Computational Structures Technology Series Structural Mechanics: Series Editor: Professor A Kaveh, Iran University of Science and Graph and Matrix Methods Technology, Tehran, Iran Third Edition 1. Structural Mechanics: Graph and Matrix Methods Third Edition Ali Kaveh 0 86380 304 0 Ali Kaveh TITLES OF RELATED INTEREST: Optimal Structural Analysis, Ali Kaveh RESEARCH STUDIES PRESS LTD. 0 86380 215 X | R Baldock, Hertfordshire, England A Partner Imprint with the Nonlinear Analysis of Offshore Structures. Institute of Physics Publishing Bjorn Skallerud and Jorgen Amdahl ere loPRESEARCH STUDIES PRESS LTD. 16 Coach House Casters, 10 Hitein Street, Baldock, Hertfordshire, SG7 AE, England ‘www reearchstudiee pres cok and Institute of Physies russ, Suit 929, The Publi Ledge Buildig, 130 South independence Mall Wes, Phiniphia, PA 19105, USA. Copyright © 2008, by Research Studies Press Ltd. Research Studies Pros Ltd. isa pater nprin with the Institute of Phyales rusts Allsights reserved [Nopatt ofthis bok may be reprodsced by any means, nor transmit, nor tensed into a machine language without the writen perma ofthe publisher, Marketing: Institute of Physies eunusine, Dina House, Temple Bask, Bristol, BSI GBE, England swims bookmarken Distribution: NORTH AMERICA ADC, 50 Winter Sport Lane, PO Box 20, Williston, VT 05495.0020, USA Tel: 1-00 632 0880 or outside USA 1-802 862 0095, Fax: BO2 N68 1626, E-mail: rder@uievt.com UK AND THE REST OF WORLD ‘Marston Book Services Ltd, P.0. ox 26, Abingdon, Oxfouishir, OX14 AYN, England “Tels +4 0)1235 465500 Fax: +44 (01235 465555 Fem: deerardeamarvon.co uk Library of Congress Cataloging-in-Publication Data Kaveh, A. (Ali), 1948 ‘Structural mechanics: graph and matrix methods / Ali Kaveh, 3d ed pcm. ~ (Computational structures technology series 1) Includes bibliographical references and inde. ISBN 0-86380-804-0 (ak. paper) 1. Structural analysis (Engineering)Matrix mothods, 1. Til. I, Series, ‘TAGA2.K38 2004 624.17 de22 2004009855 British Library Cataloguing in Publication Data ‘A catalogue record for this book is available from the British Library. ISBN 0 86380 304 0 Printed in Great Britain by CPD Wales, Ebbw Vale Cover artwork by A3 grafix td. Editorial Foreword Computational Structures Technology (CST) blends the insightful modelling of structural response with the development of computational methods. CST is an outgrowth of mathematical and mechanical methods of structural analysis, developed over the past five decades; representing one of the most significant developments in the history of the structures fie. ‘The development of CST cuts across a number of disciplines including structural mechanics, discretization techniques, numerical methods, computer seience and combinatorial optimisation, The later comprises theory of graphs, algebraic graph theory, linear algebra and matroids, The present book employs many of the above disciplines with special emphasis on combinatorial optimiation. The latter enables analysts to handle large-scale structural systems with low computational cost and time, fulfilling the requirement for faster analysis and design of future complex systems, Last but not least, I would like to express my gratitude to all my colleagues at Research Studies Press for their support of the present series, and I am looking forward to contributions from eminent authors around the globe, A. Kaveh, Series EditorPreface In order to understand the behaviour of a physical system in general, and a structure in particular, different properties associated with it should be identified and treated separately. One of the most important properties of a structure is its topology or connectivity. This property can well be understood, by studying the underlying mathematical model of a structure, which is considered to be a graph, Although some graph-theoretical concepts were implicitly applied to structural ‘mechanics in the work of Henneberg and Muller Breslau as early as 1900, the importance of graph theory has become apparent only inthe recent decades due to its particular suitability for computational mechanics. The increasing complexity of practical structures requires new tools for handling their problems, and the theory ‘of graphs is one such powerful means. ‘The author has been involved in various developments and applications of graph theory in the last 33 years. The present book contains part of this research, suitable for various aspects of matrix structural analysis. In order to limit the volume of this, book, details of some of the well-established approaches described in textbooks on ‘matrix analysis of structures have been excluded. Instead, efficient methods are presented for optimal analysis of structures, leading to the formation of structural matrices which are sparse, well-structured and well-conditioned. ‘The concepts presented in this book are not only applicable fo skeletal structures, but cam equally be used for the analysis of other systems such as hydraulic and electrical networks. These concepts can easily be extended to finite clement methods |In Chapter 1 an introduction is provided to the definitions and basic concepts of graphs. In Chapter 2, methods for determining the degree of statical and kinematical indeterminacies of structures are presented. This chapter covers some important concepts such as expansion process and a generalised intersection theorem for evaluating various invariants during the expansion process, which are also employed in subsequent chapters. In Chapter 3, a brief mathematical study of the rigidity of structures is included. Chapter 4 provides the network formulation of structures by assigning different variables with nodes and members of the graphs, In Chapter 5, the matrix displacement method is briefly described. A detailed treatment of the foree method is presented in Chapter 6, in which many graph theoretical methods are developed for the formation of sparse flexibility matrices for frame structures. In Chapter 7, various methods are presented for nodal and. element ordering for improving the structure of the stiffness and flexibility matrices. This makes the efficient and economical solution of large system linear ‘equations feasible. Conditioning of the structural matrices, when a structure contains members of widely different stilThesses, becomes very important, This is discussed in Chapter 8 and methods are presented for selecting statical and kinematical bases leading to well-conditioned flexibility and stiffness matrices. Certain problems in structural analysis require more general concepts than those ofgraph theory, Such problems are treated in Chapler 9 and the theory of matroids employed for their analysis. The Greedy algorithm of combinatorial optimisation and its applications are also included in this chapter. Configuration processing and data generation ate discussed in Chapter 10 using simple transformations, providing an efficient tool for pre-processing of space structures and finite element meshes. Finally, the use of various forms of symmetry is presented in Chapter 11 for simplifying the eigensolution of problems associated with graph matrices and vibrating systems. | would like to take this opportunity to acknowledge a deep sense of gratitude to a ‘number colleagues and friends who in different ways have helped in the preparation of this book. Mr. J.C. de C. Henderson formerly of Imperial College of Science and Technology first introduced me to the subject with most stimulating discussions on various aspects of topology and combinatorial. mathematics, Professor F. Ziegler encouraged me to write this book and enabled me to complete it in the very happy and stimulating atmosphere of his institute at the Technical University of Vienna. My gratitude is extended to Professor P.C. Kendall, formerly ‘of Sheifield university and Mrs V.A, Wallace for constructive comments on the First and second editions of this book. My special thanks are due to Mrs C. Holmes, publishing director, and William Askew, Chairman, both of Research Studies Press Limited, for editing the third edition and her unfailing kindness in the course of the preparation ofthis book. | would like to thank my former Ph.D. and M.Se, students, Dr. A, Mokhtarzadeh, Dr. HA. Rahimi Bondarabady, Dr. H. Rahami, G.R. Roosta, 1. Gaderi, M.A. Sayatinejad, P. Ashtari and B. Salimbahrami for their help in various aspects of \wrting this book, My warmest gratitude is due to my wife Mrs Leopoldine Kaveh for proof reading, the first edition and for her continued support in the course of preparing all three editions. ‘The publication of the third edition has provided another opportunity for an extensive revision of this book. However, the original structure of the book is preserved, since comments on the frst and second editions have conformed the suitability ofthe materials previously presented, Every effort has been made to render the book error free. However, the author ‘would appreciate any remaining errors being brought to his attention through the following e-mail address alikaveh ius. A. Kaveh Professor of Structural Engineering Iran University of Science and Technology ‘Guest Professor at TU-Wien ‘Vienna, May 2004 CONTENTS: Contents Editorial Foreword Preface List of Symbols, 1, BASIC CONCEPTS AND DEFINITIONS OF GRAPH THEORY 1 Introduction 1.2. Basie Definitions 1.2.1 Definition of a Graph 12.2 Adjacency and Incidence 123 Isomorphie Graphs 124 Graph Operations 12.5 Walks, Trails and Paths 12.6 — Connectedness 12.7 Cycles and Cutsets 1.2.8 Trees, Spanning Trees and Shortest Route Trees 1.3 Different Types of Graphs 14 Vector Spaces Associated with a Graph 1.4.1 Group and Fields 142 Vector Spaces 143 Veotor Space of a Graph 1.44 Cycle Subspace and Cutset Subspace of a Graph 145 Fundamental Cycle Bases 1.46 Fundamental Cutset Bases 1.4.7 Dimensions of Cycle and Cutset Subspaces 148 Orthogonality Property 1.5 Matrices Associated with a Graph 15.1 Matrix Representation ofa Graph 15.2 Cycle Bases Matrices 15.3 Cutset Bases Matrices 1.6 Directed Graphs and Their Matrices 1.7 Graphs Associated with Matrices 1.8 Planar Graphs ~ Euler's Polyhedra Formula 18.1 Planar Graphs 18.2 ‘Theorems for Planatity 19 Maximal Matching in Bipartite Graphs 19.1 Definitions 1.9.2 Theorems on Matching 1.9.3 Maximum Matching Exercises Page numberCONTENTS ‘TOPOLOGICAL PROPERTIES OF SKELETAL STRUCTURES, 2.1 Introduetion 22 Mathematical Model of a Skeletal Structure 23. Union-interseetion Method 23.1 A Unifying Funetion 232 An Expansion Process 233 An Intersection Theorem 234 A Method for Determining the DKI and DSI of Structures 23.5 Modifications on a Structure 24 Identification Method 25 The DST of Structures: Special Methods 2.6 Space Structures and Their Planar Drawings 2.6.1 Admissible Drawing of a Space Struct 2.62 The Degree of Statieal Indeterminacy of Space Frames 2.6.3 The Degree of Statical Indeterminacy of Space Trusses 2.64 Comparison of Classical and Topological Formulae 2.7 Suboptimal Drawing of a Space Structure 27.1 Introduction 272 Automatic Drawing ofa Space Structure Exercises ITY OF SKELETAL STRUCTURES 3.1 Introdvetion 32 Definitions 33 Complete Matching forthe Recognition of Generie Independence 34 A Decomposition Approach forthe Recognition of Generic Independence 35. Rigidity of Planar Trasses: Special Methods 35.1 Simple Trusses 3.52 Trusses inthe Form of 2-rees 353. Atte and its Rigidity 3.54 Grid-form Trusses with Bracings 3.6 Henneberg Sequence for Examining the Rigidity of Trusses 3.7 Connectivity and Rigidity Exercises NETWORK FORMULATION OF STRUCTURAL ANALYS! 4d 42 43 Introduction ‘Theory of Networks Basic Concepts of Network Theory 43.1 Topological Properties of Networks 6 44 CONTENTS: 43.2 Algebraic Properties of Networks 43.3 Formulation of Network Analysis Formulation of Structural Analysis Exercises MATRIX DISPLACEMENT METHOD 3.1 Introduction 52 Formulation’ 5.3. Blement Stiffness Matrices 53.1 Stiffness Matrix ofa General Element 53.2 Stiffness Matrix of a Bar Element 53.3 Stiffness Matrix of a Beam Element SA Overall Stiffness Matrix of a Structure 55 General Loading 546 Computational Aspects of the Matrix Displacement Method Exercises : MATRIX FORCE METHOD 6.1 Introduction 62 Formulation 63. Generalized Cycle Bases of a Graph 64 — Minimal and Optimal Generalized Cycle Bases 6.5 Pattern Equivalence of Flexibility and Cycle Adjaceney Matrices 6. Minimal GCB of a Graph 6.7 Selection of a Subminimal GCB: Practical Methods 67.1 Method 1 672 Method 2 6.73 Method 3 68 Force Method for the Analysis of Rigid-jointed Skeletal Structures 68.1 Cycle Bases Selection: Topological Methods 682 Cycle Bases Selection: Graph-theoretical Methods 68.2.1 Suboptimal Cycle Bases; A direct Approach 68.2.2 Suboptimal Cyele Bases; An Indirect Approach 68.23 Examples 683 Formation of By and B, Matrices 69 Force Method for the Analysis of Pin-jointed Planar Trusses 69.1 Associate Graphs for Selection of a Suboptimal GCB. 69.2 A Bipartite Graph for Selection of a Suboptimal GCB 6.10 Force Method of Analysis for General Structures 6.10.1 Algebraic Methods 6.11 Optimal Plastic Analysis an Design of Frames 2 Ls 19 3 1s 125 12s 136 1B6 bs 142 145 151 153 155 159 159 160 16s 168 169 169 170 170 m1 m1 m 173 178 186 188 189 192 197 17 200 203 203 21CONTENTS 6.11.1 Formulation 6.11.2 Optimal Safe Plastic Analysis of Frames 6.11.3 Optimal Safe Plastic Design 6.114 Formation of My and M, Matrices 6.11.5 Standard Mathematical Programming Formulation 6.11.6 Numerical Results Bxereises ORDERING FOR BANDWIDTH, PROFILE AND FRONTWIDTH OPTIMISATION 7.1 Introductio 72, Bandwidth Optimisation 73. Preliminaries 74 Pattern Equivalence of Stiffness and Cutset Adjacency Matrices 7.5 A Shortest Route Tree and its Properties 76 Nodal Ordering for Bandwidth Reduetion 76.1 A Good Starting Node 762 Primary Nodal Decomposition 763. Transversal P of an SRT 7.64 Nodal Ordering 765 Examples 7.7 A Connectivity Coordinate System for Nodal Ordering 74.1 A Connectivity Coordinate System for Planar Graphs 77.2 A Connectivity Coordinate System for Space Graphs 7.8 Nodal Numbering for Profile Reduction 19. Graph-theoreticalInexpretation of Gaussian Elimination 7.10 Element Ordering for Bandwidth Optissation of Flexibility Mates 710.1 An Associate Graph 71102 Distance Number of an Element 7.103. Element Ordering Algorithms 7104 Example 7.11 Ondeting for Bandwidth Optimisation of Finite Blement Meshes 712 Ordering Using Algebraic Graph Theory 7121» Definitions 7.122. Pigenvalues and Eigenvectors of Matrix A 7.12.3 Eigenvalues and Eigenvectors of Matrix L 7.124 A Hybrid Method for Ordering 712.5 Numerical Results 7126 _ Discussion 7.13. Bandwidth Reduction for Rectangular Matsices 212 213 214 216 217 218 219 252) 252 253 | CONTENTS, 7.13.1 Definitions 7.132 Algorithms 7.14 Substructuring for Double Borded Block Diagonal Form 7.14.1 Main Algorithm for Substructuring 7.14.2 Simplified Algorithm for Substructuring Exercises CONDITIONING OF STRUCTURAL MATRICES 8.1 Introduction 82 Condition Numbers 82.1 The Ratio of Extreme Bigenvalues 82.2 Determinant of a Row - Normalized Matrix 823 The Ratio of Determinants 83. Weighted Graph and an Admissible Member 84 Optimally Conditioned Cycle Bases 84.1 Formulation ofthe Problem 8.5. Suboptimally Conditioned Cycle Bases 83.1 Algorithms 852 Examples 8.6 Optimally Conditioned Cutset Bases 8.6.1 Mathematical Formulation of the Problem 8.7 Suboptimally Conditioned Cutset Bases 87.1 Algorithms 8.72 Numerical Results Exercises MATROIDS AND SKELETAL STRUCTURES 9.1 Iniroduction 9.2 Axiom Systems for a Matroid 9.2.1 Definition in terms of Independence 9.22 Definition in terms of Bases 9.23 Definition in terms of Circuits 9.24 Definition in terms of Rank 9.3. Mateoids Relevant to Structural Mechanics 93.1 A Basis fora Finite Vector Space 9.32 A Basis for Cycle Space of a Graph 9.3.3 A Basis for Cutset Space of a Graph 934 Cycle Matroid of Graph 9.3.5 Cocycle Matroid of a Graph 93.6 Rigidity Matroid of a Graph 93.7 Matroid for Null Basis of a Matrix 94 Combinatorial Optimisation: the Greedy Algorithm 261 262 265 266 267 268 om am 21 2m 275 276 276 277 219 279 280 284 288 288 289 290 291 294w a CONTENTS 9.5 Application of the Greedy Algorithm: 9.6 Problems with Application of the Greedy Algorithm 9.6.1 Planar and Space Frames 9.62 Planar Trusses 9.63 Space Trusses 9.7 Formation of Sparse Null Bases, 97.1 Definitions 9.72 Null Bases Formation Exercises A GRAPH-THEORETICAL APPROACH FOR CONFIGURATION PROCESSING 10.1. Introduetion 10.2 Algebraic Representation of a Graph 10.3. Representations of Operations on Graphs 103.1 Addition of Two Subgraphs. 10.3.2 Subtraction of Two Subgraphs 104 Special Graphs 10.5 Some Funetions for Configuration Processing 10.5.1 Translation Functions 103.2 Rotation Functions 10.5.3. Reflection Functions 10.54 Projection Functions 10.6 Geometry of Structures 10.7 _ Extension to Hypergraphs Exercises GRAPH SYMMETRY FOR EIGENSOLUTION (OF DYNAMIC SYSTEMS Introduction Decomposition of Matrices to Special Forms M21 Formd 1122 FormIt 1123 Formil 11.3 Laplacian Matrices for Different Forms 113.1 Symmetry and Laplacian of Graphs 113.2 Factorization of Symmettic Graphs 113.3 Form IiLas an augmented Form IL 1134 Mixed Models 308 309 310 BIL BIL 312 312 312 314 336 114 us 16 m7 us CONTENTS Bigenveetors ofthe Form Il Matrices I14.1 Bigenvectors of M and Condensed Submatrices C and D 1142 Formation of Eigenvectors Eigenveetors for Matrices of Form IIL Generalized Form Ill Matrices Vibrating Cores for a Mass-spring Vibrating System The Graph Model of a Mass-spring system 18.1 Vibrating System with Form Il Symmetry 1118.2 Vibrating System with Form If Symmetry 1183 Generalized Form IIl and Vibrating System IL84 Discussions Exercises REFERENCES INDEX 366 372 373 315 379 382 384 385 388 391 384 385 399 15:List of Symbols List of Symbols ‘The important symbols, their definitions and the page upon which each symbol first occurs are listed inthe following: A A (AEB) oT Bp Bes) bu(S) bys) cross section area adjacency matrix bipartite graph with node sets A and B, and member set E 4 as, adjacency ist incidence matrix collection ofbase of a maroid agde-member incidence mattix particular soltion ‘complementary solution bipartite graph of S ero Beli number of first Betti number of S cements of a cycle basis cycle basis incidence matrix collection of cieuits of a matroid 146 159 159 90 16 20 293co GQ a > degin) det A iS) Structural Mechanies: Graph and Matrix Methods ceulset basis incidence matrix ceulset-nember incidence matrix cyele-member incidence matrix cycle ceyele graph with k nodes collection of circuits of « matroid entry of © the number of points with j members crossing crossing number of a planar drawing S? of S eyele adjacency matrix degree matrix ofa graph ccutset adjacency matrix degree of a node n, determinant of A mension of laste modulus eccentricity of unassembled flexibility matix of a structure svaxinum fromtwdth fixed end action colletion of independent subsets ‘overall flexibility matrix ofa structure 293 20 n 64 21 248 2 27 141 208 161 24 149 292 102 GRQ) 1jg.CLN IylN Ics. K-18) Ka Ky Ke cor r “Oo Py) LHS. M(S).M List of Symbols integers modulo 2 constant matix numberof handles uit max cycle length numberof the member jk incidence numberof the member jk integer coordinate system unassembled stifness mate stiffness matrix ofthe structure total graph bipanite graph ‘complete graph wth N nodes reduced stitfness matrix stiffness coefficient stiffness matrix ofan clement nits fea coordinate system sttness matrix of element I Laplacian matrix complementary Laplacian matrix total length of the eyeles (generalized eyces) ofa basis length of the path P, let hand side ‘umber of members of S 86 6 am 186 186 ai 125 125 258 14s 137 138 43 248 248Ma(S) MAS) Ms) ML. 8) ams) MSA 210(S) eS) NS).N N NG) NS) nF) NP. Structural Mechanics: Graph and Matix Methods number af members for triangulation of numberof members inthe exterior boundary of S ‘number of members of seme ist cycle matoid of cycle space matroid of smatroid coeyele matroid of cutset space matoid of S number of nodes of $ nl raph with k nodes ‘numberof intemal nodes of umber of nodes of S, nity of F, snon-deterministie polynomial time NP-complete, NP-hard OG) Pp FQ PP, node onder of an algorithm vector of pointers permutation matices path, path graph 59 47 19 295 295 292 296 296 9 a" 293 93 93 93 19 219 PRO! cca) RYS) RS) REF ROT AB) List of Symbols projection fnetion joint load veetor redundants vector row vector rank of number of internal regions numberof regions of embedded graph $ refletionfanetion rotation function ‘member (element) force vetor in global coordinate sytem clement force vector in loal coordinate system subgraph of S elimination graph k GS; ie-amion of k subgraphs bipartite graph with node set and B raph model ofa structure, finite st Set of elements of matroid shortest route tee tree transformation mateix cotree coefficient ofthe transformation mattix T 327 158 158 293 9 28 326 321 125 140 241 a 26 292 133 139TRAN we) We Ws wim) pl aye RF Structural Mechanics: Graph and Matix Methods translation function tei strain energy member distortion vector in global coordinate system clement distortion vector in local coordinate system degree matix joint displacement vector fi fe vector space Fiedler veetor relative displacements of cus ‘work done by external loads awa iS width ofa contour C, vector space ofa graph cust space ofa graph excl space ofa graph eight of member my cardinality of X soba coordinate system local coordinate system integers modulo p, denoted GF(p) Greek Symbols 318. 129 125 140 Is 124 295 49, 162 129 26 am 88 131 131 S) % 4S).DSI re) (8),DKE de ania 2S) ns(S) WA) ws) 8) x00) as) Ys) List of Synbols integer equal to 3 or 6 orientation coeiciens rotation angles numberof constraints sparsity coefficient of a matrix liameter of S 1pproximation tothe Fiedler vector degree of statical indeterminacy image of X degres of kinematical indeterminacy algebraie comnoctivity largest and smallest eigenvalues oa matix rember connectivity “The largest eigenvalue ofthe adjaceney matrix A (8) for planar tases = aM(AyHbN(AIE unifying linear fuetion approximate eigenvalue rank ofthe graph S rank of submatrix of H corresponding tothe members of X node cover number ‘matching mumber mt 191 132 124 228 250 2 2 49 2 102 248 9 48 4B 250 a7 31 31rhs os 8 1) ur Structural Mechanies: Graph and Matrix Methods ‘composition of the algebraic representation of subgraphs Logical Symbols, binary operations ring sum union collection of the sets of eigenvalues intersection an empty set a vector ora set ‘ansposi mn of a matrix containment symbol: a subset of set 314 CHAPTER 1 Basic Concepts and Definitions of Graph Theory 1.1 INTRODUCTION Graph theory is a branch of mathematics started by Euler (45) as early as 1736, It ‘took a hundred years before the second important contribution of Kirchhoff 139} had been made for the analysis of electrical networks. Cayley [22] and Sylvester [228] discovered several properties of special types of graphs known as trees. Poincaré [195] defined in principle what is known nowadays as the incidence ‘matrix of a graph, It took another century before the first book was published by ‘Konig [141]. After the second world war, further books appeared on graph theory, Ore [183], Behvad and Chartrand [11], Tutte [240], Berge (13], Harary [70], Gould (63}, and West [245], among many others. Graph Theory has found many applications in engineering and science, such as ‘chemical, civil, electrical and mechanical engineering, architecture, management ‘and control, communication, operational research, sparse matrix technology, ‘combinatorial optimisation, and computer science, ‘Therefore many books have been published on applied graph theory such as those by Bondy and Murty [16], Chen [24], Thulasiraman and Swamy [235], Wilson and Beineke [252], Mayeda [170], Christofides [26], Gondran and Minoux [61], Beincke and Wilson (12], Deo [37], Cooke et al. [32], Kaveh [110-111] and many others. In recent yeats, due to the extension of the concepts and applications of the graph theory, many journals such as Journal of Graph Theory, Journal of Combinatorial Theory 4 & B, Discrete and Applied Mathematics, SIAM Journal of Discrete Mathematics, European Journal of Combinatorics, and. Graphs and Combinatories are being ‘published fo caver the advances made in this field,2 ‘Structural Mechanies: Graph and! Matrix Methods Im this chapter basic definitions and concepts of graph theory are presented; however, for proofs and details the reader may refer to fextbooks on this subject, Res (63,70,245}, 1.2 BASIC DEFINITIONS, ‘There ate many physical systems whose performance depends not only on the characteristics of their components, but also on their relative location. As an example, in a stuctue, ifthe properties of a member are altered, the overall bbehaviour ofthe structure will be changed. This indicates thatthe performance of a structure depends on the characteristics of its members. On the other hand, if the location of @ member is changed, the properties of the structure will again be diferent, Therefore the connectivity (topology) of the structure influences the performance ofthe whole structure. Hence it is important to representa system so that its topology can be understood clearly. The graph model ofa system provides « powerfll means for this purpose. 1.2.1 DEFINITION OF A GRAPH A graph S consists of a set N(S) of elements called nodes (vertices or points) and a set M(S) of elements called members (edges or arcs) together with a relation of incidence which associates each member with a pait of nodes, called its ends, ‘Two or more members joining the same pair of nodes is known as a multiple ‘member, and @ member joining a node to itself is called a foop. A graph with no loops and multiple meinbers is called a simple graph. If N(S) and M(S) are ‘countable sets, then the corresponding graph S is finite. Since the great majority of the results in this book pertain to finite graphs with no loop and multiple members, ‘only simple finite graphs are needed, which are referred to as graphs. ‘The above definitions correspond to absiract graphs; however, a graph may be Visualized as a set of points connected by line segments in Euclidean space; the points are identified with nodes, and the line segments without their end points are identified with members. Such a configuration is known as a ropological graph. ‘These definitions are illustrated in Figure 1.1 LIN. hile merhere toon (a)'A non-simple graph. (6) A simple graph. Fig. 1.1. Non-simple and simple graphs. CHAPTER | Basic Concepts and Definitions a 1.2.2 ADIACENCY AND INCIDENCT Two nodes of a graph are called adjacent if these nodes are the end nodes of a member. A member is called incident with a node if it is an end node of that ‘member. Two members are called incident if they have a common end node, The degree (valency) of a node n, of a graph, denoted by deg (n), is the number of ‘members incident with that node. Since each member has two end nodes, the sum, fof node-degrees of a graph is twice the number of its members (handshaking, Jemma - known as the first theorem of graph theory). Fig. 1.2 A simple graph S. As an example, in Figure 1.2 two nodes n, and ng are adjacent. Node nis incident \with member m and m, and deg (m3) =4 1.2.3 ISOMORPHIC GRAPHS ‘Two graphs S; and S; are called isomorphic if there exists a one-to-one correspondence between their node sets and adjacency is preserved, As an example, the three graphs shown in Figure 1.3 are isomorphic, The word isomorphic is derived from the Greek words same und form. @ ) © Fig. 1.3. Three isomorphic graphs.4 ‘Structural Mechanics: Graph andl Matrix Methods 1.24 GRAPH OPERATIONS, A subgraph 8; of is a graph for which N(S)) < N(S) and M(S) & M(S), and each ‘member of S} has the same ends as in . ‘The zmion of subgraphs S, 82... ofS, denoted by S* SUSU x x U Sq i a subgraph of S with N(S!) = UNS.) and MS!) = MGS). The intersection of two subgraphs S, and S, is similarly defined using intersections of node-sels and member-sets of the two subgraphs. The ring sw of two subgraphs S; @ S,= $,US,~ 5,5, isa subgraph which contains the nodes and members ofS; tnd $, except those elements common fo S, and S, These definitions are illustrated in Figure 14 kx WY N ws Osi OS, @SUS, sn, ses, Fig. 14 A graph, two ofits subgraphs, union, intersection and ring sum, ‘There are many other useful operations, such as Cartesian product, direct product and strong Cartesian product, suecessfully applied to structural engineering [129] 1.2.5 WALKS, TRAILS AND PATHS. A walk w ofS, is finite sequence w = (my 5m, my yxy th, ny} Whose ferms are alternately nodes 1 and members my; of $ for 1 Sisk, and mand nate the two tends of mA rail tin S, is a walk in which no member of S appears more than ‘once. A path P in S, is a tal in which no node appears more than once. The engi ‘ofa path P, denoted by L(D,) is taken as the number of its members. P is called the shortest path between the two nodes np and m, if for any other path P, between | CHAPTER 1 Basie Concepts and Definitions 5 these nodes L(P.) < L(P). The distance between two nodes of a graph is defined as the number ofits members ofa shortest path between these nodes. ‘Asan example, in Figure 1.5, w= (My ms, Mg Mg Mey Ne Nyy 5 Ms Me My, Ms My) isa walk between n) and ns, in which member my and nodes n» and ns are repeated, twice. (2) A walle w in. (b) A wail tin, Fig. 15 A walk, (©) A path Pin. (rail and a path in, t= (0) ,ms,ms,.mg 5m mp5 Me my, Me Myo My) isa trail between n; and ny in which node ns is repeated twice. P=(Mjy Mey Me Me Ms My Ne) jis a path of length 3 in which no node and no member is repeated, 1.2.6 CONNECTEDNESS ‘Two nodes 1, and n, are said to be connected in $ if there exists a path between these nodes. A graph S is called connected if all pairs ofits nodes are connected. A. ‘component of a graph $ is @ maximal connected subgraph, ie. it is not a subgraph Of any other connected subgraph of S. These definitions are ilustrated in Figure 2d Nw (a) A connected graph. (8) A disconnected graph. Fig. 1.6 A connected graph and a disconnected graph with 3 components.6 Structural Mechanies: Graph and Matrix Methods 1.2.7 CYCLES AND CUTSETS. ‘A cycle isa path (ny myn)... my ym) for which np = n, and p 2 1; ie. a cycle is a closed path. Similarly, a closed trail (hinged eycle) and a elosed walk can be defined, Figure 1.7 XX XX (6) A hinged cycle of (a) Acycle of 8 Fig. 1.7. Two cycles of S. A cuset in a graph $ is a set of members whose removal from the graph increases the number of connected components of S, Figure 1.8(a). Ifa cutset results in two ‘components §} and S, then itis known as prime eutset, Figure 1.8(b). A link is a ‘member with its ends in two components produced by a cutset, Links are shown in bold lines. Ina prime cutset, if one of the components S, or S; consists of a single node, then the prime cutset is called a coeyele, Figure 1.8(¢), Mx KX Kx (a) Acuset of S, (6) A prime cutset of . Fig. 1.8 Different cutsets of 8 (6) A cocyele of S, 1.28. TREES, SPANNING TREES AND SHORTEST ROUTE TREES A mree T of $ is a connected subgraph of S, which contains no eycle, A set of trees Of S forms a forest. Ifa tree contains all the nodes of S, it is called a spanning tree of S. For simplicity it will be referred to asa iree, from now on. ‘A shortest route tree (SRT) rooted at a specified node ny of Sis a tree for which the distance between every node n, of T and ny is a minimum, An SRT of a graph ‘ean be generated by the following simple algorithnn: Label the selected root as "0" and the adjacent nodes as "I", Record the members incident to "0 as tree members. Repeat the process of labelling with "2" the ‘unnumbered ends of all the members incident with nodes labelled as "I", again CHAPTER | Basic Concepts and Definitions 7 recording the tree members. This process ter labelled and all the tree members are recorded, The lubel of the lst node indicates, the lengih of the SRT and the maximum number of nodes with the same label is defined as the width of the SRT ‘The above definitions are illustrated in Figure 1.9. The length and width of the SRT in Figure 9(4) are 2 and 3, respectively It is easy to prove that fora tree T, M(T)=N(T)= 1, aay ‘where M(T) and N¢T) are the numbers of members and nodes of T, respectively. ‘The complement of Tin S is called a corree, denoted by T*, The members of T are known as branches and those of T* are called chords. For a connected graph S, the ‘number of chords is given by: M(T*) = M(S) = MCT), (12) Since, N(T)=N(S), hence, M(T*) = M(S) = N(S) +1, (3) ‘where M(S) and N(S) are the numbers of members and nodes of S, respectively. [Notice that fora sot and its cardinality the same notation is used and the difference should be obvious from the context. XX (a) A graph s. 4 (@ An SRT rooted from my, (e) A cotree shown in dashed lines. Fig. 1.9 A tree and a cotree of S. (b) A tee of S (©)A spanning tree of .8 Structural Mechanics: Graph and Matrix Methods 1.3. DIFFERENT TYPES OF GRAPHS In order to simplify the study of properties of graphs, different types of graphs have bbeen defined. Some important ones are as follows: A null graph is a graph which contains no members. Thus Ny is a graph containin isolated nodes. rap 8 A eyele graph is a graph consisting of a single cycle, Therefore C, is a polygon with k members, aie poly A path graph is graph consisting ofa single path. Hence P, i a path with k nodes and (k-1) members, ‘A complete graph is a graph in which every two distinet nodes are connected by cexaclly one member, Figure 1.10. AK K K& Ky Ks Fig. 110 Some complete graphs. A complete graph with N nodes is denoted by Ky. It is easy to prove that a complete graph with N nodes has N(N ~ 1)/2 members A raph is calle bipartite, i the comesponding node set an be spit int two set ‘and N; in sucha way that each member of Joins anode of Nyt a node of Ns ‘complete bipartite raph is a bipartite yap in which each node N, i oined © cach node of N; by exatly one member Ifthe number of noes in N, aad Near dense bya, respectely then a complet bipartite graphs denote by Kes Examples of ipaite and conpicte bipartite graphs are show a Figure 111 (@) A bipartite graph, (b) A complete bipartite graph Ky. Fig. 1.11 Two bipartite graphs, CHAPTER 1 Basic Concepts and Definitions 9 1.4 VECTOR SPACES ASSOCIATED WITH GRAPHS In this section, itis shown that a vector space can be associated with a graph, and the properties of two important subspaces of this vector space, namely eyele and coulset spaces, is studied. For this purpose, simple definitions from sets, groups, fields and vector spaces are briefly presented. The material presented in this seetion is based on the work of Thulasiraman and Swamy [235] 1.4.1 GROUPS AND FIELDS Consider a finite set S = {b,¢,..}, and define a binary operation + on S. This ‘operation assigns to every pair (a and b)eS a unique element denoted by a+b. The set $ is said to be efosed under + if the element (a*b)e S, whenever (a and b)eS. ‘The operation + is said to be associative if sX{b+c) = (atb)+e for all a, b and c in SS. The operation i called commutative ifa+b = b+a for alla and bin S. Definition 1: A set S with a binary operation +, called addition, is a group if the following postulates hol: 1. Sis closed under + 2. The operation + is associative. 3. Thete exists a unique element c€S such that a+e= e+a= a forall aes. 4, For each element a&S there exits a unique element b such that bta = atb = e. ‘The element b is known as the inverse of a, and vice versa. Obviously the identity clement eis its own inverse. A group is called abelian if the operation + is commutative. Examples: The set $= {..0-2-10/H1y2,..} consisting of all integer numbers under usual addition for + forms a group. Here, 0 isthe identity element and -a is, the inverse of a €S. This group is also abelian, Another example isthe set Zp (0,1,2,..P=1} of integers with modulo p addition operation. If a= mp+q for 0S q $ p-a, then in modulus arithmetic a = q(modulo »). In this group, 0 is the identity element and the integer p-a is the inverse of a, except 0 which is its own inverse. As an example, the addition table of Z5 is shown in Table 1.10 ‘Structural Mechanics: Graph and Matrix Methods Tablel.1 Table of Z, + oti 2 ofolifa Poa 2 | 2 0 1 Definition 2: A set F with two operations of addition (+) and multiplication (© ) is a field ifthe following postulates hold: 1. Fis abelian group under +, with the identity element denoted as. 2. The set F— {e} is an abelian group under ©, the multiplication operation. 3. The multiplication operation is distributive with respeet to addition, ie. 46 (bte) = (ao b) + (ave) for all a, band in . Example: Consider Z,={0,1,2, ... .p-1} again with addition (modulo p) and ‘multiplication (modulo p) as the two operation. Tt can be shown that the set Z-{0}={1,2, ... p=1} is.am abelian group ifp is prime, Therefore 2, isa field if isa prime. The set Z; of infegers modulo 2, denoted GF(2), is an important field in our study with 0-0 ), 140 OF1 = Land 1+1=0, 000=0,100=001=0,and Lo 1.42 VECTOR SPACES ‘Consider a set S with a binary operation G2, Let F be afield with + and © being the addition and multiplication operations, respectively. A multiplication operation, denoted by *, is also defined between the elements of F and those of S. This ‘operation assigns to each ordered pair (cs) a unique clement denoted by cs, where o: is in F and s is in S. The set $ is a vector space over F if the follow postulates hold ome m 1. Sis an abelian group under CHAPTER 1 Basic Concepts and Definitions mn 2, For any elements 0, and fin F, and any elements s, and s, in S the followings hold: es) (os, )V(a*s2) and (c+ BY) +5) EBesy) 3. For any clement cand Bin F and any element sin S: (a=P)+s= 0485) 4, For any element s in S, 1¢s= , where | isthe multiplicative identity i F Consider a veetor space S, over the field F, The elements of $ are called vectors and those of Fare known as scalars, Ifan element s of S is expressible as, 5 (its) Blues, B.... (aps), a Where ss are vectors and 04's are scalars, then sis said to be a linear combination Of $8 oo: VeoIOMS Sy» 8 ae Said to be linearly independent if no vectors in this set is expressible as 2 linear combination of the remaining vectors inthe set. Vectors 818s, ... ss form a basis in the vector space S if they are linearly independent and every vector in S is expressible as a linear combination of these vectors. The vectors s),83,-. sare known as basis vectors. The dimension of the vector space 8, denoted by dim(S), is the number of veetors in a basis of S. If S° ia subset of the vector space S over F, then S’ is a subspace of Sif 8’ is also a vector space on F, The direct sum S;E) $;of two subspaces $, and 8; of Sis the set of all vectors of the form sEls;, where s) ¢ S; and s; € Sp It can be proved that 61S; is also subspace, and its dimension is given by dim(S,Bl $y) = dim(S\) + dima(S:) ~ dim(S 85). (as) Note that $8: is also subspace whenever S, and S» are subspaces. 143 VECTOR SPACE OF A GRAPH Consider a graph S = (NM) and let Ws denote the collection of al subsets of M, including the empty set ©. Under ring sum operation © between sets, in the following it is shown that Ws is an abelian group. Defining a suitable ‘multiplication between elements of the field Z, and those of Ws, it can be shown, that Ws isa vector space over Z,,2 ‘Structural Mechanics: Graph and Matrix Methods It can be shown that W is closed under @. The operator © is associative and commutative. Further for any element M; in Ws, M;® @ = M,and M,@ M,= 2 ‘Therefore for the operation ®, @ is the identity element, and each M; is its own inverse. Hence Ws is an abelian group under ©. Let *, a multiplication operation between the elements of Z, and those of Ws be defined as follows: leM,=M, and O*M,= 2. With this definition of + one can verify that the elements of Ws satisfy the following other requirements of a vector space: 1. (or+B)*M) = (a*M,) @ (B*M)). 2, aP(M,@ M) = (aM) @ (as). 3.(@BYYM= ar(eM), 4. 14M = M, (Notice tha isthe multiplicative identity in Z) ‘Therefore Ws is @ vector space over Z:. The dimension of this space is equal to the ‘number of members of the graph S, ‘Since each member-induced subgraph of S corresponds to a unique subset of M, ‘and by definition the ring sum of any two member-induced subgraphs corresponds to the ring sum of theit corresponding member ses, itis obvious thatthe set of all ‘members-induced subgraphs ofS is also a vector space over Z, if the multiplication is defined as follows 1#Mi=M, and 0%M;= ©, the null graph having no nodes and no members. ‘This veotor space will also be referred to by the symbol Ws, 1.44 CYCLE SUBSPACE AND CUTSET SUBSPACE OF A GRAPH Now we study two important subspaces of Ws, namely eycle space and cutset space of a graph, ‘Theorem {The set of al simple cytes and union of member-dijoint eyes oF a graph, We, is a subspace Ws of 8, vores CHAPTER 1 Basic Concepts and Definitions B Consider C; and C, a8 two eyeles of We. To prove the theoren one should show that C)@ C, is algo weyele C, belonging to We. Let n be a node of C,. This node is present at least in of the two cycles C, and C;. Let M, (=1,2,3), denotes the rmemabers incident ton in C,, Let [0 shows the number of members of C;, thus [My is the number of members incident ton in C, Note that [M,| and [MA are both even and one of them may be zero Furthermore [My i non-2er0. Since C=C; ® C2, we have: My=M.@M;, Therefore: [Mi =|] © [v4] ~2)M,Mg Since [Mi and [Mi are both even, thus [Mb] is also even, This is true forall he nodes oF Cy, and it follows that isa cycle in We andthe theorem is proven, ‘Theorem 2: The set of al cutsets and the union of member-disjoint cutsets We in «graph S, isa subspace ofthe vector space Ws of S. Itcan be shown that the ting sum of any two cus in a graph is a cut in. Similarly the union of any two member-disjoint cuts in a graph $ is also a cut in S. Since We is closed under the ring sum operation thus the proof is completed 1.4.5 FUNDAMENTAL CYCLE BASES, A special cycle basis known as a findamental cycle hasis can easily be constructed corresponding to a tree T of S. In a connected S, a chord of T together with T contains a eyele known as a fimdamental eycle of S. Moreover, the fundamental cycles obtained by adding the chords to T, one ata time, are independent, because ceach cycle has a member which is notin the others, Also, every eyele C, depends fon the set of fundamental cycles obtained by the above process, for C; is the symmetric difference of the eyeles determined by the chords of T which lie in C, ‘Thus the eycle rank (cyclomatic number, frst Betti number, nullity) of graph S ‘which is the number of cycles in a basis ofthe eycle space ofS, is given by, bY(S)=M(S) - NOS) +1, a6 and if $ contains bp(S) components, then: by(S) = M(S) — N(S) + bal), (7)4 ‘Structural Mechanics: Graph and Matrix Methods ‘A formal proof is provided in Section 1.4.7. ‘As an example, the selected tree T and four fundamental eycles of S are illustrated se 2A graph $ and a fundamental cycle basis of S. 1.4.6 FUNDAMENTAL CUTSET BASES AA basis can be constructed for the eutset space of a graph $, Consider the tree T and its cotree T*, The subgraph of S consisting of T+ and any member of T (branch) contains exactly one eutset known as « findamental cutset. The set of cutsets obtained by adding branches of T to Ts, one ata time, forms a basis for the eautset space of S, known as a fiidamental cutset basis of S. The cutser rank (rank of $) is the number of cutsets in a basis for the cust space of S, which is given by p(S)=N(S)—1, (a8) and for @ graph with by(S) components (8) = N¢S) ~ bS). as) A formal proof is provided in Seetion 1.47. A graph S and a fundamental cutset basis of $ are shown in Figure 1.13, A branch of the tree subdivides the nodes of the tree into two subsets. The members of a ‘cutest should have one end in each subsets, CHAPTER 1 Basic Concepts and Definitions 15 PYU= (b) A tree T of S. (0) A graph 8. (©) Cottee T* of T. Fig. 1.13 A graph $ and a fundamental cutset basis of S, 1.4.7 DIMENSION OF CYCLE AND CUTSET SUBSPACES Consider Ts a spanning tee of a connected graph $ with N nodes and M members. The branches of T are donoted by bb... Be and the chords by 1.3. “env Let C, and Cy refer to the fundamental cycle and the fundamental cutset With espoot oe, and b, , respectively. Since each findamental eycle contains exactly one chord, and this chord is present inno other fundamental cyele, therefore the fundamental eyeles Ci, Cs, .. Catns are independent, Using a similar reasoning for eutsets, it becomes obvious that all the fundamental eutsets Cj,C%, .. «C1 ane also independent. [Now we need to prove that every subgraph in eycle (cuset) subspace of $ can be expressed as a ring sum of C, (Cj). For this purpose consider any subgraph C in the eyele space of S. Let C contain the chords ¢i,00.-.te Let C’ denote the ring sum of the fundamental eyele C),Cine.Ce Obviously the chords esi. ATE present in C’ , and C’ contains no other chord ofthe T. Since C also contains these chords and no others, C’® C contains no chords.16 ‘Structural Mechanics: Graph and Matrix Methods Now itis claimed that C’6 C is empty. If this is not tre, then by the preceding arguments, C’® C contains only branches and has no cycle. On the other hand, being a ring sum of cycles, C7@ C is a cycle of the union of member-disjoint cycles. Thetefore the assumption that C” @C is not empty leads toa contradiction Hence C’ ® C is empty. This implies that Ci C2... © Ce, ie. every subgraph inthe eycle space ofS can be expressed as a ring sum of te fundamental cycles. {na similar manner it can be proved that every subgraph inthe cutset subspace of S ccan be expressed as a ring sum of the fundamental cusets, ‘The following fect can now be concluded |. The fundamental cycles with respect toa spanning tree of S constitute a basis for the eyele subspace of S, and therefore the dimension of the eyele subspace of S is equal to M-N+L 2. The fundamental cutsets with respect to a spanning tree of S constitute a basis for the cutset subspace of S, and therefore the dimension of the eutset subspace of Sis equal to N-L For graphs which ate not connected, spanning forest will replace the spanning tee, ‘and the dimensions for cycle subspace and cutset subspace will then be nullity of S = bi(S) = M-N¢by(S) and rank of S = p(S) = N-by(S), respectively, Here, by(S) is the number of components ofS, 1.48 ORTHOGONALITY PROPERTY ‘Two vectors are called orthogonal if their scalar product is zero. It can be shown that a vector is a eyele set (cutset) vector, if and only if it is orthogonal to every ‘vector of a eutset (cycle set) basis. Since the cycle set and cutset spaces of a graph S containing M(S) members are both subspaces of the M(S)-limensional space of all vectors which represent subsets of the members, therefore the eycle set and cutset spaces are orthogonal components of each other. 15 MATRICES ASSOCIATED WITH A GRAPH Matrices play a dominant rote in the theory of graphs and in particular in its applications to structural analysis, Some of these matrices conveniently describe ‘the connectivity properties of a graph and others provide useful information about ‘the patterns of the structural matrices, and some reveal additional information about transformations such as those of equilibrium and compatibility equations, CHAPTER 1 Basic Concepts and Definitions " Im this section various matrices are studied which reflect the properties of the corresponding graphs. For simplicity, all the graphs arc assumed to be connected, since the generalization to non-connected graphs is trivial and consists of ‘considering the direct sum of the matrices for their components 1.5.1 MATRIX REPRESENTATION OF A GRAPH ‘A graph can be represented in various forms, Some of these representations arc of theoretical importance, others are useful from the programming point of view when applied to realistic problems. In this section six different representations ofa graph are described. ‘Node Adjaceney Matrix: Let S be a graph with N nodes. The adjacency matric A is an NxN matrix in which the entry in row i and column j is 1 if node m is ‘adjacent to n,, and is 0 otherwise. This matrix is symmetric and the row sums of A are the degrees of the nodes of S ‘The adjacency matrix ofthe graph S, shown in Figure 1.14, is a SxS matrix as: momp my mg ns (119) nfo 1 1.0 0 ml 0 1 1 0 Azmll 1 0 1 1 nfo 1 1 0 1 aslo o 1 1 0 Fig. 1.14 A graph S. It can be noted that A is a symmetric matrix of trace zero, For two isomorphic graphs S and 8’, the adjacency matrix A of $ can be transformed to_A' of by simultaneous permutations of the rows of A. The (ij)th entry of A? shows the ‘number of walks of length 2 with n, and 1, as end nodes. Similarly, the entry in theIs Structural Mechanies: Graph and Matrix Methods (ig) position of A* is equal to the number of walks of length k with n, and n; a8 ened nodes. The polynomial, (2) = det A), aay is called the characteristic polynomial of S. The collection of N(S) eigenvalues of Ais known as the spectrum of §, Since A is symmetric, the spectrum of S consists of N(S) real numbers. The sum of eigenvalues of A is equal 0 zero, The eigenvectors of A are orthogonal, Node-Member Incidence Matrix: Let $ be a graph with M members and N nodes. The node-member incidence matrix Bis an NxM matrix. in which the entry in row i and column j is 1 ifnode 1, is incident with member 1, and is 0 otherwise. ‘As an example, the node-member incidence matrix of the graph in Figure 1.14 isa 9X7 matrix ofthe form: my mg my my ms mg my afi ot 0 0 0 oo} HO go a 7 7 11 B=m)/1 1 0 1 1 0 0 a foo o o 1 14d nfo ro 0 0 1 0 Obviously, the pattem of an incidence matrix depends on the particular way that its nodes and members are labelled, One incidence matrix ean be obtained from ‘nother by simply interchanging rows (corresponding to relabelling the nodes) and columns (corresponding to relabelling the members). The incidence matrix B and the adjacency matrix A ofa graph $ are related by, Be ALY (13) ‘where V is a diagonal matrix of order N(S), known as the degree matrix, whose typical non-zero entry vj is the valency of the node n, of § for i=1, .. N(S). ‘The rows of B ate dependent and one row can arbitrarily be deleted to ensure the independence ofthe est ofthe rows, The node corresponding to the deleted row is called a lat (reference) node. The matrix obtained after deleting a dependent row is called an incidence marry ofS and it is denoted by B. Although A and B are of great theoretical value, however, the storage requirements for these matrices are high and proportional fo NxN and Me<(N-1), respectively. In CHAPTER 1 Basie Concepts and Definitions 19 facta large number of unnecessary 72r0s is stored in these matrices. In practice one ‘can use different approaches to reduce the storage required, some of which are described in the following. Member List: This type of representation is a common approach in structural ‘mechanics, A member list consists of two rows (or colurans) and M columns (or rows). Each column (or row) contains the labels of the two end nodes of each ‘member, in which members are arranged sequentially. For example, the member list of Sin Figure 1.14 is: my my m3 my mg me my Cs 3} 15 ilsets 2. 3 coed It should be noted that a member list can also represent orientations on members, ‘The storage required for this representation is 2M, Some engineers prefer to add a third row containing the member's labels, for easy addressing. In this ease the storage is increased to 3xM. A different way of preparing a member list i to use a veetor containing the end rhodes of members sequentially; e.g for the previous example this vector becomes: (13 53,551,523 534245 524), (ts) ‘This is a compact description of a graph; however, itis impractical because of the extra search required for its use in various algorithms Adjacency List: This list consists of N rows and D columns, where D is the ‘maximum degree of the nodes of S. The ith row contains the labels of the nodes: adjacent to node i ofS. For the graph S shown in Figure 1.14, the adjacency list is 16) z BE i how ‘The storage needed for an adjacency list is NxD. ‘Compact Adjacency List: In this list the rows of AL are continually arranged in a row vector R, and an additional vector of pointers P is considered. For example, the compact adjacency list of Figure 1.13 can be written as:20 Structural Mechanics: Graph and Matrix Methods R= 23.13 4,6.24,523,53.4), P= (1,3,6,10,13,15). (ay P is a vector (p,, ps, ps, .- ) which helps to list the nodes adjacent to each node. For node n, one should start reading R at entry p, and finish at pj ‘An additional restriction can be put on R, by ordering the nodes adjacent to each node n, in ascending order of their degrees. This ordering can be of some advantage, an example of which is nodal ordering for bandwidth optimisation. The storage required for this list is 2M+N + 1. 1.5.2. CYCLE BASES MATRICES The eyele-member incidence matrix © of « graph S, has «row for each eyele or hinged cycle and a column for each member. An entry & of © is 1 if eycle C, contains member m, and it is 0 otherwise. In contrast to the node adjacency and node-member incidence matrix, the cycle-member incidence matrix does not determine a graph up to isomorphism; ie. two totally diferent graphs may have the ‘same eyele-member incidence matrix. For a graph $ there exists 2°") —1 cycles or hinged cycles, Thus © is a (2%) <1 pat matrix. However, one does not need all the cycles of S, and the clements of a cycle basis are sufficient, For a cycle basis, a cycle-member incidence matrix becomes a by(S)xM matrix, denoted by C, known as the cycle basis incidence matrix of 8. As an example, matrix C for the graph shown in Figure 1.14, for the following cycle basis, Com (mm sm) a= (om, ms,m) y= (m4, ms,m)) is given by mm, Mm; my ms me my qfi o 1 1 0 0 o c=c,J0 1 0 0 1 1 of 00 0 1 oe (1-18) CHAPTER | Basie Concepts and Definitions 2 ‘The evele adjacency’ manrix D = CC! — W is a b,(S)xby(S) matrix, each entry dy of which is 1 iFC, and C, have at least one member in common, and it is 0 otherwise. For the above example, (19) ‘where W is a diagonal matrix, in which a typical non-zero entry wy isthe length of the eyele C;. The trace of CC is equal to the total length of the eycles in the basis, ‘An important theorem can now be proved which is based on the orthogonality property mentioned in Section 1.48, ‘Theorem: Let S have incidence matrix B and a cycle basis incidence matrix C. ‘Then: B= 0 (mod 2). (1-20) Proof: Consider the ih row of € and the jth column of BY, which is the th row of B, The rth entry i these two rows are both non-zero if and only ifm, is in cycle C, and is incident with n, If me is in C, then nis also in C,, but ifn, isin the eye, then there are two members of G, incident with n, so that the (i,j)th entry of CB is +1 = O{mod 2), and this completes the proof “Matrix € for a fundamental eyele basis with special labels for its tree members and chords, finds a patiular pattem, Let § have a tee T whose members are M(T) = (Gy, me, «and a cotee for whieh MCT*) = (Rp, mp ys) Then there i @ unique fundamental cycle C, in S ~ M(T") +m, p+l Sis M(S) and this set of ‘yeles forms a basis forthe eyele space of S. As an example, for the graph S of Figure 1.12 (page T4) whose members ae labelled as shown in Figure 1.15, the fundamental cycle basis consist of; a Cy (a, m, my) (my, ms, my) C= (mn, my mg, Ms, Mp), Cy= (5 My Mo, My Ms) ‘The corresponding C for the selected tree T is denoted by Cy and has the following form:2 ‘Structural Mechanies: Graph and Matrix Methods my mz my my ms me my My My Mp cyt 1 0 0 0 oF 1 0 0 oO Clo 00 oo oe “orelt 1 0 1 1 fo 0 1 ofr 0 le ot ooo Mery a) 2 Fig. 1.15 A graph and its tree members, 153. CUTSET Bases Marerces ‘The cutset-member incidence matrix © for a graph S, has a row for each cutset of S and a column foreach member. An entry &} of © is | iPeuset CP contains member my and itis 0 otherwise, This mati ike © does not determine a graph completely. Independent rows of ” for a cutset, basis, denoted by C*, form a. ‘matrix known as a curser basis incidence matrix, which is a p(S)xM matrix, p(S) being the rank of graph S. As an example, C* for the cutset of Figure 1.13 with ‘members labelled as in Figure 1.15, is given below: off ooo oo 1 on clo oo 1 oo 0 on eff 1 oo OOF (1-22) cJoo oo 1 oo on GJo 0 1 0 0 0010 slo 0 0 0 0 1 00 0 CHAPTER | Basie Concepts and Definitions 2B ‘The cuter adjacency mais D'=C°C™ is a p(S)xp(S) matrix. defined analogously to eycle adjacency matrix D, For a fundamental cuset basis with appropriate labelling of the members in T and 'T*, the particular pattern of C* becomes: 100 000f1010 — oor eo to Gay oe eee © a cooootooot vom the ethogonaliy condition, CC =0, hence 0 1 ter af dh|-0 Hence Cy+€2" = O(nod2),resuing in (25) Cr=ct Therefore for a graph having Co, one can construct ¢% and vice versa ‘There exis a very simple basis for the cutset space of a graph, which consists of N-I eocycles of S. Asan example, for the graph of Figure 1.14, considering ns asa datum node, we have, mom, mj my ms mg my i 0 100 0 4G) (1-26) | 0 o 1 1 oo 1 trod o 0 0 0 0 1 which is the same as the incidence matrix B of S. The simplicity of the displacement method of structural analysis is due to the existence of such a simple basis,4 Structural Mechanics: Graph and Matrix Methods 1.6 DIRECTED GRAPHS AND THEIR MATRICES [An oriented or directed graph is graph in which each member i assigned an orientation. A member is oriented ftom its initial node to its final node, as shown in Figure 1.16(a) The initial node is said to be positively incident on the member and the final node negativey incident, as shown in the Figure: @ © © Fig. 1.16 An oriented member, a directed graph and a directed tre. The choice of orientation of members of a graph is arbitrary; however, once itis chosen, it must be retained. Cycles and cutsets can also be oriented as shown in, Figure 1.160). Asan example, mis positively oriented in eyele C,, and mo is negatively oriented incutset C}. All the matrices B, B, C and C* can be defined as before, with the difference of having +1, ~1 and 0 as entries, according to whether the member is positively, ‘negatively and zero incident with a cuset ofa eycle ‘Asan example, for graph S in Figure 1.16(b) the matrix B with n, as datum node is formed: m4 5 ™ 0 nf-1 00 1 0 o 0 0 ajo 10 0 -1 10d) pemfo 01 o 10 0 a2 ng} 0 0 0 0 Ot 0 alo 0 0 1 o-11 alo 0 0 0 Ome! CHAPTER I Basic Concepts and Definitions 25 Consider a tree as shown in Figure 1.16(c), The corresponding cycle basis ‘neidence matrix ean be written as: mom: a3 my as me m7 mE MY MO qf 1 00 0 1 0 OD) co 11 0 0 ofo 1 0 Ol (128) 3) 1 0/0 0 19) Calo -11lo 0 OF Ce Obviously, CB'= (mod 2), (1-29) ‘wth similar proof as that ofthe non-orented case -Acutset-member incidence matrix is similarly obtained as: me ms eM Gf 0 0 0 0 of-1 0 -1 0 Glo 10 00 Ok (130) rl Ole 0 6 1 Oy 0 ig clo oo 0 1 ofo 0 4 1 clo oo 0 0 1Lo 0 o cp ce Itcan easily be proved tat: 31) 1.7 GRAPHS ASSOCIATED WITH MATRICES, ens re nF hee Si svete ma St bho se hy ma ee} si he fn) oe cadre26 Structural Mechanics: Graph and Matrix Methods directed graph associated with a non-symmetric matrix is usually called a digraph and the word graph is used for the undirected graph associated with a symmetric matrix Fig. 1.17 A non-symmetric matrix H and its associated digraph, + Fig. 1.18 A symmetsic matrix Hand its associated graph, For H to be viewed as the adjacency matris, due to the presence of diagonal entries, one loop should be added to each node. However, since the structural models have always non-zero diagonal entries and eontain no loops, this addition is disregarded. With a rectangular matrix & 2 bipartite graph $= (A,B) can be associated. For each row of E-a node of A and with each column of E.a node of B is associated. Two nodes of A and B are connected with a member ofS ifey is non-zero, An example of this is shown in Figure 1,19, Fig. 1.19 A rectangular matrix K and its associated bipartite graph, CHAPTER | Basic Concepts and Definitions n [A weighted bipartite graph can be defined for any mn rectangular matrix H, The nodes thfayfn and nodes C1,C..Cq of the graph correspond to rows and columns of H, respectively. If ly 0, then 5 is joined to e, by a member whose ‘weit is hy. 1.8 PLANAR GRAPHS - EULER'S POLYHEDRA FORMULA Graph theory and properties of planar graphs, were first diseovered by Euler in 1736, After 190, years Kuratowski found a criterion for a graph to be planar Whitney developed some important properties of embedding graphs in the plane. ‘MacLane expressed the planarity in terms of the graph’s eyele basis. In this seetion some of these criteria are studied, and Euler’s polyhedra formula is proven. 18.1 PLANAR GRAPHS. ‘A graph $ is called planar if it can be drawn (embedded) inthe plane in such a way ‘that no two members cross each other. As an example, a complete graph K, shown in Figure 1.20 is planar since it can be drawn in the plane as shown: N A (a) Ky (b) Planar drawings of K. Fig. 1.20. Ky and two of its drawings On the other hand Ks, Figure 1.21, is not planar, since every drawing of Ks ‘contains atleast one crossing, Be @ Ks (b) Two drawings of Ks with one crossing, 1K and two ofits drawings.28 Structural Mechanics: Graph and Matrix Methods Similarly Ky, Figure 1.22, snot planar, as ilstrated: (@) Ks. (b) Two drawings of Ks, with one crossing, Fig. 1.22 Ky3 and its drawings. A planar graph S drawn in the plane divides the plane into regions all of which are bounded and only one is unbounded, IFS is drawn on a sphere, all the regions will >be bounded; however, the number of regions will not change. The eyele bounding a region is called a regional cycle. Obviously the sum of the lengths of regional cycles is twice the number of members of the graph ‘There is an outstanding formula that relates the number of regions, members and nodes ofa planar graph, inthe form, R(S) MIS) + N(S) =2, ‘where R(S), M(S) and N(S) are the numbers of regions, members and nodes of planar graph S, respectively. This formula shows that for different drawings of S in the plane, R(S) remains constant Originally the above relationship was given for polyhedra, in which R(S), M(S) and N(S) correspond to faces, edges and corners of a polyhedron, respectively. However, the theorem can easily be expressed in graphtheoretical terms as fallows. ‘Theorem (Euler [45)}: Let S be a connected planar graph, Then: R(S) = M{S) + N(S)=2. (32) Proof: For proof, S is reformed in two stages. Inthe frst stage, a spanning tree T of S is considered in the plane for which R(T) ~ M(T) + N(T) = 2. This is true since R(T) = 1 and M(T) = N(T) ~ 1 In second stage chords are added one at a time. Addition of a chord increases the number of members and regions each by lity, leaving the left hand side of Bq, (1-30) unchanged during the entire process, and the result follows, CHAPTER | Basie Concepts and Definitions 29 1.8.2. THEOREMS FOR PLANARITY In order to check the planarity of a graph, different approaches are available which, fre based on the following theorems. These theorems are only stated and the reader ‘may refer to textbooks on graph theory for proofs. ‘Theorem (Kuratowski [148]}: A graph S is planar ifand only ifit has no subgraph contractible to Ks or Kea. Contracting a member m= (a2) is an operation in which the member is removed ‘and ni is identified with n, so that the resulting node is incident to all members {other than m,) that were originally incident with n; or n; If @ graph S’ can be obtained from § by succession of memiber contractions, then Sis contractile to’. The process of the contraction of a member (n,n) of a graph is shown in Figure 1.23(a), and the contraction of the Petersen graph to Ky is illustrated in Pigure (@) Contraction of a member my, (©) Contraction of the Petersen graph to Ks, Fig. 1.23. The contraction of a member in @ graph, ‘Theorem (MacLane (167): A connected graph is planar if and only if every block of S with at least three nodes has a eyele basis, C),C2,..,Chy¢s) and one additional cycle Co, such that every member is contained in exactly two of these by(S)+1 ceyeles,30 ‘Structural Mechanics: Graph and Matrix Methods A block is a maximal non-separable graph, and a non-separable graph is a graph that has no cutpoints. A cu-poinr is a node whose removal increases the umber ‘of components and a bridge is such a member, In Figure 1.24, a graph and its blocks are illustrated! bridge ccut-point LNV Fig, 1.24 A graph and its blocks. Definitions: A graph S* is a dual graph of a graph $ if there is a 1-1 correspondence between the members of S* and those of S, such that a set of ‘members in S* is a cycle vector of $* if and only if the corresponding set of members in Sis cutset vector ofS, ‘Theorem (Whitney [250}) - A graph is planar if and only it has « combinatorial ual (a) A planar graph S, (b) The dual graph of . Fig. 1.25 A planar graph and its dual Fora connected planar graph S, the dal graph S* is constructed as follows To each region rofS thers i comespondng node 5° of S* ad to each member tof § there isa conesponding member mi in S*, such tha ifthe member my CHAPTER 1 Basic Concepts and Definit 31 occurs on the boundary of two regions andr, then the member m} joins the corresponding nodes of and «fin St Figure 1.2, 1.9 MAXIMAL MATCHING IN BIPARTITE GRAPHS 1.9.1, DEFINITIONS As defined before, a graph is bipartite i ts set of nodes can be partitioned into two sets A and B, such that every member ofthe graph has one end node in A and other in B. Such a graph is denoted by S = (A,B). A set of members of S is called a matching if no two members have a common node, The size of any largest ‘matching in S is called the marching number of S, denoted by y(S). A subset N¢S) NG), is the node cover of 8, if each member of S has at least one end node in N(GS). The cardinality of any smallest node cover, denoted by 2(8), is known as the node covering number of 8. 1.9.2. THEOREMS ON MATCHING In this section, three theorems are stated, and the proofs may be found in the book by Lovasz and Plumner [164]: ‘Theorem 1 (Konig {1-41,142)): For a bipartite graph S, the matching mumnber y(S) is equal to the node covering mumber 1S). ‘Theorem 2 (Hall [69): Let S = (A,B) be a bipartite graph. Then S has a complete ‘matching of A into B if and only if PCX) | 2 [X| for all X cA. TX) is the image of X, ie. those elements of B which are connected to the elements of X in S. Figure 1.26(a) shows a bipartite graph for which matching exists and Figure 1.26(b illustrates a ease where complete matching does not exist, because X = (a), a2) are matched to by, ie. FOS Xt: a 26 Matching in bipartite graphs.32 Structural Mechanies: Graph and Matrix Methods AA perfect matching is a matching which covers all nodes ofS, ‘Theorem 3 (Frobenius [53]): A bipartite graph $= (A,B) has a perfect matching and only if[A| = [B| and for each X cA, [F() |> KX ze (1) Perfect matching exists (b) Perfect matching does not exist Fig. 1.27 Perfect matching in bipartite graphs. ‘This is also known as the marriage theorem. Figures 1.27(a) and (b) show eases when a perfect matching exists and does not exis, respectively, ‘Therefore Frobenius’s theorem characterizes those bipartite graphs which have a perfect matching. Hall's theorem characterizes those bipartite graphs that have a ‘matching of A into B. Konig’s theorem gives a formula for the matching number of a bipartite graph, 1.9.3. MAXIMUM MATCHING. Let M be any matching in a bipartite graph $ = (A.B). A path P is called an ‘alternating path with respect to M, or an Malternating path if its mennbers (edges) are altemately chosen from the matching M and outside M. A node is exposed (unmatched, not covered) with respect to matching M if no member of M is incident with that node. An alternating tree telative to the matching, is a tree which. satisfies the following (wo conditions: first, the tee contains exactly one exposed node from A, which is ealled its roor, second, all paths between the root and any ‘other node in the tree are alternating paths. ‘As an example, in Figure 1.28(a) the path ajbyaybsa, is an alternating path with respect to the matching shown in bold lines; a2, bs, ay and by are exposed nodes. CHAPTER 1 Basie Concepts and Definitions 3 4 b a by a bs a by — ay bs ay by a be ay be () An arbitrary matching, (b) An augmented matching. Fig. 1.28 Operation for maximum matching ‘An M-alternating path joining two exposed nodes is called an M-ougmenting path. For every such path the corresponding matching can be made larger by discarding the members of P - M and adding those of P ~ M, where P is an M-altemating path bzaibyaybsay, Figure 1.28(b). Thus, if $ contains any Mcalternating path P joining two exposed nodes, then M can not be 4 maximum matching, for one can readily obtain a larger matching M’ by discarding the members of P.M and adding those of P ~ M. ‘Theorem 4 (Berge [13-14]: Let M be a matching in a graph 8. Then M is a ‘maximum matching if and only if there exists no augmenting path in § relative to M. ‘The above result provides @ method for finding a maximum matching in S. The ‘computational procedure for construction of a maximum matching begins with considering any feasible matching, possibly the empty matching. Each exposed node of A is made the root of an alternating tree, and nodes and members are added to the trees by means ofa labelling technique. Eventually, the following two ‘cases must occur: either an exposed node in B is added to one ofthe trees, or else it is not possible to add more nodes and members to any of the trees. Inthe former case, the matching is augmented and the formation of trees is repeated with respect to the new matching. In the latter ease, the trees are said to be Hungarian and the process is terminated, ‘Asan example, consider the matching shown in Figure 1.29(a), in which bold lines represent members in the matching. Alternating trees are constructed, with the exposed nodes 2, and as of A as roots, as shown in Figure 1.29(b). An augmenting path is found, as indicated in the Figure. Naturally, several different sets of alternating trees could have been constructed, For example, the tree rooted at node ay could have contained the member (a,b).4 Structural Mechanies: Graph and Matrix Methods oe ay ab; ay by ay bs abs (a) Abipartite graph S, Fig. 1.29 A bipartite graph and is alternating tree, (b) Alternating trees Augmented matching is shown in Figure 1.30. When the alternating tree of Figure 1.31 is used for the augmented matching, it beeomes Hungarian, a Fig. 1.31 Alternating tee for augmented matching, BIPARTITE MATCHING ALGORITHM Let X be any matching, possibly the empty matching ofa bipartite graph $ = (A,B), ‘No nodes are labelled. Step 1 (labelling) Ll Give label © to each exposed node in A. 1.2 If there are no unseanned labels, go to Step 3. Otherwise, find a node i with aan unscanned label. Ifi € A, goto Step 1.3; ie B, go to Step 1.4. CHAPTER 1 Basie Concepts and Definitions 35 1.3 Scan the label on node i (ieA) as follows: for each member (ij) € X incident to node i, give node j the label "i, unless node jis already labelled. Return to Step 1.2. 1.4 Scan the label on node i (ie B) as follows: if node iis exposed, go to Step 2 Otherwise, identify the unique member (ij)€X incident to node i and give node j the label "i", Return to Step 1.2 Step 2 (Augmenting): ‘An augmenting path has been found, terminating at node i (identified in Step 1.4) ‘The nodes preveding node i in the path, ate identified by backtracking, That is, if the label on node i is "j", the second-t0-last node in the path is j. Ifthe label on node j is "k", the third-o-last node is k, and so on. The initial node in the path has the label "2", Augment X by adding to X all members in the augmenting path that fare not in X and removing those which are in X. Remove all labels from nodes. Return to Step 1.1 ‘Step 3 (Hungarian Labelling) ‘The labelling is Hungarian, no augmenting path exists, and the matching X is of ‘maximum cardinality, For further study, the reader may refer to the original paper of Hoperoft snd Karp [80] o Lawler [157]. An algorithm using a different approach may be found in Ref. [6]. EXERCISES 1.1 In the following graph, which members are incident with node 3 (identify wilh their end nodes)? Which nodes are adjacent to node 4? What is the degree of node 2? 1.2 Are the following graphs isomorphic?36 Structural Mechanics: Graph and Matrix Methods 13 Draw a tree, a spanning tree and an SRT rooted at O for the following graph, Use 0" as the root of a second SRT and compare its length and width with those of the first one. é ° 1.4 What types of graph are the following? AAO 1.5 Listall the eyeles ofthe following graph: 1.6 Prove that fora planar graph embedded ona sphere with all triangular faces, MGS) = 3N(S) 6. 1.7 Find a fundamental eyele basis of the following graph using an arbitrary spanning tree: 37 a | | 1.8 In the above example use two SRTs rooted at O and O" and compare the length of the corresponding fundamental cycle bases. 1.9. Write the adjacency and member-node incidence matrices of the graphs in Exercise 1.2. Use an arbitrary node and member numbering What can you say about the resulting matrices? 1.10 Write C, €*, Cp and Cj, matrices for the following graph and examine the orthogonality property: 1.11 Identify the planar graphs inthe following figure: KY & WX @ ©) © @ 112 Prove that Ks and Ky, are not planar. 1.13. Euler's formula as in Eq, (1-32) fils for disconnected graphs. Ifa planar raph $ has bo(S) components, how can the formula be adjusted? 1.14 Find a maximal matching for the following bipartite graphs. Which one is complete and which one is a perfect matching?Las Structural Mechanies: Graph and Matrix Methods @ (by Why is there no complete matching forthe following bipartite graphs? ©) CHAPTER 2 Topological Properties of Skeletal Structures 2.1 INTRODUCTION In the analysis of skeletal structures, three different properties are encountered, which can be classified as topological, geometrical and material. Separate study of these properties results in a considerable simplification in the analysis and leads to 1 clear understanding of the structural behaviour. This chapter is confined to partial study of the topological properties of skeletal structures, since both displacement and force methods require such a study at the beginning of the analysis. The ‘number of equations to be solved in the two methods may differ widely for the same structure, This number depends on the size of flexibility and stiffness ‘matrices, which are the same as the degree of statical indeterminacy (DSI) and the Ses!US, 38! US, > SHS'US, 9 ShStUS, (b) An expansion process. Fig. 2.3 Formation of a planar simple truss, Similarly, a simple space truss can be re-formed by joining a subgraph S) ‘containing three members connected at a node, which are not in a plane, to a triangle to form a tetrahedron, Figure 2.4, This process should be continued to obtain the entire truss. At each stage of expansion, the intersection consists of three disjoint nodes. (@) A typical subgraph Sy, vb g A SS! 4 SIUSPS' 4 S'USRS' > S'USPst (b) An expansion process, 24 Formation ofa simple space truss, In the above expansion process, the properties of Sp, and the way this typical subgraph is joined to the previously expanded subgraph, guarantee the determinacy of the simple truss.46 Structural Mechanies: Graph and Matrix Methods ‘The idea can be extended to other types of structure. AS an example, a tree-shaped frame can be re-formed by adding one member at each stage of an expansion as shown in Figure 2.5. baat SS'98 5 8 4 soo > st Fig.2.5. Formation ofa tree-shaped frame model 2.3.3 AN INTERSECTION THEOREM In a general expansion process, a subgraph S, may be joined to another subgraph S; in an arbitrary manner, For example ¥(S:) or ¥{S)) may have any arbitrary values and the union $; US; may be a connected or a disjoint subgraph. The intersection $0 'S, may also be connected or disjoint, It is important to find the properties of S; U Sz, having the properties of Sj, S) and S; © $2. However, one cannot simply ‘employ an additivity formula ofthe following form, VS) US: Si) + ($1) WSLS), as is offen thought, As an example, consider a planar truss containing S; and $; a8, its subgraphs, Figure 2.6. In order o calculate the DSI of S by, 8) =M(S) - 2N(S) + 318), we have, YS) = 5-2dH3=0, (S)=1- 20243 =0, and S198) =0-Dd#2=2, leading to, 8: US) = 0+0-@)=~2, which is an incorrect answer, However, if (S; > $) is replaced by'7 (Sy 9 Ss M(S)—2N(S) +3, then the correct result will be obtained, 15108) ~ 20043 leading to: YS) US) = 040--)= +1 (CHAPTER 2 Topological Properties of Skeletal Structures “7 US s S SiS: Fig. 2.6 A truss, its subgraphs and their intersections. ‘The relationship given in the above becomes totally wrong if $, and/or S, contain ‘more than one component, The following theorem removes all the difficulties and leads to a correct calculation of the properties of 8, US, In order to have the formula ints general form, q subgraphs are considered in place of two subgraphs ‘Theorem (Kaveh {89D} Let $ be the union of q subgraphs S$, Ss, 8, with the following functions being defined: V(S)=aM(S) + BN(S) +evs(S), VES) = aM(S) + BNIS) + evuSi 4 (A) = aM(A)) + BN(A) + evu(AD) Booth where A\= 8! Sand S“!=$)U 8; UU Sy. Ten: a a [v(S) ~ eva S)] = YIVS})~evo(S)]~ LWA) evo A] (28) Proof: According to Eq, (2-7) IM{S) + BNIS) + ev) = afMUS.) + SIMKS))—MCADIT + BINS.) m vis) + SING )=NEADH +eV4S). 2 Adding, 4 a A fe YvolS)) ~ ¢ YivolSi) 1+ Le YivolAs)—¢ YvolAi)]=0, a a a and rearranging the terms, the proof is completed as follows:48 Structural Mechanics: Graph and Matrix Methods a (15) ~ ovdS)1= SY EaM(S,)+ NGS) Fel ~ FLAMCA)) +BNCAD +e] a a 4 Lows) -evotS91 — LIVAI-evn(AT a Special Case: If S and each of its subgraphs considered for expansion (S; for i= 1.1.4) are nondisjoint (connected), then Eq. (2-8) can be simplified as, v9) a a me Ev) - LA). 2) a where VA) =aM(A) +bN(A) +e. For caleulating the DKI and DSI of a complex structure or a structure with a large number of members, one normally selects a repeated unit ofthe structure and joins these units sequentially in a connected form, Thus, Eq. (2-9) can be applic place of Eq. 2-8) 10 obtain the overall property ofthe structure, 2.3.4 A METHOD FOR DETERMINING THE DKI AND DSI OF STRUCTURES Let S be the union of its repeated and/or simple pattern subgraphs S; (F1,...). Calculate the DKT or DSI of each subgraph using the appropriate coefficients from ‘Table 2.1, Now perform the following steps ‘Step 1. Join $, t0 $2 to form $? = S, U $2, and calculate the DKI or DSI of thei intersection Az ~ 8, © S, The value of v(S') can be found using Eg. (2-8) or Eq. 2-9), 88 appropriate. ‘Step 2. Join $; to? to obiain S = S? U Sy, and determine the DKI or DSI of Ay = SAS}, Similarly to Step 1, calculate WS). ‘Step k. Subsequently join S,,, to $', calculating the DKT or DSI of Ausi= S*0 Sst and evaluating the magnitude of v(S*") a Repeat Step k until the entire structural model is reformed and its DKI or DSI is determined. CHAPTER 2 Topological Properties of Skeletal Structures 49 In the above expansion process, the value of q depends on the properties of the ‘substructures (subgraphs) which are considered for reforming S. These subgraphs have either simple pattems for which v(S.) can easily be calculated, or the value of their DKI or DSI ate already known, In the process of expansion, if an intersection A, itself has a complex pattem, further refinement is also possible; ie. the intersection can be considered as the union of simpler subgraphs. Example 1: A simple structure $ is chosen to illustrate the steps and capabilites of the method, as showin in Figure 2.7(6). The union and intersections at each step are ited in Figures 2.7(c) and (@) The crossing points atthe first and third panels are also considered as extra nodes Kix KV (2) A skeletal structure and its graph model XM MH s: 8: 8 (©) Subgraphs ofS. SUS: A= S18: (©) The union and intersection of S, and S;,oo ater al Matter (CHAPTER 2 Topological Properties of Skeletal Structures st 8) is | process is illustrated in | MUZzN od in place of Eq, (2-9), since S, is a disjoint subgraph, The expansion igure 2.8, Sus AynSa8y (@) The union and intersection of $* and S;, Fig. 2.7 An expansion process for reforming S, Ss Ss S ee eae (b) Operation ofthe first step. ‘The DKI for this truss is obtained in a similar manner using Eq. (2-1) for calculating the DKI of subgraphs and their intersections, and considering the support conditions, we obtain: . WS) =2NG)-2= 10-28, (S)=2NGS.)=8, . (Ss) = 2N(Ss)~ 10- 1 (As) = 2x2 = 4, and (Ag) = 22 S'eS=S'US; AS=S'08) Therefore (©) Operation of the second step. NS)=(8)}H8)—(4)= 12 and H(S=M8) = (12}49) - 4) = 17. Fig. 2.8 A different decomposition of . For S as the graph model ofa rgid-jonted fame, Eqs (2-5) and (2-6) are used and ‘When § is viewed as @ truss, using appropriate coefficients from Table 2.1, using the following results are obtained Bq, (2-8) results in For the plane frame 8) =30, (8) =27, LS") - 3] = [1-32] + [- 2-31] - [4-34], and for the plane truss (8) = 2, (S)= 17, or 4S)= ‘An expansion process using different decomposition ofS is also considered. In this | Similarly forthe second step, ‘ease, Sis decomposed into three subgraphs, as shown in Figure 2.8(a), and Eq. (2~ (WS) —3]= [1-341] + 10~3x1] ~ f+ 446-3],32 ‘Structural Mechanics: Graph and Matrix Methods or: yS)=18)=2. In order to calculate the DKI, Eq, (2-7) is used for determining the DKIs of the subgraphs and intersections (with appropriate coefficients from Table 2.1), and the same results asthe previous ones are obtained. Example: Let S be the graph model of a space structure. This graph can be considered as 27 subgraphs S; as shown in Figure 2.9(a), connected to each other to 27 form a graph S= US;. The interfaces of S, |o127) are shown in Figure 2.9(6), in which some of the members are omitted for the sake of clarity ‘The expansion process consists of joining 27 subgraphs S, one at time, In thi process, the selected subgraphs can have three different types of intersections, ‘which are shown in Figure 2.10(a), ‘In order to simplify the counting and the recognition ofthe types of interface, re-formed storey by storey. For the first storey, a 3%3 table is constructed to show the types of intersection occurring in the process of expansion. The numbers on each box designate the type of intersection, Figure 2.10(b). Similar tables are used for the second storey and the third storey of S, Figure 2.10(b). 7 @) AsubgaphSiofs, (6) S= J S, without some of its members, Fig. 2.9 A space structure 8. CHAPTER 2 Topological Properties of Skeletal Structures 33 Thus thee exist 6 intersections of type A, 12 intersections of type A? and & intersections of ype AP htt ‘Type 1-At Type 2~A? (@) Three different types of Kio Boy Second floor oy First floor ‘Thied floor (0) Types of intersection afer the completion ofeach storey. Fig. 10 Intersections and their types. Since each S; is @ connected subgraph, and in the process of expansion, §; is kept connected, a simplified Eg. (2-8) can be employed: zu 2 vS)= Divs) - DHA).3 Structural Mechanics: Graph and Matrix Methods {As has been shown: Ge x DT WA) = DWAD+ LAN + VA). a tm A typical S, can be considered as, Si=KU Ly ‘where Ky s the [skeleton ofa filly triangulated cube and L; is a star connecting the ‘middle node F to the corner nodes of the cube, Figure 2.11. Thus Wi) = VK) + VL) VK OL)» @ ) © Fig. 2.11 Decomposition of S=K,U Ly Similarly, A? and A} can be decomposed as: AP=AlUAL and A} = APUAL ‘When $ is viewed as a ball-jointed truss the DSI of S ean be calculated as follows: 4S) = OH 3946) ~ (0 3x 816) = 45. A simple reason for 1K.) = 0 willbe given in Section 2.5. Y(AL) = M~3N46 =8~3x 546 ~ 1, CHAPTER 2 Topological Properties of Skeletal Structures 55 F(A) = CDH (1-246) = F(AN)= CDH -3) ~ @-3x316) == 3. Hence: US) = 27(5) = [6(-1)+12(-3)+8(-3)] = 201, When $ is taken as a space rigitjointed frame, then 7(S) ~ 3564 is obtained. Similar calculations are performed for determining the DKI of S, and 1(S) = 591 and 1(S) ~ 1188 are obtained for ball jointed truss and space fiame, respectively. ‘The union-intersection method becomes very efficient for structures with repeated patterns. Counting is reduced considerably by this method. As an example, the use fof the classical formula for finding the DSI of $ in the above example requires counting 792 members and 199 nodes, which is not an easy task with a probable ‘mistake in the process of counting 2.3.5. MODIFICATIONS ON A STRUCTURE. For design of a structure, itis sometimes necessary to add or delete a node, # ‘member or a substructure, The effect of such operations on the DKI or DSI of a structure ean easily be found using the intersection theorem. Example: Suppose a member m, is added in between two existing nodes n, and n, of S, as shown in Figure 2.12; ie, S, = S Uma, From Eq. (2-9): YSI=VS) +m) - TSOM) = 3x6+ 0-30-21) =21. Similarly, ifm, is deleted fiom S, ie. S;= S~m then, 8) = 15) + 4m) — 7S.) or 8) = 15) — Hem) + 7G. Om) = 36 — (0) + 0-241) =15.56 ‘Structural Mechanies: Graph ad Matrix Methods @s Os. os, Fig. 2.12. Modifications on a planar frame ‘The idea is more general and the effect of adding or deleting a substructure can easily be determined. As an example, considera planar truss S, as shown in Figure 2.13(a). Ifa substructure, as illustrated in Figure 2.13(b), is deleted from S, then the reduced S, (Figure 2.13(¢)) is obtained, for which #(S,) can be calculated using Eq, 2-9) as: ¥S)= HS) +18) ~ 7,989. Hence: (8)- 18) FG. 8) 2-10) = (2) -( 2) + (0- 21043) = 17, (0) Main truss, (©) Substructure Su, (e) Reduced truss §,. Fig. 2.13 Modifications on a planar truss. A similar process can be used for determining the DKI for a modified structure, 24 IDENTIFICATION METHOD ‘This approach is also based on the union-intersection method and provides a simple ‘means for finding the topological properties of a structural model S, after identifying two subgraphs of S. For example, consider a model as shown in Figure 2.14(a) Identifying ab with ed as in Figure 2.14(b), results in a cylindrical space CHAPTER 2 Topological Properties of Skeletal Structures 37 graph S*, Identification of ac and bd, leads to a toruslike skeletal structure as depicted in Figure 2.14(0), ‘Sin developed form Sins (a), Srand S' 98, @)S%, sand? 9S,58 Structural Mechanics: Graph and Matrix Methods (6) A torus-ke structure S Fig. 2.14 Identifications in a structure. ‘The following equation can be employed in such an approach, which is similar to Eq, (2-9) ofthe previous seation, VS'US) = VS) + VG) - TIS'S). etn In this relation, however, S, isa subgraph of ' through which the identification has bbeen made, Obviously S'S, consists of two disjoint, Example: Let §; be the graph model of a rigid-jointed frame, Figure 2.14(a), The first Betti number of S? which is the 1-skeloton of a torus, is obtained by two identifications, as shown in Figures 2.14(b) and (c). In the case v(S) = by(S) = M(S) ~ N(S) + by(S), for the frst identification: bYS=72 bi) =0, by(S'9S)= 12-1441 =~ 1, biS! US) = bys! (7)+0)-(- = 73. For the second identification: biS)=73, — bSD=1, Buss) =12- 12+ buS? US) =bS})= (7341) ~ CH) = 73, CHAPTER 2 Topological Properties of Skeletal Structures 9 When $ is viewed as a space frame, the DSI can easily be found as 4(S) ~ 6b\(S) = 6x73 = 438, For $ viewed as a ball-jointed truss, the DSI can be determined as follows: For the firs identification: 1 !) = M—3N46= 120 ~ 3x94 4S) = = 3796 TOSS = 12-3x14+6=—24, 48! US) = 18) = 2-9) - 24 For the second identification: WS)= YS? AS) = 12 - 3K1246 = - 18, 6, 8) = 6-3x646=~6, WSUS) = 1S) = 6 HH = 6)—C 18)=6. 2.8 THE DSI OF STRUCTURES: SPECIAL METHODS, In this section, using Euler's polyhedra formula (Section 1.8.1), some useful theorems are proven, which provide simple means for calculating the DSI of ‘various types of skeletal structures ‘Theorem 1; For a fully triangulated planar truss (except the exterior boundary), the internal DSI isthe same as the number ofits internal nodes: 8) = NO). 12) Proof; Draw S in the plane, According to Euler's formula, R(S)-M(S) + N(S)=2, 13) where R(S) = RAS) + 1 is the total number of regions and R(S) is the number of ‘wiangles. Since all the intemal regions are triangles, 3R(S)=2M(S) - MAS), ey)60 Structural Mechanies: Graph and Mattix Methods MS) being the number of members in the exterior boundary of S which may be any polygon. Substituting, MAS) = NAS) = NS) ~ NS), 45) in Eq, (2-14) leads to: 3R(S) = 2M(S) ~ NUS) = 2M(S) - N(S) + NYS). (2-16) From Eg, (2-13), we have: RYS)=M(S) ~ NS) +1. en ‘Substituting from Eq, (2-17) in Bq, (2-16) yields: M(S)-2N(S) + NAS). Thus: 9)=NG). For trusses with nontriangulte internal regions (Figure 2.15(a) et M(S) be the number of members required for competion of the triangulation of the internal region ofS, thes 48) -NiS)- Md). 18) ‘The number of members required for triangulation ofa polygon is constant and independent of the way it i tangulated, This is why Eq. 2-18) can easily be established ‘Once the internal DSI of a structure is found, the extemal DSI resulting from the additional supports can easily be added, to obtain the total DSI. Example: For the truss shown in Figure 2.13(a), the application of Eq. (2-12) results in; YS)=N)=32. ‘The use ofthe classical formula (Eq. (2-2)) leads tothe same result, 3) = 129 ~2x50-+3 = 32, but it necessitates counting 129 members and 50 nodes, compared with counting 32 internal nodes, CHAPTER 2 Topological Properties of Skeletal Structures a Example: Let § be a planar truss as shown in Figure 2.15. Triangulation of the internal region in an arbitrary manner requires 9 members, shown as dashed lines. Thus 4S) =N¢S) - M,(S) = 16-9 (b) Triangulated S. Fig. 2.15 A general planar truss and its triangulation, (@) A planar truss 8. ‘Theorem 2; The DSI of a planar rigid-jointed frame S is equal to three times the ‘number ofits internal regions, 8) =3R(S). 19 Proof: A tree structure is statically determinate, For each addition of a chord the indeterminacy of the structure inereases by 3, and therefore: 1) LM(S) -N(S) + 1] ‘Using Eq, (2-17) completes the proof, ie. 5) = IRS). Example; ‘The DSI of the rigid-jointed plane frame shown in Figure 2.16(a) is obtained as, (8)= IRS) ~ 3x16 = 48, and forthe rigid-jointed arch-type structure of Figure 2.16(b), we have: 8) = 3x6= 18. Obviously, the classical formula (Eq, (2-6)) requires more counting than the above method.a Structural Mechanies: Graph and Matrix Methods KEES (@) A planar frame, (©) A planar arch-type structure. Fig. 2.16 Rigid-jointed structures. Inter-Relation of Theorem 1 and ‘Theorem 2: Theorem 1 and Theorem 2 are closely inter-telated. Let 8 be a planar rgid-jointed fame and S’ bea planar truss, obtained from S by introducing (deg (n) ~ 1] releases in each node m, of S, where deg (n) is the degree ofthe node n, Then, +8) = 3M(S) -3N(S) +3, YS)=48)—, (220) ‘where ris the total number of releases given by: No r= Yideg(a)=1). @at Using the result oF Seetion 1.2.2, r= 2M(S) = NO). Substituting rin Eq, (2-20), and using Eq, 2-19), leads to, (81) = 3IM(S) = N(S) + 1] = [2M(S) - N(S)] = M(S)~ 2N(S) +3, ‘Which is the familiar classieal formula for a planar truss. ‘Now, fora triangulated model, by Theorem 2: 4S) =3R(S). ‘Using Eq, (2-14) leads to: CHAPTER 2 Topological Properties of Skeletal Structures 6 (8) = MIS) - MAS), (On the other hand, YS) = 108) = 2M(S) ~ N(S)] = [2M(S) ~ MAS)] = BMS) ~ NS) = N(S)~ MAS) = NS) = N&S)= NAS), ‘which is the same result ns that of Theorem 1 Using a reverse process, by introducing constraints in place of releases, a truss model can be changed to a frame model S, and Eq, (2-19) can be derived from Bq, (2-12). ‘Theorem 3: A ball-jointed space truss dravn (embedded) on a sphere is internally satically determinate, if all the created regions are triangles. Proof: Since each triangular region has three members and each member is ‘contained in two regions, therefore: 3R(S) = 2M) (2-22) ‘Combining this equation with Eq, (2-13) yields, M(S)—3N(S) +6 = 4S) = which completes the proof. ‘As an example, a ball-ointed truss with K, of Figure 2.11(b) as its graph model, is statically determinate. ‘Theorem 3 can easily be generalized to graphs embedded on other surfaces such as a sphere with h handles, where Euler's formula becomes: R(S)-M(S) + N(S)=2— 2h, 223) “The substitution of Eq, (2-22) yields: M(S)~ 3N(S) + 6= 6h. Thus: 48)= 6h.64 Structural Mechanies: Graph and Matrix Methods ‘As an example, the DSI ofa triangulated space truss in the form of a torus (sphere ‘with one handle), shown in Figure 2.17, becomes (S) = 6x1 = 6. Fig. 2.17 The L-skeleton of a triangulated torus 2.6 SPACE STRUCTURES AND THEIR PLANAR DRAWINGS In this section methods are developed for transforming the topological properties of space structures into those of their planar drawings, thus simplifying the counting, process for the calculation of the DSI for space structures. 2.6.1 ADMISSIBLE DRAWING OF A SPACE STRUCTURE A drawing $° of a graph S in the plane is a mapping of the nodes of S to distinct points of S*, and the members of'S to open ares of S® such that: (i) the image of no memier contains that of any node; (i) the image of a member (n,n) joins the points corresponding to n, and ny. ‘A drawing is called admissible (good) ifthe members are such tha: (Gi) no two ares with a common end point meet; (iv) no two ares meetin more than one point; (¥) ao three ares meetin a common point. ‘The configurations probi sd by these three conditions are shown in Figure 2.18, CHAPTER 2 Topological Properties of Skeletal Structures 6s Fig. 2.18 The prohibited configurations. A point of interseétion of two members in a drawing is called a crossing, and the crossing number e(S°) of a graph $ is the number of erossings in any admissible drawing of $ in the plane, An optimal drawing in a given surface is one which exhibits the least possible crossings. In this book we will use only admissible 33) as: WS) =0x146-(643)= For S being a space frame, the DSI is calculated using Eq, (2-31) as: y(S)= 6 [166-21] =48, Fig. 2.27 A planar drawing ofa space structure 2.7 SUBOPTIMAL DRAWI (G OF A SPACE STRUCTURE 2.7.1 INTRODUCTION For a general graph, there is no known formula by which the crossing number can be calculated. There is also no algorithm by which an optimal drawing can be objained. The only partial results available are for special graphs like complete graphs, bipartite graphs and cubic graphs, For these graphs only upper and lower bounds for crossing numbers are available. In fact itis proven that for given graph ‘Sand an integer k, the question o(S°) [1k is NP-complete (soe Section 3.4 for the definition), Gary and Johnson [55]. Therefore, an approach for estimating crossing rhumbers, and the exact value of e(S") should be restricted to special cases. AS an example, for complete graphs, some results are available in Refs [43,213,246] CHAPTER 2 Topological Properties of Skeletal Structures 15 ‘Two theorems of the previous subsections provide useful lower bounds for the crossing number of space structures, and lead to the design of an efficient algorithm, for suboptimal drawings of general graph Corollary 1: The staical indeterminacy 8) ofa ball jointed space ts 8, isa lower bound tothe crossing numberof Cor ider Ba, (2-27) 05, 8) = e159) MAS", and rearrange its tems as: (8°) = 8) + MS". (234) Since MS") 2 0, therefore (8) 248). Hence 1(5) isa lower bound for (8. From Eg. 2-34) it follows that: Min e(8") © Min MAS") 235) Corollary 2: The erossing number of @ graph is related to its Betti number by: ofS") = R(S?) —b(S). 236) ‘obvious from Eq, (2-25) if its terms are rearranged, Therefore, to minimize €(S") one should minimize the number of internal regions of its planar drawing, i. Mine(S") «> Min R(S"). e317 Based on the above results, an intuitive algorithm can be designed for suboptimal drawing, However, for an automatic drawing an algebtaie graph theory method is developed [129] as presented inthe following section, 2.7.2 AUTOMATIC DRAWING OF A SPACE STRUCTURE Definitions: A. graph can efficiently be represented by its member-node list, and the adjacency matrix of the graph can easily be constructed. Consider a graph with 10 nodes and 19 members, having the following adjaceney matrix:6 Structural Mechanics: Graph and Matrix Methods 12345678910 ie roa 1 ater t 1 a| Moat 4 cig 5 ert a6 i to) 7 om et ot 8 on 9 . 10| 4 ‘where only non-zero entries are shown for clarity. When drawing this graph, the nodes are placed on a straight line and the members are placed at the top and bottom in a random manner. Members at the top and bottom of this line are identified by +1 and —1, respectively, resulting in a new matrix A*, For the graph shown in Figure 2.28, the corresponding matrix A* is constructed as: 1 2esta 3 6 78) 0) 10: 1 a + 2) et +1 3 . + 4 4 oan ard . a oy 6 . a 7 eH 8) sym ea 9] a 10 . For this drawing, the crossing number is equal to 8. This number ean also be found from A*, using the upper triangular submatrix of A*, For a row, only the right part of the diagonal entry ofthat row will be considered. CHAPTER 2 Topological Properties of Skeletal Structures ” Fig. 228 A drawing with 8 crossings. For each entry A%(i) of A*, we define a subdomain of A* called the ferritory of AMG). This territory consists of the subdomain specified by AM(Li-L, i131), denoted by SUBA*(L:i-I , i#1:j-1). The territory corresponding tothe (i)th entry contains the entries from row I to i-1, and cokumn i+! to j-I. As an example, the region corresponding to (3,10)th entry of A* contains the entries confined to rows 1 to 2 and columns 4 10 9, ie. for A%G3,10) the territory is shown by SUBA*(1:2 , 49). For each subdomain there is an associated crossing number, which is the same as ‘the number of entries having the same sign as the considered entry. As an example, for the entry (3,10) shown in A*, the number of +1s is equal to 2. Obviously if ‘A%(3,10) was =I, then the number of crossings would be equal to 3. For all non-zero entries of AY, the territory can be specified and their crossing ‘numbers can be calculated and summed up, resulting in the number of crossings of the drawing. In the above example, the member connecting 3 to 10 have crossing, points with the members I fo 8 and 2 to 9, since in this territory, both members are drawn atthe top. Present Method: For a fixed positioning of the nodes on a straight line, the number of members (entries) is constant, however, the number of erossings should be minimized [27,123], This can be achieved by reducing the erossing numbers of the territories. As an example, in matrix A*, the entries (2,9) and (1,8) can be changed to —1, however, this may have a reverse effect on the number of crossings ote other teitores, hence increasing the number of erossings of the drawing. ‘Now we associate consecutive numbers with each non-zero entry of A®, to obtain matrix B as follows8B Structural Mechanics: Graph and Matrix Methods Lf ea 3 Ole 6 3 . foe. 4 -i ou BS / pRB ow (2-40) 6 +15 16 7 ou 8} ym 19 9 : 10] : ‘The numbers 1 to 19 are associated with these entries for 19 members. For the previous example, the teritory of node 9 will contain numbers 1,2,3,5 and 6, Now ‘we define a virtual graph whose nodes are the same as the number of members of the original graph and two nodes n, and n, are connected ifn isin the subdomain of 1, For the considered graph, the node 9 is connected to the nodes 1,2.3,5 and 6. ‘Other members are formed considering the other territories ‘Since two types of numbers are used in the territory, ie, +1 and 1, the members ‘can be identified using two colours. Obviously for the virtual graph G,, ifa member hhas two nodes of the same colour, this will be equivalent toa crossing, [Now an optimal planar drawing (embedding) can be constructed as follows: Colour the nodes of G, in such a manner that no two adjacent nodes have the same colour. Obviously this is possible only if 4 colours are available (Ref. [183]. However, if'@ graph can be coloured with to colours only, it is equivalent to a planar drawing with no crossing. It can be shown that only bipartite graphs with no ‘odd cycles are two-colourable, However, ifa graph is not two-colourable, atleast ‘one should reduce the number of members with the nodes of the same colour. This is equivalent to having a pseudo-bipartte graph, each part of which contains a limited number of members. Using the adjaceney matrix A of G,, one can decompose the graph into two subgraphs with close properties to a bipartite graph, ice. two subgraphs with maximal interface. Methods for bisection with minimum interface use the Fiedler vector of the Laplacian of a graph (101,218]. The connection between two subgraphs can be 1 represented as C corresponds to Ay of L, since 2,0, and Ay corresponds to maximal C. It should be x'Lx using Rayleigh's mi mal principle. Minimal C CHAPTER 2 Topological Properties of Skeletal Structures p noted that for some graphs, A produces better results than 2a, and similarly in ‘maximal case, 2... might perform better than 2. This is observed for planarization of Ki, After calculating Aya OF Ay, the corresponding eigenvectors can be found. Here instead of ordering and partitioning as in 6, the positive or negative nature of is considered and replaced by +1 and ~1. These values specify the members of graphs fon the top and bottom of the straight line, respectively. Once A* is formed, the ‘number of crossings can easly be evaluated, ‘The problem of bisection can be formulated in the form of a maximal-flow or Minimum-cut problem, Lawler [157,32], and there are other approximate algorithms concerning this problem, Ref. [73] ‘The method presented inthis scction is designed to reduce the crossings for a given position ofthe nodes. Any change in the position of the nodes may alter the number ‘of crossings. It seems logical to locate the odes in such a manner that the corresponding subdomains be localized, ic. correspond to smaller length and width for subdomains. Reducing the bandwidth or profile of A may help such localization. This can be achieved by ordering using 2 of the Laplacian for the original graph G, The results ate shown in Figure 2.29, where the number of crossings is reduced fo 4, Ref. [68]. However, one may find locations not corresponding to minimal profile or bandwidth, and still lending to @ smaller ‘number of erossings, asin Figure 2.30, corresponding to the erossing number equal {0 2. The matrices for these cases are shown inthe following: ea ol ee Ae @-41) sym e4d ‘Matrix A® corresponding to Figure 2.29 with number of crossings equal to 4.80 Structural Mechanics: Grapla and Matrix Methods led ee | (2-42) 3 eo 4 ea 1 reg et 6 en 7 . 4 8] sym il 9 . 1 u 7 CHAPTER 2 Topological Properties of Skeletal Structures a It can be seen that the matrix in Figure 2.30 is not well structured, however, the corresponding crossing number has the least magnitude, Examples: For complete graphs, all the entries are present in A, and therefore the least crossing can be evaluated tmuch easier. The present algorithm is applied to Kio, Kis, and Kys, and corresponding graphs are drawn in Figures 2.31 to 2.33. The ‘number of crossings, satisfies the following known inequality 2-43) ‘where v(K) is the crossing number and K, is a complete graph containing n nodes [os]. Fig. 2.30 The drawing of A* of Eq. (2-42) with 2 crossings. Fig. 2.32. A planar drawing of Ky2 ‘Structural Mechanics: Graph and Matrix Methods Fig. 2.33. A planar drawing of Ks ‘As mentioned before, for the above examples no trial and error method is used to improve the positioning; and the present approach is merely employed. There is no general method for such positioning, and one can use a greedy type algorithm, although there is no guarantee in obtaining minimal crossings, which is an NP- ‘complete problem [55]. For structural models, the ratio ofthe numbers of members to the number of nodes, is far less than those of complete graphs. For such models, one can decompose the ‘models into an appropriate number of subgraphs having minimal interfaces; for ‘each subgraph the embedding can be performed separately. In the following, two examples of structural models are studied. The planar ‘drawing of the space frame in Figure 2.21 is illustrated in Figure 2.34. The planar embedding of the double layer grid shown in Figure 2.24 is provided in Figure 2.35; Fig. 2.34 A planar drawing of the space frame with 8 erossings. CHAPTER 2 Topological Properties of Skeletal Structures 3 38 A planar drawing ofthe space frame with 12 crossings, ‘The application of the present method is by no means confined to finding the DSI of structures. It can be extended to other systems such as electrical and hydraulic systems, printed circuit board layout, very large-scale integration citeuit routing and automated graph drawing. EXERCISES 2.1 Use an expansion process to find the DSI of a 3x4 planar truss S as shown. ‘The unit to be considered for expansion is also given. s 5, 22 If the truss in the previous example is an mxn planar truss, determine the corresponding DSI. 23. Find the DKI of the truss in Exercise 2.1 and compare it with its DSI, What do you conclude? 24 Derive Eq, (2-9) from Eq, (2-8).84 Structural Mechanics: Graph and Matrix Methods 2.5 Prove that for determining the DSI of a planar truss, the crossing point of ‘any two members can be regarded as an extra node. If the erossing members are ‘more than two, wity does such an operation become incorrect? 2.6 Determine the DKI and the DST of the following planar frames with some releases: @ © o 2.7 Find the DSI of the following planar truss using three different methods: classical, modification and triangulation 2.8 Determine the DSI of Sin Figure 2.9 using its planar drawing. Consi first asa space truss and secondly as a space frame. 8 29 Determine the DST of the following double-layer grid, Suppose $ is supported in statically determinate fashion, CHAPTER 3 Rigidity of Skeletal Structures 3.1 INTRODUCTION ‘The rigidity of structures has been studied by pioneering structural engineers such as Henneberg [79] and Miller-Breslau [176]. The methods they developed for examining the rigidity of skeletal structures are useful for the study of structures either with a small number of joints and members, or possessing, special connectivity properties. Rigid-jointed structures (frames), when supported in an appropriate form and containing no releases, are always rigid. Therefore only truss structures require to be studied for rigidity. Various types of methods have been employed for the study of rigidity; however, the main approaches are either algebraic or combinatoric. A comprehensive discussion of algebraic methods may be found in the work of Pellegrino and Calladine [190], The first combinatorial approach to the study of rigidity is due to Laman [150] who found the necessary and sufficient conditions for a graph to be rigid, when its members and nodes correspond to rigid rods (bars) and rotatable ppin-joints of a planar truss. Certain types of planar truss have been studied for rigidity by Bolker and Crapo [15], Roth and Whiteley [211] and Crapo [33] Although Laman theoretically solved the problem of rigidity for planar trusses, no Algorithm was given to check whether a given graph was rigid. Two combinatorial algorithms are developed by Lovise and Yemini [165] and Sugihara [227], the interrelation of which has been shown by Tay [231]. Some studies have recently bbeen made inthe direction of extending the developed concepts for planar trusses to those of space trusses. However, the results obtained are incomplete and applicable only to special classes of space truss (see Ref. [247] as an example). Therefore, in this chapter, only plana trusses are studied,86 ‘Structural Mechanics: Graph and Matrix Methods 3.2 DEFINITIONS Rigidity of trusses can be studied at different levels. The fist level is combinatorial ~ isthe graph of joints and members (bars) correct? The second level is geometrical + is the placement of joints appropriate? The third level is mechanical - are the selected materials and methods of construction suitable? This chapter is devoted to the first level rigidity analysis of planar trusses. For this purpose, simplifying assumptions and definitions are made as follows: Consider a planar truss composed of rigid members and pinned joints. Each joint connects the end nodes of two or more members in such a way that the mutual angles of the members can change freely if the other ends are not constrained. Such, ‘an assumption is adequate for the first level analysis of the rigidity. Let M(S) and [N(S) denote the set of members and nodes of the graph model S of a truss, Denote the Cartesian coordinates of a node nj N(S) by (x,y). The number of members land nodes of S, as before, are also denoted by M(S) and N(S), respectively. ‘A member connecting n, to nj constrains the movement of $ in such a way that the distance between these two nodes remains constant, (xj -x))° +9) -¥))? = const. G1) Differentiating this equation with respect to time t, we get, (ix) GD +OH- YG -HY =O, oa where the dot denotes the dfferetition with respect fot, Equation (3-2) implies that the relative velocity should be perpendicular to the member, that is, n0 rember is stretched or compressed. Writing all such equations fr the members of S, the following system of linear equations is obtained, Hw=0, G3) ‘where H is a M(SQN(S) constant matrix and w is a column veetor of unknown variables w- tye yyy V1 Ya ~ Frys t denoting the transpose. A veetor w which satisfies Eq. (3-3), is called an infinitesimal displacemem of S. The infinitesimal displacements of S with respect to, point-wise addition and ‘multiplication by scalars, form a linear veetor space R°™". The rigid body motion ina plane isa three-dimensional subspace of this linear space. The co-dimension of this subspace of rigid motions in the space of all infinitesimal motions is called the degree of freedom of 8, denoted by 8). The structure Sis rigid if {S) = 0 CHAPTER 3 Rigidity of Skeletal Structures 87 ‘As an example, consider a truss as shown in Figure 3.1, For this truss, matrix 1 land veetor w can be written as: xox} o yim yimm 8 Helx-m 0 x8) ime Oe Oommen) OYE MEY) and wet aN ‘The entties of H are real and linear fanctions of the nodal coordinates of the corresponding graph. yi yt ne Fig. 3.1 A triangular planar truss It should not be thought that the rigidity of S requires Eq. (3-3) to have only the trivial solution w = 0. Any rigid body motion with non-tivial w will also satisfy ‘this equation, As an example, consider a translation of the entre S specified by a vector {a,b}', ie, Xj =%j=%; =a and J) =5j= Hy =b. Obviously Hw = 0 still holds, since the sum of the first (or second) three columns of H is zero. Therefore rank (H) <2N(S). The rigid body motion subspace in plane has dimension 3, and for any truss we have rank (H) < 2N(S) — 3. However, if rank (H) = 2N(S) ~ 3, then § is called rigid and for rank (H) < 2N(S) — 3, itis non-rigid. In the above ‘example rank (H) = 23 — 3 = 3 holds, and therefore a triangular planar truss is rigid, Now consider other examples as shown in Figure 3.2. The truss shown in Figute 3.2(a) is rigid, while the one in Figure 3.2(b) isnot rigid, although their underlying graphs are the same, The assignment of velocities, indicated by arrows, forms an intnitsimal displacement beeause it does not violate Eq, 3-3). The nodes without aows are assumed to have zero velocities. Similary, though Figure 3.2(6)88 ‘Structural Mechanics: Graph and Matrix Methods and Figure 3.2(d) have the same graph models, (c) is rigid but (4) is not rigid. It should be noted that an infinitesimal displacement does not always correspond t0 ‘an actual movement of a mechanism. The truss (b) deforms mechanically, while truss (4) violates only Eq, (3-3), Buy @ Fig. © @ 12. Rigid and non-rigid planar trusses. The nodes ofa structure $ are in gener position if xy YR Yo Rg, Yas a algebraically independent over the rational field. When the nodes are in general Position, the definition of algebraic dependence shows that a subdeterminant of atric H is 0, if and only if itis identically 0, when Xiy Yj. «Xen Yun ME considered as variables, Therefore i the nodes of $ are in gencral postion, the linear independence of the Ea. (3-3) depends only onthe underlying graph, and consequently the rigidity also depends only on the graph model of the stuctue. From now on itis assumed that the nodes ofS are in general position, For a bal-jointed space truss, Eq, (3-3) can be written in a general form to include 4 third dimension z. For such a ease, a rigid body motion in space is a six dimensional subspace of R™®), Therefore a space truss will be rigid if rank (H) = 3N(S) = 6 and non-rigid if rank (H) < 3N(S) - 6. ‘Suppose S isthe graph model ofa planar truss whose joints are in general position. A graph $ is called stiff if the corresponding truss is rigid, For any X cM, let s(X) be the rank of submatrix of H_ consisting of the rows associated with the members of X. X is called generically independent if ps(X) = |X, and generically dependent if ps(X) <|X|. [X| denotes the cardinality of X For any subset X of M(S), define, 1309) M(x) +2N(X)~3, G4) ‘where [M(X)| =[X), Then the following basic theorem on rigidity can be stated: ‘Theorem 1 (Laman [150}): The graph S is generically independent if and only if Hs(X) 2 0 for any non-empty subset X of M(S). CHAPTER 3 Rigidity of Skeletal Structures 89 Corollary 1S is stiff i and only if there exists M' = M(S) sueh that [M"| = 2N(S) ~3 and p(X) 2 0 for every non-empty subset X of M’ Corollary 2: $ is stiff and generically independent, ifand only if (@)—s(M)=0 and (©) ¥s(X) 2 0 for every non-empty subset X of MS). Using (8) = M(S) ~ 2N(S) +3 =~ (5), Theorem 1 can be restated as follows: ‘The graph S is genetically independent if and only if (S) $0 for every subgraph S, of S, Figure 3.3(a). The graph S is stiff if and only if there is @ covering subgraph 5 of S such that(S) = 0 and (S) $0 for every non-empty subgraph S, of 5, Figuse 3.3(b), which is a statically indeterminate structure nally, the graph S is stifPand generically independent, if and only if (@) 48)=0 and (b) WS) <0 for every subgraph 8, of S, Figure 3.3(€), which is a statically determinate truss Unfortunately the application of Theorem 1 requites 2" steps to determine ‘whether a graph is generieslly independent. In the next sections, two methods are described for an efficient recognition of generic independence, which were developed by Sugihara, and Lovasz and Yemini VN EAN LN @s ()S=SUm © Fig. 3.3. Generically independent, stiff, generically independent and stiff graphs. 3.3. MATCHING FOR THE RECOGNITION (OF GENERIC INDEPENDENCE Planar trusses ae frequently used in structural engineering and therefore a method is presented in this scction for checking the rigidity of these structures, which is suitable for both determinate and indeterminate trusses. The algorithm is a90 Structural Mechanies: Graph and Matrix Methods polynomial bounded one and uses complete matching of a specially constructed bipartite graph for the recognition of generic independence, Definitions: Let B(S) = (A,E,B) be a bipartite graph with node sets A, B and ‘member set E. A subset E" of F is called a complete matching with respect to A if the end nodes of members in E are distinet and if every node in A is an end node of one member in E’ For X & A, let CX) be the set of all those nodes in B that are connected to nodes in X by some members of E. As hnas been described in Chapter 1, @ graph has complete matching, if and only if [X| < [P| for every X & A. Examples of ‘matching and complete matching are depicted in Figures 3.4(a) and (b), respectively. Using these definitions, the method is described as follows: (2) A matching, (b) A complete matching, Fig. 34 Examples of matching, Construct bipartite graph ofS. For this purpose let S be @ graph with N(S) nodes ‘and M(S) members. The corresponding node set and member set are shown with the same symbols. For each node n, of S let p; and q; be two distinet symbols. Then let BUS) = (A*,E*.B*) be the bipartite graph whose node sets A* and B® and ‘member set E* are defined as: A*=M(S), BY {191 Pas Pris, ds} E*= ((mp)smg).(m.p)m.q) |m = (n,0} € M(S)). ‘This bipartite graph is now augmented as follows: Lett, te and & be three distinct symbols, Then for any 1S i is added to S; in two different manners as illustrated in Figures 3.10(¢ and d). Using the intersection theorem for both cases we have, CHAPTER 3 Rigidity of Skeletal Structures 7 (SUS) =0+ED-CD=0, while (d) is rigid, (c) is not rigid, In order to show this property, (©) and (d) are decomposed in Figure 3.11. X ky KY a) Decomposition of (¢) (&) Decomposition of (€)_(c) Another decomposition of (4). Fig. 3.11 Rigidity control by Theorem 4 ‘Now for decomposition of Figure 3.11(2) we have, (@xt-3)+ @d-9225-3, wwhih is false and therefore (cis not rigid. For (@) we have (x53) + B2- 225-3, Which is true, Any other decomposition leads to a true inequality and hence (d) is raid. For example, for the decomposition of Figure 3.11(c) one obtains, (24-3) + (Qx2=3) + OX2-3) § 2xS—3), Which is also true. Example: Consider S; and S; as shown in Figures 3.12(a) and 3.12(b), respectively. For the decompositions shown in (d) we have, (2x4 = 3) + (2x2 = 3) + (2-3) + 2-3) 2 2X6~3), is false, and the subgraph S, U S; is not rigid, though 9( S: US) KO KORO) @S HS (SUS) (@) Decomposition of () Fig. 3.12, Rigidity control by Theorem 4 whi ‘The above two examples suggest the use of a simple method for possible detection of the violation of the rigidity condition, which consists of decomposing S'" into its members and using Theorem 4, If a member has both ends in Sit should be98 Structural Mechanics: Graph and Matrix Methods ‘added to S', since this does not inerease 2N(S") ~ 3, while decreasing the L.H.S. of Eq, (3-8) to become smaller than its R.H.S. Although such a simple rule makes the quick detection of rigidity feasible, for a ‘russ with a complex pattern one should employ a general method similar to those presented in Sections 3.3 and 3.4 3.5.4 GRID-FORM TRUSSES WITH BRACINGS Consider planar truss consisting of square panels with of without diagonal ‘members, an example of which is shown in Figure 3.13(a). The bipartite graph B(S) of Sis a graph which is constructed as follows: Associate one node with each row of squares and denote them bY fi, 3, 5 Ths With each column of squares, associate ane node and denote them by €1,€2,». Cy as depicted in Figure 3.13(6). Connect r, to ¢; if the comesponding squares in $ hhave a diagonal member. The graph obtained in this manner is called the bipartite graph B(S) ofS. in ct mp 2 1 e3 cs @ ) CHAPTER 3 Rigidity of Skeletal Structures 99 It is easy to prove that S is rigid, if and only if the corresponding bipartite graph [B(S) is a connected graph. For this purpose, consider a square grid-form truss with, ‘wo rows and two columns as depicted in Figure 14(a). A series of deformations ‘ean now be performed as shown in Figure 3.14(b). He ER @ © Fig. 3.14 A squate grid-form truss and its deformation components. Obviously, an arbitrary deformation of the truss can be considered as a combination of these deformation components, Now ifa diagonal member is added fo one of the squares, say the square corresponding 10 1 and cj, then the deformation still takes place; however, a constraint inthe form of x\=y is imposed. Ifa sufficient number of diagonal members are added, then x; 2 and no square will deform relative to the squares, Le. the entire truss will be rigid (itis properly supported). This argument holds for any square grid form truss with m rows and n. columns. Since the nodes of B(S) correspond t0 X),X:y Xan Yie Yon Ym and the adjacency of r, and c, in B(S) corresponds to the equality of xx, therefore the result follows If B(S) is spanning tee, then the corresponding S is generically independent and stiff I cam also be proven that the statical indeterminacy of S is the same as the first Betti number of B(S), Kaveh [92]. Figure 3.15(a) shows a rigid yee for a planar truss, the corresponding bipartite graph of which is illustrated in Figure 3.15(). cl et ot ot n = 2 @s. (6) BOS). Fig. 3.15 A tree and its tree bipartite graph,100 ‘Structural Mechanics: Graph and Matrix Methods 3.6 Henneberg Sequence for Examining the Rigidity of Trusses ‘The following algorithm uses the Henneberg sequence to examine the rigidity of a tziven determinate planar truss, First it deletes all the nodes of degree 2 from the raph mode! of the truss, sequentially. Then a node i of degree 3, which is ‘connected to nodes a, b and ¢, next, the incident members are omitted and a new ‘member is added to the model between one of the pairs (a,b), (bye) or (a), if that pair is not already connected in the previous model. Such a pair always exits otherwise the truss would not have been a deferminate structure, The above two ‘operations are collectively called the Henmeberg sequence. The following algorithm uses the Henneberg sequence until a single member is obtained, indicating the rigidity of the truss. The following theorem provides the proof for the existence ofa node of degree 3 in a statically determinate planar truss: ‘Theorem [89]: For a statically determinate planar truss, there is at least one node ‘of degree 2 or 3 for which Henneberg sequence can be applied, Proof: Let the average degree of the nodes of S be defined as, Ss Vi__ 2M(S) ANG NO" where Vj is the degree of the node i, For a statically determinate planar truss, the degree of static indeterminacy 3(8) = M(S) — 2N(S) +3 is equal to zero, i.e. M(S) = 2N(S) ~3, and 4N(S)-6 NGS) ‘Thus there should be a node j with V, = 2 or 3 for which the Henneberg sequence cean be applied, CHAPTER 3 Rigidity of Skeletal Structures 101 Algorithm: ‘Step 1: Number the nodes and members of the graph model of the truss, separately. ‘Step 2: While there is a node of degree 2, delete such nodes and their incident ‘members, one by one. ‘Step 3: Ifa node of degree 1 exists in S, go to Step 5. Step 4: Delete a node of degree 3 with its incident members and connect two end. nodes of the omitted members to make a non-repeated new member in S, and repeat steps 2-4, Ifsuch a node is not found, oto Step 6. ‘Step 5: If the remaining graph consists of a single member only, then the $ is a rigid statically determinate truss. Otherwise itis not a determinate rigid truss Step 6: End, Example: A statically determinate truss, shown in Figure 3.16 is examined for rigidity, using the above algorithm, Table 3.1 summarizes the operations performed in different steps of the Henneberg sequence algorithm. Member 9 remains atthe end, indicating the rigidity ofthe truss. Fig. 3.16 A statically determinate planar truss. It should be mentioned that the Henneberg sequence can also be applied in the form of an expansion, ie. a node with two incident members can be added to S, and a node i with 3 incident members (i), (,b) and (ie) ean also be replaced by a ‘member (a,b), (a6) or (bx) ofS. ‘The Henneberg sequence in the form of a mixed expansion and contraction (Collapse), provides a suitable means to perform Miller Breslau’s bar exchange ‘method, which is occasionally mistakenly associated with Henneberg in the literature. Sueh operations can also be used in the force method of structural analysis [89].102 ‘Structural Mechanies: Graph and Matrix Methods ‘Table 3.1. Operations inthe Henneberg sequence algorithm. Step | Deletod node | Deleted members | New member ia 22 78. 619 313.14 41620 0.15 521 The Boqnodes 41 ‘2N(nodes 5.9) 3.7 CONNECTIVITY AND RIGIDITY ‘The member-connecivity (8) ofa connected graph Sis the smallest number of members the removal of which disconnects S. When 4(S) =k, then the graph Sis called Kmember-connected. A similar definition can be obtained for node- connectivity x(S) by eplacing "members" with "nodes" ‘Attempts have been made to relate the connectivity of a graph to its rigidity. Some partial results have been obtained; however, no general approach is found for such an infertelation, Some of the results obtained by Ref. (165) are outlined in the following ‘Theorem 5: Every 6-connected graph is stiff ‘As an example, the graph shown in Figure 3.17 is stiff; however, deleting four ‘members makes it non-stifl, It can be proven that a S-connected graph is not necessarily sift. As an example, for the graph of Figure 3.18, the rigidity canbe checked as follows: (CHAPTER 3 Rigidity of Skeletal Structures 103 Fig. 3.17 A stiff 6-connected graph, Decompose $ into 8 complete subgraphs Ks and leave the rest of the members as single subgraphs. The application of Eq. 3-8) results in, (2X5 ~ 3) + 20(2x2 ~ 3) 2 2x40 ~ 3, ie. 76277, ‘which is false and therefore S in not rig. 18 A non-stif'S-connected graph, Finally it should be noted that many attempts have recently been made to extend these ideas to space trusses; however, thus far, no concrete result applicable 10 general space graphs has been obtained. Many open problems remain for further research, if pure graph-theorctical methods are to be developed forthe recognition of the rigidity of space trusses. The theory of matroids, which is biiefly introduced in Chapter 9 of this book, seems fo be a promising tool forthe study of rigidity.104 Structural Mechanics: Graph and Matrix Methods EXERCISES, 3.1 Use Sugihara’s matching method to verify the genetic independence of the following truss models: /\ ZN @ ) 3.2. Employ Lovisz and Yemini's decomposition method to verify the generic independence ofthe following truss: ZAIN 3.3 Why is a simple truss statically determinate and how can its rigidity be ‘guaranteed? 3.4 Examine the rigidity ofthe following trusses: CHAPTER 3 Rigidity of Skeletal Structures 105 3.5. If two rigid planar trusses are connected to each other at two nodes, why is the resulting truss rigid?CHAPTER 4 Network Formulation of Structural Analysis 4.1. INTRODUCTION Graph theoretical concepts have been widely employed for the analysis of networks in the field of electrical engineering. Kirchhoff was the first to establish the basis for the analysis of networks, as early as 1874, In introducing the concepts of circulating loop currents, independent loops (cycles, circuits) and methods for Finding branch curtents in’a network, Kirchhoft [139] employed a number of topological ideas long before graph theory became an established branch of ‘mathematics Maxwell developed both the node method and the mesh method which are described in his famous book on Electricity and Magnetism of 1892, Ku [147] studied the rules of Maxwell and Kirchhoff in the light ofthe principle of duality Synge [229] investigated the matrix theory of electrical networks. Further applications of graph theory to the conceptual analysis of electrical networks may bbe found in Trent [239], Roth [209] and Kron [143] among others ‘The first application of topology and graph theory 10 the analysis of clastic nctworks (stuctres) seems tobe duc to Kron [148], In this paper an analogy ws sade betwoen electrical networks and elastic strvetutes. Worked examples using equivalent creuits for east stuctures ean be found in Carter and Kron [18] Ina series of papers, Kron [143] presented the method of piecewise solutions of large-scale systems, so called "diakopres". in this method a physical system is torn into an appropriate number of small subdivisions, each of which is analysed and solved separately. The partial solutions are then interconnected step by step until108 Structural Mechanics: Graph and Matrix Methods the solution of the entire system is obtained, Further consideration of topological ‘concepts for elastic structures was given by Kron [144] Langefors [151-156] established the system of equations for the analysis of indeterminate continuous frames and presented a complete topological approach 10 structural analysis, Samuelsson [214] used algebraic topology for the analysis of skeletal structures, and Wiberg [251] employed a mixed method for the analysis of ‘more general structures including continuum problems Henderson and Bickley [77] related the degree of statical indeterminacy of a rigid- jointed frame to the first Betti number of its graph model. Henderson [76], Henderson and Maunder (78] and Kaveh (95] used various embedding techniques in order to overcome the difficulties involved in the force method of structural analysis, ‘The analysis for cases where the structures are viewed as general systoms or networks, was formulated by Lind [159], Dimaggio [38], Fenves [46], Fenves and Branin [47], Spillers (222-224), Mauch and Fenves (168). Oden and Neighbors [182] employed graph theory to the non-linear analysis of structures, and Fenves and Gonzalez-Caro [48], Munro [177] and Mokhtar zadeh and Kaveh [176] used it in plastic analysis and’ design. Cassell [19] and Shinghui and Guohua [217] ‘employed graph theory in finite element analysis. Different formulations of structural analysis have been made using various ‘mathematical tools. Perhaps Russopoulos (212] was the first to achieve this goal by using tensorial approaches. Argyris and Kelsey [4] employed matrix algebra, Langefors [151-156] and Samuelsson [214] used algebraic topology, Maunder [169] employed vector spaces, and Fenves and Branin [47] used graphs and networks for the formulations of structural analysis. In this chapter a network-topological formulation of structural analysis, which in the main follows that of Fenves and Branin {47], is presented. A complete matrix formulation of displacement and force methods will be provided in subsequent chapters 42 THEORY OF NETWORKS ‘The common properties of various networks such as electrical, hydraulic and mechanical, ellow a unified treatment of their properties feasible. As an example, the theory developed for the analysis of each type should be modifiable to apply 10 the analysis of the other types. Since the history of the application of graph theory to the analysis of electrical networks is much older than that of non-clectrical networks, a formulation of the theory of structures based on network theory is | | CHAPTER 4 Network Formulation of Structural Analysis 109 justifiable. This unified treatment opens the door for the extension of advances made in network theory to structural theory and vice versa. For defining a structure, three types of property should be studied separately. ‘These properties, as already mentioned in Chapter 2, consist of. 1, Topological properties: which contain information about the interconnection of the members, provided by the graph model of a structure, 2, Geomettical properties: which specify the spatial position of each member, implied by the dimensions ofthe structural drawing. 3, Mechanical properties: which consist of the load-deflection relationships ofthe ‘members forming the entire structure Each of the above-mentioned properties is independent of the other two and each plays its own important role in characterizing the structure. However, itis easy to lose sight of these roles in the usual treatment of the structures. One of the main objectives of this chapter is to separate these properties, in order to find the common aspects of structural and network theories. It is shown that the force and displacement methods of structural analysis for rigid-jointed skeletal structures can easily be developed as the counterparts of mesh and node methods of network analysis, 4.3. BASIC CONCEPTS OF NETWORK THEORY A system which is made up of several individual members interconnected in some particular way, can be defined as a network, As a result of this interconnection, the variables specifying the behaviour ofthe individual members are inter-related. This is true regardless of whether the network is electrical, hydraulic or mechanical. In a network two mathematical structures are present: (@) An underlying topological structure called a graph, e.g, the dead wires of an electrical network, or the stress free state of the structure, (b) A superimposed algebraic structure corresponding to the live quantities of a network, e.g, electromagnetic in an electrical network, or the loaded state of a structure The corresponding mathematical structures will posses distinct properties common to the modelling of a variety of physical phenomena, The reason for the abstract {description of these properties is to expose the essential elements of all types of network problems without being restricted to some particular physical phenomenon.110 ‘Structural Mechanics: Graph and Matrix Methods 43.1 TOPOLOGICAL PROPERTIES OF NETWORKS. ‘A notwork can be modelled as a graph. The connectivity relations of a network are the topological properties of its graph model. These properties were studied in Chapter 1 and are re-introduced here using slightly different notations. Let $ be the mathematical model of a network which is taken as a directed graph, Consider B as the node-member incidence matrix of S. Since each column of B contains +1 and ~1, is rows are linearly dependent, ie. a) Selecting any particular node of $ as a datum (reference) node and assigning the ‘ordinal number N to this node, one can write: bi M 42) x Diy for a “Thus the Mh row of B can be deleted to obtain the incidence matrix B, without toss of information, since, No. E byi ij for J=1,2,...M, 43) in whi bb are the entries of matrix B. Example: Consider a directed graph as shown in Figure 4.1 for which M = 6 and N=5. Matrix B is formed as: my my mz mg msm nf +1 0 +10 0 nf0 0 41 -1 +1 0 Bensl41 -1 0 0 -1 +1 alo 0 -1 0 0 -1 0 00a 44) Equations (4-1), (4-2) and (4-3) can now be verified as an exercise, using the above matrix. CHAPTER 4 Network Formulation of Structural Analysis uw @ ee bed ® Ss ce 5 ) Fig. 4.1 A directed graph S and numbers assigned to its nodes. Consider an arbitrary tree T ofS. IF the branches of Tare frst ordered, followed by numbering its chords, then B will be partitioned into two submatrices By and B, my mp_ms_my ms mg o +1 0 41) 0 07, o os tao (4s) 4-1 0 of-t at oo -1 ola By Submatrix By is an (N=1)x(N=1) matrix, which is non-singular. Its inverse Py ean ee eee fae ae Hs ms eee eee eat on ‘The most important topological property of a graph is the orthogonality property, hich can be expressed as: BC'=0 or CBY=0. (48)nz Structural Mechanics: Graph and Matrix Methods For a fundamental eyete bass, ifthe tree members are numbered frst, followed by ‘numbering the chords, then Eq, (4-8) in patitioned form can be written as fo, wale ]-« o BiB, PrB,, (4-10) ‘These equations provide enough tools for the formulation of the network problem, ‘The connectivity properties of a graph model are all contained in these relations. AAs pointed out in Chapter 1, one can also construct and use a eutset basis incidence ‘matrix in place of cocyele basis (node)-member incidence matrix, but this is not discussed here, 4.3.2 ALGEBRAIC PROPERTIES OF NETWORKS In algebraic topology and graph theory, itis customary to assign mathematical objects such as numbers to nodes (0-simplexes), members (I-simplexes), cutsets and cycles of a graph (I-complex). This leads to an algebraic structure associated ‘with a graph, which is studied in the following: Consider a graph $ and associate @ number with each ofits nodes. These numbers are represented by an N-vector W. When Wis pre-multiplied by the transpose of the nodesmember incidence matrix of $,B', the transformation induces the assignment of some numbers to the members ofS, determined by: u=Bu, 1) In this equation w is an M-vector containing the numbers assigned to members. If ‘member m has nodes s and ta its starting and end nodes, respectively, then Eq (4- 11) yields: (12) CHAPTER 4 Network Formulation of Structural Analysis us [As an example, if @ ={2,4,6,3,5} contains the numbers assigned tothe nodes of $ in Figure 4.1(b) and B is taken from the previous example (Ea, (4-4), then Eq. (4- 12) leads to: This shows that the number assigned to a member m is simply the difference between the quantities assigned to its end nodes. Now if the Nth node of $ is taken asa datum node, then Eqs (4-11) and (4-12) will be transformed to: Lays +n). 3) i tl _ 11g, = Bayi + Baty Fa ‘This relation indicates that the member quantity uy depends on the difference between the nondatum quantities and the datum quantities. Taking these differences as an (N-1)-vector u’ , Eq, (4-13) can be re-written as: u=Biu', Con) in which B is the member-node incidence matrix. For the previous example, if node 5 is taken as the datum node, then u will be given as w= (2-5, 45,6 -$}'= (-3,-1,1,-2)! and B is obtained by deleting row 5 of B. Thon Eq, (4-14 results in,14 ‘Structural Mechanies: Graph and Matrix Methods tee contains he sane numbers as those assigne previously to the member of ‘The vector wu is now pre-multiplied by the eycle basis incidence matrix C of S, "Naturally this operation will assign numbers to the elements of the considered eyele basis ofS However, hecnse of the ohogonality a the eoxjee bass and the cycle basis of a graph: aed _ : cB 0. ‘Thus the eyele quantities obtained in this way are all zeros. This is obvious since: Cu=CB'u' = Ou'=0, 15) In Figure 4.2(a, for the cycle basis of S consisting of its regional eyeles, we have, cu[? -1 0 +1 41 0 Orie o 1 i|d ’ ooh \which verifies Eq, (4-15), jer], ® 6 8 1 6 oy OO) Fig. 4.2. Numbers assigned to the eycles and! members of , CHAPTER 4 Network Formulation of Structural Analysis us For a member, apart from the number obtained by Eq, (4-11) a8 uy one can also assign an asbitrary number U,, Collecting these values in an M-vector and pre- ‘multiplying by C, leads to non-rero quantities assigned to the cycles, ie. cu= Z 16) For the previous example, consider U = {6,10,15,5,12,8)', as shown in Figure 4.2(b). Then Eq. (4-16) leads to 1045412] [7 is-12-8 | |-s Adding the derived quantity u of a member to the arbitrary quantity U assigned to it result i V-U+0, 17) and Bgs (4-15) and (416) lead to cv-u' 18) ‘The relationships obtained up to the present are summarized as shown in Figure 43, 5g (0 Fig. 4.3. Transformation diagram of Type I variables. Now a dual set of relations can be obtained by assigning an arbitrary number to teach eyele of the graph S. These numbers can be collected in a vector p’ of length (dimension) b(S) = M ~N-+ 1 for a connected S. The transformation Cp’ induces the assignment of a related set of numbers to the members of S according to: p-Cp. (419) ‘These member quantities induce quantities in the nodes of S by the following transformation: ip =BC'p'= 0p'=0. (4-20)16 ‘Structural Mechanics: Graph and Matrix Methods ‘Again, due to the orthogonality property (BC! = 0), the induced quantities in the nodes are zero, It is possible to assign a certain arbitrary number P; to each member, in addition to the derived member quantity p,. Then the transformation BP will not lead to z2r0 uantities for the nodes, i, BP B 2p in which P’ is an (N ~ 1)-vector. Summing the derived and arbitrary quantities for the members, we have: R=P+p, (422) The above relations and transformations are summarized in Figure 4.4 ate B Fig. 44 Transformation diagram of Type Il variables. At this point, the purely topological aspects of the algebraic structure that characterizes the networks have been described. Two dual sets of variables U and U'on one hand and p’, p, P and P’ on the other hand, have been independently studied. However, a distinct characteristic of a network problem is the existence of a relationship between the two sets of variables, expressed as, Vem, 423) on U+u=fP4p) 424) The inverse relation also exiss and can be written as, R=4y, (428) 0: P+p=K(U+u) 426) In the above relations, f and k are MxM matrices, whose entries are real and complex numbers; fis the reverse of k, ie, fk 27) CHAPTER 4 Network Formulation of Structural Analysis 7 Addition of the above transformation naturally reduces the types of mathematical ‘objects that can enter into this algebraic structure. Combining the diagrams of Figures 4.3 and 44, together with Eqs (4-23) and (4- 25) leads to the diagram shown in Figure 4.5. Fig. 4.5 Transformation diagram ofthe network problem, Now the arbitrary vectors w’, U, p’ and P can no longer be chosen independent of cach other, since these vectors are related through Eqs (413), (4-19), (4-23) and. Fg. (4-25), i. U+Biu'="P +C'p!), (4-28) and: PHCp'= KU + By’) (4-29) Pre-muliplication of Eg. (4-25) by C and Eq. (4-26) by B results in, cU-cr +crc'p’, (4.30) and BP = BKU + BKB'u’ Cen) ‘The above relations should be satisfied for every admissible choice of vectors u’ U, p'and P. Up to this stage, no physical significance has been assigned to the network variables, However, there are many physical phenomena that exhibit two sets of ‘variables (dual) as discussed above, and these sets will be inter-related in the same ‘way. These phenomena can be treated as network problems, and subsequent sections of this volume will concentrate upon elastic networks.18 ‘Structural Mechanies: Graph and Matrix Methods 4.33 FORMULATION OF NETWORK ANALYSIS In a network problem the following dat ae normaly given: 1A graph model for which matrices B and C can easily be formed 2. ‘The transformation matrix f and/or matrix k 3. The abiteary vectors Vand P ‘The unknowns consists of wand p vectors, which should satis the following 1. 8g. (4-24) and /or Bg. (4-26) 2.9, (4-20) and/or Ba, 4-19), 3. Ba. (+15) and/or Ba. (4-14), “Two different methods ean be used for solving the above problem, namely the mesh method and the node method. The first approach introduces auslary eycle (mesh) variables p’ and the latter uses auxiliary node variables. u’, respectively. The transformation diagram of Figure 4.5 can easily be extended as illustrated in ont A. des LJ B ig. 4.6 Extended transformation diagram. Mesh Method: This method follows directly from Eq. (4-30). This equation satisfies conditions 1 and 3, and can be written as cWU-@)=CIC'pt re ) In this equation the left hand side is known and it may be solved for p’ by the inversion of (CIC), ie. CHAPTER 4 Network Formulation of Structural Analysis 9 pt =(CIC)* CU ~). (4.33) Now employing p= C'p’ and u = f(P¥p) ~ U, the solution is competed. It should ‘be mentioned that condition 2 is also satisfied by Bq. (4-20) ‘When the computation of R and V in an explicit form is needed, then one can use, R= Ci(crcycu +[1- C(CIC) CHP, (434) in which Iisa unit matrix and V=1C (CAC) ICU + [FIC (CIC) CNP. (435) Node Method: ‘This method uses Eq. (4-31), which satisfies conditions | and 2 of the solution and can be written as BOP KU) = BKB‘ (436) Solving for u’ , we obtain, (BKB) BP KU), 43 ‘and the solution is completed by employing w= B'’ and p= k(U+w)~ P. ‘Vectors V and R can now be computed explicitly as, R=KB(BkB')"! BP + [kK (BKB)” BKIU, (438) and Y= B'(BkB)"' BP + (I BY (BKB) "BAIL. (439) 44 FORMULATION OF STRUCTURAL ANALYSIS ‘The following properties are common to networks and structures: |. The underlying mathematical models for both are graphs. 2. The sum of the forces at each joint is zero. 3, Distortions in any eycle sum to zero.120 Structural Mechanics: Graph and Matrix Methods [Network variables have the following physical interpretations: cycle (redundant) forces. yemiber forces induced by redundants, pplied member forces (Fixed end actions) R= total member forces. P’= joint loads. = cyele distortions (lack of it at closure). U= applied member distortions ((ree-ed distortion). 1 member distortions induced by joint displacements. v jotal member distortions. = joint displacements, k = unassembled stiffness matrix of the structure containing member stiffness matrices ina block diagonal form. £ = unassembled flexibility matrix of the structure containing member flexibility ‘matrices in a block diagonal form. Structural analysis is different from network analysis in that the elements ‘corresponding to nodes, members and cycles are not scalars but vectors. Naturally, for different clements, different vector bases are needed. This problem can be ‘overcome by interpreting the elements of topological matrices as identity matrices ‘of appropriate orders. The geometric transformations between vector bases, however, can be handled separately from the topological ones. Various methods are available for separating geometry from topology. As an example, using a global coordinate system for the entire structure, all variables can be’ referred 10 a ‘common point and therefore become directly additive, An alternative vector basis of the member variables and a tree formation matrix between member and nodal ‘quantities can be redefined, Spillers [222]. Finally, a slight modification can be made to the definition of the incidence matrix B whereby geometric transformations are effected only from the positive to the negative end of a ‘member, Fenves and Branin [47]. It is now easy to show that the network problem. corresponds to a complete formulation of structural analysis, Geo CHAPTER 4 Network Formulation of Structural Analysis 1a and R = KV are force Force Displacement Relationship: The oquations V = R fore placement rlaionsips which cn be rege as generalized Toke’ lw sic K. contain the flexibility and slfiness matrices of indivi aiarisse, respectively. These matrices ae known as the unassembled flexibility aa ttiens patties of the structure, which are discussed in Chapters $ and 6 equilibrium of Forces: The elationships p= C'p’ and Bp = 0 are equilibrium cquations, The frst equation tates thatthe member forces ean be expressed as Haase pintions of independent cyele forces orredundants. The later indicates thatthe sum of member forces at every joint is equal to 220, lit = Biu’and Cu = 0 are ae ‘Pare present and U, Pand Ure taken as zero, Foree Method: In this method the unknowns are taken as cycle forces (redundants) p’. By the equilibrium of forces: R=PHp=P+Cp (4-40) can be inl by ein utube fring its aetna ah LL abers are Py P’ and zero in its ‘mattix Pr. Obviously member forces in tree memt chords. Therefore P,P’ =| (ty ’ [ ® | ‘Using the flexibility matrix, member distortions can be written as: V=R=1P+IC'p’. (442) ‘Compatibility conditions can now be imposed, leading to: U'=cy= cir +cic'p’=0. 4-8)12 Structural Mechanics: Graph and Matrix Methods Solving for redundant forees, we obtain: p’=-(crcy cr. (4-44) ‘The member forces and distortions ean be obtained as, R=P-C(crcy ‘cr, (4-45) and V=f[P-Cicrc) ‘erry, (4-46) ‘hich completes the solution by the force method. Displacement Method: In this method the unknowns are joint displacements w Imposing the compatibility continuity) conditions: VeUtus0+ Bw =Bu’ 47) Using the unassembled stiffness matrix ofthe structure, we can write: RekV kB (4-48) [Now by equilibrium, the total member forces in each joint ofthe structure must be equal fo 22t0, i, P’= BR = BKB'w’ (4-49) ‘The joint displacements can now be calculated as w= (BKB) PY, (4-50) and the final member distortions and member forces can be obtained as, V=Bipne)'P’, 51) and R= KB(BKB) P’ 52) ‘This completes the analysis ofa structure by the displacement method, AC this stage, we have shown that the two methods of structural analysis correspond, directly tothe general network solution as illustrated in the diagram of Figure 4.6. Although a general formulation of the force and displacement methods has been presented in this chapter, an efficient analysis ofa structure requires modifications CHAPTER 4 Network Formulation of Stcuctural Analysis 123 nd special considerations. These together with examples in the ilustraton of the ‘methods will be given in subsequent chapters. EXERCISES 4.1 Consider the oriented graph model of a network as shown in Figure (2. “Associate the nodal quantities as depicted in Figure (b). Find the corresponding ‘member quantities, 1 5 @) 0 4.2. In the previous exercise find the quantities which are associated with the regional eycles ofthe graph. ‘The following exereises are intended to be treated as interesting esearch problems, and should not be considered as simple exercises 4.3. Formulate the analysis of rigid-jointed skeletal structures in terms of vector spaces (soe Ret, (169). 4.4 Formulate the analysis of frame structures using algebraic topology (see Refs (156,214), 4.5 Formulate the analysis of skeletal structures in terms of the theory of rmatroids (see Chapter 9 and Ref. [174] for equivatent network formulation). 4.6 How do you modify network analysis if there are some releases in the structure (see Ret. [168))CHAPTER 5 Matrix Displacement Method 5.1 INTRODUCTION In the last half-century, considerable progress has been made in the matrix analysis of structures, The topic has been generalized to finite elements, and extended to the ability, non-linear and dynamic analysis of strctutes. This progress is duc to the simplicity, modularity and flexibility of matrix methods. Many textbooks covering these methods have been published including Argyris [4], MeGiire and Gallagher [172], Livesley [160], Meek [173], Kardestuncer (88), Ziegler [254], and Vanderbilt [242] among many others. In these books the displacement method of structural analysis is thoroughly developed, and therefore only a brief introduction will be presented here. 5.2 FORMULATION In Chapter 4, the network-topological formulation of the displacement (stiffness) ‘method of structural analysis bas already been presented. In this section, a matrix fornmalation using the basic tools of structural analysis - equilibrium of forces, ‘compatibility of displacements, and force-displacement relationships - is provided. ‘The notations are chosen ffom the most popularly encountered versions in structural mechanies. Consider a structure S with M members and N nodes; each node having one degree ‘of freedom, The kinematical indeterminacy of S may then be determined as,I 126 Structural Mechanics: Graph and Matrix Methods 18) =0N-p, G1) hore Bis the numberof ens de oth ims tothe spot contons. As an exanp 14S) fr tho planar tus 8 depicted in Figure 51a is gien by MS) no 7 an fr te pc te show Figs 5.1) i eal 6 H)= 68 (2) A planar truss. (b) A space frame. Fig. 541 ‘The degrees of freedom of the joints for two structures, One can also calculate (8) by simple addition of the degrees of freedom of the inns ofthe snot, is re tas SMS) 282 eae heehee frame n(S) = 4x6 = 24. no Tan rite Let p and v represent the joint loads and 1 joint loads and joint displacements of a structure. Then the foreedispacement relationship forthe structure can bo expresed og p> Ky, (52) where K is a NNxqN symmetric matrix, ki the stifi x, known as the stiffness matrix of the structure. Expanding the ith equation of the abave system, the force p, can be expressed in terms ofthe displacements (¥).¥eunMay) a BES KM, Kia, tt Kegan 63) A typical coeficient Ky i the value ofthe force py requi value of the force p, required to be applied atthe ith component ofthe structure, in order fo produce a displacement yal at j and zero Aisplacements at all the other components, 7 CHAPTER 5 Matrix Displacement Method a7 ‘As has been shown in Chapter 4, the member forces r can be related to nodal forces p by: p=Br. G4) Similaey, the joint displacements v can be relate to member distortions w by: wey. 65) For each individual member of the stucture, the member forces can be related to member distortions by an element stifiness matrix Ig. A block diagonal matrix containing these element stiffness matrices is known as the unassembled sifnest matrix ofthe structure, denoted by k. Obviously: =u. 66) This equation together with Eqs (5-4) and (5-5) yields: p= BKB 6 Therefore, K = BKB! 68) is obtained. The matrix K is singular since the boundary conditions ofthe structure are not yet applied, For an appropriately supported structure, the deletion of the rows and columns of K corresponding the support constraints results in a positive definite matrix, known as the reduced stiffness matrix of the structure. Let us illustrate the method by means of a simple example. Consider a fixed end beam with a load P applied at its mid span. This beam is discretized as two bear elements, as shown in Figure 5.2(a). The components of element forces and element distortions are depicted in Figure 5.2(b) and those of the entire structure are illustrated in Figure 5.2(c), oo — 1 ' (9) A fixed ended beam 8.128 ‘Structural Mechanies: Graph and Matrix Methods rh TA To tet NX @N i Bh Fly ty (b) Member forces and member distortions. Praha MADAM Pet Pay (©) Nodal forces and nodal displacements ofthe entre structure. Fig, 5.2 Illustration ofthe analysis of simple structure, For each clement such as element 1, the element stiffness matrix can be writen as, ny fk ki kis kis Pur fa} | Ror on Kas kag fu = 69) te] [isi sz Kas yg rus 4, kar kas kag [ug and forthe entire structure we have: Pi] [Ki Kio Kis Kis Kis Kis Jv P2] ]Kar Koz Kas Kay Kas Koo | vo Ky Ky Kaz Kaz Kas Kye fv. Ps} _]Kar Kap Kas Kye Kas Kae va (5-10) Pal |Kar Kaz Kas Kay Kas Kas] va Ps} [Ks Ksz Ks3 Ksq Kss Kyo [vs Pol [Kor Ker Kes Koa Kes Keo hve. Element stiffness matrices cj and ky can be easily constructed using the definition ‘of ks, For a beam element, ignoring its axial deformation, these terms are shown in Figure 5.3. The structure has a uniform cross section and since both elements have the same length: z CHAPTER 5 Matix Displacement Method 9 kan kq3, yg ky aq Fig. 53 ‘The sins eoefcens ofa beam element ignoring its ail cformaton of -3/L -6/? =3/) sme 2 11 oay Mem To a ot? 3 “WL ot BL 2 ‘The unassembled stiffness matrix is an 8*8 matrix of the form I: eee (5-12) ft “] "Now consider the equilibrium of the joints ofthe structure, resulting in, + Pa Tst hy Pete tte | P= Pete G1) ‘or ina matrix form we have,130 and more compactly: Pr Pa Ps Ps Ps Ps, Structural Mechanics: Graph and Matrix Methods p=Br. ‘Consider now the compatibility of displacements as: and in a matrix form and in compact form: Wewyavy w= We have, Ev=By. (5-14) (S15) (5-16) (S17) (5-18) a CHAPTER $ Matrix Displacement Method BI ‘The reason for matrix E being the transpose of the matrix B, has already been discussed in the previous chapter, however, using the principle of virtual work, a simple proof can be obtained. Consider: W= work done by extemal loads. Lat U= strain energy = 4u'r nn Equating W and U leads to B= BY and complete the proof. “Therefore the stiffness matrix ofthe entire structure can be obtained as: oi? -3/L -6/12 -3L 00 -7L 2 WL 1 0 oO eo? WL 12? 0-61? SL gy L[-37- 1 0 4 ML 0 0-6/7 LoL 3L te oa Applying the boundary conditions, leads to the formation of the following reduced stiffness matrix: pa] _2elfi2i? ofa Rhee] a > pp therefore vq = Pig PL Since p,= 0 and : . 24E1 ~ 24ET From this simple example, it can be soon that matrix B is a very sparse boolean ‘matrix and the direct formation of BKB using, matrix multiplication requires a considerable amount of storage. In the following, i is shown that one can form [BKB with an assembling process (Known also as planting), as follows:132 ‘Structural Mechanics: Graph and Matrix Methods Consider an element "a" of a structure, as shown in Figure $4, for which the clement stiffness matrix can be written as, E a 5.21 yy 21) { and j are the two end nodes of member a. Multiplication BRB has the following elfeeton ky joo foo) oo 09) oo oo 1 ofki Kiyo 0 0 1 0 0 0 o) |r ofo 0 6 kyo Ky 0 O 0 oft kjfo 0 0 0 0 1 0 of lo ofo 00 Ks OK; 0 o or ou oo oo oo 0 | (5-22) yooo 000090 000 00 0 00 3J000 0 0 0 od : 6 4/0 60 Kj 0 ky 0 0 a sJo00 0 0 0 0 7 a 4 so 0 0 ki 0 ky 0 0 000 0 0 0 00 soo 0 0 0 0 00 3 2 1 Fig. 54 A structural model 8, ‘The adjacency marx of i ako an x8 matix and the effet of node 4 being aac ne 6 tents of unt ents tte sane etn he submatrices of the element One can build up the adjacency sin of gap the addon ofthe eet of one member ata tine. Inthe sae way, te ea lo form the ovr sess mato th strut by the adton of the conten of every member in sueesion As an example, fo the grap shown in Figure 5, the overall stifles mi haste following pater CHAPTER 5 Matrix Displacement Method 133 Olas 7.8 Sc (523) ‘Non-zero entries are shown by *. For a stifess matrix each of these non-zero entries is an onc submatrix, where a isthe degrees of freedom of each node of the structure. As an example, for a planar truss o¢= 2, and fora space frame c= 6. The formation of the stiffness matrix by the above process is known as assembling oF planting ofthe stiffness matrix of @ structure In the above example, the stiffness matrices could be assembled because both are constructed with reference to the same coordinate system. However, fora structure in general, the stiffness matrices should be prepared in a single coordinate system. On the other hand, for each element, there exists a coordinate system attached to the element, known as a local coordinate system. In Figure 5.5, local coordinate systems for members 45 and 25, and the global coordinate system for the entire structure ae illustrated, Fig. 5.5. Local X,Y and global coordinate x, y systems. A global coordinate system can be selected arbitrarily; however, it may be advantageous to select this coordinate system such that the structure falls in the first quadrate of the plane, in order to have positive coordinates for the nodes of the structure. On the other hand, a local coordinate system of a member has one of134 Structural Mechanics: Graph and Matrix Methods its ae along the member, the second ai sin ts plane of sty (i ths one) and the tid as is chosen such ht res aight funded sordinge _ The transformation from a local coordinate to global coordinate inate © a plobal coordinate system ean be performed as ilustrted in Figure 5.6, in which xyzis the global system nd xoy.zn ‘often denoted by 32, is the local system, a ‘The relation between xyyiz, and xyz can be expressed as xi] [cosa 0 sino x Bile OLY, (5-24) 2] [-sina 0 cosa z @ o) Fig. 5.6 Transformation from local coordinate system to global coordinate system, Similarly xxysz, and x,y\2y are related by, x2] [cos sin oxy y2]=|-sinB cos of yi 625) nllo 0 ify CHAPTER 5 Matrix Displacement Method 135 oo) (0k ois and y3]=|0 cosy siny} yo. (5-26) 23] |o -siny cosy) z> ‘Combining the above transformations results in, (cosccosp) Gin) (cosBsina) 1=|~inasiny reovasinpicosy) (cosBeosy) (sinyeosa~sinassin eos) (sinetcosy—cosassinBsiny) (-cosBsin7) (cosercos+ sin sin sin 7) 627) whore: a] fx ys |= Et} (5-28) al le ‘A vector in a local coordinate system T and in a global coordinate system Pare related by: 6-29) Wan easily be proved that T i an onthogonal mattx, ie. or'= rr. (630) Inthe above transformation, represents the lilt ofthe member which is quite often zero, Thus T can be simplified as: cosatcosB sinB sincrcosB. T=|-cosasinf cos —sinasinB 631) msn «0 cose ‘This matrix can easily be written in terms of the coordinates of the two ends of a vector. Considering Figure 5.6(b), Eq, (5-31) becomes, xy/L yl ay lL. Tealaxgyy (LAL LANL: ypizy/L*L}, (5:32) Oe ee136 Structural Mechanies: Graph and Matrix Methods where: AEX ORIEN BTA e 1 1 Btxhy? and L= (eh + yh +xh)? 6-33) Notice that T transforms a 3-dimensional vector from a global to local coordinate systom and 1" performs the reverse transformation. Hovever, if the element forces or element displacements (distortions) consist of p veetors, the block diagonal matrix with p submatrices should be used. As an exemple, fora beam element of a space frame, with each node having 6 degrees of freedom, the transformation matrix is a 1212 matrix ofthe form: T| (534) 5.3 ELEMENT STIFFNESS MATRICES Element stiffness matrices for skeletal structures can be obtained using various methods, For some elements, concepts from mechanics of solids are sufficient for the formation of an element stiffness matrix; for others, energy methods are more suitable. In the following, a general method for the formation ofa stiffness matrix is presented and then applied to bar and beam elements. The details of the 0 for every non-zero vvocior x. The stiffness matrix K ofa structure is positive definite since, plv= (Ky) =viK'y = vIKv=2W, ‘and W is always positive 5.3.2. STIFFNESS MATRIX OF A BAR ELEMENT ‘Consider a prismatic bar element as shown in its local coordinate system, Figure 5.8, According to the definition of such an element, only axial forces are present. ‘The strain energy of this bar ean be calculated as: CHAPTER 5 Mattix Displacement Method Hf ove antsdyte= ff eb taayae By ede u 2 ‘On the other band: Fig, 5.8. A bar element in its local coordinate system, Since the strain is constant along the bar, u, ean be expressed as: From the boundary conditions: Hence: 139 (5-46) 6-4) (5-48) (5-49) (5-50) (6-51)140 Structural Mechanics: Graph and Matrix Methods ‘and from Bq, (5-46) the strain energy of the bar can be calculated as: (5-52) Hence: (5-53) AA, and Kj =0 forall other components. ‘Therefore, the stiffness matrix of a bar element in the selected local coordinate system is obtained, and i co ca rn ale 7 fe 5 (5-54) is is is, 6, From Eq. (5-29), we have; (8-55) and w=Tw 6-56) From the definition of an element stiffness matrix ina local coordinate system: 657) By substitution of Eqs (5-55) and (5-56) inthe above equation: r=TRru=T'RTu, (5-58) CHAPTER 5 Matrix Displacement Method 141 By definition ofa stiffness matrix in a global coordinate system: r=ku. (5-59) ‘Comparison of Eg. (5-58) and Eg, (5-59) results in; rier, (5-60) Hence the stiffness matrix of a bar element in a global system, as shown in Figure fe} Fig. 5.9 A bar element of a space truss Denoting T in Eg. (5-32) by, Tr Ta Ts T=|T% Tr Tas 5-61) Ts Ts2 Tas, can be written as:142 ‘Structural Mechanics: Graph and Matrix Methods mH TaT2 sym, EA) TuTs ToT e (5-62) 2 Te Tihs Th “Tila -T Talis Tha TR Tis ~TaTis -T Tus Tats TA ‘The entries of the above matrix can be found using T, from Eq. (5.32). As an ‘example, the stiffiess matrix of bar | in the planar truss shown in Figure 5.10, ean be obiained as: th a i Oh, + yh #7)? T=" Och tyh teh)? Fig. 8.10 A planar truss and the selected global coordinate system, ‘Therefore: os -05 EA |-05 05 Ly2|-05 05 05-05 05 05 os _-05 os -05 -05 05 k= 5.33 STIFFNESS MATRIX OF A BEAM ELEMENT Consider a prismatic beam element as shown in Figure 5.11. The element forces and the element distortions, are defined by the following vector: CHAPTER 5 Matrix Displacement Method 3 Fe tnotgetyeona)!s and {Ha sett a}, ‘where 1) ~ 1) are the force components at end i and ry ~ te are moment components at end i, Also ry ~ ty are the force and rip ~ rjz are the moment components, respectively at the end j, and u, 12) ate correspondingly the translations and rotations af the ends i and j ofthe element. Fig. 5.11 A beam element in the local coordinate, Using energy methods, the stiffness matrix of the beam element in the local coordinate system defined in Figure 5.11 can be obtained as:144 ‘Structural Mechanics: Graph and Matrix Methods In which 1, I, and J are the moments of inertia with respect to the and Z axes ‘and J isthe polar moment of inertia of the section. E specifies the elastie modulus ‘and v isthe Poisson ratio. The length of the beam is denoted by L. For the two-dimensional ease the columns and rows corresponding to the third «dimension can easily be deleted, to obtain the stiffness matrix of an element of a planar frame. ‘The stiffness matrix in a global coordinate system ean be written as T tT | fe) 7 (5-4) v T r T For the two-dimensional caso, (5-65) ‘The entries of kare as follows, ky = Tiiog + Tho kay =TiMie +TaTa20§ ky = Toy + Tho kay = Tyo, kp= Tao ky = af Kay = ~The +108 ag = Moy Tagh ~TiaT ot okgs = Tye, kag = T1083 6-66) Ksy=—ThiTa20§ —TaTh yoy sq =—THio§ - They gg =TyiTyp08 +737, 04 4 Tartan *TaT ss Dad 4 TH rand + Tyg, Koy = Tat hg = Tra sg =O, og = Tye, kes =—Taaa, eg = 08 (CHAPTER 5 Matrix Displacement Method 14s in which: EA js _ 6El, 3 mat a 4 ‘As an example, consider the planar fiame, shown in Figure 5.12, with 30x10°%m* and E=2x10"'Nim?. For element 1 we have, Ts Ta=0, Test Ta= and the stiffness matrix ofthe element is obtained as, Fig. 5.12 A planar frame. 54 OVERALL STIFFNESS MATRIX OF A STRUCTURE Once the stiffness matrix of an element is obtained in the selected global coordinate system, it can be planted in the specified and initialised overall stiffness146 Structural Mechanies: Graph and Matrix Methods CHAPTER 5 Matrix Displacement Method 147 tmatrix of the structure K, using the process described in Section 5.2. This is illustrated by the following simple example: Let $ be a planar truss with an arbitrary nox clement numbering, as shown in ] EA} O 0 calculated using Eq. (5-32) and Eq, (5-33) as follows: 1 ee Forbarts hy =322L Sno i | the over ttresn mati ofthe stature is an 88 main, whch cn easly be 2 2 2 i fotmed by planting the three member stiffness matrices as follows: Partitioning K into 2x2 submatrices, it can easily be seen that it is pattern ‘equivalent to the node adjacency mattix of the graph model of the structure as Now the stiffness matrices can be formed using Eq. (5-62) as: * 0.433 0.75 0.433 0.75, | ‘This pattem equivalence simplifies certain problems in structural mechanies, such as ordering the variables for bandwidth or profile reduction, methods for increasing the sparsity using special cutset bases, and improving the conditioning of structural ‘matrices, which will be discussed in Chapters 7 and 8 02s sym. BAl-0433 0.75 | 2|-025 0433 0.25 0.433 0.75 —0.433 0.75 ‘The mattix KK is singular, since the boundary conditions have to be applied. For bar 2: ky Consider, p=Ky ‘and partition it for free and constraint degrees of freedom as:148 Structural Mechanics: Graph and Matrix Methods Pr]_[Kr Ke [vr (lle eT: en ‘This equation tas a mixed matte; an, have known values and pe and jae unknowns. Kris known as the reduced stiffness matrix of the structure, which is ‘non-singular for a rigid structure, For boundary conditions such as vz ~ 0, itis easy to delete the corresponding rows and columns to obtain, Pe=Karve, (5-68) from which vy can be obtained by solution of the above set of equations, In a computer this can,be done by multiplying the diagonal entries of Ke by a big ‘number such as 10%. An alternative approach is possible by equating the diagonal entries of K., to unity and all the other entries ofthese rows and columns to zero. If ¥. contains some specified values, pe will have corresponding v, values. A third ‘method, which is useful when a structure has more constraint degrees of freedom (euch as many supports), consists of the formation of element stiffness matrices considering the corresponding constraints, ie. to form the reduced stiffness ‘matrices of the elements in place of their complete matrices. This leads to some reduetion in storage, also at the expense of additional computational effort. AAs an example, the reduced stiffness matrix of the structure shown in Figure 5.13 can be obtained from K, by deleting the rows and columns corresponding to the three supports 1, 3 and 4, 20]_EA[LS 0 [ur 30} 2 [0 1shury ‘The solution results in the joint displacements as: a rr ‘The member distortions can easily be extracted from the displacement vector, and ‘multiplication by the stiffness matrix of each member results in its member forces in the global coordinate system, As an example, for member 3 we have: 1 4 i (CHAPTER 5 Matrix Displacement Method 7) 40/1SEA] [13.33 gm | 4orea |_| 0 1 oP ]-333 o of o o ‘A transformation yields the member forces in the local coordinate (-23.99 23.99}, rp = {-10.659 10.65)! and ry = (13.33 -13.33)". systems, Example: The ise shown in Figute 514 has member veh of he same cross, festa ae of 13000, and taste modus 210 KNin?. Veta loads of TOEN and SEN ae applied anode 3 and node 5 espetvely. Deter he ees inall member: -14 A planar truss S. ‘The force-displacement relationship for a planar bar member is obtained from Eq, (5-23) as follows (5-69) ‘The stiffness matrices forthe members of § are determined as: For members | and 2:150 Structural Mechanies: Graph and Matrix Methods 1575.0 0 ~1575.0 0} [1575.0 0 -15750 01 o 0 0 of fo 0 0 0 175.0 0 -1375.0 of |15750 0 ~1575.0 0 o 0 0 of Lo 0 0 0 For members 3 and 4 121071 53002 -1210.71 330.02 530.02 23203-53002 ~232.03 ~1210.71 83002 1210.71 530.02 |’ =53002 -232.03 530.02 232.03, 67208-58818 ~672.08 588.18, 588.18 51476 588.18 514.76 672.08 $88.18 762.08 ~ 588.18) 588.18 ~S14.76 -588.18 514,76 For members 5 and 6: 0 0 0 Oo 1210.71 -s3002 -1210.71 530.02 0 3600.0 0 ~36000] |~s30.02 232.03 s3002 —232.03 9 0 0 o ]*|-1210.71 53002 1210.71 ~330.02 0 ~3600.0 0 36000] | s30.02 -232.03 -s3002 232.03, For member 7 1210.71 -53002 ~1210.71 530.02 ~$3002 232.03 530.02 -232.03 1210.71 53002 1210.71 —s30.02 53002-23203 530.02 232.03 Assembling the stiffness matrix ofthe entre structure and imposing the boundary conditions 3} =3Y = 8 = 8% =0 results in CTIAPTER 5 Matix Displacement Method 151 0] foes. 3 0 | |-530.02 4296.09 sym. 8 ° ° 0382208 8 10] 0 3600.0 588.18 4114.76 a o | |-12t071 s3oor -1s750 © 0 2785.71 By s3002 23203 0 0 $3002 28203) ay ‘The solution ofthe above equations results inthe joint displacements Tos14x10°> 8} =-2.12241x10-, 8§ =-1.09894x10, ~1.828108x10? and 8Y =-9.521007x10 BY = 225665107? 33 Once the displacements are calculated, the member forces can easily be obtained using member stiffness matrices. 5.5 GENERAL LOADING ‘The joint load vector of a structure can be computed in two parts. The first part comes from the external concentrated loads and/or moments, which are applied at the joints defined as the nodes of S. The components of such loads are most easily specified in a global coordinate system and can be entered to the joint load vector P. ‘The second part comes from the loads which are applied on members. These loads are usually defined in the local coordinate system of a member. For each member the fixed end actions (FEA) can be calculated using the existing classical formulae ‘of tables. A simple computer program ean be prepared for this purpose. The fixed fend actions should then be rotated to the global coordinate system using the ‘transformation matrix given by Eq. (5-27). The FEA should then be reversed and applied to the end nodes of the members. These components can be added to p to form the final joint load vector, After p has been prepared and the boundary conditions imposed, the corresponding equations should be solved to obtain the joint displacements of the structure, Member distortions can be extracted for each ‘member in the reverse order to that used in assembling p vector Example; A portal frame is considered as shown in Figure 5. 15. The members are all made of sections with area A= 150cm?, moment of inertia 1, = 2x10'em' and clastic modulus E = 2xLO°kN/em’, Calculate the joint rotations and displacements152 Structural Mechanics: Graph and Matrix Methods son Eye 3 12bNn ES 7) Glas al | 5 ale = Fig. $.18 A portal frame and its loading. ‘The equivalent joint loads are illustrated in Figure 5.16: 16004 m ry 74 Fig. 8.16 Equivalent joint loads. Employing Fa. (5-66), the sifness matrices forthe members are obtained as For member 1 ‘0.008 0 075 oe keto 15 © 00 0.008 0 15 0.008 7 0-075 0 0 075 “1S 0 200 15 0 400, and for member 2; CHAPTER 5 Matrix Displacement Method 153 06 0 0.004 sym. ee) m7 0 0 Wd © 0.004 -096 0 0.004 0 096 160 0 0.96 320 For member 3: 0.008 0 075 sym, 400 Poe =0.008 0-15 0.008 Ge 075, 0 20) 075 150 200-15 0 400 Assembling the stiffness matrices and imposing the boundary conditions results in the following equations: 14 0.608 Ey ° 0 0754 sym. y 160} gq] 1S 096 720 05 ° -06 0 0 0.608 # ° 0 0.004 -096 0 0754 — ay ° 0 056 16015 -096 720) 64 Solving these equations leads to: B £85 = 0.06533767, 8Y =~2.617704E-04 and 0§ =— 1.16855E-04 (0659167, 8}=2.6177648-04, 6} -8.983453E-05, 5.6 COMPUTATIONAL ASPECTS OF THE MATRIX DISPLACEMENT METHOD ‘The main advantage of the displacement method is its simplicity for computer programming, This is due to the existence ofa simple kinematical basis formed on.Ist ‘Structural Mechanies: Graph and Matrix Methods 1 special cutset basis known as cocycle basis ofthe graph model S of the structure, Such a basis does not correspond to the most sparse sifiness matrix, however, the sparsity is good enough, not to look for a better basis in more usual cases. However, if an optimal cutset basis of $ is needed, then the displacement method has all the problems encountered in the foree method, described in Chapter 6. The algorithm for the displacement method is summarized in the following. The coding for such an algorithm may be found in textbooks such as those of Vanderbilt [242] and Meek [173] Algorithm, ‘Step 1: Select a global coordinate system and number the nodes and members of the structure. An appropriate nodal ordering algorithm will be discussed in Chapter 1 Step 2: After initialisation of all the vectors and matrices, read the data for the sructure and its members. For multi-member regular structures, data can be generated using the method of Chapter 10. Step 3: For each member of the structure: (@) compute L, L*, sina, sinBsiny, cosa, cosB, cosy, (b) compute the rotation matrix T () form the member sti 3s matix I ints local coordinate systems, (@) form the member stiffness matrix k in the selected global coor system; (©) plant k in the overall stiffness matrix K of the structure Step 4: For each loaded member: (a) read the fixed end actions; (6) transform the fixed end actions to the global coordinate system and reverse it to apply at joints; () store these joint loads in the specified overall joint load vector. Step 5: For each loaded joint: () read the joint number and the applied joint loads; + 4 1 CHAPTER 5 Matrix Displacement Method 155 (b) store it in the overall joint load veetor. Step 6: Apply boundary conditions to the structural stiffness matrix KK, to obtal the reduced stiffness matrix Kg, Repeat the same for the overall joint load vector. Step 7:. Solve the corresponding equations to obtain the joint displacements Step &: For each member: (a) extract the member distortions from the joint displacements (©) rotate the member distortions tothe loeal coordinate system; (© compute the member stiffness mats (compute the member Forces and fixed end ations, Step 9: Compute the final member forces. For an efficient displacement analysis ofa structure, special considerations must be talcen into account, which will be discussed in Chapters 7, 8 and 10 of this book. EXERCISES 5. Compute the transformation matrix T of the element "a" for the following. truss: 7 ‘ 5 ; » r & ya ae 5.2 For the following planar frame, compute the joint displacements, ignoring, the effect of axial deformations of the members. El is considered constant for all the members:156 ‘Structural Mechanics: Graph and Matrix Methods SOKN 5.3. Compute the stiffness matrix of member "a" for the following poral frame, in the selected coordinate system: 5.4 Consider a single layer grid in the xz-plane loaded in the x-direction, Derive the stiffness matrices of a member of this grid in local and global coordinate systems. 55 Find th join displacersents of the space tras in the following digs ‘where EA is considered to beth aun for al fhe monies — 5.6 Compute the joint load vector ofthe following frame: CHAPTER 5 Maitix Displacement Method 137 0eN 4m oki 40QN 2m 2m NA 5.7 Derive the stiffness matrix of a beam element of a planar frame which has ‘one end hinged, 5.8 Determine the patterns of the stiffness matrix and reduced stiffness maitix Of the following pitched roof frame using two dilTerent node numberings, chosen arbitrarily: 5.9 Perform the matrix stiffness analysis for the following truss, where P = TORN, EA = 40x10°N and L= 4m: a7 ! LON, ee 5.10 Derive the overall stiffness matrix of the following grid and calculate its joint displacements: EL= 60x10°N.m', GI = 100x10°N.n?, L = 4.0m and P= 1.0x10°N.158 Structural Mechanics: Graph and Mattix Methods ap ie 5.11 Perform the matrix stiffness analysis forthe following grid: P= 30KN, E = 2.1x10'kNiem® , G 8560em*, J=33730em' and A= 149er 1.05x10%«N/em’, 1, = l, = 25170em', “vam y CHAPTER 6 Matrix Force Method 6.1 INTRODUCTION The force method of structural analysis, in which the member forces are used as unknowns, is appealing to engineers, since the properties of members ofa structure most often depend on the member forces rather than joint displacements. This method was used extensively until 1960. After this, the advent of the digital computer and the amenability of the displacement method for computation attracted most researchers, As a result, the force method and some of the advantages it offers in non-linear analysis and optimisation has been neglected. Four different approaches ate adopted for the force method of structural analysis, ‘which will he classified as: 1. Topological force methods, 2, Algebraic force methods, 3. Mixed algebraic-combinatorial force methods, 4. Integrated force method. Topological methods have been developed by Henderson [76] and Maunder [169] for tigid-jointed skeletal structures using manual selection of the cycle bases of their graph models. Methods suitable for computer programming are due to Kaveh, [89,953,107]. These topological methods are generalized to cover all types of skeletal structures, such as rigid-jointed frames, pin-jointed planar trusses and ball- jointed space trusses (94] Algebraic methods have been developed by Denke [36], Robinson [204], Tope [238], Kancko et al. (87), Soyer and Topsu [ 221] and mixed algebraie-topological160 Structural Mechanies; Graph and Matrix Methods methods have been used by Gilbert et al. [58], Coleman and Pothen [29-30], and Pothen [196]. ‘The integrated force method has been developed by Patnaik [177-179], in whic the equilibrium equations snd the compatibility conditions are’ satisfied simultaneously in terms ofthe force variables. In this chapter, topological methods using graph-theoretical concepts are mainly presented. However, algebraic methods and mixed algebraic-topological approaches are also briefly discussed 6.2 FORMULATION In this section a matrix formulation using the basic tools of structural analysis - equilibrium, compatibility and load-displacement relationships - is described. The notations are chosen from those most commonly utilized in structural mechanies. Consider a structure $ with M members and N nodes, which is 1(S) times statically indeterminate. Select +(S) independent unknown forces as redundants. These ‘unknown forces can be selected from external reactions and for internal forces of the structure. Denote these redundants by A= Lada, dysyh GI) Remove the constrains corresponding to redundant, inorder to obtain a statically determinate structure, known as the basic (released or primary) structure of S. Obviously a basie structure should be rigid. Consider the joint loads as, DE [Py PaysPal's 62) ‘where n is the number of components for applied nodal toads. [Now the stress resultant distribution r due to the given load p for a linear analysis by the foree method can be written as, =Bop+ Ba, (63) ‘where By and By are rectangular matrices each having m rows, and n and 7(S) coluras, respectively, mu being the numberof independent components for member forces. Byp is kaown a8 particular solution, which satisfies equilibrium with the 7 (CHAPTER 6 Matrix Force Method let imposed Toad and Byq is a complementary solution formed from a maximal set of independent sel-equilibrating stress systems (S.E,Ss), known as a statieal bass. ‘As an example, consider a planat truss, as shown in Figure 6.1(a), which is 5 times statically indeterminate. EA is taken to be the same for all the members. mp (2) A planar tus. (b) Selected unknown forces. Fig. 6.1 A statically indeterminate planar truss. (One member force and one component ofa reaction may be taken as redundant ‘Allemativly, two member forces ean also be selected as unknowns, as shown in Figure 6.1(0). The corresponding Bp and By matrices can now be obtained by applying unit values of p, (1,2) and q G=1,2,3), respectively p[0 get kG wi=[ 238 38M 8 0 Sk Se Oe th and “35 0-35 0 1010-45-45 0 piel o” 35 0-35 0101 0 Was —45 6 0 1 cr 0000 0 oO oO ‘The columns of B, form a statical basis of S. The underlying suberaph ofa typical self-equilibrating stress system (For qy =1) is shown in bold lines, Figure 6.1(b). 'As a second example, consider a portal frame shown in Figure 6.2(a), which is 3 times statically indeterminate,162 Structural Mechanies; Graph and Matrix Methods if Ph (8) A simple frame and its primary structure, at 1 Aye | ir =2 ol et 4 4 2 a4 4 (b) Three self-equlibrating systems, Fig. 6.2 A statically indeterminate frame, ‘This structure is made statically determinate by an imaginary eut at the middle of its beam, The unit value of external load p, and each of the bi-actions q, (i=1,2,3) Jead to the formation of By and B, matrices, in which the two end bending ‘moments (Mj.Mj) of a member are taken as its member forces: and ‘The columns of By form a statical basis of $, and the underlying subgraph of exch self-equiibrating stress system is a eyele, a illustrated in bold lines, Figure 6.2(b). Notice that three self-equifibrating stress systems are formed on a eyele ofa planar frame. In both of the above examples, particular and complementary solutions are obiained from the same basic structure. However, this is not a necessary requirement, as imagined by some authors. In fact @ particular solution is any solution satisfying equilibrium with the applied loads, and a complementary solution is any maximal set of independent sel-equilibrating systems, The latter is | i | i (CHAPTER 6 Matrix Force Method 163 a basis of a vector space over the field of real numbers, known as a complementary solution space, Ret. {78} Using the same basic structure is equivalent fo searching for a cycle basis of a ‘graph, but restricting the search to fundamental cycles only, which is convenient bbut not efficient when the structure is complex or cycle bases with specific properties are needed, ‘As an example, consider @ 3-storey frame as shown in Figure 6.3(a). A eut system as shown in Figure 6.3(b) corresponds to a statical basis containing throo self- cequilibrating stress systems formed on each element of the cycle basis shown in Figure 6.3(b). However, the same particular solution can be employed with a statical basis formed on the eyeles of the basis shown in Figure 6.3(€). Ir ae @ ) © Fig. 63 A 3-siorey frame with different cut systems. AA base structure need not be selected as a determinate one. For a redundant basic structure, one may obtain the necessary data either by analysing i first for the loads and bi-actions qj= 1 (1,2, « ,(S)), or by using existing information, ‘The compatibility equations in the actual structure will now be derived. Using the load-displacement relationship for each member, and collecting them in the diagonal of the unassembled flexibility matrix’ Fy, one can write member distortions as: Ft = Foo Fr Bra 4) Ini fm bl-Frallto | 9) From the principle of virtual work:164 Structural Mechanics; Grapa and Matrix Methods 66) Combining Eq. (6-5) and Eq, (6-6) results in, [i] [Beaten 6 in which vy contains the displacements corresponding to the foree components of and vz denotes the relative displacements of the cuts for the basic structure, Performing the multiplication, ¥o]_[BifxBo BiB, |[P [labs sce} = Yo = (BiFnBo)P + (BF yBi)a. 69) and Ye= BiBy)P + iF )4 6-10) Consider now the compatibility conditions as wre. (6-11) ‘The above equation together with Bq, (6-10) leads to: (B}FB,) "BHF Bo) (12) Substituting in Bq, (6-9 yields Yo =[B)F Bo ~BoFnB,(B}F Bi)” BIR sBolP, 13) and the stress resultant in a structure can be obtained as, 1 =[Bp~B,(B)F,B,)"'BiF,Bolp, (6-14) where G =(B}FgB,) is known asthe lsiiliy matric ofthe structure, For an efficient force method, the matrix G should be: | f (CHAPTER 6 Matrix Force Method, 16s () sparse; (©) wellconditioned;, (©) well-structured, ie, narrowly banded. In order to provide the properties (2) and (b) for G, the structure of By should be carefully designed, since the pattern of Fy, for a given discretization is unchanged; ice, a suitable statical basis should be selected. This problem is treated in different forms. by various methods. In the following, graph-theoretical methods are described for the formation of appropriate statical bases of different types of skeletal structures. The property (c) above has a totally combinatorial nature and is studied in Chapter 8. 6.3 GENERALIZED CYCLE BASES OF A GRAPH In this section, $ is considered to be a connected graph. For 3(S) = aM(S) + bN(S) + e7p{S), the coefficients b and ¢ are assumed to be integer multiples of the coefficient a> 0. Only those coefficients given in Table 2.1 are of interest Definition 1: A subgraph S, is called an elementary subgraph, if it does not contain a subgraph Sj cS, with 7(S}) > 0. A connected rigid subgraph T of S containing all the nodes of S is called a tree, if (T) = 0. For AS) = b(S). 8 tree becomes a tree as defined in Chapter 1 Obviously, a structure whose model is a y-tree is statically determinate when (S) deseribes the degree of static indeterminacy of the structure, The enswing stress resultants can uniquely be determined everywhere in the structure by equilibrium only. Examples of trees for different functions ate shown in Figure 6.4, & (2) Y{S)=3M-3N43.—() (S)=M-2N43._(€) (S}>M-3NE6. Fig. 64 Examples of yttes. Notice that 4(T) = 0 does not guarantee the rigidity of a ree. For example, the ‘graphs depicted in Figure 6.5 both satisfy (T) = 0; however, neither is rigid.166 Structural Mechanies; Graph and Matrix Methods soe (2) fS)-M-2N13, (b) S)-M-3N46, Fig. 65 Structures satisfying y(T)=0 which are not rigid. Definition 2: A member of S—T is called a ychord of T. Definition 3: A removable subgraph S, of a graph 8, is the elementary subgraph for witich Y{S;~ S) = (Sp), i, the removal of S, from S, does not alter its DSI, A ‘yiree of S containing two chosen nodes, which has no removable subgraph, is called a path between these two nodes. (pep LE (2) (S\ralM-NH1),—(b) Y(S)-M-2N43.— (©) {S)J=M-3NH6. Fig. 6.6 Examples of paths. ‘As an example, the graphs shown in Figure 6.6 are y-paths between the specified nodes n, and 0, Definition 4: A connected rigid subgraph of $ with (Cy) ~ a, which has no removable subgraph, is termed a y-cyele of S, The total number of members of Ci, denoted by L(C,), is called the /engrh of C,. Examples of y.cyeles are shown in Figure 6.7, CHAPTER 6 Matrix Force Method. 167 (2) WS}-O(M-N41).—(b) 1{S)-M-2N43. (6) 1{5)-M-3N46, Fig. 6.7. Examples of y-cycles. Definition S: Let m; bea y-chord of T, Then T U m, contains a eyele C, which is defined as a jfimdamental y-cyele of S with respect to T. Using the Intersection ‘Theorem of Chapter 2 (Eq. (2-9), it can easily be shown that, YT Um) = 0+ (at2b+e) - 2bFc) =a, indicating the existence of a yeyele in T Um, For a rigid T, the corresponding, fundamental y-cyele is also rigid, since the addition of an extra member between, the existing nodes of a graph cannot destroy the rigidity. A fundamental y-eyele can be obtained by omitting all the removable subgraphs of TU m, Definition 6: A maximal set of independent cycles of S is defined as a generalized cycle basis (GCB) of S. A maximal set of independent fundamental cyeles, is termed a jimdamental generalized cycle basis of S. The dimension of such a basis is given be n(S) = (Sy, Asan example, a generalized cycle basis of a planar truss is shown in Figure 6.8. Ex P<] Ex bad (b) A generalized cycle basis ofS. Fig. 68 A planar truss S, andthe elements of a GCB of S.168 Structural Mechanics; Graph and Matrix Methods Definition 7. A generalized cycle basis-member incidence matrix © is an 1(S)xM matrix with entries -1, 0 and +1, where c= 1 (or ~ 1) if eycle C, contains positively (or negatively) oriented member m, and cj ~ 0 otherwise, The generalized cycle adjacency matrix is defined as D, which is an 9(S)«n(S) matrix, ‘when undirected 7-cycles are considered; then the negative entries of C become positive 64 MINIMAL AND OPTIMAL GENERALIZED CYCLE BASES A mattix is called sparse if many of its entries are zero, A generalized eycle basis C = {CCC} i called minimal if it corresponds to minimum value of ® LO= Y uc) (6-15) fat Obviously x(C) = L(C) and a minimal GCB can be defined as a basis which corresponds to minimum 4(C). A GCB for which L(C) is near minimum is called a subminimal GCB ofS, AGCB corresponding to maximal sparsity of the GCB adjacency matrix is called ‘an optimal generalized cycle basis of S. If x(CC) does not differ considerably from its minimum value, then the corresponding basis is termed suboptimal. ‘The matrix intersection coefficient «(C) of row i of GCR incidence matrix C is the ‘number of row j such that: (a) je fi#1442,.019). () CAG, #@, ic. there is at least one k such that the column k of both y-eyeles C; and C; (rows i and j) contain non-zero entries Now itean be shown that net xD) = 1S) +2 Yo(C) (6-16) a This relationship shows the correspondence of a GCB incidence matrix C and that net ofits GCB adjacency matrix. In order to minimize 4(D), the value of "'$*o;(C) should be minimized, since 9(S) is a constant for a given structure S; ie. yeyeles with a minimum number of overlaps should be selected. CHAPTER 6 Matrix Force Method. 169 65 PATTERN EQUIVALENCE OF FLEXIBILITY AND CYCLE ADJACENCY MATRICES Matrix By containing a statcal basis, in pattoned form, is pattern equivalent to -, Similarly BfF, By is pattem equivatent to CIC’ or CC’, This correspondence transforms some structural problems associated with the characterization of G = 1B FB, into combinatorial problems of dealing with CC*. ‘As an example, ifa sparse matrix G is required, this can be achieved by increasing the sparsity of CC., Similarly for a banded G, instead of ordering the elements of a statical basis (selequilibrating stress systems), one can order the corresponding cycles. This transformation has many advantages, such as 1) The dimension of CC’ is often smaller than that of G. For example, for a space frame the dimension of CC’ is six-fold and for a planar frame three-fold smaller than that of G. Therefore the optimisation process becomes much simpler when combinatorial properties are used. 2) The entries of C and CC' are elements of Zs and therefore easier to operate on, compared to B, and G which have real numbers as their entries. 3) The advances made in combinatorial mathematies and graph theory become directly applicable to structural problems. 4) A correspondence between algebraic and graph-theoretical methods becomes established. 6.6 MINIMAL GCB OF A GRAPH Theoretically, a minimal GCB ofa graph can be found using the Greedy Algorithm developed for matroids. This will be discussed in Chapter 9 after matroids have been introduced, and here only the algorithm is briefly outlined Consider the graph model of a structure, and select all ofits y-cycles. Order the selected y-eyeles in ascending order of length, Denote these cycles by a set C. Then perform the following steps Step I: Choose a y-eyele C; of the smallest length, Le. L(C;) < L(C%) for all C, eC.170 ‘Structural Mechanics; Graph and Matix Methods Step 2: Select the second yeyele C2 from C and L(C3) $ L(C) for all peyeles of C~ {Cy} {C)} which is independent of C, Step k: Subsequently choose a yeycle C, from C ~ {C,CiyCis) which is independent of Cy, Ca... C1 and L(C,) $ L{C) for all C,€ C= {CsCavee Cor After 1(S) steps, a minimal GCB will be selected by this process, a proof ean be found in Kaveh [94], 6.7 SELECTION OF A SUBMINIMAL GCB: PRACTICAL METHODS In practice, three main difficulties are encountered in an efficient implementation, of the Greedy Algorithm. These difficulties are briefly mentioned in the following: 1, Selection of some of the yeeycles for some 9(S) functions; 2. Formation of all ofthe y-eycles of S; 3, Checking the independence of -eyeles In order to overcame the above difficulties, various methods are developed. The bases selected by these approaches correspond to very sparse GCB adjacency: ‘matrices, although these bases are not always minimal 6.7.1 METHOD 1 This isa natural generalization of the method for finding a fundamental cycle basis ‘of a graph, and consists of the following steps Step 1: Select an arbitrary ree of S, and find its y-chords, Step 2: Add one 7-chord at a time to the selected ree to form fundamental y+ ceycles of S with respect to the selected tree, ‘The main advantage of this method is the fact that the independence of y-eycles is {guaranteed by using a ree. However, the selected ‘cycles are often quite long, corresponding to highly populated CCB adjacency matrices. CHAPTEI 6 Matrix Force Method. 1 6.7.2 METHOD 2 ‘This is an improved version of Method 1, in which a special 7-tree has been employed and cach chord is added to 7tree members after being used for formation of a fundamental y-eycle. Step I: Select the centre "O” of the given graph, Methods for selecting such a node are discussed in Chapter 7. Step 2: Generate a shortest route tree rooted atthe selected node O, and onder its chords aceording to their distance from O. The distance of a member is taken as the sum of the shortest paths between its end nodes and O. Step 3: Form a yeycle on the chord of the smallestistance number, and add the used chord to the tree members, ie. form T Um, ‘Step 4: Form the second y-cyele on the next neatest7-chord to O, by finding a path in T Um (not through m,). Then add the second used 7-chord m; to T Um, obtaining T Um Um, Step 5: Subsequently form the Ath y-eyele on the next unused chord nearest to ©, by finding a path in the TU my U ms U .. Um (not through my). Such a 7 path together with my forms a yyele, Step 6: Repeat Step 5 until n(S) of peyeles are selected Addition of the used j-chords to the 7-ttee members leads to a considerable reduction in the length of the selected y-cycles, while maintaining the simplicity of the independence check. In this algorithm, the use of an SRT, orders the nodes and members of the graph. Such an ordering leads to fairly’ banded member-node incidence matrices Considering the columns corresponding o tree members as independent columns, a bbaso is effectively selected for the cycle matroid of the graph, Kaveh [89,108] 6.73 METHOD 3 ‘This method uses an expansion process, at each step of which one independent 3+ cyele is selected and added to the previously selected ones. The independence is secured using an admissibility condition defined as follows’im ‘Structural Mechanics; Graph and Matrix Methods ‘A poyele Cy. added to the previous selected y-eyeles C called admissible if UU. U Cis AC UG = HC) +a, (6-17) where "a" is the coefficient defined in Table 2.1. The algorithm can now be described as follows: Step 1: Select the frst y-eycle of minimal length C), Step 2: Select the second y-cycle of minimal length C; which is independent of Cy, ie. select the second admissible -cyele of minimal length, Step k: Subsequently find the ih admissible ycycle of minimal length Contin this process until (8) independent yeylesfoming 2 submininal GCB are obtaine A ‘peyele of minimal length can be generated on an arbitrary member by adding a ‘y-path of minimal length between the two end nodes of the member (not through the member itself). The main advantage of this algorithm is avoiding the formation of all yeyeles of $ and also the independence control that becomes feasible by ‘graph-theoretical methods, The above methods are elaborated for specific (8) functions in subsequent sections, and examples are included to illustrate their simplicity and efficiency. 6.8 FORCE METHOD FOR THE ANALYSIS OF RIGID-JOINTED STRUCTURES For this type of skeletal structure, a statical basis can be generated on a cycle basis, of its graph model. The function representing the degree of state indeterminacy, (8), ofa rigidjointed structure is directly related to the first Betti number by(S) of its graph model, ‘{S)= ab (8)= alM(S)-NG)+bo()} (6-18) where = 3 of 6 depending on the structure being a planar or a space frame, respectively. For a frame structure, matrix By can easly be generated using a shortest route tree of its model, and B, can be formed by constructing 4 o 6 S.E.Ss on each element ofa cycle basis ofS, CHAPTER 6 Matrix Force Method. 13 In order to obtain a flexibility matrix of maximal sparsity, a eycle basis ccomesponding to minimum x(D) is needed, ie. an optimal cycle basis should be formed. However, because of the complexity or this problem, most of the research hhas been concentrated on minimal cycle basis selection, except those of Refs [97,98], which minimize the overlaps ofthe eyctes rather than their length only ‘Methods for formation ofa cycle basis can be divided into two groups, namely: (a) topological methods, (b) graph-theoretical approaches. Topological methods useful for the formation of cycle bases by hand, were developed by Henderson and Maunder [78] and Kaveh [89,108], Graph-theoretical ‘methods suitable for computer applications, were developed by Kaveh (93,96, 108]. For completeness of the discussion, both groups of methods are described in this section, The first group, however, requires some terminology from algebraic ‘and combinatorial topology for which the reader may refer to Cooke and Finney {31}; otherwise Section 6.8.1 can be ignored. 68.1 CYCLE BAS! SELECTION: TOPOLOGICAL METHODS Cyele selection procedures described in this section use the concept of embedding. the geometric realization of S into another polyhedron whose dissection has dimension equal to 2. The idea originates from a planar graph embedded in R?, in ‘which the cycles bounding finite regions form an efficient basis (mesh basis) For the first eycle group Z(S,R). ‘The object is to extend this approach to embedding S on polyhedrons and ‘manifolds with certain properties. These properties are measured by using the homology groups H,(K,R) of the underlying complex, which measure, roughly speaking, the number of independent p-dimensional holes of K. In other words, they measure the extent to which K has non-bounding p-cycles. A 2-Dimensional Polyhedron Embedding. Let S be the mathematical model of a ‘ructure, which is a simple graph (I-complex). The underlying polyhed:on ot ‘geometric realization ofS is often denoted by |S), However, for simplicity it is also denoted by S, ‘An embedding f: $— P is a homemorphism ofS into polybedron P. An embedding is called a 2-ell embedding ifthe components of [P-fS)] are all cells. Ifthe - cells are regular, then embedding is called a regular 2-cell embedding, Let f(S) be dissected into a’ L-complex isomorphic to the dissection of S, Then f(S) and the components of [P=f(S)] form a dissection of P into a 2-dimensional cell complex ‘Admissible Embeddings. The cycles bounding the 2-cells of K are known as regional cycles, An admissible embedding f of S is one for which the regional ‘eycles form a set of by(S) independent eyeles from Zi(K,R). The necessary and14 ‘Structural Mechanics; Graph and Matrix Methods sufficient condition for a 2-cell embedding f: $ — P to be admissible is that the homology groups with real coefficients are trivial, In the case of 2-cell embedding, this condition holds when first and second Betti number of K are zero, Using oneepts fom alae topology, the admissibility condition ean be sated follows: ‘The necessary and sufficient condition for : S —» P to be admissible is that the corresponding K be acyclic, ‘A regular complex K is called acyelic if all homology groups of K are trivial, Equivalently, any eyele of Z,(K,R) bounds in K; ie. Z)(K,R) = BYK,R). It is easy to show that « contractible complex is acyclic; hence a contractible embedding is admissible. If K is collapsible, then itis obviously contractible. Thus a collapsible embedding is also admissible, A raph can be viewed as the L-skeleton of a 3-complex. An n-cell is called collapsible if it can be shrunk into the remainder of its n-1 cells through a free n— cell. Ifa 3-complex can be collapsed into a point, then itis called collapsible. A collapsible 3-complex can be used for the formation of a eycle basis of its 1+ skeleton. This can be achieved by collapsing all the 3-cells tough free 2-cells of 2-cells being freed in subsequent steps, Example: Consider a space graph $ as shown in Figure 6.9(@). can be viewed as the L-skeleton of a 3-complex as depicted in Figure 6.9(b). After collapsing all the 3-cells through the shaded 2-cells, the bounding cycles of the remaining 2-complex (81 cycles) form a eyele basis of , (a) A space graph S, Fig. 6.9 A space graph and its collapsible embedding. (b) S embedded on a 3-complex CHAPTER 6 Matrix Force Method. 175 Modified Manifold Embedding: Edmond’ permutation technique provides a method fora 2-coll embedding of Sin an orientable 2-anifold M. The choice of node permutations in this method is made arbitrarily, which may lead to a manifold with an unnecessarily large genus. The genus of graph i the minimum number of handles of a sphere on which the graph can be embedded. In ord to reduce the gems of M, Duke [41] developed a reduction technique for transforming a 2-cel cmbedding M into M¢, where the genus of M* is one less than that of M. Youngs [253] developed an algorithm for minimal embedding by considering all the possible sets of node permutations of S, leading to genus of S. This algorithm is very lengthy and impractical. Kaveh [95] developed a practical method in which the node permutations are determined in the process of embedding S. However, this algorithm doesnot always lead to minimal embeddings. {A different approach may also be employed which is based on an intuitive regular embedding of § on a manifold. For an embedding £ S + P which dissects polyhedron M into 2-complex K, M is a manifold if; 1. Bach I-eell of K is incident wit 2. All the 2-cells and I-cells of K having a particular O-cell and a face can be ‘ordered in a sequence so that the consecutive cells are incident. exactly two 2-cells 3. Kean be oriented so that where 2; is regional eycle. FM is manifold, then b(K) = 1 and b(K) = 2, where is the genus of M, i. M is homeomorphie to a sphere with Y handles. Thus fis not admissible, but can bbe modified by adding 2¥ appropriate fillings and one perforation of order 2 to ‘make it admissible, Example: Consider a hollow box $ embedded on a sphere with two handles, ie. a double torus, as shown in Figure 6.10(a). Modifications are made by four proper fillings and one perforation (P) of order 2, as illustrated schematically in Figure 6.10 (b). (a) Sand its embedding. (b) Schematic illustration of the modified torus. Fig. 6.10 An admissible embedding of .176 Structural Mechanies; Graph and Matix Methods Seventy-nine eycles of length 4 and two cycles of length 8 are selected as a minimal eycle basis of S. However, for multi-member complex structures, this ‘method is by no means practical Example: Let $ be the graph model of a 3-storey frame as shown in Figure 6.11(a). The fitst storey can be embedded on a torus and the third storey on a second torus. Four tubes can be added to accommodate the columns of the second storey. Identifying these through one of the tubes results in a sphere with five hhandles as depicted in Figure 6.11(b), Once the embedding is achieved, ‘modifications can easily be made. In the schematic representation, fillings are shaded and the necessary perforation is denoted by P, Figure 6,1 1(b) In a manifold embedding, the quality of the selected eycle basis depends upon the genus of the manifold on which S is embedded. Thus itis ideal to have a minimal ‘embedding. However, little is known about an efficient approach to carry out sch ‘an embedding (a) Graph model ofa space frame. (b) Schematic illustration ofthe embedding, Fig. 6.11 A modified manifold embedding of a space graph, Embedding $ on a Union of Dises: S can be considered as the union of some planar comected non-separable subgraphs of S. The process of such an embedding is as follows: Step 1: _ Identify a planar subgraph S, and embed it on a dise d, whose dissection Kis isomorphic to S,, (CHAPTER 6 Matrix Force Method. 7 Step 2: The second subgraph S» is identified such thatthe corresponding Ks has 4 2-call with afieeI-face and |K,| Kal is a connected subspace ofthe frontier of 4 Step 3: The process step 2 is continued, and at the ith step Ki is joined to Kj, with K; having a free I-face and Kilo] being a connected subspace of the frontier of d;, Obviously K' is collapsible to K"! 4 ‘The process will be terminated when all Icells of $ are embedded in K9= U Ky which is collapsible, since: K=KIGKT SKIS. OKT=K, Example: Consider a space frame $ as illustrated in Figure 6.12. This model is embedded on three discs K,-Ks as shown in the Figure, resulting in a cycle basis consisting of 56 regional cycles of length 4 It is ideal fo embed S in a complex K with a minimum number of dises for all possible collapsible embeddings. This number is known as the thickness of the raph S, denoted by 1(S), Ref. [178]. Hence an embedding should be performed with q as the thickness of S, Again only partial results are available about the thickness of the graphs. Any approach to embedding on a smaller number of dises ‘would be advantageous for reducing the overlaps of the cycles. 8 Ky AR Sy” MURS KE TIY K K Fig. 6.12 A space graph and the identified dises K,-178 Structural Mechanics; Graph and Matrix Methods Remark: The computer implementation of the methods of this section for selecting a eyele basis may become uneconomical from engineering point of vi hhowever, the study of these methods provides a firm background for designing ‘many graph-theoretical approaches. The visualization of these methods helps to develop algorithms having different useful characteristics, As an example, the expansion process of the next section may be viewed as an embedding of a graph on the elements of a dise space, with dises having certain properties. Similarly, one design an algorithm for embedding a graph on the elements of a ball space, with balls having specified properties (the reverse process ofa collapsible embedding) 6.8.2 CYCLE BASES SELECTION: GRAPH-THEORETICAL METHODS Cycle bases of graphs have many applications in various fields of engineering, The amount of work in these applications depends on the eycle basis chosen. A. basis with shorter cycles reduces the time and storage required for some applications; i. it is ideal to select a minimal eyele basis, and for some other applications minimal overlaps of eycles are needed; ie. optimal eycle bases are preferred. In this seetion, the formation of minimal and subminimal eycle bases is first discussed. Then the possibility of selecting optimal and suboptimal eycle bases is investigated Minimal eyele bases were considered first by Stepanee [225] and improved by Zykov [257]. Many practical algorithms for selecting subminimal eycle bases have been developed by Kaveh (93) and Cassell etal. (21) Similar methods have been presented by Hubicka and Syslo (83] claiming the formation of a minimal cycle basis ofa graph, Kolasiska [140] found a counter example to Hubicka and Syslo’s algorithm. A siilar conjecture was made by Kaveh [93] for planar graphs; however, a counter example has been given by Kaveh and Roosta (114. Horton [82] presented a polynomial time algorithm to find minimal eycle bases of graph Kaveh and Rahami used an algebraic graph-theoretical approach (133) In this section, the merits of the algorithms developed by different authors are discussed; a method is given for selection of minimal eycle bases, and efficient approaches are presented for the generation of subminimal cycle bases, Formation of A Minimal Cycle on a Member: A minimal length eyele C, on a member’ m, called its generator, ean be formed by using the shortest route tree algorithm as follows; Start the formation of two SRTs rooted at the two end nodes n, and n, of tp and terminate the process as soon as the SRTS intersect each other (not through m, itself) at say n,. The shortest paths between n, and n., and ny and n, together with nm, form a minimal eycle C, on m, Using this algorithm, eyeles of prescribed lengths can also be generated, ‘As an example, C; is @ minimal eyele on m, in Figure 6.13. The SRT$ are shown in bold lines. Notice that the generation of SRTs is terminated as soon as n, has been found. CHAPTER 6 Matrix Force Method. 179 Fig. 6.13 A minimal eycle on a member. ‘A minimal cycle on a member my passing through a specified node n, can similarly be generated. An SRT rooted at nis formed and as soon as it hits the end nodes of ny, the shortest paths are found by backtracking between ny and n, and ny and ny These paths together with m, form the required cycle. As an example, a minimal cyele on my containing ny is illustrated by dashed lines, Figure 6.13, Different Cycle Sets for Selecting a Cycle Basis: It is obvious that a general cyele can be decomposed into its simple eycles. Therefore itis natural to confine the considered set to only simple cycles of S. Even such a cycle set, which forms a subspace of the cycle space of the graph, has many elements and is therefore ‘uneconomical for pructieal purposes df -Ci) Fig. 6.14 A graph S and selected eycles. In order to overcome the above difficulty, Kaveh [93] used an expansion process, selecting the smallest admissible (independent with additional restriction) cycles, fone at a time, until b(S) eycles forming a basis had been obtained. In this approach, a very limited number of eycles were checked for being an element of a basis. As an example, the expansion process for selecting a cycle basis of S is illustrated in Figure 6.14 Hubicka and Syslo [83] employed a similar approach, without the restriction of selecting one eycle at each slep of expansion. In their method, when a cycle has been added to the previously selected cycles, increasing the first Betti number of the expanded part by "p", then p created cycles have been formed. As an example,180 Structural Mechanics; Graph and Matrix Methods in this method, Steps 4 and 5 will be combined into a single step, and addition of cycle 5 will require immediate formation of the cycle 4. The above method is modified, and an efficient algorithm is developed for the formation of cycle bases by Kaveh and Roosta (114), Finally, Horton [82] proved that the elements of a minimal cycle basis lie in between a eyele set consisting of the minimal cycles on each member of S which ‘passes through each node of S, i.e. each member is taken in turn and all cycles of ‘minimal length on such a member passing through all the nodes of $ are generated. Obviously, M(S)M(S) such cycles will be generated. Independence Control: Each cycle of a graph can be considered as a column vector of its cycle-member incidence matrix. An algebraic method such as Gaussian elimination may then be used for checking the independence of a cycle with respect to the previously selected sub-basis. However, although this method is general and reduces the order dependency of the cycle selection algorithms, like ‘many other algebraic approaches its application requires a considerable amount of storage space. ‘The most natural graph-theretical approach is to employ a spanning tree of S, and form its fundamental eycles, This method is very simple; however, in general its use leads to long eycles. The method can be improved by allowing the inclusion of ‘each used chord in the branch set of the selected tree. Further reduction in length ‘may be achieved by generating an SRT from a centre node of a graph, and the use of ls chords in ascending order of distance from the centre node, Kaveh [89] A third method, which is also graph-theretical, consists of using admissible cycles. ‘Consider the expansion process, Ge ssc's,, sch =s, k Cr. A eyele Ci called an admissible eyel, if for C* where C cu Cos bie =b(CU Gu) =b(C)+1 (6-19) It can easly be proved that the above admissibility condition is satisfied if any of the following conditions bold: (1) Ay = Ch Cus =, where @ is an empty intersection, @) bu =1—5, where ands are the numbers of components of C* and C¥, respectively. he CHAPTER 6 Matrix Force Method. 181 8) B (Ass) =O when CE and Care connected (rs) In the above relations, 5 (A) = Wij -Nj-+ 1, where Mand Rare the mambers respectively. ‘of members and nodes of ‘As an example, the sequence of eycle selection in Figure 6.15 are as specified by their numbers. Fig, 6.18. A cycle and its bounded cycles. A different approach, is suggested by Hubicka and Syslo, in which, bic) = d(C +B. (6-20) js considered to be permissible. However, a completion is performed for p> 1. As tn example, when Cy is added to C its first Betti number is increased by 3 and therefore eyeles C; and Cy must also be selooted at that stage, before further expansion. Having discussed the mathematical concepts involved in a cycle basis selection, three different algorithms are now deseribed Algorithm 1 (Kaveh 1974) ‘Step 1: Select a pseudo-centre node of maximal degree O. Such node can be selected manually or automatically using the graph or algebraic graph-theretical ‘methods discussed in Chapter 5. Step 2: Generate an SRT rooted at O, form the set ofits chords and order them according to thei distance from O. Step 3: Form one minimal eyele on each chord in turn, starting with the chord nearest to the root node. A corresponding simple path is chosen which containsIs2 Structural Mechanics; Graph and Matrix Methods members of the tree and the previously used chords, hence providing the admissibility of the selected cycle, This method selects subminimal cycle bases using the chords of an SRT. The nodes and members of the tree and consequently the cycles are pattially ordered according to their distance fiom O. This is the combinatorial version of the Turn- back method to be discussed in Section 6.10.1 Algorithm 2 (Kaveh 1974) Slep I: Select a centre or pseudo-centre node of maximal degree O. Step 2: Use any member incident with © as the generator of the first minimal cycle. Take any member not used in Cy and incident with O, and generate on it the second minimal eycle. Continue this process until all the members incident with O are used as the members of the selected cycles. The cycles selected so far are admissible, since the intersection of each eycle with the previously selected cycles isa simple path (or a single node) resulting in an increase of the Betti number by unity for each eyele Step 3: Choose a starting node 0’ , adjacent to O, which has the highest degree, Repeat a step similar to Step 2, testing each selected eyele for admissibility. If the cycle formed on a generator m, fails the test, then examine the other minimal eyeles on mi eny seh eyeleexins no aisle minal epee canbe found fon my, then! (@) Form admissible minimal eyetes on the other members incident with OF 1m does not belong to one ofthese subsequent cycles, thea (ii) Search for an admissible minimal eycte on m, since the formation of eycles on other previous members may now have altered the admissibility of this eyele. If no such cycle can be found, leave m, unused. In this step more than one member may be left unused, Step 4: Repeat Step 3 using as starting nodes a node adjacent to O andor 0, having the hihest degree, Continue the formation of eyes until all the nodes of S have been tested for cycle selection. Ill the members have not becn used, sclect the shortest admissible yele available for an umsed member ss generator. Then test the minital eyeles on the other unused members in ease the formation ofthe longer cycle has altered the admissibility. Bach time a minimal eycle is found to be admissible, add to C’ and test all the minimal eycles on the other unused members again. Repeat this process, forming other shortest admissible cycles on unused members as generators, until S is re-formed and a subminimal eyele basis has been obiained 7 CHAPTER 6 Matrix Force Method. 183 ‘Both of the above two algorithms are ordet- 29. . Example 2: In this example, $ is the model of a space fame, considered as Sis where a typical S; is depicted in Figure 6.17(a). For §, there are 12 ‘members joining 8 comer nodes, and a central node joined to these comer nodes. CHAPTER 6 Matrix Force Method. 185 ‘The model S is shown in Figure 6.17(b), in which some of the members are ‘omnitted for clarity in the diagram. For this graph, by(S) = 270. (@) Atypical $,(1,...27) — (b) S with some omitted members, Fig. 6.17 A space frame S, ‘The selected eyele basis using any of the algorithms consists of 270 cycles of length 3, forming a minimal eyele basis of S. For Algorithm 3, the use of different starting nodes leads to a minimal cycle basis, showing the capability of this method. Example 3: $ is a planar graph with b(S)=9, as shown in Figure 6.18. The application of Algorithm 3 results in the formation of a eycle of length 3 followed by the selection of five cycles of length 4. Then member {1,6} is used as the generator of a six-sided cycle C; = (1,2,3.4,5,6,1). Member {2,10} is. then employed to form a seven-sided cycle Cy (2,11,12,13,14,15,10,2), followed by the selection ofa five-sided cycle C, = (10,5,4,3,2,10). The selected cycle basis has ‘total length of L(C)=41, and is not a minimal eycle basis. A shorter cycle basis can be found by Algorithm 4 consisting of one three-sided and five four-sided cycles, together with eycles, y= (1,2,10,5,6,1), Ce= (2,3.4,8,10,2) and Cy= (2,11,12,13,14,15,10,2), forming a basis with the total length of 40. However, the computation time and storage for Algorithm 3 is far less than that of Algorithm 4, as compared in Ref. (ud)186 Structural Mechanies; Graph and Matrix Methods Fig. 6.18 A planar graph 8, 682.1 SUBOPTIMAL CYCLE BASES ; A DIRECT APPROACH. Definition 1: An elementary contraction of graph S is obtained by replacing a path containing all odes of degice 2 with a new member. A grph S contacted 0 a graph S’ is obtained by a sequence of elementary contraction. Since in each clementary contraction, k nodes and k members are reduced the first Bett number does not change in a contraction, ie. b(S) = b(S'). As an example, S is contracted to Sin Figure 6.19, RIE @s ws Fig. 6.19. S and its contracted graph S’ ‘This operation is performed in order to reduce the size of the graph and also because the number of members in an intersection of two cycles is unimportant; a sige members enough to render Cr; nonempty, and hence to prodiceh non zero entry in CC, " a Definition 2: Consider a member m, of a graph S. On this member, p- minimal cycles of length q can be generated, p is called the incidence number and q is defined as the eyele lenge number of m,, respectively. In fact p and q are measutes (CHAPTER 6 Matrix Force Method. 187 assigned to a member to indicate its potential as a member in the elements of a ‘eycle basis, In the process of expansion for eyele selection, an atificial increase in P results in the exclusion of this element from a minimal cycle, keeping the number of overlaps as small as possible, Space graphs need special treatment. For these graphs, when a member has p-1, then the next shortest-length eycles with q’= get (1 beng the next smallest possible integer) aro also considered. Denoting the number of such eyeles byp, the incidence number and cycle length nuraber for this type of member are taken as, Ty =ptt and T= (a+p)/(I+p), (6-21) respectively, The end nodes of the considered member aro j and k. Definition 3: ‘The weight of a cycle is defined as the sum of the incidence numbers ofits members Algorithm A. Step 1: Contract $ into’, and calculate the incidence number (IN) and eyete length number (CLN) ofall is members. ‘Step 2: Start with a member ofthe least CLN and generate a minimal weight eyele fon this member. For members with equal CLNs, the one with the smallest IN should be selected. A member with these two properties will be referred to as "a ‘member of the least CLN with the smallest IN". Step 3: On the next unused member of the least CLN with the smallest IN, ‘generate an admissible minimal weight cycle. In the case when a cycle of minimal ‘weight is rejected due to inadmissiblity, the next unused member should be considered, This process is continued as far as the generation of admissible ‘minimal weight eycles is possible. After a member has been used as many times as its IN, before each extra usage, increase the IN of such a member by unity, Step 4: On an unused member ofthe least CLN with the smallest IN, generate one admissible eyele of the smallest weight. This cycle is not a minimal weight eycle, cotherwise it would have been selected at Step 3. Such a cyele is called a subminimal weight cycle. Again, update the incidence numbers for each extra usage. Now repeat Step 3, since the formation of the new subminimal weight cycle may have altered the admissibility condition of the other cycles, and selection of further minimal weight eyeles may now have become possible Step 5: Repeat Step 4, selecting admissible minimal and subminimal weight cycles, until b)(S’) ofthese eycles are generated.188 ‘Structural Mechanics; Graph and Matrix Methods Step 6; A reverse process to that of the contraction of Step 1, transforms the selected eyeles of into those of S. ‘This algorithm leads to the formation of a suboptimal cycle basis, and for many models encountered in practice, the selected bases have been optimal 68.22 SUBOPTIMAL CYCLE BASES ; AN INDIRECT APPROACH Definition 1: The weight ofa member in the following algorithm is taken as the sum of the degrees ofits end nodes. Algorithm B Step 1: Order the members of $ in ascending order of weight. In all the subsequent steps use this ordered member set, Step 2: Generate as many admissible cycles of length «as possible, where cis the length of the shortest cycle of S. Denote the union of the solected cycles by C", When otis not specified, use the value a= 3. Step 3: Select an admissible eycle of length ce+1 on an unused member (use the ‘ordered member set). Once such a cycle Cay, is found, control the other unused ‘members for possible admissible cycles of length a. Again select an admissible ceycle of length ot+1 followed by the formation of possible o-sided cycles, This process is repeated until no admissible eyeles of length a and +1 exn be found. Denote the generated cycles by C. Step 4: Select an admissible eyele C,,; of length ct+2 on an unused member. Then check the unused members for the formation of acsided eyeles. Repeat Step 2 until ro eyele of length « or a+1 can be generated. Repeat Step 3 until no cycles of Tength «, cs+1 or 042 can be found. Step 5: Take an unused member and generate an admissible eyele of minimal length on this member. Repeat Steps 1, 2 and 3. Step 6: Repeat steps similar to that of Step 4 until b,(S) admissible cycles, forming ‘a suboptimal eycle basis, are generated, Using the ordered member set affects the selection process in two ways: CHAPTER 6 Matrix Force Method. 189 (0) Generators are selected in ascending weight order, hence increasing the possibility of forming eyeles fiom the dense part of the graph, This inereases the chance of cycles with smaller overlaps being selected. (2), From cyeles of equal length formed on a generator, the one with smallest total ‘Weight (sum of the weights ofthe members ofa cycle) is selected. ‘The eyele bases generated by this algorithm sre suboptimal; however, the results are slightly inferior to those of the direct method A. 68.2.3 EXAMPLES In this section, examples of planar and space frames are studied. The eyele bases selected by Algorithms A and B are compared with those developed for generating ‘minimal eycle bases (Algorithms 1-4). Simple examples are chosen, in order 10 illustrate clearly the process of the methods presented. The models, however, can bbe extended to those containing a greater number of members and nodes of high degree, to show the considerable improvements tothe sparsity of matrix D. Example 1: Consider $ 8 the graph model of a space frame with by(S) = 12, as shown in Figure 6.20. Hence, 12 independent cycles should be selected as a basis. Algorithm 3 selects a minimal cycle basis containing the following cycles: 1.2.3), C=(12,5), C9(1 3.4), CUS), Co(23,6.7), BAT, CHES89), C6789), CH(81112), CGT, Cr=(9812,13), C(10,11,12,13), “which corresponds to: lO) = 43+8x4 = 44, and (D) = 124223 = 58. (a) A space structure. (b) The graph model S of the structure Fig. 6.20 A space frame, and CLNs and INs ofits members190 ‘Structural Mechanics; Graph and Matrix Methods Using Algorithm A leads o the formation ofa simitar basis with the difference that €4=(6,9,10,13) is generated in place of C3=(6,7,8.9, which leads to: X(C!)= 433484 = 44, X(D')= 1242420 = 52, ‘The CLNs and INs of the members used in this algorithm are illustrated in Figure 6.204a), Example 2: In this example S is a space structure with b(S) = 33, as show in Figure 6.21(a). Both Algorithms 3 and A, select 33 cycles of length 4, ve. a ainimal eyete basis with 7(C) = 4533-132 ig obtained, The basis selected by Algorithm 3, contains (inthe worst case) all 3-sided cycles of $, exeept those ‘which ae shaded in Figure 6.21(@), with 4(D) = 233, Algorithm A selects all 3-sided eycles of S except those shaded in Figure 6.21(b), with 1(D)=190, It will be noticed that, for structures containing nodes of higher degrees, considerable improvement is obtained by the use of Algorithm A. (@) A toinimal eyele bs Fi (b) A suboptimal eyele basis, 6.221 Minimal and suboptimal eyele bases of Example 3: Consider a space frame as shown in Figure 6.22, for which b\(S)}-10. ‘The minimal eyete basis selected by Algorithm 3, consists of, CH (1.23), Cm 45,6), C= (7,8,9), Cyr (10,11,12), C= (1.25.4), Ce = (458.0, C, Ce 7.8,11,10), Crr=(8,9,12,11), ‘corresponding to 1(C) ~ 4x3+6x4 = 36 and 1(D) = 10+2[040+0+2+343+ 41344) = 10x19 = 48, CHAPTER 6 Matrix Force Method. 191 However, the following non-minimal eycle basis has a higher sparsity for C, and leads to @ more sparse D, 3,64), Ci 2.3), C= (125.0), C= 23,65), Cr C= AS8,7), Co (5,6,9,8), Cr™ (4,6,9,1), C= (78,1110), C= (8,9,12,11), Cie (10,11, 12), 2xGe8x4 = 38 comesponding, to x(D') = 10+2[142+3 OND T= 44, for which 4(C’) #4283619249] ‘The above cycle basis can further be improved by selecting the following basis: C= (12,3), C= (L245), CH 235.0), CH 45.0), CF 46.7.9), C= (5.68.9), C= (7,8,9), Com (7,8,10,11), C= (7,9,10,12), Cos (10,11,12), for which 4(C’) = 2x3+8x4 = 38 comesponding to 4(D’) = 10+2[14242 124214942] = 104216 = 42, Fig. 6.22. A space fiame S, ‘Therefore, the idea of having an optimal cycle basis in between minimal eyele bases is false.192 ‘Structural Mechanics; Graph and Matrix Methods 68.3 FORMATION OF By AND B, MATRICES In order to generate the elements of a By matrix, a basie structure of $ should be selected. For this purpose « spanning forest consisting of NG(S) SRTS is used, ‘where NG(S) is the mmber of ground (support) nodes of S. As an example, for S shown in Figure 6.23(a), two SR subtrees are generated, Figure 6.23(b) @ ) Fig. 623. S and two ofits SR subtrees, ‘The orientation assigned to each member ofS is from the lower numbered node (0 its higher numbered end. For each SR subtree, the orientation is given in the direction ofits grow from its support node. MATRIX Bo: This is a 6M(S)<6NL(S) matrix, where M(S) and NL(S) are the ‘numbers of members and loaded nodes of S, respectively. Ifall the free nodes are loaded, then NL(S) =N(S) ~ NG(S), where NG(S) isthe number of support nodes. For a member, the intemal forces are represented by the components at the lower ‘numbered end. Obviously the components at the other end can be obtained by ‘considering the equilibrium of the member ‘The coefficients of By can be obtained by considering the transformation of each joint oad to the ground node ofthe corresponding subtree. [Bo] for member j and ode jis piven by a 6x6 submatrix as, Se aoe od Boi Go Boba e | ooh 2 0-010 CHAPTER 6 Matrix Force Method. 193 in which Ax, Ay and Az-are the differences ofthe coordinates of node j with respect to the lower numbered end of member i, inthe selected global coordinate system, and A, isthe orientation coefficient defined as: +1if member is positively oriented in the tree containing j, ‘45 = {-Lif members nogatively oriented inthe tree containing j, 0 if members not inthe tree containing, node j ‘The By matrix can be obtained by assembling the [By] submatrices as shown schematically in te following: ae a (623) 14s 6NLS) MATRIX By: This is a 6M(S)x6b,(S) matrix, which can be formed using the clements of a selected cycle basis. For a space structure, six self-equilibrating ‘stress systems can be formed on each cycle. Consider C; and take a member of this cycle as its genorator. Cut the generator in the neighbourhood of its beginning node and apply six bi-actions as illustrated in Figure 6.24, Fig. 6.24 A cycle and the considered bi-action ata cut. ‘The internal forces under the application of each bi-action are a self-equilibrating stress system As for the matrix By, a submatrix [Bj}j of By is a 6*6 submatrix, the94 ‘Structural Mechanics; Graph and Mattix Methods columns of which show the intemal forces atthe lower numbered end of member i under the application of six bi-actions at the cut of the generator j o 00 o 00 O00 f 000 oo Lo OL {81-6 |, 624 -an ay 1 dz 0 -Ax 0 nay Ax 00 in which Ax, Ay and Az are the differences of the coordinates x, y and z of the beginning node of the generator j and the beginning node of the member i, The orientation coefficient Ay is defined as: + 1if member ihas the same orientation of the eycle generated on j, 1 iff member i has the reverse orientation of the cycle generated on j, 0. if member iis notin the eyele whose generators j ‘The patter of B, containing (B,} submalrices is shown in the following: ne a, (625) asd) ‘Subroutines for the formation of By and B, matrices are included in the program presented in Ret. (112 Example 1: A four by four planar frame is considered as shown in Figure 6.25 | | CHAPTER 6 Matrix Force Method 195 Fig. 628. A four by four planar frame S. ‘The patterns of By, B, and B{B, formed on selected SRT and the elements of the cycle basis selected by any of the methods of the previous section are depicted in Figure 3.26, corresponding to x (B,) = 241 and 7, (B{B, )~388, Sens page guess Rees ()Patern of By (b) Pattern of By. (6) Pattern of BB, Fig, 6.26 Patterns of By By and BjB, matrices for. Example 2: A one-bay tree-storey frame is considered as shown in Figure 6.27196 ‘Structural Mechanics; Graph and Matrix Methods Fig, 6.27. A simple space frame S. “The pattems of fh and BIB} matrices formed on the elements of the eycle basis selected by any ofthe graph-heretcl algorithms ofthe previous section are shown in Figure 628, coresponing to 7 (By) = 478, (B;) = 310 and BIB) =562. —— Ay RSs. XJ Ae Sy RAN SQ [oe (a) Pattern of By. (b) Pattern of B,. (©) Pattern of BiB, Fig, 628 Paitems of By, By and BIB, matrices for. ‘The rest ofthe operations are routine workin the marx analysis of structures, The interested reader may rofer to standard textbooks such as those of MeGuire and Gallagher (172), Prezmienecki [197] or Mek [173] (CHAPTER 6 Matrix Force Method 197 6.9 FORCE METHOD FOR THE ANALYSIS. OF PIN-JOINTED PLANAR TRUSSES ‘The methods described in Section 6.8 are applicable tothe selection of generali cycle bases for different types of skeletal structures. However, the use of these algorithms for trusses engenders some problems, which are discussed in Ref. [108] In this section, two methods are developed for selecting suitable GCBs for planar trusses. In both methods, special graphs are constructed for the original graph model $ of a truss, containing all the connectivity properties required for selecting, 4 suboptimal GCB of S. 69.1 ASSOCIATE GRAPHS FOR SELECTION OF A SUBOPTIMAL GCB Let § be the model of a planar truss with triangulated panels, as shown in Figure 6.29. The associate graph of S, denoted by A(S), i a graph whose nodes are in a one-to-one correspondence with triangular panels of S, and two nodes of A(S) are connected by a member ifthe corresponding panels have a common member in S, LAIN, Nec PRR ig, 6.29 A planar truss S and its associate graph A(S), IES has some cut-outs, as shown in Figure 6.30, then its associate graph can still be formed, provided that each cut-out is surrounded by triangulated panels For trusses containing adjacent cutouts, a cut-out with eutsnodes in its boundary, ‘or any other form violating the above-mentioned condition, extra members can be added to S. The effect of such members should then be included in the process of generating its sel-equilibrating stress systems. aaa) LL Peto Aes oe peppers A] Fig. 630. S with two cut-outs and its A(S) PN198 ‘Structural Mechanics; Graph and Mattix Methods ‘Theorem A: Fora fully triangulated truss (except forthe exterior boundary), as Figure 6.30, the dimension of a statical basis (S) is equal to the number of its infernal nodes, which is the same as the first Betti number of its associate graph, 48) =N(S) = BIAS). (6-26) Proof: Let M’ and N’ be the numbers of members and nodes of A(S), respectively. By defini = R(S) =I, and M’=M,(S) = M(S)-M,(S) = M(S)-N.) = M(S)-[N(S) -N; (5). ‘Thus: by{A(S)]= MN“ = M(S)~[N(S)-N(S)}-R(S)HTHT = 2 RGS)IM(S) -N(S}+ NUS). By Euler's polyhedron formula, we have: 2-R(S) + M(S)-N(S)=0. ‘Therefore bLA(S)] = NG). For trusses which are not fully triangulated, as described in Chapter 2, we have: 48) = NAS) MAS). (627) A Cycle of A(S) and the Corresponding Cycle of S: In Figure 6.31(a), @ (riangulated truss and its associate graph, which isa eycl, are shown for which 8) = Nw L= BLAS) Since C, of A(S) corresponds to one ycycle of S, it is called a pe J cycle, denoted by C;. Typical z-eycles of S are shown by continuous lines, and their 7 ‘chords are depicted in dashed lines. (CHAPTER 6 Matrix Force Method. 19 (a) A type C, eye, (b) A type Cu eyele. Fig. 6.31 Two different types of eyele. Figure 6.38(b) shows a truss unit with one cut-out. In general, if cut-out is an m- ‘gone, then the completion of the triangulation requires m3 members. Instead, m internal nodes will be ereated, increasing the DSI by m, Hence Eq, (6-27) yields, yS=m-(m-3)=3, bias) However, in this case, $ contains three y-cycles. A path P and three y-chords (dashed lines) are depicted in Figure 6.31(b). Obviously, PUm, (i=1,2,3) form three 7peycles which correspond to a eyele of type Cyy of A(S). Thus two types of eyeles C; and Cy should be recognized in A(S) and an appropriate number of cyeles will then be generated, Algorithm AA Step 1: Construct the associate graph A(S) of . Step 2: Select a mesh basis of A(S), using an appropriate cycle selection algorithm, For fully triangulated S, Algorithms 1-3 generate cycle bases with 3-sided elements ‘Step 3: Select the y-eycles of S corresponding to the cycles of A(S). One ¥- cycle for each eycle of type Ci, and three yeycles for each cycle of type Ch, should be chosen. Once a GCB is selected, on each y-cyele one selfcequilibrating stress system can easily be formed. Therefore, a statical basis with localized self-equilibrating stress systems will be obtained,200, Structural Mechanies; Graph and Matrix Methods Example: Let § be the graph model ofa planar truss, as shown in Figure 6,30, for which (S)=12, For A(S), six eyeles of length 6 of type C, and two cycles of lengths 18 and 26 of type Cy are selected. Therefore, the total of 6 + 3x2 = 12 cycles of $ are obtained. On each -cyele one self-equilibrating stress system is constructed, and a statical basis consisting of localized self-equilibrating stress systems is thus obtained, 69.2. ABIPARTITE GRAPH FOR SELECTION OF A SUBOPTIMAL GCB. Let § be & planar truss consisting of rectangular panels with or without diagonal members, an example of which is shown in Figure 6.32(a). The bipartite graph B(S) of Sis constructed as follows: “Associate one with each row of panels and denote them by f, 13.» fy Similarly, ‘with each column of panels, associate one node and denote them by €y,€3, =. 4 illustrated in Figure 6.32(b). Connect r,t ¢; ifthe corresponding panel in S has ‘a diagonal member. The graph obtained in this manner is ealled the bipartite graph BS) of. @ © Fig. 6.32 A graph S and its bipartite graph B(S), ‘Theorem B: The degree of statical indeterminacy of a planar truss $, is the same as the first Betti number of its bipartite graph B(S), ic. 7()= bi [B)} (6-28) Proof; For a truss of rectangular shape with rectangular panels and without diagonal members, the number of members M and nodes N are as: Mi= [44301 (m-1f+ fat) +2mn-+n, N= (m+1)(o+1) CHAPTER 6 Matrix Force Method. 201 For M’ extra diagonal members, M is increased by M(B(S)), which isthe same as the number of members of B(S). By definition, the number of nodes of B(S) is N(B(S) and M(S) = m+2mntn¥¥M(B(S) = N= mnémént Therefore: (8)=M(S)-2NG)+3 = m-+2ma+n+M(B())~2(aun-+m-+n41)+3 = M(BIS))- N(BS)}+1= b, [BS A Cycle of B(S) and the Corresponding 7-Cyele of S: A cycle of B(S) corresponds to a subgraph of $ which contains a-7-cyele, In Figure 6.32, a cycle of BS) and its corresponding subgraph containing a y-cycle of S are illustrated in ‘bold lines. For this formation, r; and rare specified a8 Tain aNd fp. All cs which hhave links with two 138 with fg © 14 5 fw should be considered in the formation ‘of the corresponding 7-cycle, For this eycle, only e, and c; have such properties. FS situated in row 1, and rp, and columns ¢, and ¢ are le. For simplicity, a cycle and its 7-eycle ate denoted ‘Therefore, the subgraph those which contain 2 7 by, Cilasmsere2) and 7-Ci(q.r,¢1,¢2), (6-29) and for Cy(tstayse4), we have y—C)(0,t4€1,02,63,¢4) ‘This means that a y-eycle can be formed on a subgraph of S, which contains row 3 and 4, and columns 1, 2, 3 and 4 of S, General Cass: For an eglaity in the patter of the boundary, as tha of Figure 633, the above metod can sill be used, The additions! part contains M = 2N members, leaving (8) = M(S) ~ 2N(S)'3 unchanged. On the other hand, M(B(S) and N(B(S) ae not affected. Hence b B(S)] = M (B(S)) - N(B(S) +1 is unaltered, and y(S) = b,[B(S)] still holds,202 Structural Mechanics; Graph and Matrix Methods Fig. 6.33 A truss model S with irregular boundary Algorithm BB Step 1: Construct the bipartite graph B(S) of S Step 2: Select a subminial cycle basis of Bi ) Step 3: Find the subgraphs of $ corresponding to the cycle of the selected basis oF BGS) at Step 2 Now one S.E.S. can be formed on each subgraph of S containing a y-eyele of S. For localizing the S.E.Ss, the cycle for which lfuin — taal + [ema ~ Gn as the smallest value should be chosen. The value [tmx ~ Cas! can be Obtained once Fi and tyiy are determined. Due to the simplicity of the structure of B(S), the formation of eycle of short lengths leading y-oyeles of S becomes almost trivial Example: Let S be the graph model of a planar truss, as shown in Figure 6.32(a), ‘The bipartite graph B(S) of S is constructed as in Figure 6.32(b). For this structure, r8)=Ni(8)-M,18) and six y-cyetes are selected. The following eycles of B(S) and the corresponding ‘subgraphs of § containing the elements of a GCB are obtained: Cilirascrsee) y-Clasts.ere: Calnorrercs) y-Calnsmerer,ca.e) Cslrstenea) y- Calne c1.02,€3.04) CHAPTER 6 Matrix Force Method. 203 Caltsstyere4) 1-Caleytarere2.03.04) Cslastsenes) 1-Cslrstrseare1se2.03,04.05) Colnotsseares) 1-Calhotase3,€3,64505) Finally, it should be noted that Algorithm AA is more suitable for triangulated trusses, with or without cut-outs, while Algorithm BB is more appropriate for tnusses with rectangular panels, with or without diagonal members, 6.10 FORCE METHOD OF ANALYSIS (OR GENERAL STRUCTURES Combinatorial methods for the force method of structural analysis have been presented in previous sections. These methods are very efficient for skeletal structures, and in particular for rigid-jointed fiames. For a general structure, the underlying graph or hypergraph of a self-equilibrating stress system has not yet bbeen properly defined, and much research has still to be done. Algebraic methods, ‘on the other hand, can be formulated in a more general form to cover different types of structures, such as skeletal structures and finite element models. The main drawbacks of these methods are the larger storage requirements, and the higher number of operations. ‘These difficulties can partially be overcome by employing combinatorial approaches within the algebraic methods, whenever such tools are available, and their use ean lead to some simplifications. 6.10.1 ALGEBRAIC METHODS Consider a diserete or diseretized structure S, which is assumed to be statically indeterminate, Let r denote the m-limensional vector of generalized independent clement (member) forees, and p the n-vector of nodal loads. ‘The equilibrium conditions of the structure can then be expressed as, Ar=p, (630) where A is an mem equilibrium matrix, The structure is assumed to be rigid, and therefore A has a full rank, ie. t=m—n> O and rank A =n.204 ‘Structural Mechanics; Graph and Matrix Methods ‘The member forces can be written as r= Bip} Bag, (631) where By is an nmn matrix such that ABy is an mn identity matrix, and By is an ‘mxt matrix such that AB, is an nxt zero matrix. B, and By always exist for a structure, and in fact many of them can be found for a structure, By is called a self siress matrix a8 well as null basis matrix, Each column of By is known as a null vector. Notice that the null space, null basis and null vectors correspond to complementary solution space, statical basis and self-equilibrating stress systems, respectively, when $ is taken as a general structure. Minimizing the potential energy requires that r minimize the quadratie form, Ley Arf, (632) subject to the constraint as in Eq, (6-30). Fx is an mem block diagonal element flexibility matrix. Using Eq, (6-31), it ean be seen that q must satisfy the following equation, (B/ FB.) ~BY F,Bop, (6-33) where B] FB, © G is the overall flexibility matrix of the structure. Computing the redundant forces q from Bq, (6-33), r ean be found using Eq. (6-31). The structure of G, is again important, and its sparsity, bandwidth and conditioning govern the efficiency of the force method. For the sparsity of G one can search for a sparse By tmatrix, which soften referred to asthe sparse mull basis problem Many algorithms exist for computing a null basis B, of a matrix A. For the ‘moment, let A be partitioned so that, AP=[A,, Arh, (634) ‘where Ay is mn and non-singular, and P is a column permutation mateix that may bbe required in order to ensure that A, is non-singular, One ean wate: nf ws a [ar asl ate AB, CHAPTER 6 Matrix Force Method. 205 Obviously, a permutation P that yields a non-singular A, matrix, can be chosen purely symbolically, but this says nothing about the possible numerical conditioning of A, and the resulting B,, In order to control the numerical conditioning, pivoting must be employed. There fare many such methods based on various matrix factorizations, including the Gauss-fordan elimination, QR, LU, LQ and Tumn-back method, Some of these methods are briefly studied in the following: Gauss-Jordan Elimination Method: In this approach, one creates an nx identity matrix 1 in the first columns of A by column changes and a sequence of n pivots. ‘This procedure can be expressed as, GG, .. G:G,AP=[1,M], (636) wine Gy is the ith pivot matrix and P is an mm column permutation matrix (so P* = P)and Lis an wn identity matrix, and M is an nxt mateix. A pivot matrix Gis @ diagonal matrix whose ith diagonal entry is the th pivot. We denote G,G,, G,G, by G, where G is a diagonal matrix whose diagonal entries are the scaling constants, Sealing, means simply to perform row operations Equation (6-36) can then be written as, GAP ~=[I,M], (637) or apec-lit,M)=[6-!, Im, 638) ‘which can be regarded as Gauss-Jordan factorization of A, and By ®e and B, Example 1: ‘The four by four planar frame of Figure 625 is reconsidered. The pavers of By and BIB, formed by the Gauss-Jordan elimination method, are 491 and (BB) “1342 (6-39) depicted in Figure 6.34, corresponding to 2(B,206 ‘Structural Mechanics; Graph and Matrix Methods (a) Pattern of B,. (b) Pattern of BIB, Fig. 634 Pattors of By and BjB matrices for. Example 2: The tneestorey frame of Figue 627 is reconsidered, and the Gauss-Fondan elimination method is used. The patterns of By and BB, matrices formed are shown in Figure 6.35, comesponding to 4(B;) = 483 and 2( BB) =1592. (@) Paterno, (6) Pate of BIB, Fig, 635. Patterns of By and Bj By, matrices for (CHAPTER 6 Matrix Force Method. 207 LU Decomposition Method: Using the LU decomposition method, one obtains the LU factorization of A as: PA=LU U.Va}. (6-40) and P arc again permutation matrices of order nx and mx, respectively, Now By and By canbe writen as: sey anf Example 1: The four by four planar frame of Figure 6.25 is re-considered. The pattems of By, By and BYB, formed by the LU factorization method, are depicted in Figure 6.36. The sparsity forthe corresponding matrices are .(B,) = 280, 1(81) 18 and (BB) = 1248, | (6-41) Ny @Patem of Be (b)Pattemof By, (@) Pattern of BIB, Fig. 636 Paters of B, and B[B, matrices forS Example 2: The treestorey frame of Figue 6.27 is r-considered, and the Lt factorization method is used. The patterns of By, By and BB mares formed are208 ‘Structural Mechanics; Graph and Mattix Methods shown in Figure 6.37, corresponding to (By) = 287, 1(B,) = 504 and x( B{B,) =1530, (@ Patiem of By. (@)Pattemof (6) Patten of BiB Fig. 637 Patterns of B and B{B matrices fr. QR Decomposition Method: Using @ QR factorization algorithm with column Pivoting yields, AP=QIR:, Ro) (6-42) Where P is again a permutation matrix, and Ry is an upper triangular matrix of order n. B, can be obtained as: 6-43) Turn-Back LU Decomposition Method: Topcu developed a method, the so- called Tum-back LU procedure, which is based on LU factorization and often results in highly sparse and banded B, matrices, Heath et al. [74] adopted this ‘method for use with QR factorization. Due to the efficiency of this method, a brief description of their approach will be presented in the following. Write the matrix A = (4,0) by columns. A start column is a column such that the ranks of (@1,82..,8y1) and (88.984) are equal. Equivalently, a, is a start ‘column, if itis linearly dependent on lower-numbered columns, The coefficients of this linear dependency give a null vector whose highest numbered non-zero is in CHAPTER 6 Matrix Force Method. 209 position s. It is easy to see that the number of start columns ism — dimension ofthe null space of A. ‘The start column can be found by performing a QR factorization of A, using orthogonal transformation to annihilate the subdiagonal non-zeros. Suppose tht in carrying out the QR factorization we do not perform column interchanges but simply skip over any columns that are already zero on and below the diagonal. The result wll then be a factorization of the form li ® Aza lil ° til (6-44) ‘The start columns are those columns where the upper triangular structure jogs to the right; that is, a, is a start column ifthe highest non-zero position in eolumnn s of is no larger than the highest non-zero position in earlier columns of R. ‘The Turn-back method finds one null vector for each start column a, by “turning back" from column s to find the smallest k for which columns ayyjyndy4 ATE linearly dependent. The null vector has a non-zero only in position s-K through s. Thus, if kis small for most of the start columns, then the null basis will have a ‘small profile, Notice that the turn-back operates on A, and not on R, The initial QR factorization of A is used only to determine the start columns, and then discarded. ‘The null vector that Turn-back finds from start column a, may not be non-zero in position s, Therefore Turn-back needs to have some way to guarantee that its null ‘vectors are linearly independent. This can be accomplished by forbidding the left- ‘most column of the dependency for each null vector from participating in any later ‘dependencies. Thus, if the null vector for start column a, has its first non-zero in position sk, every null vector for a start column to the right of a will be zero in position sk. Although the term "Tum-back" is introduced in Ref. 238], the basic idea had also been used in Refs. [89,93,94]. Since this correspondence simplifies the ‘understanding of the Turn-back method, itis briefly described inthe following, For the Algorithm 1 of Section 6.8.2, the use of an SRT orders the nodes and ‘members of the graph simultaneously, resulting in a fairly banded member-node210 ‘Structural Mechanics; Graph and Matrix Methods incidence matrix B. Considering the columns of B corresponding to tree members as independent columns, effectively a cycle is formed on each ordered chord (start column) by turning back in B and establishing @ minimal dependency, using the tree members and previously used chords. The eyele basis selected by this process forms @ base for the cycle matroid of the graph, as it is described in Kaveh [89,105]. Therefore, the idea used in Algorithm I and its generalization for the formation of generalized cycle bases in Refs. [89,94] seem to constitute a similar ‘dea to that ofthe algebraic Turn-back method. Example 1; The four by four planar frame of Figure 6.25 is re-considered. The pattems of By, By and BIB, formed by the Turn-back LU factorization method are depicted in Figure 6.38, corresponding to (By) = 175, x(B,) = 240 and x( B{B) ) = 408, (@) Pater of B, (©) Pater of B, (6) Pattom of BIB, Fig. 6.38 Pateras of By By and BIB, maties fr s Example 2: The three-storey frame of Figure 6.27 is r-considered, and the Tum- back LU fuctorization method is used. The patterns of By, By and BIB matrices formed are shown in Figure 6.39, corresponding to (By) = 160, x(B}) ~ 476 and (BIB, )~ 984 CHAPTER 6 Matrix Force Method. au (© Patera of BiB, (a) Pattern of By (b) Pattern of By Fig. 6.39 Patterns of By, By and BB, matsces fr. A comparative study of various force methods has been made in Ref. (112), Many algorithms have been developed for selection of null bases, and the interested reader may refer to Refs, 29,58,87,196,238], 6.11 OPTIMAL PLASTIC ANALYSIS AND DESIGN OF FRAMES, ‘The problem of plastic analysis and design of rigid-jointed skeletal structures was, ‘east in the form of @ mathematical programming with a linear objective function subjected to linear constraint by Chames et al. {23}, as early as 1951. Further progress in this field is due to Baker and Heyman [5], Horne [81], Thierauf [233], ‘Tam [230], Mokhtarzadeh and Kaveh [176] among many others, Considerable progress has been made in the last decades; a complete list of references may be ound in Munro [177], and Livestey 160} In this section, efficient programs are developed for optimal safe plastic analysis tnd design of frames. These problems are formulated in linear programming form, ‘and graph-theretical approaches are used for the formation of highly sparse ‘equilibrium equations as constraints of the optimisation approach,212 Structural Mechanies; Graph and Matrix Methods 6.11.1 FORMULATION Tre flowing sumptions are made o simpli the fomustin of the past analysis and design of frames: " 1. Equilibrium equations are referred to the undeformed geometry of the structure, 2, Plastic hinges are considered at critical section, with unlimited ductility. 3. The loads are assumed to increase proportionally. 4. Constraints are related only to bending moment yield conditions, and to the design considerations, 5. The effect of axial forces and shear forces is neglected, Fora static approach, the distribution of the internal forces at critical sections can be written as: r= Bp+Boa, (45) Consider a two-storey frame as shown in Figure 6.40 for which 9(S) = 6, and the ‘number of critical seetions is 11, as marked Roy [ope | a | Sel | Fig. 6.40. A two-storey frame and its critical sections. Cts can be introduced at the end of two beams, to transform the model into a determinate one, in the form of two subtrees. The distribution of moments at the critical sections (M;, Mp, ... , My) of these subtrees, can then be calculated under the uit values of th applied oud, Py nd bsions Ry, Ry Ry Ry Rs ad (CHAPTER 6 Matrix Force Method. 113 Using Eq. (6-45), the bending moments corresponding to internal loads can be expressed as: M,=Map + Mig. (6-46) For the considered example, My and M, are 11x3 and 116 matrices, respectively In general, the dimensions of these matrices are ex/ and exk, respectively, where ¢, Tand fare the number of critical sections, number of applied load components and the number of redundants of the structute, respectively. 6.11.2 OPTIMAL SAFE PLASTIC ANALYSIS OF FRAMES Inthe safe analysis of frames, the least value of the load factor 2 should be found subjected to the equilibrium and yield conditions. Consider M, as @ ex! vector containing the full plastic moments of the members at critical sections, The ‘mathematical formulation ofthe problem can be stated as Maximize % subject to [A(MpP+ My) +M:X|STM,, 647) ‘where T is a Boolean matrix consisting of 0 and 1 entries. The =1 ifthe jth plastic roments has to be considered in limiting the bending moment at section i, and t=0 ‘otherwise. Mi. contains the fixed end moments, added due to the presence of span Toads. equation (6-47) canbe writen a: AIM, SMMyP-+Mg)+MXSTM,. (6-48) For unrestricted variables X, the following transformation is considered to obtain non-negative optimisation variables X : +2, (6-49) in which 0 is a large positive number, Therefore 1M, , four inequalities may result, Transformation of these conditions to a standard form of LP problem requires slack and artificial variables to be considered in the left-hand side of the constraints. In the above equations A and A; of B; and B, are matrices having +1 and —1 entries depending on a typical fenry (ij) of these mattices, corresponding to jth slack or artificial variable, respectively, and zero otherwise. Once the standard forms are obtained, any program on simplex method can be used forthe solution218 Structural Mechanies; Graph and Matrix Methods 6.11.6 NUMERICAL RESULTS Example 1: A two-storey frame is considered as shown in Figure 6.41. Four ‘2foups of plastic moments are taken as design variables. An optimal design is performed and the following results are obtained: For the columns of the frst floor Myy-3.0kN.m, and columns of the second floor My=I-SKN.m. For beams of the ‘first and second floor, M,s=4.5kN.m and My=1-SkNam are selected, respectively. ‘The total minimum weight is calculated as 51,0kN.m*. ps wm) | Ma nf wl ym Mae Mp Mo a Fig. 641 A two-storey frame. Example 2: A two-span frame is considered as shown in Figure 6.42. Four groups of plastic moments are taken as design variables. An optimal design is performed ‘and the following results are obtained: For the columns My;=27KN.m, and for beams My/=77kN.m and My:=S9kN.m, are selected, respectively. The total minimum weight is calculated as 1059 kN. Fig. 6.42. A two-span frame, ‘The methods presented in this section are simple and eflicient and require exactly the same data as kinematic approach. These methods are applicable to structures Sethe ietirags CHAPTER 6 Matrix Force Method. 219 of arbitrary configuration, and can easily be modified for structures with non prismatic members. EXERCISES 6.1 Foreach set of integers a,b and ¢ of Table 2.1, draw an atbitrary ytree and ayeycle, 62 Construct a y-tree for the following graph when itis viewed as the graph ‘model ofa planar truss: 63 _ In Exercise 4.2, select a fundamental y-cyele basis of § and form its y-eyele adjaceney matrix. 64 Find a graph for which Algorithm 3 fails to select a minimal cycle basis. Repeat this exereise for Algorthin 2 6.5 Form By and By matrices, by selecting a suitable SRT and cycle basis for the following planar frame: b-.— 100% h roo h 100K h20 Structural Mechanics; Graph and Matrix Methods 6.6 Form By and B; for the planar truss of Exercise 4,2, when it is supported in statically determinate fashion. Choose the support nodes arbitrarily. 6.7 Prove the minimality of the eycle basis selected by Horton’s algorithm. 6.8 Find a counter-example for the minimality of the cycle basis selected by Algorithm 3 for planar structures, 62. NY the rains eles of «planar graph form eye ba (avesh 6.10 Use a combinatorial method, to generate minimal and suboptimal eycle bases forthe following graphs: 4] CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth Optimisation 7.1 INTRODUCTION ‘The analysis of many problems in engineering mechanics involves the solution of a set of linear equations ofthe form, Ax . oD where A is a symmetric, positive definite and usually very sparse matrix. For large structures encountered in practice, 30-50% of the computer execution time may be devoted to solving these equations. This figure may rise to about 80% in non- Tinear, dynamic or structural optimisation problems. Different methods can be used for the solution ofthe system of equations, of which Gaussian climination is the most popular among structural analysts, since it is simple and has produced some very satisfactory error bounds. In the forward course of elimination, new non-zero entries may be created, but the back substitution does not lead to any new non-zero elements. It is beneficial to ‘minimize the total number of such non-zero elements created during the forward course of the Gaussian elimination in order to reduce the round-off errors and the computer storage. Matrix A can be transformed by means of row and column ‘operations to a form which leads to the creation of minimum number of non-zero entries during the forward course of the elimination. This is equivalent to the "a priori" determination of permutation matrices P and Q, such that PAQ=6. (7)222 Structural Mechanics: Graph and Matrix Methods ‘When A is symmetric and positive definite, it is advantageous to have G also symmetric so that only the non-zero elements on and above the diagonal of G need to be stored, and only about half as many arithmetic operations are needed in the elimination. ‘The diagonal clements of A and G are the same, only in different positions. In order to preserve symmetry, Pis taken as @ 0 that Eg. (7-2 Q'A0=6., 3) Some of the desirable forms of G are shown in Figure 7.1, consisting of the banded fom vale nde fom, dely bordered block agonal form and bordered banded form: ¥ 7 a (a) Banded form, (b) Variable banded form. q (©) Doubly bordered block diagonal form,(d) Doubly bordered banded form. Fig. 7.1 Different forms for matrix G, For transforming a symmetric matrix A into the above forms, various methods are available, some of which will be described in this chapter. However, due to the simplicity of the banded form, most of the material presented will be confined 0 optimising the bandwidth of the structural matrices, and other forms will be introduced briefly, CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 223 7.2 BANDWIDTH OPTIMISATION {In the Gaussian elimination method, the time required to solve the resulting equations by the banded matrix technique is directly proportional to the square of the bandwidth of A. As mentioned before, the solution of these equations forms a large percentage of the total computational effort needed forthe structural analysis. ‘Therefore it is not surprising that a lot of atention is being paid to the optimisation, of the bandwidth of these sparse matrices. A suitable ordering of the elements of a ‘kinematical basis for a structure reduces the bandwidth of A, hence decreasing the solution time, storage and round-off errors. Similarly, ordering the elements of a statical basis results in the reduction of the bandwidth of the corresponding flexibility matrix ofthe structure Iterative methods using different criteria for the control of the process of interchanging rows and columns of A are described by Akyuz and Utka [2], Alway ‘and Martin [3], Rosen [207] and Grooms [66]. For these methods, in general, the required storage and CPU time can be high, making them uneconomical. ‘The first direct method for bandwidth reduction was recognized by Harary [70] in 1967, who posed the following question: For a graph $ with N(S) modes, how can labels 1,2, ... NS) be ‘assigned to nodes in order to minimize the maximum absolute value of the difference henween the labels ofall pairs of adjacent nodes? For a graph labelled in such an optimum manner, the corresponding. adjacency tmatrix will have unit entries concentrated as closely as possible to its main diagonal, In structural engineering, Cuthill and McKee [34] developed the first graph- theoretical approach for reducing the bandwidth of stiffness matrices. In their ‘work, a level structure is used which is called a "spanning tree” ofa structure. The ‘author's interest in bandwidth reduction was initially motivated by his interest in generating and ordering the elements of cycle bases and generalized cycle bases of ‘graph, as defined in Chapter 6, in order to reduce the bandwidth ofthe flexibility matrices, Ref, (89). For this purpose, a shortest route tree (SRT) has been used. ‘The application of this approach has been extended to the elements of @ kinematical basis (cutset basis), in order to reduce the bandwidth of stiffness ‘matrices, Subsequently, it has been noticed that there is # close relation between Cuthill-MeKee’s level’ structure and the author's SRT. However, there is a difference between these tres in that an SRT contains additional information about the connectivity properties of the corresponding structure Further improvements have been achieved by employing special types of SRTs such as the longest and narrowest ones, Ref. [100]. Generation of a suitable SRT depends on an appropriate choice of starting node. Kaveh [89] used an end node of224 ‘Structural Mechanics: Graph and Matrix Methods 1m arbitrary SRT, which was chosen om it ast onou (eed having the ea Vane. Gibbs ea. [57] emplayed sins nde and called paca Pesperlnde, Cheng [25] set an algae approach oct singe node oe Se of nde a the fot ofan SRT. Kaveh splot siltneas RTs fo Seetng a pseuo-peripherl nod. A compar of sx ile goiin Woo ta REF LO) Ace gph tery has alo eco tr ff Skating node, Kaveh [10] Grins et al [65, Kaveh and Rahn Boner pentane aay [lol] al. [65], Kaveh and Rahimi Bondarabady. Extensions and applications ofthe nodal numbering algorithms to element odetn for bandh, profile and fontwidth optimisations ate due to Kaveh [101 102 Alias an Da 1, verin [4 Rare (20), Pina [19], Fomes a ng [49], Sloan and Randolph [219], Sloan [220], Burgess and Lai [17], and excellent books on these topics are those of Duffet al. [39] and Pissanetsky [193], 7.3. PRELIMINARIES A matrix A is called banded, when all its non-zero entries are confined within « band formed by diagonals parallel to the main diagonal. Thus Ay = 0 when li — j| > band Ais. #0 of A.s.s #0 for at least one value of k bis the half-handwidth and 2b+1 is known as the bandwidih of A. As an example, for 1 6 4) the bandwidth of A is 2b+1=22+185, ‘A banded matrix can be stored in different ways. The di in different ways, The diagonal storage of a symmetric banded axn matrix A is an mx(bi1) matrix AN. The main diagonals are stored in the last column, and lower co-diagonals are stored down-justified in the emaining columns, As an example, AN for the above matrix is: rs) CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 225 ‘When A is a sparse mattix, this storage scheme is very convenient, since it provides direct access, inthe sense. that there is a simple oneo-one Correspondence between the position of an entry in the matrix AQ) and its position in ANG, ~i#+b+. Obviously, the bandwidth depends on the order in which the rows and columns of ‘Aare arranged. This is why iterative techniques seek a permutation of the tows and ‘permutation of columns to make the resulting bandwidth small. For symmetric ‘matrices, identical permutations are needed for both the rows and the columns. ‘When a system of linear equations has a banded matrix of coefficients and the system is solved by the Gaussian elimination, with pivots being taken from the diagonals, all the operations are confined to the band and no new non-zero entries are generated outside the band. Therefore, the Gaussian elimination can be carried ‘out in place, since a memory location is already reserved for any new non-zero that ‘might be introduced within the band. ‘When a banded matrix of igh order has a wide band and a large number of zeros inside it, the diagonal storage may become wasteful. Then an emelope (variable band) scheme of Jennings (85), the so-called skyline (Felippa [50]), may be used. For each row i of a symmetric matrix A, define bimim jaa, 7) ‘where jain (@) is the minimum column index in row i for which Ay # 0. Therefore, the first non-zero of row i lies b, positions to the left of the diagonal, and b is defined as: max (b), 7) ‘The envelope of A is the set of elements Ay such that 0 < i -j 1, are in positions IN(i~ 1}+1 to IN(i. The only element of row 1 is Aaj, stored in AN(1). The elements have consecutive, easily calculable column indices.226 Structural Mechanics: Graph and Matrix Methods For example, the matrix of the previous example, has a profile equal to 4, and its envelope storage is: Postion = 1 2 3 4 5 6 7 8 9 AN 62 73 8 9 4 5 7-9) IN = 3 5 8 9 A variant of Jennings’s scheme is obtained when the transpose of the lower envelope is stored. In this case elements are stored column-wise, and since the colurmns of the matrix retain their length, the scheme is often termed skyline storage. The profile of a matrix also changes if the rows and columns are permuted. Another useful concept used in the design of storage schemes for symmetric matrices is waveffont or frontwidth. This method fas application in the clement method when a frontal solution is employed, Irons (84,121, 149,158.21 With reference to equation (7-1), column jis ssid to be active at stage i ij 2 i and there is a non-zero entry in colurnnj with a row index, k, such that k $i. Letting f denote the number of columns that are active at stage i, the mau franewidde of Ais given by: Fy = max {fi (7-10) Isisn ‘The oot mean-squared (ts) wavefon is defined as: Peg Sant ony j Assutning that n and the average value of fate reasonably large, it can be shown that a complete profile oF front factorization requtes approximately O(aF*) operations. 74 PATTERN EQUIVALENCE OF STIFFNESS AND CUTSET ADJACENCY MATRICES In previous chapters, it has been shown that the stifiness matrix K ofa structure i pattern equivalent to the eutset basis adjacency matrix N= LL, whete Lis the cust basis incidence matrix of the structural model $. Similarly, the flexibility matrix G is patern equivalent to the cycle basis adjacency matrix D = CC, where Cis the eyele basis incidence matrix of 8. CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 27 Reducing the bandwidths of N and D directly influences those of K and G, respectively. Notice that the dimensions of N and D, for general space structures, tre six-fold smaller than those of K and G, and therefore more simple to optimise. For stiftness analysis there exists a special cuset basis whose elements correspond, to stars of its nodes except for the ground node (cocycle basis). The adjacency matrix of such a basis naturally is the same as that ofthe node adjacency matrix of SS with the row and column corresponding the datum node being omitted, In this chapter such a special cutset basis will be considered, and the nodes of $ will be ‘ordered such that the bandwidth of its node adjacency matrix is reduced to the smallest possible amount, Let A be the adjacency matrix of graph. Let i and j be the nodal numbers of member k, and etc, =[i~,. Then the bandwidth of A can be defined a, biAy= Max 042 K=1,2, aMM(S)} +1, 2) where M(S) is the number of members of 8, In order to minimize the bandwidth of A, the value of b(A) should be minimized, Papademetrious [185] has shown that the bandwidth minimization problem is an NP-complete problem. Therefore, any approach to it is of interest primarily because ofits heuristic value228 Structural Mechanies: Graph and Matrix Methods 7.8 A SHORTEST ROUTE TREE AND ITS PROPERTIES AA shortest route tree rooted at a node, called the starting node (root) of the tree, has the following properties: ‘The path from any node to the root through the tree isa shortest path. An algorithm, for generating an SRT has already been described in Chapter 1 and therefore only its properties relevant to nodal number ae discussed here, ‘An SRT decomposes (partitions) the node set of S into subsets according to their distance from the root. Each subset is called a contour (level) of the SRT, denoted by G,, The contours of an SRT have the following properties: AG) (C)ECAUCn I PA forany eigenvalue Ais iif) 2 has a unique eigenvector W, which may be taken to have all positive cents. [As W, is the eigenvector corresponding 10 2, QW, = 2W, for i=l, .. NUS) “Multiplying the two sides with Q, one obtains QQW; =2,QW, =221¥. Repeating this process resus in Q*W; =2EW. Now consider any vetor x nt othogonal CW) Ws +. 0qysWys) G10. ory Multiplying the two sides with Q', and using QW, =24W, for el, .. NCS) leads to bx = Af cy Wy + Ab aWa +. +BKys) ON Wass, (7-18) and, ask->e>, we have, Q'N/ AY = 0 W, + Gala) We +. +Aoys/a) Mohs) Ways) 201i, (7-19) since 2, is the eigenvalue of strictly largest modulus, and (2/2) is less than unity anu approaches to zero when ktm other words, the ratios ofthe components of Qe approach the ratios ofthe components of W, ask increases Let ¥~ (ates then the 1h component of Q' isthe number of walks of length {beginning a an arbitrary node of S and ending at, If, i a good starting node (peripheral node), tis namber wil be smaller. Thus fork-¥=>, one should obtain Some average number, defined as the accessibility index by Gould (64), which indiates how many ‘walks go on average through « node. With a suitable normalization, Qr converges to the largest eigenvector W, of , Stafing [226]232 Structural Mechanics: Graph and Matrix Methods ‘The following algorithms are designed to obtain a good starting node of a graph: Algorithm A (a TGT approach) Step 1: Construct an SRT on each node of $ and select the narrowest SRT. In this algorithm the width numbers of all nodes are calculated, These numbers may: further be used in the process of ordering, Obviously, such an approach ean be expensive for large structures, which defeats its application Algorithm B (a TGT approach) Step I: Start with a node ofthe least valency and form an SRT. Step 2: Record all the nodes ofthe last contour of the selected SRT. ‘Step 3: Form SRT rooted at each ofthe recorded nodes and choose the one which corresponds to the narrowest SRT. The process of constructing an SRT is, terminated as soon as the width of one of its contours exceeds the width of the previously selected SRT Algorithm C (a TGT approach) Step 1: Start from an arbitrary node of S. Construct an SRT on this node and take a node of least valency from its last contour. Step 2: Form a new SRT from the selected node, and then repeat Steps 2 and 3 of Algorithm B and choose the starting node corresponding to the narrowest SRT Algorithm D (a TGT approach) Step 1: Form an SRT on each node of $ and select the narrowest SRT, Unlike Algorithm A, the construction of an SRT is terminated as Soon as its width exceeds the width ofthe previously selected SRT. This algorithm is more economical than Algorithm A; however, it does not provide the information that Algorithm A reveals, for subsequent use, Algorithm E (a TGT approach) Step 1: Start with an arbitrary node, form an SRT on this node and take a node 1 of least valency from its last contour, CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 233 Step 2: last contours sate an SRT on n, and find ll nodes contained in is even, first and Step 3: Generate an SRT on each node of these contours, and find the narrowest ‘one. The process of formation of an SRT is terminated as soon as the width of one of its contours exceeds the width of the previously selected SRT. Denote the selected node by ny Step 4: Check adjacent nodes ton, for possible reduction in width, to decide the final starting node. Algorithm F (a TGT approach) Step 1: From an arbitrary node generate an SRT, and from its ast contour select ‘anode X, of minimal valeney. Observe the width of the selected SRT. Step 2: Generate an SRT from X; and select Xz ofthe least valency from its last contour, and observe the width, ‘Step 3: Generate two SRTs simultaneously rooted at X; and X, and find the node X, which is the last node of S included in one of the SR subtrees, Once X; is found, terminate the process of forming SRTs. Generate an SRT from Xs and “observe its width, X, and Xp are called the generators of Xs Step 4; Repeat the process of Stop 3, using the pairs (X),Xs) and (X:,Xs) a the ‘generators to find X, and Xs respectively. Construct the corresponding SRT and ‘observe their widths. Step 5: Repeat the process of Step 3 for X; (= 3, 4, ..) together with the ‘corresponding generator, until no further improvement in width is observed. The owest SRT should be selected for nodal decomposition ofS. ‘An example of the application of this algorithm is depicted in Figure 7.4, where a cross-shaped grid $ is considered. Starting from an arbitrary node "O”, an SRT is generated and X; is obtained from its last contour. Generating a new SRT from Xj, node X; is chosen from its last contour, Xs is the resull of generating. two simultaneous SRT from X, and X;. Using (X,,Xs) and (%,%5), nodes Xy and Xs are obtained, respectively. The width of the selected SRTs rooted at X).X;,Xs,Xe and X; are 8,8,8,11, and 10, respectively, Therefore the process is terminated and X is taken as @ good starting node of S,24 Structural Mechanics: Graph and Matrix Methods Xs xr * xy Fig. 74 A cross-shaped grid and the selected X; (1,5) by Algorithtn F, Algorithm G (an AGT approach) Step 1: Caleulate the dominant jenvector Wy = {Wi,Wava,Why! of matrix Q Step 2: Find Min W; in W, The node corresponding to this component is taken 8a good starting node ofS, For calculating the dominant eigenvector W of Q, an iterative method is used, which starts with v= (1,1, ..,1}! and calculates Q®. This vector is then normalized and multiplied by Q. This process is repeated until the difference between two consecutive eigenvalues, obtained fiom Qu ~ 2v, is reduced to a small value which, for example, can be taken as 10-3. The author has compared the efficiency of the above algorithms through some randomly generated graphs. A comparison of the time and bandwidth for the generated examples may be found in Kaveh [103], CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 235 7.62 PRIMARY NODAL DE‘ COMPOSITION Once a good starting node is selected, an SRT is constructed and its contours {CyCovnCn} ate obtained. These subsets are now ordered according to their distances’ from the selected starting node. Obviously, many SRTS can be constructed on a node. Although all lead to the same’ nodal decompositions, different transversals will be obtained for different SRTs. Thus, in the generation process, the nodes of each contout C; are considered in ascending order of their valencies (or entries in eigenvector W,) for selecting the nodes in Cin order to provide the conditions for the possiblity of generating a minimal (or optimal) ‘wansversal as defined in the next section. Finding an optimal transversal before an SRT js fixed seems to be a time-consuming problem, However, for most of the ‘models encountered in practice, optimal transversal isin between minimal ones. In the following, two algorithms are given for selecting a suboptimal transversal of an SRT. 7.63 TRANSVERSAL P OF AN SRT {transversal of an SRT, i defined as «connected path P containing one distinct rode N, from each contout C,of the SRT. A sina transversal isthe one for which $deg(N;) is minimum, An optinal transversal isthe one leading tothe best nodal numbering, ie. « numbering corresponding to smallest bandwidth for the selected decomposition. The weight of a node is defined as its degree or its vahie W, in Wi, depending upon the use of aTGT approach or an AGT method, respectively, Algorithm AA (a TGT approach) Step 1: Take a node Ny of minimal weight from the last contour Cy of the selected SRT Step 2: Find Na from yet Which is connected to Ny, by a branch of the SRT. Step 3: Repeat the process of Step 2, selecting nodes N, representative nodes of the contours Cy et we Nip as the sy (Cho Tespectively, The above algorithm is a backtracking process from a node of minimal weight in the last contour Cy, and selects a transversal P= {Nj, No, Ne} Which ean now be used for ordering the nodes of the contours ofthe corresponding SRT.236 ‘Structural Mechanics: Graph and Matrix Methods ALGORITHM BB (an AGT approach) Let C),Cay. Cm be the selected contours of the SRT, and correspondingly put these subsets in W, into a similar order, ie. Wie (WCW), Cad}. (720) Where W(C) contains the entries of W; corresponding to the nodes of C,, Now the algorithm can be described as follows: Step 1: Label the root as Ny and assign W; of this node as its new weight denoted by W, Step2: Calculate the new weight W, of each node of C; by adding the W's fiom WC) to W, Step 3: Repeat the process of Step 2, calulating W, foreach node of CC, Cu Step 4: Take @ node Ny, of minimal weight from the last contour Cy of the selected SRT, Step 5: Find N,.1 from C,,;, which is connected to Ny by a branch of the SRT. Step 6: Repeat the process of Step 5, selecting NaaNyss representative nodes ofthe contours CyCxs, C Ni asthe P= (NNaeoaNw) forms a suboptimal transversal ofthe selected SRT. In general, the results of the Algorithm BB have been better than those of Algorithm AA, at the expense of some additional computer time, As an example, for the grid model of Figure 7.5, a suboptimal transversal of the SRT rooted at X; is shown in bold lines 7.64 NODAL ORDERING Step 1: Number, as "1", Step 2: Nz is given number "2" and an SR subtree is generated from Ns, ‘mumbering the nodes of C; in the order of their occurrence inthis SR subtree, ‘Step 3: The process of Step 2 is repeated for numbering the nodes of C3, Ch «5 Cr sequentially using Na, Ns, «» Ny as the starting nodes of SR subtrees, until all the nodes of $ are numbered, (CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 237 Now the numbering can be reversed, in a way similar to that of the Reverse Cuthill McKee algorithm, for possible reduction of fill-ins in the process of ‘Gaussian elimination, wihich will be discussed in Section 7.9. 7.6.5 EXAMPLES ‘The following two simple examples are chosen to illustrate the steps of the presented approaches, but the applications are by no means limited to such simple Example 1: Let $ be the graph model of a truss structure, as shown in Figure 7.5(a). Using a TGT algorithm, a staring node A is found, and the corresponding SRTs are depicted in Figure 7.5(b). A transversal is selected as shown in bold lines, Figure 7.5(c). Then nodes are numbered contour by contour, employing the representative nodes asthe starting nodes of SR subtrees, Figute 7.5(d). (eee ee ce eG) a of of ot fs u| fy © () Initial numbering of S. (b) The selected SRT. Ne__Ni_Ne_N_Nt_Ns Geet ees ™ Joni rbnbAlele ™ Alles see eee wl (©) The selected transversal P. (@ Final nodal numbering ofS. Fig. 75 Graph model S of Example 1238, Structural Mechanics: Graph and Matrix Methods Example 2: 5 is the model of a grid with uniform valeney distribution, as shown in Figure 7.6(a). Using algorithm G, the following dominant eigenvector is obtained for matrix Q of S, in which for simplicity only four digits ate provided: W) = (0.3344, 0.5298, 0.6161, 0.5951, 0.4791, 0.3011, 0.1180, 0.3972, 0.7432, 0.9540, 1,0000, 0.8786, 0.6183, 0.2875, 0.2875, 0.6183, 0.8786, 1,000, 0.9540, (0.7432, 0.3972, 0.1180, 0.3011, 0.4791, 0.5951, 0.6160, 0.5298, 0.3344)" ‘Thus node "7" is selected as a good starting node. An SRT is generated from this node and, using Algorithm BB, a transversal P = {7,14,21,28,27,26,25,24,23,22} is selected, which is shown in bold lines in Figure 7.6(a) Final nodal numbering is, illustrated in Figure 7.6(6). ma 2s ess {LA Ae vio “ol A], Ya Ya Aol “al oa 3 Vol Ail a| Aa Aa al Sas 8 a 52 (9) Initial numbering and the selected transversal, (b) Fig. 7.6 ‘The graph model $ of Example 2, CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 239 7.7 A CONNECTIVITY COORDINATE SYSTEM FOR NODAL ORDERING {In order fo cast the concepts developed for nodal ordering in a mathematical form, 4 connectivity coordinate sytem is defined for nodal numbering ofS, Kaveb [106] Separate study of planar and space graphs resulls in clarification of further interesting points about nodal numbering of space structures, as described inthe following. 7.7.1 A CONNECTIVITY COORDINATE SYSTEM FOR PLANAR GRAPHS Let $ be a connected planar graph, Select a good starting node "0" of S as the root of an SRT (analogous to the centre of a coordinate system). Generate an SRT of S rooted at © and find a suboptimal transversal P of the SRT (analogous to the x= axis). Now any node n, of S can be associated with two integers, namely (fd), here, is the shortest distance of n, from O (distance of the contour C; from O ‘which n; belongs), and dj is the shortest distance of n, from the representative node Nie Ci (analogous to (x,y) in a Cartesian coordinate system), ‘An order can now be defined for the nodes of S as, ny (rd) comprises of the I-weighted degrees of the nodes, and v; and vy are distances of the nodes from two pseudo- CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 255 peripheral nodes, These nodes can be obtained using diferent algorithms, Kaveh [110}, The results are provided in column 3 ofthe Tables denoted by v" For the second ease, five Ritz vectors are employed. The first four vectors are the same as those of the previous case, and the fiflh vector v; contains the 2-weighted grees of the nodes of the graph. The results are provided in column 4 of the ‘Tables labelled as v’, It should be noted that other veetors containing graph properties which influence the ordering may be considered additional to the above five vectors. However, the formation of such additional vectors may require some extra computational time, reducing the efficiency of the algorithm, Example 1; An FE mesh with one opening, comprising of 1248 nodes and 1152 rectangular elements is considered, as shown in Figure 7.17, The results for different methods and their computational time are illustrated in Table 7.1 to compare their efficiency. Fig. 7.17 An FE mesh with one opening.256 Structural Mechanics: Graph and Matrix Methods CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth... 257 Table 7.1. Results of Example 1 ‘Table 7.2. Results of Example 2. PAGM v v PAGM. v v B B & B 7 77 a P. 3a84__| 36243 | 36189 \ P a74i_ [49400 | 48936, E 28.07__[ 29.44 [29.251 E 23.97 | 25.63_| 25.32 Fou 35, 39 39 Fan 37 2 42 Time (see) _[_14003_| 28 29) Fime (see) | large [2.63 [2.89 Example 3: A two-dimensional FEM of a tunnel, comprising of 6888 nodes and Example 2: An H-shaped FE mesh, comprising of 2096 nodes and 3900 triangular 6720 rectangular elements is considered, as shown in Figure 7.19. The results of clements is considered, as shown in Figure 7.18. The results for different methods using different methods and their computational time are presented in Table 7.3. and their computational time are illustrated in Table 7.2 to compare their efficiency. Fig. 7.19 A two-dimensional FEM of a tunnel. Fig. 7.18 An FE mesh with one opening ;258 Structural Mechanies: Graph and Matrix Methods ‘Table 7.3. Results of Example 3. PAGM v y B a5 331 32 P Taina _| 733738 _| 733738 E 112.99_[112.93_ [112.93 Fass Téa 175, 175 Time (see) | 106 27.6 | 289 Example 4: A two-dimensional FE mesh with four openings comprising of 748 nodes and 1236 triangular elements is considered as shown in Figure 7.20. The results for different methods and their computational time are illustrated in Table 7.4 in order to compare their efficiency. Fig. 7.20. An FEM with four openings. CHAPTER 7 Ordering for Bandwidth, Profile and Frontwidth 259 Table 7.4. Results of Example 4. i PAGM v v | B BI w a7 i P Tis [13162 | 13126 i EF 18.42 | 1861 [18.56 Fuss 29 2 29 | “Fie (see) _[_1677 12 13 Example 5: A thtee-dimensional finite element model of a nuzzle is considered as shown in Figure 7.21, This model, contains 4000 rectangular shell elements. The results for different methods and their computational time are illustrated in Table 7.5, in order to compate their efficiency. Pe Za 7.21 A threo-dimensional FE mesh of a muzzle260 Structural Mechanics: Graph and Matrix Methods Table 7.8. Results of Example 5 PAGM v v B 3 wo a P. iis] 13162_| 13126 FE 1842_| 18.61 | 18.56 Fea 29, 29 29 Time (se¢ 1677 12 13 7.12.6 DISCUSSION The performance of the present method, illustrated in the previous section, compares well with @ pure algebraic graph method, with a substantial reduction in the computational time, Naturally, addition of extra graph parameters will increase the computational time required. Relative values of the coefficients of the Ritz vectors show the importance of the corresponding parameters in the ordering algorithm. For the examples presented in the previous section, the coefficient corresponding to v3 and v, (the distances from the pseudo-peripheral nodes) seem to be more important, since most of the examples have a more or less uniform distribution of nodal’ degrees, Naturally, for models with non-uniform degree ighs and any = maxtig —ebix #0, ka ob % @ O) Fig. 8.7 A planar frame S, a cocycle basis and a cutset basis of .290 Structural Mechani Graph and Matrix Methods 123456 123456 ewes Ife aes + ae e+ waft tee wa sae Die d= eee jo o+ +e | tae fot eae 4 oe (Dj) = 24 103) 1t will be realized that sparser stiffness mattices ean be generated using suitable ccutset bases than by employing the traditional cocycle basis, Ref. [113] In order to keep the off-diagonal terms small, the members in the overlaps of the cutsets should be as flexible as possible; ic. lower weight members should be included in the overlaps, in the following, three algorithms are designed for the formation of suboptimal cutset bases of the graph model ofthe structures, 87.1 ALGORITHMS: The formation of a cocycle basis ofa graph model $ is simple and straight-forward. For this purpose, the members incident with each free node (except the selected datum node) are taken as an element of the basis. Repeating this operation for all the free nodes, completes the process of the generation, Algorithm A Step 1: Generate a spanning tree of maximal weight. Order its members (branches) in ascending magnitude of weight, Step 2: Use a branch of the least weight, form the selected tree and form the first fundamental cutset on this branch, Step 3: Form the next fundamental cutset on the unused branch of the least ‘weight Step k: Repeat Step 3 for the other unused branches until p(S) = N(S) ~ 1 independent cutsets forming a basis are generated. CHAPTER 8 Conditioning of Structural Matrices 291 Algorithm Stop 1: Form a cocyele bass; denote the selected cocyeles by C" Step2: Take the frst cocycle Cf of C"" and combine with the remaining cocyeles of C'", For cach cocycle C} (2, .. ()) satsying the following condition, replace Cj with Ch@C} Condition: ( 1) or (La Lys and Wae< Wa), where Lu (Wa1) and Lo(W.) indicate the lengths (weights) before and after the application of the combining process. The new set of eocycles and/or cutsets is denoted by C’ Step3: Take C3 of C” and repeat a proces similar to that of Step 2. Stepk: Take CL of Cand combine with he elements of CY, The process terminates when k becomes equal o9(S) Algorithm © js algorithm is the same as Algorithm B, with the difference that the corresponding condition, is replaced by the following one: Condition: (Was Was) oF (Wa= Wa and La< La). ‘The selected bases are suboptimal and contain clements with lower weight ‘members, leading to kinematical bases corresponding to small off-diagonal terms for stiffness matrices. 8.7.2. NUMERICAL RE: ULTS A one-bay four-storey planar truss is considered as shown in Figure 8.8, with cross-sections being designated by A,, Typical member eross-sections are: Ac=20em?, Az=L0em?, Ay=Sem?, Ay=dem? and B=2.1%10°kN/em?292 ‘The patterns of the cutset bases adjacency matrices are illustrated in the following: Structural Mechanics: Graph and Matrix Methods (@) A planar truss Fig. 8.8 A planar truss and its graph model S, 4 5678 Pattern of D* by a cocycle basis. (b) The graph model S, 2345678 Patter of D* by Algorithm A. | CHAPTER 8 Conditioning of Structural Matrices 12345 678 Pattern of D* by Algorithm B. 293 78 Pattern of DY by Algorithm C, ‘The condition numbers of stiffness matrices, the sparsity, and the magnitudes of Ls and Ws forthe selected cutset bases are illustrated in Table 8.5. ‘Table 8.5 Comparison of the condition numbers and sparsities. Algorithm | PL. PN poet | x) | ts | Ws Coe. Basis | 272013 | 22965707 | aRNDAsE-G | 34 | 13 | TSHBGS ‘A | 2148762 | 1.o936396-s | 113400853 | 32 | 12 | 480887 B | 2502612 | gsovraaE-4 | o4osims-3 | 22 | 7 | 462000 © | 2288613 | yro26076-4 | saszarse-3 | _24 |__| 364000 ‘The execution times for the formation of the solected cutset bases (Tc, and the corresponding stiffness matrices (T.), are presented in Table 8.6, ‘Table 8.6 Comparison of the computational time. Time | Cocycle A B c basis Te ‘0.00 O88 076 088 Tx 043 0.65 0.48 0.60) ‘Although the sparsity of stiffiaess matrices K can be improved by the formation of special cutset bases in place of cocycle bases the improvements, in general, are not significant. On the other hand the conditioning of KC can be improved by employing, appropriate cutset bases, Algorithm B improves the conditioning of the stifiaess ‘matrices, maintaining the sparsity of the stiffness matrices. This improvement is more significant for Algorithms A and C, although the sparsity of Kis not ‘maintained.294 Structural Mechanics: Graph and Matrix Methods EXERCISES 8.1 For the grid shown in the following, the relative stiffnesses are encircled. Find a suboptimally conditioned cycle basis of this model ‘io ao 8.2 Find the condition numbers PL, PDET and PN of the stiffness matrix of the following planar frame. Take the elastic modulus as E=2x10'kNiem, ‘A-000106m7 10.0007 8.3. Repeat Exercise 8.2 for the flexibility matrix ofthe same structure, 84 Study the effect of bandwidth reduct structures and find out whether this effect is sig simple matrix chosen arbitrarily fon the stiffness matrices of ificant. Hlustrate this fact with a 8.5 Writes computer program to calculate the largest and the smallest cigenvalues for adjacency matrices of graphs, 8.6 Use Algorithms A, B and C to find suboptimally conditioned eyele bases for the following weighted graphs. The numbers | and 2 show the member types as siven in Section 8.5.2, | CHAPTER & Conditioning of Structural Matrices [Tt | © i) 295CHAPTER 9 Matroids and Skeletal Structures 9.1 INTRODUCTION The theory of matoids, which was introduced by Whitney [249] in his pioneering paper as early as 1935, is concerned with the abstract properties of independence. He conceived a "matroid” as an abstract generalization of a matrix; hence some of the language, including the name of the theory, is based on that of linear algebra. ‘AL the same {ime he refers to matroids as generalized graphs, and uses some terms, from graph theory. Matroids have received a great deal of attention from both the theoretical and the application points of view. Contributions have been made to its extension by Tutte, Rado, Welsh, Mirsky, Edmonds and many others; an excellent introduction is Welsh [243]. Matroids have been applied to vatious fields of engineering such as electrical networks by Minty [174], structural analysis by Kaveh [89,105] and rigidity of structures by Crapo [33] Reeski (202), and Whiteley (248) among many other fields of science and engineering, The author's interest in matroids has been ‘motivated by the active presence of both matrices and graphs in the matrix analysis of structures, and the need for generalizing some of the existing concepts of graph theory, Refs. [93,94] In this chapter, a matroid is defined using different but equivalent sets of axioms, Examples of matroids associated with graphs are included, due to the partial familiarity of structural analysts with the theory of graphs. The Greedy Algorithm developed by Edmonds [42] for selecting an optimal base of a matroid is described,298 Structural Mechanics: Graph and Matiix Methods and the problems encountered in the application of this algorithm in structural ‘mechanics ate discussed, Methods are suggested to overcome the difficulties involved, 9.2 AXIOM SYSTEMS FOR A MATROID. A mattoid may be defined in many different inter-refated forms, several of which were described in Whitney's original paper. Here the definitions in terms of the concepts of independence, bases, cireuits and ranks are presented, and the proof of their equivalence may be found in Refs. [243,249], Matroid theory postulates certain sets f0 be independent and develops a fitful theory ftom certain axioms which it requires to hold for this collection of independent sets. 9.2.1 DEFINITION IN TERMS OF INDEPENDENCE [A matroid visa sot of elements. ‘S(called independent sets) such that } and a collection # of subsets of W Gee 1D) Xe and YCX,then Ye & 1B) Xe F and Ye & with) X|=]¥] +1, then there exists s © X ~ ¥ such, that Y4se Hore |X | and | ¥ | denote the cardinalites ofthe sets X and Y, respectively. For a mairoid 3/= (8.2, those subsets of belonging to Fare ealled independent, and those which do not belong to % are known as dependent, A. maximal independent subset of a matroid, is known as a base of 9.2.2. DEFINITION IN TERMS OF BAS! M2) is 4 mattoid if the collect following conditions: of bases of denoted by a satisfies the Bl) 940, CHAPTER 9 Matroids and Skeletal Structures 299, 2) | By|=|Bz|forevery By, Bre at B3) If By, By ¢ @and 5; © By, then there exists a s> € By such that (Bi-sy+s:) ea A circuit of a matroid /is a minimal dependent set of 5 9.2.3. DEFINITION IN TERMS OF CIRCUITS (5:9) 8 matroid if the collection of citeuits of 4%, denoted by ¢ satisfies the following postulates; C1) No proper subset ofa circuit isa ereuit C2) IFC; and Cz ate distinet citeuits of Cand s © C; 0 Co, then there exists @ cireuit: Cyof Cuch thar Cy (CUC)=8 Corresponding to each subset F, a number r(F;) ¢ Z is defined, which is known as the rami of F, as follows: 116) = Max (KEX GR..Xe #1 oy 9.24 DEFINITION IN TERMS OF RANK 34.) forms @ matroid ifthe following conditions hold: R1) The rank of the null subset is zero. R2) Fora subset F\and any element s notin F; r(F; +8) = 1(F) +k, (k 0 oF D. fF), then (F; +5148) F). R3) Fors, synotin if B+ 8) HF +8) ‘The mulliy of F, is defined as n(F) = IFi|- x0). Obviously, a subset is independent if its nullity is zero; otherwise it is dependent, An element s © sis called dependent on F if (+8) ~ 1(F); otherwise it is independent of F. The nullity of a base is zero, and that of a eireutis unity.300 Structural Mechanies: Graph and Mattix Methods Now it is obvious that knowledge of the bases, or circuits, or rank functions is sufficient to uniquely determine the corresponding matroid, Therefore, it is not surprising that there exist axiom systems for a matroid in terms of each of these concepts. One ean also consider one of these axiom systems and prove the others as theorems. A knowledge of all the axiom systems, helps in constructing suitable ‘atroids and employing the relevant properties efficiently. Example 1: Consider the following matrix: 2345 afi oour Ooo over the field R of real numbers. The column set {1,2,34,5} of A and its independent subsets form a matroid (A). The set of independent subsets of this ‘matroid is obtained as I= (@, {1}, £2}, {4}, (3), {1,2}, (24}, (2,5), (4,5}}, and the set ofits ciruits is C= {(3), {1,4}, {1,25}, (2.4,5)) Example 2: Let S be the graph as shown in Figure 9.1. Consider a matroid 448), formed on the members {1m,,m;,ms.mys} of S, with circuit set as © = (as) {n,m} mmm}, masmyms}}. This matroid is the known as the «eyele matroid ofthe graph S, as defined in the Section 9.3.4, Fig. 9.1 A planar graph s. ‘Now compare :i(S) with 44) in Example I. It can be seen that, under bijection from (1,2,3,4,5) to {m)m.m,myms}_ defined by ¥(i) = m, a set X is a eiteuit in 24A) if and only if %(X) is a citeuit in 44S). Equivalently, a set Y is independent in (A) if and only it YY) is independent in 44S). Thus matroid A) and 44S) hhave the same structure or are isomorphic. A matroid that is isomorphic to the CHAPTER 9 Matroids and Skeletal Structures 301 cycle matroid of a graph is called graphic. Therefore, the matroid ACA) in Example 1 is graphic. 9.3 MATROIDS RELEVANT TO STRUCTURAL MECHANICS ‘Matroids have been applied to various problems in structural analysis and the study of the rigidity of skeletal structures. In this section, examples of such matroids are considered, and the properties associated with each one are discussed 9.3.1 A BASIS FOR A FINITE VECTOR SPACE A conceptual study of structural analysis using vector spaces has been made by Maunder [169]. One can easily obtain a matroidal version of this study, by constructing matroids ofthe following kind: Let 1/be a finite vector space and be the collection of linearly independent subsets of vectors of Then 4¢= (14 4 forms a matroid, The rank function of this ‘matroid is the dimension of and its base forms a basis of the vector space, Although a finite vector space always constitutes a matroid, not all matroids are realizable as vector spaces, 93.2 ABASIS FOR CYCLE SPACE OF A GRAPH A cycle basis of a graph is defined in Chapter 1, and its application for the formation of a statical basis of a structure is described in Chapter 6, In this section, 1 eycle space and its bases are formulated in terms of matroids. Let Contain all simple cycles of a graph S, and #be the collection of mod 2 independent cycles of S. Then (A forms a matroid, defined as cycle space ‘matroid 46(S) of S. A base of 46S) is a eyele basis of S, and its rank is M(S) — NiS) + biS). ‘The above matroid can be defined using the member-cycle incidence matrix of a graph. Each row of this matrix corresponds to a member, and each column represents a cycle, ‘The columns of a member-eycle incidence matrix are either dependent or independent. Take the columns of the matrix as elements of ¢; and independent subsets of columns as elements of # Then (G#) forms a cycle space matroid 4(8) of . Example: Consider a graph S as shown in Figure 9.2. This graph contains 3 cycles C= C Cay Cy} and = (C)MCIACIAC, CaYCs, Ca)(Ca, C2}. The rank of 2K(S) = 5 ~ 41 = 2, and {C), C3} is atypical base of this matroi.302 Structural Mechanies: Graph and Matsix Methods Fig. 9.2 A graph S and its cycles. 93.3. ABASIS FOR CUTSET SPACE OF A GRAPH AA cutset space of a graph is defined in Chapter 1, and its application for the formation of a ‘when the displacement method is used, is deseribed in Chapter 5, In this section, a cutset space and its bases are defined in terms of matroids Lot C* contains all cutsets of a graph S, and #be the collection of mod 2 independent cutsets of $, Then, (C*,2) forms 1 matroid, defined as entset space ‘matroid :4{S) of S. A base of -4-(S) is a cuset basis ofS, and its rank is given by N(S) ~ b(S). This matroid can also be defined using the member-cuset incidence ‘matrix of S. The tows and columns of this matrix correspond to membets and cutsets, respectively. The columns of this matrix ate either dependent or independent, Take the columns of the matrix as elements of ¢*, und independent subsets of columns as elements of 2 Then (c*,) forms a culset space matroid He{S) of. Example: Consider @ graph as shown in Figure 9.3. The non-empty elements of the set C* contains 7 eutsets {C}, C}, Ch. C4, Ch.C§, CF} and = (CH), (2) C3, (CGI. (C3), (C5 )s (C5 Ws (CF CHI (CFC), (CCD, (CLD), AC{.C§+C}))}- A typical base of the cutset space matroid, can be taken as B, = (Ch.¢.¢4) CHAPTER 9 Matroids and Skeletal Structures 303 Fig. 9.3 A graph and its eutsets 934. CYCLE MATROID OF A GRAPH Spanning trees of a connected graph (spanning forest when S is not connected) have various applications. Some of its applications in structural engineering are described in Chapters 3, 5 and 6. In the following, a eyele matroid is defined in different inter-related forms, a base of which is a spanning tree of S: Let $ be a graph, Consider $ as the set of members of S, and let X © if and only if X docs not contain a cycle of S, i. itis a cycle-free subgraph (subtree if S is connected and subforest if it is disconnected). Then # is a collection of independent sets of a matroid in J; known as the cycle marroid of S, denoted by 8). This matroid is called a polygon matroid by Tuite [241], “Alternatively, let $ be a graph and consider the set ofall spanning forests of Sas It can easily be shown that is a base set of a matroid .1/~ (5,29 on member set 48) of S, known as the eycle matroid of S. Similaely, et C denote the set of simple cycles of a graph S, then C is the set of cireuits of a matroid a/on member set M(S), called a cycle matroid of S. The rank ‘of 48) is N(S) ~ bp(S), and for a connected graph itis NES) = 1 Example: Consider a graph S as shown in Figure 9.4(a). The sets sand for the ‘eycle matroid of $ areas follows: = fm sm, (ms, 3) Cam, 9) 5 (ms, ms, my). Amst) and = {Cm )y ( {msn oy (1M). ‘A typical base of 4S) may be considered as B, citeuit of 44S) cam be taken as { mmm} oF (mts mamma}, and a typicalz (@) A graph S. (b) A typical base of 44S). Fig. 9.4 A graph $ and a typical base ofits eyele matroid, 93.5 COCYCLE MATROID OF A GRAPH Let $ be a graph and ¢* denote the set of cutsets ofS. Then, C* is the set of circuits of a matroid on (8), called a cocycle or culser matroid of $, denoted by 4S), Obviously, a set X of members of Sis a base of the cocycle miatroid (8), if and only if M(S) ~ X is a spanning forest of S. For a connected graph, the members of M(S) ~ T are known as cotrees of S. The rank of 44S) is given as 1(4¢(S)) = M(S) -NiS) + bu(S). Definition: Let /be a matroid on ¢ whose bases are B,. The collection of sets s° ~B, are bases of another matroid 4¢ on .¢ known as the dual matroid of 2% This dual matroid is unique for an and the dual of a dual matroid is itself. Circuits of A¢are called cocireuits or culsets of 44. By definition, it follows that the cycle matroid 448) is the dual of the cocyele ‘matroid 4¢(S) of a graph S. Example: Let § be a connected graph as shown in Figure 9.5. Fig. 95 A connected graph S. CHAPTER 9 Matroids and Skeletal Structures 305 ‘The circuits of 4S) and .40(S) are as follows: (8) = {mmm}, and CCS) = {amy} fom, my} fom, ma} fm, my}. 93.6 RIGIDITY MATROID OF A GRAPH Graph theoretical approaches fo the study of the rigidity of planar trusses are described in Chapter 3, Further investigation has proved the suitability of matroid theory to the study of rigidity, Refs (15,1210,249], In this section, a simple definition ofthe rigidity matroid of a graph is given. Let § be a graph, and define the support WF) of a subset F, of 4S) as the set of end nodes of members in F; Define a subset F, of 4S) to be independent, if IMCR})| S 21NCR| ~ 3 holds for all subsets of F These independent sets collectively form # and (M(S),2) form a matroid known a a two-dimensional generic rigidity matroid a(S) ofthe graph S. A graph S is called rigid i r44S)) = 2N(S)~ 3. A cireut of {S) isa minimally dependent subset of 4S), examples of ‘which are depicted in Figure 96. Ry Se eI Fig. 9.6 Examples of circuits of #(). 9.3.7. MATROID FOR NULL BASIS OF A MATRIX Matroids are employed in combinatorial approaches to the force method of structural analysis, as will be described in Section 9.5, Matroids ae also used in algebraic force methods, Ref [74], brief description of which i given here. Let Ax = b be the equilibrium equations of a structure, where x and b are the ‘veciors of internal forces and applied loads. Fora statically indeterminate structure, ‘Ais an mn rectangular matrix with n 9T 02) Where mio is chosen such that it has the smallest weight and does not form a cycle With my Um, U... U mi, A shortest spanning tree will then be obtained CHAPTER 9 Matroids and Skeletal Structures 307 ‘This method formed a basis for the Greedy Algorithm for matroids, devetoped by Rado [199} and independently proved by three different authors, Refs. [42,54,243} Greedy Algorithm Let 4/= (577) be a matroid, Assign a positive value fo each element of denoted by WG), $¢ J For asubset X € %, define a weight function as, WO) =EWis), 3) ‘where summation is taken overall elements s, of X< ‘The problem is to find @ subset Xoy of J such that Xue € and W(X yy) is ‘minimum (or maximum) over all elements of 5° The Greedy Algorithm proceeds as follows: Seleot an element s; of minimal (maximal) measure (weight) from J and let Fy = {81}. Form Fs from F, by adding an elements, of minimal (maximal) measure such that F, is an independent set from J°— {s)}, and let F; = Fy+{ss). Subsequently ‘choose $y of minimal (maximal) measure from J°~ {8182-8} sueh that Fy is still an independent set. This process is clearly finite, and the finally selected set is base of minimal (tnaximal) measure for ‘An elegant proof of the minimality of the selected base may be found in Refs. (243,244), Example: Consider a graph $ as shown in Figure 9.7(a), with some positive Weights assigned to its members. A base of minimal measure for cycle matroid oS), which is a spanning tree of minimal weight, is selected as depicted in Figure 9.10), (a) A graph s.308 ‘Structural Mechanics: Graph and Matrix Methods (©) Expansion process of the Greedy Algorithm. Fig. 9.7 A graph and the selected minimal base for its 448) matroid, 9.5 APPLICATION OF THE GREEDY ALGORITHM: A COMBINATORIAL FORCE METHOD. In Chapter 6 it is shown that, for an efficient force method, the sparsity of the Aexibility matrix ofa structure, which is pattem equivalent to the generalized cycle adjacency matrix of its graph model, should be maximized. This can be achieved by the use of a generalized cycle basis of minimal measure, where the weight of a ‘poyee is taken as its length (the umber of its members). The Greedy Algorithm is 4 powerful means for this purpose. However, its application engenders certain 4ifficulties, which form the remainder ofthis chapter. Leta ycyele ofa graph S be defined as a minimal subgraph C of S, which is rigid and {C) = 2. A maximal set of independent ycycles of S is known as a generalized cycle basis ofS, the dimension of which is equal to 9($) = Sy/a. The integer "a" is defined in Table 2.1 (p. 44). The set of all teycles ofS, together with #eontaining independent subsets of cycles, form a matroid (8), called the generalized eycle space matroid. A base of this matroid isa generalized cycle basis of S. Therefore, the Greedy Algorithm solects a minimal generalized cycle basis of Algorithm: Let the weight of a yeycle be measured by the number of its ‘members, Select all y-eycles of S, denoted by and proceed as follows: ‘Step 1: Select the first y-eycle of minimal length from Step 2: Take the second independent y-cycle of minimal length from ¢°- {C)}. Step k: Subsequently choose a y-cycle C, of the least length from @— {C1, Co » Cy} which is independent of the previously selected ‘-eyeles. Continue the CHAPTER 9 Matroids and Skeletal Structures 309 process until B(S) of yFeycles, forming a minimal generalized cycle basis, is generated, Proof, Let B be an optimal generalized cycle basis of S, and let be the basis selected by the Greedy Algorithm. Let the -eycles off be ordered and denoted by: ps Gch Ct where C= UC, e+) Then a peyele C,€ P exists such tht: Camu BOB eB es By the well-known exchange theorem of matoids, there exists Cy, © such that, 0-6) Cds forms a basis, Moreover, Cy, € C* must hold or else B” would contain only 8(S) 1 distinct peyeles. By definition, L(B) $ E(B"), which implies L(C) 2 LC). But the inequality cannot hold since C, would then have been selected in place of C by the algorithm at that stage, and therefore L(B") = LB) Henee B° is also an optimal basis which has more yeeyeles in common with B than B. Continuing to reduce in this ways it becomes evident dat L(B) = 1) and thos B forms a minima generalized cycle basis for S. 9.6 PROBLEMS WITH APPLICATIONS OF ‘THE GREEDY ALGORITHM In practice, three main difficulties are encountered in an efficent use ofthe Greedy ‘Algorithm: L. The formation of a y-eycle; provision ofits rigidity.310 Structural Mechanies: Graph and Matrix Methods 2, The formation ofall y-cycles ofS. 3, Checking the independence of each selected ypeyele. In the following, the above problems are discussed for different types of skeletal structures listed in Table 2.1: 9.6.1 PLANAR AND SPACE FRAMES For this type of structure, ¥(S) = aby(S) (G3 or 6), and c:S.E.Ss can be generated ‘on each cycle of the selected cycle basis. For maximal sparsity of flexibility matrices, optimal eyele bases should be formed. However, minimal eycle bases are often preferred due to simplicity and applicability of the Greedy Algorithm and ‘many other topological methods. Only minimal eyele bases are discussed here. First Problem: For this type of skeletal structures, a cycle is rigid both in the plane and in the space; therefore rigidity is no problem. Second Problem: Formation of all simple eycles af a graph, esp structures, is quite time-consuming and impractical. Horton [82] 1 number of eycles to be formed in conjunction with the Greedy Algorithm to M(S)M(S). The author's expansion process requires far fewer cycles t0 be ‘generated, of course atthe expense ofthe formation of slightly longer eycle bases. ‘Third Problem: The simplest method for ensuring independence is the use of the cchords of a spanning tree as the generators of independent eycles. This, in general, leads to the formation of very long cycles. This algorithm is modified by ‘employing an SRT rooted at an appropriate starting node. Fusther improvement has been achieved by using ordered chords and addition of the used chord to tree ‘branches at each stage of the expansion, Independence can also be ensured using the admissibility condition, restricting the increase of eyele rank by unity in each step of an expansion process, as has been described in Chapter 6. An alternative method is the use ofa completion process, Ref. [83]. A simple but rather expensive approach to ensure independence is the application of an algebraic method such as, Gaussian elimination, However, this approach reduces the order dependency of the expansion process, A mixed version of admissibility condition and Gaussian v}. As an example, two circuits n(b,) = {bubs.bi} and n(b,) = (bs.bsJbnby} can easily be formed. However, if n(u) does rot have WHP, then it contains a cireuit which should be identified by numerical factorization, Once the formation of a circuit becomes possible, different algorithms can be designed for the formation of a null basis. Two such algorithms are given by Coleman and Pothen (29,30), and in the following an algorithm for the formation. ‘ofa fundamental null basis is briefly discussed. Let N be the empty set, Find a complete matching M of A, partitioning the columns of Aas A= [M U]. Thea, for each uc U, construct a circuit n(u). Augment the null, basis N' with the computed null vector. This process should be repeated for all ‘members of U in order to obtain a fundamental null basis of A with WHP. When A does not have WHP, then a fundamental basis can be computed only when M has full rank. Therefore, one should choose M by a matching, but should ensure that M has full rank while factoring it to compute the null vectors. When it is rank deficient, the dependent columns in M should be rejected, and a new maximum ‘matching should be found. This will ensure the formation of @ basis and will always succeed when A has fll row rank. Remarks: Matroids make the compact formulation of many structural problems feasible, Onee a matroid has been constructed, the Greedy Algorithm can be used {or the formation of its minimal base, However, the efficient application of the Greedy Algorithm requires special considerations, similar to those discussed in Section 9.6.314 Structural Mechanies: Graph and Matrix Methods EXERCISES, 9.1 Specify Cand #for the eycle space matroid ofthe following graphy 9.2 Specify Cand for the cutset space matroid of the wraph of Exercise 9.1, 9.3. Prove that ¥(S) = M(S)~ N(S) + 1 can be considered as the rank function of amatroid 9.4 Construct a base for the eyele matroid ofthe following graph: 9.5 Forthe following graph, specily two citeuits for the eyele space matroid and its dual 9.6 Use the Greedy Algorithm to form a minimal base for the cycle matroid of the following graph. The weights are shown on the members ofthe graph: CHAPTER 9 Matroids and Skeletal Structures Bis 9.7 Using the Greedy Algorithm, form a minimal base of the eycle space ‘matroid ofthe following graph. Take the weights of all the members as unityCHAPTER 10 A Graph-Theoretical Approach for Configuration Processing 10.1 INTRODUCTION For a large system, configuration processing is one of the most tedious and time- consuming parts of the analysis. Different methods have been proposed for configuration processing and data generation, among which the formex algebra of Nooshin [180,181] is perhaps the most general and powerful tool for this purpose. Behravesh et al. [9] employed set theory and showed that some concepts of set algebra can be used to build up a general method for describing the interconnection patterns of structural systems. There are many other references om the field of data generation; however, most of them are prepared for specific classes of problem. For example, many algorithms have been developed and successfully implemented ‘on mesh or grid generation, a complete review of which may be found in a paper bby Thacker [232] and a book by Thomson et al. [234]. In this chapter, a graph-theoretical approach is presented for configuration processing, which uses similar concepts developed for the set theoretical and formex algebraic methods of Refs. [9,180,181], with the difference that it avoids the use of new terminology, and employs the most elementary definitions of graph theory. Four basic functions are presented here, which consist of translation, rotation, reflection and projection functions. ‘The algebraic representations described here not only provide the topological properties of a structure, but also provide simple means for obtaining the goomettical properties of structures using, simple transformations. Examples of skeletal structures are included to illustrate the simplicity of the concepts presented in this chapter. The application of the ‘ansformations is extended to the generation of finite element meshes.a8 Structural Mechanies: Graph and Matrix Methods 10.2 ALGEBRAIC REPRESENTATION OF A GRAPH. Various algebraic sepresentations of graphs were studied in Chapter 1. These ‘asthe node-member incidence matrix, ajaceney matrix, member list, and in particular the following compact member list, containing the end nodes ofthe members of a graph ina single row. ‘As an example, the interconnection pattern ofa graph S, shown in Figure 10.1 can ‘be represented by, 2513523534, or, ina forma compatible with the notation used inthis chapter: {U2ML3}23),340 Fig. 10.1. A graph S In this scheme, effectively a set of members is specified; however, the common ‘node numbers provide the interconnections ofthe members. All these representations specify the corresponding graph up to isomorphism. In structural engineering, however, a structural model should uniquely be represented. ‘Therefore some additional information must be provided. As an example, the coordinates of nodes in Euclidean space ean be given. In such a ease, however, the sgeometty of the structure will also be involved. To have only the connectivity, a ‘weaker formulation will be sufficient. For this purpose, one may use an integer coordinate system for specifying the interconnection of S. Further simplicity can be achieved if a rectilinear grid system is employed. Such a system may be a 1, 2 or dimensional coordinate system, an example of which is depicted in Figure 10.2. 319 Fig. 10.2. A 2-dimensional rectilinear grid system, From now on, any member of the set of such systems will be referred to as an integer coordinate system (ICS), An ICS has grid lines and grid points, as shown in Figure 10.2. ‘A single node of a graph will be shown by a grid point. A member of a graph will be spocified by its end nodes. As an example, =) represents the node nad (2.2)4.3)1 js an algebraic representation of the member my, as illustrated in Figure 10.2. This isa directed member, and, if no orientation is assigned, then one ean represent also as, my =[43.2.2)) i.e. an arbitrary order can be used for the end nodes of in genera ti pretrable fous the is quater of an ICS in onder ove postive sae ie rdnatcs ofthe nodes Howeve, fr some reasons the eames ee gat then both postive and negative values wll be past. Asan 1CS pee tic rprenetton of yeh 8 Synmeti membst With Teopedt 0 ngven as (-1.2(1.2)) my320 ‘Structural Mechanies: Graph and Matrix Methods [A subgraph (or a graph) can also be represented in an ICS by specifying its ‘members, andthe integer coordinates of ts end nodes, Figure 10.3. Asan example, the subgraph S, canbe representa as: (2.3.B.2G2.2,0M20.00 L nf ma Si mi ne me [as h Fig. 10.3 A member and a subgraph in an ICS. ‘To simplify the notation and the language, from now on we will not distinguish ‘between a graph (or subgraph) and the corresponding algebraic representation. It is interesting to note the difference between the above representation and the previous scheme which specifies the graph up to isomorphism: ie Si ((L2)23)3.4)) In fact, the integer coordinates of each node guarantee the unique representation of Si. ‘An algebraic representation such as the one above allows one 0 assign weights with nodes, members, and subgraphs of a graph, These weights can be specified loads assigned to the nodes, relative stiffnesses associated with members, or group properties of substructures. This can be done in the same way that numbers were assigned to nodes, members and cycles of a graph in Chapter 1. However, inthis chapter it is unnecessary to consider such factors, since only the intereonnections of the structural models are of interest in configuration processing. CHAPTER 10 Configuration Processing 321 10.3. REPRESENTATIONS OF OPERATIONS ON GRAPHS 103.1 ADDITION OF TWO SUBGRAPHS. Consider subgraph S,, which is joined to Sp through a member, as shown in Figure 10.4(a), resulting in $1 U Sy > > i S\US: SOS @ o Fig. 10.4 Addition of two subgraphs. Representing S, U S2 in an ICS, one writes: Si= 12V.23)MC23).0,2M0.221)- and 82= (2D.G.2M2ZV2IMG2,230. “The algebraic structure of $; U $2 is obtained by omitting the repeated intersection of two subgraphs. Therefore, S14 S.-112.0,2,9M23M1,910.2.2,0ME0.E.2NE2R2I1, in which one [(2,1),2,3)] is deleted in composition. From now on, such @ representation will be denoted by $,+S:. Ifa series of compositions as,322 Structural Mechanies: Graph and Matrix Methods 8, US; US} U.S, (10-1) is petfotmed, then the conesponding algebraic representation will be shown by different summation sign, as the following, in order to emphasis the composition nature of the operation S454. 4S, =a (10-2) In order to avoid the extra operation of deleting the algebraic representation of ‘common members, composition may be considered through the common nodes only, ie. S; and 'S; may be considered as illustrated in Figure 10.5. tn this composition, $; 7 S; contains no member, and the composition of two subgraphs is simplified and consists of simple adaltion of thei algebraic representato I>. Sin Ss: S: S\US; Fig. 10.5. A simple composition 10.32 SUBTRACTION OF TWO SUBGRAPHS Consider a subgraph Sas shown in Figure 10.6, Take Sas the star ofthe node n,, ‘A subgraph S$; ~ S: is a graph containing the star of ny. This subtraction in algebraic form is very simple and contains the deletion of the members ofthe star of, Le. for, S110 D.GB.DLE.HGILCSC3MK03,000 and S=110,).G,DBV.G3I), ee CHAPTER 10 Configuration Processing 323 Fig. 10.6. Subtraction of two subgraphs the algebraic representation of this subiraction is given by: 8) -$:=116,9).0,3)) (0,3... Cutouts frequently occur in structures. Sometimes, in order to keep tho regularity of the structure, itis advantageous to generate the entire structural model and then subtract (delete) certain parts, This can be achieved by subtracting the stars within the cut-out. As an example, for S with cut-out shown in dashed lines (Figure 10.7), the main structure, S,, can be formulated as: S_=S-Y,Starny, (03) Fig. 10.7 A plane grid with a cut-out.324 Structural Mechanics: Graph and Matrix Methods 10.4 SPECIAL GRAPHS ‘A mull graph containing only isolated nodes, denoted by So, can be represented by ifs nodes. As an example, the null graph Sp of Figure 10.8 can be represented as: S0= (E0.2,0.B,D4,0.1,3,2.2)8.2)4,2,0,923)G343)} Fig. 10.8 A null graph S;, Asa second example, the null graph containing the isolated nodes n), ns, ny and ny of Figure 10.7 ean be waitten as, Sx = (2.2),8.2.2.3.8.3)), ‘where Ss the mull graph of S, Conversely Sy may be considered as Star (Sp) in S, where NGS20) Star (Sp) =U Starn; where n, € S:, (10-4) and N(Su) is the number of isolated nodes of Sp, Therefore, in Figure 10.7, the sid withthe cut-out can be represented as Sy = 8; ~ Star (So). ‘The null graph, or the null graph of a graph, plays an important role in structural analysis, As an example, the null graph of a part of a structural model may Fepresent the boundary nodes, or those nodes which are loaded inthe structure; and therefore compact representation of such nodes is advantageous in data generation, ‘The null graph of a graph S; will be denoted by Sy, CHAPTER 10 Configuration Processing 325 AA similar convention can be applied to a complete graph. Since in a complete ‘raph, all the nodes are connected to each other by distinct members, one can specify only its null graph with an extra command to indicate the completeness of the graph, For example, COM (Sia) can be used to refer to the complete graph Sy. 10.5 SOME FUNCTIONS FOR CONFIGURATION PROCESSING ‘Many practical structures have a regular and repeated pattern, which enables their generation using simple functions. As an example, S in Figure 10.9 can be generated by translating $), shown in bold lines, in the Iy direction; ie. if the coordinates of the nodes of S, are each inereased by 2 units, the adjacent subgraph Sp will be obtained. A similar operation on S» results in Ss, and the same ope 4 on S; forms, completing the generation of | Si 10.9 A typical S, for generating $ by translation, ‘A second example is depicted in Figure 10.10, in which S is generated by rotating typical subgraph S; shown in bold Lines. Rotation about O by 90° leads to S, and 4 similar operation on S; results in Ss, ete. Therefore, S= $; U SU $y U Sy ean be easily generated.326 Structural Mechanies: Graph and Matrix Methods 3 2 1 Si (ag, (Oar Fig. 10.10 A typical S, for generating S by rotation, An operati bby means of which §; is obtained from Scan be shown as, S418: (10-5) ‘where 6 can be considered as a mapping of S; into Sj or it may be taken as a fimotion which defines the relationship between S, and $;, The following functions are defined for (a) IfS;~ 915, then $,~ 6" |i. the inverse of function 6 exists (b) ‘The identity function is supposed to map S, onto itself This function is denoted by @": Le. S\= 9°15, (©) Repeated application of on S, will be denoted by |S, which means 6 has ‘been applied r times, and each time @ operates on the result of the previous step, (@) The composition of functions 6" and 6” is equivalent to In the following, various Functions and their significances are described, which in the main follow those of Ref. [180]. 10.5.1 TRANSLATION FUNCTIONS A translation funotion operating on S, to produce §, is denoted by, S,=@1S\= TRAN (hg)l Si, (10-6) CHAPTER 10 Configuration Processing 327 in which h~ 1, 2 and 3 depending on translation taking place along I, Iz and 1), respectively. q is the number of integer units to be added to each coordinate of algebraic representation of $, in direction h. This operation can be stated as follows: Let a node of S; be denoted by (Iy.JJs) and that of $; be denoted by (I. 13,15) The function TRAN (b,q) maps the nodes ofS; onto those ofS; such tha, foreach node, 1, for i=1,2,3 except # and Kelte (10-7) Obviously, q should be taken as an integer, ifthe new node is supposed to be a grid point Example: Consider a graph S, as shown in Figure 10.11 : tea ll err tase : Fig, 10.11. Subgraphs obtained by transtations. Scan be waiten a S)={0,0.2.)M2,0,2.2)} Similarly, S; may be represented a: S)= TRAN(L 4) |$\=16,D (6,016,620). ‘Then S; can be obtained from S; by: S)= TRANG22)|82 ={16,3)63)]{(6.3)(64)]} Composition of the above functions results in Ss, when S) is given, i,328 Structural Mechanics: Graph and Matrix Methods Ss= TRAN(,2) | TRAN(4) | $i ={((5,3)(6.3)]M(63)(6.4)). Example: A simple rectangular grid is formulated. Consider a graph as shown in Figure 10.12, Fig. 10.12 A rectangular grid obtained by pure translations. Different units (subgraphs) can be considered for generation. Naturally, one should ‘consider the sinallest unit for which the translation functions can also be written in ‘simple form, Let: $)= {0,29} and S$; ={10,0,2,1))) ‘The members ofthe vertical edge of S containing S, can be represented as TRAN (2,0))8) =), TRAN (2,1); and TRAN (2,2), The above three members can collectively be shown as: TjoTRAN @H)Ss, ‘Translation of the resulting subgraph in the I direction completes the formation of all the vertical members,ie, Sy= Eh TRAN (1.i) | 2. TRAN (2,) Si, nd ‘The horizontal members can be formed by a similar approach leading to TRAN (1,i)| Ss. Su= Dj. TRAN GS)! Sf Naturally, S* Sy + Sy, CHAPTER 10 Configuration Processing 329 jo TRAN (Li)! jpg TRAN Qi) | (CD,C290) + Jeo TRAN (2) |Sf.g TRAN (Li) | ((1,1)2,00. 5 as a generating unit, Asan alternative formulation, one may consider S, US; Ss° LOD0.2MOD20). ‘Then Se= DTRAN Q5) |Dj.g TRAN (1, Ss, a generating all members except the upper and right-hand side boundary members, which can be generated as S,= TRAN (1,5) U}_g TRAN 24) |S; + TRAN (2,3) 4g TRAN (1) |S ‘Therefore, $= S, + Ss generates the entire model S, Translation functions have the following properties: 1, TRAN (hy,q)) | TRANG) | S; these functions are commutative, TRAN (haa) | TRAN (bia) | $i i 2. TRAN (hq) |S.= TRAN (h, -a) S, 3. TRAN (hq) | TRAN (hugs) |S: = TRAN (b,qrtaa) [Su Similarly, TRAN (hq) | S,= TRAN (1,4) |S 10.5.2 ROTATION FUNCTIONS This function rotates a subgraph S, with respect to a specified node to obtain S, and is denoted as: $= ROT (hy bhs.43) (10-8) Let (lish) be atypical node ofS, and (1.13.1) be the integer coordinates of the odes ofS, The rotation function maps the nodes of S; onto those of S; such that, for each node,330 Structural Mechanics: Graph and Matrix Methods 42+4)—lyp for i=hy, 2-4) + Ty for i= ho, (10.9) and = forthe remaining value of Obviously, the numbers q; and q; should be such that their sum and difference become integers, unless an even power ofthe function is used, Example: Consider S; as shown in Figure 10.13, This subgraph ean be represented S.=0.20M2D.2020. Fig. 10.13 A subgraph S; rotated to 8, For ROT (1,2,2,3), we have, Sy = ROT (1,2,2,3) | 81 (14.24. 9M43)G.3))). ‘which is depicted in Figure 10.13. ‘The rotation function ROT (hy.t.q,42) rotates S, through w/2 about an axis perpendicular to the tj, ~ tio plane and intersects this plane at a point whose coordinates are qy and q2. The sense of rotation is such tha a rotation of Ty through ‘1/2 about the origin maps the positive side of ly onto that of I. ROT (hi,bs,q,) represents the roation ofS, by kn CHAPTER 10 Configuration Processing 331 For a general ease, when the rotation isan arbitrary angle simple formulation ‘ean be made, Consider two points A(I:) and A’(ij,13) as shown in Figure 10.14 ‘The centre of rotation is taken as O'(q1,42) The following relations are obvious: 0A =0040 cc ON=00%407A" Projecting the vectors OA and OA’ on and I: axis, aon, ‘here L is the length of O'A., which is the same as OA", since rotation preserves the length. Using Eq, (10-10) and projecting the components of the vectors OA and OA’ om Iy and I; leads to: ‘41 + Leos(er+B) 1+ Leos acosB—Lsin asin 42+ Lsin(a-+B)=15 =a +LsinBeosB—LsinBeosa 12) Bq, (10-1) ad Bg, (1012) feast: 1 (lay) eos (2 —43)sinB 1043) 12 —(la +a) 00sB-+(hh-an)sinB aor) Functions for a genera ation are obained as: Y= 41 —q1cosB+qp sinB +1, cos Ip sinB co 13 =42~4p cos +4; sin = 1p cosB +1; sin Substtting B= 90° leads to the special ease of Ea, (10-).332 Structural Mechanics: Graph and Matrix Methods Fig. 10.14 General rotation, Example: The graph mode! of a planar truss, as shown in Figure 10.15(a), can be generated using rotation functions as follows () The graph model S. CHAPTER 10 Configuration Processing 333 (0) isrotated, (6 The subgraph of () is rotted, : . sh oss 7 7 a @Ss. (¢) S: is rotated. bh i fos (0 The subgraph of (e) is rotated, @s Fig. 10.15 A planar truss generated by pure rotations.34 Structural Mechanics: Graph and Matrix Methods Consider Si {14,0,5.2))) and S2= {{04,,6,0)} First Sand Sy, shown in Figures 10.14 (band), are formulated as $= YL gROP E21} ROK .26.) /YP gROT252)S, and ThgROT 285) |S% ,ROT (243) |ROTLL252} |S; Now, the tire model , we hve S=S)+S) Rotation functions have the following properties: 1. Rotation functions are notin general commutative, except for some special 2. For an integer k, ROT (hihaqua:)"S) = Si ie. ROT (hyshoquas)* is an identity Function, 3. The inverse ofa rotation function isthe cube of itself i.e. ROT (hy...) '}S;= ROT (hha ina) Si 10.5.3 REFLECTION FUNCTIONS This function reflects a subgraph S, and finds , as: Sy REF (hq) |i, (10-15) Like the previous functions, a reflection function REF (h,q) maps nodes of S, onto those of S; by the following rule: Ty = 24-1, for and for the remaining is (10-16) ‘Obviously, 2q should be integer if the new coordinates are intended to be integers. CHAPTER 10 Configuration Processing 335 Example: Different reflections of S, are illustrated in Figure 10.16 Si (2D22M2.240,291(12321)0. Fig. 10.16 Subgraphs generated by reflections Now 8) = REF (2.2.5)/8)= (2.4,2,3M23\13)HU3.24)}, and S)= REF (13.5)(8) = (6,5, 2MG6.26262)6.0) Similarly, 8, =REF 2.2.5)|$3= (6.4.53) MG363M635A]) For a reflection function, Sis the mirror image of S, is nomnal to the I, axis and intersects it ata point with th respect to a plane which 4@ Example: The graph shown in Figure 10.17 can be formulated using a single member a follows ‘Take 8; = {(2,1)]{(1.2)]} as a generating subgraph, Then: REF (2,2) |Soj-g REF (14) | Dag REF (1,2)'| Si336 Structural Mechanics: Graph and Matrix Methods 4 Fig. 10.17 A graph generated by pure reflections. Reflection functions have the following properties: 1, For hy #hs, REF (hy,qy) | REF (hq) |S, REF (hq) | REF (191) |S 2, REF (h,q)" | S;= REF (h,q) |S, also REF (b,q)* | S,= and REF (h,q)"*"' S\ = REF (hq) |. 10.5.4 PROJECTION FUNCTIONS. ‘A projection function is defined as, PROJ (bg) |S; (10-17) which is a mapping of §, onto S, such that: Tye a forich and {=I for remaining is (1.2.3). (10-18) Naturally, q should be integer if the nodes of $; are preferred to be on the grid points, Example; The projections ofS, S; and S, are illustrated in Figure 10.18 S)= CUDM220, S:=PROI (1,4) |8)= (4, D140), Ss= 2..CICICHLCA2I CHAPTER 10 Configuration Processing 337 and S.= PROJ (2,1) [$= ((2.0,G,0MG.D,Q1)MQ1),2,1))} ig. 10.18 Projections of subgraph. For a projection function S, = PROS (hq) | Ss Sis obtained from S; by projecting onto a plane that is perpendicular tothe I, axis and that intersects the plane at a point where I, q Projection functions have the following properties: 1, A projection fanction has no inverse; 2. Fort #hs, PROM (h,a}PROF (haa) | Sim PROJ (h,92)PROS (by 4))85 3. PROS (hq | | PROJ (hg:) | PROM (ha) | S,= PROJ (ha) |S: It should be noted that the useful functions are by no means limited to the four types introduced inthis section, One ean define functions which are more suitable for a particular application, Many such functions are defined in Ref. [136], Compact algebraic representation of a structural model is one of the aims of the ‘methods presented in this chapter. For this purpose, it may be necessary 10 use @ ‘combination of functions, which are presented or can be defined. However, optimal representation of an arbitrary configuration in the sense of compactness is an open problem. Example: In the following example, a combination of translation and rotation functions is used for algebraic representation of a structural model. Let $ be a double-layer grid as shown in Figure 10.19. A 3-dimensional integer coordinate system is used for its presentation, ‘The top layer of Sis represented as:338, Structural Mechanies: Graph and Matrix Methods TRAN(2i) |S} gTRANC ICL 1LGL 0), TV= Fig TRANG2 |S} gTRANC2CLIN(3,0] Fig. 10.19 A double-layer grid 8, TThe bottom layer can be formulated as BH = Sg TRANGL2i) 1} pTRANCA)|(2.2.0)42,0)}, BY= YL TRAN( i) | Lo TRANC2) | (2,2.0),2,40)] ‘The rest of the model can be represented as BT=Y}.gROTU.2,2,2)i[ (010,220) BTL = YA} TRAN(2) |, TRANG.ADIBT, bra= Y2TRANGAD | [TRAN 2.) [Br, The algebraic representation ofthe entire model ean now be shown as: S=TH+TV+BH+ BY +BTI+BT2, CHAPTER 10 Configuration Processing 339 10.6 GEOMETRY OF STRUCTURES ‘The algebraic representations described in the previous sections not only provide the topological properties of a structure, but also provide simple means for ‘obtaining other properties. As an example, the geometrical properties ofa structure cean be obfained using simple transformations. All that is needed is to establish a relation between the integer coordinate employed and the coordinate system chosen for geometrical presentation of the structure. This coordinate system ean be an orthogonal Cartesian coordinate, a polar, a cylindrical or any other curvilinear coordinate system. Example: Consider a grid mod! as shown in Figure 10.20, O§e5ggdg0D Fig, 10.20 A planar grid S. Let x and y be the geometric coordinates of a typical joint of the structure with respect to a selected two-dimensional Cartesian coordinate system, Consider the following transformation: 1) and y=2h, (10-19) ‘Then the geometry of $ will be a graph as shown in Figure 10.21340, ‘Structural Mechanics: Graph and Matrix Methods Fig. 10.21 Geometrical properties of S. Now consider the same graph and use a polar coordinate system with the following transformation: (10-20) Then S becomes a configuration as illustrated in Figure 10.22. ay Fig. 10.22 Grid S mapped to polar coordinate system. ‘The same model can be mapped into a three-dimensional eylindrical coordinate system with the following simple transformation: (10-21) ‘The resulting configuration is depicted in Figure 10.23, CHAPTER 10 Configuration Processing 34 Fig, 10.23 Grid S mapped to cylindrical coordinate system. It should be noted that the connectivity coordinate system defined in Chapter 7 can easily be combined with the concepts presented in this chapter, to obtain a favourable nodal numbering of a structure, Kaveh [91] 10.7 EXTENSION TO HYPERGRAPHS For skeletal structures, the model can be represented as a graph; however, there are other types of structure having more than two nodes per member. Finite element ‘models are examples ofthis kind. For these structures, the madel can be considered a a hypergraph, Berge [13]. A Jopergraph consists of a set of nodes, and a set of ‘members with a relation of incidence which associates some nodes with each 'member. A special case is when two nodes are associated with each member; then a hypergraph becomes a graph, Many of the methods presented thus far stay valid for hypergraphs as wel. Example: A finite element mesh consisting of triangular elements is considered as shown in Figure 10,24. This model ean be formulated as follows: A typical element m; can be represented as: m= ((,0,2,1,2,2))) Similarly: m= (((,0,2.2)1,2)])342 Structural Mechanics: Graph and Matrix Methods ‘These elements may be shown as =m m= (0 D.2),22M0,0220.200. afl Sal ath ea Lal ‘ lll Z Z) RR 5 Fig. 10.24 A hypergraph as the model ofa finite element mesh, Using translation fumetions, the alzeb : representation of the entire FEM S= Zig TRAN(LD| Leo TRANG, |S, ‘Therefore, mesh generation can be considered as a special ease ofthe configuration processing discussed in this chapter. Example: A frame structure is considered with its shear wall modelled as triangular finite etements, Figure 10.25. ‘The algebraic representation ofthe model can be written as: F= Yi gTRANG2) Y} gTRANG.2))| (ADLIMD.G3)), W= Y}oFRANG2i) Dj. pTRANG2)| EO} ROTE.22)17.0,8.2.0.D) S\o F+W, and i ao REFULLD)| Sy, 343 eats ee rato Ne sacs rion Fig, 10.25 A frame with a shear wall It can be concluded that, simple definitions and concepts of the theory of provide a powerful tool for a complete formulation of configuration processing. The ideas are by no means limited to topological graphs, but can be applied equally to abstract graphs and hypergraphs, preserving the generality of the formulation, Finally, it should be mentioned that the functions presented in this chapter can be programmed and employed in data generation of large-scale structures. Other functions suitable for particular applications can also be formulated and used. EXERCISES 10.1. Write the algebraic representation of the following graphs in an arbitrary “TO AY 10.2 Plot the graph corresponding to the following algebraic representation in an Ics: S= Lj.g TRAN (1,3i)|Zj-p TRAN (2) | ((Us1).2,2)M3,12.2)0.344 Structural Mechanics: Graph and Matrix Methods 10.3 Write the algebraic representation of the following structural model using translation functions: 10.4 Plot the graph model of the following algebraic representation in an arbitrary ICS: $= Li gROT2.6,6(6.4),2.294,0.2.20 10,5 Write the algebraic representation of the following structural model in an arbitrary ICS, using reflection functions: 10.6 Using translation and rotation functions, write the algebraic representation of the following tower structure inthe given three-dimensional coordinate system. (CHAPTER 10 Configuration Processing 345 | 4 4 1 10.7 Write a computer program to generate a configuration with the translation functions of Section 10.5.1. 10.8 Repeat 10.7 with the rotation functions of Section 10.5.2 10.9 Repeat 10.7 with the reflection functions of Section 10.5.3. 10.10 Repeat 10.7 with the projection functions of Section 10.5.4.CHAPTER It Graph Symmetry for Eigensolution of Dynamic Systems 11.1 INTRODUCTION ‘Symmetry has been widely applied to many problems in science and engineering [72,208]. Early applications appeared in chemistry [75,171,191]. In structural mechanics, symmetry is used to simplify the analysis and design [59-126]. Two ‘main mathematical tools are of great importance in studying symmetric engineering models, namely graphs and groups [255,256]. Here, graph theory is selected for this purpose to match with the scope of this book. The application of the methods presented in this chapter isnot limited to geometric symmetry. Thus the symmetry studied here can more appropriately be considered as ropological symmetry Large cigenvalue problems arise in many scientific and engineering problems [8,86,161]. While the basic mathematical ideas are independent of the size of matrices, the numerical determination of eigenvalues and eigenvectors becomes more complicated as the dimensions of matrices increase and their sparsity decreases. Thus special methods are beneficial for efficient solution of such problems, Methods are developed for decomposing and healing of the graph models of structures in order t0 calculate the eigenvalues of matrices with special patterns [124-126,134]. The eigenvectors corresponding fo such patterns are studied in reference [125]. The formation of divisors and codivisors, using a graph-theoretical approach, is developed by Rempel and Schwolow [203] and well described in reference [35]. Here, only symmetric forms are presented, since simple graph- theoretical concepts are sufficient for their formation.348 ‘Structural Mechanics: Graph and Matrix Methods ‘The theory of graphs can be applied to studying the dynamic behaviour of vibrating systems. Decomposition approaches are employed to form matrices of special pattems for mass-spring systems [126]. The problem of finding eigenvalues and eigenvectors of symmetric vibrating systems is transferred into calculating those of their modified subsystems. This decreases the size of the eigenvalue problems and ‘correspondingly increases the accuracy of their solutions, and reduces the ‘computational time. The methods of this chapter can easly be applied to the calculation of the spectrum of graphs. The second eigenvector for the Laplacian matrix of the graph, known as the Fiedler vector [51], can also be calculated. This vector is employed in nodal and element ordering of structures, as discussed in Chapter 7. The application can also be extended to graph partitioning required for parallel computing of large~ scale systems (118, 122,135,136], 11.2, DECOMPOSITION OF MATRICES TO SPECIAL FORMS In this section, an NxN symmetric matrix M is considered with all entries being real. For three canonical forms, the eigenvalues of [MJ are obtained using the properties ofits submatrices. 11.2.1 CANONICAL FORM ‘In this case, matrix M bas the following pattern: = [Tabs foe oll a avy with N=2n, Considering the set of eigenvalues of the submatrix A as (AAJ, the set of eigenvalues of [M] can be obtained as: AM} = (AV TAA} (ata) ‘Where U is the sign forthe collection ofthe eigenvalues of the submatrices. Since det M = det Axdet A, the proof becomes obvious. Form I can be generalized to a decomposed form with diagonal submatrices Ay, Aa, Ass o:+Ay of different dimensions, and the eigenvalues can be calculated as: {AMD = (Ag) T (ha) TAT. T (Aap (113) CHAPTERII Graph Symmetry for Eigensolution of. 349 "The proof follows from the fact that det M-= det Apcdet Aaxdet Agx ... det Ay, ‘Example: Consider the matrix [M] as, wana [f 3] Since (A) = (0.3723, 5.3723}, therefore (AM) ~ {-0.3723, 5.3723, -0.3723, 33723}. 11.22 CANONICAL FORM IL For this case, matrix M can be decomposed into the following form: [bea lh ioe tale . ‘The eigenvalues of [M] can be calculated as, {aM} = (AC) AD} «sy [C] and [D] ate called condensed submatrices of (M] Proof: In block form, addition of the second column of the matrix M to the first column and reducing the first sow from the second row results in: [[a+B} ey =i (B} FS [BI] (a1) UA+BI [Al (co) cal-t81]" Lio) 1) where C=IA] +B] and D=[AJ~[B} a7) and det M= det © x detD, (11-8) and the proof is complete350 Structural Mechanics: Graph and Matrix Methods Example: Consider the matrix [MI] as follows: This matrix has the pater of Form I, and is decomposed according to Eg (11-4), leading to: Matrices (Cand [D] are formed using Eq (11-5) as tel=tar+om1=[38 13), a wrm-[3 8) For these mates, the eigenvalues are, {RC} = (35.9459, 0.9459) and {AD} ~ (3.8172, 288172}, hence: (AM) = (RC) T {AD} = (09459, 8.8172, 288172, 35.9459} 11.2.3 CANONICAL FORM IIL ‘This form has an NxxN Form Ht submatrix augmented by k rows and columns, as shown in the following: Ly Tt u uy uy ly CNEL CONAN) CON+LNED, CINFLNEK) UN+KI) | ANKENY ZOV+KNSD, UN+KN+K) 19) CHAPTERII Graph Symmetry for Eigensolution of 351 Where [M] is a (NHOX(N+K) matrix, with a N>eN submatrix with the pattern of Form IT (N=2n), and k augmented columns and rows. The entries of the augmented ‘columns are the same in each column, and all the entries of [MI] are real numbers, C(ig) and Zi) are arbitrary real numbers. ‘The sot of eigenvalues for [M] is obtained as: (AM) = {AD} 0 (AE). (1-10) Proof; Similar to Form Il, here (M can be factored by rows and columns permutation, In this case first the augmented rows and columns are transformed into the central part of the matrix, The last column (in block form) is added to the first column, followed by reducing the first row from the last row. ALB P) [A imj=[Bla r|=|B QuR PB P Al> RH Q A PB) [AWB PB QR H/>/QeH Ro W>lQ+H R| HW] Cit Bpal [BHA PA ice. the matrix is now in a factored form, (E) (K}) + Le 7 [in iol vee wre and are constucted as follow D=rAl- f : . bao : ane) and dot M= dt D x det E. ana352 Structural Mechanics: Graph and Matrix Methods Although the entries of each augmenting column in the upper-right part of the matrix M are considered to have constant entry L., however, the main requirement is the repetition of the first N/2 rows and the second N/2 rows (submatrix P) of the augmented columns, Example: Consider [M] as follows: -1 05 |-0.7 ~0.7 )-10.3 34 [os 09 |-103 (M}=!=07 -07 | -1 05 |-103 (OSE t09E [Erase | 103) “13 =123 -133 13° ~57 orf Fa SLY 07 5-07 103) [17-02 103 ml aa, Seo [ay F283 and 133-113 -123+13 -37| [-246 -i1 =57) Eigenvalues for [D] and [B] are caleulated as: {AD} = (-0.9516, 3.7516}, {AB} = (1.6224, 17.6885, -21.8109}, ‘Therefore, the eigenvalues of [MI] are obtained: {2M} = {AD} G {AE} = {-0.9516, 3.7516, 1.6224, 17.6885, 21.8109}. 11.3 LAPLACIAN MATRICES FOR DIFFERENT FORMS, ‘The Laplacian matrix L{S) is an important matrix associated with a graph S. The list of eigenvalues together with their multiplicities of L(S), is known as the spectrum of 8. There are interesting relationships between the properties of a graph {and the Leplacian spectrum. Since a graph is the underlying model of a vibrating. structure or a system, therefore eigenvalues of graphs are of great importance in their study. CHAPTERII Graph Symmetry for Figensolution of 353 ‘The Laplacian L(S) = [wun of weighted graph $ is an NN matrix defined as, 148) =D(S)- ACS), (rts) ‘where N is the number of vertices ofthe graph and: (sum of the member weights connecting n; ton, forn; sum of the weights of membersconnected ton; for n; o otherwise ar-16) ‘When the weights of members are considered as one, then DS) and A(S) becomes the degree matrix and adjacency matrix ofS, respectively. 11.3.1 SYMMETRY AND LAPLACIAN OF GRAPHS. In this section, [M] is taken as the Laplacian of S, and for different forms of [Mf], the corresponding graphs are introduced. This comespondence provides an efficient means for calculating the eigenvalues of the Laplacian of graphs. Consider a symmeteic graph $ with an axis of symmetry. The following cases may ‘Symmetry of Form I: The axis of symmetry does not pass through memibers and nodes. In this case, $ is a disjoint graph and its components S; and S; are isomorphic subgraphs, Figure 11.1. In order to have the Laplacian matrix in Form I, the nodes of S, are numbered first as 1,2, ..., N/2, followed by the labelling of the nodes of S; as N/2+1, N’2¥2,...N, such that, for a typical node I in S, the corresponding symmetric counterpart in S; is labelled as 1+N/2, 81 Fig. U1.1 Symmetry of Form I. ‘Symmetry of Form I; The axis of symmetry passes through members, and the ‘graph $ has an even mumber of nodes. The members cut by the axis of symmetry ‘are called link members, and their end nodes are taken as linked nodes, Figure 11.2. Link members connect the two isomorphie subgraphs S, and S; to each other. For this ease, two different types of connections ate considered, as illustrated in354 ‘Structural Mechanics: Graph and Matrix Methods Figues 11.2() and (b). The fit one is «direct connection and the second is 1m a ditect connection, a typical node I in S; is connected to the node labelled as 14/2 in S; by a link member, In cross-connection, a typical pair of nodes I and 3 in S) is connected to J4N/2 and F+N/2, respectively. A combination of the direct tnd cross-connection is also possible, \ Lap 1A} fone | (@) A direct connection, (b) A cross-connection, Fig. 11.2 Symmetry of Form I For some graphs the axis of symmetry may pass through an even number of nodes, as illustrated in Figure 11.3(@). Then, one may still consider the graph as Form Ti by altering the axis of symmetry, as shown in Figure L1,3(b), while maintaining the topological symmetry ofthe model. (a) The axis of symmetry. (b) Altering the axis of symmetry. Fig. 11.3 Symmetry of Form If with axis passing through nodes, CHAPTERII Graph Symmetry for Eigensolution of 355 ‘Symmetry of Form IIL In this ease, the axis of symmetry passes through nodes, while the conditions of direct or eross-connections are not fulfilled, Figure 11.4 ‘The nodes on the axis of symmetry are ealled central nodes. Fig. 11.4 A symmetry of Form I A combination of Form II and Form HII connections may also be identified in a model 11.3.2 FACTORISATION OF SYMMETRIC GRAPHS, Once the three types of symmetry are identified, the isomorphie subgraphs $; and 'S; are modified such that the union of eigenvalues of the Laplacian matrices of the ‘80 modified subgraphs becomes the same as the eigenvalues of tho entte graph S. ‘The process of the modifications made to the subgraphs is called healing of the subgraphs, and the entire process may be considered as the factorisation of graph. The subgraphs obtained for a graph S after healings arc called the divisor and codivisor of 8. The following operations should be performed for different forms. Factorisation for symmetry of Form I: No healing is roquired, and S, and $; are the factors of S Factorisation for symmetry of Form Ik: For dreot connection, ink members ate removed and loops are added tothe linked nodes ofS, resulting in the subgraph C known as the divisor of S, The subgraph S; with no alteration, forms the codivisor DofS. For cross-connection, after removal of the link members, new members are addled between I and J in S; in order to obtain the divisor C. For S2, k loops are added t0 cach linked node, where kis the number of links eonnected to that particular linked ‘node, The members between a typical pairs of nodes labelled as J+N/2 and }#N/2 tare then removed, in order to obtain the codivisor D.356 ‘Structural Mechanics: Graph and Matrix Methods Factorisation for symmetry of Form IM: The nodes on the axis of symmetry are changed to neutral nodes in S; to obtain the divisor D. A neutral node is a node Which is drawn in the graph; however, it does not take part in the formation of the Laplacian matrix. In order to obtain the codivisor B, the nodes on the axis of symmetry are added to S; and each central node is connected to the corresponding, linked nodes in $; by directed members. The above rules provide simple means for factorising graphs with symmetry. Therefore, simple calculation of eigenvalues of the Laplacian matrices of such graphs becomes feasible, Example: Consider the symmetric graph S shown in Figure 11.5: es io ‘\ D AN OR, Fig. 11.5 A symmetric graph $ and its factorisation, ‘The nodes A, B, and C in the first subgraph, have the corresponding nodes A’, B’ ‘and C’ in the second subgraph. If A, B, C, A’, B’, and C’ are numbered as 1 to 6, then the Laplacian of Sin Figure 11.5 can be weitten as: uu G | In ‘The —I entries in LU correspond to the link members A.A’ and CC” oo -1l-a -1 3 The condensed matrices (C] and [D] in this form are obtained as: (C]=[6]+ (11 and (D]~(]— fn, uy CHAPTERLI Graph Symmetry for Bigensolution of 357 Matix (C] is the same as (G] with 1 added o its linked nodes A and C, and (D] is the same as [G] with -(-1) added to its linked nodes A’ and C’. Therefore, (C} and [D] ean be viewed as the Laplacian matrices of two subgraphs C and D as shown in Figute 11.5, healed with loops being added to D at A’ and C’, Thus factorisaton ofS is obtained where healings are made to include the effect of link members AA’ and CC” The matrices (C] and [D] are calculated as: 2-1) [00 20 {cl=|-1 3 -il+]o -1 0 [-1 -1 3] [oo - and 2-1-1) fo 0 0 wi=|-1 3 -1]-[o -1 0 =|-1 eit se|e:0. 0) eteet a ‘Therefore, in place of finding the eigenvalues ofthe Laplacian of , those of C and Dean be calculated, and {ALAS} = (AC(S)} T (AD(S)} (ai-t8) (AC(S)} = (0.3.8), PAKS) = (145), and (ALS) = {0,1,3,3,4,5}. Example: Consider a graph $ with symmetry of Form If and eross-connections, as shown in Figure 11.6, For this graph the Laplacian matrix is: In this example, the link members 1-3 and 2-4 have direct connections and 1-4 and 2-3 have eross-connections. Therefore, one loop in node 1 and one in node 2 are produced with respect to direct symmetry, and additional loops are due to the cross symmetry. The deletion of member 1-2 and the addition of an extra member ‘between 3 and 4 are also the healing required because of the cross-connection, The factorisation is shown in Figure 11.6 and the corresponding (C] and (D] matrices are as follows:358, Structural Mechanies: Graph and Matrix Methods CHAPTERII Graph Symmetry for Eigensolution of 359 “yp & Fig. 11.6 A graph $ and its factors C and D. Example: A graph S is considered as shown in Figure 11,7. The Laplacian of S is 4 16X16 matrix, which is put in Form IL with suitable ordering. The subgraphs corresponding to the condensed submatrices [C] and [D] are obtained as shown in Figure 11.7. Here the members in C with both ends as linked nodes are changed into multiple members, and the linked nodes in D have received the appropriate number of loops. Further decompositions result in the subgraphs illustrated in | Figure 11.7. The eigenvalues are then clculsted forthe nibgraphs as provided in| Table nena’ ashen seule forthe bap as povided | ‘The cgenvaies forthe Laplacin [Lo $ are obtained as Fig. 11.7 A graph S with symmetry and its decomposition {AL{S)}= (0, 7,4, 5, 1.4364, 9.8053, 6.3596, 1.4364, 9.8053, 4.3987, 6.3596, Table 11.1 Subgraphs of $ and their eigenvalues. | 2.1518, 5.6727, 7, 9.1755}. Subgraphs eigenvalues Using the symmetry, the Laplacian matrix of $ with dimension 16x16 having 256 a ors nies reduced to fou matics of dimecsion 4 having counted togetier 6 DC 14364, 9.8053, 6,3596 © T4364, 9.8053, 4.3987, 6.3596 DD 2ISTR, 5.6727, 79.1735 |360 Structural Mechanics: Graph and Matrix Methods 11.3.3 FORM III AS AN AUGMENTED FORM IL Consider the Laplacian for the Form II and augment it by a row and a column as: 2 « }] ule = 11-19) (i= oa ul ¢ abe : Here, we have no column with equal entries, and the only augmented row is the transpose of the augmented column. Similar to the general case, many augmented rows and columns may be included, Consider the graph shown in Figure 11. 8: i f 2 4 | Fig. 11.8 A graph S, ‘The Laplacian is formed as, where [6] = [2] and [LI ‘The condensed matrices {D] and [E] and their eigenvalues are obtained as: CHAPTERII Graph Symmetry for Eigensolution of... 361 [] =(2]-[-1]= [3] and {AD} = {3}, and 2H) -1 0) 1 -1 0 HED 3 =1] and (28) = (0,14 ovo 1} lo 17 Hence: (AL(S)} = (04143) In Form I, the matix (L) contains a submatrix (H} coresponding 0 the symmettic core of the graph. A symmetric core of a graph is a subgraph for which the corresponding Laplacian matrix is of Form II. As an example, for the graph shown in Figure 11. 9, the edge KL is the symmetric core, x Fig. 11.9 A graph with a symmetric core, ‘Node i is linked to nodes K and L in a symmetric manner. K and L. are called in ‘core nodes and iis known asthe out-of core node. In order to construct Form IIL the in-core nodes and the out-of-core nodes should be ordered. In-core nodes are numbered in a suitable manner for Form I followed bby an arbitrary ordering of the out-of-core nodes. For constructing the graph model of the condensed matrix E, it should be noticed thatthe adition ofthe last augmenting row and column results ina non-symmetric tmatrix. Therefore we should define a directed subgraph. Ina directed graph, the embers are directed, and the degree of a node isthe number of arrows leaving that ode, A member with two opposing arrows can be treated as a member with no direction. Example: Consider the symmetric graph of Figure 11.10 with an odd number of nodes. This graph has 7 nodes and contains a symmetric core of Form I The362 ‘Structural Mechanics: Graph and Mattix Methods additional node A is connected to symmetric nodes 3 and ¥, It ean be seen that the symmetry is preserved. neutral node Fig. 11.10 A symmetsic graph with 7 nodes and its factors D and B, The Laplacian of the graph shown in Figure 11.10 has Form Tl, The nodal rhumbering of the core is the same as for the Form I, followed by node A ‘numbered as 7. The corresponding Laplacian has the following patter: ‘The condensed matrices are: ‘The matrix L without partitioning leads to {2}, which is (AL} {3,3.7321,3,0.2679,4.4142,1,58580). Here, L. has 49 enti CHAPTERI Graph Symmetry for Eigensolution of. 363 entries for two condensed matrices D and B is 3%3+4x4=25. For cores with higher ‘numbers of nodes, partitioning leads to higher saving in numerical calculations. {In this example, the diagonal entry of Dn the third row is 3, while the sum of non- diagonal entries in that row is -2. Here, unlike the condensed matrix of Form Hl, Where the sum of absolute values of non-diagonal entries is bigger than the diagonal entry by an even number, the sum in bigger by an odd number. The reason is that one non-diagonal non-zero entry isin the augmented column which ‘does not takes part inthe subtraction for calculating D. Therefore a neutral node is ‘defined as a node which is drawn inthe graph; however, it does not take part in the formation of the Laplacian mates. Example: For the graph shown in Figure 11.11, the subgraphs of E and D are illustrated inthis Figure. Fig. 11.11 A graph S and its factors D and F.364 ‘Structural Mechanics: Graph and Matrix Methods For the subgraph D, the Laplacian and its eigenvalues are: 3-41-10 0000 “14 0-1 00 “10 3-1 00 pepo “hts o ol, oo 10s 4-1 o ooo ras O00) 0 oe O01 0 0 4 aD) For the subgraph E, the Laplacian and its eigenvalues are: ,4.1815,3.4613,2.8305,1.5179,5.4664,0,6958,6.8423) Po1000 09 ° 000 3 o 00 -1 000 -1 o of ° 0 0 0 o oo 0 o 000 2 {AB} = (0,5.5254,1.4470,4.4778,2.3028,3.1780,3.5584,2,0.5105}, ‘The set of eigenvalues of L is the union of the eigenvalues of Band D. ‘The main graph has 1717=289 entries in L, while the sum of entries of E and D is 9x9+8x8=145, which is nearly half ofthat of L, 11.3.4 MIXED MODELS ‘A graph may have different symmetries inthe process of sequential decomposition, In the following, examples are considered containing both Form II and Form TIT symmetries, Example 1; For this example, operations for decomposing and healing are shown in Figure 11.12, where S is first decomposed to C and D, both being Form Ill, ‘Then C is decomposed to CD and CE. Similarly, D is decomposed to DD and DE, (CHAPTERII Graph Symmetty for Eigensolution of 365 Fig. 11.12. Decomposition and healing of S. Example 2: For this model, operations for decomposing and healing are shown in Figure 11.13, where Sis first decomposed to C and D having Form II and Form Ill, respectively. Then C is decomposed to CC and CD, both being Form Il, Similarly, D is decomposed to DD and DE, which have Form If and Form Ill, respectively CC is decomposed to CCC and CCD, and CD is decomposed to CDC and CDD. ue are coc cop cece COD og ~S WV a 3 . 3 Fig. 11.13 Decomposition and healing of S. Remark: A graph may contain many symmetries, and an optimal sequence of using these symmetries for decomposition can have a decisive effect on the size and quality ofthe factors ofthe graph,366 Structural Mechanies: Graph and Matrix Methods 114 EIGENVECTORS OF THE FORM II MATRICES: In this section, the properties of the eigenvectors of the Form IT matrices are studied. For lower-dimension matrices, such as 2x2 and 4x4 rmutices, parametric study is performed, and for high-dimension matrices numerical examples are included, For a general NxN matrix M, the eigenvalues 2, ‘can be found by solving the characteristic equation: + Aa denoted by {2M}, det(M-AN) = 0. (1-20) Once 2 (1 = 1, ... ,N) is obtained, the associated eigenvector v; can be formed using: (MAD y a2n) tots, different methods may be employed, Here ¥ is notmalized by dividing the entries of v, by its largest entry, In this section, an NoeN symmetric matrix M is considered, with all entries being realy Consider a 2x2 matrix M: ab a (at-22) ‘The characteristic equation of this matrix is, 12 —2ah-+(a? —b2), (1-23) and the eigenvalues are: (AM) = (atba-b}, (ut2ay Substituting in Eq, (11-22), the associated eigenvectors are formed as follows: nonents nef sat aimn-nas vee!) (1-28) CHAPTERII Graph Symmetry for Figensolution of 367 Now consider @ 44 matrix of the Form Il Toe oe (11-26) Ly Sym. Ly In order to havea mtx in Form I, the follwing should hol: cmd and fa, ie. M should be: (11-27) ‘The characteristic equation of this mati, [Q(btetLtLabdtetb +L) +L eLrtbLfatd)' [Q2+(bre-Li-L)htelb-L) HLL bL-(a-d) (11-28) and the associated eigenvalues are y= HLL sterb) + fle bye b=2L, -2Ly)+AaHOE HL La)? ] da= EU(LyHLptetb) —y(e—bXe—b—2L) +2L9)+4(a +d)? +(Ly = Ly)? ] dys H(LetLane-b) + Yle=bye=b+2L;—2L;) 44-0? +L | 2am EU(L tL e-b)~ fle bye=b 421 ~2)+4@=0? +(L)—La)? ] (al-29)368 Structural Mechanies: Graph and Matrix Methods Substituting in Eq, (11-21), the associated eigenvectors are formed as follows: La +b-0)~ fle bye b=2L, 421g) HA He)? +L, 41g) 1204S) H [Ly —Ly +b-2)— Yle—bye—b- 2b, +21) +44? H(z Ved) _ |e ue 1 1 = oe fori ta aaa ea “(Ly -L2 +6. : “presse fear, Hey aE API - tb [Ly +h: +b-0)+ Yle ye b+ 21-21) Aes ,-L Ie) =1__ — (Ly +L, +b—e)- Yle=bye=br al, AL) raed Hy Ly 70+) Me 1 [CL +g +b-e)—Yle~byle~b + 2L; = 202) + 4(a—A)? +(Ly-L a)" 2+) (1130) A similar approach can be used for 6x6, 8. trices The above relationships eveal he following resus: 1. The absolute values ofthe entries are repeated in each eigenvector. 2. The eigenvectors ofthe Form If matties canbe divided into two groups 2) Those in which N/2 entries are repeated with the same signs; +b) Those in which N/2 entries are repeated withthe reverse signs. CHAPTERII Graph Symmetry for Eigensolution of 369 Example 1: Consider the following 44 matrix: ‘The eigenvalues and the associated eigenvectors of M are calculated as: {-0.1536 6902, at 6902] 4, =32.4513 ; yy 2g =74044 | v= ‘As mentioned earlier, the absolute values of the entries are identically repeated in vy; and vz and the entries are repeated in vs and v, with the reversed signs. As an example, in v, the absolute values of the fist and second entries are repeated as the third and fourth entries of this vector, ie. for ~0.1536=a and ~0.6902 the eigenvector v; is, y= and, for ~0.05967 = a and 0.3759 = b, the vector ¥2 has the same pattem, ‘Therefore, v; and v; belong to the fist group, For vy and vs, the first and second entries are the same as the third and fourth entries with the reversed signs, ie. for vs with ~0.2372 = a and 0.6661 = b, we have:370 Structural Mechanies: Graph and Matrix Methods aian, ‘The third and fourth eigenvectors belong to the second group. Example 2: Consider a 66 matrix as follows: 1 7 18 4 6 n 2 12 4 3 20 0 4 6 20 10 Ho 2 ee a7 18 1416 ‘The eigenvalues and the associated eigenvectors are as follows: 7.2048; y= 8976+4,L151i: 0.2132-+0.07921 02373-03014) 0.21324 0.07921 0.2250+ 0.50041 -0.2373-03014i y= 26.6364 599 = y= 3.89764. 115 15, 02132-0.07921 -0.2373+0,3014i 0.2132-0.07921 0.2250-0.5004i -0.2373-+0.30141 CHAPTER! Graph Symmetry for Eigensolution of 37 d= -0.1077; = 103441; ‘The above eigenvectors can also be considered as two groups. In the frst group, the entries are repeated with the same signs, and in the second group, the entries aro repeated with the reverse signs, Example 3: Consider the following 88 matrix in the Form Il: Te eee 7 9 wtf -1o0 1 8 lets 80) 2s 0202 1/805 7206 678 23 4 ell 0; ily 9 10a elo el} o 8 4 2 60s 7 slo 01 ‘Two eigenvalues and the associated eigenvectors corresponding to different groups are calculated as, y= 32.5249 ; [-oasez 02763 —03707 ~03671 03892 —027= -03707 -oaoni}, vers the fist 4 entries ar repeated in the sscond half ofthe vost, and aye 11.1679; wy =E0.716 06260 02104 —01856 0.1716 —06260 ~0.2101 0.1856}, where the first 4 entries are repeated with reverse signs in the next half of the vector.32 ‘Structural Mechanics: Graph and Matrix Methods 11.4.1 EIGENVECTORS OF M AND CONDENSED MATRICES C AND D In this section, the relationship between the eigenvectors of M and those of the condensed submatrices C and D is established. This simplifies the calculations for large matrices considerably. This parametric study is made for a 4x4 matrix of Eq, ar2n, ‘The condensed submatrices are obtained as: eft 214] mf co ‘Now the eigensolution of each matrix is studied separately. Eigensolution of € ‘The characteristic equation is obtained as: [O° (eet HL YA+e(O+L )+LLe-bL-(atd)"] =0.(11-33) ‘The eigenvalues and eigenvectors are calculated as: drow F(LitLtetby+ yle—by(e=b=2L, +2Lg)+4(ard)? +L) Lg) FULstLsterb) — fle bye— 6-21, +22) +4e4d)* (yl? (isa) nee fateh 0 ete aoe] 1 ae fra aT] 1 (113s) CHAPTERII Graph Symmetry for Eigensolution of .. 33 Eigensolution of D The characteristic equation is obtained as: [02+(bte-L-Lte(b-L#L La bLy{a-d)'] = 0. (11-36) ‘The eigenvalues are found as: dao E[(LytL-e-b) + ye bye—b4 21 ~2Lq) + Alaa)? +L, Lay? Dog [Ly+Ls-e-b)= Yle=bXe=b+ 2, -2Lp)+4(a—F+(Ly La)? (11-37) ‘The eigenvectors are calculated as (ai-38) nae faiatl oe E ET| A careful study of the above solutions reveals that: 1. The entries of the eigenvectors of the condensed and original matrices are proportional; 2. The entries of the eigenvectors for C are proportional to the entries of the ‘eigenvectors of the first group and those of D are proportional to the entries of the second group. 11.42 FORMATION OF THE EIGENVECTORS, ‘The eigenvectors of M can be obtained, using the following expansion process: 1. Find the eigenvectors of € and expand it by repeating the same entries, i.374 Structural Mechanies: Graph and Matrix Methods , fi yond (11-40) Exanple: For the condensed Ca typical sigenestor conesponding © 2 - 32,5249 in a normalized form is: * " “ — ve=F-1 0.7099 ~0.9525 -0.9432}', and for the original matrix we have wn{-t 07099 0.9525 ~0.9432 1 -07099 -09528 -o9432}', ie, half ofthe ents ae repeated For atypical eigenvalue of D, namely A= 11.1679, the associated eigenvec obtained as, a the associated eigenvector is 0.2741 1 0.3361 -0.2965}', ‘and by expanding and reversing the signs of the entries for the second half of the vector, atypical eigenvector forthe original matrix M is obtained as wu {-02741 1 0.3361 -0.2965 0.2741 -1 -0.3361 0.2965} CHAPTERII Graph Symmetry for Eigensolution of. 375 ‘The remaining eigenvalues can similarly be obtained. 18 ENVECTORS FOR MATRICES OF FORM It Consider M as @ matrix with a core of Form I, augmented by a row and a column, The entries of the column adjacent to the core are equal (ay Addition of one row and column in the above form preserves some eigen- properties of D and M, since: D=[AI-[B} and {XD} ¢ {AM}. (11-42) Since L is any real number, many such columns can be added to the core, with L having different values. In all these cases, the cigen-properties of the condensed matrix D of the core stay unaltered in M. It should be noted that M is now a non-symmetric matrix. However, if all the entries in the augmenting row and column adjacent to the core are selected as L, then the symmetry will be retained. tar | iB) + : 113 Mim) | tayo Coe LLP where P is any arbitrary real number. The condensed matrices D and E ean easily be obtained, Example 1; Consider the submatrix,316 ‘Structural Mechanics: Graph and Matrix Methods in Form Tl. This matrix is augmented by one row and one column such that the submatrices D and E are produced with the same eigenproperties, and the new matrix is symmetri. ‘The following matrix is considered: Where D and F are: ] and (AD}= (1), 3 | 5 4 and (ME) = (-0.3589,8.3589), (AM) ~ (1,-0.3589,8.3589) ‘The eigenvectors of M are; 1 0.8932] y=1= = 47) shy =-0.3589 v, = 0.8982, 0 -1 0.5598} 3589 = v3 = 0.5589) 1 In tis example, the eigenvector v, corresponding to hy =I is obtained from D. This is why the first and second entries are repeated with a reverse sign, The third entry is zero, since it is not the symmettic core. Therefore, it ean be concluded that the entries of this vector are independent of the augmented row and column and the ‘magnitude of their entries. The zeto entry in v; makes it independent, For the cigenvalues corresponding to the condensed matrix E, the situation is different, Since in E the submatrix [A+B] is present, N/2 of the entries (N being the dimension of the core) are repeated and the rest of the entries are independent. In the above exarple, the entry 0.5398 is repeated, since the fitst and the second rows ‘oF take part in [A+B], and the third entry is independent, ‘When k rows and columns with the above prope for the eigenvectors will hold. are added, similar properties “i CHAPTERII Graph Symmetry for Eigensolution of. 37 Remarks: The above results suggest @ simple approach for the formation of the eigenvectors forthe Form III matrices. (@) For calculating the eigenvectors of D, perform the following operations: Let the core in Form Il be an NaN matrix: ia i aay (BIT TAT cas For this submatrix, D_will be an (N/2*N/2) matrix, and the corresponding eigenvector will have N/2 entries. If k rows and columns are added for the ‘formation of Form II matrix M, then the corresponding eigenvector will have Nk entries, Now a vector of dimension N/2 is related 0 a vector of dimension Nik. For this purpose, the first N/2 entries are writen and copied undemeath with the reverse signs, and then k zero entries are added (©) For the eigenvectors obtained for B, consider the following ‘The dimension of an eigenvector of E is (N/2 + k)x(N/2 + k), and it has N/2 entries for becoming an eigenvector of M. Here N/2 enties are copied withthe same sign and added to above the eigenvector of E. Example 2: Consider a matrix N ofthe Form I as: at aes ‘Three rows and columns are added to N to form M withthe following properties: 1, Misa symmetric matrix, m(ij) = m(j); 2. M has condensed submatrices D and E. ‘The following matrix has the aforementioned properties: 2J1 378 11237 8 Mele ae 7 7018 8 8 8B Ls378 Structural Mechanies: Graph and Matrix Methods ‘The condensed matrix Dis: [Now the correspondit va =| 0 |. ‘The condensed B is formed as: 8 8 8 2 bat igenvalues and cignvectors are calculated and augmented to form the ig eigenvectors of M: 0.5230 06149 2k = 304475 vB = =M= : B= gas] > : 1 0.6241 eH 03508, SAL ay) | same] 03500 fs ~0.2237 = 0.2237 05262 05262 11407 0.8313 -1 CHAPTERL Graph Symmetry for Eigensolution of. 379 ~ 0.2827 1 gE = 0.8900 vB = =v 0.0806 -03712 Using the condensed matrices and their properties, increases the speed of calculation and results in more accurate eigensolution, For matrices with cores of high dimensions, the application of the present decomposition becomes more economical, 11.6 A GENERALIZED FORM III MATRIX "Now the question arises whether rows and columns with other properties can be added to a core of Form II pattem. When the sum of the absolute values for the ‘tries of an augmenting row is the same as that ofits column, then D and B. can be formed as before. It should be noted that such a restriction is not necessary to be imposed on the last row and column, Consider the following form, M=|[BI] [A] H] i throw, (4s) oH P. arene For the ahiom, 8 2a He Doane Y aca=haof 1 jane ivi or (1-46) Example 3: Consider a core in Form Il:380 Structural Mechanies: Graph and Matrix Methods ‘Add an augmenting row and columa to form M with the following properties: 1. Matrix M is symmetric; 2. Mihas condensed submatrices D and Es 3. Two numerical values are considered for Hin M, namely Hi; = 4 and Ho= 8. 6) 2 Wf M=[2|-6 4 aan HOH P ForH | P is taken arbitrarily as 8, i, 24 4 | {any} =(-8,-62462,10.2462} 48 [=$] M=[2 4 ‘The condensed submatrices are: D,=[-6- =[-8]= (AD =-8}. 4 4 z ele. 2) > ord=E-s2ues024} ‘These submatrices will contain the eigen-properties of part of the M. For H= 8, the numerical value of P is taken as —16, and 2-8 =e 8 |= fam}={-82.8002,~22.8062}, -8 -16 ‘The corresponding condensed submatrices are: D. 8] (Ady 4-8 Fa=l 16 Tfa|> ts) = ena} CHAPTERII Graph Symmetry for Eigensolation of. 381 ‘As mentioned before, in both M, and M; matrices some properties ofthe eore are ‘eft unaltered. These eigenvalues correspond to those of D. In this example, H= — 8, the eigenvalue corresponding to Dy has this property. [All the properties of Section 11.5 are applicable to this special pattem. As an example, the eigenvalues of M are: 1 8062>v2={ 1} -0.85| (0.4253) hy =~22,8062 = vy = {04253}. i It should be mentioned that, the above argument is still applicable when rows and ‘columns are added to the symmetric core ofthe Form I.382 ‘Structural Mechanies: Graph and Matrix Methods 11.7 VIBRATING CORES FOR A MASS-SPRING VIBRATING SYSTEM Consider a symmetric system as shown in Figure 11.14(3). This system is symmetric, and its properties can be studied using its substructures, (b) The subsystems of , 1.14 A symmetric dynamic system and two subsystems witha link spring. ‘These properties consist of the mass m, and the stiffness kj. The masses, siffnesses and their connectivity are considered to be symmetric with respect to the axis shown in Figure 11.14(a). ‘This system can be considered as two identical subsystems connected to each other with a spring, knows as a link spring, as shown in Figure 11.14(b). ‘This system has two degrees of freedom vy and v2. The natural frequencies and natural modes for the following eigenproblem, {(K] -0? (mp (44 = (0), (11-48) canbe founda {x)-0%tm]=0. (ras) ‘Where (Kis the stiffness matrix and {ml is the mass matrix of the system, The eigenvalues and eigenvectors are denoted by ¢, and 6. respectively. CHAPTERIL Graph Symmetry for Eigensolution of, 383 Since [K] and [ml are both symmetric, therefore the matrix {KC} ~ © fm] fas Form Tas the following: wean | ae er Using «@?m,=2., one can write: (ky tky)-A kp fo (si) sky ths) Since the stiffness matrix has Form Il, one ean find its eigenvalues by calculating, those of its condensed submatrices: afk, +k, ke ]=fky], D=[hy +a +k2)= Ey +k] cs) ‘The matrices © and D pastially contain the eigenvalues of S. Since these submatrices have @ nature similar to that of the overall stiffness matrix, the ‘condensed matrices C and D define the stiffness mattices of the subsystems, as shown in Figure 11.15. ‘The structures corresponding to the condensed submatrices are referred to as Vibrating cores. These vibrating cores, contain part of the properties of the vibrating system. Therefore, the eigenvalues and eigenvectors of the overall structure can be found using those of Cand D systems, Figure 11.15. =k] © D={k, +2k,] “ Fig. 11.15 Subsystems corresponding to condensed submatrices C and D.384 Structural Mechanics: Graph and Matrix Methods For system S having N degrees of freedom, m and K are NN matrices, and if the N structure is symmetric, the corresponding submatrices will be 11.8 THE GRAPH MODEL OF A MASS-SPRING SYSTEM The mathematical model of a dynamic system, consists of masses and springs. ‘These masses are connected by means of springs. As the mathematical model, a ‘weighted graph is defined as follows: 1. The supports in the mathematical model are associated neutral nodes in the rapt 2. For each mass, a node of the graph is associated, and its weight is taken as the ‘magnitude of the mass, 3. An edge is considered for each spring, and its weight is taken as the stiffness of the spring. ‘Asan example the graph mode ofa dynamic system shown in Figure 11.16) is depicted in Figure 11.16(b). mu 7 ' ky omy ka mk hot tot a 1k (2) A dynamic system. (b) The graph model S, 11.16 A dynamic system and its graph model, CHAPTERII Graph Symmetry for Figensolution of 385 For a dynamic system, we have: Ko =0%m => [K-07 my)= (a-s3) This is an eigenvalue problem for which @ is the eigenvalue and @ is its eigenvector. we consider mas a diagonal matrix, then the above equation becomes [(K-m]-o'N}o= 0. (L-s4) If [K-m]= Ley and 2=0?, then we have: [Ly -a}p=0 (155) ‘This is an eigenvalue problem corresponding, to eigenvalues and eigenvectors of Ly. This relationship can be associated with the corresponding graph. If Ly is the eneralized Laplacian of the graph, then the above problem becomes an eigenproblem of a graph, 11.8.1 VIBRATING SYSTEMS WITH FORM II SYMMETRY ‘As an example, the generalized Laplacian matrix for the graph S; in Figure 11,16(b) has the following form: ky+ka-m, ky = 11-56 lr [ =k) ky +k;—my CD For a symmetric graph, an appropriate numbering of the nodes result in a ‘generalized Laplacian matrix with Form I Example: Consider a dynamic system as shown in Figure 11.17, with graph model being 8:386 Structural Mechanies: Graph and Matrix Methods ge Fig. 11.17 A dynamic system and its graph model, ‘This graph is symmetric, and the Laplacian and generalized maces ae as kytkg tks kp ky 0 why kptky 0 ky ky 0 yt thy ky as) o ky ky ka tks kytka thy—m ky mk ° ok; Kytky-m; o mks (1-58) ky 0 ky tkatky—m ky ° ky mk Ky-thy my For symmetry in the Form Il, the generalized Laplacian matrix ean be written as (11-59) CHAPTERII Graph Symmetry for Figensolution of 387 ‘The submatrix G is called the shape mani and represents the properties of both subgraphs, which are identical, and LI is called the fink marrix and shows the way ‘wo subgraphs are connected to each other. The submatrix LI represents the effect of the springs between two subgraphs inthe stiffiaess matrix, As mentioned before, we have an eigenproblem for the matrix Ly. According to the properties of the Form I, if [G + Li] = C and (G ~ Lt} = D, then Lip ean be ec ce [e ial If Ly isthe generalized Laplacian matrix ofa graph, chen tis graph will consist of two subgraphs with N/2 nodes for cach subgraph, which are not connected to each ‘other, and Lr has eigenvalues as: EIG(Ly) = FIG(C)T EIG(D), (61) ‘Therefore, the subgraph C and D ate the dynamic cores of the model. Each core defines part of the natural frequencies @, of the whole system: {on7}= {0p} U {op} (11-62) ‘Asan example, for graph S, the cotes C and D are shown in Figure 11.18: Ad 18 The dynamic cores C and D ofS.388 ‘Structural Mechanics: Graph and Matrix Methods The Laplacian matrices of C and D are as follows: and ky thy #2 hp : 64) oa oe, ns AAs mentioned previously, the Laplackan mates ofthe corresponding graphs are the same a the stfnes matrices ofthe mathematica model foreach subgraph C and D, as shown in Figure 11.19 4 ke ky taf CE CED B Be Fig. 11.19 The subgraphs for C and D. 11,82 VIBRATING SYSTEMS WITH FORM Ill SYMMETRY For a symmetric system with odd number of masses, the corresponding graph will hhave Form III symmetry. For such a system, the vibrating cores can be identified using symmetry. As the third example, consider the model shown in Figure 11.20(a): CHAPTERLI Graph Symmetry for Eigensolution of 389 \ eee ee ee san A ani Co A vay ae Danis © m, a, m SR $m 6 be ® Fig. 11.20 A dynamic system and its graph model. ‘The corresponding graph is shown in Figure 11.20(b). The Laplacian and generalized Laplacian matrices are: tk; 0 ky L=| 0 kk -ka|, (11-65) “ky -ky “ty and Kytky-m 0 ky w=] 0 kytky=my kz |. (11-66) -k -k, %y-m, ‘As can be seen, L and L both have Form Il ‘The Laplacian matrices corresponding to the vibrating cores are given below: L(D)= Uk +k: ~0}=[k, +k2), «isn and kytky Ky 7 1.68) Le ee oe (11-68) The graphs of these matrices are shown in Figure 11.21:390 ‘Structural Mechanies: Graph and Matrix Methods Fig. 11.21 The subgraphs for D and E, If there is @ directed edge between two nodes i and j dieoted from i to j, it represents directed spring in the dynamic system, Figure 11.22. The main characteristic of such a spring is that the connection of this spring to masses is such that it does not take part in the stiffness of kj, but it affects ky, i. k= x A hunk iW i ky Fig. 11.22. A directed spring, ‘The mathematical models corresponding to the cores D and E are shown in Figure 11.23: Fig. 11.23 Models corresponding to D and E. CHAPTERI! Graph Symmetry for Eigensolution of .. 391 According to the properties ofthe cores, {Ay} = (AD) 0 (AE) (al-69) and {oly} =(@D} T(E}, (11-70) ‘nd, from the vibrating cores E and D, the natural modes of the whole system can bee found. If each vibrating system itself contains symmetry, then the cores can be decomposed accordingly. Further decomposition of the refined cores for symmetry is also possible 11.8.3 GENERALIZED FORM III AND VIBRATING SYSTEM ‘As described in Section 2, for a graph with symmetric core having Form I, ifthe ‘complement of the core is connected by the nodes of degree 1, then the nodes can be ordered to produce a Laplacian matrix of Form II, This property ean also be used for graphs corresponding to the vibrating systems. Consider the system in Figure 11.24(a), together with its graph illustrated in Figure 11.240), & i Wy — BN i WG Fig. 11.24 A dynamic system and its graph model,392 Structural Mechanics: Graph and Matrix Methods ‘The subgraph containing the nodes A, B and C is a symmetric core of Form II ‘The nodes E and D are connected to this core through C. Therefore, the Laplacian matrix of this graph will be generalized Form IIL. L and Lx are formed as follows: ree ob 0 * keh -— 0 Oo ky -ky kth ky 0 ay 0 0 ky gtky —ky a) 0 = ky kat, kytky=m = ak o o wk kythy~ ak; ° o ak wk Ak+ky—mz ky ° ° ° “ktm ky ° 0 0 keyg tks my anny ‘The condensed submatrices D and E are formed for Ly as: = [k+hy— m ~ Ck] = [2krHg- my}, (4-73) Ky thy —m—ky— —ky—ky ° 0 cke-ky 2ketky-m ky 0 o why ytlgrmy ky | O79 0 ° mk katks—my or ki-m -ky ° 0 =2ky ky tky-mz ky 0 E 7 0 mks kytky-my ky 2 0 ° ky ketks—my ‘The subgraphs associated with the cores D and E, are shown in Figure 11.25: CHAPTERII Graph Symmetry for Eigensolution of . 393, Bo me ks_ mb om, ey s B ky om, D: fies . ii Fig. 11.25. Subgraph associated with the cores D and E. ‘The form of the vibrating cores corresponding to D and E is shown in Figure 11.26: b VS cap k Fig. 11.26 Submodels D and E, Itcan be observed that, due to symmetry, the generalized Laplacian is decomposed into two submatrices of 1x1 and 4><4, and the cores are formed, LEN other nodes are connected to C in a similar manner, again the graph can be ‘decomposed into two cores D and E, as shown in Figure 11.27. It should be noted thatthe core D does not change. Graph with nodes A Fig. 11.27 A graph decomposable into D and E.394 Structural Mechanics: Graph and Matrix Methods Thus, the natural frequency of the core D and the corresponding mode of the system will be unaltered. Therefore, one can conclude that part of the natural frequency of the symmetric system with Form TI will exactly be reflected in the whole system. ‘Consider the system shown in Figure 11. 28: CHAPTERLI Graph Symmetry for Eigensolution of 395 Dah) Co) bs ky ake | ety, E Fig. 11.28 A dynamic system and its graph model. ‘The Ly,L{D) and L(B) are as follows: fy etm ty °

You might also like

• The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life

  

  From Everand

  The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life

  Mark Manson

  Rating: 4 out of 5 stars

  4/5 (5794)
• The Little Book of Hygge: Danish Secrets to Happy Living

  

  From Everand

  The Little Book of Hygge: Danish Secrets to Happy Living

  Meik Wiking

  Rating: 3.5 out of 5 stars

  3.5/5 (400)
• Tutorial Resortes de Suelo para Diseno en Acero

  

  Document39 pages

  Tutorial Resortes de Suelo para Diseno en Acero

  Juan Carlos González

  No ratings yet
• • Magazines
  • Podcasts
  • Sheet music
• Time-Frequency Analysis of Wind Effects On Structures: A. Kareem, T. Kijewski

  

  Document18 pages

  Time-Frequency Analysis of Wind Effects On Structures: A. Kareem, T. Kijewski

  Lisa Poole

  No ratings yet
• Response of Hollow-Core Slab Floors To Concentrated Loads

  

  Document16 pages

  Response of Hollow-Core Slab Floors To Concentrated Loads

  Juan Carlos González

  No ratings yet
• Topic15 4 GeotechnicalEarthquakeEngineeringHandouts PDF

  

  Document23 pages

  Topic15 4 GeotechnicalEarthquakeEngineeringHandouts PDF

  József Tóth-Horgosi

  No ratings yet
• 3D Nonlinear Transient Analysis of Concrete Dams - Paper

  

  Document15 pages

  3D Nonlinear Transient Analysis of Concrete Dams - Paper

  Julio Humberto Díaz Rondán

  No ratings yet
• Stability of Planar Structural Systems-Fred W. Beaufait

  

  Document45 pages

  Stability of Planar Structural Systems-Fred W. Beaufait

  Juan Carlos González

  No ratings yet
• Practical Three Dimensional Nonlinear Static Pushover Analysis

  

  Document4 pages

  Practical Three Dimensional Nonlinear Static Pushover Analysis

  Arvin Edsel Sioson

  No ratings yet
• Seismic Design of Liquid Storage Tanks

  

  Document112 pages

  Seismic Design of Liquid Storage Tanks

  Munna Bhai

  100% (1)
• Vickery PJ. (2009) Hurricane Hazard Modeling - The Past, Present and Futere

  

  Document14 pages

  Vickery PJ. (2009) Hurricane Hazard Modeling - The Past, Present and Futere

  Juan Carlos González

  No ratings yet
• Precast Prestressed Concrete Horizontally Curved Bridge Beams

  

  Document42 pages

  Precast Prestressed Concrete Horizontally Curved Bridge Beams

  Edward van Martino

  100% (1)
• Machine Foundations

  

  Document249 pages

  Machine Foundations

  nizamaniza

  100% (32)
• POULOS DAVIS Pile Foundation Analysis and Design PDF

  

  Document410 pages

  POULOS DAVIS Pile Foundation Analysis and Design PDF

  Sheril Chandrabose

  100% (7)
• Is 4623 2000

  

  Document26 pages

  Is 4623 2000

  Juan Carlos González

  No ratings yet
• 29 30 31 SpillwayGatesPP20121003

  

  Document76 pages

  29 30 31 SpillwayGatesPP20121003

  Juan Carlos González

  No ratings yet
• Boletín de American Meteorological Society (Tornados)

  

  Document68 pages

  Boletín de American Meteorological Society (Tornados)

  Juan Carlos González

  No ratings yet
• Wind Analysis On Chimneys

  

  Document4 pages

  Wind Analysis On Chimneys

  anilpradeep

  No ratings yet
• Effects of Plastic Hinge Properties in Nonlinear Analysis of Reinforced Concrete Buildings

  

  Document9 pages

  Effects of Plastic Hinge Properties in Nonlinear Analysis of Reinforced Concrete Buildings

  Juan Carlos González

  No ratings yet
• MCEER (1998) - Modeling of Pile Footings and Drilled Shafts For Seismic Design

  

  Document162 pages

  MCEER (1998) - Modeling of Pile Footings and Drilled Shafts For Seismic Design

  Juan Carlos González

  No ratings yet
• 0951.steel Framed Structures. Stability and Strength by R. Narayanan

  

  Document341 pages

  0951.steel Framed Structures. Stability and Strength by R. Narayanan

  xxmorph3u5xx

  No ratings yet
• Theory of Elasticity - TIMOSHENKO PDF

  

  Document263 pages

  Theory of Elasticity - TIMOSHENKO PDF

  Shaik Bademiya

  No ratings yet
• The Structures and Dynamic of Tornado-Like Vortice

  

  Document29 pages

  The Structures and Dynamic of Tornado-Like Vortice

  Juan Carlos González

  No ratings yet
• We People Here Nahuatl Accounts of The Conquest of Mexico

  

  Document40 pages

  We People Here Nahuatl Accounts of The Conquest of Mexico

  Juan Carlos González

  100% (1)
• Reinforced Concrete Deep Beams - Prof. F.K

  

  Document299 pages

  Reinforced Concrete Deep Beams - Prof. F.K

  ward_civil036694

  100% (5)
• Shell Ino

  

  Document13 pages

  Shell Ino

  ericdravenll6427

  No ratings yet
• Hydrodynamic Coefficients For Calculation of

  

  Document14 pages

  Hydrodynamic Coefficients For Calculation of

  Anoop George

  No ratings yet
• Seismic Design of Eccentrically Braced Frames

  

  Document32 pages

  Seismic Design of Eccentrically Braced Frames

  Mauricio Riquelme

  No ratings yet
• Lift On An Ascollating Cylinder in Smooth and Turbulent Flow

  

  Document12 pages

  Lift On An Ascollating Cylinder in Smooth and Turbulent Flow

  Juan Carlos González

  No ratings yet
• Practical Modelling of High-Rise Dual Systems With Reinforced Concrete Slab-Column Frames

  

  Document22 pages

  Practical Modelling of High-Rise Dual Systems With Reinforced Concrete Slab-Column Frames

  thanzawtun1981

  No ratings yet
• Crítica A Ala Escala de Fujita

  

  Document5 pages

  Crítica A Ala Escala de Fujita

  Juan Carlos González

  No ratings yet
• Morison Equation

  

  Document8 pages

  Morison Equation

  mailnewaz9677

  0% (1)
• Shoe Dog: A Memoir by the Creator of Nike

  

  From Everand

  Shoe Dog: A Memoir by the Creator of Nike

  Phil Knight

  Rating: 4.5 out of 5 stars

  4.5/5 (537)
• Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race

  

  From Everand

  Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race

  Margot Lee Shetterly

  Rating: 4 out of 5 stars

  4/5 (895)
• The Yellow House: A Memoir (2019 National Book Award Winner)

  

  From Everand

  The Yellow House: A Memoir (2019 National Book Award Winner)

  Sarah M. Broom

  Rating: 4 out of 5 stars

  4/5 (98)
• The Emperor of All Maladies: A Biography of Cancer

  

  From Everand

  The Emperor of All Maladies: A Biography of Cancer

  Siddhartha Mukherjee

  Rating: 4.5 out of 5 stars

  4.5/5 (271)
• A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story

  

  From Everand

  A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story

  Dave Eggers

  Rating: 3.5 out of 5 stars

  3.5/5 (231)
• Never Split the Difference: Negotiating As If Your Life Depended On It

  

  From Everand

  Never Split the Difference: Negotiating As If Your Life Depended On It

  Chris Voss

  Rating: 4.5 out of 5 stars

  4.5/5 (838)
• Grit: The Power of Passion and Perseverance

  

  From Everand

  Grit: The Power of Passion and Perseverance

  Angela Duckworth

  Rating: 4 out of 5 stars

  4/5 (588)
• On Fire: The (Burning) Case for a Green New Deal

  

  From Everand

  On Fire: The (Burning) Case for a Green New Deal

  Naomi Klein

  Rating: 4 out of 5 stars

  4/5 (74)
• Yes Please

  

  From Everand

  Yes Please

  Amy Poehler

  Rating: 4 out of 5 stars

  4/5 (1891)
• Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future

  

  From Everand

  Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future

  Ashlee Vance

  Rating: 4.5 out of 5 stars

  4.5/5 (474)
• Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America

  

  From Everand

  Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America

  Gilbert King

  Rating: 4.5 out of 5 stars

  4.5/5 (266)
• The Unwinding: An Inner History of the New America

  

  From Everand

  The Unwinding: An Inner History of the New America

  George Packer

  Rating: 4 out of 5 stars

  4/5 (45)
• Fear: Trump in the White House

  

  From Everand

  Fear: Trump in the White House

  Bob Woodward

  Rating: 3.5 out of 5 stars

  3.5/5 (738)
• Team of Rivals: The Political Genius of Abraham Lincoln

  

  From Everand

  Team of Rivals: The Political Genius of Abraham Lincoln

  Doris Kearns Goodwin

  Rating: 4.5 out of 5 stars

  4.5/5 (234)
• Principles: Life and Work

  

  From Everand

  Principles: Life and Work

  Ray Dalio

  Rating: 4 out of 5 stars

  4/5 (599)
• Angela's Ashes: A Memoir

  

  From Everand

  Angela's Ashes: A Memoir

  Frank McCourt

  Rating: 4.5 out of 5 stars

  4.5/5 (440)
• The World Is Flat 3.0: A Brief History of the Twenty-first Century

  

  From Everand

  The World Is Flat 3.0: A Brief History of the Twenty-first Century

  Thomas L. Friedman

  Rating: 3.5 out of 5 stars

  3.5/5 (2259)
• The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are

  

  From Everand

  The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are

  Brene Brown

  Rating: 4 out of 5 stars

  4/5 (1090)
• Rise of ISIS: A Threat We Can't Ignore

  

  From Everand

  Rise of ISIS: A Threat We Can't Ignore

  Jay Sekulow

  Rating: 3.5 out of 5 stars

  3.5/5 (137)
• The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers

  

  From Everand

  The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers

  Ben Horowitz

  Rating: 4.5 out of 5 stars

  4.5/5 (344)
• Bad Feminist: Essays

  

  From Everand

  Bad Feminist: Essays

  Roxane Gay

  Rating: 4 out of 5 stars

  4/5 (1015)
• Steve Jobs

  

  From Everand

  Steve Jobs

  Walter Isaacson

  Rating: 4.5 out of 5 stars

  4.5/5 (806)
• John Adams

  

  From Everand

  John Adams

  David McCullough

  Rating: 4.5 out of 5 stars

  4.5/5 (2409)
• The Glass Castle: A Memoir

  

  From Everand

  The Glass Castle: A Memoir

  Jeannette Walls

  Rating: 4.5 out of 5 stars

  4.5/5 (1712)
• The Outsider: A Novel

  

  From Everand

  The Outsider: A Novel

  Stephen King

  Rating: 4 out of 5 stars

  4/5 (1839)
• The Light Between Oceans: A Novel

  

  From Everand

  The Light Between Oceans: A Novel

  M.L. Stedman

  Rating: 4.5 out of 5 stars

  4.5/5 (789)
• The Sympathizer: A Novel (Pulitzer Prize for Fiction)

  

  From Everand

  The Sympathizer: A Novel (Pulitzer Prize for Fiction)

  Viet Thanh Nguyen

  Rating: 4.5 out of 5 stars

  4.5/5 (121)
• Brooklyn: A Novel

  

  From Everand

  Brooklyn: A Novel

  Colm Toibin

  Rating: 3.5 out of 5 stars

  3.5/5 (1937)
• The Woman in Cabin 10

  

  From Everand

  The Woman in Cabin 10

  Ruth Ware

  Rating: 3.5 out of 5 stars

  3.5/5 (2322)
• A Man Called Ove: A Novel

  

  From Everand

  A Man Called Ove: A Novel

  Fredrik Backman

  Rating: 4.5 out of 5 stars

  4.5/5 (4609)
• Little Women

  

  From Everand

  Little Women

  Louisa May Alcott

  Rating: 4 out of 5 stars

  4/5 (104)
• Manhattan Beach: A Novel

  

  From Everand

  Manhattan Beach: A Novel

  Jennifer Egan

  Rating: 3.5 out of 5 stars

  3.5/5 (792)
• Wolf Hall: A Novel

  

  From Everand

  Wolf Hall: A Novel

  Hilary Mantel

  Rating: 4 out of 5 stars

  4/5 (3811)
• The Perks of Being a Wallflower

  

  From Everand

  The Perks of Being a Wallflower

  Stephen Chbosky

  Rating: 4.5 out of 5 stars

  4.5/5 (2103)
• The Art of Racing in the Rain: A Novel

  

  From Everand

  The Art of Racing in the Rain: A Novel

  Garth Stein

  Rating: 4 out of 5 stars

  4/5 (4200)
• A Tree Grows in Brooklyn

  

  From Everand

  A Tree Grows in Brooklyn

  Betty Smith

  Rating: 4.5 out of 5 stars

  4.5/5 (1929)
• Sing, Unburied, Sing: A Novel

  

  From Everand

  Sing, Unburied, Sing: A Novel

  Jesmyn Ward

  Rating: 4 out of 5 stars

  4/5 (1103)
• Her Body and Other Parties: Stories

  

  From Everand

  Her Body and Other Parties: Stories

  Carmen Maria Machado

  Rating: 4 out of 5 stars

  4/5 (821)
• The Constant Gardener: A Novel

  

  From Everand

  The Constant Gardener: A Novel

  John le Carré

  Rating: 3.5 out of 5 stars

  3.5/5 (104)