You are on page 1of 9

Compufers & Sfrucrures Vol. 40. No. 4, pp.893-901, 1991 0045.7949/91 $3.00 + 0.

00
Printed in Great Britain. Pergamon Press plc

GRAPHS AND STRUCTURES


A. KAVEHt

Iran University of Science and Technology, Tehran 16844, Iran

(Received 5 June 1990)

Abstract-The theory of graphs has found many applications in structural engineering. The study of the
rigidity of structures, flexibility analysis, dynamic analysis, optimization, finite elements, graphics, sparse
matrix technology and ordering are among those which have benefited more than other subjects. This
paper reviews some of these applications. Methods are developed for determining the topological
invariants of skeletal structures. The applications are extended to the generation of statical bases leading
to sparse and banded flexibility matrices, for an efficient flexibility analysis of structures.

INTRODUCTION basis corresponding to a set of independent com-


patibility equations should be constructed. The di-
Graph theory is a branch of mathematics which has mension of such a basis, y(S), is the same as
found many applications in engineering and science. the degree of statical indeterminacy (DSI) of the
In the field of structural mechanics, it is successfully structure.
applied to the conceptual analysis of structures by Adopting a graph-theoretical approach, a statical
Langefors [l], Kron [2], Henderson [3], Samuels- (kinematical) basis can be formed on substructures
son [4], Wiberg [S] and Trent [6]. For the analysis of corresponding to the elements of generalized cycle
skeletal structures when the structure is viewed as a (cut-set) basis.
system, the formulation is made by Lind [7], The stiffness matrix K of a skeletal structure is
Spillers [8], Dimmagio [9] and Fenves [lo]. In order pattern-equivalent to the cut-set basis adjacency
to overcome the difficulties involved in the force matrix N = LL of its graph model S, where L is the
method of structural analysis, graph theory is cut-set-edge incidence matrix of the corresponding
employed by Henderson and Maunder [l l] and basis. Similarly the flexibility matrix G is pattern
Kaveh [12]. Oden and Neighbors [ 131 applied graph equivalent to the generalized cycle basis adjacency
theory to the nonlinear analysis of structures, and matrix D = CC, where C is the generalized cycle-
Fenves and Gonzalez-Caro [ 141 and Livesley [ 151 edge incidence matrix of the selected basis.
used in the plastic analysis. Cassell[16] and Shanghui For an efficient analysis of a structure by stiffness
and Guohua [17] employed the finite element analy- method or flexibility approach, the sparsity of K or
sis. Rigidity of structures via a graph theoretical G should be maximized. This can be achieved to a
approach was studied by Laman [18], Loviz and great extent by maximizing the sparsity of N and D,
Yemini [19], Sughihara [20] and Kaveh [21]. Graph respectively. The simplicity of the stiffness method is
theory has been extensively used in ordering by due to the natural existence of a special cut-set basis,
many researchers such as Cuthill and McKee [22], known as a cocycle basis, consisting of the cuts
Cheng [23] and Kaveh [24,25]. around the nodes of S, except the ground node. This
Analysis of a structure involves the study of three basis corresponds to a highly sparse N matrix,
different properties: topological (graph theoretical), although the sparsity is not maximal for all the
geometrical, and material. Separate study of these structures. However, no such a simple generalized
properties results in a considerable simplification in cycle basis can be generated for the flexibility analysis
the analysis and leads to a clear understanding of the of the structures.
structural behaviour. This paper is devoted only to For rigid-jointed frames a statical basis can be
the study of the graph-theoretical properties of generated on the elements of a cycle basis. Efficient
skeletal structures. methods are developed for selecting cycle bases corre-
For the stiffness analysis of a structure, a sponding to sparse flexibility matrices [26-281. For
kinematical basis leading to a set of independent general skeletal structures, suboptimal generalized
equilibrium equations should be formed. The di- cycle basis leading to sparse flexibility matrices are
mension of such a basis, q(S), is the same as the selected[29]. Special methods for the formation of
degree of kinematical indeterminacy (DKI) of such bases are given in Ref. [30].
the structure. For the flexibility analysis, a statical In this paper, some basic concepts of graph theory
necessary for the analysis of skeletal structures are
tGuest professor, Institut fiir Allgemeine Mechanik, presented. Methods are developed for determin-
TU-Wien, A-1040, Vienna, Austria. ing topological invariants applied to the flexibility

893
894 A. KAVEH

analysis of rigid-jointed skeletal structures. Some


concepts of graph theory are generalized to include
properties of different types of skeletal structures. A
method is presented for generating suboptimal gener-
alized cycle bases of an arbitrary type of skeletal s=s,us, s, s, s,n s,
structure. A simple approach is included for ordering Fig. 2. Union and intersection of two subgraphs.
the elements of the selected bases leading to banded
flexibility matrices. ti,, = 0 otherwise. Unlike graph theory we consider
n,,=l fori=l,...,N
DEFINITIONS FROM GRAPH THEORY
4 n2 n3 n4
For readers who are not familiar with the theory of n, 1 1 1 1
graphs, the necessary definitions for the study of this
article are described below. Ji= n3
4 I 1 1 1 01 I
A finite graph S consists of a set N(S) of elements
n4 1 0 1 1
called nodes (vertices) and a set M(S) of elements
called members (edges) together with a relation of
Deleting the row and column corresponding to de-
incidence which associates with each member a pair
pendent cocycle n, , leads to cocycle basis adjacency
of distinct nodes called its end nodes. The degree
matrix N
(valency) of a node of S is the number of members
incident with that node. The incidence relation may 1 1 0
be represented as of the following forms: N=l 11.
(a) A member list which is constructed from a list
[0 1 11
of members with their end nodes. This list is given for
the graph S shown in Fig. 1. A graph S, is a subgraph of S if N(S) sN(S),
(b) A node-member incidence matrix z = [&I, M(S,) GM(S), and each member of S, has the same
which is defined by 4 = 1 if node n, is an end node of end nodes as in S. The union of subgraphs S, , S,, . ,
member m, and 6 = 0 otherwise. For S this matrix S, of S, denoted by Sp = Up=, S, = S, US2U. . . US,, is
becomes a subgraph of S such that N(SP) = Up=, N(S) and
M(Sp) = Ur=, M(S,). The intersection of two sub-
m, m2 m3 m4 4
graphs Si and S, denoted by Si OS,, is similarly defined
4 1 1 1 0 0
using the intersection of their node-sets and member-
I= 4 1 0 0 1 0 sets (Fig. 2).
*3 0 1 0 1 1 A path Pk in S is a finite sequence Pk =
n4 0 0 1 0 1 {n,,m,,..., mp, n,} whose terms are alternately
nodes ni and members m, of S, where n,_ , and ni are
Taking n, as a ground node (dependent cocycle) a the two end nodes of mi (1 < i < p). A simple path in
new incidence matrix is created by deleting the row S is a path in which no member or node of S appear
corresponding to this node. The rows of this matrix, more than once. If end nodes no and np of a simple
denoted by L, contains independent cocycles forming path corresponds to end nodes of a member m$,
a cocycle basis. A cocycle is a set of members which then P,Umj! forms a simple cycle Ci of length
isolates the node incident with these members. Matrix L (C,) = p + 1. The length of P,, L (Pi) = p, is taken
L for graph S of Fig. 1 with n, as a ground node as the number of its members. The path Pi is called
becomes a shortest path between two nodes no and np if any
ml m2 m3 m4 m5 other path P, between these nodes L(P,) < L(P,). In
1 0 0 1 0 this paper simple cycles will be referred to as cycles.
The member m$ is called the generator of C,. A cycle
0 1 0 1 1
generated on a member mf is called a minimal cycle if
n,lO 0 1 0 11 P, is a shortest path other than the generator (see
Fig. 3).
(c) A node adjacency matrix of S, which is defined

a
by m = [AJ, where & = 1 if member (n,, n,) is in S and

Member End nodes


PC,
ml
no
Ci

Fig. 1. A planar structure S. Fig. 3. A minimal cycle C, of length 3 on m:.


Graphs and structures 895

Fg= 0 otherwise. For graph S of Fig. 1, three cycles


can be generated

\
a
Fig. 4. A shortest route tree rooted from Q
and the cycle-edge incidence matrix is formed as

1
Two nodes ni and nj are said to be connected if m1 m2 m3 m4 m5
there exists a path between these nodes. A graph S is c, 1 1 0 1 0
said to be connected if all pairs of its nodes are 1
c=c* [ 0 1 1 0
connected. A component of a graph S is a maximal
c, 1 0 1 1 1
connected subgraph, i.e. it is not a subgraph of any
other connected subgraph of S.
Obviously C, is dependent on C, and C,, and a cycle
A tree T of S is a connected subgraph of S which
basis contains only two cycles, i.e. C, and C2. Hence
contains no cycle. If T contains all the nodes of S, it

1
is called a spanning tree of S. The complement of
1 1 0 1
members of T is S are the chords of this tree. A
shortest route tree rooted from a node n, of S is a tree c= [0 1 1 0 01
for which the distance of every node nj of T and n,
is minimum (see Fig. 4). A set of trees is called a forms a cycle-edge basis incidence matrix of S, and
forest. the corresponding cycle basis adjacency matrix D is
Two cycles C, and C, can be added to form their formed as
ring sum as C, 0 C, = C, UC, - C,flC, . Equivalently,
one can identify a cycle with its incidence vector, and D=CC=
cycle addition as vector addition over the integer
modulo 2. The set of all cycles is closed under
addition, forming a cycle space. A maximal set of where
independent cycles is known as a cycle basis. The h (9
cardinality of a cycle basis is the same as the first Betti i& L(C,) = trace D = 6.
number

The genus y(S) of a graph S is the minimum


6, (S) = M - N + b,(S), number of handles which must be added to a sphere
so that S can be embedded on the resulting surface.
Obviously y(S) = 0 if and only if S is planar. The
where M, N and b,(S) are the numbers of members, thickness e(S) of a graph is the minimum number of
nodes and components of S, respectively. planar subgraphs whose union is S. The crossing
The cycle-edge incidence matrix c = [Ei,] of S has number v(S) of a graph S is the minimum number of
a row for each cycle and a column for each member pairwise intersection of its members when S is drawn
with Eii= 1 if the ith cycle contains member mj and in the plane. Obviously v(S) = 0 if and only if S is

A Plane frame Datum node

A Plane truss Graph model of the truss

Fig. 5. Structures and their graph models.


896 A. KAVEH

planar. A graph is planar if it can be embedded in the Skfl Sk,,. The subgraphs S, (i = 1, . . . , q) may con-
plane. Genus, thickness and crossing number are sist of a member, a simple cycle or any other sub-
called the topological invariants of a graph. graph having prescribed values for cz(Si) and/or
a(A,). For different purpose, imposing restriction on
THE GRAPH MODEL OF A STRUCTURE
the choice of subgraphs Sk + , and their intersection
A k+, at each step, useful information can be
The mathematical model S of a skeletal structure obtained.
is considered to be finite graph without loops and
multiple edges (members). For rigid-jointed skeletal
structures (frames), S is constructed by representing A GENERAL UNION-INTERSECTION THEOREM
each element of the structure by a member and each
joint by a node in a one-to-one correspondence, In the process of expansion, at each stage, the
except for support joints which are in many-to-one properties of expanded subgraph Sk should be inves-
correspondence. Trusses are similarly modelled, how- tigated. The following theorem provides a powerful
ever, in this case structures are assumed to be simply mean for this purpose.
supported (Fig. 5).
Theorem 1
A UNIFYING FUNCTION
Let S = UT= I S, be the graph model of a structure
Consider a linear function g(S) in the following with the following functions being defined
form
a(S) = aM + bN + ca,(S)
cc(S) = aM + bN + cm,(S),
cc(S,)=aM,+bN,+ca,(S,), fori=l,..., q
where M, N and cc,(S) are the numbers of members,
nodes and components of S, respectively. The co- and
efficients a, b and c are integer parameters depending
on both the type of the corresponding structure and r(A,)=uR,+b~i+cccr,(Ai), for i=2 ,..., q,
the property which the function is expected to rep-
resent. For example IX(S) with appropriate values for where I@, and x, are the numbers of members and
a, b and c may describe the DSI [y(S)] or DKI [q(S)] nodes of Ai, respectively. Then
of a certain type of skeletal structure as given in
Table 1, or the first Betti number b,(S) of S as defined ]cL(s) - czO(s)1= i LccCsi)
- cGIO(si)l
,=I
and employed in graph theory.
It should be mentioned that the function a(S) can
- ,t2 [a(Ai)- Cool. (2)
be written in a more general form to include higher
dimensional elements, such as the inclusion of shear A simple proof of this theorem may be found in
pannels, etc. Kaveh [ 16,201.

AN EXPANSION PROCESS
Special case
In order to study a property of the graph model S Let S and S, (i = 1,. . . , q) be connected sub-
of a structure, S can be reformed by a sequence of q graphs, then (2) can be simplified as
elementary expansions as follows:

S, = S/lS/l. ./sq = s, cc(S) = i cc(S) - i E(A,),


(1) i=, 1-2

where Sk = Uf= , Si, Sk+ = SkUSk+ , and Ak+ , = where


&(Ai) = u&&+ bni + c. (3)
Table I.
Type of structure a(S) a b c
Plane rigid-jointed frame y(S) +3 -3 +3 DEGREE OF STATICAL INDETERMINACY AND
0 +3 -3 CROSSING NUMBER
v(S)
Space rigid-jointed frame y(S) +6 -6 +6
q(S) 0 +6 -6 The concept of statical indeterminacy is central to
Plane pin-jointed truss y(S) +1 -2 f3 a clear understanding of the mechanics of a skeletal
1(S) 0 +2 -3 structure, when analysed by means of the flexibility
Space ball-jointed truss y(S) +I -3 +6
method. In the following two theorems are presented
q(S) 0 +3 -6
for calculating the DSI of space structures using the
First Betti number of S b,(S) +1 -1 +1
crossing numbers of their planar drawings.
Graphs and structures 897

Definitions to the ground node is a minimum; transferring any


joint load to the ground with least number of mem-
A drawing S* of a graph S is a mapping of S
bers with non-zero loading in the corresponding
into a surface. The nodes of S go into distinct nodes
column in B,,.
of S*. A member and incidence nodes map into a
homeomorphic image of the closed interval [0, 1] with The compatibility equations yield
the relevant node. A good drawing is one in which no
X = -(B;F,B,)-B;F,B,P, (7)
two members incident with a common node have a
common point, and no two members have more than
where F,,, defines the flexibility of the individual
one point in common. A common point of two
disconnected members, and the matrix B, consists of
members is a crossing. An optimal drawing in a given
y(S) independent columns of SESs; the ith column
surface is one which exhibits the least possible cross-
being associated with a unit value of X,. The stress
ing. The number of crossing points of S after drawing
resultant r is then obtained by
on a plane or sphere, SP, is denoted by v(P). For
cases when the drawing is optimal, v(P) becomes the
r =[B,-B,(BiF,,,B,)-BfF,B,]P, (8)
crossing number

where G = Bf F,,,B, is the flexibility matrix of the


y(S) = 6b1(S) = 6 [Ri(SP) - v(P)], (4)
structure.
The complementary solution can be formed on a
where RI(P) is the number of internal regions of Sp;
maximal set of independent y-cycles forming a gener-
i.e. R,(P) = R(S*) - 1.
alized cycle basis of the topological model of the
Theorem 2 structure, which will also be denoted by S. On each
y-cycle 1, 3 or 6 SESs can be formed, depending on
Let S be a ball-jointed truss. Then whether S is a truss, planar frame or a space frame,
respectively. For an efficient flexibility analysis, G
Y(S) = VW - K(W, (5) should be sparse, narrowly banded and well con-
ditioned. In this paper only the sparseness of G is
where M,(P) is the number of members required studied. Bandwidth of G can be reduced by ordering
for full triangulation of SP. Simple proofs of these the selected Y-cycles.
theorems together with some useful results may be
found in Refs [31,32]. Obviously, it is advantageous Embedding for selecting cycle bases
to have an optimal drawing of S, since the number of In order to form B, matrix for frames, a cycle basis
countings becomes minimum in calculating its DSI. of S should be generated. Obviously for efficient
For the case where v(P) = 0, the corresponding solution, the elements of the selected basis should be
mesh basis of S forms a suitable cycle basis of S. as localized as possible, in order to provide maximal
However, surprisingly little is known about the cross- sparsity for the flexibility matrix G.
ing number of S in general. For our purpose further
research is required to develop a systematic approach Main idea
for an optimal drawing or at least a drawing with The cycle selection procedure in this section uses
subminimal crossings. the concept of embedding the geometric realization of
S into another polyhedron whose dissection has
FLEXIBILITY ANALYSIS OF STRUCTURES
dimension equal to two. This concept originates from
the planar graphs embedded in IR*, in which the cycle
The stress resultant distribution due to a given bounding the finite regions form an efficient basis
loading on a skeletal structure S, obtained by the (known as a mesh basis) for the first cycle group
flexibility method can be given as z, (S, R).
The object is to extend this approach to embedding
r=B,,P+B,X, (6) S on polyhedra and manifolds with certain proper-
ties. These properties are measured by using the
where $P is any particular solution satisfying equi- homology groups H,(K, R) of the underlying com-
librium with the imposed loads, and B, X is a comp- plexes, which measure, roughly speaking, the number
lementary solution formed from a maximum set of of independent p-dimensional holes of K. In other
independent self-equilibrating stress (SES) distri- words, it measures the extent to which K has non-
bution known as a statical basis. The number of bounding p-cycles.
independent SESs is the same as the degree of statical
indeterminacy (DSI) of the structure, denoted by A two-dimensional polyhedron embedding

Y(S). Let S be the mathematical model of a structure


The particular solution can be found from a pri- which is a simple graph (l-complex). The underlying
mary structure corresponding to a shortest route tree polyhedron or geometric realization of S is often
of S, in which the length of the path from each node denoted by S, however, in this paper it will be shown
898 A. KAVEH

by S. An embedding f :S --) P is a homeomorphism 3. K can be oriented so that y zi = 0, where zi


of S into the polyhedron P. An embedding fis called is a regional cycle.
a 2-cell embedding if the components of [P -f(S)]
If M is manifold, then b,(K) = 2, where the genus of
are all 2-cells. If the 2-cells are regular, then embed-
M and b, (K) = 1. Thus f is not admissible, but can
ding is called a regular 2-cell embedding. Let f(S) be
be modified by adding two appropriate fillings of
dissected into a l-complex isomorphic to the dissec-
order 2 and one performation of order 2, to satisfy
tion of S. The f(S) and the components [P -f(S)]
the admissibility condition for embedding.
form a dissection of P into a two-dimensional cell
As an example, Fig. 6 shows a hollow box S
complex K.
embedded on a sphere with one handle, i.e. a torus.
Modifications are made by two proper fillings [15, 191
Admissible embeddings
and one perforation (A) of order 2.
The cycles bounding the 2-cells of K are known as Thirty cycles of length 3 and two cycles of length
regional cycles. An admissible embedding f of S is 4 are selected as a minimal cycle basis of S. However,
one for which the regional cycles form a set of b,(S) for multi-member complex structure this method is
independent cycles from Z,(K, R). The necessary and by no means practicable.
sufficient condition for a 2-cell embedding f : S + P In a manifold embedding, the quality of selected
to be admissible are that the homology groups with cycle basis depends upon the genus of the manifold
real coefficients of P are trivial [ll]. In the case of on which S is embedded. Thus it is ideal to have a
2-cell embedding, this condition holds when first and minimal embedding. Again little is known about
second Betti number of K be zero. Using algebraic an efficient approach to carry out this type of
topology the admissibility condition can be stated as embedding.
follows: the necessary and sufficient condition for
f : S - P to be admissible is that the corresponding Embedding S on a union of disks
K be acyclic. A regular complex K is called acyclic if S can be considered as the union of some planar
all homology groups of K are trivial. Equivalently, connected non-separable subgraphs of S. The process
any cycle of Z,(K, R) bounds in K; i.e. of embedding as follows:
Z,(K, R) = B,(K, R). It is easy to show that a con-
tractible complex is acyclic, hence a contractible Step 1. Identify a planar subgraph S, and embed it
embedding is admissible. If K is collapsible, then it is on a disk dl whose dissection K, is isomor-
obviously contractible. Thus a collapsible embedding phic to S,.
is also admissible. Step 2. The second subgraph S, is identified such
that the corresponding K2 has a 2-cell with
Modified manifold embedding a free l-face and 1K, 101K2 1is a connected
subspace of the frontier of d,.
Edmonds permutation technique provides a
Step 3. The process of Step 2 is continued and at ith
method for a 2-cell embedding of S in an orientable
step K, is joined to K- = Uj; / K, with K,
2-manifold M. The choice of vertex permutations in
having a free l-face and 1KiI fll K- 1 is a
this method is taken arbitrary, which may lead to a
connected subspace of the frontier of (ii.
manifold with an unnecessarily large genus. In order
Obviously K\ K-.
to reduce the genus of M, Duke [33] has developed
a reduction technique for transforming a 2-cell The process is terminated when all l-cells of S are
embedding M into N, where y(N) = y(M) - 1. embedded in K = UJ=, K,, which is collapsible, since
Youngs [34] has developed an algorithm for a mini- K=KJ\Kq-\...\K\...K=K,.
mal embedding by considering all the possible sets of It is ideal to embed S in a complex K with
vertex permutations of S leading to genus of S. This minimum number of disks. This minimum number of
algorithm is very lengthy and cannot be used as a disks for all possible collapsible embedding is an
practical method. Kaveh [35] developed a practical invariant of S, known as the thickness of S. Hence
method in which the vertex permutations are deter-
I
mined during the process of embedding S. However,
this algorithm does not always lead to a minimal
embedding.
A different approach may also be used based on an
intutive regular embedding of S. For an embedding
f : S --) A4 which dissects polyhedron M into a 2-com-
plex K. M is a manifold-if _
1. Each l-cell of K is incident with exactly two
2-cells.
2. All the 2-cells and 1-cell of K having a particu-
lar O-cell as a face can be ordered in a sequence
so that the consecutive cells are incident. Fig. 6. S embedded on a modified manifold.
Graphs and structures 899

embedding should be performed so that q is the same Definition 1


as the thickness of S. Again only partial results are Si is called elementary if every subgraph Si c Si
available about the thickness of graphs. Any system- (different from a single node) satisfying ~(3;) < 0,
atic approach for embedding on a minimum number where Si is any component of Si. A connected
of disks should be of advantage for reducing the subgraph T of S is called a y-tree of S if it can be
overlaps of the cycles. formed by joining its elementary subgraphs S,,
s,,... , S, (except a single node) such that
An expansion process for cycle selection
A major difficulty in the formation of a collapsible (a) Y(S, ) = 0
complex has been its automation. Thus a reverse (b) y(S,) = y(Ai), for i = 2,3, . . . ,p. (9)
approach, namely an expansion process has been
adopted by Kaveh [36]. This method starts from a From (2) it follows that y(T) = y(U;, , Si) = 0. If T
specified node and generates a minimal cycle on an contains all the nodes of S, it will be a spanning y-tree
edge incident on this node. The cycle C, of minimal of S. The members of S which are not included in T,
length is formed, which is connected to C, and C, UC, are called the y-chords of T.
is collapsible. The process is continued until b,(S)
independent cycles are selected. The independence Dejinition 2
check imposes restrictions such that the collapsibility
A removable subgraph S, of Si is an elementary
of c,uc,lJ . . . UC,,,, is maintained. This approach is
subgraph the removal of which leaves the DSI un-
a greedy type algorithm which embeds S on a col-
changed, i.e. y(Si\S,) = y(S). A y-tree of S contain-
lapsible set of disks. Obviously the number of disks
ing two chosen nodes which has no removable
will be bounded by the first Betti number and the
subgraph is called a y-path of S between these two
thickness of the graph. The above method is order
nodes. A connected subgraph C, of S with y(C,) = a
dependent like many other combinatorial algorithms
which has no removable subgraph is termed a y-cycle
and starting node influences the total length of the
of S. The total number of members of C, is taken as
selected cycles. Thus, further improvements have
its length. Obviously the union of T and each y-chord
been needed to improve the results by allowing the
contains a y-cycle and it becomes a y-cycle after the
selection of disjoint cycles (disjoint disks) [36]. These
removal of its removable subgraphs.
algorithms are successfully implemented and efficient
generation of cycle bases has become feasible [27]. In
the above approaches, the total length of cycles of a Definition 3
basis, is reduced in order to decrease the overlaps of A maximal set of independent y-cycles of S is called
the cycles. However, such a basis, in general, does not a generalized cycle basis of S and it contains y(S)/a
lead to a cycle adjacency matrix D of maximal of y-cycles.
sparsity. A direct method for maximizing the sparsity
of D is recently developed in [26]. For this purpose
Definition 4
weights are assigned to edges of S indicating the
length and the number of minimal cycles which can In an expansion process if subgraphs S, be y-cycles
be generated on each edge. Restrictions are imposed C, and
on the selection process to generate cycles of least
overlaps, often leading to optimal cycle bases. Re- y(Ck+)=y(CkUCk+l)=y(Ck)+aP (10)
cently, Kaveh [35] studied special cases in relation
with minimal-optimal cycle bases. However, before then C, + , is called an admissible y-cycle.
developing methods for finding the genus and thick-
ness of a graph, no general approach seems to DeJnition 5
become feasible.
A y-cycle generated on a member mj, having the
shortest length, is called a minimal y-cycle on m,.
GENERALIZATION OF SOME CONCEPTS FROM
GRAPH THEORY
Dejinition 6
For rigid-jointed frames y(S) = 1(16,(S), I++being 3 Let x(C) and x(D) denote the number of non-zero
or 6 depending on whether S is a planar or a space members of y-cycle-edge incidence matrix and y-cycle
frame. The concepts of graph theory are directly adjacency matrix of a y-cycle basis of S, respectively.
applicable to this kind of skeletal structure. However, A generalized cycle basis corresponding to minimum
for planar or space trusses y(S) can not be expressed x(C) is called minimal and the one leading to
as B&(S), and most of the concepts should be minimum x(D) is called optimal. When near mini-
generalized. In the following certain definitions of this mum sparsity is achieved, then the corresponding
kind are presented, which are suitable to all kinds of bases are referred to as subminimal and suboptimal
skeletal structures. generalized cycle bases.
900 A. KAVXH

ALGORITHM FOR THE FORMATION OF A (b) The distance of the furthest node of C, from 0,
SUBOPTIMAL GENERALIZED CYCLE BASIS
denoted by df .
The following algorithm which is based on the (c) The mean value of d; and dfk, i.e. )(d; + 4)/2 1.
above concepts and definitions, lead to the formation (d) The sum of dZ + 1L(C,)/2 (, where Jo is the
of suboptimal generalized cycle bases. length of C,.
(e) The mean value of the distances of the nodes
Step 1. Select a minimal y-cycle on each member rni for C,, i.e.
of S, and assign its length as the weight pi
of mi.
Step 2. Generate the first y-cycle on the member of
smallest weight.
Step 3. Select minimal cycles on members in the Any of the above definitions can be used as the
ascending order of their weights pi. This distance number of an element of a generalized cycle
process should be carried out as far as the basis (GCB), or a finite element model (FEM).
formation of minimal admissible y-cycles is
Algorithm
possible.
Step 4. If the number of selected y-cycles is less Step 1. Order the nodes of S with a nodal number-
than y (S)/a, then form an admissible y- ing algorithm, e.g. with the algorithm given
cycle having the least possible length on the in Ref. [37].
member of the least unused pi. This y-cycle Step 2. Use the same starting node as in Step 1 to
is subminimal, otherwise it would have been generate an SRT and find the distance
selected at previous steps. number of the elements of the considered
Step 5. Repeat Steps 3 and 4, generating admissible GCB (FEM).
minimal and subminimal y-cycles. Until Step 3. Assign these distance numbers to the
y(S)/a such cycles forming a suboptimal nearest (furthest or any other appropriate
generalized basis are selected. intermediate) nodes of the elements of the
GCB (FEM). In this process a node may
This algorithm is general and covers all types of become the representative node of q el-
skeletal structures. However, when y(S) = $6,(S), ements of the GCB (FEM). Then q different
then a generalized cycle basis becomes a cycle basis (some may be equal) distance numbers will
of S, for selection of which an efficient method is be assigned to this node.
developed in [36]. For planar trusses, simplified Step 4. Number these nodes in the ascending order
methods are developed which are more suitable from of their distance numbers. For equi-distant
programming point of view. Once an optimal or nodes, the same sequence as the nodal num-
suboptimal generalized cycle basis of S is generated, bering of Step 1 should be used, to effect the
matrices B, corresponding to highly sparse flexibility connectivity properties of S.
matrix G can easily be formed. Step 5. Order the elements of the GCB (FEM) with
the same numbers received by their repre-
sentative nodes. This provides an efficient
ORDERING FOR BANDWIDTH REDUCTION ordering for the elements of the considered
GCB (FEM).
In order to reduce the bandwidth of the flexibility
matrices, the elements of the selected generalized
cycle bases should be ordered. In this section a simple REFERENCES
approach is presented for element ordering to reduce
B. Langefors, Analysis of elastic structures by matrix
the bandwidth of the flexibility matrices of skeletal
transformation with special regards to semi-monocoque
structures. This method can also be used for nodal structures. J. Aero. Sci. 451-458 (1952). Algebraic
ordering of finite element models for reducing the methods for the numerical analysis of build-up systems.
bandwidth of stiffness matrices or frontwidth in SAAB, TN38 (1956). Algebraic topology and elastic
frontal solution of equations. networks. SAAB, TN49 (1961).
2. G. Kron, Diakoptic-piecewise solution of large-scale
system. A serial of twenty chapters, starting 7 June
Definition 1957, Electrical Journal, London (1957). Elastic struc-
tures from the point of view of topological network
The distance of d, of a node n, of S from a selected theory. RAGG Memories 3, 329-337 (1962).
node 0 is the length of the shortest path connecting 3 J. C. de C. Henderson, Topological aspects of structural
ni to 0. The distance number of an element of a linear analysis. Aircr. Engng 32, l-5 (1960).
generalized cycle basis or a finite element C, from 0 4. A. G. Samuelsson, Linear analysis of frame structures
by use of algebraic topology. Ph.D. thesis, Chalmers
is defined by one of the following:
Tekniska Hiigskola, Gdteborg (1962).
5. N. E. Wiberg, System analysis in structural mechanics.
(a) The distance of the nearest node of C, from 0, Ph.D. thesis, Chalmers Tekniska Hdgskola, Gdteborg
denoted by d;. (1970).
Graphs and structures 901

6. H. M. Trent Isomorphisims between oriented linear 22. E. Cuthill and J. McKee. Reducing bandwidth of sparse
graphs and lumped physical systems. J. Acoust. Sot. symmetric matrices. Proc. 24th Nat. Conf. kX4
Am. 27, 500-515 (1955). 152-172 (1962).
7. N. C. Lind, Analysis of structures by system theory. 23. K. Y. Cheng,Note on minimizing the bandwidth of
Proc. ASCE ST2, l-21 (1962). sparse symmetric matrices. Computing 11, 27-30
8. W. R. Spillers, Application of topology in structural (1973).
analvsis. Proc. ASCE ST4. 301-313 (1963). 24. A. Kaveh, Multiple use of a shortest route tree for
9. F. L: Dimmagio, Statical indeterminacy and stability of ordering. Commun. Appl. Numer. Meth. 2, 213-215
structures. Proc. ASCE 89, 63-75 (1963). (1986).
10. S. J. Fenves, Structural analysis by networks, matrices 25. A. Kaveh, Ordering for bandwidth reduction. Comput.
and computers. Proc. ASCE STl, 199-221 (1966). Struct. 24, 413-420 (1986).
11. J. C. de C. Henderson and E. A. W. Maunder, A 26. A. Kaveh. An efficient flexibility analysis of structures.
problem in applied topology. J. Inst. Math. Applic. 5, Comput. Struct. 22, 973-977 (lb86). -
254-269 (1969). 27. A. Kaveh, An efficient program for generating cycle
12. A. Kaveh, Application of topology and matroid theory bases for the flexibility analysis of structures. Commun.
to the analysis of structures. Ph.D. thesis, London Appl. Numer. Meth. 2, 339-344 (1986).
University (1974). 28. A. Kaveh, Suboptimal cycle bases of graphs for the
13. J. T. Oden and A. Neighbors, Network-topological flexibility analysis of skeletal structures. Compur. Meth.
formulation of analysis of geometrically and materially ADDI. Mech. Ennnz 71. 259-271 (1988).
non-linear space frames. First Int. Conf. on Space 29. A.i(aveh, A combmatorial optimization problem; opti-
Structures, University of Surrey, Sept. 1966, (Edited by mal generalized cycle bases. Comput. Meth. Appl. Mech.
R. M. Davis) Blackwell Scientific (1967). Engng 20, 39-52 (1979).
14. S. J. Fenves and A. Gonzalez-Caro, Network topolog- 30. A. Kaveh, Statical bases for efficient flexibility analy-
ical formulation of analysis and design of rigid-plastic sis of planar trusses. J. Strucf. Mech. 14, 475-488
framed structures. Int. J. Numer. Meth. Engng 3, (1986).
425-441 (1971). 31. A. Kaveh, Topology and skeletal structures. ZAMM 68
15. R. K. Livesley, The automatic design of structural 347-353 (1988).
frames. Quart. J. Mech. Appl. Math. 9, 257-269 (1956). 32. A. Kaveh, Planar drawings of space structures. Iranian
16. A. C. Cassell, An alternative method for finite element J. Sci. Tech. 14, (1988).
analysis; a combinatorial approach to the flexibility 33. R. A. Duke, The genus, regional number and Betti
method. Proc. R. Sot. Lond. A352, 73-89 (1976). number of a graph, Canad. J. Math. 18, 817-822
17 Y. Shanghui and X. Guohua, The method of topology (1966).
graph theory for the finite element analysis. Proc. 34. J. W. T. Youngs, Minimal embedding and genus of a
Comput. Mech. 659-663 (1984). graph. J. Math. Mech. 12, 303-315 (1963).
18. G. Laman, On graphs and rigidity of plane skeletal 35. A. Kaveh, On minimal and optimal cycle bases of
structures. J. Eng. Math. 331-340 (1970). graphs for sparse flexibility matrices. ZAMM 69
19. L. Lovriz and Y. Yemini, On generic rigidity in the T212-T214 (1989).
plane. SIAM J. Alg. Disc. Meth. 3 91-98 (1982). 36. A. Kaveh, Improved cycle bases for the flexibility
20. K. Sughihara, On some problems in the design of analysis of structures. Comput. Meth. Appl. Mech.
skeletal structures. SIAM J. Alg. Disc. Meth. 4,355-362 Engng 9, 267-272 (1976).
(1983). 37. A. Kaveh, A connectivity coordinate system for node
21. A. Kaveh, A cominatorial study of the rigidity of planar and element ordering. Proc. Civil-Comp 89, London, 1,
structures. Acta Mech. 62, 189-196 (1986). 149-155 (1989).

CAS 4ol4-4

You might also like