk appear in cyclic order in the anticlockwise direction. Let ei
2Our
74
(3.18)
where e
i is the vector perpendicular to ei satisfying |ei | = |ei | and
i=1
75
e1
F2
F1
p
ek
Fk
76
b1
c
b2
b3
b2
b1
v
b3
77
bedding XW T
is given by assigning to a conguration the set
78
of angles which the bars of the arm make with the rst bar. Alternatively, we may think that the rst bar [p1 p2 ] is pinned at points
W parameterizes the variety of all
p1 = (0, 0) and p2 = (l1 , 0); then X
possible positions of p3 , . . . , pn which are determined by all possible
angles 3 , 4 , . . . , n with the x-axis, see Figure 3.13. Given a congp3
pn
n
p2
p1
pn-1
vi Tpi (R2 ),
l1
p1
l2
p2
ln-1
pn-1
pn
79
i+1 R.
(3.21)
vi =
i
=
r (pr pr1 ) ,
i = 3, . . . , n.
r=3
j
=
r=i+1
j
=
r, s=i+1
If k denotes the angle between pk pk1 and the x-axis (as shown on
Figure 3.13) then one has
det(ps ps1 , pr pr1 ) = ls lr sin(s r )
and inequality (3.24) turns into
=
(3.25)
lr ls (r s ) sin(s r ) > 0.
i+1r<sj
80
p1
p k1
p km
k +1 = k +2 = = k+1 .
One may view this conguration as a robot arm with fewer vertices
p1 , pk1 , . . . , pkm , pn and apply to it Theorem 3.6. We obtain that there
81
i = j
This is obvious from the interpretation of numbers i as angular velocities i (see above) since all subdivided parts of a segment move with
the same velocity when viewed as parts of the integrated arm.
W near
Consider now admissible congurations q = (q1 , . . . , qn ) X
p. Each such conguration can be also described by a sequence of angles
(3 , 4 , . . . , n ). A tangent vector eld V near p will be described by
i =
i (3 , . . . , n ), where i =
a sequence of C -smooth functions
3, . . . , n. We set
(3.28)
i = i i + i ,
i = 3, . . . , n.
indices i + 1 < j such that the part of p is not straight between i and
j, then inequality (3.25) is satised and hence it also holds for close
congurations q. Finally consider the case when the part of p between
i and j is collinear. Then using (3.27) and (3.28) we nd
=
r
s ) sin(s r )
lr ls (
i+1r<sj
lr ls (s r ) sin(s r ).
i+1r<sj
Clearly there exists i > 0 such that for all x == 0, |x| < i one has
x sin x > 0. Hence, each term in the above sum is non-negative and
the sum vanishes if and only if the segment of conguration q between
the vertices qi and qj is collinear. This proves that V is an expansive
tangent vector eld in a small open neighborhood of p.
Now, if locally the elds VU are constructed on an open covering
W T n2 , one takes a smooth partition of unity {U } subor{U } of X
dinate to this covering and denes
=
(3.29)
V =
U VU .
U
W T n2 .
This gives the required eld in X
82
(3.30)
x(p, 0) = p,
t 0.
Here we use the well-known general theorem about existence of solutions of ordinary dierential equations with continuous right-hand side.
One may view x(p, t) as a continuous process of modication of the robot arm starting with the initial conguration p for t = 0. Our aim is
to show that x(p, t) converges to p0 for t .
Lemma 3.14. Let p1 , p2 , . . . be a sequence of admissible congurations,
W , and let tn > 0 be a sequence of real numbers. If the limits
pn X
lim x(pn , tn ) = q T n2 ,
lim pn = p T n2
83
li a
r
=
li
i=1
the point sp(a) lies on the segment [pr , pr+1 ] and satises
|sp(a) pr | = a
r1
=
li .
i=1
belongs to XW .
W is smooth and therefore (invoking a wellThe vector eld V |X
known result from the theory of ordinary dierential equations) we
conclude now that the solution x(p, t) of the dierential equation (3.30)
is smooth as a function of p and t.
W one has
Let us now show that for any initial conguration p X
(3.31)
lim x(p, t) = p0 ,
84
Tn
tn
*
|x|dt
Tn
tn
Tn
tn
and therefore
F (x(p, tn+1 )) F (x(p, Tn )) F (x(p, tn )) +
i
,
2C
85
Now suppose that t is such that x(p, ) U for all t. Using the
Mean Value Theorem we nd some satisfying t such that
F (x(p, )) F (x(p, t)) = Vx(p,) (F ) ( t) > i/2 ( t)
which again contradicts the fact that F is bounded. Hence we obtain
a contradiction to our initial assumption that q == p0 .
This argument proves that q = p0 is the only limit point of the
trajectory {x(p, t); t 0} and hence for any admissible conguration
W the trajectory x(p, t) converges to p0 as t .
pX
W is contractible. Dene a deformaOur nal goal is to show that X
tion
W [0, 1] X
W
H:X
by the formula
;
<
W , [0, 1)
H(p, ) = x p,
, for p X
1
and
H(p, 1) = p0 .
We claim that H is continuous, which is equivalent to the following
statement: For any neighborhood U of p0 there exists a neighborhood
V of p and a real number T such that for any t T and q V one
has x(q, t) U .
The negation of this statement is equivalent to the existence of
W converging to p and a
a neighborhood U of p0 , a sequence pn X
sequence tn such that x(pn , tn )
/ U . Set
S = {x(pn , t); n N, t 0}.
W . By our assumptions,
By Lemma 3.7 the closure S is contained in X
1
for all x S.
We obtain that
F (x(pn , tn )) F (x(pn , 0)) = Vx(pn ,) (F ) tn c tn .
This implies that the sequence F (x(pn , tn )) is unbounded, a contradiction.
!
CHAPTER 4
88
4. NAVIGATIONAL COMPLEXITY
: PX X X
s : X X PX
satisfying
(4.3)
s = 1XX .
89
X X = F1 F2 F3 Fk
such that:
1. The restriction s|Fi : Fi P X is continuous, i = 1, . . . , k;
2. Fi Fj = , where i == j;
90
4. NAVIGATIONAL COMPLEXITY
91
A towards B along the shortest geodesic arc. Consider now the set
A
B
--A
--A
In the case when n is even the above procedure has to be modied
since for n even any vector eld v tangent to S n has at least one zero.
However, we may nd a tangent vector eld v having a single zero
A0 S n . Write F2 = {(A, A); A == A0 } and dene s2 : F2 P S n
as in the previous paragraph. The set F3 = {(A0 , A0 )} consists of a
single pair; dene s3 : F3 P S n by choosing an arbitrary path from
A0 to A0 .
Next we explain the relation between the invariant TC(X) and the
notion of Schwarz genus of a bration. Recall that the Schwarz genus
of a bration p : E B is dened as the minimal number k such that
there exists an open cover of the base B = U1 U2 Uk with the
property that over each set Uj B there exists a continuous section
sj : Uj E of p : E B, see [93].
The concept of Schwarz genus generalizes the notion of Lusternik
Schnirelmann category cat(X); the latter is dened as the smallest
integer k such that X admits an open cover
X = U1 U2 Uk
92
4. NAVIGATIONAL COMPLEXITY
93
Definition 4.11. Let X be a topological space. Its topological complexity TC(X) is dened as the Schwarz genus of bration (4.1).
94
4. NAVIGATIONAL COMPLEXITY
Several equivalent characterizations of TC(X) are given by Proposition 4.12 in the case when X is an ENR.
Proposition 4.12. Let X be an ENR. Then TC(X) = k = Z = r
where the numbers k = k(X), Z = Z(X), r = r(X) are dened as
follows:
(a) k = k(X) is the minimal integer such that there exist a section
s : X X P X of bration (4.1) and an increasing sequence
of k open subsets
U1 U2 Uk = X X
with the property that for any i = 0, 1, . . . , k 1 the restriction
s|(Ui+1 Ui ) is continuous; here U0 = .
(b) l = l(X) is the minimal integer such that there exist a section
s : X X P X of (4.1) and an increasing sequence of k
closed subsets
F1 F2 Fk = X X
with the property that for any i = 0, 1, . . . , k 1 the restriction
s|(Fi+1 Fi ) is continuous; here F0 = .
(c) r = r(X) is the minimal integer such that there exist a section
s : X X P X of bration (4.1) and a splitting
G1 G2 Gr = X X,
Gi Gj = ,
i == j
95
96
4. NAVIGATIONAL COMPLEXITY
max
(x,y)XX
(x, y).
TC(X) 2 dim X + 1.
Proof. This corollary is obviously true for any ENR as follows from
the previous corollary. We state and prove it for polyhedra since in
this case we may give an explicit construction of a motion planning
algorithm.
Let X (r) = X r X r1 denote the union of interiors of all simplices
of dimension r where r = 0, 1, . . . , n. Dene
,
Gi =
X (r) X (s) , i = 0, 1, . . . , 2n.
r+s=i
97
The result of Corollary 4.15 would follow from Proposition 4.12 once
we show that each set Gi admits a continuous section si : Gi P X.
One has
6
Gi =
(r) (s)
r+s=i
TC(X) <
2 dim(X) + 1
+ 1.
r+1
TC(X) 3.
Y
q
98
4. NAVIGATIONAL COMPLEXITY
Proof. This follows directly from Denition 4.11, see [22], Theorem
5.
!
Exercise: Let G be a connected Lie group. Then
(4.10)
TC(G) = cat(G).
As a corollary we obtain
(4.11)
99
TC(T n ) = cat(T n ) = n + 1.
and
(4.14)
If A B X X then
(4.15)
100
4. NAVIGATIONAL COMPLEXITY
TCX (Y Y ) TC(Y ).
101
102
4. NAVIGATIONAL COMPLEXITY
103
F (x, y) = |x y|2
(x(t),
y(t))
t [0, ),
104
4. NAVIGATIONAL COMPLEXITY
M M = F1 Fk ,
and Fi Fj = , i == j. Each Fi is a submanifold (although not necessarily compact), hence an ENR. More precisely, Fi is homeomorphic
to the total space of a vector bundle over Si of rank
2 dim M ind Si dim Si
where ind Si denotes the Bott index of Si , viewed as a critical submanifold of F . The projection i : Fi Si is given by
i (A, B) = lim (x(t), y(t))
t
where (x(t), y(t)) is the trajectory (4.23) satisfying the initial conditions
x(0) = A, y(0) = B.
For a critical value r Crit(F ) denote by Cr and Br the unions
,
,
Cr =
Fi and Br =
Si .
ci =r
ci =r
By Proposition 4.12,
TC(M )
TCM (Cr )
rCrit(F )
ci =r
105
Remark 4.33. The proof above uses the observation that dierent
critical submanifolds Si , Sj M M lying on the same level set of
F can be aggregated, i.e., united into a single submanifold so that
the map qr (see above) remains continuous. This allows signicant
improvement of the upper bound replacing the summation operation
by the maximum, compare (4.22).
More generally, two critical submanifolds Si , Sj can be aggregated
for this purpose assuming that there are no orbits of the gradient ow
of F connecting them.
Note also that the proof of Theorem 4.32 does engage property (b)
from Denition 4.30.
Example 4.34. Let M = S 1 S 1 S 1 = T n be the n-dimensional
torus. Consider the navigation function F : M M R given by
F (u, z) =
n
=
|ui zi |2 .
i=1
Gij Gij ! = ,
j == j ( ,
106
4. NAVIGATIONAL COMPLEXITY
sij : Gij P M,
i = 1, . . . , k,
j = 1, . . . , ri
i = 1, . . . , k,
j = 1, . . . , ri .
t
0
|grad(F)(x(), y())|2 d.
Then the trajectory (x( ), y( )) (see (4.23)) reaches the critical submanifold Si in nite time
(4.27)
= F (A, B) Fi
x = 2Px (y x),
y = 2Py (x y),
107
(4.29)
(4.30)
m
=
ai b i ,
ai H i (X; k),
bi H ni (X; k).
i=1
ai bi = 0,
i=1
108
4. NAVIGATIONAL COMPLEXITY
Cohomology classes having higher weight can be obtained as cupproducts of zero-divisors as concluded from the following statement:
Lemma 4.39. Let u H n (X X; R) and v H m (X X; R( ) be
two cohomology classes. Then the weight of their cup product uv
H n+m (X X; R R( ) satises
(4.31)
i = 1, . . . , k
109
1, if b1 (X) = 0,
2, if b1 (X) = 1,
TC(X) =
3, if b (X) > 1.
1
Proof. If b1 (X) = 0, then X is contractible and hence TC(X) = 1. If
b1 (X) = 1, then X is homotopy equivalent to the circle and therefore
TC(X) = TC(S 1 ) = 2, see above.
Assume now that b1 (X) > 1. Then there exist two linearly independent classes u1 , u2 H 1 (X). Thus
ui = 1 ui ui 1,
i = 1, 2
: H (; R) H +i (; S),
110
4. NAVIGATIONAL COMPLEXITY
which coincides with the standard notion of excess, see [74], page 27.
As noted above, any cohomology class u H j (X; R) determines a
class
u = 1 u u 1 H j (X X; R)
where denotes the cohomology cross product. Note that u is a zerodivisor and hence wgt(u) 1. Observe that
(u) = (p2 (u) p1 (u)) = p2 ((u)) p1 ((u)) = (u),
by the naturality and additivity of (here p1 , p2 : X X X are the
projections onto each factor).
111
(4.33)
(4.34)
C
pC
1 9 p2 .
A
u|A = (pA
2 ) (u) (p1 ) (u) H (A; R).
112
4. NAVIGATIONAL COMPLEXITY
where
ni {0, 1, . . . , m 1},
) = 4n + 2.
) = 2 dim(L2n+1
TC(L2n+1
p
p
2n+1
where x = 1 x x 1, y = 1 y y 1 H (Lp2n+1 Lm
; Zp ).
The cohomology class
; <
2n
2n
n
x y = (1)
L2n+1
; Zp )
x (y n y n ) H 4n+1 (L2n+1
p
m
n
. (
is nonzero provided that the binomial coecient 2n
is not divisible
n
by p. The latter condition is equivalent to the requirement that all
digits ni in the p-adic decomposition of n are small, i.e., satisfy
ni (p 1)/2, see [32], Lemma 19. Using (4.37) and Corollary 4.40
we obtain that under the assumptions of the theorem one has
TC(L2n+1
) > wgt(x) + 2n wgt(y) 4n + 1.
p
This proves the lower bound TC(L2n+1
) 4n + 2.
p
2n+1
The reverse upper bound TC(Lp ) 2(2n + 1) = 4n + 2 holds for
all lens spaces; we refer to [32] for more detail.
!
Next we state without proof two results from [32].
Let (n) denote the number of 1s in the dyadic expansion of n.
113
) = 4n + 2
) = 2 dim(L2n+1
TC(L2n+1
m
m
4, for m = 2,
6, for m 3.
The topological complexity of rational formal simply-connected topological spaces coincides with their zero-divisors cup-length plus 1, as
shown by L. Lechuga and A. Murillo [69]. In other words, Corollary
4.40 is sharp in this case.
In a recent preprint [8] Daniel Cohen and Goderzi Pruidze computed explicitly the topological complexity of right-angled Artin groups.
Let = (V , E ) be a nite graph with no loops or multiple edges. Here
V is the set of vertices of the graph and E is the set of edges. The
right-angled Artin group associated to is the group G = G with generators corresponding to vertices v V of and relations vw = wv
corresponding to edges {v, w} E . R. Charney and M. Davis [7]
and J. Meier and L. Van Wyk [72] describe explicitly the Eilenberg
MacLane space X of type K(G , 1). The main result of Cohen and
Pruidze [8] states that
(4.39)
TC(X ) = z() + 1,
where
z() = max |K1 K2 |
K1 ,K2
114
4. NAVIGATIONAL COMPLEXITY
where n = 0, . . . , k + k ( 1. Then
= W0 W1 Wk+k! 1 = (X X) (Y Y )
and
Wn Wn1 =
115
(
)
(Ui Ui1 ) (Uj( Uj1
i+j=n
(
))
is a disjoint union and the section (s s( )|((Ui Ui1 ) (Uj( Uj1
is continuous. The result now follows from Proposition 4.12.
!
: 1 X2 Xk )
TC(X
k
=
: i ).
TC(X
i=1
Hence, if one controls simultaneously k systems having conguration spaces X1 , . . . , Xk , the total conguration space is the Cartesian
product X1 X2 Xk and its topological complexity is bounded
above by the sum of topological complexities of individual systems according to inequality (4.41).
To obtain a lower bound we introduce the following notation. For
a topological space X we denote by zcl(X) the largest integer k such
that there exist k zero-divisors u1 , . . . , uk H (X X; Q) having a
nontrivial cup-product u1 u2 . . . uk == 0 H (X X; Q). Corollary 4.40
gives the inequality
:
TC(X)
zcl(X).
(4.42)
Lemma 4.52. One has
116
4. NAVIGATIONAL COMPLEXITY
: 1 X2 Xk ) +k zcl(Xi ).
Corollary 4.53. TC(X
i=1
Corollary 4.54. Suppose that one controls simultaneously k systems
having path-connected conguration spaces X1 , . . . , Xk . Assume that
: i ) M and (2) each Xi has a
(1) for some constant M one has TC(X
(Xi ; Q) =
nontrivial reduced rational cohomology H
= 0. Then
: 1 X2 Xk ) kM
k TC(X
: 1 Xk ) grows linearly with k.
i.e., the number TC(X
Example 4.55. Imagine that we control k identical systems, each having conguration space X = R2 B, where B R2 is a ball (repre:
: n ) = n.
senting an obstacle). Then TC(X)
= 1, zcl(X) = 1 and TC(X
n
Here X denotes the n-fold Cartesian product X X.
In the case when X = R3 B, where B R3 is a ball, the answers
:
: n ) = 2n.
are slightly dierent: TC(X)
= 2, zcl(X) = 2 and TC(X
Comparing these two cases we conclude that the complexity of the
motion planning problem in R3 is twice the complexity of the planar
motion planning problem.
So far we have discussed the problem of centralized control which
is characterized by centralized decision making. Now let us consider
algorithms of distributed control, i.e., when the controllable objects
have their own motion planning algorithms and behave independently
of the behavior of the others. For simplicity we will assume that the
objects we control are identical, each object has conguration space X
and all motion planning algorithms are tame, see Denition 4.4.
117
TC(F (R , n)) =
118
4. NAVIGATIONAL COMPLEXITY
(y1 , y2 , . . . , yn ) 7
yi yj
S m1 .
|yi yj |
eij = ij [S m1 ]
e2ij = 0,
for any triple i, j, k. It follows that a product ei1 j1 ei2 j2 . . . eik jk is nonzero
if and only if the subgraph of the full graph on vertices {1, 2, . . . , n}
having the edges (ir , jr ) contains no cycles.
Hence for m 3 the conguration space F (Rm , n) has the homotopy type of a polyhedron of dimension (n 1)(m 1). Since it is
(m 2)-connected we may use inequality (4.6) to nd
TC(F (Rm , n)) <
2(n 1)(m 1) + 1
1
+ 1 = 2n 1 +
.
m1
m1
119
n
9
(
e1i )2
i=2
0
equals = (2)n1 m m where m = ni=2 e1i . The monomial m == 0
is nonzero and hence the product is nonzero.
The opposite inequality TC(F (Rm , n)) 2n 1 follows now from
Corollary 4.40. This completes the proof of Theorem 4.57 in the case
m 3 odd. Details of the proof in the case m = 2 can be found in
[26].
In a forthcoming joint paper with Mark Grant we show that
TC(F (Rm , n)) = 2n 2
for all even m.
In paper [33] we studied a more general problem of controlling
multiple particles such that there are no collisions between the particles
and a set of obstacles, which can be movable (however the trajectory
of the obstacles is known in advance).
Next we discuss the conguration spaces F (, n) where is a connected graph. These spaces were studied by R. Ghrist, D. Koditschek
and A. Abrams [39], [40], [2] ; see also [38], [90]. To illustrate the
importance of these conguration spaces for robotics one may mention
the control problems where a number of automated guided vehicles
(AGV) have to move along a network of oor wires. The motion of
the vehicles must be safe: it should be organized so that collisions do
not occur. If n is the number of AGV, then the natural conguration
space of this problem is F (, n) where is a graph.
The rst question to ask is whether the conguration space
F (, n) is connected. Clearly F (, n) is disconnected if = [0, 1]
is a closed interval (and n 2) or if = S 1 is the circle and n 3.
These are the only examples of this kind as the following simple lemma
claims:
Lemma 4.58. Let be a connected nite graph having at least one
essential vertex. Then the conguration space F (, n) is connected.
120
4. NAVIGATIONAL COMPLEXITY
Paper [24] contains a sketch of the proof and also an explicit description of a motion planning algorithm in F (, n) (assuming that
is a tree) having precisely 2m() + 1 domains of continuity.
If is homeomorphic to the letter Y , then m() = 1 and F (, 2) is
homotopy equivalent to the circle S 1 . Hence in this case TC(F (, 2)) =
2. The equality (4.45) fails in this case.
For any tree one has TC(F (, 2)) = 3 assuming that is not
homeomorphic to the letter Y . This example shows that the assumption n 2m() of Theorem 4.60 cannot be removed: if is a tree
with m() 2, then the inequality above would give TC(F (, 2)) =
2m() + 1 5.
Here are more examples. For the graphs K5 and K3,3
K5
K3,3
one has
(4.46)
121
B
O
Lines through the origin in R3 may represent metallic bars xed at the
xed point by a revolving joint; this situation is common in practical
robotics.
If the angle between the lines A and B is acute, then one may rotate
A towards B in the two-dimensional plane spanned by A and B such
that A sweeps the acute angle. Hence the problem reduces immediately
to the special case when the lines A and B are orthogonal. In this case,
if the intention is to use simple rotations, one needs a continuous choice
of the direction of rotation in the plane spanned by A and B.
Note that the Lusternik Schnirelmann category of the real projective spaces is well known and easy to compute: cat(RPn ) = n + 1,
see for example Proposition 4.5 of [53]. Using the general properties
of the topological complexity mentioned above we may write
n + 1 TC(RPn ) 2n + 1.
More precisely, one can show that TC(RPn ) 2n for all n and the
equality holds i n is a power of 2.
122
4. NAVIGATIONAL COMPLEXITY
123
126
in 2007 in the AMS series Contemporary Mathematics. This volume is a collection of papers written by participants of the conference
Topology and Robotics held in ETH Zurich in 2006.
The book [16], which appeared when these notes had been completed, is an amazing additional source of beautiful mathematics which
uses geometry and topology in service of algorithms in computer science and engineering.
Bibliography
[1] A. Abrams, Conguration spaces of colored graphs, Geom. Dedicata 92
(2002), 185194.
[2] A. Abrams, R. Ghrist, Finding topology in a factory: conguration
spaces. Amer. Math. Monthly 109 (2002), no. 2, 140150.
[3] L. Antoine, Sur lhomeomorphie de gures et de leur voisinages, J. Math.
Pure Appl. 86(1921), 221325.
[4] T. Biedl, E. Demaine, M. Demaine, S. Lazard, A. Lubiw, J. ORourke,
S. Robbins, I. Streinu, G. Toussaint, S. Whitesides, Locked and unlocked polygonal chains in three dimensions, Discrete and Computational
Geometry 26(2001), 283287.
[5] T. Biedl, E. Demaine, M. Demaine, S. Lazard, A. Lubiw, J. ORourke, S.
Robbins, I. Streinu, G. Toussaint, S. Whitesides, A note on reconguring
tree linkages: trees can lock, Discrete Applied Mathematics, 117(2002),
293297.
[6] J. Cantarella, H. Johnston, Nontrivial embeddings of polygonal intervals
and unknots in 3-space, Journal of Knot theory and its Ramications,
7(1998), 10271039.
[7] R. Charney, M.W. Davis, Finite K(, 1)s for Artin groups. Prospects
in topology (Princeton, NJ, 1994), 110124, Ann. of Math. Stud., 138,
Princeton Univ. Press, Princeton, NJ, 1995.
[8] D. C. Cohen, G. Pruidze, Motion planning in tori, Bulletin of the London
Mathematical Society, 40(2008), 249262.
[9] D. C. Cohen, G. Pruidze, Topological complexity of basis-conjugating
automorphism groups, Preprint arXiv:0804.1825.
[10] R. Connelly, E. Demaine, G. Rote, Straightening polygonal arcs and
convexifying polygonal cycles, Discrete and Computational Geometry,
30(2003), 205239.
[11] R. Connelly, Expansive motions, Preprint 2006.
[12] O. Cornea, G. Lupton, J. Oprea, D. Tanre, Lusternik Schnirelmann
category, AMS, 2003.
[13] M. Coste, An Introduction to Semialgebraic Geometry. Dip. Mat. Univ.
Pisa, Dottorato di Ricerca in Matematica, Istituti Editoriali e Poligraci
Internazionali, Pisa 2000.
[14] D. Davis, Immersions of projective spaces: a historical survey, Contemporary Mathematics, AMS, 146(1993), 3137.
[15] V. de Silva, R. Ghrist, Coverage in sensor networks via persistent homology, Algebr. Geom. Topol. 7 (2007), 339358.
[16] E. D. Demaine, J. ORourke, Geometric folding algorithms; Linkages,
origami, polyhedra, Cambridge University Press, 2007.
128
BIBLIOGRAPHY
BIBLIOGRAPHY
129
130
BIBLIOGRAPHY
BIBLIOGRAPHY
131
[80] P. Niyogi , S. Smale and S. Weinberger, Finding the Homology of Submanifolds with High Condence from Random Samples, Discrete Comput. Geom. 39(2008), no. 13, 419441.
[81] K.J. Pearson and T. Zang, Topological complexity and motion planning
in certain real Grassmannians, Applied Mathematics Letters, 17(2004),
499502.
[82] E. Rimon, D.E. Koditschek, The construction of analytic dieomorphisms for exact robot navigation on star worlds, Trans. Amer. Math.
Soc. 327 (1991), no. 1, 71116.
[83] R.T. Rockafellar, Convex analysis, Princeton University Press, 1970.
[84] E. Rodriguez, The snake charmer algorithm, Topology and Robotics,
(M. Farber, R. Ghrist et al editors), Contemporary Mathematics AMS,
vol. 438, 2007, 1932.
[85] C. P. Rourke, B.J. Sanderson, Introduction to Piecewise-linear topology,
Springer-Verlag, 1972.
[86] Y. B. Rudyak, On category weight and its applications, Topology, 38
(1999), no. 1, 3755.
[87] M. Sharir, Algorithmic motion planning. Handbook of Discrete and
Computational Geometry. J.E. Goodman, J. ORourke eds. CRC Press,
Boca Raton, FL, 1997, 733754.
[88] S. Smale, On the topology of algorithms, I, J. of Complexity, 3(1987),
8189.
[89] E. Spanier, Algebraic Topology, Mebman Hell, 1966.
[90] J. Swiatkowski, Estimates for the homological dimension of conguration
spaces of graphs, Colloq. Math., 89(2001), 6979.
[91] I. Streinu, Pseudo-Triangulations, Rigidity and Motion Planning, Discrete and Computational Geometry, 34(2005), 587635.
[92] N. Shvalb, M. Shoham, D. Blanc, The conguration space of arachnoid
mechanisms, Forum Mathematicum, 17(2005), 10331042.
[93] A.S. Schwarz, The genus of a ber space. Amer. Math. Sci. Transl.
55(1966), 49140.
[94] W. Thurston, Shapes of polyhedra, Preprint 1987.
[95] W. Thurston, J. Weeks, The mathematics of three-dimensional manifolds, Scientic American, July 1986, 94106.
[96] J.C. Trinkle, R.J. Milgram, Complete path planning for closed kinematic
chains, The International Journal of Robotics Research, 21(2002), 773
789.
[97] J.C. Trinkle, R.J. Milgram, The geometry of conguration spaces for
closed chains in two and three dimensions, Homology, homotopy and
applications, 6(2004), 237267.
[98] I. Vardi, Computational recreations in mathematics, Redwood City, CA,
Addison-Wesley, 1991.
[99] V.A. Vassiliev, Cohomology of braid groups and complexity of algorithms,
Functional Analysis and its Appl., 22(1988), 1524.
[100] K. Walker, Conguration spaces of linkages, Undergraduate thesis,
Princeton Univ., 1985.
[101] A.Wenger, Etudes des espaces de congurations de certains syst`emes
articules, Travail de dipl
ome, University of Geneva, 1988.
132
BIBLIOGRAPHY
Index
Median subset, 10
Morse Smale complex, 20
Motion planning algorithm, 88
Navigation function, 102
Normal length vector, 34
Planar linkage, 2
Poincare polynomial, 16
Polygon space, 3
Polyhedral cell complex, 43
Random linkage, 36
Reduced Euler characteristic, 43
Reduced topological complexity, 115
Relative topological complexity, 99
Robot arm, 5
Robot arm distance map, 11
Schwarz genus, 91
Short subset, 10
Singularities of M% , 15
Spatial polygon spaces, 34
Steenrod square, 110
Stratum, 32
Stress, 70
Strut, 68
Subspace arrangement, 118
Length vector, 3
Lens space, 111
Link of a simplex, 42
Linkage, 2
Long subset, 10
Lusternik Schnirelmann category,
91
Variety of quadrangles, 6
Zero-divisor, 107
Walkers conjecture, 31
Weight of cohomology class, 106
Workspace, 5