Combinatorial Rigidity PDF

Recent Titles in This Series
2 Jack Graver, Brigitte Servatlus, and Herman Servatius, Combinatorial rigidity, 1993
Ethan Akin, The general topology of dynamical systems, 1993
Combinatorial Rigidity
r
Graduate Studies
in Mathematics
Volume 2
Combinatorial Rigidity
Jack Graver
Brigitte Servatlus
Herman Servatlus
Editorial Board
James E. Humphreys
Robion C. Kirby
Lance W. Small
1991 Mathematics Subject Classification. Primary 05B35; Secondary 05C10.

ABSTRACT. Rigidity theory is introduced in an historical context. The combinatorial aspects of
rigidity are isolated and framed in terms of a special class of matriods. These matriods are a
natural generalization of the connectivity matriod of a graph. This book includes an introduction
to matriod theory and a comrehensive study of planar rigidity. The final chapter of the text
is devoted to higher dimensional rigidity, highlighting the main questions still open. This book
contains an extensive annotated bibliography.
Library of Congress Cataloging-in-Publication Data

Graver, Jack E., 1935-
Combinatorial rigidity/Jack Graver, Brigitte Servatius, Herman Servatius.
p. em. -(Graduate studies in mathematics, ISSN 1065-7339; v. 2)
Includes bibliographical references and index.
ISBN 0-8218-3801-6 (acid-free)
1. Matriods. 2. Topological graph theory. I. Servatius, Brigitte, 1954- II. Servatius,
Herman, 1957- III. Title. IV. Series.
QA166.6.G73 1993
511'.6-dc20 93-34431
CIP
Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting
for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
(including abstracts) is permitted only under license from the American Mathematical Society.
Requests for such permission should be addressed to the Manager of Editorial Services, American
Mathematical Society, P.O. Box 6248, Providence, Rhode Island 02940-6248. Requests can also
be made by e-mail to reprint-permissionGmath.ams.org.
The owner consents to copying beyond that permitted by Sections 107 or 108 of the U.S.
Copyright Law, provided that a fee of $1.00 plus $.25 per page for each copy be paid directly to
the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, Massachusetts 01923. When
paying this fee please use the code 1065-7339/93 to refer to this publication. This consent does
not extend to other kinds of copying, such as copying for general distribution, for advertising or
promotional purposes, for creating new collective works, or for resale.
©Copyright 1993 by the American Mathematical Society. All rights reserved.

The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America.
@The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
0 Printed on recycled paper.
This volume was typeset by the authors using AN(S-'IEX,
the American Mathematical Society's 'lEX macro system,
with technical and editorial assistance from the AMS.
10 9 8 7 6 5 4 3 2 1 98 97 96 95 94 93
Contents
Preface ix
Chapter 1. Overview 1
1.1. An Intuitive Introduction to Rigidity 1
1.2. A Short History of Rigidity 9
Chapter 2. Infinitesimal Rigidity 17

2.1. Basic Definitions 17
2.2. Independence and the Stress Space 24
2.3. Infinitesimal Motions and Isometries 30
2.4. Infinitesimal and Generic Rigidity 36
2.5. Rigidity Matroids 39
2.6. Isostatic Sets 47
Chapter 3. Matroid Theory 55

3.1. Closure Operators 55
3.2. Independence Systems 56
3.3. Basis Systems 59
3.4. Rank Function 60
3.5. Cycle Systems 62
3.6. Duality and Minors 68
3.7. Connectivity 74
3.8. Representability 79
3.9. Transversal Matroids 82
3.10. Graphic Matroids 84
3.11. Abstract Rigidity Matroids 86
Chapter 4. Linear and Planar Rigidity 93

4.1. Abstract Rigidity in the Plane 93
4.2. Combinatorial Characterizations of {72(n) 96
4.3. Cycles in Q2(n) 98
4.4. Rigid Components of Q2(G) 100
4.5. Representability of Q2 (n) 103
vii
viii CONTENTS
4.6. Characterizations of A2 and (A2).L 104

4. 7. Rigidity and Connectivity 109
4.8. 'frees and 2-dimensional Isostatic Sets 113
4.9. 'free Decomposition Theorems 118
4.10. Computational Aspects 123
Chapter 5. Rigidity in Higher Dimensions 129

5.1. Introduction 129
5.2. Higher Dimensional Examples 131
5.3. The Henneberg Conjecture 133
5.4. Stresses and Strains 138
5.5. 2-Extensions in 3-Space 143
5.6. The Dress Conjecture 147
5.7. Other Conjectures 149
References 153
Index 171
Preface
A framework in m-space is a triple {V,E,p), where (V,E) is a finite graph and

p is an embedding of V into real m-space. A framework is a mathematical
model for a physical structure in which each vertex v corresponds to an idealized
ball joint located at p(v), and each edge corresponds to a rigid rod connecting
the joints corresponding to its endpoints. Obviously this concrete realization
is meaningful only for m ~ 3, and may be used to describe a very general
class of physical structures, including rigid ones such as pedestals or bridges,
as well as moving structures such machines or organic molecules. Making the
distinction between frameworks whose realization is rigid and those which can
move is the fundamental problem of rigidity theory, which can also be considered
for frameworks in higher dimensions. In low dimensions one could construct an
appropriate realization of a given framework and test the model for rigidity.
Of course, the mathematical task is to develop a method for predicting rigidity
without building a model.
One would expect that whether a framework is rigid or not depends on both
the graph (V, E) and the embedding p; or, in more general terms, that the
question of rigidity has both combinatorial and geometric aspects. Our primary
interest is in the combinatorial part of rigidity theory, which we call combinatorial
rigidity. However, the two parts of rigidity theory are not so easily separated. In
fact, only in dimensions one and two has total separation of the two parts been
achieved.
In the first chapter we will give an overview of the subject, developing both
aspects of the theory of rigidity informally in an historical context. This chapter
stands apart from the rest of the book in that it contains no formal proofs. Most
of the concepts introduced here will be reintroduced in a more formal setting
later on.
The second chapter is devoted to a study of infinitesimal rigidity, a linear ap-
proximation which stands at the boundary of the combinatorial and geometric
nature of rigidity. The infinitesimal approach offers at least a partial separa-
tion of the combinatorial and geometric aspects by regarding the matrix of the
derivative of a framework motion as a matroid on the edges of the framework.
ix
X PREFACE
In general, depending on the dimension and the embedding, the edges of a graph
are the underlying set of several such matroids, all of which belong to the class
of abstract rigidity matroids, which are defined at end of chapter 2.
The fundamental combinatorial structures used to study rigidity are the var-
ious rigidity matroids. The second chapter consists of a development of matroid
theory, the theoretical foundation for much of modern combinatorics. There will
of course be a special emphasis on those parts of the subject most applicable to
rigidity matroids.
Chapter 4 is devoted to an extensive study of combinatorial rigidity dimen-
sion 2, which has a. nice analogy, via the !-dimensional case, with "traditional"
graph theory from a slightly different point of view. A thorough knowledge of
planar rigidity is essential to developing a good intuition for rigidity as a whole,
and provides an extensive collection of tractable examples. Algorithmic and
computational aspects are also treated.
In the last chapter, we will discuss combinatorial rigidity in higher dimensions.
Special attention is paid to dimension 3, in which there is the most practical
interest, but where the characterization problem is still unsolved. Many of the
results in this chapter have not yet appeared elsewhere.
The book concludes with an extensive annotated bibliography.
This text is suitable for a second graduate course in combinatorics and was
already used as such at Syracuse University and at Worcester Polytechnic Insti-
tute by the authors. Each chapter contains a variety of exercises, some letting
the reader fill in the details of the theory, some working through examples, as
well as many which point the way to aspects of rigidity theory not covered in the
text. Exercises are placed so that the reader can check his understanding of each
concept before going on to the next one. The annotations in the bibliography are
not only a valuable research tool, but also meant to stimulate a project oriented
course of study.
Each of the chapters is mathematically self-contained, and the reader may
safely peruse them in the order best suited to his background and interest.
Many thanks to Ray Adams and John Shutt for diligent proof reading and
thoughtful questions.
Chapter 1. Overview
1.1. An Intuitive Introduction to Rigidity. We will use the term graph

to denote a finite, undirected graph without loops or multiple edges. We will
use the notation (V, E) for the graph with vertex set V and edge set E, where
E is a collection of unordered pairs of vertices. We will use the term framework
(in m-space) to denote a triple (V,E,p), where (V,E) is a graph and pis an
embedding (injection) of V into realm-space.
Let (V, E, p) be a framework in m-space and assume that V contains n ver-
tices. One natural approach to the question "is the framework rigid?" is to
coordinatize m-space and write down the distance equations for the pairs of
points in P = p(V) which correspond to edges in E with the coordinates of the
points replaced by mn distinct variables. The coordinates of the given points
give one solution to this system of quadratic equations. Do other solutions exist?
The answer clearly is yes: consider any congruent framework in m-space, the co-
ordinates of the points of this framework form another solution to this system.
Do all solutions arise from congruent frameworks? The answer to this question
will depend on the particular framework under consideration. In the case of the
triangle in the plane, the Side-Side-Side Theorem of plane geometry tells us that
there are no further solutions to that system. On the other hand, the existence
of parallelograms with sides of appropriate lengths shows that there are many
"non-congruent" solutions to the system of equations of a rectangle. Clearly
then, the condition that all solutions to the system of a framework correspond
to congruent frameworks implies that the framework is "rigid". But, before we
formalize this into a definition for rigidity, let us consider another example.
Take a rectangle in the plane and insert a diagonal (see Figure l.la). In one
sense this framework is rigid: it cannot be "continuously" deformed in the plane.
However, the framework obtained by flipping one of the triangles over the diago-
nal (see Figure l.lb) is not congruent to the original framework but is a solution
to the system of quadratic equations. To reflect this distinction, we introduce
two different definitions for rigidity. We say that a framework is strongly rigid (in
m-space) if all solutions to the corresponding system of quadratic equations cor-
respond to congruent frameworks; we say that a framework is rigid (in m-space)
1
2 CHAPTER 1. OVERVIEW
{b)
FIGURE 1.1. A flip.

if all solutions to the corresponding system in some neighborhood of the original
solution (as a point in mn-space) come from congruent frameworks. Clearly,
strongly rigid implies rigid. Among our examples, the triangle is strongly rigid
and hence rigid, the rectangle with one diagonal is rigid but not strongly rigid,
while the rectangle without diagonals is neither strongly rigid nor rigid.
Let us take a closer look at our first approach to rigidity. We start by assigning
an order to the vertices in V; the easiest way to do this is to identify V with
{1, 2, ... , n}. We are then given an ordered set, P = p(V), of n points in real
in-space and a set E of the pairs of these points whose distances are to remain
fixed. If we coordinatize m-space, we may (using the order on P) identify P
with a point in real mn-space. The set E of pairs of points then yields a system
of lEI quadratic equations. This system determines an algebraic set A in mn-
space called the configuration space of P. Clearly PEA and we may describe
a physical movement of the framework in space by a path in A starting at P.
The algebraic sets described by such "linkages" are the topic of the Scientific
American article ''The mathematics of three-dimensional manifolds", by W. P.
Thurston and J. R. Weeks (129]. Readers who feel that their geometric intuition
is lacking in this area should find this paper very helpful. If we replace the system
of equations from E by the system from K (where K denotes the collection of
all pairs of points from V), we get a second algebraic set which we will denote
by C. Clearly, Cis a subset of A and it contains the point P. We say that two
frameworks {V,E,p) and {V,E,q) are ''near" one another if P and Q = q{V)
are "near" one another as points in mn-space (see Figure 1.2a). If Q is on A
~---: (a)
~ {b) (c)
:
FIGURE 1.2. A framework and some nearby frameworks.
then the corresponding rods in these two frameworks have the same length (see
Figure 1.2b). If Q lies on C, the two figures are congruent (see Figure 1.2c). Thus,
we have defined {V, E, p) to be rigid if A and C are equal in some neighborhood of
P and (V, E, p) to be strongly rigid if A = C. If Q lies on A-C and can be joined
to P by a curve on A which intersects C only in P, then we say that {V, E, q) is
a flex of (V, E, p); one may identify the intermediate points on this curve with a
deformation of the framework {V,E,p) into the framework {V,E,q).
1.1. AN INTUITIVE INTRODUCTION TO RIGIDITY 3
In order to avoid uninteresting special cases, we will assume that P contains

a subset S consisting of m + 1 points in general position, i.e. S is the set of
vertices of a non-degenerate m-simplex in m-space. Under this assumption, it
is not difficult to show that C has dimension m(m + 1}/2. A heuristic proof of
this fact counts the number of "independent" choices one makes in moving from
one point on C to a nearby point on Cor, equivalently, in moving the ''physical"
framework in m-space to a nearby position:
Order the points of Q and note that them coordinates of the new
position of the first point may be chosen independently. Once
the first point is fixed, the second point is constrained to move
along the (m- 1}-sphere with the first point as center and the
length of the rod between the first and second points as radius.
Hence we have only (m - 1) independent choices in locating the
second point. Similarly, there are (m-i+1} independent choices
in locating the i'th point, for i = 1, ... m. The last point could
occupy either of two positions, one being the reflection of the
other through the hyperplane containing the first m points of
Q. However, if we restrict the new position to a small enough
neighborhood of the initial position, one of these choices is ruled
out. Now, if we add up the total number of independent choices
one has in moving the structure (counting no choices for the last
point}, we get m(m + 1}/2.
Thus, a framework (V,E,p) will be rigid if the corresponding algebraic set
A has dimension m( m + 1) /2 in a neighborhood of P = p(V). Since each
constraint equation "usually" reduces the dimension of the solution variety by
one, we should expect that a framework with mn- (m(m + 1}/2} "properly
placed" edges would be rigid. For the plane (m = 2} this formula becomes
2n- 3. The triangle, which is rigid, has 3 = 2{3} - 3 rods as expected. The
formula indicates a need for 2( 4}- 3 or 5 rods for a framework with four vertices.
The rectangle has only 4 rods and is not rigid while the rectangle with diagonal
has 5 rods and is rigid. This concept of ''properly placed" rods must be discussed
further; as well as the question: can we make a framework rigid using fewer than
the expected number of rods?
In Figure 1.3a below, we illustrate the fact that one can indeed induce rigidity
wv (a) (b)
~[] (c) (d)
FIGURE 1.3. Rigid and non-rigid frameworks in 2-space.

with fewer than the prescribed number of rods. All four planar frameworks in
Figure 1.3 have six vertices; hence we should expect that 2(6) -3 or 9 rods would
be required to make these frameworks rigid. However, 1.3a is rigid with only 8
rods and framework 1.3c with 9 rods is not rigid. Both of these frameworks
are singular in that there exist arbitrary close frameworks with the same graph
structures, (frameworks 1.3b and 1.3d respectively), which exhibit the predicted
rigidity properties. The framework 1.3a is rigid because the "chain" of three rods
is pulled taught. If the slightest slack is permitted, as in framework 1.3b, the
framework is no longer rigid. Note that two rods cross in framework 1.3c. This
crossing has no significance and adds no constraint to the framework; we think
of the rods as being able to slide across one another. The framework 1.3c is not
rigid because the three vertical rods are equal in length and parallel and hence
permit a horizontal shear. If the uniform length or the parallelism is destroyed,
as in framework 1.3d, the framework becomes rigid.
We say that a given framework (V, E, p) is generic if all frameworks corre-
sponding to points in a neighborhood of P = p(V) in Rnm are rigid or not rigid
as is (V, E, p). A set of points P in m-space is said to be generic if each frame-
work (V, E, p) with p(V) = P is generic. The fundamental facts that we will
prove are: first, almost all sets of points in m-space are generic; and second, if
(V,E,p) and (V,E,q) are two generic frameworks with the same graph (V,E),
then either they are both rigid or they are both non:rigid. Assuming this last
result, we define a graph (V, E) to be rigid (in dimension m) if the frameworks
corresponding to the generic embeddings of V into m-space are rigid. We say
that a framework (V, E, p) in m-space is generically rigid (in dimension m) if the
graph (V, E) is rigid in dimension m. The four frameworks in Figure 1.3 show
that rigidity and generic rigidity are quite distinct concepts: framework 1.3a is
rigid but not generically rigid; framework 1.3b is neither rigid nor generically
rigid; framework 1.3c is not rigid but it is generically rigid; and framework 1.3d
is both rigid and generically rigid.
While it is clear that generic rigidity is a combinatorial concept for all dimen-
sions, it is only for dimensions one and two that this fact has been used exten-
sively. The reason is that a combinatorial characterization for generic rigidity has
been found only for these dimensions. In 1-space, all embeddings are generic and
the concepts of rigidity and generic rigidity coincide with one another and with
the concept of connectedness for graphs. In dimension 2, we have the following
characterization of generic rigidity:
THEOREM (G. LAMAN, 1970, [64]). A graph (V,E) is rigid/or dimension 2
if and only if there is a subset E' of E such that:
1. IE' I = 2IVI- 3;
2. IFI ~ 2jV(F)I-3 for all non-empty subsets F of E' {where V(F) denotes
the support ofF, i.e. the collection of all end points of the edges in F).
One may interpret condition 1 as requiring that E contain enough edges to

be rigid and condition 2 as requiring that none of these edges be "wasted" by

packing too many between the vertices of any subset of V.
We turn now to yet another concept of rigidity. Let a framework (V, E, p) be
given in m-space and consider an assignment of an m dimensional vector u(i) to
each vertex i in. V; we actually think of u(i) as a velocity vector for the point
p(i) in m-space. Such an assignment u is said to be an infinitesimal motion
of the framework (V,E,p) if, for every edge (i,j) of (V,E) the projections of
u(i) and u(j) onto the line through p(i) and p(j) have the same lengths and
directions. One may interpret this condition as follows: the vector assignment
u is the set of initial velocities of a motion for the points in P = p(V) which
neither compress nor stretch the "rods" of the framework.
The condition that a vector assignment u is an infinitesimal motion may be
restated as a system of homogeneous equations:
(u(i)- u(j)) * (p(i)- p(j)) = 0, for all (i,j) EE .
Hence it is clear that V, the set of infinitesimal motions of the framework

(V, E, p), is a subspace of the nm-dimensional vector space of all functions from
V into Rm, where n = lVI· The role of V here is analogous to that of the alge-
braic set A in our initial approach to this problem. As in that earlier approach,
replacing E by the set of all pairs K yields a second subspace which we will
denote by 'D, see Figure 1.4. Clearly, 'D is a subspace of V. It is not difficult
FIGURE 1.4. The framework spaces.

to show that the vector assignments in 'D correspond to the initial velocities of
the points in P under the direct isometries or rigid motions of m-space. We say
that a framework (V, E, p) is infinitesimally rigid if the two ~ubspaces V and 'D
are equal. If V does not equal 'D, we say that the framework (V, E, p) is not in-
finitesimally rigid and we call an infinitesimal motion u not in 'D an infinitesimal
flex of the framework.
The two concepts, rigidity and infinitesimal rigidity, are closely related for a
framework (V,E,p). Intuitively, one may think of the space Vas the tangent
space to A at the point P = p(V) and 'D as the tangent space to C. Now
suppose that (V,E,p) is non-rigid. Since C has a lower dimension than A in a
neighborhood of P, one can prove the existence of a curve on the algebraic set A
starting at P and moving off the subvariety C with the property that its partial
derivatives of sufficiently high order exist and are non-trivial. These partial
derivatives can then be seen to be an infinitesimal flex of the framework. Thus,
if (V, E, p) is not rigid, it is not infinitesimally rigid; or equivalently, infinitesimal
rigidity implies rigidity. However, the converse is not true. To see this, consider
framework 1.3a. Assign the zero vector to all points except to one of the two
points on the "chain"; to this point assign any non-zero vector perpendicular to
the chain. One easily sees that this is an infinitesimal flex of that framework.
Thus, while the framework 1.3a is rigid, it is not infinitesimally rigid. The
relationship between infinitesimal rigidity and generic rigidity is not so easy
to see. However, we will prove later that infinitesimal rigidity implies generic
rigidity. In fact, if a set of points P in IR.m is generic, then the concepts of rigidity,
generic rigidity and infinitesimal rigidity agree for all frameworks (V, E, p) with
p(V) = P. Strongly rigid remains a concept apart even on generic sets. In
summary:
• strong rigidity => rigidity;
• infinitesimal rigidity => rigidity;
• infinitesimal rigidity => generic rigidity.
That these are the only implications among these concepts is demonstrated by
the list of examples in table 1.1.
Example Rigid Strongly Infinitesimally Generically
Rigid Rigid Rigid
Framework 1.3b No No No No
Framework 1.3c No No No Yes
Framework 1.5a Yes No No No
Framework 1.5b Yes No No Yes
Framework l.la Yes No Yes Yes
Framework 1.5d Yes Yes No No
Framework 1.5c Yes Yes No Yes
Tetrahedron Yes Yes Yes Yes
TABLE 1.1.
(a) (b) (c) (d)
FIGURE 1.5. Some singular frameworks.

As illustrated in Figure 1.4, we may regard infinitesimal rigidity, that is com-
paring V to "D, as a first order approximation to rigidity, which compares A to
C. One approach to answer the question "Is (V, E, p) rigid?" is to pursue better
approximations by considering higher order terms, see [23]. Define a 2 'nd order
flex to be an assignment of two vectors 11i and at to each each vertex i such that
the equations
(at- a;)* (Pi-P;)+ (ui- u;) 2 = 0,
are satisfied for all {i,j) E E. These equations would have to be satisfied by
the velocities {the ui's) and accelerations {the at's) of a physical motion of the
framework. Let V2 be the set of 2'nd order flexes of (V, E, p) and let 'D2 be
the set of 2'nd order flexes of (V,K,p). The framework (V,E,p) is said to be
2'nd order rigid if V2 = 'D2 in a neighborhood of 0. While 2'nd order rigidity
has been successfully developed, there exist highly singular frameworks which
make extension to higher orders difficult, [26]. In any case, higher order rigidity
belongs to the geometric rather than the combinatorial nature of rigidity, and is
not considered in this text.
One extension of "rod and joint" rigidity which has received a lot of atten-
tion is the study of tension frameworks. Here one admits cables in addition to
rods. Upon a closer look, one observes that a cable is equivalent to a chain
of rods, which is modeled by a path of edges. Hence, one could study tension
frameworks in the "framework" of rigidity. However, this is not a very practical
way to approach tension framework problems. Many special techniques have
been developed for considering these tension frameworks and a good bit of the
literature is devoted to these frameworks. While tension frameworks will not
be discussed in this book, references to the literature on tension (or tensegrity)
frameworks are included in the bibliography.
Another extension of "rod and joint" frameworks, in higher dimensions, is
obtained by including other rigid building blocks such as ''polygonal plates".
In 3-space, one may consider three dimensional structures consisting of rigid
polygons joined by hinges along common edges. Without going through the
process of developing the formal definitions for the various types of rigidity, one
can, with a few examples, develop usable intuitive definitions for these concepts.
Consider the structure consisting of four congruent triangles and a square hinged
together to form a pyramid. This structure is rigid, in fact, strongly rigid. If
the square is removed the structure is no longer rigid. It is perhaps surprising
that some of the earliest results in rigidity theory dealt with these "plate and
hinge" structures. In fact, the first major result published in rigidity theory is a
theorem of Cauchy about plate and hinge structures in 3-space. This and other
early results will be discussed in the next section.
Exercise 1.1. Consider the !-dimensional framework consisting of two ver-

tices joined by an edge: specifically (V, E, p) where V = {1,2}, E = {{1,2)},
p{l) = 0 and p{2) = 1.
a. Show that the algebraic set A of this framework is the union of the two
lines x2 = x1 + 1 and x2 = x 1 - 1 in IR2; locate the point on A which
corresponds to the framework (V, E, p).
b. Show that the space V of infinitesimal motions of this framework is the
!-dimensional subspace given by x2 = x1.
Exercise 1.2. Consider the !-dimensional framework: (V, E, p) where V =
{1,2,3}, E = {(1,2), (2,3)}, p(1) = 0, p(2) = 1 and p(3) = 3.
a. Describe the algebraic set A in IR.3 for this framework and locate the point
on A which corresponds to the framework (V, E, p).
b. Describe the algebraic set C.
c. Is this framework strongly rigid?
d. Describe the space V for this framework.
Exercise 1.3. Consider the framework described in Exercise 1.1 as a frame-
work in 2-space specifically (V,E,p) where V = {1,2}, E = {(1,2)}, p(1) =
(0, 1) and p(2) = (1, 0). For an arbitrary embedding q of V into IR2 , let q(l) =
(x1,y1) andq{2) = (x2,y2).
a. Give the equation which defines the algebraic set A in IR4 •
b. Describe the intersection of A with the (x1,yt)-plane {x2 = Y2 =OJ.
c. Describe the intersection of A with the plane given by X2 = 0 and Y2 = a,
for some constant a.
d. Describe the intersection of A with the plane given by x1 = b and x2 = 0,
for some constant b.
e. Describe the intersection of A with the 3-space given by Y2 = 0 and locate
the point on A which corresponds to the framework (V, E, p).
Exercise 1.4. Consider the framework from Exercise 1.3.
a. Describe algebraically the space V of infinitesimal motions of this frame-
work. (Use the notation u(l) = (Xt.Yl) and u(2) = (x2,Y2) for the
mapping u of V into JR2 •
b. Describe geometrically the intersection of V with the 3-space Y2 = 0.
c. Describe geometrically the tangent plane to the intersection of A with the
3-space X2 = 0 at the point corresponding to the framework (V, E, p).
d. Describe the relationship between the answers to parts b and c.
Exercise 1.5. Let (V, E, p) be any !-dimensional framework.
a. Letting q( i) = Xi for all i in V and all embeddings q of V into IR2 , show
that A is described by the system of equations: !xi - x;l = Ci;, for all
(i,j) in E, where Ci; = IP(i) - p(j)l.
b. Letting u( i) = Xi for all i in V and all vector assignments V, show that
V is the solution space of the system of equations:
Xi- x; = 0, for all (i,j) in E.
c. Describe the relation between A and V.
d. Also describe the relation between C and 'D.
1.2. A SHORT HISTORY OF RIGIDITY 9
e. Use your answers to c. and d. to explain why rigidity and infinitesimal

rigidity are identical in dimension one.
Exercise 1.6. Consider Table 1.
a. Verify that the table is complete, i.e. that it includes all possibilities con-
sistent with the three implications between the various types of rigidity.
b. Verify all of the claims in Table 1.1.
1.2. A Short History of Rigidity. In 1766, Euler conjectured: "A closed
spatial figure allows no changes, as long as it is not ripped apart." [43). As
stated at the end of the last section, the first major result published in rigidity
theory was a theorem of Cauchy. It was a first step toward proving the Euler
Conjecture.
THEOREM (A. L. CAUCHY, 1813 [17]). If there is an isometry between the

surfaces of two strictly convex polyhedra which is an isometry on each of the
faces, then the two polyhedra are congruent.
By a strictly convex polyhedron, we mean a polyhedron which has the prop-
erty that, for each vertex, there exists a plane which intersects the polyhedron in
that vertex and no other point of the polyhedron. The proof involved a mixture
of topological arguments and arguments from elementary geometry. Unfortu-
nately, there were several minor errors in the proof; the first complete proof
was published by Steinitz and Rademacher in 1934, [109). A simpler complete
proof of a slight generalization of Cauchy's Theorem was given by Alexandrov
in 1950, [1).
It is easy to see that the convexity condition cannot be dropped from the
hypothesis of Cauchy's Theorem. Consider the pair of polyhedra constructed as
follows. Take a cube and a pyramid with base congruent to a face of the cube
and with height less than the height of the cube. Remove the base from the
pyramid and remove one face from the cube. Then join these two surfaces along
their boundaries, once with the apex of the pyramid pointing out and once with
it pointing in. There is an obvious homeomorphism between these two surfaces
which restricts to an isometry on each face; but these two polyhedra are not
congruent. The following corollary is an immediate consequence of Cauchy's
Theorem.
COROLLARY. The 2-skeleton of a strictly convex polyhedron in 3-space is

rigid.
With the Euler Conjecture in mind, it is natural to ask if "strictly convex"
may be dropped from the hypothesis of the corollary. One may think of Cauchy's
result as the first step toward a proof of Euler's Conjecture. The next step
was due to R. Bricard. In 1897, he published a complete study of the rigidity
properties of the octahedron. The main result of that study can be stated as
follows.
THEOREM (R. BRICARD, 1897, (15]). The 2-skeleton of any polyhedral em-
bedding of the octahedron in 3-space is rigid. However, the !-skeleton of the
octahedron has a non-rigid embedding in 3-space.
At first glance the two parts of this theorem seem to be in contradiction to

one another: since the faces of the octahedron are all triangular, deleting them
could not change a rigid plate and hinge framework into a non-rigid rod and
joint framework. It would then seem to follow from the first part of the theorem
that any embedding of the 1-skeleton of the octahedron would also be rigid,
however, not every embedding of the 1-skeleton extends to an embedding of
the 2-skeleton. The non-rigid embeddings of the 1-skeleton of the octahedron
are easy to describe. Consider a non-regular pyramid with a square base and
assume that the perpendicular projection of the apex onto the base does not lie
any line of symmetry of the square. Now construct a second pyramid on the
same base which is isometric to the first under the 180 degree rotation about
the line perpendicular to the base at its center. Note that both apices are on
the same side of their common base. The reason that this octahedral1-skeleton
is non-rigid is that both pyramids produce the same family of distortions of the
rectangular base as they flex.
To understand the relationship between the two parts of Bricard's Theorem,
consider the 1-skeleton of the octahedron embedded as just described. Since
the faces of the octahedron are triangles, adding the faces to this 1-skeleton can
be done without destroying the planarity of those faces and without limiting
any motions of the structure. We thus have a copy of the 2-skeleton of the
octahedron in 3-space which is not rigid! However, it is clearly not embedded:
it has self-intersections.
The next new result on the Euler Conjecture did not occur until many years
later when Alexandrov gave his generalization of Cauchy's Theorem.
THEOREM (A. D. ALEXANDROV, 1950 (1]). If vertices are inserted in the

edges of a strictly convex polyhedron and the faces are triangulated, then the
1-skeleton of the resulting polyhedron is infinitesimally rigid.
This result has the following corollary:
COROLLARY 1.2.1. If a convex polyhedron in 3-space has the property that the
collection of faces containing a given vertex do not all lie in the same plane, then
the 2-skeleton of that polyhedron is infinitesimally rigid.
The Euler Conjecture was finally settled in 1977, but just prior to that a
result was published which greatly extended the set of configurations for which
the Euler Conjecture was known to be valid.
THEOREM (H. GLUCK, 1975, [45]). Every closed simply connected polyhe-
dral surface embedded in 3-space is generically rigid.
Gluck's Theorem tells us that the Euler Conjecture is almost always true for
closed simply connected polyhedral surfaces. Just two years later, a counterex-
ample to the Euler Conjecture was found by R. Connelly [20]. The counterex-
ample is based on Bricard 's flexible octahedron and, of course, it is non-convex
and non-generic. Topologically Connelly's surface is a sphere, thus showing that
Gluck's Theorem is best possible.
In 1978, Asimow and Roth showed that the triangulation condition in the
hypothesis of Alexandrov's Theorem is necessary.
THEOREM (L. ASIMOW AND B. ROTH, 1978 [2]). The 1-skeleton of a strict-
ly convex polyhedron embedded in 3-space which has at least one non-triangular
face is not rigid.
The proof of this theorem has two parts. The first part is combinatorial. Using
Euler's formula for the sphere, Asimow and Roth show that the number of edges
in such a 1-skeleton is less than 3v - 6, where v is the number of vertices of the
polygon. (A triangulation of a sphere has exactly 3v-6 edges.) By the dimension
arguments given above, in conjunction with our discussion of Laman's Theorem,
it is clear that there are not enough constraints to make this framework rigid
unless it represents a singular point on the algebraic variety A. In the second,
geometric, part of their proof, Asimow and Roth show that a strictly convex
framework cannot correspond to a singular point. To see that strict convexity
is essential to their result, consider a tetrahedron with equilateral faces. Insert,
centered in one face, a smaller equilateral triangle with sides parallel to the sides
of the face. Now, join by edges the corresponding vertices (see Figure 1.6). This
figure has four non-triangular faces. Nevertheless it is easily seen to be rigid as
the small triangle is held in place by tension.
FIGURE 1.6.
'Thming to generic rigidity, we have already noted that Laman characterized

generic rigidity for dimension two in 1970. It is natural to ask if that resu\t
could be extended to higher dimensions. Unfortunately, the answer is no. We
next demonstrate the non-extendability of this result to three space. Simply
translating the statement of Laman's theorem to 3-space yields:
A graph (V, E) is rigid for dimension 3 if and only if there is a subset E' of E
such that:
1. IE'I = 3IVI - 6;
2. IFI :5 3IV(F)I- 6 for all non-empty subsets F of E', where V(F) is the
set of endpoints of the edges in F.
Again, we interpret condition 1 as requiring enough edges to be rigid and
condition 2 as preventing any of these edges from being wasted by packing too
many of them between a given set of vertices. The framework in Figure 1. 7
is the simplest counter example to this conjecture. It consists of 4 tetrahedra:
FIGURE 1. 7. The double banana.

they have been joined in pairs along common triangular faces, and the two pairs
are joined at the antipodal vertices. One can easily check that it satisfies both
conditions 1 and 2 above where E' is taken equal to E. Yet this framework is
not rigid; observe that the two halves may rotate independently about the line
containing the two points that they have in common. Let e denote the edge
(not in E) which joins the two vertices that the two halves have in common.
We say that e is an "implied" edge of the framework since the distance between
the images of its end points is fixed by the framework. In fact, e is an implied
edge of both the right and left hand parts of this framework. So we have, in
3-space, another way of ''wasting" edges: two different sets of edges "imply" the
same edge. Several attempts have been made to extend Laman's Theorem to
3-space by classifying all of the new ways to "waste" edges. To date, none of
these attempts has been successful.
Another thread in the history of the development of rigidity theory starts with
the work of J. C. Maxwell (75] and (76]. Recall the discussion above concerning
the vector space V of the infinitesimal motions of a framework (V, E, p). If we
define L to be the linear transformation of mn-space (specifically, the space of
all functions u: v -am) to lEI-space (specifically, the space of all functions
u: V- R) where the coordinate of L(u) corresponding to the edge (i,j) is
(L(u))(i,j) = (u(i)- u(j)) * (p(i)- p(j)),

then V is the kernel of L. Since our basic interest in V is its dimension, we
need only compute the dimension of the image of L in lEI-space. The heart
of Maxwell's work is the characterization of the orthogonal complement of the
image of L as the "space of stresses" of the framework (V, E, p).
By a stress on a framework (V,E,p), we mean an assignments ERE (the
set of functions from E into the reals) such that for each i E V the sum
Es(i,j)(p(i)- p(j)), taken over all vertices j adjacent to i in (V,E), is the

zero vector. It is not difficult to see that the collection of all stresses of (V, E, p)
is a subspace of lEI-space. We call this the stress space of the framework and
denote it by S. Maxwell observed that Sis the orthogonal complement of the
image of L and used this fact to prove the following theorem.
THEOREM (J. C. MAXWELL, 1864, [75]}. Let (V, E, p) be a framework in
m-space. Then
dim(S) ~ lEI- m!VI + 1) (m;
with equality if and only if the framework is infinitesimally rigid.
A second inequality:
dim(S) ~ 0,
is obvious. Equality in this second case is equivalent to the condition that the
set of edges be independent, that is the deletion of any edge from the framework
results in a framework with a larger set of infinitesimal motions, or, in other
words, none of the edges in E is an implied edge in the framework obtained
by deleting it. This concept of independence will play a central role in our
development of infinitesimal rigidity. Through these two inequalities, the study
of the infinitesimal rigidity and independence properties of a framework can be
reduced to a study of the dimension of the space S. Maxwell's work concentrated
on the space S. He introduced the concept of "reciprocal figures" which is
closely related to duality in graph theory. Using this tool he proved the following
interesting result.
THEOREM (J. C. MAXWELL, 1864, [75]}. If a planar framework is the pro-
jection of the !-skeleton of a polyhedron in 3-space, then it admits a non-trivial
stress.
Maxwell continued the study of reciprocal figures in [76] and his concept of
stresses remains a major tool of research in rigidity theory.
Closely related to rigidity theory is the theory of planar and spatial linkages.
By a planar (spatial) linkage, we shall mean a rod and (ball) joint framework in
the plane (3-space) with some vertices held fixed. The interest is not in finding
linkages which are rigid but rather in finding linkages which will trace specific
curves (curves or surfaces). For example, the linkage consisting of a single rod
pinned at one end can be used to trace a circle in the plane (sphere in 3-space).
By far, the most famous problem in the theory of linkages was the problem
of finding a planar linkage that would trace a straight line segment. In 1784,
James Watt found a linkage which gave a good approximation to straight line
motion. This approximation was sufficiently good that the linkage could be used
in the construction of engines where one wished to constrain reciprocal motion
to a straight line. The first exact solution was found by another engineer, M.
Peaucellier, in 1864. His linkage is illustrated in Figure 1.8b.
p'
Peaucellier's Cell Peaucellier's Apparatus

FIGURE 1.8.
In part a of that figure, we introduce Peaucellier's Cell. In both parts of the
figure, the pinned joints are indicated by a square. Peaucellier's cell consists of
six rods of two different lengths linked to one another as illustrated in l.Ba. The
joint where the two longer rods meet is pinned. If the two longer rods are spread
as far apart as is possible, the points p and p' coincide. If in this configuration
the cell is rotated about the pinned vertex, the point p = p' traces a circle C.
The critical feature of Peaucellier's cell is that the two vertices p and p are
inversions of one another through the circle C. Thus, if as in Figure l.Bb, pis
constrained to move in a circle through the center of C, then p' is constrained
to move in a straight line.
The early history of the theory of linkages is beautifully described in a pub-
lished lecture "How to draw a straight line" by A. B. Kempe, [63] which was
originally given to a group of science teachers in the summer of 1876. Prior to the
presentation of this lecture, Kempe had shown that any algebraic curve could
be traced by a suitable linkage [62]. Kempe's lecture only briefly touches on
linkages to trace curves and surfaces in space. A better reference for these prob-
lems is Chapter V, Kinematics, in Geometry and the Imagination by Hilbert and
Cohn-Vossen [54]; one may also consult the aforementioned Scientific American
Article [129].
As one can see from this short history, the results in rigidity theory have been
rather sparse during the period of time starting with Euler's Conjecture in 1766
and ending with its resolution, by Connelly, in 1977. However, since the mid
nineteen-seventies, a great deal of research in rigidity theory has been published.
The portion of those works of direct concern to generic rigidity will be discussed
in the remainder of this book. All of them will be described in the annotated
bibliography at the end of the book.
Exercise 1.7. Consider Peaucellier's Cell (Figure l.Ba} in the plane with the
pinned vertex at o, the origin, and let C be the circle with center at o and radius
a, equal to the distance from o to p when p and rl coincide.
a. Prove that p and p' are inverses with respect to C, that is o, p and p'
are collinear and d(o,p) x d(o,p') = a2 •
b. Using part a, prove that, if p is constrained to move in a circle through
o, then p' will move along a straight line.
Chapter 2. Infinitesimal Rigidity
2.1. Basic Definitions. Instead of starting our formal development of rigid-

ity with one of the definitions given in Section 1.1, we will start with the con-
struction of a rather elaborate mathematical foundation for rigidity. Not until
Section 2.5 will we return to those original definitions. However, one of the fun-
damental questions raised in Section 1.1, namely ''when is a framework generic?",
is dealt with here.
As one studies the infinitesimal rigidity structure of a framework (V, E, p),
one soon discovers that this study also involves the infinitesimal rigidity of other
frameworks obtained from (V, E, p) by adding or deleting edges or vertices to
the graph (V, E) and adjusting p. Hence we will often fix one underlying set
V and one injection p of V into Rm and then consider all frameworks of the
form (U, E, PlU) where (U, E) is a subgraph of (V, K), the complete graph on
the vertex set V, and PIU is the restriction of p to U. With this motivating
factor in mind, we will proceed to develop the basic notation that we will use
throughout this text.
Let V be the fixed n-element set {1, 2, ... , n} and let K be the collection
of all unordered pairs of elements of V. Then (V, K) is a complete graph. We
will use the notations e;.; and (i,j) interchangeably to denote the edge of (V, K)
with vertices i and j as endpoints where i < j. In a few instances, it will be
convenient to use e;.; or (i, j), where j < i, to represent the edge (j, i). This will
usually happen when i is fixed and we sum or collect over all j =F i, i.e. over
all edges with i as an endpoint. These sums or collections would be awkward to
describe if we insisted on the convention that (i,j) denotes an edge only when
i < j.
There are three operators on the subsets of V and K which will be of use to
us. For any edge set E ~ K,
V(E) = {i 13 j E V with (i,j) E E or (j,i) E E}.
The set V(E) is called the support of E. For any vertex set U ~ V,
K(U) = {(i,j) I i < j and {i,j} ~ U};
17
18 CHAPTER 2. INFINITESIMAL RIGIDITY
for any fixed edge set E ~ K and any vertex set U ~ V,
E(U} =En K(U).

In this notation (U, K(U}} is the complete subgraph of (V, K) on the vertex
set U; similarly (U, E(U}} is the subgraph of (V, E) induced by the vertex set
U; and (V(E}, E) is the subgraph of (V, K) induced by the edge set E. These
operators have several useful properties which are listed in Lemma 2.1.1 below.
Their proofs involve only straight forward arguments from elementary set theory
and are left as an exercise for the reader.
LEMMA 2.1.1. Let (V, K) denote the complete graph and let E, F ~ K.
a. IfF ~ E, then V(F} ~ V(E).
b. V(F U E) = V(F) U V(E).
c. V(F n E) ~ V(F) n V(E).
Exercise 2.1. Consider the operators K(·}, E(-) and V(·).

a. Prove Lemma 2.1.1.
b. Show by an ezample that equality need not hold in Part c of Lemma 2.1.1.
c. State and prove the results for the operators K(·) and E(-) which are
parallel to the above results for V (·).
Let S be any set; an operator (-} mapping the power set of S onto the power
set of S is called a closure operator on S if the following three conditions are
satisfied:
Cl: IfT ~ S, then T ~ {T};
C2: If R ~ T ~ S, then {R} ~ {T};
C3: If T ~ S, then {{T}} = {T).
Exercise 2.2. Let (V, K) denote the complete graph.
a. Show that{·}, defined by {E)= K(V(E}}, is a closure operator on K;
b. Show that (·}, defined by
(E)= {(i,j)li < j and i is joined to j by a path in (V, E)},

is a closure operator on K.
The closure operator described in Exercise 2.2b is called the connectivity clo-
sure operator (on K). Given a closure operator (·} on the setS, we may define
the closed sets (with respect to (·}) as the subsets of S of the form (T} for some
subset T of S.
Exercise 2.3. Describe the closed sets for the two operators defined in Exer-
ciSe 2.2.
Exercise 2.4. Let S be a finite set. Let (·} be a closure operator on S and
let C be the collection of closed sets with respect to (·}.
a. Show that
Cl': SEC,
2.1. BASIC DEFINITIONS 19
C2': C is "closed" under intersection.

b. Show that, for any T ~ S, (T) is the intersection of all the closed sets
containing T, i.e.,
(T) = n
ReC,T~R
R.
Exercise 2.5. LetS be a finite set and let C be any collection of subsets of
S satisfying:
Cl': SEC,
C2': C is "closed" under intersection.
Show that the operator (-) of Exercise 2.4 is a closure operator on S with C as
its closed sets.
A clo.~ure operator ( ·) on a set S is a matroid closure operator if it satisfies
the additional condition:
C4: If T ~Sand s,t E (S- (T)}, then s E (TU {t}) if and only if
t E (TU {s}).
Exercise 2.6. Recall Exercise 2.2.

a. Show that the closure operator in Exercise 2.2a is not a matriod closure
operator when n is greater than 3.
b. Prove that the connectivity closure operator {described in Exercise 2.2b}
is a matroid closure operator.
Exercise 2.7. Let C be a collection of subsets of S.
a. Show that if C is the collection of closed sets of a matroid then C satisfies:
Cl': SEC
C2': C is "closed" under intersection
C3': For any C E C the union of the minimal sets in {C' E C I
C c C'} isS.
b. Suppose C satisfies the above three conditions. Show that C is the collec-
tion of closed sets of a matroid.
The prototype matroid closure operator is the spanning set operator on a set
of vectors: Let S be a finite set of vectors in a vector space; for any subset T of
S, let (T) = span(T) n S, where span(T) denotes the subspace spanned by T.
The first three conditions for a closure operator follow from the fact that (·) is
a closure operator on the subsets of a vector space. Now, assume that vectors s
and t belong to S but not to (T). It follows that
(2.1.1} dim((T U {s})) = dim((T U {t})) = dim(span(T)) + 1.
Next assume that s E (T U {t}). It follows that T U {s} ~ (T U {t}); that

(TU {s}) ~ (TU {t}); that (TU {s}) = (TU {t}), by (2.1.1} above; and, finally,
that t E (T U {s}). By a symmetric argument, t E (T U {s}) implies that
s E (T U {t}). We have proved:
LEMMA 2.1.2. LetS be a finite set of vectors from some vector space and, for
any T ~ S let (T) = span(T) n S. Then (-) is a matroid closure operator on S.
The reader may well ask, just how matroid closure operators fit into rigidity
theory. In our earlier discussions we mentioned "implied" edges or rods: the
edge between p(i) and p(j) in the framework (V, E, p) is implied if the dis-
tance between p(i) and p(j) is unaltered by any of the (infinitesimal) motions
of (V, E, p). The operator on K which assigns to E all edges in or implied by E,
is a matroid closure operator. The framework will then be infinitesimally rigid
if and only if the closure of E is K.
Throughout this text we will use Y X, where Y and X are sets, to denote
the collection of all functions from X into Y. Usually Y will be the reals R or
the real vector space :Rm; in some cases, particularly in the exercises, the reals
may be replaced by some other field, F. We note that, for any finite set X, :Rx
(Fx) is an IXI-dimensional vector space over R (over F). We will be particularly
interested in the space of functions from V into :Rm, (Rm) v. Given p E (Rm) v,
we will use Pi to denote the vector p(i), for i = 1 ... n. This space also has the
structure of a real vector space and has dimension miVI = nm. In fact, we have
a natural isomorphism, I: (:Rm)V--+ :Rnm, given by
I(p) =(Pl. P2, .. · , Pn)·

In other words, if p maps V into :Rm, we identify p with the nm-tuple of real
numbers:
where (pit 1 Pi2•· .. 1 Pim) are the coordinates of pi. Again, the reals may replaced
by some other field in the above discussion.
Let p be an embedding of V into :Rm and let P denote set p(V). We say
that the set P is in general position and that p is a general embedding if, for any
q-element subset Q of P with q < (m + 1), the affine space spanned by Q has
dimension q - 1. That is to say: if q = 2, Q spans a line; if q = 3, Q spans a
plane, etc.
Exercise 2.8. Let p be an embedding of V into :Rm and let P = p(V). Verify
the following observations.
a. If m = 1, P is in general position.
b. If m = 2, then P is in general position if and only if no 3-subset of P
lies on a line.
c. If m = 3 and n > 3, then P is in general position if and only if no
4-subset of P lies on a plane.
As noted above we may identify an embedding p of V into :Rm with a point
in Rmn. We may also measure the length of the edges by evaluating the rigidity
function p: Rmn --+ RIE(K)I defined by p(p)ij = (PI- PJ) 2 , (where the coordi-
nates of RIE(K)I are indexed by the pairs ij in, say, lexicographical order.) So the
ij'th coordinate of p(p) is the square of the length of the edge ij inK. Clearly
p is continuously differentiable and we define R(p), the rigidity matrix for the
embedding p, by p'(p) = 2R(p). R(p) is an n(n- 1}/2 by nm matrix whose
entries are functions of the coordinates of p as a point in :R.nm. In the matrix
below we write out coordinates of p and of the rigidity matrix R(p}, in the case
n=4 andm= 2.
P = (Pl!P2,pa,P4) = (plt,Pt2,P2l!P22,P31!P32,P41!P42)i
Pn-P21 P12-P22 P21-Pll P22-P12 0 0 0 0

Pn-P31 P12-P32 0 0 P31-Pll P32-P12 0 0
Pn-P41 P12-P42 0 0 0 0 P41-Pll P42-P12
0 0 P21-P31 P22 -p32 P31-P21 P32-P22 0 0

0 0 P21-P41 P22-P42 0 0 P41-P21 P42-P22
0 0 0 0 P31-P41 P32-P42 P41-P31 P42-P32
Using the vectors Pi and P; instead of writing out the actual coordinates, we
can write R(p) in a condensed form:
Pt-P2 P2 -Pl 0 0
Pt-Pa 0 Pa -pl 0
Pt-P4 0 0 P4 -pl
R(p) =
0 P2 -pa Pa -p2 0
0 P2 -p4 0 P4 -p2
0 0 Pa -p4 P4 -pa
where 0 stands for the zero vector. This shorthand notation is ambiguous in that
R(p) looks the same for all m. Nevertheless, we will use this notation because
of its convenience.
Exercise 2.9. Write out the full matrix for five points in :R.2 and the short-
hand matrix for five points in :R.m.
Consider the following set up: the integers n and m are fixed but p is permit-
ted to vary over all mappings of V into :R.m, i.e. p is permitted to vary over all
points of :R.nm. How do we view the restriction of p to the set of embeddings? If
i and j are distinct vertices, the vector equation Pi = P; corresponds to a system
of m linear equations in the coordinates of p. The solution to this system is some
(nm - m }-dimensional hyperplane in :R.nm. Considering all pairs of vertices, we
have a collection of n(n- 1}/2 such hyperplanes and pis an embedding if and
only if P does not lie on the union of these hyperplanes. Since each of these hy-
perplanes, and hence the union of these hyperplanes, is a closed set of measure
zero, the embeddings form an open dense subset of :R.nm. (As a measure of the
topological complexity of this space of embeddings, we note that for m = 2 its
fundamental group is the unpermuted n-string braid group.)
Now consider a set of edges E ~ K such that, for some embedding p, the rows
of R(p) corresponding to E are independent. Let o(p) denote the determinant
of some lEI by lEI minor of the submatrix consisting of those rows. The solution
set to o(p) = 0 is either all of Rnm or an algebraic curve in Rnm. The set of all
embeddings p, for which these rows are independent, lie on XE, the intersection
of the algebraic curves corresponding to the lEI by lEI minors of these rows.
Since the rows corresponding toE are independent for some embedding, XE is a
closed algebraic set of measure zero. Let X be the union of the XE for all E ~ K
such that the rows corresponding to E are independent for some p. Then X is
a closed set of measure zero. We say that an embedding p is generic if p ¢ X.
We have:
LEMMA 2.1.3. Let the n-set V and the integer m be given. The set of all
generic embeddings of V into Rm is an open dense subset of Rmn.
Exercise 2.10. Prove that, if an embedding pinto R 2 is generic, then it is a
general embedding into R2 • Can you generalize your proof to higher dimensions?
One simple but very important observation concerning generic embeddings is:
LEMMA 2.1.4. If p is a generic embedding of V into Rm and U ~ V, then
the restriction of p to U, PIU, is a generic embedding of U into Rm
To see this simply note that a set of rows of R(piU) is independent if and only
if the corresponding rows of R(p) are independent.
We close this section by relating two of the fundamental ideas that we have
introduced. Consider a fixed embedding p of V. The rows of the rigidity matrix
R(p) are vectors in Rmn; and we may then consider the "prototype" matroid
closure operator on this set of vectors. Since each of these vectors is identified
with an edge of (V, K), we may identify this closure operator with a natural
matriod closure operator on K, called the rigidity (matroid} closure operator
(on K given by the embedding p): ForE~ K, an edge e belongs to the closure
of E if the row of R(p) associated with e is in the subspace of Rmn spanned by
the rows of R(p) associated with the edges in E.
The connection between this closure operator and infinitesimal rigidity will be
made clear in Section 2.2. There we prove that a framework (V(E), E, PIV(E)) is
infinitesimally rigid if and only if the closure of E, under the rigidity closure op-
erator on K given by p, is K(V(E)). However, we need not wait until Section 2.4
to verify this relationship for !-dimensional frameworks:
THEOREM 2.1.1. Let the complete graph (V, K) be given and consider any
embedding p of V into R1 . Then
a. p is generic;
b. the matroid closure operator on K given by the rows of R(p) is the
connectivity closure operator on K. (Described in Exercise 2.2b.)
PROOF. a. Let V and p be given and let (i,j) E K. We observe that
the row of R(p) corresponding to the edge (i, j) has just two nonzero entries,
(Pil - P;l) and (P;t - Pil), (these are nonzero since p is an embedding). If we
multiply through this row by 1/(pu- P;1),we will not change the determinant of
any submatrix from zero to nonzero or from nonzero to zero. Let M denote the
matrix obtained by multiplying through the row indexed by (i,j) by 1/(pil-P;l),
for all 1 :::; i < j :::; n. M is then a {0, ±1 }-matrix, in fact, it is the edge-
vertex adjacency matrix of (V, K) in which the edges are given an orientation by
the ordering on V and the sign indicates that orientation. Since corresponding
subdeterminants of M and R(p) are either both zero or both nonzero for all
embeddings p, we conclude that all embeddings p are generic.
b. Let {·) be the rigidity closure operator on K given by p and let {{ ·))
denote the connectivity closure operator on K. Let ri; denote the row of R(p)
associated with the edge (i,j); let 1 :::; i < j < k :::; nand consider the rows of
R(p) associated with the edges (i,j), (j, k) and (i, k). We have:
We conclude that if (i,j),(j,k) E E, then (i,k) E {E). Using this fact several
times, we can conclude that, if i and h are vertices in V which are joined by a path
in (V, E), then (i, h) E {E). In other words, we have shown that (i, h) E {{E))
implies (i, h) E {E).
Now suppose that (i, h) E {E). We must have:
(2.1.2) rih = L Sere, for some edge set F ~ E.

eEF
We may assume that F is as small as is possible and we will show that un-
der this assumption, F is the edge set of an elementary path from i to h.
First, we note that, F cannot contain the edges of a circuit: Suppose that
i11 (jl!h),h, (h,ja), ... , Us-l!is),js, Us.id is a circuit whose edges belong to
F. One may easily check that
L(1/(P;i,l - P;(i+l),l))r;i,;(i+l) = 0.
i=l
This relation could then be used to eliminate r; 1,;2 from the sum in (2.1.2),
contrary to the minimality of F. We conclude that F is the edge set of a forest
in (V(E), E). Suppose that j is a pendant vertex of this forest. Then j occurs
in only one term of the right hand side of (2.1.2) and the j'th component of the
right hand side is not zero. But then j = i or j = h. We conclude that F is the
edge set of a path from ito hand that (i, h) E {{E)).
Hence, {E)= {{E)) and{·) is the connectivity closure operator. 0
Combining this result with Exercise 1.5b we see that, in dimension one, the
concepts of rigidity, infinitesimal rigidity, and generic rigidity coincide.
2.2. Independence and the Stress Space. As in the previous section,

V = {1,2, ... , n}, K = {(i,j)ll ~ i < j ~ n}, and pis a fixed embedding of V
into Rm. Thus, (V, K) is the complete graph on V and, for any E ~ K, (V, E, p)
is an m-dimensional framework. As above, we identify the edges of K with the
rows of the rigidity matrix R(p). This identification enables us to define an edge
set E ~ K to be independent with respect to p if the corresponding set of rows
is independent as a set of vectors in Rmn.
We will see later that the concepts of independence and rigidity are closely
related; for the present we embark on an "independent" investigation of inde-
pendence. We start our investigation by stating a result which follows directly
from the definition of generic embedding.
THEOREM 2.2.1. Let the finite set V be given and let K = K(V). If a set of
edges, E ~ K, is independent with respect to some embedding of V, then that
set is independent with respect to all generic embeddings of V.
In particular, if E is independent with respect to one generic embedding into
m-space, then it is· independent with respect to all generic embeddings into m-
space. This leads us to define an edge set E to be generically independent for
dimension m if it is independent with respect to all generic embeddings into
m-space and to state the corollary:
COROLLARY 2.2.1. If E ~ K is independent with respect to one m-dimen-
sional embedding of V, then E is generically independent for dimension m.
We say that an edge set E ~ K which is not independent with respect to the
embedding p is dependent with respect to p. A minimal, nonempty dependent
edge set, with respect to p, is called a cycle with respect to p. Since an edge set
which is independent with respect to any embedding into m-space is independent
with respect to all generic embeddings into m-space, any edge set which is de-
pendent with respect to one generic embedding into m-space is dependent with
respect to all embeddings into m-space. We define an edge set to be generically
dependent (or a generic cycle) for dimension m if it is dependent (or a cycle)
with respect to one, and hence all, generic embeddings into m-space.
COROLLARY 2.2.2. If E ~ K is dependent with respect to one m-dimensional
generic embedding of V, then E is generically dependent for dimension m and
is, in fact, dependent with respect to all embeddings of V into m-space.
COROLLARY 2.2.3. If E ~ K is a cycle with respect to one m-dimensional
generic embedding of V, then E is a generic cycle for dimension m.
Exercise 2.11. Prove Corollary 2.2.3.
Let E be a set of edges. Considering the minors of the set of rows of R(p)
which correspond to E is not a very practical way of checking for independence.
One useful approach to independence is to check that a set of vectors satisfies
2.2. INDEPENDENCE AND THE STRESS SPACE 25
no dependency relations. An edge set E will be independent if and only if the

corresponding set of rows of R(p) satisfies no non-trivial dependency relation.
A non-trivial dependency relation for these rows has the following form:
(2.2.1) L Sijrij = 0,
(i,j)EE
where rij is the row corresponding to the edge (i,j), Sij is a scalar, and some
Sij =F 0. To simplify the notation we will define Sij to be zero for all edges
(i,j) E (K -E). Then the sum in (2.2.1) can be taken over K. We will consider
the rigidity matrix in its condensed form where the columns are indexed by the
vertices: if the vertex i is not equal to either endpoint of the edge (j, k), then
the entry in row (j, k) and column i is the zero vector; the entry in column i and
row (i,j) is Pi- Pi and the entry in column j and row (i,j) is Pi- Pi· If we
consider the sum in (2.2.1) one column at a time, we see that, for each i E V(E),
we must have:
(2.2.2) L Sij(Pi- Pj) = 0.

j;f:.i
If j is not in V(E), then Sij = 0. Hence, we may visualize our dependency

relation on the framework (V(E), E, p) in Rm as follows: we have assigned the
scalars Sij to the edges of the framework so that, at each vertex, the vectors
formed by directing the edges of E toward that vertex and multiplying them by
the associated scalars must sum to zero. We consider several examples.
Let V = {1, ... , 4}, m = 2, and let p embed V onto the four corners of the
unit square in 2-space. (See Figure 2.9) Note, there is no vertex at the center,
pl = (0,0) p2 = (1,0)
FIGURE 2.9. A resolvable stress on a tetrahedron.

the edges (1, 4) and (2, 3) simply cross there. As indicated in the figure, we
assign the scalar 1 to the edges corresponding to the sides of the square and -1
to the edges corresponding to the diagonals. By direct computation, one easily
checks that equation (2.2.2) is valid at each vertex. We conclude that for this p
the set K is dependent.
Physically, we may think of the assignment of Sij 's as corresponding to the
replacement of each edge (i, j) with a spring whose compression or extension
is indicated by the value 8;,;. In the above example, the horizontal and ver-
tical members are compressed springs, and the diagonal members are extended
springs. The dependency condition at each vertex says that the sum of the forces
there is zero, so that the structure is stable.
Now suppose that E consists of all of the edges of K except the edge (1, 2),
and let 8;,; be any dependency relation for the set of remaining edges. In order
that (2.2.2) is satisfied at Pt, we must have s13 = 814 = 0: two nonzero vectors
can sum to zero only if they are negatives of one another; but, no nonzero
multiples of (Pt - P3) and (Pt - P4) can be negatives of one another since these
vectors have non-parallel directions. Similarly, 8 23 = 8 24 = 0 and it follows that
s 34 = 0. Hence, the rows corresponding to the edges of E satisfy no non-trivial
dependency relation. This last example, coupled with Theorem 2.2.1 shows that
the tetrahedron minus an edge is generically independent in the plane.
In Figure 2.10, we have two embeddings of the complete bipartite graph K3,3,
i.e., a graph on six vertices with nine edges which may be described as a 6-circuit
with its three major diagonals. To the embedding illustrated in 2.10a we have
2 3
2
FIGURE 2.10. Two frameworks on K3,3·
assigned scalars which yield a dependency relation. However, with respect to

the embedding illustrated in 2.10b, the nine edges are independent.
Exercise 2.12. Consider Figure 2.10.
a. Verify equation {2.2.2} at the vertices of the planar framework in Fig-
ure 2.10a.
b. Prove that the nine edges of the framework in Figure 2.10b are indepen-
dent. Hint: suppose that {s;,;} gives a relation among these edges; show
that 836 = 0 then show that s23 = S34 = 0, etc.
c. Prove that the hexagon with diagonals is generically independent in the
plane.
d. Prove that any connected trivalent graph on 6 or more vertices is gener-
ically independent in the plane.
Clearly, finding non-trivial dependency relations, or showing that none exist,
is a useful method for deciding the independence of an edge set. Hence, we
formalize this technique. Let E ~ K be given and consider the vector space
JRE of all functions from E to the reals. For s E JRE, we denote 8(i,j) by s;,;i
s is called a set of stresses for E and 8;,; is the stress on the edge (i,j). A set
of stresses for E, not all zero, which satisfy the equations in (2.2.2) above, is
often called a non-trivial resolvable set of stresses for E. Hence an edge set E is
dependent with respect to an embedding p if and only if it admits a resolvable
set of stresses. Clearly, the collection of resolvable stresses for E form a subspace
of RE; we denote this subspace by S(E).
In the special case E = K, we consider the linear transformation T : RK -+
(Rm) v where (T( s ))i is given by:
(2.2.3) (T(s))i = ~::::si;(Pi- P;).

#i
To illustrate this linear transformation, consider again the case n = 4, m = 2,
and the slightly different embedding pas indicated in Figure 2.11. Here pis a
P3 = (0,1)
P4 = (x,y)
P1 = (0,0)
FIGURE 2.11. Stressing a general tetrahedron.
general embedding, i.e. x "' 0, y "' 0, and x + y "' 1. One easily checks that, for
any s E RK, we have:
T(sh = (-s12- xs14, -s13- ys14)
T(s)2 = + s23 + (1- x)s24, -s23- ys24)
(s12
T(s)a = (-s23- XS34,s13 + s23 + (1- y)s34)
T(s)4 = (xs14 + (x -1)s24 + xs34, ys14 + ys24 + (y- 1)s34)
We may now ask if the edge set K is independent with respect to this em-
bedding, or alternatively, if it is dependent, i.e. if there exists a non-trivial s so
that T(s) is the zero vector. To answer this second question, we check to see if
the set of eight equations in six unknowns obtained by setting each coordinate
of T(s)i to zero for each i, has a non-trivial solution. The answer is yes and it
is left as an exercise for the reader to show that the solution set to this system
has the description given below.
Exercise 2.13. The solution set to the system just described has the form of
all S E RK such that·. s12 -- -..,"'S 14.• s13 -- - ys 14.· s23 -- (1-a:-y)'
-a:11811 • s24 -- (1-a:-y)'
z8' 4 •
and s34 -- (1 11811
a: y).
We conclude, from Exercise 2.13, that K is dependent with respect to this
embedding. In view of the examples in Figure 2.10, it is natural to ask if there is
any planar embedding, with respect to which, K is independent. The answer is
no. This will follow once we prove Lemma 2.2.2 which implies that the generic
embeddings are invariant under a change of coordinates. Hence we may assume
that the embedding in Fig 2.11 is generic.
We return to the general case where n and m are arbitrary, and consider
E ~ K. To investigate the possible dependence of E, we consider 1jE. the
restriction of the linear transformation T to the subspace RE (of RK), which,
of course, is also a linear transformation, and call it the stress operator of the
framework (V,E,p). In order to simplify the notation we will continue to write:
(2.2.4) (1jE(s))i = ~:::>i;(Pi- P;),

ioFi
where we understand Sij to be 0 whenever (i,j) is not in E. With these simpli-
fying conventions, we have that S(E) = ker(1jE)· Thus, E will be independent
with respect to p if and only if the kernel of the linear transformation 1jE is
trivial or, equivalently, E will be dependent if and only if the kernel of 1jE is
non-trivial. Since we will need to refer to this fact often, we formalize it as a
lemma.
LEMMA 2.2.1. Let the framework (V,E,p) be given and let 1jE be the stress
operator defined in 2.2.4 above. Then:
a. S(E) = ker(TIE)·
b. E is independent with respect to p if and only if the kernel of 1jE is
trivial.
c. If IV(E)I :::; m + 1 and p(V(E)) is in general position, then E is inde-
pendent.
Exercise 2.14. Prove conclusion (c) of Lemma 2.2.1. Hint: Note that for
each i, the vectors {Pi-P; I (i,j) E E} are independent.
Exercise 2.15. Prove that the matrix of 1jE is the transpose of the rigidity
matrix.
It follows from Lemma 2.2.1 that E is independent with respect top if and
only if S(E) is trivial. We use this fact now to show that the set of generic
embeddings are invariant under a change of coordinate systems. By an affine
transformation of Rm we mean a function A : Rm - Rm of the form A(x) =
xM + t, where M is a nonsingular m x m matrix and t is any point of Rm.
This induces a transformation of Rnm which we also denote by A and define by
A(p)i = A(pi)·
LEMMA 2.2.2. Let V = {1, ... , n} and p any embedding of V into Rm be
given; let A be an affine transformation of Rm. Then
a. For any edge set E ~ K, the rows of R(A(p)) corresponding toE are
independent if and only if the rows of R(p) corresponding to E are in-
dependent.
b. A(p) is a generic embedding of V if and only if p is generic.
PROOF. Let E ~ K and assume that the rows of R(p) corresponding to E

are dependent i.e. S(E) contains a non-zero element. Then for each i in V,
TjE(S)i = :~~::>iJ(Pi- P;) = 0.

;.;i
Multiplying both sides of the last equality by M on the right and adding then
subtracting t inside the parentheses gives
:~::>i;(A(p)i- A(p);) = 0.
j=i
We conclude that the rows of R(A(p)) which correspond toE are dependent.
Part a now follows by symmetry; specifically the invertability of affine transfor-
mation. (See Exercise 2.16 below). Part b follows at once from part a. 0
Exercise 2.16. Let A be an affine transformation of am.

a. Prove that A has an inverse, A -l, which is also an affine transformation.
b. Let p be an embedding of V into am. Show that p is a general embedding
if and only if A(p) is a general embedding.
This ability to change coordinates now enables us to give a simple proof of the
important fact that generic embeddings are general embeddings, (the converse,
of course, is not true except in dimension one).
LEMMA 2.2.3. Generic embeddings are general embeddings.
PROOF. Let V = {1, ... ,m} and let p: V--+ am be an embedding which is
not general. Without loss of generality, we may assume that P1, ... , Pk span a
(k- I)-dimensional affine space and that Pk+l lies on that affine space where
k $ m. Let U = { 1, ... , k + 1}. Changing coordinates if necessary, we assume
that Pi = ei, for i = 1, ... , k. It follows that Pk+l = (x1, x2, ... , Xk, 0, ... , 0)
where x1 + x2 + · · · + Xk = 1. Let Sij = -XiXj, for 1 $ i < j $ k and let
si(k+l) =Xi, for 1 $ i $ k. One easily checks that for each i = 1, ... , k + 1
L Sn;(Pi- P;) = 0.
;.;i
Thus, K(U) is dependent. However, by Corollary 1 to Theorem 2.2.1 and
Lemma 2.2.1c, K(U) is generically independent. We conclude that p is not
generic and, if coordinates have been changed, the original embedding was not
generic. 0
Exercise 2.17. Verify the vector computations in Lemma 2.2.3.

Exercise 2.18. Give an example of a general embedding in the plane which
is not generic.
In view of the standard formula
(2.2.5) dim(ker(TjE)) + dim(im(TjE)) = dim(domain(TjE))

for a linear transformation and the observation that dim(domain(TjE)) =lEI, we
need only compute the dimension of the image of 11E in order to. decide whether
E is independent or not. Now the space, im(TjE), is a subspace of Rmn and its
dimension can be related to the dimension of its orthogonal complement in Rmn
using another standard formula:
(2.2.6)
Combining 2.2.5 and 2.2.6 and Lemma 2.2.1, we see that E is independent
if and only if dim(im(TjE).L) = mn -lEI. In the next section, we will see that
im(TjE).L is none other than the space of infinitesimal motions of the framework
(V(E), E, p)! We close this section with a couple of exercises, the first of which
is to verify a lemma which will prove to be most useful in the next section.
LEMMA 2.2.4. For a given embedding p of V into Rm and a given E ~ K, the

image ofTjE is spanned by the rows of the rigidity matrix R(p) which correspond
to the edges in E.
Exercise 2.19. Prove Lemma 2.2.4.

Exercise 2.20. Let p map V into R by p(i) = i, for all i = 1, ... , n. By
Theorem 2.1.1, this is a generic embedding of V into R 1 ; hence, results relative
to this very simple embedding are, in fact, generic results.
a. Show that E ~ K is a cycle, with respect to this embedding, if and only
if E is the edge set of an elementary circuit in (V, K).
b. For any edge set E, show that the collection of the sets of stresses de-
termined by the elementary circuits of (V(E), E) span ker(TjE)·
2.3. Infinitesimal Motions and Isometries. Let V = {1, ... ,n} and p
mapping V into Rm be given. Let E be an edge set of (V, K) and consider the
framework (V,E,p). Recall from the introductory chapter that u E (Rm)v is
an infinitesimal motion of (V, E, p) if
(2.3.1) (Ui- u;) *(Pi-P;)= 0, for all (i,j) in E,

where* denotes the usual inner product in Rm. If we identify u E (Rm)v with
then-tuple of vectors (u1, u2, ... , Un), we see at once that equality (2.3.1) holds
if and only if for each edge (i, j) we have
(ut.U2,··· ,un)*(O, ... ,O,pi-P;,O, ... ,O,p;-Pi,O, ... ,0)=0,

where here * is the usual inner product in Rnm and the right hand factor is
the row of the rigidity matrix R(p) corresponding to the edge (i,j). Applying
Lemma 2.2.4, we have:
2.3. INFINITESIMAL MOTIONS AND ISOMETRIES 31
THEOREM 2.3.1. Let V, the embedding p of V into IRm, and the edge set
E ~ K be given. Let 11E be the stress opemtor from JR.E into (JRm) v. The set
of infinitesimal motions of the framework (V, E, p) is a subspace of (Rm) v. In
fact, it is the orthogonal complement of the subspace spanned by the rows of the
rigidity matrix R(p) which correspond to the edges of E or, equivalently, the
orthogonal complement of the image of 11E.
We will denote this space of infinitesimal motions of E by V(E), (denoted
simply by V in the introduction). Combining this theorem with equations 2.2.5
and 2.2.6 of the previous section yields:
COROLLARY 2.3.1. Let V, the embedding p of V into Rm and the edge set
E ~ K be given. Then:
dim(V(E)) = nm -lEI+ dim(S(E)),
where S(E) denotes the space of resolvable stresses for E.
In order to decide whether or not (V(E), E, p) is rigid, we must compare V(E)
with V(K(V(E))), (denoted by V in the introduction). To simplify the notation,
we will denote V(K(V(E))) by V(E); we call this the space of infinitesimal
isometries of V (E), the reason for selecting this terminology will soon become
clear. Since E ~ K(V(E)), the orthogonal complement of the space spanned by
the rows of R(p) corresponding to E contains the orthogonal complement of the
space spanned by the rows of R(p) corresponding to K(V(E)). We have:
LEMMA 2.3.1. Given V, the embedding p of V into IRm and E ~ K; the space
V(E) is a subspace ofV(E).
As in the introduction, we will say that the framework (V(E),E,p) is in-
finitesimally rigid (with respect to the embedding p) if V(E) = V(E) and we
will call an infinitesimal motion in V(E)- 'D(E) an infinitesimal flex of E.
The natural place to start our investigation of the space V(E) is with a con-
sideration of the infinitesimal direct isometries of IRm. By an isometry oflR.m, we
mean a one to one function from Rm onto Rm which preserves the distance be-
tween pairs of points. These can be classified as orientation preserving (direct),
e.g. translations and rotations, or orientation reversing (opposite), e.g. reflec-
tions. Of these only the direct isometries can be achieved by a motion within
IRm. Thinking of a direct isometry as a motion, we may then consider the vector
field of the initial velocities of the points of Rm under this motion. We can give a
straight forward definition of such a vector field: a vector field U : IRm -+ IRm is
an infinitesimal isometry oflRm if (U(x)- U(y))*(x-y) = 0, for all x,y E Rm.
If U is an infinitesimal isometry of Rm, then it is clear that u E (JRm) v defined
by 11i = U(pi), for i = 1, ... , n, is an infinitesimal motion for (V, E, p), for all
E ~ K. With the next few results, we will prove that the infinitesimal motions of
(V, E, p) obtained by restricting the infinitesimal isometries of Rm to V (E) are
the infinitesimal motions in V(K(V(E))). To facilitate these proofs, we define,
for any subset S of IRm, a function U : S --+ IRm to be an infinitesimal isometry

of S if (U(x) - U(y)) * (x- y) = 0, for all x, y E S.
Exercise 2.21. Prove that the set of all vector fields of Rm is an infinite di-
mensional vector space over R under the natural addition and scalar multiplica-
tion: (aU+bW}(x) = aU(x)+bW(x). Then show the collection of infinitesimal
isometries of Rm, or of any subset S of IRm, is a subspace of this space.
While the reader may be familiar with the mathematics of the isometrics of
IR, the infinitesimal isometrics are likely to be unfamiliar objects; hence, we take
time to explore them here. We will denote the space of infinitesimal isometrics
of IRm by zm. We start our investigation by proving a very useful lemma.
LEMMA 2.3.2. Let S be any set of m points in IRm which is in general position
and let T be a subset of IRm containing S and also containing a set of m + 1
points in general position.
a. If V is any infinitesimal isometry of S, then there is at most one in-
finitesimal isometry U of Rm such that U restricted to S equals V.
b. If U and W are two infinitesimal isometries ofT such that U(x) =
W(x}, for all x inS, then U = W.
PROOF. LetS= {p1, ... ,pm}· Suppose that U and Ware two infinitesimal
isometries ofT such that U and W simply agree on the set S. Let x be any point
of T not on the hyperplane containing {p1 , ••• , Pm} and consider X = U - W.
Thus X is an infinitesimal isometry of IRm and we have:
(X(x) - X(pi)) * (x - Pt) = 0, for all i = 1, ... , m.

Since X(pi) = 0, we conclude that X(x) is perpendicular to each of them vectors
(x - Pt}, ... , (x - Pm)i since {Pt. ... , Pm} is in general position and x is not
on the hyperplane containing {Pt. ... , Pm}, these vectors are independent and
X(x) must be the zero vector.
If y E T lies on the hyperplane spanned by {p1 , ••• , Pm} it does not lie on
one of the hyperplanes spanned by an m-subset of {x, P1, ... , Pm}· Now the
above argument can be used to show that X(y) = 0. Thus X is identically zero
on T and U = W. This proves Part band, by taking T = IRm, it proves Part
a. D
The simplest of the infinitesimal isometries are the infinitesimal translations,

which are the constant vector fields. For any vector t E IRm, Tt is the vector
field defined by Tt(x) = t.
LEMMA 2.3.3. Let V and the embedding p ofV into Rm be given so that either
p(V) is in general position or contains a set of m + 1 points in general position;
let E ~ K such that V(E) = V. Then:

n{2m-n+ 1)
dim(V(E)) = ifn::::; m+ 1;
2
dim{V{E)) = (m;1) ifn= m,m+ 1;
dim(V(E)) ::::; (m;1) ifn>m+l.
PROOF. Assume n ::::; m + 1: By Lemma 2.2.1, S(K) is trivial. Then by

Corollary 1 to Theorem 2.3.1, dim{V(E)) = dim(V(K)) is given by nm- (;) =
n( 2m;n+l). The second equation then follows by substituting n =morn= m+1
into the first equation.
Now assume n > m + 1 and let S ~ p(V) be a set of m + 1 points in general
position. We have just shown that the dimension of the space of infinitesimal
isometries of S is (m: 1). We note that the space of infinitesimal isometries of
p(V) projects into the space of infinitesimal isometries of S and that it follows
from Lemma 2.3.2a that this projection is one to one. Hence the dimension of
the space of infinitesimal isometries of p(V) is less than or equal to the dimension
of the space of infinitesimal isometries of S. D
There are several ways to see, in fact, that equality holds in the third case of
Lemma 2.3.3: One could compute dim{S{K)) or one could show directly that
every infinitesimal isometry of S extends to an infinitesimal isometry of p(V).
Both these approaches involve extensive computations. We will show that every
infinitesimal isometry of S extends to an infinitesimal isometry of p(V) indirectly
by proving that the dimension of the space of infinitesimal isometries of :R.m is
{m:l).
The reader familiar with the isometries of :R.m should take care not to con-
fuse the infinitesimal transformation T t with the translation by t. The former
assigns t to the point x while the latter assigns x + t {its new position under the
translation) to x.
The set of all infinitesimal translations {like the set of translations) is an m-
dimensional real vector space. We let ei denote the i'th standard basis vector
for Rm (e 1 = {1, 0, ... , 0), etc.), we let e0 denote the zero vector, and we let Ti
denote the infinitesimal translation T 81 , for i = 0, ... , m.
Exercise 2.22. Consider the infinitesimal isometries of :R.m.
a. Show that the infinitesimal translations are infinitesimal isometries and
that To is the zero infinitesimal isometry.
b. Show that the infinitesimal translations form an m-dimensional subspace
ofzm with T 11 •.. , Tm as basis.
The infinitesimal rotations are a bit more complicated to describe; in fact, we
will not give the definition of an arbitrary infinitesimal rotation (that will be left
as an exercise).
To motivate the following definition of some special rotations, we first consider

the plane R 2 and a counterclockwise rotation about the origin. If (x, y) is any
point in the plane, the velocity vector of this motion at (x, y) is perpendicular to
the line through (0, 0) and (x, y) and pointing to the left as one views it from the
origin. See Figure 2.12. It is also clear that the length of the velocity vector at
(x', y') \ (2o:,2y)

/ \ (x,y)
(0,0)
(-x,-y) \
FIGURE 2.12. An infinitesimal rotation.
(x, y) is proportional to the distance of (x, y) from the origin. With this example
in mind, we define the vector field Ron R 2 by R(x, y) = ( -y, x), for all (x, y).
Exercise 2.23. Consider the actual rotation u of the plane defined by
u(x,y) = (xcos(8) -ysin(8),xsin(8)+ycos(8)).
If we fix x and y and permit 8 to vary from 0 to some positive value, we get a
parametric description of the trajectory of the point (x, y) under this rotation.
Differentiate this trajectory with respect to 8 and evaluate the derivative at 0.
Exercise 2.24. Consider the vector field R on R2 defined by R(x, y) =
(-y,x).
a. Show that the vector field R is an infinitesimal isometry.
b. Show that, ifU is any infinitesimal isometry of the plane with U(O, 0) =
(0, 0), then U = cR for some real number c.
c. Show that {Tl> T2, R} is a basis for the space of infinitesimal isometries
of the plane. (Hint: Let U be any infinitesimal isometry of the plane
and let U(O,O) = (a,b). Now apply Part b of this exercise toW=
(U- aT1- bT2).J
d. Use your work in Part c to show that, if U is any infinitesimal isometry
of the plane, then U(x, y) = (a- cy, b +ex) for some set of constants
a,b,c.
e. Use the formula developed in Part d to show that either U is a translation
(if c = 0} or U is an infinitesimal rotation about the point ( -b/c, afc).
(Note: a fixed point of an isometry corresponds to a "zero" of the asso-
ciated infinitesimal isometry.)
In !Rm, the axis of a rotation is an (m- 2)-dimensional hyperplane or affine

subspace A (a point when m = 2 and a line when m = 3). The velocity vector
of a point p under a rotation about A may be visualized as follows: Construct
the (m - 1)-dimensional hyperplane containing A and p and consider a vector
at p perpendicular to this hyperplane pointing in the direction of rotation.
The length of the vector will be proportional to the distance of p from the axis
A; of course, the velocity vector is the zero vector at a point of A. In order to
construct a basis for the infinitesimal isometries of !Rm, we will need to consider
only a finite set of special rotations.
The rotations that we will consider are those with an axis of rotation spanned
by m - 2 coordinate axes. Let ~;, where 1 $ i < j $ m, be defined by
~;(x 11 ••• ,xm) = Xie; -x;ei. We may easily verify that~; is an infinitesimal
isometry:
(~;{x)- ~;(Y)) * (x- y) = ((xi- Yi)e; - (x; - Y;)ei) * (x- y)

= (xi- Yi)e; * (x- y)- (x;- Y;)ei * (x- y)
= (xi- Yi)(x;- Y;)- (x;- Y;)(xi- Yi) = 0.
In the next lemma, we observe that these rotations along with the coordinate
translations form a basis for the space of all infinitesimal isometries of !Rm.
THEOREM 2.3.2. a. The space zm of infinitesimal isometries of !Rm has

dimension (mt 1).
b. If S is a subset of m points of !Rm in general position or contains a set
of m + 1 points in general position, then each infinitesimal isometry of
Sis the restriction of an infinitesimal isometry of!Rm to S.
PROOF. If lSI = m, let S' = S; otherwise let S' be an {m + 1)-subset of S

in general position. Denote the infinitesimal isometries of S' by I(S') and let
L : zm - I(S') denote the restriction map L(U)(x) = U{x), for all x E S'.
One easily checks that L is a linear transformation and, by Lemma 2.3.1, L
is one to one. By Lemma 2.3.3 dim(I(S')) = (mt 1) and we conclude that
dim(zm) $ (mt 1). If we can show that the (mt 1) infinitesimal isometries {Ti I
1 $ i $ m} U{ it; 11 $ i < j $ m} are independent, then we can conclude that
dim{zm) = (mt 1) and Lis onto.
Suppose we have a relation among the infinitesimal isometries in this set:
L ~Ti+ L bi;~; =To.

1:5i:5m 1:5i<j:5m
Evaluating both sides of this equation at e 0 yields:

and we conclude that a1 = a2 = am = 0. Substituting these values into the

equation and evaluating both sides at ei gives:
( -bli, · · · 1 -b(i-l)i 1 0, bi(i+l) 1 ••• 1 bin) = (0, 0, · · · 1 0),
We conclude that all of the coefficients must be zero and the infinitesimal isome-
tries are independent.
Finally, let U be any isometry of S. We have just shown that the restriction
of U to S' is the restriction of an isometry of Rm to S'. Hence, by Lemma 2.3.2a,
U is the restriction of an isometry of Rm to S'. 0
COROLLARY 2.3.1. Let V, the embedding p of V into Rm and E ~ K be

given, where V(E) is a set of m points in general position or contains a set of
m + 1 points in general position. Then
a. dim(V(E)) = mn- mjV(E)I + {mt 1).
b. If W is any infinitesimal isometry in V( E) (an infinitesimal isometry
ofV(K(E))), then there exists one and only one infinitesimal isometry
U ofRm su.ch that U(pi) = W(pi), for all i E V(E).
PROOF. Part b follows directly from the theorem as does the fact that the
space of isometries of p(V(E)) has dimension (mt 1). Note that, in constructing
an infinitesimal isometry W of E, we may define W independently on each
vertex in p(V- V(E)) and assign it to be any infinitesimal isometry of p(V(E))
on p(V(E)). Hence, the dimension of V(E) is equal to m(n -IV(E)I) plus the
dimension of the space of infinitesimal isometries of p(V(E)). 0
2.4. Infinitesimal and Generic Rigidity. For E ~ K, we define the de-

pendency number of E to be the number of independent relations among the
rows of R(p) corresponding toE or, equivalently, as the dimension of S(E); we
denote it by dn(E). We define the degree of freedom of E to be the dimension
of V(E) minus the dimension of V(E) and denote it by df(E).
THEOREM 2.4.1. Let V, the general embedding p of V into Rm and E ~ K

be given.
a. dn(E) 2:: 0, with equality if and only if E is independent.
b. df(E) 2:: 0, with equality if and only if E is rigid.
c. If IV(E)I2:: m, then lEI = mjV(E)I- {mt 1) + dn(E)- df(E).
d. If IV(E)I ~ (m + 1), then dn(E) = 0, (i.e. E is independent) and
df(E) = (IV~E>I) -lEI.
PROOF. Parts a and b follow at once from the definitions of dn(E) and df(E).
2.4. INFINITESIMAL AND GENERIC RIGIDITY 37
Assume IV(E)I = n ~ m. By the corollaries to Theorems 2.3.1 and 2.3.2, we

have:
df(E) = dim{V{E))- dim{'D(E))
= [nm -lEI+ dn(E)]- [m(n -IV(E)I) + (m; 1)]
= miV(E)I-IEI- (m; 1) + dn{E).
Now assume IV(E)I :5 m + 1. By Lemma 2.2.1c, E is independent and

dim{S) = 0. Thus, by Corollary 1 to Theorem 2.3.1, dim{V{E)) = nm -lEI.
Similarly, dim{V{E)) = dim(V(K(V(E))) = mn- (;) and part d follows by
direct computation. D
Theorem 2.4.1 can be used to produce most of the standard theoretical results
about infinitesimal rigidity. However, we will prove these results later in a more
general setting. Hence, we will not pursue this investigation beyond using this
result to prove the rigidity result parallel to Theorem 2.2.1 and to demonstrate,
with the following exercises, the relationship between stresses and infinitesimal
motions in working with specific exan1ples.
COROLLARY 2.4.1. If a framework (V(E), E, p) is infinitesimally rigid for

some geneml embedding p of V into Rm then (V (E), E, q) is infinitesimally
rigid for all generic embeddings q of V into Rm.
PROOF. Let dfp(E), dfq{E), dnp(E) and dnq{E) denote the degree of free-
dom and dependency number of E with respect to the embeddings p and q, where
pis any general embedding of V into Rm and q is any generic embedding of V into
Rm. We wish to show that dfp{E) = 0 implies that dfq{E) = 0. If IV{E)I < m,
we have from Theorem 2.4.1d: dfq{E) = IV(E)I(IV(E)I- 1}/2 -lEI = dfp(E).
Now assume that IV(E)I > m. Similarly, from Theorem 2.4.1c we get that
dnp(E) - dfp{E) = dnq{E) - dfq{E).
Since q is a generic embedding, dnp(E) ~ dnq{E) (see Exercise 2.25 below).
Hence, dfp(E) ~ dfq{E) ~ 0. D
This result leads us to define an edge set E to be generically rigid {for dimen-
sion m) if it is rigid with respect to all generic embeddings into m-space and to
reformulate the corollary:
COROLLARY 2.4.2. If a framework (V(E),E,p) is infinitesimally rigid for

some geneml embedding p of V into Rm then E is generically rigid for dimension
m.
Exercise 2.25. Let V, E ~ K(V), and p a geneml embedding of V into Rm
be given.
a. Show that the number of edges in a largest independent subset of E is

lEI- dnp(E).
b. Use this equality to show that dnq(E) ~ dnp(E), for any generic em-
bedding q of V into !Rm.
c. Consider the frameworks in Figure 2.10 on page 26. Use Theorem 2.4.1c
to show that, for both frameworks, df(E) = dn(E). Hence, in each
case, the framework is infinitesimally rigid if and only if the edge set is
independent.
In working Exercise 2.12, you showed that the edge sets were dependent
(2.12a) and independent (2.12b) by considering the space of stresses. We now
may deduce by Theorem 2.4.1 that the framework in Figure 2.10a is not infinites-
imally rigid while the one in 2.10b is infinitesimally rigid. Let us verify this di-
rectly by producing an infinitesimal flex for the framework 2.10a, whose vertices
lie one a regular hexagon, and showing that none exist for the framework2.10b.
In Figure 2.13a, we show an infinitesimal motion for the first framework. (Since
P1
Ps=(-1,1)
Pa=( -1,0) P4=(0,0) P3=(1,0)

P4
FIGURE 2.13. Two frameworks on Ka,a·

each triangle is an equilateral triangle, one easily checks that the indicated vector
assignment is an infinitesimal flex.) In Figure 2.13b, we illustrate an argument
that the second framework admits no infinitesimal flex. The argument goes as
follows: Let W be any infinitesimal motion of this framework. The restriction
of W to {p11 p4} is an infinitesimal isometry of that set and hence there is a
unique infinitesimal isometry of !Rm which agrees with W on this set. Let X
be the restriction of that infinitesimal isometry to p(V) and let U = W - X.
Note that U(p1) = U(p4) = 0 and that W is an infinitesimal isometry of the
framework if and only if U is identically 0. Hence, we may restrict our attention
to the simpler infinitesimal motion U. We wish to show that U is in fact iden-
tically zero. The edge joining Ps to p 4, the vector U(ps) must be of the form
(0, a), since if U(ps) = (x, y), then
0 = ((x,y)- (0,0)) * ((-1,0)- (0,0)) = -x.
2.5. RIGIDITY MATROIDS 39
Similarly, U(p3 ) = (0, b). Now let U(p6 ) = (x, y). Each of the three edges
containing p 6 yields an equation:
0= ((x,y)- (O,a)) * ((-1, 1)- (-1,0)) =y-a

0= ((x,y)- (0,0)) * ((-1, 1)- (0,2)) =-x-y
0= ((x, y)- (0, b))* (( -1, 1)- (1, 0)) = -2x+y-b
Combining these yields y = a, x = -a and b = 3a. A symmetric argument, with

P2 instead of P6, gives a = 3b. We conclude that a = b = 0 and in fact that U
is identically zero on V.
Exercise 2.26. Use the techniques of this section to explore the frameworks
in Figure 1.3.
Exercise 2.27. Use any of the techniques now available to classify all of the
graphs on 6 or fewer vertices as follows: Class 1: dependent for all embeddings.
Class 2: independent for all general embeddings. Class 3: generically indepen-
dent but admits "dependent" general embeddings.
2.5. Rigidity Matroids. In the preface, we stated that our primary interest
is in the combinatorial aspects of rigidity theory. However, as we noted then, the
geometric and combinatorial parts of rigidity theory are not so easily separated.
In fact, only in dimensions one and two has total separation of the two parts been
achieved. To illustrate just what we mean when we talk about the combinatorial
aspects of rigidity, consider a specific graph (V, E). We can ask the question:
is (V, E) generically rigid (generically independent) in dimension m? If m = 1
there is a simple answer: yes, if and only if (V, E) is connected (yes, if and only
if (V, E) is a forest). If m = 2 these question also have "combinatorial" answers.
In Chapter 4 we will give necessary and sufficient conditions for a graph to be
generically rigid (generically independent) in 2-space and these conditions will
be stated in terms of the structure of the graph. On the other hand, if m is three
or more, the only technique for answering these questions that is known to work
for all graphs is to chose a generic embedding p of V into IRm and to check if the
framework (V, E, p) is infinitesimally rigid or infinitesimally independent. Thus,
one of the more important open problems in rigidity theory is the formulation of
necessary and sufficient conditions for generic rigidity or generic independence
in dimension 3 and higher. In Chapter 5, we will discuss the search for such
conditions.
In general, combinatorial conditions can often best be expressed in the lan-
guage of matroid theory, and Chapter 3 is devoted to an introduction to matroid
theory. For our purposes here, a matroid consists of a finite set S and a matroid
closure operator 0 on S, i.e. 0 satisfies the four conditions:
Cl: If T ~ S, then T ~ (T).
C2: If R ~ T ~ S, then (R) ~ (T).
C3: If T ~ S, then ((T}} = (T).
C4: If T ~ S and s, t E (S- T), then s E (Tu{t}) if and only if

t E (Tu{s}).
We observed in Section 2.1 that an embedding p of V induces a matroid closure
operator on K = K(V): an edge (i,j) belongs to the closure of the edge set E
if the row of the rigidity matrix R(p) which corresponds to (i, j) belongs to the
subspace of anm spanned by the rows of R(p) corresponding to the edges in E.
Our approach to combinatorial rigidity is to try to characterize the closure
operator associated with the generic embeddings of V into am. This approach
necessitates a slight change i'n point of view. From now on we will concentrate
on the edge sets rather than the frameworks. For example, we will say that
the edge set E is rigid (or not rigid), with respect to p, instead of saying that
the framework (V(E),E,p) is rigid (or not rigid); and, when pis understood,
we will simply speak of E as rigid or not rigid. Similarly, we will now consider
infinitesimal motions of E instead of infinitesimal motions of the framework
{V{E),E,p).
If ( ·) is a matroid closure operator on a set S then there is an associated
concept of independence. We say T ~ S is independent (with respect to the
matroid given by(·)) if, for every sET, s ¢ (T- {s}).
Exercise 2.28. Let V be a finite set, let K = K(V), let p be an embedding of
V into am and let (·) be the closure operator on K given by p. Then, for every
E ~ K, we have two concepts of independence. One from the independence of
the rows of the rigidity matrix R(p) and a second, just defined, from the matroid
closure operator (·). Prove that the two concepts of independence agree for all
E~K.
We start the process of isolating the combinatorial aspects of rigidity by prov-

ing two useful lemmas relating some of the concepts developed in the last section
and the closure operator of a given embedding. To do this we need a slight ex-
tension of the concept of independence. Let (·) be a matroid closure operator
on the finite set S. Let T ~ S and s E S; we say that s is independent ofT
if s is not in (T). We also need some simplifying notation. Let V be a finite
set, let K = K(V) and let p be an embedding of V into am. If u is a vector
assignment in (am)v, we adopt the notation: 'Ui; = (ui- u;) *(Pi-P;), for
each edge {i,j) E K(V). Hence, u is an infinitesimal motion of E if and only if
'Ui; = 0 for all {i,j) E E.
LEMMA 2.5.1. Let V and the embedding p ofV into amn be given. Let K =
K(V) and let 0 be the closure operator of this embedding. Let E ~ K and
(i, j) E K. Then (i, j) is independent of E if and only if there is an infinitesimal
motion u of E such that 'Ui; =F 0.
PROOF. Let R(p) denote the rigidity matrix of {V,K,p). By Theorem 2.3.1,
V(E) is the orthogonal complement of the space spanned by the rows of R(p)
corresponding to the edges in E and V( EU {(i, j)}) is the orthogonal complement
of the space spanned by EU{ (i, j)}. It follows by a simple linear algebra argument
that V(EU{(i,j)}) ~ V(E) with equality if and only if the row of R(p) associated
with (i, j) is in the span of the rows associated with E, i.e. if and only if (i, j) is
in the closure of E. Thus (i,j) is independent of E if and only if there exists a
vector assignment u in V(E) but not in V(E U {(i,j)} ), i.e. if and only if there
exists an infinitesimal motion u of E such that 'Ui; =/: 0. 0
Exercise 2.29. Fill in the linear algebra aryument omitted in the proof of
Lemma 2.5.1.
LEMMA 2.5.2. Let V and the embedding p of V into :R.m be given. Let K =
K(V) and let(-) be the closure operator of this embedding. Then, for all E ~ K,
{E)~ K(V(E)); furthermore, E is rigid if and only if {E)= K(V(E)).
PROOF. If K(V(E)) = K, then {E) ~ K(V(E)). Assume then that (j, k)

is an edge not in K(V(E)). Without loss of generality, we may assume that
j is in V- V(E). Define u : V -+ Rm to be the zero vector at all vertices
except j and (P; - Pk) at j. Since u is zero at all vertices in V(E), u is an
infinitesimal motion of E; on the other hand, tt.;k equals ((p;-Pk)-O)*(P;-Pk)
which is not zero. By the previous lemma, (j, k) does not belong to {E). Thus,
{E) ~ K(V(E)). Finally, by Theorem 2.4.1, E is rigid if and only if V(E) =
'D(E). But V(E) = 'D(E) is equivalent to the statement that the subspace of
Rnm spanned by the rows of R(p) corresponding to edges in E is equal to the
subspace spanned by the rows corresponding to the edges in K(V(E)), which is
equivalent to {E) = K(V(E)). 0
We next develop two of the additional conditions satisfied by this closure
operator associated with rigidity.
THEOREM 2.5.1. Let V and the general embedding p of V into Rm be given.
Let K = K(V) and let (-) be the closure operator of this embedding. Then (-)
satisfies condition C5:
C5: If E, F ~ K and IV(E) n V(F)I < m, then
{E U F) ~ (K(V(E)) U K(V(F))).
PROOF. By Lemma 2.5.2, {EUF) ~ K(V(EUF)). Now suppose that (j, k) E
K(V(E U F)) - (K(V(E)) U K(V(F))). It follows from the properties of the
operators K(·) and V(·) that j belongs to V(E) but not V(F) and k belongs
to V(F) but not V(E) (or vice versa). We must show that (j, k) is not in the
closure of E U F.
Let S denote p(V(E) n V(F)). Since p(V) is in general position and lSI <
(m- 1), we may add a set T of m- 1 -lSI points to p(V) so that T U p(V)
is also in general position. Let A be the (m- 2)-dimensional affine space which
intersects p(V) U T in S U T and let U denote a nontrivial infinitesimal rotation
with axis A. Finally, we define the vector assignment u on V by 11i = U(pi), for
all i in V(F), and 11i = 0, for all i in V- V(F). Clearly, u is an infinitesimal
motion of F. Since U(pi) = 0 whenever Pi is in A, u is identically zero on V(E)

and, hence, an infinitesimal motion of E. We conclude that u is an infinitesimal
motion of E U F. Since j is in V(E), Uj = 0. Since p(V) U T is in general
position, and since Pk is not in S U T, uk is a nonzero vector perpendicular to
the affine hyperplane spanned by S U T U {Pk}· Since Pi does not lie in this
hyperplane, Uk is not perpendicular to (Pi- Pk) and Ujk is not zero. Then, by
Lemma 2.5.1, (j,k) is not in (EUF}. 0
THEOREM 2.5.2. Let V and the general embedding p of V into Rm be given.
Let K = K(V) and let (·) be the closure operator of this embedding. Then (-)
satisfies condition C6:
C6: If E,F ~ K are rigid and IV(E)nV(F)I ~ m, then EUF is rigid.
PROOF. Let u be an infinitesimal motion of EUF; we must show that u is an
infinitesimal isometry of p(V(EU F)). Clearly, u is an infinitesimal motion of E
and, since E is rigid, an infinitesimal isometry of p(V(E)). By Theorem 2.3.2b,
there exists a unique infinitesimal isometry U of Rmn so that ui = U(pi}, for
all i in V(E).
Similarly, there is a unique infinitesimal isometry W of Rm such that Ui =
W(pi}, for all i E V(F). But, then U and W agree at the points p(V(E)nV(F))
and, since this is a set of m or more points in general position, we have, by
Lemma 2.3.2b, U = W. Hence, Ui = U(pi), for all i E V(E) U V(F) =
V(EUF). 0
Let V be a finite set and m a positive integer. Let (-) be a matroid closure
operator on K which satisfies condition C5. Let E ~ K; taking F to be the empty
set, we have, by condition C5: (E) ~ K(V(E)). Thus, in any matroid whose
closure operator satisfies condition C5, we may define an edge set E to be rigid
(in that matroid} if (E) = K(V(E)). A matroid Am on K = K(V) whose closure
operator satisfies the conditions C5 and C6 is called an m-dimensional abstract
rigidity matroid for V. Given a framework (V, K(V), p}, with p: V--+ Rm, then
the rows of the rigidity matrix R(p) give a matroid on the edge set K (V) called
them-dimensional infinitesimal rigidity matroid ojp, and denoted by .F(p). lfp
is a general embedding of V into Rm, then, as we have just seen .F(p) defines an
m-dimensional abstract rigidity matroid on K (V). We denote the infinitesimal
matroid on K corresponding to a generic embedding of V into Rm by gm (n),
where n = lVI, and call it them-dimensional generic rigidity matroid for V. As
we have seen, this matroid is independent of the choice of generic embedding.
Exercise 2.30. Give an example of a (non-general} framework in the plane
for which .F(p) is not an abstract rigidity matroid. Conversely, does every 2-
dimensional abstract rigidity matroid arise as the infinitesimal rigidity matroid
of some planar framework?
In the next chapter as we develop matroid theory, we will investigate the
properties of abstract rigidity matroids. However, in pursuing the main goal
of this section, the identification of additional conditions that one may impose
in order to characterize the m-dimensional generic rigidity matroid, we must
develop some of the fundamental properties of abstract rigidity matroids. This
development will include some results already familiar to us, as we have proved
them for infinitesimal rigidity matroids. Hence many of the results in this section
should be viewed as generalizations or extensions of similar results in the previous
sections.
Let ( ·) be the closure operator of an m-dimensional abstract rigidity matroid
on V, let U be a subset of V and let E ~ K(U). As we have noted, (E) ~
K(V(E)). Since V(E) ~ U, we have (E) ~ K(U); Thus, the restriction of(-) to
the subsets of K(U) is an operator on K(U). Clearly, conditions Cl through C6
hold for the restriction of(-) to the subsets of K(U); hence, this restriction is the
closure operator of an m-dimensional abstract rigidity matroid on U. We have
proved Lemma 2.5.3a. Parts band c of that lemma follow from the fact that
the restriction to U of a general (generic) embedding of V into Rm is a general
(generic) embedding of U into Rm.
LEMMA 2.5.3. Let the finite set V be given and let U ~ V. Let (·) be an
opemtor on· K = K(V).
a. If (-) is the closure opemtor of an m-dimensional abstmct rigidity ma-
troid for V then the restriction of(-) to the subsets of K(U) is an m-
dimensional abstmct rigidity matroid for U.
b. If (-) is the closure opemtor of an m-dimensional infinitesimal rigidity
matroid for V then the restriction of(·) to the subsets of K(U) is an
m-dimensional infinitesimal rigidity matroid for U.
c. If (·) is the closure opemtor of an m-dimensional generic rigidity ma-
troid for V then the restriction of(·) to the subsets of K(U) is an m-
dimensional generic rigidity matroid for U.
The next lemma is devoted to some of the more useful properties of closure
and independence in a matroid.
LEMMA 2.5.4. Let(·) be the closure opemtor of a matroid on the set K.
a. If E ~ K is independent and F ~ E, then F is independent.
b. If E ~ K is independent and e E K - E is independent over E, then
E U {e} is independent.
c. If A ~ K and E ~ A is a maximal independent subset of A, then
(A)= (E).
d. If A ~ K, then all maximal independent subsets of A have the same
cardinality.
e. If((·)) is a second closure opemtor on K with the same closetlsets as(·),
then ((A)) = (A) for all A ~ K.
PROOF. Parts a and b follow directly from the definitions and are left as an
exercise for the reader.
Part c. Since E ~ A, (E) ~ (A}. If E = A, we are done. Assume then that

e E A - E. If e ¢ (E), E U {e} would, by Part b, be a larger independent subset
of A. Hence e E (E). We conclude A~ (E) and hence (A}= (E).
Part d. We proceed by induction on IAI. If A has only one element, or is
empty, then A has only one maximal independent set, and we are done. Assume
then that E and Fare two maximal independent subsets of A. Let e E E- F.
Consider A- {e} and assume (inductively} that all maximal independent subsets
of A - { e} have the same cardinality. Clearly, F is a maximal independent
subset of A- {e}. By part c, (A- {e}} =(F)= (A). However, e ¢ (E- {e}},
so (E- {e}} =f (A- {e}} and we conclude, by part c, that E- {e} is not a
maximal independent subset of A- {e}. Thus IE- {e}l < IFI and lEI ~ IFI·
By symmetry IFI ~ lEI.
Part e. Let E be independent, and e ¢E. Then e E (E) if and only if EU {e}
is dependent, and EU {e} is dependent if and only if e E ((E)). Thus the closure
operators (-} and ((·}} agree on all independent sets and, by part c, on all subsets
ofK. D
Exercise 2.31. Prove parts a and b of Lemma 2.5.4.

Let the finite set V be given and consider the complete graph (V, K). Let u
be a permutation of V and note that u induces a permutation of K (which we
also denote by u): u(i,j) = (u(i),u(j)), for all (i,j) E K, In fact, we may think
of u as an operator on K: u(E) = {u((i,j))l(i,j) E E}. If(-} is the closure
operator of an abstract rigidity matroid on V, we may ask if (·} commutes
with each permutation u, i.e. (u(E)} = u( (E)), for each E ~ K and each
permutation u of V. If this is the case, we say that the matroid and the closure
operator are symmetric. Even a general embedding can yield an infinitesimal
rigidity matroid which is not symmetric, for example see Figure 2.14. Generic
embeddings, however, give symmetric matroids.
FIGURE 2.14.
THEOREM 2.5.3. Let the finite set V and the integer m be given; then gm(V)
is symmetric.
PROOF. Let (·} be the closure operator of gm(V), let u be a permutation
of V. Let q = pu. Thus, for every E ~ K, the frameworks (V(E),E,q) and
(V(u(E)), u(E), p) are identical and u(E) is independent with respect to p if
and only if E is independent with respect to q. Since the rigidity matrix R( q)

is obtained from R(p) by permuting its rows and columns, q is also a generic
embedding of V. It follows that E is independent with respect to q if and only
if E is independent with respect top. We conclude that, for all E ~ K, O'(E) is
independent if and only if E is independent. The result now follows from Part c
of Lemma 2.5.4. 0
While the generic rigidity matroids are symmetric, this property is not suffi-
cient to characterize them, i.e. there are other abstract rigidity matroids which
are symmetric. In fact there are 2-dimensional infinitesimal rigidity matroids
which are symmetric but not generic.
Let the finite set V be given and consider the set of embeddings of V into the
unit circle of IR.2 given by x 2 +y2 = 1. As in the definition of generic embeddings,
we assign to each edge set E the algebraic set XE consisting of those embeddings
for which E is dependent. We say that pis generic if p ¢ UXE, where the union
is taken over all E such that XE ¥: IR.nm. We define p to be circle generic if
p ¢ UXE, where the union is taken over all E such that XE does not contain all
embeddings of V into the unit circle. Clearly almost all embeddings into the circle
are circle generic. We define the 2-dimensional circle rigidity matroid on V to be
the infinitesimal rigidity matroid on V given by any circle-generic embedding.
We denote this matroid by g~(IVI). In the next exercise, you will show that
g~(IVI) is symmetric but that it is not equal to g2(V) whenever lVI ; : : 6. The
latter conclusion will follow from the fact that the complete bipartite graph K 3 ,3
is generically independent (Exercise 2.12c) but dependent in g~(IVI).
Exercise 2.32. Let the finite set V be given.
a. Prove that g~(IVI) is symmetric.
b. Prove that, if lVI ; : : 6, then the edge sets of the subgraphs of (V, K)
which are isomorphic to the complete bipartite graph Ka,a are dependent
in g~(IVI).
An important function associated with a matroid on a set S is therank func-
tion: for any T ~ S, r(T) is the size of a maximal independent subset ofT.
As we will see, the matroid is uniquely determined by its rank function. Some
of the fundamental properties of an m-dimensional abstract rigidity matroid
(Lemma 2.5.6) are stated in terms of the rank function. But, before we prove
that lemma, we assemble several of the simple properties of the rank function of
an m-dimensional abstract rigidity matroid.
LEMMA 2.5.5. Let the finite set V be given, let E ~ K = K(V), let(·} be the
closure operator of an m-dimensional abstract rigidity matroid for V, and let r
be its rank function.
a. E is independent if and only if r(E) = lEI.
b. E is independent if and only if, for all (i,j) E E, r(E- (i,j)) < r(E)
c. IfF~ E then r(F)::;; r(E).
d. Let {i,j) E {K -E), then r(EU {{i,j)}) is either r(E) or r(E) + 1, with
the latter case holding if and only if (i, j) is independent of E.
e. r{(E)) = r(E).
LEMMA 2.5.6. Let the finite set V be given and let r be the mnk function of
an m-dimensional abstract rigidity matroid for V.
a. Let E ~ K(V), io E (V- V(E)), and let it. ... , ik E V(E) with k ~ m.
Then r(E U {{io, il), ... , (io, ik)}) = r(E) + k.
b. For any U ~ V:
(k)
{ m2k- if lUI = k ~ m + 1;
r(K(U)) = (m2+1) if IU I =k2::m+l.
PROOF. Part a: Let E' be a largest independent subset of E, let Fo be the
empty set and, for j = 1, ... , k, let F; = {(i0 , il), ... , {io, i;)}. Finally, for
j = 0, 1, ... , k, let H; = E' UF;. Clearly, H 0 = E' and is therefore independent.
We wish to prove that H; is independent, for all j. We proceed by induction
on j. Note that H;+1 = H; U {{io, i;+l)}. Assuming that H; is independent,
we need only show that (io, i;+l) does not belong to the closure of H;. But,
we have, by condition C5: (H;) = (E' U F;) ~ (K(V(E'))) U (K(V(Fj))). We
have that i;+l ¢ V(F;) and, since (E') ~ (E), we have that io ¢ V(E'). Hence,
{io, i;+l) ¢ (H;) and H;+l is independent. Since Hk is an independent subset
of E U Fk, we conclude that r(E U Fk) 2:: IHkl = IE' I+ k = r(E) + k. Finally by
repeated applications of Lemma 2.5.5d, r(E U Fk) ~ r(E) + k.
Part b: We consider first the case k ~ (m + 1). Let U = {i 1, ... , ik} and let
U; ={it.··· ,i;}, for j = 1, ... ,k. Since, K(U1 ) is empty, it is independent.
We proceed by induction and assume that r(K(U;- 1 )) = (j - 1)(j - 2)/2 and
note that
K(U;) = K(U;- 1) u {{it. i;), · · · , (i;-t. i;)}.
Thus, by Part a, r(K(U;)) = r(K(U;- 1 )) + (j -1) =g).
Now assume that k 2:: {m+1) and let U = SU{it. ... ,ih}, where lSI= m and
h = k-m. Let Eo= K(S) and inductively define E; to be E;-1 U{{s, i;)ls E 8}.
We wish to show that E; is both independent and rigid, for j = 0, 1, ... , h. We
proceed inductively noting that Eo is clearly rigid and, by the previous case,
independent. Assume that E;- 1 is independent and rigid. Then, by Part a, E;
is independent. Observe that E; = E;- 1 U K(S U {i;} ). Thus, by condition C6,
E; is rigid.
One easily verifies that IEhl = mk- (mt 1). Since Eh is independent, r(Eh) =
mk- (mt 1); since Eh is rigid,
r(K(U)) = r(K(V(Eh))) = r((Eh)) = r(Eh) = mk- ( m+1)

2 .
0
2.6. ISOSTATIC SETS 47
This Lemma permits us to establish a necessary condition for a set to be

independent in any abstract rigidity matroid.
THEOREM 2.5.4. Let V be a given finite set and let E ~ K = K(V) be an
independent set in an m-dimensional abstract rigidity matroid for V. Then, for
all F ~ E with IV(F)I ~ m, IFI ~ miV(F)I- (mi 1).
PROOF. Suppose that E is independent and that F ~E. Then F is indepen-
dent by Lemma 2.5.5c. Hence IFI = r(F) ~ r(K(V(F))). If IV(F)I ~ m, then,
by Lemma 2.5.6b, r(K(V(F))) = miV(F)I- (mi 1). D
Given a finite set V and a positive integer m, we say a set E ~ K = K(V)
satisfies Laman's condition if, for all F ~ E with IV(F)I ~ m, IFI ~ miV(F)I-
(mt 1). In Chapter 3, we will discuss the converse to Theorem 2.5.4. In particular,
we will show that, form equal one or two, the subsets of K which satisfy the
Laman condition are the independent sets of an m-dimensional abstract rigidity
matroid, specifically, the independent sets of gm(V). We will also show that, for
m greater than two, these sets cannot be the independent sets of any matroid.
In dimensions one and two, we may view gm (V) as the m-dimensional ab-
stract rigidity matroid for V with the maximal number of independent sets.
Alternatively, we may think of an arbitrary m-dimensional abstract rigidity ma-
troid (m = 1,2) as obtained from Qm(V) by adding some dependency conditions.
This leads us to the following definition: Consider two matroids M 1 and M 2 on
the same setS; we say Mt majorizes M2 and write M 1 ~ M2 if every subset
of S which is an independent set for M 2 is also an independent set for M 1 • In
this terminology, Theorem 2.2.1 may be restated as follows:
THEOREM 2.5.5 (RESTATEMENT OF THEOREM 2.2.1). Let p : V --+ Rm be
any general embedding. Then Qm(V) ~ F(p).
This restatement motivates the following conjecture:
CONJECTURE 2.5.1 (THE MAXIMAL CONJECTURE). Let the finite set V and
the integer m be given. If Am is any m-dimensional abstract rigidity matroid on
V, then Qm(V) ~Am·
2.6. Isostatic Sets. In the next chapter, we will prove that any matroid is
uniquely determined by its collection of independent sets or its rank function.
We will also show that the abstract rigidity matroids are uniquely determined by
the collection of those independent sets which are rigid. Such edge sets are called
isostatic. In the next lemma we state some of the useful facts about isostatic
sets.
LEMMA 2.6.1. Let V be a given finite set, let E ~ K = K(V) and consider
an m-dimensional abstract rigidity matroid for V.
a. Assume that IV(E)I ~ (m+ 1). Then E is independent; furthermore, E
is rigid, and hence isostatic, if and only if E = K(V(E)).
b. Assume that IV{E)I2:: {m + 1) and that E is isostatic. Then:

{i) lEI= miV{E)I- (mtt);
(ii) Each vertex of (V(E), E) has valence at least m;
(iii) (V(E), E) has a vertex with valence less than 2m.
c. Assume that IV( E) I 2:: {m + 1). Then, if any two of the following con-
ditions hold, all three hold and E is isostatic.
{i) lEI= miV{E)I- (mtt);
(ii) E is independent;
(iii) E is rigid.
PROOF. Part a: Let IV(E)I :5 {m + 1). By Lemma 2.5.6b, r(K(V(E)) =
IK(V(E))I. By Lemma 2.5.5a K(V(E)) is independent; then by, Lemma 2.5.5c,
E is independent. For any {i,j) E (K(V(E))-E), EU{(i,j)} is also independent
and, hence, {i,j) ¢(E). We conclude that (E)= E. It follows at once that E is
rigid, and hence isostatic, if and only if E = K(V(E)).
Part b: Now let IV{E)I be greater than m and assume that E is isostatic.
Since E is independent, lEI = r(E) (Lemma- 2.5.5a). Since E is rigid, r( (E)) =
r(K(V(E))) = miV(E)I- (m~ 1 ) (Lemma 2.5.6b). By Lemma 2.5.5f, r(E) =
r((E)). Combining these inequalities gives lEI= miV{E)I- (m~ 1 ).
Suppose that (V(E), E) contained a vertex i of valence less than m. Let
U = V(E)- {i}; let F = E(U); let H = E- F. We have IHI < m and
lUI = IV{E)I - 1 and V(F) ~ U. We have by Theorem 2.5.4 that IFI :5
mjV{F)I- (m~ 1 ). Combining this and the fact that V(F) ~ U, we have:
lEI= IFI + IHI :5 [miV{F)I- (m; 1 )1 + (m -1)

:5 [miU1-(m; 1)1+(m-1)
= [m(IV{E)I-1)- (m; 1)1 + {m -1)
= [miV{E)I- (m; 1)1 -1, contradiction!
Thus E every vertex in E has valence at least m.

Finally, we note that the sum of the valences of the vertices in (V(E), E) is
2IEI. Hence, the average valence is:
lEI - miV{E)I- (m~l) - (m~l)
2 1V{E)I - 2 IV{E)I - 2m- IV( E) I'
Thus, the average valence is less than 2m and there must be a vertex of valence
less than 2m.
The proof of part c is left as an exercise for the reader. D
Exercise 2.34. Fill in the missing step in the proof of Lemma 2.6.1b, and
prove Lemma 2.6.1c.
Let the finite set V be given and let K = K(V). The isostatic sets of{;h(V) are
easy to describe. As we stated at the beginning of this section, the independent
sets of {h (V) are the edge sets of subforests of (V, K) and the rigid sets of gl (V)
are the edge sets of connected subgraphs of (V, E). Hence, the isostatic sets of
gl (V) are the edge sets of subtrees of (V, K). We may verify the conclusions of
Lemma 2.6.1 in dimension one: Let (V(E), E) be a tree. If IV(E)I = 2, then
(V(E), E) is the complete graph consisting of a single edge; if IV( E) I ~ 2 then
lEI = IV(E)I-1, each vertex of (V(E), E) has valence at least one and (V(E), E)
contains at least one pendant vertex.
This very last observation enables us to describe a method for constructing all
!-isostatic sets (trees) in (V, K). If (V(E), E) is a subtree of (V, K), then it must
have a pendant vertex. Let i be this pendant vertex and let j denote the single
vertex adjacent to i. Then one easily sees that (V (F), F), where F = E- { (i, j)},
is also a tree. Conversely, if (V(F), F) is a subtree of (V, K) and if i E V- V(F)
and j E V(F), then (V(E), E), where E = F U {(i,j)}, is also a tree. Thus,
starting with a single edge and attaching several pendant vertices, one at a time,
always results in a tree. Furthermore, all trees are constructed in this way. The
natural question to ask is whether the construction can be extended to higher
dimensions.
Let the finite set V be given and consider an m-dimensional abstract rigidity
matroid Am for V. Let E ~ K = K(V), let U ~ V(E) where lUI = m and let
i E (V- V(E)). The edge set F = E U {(i,j)lj E U} is called a 0-extension of
E (in dimension m).
THEOREM 2.6.1. Let the finite set V and the positive integer m be given and
let Am be an m-dimensional abstract rigidity matroid for V. Let F ~ K = K (V)
be an isostatic set in Am and let E be a 0-extension ofF with V(E) ~ V. Then
E is isostatic in Am. Conversely, if E is isostatic in Am and (V(E), E) has a
vertex of valence m, then E is a 0-extension of some isostatic set in Am·
PROOF. Let U bean m-subset ofV(F), i E (V -V(F)) and E = FU{(i,j)lj E

U}. Assume that F is isostatic. Since IV(F)I ~ m, we have, by Lemma 2.5.6b,
IFI = r(F) = miV(F)I - ( m+l)

2 ·
By Lemma 2.5.6,
r(E) = r(F) + m = m(IV(F)I + 1)- ( m+l)

2 = miV(E)I- (m+l)
2 .
By direct computation,
lEI = IFI + m = m(IV(F)I + 1)- ( m+l)

2 = miV(E)I- (m+l)
2 .
By Lemma 2.5.5a, E is independent and then, by Lemma 2.6.1c, E is isostatic.

Conversely, if E is isostatic, then F as a subset of E is independent. But

then, by a direct count and Lemma 2.6.1c, we conclude that F is isostatic. 0
COROLLARY 2.6.1. Let the finite set V and the positive integerm be given and
let Am be an m-dimensional abstract rigidity matroid for V. Let F ~ K = K(V)
be independent in Am and let E be a 0-extension ofF with V(E) ~ V. Then
E is independent in Am· Conversely, if E is independent in Am and (V(E), E)
has a vertex of valence m, then E is a 0-extension of some independent set in
Am.
COROLLARY 2.6.2. Let the positive integers m and h and the finite set V,
with lVI ~ m + h, be given and consider Am, an abstract rigidity matroid for
V. Let Uo be an m-subset ofV, let it, ... , ih be distinct vertices in V- Uo. Let
Eo = K(Uo) and, for j = 1, ... , h, let E; be a 0-extension of E;-t· Then, for
j = 1, ... , h, E; is isostatic in Am.
Exercise 2.35. Prove these two corollaries.

It follows from the second corollary that any edge set in K(V) which is con-
structed from a single edge by a sequence of 0-extensions in dimension two is
isostatic in any 2-dimensional abstract rigidity matroid for V. One might ask if
all isostatic sets of Ch (V) could be constructed in this way. If the answer were
yes, then there would be only one 2-dimensional abstract rigidity matroid for V.
In fact, this is the case when lVI ~ 5. However, we have shown that the edge
set of the complete bipartite graph K 3 ,3 (Exercise 2.12d) is 2-independent and
hence, by Lemma 2.6.1c, 2-isostatic. Since Ka,3 has no vertex of valence two, it
could not be constructed by a sequence of 0-extensions.
Exercise 2.36. Let V be a six element set, let E ~ K = K(V) and consider
Ch(V).
a. Show that if E is isostatic and 3 ~ jV(E)I ~ 5 then (V(E), E) has a
vertex of valence two.
b. Starting with the single edge, list (up to graph isomorphism) all six iso-
static sets E with IV(E)I ~ 5.
c. List (up to graph isomorphism) all nine isostatic sets E with IV(E)I = 6
where (V(E), E) contains a vertex of valence two.
d. Show that (up to graph isomorphism} there are exactly two isostatic sets
E with IV(E)I = 6 where (V(E), E) contains no vertex of valence two.
In order to construct all 2-isostatic sets we would have to be able to attach
a vertex of valence 3. However, ifF is isostatic, if i E (V- V(F)), if U is a
3-subset of V(F) and E = FU {(i,j)lj E U}, E would have one too many edges
to be isostatic. Hence, we would have to delete an edge from F before attaching
a vertex of valence three if we hope to get an isostatic set. This leads us to the
general definition of a !-extension. Let the finite set V be given and consider
an m-dimensional abstract rigidity matroid Am for V. Let F ~ K = K(V),
let U ~ V(F) where lUI = m + 1, let (h, k) E F(U) and let i E {V - V(F)).
The edge set E = F- {(h,k)} U {{i,j)lj E U} is called a !-extension ofF (in
dimension m).
THEOREM 2.6.2. Let the finite set V and the positive integer m be given. Let
F ~ K = K(V) be an isostatic set in Qm(V) and let E be a !-extension ofF
with V(E) ~ V. Then E is isostatic in Qm(V). Conversely, if E is isostatic in
Qm(V) and (V(E), E) has a vertex of valence m + 1, then Eisa !-extension of
some isostatic set in Qm(V).
PROOF. Without loss of generality, we may assume that V = {0, 1, ... , n},
that n is a vertex of valence m + 1 in (V(E), E),that the set of neighbors of n
in this graph is U = {0, 1, ... , m} and that F = E- {(n, i)li = 0, 1, ... , m}.
Now choose an infinitesimal generic embedding p of V. By Lemma 2.2.2 we
may assume without loss of generality that Po is the origin and that Pi = «!i,
for i = 1, ... , m. In this proof we will consider the Pi as column vectors and
the vectors 11i of an infinitesimal motion as row vectors. Hence the usual inner
product of 11i with P; will be written as a matrix product UiPj
Assume first that E is isostatic. Then F is independent and there exists an
infinitesimal flex u of E- {{0, n)} and hence ofF which has Uon =F 0. Suppose
that u;k = 0 for all j, k E U. This means that u agrees with an infinitesimal
isometry w oflR.m restricted to U (Theorem 2.3.2b). Since (wn-wi)(Pn-Pi) = 0
for i = 1, ... , m, we may solve this system of equations for wn:
Wn = {w1{Pn- pl), · · · ,wm{Pn- Pm))M- 1,
where M is the m by m matrix with the vectors Pn - Pi as columns. Since the
points of p(V) are in general position M is invertible. Similarly:
Un = {u1 {Pn - pl), · · · , Um{Pn - Pm))M- 1,
Since w1 = u1, ... , Wm = Um, we conclude that Un = Wn· Replacing n by 0 in
this argument we also get Uo = wo. It follows that Uon =Won· But, since w is an
infinitesimal isometry, Won = 0 which is a contradiction. Hence our supposition
that u;k = 0 for all j, k E U must be false. Thus, for some j, k E U, F U {(j, k)}
is independent and, by a direct count and Lemma 2.6.1c, F U { (j, k)} is isostatic.
Now assume that, for some j and k E U, (j, k) is not in F and F U {(j, k)}
is isostatic. By relabeling if necessary, we may assume that {0, 1) is not in F
and F U {{0, 1)} is isostatic. By Corollary 1 to Theorem 2.6.1, E - {{0, n)} is
independent. Now, let u be an infinitesimal flex ofF which has Uo1 =F 0.
By solving the system
{2.6.1)
for Un, we may extend u to an infinitesimal flex of E- {(0, n)}. We may expand
each equation in {2.6.1):
((Ui - Uo) - (un - uo)) * ((pi -Po) - (Pn -Po)) = 0.
Then we rewrite each equation as:
UOi - (Un - Uo) *(Pi -Po) - (ui - uo) * (Pn -Po) + UOn = 0.
Using the equations in this form and recalling that Po = 0 and Pi = ~. for
i = 0, ... , m, we rewrite the system (2.6.1) as a matrix equation:
(2.6.2)
where Q is them x m matrix with (Ui- u 0 ) as i'th row and J is the 1 x m row
vector of l's. Solving (2.6.2) for (Un- Uo) gives:
(un- uo) = (u01, ... , Uom)- p~QT + uanJ.

Multiplying both sides on the right by Pn = (Pn- Po) and collecting terms, we
get:
(2.6.3)
Observe that the right hand side of (2.6.3) is a quadratic polynomial in the
coordinates of Pn, that the coefficients involve only the coordinates of the vectors
Ui, for i = 0, ... , m, and that, since Uol is not zero, this is not the trivial
polynomial. Since p is generic, we may assume that the coordinates of Pn are
not a zero of this polynomial, i.e. that the right hand side is not zero. We
conclude that UOn is not zero. Thus (0, n) is independent over E- {(0, n)} and
E is independent. Hence by computing lEI and applying Lemma 2.6.1c, we have
that E is isostatic 0
COROLLARY 2.6.1. Let the finite set V and the positive integer m be given.
Let F ~ K = K(V) be independent in Qm(V) and let E be a !-extension of
F with V(E) ~ V. Then E is independent in Qm(V). Conversely, if E is
independent in Qm(V) and (V(E), E) has a vertex of valence m + 1, then E is a
!-extension of some independent set in Qm(V).
COROLLARY 2.6.2. Let the positive integers m and h and the finite set V,
with lVI ~ m + h, be given. Let Uo be an m-subset of V, let i1, ... , ih be distinct
vertices in V - Uo. Let Eo = K(Uo) and, for j = 1, ... , h, let E; be either a
0-eztension or a !-extension of E;-1· Then, for j = 1, ... , h, E; is isostatic.
Exercise 2.37. Prove these two corollaries.
Intuitively, we may interpret a !-extension as follows: Given a graph (V(F), F)
where F is generically isostatic (independent) and given an edge ij in F we "split
it", i.e., insert a new vertex n in that edge and then add m- 1 additional edges
between nand other vertices in V(F). Then, the edge set of the graph (V, E),
which is obtained by this process, is also generically isostatic (independent)
Exercise 2.38. Prove that the complete bipartite graph Ks,t is independent
when either s or tis m or less and n ~ s + t. Which of these are rigid?
Exercise 2.39. Prove that the complete bipartite graph K 8 ,t, where s = m+l
and t = {mi 1), is isostatic in Qm(n), where n ~ {mi 2). Hint, start with the
complete graph on m+ 1 vertices and split each vertex.
The next chapter is devoted to the development of matroid theory. The ab-
stract rigidity matroids will be our primary set of examples of matroids and we
will prove a variety of results for these matroids. For the reader familiar with
matroid theory who may wish to skip Chapter 3, we have added a summary sec-
tion at the end of that chapter containing all matroid results concerning abstract
rigidity matroids.
Chapter 3. Matroid Theory
3.1. Closure Operators. Given a finite set E, the power set of E, denoted
by 'P(E), is the collection of all subsets of E. A matroid M is a finite set E
together with an operator (·) mapping 'P(E) into 'P(E) such that the following
four conditions are satisfied for each subset T of E.
Cl: T ~ (T);
C2: If R ~ T, then (R) ~ (T);
C3: ((T)) = (T).
C4: If s, t E (E- (T)), then s E (T U {t}) if and only if t E (T U {s}).
An operator satisfying these four conditions is called a matroid closure oper-
ator, the set (T) is called the closure of T; a subset F of E is called closed in
M ifF is the closure of some subset of E or, equivalently, if (F) =F. In the
next theorem we list several useful properties of closed sets. Some of these were
discussed earlier; see Exercise 2.4.
THEOREM 3.1.1. Let 0 be the closure operator for the matroid M onE, then
a. F is closed if and only ifF = (F).
b. E is closed.
c. If S, T ~ E are closed then S n T is closed.
d. (S) is the intersection of all closed sets containing S.
PROOF. Assume (·) satisfies axioms Cl-C4.

a. This is a direct consequence of axiom Cl.
b. This is a direct consequence of axiom Cl also.
c. By using the first two closure axioms, we have: S n T ~ (S n T) ~
(S) n (T) = S n T for closed sets S and T.
d. Let F be the intersection of all closed sets containing S. Since (S) is
closed and contains S, F ~ (S). On the other hand, S ~ F and, taking
the closure of both sides, (S) ~ (F). By part 2, F is closed and hence
(S) ~(F)= F. We conclude that (S) =F. 0
In Section 2.1 (see Lemma 2.1.2) we proved that the closure operator as-
sociated with a framework is the closure operator of a matroid. In exercises
55
56 CHAPTER 3. MATROID THEORY
throughout Chapter 2, we considered further examples. Below we list those

examples and a few additional examples.
EXAMPLE 3.1.1 (THE UNIFORM MATROID U(n, k)). Let E be an n-set and
let 0 :::; k :::; n. For S ~ E,
(S} = { S if lSI :::; k

E if lSI> k.
EXAMPLE 3.1.2 (THE COMPLETE CONNECTIVITY MATROID M(Kn)}. Let K
be the edge set of the complete graph Kn = (V, K) with lVI = n. For S ~ K,
(S} = { e E Klboth endpoints of e lie in the same component of (V, S)}
(See Exercises 2.2 and 2.6 in Chapter 2).
EXAMPLE 3.1.3 (THE VECTORIAL MATROIDS). Let E be a finite set of vec-
tors from some vector space. For any S ~ E, let (S} = span(S) n E. (See
Lemma 2.1.2 in Chapter 2).
EXAMPLE 3.1.4 (THE RANK 3 FANO MATROID). The rank 3 Fano matroid is
the vectorial matroid on the set E of nonzero vectors of a 3-dimensional vector
space over Z 2 , the field of two elements. This set of vectors may be interpreted
as the projective plane over Z 2 as illustrated in Figure 3.15. The vectors are
denoted by the seven points labeled a, . .. , g. Each of the seven lines of the
projective plane contains three points: abc, agd, aje, bge, bjd, cgj, and cde.
Geometrically, we may describe the closure operator as follows: the closure of a
point is that point; the closure of any two points or three collinear points is ''the
line" containing them; the closure of three non-collinear points or a set of more
than three points is the whole plane.
FIGURE 3.15. The Fano plane.
Exercise 3.1. Consider the closure operators, described in the previous ex-
amples, of the vectorial matroids and the complete connectivity matroids.
a. Show that each of these closure operators is a matroid closure operator.
b. In each case describe the closed sets of the matroid.
3.2. Independence Systems. The concept of a matroid has several formu-
lations, each emphasizing a different aspect of the structure, and we will discuss
most of them in this chapter. The formulation we discuss next is due to H.
Whitney. In 1935 Whitney, see (151), was the first to introduce the concept of
a matroid. He thought of the set E as a generalization of the set of columns of
3.2. INDEPENDENCE SYSTEMS 57
a matrix. The subsets of E are then of two types, dependent and independent.
Three conditions satisfied by the sets of columns of a matrix which are inde-
pendent are: the empty set is independent; any subset of an independent set is
also independent; and, given two independent sets one smaller than the other,
some element of the larger may be added to the smaller set to produce a new
independent set. Whitney observed that these three conditions alone were suffi-
cient to prove the uniqueness of the concept of rank: given a matrix all maximal
independent sets of columns contain the same number of columns. Whitney's
formal definition of a matroid is as follows:
A matroid is a pair (E,I) where E is a finite set and I is a collection of
subsets of E such that
11: 0 E I;
12: If ft E I and I2 ~ I11 then I2 E I
13: If It and I2 are members of I and lit I < II2I, then there exists an
element e in I2 -It such that I1 + e is a member of I.
For now we will call a collection I satisfying 11, 12, and 13 an independence
structure for E; the sets in I are called the independent sets of the independence
structure. We have defined matroids in terms of the closure operator because it
is natural to do so from the point of view of rigidity theory. We must now show
that our definition and Whitney's original definition agree.
Let (·} be the closure operator of a matroid M on E. I ~ E is said to be
independent in M if, for all e E I, e ¢ (I - e}. Let I( (·}) denote the collection
of independent sets in M. On the other hand, suppose we start with I, the
collection of independent sets of an independence structure for E. We may
define the closure operator (-}z on the power set of E as follows: For any subset
A of E, define (A}z by x E (A}z if either x E A or there exists an independent
subset I of A such that I + x is dependent.
LEMMA 3.2.1. Let I be the independent sets of an independence structure on

E. Let I be a largest independent set in A ~ E. Then I is a largest independent
set in (A}z and (A}z = (I}z.
PROOF. Let I be a largest independent set in A. Let x E (A}z and suppose

that I+ x is independent. There exists a J ~ A independent so that J + x
is dependent. By repeated application of condition 13 we may construct a set
H ~(I +x)- J so that JUH is independent and IJUHI =II +xl. Since J +x
is dependent, x ¢ JUH and JUH ~A contrary to the choice of I. We conclude
that x E (A}z implies that I+ xis dependent, i.e. {A}z = (I}z. Now suppose
that I'~ (A}z is independent and larger than I. By condition 13 again there is
an x E I' - I so that I+ x is independent. But x E (A}z = {I}z and I+ x must
be dependent. We conclude that I is a largest independent set in {A}z. D
THEOREM 3.2.1. Let(-} be the closure operator of a matroid M onE and let
I be the independent sets of an independence structure for E. Then
a. I( {·)) is an independence structure for E

b. {·)z is a closure opemtor of a matroid onE
c. {·h((·)) = {·)
d. I({·}z) =I
PROOF. Let the matroid M onE with closure operator{-) be given. Clearly,
the empty set is in I({·)), and I({·)) is closed under~- To show, that I({·)) also
satisfies the third independence axiom, assume there are two sets It and I2 in
I({·)) with II2I < lit I and It+ x ¢I({·)) for all x E It- h Among all such
pairs of sets choose It and I2 such that their intersection is as large as possible.
Clearly, {It) 2 {I2) and I2 is not a subset of It. Moreover, by the definition of
independence, we have that for all y E {I2), {I2 - y) is a proper subset of {I2).
Therefore, for y E (I2- It), there must be an x in It with x ¢ {I2- y). But then
I2 -y+x is in I({·)) because z E {I2- y- z+x) together with z ¢ {I2 -y- z)
implies by the fourth closure axiom that x E {I2 - y), contradicting the choice
of x. Now It and I2- y + x are independent sets whose intersection is larger
than that of It and h Thus there is a z E (I2- y + x) -It ~ I2- I11 so that
It+ z E I({·)), a contradiction.
Now we want to show that {-)z is a matroid closure operator. The first closure
axiom is clearly satisfied. Assume that A~ Band that x E {A)z. If x E A then
x E B ~ {B)z; if I ~ A is independent and I + x is dependent, then I ~ B
and x E {B)z. This shows that {A)z ~ {B)z. If I is a largest independent set
in A, then, by Lemma 3.2.1, it is a largest independent set in B = {A)z and
{A)z = {I)z. Thus {{A)z)z = {B)z = {I)z = {A)z, so that condition C3 holds.
To check C4, assume x, y ¢ {A)z, y E {A+x)z. Then there exists an indepen-
dent subset I of A + x containing x such that I+ y is dependent. I' = I - x + y
must also be independent, since y ¢ {A)z. But, I' +xis dependent, which means
that x E {A+ y)z and C4 is satisfied.
Checking that {·)z((·)) = {-) and I({·)z) =I is left as an exercise for the
reader. 0
Exercise 3.2. Verify c and d of the Theorem above.

Henceforth, a matroid M on E may be given in terms of its closure operator
{·) or the collection of its independent sets I.
Exercise 3.3. Describe the independent sets for each of the Examples 3.1.1
through 3.1.4.
EXAMPLE 3.2.1 (THE COMPLETE TRANSVERSAL MATROIDS). Let X be a fi-
nite set and let E be the power set of X, i.e. the collection of all subsets of X.
A collection {et, ... , eA:}, is said to satisfy Hall's conditions if I U~=t eil ~ k. A
collectionS is said to be independent if each of its subcollections satisfies Hall's
condition. Let I = { S I S is independent}
Exercise 3.4. Consider Example 3.2.1.
3.3. BASIS SYSTEMS 59
a. Adapt one of the standard proofs of (Philip) Hall's Theorem to show that
I, as defined in Example 3.2.1, is the collection of independent sets of a
matroid on E.
b. Describe the closure operator for this example.
c. For the two element set X = {a, b}, list the sixteen possible collections
of subsets of E = { c/>, {a}, {b}, X} and identify each as independent or
dependent.
EXAMPLE 3.2.2 (THE EUCLIDEAN MATROIDS).
a. Let E be a set of n points in real 2-space. A subset S of E is dependent
if S contains three points on a line, or more than three points. Let
I = { S I S is not dependent}
b. Let E be a set of n points in real 3-space. A subset S of E is dependent
if S contains three points on a line, or four points on a plane, or more
than four points. Let I= {SIS is not dependent}
EXAMPLE 3.2.3 (THE PARTITION MATROIDS). Let E be a finite set, h a pos-
itive integer, and P1 , ... , P~c a partition of E. Let
I= {I: IInJ'il::; h for all i = 1, ... ,k}.
Exercise 3.5. Consider Example 3.2.2a.
a. Show that I as defined in Example 3.2.2a is the collection of independent
sets for a matroid on E. Do the same for Examples 3.2.2b and 3.2.3.
b. Describe the closure operators of these matroids.
c. Show that if the points in Examples 3.2.2a or 3.2.2b are in general posi-
tion, then the matroid is the uniform matroid U(n, 2) or U(n, 3).
Exercise 3.6. Consider Example 3.2.3. If the condition that P1, ... , P~c form
a partition of E2 is relaxed and P1, ... , P~c are simply subsets of E, is I still a
matroid~
3.3. Basis Systems. Consider a matroid M on E with independent sets I.

Since every subset of an independent set is independent, I is uniquely deter-
mined by its maximal elements. Because of the condition 12, all of the maximal
independent sets have the same cardinality. These maximal independent sets are
called the bases of the matroid. The bases may be described directly: Let E be
a finite set, a nonempty collection B of subsets of E is called a basis system for
E if
Bl: B =F 0
B2: For all B1,B2 E B, IB1I = IB2I
B3: For all Bt. B2 E B and e1 E B1 - B2, there exists e2 E B2 - B1
such that B1 - e1 + e2 E B.
Condition B3 is sometimes called the exchange axiom. It also has a slightly
different but equivalent formulation:
B3': For all B1, B2 E B and e2 E B2 - B1, there exists e1 E B1 - B2
such that B 1 - e1 + e2 E B.
In condition B3 an element of B 1 is deleted and then a replacement is found

from B2; in condition B3' an element of B2 is added to B 1 and then an element
of B 1 is found to delete, leaving a new basis.
Exercise 3. 7. Let 8 be a collection of subsets of E.
a. Prove that if 8 satisfies Bl, B2, and B3, then {E- BIB E 8} satisfies
Bl, B2, and B3'.
b. Prove that if 8 satisfies Bl, B2, then it satisfies B3 if and only if it
satisfies B3'.
THEOREM 3.3.1. Let I be the independent sets of a matroid on E and let 8

be a basis system for E. Then
a. The maximal sets in I form a basis system for E.
b. The collection subsets of sets in 8 is the collection of independent sets
of a matroid for E.
c. I is the collection of subsets of sets in 8 if and only if 8 is the collection
of maximal sets in I.
PROOF. Let I be the independent sets of a matroid and 8' the collection of
maximal independent subsets of I. Since 0 is independent, 8 is not empty. Let
B1 and B2 be maximal independent sets. IB1I = IB2I, since otherwise we could
use 12 and augment the smaller one of these sets by elements of the larger one
to increase its size, contradicting maximality. Let e 1 be an element of B 1 - B2.
Then B 1 - e1 is independent, and by 12, there is an e2 in B2- B 11 such that
B 1 - e 1 + e2 is independent, and B3 is satisfied.
Let 8 be a nonempty collection of subsets of E such that Bl and B3 are
satisfied, and define the collection I' to be the collection of subsets of members
of 8. The first two independence axioms are trivially satisfied. To show 13, let
It and I2 be subsets of members of 8 with lit I< II2I· Choose bases B1 and B2
containing It and I2 respectively, such that B 1 n B2 is maximal. Let e 1 be an
element of B1- It. By B3, there is an e2 in B2 such that B1 - e1 + e2 is a basis
Ba, and by our maximality assumption e2 must be contained in /a. So It+ e2
is a subset of Ba and 13 is satisfied. Part 3 follows from a simple set theory
argument. D
Note that we used only properties Bl and B3 to prove part b. Combining

part b with parts a and c, we see that condition B2 must follow from conditions
Bland B3. You may wish to prove this directly.
Exercise 3.8. Describe the bases for each of the ezamples 3.1.1 through 3.2.3.
3.4. Rank Function. There is a natural function associated with a matroid
which also characterizes the matroid. Let M be a matroid on E with independent
sets I and define rz, a function from the power set of E into the nonnegative
integers by rz(S) = max{III : I E I, I ~ S}. The function r = rz is called the
rank function of M.
3.4. RANK FUNCTION 61
In general, let E be a finite set and r a function from the power set of E into
the nonnegative integers so that
Rl: r(0) = 0;
R2: r(S) ~ lSI;
R3: if S ~ T then r(S) ~ r(T);
R4: r(S U T) + r(S n T) ~ r(S) + r(T);
then r is called a rank function on E. If r is a rank function on Ewe define
I(r) = {I~ E I r(J) = III}
Condition R4 is called the submodular inequality. In the next theorem we
show that these two notions of rank function are equivalent.
THEOREM 3.4.1. Let I be the independent sets of a matroid onE and let r
be a rank function on E. Then
a. rz is a rank function E
b. I(r) is the collection of independent sets of a matroid
c. I(rz) =I
d. rz(r) = r
PROOF. Let M be a matroid defined on a set E by specifying a collection I
of independent sets. We will first show that its rank function, rz, satisfies Rl
through R4. The first three are trivially satisfied, so, to show R4, consider a
maximal independent subset I of SnT and augment I to a maximal independent
subset Is of S and augment Is to a maximal independent subset 1ST of S U T.
Then IT= (ISTnT) is an independent subset ofT but not necessarily maximal.
Thus, r(S n T) = Ill = lis nIT I, r(S u T) = lisT I = lis u IT I, r(S) = lis I and
r(T) ~ liT I· So, r(S n T) + r(S U T) = lis n ITI + lis U ITI = llsl + IITI ~
r(S) + r(T).
Conversely, consider a rank function ron E and consider the collection I(r).
By Rl, the empty set in I(r). If r(X) = lXI and Y ~ X we need to show that
r(Y) = IYI· Apply R3 to Y and X - Y to obtain r(X) ~ r(X - Y) + r(Y),
which, together with Rl, implies that r(Y) = IYI· To show 13, assume there are
in I(r) sets X andY such that lXI < IYI· Let Y- X= {x1, ... ,x~c} and let
xi = X+ X1 + ... +Xi, for i = 1, ... 'k. Apply R4 to xi and (X+ Xi)· Note
that Xi-1 u (X+ Xi) = xi and Xi-1 n (X+ Xi) = X. Thus, r(X,) + r(X) ~
r(Xi-1) + r(X +xi) or r(X,) ~ r(Xi-1 + [r(X +xi) - r(X)). Inductively we
have IYI = r(Y) ~ r(X~c) ~ r(X) + E~= 1 [r(X +xi)- r(X)] = lXI + E~= 1 [r(X +
x,) - r(X)]. Therefore, r(X + xi) > r(X) for some i and we easily see that
r(X +xi)= IX+ Xil and X+ Xi E I(r).
Therefore I(r) is the collection of independent sets of a matroid onE whose
rank function clearly is r. It is left to the reader to verify that I = I(r) and
rz(r) = r. D
Exercise 3.9. Prove c and d of the preceding theorem.

Exercise 3.10. Describe the rank functions for Examples 3.1.1 - 3.2.3.
Let M be a matroid on E and let r be the rank function of this matroid then
r(E) is called the rank of the matroid, and the notation r(M) is often used.
LEMMA 3.4.1. Let M be a matroid onE with closure operator(-}, rank/unc-

tion r, independent sets I and bases B. Then
a. r(8) :5 r(E), for all 8 ~ E
b. r(8) = r(E), if and only if B ~ 8 for some BE B
c. IE I if and only if r(I) = III
d. IE I if and only ifr(I- e)< r(I) for all e E I
e. r(8) = r((8)) for all 8 E E
f. (8) = {e I r(8 +e)= r(8)}
Exercise 3.11. Prove.Lemma 3.4.1

Exercise 3.12. Describe completely all matroids of rank 0, 1, or 2.
Exercise 3.13. Are all rank 3 matroids described in Example 3.2.2a?
Given a matroid M on E with rank function r, we say a set 8 ~ E is
a spanning set if r(8) = r(E). One may define a matroid in terms of these
spanning sets. This approach to matroids is uncommon. We shall see, however,
spanning sets and rigid sets are closely related in rigidity matroids. Hence, we
devote the next. exercise to developing the properties of spanning sets.
Exercise 3.14. Let M be a matroid on E with closure operator 0, bases B
and spanning sets S. Then
a. S = { 8 I (8) = E}
b. S = { 8 I 8 '2 B for some B E B}
c. B = {minimal sets in S}
d. S satisfies the following properties
S1: E E S
S2: 81 E S, 82 '2. 81 then 82 E S
S3: 81182 E S, l81l > l82l then there exists e E 81-82 such that
81 - e is a spanning set.
e. If S is any collection of subsets of E satisfying the properties in d, then
S is the collection of spanning sets in some matroid on E.
3.5. Cycle Systems. We now tum to the last general method for of de-
scribing a matroid that we shall introduce. Let M be a matroid on E. The
subsets of E are partitioned into two classes: the independent sets I and the
dependent sets. Clearly if we know which sets are in one class, we know which
are in the other. Hence, the dependent sets completely determine the matroid
M. Since subsets of independent sets are independent, it suffices to know the
collection of maximal independent sets B. Analogously, supersets of dependent
sets are dependent, so the collection 'Dof dependent subsets is determined by its
minimal members. The collection C of minimal dependent sets of the matroid
Mare called the cycles of M.
3.5. CYCLE SYSTEMS 63
LEMMA 3.5.1. Let M be a matroid onE, let (-) denote the closure operator
of M and let C be the collection of cycles of M.
a. 0 ¢C
b. If C1 and C2 are different cycles, then C1 ~ C2
c. If C is a cycle and e E C, then C - e is independent
d. If D ~ E with e,e' E D where e E (D-e) and e' ¢ (D-e'), then
e E ((D-e') -e)
e. IfC is a cycle and e E C, then e E (C- e)
f. If D ~ E satisfies the condition that e E (D-e), for all e ED, then D
is the union of cycles
g. If C1 and C2 are cycles withe' E C1 n C2 and e E (C2- C1),then there
is some cycle C so that e E C ~ (C1 U C2- e').
PROOF.
a. Since 0 is independent, 0 is not dependent and 0 ¢C.
b. This follows from the minimality of C2 under inclusion.
c. C- e as a proper subset of C must, by minimality of C, be independent.
d. LetT= D-e-e'. Note that e' ¢ (T +e) and hence e' ¢ (T). On the
other hand, e E (T + e'). In view of condition C4 of the closure operator,
we must have e E (T).
e. Since C is a cycle it is dependent, hence e E (C - e) for some e E C.
We must show that this holds for all e E C. Suppose e' E C and
e' ¢ (C- e'). Then by the previous result, e E ((C- e') - e), i.e.
(C - e') is dependent, contradicting c above.
f. Let D satisfy the property that e E (D-e) for all e E D. We wish to
show that given eo E D then there is a cycle Co so that eo E Co ~ D.
Since Dis dependent, there is some cycle C ~D. If eo E C we are done.
Assume e0 ¢C. Let e' E C and note that e' E (C- e') ~ (D-e'- e0 ).
Hence D-eo ~ ((D-e')-eo) and (D'-eo) = (D-eo) where D' =D-e'.
Thus eo E (D'- eo). Suppose e" E D' and e" ¢ (D'- e"). Then by d
above, e0 E (D"- eo), where D" = D'- e". Repeating this last step as
often as needed results in D* a proper subset of D containing eo and
satisfying the property that e E (D* - e) for all e E D*. If D* does
not contain a cycle containing e0 , we may repeat the entire construction.
But the process must stop after a finite number of applications.
g. The proof of this part is very similar to the previous proof; hence it is
only sketched. Let C1 and C2 be cycles with e' E C1 n C2 and eo E
C2- C1. Let D = C1 U C2- e' and let e E C2- C1. Since e ¢ C1,
e' E (C1 -e') ~(D-e). But then (D-e)= (D- e+e') = (C1 U (C2-
e)) 2 (C2 -e) = (C2). Hence e E (D-e). If e E C1 - C2, the same
argument will again yield e E (D-e). However, if e E C1 n C2 - e',
the argument is not valid and e need not belong to (D- e) in this
case. If e E D and e ¢ (D-e), we may delete e and repeating this
step yields D* ~ D so that e0 E D* and D* satisfies the condition
that e E (D# - e) for all e E D#. This D# is the union of cycles and
eo E C ~ D# ~ Ct U C2 - e'. 0
As we will soon show, the properties a, b, and g from this lemma characterize
the cycles of a matroid. Given a finite set E, we call a collection C a cycle system
forE, if the following three conditions are satisfied:
Zl: If C E C then C -::f 0
Z2: If Ct and C2 are members of C then Ct ~ C2
Z3: If C 1 and C2 are members of C and if e' is an element of Ct n C2,
then for each e E (C2- Ct) there is an element C E C, such that
e E C ~ (Ct n C2- e'}.
EXAMPLE 3.5.1 (THE BINARY MATROIDS). Let E be a finite set, and let 'D be
a collection of subsets of E which is closed under b., the symmetric difference.
Then the minimal non-empty sets in 'D form a cycle system for E.
EXAMPLE 3.5.2 (THE CYCLE MATROID OF A GRAPH). Let G = (V,E) be a
graph, then the edge sets of the elementary circuits in G form a cycle system for
E.
Exercise 3.15. Verify that the collections of cycles described in both of the
Examples 3.5.1 and 3.5.2 are cycle systems.
Let (-} be the closure operator of a matroid M on E. Recall that D ~ E is
said to be dependent in M if there exists an e ED such that e E (D-e). Let
C( (·}) denote the collection of cycles, i.e. minimal dependent sets in M.
Let C be a cycle system for E. We define the closure operator (·}c in the
power set of E as follows: For any subset A of E, define its closure by (A}c =
{e E E I e E A ore E C E C such that C- e ~A}
THEOREM 3.5.1. Let(-} be the closure operator of a matroid M onE and let
C be a cycle system for E. Then
a. C((·}} is a cycle system forE
b. (·}c is a closure operator of a matroid on E
c. (·}c((·)) = (-}
d. C((·}c) = C
PROOF. Part a follows from Lemma 3.5.1.

Assume that Zl, Z2 and Z3 hold for the collection C of subsets of E. Clearly
Cl and C2 are satisfied by (-}c.
To show C3, assume that there is an element e E ((T}c}c- (T}c. Then there
is a Ce E C containing e, such that Ce - e ~ (T}c. Choose Ce such that Ce
contains as few elements e1, ... e~c of (T}c - T as possible. For each ~ there
exists a Ci E C such that C;, - e;, ~ T. By Z3 there is a cycle C containing e
and not containing e;,, which is a subset of Ce U C;,, contradicting the minimality
assumption. We conclude that Ce- e ~ T and e E (T}.
To show C4, assume s, t ¢ (T)c, but s E (T + t)c. Then there is an element

C E C containing s such that C - s ~ T + t. Since C- s ~ T, t E C, which
means that t E {S + s)c.
Therefore Oc is a closure operator of a matroid on E. Parts c and d of the
theorem are easily verified. D
Exercise 3.16. Given a matroid M on E by specifying its collection I of
independent sets and given a cycle system for E, formulate and prove a theorem
analogous to Theorem 3.5.1. Your proof should only use independence and cycle
axioms.
EXAMPLE 3.5.3 (THE COMPLETE COCYCLE MATROID). Let (V, K) be a com-
plete graph. For any set of vertices U ~ V let Cu = {e = (x, y) I x E U, y ¢ U}
let C = {Cu I U ~ V, U =f c/J, U =f V}. In the following exercise you will show
that Cis the set of cycles of a matroid on K.
Exercise 3.17. Refer to Example 3.5.3
a. Show that C is the set of cycles for a matroid on K.
b. What are the bases for this matroid~
c. Describe its independent sets
d. Describe its rank function
e. Describe its closure operator
The following lemma sets out several of the more frequently used relationships
between the cycles and dependent sets of a matroid and the rank, basis, and
closure operator of that matroid.
LEMMA 3.5.2. Let M be a matroid onE with rank function r, closure operator
{·), dependent sets 'D, bases B, and cycles C. Then
a. ifC E C then r(C} = ICI-1
b. ifr(S) = ISI-1 and, for all e E S, r(S- e)= ISI-1, then SEC.
c. {S) = S U { e I there is C E C s. t. e E C ~ S + e }
d. B E B if and only if B ¢ 'D and B + e E 'D for all e E E - B
e. BE B if and only if B ¢ 'D and {B)= E.
f. If B E B and e E E - B, then B + e contains exactly one cycle C E C.
Moreover, C contains e.
PROOF.
a. This part follows directly from lemmas 3.4.1c, 3.4.1e, and 3.5.1c.
b. By Lemma 3.4.1c, S- e is independent for all e; by Lemma 3.4.1f,
e E {S- e), for all e. It follows at once that Sis a minimal dependent
set.
c. This part follows from the definition of Oc and Theorem 3.5.1c.
d. This part is simply a reformulation of the definition of a basis.
e. This follows from the previous two parts.
f. Let B be a basis and e E E- B. Since (B) = E, there is a cycle C such
that e E C ~ B + e. Suppose C' E C, C' =f C and e E C' ~ B + e.
Let e' E C'- C. By Lemma 3.5.1g, there exists C" E C such that
e' E C" ~ C' U C - e ~ B, a contradiction. 0
We have now introduced six different, but equivalent ways to define a matroid.
The flexibility that these definitions afford is very useful: the definition most
natural for a matroid will depend on that matroid. It just may turn out that,
for example, the cycles of a matroid defined in terms of its independent sets have
no "nice" description. This will become apparent as one works through the next
exercises.
Exercise 3.18. Describe the cycles for Examples 3.1.1 through 3.2.3.
Exercise 3.19. Describe the closure, independent sets, bases, and the rank
function for Examples 3.5.1 through 3.5.3.
In the next exercise we explore some of the properties of the dependent sets
of a matroid.
Exercise 3.20. Let M be a matroid onE with independent sets I, bases 8,
dependent sets 'D, cycles C. Show that the following statements hold:
a. 0¢ 'D
b. if Dt E 'D and D2 2 Dt then D2 E 'D
c. if Dt. D2 are distinct sets in 'D and e E Dt n D2, then Dt U D2 - e E 'D
d. 'D = { D I D 2 C for some C E C}
e. I = {I I I ~ C for all C E C}
LEMMA 3.5.3. Let E be a finite set.
a. A collection C of subsets of E is a cycle system for E if and only if C
satisfies Zl, Z2 and
Z3': If Ct and C2 are members of C and e E Ct n C2 then there
exists a C E C such that C ~ Ct n C2 - e.
b. If 'D is a collection of non-empty subsets E which satisfy condition c of
Exercise 3.20, then C, the collection of minimal non-empty subsets of 'D,
is a cycle system for E.
PROOF. Part a. If C satisfies Zl, Z2 and Z3, it clearly satisfies Z3'. Assume
that C satisfies Zl, Z2 and Z3' but not Z3. Then there are members C 1 and C 2
of C and elements e and e' of E, e E Ct nC2, e' E Ct- C2 and no cycle contained
in Ct u C2 that avoids e contains e'. Choose such Ct and C2 such that Ct u C2
is minimal. Z3' insures the existence of C3 ~ C1 U C2 - e. Ca does not contain
e' by our assumption, but Z2 implies that its intersection with C1 - C2 as well
as C2- Ct is nonempty. Let e" E (C2- Ct) n Ca. C2 U Ca is a proper subset of
Ct u C2, therefore, by our minimality assumption, Z3 holds for C2, Ca, e" and
e, so there is a cycle C4 ~ C2 U Ca - e" with e E C4. Now consider Ct. C4, e
and e'. Again Ct U C4 c Ct U C2, so there is a cycle Cs ~ Ct U C4 - e which
contains e', a contradiction.
The proof of part b is left as an exercise for the reader. 0
Exercise 3.21. Prove part b of Lemma 3.5.3.

Shortly after Hassler Whitney published the paper in which he defined the
concept of a matroid, he changed his research interests to algebraic geometry.
For over a decade, no papers on matroid theory were published. Then in 1948
William 'Thtte published a long paper in which he reintroduced matroids from
the point of view of cycles. Just as the differing points of view of the calculus lead
to different results, the two approaches to matroid theory encouraged growth in
different directions.
In the Whitney approach the elements of E are thought of as vectors, as in
Example 3.1.3. Independent sets, spanning sets and bases are naturally derived
from the vector space concepts. Cycles may be thought of as minimal sets of
vectors satisfying a nontrivial relation.
'Thtte, on the other hand, thought of the cycles as vectors. To explain his
approach, we look at the following generalization of Example 3.5.1. Let E be
a fixed finite set and F some field (not necessarily the reals}. Then FE, the set
of all functions from E into F, has the structure of an lEI-dimensional vector
space over F under the addition and scalar multiplication of functions. To each
function f E FE, we may associate the subset of E to which f assigns nonzero
values in F. This is called the support off and is denoted by u(/):
u(f) = {e I /(e)=/: 0}
If £ is any subspace of FE, we may consider the collections
'D = { u(/) I I E £, f not the zero function}

and C = {minimal sets in 'D}. Now let D 1 and D2 be distinct sets in 'D with
e E D1 n D2. We have /i E £ so Di = u(fi) for i = 1, 2. Let f = (h(e)ft -
ft(e)/2). Clearly f E £.It is also clear that u(/) ~ D1 UD2 and that e ¢ u(/).
Furthermore, since D 1 and D2 are distinct, It is not a scalar multiple of /2, and
f is not identically zero. Thus u(/) =/; 0. Thus, by part b of Lemma 3.5.3, C is
the set of cycles of some matroid on E.
If we take the field F to be the field Z2, we obtain Example 3.5.1: each vector
in Zf is uniquely determined by its support, i.e. u is one to one and onto. Thus
the vectors in Zf may be identified with the subsets of E. The vector sum of
two subsets SandT, S !::J. Tis the symmetric difference:
Si::J.T=SUT-SnT
The reader unfamiliar with this vector space may wish to verify the vector space
axioms, noting that the only scalars, 0 and 1, have the obvious properties under
scalar multiplication. OS = 0 and 18 = S for all subsets S. It follows then
from the above discussion that, if S is a subspace of Zf, the minimal nonempty
sets in S are the cycles of a matroid on E, a fact you have already verified in
Exercise 3.15
Exercise 3.22. Let E be given and let P 1 , ••. , P~c be a partition of E. Let
V = {F ~ E: IS n J'il is even for all i }
a. Show that V is a subspace of Zf
b. Describe the cycles of the matroid defined by this subspace
c. Describe the independent sets and bases
d. Describe the rank function
e. Describe the closure operator.
f. If the cells of the partition are replaced by an arbitrary collection of sets,
will the conclusion in part a remain valid?
3.6. Duality and Minors. We start this section by considering the matroid
introduced in Example 3.5.2. The cycle matroid of a graph was the paradigm for
both Whitney and Tutte. We will later use this example to illustrate the ideas
developed in this section.
Exercise 3.23. Let r = (V, E) be a graph.
a. Prove that the edge sets of the elementary circuits of r form a cycle
system for E.
b. Show that the independent sets of this matroid are the edge sets of the
subgraphs of r which are forests.
c. Show that, if r is connected, then the bases of this matroid are the edge
sets of the spanning trees of r.
d. Describe the bases in the case r is not connected.
e. Describe the bases in the general case.
f. Prove that if r is the complete graph then its cycle matroid is none other
than the complete connectivity matroid M(K), {Example 3.1.2}.
The first construction that we discuss is that of the dual matroid.
THEOREM 3.6.1. Let M be a matroid onE and let 8 be the bases forM. Let
8* = { E - B I B E 8}. Then 8* is the collection of bases for a matroid M* on
E.
PROOF. The result follows at once from Exercise 3.7a and b. We include a
short proof here based on results not available when basis systems were first
introduced.
If 8 is a nonempty collection of equicardinal subsets of E, then so is 8*, we
therefore only have to show that 8* satisfies B3.
Let Bi, B2 be elements of 8* with e1 E Bi - B:j. Since E - Bi E 8,
E- Bi + e1 contains exactly one cycle C of M and e1 E C, by Lemma 3.5.2.
We have C n B:j =F 0, because E - B:j is a base and cannot contain a cycle. Pick
any element e2 E C n B:j. Then E - Bi + e1 - e2 E 8 or Bi - e1 + e2 E 8*. D
We call M* the dual to the matroid M. It is clear that every matroid has
a dual, and that the dual is unique. It therefore makes sense to introduce the
following co-notation: Given a matroid M, we call the bases of M* cobases of
3.6. DUALITY AND MINORS 69
M, the cycles of M* cocycles of M, the rank function of M* the corank function

of M, etc.
LEMMA 3.6.1. Let M be a matroid onE with independent sets I, bases B,

spanning sets S; let M* be the dual of M and let I*, B*, S*, V* denote inde-
pendent sets, bases, spanning sets, and dependent sets respectively in M*. Then
a. (M*)* = M
b. r*(F) = IFI + r(E- F)- r(E), for all F ~ E
c. I* = {E - s s
I E S}
d. V* = {D I D n B =F 0 for all B E B}
e. S* = { E - I I I E I}
Exercise 3.24. Prove Lemma 3.6.1

Let r = (V, E) be a connected graph and let M be its cycle matroid. Com-
bining Exercise 3.23c and Theorem 3.4.1 we see that the complements of the
spanning trees are the bases for M*. One easily observes that these cobases are
maximal edge sets whose removal will not disconnect the graph. It follows that
the independent sets of M* are the edge sets whose removal does not disconnect
the graph. It follows then that a set F of edges is independent in M* if the
graph (V, E - F) is connected. The cycles, the minimal dependent sets, would
then be the edge sets which satisfy the following condition: the removal of all
of the edges in the set would result in a disconnected graph, but the removal of
all but one would leave the graph connected. The simplest example of a cocycle
(cycle of M*) is the set of edges with a common vertex x, where x is not a
cutvertex of r. We call these cocycles the vertex cocycles of r. We call M* the
cocycle matroid on r; in Example 3.5.3 we considered the special case when r is
a complete graph.
Exercise 3.25. Let r be a graph with several components and let M denote
its cycle matroid. Describe the bases, independent sets, cycles and rank function
ofM*.
Exercise 3.26. Extend this definition of the cycle and cocycle matroids of a
graph to multigraphs which admit loops. Describe the bases, independent sets,
cycles and rank function for these matroid.
Exercise 3.27. Let r = (V, E) be a planar graph and consider a particular
planar embedding. Let r• = (F, E) be the dual graph of r with respect to this
embedding. (Note r• may have loops or multiple edges.)
a. Show that the cycle matroid of r is the cocycle matroid of r• and vice
versa.
b. Note that the vertex cocycles of r• are the edge sets of the faces of the
embedding of r and vice versa.
One way to view the concept of matroid is that it is a generalization of graphs
which have a full duality theory: While only planar graphs have dual graphs,
the matroid of any graph has a dual. From the graph theory paper written by
Whitney in the early thirties (150], it seems reasonable that this fact played an
important role in the development of matroid theory.
Exercise 3.28. Describe the duals to the matroids in Examples 3.1.1, 3.1.3,
3.1.4, 3.2.1 and 3.2.3 {The dual of the rank 3 Fano matroid is of rank 4 and is
called the rank 4 Fano matroid.)
Exercise 3.29. Define a matroid M on the vertices of a regular octahedron
by declaring that the vertices of any triangular face are a basis. Show that M* =
M.
The term dual matroid probably was chosen by Whitney because of its relation
to duality for planar graphs. However, orthogonal matroid would have been a
better choice. This will become apparent as we consider duality from the Thtte
point of view. Return to the Thtte prototype: FE for some field F and some finite
set. This vector space FE has a natural inner product; (/,g)= 'EeeE f(e)g(e),
and, hence, a subspace C 2 F.c has an orthogonal complement £1.. Both C and
Cl. give rise to matroids onE and, as we will show, these matroids are the duals
of one another! However, this is far from obvious and several steps are needed
to prove this result.
First we note an interesting relationship between the cycles C of the matroid
defined by C and the cycles Cl. of the matroid determined by £1.: If C E C
and D E cl. then I D n c I =F 1. To see this, let I be any function in c so
that u(f) = C and let g E £1. with u(g) = D. Now {!,g) = 0. By definition
{!,g) = 'EeeE f(e)g(e). But, 'EeeE f(e)g(e) = 'Eeeu(J)nu(g) f(e)g(e). Now in
order for this last sum to be zero, either u(f) n u(g) = 0, or there is more than
one element in u(f) n u(g) (a single nonzero element cannot "sum to zero"). It
was this simple observation that Thtte exploited in developing duality theory
from the point of view of cycles.
LEMMA 3.6.2. Let M be a matroid on E and let C be the cycles of M.
c
Let 'Dl. = {S : IS n Cl =F 1 for all E C} then 'Dl. satisfies Z3' and cl. =
{minimal non-empty sets in 'Dl.} is the collection of cycles of a matroid on E.
PROOF. Let D1 and D2 be members of 'Dl. witheE D1 n D2. Let X be the

set of elements :r; in D1 U D2 - e with the property that there exists a cycle Cz
such that (Dt UD2- e) nC:~: = {x}. LetS= (Dt UD2 -e)- X. We claim that
8 E 'Dl..
First, X must be contained in D 1 nD2 , because if there is an :r; in X with, say
:r; E D 1 -D2, then CznD2 = {x}, or C:~:nD2 = {x, e}. The first case contradicts
the assumption that D2 E 'Dl.. The second case implies that Cz n D 1 = {e},
contradicting the assumption that D 2 E 'Dl.. Now assume that there is a cycle C
such that SnC = {y}. Among all such C choose one with minimal intersection
with X. Since y ¢X, C n X =F 0. If :r; EX n C, then, by Z3, there exists a
cycle C' ~ C:~: U C - :r; with y E C'. But then C' n X c C n X, contradicting
the choice of C. Thus 'Dl.. satisfies condition Z3'. Thus, by Lemma 3.5.3b, Cl.. is
a cycle system for E. D
We denote the matroid on E with cycles Cl.. by Ml... At this point it is not at
all obvious that Ml.. = M*, in fact it is not even easy to show that (Ml..)l.. = M.
THEOREM 3.6.2. Ml.. = M* and Cl.. = C*.
PROOF. We shall show that dependent sets in Ml.. are also dependent in M*
and that independent sets in Ml.. are also independent in M*.
Let D be a cycle of Ml... If D intersects every BE 8, then Dis dependent
in M* by Lemma 3.6.1d. Assume there is a base B of M with D n B = 0,
then for any element d E D, B + d contains a cycle Cd of M. d E Cd and
D n Cd = {d}, contradicting the assumption that D E c1... Hence all cycles of
Ml.. are dependent in M*. It follows at once that all dependent sets in Ml.. are
also dependent in M*.
Let I be a subset of E which contains no member of Cl... This means that for
every element e E I there is a cycle Ce in M whose intersection with I is {e}.
We conclude that {E- I) = {E). Therefore there is a base B of MinE- I,
hence I is contained in a base E - B of M*, i.e. is independent in M*. D
Exercise 3.30. Some of the classes of matroids described in Examples 3.1.1
to 3.5.3 are "closed" under duality, e.g., the dual of a uniform matroid is a
uniform matroid. Which of these classes are closed under duality?
The prototypes for the next matroid constructions that we will consider are
restrictions and contractions for graphs. We review these two concepts before
generalizing them to matroids. Let r = (V, E) be any graph and let F ~ E.
We will illustrate our constructions with the graph r in Figure 3.16; the edges
FIGURE 3.16.
ofF are indicated by the heavier lines. The restriction of r to F, rF = (V,F)

is obtained from r by simply deleting the edges of r not in F. The contraction
of r to F is obtained by contracting each connected component of r(E-F) to a
point.
In the next exercise you are asked to prove in general several conclusions one
may draw from this example.
Exercise 3.31. Let r = (V, E) be a graph and let F ~ E. Let M, MF and
M[FJ denote the cycle matroids ofr, rF and r(FJ respectively and use the same
indexing convention for the independent sets, cycles, and rank functions.
a. Show IF =In'P(F) ={I I I EI and I~ F}
b. ShowCp=Cn'P(F)={CICEC andC~F}
c. rp(F') = r(F'), for F' ~F.
d. Show C[F) = the minimal non-empty sets in {F n C I C ~ C}
e. r(FJ(F') = r(F' U (E- F))- r(E- F).
f. Describe I[FJ·
We now generalize these constructions to matroids.
Exercise 3.32. Let M be any matroid onE and let I, 8, C and r denote
the independent sets, bases, cycles, and rank function of M.
a. Show that Ip = In 'P(F) is an independence structure and hence IF is
the collection of independent sets of a matroid MF on F.
b. Show that the cycles of this matroid, Cp, are given by Cp = C n 'P(F).
c. Show that the rank function of Mp, rp, is simply the restriction of r,
the rank function of M, to F.
d. Describe the bases of MF
The matroid described in the previous exercise is called the restriction of M
to F and is denoted by Mp. Before defining the contraction for an arbitrary
matroid we reconsider our graphic prototype. Let r = (V, E) be a planar graph
and consider a specific planar embedding. Let r• = (P, E) be the dual graph
with respect to this embedding. Let F = E - {e} for the edge e. One easily sees
that the dual to r F with respect to the induced embedding is riFJ 1 i.e. deleting
an edge and contracting an edge are dual operations. This leads us to define
the contraction of M to F, M[FJ• in terms of its cocycles: (C[FJ)* = C* n 'P(F).
Thus M[F) is in fact defined to be ((M*)F )*.
THEOREM 3.6.3. Let M be a matroid onE with cycles C. Let F be a subset

of E and let MF and M[FJ be defined as above. Then
a. (M*)F = (M*)[F) and (M[FJ)* = (Mp)*
b. (C)F = C n 'P(F)
c. C[FJ consists of the minimal nonempty members of {C n F : C E C}
d. Ip = {I: I~ F, IE I}
e. I[FJ = {InF,I E I: I(In (E- F)l = r(E- F)}
f. 8p = {BnF,B E 8: IBnFI = r(F)}
g. B[FJ = {B n F, Be 8: I(B n (E- F) I = r(E- F)}

h. rp(S) = r(S) for all S ~ F
i. r(FJ(S) = r(S U (E- F))- r(E- F)
Exercise 3.33. Prove the parts of Theorem 3.6.3 not already proved in Ex-
ercise 3.32.
LEMMA 3.6.3. Let M be a matroid onE and letS~ T ~E. Then

a. Ms = (MT)s
b. M[s] = (M[TJ)[s]
C. (MT)[S) = (M[E-(T-S)J)S
d. (M[TJ)S = (ME-(T-S))[S]
PROOF. The first two parts are direct consequences of the definition of re-
striction and contraction.
Part c. Let C be a cycle of (MT)[S)· Then there is a cycleD of M such that
D ~ T and DnS =C. One easily checks that Dn(E- (T-S)) = C and hence
that there is some cycle C' of M[E-(T-S)J with C' ~ C. Hence each cycle of
(MT)[SJ is dependent in (M[E-(T-S)J)s. Next let C be a cycle in M[E-(T-S)J·
Then C ~ S and C is a cycle of (M[E-(T-S)J)s. Thus there is a cycle C' of M
such that C' n [E- (T- S)] =C. We conclude that C' ~ T and then that C
is dependent in (MT)[SJ· It follows that (MT)[s) and (M[E-(T-S)J)S have the
same dependent sets.
Part d is left as an exercise. D
Exercise 3.34. Prove part d.

An element e e E of a matroid M on E is called a loop if {e} is dependent
in M. Two elements e and e' of M are called pamllel if e and e' are not loops
and {e, e'} is dependent in M.
Exercise 3.35. Let M be a matroid onE and let e, e' E E
a. Show that e is a loop for M if and only if e belongs to no cycle of M*.
b. Show that e and e' are pamllel elements for M if and only if each cycle
of M* which contains one of e or e' must also contain the other.
A matroid M is said to be cogmphic if it is the cocycle matroid of a graph.
One important result of Whitney can be restated in the following form: A gmph
is planar if and only if its cycle matroid (the dual if its cocycle matroid} is
cogmphic. The existence of the dual graph in the case that r is planar, gives the
implication in one direction. To prove the other implication would take us too
far afield but the interested reader is encouraged to look it up (see [151]).
Exercise 3.36. Which of the matroids in Examples 3.1.1 to 3.5.3 are "closed"
under restriction? contmction? minors?
3.7. Connectivity. Let Mt be a matroid on E1 and M2 a matroid on E2,

where Et n E2 = 0. Let E = Et U E2 and C = Ct U C2 where Ct is the set of
cycles of Mi, (i = 1, 2). Then one easily checks that C satisfies the axioms for a
cycle system on E. If M is the matroid with cycles C, we write M = M1 E9M 2;
we say that M is the direct sum of M1 and M 2 and that Mt and M2 are direct
summands of M.
LEMMA 3.7.1. Let Mi be a matroid on Ei, i = 0, 1, 2 where E1 n E2 = 0,

Eo= E1UE2 andMo = M1ESM2. Letii, Bi, (·}i andri denote the independent
sets, bases, closure operator and rank function of Mi, (i = 0, 1, 2}. Then
a. Io ={It Ul2l li Eii}
b. Bo ={Btu B2 I Bi E Bi}
c. (F}o = (F n Eth U (F n E2}2
d. ro(F) = rt(FnEt) +r2(FnE2).
Conversely, if E1 and E2 partition E and if any one ofio, Bo, (·}o or ro is given
by the above formula then Mo = Mt E9 M2.
The proof of this lemma is straightforward and left as an exercise for the
reader.
Let Mi be a matroid on Ei for i = 1, ... , n and assume that E;. n E; = 0
whenever i ::f j. It follows at once from the commutativity and associativity of
the union that E9 is also commutative and associative. Hence M 1 E9 ... E9 MA is
well defined and invariant under reordering.
The notion direct of sum often has a connectivity theory associated with it.
Such is the case here. Let M0 denote the trivial matroid, i.e. the unique matroid
on the empty set. For any matroid M we have the direct sum decomposition
M = M E9 M0 = Me E9 M. If this is the only direct sum decomposition of M,
we say that M is connected. The first main result of this section is:
THEOREM 3.7.1. Let M be a matroid on E. Then there is a unique (up to

order) direct sum decomposition of M:
M =Mt E9 ... E9Mn,

where Mi is connected and nontrivial fori = 1, ... , n.
Before we prove this result, we prove several lemmas.
LEMMA 3.7.2. Let Mo = Mt ESM2 where Mi is a matroid on Ei, i = 0, 1, 2.

Then MEi = M[Ei) = M1., i = 1, 2.
PROOF. Let Ci denote the set of cycles of M1., i = 0, 1, 2. By definition,

Et n E2 = 0, Eo = Et U E2 and Co = Ct U C2. Thus Ct = Co n 'P(Et) which
by Theorem 3.6.3 implies that ME 1 = M 1 • We also have that, if C E Co, then
3.7. CONNECTIVITY 75
CnE1 = 0 or CnE1 =C. Thus the non-empty member of {CnE1 ICE Co} is
C1. Hence M[E1 J = M1. By a symmetric argument, M~ = M2 = M[~J· 0
LEMMA 3.7.3. Let M be a matroid onE and let F ~ E. MF is a direct
summand of M if and only if MF = M[FJ·
PROOF. If MF is a direct summand of M, then by the previous lemma

MF = M[FJ· Assume then that MF = M[F)· Let C denote the cycle space of
M and let C E C.
Let C1 = C n 'P(F) and C2 = C n 'P(E- F). Note that C1 is the cycle space
of MF. while c2 is the cycle space of ME-F· Thus cl u c2 is the cycle space
of MF E9 ME-F· Now suppose C E C. C n F is either empty or dependent in
M[FJ· IfCnF is empty thenCE C2. IfCnF is dependent in M[FJ = MF then
C n F contains a cycle of M. Hence C n F = C and C E C1. Thus C = C1 U C2
and M = MF EBME-F 0
LEMMA 3.7.4. Let M be a matroid onE; let M = M 1 E9 M~ and M =
M2 E9 M~ where M1 and M2 are matroids on E1 and E2 respectively. Then
M(E1 nE2 ) is also a direct summand of M, of M1 and of M2.
PROOF. In view of the previous lemma we need only show that MF = M[FJ•
(Ml)F = (Ml)[F)• and (M2)F = (M2)[F) where F = E1 n ~.
We have
(3.7.1) (Ml)F = (MEl)F = MF

(3.7.2) (Ml)[F) = (M[El))[F) = M[F)
(3.7.3) (Ml)[F) = (MEJ(F) = (M[E-(E 1 -f)J)F
These follow in order from Lemma 3.7.3 and Lemma 3.6.3, Lemma 3.7.3 and
Lemma 3.6.3, and Lemma 3.7.3 and Lemma 3.6.3. Now suppose that Cis a cycle
of (M[E-(E1 -F)J)F· Then there is a cycleD of M so that Dn(E-(E1-F)) =C.
But by the definition of direct sum, if D n E 1 :F 0 then D ~ E 1. We conclude
D =C. It follows that M[E-(E 1 -F)J)F = Mp.
Thus:
(Ml)[FJ =Mp
Combining equations (3.7.1), (3.7.2) and (3.7.3) we have
(Ml)[F] = (Ml)F = M[F) = MF

Thus by Lemma 3.7.3 MF = M[F) is a direct summand of M, MF =
(Ml)F = (M 1)[F) is a direct summand of M1 and, by a similar argument,
MF = (M2)F = (M2)[F] is a direct summand of M2. 0
We will now proceed to the proof of Theorem 3.7.1.
PROOF. If M is connected M = M is the decomposition we postulate. If M

is not connected, M = M1 E9 M2. If M2 is not connected, M2 = M~ E9 M3
and M = M 1 E9 M~ E9 M3. If any of these summands is not connected we may
replace it by its decomposition. Hence we have M = M 1 E9 M2 E9 ... E9 Mm
where each summ,a.nd is connected. We must now prove the uniqueness of this
decomposition.
Suppose M = M~ E9 ... E9 M:n, is a second decomposition of M. Let Mi be
a matroid on Ei and Mj, a matroid on Ej. In view of Lemma 3.7.4 and the fact
that Mi and Mj are connected, either Ei n Ej = 0 or Ei = Ej. We conclude
that the partitions {Ei I i = 1, ... , m} and {E~ I i = 1, ... , m'} are identical
and, hence, the two decompositions are simply reorderings of one another. 0
LEMMA 3.7.5. If M = M1 E9 M2 then M* = Mi E9 M2

Exercise 3.39. Let r = {V, E) be a graph. Let ri = {t-i, Ei) be the bicon-
nected components of r fori = 1, ... , m. Prove that {Ell ... , Em} is a partition
of E and that M = M1 E9 ... E9 Mm, where M is the cycle matroid of r and
Mi is the cycle matroid of ri, i = 1, ... , m, is the decomposition of M into
connected components.
It follows from this exercise that the cycle matroid of a graph is connected
if and only if the graph is biconnected or consists of a single edge. A standard
result from graph theory states that a graph is biconnected if and only if each
pair of edges belong to a common elementary circuit. Thus, the cycle matroid
of a graph is connected if and only if each pair of edges belongs to a common
cycle. This is actually true for all matroids as we now prove.
LEMMA 3. 7.6. Let M be a matroid on E with cycle structure C Let rv denote

the relation on E defined by x rv y if there exists C E C so that {x, y} £; C. Then
rv is symmetric and transitive.
PROOF. By its definition rv is symmetric. Now assume that x rv z and z rv y.

We wish to show that x rv y. Since x rv z there is a cycle containing x and
z, and since y rv x, there is a cycle containing y and z. Let C:c and C11 be
chosen so that x E C:c, y E C71 , C:c n C11 :f. 0 and, subject to these conditions,
C:c U C11 is as small as possible. If x E C11 or y E C:c, we have x rv y. Suppose
x :f. C11 andy :f. Cz. We will show that this supposition leads to a contradiction.
Let wE C:c n C11 and let C': be a cycle which contains x, avoids wand lies in
C:cUC11 • By minimality ofC:cUC11 , C:: 2 (C:c- (C:cnq,)) and since C': ~ C:c,
C'; n {C11 - {C:c n q,)) :f. 0. Similarly we define a;'. But then x E C::, y E a;',
C'; n a;' :f. 0 but C:: U c;: £; (C:c u C11 ) - z. Contradiction! 0
THEOREM 3.7.2. A matroid M onE with cycle set C is connected if and only
if for every e1, e2 E E there exists C E C so that {ell e2} £;C.
3.7. CONNECTIVITY 77
PROOF. Suppose M is not connected. Then M = M1 Ea M2 where M, is a

matroid in E,, E1 n E2 = 0. Let e1 E E1. and e2 E E2. Since C = C1 UC2, there
is no cycle inC which can contain both e1 and e2.
Now suppose M is connected and let e 1,e2 E E. Let C denote the cycle
system of M, let C1 = {C E C I e1 E C} and let E1 = Uoec1 C. We must show
that e2 E E1, i.e. that E1 = E. By the previous lemma if C n E1 =F 0, then
C ~ E1. Hence
c= {C E c 1 c ~ E1} u {C E c 1 c n E1 =F 0}.
But then M = ME 1 Ea ME-E 1 • Since M is connected, ME-E 1 is trivial and
E-E1 =0. 0
Returning to Exercise 3.39 and the observation that biconnectivity in graphs
corresponds to connectivity in their cycle matroids, it is natural to ask if simple
connectivity or n-connectivity, for n > 2, have matroid analogues. In Figure 3.17
we have drawn two graphs, one connected and one with two components which
FIGURE 3.17.
have identical cycle matroids. Thus, simple graph connectivity is not reflected in
its cycle matroid. Higher connectivity in graphs is defined in terms of separating
vertex sets. The diagrams in Figure 3.18 illustrate two element and three element
r n
FIGURE 3.18.
separating vertex sets. Each of these separating sets correspond to a partition

of the edge set. The question is: given the cycle matroid of the graph, but not
the graph itself, and the partition of the edge set, can we deduce the number
of vertices in the separating set? Surprisingly the answer is yes. Assume for
this discussion that each of the subgraphs r 1, r 2, 0 11 and 0 2 are connected, see
Figure 3.18. Then r and n are also connected. The bases of the cycle matroid
of a connected graph are the edge sets of the spanning trees, and hence the rank
of the cycle matroid of a connected graph is one less than the number of vertices
of the graph. Let E1 and~ be the edge sets of r 1 and r2 and E = E 1 u E 2

be the edge set of r; similarly let F1 and F2 be the edge sets of 01 and 02 and
F = Fl u F2 be the edge set of n Then IV(r)l = IV(rl)l + IV(r2)1 - 2. So
IV(r)l - 1 = (IV(rl)l - 1) + (IV(r2)1- 1) - 1 and r(E) = r(E1) + r(E2) - 1
where r is the rank function of the cycle matroid of r. Similarly r(F) = r(F1) +
r(F2)- 2 for n. So in these examples the size of the cutset which correspond
to the partition {Ell E2}, (respectively {F1, F2} ), is r(E1) + r(E2) - r(E) + 1,
(respectively r(F1) + r(F2) - r(F) + 1).
Exercise 3.40. Let r = (V, E) be a graph and let U ~ V separate r into
two connected components r1 and r2, and Let E 1 and E2 be the edge sets of
r 1 and r 2 and r be the rank function of the cycle matroid of r. Show that
lUI = r(E1) + r(E2) - r(E) + 1.
We say that a graph r = (V, E) is n-connected if, for all vertex separating
sets U ~ V, lUI 2: n. By Exercise 3.40 we can translate this to: r(E1) +r(E2)-
r(E) 2: n- 1 for all partitions {Ell E2} of the edge set which arise from a
vertex separating set of r. Of course, without referring to the graph, we cannot
tell which partitions of E come from vertex separating sets, i.e., we cannot use
the matroid structure to distinguish between the partitions of E which come
from vertex separating sets and those which do not. Hence, n-connectivity for
arbitrary matroids will have to be based on general partitions of the underlying
set.
Let M be a matroid on E. We say that a partition {Ell E2} of E is k-
separating if
Tl: k < min{IE1I, IE21}
T2: k = r(E1) + r(E2) - r(E)
We then define the connectivity of M by
tt(M) = min{k EN I there exists a k-separating partition of E}

with tt(M) = oo if, for every k 2: 0 there are no k-separating partition. We say
that a matroid M is k-connected if tt(M) 2: k.
LEMMA 3.7.7. Let M be a matroid on E. Then tt(M) > 0 if and only if M

is connected.
PROOF.Suppose tt(M) = 0. This is equivalent to the existence of a partition

{E1, E2} of E which is non-trivial, by Tl, and for which r(E) = r(E1) + r(E2),
by T2. We assert that this last condition is equivalent toM = ME 1 E9 ME2 •
By Lemma 3.7.2, M = ME 1 E9ME:z implies r = r(E1) +r(E2), so we now need
only show the converse.
Let B be any basis for M. Then B n Ei is independent in MEo i = 1, 2.
Thus IBnEil:::::; r(Ei)· But since IBI = r(E) equality must hold and BnEi is a
basis for MEo i = 1, 2. It now follows by the second part of Lemma 3.7.1 that
3.8. REPRESENTABILITY 79
M = ME 1 E9 ME2 • Thus ~~:(M) = 0 if and only if there is a non-trivial partition

of E so that M = ME 1 E9 ME2 • 0
Let M be a matroid on E and let {E 1, E2} be a partition of E. In Lemma 3.6.1
you proved that r*(E) = lEI - r(E) and r*(Ei) = r(E;) + IEil - r(E) where
{i,j} = {1,2}. Thus r*(Ei) +r*(E2) -r*(E) = r(E1) +r(E2) -r(E). The next
result now follows at once.
THEO~M 3.7.3. Let M be a matroid. Then ~t(M*) = ~~:(M).

While our definition of connectivity was a natural extension of graph connec-
tivity to matroids, the two connectivity theories are distinct. Of course there
is the shift by one: a graph r is 2-connected if and only if its cycle matroid
is !-connected, (or simply connected), but there are other differences. In fact,
matroid connectivity induces a second connectivity theory for graphs, which is
sometimes called 'futte-connectivity. It is defined by
~~:(r) if ~~:(M) = 0
r(r) ={ ~~:(M) + 1 if ~~:(M) > 0 I
where M is the cycle matroid of r, ~~:(r) is the graph connectivity of rand ~~:(M)
the matroid connectivity of M.
Exercise 3.41. Let r be a graph and let -y(r) denote the girth ofr, i.e., the
smallest elementary circuit in r. Prove the following:
a. Ifr is a complete graph on 1,2, or 3 vertices, then r(r) = oo,
b. If r is a complete graph on more than 3 vertices, then r(r) = 3,
c. If r is any graph, then r(r) :5 ~~:(r),
d. If r is any graph except a complete graph on 1, 2, or 3 vertices, then
r(r) = min{~~:(r),-y(r)}.
Exercise 3.42. Compute the connectivity of the following matroids
a. U(n,k),
b. the Fano Matroid,
c. the complete transversal matroid.
3.8. Representability. We say that a matroid M is representable over the
field F if there is a finite dimensional vector space V over F and a function
w : E --+ V, (called a Whitney function), so that F ~ E is independent in M
if and only if w is an injection when restricted to F and w(F) is independent
in V. If we coordinatize V, write the vectors in V as column vectors, order the
elements of E, E = {e 11 ... , en}, and construct the matrix M with w(ei) as
i'th column, for i = 1, ... , n, then we would have identified M with a Whitney
prototype matroid. In short a matroid is representable over a field F if and only
if it is isomorphic to the matroid of the columns of a matrix over IF.
As our first example of representability, we show how to represent the cycle
matroid of any graph over any field. Let r = (V, E) be a graph and let IF be a
field. To simplify the notation we may let V = {1,2, ... ,n} and E ~ {(i,j) I
1 ~ i < j ~ n}. Let V be then-dimensional vector space of n-tuples over IF,
with its elements represented as column vectors. Consider the Whitney functions
w: E- V, where w(i,j) is the column vector whose only non-zero entries are
a 1 in the j'th position and -1 in the i'th position, (of course, if the field has
characteristic 2, then -1 = 1). We leave it as an exercise for the reader to show
that this Whitney function actually gives rise to the cycle matroid of r.
Exercise 3.43. Let r = (V, E), F, V, and w be as defined above. Show that
the minimal dependent sets, i.e. the cycles of the matroid onE defined by w,
are the edge sets of the elementary circuits of r.
Exercise 3.44. Consider Example 3.1.4.
a. Show that the seven non-zero vectors in the three dimensional vector
space over Z2 give a representation of the Fano matroid over Z2.
b. Show that the Fano matroid cannot be represented over any field of char-
acteristic other than 2. It will follow that the Fano matroid is not rep-
resentable over the reals or the rationals.
As we will show below in Theorem 3.8.1, a matroid is represented over Z 2
if and only if it is a binary matroid. Recall that a matroid is binary if the
symmetric difference of two cycles is the disjoint union of cycles. A matroid
which is representable over every field is called a regular matroid. So the cycle
matroid of a graph is binary and regular while the Fano matroid is binary but
not regular.
Exercise 3.45. Show that the uniform matroid U (n, k) is representable over
any "sufficiently large" field. For what values of n and k is it regular?
Let M be a matrix over the field IF, let Me be the matroid represented by the
columns of M and let Mr be the matroid represented by the rows of M, that
is, Mr is represented by the columns of AfT. Clearly, then Me and Mr have
the same rank. What else can we say about the relationship between Me and
Mr? Surprisingly, the answer is "nothing else can be said about the relationship
between Me and Mr", as you will now show.
Exercise 3.46. Let M1 and M2 be any two matroids of the same rank both
representable over the field F. Then there is a matrix M over IF so that M1 is
isomorphic to Me and M2 is isomorphic to Mr.
As we discussed earlier, there were two linear algebra prototype matroids, the
first due to Whitney and the second due to Tutte, and so there should be a
second representation theory corresponding to the Tutte approach. Let M be a
matroid on E and let F be a field. A subspace T of the vector space of functions
from E into F is called a Thtte subspace forM over IF if the cycles of M are the
minimal non-empty sets in the collection of supports of the vectors in T. Not
surprisingly, this is not a distinct representation theory.
3.8. REPRESENTABILITY 81
THEOREM 3.8.1. Let M be a matroid then M is representable over F if and

only if there is a Thtte subspace for M over F.
PROOF. Let M be a matroid on E and let T be a Thtte subspace for M over

F. We must construct a Whitney function for M. We outline the construction
here leaving the details of the proof as an exercise for the reader.
Without loss of generality we may index E so that E = {ell ... , en} with
{ell ... , eb} being a basis for M. Let V be a b-dimensional vector space over IF,
let v1, ... , 'Vb be a basis for V and define w( ei) = Vi, i = 1, ... , b. Now consider
e; for j > b. As we have seen {e1, ... , eb, e;} contains a unique cycle C which in
turn contains e;. As one easily verifies, T contains a unique function f (up to
scalar multiplication) with Cas its support. Since e; E C, f(e;) =f. 0. We define
b
'"' f(e,)
w(e;) = L...., -f( ·)w(ei)·
i=l e,
Verifying that w, as defined, is a Whitney function forM is now straightforward.
Conversely, assume that M is a matroid on E and w is a Whitney function
for M over F. Specifically let w : E --+ V. Let IFE denote then-dimensional
vector space of all functions from E into F and define w : FE --+ V by w(f) =
Ei f(ei)w(~). One easily verifies that w is a linear transformation and that T,
the kernel of w, is a Thtte subspace forM over F. D
Exercise 3.47. Fill in the details of the proof of Theorem 3.8.1.

Exercise 3.48. Let T be a Thtte subspace for the matroid M on IF. Show
that T J. is a Thtte subspace for M*.
Let M be a matroid on E and let F ~ E. If w : E --+ V is a Whitney
function for M over F, then it follows at once that the restriction of w to F is
a Whitney function for MF over F. Combining this fact with Exercise 3.48 and
Theorem 3.6.3b gives the following corollary to Theorem 3.8.1.
COROLLARY 3.8.1. Let M be a matroid onE and let F ~E. If M is repre-

sentable over F, then Mp, M* and M[F) are also representable over IF. In fact
any minor of M or M* is representable over IF.
Exercise 3.49. Find a Thtte subspace for the cycle matroid of a gmph over
the mtionals, over F2.
If a matroid is representable over a field F, it is often easy to produce either
a Whitney function or a Thtte subspace which will demonstrate this fact. To
demonstrate that a given matroid is not representable over a given field is, in
general, not so easy. The following lemma can be useful in showing that a
matroid is not representable over a given finite field.
LEMMA 3.8.1. Let M be a matroid onE, representable over the finite field IF,
and let F ~E. Then the number of distinct cycles of M which are subsets ofF
is at most {IFik- 1)/{IIFI- 1) where k = IFI- r(F).
PROOF. Let 'I be a 'Thtte subspace forM and let 'IF={! E 'I I u(f) ~ F}.
One easily checks that 'IF is a subspace of dimension k = IFI - r(F). If two
functions in 'IF are scalar multiples of one another, then they have the same
support. There are exactly IFik - 1 non-zero vectors in 'Tp and hence there are
(IFik- 1)/{IIFI- 1) rays from the origin, ( all vectors which are scalar multiples
of one another). Thus 'IF contains (IFik - 1)/{IIFI - 1) functions with distinct
non-empty supports. D
We may use this lemma to show that U{4,2) is not binary, i.e. not rep-
resentable over Z2. Let E be the underlying 4-element set of U{4,2). Since
lEI- r(E) = 4-2 = 2, we have from the previous lemma that, if U{4,2) were
binary, it would admit at most {2 2 -1)/{2- 1) = 3 distinct cycles. But, each of
the four 3-element subsets of E is a cycle. Since U( 4, 2) is not binary, it follows
from the Corollary to Theorem 3.8.1 that if a matroid M has U(4, 2) as a minor
then M is not binary. In [130], 'Thtte provides the converse to thi~ observation,
giving the following characterization of binary matroid:
"A matroid M is binary if and only if the uniform matroid
U{4,2) is not a minor of M."
We will not complete the proof of this result in this text. The interested reader
can find a proof in most any book on matroid theory or may refer to 'Thtte's
original paper. In this same paper 'Thtte went on to characterize regular ma-
troids:
"A matroid M is regular if it is binary and contains no minor
isomorphic to the Fano matroid or its dual."
Again only one direction (Exercise 3.44b) is included in this text.
3.9. Transversal Matroids. As we have stated several times, a matroid is
representable over a field IF if and only if it is isomorphic to the column (or
row) matroid of a matrix over F. Recall that the column matroid of a matrix
may be defined as follows: the rank of a collection of columns is the rank of the
submatrix consisting of just these columns.
Consider now a matrix M of O's and 1's. If we select any field F, we may
think of M as a matrix over IF and thereby introduce a matroid structure on the
columns of M. Since different fields may yield different rank functions on the
columns of M, we may get several different matroid structures on the columns
of M. In [105], Ryser introduced the concept of term rank for a 0,1-matrix:
A collection of 1's in the matrix are said to be non-adjacent if no two lie in
the same row or column. The term rank of a 0,1-matrix M is the number of
elements a largest set of non-adjacent 1's in M.
THEOREM 3.9.1. Let M be a 0, 1-matrix and let E = {e11 ... , en} denote
3.9. TRANSVERSAL MATROIDS 83
the columns of M. We say that a collection of columns is independent if the

term rank of the submatrix consisting of these columns is equal to the number of
columns. These independent sets of E are the independent sets of a matroid on
E.
PROOF. Let I denote this collection of independent sets. Clearly 0 E I. If

I E I, then there is a set S of non-adjacent 1's with one entry from each column
of I. If J ~I, then the 1's inS which lie in the columns of J are a non-adjacent
set which shows that J E I.
Finally, suppose that I, J E I and III < IJI. LetS and T be non-adjacent
sets of 1's in I and J respectively. Consider the directed bipartite graph with
vertex set SUT and a directed edge from 8 E S totE T if sand t lie in the same
column, and a directed edge from t to 8 if 8 and t lie in the same row. One easily
checks that each vertex has in and out degrees less than or equal to one, so the
components of this graph are loops (corresponding to vertices of S n T), circuits,
paths and isolated points. Since ITI > lSI, either there is an isolated vertex in
T or a path starting and ending at vertices ofT. Finally, one easily checks that
the t vertices of this path form a non-adjacent set covering all the columns of I
plus one additional column of J, thereby verifying the exchange axiom. 0
A matroid defined on the columns (or rows) of a 0, 1-matrix by term rank is

called a transversal matroid.
Exercise 3.50. Prove that the complete transversal matroid, Example 3.2.1,
is indeed a transversal matroid.
Exercise 3.51. Consider the matrix
1 1
1
M= [ 1 1 1
1 1 1
1 0 0
01
1
1
01
0
1
l
Find the cycles of the column and row matroids of M over Z2, over Q, and as
transversal matroids.
Exercise 3.52. Are the uniform matroids U(n, k) transversal matroids?
Exercise 3.53. Consider the edges of a graph as two element subsets of the
vertex set. Then the edge set of a graph has a second natural matroid induced
on it by the graph structure, namely the transversal matroid of this collection of
sets. We call this the transversal matroid of a graph. Describe the independent
sets, bases, cycles, and rank function of this matroid.
LEMMA 3.9.1. Let M be a transversal matroid onE, then M is representable

over the real numbers.
PROOF. Let M be a 0, 1-matrix with Mas its column matroid with respect
to term rank. Let n be the number of 1's in M and let x 1 , ••• , Xn be independent
transcendentals over the rationals. Replace the 1's in M by x1, ... , Xn to get
M'. One easily sees that the terms in the determinant of any minor cannot
cancel out. Thus, for any minor of M, the term rank of that minor is the rank
of the corresponding minor of M'. 0
3.10. Graphic Matroids. We now consider an important subclass of regular

matroids: those that are isomorphic to the cycle or cocycle matroid of a graph.
But, first we relax our definition of graph to include multiple edges and loops.
Thus, the class of graphs will be closed under the operation of contraction. Also
the planar graphs will be closed under the construction of the planar dual, which
may also introduce multiple edges or loops. We define a matroid to be graphic if
it is isomorphic to the cycle matroid of a graph and co-graphic if it is isomorphic
to the cocycle matroid of a graph. The difficulty in identifying a graph as the
cycle or cocycle matroid of a graph is that the vertices are "lost" when one focuses
on the matroid. Can they be recovered? The answer is "yes"; specifically they
can be identified with special elements of the cocycle matroid.
Recall that the cocycles of a graph G are the minimal sets of edges whose
removal will increase the number of components. Among these cocycles are the
vertex cocycles: the collection of edges with a common vertex. These vertex
cocycles can be characterized in matroid terms only and, if the graph is 3-
connected, completely recovered from the cycle or cocycle matroid of the graph.
Our enlargement of the class of objects we call graphs is closed under sub-
contraction. Subcontraction of graphs is the prototype of the notion of matroid
minor. We see that the graphic matroids, as well as co-graphic matroids, are
closed under the taking of minors. This observation supplies half of the proof
of the following lemma.
LEMMA 3.10.1. Let M be a matroid. Then M is graphic (co-graphic} if and

only if each of the components of M is graphic (co-graphic).
Exercise 3.54. Complete the proof of this lemma.

In view of Lemma 3.10.1 and Exercise 3.40, we may restrict our attention
to biconnected graphs and connected matroids. Note that a biconnected graph
contains no loops, unless it consists of a single loop. The cycle matroid of a
graph consisting of a single loop is a matroid on a one element set with that
one element as a cycle. Hence, single loops are easily identified by their cycle or
cocycle matroids.
LEMMA 3.10.2. Let G = (V, E) be a biconnected graph, but not a single loop
and let M be its cycle matroid. Let C 1 , ••• , Cm be the vertex cocycles of M*.
Then:
a. each edge e E E lies in exactly two of the Ci;
3.10. GRAPHIC MATROIDS 85
b. the cycles of M are the minimal nonempty sets in
V = {Z ~ E I Z =f 0 and IZ n Gil =f 1 for all i = 1, ... , m}

PROOF. Since each edge has two end points, it lies in exactly two vertex
cocycles. It is also clear that if Z is the edge set of any elementary circuit of
G then IZ n Gil is 0 or 2. Hence the cycles of M belong to V. Next assume
that F E V and consider the subgraph G' = (V(F), F). By the definition of
V, each vertex in G' has valence at least two. A simple graph theory argument
demonstrates that a graph with this property must contain a cycle of M. Thus
the minimal elements in V are precisely the cycles of M. 0
THEOREM 3.10.1. Let M be a connected matroid on the set E and let C denote
the cycles of M. Then M is the cycle matroid of a biconnected graph if and only
if there exist subsets C1, ... , Cm of E satisfying:
a. Each edge e E E lies in exactly two of the Cii
b. C is the collection of minimal sets in
V = {Z ~ E I Z =f 0 and IZ n Gil~ 1 for all i = 1, ... , m}

PROOF. As we have just proved, if M is the cycle matroid of a graph, then
the conditions are met. Assume then that M is a matroid on E with cycles
denoted by C and that C 1 , ••• , Cm are subsets of E satisfying conditions a and
b. Let V = {1, ... , m} and identify e E E with the pairs (i,j) when e E Cinc3.
With this identification we have a graph G = (V, E). Furthermore Ct. ... , Cm
are the vertex cocycles of G. Let C' denote the cycles of the cycle matroid of G.
By Lemma 3.10.2, and by condition b of the hypotheses, C = C'. 0
As we alluded to above, the vertex cocycles may not be recoverable from the
matroid structure. That is, there may be other collections of cocycles which
satisfy condition a and b. In such a case there will be several graphs with the
"same" cycle and cocycle matroids. This is illustrated in the next exercise.
Exercise 3.55. List the vertex cocycles of the graphs in Figure 3.19.
a. Consider the graph in Figure 3.19a. Show that the following sets also
satisfy the two conditions for the vertex cocycles of a graph; C1 =
{et.e4}, C2 = {e1,ea,e7}, Ca = {e2,ea}, C4 = {e2,es,e5,e7}, and
Cs = {e4,es,e5}.
Draw the graph based on those sets using the construction in the proof
of Theorem 3.10.1. Find all other such collections of sets and construct
the associated graphs.
b. Prove that the vertex cocycles of the graph in Figure 3.19b are the only
collections of sets which satisfy both properties a and b of Theorem 3.10.1.
The first example discussed in Exercise 3.55 is not 3-connected while the sec-
ond example is 3-connected, or, in matroid terms, the cycle and cocycle matroids
a. b.
FIGURE 3.19.
of the first example are connected but not 2-connected while the cycle and cocy-
cle matroids of the second are 2-connected. The significance of this is explained
as follows: let M be the cycle or cocycle matroid of the graph G = {V, E) and
let C ~ E be a cocycle which is not a vertex cocycle, then the deletion of the
edges in C leaves a disconnected graph and a disconnected cycle or cocycle ma-
troid. In short, M(F-C) is not connected. If, on the other hand. Cis a vertex
cocycle, then one of the compon~nts of the graph obtained by deleting C is a
single vertex-undetectable in the matroid. Hence, if G is 3-connected and C is a
vertex cocycle, then G E-C consists of a biconnected graph plus an isolated ver-
tex and ME-Cis connected. Thus, if G is 3-connected, the vertex cocycles may
be characterized in matroid terms alone: they are the cocycles whose deletions
do not disconnect the matroid. Thus, if M is a 3-connected graphic matroid,
it is the cycle matroid of a unique graph (up to isomorphisms). This was first
proved by Whitney in a different form before he invented matroids.
THEOREM 3.10.2. Let G = {V, E) be a biconnected graph and let M be its

cycle matroid. Then G is planar if and only if M is co-graphic.
PROOF. Since this discussion is getting a bit off track, we only outline the
proof. First, if G is planar, embed it in the plane and construct the dual of G
with respect to this planar embedding. Note that the dual G* = ('W, E) can be
thought of as a graph with the same edge set as G. But, the vertex cocycles of
G* correspond to the cycles which bound faces. In fact the cocycle space of G*
is the cycle space of G and vice versa. Thus M is co-graphic.
Now suppose that M is co-graphic and let C 11 ••• , Cm be a possible collection
of vertex cocycles. These cycles in G can then be taken as the sets of edges
bounding the faces of a planar embedding of G. 0
Now, the original theorem of Whitney, a generalization of which we discussed

above, implies the well known result:
A 3-connected planar graph has a unique embedding in the plane
3.11. Abstract Rigidity Matroids. The purpose ofthis section is to gather
together all of the results about rigidity matroids which we have already proved
and to include a few additional general results about these matroids. We sum-
marize everything that went before which will be used in that which follows.
3.11. ABSTRACT RIGIDITY MATROIDS 87
Let V be a finite set and m a positive integer. Let K denote the set of all
unordered pairs of elements of V; thus, (V, K) is the complete graph on V.
A matroid on K with closure operator (·} is called an m-dimensional abstract
rigidity matroid for V if it satisfies the additional conditions C5 and C6 listed
below.
C5: If E, F f; K and IV(E) n V(F)I < m, then {E U F) f; (K(V(E)) u
K(V(F))).
IfF is empty, we have the special case that if E f; K, then {E) f; K(V(E)).
We say that E f; K is rigid if {E) = K(V(E)).
C6: If E, F f; K are rigid and IV(E) n V(F)I 2::: m, then {E u F) is
rigid.
One of the fundamental results of Chapter 2 was that infinitesimal rigidity
of frameworks on a fixed set of points generally embedded in m-space yields an
m-dimensional abstract matroid: Let p be an embedding of V in real m-space.
Let E f; K and consider the framework with p(V(E}} as points and include all
rods which correspond to edges in E. Now define {E) to be E plus all pairs
whose distance does not change infinitesimally, under infinitesimal motions of
the framework, that is, all pairs (i,j) such that (Pi - P;) * (p~ - pj) = 0 for
all infinitesimal motions p'. This closure operator defines the m-dimensional
infinitesimal rigidity matroid determined by p, which we denote by .1'(p). Ifp is
a generic embedding, this matroid is called "the" m-dimensional generic rigidity
matroid on V and we denote it by Qm(IVI).
Section 2.5 contained the following important results. These are renumbered
here for easy reference later.
THEOREM 3.11.1 (SEE 2.5.6}. Let the finite set V be given and let r be the
rank function of an m-dimensional abstract rigidity matroid for V.
a. Let E f; K(V), io E (V- V(E)), and let it, ... , i~c E V(E) with k:::; m.
Then r(E U {(io, i1), ... , (io, i~c)}) = r(E) + k.
b. For any U f; V:
(k)
{ ~k- if lUI = k :::; m + 1;
r(K(U)) = (mt.1) if lUI = k 2::: m + 1.
THEOREM 3.11.2 (LAMAN'S CONDITION, SEE THEOREM 2.5.4). Let V be a
finite set and let E f; K = K (V) be an independent set in an m-dimensional
abstract rigidity matroid for V. Then, for all F f; E with IV (F) I 2::: m, we have
that IFI:::; miV(F)I- (mi 1).
Let Am be am m-dimensional abstract rigidity matroid for V. A set E f;
K = K(V) which is both independent and rigid is called an isostatic set.
THEOREM 3.11.3 (SEE LEMMA 2.6.1). Let V be a finite set, let E f; K =
K (V) and consider an m-dimensional abstract rigidity matroid for V.
a. Assume that IV(E)I :::; (m + 1). Then E is independent; furthermore, E
is rigid, and hence isostatic, if and only if E = K(V(E)).
b. Assume that IV(E)I ~ (m + 1) and that E is isostatic. Then:

(i) lEI= miV(E)I- (mi 1);
(ii) Each vertex of (V(E), E) has valence at least m;
(iii) (V(E), E) has a vertex with valence less than 2m.
c. Assume that IV(E)I ~ (m + 1). Then, if any two of the following con-
ditions hold, all three hold and E is isostatic.
(i) lEI= m!V(E)I- (mi 1);
(ii) E is independent;
(iii) E is rigid.
This theorem has a useful corollary not included in Chapter 2, but included
here.
COROLLARY 3.11.1. Let the finite set V be given, let E ~ K = K(V), and
consider an m-dimensional abstract rigidity matroid for V. Then E is rigid if
and only if either IV(E)I ~ m + 1 and E = K(V(E)) or IV(E)I ~ m + 1 and
r(E) = m!V(E)I- (~).
Exercise 3.56. Prove this corollary.

If M and M' are two matroids on the same base set and if each set which
is independent in M' is also independent in M, then we say that M majorizes
M' and write M ~ M'.
THEOREM 3.11.4 (SEE THEOREM 2.5.5). Let the finite set V and the integer
m be given, and let p : V -+ Rm be a geneml embedding. Then the m-dimensional
infinitesimal rigidity matroid on p satisfies gm(V) ~ F(p).
This fact leads to:
CONJECTURE 3.11.1 (SEE THEOREM 2.5.1). Let the finite set V and the in-
teger m be given. If Am is any m-dimensional abstmct rigidity matroid on V,
then gm(V) ~ .Am.
Exercise 3.57. Consider the case in dimension 1.

(i) Define a matroid on K(V) that satisfies C5 but not C6. Does such a
matroid necessarily majorize g1 (V) ?
(ii) Define a matroid on K(V) that satisfies C6 but not C5. Does g 1 (V)
necessarily majorize such a matroid?
In Chapter 2 we introduced the concepts of 0-extensions and !-extensions of
sets in an m-dimensional abstract rigidity matroid. Let Am be an m-dimensional
abstract rigidity matroid for V; let E ~ K = K(V); and let i E (V- V(E))
while U ~ V(E). If lUI = m, F = E U {(i,j) I j E U} is called a 0-extension of
E. If lUI= m+ 1 and (j,j') E EnK(U), then F = (E- (j,j')) U {(i,j)li E U}
is called a !-extension of E.
THEOREM 3.11.5 (SEE THEOREMS 2.6.1 AND 2.6.2}. Let V be a finite set and
let Am be an m-dimensional abstract rigidity matroid for V. Let E ~ K = K (E).
a. Let F be a 0-extension of E. Then F is independent (rigid, isostatic) if
and only if E is independent (rigid, isostatic).
b. Assume Am is generic and let F be a !-extension of E. Then F is
independent (rigid, isostatic) if and only if E is independent (rigid, iso-
static).
This last result was used in two exercises to study bipartite graphs. We
summarize and extend the conclusions of these exercises:
THEOREM 3.11.6 (SEE EXERCISES 2.38 AND 2.39}. Let V be a finite set and
let Am be an m-dimensional abstract rigidity matroid for V. Let Kab be the
subset of K = K(E) consisting of the edges of a complete bipartite subgraph with
vertex classes of sizes of a and b with a::::; b.
a. If a ::::; m then Kab is independent.
b. If Am is generic, a = m + 1 and b = {mil), then E is isostatic.
c. If Am is generic, a = m + 1 and b 2:: {mil), then Kab is rigid.
In Example 3.1.2 of the first section of this chapter, we introduced the com-
plete connectivity matroid M(Kn) on the edge set of the complete graph (V, K),
where V is an n-set. One easily verifies that the closure operator for this matroid
satisfies condition C5 with m = 1. It follows that E ~ K is rigid if and only
if (V(E), E) is connected and that condition C6 also holds form = 1. Thus,
M(Kn) is a !-dimensional abstract rigidity matroid for V. It is also easy to
verify that an edge set E is independent in this matroid if and only if (V(E), E)
is a forest. Thus, E is isostatic if and only if (V(E), E) is a tree. Finally, the
cycles of this matroid are the edge sets of elementary circuits of (V, K). These
and further facts about this fundamental matroid are summarized in the next
theorem.
THEOREM 3.11.7. Let M be the complete connectivity matroid on the edge set
of the complete graph (V, K), where V is ann-set.
a. M is a !-dimensional abstract rigidity matroid for V, i.e. M = gl (lVI}.
b. E ~ K is independent if and only if (V(E), E) is a forest.
c. E ~ K is rigid if and only if (V(E), E) is connected.
d. E ~ K is isostatic if and only if (V(E), E) is a tree.
e. E ~ K is a basis if and only if (V, E) is a spanning tree.
f. E ~ K is a cycle if and only if (V(E), E) is an elementary circuit.
g. r(E) = lVI- c, where c is the number of components of (V(E), E).
Exercise 3.58. Prove the above Theorem.

The importance of this matroid is made clear in the next theorem.
THEOREM 3.11.8. If A1 is any !-dimensional abstmct rigidity matroid for V

then M is the complete connectivity matroid on the edge set of the complete
gmph (V,K(V)).
PROOF. The result will follow from the previous theorem once we show that
E ~ K is isostatic in A 1 if and only if (V(E), E) is a tree. We proceed by
induction on IV(E)I, the conclusion being obvious if IV(E)I < 3. Let E ~ K
and assume that IV(E)I ?:: 3.
Suppose that E is isostatic in A 1 . By Theorem 2.6.1b, (V(E), E) has a
pendant vertex. By Theorem 2.6.2a, the deletion of this pendant vertex results
in an isostatic set with smaller support. By the induction hypothesis this isostatic
set is a tree. Since attaching a pendant vertex to a tree yields a tree, (V(E), E)
is a tree.
Suppose that E is the edge set of a tree. Then E contains a pendant edge e
and E - e is also a tree. By the induction hypothesis, E - e is isostatic and, by
Theorem 2.6.1a, E is isostatic. D
It follows at once from this result that the complete connectivity matroid is in
fact 'h(IVI), the !-dimensional generic rigidity matroid for V. The 0-extensions
played a central part in the above proof and, since (h (n) is generic, !-extensions
of isostatic sets are isostatic. In fact a very general extension property holds for
this matroid as you will now show.
Exercise 3.59. Consider {h(n). Let E ~ K be isostatic and let i E V(E)
have valencek > 1. LetU = {h, ... ,j~~:} be the set of neighbors ofi in (V(E),E)
and let F be the edge set of any tree with vertex set U. Then
(E-{(i,j1), ... ,(i,j~~:)})UF
is isostatic. Let E ~ K be isostatic; let i E V- V(E); and let (V(F), F) be any

subtree of (V(E), E). Then (E- F) U {(i,j) I j E V(F)} is isostatic.
Let Am be any m-dimensional abstract matroid for V. If B ~ K = K(V) is
a basis of Am then its closure is all of K. Hence bases are also rigid, i.e. bases
are isostatic. A second useful observation can be made about the rank function
of .Am. Let E ~ K with IV(E)I ?:: m and let F be a maximal independent
subset of E. We first show that V(F) = V(E). Clearly, V(F) ~ V(E). Now
suppose that i E V(E)- V(F) and let e be an edge in E having e as end point.
It follows at once from Theorem 3.11.6a that F U {e} is independent contrary
to the maximality of F. Thus, V(F) = V(E). Combining this with Laman's
condition, Theorem 3.11.2, we get:
r(E) = r(F) :::;; miV(F)I- ( m+

2
1) = miV(E)I- (m+ 1) 2
For easy reference we state these two observations as a lemma:

LEMMA 3.11.1. Let V be a finite set and let Am be any m-dimensional abstract
matroid for V.
a. If B ~ K = K(V) is a basis of Am then it is rigid and isostatic.
b. If r is the rank function of Am and E ~ K with IV(E)I ~ m, then
r(E):::;; miV(E)I- (mt 1).
The next two results contain a summary of facts we may easily conclude about
the cycles of abstract rigidity matroids.
THEOREM 3.11.9. Let the finite set V be given and let Am be anm-dimensional
abstract rigidity matroid for V. Let D ~ K = K(V) be a cycle of .Am. Then
a. IV(D)I ~ m + 2, with equality if and only if D = K(V(D)).
b. IDI- 1 = r(D) :::;; miV(D)I- (mt 1), with equality if and only if D is
rigid.
c. Each vertex of (V(D), D) has at least m + 1 neighbors.
d. For any edge e ED, V(D- e)= V(D) and (D-e)= (D).
e. (V(D), D) has a vertex with valence less than or equal to 2m.
f. If m > 1, (V(D), D) has a vertex with valence less than 2m.
PROOF. Conclusion a follows at once from Theorem 3.11.5a; while the in-
equality in conclusion b follows from Theorem 3.11.2 and the fact that deleting
any edge must yield an independent set. Let i be a vertex of (V(D), D) and let
F be the set of edges in D which have i as end point. Clearly, D - F is indepen-
dent. Suppose that IFI:::;; m. Then by Theorem 3.11.5a, D = (D-F)UF would
be independent. We conclude that IFI ~ m + 1. The first equality in d follows
from c. Since D is a minimal dependent set, e E (D-e); thus D ~ (D-e) and
the second equality of d follows. Since D - e is independent,
r(D) = r(D- e)= IDI-1:::;; miV(D)I- (m; 1) = r(K(V(D))).
Thus D is rigid ((D) = K(V(D))) if and only if equality holds throughout this
string of inequalities (completing the proof of condition b). Finally, the we denote
the average valence in (V(D), D) by v, we have, using b above:
2IDI m 2 +m- 2
2
v = V(D) :::;; m- IV(D)I
The last two conclusions are now obvious. D
THEOREM 3.11.10. Let Am be them-dimensional infinitesimal rigidity ma-

troid on V induced by p: V-+ Rm and let E ~ K(V). Then
a. E is independent if and only if dn(E) = 0, i.e., if and pnly if S(E), the
space of resolvable stresses forE, has dimension 0.
b. E is a cycle of Am if and only if each subset of E is independent and
dn(E) = 1.
The last theorem that we will include here relates some of the properties of
an abstract rigidity matroid with the connectivity of the underlying graph.
THEOREM 3.11.11. Let V and Am, an m-dimensional abstract rigidity ma-
troid for V be given and let E ~ K = K (V).
a. If (Am) IE is connected then (V(E), E) is biconnected.
b. If E is rigid then (V(E), E) is (m +I)-connected.
Exercise 3.60. Prove this theorem.
Exercise 3.61. Aside from (h(n), which of the ten example matroids of this
chapter are abstract rigidity matroids '?
Exercise 3.62. Are restrictions (contractions, minors) of abstract rigidity
matroids themselves abstract rigidity matroids '?
Exercise 3.63. Prove the following: Let G = (V, E) be a graph on n vertices
containing a vertex v of valence (n- 1) and let M be the restriction of the 2-
dimensional generic rigidity matroid for V toE, then the contraction of M to
G- v is the !-dimensional generic rigidity matroid on G- v.
Let (V, E) be a graph. An edge setH ~ E is said to be complete if (V(H), H)
is a complete subgraph; H ~ E is a clique of {V, E) if it is a maximal complete
subset of E.
Exercise 3.64. Let Am be any m-dimensional abstract rigidity matroid on
the vertex set V. Let E ~ K(V) be a closed set of Am and let H1, ... , H~c be the
cliques of E. Prove that
a. E = H1 U ... U H~c,
b. V(E) = V(Hl) U ... U V(H~c),
c. IV(Hi) n V(H;)I < m for alll ::::; i < j ::::; k, and
d. r(E) ::::; r(Hl) + ... + r(H~c).
Chapter 4. Linear and Planar Rigidity
4.1. Abstract Rigidity in the Plane. Theorem 2.1.1 showed that in di-
mension 1 the concepts of rigidity, infinitesimal rigidity and generic rigidity are
equivalent. In Theorem 3.11.8 it was shown, in fact, that the generic rigidity
matroids are the only examples of abstract rigidity matroids in dimension 1 and
that these matroids coincide with the connectivity matroids. There are, by con-
trast, many abstract rigidity matroids in dimension 2, whose properties are the
subject of this chapter, with particular focus on (12{n). Those abstract rigidity
matroids which arise as infinitesimal rigidity matroids of generally embedded
vertex sets are often dependent on the geometry of the embedding. The generic
rigidity matroid, however, can be described combinatorially. The following the-
orem summarizes the main results of the first few sections of this chapter.
THEOREM 4.1.1. Let A2 be a 2-dimensional abstract rigidity matroid on n

vertices. The following are equivalent:
(i) A2 = (12{n);
(ii) A 2 has the 1-e:&tendability property;
(iii) The independent sets of A2 are those sets which satisfy Laman's condi-
tion;
(iv) All cycles of A2 are rigid;
(v) For any closed set E of A2 with cliques E 11 ••• , Ek, r(E) = r(El) +
· · · +r(Ek)i
(vi) r(E) = minE~= 1 (2IV(Ei)l- 3), where the minimum is taken over all
collections {Ei} of nonempty sets such that E = UEi.
While developing this theory of planar rigidity, we will also consider which
results about graph connectivity have analogous formulations in terms of rigidity
in dimension 2.
In the closing section of Chapter 3, we listed all of the results produced to
date about abstract rigidity matroids. Here we summarize those results as they
apply in dimension two. It is left to the reader to verify the accuracy of this
summary and to fill in a few missing, but elementary proofs.
93
94 CHAPTER 4. LINEAR AND PLANAR RIGIDITY
Let (V, K) be a complete graph, and consider a 2-dimensional abstract rigidity

matroid A 2 on K, i.e. a matroid on K whose closure operator satisfies the
following conditions for any two edge sets E, F ~ K.
C5: If IV(E) n V(F)I ~ 1 then (E u F) ~ K(V(E)) u K(V(F))
C6: If (E) = K(V(E)) and (F) = K(V(F)) and IV(E) n V(F)I ~ 2,
then (E U F) = K(V(E U F)).
From Theorem 3.11.1b, we see that, for any subset U ~ Von two or more
points, r(K(U)) = 21UI - 3. Thus, A2 has rank 2IVI - 3. We also note that
E E K is rigid if and only if r(E) = 2IV(E)I- 3. We conclude that all bases for
A2 have cardinality 2IYI- 3 and are rigid.
In dimension two, Laman's condition (Theorem 3.11.2) has the following form:
if E ~ K is independent, then IFI ~ 2IV(F)I- 3, for each nonempty subset F
of E.
Combining several of the results, we see that the smallest cycles of A2 are
the edge sets of the complete subgraphs on four vertices. We also note that any
other set of six edges or any set of five or fewer edges is independent.
Attaching a vertex by one or two edges to a subgraph with an indepen-
dent edge set results in a subgraph which also has an independent edge set
(0-extensions). If A2 is generic then 1-extensions of independent sets are inde-
pendent. A 1-extension of (V(E), E) is obtained by deleting an edge from E and
attaching a vertex x ¢ V(E) to the endpoints of the deleted edge and one other
vertex in V(E).
The properties of isostatic sets are listed in Theorem 3.11.3. Let E be isostatic
and consider (V(E), E). First of all lEI = 2IV(E)I - 3. We also note that each
vertex of (V(E), E) has valence at least two and that at least one vertex of this
graph has valence less than four. A larger isostatic set can be obtained from
(V(E), E) by attaching a vertex x ¢ V(E) to two vertices in V(E), in other
words by performing a 0-extension of (V, V(E)). In the generic case, one can
make a larger isostatic set by deleting an edge from E and attaching a vertex
x ¢ V(E) to the endpoints of the deleted edge and one other vertex in V(E), in
other words by performing a 1-extension of (V, V(E)).
The properties of cycles reported in Theorem 3.11.9 and suitably restricted to
dimension two are listed next. Let D be a cycle of A2. Then IDI ~ 2IV(D)I- 2
with equality if and only if D is rigid. Each vertex of (V(D), D) has valence at
least three and at least one vertex has valence exactly three.
Exercise 4.1. Show that each cycle in A2 has at least four vertices of valence
three.
Exercise 4.2. Prove that, for each n ~ 5, there is only one abstract rigidity
matroid, namely g2 (n).
Exercise 4.3. Consider the general embedding Pl = (0,0), P2 = (1,-1),
Pa = (-1, -1), P4 = (0, 3), Ps = (1,2), and P6 = (-1,2) of the complete graph
on 6 vertices. Show that the infinitesimal rigidity matroid of this embedding
4.1. ABSTRACT RIGIDITY IN THE PLANE 95
is not isomorphic to the one obtained after changing the coordinates of P6 to

(-2,3). Hint: Find a basis of one matroid, which is dependent in the other, in
fact a nonrigid cycle of the other.
This last example illustrates the fact that there are 2-dimensional infinitesimal
rigidity matroids which are not generic. In Exercise 2.12 you examined another
such example: embed V = {1, ... , 6} as the vertices of a regular hexagon, then,
in the resulting 2-dimensional infinitesimal rigidity matroid, the edges of the
hexagon plus the three (main) diagonals, a copy of K 3 ,3 , is dependent. This is a
special case of a much stronger result due to Bolker and Roth [12] which states
that, if V = { 1, ... , 6} is embedded on a conic in R 2 , then each copy of K 3 •3
in this 2-dimensional infinitesimal rigidity matroid is dependent. Since all of
these non-generic examples are infinitesimal rigidity matroids, it is very natural
to ask: Does there exist a 2-dimensional abstract rigidity matroid which is not
an infinitesimal rigidity matroid? The answer to this question is "yes"; but, the
verification of this will be put off until we have developed a few more tools for
the study of 2-dimensional rigidity.
We will see in the next section that the 2-dimensional generic rigidity matroid
is the unique maximal 2-dimensional abstract rigidity matroid on n vertices. It
would also be of interest to describe the minimal abstract rigidity matroids. We
know that all abstract rigidity matroids on n vertices have rank 2n - 3 and that
those edge sets obtained from a single edge by a sequence of 0-extensions must
therefore be bases. Any 2-dimensional abstract rigidity matroid on n vertices,
and hence any minimal 2-dimensional abstract rigidity matroid on n vertices,
must include these edge sets as bases. In general, however, additional bases are
necessary, as illustrated in Figure 4.20. B1 and B2 are b.•ses for any abstract
Ba
FIGURE 4.20.
rigidity matroid on seven vertices since they may be obtained by five 0-extensions
on a single edge. The basis axioms assert that B1 - e can be augmented to a
basis with an edge from B 2 , which, since tetrahedra are dependent, must be
Ba = B1 - e + f. Thus Ba is also a basis in any 2-dimensional abstract rigidity
matroid, but B3 is not the result of a sequence of 0-extensions of an edge since
it has no vertex of valence 2.
Exercise 4.4 (Open Question). Characterize the minimal abstract rigidity

matroids in dimension 2.
4.2. Combinatorial Characterizations of Q2(n). In Exercise 4.3 above

we saw that the generic rigidity matroid is not the only 2-dimensional abstract
rigidity matroid. Our goal is to characterize generic rigidity in the plane and our
first step is to relate 1-extendability to Laman's condition.
LEMMA 4.2.1. Let V be a finite set and let A 2 be any 2-dimensional abstract
rigidity matroid for V that satisfies the 1-extendability property. Then E ~ K is
independent in A2 if and only if E satisfies Laman's condition for dimension 2.
PROOF. By Theorem 2.5.4 we need only prove that every edge set E which
satisfies Laman's condition for dimension 2 is independent in A2. We give a proof
by contradiction. Suppose that E is a vertex and edge minimal dependent set
which satisfies Laman's condition. By minimality, E has no vertex of valence 1 or
2, so E has a vertex v of valence 3. Let F = E- v and denote the set of neighbors
of v by N = {x, y, z}. We have K(N) g; F, since otherwise K(N + v) ~ E and
tetrahedra violate Laman's condition. Moreover, K(N) ~ (F), since otherwise
we have F + (:c, y), say, is independent and E is a 1-extension ofF+ (:c, y) and
so is also independent. ForeE K(N), define Xe ~ F to be the minimal set such
that e E (Xe}· If Xe =F e, then e ¢ Xe and Xe + e is a minimal set violating
Laman's Condition. So IXel = 2IV(Xe)l - 3, hence {Xe} = K(V(Xe)). Thus
(Xe 1 UXe 2 UXe 3 } = (Xe 1 UXe 2 UXe 3 U{et.e2,e3}} = K(V(Xe 1 UXe 2 UXe 3 )) by
axiom C6. We conclude that (Xe 1 U Xe 2 U Xe 3 ) is rigid. But, Xe 1 U Xe2 U Xe 3 is
also independent. Hence, we have IXe 1 UXe 2 UXeal = 2IV(Xe 1 UXe 2 UXe 3 )l-3.
But then it follows that (Xe 1 U Xe 2 U Xe 3 ) + v violates Laman's Condition, a
contradiction. D
The preceding lemma serves to characterize Q2(n) combinatorially, and the
following theorems are immediate consequences.
THEOREM 4.2.1 (LAMAN'S THEOREM). Let V be a finite set and let Q2(n) be
the 2-dimensional generic rigidity matroid for V. Then E ~ K is independent
(isostatic} in Q2(n) if and only if E satisfies Laman's condition for dimension 2
(and lEI = 2IV(E)I- 3}.
THEOREM 4.2.2. Q2(n) is the unique maximal2-dimensional abstract rigidity
matroid on n vertices.
THEOREM 4.2.3. Let V be a finite set and let A 2 be any 2-dimensional abstract
rigidity matroid for V. Then A 2 is the 2-dimensional generic rigidity matroid
for V if and only if A2 satisfies the 1-e:ctendability condition.
Exercise 4.5. Fill in the proofs of these three theorems.

These last results show the equivalence of the first three characterizations in
Theorem 4.1.1.
Actually, in dimensions 1 and 2, the generic rigidity matroids also satisfy a
"k-extendability" condition for all k, as you will now show.
4.2. COMBINATORIAL CHARACTERIZATIONS OF g2(n) 97
Exercise 4.6. Considergm(n) on V = {1, ... , n} wherem = 1,2. Let (V, E)

be a graph without isolated vertices, let v be a vertex of V, and S the set of
neighbors of v. Let (U, F) be the subgraph obtained from (V, E) by deleting v and
the edges containing it. Assume that, for every T ~ S with ITI > m, there exists
a subset FT of K(T) such that 1FT I = ITI - m and FT n F = 0 and that F U FT
is independent. Then E is independent.
The Laman characterization of the isostatic sets in g2 (n) is a powerful tool
in the study these isostatic sets and, in view of the maximality of g2(n), tells us
much about the structure of isostatic sets in an arbitrary 2-dimensional abstract
rigidity matroid.
LEMMA 4.2.2. Let A2 be an abstract rigidity matroid for the vertex set V and
let I be an isostatic set inK= K(V). Then:
a. (V (I), I) is 2-connected.
b. (V(I), I) is 2-edge-connected.
c. If the removal of two edges disconnect (V(I), I) then the two edges have
a common endpoint v, and v has valence 2 in I.
Exercise 4. 7. Prove this lemma.
Exercise 4.8. Let g2(n) be the generic rigidity matroid for the vertex set V.
Let h and I2 be isostatic sets in K = K(V) such that V(h) n V(I2) = 0. For
i = 1,2,3, let (xi, Yi) E K where Xi E V(h) and Yi E V(I2). F'u.rthermore,
assume· that l{x1,x2,xs}l > 1 and HY1tY2.Ys}l > 1. Show that E = I1 U I2 U
{(xlt Y1), (x2, Y2), (xs, Ys)} is isostatic.
Exercise 4.9. Show directly that the collection of edge sets of a graph G,
which satisfy Laman's condition in dimension 2 is the collection of independent
sets of a matroid.
It is natural to ask if the edge sets that satisfy Laman's inequalities for dimen-
sion m > 2 also form the collection of independent sets of some matroid, since
such a matroid would be a logical candidate for the generic rigidity matroid in
dimension m. Unfortunately we have the following negative result.
THEOREM 4.2.4. The collection of edge sets that satisfy Laman's condition for
dimension m cann~t be the collection of independent sets of any matroid on K
i/m"?_3.
PROOF. Assume there exists a matroid M on K whose independent sets
are the subsets E of K satisfying Laman's condition for dimension m, that is
IE'I ~ miV(E')I- {mt 1) for all subsets E' ~ E with IV(E')I "?. m. Let 01 and
0 2 be cycles of M such that they have exactly one edge e in common and their
supports intersect only in the endpoints of e. Then (01 U 02) - e must contain a
subset E violating Laman's condition for dimension m. Let Ei =(En Oi) and
let ~ denote the cardinality of the support of Ei. Then lEI = IE1I + IE2I ~
m(n1 +n2)-m(m+1), since both Ei's satisfy Laman's condition for dimension m.
Since E violates Laman's condition and the supports of E1 and E2 have at most
two vertices in common, we have m{n1+~)-m(m+1) > m(n1 +n2-2)- m(~+l).
The last inequality is only satisfied if m < 3. 0
In the case m = 3, one may observe that the proposed rank bound, r(E) :::;
3jV{E)I-6 fails for single edges. On the other hand, the bound r(E):::; 3jV{E)I-
5 works for single edges, as you will show:
Exercise 4.10. Let (V, K) be a complete graph and let
I= {E: IFI :5 3IV(F)I- 5 for all0 C F ~ E}
Prove that I is the collection of independent sets of a matroid on K.

4.3. Cycles in ~12(n). Laman's characterization of {#2{n) allows us to com-
binatorially identify the independent sets, from which it is easy to deduce a
characterization for the cycles of Ch (n). We have already used the following fact
within the proof of Lemma 4.2.1, but we repeat it here for clarity and future
reference.
THEOREM 4.3.1. An edge set C is a cycle in g2(n) if and only if IE(C)I =

2jV{C)I- 2 and IFI :::; 2jV{F)I- 3 for every proper subset F of E(C).
PROOF. An edge set satisfying these conditions is a cycle by Laman's Theo-

rem.
Let C be a cycle g2 (n). Every subset of C is independent and so satisfies
Laman's condition. C itself must violate Laman's condition so that 2IV(C)I-3 <
ICI = IC- el + 1 :5 2IV(C- e)l - 2 = 2IV(C)I - 2 for any edge e E C. So
ICI = 2jV(C)- 2. 0
The conditions listed in this result, along with the condition that all vertices
of a cycle have valence at least 3, enable us to list all small generic cycles. In
Figure 4.21, all cycles in g2(n) having fewer than 7 vertices are described.
FIGURE 4.21. The generic rigidity cycles on up to 6 vertices.
Exercise 4.11. Verify that indeed all the generic rigidity cycles on six or
fewer vertices are described in Figure 4.21.
Exercise 4.12. Let A2 be a 2-dimensional abstract rigidity matroid. Show
that A2 admits no nongeneric cycles on fewer than si:1: vertices. Show further
that the only possible nongeneric cycles on six vertices are K 3 ,3 and the prism
(as described in Exercise 4.3).
4.3. CYCLES IN Q2(n) 99
Exercise 4.13. Let Wn denote then-wheel, that is graph obtained from an

n-gon by attaching a single new vertex which is adjacent to all the vertices of the
n-gon. In Figure 4.21 the first three graphs are wheels. Show that Wn is a cycle
in any abstract rigidity matroid in which it is contained.
It follows immediately from Theorem 4.3.1 that generic cycles are rigid, in
fact over braced in a homogeneous way, namely the removal of any edge leaves
a rigid graph, whose edge set is independent in g2 (n). In the next theorem we
show that the rigidity of the cycles in fact characterizes g2 (n) and extend the
circle of equivalences from Theorem 4.1.1 to include the fourth condition.
vertices. Then A2 = g2 (n) if and only if all cycles in A2 are rigid.
PROOF. Suppose A2 = Q2(n), we see from Theorem 4.3.1 that a cycle C in
Q2(n) has rank 2IV(C)I- 3 and is thus rigid.
Suppose, on the other hand, that every cycle of A2 is rigid. Consider a cycle
C. C and C - e have the same closure for all edges e in C. So if C is rigid,
C- e is independent and rigid, and IC- el = r(C- e) = 2IV(C- e)l - 3 =
2IV(C)I- 3. Thus C violates Laman's condition. Hence, no set which satisfies
Laman's condition can contain a cycle. 0
The following lemma lists some simple but useful properties of cycles in Q2 (n).
LEMMA 4.3.1. Let C be a cycle in Q2(n). Then:
a. (V (C), C) is 2-connected.
b. (V(C),C) is 3-edge-connected.
c. If the removal of 3 edges disconnect (V (C), C), then the 3 edges have a
common endpoint v, and v is of valence 3 in C.
Exercise 4.14. Prove this lemma.
Exercise 4.15. Show that a !-extension of a cycle is a cycle.
Exercise 4.16. Referring to Figure 4.21, show that the second cycle in the
list is a !-extension of the first cycle. Show that each of the last four cycles is a
!-extension of the second cycle.
Exercise 4.17 (Open Question). Can all 3-connected cycles in Q2(n) be
obtained from the tetrahedron by a sequence of !-extensions? {Posed by R. Con-
nelly}.
If this could be proved to be true, it would be a very useful tool in investigating
the structure of generic cycles. One might suspect that an even stronger result
is true: pick any vertex v of valence 3 in a cycle C from Q2(n), is C a !-extension
"at v"? This stronger conjecture is false as you will now show.
Exercise 4.18. Consider C, the fifth cycle in Figure 4.21. Let v be the upper
right-hand vertex of C. Show that there is no !-extension giving C with v as the
new vertex being attached.
Exercise 4.19. Let Ct and C2 be cycles in gm(n) such that E(Ct)nE(C2) =

{e} and IV(E1 ) n V(E2)I = 2, i.e. Ct and C2 have exactly one edge and no ver-
tices other than the endpoints of this edge in common. Prove that the symmetric
difference of 0 1 and C2 is an m-cycle for m = 1, 2. What can be said about
other values of m? What happens if Ct and C2 intersect in one edge but more
than two vertic!es?
Exercise 4.20. Show that, if a cycle in Q2(n) is not 3-connected, then it is
the symmetric difference of two cycles in Q2 (n) which have exactly one edge and
its endpoints in common.
Exercise 4.21. Let C be a cycle in Q2(n) such that (V(C), C) is planar.
Show that the edge set of the geometric dual of(V(C),C) is also a generic cycle.
4.4. Rigid Components of Q2(G). In this section we will describe the
closed sets in Q2(n). We will also show that, in complete analogy to the !-
dimensional case, the restriction of Q2(n) to a particular subgraph G of Kn can
be written as a direct sum of restrictions of Q2(n) to the maximal rigid subgraphs
of G.
Let G be a graph. We regard Gas a subgraph of Kn for some n > IV(G)I and
denote the restriction of Q2(n) to E(G) by Q2(G). Since Q2(n) is a submatroid
of Q2(m) form> n, this definition is independent of n, and we may define a set
F in Q2(G) to be rigid if r(F) = 2IV(F)I - 3, where r is the rank function of
Q2(n) restricted to E(G).
A maximal rigid subgraph of G is called an r-component. It is an immediate
consequence of Axiom C6 that two r-components have at most one vertex in
common, hence the r-components may be regarded as a partition of the edges of
G. A non-rigid graph has at least two r-components.
Exercise 4.22. Show that a closed set in g2 (G) cannot contain three r-com-
ponents each two of which, but not all three, share a vertex.
Since the cycles of g2 (G) consist of those cycles of g2 (n) which are completely
contained in E(G), and since cycles are rigid by Theorem 4.3.2, it follows that
every cycle of Q2(n) is completely contained in some rigid component. This gives
the following theorem.
THEOREM 4.4.1. Let G be a graph and let E 1 , ••• , E1c be the rigid components
of G. Then Q2(G) = Q2(Gt) E9 ... E9 Q2(G~c), where Gi = (V(Ei), Ei)·
If G' = (V', E') is a subgraph of G, and G~ = (~', ED are the rigid compo-
nents of G', then the closure of E' in Q2(G) is E' together with all edges in E
both of whose endpoints lie in some G~, and the closure of E' in Q2 (n) is the
union of the completions of the components E~. This leads to the following char-
acterization of Q2(n), bringing the fifth condition into the circle of equivalences
for Theorem 4.1.1.
4.4. RIGID COMPONENTS OF Q2(G) 101

vertices. Then A2 = ~h(n) if and only if for any closed set E with cliques
Ell ... , Ek we have r(E) = r(E1) + · · · + r(Ek)·
PROOF. Necessity follows from Theorem 4.4.1. Suppose r(E) = r(E1 ) +

· ·· + r(Ek)for any closed set E with cliques Ell ... , Ek. Let C be a cycle of
A2 with whose closure has rigid components E 1 , .•. Ek, k > 1. Then r( C) ~
L:r(Ei n C) ~ L:r(Ei) = r((C)) = r(C), so, since every subset of a cycle is
independent, r(C) = L:r(Ei n C) = E lEi n Cl = ICI, a contradiction. Thus
cycles in A2 are rigid and A2 = g2(n) by Theorem 4.3.2. D
It follows from this result that, if A2 is a 2-dimensional abstract rigidity

matroid which is not generic, then it must contain a closed set violating the the
equality in this theorem. You will demonstrate this in the next two exercises.
Exercise 4.23. Consider V = {1, ... , 6} embedded on a conic. Then, as we
have noted, each copy of Ka,a in K(V) is a cycle. Show that Ka,a is closed and,
therefore, that its r-components are simply its edges. Verify that the equality in
the theorem is violated for Ka,a·
Exercise 4.24. Show that the infinitesimal 7"igidity matroid of the framework
in Exercise 4.3 has a closed set E with 9 edges, r(E) = 8, whose cliques {Ei}
satisfy L:r(Ei) = 9.
In many ways, the r-components are analogous to the connected components
of a graph. However, the two concepts do have some important differences: The
removal of any edge from a graph increases the number of connected components
by at most one. By contrast, the removal of an edge from a rigid graph can
result in a graph with many r-components. For example, if G is a quadrilateral
with 1 diagonal, then the removal of the diagonal results in a graph with four
r-components, all of which are single edges. Removing any other edge of G
leaves two r-components, a triangle and a single edge. It is not difficult to show,
however, that deleting an edge from a minimally rigid graph in the plane always
yields an even number of r-components.
Exercise 4.25. Show that a graph with one degree of freedom has an even
number of r-components. For every natural number k give an example of an
isostatic set I with the property that for some e E I, I - e has 2k r-components.
The number of r-components does not give us any information on the rank of
a given set. Of course, we could compute the rank by summing the ranks of the
rigid components. Sometimes it is of interest to know how many additional edges
are needed to achieve rigidity of (V(E), E). Recall that the degree of freedom of
an edge set E, df{ E), in g2 (n) is defined by
df(E) + r(E) = 2jV{E)I- 3.

Exercise 4.26. A graph G = (V, E), of degree of freedom 1, with 2k r-

components, each of which has ~ vertices, has
2k
lVI =I:~ - 3k' + 2.
i=l
Exercise 4.27. Prove that a tree with k edges has degree of freedom k -1 in
the plane.
Using the direct sum decomposition of g2 {G) induced by the r-components
we can give a combinatorial description of g2 (G) via the rank function.
THEOREM 4.4.3. Let G = (V, E) be a graph and let r denote the rank function
ofg2(G). Then
k
r(E) =min L{21V{Ei)l- 3),
i=l
where the minimum is taken over all collections {Ei} such that E = UEi.
PROOF. Let {Ei} be a collection of edge sets such that E = UEi. We may
assume without loss of generality that {Ei} partitions E. Let B be a basis for
g2(G). It follows from Theorem 2.5.4 that
r(E) = IBI = L lEi n Bl ~ I:<2IV(Ei n B) I- 3) ~ I:<2IV(Ei)l- 3).

Thus r(E) ~ minE(21V{Ei)l-3) for any abstract rigidity matroid. Now let~
denote the rigid components of G. By Theorem 4.4.1 and the fact that each Ri
is rigid we have that
r(E) = L r(~) = L{2IRil- 3),

and the result follows. 0
With this result, the proof of Theorem 4.1.1 is complete.
Exercise 4.28. State and prove a !-dimensional analogue to Theorem 4.4.3.
As noted in the proof, the minimum is achieved if {Ei} coincides with the
collection of rigid components of G, however other collections may also give the
minimum, as seen in Figure 4.22, in which the count at the left is via the rigid
components. From Theorem 4.4.3 we have immediately the following result first
due to Lovasz and Yemini, see [73).
COROLLARY 4.4.1.
k
df{E) = 2IV{E)I- 3- min L{2IV(Ei)l- 3),
i=l
where the minimum extends over all systems {E 1 , ••• , Ek} of subsets of E such
that E1 U ... UEk =E.
4.5. REPRESENTABILITY OF Q2(n) 103
----.
~~~A
v~~~
4(2. 5-3)
----.
4(2 . 4 - 3) + 8(2 . 2 - 3)
FIGURE 4.22.
4.5. Representability of (h (n). Q1 ( n) has the nice property of being binary,

in fact, regular. We know that Q2 (n) is representable over the real numbers, since
a representation is obtained from a generic embedding p by the rigidity matrix
R(p). We will now investigate if g2 (n) is representable over finite fields.
THEOREM 4.5.1. No 2-dimensional abstract rigidity matroid on more than 4

vertices is binary.
PROOF. Recall that a matroid is binary if and only if the symmetric difference
of two cycles is the disjoint union of cycles. If n > 4, we can find in Kn two
K4 's having 3 vertices in common so that their disjoint union is independent.
See Figure 4.23. 0
5
FIGURE 4.23. Two cycles whose symmetric difference is independent.
In any matroid M on E representable over a finite field F, the number of

cycles is bounded by 1 jJ~:l, where k =lEI- r(E), see Lemma 3.8.1. So if k is
small, there are relatively few cycles.
The next theorem uses this fact to show that no finite field is large enough to
represent all generic rigidity matroids in dimension 2.
THEOREM 4.5.2. There is no finite field F such that Q2(n) is representable

over F for all n.
FIGURE 4.24.
PROOF. The graph (V, E) in Figure 4.24 is rigid and has k = lEI- r(E) = 2.
The removal of any edge with two endpoints of valence 4, as well as removal of
a vertex of valence 3, yields a cycle. Clearly, we can make the string of triangles
as long as we wish, so if a graph G of this form has 2n vertices, g2(n) contains
2(n + 1) cycles. Now assume that g2(n) is representable over IF. We have:
I1FI 2 - 1
IIFI + 1 = IIFI- 1 ~ 2(n+ 1).
D
Exercise 4.29. Prove that g2(n) is representable over the rational numbers.
Exercise 4.30 {Open Question). Is every 2-dimensional abstract rigidity
matroid representable over JR. 'I
Exercise 4.31 {Open Question). Given a finite field IF, do there exist 2-
dimensional abstract rigidity matroids with arbitrarily large vertex sets which are
representable over IF?
Exercise 4.32 {Open Question). Is g2(n) representable over some field of
characteristic 2? of characteristic q?
4.6. Characterizations of .A2 and (.A2)1.. Given a general embedding p:
V -+ .IR.2 , the cocycles of an infinitesimal rigidity matroid F(p) can, as developed
in Section 3.8 be described using the rigidity matrix R(p) corresponding to p.
Edge sets corresponding to the nonzero entries of the elements of the vector
space generated by the columns of the rigidity matrix are the dependent sets of
F(p)l.. Using just the two columns of R(p) which correspond to the vertex v,
we see that both of these columns have at most n - 1 non-zero entries, and by a
linear combination of just these two columns we can obtain a vector with exactly
n- 2 non-zero entries. So the star S(v) of a vertex v minus any one of the edges
incident with v, is dependent and, as can be shown, a cocycle. We call these
n- 1 cocycles the vertex cocycles of v; and, for each edge e E S(v), we denote
the vertex cocycle S(v)- e by Se(v)
Exercise 4.33. Let V = { 1, ... , n} and let p : V -+ .IR.2 be a general embed-
ding. Prove each of the following:
a. Each column of R(p) has exactly n - 1 or exactly n - 2 nonzero entries.
b. Each linear combination of the two columns of R(p) corresponding to a
vertex v has exactly n - 1 or exactly n - 2 nonzero entries.
4.6. CHARACTERIZATIONS OF .A2 AND (.A2).L 105
c. Given an edge e and an endpoint v of that edge, some linear combination

of the two columns R(p) corresponding to v has a zero entry correspond-
ing to e.
d. Show that the "vertex cocycles" are cocycles.
The star of a vertex minus an edge is actually a cocycle in any 2-dimensional
abstract rigidity matroid and, in fact, these cocycles are of use in characterizing
2-dimensional abstract rigidity matroids.
LEMMA 4.6.1. Let (V, K) be the complete graph on n vertices, for n > 4 and
M be any matroid on K of rank 2IVI- 3. Then each of the following three
conditions are equivalent:
a. For each v E V and each e E S(v), Se(v) is a cocycle of M.
b. No cycle of M contains a vertex of valence less than three.
c. Each 2-valent 0-extension of an independent set of M is also an inde-
pendent set of M.
PROOF. a => b: Let E ~ K, let v be a vertex of valence one or two in
(V(E), E) and let e be an edge in E containing v. If v has valence two, let e'
be the other edge in E containing v, otherwise let e' be any other edge in K
containing v. Clearly, IE n Se•(v)l = 1. Thus, if Be' is a cocycle, E cannot be a
cycle, see Lemma 3.6.2.
b => c: A 0-extension of an independent set F contains no cycle, since, by b,
this cycle would have to be a subset of F as well.
c => a: Let v be a vertex and e and edge with v as an endpoint. v is not
a pendant vertex in any basis, since the pendant vertex could be removed and
replaced with a 2-valent 0-extension giving a larger basis. So Se(v) intersects
every basis and so contains a cocycle. This cocycle cannot be a proper subset
since, for any other edge e' incident with v, there is a basis formed by a sequence
of 0-extensions having exactly e and e' incident with v. Note that any maximal
sequence of 0-extensions is a basis because r(K) = 2IVI- 3. 0
Note that, in the previous lemma, the assumption that r(K) = 2IVI - 3 is
necessary because conditions band c hold in Q3(IVI) while condition a does not.
LEMMA 4.6.2. If (V, K) is the complete graph on n vertices, n > 4 and M is
any matroid on K of rank 2IVI - 3 satisfying any of the conditions a-c in the
previous lemma, then for any U ~ V with lUI ~ 2 we have r(K(U)) = 2IUI- 3.
In particular all K4 'sinK are cycles of M.
PROOF. By c we can make an independent set in M by starting with a
single edge performing a sequence of 0-extensions, each 0-extension adding a
new vertex to the support of the edge set. When all vertices have been added
the independent set has 2IVI - 3 edges, and hence is a basis. If we start with
the vertices of U we see that r(K(U)) ~ 2IUI - 3. On the other hand, if K(U)
has a and independent set larger than 2IUI- 3, we can augment it by a sequence
of 0-extensions, one for each vertex of V - U, to an independent set larger than

2IVI - 3, a contradiction. Thus r(K(U)) = 2IUI - 3. It follows that K4 is
dependent and in fact a cycle, since every proper subset is independent. 0
THEOREM 4.6.1. Let (V, K) be the complete graph on n vertices and let M
be any matroid on K of rank 2IVI- 3. If M is a 2-dimensional abstract rigidity
matroid on K, then each of the following three conditions hold:
a. For each v E V and each e E S(v), Se(v) is a cocycle of M.
b. No cycle of M contains a vertex of valence less than three.
c. Each 2-valent 0-e:ctension of an independent set of M is also an inde-
pendent set of M.
Conversely, if any one of the conditions a, b or c hold then M is a 2-dimensional
abstract rigidity matroid on K.
PROOF. Assume that M is a 2-dimellsional abstract rigidity matroid. By

Theorem 3.11.5, condition c holds and then, by the previous lemma, conditions
a and b hold.
Conversely, suppose that one, and hence all of conditions a-c hold. We must
show that M satisfies axioms C5 and C6. We first note that, since a maximal
independent set of K(U) can be extended by 0-extensions to an independent set
containing any given edge not in in K(U), that K(U) is a closed set for any
U ~ V, in other words that (E) ~ K(V(E)). We also note that the previous
lemma implies that any edge set E which can be obtained from a single edge by
a sequence of 0-extensions is a basis forM.
Let E, F ~ K, and set U = K(E), W = K(F). Suppose that IU n WI ::::; 1.
Any edge of (K(U) U K(W)) not in K(U) U K(W) must be of the form {u, w)
for some u E U- Wand wE W- U. Let x E U- u andy E W- w, choosing
x = y if IU n WI = 1. Start with a basis for K(U). Adjoin the vertex w by
a 2-valent 0-extension to the vertices u and x. If x =F y then adjoin y by a
2-valent 0-extension to w and x. Adjoin the remaining vertices of W by 2-valent
0-extensions tow andy. The result is a basis for K(U U W) which contains the
edge {u, w) as well as bases for K(U) and K(W), hence a basis for K(U)UK(W).
Thus {u, w) is not in (K(U) U K(W)) and C5 is satisfied.
Suppose that (E) = K(U) and (F) = K(W). If IU n WI ~ 2, then chose an
edge e in K(U n W) and extend it to a basis B for K(U U W) by 2-valent 0-
extensions onto the endpoints of e. Since B ~ (K(U)UK(V)), C6 is verified. 0
At this point we can answer the question raised in the first section:
COROLLARY 4.6.1. There exists a 2-dimensional abstract rigidity matroid that

is not an infinitesimal rigidity matroid.
PROOF. Consider the set V = {1, ... ,6} and g 2{6) on K = K(V). Let 8 be
the set of bases for g2{6) and let 8' = 8- B where B is the edge set of Ka,a
illustrated in Figure 4.25. Equivalently, think of replacing the six cycles of g2{6)
4.6. CHARACTERIZATIONS OF .A2 AND (.A2).L 107
1 3 5
~
2 4 6
FIGURE 4.25.
which contain B by the single cycle B. If the resulting collection B' is the set
of bases for a matroid on K, then it will follow easily from the theorem that
it is a 2-dimensional abstract rigidity matroid. It will follow, from the Balker-
Roth result [12] on bipartite graphs, that this matroid is not an infinitesimal
rigidity matroid: their result states that, in a 2-dimensional infinitesimal rigidity
matroid, a copy of Ka,a is a cycle if and only if its vertices lie on a conic section.
In this example, only one of several different copies of Ka,a on the same vertex
set is a cycle. It remains only to show that B' satisfies the axioms for the basis
system of a matroid.
Clearly, 0 ¢ B' and all of the sets in B' are of the same size; hence, only the
exchange axiom must be verified. But, we know that 8 does satisfy this axiom.
The question then is: is B absolutely necessary? Specifically, given B1, B2 E B'
and e1 E B1-B2 is B1 -e1 +e2 either dependent or equal to B, for all e2 E B2?
Suppose that the answer were yes. Then B 1 - e1 would have to equal B minus
an edge. Since B2 =f B, B2- B1 contains at least one edge e not in B. As
you will show in the next exercise, adding any other edge to B minus an edge is
always a basis. D
Exercise 4.34. Refer to Figure 4.25. Let B denote the edge set depicted here
and let e = (1,6}. Show that B-e+ e', where e' is any one of the six edges in
K- B, belongs to B' (as defined in the proof to the corollary).
Another proof of this corollary can be based on the observation that, in the 2-
dimensional infinitesimal rigidity matroid for V = {1, ... , 6}, only one prism can
be dependent. One can show that starting with the bases for the 2-dimensional
generic rigidity matroid of V and deleting all prisms results in the bases for a
2-dimensional abstract rigidity matroid for V which is clearly not infinitesimal.
Exercise 4.35. Fill in the details of the proof just outlined.
Observe that the existence of the vertex cocycles in Theorem 4.6.1 forces the
rank of the matroid to be at least 2n - 3. Can we specify a collection of cycles
whose existence bound the rank from above?
Exercise 4.36 (Open Question). Is the following statement a theorem'!
A matroid M of on the edge set of the complete graph on n vertices is a 2-
dimensional abstract rigidity matroid if and only if all of the K4 's are cycles and
all of the vertex stars minus an edge are cocycles.
Both of these conditions are necessary. The uniform matroid on K of rank

5 has each K 4 as a cycle but, when lVI > 4, it is not a 2-dimensional abstract
rigidity matroid; and the uniform matroid on K of rank (1~1) - (lVI- 3) has
each vertex star minus an edge as a cocycle but, when lVI > 3, it is not a
2-dimensional abstract rigidity matroid.
In proving Theorem 4.6.1 we demonstrated that condition b implied the
weaker condition d:
r(K(U)) = 2IUI- 3, for all U ~ V.
Exercise 4.37. Show by example that condition d is actually weaker than
condition a.
It is natural to ask if the following stronger result, parallel to the last open
question, is valid:
Exercise 4.38 {Open Question). Is the following statement a theorem? A
matroid M on the edge set of the complete gmph (V, K) is a 2-dimensional
abstract rigidity matroid if and only if all of the K4 's are cycles and r(K(U)) =
2IUI - 3, for all u ~ v with lUI 2:: 2.
We must require that the K4's are cycles to avoid cases like the matroid on
the edges of K4 with bases indicated in Figure 4.26.
·rsr -rzr
d Bl c d B2 c
FIGURE 4.26.
We now turn our attention to an arbitrary rigidity matroid A 2 for V =
{ 1, . . . , n}. Surprisingly, the complete set of cocycles can be easily described.
Recall that we say that a spanning edge set E ~ K (V) has degree of freedom 1
when r(E) = 2n- 4 = r(K(V)) - 1.
THEOREM 4.6.2. Let V = {1, ... , n} be given. The cocycles of A2 are the
complements of closed spanning subsets with degree of freedom 1 in A2.
PROOF. H Dl. is a subset of K = K(V)) which intersects every basis of
Q2(n), then r(K- Dl.) < 2n- 3 and conversely. So if Dl. is dependent in Q2(n)
thenK- Dl. has at least degree of freedom 1. Now consider a minimal dependent
set Cl. of Q2(n)1.. Clearly K- Cl. is spanning and closed since otherwise we
could enlarge K - Cl. through 0-extensions to a spanning set whose closure has
degree of freedom at least 1 and c1. would not be minimal. Similarly, if K- c1.
had a degree of freedom greater than 1, the addition of an edge to K- Cl.
would result in a set whose complement is dependent in A2, contradicting the
minimality of Cl.. D
4.7. RIGIDITY AND CONNECTIVITY 109
Let G = (V, E) be a graph, lVI :::; n. We can identify G with a subgraph of

Kn and restrict g2(n) to the edge set E. We denote this restriction by g2(G).
To get a collection of cocycles of g2(G), G = (V, E), we take the minimal sets
in the collection {Encl. I cl. is a cocycle of g2(n)}.
In dimension 1 there is a particularly appealing description of graphs G for
which g1 (G)l. = g1 (Gl.)), namely these are exactly the planar graphs. In order
to get this nice duality theory, one has to of course allow parallel edges and
loops. In the case of g2 (G), even if we allow loops and parallel edges, we cannot
obtain such a planarity theory for 2-dimensional generic rigidity matroids since
the existence of the vertex cocycles requires the existence of cycles in g 2 (n)l.
such that their union minus any of its edges is a cycle. So we require in g2 (n)
the existence of two cycles on the same support whose union only has two edges
more than it needs to be rigid, and deletion of any edge of this union yields a
cycle. But this is impossible in an abstract rigidity matroid, since this union
must contain at least one vertex of valence 3. So g2(G)1. is not an abstract
rigidity matroid as soon as any vertex cocycle has more than two edges. If all
vertex cocycles of g2 (G) are of cardinality 2 or less, we may trivially interpret
g2 (G)l. as an abstract rigidity matroid of a graph on 2 vertices connected by
parallel edges and with some loops attached, see Figure 4.27.
--+
a h
FIGURE 4.27.
Exercise 4o39o Verify that all cocycles of Ks are isomorphic to one of the
following four graphs.
Exercise 4o40o Make a complete list of the cocycles for g2(6).

4o 7 o Rigidity and Connectivityo g2 (G) is a matroid defined on the edge
set of G. The vertex set of G is used only, via the support function, to define
independent sets. Consequently, there is no property of g2 (G) that corresponds
directly to the connectivity of G. However, it seems intuitively clear that a highly
connected graph is more likely to be rigid than a graph with low connectivity
and one may wonder how high the connectivity of a graph has to be in order to
assure that its edge set is rigid in g2(n).
110 CHAPTER 4. LINEAR AND PLANAR RlGIDITY
THEOREM 4.7.1 (LOVASZ AND YEMINI). Every 6-connected graph is generi-

cally rigid in 2-space.
PROOF. Suppose G = (V, E) is 6-connected but not rigid. Assume G has

the smallest number of vertices among all such graphs. We may also assume
that all r-components of G are complete, since completing the rigid components
neither changes the rank of G, nor decreases the connectivity of G. Moreover,
since replacing the rigid components of G by complete graphs on the vertices of
attachment does not alter the rigidity properties and since a disconnecting set
of 5 or fewer vertices of this smaller graph would also disconnect G, we conclude
by minimality that every vertex of G is a vertex of attachment, i.e. belongs to
at least two r-components.
By Theorem 4.4.1, we may select a basis Bin E by selecting a basis Bi for
each r-component and taking their union. In r-components consisting of a single
edge, we must, of course, select that edge; in r-components on 3 to 5. vertices,
we only note that each vertex has valence at least 2 in the basis selected for that
component; in all larger r-components, we take care to select a basis in which
every vertex has valence at least three. A vertex v which is contained in four
or more r-components, necessarily has valence greater or equal to four in B. If
v is contained in only three r-components, then at least one of them must be
larger than an edge, so its basis contributes at least two to the valence of v in
B, and again v has valence four or more in B. If v is contained in only two
r-components, then either both of them must be nontrivial, or one of them must
contain at least 6 vertices. In either case, v has valence 4 or more in B. Thus,
every vertex of B has valence at least 4 in B contradicting the fact that any
independent set contains a vertex of valence 3 or less. 0
This result of Lovasz and Yemini is best possible, as you will show in the next
exercises.
Exercise 4.41. Explain exactly where the above proof breaks down if only
5-connectivity is assumed.
Exercise 4.42. Construct the graph G as follows: replace each vertex in K 5 ,5
with a copy of Ks, attaching one edge of the original Ks,s to each vertex of
the inserted copies of Ks. Thus, G has 50 vertices, each of valence 5. Show
that G is 5-connected. Show that an independent set in G can contain at most
10 x 7 + 25 = 95 edges and conclude that G is not rigid.
Exercise 4.43. Find a smallest 5-connected graph which is nonrigid in Q2 (n).
Can you construct infinitely many 5-connected graphs which are not generically
rigid in 2-space? {See [73]}
Recall that a matroid MonS is connected if r(F) +r(E- F) > r(E) holds for
every non-empty proper subset F of E. With this definition the !-dimensional
rigidity matroid of a graph G is connected if and only if G is biconnected. It
4.7. RIGIDITY AND CONNECTIVITY 111
is natural to ask for relations between the rigidity of G and the connectivity of
{h{G).
G is called (vertex) birigid if its edge set is rigid in {h(n) and the removal of
any vertex of G, and the edges incident with it, leaves a rigid graph.
G is called edge-birigid if its edge set is rigid in {h(n) and the removal of any
edge of G leaves a rigid graph.
THEOREM 4. 7.2. If G has no isolated vertices and more than one edge and if
Q2 {G) is connected, then G is edge-birigid, but not conversely.
PROOF. G = (V, E) is rigid, otherwise Q2(G) could be written as the direct

sum over the rigid components of G. Hence r{E) = 2IVI- 3.
Assume that there is an edge, e, such that G- e is not rigid. Then r{ E- e) =
2IVI-4 and r(E-e)+r(e) = r(E). The last equation contradicts the connectivity
of Q2(G).
The converse is not true: Let Go be minimally rigid, having n 0 vertices and
2no - 3 edges. We attach to each edge ei a cycle ci I 1 ~ i ~ (2no - 3) I ci having
ni vertices, by identifying one edge of each Ci with one edge of G 0 • Then the
resulting graph is clearly rigid and hence has rank 2n - 3, where
2n 0 -3
n =no+ L (n, - 2).
i=l
So E~,:i- 3 n, = n + 3no - 6. The rank of each C, is 2n, - 3.
If we sum over the ranks, we get
2n0 -3 2n 0 -3
L (2ni - 3) = -3{2n0 - 3) + 2 L n,
i=l i=l i=l
= -6n0 +9+ 2n+6n0 -12 = 2n- 3 = r(G).
So Q2(G)is not connected. On the other hand, G is clearly edge-birigid. D
Exercise 4.44. Construct the simplest example of the above construction:

take Go to be a triangle and each Ci to be a tetrahedron.
Clearly, cycles are edge birigid. In fact they are edge minimal edge birigid
graphs. That is, deleting any edge from a cycle results in a graph which is not
edge birigid. Are there edge minimal edge birigid graphs which are not cycles?
The answer is yes as you will now demonstrate.
Exercise 4.45. Observe that K3,4 is not a cycle. Show that K3,4 is edge
birigid; then show that the graph obtained by deleting an edge from K3,4 is not
edge birigid.
THEOREM 4.7.3. lfG = (V, E) is birigid and lVI > 3, then Q2(G) is connected
but not conversely.
PROOF. Assume that G is birigid and that Q2(G) is not connected.

Consider the connected components Mi of 9 2 (G). Then there is a partition
of E, E = E 1 U & U · · · U E," such that
where Mi = 92(Gi}, with Gi = (V(Ei}, Ei)· Every Gi is rigid, so it follows that

k k
(4.7.1} 2IVI- 3 = r(G) = L r(Gi) = L(2IV(Ei)l- 3).
i=l i=l
Let r&i be the number of vertices of attachment of Ei, that is, the number of
vertices in the support of Ei which are also contained in the support of some
E;, i ::f: j, and let f'ni be the number of vertices contained only in the support of
Ei. Denote by m the number of vertices of G which are contained in exactly one
of the V(Ei)'s, and by n the number of vertices which occur in more than one
of these supports. So mi, m, ni, and n are defined by the following equations:
mi = IV(Ei)- U#.V(E;}I, IV(Ei)l = ni + mi,
m = E~= 1 mi, lVI = n+m.
Rewriting 4.7.1 in this new notation we obtain
k
2n +2m-3= L(2(ni +mi)- 3}
i=l
or
k
(4.7.2} 2n = 3(1 - k} + 2 L ni.
i=l
Since each vertex of attachement belongs to at least two Ei 's, we have E~=l ni ~
2n, and combining this with and 4.7.2 gives
k k
L: ni ~ 3(1 - k} + 2 L: r&i,
i=l i=l
or
k
(4.7.3} L:ni ~ 3(k -1).
i=l
On the other hand, since every cutset in a birigid graph has cardinality at
least 3, we have that ni ~ 3 for all i. This, combined with 4.7.3, gives
k
3k ~ L ni ~ 3(k- 1},
i=l
a contradiction.
4.8. TREES AND 2-DIMENSIONAL ISOSTATIC SETS 113
Conversely, if (h (G) is connected, G need not be birigid: let G be a wheel with

at least 4 spokes, then Y2(G) consists of a single cycle and hence is connected.
But the removal of the center vertex leaves a non-rigid graph. 0
Thus, we have the following sequence of implications:
G birigid 't Y2(G) connected 't G edge birigid

In reviewing the proof of the Lovasz Yemini Theorem, we see that, if a graph
is 6-connected, then it is not only rigid but over braced. This observation leads
to the first of the following open questions.
Exercise 4.46 (Open Question). Is every 6;connected graph birigid?
In Chapter 1 we introduced the concept of strongly rigid, that is sufficiently
rigid so that any two embeddings with the same edge lengths must be congruent.
Exercise 4.47 (Open Question). Is every 6-connected graph strongly rigid?
4.8. 'frees and 2-dimensional Isostatic Sets. Throughout this section
and the remainder of this chapter, we will restrict our attention to generic rigid-
ity in dimensions 1 and 2. Since we are particularly interested in the analogies
between 2-dimensional generic rigidity and !-dimensional rigidity (graph connec-
tivity), we will adopt a more graph theoretic notation. In stead of considering
an edge set Ewe will will consider the graph G = (V,E), where V = V(E). We
will speak of the graph G as being rigid, independent, isostatic, or a cycle when
E is generically rigid, independent, isostatic, or a generic cycle in Y2(V). For
9t(V), we will use the conventional graph theoretic terms: connected (for rigid),
a forest (for independent), a tree (for isostatic) and circuit (for cycle).
As we have just noted, the !-dimensional isostatic sets are simply the trees,
circuit free, connected subgraphs. Since every tree contains at least one pendant
vertex, and pruning the leaves of a tree yields a smaller tree, all trees can be
constructed from a single edge by 0-extensions.
The 2-valent 0-extensions are clearly not enough to construct all isostatic sets
in Y2(n), since there are, as we have seen, isostatic sets which contain no vertex
of valence 2. But 0- and !-extensions are enough to generate all isostatic sets
of any 2-dimensional abstract rigidity matroid. However, only in Y2(n) does
splitting an edge of an independent set always yield an independent set. We
obtain, therefore, another characterization of the 2-dimensional generic rigidity
matroid, first given by Henneberg [53]. A sequence of graphs (Vt. Et) ... (Vn, En)
is called a 2-dimensional Henneberg sequence if (Vi, E 1 ) consists of a single edge
and its endpoints and if, for each index j,2 ::::; j ::::; n, (Vj, E;) is obtained from
(V;-t. E;-t) by attaching a new vertex by two edges or by deleting an edge from
(V;-t. E;-t) and attaching a new vertex to the endpoints of the deleted edge
plus one other vertex in V;-t· We have as an easy corollary to Theorem 4.2.3:
THEOREM 4.8.1. Let the complete graph (V < K) be given. An edge set
E ~ K is independent and rigid in Ch(IVI) if and only if there is a 2-dimensional
Henneberg sequence {Vt, Et) ... (Vn, En) such that (V(E), E)= (Vn. En)·
Exercise 4.48. Fill in the details of the proof of this theorem.
Exercise 4.49. State and prove a !-dimensional version of Theorem 4.8.1.
There are thirteen different isostatic graphs on six vertices, as shown in Fig-
ure 4.28
FIGURE 4.28. Isostatic graphs on 6 vertices.
Exercise 4.50. Find Henneberg sequences for each of the graphs in Fig-
ure 4.28. Which can be done with only 0-extensionsf
Exercise 4.51. Verify that the graphs of Figure 4.28 are a list of the distinct
isostatic sets in Q2(6).
Besides its recursive description, a tree has many useful characterizations,
e.g., G is a tree if and only if every two vertices of G are joined by a unique path.
This raises the question: What is a suitable generalization of a path to generic
rigidity matroids of higher dimension?
Let G = (V, E) be a graph with lVI =nand consider the matroid Q2(n) for
V. If x andy are vertices of G, we say that a subset P of E is an isostatic xy-set
if either PU (x, y) is a cycle in Q2(n) or P = (x, y). If P = (x, y) then Pis called
the trivial isostatic xy-set.
Recall that G is connected if and only if every two vertices of G are joined by
a path, and that G is edge biconnected if and only if every two vertices of G are
joined by a path of length greater than one. Analogously, we have:
THEOREM 4.8.2. Let G = (V, E) be a graph and consider the matroid Q2(G).
Then G is rigid if and only if every two vertices of G are joined by an isostatic
xy-set, and G is edge birigid if and only if every two vertices of G are joined by
a non-trivial isostatic xy-set.
PROOF. Assume that G is rigid and that x, y E V. Since (E} = K(V), either
(x, y) E E orE U (x, y) contains a cycle containing (x, y). In either case, x and
y are joined by an isostatic xy-set. Conversely, assume that, for all x, y E V, x
andy are joined by an isostatic xy-set. It follows at once that, for all x, y E V,
=
(x, y) E (E}, i.e. (E} K(V).
The birigid case is left as an exercise for the reader. 0
Exercise 4.52. Complete the proof of this theorem.
Exercise 4.53. Show that Ka,a is an isostatic xy-set for any pair of non-
adjacent vertices x and y.
An isostatic xy-set is meant to model a path in {h(n). A path is simply
described as a tree with exactly two pendant vertices, however we will see that
there is no 2-dimensional analogue to this observation. Since a rigidity cycle
cannot have any vertex of valence 2, it follows that an isostatic xy-set can have
at most 2 vertices of valence 2. If an isostatic xy-set does have exactly 2 vertices
of valence 2, then those vertices must in fact be x andy, however Figure 4.29
shows that an isostatic set with exactly 2 vertices of valence 2 need not be an
FIGURE 4.29.
isostatic xy-set. Exercise 4.53 shows that there are isostatic xy-sets in which the
valence of both x and y is greater than 2. In Figure 4.28, all graphs with fewer
than two vertices of valence two are indeed isostatic xy-sets. Figure 4.30 shows
an example of an isostatic set without a vertex of valence two which is not an
isostatic xy-set, since any possible additional edge is contained in a proper rigid
subgraph.
FIGURE 4.30.
Let us call a pair of vertices (v, w) of an isostatic xy-set P end vertices of P if

the union of P and the edge (v, w) is a cycle. Some isostatic xy-sets have exactly
one pair of endpoints, namely (x, y), for example if the xy-set has exactly 2
vertices of valence 2. In general, an isostatic xy-set has several pairs of endpoints,
for example in Exercise 4.53 we see that in Ka, 3 every pair of nonadjacent vertices
is a pair of endpoints.
Exercise 4.54. Verify that Ka,a is an isostatic xy-set with the property that
euery nonadjacent pair of vertices is a pair of endpoints. Identify all other iso-
static xy-sets of Figure 4.28 with that property.
Exercise 4.55. Show that if {x, y} is a separating set for an isostatic graph,
then {x, y} are not end vertices.
While isostatic xy-sets only partially conform to our notion of what a path
should be, the following characterizations of isostatic sets show that isostatic
sets are in all respects analogous to trees.
THEOREM 4.8.3. The following statements are equivalent for a graph G =

(V,E);
(i) G is an isostatic set;
(ii) Every two vertices x and y of G are joined by a unique isostatic xy-set;
(iii) G is rigid and lEI = 2IVI- 3;
(iv) lEI = 21¥1- 3 and G does not contain a cycle,
(v) G does not contain a cycle and G + e has exactly one cycle for all e E
K(V)- E;
(vi) G is rigid, is not KIVI for lVI > 3 and, if e E K(V)- E, then G + e
contains exactly one cycle;
(vii) lEI= 2IVI- 3, G does not contain a copy of K4 and, if e E K(V)- E,
then G + e has exactly one cycle.
Before proving this result the reader should take a moment to write down the
one dimensional analogue to this theorem and to verify that it includes many of
the standard characterizations of trees.
PROOF. The proof is a circle of implications.

a ::::::} b: Let x and y be vertices. If x and y are adjacent, and then they
belong to a trivial xy-set, and G cannot contain a non-trivial xy-set since then
G would contain a cycle. Assume that x and y are not adjacent. Then, since G
is rigid, G+(x,y) is dependent, and (x,y) is contained in a cycle in G+(x,y), so
G contains an xy-set. This isostatic xy-set is unique since otherwise G + (x, y)
contains two distinct cycles each containing (x, y) and, by the cycle axioms, G
contains a cycle.
b ::::::} c: Suppose G is not rigid. Then there is exists (x, y) ¢ K(V)- (E),
and so (x, y) ¢ (F) for all F ~ E, and so F + (x, y) is not a cycle and, contrary
to assumption, G admits no isostatic xy-set. Therefore G is rigid and lEI ~
r(K(V)) = 2IYI - 3. If E > 21¥1 - 3 then E violates Laman's condition and
hence contains a cycle C, but this leads to a contradiction since any adjacent
pair (x, y) in C is contained in two isostatic xy-sets, namely (x, y) and C- (x, y ).
c ===> d: Suppose G is rigid and lEI = 2IVI - 3. If G contains a cycle,
then r(E) < lEI but since G is rigid, E contains a subset F such that r(F) =
2IVI - 3 = lEI, a contradiction.
d ===> e: Suppose lEI= 2IVI- 3 and G does not contain a cycle. So E is
independent and is of full rank, hence a basis. Thus E + e contains a cycle for
all e E K(V)- E. If there were two, then the cycle axioms would imply the
contradiction that G contains a cycle.
e ===> f: Suppose G does not contain a cycle and G + e has exactly one cycle
for all e E K(V) -E. If e E K(V) - (E), then e is not in the closure of any
subset F of E, so that G + e has no cycle containing e, hence contains no cycle,
a contraction. So (E) = K(V). Since K 4 is dependent, hence contains a cycle,
G is not complete if lVI > 3.
f ===> g: Suppose G is rigid, is not KIVI for lVI > 3 and G+e has exactly one
cycle for all e E K(V) - E. Since G is rigid, lEI ~ 2IVI - 3. If lEI > 2IVI - 3,
then E is dependent, and contains a cycle. Thus G + e has no cycle containing
e for any e E K(V)- E, so e ¢ (E), contradicting the rigidity of G. Suppose
that G contains a copy of K4. addition of one edge either creates a second cycle
or increases the rank. By assumption, both options are excluded. Thus, it must
not be possible to add an edge. So, if G contains a copy of K 4 , it must equal
K4, which is also excluded by assumption.
g ===> a: If E is independent, then r(E) = lEI = 2IVI- 3 and E is also rigid.
Suppose that E is not independent, then E contains a cycle C. This cycle cannot
be K4. So there exists e E (C) -C. But then G U e contains two cycles. 0
Exercise 4.56. For each statement in Theorem 4.8.3 give the analogous state-
ment for gl (n) and show that these statements are also equivalent in gl (n)
Given a graph G = (V, E) and a cost function $ : E -+ R, it is well known
that the task of finding a minimum cost spanning tree of G can be accomplished
by a greedy algorithm. So the first step is to choose the cheapest edge, and the
k'th step is to choose the cheapest unchosen edge that does not create a cycle
among those edges chosen already. If the cost function $ is an injection, then
the optimal spanning tree is unique. (For more details see for example [70]).
This turns out to be an illustration of a general matroid theoretic result, and it
may be shown that a matroid is an independence structure for which the greedy
algorithm always produces a minimum cost spanning set. In terms of rigidity
we may assign a cost function $ : E -+ R to the edges of G and find a minimum
cost spanning rigid subgraph by the following greedy algorithm: At the i'th step
choose an edge of minimal cost which is not contained in this closure.
Exercise 4.57. Let V = {1, ... ,6}, and let G = (V, K(V)) be a graph with
cost function$: K(V)-+ R defined by $({i,j}) = li- il. Find the minimum
spanning rigid subgraphs of G.
4.9. Tree Decomposition Theorems. Nash-Williams proved in 1964 that

a graph G = (V, E) is decomposable into k spanning forests if and only if IFI ~
k(IV(F)I-1) for all nonempty subsets F of E. Fork= 2 this looks remarkably
similar to Laman's characterization of independent sets in g2 (n). In this section,
we explore the relationships between these two results. The graphs described
by Nash-Williams have the same number of edges, relative to the number of
vertices, as a cycle in Q2(n). For this reason, we define any graph G = (V, E)
such that lEI = 2IVI - 2 and IFI ~ 2IV(F)I - 2, for all 0 C F ~ E, to be
a pseudocycle. In addition, we include single vertices among the pseudocycles.
(The characterization of pseudocycles in the next exercise explains the inclusion
of single vertices.) It is convenient to consider pseudocycles in the context of
multigraphs. One easily observes that, in a multigraph, a pseudocycle may have
multiple edges, but never more than two edges sharing a given pair of vertices.
In fact the simplest pseudocycles, other than single vertices, consist of two edges
with the same end points.
Exercise 4.58. Prove that a nonempty multigraph G = (V, E) is a pseudo-
cycle if and only if lEI = 2IVI - 2 and IE(U)I ~ 2IUI - 2, for all 0 C U ~ V.
Recall E(U) =En K(U).
Exercise 4.!)9. Let the multigraph G = (V, E) be a pseudocycle. Prove that
G is connected and edge 2-connected. Prove that, if G is not a single vertex,
then every vertex of G has valence at least two.
Unlike cycles, pseudocycles may contain proper submultigraphs which are
themselves pseudocycles. A single vertex or a double edge would be an example
of a submultigraph which is a pseudocycle. Such subpseudocycles will be called
the modules of a pseudocycle.
LEMMA 4.9.1. Let G = (V, E) be a pseudocycle and let Mi = (Vi, Ei) be a

module of G, fori = 1, 2. Then either Vt n V2 = 0 or both Mtn2 = (Vt n V2, Et n
E2) and Mtu2 = (Vt UV2,E1 UE2) are modules of G.
PROOF. Assume that Vt n V2 ::f: 0. Since G is a pseudocycle, we have:
if E 1 n E 2 = 0, we have by direct computation:
Adding this inequality to the first one and using inclusion-exclusion to simplify
both sides, we get:
4.9. TREE DECOMPOSITION THEOREMS 119
with equality here if and only if the equalities hold throughout the previous
strings of inequalities. But, since both M 1 and M 2 are pseudocycles, equality
holds here and we may conclude that both MlU2 and M1n2 are pseudocycles. D
Exercise 4.60. Show that, in a pseudocycle, any two cycles have at most one
vertex in common and that the union of a module and an isostatic set which are
not disjoint is a module.
As we noted earlier, it is not known if all cycles can be constructed from a
K4 by a sequence of 1-extensions. The corresponding result for pseudocycles
is valid. However, the concept of "k-extension" must be extended! Let G =
(V, E) be a pseudocycle and let x denote a vertex not in V. The multigraph
H = (V U {x},E U F), where F = {(x,u), (x,v)} and v,u E V, is called a
2-valent pseudoextension of G; and H = (V U {x}, (E- e) U F), where F =
{(x,u),(x,v),(x,w)}, v,u,w E Vande E K({u,v,w}), is called a 3-valent
pseudoextension of G. Note that, in a 2-valent pseudoextension, v and u need
not be distinct and that, for a 3-valent pseudoextension, we only require that
l{u,v,w}l;::: 2.
Exercise 4.61. Show that a 2-valent or 3-valent pseudoextension of a pseu-
docycle is a pseudocycle.
Exercise 4.62. Show that deleting a 2-valent vertex and incident edges from
a pseudocycle results in a pseudocycle.
LEMMA 4.9.2. A pseudocycle G = (V, E) with E =F 0 is a 2 or 3-valent pseu-

doextension of a smaller pseudocycle.
t
PROOF. By direct computation (~ = 4- 1 1), we see that the average
valence of G is less than 4. As you have shown, G admits no isolated or pendant
vertices; hence, G has a vertex of valence 2 or one of valence 3. Also, as you
have shown, if G has a vertex of valence 2, it is a 2-valent pseudoextension of a
smaller pseudocycle. Assume then that G admits a vertex x of valence 3.
Assume that the neighbors of x are u, v and w and that they are distinct. (The
simpler case that they are not all distinct is left as an exercise for the reader.)
Let V' = V- {x} and E' = E- {(x, u), (x, v), (x, w)}. Then IE' I = 2IV'I - 3
and E' + (u, v) will be a pseudocycle unless there is a module M(U, F) of G
which contains both u and v but not x. Assume that M is just such a module.
We first observe that w ¢ U; if it were, (U + x, F U {(x, u), (x, v), (x, w)} would
violate the defining inequality. Now suppose that N is a module containing v
and w but not x. But then M UN is a module containing u, v and w but not x,
leading to a contradiction. Thus, we may add an edge between v and w to get
a pseudocycle and G is a 3-extension of this pseudocycle. D
Exercise 4.63. Complete this proof: show that, in a pseudocycle G, a vertex

x of valence 3 has at least two distinct neighbors; then show that, if x has just
two neighbors, then no module of G can contain both of these neighbors but not
x.
Exercise 4.64. Define "Henneberg sequences" for pseudocycles and show that
each pseudocycle is the last multigraph in such a "Henneberg sequence" starting
with a single vertex.
We now return to the Nash-Williams result. Let G be a multigraph. Two
spanning trees of G which are edge disjoint and have G as their union are called
a 2-tree decomposition of G.
LEMMA 4.9.3. A multigraph G = (V, E) is a pseudocycle if and only if G

admits a 2-tree decomposition.
PROOF. Assume that G is the edge disjoint union of two of its spanning trees,
then any induced subgraph on k vertices has at most 2k - 2 edges, with equality
only if the induced subtrees have the same span. It follows at once that G is a
pseudo cycle.
We prove the converse by induction. Let G = (V, E) be a pseudocycle If
E = 0 and V = {v}, G is the edge disjoint union of two copies of the same
spanning tree - itself. Now assume that all pseudocycles on fewer vertices are
the edge disjoint union of two spanning trees.
By the previous lemma, G is a 2 or 3-valent pseudoextension of a smaller
pseudocycle H. Let T1 and T2 be two edge disjoint spanning trees of H with
H as their union and let x denote the vertex of the pseudoextension. If G is a
2-valent pseudoextension of H, simply add one of the edges containing x to T1
and the other to T2. The resulting trees are edge disjoint spanning trees of G
and have G as their union. If G is a 3-valent pseudoextension of H, let (v, w)
denote the edge of H deleted in the extension. Suppose (v, w) belongs to T1 •
Then replace (v, w) in T1 by the path (v, x), x, (x, w) and add the remaining edge
at x to T2. Again, one easily checks that the resulting trees are edge disjoint
spanning trees of G and have G as their union. 0
Cycles of (h(n) can, by Theorem 4.3.1 and Lemma 4.9.3, be decomposed

into two spanning trees. But the edge disjoint union of two trees on the same
vertex set is a pseudocycle and may not be a cycle. A pseudocycle is certainly
a dependent set in Ch (n). So, if it is not a cycle, it must contain a cycle, which
would be a module of the pseudocycle not equal to a single vertex or the entire
pseudocycle.
Exercise 4.65. Prove that a pseudocycle G is a cycle if and only if G is
not a single vertex and the only modules of G are G itself and the single vertex
subgraphs of G.
Exercise 4.66. Let G = (V, E) be a pseudocycle and let T1 and T2 be the
edge sets of a 2-tree decomposition of G. Let M(U, F) be a submultigraph of G.
4.9. TREE DECOMPOSITION THEOREMS 121
Prove that M is a module of G if and only if (U, T1 n F) and (U, T2 n F) are

spanning trees of M.
A 2-tree decomposition of G is said be a proper 2-tree decomposition if no pair
of proper subtrees, excepting single vertices, have the same span.
THEOREM 4.9.1. A graph G on n vertices is a cycle in Q2(n) if and only if it

admits a proper 2-tree decomposition.
PROOF. If G is a cycle in Q2(n), then it is a pseudocycle. So by Lemma 4.9.3

G is the disjoint union of two of its spanning trees. Since G admits no modules
other than G itself and single vertices, no two subtrees have the same span.
Conversely, if G is the union of two of its spanning trees, then it is a pseu-
docycle. Since no two subtrees have the same span, G admits no modules other
than G itself and single vertices. Thus G is a cycle. 0
Exercise 4.67. Decompose each of the six cycles on six or fewer vertices into
two edge disjoint trees no two of whose subtrees have the same span.
Exercise 4.68. Use a 2-tree decomposition to construct a cycle on n vertices
from two paths of length n. Find the smallest cycle which will not decompose
into two paths.
Exercise 4.69 (Open Question). Does every cycle with vertices of valence
3 and 4 only, have a two tree decomposition into two paths?
If an edge set E is isostatic, x, y E V(E), (x, y) ¢ E, then E + (x, y) contains
exactly one cycle of Q2 (n), hence satisfies the hypothesis of Lemma 4.9.3 and, in
fact, we have the following result.
THEOREM 4.9.2. Let G = (V, E) be a graph which does not contain a copy of
K4. Then E is isostatic in Q2(n) if and only if, for any edge e E K(V) - E,
G + e admits a 2-tree decomposition.
PROOF. If G is lVI < 4, then the result may be directly verified. Assume
then that lVI ;::: 4. If E is isostatic, then one easily verifies that the addition of
any edge toE creates a pseudocycle which admits a 2-tree decomposition.
Conversely, since G is not complete, there is an edge e E K(V)- E. Since
E + e is the union of two spanning trees, we conclude that E has cardinality
2IV(E)I- 3. We need to show that all subsets of E satisfy Laman's condition.
Assume F ~ E violates Laman's condition. Since G contains no copy of K4, one
easily checks that (V(F), F) is not a complete subgraph. Thus there exists an
edge e E K(V(F))- F and F + e violates Laman's condition by at least 2. so
by Lemma 4.9.3 E + e can not be decomposed into two trees. 0
Exercise 4. 70. Show, by example, that the condition that G not contain a
copy of K4 is essential to the previous result.
The next result seems less natural, since it involves multigraphs, but is more
useful, since it has no exceptional cases. Furthermore, it is slightly easier to
prove.
THEOREM 4.9.3. Let G = (V, E) be a graph. Then E is isostatic in g2(n) if
and only if, doubling any edge e E E, results in a multigraph which admits a
2-tree decomposition.
Exercise 4. 71. Prove this theorem.
It is clear that an edge set E all of whose subsets violate Laman's condition by
at most 1, can have cardinality at most 2IV{E)I-2, and that if lEI < 2IV{E)I-2,
E can be extended to an edge set with the same property on the same support
with cardinality 2IV{E)I- 2. We formulate this fact as a routine exercise.
Exercise 4. 72. Prove the following: Let E be an edge set such that for all
0 C F ~ E, IFI ~ 2IV{F)I - 2 and let V = V(E). Then there exists E' 2 E
such that V(E') = V and G'(V, E') is a pseudocycle.
Combining Theorems 4.9.2 and 4.9.3 with the statement of the previous ex-
ercise we immediately obtain:
THEOREM 4.9.4. Let G = (V, E) be a graph which does not contain a copy of
K4. Then E is independent in g2(n) if and only if, for any edge e E K(V)- E,
G + e is the edge disjoint union of two spanning forests.
We also obtain the following similar result.
THEOREM 4.9.5. Let G = (V, E) be a graph. Then E is independent in g2 (n)
if and only if, doubling any edge e E E, results in a multigraph which is the edge
disjoint union of two spanning forests.
Theorem 4.9.3 implies that an isostatic set is the union of three trees. One of
the trees is spanning, the other two arise from removing the extra edge from the
second tree in the decomposition, so one of the small trees could actually be an
isolated vertex. Crapo [32] noted that many other 9-tree decompositions exist
and went on to prove the 3-tree decomposition theorem below. In Figure 4.31
we illustrate his result by listing three 3-tree decompositions for Ka,3·
THEOREM 4.9.6. A graph G = (V, E) is isostatic if and only if G is the edge
disjoint union of three trees such that each vertex of G is contained in exactly
two of the trees, and no two subtrees have the same span.
PROOF. If E is isostatic then, by Theorem 4.9.3, for any edge e E E, we have
that the multigraph G obtained by doubling e is the edge disjoint union of two
spanning trees. Removing the duplicate e disconnects one of the trees, so that
E is the edge disjoint union of three trees with each vertex belonging to exactly
two of these trees. Furthermore, G contains exactly one cycle C of g2 (V), and
e E C, so no two subtrees have the same span.
4.10. COMPUTATIONAL ASPECTS 123
® ]*·
®vX. u
®ox.
FIGURE 4.31. Three 3-tree decompositions of K 3,3 .
Conversely, if E is the union of 3 trees T1, T2 and Ta such that every vertex
of G is contained in exactly 2 of the trees, then
and
IE(G)I = IV(T1)I + IV(T2)I + IV(Ta)l- 3 = 2jV(G)I- 3.
Since no two subtrees have the same span, all edge subsets satisfy Laman's
condition. D
Exercise 4.73. For which values of n could the trees of a 9-tree decompo-
sition be paths of equal length? Can you give an isostatic set with a ,9-path
decomposition for each of these values of n?
4.10. Computational Aspects. Given a graph G = (V, E), how can we
decide whether or not E is independent in g2(G)? We can use any one of our
characterizations for g2(n) and ask if it provides us with a useful algorithm to
test the independence of an arbitrary edge set. Take, for example, Laman's
condition. Na.lvely, to check if E is independent we have to check if IFI ~
2IV(F)I- 3 for each set F in the collection 'P(E) of all subsets of E. This is
clearly not a very efficient way of answering the question of independence. A
more efficient method is the following: Consider the function l/J : 'P(E) -+ Z
defined by l/J(F) ~ 2jV(F)I- 3 -IFI· This is a submodular set function:
l/J(F U E)+ l/J(F n E) ~ l/J(F) + l/J(E).
There exist polynomial time algorithms to find the minimum of any submodular
set function, see [48]. If this minimum is non-negative, then E is independent.
Exercise 4. 74. Verify that l/J, as just defined, is submodular.
It was first pointed out by Lovasz and Yemini [73] that an algorithm by
Edmonds [41] can be used in conjunction with Theorem 4.9.2 or Theorem 4.9.3 to
efficiently resolve the question of whether a given edge set is isostatic. Edmonds'
algorithm finds a decomposition of a set in a matroid into k disjoint independent

sets, or decides that such a decomposition is not possible. Taking k = 2 and
considering the rank function of {}t (n) on E, the Edmonds algorithm will search
for the decomposition of an edge set into two forests. To use the Edmonds
algorithm in conjunction with Theorem 4.9.2 to verify that an edge set is isostatic
would require running the algorithm up to (1~1) - lEI times while using it in
conjunction with Theorem 4.9.3 would never require more than lEI applications.
The reason that the algorithm must be run many times is that E + e may have
a 2-tree decomposition and E may still contain a cycle (two subtrees with the
same span). Only when the added or doubled edge has both endpoints on this
cycle will the algorithm terminate with a negative result. Crapo (32] observed
that, with a slight modification, one application of the Edmonds Algorithm will
determine generic independence. It is this modified Edmonds Algorithm that we
will develop here.
The key to the Edmonds Algorithm is the following subroutine: Let a graph
G = (V, E) be given and let Fo and F1 be two edge disjoint subforests. Construct
the following sequences of nested sets:
Fo 2 F2 = Fo n (F1) 2 F4 = Fo n (Fa) 2 · · ·;
F1 2 Fa = F1 n (F2) 2 Fs = F1 n (F4) 2 · · · ;
where (-) denotes closure in {}t(G). (We see that F2 2 F4, since F1 2 Fa; then
Fa 2 Fs since F2 2 F4; etc.)
Because of the finiteness of G, each of these sequences must eventually stabilize
by repeating the same subforest from some index on. In fact, one easily checks
that, if F;+2 = F;, then
F;+a = F(J+a) (mod 2) n (F;+2) = F(j+1) (mod 2) n (F;) = F;+1·

Hence, if F;+2 = F;, then Fi+2 = Fi, for all i;:::: j. Next, we observe that, since
Fi = Fi (mod 2) n (Fi-1), Fi ~ (Fi-1) and (Fi) ~ (Fi-1)· Thus, we have the
associated nested sequence:
Exercise 4. 75. Prove that all three sequences stabilize at roughly the same
time: Specifically, show that, if F;+2 = F;, then (F;+l) = (F;); and that if
(F;+l) = (F;), then F;+2 = F;. Also show that, when they stabilize, either they
are all empty or none are empty.
Suppose that these sequences stabilize before reaching the empty set, i.e.
Fi+2 = Fi =F 0. Then Fi+1 ~ (Fi) and Fi ~ (Fi+1)· To simplify the notation,
assume that D and E are two edge disjoint subforests of G such that D ~ (E)
and E ~ (D). Let (U11 D1), ... , (Uk, Dk) denote the components of (V(D), D).
The components (U11 D1), ... , (Uk, Dk) are then disjoint trees. Since E ~ (D)
each edge in E has both endpoints in the same U;. Thus, the vertex sets of
the components of (V(E), E) are subsets of the vertex sets of the components
of (V(D), D). Then, by symmetry, we may assume that the components of
(V(E), E) are (U11 E 1 ), ••• , (U~c, Etc)· It follows that, each subgraph of G of the
form (Ui, DiU Ei) is a pseudocycle. We conclude that, if G = (V, E) admits two
edge disjoint subforests Fo and F1 such that the associated sequences stabilize
at subforests which are not empty, then the edge set Fo U F 1 is dependent in
g2(G).
The converse is also true. Suppose that Fo and F 1 are edge disjoint subforests
of G and suppose further that (V(Fo) U V(F1 ), Fo U F 1 ) contains a pseudocycle,
(U, D). Since IDI 2= 2IUI- 2 while ID n Fil :::; lUI- 1, fori= 0, 1, we conclude
that (U, D n Fi) is a tree, for i = 0, 1. Thus, G admits subtrees 7i of Fi for
i = 0,1 with V(To) = V(Tl). It follows at once that (To) = (T1 ) and a simple
induction argument shows that F; 2 T; (mod 2) for all j.
There is one more fact about these sequences that is needed to understand the
algorithm. Assume that we have been given Fo and F1 and have constructed the
sequences. Suppose that e E (F;-1) but e ¢ (F;). Since F; = F; (mod 2)n(F;-1),
e cannot belong to F; (mod 2). On the other hand,
e E (F;-1) ~ (F;) ~ · · · ~ (F; (mod 2))·

Hence, there is a unique cycle C ~ F; (mod 2) + e. Furthermore, (C - e) g;
(F;- 1 ): if (0-e) ~ (F;- 1 ), thenC-e~ F; which would imply the contradiction
e E (F;). Thus, there is a first index j 1 such that (C- e) g; (F;')· Hence there
must be some edge e' E C- e but not in (F;')·
We now describe the modified Edmonds Algorithm. Again note that, in this
description and in the proof that it "works", we use rank and closure in g1(G).
INPUT: A graph G = (V, E)
OUTPUT: A 2-forest decomposition of G when G is independent and
the message "DEPENDENT" when G is dependent.
Step 1: Set Fo = F1 = 0.
Step 2: If Fo U F 1 = E, GOTO Step 7; otherwise, let e be the edge of
least index in E- (Fo U F1).
Step 3: If e ¢ (F0 ), replace Fo by Fo + e and GOTO Step 2.
Step 4: If e ¢ (F1), replace F1 by F1 + e and GOTO Step 2.
Step 5: Construct the sequences of nested sets:
Fo 2 F2 = Fo n (F1) 2 F4 = Fo n (Fa) 2 · · · ;
F1 2 Fa = F1 n (F2) 2 Fs = F1 n (F4) 2 · · · ;
until the sequences become stationary. If the stationary sets are not
empty, STOP and output "DEPENDENT".
Step 6: Let j be the first index so that e ¢ (F;). Let C be the unique
cycle contained in F; (mod 2) + e and let j' be the first index so that
C- e g; (F;' ). Next let e' be the edge in C- e with smallest index
such that e' f/. {F;')· Replace F; (mod 2} by F; (mod 2) + e- e', replace
e by e' and GOTO Step 3.
Step 7: construct the sequences of nested sets:
Ft 2 Fa = Ft n {F2) 2 Fs = Ft n {F4) 2 ··· ;

until the sequences become stationary. If the stationary sets are not
empty, STOP and output "DEPENDENT", otherwise STOP and out-
put Fo and Ft.
Before we prove that the algorithm actually terminates as described in the
output statement, we illustrate just how the algorithm works. To do this, we
apply the algorithm to the graph in Figure 4.32. Since this graph has 7 vertices
and 12 = 2 x 7-2 edges, it is clearly dependent. However, it serves the purpose
of illustrating the algorithm very well.
FIGURE 4.32.
EXAMPLE 4.10.1. Starting with Fo = Ft = 0, the algorithm iterates Step 2

eleven times before it can no longer simply add an edge to either Fo or Ft. At
this point, we have Fo = {1, 2, 3, 4, 6, 7} and Ft = {5, 8, 9, 10, 11 }. (See the table
at the end of this example.) Edge 12 is in the closures of both Fo and Ft; so,
the algorithm moves on to Step 5 (for the first time). The closures of the sets it
generates are:
{Fo) = E,
{Ft) = {4,5,6,7,8,9,10,11,12},
{F2) = {Fo n {Ft)) = {4, 6, 7, 11, 12},
{Fa) = {Ft n {F2)) = {11},
{F4) = {Fo n {Fa)) = 0.
In step 6, we note j = 3; we add edge 12 to F 1 and find the cycle C = {9, 10, 12};
we note j' = 2: we then replace edge 9 by edge 12 in F 1 and return to Step 3
with e equal to edge 9. Since edge 9 is in the closures of both Fo and the new
Ft, we again proceed to Step 5. This time we get the sequence:
(Fo} = E,
(Ft} = {4,5,6,7,8,9,10,11,12},
(F2} = (Fo n (Ft}} = {4, 6, 7, 11, 12},
(F3} = (Ft n (F2}} = {11,12},
(F4} = (Fo n (F3}} = 0.
This time j equals the "old" j' or 2. Edge 9 is added to Fo creating the cycle
C = {2, 3, 4, 7, 9}. Then j' = 1 and edge 9 replaces edge 2 in Fo. We move
to Step 3 with e equal to edge 2 and then to Step 4 where edge 2 is added to
Ft. The algorithm then returns to Step 2 and moves directly to Step 7. We
have a 2-forest decomposition of E. But, since (Fo} = (Ft} = E, the algorithm
terminates with the message "DEPENDENT".
In Table 4.2 we list the sequence of sets Fo and Ft which occur.
iteration Fo Ft
of Step 2
0 0 0
1 {1} 0
2 {1,2} 0
3 {1,2,3} 0
4 {1,2,3,4} 0
5 {1,2,3,4} {5}
6 {1,2,3,4,6} {5}
7 {1,2,3,4,6,7} {5}
8 {1,2,3,4,6,7} {5,8}
9 {1,2,3,4,6,7} {5,8,9}
10 {1,2,3,4,6, 7} {5,8,9,10}
11 {1,2,3,4,6,7} {5, 8, 9, 10, 11}
12 {1,2,3,4,6,7} {5,8,10,11,12}
12' {1,3,4,6,7,9} {5,8,10,11,12}
12" {1,3,4,6,7,9} {2,5,8,10,11,12}
TABLE 4.2. Edmonds' Algonthm
We have already seen that, if the algorithm terminates, it does so with the
"correct" answer: "DEPENDENT'', if E is dependent, and a 2-forest decompo-
sition, if E is independent. We still must show that the algorithm terminates in
a finite number of steps. Specifically, we must show that, between two iterations
of Step 2, the algorithm can return to Step 5 only a finite number of times.
In order to follow the steps of the algorithm, we will not rename our new sets
and edges with the old names: e.g. e' will not be renamed e. Also, for this
discussion, let us assume that first index j so that e f/. (F;} is even, so the "new
Fo", which we now denote by F~, is Fo + e - e'. We also have that F{ = F1.
Note that (Ff} = (Fi} fori= 0,1. Now consider F 2 = F~ n (F{}. If e' E (F~},
thenC-e~ (F2} and (F~} = (F2}. Inductively, (Ff} = (Fi}, fori= 0 .. . j',
where j' is the first index so that e' f/. (F;'}· Thus the "new j" is j' and j' < j.
We conclude that Step 5 can only be iterated lEI or fewer times between any
two iterations of Step 2. We have proved:
THEOREM 4.10.1. Given a graph G = (V, E), the Edmonds algorithm, as
adapted above, terminates in a finite number of steps. It terminates with a
partition of E into two forests, whenever E is independent and with the message
"DEPENDENT" whenever E is dependent.
Exercise 4. 76. Consider the graph pictured in Figure 4.33 below.
a. Verify that E is independent using the algorithm.
b. Replace edge 7 by a new edge 7 joining vertices u and v; replace edge 9
by a new edge 9 joining vertices v and w; and rerun the algorithm.
FIGURE 4.33.
Reconsider the graph in Figure 4.32. Is this a cycle? Can the algorithm be
altered to check for cycles? The answers are "yes" and ''yes". If we are applying
the algorithm to a graph G = (V, E) with lEI = 2IVI - 2, the application of
Step 7 is superfluous: any proper subpseudocycles are masked by the fact that
E itself is a pseudocycle. How can we search for proper subpseudocycles which
we did not run across while constructing the 2-tree decomposition? To answer
this question, we must make an important observation: Let G = (V, E) be a
graph with lEI= 2IVI-2 and suppose that G contains a proper subpseudocycle
F. Since IFI = 2IV(F)I- 2 and IFI < lEI, we can conclude that IV(F)I < lVI·
Thus, any proper subpseudocycle must be contained in one of the subgraphs of
G obtained by deleting a vertex.
Exercise 4. 77. Alter Step 7 of the Edmonds Algorithm to check for all proper
pseudocycles, in the case lEI = 2IVI - 2. (If you are clever, you can run the
subroutine considembly fewer than lVI times.)
Exercise 4. 78. Use your altered algorithm to show that the graph in Fig-
ure 4.32 is a cycle.
Chapter 5. Rigidity in Higher Dimensions
5.1. Introduction. In the preceding chapter, several characterizations of 2-

dimensional generic rigidity were given. Of these the Laman characterization
is known to fail in all higher dimensions, a fact we have noted several times.
All of the remaining characterizations of generic rigidity in 2-space which we
discussed in Chapter 4 may be extendable to 3-space. That is, they are not
known to be false in 3-space. In so far as possible, we will frame our discussions
in m-dimensional space. To that end it is convenient to introduce the following
notation. Let the finite set V be given. An edge set E !;;;;; K (V) which is isostatic
{independent, rigid, or a cycle) in them-dimensional generic rigidity matroid for
Vis said to be m-isostatic {m-independent, m-rigid, or an m-cycle).
The second characterization of the 2-dimensional generic rigidity matroids
which we considered was the Henneberg construction. In 1911, Henneberg [53)
described a method for constructing all 2-isostatic sets: a sequence of graphs
{V1, E1), ... , (Vn, En) is called a 2-dimensional Henneberg sequence if (V1, E 1)
consists of a single edge and its endpoints and if, for each index j, 2 ~ j ~ n, the
graph (V;, E;) is obtained from (V; _~, E;- 1 ) by a 0-extension or a !-extension.
In 3-space, a sequence of graphs {Vi, E1), ... , (Vn, En) is called a 3-dimensional
Henneberg sequence if (V~, E 1) consists of a triangle and its vertices and if, for
each index j, 2 ~ j ~ n, the graph (Vj,E;) is obtained from (V;-t,E;-1) by a
0, 1 or 2-extension.
Intuitively, a 2-extension involves deleting two edges and attaching a new, 5-
valent vertex to the three or four endpoints of the deleted edges and two or one
other vertex. The precise definition of a 2-extension will be given in Section 5.3.
There we will show that each 3-isostatic set is the edge set of the terminal graph
of some 3-dimensional Henneberg sequence. The Henneberg Conjecture is that
every 3-dimensional Henneberg sequence terminates in a graph whose edge set
is 3-isostatic. Henneberg sequences in m-space are the subject of Section 5.3.
In Section 5.4, we restrict our attention to Henneberg sequences in 3-space and
prove a weak form of the Henneberg Conjecture. This result is then used to
prove Gluck's Theorem [45).
Another characterization of generic rigidity discussed in the previous chapter
129
130 CHAPTER 5. RIGIDITY IN HIGHER DIMENSIONS
is due to Dress and is based on properties of closed sets. If E is any edge set, the
cliques of E are defined to be the edge sets of the maximal complete subgraphs
of the graph (V(E), E). In Exercise 3.64, you proved that, if Am is any m-
dimensional abstract rigidity matroid for V, if E ~ K = K(V) is a closed set of
Am and if K1. ... , Kh are the cliques of E, then:
{i) E = K1 U ... U Kh;
(ii) V(E) = V(K1) U ... U V(Kh)i
(iii) IV(Ki) n V(K;)I < m, for all distinct i and j;
(iv) r(E) ~ r(K1) + ... + r(Kh)·
If m = 1 above, then equality holds in (iv). In Theorem4.4.2, we proved that if
A2 is a 2-dimensional abstract rigidity matroid for V, then A2 is generic if and
only if equality holds in {iv), for every closed set E of A2.
Let Qm(V) be the 3-dimensional generic rigidity matroid for V and let E ~
K = K(V) be closed. One cannot expect equality to hold in (iv) above. To
see this consider two triangles which share a common edge: each has rank 3
so the sum of their ranks is 6; but, their union clearly has rank 5. The edge
common to the two triangles is counted twice in the sum of the ranks. At a
conference on rigidity held in Montreal in 1987, Dress pointed out that, in 3-
space an "inclusion-exclusion" type formula might work. To see what Dress
had in mind, let Aa be a 3-dimensional abstract rigidity matroid for V and let
E ~ K = K(V) be closed. Let K 11 ••• , Kh be the cliques of E and let H be the
edges of E which belong to more than one of these cliques. For each e e H, let
77(e) denote one less than the number of cliques which contain e. Now suppose
that H is independent and extend it to a maximum independent subset F of E.
If Fi = F n Ki, for each index i, we have:
h h
r(E) = IFI = L IFil- L TJ(e) ~ L r(Ki)- L TJ(e).
i=l eeH i=l eeH
Dress conjectured that the inequality

h
r(E) ~ L r(Ki)- L 77(e)
i=l eeH
holds in all cases and that Aa will be generic if and only if equality holds for all
closed sets E. This conjecture is the subject of Section 5.6 of this chapter.
In the last section, we will consider the possibility of extending the character-
ization of 2-isostatic set in terms of a decomposition into trees. Also discussed
there is the maximal conjecture: Let V be given and let Aa be a 3-dimensional
abstract rigidity for V; then any set E ~ K(V) which is independent (rigid) in
A 3 is independent {rigid) in Qa(IVI). If this conjecture could be proved, it would
characterize the 3-dimensional generic rigidity matroid for V as the maximal
3-dimensional abstract rigidity matroid for V under the natural partial order on
matroids.
5.2. HIGHER DIMENSIONAL EXAMPLES 131
5.2. Higher Dimensional Examples. While examining possible combina-

torial characterizations of generic rigidity in higher dimensions, it is helpful to
keep in mind two critical examples. The double banana illustrated in Figure 1. 7
is clearly an important graph: it is the simplest example which demonstrates
that Laman's Theorem cannot be extended to dimension 3. It and its general-
izations will also have a significant role to play in the consideration of possible
generalizations of the Henneberg Conjecture. Another important set of exam-
ples for the study of rigidity in higher dimensions is the class of the complete
bipartite graphs. Fortunately, the generic rigidity properties of all complete bi-
partite graphs are known in all dimensions. These results are contained in a
1980 paper by Bolker and Roth [12]. We present the Bolker-Roth results which
are necessary for our discussions.
In Exercises 2.38 and 2.39, you computed the rigidity properties of many com-
plete bipartite graphs. These results were then summarized in Theorem 3.11.6.
The simplest case not covered by the above theorem is the complete bipartite
graph Km+2,m+2· In dimensions m = 1, 2 and 3, Km+2,m+2 has too many edges
to be independent i.e. it violates the Laman constraint. However, in dimensions
m = 4 and above, Km+2,m+2 does not violate the Laman inequality and could
conceivably be independent. As it turns out Km+2,m+2 is a cycle in dimension
m, form~ 3.
THEOREM 5.2.1. Let m ~ 3 and n ~ 2m+ 4. Then Km+2,m+2• the edge set
of the complete bipartite graph, is an m-cycle.
PROOF. Let V = { 1, 2, ... , n} and let p be a generic embedding of V into

Rm. Without loss of generality, we may assume that n = 2m+4 and and we may
restrict our attention to the copy of Km+2,m+2 which has U = {1, ... , m + 2}
and W = { m + 3, ... , n} as the its two vertex sets.
We start the proof by showing that all proper subsets of Km+2,m+2 are in-
dependent. By Theorem 3.11.6 and the fact that m ~ 3, the smaller graph
Km+l,m+l• ism-independent. Moreover,
and so, at least two more edges will be required to make it rigid. Let U' s;;; U and
W' s;;; W denote the two vertex classes of this graph and let e be an edge between
two vertices in U'. If e were in the closure of Km+l,m+l> then, by symmetry,
each edge between two vertices of the same block would be in the closure and
Km+l,m+l would be m-rigid. Since this is not possible, we may conclude that
E = Km+l,m+l U {e} ism-independent. Next, let e' be an edge between two
vertices in W. Observe that E is still one edge short of the number needed to
be m-rigid. Suppose that e' were in the closure of E. Then, by symmetry, each
edge between two vertices in W would belong to the closure of E. Again this
leads to the contradiction that E is m-rigid. We conclude that e' is not in the
closure of E and that F = Km+l,m+l U {e, e'} is m-independent.
Now let u denote the only vertex in U- U' and w the vertex in W- W' and
let H be the edge set obtained by attaching u with edges to each vertex in W
while deleting e' and attaching w to each vertex in U while deleting e. That is,
His obtained from F by two !-extensions and is, therefore, m-independent. We
also recognize Has Km+2,m+2- (u,w). By symmetry, all subsets of Km+2,m+2
obtained by deleting one edge are m-independent. Hence, either Km+2,m+2 is
m-independent or an m-cycle. To show that Km+ 2,m+2 is actually an m-cycle
we employ Theorem 2.4.1. We can establish that Km+2,m+2 ism-dependent,
and hence an m-cycle, by showing the existence of a nontrivial resolvable stress
for Km+2,m+2·
Recall that a set of resolvable stresses. for Km+2,m+2 is a set of scalars Buw 1
for u E U and w E W, so that:
L Buw(p(u)- p(w)) = 0, for all wE W
ueu
and
L Buw(p(w)- p(u)) = 0, for all u E U.
weW
Since p(U) is a set of m + 2 points in Rm, the points in p(U) must satisfy two
"independent" relations:
L aup(u) = 0 and L bup(u) = 0,
ueu ueu
where the bu 's are not simply a scalar multiple of the au's. Let a = Eueu au
and b = Eueu bu and, for each u E U, let ru = aub- bua. One easily checks
that:
L
rup(u) = 0 and L
ru = 0.
ueu ueu
Similarly, we construct scalars tw, for w E W, such that:
Ltwp(w) = 0 and L
tw =0.
weW weW
Finally, let Buw = rutw, for all u E U and w E W. The Buw's form a set of
resolvable stresses since
L Buw(p(u)- p(w)) = tw[L rup(u)- (L ru)p(w)]
ueu ueu ueu
= tw[O + Op(w)]
= 0,
for all w E W and similarly for all u E U. 0
We may conclude from this that all suitably large bipartite graphs are depen-
dent in all generic rigidity matroids which contain them. Specifically:
5.3. THE HENNEBERG CONJECTURE 133
COROLLARY 5.2.1. Let m ;?: 3, let h, k ;?: 2 and let n ;?: 2m+ h + k. Then
Km+h,m+k is dependent in Qm(n).
Exercise 5.1. Form= 1, 2 and 3, list the values of h and k for which Kh,k
is m-independent, for which it is m-rigid and for which it is an m-cycle. Do the
best that you can to compile such a list for m = 4.
5.3. The Henneberg Conjecture. The fundamental observation underly-
ing the Henneberg Conjecture is that each 3-isosta.tic set containing more than
three edges has either a vertex of valence three, a vertex of valence four or a
vertex of valence five (Theorem 3.11.3). Theorem 3.11.5 tells us that we may
attach a vertex of valence three to any 3-isostatic set to get another 3-isostatic
set. It also describes just how to attach a vertex of valence four: given a 3-
isostatic set E on four or more vertices, deleting an edge and attaching a new
vertex of valence four to the endpoints of the deleted edge and any other two
vertices in V(E) produces a new 3-isostatic set. What is needed is a set of
conditions under which one may attach a vertex of valence five. And, if we are
to extend the Henneberg construction to even higher dimensions, we need to
fully understand the conditions under which we can attach vertices of valence
k, for (m + 2) ~ k < 2m. A natural set of necessary conditions are easy to
describe. Let the complete graph (V, K) be given, let Am be an m-dimensional
abstract rigidity matroid for V and let F and G be subsets of K. We say that
G is independent over F if r(F U G) = r(F) + IGI. Let S ~ V(F); we say that
S is free in F if, for any U ~ S with lUI > m, there exists a set G ~ K(U)
such that IGI = lUI - m and G is independent over F. There are several simple
consequences of these two definitions:
LEMMA 5.3.1. Let the complete graph (V, K) be given, let Am be an m-di-
mensional abstract rigidity matroid for V and let F and G be subsets of K.
a. If G is independent over F, then G n F = 0 and G is independent.
b. If F is independent and G is independent over F, then F U G is inde-
pendent.
c. LetS~ V(F) and let lSI~ m, then Sis free in F.
d. Let S ~ V(F), let lSI = m + 1 and let H consist of a single edge in
K (S) - F. If H is independent over F, then S is free in F.
e. LetS ~ V(F), let lSI = m + 2 and let H consist of a pair of (vertex)
disjoint edges in K (S) - F. If H is independent over F, then S is free
in F.
PROOF. Parts a, b and c are immediate consequences of the definitions, as
one easily checks.
Let F and S be as described in e above. To verify that S is indeed free over F,
we must verify that the defining condition holds when U is an (m + 1)-element
subset of S. Suppose that U E Sand lUI= m + 1. Since the two edges in H
have no common vertex, K(U) n H is not empty. Let G consist of a single edge
from K(U) n H. Then r(F U G) = IFI + 1. Thus, G is independent over F, as

required. The proof of part d is similar and simpler and is left to the reader. D
Let the complete graph (V, K) be given, let Am be an m-dimensional abstract

rigidity matroid for V. Let F be a subset of K and let S be a subset of V(F)
of cardinality m + k so that K(S) n F contains a k-subset I with the property
that S is free over F- I. Let i e V- V(F), let L = {(i,j) I j e S} and let
E = (F- I) U L. We say that Eisa k-extension of F.
LEMMA 5.3.2. Let the complete graph (V, K) be given and let Am be an m-
dimensional abstract rigidity matroid for V. Let E ~ K, let v e V (E), let S
be the set of neighbors of v, let L be the set of edges with v as endpoint and let
F = E - L. If L is independent over F, then S is free in F.
PROOF. The result is trivially valid if the valence of v is m or less. We assume

that v has valence at least m + 1 and proceed by induction on the valence of v.
In showing that S is free in F, we need only consider the case U = S, the cases
with U a proper subset of S being taken care of by the induction hypothesis.
Choose i e S, let E' = E- (i,v), letS'= S- {i} and let L' = L- (i,v).
Since Lis independent over F, L' is independent over F and (i,v) ¢ (E'). Our
first task is to show that K(S) is not a subset of (E'). Suppose that it is. We
would then have that K(S) and K(S' U {h}) are both subsets of (E'). But,
IV(K(S)) n V(K(S' u {v} ))I= IS- {i}l2:: m.
implying:
(i,v) E (K(S)UK(S'U{v})) ~ (E').
We conclude that there is an edge (i,j) E K(S) so that (i,j) is not in (E'). Let
E" = E' + (i,j) and note that r(E") = r(E). Let F" = F + (j, k). Since
r(E") = r(E') + 1 = r(F) + IL'I + 1 = r(F') + IL'I,

we conclude that L' is independent over F'. By the induction hypothesis, there is
a set I' ~ K (S') so that II' I = IS' 1-m and I' is independent over F'. One easily
checks then that I= I'+ (i,j) has the correct cardinality and is independent
over F thus demonstrating that S is free in F. D
A sequence of graphs (V11 E 1 ), ••• , (Vn. En) is called am m-dimensional Hen-

neberg sequence if (VI. Et) is a complete graph on m vertices and if, for each
index 2:::; j:::; n, (V;,E;) is a k-extension of (V;- 1 ,E;-t) with 0:::; k < m.
THEOREM 5.3.1. Let Am be an m-dimensional abstract rigidity matroid for

V. If the edge set E ~ K = K(V) is isostatic and IV(E)I2:: m, then there is an
m-dimensional Henneberg sequence (VI. Et), ... , (Vn, En) such that (Vn. En) =
(V(E),E).
PROOF. If E is isostatic and IV(E)I = m, then (V(E), E) is the complete

graph on V(E) and (V1 ,Et) = (V(E),E) is the required Henneberg sequence.
We assume then that IV(E)I > m and proceed by induction on IV(E)I. By
Theorem 3.11.5, we may conclude that (V(E), E) contains a vertex i of valence
m+k with 0 :5 k < m. LetS be the set of neighbors of i and let F = E- {(i,j) I
j E S}. Clearly, V(F) ~ V(E) - {i}. Also, it is clear that F is independent.
By the previous lemma, S is free in F. Hence, there exists G ~ K (S) such
that IGI = lSI - m, that G n F = 0 and that H = G U F is independent. By
direct computation, we see that IV(H)I = IV(E)I- 1 and that H is isostatic.
By the induction hypothesis, there is an m-dimensional Henneberg sequence
(Vlt E1), ... ,·(Vn-1 1 En-1), with (Vn-1 1 En-1) = (V(H), H). But, (V(E), E) is
a k-extension of (V(H), H). 0
Exercise 5.2. Fill in the missing computation in the above proof.
If the m-dimensional abstract rigidity matroid Am is not gm (n), we can not
expect each Henneberg sequence to terminate in an isostatic set. For example,
in IR2 , the edge set of the complete bipartite graph K 3 ,3 is generically isostatic
and, hence, the terminal graph in some 2-dimensional Henneberg sequence in
Q2(6). However, if .F(h) is the infinitesimal rigidity matroid obtained by an
embedding h of vertices of K 3 ,3 as the vertices of a regular hexagon with 3
diagonals, this edge set is dependent. Nevertheless it is still the terminal graph
in some 2-dimensional Henneberg sequence.
Exercise 5.3. Produce a 2-dimensional Henneberg sequence for the edge set
of the complete bipartite graph Ka,3· In the infinitesimal rigidity matroid de-
scribed above, identify the first graph in the sequence whose edge set is not iso-
static.
In dimensions 1 and 2, each Henneberg sequence terminates in an edge set
which is generically isostatic and, in his paper, Henneberg [53] used 3-dimen-
sional Henneberg sequences to produce 3-isostatic sets. Hence:
CONJECTURE 5.3.1 (THE HENNEBERG CONJECTURE). Every 3-dimensional

Henneberg sequence ends in an edge set which is generically independent in 3-
space.
If this conjecture is true, then all3-isostatic sets would be uniquely determined

and Qa(n) would be characterized. It seems natural to extend this conjecture
to higher dimensions and this more general conjecture would follow if we could
prove that, for k = 0, ... , m- 1, a k-extension of am-isostatic set is m-isostatic.
By Theorem 3.11.5, we know this to be true fork= 0 and k = 1. In general, we
wish to know the answer to the following question:
If we are given the complete graph (V,K), an edge set E ~ K
and a vertex v of valence m+k in (V(E), E) such that F, the set
of edges in E not containing v, ism-independent and S, the set
of neighbors of v in {V{E), E)), is free in F. May we conclude

that E ism-independent?
Actually, the only case for which the answer to this question is not known is the
case m = 3 and k = 2, i.e. the case needed to attach a vertex of valence five in
dimension three. The answers to this question are listed in Table 5.3. We have
0 1 2 3 k
1 yes yes yes yes yes
2 yes yes yes yes yes
3 yes yes ??? no no
4 yes yes no no no
m yes yes no no no
TABLE 5.3. Is a k-extension of an m-isostatic set always m-isostatic?
already explained the "Yes" entries in the first two columns. The "Yes" answers,
for all k, in the first and second rows follow from the Laman characterization in
dimensions one and two, as we will show with this next lemma.
LEMMA 5.3.3. Let the complete graph (V, K) be given and consider Ym{V).
Let E ~ K, let v be a vertex of valence m + k in (V(E), E), let F be the set of
edges in E not containing v and letS be the set of neighbors of v in (V(E), E).
IfF is independent and S is free in F, then E satisfies the Laman condition.
PROOF. Let H be a nonempty subset of E, we must show that H satisfies
Laman's inequality. If v ¢ V(H), then H ~ F and the conclusion follows
directly from the fact that F is independent. Assume then that v E Hand let
S' = S n V{H). Suppose that IS' I = m + k' with k' > 0. (The simpler case,
IS'I ~ m, will be left as an exercise for the reader.) Since Sis free over F, S'
is free over F and there exists G ~ K(S') such that G n F = 0, IGI = k', and
FUG is independent. We also have V{G) ~ S' ~ V(H). Combining these facts
gives:
IHI = IF n HI+ (k' + m) = I{F n H) u Gl + m

~ (miV(H)- {v}l- (m; 1)) +m
~ miV(H)I- (m; 1),
as required. 0
Exercise 5.4. Complete the proof of this lemma.
Exercise 5.5. Let the finite set V be given and consider Q1{V).
a. Let E ~ K = K(V) and letS~ V(E). Prove that S is free in E if and

only if no two vertices of S lie in the same component of (V(E), E).
b. Let E ~ K be isostatic, i.e. (V(E), E) is a tree. Let i E V- V(E) and let
F be the edge set of any subtree of(V(E), E). Prove that (E-F)U{(i,j) I
j E V(F)} is isostatic.
The double banana, see Figure 1.7, shows that a 3-extension of a 3-isostatic
set need not be 3-isostatic: You will easily verify that the set of neighbors of a
vertex of valence six in this graph is free in the edge set obtained by deleting the
six edges containing that vertex.
Exercise 5.6. Verify that the double banana is a 3-extension of a 3-isostatic

set.
This example may be altered as follows: Replace the righthand most triangle
by a k-circuit and add m - 3 vertices attached to all other vertices in the graph.
The resulting family of graphs demonstrate that, for m ~ 3 and k ~ 3, a k-
extension of an m-isostatic set need not be m-isostatic. This accounts for all of
the "No" entries except those in the column labeled 2.
Exercise 5.7. Consider the extension of the double banana graph described
above and show that it is the asserted counterexample. First, consider the case
where m = 3 and k ~ 3 and note that the righthand "banana" in the k+ 1 example
is a 1-extension of righthand "banana" in the k example. Use this observation
to prove that the counterexample is valid for m = 3 and all k ~ 3. Then verify
that the same observation holds when m > 3.
Maehara [79) and Woodall [private communication) demonstrated the "No"

in the (4,2) position. The relevant example is easy to describe. Start with the
edge set of the complete graph on six vertices from which one edge has been
deleted. This edge set is easily seen to be 4-isostatic. Then one can make seven
2-extensions deleting the fourteen original edges and ending with the complete
bipartite graph Ks,7· In the introduction to this chapter, we proved that Ks, 6
is a 4-cycle; so Ks,1 is not 4-isostatic. Hence, at some step in this Henneberg
sequence, a 2-extension of a 4-isostatic set did not result in a 4-isostatic set. This
example may be easily extended to higher dimensions. Hence, the only open case
is the Henneberg Conjecture.
Exercise 5.8. Show that the Maehara- Woodall counterexample fails before
the last vertex is added. Identify exactly where it fails.
Exercise 5.9. Let m > 4 and adapt this construction to produce an m-

dimensional Henneberg sequence which terminates in the edge set of a graph on
2m+ 4 vertices which contains the cycle Km+2,m+2. This verifies the remaining
"no's" in the column labeled 2 in the table.
5.4. Stresses and Strains. In this section we derive a fundamental formula

for working with infinitesimal rigidity matroids.
Let V = {1, 2, ... , m + 2} and let p be any general embedding of V into Rm
and consider the (mr) x m(m + 2) rigidity matrix of the complete graph on V.
In its condensed form, we have:
Pt-P2 P2 -Pl 0 0
Pt-Ps 0 Ps -Pl 0
Pt-P4 0 0 0
R(p) = Pt-Pm+2 0 0 Pm+2 -Pl

0 P2 -pg Ps -P2 0
0 P2-P4 0 0
0 0 0 Pm+2 -Pm+l
We know that the complete graph on m + 2 vertices is a cycle in any m-
dimensional abstract rigidity matroid and, hence, in any m-dimensional infinites-
imal rigidity matroid. We conclude that the rows of R(p) satisfy exactly one
relation. In the next sequence of lemmas, we will develop an explicit description
of this relation.
The first lemma is a standard result from linear algebra; its proof is included
here for the sake of completeness.
LEMMA 5.4.1. Let M be any matrix with m columns and m+ 1 rows. Denote
the i 'th row byr,
and let di denote ( -1)i times the determinant of the matrix
obtained from M by deleting the i 'th row. Then: E~ 1 c4ri = 0.
PROOF. Append to M as a first column a duplicate of the j'th column. Call

this new (m + 1) x (m + 1) matrix M'. Since M' has a repeated column, it
is singular and has a zero determinant. However expanding the determinant
about the first column gives the negative of the j'th component of the sum
E~1 c4ri. D
LEMMA 5.4.2. Let V = {1, 2, ... , m + 2} and let p be any general embedding
of V into Rm and consider R(p), the rigidity matrix of the complete graph on V.
Let N denote the (m+2) x (m+ 1) matrix with rows (p,,1), fori= 1, ... , m+2,
and let d, denote ( -1)' times the determinant of the matrix obtained from N by
deleting the row (p,,1). Then
{siJ =did; 11 ~ i < j ~m + 2}

is the set of coefficients of a dependency relation for the rows of R(p), i.e. a set
of resolvable stresses for the corresponding complete framework, (V, K (V), p).
5.4. STRESSES AND STRAINS 139
PROOF. Let R denote R(p), let ~; denote the row which contains the entries
Pi - P; and P; -Pi and let Mi denote the {m + 1) x m submatrix of R consisting
of the non-zero rows of the submatrix of the i'th set of m columns of R. That is
Mi is the matrix with (Pi-P;) as rows, for 1 ~ j ~ (m + 2),j :/: i. Finally, let
Mi(j) denote the row (Pi-P;) of~ and Mi(-j) the submatrix of Mi with
this row deleted.
Clearly, if E#i 8ijMi(j) = 0, then E 1 ~i<j~m+2 8ij~J = 0. It is also easy
to see that E#i 8i;~(j) = 0 whenever E#i d;Mi(j) = 0. It is this last
equality that we must now show.
If we can show that d; = {-1)i{-1)iui;det(Mi(-j)), where O'ij = -1 when
i > j and +1 otherwise, then our equality will follow from the previous lemma.
Recall that d; = (-1)i det(N( -j)) where N( -j) denotes the matrix obtained
from N by deleting the j 'th row. Thus, it remains only to show that
( -1)iui; det(Mi( -j)) = det(N( -j)).
Subtracting the row [pi 1] from each row of N( -j), yields the matrix:
Pi-1- Pi 0
0 1
Pi+1- Pi 0
Pm+2- Pi 0
where the row [P; - Pi OJ deleted. The required equality now follows easily. 0
Exercise 5.10. Fill in the details of the proof of this last lemma.
Exercise 5.11. Let V = {1,2,3,4} and let p be any general embedding ofV
into R 2 • As we saw in Chapter 2, we may apply an affine transformation so that
P1 = {0,0), P2 = {1,0), P3 = {0, 1) and P4 = (x,y).
a. Write out R(p), the rigidity matrix for the complete graph on V, under
this embedding.
b. Write out the matrix N, as defined above, compute the (signed) deter-
minants d1 , ..• , d4 and compute the coefficients
{ 8· ·
'·'
= ""i.~.dJ· 11 <- i < J. <- 4}.
c. Verify that sR(p) = 0, where s = {812 1 813,814,823,824, 834).
Exercise 5.12. Using a symbolic computation program - such as Mathe-
matica or Maple} -repeat the previous exercise for an arbitrary embedding p,
i.e. do not apply the affine transformation.
Exercise 5.13. Let V = {1, 2, 3, 4, 5} and let p be a general embedding of

V into IR3 with Pl = (0,0,0), P2 = (1,0,0), P3 = (0,1,0), P4 = (0,0,1) and
Ps = (x,y,z).
a. Write out R(p), the rigidity matrix for the complete graph on V under
this embedding.
b. Write out the matrix N, as defined above, compute the {signed} deter-
minants d1, ... , ds and compute the coefficients {SiJ = dtd; I 1 ~ i <
j ~ 5}.
c. VerifythatsR(p)=O, wheres=(s12, ... ,s45).
We return to our original setting: let V = {1, 2, ... , m + 2} and let p be any
general embedding of V into IRm. Also let {s,3 I 1 ~ i < j ~ m + 2} be a set
of resolvable stresses for the complete graph (V,K(V),p). Let u be any vector
assignment to the images of V under p, i.e. u E (IRm) v. As usual, we will denote
u(i) by lli· It is convenient to think of the ut as m x 1 column vectors and u as
the m(m+2) x 1 column vector obtained by stacking these vectors on top of one
another in order with u1 at the top. As in Chapter 2, we denote the product
(Pi - P;) (ut - u;) by Ui;. We call Ui; the strain induced on ij by u and we
let ii denote the (mi 2) x 1 column with the Uij as entries ordered to match the
ordering of rows in R(p). Then, we have:
R(p)u = ii.
Multiplying both sides on the left by s = (s12, •.. , Sm+l,m+2), we have:
0 = O'ii= sR(p)'ii= sii=
Thus, the strains induced by a vector assignment on the points in p(V) satisfy
a relation with a set of resolvable stresses for the complete graph (V, K(V), p)
as coefficients. The "converse" is also true as we now show.
Assume that {u~; 11 ~ i < j ~ m + 2} is any set of scalars which satisfy the
relation
L SijU~j = 0.
l~i<j~m+2
We wish to show that the u~3 are the strains induced by some vector assignment
u on the points in p(V). Since any two vector assignments which differ by an
infinitesimal motion will induce the same set of strains, the choice of u will not
be unique. Our task is to construct some vector assignment which induces these
strains. We start by choosing u1 to be any vector in IR3 . Next choose u1 to be
any vector among the solutions to the vector equation:
5.4. STRESSES AND STRAINS 141
We contin..ue to define 11i inductively, fori= 1, ... , m+ 1, as a solution to the

system of equations:
(Pi-P;)* (Ui- u;) = uji, for 1 5: j < i.

Since p is a general embedding, this system cannot be inconsistent and must
always have at least one solution. Next we define 11m+2 as the solution to:
(Pm+2- P;) * (11m+2- u;) = uj,m+2• for 1 5: j < m.

We now have defined u so that Ui; = u~; with the possible exception of the pair
(m + 1, m + 2}. However:
L
l~i<j~m+2
Sij~j = 0 = L
l~i<j~m+2
SijUiji
and we conclude that Sm+l,m+2Um+l,m+2 = Sm+l,m+2u:n+l,m+2" Since (V, K(V))

is a cycle, each proper subset of rows of R(p) is independent. Thus none of the
Sij can be zero, we conclude that Um+l,m+2 = u:n+l,m+2· We have proved:
LEMMA 5.4.3. Let V = {1, 2, ... , m + 2} and let p be any general embedding
of V into Rm. Let N denote the (m + 2} x (m + 1} matrix with rows (Pi,1}, for
i = 1, ... , m + 2, and let c4 denote ( -1)i times the determinant of the matrix
obtained from N by deleting the row (pi,1} and let
{si,; = c4d3 11 5: i < j 5: m + 2}.

Let u be any vector assignment to the points in p(V} and let Ui,;, for 1 5: i <
j 5: m + 2, denote the induced strains. Then E 1 ~i<j~m+ 2 Si;Ui; = 0.
Conversely, if {Ui,; 11 5: i < j 5: m + 2} is any set of scalars which satisfy
E 1 ~i<j~m+2 si;Ui; = 0, then there is a vector assignment to p(V) with induced
strains equal to the Ui; .
In this sequence of lemmas, we restricted our attention to frameworks on m+ 2

points. However, these results are clearly applicable to any subset of m+2 points
in a larger framework. We restate the conclusions of our investigation in this
more general setting.
THEOREM 5.4.1. Let the finite set V = { 1, ... , n} and a general embedding p
ofV into Rm be given. Let U = {i1.i2, ... ,im+2} be an (m+2)-subset ofV; let
N denote the (m + 2} x (m + 1} matrix with rows (Pii ,1}, for j = 1, ... , m + 2,
and let d; denote (-1 ); times the determinant of the matrix obtained from N by
deleting the row (Pii , 1} and let
{s;,k = d;dk 11 5: j < k 5: m + 2}.

Let u be any vector assignment to the points in p(V) and let Uiii" denote the
strain induced on the edge (i;, ik). Then E 1 ~j<k~m+2 s;kUiii" = 0.
Conversely, let {Ui; 11 :::; i < j :::; n} be a set of scalars so that
L SjkUiJik = 0,
t:5j<k:5m+2
for each (m+2)-subset {it. i2, ... ,im+2} ofV. Then there is a vector assignment
to the points p(V) with the uii as its induced strains.
Exercise 5.14. Fill in the details of the proof of the converse. Then state and
prove a stronger converse, i.e. one in which fewer conditions must be satisfied.
To illustrate the uses of this last result, we consider Ka,a in the plane. Specifi-
cally, let V = {1, ... ,6} and choose p to be a general embedding with Pt = {0, 0),
P2 = (1, 0), Pa = (0, 1) and Pi = (xi, Yi), for i = 4, 5,6. Suppose that u is an in-
finitesimal motion of the framework (V, E, p), where E = {(i,j) 11:::; i:::; 3,4:::;
j :::; 6}. Then only Ut2, Uta, u2a, U45, U45 and U55 can be nonzero. Restricting
our attention to the 4-subset U = 1, 2, 3, i, where i is 4, 5 or 6, we may construct
N and compute the corresponding coefficients:
dt = -(1- Xi- Yi)
da = -Xi
da = -yi
~ = 1
This gives us three relations for Ut2 1 Uta and u2a:
[(1 -xi - Yi)xi]Ut2 + [(1 -xi - Yi)Yi]Uta + [XiYi)u2a = 0,

for i = 4, 5, 6. Thus, if the framework admits an infinitesimal motion u, the
vertices p4, Ps and P6 must lie on the conic given by:
Ut2(1 -X- y)x + Uta(1 -X- y)y + U2axy = 0;
and one easily checks that Pb P2 and Pa also lie on this conic.
Conversely, if Pt •... , P6 all lie on a conic, then the conic must have the form
a(1 - x- y)x + b(1 - x- y)y + cxy = 0. To see this let
Ax2 +By2 +Cxy+ax+by+d= 0
be the equation of a conic containing the points with coordinates (0, 0), (1, 0)
and (0,1). Substituting 0 for x and 0 for y gives d = 0; substituting 1 for x and
0 for y gives A = -a; and substituting 0 for x and 1 for y gives B = -b. Thus
the equation of the conic can be rewritten as
ax - ax 2 - axy + by - by2 - bxy + cxy = 0, where c = C +a+ b.
Hence, using the construction in the proof of the lemma which precedes the
theorem, we may construct an infinitesimal motion for p 11 P2 and Pa with Ut 2 =
a, Uta = b and u2a = c and then extend this to an infinitesimal motion of the
enti.re framework.
5.5. 2-EXTENSIONS IN 3-SPACE 143
Exercise 5.15. Note that, in the above example, d1 + d2 + ds + d4 = 0. This

is not an accident and is a good check that the computations are correct. Prove
that, in general, L::.i
2 ~ = 0.
Exercise 5.16. Use these techniques to find the conditions under which the
'prism" is not infinitesimally rigid. Hint: let P1, P2 and Ps be the vertices
of one of the triangles of the prism; let p4, P5 and Pe be the vertices of the
other triangle; and let {(1,4), (2, 5), (3,6)} be the edges joining these triangles.
Produce an equation for each of the following 4-subsets: {1,2,3,4}, {1,2,3,5},
{1,2,3,6}, {1,2,4,5}, {1,3,4,6} and {2,3,5,6}.
Exercise 5.17. Let V = {1,2,3,4}, let E = {(1,2),(1,3),(2,4),(3,4)} and
consider a quadrilateral framework (V, E, p). Let u be an infinitesimal motion of
this framework and compute the coefficients of the relation: S14U14 + s23U23 = 0.
Use this relation to explain the motions of this framework. Specifically: When
are the distances between P1 and P4 and between P2 and Ps both increasing or
both decreasing? When does one increase while the other is decreasing? Where
does the transition between these different modes occur?
5.5. 2-Extensions in 3-Space. We now turn to a discussion of 2-extensions
in 3-space. Our plan of attack on 2-extensions in 3-space will proceed as follows:
Let V = {0,1, ... , n} and assume that p is a generic embedding of V into
R3 adjusted so that Po is the origin while {p 11 P2 1 Ps} are the standard basis
vectors. For 4 ~ i ~ n- 1, we let Xi, Yi and Zi denote the coordinates of the
vector Pi and we let Pn = (x, y, z). Let E ~ K = K(V) with lEI = 3(n + 1)- 6.
Assume that n is a vertex of valence 5 in (V, E) and that S = {0, 1,2,3,4}
is the set of neighbors of n in this graph. Assume further that S is free in
F = E - { (i, j) I j E S} and that F is 3-independent. We wish to prove that E
is 3-isostatic. Since E - { (4, n)} is a !-extension of F, it is 3-independent. If we
can find an infinitesimal motion u of (V, E - { (4, n)}, p) with U4n =F 0, then, by
Lemma 2.5.1, we can conclude that E is 3-independent.
Next we describe a technique for constructing just such an infinitesimal mo-
tion. Since lSI= 5 and Sis free over F, there exist two edges (i,j) and (i 1 ,j1 ) in
K(S) but not in F so that F U {(i,j), (i1 ,j1 ) } is 3-isostatic. Hence, we can find
infinitesimal flexes w ofF U {{i,j)} and w 1 ofF U {{i 1 ,j1 )} with wi'j' and w~;
both nonzero. Next we extend each of these infinitesimal flexes to infinitesimal
flexes of E- {{0, n), (4, n)}. Since Won and w&n are nontrivial rational functions
in x, y and z and since p is generic, we conclude that Won =F 0 and w&n =F 0. Let
u = WonW 1 - w&nw.
Thus, u is an infinitesimal motion of E 1 = E - { (4, n)}. We have:

Uij =
I I
WonWij- WanWij =
I
WonWij r
..J., 0·
Thus, u is actually an infinitesimal flex. Furthermore, since E 1 is 3-independent,

u is the only flex of this framework, up to multiplication by a scalar. If we
can show that u 4n =F 0, then we may conclude that E is 3-isostatic. Using the
techniques we have just developed, we may compute u 4n as rational function in
x, y and z. Since the embedding is generic, we may conclude that U4n =F 0 unless
the numerator of this rational function is identically zero. In this latter case,
U4n = 0 and we conclude that E contains a 3-cycle.
So far this approach has been unsuccessful in establishing that a 2-extension
of an arbitrary 3-isostatic set is 3-isostatic. However, it has been successful
in establishing a special case: the case where the neighbors of the vertex of
valence 5 (Pn in the above notation) form a pentagon. Specifically, we will
assume that { (0, 1), (0, 2), (2, 3), (3, 4), (1, 4)} ~ F and that F U { (1, 2), (1, 3)} is
3-independent. See Figure 5.34. Let w be an infinitesimal motion ofF U { (1, 2)}
Pn
Po P4
FIGURE 5.34.
and let w 1 be an infinitesimal motion ofF U {(1, 3)}. Then:
W01 = W02 = W14 = W23 = W34 = W1n = W2n = W = W12 = 0,

with w1a =F 0 and
I I I I I I I I I 0
~=~=~=~=~=~=~=~=~=,
with wb =F 0. Now apply Theorem 5.4.1 to the vertex sets {p1,P2,Pa,Pn,Po}

and {Pl! P2 1 pa, Pn 1 P4}· For the first set, the only relevant coefficients are s12,
s1a, soa and son· These are easily computed, from the matrix N below, to be
s12 = xy, s1a = xz, soa = -zf and son = f, where f = f(x, y, z) = (x+y+z-1).
The relevant coefficients for the second set are computed from N 1 and are: t 12 =
[x/4 - X4j)[yf4 - Y4f), t13 = [x/4 - X4j)[zj4 - Z4), t24 = [Y/4 - Y4f)f and
t4n = -/4/, where /4 = (x4 + Y4 + Z4 - 1).
5.5. 2-EXTENSIONS IN 3-SPACE 145
The resulting equations are:
fwon = -xzw1a + zfwoa,

fw~n = -xyw~2 + zfw~;
f f4w4n = [xf4 - x4f)[zf4 - z4f]w1a + [Yf4 - Y4f]fw24i
f f4w~n = [xf4 - x4f)[yf4 - Y4f]w~2 + [Yf4 - Y4f]fw~4·
We wish to show that wan w~n - w6n W4n is not identically zero (as a function
of x, y, and z). Suppose this expression is identically zero; then so is the poly-
nomial P f4[WonW~n- w6nw4n]· We will show that this supposition leads to a
contradiction by considering the coefficients of the terms involving the variables
x andy. To isolate these terms, set z = 0. This gives:
-xyw~2f{ -z4[xf4 - X4f]w1a + [Yf4 - Y4f]w24} = 0
or, since wb =f 0 and f = (x + y- 1),
-xy(x+y-1){ -z4[(!4 -x4)x-x4y+x4]w1a+ [-y4x+(I4-Y4)Y+Y4]w24} = 0.

One easily computes the coefficient of the xy term to be
-Z4X4W13 + Y4W24i
and the coefficient of x 2 y to be
(z4X4W13 - Y4W24) - z4(!4 - x4)w1a - Y4W24 = 2(z4X4W13 + y4w24) - z4f4w13·

Setting each of these equal to zero, we see z4f4w 1a = 0. However, we assumed
that w1a =f 0 and since p is a general embedding, Z4 and !4 are both different from
zero. Thus, in this special case, a 2-extension of a 3-isostatic set is 3-isostatic.
We now use this special case to prove Gluck's Theorem.
THEOREM 5.5.1 (GLUCK [45]). Let (V,E) be a planar graph with more than
one edge and without isolated vertices, i.e. with V(E) = V. Then E is 3-
independent. Furthermore, E is 3-isostatic if and only if the embedding is a
triangulation.
PROOF. Fix a planar embedding of (V, E). Suppose that (V, E) is not a
triangulation and add a set of edges H so that (V, E U H) is a triangulation.
Let T denote the set of (triangular) faces of this new graph. Counting the edges
around each face and summing over all faces, we have:
3ITI = 2(IEI + IHI) or ITI = ~(lEI+ IHI).

Substituting this value for ITI into Euler's formula, we get:
lVI- (lEI+ IHI) +~(lEI+ IHI) = 2.

Solving for lEI gives:

lEI = 3IVI- 6 -!HI.
It follows at once that, if (V, E) is not a triangulation, then E is not rigid and
therefore not 3-isostatic. If we show that all triangulations are 3-isostatic, then
E U His 3-isostatic and E is 3-independent.
All that remains to prove then is that if (V, E) is a triangulation then E is
3-isostatic. We have just seen that, if (V, E) is a triangulation, then lEI has the
correct number of edges to be 3-isostatic. We proceed by induction on lVI· The
simplest two triangulations are the complete graphs on three and four vertices,
both of which are 3-isostatic. Hence, we assume that (V, E) is embedded as
a triangulation, lVI ~ 5 and all triangulations on fewer than lVI vertices are
3-isostatic.
Suppose that (V, E) contains a 3-circuit C which is not a face. If we "cut"
the graph along C, we get two planar triangulations each with C bounding a
new triangular face. Each of these triangulations has fewer vertices than (V, E).
Hence the edge sets of these triangulations, call them E 1 and E2, are 3-isostatic
and consequently rigid. Since
IV(Et) n V(~)l = IV( C) I = 3,

we conclude that E = Et U ~ is rigid. Thus, by part c of Theorem 3.11.3, E is
isostatic.
Now assume that (V, E) admits no 3-circuit which is not a face. Since the
neighbors of a vertex of valence 3 form a 3-circuit which could only be a face
if (V, E) were the complete graph on 4 vertices, we conclude that {V, E) admits
no vertices of valence 3. By part b of Theorem 3.11.3, we conclude that (V, E)
has a vertex of valence 4 or one of 5. We assert that (V, E) is obtained by a 1 or
2-extension from a triangulation with one vertex fewer.
For the sake of this discussion, let i be a vertex of valence be 5. Let
{j1 ,h,ia,j4,js} be the vertices adjacent to i in cyclic order so that each of
the consecutive pairs, modulo 5, is joined by an edge. These edges form a circuit
which bounds a face in the inherited embedding of
(U, F)= (V- {i}, E- {(i,jt), ... , (i,js)} ).
Suppose (iJ,ja) E F, Then {i,jt,ia} would be the vertices of a 3-circuit in (V,E)

which does not bound a face, contrary to assumption. By symmetry, we may
conclude that none of the five diagonals of this pentagonal face belongs to F. It
follows that (U, F U {(j1, ja), (jt, j4)}) is a triangulation and, by the induction
hypothesis, is 3-isostatic. We apply the special case Henneberg conjecture proved
above and conclude that {iJ,h, ... ,js} is free over F and that Eisa 2-extension
of F.
Similarly, if (V, E) has a vertex of valence 4, E would be a !-extension of F.
In this case, we apply Theorem 3.11.5 to conclude that E is 3-isostatic. D
5.6. THE DRESS CONJECTURE 147
In his Ph.D. thesis, Allen Fogelsanger proved that the 1-skeleton of any tri-
angulation of a 2-dimensional manifold is 3-rigid. The manifold need not be
embedded in 3-space and, therefore, need not be orientable. Of course, if the
2-manifold is not a sphere, the 1-skeleton will have too many edges to be inde-
pendent and hence cannot be 3-isostatic.
Exercise 5.18. Adapt the arguments used above to prove Gluck's Theorem
to prove that the !-skeleton of a triangulation of the projective plane is 3-rigid.
5.6. The Dress Conjecture. We turn next to the Dress Conjecture in di-
mension three and higher. As previously noted, this conjecture was put forward
by Dress in the 1987 Montreal Conference. Dress pointed out that, for a closed
set E with cliques K11 ... , Kh, the rank of E seems to be given by the "inclusion-
exclusion" formula based on rank instead of cardinality. Let H denote the col-
lection of indices {1, 2, . . . , h}. By the Principal of Inclusion-Exclusion, we have
the following formula for the cardinality of E in terms of the cardinalities of its
cliques:
lEI= L (-1} 111 +11 niei Kil
Ir;,H,I-#0
If E is also independent, rank and cardinality are the same for E and all of its
subsets. Giving: The Dress Equation
r(E) = L (-l)III+lr(nieiKi)·
Ir;,H,I-#0
Dress noted that this equation did not break down for the double banana. Let
E be the closure of the double banana. E is obtained from the double banana
by adding the edge between the two vertices of valence 6. E has just two cliques:
two copies of the complete graph on 5 vertices. These two cliques intersect in a
single edge - the edge that was added. Now the rank of E equals the rank of
the double banana which is 17. The rank of each clique is 9 while the rank of
their intersection is 1. Thus, 17 = 9 + 9 - 1 and the Dress Equation holds. No
closed set in g3 (n) is known to violate the Dress Equation.
In 3-space the Dress equation can be simplified. Let E be a closed set and let
K1, ... , Kn be the cliques of E. As noted in the introduction, two cliques can
intersect in at most a single edge. It follows that any set of more than one clique
will intersect in the empty set, a single vertex or a single edge. Furthermore,
only those sets which intersect in a single edge will contribute to the right hand
side of the Dress equation. Let F be the set of edges in E which belong to more
than one clique of E and, fore E F, let He denote the set of indices of the cliques
which contain e. We may rewrite the Dress Equation as:
r(E) =L r(Ki)- L 17(e),

iEH eEF
where
1J(e) = L
l~He,lll>l
( -1)1I1+1r(nieiKi)
= L
I~H.,III>l
(-1)1111
= c~el) - c~el) + ... ± IHel =F 1

= (1- 1)1Hel - (1 -IHel)
= IHel-1
As we will see, the Dress Equation is known to fail in dimension m, for m > 3.
Hence, the Dress Conjecture concerns only 3-space. It has two parts:
CONJECTURE 5.6.1 (THE DRESS CONJECTURE). Let the finite set V be given
and let M be a 3-dimensional abstract rigidity matroid for V. Let E be a closed
set in M, let K11 ... , Kh be the maximal complete subsets of E, and, for each
e E E, let 17( e) be one less than the number of maximal complete subsets which
include e.
(i) r(E) ~ EieH r(Ki)- EeeF 17(e).
(ii) equality holds for all closed sets E of M if and only if M = Qa(V).
It is possible that the first part of the Dress Conjecture may extend to higher
dimensions. However, the second part, which would characterize the generic
rigidity matroid, fails in all higher dimensions. The example which demonstrates
this is the generic cycle Km+2,m+2 in Qm(n), for n;;::: 2m+ 4. If m;;::: 4, we have
noted (and you can easily check) that Km+2,m+2 is not rigid. From this we may
conclude that Km+2,m+2 is a closed set in Qm(n): By symmetry, if one of the
edges between two vertices of one of the vertex set were to belong to the closure
of Km+2,m+2• then all of the edges between vertices in this vertex set would have
to belong to the closure of Km+2,m+2· But, this would imply that Km+2,m+2
were rigid. We conclude that Km+2,m+2 is closed and that the maximal complete
subsets of Km+2,m+2 are its individual edges. In this case, then the right hand
side of the Dress inequality sums to (m + 2) 2 while r(Km+2,m+2) is one less.
Exercise 5.19. Km+2,m+2 is a cycle for m=9. Why is it not a counterex-
ample to the Dress Conjecturef
Exercise 5.20. Let the finite set V be given and let E be a closed set in
Qa(V). Let K1, ... , Kh be the cliques of E and let F denote the set of edges in
E which belong to more than one clique. Show that, IfF is a subset of a single
clique, then equality holds in the Dress Inequality.
One application of the 2-dimensional version of the Dress formula was the
proof that all 6-connected graphs are 2-rigid. Perhaps the validity of the Dress
Conjecture would imply a similar theorem for 3-space. Lovasz and Yemini [73]
5.7. OTHER CONJECTURES 149
suggested that m(m +I)-connectivity might also imply m-rigidity when m > 2.
They noted in their paper that the construction of a 5-connected, but not 2-rigid
graph, generalizes to an (m(m +I)- I)-connected graph which is not m-rigid,
showing that their "suggestion" is the best one can hope for.
Exercise 5.21. Construct the graph G as follows: replace each vertex in Kh,h
with a copy of Kh, attaching one edge of the original Kh,h to each vertex of
the inserted copies of Kh. Thus, G has 2h2 vertices, each of valence h. Show
that G is h-connected. Show that an independent set in G can contain at most
2h(mh- (mt 1)) + h 2 edges but needs to have rank m{2h2 ) - (mt 1), if G is to
be rigid. Conclude that G is not m-rigid if h = m( m + I) - I.
5. 7. Other Conjectures. Another characterization of generic rigidity in 2-
space that is a candidate for extension to 3-space is the "three tree characteri-
zation" due to Crapo:
The edge set E is 2-isostatic if and only if graph (V(E), E) is
the union of three, edge-disjoint trees such that each vertex in
V(E) belongs to exactly two of these trees and that no two of
the trees have subtrees (on two or more vertices) with the same
span {Theorem 4.9.6).
The corresponding decomposition for an m-isostatic set in m-space would be into
(~ 2 ) trees with each vertex belonging to exactly m of these trees and satisfying
some nondegeneracy condition. The next exercise verifies that these are the
appropriate numbers and gives us some insight into the degeneracy condition.
Exercise 5.22. Let the positive integer m and the finite set V be given and let
h ;::: (mr! 2) be a second integer. Fori = I to h let 7i ~ K(V) so that (V(7i), 7i)
is a tree. Assume that the trees are edge disjoint and that each vertex in V is in
V(7i), for exactly m of the indices. Prove that: lEI ~ m!V(E)I- (mr! 2), with
equality holding if and only if h = (mr! 2) •
At the very least, the degeneracy condition must be designed to insure that
Laman's condition is satisfied. The exercise gives us the least restrictive degener-
acy condition needed to insure Laman's condition. Given a finite set V, a subset
U of V and a family {7i I i = I, ... , h} of edge disjoint trees with V(7i) ~ V,
the family of trees consisting of the (nonempty) components of 7i n K(U), for
i = I, ... , h, will be called the family of trees induced by {7i I i = I, ... , h} on
U. A family of edge disjoint trees is said to be an m-dimensional C-L {Crapo-
Laman) family for V if, for each subset U of V of cardinality m + 2, the family
of trees induced on U contains at least (mt2) trees.
THEOREM 5.7.1. Let the positive integer m and the finite set V be a given
and let {7i I i = I, ... , h} be an edge disjoint family of trees with V(7i) ~ V,
for i = I, ... , h. Then E = Ut=l, ... ,h7i satisfies Laman's condition if and only
if {7i I i =I, ... , h} is an m-dimensional C-L family for V.
Exercise 5.23. Prove this theorem.

Note that this is the natural extension of the Crapo characterization theo-
rem for 2-space and has the Crapo Theorem as a corollary. Another corollary,
originally proved by Crapo using a decomposition theorem of Thtte, is:
COROLLARY 5.7.1. IfV is a finite set and E ~ K(V) ism-independent then
E is the union of the edge sets of an m-dimensional C-L family for V.
Crapo made no specific conjecture for higher dimensions. However, he stated
that there may be a characterization of generic rigidity in higher dimensions in
terms of this tree decomposition if the appropriate degeneracy condition could
be found. It is clear, from the theorem, that the degeneracy condition just
introduced is not the appropriate condition: By the theorem, we see that the
double banana graph Figure 1. 7, which satisfies the Laman cqndition, must admit
a 3-dimensional C-L family of trees.
Exercise 5.24. Find a 3-dimensional C-L family of trees for the double ba-
nana graph.
The final proposed characterization of generic rigidity in m-space that we will
discuss is the Maximal Conjecture. It was stated in Section 2.5. If M 1 and M 2
are both matroids on K, we say that M1 majorizes M 2 and write M 1 ~ M 2 if
each independent set in M2 is also independent in M 1 . Theorem 2.5.5 character-
izes 9m (n) as the unique maximal m-dimensional infinitesimal rigidity matroid
on K(V) (where V = {1, ... , n}) relative to the relation "~". Since the concept
of infinitesimal rigidity matroid has not been characterized combinatorially, this
result does not yield a combinatorial characterization of generic rigidity. But
this result leads to:
CONJECTURE 5.7.1 (THE MAXIMAL CONJECTURE). Let the finite set V and
the integer m be given. If .Am is any m-dimensional abstract rigidity matroid on
V, then 9m(V) ~.Am·
As we saw, this conjecture follows as simple consequence of Laman's Theorem
in dimension 2. In dimension 3 it is not difficult to verify that the Maximal
Conjecture is implied by both the Henneberg Conjecture and Dress Conjecture.
THEOREM 5.7.2. If the Henneberg Conjecture holds then the Maximal Con-
jecture is valid in dimension 3.
THEOREM 5.7.3. If the Dress Conjecture holds then the Maximal Conjecture
is valid in dimension 3.
The Maximal Conjecture is false in 4-space. This was proved by N.J. Thurston
in his Bachelor of Arts Thesis [128]. Fundamental to his counterexample is the
following observation concerning the nature of the important counterexamples
in rigidity theory: the double banana in 3-space and K6,6 in 4-space. Each is
an "unexpected" cycle. The double banana is predicted to be 3-isostatic by
5.7. OTHER CONJECTURES 151
Laman's condition and Ks,s is predicted to be 4-independent by both Laman's

condition and the extension of the Henneberg conjecture to 4-space. That the
double banana is a 3-cycle follows from a matroid argument: apply the third
cycle axiom to copies of the cycle intersecting in a single edge. That Ks,s is a
4-cycle follows from the structure of 4-space but not from a matroid argument.
What Thurston showed was that, if you started with Q4 (V) on a 12-set V and
simply add the edges sets of each Ks,s to the collection of bases, you get another
4-dimensional abstract rigidity matroid which clearly majorizes Q4(V) but is not
equal to it.
The situation in 3-space, however, is more promising. The only known non-
rigid cycles in Y3 (n) arise from amalgamations of cycles forced by Laman's con-
dition. If, however, the Maximal Conjecture is also false in dimension 3, then
the theorems above imply that none of the approaches described here for charac-
terizing generic rigidity in 3-space are sufficient, and new insight into the nature
of 3-space will be required.
References
1. A.D. Alexandrov, (1950). Convex polyhedra (in Russian), Moskow.

German translation: Convexe Polyeder. In Chapter 10 the au-
thor proves that if vertices are added to the edges of a strictly
convex polygon and the faces are then triangulated, the result-
ing !-skeleton is infinitesimally rigid, and hence rigid. (This
description is taken from (2).)
2. L. Asimow and B. Roth, (1978). Rigidity of graphs, 'Irans. Amer. Math.
Soc. 245, 279-289.
Introduces rigidity from the point of view of algebraic geom-
etry. Proves that if lEI < miVI - (m + l)m/2, then (V, E) is
not generically rigid, as well as the fact that complete graphs
are the only graphs rigid in all embeddings in all dimensions.
3. L. Asimow and B. Roth, (1979). Rigidity of graphs II, SIAM J. Appl. Math.
68 1, 171-190.
Introduces infinitesimal rigidity. Defines structures (rods, pan-
els, etc.), and general position (generic for all edge sets). Proves
that infinitesimal rigidity, generic rigidity, and rigidity are
equivalent if the vertices are in general position, in this sense.
Proves for dimension 2: 0 and !-extensions of independent sets
are independent. If E 1 and ~ are rigid edge sets and have 2
common vertices, then E1 U ~ is rigid. If E 1, E2 and E3 are
rigid and if each pair has a distinct point in common, then
E1 U E2 U E3 is rigid. A totally dependent set (cycle) D con-
tains 2jV(D)I-2 elements any IDI-1 of which are independent
and rigid. Proves for dimension 3: Alexandrov's extension of
Cauchy's Theorem, Gluck's Theorem, and that independent
sets satisfy Laman's Condition.
4. J. A. Baglivo and J. E. Graver, (1983). Incidence and symmetry in design
and architecture, Cambridge Univ. Press.
Contains an exposition of the grid bracing problem.
5. D. Barnet and B. Griinbaum, (1969). On Steinitz's Theorem concerning
convex 3-polytopes and on some properties of planar graphs, in "Many
153
154 REFERENCES
Facets of Graph Theory", Springer Lecture Notes 110, 27-40.

This paper is devoted to two proofs of Steinitz's Theorem: A
graph is isomorphic to the 1-skeleton of some convex poly-
hedron in 3-space if and only if the graph is planar and a-
connected.
6. G. T. Bennett, (1912). Deformable octahedra, Proc. London Math. Soc. (2)
10, 309-343.
7. W. Blaschke, (1920). Wackelige Achtftache, Math. Z. 6, 85-93.
8. E. Bolker, (1979). Bracing rectangular frameworks II, SIAM J. Appl. Math.
36, 491-508.
This paper describes the minimal redundant sets of diagonal
braces in a grid of cubes in space. No proofs.
9. E. Bolker, (1977). Bracing grids of cubes, Environment and Planning B 4,
157-172.
Simplified version of [8].
10. E. Bolker and H. Crapo, (1977). How to brace a one story building, Envi-
ronment and Planning B 4, 125-152.
Simplified version of [11].
11. E. Bolker and H. Crapo, (1979). Bracing rectangular frameworks, Siam J.
Appl. Math. 36, 473-490.
An introduction to infinitesimal rigidity applied to the grid
bracing problem. Shows equivalence to bipartite graph model.
Solves bracing problem for a planar grid in space supported
by vertical rods of uniform length from a solid floor. (Contains
[28].)
12. E. Bolker and B. Roth, (1980). When is a bipartite graph a rigid frame-
work'!, Pacific J. Math. 90, 27-44.
The relationship between infinitesimal rigidity and the stress
space of a framework is studied in the special case that the
framework· is bipartite. The following results are obtained:
K4,n (for n > 5) and Km,n (for m, n > 4) are generically rigid
in 3-space. K6,7 has enough edges to be generically rigid in
4-space but is not. A Ka,a framework in the plane is infinites-
imally rigid unless the vertices lie on a conic. A Kd+I,d(d+l)/2
framework in d-space such that each vertex set spans d-space
is infinitesimally rigid unless the vertices lie on a conic. Several
examples are discussed. (See also [135] & [139].)
13. 0. Bottema, (1967). Flexible hexagons, lndigat. Math. 29, 151-155.
Shows that there are classes of non-singular flexible hexagons
the opposite elements of which are not equal. (See also [66].)
14. R. Bricard, (1895). Reponse a Question 976, Intermediaire Recherches
Math. 2, 245.
REFERENCES 155
15. R. Bricard, (1897). Memoir sur la theorie de l'octaedre articule, J. Math.

Pure and Appl. 5 3, 113-148.
A complete study of the embeddings of the 1-skeleton of the oc-
tahedron in 3-space. Two main results: All embeddings which
extend to embeddings of the 2-skeleton are rigid. There is an
embedding of the 1-skeleton which is not rigid.
16. C. R. Calladine, (1978). Buckminster Puller's "tensegrity" structures and
Clerk Ma:cwell's rules for the construction of stiff frameworks, Intemat. J.
Solids and Structures 14, 161-172.
A study of some particular frameworks in space.
17. A. L. Cauchy, (1905). But les polygons et les polyedres, Oevres Completes
d'Augustin Cauchy 2e Serie Tom 1, 26-38.
A proof of: The 2-skeleton of a strictly convex polyhedron in
3-space is rigid.
18. R. Connelly, (1975). An attack on rigidity I, II, Bull. Amer. Math. Soc. 81,
566-569.
A proof of: Any embedded suspension of a polygon circle is
rigid in 3-space.
19. R. Connelly, (1976). An immersed polyhedral surface which flexes, Indiana
Univ. Math. J. 25, 965-972.
A review of Cayley's result and some examples.
20. R. Connelly, (1978). A counter example to the rigidity conjecture for poly-
hedra, Inst. Haut. Etud. Sci. Publ. Math. 47, 333-335.
The first counter example to Euler's conjecture.
21. R. Connelly, (1978). A flexible sphere, Math. Intelligencer 1, 130-131.
22. R. Connelly, (1979). The rigidity of polyhedral surfaces, Math. Mag. 52,
275-283.
A short history of Euler's conjecture and a discussion of several
counter examples.
23. R. Connelly, (1980). The rigidity of certain cabled frameworks and the
second-order rigidity of arbitrary triangulated convex surfaces, Advances
in Math. 37, 272-298.
The concept of n'th-order rigidity is introduced to deal with
frameworks that are rigid but not infinitesimally rigid, i.e. first
order rigid. He proves that n'th order rigidity implies rigidity.
He then extends Alexandrov's result proving that all convex
polyhedra with triangular faces are rigid. Finally, he applies
the theory of second-order rigidity to tensed frameworks.
24. R. Connelly, (1982). Rigidity and energy, Invent. Math. 66 1, 11-33.
25. R. Connelly, (1990). On generic global rigidity, in "Applied Geometry and
Discrete Mathematics", DIMACS Ser. Discrete Math. Theoret. Comput.
Sci. 4, 147-155.
Ks,5 is shown to be not generically globally (i.e. strongly) rigid
156 REFERENCES
in dimension 3, which disproves Hendrickson's conjecture that

a graph is globally rigid in Rd if it is (d + 1)-connected and
generically redundantly rigid in Rd.
26. R. Connelly and H. Servatius, {1993}. Higher-order rigidity- What is the
proper definition 'I to appear in Discrete & Computational Geometry.
Exhibits an example of a finite mechanism whose configuration
space contains a cusp. The cusp point is a framework which
turns out to be third order rigid, even though the framework
is not rigid, calling into question the whole notion of higher
order rigidity.
27. R. Connelly and W. Whiteley, {1990}. Second-order rigidity and pre-stress
stability for tensegrity frameworks, preprint.
Shows that first-order rigidity implies pre-stress stability im-
plies second-order rigidity implies rigidity, for any framework,
and that none of these implications is reversible, even for bar
frameworks.
28. H. Crapo, {1977). More on bracing a one story building, Environment and
Planning B 4, 153-156.
Considers the bracing problem for planar grids in space sup-
ported by vertical rods of uniform length from a solid floor.
(Incorporated into (11).)
29. H. Crapo, {1979}. Structural rigidity, Structural Topology 1, 26-45.
This paper summarizes the then available general theory of
rigidity in 2 and 3 space.
30. H. Crapo, {1982}. The tetrahedral/octahedral truss, Structural Topology 7,
52-60.
He shows that this truss is not infinitesimally rigid.
31. H. Crapo, {1991}. Invariant-theoretic methods in scene analysis and struc-
tural mechanics, Journal of Symbolic Computation 11 5/6, 523-548.
Contains the derivation of pure conditions via complementa-
tion (Hodge star) in the exterior algebra of an independent
configuration. Remarks on Cayley factorization of projective
conditions.
32. H. Crapo, {1993}. On the generic rigidity of structures in the plane, Ad-
vances in Applied Math., to appear.
Gives a new characterization of generic rigidity in the plane:
A graph G is generically rigid in the plane iff it is the edge
disjoint union of three connected subgraphs G, of G such that
every vertex of G is in exactly two of those subgraphs and no
subgraph of G having at least one edge is spanned by subgraphs
of two of the three subgraphs Gi. This characterization leads
to an efficient algorithm to decide whether a given graph is
generically rigid in the plane.
REFERENCES 157
33. H. Crapo and J. Ryan, (1986). Spatial realizations of linear scenes, Struc-
tural Topology 13, 33-68.
34. H. Crapo and W. Whiteley, (1982). Statics of frameworks and motions of
panel structures: a projective geometric introduction, Structural Topology
6, 42-82.
Two traditionally separate fields, the statics of bar and joint
frameworks and the instantaneous kinematics of articulated
panel structures, are here united in a novel and appropriate
projective geometry setting. Exterior algebra (or Cayley al-
gebra) is employed to represent equilibrium loads and static
stresses on a bar and joint framework, and to represent in-
stantaneous motions of an articulated panel structure. In the
special case of a framework and a panel structure, of a spherical
polyhedron, a natural algebraic isomorphism between static
stresses in the framework and instantaneous motions of the
panel structure is exhibited. Under a suitable path condition
on the stresses, this isomorphism is extended to frameworks
and panel structures based on arbitrary oriented polyhedra.
35. H. Crapo and W. Whiteley, (1993). Plane stresses and projected polyhedra,
Structural Topology, to appear.
36. H. Crapo and W. Whiteley, (1993). Stressed frameworks and projected 4-
polytopes, to appear.
37. H. Crapo and W. Whiteley, (1993). The Geometry of Rigid Structures,
Encyclopedia of Math., Cambridge University Press.
38. H. R. Cox, (1936). On the synthesis and analysis of simply-stiff frameworks,
Proc. London Math. Soc. (2) 40, 203-216.
A study of generic structures in 2 and 3-space. A variety of
results is proved concerning ways to add vertices to generically
rigid frameworks or join rigid frameworks which result in larger
generically rigid frameworks.
39. M. Dehn, (1916). tiber die Starrheit konvexer Polyeder, Math. Ann. 77,
466-473.
40. A. C. Dixon, (1899/1900). On certain deformable frameworks, Messenger
Math. 29, 1-21.
Certain planar linkages are constructed and then used to prove
a geometric theorem. Includes a study of all linkages with 1
degree of freedom, 4 pivots, 4 joints, and 8 rods.
41. J. Edmonds, (1965). Minimum partition of a matroid into independent sub-
sets, J. Res. Nat. Bur. Standards B 69, 67-72.
Describes an efficient algorithm to decompose a given graph
(matroid) into k disjoint trees (k bases) whenever possible.
42. N. V. Efimov, {1962). Qualitative problems in the theory of deformable
structures, Translations Amer. Math. Series 1 6, 274-423.
158 REFERENCES
43. L. Euler, (1862). Opera Postuma I, Petropoli, 494-496.

This is the paper in which Euler made his conjecture on the
rigidity of closed polyhedral surfaces in 3-space.
44. A. Fogelsanger, (1988). The generic rigidity of minimal cycles, Ph.D. The-
sis, Cornell University.
Proves that the 1-skeleton of any abstract triangulation of a
2-manifold is generically rigid in 3-space.
45. H. Gluck, (1975). Almost all simply connected closed surfaces are rigid, in
"Geometric Topology", Springer Lecture Notes 438, 225-239.
The author introduces both rigidity and infinitesimal rigidity
and proves that the 1-skeleton of any triangulated simply con-
nected closed surface in 3-space is generically rigid.
46. M. Goldberg, (1978). Unstable polyhedral structures, Math. Mag. 51, 165-
170.
Discusses several specific surfaces: rigid but not strongly rigid;
rigid but not infinitesimally rigid; and Connelly's counter ex-
ample.
47. Jack E. Graver, (1991). Rigidity Matroids, SIAM J. Discrete Math. 4, 355-
368.
Introduces the concept of an abstract rigidity matroid.
48. M. Grotschel, L. Lovasz and A. Schrijver (1988). Geometric Algorithms and
Combinatorial Optimization, Springer-Verlag, Berlin.
49. B. Griinbaum, (1978). Lectures in lost mathematics, Univ. of Washington.
Discusses the history of rigidity theory.
50. B. Griinbaum, (1988). Rigid plate frameworks, Structural Topology 14, 1-8.
Studies plate frameworks in the plane in which every vertex
intersects exactly two plates.
51. Open University, (1981). Kinematics, The Open University Press.
This text forms part of the correspondence element of an Open
University Third Level Course. Investigates ways in which var-
ious rigid mechanical components can be interconnected so as
to form a kinematic system of specific type. Kinematic struc-
tures are modeled by two different graph representations, the
direct graph and the interchange graph. Emphasis is given to
the combinatorial aspects of the subject.
52. B. Hendrickson, (1992). Conditions }or unique graph realizations, SIAM J.
Comput. 21, 65-84.
Shows that flexes of generic structures are manifolds, and iden-
tifies generic redundant rigidity as a necessary condition for a
generic structure to have a unique realization.
53. L. Henneberg, (1968). Die graphische Statik der starren Systeme Leipzig
1911, Johnson Reprint.
Summary of results of the German school of geometric statics
REFEI_tENCES 159
from the last century, statics and mechanics of frameworks

with replacement principles.
54. D. Hilbert and S. Cohn-Vossen, (1952). Kinematics, Chapter 5, in "Geom-
etry and the Imagination", Cheslea, New York.
Discusses linkages in the plane and in 3-space.
55. M. Homma and H. Maehara, (1990). Algebraic distance graphs and rigidity,
Trans. Amer. Math. Soc. 319 2, 561-572.
Proves that an algebraic distance graph, i.e. a graph in which
two vertices are adjacent if the distance between them is an
algebraic number, is complete if an only if it is rigid.
56. A. W. Ingleton, (1971). Conditions for representability and transversality
of matroids, Proc. Fr. Br. Conf. 1970, Springer Lecture Notes 211, 62-67.
57. P. J. Kahn, (1979). Counting types of rigid frameworks, Invent. Math. 55,
297-308.
This paper includes proofs of: For a fixed, graph there is a
finite algorithm which will decide if a given embedding is rigid
or not; For a fixed graph, the number of embeddings is finite
up to homotopic embeddings.
58. G. Kalai, (1984). Weakly saturated graphs are rigid, Annals Discrete Math.
20, 189-190.
A graph G of order n is weakly d-saturated if there is a se-
quence of graphs G = Go f; ... f; Gt = Kn such that Gi+l has
one edge more than Gi and the number of complete subgraphs
of order din Gi+l is strictly greater than that number in Gi.
Shows that every general embedding of a weakly d-saturated
graph Rd- 2 is rigid.
59. G. Kalai, (1985). Hyperconnectivity of graphs, Graphs and Combinatorics
1, 65-79.
A new notion of connectivity is introduced by defining a ma-
troid M: on the edges of K1c via exterior algebra. A graph
on n vertices is defined to be k-hyperconnected if its edge set
spansM:. The connection between k-hyperconnectivity and
rigidity is explored. For example, Mf is the connectivity ma-
troid, and a 2-hyperconnected graph is generically rigid in the
plane.
60. G. Kalai, (1990). Symmetric matroids, J. Comb. Theory B 50, 54-64.
Considers symmetric matroids, including rigidity matroids,
and their growth functions.
61. R. Kanayama, (1933). Bibliography on the theory of linkages, Tohoku Math.
J. 37, 294-319.
A bibliography of 306 papers on linkages published between
1631 and 1931.
160 REFERENCES
62. A. B. Kempe, (1876). On a geneml method of describing plane curves of

then 'th degree by linkworks, Proc. London Math. Soc. 7, 213-216.
This paper contains a description of several basic linkages
which in various configurations yield any curve defined by a
polynomial in x and y set equal to zero.
63. A. B. Kempe, (1877). How to dmw a stmight line, London, Macmillan.
An elementary lecture on the theory and history of linkages.
64. G. Laman, (1970). On Gmphs and rigidity of plane skeletal structures, J.
Engrg. Math. 4, 331-340.
This paper gives the first combinatorial characterization of
generic rigidity in the plane.
65. H. Lamb, {1928). Statics, Cambridge Univ. Press.
66. H. A. Lauwerier, (1966). A note on flexible hexagons, Proc. Nether. Acad.
Wet. 69, 33Q-334.
Shows that a spatial hexagon the sides and angles of which are
given is flexible if and only if its opposite elements are equal.
(See also [13].)
67. H. Lebesgue, {1907). Octaedres articules de Bricard, Enseign. Math. (2)
13, 175-185.
This paper discusses Bricard's flexible octahedron.
68. H. Lebesgue, (1909). Demonstmtion completes du theorem du Cauchy sur
l'egalite des polyedres convexes, Intermediaire Recherches Math. 16, 113-
120.
69. H. Liebmann, {1920). Ausnahmefachwerke und ihre Determinante,
Sitzungsber. Bayer. Akad. Wiss. 187-227.
70. L. Lovasz, {1979). Combinatorial Problems and Exercises, North Holland.
71. L. Lovasz, {1977). Flats in matroids and geometric gmphs, Combinatorial
Surveys, Proc. 6th British Combinatorial Conf., P. Cameron, ed., Academic
Press, New York, 45-69.
72. L. Lovasz and A. Recski, (1982). Selected topics of matroid theory and its
applications, Rendiconti del circolo Matematico di Palermo, II 2, 171-185.
73. L. Lovasz andY. Yemini, (1982). On Generic rigidity in the plane, SIAM
J. Alg. Disc. Methods 3, 91-98.
This paper introduces infinitesimal and generic rigidity. It
shows that generic rigidity is combinatorial. It contains a proof
of the cover theorem for the planar generic case and uses this
result to prove: A graph is generically independent if and only
if doubling any edge results in a graph which is the union of
two forests. Every 6-connected graph is generically rigid in the
plane. With an example the last result is shown to be best
possible.
74. L.A. Lyusternik, {1963). Convex figures and polyhedm, Dover Publications.
REFERENCES 161
75. J. C. Maxwell, (1864). On reciprocal figures and diagrams of forces, Philos.

Mag. 4 27, 250-261.
also XXVI COLLECTED PAPERS, Cambridge Univ. Press
(1890). Let (V, E) be a graph and let f be the map from real
miVI-space to real lEI-space which has V, the space of infinites-
imal motions of an embedding of (V, E), as kernel. Then the or-
thogonal complement of im(/) is the space of internal stresses
of that embedding. It is not clear that Maxwell thought of
things in this way. However, his study of internal stresses does
give rigidity results because of this relation. Maxwell studies
the space of stresses for planar frameworks by means of re-
ciprocal figures. Given a framework, a second framework is
reciprocal to the first if there exists a 1-1 correspondence be-
tween the edges such that: 1, corresponding lines are parallel;
2, the edges corresponding to a vertex bond in one figure form
a circuit in the other figure. (See [136).)
76. J. C. Maxwell, (1869-1871). On reciprocal figures, fmmes and diagrams of
force, 'frans. Royal Soc. Edinburgh 26, 1-40.
This paper is an extension of [75).
77. H. Maehara, (1988). A simple proof for the maximal planar case of Steinitz's
theorem, Bull. Inst. Math. Acad. 16 4, 373-376.
78. H. Maehara, (1989). On the rigidity of complete multipartite graphs,
Ryukyu Math. J. 2, 29-47.
79. H. Maehara, (1991). On Graver's conjecture concerning the rigidity problem
of graphs, Discrete & Computational Geometry 6 4, 339-342.
This paper shows that a 2-extension of a 4-isostatic set is
not necessarily 4-isostatic. The counterexample presented here
may easily be extended to higher dimensions.
80. H. Maehara, (1991). A rigid unit-bar-fmmework without triangles, Mathe-
matica Japonica, 36 4, 681-683.
An elementary construction of a rigid triangle-free bar and
joint framework in the plane in which all bars have equal
length.
81. H. Maehara, (1991). Distances in a rigid unit-distance graph in the plane,
Discrete. Applied Math. 31 2, 193-200.
82. H. Maehara, (1992). Extending a flexible unit-bar fmmework to a rigid one,
Discrete Math. 108 1/3, 167-172.
83. E. W. Parks, (1965). Braced fmmeworks, Pergammon Press, London.
84. A. V. Pogorelor, (1973). Extrinsic geometry of convex surfaces, 'fransl.
Math. Monographs 35, Amer. Math. Soc.
Alexandrov's work is presented here.
85. A. Pugh, (1976). An introduction to tensegrity, Univ. of Calif. Press.
Some architectural examples are discussed here.
162 REFERENCES
86. A. Recski, (1982). Engineering applications of matroids- a survey, A. Bar-

lotti, (ed) Matroid Theory and its Applications, Liguori editore, Napoli,
299-321.
87. A. Recski, (1984). A network theory approar.h to the rigidity of skeletal
structures. Part 1. Modeling and interconnection, Discrete Applied Math.
7, 313-324.
88. A. Recski, (1984). Statics and electrical network theory: A unifying role of
matroids, W. R. Pulleyblank (ed), Progress in Combinatorial Optimization,
Academic Press, London, 307-314.
89. A. Recski, (1984). A network theory approach to the rigidity of skeletal
structures. Part 2. Laman's Theorem and topological formulae, Discrete
Applied Math. 8, 63-68.
90. A. Recski, (1984). Applications of combinatorics to statics- a survey, Ren-
diconti del circolo Matematico di Palermo, II, 3, 237-247.
91. A. Recski, (1984). A network theory approach to the rigidity of skeletal
structures. Part 9. Electric model of planar frameworks, Structural Topol-
ogy 9, 59-71.
92. A. Recski, (1988/89). Bracing cubic grids - a necessary condition, Discrete
Math. 73, 199-206.
93. A. Recski, (1989). Matroid Theory and its Applications in Electrical Net-
work Theory and in Statics, Springer Verlag, Berlin.
94. A. Recski, (1989). Symmetric bmcing of one-story buildings with cables
and asymmetric bmcings of one-story buildings with rods, Symmetry of
Structures, Budapest, II, 471-472.
95. A. Recski, (1991). One Story buildings as tensegrity frameworks II, Struc-
96. A. Recski, (1991). Applications of combinatorics in structural engineering,
in "First Workshop on Applications of Combinatorial Optimization in Sci-
ence and Technology", Rutgers University, New Brunswick NJ, 275-282.
97. A. Recski, (1992). Applications of combinatorics to statics-A second survey,
Discrete Math. 108 1/3, 183-188.
98. A. Recski, (1993). Combinatorics in Electrical Engineering and Statics,
Handbook in Combinatorics, R. Graham, M. Grotschel, and L. Lovasz
(eds.) North Holland, Amsterdam.
99. A. Recski and W. Schwarzler, (1992). One-story buildings as tensegrity
frameworks III, Discrete Applied Math. 39, 137-146.
100. I. G. Rosenberg, (1979). Structuml rigidity of bar & joint and cabled frame-
works, Proc. lOth S-E Conf. Combinatorics, Graph Theory and Computing,
843-848.
The rigidity of frameworks is studied in terms of an infinite sys-
tem of equations obtained from power series expansions of tra-
jectories. The system is used for deriving sufficient conditions
for rigidity going beyond the standard infinitesimal rigidity.
REFERENCES 163
101. I. G. Rosenberg, (1980). Structural rigidity I, Annals Discrete Math. 8,

143-161.
Structural rigidity of tensegrity (i.e. bar, cable, and strut)
frameworks is studied. The frameworks are fixed in space so
that no euclidean rigid motions may occur. A rigidity criterion
reduces the problem to the check whether an infinite system of
equations has trivial solutions only. The paper concludes with
the infinitesimal rigidity of tensegrity frameworks.
102. I. G. Rosenberg, (1986). Structural rigidity II, Discrete Applied Math. 13,
41-59.
Structural rigidity of bar frameworks is studied in the case
that the corresponding rigidity matrix has rank one less than
full. A rigidity criterion for these almost infinitesimally rigid
frameworks is derived from a similar criterion in [101).
103. B. Roth, (1981). Rigid and flexible frameworks, Amer. Math. Monthly 88,
6-20.
An expository paper on rigidity.
104. B. Roth and W. Whiteley, (1981). Tensegrity frameworks, Trans. Amer.
Math. Soc. 256, 419-445.
This paper establishes some basic results concerning the rigid-
ity, flexibility, infinitesimal rigidity and infinitesimal flexibil-
ity of tensegrity frameworks. These results are then applied
to a number of questions, problems and conjectures regarding
tensegrity frameworks in the plane and in space.
105. H. J. Ryser, {1963). Combinatorial Mathematics, Carus Mathematical
Monographs 14, AMS.
106. D. Sachs, {1970). Graphs, matroids, and geometric lattices, J. Comb. The-
ory 9, 192-199.
107. I. Schoenberg and S. K. Zaremba, (1967). On Cauchy's lemma concerning
convex polygons, Canadian J. Math. 19, 1062-1071.
This paper contains several correct proofs of a lemma used by
Cauchy in proving his rigidity theorem.
108. B. Servatius, (1989). Birigidity in the plane, SIAM J. Discrete Math. 2 4,
582-589.
Defines edge and vertex birigidity, and investigates the rela-
tionship between the connectivity of the rigidity matroid and
and the birigidity of the underlying graph.
109. E. Steinitz and H. Rademacher, (1934). Vorlesungen ii.ber die Theorie der
Polyeder, Springer Verlag.
This book contains a correct proof (the first?) of Cauchy's
Rigidity Theorem.
110. J. J. Stoker Jr., (1968). Geometrical problems concerning polyhedra in the
large, Comm. Pure Applied Math. 21, 119-168.
164 REFERENCES
The purpose of this paper is to prove and at the same time

generalize Cauchy's Rigidity Theorem and to extend his ba-
sic ideas to a number of other cases, including polyhedra with
boundaries, certain non-convex polyhedra, and special exam-
ples of polyhedra of arbitrary genus.
111. K. Sugihara, (1980). On redundant bracing in plane skeletal structures, Bul-
letin of the Electrotechnical Laboratory 44 5/6, 78-88.
Develops the idea of cycles in the 2-dimensional generic rigidity
matroid and provides several nice examples.
112. T. Tarnai and E. Makai, (1989}. A movable pair of tetrahedra, Proc. Royal
Soc. London Ser. A 423 1865, 419-442.
Analyses the example of two regular tetrahedra in dual posi-
tion constrained to move such that the dual edges intersect.
113. T. S. Tay, (1976). Circuits in three dimensional structural geometry, M.
Math. Thesis, Univ. Waterloo.
114. T. S. Tay, (1981}. Rigidity problems in bar and joint frameworks, Ph.D.
Math. Thesis, Univ. Waterloo.
By a rigid body in n-space, the author means an n-dimensional
solid. An abstract linkage is a multigraph and a realization of
a linkage in n-space consists of embedding the multigraph in
n-space and replacing the vertices by rigid bodies. The main
result is: an abstract linkage has a rigid realization in n-space
if and only if it contains n(n + 1)/2 edge disjoint spanning
trees.
115. T. S. Tay, (1984). Rigidity of multi-graphs I: Linking rigid bodies in n-
space, J. Comb. Theory B. 36, 95-112.
116. T. S. Tay, (1984). Rigidity of multigraphs II, Graph Theory Singapore,
Springer Lect. Notes in Math 1073, 129-134.
This paper gives a short proof of the characterization theorem
of an n-linkage. These structures are now known as bar and
body frameworks in n-space.
117. T. S. Tay, (1989). Linking (n - 2}-Dimensional Panels in n-Space II:
(n- 2, 2)-Frameworks and Body and Hinge Structures, Graphs and Com-
binatorics 5 3, 245-273.
118. T. S. Tay, (1991). Linking (n- 2)-Dimensional Panels in n-Space I: (k-
1, k)-Graphs and {k-1, k)-Frames, Graphs and combinatorics 7 3, 289-304.
119. T. S. Tay, (1991). Henneberg's method for bar and body frameworks, Struc-
The main result is an inductive construction of generic isostatic
bar and body frameworks, namely frameworks of rigid bodies
linked by bars.
120. T. S. Tay, (1993}. On Generically dependent bar and joint frameworks in
space, To appear in Structural Topology.
REFERENCES 165
121. T. S. Tay, {1993). On the infinitesimal rigidity of ann-ring, Acta Mechan-

ica, to appear.
An n-ring in 3-space consists of an n-gon in space supported
at each vertex by two bars attached to the ground. This paper
gives a characterization of infinitesimally rigid n-rings.
122. T. S. Tay, {1993) A new proof of Laman's theorem, preprint.
Gives a direct proof of Crapo's characterization of generic iso-
static bar frameworks in the plane as the union of 3 trees such
that every vertex is covered by exactly 2 of them.
123. T. S. Tay, N. L. White and W. Whiteley, {1993). Skeletal rigidity of sim-
plicial complexes I, preprint.
124. T. S. Tay, N. L. White and W. Whiteley, {1993). Skeletal rigidity of sim-
plicial complexes II, preprint,
The central theme of the these two papers is a generalized con-
cept of rigidity, namely skeletal rigidity of simplicial complexes.
Five different but equivalent rigidity matrices are defined. The
relationship between the g-theorem of convex polytopes and
skeletal rigidity is also established.
125. T. S. Tay and W. Whiteley, {1984). Recent advances in generic rigidity of
structures, Structural Topology 9, 31-38.
126. T. S. Tay and W. Whiteley, {1985). Generating isostatic frameworks, Struc-
This paper contains several conjectures and a useful bibliogra-
phy.
127. S. Timoshenko and D. H. Youngs, {1945). Theory Of Structures, McGraw-
Hill, New York.
128. N. J. Thurston, {1991). On the rigidity of graphs, Bachelor of Arts Thesis,
Reed College.
Shows that the maximal conjecture fails in dimension four.
129. W. P. Thurston and J. R. Weeks, {1984). The mathematics of three-
dimensional manifolds, Scientific American 251 1, 108-120.
A readable introduction to the geometry of manifolds in 3
and higher dimensions. Configuration spaces of frameworks are
used as examples.
130. W. T. Tutte, {1958). A homotopy theorem for matroids I and II, Trans.
Amer. Math. Soc. 88, 144-174.
131. W. T. Tutte, {1965). Lectures on Matroids, J. Res. Nat. Bur. Stand. 69B,
1-48.
132. D. J. A. Welsh, {1976). Matroid Theory, Academic Press, London.
133. N. White and W. Whiteley, {1983). The algebraic geometry of stress in
frameworks, SIAM J. Algebraic Discrete Methods 4, 481-511.
This paper studies the algebraic - geometric conditions called
the pure condition for infinitesimal rigidity or for stress on a
166 REFERENCES
bar and joint framework.

134. N. White and W. Whiteley, {1987). The algebraic geometry of motions of
bar and body frameworks, SIAM J. Algebraic Discrete Methods 8, 1-32.
A study of the algebraic-geometric conditions for infinitesimal
rigidity, and a method for determining the motions of a bar
and body framework with 1 degree of freedom.
135. W. Whiteley, {1979). Motions of bipartite frameworks, Structural Topology
3, 62-63.
Some examples of the result obtained in [12] are given.
136. W. Whiteley, {1982). Motions and stresses of projected polyhedra, Structural
Topology 7, 13-38.
Using infinitesimal motions of panel structures, a new proof
is given for Maxwell's theorem that the projection of an ori-
ented polyhedron from 3-space gives a plane diagram of lines
and points which forms a stressed bar and joint framework.
The methods extend to prove a simple converse and a general
converse for other polyhedra under appropriate conditions on
the stress.
137. W. Whiteley, {1982). Motions of trusses and bipartite frameworks, Struc-
A continuation of the work of Crapo, [30]. The same truss is
studied after some of the symmetry is destroyed.
138. W. Whiteley, {1983). Cones, infinity and one story buildings, Structural
Topology 8, 53-70.
Consider a one story building constructed with a series of ver-
tical columns (possibly of different lengths), a roof framework
connecting the tops of the columns, plus a minimum of three
additional wall braces going to the tops of some columns. We
examine the static rigidity (i.e. infinitesimal rigidity) of this
framework by viewing it as a cone from the point at infinity.
139. W. Whiteley, {1984). Infinitesimal motions of a bipartite framework, Pacific
J. Math. 110, 233-255.
A recent established criterion [12] for the stresses on any bar
and joint framework with a complete bipartite graph is con-
verted into explicit criteria for infinitesimal motions of the
framework.
140. W. Whiteley, {1984). Infinitesimally rigid polyhedra I: statics of frame-
works, Trans. Amer. Math. Soc. 285 1, 431-461.
A proof of Alexandrov's Theorem using statics; also an exten-
sion of this result to 4-space.
141. W. Whiteley, {1987). Rigidity and polarity I: Statics of sheet structures,
Geometriae Dedicata 22 3, 329-362.
REFERENCES 167
Relates the rigidity of polarized frameworks in projective space

to euclidean sheet structures.
142. W. Whiteley, {1988). The union of matroids and the rigidity of frameworks,
SIAM. J. Discrete Math. 1 237-255.
Shows an m-dimensional infinitesimal rigidity matroid con-
tains m-copies of the cycle matroid.
143. W. Whiteley, (1988). Infinitesimally rigid polyhedra II: Modified spherical
frameworks, Trans. Amer. Math. Soc. 306 1, 115-139.
Replacement techniques are developed and used to prove that
an inseparable triangulated sphere, with one edge removed,
flexes so that all dihedral angles change, resolving a conjecture
of Kuiper.
144. W. Whiteley, (1989). A matroid on hypergraphs, with applications in scene
analysis and geometry, Discrete Comput. Geom. 4 1, 75-95.
Resolves a conjecture by Sugihara that the matroid defined for
liftings of 'polyhedral configuations' of plane points and some
collections of points designated faces, to spatial configurations
with plane faces is characterized by a simple submodular func-
tion on the bipartite incidence graph. This result is extended
to study liftings between adjacent dimesnions, and a new proof
of Laman's Theorem results from the study of liftings from the
line to the plane. Connections are made to the dual parallel
redrawing matroid.
145. W. Whiteley, (1989). Rigidity and polarity II: Weaving lines and tensegrity
frameworks, Geometriae Dedicata 30 3, 255-279.
A weaving is a pattern of straight lines in the plane with speci-
fied over and under crossings marked at each intersection. The
question of which weavings arise from lines embedded in space
is related to tensegrity frameworks.
146. W. Whiteley, (1990). Vertex splittings in isostatic frameworks, Structural
Topology 16, 23-30.
Shows how to extend an independent or rigid set by splitting
a vertex.
147. W. Whiteley, (1991). Weavings, sections and projections of spherical poly-
hedra, Discrete Applied Math. 32 3, 275-294.
A proof that the projection of a polyhedron and a cross sec-
tion of the same polyhedron are 'reciprocal' - a self stress in
the projection corresponds to a weaving pattern in the cross
section. A form of the theory is invariant under plane polarity.
148. W. Whiteley, (1991). The combinatorics of bivariate splines, in "Applied
Geometry and Discrete Mathematics", DIMACS Ser. Discrete Math. Theor.
Comp. Sci. 4, 587-608.
A description of the generic matroid for piecewise quadratic,
168 REFERENCES
globally C 1 functions over a dissected polygon in the plane.

Includes a proof of a series of properties reflecting a strong
analogy to the generic rigidity in 3-space of the graph of the di-
section if the boundary polygon is a triangle. Analogous prop-
erties include: rank 3IVI- 6 on complete graphs of more than
one vertex; vertex splitting; the independence of triangulated
spheres; etc. Some discussion of related matroids analogous to
rigidity in higher dimensions, but defined for realizations of
the graph in the plane. (Note that, by Maxwell's Theorem,
piecewise linear, globally continuous functions are analogous
to problems of plane stresses).
149. W. Whiteley, (1992). Matroids and rigid structures, in "Matroid Applica-
tions", Neil White, ed. Encyclopedia of Mathematics and its Applications
40, Cambridge University Press, 1-53.
A summary of connections between matroids and rigidity, as
well as related concepts such as polyhedral pictures and bivari-
ate splines.
150. H. Whitney, (1932). Congruent graphs and the connectivity of gmphs, Amer.
J. Math. 54, 15Q-168.
151. H. Whitney, (1935). On the abstmct properties of linear independence,
Amer. J. Math. 57, 509-533.
152. W. Wunderlich, (1954). Ein merkwii.rdiges Zwolfstabgetriebe, Osterr. Ingen.
Archiv. 8, 224-228.
153. W. Wunderlich, (1965). Starre, kippende, wackelige und bewegliche Acht-
ftache, Elem. Math. 20, 25-32.
154. W. Wunderlich, (1971). Starre, kippende, wackelige und bewegliche Ge-
lenkvierecke im Raum, Elem. Math. 26, 73-83.
155. W. Wunderlich, (1976). On deformable nine bar linkages with six triple
joints, Proc. K. Nederl. Akad. Wetensch Proc. A. 79, 255-262.
156. W. Wunderlich, (1977). Gefiihrliche Annahmen der Trilatemtion und be-
wegliche Fachwerke I, II, Z. Angnew. Math. Mech. 57, 297-304, 363-367.
157. W. Wunderlich, (1978). Sur une deformation remarquable du system des
genemtrices d'un cone du second degre et un probleme de geodesie, Bul.
Inst. Polit. Iasi. 24, 81-85.
158. W. Wunderlich, (1979). Snapping and shaky antiprisms, Math. Magaz. 52,
235-236.
159. W. Wunderlich, (1980). Neue Wackelikosaeder, Anz. Osterr. Akad. Wiss.
117, 28-33.
160. W. Wunderlich, (1980). Wackelige Doppelpymmiden, Anz. Osterr. Akad.
Wiss. 117, 82-87.
161. W. Wunderlich, (1980). Zur projectiven Invarianz von Wackelstrukturen,
Z. Angnew. Math. Mech. 60, 703-708.
REFERENCES 169
162. W. Wunderlich, (1981). Wackelikosaeder, Geometriae Dedicata 11, 137-

146.
163. W. Wunderlich, (1982). Projective invariance of shaky structures, Acta Me-
chanica 42, 171-181.
164. W. Wunderlich, (1982). Ringartige Wackelpolyeder, Anz. Osterr. Akad.
Wiss. 119, 71-77.
165. W. Wunderlich, (1982). Wackeldodekaeder, Elem. Math. 37, 153-163.
166. W. Wunderlich, (1983). Uber Ausnahmefachewerke, deren Knoten auf
einem Kegelschnitt liegen, Acta Mechanica 47, 153-163.
167. W. Wunderlich and C. Schwabe, (1986). Eine Familie von geschlossenen
gleichftii.chigen Polyedern, die fast beweglich sind, Elem. Math. 41, 88-98.
168. W. Wunderlich, (1990). Shaky polyhedra of higher connection, Publ. Math.
Debrecen 37, 355-361.
Index
0, !-extension, 94, framework, 1, 4,

2-tree decomposition, 120, free, 133,
3-tree decomposition, 122, general embedding, 20, 28,
affine transformation, 28, general position, 20,
basis, basis system, 59, generic, 4, 22,
bipartite graphs, 53, circle generic, 45,
birigididity, 111, cycle, 24,
circle generic, 45, dependence, 24,
C-L family, 149, framework, 45,
clique, 92, 130, independence, 24,
closed sets, 18, rigidity, 37,
closure operator, 18, 55, rigidity matroid, 42,
for connectivity, 18, generically rigid, 4,
in a matroid, 19, graphic matroid, 84,
rigidity, 22, Henneberg sequence, 113,
cobasis, 68, independence structure, 57,
cocycle, 69, independent, 40, 57,
cographic, 73, 84, generic, 24,
configuration space, 2, in a framework, 24,
connected matroid, 74, 110, with respect to closure, 40,
connectivity, 78, over a set, 133,
contraction, 72, infinitesimal
corank function, 69, flex, 5, 31,
cycle system, 64, isometry, 31-32,
cycle, 30, 62, motion, 5, 30-31,
generic, 24, rigidity, 5, 31,
of a framework, 24, rigidity matroid, 42,
degree of freedom, 36, 101, translation, 32,
dependence, rotation, 33,
generic, 24, isometry, 31,
in a framewo1 k, 24, infinitesimal, 31-32,
in a matroid, 64, isostatic, 47, 49, 87,
dependency number, 36, K{·), 17,
dependency relation, 25, k-extension, 134,
direct sum, 74, k-separating set, 78,
dual68, 69, Laman's condition, 4, 11, 47,
E{·), 18, Laman's theorem, 96,
edge-birigididity, 111, linkage, 13,
exchange axiom, 59, loop, 73,
extension, 49, m-cycle, 129,
flex, 2, m-independent, 129,
171
172 INDEX
m-isostatic, 129, spanning set, 62,

m-rigid, 129, strain, 140,
matroid, 39, 55, stress, 12, 26,
basis axioms, 59, resolvable, 27,
binary matroid, 64, stress space, 13, 30,
circle rigidity matroid, 45, stress operator, 28,
closure axioms, 55, strongly rigid, 1,
cobasis, 68, submodular, 61, 123,
cocycle, 69, support, 17,
co-graphic, 84, term rank, 82,
connectivity, 78, 110, transversal matroid, 83,
connectivity matroid, 56, Thtte prototype, 67, 70,
corank function, 69, Thtte subspace, 80,
cycle axioms, 64, V(·}, 17,
cycle matroid, 64, vertex cocycle, 104,
dependent sets, 64, wheel, 99,
dual, 68, Whitney function, 79,
Euclidean matroid, 59,
Fano matroid, 56,
graphic, 84,
independence axioms, 57,
partition matroid, 59,
rank axioms, 61,
representable, 79,
symmetric, 44,
transversal 58, 83,
uniform matroid, 56,
vectorial matroid, 56,
module, 118,
parallel element, 73,
Peaucellier's Cell, 14,
powerset, 'P, 55,
proper 2-tree decomposition, 121,
pseudocycle, 118,
~component, 100,
rank function, 45, 61,
restriction, 72,
rigid edge set, 40,
rigidity function, 20,
rigidity matrix, 21,
rigidity, 1, 4, 87,
closure operator, 22,
generic, 37, 42,
in a matroid, 42,
types of, 4-6,
singular framework, 3,

Combinatorial Rigidity PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Combinatorial Rigidity PDF

Uploaded by

Copyright:

Available Formats

Recent Titles in This Series

1991 Mathematics Subject Classification. Primary 05B35; Secondary 05C10.

Library of Congress Cataloging-in-Publication Data

©Copyright 1993 by the American Mathematical Society. All rights reserved.

Chapter 2. Infinitesimal Rigidity 17

Chapter 3. Matroid Theory 55

Chapter 4. Linear and Planar Rigidity 93

4.6. Characterizations of A2 and (A2).L 104

Chapter 5. Rigidity in Higher Dimensions 129

A framework in m-space is a triple {V,E,p), where (V,E) is a finite graph and

1.1. An Intuitive Introduction to Rigidity. We will use the term graph

FIGURE 1.1. A flip.

In order to avoid uninteresting special cases, we will assume that P contains

FIGURE 1.3. Rigid and non-rigid frameworks in 2-space.

One may interpret condition 1 as requiring that E contain enough edges to

be rigid and condition 2 as requiring that none of these edges be "wasted" by

Hence it is clear that V, the set of infinitesimal motions of the framework

FIGURE 1.4. The framework spaces.

(a) (b) (c) (d)

FIGURE 1.5. Some singular frameworks.

(at- a;)* (Pi-P;)+ (ui- u;) 2 = 0,

Exercise 1.1. Consider the !-dimensional framework consisting of two ver-

e. Use your answers to c. and d. to explain why rigidity and infinitesimal

THEOREM (A. L. CAUCHY, 1813 [17]). If there is an isometry between the

COROLLARY. The 2-skeleton of a strictly convex polyhedron in 3-space is

At first glance the two parts of this theorem seem to be in contradiction to

THEOREM (A. D. ALEXANDROV, 1950 (1]). If vertices are inserted in the

This result has the following corollary:

'Thming to generic rigidity, we have already noted that Laman characterized

FIGURE 1. 7. The double banana.

(L(u))(i,j) = (u(i)- u(j)) * (p(i)- p(j)),

Es(i,j)(p(i)- p(j)), taken over all vertices j adjacent to i in (V,E), is the

Peaucellier's Cell Peaucellier's Apparatus

2.1. Basic Definitions. Instead of starting our formal development of rigid-

V(E) = {i 13 j E V with (i,j) E E or (j,i) E E}.

K(U) = {(i,j) I i < j and {i,j} ~ U};

for any fixed edge set E ~ K and any vertex set U ~ V,

E(U} =En K(U).

Exercise 2.1. Consider the operators K(·}, E(-) and V(·).

(E)= {(i,j)li < j and i is joined to j by a path in (V, E)},

C2': C is "closed" under intersection.

Exercise 2.6. Recall Exercise 2.2.

(2.1.1} dim((T U {s})) = dim((T U {t})) = dim(span(T)) + 1.

Next assume that s E (T U {t}). It follows that T U {s} ~ (T U {t}); that

I(p) =(Pl. P2, .. · , Pn)·

Pn-P21 P12-P22 P21-Pll P22-P12 0 0 0 0

0 0 P21-P31 P22 -p32 P31-P21 P32-P22 0 0

0 0 0 0 P31-P41 P32-P42 P41-P31 P42-P32

(2.1.2) rih = L Sere, for some edge set F ~ E.

2.2. Independence and the Stress Space. As in the previous section,

no dependency relations. An edge set E will be independent if and only if the

(2.2.2) L Sij(Pi- Pj) = 0.

If j is not in V(E), then Sij = 0. Hence, we may visualize our dependency

FIGURE 2.9. A resolvable stress on a tetrahedron.

assigned scalars which yield a dependency relation. However, with respect to

(2.2.3) (T(s))i = ~::::si;(Pi- P;).

FIGURE 2.11. Stressing a general tetrahedron.

(2.2.4) (1jE(s))i = ~:::>i;(Pi- P;),

PROOF. Let E ~ K and assume that the rows of R(p) corresponding to E

TjE(S)i = :~~::>iJ(Pi- P;) = 0.

Exercise 2.16. Let A be an affine transformation of am.

LEMMA 2.2.3. Generic embeddings are general embeddings.

Exercise 2.17. Verify the vector computations in Lemma 2.2.3.