Matrix Structural Analysis 2nd Edition PDF

&**',? :&4 ?
53"!8835?3"? (<(+?0$(0!!5(0$? .!5(9;8?
36!++? 0(<!58(9>?
&%4? ? **#% 4?
9!?53"!8835?0?5!8(!09?
+5)830? 0(<!58(9>?
1/*?
? & -&/?
883(9!?53"!8835?3"? (<(+?0$(0!!5(0$?
;)0!++? 0(<!58(9>?
2%/? &* =? ? 1/7? /?

Copyright**2014*by*Ronald*D.*Ziemian*
*
All#rights#reserved.##This#book#or#any#part#
thereof#must#not#be#reproduced#in#any#form#
or#redistributed#without#the#wri;en#
permission#of#Ronald#D.#Ziemian.#
To
Barbara, Terry, and Constance
Preface to the Second Edition
The first edition of this book was written 20 years ago, but our aims remain the same:
to place proper emphasis on the methods of matrix structural analysis used in current
practice and to provide the groundwork for forthcoming practice and allied, more
advanced subject matter. The present edition accounts for changes in practice that
have taken place in the intervening years, and it incorporates advances in the art of
analysis we regard as suitable for application now and of increasing importar.ce to
practitioners in the years just ahead.
Among the major changes from the first edition is the addition of chapters on the
nonlinear analysis of framed structures, treatment of the solution of nonlinear prob
lems, and packaging of a compact d(sk containing the computer program, MASTAN2.
The major reductions include the deletion of Chapter 7 on the flexibility method,
Section 10.7 on the transfer matrix method, and Chapter 12 on the finite element
method. The entire text and illustrative examples have also been edited extensively,
and several chapters have been repositioned. These changes deserve some explanation.
The reduction in coverage of the flexibility method was presaged by a comment in
the preface to the first edition: "It is hardly used in practice." Although the method is
advantageous in certain circumstances and we believed it might receive greater atten
tion if cast in a more efficient form, such has not been the case. Therefore, the de
velopment of flexibility method equations of global analysis has been deleted. But
extensive coverage of the flexibility approach to the determination of element force
displacement relationships has been retained because of the value that it has in many
instances. Elimination of the global flexibility method formulation also led to recon
sideration of the chapter on the equations of statics and kinematics (Chapter 6 of the
first edition) which was in large measure preparation for the identification of redundant
forces in the flexibility method. That material has been deleted also. The other dele
tions-material on the transfer matrix method and the general finite element concept
were made in the interest in retaining a reasonably wieldy volume.
The additions represent some departures from the spirit of the first edition. In that
edition's preface it was noted that, in concentrating on the direct stiffness method, we
were dealing with an accredited, widely practiced procedure, whereas an earlier gen
eration of texts frequently dealt with subject matter that was not fully crystallized.
Nonlinear structural analysis has a long history and it is the subject of a number of
books and some highly sophisticated computer programs, but it can hardly be called
"crystallized." Many of its features are ad hoc-in the good sense of the term-and
are in need of reduction to more routine procedures through further research and
generalization. Others may always require the expertise of specialists in analysis. as
opposed to the design engineer who has many additional concerns. Nevertheless. it is
believed to be timely, and indeed necessary, to include an introduction to nonlinear
analysis in a text on the analysis of framed structures, particularly those of the civil
engineering variety. Techniques for its practical application are at hand. and there is
a growing awareness of their place in design.
Similarly in the first edition we decided to exclude computer programs, citing in its
preface their rapid obsolescence and the widespread availability of commercial pro
grams such as STRUDL. Advances in all areas of computer technology over the past
vi Preface to the Second Edition
twenty years have changed this. Particularly relevant to our subject is the capacity for
packaging structural analysis programs of significant power in an interactive graphic
supported medium suitable for running on readily available personal computers. The
computer is not a substitute for mastery of the subject matter, but in the ability for the
user to control the terms of the analysis and to picture the results that it now provides,
it can be an invaluable aid to understanding the theory and appreciating the physical
significance of the outcome. We believe these pedagogical advantages outweigh the
inevitability of obsolescence and are therefore packaging with every copy of the text
MASTAN21, an interactive graphics program with provisions for geometric and ma
terial nonlinear analysis, as well as conventional linear elastic anlysis. MASTAN2 is
based on MATLAB2, a premier software package for numeric computing and data
analysis.
Also, whereas we used SI units exclusively in the first edition, in this one many of
the examples and problems-particulary those of a nonlinear nature that have strong
implications for design-are in Imperial units. We do this because the SI system has
not been fully assimilated in practice, and we believe that appreciation of the physical
significance of the results is essential to their understanding.
Our aims for the present edition are explained further in the following discussion of
the motivation and salient features of the respective chapters:
Three purposes are served by Chapter 1, the introduction. First, we present a concise
sketch of the development of the subject. One intent of this history is to emphasize
that computerized structural analysis methods are merely one part of a continuing
process that extends back more than 150 years. Second, the role that computerized
structural analysis has played in the design of existing structures is outlined. Finally,
the computer capabilities themselves are tied to the programs written for structural
analysis. Particular attention is given to the development and use of interactive com
puter graphics.
Chapters 2-5 represent closely allied subject matter related primarily to the direct
stiffness method. Chapter 2 serves to define terminology, coordinate systems, and the
most fundamental notions of structural behavior, but it also contains two developments
of great generality. The first is the basic character of elemental relations in the form
of stiffness and flexibility and their transformability from one to another and even to
alternative formats. The second is the fundamental idea of direct stiffness analysis,
described here by means of the simplest structural element.
A more formal treatment is given to direct stiffness analysis in Chapter 3 and, con
sequently, it is possible to examine more closely the implications for large-scale prac
tical computation. The latter include considerations such as the characteristics of the
algebraic equations that are to be formed and solved. In Chapters 4 and 5 the remain
ing tools needed for the linear elastic stiffness analysis of complete frames are estab
lished. The stiffness matrix of a rather general space-frame element is formulated and
then applied, in illustrative examples, to a variety of specific situations.
Two comments on Chapters 2-5 are appropriate. The first is that, except for some
editing of the text and modification of the examples and problems, they are identical
to the same chapters of the first edition. The second is that this initial development of
the subject features the basic physical conditions of structural analysis: equilibrium,
compatibility, and the mechanical properties of the material. Reliance on work and
energy concepts is kept to a minimum. Our feeling is that it is preferable to help the
1The name MASTAN2 is an acronym for Matrix Structural Analysis, 2ct Edition. It was developed by the
authors for educational purposes only. It has no relationship to any existing commercial structural or finite
element analysis program.
2MATLAB is a registered trademark of The MathWorks Inc., 3 Apple Hill Drive, Natick, MA 01760-2098.
'
Preface to the Second Edition vii
student become attuned to the overall approach to computerized structural analysis in

this way, before proceeding to the more powerful and versatile-but more mathe
matical-virtual work-based methods.
We do introduce the formulation of matrix structural analysis on the basis of virtual
work concepts in Chapters 6 and 7. A treatment of this type is necessary to give
maximum scope to various aspects of practical design analysis of frameworks, such as
tapered members and distributed loads, as well as to the study of nonlinear analysis
and the finite element approach. The theoretical groundwork of the virtual work prin
ciple is laid in Chapter 6. Both virtual displacement and virtual force concepts are
covered, but far greater attention is given to the former because of their role in stiffness
formulations. Chapter 7 examines the implementation of the virtual work principle in
linear elastic matrix structural analysis.
Chapters 8-10 comprise an introduction to the nonlinear matrix analysis of frame
works. The approach is one of using selected structural systems and types of material
as the basis for explaining how theory can be translated into procedures useful in the
design office. In the interest of focusing on sources and end results in these chapters,
we defer to Chapter 12 the treatment of methods and details of nonlinear equation
solution and to appendices the coverage of some aspects of the basic structural me
chanics and behavior
An overview of sources of structural nonlinearity, levels of analysis used in dealing
with them, and a system for casting the required equations in matrix stiffness method
form is presented in Chapter 8. In the course of doing this the notions of the geometric
stiffness matrix and a plastic reduction matrix are introduced. Also, the range of con
ditions to be considered is illustrated by the solution of some elementary problems by
traditional methods. In Chapter 9, the groundwork prepared in the previous chapter
is used as the base for the development and application of a matrix method for the
nonlinear elastic (geometric nonlinear) analysis of both planar and space frames. The
calculation of the elastic critical load of ideal systems is also treated. Chapter 10 em
ploys the same base, plus the concepts of plastic hinge theory, for the material and
geometric analysis of planar systems. Included is the determination of the inelastic
critical load of ideal systems by an adaptation of tangent modulus theory.
Some of the more popular methods of solution of linear algebraic equations are stud
ied in Chapter 11, and the same is done for nonlinear equations in Chapter 12. Equation
solving is more in the realm of applied mathematics, numerical analysis, or computer
science than in structural engineering. Nevertheless, since primary responsibility for
the entire analysis generally belongs to structural engineers, they should have more
than superficial knowledge of the important aspects of equation solving. Operations
peculiar to the handling of equations in MASTAN2 are explained in detail. Otherwise,
these chapters are introductions to the subject that summarize useful methods, men
tion pitfalls that may be encountered, and give guidance to more specialized sources.
In the interest of developing basic stiffness methods of analysis in the earlier chap
ters, we put off consideration of some procedures essential to the efficient and realistic
solution of practical design analysis problems. Chapter 13 is devoted to the exposition
of certain of these practices, including condensation of analysis equations prior to
solution by elimination of specified unknowns, substructuring, the imposition of con
straints, connection stiffness and finite joint size, the exploitation of symmetry and
antisymmetry, and procedures for the economical reanalysis of structures when in an
iterative design sequence. Attention is also given to various types of coordinate systems
that are alternative to global coordinates, such as the local coordinate systems that are
convenient for sloping supports.
In Appendix A, principles of structural mechanics and geometry fundamental to
solving the problem of equilibrium of a deformed structure are reviewed and then
viii Preface to the Second Edition
employed in the derivation of terms that were incorporated without proof in the geo
metric stiffness matrices of Chapter 9. There are several reasons for-placing this ma
terial in an appendix rather than in the main body of the text. Throughout the text we
adhered to standard matrix notation. Concise treatment of the appended material
requires use of some elementary concepts of tensors and indicial notation. It also
requires the use of some less obvious definitions of stress and strain, such as second
Piola-Kirchoff stresses and Green-Lagrange strains, and an excursion into the analysis
of finite rotations. These changes in style and alteration of the analytical base are not
abstruse, but we believe their introduction in the earlier chapters would have been an
unnecessary intrusion on the main agrument. Nevertheless, although it is material that
is not essential to an initial study, it is fundamental to a true understanding of the
processes of nonlinear analysis and to their extension beyond the limits explored in
the main text.
The problem of nonuniform torsion in three-dimensional frameworks is also ad
dressed in this appendix but for a somewhat different reason. Practically, it is a key
element in the resistance of many systems to the destabilizing effects of combined
torsion and flexure. Failure to treat the subject would be a serious omission. But con
siderable theoretical and experimental research is needed to reduce its analysis to an
established procedure. The approach described has a sound basis. It is worthy of ap
plication, but it must be viewed in the context of an area still very much under
development.
Appendix B contains a discussion of the problem of distinguishing between the
effects of rigid body motion and displacements resulting from the deformation of a
body. It is a presented in an appendix rather than in the main body of the text because
it is more of a critique rather than the presentation of methods, something that was
done in Chapter 12.
The contents outlined here represent subject matter on the scale of two three-credit
hour, one-semester courses, with considerable selectivity of coverage and latitude in
arrangement available to the instructor. For example, the first seven chapters plus
selected portions of Chapters 11 and 13 might serve for an introductory junior- or
senior-level course. At the same level, but for classes of students with a previous,
limited introduction to the subject, the first five chapters could serve for review and
reinforcement of understanding of basic concepts, with formal instruction starting with
the virtual work approach (Chapters 6 and 7). The econd level course, at the elective
senior or introductory graduate level, might feature the virtual work approach to both
linear and nonlinear elastic analysis, material nonlinear analysis, and stability studies
(Chapters 6-10, Chapter 12, additional portions of Chapter 13, and reference to the
appendices). The program MASTAN2 can be a valuable adjunct at all levels: for dem
onstration, for review of the text's numerous examples, for problem solving, as the
base for analysis and design studies, and in support of individual programming. An
attractive application is to have students develop their own programs, either as alter
natives to the analysis routines in MASTAN2 or as additions to the program's existing
routines.
Our co-author, Professor R. H. Gallagher, died in September 1997, while this edition
was in preparation. The idea for the original book was his, and he was active in the
planning of this edition and in the revision of several chapters. His contributions are
vital to the work. The remaining authors prepared the material on nonlinear analysis
and we bear sole responsibility for any of its shortcomings.
William McGuire
Ronald D. Ziemian
Acknowledgments
We are indebted to a number of people for their help in the preparation of this edition.
Our work would not be complete without an acknowledgment of this debt and a par
ticular offering of thanks to the following:
To Professors J. F. Abel and G. G. Deierlein for using drafts as classroom notes and
for their wise advice throughout.
To Professors F. C. Filippou, J. F. Hajjar, E. M. Lui, and G. H. Paulino for their
thorough, constructive reviews of an early draft and to Professor K. M. Mosalam for
the benefit of his experience in using it in class.
To the men whose Cornell graduate research contributed to the development of
material in the book: J. L. Gross, C. I. Pesquera, M. Gattass, J. G. Orbison, Y. B. Yang,
S. I. Hilmy, J. L. Castaner, S. N. Sutharshana, D. W. White, C. S. Chen, C. Chrysos
tomou, and M. R. Attalla.
To Ms. N. Bulock and her colleagues at the MathWorks, Inc. and Professor
J. Maneval of Bucknell University for their MATLAB support in developing
MASTAN2.
To Ms. P. Welzel for her secretarial assistance in preparing the manuscript for
publication.
ix
Contents
Symbols xvii
Chapter 1 Introduction 1
1.1 A Brief History of Structural Analysis 2

1.2 Computer Programs 3
1.2.1 Computational Flow and General Purpose Programs 4
1.2.2 The Program MASTAN2 6
References 8
Chapter 2 Definitions and Concepts 9
2.1 Degrees of Freedom 9

2.2 Coordinate Systems and Conditions of Analysis 11
2.3 Structure Idealization 14
2.4 Axial Force Element-Force-Displacement Relationships 16
2.4.1 Element Stiffness Equations 16
2.4.2 Element Flexibility Equations 18
2.5 Axial Force Element-Global Stiffness Equations 19
2.6 Examples 20
2.7 Problems 26
Chapter 3 Formation of the Global Analysis Equations 31
3.1 Direct Stiffness Method-The Basic Equations 31

3.2 Direct Stiffness Method-The General Procedure 39
3.3 Some Features of the Stiffness Equations 46
3.4 Indeterminacy 47
3.5 Problems 49
References 55
Chapter 4 Stiffness Analysis of Frames-I 56
4.1 Stress-Strain Relationships 56

4.2 Work and Energy 58
4.3 Reciprocity 60
4.4 Flexibility-Stiffness Transformations 62
4.4.1 Stiffness to Flexibility Transformation 62
4.4.2 Flexibility to Stiffness Transformation 63
xi
xii Contents
4.5 The Framework Element Stiffness Matrix 66

4.5.1 Axial Force Member 67
4.5.2 Pure Torsional Member 67
4.5.3 Beam Bent About Its z Axis 68
4.5.4 Beam Bent About Its y Axis 72
4.5.5 Complete Element Stiffness Matrix 73
4.6 A Commentary on Deformations and Displacement Variables 74
4.6.1 Neglected Deformations 74
4.6.2 Displacement Variables 75
4.7 Examples 76
4.8 Problems 88
References 92
Chapter 5 Stiffness Analysis of Frame-II 93
'i l Coordinate Transformations 93

5.1.1 Transformation Matrices 94
5.1.2 Transformation of Degrees of Freedom 98
5.1.3 Transformations and Energy 98
5.1.4 Rectangular Transformation Matrices 100
5.2 Loads Between Nodal Points 108
5.3 Self-Straining-Initial and Thermal Strain Conditions 120
5.3.1 Initial Strain Problems 122
5.3.2 Thermal Strain Problems 125
5.4 Problems 130
Chapter 6 Virtual Work Principles 137
6.1 Principle of Virtual Displacements-Rigid Bodies 138

6.2 Principle of Virtual Displacements-Deformable Bodies 142
6.3 Virtual Displacement Analysis Procedure and Detailed Expressions 144
6.3.1 General Procedure 144
6.3.2 Internal Virtual Work 145
6.3.3 External Virtual Work 148
6.4 Construction of Analytical Solutions by the Principal of Virtual
Displacements 150
6.4.1 Exact Solutions 150
6.4.2 Approximate Solutions and the Significance of the Chosen Virtual
Displacements 153
6.4.3 Examples 154
6.5 Principle of Virtual Forces 158
6.5.1 Equations of Equilibrium 158
6.5.2 Characteristics of Virtual Force Systems 161
Contents xiii
6.5.3 Formulation of the Virtual Forces Principle 162

6.5.4 Construction of Analytical Solutions by the Virtual Forces
Principle 164
6.6 Problems 170
References 173
Chapter 7 Virtual Work Principles in Framework Analysis 174
7.1 Description of the Displaced State of Elements 175

7.1.1 Definition of the Shape Function Mode of Description 175
7.1.2 Formulation of Shape Functions 176
7.1.3 Characteristics of Shape Functions 180
7.2 Virtual Displacements in the Formulation f Element Stiffness
Equations 181
7.2.1 Construction of Expressions for Real and Virtual Displacements 181
7.2.2 Virtual Displacements Formula for an Element Stiffness Matrix 182
7.2.3 Application to Standard Axial, Torsional, and Flexural Elements 183
7.3 Nonuniform Elements 184
7.4 Nonuniform Torsion 186
7.4.1 An Element Stiffness Matrix 187
7.4.2 Application and Examples 190
7.5 Loads Between Nodal Points and Initial Strain Effect-A General
Approach 194
7.6 Virtual Forces in the Formulation of Element Force-Displacement
Equations 199
7.6.1 Construction of Element Equations by the Principle of Virtual
Forces 199
7.6.2 Further Applications-Shearing Deformations and Curved
Elements 203
7.7 Problems 209
References 214
Chapter 8 Nonlinear Analysis of Frames-An Introduction 216
8.1 Nonlinear Behavior, Analysis, and Design 216

8.1.l Sources of Nonlinearity 217
8.1.2 Levels of Analysis 218
8.1.3 Examples from Established Theory 220
8.1.4 A Commentary on Stability 230
8.2 A Matrix Approach 234
8.3 The Equations of Analysis and Their Solution 235
8.3.1 Equation Solution-The Options 236
8.3.2 A Fundamental Problem 237
8.4 Problems 239
References 241
xiv Contents
Chapter 9 Geometric Nonlinear and Elastic Critical Load Analysis 242
9.1 Geometric Stiffness Matrices for Planar Elements 243

9.1.1 Axial Force Member 243
9.1.2 Combined Bending and Axial Force 245
9.1.3 Examples of Plane Structure Analysis 246
9.2 . Combined Torsion and Axial Force 254
9.3 Three Dimensional Geometric Nonlinear Analysis-An Overview 257
9.4 Examples of Three Dimensional Structure Analysis 259
9.5 Problems 263
References 268
Chapter 10 Material Nonlinear Analysis 269
10.1 Nonlinear Material Behavior 269

10.1.l Plasticity Theory 270
10.1.2 Plastic Analysis 272
10.1.3 Further Considerations 273
10.2 A Plastic Hinge Method For Ductile Frames 274
10.2.1 The Yield Surface and a Plastic Reduction Matrix 274
10.2.2 Definition of the Yield Surface 277
10.3 Inelastic Critical Load Theory 278
10.4 Examples 279
10.5 The Yield Surface Concept-A Brief Survey of Further Applications 290
10.5.1 Spread of Plasticity 290
10.5.2 Multiple Yield Surfaces 292
10.5.3 Reinforced Concrete Members 293
10.6 Problems 294
References 299
Chapter 11 Solution of Linear Algebraic Equations 301
11.1 The Basic Choice-Direct Inversion versus Elimination or Iteration 302

11.2 Direct Elimination Methods 302
11.2.1 Gauss Elimination 304
11.2.2 Cholesky Method 307
11.3 Iterative Meiiivus 310
11.3.1 Gauss-Seidel and Jacobi Iterations 310
11.3.2 Conj ugate Gradient Method 312
11.4 Sparseness and Bandedness 316
11.5 Frontal Solvers 317
11.6 Errors 322
11.6.1 The Condition Number 326
11.6.2 Estimate of Condition Number 327
Contents xv
11.6.3 Error Estimates and Preconditioning 329

11.6.4 Detecting, Controlling, and Correcting Error 333
11.7 Problems 325
References 337
Chapter 12 Solution of Nonlinear Equilibrium Equations 339
12.1 Incremental Analysis 339

12.2 Incremental Single-Step Methods 340
12.2.1 Euler Method 341
12.2.2 Second-Order Runge-Kutta Methods 341
12.3 Incremental-Iterative Methods 345
12.3.1 Load Control Method 347
12.3.2 Displacement Control Methods 348
12.3.3 Work Control Method 349
12.3.4 Constant Arc Length Methods 349
12.3.5 Modified Iterative Technique 349
12.3.6 Convergence Criteria 350
12.4 Automatic Load Incrementation 351
12.4.1 Change in Stiffness 351
12.4.2 Number of Iterations 352
12.5 Element Result Calculations 352
12.5.1 Updating Deformed Geometry 352
12.5.2 Force Recovery 352
12.6 Plastic Hinge Constraints 355
12.7 Limit Point and Post-Limit Analysis 358
12.8 Critical Load Analysis-An Eigenproblem 360
12.8.1 Reduction to Standard Form 361
12.8.2 Polynomial Expansion 363
12.8.3 Power Method 364
12.8.4 Inverse Iteration 367
12.9 Problems 370
References 373
Chapter 13 Special Analysis Procedures 375
13.l Condensation 376

13.2 Substructuring 377
13.3 Constraints 385
13.4 Joint Coordinates 388
13.5 Connections and Joints 393
13.5.1 Flexible Connections 393
13.5.2 Finite Joint Size 398
13.6 Symmetry and Antisymmetry 403
m Contents
13.7 Reanalysis Techniques 408

13.8 Problems 413
References 418
Appendix A Nonlinear Analysis-A Further Look 419
A.1 Virtual Displacement Principles in Lagrangian Formulations 420

A.2 An Updated Lagrangian Formulation and Its Linearization 425
A.3 Application to the Framework Element 426
A.4 Finite Rotations and Equilibrium in the Deformed Configuration 431
A.4.1 The Finite Rotation Equation 432
A.4.2 Application of the Finite Rotation Equation to the Framework
Element 433
A.5 Nonuniform Torsion 437
References 441
Appendix B On Rigid Body Motion and Natural Deformation 444
B.1 The Nature of the Problem 444

B.2 Distinguishing Between Rigid Body Motion and Natural Deformation 445
B.3 Critique of Force Recovery Methods 447
B.3.1 The Natural Deformation Approach 447
B.3.2 The Elementary Approach 448
References 450
Author Index 451

Subject Index 453
Symbols
In matrix structural analysis, many physical quantities and mathematical operations

must be represented symbolically. Preparation of the equations of analysis in a form
suitable for computer solution requires that all symbols used be defined in a rigid
fashion amenable to numerical interpretation. On the other hand, the development of
these equations, with stress upon their physical significance, is often accomplished
through the use of simple, less formal symbols-symbols that vary with the principle
under discussion and have a clear physical connotation in the case at hand. In the
interest of generality and uniformity, we shall use some basic symbols to denote certain
quantities throughout the text. But the precise interpretation of any of these symbols
must be obtained from the local context in which it is used, and in which it will be
explained.
In general, we use the letter Pto designate applied direct forces and Pm to designate
applied moments. Similarly, Rand Rm are used for direct and moment reactions. At
their ends and within elements, we generally denote direct forces by F and moments
by M. All of these symbols may carry clarifying subscripts and superscripts to indicate
direction, point of application, or member to which the symbol applies. The symbols
may appear in either single component or vector form. The symbols u, v, and w will
designate translational displacements in the x, y, and z directions and Bx, By and fJz
r.Qtational displacements about these axes. Generally, the letter k will refer to a stiffness
quanllty-a,_nd d to a flexibility quantity.
Matrices are denoted by a boldface letter within the symbols [ ] (for a square or
rectangular matrix), { ) (for a column vector), L J (for a row vector), and I J for a
diagonal matrix. Matrix operations are represented by [ r1 as the inverse, [ ]T as the
transpose, I I as a determinant, and II II as a norm.
As a further guide, the following is a list of the principal symbols used in the text.
As indicated above, most of these may contain clarifying or modifying subscripts,
or supplementary marks (overbars -, hat symbols, etc.) that will be defined in con
text. The same applies to the individual components of matrices and vectors listed
below.
A Area
B Bimoment
Cw Warping constant
[d] Element flexibility matrix
[D] Global flexibility matrix, diagonal matrix
E, [E] Elastic modulus, matrix of elastic constants
e Normal strain
e Infinitesimal strain tensor
F Normal or shearing force
{F) Vector of element nodal forces
xvii
xviii Symbols
G Shear modulus
\G) Gradient vector
[G] Constraint equation coefficients

HBW Half-bandwidth
I Moment of inertia
IP Polar moment of inertia
[I] Identity matrix

J St. Venant torsion constant
k Spring stiffness, rotational spring stiffness
[k] Element stiffness matrix

[K] Global stiffness matrix
L Length
[L] Lower triangular matrix

l, m, n Direction cosines
m Ratio of bending moment to plastic moment
Bending or twisting moment
Plastic moment
Vector of element shape functions
n Number of degrees of freedom, number of nodes
[OJ, \OJ Null matrix and vector

{ P) Vector of global nodal forces
p Axial force
PY Axial yield or squash load
p Ratio of axial force to squash load, number of elements
{Q) Vector of element force distribution functions
q Distributed load intensity
IR) Vector of reaction forces, unbalanced load vector
lrl Residual vector

r Radius of gyration
s Current stiffness parameter
Is l Basis vector
T Temperature change from stress-free state, torsional moment
T Cauchy stress tensor
t Second Piola-Kirchoff stress tensor
[T] Force transformation matrix

U, U* Strain energy and complementary strain energy
[U] Upper triangular matrix

U, V, tt' Displacement components
v Potential energy of applied loads
Vol Volume
W,W* Work and complementary work
x,y,z Cartesian coordinates

z Plastic section modulus
Symbols xix
Greek Symbols
a Coefficient of thermal expansion, factor defined in context
a, {3, 8 Direction angles
/3 Rate of twist, relaxation factor
[r], [ 'Y] Transformation matrix
'Y Shear strain
{&} Vector of nodal point displacements

Displacement
8 Virtual quantity, variation

Green Lagrange strain tensor
Absolute percent relative error
e Angular displacement
K Curvature, condition number
Eigenvalue, load ratio, plastic strain factor, effective length factor
JI Poisson's ratio
Nondimensional coordinate
Total potential energy
7T 3.1416...
p Radius of curvature
T Shear stress, fraction of load increment
Normal stress
Uy Yield stress
{Y} Eigenvector
 Yield surface function
[cl>] Static equilibrium matrix
</> Angle of measure
? Acceptable percent error tolerance

w Eigenvalue
In addition to the above literal and matrix symbolism, we shall use the following
graphic symbols wherever it is desired to indicate or to stress some particular charac
teristic of a force or structure.
Direct force and moment components
0000 )l Direct force and moment resultants
I
I
)l Reactive force and moment components
Alternative representation of moments

C'
xx Symbols
Pinned joint

Rigid joint
"\r
Pinned connection
'
Roller support

Pinned support

Elastic support
I
Fixed support
Chapter 1
Introduction
One of the responsibilities of the structural design engineer is to devise arrangements

and proportions of members that can withstand, economically and efficiently, the con
ditions anticipated during the lifetime of a structure. A central aspect of this function
is the calculation of the distribution of forces within the structure and the displaced
state of the system. Our objective is to describe modern methods for performing these
calculations in the particular case of framed structures. The number of structures that
are actually simple frameworks represents only a part of those whose idealization in
the form of a framework is acceptable for the purposes of analysis. Buildings of various
types, portions of aerospace and ship structures, and radio telescopes and the like can
often be idealized as frameworks.
In design, both serviceability limit states and strength limit states should be considered.
A serviceability limit state is one in which the structure would become unfit for normal
service because of excessive deformation or vibration, or problems of maintenance or
durability. Generally, linear elastic analysis gives a good picture of the actual response
to service loads. A strength limit state is one in which the structure would become
unsafe. Except for sudden brittle or fatigue crack propagation, the attainment of a
strength limit is generally the result of increasingly nonlinear elastic or inelastic re
sponse that culminates in structural instability. Most often, the internal load distribu
tion at a strength limit state is calculated by a linear method. In proportioning mem
bers, the analytical results are modified in an empirical or judgmental way to account
for the effects of nonlinearity. Whereas this practice is common, the availability and
application of practicable methods of nonlinear analysis is steadily increasing. One of
the aims of this book is to provide an introduction to typical methods of this type.
Fundamentally, the behavior of all types of structures-frameworks, plates, shells,
or solids-is described by means of differential equations. In practice, the writing of
differential equations for framed structures is rarely necessary. It has been long estab
lished that such structures may be treated as assemblages of one-dimensional members.
Exact or approximate solutions to the ordinary differential equations for each memqer
are well-known. These solutions can be cast in the form of relationships between the
forces and the displacements at the ends of the member. Proper combinations of these
relationships with the equations of equilibrium and compatibility at the joints and
supports yields a system of algebraic equations that describes the behavior of the
structure.
Structures consisting of two- or three-dimensional components-plates, membranes,
shells, solids-are more complicated in that rarely do exact solutions exist for the
applicable partial differential equations. One approach to obtaining practical, numer
ical solutions is the finite element method. The basic concept of the method is that a
1
2 Chapter 1 Introduction
continuum (the total structure) can be modeled analytically by its subdivision into
regions (the finite elements), in each of which the behavior is described by a set of
assumed functions representing the stresses or displacements in that region. This per
mits the problem formulation to be altered to one of the establishment of a system of
algebraic equations.
The practical, numerical solution of problems in structural analysis thus is seen to
involve the formation and solutions of systems-sometimes very large systems-of
algebraic equations. Also, it should be fairly clear that a member of a framed structure
is simply one example of a more broadly defined family of finite elements.
Viewed in this way, structural analysis may be broken down into five parts:
1. Basic mechanics. The fundamental relationships of stress and strain, compati

bility, and equilibrium.
2. Finite element mechanics. The exact or approximate solution of the differential
equations of the element.
3. Equation formulation. The establishment of the governing algebraic equations
of the system.
4. Equation solution. Computational methods and algorithms.
5. Solution interpretation. The presentation of results in a form useful in design.
This book deals chiefly with parts 3, 4, and 5. Specifically, it is on matrix structural
analysis. This is the approach to these parts that currently seems to be most suitable
for automation of the equation formulation process and for taking advantage of the
powerful capabilities of the computer in solving large-order systems of equations. An
understanding of basic structural mechanics and basic matrix algebra is presumed. Only
that segment of the finite element method relating to framed structures will be in
cluded, other aspects being left to texts specializing in the fundamentals of finite ele
ment mechanics and procedures (e.g., Refs. 1.1 and 1.2). Computational methods and
algorithms will be discussed in Chapters 11 and 12, but more comprehensive coverage
can be found in books on numerical analysis. (e.g., Ref. 1.3).
An appreciation of the approach to structural analysis we are taking requires some
understanding of the history of this and related subjects. The following brief review
may help.
1.1 A BRIEF HISTORY OF STRUCTURAL ANALYSIS
Although it was immediately preceded by the great accomplishments of the school of

French elasticians, such as Navier and St. Venant, the period from 1850 to 1875 is a
logical starting point for our review. The concept of framework analysis emerged dur
ing this period, through the efforts of Maxwell (Ref. 1.4), Castigliano (Ref. 1.5), and
Mohr (Ref. 1.6), among others. At the same time, th concepts of matrices were being
introduced and defined by Sylvester, Hamilton, and Cayley (Ref. 1.7). These concepts
are the foundations of matrix structural analysis, which did not take form until nearly
80 years later.
An excellent chronicle of developments in structural mechanics in the period 1875
to 1920 is found in Timoshenko's History of Strength of Materials (Ref. 1.8). Very little
progress was made in this period in the development of theory and the analytical
techniques subsidiary to matrix structural analysis. To a great extent, this was due to
practical limitations on the solvability of algebraic equations with more than a few
unknowns. For the structures of primary interest in that period-trusses and
frames-an analysis approach based on member forces as unknowns was almost uni
versally employed.
Around 1920, as a result of the efforts of Maney (Ref. 1.9) in the United States and
Ostenfeld (Ref. 1.10) in Denmark, the basic ideas of a truss and framework analysis
approach based on displacement parameters as unknowns took form. These ideas
represent the forerunners of matrix structural analysis concepts in vogue today. Severe
limitations on the size of the problems that could be handled by either force or dis
placement unknowns nevertheless prevailed until 1932 when Hardy Cross introduced
the method of moment distribution (Ref. 1.11). This method made feasible the solution
of structural analysis problems that were an order of magnitude more complex than
the most sophisticated problems treatable by the prior approaches. Moment distribu
tion became the staple of structural frame analysis practice for the next 25 years. Today
it remains of value as a ready method for solving small problems and as an aid in
visualizing behavior, but it is primitive in power compared to computer methods and
has been superseded by them in the solution of large problems.
Association of the mathematical concepts of matrix theory with the engineering
concepts of structural analysis appeared in the 1930s, chiefly in the work of Frazer,
Duncan, and Collar (Ref. 1.12). The liaison developed erratically through the 1940s,
but there was no motivation for a firm union until digital computers were born in the
early 1950s. Individuals who foresaw the impact of computers on both theory and
practice then undertook the codification of the well-established framework analysis
procedures in the format best suited to the computer, the matrix format. Two note
worthy developments were the publications of Argyris and Kelsey (Ref. 1.13) and
Turner, Clough, Martin, and Topp (Ref. 1.14). These papers wedded the concepts of
framework analysis and continuum analysis and cast the resulting procedures in a
matrix format. They were a strong influence on developments in subsequent years. An
important factor in the acceptance of computerized structural analysis in the civil en
gineering profession has been the series of ASCE conferences on electronic compu
tation, especially the first three (Refs. 1.15-1.17).
The desirability of accounting for nonlinear elastic and inelastic behavior was rec
ognized from the earliest days of the development of structural analysis, and significant
contributions were made in the nineteenth century. By 1888 a practical theory for
including the effect of elastic displacements in the analysis of suspension bridges had
been evolved by Melan (Ref. 1.18), and in 1899 Goodman presented a clear expla
nation of the plastic hinge concept (Ref. 1.19). In the early years of the twentieth
century, Steinman published an English translation of Melan's deflection theory (Ref.
1.20) and, as reviewed in Reference 1.21, contributions to plastic analysis and design
were made by Kazinczy, Maier-Liebnitz, and others. But it was not until the latter half
of this century that reasonably general, practicable methods for including both geo
metric and material nonlinearity in analysis became a reality.
Among the seminal events of this period was Turner, Dill, Martin, and Melosh's
1960 publication of a geometric stiffness matrix for the uniform axial force member
(Ref. 1.22). On the material behavior side, a dramatic thrust occurred between the end
of World War II and 1965. Most of the significant developments of that period are
reported in References 1.21, 1.23, and 1.24. Since then, there has been a steadily in
creasing flow of contributions to the theory and application of nonlinear methods of
analysis and design. Most of them have been incremental, but their cumulative effect
has been profound.
1.2 COMPUTER PROGRAMS
When the structural engineer leaves the university and enters the design office, it is
probable that he or she will encounter computational hardware and software of con-
siderable sophistication. The available programs may require only the simplest data in
order to bring about solutions for the problems at hand. With such tools available, it
is natural for the student of engineering to question the utility, or even the necessity,
of what may appear to be a study of already-programmed procedures. The answer to
this question is in the responsibilities of the design engineer. The decisions that must
be made by this individual, extending from the analytical description of the structure,
thence to the evaluation of the correctness of the numerical analysis, and through to
the interpretation of the output, are not amenable to computerization. The structural
engineer must understand the principles of analysis. Computerization has relieved the
burdens of the rote operations in design analysis, but it cannot relieve the engineer of
responsibility for the designed structure.
Throughout this book, therefore, illustrative numerical examples and problems are
used. Most of them are simple. The aim is to avoid obscuring the point or principle
under study in a cloud of numerical computations. But in the process the advantage
of matrix notation and the matrix approach may not become immediately apparent
when it is clear that the algebraic equations for these elementary structures could have
been compiled in a less formal, and perhaps more physically obvious, way. Such is not
the case for most structures encountered in practice, however. They are larger, more
complicated, and generally require more extensive calculations. It is the efficient anal
ysis of such systems that we are mainly concerned with. Although adopting the matrix
approach may not influence the way we think about structural behavior, one of its
main virtues is that mentioned earlier: It provides the best format presently known for
analyzing structures by computer.
1.2.1 Computational Flow and General Purpose Programs
The equations of the matrix/finite element approach are of a form so generally appli
cable that it is possible in theory to write a single computer program that will solve an
almost limitless variety of problems in structural mechanics. Many commercially avail
able general-purpose programs attempt to obtain this objective, although usually on a
restricted scale. The advantage of general-purpose programs is not merely this capa
bility but the unity afforded in the instruction of prospective users, input and output
data interpretation procedures, and documentation.
The four components in the flowchart of Figure 1.1 are common to virtually all
general-purpose, finite element analysis programs. As a minimum, the input phase
\
should require of the user no nformation beyond that relating to the material of
construction, geometric description of the finite element representation (including sup
port conditions), and the conditions of loading. The more sophisticated general-pur
pose programs facilitate this input process through such features as prestored material
property schedules and graphical displays of the finite element idealization so that
errors in input may be detected prior to performance of calculations.
In the phase comprised of the library of finite elements is the coded formulative
process for the individual elements. Most general-purpose programs contain a variety
of elements, samples of which are illustrated in Figure 1.2. The only element we are
concerned with here is the framework element. Texts such as References 1.1 and 1.2
describe the use of other elements in analytically modeling a continuum. The element
library phase of the general-purpose program receives the stored input data and es
tablishes the element algebraic relationships by application of the relevant coded for
mulative process. This phase of the program also includes all operations necessary to
position the element algebraic relationships for connection to the neighboring ele
ments as well as the connection process itself. The latter operations produce the al
gebraic equations that characterize the response of structure.
1 Input I
Definition of physical model
geometry, material, loading
and boundary conditions
i
2 Library of elements I
Generation of mathematical
I
models for structural elements
3 Solution and applied loadings
Construction and solution of

mathematical model for
I
structural system
t
4 Output I
Display of predicted
displacements and forces
Figure 1.1 Structural analysis computational flow.
The solution phase operates on the equations of the problem formed in the prior
phase. In the case of a linear static analysis program this may mean no more than the
single solution of a set of linear algebraic equations for a known right-hand side. Non
linear static analysis normally require s a series of incremental linear analyses. Solutions
for dynamic response may require very extensive computations over a time-history of
applied loads. In still other cases, particularly where the number of unknowns is very
large, it may be advantageous to divide the total structure into several substructures.
A finite element idealization of each substructure is made and analyzed, and the results
are properly recombined into a solution for the full structure. Included in the solution
phase are the back-substitution operations needed to obtain all the desired aspects of
the solution.
I
t
I
, I
/
143 --
z I
r -
"-
y
,
.1.-x
---
(a) (b)
z
t
I
?(j
/
/ {j!'y t
- y
x
(c) (d) (e)
Figure 1.2 Sample finite elements. (a) Framework member. (b) Plane stress.
(c) Flat plate bending. (d) Solid element. (e) Curved thin shell.
The output phase presents the analyst with a numerical or pictorial record of the
solution upon which engineers can base decisions regarding the proportioning of the
structure and other design questions. The numerical record is commonly a printed list
of forces or stresses at points or within the finite elements, of displacements of specified
points, and of other desired information. Pictorially, data such as moment diagrams,
contour plots of the principal stresses, or plots of the displaced state of the structure
may be displayed on the computer screen and printed as permanent records.
The phases of the general-purpose program have one feature in common: the mod
ularity of their component aspects. Insofar as possible, the procedures treated in this
text are formulated in a manner and form consistent with those found in widely dis
tributed general-purpose programs. But they should also be of use to the student in
preparing more limited or special-purpose programs. Often, programs for a particular
purpose can be designed to treat that problem more efficiently than the general
purpose programs can.
1.2.2 The Program MASTAN2
A CD containing a structural analysis computer program entitled MASTAN2 is bun

dled with this textbook. As shown in Figures 1.3 and 1.4, MASTAN2 is an interactive
graphics program that provides preprocessing, analysis, and postprocessing capa
bilities. Preprocessing options include definition of structural geometry, support con
ditions, applied loads, and element properties. The analysis routines provide the
user the opportunity to perform first- or second-order elastic or inelastic analyses
of two- or three-dimensional frames and trusses subjected to static loads. Post-
N19
Figure 1.3 Preprocessing capabilities of MASTAN2.

N3 0.848 N7
0.739
N2
(a) Deformed shape and plastic hinge locations
r--;'- - - - - -
,'/ /
r
0.8
I
2
""
I
c::.:: 0.6 I
"tl
"' - - - - - - - l. 1st-Order Elastic
0
..J
I - - - - 2. 2nd-Order Elastic
"tl
- 0.4
I - - - 3. 1st-Order Inelastic
---- 4. 2nd-Order Inelastic
0.
0. I
--<
/,
0.2
2 4 6 8 10 12 14 16
Lateral Displacement (Node 7)
( b) Response curves
Figure 1.4 Postprocessing capabilities of MASTAN2.
processing capabilities include the interpretation of structural behavior through de

formation and force diagrams, printed output, and facilities for plotting response
curves.
In many ways, MASTAN2 is similar to today's commercially available software in
functionality. The number of pre- and post-processing options, however, has been
limited to minimize the amount of time needed for a user to become proficient with
it. Many of the theoretical and numerical formulations presented in the context of the
following chapters provide a basis for the program's linear and nonlinear analysis
routines. In this regard, the reader is strongly encouraged to use this software as a tool
for demonstration, reviewing examples, solving problems, and perhaps performing
analysis and design studies. MASTAN2 has been written in modular format. There
fore, the reader is also provided the opportunity to develop and implement additional
or alternative analysis routines directly within the program. Finally, it should be noted
that MASTAN2 will execute on any computing platform where MATLAB is available.
REFERENCES
1.1 R. H. Gallagher, Finite Element Analysis: Fundamentals, Prentice-Hall, Engle

wood Cliffs, N.J., 1975.
1.2 K.-J. Bathe, Finite Element Procedures, Prentice-Hall, Englewood Cliffs, N.J.,
1996.
1.3 G. H. Golub and C. F. Van Loan, Matrix Computations, Third Edition, The Johns
Hopkins University Press, Baltimore, MD., 1996.
1.4 J. C. Maxwell, "On the Calculations of the Equilibrium and Stiffness of Frames,"
Phil. Mag. (4), 27, 294 (1864).
1.5 A. Castigliano, Theorie de l'Equilibre des Systemes Elastiques, Turin, 1879 (En
glish translation, Dover Publications, New York, 1966).
1.6 0. Mohr, "Beitrag zur Theorie der Holz-und Eisen Konstruktionen," Zeit des
Architekten und Ingenieur Verienes zu Hannover, 1868.
1.7 A. Cayley, "A Memoir on the Theory of Matrices," Phil. Trans., 148, 17-37
(1857).
1.8 S. Timoshenko, History of Strength of Materials, McGraw-Hill, New York, 1953.
1.9 G. B. Maney, Studies in Engineering-No. 1, University of Minnesota, Minne
apolis, 1915.
1.10 A. Ostenfeld, Die Deformationsmethode, Springer-Verlag OHG, Berlin, 1926.
1.11 H. Cross, "Analysis of Continuous Frames by Distributing Fixed-End Moments,"
Trans. ASCE, 96, 1-10 (1932).
1.U R. A. Frazer, W. J. Duncan, and A. R. Collar, Elementary Matrices and Some
Applications to Dynamics and Differential Equations, Cambridge University
Press, London, 1938.
1.13 J. H. Argyris, and S. Kelsey, Energy Theorems and Structural Analysis, Butter
worth Scientific Publications, London, 1960.
1.14 M. J. Turner, R. Clough, H. Martin, and L. Topp, "Stiffness and Deflection Anal
ysis of Complex Structures," J. Aero. Sci., (23), 9, 805-823 (Sept. 1956).
1.15 Proceedings of the Conference on Electronic Computation. Held in Kansas City,
Mo., 1958, ASCE, N.Y.
1.16 Proceedings of the 2nd Conference on Electronic Computation. Held in Pittsburgh,
Pa., 1960, ASCE, N.Y.
1.17 Proceedings of the Third Conference on Electronic Computation. Held in Boulder,
CO., 1963. Proc. ASCE, J. of the Struct. Div., 89, No. ST4 (Aug. 1963).
1.18 J. Melan, Theorie der eisernen Bogenbrucken und derHiingebrucken, Handbuch
der Ingenieurwissenschaften, Leipzig, 1888.
1.19 J. Goodman, Mechanics Applied to Engineering, Longmans-Green, London,
1899.
1.20 J. Melan, Theory of Arches and Suspension Bridges, translation by D. B. Stein
man, M. C. Clark Pub., Chicago, 1913.
1.21 Plastic Desig>t in Steel, A Guide and Commentary, Manual No. 41, 2nd edition
ASCE, New York, 1971.
1.22 M. J. Turner, E. H. Dill, H. C. Martin, and R. J. Melosh, "Large Deflections of
Structures Subjected to Heating and External Loads" J. Aero/Space Sciences, 27,
February, 1960.
1.23 J. F. Baker, M. R. Horne, and J. Heyman, The Steel Skeleton, Vol. 2, Cambridge
University Press, Cambridge, 1956.
1.24 Plastic Design of Multi-Story Steel Frames, Vols. 1 and 2, Department of Civil
Engineering, Lehigh University, Bethlehem, PA., 1965.
Chapter 2
Definitions and Concepts
This and the following three chapters contain the fundamentals of the displacement
method of matrix structural analysis. Particularly useful components of the flexibility
method are also treated. The terms displacement method and flexibility method refer
to general approaches to analysis. In their simplest forms, both approaches can be
reduced to sets of formal rules and procedures. However, a particular form of the
displacement approach-the direct stiffness method-is dominant in general structural
analysis and is therefore emphasized here. Use of the procedures developed in these
chapters can give one a certain facility in analyzing simple structures, but further study
is needed to appreciate the scope and power of these methods and their under
lying principles of structural behavior. To promote such understanding, we later
explore some of the most important ramifications of the fundamentals developed in
Chapters 2-5.
We begin the present chapter with an explanation of the concept of degrees of
freedom and then follow with a description of the principal coordinate systems and
sign conventions to be used throughout the book. Idealization of framework structures
for the purpose of analysis is examined next. We then proceed to the definition of
influence coefficients: relationships between forces and displacements. Global stiffness
equations for axial force members are developed and their use is illustrated in simple
examples.
2.1 DEGREES OF FREEDOM
In this chapter we are concerned with the overall behavior of the elements of a struc
ture as defined by the displacements of the structure's joints under the action of forces
applied at the joints. The study of the relative displacements of points within individual
members-the strains-and the distribution of forces per unit area-the stresses-is
the subject of later chapters.
Displacement components required for the definition of the behavior of typical
structures are illustrated in Figure 2.1. In the pin-jointed plane truss (Figure 2.la), the
members are stretched or compressed by the applied load. The net effect is the dis
placed structure shown, to exaggerated scale, by the solid lines. Except at supports or
joints that are otherwise constrained, the movement of each joint can be described by
two translational displacement components, such as u and u in the figure. As we'll
demonstrate later, once the u and u components of all joints have been determined,
the entire response-the reactions and member forces as well as th resultant displace
ments-can be defined. For a pin-jointed space truss, three components, such as u, u,
and w in Figure 2.lb, are required at each joint.
9
10 Chapter 2 Definitions and Concepts
(a) (b)
;,;
c
(c) (d)
Figure 2.1 Joint displacements. (a) Pin-jointed plane truss. (b) Pin-jointed space truss.
(c) Plane frame (in-plane loading). (d) Plane frame (out-of-plane loading).
In beams and rigid framed structures the elements are bent and perhaps twisted, in
addition to being stretched or compressed. Figure 2.lc shows the rotational displace
ment component, 8, which, along with the translational components u and u, is neces
sary for the complete definition of the displacement of a joint in a plane frame. In a
frame loaded normal to its plane, as in Figure 2.ld, two rotational components are
needed at a joint. As shown, 81 is the common slope at the b end of member be and
twist at the b end of member ab. Likewise, 82 represents both a slope in ab and twist
in be. By an obvious extension, three translational and three rotational components
will be required at a joint in a rigidly connected space frame.
Each displacement component illustrated in Figure 2.1 is a degree of freedom. In
principle, the number of degrees of freedom of a system is the number of displacement
components or coordinates needed to define its position in space at any time under
any loading. Viewed in this way, all of the structures in Figure 2.1 have an infinite
number of degrees of freedom since each one of the members in each structure is
capable of deforming in an infinite number of modes if suitably excited. In dynamic
analysis it is normally acceptable to avoid this complexity by lumping masses at selected
points and considering only a finite number of degrees of freedom. In static analysis,
in which displacement under a given loading will be in a single, forced mode, further
simplification is possible through pre-analysis of each element and reduction of its
behavior to a function of certain degrees of freedom at the element ends. Also, it is
known that, for many structures, static response is more strongly dependent upon some
types of deformation than upon others. In rigid frames, for example, displacements
resulting from flexure are generally orders of magnitude larger than those due to uni-
form axial straining. In such cases it may be possible to neglect some degrees of free
dom without noticeable loss of accuracy. Therefore, in practice, the number of degrees
of freedom is not unique but is, instead, a function of the way in which the real structure
has been idealized for analysis. Clearly, in this process the judgment and experience
of the analyst come into play, but in any structure there will be a minimum number
of degrees of freedom that must be considered to arrive at an acceptable result.
2.2 COORDINATE SYSTEMS AND CONDITIONS OF ANALYSIS
We shall most often work with a right-hand set of orthogonal axes identified by the
symbols x, y, and z as shown in Figure 2.2. These axes remain fixed throughout the
deformation of the structure, and displacements of points on the structure are referred
to them. Consider a particle located at point g when the structure is in the unloaded,
.
undeformed state. A force vector with components Fxg, Fyg, F,g is applied to this par
ticle. Under the action of this force the particle displaces to the point h. The transla
tional displacement components of the particle are: ug = xh - xg, vg = Yh - yg, and
wg = zh - Zg These displacements could also be shown as vector components at point
g, just as the force components are. Positive values of the force and displacement
components correspond to the positive sense of the coordinate axes.
Except where otherwise noted, we are limiting our attention to linear elastic behav
ior. That is, we are assuming that deformations are small, that material properties such
as the modulus of elasticity remain constant during loading, and that the structure is
nowhere stressed beyond its elastic limit. In Chapters 8-10 we shall discuss the con
sequences of these assumptions, and we'll develop several methods for incorporating
sources of nonlinearity in the analysis. But as long as we restrict our consideration to
linear elastic action, the components of the force vector may be considered to remain
unchanged as the particle moves from g to h. Also, the mechanical aspects of this
behavior, such as the work done by the forces Fx8, Fy8, F,8 acting through the displace
ments u8, v8' w 8' are not dependent upon the path taken to point h. Limiting consid
eration to linear behavior also implies that all equations will be formulated with respect
to the geometry of the original, undeformed structure. Thus, the effect of joint dis
placements on these equations is not considered and the principle of superposition
applies. Under this principle, the response of a structure to the application of a system
of forces is identical to the summation of the responses of the same structure to the
separate application of every force of the system.
Figure 2.2 Displacement from point g to point h.

Figure 2.3 Forces, moments, and corresponding displacements.
When it is necessary to consider moments and rotational displacements, it may be

done as shown at the 2 end of member 1-2 in Figure 2.3. 1 Generalizing the term force
to include moments as well as linearly directed forces, the force components corre
2
sponding to the rotational displacement are the moments Mxz. My2, and M,2 Positive
rotations and positive moments are defined in terms of the right-hand rule. In frame
analysis the simplifying assumption is generally made that a line drawn normal to the
elastic line of the beam or frame element remains normal to that line as the structure
deforms under load. A measure of displacement at the joints of such structures is the
rotation, 8, of the normal with respect to the undeformed state. As thus described in
the coordinate system of the figure, the rotational displacement components for the
point 2 may be represented by
8x2 = aw
ay z
l 8y2 =_aw
ax
l 2
(2.1)
The right-hand rule accounts for the negative sign in the definition of 8y2, since a
positive displacement increment gives a negative rotation 8y23
w
As already indicated, the overall description of the behavior of a structure is accom
plished through the medium of force and displacement components at designated
points, commonly called node points, or just nodes. They are generally the physical
joints of the structure, since these are the points of connection of the elements of the
total, or global, analytical model. There are cases, as in the interior of members of
varying cross section, in which the nodes are arbitrarily selected and do not have a
1The picture is simplified in the interest of focusing on these effects. Direct forces and moments may of
course be present at both ends, as in Figure 2.7.
2
The term action is also used in the same generalized sense. We'll use it, rather than force, on the occasions
in which it is more descriptive of the case at hand.
3Equation 2.1 is based on the analysis of infinitesimal strains. We shall use it consistently. An equally valid
alternative is
0x2 = -ilv
az
l 0
y
2
a
= u
az
l 0
z
2
a
=- u l
ay 2
(2.la)
2 2
The relationships between the two equations and the limitations of both are discussed in Section 4.6.2.
physical significance. Nevertheless, it is common to use the terms node and joint
synonymously.
Similarly, the terms member and element are used interchangeably, but frequently,
the former will connote a complete, physical component of a structure whereas the
latter may refer to a segment of that component. The distinction should always be
clear from the context.
The forces and displacements at the nodes of a given element form column vectors
which we designate as IF) and la}. (Braces, I }, denote column vectors.) For the element
in Figure 2.3, for example, with direct forces at point 1 and moments at point 2,
IF} = LFxl Fy1 Fzl Mx2 My2 Mz2JT
la} = Lu; UI WI {)x2 {)y2 {)z2JT
where, as we shall customarily do to save space, we have listed the contents of these
vectors in transposed row vector form. An individual entry, '1;, in the vector of joint
displacements, la}= L.11 '1; ... '1nY, is termed the ith degree of freedom. For the
case of direct forces and moments at both ends, IF} and la} would be 12 x 1 vectors.
The term release is often used in framework computer analysis programs. In a sim
plified sense, this term, which relates to support conditions, implies that all of the
degrees of freedom at a joint are initially fixed and that certain components of dis
placement at the joint are then recognized as being actually free from constraint, that
is, they are "released." In the roller support condition of point b of Figure 2.la, for
example, the horizontal displacement is a release. Thre is considerably more to the
subject of releases. They will be discussed further and more precise definitions will be
given in Section 13.5.
The first step in the formation of the force and displacement vectors is the definition
of the nodal points and their location with respect to a coordinate system. We distin
guish between global and local coordinate systems.4 The global system is established
for the complete structure. The local (member or element) axes are fixed to the re
spective elements and, since the members are in general differently oriented within a
structure, these axes may differ from element to element. This is illustrated in Figure
2.4. When different types of axes are being compared or when they appear in the same
portion of the text, the local axes will be identified by primes, as in the figure. No
,
primes are used when local axes alone are used.
We use local axes in almost all of our formulations of element equations. Our con
vention for relating them to the elements and for numbering the joints of an element
is illustrated in Figure 2.4b. The local x' axis is directed along the axis of the member.
The joint at or closest to the origin of local coordinates is designated as joint 1 and the
other node as joint 2. This convention is carried over into space structures, but in these
' '
it will be necessary, additionally, to define the orientation of the local y and z axes
with respect to certain directions in a cross section of the element (see Chapter 5).
Global axes figure prominently in the development of the global equations (the
equations of the complete structure) in Sections 3.1-3.3. Given the elastic properties
of an element or structure in terms of local coordinates, the transformation of forces
and displacements from these directions to the global directions can be constructed
easily. The transformation procedures could be introduced at this point. However, in
the interest of proceeding to overall analysis as quickly as possible, we postpone the
subject of coordinate transformation to Section 5.1.
4 A third type o f coordinate system, joint coordinates, will b e discussed i n Section 13.4.
Local axes for

---,
element A
Local axes for
I y I
I v elementB
I
I
I
I I
I
I I
I
I
I
I
I
I
I Global axes
I
------ --,
I I
' x
I 0 ' ,1 ; 0 I
' +-';---- I
-------------- -- --------------
'
J
,_,
(a)
y'
Element
o' 1 2 x'
(b)
Figure 2.4 Coordinate axes and joint numbering. (a) Types of coordinate
axes. (b) Joint numbering scheme in local coordinates.
2.3 STRUCTURE IDEALIZATION
To permit analysis, the actual structure must be idealized. Members, which have width
and depth as well as length, are conventionally reduced to line elements. Their resist
ance to deformation is represented by material properties such as Young's modulus
(E) and Poisson's ratio (v), and by geometrical properties of the cross section, such as
area (A), moment of inertia (/), and torsional constant (J). The behavior of the con
nections, that is, whether pinned, semirigid, rigid, yielding, etc., must be stipulated.
How these idealizing decisions are made, or the necessary properties determined, is
extremely important and involves considerable judgment. In the interest of focusing
on the techniques of analysis, it will be assumed that the line diagrams and member
properties used are valid idealizations of the real structure.
The concepts of idealization and analysis can be described with respect to the simple
ttuss in Figure 2.5, in which the common assumption of pinned joints is implied. For
the purpose of forming a mathematical model the state of stress in the members is
represented by forces at the element ends. The corresponding displacements of these
nodes-the degrees of freedom-are employed in the characterization of the displaced
state of the element. The individual truss elements are isolated, and these forces and
displacements are identified symbolically, as for the typical member in Figure 2.5b.
Then, using the principles of elasticity and the laws of equilibrium, relations are formed
between the joint forces and displacements. The truss is next reconstructed analytically
by examining the equilibrium of member forces at each joint. By summing up the truss
element forces in each direction at each joint in each coordinate direction and equating
the result to the corresponding applied loads, the conditions of equilibrium within the
truss are fully accounted for. The joining of truss members at the joints also ensures
that the truss displaces as a structural entity without any discontinuities in the pattern
2.3 Structure Idealization 15
(a) (b)
Figure 2.5 Idealized truss. (a) Truss. (b) Typical truss member.
of displacement, that is, that the conditions of internal compatibility are also accounted
for. The solution is exact within the confines of linear elastic analysis, the assumption
of pinned j oints, and any other idealizing assumption made in converting the actual
structure to a line diagram.
Although it hardly seems necessary to apply such a formal description as the fore
going to such a simple truss, the point has been to illustrate general concepts. They
will apply to the most complex framed structure.
Framed structures are almost limitless in variety. A few examples of idealized frame
works are shown in Figure 2.6. All can be broken down into line elements, and all are
x
)C
)C
x
x
x
t f )C
)C
x
0
(a) (b)
(c) (d)
Figure 2.6 Typical framed structures. (a) Rigidly jointed plane frame.
(b) Multistory frames-rigidly jointed and trussed. (c) Trussed space frame.
(d) Rigidly jointed space frame.
Figure 2.7 Framework element.
within the purview of this text. In Chapters 2-5 we treat only cases in which the
component elements are prismatic, that is, straight and of uniform cross section from
node to node. Nonprismatic elements are covered in Chapter 7.
The complete framework element has 12 nodal degrees of freedom and 12 nodal
force components, as indicated in Figure 2.7.5 Usually, however, members of the real
structure are joined in ways such that, in the idealization, a number of these force
components and degrees of freedom may be disregarded. For example, for the purpose
of analysis, we may often assume that trusses are pin jointed. In beams, torsional
moments may often be disregarded. Other reasons for neglecting certain effects have
been cited in Section 2.1.
2.4 AXIAL FORCE ELEMENT: FORCE-DISPLACEMENT RELATIONSHIPS
Force-displacement relationships for the complete framework element will be devel

oped in Section 4.5. First, however, the simple axial force element (also called the truss
element) will be used for further demonstration of concepts and definitions.
Force-displacement equations, the relationships between joint forces and joint dis
placements are most commonly written in either of two forms: stiffness equations or
flexibility equations.6
2.4.1 Element Stiffness Equations
Element stiffness equations are linear algebraic equations of the form
{F} = [k]{.1} (2.2)

The matrix [k] is the element stiffness matrix, and {F} and {.1} are element force and
displacement vectors. An individual term of the [k] matrix, k;j, is an element stiffness
coefficient. If a displacement !lj of unit value is imposed and all other degrees of free
dom are held fixed against movement (ilk = 0 for k =I= j) the force F; is equal in value
to k;j The spring constant of elementary mechanics is a stiffness opefficient.
Consider a simple truss member directed along a local axis x antj having a known
5This element is "complete" in the sense that it is capable of accounting for axial force, Jhear in two direc
tions, biaxial bending, and uniform (St. Venant) torsion. See Section 7.4 for discussion of nonuniform torsion
and the additional two degrees of freedom needed to account for warping torsion.
6A description of a third form, mixed force-displacement equations, may be found in the first edition of this
text. Because of its limited use, it has not been included in this edition.
2.4 Axial Force Element: Force-Displacement Relationships 17
,..._
._ ___
L ----+-<
(a)

'---+-....
.. -..
(b)
Figure 2.8 Axial force element
r-- L
(c)
-
-++-- I1-
. member on x axis. (a) Axial force
element. (b) Unit displacement at 1.
(c) Unit displacement at 2.
area A, and Young's modulus E ( Figure 2.8a). Since it is assumed to be pin-ended and
therefore capable of resisting only an axial force, it is a two-degree-of-freedom mem
ber. In Figure 2.8b a unit displacement in degree-of-freedom 1 is imposed with
the other degree of freedom held fixed (d1 = u1 = 1, d2 = u2 = 0). The column of
j oint forces for this case is equal to the column of stiffness coefficients corresponding
to di. or
IFJ =!kill (i = 1, 2)
where
IFJ = LF1 F2Y

\k;iJ = Lk11 k21Y
Clearly, F1 = k11 is the force required to impose a unit value of di. and F2 = k21 is the
reactive force. Thus the column of stiffness coefficients \k;i) represents a system of
forces in equilibrium. The same interpretation can be placed on the second column of
the element stiffness matrix (see Figure 2.8c). In Figures 2.8b and 2.8c the forces F1
and F2 are shown in the directions that correspond to the imposed positive displace
ments. In analysis, unknown forces are generally assumed to act in their positively
defined directions, and stiffness coefficient signs are established accordingly. Thus the
correct direction of any force can be determined from the sign of the analytical result.
To write the stiffness equations for this member, we have, from Figure 2.8b and the
basic mechanics of deformable bodies:
or since we wish an expression for the end forces
and, by equilibrium
In the same way, from Figure 2.8c
EA
F2 -F, Uz
L
= =
Collected in matrix form, these equations are
k k 2 u
{F,} [
F2
=
k2
11 1
k22
]{ }
u2
1 =
EA [ -l J{ }U1
u2
(2.3)
1
L 1 1
Note that the pertinent equilibrium equation is 2-Fx = 0, and that this results in a zero
sum of the terms in each column of the stiffness matrix. The matrix is also symmetrical.
This is a consequence of the necessary reciprocity of displacements, the laws of which
will be discussed in Section 4.3.
A complete set of stiffness equations for an element, such as Equation 2.3, associates
all of the effective node-point forces with all of the degrees of freedom. Mathemati
cally, the stiffness matrix is singular. The reason is that, physically, the degrees of
freedom include rigid-body-motion terms. For example, in defining the first column
in this stiffness matrix, the member was restrained at node 2 by suppressing the dis
placement at that point (Figure 2.8b). If it had not been, any attempt to impose a
displacement at node 1 would have been unresisted. The member could respond as an
undeformed body and translate to the right an unlimited amount under the action of
any nonzero value of the force F Likewise, the displacement at node 1 was suppressed
1
in defining the second column of the stiffness matrix. The suppressed displacements
associated with any such support conditions are sets of displacements associated with
rigid body motion. Such degrees of freedom, when extracted, enable a more concise
description of the element stiffness properties (see Section 4.4).
2.4.2 Element Flexibility Equations
Element flexibility equations express, for elements supported in a stable manner, the
joint displacements, {1}, as a function of the joint forces, {F1}:
(2.4)
where [d] is the element flexibility matrix. An element flexibility coefficient, d;j, is the
value of the displacement !!..; caused by a unit value of the force Fj. The subscript f on
the force and displacement vectors refers to the degrees of freedom that are free to
displ,ace. It is used to emphasize the fact that these vectors exclude components related
to the support conditions, that is, that the constrained degrees of freedom are absent
from Equation 2.4.
Flexibility relationships can be written only for elements supported in a stable man
ner because rigid body motion of undetermined magnitude would otherwise result
from application of applied forces. They can be applied in as many ways as there are
stable and statically determinate support conditions. In contrast to the stiffness matrix
of Equation 2.2, the flexibility matrix is not unique. The difference is that, in compiling
the element stiffness matrix, displacements are specified (symbolically), and therefore
rigid body motion presents no problem of definition.
Flexibility relationships can also be derived for elements supported in a statically
indeterminate manner but they would not be of general use since, with certain excep
tions, they could not be combined with other elements in the representation of a
complex structure (see Section 4.4.1).
2.5 Axial Force Element-Global Stiffness Equations 19
For the axial force member, the flexibility relationships assume the simple forms
associated either with a pinned support at node 2 (Figure 2.8b ) :
{ui} [:AJ{Fi)
=
or a pinned support at node 1 (Figure 2.8c):
{u2} [:AJ{F2}
=
When the complete element framework element is considered in Section 4.5 more
complex flexibility equations will be found.
2.5 AXIAL FORCE ELEMENT-GLOBAL STIFFNESS EQUATIONS
Normally, the simplest way to form the global stiffness equations for the analysis of a
structure is to start with element stiffness equations in convenient local coordi
nates-relationships such as Equation 2.3-and then to transform them to global axes
by methods to be developed in Section 5.1. Mathematical transformations are not
essential, however. Element stiffness equations in arbitrary coordinates-either local
or global-can be developed from basic principles. Consider the axial force member
in Figure 2.9, which is identical to the member in Figure 2.8. It has known properties
and lies in an x-y plane, but in this case it makes an angle <f>--which may be different
from zero-with the x axis. Defining the degrees of freedom as displacement com
ponents parallel to the coordinate axes, there are in this instance four relevant quan
tities (ui. Vi. u2, v2) rather than the two (ui. u2) that are sufficient for the member
oriented as in Figure 2.8. Correspondingly, the resultant force acting on each end.
y
Fy2 Vz F2
Fx2 Uz
Fxlul x
?tyl>Vl (a)
y y
(b) (c)
Figure 2.9 Axial force element-member inclined to axis. (a) Inclined axial
x
element. (b) Small displacement, 11. (c) Small displacement. v.

which must be directed along the bar, may be resolved into its components in the
coordinate directions, as shown in Figure 2.9a. The element stiffness matrix must now
be of order 4 x 4.
To determine the column of stiffness coefficients relating the force components to
u , impose a small displacement in the x direction at node 2 and hold all the other
2
degrees of freedom fixed against displacement (Figure 2.9b ). Making the usual as
sumptions of small displacement theory, the lengthening of the bar,
6.F2, L, 2 2 = u cos</>
6.Lu2, is
The resultant force in the bar, is
F2 "Lu2F2
All of the force components are related to
_
-
EA
L
Ll
_
-
EA cos</>
L
Uz
by equilibrium. Following our assumption

that the components of the force vector remain unchanged as node 2 moves frcm 2 to
2', we have
FF22 2 u2 u2
= cos</>
,1..
sm ..,,
=
EA
L
EA
cos </>
. ,1.. ,1..
sm ..,, cos ..,,

L
= =

Likewise, for a small displacement in they direction at node 2 (Figure 2.9c), it follows
that
EA .
Fx2 Fxl F2
=
-
=
cos</>
.
=
L
EA
sm</>cos .
. 2 ,1..
Vz
Fy2 Fy1 F=
-
=
2 sm</> =
Corresponding displacements can be imposed at node 1 and the results compiled in

L
sm ..,, v2
the element stiffness equations:
{FFr<12} r cos2 </> sin</>cos</> -cos2</> -sin cos</> ]{ }

u1
t;.1
=
EA sin</>c s </> sin2</> -sin</>cos</> -sm2</> v1
L cos </> -sin</>cos</> cos2</> sin</>cos</> u
2
-
Fy2 -sin</>cos</> -sin2</> sin</>cos</> sin2</> V

z
(2.5)
Equation 2.5 is a more general form of the axial member stiffness equation than Equa
tion 2.3. It yields the element stiffness equations in arbitrary coordinates directly. For
</> = 0, Equation 2.5 reduces to Equation 2.3 after the deletion of null vectors, as of
course it should.
2.6 EXAMPLES
The following examples illustrate the application of the concepts discussed in this
chapter. Although not in themselves examples of matrix structural analysis, they con
tain previews of procedures that are part of formal matrix analysis methods.
Example 2.1 shows how member stiffness equations may be used to evaluate dis-
2.6 Examples 21
placements in cases in which internal forces have already been calculated. The require
ment of compatibility of displacements is satisfied through the simultaneous solution
of two equations. Note that superscripts are used for defining unambiguously the el
ement on which the force component acts.
Example 2.2 is an elementary application of the displacement method of analysis
for forces and displacements.
In Example 2.3 stiffness equations are used to determine directly the force needed
to obtain a desired displacement.
Example 2.4 is an elementary application of flexibility equations to the solution of
a thermal loading problem.
Example 2.5 illustrates that structural equations may sometimes be used advanta
geously in the solution of small displacement kinematic problems.
In Example 2.6 some of the many different ways in which element stiffness equations
are combined to form force-displacement relationships are illustrated.
EXAMPLE2.l
A statically determinate truss is subjected to the load and resulting bar a Pxa = 500 kN
200,000 MPa.
l
forces shown. What is the displacement of a? E =
4m
J
k---- 6 m ----+.--- 3 m __..j
Consider ab:
200 x 6 x 101
, = 166.4 kN/mm
VO- T q- X lQ-
,1..
'!'ab = tan-1 ( ) -4
_6
= 213 - 69
From the first part of Equation 2.5 (with u,, = u,, = 0),
F:; = 166.4(cos 2 <!>ab u,, + sin</>,,,, cos</>,,,, u,,)

333.3 = 166.4(0.6923u,, + 0.4615u,,) (a)
Consider ac:
200 x 8 x 101
= 320_0 kN/mm
5 x 10,
<l>ac = tan - 1 ( -) = 306.87
From the first part of Equation 2.5 (with uc = uc = 0).
166_7 = 320_0(0.3600u,, - 0.4800u,,) (b)

Solve Equations a and b simultaneously:
0.6923u. + 0.4615u. = 2.003 (a)

0.3600u. - 0.4800u. = 0.5209 (b) 0.72mm
t
u. = 2.41 mm
Ua = 0.72 mm j
aa' = 2.52 mm /'
[EXA MPLE 2.2

A statically indeterminate truss is subjected to the load shown. What
are the bar forces and the displacement of a? E = 200,000 MPa.
By symmetry, joint a must displace horizontally. Therefore, there is only one unknown
degree of freedom, u. From the first part of Equation 2.5, using the nomenclature
indicated, and denoting q as the typical support point
F =
(EA) . (cos </Jaq) 2u.
L aq
thus, for q = b
a 200 x 3 x 103 (cos 45)2

Fxab = Ua = 70 . 71 Ua kN
3y'2 X 103
pc 200 X 3 X 103 (cos 315)2
xa = ua = 70.71ua kN
3\1'2 X lOJ
pd 200 X 5 X 103 (cos 210)2
xa = Ua =
125 00Ua kN
6 X 103

ae 200 X 5 X 103(cos 150)2

Fxa =
Ua =
125 OO Ua kN
6 X l03

Write the equation of horizontal equilibrium of joint a with u. as an unknown:
P,;, = 125.0 u. F'.: = 70. 71 u.

'LH,, =
1000 - 2(70.71 + 125.0)u,, =
0
a -------i 1000 kN Solving for u,,,
u,, =
2.55 mm
P, = 125.0 u,, P:;, = 70.71 u,,
2.6 Examples 23
Calculate the bar forces using the member stiff

ness equations:
What is the magnitude and direction of the force P at point a

required to displace that point vertically downward 5 mm without
any horizontal displacement?
E
200,000 MPa.
=
All degrees of freedom, except v0, equal zero. v0 -5 mm. From the first and = y
second parts of Equation 2.5, using the nomenclature indicated, with q a typical
support point
F (E:t (sin </>aq COS </>aq ) V0
C
=
/ !/laq
a /
/
paqv
paqa
y
= (EA)
L aq
. (sin2 </>aq )
va
XO
'a x
For ab, </>ab 45: t paq

=
ya
200 x 1000 .
pab (sm 45 cos 45)(-5) -176.8 kN
2\/2 x 103
= =
XO
F = 70.71(sin2 45)(-5) = -176.8 kN
For ac, <l>ac = tan-1(2/-4) = 153.43:
200 x 4000 .
F (sm 153.43 cos 153.43)(-5) 357.8 kN
x 103
= =

F; = 178.89(sin2 153.43)(-5) = -178.9 kN
By equilibrium, the required force is the vector sum of these

components:
P=399.1 kN
a=tan-1 (181.0/35-5.7)
=26.97
EXAMPLE2.4
Two bars ab and be are pinned together as shown. Bar be is cooled A=1200mm2 A=2000mm2
40C. Determine the displacement of b and the force in the bars. C C
Thermal expansion coefficient a = 1.17 x 10-5 mm/mm0C.
E = 200,000 MPa.
Assume the pin at b is disconnected and calculate the gap due to

free thermal contraction of bar be due to a temperature change,
T = 40C:
U = aLbcT
= 1.17 X 10-5 X 3 X 103 X 40 = 1.404 mm
Apply equal tensile forces in each bar to close the gap:
1--) -1 "1b_
Using flexibility relationships for u0b and uh0
--F
lb'

F

1 x 103 3 x 103 ] -
1.404 = F +- ---
[ 200 x 1.2 x 103 200 x 2 ;( 103
" Ube ---

-
"
Solving for F and using the flexibility relationships for u0h and u1"' 1r 0.501mm
F 120.3 kN
b
=
a-....-_
_., ,_-
.... .. ....------------

Uah = 0.501 mm
Uhc = 0.903 mm
2.6 Examples 25
EXAMPLE2.5
A weightless bar that is 5 m long rests against a slope as shown. If the
point a is moved
move up the slope?
10 mm to the right as indicated, how far will point b
LJ lOmm
No forces are required; the problem is one of kinematics. From the first part of Equation 2.5,
with the above notation,
Fxa= :
E (cos2</>
Ua + sin</> cos</> Va - cos2 t/J ub - sin</> cos</> vb)
But
= 30
<P u" = 10 mm Va = 0
Thus
0= : (10 30 -
E cos2 (cos2 300 + sin 30 30)abtv'2]
cos
Solving for ab,

ab= 8.97 mm
EXAMPJ;E'.2.6)
Two straigh't bars of different properties are pinned together as shown. De a b c
velop the force-displacement relationships for three prescribed loading and

support conditions.
First write the element stiffness equations using Equation 2.3: paab pbab
b
Bar ab:
{pabb}=
a
kab [ -11 -lJ{U } --- ---
EA
( 1 1) kab
Fi 1 u: Li
(a) =
{ppgcc}= kbe [ -11 -lJ{ub} b pcbc

1 Uc (b)
Bar be:
EA
c
Case I:
( 2 2) kbe
Lz
=
From (a),
F'// = P = kab U b
From (b),
a
,__ _____ f-----------.
Boundary condition: ua 0
Equilibrium requirement:
=
Eliminating ub, pgb =p F/'c =
P = - -kkabbc P + kbcUc
The two force-displacement relationships are:
p = kabUb
kab" kbc
p = Uc
kab + kbc
Case II:
From (a),
From (b), p
or
The two force-displacement relationships are: Boundary condition: ua 0 =
Equilibrium requirements:
p = kabUb Ftb P, Ffc 0
= =
P =
kabUc
Case ID:
From (a),
p
From (b),
Boundary condition: ua uc = = 0
From equilibrium, Equilibrium requirement
at node b:
P = F/} + Fgc = kabUb + kbcUb
The force-displacement relationship is:
P = (kab + kbc)ub b
2.7 PROBLEMS
In the following problems all of the numerical calculations can be done with a pocket
calculator, but it is suggested that if the reader is not familiar with the use of computer
programs, the results of sample problems be checked with a program such as
MASTAN2.
2.1 In each case shown, the vertical component of reaction at bis 2P/3. Using stiffness
equations for the support elements, calculate the displacement at b. Each supporting
link has the same A, E, and L. Consider the beam itself undeformable, that is, capable
of rigid body motion but not internal straining.
(a) (b) (c)

p
a----'-
(d) (e) (f)

Problem 2.1
2.7 Problems 27
2.2 For the system shown, Eis constant. Calculate (a) the horizontal force at b re
quired to produce a displacement ub = O.OOlL, (b) the displacement at c under this
force, and (c) th reactions at a and d.
yt pin pin
a b c ,...--- d -
1.2A A l.4A x
L
Problem 2.2
O.BL----L
- -------O.SL _ j
2.3 The system and the displacement ub are the same as in Problem 2.2, but the
horizontal force is applied at c. Calculate its magnitude, the displacement at c, and the
reactions.
2.4 The truss shown is the same as in Example 2.1 except that the area of bar ab is
a variable. E= 200,000 MPa. (a) For what value of Aab is v0 = O? Is this a function of
the magnitude of the applied load? (b) Is it possible to obtain zero u0 by varying Aab?
(c) Suppose the applied load were vertical rather than horizontal, would it then be
possible to obtain zero u0 by varying Aab? If so, for what value of Aab? What would
be the corresponding va?
SOOkN
a
1+-----6m----r- 3 nr-----J
Problem 2.4
2.5 The two-bar axial system is the same as in Example 2.6 except that external axial
forces are shown acting at each of the three nodes. Write the stiffness equations for
the assembled system: {P} = [K]{..1}, where {P} = LP0 Pb PcY and {..1}
Lu0 ub ucJT. Compare elements of [K) with values obtained in Example 2.6.
Pa

a E1A1 b E2A2 c
L1
Problem 2.5
__ _..,.___
.. L2 ----
2.6 Using the matrix equation assembled in Problem 2.5 and assuming Pb= 0, elim
inate ub from the equations by algebraic operations and develop condensed matrix
equations of the form {P,} = [K,]{..1,}, where {P,] =
LP0 PcJT and {..1,} =
Lu0 ucY
Use the condensed equations to solve Case 1 of Example 2.6.
2.7 What is the magnitude and direction of the force P required to obtain displace
ment components u0 = 2 mm and v0 = -3 mm? Cross-sectional areas (mm2 x 103)
are indicated on each bar. E = 200,000 MPa.
r- 5m ---+Jm----

Sm
1 x
Problem 2.7
2.8 Using flexibility equations for the supporting links, calculate the displacement at
b for part a of Problem 2.1. If the flexibility equations are used to calculate the dis
placement at b in part b of Problem 2.1, what additional condition must be invoked?
Calculate the displacement at b in part b. As in Problem 2.1 consider the beam itself
undeformable.
2.9 Comparing the flexibility and stiffness approaches to part a of Problem 2.1, at
what tJOints were the requirement of joint equilibrium and joint displacement com
patibility invoked in the respective solutions?
2.10 Compare the two systems shown: a structural system subjected to an applied
load Pand a resistor network carrying a specified current I. Demonstrate that there is
a mathematical analogy between force and current, flexibility and resistance, and dis
placement and voltage. Assume that the structural arrangement is such that the two
parallel bars must elongate equally.
Lz
=dz
E2A2
Rigid I
block
b
a ...________
_ _, a
Problem 2.10
2.11 Compare the structural system shown with a capacitor network subjected to a
battery voltage V. Demonstrate that there is a mathematical analogy between force
2. 7 Problems 29
and charge, stiffness and capacitance, and displacement and voltage. Make the same
assumption regarding the elongation of the parallel bars as in Problem 2.10.
Rigid
block
Problem 2.11
L
2
2.U A steel bar having a cross-sectional area of 200 mm is secured at its ends to an
2
aluminum cylinder having a cross-sectional area of 350 mm The cylinder is heated
50C while the temperature of the bar remains unchanged. Calculate the changes in
lengths and the internal forces in bar and cylinder. Es 200,000 MPa, Ea,
= =
70,000 MPa, as 1.17 x 1 0 s mm/mmC, aa1

= - 2.31 x 10-s mm/mm0C.
=
..._2m---'
rz;;;;;;zJ
Problem 2.U
2.13 The composite member of Problem 2.12 is subjected to an external tension T as

shown; calculate the changes in lengths and the internal forces in bar and cylinder.
T T
Problem 2.13
2.14 Two steel cylinders are clamped together by a steel bolt as shown. The bolt is
loaded to an initial tension of 200 kN by tightening the nut. Plot a diagram of the force
in the bolt versus T, an external load applied to the composite system after bolt tight
ening. Abo1t 2 2
= 500 mm , Acylincter 4800 mm , E
= =200,000 MPa.
T
..,.i-- - - ifitiiEtjf> -- -T...
"-30mm
Problem 2.14
2.15 Use flexibility relationships, geometrical considerations, and equilibrium equa

tions to calculate the displacement at a and the bar forces. E = 200,000 MPa.
2m A= 2000mm2
Problem 2.15
Chapter 3
Formation of the Global Analysis

Equations
In this chapter we discuss the fundamentals of the direct stiffness approach to the
formation of the equations of analysis, equations in which joint displacements play the
role of unknowns. This approach requires only the notion and the algebraic form of
the element stiffness matrix. And it merely involves the application of the conditions
of equilibrium and continuity of displacement at the joints of the analytical model.
The immediate objective of the chapter is to furnish readers with a means of forming
the equations of analysis and to help them see their purpose by illustrating their so
lution and the interpretation of results in simple cases. Only axial force members are
considered. Nevertheless, the text and the examples illustrate the essence of matrix
structural analysis, and the ideas contained in them are part of the foundation on which
the whole system is built.
In extending the scope and power of structural analysis, computer-oriented methods
such as the direct stiffness method require an order in the preparation of a problem
for analysis that was not always essential in the older, classical methods. The same
skill, understanding, and imagination are needed, but it is also necessary to establish
and obey a formal routine in the statement of the problem and in its solution.
3.1 DIRECT STIFFNESS METHOD-THE BASIC EQUATIONS

A complete set of force-displacement relationships for a framework element with'
degrees of freedom is, from Equation 2.2,
(3.1)
lnwriting Equation 3.1 it is assumed that the degrees of freedom refer to the global
axes of the structure. This is accomplished either by writing the equations directly in
global coordinates as in Figure 2.9 and Equation 2.5, or by applying mathematical
31
32 Chapter 3 Formation of the Global Analysis Equations
coordinate transformations to force-displacement relationships that have been written

in local coordinates, as in Equation 2.3.1
The numbers 1, i, n identify the degrees of freedom at the joints of the element, and
in this example they correspond to some convenient global numbering system for these
joints. The global numbering system is completely independent of the local element
numbering system. There are no inherent restrictions on the ways of designating global
joints but, in any problem, there will be preferred systems. The order of degree-of
freedom numbering can have a strong effect on the efficiency of the equation solving
process (see Section 11.4).
When the global analysis equations are formed using the direct stiffness method, all
of the degrees of freedom appear in each of the rows of Equation 3.1: The element
does not have a defined support condition. Thus, as shown in Section 2.4.1, the equa
tions include rigid-body-motion terms.
Once the element force-displacement relationships have been numerically evaluated
_for al! members of the structure, application of the direct stiffness method consists of
their combination in an algebraic form that satisfies the requirements of static equilib
rium and joint compatibility at all of the junction points of the assembled model. To
illustrate the procedure, consider the formation of the force-displacement equations
of the point q in the global x direction of the analytical model of the portion of a
pin-jointed truss shown in Figure 3.1. The quantities in the x direction at this point are
designated by the subscript i. The truss bars shown all lie in the x-y plane. For con
venience, we have identified the degrees of freedom at each joint, but the only force
that has been identified is the externally applied load, P;, that acts in the x direction
at point q.
L'>5
L'ig
L'i4
'L x
Figure 3.1 Representative interior joint of a plane truss.
1We defer formal development of the transformation equations to Section 5.1. For the axial fqrce member
the transformation matrix is rectangular (see Section 5.1.4) and in Example 5.1 it is used to relate the stiffness
matrix of Equation 2.3 to that of Equation 2.5. Even without the benefit of the mathematics that precedes
it, perusal of that example at this point will provide insight into the coordinate transformation process.
(C)
P;

/if
(0) / (A)
Figure 3.2 Study of equilibrium in
the Pr!l; directions.
For junction point equilibrium, ihe applied load must be equal to the sum of the
internal forces acting on the bars meeting at that point.2 To emphasize this operation.
in Figure 3.2 we show the bars separated from the joint. From the condition of
x-direction equilibrium,
P; = F1 + Ff + Ff + Ff (3.2)
where F1 is the global, x-direction internal force component on bar A, and so on. The
force-displacement equations for the elements, each of the form of Equation 3.1, yield
expressions for F1 ... Ff in terms of the corresponding element degrees of freedom
A1 . . . M/. Substitution of such expressions into Equation 3.2 results in:
P; = (ktA1 + k'/iA + k'AAt + kA')

+ (kf;Af + kAf + kfsAf + kt:,Ai)
(3.3)
+ (kfA( + k(zAf + kAf + kgAf)
+ (kf/Af + kf?Af + kgAf + kA!})
Since, by the condition of compatibility of displacement, A; is the same for bars A, B,

C, and D in each degree of freedom, that is, A1 = iif = A( Af A1, it follows that
= =
P; = (kt + kf; + kf + kf/)A; + kf?A1

(3.4)
+ (k'/i + k + k(z + kg)t..2 + k'AA3 + + k{?iA9
or
(3.5)
This is the final form of the desired equations. The capitalized terms K;;, Kn, Kr?._, ... ,
K;9, are global stiffness coefJicients and Equation 3.5 is a global stiffness equation.
Note that each of the bars meeting at the indicated junction points possesses stiffness
coefficients with common subscripts, e.g., kt, kff, kf, kf/. When the subscripts of the
coefficients of two or more different elements are identical, the elements have a degree
of freedom in common, designated by the second subscript, and such coefficients are
added to form one coefficient of the stiffness equation for the force represented by
the first subscript.
Consider now what happens when an additional member, designated as H, frames
into joint q (Figure 3.3). In Equation 3.2 this merely means that a force Ff 1 is added
2A feature of the nomenclature is that internal (or element) forces are identified by the symbol F and external
forces by the symbol P. Appropriate subscripts or superscripts are assigned in each case.
(E)
t.4
t.,
t.9
Figure 3.3 Effect of adding a member to a joint.
(Figure 3.4) and Equation 3.4 is supplemented by stiffness terms k1;!, kt{, kt{0, kt{1 A
force F? would be added in they-direction equilibrium equation at point q, and sim
ilarly for forces Fi and Ff at the opposite end of the member. There are no other
contributions of member H to joint equilibrium equations. Thus it is a simple matter to
revise the stiffness equations to include another member, and the effects are quite local.
Note, however, that after the equations have been solved for the displ?cements as
functions of the applied loads, no simple modifications can be made to the resulting
equations to account precisely for the presence of the new member. When a new
member is added, the precise solution requires starting with stiffness equations that
account for this member. But there are simple procedures for obtaining approximate
solutions when structural changes are made. These will be taken up in Section 13.7.
In summary, joint equilibrium is satisfied through the formation of equations such
as Equation 3.2, and joint displacement compatibility is satisfied by equating degrees
of freedom, as in going from Equations 3.3 to 3.4. Therefore, if the element force
displacement relationships of Equation 3.1 satisfy the proper laws of materials (the
(C)
68) I

q
) i\ plJ pC pD pH
Fi, i, i,, j, i
A)
Figure 3.4 Equilibrium in the P;-tl, direction, member
added.
constitutive relationships) and the requirements of equilibrium within the elements,

and if the proper supports (boundary conditions) are assigned to the assembled struc
ture, then all the requirements for an exact solution-within the limits of linear elastic
3.1-3.3.
theory -are satisfied.
3.1 the global stiffness equations for an unsupported truss are obtained
These concepts are illustrated in Examples
In Example
from element stiffness equations, and it is demonstrated that the resulting stiffness
matrix contains rigid-body-motion terms. Note that the axes used in forming the mem
There is no need for coincidence; Equation 2.5 yields the correct results without further
ber stiffness equations are parallel to, but not coincident with, the overall global axes.
5.1.
transformation, provided that the member and overall global axes are parallel. This
3.2 is an application of the equations of Example 3.1 to an adequately

point will be discussed in Section
Example
supported system. The ways in which various stiffness equations are used to determine
displacements, reactions, and bar-force components are illustrated. Note that multi
pliers of known zero displacements have no effect on the results and thus may be
3.3 demonstrates the simplicity of modifying the stiffness matrix to account

eliminated from consideration.
Example
for structural changes. In the determination of the displacements, the inversion of the
stiffness matrix is indicated symbolically but, for brevity, the actual calculations are
not reproduced.
EXAMPLE3.1
For the system shown: a
-------
1. Write the member force-displacement relationships in global

I
coordinates.
4m
2. Assemble the global stiffness equations.
200,000
3. Show that the global stiffness equations contain rigid-body-motion
= 1
terms. E = MPa. c ____...__________
_ ...___,.
1,..,, A_ s9 8 ' _m _m_2

:: _,f
1
_____ ____
Define the coordinates, degrees of

freedom, and external forces as
follows:

=
_...i.::;.__ b_,.._
=
... _ __
D.,
_ ______
D._1 c x c b p_,
1. Member force-displacement relationships (see Equation 2.5): If, y
200 x 20 x 10-'
Member ab
707.11
4v'2 x 10-'
(EA) kN/mm
a L
-0.500 0.500
=
{F'F2!hh} h l
F",, 707.11
0.500
0.500
-
0.500
osoo
-0.500
lr}
2
pth
-l
F ,,
0.500 - 0 . 500 D.,
=
P{"
Sym. 0.500 .
h
Member be
(EA)
\T
200 x 18 x 10 3 329 43 kN/mm
-i.o o ol{d3}
10.928 x 103
=
be
{FFbcc:}-
5
[
- 329.43
1.000
d4dd:; 0
1.000
0
0
Sym. 6 0
Member ac
( E: : 3
t
200 15 l0
375.00 kN/mm
x
=
1 .
{FF52c:} [ 375.00
0.750 0.433
0.250
1{dd2} -0.750
-0.433
-0.433
-0.250
.:"
x
Fp'(,c =
Sym. d6s 0.750 0.433

0.250
2. Global stiffness equations in matrix form (see Equation 3.5):
Pi
P2
6.348 -1.912
4.473
-3.536
3.536
3.536
-3.536
-2.812
-1.624
-1.624
-0.938
ddd3i2
P4P3
Ps
= 102
Sym.
6.830 -3.536
3.536
-3.294
0
6.107
0
0
1.624
d4dds
p6
3. Rigid body motion. Adding rows 1 and 3 of the global stiffness
0.938 6
matrix yields the vector:
L2.812 1.624 3.294 o -6.107 -1.624J
which is the negative of row 5. Therefore, there is linear depen

dence, the determinant is zero, and the matrix is singular. This is
a signal that, under an arbitrary load, the displacements are in
definite; that is, there may be rigid body motion.
EXAMPLEJ.2
The truss of Example 3.1 is supported and loaded as shown. F0 = 500 kN
1. Calculate the displacements at a and b.

2. Calculate the reactions.
3. Calcuiate the bar forces. Use equations of Example 3.1.
x
t P2 = 321.4 kN
Boundary conditions:
L':.4 =
L':.5 = L':.5 0
=
1. Displacements. The upper three global stiffness equations can be written as follows:
383.0
{ } r6.348 -1.912 -3.536 Li1
]{ } [ 3.536 -2.812 -1.624 0
]{ }
321.4 = l02 l 4.473 3.536 Li
2
+ 102 -3.536 -1.624 -0.938 0
0 Sym. 6.830 D.3 -3.536 -3.294 0 0
Inverting the first matrix and solving for the displacements yields
2. Reactions. The lower three stiffness equations now yield the reactions:
D.1 Li d3
2
{RRxybc} [ 3.536 -3.536 -3.536 ]{ } 0.871
= 102 -2.812 -1.624 -3.294 1.244
Rye -1.624 -0.938 0 -0.193
d4 d5 D.6
[ 3.536 o o ]{o} { } -63.6
+ 102 6.107 1.624 0 = -383.4 kN
Sym. 0.938 0 --258.1
3. Bar forces. The bar forces may now be obtained from the member stiffness equations:
Member ab
Li1 Li Ll3 { } 0.87 1

2 -6'1
{FF7_l:} = 707.ll
[ 0.500 -0.500 -0.500 ] 1.244 =
{ ; : } 6
kN
-0.500 0.500 0.500 -0.193
Fab =
F'!/ V2 = +90.0 kN (tenion)
Member be
Fhc = F' = -63.6 kN (compression)

Member ac
ll, ll2
{Fye} 375.00 [0.0.473350
F'ZC =
0.433 244 {447.
0.250] {0.1.871} 258.00}
= kN
2 .
Fa, = Fye \/3 = +516.2 kN (tension)
321.4kN
0.87 mm
3.0kN
1.24 mm
383.4kN
258.1kN
63.6kN
EXAMPLE3.3
The truss shown is the same as in Example 3.2 except for the ad
6.928 m
Pa= 500 kN
dition of the horizontal tie ad.

y !
Member ad. Member force-displacement relationships at node a (see

Equation 2.5):
F"" 2006. 8 10-' (1.000) .:l 577.37.:l

I
=
x 20 x
92 X lQ-1
I
=
1 kN
<flad = 180
F2" 0 =
" d x
1. Displacements. No nonzero degrees of freedom have been added. The

stiffness coefficient of ad can be added to the stiffness matrix of Ex
ample 3.2, resulting in the following:
{Illl2l} 10-2l( +5.6.377448) - 1 1 -3.536 -1 383.0

. 9 2 1{ }
} 4.473 3.536 321.4
=
Ll, Sym. 6.830 0

{
0.383
= 1.228 mm
-0.437
2. Reactions. Reactions may be obtained by adding an equation for Rxrl,

the horizontal reaction at d, to the equations of Example 3.2:
{RRxybc} [ 2
C.1
3.536
. 1
-2 8 2
ii2
-3.536
-1.624
-
3 3
6 ]{ 0.383 } r - 44.
3
} kN
-r l =!;
10
=
RvRxrlc -1.624 -0.938

.
-5 774
0 -:
321.4 kN
3.2 DIRECT STIFFNESS METHOD-THE GENERAL PROCEDURE
The previous discussion and examples suggest the following automatic approach to
calculating the applied load versus displacement equations for the complete structure.
Parts of its mathematical basis that have not been covered already are best explained
after outlining the procedure:
1. Each element stiffness coefficient, after numerical evaluation, is assigned a double

subscript (k;i). The first subscript, i, designates the force for which the equation
is written, while the second, j, designates the associated degree of freedom.
2. Provision is made for a square matrix whose size is equal to the number of degrees
of freedom in the complete system, with the possibility that each force will be
related to every displacement in the system. Each term in this array is identified
by two subscripts. The first subscript (row) pertains to the force equation, the
second (column) to the degree of freedom in question. The array is illustrated in
Figure 3.5 for a structure with a total of n degrees of freedom. A search is then
made through the member stiffness coefficients. When a coefficient is reached
whose designating first subscript is 1, it is placed in row 1 in the column headed
by its second subscript. For example, k13 is positioned as illustrated in the figure.
3. The procedure of step 2 is continued until all elements have been searched. Each
time a coefficient is placed in a location where a value has already been placed,
it is added to the latter. At the completion of this step all terms in row 1 have
achieved their final value. Hence, for the ith degree of freedom,
where the summation extends over all the members meeting at degree of free
dom i.
4. The process of steps 2 and 3 is r e peat e d for all other rows in order. The result is
a complete set of coefficient:; of the stiffness equations for the entire structure
jth degree of freedom

,
ti
P, -- k13 --------+--+--
ith force P; ------+---- Lk;j
Figure 3.5 Formation of a global stiffness coefficient k;1 and insertion of k13
(the global stiffness equations) but with no recognition of the support conditions
as yet. This process, by which the global equations are obtained from the element
equations, is referred to as assembly.
5. The support conditions are accounted for by noting which displacements are zero
and then removing from the equations the columns of stiffness coefficients mul
tiplying these degrees of freedom.3 This is a useful way of recognizing the fact
shown in Equation 3.1 (and in Example 3.2) that multipliers of zero displacements
have no effect on the force system. The immediate result, however, is to have
more equations than unknowns. The surplus equations are those that pertain to
the external forces at the support points, that is, the reactions. The rows repre
senting these equations are now removed and saved for subsequent evaluation.
Thus it will be found that an equal number of rows and columns have been
extracted from the array, leaving a square, nonsingular matrix.4
6. The set of equations remaining after step 5 is solved for the remaining degrees
of freedom. The internal forces acting on the ends of the elements are determined
by back substitution of the solved degrees of freedom in the element force
displacement equations. These quantities may require a further transformation
from global to local coordinates and finally a transformation into stresses.
The logic of the foregoing process will now be reviewed using a matrix formulation.
It is assumed that the rationale for steps 1-4 is clear from the previous discussions of
Equations 3.1-3.5 and that these steps have been performed, that is, that the global
stiffness equations have been compiled in the form:
!Pl = [K]{.1} (3.6)

Now we assume that the support degrees of freedom !'1s} can be grouped together.
After reordering rows and columns to separate quantities pertaining to the s11p-
'The treatment of nonzero supports that follow a prescribed behavioral rule or are otherwise constrained is
covered in Section 13.3.
4 An alternative to this procedure is to take cognizance of the support conditions at the outset, that is, to
form the element stiffness matrices only with respect to the unsupported degrees of freedom. Steps 2-4 then
lead to the reduced stiffness matrix and step 5 is eliminated. This approach was implicit in Example 3.3, in
which it was found possible to form the necessary stiffness matrix without adding terms associated with the
degrees of freedom at support d.
ports from the remainder, we partition Equation 3.6 in a conformable fashion to

yield:
(3.7)
where all quantities pertinent to the supports are assigned the subscript s and those
relating to the remaining degrees of freedom have the subscript f. (Note the separation
made in Example 3.2 where, in addition to using the above partitioning, the degrees
of freedom were conveniently numbered in advance. Assigning the last sequence of
numbers to the support degrees of freedom made subsequent renumbering unneces
sary. Such a numbering scheme may be advantageous, but it is not essential.)
Expanding Equation 3.7 and noting that {As} = 0, we have
{PJ) =
[Kff]{AJJ (3.8a)
\P,) = [KsJJIAJ} (3.8b)
The general solution to Equation 3.8a is obtained symbolically by
{A.r} =
[Kffr1{PJ) = [D]{PJ) (3.9)
where the matrix [D] is the set of global flexibility (displacement) coefficients.
We have emphasized that the operation of matrix inversion is symbolic. In practice,
the process is normally one of equation solving by a method of the type described in
Chapter 11. Also, the grouping of terms and partitioning indicated in Equation 3.7 was
adopted for the sake of logic and clarity in presentation. In practice, it may be neither
necessary nor convenient.
When {AJ), the vector of displacements at all unsupported nodes, has been found
from Equation 3.9, the support reactions {P,) are found by substitution of the result
in Equation 3.8b, yielding
(3.10)
To obtain the internal force distribution in the ith element, one may multiply the
calculated degrees of freedom for that element, designated by {Ai}, by the element
stiffness matrix [k;], resulting in the numerical evaluation of the joint force components
{Fl Thus
(3.11)
These forces (direct forces and moments) will be vector components in the directions
of the global coordinate axes. Separate operations are required to transform them into:
(1) resultant forces, (2) components oriented with respect to the member axes (normal
force, shear, bending moment, and torque), or (3) unit stresses. These transformations
are often made part of the basic analysis, but occasionally it may be preferable to leave
them as separate tasks for the analyst or designer.
The application of the general procedure is illustrated in Examples 3.4-3.6. The
structures used are elementary pin-jointed trusses, but the examples contain most of
the features found in the analysis of complex systems.
Example 3.4 is similar to Example 3.1 except that the calculations are arranged in a
more formal fashion to follow the procedures outlined in this section. The details of a
basic system tor forming the global stiffness equations should be clear from this
example.
Example 3.5 is similar to Example 3.2 except that the resultant bar-force calculations
now follow a clear matrix formulation. The five displacements were calculated by com
puter, but the results may be verified easily by substitution in the stiffness equations.
In Example 3.6 it is shown that an increase in the number of constraints reduces the
labor of solving for the displacements even though it increases the static redundancy
of the system. Questions of indeterminacy are introduced in Section 3.4
EXAMPLE3.4
For the system shown:
-------d
mm2
A= 10x103
1. Write the force-displacement relationships in global coordinates.

2. Assemble the global stiffness equations.
3. Show that the stiffness equations contain rigid-body-motion terms.
E =
200,000 MPa.
I
a ---'----41
A=10xl03
lsm-
Define the coordinates, degrees of
follows:
1. Member force-displacement relationships (see Equation 2.5):

Member ab
F f. ,
mn
-1.000 x
0
1.000
l
Member cd
1{ 1 =
400
(EA)
L cd
= kN/mm
1'' .
]{fl
.
:: :: :: :
k76 kn k?K !::.7
4 0.
0 00
r l OOO
0
0
-1.000
0
1.000
Frl c x d
F5rl
kx6 k87 kxs t:.K L Sym.

Member ac
f"!.<"
6
ct----+-
LYI C
r5
y
x
Ff'
=
[ 0 . 2 5 0 0 . 4 3
30 .0 0.250 00..47530 621}
0 . 7 5 0 -0
-4 .
. 2
3 5 0
3 -0
-0 .
. 4
7 3
5 0] {6
6,
Sym. 66
Member bd
d
Fif"
=
[0 . 2
30 .0 0.250 5 0 0
0..4
7350 -0
!. .. 0.
. 2
4 5
3 0
Sym.
k- [ "
Member be
(E:t
2 0 5 V3 1 5 1 0 1
1 0
x
x
3x
4 63
. 4=
1 kN/mm
Flc
c
0 0 00 +1.0 0o]
0 F5hc
r1 1n r {q
k34 k35 k
pc
pc
Fie
_ k43
ksJ
k61
k44
ks4
k64
k45
kss
k65
k46
k,6
k66
3 4 6
64
6,
66
. 4=
1 1.0
Sym.
0 66
y
r;:,

F:ibc x
2. Global stiffness equations in matrix form (see Equations

P1
3 . 5 3 .
and 6 )
4.750 12..22950 -440..07500 1.2009 -1-00..27950 -1-20..22590 -000.750 -100.29: 61
P2 62
P:, 6_,
?4
P,
102
=
5.714 40.750 -315...27491644 -1-40..0290 -200.250 64
6,
66
p6
P1
PR Sym.
4.750 21..22950 67
68
3. Rigid body motion. Adding rows 1, 5, and 7 of the global stiffness matrix yields the vector:
L4.ooo o -4.750 -1.299 o o o.750 i.299J
which is the negative of row 3. Therefore, the matrix is singular. The displacements are indefinite-there
may be rigid body motion.
The truss of Example 3.4 is supported and loaded as shown: d
1. Calculate the displacements at b, c, and d.

3. Calculate the bar forces.
Use equations of Example 3.4.
P8 = -282.84 kN
d P7 = 282.84 kN
Boundary conditions: 1 = 2 = 4 = 0
1. Displacements. Remove columns and rows 1, 2, and 4 from the stiffness

equations, leaving
0 4.750 0 0 -0.750 -1.299 r 3

0 4.750 1.299 -4.000 0 5
0 = 10 2 5.714 0 0 l 6
282.84 Sym. 4.750 1.299 1
-282.84 2.250 8
Solving for { } on a computer yields the following results, which may be checked by substitution in the above equations:
L.1.J =
L:i " 6 7 8J =
L -0.407 9.809 -2.232 10.926 -7.801j mm
1, 2, 4)
2. Reactions. The remaining stiffness equations ( rows and are used as follows:
d3
-0.407
p 750 -1.299 0
ds d6 d1
9.809
{ l} {RRyxaa} 10-'[-4.0000 -0.
P2 -1.299 -2.250 0 -2.232
10.926
Ryb 1.299 0 -3.464 -1.299
= =
P4
-7.801
{-282. 9
-772.0} kN
1056.2
=
3. Bar forces. Develop a formula for calculating bar forces from displace
ments: From equilibrium at the
force F12 is
2 1-2, end of a general member the bar
F12 = + F2 = Fx2 cos <f> Fy2 cos ( 90 - <f>)

In matrix form this is
F12 = L cos <f> sin { }

<f>J F
2
Fy2
(a)
The member stiffness equations are, from Equation 3.11,

(b)
Substituting the last two of Equations b in Equation a gives the desired formula:
F12 = L eos <f> sin <f>J [ k31

k41
kJ2
k42
{1 k33
k43
k34
k44 J
:
U2
V2
(c)
<f> o
Member ab = o
Fab 400.0L1 = OJ
lq.. {
d2 0 } d3
[ 0 0 ] -0:407 -162.8
0-1
kN
Member cd <f> o
= o
Fed 400.0L1 0
d,
-1
. .
[ 0 0 00 {"-2.2321 +446.8
d6 d1
OJ
J -7.
10.926 kN
J
= =
801
Member ac <f> = 60
d1 d1 d,
Fae 300.0L0.500 0.866 r-0.

=
250 -0.433 0.250 0.433 0
-0.433 -0. 750 0.433 0.750 J 9.809 +891.4
J kN
.. { }
-
2 232
.
=
46 Chapter 3 Formation of the Global Analysis Equat
{ }
ions
Member bd </> 60
-0.407
=
-0. 2 50
A3
-0.433 0. 2 50 0.433
A4
300.0L0.500 0.866J [-0.433 -0.750 0.433 0.750] 10.0926 -326.8

A1 As
Fbd =
kN
{ ]
-7.801
=
Member be </> 90
-0.407
=
0 0 0
346.41LO lj [ -1 0 1 J 9.0809 -773.2
A3 A4 As A6
Fbc kN
-2.232
= =
!
Th truss shown s the same as in Example 3.5
except for the addition of
honzontal constraints at b and c. Calculate the displacements at c and d.
Remove columns and rows 1 to

leaving
5 from the stiffness equations,
Pa= -282.84 kN
{:} {-282.22.8844}
=
d
P1 = 282.84 kN
102 [ 4.750 ]{::}

P8
5.714 0 Boundary conditions:
2.250
= l. 99 I= 2 = 3 = 4 = 5 =0
Sym. A8
[
Solving for {4.},
0.175
{::} 10-2 o.5o -00.154428]{ -282.22.8844} {-1.1191401}
As
=
Sym.
= mm
3.3 SOME FEATURES OF THE STIFFNESS EQUATIONS

As has been shown, both element and global stiffness matrices are almost always
symmetrical.5 The conditions under which a set of stiffness coefficients for a structure
51dentification of the occurrence and the treatment of nonsymmetrical stiffness matrices are beyond the
scope of this text. See Reference 3.1 for discussion of typical cases.
3.4 Indeterminacy 47
possesses the property of symmetry will be discussed in Section 4.3. Practically, this
means that only the main diagonal terms and terms to one side of the main diagonal
need to be stored in a computer program (see Section 11.4).
Also, the stiffness (equilibrium) equation for a given degree of freedom is influenced
by the degrees of freedom associated with the elements connecting to that degree of
freedom. The members shown in Figure 3.1 or in Example 3.4, could comprise but a
small region of what is actually a very large structure. Bars such as E, F, and G of
Figure 3.1, and any others that might exist beyond these, have no effect on Equation
3.5. In other words, the nonzero terms in a given row of a stiffness matrix consist only
of the main diagonal and the terms corresponding to degrees of freedom at that joint
and at other joints on the elements meeting at that joint. All other terms in the row
are zero. When there are many degrees of freedom in the complete structure, the
stiffness matrix may contain relatively few nonzero terms, in which case it is charac
terized as sparse or weakly populated.
Clearly, in the solution phase of the analysis it is advantageous to cluster all nonzero
terms as close to the main diagonal as possible (see Figure 3.6), thereby isolating the
zero terms and facilitating their removal in the solution process. This can be done by
numbering the degrees of freedom in such a way that the columnar distance of the
term most remote from the main diagonal term in each row is minimized, that is, by
minimizing the bandwidth.
. .
I
' Band width
ofrowi
Figure 3.6 Typical final form of row

of global stiffness matrix.
Bndwidth minimization is but one strategy for achieving efficiency in the equation -
solvng phase. Whatever the approach employed, in large-scale applications it is es-
sential to the economy of the solution process that account be taken of the symmetry
and spaseness o the stiffness matrix. Even the very elementary problem in Example
.
3 5 requued the mves10n of 5 X 5 ?Iatrix. eal structures are often orders of mag-
:
. mtude larger than this. Equat10n-solvmg algorithms will be discussed in Chapter 11.
3.4 INDETERMINACY
l!P to this pint, the conventional concept of static indeterminacy has not been men
tioned. Cogmzance of staic indeterminacy is in fact unnecessary in the direct stiffness
pproach. The structures m Examples 3.2 and 3.5 are statically determinate and those
m Examples 3.3 and 3.6 are indeterminate, yet all were analyzed by identical proce-
dres. The displacment approach rests on a comparable concept that may be called
_
kmemattc_ mdetermmacy. Both can be clarified by comparing definitions of the two.
In the early, classical methods of analysis, a flexibility approach was normally em
ployed. In tt approach, one first determines the number of equations, in addition to
hose of eqmhbnum, that are re9uired for analysis. The number of additional equations
_ _
is the degree of
stattc mdetermmacy, that is, the number of redundant forces. If the
convenional procedure of cutting the structure to reduce it to an equivalent statically
_
detemmate on 1 used, the additional equations are developed by employing the
elas1c charactenstlcs of the system and applying the requirement of restoring the con
_
tmmty that was destroyed when the system was cut. The unknowns in the analysis are
the redundant forces at the cut sections or removed supports.
Kinematic indeterminacy refers to the number of displacements that are required to
define the response of the structure. The degree of kinematic redundancy is equal to
the number of degrees of freedom that must be conceptually constrained to reduce
the system to one in which all joint displacements are zero or have predetermined
values.6 In the displacement approach the system is first reduced to this kinematically
determinate condition by considering all unspecified degrees of freedom equal to zero.
The number of such degrees of freedom is the number of equations that are required
for analysis, that is, the number of rows of the [Kff] submatrix of Equation 3.7. These
equations are developed by employing the elastic characteristics of the system and
applying the requirement of restoring the equilibrium that was impaired when the
conceptual restraints were employed to reduce the system to an equivalent kinemati
cally determinate one. Mathematically, the restoration of equilibrium was expressed
in Equations .3.2-3.5. The unknowns in the analysis are the degrees of freedom.
From the preceding, two observations may be made:
1. The displacement approach appears more automatic since, in it, all of the un
known quantities are first reduced to zero to produce the "determinate" counterpart,
whereas in the flexibility approach selectivity may be involved in designating
"redundants. "7
2. Since the labor of solution is largely a function of the number of unknowns, the
flexibility approach would appear to be advantageous in structures in which the num
ber of force redundants is less than the number of unknown degrees of freedom, and
the displacement approach in all other cases. This view, however, fails to account for
differences in effort required to form the equations to be solved in the respective
approaches.8
The comparative examples in Figure 3.7 indicate that, except in small structures, the
_
number of redundant forces is usually considerably greater than the degree of kme
matic redundancy. In the figure, all relevant degrees of freedom are included. In eval-
in the problem
6Support settlement problems in which estimated displacements of the supports are specified
are cases in which displacements have predetermined values.
and flexibility
7Procedures for the automatic selection of redundant forces and the formulation of the stiffness
methods along parallel lines may be found in the first edition of this book. . . . .
mdetermmate,
8Note that in Example 3.5, the system is statically determinate but five times kmemat1cally
three times kmemat1cally
whereas in Example 3.6 a similar system is two times statically indeterminate and
in Example 3.6 it would
indeterminate. A flexibility approach to Example 3.5 would be very simple, whereas _ _
be relatively time-consuming. Conversely, the displacement solution of Ex mple
3.6 is simpler t_han it is m
.
each approach, m comput
Example 3.5. But, regardless of the differences in number of unknowns w1thm
except for the determmat10n of
erized frame analysis the displacement method is at least as efficacious,
forces in statically determinate systems.
3.5 Problems 49
Degrees of Indeterminacy
Freedom Static Kinematic
Structure Forces
(a)
-
\ i&h
\
m
0 2
(b)
\
; - t-rt 2- m 3
\ \ 3 0
(c) \ {
t
(d)
h tlr-f1 3 3
\ \
63 35
(e)
y;;
Figure 3.7 Static and kinematic indeterminacy.
uating kinematic indeterminacy, we make the common assumption that axial defor
mation may be neglected in analyzing flexurally loaded frames. As an example of the
consequence of this, one need consider only one translational degree of freedom at
each floor of the multi-story frame of Figure 3.7e.9
3.5 PROBLEMS
It is suggested that, in several of the following problems, particularly 3.6 and 3.15. a
computer program such as MASTAN2 be used. In all cases, the computed results
should be verified by manually checking key equilibrium and/or compatibility
conditions.
9ln tall buildings, the axial deformation of columns should be taken into account. This would add additional
degrees of freedom to the analysis.
3.1 Com ute the displacement , react

:
ions, and bar forces for the trusse
s shown.
Cross-sect10nal areas (mm 2 x 10 ) are
shown on each bar. E 200,000 MPa.
=
,.;;;.+--7
.;;.. --+
100kN
a
b i200kN
J-3m-J
(a ) (b) (c )
b.r---c--. 30 500kN
--.1
6m
J
l-6m-J
(d) (e) (f)
Problem 3.1
3.2 The areas of bars ab and ac in mm2 x 103 are as indicated. A varies from 0 to 40
X 103 mm2 (a) Plot the variation in the force in each bar versus A. (b) If the allowable
stress is 140 MPa, for approximately what value of A does the system have the highest
ratio of load-carrying capacity to weight? Assume the same material is used for all
three bars.
Problem 3.2
3.3 Repeat Problem 3.2 but, in addition to varying A, vary the angle of inclination
of bar ad from 30 to 60 by 10 increments. What are the approximate values of A and
3.5 Problems 51
g capacity to
the angle of inclination of ad that yield the highest ratio of load-carryin
weight?
the truss shown. E and A of all
3.4 (a) Assemble the global stiffness equations for
will be no force applied at e, eliminate the
bars are the same. (b) Assuming there
the global equatio ns by proper algebraic op
corresponding degrees of freedom from
equatio n contain ing only the forces and
erations and develop a condensed matrix
te the reaction s and bar forces for a 200-kN
degrees of freedom at c and d. (c) Calcula
horizontal force acting to the right at d.
Problem 3.4
3.5 (a) Assemble the glbal stiffness equations for the truss in part a of the figure.
.
b) Modify the abov to 1.nclude bars db and bf as in part b.
.
(c) Modify the above to
include bars gh and lJ as m part c. (d) Discuss the manner in which the changes may
or may not affect the labor of solution. E and A of all bars are the same.
(a) (b) (c)

Problem 3.5
3.6 Use a computer program to calculate the displacements, reactions, and bar forces
of the truss shown. A = 40 x 103 mm2 for all chord members and 25 x 103 mm2 for
all web members. E = 200,000 MPa.
100 kN 100 100
Sm r
T------+--'- 16 m
7--!L-__
__J!__i J
3.7 Asemble the [Kf!l matrices for the

.
two trusses shown. Tst to see whether each
one 1s smgular or not. What is the significance of your
results? A and E are the same
for all bars.
3 m
-t
3 m
(a) (b)
Problem 3.7
3.8 The symmetrical truss shown is made of three rods whose compressive resistance
may be considered negligible. In its externally unloaded condition (part a of the figure)
it is pretensioned so that the force in ab is 50 kN. E 200,000 MPa. (a) Calculate the
=
pretensioning force in ac and ad by statics. (b) Plot the vertical displacement at a versus
the applied load P. Assume linar elastic behavior.
5 m A = 0.3 x 103 mm2
Lb b
(a) (b)
Problem 3.8
3.5 Problems 53
it is now supported as shown.

3.9 The truss shown is the same as in Example 3.4 but
10 mm. (a) Are any forces develod
Assume that the support at bsettles vertically
of joint c and d. (c) Sketch the ms-
in the members? (b) Calculate the displacements
placed structure.
-sm-1
Problem 3.9
3.10 A and E are the same for all bars. (a) Assemble the [Kff] matrix. (b) Sketch the
displaced structure under the loads P. (c) Use considerations of symmetry to reduce
p
the order of the [Kff] matrix. (d) Calculate the displacements at band c.
r
7m
I
_i__ a
I....- Sm --6 m ------ sm -

Problem 3.10
3.11 Consider the structure of Problem 3.10, carrying-instead of the loads P-two
horizontal loads H acting to the right at b and c. Sketch the displaced structure, use
considerations of antisymmetry to reduce the order of [Kff], and calculate the displace
ment at band c.
3.U Three bars of equal A and E are arranged as shown. (a) Assemble the [Kff]
matrix and test for singularity. (b) Can a useful solution be obtained when Pb Pc?=
Demonstrate. (c) What happens when Pc > Ph?
t
I
3m
b c
,\ 4m
Problem 3.U
+Ph. 4m
+pc 4m
3.1 _:
Com ute t e displacements, reactions, and bar forces for the system shown.
A - 2 X 10 mm and E = 200,000 MPa for each member.
I
2m
f
2m
_J
Problem 3.13
3.14 If the system shown were modeled as an analogous capacitor network, what
would be the relative values of the required capacitances (see Problem 2.11)? Consider
vertical displacement only. E = 200,000 MPa.
A = 2 x HY mm2 3m
Problem 3.14
L
3.15 Use a computer program to calculate the displacements, reactions, and bar
3.5 Problems 55
forces of the trusses shown. A = 40 x 103 mm2 for all chord members and 25 x 103
mm2 for all web members. E = 200,000 MPa.
(a) (b)
(c)
(d)
i
I.
SO
8@ 7m
------'
(e)
Problem 3.15
REFERENCE
3.1 K.-J. Bathe, Finite Element Procedures, Prentice Hall, Englewood Cliffs, N.J., 1996.
Chapter 4
Stiffness Analysis of Frames-I
In this and the following chapter we develop the remaining tools needed for the linear
elastic stiffness analysis of complete frames. Here we are using the term frames in a
generic sense to include structures such as pin-jointed planar and space trusses,
rigid-jointed planar and space frames, and combinations of these forms; in short, frame
works of any shape. jointing scheme, and loading. By the end of Chapter 5 we can
demonstrate the solution of complex, realistic problems. Nevertheless, there will still
remain much more to be covered in later chapters-in fundamental areas of structural
mechanics and behavior. in nonlinear analysis, and in methods of application such as
the techniques of equation solution.
Our main purpose in the present chapter is the development and application of the
stiffness matrix of the prismatic, bisymmetrical, 12-degree-of-freedom member (see
Figure 2.7). This will be done in local coordinates. Formation of this matrix requires
an understanding of the material stress-strain relationships, and it is facilitated by the
use of energy concepts and the reciprocal theorem. These are reviewed to the extent
useful to the development of the stiffness matrix. The subjects of coordinate transfor
mation and the effects of loads applied between nodal points, self straining, and tem
perature change will be introduced in Chapter 5.
4.1 STRESS-STRAIN RELATIONSHIPS
Theories of structural behavior must incorporate Jaws relating stresses to strains. Ap

plication of any theory therefore requires knowledge of physical properties of the
material used. As mentioned in Chapter 2, in these early chapters we are limiting our
attention to linear, elastic structures. We also assume that the material used is ho
mogeneous and isotropic. It follows from these restrictions that there arc three prop
erties of intc. est, only two of which are independent. The three properties are:
E Young's modulus. the ratio of normal stress to the corresponding strain in a

uniaxially stressed element.
G The modulus of rigidity (shear modulus), the ratio of shearing stress to shear-
ing strain.
v Poisson's ratio. the ratio of transverse strain to axial strain in a uniaxially
stressed element.
Figure 4.1 shows a small element of unit length. width, and thickness stressed in two
different ways: uniaxial tension of intensity a (Figure 4.la) and pure shear of intensity
T (Figure 4.lb). For convenience the unit thickness is shown to reduced scale. From
56
4.1 Stress-Strain Relationships 57
tttttttt
(a) (b)
Figure 4.1 Basic deformations.
the figures and the above definitions, we have the following equations relating strain
to stress:
(J
e=- (4.la)
E
IJ(J
e, = -ve= (4.lb)
E
T
y= - (4.lc)
G
The minus sign of e1 signifies that the material contracts.

By equilibrium, the state of pure shear in the element of Figure 4.lb is equal to a
state of equal tensile and compressive stresses in the diagonal directions of the figure.
Analysis of the small elastic strains in the element leads to the following relationship
between the modulus of rigidity and the other two properties (see Section 6 of Ref
erence 4.1):
E
G= --- (4.2)
2(1 + v)
Equation 4.2 is the most commonly used form of the dependency relationship, but any
of the three properties of the material could be made the dependent variable if desired.
The most widely used civil engineering structural materials-, steel and concrete, have
uniaxial stress-strain diagrams of the types shown in Figure 4.2. Mild steel yields with
a pronounced permanent elongation at a stress <Tym whereas high-strength steel yields
gradually, requiring an arbitrary definition of its yield stress, <Tyh such as the commonly
used 0.2% offset criterion (Figure 4.2a). Yield stresses of structural steel used in gen
eral practice vary from less than 250 MPa (36 ksi) to more than 700 MPa (100 ksi),
with 345 MPa (50 ksi) yield a commonly specified value. At stresses below the yield,
steel stressed either in tension or compression comes close to satisfying the criterion
of ideal elasticity. The Young's modulus and Poisson's ratio of steel are always close
to 200,000 MPa (29,000 ksi) and 0.3, respectively.
The ultimate compressive strength, u" = f ;, of normal weight concrete is usually in
the 21 to 55 MPa (3 to 8 ksi) range (Figure 4.2b ). The most common ranges for cast
in-place buildings have been from 21 to 41 MPa (3 to 6 ksi), but with a trend toward
58 Chapter 4 Stiffness Analysis of Frames-I
High strength
steel
e e
(a) (b)
Figure 4.2 Typical <r-e diagrams. (a ) Steel. (b) Concrete (compression) .
using higher strength values, for example 69 MPa (10 ksi) in the vertical components
of high-rise buildings. The mechanical properties of concrete are less predictable than
those of steel, but under short-duration compressive stress not greater than f /2, its
behavior is reasonably linear. For normal weight concrete, the values Ee = 4700
MPa (1800 ksi, with expressed in ksi) and v = 0.15 are often used in the
analysis of concrete structures. Concrete creeps-that is, its strain increases-at a
diminishing rate, under constant compressive stress. Steel doesn't creep at ordinary
temperatures. Both materials fall short of the ideals of homogeneity and isotropy,
concrete more so than steel. In most analyses these shortcomings and the complications
they cause are ignored. But there are many practical problems in which doing this can
compromise the validity of the results.
4.2 WORK AND ENERGY

The work, dW, of a force F acting through a change in displacement dt::.. in the direction
of that force is the product F dt::... Over a total displacement t::..1 the total work W is
f' F dt::.. (Figure 4 . 3 a).
In this text we are concerned with static behavior, that is, with forces that are applied
to a structure in a gradual manner (inertia forces due to dynamic behavior are assumed
(a) (b) (c)
:. :
.
* .
6.1 6.
e e
1. Work and 2. Strain energy and

complementary work complementary
strain energy densities
(d)
Figure 4.3 Force-displacement relationship.
4.2 Work and Energy 59
negligible). Also, as long as we limit our attention to linear behavior, displacements

and forces are proportional by definition. Thus a single force varies linearly with dis
placement from zero to its final intensity F1 This relationship can be written as F =
kt::.,. where k is a constant (Figure 4.3b). In this case, W= kf' t::.. dt::.. = k (t::.. 1 ) 2/ 2. Also,
letting k = F1l!::..1
W= !F1!::.1.
Both the force and the displacement at a point are representable as vectors F and
A and, in the symbolism of vector algebra, the work of the force is then represented
by the dot product,
W= !F A
We deal with matrix symbolism, however, where the dot product has the form
W= !LFJ{A} (4.3)
In the illustration of Figure 4.3c, where Py = 0, this product is given by
W= !LFx oj {} = !Fxu
The above considerations, which are couched in terms of a force acting on one
element, are readily generalized to the case of a system of forces acting on a structure
and the displacements of the points of application of these forces. In such cases the
force and displacement vectors are of the form:
LP J = LPxl Pyl Pzl Px2 ''' PznJ
{A} = Lui V1 W1 U2 w nJ T
and Equation 4.3, with LFJ= LP J, gives the work of the system of forces LP J.
Equation 4.3 can be transformed into expressions exclusively in terms of either the
forces or the displacements by substitution of the stiffness (Equation 2.2) or the flex
ibilities (Equation 2.4). Therefore, from Equations 4.3 and 2.2,
w= !LAJ[k]{A} = u (4.4a)
and, Equations 4.3 and 2.4,
(4.4b)
in which {F1} is the vector of forces not related to the support conditions.
In Equations 4.4, U describes the strain energy of deformation corresponding to the
work W, W* the complementary work done by the forces acting on the element, and
U* the complementary strain energy of deformation corresponding to W*. From the
development of the equations it is obvious that, for linear elastic behavior, W*= W
and thus U*= U, provided there is no initial strain (see Sections 5.3 and 7.5). But
Figure 4.3d illustrates why we have defined the terms of Equation 4.4b as we have just
done. When the response to a force is nonlinear, W* is unequal to W, as indicated in
Figure 4.3dl. Correspondingly, the strain energy and complementary strain energy
densities (the values per unit volume, U and U* in Figure 4.3d2) and thus their inte
grated resultants, U and U*, are also unequal. Complementary energy has no direct
physical meaning but, as shall be shown in Section 6.5, it is a concept that is the key
to the development and understanding of the principle of virtual forces, one of the
foundations of the energy methods of structural analysis.
Equations 4.4a and 4.4b are of quadratic form, that is, their expansion yields U
and U* as homogeneous, quadratic polynomials in the parameters LAJ and LF1j. These
equations were developed in terms of work done on an element. It is clear, however,
that they apply to entire systems as well, once appropriate substitutions are made for
the definitions of forces, displacements, stiffnesses, and flexibilities, as in writing the
system force and displacement vectors earlier.
Example 4.1 illustrates the necessary equality of work and energy calculated in al
ternative ways.
EXAMPLE4.I
For the truss of Example 3.2 demonstrate the equivalence
of:
1. Work calculated by Equation 4.3.

2. Work calculated by Equation 4.4a.
3. Strain energy calculated using a basic
definition.
Use results of Examples 3.1 and 3.2.
1. Equation 4.3, redefining terms with respect to the whole structure, is
w = !LPJ IAl = l 383.4 321.4 OJ

{ } 0.871
1.244 = 367.0 J
-0.193
2. Equation 4.4a, redefining terms with respect to the whole structure, is
W =
!LAJ[Kff]IA)
10
2
L0.871 1.244 -o.193 J
[ 6.348 -1.912
4.473
-3.536
3.536
]{ } 0.871
1.244 367.0 J
T
= =
Sym. 6.830 -0.193
3. Strain energy. Designating a bar force by the symbol F and summing over all members, the
j
strain energy of this pin- ointed truss is
u
=
L
F2 L
2EA
=
2
[
( 90.0 ) 2
707.11
+ ( -63.6) 2
329.43
+ ( 516.2 ) 2
375.00
] = 367.0 J
4.3 RECIPROCITY
As has been noted, flexibility and stiffness coefficients for linear elastic behavior have
the property of reciprocity (d;i = di; and k;i = ki;). This property is important from the
standpoint of computational efficiency and it is useful in checking formulated or com
puted coefficients. To prove reciprocity and thereby define its scope and limitations,
consider the work done on the supported structure in Figure 4.4 as load F, is first
applied, followed by load F21 We designate the total amount of work as W,. From
Figure 4.4a, for gradual application of F,,
(4.5)
1 Even though we view the system of Figure 4.4 as a complete structure. we retain our single clement notation.
Similarly, in the discussion to follow we retain the element notation in the reference to Equation 3.7, which
was written for complete systems. Justification for this is given in the next section.
4.3 Reciprocity 61
(a) (b)
Figure 4.4 Sequential application of loads. (a) Force F1 applied at 1. (b) Force F2
applied at 2.
where the subscript on W1 and on (d1) means du e to.force 1. Now applying F2 with Fi
held constant, and using similar designations, the increment of work is (Figure 4.4b)
W12 = !(d2)zF2 + (di)zFi (4.6)

= !(d22F2)F2 + (d12F2)F1
The factor 1/2 is not applied to the second term of Equation 4.6 because F1 exists, in
full force, all during the time F2 is gradually applied. The total work, W1 is
(4.7)
Reversing the sequence of application of forces and recomputing the components
of work, we have for initial application of F2 (the work in this second sequence is
identified as W11)
2
W112 = !{d2)zF2 = !d22(F2) (4.5a)
followed by the application of Fi:

2 (4.6a)
W111 = !(d1)1F1 + (d2)iF2 = !di1(Fi) + dz1FiF2
so that
(4.7a)
Since, for a linear system, the sequence of application of loads is immaterial from
the standpoint of the work performed, we can equate the two expressions for Wand,
after canceling terms,
And, generally,
(4.8)
This algebraic statement is known as Maxwell's reciprocal theorem.

As shall be demonstrated formally in the next section, the stiffness matrix for a given
set of force conditions (the kff matrix in Equation 3.7) is the inverse of the flexibility
matrix of the system when supported in the same way. Since it has just been demon
strated that the flexibility matrix is symmetric and since it is known that the inverse of
a symmetric matrix is symmetric, it follows that the stiffness matrix is symmetric, or
(4.9)
Maxwell's reciprocal theorem is usually defined as a special case of Betti's law, which
states that the work done by a system of forces {Pi) acting through the displacements
{Ai)2 that are caused by the system of forces {P2}, is equal to the work done by the
forces {P2l, acting through displacements {A2)i that correspond to {Pi}, that is,
(4.10)
4.4 FLEXIBILITY-STIFFNESS TRANSFORMATIONS
Given one form of element force-displacement relationship, it is possible to obtain the

alternate forms through simple operations. These transformations have many uses.
Since they rest only on general mathematical operations and the principles of equilib
rium and energy, they apply not only to simple elements but also to complete systems,
the members of which undergo compatible deformation. They are developed here
because they are particularly useful in forming the framework element stiffness matrix.
4.4.1 Stiffness-to-Flexibility Transformation
To illustrate the transformation of stiffness to flexibility, we use a three-bar system of

the type analyzed in Examples 3.1 and 3.2, but for simplicity, we treat it as a single
element by using the notation of Chapter 2 (see Figure 4.5). This is permissible for the
reason given in the preceding paragraph. The bar assemblage may be thought of as
one component of a larger structure. To comply with the requirement that an element
must be supported in a stable, statically determinate manner in order to obtain a useful
flexibility relationship (see Section 2.4.2), we support it as in Figure 4.5b.
Quantities related to the supports will be assigned the subscripts and those related
to the remaining degrees of freedom the subscript f. Thus the element stiffness matrix,
Equation 2.2, may be partitioned as follows (compare with Equations 3.6 and 3.7):
(4.11)
where, for the condition shown in Figure 4.Sb, each of the submatrices ([kff], etc.) is
a 3 x 3 matrix and
{Fil = LFxl {Fsl =LFy2 Fx3 Fy3 JT
(4.12)
{Ail =Lui {A,l =Lv2 u3 u3Y
Y"LJ:
2
x,u
(a) (b)
Fire 4.5 Three-bar element. (a) Unsupported element. (b) Supported element.
4.4 Flexibility-Stiffness Transformation!' 63
Now, since /as) = /OJ from the nature of the supports,
(4.13)
The equations above the partition line are an independent set of equations relating
the external forces /F1) to the corresponding node point displacements (compare Ex
ample 3.2). Solution of these by inversion of [kff] yield
(4.14)
where
(4.15)
The matrix [d] is the desired set of element flexibility coefficients. Thus the flexibilities
are derived from the stiffnesses by merely defining a stable, statically determinate
support system, removing from the stiffness matrix the rows and columns correspond
ing to the support components, and inverting the remainder.
Of the two conditions on the supports, it was pointed out in Section 2.4.2 that the
first-stability-is require,d to prevent unrestricted rigid body motion under the ap
plication of applied forces. Regarding the second-determinacy--it can be shown that
when an element is supported in a statically determinate manner, the complementary
strain energy has a value that is independent of the specific set of support conditions
that has been chosen (Example 4.6 to follow). If a statically indeterminate support
condition is used, then strains that are dependent upon the chosen supports are intro
duced into the element. The complementary strain energy is then different for each
support condition. Since the global analysis equations can be obtained through a sum
mation of the element strain energies these must, in general, be independent of local
element support conditions.
4.4.2 Flexibility-to-Stiffness Transformation
To reverse the above process, that is, to develop the complete stiffness matrix from a
given flexibility matrix, we begin by inverting the flexibility matrix. But then we have
to add something since, obviously, inversion of the flexibility matrix yields no direct
information on stiffness coefficients related to degrees of freedom that have been
equated to zero in defining the support system.
We start by inverting a given flexibility matrix:
(4.16)
or
(4.17)
Now, it will be recalled that each column of a complete stiffness matrix represents
a system of forces in equilibrium (Section 2.4.1). Also, the forces in the "s" portion of
each column are the statically determinate support forces. Hence we can use the equa
tions of equilibrium to construct equations of the form:
IF{) = [<l>]{Fj) ( 4.18)

where the superscript j identifies the column in question of the stiffness matrix and []
is an equilibrium matrix, or matrix of coefficients of the equations of static equilibrium.
Construction of equilibrium matrices is a simple process which will be illustrated in
Example 4.2 and other, later, examples. It follows that any set of forces {F1}, stemming
from one or a combination of columns of the element stiffness matrix, are related to
the forces {Fs} in the lower portion of the matrix by a relationship of the above form,
that is
(4.19)
By substitution of Equation 4.16 in 4.19,
(4.20)
or, from Equation 4.13,
(4.21)
in which
(4.22)
In Equations 4.17 and
4.22 we have the necessary relationships for all terms in the
columns that premultiply {A1} in Equation 4.11.
Next we seek the submatrices that premultiply {As} in Equation 4.11; that is, [k1s]
and [kss] To establish [k1,], we simply invoke the reciprocal theorem (Equation 4.9),
which in matrix form may be written as
(4.23)
(Note that the transpose of a product oJ matrices equals the reversed product of the
transposes of the respective matrices. Also, since [dr 1 is symmetric, its transpose is
itself).
In forming
[kss], the lower portion of the columns that premultiply {As}, we recall
from Equations 4.18 and 4.19 that such terms are simply given by the premultiplication
of the corresponding top portion of the stiffness matrix by the equilibrium matrix [].
Hence
(4.24)
Thus, from Equations 4.11, 4.17, 4.22, 4.23, and 4.24, the constructed stiffness matrix
assumes the form:
(4.25)
In summary it is seen that the stiffness matrix is constructed from the inverse of the
flexibility matrix [d) and a matrix that derives from the element static equilibrium
relationships-the equilibrium matrix [). The property of symmetry was invoked in
constructing [k1s] from [ks1]. The [kss] portion of the result is given by a matrix triple
product in which the premultiplier of the central matrix is equal to the transpose of
the postmultiplying matrix. This triple product, known as a congruent transformation,
produces a symmetric matrix when the central matrix is symmetric; thus [kss] is assured
to be symmetric. Equation 4.25 is a general formula for transformation from flexibility
to a stiffness form that includes rigid-body-motion degrees of freedom. The number,
s, of support forces is dictated by the requirements of stable, statically determinate
support, but there is no limit on the number, f, of external forces, that is, the order of
the flexibility matrix.
The use of Equation 4.25 is illustrated in Example 4.2. Included as part of the ex
ample is an illustration of obtaining flexibility matrices from stiffness coefficients.
4.4 Flexibility-Stiffness Transformations 65
EXAMPLE4.2
Using Equation 4.25, develop the stiffness matrix for the axial force
member in arbitrary coordinates (Equation 2.5). E, A, and L are
constant.
'
Support member as shown and write equations for [) and [d): 2
I F1I =
1Fx2l {Fs} = LFxl Fyl Fy2Y
{Iii} = {u2} {lis} = Lui Vi v2JT
Thus
or
{} [ ]
Fx1
Fy1 =
-1
-tan</> IFxzl
Fy2 tan</>
Applying Fx2, the only displacement is u From Figure 2.9 and accom
2
panying equations,
EA
Fx2 cos2 "'
'I' u
2
= -
Thus
[d) = [ sec2</> J
Apply Equations 4.17, 4.23, 4.24:
EA
[kff] = [dr1 = - [cos2</>] (4.17)
L
[k1sl = [kstlT = [dr1[<1>]T

EA
(4.23)
-L -cos2</> -sin</>cos</> sin</>cos</>J
[ ]
=
[kss] =
[<l>)[dr1[<1>]T
cos2</> sin</>cos</> -sin</>cos</>
= :
E
sin</>cos</> sin2</> -sin2</> (4.24)
-sin</>cos</> -sin2</> sin2</>
Collecting terms as in Equation 4.11 yields
which is identical to Equation 2.5 after reordering rows and columns.

4.5 THE FRAMEWORK ELEMENT STIFFNESS MATRIX
We now consider a 12-degree-of-freedom element that is straight, prismatic, and sym

metrical about both principal cross-sectional axes (bisymmetrical). For such a section,
the shear center coincides with the centroid. We shall consider displacements resulting
from uniform axial deformation, flexural deformation, and twisting deformation. We
will neglect displacements resulting fro:n transverse shearing deformations and those
resulting from the out-of-plane (longitudinal) warping of a cross section that torsional
forces may cause. These geometrical limitations and disregarded deformations limit
the applicability of the resulting equations to some extent, but they are sufficient for
the analysis of the majority of the frameworks encountered in structural engineering
practice. Cases in which consideration of transverse shear and torsional warping may
be necessary or desirable are discussed briefly in Section 4.6. Methods for including
them in the analysis are presented in Sections 7.6 and 7.4, respectively.
The type of member studied, its orientation with respect to a local coordinate system,
and the nomenclature used is illustrated in Figure 4.6. A wide flanged member is
shown, but the analysis applies to any bisymmetrical section. The local x axis coincides
with the centroidal axis of the element. For illustrative purposes the local y axis i:>
assumed to define the minor principal axis of a cross section and the local z axis its
major principal axis. Positive signs are in the directions indicated. The right-hand rule
is used for moments and rotational displacements. The restrictions described earlier
leave us with an element that is sufficient for most practical purposes and, moreover,
they greatly simplify the development of the element stiffness matrix in that they yield
an analytical problem in which a number of effects are uncoupled. By this we mean a
situation in which a particular force vector causes a displacement only in the same
vector direction. For example, the axial forces Fxi and Fx2 in Figure 4.6 can only cause
axial deformation, and hence only the axial displacements u, and u2 in the element.
Also, the torques Mx, and Mx2 only cause torsional deformations and hence only twist
ing displacements Bx, and Bx2. This means that the stiffness coefficient relating any of
these force components to a degree of freedom to which it is not coupled must be zero.
From this, plus the principle of reciprocity, it follows that bending moments and trans
verse shearing forces are coupled neither to axial nor to torsional effects in a single
element. Furthermore, since the section is bisymmetrical, it is known from elementary
strength of materials that bending moments and shearing forces in one principal plane
cause no deformation outside of this plane; therefore, they are not coupled to rota
tional or translational displacement in the other principal plane. Consequently, our
problem devolves into the consideration of four separates cases, with the assembly of
t Myl By1 t My2 By2
t --t:

Fylvl v2v2
J
M Fx l------------- '---; '
h
I
/ F,,, w, Young's modulus= E / F,z, wz
Shear modulus= G
/'/ M zl 8 zl M z2 8 z2
z
L--------1
Figure 4.6 Bisymmetrical framework element.
the results in a 12 x 12 stiffness matrix having zero coefficients for all of the uncoupled
forces and displacements.
The four cases are:
1. An axial force member.

2. A pure torsional member.
3. A beam bent about one principal axis.
4. A beam bent about the other principal axis.
4.5.1 Axial Force Member
The stiffness matrix for the axial force member in local coordinates was developed in
Section 2.4.1 (Equation 2.3). It may also be written using Equation 2.5 or the results
of Example 4.2. Nevertheless, it will be redone here as a further illustration of the
flexibility-stiffness transformations of Section 4.4. Consider the member in Figure 4.7
in which the nomenclature of Figure 4.6 has been retained and a stable, statically
determinate support system has been prescribed. The displacement at point 2 under
the applied force {F1) = Fx2 is
(01. ( f 1.
EA EA
dx Fx2 dx Fx2L
u
e
U2
Jc dx
)o
= = = =
E u
Hence, (d] LIEA or, from Equation 4.17,
(kJJJ EA
=
T
=
By equilibrium, the reactive force = -

{F,) Fxi Fx2. And, from Equation 4.18,
[] =
=
-1
Thus, from Equation 4.25, in which the equilibrium relationships are invoked
[k]=E:[- --] (4.26a)
and
{:J =E: [- -J{J (4.26b)
4.5.2 Pure Torsional Member
Mathematically, the simple torsional member (a shaft) is identical to the axial force
member since, by comparison of Figures 4.7 and 4.8, we see that in the two casl':s the
dx
F F,2
4 _____.
I. L fo fo
(a) (b)
Figure 4.7 Axial force element. (a) Support system. (b) Free-body diagram.
dx
I -
...- - L _ _____,
(a) (b)
Figure 4.8 Pure torsion element. (a) Support system. (b) Free-body diagram.
forces and displacements can be represented by sets of vectors of the same type. The
physical difference is that whereas the axial member stretches or shortens uniformly,
the shaft twists uniformly. From elementary strength of materials we know that, for a
shaft subjected to a pure torque Mx2, the rate of twist, {3, that is, the change in rotation
of the cross section about the longitudinal axis per unit of length along that axis, can
be expressed as
= Mx2
/3 GJ
where Mx2 is the torque at the section considered, G is the modulus of rigidity, and J
is the torsional constant-a geometric property of the cross section. The dimensions
of J are length units to the fourth power. For the special case of the circular, cylindrical
shaft, it is equal to the polar moment of inertia of the cross section. Its value in other
cases will be illustrated in later examples.
The rate of twist is the measure of the torsional strain. If it is integrated along the
full length of the member portrayed in Figure 4.8, the total rotational displacement at
point 2, Bx2, is obtained:
Mx2 dx
(L (L Mx2L
(Jxz Jo /3 dx = Jo
= =
GJ Gf
Thus, by comparison with the axial force member, [d] = L/GJ, [cl>]= -1, and it follows
that
(4.27a)
{ } [_ { }
and
J
Mxl GJ 1 -l (Jxl
= (4.27b)
Mx2 L 1 1 Bx2
4.5.3 Beam Bent About Its z Axis
Since we are considering a bisymmetrical member, we only need to treat bending about
one axis in detail. Relationships for bending about the other follow readily by replace
ment of the relevant subscripts and adjustment of algebraic signs. But since this section
also introduces the basic force-displacement relationships for the flexural member, it
is appropriate to discuss certain fundamental aspects of beam flexure and to define
conditions and terminology that will be used here and at a later time.
The stresses and strains at any cross section caused by bending about the z axis are
directed along the x axis of the member. They vary linearly with respect to they axis
and are constant in the z direction for a fixed value of y. In elementary mechanics it
is shown that the strain, ex is given by
y
ex = -- = (4.28)
p
where p, the radius of curvature, is approximated by l/(d2v/dx2).2 Also, <Ix = Eex, so

that
d2v
<Ix = -Ey (4.29)
dx2
The direct stress <Ix has a zero stress resultant (JA <Ix dA = 0), but the moment about
the z axis produced by <Ix is nonzero and equal to M, that is,
M, = - L <IxY dA (4.30)
In the analysis of bending, one customarily works with stress resultants such as bending
moments and associated transverse shears, rather than with unit stresses. For this rea
son, in subsequent portions of the text, we frequently refer to the moments as the
"stresses" of bending, their relationship to the actual stresses being given by Equation
4.30.
One further analogy will prove useful in later work. Substitution of Equation 4.29
into Equation 4.30 gives
d2v
M, =
J A
E
dxz
yz dA (4.31a)
and, since E is constant and d2vldx2 is not a function of the cross-sectional coordi-
natey,
d2v d2u
M, = E
dx2 J A
y2 dA = EI,
dx2
(4.31b)
where I, = JAy2 dA is the moment of inertia of the section about the z axis. If, in
Equation 4.3lb, M, is regarded as the "stress" of bending, then d2vldx2 is correspond
ingly the "strain" of bending. The "elastic coefficient" connecting "stress" to "strain"
is no longer just E; it is now EI,. The factorI, reslts from the integration of behavior
on the cross section, just as M, is the integrated bending effect of the stresses <Ix.
The development of the stiffness matrix of a beam element is somewhat more dif
ficult than that for the axial and torsional elements. The stress is related to the strain
by a second-order, rather than first-order, differential equation (Equation 4.31b ). In
tegration produces two constants of integration that must satisfy the specified support
conditions. Also, several possibilities exist for the stable, statically determinate support
that is needed for the generation of a flexibility matrix. The most obvious choices are:
(1) to place simple supports at each end (the simple beam), and (2) to fix either end,
leaving the other free (the cantilever). Different degrees of freedom are suppressed in
each case and, consequently, different flexibility matrices will be obtained. But the
2The negative signs in Equation 4.28 are occasioned by the fact that we are assuming tensile strain to be
positive. Likewise, in Equation 4.30 to follow, we are assuming positive bending moments to be those that
cause positive curvature.
resulting complete stiffness matrix must be the same, since it inciudes all degrees of
freedom. We will use the cantilever system shown in Figure 4.9.3
ty
]--
/ -'
,=constant
M,, -
y l -
,J,___
- L ------+< ----L------+1
(a) Support system (b) Free body diagram
Figure 4.9 Beam bent about: axis. (a) Support system. (b) Free-body diagram.
The flexibility matrix sought is the one that satisfies Equation 4.14, which in this
case reads
{ }
V2 = (
d] { } Fy2
8,2 M,2
The terms in [d] are developed in Example 4.3. The result is

2
[d] =
!::.._
EI,
[ i
!::_
2
l
The remainder of the process of forming [k] is presented in Example 4.4. Thus, after
reordering the rows and columns of that example, we have
12 6 12 6
Fy1 2 2 v,
L L L L
6 6
Mo, I
4 2 8,1
EI, L I L
---------- 1 ----------
(4.32)
L 12 6 12 6
Fy2 2 2 V2
L L L L
6 6
M,2 2 4 8,2
L L
Develop the flexibility matrix for a prismatic beam supported as a

cantilever.
3In this case the choice of support conditions is immaterial; it's a matter of personal preference. But there
are cases in which certain choices are preferable from the standpoint of integration and satisfaction of
support conditions.
From the differential equation of the elastic curve, Equation 4.3lb
d2u M, 1
[F 2(L - x) Mzi]
dx2
=
El := EI, y
+
Integrating,
]
x
_jI L
"'1
du -[ ( 1 x2
= F Lx - + M x + C1
dx EI, yZ 2
) '2 -x dx
and
t]
L 2 M 2
C1x
u =
; [ ( ; ')
I,
F2
y
- + + + C2
du
l
dx x-o
- 0 . . C1 = 0 and ulxo = 0 .. C2 = 0
Elastic curve equations:
F 2 x2 (L )
u
_
= y
21, 3
and
du F x M,2 x
y2
=
(L ) _
dx El, 2 El,
Determine flexibility influence coefficients:
and
Assemble equations in matrix form:
EXAMPLE4.4
Using Equation 4.25 and the results of Example 4.3, develop

the stiffness matrix for the prismatic beam shown.
Write equations for [] and [di-1:

Apply Equation 4.25
Collecting terms,
V2 0,2 V1 0,1
12 6 12 6
1
2 2
-- --
L L'I L L
I
6
I 6
I
I 4 -
2
EI, L I L
( k] =
L
---------
I
---------
12 6 12 6
2 2
L L L L
6 6
2 4
L L
4.5.4 Beam Bent About Its y Axis
The development of the stiffness matrix for the beam bent about its other principal
axis is identical to the previous case and need not be repeated. There is, however, a
complication in signs that is impossible to avoid if we are to employ uniform notation.
The source of the problem may be seen by comparing Figures 4.10 and 4.9b. In each
case the member is viewed from the positive end of the coordinate axis that is normal
to the plane of the paper, and the forces and degrees of freedom are shown in their
positive directions. All moments and rotations are positive counterclockwise on the
ends of the members. In the case of the beam bent about its z axis, however, direct
forces and translational degrees of freedom are positive upward, whereas in the other
case they ;::- pvsitive downward. Therefore, the conversion of flexibility and stiffness
matrices devc!o;-erl for the case in Figure 4.9 to the case in Figure 4.10 requires a
change in the sign of all influence coefficients relating direct forces to rotations and,
F,, lf4-----
Ely= constant
L _____
_____..
Figure 4.10 Beam bent about y axis.

for reciprocity, those relating moments to translational displacements. Thus we have,

for the beam bent about its y axis,
I
12 6 I
12 6
Fz1
I
W1
L2 L2
I
L I
I
L
I
6 I 6
My1 4 I -
2 f)yl
L I
L
= g_ I
-------- - --1----------
(4.33)
L 12 6 I
12 6
Fz2 W2
L2 L L2 L
6 6
My2 2 4 f)y2
L L
It is left as an exercise for the reader to demonstrate the validity of the sign changes
made.
4.5.5 Complete Element Stiffness Matrix
All that remains of this development is to assemble the foregoing results in the com-
plete stiffness equation for the 12-degree-of-freedom, bisymmetrical member. Since
for small displacements the axial force effects, torsion, and bending about each axis
are uncoupled, the influence coefficients relating these effects are zero. Therefore,
taking Equations 4.26b, 4.27b, 4.32, and 4.33, suitably reordering rows and columns,
and letting G =
E/2(1 + v) (Equation 4.2), we arrive at the result shown as Equation
4.34 in Figure 4.11.
A A
I
I
Fx1 -
0 0 0 0 0 -- 0 0 0 0 0 U1
L I
L
I
I
12/z 61, I 12/, 61,
Fri 0 0 0 0 i
,
I 0 - 0 0 0 2 V1
L3 L I
I
L3 L
12/y 6/r
I
I 12/y 6Iy
F ,1 0 0 0 i 0 I
0 0 0 i 0 W1
L3 L
I
I L3 L
J J
Mx1 0 0 0 0 0 0 0 0 0 0 Oxl
2(1 + v)L 2(1 + v) L
6ly 4ly 6ly 2ly
My 1 0 0 2 0 0 0 0 2 0 0 Ori
L L L L
61, 4 1, 61, 21,
M ,1 0 2 0 0 0 0 - 2 0 0 0 0,1
L L L L
=E ----------------------------------------------------------------------------
A A
Fx2 -- 0 0 0 0 0 -
0 0 0 0 0 U2
L L
12/, 61, 12/, 61,
Fr2 0 0 0 0 - 0 0 0 0 - V2
_
L2 2
L3 L3 L
12/y 6ly 12/y 6ly
F,2 0 0 0 2 0 0 0 0 i 0 W2
L3 L L3 L
J J
Mx2 0 0 0 0 0 0 0 0 0 0 Ox2
2(1 + v)L 2(1 + v) L
6/r 2ly 6ly 4ly
Mr2 0 0 2 0 0 0 0 2 0 0 Oy2
L L L L
61, 21, 61, 41,
M,2 0 2 0 0 0 0 - 2 0 0 0 0,2
L L L L
(4.34)
Figure 4.11 Bisymmetrical framework element stiffness matrix.
4.6 A COMMENTARY ON DEFORMATIONS AND DISPLACEMENT VARIABLES
The deformations neglected in the development of Equation 4.34 (Figure 4.11) and
the particu!ar variables used in the definition of displacements and their derivatives
deserve further explanation at this point.
4.6.1 Neglected Deformations
As mentioned earlier, in developing Equation 4.34 deformations due to transverse

shearing stresses and torsional warping were neglected. These effects are illustrated in
the sketches of Figures 4.12 and 4.13.
(a)
t

(b)
(c) Figure 4.U Warping caused by transverse shear.
When a beam is flexed, the transverse shear stresses in the web cause cross sections
t0 warp in the longitudinal direction as in Figure 4.12b. We have neglected this, em
ploying instead the conventional beam theory, which assumes that plane sections re
main plane in flexure (Figure 4.12c and Example 4.3). Except in beams that have large
depth-to-span ratios or local regions of high shear-to-moment ratios, the additional
displacement due to web shear strain is of little consequence. Approximate procedures
for including it are presented in Section 7.6.
Similarly, when all but circular cross-sectional shafts are twisted, cross sections warp
longitudinally as shown in Figure 4.l3b. In shafts of closed cross section, such as cylin-
I (a)
I
I I l
I I I
(b)
I (c) Figure 4.13 Warping caused by torsion.

4.6 A Commentary on Deformations amt Displacement Variables 75
drical or rectangular tubes, or rectangular reinforced concrete beams, the increased

flexibility due to this type of deformation is generally small and may be neglected
(Figure 4.13c). In members of open cross section, such as steel wide-flange or light
gage beams, resistance to out-of-plane warping, which may arise either as a result of
fixing certain cross sections or varying applied torque along the member, may alter
substantially the manner in which the member and the entire framework respond to
torque. When torsional effects are small, as they are in many civil engineering frames,
the assumption of a constant rate of twist used in the development of Equation 4.27
is adequate. When torque is significant, the warping resistance effect should be in
cluded. A procedure for doing this is presented in Section 7.4.
4.6.2 Displacement Variables
In Figure 4.6 and throughout Section 4.5 the rotational degrees of freedom were listed
as angles, but in the development of the flexural stiffness relationships derivatives of
lateral displacement were used. Oniy in the analysis of torsion was an angular mea
surement (the rate of twist) used. The connection between components of rotation
and lateral displacement derivatives was explained in Section 2.2, in which it was noted
that in conventional frame analysis it is generally assumed that a line drawn normal
to the elastic line of a frame element remains normal to that line as the structure
deforms under load. In that case the rotational compor..ents at a node such as the 2
end of the element of Figure 4.6 can be represented as in Equatio'1 2.1, which is
reproduced here for reference:
8 2 =
x
aw
ay 2
l 8 2 =-
y
aw
ax
l 2
8 2 =
z
av
ax 2
l (2.1)
Not only are lateral displacements and their derivatives at the core of elementary
beam theory but, as shall be shown in Chapter 7, they are indispensable in the "shape
function" approach to contemporary finite element analysis. For this reason, it is de
sirable to summarize the background and limitations of Equation 2.1.
During the general homogeneous deformation of a body, any element in it may be
strained, translated, and rotated. If the strains are infinitesimally small, the unit elon
gations in three orthogonal directions and the engineering shearing strains related to
the same directions are (see Sections 81-83 of Reference 4.1):
au av aw
e =- e =- e =-
x ax y ay z az
=
( au av
) =
( au aw
) =
( av aw
)
(4.35a)
'Yxy ay + ax 'Yxz az + ax 'Yyz az + ay
Also, the components of rotation about the principal axes of strain are
Bx = e; - :) 8 =
y ( :: - :) 8 =
z (: - :;) (4.35b)
By definition, the shearing strains in the principal planes of strain are zero. In that
case, au/ay = -(av/ax), au/az = -(aw/ax), and av/az = -(aw/ay) from Equation 4.35a,
and, from Equation 4.35b,
aw av aw au av au
8 = - = 8 = a ---
- - (4.35c)
x ay az y ax az z ax ay
Thus it is seen that the reprec:entation of rotational displacement components by dis

placement derivatives as in Equation 2.1 or 2.la is based on the analysis of infinitesimal
strains and that only when they are infinitesimal is it rigorously correct. But in metallic
and similarly behaving constructional materials the actual strains are small. For ex
ample, unit extensions are generally of the order of a few thousandths of an inch per
inch. For this reason, conventional linear elastic analysis, which is based in part on the
relationships of Equation 2.1, can give a good picture of the response to service loads
of most structures. In nonlinear analysis we include finite strains that are the source
of higher-order terms neglected in the infinitesimal strain theory. These will be iden
tified and treated in Chapter 9 and Appendix A.
4.7 EXAMPLES
The following examples illustrate the applications of many of the concepts and results
developed here. The parts, but not the whole, of Equation 4.34 are used. Use of the
entire set of coefficients of Equation 4.34 occurs in space frame analysis. Discussion
of this is postponed until the next chapter, where the coordinate transformations
needed for the assembly of the global stiffness equations of complete space frames are
developed. We should note that in these examples and throughout the book we will
follow the convention of plotting bending moments on the tension side of the member.
12xample 4.5 demonstrates that flexibility matrices may be generated readily by tech
niques of classical analysis, such as the conjugate beam method.
Example 4.6 verifies a statement made in Section 4.4 regarding the equality of com
plementary energy under different statically determinate support systems.
In Example 4.7 it is shown that the beam stiffness matrix may be obtained directly
from the conventional slope deflection equations.
Example 4.8 is a preparatory example similar to Example 3.4. The matrices devel
oped here will be used in a number of the examples that follow in this and later
chapters.
In Example 4.9 the results of the previous example are applied to a simple case. To
illustrate some of the possibilities of matrix solutions, this five-degree-of-freedom sys
tem is solved by partitioning. Solution techniques will be discussed in Chapter 11.
Example 4.10 is a further application of the same basic stiffness equations, and in
Example 4.11, the same system is investigated for the effect of an assumed suppor
settlement. In Example 4.12 a bracket is added to the two-element beam, and a simple
torsional effect is investigated.
Example 4.13 is a preview of frame analysis problems. Here the emphasis is on the
physical interpretation of the stiffness influence coefficients. Note that although in
Example 4.13 the nodes are lettered a, b, and c as in Example 4.8 and the member
sections correspond, the orientation of member be with respect to the coordinate sys
tem is different than it is in Example 4.8. This ad hoc arrangement should cause no
difficulty in interpretation of the influence coefficients. But after transformation pro
cedures are introduced in the next chapter, problems of this sort will be solved in a
more automatic way.
In Example 4.14 some of the versatility of matrix procedures in combining special
elements is introduced. Later chapters contain additional examples of this powerful
aspect of the matrix formulation.
Example 4.15 is an application of matrix analysis to the classical problem of a beam
on a continuous elastic foundation in which it is assumed that the stiffness character
istics of the supporting medium are known. As they should be, the results obtained
are in almost precise agreement with those of the classical solution. Problems of this
type first became apparent in attempts to analyze the forces in railway rails. The same
analytical problems are encountered in systems such as cylindrical shells and structural
4.7 Examples 77
grillages (References 4.3 and 4.4), so the matrix approach can be used to advantage
in addressing those as well.
EXAMPLE4.5
Develop the flexibility matrix for a prismatic beam on simple supports.
------ L ____ __..,
Determine flexibility influence coefficients for Mz, and Mz2 using the conjugate beams and elastic
weights shown:
"EXAMPLE 4.6
Show that the alternative forms of the beam flexibility matrix in Examples 4.3 and 4.5 yield the
same complementary energy.
From Example 4.3, using Equation 4.4b,
Vi = (
2 /
z
Mz2
2 +
L . Mz2 . Fy2 +
2 2
L y2:) (a)
{ }
From Example 4.5,
_!::__)[
2
-l Mz'
Vi= !LF1J(d]/F1l = !LMz, Mz2 l\61, -1
2
J Mz2
(b)
But, by equilibrium
Mzl
yl ;S 2
M,2
--------
(c)
II+---- - L -- I
Substituting Equation c in Equation b,
( L F )
2 2
=
L
2EI,
M,/ + L M,2 Fy 2 + --f- = Uf QED
Using the "slope-deflection" equation, write the stiffness matrix for a prismatic beam element.
The straight element 1-2 is displaced by the forces shown. There are no loads between nodal
points. The "slope-deflection" equations for the member are as follows (see Ref. 4.2)
M,1
EI, ( 46,1 + 26,2
68 ) and M,2
EI, ( 26,1 + 46,2
68
-L
)
T L T
= - =
Letting 8 = u2 - vi.
M,1
EI, ( 46,1 + 26,2 +
6u1 6u2 ) and
T L -L
=
By equilibrium,
nd
FY2 =
- ( M,1 + M,2 ) =
EI, ( -
66,1 _ 66,2 _ 12u1 + 12u2
2 2
)
L L L L L L
12 6 12 6
Fy1 2 2 U1
L L L L
6
Mz1 EI, 4 2 6,1
L
L
=
12 6
Fy2 2 U2
L L
M,2 Sym. 4 6,2
which is identical to Equation 4.32.

4.7 Examples 79
[EXAMPLE 4.8]
For the system shown: a b c
1. Write the element stiffness matrices-assume no bending normal

4x HP mm2
;:2m4
to the plane of the paper. A
Iz mm4
=
2. Assemble the global stiffness equations. 50 x 106

HP mm4 HP mm4
=
E = 200,000 MPa and v = 0.3. OOx J = 100 x
m8 -------5 m ---
Define the coordinates, degrees of ty P
eaz v:
u ;;.
a ... -+...;_.;. Pmaz ya- Pax
follows:
9xa
;... .. .._ -,, -t--
t-P-mx . --t----;,-
a a c
1. Element stiffness equations. Use Equation 4.34, omitting out-of-plane shear and bending degrees of freedom, that is,
Wi, 8yi. w2, and 8y2
Member ab
0.750 I 0
Px 0 I -0.750
0 0 0 0 u.
I
Py 0.00469 I 00 18.750
-0.00469 0 18.750 Ua
I
Maxab I
14.423
I 0 0 0 -14.423 0 8xa
Mab
za 1.0 x 105 : 0 -18.750 0 0.5 x 105 8,a
= 200 --------------------------------------------------------------
I 0.750
Pxg 0 0 0 b u
Py Sym 0.00469 0 -18.750 ub

Maxbb 14.423 0 8xb
Mazbb 1.0 x 105 8,b
Member be
Fx/, 0.800 0 0 0 -0.800 0 0 0 ub

Py/, 0.00480 0 12.000 I 0 -0.00480 0 12.000 ub
Mbc I
xb 7.692 0 I 0 0 -7.692 0 8xb
I
M bc 0.4 x 105 : 0 -12.000 0 0.2 x 105
--1! = 200 ------------------------------------------------------------ !!b-
Fx I 0.800 0 0 0 Uc
I
Sym
I 0.00480 0 -12.000 Uc
Py I
I
Mbc
xc I
I 7.692 0 8xc
Mbc I
zc I 0.4 x 105 8 ,c
I
2. Global stiffness equations
I I
P,a 0.150 o o :-0.150
o o o o : o o 0 0
P,a 0.00469 0 18.750 0 -0.00469 0 18.750 : 0 0 0 0
I
P=a 14.423 0 0 0 -14.423 0 I 0 0 0 0
!'!!1:.a. 1.0 x 105: 0
---------------------
-18.750
----------------------
0 0.5 x- 105: 0
--
0 0
--------------------------
0
P,. : 1.550 o o o :-0.800 o o o
P,. : .0.00949 0 -6.750 : 0 -0.00480 0 12.000
p=b 200 22.115 0 : 0 0 -7.692 0
:
=
Pmzb 1.4 x 10': 0 -12.000 0 0.2 x 10-'

---------- --- - -- --- ---;: --- -- - -- --- - - - ----- ------; - 0.800- - - - - - - - - - - - - - - - - ---
Pxc 0 0 0
P,, Sym 0.00480 0 -12.000
Pmxc 7.692 0
Pmzc 0.4 x 10-'
EXAMPLE4.9
The beam of Example 4.8 is supported and loaded as shown.
1. Calculate the displacements at b and c.
= (}xc =
Boundary conditions, Ua = Va = (}xa = = vb =e,. 0
Jz
Y
yc -
(}xb 0 R m za
=
Also, because of absence of torque.

R
+a ---------------- --..
c p -2- +b xc - ../2
Rya Ryb
1. Displacements. Remove the u0, u0, (}xa 6,0, vb , (}xb (}xc rows and columns from the stiffness equations leaving
Pxb 0 1.500 0 -0.800 0 0 ub

6,b
pmzb
=
0 1.4 x 105 0 -12.000 0.2 x 105
P xc 51\12 200 0.800 0 0 Uc
Pye -51\12 Sym. 0.00480 -l.000 Uc
pmzc 0 0.4 x 105 6,c
As an alternative to solving on a computer as in Example 3.5, solve by operating on partitioned matrices after reordering;
thus
Pxb 0 1.550 -0.800: 0 0 0 ub

I
P xc
=
51\12 -0.800 0.800: 0 0 0 Uc
o - - --6---i- -ooM8o_iI _.::
12.000-
- - - --
-
-_-51V2 - ::_12000-
-
Pye 200
L --------- L -----------------
Uc
Pmzb 6,b
--------------
0 0 0 I
I -12.000 I
I x 1.4
x 105 0.2 105
pmzc 0 0 0 -12.000 : 0.2 x 105 0.4 x 105 (},c
I
Expanding the top partition and solving for L ub uc Y yields
{ ub } = 3 33 X 10 [
_3 0.800 0.800 ]{ } { }0 0.024
mm
Uc 0.800 1.550 51\12 0.046
=
Expanding the bottom P\Htition and solving for L 6,b 6,cf , in terms of Ve
{ (}6,zb } = 1.923 X 10-5 [ 0.4 -0.2 ]{ 12.000

uc } 23.077 X 10-5
{ }
0.2
uc
c -0.2 1.4 12.000 1.2
=
Expanding the middle partition and using the above expression for L 6,b 6,cy,
Ji= [ 200 0.00480uc - 23.077 x 10-5L12.000 12.000J {: } ] uc or uc = -19.15 mm
Substituting Uc in the expression for L 6,b 6,cY yields

{ (}6,czb } {= -0.00088
-0.00530
} rad
4. 7 Examples 81
2. Reactions. The remaining rows of the stiffness equations are used as follows:
ub Uc Uc e,b e,c
u
0.024
r} rl }
-0 750 0 0 0
0.046
Rya 0 0 0 18.750
= 200 -19.15
Rmza 0 0 0 0.5 X 105
-0.00088
Ryb 0 0 -0.00480 -6.750 12
-0.00530
60kN
-3.30kN
=
-8.80kNm
6.85kN
Moment diagram
8.80kNm V b c
Deflected structure
3.6 8.80kNm
0.00088 rad
r
f 3.30kN
19.15 mip
51-fJ.
!* 5!-f2.
0.00530 rad
The beam of Example 4.8 is supported and loaded as shown.
1. Calculate the rotation at b.

2. Calculate the rotation and bending moments.
Boundary conditions: u. = Va = (Jxa = e,. = vb = Uc =

(Jxc
=
e,c = 0
Also, ub = (Jxb = 0 because of absence of axial force and torque.
1. Displacements. Remove all except the e,b row and column from the stiffness equations, leaving
or e,b = 0.001786 rad

2. Reactions. The remaining rows of the stiffness equations yield
Rya 18.750 6.70 kN
Rmza 0.5 x 105 17.86 kNm
Ryb = 200 -6.750 8zb = -2.41 kN
Rye -12.000 -4.29 kN
Rmzc 0.2 x 105 7.14 kNm
Obtain Mg and M from the member stiffness equations, Example 4.8:

Mg = 200[1 X l05]8zb = 35.72 kNm M = 200[0.4 X 105]8zb = 14.28 kNm
Moment diagram 17.86kNm 14.28kNm
t=::-::---..
a
<:::c:::J b
7.14kNm
35.72 kNm
Deflected structure 0.00179 rad
2.41kN
EXAMPLE 4. .g
The beam of Example 4.8 is supported as shown. The support at b ----------------____,
settles 20 mm, carrying the beam with it. a b c

I
1. Calculate the rotation at b. ,..___ 8m
___ ,....._ 5 m ---J ____ __
2. Calculate the reactions and bending moments.
Boundary conditions: Ua = Va = Bxa = Oza = Uc = Ve=

Bxc 8zc = 0.
=
Also ub = 8xb = 0 because of absence of axial force and torque.
1. Displacements. Remove al! except the vb and 8zb rows and columns from the stiffness equations leaving
Ryb Ryb [ 0.00949 -6.750 ]{ }-20

{ } { } = = 200
Pmzb 0 -6.750 1.4 X 105 8zb
Solving the lower equation for 8zb,
8zb = -0.0009643 rad
2. Reactions. The remaining rows of the stiffness equations yield
Vb 8zb
Rya
() l ... -0.00469 18.750 ) 15.14 kN
( )
Rmza -18.750 0.5 X 105 6 .36 kNm
5
_20 O
Ryb = 200 0.00949 -6.750 {
-o. 009643
} = -36.66 kN
Rye -0.00480 -12.000 21.51 kN
Rmzc 12.000 0.2 X 105 -51.86 kNm

4. 7 Examples 83
Mg and MZ from the member stiffness equations, Example 4.8:

Mg = 200(-18.750vb + 1.0 x 1058,b] = 55.71 kNm
Obtain
MZ = 200(12.00vb + 0.4 x 1058,b] = -55.71 kNm
Moment diagram 6kNm

&
a

c
55.71 kNm
Deflected structure
65.36kNm 0mm
2
G-=.:::.
t .::
15.14kN
BXAMPLE 11.12
The beam of Example 4.8 is supported as shown and loaded by a 1 kN mm
force applied to a rigid bracket projecting from the beam at b.
lk: y
L Calculate the displacement at b.
2. Calculate the reactions, torques, and bending moments.
a
b
: c
m z/ .I. 5 m _J
I
1--- 8 ----
--
Boundary conditions: Ua = Va = 8xa = 8,a = Uc = Ve= 8xc = 8zc = 0.

Aho, u, 0 booaure of abren<0 of axial '"' ------ Py_b _=_-_lk_ _N_l....._ 4_ _P_:_ b_ =-_04 kN_mm_ -1+--+
R b
+;ya
1. Displacements. Remove all except the vb, 8xb and 8,b rows and columns from the stiffness equations leaving
=
{::b} { } [ = 200
000949

22. 15
-6.750
0
vb
]{ }
8xb
Pmzb 0 Sym. 1.4 x 105 8,b
Solving the second equation for 8xb
8xb = 0.009044 rad
Solving the first and third equations for Lvb 8,bf,

2. Reactions. The remaining rows of the stiffness matrix yield
vb f)xb fJ,b
Rya -0.00469 0 18.750 0.423 kN
Rmxa 0 -14.423 0 -26.1 kNmm
r,., 1
Rmza -18.750 0 0.5 X 105 1.781 kNm
= 200 0.009044 =
Rye -0.00480 0 -12.000 0.586 kN

-0.0000263
Rmxc 0 -7.692 0 -13.9 kNmm
Torque diagram
Rmzc 12.000 0 0.2 X 105 -1.413 kNm
1: 61kNmm
'
___.I
a
13.9kNmrn .______
Moment diagram 1.41kNm
b L:-1

a c
1.52 kNm
Deflected structure
EXAMPLE 4.13i
The rigid frame shown is made of the elements of Example 4.8.
1. Using the influence coefficients of Example 4.8 assemble the stiffness
}
equations relating the forces applied at b to the degrees of freedom
a
at that j oint.
A= 6x 103mm2
2. Calculate the displacement at b. I= 200 x 106 mm4
A= 4 x HP mm2
I= 50 x 106 mrn4
}
Nonzero degrees of freedom:

4.7 Examples 85
1. Stiffness equations. Select coefficients from Example 4.8 on the basis

of physical behavior. Apply a horizontal displacement, ub:
Pxb = 200(0.750 + 0.00480)ub = 200 X 0.7548ub
Pyb = 200(0 + O)ub = 0
Pmzb = 200(0 + 12.000)ub = 200 X 12.000ub
Apply a vertical displacement, ub:
Pxb = 200(0 + O)ub = 0
Pyb = 200(0.00469 + 0.800)ub = 200 x 0.80469ub
P mzb = 200(-18.750 + O)ub = -200 X 18.75ub
Apply a rotation, O,b:
Pxb = 200(0 + 12.000)0,b = 200 x 12.000,b

a
Pyb = 200(-18.750 + 0)0,b = -200 x 18.750,b
pmzb = 200(1.0 x 105 + 0.4 x 105)0,b = 200 x 1.4 x 1050,b
{ } {
Pxb
Pyb =
100/\12
-1001\12
} [
Assemble equations in matrix form and apply joint loads:
= 200
0.7548 0
0.8047
12.000
-18.750
]{ }
Ub
ub
Pmzb 50 x 103 Sym. 1.4 x 105 O,b
{} {
2. Displacements. Solving the stiffness equations yields
}
100/"2 kN
ub 0.4414 mm
ub = -0.3998 mm
o,b 0.00169 rad
100/"2 kN
(Note:ub and vb
not shown)
fEXAMPlE 4.14]
The two beam elements shown are joined by a hinge at
l
0. Develop the stiffness matrix for the combined
member. Hinge Fy2, v2
( r)
Fy2 Vz
E_ I -o..- E _I M ,2 A 2 x'
M,i,8,i 1-,..._ ___,-t
l O 2
____ ____ __
:I==-a-L . j
I _ b
_
Write separate stiffness equations for each element using Equation 4.32. Account for:
1. Equilibrium conditions.
(a) Zero bending moment at hinge.
(b) Equal shear on two sides of hinge.
2. Compatibility condition; equal transverse displacement at hinge.
3. Special conditions; slope discontinuity at hinge.
Equations. Incorporating the above

conditions, we have
------_..,
a ---- --_..,
b
12 6 12 6 12 6 12 6
2 2 2 2
aJ a3 b3 b3
Vo
a a b b
EI
4
a
6
a
2
2
a
0
EI
-
4
b
6
b
2
2
b
12 6 12 6
= =
2
aJ 3
Vo Vz
b b
4 4
0 Sym. Sym.
a b
Assemble two sets in a single matrix format, ordered and partitioned as follows:
12 6 12 6
J 2 0 0 2 0
a a aJ a
6 4 6 2
2 0 0 2 0
a a a a
i2 6 12 6
0 0 2 0 2
b3 b3
Vz
b b
M,2 EI 0 0 - 6
2
4
b L
6
2 0
2
( 1-)
b b b
=
------------------- ---------------------
12 6 12 6 6 6
0 12 !3 +
a3
2 b3 b2 b3
2 2 Vo
a a a b
6 2 6 4
0 2 0 0 2 0
a a a a
6 2 6 4
0 0 0 0
b2 b b
2
b
4.7 Examples 87
Eliminate degrees of freedom at hinge, Lii0Y =Luo 0,01 0,0,JT, by operating on partitioned equations. Do symbolically.
Submatrices are as indicated above.
Expand lower partition and solve for {.10}:

(a)
Expand upper partition and substitute Equation a for {'10}:
Thus the stiffness matrix [K] of the combined member is
Carrying out the indicated operations yields (Do as Problem 4.5)
{F.} = [K]{a.}
{FyFy21} r
M,1
= 3
3El
(a + b3)
1 a
a 2
-1
-a
M,2 Sym.
Member ab is a beam on an elastic foundation lab = 128.5 in4 Foundation

modulus k = 1.5 k/in2 E = 29,000 ksi.
Determine the deflection and the bending moment in the beam at c by:
1. Elastic theory.
2. The stiffness method.
1. Assuming linear elastic behavior of the beam and the supporting foundation it may be shown that (see Ref. 4.3):
Pf3 sinh f3L - sin f3L .!._ sinh f3L + sin f3L
V = (a) and Mc = (b)
e
2k cosh f3L + cos f3L 4{3 cosh f3L + cos f3L
in which
f3 = 4 = 4
1.5
= 001781 h;"
1
v 4Ei;;;, 4 x 29,000 x 128.5
From (a) V 0.237 in. from (b) Mc = 554 in. kips

e
=
Model as a beam supported on 19

2.
lYWlil
ok lfrrum
1 5"
equally spaced axial members of
a c b
L = 15 in.
x
01
I . 20 @ 1s dOO"
22.5 x 15
k per support = 1.5 x 15 = 22.5 k/i.n,, A sup. = = 0. 0116 m
. 2
29,000
Deflected structure:
Moment diagram:

-
0
. I I I I lTI::J/Fl I I I I I .
Mc= 547 in. kips
4.8 PROBLEMS
As in Chapter 3, it is suggested that, in several of the following problems, a computer
program such as MASTAN2 be used and that the results be manually checked.
4.1 Assemble the global stiffness equations for the beam shown. Neglect axial de
formation. Compute the displacements, reactions, and internal forces for the loading
and support conditions indicated.
a b
=30 0 LP 2 0
c d
P=1100'163o0 l=11005630 =11003063 0

mm4
x
mm4
J
mm4
mm4
x J
mm4
x
mm4
lOm
=20 =0.3
E
15m
GPa
(a )
v
lOm

a
5 kNm (applied at
(b)
b)

b m 0
c d
(Torsional restraint at a and d only)
kN
a r
b
c d
(c)
m i& J-25mm settlement

(d)

a --- i'/-4---- ---d c
0
b
Internal (e)
kN hinge
a
r
b 0
4 m--1
20kNm
(j) (g)
Problem 4.1
4.2 Develop the stiffness matrix for the beam in Figure 4.10, that is, verify Equation
4.33.
4.8 Problems 89
4.3 One way to support a beam in a stable, statically determinate fashion is as indi
cated (rotation at the left end is prevented). (a) Verify that the flexibility matrix is
{Vt} L [ 2L2 -3L Fy1

]{ }
8,2 = 6EI, - 3L 6 M,2
(b) Verify that the complementary strain energy equals that of the cantilever and
simply supported element (see Example 4.6). (c) Develop the beam stiffness matrix
using this flexibility matrix and verify Equation 4.32.
r M,2, B,2
t, x
EI,

rF,, ''
L
Problem 4.3
4.4 Compute the displacements, reactions, and internal forces for the beams shown
in the figure. Insert nodes as required. E and I are constant.
!:
1-13
1..---a b
--.++
(a)
(c)
(b)
Problem 4.4
4.5 Carry out all of the algebraic operations needed to obtain the final combined
member stiffness matrix of Example 4.14.
4.6 (a) By algebraic operations on Equation 4.32 develop a condensed stiffness ma

trix for a beam element in which the bending moment at one end is always zero
(part a of the figure below). (b) Use this stiffness matrix in the analysis of the beam
in part b.
Pu;
a c
2
1
#a r,I
y;
EI
dh
L L L
(a) (b)
Problem 4.6
4.7 Compare the stiffness method with the method of moment distribution.
4.8 Compute the displacements, reactions, and internal forces for the systems shown.
E = 200,000 MPa.
b
!: OkN
b
/=200x106mm4 r
5m Sm
103 mm2
103 mm2 /, c _J _J_
0
Sm---<
(a) (b)
Problem 4.8
For the beams I= 700 X 106 mm4, A = 15 x 103 mm2. For the struts, A = 10 x 103 mm2,
E = 200,000 MPa.
Problem 4.9
4.10 Compute the displacements, reactions, and internal forces for the system shown.
For the beam I= 500 X 106 mm4, A = 12 X 103 mm2 For the struts, A = 8 X 103 mm2,
E = 200,000 MPa.
1 12m 12m
20kN
12m
a b c
Problem 4.10
4.11 Same as Problem 4.10 but place an internal hinge halfway between a and b.
4.U Compute the displacements, reactions, and internal forces for the systems
shown. E = 200,000 MPa.
SOkN
(a) (b)
Problem 4.U
4.8 Problems 91
4.13 Plot the bending moment at a versus /1//2 as /1//2 varies from 1 to 10.
a El1
+: El2 t: El1 , d
t-f L
3
L
f
Problem 4.13
4.14 Plot the ratio of the bending moment at a to the bending moment at c versus
11112 as /1//2 varies from 1 to 10. Assume A is infinite ( neglect axial deformation of the
strut ) .
a
'!-------.,
!:
EA L
4
c d e _l_
't--------------t'#,

* '----- l: -r
Problem 4.14
4.15 Compute the displacements, reactions, and internal forces for the system shown.
E 200,000 MPa.
r
=
40kN
4 rn --.b-----4 rn --1---_,.8 rn ----
6
l 200 x 310 rnrn4
6
=
l 50 x 10 6 x 10 rnrn2
3 rnrn4
= =
A = 4 x 10 rnrn2
Problem 4.15
4.16 Members ab and be only possess a torsional stiffness k = GJ/ L. Calculate the
angles of twist ()x and ()Y at node b due to the applied twisting moments Pmxb and Pmyb
Problem 4.16
REFERENCES
4.1 S. Timoshenko and J. N. Goodier, Theory of Elasticity, Third Edition, McGraw

Hill, New York, 1970.
4.2 C. H. Norris, J. B. Wilbur, and S. Utku, Elementary Structural Analysis, Third
Edition, McGraw-Hill, New York, 1976.
4.3 S. Timoshenko, Strength of Materials, Part II, Advanced Theory and Problems,
Third Edition, Van Nostrand, Princeton, N.J., 1956.
4.4 J.P. Den Hartog, Advanced Strength of Materials, McGraw-Hill, New York, 1952.
Chapter 5
Stiffness Analysis of Frames-II
In this chapter, with the introduction of methods of performing coordinate transfor

mations and for treating loads applied between nodal points, self-straining, and tem
perature change, we complete the development of the basic tools required for the
linear elastic analysis of a framework of any shape, connection type, and loading.
The transformation of degrees-of-freedom, forces, and force-displacement relation
ships from one coordinate system to another can be treated as an exercise in analytical
geometry. It is important, however, to keep in mind the physical problem being ad
dressed, that is, to remember that merely changing the frame of reference can in no
way alter the actual response of the structure.
The concept of fixed-end forces will be introduced in dealing with distributed loads,
self-straining, and thermal effects. Understanding this notion requires that one visu
alize the application and subsequent removal of temporary, fictitious restraints on the
degrees-of-freedom and grant to such constraints the ability to resist force. It is a
concept that is at the heart of the displacement approach to analysis.
5.1 COORDINATE TRANSFORMATIONS
We have already discussed the reasons for employing a number of coordinate systems
in the analysis of a particular structure and, as in Figure 2.4, we have indicated several
logical choices. The problem can be illustrated further by Figure 5.1, a tripodal space
frame with fixed supports and a rigid joint at a. The structure is shown placed in a
convenient global coordinate system with axes x, y, and z. A certain natural action is
exerted on the a end of member ab as the system deforms under the applied loads.
This action can be portrayed graphically or its components can be listed in vector
fashion. There are many ways to do this, but the action itself is immutable.
To illustrate some possibilities we first consider the case in which the resultant action
on ab is a single force and not a couple, that is, it is the bound vector F0b with an origin
at point q (Figure 5.lb ). The position of q with respect to a can be stated in terms of
the radial distance p, the azimuth angle a, and the elevation angle {3. Similarly, we
could list the azimuth and elevation angles, y and S, to the positive direction of the
force. The six quantities a, {3, p, y, S, and the magnitude F0b of the resultant force
completely define the natural action on the a end of this member. This is physically
clear, and it would be a useful way to define the action if ab were an isolated member
and Fab a load applied to a fixed external loading point. But it doesn't lend itself to
the representation of element stiffness, and therefore it will not be considered in this
connection.
Two other options are illustrated in Figures 5.lc and 5.ld. In the former a convenient
93
94 Chapter 5 Stiffness Analysis of Frames-II
y y
z
z
(a) (b)
y y
y'
z'
(c) (d)
Figure 5.1 Alternative representations of force. (a) Structure. (b) Resultant force.
( c ) Local coordinates. (d) Global coordinates.
set of local coordinates is established and the action is described in terms of the six
quantities Fx'a, Fya, F,.0, Mxa, My'a, and M,.0, that is, direct force and moment com
ponents acting parallel to the local axes. In the latter we view the same action as the
six quantities Fx0, Fy0, Fza, M_w, My0, Mza, direct force and moment components acting
parallel to the global axes.
The following vectors are thus alternative representations of the same quantity:
IF'} = LF x'a Fy'a Fz'a Mx'a My'a M,0JT

IF} = L Fxa Fya F,a Mxa Mya MzaY
Having either of these, the other can be obtained by a suitable transformation. We
use both repeatedly since we are interested in transforming force and displacement
vectors and, by logical extension, stiffness matrices, from local to global coordinates
and vice versa.
5.1.1 Transformation Matrices
Since, as shown in Figure 2.7, we can portray displacements (translations and rotations)
by line vectors in exactly the same way as we portray forces (direct forces and mo-
5.1 Coordinate Transformations 95
y y
y'
z z
z'
(a) (b)
Figure 5.2 Direct force at a. (a) Local coordinates. (b) Global coordinates.
ments), the rules for transformation of displacements and forces are identical.' Fur
thermore, as shown below, the necessary transformations may be accomplished
through rotation matrices that consist of direction cosines. This is the result of two
things. First, the equations that comprise the set of global analysis equations involve
equilibrium and compatibility at particular joints (see Figure 3.2 and Equations
3.3-3.5) and, in these, we simply need to transform between vectors passing through
the joint: rotation, but not translation, is required. Second, we are using orthogonal
coordinate systems.
For the reasons given above, the direct force components in Figures 5.lc and 5.ld
have the same resultant. In the same way, the resultants of the two sets of moment
components are equal. Consequently, in developing transformations linking local co
ordinates to global coordinates, one may treat direct forces and moments separately.
Rules developed for direct forces also apply to moments, since they are each repre
sented by sets of orthogonal vectors.
In Figure 5.2, the direct force components in the two systems are compared.
The resultant direct force (Sin the Figure), can be resolved into either its local com
ponents Fx'a Fy'a and F,a (Figure 5.2a) or its global components Fxa Fya and F,a
(Figure 5.2b ). It follows that each of the global components may in turn be resolved
into components in the local directions, and vice versa. The easiest way to express one
set of components in terms of the other is through direction cosines. Using the no
menclature for direction cosines defined in Figure 5.3, we have, for the local
components.
Fx'a = Fx a COS ll'x + Fya COS f3x + F,a COS 8x'

Fy'a = Fxa COS O'y + Fya COS /3y + Fza COS 8y'
F,a = Fxa cos a,. + Fya cos /3, + F,a cos 8,.
'This statement is valid for linear analysis. However, as shown in texts such as Reference 5.1, finite rotations
of a rigid body do not obey the law of vector addition. The finite rotation effect, which requires special
consideration in three-dimensional nonlinear an al ysis, is studied in Section A.4 of the Appendix.
Figure 5.3 Direction a ngle s.
Using, respectively, the symbols l, m, and n, with appropriate subscripts, to designate

the cosines of the nine direction angles between the global x, y, and z axes and the
subscripted local axes, we have
Fxa = fx.Fxa + mx.Fya + nx.Fza

Fya = fy.Fxa + my.Fya + ny.Fza (5.1)
F,a = l,Fxa + m,Fya + n,.Fza
In matrix notation this becomes
(5.2)
or, in short form,
(5.3)
The matrix [y] of the nine direction cosines relating the local coordinate system to the
global one is a rotation matrix.
Recalling that the sum of the squares of the direction cosines for any axis is unity,
we have
1;. + m;. + n;. = 1
/2.
y + my2 + n2
y = 1 (5.4)
t;. + m;. + n;. = 1
;fhe length of any vector is the square root of the sum of the squares of its components.
Thus each row of ['Y] has unit length. Also, for any two orthogonal axes such as x' and
'
y ', x' and z', and z and y ' , the sum of the products of the corresponding direction
cosines (the scalar product) is zero:
fx,fy' + mx.my' + nx.ny' = 0
lyf,. + mym,. + nyn, = 0 (5.5)

f,.fx + m,mx + n,.nx = 0
Together, Equations 5.4 and 5.5 state that the [-y] matrix consists of rows of orthogonal
-uni.Ly_ctors. Such a matrix is called an orthogonal matrix. It is a square matrix having
the distinguishing property of an inverse equal to its transpose:
(5.6)
5.1 Coordinate Transfonnations 97
yr
f Mr1> er, f My2, er2
t Fyl
---t' y2
V2
/1
v,
M F' ------------ iJ
h ' tl----;
F,1> w, F,2, w2
/ Young's modulus= E /
Shear modulus= G Mz2 ez2
//Afzl ezl
z
L-----
Figure 4.6 Bisymmetrical framework element.
This property may be verified readily by the identity ['YH'Yt1 = [I) = ['Y]['Y)T. It will
be shown later ( Equation 5.14 ) that use of the orthogonality of the ['Y] matrix simplifies
considerably the labor of transforming coordinates.
Recognizing again that direct forces and moments transform independently, and that
all that has been said about forces applied to one end of a member applies equally
well to the other end, we can use Equation 5.3 directly in compiling the force trans
formation equation for the 12-degree-of-freedom framework element. Referring to
Figure 4.6 ( fepeated here ) for general nomenclature, and using primes to designate
local coordinates, this equation is
I
Fx'I I F x1
I
Fy'I ['Y] : 0 : 0 0 Fr,
I I
Fz I I I F z1
'' ____ L ___ J ____ L ___ _
I I I
Mx'I I I I M x1
I I I
My'I 0 :I ['Y] :I 0 :
I
0 My1
I I I
!-!::. ----
l
r ---
I
, ----
I
r ---- !1::.1_ (5.7)
Fx2
'
I
I
I
I
I
F x2
F r2 o:o
I
:[y):o
I I
Fr2
I I I
Fz'2 I I I Fz2
____ L ___ J ____ L ___ _
I I I
Mx2 I I Mx2
Mr2 0 0 0 ['Y] My2
M,.2 Mz2
or, in shorthand
{F'J [r](FJ (5.8)

where
I [y1
['Y]
(5.9)
[rJ = ['Y]
[y]J
(I J symbolizes a diagonal matrix in the text ) . Just as ['Y] isorthogonal, so too is [r),
or
(5.10)
Also, considering that the 12 degrees of freedom of the member in Figure 4.6 are
portrayed by vectors just as the forces are, we may use the same transformation for
these, or
{A'} = [r]{A} (5.11)
5.1.2 Transformation of Degrees of Freedom
Having the necessary transformation matrix, the transformation equations for the
member stiffness matrix follows directly. If we write Equation 2.2 in local coordinates,
{F'} = [k']{A'} (5.12)
From Equation 5.11
{F'} = [k'][r]{A} (5.13a)
and, from Equation 5.8
[r]{F} = [k'][r]{A} (5.13b)
or
{FJ = [rr1[k'J[r]{aJ (5.13c)
Using the property of orthogonality (Equation 5.10),
{F} = (r]T[k')(r]{A} (5.14)
Thus
{Fj = [k]{A} (5.15)
where
[kJ = frY[k'J[rJ (5.16)
This is the equation most generally used for finding the member stiffness matrix in
global coordinates once one has formulated it in local coordinates and knows the
geometrical relationship between the two sets of axes.
5.1.3 Transformations and Energy
We have arrived at the necessary transformations in a direct and rigorous way, but it
is useful to reconsider the subject from an energy approach. It will bring out certain
principles that may not be apparent from the above and which will be used later.
The transformed degrees of freedom are not necessarily equal in number to the
original degrees of freedom. For example, compare Figures 2.8 and 2.9. In the first
figure the behavior of the axial force member is defined by two degrees of freedom,
whereas in the second figure, in which the only difference is the coordinate system,
four degrees of freedom are required. The relationships connecting the two displace
ment vectors may be written symbolically, as before
{a'J = [r]{aJ (5.11)
where, now, [r] need not be the square matrix defined in Equation 5.7 and, if it isn't,
the inverse, [r]-1 of Equation 5.13c, does not exist. To treat this condition we again
suppose the equations to be transformed are
{F'} =
[k']{A'} (5.12)
Also, we assume that each force component F; of the vector IF'} produces the work
during the displacement a;, with no work done under any other displacement com
ponents in Iii'}. This is another consequence of employing components that act along
orthogonal axes. Such force and displacement vectors are called conjugate vectors.
Both sets, that is, Iii'}, IF'} and Iii}, IF} are taken to be conjugate sets of vectors.
Changing the frame of reference cannot alter the amount of work done. Therefore,
for the work to remain invariant under the imposed transformation, the following
equality must be satisfied:
!LF'Jlii'J = !LFJlii}
and, from Equation 5.11
LF'J[r]{ii} = LFJlii}
Thus
L F'J[r] = LFJ
or, by transposition,
(5.17)
Since Equation 5.11 was invoked in this derivation, the transformation of displace
ments presented in Equation 5.11 implies the transformation of forces given by Equa
tion 5.17. The force and displacement transformations are termed contragredient under
the stipulated conditions of conjugacy. The conclusion of immediate interest is that, if
the force transformation is first defined as in Equation 5.17, then the displacement
transformation matrix of Equation 5.11 is given by the transpose of the force-trans
formation matrix. The principle of contragredience is of considerable importance when
the displacement (or force) transformation is readily constructed from physical mean
ing, but the formation of the conjugate vector is not readily perceived. This occurs, for
example, when the condensation of degrees of freedom is accomplished by means of
a transformation process (see Section 13.3).
To follow this approach in developing element stiffness transforitiatigns, it is con
venient to deal with strain energy and external work, quantities that were introduced
in Section 4.2. Again we require that work remain invariant under the imposed trans
formation, permitting direct substitution of Equation 5.11 into Equations 4.3 and 4.4a.
Thus, from Equation 4.3,
W = !Lii'JIF'} !lii'}TIF'}
=
(5.18a)
= !LiiJ[r]TIF'} !LiiJIF}
=
and, from Equation 4.4a,
W = !Lii'J[k']{A') !LAJ[r]T[k'][r]{A) (5.18b)

=
= !LAJ[k]{A)
Hence the transformed stiffness matrix is again given by
(5.16)
The force vector is of course transformed by Equation 5.17. The transformation of [k'] /
into [k] by Equation 5.16 is a congruent transformation (see page 64). Therefore, if
[k'] is symmetric, [k] will also be symmetric.
5.1.4 Rectangular Transformation Matrices
The main reason for introducing the energy approach at this juncture is to show that,
under the condition of contragredience, it is not necessary to use the condition that
the inverse of the transformation matrix equals its transpose. Transposition can be
invoked directly. This allows convenient definition of nonsquare coordinate axis trans
formation matrices. As noted earlier, the stiffness matrix of the axial member in Figure
2.8 features the two axial displacements. The same element disposed in an x-y plane
as in Figure 2.9 is described by four displacement components. Arranging the local
'
x axis along the member of Figure 2.9 and defining the direction cosines as we have
here, the relationship between the local and global degrees of freedom becomes
The validity of this transformation may be verified by inspection, but it is done formally
in Example 5.1 by deleting irrelevant rows and columns from the complete transfor
mation matrix of Equation 5.7. In this example the same transformation matrix is used
to demonstrate the equivalence of Equations 2.3 and 2.5, which were developed in
dependently from basic physical principles in Chapter 2.
Additional demonstrations of basic transformations are contained in Examples 5.2
through 5.5.
Example 5.2 is similar to Example 5.1 in that it illustrates the reduction of the com
plete 12-degree-of-freedom transformation matrix to a special situation-in this case
the planar framework member. The partitioning of the [r] matrix follows Equation 5.7.
In Example 5.3, the results of several previous examples are used to demonstrate
the assembly of the global stiffness equations for a typical rigid frame. Axial defor
mation effects are included for completeness even though, in most practical civil en
gineering frames of this type, they may be neglected in linear analysis. The example
contains merely a listing of representative matrices and not details of their manipula
tion. The primed degrees of freedom refer to the local coordinates indicated.
The small space truss analysis of Example 5.4 is illustrative of the application of the
.same transformation techniques to space structures.
In Example 5.5 the problem of determining the direction cosines of the principal
axes of an oblique member is solved by viewing the total transformation as a series of
rotations. Most general-purpose computer programs have some built-in method for
transforming coordinates that require only simple descriptive input. Nevertheless, the
analyst should understand the geometrical and analytical problems involved.
Verify Equation 2.5 using Equation 2.3 and the transformation Equations 5.7 and 5.16.
Review development of Equations 2.3 and 2.5.
y
'
y
(a) (b)
Equation 2.3 was developed from a sketch similar to sketch a of the above figure. It may be
written
{F'} = [k']{ii'}
where [k'J is a 2 x 2 matrix.

Equation 2.5 was developed from sketch b. It may be written
{F} = [k]{a}
where [k] is a 4 x 4 matrix.

Transform [k'] to [k] using Equation 5.6
1 . [k] = [rF[k'][r], as follows: In Equation 5.7 delete
all columns except those corresponding to Fxi. Fy1' Fx2, and Fy2 and all rows except those cor
responding to Fxi and Fxz, leaving
Thus
which is the transformation matrix of this section.

From Equations 2.3, 5.16 and the above figure,
rcos<f>
,Ll[-
J J
EA sin<f> -1 [cos<f> sin<f> 0 O
[k]
0 1 0 0 cos<f> sin</>
=
0 sin<f>
or
rcos2<f> sin</> cos<f> -cos2<f> -sin<f> cos<f>l

EA sin2<f> -sin<f> cos<f> -sin2<f>
[k]
cos2 <f> sin</> cos<f>
=
L
Sym. sin2 <f>
which is the stiffness matrix of Equation 2.5.

EXAMPLE S.i,
The member shown is part of a frame located and loaded in the x-y plane.
1. Reduce the general Equation 5.7 to the particular transformation equation for
this case.
2. Demonstrate the orthogonality of the transformation matrix.
Define the local and global coordinates,

degrees of freedom, and forces:
Local system Global system

1. Deleting all irrelevant rows and columns from Equation 5.7 leaves
Fx'I Ix' : 0:0

fflx 0 :0 Fx1
F 'I
y
1y. m y: . I 0: 0 o:o Fy1
-0- - - - -:-To-
I -
-0- I: o
o M ,1
_ _ _
M:!. -------1-----------1---
Fx2 0 0 : 0 : Ix' fflx:0 Fx2
Fy2 0 0 : 0 : ly my : 0 F 2
y
-------L----------' ---
M,2 0 0 : 0:0 0 : n,. M, 2
or
I I I
cos cf> sin cf>:0: 0 0 :0
-sin cf> cos cf>:0: 0 0 : 0
------------ + --
------------ --
0 0 :i: 0 0 : o
[r] = ------------ , r --
------------ --
0 0 :0 : cos cf> sin cf> : 0

0 0 :0 : -sin cf> cos cf>:0
----------------------------
0 0 :o: 0 0 :i
2. Orthogonality.
It may be seen that [r]T[r] = [I]
Thus [r]T = [r]-1
,EXA MPLE..5:3
The rigid frame shown has E = 200,000 MPa and the following member prop
erties: for ab, cd, ed, A = 4 x 103 mm2, I= 50 x 106 mm4; for be, A = 6 x
103 mm2, I= 200 x 106 mm4 Using the results of Example 4.8 and the trans
formation matrix of Equation 5.2, develop the global stiffness equations for the
structure, including flexural and axial deformations.
0 x 0
r---- 7.416m- j-4 m-j
f.-- 1I.416m
Consider member ab in detail; show [k] only for

members be, ed, and ed:
ab
Member ab
[k'): [r]:
u'a u'a (Jz'a u/, vi, (Jz'b I I
0.800 0 0 -0.800 0 0 0 1: 0: 0 o: o I
0.0048 12.00 0 -0.0048 12.00 -1 0 : 0 : 0 o :o

------ - ------ 4 --
0.4 x 105 0 -12.00 0.2 x 105 0 0:1: 0 o:o

200 ------r-1--------
0.800 0 0 0 0101 0 110

I I I
Sym. 0.0048 -12.00 0 0:

----- 0: -1 0: 0
-1--4--------
0.4 x 105 o 0:0 : 0 0: 1
[k]:
Ua Va O, a ub vb o ,b
0.0048 0 -12.00 -0.0048 0 -12.00
0.800 0 0 -0.800 0
0.4 x 105 12.00 0 0.2 x 105
200
0.0048 0 12.00
Sym. 0.800 0
0.4 x 105
Member be
[k]:
ub vb o,b Uc Uc o, c
0.6452 0.2591 -7.0313 -0.6452 -0.2591 -7.0313
0.1095 17.381 -0.2591 -0.1095 17.381
1 x 105 7.0313 -17.381 0.5 X 105
200
0.6452 0.2591 7.0313
Sym. 0.1095 -17.381
1 x 105
Membered:
[k]:
Uc Uc o,c ud ud o,d
0.5137 -0.3817 7.200 -0.5137 0.3817 7.200
0.2911 9.600 0.3817 -0.2911 9.600
0.4 x 105 -7.200 -9.600 0.2 x 105
200
0.5137 -0.3817 -7.200
Sym. 0.2911 -9.600
0.4 x 105
104 Chapter S Stiffness Analysis of Frames-II
Membered:
[k]:
u, v, (},. ud vd (J,d
0.0048 0 -12.00 -0.0048 0 -12.00
0.800 0 0 -0.800 0
0.4 X 105 12.00 0 0.2 X 105
200
0.0048 0 12.00
Sym. 0.800 0
0.4 X 105
Assembled global stiffness equation:
P,, : Pxb P,. Pmz.b : Pxc P,d P,d

I I
I
Ub
: Uc 8,Jt
u, v, I 6:,b ud u, v,
I I
0.0048 0 -12.00 I -0.0048 0 -12.00 I u,
I
0.800 0 0 -0.800 0 I 0 0 0 v,
I
0.4 X 10' : 12.00 0 0.2 X 10' : I
----- - ---------
: - 0.6s00 - -02591- - - 4.969- : --=-oM.52- --=O.is91- - -_7.0313 I: - - ----- - ---------:----- -------------
: 0.9095 17.381 : -0.2591 -0.1095 17.381

: 0 0
: 1.4 X 10' :
7.0313 -17.381 0.5 X 10' I :
------------ --------------------1------------------------------------ 1 -----------------
200 1 I 1.1589 -0.1226 14.231 I -0.5137 0.3817 7.200 I
I I
Sym 0.4006 -7.781 0.3817 -0.2911 9.600 :
0
:
1.4 X 10' : -7.200 -9.600 0.2 X JQ' :
------ - --------i-
I ----------------I- -
: - - - - - - - - - - - - - - - - i - 05J.s5--=-0:3817 - - -480 - -:-:::o.0048- --o- - - - 1200 -
,
I 1.0911 -9.60 0 -0.800 0
I
I I I 0.8 X JQ' 1-J2.00 0 0.2 X JQ'
---------------L-----------------1------------------L-----------------'------------------
: : : : 0.00480 0 -12.00 u,
: : : : 0.800 0 v,
I I I I
I I I I 0.4 x JO'
A pin-jointed space truss is supported and loaded as shown. E = 200,000 MPa. Bar
areas are:
2 rn
Aab = 20 x 103 mm2
Aac 30 x 103 mm2
4
=
rn
Aad 40 X 103 mm2
!
=
A0, = 30 x HP mm2
1. Calculate the displacement at a.

a
ip za= -800kN
----+- ------

Pxa=200kN
1
8 rn

_j 2 rn
Develop the three-dimensional stiffness matrix for an axial

force member. Stating Equation 4.34 in the local coordinates
shown and eliminating all irrelevant degrees of freedom,
{ } [_ { '. } { '. }
J
F x1 EA 1 -l u u
[ k' )
= =
Fx'2 L 1 1 U2 Uz
In Equation 5.7 delete all columns except those corresponding

to Fx1' Fyh F,1> Fx2, Fy2, and F,2 a d all rows except t os cor
responding to Fxi and Fxi (compre Example 5.1) y1eldmg
[
J
Ix mx nx 0 0 0
[r)
=
0 0 0 Ix mx nx
5.16, [k) [r]T[k')[r],

=
Using Equation
U1 Vi W1 Uz Vz Wz
1;. lx.mx' lxnx : -1;. -lx'mx' -lxnx'
I
m;. mxnx : -fx.mx' -m;. -mx.nx'
EA n;. : -lx'nx -mx'nx' -n;.

[k)
=
--------------- ----------------------
L : 1;. lx.mx' lxnx'

Sym
1. Displacements. Record direction cosines for,each member. Locate local origin for each mem-
ber at a.
Member x y Z: L 1..- mx' nx'
ab -2 -4 -8 9.165 -0.2182 -0.4364 -0.8729

ac 6 -4 -8 10.770 0.5571 -0.3714 -0.7428
ad 6 2 -8 10.198 0.5883 0.1961 -0.7845
ae -2 2 -8 8.485 -0.2357 0.2357 -0.9428
Record direction cosine products multiplied by AIL. (Values of AIL are in mm.)
Member (AIL) x i;. ;

m . ni.x lx'mx' lx'nx' mxnx
ab 2.182 0.1034 0.4156 1.663 0.2078 0.4156 0.8312

ac 2.785 0.8643 0.3841 1.537 -0.5762 -1.1525 0.7683
ad 3.922 1.3574 0.1508 2.414 0.4524 -1.8101 -0.6034
ae 3.536 0.1964 0.1964 3.143 -0.1964 0.7858 -0.7858
}; 2.5215 1.1469 8.757 -0.1124 -1.7612 0.2103
Referring to the [k] matrix and adding stiffnesses in the usual way, the global stiffness equa
{} { } [
tions for the nonzero dc:;grees of freedom are
Pxa
Pya -
200
600 - 200
2.522 -0.1124
1.147
-l.7612
0.2103
]{ }
u"
Va
P,a -800 Sym. 8.757 w.
Solving for the displacements,
LdJ =Lua Va waJ =L0.1783 2.722 -0.4863J mm

2. Reactions. Using the properties of the element stiffness matrices (see above table), the global
components of the reactions are
-76.4
0.1034 0.2078 0.4156
Ua Va w.
Rxb
Ryb 0.2078 0.4156 0.8312 -152.8
R ,b 0.4156 0.8312 1.663 -305.6
Rxc 0.8643 -0.5762 -1.1525 170.8
-0.5762 0.3841 0.7683 } -113.8
-----------------------
0 1783
Rye
-1.1525 0.7683 1.537 -227.7

- 200 2.722
R,c
1.3574 0.4524 -1.8101 { -470.7

-0.4863
Rxd
=
kN
0.4524 0.1508 -0.6034 -156.9

----------------------- ------
Ryd
R ,d -1.810 -0.6034 21414 627.8
Rxe 6 1964 -.: 01964- -07858- 176.3
-0.1964 0.1964 -0.7858
-176.3
- . - -
R ye
0.7858 -0.7858 3.143

705.5
R,.
800kN
0.486 mm
Displaced
configuration
_/ y
I
I
I
I
76.4
I
I
l; r 117.
R .v f 170.8
x
+305.6 227.7
EXAMPLES.5
Find the direction cosines of the member shown. x
' y
and z are horizontal; y is vertical. The x axis is along
' '
the member. The x y plane makes a dihedral angle
of 30 with a vertical plane through the member axis.
Demonstrate that the matrix of direction cosines [y]
can be generated by three successive rotations.
From the figure on the right y
OA = V102 + 52 + 32 = 11.576
ix = 10/11.576 = 0.8639 ax =COS-I ix = 30.248

r
mx = 5/11.576 = 0.4319 f3x =COS-I mx = 64.411
nx "" 3/11.576 = 0.2592 Ox = cos-I nx = 74.930 x
In the figure at the right, subscripts indicate the position of

each axis at the end of each rotation. Following our sign
convention, p is a negative angle as pictured. y3, y
'
The succesSi.ve rotations are as follows:
1. An angle p about they axis, bringing the x' axis into the x1-y plane. p = -tan-1(3/10) = -16.699.
2. An angle x about the z1 axis, bringing the x axis into coincidence with the x' axis. x = (90 - f3x) = 25.589.
3. An angle w about the x axis, bringing they axis into coincidence with they' axis and the z axis into coincidence with
2
the z' axis. w = 30.
Symbolically we may write
where {x} and {x'} are coordinate vectors and ['YP], ['Yxl and ['Yw] are transformation matrices defining the three rotations
described above. Using the definitions of Equation 5.2,
0
cos w
0
sin w
][ cosx
-sin x
sin x
cos x
O
0
][ cos p
0
0
1

-s n p ]
-sin
0
0.8660
w cos
0.5000
w
0 ][ 0
0.9019
-0.4319
0 1
0.4319
0.9019
][
sin p 0

6. 578
cos p
0.2 73 ]
=
[ -0.5000
0.8638
-0.5019
0.8660
0.4319
0.7811
0.2591
0.3714
0
] 0 1 -0.2873 0 0.9578
-0.0420 -0.4510 0.8915

5.2 LOADS BETWEEN NODAL POINTS
Generally, structures must resist loads applied between joints or natural nodal points
of the system. Three possibilities are shown in Figure 5.4. Loading may range from a
few concentrated loads to an infinite variety of uniformly or nonuniformly distributed
loads.
(a) (b) (c)

Figure 5.4 Loads between nodal points.
One way to treat intermediate concentrated loads is to insert artificial nodes, such as
p and q in Figure 5.4a. The solution then proceeds as in many of the earlier examples.
The degrees of freedom at the artificial nodes are added to the total inventory, and the
necessary additional equations are written by considering the requirements of equilib
rium at these nodes. The internal element forces on each side of nodes such asp and q
must equilibrate the external loads applied at these points. Assigning the identical de
gree-of-freedom designation to the corresponding unknown displacement components
at the two sides of the node ensures the satisfaction of compatibility. This procedure
was used in Example 4.12 and its use was implied in several problems of Chapter 4.
Additional nodes may also be used when the load is distributed as in Figures 5.4b
and 5.4c. Following a process selected or devised by the analyst, the distributed loads
are "lumped" as concentrated loads at suitably selected arbitrary nodes, and the de
grees of freedom at these and the actual joints are treated as the unknowns of the
problem. There are various ways of approaching or obtaining equivalence between
the lumped and the actual loading. In all cases the lumped loads must be statically
equivalent to the distributed loads they replace. For example, the transverse load at
the artificial joint must equal the total transverse component of the distributed Joa.::
associated with that joint.
Use of artificial nodes may be the most expeditious route to a solution that is in
herently approximate but satisfactory for all practical purposes. It is not a very elegant
procedure, however, and the addition of the displacements at the arbitrary nodes to
the unknown degrees of freedom may increase the labor of solution unduly. For these
reasons, it is desirable to have recourse to a rigorous method that eliminates the need
for "artificial" joints.
The approach that is of most general use with the displacement method is one that
employs the related concepts of fictitious joint restraint, fixed end forces, and equiv
alent nodal loads. It will be explained by example. But in preparation for doing this,
it is useful to recapitulate an approach we have already used by illustrating its appli
cation to the continuous beam of Figure 5.5a. The solution of this problem of a system
loaded only at a natural node would start with the writing of Equation 3.6:
{P) = [K]{d) (3.6)
The support and the remaining degrees of freedom would then be grouped as in Equa
tion 3.7:
{[} [_fj_i_ J:'J {t:}

PS
=
Ksf : Kss as
(3.7)
Yf
Pi
ta _,b....- _,c....-----d --X
L
____ ___
I
L>+- L14- L----+i -
(a)
Free
body
diagram
(b)
Moment diagram
(c)
Deflected
structure
(d)
Figure 5.5 Continuous beam-nodal-point load.
For the particular loading, coordinate designations, and support conditions shown, this
becomes
-P Va
0 (}a
0 (}b
0
Ryb
lf- -lKsss]
[-Ksf: (}c
0
(5.19)
Rye 0
R yd 0
Rmd 0
Solution for the unknown displacements, reactions, and internal forces proceeds in the
usual way. The moment diagram and the deflected structure are drawn in Figures 5.5c
and 5.5d.
Now consider the same structure but with a uniformly distributed load of intensity
q in the center span, as in Figure 5.6. This loading, which is between nodal points, will
be treated in two stages and the results summed, as indicated in the sketches of parts
a, b, and c of the figure.
First, as in Figure 5.6a, grant the existence of fictitious external constraints capable
of reducing the nodal degrees of freedom to zero (clamping the joints). The constrain
ing forces, which in this case consist of two direct forces and two moments, are shown
in their positive sense under our sign convention. It should be clear that p:;,c must
q
yr
a YJJl f Hf+.u d
L
___ x
L--1-+-
L- L--- ---+-II
Moment diagrams
Deflected structures
( a ) System fully constrained ( b ) Removal of artificial (c) Resultant system

constraints
+ =
Figure 5.6 Continuous beam-intermediate load.
actually be negative in order to constrain the rotation at point c. It should also be

accepted that these forces are completely independent of the system-they are not
being supplied by the real beam or its real supports. We presume that the solution for
the fixed-end forces is known. In the case shown, the absolute values of the end re
actions and fixing moments are qL/2 and qL 2/12, respectively. Solutions for some other
cases are summarized in Table 5.1. Knowing the fixed-end forces, the internal forces
and deformations corresponding to the assumed constraints can be calculated, as por
trayed in the bending moment diagram and deflected structure of Figure 5.6a.
Now it is necessary to remove the fictitious constraints by applying to the nodes
loads that are equal and opposite to the fixed-end forces and permitting the system to
deform under the action of these likewise fictitious loads. The reversed fixed-end forces
are called equivalent nodal loads. They are indicated in Figure 5.6b of the example
structure by use of the symbol pr:: and the appropriate subscripts. Solving in the usual
way for the displacements, reactions, and internal forces caused by these loads results
in the bending moment diagram and deflected structure of Figure 5.6b.
Last, the total solution is obtained by summing the two parts. In arriving at the free
body diagram of Figure 5.6c, the two sets of fictitious forces have canceled each other
and we are left with the real system in which all the requirements of equilibrium and
compatibility are satisfied. The bending moment diagram and deflected structure of
this part of the figure are the final, correct ones.
There are several ways to represent this physical process algebraically. We must first
recognize that the internal forces and displacements of the fixed-end problem (Figure
5.6a) must be obtained by some means not detailed here, and reserved for addition
,to the solution of the nodal displacement problem (Figure 5.6b) . One way to formu-
Sign convention: All loads and

reactive forces are positive in
the directions shown. 2 a-b 2
R2= [(2b+L)a -( 4 )d ]
L3
Table 5.1 Fixed-end forces.
late the solution to the displacement problem is after the fashion of Equation 3.6,
but with the addition of the vector of fixed-end forces to the right-hand side of the
equation; thus
{PJ = {KJ{.1) + {P'.J (3.6a)
Physically, this states that, in the absence of any nodal displacement, i.e., {.1) 0, {PJ =
would be equal to the vector of fixed end forces, {P'"J. Conceptually, this formation is
useful because it helps to keep clear the distinction between any real nodal loads and
reaction components of {Pl and the f:.ctitious components that comprise {P'l
The support and the remaining degrees of freedom of Equation 3.6a may be grouped
as before:
{1-} lu--
P,
1l 1- f!1{
[ K 1 , K,,.j {.1_, } l P_, }
=
,
+ (3.7a)
For the particular loading and coordinate designations in Figure 5.6, the above may
be written as (compare with Equation 5.19):
0 Va 0
0 ea 0
0 eh qL 2/12
0
Ryb
=
[u-1]
sf I SS
ec
0
+
-qLz/12
--------
qL/2
(5.20)
Rye 0 qL/2
Ryd 0 0
R,,,d j 0 0
lU Chapter 5 Stillness Analysis of Frames-II
Transference of the {PF} vector in any of the above formulations (Equations 3.6a, 3.7a,
or 5.20) to the left-hand side of the equation is the algebraic equivalent of applying the
reversed fixed-end forces as nodal loads in Figmc 5.6b. Hence
(5.21)
where {PE} is the vector of equivalent nodal loads defined in the discussion of Figure
5.6b.
For the illustrative example of Figure 5.6, the statement of the displacement problem
is therefore
0 Va
0 (}a
-qL 2/12 (}b
+qL 2/12
------ -
Ryb - qL/2
--- tKsf: Kss
]
fl -[s_ (}c
0
(5.22)
Rye - qL/2 0
Ryd 0
Rmd 0 J
Solution for the unknown nodal displacements, the real reactions (Ryb, Ryo Ryd,
Rmd), and the internal forces proceeds in the usual way, but with appropriate account
ing for the loads between nodes. Thus, from Equation 3.7a, for /.ds) = 0, we have
f.dtJ =
[Kffr1/Pf - Pj} (3.7b)
and /Ps) =
[KsfHi1tl + {P{} (3.7c)
In determining forces and displacements within the loaded members, one must re
member to add the results of Equation 5.22 to the solution of the fixed end problem.
Formally, this part of the problem can be symbolized by augmenting the element
Equations 3.11 in the same way that Equations 3.6 and 3.7 were modified to obtain
the above global equations. Thus
(3.lla)
These concepts are illustrated in Examples 5.6-5.9. We note that Equations 3.6a,
3.7a, 5.20, and 5.21 are cast in a global analysis form. In accordance with our convention
'of designating forces acting on the ends of elements, we use the symbols {FF} and {FE}
to designate, respectively, vectors of fixed end forces and equivalent nodal loads for
mulated on an element basis. Components of these vectors will be symbolized by FF,
MF, FE, and ME, with appropriate subscripts. As shown in the examples, joint equilib
rium requires that element fixed-end forces appear in the global equations with mag
nitudes and signs unchanged.
Example 5.6 is a straightforward beam analysis problem. In Example 5.7, a simple
planar rigid frame is studied. In forming the stiffness equations, only those parts needed
at each stage are written. Axial deformation effects are included for illustration. The
example shows that there are no restrictions against combining real and equivalent
nodal forces, but it must be done in algebraically proper fashion.
Example 5.8 illustrates the analysis of a plane frame loaded normal to its plane. Such
a structure is often called a grid. The primed degrees of freedom refer to the local
coordinates indicated. The partitioning of the [r] matrix follows Equation 5.7. As in
Example 5.3, much of the detail of matrix formulation and manipulation is omitted.
In Example 5.9, two analyses of a portal frame are simplified considerably by using
two artifices commonly employed in classical frame analysis: (1) suppressing ( equating
to zero) displacements that are known in advance to be very small with respect to the
remainder, and (2) taking advantage of symmetry and antisymmetry to combine co
efficients of related degrees of freedom. The concepts of symmetry and antisymmetry
are discussed further in S,ection 13.6.
Although the devices used in Example 5.9 are used correctly and effectively, a word
of caution regarding the employment of presumed shortcuts in a stiffness analysis is
in order. Unless proper procedures are followed, significant contributions to the stiff
ness of the structure may be overlooked or incorrectly represented in the analytical
model. In some cases solution of the stiffness equations will prove impossible but, in
others, plausible but unrealistic results may be obtained. Problems 5.6 and 5.7 relate
to some of the consequences of improper modeling.
The beam of Example 4.8 is supported and loaded as shown. ! 20kN

J:]IJ 2kN/m J I
2. Calculate the reactions and bending moments.
a b c
---- Sm I2mI 3m
! 20kN
I J I J 2 kN/m J I I l
Fixed end forces:

! 20kN
M{,, = 20xx3 =14.40kNm

2
F= 2x82 =10.67kNm
Mza 5
12 M{c = -20x2 x3 =-9.60kNm
M{,, =-10.67kNm 5
F= 3x20+(14.4-9.6)=12.96kN
Fyb
Ff.= 8.00kN 5 5
FJ',, = 8.00 kN F{c = 7.04kN
1. Displacements. Use Equation 3.6a and the relevant stiffness equations of Example 4.8.
Rya 0.00469 18.75 -0.00469 18.75 0 0 Va 8.00
Pmza 1 x 105 -18.75 0.5 X 105 0 0 8,a 10.67 x 103

0.00949 -6.75 -=-0.00480 12.00 20.96
Ryb vb
pmzb
= 200
Sym. 1.4 x 105 -12.00 0.2 X 105 8,b
+
3.73 x 103
Rye 0.00480 -12.00 Ve 7.04
Rmzc 0.4 X 105 8,c -9.60 x 10'

Let Pmza = Pmzb = O; Va = vb =

vc = 6,c = O; reorder and partition:
6, a 6, b Va vb Uc 6,c
G, a
l
0 1 105 0.5 105: 18.75
X X
I
-18.75 0 0 10.67 103
x
0 1.4 105 : 18.75

----------------
X -6.75 -12.00 0.2 105
, ------------------------------------
X
!lb_ 3.73 103
------
x
----
Ry a 0.00469 -0.00469 0 0 8.00
H
I
Ry b
= 200 Sym
I
0.00949 -0.00480 12.00 +
20.96
Rye 0.00480 -12.00 7.04
Rmzc 0.4 105 X -9.60 103
x
Expand the upper partition and solve for L 6,,, 6,bJT:
{ 1 - 10-
6za 2 [] 0.5]-1{-10.67} {-5.684 10-4} =
X
rad
200 0.5 1.4 -3.73 0.698 10-4
6,b ) x
{ } 2 10_2[-18.:750 75 -12.-6.:757005 ]{-5.0.6684}98 { 20.7.8.090460 }

2. Reactions. From the lower partition.
R ya
R,b
x +
Rye
=
{-9.322:6.87103 }
R,nzc 0 0.2 105 -9.60 103X X
kN
x kNmm
Bending moments. Use the member stiffness equations of Example 4.8.

6,0 6,b
{ ;i}Mab [ 1 105 0 5 X 105 {- X { 10.67 103 X
0.4 105
M = 2 X 10-2 0.5 105 X 105 ]
X : :} -10. 67 103} + X
M'f, 0 x 14.40 103 x
=
{-14.9 103} x kNmm
14.96 103 x
14.96
/f\ ;jkNm
a' c
9.39
11.29
20 kN
5.684 x 10-4 rad 0.698 x 10-4 rad
::'-''-'---'-'-:'.:::==,:::::r=" 9.32 kNm

6.13 kN 23.00 kN 6.87 kN
EXAMPLES.7
The rigid frame shown is made of elements studied in
CT
Examples 4.8 and 5.3.
I 3m
1. Calculate the displacement at b. Include flexural
and axial deformation effects.
>----- 7.416 m --.___ __ 8 m --__.,,! 1
Pby= -46.35 kN Fixed end forces:
x
'
Mb= - Mc = 4i'28 = 21.33 kNm

F{b=F{c=l6kN
1. Displacements. Use Equation 3.8a and stiffness equations for the nonzero degrees of freedom drawn from Examples
4.8 and 5.3.
{ PPyxbb }- {- }
-
18.75
46.35
Pmzb [ 0
= 200
(0.645 + 0.750) (0.259 + 0)
(0.109 + 0.00469)
(7.031 + 0)
(-17.382 + 18.75)
]{uvbb} { + 16
0 }
Sym. (1 + 1)105 (}zh 21.33 X 103
{uvbb } = 0.5 X 10-2

[ 1.394 0.259
0.114
7.031
1.368
]-i{ 18.75
-62.35
} {
=
0.09982 mm
-4.996 mm
}
6,b Sym. 2 x 105 -21.33 x 103 -0.000534 rad
116 Chapter 5 Stiffness Analysis of Frames-Il
2. Reactions. Use member stiffuess equations from Examples 4.8 and 5.3.
ub vb 6,b
{ }
Rxa -0.644 -0.259 -7.031 0
Rya -0.259 -0.109 17.382 0
Rmza
Rxc
= 200
7.031
-0.750
-17.382
0
0.5 x 105
0
0.9982
-4.996 +
0
0
-0.000534
Rye 0 -0.00469 -18.75 16.0
Rmzc 0 18.75 0.5 X 105 -21.33 x 103
131.0 kN
55.4 kN
13.43 x 103 kNmm
-149.7 kN
22.7 kN
-45.41 x 103 kNmm
46.35 kN
13.43 kNm
13 1 0
55.4 kN
The rigid frame (grid) shown lies in a horizontal plane. The loads act
vertically. Member ab has the properties of member be of Example 4.8, z
and member be has the properties of member ab of that example.
1. Calculate the displacement at b.

x
r-- 5 m---->4
Fixed end forces:
)
F
My0-
_ F
-Myb-
_
-
3 X 52
- -6.25 kNm
12 -
_
Ff0 =Ffb = 7.50 kNm

1. Displacements. Following the y,x'
style of Example 5.3 and using the z,y'

results of Example 4.8, develop
the member stiffness matrices in
global coordinates.
z' Mya' a
ab
Member ab
(k'): [rJ:
u'a (}x'a (}z'a u/, (}x'b (},.b
0.0048 0 12.00 -0.0048 0 12.00 1:
_J
0 ______
o:
i
o:o
L__ ______
0
7.692 0 0 -7.692 0 0:1 o
o: : o 0
0.4 x 105 -12.00 0 0.2 x 105 o:I o -1: I
0:I
0 0
200 0-:-0- --a :-1 :o- - - - -a
0.0048 0 -12.00
- ----- - + -- ------
Sym. 7.692 0 o:o 0:0: 1 o
0.4 x 105 o:I o o:I o:o
I
-1
[k):
Wa 6xa Oya wb (}xb (}yb
0.0048 0 -12.00 -0.0048 0 -12.00
7.692 0 0 -7.692 0
0.4 x 105 12.00 0 0.2 x 105
200
0.0048 0 12.00
Sym. 7.692 0
0.4 x 105
Member be
(k'):
u;, (}x'b (}z'b u'e (}x'e (}z'e
I I I
0.00469 0 18.75 -0.00469 0 18.75 110 010 10 0
--L----..L.--..L.----
14.423 0 0 -14.423 0 o :o 1:0 :0 o
1 x 105 -18.75 0 0.5 x 105 0: 1 o: o :lo 0
200 __ L !____ __ ___ _
0.00469 0 -18.75 o :o 0:1 :0 0

--t-----+--+----
Sym. 14.423 0. o :o o :o :o 1
1 x 105 o :I o 0:
I
0:1I
0
[k]:
wb (}xb (}yb -we (}xe (}ye
0.00469 18.75 0 -0.00469 18.75 0
1 x 105 0 -18.75 0.5 x 105 0
14.423 0 0 -14.423
200
0.00469 -18.75 0
Sym. 1 x 105 0
14.423
{ 7.050 }
wh, Bxb Byh:
Assemble global stiffness equations for the nonzero degrees of freedom:
{PmxP,bb}- {-5.000}- 200[0.00949 118.71055 12.00 ]{wb}

X 0 (Jx b +
Pmyb 0 Sym. 0.4 105 Byb 6.25 103
X X
{w(Jxhh} 5.264 10-s[0.4 108 -7.2.5365103 -122.25 103]{ -12.0 50 }

X
X x x
&yb { -22.37 }
=
mm
Sym. 5.974 -6.25 103 x
5.4.913195 10-3
x
x
rad
10-3
rad
2. Reactions. Use member stiffness equations:
R,Rmxaa -0.w00h0480 -7.(J0xh692 -12.B0yb00

310-37 } -6.25 103
7.050
RmR,yca 200 -0.12.000469
=
0 -18.075 0.2 0 105 {4.1-22.
x
95 x +
x
RmxRmycc 18.075 0.5 0 105 -14.0423 5.931 10-3 000

x
x
14.7445
-6. kN
kNmm
-36.215.25103
x
kN
kNmm
-41.-17.94 11103
x
kNmm
kNmm
Torque diagram:
6.45 kNmm
Moment diagram: 41.94 kNm

Deflected structure: 0.017 kNm y

c /(11
t f4t.94kNm
1
S.2S kN
36.21 kNm
14.74 kN
EXAMPLES.9
In the planar rigid frame shown, member ab corresponds to member ab of Ex
ample 5.3, member be corresponds to member ab of Example 4.8, and member I 200 106mm4
de corresponds to member ed of Example 5.3. Neglect axial deformations.
= x
1. Calculate the displacements under a uniformly distributed vertical load of

2 kN/m. Use symmetry to reduce number of unknowns.
Sm
I SO 106mm4
2. Calculate the displacements under horizontal loads of 2.5 kN at b and e. Use
= x
antisymmetry to reduce the number of unknowns.

a
x
---8m-----i
0.
{} [ ]{ } { }
Neglecting axial deformations, vb Ve Assemble global stiffness
4.8 5.3.
= =
equations for nonzero degrees of freedom. Use Examples and
Pxb 0.00480 12.000 0 0 Ub P';b

pmb
200 1.4 105 x 0 0.5 105 x ()b p;;,b
Pxc
=
Sym. 0.00480 12.000 Uc

+
P';c
Pmc 1.4 105 X Oc p;;,c
1. Uniform vertical load. By symmetry, uc -ub and ()c -Ob. Con
0.
= =
{} [ ]{ }
sidering both symmetry and zero axial displacements, ub = uc =
Stiffness equations may be written
0.00480 20 0
200 1. 05 0 0.5 105 b
0 =
0.00480 12.000 0
{ }
Sym.
0 1.4 105 X -()b
+
10.6 103 x
kNmm
-10.6 103 x
Expand second equation and solve for Ob:

0 = 200(1.4 - 0.5)l050b + 10.67 x 103
8b = -0.000593 rad
UO Chapter 5 Stiffness Analysis of Frames-II
(}{cPxblOb, {2.5} [0.u0c480ub 12.0 0 0 0 , 5]{Ubl

2. Horizontal load Letting
=
= and noting that, by antisymmetry,
stiffness equations may be written:
PxPmmcbc 200.5 20 1.4 10 0.00480 10..1452.0 1005 u(}(}bbb

p
= =
Sym.
X X 1Q-
Simplify the first two equations and solve for L ub o b y

{20.5} 20 [102..00 0480 1.192.0 1005] {u(}bb}
=
2.5 kN
x
{uObb} {-0.30.090195 }
=
mm
rad
5.3 SELF-STRAINING-INITIAL AND THERMAL STRAIN CONDITIONS
A self-strained structure may be defined as any system that is internally strained and
in a state of stress while at rest and sustaining no external load. Several examples are
illustrated in Figure 5.7. A bar fabricated overly long and forced into place between
fixed ends is one case ( Figure 5.7a). To ensure tightness in the diagonal bracing of
truss systems, some tension bars may be fabricated short and then drawn into position,
producing self-straining of the type indicated in Figure 5.7b. If, for some reason, a
beam is fabricated with an upward bow and then pulled into place, as in Figure 5.7c,
-----:-1
I
Fabricated length

a
L Assen.l.iled length __jb:L
tJ.L
(a) (b)
JI--- ! -===
A ssembled position
(c)
Figure 5.7 Self-strained structures. ( a ) Compressed fit. (b) Draw of tie rod.
(c) Fabricated bow.
self-straining occurs. Shrink-fit problems, prestressed concrete girders, and cable

stayed bridges, in which the cables are adjusted to produce a bow in a girder that is
fabricated straight, are other examples of self-strained structures. The analysis of all
of these circumstances is within the scope of the approach to be described below. The
state of residual stress that may exist in an unloaded member that has previously been
loaded until part of the system has yielded is another example of self-straining, but it
is not one that is treated in this chapter (Example 10.9 illustrates this type of self
straining).
Temperature changes with respect to the unloaded, stress-free state of a structure
may also cause self-straining. If the natural expansion or contraction of a system under
such temperature changes is resisted in any way, internal forces develop and the system
may be self-strained. If the bar in Figure 5.8 is heated, it tries to expand an amount
!J..L = a.LT, where a is the coefficient of thermal expansion and Tis the temperature
change, presumed to be uniform in the illustration. If this expansion is prevented by
fixed ends, an axial force is transmitted to the bat and it is correspondingly stressed.
aLT=6.L
f.<------L--------f+--i
Figure 5.8 Temperature restraint:
All of the structures in Figures 5.7 and 5.8 have one thing in common, they are
statically indeterminate. It is impossible for a statically determinate elastic structure to
be self-strained. In the statically determinate, pin-jointed truss of Figure 5.9a, an im
posed shortening or lengthening of any bar, from any cause, may be accommodated
by a stress-free change of geometry. Likewise, the right end of the simply supported
beam in Figure 5.9b can settle, or the end can expand or contract thermally, without
any stress change. Although no -stresses are produted in such cases, design require
ments might prompt the calculation of the displacements that occur.
At this point it is important to recall that strain is a kinematic quantity. As defined
by Love (Ref. 5.2, p. 32): "Whenever, owing to any cause, changes take place in the
relative positions of the parts of a body the body is said to be strained." In almost all
of the problems treated earlier, the causative agent was stress. But stress is not nec
essary. Strain may also result from temperature change, for example. As soon as we
recognize temperature change as a source of strain, it becomes clear that we must
b'
(a) (b)
l
Figure 5.9 Displacements of staticai y determinate structure. (a) Simple
truss-result of elongation of bar ab. (b) Simple beam-result of beam
expansion and support settlement.
U2 Chapter S Stiffness Analysis of Frames-II
measure strain from some reference temperature and the geometry of the system at
that temperature. From this it is a simple step to the recognition that any geometrical
configuration of a system that is different from a defined reference configuration and
that is not the result of rigid body motion must be the result of the strain, that is,
changes in the relative positions of the parts of the body.
To illustrate these concepts, we define the assembled (solid line) configuration of
each of the structures in Figures 5.7 and 5.8 as its reference configuration. Each can
be viewed as having undergone a sirain prior to assembly. In the members ab of Figures
5.7a and 5.7b, tor example, the displacements /:,,,L can be regarded as having arisen
because of unit strains /:,,,L/ L. Initial strains can likewise be regarded as the source of
the curved fabricated state of the beam in Figure 5.7c. The thermal expansion of the
member in Figure 5.8 is due to a unit thermal strain aT. The forces required for the
assembly of these initially strained members cause further strains throughout the sys
tem. Hence, the reference configuration is internally strained and in a state of stress
while sustaining no external load; that is, it is self-strained. The strains both prior and
subsequent to assembly must be accounted for in the structural analysis.2
5.3.1 Initial Strain Problems
Structural analysis for the effects of self-straining can be accomplished through an

approach that, in principle, is identical to the one developed earlier for loads applied
between nodal points:
1. We first assume the presence of fictitious constraints at each of the degrees of

freedom of the reference configuration. We then determine the magnitude of the
forces required of these constraints to counteract the source of the initial strain
and maintain compatibility in the reference state with no nodal point displace
ment, that is, in a fixed-end condition. For example. suppose one bar of an in
determinate truss is fabricated short a prescribed amount (Figure 5.7b). The fixed
end forces would be those needed to draw the two ends of the bar into their
compatible, but undisplaced, position in the reference configuration and thus to
permit them to be joined to the other members, so that, subsequently, they can
function as regular force resisting elements of the assembled structure.
2. We next annul these fixed-end forces by applying forces equal and opposite to
them as equivalent nodal loads acting on the assembled structure, and we cal
culate the displacements they cause.
3. The resultant displacements, reactions, and internal forces are, just as in the case
of distributed loads, the sum of those found in the fixed-end analysis and the
nodal displacement analysis (see again Figure 5.6).
The process is illustrated in Example 5.10. Bar ad is fabricated 3 mm too short. What
are the consequences? As seen in the example, large forces and substantial displace
ments are developed, purely as the result of self-straining. Example 5.11, a support'
settlement problem, is an additional illustration. Here the fixed end forces are those
caused by the prescribed settlement with all other degrees of freedom restrained. The
coefficients needed for determining these fictitious forces may be taken directly from
2The view that initial strains are the origin of self-straining conditions in structures is important in the
development of general procedures in matrix structural analysis and will be adopted in Section 7.5. Consis
tent with the approach we have taken in these early chapters, however, we are following a simple physical
approach here.
5.3 Self-Straining-Initial and Thermal Strain Conditions U3
the beam element stiffness matrix as shown in the example. Comparison of this ex
ample and Example 4.11 discloses that support settlement problems can be solved as
either self-straining or specified joint-displacement problems.
EXAMPLE 5.10
The truss shown is the same as in Example 3.3 except that the bar ad is
fabricated 3 mm too short.
1. Calculate the displacement at a and b after ad is joined to the remainder.

E = 200,000 MPa.
><------ 10.928 m ----
Assembled structure:
Fixed end forces (see Example 3.3):

----+.-- 3 mm
I d --- 6.928
a
11+- m ----
F;0 = -F;d = (Et) tH0d = 577.37 x 3

ad = 1732.J kN
{} {} [ ]{ } { }
1. Displacements. Following Example 3.3 and Equation 3.8a
12.122 - -3.536 A1 1732.1

= = 10
2
:: 3.536 A2 + 0
P3 0 Sym. 6.830 A3 0
LA1 A2 A3j = L -1.683 -0.0518 -0.8448J mm

{ l [ {} }
2. Reactions. See Example 3.3.
2
3

{
J
Ryb 3 36 -3.536 -3.536 0
-1.683
{ }
R xc l 2 -2.812 -1.624 -3 294 0
=
0
-1.624 -0.938
-o 0518 +
0
Rye
-o.8448
Rxd -5.774 0 0 -1732.1
-278.1
760.0
= kN
278.1
-760.2
EXAMPLE 5.ll..
The beam of Example 4.8 is supported as shown. The support at b a b c
settles 15 mm, carrying the beam with it. ...0:: :U:

1. Calculate the displacements at
a and b.
>+---- 8 m -------'-5m
Fixed-end forces-Impose a 15 - mm

15mm
vertical displacement at b with no
other joint displacement. Using Ex-
ample 4.8 or Table 5.1, the fixed-end
forces are as follows:
...._-____
"
___
=:--_______
... .L__r::--=----
:U...

---.-------+-
- 15mm
T
Ma =
Mb = 200(-18.75)(-15) Mb =
Mc = 200(12.00)(-15)
56.25 x 103 kNmm -36.0 x 103 kNmm

=
=
F;a =
- F;b
=
200(-0.00469)(-15) F;b =
-F;c
=
200(0.00480)(-15)
14.07 kN -14.40 kN
=
=
1. Displacements. Use the stiffness equations of Example 5.6 (which is the same structure) and allow no further variation
in vb.
{PP'"'"} {O} =
200 x 10
5 [ 1 0.5 ]{ } {
8'"
+
56.25 x 10 :}
0.5 1.4 8,b 20.25 10
=
mzb 0 X
L 8,a 8,bj = L -0.002984 0.000342J rad

{ } [ :: ]{ } { -:: }
(},. e,b
Rya 18.75
Ryb -18.75 -29.84 +
{ }
2
2 x 10-
3.42 14.40
=
Rye 0 -12.00
Rmzc 0 0.2 x 105 -36.0 x 103
4.16 kN
-17.74 kN
-
13.58 kN
-34.63 x 103 kNmm
0.000342 rad 17.74 kN .
5.3.2 Thermal Strain Problems
To treat thermal problems in the way just described, a method for computing the fixed
end forces resulting from temperature change is needed. A general method is given in
Section 7.5. Here, we develop the required relationships for two cases of general in
terest through simple reasoning.
5.3.2.1 Uniform Temperature Change in an Axial Member We consider a prismatic

bar of area A, Young's modulus E, and thermal expansion coefficient a (Figure 5.lOa).
E :-:.j x
,______ ; ___ ___.,1:_ , ,[

__ tli,r=aLT
(a)
0
F/; =-EAaT
l+-----L-------+<
(b)
Figure 5.10 Uniform temperature expansion.
Under a uniform temperature increase T the bar would elongate an amount

D.L = aLT
U6 Chapteli 5 Stiffness Analysis of Frames-II
if permitted to expand freely. But this is the elongation that would be suppressed by
fictitious nodal constraints. For the axial force member, the constraining force would
be
EA
F=-t::..L
L
Combining the two equations and following our usual local sign convention,
F;1 = - F;2 = EA aT (5.23 )
Note that these forces are independent of the bar's length.
5.3.2.2 A Thermal Gradient Through the Depth of a Beam Here we consider a

bisymmetrical beam of moment of inertia I, depth h, Young's modulus E, and expan
sion coefficient a (Figure 5.lla). The entire bottom surface is heated an amount T1
and the top surface an amount Tu, with a linear temperature gradient!::.. T = (T1 - Tu)
in between. If free to expand, the axis of the beam would elongate an amount
aL( T1 + Tu)/2 and, in addition, the beam would curve in a circular arc as in Figure
5.llb. The fixed-end forces necessary to suppress the longitudinal expansion can be
calculated from Equation 5.23, with Tin that formula taken as (T1 + Tu)l2. The fact
that the curvature is circular needs to be proved and then we have to show what forces
are necessary to constrain it. To do this, consider the thermally induced curvature of
an element originally dx long when it is permitted to expand freely. The element is
shown to enlarged scale in Figure 5.llc. The angle change between its two faces is
a(T1 - Tu)dx
de=
h
or
de a(t::..T)
dx h
I+--- L -----><
Al
(a)
(b)
EJ,a(T, - Tu)
[a
I. . I (1 + aTi)dx
(c)
(d)
2
Figure 5.11 Thermal gradient tlirough a beam.
where, here, Tis the difference between lower and upper surface temperature. But,
neglecting second order effects, de/dx = lip, the curvature of the section. Thus
The element selected is typical of conditions all along the span. Thus, since the cur
vature is constant, the beam is bent in a circular arc. To constrain this we need simply
add end moments as in Figure 5.lld. From the differential equation of the elastic curve,
Equation 4.3lb,
Therefore, combining the two equations for curvature, we have, with proper regard
for the signs of the fixed-end moments (see Figure 5.lld in which the constraining
moments are shown in their physically correct directions),
(5.24)
Again, the fixed-end forces are independent of the length.

The methods of thermal analysis are illustrated in Examples 5.12 and 5.13.
Example 5.12 is an example of the thermal-force analysis of a small truss. The primed
forces represent forces in the local coordinate directions, and the unprimed forces
represent components in the global coordinate direction. For each member, the origin
of local coordinates is taken at the a end.
EXAMPLE 5.12
In the pin-jointed truss shown, all bars are cooled 20C. y

a = 1.2 X 10-s mm/mm0C. E = 200,000 MPa.
1. Calculate the displacement at a.

3. Reduce the thermal displacement under uniform temperature change
a x
to zero by suitably altering the areas of bars ac and ad.
L=6m
A=6000mm2
Fixed-end forces-Following Equation 5.23,
F:a = F:b = 0
Fa = -Fb = -F: .
a
= -200 x 6 x 103 x 1.2 x 10-sc-20) = 288.0 kN

F[. = - F[c = - F[ . cos 30
= -200 x 5 x 103 x 1.2 x 10-5(-20) cos 30 = 207.9 kN

F;. = - F;c = F1;. sin 30
= 200 X 5 X 103 X 1.2 X 10-5(-20) sin 30 = -120.0 kN
a Ffa
l Ff.a
F[. = F;. = - F[d = - F;d = F1;. cos 45

F{d
= 200 x 4 x 103 x 1.2 x 10-5(-20)cos 45 = -135.8 kN
F[d
1. Displacements. Record direction angles, direction cosines, and products multiplied by AIL
( see Example 5.4). AIL is in mm. See Figure 5.3 for definition of direction angles.
Member ax' lx' mx' (AIL) z;.(AIL) m;.(AIL) lxmx-(AIL)

ab 270 0 -1 1.000 0 1.000 0
ac 150 -0.8660 0.5000 0.6250 0.4688 0.1563 -0.2706
ad 45 0.7071 0.7071 0.8000 0.4000 0.4000 0.4000
L 0.8688 1.5563 0.1294
Stiffness equations for displacement at a ( see Equation 3.6a):
{ Pxa } { O } 200
[ 0.8688 0.1294 ] {u} { +
(0 + 207.9 - 135.8) }
Pyo
= =
0 0.1294 1.5563 Va (288.0 - 120.0 - 135.8)
{u} { -0.4045 }
= mm
Va -0.0698
2. Reactions. Use negatives of member influence coefficients in the above table.
u. Va
Rxb 0 0 0 0
Ryb 0 -1.000 -288.0 -274.0
Rxc -0.4688 0.2706 {-0.4045} -207.9 -173.8

= 200 + kN
Rye 0.2706 -0.1563 -0.0698 120.0 100.2
Rxd -0.4000 -0.4000 135.8 173.8
Ryd -0.4000 -0.4000 135.8 173.8

For uniform temperature

3. Nullify displacements. Net, thermal, fixed-end forces must be zero.
fixed geometry (truss configuratio n), fixed-end forces are a function of the bar
change and
areas alone (see Equation 5.23). Thus we have
-Aac COS 30 + Aad COS 45 = 0
Aac sin 30 + Aad sin 45 = 6000
-0.8660 Aac + 0.7071 Aad = 0

6000
0.5000 Aac + 0.7071 Aad = 6000
Aac = 4393 mm2 Aad = 5380 mm2
In Example 5.13 the differential heating of a continuous beam is studied. Although

a point of inflection appears in the beam, the bending moment is positive throughout.
Stressing may be thought of as occurring as the beam is strained from the position it
would assume if the thermal deformation were permitted to occur in some unrestrained
way (say, through removal of statically redundant supports).
[EXAMPLE 5.131
The beam is the same as in Example 5.6. The depth of span ab is a b c
400 mm and that of span be is 200 mm. The bottom surface of both mm4 5 mm4 I= 200 106 I= 0 106
spans is heated 10C and the top surface 20C, with a uniform gradient
x x
in between. a= 1 2 X 10-5 mm/mm0C.

.

- sm-----L-5m-
Fixed-end forces (see Equation 5.24):
G--0Ma
F - F 200 x 200 x 106 x 1.2 x 10-5(-10) -
Mza =
Mzb =
- - 12.0 X 10
3
kNmm
400
G--2)
Mb Mc
F - F 200 x 50 x 106 x 1.2 x 10-5(-10)
Mzb =
Mzc = -
- -6.0 x 103 kNmm
200
1. Displacements. Use the stiffness equations of Example 5.6:
{Pmw} {O} =
0
=
200 x 10s [ l
0.5
0.5 ]{ } {
(}'a -12.0 X 103 }
1.4 o,b + (12.0 - 6.0)103
L Ow O,bj =
L0.0008609 -0.0005217J rad
{ } Ry
[
Ryba
200
:
-18.75
0
75
(}zb
18.75
-6.75
-12.00
]{ 0.0008609
-0.0005217
{} }
+ 0
0
0
Rye
{ }
=
Rmzc 0 0.2 x 105 6 x 103
1.27 kN
-2.52 kN
1.25 kN
3.913 x 103 kNmm
3.9 1 kNm
10.17kNm
Stressed configuration
0.00086 1 rad
{
1.27kN 2.52 kN 1.25 kN
5.4 PROBLEMS
As in the previous chapters, it is suggested that, in several of the following problems,

a computer program such as MASTAN2 be used and the results checked manually.
5.1 In the sketches below, the x and z axes are horizontal and they axis is vertical.
In part of the figure the x'y' plane makes a dihedral angle of 75 with the vertical
plane through OA and in part b the xy
a
' ' plane makes a dihedral angle of 90 with the

vertical plane. Determine the direction cosines for the two elements OA.
y y
XA =-4 Q
x
YA =-5
ZA
x
'
(a) x (b)
Problem 5.1
5.4 Problems 131
5.2 Two sets of parallel, orthogonal axes are shown. Six vector compo ents ct at

O', the origin of one set. Develop the transformation matrix hat relates six statically
equivalent parallel components (not shown on the sketch) actmg at 0 to those shown.
y'
y
'
x
X'
Zo
/
z
Problem 5.2
5.3 The two sets of axes shown lie in the same plane. Using the rotation matrix ['Y]
and the results of Problem 5.2, develop a transformation matrix that makes P.0 Py, and
Pmz statically equivalent to Px, Py, and Pmz' Show that it can be expressed as the
product of a translation and a rotation matrix.
Yo
p::
x
Problem 5.3
5.4 The member A is the same as in Example 5.5. Using the results of that example,
plus an extension of the reasoning used in Problem 5.3, construct an equilibrium matrix
[] that relates reactive components acting at 0 parallel to the global axes (these
components are not shown in the figure) to the forces acting at A that are expressed
in local coordinates.
y'
'
z
Problem 5.4
E = constant.
SOkNm b
c
2/ E a
I
' f-- 6m-+ 4m-- 8m-----j

(a) (b)
Prohlem 5.5
5.6 Assume E = constant and A = zero for all members; calculate u, v, and (}for
joint a. Also calculate the distribution of internal forces. Why isn't the answer realistic?
What is the shortcoming of the solution and how can it be corrected? What answers
would one obtain to the same problem using moment distribution? What assumptions
as to deformation are implicit in the moment distribution solution?
I T 2L
I d
Problem 5.6
5.7 Assume E = constant and A = zero for both members. Assemble the stiffness
equations required for the solution of u, v, and (} at joints b and c. Show that this
solution fails in this case. What is the difficulty and how can it be corrected? Why was
5.4 Problems 133
a stiffness solution (albeit an unrealistic one) obtainable in Problem 5.6 but not in this
case?

Problem 5.7
Neglect axial deformation (i.e., assume it to be zero) except in the structures for which
areas are presented. E = 200,000 MPa.
(see Problem 4.12a for properties)
JlITffi$1D1l (b)
1 t kN
SkN/m
60
b .-------l--'---.1 b
cl
L. .,11
2
I
a d _l
0 0
f.-Sm Sm
(c) 2m
(d)
(f)
Problem 5.8
5.9 The structure shown is the same as in Example 5.3. Using a computer, calculate
the displacements, reactions, and internal forces. Solve for (a) axial deformation in
cluded, and (b) axial deformation neglected. What difference does it make? Double
the moment of inertia of the columns, solve, and compare results. Double the moment
of inertia of the beams (with column l's at original value) and do likewise.
Problem 5.9
5.10 Compute the displacements, reactions, and internal forces for the systems
shown.
y
r
BeamsWl6 x 36
ColumnsW 12 x 53
E = 29,000 ksi
L 1ookN
All members:
A =2000mm2
E = 200,000 MPa
Sm
J
(a) ( b)
Members ab, cd
A = 12,000mm2
6 4
Ix= ly = 300 x 10 mm
J = 500 x 106mm4 r
I
2
Members be A = 20,000 mm
6 4 Member abWl8 x 35
Ix= ly = 500 x 10 mm bcW12 x 50
J = 900 x 106mm4 E = 29,000 ksi 18 ft
E = 200,000 MPa
(c) (d)
Problem 5.10
5.4 Problems 135
5.11 Bars ae and ee are heated 30C. Compute the nodal-point displacements and
the bar forces. A = 6000 mm2 all bars, E = 200,000 MPa, a = 1.2 X 10-5 mm/mm0C.
3.5m
4m
I _J b
C-3mj__5m__j
Problem 5.11
5.12 The depth of spans ab and ed is 200 mm and that of span be is 400 mm. The
bottom surface of span be is heated 40C and the top surface 15C, with a uniform
temperature gradient in between; the other spans remain unheated. a = 1.2 x 10-5
mm/mmC, E = 200,000 MPa. Compute the displacements, reactions, and internal
mm4 mm4
forces.
I= 50 x 106
a
m
b I= 200 x 106 mm4 m
c
I= 50 x 106
d

Lsm----
I -8m
Problem 5.U
-- 5m j
5.13 The problem is the same as the previous one except for the struts at b and e.
These are heated 15C in addition to the beam being heated as prescribed in Problem
5.12.
b c d
Smt
a
1
Problem 5.13
5.14 The beam shown is of constant section throughout its length and it is subjected
to a depthwise thermal gradient ii T all along its length. Calculate the displacements,
reactions, and internal forces for (a) L2 =Li, and (b) L2 = 2L1 Compare results.
a b c
I m I
l ___
Problem 5.14
L, ___,.L
._ , - - _J
5.15 (a) It is desired to pretension the system in part a of the figure so that, when
the geometrical configuration shown is obtained, the tension in ab is 50 kN. To what
lengths should the individual bars be fabricated if this is to be attained? E =
200,000 MPa. (b) The problem is the same but, in this case, there is an additional bar
ae, as shown in part b.
11
r
6m
f
Sm
A= 300mm2
T;=50kN
-----1b
(a) (b)
Problem 5.15
5.16 Two steel cylinders are clamped together by a hot driven rivet. The heads are
formed on the rivet at a temperature 1000C higher than the cylinder. The rivet cools
against the resistance of the cylinder. Assume that the rivet is elastic throughout the
cooling process. What are the forces in the system after the rivet has cooled 1000C?
Arivet 500 mm2, Acylinder 4800 mm2 E 200,000 MPa, a 1.2 X 10-5 mm/mC
= = = =
~ ,..._____
._ 200mm--------+<
Problem 5.16
5.17 Given the axial force member stiffness equations in orthogonal coordinates x
and y (Equation 2.5), develop the transformations necessary to transform these equa
tions into the oblique coordinate system defined by the axes 7J and and the included
angle If!. Compare the transformed equations.
y 11
x,
Problem 5.17
REFERENCES
5.1 F. P. Beer and E. R. Johnson, Vector Mechanics for Engineers: Statics, Sixth, Edi
tion, McGraw-Hill, New York, 1996.
5.2 A. E. H. Love, A Treatise on the Mathematical Theory of Elasticity, Fourth Edition,
Dover Publications, New York, 1944.
Chapter 6
Virtual Work Principles
In prior chapters we established the relationships of framework analysis by dealing

directly with the basic conditions of equilibrium and displacement continuity for elastic
structures. That approach has considerable appeal because of its straightforward na
ture, but another method-virtual work-can be exploited for the same purpose as
well as for the development of important analytical tools that are beyond the reach of
the simple physical approach. Virtual work is a general term that includes two distinct
principles, those of virtual displacements and virtual forces.
Virtual work concepts, in one form or another, will have been encountered by any
student who has been exposed to elementary structural analysis. Because our purpose
in dealing with such principles is to establish a basis for general framework analysis,
it is necessary to take a closer look at theoretical questions than is usually the case in
introductory structural analysis. In Chapter 7 we demonstrate the utility of these the
ories in matrix structural analysis. ( References 6.1-6.3 contain further discussions of
the theories and their application. )
This chapter begins with an examination of the concept of work in structural me
chanics, a topic treated briefly in Section 4.2. The notion of a virtual, or imaginary,
displacement is introduced and, by study of the response of a rigid body to this type
of displacement, the most elementary form of the principle of virtual displacements is
established. The usefulness of this form of the principle is limited to the calculation of
support reactions and member forces in statically determinate structures.
The principle of virtual displacements is of value principally in the formulation of
stiffness equations, which imply deformable behavior. Therefore, the concepts devel
oped for rigid bodies are established for deformable bodies in Section 6.2. Detailed
virtual work expressions are constructed for axial, torsional, and flexural elements in
Section 6.3. Although we emphasize the theoretical foundations of virtual work, a
proper appreciation of certain of its features and subtleties can be gained only through
specific examples. Hence, Section 6.4 examines the principle of virtual displacements
as applied to a few classical analysis problems.
The principle of virtual forces is the alternative, or dual, of the principle of virtual
displacements and is the basis of the direct formulation of element flexibility relation
ships. Although the stiffness formulation is emphasized in computerized structural
analysis, there are many cases in which the formulation of flexibility relationships and
their transformation to stiffness terms as described in Section 4.4 is simpler than the
direct formulation of stiffness relationships. The theoretical basis of the principle of
virtual forces is therefore established in Section 6.5. We treat the principle in terms of
its role in element flexibility formulation rather than as an exposition of its historical
place in classical indeterminate structural analysis.
137
138 Chapter 6 Virtual Work Principles
6.1 PRINCIPLE OF VIRTUAL DISPLACEMENTS-RIGID BODIES
To develop the principle of virtual displacements we must amplify the definition of

work given previously in Section 4.2. There, it was stated that the work of a force Fis
{C>.o
W = o F d!l. (6.1)
J
.., -,.::re A is the displacement of the point of application of the force in the direction of
t1e force, with A0 the final amplitude of displacement. This development will be limited
r forces that are gradually applied, so that the displacement grows in proportion to
he force as shown in Figure 6.la. The figure refers to conditions of linear elastic
behavior. This restriction is unnecessary in the development of virtual work principles
in their more general form, but it is adopted here because we deal only with linear
elastic structures in this and the next chapter.
For a given intensity of the load F0 the amount of work produced, W0, is given by
the shaded area of Figure 6.la. If a small increment of displacement, dA, is imposed
through the action of a small increment of force, dF, the change of work is
dW0 = F0(d6.) + 1!2(dF)(d6.) (6.2)
If we discard the second term on the right side, based on order-of-magnitude consid
erations, we have the "first-order" change of work:
dW0 = F0(d!l.) (6.2a)
The same form of relationship is obtained if, as pictured in Figure 6.lb, it is imagined
that the displacement increment occurs without the action of a corresponding force
increment. Such an imaginary or hypothetical displacement is termed a virtual dis
placement, cSA, and the increment of the work of the existing force acting through the
virtual displacement, cSW0 = F0( cSA), is known as virtual work. The virtual displacement,
being imaginary and unrelated to any motivating force, need not be restricted to small
ness but will, instead, be arbitrary in direction and undefined in magnitude. If the
virtual displacement is not small, due account must be taken of this fact in the resulting
formulation. Only small virtual displacements are treated here.
In the following we will use cSA and the like to designate virtual displacements. 1
F F
--------
LM
(a) (b)
Figwe 6.1 Force-displacement relationship.
'In the first edition of this text an overbar was placed on all virtual displacement symbols to distinguish
them from real displacements. Overbars are omitted in this edition because we will use the 5 symbol as an
indicator of virtual quantities and not as a differential operator. Therefore, there should be no danger of
misinterpretation.
Virtual work will be designated as 8W, frequently with a subscript indicative of its
source.
We first consider a single particle acted upon by a system of forces F,, ... , F;, ... ,
F., which are in equilibrium. Denoting the direction cosines between an arbitrary di
rection and the forces as AI> ... , A;, ... , A., the condition of equilibrium of the forces
is L F;A; = 0.
i=l
Suppose the particle is subjected to a virtual displacement M. in an arbitrary direc

tion. The virtual work is then
8W = L F; (M) (6.3)
i=l
where denotes the dot product. Introducing the direction cosines that give the com
ponents of the forces F; in the direction of 8Li this can be written as
8W = F1A1(M) + ... F;A;(M) + ... FsAs(M) + ... ( * )

= F;A; (M) (6.3a)
and, since the bracketed terms that multiply 8Li are zero in accordance with the con
dition of force equilibrium, it follows that 8W = 0. This is the virtual displacement
principle, which can be expressed as follows:
For a particle subjected to a system of forces in equilibrium, the work due to a

virtual displacement is zero.
It is of interest to ascertain whether the virtual displacement principle can be used
to determine the converse, that is, if a system of forces acting on a particle is in equi
librium. This would appear to be difficult, since the virtual displacement is arbitrary,
implying an infinity of directions in which the virtual work can be assessed. Any dis
placement, however, can be defined by components in the independent coordinate
directions, such as three independent components in a general three-dimensional state.
Thus one can establish the conditions of equilibrium in the latter case by writing virtual
work equations for three independent virtual displacements. The converse of the above
statement of the principle of virtual displacements can therefore be written:
A particle is in equilibrium under the action of a system of forces if the virtual work
is zero for every independent virtual displacement.
The above statements apply also to a system of particles under the action of an
equilibrated set of forces, since the virtual work equations can obviously be written
for each and every particle. Consideration of such systems of particles, or rigid bodies,
demands a more careful examination of the description of the virtual displacement.
Consider, for purposes of this examination, the beam pictured in Figure 6.2a. The
member shown is nondeformable; it is supported in a statically determinate manner
and is subjected to a concentrated applied load of Py3 There would appear to be no
scope for application of the virtual displacement principle, since no displacement seems
possible. The support forces can replace the support constraints, however, as in Figure
6.2b. If the conditions of equilibrium are written and solved, these support forces can
be calculated. Then the structure can be represented, as in Figure 6.2b, as a free body
under the action of an equilibrated force system without any consideration of support
conditions. The structures represented in Figures 6.2a and 6.2b are entirely equivalent
from the point of view of structural analysis. Hence, a virtual displacement can be
defined for Figure 6.2b as shown in Figures 6.2c. Since the beam is rigid, this virtual
y, u Py3
x u 3 2
(a)
(b)
Figure 6.2 Rigid body analysis of simply

supported beam by principle of virtual
displacements: (a) Actual structure.
(b) Structure with defined support
(c) reactions. (c) Virtual displacement.
displacement must represent rigid body motion, which is a straight line described by
the equation:
'
av = ( 1 - f ) av1 + f av2 ( 6.4 )
The virtual displacement principle can now be written as
y av1 + Fy2av2 - Py3au3 = o

aw= F1 (6 5)
and, from Equation 6.4 with x = x3,
aw= F1 y av2 - Py3

y av1 + F2 ( 1 - 7 ) av1 - Py3 7 &2
( 6.5a )
= [ Fy1 - Py3 ( 1 - 7 ) J av1 + ( Fy2 - Py3 7 ) av2 = O
Now since the virtual displacements are arbitrary, the terms av1 and &2 can take on
any value, including zero. This means that the multipliers of &1 and av2 in Equation
6.5a must each equal zero. Thus we have
and
P3
y X3
F2
Y =--
Application of the principle of virtual displacement has produced the two relevant
equations of equilibrium of this body. If we had introduced just one support force (F2
y )
in place of the support constraint, then the admissible virtual displacement would lead,
by means of the principle of virtual displacements, to only one of the above two equi
librium equations. On the other hand, if the virtual displacement state were generalized
to include an axial component, then all three equilibrium equations of this rigid planar
structure would be produced.
Example 6.1 demonstrates that the principle of virtual displacements can be em
ployed to calculate directly the force in a member in the interior of a statically deter
minate structure. The force in the member is represented as acting on the joints at the
ends of the member, but the virtual displacement takes place as if the constraining
action of the member itself were absent.
EXAMPLE6.l
Calculate the force in member 3-6 of the truss shown, using the principle of virtual work .
2J r8u5
Replace member 3-6 by the equal and opposite forces F3_6 acting on the joints to which
the member is connected. Assume, for purposes of defining the virtual displaced state,
-----------r:,""9
that the constraining action of the member is absent. Since the member has been ef
I I .-'1 I
fectively removed through this step, the indicated virtual displacement can be enforced. I I .,,,.""' I I
I I
[
,,,,"" I I
By the principle of virtual displacements, I I .,,,
"" I I
I I ,..- .,, I I
P I I_..- I I
1.5 ---;+?' ----------"'1/1
2P( ou,) + P( ou,) -
F3-6 =
F.1-6
2\ITI5P
vTI5 (OU s )
J = 0 I
I
I

/
I
I
I
F3_6
KI I
F3 -61 /
I/
''
_.,...,,,...
-----------
,
-=-
, ,,
I ,.,, I
I .,,,..,,""' I
I .,,,.-' I
I .,,,,,,,,. I
.........
In summary, the principle of virtual displacements can be employed to establish the
equilibrium equations whose solution gives reactions or statically determinate internal
member forces. In either case the reactive or member force is represented, but the
support or constraining action is removed to permit participation of the corresponding
displacement component in the definition of the virtual displacement state.
The principle of virtual displacements, as an approach to the construction of rigid
body equilibrium equations, is of no direct value in matrix structural analysis. Element
stiffness formulations include all of the element degrees of freedom and incorporate
within themselves the element rigid-body equilibrium equations. Nevertheless, the ex

amples given above illustrate the most fundamental application of the principle and
emphasize its relationship to the conditions of equilibrium.
Our interest lies in deformable bodies rather than in rigid bodies because only by
taking into account deformability are we able to calculate the response of statically
indeterminate structures. Therefore, we devote the next section to the extension of
the principle of virtual displacements to deformable bodies.
6.2 PRINCIPLE OF VIRTUAL DISPLACEMENTS-DEFORMABLE BODIES
Application of the principle of virtual displacements to deformable bodies requires

that the total virtual work of a system be clearly defined, which means that distinctions
must be made between external and internal work. The relationships that describe
these terms are developed in the following.
To establish the desired relationships we examine the simplest arrangement of struc
tural elements, a pair of axial members connected in series as shown in Figure 6.3a.
The right end (joint 3) is fixed, and loads Pi, P2 are applied to joints 1 and 2. The
stiffnesses of the axial members are designated as k1 and k2
Figure 6.3b shows the free body diagrams of the joints and elements. All forces are
initially shown as acting in the direction of the positive x axis. The internal forces
exerted by the joints on the members are designated as F12, F2i, F23, while primes are
employed to distinguish the action of the internal forces on the joints. The force on a
joint and the counterpart member force must be equal and opposite. Thus
F12 =
-F;2
F21 =
-F!2.1 (6.6)
F23 =
-F3
Also, the condition for equilibrium of element 1 is
F12 + F21 =
0 (6.7)
or, in view of Equations 6.6,
(6.7a)
The conditions of equilibrium of joints 1 and 2 are, respectively,
P1 + F;2 = 0 (6.8a)
P2 + F1 + !23 =
0 (6.8b)
p k1
--41-
1 x,u 3
F--;-+-
21 F-;-+-
2:1
(b)
Figure 6.3 Two-element axial structure. (a ) Actual structure. (b) Free-body
diagrams of joihts and elements.
6.2 Principle of Virtual Displacements-Deformable Bodies 143
Now consider the effect of virtual displacements l>u1 and l>u2 of the joints 1 and 2.
Work is a scalar quantity so that the total virtual work of the system of joints is simply
the sum of the virtual work of the component parts. In this case, therefore, the total
virtual work is
(6.9)
= (P1l>u1 + P2l>u2) + (F;2l>u1 + F1l>u2 + F3l>u2)
The first term on the right side of Equation 6.9 represents the virtual work of the
applied loads and is designated as l>Wext that is
(6.10)
The second term on the right side is the virtual work of the internal forces acting
on the joints. By substitution of Equations 6.6, this term can be transformed into one
expressed in terms of the forces exerted by the joints on the member, that is
(6.11)
The expression within the parentheses on the right represents the virtual work of the
2
internal forces acting on the members, or the internal virtual work, l>W;n1 Hence in
the present case
l>Wint = (F12l>u1 + F21l>u2 + F23l>u2) (6.12)

Furthermore, in view of Equations 6.9 through 6.12, for the conditions under study,
l>W = l>Wext - !)Wint

By application of the above procedure to any other type of deformable structure, it
can be shown that this is a general expression for the total virtual work. To establish
the properties possessed by this expression it is only necessary to return to the earlier
statement of the total virtual work (Eq. 6.9) and substitute into the second term on
the right (i.e., -l>W;n1) the equilibrium relationships given by Equations 6.8a and b.
We then have
or
l>W = l)Wext - !)Wint = 0 (6.13)

This is the algebraic statement of the principle of virtual displacements for deformable
structures. Again, although it has been established for a special case, its validity can
be confirmed by application of the above procedure to any other type of structure. It
is completely consistent with the statement of the virtual displacements principle of
the previous section. It is now necessary to recognize that it encompasses a system of,
particles and the virtual work is the total virtual work, both internal and external. The
verbal statement ofEq uation 6.13 is as follows:
For a deformable structure in equilibrium under the action of a system of applied

forces, the external virtual work due to an admissible virtual displaced state is equal
to the internal virtual work due to the same virtual displacements.
Observe that the qualifying term, admissible, prefaces virtual displaced state. One
would expect that there are some limitations on the choices that can be made of the
2This quantity. may be viewed as energy stored within the member as a result of the virtual displacements.
For this reason it is also called the virr11a/ strain energy. 5U. a designation that was used in the first edition
of this text.
virtual displacements, and it is indeed the case. The term admissible is employed to
emphasize this consideration. We define these limitations in the next section, where
the method followed in using the virtual displacements principle is outlined, and virtual
work expressions are given for the specific modes of structural action with which we
are concerned.
6.3 VIRTUAL DISPLACEMENT S ANALYSIS PROCEDURE

AND DETAILED EXPRESSIONS
6..1 General Procedure
Although the stiffness properties of the elements in Section 6.2 were defined at the
outset, no use was made of them in the development of the principle of virtual dis
placements for deformable bodies. This circumstance can be used to advantage in
certain situations, but the use to which the principle of virtual displacements will be
put in this text requires the introduction of the structural stiffness properties. They are
introduced to transform the internal virtual work from an expression written in terms
of forces and virtual displacements into an expression in terms of displacements and
virtual displacements.
As an example, in the structure of Figure 6.3a, we have the following relationships
between the member stiffness and the joint forces and displacements:
F12 = k1(U1 - Uz)

F 23 = kzUz
Hence, by substitution in Equation 6.12,
8Wint = k1(U1 - U2)8u1 - k,(u1 - Uz)DUz + kzUzDUz

This is an expression in terms of the displacements (ui. u2) and the virtual displace
ments (8ui. 8u2).
It is tempting to refer to displacements such as u1 and u2 as actual displacements.
All of the developments thus far have either explicitly dealt with or implied a proce
dure leading to the solution that is exact within the limits of linear elastic behavior.
For these cases the displacements are indeed the actual displacements of the linear
system. However, the principle of virtual displacements finds its most powerful appli
cation in the development of approximate solutions. In these, the state upon which
the virtual displacement is imposed may itself be an approximation of the actual dis
placed state (see Section 6.4.2). Thus, rather than calling them actual displacements,
we will use the adjective real-in its sense of something rooted in nature-to distin
guish between tangible and virtual displacements.
' The character of both the real and virtual displacements, and the limitations imposed
upon them, deserve careful attention. As emphasized previously, the solution to a
problem in structural mechanics is exact if it meets all relevant conditions of equilib
rium and displacement continuity. The foregoing development of the principle of vir
tual displacements gives explicit attention only to the condition of equilibrium. The
condition of displacement continuity, which requires that the displacements are con
tinuous functions of the coordinates of the structure (x, y, z), must be met implicitly.
That is, the real and virtual displacements must be of a form such that the displacement
3
continuity conditions are satisfied from the outset. Admissible displaced states are
those that meet these conditions.
'It should be noted that this requirement still permits the virtual displacement of supports and joints as in
Figure 6.2 and Example 6.1. provided that the supports are represented by support-reaction forces rather
than by the actual constraints against displacement.
6.3 Virtual Displacements Analysis Procedure and Detailed Expressions 145
With the above in mind, an outline can be given of a general procedure of structural
analysis based on the principle of virtual displacements. One begins with a description
of the real displaced state by means of admissible functions that have undetermined
multipliers. The virtual displaced state is similarly described, except that the chosen
admissible functions have arbitrary multipliers (e.g., Bu1 and Bu2 in the development
above). These functions are, in turn, used in the construction of the terms BWext and
BWint Equating BWext and BWint produces conditions that enable the evaluation of the
undetermined multipliers of the real displaced state. These values are such that the
conditions of equilibrium are satisfied. It will be found that the arbitrary multipliers
may be canceled from the resulting expressions or that they may lead to conditions
that can only be satisfied by certain relationships among the real quantities of t-he
problem.
The tasks that remain before this general procedure can be applied to specific prob
lems are the formulation of explicit expressions for the internal virtual work BWint and
the external virtual work BWext for the various modes of structural action.
6.3.2 Internal Virtual Work
The primary modes of structural action with which we are concerned are axial, tor
sional, and flexural behavior. In the following discussion, we develop relationships for
BWint for members of length L sustaining these modes of behavior.
Consider first an axial member, a differential segment of which is shown in Figure
6.4a. The stress ax that acts upon the segment throughout the virtual displacement is
the stress corresponding to the real displacement. The virtual displacement of the left
end of the segment is designated as Bu. The virtual displacement will, in general, change
from one point to the next along the segment so that at its right end it will be, to the
first order in dx, Bu + [ d( 5u )/dx ]dx.
To demonstrate the relationship between virtual displacements and virtual strains
we will first treat the differential segment as a free body. The real forces at its ends
(a)
z
M_
(c)
Figure 6.4 Virtual strain conditions. ( a ) Axial behavior. (b) Torsional behavior. (c) Flexural
behavior.
are Fxo with a negative sign assigned to the force on the left face, since it acts in a
x direction. The virtual work of the force on the left
direction opposite to the positive
-Fxou and that of that force on the right face is Fx{ou + [d(ou)!dx]dx).
face is therefore
Thus for the segment as a free body, oWext -ouFx + {ou + [d(ou)!dx]dx)Fx
= =
[d(ou)!dx]Fxdx and, since oWext oWint from Equation 6.13,

=
d(ou)
oWmt Fdx (6.14a)
x
=
dx
Now, axial strain is defined as the rate of change of the axial displacement with respect
to the axial coordinate; that is, for linear behavior ex du/dx. In the present case
=
we have a virtual axial displacement ou. Defining the virtual strain 8ex in an analo
gous manner, we have 8ex = d(ou)!dx. Also, Fx uxA. Thus, Equation 6.l4a can be
=
written as
(6.14b)
This relationship pertains to a differential segment dx. For a complete axial member
of length L, it follows that the change in internal virtual work due to virtual displace
ment is
(6.15a)
Finally, by use of Hooke's law (ux = Eex), this can be written entirely in terms of
strain:
(6.15b)
Since the real and virtual strains are the derivatives of the real and virtual displace
ments [e x
=
duldx, Oex = d(ou)/dx], the internal virtual work can also be written as a
[]
function only of the real displacements and virtual displacements:
(L d(ou) du
oWint = Jo EA
dx
dx (6.15c)
This means that the internal work can be calculated if expressions are availabie for
the real and virtual displacements, as will be the case in the work to follow.
In the case of torsion, the "strain" {3 is the rate of change of rotation of the cross
section about the longitudinal axis (see Section 4.5.2). If, in the differential segment
of a torsion element, Figure 6.4b, the left face undergoes rotational displacement Bx,
the right face is displaced Bx + (dBxldx)dx, and the rate of change of rotational dis
placement is
(6.16)
Correspondingly, the torsional strain due to a virtual displacement in torsion is
d(oBx)
o{3 = (6.17)
dx
The change in internal virtual work due to a virtual twisting displacement in the
presence of a real twisting moment Mx is, in an element of length L,
(L r d(oBx)
oWint = Jo (o{3)(Mx)dx = Jo Mxdx (6.18)
and, since Mx GJ(d O)dx), this can be written entirely in terms of displacement
qualities:
. f L d(88x) d()x
8Wm1 = GJ dx (6.18a)
0 dx dx
In examining beam flexure we consider a member whose axis is coincident with the
x axis, has a cross section symmetric about they axis, and is subjected to bending about
the z axis. As explained in Section 4.5.3, the curvature d2u!dx2 = K0 can be regarded
as the strain for bending behavior. Figure 6.4c, which describes the displacements of
a differential segment of a flexural member, is further illustration of this point. The
deformation of the segment is characterized by the rate of change of rotation Oz with
respect to the x coordinate. The rotation of the cross section is equal to the slope of
the neutral axis, 8, = duldx.4 The rate of change of 8, with respect to x is, then
l_
dx
[(e '
+
d(),
dx
dx ) () J
-
'
=_!-
dx
[( du
dx
+
d2u
dx2
dx ) -
du
dx
] =
d2u
dx2
= K.
-
(6.19)
It follows that the virtual bending strain l'iK, is, for the virtual transverse displace
ment 8u,
(6.20)
It was also explained in Section 4.5.3 that the "stress" in flexure is the bending
moment M,. Therefore, in an element of length L, the internal virtual work due to
virtual displacement of the flexural member is
( (' d2(8u)
8W;n1 = Jo (8K, )(M,)dx = Jo - M,dx (6.21)
dx2
Also, since the moment is related to the curvature by M, = EI,(d2uldx2) (see

Eq. 4.31b), this can be written as
(6.21a)
A comment is in order regarding the conditions of continuity in flexure: The trans

verse displacements, u and 8u, must of course be continuous. The flexural deformation
state is also characterized by the rotations 8, and 8 8
equal to the slopes of the elastic axis, duldx and d(8u)/dx. Hence, in order for the real
, which, as we have noted, are
and virtual displacement states to be admissible, they must also possess continuous
first derivatives. Furthermore, the displacement boundary conditions in flexure in a
given problem might involve not only u and 8u but also duldx and d(&)/dx.
It is useful in the work that follows to have available a general formula for SW;n,
applicable to any or all of the above cases and also to problems of structural mechanics
wherein the state of stress might be two- or three-dimensional. A general designation
of the state of stress at a point in a structure is in the form of a column vector l<TJ,
which includes all components of stress that are present. For a three-dimensional state
T
of stress, for example, we have l<TJ = L ux O'y u, Txy Tyz Tzx J . Similarly, the vir
tual strains can be designated by a row vector L 8e J which, for a three-dimensional state
of strain is L&x &y &0 l'iYxv 8y,,0 8y0xJ. Now, the internal virtual work per unit
volume of a multiaxial stress state is given by &., O'x
+ & y O'v +
, which can be
...
"As noted in Chapter 4, this statement implies neglect of deformation due to transverse shear.
written as L<5ej{<T}. For a structure whose volume is symbolized by (vol), the internal
virtual work is, therefore,
8Wint = f vol
L<5ej{<T)d(vol) (6.22)
The symbolism of Equation 6.22 is consistent with each of the definitions given
above
for the cases of axial, torsional, and flexural behavior. For axial behavior, for
example
L<5eJ = &x. {<T) = O',., and d(vol) =A dx, yielding 8W;n1 = L

L &x O'xA dx,
as before.
The principle of virtual displacements in the formulation of stiffness equations re

quires that 8W;n1 be expressed solely in terms of strain. For this purpose one can
introduce the stress-strain relationships {<T) = [E]{e) into Equation 6.22, resulting in
8W;n1 = f vol
L<5eJ[E]{e}d(vol) (6.23)
where [E] is the matrix of elastic constants. For framework elements [E] is merely the
scalar value E, or in the case of torsion, /2(1 + v) .
6.3.3 External Virtual Work
The calculation of the external virtual work 8Wext is a simple matter in the case of
concentrated loads and can be represented symbolically as
i=I
where 86.; refers to the virtual displacement of the ith degree of freedom and P; is the
applied load in that degree of freedom, in a situation in which loads are applied to s
degrees of freedom. For applied moments, of course, the virtual displacement is an

angular displacement 88x; (or My; or 88,; ) and the applied loads are moments, Pmx (or
Pmy or Pm,).
Note especially those cases in which the virtual displacements include rigid body
motion to enable the construction of equations for determination of support reactions.
This was illustrated in Section 6.1 and will again be illustrated in the next section. In
such cases, the reaction at the support that is released to enable rigid-body virtual
displacement must be designated and included in the expression for 8Wext
Distributed load cases also deserve attention. Denoting the distributed loads as q
and combining with the virtual work of concentrated loads,
8Wext = f M. qdx + (M;)P; (6.24)
where the limits of the integral are taken to define the portion of the member on which
the load acts. Use of Equation 6.24 is illustrated below in Example 6.5 for a case of .
flexure. It is of central importance in the exploitation of virtual work concepts in the
formulation of fixed end forces described in Section 7.5.
It was observed in Section 6.2 that work (and therefore energy) expressions produce
scalar quantities. Thus, when an element is subjected to the combined action of axial
load, torsion, and flexure, the change in internal virtual work due to virtual displace
ment is simply the sum of the <5W;n1's for the respective actions. Correspondingly, the
change in external virtual work is simply the sum of oWext for the component external
actions. This point is brought out in Example 6.2.
EXAMPLE6.2
y,v
(FxA, 2),ly, (My2 M,2(Mxi),

A prismatic member of uniform cross section, with geometric
properties
axial load
J,
a twisting moment
and ) shears (Fv2 F,2),
and I, is fixed at one end and subjected to an
and
at the opposite end. The
and bending moments
displacements that result from the application of the forces are
as given below. The virtual displacements chosen are of the same
general shape. Construct the expressions for the internal virtual
work in the member due to axial, torsional, and simple flexural
behavior. z, w
The real displacement state can be described by the following functions.'
v [3(Y - 2(trJv2 x[(Y - tJe,2

=
+
w [3(Y - 2(trJw2 - x[(Y - tJey2

=
In accordance with the problem statement, the virtual displacements are written in the same
form:
OU =
x 2 08x x 08x2
OU =
L L
OU = [ 3(zY - 2(trJou2 x[(Y - z]oe,2

+
OW = [3(zY - 2(trJow2 - x[(Y - z]oey2

In this case, the desired quantity is given by the sum of Equations 6.15c, 6.18a, and 6.2la, plus
Equation 6.2la written with respect to bending about they axis:
oWint f [Ld:)]EA(:)dx f [ d:x)

=
'- ]a1(;)dx +
J, d2(ou)]E1 d2u)dx J, d2(ow) 1 (d2w)dx

[ dx2 '(dx2
+
o
[ dx2 ]E dx2 +
o Y
o8x, ou, ow), (u, e... u, w,

Upon performance of the indicated differentiation on the given displacements
substitution into this expression, and integration, we have
OU,
oWint ou2(E:)u2 Mx2(a;)ex2 ou2(1!,)u2 - ou2(6::')0,2

= +
+
5The derivation of these expressions is presented in Section 7.1.

6.4 CONSTRUCTION OF ANALYTICAL SOLUTIONS BY THE PRINCIPLE

OF VIRTUAL DISPLACEMENTS
6.4.1 Exact Solutions
We now have available the general statement of the principle of virtual displacements
(Eq. 6.13) and the ingredients of the terms that comprise the virtual work. The purpose
of this section is to illustrate the use of the principle. The approach taken is one that
seeks "analytical" rather than "numerical" solutions. The structures studied are
treated as a whole, instead of being subdivided into segments or elements. The ana
lytical approach enables us to identify key theoretical aspects of the principle of virtual
displacements that require clarification or have escaped identification in the foregoing
development of the principle.
Application of the principle of virtual displacements leads directly to expressions
for loads in terms of displacements, that is, stiffness equations. These expressions are
not in themselves solutions to problems, except in cases where the displacements are
specified and the loads are to be calculated. Generally, when the expressions are ob
tained, they must then be solved to give the displacements in terms of the specified
loads. The adaptation of the virtual displacements principle to the formulation of
framework element stiffness equations and to the construction of global stiffness equa
tions is given in the next chapter.
As a first example of a case in which an exact solution is possible, consider the
uniform axial member of Figure 6.5a. The element is fixed at point 1 and subjected to
a load Fx2 at point 2 where the displacement is u2 We know that under these circum
stances the displacement varies linearly between points 1 and 2. Therefore, we adopt
the form of the actual displacement to describe the real displacement u:
x
u= Uz
L
This gives a strai' n:
du u2
ex = dx = L
Now, the construction of the expressions for 8Wint and 8Wext requires the definition
of the virtual displacements and corresponding virtual strains. One convenient way
to define these is to use the same form as that of the real displacements. (Alter
native modes of description are discussed later.) Thus we choose Bu= (x!L)8u2, and
[d(8u)ldx] = ou2/L.
Figure 6.5 Uniform axial member. (a) Member. (b) Virtual

displaced state.
6.4 Construction of Analytical Solutions by the Principle of Virtual Displacements 151
Writing now the expression for 8Wint for the axial member (Equation 6.lSc)
8Wint J(Lo (d(8u) )EA du dx (L 8u2 EA U2 dx

dx dx Jo
=
=
L L
and, since 8ui. u2, E,

A, Jtdx
and L are all constant within the member and
=
L
8Wint (8u2) ( E:) =

Uz
Also
8Wext (8u2)Fx2
=
Equating 8Wint and 8Wext consistent with the principle of virtual displacements,
(8u2)u2 (E:) (8u2)Fx2 =
and, since 8u2 is arbitrary (it can take on any desired value), it follows that
Uz
ELA Fx2 =
This is a stiffness equation. Solving for the displacement in terms of the applied load,
L
Uz
=
Fx2 EA
which is the well-known exact result for this problem.
The virtual displacement met the same boundary condition (constraint of point 1)
as the real displacement. Suppose the virtual displacement is chosen of a form that
does not meet this condition. For example, if we choose
8u ( i)8u1 i 8u2
=
1 - +
we have a virtual displacement that still varies linearly but also involves displacement
of point 1. The new virtual displacement parameter 8ui, which has been introduced,
permits rigid-body virtual displacement. As Section 6.1 and its illustrations disclosed,
such a virtual displacement of a support point requires replacement of the support
designation by the support-reaction force designation Fxi (Fig. 6.Sb . We now have )
\Wext = Fxl 8U1 + Fx28u2
Also, since now
I
d[ ( i)8u1 (i)8u2 J
1 +
8u1 8u2
&x =
--
dx
---
- -
L
+
we have
8Wint =
(L
Jo ( 8u, 8u2)EA dx
-
L
+
L
Uz
L
-(ou1)u2
EA (8u2)u2 EA +
L L
Equating l>Wext and 8Wint results in
and, because of the arbitrary character of ou1 and ou2,

EA
f,2 Uz
L
=
EA
Fx1 -u2 -Fx2
L
= =
The second equation is the equation of overall equilibrium of the element.

We find, therefore, that the virtual displacement is not required to meet the same
support conditions as the total displacement. If it does not meet these conditions,
however, the pertinent support forces must be introduced and will appear in the ex
pression for oWext Application of the principle of virtual displacements then estab
lishes, for these terms, equations of equilibrium of the overall structure.
The real displacement of the above example was constrained by support conditions
and the virtual displacement was not. The extra degree of freedom of the latter led to
a separate statement of the rigid-body equilibrium conditions. In the formulation of
element stiffness equations, as we observed in Section 2.4 and subsequently, it is de
sired that all joint displacements be represented (no support constraint) and that the
rigid-body equilibrium conditions be contained within these stiffness equations. It can
be surmised that the proper approach to take in such cases is to define unconstrained
representations of both the real and virtual displacements.
Reference was made earlier to the notion of employing virtual displacement states
of a shape different from the real displacement state. For example, consider in the
above the use of a function that meets the same boundary condition as the real dis
placement but is of quadratic form:
OU =
(f) \ OUz)
' then
In this case,
which is identical to the expression obtained previously.

Still another form for the virtual displacement for this case is
for which the strain is
d
- (ou) - (8u2)
dx
=
1T
2L
cos
1TX
2L
In this case,
= (8u2)EA
Uz f 1T
cos
1TX
2L
dx
L 0L 2L
EA
= (8u2)u2 L
which is again identical to the expression obtained when the virtual and real displace
ments are of the same form.
The above two forms of virtual displacement were applied to a real displacement
that was of the form of the exact solution. Not only were the real displacements con
tinuous but the stresses associated with them satisfied the condition of equilibrium.
Although it is worthwhile to assign a physical meaning to virtual displacements, an
other view that can be taken is that they are simply arbitrary multipliers of the equi
librium equations. Reexamination of the development of Section 6.1 and 6.2 should
confirm the latter view. Hence, if the chosen real displacements correspond to stresses
that satisfy identically the conditions of equilibrium, any form of admissible virtual
displacement will suffice to produce the exact solution.
6.4.2 Approximate Solutions and the Significance of the Chosen Virtual Displacements
Different forms of admissible virtual displacement generally produce different solu

tions when the real displacements are themselves approximations to the exact displace
ments. Recognition of this fact is important because the customary practical formu
lations of stiffness relationships for more complex phenomena in frameworks (e.g.,
taper, nonlinearity, instability, and dynamics) and for nearly all finite element repre
sentations are based on real displacements that are approximations.
This point can be illustrated quite simply by assigning a taper to the axial member
of the above problem and solving for the end-point displacement u2. The taper
(Fig. 6.6) is given by A A 1 (1 - x/2L). Because of it, the stress and strain vary with
=
the axial coordinate, and the axial displacement is no longer accurately described by
a linear function u xu2/L. The linear function gives constant stress, ux
= Edu/dx = =
(ElL)u2, which violates the condition of equilibrium,6 dF)dx d[ uxA 1 (1 - x/2L)]/

=
dx = 0. Nevertheless, to illustrate the significance of the principle of virtual displace

ments when the real displacements are approximate, we employ u xu2/L. =
First, if we use a virtual displacement of the same form as the real displacement
(8u x8u2/L) and consider that the area A of Equation 6.15c is now a variable, we have
=
(L 8u2 ( ) (Uz) ( x ) 3EA I

8W;n1 = Jo L E L A1 1 dx (8u2)(u2) 4L
2L
- =
Figure 6.6 Tapered axial member.
6The equation of equilibrium for axial behavior is given in Section 6.5.1, Equation 6.25.
and since 8Wex1 is still (8u2)Fx2, the principle of virtual displacement gives
3u2EA1
= Fx2
4L
2
Next, we try 8u (x/L) (8u2)
= and retain u = (x/L)u2 This results in 8 Win1
2(8u2)(u2)EA,13L and
2u2EA1 _
- Fx2
3L
Finally, for (8u) = (8u2) sin(1Txl2L) and u = (x/L)u2, 8 Win1 = (1 - 111T)(8u2)(u2)EA1/L
and
Clearly, three different solutions have been obtained for three different choices of
virtual displacement. The exact solution is 0.72lu2EA1/L = Fx2 (see Section 7.3). None
of the above represent the exact solution, but each is a valid approximate solution.
The explanation of why these different solutions are obtained is as follows. The
statement of the principle of virtual work, given previously, implied that the solution
would be correct for 8Wint 8Wext What, in fact, occurs here is that both 8Wint and
=
8Wext are approximate because of the approximate nature of the chosen real displace
ment function. The condition 8 Wi n1 8 We xt is then enforced, providing a basis for the
=
calculation of the undetermined parameter (u2) of this function. Although the ap

proximate real displacement function cannot give exact satisfaction of the equilibrium
conditions at every point in the structure, it can be shown that the enforcement of the
condition 8 Wint = 8 Wext results in a weighted-average satisfaction of the equilibrium
conditions throughout the structure. If more terms are employed in the approximation
of the displacement, the resulting weighted averages come closer to the correct solu
tion, provided the proper conditions on a mathematical series are met.
If different solutions are obtained for different choices of virtual displacements for
given approximate real displacement, what choice shouid be made? The standard pro
cedure is to adopt a virtual displacement of the same form as the real displacement. This
proves convenient in the integration of the resulting expression for 8Wint Furthermore,
alternative choices may bring about unsymmetric forms of the resulting stiffness equa
tions, in violation of the reciprocal theorem (Problem 7.12).
6.4.3 Examples
Examples 6.3-6.5 amplify the considerations established here. In Example 6.3 the
internal virtual work established in Example 6.2 is supplemented by the virtual work
of the applied loads, and the principle of virtual displacements is invoked to produce
relationships between the applied loads and the displacements at their point of appli
cation. The example serves to demonstrate application of the principle to cases in
volving combmed axial, torsional, and flexural behavior.
Construct, using the principle of virtual displacements, the expressions for the displacement of
the tip (point 2) of the cantilever beam of Example 6.2.
To solve this problem we need expressions for 8W;n1 and 8W0xi. 8W;n1 was established in Example
6.2. For 8W0x,:
Equating this to the expression for8W;nt found in Example 6.2, we have:
For independent8u2 (8v2, ...88,2 = 0)

EA
Fx2 = L U2
For independent8u2
12/, 6EI,
Fy2 = -U v2 - 8,2
U
For independent8w2
For independent Mx2
For independent88y2
For independent88,2
6EI, 4El,
M,2 - 7 V2 + 8,2
L
=
The conditions of equilibrium of internal and external forces can be established by replacement,
in the virtual work expression, of the terms such as E(du/dx)A by the corresponding forces (in
this case, Fx) Overall equilibrium equations for this member can be established by including
the support reactions (at point 1) in the analytical model and by imposition of displacements
and virtual displacements in the form of rigid body motions.
Example 6.4 describes how the principle of virtual displacements can be used to
advantage in the analysis of a statically indeterminate truss structure where all mem
bers are joined to a single point. The resulting pair of equations is coupled and must
be solved simultaneously. Although in this simple case the solution is easily obtained,
in general the virtual displacements principle is not advantageous in hand-calculation
analysis of statically indeterminate trusses. This is because of the complexity of defining
the external and internal work resulting from a system of arbitrary joint displacements.
[EXAMPLE 6.41
Calculate the displacement of point 4 and the member forces through the
application of the principle of virtual displacements.
2
E = 200,000 MPa. All areas: 15 x 104 mm
f.-- 16 m -____,.. 12 m ------j

Form the internal virtual work due to virtual displacements Bu and

OV4. This first requires the calculation of the elongation of me bers
A, B, and C due to the actual and virtual displacements of point 4.
For element A (see figure),
U4 aA V4 sin aA
+
COS
eA
LA LA
(a)
OU4
cos aA +
OV4
OeA = sin aA
LA LA
For element B,
"- - OV4
oea - (b)
La
For element C (same sketch as element A),
U4
cos ae + -
V4 .
ee = - sm ae
L e Le
-
(c)
Bu. 8v4 .
&e = - cos ae + sm ae
Le Le
Thus
8W;n1 = L f(&CT)d(vol.) = L ( 8eCTvol.)
= &AeAEALA + &aeaEALa + &ee eEALe
,, EA ,, EA . 2 ,, EA .
cos aA + uv4v4 sm aA + uu4v4
2
= uU4U4 sm aA cos aA
LA LA LA
,, EA . EA EA
+ uv4u4 8v4v4 Ls
2
8u4u4 L cos ae
+ + e
sm aA cos aA
LA
.
,, EA EA . EA .
+ uV4V4 ae + OV4U4 cos ae sm ae + 8u4v4
2
sm cos ae sm ae
Le Le Le
With LA = 22.626 m, La = 16 m, and Le = 20 m,
)
cos aA = sin aA = 0.707 cos ae = -0.6 sin ae = 0.8
0
- (_Q2_
wint - OU4
22.626
U4 + _Q2_ V4 +
22.626
0.36
20
U4 -
0.48
20
V4 EA
+
( "
uV4
0.5
V4
0.5
+ 22.626
U4
v4
+ 16 +
0.64
ZO V4
-
0.48 )
U4 EA
22.626 20
Form the virtual work due to applied loads:
8Wext = OU4 320 + OV4 240
Equating 8W;n1 and 8Wcx using values for A and E, and noting the arbitrary character of 8u4
and 8v4 gives
1203u4 - 57v4 = 320
-57u4 + 3498v4 = 240
Solution of these yields u4 0.2695 mm and v4 0.0730 mm. The member forces are obtained
by substituting u4 and v4 into Equations a, b, and c to give strains and multiplying these by EA.
= =
The final results are: FA = 321 kN, Fa = 136.9 kN, and Fe = -155 kN.
Example 6.5 is a problem that involves only one form of behavior-bending-but

displacements that are a function of two undetermined parameters. Because of the
choice of trigonometric functions used to describe the approximate real displaced state
and the virtual displaced state, the resulting equations are uncoupled, and the unde
termined parameters can be solved for independently. This approach, which is based
on a series representation of the displaced state, is known as the Rayleigh-Ritz method.
Because it is approximate, the chosen displacement function violates the conditions of
equilibrium.
Using the principle of virtual displacements and the following two-term trigonometric
expansion in description of the transverse displacement, calculate the displacements at
joints 2 and 3. Compare this solution with the exact answer (El, constant).
v= a1 . 'TTX
sm
L
-+ a2 sm . 21TX
-
2( 2 2
BK,=
d B
-;J;2 =
u)
-001(L) m L 0024(L) sm L
'TT
s
. 'TTX
-
'TT . 21TX
Form the internal virtual work
8Wint = LL OK, K,El,dx
= r [001(r 7 002 4(r z ] sin + sin

2 x
x [a1(Y 7 a2 4(Y z ] sin +

sin
2 x
El,dx
-U- J [001 a1 sm2 16002 a2 sm2 L

El,1T4 (L . 'TTX . 21TX
BWint = +
o L

+ 4(001 a2 002 a1)

+]
sin
'TTX
L
sin
21TX
L
dx
and since
r sin(7Tx/L) sin(27Tx/L) dx = O
and
IoL sm. 2 -
'TTX
L
dx = IoL sm2 . 21TX
-
L
dx =
L
-
BWint =
El 1T4
2
3 001 a1 -U 002 a2

+
8El,7T4

Form the virtual work of the applied loads:

By the principle of virtual displacements,
4
EI,11'4 BEl,11'
2L
3 001 'a1 + 002 a2
= 001(0.707P2 + 0.966P3) + oo2(P2 - 0.5P3)
And, because of the independent nature of 001 and 002,
so that
u =
L
4
3
[ (1.414P2 + l.932P3) sm
7rX . 21T'X
+ (0.125P2 - 0.0625P3) sm
]
El,11' L L
3 3
L L
U2 =
EI,11'
4 (1.125P2 + l.303P3) = - (0.0116P2 + 0.0134P3)
EI,
[ versus ;, (0.0117P2 + 0.0133P3) exact

]
3 3
L L
V3 = 4 (1.303P2 + L898P3) = (0.0134P2 + 0.0195P3)
El,11' El,
[ versus
L
3
Ell
(0.0133P2 + 0.0197P3) exact
]
It can be observed that in Examples 6.3 and 6.5 different approaches were taken to
the definition of the displaced state. In Example 6.3 the displaced state is written
directly in terms of the physical displacements, that is, the displacements u2, v2, ,
(),2 of point 2. In Example 6.5, the displaced state is written in terms of the parameter
a1 and a2, known as generalized displacements or generalized parameters. These param
eters are carried through to a solution in terms of the applied loads, at which point
they are substituted into the originally chosen expression for the displaced state. It
would have been possible, alternatively, to fit the parameters a1 and a2 to the transverse
displacements at two points on the beam ( e.g., the center and a quarter point) and
obtain an expression for the displacement at any point in terms of the displacements
at these two points before performing the virtual work calculation.
In Chapter 7, where the virtual work principle is employed to formulate element
stiffness equations, it will be necessary to work with the displaced states in terms of
physical parameters, the element joint displacements. Thus, if the displaced state is
first described in terms of generalized parameters, it must then be transformed to an
expression in terms of physical parameters. A detailed examination of the procedure
to be followed is presented at the start of that chapter, in Section 7.1.
6.5 PRINCIPLE OF VIRTUAL FORCES
6.5.1 Equations of Equilibrium
Stresses and strains are the dual parameters of structural analysis. They are connected
to each other by the elastic constants. In the principle of virtual displacements the
internal virtual work resulted from virtual strains and real stresses. It is logical to
expect, and it is indeed the case, that there is a dual work principle in which the internal
virtual work is due to the action of virtual stresses on real strains. This dual principle
is the principle of virtual forces.
In the principle of virtual displacements we examined the changes in external and
internal work that took place on account of a perturbation, or virtual change, of a
displacement state about a real state of stress and strain. This process gave equations
for the loads in terms of the displacements, that is, stiffness equations. We now examine
an imaginary, or virtual, state of applied loading and its associated internal stress sys
tem. The loading and its associated internal stress system are designated the virtual
force system in what follows. Application of this system in conjunction with the prin
ciple of virtual forces leads to expressions for displacements in terms of forces, that is,
flexibility equations.
The conditions associated with a displacement state are the conditions of displace
ment continuity. Consequently, the virtual displacements were required to meet these
conditions, consistent with the chosen description of support conditions. The condi
tions associated with a stress state are those of equilibrium; therefore, the fundamental
requirement on virtual force systems is that they meet the relevant conditions of equi
librium. Thus, preliminary to the definition of the principle of virtual forces, we must
first discuss the relationships for the force resultants of axial, torsional, and flexural
behavior, especially those representing the conditions of equilibrium.
Consider first the axial member, a differential segment of which is shown in Figure
6.7a. For generality, the member is assumed to be of nonuniform cross section, with
area A, subjected to a distributed axial load q, expressed in units of force per unit
length.
(b)
x, u
(a)
(c)
Figure 6.7 Forces acting on differential elements of axial, torsional, and flexural members. (a) Axial behavior
(b) Torsional behavior. (c) Flexural behavior.
The internal force on the left face is Fx. (Fx is positive when tensile.) Because of the
presence of the applied load q, this force will change along the length and at the right
face it is Fx + (dFxldx)dx. The condition of equilibrium in the horizontal direction can
now be written as
or
dFx
-+ q =O (6.25)
dx
The more fundamental measure of structural action on a cross section is stress, which
is related to the force by Fx = Aux.
Similar relationships hold for the case of torsional behavior (see Figure 6.7b). The
left face of the illustrated differential element is subjected to a twisting moment Mx
and, because of the presence of a distributed applied twisting moment mx, the moment
on the right face is Mx + (dMxldx)dx. The condition of equilibrium of moments about
the x axis gives
or
dMx
+ m x=0 (6.26)
dx
Again, there is a more fundamental measure of structural action, which in this case is
the shearing stress r = Mxrll, where J is the torsion constant and r is the radius from
the x axis to the point at which r is evaluated.
In examining beam flexure we consider, as before, a differential element (Fig
ure 6.7c) that has a cross section that is symmetric about the y axis. A moment M,
and a shear FY act on the element's left face. A distributed transverse load q acts
on the element in the y direction. Because of the action of an applied distributed
load, q, the shear at the right face is FY + (dFyldx)dx. The moment at the right face is
M, + (dM,!dx)dx. From y-direction force equilibrium,
(
""-' FY = FY +
LI
dFy
- dx
) - F
Y
- q dx = 0
dx
so that
dFy
- q =O (6.27)
dx
Also, from the equilibrium of moments about the z axis at the right end of the element
( )
(the higher-order effect of q is disregarded),
dM,
""-'
.., M, = M + dx - M - F dx = 0

Y
' dx '
or
dM,
= Fy (6.28)
dx
Finally, by substitution of (6.28) into (6.27),

d2M,
(6.29)
dx2 q
=
Equations 6.27 and 6.28 are the basic equations of equilibrium for beam bending.
When the internal bending moment satisfies Equation 6.29, both Equations 6.27 and
6.28 are satisfied.
The stresses due to the shear and bending moment are, respectively, Txy = FyQ,11,b
and <rx = M,yll,. In the equation for Txy b is the width of the element at the point
where the shear stress is calculated and Q, is the first moment of area beyond that
point, taken with respect to the neutral axis.
6.5.2 Characteristics of Virtual Force Systems
We can now proceed to the definition of the characteristics of virtual force systems.
The fundamental requirement on any such system is that it meet the relevant condi
tions of equilibrium. These are represented, externally, by the overall equations of
equilibrium ( e.g., "1:.Fx = 0, etc. ) and internally by Equations 6.25, 6.26, or 6.29, as
appropriate.
Consider, in illustration of this requirement, the imposition of a virtual force fiFy, to
the left end of the cantilever beam of Figure 6.8a. The appropriate virtual force system,
in terms of internal shears and moments, is simply that given by elementary statics and
shown in Figures 6.8b and 6.8c. If the bending moment distributions were, however,
assumed to be of the form fiM, (x2/L )fiFy1 rather than the correct form x( fiFy, ), then
=
the relevant equilibrium condition ( Equation 6.28) gives for fiFy1> the virtual shear at
any section:
_ dfiM, _ 2x
fiFY fiFy1
dx L
- -
which is not consistent with the correct relationship fiFy = fiFy,. This assumed virtual
force system would not be an acceptable virtual force system.
Generally, an? in particular for statically indeterminate structures, there is more

\
(c)
.. . . '
Figure 6.8 Virtual force systems for cantilever
beam. (a) Structure. (b) Shear. (c) Bending
moment.
(a)
(b)
,"(-x)
(c)
liM,
K Figure 6.9 Fixed beam under concentrated load
alternative virtual force systems. (a) Structure.
( b) Moment diagram-simple support conditions.
(c) Moment diagram--cantilever support conditions.
(d) Moment diagram-fixed support at left and hinge
(d) at right.
than one acceptable choice of a virtual force system. Figure 6.9 illustrates this point.
The given analysis condition ( Figure 6.9a) is a fixed end beam under the action of a
concentrated load Py3 The moment diagrams of Figures 6.9b-6.9d are, respectively,
the solutions for the beam with simple supports, cantilever support, and fixity at the
left end and a hinge at the right end, each under the action of the applied load Py3 It
is easily verified that application of the equilibrium conditions ( Equations 6.27 and
6.28) gives a statically consistent force-distribution relationship for each support sys
tem. Hence, in the problem at hand, any one of these is acceptable as a virtual force
system for a virtual force 8Py3
Despite the wide number of choices available in the definition of virtual force sys
tems for statically indeterminate structures, practical considerations favor the use of
systems calculated for a statically determinate form of the same structure, such as the
simple beam ( Fig. 6.9b) of the problem of Figure 6.9a. The use of internal force dis
tributions corresponding to the statically indeterminate fonns of the structure are gen
erally not practical. They are certainly not of use in the development of computer
oriented analysis tools since, most often, it is precisely this information that is being
sought through the solution.
6.5.3 Formulation of the Virtual Forces Principle

It was stated at the beginning of this section that the principle of virtual forces stems
from the investigation of the changes in work and energy that occur on account of the
imposition of a system of virtual forces on the state achieved under the real applied
(]
(a) (b)
Figure 6.10 Complementary external and internal virtual work.
loads. The virtual force system produces external complementary virtual work, 8W:x1
and internal complementary virtual work, 8Win17
The term "complementary" can be explained by reference to Figure 6.lOa, a plot
of the relationship between a force F and its corresponding displacement A, for linear
elastic behavior. As explained in Section 4.2, Figure 4.3, and again in Section 6.1, Figure
6.1, the area between the F-A line and the A axis is the external work. The shaded area
between the F-A line and the F axis is complementary to the external work and is
designated as the external complementary work. Thus, if the intensity of the force is
incremented by an amount 8F, the virtual force, then the external complementary
virtual work cSW:xi is ( 8F)(A0), where Ao is the intensity of the displacement at the total
load F0
A general expression for the internal complementary virtual work cSWini can be
established by reference to Figure 6.lOb, a plot of stress versus strain at a representative
point in a structure. In analogy to the description of external virtual load-displacement
circumstances, the action of the virtual stress8u0 (caused by the external virtual load)
on the real state of straine0 results in a complementary virtual strain energy per unit
of volume equal to L oo0J{e0}. The internal complementary virtual work in the complete
body of volume (vol) is, therefore
8Win1 = J vol
Loo0j{e0}d(vol) (6.30)
Because we wish to deal with force states rather than displacement states, the re
lationship between stress and strain is invoked to replace {e0} by stress {u0}. This has
been designated as {u0} = [E]{e0} for the general state of stress and, since we require
strain in terms of stress, we write it as {e0} =
[Er1{u0}. Hence, Equation 6.30 becomes
8Win1 = Jvol
LoooJ[Er1{uo}d(vol) (6.31)
We are now in a position to state the principles of virtual forces, which is simply a
condition for the balance of internal and external complementary virtual work for the
condition of an imposed virtual force system:
cSlf!xt =
8Wint (6.32)
7This quantity is also called complementary virtual strain energy, f>U*, a designation that was used in the first
edition of this text.
It has already been emphasized that the real and virtual force states must each satisfy
the conditions of equilibrium. Equation 6.32 represents the conditions of compatibility.
When the real force state corresponds to a state of strain which meets the conditions
of compatibility, the solution is exact.
The solutions to be established by use of the principle of virtual forces are all exact
(within the limits of the assumption of linear behavior) and we use Equation 6.32 to
achieve the evaluation of the parameters that characterize such solutions. This is done
in the next section and again in Section 7.6. Therefore, with exact solutions in mind,
we can give the following verbal statement of the principle of virtual forces (see
Ref. 6.1).
The strains and displacements in a deformable system are compatible and consistent
with the constraints if and only if the external complementary virtual work is equal
to the internal complementary virtual work for every system of virtual forces and
stresses that satisfies the conditions of equilibrium.
If the real force state does not correspond to a deformational state that exactly
satisfies compatibility, then Equation 6.32 can be used to enforce an approximate sat
isfaction of the conditions of compatibility. This is analogous to the state of affairs in
the virtual displacements approach, when the real displacement corresponds to a stress
state that does not satisfy equilibrium, and enforcement of the virtual displacement
condition (Eq. 6.13) results in an approximate satisfaction of equilibrium. In this text,
however, we are not concerned with approximate solutions by means of the principle
of virtual forces.
To make use of this principle it is necessary to establish specific expressions for the
respective modes of structural behavior that are of interest in framework analysis. In
so doing, the subscript 0 is removed from all terms. Also, the stresses are replaced by
the respective stress resultant terms.
For axial members (Fig. 6.7a), we have {O"j = F)A, LcSuJ = cSFxlA, [Er1 = 1/E, and
d(vol) =A dx . Therefore, Equation 6.31 becomes
( L
F
cSifint = Jo cSFx
dx (6.33)
E
For the flexural member (Fig. 6.7c), we have u = ux = M, y l l ,, LcSuJ = cSux =
cSM, yll,, and [Er1 = 1/E. Thus, Equation 6.31 becomes
)( )
y y d(vol)
cSifint =
(cSM, /, M, I, ----e-
vol

(6.34)
) = cSM,. Z:
=
r (5 ' )(' )( L r y2dA dx dx
Through similar reasoning we find that, for torsion,
( L
M
c5ifi01 = Jo cSMx (6.35)

G; dx
6.5.4 Construction of Analytical Solutions by Virtual Forces Principle
. As in Section 6.4, where we explored the practical implications of the principle of

virtual displacements through the medium of analytical solutions, we now examine
6.5 PJinciple of Virtual Forces 165
solutions by the principle of virtual forces. The solutions to be obtained are for dis
placements due to the action of specified forces.
To use the principle of virtual fon;es we must have expressions for the real and
virtual internal force systems. The total internal force system is that resulting from the
given applied loads. The virtual internal force system results from the application of a
virtual load that corresponds to the desired displacements.
The simplest example concerns the calculation of displacement u2 of the uniform
axial member (Fig. 6.5a) due to the end load Fxz The real internal force system is,
from statics, Fx Fxz The virtual force is chosen in correspondence with the desired
=
displacement and therefore is SFx2, resulting in an internal virtual force system SFx =
SFxz We have, from Equation 6.33,
(L SFx2 Fx2 d _!::__

SW'.mt =
J0
EA
x = SFx2 . Fx2
EA
The external complementary virtual work is simply SFx2 u2 Thus, from the principle
of virtual forces
or
which is the correct solution.

Consider next the uniform flexural member, Figure 6.11, simply supported at either
end and subjected to the central load, - Pc. We seek to calculate the displacement
under this load so that the virtual force is -SPc. The internal force system, character
ized by the moments M, is (Fig. 6.llb)
1:
L
.1,_
2
rv l
PcoroPc
0
x

c

(a) Centrally-loaded beam
or
oMz =
oPcx
Figure 6.11 Simply supported
beam. Total and virtual force
2 systems. (a ) Structure. (b) Moment
diagram for simple support
(b) Internal moments, real and virtual conditions.
and we choose the virtual internal force system to be of the same form:
x
8Mz = c -
(8P)
2
Since these are derived from statical analysis, they meet the conditions of equilibrium.
We have, from Equation 6.34 and invoking symmetry about c,
t>Wfnt = - lu2 (
E/z
2
0
8Pc
x
-
2
)( )
Pc
x
- dx
2
= 8Pc Pc --
48/z
L3
The external complementary virtual work is t>W:xt = -8Pc Ve, and the principle of
virtual forces gives
L3
-8Pc vc = 8Pc Pc --
48E/z
0r, upon cancellation of 8Pc from both sides,
vc =
In the case described above we placed a single virtual external load 8P; at the point
i at which a displacement solution was desired. (In flexure, when an angular displace
ment is sought, the virtual force is a moment.) Since the virtual displacement is can
celled from both sides of the equation t>W:xt = t>W'int its value is immaterial. Therefore,
it is customary to set 8P; equal to unity, and for this reason the principle of virtual
forces, in practical application, is generally termed the unit load method. The procedure
of setting the virtual external load to unity (with due account being taken of the sign
of the virtual force) is followed in Examples 6.6-6.8.
Using the principle of virtual forces, calculate the displacements of points 2 and 3 of
the beam in Example 6.5.
The internal moments due to the applied loads are
x
M, = 1 (9P2 + 5P3)
- .
2
1 5
M, = 4 (L x)P2 +
12
P3x ( s x s )
M,
P2
= 4 (L - x) +
7P3
U
(L - x) G s x s L )
v2 V3'. we V2
8P2 -1
n of and respectively. For
Two systems of virtual forces are needed for calculatio .
at point 2. The correspondmg virtual moments are then
=
3x (0 x L)
place a virtual external load
BM = - :S :S -
' 4
4
=
L - x ( x L)
-- :S :S
4
-1 { Jo(u4 (3x)()(9P2
In this case, the internal complementary virtual work
1 =
12 5 P3)dx +
fu1u14 2 (L )(41)[(L - x)P2 35 PJX J dx

m EI, 4
x
+ +
2 ( L x)(L x)(P2 P3)dx}

+
J1:11 +
=
L- 3 (0.0117P2 0.0133?3)
EI,
+
Also, the external complementary work is BW:xi ( 8P2) V2 -1 v2.

= = Equating BW'l'n1 and
BW:xi we have
V2 --L3 (0.0117P2 0.0133?3)

=
EI,
+
Similarly, by placing a virtual force 8P3 -1 = 3 at point and again pursuing the virtual forces
V3 - L3 (0.0133P2 0.0197P3)
procedure, we find
= +
EI'
Just as the virtual displaced state in the principle of virtual displacements need not
correspond to the real displaced state, the virtual force state need not correspond to
the real force state in the application of the principle of virtual forces. This circum
stance often proves useful in the analytical determination of the displacements of stat
ically indeterminate structures. For example, if a solution is available for the internal
force distribution due to the applied loads, then a statically determinate virtual force
system can be employed in calculation of displacements by means of the principle of
virtual forces.
This is illustrated in Example 6.7. The exact solution for the desired displacement
is obtained even through the chosen virtual stress field does not correspond to a dis
placement field that meets the displacement continuity conditions of the problem. (The
displaced state of the simply supported beam is in violation of the fixed-support con
dition at point 2.)
Example 6.8 demonstrates that care must be exercised, however, in the choice
of virtual force fields so that the results obtained have practical value. The logical
choice of a virtual internal force distribution is the force associated with the actual
cantilever support (virtual force system 1). This leads to the conventional solution
Ve = 5P1L3!48EI,. Virtual force system (2), however, which is associated with simple
support, gives the solution v1 /2 - Ve = P1L3/16EI,. Although this is a valid solution,
its form is not of direct practical value. It requires a separate determination of Ve
and this cannot be accomplished using simple support conditions for the virtual force
system.
Calculate the displacement V

e
at the center of the beam shown, using the principle of
virtual forces.
c
The beam is statically indeterminate, and the determination of the internal force distribution is,
in itself, a problem for solution by the principle of virtual forces, using redundant force concepts.
Since the problem seeks only a value of displacement, it will be assumed that the indeterminate
structural analysis has already been performed, giving the following relationships for the internal
moments in terms of the applied loads P3
(osxs)
M' = -7 -7x) (s s )
?3
(5 L x L
o e
2
l'1 order to obtain the displacement v0 a virtual force P =
-1 is placed at point c. Since the
associated internal moments need only satisfy the conditions of equilibrium, we calculate them
x
as if the beam were simply supported at both ends. This gives
oM' (osxs)
= -
L-x (sxsL)
2
Thus
[ Jo(w (x2) u 2x
ow:'ni =
1
l, 27 dx J ( 2) -7x) 27
(5x - L )
4P3
+
L/3
(5 L
P3
dx
fu2 - 7x) 27 dx
E
L 3
J
L ?3 5P3L
--
x
+ (SL - =
2 1296/,
Also,
0W:x1 = -1 Ve
Hence, by 0Win1 = 0W:x1

vc =
Calculate the displacement at point c of the cantilever beam shown, using the
based on the actual support condition and

of the beam.
(2)
virtual forces approach with the following systems of virtual forces:
a system based on simple support
(1) a system
L
k---2----.
(a)
The actual internal force system, defined by M,, is M,=P

Real force system
shown in part b of the figure. Proceeding to the so
lution for virtual force system (1), we place a unit
(b)
load at point c and obtain the internal virtual force

system shown in part c of the figure. Thus tcPe = 1
----
1
LBW:xt = Ve .SMz - (x-.L)
1 ( L ) 5P1L3
- -.SPc
2
BW'fn1 = EI
, I w x
-
2 (Pix)dx = 48EI
,
Virtual force system for
given support cond1t1on
(c)
Hence, by BW:x1 = BW'fnt 1 1

2 2
Virtual force system for
simple support conditions
(d)
The virtual force system (2) gives the internal virtual
force system shown in part d of the figure. In this
case, we have
also
1
0W'f01 = EI Jo
( Ll2 ( ) 1
(P1x)dx + El u2
L fL + x
(P1x)dx
, ,
P L3
= 161EI
,
Hence, by 0W:x1 = BW'fn1

Vi P L3
- Ve= 1
2 16EJ z
Our intention in introducing the principle of virtual forces has been to establish a
basis for the formulation of element flexibility equations, a topic which will be taken
up in detail in Section 7.6. In such cases the element will always be a statically deter
minate structure, and the virtual forces will be chosen in the same form as the real
forces. As mentioned in the introduction to this chapter, our principal interest in flex
ibility relationships is in their value in the formation of element stiffness equations
through simple flexibility-stiffness transformations. The role of the principle of virtual
forces in statically indeterminate structural analysis, where it is most prominent in
classical methods, has only been mentioned briefly. To use the principle effectively in
large-scale, computerized indeterminate structural analysis it is necessary to have a
more organized scheme than the one that is customarily adopted in classical treatments
of the principle. The essential components of one such scheme may be found in Chap
ters 6 and 7 of the first edition of this text.
6.6 PROBLEMS

6.1 Calculate, by use of virtual work, the force in membe
r 4-6.
Q
3@a ,. Q
J
2
P 2
._
t l
_L_ /,1 3 5 7 9
2P
14----- 4@a= 4a --------+1
Problem 6.1
6.2 Calculate, by virtual work, the reaction at point 5.
j50kN 180kN
2 3 t4 s 6 !1 s
l 1 a I 1
-- 6m ----3m +2m +2m -l-- 4m -----+-3m -+-3m --J
i.-----
Problem 6.2
6.3 Calculate, by virtual work, the reactions at points 1 and 2 of Problem 6.2.
6.4 Establish expressions for the total strain energy, U, of combined axial, torsional,
and flexural behavior (Example 6.2) by treating 8Wint as a diffen;:ntial and assuming
linear behavior.
6.5 A nonlinear stress-strain law is defined as u = E0(e + 100e2) where E0 is the

modulus of elasticity at e = 0. Develop an expression for the internal virtual work for
an axial member composed of this material.
6.6 Calculate the displacement at point A, using the principle of virtual displace
ments. Use v = vA sin 7rx/L and a similar form for&.
\
jPA
ly,v 2 tA 3 4
Problem 6.6
6.7 Solve Problem 6.6 by means of the following expansion for v and a similar ex
pansion for&.
6.6 Problems 171
6.8 Calculate, by means of the principle of virtual displacements, the deflection of

the end point of the cantilever beam shown. Use the following approximation of the
transverse displacement (EI constant ) . (First fit the coefficients a1 and a2 to the dis
placements at points 2 and 3, v2 and v3). Check for satisfaction of the conditions of
equilibrium and discuss the results obtained.
v = ( 1 - cos ;) a1 + 1 ( - cos
3
2
"';,) a2
1
_J
-xu
2
L
C
Problem 6.8
L
.
3
v = a1x2 +a2x . (Fit
x
6.9 Solve Problem 6.8, using the coefficients a1 and a2 to the
displacements at point 3, v3 and 8 3. )
6.10 Calculate, by means of the principle of virtual displacements, the displacement
2
at point A, using v = x(x - L)a1+x (x - L)a2 and a similar form for c5v1 (EI constant) .
J1_
l I I I I I I I I I!
_

.L_jA
,
xu
------L ______ 2
_,___
2
Problem 6.10
6.11 Using the principle of virtual displacements, solve for the displacements v1 and
8zi of the cantilever beam and for the support reactions Fy2 and M,2
Use
v=2
1
( 1 +cos
1TX
L
) Vi.
1 1TX
(
&=- l+cos- &1
2 L
)
1 2
FYJlf+ --x-
, - -- L
-----
Problem 6.11
6.12 Determine, by means of the principle of virtual displacements, the approximate

value of the transverse displacement at point 2. Use v = -sin( 7Tx/10)v2 in description
of the displaced shape of the beam. E = 200,000 MPa.
I= 40x106 mm4
3
-__:;_ __________________ ----.J... 4
..-0-' ----
Problem 6.12
5 m----2
-++- m L __J 3m
6.13 Calculate, by means of the principle of virtual displacements, the displacement

of point 5. All member areas are A = 10 X 104 mm2, E = 200,000 MPa.
lOm
0
,_J 0
L 1om __j
Problem 6.13
6.14 Calculate, by virtual work, the transverse displacement of point 3. E

200,000 MPa, I = 100 x 106 mm4.
7TX 7TX
v sm sm 4

= a1 + a2
12
---
I.- -;,l
4m
Problem 6.14
4m l
4 m -+1:.----- 12 m ___ ____.,
6.15 Solve for the displacement at point 3 by use of the principle of virtual forces.
The exact solution for the moments in the beam, due to the applied loads is,
M, = -45 + 22.Sx (0 =-:; x =-:; 6)

= 270 - 30x (6 =-:; x =-:; 9)
References 173
Use for the virtual force distribution a statically determinate bending moment distri
E
tI
bution. = 200,000 MPa.
V
ty Py3 = 30 kN

J
I:.u
...._
._ __ 6 m ____ ..,..__ 3 m
Problem 6.15
6.16 Derive expressions for the angular and transverse displacement, 8,2 and v2 by
use of the principle of virtual forces. The member is of rectangular cross section and
has unit width.
u
r y3
,-
P
t
3 3
2
x,u
l
1 2
Problem 6.16
L .I. L _J
6.17 Calculate the displacement at the center (x = L/2 ) of the structure of Example
6.5, using the principle of virtual forces.
6.18 Calculate the displacement 8,b of Example 4.10, using the solution of that ex
ample for the real force system and a statically determinate virtual force system.
REFERENCES
6.1 J. T. Oden and E. A. Ripperger, Mechanics of Elastic Structures, 2nd Edition,

Hemisphere Publishing Co., N.Y., 1981.
6.2 J. S. Przemieniecki, Theory of Matrix Structural Analysis, McGraw-Hill, N.Y., 1968.
6.3 J. H. Argyris and S. Kelsey, Energy Theorems and Structural Analysis, Butterworth
Scientific Publications, London, 1960.

Chapter 7
Virtual Work Principles in

Framework Analysis
In this chapter we apply the principles of virtual work established in Chapter 6 to the
construction of the algebraic relationships of framework matrix structural analysis. A
development that is entirely parallel to the direct formulation procedure given earlier
can be accomplished through virtual work. To include such parallel developments is
desirable since it equips the reader to cope more easily with continuing developments
in this field, which might be based on either direct formulation procedures or on virtual
work concepts.
The principles of virtual work are applied herein principally for their facility in the
formulation of approximate solutions. This facility has no identifiable counterpart in
the direct formulation procedures. For example, an expeditious and relatively accurate
approximate formulation can be established for tapered members, and a quite general
approach can be made to the treatment of distributed loads. Also, it is the approach
we shall use in the formulation of the equations of geometric nonlinear and elastic
critical load analysis in Chapter 9.
The starting point of this chapter, Section 7.1, delineates the manner of describing
the displaced state of framework elements. When the resulting element formulation is
intended to be approximate, the approximation is nearly always in the form of a de
scription of the displaced state, or "shape function," of the elements.
With a means of description of the displaced state of elements in hand, we proceed
in Section 7.2 to the formulation of the standard stiffness relationship of constant
section framework elements, using the principle of virtual displacements. Relationships
already established in Chapters 2-5 are reformulated in this manner. The material
presented merely serves to illustrate the details of the procedure. The next section,
however, takes up the formulation of approximate stiffness equations for tapered el
ements, an entirely new topic. Similarly, nonuniform torsion, the condition in which
torque is resisted by a combination of the simple torsional shearing stresses studied in
Section 4.5.2 and resistance to out-of-plane warping, is treated in Section 7.4. Then,
on the basis of virtual displacement considerations, a completely general approach to
the treatment of loads applied between the end points 9f framework elements is pre
sent d. This approach encompasses the extremes of point loads and distributed loads
f
and1confirms the validity of expressions derived in Section 5.2.
The element flexibility relationships employed up to this point were those for the
uniform axial member and beam element and were established in Section 4.5 through
direct reasoning. Practical structures, however, may contain tapered or curved ele
ments or elements of such form and behavior that flexibility equations are not readily
174
derivable through that route. Nevertheless, a general approach to the formulation of

element flexibility relationships is available on the basis of the principle of virtual forces
presented in Section 6.5. In Section 7.6 it is applied to the formulation of element
flexibility equations. Included is the treatment of transverse shearing deformation in
beams.
7.1 DESCRIPTION OF THE DISPLACED STATE OF ELEMENTS
7.1.1 Definition of the Shape Function Mode of Description
In linear elastic analysis the only ingredients needed for the construction of element
stiffness equations by the virtual displacements approach are:
1. The elastic constants relating the stresses and strains of the material.
2. Descriptions of the real and virtual displacecl states of the element.
3. The relevant differential relationships between strain and displacement.
The elastic constants are known from laboratory testing. The differential relationships
between strain and displacement are basic relationships in structural mechanics and
have already been defined for the cases of interest. It has been pointed out (Section
6.4) that the virtual displaced state is logically chosen to be of the same shape as
the real displaced state. Thus, to form element stiffness equations by means of the
virtual displacement approach, the only preliminary task that remains is the description
of the displaced shape of the element, which is then employed for both the virtual and
real displaced states.
The displaced state of the simplest structural elements, such as those studied pre
viously, can be found by solution of the differential equations that govern the behavior
of these elements (e.g., Equation 4.31b in the case of bending). To use the virtual
displacement approach would appear to be redundant in part, since, as demonstrated
in Section 4.5, the element force-displacement equations can be constructed directly
from these solutions. It will be shown in Sections 7.3 and 7.5, however, that a descrip
tion of the displaced state of an element that is exact for such conditions can be ex
tremely useful in the construction, by the virtual displacement principle, of terms that
account for nonuniform cross-sectional properties and distributed loads.
Alternatively, it is possible to- begin with an assumption as to the displaced state of
the element that is not obtained from a solution to a differential equation. In the case
of the simplest form of an element it might, fortuitously, correspond to the exact
displaced state. Generally, however, the assumption made will be an approximation.
Whatever the approach taken to the definition of the displaced state of an element,
our objective is to produce an algebraic expression in terms of all of the displacements
at the element node points, that is, an expression of the form
a = N1a1 + N2d2 + N;d; + Nnan

n (7.1)
= 2: N;d; = LNJ{.1}
i=1
where a is the displacement component in question (such as u in the case of an axial

member, or vin the case of the beam), d; is the ith degree of freedom of the element,
N; is the "shape function" corresponding to d;, and n is the total number of degrees
of freedom at the node points of the element. We elaborate on the physical meaning
of a shape function below; for the present, it is simply identified as a function of the
element coordinates.
176 Chapter 7 Virtual Work Principles in Framework Analysis

1

Fxl> U 1 1
-
-
---- ----- - L ----1 - - 2 Fxz, Uz
I
x,u Figure 7.1 Axial force element.
7.1.2 Formulation of Shape Functions
Consider, for example, the axial member (Fig. 7.1). Here, d = u, d1 = u1 and dz =
Uz. Hence, the form of the displacement expression is
(7.2)
The pure torsion member ( Figure 7.2) is described by end-point displacements that
are the angular displacements 8x1 and Bxz The angular displacement at any point be
tween nodes 1 and 2, ex, can be expressed as
(7.3)
Another example is the flexural element ( Fig. 7.3), where d v, d1 vi. dz = = = Vz,
d3 = 8,i. and d4 = 8,z. In this case the displacement would be described by
(7.4)
The derivation of the general frame element stiffness matrix in Chapter 4 was based
upon the superposition of simple axial and torsional behavior and flexure about two
perpendicular axes. Consequently, the axial, torsional, and flexural displacement func
tions cited above are sufficient for the derivation of the same relationships through
application of virtual work concepts. It now remains to establish the algebraic form
and physical significance of the shape functions (N;) for these cases. We begin with
the simplest case, that of the axial member of uniform cross-sectional area A.
For this element it is known that the axial strain is constant. Since ex = du/dx, u is
a linear expression in x, the most general statement of which is
(7.5)
The terms a1 and az are constants that, as yet, have no assigned physical meaning. We
also observed that two constants are needed to match the two node-point displace
ments, u1 and Uz. Proceeding to the evaluation of Equation 7.5 at the node points we
have, at point 1, where x = 0,
Also, at point 2, where x = L and with the above result,
so that
llz = ( uz - u1 ) /L
By substitution of these expressions for a1 and az into Equation 7.5
(7.6)
Figure 7.2 Pure torsion element.

'4------L-------+<
Comparison of Equations 7.2 and 7.6 discloses that N1= (1 - (x!L)), N2

is instructive to examine these shape functions more closely for the sake of establishing
= x/L. It
properties applicable to all element-shape functions. First, the shape functions, which
for this case are nondimensional, have unit value when evaluated at the nodal point
to which they refer and are zero when evaluated at all other nodal points. This must
be the case for u= u1 when Equation 7.6 is evaluated at point 1 (x= 0) and u= u2
at point 2 (x= L).
Figure 7.4 shows the variation of N1 and N2 as a function of the element coordinate
x. Each plot describes the displaced shape of the element when the element is given
a unit displacement at the node point corresponding to N; (point 1 in the case of N,,
point 2 in the case of N2), and the displacements at all other degrees of freedom are
held to zero. It describes the displaced shape of the element under these circumstances,
hence the designation shape function for N;.
Definition of the displaced state of the pure torsional element (Fig. 7.2) follows along
the same lines as for the axial member. The angular displacement (Ox) is given in terms
of two node-point displacements (Eq. 7.3), and so it is logical to employ a linear de
scription of the angular displacement at any point along the element:
Ox= ll1 + ll2X

Clearly, the process of evaluation of a1 and a2 results in
e;= ( - i) Ox1 i Ox2

1 + (7.7)
so that N1= (1 - (x/L)), N2= x/L as in the case of the axial member.
Construction of the shape functions for flexural behavior involves more extensive
algebra than the above cases but follows along the same lines. The element (Fig. 7.3)
is described in terms of four joint displacements, Vi, V2, ez1' ez2 Note, as emphasized
in prior chapters, that in flexure the node-point angular displacements are the deriv
"
atives of the transverse displacements, that is ezl = dv!dxli. ez2= dvldxb.
A polynomial expression has been used in the case of the axial and torsional ele
mljlts, and it is logical to adopt a polynomial expression for the flexural element. Since
\there are now four joint displacements, the polynomial in question is cubic:
(7.8)
Figure 7.4 Shape functions for uniform axial element.

Evaluation of v, at point 1, where x = 0,
(7.8a )
At point 2, where x = L,
a2L +a3L2 3
V2 = a1 + + a4L (7.8b )
To evaluate 0,1 and 0,2, we need an expression for slope:
dv 2
dx
O, - - a2 2a3x 3a4x
_ _
+ +
, Hence, at point 1,
(7.8c )
and at point 2,
(7.8d )
Equations 7.8a-7.8d are four equations in four unknowns, which can be solved to yield
ll1 = V1
a2 = 0,1
1
a3 = L2 (-3v1 + 3v2 - 20,1L - 0,2L)
and
1
a4 = L3 (2v1 - 2v2 + Oz1L + 0,2L)
Substitution of these into Equation 7 .8 and collection of terms gives
v= r 3(tY (tf } [ (tY (tYJ

1 - + 2 1 +
3 - 2 v2
( ) [( ) ]
2 2 (7.9 )
x t - x -
+
L 0z 1 +
L L 0'2
Thus, by comparison with Equation 7.4,
N
i =
1 -3 (zY (iY + 2 N3=x 1 ( iY
[ (iY - i]
-
Ni
=
3 ( zY ( i Y
- 2 N4 = x
Figure 7.5 Shape functions for

flexure of uniform beam element.
These shape functions are plotted in Figure 7.5. It is instructive for the reader to sketch
these on the basis of the definition of shape functions given above. The process de
scribed above for the axial, torsional, and flexural elements is illustrated further in
Examples 7.1 and 7.2.
It is desired that a stiffness matrix be formed, using the virtual displacement prin 1 2 3
. +
ciple for an axial member with three node points. Construct the shape function on E+
which the formulation will be based.
L ----- L_j
Since the shape function will involve three j oint displacements ui. u2, u3, and it is appropriate
( )
to choose a quadratic three-term polynomial as the basis for the shape function. Thus
By evaluation at =0,
x x L, x = and = 2L,
U1 = a1
U2 = a1+ a2L+ a3L2
U3 = a1+ 2a2L+ 4a3L2
Solving,
u
Substitution of these into the original expression for gives
u (1 - 2 L+ )
= u + ( 2x - x2)u2 - ( - )u3 _.::.._
2L2 L L2 2L 2L2
I
This expression can be checked to confirm that N; =

1 x; at and N; =0 at points other than i.
The illustrated tapered beam is of rectangular cross section, of width

b, h1(1+ x/L). t
w. I(x)

and tapering depth Construct an expression for the
displaced shape that satisfies exactly the governing differential "1
Mz1
equation
Fy1 M
Fy2
L
The moment of inertia is
I
_ bhj (1+-x)3
--
12 L
Thus
Ebhf d2 [(1+-x)3 -d2u

12 dx2 L dx2]
---- =0
or
d2[( + x)3 dzu] =

dx2 i 1 dx2 o
Integrating this expression twice,
( 1+
x ) 3
dzv
dx2
= Ci + C2x
L
where Ci and C2 are constants of integration. Rearranging,
dzv
=
( 1+
x ) -3
Ci + x 1+
( x ) -3
C2
dx2 L L
By integration,
L2
du
dx
=
-
L
2 ( x
1+-
L
Ci
2+ 2 ) r ( x
1+-
L
1
) 2 -
( x
l+-
L
1
)] C2 + C3
Integrating again,
L2 r( ( )]
)
Ci
( )
3 1 x
v = 2 L + In 1+ C2 + C3x + C4
x -
x L
1+- 2 l+-
L L
This expression is evaluated at the end points (vi = vat x = 0, v2 = vat x = L), differentiated
and evaluated at the end points (dvldx = Oz1 at x = 0, dv!dx = 8,2 at x = L), and the resulting
equations are solved for Ci. C2, C3, C4 Substitution of the solution for Ci. ... , C4 into the
equation for u gives the desired expression in terms of shape functions (Problem 7.4).
7.i.3 Characteristics of Shape Functions
Some observations should be made regarding the foregoing. First, although polynomial
expressions were the appropriate bases of the shape functions for simple axial and
flexural members (Eqs. 7.5 and 7.8 and Ex. 7.1), a logarithmic expression gave the
exact basis for the tapered beam shape function (Ex. 7.2). Other types of functions,
such as trigonometric expressions, also find application. Nevertheless, polynomial func
tions are prevalent and are used both when they represent the exact displaced shape
and in approximation of logarithmic, trigonometric, or other functions. Examples of
the latter are given in Section 7.3.
Second, the above shape functions (Eq. 7.6 and 7.7 and the results of Examples 7.1
and 7.2) each include representation of both rigid body motion and elastic deforma
tional behavior. These two classes of behavior are seen more clearly in the underlying
polynomial expression. For example, in the case of the axial member (Eq. 7.5) the
term a1 portrays rigid body motion and a2x embodies the strain. Jn the flexural member
(Eq. 7.8), these behaviors are characterized by ai. a2x and a3x2 a4x3, respectively.
Third, since shape functions are usually nondimensional mQftipliers of the joint dis
placements, it may be convenient to express them in terms. of nondimensional coor
dinates. For the axial member, for example, one can introduce the coordinate g = x/L,
which then has a range of 0 to 1 from one end of the element to the other. Equation
7.6 can then be written
u = ,(1 - g)u1 + gu2

In the case of flexural members, there is a mixture of translational and angular dis
placements (Eq. 7.9). The multipliers of the angular displacement must have the units
of a length dimension. Thus, N3 can be written as x(l - x/L)2 = x(l - g)2, which
mixes the x and g coordinates. To avoid this one can express the angular displacements
as OiL and 0 L, in which case N3 (x/L)(l - x/L)2 (1 - fl2. This mode of
2
= =
7.2 Virtual Displacements in the Formulation of Element Stiffness Equations 181
expression of the angular displacement is advantageous in scaling a system of equations

in attempts to improve solution accuracy (see Section 11.6.4).
The approach to the construction of element shape functions taken in this section
can be summarized as follows. The displaced state of the element is first described in
terms of parameters a;, one for each element joint displacement. The choice of the
displaced state may be based upon an intuitive understanding of the behavior of the
elements, solution of the governing differential equation (Ex. 7.2), or simply through
the choice of a polynomial or other expansion containing as many terms as there are
element degrees of freedom. The expression for the displaced state is evaluated at
each degree of freedom d;, producing equations that are sufficient in number to de
termine a; in terms of d;. The multipliers of the joint displacements d; in the resulting
expression for the displaced state are the shape functions.
There is a more elegant way of producing expressions for the displaced state of
elements, directly in the desired form a = LNJ{..:1}. This is through the use of interpo
lation formulas that fit an expression for a variable through specified points at which
the variable is known. The interpolation approach is detailed in Reference 7.1. The
route of first selecting an expression in terms of parameters a; and then eliminating
these parameters has been adopted here for simplicity and because it makes clear the
shape of displacement upon which each term of LNJ is based.
7.2 VIRTUAL DISPLACEMENTS IN THE FORMULATION

OF ELEMENT STIFFNESS EQUATIONS
7.2.1 Construction of Expressions for Real and Virtual Displacements
Having established expressions for the displacement of any point on an element in

terms of the element node-point displacements, we now employ these expressions in
the principle of virtual displacements to produce element stiffness equations.
The expression for internal work is written in terms of strains. To obtain these, the
displacements must be differentiated in the manner appropriate to the type of struc
tural action under study. The relevant strain-displacement equations were discussed
in Section 6.3.2. For an axial member, for example, the displacement is u and the strain
ex = duldx. For torsion, the displa'te nt is the rotation about the x axis, Ox, and the

straip {3 is the rate of change of this rotation with respect to the axial coordinate, that
is, {3 = d Ox ldx . For beams, the displacement is denoted as v and the "strain" is the
curvature Kz = d2v/dx2 ,
In general, then, the strain ( e ) is obtained by differentiation of Equation 7.1 and we
symbo)ize this operation by
e = a' = LN'J{..:1} (7.10)

where the prime denotes the appropriate differentiation with respect to the spatial
coordinate. This may require first derivatives (as in the case of axial and torsional
elements) or second derivatives (as in bending). Also, we note from Equations 7.6-7.9
that only the shape functions LNJ are functions of the spatial coordinates and are
therefore the only terms subject to differentiation. Since the joint displacements are
specific values and not functions of the coordinates, they are not subject to
differentiation.
To illustrate Equation 7.10 we first examine the axial member where, from Equation
7.6,
so that
(7.lla)
Similarly, for torsion, from Equation 7.7,
f3
=
d8x
dx
=
l -
_!_ _!_
L L {J }8x1
8x2
(7.12a)
For the beam element, from Equation 7.9 (with "strain," e = K, = d2v/dx2),
Kz = : = l:2 ( 1 -
) I I( ) I :2 ( ) I I ( ) J ':
- 1 - 1 - 2
j)
8,1
(7.13a)
(The order of terms in the displacement vector has been changed to conform to the
order employed in the stiffness matrix derivation of Example 4.4).
To form the complete statement of internal virtual work, it is necessary to have
expressions for the virtual strains. The approach taken is to adopt the same form of
displaced state for the virtual displacements as for the real displacements. Thus, par
aphrasing Equation 7.1,
8A = LNJ{.Sd} (7.14)
where {.Sd} denotes the element node-point virtual displacements. The virtual strain,
&, is then
8e = LN'J{8d} (7.15)
The specific forms of this expression for axial, torsional, and bending behaviors are
8e x = (7.llb)
.S{3 =
l-i iJ{:::J (7.12b)
.SK, =
l:2 ( - )II( - )\ : ( ) I I ( )
1 1 2
- 1 -
2
J{:}
88,1
(7.13b)
7.2.2 Virtual Displacements Formul& f<;: an Element Stiffness Matrix
The expression of the principle of virtual displacements can now be written for an
individual element, treating it as if it were a structure in isolation. The basic statement
of this principle was established in Section 6.2 as
8W = .SWext - .SW;n1 = 0 (6.13)
Also, it was shown in Section 6.3 that
.SW;n1 =
JvoI LcSeJ[E]{e} d(vol) (6.23)
7.2 Virtual Displacements in the Formulation of Element Stiffness Equations 183
where {e} and L5eJ are the real and virtual strains that are characteristic of the mode
of behavior under study, for example {e} = ex for axial behavior. [E] is the relevant
elastic constant. The element is loaded by the joint forces Fi. ..., F;, ... , Fn. (In Sec
tion 7.5 we take up the matter of distributed loads and loads applied at points other
than the element node points).Thus, tSWext = L7= 1 tSf:.. ;F;, and the virtual work principle
becomes
L01 L5eJ[E]{e} d(vol) =

M;F; =
LtSJ:1j{F}
By substitution of Equation 7.10 and 7.1 5 we have, for an element described by n
degrees of freedom, 1
(7.16a)
or
LtSi:1J[k]{i:1 l =
LtSi:1J{F} (7.16b)
where
[k] =
[Jvol {N'}[E]LN'J d(vol) J (7.17)
Under the condition that the virtual displacements are completely arbitrary,
[k]{i:1} =
{F} (7.18)
Equation 7.17 is the general expression for an element stiffness matrix, derived through
use of the virtual displacements principle.Equation 7.18 represents the set of element
stiffness equations to which it refers.
7.2.3 Application to Standard Axial, Torsional, and Flexural Elements
For an axial member of uniform section (Fig. 7.1), [ E] =

E, d(vol)
expressions for real and virtual strains are given by Equations 7.l la and 7.l lb. Thus,
by substitution into Equation 7.17,
= A dx, and the
[k] [r[-/j+I IJA ] dx

E : [ : -:J
_
which is the already established result for the axial force member.
For the torsional element [E] = G, d(vol) is replaced by J dx, and the real and virtual
strains are defined by Equations 7.12a and 7.12b. Again, using these in Equation 7.17,
[k]
[L'[-/j+I IY ] dx

[ : -:]_
This was derived, in a different manner, in Section 4.5 (Eq. 4.27a).
'Generally, [NJ and [N'] are matrices. Since in this chapter we deal only with separate modes of behavior,
for each of which {e} has a single component, they are vectors here.
For the beam element, [E] = E and, after integration across the depth, d(vol) leads
to Iz dx. The real and virtual strain expressions are given by Equations 7.13a and 7.13b.
After substitution in Equation 7.17,
(l _ 2x )
L2 L
[k]
i ( - 1)
=
( 2x _
1
)
L2 L
i (
[
- 2 )
[k] =
E:z
V2
WL'
Sym.
-6/L
(Jz2
4
V1
-12/L2
6/L
12/L2
!
-
6/L
4
L
]
In this case, the external virtual work is L &2 88z2 &1 88ziJ
LFy2 Mz2 Fy1 Mz1Y The result is the one obtained by alternative means in Ex
ample 4.4 (see also Eq. 4.32).
7.3 NONUNIFORM ELEMENTS
Practical structures may contain members of nonuniform cross section, as shown in

Figure 7.6a. It is, of course, possible to represent such members by a series of uniform
section elements, as shown in Figure 7.6b. But this may require an inordinate number
of unknown displacement components to yield a solution wherein the error due to the
approximation is within acceptable limits. The formulation of exact stiffness equations
for a nonuniform element is an alternative, but, as Examples 7.2 and 7.9 disclose, this
can be rather complicated. A convenient approach, described below, uses the principle
of virtual displacements. It enables the construction of exact or approximate stiffness
relationships for nonuniform members.
The expression for the element stiffness matrix (Eq. 7.17), derived from the principle
1- (a)
(b}
Figure 7.6 Continuous beam with nonuniform cross section. (a) Actual structure.
(b) Stepped representation.
7.3 Nonuniform Elements 185
of virtual displacements, requires the derivatives of the displacement shape functions

LN'J and a description of the element geometry as embodied in d(vol). These repre
sentations of element behavior and geometry can be either exact or approximate. (See
Section 6.4 for the requirements of an exact solution).
To illustrate the approximate procedure we examine the case of a linearly tapering
axial member whose cross-sectional area at any point is given by
where A1 is the cross-sectional area at point ,1 and r is a parameter that can be adjusted
to give a specific ratio of end areas. This is similar to the description of taper as used
[N1
for the eleme
= ( 1 -x/L), N2 x L6.6.]
: nt of Figure
= !
Using the shape functions for the uniform axial member
, the expression for the approximate stiffness matrix is
[k] [Lll)+I I J A. ( 1 - ) dx l
and, upon integration
[k] =
E
:i ( 1 - )[- -J
The result, in this case, is a stiffness matrix that is of the same form as the stiffness
matrix for a uniform axial force element with "effective area"
is due to the circumstance of constant strain arising from the chosen approximate
A = A1(l - r/2). (This
displaced state. It does not occur for flexural elements where the strain due to the
chosen displaced state, in the form of the curvature, usually varies over the length of
the element.)
In this case the exact displaced shape can be found through solution of the differ
ential equation for axial behavior. This equation has not been developed previously,
but it is established quite easily. The condition of equilibrium for a differential segment
d(Aux)ldx 0 (see Section 6.5). The strain-displacement equa
tion is ex
gives
= u
of an axial member is =
d / dx and, by Hooke's law,' <Tx Eex. Combination of these relationships

=
( ddxu) 0 .!!._ EA
dx
= ( 7.19)
conditions u u1 x A0 A1(lu -rx/L)

In the present case, with
at u2 x, L
=
and at
the solution of this equation for boundary
is
U - [ 1 ln(lln(l--rx/L)r) ] U1 [lnln(l
(l - rx/L) ] Uz
= = = =
_ _
- r) +
Using this i n the formula for the stiffness matrix (Equation 7.1 7 ), w e obtain
[k] -L. EA1

=
r [
ln(l - r) -1 1 ]
1 -1
For r = 11 2 this gives a value of 0.7213EA1/L for each stiffness coefficient, while the
approximate formulation gives 0.75EA1/L, an error of 4%.
Example 7.3 illustrates the approximate approach for the case of the tapered can
tilever beam.
Determine the deflection (v ) at point 1 of the member of Example 7.2 due to Fy1 using an
(M,1 =
approximate formulation, based on the shape functions of a uniform beam (Equation 7.9).
0). The member is supported as a cantilever at point 2.
Since v 2 = = 0,2 0, the relevant stiffness equations are
{I}=[:: ::JU,J
with I= ( bhf/12)(1 + x!L)3 and using the curvature shape functions of Equation 7.13a,
k11 = r [2 ( ) r E; ( ir =
_ 1
3b
1
+
dx Eh13
3
(L (2x
k12 = k21 =
Jo L2 L) (3x ) ( ) dx
_
l
X
L L
_ 2
Eh13b
12
l
+
.::_ 3
L
=
29Eh13b
20L
2
= J(Lo [(3x ) ] ( )
2 3
Eh13b l + .::_ 3Eh13b
k22 _
2 dx
=
L L 12 L 4L
]{ v1}
Thus, upon evaluation of the terms of the stiffness matrix,
{ }= [
Fy1
0
Eh13b
20L3
81
29L
29L
15L2 0,1
The solution of these equations gives
V1 = 0.06684
Fy1L3
Eli
This compares with the exact solution 0.06815Fy1L3/E/1 (1.92'}'.o error), obtained as a solution
to Example 7.9
The solution errors in Example 7.3 and in the above tapered axial member are due
to the failure of the chosen polynomial displacement state to satisfy the condition ot
equilibrium. For the axial member, this condition is represented by Equation 7.19,
u (1 - rx/L)u1 (rx/L)u2,
( [ (1 - rx ) u1
which gives, for =
+
r2
A1
d
dx
1
rx
- L
E
d
dx ) L
+
rx
L
u2
J = EA1
L2
(u1 - u2)
The term on the right is zero only when r equals zero. A similar exercise for the shape
function used in Example 7.3 will disclose that it does not meet the equilibrium con
dition for flexural behavior (Problem 7.9). As the length of these elements becomes
smaller, because of the refinement of the analysis grid, the error in equilibrium be
comes smaller and the exact solution is approached.
7.4 NONUNIFORM TORSION

The stiffness matrix for an element in pure torsion was developed from simple strength
of materials principles in Section 4.5.2 and by the virtual displacement approach in
Section 7.2.3 of this chapter. It was pointed out in Section 4.6 that whereas that matrix
is adequate for the analysis of many systems in which torsional effects are small, it is
limited by the fact that it neglects resistance to cross-sectional out-of-plane warping
(Figure 4.13). Warping resistance can be significant in the response of many frames
consisting of members of open cross section, such as the wide flange steel shape. In
deed, it can be the dominant factor in their resistance to torsion and combined tor
sional-flexural effects if warping deformation is restrained in any way. The relevant
elastic theory is developed in References 7.2 and 7.3, and its application is illustrated
in Reference 7.4.
7.4.1 An Element Stiffness Matrix
In the following the principle of virtual displacements is used to develop an element

stiffness matrix that accounts for warping restraint in bisymmetrical wide flange shapes.
Elementary engineering theory is used in formulating the governing differential equa
tion, but the result is the same as that obtained by the more general methods of the
cited references. The numerical solution of the equation is approximate, but it will be
found to be in good agreement with analytical solutions of representative problems.
A wide flange member subjected to an applied torque at the right end is shown in
Figure 7.7. If there is no restraint against longitudinal displacement of the fibers, its
rate of twist is constant, all cross sections are in a state of pure shear, and the stiffness
matrix of Equation 4.27 and Section 7.2.3 applies. But the cross sections will warp out
of plane uniformly, and as the member twists, its flanges will displace laterally as
indicated in Figure 7.7a, in which twisting but not warping is prevented at the left end.
If, however, longitudinal displacement is restrained in any way, the rate of twist is no
longer constant, and both normal and shearing stresses can exist on cross sections.
Cross-sectional out-of-planf warping can still occur, but it will vary along the length.
As a consequence, the flanges will bend as they displace laterally as in Figure 7.7b, in
which both twisting and warping are prevented at the left end. In this condition, known
as nonuniform torsion, internal resistance to the applied torque becomes more com
plex, but it can be analyzed by the introduction of an additional degree of freedom,
the rate of twist 8, and its conjugate variable, a self equilibrating action B, called the
bimoment.
Without undue simplification, the wide flange member can be viewed as one in which
the applied torque is resisted by a combination of the continuous flow of shear stresses
on a section and the resistance of the flanges to lateral bending (Figure 7.8). The former
- --_ ,____
11r==::==== M x2_._..
lz
/xi._ 111------------ _
_ __,:1-2>
, 11 2
(a) Warping unrestrained (b) Warping restrained

Figure 7.7 Torsion member.
I h
_J
y
(b)
A-A
(c)
Figure 7.8 Nonuniform torsion.
is commonly called St. Venant torque, after the developer of the theory of pure torsion,
and the latter warping restraint torque or just warping torque. The normal stresses due
to flange bending must be self equilibrating since the net action on the cross section is
a torque. But, as seen in Figure 7.8b, the product of the flange shears associated with
transverse bending and the effective depth of the section constitutes a twist resisting
couple. For a small angle of twist the quantities of interest can be expressed as follows:
From Section 4.5.2, noting that {3 = 8, the St. Venant torque is
Tsv = GJ8 (7.20)

In the development of the warping restraint torque, we will apply the conventional
moment-curvature relationship to the flanges. Noting that the flange moment of inertia
is approximately half that of the section, and letting w1 = 8xhl2 (Figure 7.8c)
(d2w f \ Efyh2 (d28x)

M1h Eli h
dxz } - 4 - dx2
= =
Defining the self-equilibrating action M1h as the bimoment, B, and designating the
section property lyh2/4 as a warping constant, Cw, of the wide flange section, we have
(7.21)
Values of the warping constant for other sections may be found in the cited references.
Now, noting that Twr = Hh = -(dM11dx)h = -dB/dx
(7.22)
Thus, since the total torsional resistance is the sum of the St. Venant and warping
resistance torques we have, from Equations 7.20 and 7.22:
(7.23)
In employing virtual displacements to formulate the element stiffness equations for

a member of this type, the internal and external virtual work expressions developed
in Section 6.3 may be used for the virtual work associated with St. Venant torsion. For
warping torque we have, using Figures 7.7b and 7.8 and the above relationships
(7.24)
in which Mt1 is the flange moment at end 1, and
which can be reduced to
(7.2S)
Thus, for a torsional element with warping constraint at both ends, the external virtual
work of nonuniform torsion is 8Wext = L 88x1 88x2 881 882JLMx1 Mx2 B1 B2JT.
From Equation 7.17 and the internal virtual work equations for St. Venant and
warping torsion, it follows that the stiffness matrix for such a four degree of freedom
element can be expressed as
[k] = [f {N'}GLN'jl dx + r {N"}ELN"JCw J dx (7.26)
For a torsional member in which warping is unrestrained, the rate of twist is constant.
Therefore, it was correct to employ a linear description of the twist in the development
of the shape function used in the earlier analysis (see Equation 7.7). This is not the
case in nonuniform torsion, however. In this the shape functions must be approximate.
In the following, both the real and the virtual angle of twist are represented by a cubic
polynomial. Thus
Bx = ai + a2x + a3x2 + a4x3

(7.27)
8 = a2 + 2a3x + 3a4x2
Noting the identity of these expressions and those of Equation 7.8 for the deflection
and slope of a beam element, it is seen that LNJ = LN1 N2 N3 N4j, in which the
shape functions are those of Equation 7.9. Substituting the derivatives of these in
-
Equation 7.26 and integrating results in
Bxl Bx2 81 82 Bxl Bx2 8 1 82

6 6 1 1 12 12 6 6
L2 L2
-
SL SL 10 10 L L
6 1 1 ECw 12 6 6
[k] GJ L2 (7.28)
--
+
=
SL 10 10 \ L L L
2L L
Sym. --
Sym. 4 2
lS 30
2L
4
15
190 Chapter 7 Virtual Work Principles in framework Analysis
The previously developed 12-degree-of-freedom stiffness matrix of Equation 4.34

can be modified and augmented by the elements of [k] to form a 1 4-degree-of-freedom
stiffness matrix for the linear elastic analysis of members subjected to combined axial
force, bending, and nonuniform torsion. The result is Equation A.55 of Appendix A
in which the St. Venant torsional terms of Equation 4.34 have been replaced by the
above, and columns and rows related to the rate of twist and bimoment have been
added. For our present purpose-the linear analysis of members under torsion
alone-the four-degree-of-freedom stiffness matrix of Equation 7.28 is sufficient.The
essential stiffness equation may be written in a form that emphasizes the relative roles
of St. Venant and warping torsion. Thus, letting
(7.29)
and combining terms in [k], we have
l l
( 12a) -(
+ + 21 a ) ( L lO + 6a ) ( L lO + 6a )
l l
r'l (")
Mx2 GJ
( + 21 a ) C -L o + 6a ) C -L o + 6a ) ex2
(7.30)
=
L
B1
B2
Sym. (u
2
i5
+ 4a ) ( Li -
30
1
+ 2a ) e;1
e;2
( u
2
i5
+ )
4a
7.4.2 Application and Examples
Judicious treatment of torsion in frame members requires consideration of the

following:
1. The relative magnitude of the warping restraint and St. Venant effects. As seen
from the parameter a of Equations 7.29 and 7.30, for a given material the significance
of warping restraint increases with the ratio CjJL2 and, for a given cross section, the
effect of warping torsion decreases with increase in beam length. But also, since the
bimoment is self-equilibrating, the effect of a given source of restraint is, in principle,
a localized one.For deep, relatively thin-walled wide flange sections, the warping con
stant is so much larger than the torsional c;onstant that warping restraint such as fixing
the flanges at the end of member may have a major influence on the torsional response
of spans of practical length. It may still be significant for shallower, stockier sections.
But for closed sections, such as the rectangular tube, it is generally inconsequential.
Warping resistance of open sections in which the component elements meet at a point,
such as angle and cruciform sections, is also negligible.
2. The boundary conditions. In principle, ,torsional and flexural support conditions

are independent of each other.For example: the torsionally simple supports of Figure
7.7a will provide flexural fixity about the section's major axis if the two idealized hinges
which constitute the support are restrained in the x direction. On the other hand, the
torsionally fixed support of Figure 7.7b could be flexurally simple with respect to ro
tation about the section's minor axis. This would be the case if the member were
welded to a plate thick enough to prevent warping of the member but that itself would
be supported by hinges that would permit it to rotate about the member's minor axis.
Additional questions arise in the definition of torsiohal boundary conditions at the
junction of noncolinear members: for example, at a beam-to-column joint. For these
reasons, we shall consider only the two extremes of torsional support: warping-free
and warping-fixed at the ends of assemblages of colinear elements. A scheme for ac
commodating noncolinear elements and conditions of partial warping restraint is pre
sented in Reference 7.6, and other possibilities for defining boundary conditions are
described in Reference 7.7.
3. Solution accuracy. It should be noted that the underlying assumption of a cubic

polynomial for the angle of twist does not satisfy a requirement of internal equilibrium.
For the member considered, in which there is no load between the nodal points,
dM)dx must be equal to zero. Substitution of the derivatives of the cubic polynomial
of Equation 7.27 into Equation 7.23 reveals that this condition is not satisfied. Also,
the accuracy of the terms of the solution obtained through the use of Equation 7.30
decreases as the order of the derivative to which they are related increases. Thus the
calculated distribution of the bimoment and the warping torque may not be as reliable
as the nodal forces, the angle of twist, and the St. Venant torque: But Equation 7.30
is efficient in computation and it can produce results in good agreement with known
analytical solutions (Refs. 7.5 and 7.6). Further, the use of higher order approximating
polynomials may lead to other, more serious, computational problems (Ref. 7.8). For
these reasons, use of the cubic polynomial can be recommended. Methods for over
coming some of its shortcomings are incorporated in the following examples.
In Example 7.4, Equation 7.30 is applied to the analysis of a shaft using a single
element idealization of the member. The distribution of twist is determined from the
calculated end effects and the cubic shape function. The derivative of this function is
then used in the calculation of the St. Venant torque (Eq. 7.20). To avoid reliance
upon the higher derivative of Equation 7.22, the warping torque is determined as the
difference between the total and the St. Venant torque.
Example 7.5 illustrates the influence of the relative values of ex. For the case in which
warping is restrained there is little difference in the twisting of the three sections. But
comparison of the warping fixed and warping free response shows that St. Venant
torsional resistance dominates in the shallowest section, whereas warping resistance
does in the deepest one. Related to this is the fact that the effect of tensionally fixing
the left end decays most rapidly in the shallowest section. In this example, a- four
element idealization of the members was used. Member subdivision of this sort is
another way to improve the accuracy of an approximate solution. In this case, however,
there is very little difference between the results obtained by it and by those from the
single element idealization of Example 7.4. Both are excellent approximations of the
analytical solution of the differential equation (see Ref. 7.4).
f" 1
The member shown is fixed against warping and twisting at the left end and free at
_j12--
1
the right end.
W14x68
1. Determine the ractions.
2. Plot the angle of twist and the apportionment of T between St. Venant and i---- L = 11'-4
warping torque.
1. Section properties: J = 3.02 in.4, Cw = 5380 = 29,000 = 0.3

in.6, E ksi, 11
0. = (2.6 5380)/(3.02 1362)= 0.250

x x
From Equation 7.30

00 -1.600L2 { 8x2}
{T}0 = L [ -1.4.6200L l.133L J 0;2
GJ
Thus
Bx2 0.515 TL 8;2= 0.727 T

=
GJ
,
GJ
and
{Mxi}= T[-4.200 1.600L2J{0.0.52715}= { 4-T4 }

B1 -l.600L 0.467L -0. 8 TL
Reactions: M1= 0.484 T x 136/13.32 = 4.94 T
7.9
2. Twist and torque distribution:
From Equation
ox= [3(zY - 2(zYJox2 + [ (iY (iYJL0;2 -
o;= [ 6(i) - 6(zYJ i2 + [3(iY - 2(i) Jo;2

and Equation 7.20:
Angle of twist
-:-oso
Torque distribution
l:dfff J: I
Twr = .273 T
T
Tsu= .727 T
0 L/4 L/2 3L/4 L

...... "'--
2
1
Compare the torsional response of three shafts of essentially equal weight:
L W8 x 67
f
"'
3
I
2. W14 x 68 L = ll'-4 -----l
3. W24 x 68
E = 29,000 ksi, v = 0.3
Section properties:
W8 x 67 W14 x 68 W24 x 68
5.06 in4 3.02 in.4 1.87 in.4
1440 in.6 5380 in.6 9430 in.6
a 0.040 0.250 0.709
Using Equation 7.30 and four element

discretization:
1.% (W 24 x 68)
2.08 (W 14 x 68)
1.93 (W 8 x 67)
0 L/4 L/2 3L/4 L

(O)T) x 1<>3
rr------=====, .987 T
.734 T
.442 T
TSVIT
0 L/4 L/2 3L/4 L

Tsv/T
Example 7.6 is another case for which analytical results may be found in Reference
7.4. Again, torsionally fixing the ends has a profound effect on the twisting of the
member.
------
A torque, T, is applied at the quarter point of the member shown. Com
pare its torsional/ response for warping free and warping fixed conditions
":
at both ends.
L/4 3 L/4
E = 29,000 ksi, v = 0.3

L = 11'-4"
..
Use Equation 7.30 and four element discretization: Warping free .164
o '-0::::::'----=== o
0 L/4 3L/4 L
.163
-.106
7.5 LOADS BETWEEN NODAL POINTS AND INITIAL STRAIN EFFECTS

A GENERAL APPROACH
The principle of virtual displacements, when intermediate loads act on members or

when initial strains are present, gives a particularly convenient transformation of these
loads and initial strains into equivalent joint forces. Such transformations were devel
oped earlier (Sections 5.2 and 5.3) for simple cases. The formulas of Sections 5.2 and
5.3 can be duplicated with ease by means of the virtual work approach. The significant
attribute of the virtual work approach, however, is that it is able to deal with rather
complicated conditions in a unified and convenient manner.
We examine first the treatment of intermediate loads on members. Consider, for
purposes of the development, a framework member subjected to a distributed load q,
oriented arbitrarily with respect to the x, y, z axes. Consider also the virtual displaced
state of the element, designated by 8.i. For axial behavior, for example, 8.i = &t. The
external virtual work of the load q in a differential length of the element dx is the dot
product 8.i q dx; and for the total length of the element, it is (from Equation 6.24)

(7.31)
Adopting now the description of 8.i in terms of the joint virtual displacements (Equa
tion 7.14), we have
(7.31a)
7.5 Loads Between Nodal Points and Initial Strain Effects-A General Approach 195
so that
(7.32)
This is a general formula for the effective nodal loads {FE}. Clearly, the joint forces
{FE} produce the same virtual work as the distributed loads. For this reason it is cus
tomary to describe such joint forces, when calculated by use of Equation 7.32, as
"work-equivalent" loads.
A specific case of the above is the uniformly loaded axial member, where q = qx
and
Hence, by Equation 7.32
In this case the equivalent nodal loads are those that one would obtain by simple
apportionment of the total load to the two joints.
The more significant conditions are encountered in the case of flexural members,
where 8.li = 8v and the distributed loads (qy) act in they direction (Fig. 7.9). Also,
concentrated loads P3, . , P;, ... , P, may be present. Thus the external virtual work
is, using Equation 7.31 for the distributed load portion,
where 8v symbolizes the virtual transverse displacement at all points along the beam,
and 8v; is the virtual transverse displacement at the point of application of P;. There
fore, 8v; denotes the evaluation of 8v at point i.
As in previous developments, we now assume that the virtual displaced state of the
beam element can be described in terms of the four joint displacements 8v" 8u2, 8(J0"
and 80,2, that is, in the form of Equation 7.13b.
Figure 7.9 Flexural member subjected to intermediate loads.

Substitution of Equation 7.13b into the expression for the virtual work of the applied
loads results in
J N1qy dx 2: N\P;
J N2qy dx 2: NP;
+ (7.33)
J N3qy dx 2: NP;
J N4qy dx) 2: NP;

where the summations extend from i = 3 to i = r, and N\, . . . , N designate the shape
functions evaluated at each point i. This can be written more concisely as
. (7.33a)
where {FE), as in Section 5.2, is the vector of effective nodal forces corresponding to
the distributed loads qY and the within-span forces P;. Thus
(7.34)
and
J N1qy dx 2: N\P;
J N2qy dx 2: NP;
{Fd) = (7.34a) {Fe}= (7.34b)
J N3qy dx 2: NP;
J N4qy dx 2: fl/pi
If Equation 7.33a is used in the principle of virtual work (Eq. 6.13), as a supplement
to the 8Wext of the nodal forces (L8.1J{F)), the element stiffness Equation 7.18 is ex
tended to2
(7.18a)
Initial strain effects, which have been discussed in Section 5.3 bear a similarity to
distributed load effects in that they are represnted by vectors of equivalent joint
forces. They derive, however, from the internal virtual work rather than from the
external virtual work.
Initial strain effects are basically represented in the stress-strain law. Consider, for
purposes of explanation, direct strain behavior in the x direction. When initial strain
is present, the total strain is the sum of the strain due to stress, a)E, and the initial
strain, e, that is
ax .
ex = - + e'x
E
Solving this expression for stress, we obtain
CTx = Eex - Ee (7.35)
2ln Equation 5.21, {PE} was used to designate equivalent nodal loads whereas in Equation 7.18a {FE} is used
for the same quantities. The only difference is one of symbolism: Equation 5.21 was developed in the context
of a global analysis example and Equation 7.18a from the analysis of element joint forces.
7.5 Loads Between Nodal Points and Initial Strain Effects-A General Approach 197
and generally,
(7.35a)
Consequently, the internal virtual work, given previously by Equation 6.23, is now
oWint =
fvol Loej{ul d(vol)
L01 Loej([E]{el - [E]{ei)) d(vol) (6.23a)
=
Jvoi Loej[E]{el Jvol Loej[E]{ei}
d(vol) - d(vol)
Now, the first integral on the right side is the same integral as established earlier for
conditions where initial strains are absent and is the basis for the element stiffness
equations (see Section 6.3, Equation 6.23). The second integral leads to a vector of
"initial forces" {Fil by substitution of the expression for the virtual strains in terms of
the joint displacements, given by Equation 7.15:
Loi oe[E]{ei} d(vol) =

LoJ
Loi {N'J[E]{eil d(vol)
(7.36)
=
Lo.:j{Fil
where
(7.37)
Thus, upon substitution of the results of Equation 7.36 into Equation 6.23a, and ap
plication of the principle of virtual displacements, we have
(7.18b)
For an axial member, for example, {N'I = -i {-} and [E] = E. Suppose the initial
strain is due to a uniform temperature change T. Then {eil = aT and the initial forces
are
This is the result obtained in Section 5.3, Equation 5.23, in a more intuitive manner.
The present development, however, is based on Equation 7.37, which permits a more
general formulation. For example, initial strains that are a function of the coordinates
of the element can be handled.
We consider, in the case of bending, a member with a bisymmetrical cross section
and a linear variation of initial strain across the depth (h) of the element. These strains
have a value of (Ad2) at the top and bottom of the section and zero value at the
neutral axis. It was shown in Section 5.3 that for this case the curvature due to initial
strain is Ki = eiJh. Adopting, in the formula for initial strains, Equation 7.35, the same
analogies adopted previously for the flexural element stiffness matrix, we have
{Fil = r {N'IEIKi dx (7.38)
{N') now represents the second derivatives of the shape functions for transverse dis
placement, such as those given in Equation 7.13a for the simple flexural element.
Example 7.7 illustrates the application of the work-equivalent load approach to cases
that were studied by an alternative approach in Chapter 5. Example 7.8 illustrates the
calculation of the initial forces for a linear temperature change across the depth of a
flexural element where the change varies along the length of the member.
EXAMPLE7.1
Verify, by use of the virtual work approach, the effective nodal loads {FJ for the b
r,
beam element and loadings shown. Results for these cases are given in Table 5).
1 tp 2
Case a
q
t I I I I I t
I.
Case b
L .
For Case a, using the shape functions of Equation 7.9.
1[ - 3(i) 2 2(iYJ lxG P

+
b2 (3 b)P
3
L
a + Fi1
(iY(3 - ) lxap L3 3b)P

a2
(a+ F0_
{FJ =
x(1 - Lr1xa ab2 P

- M

p Lz
x \ - 1) I p --a2b P M:2
L L xa
Lz
For Case b,
qL
r [1 - 3(iY 2(iYJq dx+ 2 yl
rF"
qL
r [3(iY - 2(iYJq dx 2 y2
FE
qL2
r x(1 - iYq dx 12 M:1
lL x(: i)q dx
-
- qL2
12 M:2
The simple flexural element ( Figure 7.3), D.T (1 - x/L)(D.Ti)

which in this case is of bisymmetrical cross section and
of depth h, sustains a temperature difference across the depth given by
+ (x!L)(6.T2).
initial forces.
(The
=
neutral axis temperature change is zero. ) Calculate the relevant vector of

7.6 Virtual Forces in the Formulation of Element Force-Displacement Equations 199
It is shown, in Section 5.3, that the initial curvature in this case is

Equation 7.38 and using Equation 7.13a for {N'}, we have
K; = a(!l T)lh. Hence, by
r :2 ( - )[ ( -i)<llT1) +I (ilT2)J
1
1
dx
Ela r - :2 ( - )[ ( -i)(ilT1) +I (ilT2)]

. -
{ F' } =
1 i dx
=-
Ela
h hL -L(!l T)1
r I( - 2)[ ( -i)(ilT1) +I (ilT2)J i dx
r I( - )[ ( -i)(ilT1) +I (ilT2)]
1 i dx L(ilT)i
(The ordering of terms in the vector is the same as in Example 7.7)
7.6 VIRTUAL FORCES IN THE FORMULATION OF ELEMENT

FORCE-DISPLACEMENT EQUATIONS
7.6.1 Construction of Element Equations by the Principle of Virtual Forces
We have established that the virtual displacement principle is the basis for the direct
formulation of element stiffness equations. The dual principle of virtual forces is, con
versely, the basis of the direct formulation of element flexibility equations.
Element flexibility equations are of considerable value in practice since they can be
transformed into element stiffness equations by inversion and by supplemental matrix
operations detailed in Section 4.4. There are many conditions under which the for
mulation of element flexibility equations is straightforward, while the direct formula
tion of the corresponding element stiffness is extremely difficult. Such conditions in
clude the formulation of the beam element for combined transverse shear and flexural
deformations, tapered members, and elements with curved or irregular axes.
The principle of virtual forces was stated in Section 6.5 as follows
(6.32)
where
8Wfn1 = Jvol
LooJ(Er1{cr} d(vol) (6.31)
The real ( CJ' ) and the virtual ( oo) stresses are, for a given mode of structural behavior,
single components of stress: uniaxial direct stress for axial or flexural behavior and
shear stress for simple torsion. [E]-1 denotes the inverse of the relevant elastic con
stant. For the particular case of element flexibility formulation, where the joint dis
placements are Ai. ... , A;, ... , A1 and the virtual joint forces are 8Fi. ... , 8F;, ... ,
8F1, sw:.1 becomes
f
sw:.1 = 2: ( 8F;)(A;) =
L 8FtJl&J (7.39)
i=l
We have changed the symbolism from {Pf) to {FJ} because the latter is reserved for
designation of element joint forces. The subscripts f emphasize that the virtual joint
forces and associated displacements refer to the points that are free to displace on
elements that are supported in a stable, statically determinate manner. Thus, the num
ber of node points, f, represented above is s fewer than for the stiffness formulation,
where s designates the number of rigid body motions. For flexure, the displacements
include angular displacements 8; for which the corresponding virtual forces are the
moments 8M;.
In the following we direct our attention to the complementary internal virtual work
8Wfn1, since this is the source of coefficients h of the element flexibility relationships.
To evaluate 8Wfni for this purpose, expressions must be constructed that give the
virtual and real stresses, L DO'J and {O'} in terms of joint real and virtual forces, !Ftl and
{ 8Ftl An intermediate step, however, is the transformation of the stresses into the
stress resultants of the structural action in question. This step was detailed in Section
6.5 and gives, for axial behavior (Fig. 7.lOa),
8Wfn1 a = r 8Fx (:) dx (6.33)
where 8Fx and Fx are the virtual and real internal axial forces. For a pure torsional
element oriented as in Figure 7.lOb,
8Wfnt I = r 8Mx (;) dx (6.35)
where 8Mx and Mx are the virtual and real torsional moments (about the x axis), and
J is the torsional constant. For a flexural element with coordinate axes as shown in
Figure 7.lOc,
(6.34)
---- Fx -r-- - -Fx- - 2

x, u (b)
[_____ LdxJ
y, v
(a )
x
L
z
(c)
Figure 7.10 Internal force resultants for axial, torsional, and flexural elements (virtual forces 8Fx2, 8Mx2, 8Fy2, 8M,2
corresponding to Fx2, . , M,2 are not shown). (a) Axial force element. (b) Torsional element. (c) Flexural element
(x-y plane).
where 5Mz and Mz are the virtual and real internal moments at an arbitrary point
within the element.
The approach taken in the principle of virtual displacements, where the displace
ment at an arbitrary point within the element is related to the joint displacements by
means of shape functions, suggests the use of analogous functions that relate the in
ternal forces to the joint forces in the principle of virtual forces. Symbolically, this can
be written as
(7.40)
for axial force members, or torsional or flexural members, as the case may be. L QJ, the
vector of functions of the force distribution must be chosen in such a way tht the
conditions of equilibrium are satisfied, a consideration that was emphasized in Section
6.5. It will be shown below that the choice of L QJ to meet such conditions presents no
difficulty.
It is both convenient and appropriate to choose the same shapes of virtual and real
force systems. The equilibrium conditions, cited above, are then sought only once, the
integration of 5W'l'ni is simplified, and the resulting flexibility matrix is guaranteed to
be symmetric. The chosen form of the virtual force system is, consequently,
(7.41)
Substitution of Equations 7.40 and 7.41 into 5W'l'ni yields
(7.42)
where, for the axial force member (Eq. 6.33, Fig. 7.lOa),
(7.43a)
(7.43b,c)
and for the torsional element (Eq. 6.35, Fig. 7.lOb),
(7.44a)
(7.44b,c)
and, for the flexural element (Equation 6.34, Figure 7.lOc),
(7.45a)
(7.45b,c)
Note that each of these elements could have been supported differently, resulting in
different specific forms for {Ff} and {5Ftl Note also that they are portrayed as having
nonuniform cross sections. It is demonstrated in the following that the principle of
virtual forces enables a very expeditious treatment of nonuniform elements.
A convenient illustration of the use of the principle of virtual forces in the formu
lation of element flexibility relationships is given by the tapered axial member (Fig.
6.6). We choose
internal force state is
A FxA1(l Fx-2 rx/L). LQJ
=
= Hence,
The relationship between the joint force and
in this case is merely unity. The flexibility
matrix is a single term and is obtained by the substitution of the foregoing into Equa
tion 7.43a, resulting in
[d] = d22 A1 lL -rxxd IL)

=
E
-
1
o (l
=
-E-
L -r)
A1r ln(l
This is the exact solution for the flexibility coefficient.

To establish the stiffness matrix for this element we now use the procedure of Section
4.4, Equation 4.25. Here, !Fs}
4.25,
= xF i -Fx2, [<I]= so that = -1. Hence, by Equation
[k] = [-2dd 222=11 -d22 J EA1 r [ -J

d22 L -r)
-
-
i
i
=
-
ln(l
1
-1
which is the same as the result obtained in Section 7.3.

It is important to observe why this exact solution was obtained with ease, whereas
the solution by use of the virtual displacements principle (Section 6.4) was considerably
more difficult to obtain. The virtual and real forces required by the virtual forces
principle are given by the equation of statics; the variations that may be present in the
section or material properties , (A,
I , E, etc.) do not affect these equations. Exact
displacement functions for elements with such variations can only be obtained through
solution of differential equations with varying coefficients and are generally of rather
complex form.
Still another simple illustration is given by the construction of the flexibility matrix
for the flexural element, Figure 7.lOc, but with constant section properties. From stat
ics, we can write
M, = L(L - x)
so that LQJ L(L -x)
= lj. Then,
5M, = L(L - x)
With the above relationships Equation 7.45a becomes, for an element of length L,
[d] ; [ f { (L x) }L(L -x)
I
,
1J dx J '
[ r !l (7.46)
This matrix was derived in Example 4.3 by solution of the governing differential equa
tion of beam bending.
Example 7.9 is a further illustration of the above. It deals with the tapered beam
element for which exact shape functions were established in Example 7.2 and an ap
proximate stiffness matrix and a solution for the tip displacement were obtained in
Example 7.3. With virtual forces the exact flexibility matrix is established quite easily.
The exact stiffness matrix can be obtained by inversion of this flexibility matrix and
use of Equation 4.25. The same result will be obtained by use of the exact displacement
functions in the virtual displacements principle, Equation 7.17.
EXAMPLE7.9
Formulate the flexibility matrix for the tapered cantilever beam of Example 7.2 using the prin
ciple of virtual forces. Calculate the displacement u1 due to the force Fy1 and compare with the
solution of Example 7.3.
In this case the relationship between the internal real and virtual moments and the joint real
and virtual forces are
M, = Lx
so thatLQJ = Lx -lj.
From Equation 7.45a,
For M,1 = 0, this gives u1 0.068l5Fy1L3/EI,. The solution for Example 7.3 is
U1 = 0.06684Fy,L3/EI,.
7.6.2 Further Applications-Shearing Deformations and Curved Elements
The principle of virtual forces is a highly advantageous approach in situations in which

more than one mode of action may require representation within the element and in
the case of "line elements." The first is exemplified by the combination of flexure and
transverse shear deformation, or the combination of axial, torsional, and flexural be
havior. In such cases the component modes of structural action each produce a con
tribution to the virtual change in complementary internal work <'>Wint Therefore, for
a bisymmetrical element of length L, in the presence of axial, torsional, and flexural
deformations,
<'>Wint = LL <'>Fx (: ) dx
+ LL c5M ( , Z:) dx + f0L <'>Mx (;) dx + LL c5My (;) dx
For line elements (framework elements, but especially complicated elements of the
kind illustrated in Fig. 7.11) one need only fix one end of the member, say, point 2 for
the case shown. Expressions for the real stresses and virtual stresses at any point
in the member in terms of the free-end real and virtual forces (the forces at point 1)
,71
,/ I
Figure 7.11 Nonprismatic
framework element.
can then be constructed straightforwardly from statics. These relationships are of the
form of Equations 7.40 and 7.41, where {F1l and {8F1l are the real and virtual forces
at point 1.
These features of the virtual forces approach can be illustrated by its application to
two problems: (1) the effect of shearing deformation on the deflection of a beam, and
(2) the analysis of a ring beam.
7.6.2.1 Shearing Deformation of a Beam The flexural element of Figure 7.3 will
be the basis for analyzing the effect of shear. We define a stable, statically determinate
support condition by fixing the left end (point 1 in Figure 7.3) and use the principle of
virtual forces to develop the relevant flexibility matrix. The counterpart stiffness matrix
is then constructed by the procedure of Section 4.4.
In dealing with transverse shear deformation we adopt the simplified notion of an
"equivalent shear area," A,. This area, multiplied by the shearing stress ( Tna) at the
neutral axis gives the total shear force on the cross section. Thus there is assumed to
be a uniform shear stress Tna = Fy21A, on the cross section. The corresponding neutral
axis shear strain is y = Fy2IA,G and in the expression for the component of comple
mentary internal virtual work due to transverse shear, 8Wf01 , , we have d(vol) =A, ax,
u = 'Tna 8u = 8Tna E-1 = l/G, {Fil= Fy2, {8F1l= 8Fy2, LQJ = 1. Hence
*
8W int
_ (L ( ) _!_ ( )
8Fy2 Fy2
A, dx
_ 8Fy2 Fy2L
s
-
Jo A, G A, -
A,G
To this must be added the complementary internal virtual work due to bending, de
[ ]{
noted here by 8Wf01 b The flexibility terms for 8Wf01 b have already been established
and are given by Equation 7.46. Therefore,
}
L3 L ; L
8M 2J !-__z,_ -2 ,g ! !!.
_ _
Fy2
(7.47)
z L : L M,2
-- ,_
2EI, : EI,
I
The desired flexibility matrix [d] is the central matrix of this product. The equilibrium
matrix [] needed for the transformation of this matrix into a stiffness matrix is un
affected by the presence of shearing deformation and is thus the same as that presented
in Example 4.4. Using that matrix and the above matrix [d] in Equation 4.25 yields,
with EI,IAsG =
11,
L L
1 -1
2 2
ri () '.')
M, EI,
(2 )+ 11
L
2
(2 ) - 11
:;,121
= (7.48)
L
L ( )
+ 11
Sym. 1
2 8, 1
(2 + 11
Shearing deformation may be included in general frame analysis by replacing terms

in Equation 4.34 by the corresponding terms of Equation 7.48 and their counterparts
in the x-z plane. For most framework members of practical length the influence of
transverse shear deformation may be disregarded as negligible. It can, however, be
significant in members of small span-to-depth ratio or-a similar and often related
criterion-a region in which the shear-to-moment ratio is high. For these, use of ap
proximate values of the equivalent shear area is usually satisfactory.
In Reference 7.9 it is shown that As for the rectangular section (Figure 7.12a) may
be calculated as either 2bd/3 or Sbd/6, depending upon how resistance to cross-sec
tional warping due to transverse shear is accounted for in the assumed behavioral
model. The difference is small. In the same reference it is suggested that, for the wide
flange section of Figure 7.12b bent about its horizontal axis, the equivalent shear area
may be taken as
(7.49)
Reference 7.10 contains several similar suggestions, including one that satisfactory
approximate results for the wide flange section may be obtained by taking for As the
area of the web.
Application of these recommendations is illustrated in Examples 7.10 and 7.11. In
the first of these it is shown that, for an end-loaded cantilever beam of rectangular
cross section, the effect of shearing deformation is small, even for a relatively deep
beam. The second example illustrates that the ratio of equivalent shear area to full
area may be relatively small in wide flange sections, indicating that shearing defor
mation may require more frequent consideration in members of this type. In each of
the sections considered, the two recommendations for equivalent area yielded ap:
proximately the same results, indicating that either should be satisfactory in design.
I
(a) (b) Figure 7.12 Cross section dimensions.
The beam shown has a rectangular cross section. Illustrate the effect of depth-to-span y, v
ratio on the relative magnitude of flexural and shearing deformation.

r
a b

-- ----
-.t-;-
..--
-. x
- L -- +-1
From Equation 7.47
=
PL3 ( l
3/ )
vb 3EI + GL2A,
For
where,
for - -
d
L
=
1
5'
'Y = 1.039
for -
d
L
= -
1
10'
'Y = 1.010
Calculate A, for W27 X 102 and WlO x 100 sections using Eq. 7.49. Compare with the web
area.
Sections:

dl l.10
27.09 w 10 x 100
W27x 102: A = 30.0 in2

I= 3620 in4
W 10 x 100: A = 29.4 in2

W27x 102 I= 623 in4
W27 x 102:
8(0.515)(3620)
Equation 7.49 A, = = 1237 in.2
[(10.015)(27.09)2 - (25.43)2(10.015 - 0.515)]
Web area A, = (0.515)(25.43) = 13.10 in.2
WlO x 100:
8(0.680)(623)
Equation 7.49 A, = = 6.57 in.2
[(10.34)(11.10)2 - (8.86)2(10.34 - 0.680)]
Web area A, = (0.680)(8.86) = 6.02 in.2

7.6.2.2 The Circular Ring Beam To illustrate the application of the principle of
virtual forces to a more general line element we will formulate a flexibility matrix and
define the relationship needed to construct the corresponding stiffness matrix for an
arc of the circular ring beam of Figure 7.13. Elementary flexure theory will be used;
axial behavior, transverse shear deformation, and curved beam theory will be
disregarded.
A stable, statically determinate support condition is defined by fixing point 2. The
applicable expression for the internal virtual work is Equation 6.34:
where ds R d</J. From statics, the relationship for the internal bending moment,
Mz is
or, in matrix form,
{FxFy11 }
R
Mz = L -sin </J (1 - cos </J) -lj R = LQJ{F1}
Mz 1
The virtual forces are chosen of the same form:
R
{8F8Fyx11 }

8Mz = L -sin </J (1 - cos </J) -lj R = LQJ{8F1}

8Mz1
We can now write
8Wfnt = i.p 8MzCMz/Elz)R d<jJ

8F,,R 8M,, {r { : +
(! = ., -,in"' (1 - cos </J)
Figure 7.13 Ring beam.

which, upon integration, gives L 8F1j[d]{F1} where
1/1 1 . : 3 1
- - - sm 21/11 cos 1/1 - - - - cos 21/1 1 - cos 1/1
2 4 : 4 4
I
-----------------------------------+
---------
: 3 1 :
i 2 1/1 - 2 sin 1/1 + 4 sin 21/1 i sin 1/1 - 1/1 (7.50)
I

I
------------
'
--------------------
--- + ----------
I
Sym
and [d] is the desired flexibility matrix. For convenience we have chosen to write the
joint direct forces as Fx1R and Fy1R rather than Fxi and Fyi The corresponding dis
placement vector is, therefore, Lu1/R v1/R ll,iJ.
The establishment of the element stiffness matrix requires, in addition to the inverse
of [d], the static equilibrium relationship, or
(7.51)
Application of these equations is illustrated in Example 7.12.
EXAMPLE 7.12
For the circular arch shown, I = 150 x 106 mm4, E = 200,000 MPa
1. Determine an element stiffness matrix

2. Analyze the structure
y
7.50 7.51, 90
] [
1. From Equations and fort/!=
[d] = _!!.__
4EI
[-;4
-2
(3?T - 8)
(4 - 27T)
4
(4 - 27T)
27T
[) =
-1
0
-1
From Equation 4.25

10.719 9.843 -3.247 -10.719 -9.843 2.371
10.719 -2.371 -9.843 -10.719 3.247
1.365 3.247 2.371 -0.429
[k] = 107
Sym 10.719 9.843 -2.371
10.719 -3.247
1.365
7.7 Problems 209
2. From symmetry, the only nonzero displacement is vC' Therefore, analyzing one half the
structure:
Displacements: {A1} =
(k!!i-'!P1}
1
25R) :. -33.59 mm
10_719 X 107 (-
Ve =
Reactions: {P,} =
(k,1JIA1l
:. Rxa = 22.96 kN, Rya = 25 kN, M,. = -66.4 kN m
Moment diagram:
66.4kNm
7.7 PROBLEMS
2 3 4
7.1 Using the quartic polynomial, v = a1 + a2x + a3x + a4x + a5x , construct the
shape function for the illustrated flexural element.
L L
- -----2___,
1+--2
Problem 7.1
7.2 Formulate the shape function for the four-jointed axial member using u = a, +
2 3
a2x + a3x + a4x .
+ -- x.u +
L- .
L--+++- I - L --+++-1- _j L
Problem 7.2
5
7.3 Using the quintic polynomial v = a1 + a2x a(;X , construct the shape function
for a beam segment of length L. The node point degrees of freedom are
/ d2v
I
du
V; (i 1, 2)
dx2 ;
=
dx;
7.4 Complete Example 7.2 by establishing the explicit form of the shape functions
for v in terms of V i. v2, 8,i. 8,2.
7.5 Formulate the shape functions for the flexural element of Figure 7.3 on the basis
of the following:
7.6 Formulate the stiffness matrix for the element of Problem 7.1.
7.7 Formulate the stiffness matrix for a four-noded torsional element, using for the
rotational displacement (}x the following shape function.
where
(L - x)( 2L - x)(3L - x) x(L - x)(x - 3L)

N1 N
6L3 3
=
2L3
x( 2L - x)(3L - x) x(L - x)( 2L - x)
Nz = N4 =
2L3 6L3
Problem 7.7
7.8 Formulate the stiffness matrix for a uniform flexural element, based on the fol
lowing displacement field.
v = [ ( ;2 - ;2) ( )}
1 -
sin +
si n
3
2 -
3
2 1
3
+ [ ( ;2 - ;2) ( ) ]
sin - sin
3
2 - 2 V2
3
+ [ ( ;2 ;2) ( ) ]0,1
x +
L
2 7T sin - +
L
2 7T sin
3
2 - 2
3 7T 3

3
[ ( ) ( ;2 - ;2) ( ) (
L
2 7T 1 + 2 sin +
L
2 7T 1
- sin
2 - 2 )] 8,2
7.7 Problems 211
7.9 The governing differential equation for beam bending, representing the condition
of vertical force equilibrium, is (d2/dx2)EI,(d2vldx2) = 0. Test the shape function and
moment of inertia variation used in Example 7.3 for satisfaction of this condition.
7.10 Establish the exact displacement expression and shape functions for the given
tapered axial member. Formulate the relevant stiffness matrix.
i.---L
--
Problem 7.10
7.11 Formulate the stiffness matrix for the tapered axial member of Problem 7.10
using the shape function of Equation 7.6.
7.U Form the stiffness equations for Fx and Fx3 for the illustrated three-noded axial
2
member, using for the total displacement the expression:
u=
( 1--+-
L
x 2 2)
3 ( Lx Lx2 ) u2 + (2x2L --Lx ) u3
L u +4 ---
x
I
-
2 2 2
and for the virtual displacement (8u) the following expression.
1
8u= V2
[ . 7TX
sm + sm
. 37TX ] 8u2 + 1 [ . 7TX . 37TX
sm - - sm --
]
2L
-
2L 2 2L 2L
Discuss the implications of the result.
-x,u
---!--- j
Problem 7.12
L/2 L/2 3
7.13 Compare the torsional response of a WlO X 45 and a W21 x 44 section for
warping free and warping fixed end conditions. E = 29,000 ksi and v = 0.3.
12'
Problem 7.13
----T
---+------ 12' __ ___...
7.14 R epeat Problem 7.13 for warping fixed conditions at one end and warping free
at the otber.
7.15 An axial member is subjected to distributed loads as illustrated. The shape func
tions for Case a of the figures are the linear expressions (1 - x/L) and x/L. For Case
b they are given by u in Problem 7.12. Discuss the relationship of the results obtained
with those established by simple prorating of forces to the joints.
q(x) = q1 1 ( +
2
: + ( :))
2
and q(x) = q2 ( sm
. 27TX
L
+ sm
. 47TX
L
)
q(x) q(x)
-- -- ......
. ......... ......
... ............
1--- 1+--- +3
1. X,U
L-----<
l=ci
(a) (b)
Problem 7.15
7.16 Formulate, by use of the work-equivalent load approach, the effective nodal
forces for the beam flexure in selected cases of Table 5.1.
7.17 Formulate, by use of the work-equivalent load approach, the effective nodal
forces for the beam flexure case for the following loads in the interval 1-2.
. 2 n7TX
(b) q = qn sm
L
------L------.<
Problem 7.17
7.18 A space-frame element is subjected to the illustrated distributed loads. Calculate

the relevant effective nodal forces.
lrX
q = 2(1 + ) + 1.5 sin
/ 10 10
lOm
Problem 7.18
7.19 The axial element of Example 7.1 is subjected to a temperature change given
by T (1 - x/2L)(T1) + x/2L(T3), where (T1) and (T3) are the temperature changes
=
at points 1 and 3. Calculate the relevant initial force vector.

7.7 Problems 213
7.20 Calculate, for the tapered flexural element of Example 7.2, the vector of initial
forces for an initial transverse displacement of the form:
. . 1TX 31TX
v' = v'A sin + v'B sm
.
- --
L L
where v and v k are measured amplitudes of initial displacement.
7.21 The temperature distribution across the depth of practical flexural elements is
usually nonlinear. Also, such members are not symmetric about the z axis. Using the
symbolism shown, formulate the expression for initial curvature, K\ to be used in the
calculation of initial forces by means of Equation 7.38.
y(G.. symmetry)
Geometry x-Direction temperature profile

Problem 7.21
v
7.22 Develop a flexibility matrix for the beam shown.
r El1 El2 El1
bt
lf - x,u ,2
"=l
L
3
L
Problem 7.22
7.23 Formulate the flexibility 'matrix for the ring element of Section 7.6.2.2, account
ing for both simple flexure and axial deformation. Designate the cross-sectional area
of the member as A.
7.24 Construct the final algebraic form of the stiffness relationships for the ring beam
element of Example 7.12.
7.25 Formulate the flexibility matrix for the element shown below. Consider only
simple flexural behavior. The moment of inertia I and area A are constant.
Problem 7.25
7.26 Formulate the flexibility matrix for the element shown below. Consider only
flexural and torsional behavior. All section properties are constant.
My1
Problem 7.26
7.27 The wide flange beam shown is loaded by a concentrated force at the center.
Calculate the length ratio, Llh, for which the central deflection caused by shear be
comes 20% of that due to bending.
L/2 ---;....---- L/2

l
----I
Problem 7.27
7.28 Repeat Problem 7.27 for a uniformly distributed load, q, instead of the concen
trated load.
7.29 Calculate the displacement under the load at point A of the continuous circular
arch shown. Consider only flexural behavior.
180kN
A
I= 200 x 106 mm4
E 200,000 MPa
=
---25 m ---.,.._----25 m ----"1
Problem 7.29
REFERENCES
7.1 R.H. Gallagher, Finite Element Analysis Fundamentals Prentice-Hall, Englewood

Cliffs, N.J., 1975.
7.2 V. z. Vlasov, Thin Walled Elastic Beams, 2nd Edition, Israel Program for Sci
entific Translations, Jerusalem, 1961.
7.3 S. P. Timoshenko and J. M. Gere, Theory of Elastic Stability, 2nd Edition, Mc
Graw-Hill, New York, 1961.
7.4 P. A. Seaburg and C. J. Carter, Torsional Analysis of Structural Steel Members,
Steel Design Guide Series 9, American Institute of Steel Construction, Chicago,
1996.
7.5 R. S. Barsoum and R. H. Gallagher, "Finite Element Analysis of Torsional and
Torsional-Flexural Stability Problems," Intlll for Num Meth in Engr, Vol. 2, 335-
352, 1970.
References 215
7.6 Y.-B. Yang and W. McGuire, "A Procedure for Analyzing Space Frames with
Partial Warping Restraint," Intl fl for Num Meth in Engr, Vol. 20, 1377-1398,
1984.
7.7 D. A. Nethercot and K. C. Rockey, "A Unified Approach to the Elastic Lateral
Buckling of Beams" Engr fl, Amer. Inst of Steel Const., Vol. 9, No. 3, 1972.
7.8 R. S. Barsoum, "A Finite Element Formulation for the General Stability Analysis
of Thin Walled Members," Ph.D. Thesis, Cornell University, Ithaca, N.Y., 1970.
7.9 S. Timoshenko, Strength of Materials, Part I, 3rd Edition, Van Nostrand, Prince
ton, N.J., 1955.
7.10 R. J. Roark and W. C. Young, Formulas for Stress and Strain, 5th Edition,
McGraw-Hill, New York, 1975.
Chapter 8
Nonlinear Analysis of Frames

An Introduction
In this chapter we outline some of the ways in which the direct stiffness method of
matrix analysis may be used in the nonlinear analysis of framed structures. The chapter
scarts with a brief discussion of nonlinear behavior, the aims of nonlinear analysis, and
its place in structural engineering. Sources of nonlinearity that are of common concern
in frame design are listed, and the levels of analysis to be considered are defined. To
provide perspective and a basis for comparison with the matrix methods, established
classical solutions of some elementary problems are illustrated. Following this a sym
bolic matrix stiffness method equation for each of the levels of analysis is presented.
Strategies for the numerical solution of nonlinear equations are outlined, but details
of equation solution are left to Chapter 12. The chapter ends with a brief discussion
of problems associated with formulating the equations of equilibrium on the deformed
structure, but a more thorough exploration of equation formulation is deferred to
Chapters 9 and 10.
8.1 NONLINEAR BEHAVIOR, ANALYSIS, AND DESIGN
Most civil engineering structures behave in a linear elastic fashion under service loads.
Exceptions are slender structures such as some suspension systems, arches and tall
buildings, and structures subject to early localized yielding or cracking. But prior to
reaching their limit of resistance, almost all structures would exhibit significant nonlin
ear response. Therefore, if linear elastic analysis is the highest level available, the
design engineer must find another way to account for the effects that the analysis is
incapable of simulating. The answer may lie in any of the following: a) individual
judgment; b) code formulas that accept the results of a linear elastic or simpler analysis
and make allowance for nonlinearity in some empirical or semi-empirical way; or
c) supplementary theoretical or experimental studies.
In nonlinear analysis an attempt is made to improve the analytical simulation of the
behavior of a structure in some respect. The fundamental aim is to improve the quality
of design by providing the engineer with a more reliable prediction of the performance
of a system that is under design or investigation. In making this closer link between
structural analysis and actual behavior, the traditional distinction between the terms
"analysis"-the determination of forces and displacements under given loads-and
"design"-thc proportioning of members and connections to resist the determined
effects-becomes blurred (see Ref. 8.11). We will emphasize the analytical side of the
216
problem but the determination of some aspect of behavior is the primary objective in
the structures we consider.
Reflecting on the contents of the previous chapters we can see that, in linear elastic
analysis, there is art in the reduction of the actual frame to a line diagram and in the
assignment of boundary conditions and member properties. But when that has been
done, the result is an analytical problem that has a unique solution, the one that satisfies
the requirements of co patibility and equilibrium on the undeformed structure. That
is all that is meant by tlie term exact solution in reference to a linear analysis. In the
two preceding chapters we have seen that there may be mathematical difficulties in
the formulation of the linear equations and that we may have to settle for an approx
imation of that solution. Nevertheless, although there are many different approaches
to linear elastic analysis, and although they may differ in mode and amount of com
putational effort, all methods that are capable of producing "an exact solution" rest
on the same few principles of mechanics. But the premise of linear elastic behavior
forecloses the possibility of revealing any manifestation of nonlinearity: A problem
has been solved, but the solution may not tell us all we would like to know regarding
the structure. Indeed, crucial information may be missing.
In using nonlinear analysis the uncertainty regarding actual behavior may be re
duced. In the process, however, the element of art in modeling the structure and in
handling the equations of analysis is increased. In modeling, the analyst must decide
what sources of nonlinearity are apt to be significant and how to represent them. In
dealing with the resulting nonlinear equations, decisions must be made regarding how
to reduce them to a system suitable for practical computation and, then, the method
for solving the reduced system. In this and the following two chapters, procedures for
the treatment of common situations will be presented. But the conditions encountered
in practice are so varied, and the devices for dealing with them so many, that the
coverage cannot be comprehensive.
8.1.1 Sources of Nonlinearity
In linear elastic analysis the material is assumed to be unyielding and its properties
invariable, and the equations of equilibrium are formulated on the geometry of the
unloaded structure or, in the case of self-strained structures, on an initial reference
configuration (see Section 5.3). Subsequent deformations are assumed to be so small
as to be insignificant in their effect on the equilibrium and mode of response of the
system. One consequence of this was our ability to treat axial force, bending moments,
and torques as uncoupled actions in developing the stiffness equations for elements of
bisymmetrical section in Chapters 4 and 7.
Nonlinear analysis offers several options for addressing problems resulting from the
above assumptions. We may consider only the geometric nonlinearity. That is, we may
continue to treat the structural material as elastic but include the effects of deforma
tions and finite displacements in formulating the equations of equilibrium. It is also
possible to consider only the material nonlinearity, that is, the effect of changes in
member material properties under load. And, as a third general option, we may include
effects of both geometric and material nonlinearity in the analysis. In each case, the
possibility of coupling of internal actions must be considered; it may be a dominant
feature of the analysis. Among the many sources of each class of nonlinearity are the
following.
Geometrical effects:
1. Initial imperfections such as member camber and out-of-plumb erection of' a

frame.
218 Chapter 8 Nonlinear Analysis of Frames-An Introduction
2. The P-!;,,. effect, a destabilizing moment equal to a gravity load times the horizon
tal displacement it undergoes as a result of the latera displacement of the sup-
)
. '
portmg structure.
3. The P-fJ effect, the influence of axial force on the flexural stiffness of an individual
member.1
Material effects:
1. Plastic deformation of steel structures.

2. Cracking or creep of reinforced concrete structures.
3. Inelastic interaction of axial force, bending, shear, and torsion.
Combined effects:
1. Plastic deformation plus P-!;,,. and/or P-fJ effects.

2. Connection deformations.
3. Panel zone deformations.
4. Contributions of infilling and secondary systems to strength and stiffness.
8.1.2 Levels of Analysis
Rarely, if ever, is it possible to model all sources of nonlinearity and portray the actual
behavior of a practical structure in all of its detail. Normally, the problem is one of
selecting a method that falls short of the ideal in one way or another but that does
provide adequate analytical simulation of the case at hand. The most common levels
of analysis are represented in Figure 8.1 by schematic response curves for a statically
loaded frame. The degree to which they can model true behavior differs, but each can
yield information of value to the engineer.
By definition, first-order (linear) elastic analysis excludes nonlinearity, but it gener
ally represents conditions at service loads very well.
The elastic critical load is the load at which both the original and an alternative
loading path become mathematically valid2 and it can be shown that the path taken
from that point will be the alternative one. The load is commonly determined from an
eigenvalue analysis of an idealized elastic model of the structure3. The analysis also
yields the eigenvector, that is, the shape that the system assumes in the post-critical
state, but it doesn't determine its amplitude. The inelastic critical load is similarly de
fined and calculated, but it is one in which the possibility of precritical inelastic material
behavior is considered in the analysis. A critical load analysis may not include the
nonlinear phenomena that must be taken into account to determine accurately the
magnitude of the load that would cause failure, but frequently, it gives adequate-and
valuable-pictures of possible types of failure.
In second-order elastic analysis the effects of finite deformations and displacements
'The P-t:. and P-5 effects described here may be encountered and treated as distinct-phenomena so we define
them as such. But they often occur in combination, justifying a blanket "P-delta" definition such as "the
secondary effect of column axial loads and lateral deflection on the moments in members" (from the Glos
sary of Reference 8.1 ).
2This state is generally referred to as bifurcation of the loading path.
3In this and other comparable methods of analysis it may be found that a number d values of the applied
load satisfy the bifurcation condition (e.g., see Example 8.3). Each may be called a "critical load." Here,
we are reserving the term for the smallest physically meaningful value.
4Following what has become common in civil engineering structural analysis, we use the term second-order
not in a precise mathematical sense but merely as shorthand to designate any method in which the effects
referred to are accounted for.
H First-order
elastic analysis
Bifurcation
_J::gs.!!cgi!_icl !Q'!Q
Hee
____
E sc i }'_liit
Hes :7"':::,...:-=--...-..:
::; ::;:
:r:
-0
"'
..Q H;e
<ii
....
<U Plastic limit load
HP
-l
Inelastic stability limit ::--;:1 Ll r-
H;s
ioolk Y'----------------::R
: ---+ " "
Lateral deflection, Ll
Figure 8.1 Levels of analysis.
of the system are accounted for in formulating the equations of equilibrium. A second
order elastic analysis can produce an excellent representation of destabilizing influ
ences such as the P-tl. effect, but it has no provisions for detecting material nonlinearity.
Several of the possible modes of nonlinear elastic behavior are indicated in Figure 8.1:
(a) bifurcation (branching) of the loading path with the system following an alternative
path in the post-critical state; (b) gradually increasing nonlinearity culminating in elas
tic instability at a limit point; and (c) increasing stiffness either from the onset of
loading or, as shown, following a period of gradual softening.
In first-order inelastic analysis the equations of equilibrium are written in terms of
the geometry of the undeformed structure. Inelastic regions can develop gradually or,
if the plastic hinge concept is adopted, as abrupt changes in the structure's response.
When the destabilizing effects of finite displacements are relatively insignificant,
first-order inelastic analysis can produce an excellent representation of simple elas
tic-plastic behavior and failure through mechanism formation, that is, the simple plastic
limit load of Figure 8.1. But it has no provisions for detecting geometric nonlinear
effects and, of paramount concern, their influence on the stability of the system.
In second-order inelastic analysis the equations of equilibrium are written in terms
of the geometry of the deformed system. It has the potential for accommodating all
of the geometric, elastic, and material factors that influence the response of a structure.
Thus, in principle-and in a deterministic sense-it enables the preparation of ana
lytical models capable of faithfully simulating actual behavior and calculating the in
elastic stability limit, that is, the point at which a system's capacity for resistance to
additional load is exhausted and continued deformation results in a decrease in load
resisting capacity.
The path of increasingly nonlinear response (elastic or inelastic) culminating in in
stability is probably the most common mode of failure in civil engineering structures.
For this reason, methods for detecting limit points will be a major concern of this text.
8.1.3 Examples from Established Theory
Underlying contemporary nonlinear matrix analysis are long established methods of

structural mechanics and applied plasticity. A few of these are applied to elementary
systems in the following examples. The purpose is to illustrate types of physical be
havior and some of the concepts and mathematical techniques used in dealing with
them. The examples will also be used in appraising the matrix methods to be developed
in subsequent chapters.
In some of the examples large changes in geometry are studied for the purpose of
defining or illustrating a particular phenomenon, but it is important to stress that even
at the ultimate, practical civil engineering structures are rarely called upon to resist
more than moderately large displacements and small strains. Both of these terms are
qualitative; they depend upon the individual situation and cannot be defined precisely.
In the interpretation of the examples to follow-and in any nonlinear analysis-it may
be helpful to think of a moderately large displacement as one in the range of a hun
dredth to a tenth of the span or height of the structure, and a small strain as anything
from the infinitesimal to one of the order of ten times the proportional limit of the
material.
In Example 8J. an ideally straight and rigid member supported by an elastic spring
at the top is subjected to two types of loading. In the first, in which the applied hori
zontal force is zero, a critical load is determined by postulating a lateral displacement
and studying the equilibrium of the displaced system to determine the conditions under
which that displacement could occur. The system is stable for Pless than kL, but when
the two become equal, the loading path can bifurcate. Since in the deflected position
the overturning moment, Pt:., will be greater than the restraining moment, kt::.L cos 8,
the member will deflect. Therefore P = kL is the critical load.
EXAMPLES.I
Member ab is a rigid bar and member be has a constant axial stiffness coefficient k. p
Analyze the behavior of the system for:
c b aP
1. a= 0
2. a* 0 ,V'-------<;J
I
L
1. a = 0. Assume the possibility of lateral displacement of b and

consider the equilibrium of ab'.
2.M.: P!;). - k!;).L cos 8 = 0, or !;).(P - kL cos 8) 0

=
For 8 = 0, two equilibrium states exist:

(1) /;). = 0, the undeflected configuration
(2) P Per kL, the critical condition
= =
The critical condition is unstable since, for /;). * 0
P = kL COS 8 < Per (a)

2. For a * 0 there is lateral displacement from the outset p
"LM.: Pil + aPL cos 8 - kilL cos 8= 0 or, letting fl= L sin 8 k b b' aP
""----+-
r
kL cos 8 I
P= ---- (b) I
1 + a cot 8 I
L cos 9 ____..I
. . d(PlkL) I
Solving for hm1ts: = 0 when (J
dfJ I
I
8 = tan-1 (c)
From Equation b, the equilibrium paths for several val ues of a are: Plkl
a=O
sin9
In the second case of Example 8. 1, the presence of an active horizontal load results
in lateral deflection from the outset. The destabilizing P-6. effect is nonlinear and it is
seen that stability limits that depend on the relative values of the horizontal and vertical,
loads are obtained.
The system of Example 8.2 is a shallow elastic arch consisting of two axial force
members. Analysis of the geometry and the equilibrium of the displaced structure,
without restricting it to small displacements, leads to the nonlinear elastic response
illustrated by the solid curve. From the unloaded state, the rate of resistance to down
ward load decreases until, at stage 1, an elastic stability limit is reached. Deflection
under decreasing load can then take place until, at stage 2, the structure is flat an<!
unloaded but self-compressed. From this stage until stage 4, further deflection would
require an upward load-with a maximum at stage 3-to equilibrate the decreasing
compression in the members. At stage 4 the structure would again be unloaded and
unstressed. It could then deflect further under increasing downward load. Point 1 is
an obvious elastic stability limit, and the instability of response from points 1 to 3 could
be demonstrated by further analysis (Example 8.8). But even without such an analysis,
it is clear that attainment of equilibrium beyond point 1 would require either an ex
ternal agent to control the unloading or the displacement subsequent to that stage or
a constant load and a system capable of undergoing the displacements required to
"snap through" to stage 5, at which point it would function as a stable suspension
system under further loading.
EXAMPLE8.2
The system is a shallow three-hinged arch. Members ab and be have a constant
axial stiffness coefficient k. Analyze the behavior of the system.
Consider the equilibrium of the displaced structure:

p
b
From the geometry of the displaced structure
t:.b = L[sin 8 sec(a - 8)] (a)
t:.Lab = L[l - cos a sec(a - 8)] (b)
From the equilibrium of joint b'
P = 2Fab sin(a - 8)
Letting Fab = k ilLab
P = 2kL[sin(a - 8) - cos a tan(a - 8)] (c)
Solving for limits, dP/d(a - 8) = 0 when
cos3(a - 8) = cos a (d)
The equilibrium path for given values of a, L, and k is, P

from Equations a and c: Limit point
---+ ____
Sa_E t_ru
h h _ --+ __
Displaced shape at key points: b

.....
r
,.,...
.-- .....
..,.,.....,.... J.:1im ' -..... ...... J
_
_
__ 1 ............ Lsina
!
/
a ::=:: ::_ :_ :_ ____________ ______ 2 ::::;

--.....,
3 ---.,,,.,,...-
4
-
..... - -
...-
__
.,,..,.. Lsina
....... -. Yr
__
-. 1 P1im ..- .,.... ,,,.,...

..... ...- 5
In Example 8.3 stability is provided by the flexural resistance of the member itself,
and we limit consideration to the classical small displacement theory. The technique
used to solve the first case is the same as that used in the first case of Example 8.1:
The possibility of a small lateral displacement of an ideally straight, axially loaded
member is assumed. Then, by studying the requirements of equilibrium for the straight
and flexed states, the mathematical condition under which they can coexist is deter
mined and, either by further mathematical analysis of the displaced condition or by
logical deduction, a conclusion is reached regarding the relative stability of the two
states. This technique was developed by Euler 250 years ago. Equations a-d summarize
his theory and Equation c, which represents the governing state-neutral equilibrium
in the deflected po1>ition-is usually called the Euler load in his honor.
Except for Euler's use of a physically undefined elastic constant, C, rather than the
El, Equation c comes directly from Reference 8.4. The same theory
flexural rigidity,
Peri a
has been used as the basis for the determination of an inelastic critical load,
value generally obtained by employing the tangent modulus, E,, rather than E in the
Euler equation. Physically, this is a means of accounting for pre-critical inelastic be
havior of an ideal member ( see Ref. 8.5 and Section 10.3). Computationally, an iter
ative solution is generally required since E, is a function of the axial stress in the
member.
Member ab is an ideally straight elastic bar. Summarize the small displacement theory of its flexural
M ,
r-
behavior for:
1. M0 = 0
2. M0 = PL
a b
L E,l
L.
1. For a = 0 assume the possibility of lateral displacement. The bending moment
at xis M = Pu and, from small displacement theory,
-
d2u +Pu
= 0 (a )
dx2 EI
In the general solution
u = C1 sin x + C cos
2
x (b)
The boundary conditions x = 0, u = 0 and x = L, u = 0 yield
and C1 sin L = 0
---y
i:PL
The smallest load that will admit deflection, i.e., C1 * 0 is the Euler load
n2EI
Pere= V (c )
From Eq. b, the equation of the deflected curve is
. 7TX
u= C1 Sill (d )
L
2. For a* 0 there is displacement from the outset. M = -Pu - aPL, or
d2u Pu aPL
-+-= --- (e )
dx2 EI EI
with the general solution
u= C1 sin /fI x+ C
2
cos /fI x - aL
Utilizing the boundary conditions and recognizing that, by symmetry, Umax =

ulxuz = 8
(f )
Graphically:
The second case of Example 8.3 also parallels that of Example 8.1 except that,
whereas in Example 8.1 large displacements were accounted for, in this example small
displacement theory is used. As a consequence, the calculated response becomes as
ymptotic to the critical load rather than exhibiting limit point characteristics. As a
practical matter there can be no objection to this since any nonlinear elastic behavior
encountered in realistic civil engineering structures will be represented adequately by
mall displacement theory.
The first two cases of Example 8.4 are identical in principle to the corresponding
parts of Example 8.3. The mathematical difference is in the boundary conditions and
the practical difference is that, in the second case, for members of equal length and
flexural rigidity the relative influence of the axial force on lateral bending is greater in
this example than in the previous one. There is a nonlinear effect in both members,
but as mentioned in Section 8.1.1, it is often called the P-fi effect when lateral trans
lation of the applied load is possible, as in this example, and the P-8 effect when it is
prevented, as in Example 8.35.
5Jn members with translating jointsythe term P-8 is also used to describe the nonlinear deflection of the
member with respect to the chord joining its two ends.
,r
Member ab is an ideally straight elastic bar. Summarize the small displacement theory of its flexural
behavior for:
af
1. a= 0
r-
2. a* 0
3. a* 0 and P directed upward
L ,/
L,
1. For a = 0. Proceeding as in Example 8.3, the same differential equation is obtained. The c, p
boundary .::onditions are: x = 0, v = 0 and x = L, dvldx = 0. Solving and satisfying the
boundary conditions results in
b aP
y
I
I x
(a)
I
I k-----+--
L
L,:
and the equation of the deflected curve
. 1TX
V = C1 SID (b)
ZL
x
2. For a* 0, proceeding as in Example 8.3 leads to the differential equation
d2v Pu aP x
+ = (c)
dx2 EI EI
and the general solution
v = C1 sin x + C2 cos x - ax (d)
a
0 and C1 =

The boundary conditions yield C2 =
-cos -L
EI EI
or, from Eq. (d), the deflection
.:i = l
v xL = a
L
l]
tan
EI L
L
- 1 (e)
3. For a* 0 and P upward, M =Pu - aPx or, proceeding as above
v = C1 sinh /i x + C2 cosh /i x + ax (f)
and
L
[ ]
tanh
EI
=v l x=L = aL 1 - (g )
jiLp
Graphically:
Case 3
aL
The third case of Example 8.4 is similar to the second but for the fact that the vertical
component of load is a stabilizing one. As a consequence of starting from the small
deflection equation of flexure, in this case the calculated response becomes asymptotic
to the displacement that would obtain in a pure (unftexed) axial force member. All
three cases of this example will be referred to in later applications of numerical anal
ysis, but since they are well documented in standard texts (e.g., see Ref. 8.3), only
those steps considered essential to an appreciation of the theory and its implications
have been included here.
Example 8.5 is a study of the behavior of a rectangular section steel beam in which
the material is assumed to have elastic-perfectly plastic characteristics. The nonlinear
effect of gradually spreading plasticity is traced through the use of elementary equa
tions that imply satisfaction of equilibrium on the undeformed structure. The theoret
ical limit of resistance is reached when the bending moment at the left end becomes
the plastic moment, MP, which is equal to the yield stress, uy, times the plastic section
modulus (bd2/4 in this case). At this point the cross section is fully plastified, and a
"plastic hinge" forms. The theoretical limit is approached asymptotically, but in the
numerical illustration resistance to further loading becomes insignificant by the time
the end deflection becomes equal to a tenth of the span.
1. Determine the load-deflection relationship for the beam shown. The material is elas
tic-perfectly plastic. Consider partial plastification.
2. Calculate the result for L = 8 ft., b = 2 in., d = 8.25 in., <Fy = 36 ksi, E = 29,000 ksi
D C
I I x v
I '
fd +Y
_!!__, c b
I
- I
ty

:I
Material 'e
I.
L,
viI
----i 3J c
-----
v
-8- Stress at section D - D
1. By moment-area principles:
A= A
l.l I.lac + A
l.lcb
= J
L,
L
Mx dx
Eleff
+
vL;
3EI
(a)
where
(b)
At Section D-D:
M = Vx =
[d42 y32]bo-y
-
Using Eq. (b)
= 3L.]1/2
y \!32 d [1 - 2:_ (c)
Also,
Jeff
= --
2by3 ( d)
3
From Eqs. (a)-(d)

.:i r-1 (L [ X2 dLxr2 L.
}]
-
_
+ (e)
3E dL. JL, 1 - 32
V3
Integrating and evaluating limits
.:i
=
2L2o-r [18(L./L)2 - 6(L.IL)) - 1 rn(Le)2]
3J ( 32LL,
(f)
_
3E d 1- -
L
2. For the given dimensions V(kips)

S = 22.69 in\ Z = 34.03 in3 12.76
For L, = 96 in:
My = 816.8 in. kips, VY = 8.51 kips
.:i = .:iy = 0.925 in
For L. = 64 in
MP = 1225.3 in. kips, VP = 12.76 kips
.:i 00
Example 8.6 is an exercise in simple plastic analysis, a methoct' in which the devel
opment of zones of plasticity is ignored and cross sections are assumed to be either
fully elastic or fully plastic. In the beam-column studied, mechanism motion can occur
when plastic hinges form at the fixed end and under the transverse load. The inter
action equation used to define full plastification of the cross section (Eq. 10.5 from
Section 10.1.2), is one commonly used for steel wide flange sections subjected to axial
force and major axis bending (see Ref. 8.6). The plastic limit load is obtained by
assuming the hinge locations to be fully plastified and satisfying equations of equilib
rium written for the member in its undeformed position.
p;qlPLE8.
For the member shown, determine the plastic limit load by con
ventional plastic analysis. Use the interaction curve of Equation
10.5.
Member properties: A = 19.1 in.2, I= 533 in.4, Z = 96.8 in.3,
E= 29,000 ksi, <Ty = 50 ksi
(a) The first hinge will form at a
Py = 19.1 x 50 = 955 kips
MP = 96.8 x 50 = 4840 in. kips

c
From Eq. 10.5:
_.!.__
955
+
0.85Ma
4840
= l ------ L = 288" ------+!I
Elastic moment-diagram
:. P = 259.3 kips, Ma = 4148 in. kips,
Mb= 2212 in. kips
(b) The second hinge will form at b, resulting in a

mechanism.
Satisfy the equilibrium requirements:
"i.Mb,: Re= Mb/192
"i.Mb,: (Ma + Mb) = (.3P - Rc)96

For hinges at a and b, Mb= Ma, or
Ma= ll.52P
Thus, from Equation 10.5
P = 325.7 kips, Ma= Mb = 3752 in. kips

(c) Results: P (kips)

y
_E_ + 0.85M1 = l
955
-.... P M
..... y P

.....
.....
.....
.....
M
)_____...,, p
N N 00 0
M (in. kips)
N
("') -q-
N
Interaction diagram
p
0.605
1.204 6.b (in.)
Load-displacement diagram
Before the limit is reached, formation of the first hinge results in redistribution of
internal resistance. This is illustrated in two sketches in the example: (1) a comparison
of elastic and final moment diagrams; and (2) an interaction diagram containing traces
of the loading paths (the force points) of the end and interior cross sections. Also
shown is a response curve, which indicates that since equilibrium equations are written
on the undeformed structure and the member is assumed to behave elastically between
plastic hinges, the response between levels of hinge formation is linear.
Example 8.7 is an analysis of the combined effect of the geometric nonlinear be
havior of an elastic beam-column and the material nonlinearity of a semirigid connec
tion of prescribed moment-rotation characteristics. Equilibrium is formulated on the
deflected structure and classical small deflection elastic theory is used for the beam
column. The nonlinear response equation, Equation c, is seen to have a limit. In the
numerical illustration the system becomes unstable at an applied vertical load equal
to 63% of the elastic critical load of the same member in a fixed base condition.
[EXAMPLES.
Member ab is an elastic bar connected to its base by a nonlinear rotational spring having the moment
rotation characteristic:
Analyze the bar under the action of a fixed force H and a varying force P
From Equation d of Example 8.4, letting = HIP
. G, G,
a
Hx
u= C1 sm - x + C2 cos - x - -
EI EI P
du
For x = 0, u 0 and C2 = 0. For x = L, - = 80, or
dx
=
Thus
x
Hx
p
(a)
Also, M0 = HL + Pt:l., or
f3Wa - HL
t:l.= ---- (b)
( ) [( ]
p
Solving for !!.. from Equations a and b:
!!..=
H
312 tan {P L
...j Ei
_ {P L
...j Ei
) +
/32(1 Yl - 4(HEI tan2VPiFJL)/f32)2
(c)
EI
p {P L
4HEI tan El
EI V
Plot Equation c for: L = 12 ft, I= 88.6 in.4, E = 29,000 ksi, f3= 10 x 104 in. kips/rad, and
H= 10 kips
7?(29,000)88.6
(Note that for a fixed base Pere = 305.7 kips
. )
4(144)2
=
P (kips) limit point
100
0
o 3.9 10 20 30 40 50 (in)
28.l
8.1.4 A Commentary on Stability

A maj or theme of this section has been structural instability, ways in which it may
occur, and analytical methods for predicting-and forestalling-the possibility of its
occurrence. We have used a case study approach in which individual situations have
been considered and states of stable and unstable equilibrium deduced from the phys
ical results of analyses of the case at hand. More mathematical criteria for stability
exist but none is all encompassing, and in engineering applications it is rarely necessary
to resort to them since stable equilibrium is usually evident from the physical facts.
But to establish a relationship between the findings of our individual studies and to
illustrate their underlying coherence, we present the following summary of a basic
concept of stability and a criterion based on the total potential energy of a system.
Works on stability such as References 8.2, 8.7, 8.8, and 8.9 contain more extensive
treatment of this and other mathematical criteria.
Stability of equilibrium may be defined by reference to three states of a ball at rest
on a surface (Fig. 8.2). If the ball is on a concave spherical surface (Fig. 8.2a), any
displacement from the position shown will require a certain amount of work; the po
tential energy of the system will be raised. On this basis it is reasoned that it is in stable
equilibrium inthe defined position. If it is on a horizontal plape 8.2b);i4;..m:flie
(F.
displaced without the expenditure of work; there will he no 1change in the potential
energy of the system. In this case it is said to be in indifferent or neutral equilibrium.
In the third case (Fig. 8.2c), the ball is on a convex spherical surface. Any displacement
will lower it; the potential energy of the system will be decreased. It is in unstable
equilibrium.
This definition of the stability of equilibrium of a rigid body has been extended to
deformable ones by introducing the concept of total potential energy from theoretical
mechanics:
Il=U+V
in which Il is the potential energy of the system, U is its strain energy, and Vis the
potential of the applied loads.
We may relate this expression to virtual work principles by recalling the virtual
displacement statement of equilibrium
(6.13)
in which we have changed the signs of both the internal and external virtual work
terms of the original Equation 6.13 to facilitate comparison. We first recognize that
the internal virtual work of Equation 6.13 is the change in the strain energy of a
structure as it passes from a configuration in a state of equilibrium to one defined by
infinitesimally small virtual displacements and that the external virtual work is the
negative of the corresponding change in potential of the applied loads. Comparing SW
and Il it follows that, at equilibrium, the total potential energy assumes a stationary
value, that is, in variational terms, sn = 0.
Although this is a mathematical condition for static equilibrium, it is not sufficient
to determine whether the equilibrium state is stable, neutral, or unstable in the sense
defined above. But reasoning in the same way, it is seen that this depends upon
whether, at equilibrium, the total potential energy is, respectively, a minimum, neutral,
or a maximum and, further, that the distinction can be made by investigating the second
or higher variations of Il. The process will be illustrated by application to two of the
earlier examples of this chapter. Its extension and use in the analysis of the stability
of multiple-degree-of-freedom systems may be found in the references cited.
OQQ (a) (b) (c) Figure 8.2 Equilibrium states.

In Example 8.8 the physical conclusions regarding the stability of the system of
Example 8.2 are reexamined using the total potential energy criterion. The members
are constant stiffness springs whose strain energy is a function of their shortening and
the stiffness coefficient. The datum for measurement of the potential energy of the
load is the unloaded configuration. Thus the total potential energy can be expressed
as a function of the single variable (} and variations of the potential as derivatives with
respect to (}. When equated to zero, the first derivative confirms the equilibrium path
determined in Example 8.2. The second derivative produces the condition for stability
of equilibrium. For any point on the equilibrium path at which the second derivative
is positive the total potential energy is a minimum, therefore the equilibrium is stable.
When the second derivative is zero, equilibrium is neutral, which in this case identifies
the limit points 1 and 3 of Example 8.2. Similarly, negative values of the second de
rivative between points 1 and 3 signify maxima of the total energy in that region, that
is, an unstable state.
For the structure of Example 8.2 use total potential energy criteria to:
1. Determine the equilibrium equation

2. Analyze the stability of the equilibrium path
1. Let 0 of Example 8.2 be the basic displacement variable. Using Eq. 4.4a for U and letting
V = 0 in the unloaded state
II = u + v = G
2 k 6.Lb ) - Piib
= kL2[1 - cos a sec(a - 0)]2 - PL[sin 0 sec(a - 8)] (a)
arr =
dII
= 2kL2
{ [cos(a - 0) - cos a][cos a tan(a - 8)] } _ PL cos a
(b)
dO cos2(a - 0) cos2(a - 8)
Letting arr = 0 and solving for P
P = 2kL[sin(a - 0) - cos a tan(a - O)] (c)
as in Example 8.2
2. The second variation of II is, from Eq. (b)
d2IT cos a . .
{2kL2[cos a + 2 cos a sm2(a - 0) - cos3(a - O)
d,.i
rr cos4(a - 8)
(d)
- 2 cos(a - O) sin2(a - 8)] + 2PL[cos(a - 0) sin(a - 8)]}
Substituting Eq. (c) for P, 2 on the equilibrium path is
d2II 2kL2[cos a - cos3(a - O)]

(e)
dO cos4(a - 0)
Thus stability is governed by the sign of the numerator within the physical range la - Bl < TTl2:
d2II
> 0 for cos3(a - 8) < cos a the system is sable
d02
d2II
= 0 for cos3(a - 0) = cos a equilibrium is neutral
d02
d2II
< 0 for cos3(a - 0) > cos a the system is unstable
d02
which confirms the physical evidence of Example 8.2.
I
A similar reexamination of the first part of Example 8.4 is conducted in Example

8.9. We draw on an energy approach to the determination of the critical load in Ref
erence 8.7 for the definition of U and V. Again, the total potential energy can be
expressed as a function of a single variable, in this case the tip deflection, Li. The
bifurcation condition of Example 8.4 is confirmed, as is the neutrality of equilibrium
in the deflected state. In both examples, however, there remains the question of the
physical significance of "neutral equilibrium." In Example 8.4 we addressed it by an
alyzing the member under the presence of a small lateral force and demonstrating the
destabilizing effect of a compressive load. In Example 8.9 our reliance upon small
displacement theory limits the application of the total potential energy criterion. Closer
examination of the static equilibrium of an ideal elastic bar in the deflected position
is needed, and this requires consideration of larger displacements in an approach com
monly known as the theory of the elastica ( Ref. 8.7). By that theory it may be shown
that the bifurcation load of an ideal elastic bar is the same as in the small displacement
theory and that the deflected configuration is stable-but only weakly so. A very small
increase in the load would cause a large deflection. In a structure of realistic dimen
sions, either the elastic displacement would become untenable or the P-Li effect would
result in destabilizing inelastic behavior. In either event, the critical load can be viewed
as a practical limit on the stability of a framework element.
Use total potential energy criteria to analyze the stability of equilibrium of the column of Ex
ample 8.4
For present purposes use the coordinate system shown. From Example 8.4, the buckled config- x
uration is
The bending moment at x is
'TTX L
M = P(li - v) = Pli cos
L1
2L
In Reference 8.7 it is shown that
U =
f,L M2
--
dx
=
p2 Ji2L
--
----y
o 21 41
and that, for small displacements
Thus, letting V = 0 prior to buckling
Therefore:
(a)
dTI
2
P Lli p7T Ji 2
----- (b)
dli 2/ 8L
d2IT p2L
--- P7T2
(c)
dli2 2/ 8L
dll . . 7T2/
Lettmg. = O m EquatJon b gives, fort:. * 0, Pere
.
4 2 , which confirms the Euler solution.
dt:.
=
Substituting Pere in Equation c gives, fort:. * 0, : = 0, which indicates a state of neutral equilibrium at the critical load.
8.2 A MATRIX APPROACH
In the next two capters a number of the concepts and techniques of nonlinear analysis
demonstrated in the above examples will be develope.d and applied in the context of
the matrix stiffness method. For coherence and versatility, the approach taken is keyed
to the levels of analysis illustrated in Figure 8.1. Each level can be defined by a symbolic
matrix equation, as follows.
Linear elastic analysis will continue to be required. To emphasize its nature, we shall
designate the linear elastic stiffness matrix as [Ke] or, in global analysis terms
(8.1)
If we assume that, in the higher-level methods, the underlying nonlinear equations
of equilibrium have been reduced to a set for which we can adapt techniques used for
solving simultaneous linear algebraic equations, then behavior can be traced incre
mentally and each method can be stated symbolically as a variant of the global stiffness
equation
[K1]{da} = {dP} (8.2)

In Equation 8.2 [K,] is a tangent stiffness matrix, {da} is a vector of incremental nodal
point displacements, and {dP} is a vector of incremental nodal point loads and reac
tions. Typically, [K,] has a linear elastic component and one or more additional com
ponents that are functions of the loads and/or the displacements in existence at the
start of the increment. Thus the computational problem can be addressed as the con
ventional one of solving for unknown displacements and back substitution of the re
sults in element stiffness equations to determine element forces, but doing so in a
stepwise fashion in which the total response is determined through summation of
increments.
The various levels differ in the types of nonlinearity included in [K,], the way in
which the equilibrium equations are formulated, the details of equation solution, and
the way in which members may have to be subdivided into discrete elements to achieve
satisfactory results.
In second-order elastic analysis the effects of finite deformations and displacements
are accounted for in formulating the equations of equilibrium and Equation 8.2
becomes
(8.3)
in which [Kg], geometric stiffness matrix, represents the change in stiffness that
the
results from these effects. The matrix [Kg] can be developed in a number of ways, and
various procedures can be used to solve Equation 8.3. Generally, members must be
subdivided into several elements to produce satisfactory results.
In first-order inelastic analysis the equations of equilibrium are written in terms of
the geometry of the undeformed structure, and Equation 8.2 becomes
(8.4)
where (Km], which we shall call the plastic i n matrix,
re d u ct o represents the change in
stiffness that results from inelastic behavior of the system.
8.3 The Equations 9f Analysis and Their Solution 235
In second-order inelastic analysis both geometric and material nonlinearity are ac

counted for. The equations of equilibrium are written in terms of the geometry of the
deformed system and Equation 8.2 becomes
(8.5)
For the calculation of elastic critical loads the global stiffness equation is cast in the
form of a generalized eigenvalue problem in which the equation of equilibrium at the
critical state is
(8.6)
where [Kg] is the geometric stiffness matrix computed for a reference load {Prer)6,
A eigenvalue) is a load factor with respect to {Pref), and {A) (an eigenvector) is the
(an
buckled shape. The lowest value of A that satisfies the equation for {A) 1=- 0 yields the
elastic critical load vector A{Pref), and the corresponding {A) defines the buckled con
figuration. By modifying the material constants in [Ke], Equation 8.6 can be used to
calculate inelastic critical loads in agreement with an accepted concept such as the
tangent modulus theory. Methods commonly used for the calculation of both eigen
values and eigenvectors are based on the following reduced and manipulated form of
Equation 8.6.
(8.7)
in which the subscript f symbolizes that the respective matrices and vectors relate to
the free degrees of freedom only, and it is recognized that [K.1] must thus be non
singular. In general, however, the product [K.1]-1[Kg1] is not symmetric. Therefore, to
cast Equation 8.7 in a form amenable to convenient solution it must be transformed
in a way that avoids the complications of dealing with unsymmetric matrices. Processes
by which this may be accomplished and the resulting equations solved are presented
in Section12.8. In the examples of the following chapters we will continue to refer to
Equation 8.7 as the source of calculations of critical loads and buckled configurations.
But except for simple cases in which the answers are self-evident, Equation 8.7 is being
referred to in a symbolic sense; the actual solution requires some intermediate steps
and an iterative (or comparable) procedure.
8.3 THE EQUATIONS OF ANALYSIS AND THEIR SOLUTION

In developing methods of linear elastic analysis in earlier chapters we dealt with as
semblages of one-dimensional framing members for which exact or approximate so
lutions of the ordinary differential equations for each member are well known. The
solutions were cast in the form of linear algebraic relationships between forces and
displacements at the end of each element. As we have seen in the illustrative examples,
in nonlinear analysis the response of the elements is a continuously or intermittently
changing function of the applied loads. Unfortunately, it is often impossible to solve
the underlying equations in any direct analytical way. And even if such solutions did
exist, we would be confronted with the problem of solving a set of algebraically non-
6 As defined here, the internal forces upon which [K1] is based are calculated by a linear elastic analysis for
the reference load (P ,.1} and it is assumed that this distribution obtains at the critical load, that is, any
internal redistribution of forces between the reference and critical load is disregarded. For this reason, this
approach is often called linear stability analysis even though it is one device for the analysis of a physically
nonlinear phenomenon.
linear or transcendental equations, another forbidding task. Normally, therefore, it is

necessary to deal with them in some piecewise linear way, as in Equations 8.2-8.5.
The strategy we shall use is indicated by the way in which those equations are for
mulated. Initially, it requires obtaining symbolic analytical solutions of the differential
equations of elements and reducing the degree of these inherently nonlinear solutions
by eliminating higher-order terms deemed to be insignificant. Then, as in earlier work,
assumed shape functions are used to cast the reduced solutions in the form of algebraic
relationships between the forces and displacements at the element ends. In this case,
however, the force-displacement relationships may, in themselves, be functions of the
end forces. It also involves adaptation of features of plasticity theory such as the plastic
hinge concept and the normality criterion for plastic flow. The immediate objective is
the development of a load-dependent element tangent stiffness matrix that can be
evaluated for given element forces and is a reliable predictor of a type of behavior in
the neighborhood of the loads to which those forces correspond.7 Conventional stiff
ness method procedures are then used to assemble a linear global stiffness equation.
The process to this point encompasses the first three parts of the structural analysis
procedure defined at the start of Chapter 1: basic mechanics, finite element mechanics,
and equation formulation. The fourth part, equation solution, can use a variety of
methods. Following is a summary of a few of them, with an emphasis on those we shall
use later.
8.3.l Equation Solution-The Options
The available choices for solving Equations 8.3 to 8.5, and some of the particular
problems associated with each choice are illustrated in Figure 8.3.
The most elementary approach is an Euler or simple-step one (Fig. 8.3b ), in which
the tangent stiffness matrix is formulated for a given set of forces, increments of either
load or displacement are specified, and the equation is solved for the corresponding
unknowns. These are then used in the element equations to determine new values of
internal forces, the tangent stiffness matrix is reformulated using these forces and the
deformed geometry, and the process is repeated. The total loads, forces, and displace
ments at the end of each increment are determined by summing these entities over all
increments. Equilibrium between external loads and internal forces is not checked,
therefore the calculated response curve can drift from the true one. However, if the
incremental steps are adequately small and for structures of moderate nonlinearity
which is generally the case for civil engineering systems-good accuracy may be ob
tained with a modest amount of computational effort. Single step methods are de
scribed further in Section 12.2.
A higher level of computational sophistication is shown schematically in Figure 8.3c.
It requires use of one of the multitude of multistep or iterative schemes, of which some
of the more prominent are described in Section 12.3. The basic objective of all of these
schemes is the establishment of equilibrium at the end of the load increment. They do
so by analyzing the imbalance between the applied loads at the end of a linear step
and the internal (element-end) forces calculated from the results of that step. Each
scheme is intended to correct, in an iterative fashion, for the imbalance between the
71mplicit in this discussion and throughout Chapters 9 and JO is the use of an "updated" analysis in which
at each step all quantities are referred to a loaded and deformed reference state. In incremental analysis an
alternative approach is to refer all quantities to the initial configuration, which may be self-strained. In that
case it is necessary to include an initial strain matrix in the tangent stiffness matrix (see Sections A.I, A.2,
and Reference 8.10).
8.3 The Equations of Analysis and Their Solution 237
H
( Calculated
(a) (b) Simple step analysis
H H Calculated H
Calculated Initial load path

,,.. -----
/
(c) Incremental-iterative (d) Limit point (e) Branching point

analysis determination
Figure 8.3 Equation solution methods and problems.
linear approximation and the actual nonlinear response or, more commonly, to reduce
it to a tolerab!e level.
Common to all of these approaches are
1. The problems of numerical convergence just mentioned.

2. Detection of limit points and tracing post-limit point behavior (Figure 8.3d).
3. Detection of states of bifurcation, that is, points at which the response may branch
off from its initial stable mode to a different one-either stable or unstable
with little or no numerical forewarning (Figure 8.3e).
4. Detection of changes in the constitutive relationships, such as the onset of ma
terial yielding.
The first three will be treated in Chapter 12 and the fourth will be dealt with in Chap::
ter 10.
8.3.2 A Fundamental Problem
In nonlinear analysis the aim is to trace the history of all material points in a structure
as it undergoes progressive loading. We have argued that, generally, the only practical
way to do this is by a series of linear analyses that employs one or more of the equation
solution schemes referred to above. Bui inherent to this approach is the fundamental
problem illustrated by Figure 8.4, in which we examine the response of a typical mem
ber of the frame of Figure 8.4a, the rafter ab. Its initial, unloaded, undeformed state
(Configuration 0) can be defined either in terms of the fixed global coordinate system,
(b)
Oy'
Configuration 0
(c)
Figure 8.4 Element deformations.
x, y, or the local coordinate system 0 x , 0y', in which 0x' is directed through the member
'
ends a0, b0 ( Fig. 8.4c). Let us accept for the moment that, following the procedure of
defining properties in local coordinates, transforming and assembling them for global
analysis, and using one of the forms of incremental solution, we have been successful
in tracing the exact behavior of the structure to Configuration t. The frame's correct
load-displacement relationship is as indicated in Figure 8.4b, and now the properties
of the rafter can be expressed either in terms of the global coordinate system or the
' ' ' '
new, updated local coordinate system x , 'y', with x directed through the ends of the
displaced member. The member will have changed shape and size in the process, but
the equations of equilibrium formulated on this shape are satisfied and the position of
any material point, p, has been traced from its initial position, p0, to its new position,
p,. But, in starting the next step (and in reality, of course, any previous step ) we are
faced with a dilemma. We would like to formulate the equations of equilibrium on
Configuration t + /:J.t, but we can't since that configuration is still unknown.
To circumvent this problem we shall use the approach implied in the previous sub
section. Configuration t is defined as a "reference configuration" and all stresses, loads,
and displacements of the member in the next step are expressed in terms of this local
' '
x , 'y' coordinate system. The global tangent stiffness matrix is comprised of element
matrices determined from their local reference configurations and the global equilib
rium equation is therefore an approximate one that is linear in the incremental dis
placements. If a simple step solution procedure is adopted, the updated configuration
determined from a single application of this equation becomes the new reference con
figuration and the process advances to consideration of the next load increment. But
if one of the iterative schemes is followed, the first updated configuration becomes a
trial "deformed configuration" that may then be used as the basis for a successive
approximation scheme designed to converge on the correct Configuration t + /:J.t.
8.4 Problems 239
In the development of the requisite element matrices in the next two chapters we
shall generally use physical arguments appropriate to the case at hand. They will satisfy
our objective: the construction of equations that are reliable and useful in the design
of civil engineering structures. But in a few instances, notably in accounting for finite
rotations and the nonlinear interaction of bending moments and torques in Chapter
9, it will be necessary to turn to a more general approach, one that deals directly with
the fundamental problem of formulating the equations of equilibrium on the deformed
structure. We will use the results of that approach in Chapter 9, but we leave the details
of its development to Appendix A.
8.4 PROBLEMS
8.1 The system shown has a small initial displacement 0 Member ab is a rigid bar
and member be has a constant axial stiffness k. Determine the system's response as
the load Pis applied. Compare with the results of Example 8.1.
01' p
Problem 8.l
8.2 (a) Write the linear elastic global stiffness equation for the single-degree-of
freedom system shown. Compare with the comparable equation of Example 8.2.
(b) Nonlinear elastic behavior of this system will be analyzed in Example 9.2
using Equation 8.3. Explain how the same result could, in principle, be obtained solely
from the equation written in answer to part (a). Comment on the computational prob
lems that would be encountered and methods by which they might be solved in ob
taining such a solution.
Problem 8.2
8.3 The conventional linear theory of flexure used in Examples 8.3 and 8.4 is based
on the small displacement approximation of the radius of curvature (Eq. 4.28). Recall
the nature of the approximation and review the significance of the neglected term by
studying the more rigorous "theory of the elastica" as presented in texts on stability
such as Reference 8.7. Comment on the practical differences. In particular, in what
sense and to what degree does the neglected term affect the critical load and post-
,
critical behavior of an elastic column?
8.4 An elastic member has the small initial curvature shown. Using small displace
ment theory, analyze its response as the load Pis applied. Compare with the results
of part 2 of Example 8.4.
il
Problem 8.4
8.5 Using small displacement theory, analyze the response of the member shown.
Compare the result with Example 8.3.
p
M0= aPL
Problem 8.5
8.6 Member ab is elastic, but members bd and cd are rigid. Determine the critical
load of the system. Compare the result with Example 8.4.
p
b
d
T
Rigid
L E,I Rigid
1 a
0
c
0
Problem 8.6
8.7 Member ab is a rigid bar connected to a fixed base by a linear rotational spring
of stiffness k = MIO. (1) For a = 0 determine the critical load. (2) For a * 0 deter
mine the load-displacement response (consider small displacements only; assume
cos () = L).
p
! aP
I
I
L
Rig1ct
1 0
k=MIB
a
Problem 8.7
References 241
8.8 The beam shown is assumed to be made of elastic-perfectly plastic steel with
uy = 36 ksi and E = 29,000 ksi. For cross section (a), a wide flange section, analyze it
by the conventional elastic-plastic hinge method and plot the curve of P versus the
deflection at point b. Sketch the corresponding curve for the rectangular cross section
(b). Note the relative values of first yielding, second hinge formation, and plastic limit
load.
p
111.22"
!. ' JE8x71 ]
8. L i6._j (a) (b)
Problem 8.8
8.9 Repeat Example 8.6 for a uniformly distributed load, q = .02P kips/ft instead of
the concentrated transverse load.
8.10 Repeat Example 8.7 but with member ab assumed to be a rigid bar ( consider
large displacements ) .
REFERENCES
8.1 Load and Resistance Factor Design Specification for Steel Buildings, American
Institute of Steel Construction, Chicago, 1993.
8.2 J. G. Croll and A. C. Walker, Elements of Structural Stability, John Wiley and
Sons, New York, 1972.
8.3 S. P. Timoshenko, Strength of Materials, Parts I and II, 3rd edition, Van Nostrand,
Princeton, 1956.
8.4 A. L. Euler, "Methodus Inveniendi Lineas Curvas. . .", Bousquet, Lausanne,
1744.
B. W. A. Oldfather, C. A. Ellis, and D. M. Brown, Leonhard Euler's Elastic
Curves, Translated and Annotated, ISIS, Vol. XX, 1933, Saint Catherine Press,
Bruges.
8.5 T. V. Galambos, editor, Guide to Stability Design Criteria for Metal Structures,
5th edition, John Wiley and Sons, New York, 1998.
8.6 Plastic Design in Steel, A Guide and Commentary, Manual of Practice No 41,
American Society of Civil Engineers, New York, 1971.
8.7 S. P. Timoshenko and J. M. Gere, Theory of Elastic Stability, 3rd edition,
8.8 D. 0. Brush and B. 0. Almroth, Buckling of Bars, Plates, and Shells, McGraw-
8.9 H. Ziegler, Principles of Structural Stability, Birkhauser, Basel, 1977.
8.10 K.-J. Bathe, Finite Element Procedures, Prentice Hall, Englewood Cliffs, NJ, 1996.
8.11 R. D. Ziemian, W. McGuire, and G. G. Deierlein, "Inelastic Limit States Design,
Part I: Planar Frame Studies, Part II: Three-Dimensional Frame Study," Jl. of
Struct. Engr., ASCE, Vol. 118, No. 9, 1992.
Chapter 9
Geometric Nonlinear and Elastic

Critical Load Analysis
A stiffness method approach to the geometrical nonlinear analysis of planar and space
structures and the calculation of elastic critical loads is developed in this chapter. Al
though the results of the two levels of analysis differ-a continuous response curve in
the first case and one point in the response of an ideal model of a system in the
second-they employ the same form of the elastic and geometric stiffness matrices.
And both are useful in design.
The approach is based on Equations 8.3 and 8.6. The required elastic stiffness ma
trices will be drawn from earlier chapters, but geometric .stiffness matrices will be
derived in this chapter using virtual displacement principles and assumed shape func
tions for the purpose. In-plane behavior will be treated before out-of-plane behavior
for two reasons: (1) whereas, currently, geometric nonlinear methods are used rather
widely in the design of planar systems, their use in space structures is still limited; and
(2) consideration of finite rotations and the interaction of torsion and flexure, which
is essential to the nonlinear analysis of a space system, presents problems that require
special attention. Accordingly, in Section 9.1 we'll develop and apply geometric stiff
ness matrices for the axial force element and the element subjected to axial force and
bending about one axis. The interaction of axial force and St. Venant torsion is con
sidered in Section 9.2. The geometric nonlinear analysis of three-dimensional frames
is then taken up in Section 9.3. Some elements of the requisite 12-degree-of-freedom
geometric stiffness matrix are drawn from the findings of the two preceding sections
and some, specifically those relating to the interaction of torsion and bending, from
Appendix A. Proper treatment of this interaction, which involves subtleties in the
establishment of equilibrium on the deformed structure, requires a more mathematical,
structural mechanics approach than that which has sufficed to this point. The approach
referred to is important from the standpoint of satisfying our immediate need and,
even more so, for its clarification of the fundamentals of nonlinear analysis. We present
it in an appendix, however, because to develop it here would delay the application of
the results, which is equally important to the appreciation of their significance. These
results are illustrated in Section 9.4 through a series of examples of geometric nonlinear
analysis and elastic buckling.
As in Chapters 4 and 7, all of the stiffness matrices developed in this chapter are in
local coordinates.
242
9.1 GEOMETRIC STIFFNESS MATRICES FOR PLANA R ELEMENTS
We consider a prismatic element of bisymmetrical cross section subjected to an axial

force and flexed about its z axis (Fig. 9.la) and, in doing so, we'll disregard shear
deformation. This element and loading furnish a simple illustration of the formulation
of stiffness properties for geometric nonlinear analysis while giving insight into key
aspects of analysis common to all structural forms. As load is applied to the structure
of which the element is a part, its displacement from the reference configuration of
Figure 9.la is a function of how much it: (1) rotates as a rigid body in compliance with
the relative displacement of its ends; (2) stretches or shortens; and (3) bends. All three
occur simultaneously and they may influence each other. But to focus on the most
significant effects, we will neglect or approximate some minor coupling and treat the
total displacement as the resultant of two sequential actions: (1) stretching and rigid
body rotation of the axis (Figure 9.lb); and (2) flexure of the element with respect
to the chord joining the displaced ends (Figure 9.lc). And for the same purpose we
shall develop two geometrical stiffness matrices: (1) for the case in which there is no
bending and the problem reduces to that of a straight member undergoing the rota
tion and axial straining of Figure 9.lb; and (2) for a member subjected to combined
axial force and bending, in which case the deformations of Figures 9.lc must also be
accounted for.
9.1.1 Axial Force Member
In contrast to the infinitesimal strains that formed the basis of linear elastic analysis,
here we start with a study of small finite strain of a differential length of the element's
material. We are assuming the element to be straight and strain free in the reference
y
x+u+dx+ </J; dx l
x+u l
v 2 -r--- ---- --
-,----
-- .,-- --.,..,.. 2 '
(v + du dx ) ======::'.;
I ;iii'
dx v+ b'
- - 'a
l' --
VJ+---<...-
._...u.,...., 1--+------.
a ----.2----+-- x
b x
dx
U2 (c) flexure
x--------+<
t+----L------i
(b) stretching and rigid body
rotation of axis
Figure 9.1 Finite strain of planar element.
244 Chapter 9 Geometric Nonlinear and Elastic Critical Load Analysis
configuration and, since we consider only axial strains, we can again treat the problem
as mathematically one dimensional.
9.lb, the length of the material segment, dx, in the reference configuration
In Figure
ab. But after rigid body rotation and axial straining, its length is
is designated
2 2 1'2
a'b' 1 + 2
d
d + d
=
d
+[ d
d
dx
; ( ; ) ( ) ]
2 2
. . .
Des1gnatmg the mcrement, 2
du
dx
+
du
-
dx
+
du
dx ( ) ( )
, as dab, we have
- -
a'b'
[1 + dab ]112
dx
- =
Or, expanding by the binomial theorem and neglecting higher-order terms

2 2
a'b'
dx
1 +
du
dx
+
!
2
[( ) ( ) ]
du
dx
+
du
dx
Now, defining the extension per unit length, (a'b' - ab)/dx, as the finite strain, er;m
referred to the reference configuration, 1

2
er
m
=
du
dx
+
!
2
[( ) du
dx
+ (9.1)
Applying the principle of virtual displacements to the reference configuration (com

pare with Equation 6.15a)
8Wint = f Ux &r;n
vol
d(vol) (9 2)
.
Or, using Equation 9.1 and integrating over the depth of the member, we have
8Wint =
r <rxA(d::) dx r <rxA[ 8 (:r 8( rJ dx
+ + (9.3)
In the first integral of Equation 9.3, we have let 8(duldx) = (d8uldx), which is valid
for infinitesimal displacement. Now, using the conventional elastic stress-strain rela
<rxA x
tionship in the first integral and letting = F 2 in the second, the result is
8Wint = r (;)EA(d::) x LL [ (: r (:rJ dx + F 2 8 + 8 dx (9.4)
It may be seen that Equation 4.26a, the linear elastic stiffness matrix of the axial force
element, follows directly from the first integral of Equation 9.4 (see also Equ;ition
6.15c). Therefore, we need only concern ourselves with the second integral, which
produces the geometric stiffness matrix [kg]. Noting that the virtual operator 8 may be
treated as a differential operator with respect to the variables du/dx and duldx (Ref
erences 9.6 and 9.7), the internal virtual work for this term may be written as
8Win1 g = F x Jo(L [(
2
dBu du
dx dx ) ( +
d8v du
dx dx )] dx (9.5)
1The significance of neglecting higher-order terms in the binomial expansion is discussed in Section Al of
Appendix A. There it is shown that, whereas the use of strains based on Equation 9.1 (and Equation 9.la
of Section 9.2) in virtual displacement equations written on the reference configuration is satisfactory for
present purposes, the result is strictly valid only for small finite strains.
From the development of the virtual displacement formula for an element stiffness
matrix through the use of shape functions in Section 7.2 it follows that
[k ]
g = Fxz r [{N}LNJ + {N}LNJ] dx (9.6)
where LNuJ and LNuJ are shape functions for the respective displacement coordinates,
which in this case are those for a simple axially loaded member. Thus, from Equation
7.6 (with= x!L)
and (9.7)
Therefore
LNJ = l- J and LNJ = l- J (9.8)
Use of Equations 97 and 9.8 in Equation 9.6 results in
-1 0 1
[k
g]
= Fxz [ 0 -1 (9.9)
L -l 0 1 0
0 -1 0 1
Thus the geometric stiffness matrix is a function of the total axial force acting on
the element in the reference configuration. The components of nodal point displace
ment are incremental values referred to this configuration; in the equations of nonlin
ear analysis, they will be specifically designated as such. Application of Equation 9.9
is illustrated in Examples 9.1 and 9.2 of Section 9.1.3.
9.1.2 Combined Bending and Axial Force
To include the effects of bending on the geometric stiffness matrix, the strain due to
flexure (Fig. 9.lc) must be added to Equation 9.1. In doing this we shall neglect the
effects of the rotation and stretching of the element's axis on this quantity and use
Equation 4.28, the elementary mechanics approximation for infinitesimal flexural
strain. Thus the total finite strain becomes
efin =
: - (:) + H (:Y + (YJ
Adding the new term to Equation 9.2, letting
y
(9.10)
ux = -yM,11, in that term, noting that
fA y2 dA = I,, integrating over the depth of the member, and letting 8(d2v!dx2) =
(d28v!dx2), we have, in place of Equation 9.3
8W;n1 =
(L
Jo ( ) + J(oL ( )
UxA
d8u
dx
dx M,
d28v
dx2
dx
[ (r + (YJ
L (9.11)
+L UxA 8 8 dx
Using the elastic stress-strain relationship of Equation 4.31b in the new term and treat
ing the others as before:
8W;n1 = ( ) ( )
(L du
A
d8u
Jo dx E dx dx + ( ) ( )
(L d2v d28v
J o dx2 EI, dx2 dx
r [ ( r ( rJ
(9.12)
+ Fxz 8 + 8 dx
We have dealt with the first integral in the previous subsection, and we now note that
Equation 4.32, a linear elastic stiffness matrix, follows from the second integral (see
Eq. 6.21a). Thus reevaluation of the third integral using appropriate flexural shape
functions is the only task remaining. For this purpose we adopt the functions used for
the analysis of an element in simple flexure or, from Equation 7.9,
u (1 - g)u1 + gu
2
=
(9.l3)
V (1 - 3g2 + 2g3)v1 + (1 - 2g + g2)x0z1 + (3g2 - 2g3)v - (g - g2)x0z
2
=
2
Use of the derivatives of these shape functions in Equation 9.6 results in
U1 V1 ozl Uz Vz Oz2
1 0 0 -1 0 0
6 L 6 L
0
5 10 5 10
2L2 L L
z
[kg] = i2 -
15
0 - -
10 30 (9.14)
1 0 0
6 L
Sym.
5 10
2L2
-
15
9.1.3 Examples of Plane Structure Analysis
Application of Equation 8.3 for geometric nonlinear analysis and Equation 8.7 for the
determination of elastic critical loads is illustrated in the following elementary but
representative examples. The appropriate equations of Chapter 4 are used for the
formation of element elastic stiffness matrices and Equations 9.9 and 9.14 for the
geometric ones. The computer results shown have been obtained from the program
MASTAN2 and one of the equation solution schemes explained in Sections 12.2 and
12.3 or, for critical loads and buckled configurations, the algorithm for calculating
eigenvalues and eigenvectors described in Section 12.8.
Whereas in the examples of linear numerical analysis in earlier chapters it was gen
erally feasible to include extensive details of the formulation and solution of the prob
lem, normally this is not the case in nonlinear analysis. Even in simple problems, a
solution process composed of several increments and the use of subdivided members
may make detailed explanation impractical. The aim will be to give a clear statement
of the problem, key features of the approach used to solve it, and major results (e.g.,
a numerical answer or a graphical display of a particular behavioral phenomenon).
The statement of the problem and approach taken should be sufficient to enable the
reader to obtain the same results or, recognizing the possibility of reasonable differ
ences in nonlinear analysis, at least comparable ones.
Example 9.1 is similar to Example 8.1 except for features essential for computer
analysis: numerical specification of data-including a very large but not infinite area
for member be-and a pinned rather than a vertically sliding support at a. For the case
of a vertical load only, Equation 8.7 becomes elementary if it is assumed that the effect
of the vertical component of displacement on the critical load can be neglected. The
critical load becomes identical to that of Example 8.1 and the horizontal component,
ub, remains indefinite. For the case of a combined vertical and horizontal load, a com
puter solution of Equation (e) results in the detection of a limit point at 85% of the
critical load, which is approximately the same as the limit determined by Equations b ( )
()
and c of Example 8.1 for the comparable case of that example. The difference can
be attributed to the difference in specified boundary conditions, and more important,
the fact that, whereas the elastic stiffness coefficient of member ab was defined as a
constant in Example 8.1, it was continually updated in the present example. For a small
displacement problem, the change would be negligible, but in this large displacement
one it is significant.
Both members are axial force members

Ip
Aab = 2 mm2, Abe= 5 103
X mm2, E = 200,000 MPa
r:. '---ob
f aP
Analyze the response of the system for
La= 0 4m
a= 0.05
2.

4m --I
La= 0 8.7.
4.26b:
Calculate the critical load using Equation
Equation
b
[Ket] [[ J [
= 200 400
0 0
O
+
O
4000
X
O
5 01 3 = I_ [
]] lO O
( a)
Equation 9.9:
u0 v b
[Kv J = _
Pref[1 0
4000 0 1 ] (b )
Equation 8.7
(c )
Assuming vb = 0 and ub 0
=F
Pere = Ji.Pref = 400 kN (d )

From Example 8. 1
Pere = = 2 200 = 400
kL X kN
2. a= 0.05
Equation 8.3:
(e)
Solving iteratively: P (kN)

t--f>c...._=_40
__0
400
(1692, 340)
1 000 2000 3 000 ub (mm)
Example 9.2 is similarly related to Example 8.2. The member bd has been added,
and dimensions and properties have been specified. The computer solutions illustrated
contain provisions for continually updating stiffness coefficients and tracking physically
unstable regions of the response curves (see Section 12.7). In recognition of its sym
metry, the system is analyzed as a single-degree-of-freedom one. Increasing the stiff
ness of bd changes the response from a mode that is directly comparable to that of
Example 8.2 (Abd = 0) to one in which the region of instability is gradually decreased
and finally eliminated. It is of interest to note that in the case of Abd = 0 the numerically
calculated values of the displacement at the stability limit point (212 mm) and the
stability limit (2564 kN) are in agreement with the terms of the analytical solution
(Examples 8.2 and 8.8).
EXAMPLE9.2
All members are axial force members. E = 200,000 MPa. Analyze the large
displacement response of the system for Aab and Abe = 1000 mm2 and Abd =
0, 15, 30, and 45 mm2.
For equilibrium in a reference configuration-using symmetry and assuming the

geometry, member lengths, and bar forces have been updated-the stiffness ma
trices are:
From Equation 8.3:

Computed results: P(kN)

Abd= 45mm2
12000
9000
Abd= 30mm2
6000
3000
-3000
Example 9.3 is essentially a computerized version of Example 8.3. The member is

subdivided into two elements of equal length and the global stiffness equations (Equa
tion 8.7 for the first case and Equation 8.3 for the second) are assembled from the
basic element equations (Equation 4.34 for the elastic stiffness matrix and Equation
9.14 for the geometric one). The two-element idealization proves to yield anexcellent
approximation of the critical load.
EXAMPJ,,E 9.3
Member ab is a straight elastic bar
A = 1.27 x 104 mm2, I= 3.66 X 107 mm4, E = 200,000 MPa
Analyze its behavior for
1. a= 0
2. a= 1.25 X 10-4
Model the member as two elements:
1. a= 0
From Example 8.3
From Equation 8.7
APref = 1,137 kN
2. a= 1.125 X 10-4
Computed results are: P(kN) Pere= 1137 kN
1200
Eq.8.3
400
0
o 10 20 30 40 50 u2 (mm)
Example 9.4 is similarly related to Example 8.4. The nonlinear analyses are quan
titative demonstrations of the destabilizing effect of a compressive force and the stiff
ening influence of a tensile one.
The member is a straight elastic bar

r
A 1.27 x 104 mm2, I= 3.66 x 107 mm4, E 200,000 MPa
r
= =
rr
Analyze its behavior for: aP
1. a= O;
2. a= 0.005;
l_
3. a= 0.005 and P directed upward.
a
_x
0
Model the member as two elements
1. a= 0
From Example 8.4:
_,
ty
7'1-EI
Pere =
4L 2 = 1,129 kN
From Equation 8.7
APref = 1,129 kN ,
2 and 3. For a= .005 computed results are

P(kN) P(kN)
P re e= 1129 10000
8000
6000
4000
400
2000
O 0
O 100 200 300 400 500 U3 (mm) 0 5 10 15 20 u3 (mm)
Case2 Case3
The behavior of a slender two-story frame is studied in Example 9.5, first by calcu
lating the elastic critical load under vertical load alone, and then by using first- and
second-order methods to analyze it under 60% of that load plus proportionately ap
plied horizontal loads. A substantial P-ti effect is observed; the second-order theory
moments are significantly greater than the first-order ones. Analyses of this type may
be used in satisfying the requirements for determining the second-order effects of
lateral translation of building frame joints in standard design specifications (see Chap
ter C of Reference 9.2). The effect of column bowing has not been included in the
display of results, and in this case it would not be significant. It could be approximated
by using the shape function implicit in the geometric stiffness matrix of Equation 8.3
as an interpolation function. But where member bowing might be important, a
preferred approach would be to subdivide the member. The refined model would
give a more accurate representation of the member's stiffness and, as illustrated in
Example 9.6, it yields the displacements of internal loads directly.
Members are straight elastic bars.
ab and be: A = 2.5 X 104 mm2, I= 6.36 x 108 mm4 [ --r.

r---aP e
bd and ce:
E = 200,000 MPa
A = 1.76 X 104 mm2, I= 8.61 x 108 mm4
T4m
1. a= 0 Compute Pere = APref
2. a = 0.01 and P = 0.6Pcre
Compare first- and second-order elastic analyses.
t lr--
b ----,.!..
6m
l---x
0
l-- 4m
Model each member as a single element.
1. a= 0 From Equation 8.7 APref = 6630 kN P(kN)

2. a= 0.01, P = 3978 kN 4000 Eq8.l
Computed results are
Moment diagrams 30 60 90 120 150 u, (mm)

(kNm)
The single-story, single-bay portal frame analyzed by first- and second-order elastic
methods in Example 9.6 will be analyzed by inelastic methods in Chapter 10 (Example
10.5). In all cases the columns are modeled by two elements and each segment of the
gir der by a single element. As in the preceding example, a significant second-order
elastic effect is observed. The elastic redistribution of bending moments may also be
noted: the moment at b changes sign, for example, and there is a shift in the location
of the point of inflection in girder segment ed. _,-
EXAMPLE9.6
Members are steel wide flange sections !:o kips !

r:
20 kips
ab and de: A
be and cd: A
=
=
13.3 in.2, I
24.8 in.2, I
=
=
248 in.4
2850 in.4

t W27x84
24' W10x45 Wl0x45
E 29,000 ksi
l_ E____!
=
1. Compute Pere = APref r;

2. Compare first- and second-order elastic analyses
l- 20' 40' _____.j
Model columns as two elements and girder segments as one element.
1. From Equation 8.7 with Pref equal to the vector of applied loads, A = 2.20
2. Computed results are Load Factor
2661
044
519
I
I
I
0.4
Moment diagrams 0.2

(in. kips)
Example 9.7 presents the results of critical load and first- and second-order analyses
of a planar trussed framework. It is one of seven slender frames studied analytically
and in the laboratory by Hoff et al. (Ref. 9.3). They calculated critical loads by deter
mining the limit of convergence of the Hardy Cross moment distribution process as
applied to members subjected to axial force and bending, and they then compared
their analytical and test results. For the specimen considered here the calculated and
measured values were 203 pounds and 220 pounds, respectively. In the present com
puter analysis, each lower chord member has been modeled by two elements and all
of the other members by one. The critical load obtained by Equation 8.7 is 210 pounds.
As shown by a representative load-deflection curve, nonlinear response becomes
perceptible at less than one-half the critical load. Also, this particular deflection com
ponent reverses direction under the secondary effect and increases markedly in mag
nitude as the critical as approached.
EXAMPLE9.7
Members are elastic bars with
A = 9.3482 x 10-2 in.2, I = 6.9542 x 10-4 in.4,
E = 29,000 ksi
Compute Pere = AP,ef and compare first- and second-order elastic

analyses.
Model bottom chord members as two elements each and other

members as single elements.
N[2jS]/j
NS N6 N7 N8 N9 NlO Nll x
From Equation 8.7 Pere = 210 lbs

The buckled configuration from Equation 8.7 and the deflected
shape from Equation 8.3 at P = 210 lbs are:
-----
A
k21SJ21 Buckled shape

Eq 8.7
y
z
x
Deflected shape at P
Eq 8.3
= 210 lbs
Vertical deflection at Node 6: P (lbs)
Eq 8.3
so
-4 0 4 8 12
The effectiveness of a lateral brace of a compressed member is a function of its

strength and stiffness. Example 9.8 illustrates methods for calculating each in the case
of a column braced at midheight. It can be shown that the elastic critical load for an
2 2
ideally straight column of this type is 7T EllL , provided that the stiffness of the brace
in the lateral direction is equal to or greater than two times that load divided by L
( see Ref. 9.4 or 9.5) . For a smaller stiffness, the column will buckle in a single half
2 2
wave at a lesser load, with a minimum of 7T El/4L for no brace at all. This is dem
onstrated in the first part of the example. Since in the ideal case there is no force in
the brace at incipient buckling, analysis of the strength demand requires the assump
tion of an imperfect system. This is done in the second part of the example, in which
the midpoint of the column is assumed to be initially offset l/500th of the total length
( and nodes at the quarter points 3/4 as much ) . Second-order elastic analysis results in
a brace force of approximately 2 kips at the critical load of 215 kips. It is seen that, in
this particular example, both the stiffness and the strength required for the effective
bracing of the elastic column are modest.
EXAMPLE 9.8]
Member ac is an elastic bar braced at midheight by the elastic strut bd. E = 29,000 ksi.
1. Assume ac
of the brace.
ideally straight. Determine the variation of the critical load with the axial stiffness
T
!p
c
A = 7.08 in2
2. Assume an initial offset of 2L/500 at point b and a brace area, Ab,, of 1.27 in.2 Determine the I 18.2 in4
L l3'
=
=
variation of force in the brace with the applied load.

L Abr d
t
L = 13' Lb, = 10'
1. Buckling in two half waves will occur for kb, 2: 2PelL, in which Pe= -rr2-EIIL2 and kb,= Ab,E/Lbr The minimum condition
for this is
kb, =
( )
2 7T2(29,000)(18.2) .
= 2.744 k1ps/m.
.
156 (l56)2
or
Abr
_
-
(2.744)(120) _
- 0.01136
.
m. 2
29,000
Modeling ac by four elements and applying Equation 8.7 for a series of values of Ab, produces the Pere vs. Ab, relationship
of sketch (a).
2. For b offset (2 X 156)/500 = 0.624 in. and Ab, = 1.27 in.2, application of Eq. 8.3 to the same model produces the P vs.
Fb, relationship of sketch (b).
P(kips)
/
/
215.7
200 200
100 100
53.5
.005 .010 .015 1.0 2.0 3.0 Fb,(kips)

(a) (b)
9.2 COMBINED TORSION AND AXIAL FORCE
In Section 9.1 we treated structures in which the response to an action was in a single
plane. As seen in Sections 4.6 and 7.4, even in the simplest cases twisting about a
straight axis can be accompanied by displacement of points in the body of the element
in three coordinate directions. The interaction between torque and other components
of the applied force can have a profound effect on the behavior of a system. The
problem is a three-dimensional one that will be dealt with in Section 9.3, but one
important factor, the influence of axial force on the twisting of a shaft in which the
only mode of resistance is St. Venant torque, can be analyzed by a simple extension
of the approach of Section 9.1.
Consider the rectangular tubular element of Figure 9.2 subjected to an axial force,
Fx2, and torque, Mxz Twisting and net longitudinal displacement of the left end are
9.2 Combined Torsion and Axial Force 255
t=
x dx
A L-------+1
(a)
z --1+-W-z_.o
A-A
(b)
Figure 9.2 Combined Torsion and Axial Force.
prevented. A cross section at a distance x from that end will rotate about the centroidal
axis through an angle Bx and a point p in the tube's wall will translate as shown in
Figure 9.2b. For a small rotation, the components of translation in the plane of the
cross section are
v = -z8x and W =y8x

Thus
du d9x dw d8x
-z- =
dx ydx
-
and (9.15)
dx dx
By an obvious extension of Equation 9.1, the finite strain of a longitudinal fiber passing
through point p is
du 1 du ) 2
[(
efin = + + (9.1<0
dx 2 dx
Limiting consideration to the geometric effect of interest here we have, from Equa
tion 9.15
.
erm & =
!
2
[ z (ddx8x) 2 yz (ddx8x) 2] = 2! [
Z
+
Z
2
+
y2]
(ddx8x) 2
This is a warping strain, which, in the presence of an axial stress, is a source of internal
virtual work. Thus, following Equation 9.2, the relevant virtual work equation may be
(d8x) 2[ 2
written as
Loi
1
8Wint -l 8 Z + y2]dAdx
_
<Ix
g
dx
or, letting ax =
Fx21A and integrating over the cross section
(9.16)
in which fA (z2 + y2) dA =

IP, the polar moment of inertia of the cross section.
From Equations 9.4-9.8 it may be seen that, if a linear shape function is assumed
for 6x, Equation 9.16 leads to the stiffness matrix
[k ]
g
=
Fx2lp
AL
[ (Jxl
-l
1
(9.17)
In Example 9.9, this equation is used in conjunction with Equation 4.27a to deter
mine the torsional critical load of a column, a bifurcation condition in which twisting
about the axis of an axially loaded member becomes possible. A cruciform section is
used for illustration because, since the ratio of its torsional to flexural rigidity is rela
tively low, it is a shape prone to becoming unstable through twisting. For consistency
with earlier numerical examples Equation 8.7 is used in the formulation of the problem,
but in this case it is seen to lead immediately to the classical analytical solution (see
Ref. 9.4). Also, although for most structural shapes simple torsional buckling is nor
mally not the controlling mode, in the numerical example used for illustration, com
parison with the flexural critical load shows that it would be in this particular case.
EUJIPLE 9:9J
The member shown is a bisymmetrical steel cruciform section fixed against twist tx
JP
ing at the top and bottom.
A 48 in.2, Iy I, 1152 in.4
I
= = =
JP = 2304 in.4, J = 16 in.4, E = 29,000 ksi, v = 0.3
Determine the elastic critical load in pure torsion. Compare with the flexural L=20'
critical load. l -i
A
--
y
A-A
Model the member as two elements
From Equation 4.27a
From Equation 9.17
From Equation 8.7

9.3 Three Dimensional Geometric Nonlinear Analysis-An Overview 257
Thus
GIA (29000)(16)(48)
3718 ki
= =
Pert
=
APref
JP
=
s
2(1.3)(2304) P
compared with
7T2Ely 7T2(29000)(1152)
5724 kips
=
Pery
=
=
L2 (240)2
9.3 THREE DIMENSIONAL GEOMETRIC NONLINEAR ANALYSIS-AN O VERVIEW

As may be seen from Chapters 4 and 5, in linear elastic analysis the difference betwe'en
planar system and three-dimensional system analysis is essentially quantitative: The
computations are lengthier, but the basic equations are the same. In nonlinear analysis
the quantitative differences remain, and are indeed magnified, but there are also qual
itative differences. We already have the three-dimensional elastic stiffness matrix of
Equation 4.34 and we may write a three-dimensional geometric stiffness matrix simply
by adding to Equation 9.14 terms obtained by applying the same approach to a member
bending in the x-z plane and the effect of St. Venant torsion-axial force interaction.
With due attention to signs, the result is
U1 V1 W1 8xl 8y1 8,1 U2 V2 W2 8x2 8y2 8,2

1 0 0 0 0 0 -1 0 0 0 0 0
6 L 6 L
-
0 0 0 0 0 0 0
5 10 5 10
6 L 6 L
-
0 - -
0 0 0 - -
0 - -
0
5 10 5 10
Ip Ip
0 0 0 0 0 0 0
A A
2L2 L Lz
-
0 0 0 -
0 0
15 10 30
Fx2 2L2 L Lz
[kg) -
0 0 0 0
30
=
L 15 10
1 0 0 0 0 0
6 L
0 0 0 --
5 10
6 L
Sym 0 0
5 10
Ip
-
0 0
A
2L2
-
0
15
2L2
-
15
(9.18)
Equation 4.34 remains valid and Equation 9.18 is satisfactory insofar as it goes. It
may be noted that it contains a second order term corresponding to every first-order
term of Equation 4.34. In addition to accounting for the interaction of axial force and
St. Venant torsion just mentioned, it has provisions for accounting for the geometric
interaction of axial force and bending in the principal directions as the element dis
places and, since shearing forces are treated as reactions to the variation in bending
moments, for accounting for axial force-flexural shear interaction. The element stiff
ness matrices of Equations 4.34 and 9.18, when used in conjunction with the global
equations of Chapter 8, are adequate for the analysis of many of the three-dimensional
geometric nonlinear frame problems encountered in practice. Nevertheless, they are
less than fully comprehensive.
Among the ways in which these equations fall short are some that are beyond the
scope of this text, such as the treatment of nonsymrnetrical cross sections. Others,
however, relate to features that should be considered in any introductory work on
nonlinear frame analysis. They include the need to consider the interaction of other
components of force, such as bending moments and torque, the need to ensure the
ability to satisfy equilibrium on the deflected structure, and the desirability of including
provisions for the analysis of nonuniform torsion. To address these needs and desires
means going beyond the condition-specific physical arguments used to this point and
calling on more comprehensive fundamental principles of the mechanics of structures
and finite displacements. Full understanding of the equations of analysis and the as
sumptions made in their development requires that the theory behind them be ex
plained. But, for the reasons given in the introduction to this chapter, we are putting
a review of the essential theory and the development of the additional terms in Ap
pendix A. The terms relevant to the interaction of St. Venant torsion and flexure are
summarized in Equations A.38 and A.48. Adding these to Equation 9.18, we have:
U2 V2 W2 o ,, Oy2 0,2
u, v, w, o,, Oy1 o, ,
F,2 F,2
0 0 0 0 0 0
0 0 0 0
L L
6F,2 My2 M,2 F,2

6F,2 My, M,2 F,2
0 0 0
10 SL L L 10
SL L L
6F,2 M,2 F,2 M,2

6F,2 M,, F,2 M,2
0 0 L
10 L SL L 10
SL L
F,2/p M,1 + M,2 My1 + My2
F,2/p 2M,1 - M,2 2My1 - My2 My1 Mz1
0 6
6 L L AL 6
AL 6
_,Mxi F,, M,1 + M,2 F,2L M,2
2F,2L
0 0 2
L 10 6 30
lS
F,, M,2 My1 + My2 M,, F,2L

2F,2L
0 30
lS 10 L 6 2
(kJ =
F,,
0 0 0 0 0
L
6F,2 My2 M,2 F,,

Sym. 0
SL L L 10
6F,2 M,, F,2 M,2
SL L 10 L
F,2/p M,1 - 2M,2 My1 - 2My 2
AL 6 6
2F,2L
0
lS
2F,2L
lS
(9.19)
9.4 Examples of Three-Dimensional Structure Analysis 259
9.4 EXAMPLES OF THREE-DIMENSIONAL STRUCTURE ANALYSIS

Examples 9.10-9.12 are companion examples of the calculation of elastic critical loads
in elementary structures: a simply supported beam subjected to uniform bending about
the principal axis, an end-loaded cantilever beam, and a beam-column subjected to a
combination of axial load and uniform bending about the principal axis. In each, buck
ling occurs through a combination of lateral bending and twisting. Results obtained by
the application of Equation 8.7 are compared with classical solutions (see Ref. 9.4).
Since provision for nonuniform torsion is not included in the geometric stiffness matrix
used in the analysis (Equation 9.19) the neglect of warping resistance requires com
ment. In the first two examples the cross section is a narrow rectangle whose resistance
to warping is essentially nonexistent. A wide flange member with significant warping
resistance is used in the third example. However, as shall be shown in the analysis of
the results, for the prescribed loading condition neglect of warping restraint has only
a small effect on the calculated critical load. Nonuniform torsion in nonlinear analysis
will be treated in Section A.5 of Appendix A. Included in Example A.3 of that ap
pendix is an analysis of the member of Example 9.12 under loading conditions in which
the warping restraint effect is substantial.
In Example 9.10 a three-element discretization of the member produces a calculated
critical load within 5% of the theoretical value, which is often satisfactory for practical
purposes. Also shown in the example are plots of the midspan lateral displacement
and twist obtained from a second-order elastic analysis of the four-element model of
the member in which a small proportionate lateral (out-of-plane) load was added at
midspan. It is seen that the response becomes essentially asymptotic to the critical
load, lending support to that value as the member's limit of elastic stability. It should
be noted, however, that there are cases in which beams can undergo substantial in
plane deflection before buckling laterally and that this can have a significant influence
on the critical load (see Appendix A, Section A.5, Example A.2). The prebuckling
deflection effect was not included in the development of the classical solution referred
to in Example 9.10, nor in the formulation of the stiffness equations, in which the
undeflected member is taken as the reference configuration.
EXAMPLE 9.10
The member shown is a rectangular steel section fixed against twist

ing at both ends.
A = 20 in.2, I, = 166.67 in.4, ly = 6.67 in.4
J = 26.67 in.4, E = 29,000 ksi, 11 = 0.3
1. Determine the critical lateral buckling load.

2. Determine second-order response (add a perturbing load at
midspan).
1. The theoretical critical moment is(Ref 9.4)
'TT 290007T 6.67 x 26.67

Mer = YElyGJ = 6,280 in. kips
L
=
2.6
The calculated critical moments using Equations 4.34, 9.19, and 8.7, and three levels of discretization are
No. elements Critical moment (in. kips) Percentage error(%)
3 6570 5
4 6440 3
8 6320 1
2. Midspan response
M (in. kips x HP)
(Using 8-element model with a lateral load of M/800 L Y 7 Mer= 6320
at midspan)
ex 6
5 Eq8.3
4
3
2
1
0
0 0.2 0.4 0.6 w (in.)
0 0.004 0.008 0.012 ex (rad)
Example 9.11 is similar to Example 9.10 with perhaps the most significant difference
being one that is not obvious from the summary of results. The solution of the differ
ential equation of the end loaded cantilever is much less straightforward than that of
the uniformly flexed, simply supported beam (see Ref. 9.4) whereas in the formulation
and solution of the matrix equations there is no difference between the two.
EXAMPLE 9.11
The member shown is a cantilever beam of the same cross section as in Example 9.10.
Determine the critical lateral buckling load.
The theoretical elastic critical load is (Ref. 9.4)
p 4.013 y'El J 4.013 x 29,000 6.67 x 26.67

G - 267 k"1ps
= =
L2
er Y
602 2.6
The calculated critical load using Equations 4.34, 9.19, and 8.7 and three levels of discretization
are
No elements Critical load (kips) Percentage error ( % )
2 286 7
3 275 3
4 271 1
The buckled configuration of the member in Example 9.12 is again a combination

of twisting and flexing. Equation A.58 of Appendix A is the clasical critical load
equation for the member and loading shown. Controlling parameters in that equation
are Pcr and Pc,, the critical loads in minor axis bending and torsion, respectively. For
given end moments, one of the two solutions to the resulting quadratic equation must
be less than Per and Pc, and the other larger than both. Thus, the governing tor
sional-flexural critical load, Pc1r, is the lesser of the two solutions. In Example 9.12 the
warping resistance term in Pc, has been neglected. Also, end nwments proportional to
the axial load have been prescribed, and a value of the proportionality factor that
results in a critical load equation in which one root is infinitely large has been used.
9.4 Examples of Three-Dimensional Structure Analysis 261
The finite root is the simple function of Per and Pc, shown. This value is approximated
to within 2% by the matrix analysis of a three-element discretization of the member.
EXAMPLE 9.12
Member ab is a wide flange section subjected to an axial force and bending about its major axis.
A = 22.4 in.2, I, = 2100 in.4, ly = 82.5 in.4,
J = 2.68 in.4, C.., = 11,100 in.6, E = 29,000, v = 0.3
Calculate the critical torsional-flexural critical load. Neglect warping resistance.
The critical load is the smallest load that satisfies the following equation (Ref. 9.4 and Equa
tion A.58)
(a)
in which
Per=
TT2Ely
-U ,
GJA
and Pc, = -1 -
(neglecting warping resistance)
p
Letting M = aPL, Equation a becomes
( 1 -
a2
)L2
P2 - (Per + Pc,)P + (PccPc,) = 0 (b)
For illustration, let -

a2AL2
1- = 1. Thus
p
(c)
in which
p
er=
Tr2(29,000)(82.5) -
(24 X 12)2
.
- 284.7 kips and Pc, =
2.6
--
29,000 (2.68 x 22.4)
(2100 + 82.5)
.
= 306.8 kips
Thus
p
ctf =
(284.7)(306.8) - .
- 147.7 kips
(284_7 + 306_8)
he alculated critical load using Equations 4.34, 9.19, and 8.7 and three levels of discretiza
tion 1s
No. elements Critical load (kips)
2 155.3
3 151.1
4 149.6

It was noted that t e validity of this example would be influenced by the fact that
_ _
tors10nal warpmg resistance was neglected. Although true, the effect is not great in
'\
this case because the dominant contribution to displacement was minor axis bending.
Application of Equation A.58 to the same problem, but with warping resistance in
cluded, shows that, whereas Pct is more than twice the value used here, the critical
torsional flexural buckling load is only about 7% higher.
Example 9.13 is similar to the first part of Example 9.8 in that it is a study of the
effectiveness of lateral bracing, but in this case it involves the possibility of out-of
plane buckling of a planar system. The question is whether the tension diagonal has
sufficient stiffness to prevent transverse displacement of the compression diagonal at
its elastic critical load. It is found that if the tension diagonal is pin-connected at the
junction of the two and the compression diagonal is continuous, it does not. But if the
tension diagonal is continuous it does, regardless of the rigidity of the compression
diagonal's connections. The reason for the small difference between the critical loads
of cases 2 and 3 is that, as a continuous member, the compression diagonal derives
some increased resistance from flexural restraint at the junction.
EXAMPLE 9.13
All members are straight elastic bars. Beam-to-column joints are rigid. Diagonals are
pinned at each end and connected ate. /diagonals = 10.7 in.4, E = 29,000 ksi.
1. Calculate the elastic critical load, H, for

(a) Tension diagonals pinned ate.
(b) Compression diagonals pinned ate.
(c) Both diagonals continuous ate.
2. Compare results with classical theory.
1. Model all members as two elements and assume the beam is braced out of plane. From Equation 8.7 the critical loads
and buckled shapes are:
(a) Her= 86.0 kips (b) Her= 150.0 kips (c) Her= 164.0 kips
2. Compare axial force in be at H = 150 kips with its critical load as a pin-ended column:
From linear analysis for H = 150 kips, Fbe= 101.0 kips.
For be, L = 174.9 in.:
7T2(29,000)(10 7)
.
Per = 100.1 kip s
(174.9)2
=
Elastic force redistribution of internal forces in a space frame is illustrated by Ex

ample 9.14. As member ae buckles under load, the remainder of the system, paticu
larly member ab, assumes an increasing portion of the resistance to further loadmg.
9.5 Problems 263
[EXAMPLE 9.14
The member properties of the frame shown are
2 x 103 mm2, ly I, 3 X 106 mm4, I 6 X 106 mm4

ab, ac, ad: A = =
= =
2 x 102 mm2, fy I, 3 x 105 mm4, J 3 X 105 mm4

ae: A
= x
= = =
E = 200,000 MPa, 11 = 0.3.
1. Calculate the elastic critical load.

P and a vertical load
2. Determine the second-order response under the load
of P x 10-6 at the midspan of ae.
1. Model all members by two elements.

From Equation 8.7, Pere = 116.1 kN (sketch (a))
2. From Equation 8.3, force redistribution starts at approximately 110 kN (sketch (b))
P (kN)
200
Fae
(compression)
150
100
d
50
e
100 200 300 400 F(kN)
(a ) (b)
9.5 PROBLEMS
Note: In the following all references are to elastic methods of analysis and, unless
otherwise mentioned, to in-plane behavior only. Members are aligned or oriented
for major axis bending.
9.1 Compute the critical load and the nonlinear response for the truss shown ( mem
ber ab is elastic and be is essentially rigid ) . Use a = 0 in the critical load analysis and
a = 0.0025 in the nonlinear analysis.
r aP
R
2
4m A= 5x103 mm2
_!_, 0
E=
0
200,000 MPa
2m_j
Problem 9.1
9.2 For the frame shown compute the critical load factor for the following
reference
loads:
Pb Pc
1 2000 kN 0
2 1333 667
3 1000 1000
For each case use several levels of element discretization and compare results.
r b e
IA
b =1.76x104mm2
r
I =8.
61x108 mm4
A
1
5m E =200,000 MPa
L
=2. 5x104mm2
I =6.36x108 mm4
a d
0 0
I--- 4 m ------I
Problem 9.2
9.3 The same as Problem 9.2 except for the reduction of the moment of inertia of
member cd to 3.18 x 108 mm4 and use of the following reference loads:
Pb Pc
1 2000 kN 0
2 1000 1000
3 0 2000

9.4 (a) Compute the critical load and the buckled shape of the frame shown.
(b) Compare the results of linear and nonlinear analyses for P = 0.5Pcre
r!
2
T
...P_/__.... ... ;...._ _ _ - - - - ..,,. _....,
!,;;:.
: !P/2
A =24.8in2
I =2, 805 in4
24' .-------.--A = 13.3 in2
l
____./I =248in4
E
a c
0 0 =29,000 ksi
20' -i--- 40' __ ____,
Problem 9.4
9.5 Problems 265
9.5 Compare the results of linear and nonlinear analyses for the frame and loading
shown.
3m
A =4x HP mm2
I = SOx 106 mm4
1----'"1
1 Sm
a E = 200,000 MPa ej_

0
7.416m--+<--4m j
k---1-- 1.416m
Problem 9.5
9.6 Use nonlinear analysis to calculate the ratio of the maximum moment in the span
to Mb for the following cases: MjMb = -1.0, -0.5, 0, 0.5, 0.8 and PfPere from 0 to 0.9,
in which Pere = 7T2 Ell L2 Compare results with classical solutions (see the Commentary
of Reference 9.2 and Reference 9.4)
>-----L-----
Problem 9.6
9.7 (Note: The frame shown is one of those analyzed and tested by Hoff, et al.; see
Reference 9.3.) (a) Compute the critical load for the rigidly connected frame shown.
(b) Compare linear and nonlinear analyses of the fra1ne. (c) Compare the above results
with those obtained by assuming the system to be a pin-connected truss.
Problem 9.7
9.8 The member shown is restrained from twisting at each end. Compute its
critical
lateral buckling load. Compare the result with a theoretical solution in which
it is shown
that Pere =16.94YElyGIIL2 ( Reference 9.4).
i: A=22.4in2
r ,
---x
4

I z : i
I
8'
7=2.94in4
E=29,000 ksi
Problem 9.8
8' J
9.9 Compute the torsional-flexural buckling load of the frame shown. The frame is
prevented from twisting at a and c and it is laterally supported at b.
y p
t
b ir---'--c tA3 : in2
2 4
0 ly=115in4
J=2.94in4
ab: A=22.4in2
24' Iz=210in4.
ly=82.5m4
L
J=2.68in4
a E=29,000 ksi
0 I
x
5' -l- 5'
Problem 9.9
9.10 Compare linear and nonlinear analyses of the space frame shown.
All columns: A=10.4in2

I, =Ir=50.5in4
J=85.6in4
All beams: A=17.4in23
I,=ly 15 in4
J=47.2in4
=
E=29,000 ksi
Problem 9.10
9.11 Compute the critical load and nonlinear response of the system of Problem 4.15
loaded by loads of 1500 kN at nodes b and c.
9.5 Problems 267
9.U Study the response of the space frame of Problem 5.lO(b). Vary the boundary
conditions and the lateral support for the individual members. Compute the critical
load and full nonlinear response for each case studied and comment on the results, for
example, on the mode of behavior and the factors contributing to strength or weakness.
9.13 As in Problem 9.12, study the response of the space frame of Problem 5.lO(d).
Do not vary the boundary conditions in this study but do consider the effects of varying
the relative axial, flexural, and torsional properties of the two members.
9.14 Member ab is a straight elastic column supported by a linear spring at b. E =
200,000 MPa for both members. (a) Determine the minimum area of the strut required
to develop the Euler load of the column (See Examples 8.1 and 8.3). (b) Analyze the
resistance of the column for smaller values of the strut area. What is the resistance for
larger values? (c) In all cases, what is the force at incipient buckling?
Problem 9.14
9.15 The system is the same as in Problem 9.14 except that b is initially offset 15 mm
in the x direction. Analyze the resistance of the column for: (a) the minimum value of
Abe determined in Problem 9.14; (b) one-tenth of the minimum; and (c) 10 times the
minimum.
9.16 Member ab is a straight, slender nonprismatic column. E 200,000 MPa, A 1 =

=
1.27 X 104 mm2, 11 = 3.66 x 107 mm4, A2 3.175 X 104 mm2, 12 = 9.15 X 107 mm4
=
Calculate the critical load. Compare the result with a theoretical solution (see Refer
ence 9.4) in which it is shown that Pere mEl2/L 2 and that the parameter m = 6.68
=
for the proportions of this problem ( a = 0.4 and 11112 = 0.4).
L
I 'f:- A
, ,
J
j t:-
= 8 m aL = 3.2 m A2.I2
2 A I
, ,
Problem 9.16
9.17 Compare the linear and nonlinear response of this simplified suspension system
for (a) The load P applied at: (1) successive panel points; and (2) simultaneously at all
panel points; and (b) lg = 5,000, 10,000, and 20,000 in4
Ac= 5.40 in2 d
Ag= 50 in2
lg = (vanable)
f+------6@ 40' = 240' --------i
Problem 9.17
9.18 (a) Compare the linear and nonlinear response of this simplified cable-stayed
system for the load P applied successively at 40-ft intervals along the span and simul
taneously at all points. (b) Calculate the critical load for the system with P applied
simultaneously at all points.
All cables:
A= 5.40 in2
_J
0
E = 29,000 ksi
-------80'
Problem 9.18
REFERENCES
9.1 Y. C. Fung, A First Course in Continuum Mechanics, 2nd edition, Prentice-Hall,

Englewood Cliffs, N.J., 1977.
9.2 Load and Resistance Factor Design Specification for Steel Buildings, American
9.3 H. J. Hoff, B. A. Boley, S. V. Nardo, and S. Kaufman, "Buckling of Rigid Jointed
Plane Trusses," Trans., American Society of Civil Engineers, Vol. 116, 1951.
9.4 S. P. Timoshenko and J. M. Gere, The Theory of Elastic Stability, 2nd edition,
9.5 W. McGuire, Steel Structures, Prentice-Hall, Englewood Cliffs, N.J., 1968.
9.6 J. T. Oden and E. A. Ripperger, Mechanics of Elastic Structures, Second Edition,
Hemisphere P:.iblishing Corp., New York, 1981.
1996.
Chapter 10
Material Nonlinear Analysis
A stiffness method approach to the inelastic analysis of planar frames is developed in

this chapter. Equation 8.4 will be the base for considering the effects of material non
linearity alone and Equation 8.5 for the treatment of combined material and geometric
nonlinearity. The focus is on ductile systems: structures with members that, after a
period of initial elastic response, can undergo substantial inelastic deformation without
fracturing or becoming locally unstable. The material is assumed to be elastic-perfectly
plastic and coverage is limited to plastic hinge methods; that is, methods in which cross
sections are assumed to be either fully elastic or perfectly plastic. These methods are
sufficient for the analysis of a great number of the problems encournered in the design
of framed structures. Detailed analysis of strain hardening and distributed plasticity
effects is generally accomplished through more complex nonlinear finite element pro
grams, but procedures for bringing them within the purview of the concentrated plas
ticity approach have been advanced. Several will be summarized in Section 10.5 of this
chapter.
Aspects of the theory of plasticity and ways in which they have been adapted for
use in structural analysis are outlined in Section 10.1. Emphasis is on the background
and justification for the plastic liinge theory. In Section 10.2 a particular adaptation of
yield surface theory and the normality criterion is used in the development of a plastic
reduction matrix suitable for incorporation in Equations 8.4 and 8.5. The tangent mod
ulus theory approach to the determination of the inelastic critical loads of ideal struc
tures and its use in material nonlinear analysis is described in Section 10.3. Section
10.4 contains examples illustrative of the methods presented in the previous sections.
10.1 NONLINEAR MATERIAL BEHAVIOR
Whereas many of the sources of geometric nonlinearity are sufficiently similar that the
geometric stiffness matrix concept introduced in the previous chapter has widespread
applicability in both elastic and inelastic nonlinear analysis, the sources of nonlinear
material behavior differ in basic ways and thus require specialized treatment. For ex
ample, yielding of steel and cracking of concrete are fundamentally different phenom
ena, and further, each may be handled in a variety of ways. In steel the treatment of
yielding may range from a simple rigid member-plastic hinge analysis to a three
dimensional finite element analysis based on rigorous continuum mechanics and plas
ticity theory. In reinforced concrete there are numerous schemes for either modeling
individual cracks or for "sm aring" a field of cracks to obtain their gross effect. We
will concentrate on the devefppment and application of one approach to the matrix
analysis of frames of ductile material. Although limited in scope, the approach taken
269
270 Chapter 10 Material Nonlinear Analysis
is of significant practical utility, and it provides insight into problems common to all
types of material nonlinear analysis and guidance for their treatment.
In the engineering of framed structures we consider behavior at four levels: (1) at a
point and the differential region around it; (2) on a cross section; (3) throughout the
member's length; and (4) the system as a whole. In the chapters on linear elastic
analysis consideration of stresses at a point was bypassed by starting with accepted
engineering theories of their integrated effect on cross sections: axial force, bending
moment, and torque. These stress resultants were used in the development of element
equations, which then became the basis for system analysis. Since the subject was
structural analysis alone and the sequence followed was a straightforward adaptation
of classical displacement-based methods of elastic analysis, there was no need for ex
plicit categorization. But there is in nonlinear analysis, in which, at each of the four
levels, there are options for modeling some aspect of behavior. Proper interpretation
of the final analysis depends on an understanding of the choices made and where they
were made.
This may be illustrated and the role of various approaches to nonlinear analysis
clarified by the following summary of some of the ideas used in modeling inelastic
behavior and the schemes by which they are incorporated in systems analysis.
10.1.1 Plasticity Theory

The portion of the theory of plasticity of interest starts with consideration of a differ
ential element at a point in a body of elastic-perfectly plastic material subjected to
multidirectional stress. The material is presumed to have a known uniaxial stress-strain
relationship (Fig. 10.la) and, at the point considered, the principal stresses of Figure
10.lb. The theory involves the definition of a yield function and a corresponding.flow
rule. The most commonly accepted yield function is derived from the von Mises cri
terion and the flow rule from the normality criterion (e.g., Refs. 10.1 and 10.2).
The van Mises criterion postulates that the differential element will be elastic pro
vided that the function
(10.1)
(a) (b)
I
I
I
(c) (d)
Figure 10.1 Elements of plasticity theory.
is less than 2u;, that yielding can occur when it is equal to 2u; and that, for the ideally
plastic material, values of f greater than 2u; are inadmissible. Depending on the di
rection of subsequent loading, an element in a yielded state will respond in one of
three ways: 1) simple plastic straining; 2) a combination of plastic and elastic straining;
or 3) simple elastic straining (a return to an elastic state).
To simplify visualization of this process, consider the case of plane stress in which
u3 is equal to zero. For this, the yield function, fY' can be written nondimensionally as
2 + 2
0" 1 - 0"10"2 O"z
2
= l (10.2)
0" y
This is the ellipse shown in Figure 10.lc. It is generally called the yield locus or, inter
preting it as a symbol of the full three-dimensional theory, the yield surface.
The flow rule, which relates plastic strain increments to current stresses and stress
increments, is based on the idea that, in the plastic domain, the strain in a component
direction is the sum of a recoverable elastic strain and an irrecoverable plastic strain,
or, as in Figure 10.ld
(10.3)
For stress on the von Mises yield surface, the normality criterion states that, if plastic
strain occurs, its resultant must be normal to the surface at that point. Thus, for the
planar case of Figure 10.lc, its incremental components are
afy
deP.
l) = A (10.4)
auij
in which the factor A is indefinite if plastic flow is not constrained and a function of
the resistance offered by the surrounding body if it is constrained. It follows that the
incremental elastic strain must be tangent to the surface or, if there is no plastic strain,

directed inwards.
For a material that is not ideally plastic in the yielded state, for example, a material
with the uniaxial bilinear elasto-plastic characteristics of Figure 10.2a, the same rules
apply, but in addition, a hardening rule is required to specify how the yield surface is
modified during plastic flow. There are a number of these, but the most elementary,
and probably most commonly used, are isotropic hardening, in which the yield surface
is assumed to expand uniformly, and kinematic hardening, in which it translates in
space (Fig. 10.2b).
It is seen that these concepts-the yield surface, a flow rule, and a hardening
rule-were developed from consideration of conditions at a point in a body of ductile
..__
(a) Isotropic hardening Kinematic hardening

(b)
Figure 10.2 Hardening material.
material. Supported by experimental evidence, they and the underlying idea that
strains can be treated as the sum of elastic and plastic parts are the basis for much of
the theory of plasticity applicable to structural analysis. They may, for example, be
incorporated in the constitutive relations of inelastic finite element programs, carried
to the member level through a process of numerical integration and the system level
by global analysis (Ref. 10.3).
10.1.2 Plastic Analysis
What is commonly called "plastic analysis" in structural engineering treats of an ideally

elastic-plastic material (Figure 10.la) and, in its simplest form, involves two concepts:
the plastic hinge and mechanism formation (Ref. 10.4).
The plastic hinge concept postulates that the cross section of a member, say, the
section of a steel beam subjected to axial force and bending as in Figure l0.3a, can
have only two states: (1) completely elastic if the maximum stress as computed by
common engineering theory is equal to or less than uy , or (2) fully plastic under a
distribution of tensile and compressive stresses of magnitude u that equilibrates the
y
forces on the section. The latter case defines a plastic hinge, a section that can undergo
indefinite plastic strain under these forces if it is not constrained by the resistance of
the remainder of the system. The conditions under which such a hinge can form are
cross-section dependent. A commonly used bilinear formula for the case of a wide
flange beam bent about its major axis is (see Ref. 10.4):
1
M us( )Mp :::; Mp
= - (10.5)
Where P , the squash load, and MP, the plastic moment, are, respectively, the section's
y
area and plastic section modulus times uy. The graphical representation of Equation
10.5 in Figure l0.3b is an interaction diagram or, by extension of the terminology of
plasticity theory, a stress resultant yield surface.
Mechanism formation is a member or system process. In its conceptually most ele
mentary form, rigid-plastic analysis, elastic deformations are neglected and a search is
made for the smallest value of the load required to form a mechanism, that is, a pattern
of plastic hinges that will satisfy the conditions of equilibrium of the undeformed struc
ture and, at the same time, admit departure from that configuration. Application of
this process for finding the plastic limit load was illustrated in Example 8.6, which
employed the yield surface defined by Equation 10.5. That example was a very simple
one, however; the potential location of critical plastic hinges and the corresponding
0.15Py -
(a)
Figure 10.3 Plastic hinge formation.
p
rigid plastic analysis
first order-elastic-plastic analysis
(a ) (b)
Figure 10.4 Mechanism formation and first-order elastic-plastic analysis.
requirements for satisfying equilibrium on the undeformed structure were obvious

(Fig. 10.4a). That is rarely the case. Therefore, in routine plastic analysis it is generally
more practical to use a first-order inelastic analysis that produces the same limit load
and gives a more detailed picture of system behavior (Figure 10.4b). Essentially, it is
a piecewise linear elastic analysis with the capability for detecting plastic hinges as they
form and modifying them to deform plastically under further loading, with redistri
bution of the components of cross-sectional resistance following the dictates of the
yield surface. Preferably, it should also be capable of detecting and accommodating
the possibility of an elastic return from the yield surface.
10.1.3 Further Considerations
Plastic analysis as just described is practical, but it does not account for the gradual
development of inelastic zones (distributed plasticity as in Example 8.5) nor for the
second-order effects that contribute to instability (Example 8.7). There are finite ele
ment programs that include provisions for geometric nonlinearity and plasticity theory
based material nonlinearity that do not have these fundamental limitations. However,
their use in general structural engineering practice may be limited by the need for
specialized expertise in such tasks as the definition of realistic constitutive relationships
and boundary conditions, selection and control of the equation solution process, and
interpretation of results.
In conventional plastic design it has been customary to compensate for the inade
quacies of simple plastic analysis in some hybrid fashion. For example, under the plastic
design provisions of the 1989 AISC Specification for Structural Steel Buildings (Ref.
10.5), the maximum strength of one and two story steel frames "may be determined
by a routine plastic analysis procedure and the frame instability effect may be ignored"
and "members subjected to combined axial force and bending moment shall be pro
portioned to satisfy the following interaction formulas":
P
Per
+ -( P-)CmM
=::; 1.0
1- Mm
p (10.6)
---
e
p M
- + =::; 1.0, M =::; MP
Py 1.18MP
in which P and M are the calculated axial force and maximum moment in the member.
The first of these equations is an empirical one that contains several coefficients and
terms defined in the AISC specification and intended to account for member desta
bilizing effects that simple plastic analysis is incapable of simulating. The second, how
ever, defines a plastic hinge (Eq. 10.5) and thus will be an integral part of the analysis.
10.2 A PLASTIC HINGE METHOD FOR DUCTILE FRAMES
To develop a plastic-hinge-based method of analysis, we again consider the element

studied in Section 9.1, but in this case we define it to be of a ductile material. The
assumptions made in that section apply to this one as well, and fot: the treatment of
material nonlinear behavior, we add the following:
1. Plastic deformations are. confined to zero-length plastic zones at the element ends
(Fig. 10.Sa).
2. The material is assumed to be linearly elastic-perfectly plastic with no strain
hardening (Fig. 10.Sb ).
3. The effects of shearing stresses and direct stresses normal to the axis of the mem
ber on plastic deformation are disregarded.
4. As a consequence of the above, the end sections can undergo an abrupt transition
from a fully elastic to a fully plastic state. Combined forces (axial force and bend
ing) that initiate yielding on a cross section are assumed to produce full plastifi
cation of the section (Fig. 10.Sc).
5. Plastic deformations are governed by the normality criterion.
10.2.1 The Yield Surface and a Plastic Reduction Matrix
An initial task is the development of a plastic reduction matrix for use in the global
analysis equations. We will follow an approach devised by Porter and Powell (Ref.
'I Fy Ft y2
t1
01
I M,, E,A,I,,Z,
I
I o
x
L .r
(a)
y y
o
<1 max < <1 <1
(1
A B
H H
(b)
e
m A B
J JJ
A-A

B-B
(c)
Figure 10.5 Concentrated plasticity (plastic hinge) element.
10.2 A Plastic Hinge Method for Ductile Frames 275
p
1
cI>(p, m ) = 1
Plastic deformation Force increment
(b) (c)
(a)
Figure 10.6 Yield surface, force increments, and plastic deformation.
10.7) which, in turn, uses features of plasticity theory and plastic analysis summarized
in Section10.1.
Included is the concept of a stress resultant yield surface, which, for the element
considered, is assumed to be a continuous, convex function of the axial force and
bending moment on a cross section (Fig. 10.6a). Symbolically, it can be represented as
(p, m ) = 1 (10.7)
in which p = FxfPy, the ratio of the axial force to the squash load (Auy), and m =
M,IMP'' the ratio of the bending moment to the plastic moment (Z,uy)
As noted earlier, for practical application the function must be determined from
studies of actual sections. Several possibilities will be considered later. But of interest
at the moment is the basic concept that: (1) cross sections for which the force point ( a
point with the coordinates p, m ) lies within the surface are elastic; (2) sections with
force points on the surface are fully plastic; and (3) points outside the surface are not
admissible because of the assumption of an elastic-perfectly plastic material.
To illustrate, let us assume that at a particular stage in a global analysis, the "l" end
of the element has reached the yield surface while the "2" ( Fig.
end is still elastic
10.6a). Under the next increment of loading the force point at the 2 end may move in
any direction, but there are only two possibilities for the 1 end. Depending on the
response of the rest of the structure it may either remain plastic with a force point
constrained to move along the yield surface or unload elastically with its force point
returning to the interior. Both possibilities llato be considered, but we'll first deal
with continued plastic action. It is assumed that the resultant incremental displacement
at a plastified end can be treated as the sum of an elastic and a plastic contribution:
(10.8)
It is also assumed that the normality criterion applies. Thus from Figure 10.6b, for end
1:
{d4.Pi) = Ai{Gi} (10.9)
{'}
where {Gi} is the gradient to the surface at that point
{Gi) =
(10.10)
aM,1
and A1 is the magnitude of the plastic deformation.1 For an otherwise unrestrained

location, such as the fixed end of a cantilever beam, it would be theoretically unlimited.
In a stable system it will be constrained by contiguous elements and its value deter
mined by Equation 10.14 below.2 Admitting the possibility of plastification of both
ends of the element, we have
(10.11)
The matrix [G] is at the core of the development of the plastic reduction matrix. Its
only nonzero elements are at plastified element ends. At these points it reduces the
axial and rotational resistance and ensures that the other contribution to the total
deformation, the elastic one, is tangent to the yield surface at the force point. For this
reason, any calculated change in the force vector at such a point must follow the elastic
relationship:
(10.12)
where [ke] is the element's elastic stiffness matrix. Thus, for a point on the yield surface
the increment in force is tangent to it. As shown in Figure 10.6c, for a curved convex
surface this violates the requirement that the force must move along the surface. Pro
cedures for correcting for this discrepancy are discussed in Section 12.6.
From Equation 10.11 and the orthogonality of the plastic deformation and the cal
culated incremental force vectors, we have
or, since LA-J is arbitrary,

[GY{dF} = o (10.13)
From the above relationships, equations that relate the magnitude of plastic defor
mation and the force increment to the total deformation increment may be developed
by straightforward matrix manipulation. Thus, using Equations 10.8, 10.11, 10.12, and
10.13, and solving for {A.} yields
(10.14)
Similarly, using Equations 10.8, 10.11, 10.12, and 10.14 and solving for {dF} results
m
(10.15)
in which
(10.16)
is the element plastic reduction matrix. It is left to the reader to verify the development
of Equations 10.14 and 10.16 (Problem 10.la).
If, after a plastic hinge has formed at a particular section, the force redistribution
1 Justification for the use of the normality criterion in stress resultant space may be found in Reference 10.8.
To illustrate its application symbolically as in Figures l0.6b and c, a dimensional rather than a non
dimensional surface is employed.
2Although the "magnitude" of a deformation over "zero length" has no physical significance, the A's are
proportionality factors that indicate the magnitude of the theoretical "kink" in the deformed configuration
at a plastic hinge. More important, their signs indicate whether the calculated plastic flow can take place or
not (see Eq. 10.14 and its interpretation in the examples of Section 10.4).
10.2 A PlastiC Hinge Method for Ductile Frames 277
that takes place under subsequent loading of the structure is such as to reduce the
resultant action on that section, it may unload elastically. The appearance of a negative
element in the {A} vector computed by Equation 10.14 at the end of a load increment
is a signal that this may be occurring and that the increment should be re-done and
the analysis continued after ret.0.:r.ing the corresponding surface gradient to a null
vector.
10.2.2 Definition of the Yield Surface
Application of Equation 10.16 requires the definition of the stress resultant yield sur
face. Its dimension is equal to the number of stress resultants considered in the deter
mination of cross-section plastification and its form is a function of the section's shape.
In principle, therefore, it could be a six-dimensional "hypersurface." But in a plastic
hinge analysis of ductile frames in which the effects of shearing strain due to torsion
and transverse shear are not considered, it reduces to a three-dimensional one: axial
force and bending about two axes. The function's form is established by application
of the corresponding three equations of cross-sectional equilibrium to a given shape
of given material, say an elastic-perfectly plastic material in which the yield points in
tension and compression are assumed to be equal. A number of studies of this type
are presented in Reference 10.9. For a bisymmetrical wide flange shape, they may be
represented symbolically by one octant of a three-dimensional surface (Figure 10.7a).
Traces of the surface for a W12 X 3 1 section on the major and minor axis bending
planes are shown in Figure 10.7b.
Also plotted on Figure 10.7b are some equations useful in system analysis. The major
axis bending equation, Equation 10.5, and its companion minor axis equation from
Reference 10.4
(10.17)
are seen to be close approximations of the two traces. In the numerical examples of
Section 10.4 the major axis bending terms of the following continuous function have
been used.
<l> (p, my, m,) = p2 + m + m; + 3.5p2m + 3p6m; + 4.Smim; = 1 (10.18)
,... W12 x 3l{Ref 10.9) W12 x 31(Ref 10.9)

/p2 +m,2+3.5p2m,2=1 p2 +my4+3p6m/=1
p
,.,
M,
Axial force Axial force
major axis bending minor axis bending
(a) {b)
Figure 10.7 Wide flange section yield surface.
It is seen that this, too, is a good approximation of the traces. Equation 10.18 is a
modified version of a yield surface described in Reference 10.10, in which application
of the surface under conditions of biaxial as well as uniaxial bending is demonstrated.
10.3 INELASTIC CRITICAL LOAD THEORY
In Example 8.3 it was shown that the elastic critical load of the ideally straight column
of Figure l0.8a is the Euler load, Pere = Tr2EIIL2. If the column were made of the
elastic-perfectly plastic material of Figure 10.1, its theoretical limit of resistance would
be the lesser of that load or the squash load, Py = AuY" A long slender column would
buckle, and a short stocky one would fail by starting to squash. It is known, however,
that even short columns can buckle. The generally accepted explanation for this is the
tangent modulus theory (Ref. 10.11). It is based on the observation that the load
shortening relationship for a section made of a material such as structural steel is
affected by the residual stresses that result inevitably from the manufacturing process.
Thus the effective stress-strain diagram of the material is as in Figure l0.8b. Below a
proportional limit, up, it is elastic. Above that point it is inelastic with a gradually
decreasing resisance, measured by the tangent modulus, E,. The theory postulates
that, for an ideally straight column with an elastic critical stress greater than uP, bifur
cation of equilibrium can occur and the column will start to bend at a load
Tr2 EI
Peri= (10.19)
in which E, is the tangent modulus at the stress <Teri = Per/A.

Except for the use of E, rather than E, the development leading to Equation 10.19
is mathematically identical to that leading to the Euler load (Example 8.3). Physically
there is a difference in that, whereas E is only a function of the type of material, E, is
also stress dependent. A frequently used relationship between the two moduli is
(10.20)
This is an empirical expression designed to represent the performance of structural

steel columns in the inelastic range (Reference 10.11). Implicit in it is the assumption
that the proportional limit is half the yield point. In Examples 10.1-10.3 of the follow
ing section it will be used in the calculation of inelastic critical loads of ideal systems
by Equation 8.6 and the iterative algorithm of Section 12.8. Equation 10.20 may also
be invoked in second-order inelastic analysis to approximate the reduction in stiffness
of regions that are within the yield surface but subjected to an average axial stress
r a
T
L E,I,A
ay r-------:=---
a; 1--..-
(a) (b} Figure 10.8 Tangent modulus concept.

10.4 Examples 279
greater than one half the yield point. This, too, is done in the examples of the next
section.
10.4 EXAMPLES
Example iO.l is a study in the behavior of a pinned end column. In the first part of
the example, elastic and inelastic critical loads are calculated by Euler and tangent
modulus theory for a length of 10 feet, which places the member in the intermediate
slenderness range. It is seen that the tangent modulus load can be obtained directly
from the simultaneous solution of Equations 10.19 and 10.20. A four-element model
of the column of the same length, but with an assumed parabolic imperfection of
amplitude equal to 1/1000 of its length, is then analyzed by second-order elastic and
inelastic methods. In the inelastic solution, the effective modulus under axial stress
greater than 25 ksi is reduced according to Equation 10.20. For the case analyzed, the
calculated limit point is almost identical to the inelastic critical load.
Consider weak axis behavior of the column shown
A = 9.13 in.2, I = 37.1 in.4, Z = 14.1 in.3, r = 2.02 in., E = 29,000 ksi, ay = 50 ksi
L For L = 10'
(a) Calculate the elastic and inelastic critical loads.
(b) Assume an init411. parabolic bow of L/1000 amplitude. Analyze by second-order elastic and
inelastic metho
2. Use second-order inelastic analysis to determine the resistance of the bowed member for
0 s L/r s 200. Compare with Equation 10.21.
1. Use four element modeling

p
(a) ere
71'2El
L2
=
2
71' (29,000)(37.1)
=
(120)2
=
'
737.4 klpS r
.
From Equation 10.19 aeri =
Peri
A
=
71'2 ,(37.1)
(120)2(9.13)
(1)
r
b
From Equation 10.20 E, = 4(29,ooo {; ( - ) J

1 (2) c
From (1) and (2):

a x
E, = 15,200 ksi, aeri = 42.26 ksi, Peri = 385.8 kips
(b) Response curves
200
Eq 8.5
2 3 4 Uc (in)
2. Resistance curves P (kips)

500 Py
------- ,
400 \
\Pere
\
300 \ Eq8.S
200
Eq 10.21
100
so 100 150 200 L/r
In the second part of Example 10.1, the inelastic analysis procedure is used to cal
culate the resistance of the imperfect member over the range of L/r, the slenderness
ratio, from zero to two hundred. The result is compared with the AISC LRFD column
formula ( see Ref. 10.6)
for A :5 1.5
for A> 1.5 (10.21)
in which
,\=
KL Ju;,
TTr {
For the effective length factor K = 1, and the assigned properties, the two parts of the
column equation become
and
O'er
=
[5020.2]
(L/r)2 O'y
respectively.
In comparing results it should be recognized that the resistance of a column is a
function of its cross-sectional shape, the nature of its supports, and its geometrical
imperfections and residual stresses. Even among simply supported columns, there is a
wide range of behavior ( see Ref. ) . Behind the reduction of this complex behavior
10.11 10.21,
to the single design formula, Equation there is an extensive body of analytical
and experimental research. In contrast to this, the calculated curve is the result of a
plastic hinge analysis of a single, arbitrarily selected, section. The agreement between
the two curves is good: the difference in resistance is generally less than 10% and the
maximum is 14% , which is within the range of variability-the "scatter band" -of the
actual resistance of a column of given nominal dimensions and properties. This could
be viewed as fortuitous, but a more favorable interpretation is that it is evidence that,
in cases in which the columns are components of a system with members and boundary
conditions that can be reasonably modeled, an inelastic analysis of the type illustrated
can be a practically useful predictor of nominal strength.
In Example 10.2, the proportions of the ideally straight two-story columns are such
that, at the critical load, the upper story is inelastic while the lower one remains elastic.
10.4 Examples 281
It demonstrates that a mixture of "Euler" and "tangent modulus" behavior can be

accommodated by the algorithm of Section 12.8. The calculated critical load is between
the values for the two spans treated as independent pin ended columns. The inference
is that the upper, lower capacity, segment derives some restraint from the lower one.
fBXAMPLE 10.:i]
For both spans of this continuous column, E = 29,000 ksi and ay = 33 ksi. For be, A = 9.13 in.2, I= 37.1
in.4 For ab, A = 17.6 in.2, I= 116.0 in.4
1. Determine the critical load of the system and the corresponding stress in each span.
2. Compare with the result obtained by treating each span as an isolated pin ended column.
1. From Equation 8.7, using a four-element idealization for each span p
Per = 265.5 kips
be: U'er = 265.5/9.13 = 29.1 ksi
ab: U'er = 265.5/17.6 = 15.1 ksi
span be is inelastic and ab is elastic at the critical load.
2. As pin ended columns the critical loads would be

4
be: Following Example 10.1, E, = 1 ,500 ksi, aeri = 28.1 ksi, and
2
. 7T (14,500)(37.1)
pen =
44 2 = 256.1 k"lpS < per
(1 )
ab:
p 7T2(29,000)(ll6)
ere = 2 = 294 1 kl" pS > per

(336)
The "no sidesway" and "sidesway" elastic and inelastic critical loads of two frames
are the subject of Example 10.3. In its sidesway-prevented state the columns of the
first frame are effectively "short," and the material strength is the determining factor
in its resistance. When sidesway is not prevented, the column effective length is much
greater; the strength of the material has little effect on resistance. The second frame
presents a different situation. Even without an external sidesway constraint, its pro
portions and loading are such that the exterior columns brace the heavily loaded in
terior one. The inelastic critical load at which the interior column is essentially fully
supported is the governing condition.
Determine critical loads and buckling 1 2
29,000 ksi, ay =
modes for the following frames. E =
36 ksi. Use Equation

8.7 and two-element idealization for
r r
all members. r A= 13.5 in,/= 712 in
A= 9.13 in2
r All members
15 ft 10 ft A= 35.3 in
l :___:_ = 112 in
l I= 1380 in
x x
0 0 0 0
----- 40 ft --- 10 ft-..+.-1-- 0 ft --I
Critical loads and buckled configurations:
1. Consider sidesway prevented and sidesway permitted

cases.
r er
_____ _.,. _____
Pere= 1,510 kips Pere= 213 kips

Peri= 315 kips Peri= 202 kips
2. Consider sidesway permitted case. Pere= 14,400 kips Peri= 1,346 kips
i i
Application of Equation 8.4 for material nonlinear analysis and Equation 8.5 for
geometric and material analysis is illustrated in Examples 10.4-10.8. The appropriate
equations of Chapter 4 are used for the formation of element elastic stiffness matrices,
Equation 9.18 for the geometric ones, and Equation 10.16 for plastic reduction matri
ces. The basic equation solution routines are the same as those of the examples of
Chapter 9, but additional features associated with consideration of material nonli
nearity include (1) detection of plastic hinge formation; (2) tracking the yield surface;
and (3) detection of possible elastic unloading from the yield surface. Summaries of
devices used for treating these conditions are presented in Chapter 12.
3P
i:
The member and loading are those of Example 8.6.
1. Compare first- and second-order inelastic analyses.

:
c p
-
x
Ls, .I.
2. Determine the plastic deformation vector and the incremental W12x65
force relationship at first hinge formation in the first-order
analysis. 16'
10.4 Examples 283
1. Idealize ab and be as single elements.

From Example 8.6 Py = 955 kips MP = 4840 in. kips
From Equation 10.18, the yield surface is

( )2 ( )2
p
+
M
+ 3.5
( PM )2 1 (a)
955 4840 4_62 x 106
=
From Equations 8.4 and 8.5, response curves and force point traces are:
P (kips) P (kips)
400 First-order inelastic 955
4840 M (in. kips)

2. From first order analysis, first hinge forms at a when P 259.7 kips, M0
= = 4151.2 in. kips
p = 259.7/955 = .2719 m = 4151.2/4840 0.8507

=
a 2P 7P m 2
+ 0.002036
aP
=
Py Py
=
.l.0{Ga}
a 2m
+
7p m 2 0.0004461
,...___{Fa } {2541951..72}
=
aM Mp Mp
= =
. {0.002036}
Equation 10.9: {dapa) .A0{G0} ..\0
0_000446
= =
. {dP0} {dP0}
Equation 10.13: LG0J L.002036, .0004461 J 0.
dM dM
= a =
. . dM
" 0 = -4.564 dP0
Example 10.4 is the same as Example 8.6 except for the use of a different yield
surface and both first- and second-order matrix analyses. As expected from the close
agreement of the yield surfaces, the difference in first-order analysis results is incon
sequential In the second-order matrix analysis the limit point is found to be approx
imately 5% less than the simple plastic limit load; the destabilizing P-8 effect is per
ceptible but not very large.
In this example the member is subdivided into the elements ab and be in advance
recognition of the potential location of plastic hinges. Initially the [Km] matrix of Equa
tion 8.4 or 8.5 is null and the analysis starts out as a conventional elastic one. The first
hinge is detected with the aid of an algorithm based on the determination of the dis
tance from a calculated state to the yield surface, plus an iterative procedure designed
to converge on that surface to within an acceptable tolerance (see Section 12.6). The
situation that exists once that has been accomplished is illustrated by the case of the
first plastic hinge in the first-order inelastic analysis. The form of the plastic defor
mation vector and the incremental force relationship at that hinge are determined from
Equations 10.9 and 10.13, respectively. The magnitude of the components of plastic
deformation and incremental force remain unknown until computed in the next step
of the analysis, but [Km) is no longer null. When calculated by Equation 10.16 and
used in the global analysis through Equation 10.15, the force point at a is constrained
to move tangent to the yield surface at that point. Steps that may be taken to correct
for the spurious outward drift of a hinge point from a convex yield surface are de
scribed in Section 12.6.
The plastic reduction matrix of Equation 10.16 is seen to have its roots in elementary
plastic analysis. For this simple case, the elementary approach of Example 8.5 is as
expeditious as any, but its application to structures of any size verges from tedious to
impossible. The matrix approach has i.ts own numerical and behavioral problems: those
illustrated by this example and also, for example, the detection and treatment of elastic
unloading at a hinge and the avoidance of spurious "joint mechanisms" (a false signal
of instability from the formation of plastic hinges on two sides of a node). But most
of these are problems that can be dealt with effectively in the computer programs that
are essential to its application.
The structure of Example 10.5 is the same as that of Example 9.6, but in this case it
is subjected to two levels of inelastic analysis. In both first- and second-order inelastic
analysis, failure is the result of plastic hinging at the same locations. The P-11 effect is
seen to reduce the load carrying capacity by 12%. It should be noted that the yield
surface is represented nondimensional!y because the force point traces pictured are
for two different cross sections. The first plastic hinge forms at the top of a column
under a combination of bending moment and a substantial axial force, the second hinge
in the girder, in which the axial force is insignificant. Between the formation of the
first and second hinges the moment at the column top decreases as the axial force
increases, as dictated by the yield surface. In the plot of the response curves, results
of the elastic analysis of Example 9.6 are added for comparison.
(!i'XAMPLEJ0.5)
The frame and loading are the same as in Example 9.6. ay = 36 ksi. 1 60k
Compare first and second-order inelastic analyses. Use the yield surface of
Equation 10.18
r
6k r: l t c
21xs4
24' W10x45
l__ E..._____.1 e
0 ::::::;:
20' 40' ----..j
Model columns as two elements and girder segments as single elements
WlO x 45: Py Aay 13.3 X 36 478.8 kips, MP Zay 54.9 X 36 1976 in. kips
Py MP
= = = = = =
W27 x 84: = 24.8 x 36 = 892.8 kips, = 244.0 x 36 = 8784 in. kips
(p)2 (M)2 (PM )2

From Equation 10.18:
Py MP + + 3.5
PYMP = 1
10.4 Examples 285
From Equations 8.4 and 8.5, results are: 1665 } PfPy

1720 1
0.8
0.6
{8783 first-order 0.4
8783 second-order 0.2
second-order
Limiting moment diagrams
Eq8.4
Second hinge
Mechanism 0o 5 10 15 20 25 ud (in)
Example 10.6 is similar to Example 10.5 except for the addition of a column and
specification of a loading pattern designed to illustrate further the P-ll effect. The two
levels of analysis-first- and second-order inelastic-yield two different failure modes.
In the first case it is a simple story-wide panel mechanism. Under the combined de
stabilizing effects of geometric and material nonlinearity included in the second case,
a stability limit is reached prior to mechanism formation.
Members ab, cd, and ef are WlO X 45 sections. Members bd and df are W27 X
84 sections (see Example 10.5 for section properties). E = 29,000 ksi, <Ty= 36 ksi
Compare first and second-order inelastic analyses. Use the yield surface of
Equation 10.18.
-=ma.,,. ___! c e
0
20' ---- 40' -----1
Model columns as two elements and gird

ers as single elements. From Equations 8.4
and 8.5 results are:
Limiting moment diagrams (in. kips).
1975 943 1572 1975 1116 1657

First-order inelastic Second-order inelastic
Plastic hinge formation sequence:

2 2
First-order inelastic Second-order inelastic
Response curves: P (kips)

312 First-order inelastic
240
Second-order inelastic
160
80
3 6 9 12 ub (in)
Example 10.7 presents the results of a second-order inelastic analysis of a frame

subjected to non-proportional loading. It is one studied analytically and experimentally
by Arnold et al. (Ref. 10.12). Beam loads of 20 kips each were applied first and held
on the structure while 60 kip loads were placed on each column. With these loads in
place, the frame was then tested to failure under horizontal loading. Shown in the
example are a moment diagram, deflected configuration, and a response curve calcu
lated by Equation 8.5. Member sizes and yield strengths reported in Reference 10.12
were used in the analysis. The agreement between analysis and test is good. The re
ported failure load was 7% higher than the calculated value, a difference attributable
mainly to the actual finite size of the joints as compared to the point assumed in the
analysis. The test frame was also somewhat more flexible than the analytical model.
This may be attributed to small column base rotation observed in the test.
fi__________________ _
rf- -T,9
The properties of the frame shown are:
W5 x 18.5: A = 5.45 in.2, I = 25.4 in.4, 3W

r ,, W ,, 3W
9 49
Z = 11.4 in.3, u
y
= 56.2 ksi 81"
SlO x 25.4: A = 7.38 in.2, I = 122.1 in.4, b e c H
I
S10x25.4
Z = 28.0 in.3, u
y
= 38.6 ksi
E = 29,000 ksi
105" W5 x18.5
1
Loading sequence:
d
1. Apply beam loads, W 20 kips. -x
0
=
2. Add column loads, 3W = 60 kips. a I

t+----179" --+j
3. Add H and load to collapse.
Calculate the response of the frame by second-order inelastic analysis.

10.4 Examples 287
Model columns as two elements; be, ef, and Limiting moment diagram ( in. kips ) Plastic hinge formation sequence
fc as one element. Each from Equation 8.5; 349 546
using the yield surface of Equation 10.18: I
4
I
I
I
I
I
3
564 546
Response: H (kips)
20
Test
16.8
15.
71-=-5+------->3...4...._
10
1 2 3 4 Uc (in.)
The use of distributed rather than concentrated loads in nonlinear analysis is dem
onstrated in Example 10.8. In this, as in all the systems analyzed in this section, plastic
hinges can only form at defined nodal points. A burden is placed on the analyst to
define nodes reasonably close to potential hinge locations. Few practical systems are
so sensitive to hinge location that this is a problem, however.
All members are W30 x 108 sections
A = 31.7 in.2, I = 4470 in.4, Z = 346 in.3, uy = 50 ksi, E = 29,000 ksi
Assume proportional loading.

Compare first and second-order inelastic analyses.
Model columns as two elements and girders as four elements. Use the yield surface of Equation 10.18.
,
Define uniform load in local coordinates:
From Equations 8.4 and 8.5: Response curves
Plastic hinge formation sequence

Load Factor
2.5
2.13 "--t-----,..--
, --
2
1.86 T----?"-:--::;;>
l.5
1
1
0.5
0
o 5 10 15 20 25 30 ub (in)
Example 10.9 is a study in deflection stability, a phenomenon related to the repeated

application of a sequence of loads. If, as in the example, a two-span beam is subjected
to the loading pattern shown, it will remain elastic provided the maximum moment
under either loading is less than the plastic moment. At the other extreme, a plastic
limit will be reached under the initial application of the first load if it is sufficiently
large to cause a beam mechanism to form. For loads between the elastic and inelastic
limits, plastification at either potential hinge point may occur and then be relieved as
the sequence progresses. Because of the irrecoverable plastic rotation at an active
hinge, the member will be permanently deformed and a residual moment will remain
after the removal of the particular load. The situation will stabilize (colloquially
"shakedown") if the magnitude of the load is equal to or less than a certain value,
otherwise the permanent deformation will increase continuously, with the potential of
becoming excessive. It can be shown that shakedown will occur if, at points of maxi
mum moment, the absolute value of the sum of the residual moment, M,, and the
elastic moment, Me, produced by the loads does not exceed the plastic moment, MP.
Thus, for a general point i
(10.22)
The illustrative example used is from Reference 10.4. The only possible residual mo
ment diagram is the one shown. In the response curves an application or removal of
a pattern of loads is designated as a step in the "loading sequence." It is seen that, at
the shakedown load, stability is obtained after the first full cycle whereas for a load
between it and the plastic limit, permanent deflection increases continually.3
The continuous beam is subjected to the repeated loading pattern shown

t d b c
I = 533 in.4, Z = 96.8 in.3, ay 50 ksi, E = 29,000 ksi
w12x65 ---X
1. Determine the elastic limit, the plastic limit, and the shakedown load. 1-- L = 24' -I-- L 24' __j =
2. Study the system using first-order inelastic analysis for

(a) the shakedown load r LoadA
(b) P = 95 kips
LoadB

3Although, as shown, prQgressive plastification of this sort is possible, it is not often a factor that requires
"
investigation in design. See l(ference 10.4 for further discussion of this subject.
10.4 Examples 289
1. Moment diagrams (elastic ordinates shown for Loads A and B)

___ PL A\12
64PL

64PL
M,
Load A LoadB Residual moment
Limiting conditions for MP =96.8 X 50 =4840 in. kips:
64 Mp 64(4840)
Elastic limit: P = = =82.73 kips
. 13 L 13(288)
PPL
Plastic limit: Mp +RaL--=0
2
and
RLa
=O
_
Mp
2
:. PP =6MPIL=100.83 kips
13 1
Shakedown load: From Eq. 10.22 PsL + 2 M, =MP
64
and
12
Ps - M, =MP
64
Mp 96 MP
:. M, = -19 =254.8 m. k'1ps and Ps = 84.91 k'1ps
19 L
=
2. Response
l>'
Loading sequence: (1) Load A, (2) unlad, (3) Load B, (4) unload, etc. o'ri/ l>/
14 'V/ o'ri
y'/
12
p. 10
B
"'
"O 8
"'
0
...l 6
4
2
0
0 2 4 6 8 10 12 14 VJ (in)
Example 10.10 is a summary of the results of a comparative study of a "fully braced"

and an "eccentrically braced" frame. The two differ only in the insertion of a flexure
and shear resistant link in the center of the second. Both are simplified representations
of framing and loading in a lower story of a multistory diagonally braced system.
Although the example is unrealistic to the extent that the effect of shear on yielding
is not included in the second-order inelastic analyses to which the frames are subjected,
there are lessons in the different modes of behavior that are of use in actual design.
Frame A is stiffer and slightly stronger than Frame B, but its collapse is precipitated
by inelastic buckling of member cd and it has little post limit bucking resistance. Frame
B absorbs a greater amount of energy through progressive yielding of its beams;
its failure is more gradual. Considerations of this sort are particularly important in
earthquake-resistant design in which a key factor is the amount of energy required

to destabilize a system.
Compare the behavior of the follow

ing frames, E = 29,000 ksi. For 1200kfY
r- 14' ----r- 14' -12400k 1200kr13' 2400k
beams, <Ty = 33 ksi. For columns and
diagonals, <Ty = 100 ksi. t Wl0x30 l e h5ok 150k t WlO x30
T' b

b
12
-
i
Frame A Frame B
'
Model ee by one element, all other mem
bers by two elements. Use Equation 8.5. [7178
Account for elastic shearing deformation
(see Example 7.11). 1.78
Failure modes and plastic hinge formation
sequence (by load factor):
1.78
Response curves: Load factor
2
1.78----+-.
--
1.66 1.5
1
0.5
0.5 1 1.5 ub (in)
10.5 THE YIELD SURFACE CONCEPT-A BRIEF SURVEY OF FURTHER

APPLICATIONS
The stress resultant yield surface concept has been adapted, in various ways, to the
treatment of additional problems in material nonlinear analysis, including the spread
of plasticity, strain hardening, and the analysis of reinforced concrete members. For
several reasons-limited applicability, the need for extensive empirical support, or
special computational requirements-application of these extensions will not be un
dertaken here. But the following summaries of a few of them illustrate the concept's
potential.
10.5.1 Spread of Plasticity
A zone of gradually developing plasticity of the type studied in Example 8.5 can be
reduced to a "quasi-plastic hinge" and analyzed by the matrix methods of this chapter
following a procedure presented in Reference 10.13. The reduction is.accomplished
through a process in which nonlinear equations for the force-strain relationships of
partially plastified cross sections are integrated numerically along the element to obtain
inelastic flexibility coefficients, which are then used in generating an inelastic stiffness
matrix. To illustrate, we consider the beam-column element of Figure 10.9a with
10.5 The Yield Surface Concept-A Brief Survey of Furtlier Applications 291
Full plastification surface
(b)
(c) (d)
Figure 10.9 Quasi-plastic hinge model.
regions near each end in which the maximum stress is greater than the proportional
limit. As shown in Figure l0.9b, the force point (M, P) of a typical cross section in one
of these regions is between the initial yield surface, a surface bounding the region of
elastic behavior, and the full plastification surface. The initial yield surface is deter
mined from elementary engineering theory (taking into account the effect of residual
stress) and the full plastification surface from the appropriate empirical relationship
(e.g., Eq. 10.18). For a given axial force, the moment-curvature relationship of the
material considered has the general shape illustrated in Figure 10.9c and, after sub
tracting the elastic curvature, the shape of Figure l0.9d. The axial force-strain rela
tionships for a given bending moment are assumed to have similar shapes.
In Reference 10.13, the moment-plastic curvature and axial force-plastic strain re
lationships are modeled by the equations
d.4.
'+'p c
( M - M,c
)n
- =
(10.23a)
dM Mpc - M,c Mpc - M
and
(10.23b)
in which c and n are functions of PfPP and cP and nP are functions of M/Mw These
parameters are obtained through calibration with the results of analyses of the stress
distribution on partially plastified cross sections conducted on an independent fiber
element program, a program in which the cross section is subdivided into a number of
small elements. Adding the inelastic relationships of Equation 10.23 to the conven
tional elastic ones in the partially plastified regions, integrating over the length of each
region and satisfying the compatibility and boundary conditions for the regions, ex
pressions for the elongation and the end rotations of the member can be obtained. For
example:
in which
M
L2
l2(M) =
(M I + M2)2 J
M,c
(M - m)F(m) dm
M
L3
f (M)
3
=
2(M I+ M2)3 J M,,.

(M - m)2F(m) dm
where m is an integration variable for moments and
F(m) C
( )
m - M,c n
(Mc - M,c) Mpc - m
=
The incremental flexibility equation for the member of Figure l0.9a, which may be
obtained by differentiation of Equation 10.24 and the comparable equations for (}2 and
u is, symbolically
(10.25)
The four flexibility terms ae/aP, ae2JaP, au/aMi. and au/aM2 have only plastic con
tributions whereas the remainder have both elastic and plastic parts. Studies in Ref
erence 10.13 indicate that the purely plastic interaction terms have an insignificant
effect on in-plane response. Neglecting them and using the remainder of the flexibility
matrix in the flexibility to stiffness transformation process of Section 4.4.1 leads to the
element stiffness equation
[kemJldA} = {dP} (10.26)

where [kem] is a symmetric 6 X 6 elastic-plastic matrix in which the zones of distributed
plasticity have been converted into equivalent plastic hinges. Equation 10.26 is a for
mula for first-order inelastic analysis. For second-order inelastic analysis it may be
modified to
[kem + kg]{dA} = {dP} (10.27)

in which [kg] is a conventional geometric stiffness matrix.
This approach requires a precalibration process to determine the cross section pa
rameters of Equations 10.23, modification of Equation 10.24 and its companion equa
tions for the case of single curvature, and numerical integration for the determination
of the terms of [keml Ne vertheless, it can be computationally more efficient and pos
sess an accuracy comparable to that of a fiber element program ( see Ref. 10.13 ) .
10.5.2 Multiple Yield Surfaces
There is a large body of research, both in plasticity theory and structural analysis,
directed toward modeling the behavior of a material as its stress-strain relationships
change under load. And much of it is based on theories or experimental evidence of
how the yield surface changes. The representation of strain hardening by isotropically
or kinematically hardening surfaces in Figure 10.2 is an illustration of two such theories.
In the following we summarize a proposal for adapting one theory of plasticity, the
Mroz hardening theory ( Ref. 10.14), to inelastic structural analysis by combining the
plastic hinge concept with the notion of multiple stress resultant yield surfaces.
10.5 The Yield Surface Concept-A Brief Survey of Further Applications 293
p p
- - - - - -:_;-___.,__..,,....._
P P
ld
(a ) (b)
(J
Beam
Force
I if'- -
HEiL -
I .--_..---
_..

Rigid-plastic hardening hinge
t o()o
Exler Elastic beam
: l
node
J
lnternal n
o o Deformation
(d)
(c)
Figure 10.10 Generalized plastic hinge model.
The Mroz theory is more complex than the theories pictured in Figure 10.2 in that
it postulates a series of yield surfaces with constant, but direction dependent, hardening
between each pair. By extrapolation to a differential segment of a framework element
subjected to axial force and bending it would imply the differential hardening behavior
illustrated in Figure 10.lOb. In Reference 10.15, Powell and Chen apply this idea to
the analysis of a framework element and, in doing so, they assume that yielding and
hardening, which may actually be present along the length of the member, can be
treated as concentrated zero-length hinge subelements at the element ends (Fig. 10.lOc ).
The resultant action of the element is thus a combination of the response of an elastic
beam and inelastic hinges, as indicated in Figure 10.lOd. The analysis involves inclusion
of the plastic flexibility of the hinges and, ultimately, inversion of flexibility equations
to obtain a stiffness relationship for the full element. Assumptions regarding hinge
behavior or calibration against typical patterns of distributed plasticity are required
for application of the method. Examples of both are presented in Reference 10.15.
10.5.3 Reinforced Concrete Members
Generally, the material ductility upon which the validity of the inelastic analysis
method of Section 10.2 depends cannot be relied upon in conventional reinforced
concrete structures. But the concept of a stress resultant failure surface, which is similar
to a yield surface in appearance and in some of the ways it may be used in analysis,
does have a place in the design of reinforced concrete structures. An approach com
monly used in the development of a failure surface may be summarized by consider
ation of a doubly reinforced rectangular section subjected to axial force and bending
about a principal axis (Figure 10.lla). Basic assumptions, which are implicit in Figure
(a) (b)
Compression
controlled limit
"'-- Mn
Tension
controlled limit
(c)
Figure 10.11 Reinforced concrete section failure surface.
10.llb, are: 1) plane sections remain plane in bending; 2) tensile resistance of the
concrete is ignored; 3) the ultimate compressive resistance of the concrete can be
represented by an equivalent uniform stress block; and 4) the limiting resistance of the
reinforcement is its yield strength. These assumptions, plus the conditions of axial force
and flexural equilibrium, are sufficient to calculate the resistance of a section of given
dimensions and material properties; that is, to generate the surface of Figure 10.llc.
In the region ab the limit of resistance is governed by compression in the concrete and
in region be by tension in the lower reinforcement. Point b is the balanced failure point,
the condition under which crushing of the concrete and yielding of the lower row of
steel would occur simultaneously.
The above approach and the physical consequences of the different modes of failure
initiation are treated in detail in texts on reinforced concrete structures such as Ref
erence 10.16. A three dimensional surface of the same type, but for the design of
biaxially loaded steel-concrete composite columns, is developed in Reference 10.17.
Included is a generalized empirical equation for the representation of the surface.
10.6 PROBLEMS
10.1 (a) Verify the development of Equations 10.14 and 10.16. (b) For element ab of
Example 10.4, develop the [km] matrix for the case of a plastic hinge at end a using a
symbolic mathematics program.
10.2 Equation 10.18 is a polynomial approximation of the yield surface of a W12 X 30
section of elastic-perfectly plastic steel subjected to axial force and bending. Using
elementary beam theory, write a program to generate the surface and compare the
results with Equation 10.18.
10.6 Problems 295
10.3 Determine the elastic and inelastic critical loads for the pin jointed plane frame
shown.
c
A= 9.13 in2
I= 110 in4
E = 29,000 ksi
<1y = 50 ksi
Problem 10.3
10.4 - r-the columns

Determine the major axis elastic and inelastic critical foaas fo
shown. All loads are applied concentrically. uy = 50 ksi, E = 29,000 ksi.
p p
J.
b b
t= t
W12x26 2P W12x26
8' W12x65
1
.t
18' W12x65
W12x26
L a
0
0
(a) (b )
Problem 10.4
10.S Determine the major axis elastic and inelastic critical loads for the continuous
columns shown. All loads are applied concentrically. uy = 50 ksi, E = 29,000 ksi.
p p
(a) (b)
Problem 10.S
10.6 Calculate the first order inelastic response of the laterally supported beams
shown. u
y
= 50 ksi, E = 29,000 ksi. For case (d) determine an approximate maximum
strength-to-weight ratio by analyzing the system for several values of x.
0.01
r
W16x67
p klft
0 0
l-2@8' 16'_.)__3@8' 24'___J_2@8' 16'_.j

= = =
(a)
l l cl ;;g;; l l W30l lxl261l l !

l: W30x q klft
W30x13
q klft
1 W30x261
0 0 0
L4o 5j. 60.__J L 40' 40'----J

x x
(c) (d)
Problem 10.6
Pk/ft I I I 'l
I I I I.
J)W36x131
---1-
Ol
w16x61 , \j
p
I 10. l:36 x25I

.,..,4J.,.,
.
8' 8' ---'--16' ______j
W@ 0 0
'1
f..-
L 40 4
(a) (b)
0.2Pt t0.2P
[ l l l l u l 0.01 l kl l J l l :k2-
W16x67
- W16x67
p klft p
i
16' 3@8 24' 16'____.j L 16'

0 0
---+l+--- = ---'-'-- = ---W--
(c) (d)
Problem 10.7
10.8 For the frame and loading conditions of Problem 9.2, calculate the inelastic
critical loads. u
y
= 250 MPa.
10.9 Calculate the inelastic critical loads for the frame and loading conditions of
Problem 9.3. u
y
= 250 MPa.
10.10 (a) Compute the inelastic critical load for the frame of Problem 9.4. Uy= 36 ksi.
(b) Calculate the response of the frame using first- and second-order inelastic analyses.
Compare the results. Beams Z = 244 in3, columns Z = 54.9 in3, u
y
= 36 ksi.
10.11 Use first- and second-order inelastic methods to analyze the frame of Problem
9.5. Compare the results. Rafters Z = 1.28 x 106 mm3, columns Z = 4.24 X 105 mm3,
u = 250 MPa.
y
10.6 Problems 297
10.12 Analyze the frame shown by first- and second-order elastic and first- and
second-order inelastic methods. <T
y
60 ksi for columns and 36 ksi for all other mem
=
bers, E = 29,000 ksi. ( Account for elastic shear deformation) .
3.l--12____..).3
Problem 10.U
10.13 The rigid frames shown are laterally braced. <T

y
= 36 ksi, E = 29,000. Analyze
each frame by first and second order inelastic analysis methods and compare the
results.
+5 pr
r r r r rp
lJl\ I i;:1 11r
\
0 0 0
l \
l tts 6x6x HSS14x14x 1
.1
ttss 4x14x HSS lOx lOx
I---- 6@ 8' = 48' 1------ 6@ 8' 48' = .
(a) (b)
Problem 10.13
10.14 The beams, columns, and bracing members of tne rigidly jointed frame shown
have the sizes indicated. <T
y
= 36 ksi, E = 29,000 ksi. Analyze the frame by first- and
second-order inelastic methods.
W27x 194
l
Wl8x86
0 0 0
30' 20+-4o_j
Problem 10.4
298 Chapter 10 Material NoDiinear Analysis
10.15 Study the deflection stability of the frame shown under the following loading
pattern: 1) a vertical load (P = 30 kips) applied at c and retained on the structure; and
2) a horizontal load, H 15
= kips, applied repeatedly at b (H = 0, 15, 0, 15, etc).
uy = 50 ksi., E = 29,000 ksi.
H b c d
I 3.5'
W4x13
6'
el
l
0
,.J
a
0
f--- 4
Problem 10.15
10.16 The rigid frame shown is laterally braced. uy = 36 ksi, E = 29,000 ksi. Analyze
it by a second-order inelastic method. Attempt to improve the overall strength-to
weight ratio of the structure by considering and analyzing it for several different sizes
of web members.
HSS6x6x !
8'
HSS14x14x i HSS4x4x i
8' 8'
Problem 10.16
10.17 The arch shown is laterally braced. uy = 50 ksi, E = 29,000 ksi. Compare its
resistance to two loading conditions: (a) the three loads shown; and (b) a load at point
b only.
Problem 10.17
10.18 The frame shown is laterally braced, u

y
= 50 ksi, E = 29,000 ksi. Analyze it by
a second-order inelastic method. Attempt to improve the overall strength-to-weight
References 299
ratio by considering and analyzing it for several different sets of member sizes (vary
member sizes individually if desired).
2W kips .--..-....-.---r--:=;-;';:'."-.--.---,,-,
......_
. _.._.....,._...;.;.
.. ..==_..__.,__.__........,.
4 W kips 1--..---.....-.-rr-=-.--.---,,-;
.,_.._....Ir'"-';;.;.;...;...
... ..,__..__.
___._,
.
0
20' ___._
., __
Problem 10.18
10.19 Analyze the system of Problem 9.17 by first- and second-order inelastic
methods. Study the effect of varying girder size. Girder <T
y
= 50 ksi, cables <T
y
= 150 ksi.
10.20 Analyze the system of Problem 9.18 by first- and second-order inelastic
methods. Study the effect of varying cable and girder sizes. Girder <T
y
= 50 ksi, cables
<T
y
= 150 ksi.
REFERENCES
10.1 A. Mendelson, Plasticity: Theory and Application, Macmillan, New York, 1968.
10.2 M. Zyczkowski, Combined Loadings in the Theory of Plasticity, PWN-Polish
Scientific Publishers (ARS Polona-Ruch distributors), Warsaw, 1981.
1996.
10.4 Plastic Design in Steel, A Guide and Commentary, ASCE Manual No. 41, ASCE,
New York, 1971.
10.5 Manual of Steel Construction, Allowable Stress Design, 9th edition, American
10.6 Manual of Steel Construction, Load and Resistance Factor Design, 2nd edition,
American Institute of Steel Construction, Chicago, 1994.
10.7 L. Porter and G. H. Powell, "Static and Dynamic Analysis of Inelastic Framed
Structures," Report No. EERC 71-3, Earthquake Engineering Research Center,
f
Univ. o California, Berkeley, 1971.
10.8 P. G. Hodge, Plastic Analysis of Structures, McGraw-Hill, New York, 1959.
10.9 W.-F. Chen and T. Atsuta, Theory of Beam-Columns, Vol. 2: Space Behavior
and Design, McGraw-Hill, New York, 1977.
10.10 J. G. Orbison, W. McGuire, and J. F. Abel, "Yield Surface Applications in
Nonlinear Steel Frame Analysis" Computer Methods in Applied Mechanics and
Engineering, Vol. 33, Nos. 1-3, 1982.
10.11 T. V. Galambos, editor, John Wiley and Sons, Guide to Stability Design Criteria
for Metal Structures, 5th edition, New York, 1998.
10.12 P. Arnold, P. F. Adams, and L.-W. Lu, "Strength and Behavior of an Inelastic
Hybrid Frame," JI. of the Struct. Div., ASCE, Vol. 94, No. STl, 1968.
10.13 M. R. Attalla, G. G. Deierlein, and W. McGuire, "Spread of Plasticity: A Quasi
Plastic-Hinge Approach," JI. of Struct. Engr., ASCE, Vol. 120, No. 8, 1994.
10.14 Z. Mroz, Mathematical Models of Inelastic Material Behavior, Solid Mechanics

Division, University of Waterloo, Waterloo, Ontario, 1973.
10.15 G. H. Powell and P. F. S. Chen, "3D Beam Column Element with Generalized
Plastic Hinges," JI. of Engr. Mech., ASCE, Vol. 112, No. 7, 1986.
10.16 J. G. MacGregor, Reinforced Concrete, Mechanics and Design, Prentice Hall,
Saddle River, New Jersey, 1997.
10.17 P. R. Munoz and C-T T. Hsu, "Biaxially Loaded Concrete Encased Composite
Columns: Design Equation," JI. of Struct. Eng., ASCE, Vol. 123, No. 12, Dec.,
1997, Reston, Virginia.
Chapter 11
Solution of Linear Algebraic

Equations
We have been concerned mainly with the physical aspects of the structural analysis
problem-the formation of equations that satisfy the laws of equilibrium, compatibility
and the constitutive relations of the material, and the interpretation of the solution of
such equations. We now shift our attention to the mathematics of equation solving.
In Chapters 2-5, first-order elastic analysis included solving a system of linear si
multaneous equations for the unknown displacement components. In Chapter 8, it was
shown that more realistic modeling of structural systems results in nonlinear equilib
rium equations. As a practical means for solving these algebraic and transcendental
equations, the analysis is reduced to a series of linear increments in which the displace
ment components are determined by employing the same equation solution strategies
as those used in first-order elastic analysis.

Although the subject of equation solving may be more in the realm of numerical
methods or computer science than in structures, the algorithm employed can have a
dramatic impact on many aspects of analysis, ranging from overall efficiency to accu
racy and reliability of the results. For the larger problems common in practice, it has
been estimated that 20% to 50% of the computer execution time may be devoted to
solving systems of equations. Where a computer is limited to representing all values
by a fixed number of significant figures, numerical results can be susceptible to accu
mulated error. Since primary responsibility for the entire analysis generally belongs to
the structural engineer, more than a superficial acquaintance with the different possi
bilities for solving equations is required.
We begin this chapter with a review of the basic problem. Direct and iterative meth
ods for solving large systems of equations are described and illustrated. We then pres
ent ways to improve solution efficiency by taking advantage of peculiar characteristics
of most stiffness coefficient matrices such as symmetry, bandedness, and sparseness.
Finally, we consider the questions of accuracy and error, and what may be done about
them. We shall see that there are ways to detect sys'tems of equations that may be
particularly troublesome numerically, to estimate whether the solution errors may be
intolerable and, if so, to avoid or at least ameliorate the situation.
301
302 Chapter 11 Solution of Linear Algebraic Equations
11.1 THE BASIC CHOICE-DIRECT INVERSION VERSUS

ELIMINATION OR ITERATION
The problem in numerical analysis that we have addressed many times can be stated
as
[A]{x} =
{b} (11.1)
where the elements of [A] and {b} are known and those of {x} unknown. We limit
consideration to cases in which the coefficient matrix [A], while not necessarily sym
metrical, is square and nonsingular. The object is to solve for {x}. The solution may be
written symbolically as
(11.2)
Interpreting [Ar1 literally as the inverse of [A], it can be evaluated through Cramer's
rule as the general equation
[adj A]
[Ar' (11.3)
IAI
=
In Equation 11.3, [adj A] is the adjoint matrix of [A]; that is, the matrix of transposed
cofactors,' IAI is the determinant of [A]. The derivation of Equation 11.3 can be
and
found in most texts on matrix algebra and many on numerical analysis (see Ref. 11.1).
Direct inversion of [A] in this fashion is straightforward. Unfortunately, for almost any
matrix of practical interest, it is an inefficient way to solve Equation 11.1 since it
requires the evaluation of a number of determinants of high order in calculating [Ar'.
Many more efficient schemes for determining {x} without constructing the inverse of
[A] explicitly have been devised. Generally, the most attractive are divided into two
groups, direct elimination methods and iterative methods.
11.2 DIRECT ELIMINATION METHODS
Elimination methods employ the concept of factoring the matrix [A] into products of
triangular and-sometimes-diagonal matrices. The decomposition theorem (Ref.
11.1) states that a general square matrix [A] is expressible in the form:
[A] =
[L][D][U] (11.4)
where [L] and [U] are unit lower and upper triangular matrices, and [D] is a diagonal
matrix.
Variants of this basic decomposition equation can be obtained by associating [D]
with either [L] or [U]. Further, limitless possibilities are opened up by considering [D]
as the product of two diagonal matrices and associating one of this diagonal pair with
[L] and the other with [U]. Thus [A] may be factored into lower and upper triangular
matrices:
[A] =
[L][U] (11.5)
where in this formulation, [L] and [U] are not necessarily the same as the [L] and [U]
of Equation 11.4. More importantly, they are not unique. Instead, they depend on the
i
'The cofactor of a matrix element a;i is the signed determina'lt defined as A;i = ( -1y+ M;i, where M;i, the
minor of a;i, is the determinant of the matrix formed by deleting the ith row and the jth column of [A].
way in which the [D] matrix of Equation 11.4 is distributed between the lower and
upper triangular matrices of Equation 11.5. For a 3 x 3 matrix, the expanded form of
Equation 11.5 is
(11.6)
In Equation 11.5 either [L], [U], or neither may be a unit triangular matrix. Usually
the diagonals of both are not unity.
Differences among the many elimination methods found in the literature are simply
differences in the way in which [A] is factored into [L] and [U]. Before showing how
[L] and [U] are formed under two different procedures and describing the other fea
tures of these equation-solving techniques, we should point out the general advantage
of matrix decomposition.
Assume that we have a square, nonsingular, possibly nonsymmetrical [A] matrix,
and its corresponding [L] and [U] matrices have been generated by some proper de
composition scheme. Equation 11.1 may be rewritten:
[A]{x} = [L][U]{x} = {b} (11.7)
For convenience, let us introduce an unknown auxiliary vector {y} which is defined by
[U]{x} = {y} (11.8)
From Equations 11.7 and 11.8, we have
[L]{y} = {b} (11.9)
Since [L] is a known lower triangular matrix and {b} is given, we may determine {y}
element by element starting from the top equation in a simple process of forward
substitution
i-1
b; - 2: l;jyj
j=l
Y; = for i = 1, . . . , n (11.9a)
I;;
With [U] a known upper triangular matrix and {y] now determined, we return to
Equation 11.8 and solve for {x} element by element starting from the bottom equation
in a process of back substitution
j=i+l
X; = for i = n, ..., 1 (11.8a)
U;;
We note that the whole process of solving for the unknown {x} has been completed
without calculating the inverse of [A]. Instead, Equation 11.1 has been solved with two
basic steps inherent to most elimination methods: (1) decomposition of the coefficient
matrix, typically done by a process of forward or backward elimination; and (2) forward
and/or backward substitution. These steps should be clear in the descriptions of the
most popular Gauss and Cholesky methods that follow. As for the relative merits of
decomposition and substitution versus explicit inversion, studies in which the required
arithmetical operations have been counted and compared attest to the general supe-
2
riority of the elimination method for solving systems of practical interest (see Ref.
11.2).
11.2.1 Gauss Elimination
Basic Gauss elimination commences with a forward elimination process. Given a set
of equations such as
ll11Xi ll12X2 ll13X3 b1

+ + =
ll21X1 ll22X2 ll23X3 b2

+ + = (11.10)
ll31X1 ll32X2 ll33X3 b3

+ + =
eliminate the first unknown from the second and third equations by multiplying the first
equation by
first equation by
a21/a11a31/a11
and subtracting it from the second and then multiplying the original
and subtracting it from the third, yielding
( - ll--
ll22
21ll12)X2 +
(ll23 a21ll13)x3 b2 - a21 bi
_
ll11 ll11 ll11 (11.lla)

(ll32 - --ll31lli2)X2 +
(ll33 ll31ll13)x3 b3 - ll3i b1
_
ll11 ll11 ll11

Operating similarly on these two equations,
with the unknown x3 x2 can be eliminated, leaving one equation
where
The completed elimination-or reduction-process has resulted in converting the

original system of equations to the form
U1iXi Ui2X2 + + U13X3 = Yi

U22X2 + U23X3 = Y2 (11.12)
U33X3 = Y3
where u1i ai1> u22 (a22 - a2iai21a11), Yi
= = = bl> Y2 =(b2 - a21b1/a11), etc. Since we
have performed only a series of algebraic manipulations, the solution to the system
shown in Equation 11.12 is identical to that of Equation 11.10.
This solution is found by solving the third equation of the system shown in Equation
11.12-or equivalently, Equation 11.llb-for its only unknown
unknowns x2 Xi and
x 3.
can now be obtained, in that order, by backward substitution
The remaining
2The total number of floating point operations, flops, is o ten used to compare the computational efficiency
J
of different solution algorithms. If n is the number of equations to be solved, an approximate number of
flops required for "the procedures presented in this section are as follows: Cramer's Rule, n!; Gauss elimi
nation, n3/3; Cholesky decomposition, n3/6; and banded Cholesky decomposition, n HBW2, where the half
bandwidth HBW is defined in Section 11.4.
working from the bottom up in Equation 11.12. The same procedure may be applied
to systems of any size, provided the coefficient of the leading unknown at any stage of
the elimination does not equal zero, a case we will consider later in this section.
Comparing Equations 11.8 and 11.12, we see that they are identical; the former is
just a matrix statement of the latter. It is clear, therefore, that during the Gauss elim
ination process the coefficient matrix, [A], has been factored into upper and lower
triangular matrices and, further, the lower triangular matrix has been inverted and
postmultiplied by the right-hand side vector, {b}.
Gauss step-by-step decomposition of [A] into the product [L][U] can be represented
algorithmically in several ways. Where [A] is typically the stiffness matrix in the struc
tural analysis, one convenient scheme for performing the decomposition of this sym
metric matrix is the following:
l;i
U = - for j = 1, ... , i - 1
JI Iii
U;; = 1 (11.13)
i-1
for j = i, ... , n
This scheme is applied and verified in Example 11.1. Example 11.2 illustrates the matrix
interpretation of Gauss elimination through Equations 11.7-11.9.
Given
rlO 30 20 10 ]
30 178 201 -36
[A]
=
20 201 485 21
10 -36 21 350
factor into [L][U] using Equation 11.13 and verify results.
Column 1: u11 1
=
f11 G11 100

=
=
f21 G12 30
=
=
/31 au 20
= =
/41 G14 10
= =
Column 2: u12 12/111 30/100 3/10

= = =
U22 1
=
f22 G22 - U12 X f21 178 - (tli) X 30 169

= = =
/32 G23 - U12 X [31 201 - (tli) X 20 195

= = =
/42 G24 - U12 X /41 -36 - (tli) X 10 -39

= =
=
Column 3: u13 13/l,, 20/100 1/5

= = =
U23 /32/!22 195/169 15/13

= = =
UD = 1
/33 G33 - U13 X /31 - U23 X /32 485 <*) X 20 (ft) X 195 256
= =
=
- -
[43 G34 - Uu X f41 - U23 X /42 21 - <*) X 10 (ft) X (-39) 64

= =
= -
Column 4: U14 = /41/111 = 10/100 = 1/10
U 24 = /42/122 = -39/169 = -3/13
U34 = /43/133 = 64/256 = 1/4

U44 = 1
r lr 1r 1
/44 = a44 - U14 X /41 - U24 X /42 - U34 X /43
= 350 - eta) x 10 - c-f:i) x c -39) - m x 64 = 324
Check: lOO 0 0 o l 3/10 1/5 1110 lOO 30 20 10

30 169 0 0 0 1 15/13 -3/13 30 178 201 -36
=
20 195 256 0 0 0 1 1/4 20 201 485 21

10 -39 64 324 0 0 0 1 10 -36 21 350
(L](U] = (A) Q.E.D.
Given
lOOx1 + 30x2 + 20x3 + l0x4 = 490
30x1 + 178x2 + 20lx3 - 36x4 = -347
20x1 + 20lx2 + 485x3 + 2lx4 = -1112
lOx1 - 36x2 + 2lx3 + 350x4 = 651
solve by Gauss elimination and verify results.
r
The coefficient matrix is identical to (A) of Example 11.1-use results of that example. From
Equation 11.9, (L](y) = {b}, solve for (y) by forward substitution
1{:} {-::}
lOO 0 0
30 169 0
=
20 195 256 y3
0 -1112
10 -39 64 324 Y4 651
Yi = 490/100 = 49/10
. Y2 = ( -347 - 30 x ( Tt ) ) /169 -38/13 =
Y3 = ( -1112 - 20 x c*) - 195 x c-m11256 = -512

Y4 ( 651 - 10 x (Tt) - (-39) X (-) - 64 (-))/324 2
l{ } }
= X =
From Equation 11.8, (UJ(x) = (y), solve for (x) by backward substitution
[ T {
115 lflQ X1 49/lQ
15/13 -3/13 X2 -38/13
=
1 l/4 X3 -5/2
0 1 X4 2
X4 = 2
X3 = - - () X 2 -3 =
x2 = - - cm x c -3) - c-f:i) x 2 = 1
X1 = *- cfo) x 1 - m x c-3) - Cta) x 2 = 5
Check:
rlO30
30
178
201
20
201
485
-36
10
21
]{ } { }
5
1
-3
490
-347
-111_2
20
=
10 -36 21 350 2 651
[A]{x} =
{b} Q.E.D.
It was mentioned earlier that if at any stage in the elimination process the coefficient
of the first equation for the leading unknown, often termed the pivot element, becomes
zero, the basic process will not work. When checks are not made to avoid the resulting
division by zero that will occur in Equations 11.11-11.13, the above process is called
"naive" Gauss elimination. This difficulty can be avoided by employing partial pivot
ing, that is, by interchanging the offending equation with another having a nonzero
coefficient for the same unknown. This will always be possible if the submatrix under
consideration is nonsingular. Note that partial pivoting will destroy any existing sym
metry of a coefficient matrix, thereby making it impossible to use algorithms based on
symmetry, such as the one expressed by Equation 11.13. To preserve symmetry, this
concept may be extended to complete pivoting where both rows and columns are in
terchanged. As discussed later in this chapter, pivoting strategies are also often em
ployed to improve solution accuracy.
In Chapter 13 we will present static condensation and substructuring as one way of
handling large systems of equations. The role of that technique in equation solving
will not be discussed here other than to define matrix condensation as a process of
contracting the size of a system of equations by eliminating certain degrees of freedom.
The relationship between this definition and Gauss elimination should be fairly obvi
ous. The tie between static condensation, substructuring, and Gauss elimination is
developed in Reference 11.3 and numerous papers in the literature.
11.2.2 Cholesky Method
Although there exists Gauss decomposition schemes that can be applied to any square
coefficient matrix, the Cholesky method applies only to cases where [A] is symmetrical
and positive definite.3 Its distinguishing feature is that [A] can be factored efficiently
into upper and lower triangular matrices that are the transpose of each other; thus
[A] =
[L][L]T (11.14)
Once this decomposition has been performed, equation solution proceeds by forward
and backward substitution following Equations 11.7-11.9, with [U] in those equations
becoming [L]T.
3For all nonzero vectors {z}, [A] is a positive definite matrix as long as the inner product {z}T[A](z} is always
greater than zero. If this inner product is always greater than or equal to zero, [A] is positive semi-definite
and if it can be less than zero, [A] is indefinite. Alternatively, the definiteness of a matrix can be defined in
terms of its eigenvalues (see Section 12.8). If all the eigenvalues of [A] are positive, it is positive definite.
(A] is positive semi-definite if its eigenvalues are greater than or equal to zero. And, (A] is indefinite when
its eigenvalues are either positive, zero, or negative.
A convenient recurrence relation for Cholesky decomposition is as follows. Each

successive column of [L] is calculated as
i-1
tu = a;; - 2: t7k
k=I
i-l (11.15)
aji 2: tjktik
k=I
-
for j = i + 1, . . . , n
l;;
It should be noted that if the above evaluation of!;; involves taking the square root of
a negative number, this implies that [A] is indefinite (footnote 3). With respect to
structural analysis in which [A] = [KJJ], this indicates that the system is unstable (see
Section 8.1.4). An additional benefit of working with positive definite symmetric ma
trices is that a zero will never exist in the denominator of Equation 11.15; hence this
algorithm can be employed without the use of pivoting.
Equation 11.15 may be modified to a slightly more efficient root-free version of
Cholesky decomposition. In this scheme the coefficient matrix is factored into the form
of Equation 11.4 with [U] being [L]T; thus
[A] =
[L][D][L]T (11.16)
Starting with i = 1, each column of [D] and [L] can be calculated as

i-1
d;; = au - 2: dkkt7k
k=I
Lu = 1 (11.17)
i-1
aji 2: dkktjktik
k=I
-
lji = ------ for j = i + 1, . . . , n

d;;
With [A] factored into the above triple product, a solution can still be obtained fol
lowing Equations 11.7-11.9 with [U] in those equations now becoming [D][L]T. If this
method is employed with a pivoting strategy, it can also be used to solve symmetric
indefinite systems. Beyond being a more general solution method, root-free Cholesky
decomposition can provide insight for solving eigenproblems associated with critical
load analysis (see Section 12.8).
Equations 11.15 and 11.17 are illustrated and verified in Example 11.3.
[ ]
EXAMPLE 11.3
Given
100 30 20 10
30 178 201 -36
(A]
=
20 201 485 21
10 -36 21 350
1. Factor into (L][L]T using Equation 11.15 and verify results.

2. Factor into [L][D][Lfr using Equation 11.17 and verify results.
1. Column 1: /11 = =VlOo =10

121 =ll21ll11 =30/10 =3
/31 = ll31//11 =20/10 =2
/41 =ll41//11 =10/10 =1
Column 2: /22 =Ya22 - /1 =Vl78 - 32 =13

/32 =(a32 - (/31 X 121)]//22 =(201 - (2 x 3)]/13 =15
/42 =(a42 - (/41 X 121)]//22 =(-36 - (1 X 3)]/13 =-3
Column 3: /33 =Ya33 - /1 - /2 = v485 - 22 - 152 = 16

/43 =(a43 - (/41 X /31) - (/42 X /32))//33 =(21 - (1 x 2) - (-3 x 15)]/16 =4
Column4: /44 = Va44 -11 - /2 - /3 =V350 - 12 - (-3)2 - 42 = 18
[ ][ -] [ ]
Check:
1 1 1 30 20 10
1 1 1
178 201 -36
2 15 16 0 0 0 16 4 20 201 485 21
1 -3 4 18 0 0 0 18 10 -36 21 350
(L](LF =(A] Q.E.D.
2. Column 1: d11 = a11 = 100

111 = 1
/21 = ll21fd1 I =30/100 = 3/lQ
/31 =ll31ld11 = 20/100 = 1/5
/41 =ll41/d1 I =lQ/100 = 1/lQ
Column 2: d22 = a22 - d11 x n1 = 178 - 100 x (fo)2 = 169
122 = 1
l32 = (a32 - d11 x l31 l2i]ld22 = (201 - 100 x m x (fo)]/169 =15/13
x
l42 =(a42 - d11 X l41 X l2i]ld22 =(-36 - 100 X (ta) X (fo)]/169 =-3/13
2
Column 3: d33 = a33 - du x /1 - d22 x /2 =485 - 100 x (ff - 169 x (H} = 256
/33 =1
/43 = (a43 - du X l41 X /31 - d22 X /42 X IJZ]/d33
=(21 - 100 x Ck> x G> - 169 x <-fi> x cm11256 =114
Column 4: d44 =a44 - d11 X /1 - d22 X /2 - d33 X /3
= 350 - 100 x cia>2 - 169 x c-n>2 - 256 x m2 = 324
[ ] [ [] []
/44 = 1
]
Check:
0 0 0 100 0 0 0 1 fa k ta 100 30 20 10
IO 1 0 0 0 169 0 o o 1 H -fi - 30 178 201 -36
-
k H 1 o o 0 256 0 0 0 1 20 201 485 21
to -fi 1 0 0 0 324 0 0 0 1 10 -36 21 350
(L](D](L]T =(A] Q.E.D.
11.3 ITERATIVE METHODS
For larger systems, typically in excess of 1000 equations, the elimination methods can
sometimes be inadequate for reasons that include available computer memory, exe
cution time, and accumulated round-off error. To circumvent these problems, iterative
methods can often be used as an alternative solution scheme. The primary reason these
methods are effective is that a solution can be obtained to within some prespecified
error tolerance.
11.3.1 Gauss-Seidel and Jacobi Iterations
One of the most common iterative schemes is the Gauss-Seidel method. The method
begins with an initial estimate or guess at the solution vector, /x}. Then each of the
system's equations is repeatedly used to solve for or update one of the unknowns in
terms of previous values of the others. For the system provided in Equation 11.10, this
yields
1 1
x'{' = (b1 - a12x2- - a13x)- )/a11
Xz = (b2 - a21X7' - a23X)-I )/a22 (11.18)
x) = (b3 - a31x7' - a32x2')/a33

where m and m - 1 are the present and previous iteration numbers and, for example,
1
xzi- and x2 are the previous and new estimates of the second element of the solution
vector. Note that the original equations must be ordered so that the diagonal of [A]
contains only nonzero values, a situation that will always be possible to attain in the
analysis of stable structures.
In general, the Gauss-Seidel iterative form for a system of n-equations can be stated
as
(11.19)
For convenience, a vector of zeros is often used as an initial guess at the solution. It
should be noted, however, that in a nonlinear incremental analysis the solution vector
from the previous load step can often provide an excellent initial guess for the current
system of equations. Iterations are repeated until the solution converges "close
enough" to the true solution. For an acceptable percent error tolerance , a recom
mended convergence criterion is
(11.20)
with
(11.20a)
and
= 0.5 x 102-q % (11.20b)
where q is the desired number of accurate significant figures (Ref. 11.4).

Obviously the number of iterations required and the possibility of convergence can
depend on several factors, including the initial guess and the required convergence
tolerance. For systems where [A] is positive definite, however, the iteration process
11.3 Iterative Methods 311
will always converge, barring any significant truncation error. To improve the rate of
convergence, relaxation schemes can be employed. In this approach, a newly calculated
value is modified by taking the weighted average of its current and previous values:
xr = {3x'(' + (1 - {3)xr-l (11.21)
where {3 is the relaxation or weighting factor. When {3 is defined between 1 and 2, it

is assumed that the current value x'(' is slowly approaching the exact solution and hence
if more weight is placed on it the solution will converge faster. This modification is
called overrelaxation. Setting {3 to a value between 0 and 1 is called underrelaxation,
an adjustment that is often employed to damp out oscillations and attempt to make a
nonconvergent system converge. Although the proper choice of {3 is problem specific,
overrelaxation schemes are generally applied in solving structural analysis problems.
Instead of immediately using the most recently calculated elements of the solution
vector, Jacobi iteration calculates all of the new values on the basis of all of previous
values
(11.22)
Although the Gauss-Seidel method typically converges faster, Jacobi iteration is worth
mentioning since it is ideally suited for computer implementations that use parallel
processing.
The Gauss-Seidel method with and without overrelaxation is illustrated in Example
11.4.
EXAMPLE 11.4
Using an acceptable error tolerance of= 0.1 %, solve the system of equations given in Example
11.2 by the Gauss-Seidel method for relaxation factors of:
1. f3 = 1.0
2. f3 = 1.2
In each case, combining equations 11.19 and 11.21 yields
1 1
x'i' = /3[ b1 - (a12x2- + a1 3x3-i + a14X'.j- )]/a11 + (1 - f3)x'i'-1
1 1 1
x2 = f3[b2 - (a21X'i' + a23X)- + a24x4- )]/a22 + (1 /3)x2- -
1
x) = f3(b3 - (a31X'i' + a32X2' + a34X4-1)]/a33 + (1 - f3)x3-
1
{}
x4 = f3(b4 - (a41X'i' + a42x2' + a43x3)]/a44 + (1 - /3)x4-
1. u,;ng '" ;n;,;,1 ""'" of {x") nd p 1.0
The first iteration is
xl = 1.0[490 - (30 x 0 + 20 X 0 + 10 x 0)]/100 + (1 - 1) X 0 = 4.900
xi= 1.0[-347 - (30 x 4.9 + 201X0 + (-36) x 0)]/178 + (1 - 1) x O = -2.775
xi = 1.0[-1112 - (20 X 4.9 + 201 x (-2.775) + 21 X 0]/485 + (1 - 1) x O = -1.345
x! = 1.0[651 - (10 x 4.9 + (-36) x (-2.775) + 21 X (-1.345))]/350

+ (1 - 1) x 0 = 1.515
3U Chapter 11 Solution of Linear Algebraic Equations
A second iteration yields
{x2} = {-2.-1.7107391
:
.
h a maximum
wit
.
percent relalive error of e,,
1-1.111-(-2.775)1 100% 149.8%
=
-1.111 =
After 14 iterations the error tolerance C 0.1% is satisfied with
4
5.0004
_2.0.99996
993
} =
5
_31
{}
{x1 J =
2. Using the same initial guess and {3

{ 1.9999
compared to the true solution
=
1.2
{xT} =
2
The first iteration yields
5.880
-3.529
}
The second iteration would proceed with
-1.287
1.688
{x1} =
{
XT =
1.2[490 -(30 (-3.529) 20 (-1.287) 10 1.688))/100 (1 - 1.2)5.880
x + x + x +
=
6.081
Calculating the remaining elements of {x2) gives
{x2) =
{-2.-1.:5730811
. . .
with a maximum percent relalive error of Ea =
_0.709 529)1 100% 397.7%
1-0.709-(-3. =
The benefits of over-relaxation are realized by obtaining a solution in only

satisfies the given error tolerance
8 iterations that
{xs} =
{-3.1.4.900999000001}
2.0000
11.3.2 Conjugate Gradient Method
The conjugate gradient method is based on the mathematical property that any vector
(x} with n elements can be expressed in an n-dimensional space by a scaled combination
of n linearly independent or basis vectors { s; }
n
(x} = L ( }
a; s; (11.23)
i=I
Hence, the true solution to the system of n equations posed in Equation 11.1 can be
expressed by Equation 11.23 or by
n
{x} = {x0} + L ads;) (11.24)

i=I
where {x0J is an initial guess at the solution vector and the a; scale factors of Equations
11.23 and 11.24 are not necessarily the same.
With this in mind, Equation 11.24 can be estimated in a first iteration as
(11.25)
Closer approximations could be achieved in successive iterations by calculating and
including additional scaled basis vectors according to
(11.26)
This process could then be repeated until a solution is obtained that is once again
considered "close enough." It is important to note that there exists a limit to the
number of iterations required; the exact solution will be obtained in a maximum of
n iterations since all n basis vectors would be included.
Clearly, the crux of the matter is to provide a scheme for calculating each additional
basis vector {s;) and its corresponding scale factor a;. To find these entities, the con
jugate gradient method employs an optimization strategy. Assuming that [A] is a pos
itive definite symmetric matrix, the solution to Equation 11.1 will occur when the
quadratic
(11.27)
is at a minimum. Where the gradient vector or slope of this function is defined by
{G} = [A]{xj - {bj (11.28)

we are simply restating Equation 11.1 by indicating that Equation 11.27 will be at a
minimum when {G} is the null vector. It is important to note that in structural analysis
terms, Equation 11.27 represents the total potential of the system and the minimum
given by setting Equation 11.28 to the null vector results in stiffness equations of
equilibrium (see Section 8.1.4).
Instead of calculating the inverse of [A] to locate this minimum, it can be shown
that for a given positive definite quadratic, such as the one in Equation 11.27, the
unique minimum point may be defined by (Ref. 11.5)
i-1
-{s;}T{G }
_ n
{xj - {x0} +
{s;}T [A]{s;){s;} (11.29)
where {s;} are the n linearly independent A-conjugate directions.4 Comparing Equa
tions11.24 and 11.29, the scale factor a; has now been defined in terms of its corre
sponding conjugate direction {s;}
;
-{s;}T{G -1}
Ct; (11.30)
{s;}T[A]{s; J
=
"The conjugate direction vectors for an n x n coefficient matrix [A] are all orthogonal in the inner product
of (A] and are defined by
To determine the successive conjugate directions that will be used in each iteration,
we introduce a residual vector
(11.31)
Note that this residual vector is a measure of how well the solution obtained in the
previous iteration satisfies Equation 11.1. This vector will either be zero, which means
we have found the true solution, or nonzero and hence linearly independent of all of
the previously calculated conjugate directions. Where the residual vector is the nega
tive of Equation 11.28, Equation 11.30 may be written more conveniently as
{s;}T{r;-1}
a; = (11.32)
{s;}T[A]{s;}
By further employing this residual vector, it can be shown that the conjugate direction
for the ith iteration may be calculated using the recurrence equation
i {s;_i)T[A]{-1)
{S;} - {r -1) - . {s; (11.33)
{s,_ }T[A]{s,. } -d
1 -1
For the first iteration {si} would be assumed equal to {r0}. A derivation of Equation
11.33 is provided in Reference 11.2.
To minimize the number of matrix-to-vector multiplications, two simplifications are
often employed. First, an auxiliary row vector Ly;J = {s;} T[A] is retained during each
iteration and used in the denominators of Equations 11.32 and 11.33. If the coefficient
matrix [A] is symmetric, the transpose of this row vector Ly;JT = [A]{s;} may be used
to determine the residual vector for the next iteration by a simple vector addition
according to
{r;} = {b} - [A]{x;} = {b} - [A]({x;-1) + a;{s;)) = {-1) - a;[A]{s;) (11.34a)
or
(11.34b)
Iterations in the conjugate gradient method are often repeated until the magnitude
of the residual vector drops below a prescribed error tolerance according to
e<? (11.35)
with
e=V/?m (11.35a)
and ? being a quantity rather than a percentage as defined in Equation 1 l.20b. Similar
to the Ga11ss-Seidel method, the number of iterations required to converge on an
acceptable solutiori is a function of several factors, including the initial guess and the
prescribed error tolerance. In the absence of a good estimate of the solution, the null
vector is often used as a starting point. It is important to note that the condition5 of
the coefficient matrix can also have a major impact on the rate of convergence. To this
end, several preconditioning schemes have been developed to improve systems of
equations that are poorly conditioned (see Reference 11.2). One such method is pre
sented in Section 11.6.3 on error analysis. Use of the conjugate gradient method as
presented in Equations 11.26 and 11.31-11.35 is illustrated in Example 11.5.
5Briefiy stated, the condition of a system of equations can be defined as the significance of small changes in
the coefficients that often occur during the solution process. This topic will be explored in more detail in
Section 11.6.
rEXAMPLE 11.5]
Solve the system of equations given in Example 11.2 by the conjugate gradient method using
t = 0.001.
f100

;Oo
30
178 201
20
-36
10 ] { }
,
-347
490
, and an initial guess {o}

0
l
(A] {b}
201 485 21 1112 0
With = = (x0) =
_
10 -3o 21 350 651 o
First iteration:
Equation 11.31, {b}
Assume (si} =
{r0)
{r0} =
=
{ }
{b} - (A]{x0)
490
-347
_1112
=
651
Calculate auxiliary row vector Ly1J = {si}T[A]

{si}T(ro}
.
Equation 11.32, a1 =
LYiJ(si}
= 2.2235 X 10
1.090
-o.772
- 3
}
Equation 11.26, +
Equation ll.34b,
{x1}
(r1}
=
=
{x0)
(r0} - a1LY1f
ai(si} =
{ -2.473
1.448
Equation 11.35, e = \/fr1l1Tr11 = 589.86 1:: t = 0.001
Second iteration:
Equat.10n 11.33 , {s2} - {r1) - L

_
1
yiJ{r }
{s1} = { } 523.536
246.999
-1.378
LY1J{si}
--
269.710
a2 = 5.0182 X 10 - 3
3.716
}
2
{x } = {x1} + a2{Sz} = { 0.468
_2.479
2.801
with e = 355.50 1:: ? = 0.001
Third iteration:
{s }
3
= {-::t:},
-199.628
a3 = 3.9667 X 10 - 3 {x3}
, = {-rn}'
.
2.009
E = 4.009 1:: [ = 0.001
-
-
{ }
Fourth iteration:
1.8400
-3.1986
, - 1.6118,
5
1
}
{s4}
1.4636
-0.5650
a4
As expected, the true solution is obtained after

-
4
{x } =
{ _3
n =
, e = 0 <
4 iterations.
t = 0.001
11.4 SPARSENESS AND BANDEDNESS
As indicated in Section 3.3, the global stiffness matrix is rather sparsely or weakly
populated in most structural analysis problems. Again, this is because each row or
equilibrium equation for a particular degree of freedom is only influenced by degrees
of freedom associated with the often small number of elements connecting to that
degree of freedom. All other degrees of freedom for the remaining unattached ele
ments have no effect on this equilibrium equation and hence have zero coefficients in
that row.
Sparseness often results in the stiffness matrix having a banded form. This is illus
trated in Figure 1 1.1, in which crosses in the figure show the location of potentially
nonzero stiffness coefficients. The bandedness of the stiffness matrix is indicated by
the clustering of these coefficients about the main diagonal. As shown in Figure 11.lb,
we will denote HBW as the half-bandwidth of a symmetric matrix.
The opportunity to gain efficiency in solving linear algebraic equations can be re
alized if we note that all coefficients outside the bandwidth will always retain a zero
value during any of the solution procedures described in the previous sections.Hence,
the performance of these schemes can be improved by modifications that avoid the
storage and manipulation of these useless zeros. For example, Equation 11.15 may be
modified as follows:
I;;= a;; -
k=max( 1,i-11/JW)
i-1 (11.36a)
2: 1jk1ik
k=max( 1,j-11/JW)
lji = ------ for j= i + 1, ..., min(i + HBW, n )
!;;
1 - HBW:t1=12 1
1 2 3 4 5 6 7 8 9 1011 12131415161718
1 x x x x x x xx x
7 2 x x x x x x x x
3 x x x x x x x
4 x x x x x x x x x
5 x x x x x x x x
15 6 x x x x x x x
4 13 7 x x x x x x
8 x x x x x
9 x x x x
10 x x x x x x
12 11 x x x x x.
10 12.___.___ x x x x
13 .___.___ Sym. x x x x x x
14 x x x x x
15 x x x x
16 x x x
17 x x
18 x
(a) (b)
Figure 11.1 Banded [Kff) stiffness matrix (HBW = 11).
11.S Fronta_ Solvers 317
HBW+l=9-1
I1 2 3 4 9 5 6 7 8 1011 12131415161718
1 xx x x xx x X! x
13 x x x x xx xx
2
3 xx x x xx x
4 x x x xx x
5 x x x xx
6 x xx x
7 xx x xx x x x x
9
7
8 xx x x x x xx
xx x x x x x
'
10 xx x x x x
11 xx x x x
12 -- x ' x x x
13-- Sym. xx x x x x
14 xx x x x
15 xx xx
16 xx x.
17 xx
18 x
(a) (b)
Figure 11.2 Improved handedness (HBW = 8).
Equations 11.13 and 11.17 can be adjusted in a similar manner. In all cases, the half
bandwidth may he calculated as
HBW = max (maxdofe1 - mindofe1) (ll.36b)

lse/sm
where m is the total number of structural elements and maxdofei and mindofet are
the maximum and minimum global degree of freedom numbers associated with ele
ment el.
The shaded area of the matrix in Figure 11.1 covers all the stiffness coefficients of
[Kff] that need to be stored and manipulated. Therefore, it is obvious that it would be
advantageous to renumber the degrees of freedom in such a way that the bandwidth
of the system is minimized. For most structures, near optimum numbering is easily
attainable through visual inspection of the system. This can be done by noting that _the
half-bandwidth of the global stiffness matrix parallels the maximum difference in
global degree-of-freedom numbers pertaining to any element in the structure. Figure
11.2 illustrates the improvement in bandedness that can result from small changes to
the degree-of-freedom numbering shown in Figure 11.l. Automatic numbering
schemes that attempt to minimize the bandwidth are employed in most commercially
available computer programs for structural analysis. For reviews of solution procedures
that exploit bandedness and employ minimum storage schemes including skyline or
envelope storage and banded storage, the reader is referred to References 11.2 and
11.6-11.8.
11.5 FRONTAL SOLVERS
In performing direct elimination procedures, such as Gauss elimination or others, it is

important to note that we do not need to work with, and hence computationally store,
all of the stiffness equations at one time. That is, we can assemble the stiffness equa
tions on an element-by-element basis while at the same time successively eliminating
degrees of freedom.
Starting with an element having one or more degrees of freedom for which all stiff
ness contributions have been assembled, we eliminate these degrees of freedom by
expressing them as functions of the degrees of freedom common to this and another
element. We then add, in proper fashion, the remaining, reduced stiffnesses of the first
element to those of the second element. We may then eliminate any degrees of free
dom for which the stiffnesses are fully assembled in the combined system. We proceed
in this fashion, adding one element at a time-and nodal forces as they are encoun
tered-until the entire structure has been covered. We will then have developed a set
of equations that can be solved for one unknown at a time by back substitution. The
process is called a wave front solution, since it may be depicted as a wave spreading
gradually over the structure.
It is apparent that, in a frontal solution, the entire assembled global structure stiff
ness matrix is not needed. Although frontal solvers may be used effectively with
banded systems, banding is irrelevant. Instead of node numbering being important, as
in band solvers, the order in which elements are introduced becomes critical.
Before discussing features of frontal solvers, typical characteristics of a wave front
solution will be illustrated through Example 11.6. To demonstrate the irrelevance of
banding in a frontal solution, the degrees of freedom are labeled in an unusual way.
Although it is not needed in the subsequent solution and, normally, would not be
assembled, the global stiffness matrix is presented to show how a clumsy numbering
system can obscure the natural bandedness of a structure.
ttiXMPLE U.'61
Write the global stiffness matrix. Using
a wave front approach, develop equa
tions for calculating the displacements,
L = 5 m, E = 200,000 MPa,
I = 100 x 106 mm4
Degrees of freedom
Global stiffness matrix

'110 '111 '112 '113 '114 '115 '116 '111 '118
0.004 0 0 0 0 0 0 0 0.002
8 0 2 0 0 0 2 0
' ',
',, ! 9.2 x 10-1 0 2.4 x 10-3 0 0 -2.4 x 10-3 0
EI 8 0 0 0 0 2
(K]=-
L 12 0 4 0 0
Sym. 4 0 0 2
16 4 0
12 0
8.004
Wave front solution. Use primes to indicate degrees of freedom where stiffness coefficients are incomplete at stage consid
ered. (See page 321 for a step-by-step description of the stages.)
1. Member ib:
EI [ 0.004 0.002 ]{ } {o}A'o =
L 0.002 0.004 A;8 0
which reduces to
[ OJ [
1 0.004 0.002 ]{ } {O}
A10 =
or
[ 0.004 o.002 ]{ } {o}
A10 =
0.5 1 0 0.003 A;8 0 0 0.003 A;8 0
2. Add member ab:
EI [ 4 2 ]{ } {A15 = 120 x 103 }

L 2 4.003 A;8 0
reduces to
[ 4 2 ]{ } A,5 = L { 120 x 103 }

0 3.003 Ai8 EI -60 x 103
3. Add member be:
EI [ 7.003 2 ]{ } {A 8
:
= -60 x 103 }
L 2 4 A1 0
3
reduces to
4. Add member cd:
EI [ 7.4288 2 ]{ } {
A
:3
=
17.136 x 103 }
L 2 4 A11 0
reduces to
[ 7.4288 2 ]{ } A1
3
L { 17.136 x 103 }
0 3.4616 Ai1 = EI -4.6134 x 103
5. Add member de:
EI [ 7.4616 2 ]{ } {
A '
:
= -4.6134 x 103 }
L 2 4 A17 0
reduces to
[ 7.4616 2 ]{ } A11 =
L { -4.6134 x 103 }
0 3.4639 Ai7 EI 1.2366 x 1D3
]{ } { }
6. Add element ef:
EI
T
[ 11.4639
-2.4 x 10-3
-2.4 x 10-3
9.6 x 10-7 -2.4
4
X 10-3
A17
Ai2
=
l.2366 x 1D3
0
4 -2.4 x 10-3 s a;6 o
]{ } { }
Eliminating A17, this reduces to
-2.4 x 10-3 4 A17 1.2366 x 103

7
4.5755 x 10- -1.5626 x 10-3 Ai2 = 0.259
-1.5626 x 10-3 6.6043 Ai6
EI -4.3148 x 10 2
[ ]{ } { }
7. Add element fg:
6
1.4176 x 10- 8.3740 x 10-4 2.4 x 10-3 A12 -39.741
8.3740 x 10-4 14.6043 4 A16 - -4.3148 x 102
2.4 x 10-3 4 8 Ai4 0
]{ } { }
Eliminating A12 and A16, this reduces to
8.3740 x 10-4 2.4 x 10-3 A12 -39.741

14.110 2.5823 A16 = 23.044 x 103
0 3.4642
I
Ai4 63.064 x 103
8. Add member gh:
[7.4642]{A14} = {63.064 x 103)
reduces to
[7.4642]{A14} = I (63.064 x 103)
9. Collect complete set of equations:
4A10 + 2A18 = 0
4A1s + 2A18 = 120 x 103 X LIE/
7.003A18 + 2A1 -60 x 103 x LIE/

3
=
7.4288A13 + 2A11 = 17.136 x 103 x LIE/
7.4616A11 + 2A17 = -4.6134 x 103 x LIE/
11.4639A17 - 2.4 X 10-3A12 + 4A16 = 1.2366 x 103 x LIE/

6
1.4176 x 10- A12 + 8.374 x 10-4A16 + 2.4 x 10-3A14 = -39.741 x LIE!
14.110A16 + 2.5823A14 = 23.044 x 103 x LIE/
7.4642A14 = 63.064 x 103 x LIE/
10. Solve by back substitution.
A10 1.1366 x 10-3 rad

A11 0.4348 x 10-3 rad
A12 -10.597 mm
A1 0.4596 x 10-3 rad
3
{.:i} = A14 2.112 x 10-3 rad
A1s e.6366 x 10-3 rad
0.02173 x 10 3 rad
... j
-
A11 -2.199 x 10-3 rad

A1R -2.2732 x 10-3 rad
In the example, the elements are considered in the order ib, ab, be, cd, de, ef, fg,
and gh. The element stiffness equations are written, added to equations from pre' ious
stages that remain incompletely reduced, and the combination is then further reduced
to the extent permissible at the current state. Gauss elimination is employed, but other
reduction algorithms could have been used. The extent to which the current set of
equations can be reduced depends upon whether all stiffness coefficients relating to a
given degree of freedom have been accumulated, and whether we have duly entered
all loads acting at the nodal points for which we are compiling-in piecemeal fashion
equilibrium equations. For instance, in considering the initial element, ib, the first
equation is complete. ll10 may be expressed as a function 9f ll18 Once this has been
done, ll10 may be considered passive, and the equation containing it may be stored for
later use in back substitution. In element ib, the second equation is not complete since
ll18 is at an interface with the deformable elements ab and be. Its stiffness coefficients
have components yet to be accumulated from these elements. Thus, at this stage, rhis
equation can be only partially reduced by considering the effect of ll10 (the rotation
at i) upon ll18 Primes are used as temporary indicators of degrees of freedom ap
pearing in equations in which the stiffness coefficients or the load vector term is in
complete. When the second element ab is brought into consideration, ll15 appears and
may be eliminated immediately since it is in.rnal to the two-element system and the
load vector term is the proper one for node a. The elimination of ll1rs r.11st await t.he
addition of the third element, be, and the completion of the reduced load term at node
b. Later, when we encounter element ef, only one of the three variables under con
sideration, that is, ll17, may be eliminated. The other two, a12 and a16' must remain
active until all of the equations in which they appear are complete through the addition
of element fg and the load at node f.
Working through the structure in the above way, we arrive at nine equations that
can be solved, one at a time, by back substitution.
The frontal solution concept is pictured in Figure 11.3. The wave front is the region
between the two dashed lines in each sketch. The degrees of freedom in this region
are the active variables. We start with the wave covering element ib. As soon as ll10
has been eliminated (Figure ll.3a), ll18 is the only active variable until the wave
spreads over ab to include ll15 in the front (Figure ll.3b). Progress at selected later
stages is indicated in the remaining sketches. It may be noted that, each time an ad
ditional element is added, new variables become active. Some of these may be elimi
nated forthwith, for example, ll15; others, such as ll18, may remain active, that is, within
the wave front, through the addition of a number of elements.
There is, of course, much more to practical frontal solvers than can be illustrated by
a simple example. We have focused on the basics of the approach to equation solving
that may be depicted by the wave frot concept. Reference 11.9 presents a sophisti
cated procedure for solving large systems of equations.
Frontal solvers generally provide for the passage of the decomposition array through
high-speed core memory three times-once when it is developed, once during forward
elimination, and once during backward substitution. Among the advantages of frontal
solvers are those already observed: (1) only a very small portion of the entire stiffness
matrix need be considered at any one ti.me; the entire matrix is never assembled; (2)
efficiency does not depend upon banding; and (3) variables may be eliminated in a
different order from that in which they are activated. In using frontal solvers, one is
apt to encounter a rather formidable amount of bookkeeping, that is, programming
related to data handling. Solvers that exploit handedness are generally simpler in this
respect.
a c d e f
b I
I
I
I '- I
10\ I
/
i
l I, 01 I
-- - (a) After reduction of ib

' -//
...- ----- .......

I 15 18
I
\ 0 I
" I
\ I
I / ;Q1 I
11
\' 01/
I
' ::. -:.. (b) Before reduction of ab
(c) After reduction of ab
(d) After reduction of be

------------------------ -
---18----13---11- - -l7--..\
/ '
'115 12,161
11
11 0
\'-----.... ,,- ---------- -- --' I
------.. /-----------------
I I
\ I
1 11
I \
1 1011
0II
'...._-.::,/ (e) After reduction of ef
Figure 11.3 Wave front concept.
11.6 ERRORS
In structural analysis, as in modeling all physical problems, results are never exact.
Recognizing that perfection is unattainable, the analysis must, however, provide an
. swers sufficiently accurate to make a reliable estimate of structural behavior and, in
turn, permit safe and economical design. Major sources of error in framework analysis
may be grouped as follows:
Idealization. The process whereby an actual structure is converted to a computa

tional model. Accuracy is affected by all simplifications and personal judgments
11.6 Errors 323
in representing the loading, geometric and material properties, and support

conditions.
Input/Output. These include errors that range from mistakes made while entering
the idealized data into a computer analysis program to interpreting the results
provided by such a program. Note that the two are intrinsically related; input
errors almost always lead to incorrect output.
Procedural. Errors that arise due to approximations employed in the solution al
gorithms. These can be both theoretically and numerically based.
Manipulation. Those errors that occur during computational processes when irra
tional numbers, and in some cases rational numbers, are represented by an ap-
. proximate value that has a fixed number of significant figures. They derive fun-
damentally from truncation and round-off operations.
It should be clear that this brief catalog is not comprehensive, nor is it completely
definitive. There is in reality no sharp interface between categories. Although all of
the above sources of error are important, we will now focus on errors of the manipu
lation type. The relevance of procedural errors will be illustrated in the next chapter.
The significance of manipulation errors in solving systems of equations is directly
related to the condition of the problem (for a treatise on the effect of manipulation
errors in algebraic processes see Ref. 11.10). If small errors in the coefficients of the
equations, or in the solving process, have little effect on the solution, the problem
is well-conditioned. On the other hand, if the effects are large, the problem is ill
conditioned. As shown in Figure 11.4, we may picture these ideas graphically by con
sidering the system
a11X1 + a12X2 = bi
az1X1 + az2X2 = b2
If the equations-or rows of the coefficient matrix-are linearly dependent, no so

lution exists, the coefficient matrix is singular. This is illustrated in Figure ll.4a by two
Xz
(b)
Xz Xz
(c) (d)
Figure 11.4 Problem solution. (a) Singularity.

(b) Orthogonality. (c) Well-conditioned.
(d) Ill-conditioned.
parallel lines. This should only occur in a framework analysis when the structure is
unstable. Optimum conditioning is obtained if the two lines are orthogonal, a rare
condition that is illustrated in Figure ll.4b.
The normal well-conditioned problem is represented by the intermediate case shown
in Figure ll.4c. Although there will be inevitable solution errors and the problem
actually solved will not be the mathematical one presented by the solid lines, but rather
the imperfect physical one indicated by dashed lines, the error should be tolerable.
That is, the calculated solution vector {xA} will he sufficiently close to the true solution
{x7}. When the angle between the lines becomes small, as in Figure ll.4d, the problem
becomes ill-conditioned. The solution is then sensitive to small errors of representation
or computation and it may be quite inaccurate. In other words, a wide range of answers
can closely satisfy an ill-conditioned system of equations (see Example 11.12).
These concepts apply with equal validity to problems of large, practical size. In
framed structures, conditioning problems may be encountered in systems where: (1)
adjacent elements have widely varying stiffness; (2) it is possible for large rigid body
rotations to occur without causing significant strain; or (3) there are many degrees of
freedom (perhaps thousands). In Exam.pie 11.7, the behavior of a number of combi
nations of two axial force members of different stiffness is compared. The range of
stiffness ratios is extreme and designed only to bring out different numerical
characteristics.
EXAMPLE 11. 7!
For the structure shown, vary the relative stiffness of
the two elements and determine the relative errors .
..---L
--- i----.,..--- Lz---+i
of 10-digit and 5-digit Gauss elimination solutions.
Use a 10-digit closed form solution as a standard
of comparison. -Tl each case, input and round-off to the
defined number of digits after every addition, subtrac
tion, multiplication, and division. Let E 1 A1/ L1 = k1 and
E2A2/ L2 = k2.
Closed form solution:
Thus
( a)
Gauss elimination solution:
(k1 + kz)l::.b - kz!::.c = 0
-k2!::.b + kzl::.c = P
p
!::. c =
[
-------
k2 -
(-k2)(-k2) ] ( b)
(k1 + kz)
11.6 Errors 325
Comparison:
1. Let k1 = v'3. Vary k2 from V2x10-6 to V2 X 1010

2. Closed form solution. Use Equation a and 10-digit accuracy.
3. JO-digit Gauss elimination. Use Equation band 10-digit accuracy.
4. 5-digit Gauss elimination. Use Equation band 5-digit accuracy.
Results:
10-Digit Gauss 5-Digit Gauss

Closed Form
Case k1lk2 A A for P = 1 A for P = 1 % Relative Error* A for P = 1 % Relative Error
1 v'31V2x10-6 Ab 0.5773502690 0.5773502692 -3.46x10-8 0.57733 3.51x10-3

= 1.225x106 Ac 707107.3588 707107.3589 -1.41x10-8 707110. -3.74x10-4
2 v'31V2x10-4 Ab 0.5773502690 0.5773502694 -6.93x10-8 0.57734 1.78x10-3

= 1.225x104 Ac 7071.645164 7071.645166 -2.83x10-8 7071.6 6.39x10-4
3 v'31V2x10-2 Ab 0.5773502690 0.5773502688 3.46x10-8 0.57736 -1.69x10-3

= 1.225x102 Ac 71.28802841 71.28802840 1.40x10-8 71.291 -4.17x10-3
4 v'31V2 Ab 0.5773502690 0.5773502694 -6.93x10-s 0.57736 -1.69x10-3

= 1.225 Ac 1.284457050 1.284457051 -7.79 X 10-s 1.2845 -3.34x10-3
5 v'31V2x102 Ab 0.5773502690 0.5773502585 1.82x10-6 0.57773 -6.58x10-2

= 1.225x10-2 Ac 0.5844213368 0.5844213262 1.81x10-6 0.58480 -6.48x 10-2
6 v'31V2x104 Ab 0.5773502690 0.5773498363 7.49 X 10-s 0.49993 13.41

= 1.225x10-4 Ac 0.5774209797 0.5774205469 7.50 X 10-s 0.50000 13.41
7 v'31V2x106 Ab 0.5773502690 0.5773664984 -2.81x10-3 co -co
= 1.225x10-6 Ac 0.5773509761 0.5773672055 -2.81x10-3 co -co
8 v'31V2 X 108 Ab 0.5773502690 0.5882352870 -1.89 co -co
= 1.225 X 10-s Ac 0.5773502761 0.5882352941 -1.89 co -co
9 v'31V2x1010 Ab 0.5773502690 co -co co -00
= 1.225x10-10 Ac 0.5773502691 co -co co -co
% Relative Error = 100 (true - approximate)/true
The reasous that problems are encountered in the Gauss solution process but not
in the closed form solution can be explained. First, it is emphasized that both solutions
I
are algebraically equivalent; the closed form solution ( Equation a ) is a simplification
of the expressions obtained by Gauss elimination ( Equation b ) . Hence, manipulation
errors are the only source of any problems.
In the closed form solution the displacement Llc depends on the sum of the recip
rocals of the two stiffnesses. The only manipulation errors are those that result from
truncating the smaller reciprocal term when it is negligible and rounding of the cal
culated sum and its product with the axial force P. The truncation error directly cor
responds to the physics of the problem; if the two stiffnesses vary significantly, the
member with smaller stiffness will control the size of Llc.
In using a numerical equation-solving scheme such as Gauss elimination, the solution
first requires stiffnesses k1 and k2 to be added. If one value is much smaller than the
other, significant information is lost irretrievably through the necessity of truncating
this sum to the number of figures retained by the computer. If the mismatch is so great
that the significance of k1 is lost in this addition, the displacement Llc may become
extremely inaccurate or even unattainable. To make matters worse, any error in Llc is
propagated into the calculation of displacement !J.b as a result of the back-substitution

phase of the Gauss solution process.
Perhaps it is necessary to indicate that Example 11.7 is not intended to imply that
element stiffnesses are actually known to 10 digits. The example is intended only to
illustrate cases of well-, poorly-, and ill-conditioned problems. Rarely can we be con
fident of more than two- or three-digit knowledge of stiffness quantities. The point is
that, for a given level of confidence in these quantities, how can we detect and avoid
or compensate for attrition produced by numerical error and thus assure a solution
with at least the same level of reliability?
11.6.1 The Condition Number
We now turn to searching for general methods for detecting ill-conditioning in large
order systems. At the same time we will be estimating the effect of poor conditioning
on the accuracy of results obtained in computations that retain only a fixed number of
significant figures. Truncation errors will be emphasized because it has been found that
initial truncation errors appear to be the main source of manipulation errors in prob
lems of structural analysis (Ref. 11.11). To adequately describe and assess this error
!" first have to take a brief excursion into the subject of norms.
A norm is a function that provides a measure of the size or length of a vector or
matrix. For present purposes a norm must exhibit properties of
(i) Scaling: lla{xJll lal llxll

=
(11.37a)
(ii) Positiveness: llxll > 0 for any {x} * 0 (11.37b)
(iii) Triangle Inequality: llfx} + {yJll =:::; llxll + llYll (11.37c)
As examples, the Euclidean norm for a vector is expressed as
llxlle = \/{XrTxJ = J xt (11.38)
and the Frobenius norm for an n x n matrix [A] is
n n
llAllF = L L at (11.39)
i=l J=I
Note that the convergence criterion given earlier in Equation ll.35a can be expressed
as simply the Euclidean norm of the residual vector llr;ll . . Another popular and simpler
norm to apply is the uniform or infinity norm. For a vector this norm is expressed as
the maximum magnitude
llxlloo =
max lx;I (11.40)
I sisn
and for an n x n matrix [A] as the maximum row-sum

n
llAlloo =
max L la;1I (11.41)
lsi:sn j=l
Finally, if a vector norm and a matrix norm are consistent then it can be shown that
(Ref. 11.2)
li[A]{xJll =:::; llAll llxll (11.42)
With this understanding of norms, an error analysis can commence by assuming that
Equation 11.l could be solved exactly for {x}. First let us suppose that the coefficients
11.6 Errors 327
in{b} are also known exactly but those in [A) are subject to some small uncertainty
[8A). Equation 11.1 is modified to
[A + 8A]{x} = {b} (11.43)
The difference between the solutions to Equations 11.1 and 11.43, {x} and {x} respec
tively, would in general reflect how sensitive Equation 11.1 is to small changes in the
coefficient matrix [A). If we denote this difference by
{ 8x} = {x} - {x} (11.44)
and substitute this in Equation 11.43, we obtain
[A + 8A]{x + 8 x} = {b} (11.45)
Solving this for the error {8x} gives

{8x} =
-[Ar'[8A]{x + 8x} (11.46)
Taking the norm of both sides of Equation 11.46 and employing Equations 11.37 and
11.42 yields
ll8xll ::; llA-'11 ll8All llx + 8xll (11.47)
This can be simplified to a comparison of relative errors given by
ll8xll
<K -- ll8All (11.48)
llx + 8xll - llAll
or by further employing Equation 11.37c
ll8xll < ll8All

-
K
(ll.48a)
llxll ll8 All llAll
l - K
llAfl
where K, the condition number of the matrix [A] with respect to a given norm, is given
by
llAll llA-'11
K =
(11.49)
By further subjecting the coefficients in both [A) and {b} to uncertainties [8A] and
{8 b}, it can be shown that
ll8xll
::;
K
( 118All
+
ll8 bll ) (11.50)
Txif l - K
ll8All llAfl libil
llAfl
It is important to note that in both Equations ll.48a and 11.50 the "error amplifi
cation" term Kl(l -
KlI8All!lIAII) can be approximated by the condition number K when
A A
ll8 ll!ll ll is sufficiently small. Hence, it should be clear that a bound can be placed on
the relative error of the solution and more importantly it is directly proportional to
the condition number K of the coefficient matrix [A). Obviously, the smaller the con
dition number (that is, the closer it is to unity) the more well-conditioned the system.
11.6.2 Estimate of Condition Number
To compute the condition number using Equation 11.49 would be a formidable task
because it requires determining [Ar'. This is further complicated by the fact that if
[A] is ill-conditioned, the inverse [A]-1 would be most likely computed with large
32S Chapter 11 Solution of Linear Algebraic Equations
errors and hence the accuracy of the condition number would be at best questionable.
Thus we cutomarily rely on estimates or bounds on the condition number.
One reasonable estimate can be obtained from a lower bound estimate of the norm
of [At1 (Ref. 11.5)
(11.51)
where [BJ is a singular matrix which is "near" or close to the nonsingular matrix [AJ.
Combining Equations 11.49 and 11.51 provides a bound on the condition number
> llAll
(11.52)
K - llA - Bii
The singular matrix [BJ may be formed by first setting it equal to [AJ and then making
small changes so that it becomes singular. For example, this can usually be done by
choosing [BJ to have the same elements as [AJ except that the last row is replaced by
a linear combination of all of its other rows such that it matches the last row of [AJ in
!!but one position. Example 11.8 illustrates the use of Equations 11.41 and 11.52.
Use Equations 11.41 and 11.52 to estimate the condition number for the following:
10x1 + 8x2 + 7x3 + 7x4 = 32
8x1 + 10x2 + 9x3 + 6x4 = 33
[ ]
7x1 + 9x2 + 10x3 + 5x4 = 31
7x1 + 6x2 + 5x3 + 5x4 = 23
10 8 7 7
8 10 9 6
( )
[A) 7 9 10 5
=
7 6 5 5
1101 + 181 + 171 + 17 1
181 + 1101 + 19 1 + 161 33
llAll oo max
171 19 1 1101 151
=
=
+ + +
l?i + 161 + 151 + 151

To find a singular matrix [B) close to matrix [A]:
Partition [ A) according to [1r-1l]

[TY: 5
with [S) 1
9 10
]. [T) [H ond
rn { }
[R]
0.6029
Solve [S]{y} =
[T], (y} = 0.2500
-0.1471
11.6 Errors 329
Let
B4.4
{ }[]
[BJ be identical to [A] except Bn.n {y}T[R]
0.6029 T 7
0.2500
= 6 4.9853
=
-0.1471 5
Hence , (BJ = [l
7
7
1

9 10 5
6 5 4.9853
] which is close to (A] and singular.
llA - Bii = IAn,n - Bn,nl = IS - 4.98531 = 0.0147

With
llAll 2.24 x 103

Bii 0.0147
= =
llA -
K 2.24 x 103 which indicates the system is ill-conditioned.
Other more general methods for estimating a condition number are presented in
the literature ( see Ref. 11.2). An in-depth study of solution error ( in Ref. 11.5) shows
that the lower bound on the condition number is
max j A;I
1 si:s.n
K ---
(11.53)
min I A;I
l'S.i'!E;n
where A; are the eigenvalues of the coefficient matrix [A]. Eigenvalues may be obtained
by expanding the determinant
i[A] - A [I]I = O (11.54)

where [I] is the identity matrix, and then solving for the roots of the resulting char
acteristic equation. Since eigenvalues are of fundamental importance in critical load
analysis, an introduction to them and to eigenproblems in general will be presented in
Chapter 12.
It should be noted, however, that as the number of equations n increases, calculating
the eigenvalues using Equation 11.54 can become a task <..:> formidable as calculating
[Ar 1 . Fortunately, there are simple ways of estimating these values. For example, an
upper bound on the largest eigenvalue can be obtained from the infinity norm
max j A;I :::;,; llAll"' (11.55)

lsisn
An estimate of the smallest eigenvalue can be calculated using the relatively simple
inverse iteration procedure presented in Section 12.8.4.
11.6.3 Error Estimates and Preconditioning
An approximate condition number is often all that is needed in estimating potential

error. Suppose that the coefficient matrix [A] is represented in the computation by p
decimal places, that is llcSAll!llAll 10-p, and similarly, s is the resulting number of
=
correct decimal places in the solution, llcSxll!llxll 10-s. From Equation 11.48, the
=
relationship between p ands-the effect of the initial truncation error-may then be

estimated as
S ;:::: p - Jog IQK (11.56)
Since the formula contains log 10K and, at most, s and p are on the order of ten, it is
clear that great precision in calculating K is not required. The use of Equations 11.53
and 11.56 is illustrated in Example 11.9.
EXAMPLE 11.9
Using the stiffness matrix of Example 11.7, calculate the eigenvalues, condition numbers, and
estimated number of correct decimal places in Cases 1, 4, and 7 of that example. Compare with
results of Example 11.7.
Eigenvalue calculation:
[[ ]
((K] - A(I]]{x}= 0
J
(k, + k2) -k2
- A[I] {x} = 0
_ k2 kz
[ (k, + kz - A)
-k2
-k2
(k2 - A)
J
{x} = 0
Characteristic equation:
2
(k, + kz - A)(k2 - A) - (-k2) = 0
2
A - (k, + 2k2)A + k1k2 = 0
Solving for A,
Results:
p=10 p=5
k, s s s s
Case kz A, A2 K = A,IA2 log,o K (p - Iog10 K ) ( Ex. 11.7) (p - log,o K) ( Ex. 11.7)

v'3
1 1.73 1.41 x 10-6 1.22 x 106 6.1 4 9 0 4
\/2 x 10-6
v'3
4 3.94 G.:52 6.33 0.8 9 9 4 4
\/2
v'3
7 2.83 x 106 0.87 3.27 x 106 6.5 4 4 0 0
\/2 x 106
Example 11.9 indicates that Equation 11.56 is accurate most of the time. However,
in some cases, such as Case 1, it does forecast ill-conditioning even thocgh the results
of Example 11.7 indicate this problem is well-conditioned. This is a situation of artificial
ill-conditioning, or skewness. This misleading behavior may be detected and rectified
by a procedure known as preconditioning or equilibrating. It is important to note that
11.6 Errors 331
preconditioning does not change the nature of the problem; a truly ill-conditioned
system will remain so even after preconditioning. Hence, when the condition number
of a properly equilibrated matrix is used in Equation 11.56, it should provide a reliable
estimate of the initial truncation error.
Preconditioning can be explained as the following series of matrix manipulations.
First, we premultiply both sides of Equation 11.l by [Di]
[Di][A]{x} = [D1]{b} (11.57)
and then substitute the solution vector {x} according to
{x} = [Dz]{x} (11.58)

where [Di] and [Dz] are square matrices to be prescribed. The resulting preconditioned
system of equations, one with a more reliable and often improved condition number,
is
[A]{x} = {b} (11.59)

where [A] = [Di][A][Dz] and {b} = [Di]{b}.
Although it goes beyond our present needs, it may be observed that, for any [Di]
and [Dz], Equation 11.59 may be solved for {x} and the result substituted in Equation
11.58 to find the solution {x) to the original system of equations. As suggested in Section
11.3.2, employing such a preconditioning strategy within the conjugate gradient
method can make this iterative solution scheme more effective by significantly im
proving its rate of convergence.
Several schemes have been proposed for the selection of [Di] and [Dz]. In Reference
11.12 it is suggested that good results can be obtained by letting [Di] = [Dz] =[D]
where [D] is a diagonal matrix whose elements are defined by
d;; = 11\/ll;; for i = 1, . . . , n (11.60)

-
In Example 11.10 this preconditioning scheme is applied to the cases presented in

Example 11.9. A dramatic change in K for Case 1 is observed and the system is now
corretly predicted as well-conditioned. The ill-conditioning of Case 7 p_roperly
remams.
Example 11.11 is an illustration of the above statement that preconditioning does
not change the nature of the problem. Several cases treated in Example 11.7 are solved
again after equilibration by the scheme just described. There is seen to be no essential
difference in the results obtained in the two examples.
Repeat Example 11.9 using a stiffness matrix equilibrated by Equation 11.60.
[\/<+
Preconditioning:
k,
[K] - [D][K][D]
[
-
J
=
1 -
Vk + kz
VT,_
v7<i+Tz VT,_
Determine eigenvalues of equilibrated matrix:
[ (1 - A)
[(K] - A(I]]{x)= _ \/k; Vk:+k-i {x} = 0
(1 - A)
Vk;+k; Vk; ]
Characteristic equation:
(1 -
A2 - 2 A
A)2 -
+
__ki _kk_i2_k2
+
- =0
=0
+
k 1 k2
Solving for A,
and
Results:
p=10 ( p=5
1
kk2 s s s s
Case Ai A2 K = A1/A2 logtO K (p - log10 K ) (Ex. 11.7) (p - logto K ) (Ex. 11.7)
V3
1 1.001 0.999 1.002 =O 10 9 5 4
\12 x 10-6
V3
4 1.67 0.33 5.07 0.7 9 9 4 4
\12
V3 0 0
7 2.00 6.12 x 10-7 3.27 x 106 6.5 4 4
\12 x 106
For Cases 1, 4, and 7 of Example 11.7, investigate the accuracy of results obtained by 10- and
5-digit Gauss elimination following preconditioning b y Equations 11.57 to 11.60.
Start with equations that have been preconditioned algebraically:
[ -]{:}{} : {:}{V ,}
k d
Inserting numerical values and solving for {:ii} and then {Ii), the results are
10-Digit Gauss 5-Digit Gauss

kk12 Closed Form Preconditioned Preconditioned
Case A A for P=1 Equations Equations
V3 Ab 0.5773502690 0.5773502695 0.57733

1
\12 x 10-6 Ac 707107.3588 707107.3586 707110.
V3 Ab 0.5773502690 0.5773502698 0.57735

4
\12 Ac 1.284457050 1.284457051 1.2845
V3 Ab 0.5773502690 0.5775121819 00
7
\12 x 106 Ac 0.5773509761 0.5775128889 00
11.6 Errors 333
11.6.4 Detecting, Controlling, and Correcting Error

We close this section on error analysis by shifting our attention from predicting possible
error to detecting and, if possible, controlling and even correcting for numerical errors
actually committed.
. .
Perhaps the most obvious and commonly used procedure for error detection 1s sub
stituting the solution vector obtained, say,{xA}, back in the original system of equations
and calculating the corresponding right-hand side vector {bA} If the elements of {bA}
differ substantially from those of the original right-hand side vector {b}, it is clear that
significant errors have occurred in the solution process and we have not found the true
solution {xrl Unfortunately, the converse is not necessarily true; small differences
between {bA} and {b} do not prove that {xA} is accurate. By the mere definition of ill
conditioning given earlier, several solutions which may vary widely can provide a {bA}
that is acceptably close to {b}. In this same regard, according to
(11.61)
the elements of [At1 may be so large-another sign of possible ill-conditioning-that

a small discrepancy in the right-hand sides {c5b} may be severely amplified, producing
a wide range in solution vectors.
In a structural analysis, back substituting the solution vector is the equivalent of
performing an equilibrium check. Although this check may be inconclusive, extending
it to a check for displacement compatibility will confirm that a solution is the correct
one for a given idealization. Where a complete equilibrium and compatibility check
can be a time consuming process, a distributed sampling of checks can often confirm
the merit of a solution.
When a poorly conditioned problem is encountered, several measures can be used
to correct or at least control the amount of numerical error. One scheme is to use
scaled6 equations as a criterion for employing partial pivoting (see Section 11.2.1).
Solution accuracy may then be improved by interchanging rows of original coefficient
values so that the divisor is always the most significant coefficient in the submatrix
column of the unknown being eliminated. This concept may be extended to complete
pivoting, where columns are also interchanged so that the dividing coefficient is the
largest in the entire submatrix of coefficients to be operated on. In complete pivoting,
symmetry is preserved.
Another method for increasing solution accuracy is to employ a technique known
as iterative improvement (Ref. 11.12). After factoring [A) according to algorithms such
as Equations 11.13 or
11.15, the solution {x1} to Equation 11.1 is first computed. The
residual of this solution {r1} is then computed according to Equation 11.31. Using {r1}
and the original decomposition of [A], a solution {z1) is obtained for
(11.62)
A more accurate solution to the original system of equations can then be furnished by
(11.63)
The process would then be repeated until the change in the solution {z') becomes
negligible. In a structural analysis, this approach is equivalent to computing residual
6Scaling refers to normalizing the equations so that the maximum coefficient in a row equals unity. Because
scaling may introduce rounding error, it is recommended that the scaled coefficients be used only as a
criterion for pivoting and the original coefficient values be retained within the elimination and substitution
computations.
loads from an equilibrium analysis and then applying these loads to the structure as
forces to obtain displacement corrections.
Considering that a majority of numerical errors in framework analyses are the result
of truncating valuable information, it is important to note that there are no schemes
available for retrieving or appropriately compensating for this lost information. In this
regard, perhaps the most effective approach in controlling numerical error is to use
the greatest computational precision available. In doing so, one must be sure to for
mulate all load vectors, element properties, and stiffness coefficients with the same
high-level precision used during the solution process.
Example 11.12 illustrates the above point that small differences in {bA) and {b) do
not prove that {xA) is necessarily accurate. The use of iterative improvement is illus
trated in Example 11.13.
EXAMPLE 11.12
Show that the following two solutions closely satisfy the system of equations given in Example
1 .8 L6 2.9 -0.l -7.2jT
1.
and that neither is close to the true solution
{xA} =
L1 1 1 1
{xr} = Jr
2. Ll.50 1.19 0.89 o.18y 8 7

{xA} =
[AJlx) =
['! 1690 1905
[b} with [A] =
1. with L6 2.9 -0.1 -7{3.2J2T.1}

{xA} =
{bA}
33221...991
= (A]{xA} =
{0:.1}
{ob}= {b} - {bA} = when {Bx}= {xr} - {xA}
{=::}8.2 =
2. with Ll.50 1.19 0.89 {30.21.80J11}

{xA} =
{bA}
33221...909l
= (AJ{xA} =
{Bb} =
{-00..0011}
{b} - {bA} =
-O.Ol
when
{-0-0O...51l9}
{ox} = {xr - xA} =
0.01 0.82
Use iterative improvement on the second solution in Example
solution.
1 .12 to converge on the true
11. 7 Problems 335
With {xi}= Ll.50 1.19 0.89
{ri} = {b} - [A]{xi} =

{ }
0.18Y
-0.Ql
0.01
-O.Ol
-I
{zi} = [A] {ri} =
{ } -0.50
-0.19
O.ll
0.01
0.82
(x,1 = {xd + '" ' =m

("} = {bl - [A}{x,J = { } indi"ting that {x,] '' tho trno 'olution.
11.7 PROBLEMS
11.1 Solve the following systems of equations by (1) Gauss elimination; (2) the Cho
lesky method; and (3) the root-free Cholesky method.
(a) 64x1 + 32x2 - 16x3 = 140
32x1 + 416x2 + 72x3 = 170
-l6x1 + 72x2 + 120x3 = 5
(b) 25x1 + 5x2 - 2.5x3 = 60
5x1 + l7x2 + ll.5x3 = -8

-2.5x1 + ll.5x2 + 13.25x3 = 11
11.2 Attempt to solve the following system of equations by the Cholesky method.
What can be concluded?
20x1 + l0x2 - 40x3 = 100
l0x1 + l5x2 + 30x3 = -110
-40x1 + 30x2 - 20x3 = -300
11.3 Solve the equations of Problem 11.2 by the root-free Cholesky method.
11.4 Solve the equations of Problem 11.1 by (1) Gauss-Seidel iteration; and (2) Jacobi
iteration. Obtain a solution which is accurate to within three significant figures.
11.5 Solve the equations of Problem 11.lb by (1) Gauss-Seidel iteration; and (2)
Gauss-Seidel iteration with overrelaxation f3 = 1.4. Use ? = 5%.
11.6 For the equations of Problem 11.2, perform five iterations of the Gauss-Seidel
method. Based on this solution and the conclusion made in Problem 11.2, comment
on the apparent effectiveness of using this scheme.
11.7 Solve the equations of Problem 11.1 by (1) the conjugate gradient method; and
(2) the conjugate gradient method with preconditioning. Use ( = 0.1. Discuss the effect
that preconditioning has on the error calculated at the end of each iteration.
11.8 Solve the equations of Problem 11.2 by the conjugate gradient method. Use
'= 0.1.
11.9 Solve the following systems of equations by the Cholesky method using (1)
Equation 11.15 and (2) Equation 11.36a, which accounts for handedness.
(a) = -100
100x1 + 409x2 - 9x3 = -445
-9x2 + 10x3 + 3x4 = 63
3x3 + 13x4 = 74
(b) 16x1 - 48x2 16

-48x 1 + 180x2 + 72x3 = -192
72x2 + 153x3 + 36x4 = -171
36x3 + 148x4 + 4x5 = 492
4x4 + 29x5 = 74
11.10 For the frames shown, label the degrees of freedom so that the bandwidth of
[Kff] is at a minimum. Use Equation 11.36b to calculate the half-bandwidth.
(a) (b)
Problem 11.10
11.11 Solve the structure shown using a wave front approach. E = 200,000 MPa
r -50 kN -30 kN -IOkN

t==:
:= f ti x===:'.:
200 f::::
:
106 mm4
I=
:=
::=;:
x
==
=::::
=
L- 4@2m=Bm --
Problem 11.11
11.12 Use Equations 11.41 and 11.52 to comment on the condition of the equations
of (1) Problem 11.1; and (2) Problem 11.2.
11.13 Comment on the condition of the following systems of equations using (1)
Equation 11.52 with norms calculated by Equation 11.41; and (2) Equation 11.53.
(a) 1.2x1 + 0.4x2 2
= (b) X1 + O.Bx2 = 3.1
23.9x1 + 7.95x2 = 39.8 99.9x1 + 1000x2 = -2.5
11.14 Repeat Problem 11:13 after applying preconditioning.
References 337
11.15 Solve the following system of equations and discuss the condition of the system:
1 .00 x1 + 0.50x2 + 0.33x3 = 6

0.50x1 + 0.33x2 + 0 . 25 x3 = 4
0.33x 1 + 0.25x2 + 0. 20 x3 = 2
11.16 The following system of equations is similar to that in Problem 11.15. Solve it
exactly by retaining the fractional notation. Compare your solution with that obtained
in Problem 11.15. Approximately how many digits would have to be used in a Gauss
elimination solution using decimal notation in order to obtain four-digit accuracy?
x, + !x2 + ! x3 = 6
!xi + !x2 + ! x3 = 4
!xi + ! x2 + !x3 = 2
11.17 Repeat Example 11.9 using Equations 11.41 and 11.52 to calculate the condi
tion numbers.
11.18 The structure shown consists of three axial force members in series. The forces
are Pb = Pd = 1 and P. = 0. The stiffnesses are kab = kc,1 = 1 and kbc = 1 X 104
(1) Calculate the displacement vector using Gauss elimination and four-digit arithmetic
(truncate to four digits after every operation-be sure that truncation is done and not
merely on the calculator's display-this may require reentry after every operation).
(2) Repeat using eight-digit arithmetic. (3) Based on these solutions, discuss the con
dition of the system.
c
Problem 11.18
11.19 Use Equations 11.52 or 11.53 and Equation 11.56 to estimate the truncation
error for Problem 11.18. Compare with results obtained in Problem 11.18.
11.20 Precondition the coefficient matrix of Problem 11.18, and repeat Problems
11.18 and 11.19.
11.21 Repeat Problems 11.18 to 11.20 with kab = kct1 = 1 X 104 and kbc = 1.
REFERENCES
11.1 L. Fox, An Introduction to Numerical Linear Algebra, Oxford University Press,

New York , 1965.
11.2 G. H. Golub and C. F. Van Loan, Matrix Computations, 3rd edition, The Johns
Hopkins University Press, Baltimore, Md., 1996.
11.3 K. J. Bathe, Finite Element Procedures, Prentice-Hall, Englewood Cliffs, N.J.,
1996.
11.4 J. B. Scarborough, Numerical Mathematical Analysis, 6th edition, The Johns
Hopkins University Press, Baltimore, Md., 1966.
11.S J. L. Buchanan and P. R. Turner, Numerical Methods and Analysis, Mc Graw
11.6 A. G e orge and J. W. H. Liu, Computer Solution of Large Sparse Positive Definite
Systems, Prentice-Hall, Englewood Cliffs, N.J., 1981.
11.7 K. J. Bathe and E. L. Wilson, Numerical Methods in Finite Element Analysis,

Prentice-Hall, Englewood Cliffs, N.J., 1976.
11.8 G. H. Paulino, I. F. M. Menezes, M. Gattass, and S. Mukherjee, "Node and
Element Resequencing Using the Laplacian of a Finite Element Graph," Intl.
JI. Num. Meth. in Engr., Vol. 37, No. 9, 1994, pp. 1511-1555.
11.9 B. M. Irons, "A Frontal Solution Program for Finite Element Analysis," Intl JI
for Num. Methods in Engineering, 2, 5-32, 1970.
11.10 J. H. Wilkinson, Rounding Errors in Algebraic Processes, Prentice-Hall, Engle
wood Cliffs, N.J., 1963.
11.11 R. A. Rosanoff, J. F. Gloudeman, and S. Levy, "Numerical Conditioning of
Stiffness Matrix Formulations for Framed Structures," Technical Report
AFFDL-TR-68-150, USAF Flight Dynamics Laboratory, Wright-Patterson Air
Force Base, Ohio, 1968.
11.U G. Forsythe and C. B. Moler, Computer Solution of Linear Algebraic Systems,
Chapter 12
Solution of Nonlinear Equilibrium

Equations
In Chapter 8 we outlined strategies for the numerical solution of the equilibrium equa
tions associated with a nonlinear analysis. The main purpose of the current chapter is
to provide details of these schemes. Two distinctly different mathematical approaches
will be explored.
We begin with the incremental method and present the single-step and iterative
(multi-step) procedures that are most commonly used to calculate nonlinear structural
behavior. Particulars considered crucial to the successful implementation of these
methods are examined. They include selection of appropriate load increment size,_
updating the deformed geome!ry, force recovery algorithms, and strategies for de
tecting limit points and tracing post-limit point behavior.
We then proceed to the second approach, in which a predicted limit point or critical
load is determined from an eigenvalue analysis. After providing an introduction to
basic concepts, the nonlinear equilibrium equations are cast in different forms of
eigenproblems and direct and iterative schemes for solving them are presented.
This chapter is not intended to be comprehensive or to replace textbooks dedicated
to these and related nonlinear solution topics. It should, however, provide an engineer
with information sufficient to perform successful nonlinear analyses of frameworks.
U.1 INCREMENTAL ANALYSIS
As shown in the preceding chapters, formulation of equilibrium on the deformed and

perhaps yielded geometry of a structure will result in a system of nonlinear stiffness
equations. One method for solving these equations is to approximate their nonlinearity
with a piecewise segmental fit (see Figure 12.1 in which the approach is illustrated by
a representative load component-displacement diagram) . In doing so, the total load on
a structure, {P}, or its equivalent, the product of a total load ratio A and a given ref
erence load {Pree}, is applied through a series of load increments {dP;). Mathematically,
this is written as
n
{P} = A{Prer} = 2: {dP;} (12.1)

i=1
where n is the total number of load increments employed. The structure's correspond
ing displacement response {a} then follows as
{a} =
2:: {da;} (12.2)
i=1
339
340 Chapter ll Solution of Nonlinear Equilibrium Equations
p
Calculated Equilibrium Path
Actual Equilibrium Path
Flpre ll.1 Piecewise segmental tit.
where {dA1} is the vector of displacements resulting from applying the ith load incre
ment {dP1}.
The degree to which a piecewise fit approximates the actual equilibrium solution is
clearly a function of how accurately the nonlinear relationship between {dP1} and {dA1)
is represented during each load increment. To adapt the relatively simple techniques
presented in Chapter 11 for solving linear algebraic equations, the nonlinear behavior
that occurs during each increment will be approximated by performing one or more
linear analyses. The number of analyses performed and how their results are used to
evaluate the displacement {dA,) produced by a given load increment {dP,) is what
distinguishes various nonlinear solution techniques. In this regard, general categories
of single-step methods and iterative methods will be presented.
12.2 INCREMENTAL SINGLE-STEP METHODS
The single-step methods all employ a strategy that is analogous to solving systems of
linear or nonlinear differential equations by the Runge-Kutta methods. In general they
can be cast in the form
(12.3)
where {A,_ 1) and {A,) are the total displacements at the end of the previous and current
load increments, respectively. In Equation 12.3, the increment of unknown displace
ments {dA,) is found in a single step by solving the linear system of equations
[K,]{dA,) = {dP,) (12.4)
where the load increment {dP,) is given by
{dP,J = dA,{Prer} (12.4a)
The size of the load ratio for the first increment, dA1i which must be prescribed by
the analyst, should normally be about 10%-20% of the anticipated maximum applied
load. The same value may be used for dA1 in subsequent increments of the analysis,
but as shown in Section 12.4 there are also automated procedures for calculating this
quantity that may be used to advantage.
In Equation 12.4, [K1] can be interpreted as the representative stiffness for the entire
load increment. Taken as a weighted average, [K ,] is written as
m
[K,] = a1[K1] (12.S)

lI
where a1 is the weighting coefficient corresp onding to a stiffness [K1] that was calcu
lated at one of m sampling points within the increment. Each [KJ] can be defined by
12.2 Incremental SlnaJeStep Method 341
typical Runge-Kutta type recurrence relationships. That is, the sampling point used to
calculate [K1] is obtained by using some or all of the stiffness matrices that correspond
to the previous j -1 sampling points. The number of sampling points, m, employed
in each step defines the order of the method. In general, the accuracy of the nonlinear
relationship between {dP1} and {dA1} will improve with an increase in this order.
By varying the weighting factors and the number and location of stiffness sampling
points, various types of single-step strategies can be devised from Equation 12.S. Two
popular schemes are the Euler and midpoint Runge-Kutta methods.
12.2.1 Euler Method
The most elementary of the single-step strategies is the Euler, or simple step, method.
Using Equation 12.5 with m = 1 and a1 = 1, the Euler method can be cast as a ftrst
order Runge-Kutta approach with
[K,] = 1 [Ki]
(12.6)
In this case, [Ki], the tangent stiffness matrix, is calculated using the deformed ge
ometry and corresponding element forces that exist at the start of the increment.
Hence, the displacement vector {dA1l appearing in Equation 12.3 is determined in each
load increment by performing the single linear analysis represented by Equation 12.4.
This method is illustrated in Figure 12.2.1
12.2.2 Second-Order Runge-Kutta Methods
The second-order form (m = 2) of Equation 12.5 is given by
[K,] = a1[K1] + a2[K2] (12.7)

where [Ki] is the tangent stiffness at the start of the increment and [K2) is the stiffness
calculated using the deformed geometry and corresponding element forces at some
point within the increment. Noting that the displacements at this second sampling point
are not initially known, an additional global frame analysis must be performed before
Actual Equilibrium Path
F1gure 12.2 Euler method.
1Figures 12.2-12.6 are schematic load-displacement diagrams in which multi-dimensional response has been
represented as uni-dimensional for illustrative purposes. Vector notation is employed to coordinate the
forces, displacements, and stiffnesses shown with the related equations of the text.
342 Chapter U Solution of Nonlinear Equilibrium Equations
I /
/ Actual Equilibrium
1/
I I/
Path
(dP;}
ldP;}
Figure U.3 Midpoint Runge-Kutta

method.
Equations 12.7 and 12.4 can be applied. This is achieved by first solving the following
equation for {dA,J
[Ki]{dA,..J = ,{dP;} with 0 < , :5 1 (12.8)
where the prescribed load ratio , defines the relative location of the sampling point
within the load increment. The displacements at the second sampling point {A2), which
are needed to formulate [K2], are obtained by adding {dA1...} to the displacements at
the end of the previous increment {A;_i}:
(12.9)
[Ki] and [K2] evaluated and given a set of prescribed weighting factors a1 and
With
a2, [K;]. The incre
Equation 12.7 can be used to calculate the representative stiffness
mental displacements {dA;} appearing in Equation 12.3 are then obtained by solving
Equation 12.4.
One of the most common versions of the second-order schemes is the midpoint
Runge-Kutta method. As the name implies, the second sampling point is taken at the
midpoint of the load increment(,= 112). By further using weighting factors of a1 = 0
and a2 = 1, this method relies fully on using the midincrement stiffness as the repre
sentative stiffness for the entire load increment. Also known as the predictor-corrector
or improved polygon method, this procedure is illustrated in Figure 12.3.
Other popular second-order schemes include Heun's method with a single corrector
(,= 1, a1 = a2 = 1/2) and Ralston's method(,= 0.75, a1 = 1/3, and a2 = 2/3). Using
a single-degree-of-freedom system, the Euler and midpoint Runge-Kutta methods are
illustrated in Example 12.1.
f)XAMPLE 12.1
A nonlinear spring has a stiffness that is related to the force in the spring by k = 112 (P + 1). Using
three increments of the following methods, determine the extension of the spring A when subjected
to a force of P = 3.
1. Euler method
2. Mid-point Runge-Kutta method
3. Compare both solutions to the exact solution A = (P + 1)2 - 1
12.2 Incremental Single-Step Methods 343
1. Euler method with three increments, dA;=1/3

- 1
First increment: Po=0 k1= =0.5
2(Po + 1 )
1
dP1=dA,P=1.0 dA1=k! dP1=2.0
P1=Po+ dP1=1.0 A1=Ao+ dA1=2.0
- 1
Second increment: k2= =0.25
2(Pi + l)
dP2=dA2P=1.0 dA2=ki1dP2=4.0
P2=P1 + dP2=2.0 A2=A1 + dA2=6.0
- 1
Third increment: k3 = =0.1667
2(P2 + l)
1
dP3=dA3P=1.0 dA3=k3 dP3=6.0
P3=P2+ dP3=3.0 A3=A2+ dA3=12.0
When P=3.0, the deflection is A=12.0.
2. Mid-point Runge-Kutta method with three increments, dA;=1/3 and = 112
First Increment:
1
Predictor step Po=0 ks= =0.5
2(Po + 1 )
1
,dP1=,dA1P=0.5 dA=k; ,dP1=1.0
Pm=Po+ ,dP1=0.5 Am=Ao+ dA=1.0
1
Corrector step km= =0.3333 k1=O.Oks + l.Okm=0.3333
2(Pm + l)
1
dP1=dA1P=1.0 dA1=k! dP1=3.0
P1=Po+ dP1=1.0 A1=Ao + dA1=3.0
Second increment:
Predictor step P,=1.0 ks=0.25
1
dP2=0.5 dA=k; ,dP2=2.0
Pm=P,+ dP2=1.5 Am=A1+ dA=5.0
Corrector step km=0.2 k2=O.Oks+ l.Okm=0.2

1
dP2=1.0 dA2=ki dP2=5.0
P2=P1+ dP2=2.0 A2=A1+ dA2=8.0
Third increment:
Predictor step P2=2.0 ks=0.1667
,dP3=0.5 dA=k; I ,dP3=3.0
Pm=P2 + dP3=2.5 A,,, =A2 + dA=11.0
Corrector step km=0.1429 k3=O.Oks+ l.Okm=0.1429

1
dP3=1.0 dA3=k3 dP3=7.0
P3=P2+ dP3=3.0 A3=Az + dA3=15.0
When P=3, the deflection is A=15.0.

344 Chapter 12 Solution of Nonlinear Equilibrium Equations
3, Comparison with exact solution p

Ex11ct Solution
3
12 14 16 t.
Major advantages of the single-step methods are their simplicity and efficiency. In
most cases, only one or two analyses are performed in each increment. In this regard,
these schemes, especially the midpoint Runge-Kutta method, are particularly attractive
for the analysis of structures exhibiting moderate nonlinearity. A drawback of these
techniques is that the error (see Fig. 12.2) resulting from the use of a single represen
tative stiffness in each load increment can accumulate. Hence, the total internal ele
ment forces, which are a by-product of the potentially inaccurate displacements, are
not necessarily in equilibrium with the externally applied forces. Although this so
called drift-off error can be reduced by using a smaller load ratio dA1, the additional
number of increments required for analyzing highly nonlinear systems may become
unreasonable. In these cases the use of an iterative scheme may be more appropriate.
Examples 12.2 and 12.3 illustrate these concepts by comparing results obtained by the
Euler and midpoint Runge-Kutta methods with results from an iterative solution
method presented in the next section.
[&UMPLB U.2
Perform second-order elastic analyses of the steel frame shown using
1. Euler method with d>. = O.S r1300

kips
2. Euler method with d>. = 0.25 30 kips t b c
3. Euler method with d>. = 0.1 -----.......
4. Work Control method
Properties for both members

L = 180 in
A= 15 in2
a __; I = 250 in4
E = 29,000 ksi
12.3 lncrementallterative Methods 345
Modeling column as two elements and beam as a single element Euler,d.l.= 0.1
Euler,d). = 0.25
\.
0.8
.g

.,, 0.6
.
.,,
Jl 0.4
:
<
0.2
10 20 30 40
Lateral displacement at point b (in.)
Using the frame shown in Example 12.2, perform second-order elastic analyses using
1. Euler method with d>.. = 0.25

2. Midpoint Runge-Kutta method with d>.. = 0.5
3. Work Control method
Modeling column as two elements and beams as a single element. Work Control
0.8
.g

.,, 0.6
.
] 0.4
:
<
0.2
Lateral Displacement at point b (in.)
12.3 INCREMENTAL-ITERATIVE METHODS
In contrast to the single-step schemes, the iterative methods need not use a single
representative stiffness in each load increment. Instead, increments are subdivided into
a number of steps, each of which is a cycle in an iterative process aimed at satisfying
the requirements of equilibrium to within a specified tolerance. Equation 12.3 is thus

modified to
m;
(12.10)
j=l
where m; is the number of iterative steps required in the ith load increment.2 In each
step j, the unknown displacements {d4.1) are found by solving the linear system of
equations
(12.11)
where (K1-1] is the stiffness evaluated using the deformed geometry and corresponding
element forces up to and including the previous iteration, and {R1-1} represents the
imbalance between the existing external and internal forces. This unbalanced load
vector can be calculated according to
(12.12)
where {P1-1} is the total external force applied and {F1-1} is a vector of net internal
forces produced by summing the existing element end forces at each global degree of
freedom. Similar to Equation 12.4a, the load applied in each step is determined by
(12.13)
where di..{ is now taken as the load ratio for the current iteration.
The methods presented in this section represent a sample of the different types of
multistep or iterative methods available; for a comprehensive review, the reader is
referred to Reference 12.1. All of these procedures will follow Equations 12.10-12.13.
In discussing these methods we will assume that the load ratio for the first step of the
ith increment, d>..{=1, has been prescribed according to either experience or an automatic
strategy (see Section 12.4). The difference between the procedures can then be defined
simply by the algorithm or constraint equation used to determine the iterative load
ratios d>..1""'2 for the remaining steps in an increment.
To provide a summary of the iterative methods, it is convenient to replace Equation
12.11 with the following two equations:
(K1-1]{d4.1} = /Preti (12.14a)
(K{-1]{d} = {W;-1) (12.14b)
Summing the solutions obtained in Equations 12.14a and 12.14b yields the displace
ment vector for each iteration
{d4.1} = dA1{d4.1) + {d} (12.15)
Application of Equations 12.12-12.15 is illustrated schematically in Figure 12.4.

The ith increment starts by assuming the unbalanced load {R?J = 0 and calculating the
displacements for the first iteration {d4.)} that correspond to an applied load dA.f/Prec)
and a tangent stiffness (K?], which is based on the results of the previous increment.
After updating the deformed geometry and element forces (see Section 12.5), the
(R!}. This load is then
second iteration begins by determining the unbalanced load
applied to the structure and the resulting displacements {d4.7} are determined using an
updated tangent stiffness [K)]. This stiffness [K!) is then used again to calculate the
displacements {d4.7} that correspond to an applied load !Prerl The iterative load ratio
2 A subscript will be used to indicate a particular increment and a superscript will represent an iterative step.
U.3 Incremental-Iterative Methods 347
I
{dii2;}
j [K\]
I dA.2;{dLi2;}
End of I
I I
---- -o
I
{R\}
{dP;}
- increment
{dli\l--1- {dli2;}
I {dLi3;}
'4----- {dli;} ------+<

Figure U.4 Schematic of iterations.
dAf is computed using one of the methods of the following subsections (Equations
12.16, 12.17, 12.19 or 12.21J.:..Jhe net displacements for the second iteration {daf} is
the sum of {dafl and dAf{dafl. The second cycle of iteration concludes by checking
convergence criteria (see Section 12.3.6). Necessary additional iterations are per
formed following the same procedure. The total load applied in the ith increment is
{dP;} = ( ) dA1 1Prer} and the resulting displacement is {d.'1;} = j {d.11}.
12.3.l Load Control Method
In the load control or Newton-Raphson method, a fixed amount of load is employed

in each increment. All of this load is applied in the first step j = 1 and additional
iterations j 2 are only performed in an attempt to satisfy equilibrium requirements.
In this regard, the iterative load ratio in Equation 12.15 is given by
d>..1 = 0 for j 2 (12.16)
This procedure is illustrated in Figure 12.Sa.

A significant shortcoming of this method becomes apparent when attempting to
solve problems with limit points. Once a fixed load is defined in the first step, there is
no way to modify the load vector should a limit point occur within the increment.
348 Chapter U Solution of Nonlinear Equlllb_rlum Equations
p
p
(a) Load control (Newton-Raphson) (b) Displacement control
p p
{M;) {M;)
(c) Work control (d) Arc length
Figure U.5 Iterative methods.
Although repeating the entire increment with a reduced initial load ratio can enable
one to approach the limit point slowly, the resulting near singular nature of the stiffness
matrix makes it difficult to trace the post-limit state response of a structure.
U.3.2 Displacement Control Methods
In a tradit!<,,ial displacement control method, the load ratio in the first step of an
increment is defineri so that a particular "key" displacement component will change
only by a prescribed amount. The load ratios for the remaining iterations are then
constrained so that this key displacement component will not change. By requiring
Equation 12.15 to be zero for a particular free degree of freedom, say, du, the iterative
load ratio is given by
dA{ = -du{1du1 for j i:!:: 2 (12.17)

where and
du7i are single elements in the solution vectors of Equation 12.14. The
du{
method is illustrated in Figure 12.Sb.
Clearly, the disadvantage of this method is having to select the key displacement
component; in some cr.ses it may be obvious but in others not. One simple technique
12.3 Incremental-Iterative Methods 349
is to use whichever degree of freedom incurred the absolute largest change during the
first iteration of the current increment. Other more sophisticated techniques for over
coming this and other problems associated with the displacement control methods are
available in the literature (see Refs. 12.2 and 12.3).
12.3.3 Work Control Method

As its name implies, the work control method uses botr; loads and displacements. In
this case, the constraint condition requires a zero increent in external work for each
equilibrium iteration. This is represented by
dW{ = {dPW{dA.{} = O for j;:::: 2 (12.18)

By substituting Equations 12.13 and 12.15 into Equation 12.18, the following iterative
load ratio is obtained:
-{Pref}T{d.:i{}
d }./ = for j;:::: 2 (12.19)
1 /Prer}T{dA.{}
The work control method is illustrated in Figure 12.5c. Both this method and the
displacement control schemes are suitable for calculating post-limit state response (see
Ref. 12.3).
12.3.4 Constant Arc Length Methods

As with the work control procedure, the arc length methods iterate neither at fixed
load nor at fixed displacement. One of the many constant arc length methods currently
available is based on defining and further constraining an arbitrary arc length ds in
each equilibrium iteration according to
for j;:::: 2 (12.20)

Substitution of Equation 12.15 into this constraint equation results in an orthogonality
equation that can be used to calculate the following iterative load ratio:
. - {dA_ 1 }T{dA_/j
dA = -1-
I
for j ;:::: 2 (12.21)

I
{dA.) }T{dA.1} + dAf

A schematic of the constant arc length method is presented in Figure 12.5d. In many
cases, arc length procedures can solve problems that not only exhibit limit point be
havior, but also snapthrough and snapback response (see Ref. 12.4).
12.3.5 Modified Iterative Technique

In many cases the efficiency of the iterative methods can be enhanced by replacing
the stiffness matrix [K{-1] used in each iteration (see Eqs. 12.11 and 12.14) with the
tangent stiffness [K7] from the initial step of the increment. Although this modified
iterative technique3 will typically require more steps per load increment (see Figure
12.6), the additional effort is often offset by the substantial savings realized as a result
of not having to assemble and decompose a new global stiffness matrix during each
iteration. It should be noted, however, that the use of this technique may be inadequate
"This method is sometimes called a modified Newton-Raphson approach. We choose not to use this descrip
tion since the technique can be applied to solutions schemes other than the Newton-Raphson method.
End of
Actual Equilibrium
Path
r
I Start of ith
increment
r 1M1 1
1
{M 21
,
IMrl
f-
--
{M;) ---+1
Figure U.6 Modified iterative approach.
for the analysis of structures that exhibit extreme deformations or significant stiffening
behavior as a result of large internal tensile forces.
12.3.6 Convergence Criteria

We now tum our attention to the determination of the appropriate number of itera
tions needed to satisfy the requirements of equilibrium. Given that the deformations
in each iteration are at the heart of any equilibrium requirement, it seems reasonable
to base convergence criteria on these calculated values. In Reference 12.5, three al
ternative norms are suggested for testing convergence in this way:
Modified absolute norm:
1 N
llell (12.22)
N
=
t-1
Modified Euclidean norm:
llell =
(12.23)
Maximum norm:
llell (12.24)
U.4 Automatic Load Incrementation 351
In these equations, N is the total number of unknown displacement compo ents, d!J.k
is the kth element of the incremental displacemeni vector {d.:i{}, and dependmg on the
units of d!J.k> the value of 6.ref is taken as either the largest component of translation
or rotation within the total displacement vector {.:i;}.
Using any of these norms, a convergence criterion of
(12.25)
llell s t
can be used where the acceptable tolerance tis usually on the order of 10-2 to 10-6,
depending on the desired accuracy. As an alternative to comparing displacements,
reasonable equilibrium criteria may also be based on assessing the unbalanced load
vector (Ref. 12.6) or by studying the increment of internal work (Ref. 12.4).
12.4 AUTOMATIC LOAD INCREMENTATION
The size of the load ratio employed in each increment of the analysis can have a
dramatic effect on the solution. In the single-step methods, proper selection of dA is
the only means for controlling drift-off error. In the iterative schemes, a poor definition
of dJ\.l could result in the solution not converging within a practical number of itera
tions. In both methods, an excessively small load ratio may result in significant com
putational effort with a negligible increase in accuracy. To provide assistance in de
termining dAJ, there are several types of automated procedures that may be employed.
Two of these schemes &re presented. The first may be applied to both the single-step
and iterative methods. The second is intended only for iterative procedures. In both
schemes, it is assumed that the size of the load ratio for the first increment dA1 or the
first iteration of the first increment dA has been prescribed by the analyst. As stated
earlier, a value of 10%-20% of the anticipated maximum applied load is generally
satisfactory. Section 12.6 also provides load ratio size constraints that are required
specifically for a material nonlinear analysis based on the plastic hinge method.
12.4.1 Change in Stiffness
The load ratio dA at any point in the analysis should reflect the corresponding state of
nonlinearity. A simple scalar measure of the degree of nonlinearity can be obtained
from a current stiffness parameter given by (Ref. 12.7)
s = {dir!Prerl
(12.26)
1 {d.:il}T{Pref}
Since the parameter S; will always have an initial value of unity, stiffening or softening
of.., the structural system will be indicated by values greater than or less than one,
respectively. With the exception of bifurcation points, S; will become zero at a limit
point. For a more sophisticated and perhaps more rigorous measure of nonlinearity,
Reference 12.3 provides details on a generalized stiffness parameter.
Using either Equation 12.26 or the stiffness parameter of Reference 12.3, load ratios
for a single-step method or the first cycle of an iterative method can be obtained from
(12.27)
where d>..1
is the load ratio prescribed at the start of the analysis, and the exponent y
typically equals0.5 to 1 (Ref. 12.1). Strategies for selecting the appropriate sign in
Equation 12.27 will be discussed in Section 12.7.
U.4.2 Number of Iterations
?
In t
?
ultistep ethods t e number of iter tions in each increme
nt needed to satisfy

eqmhbnum reqmrements is usually proportional to the
size of the initial load ratio
dAf. In this regard, the following parameter can be employed:
Nd
r/l;-1 =
-
(12.28)
N;-1
where Nd is the number of iterations the analyst desires for convergence and N;_1 is
the number of iterations required for convergence in the previous or i-1 increment.
Using this parameter, the initial load ratio is obtained by
(12.29)
where the exponent 'Y typically ranges from 0.5 to 1.
Because the values provided in Equations 12.26 and 12.28 may result in initial load
ratios significantly larger than 1, it is always desirable to prescribe an absolute maxi
mum limit on the initial load step. It should be noted that other similar but more
sophisticated schemes based on fixed increments of displacement, external work, or
arc-length are available in the literature (see Ref. 12.1).
U.5 ELEMENT RESULT CALCULATIONS
U.5.1 Updating Deformed Geometry
To employ the stiffness relations of Chapters 8 and 9 in conjunction with the incre
mental analysis procedures presented in Sections 12.1-12.3 requires that the structural
geometry include all accumulated deformations. For moderate displacement problems,
this can be achieved by completing each step of the single-step or iterative method
with a process of coordinate updating. That is, the coordinates of each node are mod
ified or updated to include the translational displacement components that occurred
during the step.
In updating the coordinates of nodes or element ends, we are representing the de
formed geometry of the structure by changing the position and hence the orientation
of each element with respect to the global coordinate system. This repositioning should
also account for any changes in the element's web orientation angle (see Example 5.5).
The web orientation may change in each step and for elements subjected to twisting
it will vary along the element length. For moderate displacement problems this effect
may be approximated satisfactorily by the average of the end rotations taken about
the element's local x-axis, (d&x1 + d&x2)/2. In all cases the element is assumed to be
straight at the start of the next step or iterative cycle.
U.5.2 Force Recovery
In linear analysis, element forces are conventionally calculated by substitution of dis

placements obtained from the global analysis in the element stiffness equation, Equa
tion 3.11, which for present purposes may be written
{F) = [k]{a) (12.30)
The components of {&) may be in either the global axes directions or transformed to
the element's local axes. In the first case [k) must be expressed in global coordinates,
{F) will be ,likewise, and a further transformation will be required to convert {F) to the
12.5 Element Result Calculations 353
vector of axial forces, end shears, and moments that is usually needed for design. In
the second case, [k] must be in local coordinates and it follows that {F] will be also.
We may illustrate the extension of this conventional procedure to nonlinear analysis
by considering the element of Figure 12.7. The first sketch, Figure 12.7a, illustrates the
element's orientation and the forces acting on it at the start of a linear step of a single
step or iterative procedure. Figure 12.7b shows its orientation at the end of the step
with the corresponding forces referred to the initial local axes. Figure 12.7c is of the
(c)
(b)
(a)
Figure 12.7 Element end forces.
same state with the forces transformed to the new local axes. This is the form needed
for ue in design as ell as for updating the tangent stiffness matrix. To simplify the
otatwn we have omitted all subscripts and superscripts except those needed to iden
tify the ends of the element and its orientation at the beginning and end of a linear
step. In discussing this figure the following vectors will be used:
1. The forces at the start of the step:
{1F} L1Fa iva iMa l b ivb IMbJT (12.31a)

=
p
2. The increment of forces:
{dF} = LdFa dV0 dMa dFb dVb dMbjT (12.31b)
3. The forces at the end of the step referred to the initial configuration:
(12.31c)
4. The forces at the end of the step referred to the final configuration:
{2F} = L2Fa 2Va 2Ma 2Fb 2Vb 2MbJT (12.31d)
5. The displacement increments transformed to the initial configuration:
(12.31e)
Following Equation 8.2, we have {dF) = [k1]{d.i) in which [k1] is the element tangent
stiffness matrix in the local coordinates at the start of the step. Thus for nonlinear
elastic analysis Equation 12.31c becomes !TFJ = {1F) + [ke + kg]{d.i) where the tangent
stiffness matrix has been separated into its elastic and geometric components, both
calculated at the start of the step. {2F) may be obtained from gF) by the transformation
(12.32)
in which [Tr] is the element transformation matrix4 (Ref. 12.8). Use of Equation 12.32
is illustrated in Example 12.4.
EXAMPLE 12.4
Given the geometry and force information shown, use Equation 12.32
to calculate the element's forces at the end of the current step.
Forces at start of step:,
4Rather than constructing (Tr] directly, it is generally more convenient to calculate it as the product of a
transformation from the initial local coordinates to the global system and one from the global system to the
final local coOidinates. Recognizing the orthogonality of the transformation matrices, it can be demonstrated
readily that the resuit is
{2F} = [;r][!rY{tFJ (12.33)
In these transformation matrices the subscript x refers to the global system.
U.6 Plastic Hinge Constraints 355
Increment of element forces: {dF;} = L2 5 100 -2 -5 400 Y

Forces at end of step referred to final configuration:
2Fa
{1F} 500 -8 -20 15ooy

=
L8 20
{TF) = {1F) + {dF} = L10 25 600 -10 -25 19oo y
[ '' 15 sin 15 O] [ 0.9M 0.259
with 1</> = 15, [''Y] = -si 15 cos 15 0 = -0.259 0.966

0 1 0 0 ]
[ 0.940 0.342
with 2</> = 20, (2-y] = ;
-0 42 0.940
0
From Equations 12.32 and 12.33
10 12.14
25 24.03
600 600
{2F} =
(Tr){IF} =
(2r]('r)T -10 -12.14
-25 -24.03
1900 1900
This method of force recovery is straightforward in application and conventional in

the sense that it is a direct extension of the standard linear elastic procedure. It will
be recognized, however, that there is nothing in it to distinguish between displacements
resulting from rigid body motion and those due to deformation: It is an approximate
approach to force recovery. Methods that do distinguish between the two sources of
displacement are discussed in Appendix B and the limitations of this conventional one
are explored. It is shown that, for small strain, moderate displacement analysis, the
rigid body motion effects are normally small. For this reason, and because it has been
found adequate in many examples of both elastic and inelastic nonlinear analysis,
conventional force recovery in conjunction with a single step method is considered
suitable for general framework analysis. In cases of highly nonlinear structures, an
iterative analysis method and the force recovery procedure defined in Appendix B are
recommended.
12.6 PLASTIC HINGE CONSTRAINTS
For the material analysis model developed in Section 10.2, the load ratio is typically
reduced to prevent plastic hinge formation from occurring within a load increment,
and thereby avoid any accompanying abrupt changes in stiffness. To calculate a load
ratio that results in plastic hinges forming only at the end of an increment requires a
comparison of the total element forces at the end of each step with the plastic hinge
criterion employed (e.g., the yield surface of Equation 10.18). If no hinges form, the
solution scheme continues. In some cases, as illustrated in Figure 12.8, one or more
sets of element end forces will breach the yield surface and signify the formation of
plastic hinges at a fraction T of the current load ratio. If it is the first step of the
increment, the load ratio should be reset to the product of the minimum of these
fractions Tmin and the current load ratio. In situations where hinges are detected in
analyses that follow the first step, it is suggested that the entire increment be repeated
using an appropriately reduced load ratio.
Calculation of the fraction T for a given set of element end forces reduces to a
relatively simple rootfinding problem. For example, Equation 10.7 can be rewritten as
<P( p + T dp, m + T dm) - 1 = 0 (12.34)
where p, dp, m, and dm are the known quantities of nondimensionalized forces and
moments shown in Figure 12.8, and Tis the unknown root. To ensure proper conver
gence, a rootfinding method that always brackets the root should be used. The method
of false position, or regula falsi, is recommended. Starting with initial guesses of T1 0
=
and Tu = 1, this technique provides an improved estimate of the root Tr according to
[CfJ(Tu) - l](T1 - Tu)

(12.35)
c"fJ( T1) - c"fJ( Tu)
where, for example, <P( Tu) = <P ( p + Tudp, m + Tudm). Using Equation 12.35, successive
iterations would continue with Tr replacing whichever of the two guesses, T1 or Tu,
provides a function value <P( T) - 1 with the same sign as <P( Tr) - l. The use of Equa-

tions 12.34 and 12.35"is illustrated in Example 12.5.
p = PIPy
1.0 m = MIMP
Figure 12.8 Plastic hinge formation at fraction T of attempted

load increment.
U.6 Plastic Hinge Constraints 357
EXAMPLE 12.S
Given the non-dimensionalized quantities p = 0.3, dp = 0.05, m = and dm - 0.1 use O.,
. .
Equation 12.35 to calculate the fraction 'T so that Equation 12.34 1s satisfied to four decimal
places. Assume <t>(p, m) = p2 + m2 + 3.5p2m2.
First iteration:
Ti = O <t>(Ti) = <l>(p + T1dp, m + T1dm) = 0.9316

'Tu = 1 <l>(Tu) = <l>(p + 'Tudp, m + 'Tudm) = 1.2798
. 1 (1.2798 - 1)(0 - 1)
., =

From Equation 12.35, -

- i - 0 1964

(0.9316 _ 1.2798)
<t>( T,) = <t>(p + T,dp, m + T,dm) = 0.9935
Second iteration:
<t>('T1) - 1 = -0.0684 <l>('Tu) - 1 = 0.2798
<t>('T,) - 1 = -0.0065
since <t>( T, ) - 1 and <t>( T1) - 1 are both negative, let Ti = T,
'Tt = 0.1964, <t>( 'T1) = 0.9935
'Tu = 1.000, <l>('Tu) = 1.2798
(1.2798 - 1)(0.1964 -
'T, = 1 - 1) = 0.2146
(0.9935 - 1.2798)
<t>('T,) = 0.9994
Third iteration:
<t>( 'T1) - 1 = -0.0065 <l>('Tu) - 1 = 0.2798
<t>('T,) - 1 = .-0.0006
since <t>( T,) - 1 and <t>( T1) - 1 are both negative, let T1 = T,
(1.2798 - 1)(0.2146 - 1)
'T, = 1 - = 0.2163
(0.9994 - 1.2798)
<t>( T,) = 0.9999 After 3 iterations, T= 0.2163 satisfies Equation 12.34.
Another situation in a material nonlinear analysis that may require a reduction in

the size of the load ratio occurs when plastic hinges exist at the start of an increment.
The use of a plastic reduction matrix will constrain the element forces at a plastic hinge
, to move tangent to the yield surface. If the surface is curved, these forces will not
remain on the surface with a change in applied load and, as a result, may exceed a
tolerable amount of yield surface drift (see Figure 12.9, forces move from point A to
B). To avoid this it may be necessary to once again reduce the current load ratio by a
--. --. Maximum Tolerable Yield Surface, '

l.O
', B
'
'
'
'
'

1.0
m =MIMP
Figure 12.9 Controlling yield surface drift.
factor of 7, so that the increment in forces move from point A to C. This can be
achieved by first isotropically expanding the original theoretical yield surface <l> by a
prescribed amount to obtain a maximum tolerable surface <l>'. The above root-finding
strategy can then be employed on this expanded surface <l>' to find the permissible
fraction 7. Note that a yield surface can be isotropically expanded by artificially in
creasing the yield strength of the material.
Obviously, for the analysis to continue the forces residing on the tolerable yield
surface must be returned to the original yield surface. For practical structures, these
small deviations in force can probably be neglected without a significant impact on
overall results. In any case, the drifted. force point should be returned to point D on
the yield surface via a path that is approximately normal to the surface (see Fig. 12.9).
More sophisticated one-step and iterative procedures for circumventing problems of
yield surface drift are found in References 12.9 and 12.10.
In addition to reaching the yield surface or moving tangent to it, the distribution of
element forces may change under subsequent applied loading to the extent that one
or more sections with plastic hinges may unload elastically (see Section 10.2.1). In this
event it is suggested that the entire load increment be repeated using elastic properties
for these sections.
U.7 LIMIT POINT AND POST-LIMIT ANALYSIS
In Section 8.1.2 a physical description of a stability limit point was given as the point
at which a system's capacity for resistance to additional load is exhausted and continued
deformation results in a decrease in load-resisting capacity (see Figure 12.10). It may
be characterized numerically as a point in the analysis at which the global stiffness
U.8 Limit Point and Post-Limit Analysis 359
p
B
-.
/Limit Point
Stable Unstable, Post-Limit

Region -'--Region
Figure U.10 Limit point and tracing
post-limit response.
matrix ceases to be positive definite.5 Limit points can be detected by the presence of
one or more nonpositive coefficients on the main diagonal of the stiffness matrix fol
lowing a Gauss or a root-free Cholesky decomposition (see Section 11.2).
Once a limit point is reached, several methods are available for continuing the anal
ysis into the post-limit region. A simple approach, given in Reference 12.11, is to
continue using a positive initial load ratio dAJ and obtain a solution to the nonpositive
definite system by performing the standard decomposition and substitution steps pro
vided in Section 11.2. As shown in Figure 12.10, applying this action in the post-limit
region results in moving from point A to B. Reversing the sign of the load increment
and corresponding displacements correctly shifts the results from point A to C. This
process is repeated until either the analysis is complete or the stiffness matrix returns
to being positive definite. If the latter occurs, this stiffening behavior is represented by
reversing the sign of the initial load ratio back to positive.
Note that the global stiffness matrix will be singular at the theoretical limit point,
thereby making it impossible to continue directly into tracing post-limit state behavior.
Although a zero-stiffness limit point is rarely encountered in the incremental methods
commonly employed, the approach to it indicates that the numerical accuracy of the
solution may diminish in the neighborhood of a limit point because of ill-conditioning.
Techniques for detecting, controlling, and correcting for this situation were presented
in Section 11.5.
As indicated in Figure 12.11, a bifurcation state can occur prior to reaching a smooth
limit point. In these situations the response of the structure will branch off from its
initial stable mode to a different stable or unstable mode with little to no numerical
forewarning. Detection and treatment of bifurcation points is discussed in Reference
12.12. However, beyond recognizing this possibility, it is stressed that few structural
systems display ideal bifurcation. The effect of inevitable imperfections in geometry,
material, and loading greatly reduce the probability of its occurrence. As illustrated in
several of the examples of Chapters 9 and 10, reasonable simulation of actual behavior
can generally be obtained by a nonlinear computational analysis that incorporates
small imperfections within the model.
'The global stiffness matrix may also become singular when all elements that are connected at a common
node have formed a plastic hinge at this node. To prevent this local joint instability from controlling the
overall limit point of the structure, an elastic element with a relatively small stiffness is often used in parallel
to any elements with yielded ends.
Bifurcation
Point
Stable mode
Unstable mode
6 Figure 12.11 Bifurcation state.
12.8 CRITICAL LOAD ANALYSIS-AN EIGENPROBLEM
In Chapter 8 general concepts regarding the stability of structural systems were intro
duced. As shown in Chapters 9 and 10 and Section 12.7, the limit point or point of
1.;:utral equilibrium that usually identifies the transition from a stable to an unstable
structure can be determined from an incremental analysis. Mathematically, this point
may be characterized by Equation 11.50 when the condition number of the tangent
stiffness matrix becomes infinite and small changes to the load vector will result in
unbounded changes to the displacement vector. That is, the current system of equilib
rium equations is singular and displacement vectors other than the null vector satisfy
[K1.f!H..1.r} = {O} (12.36)

in which [K1,ff] is the tangent stiffness matrix and the subscript ff indicates that only
the free degrees of freedom are included.
As an alternative to performing a complete incremental analysis, insight into the
stability of many structural systems can be obtained by neglecting material nonlinear
behavior (a qualification that will be modified below) and assuming that the relative
distribution of internal force is the same at all ratios of applied load. Combining these
simplifications with the fact that all of the element geometric stiffness matrices are
linear functions of their end forces (see Eqs. 9.17-9.19) permits Equation 12.36 to be
rewritten as
(12.37)
or
(12.38)
where [ Kg.ff]
- is calculated from element forces that were obtained from a linear
elastic analysis for a known reference load {Pref}, and A represents the ratio of the
elastic critical load to this reference load.
Equation 12.38 is the general form of an eigenvalue problem. In most cases it is
easier to solve this type of problem than to perform an incremental analysis. Three
basic approaches for solving the problem as it relates to a stability or critical
load analysis are provided in the following sections. The first two are based on con
version of the general eigenproblem of Equation 12.38 to one of the standard form
[H]{Y} = w{Y}. The third method, which is deduced from concepts presented in the
first two, provides a solution scheme that does not require conversion.
Before proceeding, however, we note that the above process can be oified to
include some degree of material nonlinear behavior (see Section 10.). This will affct
Equation 12.38 in three ways; (1) [Kef . f] is replaced y [Ktf. f] reftectmg any reucton
in material stiffness at the critical load, (2) a nonlinear mcremental analysis usmg
[K ] is required to determine internal force distribution, and (3) the internal element
for1 i:s and hence [-K .ff] are no longer linear functions of the reference load {P rerl.
With these points in ind, the objective of an inelastic critical load analysis is to de
termine the minimum load ratio A that will satisfy the following equation with A = 1:
(12.39)
This can be done systematically by prescribing different values of A and performing
the associated eigenvalue analyses until the load ratio A in Equation 12.39 takes on a
value of unity. The minimum load ratio A meeting the requirements of Equation 12.39
represents the ratio of the inelastic critical load t<:> the reference load.
U.8.1 Reduction to Standard Form
The standard form of a matrix eigenvalue problem is
[H]{Y} = w{Y} (12.40)

where [H] is a known n x n matrix, w is an unknown constant, and {Y} is a vector with
n unknowns. The objective in solving an eigenvalue problem is to go beyond the trivial
solution {Y} = {OJ and determine those pairs of eigenvalues w; and eigenvectors {Y;)
that also satisfy Equation 12.40. There will be in general n different eigenpairs that
satisfy this equation. In all cases, eigenvectors can only be determined to within a scalar
multiple since both {Y;) and c{Y;) satisfy Equation 12.40, where c is a scalar quantity.
It is important to note that a symmetric matrix [H] will always yield real eigenvalues
and real orthogonal eigenvectors. This fact considerably simplifies the computational
details associated with the more practical methods for solving eigenvalue problems.
With this introduction, these concepts will be applied specifically to critical load
analysis. We begin by converting Equation 12.38 to standard form. One way of doing
this is to premultiply both of its sides by the inverse of [Kef. f] yielding
-I 1
[Keff] [-Kg,ff]{.dfj =A {.dfj (12.41)

Thus, this equation takes on the form of Equation 1.40 with [H] = [Ke.ff]-1 [-Kg.ff]
and w = 1/A. Unfortunately, the product [Kef. fr I [-Kg.ff] will not in general be sym
metric. Since from a computational point of view a symmetric coefficient matrix [H]
is desirable, an alternative conversion is presented.
Considering that [Kef. f] is always positive definite, it can be factored according to
the Cholesky method (see Section 11.2.2) as
(12.42)
Noting also that [L - Ir = [L Tr I and [L]- I [L] = [L Tr I [L]T = [I] where [I] is the
identity matrix, Equation 12.38 is rewritten as
(12.43)
where w= 1/A. Equation 12.43 can then be expressed in the standard form of Equation
12.40
[H]IYJ = w{YJ (12.40)
where
[HJ= [Lr1[-Kg,JJ][UJ-l (12.44a)

{Y} = [Lf{Atl (12.44b)
Since [Kg.ffJ is sym etri , it can be shown that [HJ will
also be symmetric, and we
have met a essential ob1ective .
? of converting Equation 12.38 to standard form. Ex
ample 12.6 illustrates this process.
EXAMPLE li.6
For the structural system shown, prepare the eigenproblem in standard form.
3000kN
10-,l- 200,:.1520xx000106
m
A=
I=
E=
Io' mm'
mm4
MP'
,-;
Using the degree-of-freedom labels provided
[K ff] =
.
144
[ 0
0
0
0.36
1800
0
1800
1.2 x 107
]

12 0
From Equation 11.15: [L] = [ 0.6
3000 173J
]
0.0833
From Equation 12.46: [Lt1 = [ 0 167
0 -2.89 5.77 x 10-4
With Pref= 1000 kN, Fx2 = -1000 kN
From Equation 9.14: [Kg.ff] = [-.l -i12

0 -100
-00
-1.33 x 106
]

6.944 x 10-4 0
From Equatio1112.44a: [HJ= [Lt1[-Kg.ff][L-1]T = [ 0.333 -0481]

-0.481 1.111
Standard form of eigenproblem: [H]{Y} = w{Y}
The eigenvalues w calculated from Equations 12.43 and 12.40 are identical. The
reciprocals of these values l/w represent the load ratios A that satisfy equation 12.38.
Therefore, the reciprocal of the largest eigenvalue calculated from Equation 12.40 is
equal to the smallest A, that is, to the critical load ratio for the structural system. The
corresponding eigenvectors {Y} obtained by solving Equation 12.44 are related to the
buckled configurations or modes {A1) of the structure according to
(12.45)
U.8 Critical Load Analysis-An Eigenproblem 363
Note that the Cholesky decomposition given in Equatio

n 12.42 i one _ of t e ost
-
. This step c n be avo ded if e m1tially
time-consuming parts of this conversion process _
perform the same decomposition as part o the linear elastic
a alys1s tha is reqmred
f
_
g Equation 11.15 to
to obtain forces for the calculation of [-Kg.ff] After employm
from
obtain [L], each successive ith row of [Lr1 can be calculated
r;;1 = 111;;
(12.46)
for j = i - 1, i - 2, ..., 1
ljj
We now proceed by presenting popular methods for solving standard and general
forms of eigenvalue problems.
12.8.2 Polynomial Expansion

A formal solution for the eigenpairs can be obtained by rewriting Equation 12.40 in
the form
[H - wl]{Y} = {OJ (12.47)

where [I] is an n X n identity matrix. A nontrivial solution can exist if and only if the
matrix [H - wl] is singular; that is, the determinant of the coefficient matrix is zero:
IH - wII = 0 (12.48)
Expansion of this determinant yields a polynomial of degree n in the unknown scalar
w, where n is the order of the coefficient matrix. Solution of this characteristic equation
then produces n roots which are the eigenvalues. For each eigenvalue w;, an associated
eigenvector {Y;] can be obtained by assuming a value of unity for one of its elements,
usually the nth element, and then using Equation 12.47 to solve for its remaining
elements. Example 12.7 illustrates the use of polynomial expansion.
JEX'A.MPLE 12.1J
For the column and results provided in Example 12.6, use polynomial expansion to determine
the critical buckling load and corresponding buckled shape.
[H - wl] =
[6.944 x
0
10-4 - w 0
0.333 - w
0
-0.481
]
0 -0.481 1.111 - w
IH - wll = (6.944 X 10-4 - w)[(0.333 - w) X (1.111 - w ) - (-0.481)2] = O
= (6.944 X 10-4 - w)[w2 - 1.444w + 0.1386] = 0
The roots of this characteristic equation are
w, = 6.944 10-4
x

1.444 Vl.4442 - (4 o.1386)

+ x
Wi = = 1.341 ( maximum )
2
1.444 - Vl.4442 - (4 x 0.1386)
W3 = = 0.1036
2
Maximum w corresponds to minimum critical load ratio
Am;n = 1/Wmax = 0.746

APcer 746 2El
Pc, = = kN which is within 1.5% of the theoretical value P."' -rr = 740 kN
4L2
=
The corresponding buckled shape is obtained from

12.47.
[-1.-03399 -1.00076 -0.810 {
Equation
[H wl]{Y) = {O}
0 -0.4810 -0.2296] 1 {0} y3

=
After performing Gauss Elimination ( see Section 11.2.1) this system of equations reduces
[-l.3399 -1.0076 -0.0810 Y1

to
]{:} {0} =
0,
fH - wlf
Y, 1,
As expected, this system of equations is singular, and we will need to assume
{-0.:774}
=
v.luo '' ono domont of {Yj. WHh bcl< rnb,.itution yiold' {Yj
Buckled shape:
12.45, [Li-1 12.6 Per= 746 kN
[0.00833 1.067 00 ]T -0.4774 -3.687

From Equation with computed in Example
l.Omm
-1.56 1 o-4 rads
0 -2.89 5.77 10-4 { 1 } {5.77 10-4}

x
Q Q
{.11) =
x x
=
which may be expressed to within a scalar multiple as
U.8.3 Power Method

Polynomial expansion is only recommended for situations where the order of [H] is
very low (n < 5). Typically, this is not the case in critical load analyses of practical
structures. Also, it is generally true that only one or a few of the largest eigenvalues
and associated eigenvectors of Equation 12.40 are sought. For these reasons more
efficient numerical schemes are desired.
The power method is often the simplest numerical algorithm to use when solving
eigenproblems of standard form. It further suits our needs since it always provides the
dominant ')r absolute maximum eigenvalue and its companion eigenvector. This iter
ative scheme st<>.rt with an initial guess \Y0} as the dominant eigenvector. This esti
mate, typically a vector of ones, is then substituted into the left side of Equation 12.40
yielding
[H] (Y0} = (Y1} .

(12 49)
A better estimate \Y1} of the dominant eigenvector is obtained by normalizing, or

dividing all elements of \Y1} by llY1 lie (see Eq. 11.38). A first approximation w1 of the
dominant eigenvalue is given by
(12.50)
Using {Y1 }, the procedure is then repeated in the next iteration to define the dominant
eigenpair more precisely.
Iterations continue until a solution of acceptable accuracy is obtained. In this regard,
the following convergence criterion based on successive approximations of the desired
eigenvalue is recommended:
(12.51)
where
(12.52)
and tis a prescribed percent error tolerance (see Eq. ll.20b). Example 12.8 illustrates
the power method and the use of Equations 12.49 and 12.52.
rEXAMPLE 11.8]
Using the results of Example 12.6, use the power method to calculate the.column's critical load
ratio and buckled configuration. Use ( 0.1%.
=
(Y1} [H](Yo}
{ -6.944 x 10-4
-0.1478
}
{ }
= =
-0.6300
0.0011
(Y1} = (Y1J!llY11i. = (Y1} / 0.6471 = -0.2284
0.9736
and
W1 = (Y1}T[H]{Y1} = 1.2845
Results of the second iteration:
(Y2J = L7.453 x 10-1 -0.5445 1.1916f

llY2ll. = 1.3102
(Y2} = L5.6884 x 10-7 -0.4156 0.9095f
w2 = 1.3405
and
e.
1 1.3405 - 1.2845
100%
1 4.18% 1: ( 0.1%
1.3405
= = =
After a third iteration:
(Y3} = L2.9465 x 10-10 -0.4297 0.9030f

w3 = 1.3409
e. = 0.03% < ( 0.1% =
Critical load ratio:

366 Chapter U Solution of Nonlinear Equilibrium Equat
ions
Buckled configuration:
{ }
From Equation 12.45
Per= 746 kN
2.455 x 10-11
{.11) = [L -IJT{Y3} -3.3228 1.0mm
10-4 rads
=
-1.57 x
5.2132 x 10-4
which may be expressed to within a scalar multiple
as
{.if) = cl0 1.0 -1.57 X 10-4JT
In some caes, the power method is slow to converge. One approach for accelerating
.
this process is to use one of the relaxation schemes presented in Section 11.3.1. A
better estimate of the current normalized eigenvector can be obtained by
(12.53)
where the relaxation factor f3 has a value between 0 and 2. If successive iterations are
found to result in oscillating estimates of an eigenvalue, underrelaxation (0 < f3 < 1)
should improve the convergence rate. Overrelaxation (1 < f3 :s; 2) will provide similar
benefits when successive estimates of this eigenvalue tend to change in a monotonic
fashion.
For some structural systems, the first few critical load ratios and their corresponding
buckled shapes are often desired. After calculating the dominant eigenpair w1 and
{Yi}, subsequent eigenpairs can be found by employing a deflation scheme. In this
approach [H] is transformed into a different coefficient matrix, say [H2], which has the
same eigenpairs as the original coefficient matrix [H] except that the current dominant
eigenvalue becomes zero. Thus, the second largest eigenvalue w2 and the associated
eigenvector {Y2} of [H] can be obtained by using the power method to find the dom
inant eigenpair of [H2]. The process is then repeated until the desired number of
eigenpairs of [H] is found. An example of a deflation scheme is Hotelling's method
given by
(12.54)
11t should be noted that in all cases the buckled configurations {i11} corresponding to
the calculated eigenvectors {Y} are obtained by employing Equation 12.45.
The Sturm sequence property provides a useful and convenient means for confirm
ing the calculation of multiple eigenvalues. The property is based on performing an
[L][D][LY factorization (see Eq. ll.16)of [H - wl] for a given value of w. The number
of negative elements in[D] will equal the number of eigenvalues less than w. Examples
12.9 and 12.10 illustrate Hotelling's deflation method and the Sturm sequence property,
respectively.
Using the results of Examples 12.6 and 12.8, employ Hotelling's deflation and the power method
to determine the column's next sub-dominant critical load ratio and buckled configuration. Use
c = 0.1%
U.8 Critical Load Analysis-An Eigenproblem '367
From Equation 12.54 and using
[H2] = [H]
_
w and {Y }
wHYmYW
= 10-3 x
[ 0.6944
0
0
85.71
39 18 ]
{YW(Yi} 39.18 17.88
0
Using [H2] in place of [H] in Equations 12.49 to 12.52, the power method converges after
two iterations
w = 0.1036
{Yl = L3.3892 x 10-s o.9095 0.4156l
1
w produces a critical load ratio of A2 = 2 = 9.6511 or
W2
and a buckled configuration of

Pcr.2 = A2Pref = 9651 kN
{a[.2} = [L -1){Y} =
{ 2.8244 x 10-6
0.3160
} Pcr;2 = 9651 kN
2.3997 x 10-4 7.59 x 10-4 rads
which may be expressed to within a scalar multiple as

l.Omm
{af.2 } = cLO 1.0 7.59 x 10-4y
Note that Pcr.z = 9651 kN is approximately one and half times the theoretical solution of
9-i2EI
per,2= 2 = 6662 kN. By subdividing the column into more elements, a more accurate
4L
solution can be obtained. For example, using two elements provides Pcr,2= 6884 kN which
is within 3.3% of the theoretical solution.
Using the results of Example 12.6, apply the Sturm sequence property to show that [H] has two
[
eigenvalues that are smaller than w = 0.5
]
6.944 x 10-4
w = 0.5,
[
With [HJ= 0 333 -0 481 and
]
-0.481 1.111
-0.4993
[H - wl] =
-0. 667 -0. 811
-0.4811 0.6100
From Equation 11.17,
[
[H - wlJ= [L][D][LjT
[
LL]
1 0 O -0.4993
J
0
with [LJ= 0 1 0 and [DJ= . 0 -0.1667
0 2.8868 1 0 0
The number of negative elements in [DJ is two, which correctly indicates that [HJ has two
eigenvalues that are less than w = 0.5. These two eigenvalues were obtained in Example 12.7
as w1 =6.944 x 10-4 and w3 = 0.1036.
12.8.4 Inverse Iteration
The simplicity and efficiency of the power method can be offset by the necessity of
having to convert the eigenproblem to standard form. This, coupled with the iact that
the handedness of [Ke.ff] and [-Kg.ff] is not carried through in [HJ, suggests the need
for an iterative algorithm that can be used to solve the eigenproblem in generalized
form.
One such method is inverse iteration. It is also well suited for the eigenproblem
defined by Equation 12.38 because it converges on the eigenvector that corresponds
to the minimum eigenvalue or critical load ratio A. The method begins by assuming
an initial estimate {.:iJJ of the controlling eigenvector, typically a vector of ones, and
then calculating an auxiliary vector {y1 J according to
(12.55)
An improved estimate {.1}) of the eigenvector is obtained by first solving the system
of equations
(12.56)
and then normalizing the solution {.i}J according to
(12.57)
An estimate of the corresponding eigenvalue can be obtained from
AI = {.:i}JT[Ke.ff]{.1}} (12.58)
Similar to the power method, the procedure is repeated using {at 1} as the initial
estimate for the ith iteration. The scheme continues until the convergence criterion
defined in Equations 12.51 and 12.52 is satisfied. Example 12.11 illustrates the use of
the inverse iteration method to solve the problem defined in Example 12.6.
Employ inverse iteration to compute the critical load ratio and buckled configuration for the
column defined in Example 12.6. Use (= 1 %.
Uing an initial "timato of

{:}
{&j) tho "'ul" of tho firnt iterntion "'
[0.o 1 0.012 1000 ]{11} { 100.0.1102 }

{y1) = [-K11.ff]{a7(= =
0 100 1.33 106 1 1.334 106 x x
Solve
[1440 0.036 18000 ] { 100.0.1102 }, { -1.6.914410 10-4103 }

[K ff]{l}) = {y1}
.
ti}) = {i}} =
x
x
0 1800 1.2 107 1.334 106

x x 0.2776
Normalizing according to Equation 12.57:
{af1} =
{l.598-2.55310-6}
{a})
A
=
x
V1lW[-Kg.ff]{i}J
6.386 10-4 x
A first estimate of the minimum eigenvalue is
Results of the second iteration:

=
{y2J Ll.598 x 10-6 -0.2425 596.161f
=
{lJ] L uo9 x 10-9 -3.689 6.o3o x 10-4f
=
1a7i L8.578 x 10-10 -2.852 4.662 x 10-4r
=
A2 0.750
and according to Equation 12.51
e =
a
1 0.750 - 1.371
0.150
1 100% = 82.8% <t (
=
1%
Results of third iteration:

=
{y3J L8.578 x 10-11 -0.2956 336.43f
=
1l}1 L5.957 x 10-13 -3.846 6.049 x 10-4r
=
1a}1 L4.444 x 10-3 -2.869 4.512 x 10-4r
A3 = 0.746
and
e
a
= 1 0.746 - 0.750
0.746
1 '.10%
=
0.54% < ( =
1%
=
With a critical load ratio of A =
0.746, Per APref = 746 kN and the buckled configuration Per= 746 kN
{A}) expressed to within a scalar multiple is
= l.Omm
{AJJ clO LO -1.57 x 10-4y -1.57 x 10-4 rads
Finally, it should be noted that there are other more complex but possibly more
efficient solution techniques available for solving eigenvalue problems. In general,
methods for solving eigenproblems can be classified in one of four groups: vector
iteration approaches such as the power and inverse iteration methods, transformation
strategies, polynomial iteration techniques, and Sturm sequence-based methods. Since
there are no direct methods for determining the roots of a polynomial of degree 5 or
higher, all schemes are iterative. Several of them employ variations of the iterative
methods as well as the acceleration and deflation schemes presented in this section.
However, there is no single algorithm that is best suited for all eigenvalue problems.
Reference 12.13 provides a review and criteria for selecting appropriate numerical
algorithms for solving eigenvalue problems related to structural engineering.
U.9 PROBLEMS

U.1 For the nonli ear spring stif ness provided, use three
increments of the following
methods to determme the extens10n of the spring a
when subjected to a force of
P=3
(a) Euler method
(b) Midpoint Runge-Kutta method
(c) Compare both solutions to the exact solution, a = (P + 2)3
- 8
3(P 2)2
1
k=
+
.:l

-::: I P
-
Problem U.1
U.2 Repeat Problem 12.l for k = lfeP with an exact solution of a =eP 1.
U.3 Repeat Problem 12.l for
[2 ( ) -i
J
3 3P
k= cosh
T
with an exact solution of a = sinh(3P/2).
U.4 Member ab is a rigid bar connected to its base by a nonlinear rotational spring
with stiffness
(106 - M2)2 in. kips

k=
106 + M2 rad
Using three increments of the following methods, determine the vertical displacement
at b for P= 8 kips and Lab =10 ft.
(a) Euler method
(b) Midpoint Runge-Kutta method
(c) Compare both solutions to the exact solution,
M
=
(} 106 - M2
br
a---e
Problem U.4
U.5 Use the Euler method to calculate the second-order elastic response of the can
tilevered beam show (adapted from Ref. 12.14). Start with an increment size of
dA =0.5 and repeat analyses with the initial increment size reduced by one-half until
there is a negligible change in the load-displacement response. Confirm your solution
with the work control method.
A = 10 x 104 mm2
0.4 kN
I= 10 x HP mm4
E= lOOMPa
"------
i
----1 m --------i
Problem U.5
12.9 Problems 371
12.6 Repeat the convergence study of Problem 12.5 using the midpoint Runge-Kutta
method. Compare the results.
12.7 Repeat Problems 12.5 and 12.6 for the frame shown (adapted from Ref. 12.15).
i 20 kN
r
0
H---+0+o>---- .96m --1
0.24m
l.2m
A =600mm2
l
/=2x 104mm 4
E = 72,000 MPa
0
Problem 12.7
12.8 Use the Euler method to calculate the second-order inelastic response of the
frame shown. Start with an increment size of dA = 0.5 and repeat analyses with the
initial increment size reduced by one-half until there is a negligible change the load
displacement response.
1
I
v
0.05 4t_______ ,..._
W18x35
0
15'
V = 365.4 kips
J
E =29,000 ksi
cr = 36 ksi
y
0
I 15---
Problem 12.8
12.9 Repeat the convergence study of Problem 12.8 using the midpoint Runge-Kutta
method. Compare the results.
12.10 Repeat Example 12.5 using p = 0.6, dp = 0.1, m = -0.5, and dm = -0.08.
......_
12.11 Given that the nondimensionalized quantities p = 0.378 and m = 0.756 satisfy
a yield surface defined by <l>(p, m) = p2 + m2 + 3.5p2m2 1, determine the fraction =
'T that must be multiplied by increments dp = 0.1 and dm = 0.2 in order to satisfy a
yield surface drift tolerance of 0.01. In this case, the maximum tolerable yield sur
face is
<l>'(p m)
' =
_ 1 _ p2
1.012
( + m2 +
p2m2
1.012
) = 1
12.12 Using two elements to model each of the columns shown, prepare the eigen
problem in standard form and solve for the elastic critical load ratio and corresponding
buckled shape by (1) polynomial expansion and (2) the power method with?= 0.1 %.
Assume a compressive axial force of Pref = 500 kips and let E = 29,000 ksi,
I= 1200 in.4, and L 29 ft. Neglect axial deformation and compare with exact
solution
=
.
provided.
_ 4i2EI
Per-
L2
(a) (b) (c)
p _ 2.li2EI
er - L2
(d) (e)
Problem 12.12
12.13 Using the results of part (2) of Problem 12.12, employ Hotelling's deflation
and the power method to determine the next subdominant critical load ratio and buck
led configuration. Use t = 0.1 %.
12.14 Show that the dominant and subdominant eigenvectors obtained for each col
umn in Problems 12.12 and 12.13, respectively, are orthogonal, {Yi}T{Y2} = 0.
12.15 Apply the Sturm sequence property to the results of Problem 12.13 to
confirm that the two eigenvalues calculated for each column are both less than
w = 1.05/max(Ai. A2).
12.16 For the columns shown in Problem 12.12, use inverse iteration to compute the
critical load ratios and buckled configurations. Use t = 0.1 %.
12.17 Repeat Problem 12.16 using an estimate for the initial eigenvector that is more
reasonable than a vector of ones.
References 373
corresponding
U.18 For the two-bar truss shown, calculate the critical load ratio and
buckled configuration by employin g the following methods with?= 0.1 %:
(a) Reduction to standard form and the power method
(b) Inverse iteration
(c) Structural analysis program
ipref = 500 kN
E = 200,000 MPa
A = 2500mm2
lOm

Problem 12.18
U.19 Repeat Example 9.10 by employing two elements and the following methods
to calculate the beam's critical load ratio and corresponding buckled configuration.
Use?= 0.1%:
U.20 Repeat Problem 12.19 using one element to model the beam defined in
Example 9.11.
U.21 Repeat Problem 9.16 by employing three elements and the following methods
to calculate the column's critical load ratio and corresponding buckled configuration.
Neglect axial deformation and use?= 0.1%.
REFERENCES
U.1 M. J. Clarke and G. J. Hancock, "A Study of Incremental-Iterative Strategies

for Nonlinear Analyses," Intl. JI. Num. Meth in Engr., Vol. 29, 1990, pp. 1365-
1391.
U.2 0. C. Zienkiewicz, "Incremental Displacement in Nonlinear Analysis," Intl. JI.
Num. Meth. in Engr., Vol. 3, 1971, pp. 587-592.
U.3 Y-B. Yang and S-R. Kuo, Theory and Analysis of Nonlinear Framed Structures,
U.4 M. A. Crisfield, Non-linear Finite Element Analysis of Solids and Structures,
Volume 1: Essentials, John Wiley and Sons, New York, 1991.
U.5 P. G. Bergan and R. W. Clough, "Convergence Criteria for Iterative Processes,"
A/AA JI, Vol. 10, 8, August, 1972, pp. 1107-1108.
U.6 K. J. Bathe, Finite Element Procedures, Prentice-Hall, Englewood Cliffs, N.J.,
1996.
U.7 P. G. Bergan, G. Horrigmoe, B. Krakeland, and T. H. Sreide, "Solution Tech
niques for Nonlinear Finite Element Problems," Intl. JI. Num. Meth. in Engr.,
Vol. 12, 1978, pp. 1677-1696.
U.8 C. I. Pesquera, Integrated Analysis and Design of Steel Frames with Interactive
Computer Graphics, Ph.D. Dissertation, Cornell University, Ithaca, N.Y., 1984.
U.9 J. G. Orbison, W. McGuire, and J. F. Abel, "Yield Surface Applications in
Nonlinear Steel Frame Analysis," Computer Meth. in Appl. Mech. and Engr.,
Vol. 33, 1982, pp. 557-573.
U.10 R. D. Krieg and D. B. Krieg, "Accuracies of Numerical Solution Methods for
the Elastic-Perfectly Plastic Model," JI. of Pressure Vessel Tech., Trans. ASME,
November, 1977, pp. 510-515.
ll.11 P. G. Bergan, "Automated Incremental-Iterative Solution Schemes," Numerical
Methods for Nonlinear Problems, Proceedings of the International Conference,
Swansea, September, 1980.
U.12 M. A. Crisfield, Non-linear Finite Element Analysis of Solids and Structures,
Volume 2: Advanced Topics, John Wiley and Sons, New York, 1997.
U.13 K.-J. Bathe and E. L. Wilson, Numerical Methods in Finite Element Analysis,
U.14 K. Mattiasson, "Numerical Results from Large Deflection Beam and Frame
Problems Analysed by Means of Elliptic Integrals," Intl. JI. Num. Meth. in Engr.,
Vol. 17, No. 1, 1981.
U.15 S. L. Lee, F. S. Manuel, and E. C. Rossow, "Large Deflection and Stability of
Elastic Frames" JI. Engr. Mech. Div., ASCE, Vol. 94, No. EM2, April, 1968.
Chapter l3
Special Analysis Procedures
In earlier chapters many of the basic principles of structural mechanics were reviewed
and the fundamentals of the stiffness method were developed. Now we wish to look
at some of the techniques used in coping with complications encountered in the prac
tical application of this method. The complications can usually be associated with one
or more of the following factors: problem size, solution efficiency, and solution accu
racy. Most real structures have a large number of degrees of freedom. A structure may
be so large that an analysis can be achieved most effectively if it is broken down into
parts, with each part analyzed separately and the results combined in a way that sat
isfies all of the equilibrium, compatibility, and boundary conditions. More often, the
analyst may adopt this approach because the structure lends itself to it naturally. The
second factor, solution efficiency, refers to the relative cost, in terms of programming
and computer expenditures, of the many different ways of ordering the numerical
operations leading from the input of data to the output of the solution. Solution ac
curacy refers to the relative precision of these different methods and to ways in which
errors may be analyzed and corrected.
This chapter deals with some of the methods for treating problems of large size. It
also covers a few logically or mathematically related techniques that are useful in the
treatment of some special but frequently encountered problems. Most of the methods
discussed are strictly valid for linear elastic analysis only, but some are unrestricted.
Additional ways for treating large problems and the subj ects of efficient and accurate
solution of equations were treated in Chapters 11 and 12.
One way to handle large structures is to use a coarse structural idealization, that is,
to disregard, suppress, or approximate the effect of degrees of freedom that, in the
opinion of the analyst, have only a minor bearing on the result. The many different
ways in which this may be done are so dependent upon the individual structure that
they cannot be discussed usefully in a general text. Here we present some schemes for
reducing the order of the systems of equations that have to be solved at any one time
once the structure has been idealized. This means that generally we will be discussing
methods for reducing the order of the stiffness matrices to be inverted. We shall first
discuss matrix condensation as a general technique. Then we shall show how it is em
ployed as part of a routine-the substructuring meth od-for breaking the entire struc
ture into parts, analyzing each part in turn, and putting the results together in the
proper way.
After this, we shall look at some of the practically important features of support
conditions and coordinate types, connection stiffness and joint size, and the application
of the principles of symmetry and antisymmetry. It is a mixed group of topics. Some
are related mathematically, but all have the merit of either contributing to the effi-
375
376 Chapter 13 Special Analysis Procedures
ciency of an analysis or to the realism of the structural model. We shall also consider
the problem of revising structural properties in the light of the results of preliminary
analysis, an essential task in the normal iterative analysis and design process.
13.1 CONDENSATION
The term condensation refers to the contraction in size of a system of equations by

elimination of certain degrees of freedom. The condensed equations are expressed in
terms of preselected degrees of freedom !Ac} that, together with the eliminated quan
{Ab}, comprise the total original set of degrees of freedom, that is, LAJ = LAb : AJ
tities
Mathematically, the process takes the original set of equations, [K]IA} = {P}, which
may be expressed in P artitioned form as
.
(13.1)
and condenses them to the form:
(13.2)
To reduce Equation 13.1 to Equation 13.2, first expand the upper partition and solve
for {Ab} Thus
(13.3)
Substituting this value of {Ab} in the expanded lower partition of Equation 13.1 yields
(13.4)
Letting
(13.5)
and
(13.6)
Equation 13.4 becomes identical to Equation 13.2.
One way of using the condensation procedure is immediately obvious. If Equations
13.l are stiffness equations in which LP J = LPb : PcJ is a known load vector, one may
solve for the unknown displacements in two steps. First, from Equation 13.2,
(13.7)
Second, by substituting the result in Equation 13.3,
Ii\.,) = [Kbbr1IPb) - [Kbb]-1[KbcHKccr1!Pcl (13.8)

Although it can be shown that the two-stage solution always requires a greater
number of arithmetical operations than does the direct solution of the unpartitioned
form of Equation 13.1, it has the virtue of enabling the solution to be effected through
the inversion of matrices [K bb] and [Kee], which are each of lower order than the
original unpartitioned matrix [K). This can be important in treating problems of very
large size or in using limited computer-core storage.
The concepts of condensation have already been used although they were not iden
tified by that term. For instance, Examples 4.9 and 4.14 contain applications of con
densation techniques. Example 13.1 is a further illustration of condensation.
From the results of Example

condensation.
4.3, remove the degree of freedom 0,2 by
EI,
.,.______ __,
L ____
Solving for LM,2 Fy2J from Example 4.3,

{M,2 } 2EI, [ 2
= -3/L
2
]{ 0,2 }
Fy2 L 3/L - 6/L V2
From Equation 13.5,
From Equation 13.6,
From Equation 13.2,

[3/,J{V2 { Fy2
L3
} = + M,2
2 L
} or
It is of interest to observe that condensation of a stiffness matrix means satisfaction

of the equilibrium equations corresponding to the elements eliminated. Also, it should
be clear that the "eliminated" degrees of freedom are not discarded. They are ex
pressed as functions of the corresponding forces, the remaining degrees of freedom,
and the coefficients of the equations, with the result substituted in the original equa
tions (see Equations 13.3 and 13.4). This process is the matrix analogue of traditional
Gauss elimination for the solution of simultaneous linear algebraic equations (Section
11.2). Nothing is "lost" or "approximated" in the process of condensation. The effect
of the forces that correspond to the eliminated degrees of freedom (the forces {Pb})
on the remainder of the system is incorporated in the force vector {Pc} of Equation
13.6. In Equations 13.7 and 13.8 all of the elem<!nts contributing to the stiffness of the
system an:: properly represented. Finally, selection of the degrees of freedom to be
eliminated (the displacement components {.db}) is at the analyst's discretion. Usually,
however, the selection is not arbitrary; there are generally logical reasons for choosing
certain degrees of freedom for elimination.
13.2 SUBSTRUCTURING
When, for one of the reasons given in the introduction to this chapter, it is either
impossible or undesirable to analyze a structure in its entirety in one stage, it may be
analyzed by the substructuring method. In this method, major components of the struc
ture, called substructures, are first analyzed separately and the results are then com
bined. The mathematics of substructuring follow directly from the equations of
condensation.
378 Chapter 13 Special Anadysis Procedures
'fYpical point in the

interior of a substructure
p--P---1P---1
(a)
Interface (c) nodes

( \ / /
' '
Internal (b) nodes
,.
I ----!'+-----
'
,.
I '-+-----'-'-+----....._,
'
(b)
Figure 13.1 Substructure analysis. (a) Substructure analysis conditions
(applied load not shown). (b) Typical interior and interface nodes
substructure N.
Figure 13.la shows a structure partitioned into three major substructures M, N, and
0. There is no physical significance to the fact that the interface members are shown
heavier. They are just drawn that way to emphasize the divisions we have made for
purposes of analysis. Consider first the stiffness properties of the Nth substructure. The
following subscripts are employed:
b-free degrees of freedom internal to a substructure, that is, not associated with
any other substructure.
c-free degrees of freedom located on an interface between two substructures.
Figure 13.lb illustrates the use of these subscripts in substructure N. The stiffness
equations for this substructure can be written in the partitioned form:
(13.9)
Observing that, except for the addition of superscripts, Equation 13.9 is identical to
Equation 13.1, we may use the results of Section 13.1 to write the equations that effect
the elimination of the interior nodes of substructure N. From Equation 13.2,
[K]{dJ = {PJ (13.10)

where, from Equation 13.5,
[K] = [K] - [K][Kbr1[K;%] (13.11)
It is useful to list the contributions to {P}. Using Equation 13.6 to do this, Equation
13.10 becomes
[K]{A} =
{P} - [K][K;;b]-1{P;;}
To simplify the notation let
(13.12)
We then have
[K]{A} =
{P} - {R} (13.13)
We see that the condensed stiffness matrix [K] relates the interface degrees of
freedom {A} to the forces acting at the interface nodes {P} plus fictitious nodal loads
-{R}, which are equivalent to the real forces acting at the interior nodes (compare
with Equation 5.21). It should be clear from the development of Equation 13.13 that
we may treat the entire substructure N as a single element having the substructure
stiffness matrix [K] and that we may combine all such substructure stiffness matrices,
interface nodal forces, and equivalent nodal loads to form the stiffness equations for
the juncture degrees of freedom for the complete structure:
(13.14)
The rules of combination and manipulation of substructure equations are the same
as those developed in Chapters 3-5 for single members. In forming [Kee], rigid body
motion of the structure is prevented by taking into account the actual boundary con
ditions as, for example, the zero displacements at the bottoms of the columns of sub
structure 0 in Figure 13.1. The matrix [Kee] includes only the free degrees of freedom
and is therefore nonsingular. Care must be taken to avoid inadvertent duplication of
loads acting at interface nodes or stiffnesses of elements situated on the boundaries
between substructures.
Solution of Equation 13.14 yields ,all of the displacements, {Ac}, of the interface
nodes. Thus
(13.15)
Extracting from {Ac} the interface displacements for each of the substructures in
turn, we may then obtain the displacements of interior nodes from Equation 13.3.
Hence, for substructure N of Figure 13.1,
{A;;} = -[K;;br1[K/%]{A} + [K;;b]-1{P;;} (13.16)
Having all of the nodal-point displacements, member forces and reactions may be
found in the usual way.
In Example 13.2, substructuring is used in the analysis of a fixed-end beam subjected
to concentrated loads. The original system, which has six free nodal-point degrees of
freedom, is arbitrarily divided into two substructures, each of which has four free
degrees of freedom. By eliminating the interior degrees of freedom in each substruc
ture, stiffness equations having as unknowns the two common degrees of freedom are
formed and solved. Therefore, instead of solving the problem through the inversion
of one six-by-six matrix, we arrive at the same result after inverting three two-by-two
matrices. It may be observed that in the course of computing the equivalent nodal load
contribution of each substructure, we calculate [K;;br1{P;;). This is a displacement
vector that, when added to displacements resulting from movement of the interface
nodes, -[K;;1b]-1[K]{A}, yields the resultant displacement of the interior nodes of
each substructure (see Eq. 13.16).
EXAMPLE 13.2
r,
_
Analyze the fixed end beam shown using the substructure v -50 kN -30 kN -10 kN
! -i_-_
! -
method. Calculate the deflection and slope at each load point.
-
I/f---
----'I----<
E = 200,000 MPa.
_
_ _
x
_ l= 200x 106 mm4
------ 4@ 2 m = 8 m
The degrees of freedom, substructures, and element stiffness

matrix are as follows.
Substructure A Substructure B
From Equation 4.34, the segment stiffness matrix is
--
--
12 6 12 6
-
(2000)2 (2000) (2000)2 (2000)
r1 --
6
4 - 2
Mza = 200 X 2 X 108 (2000)
]
Fy b 2000 12
[
6
-
:
M,b Sym.
(2000)2 (2000)
1.5 x 10-6 1.5 x 10-3 -1.5 x 10-6 1.5 10-3
2 -1.5 x 10-3
4 x 107
1.5 x 10-6 -1.5 x 10-3
=
Sym.
2
Analysis of Substructure A. From Equation 13.9:
[
U2
7 0.1875 X 10-6
[KeAe] 4 X 10

-0. 375 X 10-J

=
From Equation 13.12, b][Ktbi-1/Pt }

{R} = [K
- -1.5 10-3
: ]{ } { : } 5
-
{R} = 4 X 107 [ : : =: =
-2. 104
Analysis of Substructure B. From Equation 13.9:
A
Uz
0.1875 x 10-6
B] = 4
[Kee X 107 [ 0.375 X 10-3
From Equation 13.12, {R: } = [K:b][K:bJ-1{P: }
-l.5 x 10-6 1.5 x 10-3 -h_ 5.0

{ReB} = 4 x 101 =
[ -1.5 x 10-3 1
]{ } { 0 0.5 x 104
}
Analysis of Assembled Substructures. From Equation 13.14 [Keell.ie} =/Pe} - {Rel Summing contributions of substructures
A and B,
1 0.375 X 10-6 0 -30 30.0 }

4 X 10 [ 0 2
]{ } { } {
u2
8 2
,
=
0
_
-2
x 104
Solving,
{ -2.54 mm }
=
-1.56 X 10-3 rad
and
-l.96mm
{aB} B]-l[KB]{aB}
b = -[Kbb be e + [KbbB]-l{PbB} = {U3} {
8
=
1.48 x 10-3 rad
}
,3
Deflected Structure. 4.0

In Example 13.3, the same method is applied to the solution of a three-story frame.
Neglecting axial deformations, there remain nine free degrees of freedom in this struc
ture. Substructuring is again arbitrary. Two substructures, each having six free degrees
of freedom, are identified. After condensation, stiffness equations in the three common
degrees of freedom are formed and solved. The similarity between this solution and
that of Example 13.2 should be apparent. So, too, should the similarity between the
fictitious nodal forces {Re} and the fixed end forces {PF} discussed in Sections 5.2 and
7.5. A substructure carrying loads on interior nodes is treated in the same way as a
single element loaded between nodal points. A single member is, in fact, a substructure
of a larger system. Multimember substructures may be combined with single-element
"substructures." Another way of saying the same thing is that, in condensation, we
have a scheme for reducing multi-element systems to equivalent single members.
In Example 13.3 we draw numerical values of stiffness coefficients from previous
examples, particularly Examples 5.9, 4.13, and 4.8. Figure 13.2 illustrates the relation
ships between the forces, reactions, and displacements we are dealing with. It also
-2.83x10-2
I
: 83 1

=
I
0.00 0
I
-;!). -,-
12.0 12.0 12.247 x 104 3.691 x 104
Typical elements TYpiciJ:l elements Elements of
of[KA] +200 of[K] +200 fPtl and {R}
(a) Analysis ofsubstructure A.
1$ I
12.0
12.0
0 0 0 0 0
TYpical elements TYpicl elements Elements of
of[K8] +200 of[K] +200 {Pf} and {R}
( b) Analysis ofsubstructure B.
4.396
-3.652
+4
x
-1.758
10-4
2kNm -2.510
0 0 0 0
Elementsof Elements of {6cl
fPcl {-Rel
and
(c) Analysis of assembled substructures.

Figure 13.2 Deflected structures for Example 13.3.
pictures the physical response of the various subdivisions as well as of the total assem
blage. Some of the elements of the [K] and [K] matrices of substructure A are shown
in Figure 13.2a. The effect of forces acting on the interior nodes-assuming that
the boundary nodes are restrained-is also shown. Substructure Bis similarly analyzed
in Figure 13.2b. Figure 13.2c shows the effect of the equivalent nodal forces on the
combined system. The resultant displacement components of each node and the mo
ment diagram, which is calculated from the member stiffness equations, are shown in
the example itself. It is suggested that the example be studied in conjunction with
Figure 13.2.
Examples 13.2 and 13.3 are, of course, merely illustrations of a technique. In practice,
such simple systems would be solved without substructuring. The method comes into
its own only in the analysis of large systems.
The relationships for a given substructure (Eqs. 13.1-13.6) are also useful in prelim
inary design studies. Such applications will be described in Section 13.7.
1-;;,::-.'11: --!'.i'-, Jt
J,_ "' - .'J!L,.._...
Analyze the rigid frame shown using the substructure method. Neglect axial de
E
6
formations. I of each beam is 200 x 10 mm4 and I of each column is 50 x 10 mm4.
200,000 MPa. Refer to Examples 5.9, 4.13, and 4.8 for numerical values of
6 r: _____ ---. 2
I
=
element and substructure stiffness influence coefficients. See also Figure 13.2 for 2kNm
Sm
physical interpretation of selected influence coefficients.
3
---- 4 i
2kNm I Sm
s
-----4 Ll
2kNm I Sm
1
-x,u
s_l
0 0
I--- 8 _____jm
Degrees of freedom and substructures. 3

1 2
2kNm\;.
3

0 Substructure A
_
3 4
2kNm\;_
3 4
3
k

s
___
_6
Substructure B
2kNm
7 8
Analysis of Substructure A.
{!-} [-!-!'s] {A}

Pc
=
Kcb 1 Kee Ac
3 0.00960 12.0 12.0 : -0.00960 12.0 12.0
I
2 x 103 1.4 X 1()5 0.5 X 105 : -12.0 0.2 X 105 O

0 1.4 X 105 : -12.0 0 0.2 X 105
=
200 ------------------------- ---------------------------
3 0.00960 -12.0 . -12.0

2 x 103 1.4 X 105 0.5 X 105 8
3
0 Sym. 1.4 X 105 8
4
[K] =
[K] - [Kb][Ktbi-1[Ktc] = 200 [ O
8
0
3
12.247 x 104
8
0
4
3.691 x 104
]
Sym. 12.247 x 104
(R} =
[
[Kb][Ktbi-1{Pt}
-0.0096
12.0
-12.0
0.2 X 105
-12.0
0
]{ 355.47
-6.076 x 10-3
} {
=
-3.00
4.144 x 103
}
12.0 0 0.2 X 105 -2.83 X 10-2 3.700 X 103
Analysis of Substructure B.
3 0.0192 0 0 : -0.00960 -12.0 -12.0

I
2 x 103 1.8 X 105 0.5 X 105 : 12.0 0.2 X 105 0

0 1.8 X 105 : 12.0 0 0.2 X 105 - 8_ 6 _
200
- - ---- - ------ - ------- - ------ - -- --------------------------
=
0 : 0.00960 12.0 12.0

I
0 Sym. I
I
0.4 X 105 0
I
0 I 0.4 x 105
'
8
4
4.961
-0.683 x 104
]
3.010 x 104
12.0
0
]{ 156.24
1.204 x 10-2
}{=
-1.396
-1.634 x 103
}
0.2 X 105 - 3. 344 X lQ-;-3 -1.942 X 103
Analysis of Assembled Substructures: [i ] { i1 }

tc c = {P l
e - {R }e
[ 0 00355 4.961 4.961

]{U3} { } { 3 -4.396 }
200 15.257 x 104 3.008 x 104 83 = 2 x 103 - 2.510 x 103
3
Sym. 15.257 x 104 84 0 1.758 x 10
u3
{} { } 11.36 mm
{a }
c = 83 = -3.141 x 10-4 rad
8 -3.652 x 10-4 rad
4
Deflected structure and moment diagram.

-1.57 x 10-4
Moments obtained from Equation 4.34.

Beams:
Columns:
13.3 CONSTRAINTS
We have used the term constraint to designate the suppression of a particular com
ponent of displacement. Now, we view the definition of constraint in the broad sense
as a set of relationships between degrees of freedom that are supplemental to the basic
stiffness relationships. Algebraically, a typical constraint equation-the ith such equa
tion in a given problem-will be written in the form:
(13.17)
where Gn, G;2, G;i, G;n, H; are specified coefficients and .11> '12, Llj, Lln are the degrees
of freedom of the structure. It should be noted that Equation 13.17 is a linear algebraic
equation and is thus applicable only to the treatment of linear problems.
In the simplest type of constraint, the familiar one of a single constrained degree of
freedom at a support point, all G;j = 0 except the coefficient that multiplies that degree
of freedom. The latter takes on the value 1.0. Also, H; = 0. For example, if degree of
freedom 2 is suppressed, we have G;2 = 1, and the constraint equation becomes
'12 = 0
Practical circumstances under which the general definition of a constraint given by

Equation 13.17 plays a role are illustrated in Figure 13.3. In Figure l3.3a, for example,
the column bases are assumed to settle in such a manner that their vertical displace
ments are related to one another as indicated. In Figure l3.3b we see that, because of
the relatively great axial stiffness of the horizontal members of the frame, it is accept
able to set all horizontal displacements at a given level equal to one another.
Not only does the general constraint condition involve more than one degree of
freedom, but a given structural analysis situation may feature more than one constraint
condition. Thus, in developing the analytical approach to the treatment of constraints,
we consider the case where r such conditions are present in a structure described by
n degrees of freedom. The r linear constraint conditions are, for this case,
[G]{A} = {HJ (13.18)

where [G] is an r x n matrix of the constraint equation coefficients, {A} is the list of
joint degrees of freedom of the structure, and {H} is a vector of known constants. For
simplicity, we deal only with the case {HJ = 0. Development of the more general case,
{H} * 0, is left as an exercise for the reader (Problem 13.11).
Now, each constraint equation can be viewed as a means to eliminate one of the
degrees of freedom in favor of the remaind'- r, simply by solving the constraint equation
for the former and using the result in the stiffness equations. Hence, with r constraint
conditions, it is possible to eliminate r degrees of freedom from the stiffness equations.
To accomplish this elimination we partition Equation 13.18 as follows:
[Ge : Ge] {1} = {O} (13.19)
where [Ge] is a solvable (nonsingular) r X r matrix and [Ge] is the r X ( n - r ) remainder

of the original matrix[G]. {Ae} and {Ac} are the corresponding degrees of freedom.
Solving Equation 13.19 for {Ae},
{Ae} =
-[Ger1[Gc]{Ac}
13 14 15 1" ''13 = U14 =U15 = Ul6

9 10 11 12 U9=U10=Utl =Utz

5 6 7 '8 U5 = U6 =U7 = Ug

u3 = 2u2
u4 3u2
=
8 0 0 0 0
(a) (b)
Figure 13.3 Representative constraint conditions.
or
(13.20)
with
(13.21)
To account for the constraint condition in the stiffness analysis, the stiffness equa
tions must be partitioned to separate the degrees of freedom {Ae} and {Ac}
(13.22)
and, by substitution of Equation 13.20,
\Pe) = [[Kee][rec] + [Kec]]{Ac} (13.22a)
\Pc} = [[KceHrec] + [Kcc]] {Ac} (13.22b)

,
The degrees of freedom to be eliminate / l Ae}, have been expressed in terms of the
degrees of freedom to be retained, {Ac}, by the transformation relationship, Equation
13.20. Equation 13.22b could be solved, as given, for {Ac}, and this solution could then
be inserted in Equation 13.20 to yield {Ael The matrix on the right side of Equation
13.22b is unsymmetric, however, so that the solution of this equation is relatively
expensive. To produce a symmetric set of equation coefficients and to establish rela
tionships between {Ac) and the counterpart forces, there must be a corresponding, or
contragredient, transformation of joint forces. This transformation is given by the
transpose of the matrix that transformed the displacements; that is (see Section 5.1),
(13.23)
We have placed the "hat" n over {Pc} to emphasize that it represents the transference
of forces{Pel to the location of the degrees of freedom c. Such forces supplement the
loads !Pel initially at these degrees of freedom. Hence, the total effective loads at
degree of freedom c, {Pc} are
(13.24)
and, by use of Equations 13.22,
{Pc} = \Pc} + [Cc]T{Pe}

(13.25)
= [[Kee] + [Kce][rec] + [rec]T[Kec] + [rec]T[Kee][rec]]{Ac)
Equation 13.25 discloses that if the transformation of degrees of freedom, [rec] can
be established from a set of constraint conditions, then its use in this equation will
result in a reduction of the degrees of freedom to be solved in the subsequent stiffness
analysis. This reduction is equal to the number of constraint conditions. The eliminated
displacements can subsequently be determined from Equation 13.20. Example 13.4
illustrates this.
The beam shown is constrained to displace in such a

manner that v2 = v3 and 82 = - 83 Calculate v3 and the
bending moments at points 2 and 3. EI = constant.
?
T e constraint conditions will be used to eliminate the v2 and 02 degrees of freedom. The global
.
stiffness equat10n of the structure, arranged so as to facilitate the elimination, is
The stipulated constraints can be written as
Solving for v2, 02,
Now, in accordance with Equation 13.25,
+
[ OJ [
1 -6 3L
2
] [
+
1 ]]{ : }
0 -1 -3L L 0 -
And, after forming the products and collecting terms,
{ Py3 + Py2 } =
2 / [ 12 6L
2
]{ }v3
Pm3 - Pm2 L 6L 6L 03
After inversion,
and, with pm3 pm2

=
0,
3
(Py3 + Py2)L
V3
=
12/
2
(Py3 + Py2)L
03
=
12/
13.4 JOINT COORDINATES

There are many important practical circumstances under which the coordinate axes
used to describe behavior at a joint in the global equations must be different from the
global axes. Such joint axes were mentioned in Section 2.2. We now describe some of
these circumstances and develop procedures to deal with them.
A typical situation is shown in Figure 13.4a. The behavior at joint i would usually
be described in terms of the x- andy-direction displacements u; and V;. Because of the
indicated support condition, however, it is necessary to impose a constraint in they"
" Y
l
\f3x
y,v
y , V/' V;
* *
ay x
*
X ,U;
ax
i x
x, u
(a)
Z, w
y,v
x, u
(b)
Figure 13.4 Joint coordinates.
direction, v'[ = 0. This support constraint cannot be implemented merely by setting

one global-direction degree of freedom to zero, thereby eliminating a single column
of the global stiffness matrix. But it can be accomplished by using u'[ and v'[ as degrees
of freedom in the global equations and then setting the single degree of freedom v'[
equal to zero.
Another situation is described in Figure 13.4b. This portrays a segment of a planar
truss that is a portion of a large space truss referred to global axes x-y-z. Local axes
" " "
x -y -z , differing from the global directions, are positioned at joint i of the planar truss
" "
with x and y in the plane of the truss. Clearly, the structure possesses no stiffness in
"
the z direction. If the stiffness equations are written with respect to the global axes,
"
this condition will not be apparent from inspection of such equations, since the x and
"
y stiffnesses have components in all of the global directions. What is desired is to have
a zero row and column in the global stiffness relationships, which can then be struck
out before solution of the equations. This can be accomplished if the joint axes are
used to describe, in the global equations, the behavior at joint i. The row and column
"
corresponding to stiffness in the z direction will then be zero and can be eliminated
directly.
The employment of joint coordinates requires a selective transformation of degrees
of freedom and corresponding forces. These transformations must follow the algebraic
procedures developed in Section 5.1.2 (Eqs. 5.12-5.16). Here we may view the process
as one of transforming from a set of equations written entirely in global coordinates:
{P } = [K]{d} (13.26)
to one that is written in a set of mixed coordinates, some global and some referred to
particular joints. We designate this set as
{P"} = [K"]{d"} (13.27)

A question of immediate significance is how [K] is transformed into [K"]. Following

Equation 5.16, which is completely general, we may write
{K"} = [rY[K][r] (13.28)
where, now, the transformation matrix may be defined as a square matrix having:
(1) unity on the diae;onals and zeros elsewhere in all rows and columns corresponding
to unaltered degrees of freedom; and (2) appropriate direction cosines in the positions
corresponding to all pairs of altered degrees of freedom (one global and one joint
degree of freedom).
The manner in which the global stiffness matrix is modified to accommodate joint
coordinates may be detailed by using the problem of Figure 13.4a as an example. The
direction angles linkir.g the global axes to the joint coordinate axes at node i are defined
in the figure. Using the notation of Section 5.1 and the coordinates of the points i and
j, the direction cosines of the x-y axes with respect to the x"-y" axes are Ix" my = =
(x; - xj)IL, mx" = -Ly = -(y; - yj)IL, where L V(x; - xY + (y; - yy. The
=
transformation matrix
(13.29)
Substituting this value of [r] in Equation 13.28, the transformation may be made.
The process is illustrated in Example 13.5.
The illustrated truss was analyzed in Examples 3.1 and 3.2 for
support conditions of a hinge at point c and a roller on a horizontal
plane at point b. Analyze the same structure for a hinge at point
c and a roller on an inclined plane at point b.
The global stiffness matrix for the unconstrained state of this structure was established in Ex
ample 3.1. For convenience, the boundary conditions at point Uc 0, are applied to the
]{ }
c, Ve
{} [
= =
matrix, since these conditions exist in both Example 3.2 and the present example and do not
affect the operations to be performed. Thus, from Example 3.1,
{P} = [K]{.d} Pxa 6.348 -1.912 - 3.536 3.536 Ua
P a 4.473 3.536 -3.536 Va

y 2
10
=
Pxb 6.830 -3.536 ub
P b Sym. 3.536 vb
y
{} [ ]{ }
Now, from the relationship between the above degrees of freedom and those pertaining to the
inclined plane,
(.1) = [r](.1"} Ua 1 0 0 0 Ua
Va _ 0 1 0 0 Va
-
ub 0 0 0.866 -0.500 ub
vb 0 0 0.500 0.866 vb
{} [ ]{ }
Apply the transformation to the global stiffness matrix:
(P} =
[K)[r](.1"} Pxa 6.348 -1.912 -1.294 4.830 Ua
Pya 2 -1.912 4.473 1.294 -4.830 Va
= 10
Pxb -3.536 3.536 4.147 -6.477 ub
Pyb 3.536 -3.536 -1.294 4.830 vb
{} [ ]{ }
Now, let (P"} = [r]T(P}, by Equation 5.14 ( or Equation 13.28):
(P"} = [rY[K][r](.1"} Pxa 6.348 -1.912 -1.294 4.830 Ua

Pya 2 -1.912 4.473 1.294 -4.830 Va
= 10
P';b -1.294 1.294 2.944 -3.194 ub
P'{,b 4.830 -4.830 -3.194 7.421 vb
{ } [ ]{ }
The remaining boundary condition, vb = 0, is now enforced. Thus
383.0 6.348 -1.912 - l.294 Ua

2
321.4 = 10 -1.912 4.473 1.294 Va
0 -1.294 1.294 2.944 ub
{} { }
Solving:
Ua 0.928
Va = 1.143 mm
ub -0.094
It may be demonstrated readily that it is not necessary to carry out the matrix mul
tiplication of Equation 13.28 in full, since only a limited number of rows and columns
are affected. The effects on the rows and columns of the global stiffness matrix are as
follows.
1. The column of the original global stiffness matrix multiplied by U; is multiplied

by Ix" and added to the product of ly" and the column multiplied by V;. The
.. resulting column vector, which corresponds to u';, replaces the column vector
corresponding to U;. A similar operation replaces V; with v';.
2. A new row in the Px; location is constructed by multiplying the row corresponding
to Pxi by Ix and the row corresponding to Py; by ly" and adding the two. A new
row in the Py; location is also formed by performance of corresponding opera-
tions. In following the sequence described, the rows operated upon must contain
u'I and ui' columns before products are formed.
These operations are illustrated schematically in Figure 13.5. It is suggested that the
reader use them to verify Example 13.5.
P,; -f - --
--+
1 I
P,, - --
(a) Original global stiffness matrix
(
This column obtained from This column obtained from
"
U; column times Ix plus U; column times mx plus
u, rol- '"" 1,. o, rolu= limo. m ,.

"
1 1 This row obtained from

Px; row times lx plus
-+-7- Py; row times I y
"
These terms contain
This row obtained from
::P"d::f
Px; row times mx plus
Py; row times my
(b) Modified global stiffness matrix
Figure 13.5 Replacement of global-coordinate displacements by

local-coordinate displacements.
The element stiffness matrices could alternatively be formed directly in terms of the
relevant local axes. This presents certain difficulties in input data assimilation because
the support conditions am referred to the joints rather than to the elements, but the
procedure is nevertheless concise and efficient. Formation of the global stiffness matrix
and all other operations then proceed in the usual way. Still another approach is to
treat Equations 13.26 and 13.27 as constraint conditions, in the manner described in
the Section 13.3.
One scheme that avoids the algebraic complexity of the preceding approaches is to
introduce a special boundary element, as shown in Figure 13.6. This element is then
Figure 13.6 Artificial supports.

assigned a value that is three or four orders of magnitude greater than the other stiff
ness coefficients.
13.5 CONNECTIONS AND JOINTS
Throughout the text we have emphasized that the analytical model is an idealization
that approximates the behavior of the actual structure. For example, members that
have depth and width have in all cases been reduced to line elements. The objective
of this section is to introduce basic concepts of modeling the behavior of connections.
Emphasis is placed on elementary methods for incorporating connection stiffness and
joint size within large-scale framework analyses.
13.5.1 Flexible Connections
With respect to stiffness, connections may be classified according to their moment

rotation relationships into one of three categories: ideally rigid, ideally pinned, and
semi-rigid (see Figure 13.7). These curves express the moment M transmitted through
the connection as a function of the relative rotation (}, between two connected mem
bers. In a rigid or fully restrained connection, no relative rotation occurs for the
moment resisted and the M-0, curve is a vertical line. For a pinned or unrestrained
connection, no moment develops for the relative rotation imposed and hence the
M
Fully Restrained
I
j k = constant
/ 1
/
/
/
/
/
/
/,
M M ,;
r--
I
I
I
frame I---. Undeformed
e,
I frame
I 1 (b) Classification of connections
(a) Definition of Mand 8,
Figure 13.7 Connection characteristics.

M-8, curve is a horizontal line. All intermediate conditions are, in principle, semi-rigid
or partially retrained. In practice, however, moderate departures from the ideals are
ignored, and most analyses are performed by assuming either rigid or pinned
connections.
Other systems for classifying connections that also include strength and ductility
attributes are available in the literature (see Refs. 13.1 and 13.2). The process of clas
sifying a particular connection and, in turn, determining its corresponding M-8, curve
is important and requires considerable judgment and knowledge of connection behav
ior. Since this process is not within the purview of this text, we will assume that the
classification has been made and we'll focus on implementation of the results.
One method for approximating the effect of connection flexibility within an analyt
ical model is to include a linear rotational spring at each end of a beam element (Figure
13.8). The stiffness k of these springs is the ratio of the transmitted moment to the
rotation within the connection. That is,
M = k8, (13.30)
in which, at the left and right ends of the element shown in Figure 13.8, Equation 13.30
is Mz1 = k1(8z1 - 8z;) and Mz2 = k2(8z2 - 8zi), respectively.. In this approximation, the
length of the connection is ignored and the spring stiffness k is generally taken as an
empirically determined initial stiffness of the connection (Figure 13.7b). By varying
the stiffness of the springs from k = 0 to k = oo, the above three general connection
categories can be defined. This approach is usually limited to linear analysis under
service loads.
By letting k1 = a1Elz/L and k2 = a2Elz/L, the first-order elastic force-displacement
relationship for a beam with flexible end-connections is
12 (
-2 1+
0'.1+ 0'.2 ) (1+)
L
-12
L
(1+ + ) (1+ )
2
0'.1
a1a2
0'.2
L a1
L a1a2 a2
-6 (
- 1+- )
{q
4 ( 1+ :J
r1
2
2
Mz1 aElz L a2
(13.31)
1 ( + ) -6 (
- 1+- )
=
Fy2 L 20'.1 0'.2

1+
Mz2 L a1a2 L a1 8z2
Sym.
(1+ :J 4
where a = a1a2/( a1a2 + 4a1 + 4a2 + 12). The development of Equatit"n 13.31 is
provided in the following two examples. In Example 13.6, the flexibility-stiffness trans
formations presented in Section 4.4 are employed and in Example 13.7 the same result
is obtained by the condensation method of Section 13.l.
Using Equation 4.25, develop the stiffness relationship for a beam with rotational springs at its
ends (M= k8) and bent about its z axis.
Modify the result of Example 4.3 to include the effects of the rotational
springs, the flexibility influence coefficients are determined as:
EI, EI,
With k1 = a1 and k2 a2 , assemble equations in matrix form
L L
=
{a}= (d]{F}:
r( )
1
Modifying the results of Example 4.4, [d]- and [cl>] are:
12 a1 + a2 -6 ( 2
J
1 aEI,
-2
L
1+ ---
a1a2
) --
L
1+-
a1
[d]- =
L -6
L
( 1+ 2
a1
) 4
( 1+
3
a1
)
1
{ } [ _OJ{ }
F1
y =
M,i L 1
F2
y
M,2
with [cl>]= -[ ]
Apply Equation 4.25:
-6
-
L
( 2
1+- )
az
Sym.
Show that Equation 13.31 can also be obtained by removing the internal degrees of freedom
(u;, 8,;, vi, 8,) through the condensation method of Section 13.1.
. EI, EI, .
With k1 = a1 and k2 = a2 , formulate stiffness matrix for each component:
L L
{ }
M,1
=
a1EI, [ _ 1 -1 ]{ }
8z1
and
f M,i } =
a2EI, [ 1 -1 ]{ }
8,j
M,; L 1 1 8,, LM,2 L -l 1 8,2
12 6 -12 6
2 2
L L L L
6 -6
C'} {:}
4 2
M,; EI, L L
=
Fyi L -12 -6 12 -6
2 2 8,1
M,i L L L L
6 -6
2 4
L L
Using the equilibrium requirements Fy1 = Fy; and Fy2 = FYi and the compatibility requirements
u1 = u; and u 2 = vi, assemble stiffness relationships for the entire system:
6 -6
8,;
M,; 4 + a1 2 -a1 0
L L
6 -6
-a2 8,j
M,i 2 4 + a2 0
L
I
I
L
------------- 1 -------------------
6 6 12 -12
EI,
I
Fy 1 2 0 2 0 U1
I
= L L
I
L L
L
I
I
-a1 a1 0 0 8,1
Mz1 0 0
I
I
I
-6 -6 :-12 12
Fy2 2 0 2 0 U2
L L L L
a2 8,2
M,2 0 -a2 0 0 0
To remove 8,; and 8,i, use Equation 13.5 with
-6
L
-6
L
and
12 -12
0 0
Lz Lz
EI, 0 a1 0 0
[Kee]= L -12 12
0 0
Lz Lz
0 0 0 a2
1
[Keel = [Keel - [KcbHKbbi- [Kbc]
12
L2
_- _ ( 1 1+ a
+ _a_ _ __ 2
a1a2
)
L
-
_6 ( + a2 )
1 2
aEI,
[Kccl = L
Sym.
with a = a1a2/( a1a2 + 4a1 + 4a2 + 12) this is the stiffness matrix of Equation 13.31.
To predict the strength limit state of frames with semi-rigid connections, nonlinear
M-8, relationships should be employed. One method for accomplishing this is to per
form an incremental analysis (see Sections 12.1-12.3) with equation 13.30 modified to
dM = k d8, (13.30a)
where k is the rotational spring stiffness for a particular load increment (see Figure
13.7b ). In this case, k would be a prescribed nonlinear function of either the total end
moment M or relative rotation 8, (see Example 8.7, Ma = k8a with k = (3/Wa). Note
that if the coefficients a1 and a2 are defined as functions that can be calibrated to
model the nonlinear connection stiffnesses k1 and k2, an incremental form of Equation
13.31 can be employed. It should also be noted that for a nonlinear analysis that
includes elements with linear or nonlinear rotational springs at their ends, the con
densation method of Section 13.1 may be employed. That is, the approach illustrated
in Example 13.7 could be used with the element's first-order elastic stiffness replaced
with its appropriate nonlinear stiffness matrix.
The special case of a pinned connection at one end of an element and a rigid con
nection at the other end can be derived from Equation 13.31. For example, substitution
of a1 = 0 to represent a pin at end 1 and a2 = oo for a rigid connection at end 2 results
m
3 -3 3
{U
2 0 2
L L
0 0
3 (13.32)
2
L
Sym. 3
Likewise, the stiffness relationship for a rigid connection at end 1 ( a1 = oo ) and a pin
at end 2 ( a2 = 0) is
3 3 -3
2 2 0
L L L
ri [:)
-3
M,1 EI 3 0
L
=
(13.33)
Fy2 L 3
M,2 2 0
L 0,2
Sym. 0
Equations 13.32 and 13.33 can also be obtained using the plastic hinge concepts
presented in Section 10.2. The stiffness matrix for an element with the bending moment
released at one or both ends may be obtained from
(13.34)
where [k;j] is the stiffness matrix for the unreleased flexural element and the matrix
[G] only contains nonzero values for the rotational degrees of freedom being released.
For example, the [G] matrix for a planar element subjected to axial force and flexure
with a pinned connection or, equivalently, zero moment at end 1 is
[G]T = [0 0 1 0 0 0] (13.35)
Likewise, for a three-dimensional element with both y- and z-moments released at
end 2, [G] is defined as
[G]T =
[ o o o o o o o o o o 1 o ] (l3.36)
0 0 0 0 0 0 0 0 0 0 0 1
A benefit of using Equation 13.34 to incorporate pinned connections is that the
stiffness matrix [k] can be generalized for nonlinear analysis by accounting for geo
metric and/or material nonlinear effects directly within [k;J as described in Chapters
8-10. Equation 13.34 can also be used to incorporate pinned connections in a linear
elastic analysis in which [k;J has been modified to account for shear deformations (see
Section7.6.2). It is left as an exercise for the reader to demonstrate that the use of
Equations 13.34 and 13.35 result in the form of Equation 13.32 (Problem 13.16).
13.5.2 Finite Joint Size
The joints in an actual structure are not mathematical points or nodes as assumed in
the typical analytical model. Three cases are illustrated in Figure l3.9a:
1. Structural joints are of a finite size that can be on the order of 5% to 10% of the
lengths of the intersecting members. This region or panel zone may represent a
structural component with a stiffness much greater than that of the connected
members.
2. In floor systems, it is common for the tops of beams and not their, centroidal axes

to share a common elevation.

3. For reasons related to connection geometry, the centroidal axis of a diagonal
member may not intersect the nodes representing its ends.
Other cases include situations where the shear center of the member does not pass
through its centroidal axis or when applied loads are eccentric to the nodes or the
centroids of members.
,__L
- L
l+----L,---+i
'r l+-
-
-
--L,---+i
- 1
e
Nodes
- - Centroid Axis a Offset points
- Rigid Jinks
- Elements
(a) (b)
Figure 13.9 (a ) Frame with moment-resisting connections. (b) Analysis model with rigid links
used to model panel zones.
All of the above suggest the occasional need to offset the ends of elements from the
arbitrary location of their defining nodes in the analytical model. As shown in Figure
l3.9b and in detail in Figure 13.lOa, one method for accomplishing this is to incorporate
rigid links at the ends of an element. Equilibrium of the free-body diagrams in Figure
13.lOb yields the relationship between forces at the ends of the element {F;j} and the
forces at the ends of the rigid links {Fu} as
'
{Fu} = [T]{Fij} (13.37)
yi
Mzl
FIC;
(a)
T
(b)
Figure 13.10 (a ) Offset element with rigid links. (b ) Free-body diagrams of
rigid links.
where the force transformation matrix [T] is given by
1 0 0 0 0 0
0 1 0 0 0 0
-e ex; 1 0 0 0
y;
[T] =
(13.38)
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 e yj -e 1
xj
Note that this matrix is not orthogonal, [T]-1 -4= [T]T. From the principle of virtual
work, the contragredient transformation ( see Section 5.1.3) for the displacements is
(13.39)
As illustrated in Example 13.8, these transformations can also be derived using the
constraint method of Section 13.3.
By assuming rigid links at the ends of element ij, develop six linear constraint
conditions in the form of Equation 13.20 with {A,} = (Ad and {Ac} = {A12}.
Show that these results may be used to develop a set of transformations equiv
alent to Equations 13.37 to 13.39.
With {A,} = (Ad_= Lu; V; O,; ui vi O,if

(Ac} = (A12} = Lui u, 0,1 u2 u2 0,2f
Write three constraint conditions for each rigid link.
Rigid link 1-i:
U; u, - [L1; cos c/Jli - L1; cos(cpli + o,,)]
1=
with Oz1 small, cos 0,1 1 and sin 0,1
= 0,1 =
L1;sin (4>!i+ e,1
u; = u1 - (Lli sin c/Jli) Oz1 -L1;sin ( 4>1;
R
V; = v1 + [Lli sin( c/J1; + Oz1) - Lli sin cp,;] /
/
U; = u, + (Li; cos c/J1;) 0,1 /
/
O,; = 0,1 /
/
L1; cosq>li -
L1;COS (4>1;+6z1)
2
Rigid link 2-j:
ui u2 + [L2i cos <P2i - L1i cos( <P2i +. 8,2)]
/
=
8,2 8,2 8,2 8,2 /

//
with s mall, cos = 1 and sin =
U2 + (L2i <P2) 8,2

ui = sin
/
I//
vi = v2 - [L2i sin( <P2i + 8zi) - L1i sin <P2J
vi = v2 - (L2i cos <P2) 8,2 d
L2i sin ( 4'2i + B,-})

8,i 8,2
-L2i sin ( 4'2i)
=
4'2j
In the form of Equation 13.20
U; 1 0 - L1; sin c/J1; 0 0 0 U1

V; 0 1 Lli cos cpli 0 0 0 V1
8,; 0 0 1 0 0 0 8,1
Ui 0 0 0 1 0 L1i sin c/J2i U2
vi 0 0 0 0 1 - L2i cos <P2i V2
8,i 0 0 0 0 0 1 8,2
forces are
From virtual work, the contragredient transformation for the
Fx1 1 0 0 0 0 0 ( Fxi
0
r
Fy1 0 1 0 0 0
M,1 -Lli sin c/J1; Lli co s c/J1; 1 0 0 0 M,;

Fx2 0 0 0 1 0 0 Fxi
Fy2 0 0 0 0 1 0 Fyi
M,2 0 0 0 L2i sin <P2i -L1i co s <P2j 1 M,i
With reference to Figure 13.lO(b) ex; Lli cos c/J1; exi L2i cos c/Jii
= L2i
=
=
eyi = Lli sin <Pli eYi sin 2;
tions.
and hence Equations 13.37 to 13.39 are equivalent to the above transforma
From Equations 13.37 and 13.39, the stiffness relationship in global coordinates for
an element with rigid offsets can be expressed by
( 1 3. 40)
where the local stiffness matrix [k;J and the rotational transformation matrix [r;i] are
both calculated using the length and orientation of the element as defined by offset
points i and j (see Fig. 13.lOa).
After nodal displacements have been calculated, use of Equation 13.39 yields the
end forces associated and aligned with the offset element
(13.41)
Another method for incorporating the effects of finite joint sizes within a framework
analysis is to represent the above rigid links by additional elements that are assigned
stiffness coefficients three or four orders of magnitude greater than those of the offset
elements. By employing this approach, Example 13.9 demonstrates that accounting for
this factor typically increases the predicted stiffness and strength of the structure.
[EXAMPLB J33'
For the frame shown, compare results of analyses that do and do 400 kips 400 kips
not include rigid links to model the panel zones.
W18x 35
100 kips ------------- -
I I
I I
I I
I I
I I
2' I W14x68 W14x68 I
I I
L
I I
I I
I I
l I
//, /////// ;., '7///fi $
----- 15' _____ ____.,
Analytical Models: 7.02'(typ.)

Without Rigid Links: With Rigid Links:
JI
b b c
c
--.t- k .,.
.-.1..... ..
I
_j 8.85"(typ.)
I
(4 nodes) 144 (8 nodes)
a d 1 a d
L-1so-I L-1so-I
Material All elements E = 29,000 ksi Elements ai, jk, dl E 29 x 103 ksi
=
a 36 ksi a 36 ksi
y y
Properties: = =
Elements ib, bj, kc, le E .29 x 106

= ksi
ay 36 x 103
= ksi
(increased by several orders of
magnitude to simulate rigid links)
Comparative Results:
(i) Elastic Analysis
Type Quantity Measured w/o Rigid Links w/ Rigid Links % Diff.
First-order Max. Lateral Disp. 0.630 in. 0.543 in. -16.0

Max. Column Moment 3869 in. kips (pt. a) 3603 in. kips (pt. a) -7.4
Max. Beam Moment 2232 in. kips (pt. b) 2302 in. kips (pt. j) +3.0
Second-order Max. Lateral Disp. 0.659 in. 0.564 in. -16.8

Max. Column Moment 4020 in. kips (pt. a) 3723 in. kips (pt. a ) -8.0
Max. Beam Moment 2338 in. kips (pt. b) 2394 in. kips (pt. k) +2.3
Eigenvalue Critical Load Ratio, A 21.15 24.58 +14.0
13.6 Symm etry and Antisymmetry 403
(ii) Second-order Inelastic Analyses

0.9
0.8 -04L0.81
0.7
M
...
0 0.6
u -- With rigid links
0.5
-0
"'
0.79 0.78
0
...J
0.4
M'
0.3
0.2 - - Without rigid links
0.1 0.76 0.75
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Lateral Displacement (in.)
Finally, it should be noted that panel zones are not necessarily rigid components.
Experimental and analytical studies have demonstrated that the high shear forces and
corresponding deformations in these zones can have a pronounced effect on the ulti
mate strength of a frame. In this regard, nonlinear models that treat the panel zone as
a subassemblage have been proposed. Reference 13.3 contains a comprehensive review
of such models.
13.6 SYMMETRY AND ANTISYMMETRY
Many bridges, buildings, aircraft, ships, and structures such as cooling towers and radio
telescopes possess some form of symmetry. Recognition of symmetry permits a com
plete structural analysis to be made by considering only a portion of the total structure:
!, 'or even less, depending on the type and degree of symmetry. Example 5.9 was an
elementary illustration of the application of the principles of symmetry. We now con
sider the subject in a somewhat more formal way.
There are three aspects to the problem: (1) recognition and definition of the type
of symmetry; (2) manipulation of the loads and forces in a way that enables one to
take advantage of symmetry; and (3) prescription of the proper boundary conditions
on the segment of the total structure to be analyzed.
Structural symmetry involves some balanced arrangement of geometry, structural
properties, and support conditions. In the dictionary sense it is the correspondence in
size, form, and arrangement of parts on opposite sides of a plane, line, or point. Here
we only consider the symmetry of structures that have a balanced arrangement about
some plane and consist of members that have bisymmetrical cross-sections. Our ex
amples are limited to plane structures. Reference 13.4 contains a useful further dis
cussion of the principles of symmetry and a comprehensive list of publications on
structural applications.
The plane structures of Figure 13.11 are obviously symmetrical. Their symmetry may
be defined either in terms of rotation about the y axis or reflection of every material
point through the y-z plane (the vertical plane that is normal to the paper and passes
through point 0). In Reference 13.4, Glockner notes that three common symmetry
operations, reflection in a plane, rotation about an axis, and inversion through a center,
r
Ii I2 Ii o"
0
I
-- ----X
Li L2 Li--1
Figure 13.11 Symmetrical Structures.
are equivalent to one or more reflections. Thus reflection may be treated as the fun
damentai operation of symmetry. For simplicity, we use the concept of rotation in the
following discussion, since it seems easier to visualize.
In working with symmetrical structures, it is useful to employ the concepts of sym
metric and antisymmetric loads. The following definitions of the two conditions are
similar to Glockner's:
1. A system of loads acting on a symmetric structure is symmetric if it is brought

into an equivalent configuration as the symmetry conditions of the structure are
applied to the system.
2. A system of loads acting on a symmetric structure is antisymmetric if it is brought
into an equivalent configuration as the symmetry conditions of the structure are
applied to the system and the signs of all load vectors are reversed.
These definitions are illustrated in Figure 13.12. For the symmetrical loads in Figure
13.12a, rotation about they axis of structural symmetry produces an equivalent system
of loads, except for the insignificant reversal of the identifying symbols. For the anti
symmetrical loads in Figure 13.12b, rotation about the y axis produces the situation
shown. Reversal of the direction of the rotated loads would produce a system equiv
alent to the original, unrotated one.
More often than not, the loads on a symmetric structure are neither symmetric nor
antisymmetric. Nevertheless, it is still possible to take advantage of structural sym
metry for all cases in which the principle of superposition applies. As shown in Figure
13.13, any load on a structure symmetrical with respect to a plane may be treated as
the sum of a symmetrical loading (Figure 13.13b) and an antisymmetrical loading (Fig
ure 13.13c).
Given symmetrical or antisymmetrical loads or having decomposed the natural load
ing into such components, there remains the problem of determining the boundary
conditions that must be applied on the plane of symmetry to permit one-half of the
structure to be analyzed properly. In terms of displacements these are:
1. For symmetric loading there can be

(a) No translation normal to the plane of symmetry.
(b) No rotations about two orthogonal axes in the plane of symmetry.
;- -
(b)
Figure 13.12 Load systems. (a) Symmetrical loads. (b) Antisymmetrical loads.
+
structural
symmetry,
(a) (b) (c)

Figure 13.13 Decomposition of loading systems. (a ) General loading. (b)
Symmetrical part. (c) Antisymmetrical part.
2. For antisymmetric loading there can be

(a) No translation in the plane of symmetry.
(b) No rotation about an axis normal to the plane of symmetry.
Although offered without proof, most of these requirements should be clear on phys
ical grounds. As an exercise in verifying them, consider the possible displacements of
the node a of the planar symmetrical space frame of Figure 13.14 under the separate
actions of the symmetrical load V and the antisymmetrical load H. Schemes for deter
mining symmetry boundary conditions are discussed further in Reference 13.4.
To solve a problem once symmetry has been recognized and the loads have been
suitably decomposed, the characteristic segment of the structure is analyzed twice: once
for the symmetric loading condition and once for the antisymmetric condition. The
resuits are superimposed using proper signs for all displacements and forces.
In Example 13.10, this procedure is applied to a symmetric continuous beam. In this
case, nothing is gained through the application of symmetry since there are two un
known degrees of freedom in each analysis, which is the same as the number of un
knowns (Ob and Be) in the original structure. It is merely an illustration of the application
of the principles of symmetry and a demonstration of the treatment of loads on an
Vertical plane
of symmetry
Plan
v
H
b,c
h,i
x 0 y
Elevations
Figure 13.14 Planar symmetry conditions.
element bisected by a plane of symmetry. It should be clear, however, that had the
number of spans been greater than three, the reduction in degrees of freedom would
have been significant.
Example 13.11 is an application of symmetry principles to a rigid frame. The number
of unknown degrees of freedom in the complete frame is nine, yet the complete anal
ysis is accomplished by considering no more than five. Further reductions could have
been effected by taking advantage of the usual frame assumption that axial deforma
tions are negligible. This was not done since our interest at this point is simply in
illustrating the principles of symmetry. Close attention should be paid to the signs used
in combining displacement components. Note that in crossing the plane of symmetry,
the signs of the u and 8 symmetrical components, and the v antisymmetrical compo
nent, are reversed.
Finally, it will be observed that in most of the general-purpose computer programs
in common use, it is possible to take advantage of symmetry by specifying appropriate
boundary conditions.
Using symmetry principles, determine the slope at b and

shown. I = 50 106
x mm4, E = 200,000 MPa for all members.
c for the beam
a
rO kN rO kN c d
I f.1.
2m
-...l.-- 5m
3m #& __j
]_/
3m
Symmetrical part: Plane of symmetry

By symmetry, 00 0.
Unknown degrees of freedom:
Fixed-end forces:
Ob,=
Vo.
a
rO kN b lO! I rO c rO d
-
l- 2i_11Li
3--i 2 _} 2-i- 3
Mfb -10 22 3/52 -4.80
ab = X x = kNm
' 2-.1 ,
Mfb 10 2 o.5212.52 0.80
bo = x x =
"'---
0.5 0.5
Mf0 -10 22 0.5/(2.5)2 -3.20
= x x =
F;b 1.04 F;0 8.9 kN 6 kN
[ ]
= =
Element stiffness equations ( from Equation 4.34):

{M,b} 200 4 250050 106 - (2500)2 50 106 { b}
ab M,b 200(4 50 106/5000)0b 200(4 104)0b
= x x = x
x x 6 x x (}
bo
Fy
=
o
50 106
Sym.
12 x x
Vo
(2500)3
200 [8-48.1040 .,.-48.0.00384 J { (Jb}
=
x
V0
Global stiffness equations:
{o} 200[12 104 -48.o ]{ob} {-4.oo 103}

=
x + x
0 -48.0 0.0384 V0 8.96
Solving:
{(Jb} {-0.600 10-3 }

=
x rad
v0 -1.917 mm
y
Antisymmetrical part:
By antisymmetry, v0 = 0.
Unknown degrees of freedom: Ob, 80.
Element stiffness equations (Equation 4.34):
ab
y
bo
Global stiffness equations:
{} [ { } { }
J
o 12 x 104 4 x 104 eb -4.oo x 103
= 200 +
0 4 X 104 8 X 104 80 -3.20 X 103
Solving:
{} {
(Jb
=
0.120 X 10-3 } rad
80 0.140 X 10-3
Combined analysis:
Ob = (-0.600 + 0.120)10-3 = -0.480 x 10-3 rad
(Jc = ( +0.600 + 0.120)10-3 = +0.720 X 10-3 rad
v0 = -1.92 mm
0.140 x 10-3 rad
Using symmetry principles, determine the displacements at the IOkNI

t 8m -
joints of the rigid frame shown. For members ab and de, A = 4
X 103 mm2, I= 50 x 106 mm4 For members be and cd, A = 6
X 103 mm2, I= 200 X 106 mm4. E = 200,000 MPa for all mem
m
bers. Use the results of Example 5.3.
' 3; '5kN
Sm
a e l
7.416 m --7.416 m ___.j
0 0
Symmetrical part:
By symmetry, Uc= (Jc= 0.
Unknown degrees of freedom: ub, vb, Ob, Ve- Y
From the element equations of Example 5.3, the global stiffness equations are
Plane of
{} l o o.6500 0.2591 4.969 -0.2591 ]{ }
ub
b symmetry
0 0.9095 17.381 -0.1095 vb
= 200
0 Sym. 1.4 x 105 -17.381 (Jb
-5 0.1095 Ve
a
x
Solving,
Lub vb Ob vcJ = L -2.208 mm -0.0313 mm -0.611 X 10-3 rad -5.583 mmJ

Antisymmetrical part:
By antisymmetry, Ve = 0.
Unknown degrees of freedom: ub, vb, ()b, Uc, ()c
From the element equations of Example 5.3, the global stiffness equations are Plane of
symmetry
5 0.6500 0.2591 4.969 -0.6452 -7.0313 ub
0 0.9095 17.381 -0.2591 17.381 vb
0 = 200 .1.4 x 105 7.0313 0.5 x 105 ()b a
0 Sym. 0.6452 7.0313 Uc 0 x
0 1 x 105 ()c
Solving,
Lub vb ()b Uc 9cJ

= L7.085 mm 0.0093 mm -0.7403 x 10-3 rad 7.093 mm 0.3680 x 10-3 radJ
Combined analysis: 7.09

ub = -2.20 + 7.08 = 4.88 mm 5.58
vb = -0.03 + 0.01 = -0.02 mm
()b = (-0.611 - 0.740)10-3 = -1.351 x 10-3 rad
Uc = 0 + 7.09 = 7.09 mm
Ve = -5.58 + 0 = -5.58 mm +0.0004

.02 0.04
()c = 0 + 0.368 X 10-3 = 0.368 X 10-3 rad
ud = +2.20 + 7.08 = 9.28 mm -0.00013
vd = -0.03 - 0.01 = -0.04 mm
()d = ( +0.611 - 0.740)10-3 = -0.129 x 10-3 rad
13.7 REANALYSIS TECHNIQUES

The term reanalysis techniques refers to procedures for assessment of the effect of
modest structural design changes without incurring the computational costs of a com
plete reanalysis. Up until now, our work has been concerned with situations in which
the arrangement arid proportions of the structural members and the support conditions
are known. In practice, however, the designer must first estimate the member sizes,
analyze the structure, and compare the calculated and allowable forces and displace
ments. Then, if certain calculated values are greater than the allowables, or if the
allowables are unacceptably greater than the calculated values, changes will be made
in the structural analysis conditions. Indeed, it may prove necessary to repeat this cycle
a number of times. It is useful, therefore, to consider procedures, alternative to com
plete conventional reanalysis, by which the results for the modified structures can be
obtained.
In the case of support condition changes, the procedure already described for sub
structuring is applicable. Consider the structure shown in Figure 13.15a, where all
column bases are assumed to be fixed. The degrees of freedom of the remaining joints,
comprising all of the displacements to be determined in this analysis, are designated
as {bl The global stiffness equations, already modified to account for the support
conditions, are then
(13.42)
and, after solution,
(13.43)
18 19 16 17 18 19
16 17
14 15 11 12 13 14 15
11 12 13
8 9 10 6 7 8 9 10
6 7
2 3 4 5 1 4 5
1
29 30 32 31 32
31
0 0 0 0 0
(a) (b)
Figure 13.15 Modification of support conditions.
Now, if it is decided to examine the effect of changing the support conditions of the
structure from fixed to hinged supports at points 28, 29, and 30 (see Fig. 13.lSb), there
is no change in the stiffness properties [Kbb] associated with the degrees of freedom
{Ab) One must supplement global stiffness equations, however, to account for the
degrees of freedom {Ac), where {Ac) = L (}28 (}29 (}30j. Thus the global stiffness equa
tions become
(13.44)
These equations are exactly the same form as in substructure analysis, where the so
lution for {Ac} is given by
{Ac} =
[[Kee] - [Kcb][Kbbr1[Kbclr1[{Pc} - [Kcb][Kbbr1!Pb)] (13.15a)
and
(13.16a)
Note that [Kbbr1 is already available from the solution of the original structure (Fig.
13.lSa), and the solution for the modified structure (Fig. 13.lSb) requires only the
inverse of a matrix of the same order as the "relaxed" support degrees of freedom
{Ac}
A general procedure for investigating the effects of different support possibilities
emerges from the above. All degrees of freedom that will not be among those consid
ered for support conditions are first isolated and designated as {Ab) The associated
stiffness coefficients [Kbb] are evaluated and solved, [Kbbr1 The degrees of freedom
of any support condition to be examined are then designated as {Ac), and the solution
for the full set of joint displacements is obtained from Equations 13.lSa and 13.16a.
We consider now reanalysis techniques intended to deal with situations in which the
element stiffnesses are modified on account of changes in cross-sectional properties or
material properties. Two classes of such techniques can be identified: exact and ap
proximate procedures.
Exact reanalysis techniques have appeared often in the literature, described by
various names and formulated through different strategies. Each depends upon ex
ploitation of the zero terms of the matrix of changed element stiffnesses. One such
approach is described as follows.
If the changes to the original stiffness matrix [K0] are denoted as [8K0], the new
.
stiffness equations can be written in the form:
{P} =
[[K0] + [8K0]]{A} (13.45)
Note that in this development the load vector {P} is assumed to be unaltered by the
changes in dead load due to changes in member size represented as [8K0]. By factoring
out [K0] on the left side, this can be written as
1
{P} [K0][[I] + [K0]- [8K0]]{A}
= (13.45a)
and, by inversion (recalling that the inverse of a product of matrices is the product of
their inverses in reversed order), we obtain the solution for the displacements {A} of
the modified structure:
(13.46)
Furthermore, since [K0t1{P} {A0), where {A0) symbolizes the solution for displace
=
ments of the original (unmodified structure),

(13.46a)
1 1
Since {A0} is already available from the solution of the original structure, it is only
necessary to develop an expression for [[I] + [K0t [8K0]t In so doing, it is assumed
that the changes in stiffness can all be collected in the upper left portion of [8K0]. In
be written as [u7 v7 87 u12 v12 812

Figure 13.15a, for example, if member 7-12 is altered, the vector of displacements will
] with a corresponding arrangement of the
vector of applied loads. If all degrees of freedom associated with stiffness changes are
designated as {Ai} and the remainder as {A2}, we can rewrite Equation 13.45 in the
partitioned form:
(13.45b)
[
where 8KZ1] contains all of the modified element stiffness terms. Also, we write the
already available inverse of [K0] in the form:
[[I]
With the above expanded relationships, we can write
1
[K0]- [8K0]] as
[[I J [ Z22 ][ Z1 J] [
+
021 02
DZ1 D 8K q
J
O O O
+
01
I D21 8K1 I
=
0 0 0
where [q] [I +
= 8Ki]. Note that here the size of [q] and the identity matrix [I]
is the same as the number of degrees of freedom associated with the modification. The
inverse of the above is
[ ----q ---=-1- -
-D21 8K11q I

I
] =
[[I] + [Kotl[8Ko]]-l (13.47)
Substitution of Equation 13.47 into 13.46a, with {A} partitioned intoLA1 : A2f, gives
{A} = {;} {2--orbi-=-1--}

=
(l3.4S)
Observe that the only matrix inverse appearing in this expression is that of [q]. The
necessary operations are otherwise matrix multiplications.
Example 13.12 illustrates the above procedure. In order to present the procedure in
detail, the structure consists of only three members and the changes affect a large part
of the total stiffness matrix. To be useful, however, the reanalysis situation should
involve a relatively small number of members of the original structure.
The stiffness of member 1-2 is to be changed from k =
1 to k = 2. Using the "exact" procedure, calculate the

displacements in the changed structure.
Original structure
Stiffness equations for the original structure are
{:} [- - -]{::}
P4
=
0 -3 5 U4
Flexibility equations for the original structure, (4) = (D0](P}, are
(13.1)
[Did =
(16) [SK0] =
1[ OJ 0
0 0 0 [<'>Kid =
(1)
2
0 0 0
[q] =
[[I] + [Dl1][<'>Kl1Jl =
( (1) +
2
(16)(1) ) =
Thu. with i":l = L16 10 J {'.l

i"I =
[qJ-'l":J =
L16 10 6J {'.}= l !J{'.}
.
Also, with {42) = -
1 [ 10 15
28 6 9
=
[
_!_ 10 15
28 6 9
4U Chapter 13 Special Analysis Procedwes
To summarize,
{a1}
---
d2 { }
= u
3
=
1
44
-
l!
llO 1o
__ ____
20 12
j{P2}p
3
U4 6 12 16 P4
This result agrees with that obtained by direct inversion of the stiffness matrix of the changed
structure.
An approximate reanalysis formula can be obtained as follows. In Equation 13.45

we first replace {A}, the total displacements of the modified structure, with {A0} +
{ cSA0}, where (A0) designates the already-solved-for displacements of the original struc
ture and {c5A0} are the as-yet unknown changes in displacements due to the structural
modifications. Equations 13.45 therefore is now written as
{P} = [[K0] + [c5K0]]{A0 + c5A0} (13.45c)
Expansion of the right side gives
{P} =
[K0]{A0} + [c5K0]{A0} + [K0]{8A0} + [8K0]{8A0} (13.45d)
Now, (K0]{A0} {P}. (The displacements {A0} are the solution to the problem for the
=
original stiffness [K0].) These terms therefore can be cancelled from the two sides of
Equation 13.45d. Also, we discard the higher-order term [8K0]{c5A0); this comprises
the approximation being made. Equation 13.45d is then reduced to
[8K0]{A0} + [K0]{8A0} = {O}
or
(13.49)
All terms on the right side of Equation 13.49, [K0r1, [8K0], and {A0}, are known,
and the change in displacement {8A0} due to the modification [8K0] is obtained simply
from a matrix product Equation 13.49 has been developed without any assumptions
being made regarding the location of zero terms in [8K0]. In large-scale practical design
analysis, however, the changes in stiffness may occupy a small portion of [8K0], and in
such cases it is possible to achieve significant efficiencies in formation of the product
on the right side.
The calculated changes in displacement can be applied to the element stiffness ma
trices to determine the associated changes in the internal forces. In some cases, the
analyst may wish to have the revised flexibility matrix. This can be obtained from
{Ao} + {8Ao} = [Kor1{P} - [Kor1[8Ko]{Ao}

= [Kor1\P} - [Kor1[8Ko][Kor1!P}
= [[Kori - [Ko][8Ko][Kor1HP}
= [Ko-1][[1] - [8Ko][Kor1]{P} (13.50)
Thus, the revised flexibility matrix is
(13.51)
The use of the above approximate reanalysis technique is illustrated in Example

13.13. This problem, which was solved exactly in Example 13.12, shows that the ap-
13.8 Problems 413
proximate method is not always sufficiently accurate. The design analyst must exercise
judgment in identifying such circumstances. To improve the accuracy of this method,
Equation 13.49 can be employed iteratively, that is, {.1) + {8.1) can be used in place
of {.1) to obtain a better estimate of the displacement changes. This approach adds to
the computational cost, however, which might then exceed the cost of a completely
new analysis. Indeed, the cost of reanalysis techniques versus completely new analysis
have been questioned in some studies (Refs. 13.5, 13.6). Note also that a wide variety
of reanalysis techniques has been proposed (e.g., Refs. 13.5-13.10), of which the above
is merely a simple example.
[EXAMPLE 13.13]
Using the data of Example 13.12 and the "approximate" method (Equation 13.50), calculate
the displacements in the changed structure. Compare the results with the "exact" solution.
From Equation 13.50:
{} {Ao} +
[ {oAo} =
][
(Kott([I] - (oKo][Kott]{P}
][[ ]- [
U2 l 16 10 0 16
1
U3 10 15 0 10
[
=
28 28
U4 approx. 6 9 11 0 0 1 0 0 0 6
192 120
=
(2 ) 2 120
72
320
192
or, in decimal form,
{::} [:; : :!]{:}

U4 approx.
=
0.092 0.245 0.347 P4
{} [ ]{ }
The exact solution, in decimal form, is
U2 0.363 0.227 0.136 P2

U3 =
0.227 0.454 0.273 P3
U4 0.136 0.273 0.364 p4
13.8 PROBLEMS
13.1 Apply the condensation equations of Section 13.1 to the solution of Problem
3.4.
13.2 Con ense the angular displcement (}b from Enmple 5.7 to produce stiffness
?
.
equations m terms of ub and vb; then solve for the latter.
13.3 Fo t e sti fess matrix for the three-jointed axial member of Example 7.1,

and then ehmmate JOmt 2 by use of the condensation procedure of Section 13.1. Com
pare the result with the axial member stiffness matrix.
13.4 Use substructuring to analyze the structures of Problem 4.9.
13.5 Use substructuring to apalyze the structure of Problem 4.15.
13.6 Combine substructuring with, where appropriate, the methods of Section 5.2 to
analyze the structures of Problem 5.8.
414 Chapter 13 SpeciaJ AnaJysis Procedures
13.7 Use substructuring to analyze the structure shown.
I = 100 x 106 mm4 E = 200,000 MPa
lOkN 2/
5m
5kN
I I
2/ 2/
I I
-i5m
I
0 0 0 _J
/-3m-1-4m--J
Problem 13.7
13.8 Use substructuring in conjunction with a computer program to analyze the

trusses of Problem 3.15.
13.9 (a) Compute the displacements, reactions, and bar forces for the truss shown
using substructuring. Cross-sectional areas (mm2 X 103) are shown on each bar. E =
200,000 MPa. (Same as Problem 3.1).

(b) Solve the same problem but for a vertical load of 300 kN at d.
b 300kN
7m
20
0
L10m---L10m_j
Problem 13.9
13.10 Remove the hinge from Example 4.14 by enforcement of the constraint con
dition ()to, - ()to, = 0 and demonstrate that the resulting stiffness equations are those
of a member of length L.
13.11 Establish the matrix equations, similar to Equation 13.25, to accommodate
constraint conditions of the type given by Equation 13.18 (i.e.,
{HJ * 0).
13.12 Assume that the joints a and b of Example 5.6 are so connected that ()za =
-6(),b Enforce this constraint condition and solve for 8,b

13.8 Problems 415
13.13 A roller support is placed at joint b of the rigid frame of Example 4.13 as shown
below. Using the "joint coordinate" approach of Section 13.4 and the data of Example
4.13, solve the problem to account for the presence of this support.
a b
c

Problem 13.13
13.14 Re-solve Example 5.7 after the assembled stiffness matrix at joint b has been
revised so that the coordinate axes are aligned in the direction of member ab and
perpendicular to it.
13.15 Example 5.10 has been revised so that the hinge at point c is replaced by a
roller that is oriented as shown below. Calculate the displacements for this revised
support condition.
t;;;.d _____ -"
Problem 13.15
13.16 Demonstrate that the use of Equations 13.34 and 13.35 result in the form of
Equation 13.32.
13.17 Compute the displacements, reactions, and internal forces for the beam shown.
E = 200,000 MPa and I = 10 x 106 mm4
lOkN
!_b _c
,._ __ 4 m----- 2 m ____..j
Problem 13.17
13.18 For the beam shown, plot the relationship between a1 and the vertical displace
ment vb. For what value of a1 is the beam essentially fixed at a?
p
k a1EI
a
=
L EI
!b
L---
Problem 13.18
13.19 Compute the displacements, reactions, and internal forces for

the frame shown.
E = 29,000 ksi. Use the following connection stiffnesses:
(a) k =1200 in. kips/rad
(b) k =12, 000 in. kips/rad
(c) k =120,000 in. kips/rad
r li .
c
15' All members:

W 12x16
L b
'-----10' -------+!
Problem 13.19
13.20 Compute the displacements, reactions, and internal forces for the frame shown.
E = 29,000 ksi. Use the following connection stiffnesses:
(a) k1 = k2 = 14.4 X 1C>4 in. kips/rad
(b) k1 = 2k2 = 14.4 X 104 in. kips/rad
(c) k1 = 0 and k2 =oo
(d) k1 = 0 and k2 =14.4 X 104 in. kips/rad
75 kips k1
l
b c
..ft
All members:

10,
J
W14x82
a d
-- 5'----+<
!+---1
Problem 13.20
13.21 Use Equation 13.38 to analyze the beam-column shown. Compare results with
an analysis that uses an equivalent axial force and moment acting at the centroid of
lthe member. E = 29,000 ksi.
300kips
l
""" 20 '

J
Problem 13.21
13.8 Problems 417
13.22 Compute the displacements, reactions, and internal forces for the beam-column
shown. E = 29,000 ksi.
Problem 13.22
13.23 Repeat Example 13.9 for the frame shown. Assume all connections are rigid.
<Ty = 36 ksi, E = 29,000 ksi.
W16x26
I W18x50
I ==-- 0
11===1===;1 ll===15
x x
12' 00 00

--- 4'----___,- 6'

----2
Problem 13.23
13.24 The area of member ca of Example 5.10 is changed from 15 X 103 mm2 to
7. 5 x 103 mm2. Recalculate the displacements, using both the exact and approximate
methods of Section 13.7.
13.25 Using the approximate method of Section 13.7, calculate the displacements and
internal forces of the structure of Example 5.7, for a change in the size of member ab
to A = 800 mm2, I = 300 x 106 mm4
13.26 Using the approximate method of Section 13.7, calculate the displacements and
internal forces of the structure of Example 5.4 for a change in the size of member ad
to 20 x 103 mm2. Confirm the accuracy of the approximate solution by performing a
computer analysis of the revised structure.
13.27 (a) Re-solve Problem 4.lg using symmetry and antisymmetry conditions. (b)
Calculate the displacements at joint b of the second structure in Problem 4.9 using
symmetry and antisymmetry conditions.
13.28 Re-solve Problem 5.8c using symmetry and antisymmetry conditions.
13.29 Sketch the deflected shapes of the structures in Problems 5.lOa, 5.lOc, and
5.lOd. State the conditions of symmetry and antisymmetry that can be invoked in each.
Use sketches to illustrate how they would be employed, that is, what the resulting
idealized structures would look like.
REFERENCES
13.1 Y. Goto and S. Miyashita, "Classification System for Rigid and Semirigid Con
nections," ASCE, fl. of Struct. Engr., Vol. 124, No. 7, July, 1998.
13.2 R. Bjorhovde, A. Colson, and J. Brozzetti, "Classification System for Beam
to-Column Connections," ASCE, fl. of Struct. Engr., Vol. 116, No. 11, Nov.,
1990.
13.3 W. F. Chen and E. Lui, Stability Design of Steel Frames, CRC Press, Boca Raton,
Fla., 1991.
13.4 P. G. Glockner, "Symmetry in Structural Mechanics," ASCE, fl. of the Struct.
Div., Vol. 99, No. STl, Jan., 1973.
13.5 D. Kavlie and G. H. Powell, "Efficient Reanalysis of Modified Structures,"
ASCE, fl. of the Struct. Div., Vol. 97, No. STl, Jan., 1971.
13.6 J. H. Argyris and J. R. Roy, "General Treatment of Structural Modifications,"
ASCE, fl. of the Struct. Div., Vol. 98, No. ST2, Feb., 1972.
13.7 U. Kirsch and M. F. Rubenstein, "Reanalysis for Limited Structural Design
Modifications," ASCE, fl. of the Engr. Mech. Div., Vol. 98, No. EMl, Feb., 1972.
13.8 A. K. Noor and H. E. Lowder, "Approximate Techniques of Structural Re
analysis," Computers and Structures, Vol. 4, 1974.
13.9 J. Sobieszczanski, "Structural Modification by the Perturbation Method,"
ASCE, fl. of the Struct. Div., Vol. 94, No. ST12, Dec., 1968.
13.10 R. J. Melosh and R. Luik, "Multiple Configuration Analysis of Structures,"
ASCE, fl. of the Structural Div., Vol. 94, No. STll, Nov., 1968.
Appendix A
Nonlinear Analysis
A Further Look
Geometric nonlinear response, and in particular the stability of framed structures, is

a function of the way in which components of force interact with each other as the
structure deforms and displaces under load. The effect of axial force on the uniaxial
bending and simple twisting of a member is a major concern, but it is one that lent
itself to the relatively simple physical analyses of Sections 9.1 and 9.2. Equally impor
tant are the phenomena whereby a beam subjected to primary bending about one
principal cross sectional axis can become unstable through a combination of bending
about the other one and twisting about the longitudinal axis, and the closely related
mode in which the primary action is a combination of axial force and bending about
one principal cross-sectional axis of a beam-column.1 These, too, were treated in Chap
ter 9, but in the interest of obtaining tangible results we employed without proof
various terms representing the spatial interaction of bending, torque, and other force
components and we limited our analysis of torque to the pure St. Venant type. The
requisite proofs require taking a close look at the way in which the stress resultants
used in engineering analysis are derived from the principles of structural mechanics
and the effect of displacements, particularly finite rotations, on their interaction.
Thus this appendix has a twofold purpose. The immediate one is the development
of the terms of the geometric stiffness matrices of Chapter 9 that were beyond the
reach of the physical arguments of that chapter and the analysis of nonuniform torsion.
But in the course of meeting this objective, a fundamental approach to the develop
ment of the equations of elastic nonlinear analysis will be described and key problems
encountered in applying it to framed structures will be treated. Therefore the broader
purpose is to provide deeper insight into the process of nonlinear analysis.
The topics covered include
1. Use of virtual displacement principles in Lagrangian formulations.

2. Linearization of the virtual displacement equations.
3. Application of the linearized equations to the framework element.
4. The nature of finite rotations and the need to consider them in establishing equi
librium on the deformed structure.
5. Nonuniform torsion.
'Both are forms of torsional-flexural buckling, but that term is often reserved for the second. with the first
commonly called lateral buckling.
419
420 Appendix A Nonlinear Analysis-A Further Look
A.1 VIRTUAL DISPLACEMENT PRINCIPLES IN LAGRANGIAN FORMULATIONS
In Section 8.3.2 we discussed the fundamental problem of formulating the equations

of equilibrium on an unknown configuration and we described a scheme used to cir
cumvent it in the analysis of the interaction of axial force with bending and twisting
moments. In this appendix we review a more rigorous method: use of the principle of
virtual displacements in a Lagrangian formulation.2 This is an approach upon which
broad formulations of the nonlinear analysis problem can be based. Here it will be
used as the basis for the analysis of additional forms of interaction between compo
nents of force acting on the framework element. The review draws on References A.l,
A.2, and A.3. Further details may be found in those works.
The rigorous use of virtual displacement theory to establish equilibrium on the de
formed configuration would require that it be applied to that state, that is, to Config
uration t + !::.tin Figure 8.4. The idea of applying a virtual displacement to an unknown
configuration would appear to be compounding the problem. But it has been found
possible to express these virtual displacements and the sought after real force and
displacement increments in terms of the known conditions and coordinate system at
configuration t. As shall be demonstrated, a particular measure of finite strain-the
Green-Lagrange strain tensor-and stress-the second Pio/a-Kirchhoff stress ten
sor-are conjugate variables; that is, their scalar product (an operation defined below)
is a measure of the real or virtual work done during a corresponding real or virtual
displacement (see Section 5.3.1). And both tensors satisfy the condition of being ex
pressible in terms of the known conditions at configuration t.
For simplicity of notation in the demonstration we will consider Configuration t to
be the unknown deformed state and Configuration 0 the known unloaded, undeformed
state, as in Figure A.l. At the end we will discuss the rather straightforward adaptation
of the results to the case in which Configuration t is a known intermediate state and
Configuration t + !::.tis sought. In Figure A.1 the components of all vectors and matrices
Virtual displaced configuration
\---"
/
/
/
Configuration 0
Xt
Figure A.1 Virtual displacements in nonlinear analysis.
2The term Lagrangian formulation refers to one that uses material coordinates defined in terms of the
undeformed or a Eulerian
temporary reference configuration. The Lagrangian approach is in contrast to the
formulation, which employs spatial coordinates in the deformed configuration and is commonly used in fluid
mechanics for analysis of the motion of material through a stationary control volume.
A.l Virtual Displacement Principles in Lagrangian Formulations 421
are stated in terms of a fixed Cartesian coordinate system. If we wish to formulate the
equations of equilibrium on the deformed state and to use virtual displacement prin
ciples to do so, then the displacements should be applied to Configuration t, as indi
cated. Thus the internal virtual work is
(A.1)
in which T is the real (physical or "Cauchy") stress tensor and e is the infinitesimal
strain tensor. T:8e is defined as the scalar product of two tensors, that is, the sum of
the nine products of components or, in the usual indicial notation, T;i8e;i, and integra
tion is over the volume of the deformed configuration. In matrix form, the stress tensor
may be written as
(A.2)
+
Noting that 'x; = 0x; u;, in which u; is a Cartesian component of the increment of
displacement between the undeformed and the deformed configuration, alternative
forms of the infinitesimal strain matrix are3
1 1 au1 ! ( au1 + au2 ) !( au1 + au3 )

2 /'xy 2 'Yxz
ex
a'x1 2 a1x2 a1x1 2 a1x3 a1x1
1 1 !( au2 + au1 ) au2 !( au2 + au3 )

e or
2 /lyx
e=
y 2 /lyz 2 a1x1 a'x2 a'x2 2 a1x3 a1x2
1 1
e,
!( au3 + au1 ) !( au3 + au2 ) au3
2 'Yzx 2 'Yzy 2 a1x1 a1x3 2 a'x2 a1x3 arx3
(A.3)
or, in tensor component form:
(A.4)
Thus
Se =
! ( a8u; + a8ui ) (A.5)
I) 2 a1xj a'xj
In the above equations the superscript t signifies that the tensors apply to the de
formed configuration and that they are written as functions of the components of
position vectors of points in that configuration. Also, integration is over the deformed
volume. For these reasons, although Equation A.1 is a valid statement of the principle
of virtual displacements, it can't be used directly to determine the physical stresses in
configuration t since that configuration is still unknown. Therefore, we turn to an al
ternative expression of the internal virtual work
(A.6)
'To ensure that e transforms as a second-order tensor, the factor1/2 is included in the definition of the shear
strain terms (see Reference A.1 and compare with Figure 4.Ib and Equation 4.lc).
in which t is the second Piola-Kirchhoff stress tensor and E is the Green-Lagrange

strain tensor. Here, the superscript t again signifies that the tensors apply to the de
formed configuration, but the subscript 0 designates that they are referred to the un
deformed configuration; that is, their components are written as functions of the known
Cartesian coordinates of p0 and integration is over the undeformed volume. Our im
mediate aim is to demonstrate that the internal virtual work symbolized by Equation
A.6 is indeed identical to that of Equation A.1.4 Throughout, we assume that the
external virtual work is deformation independent (see Section A.3 for a comment on
the implications of this restriction).
Consider strains first: The Green-Lagrange strain tensor is a measure of finite strain
defined as
aE = !(&xT 0x - 1) (A.7)
in which 0X is the deformation gradient

a'x1
aoxl
a'x1
aox2
a'x1
aox3
l
a'x2 a'x2 a'x2
oX (A.8)
a0x1 aox2 aox3
=
a1x3 a1x3 a1x3

aoxl aox2 aox3
The deformation gradient describes the elongation and rotation that a material fiber
undergoes between Configuration 0 and Configuration t. As illustrated in Figure A.2,
a differential material element, d0x, at point p0 would become d'x at point p,. There
'
fore, by the chain rule of differentiation
{ d'xJ = [0X]{d0xj
Also, noting that the squares of the lengths of the element in the two states are
2 2
(d0x) = Ld0xj{d0xj and (d'x) = Ld'xJld'xJ
it follows that
2 2
(d'x) - (d0x) = Ld0xJ[oXT oX - I]{d0xJ (A.9)
Through comparison of Equations A.7 and A.9 it may be seen that the Green
Lagrange strain tensor was defined to give the change in the squared length of the
material element (see also Reference A.4). From Figure A.2 and the manner in which
Equation A.9 was derived, it is clear that it contains information on the rotation of the
d'x =
{ }
d'x
d'x
d'x3
Figure A.2 Displacement, stretch, and rotation of a

differential material element.
4lt is suggested that the reader verify the key equations of this demonstration by carrying out the interme
diate operations.
A.1 Virtual Displacement Principles in Lagrangian Formulations 423
element as well as on its elongation. But it can be shown that the deformation gradient
can be decomposed into the product of a symmetric stretch matrix, 0S, and an orthog
onal rotation matrix, 0R, such that 0X = 0R 0S. Therefore 0XT 0X = 0S 0RT oR oS =
0S 0S, which means that although the stretch tensor itself produces some rotation of
all material vectors except those in the principal strain directions, the Green-Lagrange
strain tensor is independent of their rigid body motions (see also Refs. A.1 and A.2).
0x; +
Using Equation A.8 and letting 'x; = u;, the Green-Lagrange strain in tensor
component form becomes
I 1 +I +I I
2(10U l ) (A.10)
_
Oe--
l) l,) 0U
j.l oUkoUk
,l
in which commas denote differentiation with respect to the coordinate following. For
comparison with the infinitesimal strain tensor we have, for example:
2
o e11 =
:) [ (:olr
+
(:olr (:::J ]
+ +
(A.lla)
oe12 =
! ( au1 + au2
a0x2 a0x1
) ![ +
au1 au1 + au2 au2 + au3 au3 ] (A.llb)
2 2 a0x1 a0x2 a0x1 a0x2 a0x1 a0x2
Also from Equation A.7 we can write the virtual strain as
(A.12)
Now, defining 8,U as

a8u1 a8u1 a8u1
a'xi a'x2 a1x3
a8u2 a8u2 a8u2
8,U = (A.13)
a'xi a'x2 a1x3
a8u3 a8u3 a8u3
a'xi a'x2 a1x3
from the chain rule for the partial differentiation of the elements of 8,U with respect
to those of {p0} it may be seen that
8oX = 8,UoX (A.14)
Thus, from Equations A.12 and A.5
+ +
80 = HQXT(8,ur o X o XT(8,U)oX] = oXT{![(8,U)T (8,U)])oX (A.15)
= o XT8eoX
which relates the virtual Green-Lagrange strains to the virtual infinitesimal strains.
Now consider the second Piola-Kirchhoff stress tensor. In the context of the two
states considered it is defined as
0
0x TI0XT
0'T = .!!.
f I (A.16)
p
in which 0p and 'pare the material densities in the undeformed and deformed config
urations respectively and ?X is the spatial deformation gradient, which may be verified
to be the inverse of the deformation gradient (Equation A.8). In component form the
stress tensor is
I
oT;i
- _
-1
0
P
p
0
,x;,m Tmn 0
,xJ.n (A.17)
Substituting Equations A.15 and A.16 in Equation A.6, we have
{ (iT:80dV0 { p [?XT?XT]:[0XT 8e0X]dV0 (A.18)

Jvo Jv0 p
=
0
Assuming no loss of mass in the transition, that is, that pV0 = 'pV" and recognizing
the orthogonality of the gradient matrices and the symmetry of the stress and strain
tensors, we see that
{ 0T:80dV0 { T:8edV, (A.19)

Jv0 Jv,
=
which satisfies the objective of demonstrating that Equations A.6 and A.1 yield iden
tical values of the internal virtual work.
As a commentary on the above it will be observed that Green-Lagrange strains are
but one of several measures of finite strain, logarithmic (Hencky) strains being another.
All have a physical base, but the Green-Lagrange variety is perhaps the most useful
in the development of general methods of nonlinear analysis.5
As measures of stress, on the other hand, the Cauchy stresses are the ones that have
physical meaning and thus they or their resultants are the ones we must ultimately
calculate. As shown in References A.l and A.2, second Piola-Kirchhoff stresses are
ingenious constructions based on the transformation of stresses and the deformation
of the element on which they act from one state to another, but they have little physical
meaning. Their primary virtues are those just described: in combination with Green
Lagrange strains they are valid measurements of virtual work and thus provide a basis
for the formulation of nonlinear equations of equilibrium on a known configuration.
The Lagrangian formulation is used in both Figures 8.4 and A.1 and each illustrates
variations on the same basic idea of using material coordinates in the undeformed or
a temporary reference configuration-as opposed to spatial coordinates in the de
formed configuration-to trace the nonlinear behavior of a system. The process shown
in Figure A.1 is generally termed a total Lagrangian one in that the reference config
uration is the actual initial one, whereas that in Figure 8.4 is called an updated La
grangian approach. All of the equations developed from Figure A.1 can be applied to
the case of Figure 8.4 by redefining Configurations 0 and t of Figure A.1 as Configu
rations t and t + tlt, respectively.
5It may be noted that Equation 9.la of Section 9.2 is apparently equivalent to Equation Alla, the com
parable Green-Lagrange strain component. There is a fundamental difference, however. The Green
Lagrange strain components involve only linear and quadratic terms of the displacement derivatives. No
higher-order terms were neglected in the development of Equation A.11. The Green-Lagrange strain tensor
is a complete finite strain tensor, and it is for this reason that there are no restrictions on its use in conjunction
with second Piola-Kirchhoff stresses in the formulation of large displacement problems. The definition of
strain as the unit extension of an element, as in Equations 9.1 and 9.la, produced the same result, but those
equations are approximations obtained by neglecting higher-order terms. When used in conjunction with
Cauchy stresses in a virtual displacement equation formulated on the reference configuration, as we did in
the development of Equations 9.4 and 9.16, it must be recognized that the result is only valid for small finite
strains.
Equation 9.1 does furnish a simple geometric interpretation of the direct components of the Green
Lagrange strain. When strains are small, the measure of strain based on the change in the squared length
of a material element is a close approximation of the unit extension. Similar analysis of the angle change
between two material elements leads to the conclusion that, when the unit extensions and angle changes
are small, Green-Lagrange components such as Equation A.llb are approximately one-half of the change
in angle in the plane defined by the elements (see Refs. A.4 and A.5 for further geometrical interpretation
of the strain components).
A.2 An Updated Lagrangian Formulation and Its Lineari:iation 425
In Figure A.1 all variables have been stated in terms of components of a single
Cartesian coordinate system, whereas in Figure 8.4 local element coordinate systems
have been shown for each stage of loading, with the understanding that conventional
transformation procedures will be used to transform the resulting matrices to a single,
fixed system for the formation of the global equations of analysis. But the coordinate
system employed is not a distinguishing feature of the total and updated Lagrangian
approaches; the two schemes were used merely to illustrate the choices available.
Since, in practice, linearized approximations of the basic equations are normally
used, both Lagrangian approaches will be applied incrementally. When used consis
tently, both yield the same results.6
A.2 AN UPDATED LAGRANGIAN FORMULATION AND ITS LINEARIZATION

An essential step between the statement of the principle of virtual displacements and
its application to specific elements is the development of general incremental equations
of equilibrium. An updated Lagrangian formulation of this type is outlined in this
section.
As in Section 8.3, we start at an intermediate stage of the analysis, Configuration t,
a known, loaded and deformed reference state; one in which the material's constitutive
laws and the conditions of equilibrium and compatibility are presumed to have been
satisfied by prior incremental analysis (see Figure 8.4). The objective is to write equa
tions of equilibrium for the next increment, Configuration t + At, but with all static
and kinematic variables expressed in terms of the reference state and, having done
that, to reduce the equations to a linear form suitable for incremental and incremental
iterative analyses.
As noted above, the equations of Section A.1 may be applied to the present case
symbolically by shifting 0 and t to t and t + At respectively. Thus, for equilibrium on
Configuration t + !J..t, from Equation A.1 and the basic virtual work equation, 8Wint =
8Wext (Eq. 6.13), we have
in which '+c>.'R is the external virtual work. The expression for external virtual work
will be retained in this form throughout the section. Implicit in this is the assumption
that the external loading is deformation independent. The significance and limitations
of this will be considered in Section A.3. From Equation A.19, the alternative expres
sion for internal virtual work, we have
(A.20)
or, in indicial notation
(A.20a)
in which component forms of i+t!.:fij and 01+6:sij may be obtained from Equations A.17
and A.10 by the appropriate changes in sub- and superscripts.
6The effect of initial strains has not been included in this discussion. When present, they must be considered
in the analysis of the undeformed state (see discussion of initial strain matrices in Ref. A.2).
The stresses may be decomposed as

1 1
l+t;.iT
l -ij - I T-ij +
- I T-ij -
- Tij + I T-ij
in which T;j, the second Piola-Kirchhoff stress at Configuration t, is identical to the

Cauchy stress, 7;j, at that configuration, and 1T;j is the increment of Piola-Kirchhoff
stress between Configurations t and t + 6.t.
For Green-Lagrange strains and incremental displacements expressed in terms of
the reference state, i+t;.eij = 1e;j Thus, from Equation A.10
(A.21)
Letting 1eij 1eij + T/ij Equation A.21 may be decomposed into linear and nonlinear
r
=
components in which
and (A.22)
The virtual strains may be decomposed in the same way.

Decomposing stresses and strains in Equation A.20a in this fashion we have
1
{ 1T;jf>1e;jdV, + { 7;jf>,e,,dV1 + { T;jf>rT/ijdV1 r+t;.rR (A.23)
J J J
=
Equation A.23 remains a consistent statement of the principle of virtual displacements

in which all integrals are over the known volume of Configuration t. For a given dis
placement variation, the second integral is a known quantity. It is customary to trans
pose it to the right-hand side of the equation as an expression of the virtual work of
the existing forces on Configuration t, but we shall leave it on the left-hand side for a
reason to be explained in Section A.4.2. The third integral is linear in the incremental
displacements, but the first one is not, so it must be approximated in some way for the
purpose of practical computation. Approximations suggested in Reference A.2 are
,T;j 1Cijrs 1e,5 and f>1e;j
= f>1e;j, in which 1C;jrs is the incremental stress-strain tensor at
=
Configuration t. The result is the linearized equation
(A.24)
The development of the revised integral and the approximations involved are ex
plained further in Reference A.2. Equation A.24 admits incorporation of variable
materials properties, but for linear elastic materials 1C;jrs is constant and the first in
tegral leads to the conventional elastic stiffness matrix.
A.3 APPLICATION TO THE FRAMEWORK ELEMENT
Equation A.21 or a suitably linearized version of it, such as Equation A.24, can be the
base for the development of finite element equations for the nonlinear analysis of many
line, plate, and solid elements (see Reference A.2). Here we shall use Equation A.24
in a limited way: to fill in the gaps in the nonlinear framework element stiffness ma
trices that could not be developed through simple physical arguments.
Figure A.3, a copy of Figure 4.6. shows the bisymrnetrical element and the nodal
forces and degrees of freedom considered. All q u antiti es shall be referred to the axes
shown, thereby eliminating the need for the sub- and superscripts of Equation A.24.
As before, deformations due to flexurally induced shear will be neglected and, in this
section, we shall only consider simple St. Venant torsion. These limitations, plus the
use of common engineering theory for converting stresses to cross-sectional stress re
sultants, admits considerable simplification of Equation A.24.
y
t
+ Myl lly1
tFy1,V1
Figure A.3 Bisymmetrical framework element.
As shown in Figure A.4a, the stresses at a poin on a cross section are ax, 'Txy and
'Txz Because a , a,,

y
and 'Tyz are not considered, and since 'Tyx= 'Txy and 'Tzx= 'Txz these
are the only independent stress components. They and the corresponding strain com
ponents can be written as simple vectors
{77} = L7Jxx 1Jxy 1JxzJT (A.25)
For elastic behavior, the stress-strain tensor becomes
ICJ=IEGGJ (A.26)
If applied forces are deformation independent, '+t>.'R, the external virtual work can
be expressed as the scalar product L'+t>.'Fj{Su}, with
L'+MFJ= LFxl Fyl F,1 Mxl Myl M,1 Fx2 Fy2 F,2 Mx2 My2 M,2J (A.27)
in which L'+'FJ is the vector of forces at the end of the load step, and
{Su} = SL U1 Vi W1 Oxi Oy1 8,1 Uz Vz Wz Ox2 Oy2 0,2Y (A.28)
in which {Su) is the vector of virtual displacements from the reference configuration.
The condition of deformation independence is satisfied by path independent conser
vative loads, which are the only loads to be considered here. Reference A.2 contains
suggestions for treating forces that do not satisfy this restriction.
From the above it may be seen that the first integral of Equation A.24 produces the
conventional elastic stiffness matrix and the second integral produces the forces acting
on the element in the reference configuration. The components of these may be eval
uated by the methods of the earlier chapters. Our immediate interest is therefore in
y y
(a) (b)
Figure A.4 Stresses and displacements.
the third integral alone. From the definition of strains and the use of Equation A.22
we have
L Tij8ijdV= L CTx8
[ e:r er ( rJ
+
+ :
d
v
f au au aw aw au au av av
(A.29)
+
v [
" --+-- dV +
Txy U
ax ay ] J [
ax ay v
Txz8 --+-- dV
ax az ax az ]
In which the fact that Tyx= Txy and Tzx = Txz and the engineering assumption of no in
plane distortion that results in av/ay = aw/az= 0 have been accounted for.
In using Equation A.29, we shall need the incremental displacements at a point on
the cross section and their derivatives. From Figure A.4, the displacements are
(A.30a)
with the derivatives
av ave aex aw awe aex

-=--z = + y
ax ax ax ax ax ax
au ave av aw
= 0 (A.30b)
ay ax ay ay
au awe av aw
=0
az ax az az
-
We will also convert stresses to stress resultants by integration over the cross section,
giving
My 1 + My
JA
CTxdA = Fx J
A
TxydA = Fy=-
M,1 +
L
M,2
J A
TxzdA = Fz=
L
2
(A.31)
L uxzdA = My L CTxydA =-M, L TxzYdA = aMx - L T,y zdA = (1 - a)Mx
in which the flexural shears are treated as reactions to bending moments, and we define
by a and (1 -a) the portions of the total torque, Mx, resisted by the stresses Txz and
Txy, respectively. For pure St. Venant torsion a= 112, but for reasons to be explained
later, we will retain it as an undetermined parameter for the present.
To summarize the development of the desired terms of the geometric stiffness matrix
we will consider representative components of Equation A.29. For example, using
Equation A.30 in the first integral of the right-hand side of that equation yields
1
- (
2 Jv
Ux
8
r( ) (
au 2
ax
- +
av
ax
- - Z -
aex 2
ax ) ( +
aw
-+ y -
ax
aex 2
ax )] dV
in which the subscript c has been dropped to simplify the notation and the effect of
the higher-order derivatives in au/ax is assumed to be negligible. Expanding the terms
of this integral and using Equation A.31 to convert stresses to stress resultants, we
have
L Fx8
[ e:r e:r ( rJ
+ +
:
dx +
Fp L 8 ;( r dx
(A.32)
f av aex f 8r aw aex
- My
L
8 [ ax ax ]
_
dx - M,
L
L ax ax
] dx
Comparing the first two integrals with Equations 9.4 and 9.16, it is seen that they are
the source of the terms of Equation 9.18 that relate to axial straining and the interaction
of axial force with flexure and torsion. The last two integrals, which remain to be
evaluated, are sources of flexure-torsion interaction terms. To avoid repetition in the
evaluation, we will consider only the last in detail. Noting that for constant shear
M, = -M,1 ( i)1 - (1:)+ M,

2
gives
M' 1
J [ ] (
L
8
aw aex
ax ax
dx _
M,1 + M,
L
2 )J [ L
xs
aw aex
ax ax
dx ] (A.33)
We'll return to the evaluation of the integrals in Equation A.33 after similarly analyz
ing the other integrals of Equation A.29. Using Equation A.30 in the second one on
the right-hand side yields
f" Tx 8
[ ( )(
av
--
au a2w
- - z - - y
a2v
) ( aw
+ ex - + y -
aex
)] dV
y ax ax ax2 ax2 ax . ax
-
Neglecting products of the derivatives of the same variable as of higher order, noting
that for a bisymmetrical cross section L ( 'Txyy)dA = 0, and using Equation A.31 to
convert stresses to stress resultants, we have
J L
F 8
y
-
av
[( )( ) ( )]
ax
au
ax
+ ex
aw
ax
dx _
JL
(l _
o:)Mxl> [ av azw
ax ax2
] dx
Similarly for the third integral on the right-hand side of Equation A.29,
In both of these, the first term in the first integral represents a degree of flexure-axial
force interaction that is assumed to be insignificant and is therefore neglected. The
remaining terms involve torsion-flexure interaction. Adding_ these terms, and replacing
FY and F, with their equivalent end moments results in
-( ) L [ :] ( : ) L [ ]
M,i M,
2 s
ex dx -
M
yi
M
y2 s
ex dx
(A.34)
-L [ ::] L [: :]
(1 - o:)Mx8 dx
+ o:Mx8 dx
Application of the shape function approach to the determination of geometric stiff

ness matrix coefficients can be illustrated by selection of representative terms in Equa
tions A.33 and A.34. For example, we may analyze the terms in which w is the real
displacement component. Thus, first selecting the integrals in which w appears
M,1
J L
8 [ aw aex
ax ax
dx
] _
( M,1 + M,
L
2 )( J [L
xs
aw aex
ax ax
dx
+ ] JL
s
[ ] )ex
aw
ax
dx
av a2w aw a 2v
-
JI.
(1 - o:)Mx8 - -
ax .ax2
[ ] dx +
J L
[ ]
o:M x8 - --
ax ax2
dx
and then retaining only the terms in which w is real, we have
Mz1
J aw
-

u
a ex
-
dx -
( Mz1 + Mz2
) (J ( X
aw
-

u
a ex
-
) dx
+
J- 8()xdx )
aw
ax ax L ax ax ax
a2w av
L L L
-
J (1 - a)Mx - 8 -
dx
+
J
aw a2 v
aMx - 8 - dx
(A.35)
L ax2 ax L ax ax2
a might have a value other than 1/2 and
Later we will investigate the possibility that
the consequences of that event. But since in this section we are limiting consideration
to pure St. Venant torsion, we will letMx equal its equivalent, Mx 2 and a equal 1/2,
and we'll assume a linear shape function for the twist. Cubic functions will be assumed
for the transverse displacements (see Equation 7.9). Thus
()x = LNiJ{Ox} = L(l - ) JL8x1 8x2Y

v = LN3j{v} = L(l - 3e + 2g3) (1 - 2 + e)x (3e - 2g3) -( - e)xj
LVt ()zl V2 8z2JT (A.36)
w = LN3Jfw} = L(l - 3e + 2g3)
- (1 - 2 + e)x (3e - 2g3) ( - e)xJLw1 8y1 Wz 8y2Y
in which = x/L.
Following the procedures of Section 7.2, we have
[ L
L86xJ Mz1 {Nl}LN3Jdx - ( )(L
Mzt Mzz
x{Nl}LN3Jdx + L {Ni}LN3Jdx ) J tw}
(A.37)
[ ( )(L
L8vj -
x2
{N3}LN3J -L )J {N3}LN3J t w}
Integrating and combining terms results in the geometric stiffness matrix
W1 8y1 Wz 8y2
Mx 2 Mx 2
0 0
L L
M _Mz1 M M
M
zi Mz1 + z2 z1 + z2
L 6 L 6
Mx 2 Mx2 Mx2
,0 (A.38)
L L 2
[kg)= Mx2
Mx2
0 0
'L L
Mz2 Mz1 + Mz2 Mz2 Mz1 + Mz 2

L 6 L 6
Mx2 Mx2 Mx2
0
L 2 L
Analysis of the terms in which v and ()x are the real displacement components and
treating the remainder of the integrals of Equation A.29 in the same way will, with
one exception, result in all of the flexure-torsion interaction terms considered to be
significant. These ari the contributions that have been added to Equation 9.18 in the
production of Equation 9.19. The excepted terms, which have also been added to
Equation 9.18, are developed below (Eq. A.48).
.
A.4 FINITE ROTATIONS AND EQUILIBRIUM IN THE DEFORMED CONFIGURATION
A problem not considered in the above is illustrated by Figure A.S. The structure
shown is a plane frame subjected to a transverse load (Fig. A.Sa). Assume that at a
particular load step it has been established that joint b is in equilibrium. The bending
moment, M, at the end of member be is balanced by the torque, T, at the end of
member ab (Fig. A.Sb). If these quantities are determined by conventional engineering
theories for beams and for shafts subjected to St. Venant torsion, each can be repre
sented as in Figure A.Sc, that is, by couples consisting of direct and shearing forces,
M and T/2, acting on the ends of rigid links of unit length. If the joint is now subjected
to a small rigid body rotation, Bx, about the x axis, equilibrium will be destroyed be
cause the incremental moment about the y axis generated by the bending moment is
twice that generated by the torque (Fig. A.Sd).
The problem is subtle because there are several suspects for the source of the
difficulty:
1. Inconsistencies between the conventional engineering theories of flexure and

torsion.
2. Flaws in the linearization of the basic continuum mechanics equation, for example
in going from Equation A.23 to A.24.
(a).
z/ (b) (c)
y y
(Jx
;1 T
z.
1
z
M
\_---
M =-'MOx M = T(Jx
y y 2
(d)
Figure A.S Finite rotation of a joint.
3. The fact that finite rigid-body rotations do not obey the commutative law of
addition and thus are not true vector quantities (see Ref. A.1).
4. The fact that in linear analysis in general, and in the use of the shape function
approach in particular, components of rotation are approximated by displacement
derivatives (see Eq. 2.1).
Several approaches to the treatment of the problem have been proposed. In matrix
analysis most of them involve the formulation of the geometric stiffness matrix (Refs.
A.6-A.9). The following, which is patterned after one in Reference A.9, considers
both the nature of finite rotations and the distinct characteristics of the engineering
theories of bending and torsion.
A.4.1 The Finite Rotation Equation
In 1775, using spherical trigonometry and a fixed coordinate system, Euler developed
the following relationship between the locations of a vector as it is rotated through an
angle <p about an arbitrary axis through the origin (Ref. A.10).
cos C = cos {(cos2 aa a + sin2 cos q;) + cos 77(cos a cos 13(1 - cos q;) - cos ')'sin q;)
+ cos it(cos cos y(l - cos <p) + cos 13 sin <p)
cos rJ' = cos 77(cos2 13 + sin2 13 cos q;) + cos it(cos 13 cos y(l - cos q;) - cos a sin q;)
(A.39)
+ cos {(cos a cos 13(1 - cos q;) + cos')' sin q;)
cos it' cos it(cos2 'Y + sin2 'Y cos q;) + cos {(cos cos cos - cos sin
=
+ cos 77(cos 13 cos y(l - cos q;) + cos aasin q;)

y(l - q;) 13 q;)
in which {, 77, and it, and (', 771, and it' are the direction angles of the vector before
and after the rotation,
a, 13, and 'Y are the direction angles of the axis of rotation, and
q; is the angle of rotation. .
Euler's equations may be put in compact matrix form by referring them to Figure
A.6 and making the following substitutions:
cos
{ a} {I} { } {axay} { ('} {bx}by cos { cos
1 1
cos 13 = m cos 77 = j;j cos 771 = fbl (A.40)
cos 'Y a, b,
n cos it cos it'
Axis of rotation
Figure A.6 Finite rotation in a

z global coordinate system.
Noting that lal = lbl and letting (l - cos cp) = 2sin2 () ,EquationA.39may be written
{b} = [I + sin cpV + 2sin2 V2 }a} (A.41)
in which {a} and {b} are the position vectors of points A and B, and
-I]
-
Further, by scaling the direction cosine vector of the rotation axis by the magnitude
of the rotation, cp7, that is, cp = cpl/ m n J = L 'Px 'Py cp,j we have
in cp sin( cp/2) 2
{b} = [ + s w + ( w2
) ] {a} (A.42)
cp 2 cp/2
in which
(A.43)
'Px
Equations A.41 and A.42 are alternative ways of expressing Equation A.39, the
original Euler equation. All apply to a rotation of any magnitude. For the small to
moderate rotations of interest in normal structural engineering applications, the ap
proximations sin cp = cp and sin cp/2 = cp/2 are satisfactory. Thus, Equation A.41 reduces
to {b} = (T]{a} in which
(T] = (I + W + !W2] (A.44)

It may be seen that it is the third matrix, W2, that makes the difference between
viewing rotations as infinitely small and small but finite.
A.4.2 Application of the Finite Rotation Equation to the Framework Element
To investigate the role of finite rotations in the problem of interest, we start with the
second integral on the left-hand side of Equation A.24
t T;/>e;idV
in which all terms refer to a known reference configuration. As mentioned in the
discussion of Equations A.23 and A.24, for a given displacement variation it can be
integrated to produce an expression of the virtual work of the forces acting on the
element prior to the application of a load step. In doing this, the approximation of
rotation components by displacement derivative.s is used and the finite rotation effect
is not considered. Nothing of consequence with respect to the interaction of axial force
and bending moments or torque is lost in the approximation. But it has been shown
in Reference A.9 that, in the determination of strains, inclusion of the finite rotation
7In view of the noncommutative nature of finite rotations, the vector cp has been designated variously as a
"pseudo vector" and a "rotational pseudo vector" (see Ref. A6). We believe that in the context in which it
is being used here, that is, in the analysis of the incremental response to a given set of loads, the term rotation
component vector is more descriptive and not subject to misinterpretation.
effect as represented in the w2 matrix results in virtual work terms that are of the
same order as those found in the analysis of t T;/31J;idV, the third integral of Equa
tions A.23 and A.24.

Further, these additional terms can be used to explain and rectify the moment-torque
equilibrium imbalance illustrated in Figure A.5 and to demonstrate that, to the order
of terms considered in the analysis, the bending moment-torque interaction is not a
function of the parameter a. A summary of these findings follows.
For the Cauchy stresses and infinitesimal strains of Equation A.25 and the expression
of these strains in terms of displacements in Equation A.4, the above integral becomes
r T; 8e; dV
Jv j j
=
Jv
( )
r <Tx8 au dv + r Txy8 au
ax Jv ay
( +
ax
)
av v
d +
Jv
(
r Txz8 au
az
+ )
aw v
ax
d
(A.45)
To determine the relevant virtual strain terms, consider the finite rotation of a cross
section of a bisymmetrical member (Figure A.7). The section is assumed to remain
plane and the rotation is about the centroid. From Equation A.44
{b} =
[I + W + !W2){a}
in which, as above, {a} and {b} are the position vectors of a point on the section before
and after rotation. The incremental displacement of the point is equal to {b} - {a}, or
{u} =
[W + W2/2]{a}. Now assume that [W) has been accounted for in the analysis of
infinitesimal strains and that the only displacements that are of interest here are the
additional ones associated with the finite rotation effect. Thus let
or in expanded form and in the notation of Figure A.7:
(8x8y)
-(e-; + eD
(8y8z)
Axis of
y rotation
Figure A.7 Finite rotation of a cross-section.

The partial derivatives required in Equation A.45 are

au 1 ( Ox
aoy
+ Oy
aox
y
) Z( +
1
Ox
aoz .
+
(J
z
aox
z
) au au
ax Z ax ax ax ax ay az
=
av - ( ()x
aoz
+ ()z
aox
y
) +!( ()y
aoz
+ ()z
aoy
z
)
ax ax ax 2 ax ax
=
aw !(
oy
aoz
+ ()z
aoy
y ) (
-
ox
aox
+ ()y
aoy
z
)
ax 2 ax ax ax ax
=
Substituting these quantities in Equation A.45 and using abridged notation in which
a prime represents differentiation with respect to x, its right-hand side becomes
Iv (S(()x()y)'y + S(()x()z)'z)dV Iv T;y (6(0x0y) - 28(()x()

+ + ()z();)y + S(OyfJ,)'z)dV
xz
+ { T (S(()x()z) + S(OyOz)' y - 28(()x() + OyO;)z)dV
Jv 2
Using Equation A.31 and noting that for bisymmetrical sections L TxyydA
L TxzZdA we have
= 0,
- L z S(fJxfJy)'dx L Y S(()x()z)'dx L ; S(()x()y)dx

+ +
(A.46)
- L (1 ; ll) Mxll(Oy()z)'dx L z S(()x()z)dx + L MxS(OyfJz)'dx
+
A.4.2.l The Effect of a on the Internal Virtual Work Equation The net effect of
the torsional shearing stress distribution on the internal virtual work expression for
torsion-flexure interaction is represented in the sum of four integrals containing a
the two from Equation A.46 and two from Equation A.34. Thus
-(1; ) L a
MxS(fJy()z)'dx + L MxS(OyOz}'dx
- (1 - a ) L Mxll(v'w")dx + L Mxll(w'v")dx
a
Combining the first two integrals, making the approximations Oy -w' and ()z v' = =
(an approximation that is consistent with our replacement of rotation components by

displacement derivatives in previous equations) and differentiating the result, we have
G )L- a Mxll(v'w" + w'v")dx - (1 - a ) L Mxll(v'w")dx + a L Mxll(w'v")dx

Combining terms gives
!f
2 L
Mxll(-v'w"
.
+ w'v")dx
which is the same result as that obtained by assuming a to be 112, as in the case of
pure St. Venant torsion (see Eq. A.35). Thus, to the order of approximation implicit
in our analysis, the distribution of torsional shearing stress does not enter into the
expression for the differential of internal virtual work. In numerical analysis using the
assumed shape function approach it of course remains a factor, as we shall see when
we consider the case of nonuniform torsion in Section A.5.
A.4.2.2 The Effect of Finite Rotations on Joint Equilibrium The apparent joint
equilibrium imbalance illustrated in Figure A.5 can be explained and rectified by con
sidering the remaining four finite rotation effect integrals of Equation A.46.
L J -f
Integrating the bending moment terms by parts yields
IL JL -f
M' M M'
I L
M
--;f 8((Jx(Jy) 0 + 8((Jx(Jy)dx + -f 8((Jx(J,) 0 - 8((Jx(J,)dx
+ L; 8((Jx(Jy)dx + L ' 8((Jx(J,)dx

But, M =
- FY and M; = F,. Therefore the integral terms cancel and the evaluation
of the portion of Equation A.45 associated with the finite rotation effect reduces to
( -' 8((Jx(Jy) + y 8(0x(J,) ) 1: (A.47)
Letting
x
M, -M,1 + (M,1 + M,z) and
L
=
this becomes
(-M,1 8(0x1 Oy1) + Myl 8(0x10,1) - M,2 8(0x20y2) + My2 8(0x20,2))

This can be written in matrix form as L 88J[kg]{8} in which
(Jxl Oy1 0,1 (Jx2 Oy2 0,2

0 -M,1 My1 0 0 0
0 0 0 0 0
1 0 0 0 0 (A.48)
[k ]
g 2 0 -M,2 My2
=
Sym.
0 0
0
These contributions have also been added to Equation 9.18 in the. formation of
Equation 9.19.
The significance of the terms of Equation A.48 is demonstrated in Example A.l, a
frame studied in Reference A.6. Since the rectangular section is one in which warping
resistance is essentially zero, the frame is also a test of the validity of Equation 9.19,
an equation that only has provisions for handling St. Venant torsion. The elastic tor
sional flexural critical load of 1.093 N found in part 1 of the example (and shown to
be attainable by the accompanying nonlinear analysis) is identical to that reported in
Reference A.6 for the same four-element per member idealization. It is further sup
ported by values ranging from 1.14 to 1.18 N determined in independent finite element
analyses of multi-element models (Refs. A.6 and A.9). The somewhat higher values
of the finite element analyses may be attributed to the fact that they account for the
finite size of the junction at point b, whereas the frame model used in the example
does not. When, as in part 2 of the example, the finite rotation effect terms are excluded
fro f
the geometric stiffness matrix usd i the ana ysis the result is drama ically dif
f
fer nt. The critical load of 0.551 N, which 1s approximately half that found m the first
parlt, must be considered incorrect in view of the independent corroboration of the

hig er value.
!
71,240 N/mm2,
40mm -j
Members ab and be have the cross section shown. E = v = 0.3.
1. (a) Determine the elastic critical load using Equation 8.7 and the [kg] of
Equation 9.19.
,
bl +, ' le 0.6mm
(b) Perform a nonlinear analysis with a lateral load of O.OlP added at point
c.
240mm
I A
t II
2. Determine the elastic critical load omitting the contribution of Equation
A.48 to 9.19.
m
A-A
Model each member as four elements
1. (a) Per = 1.093 N. Buckling is in a torsional-flexural mode.

(b) In second-order analysis, a load-displacement plateau is reached at approximately the critical load.
2. Buckling in a torsional-flexural mode occurs at Per = 0.551 N.

Results: P(N)
1.25
Per= 1 093 N
.
0.75
Eq8.3
Per= 0.551 N
0.5
0.25
00 2 4 6 8 10 w, (mm)
A.5 NONUNIFORM TORSION
The subject of nonuniform torsion was introduced in Section 7.4 and a stiffness matrix
sufficient for the linear analysis of members subjected solely to torsion was developed
there (Eq. 7.28). That matrix will be incorporated in the elastic stiffness matrix required
for the application of Equations 8.3 and 8.7 to nonlinear and critical load analysis of
problems in which nonuniform torsion is considered. First, however, the corresponding
geometric stiffness matrix must be developed. For members of bisymmetrical section
this may be done by adding two degrees of freedom to the framework element and
using a cubic shape function for twist in virtual displacement integrals of Equations
A.32 and A.34. Although approximate, it yields reliable results in typical test problems.
As shown in Figure A.8, the added degrees of freedom are the rates of twist at the
element ends and their corresponding actions are the bimoments defined in Section
7.4. As in Section A.3, we will assume that the applied forces are deformation inde
pendent and express the external work as L'+A'Fj{8u}, with
L'+A'Fj = LFxl Fyl F,1 Mxl Myl M,1 Fx2 Fy2 F,2 Mx2 My1 M,2 B1 B2J (A.49)
Figure A.8 Bisymmetrical 14 degree of freedom framework element.
and
{80} = BLU 1 V1 W1 (}xl Oyl (}zl Uz Vz Wz (}x2 Oy2 (}z2 e;l 0;2Y (A.50)
The relevant internal virtual work integrals are, from Equation A.32
! Fxlp f a( aex) 2dx - Myf a [ av aex ] dx - Mz f a [ aw aex ] dx (A.51)

2 A L ax ax axL ax ax L
and from Equation A.348
(A.52)
Application of the shape function approach and analyzing terms in which w and Ox are
the real displacement components we have, proceeding as in the development of Equa
tion A.37
F
LBOxJ [L P {Ni}LNiJdx } o}
+ [ L
L80xJ Mz1 {Ni}LNiJdx - ( )(L
Mzl Mzz
x{Ni}LNiJdx + L {N3}LN3jdx )} w}
- L8vJ [ (i2)(L {Ni}LNiJdx +L )J

{Ni}LNiJdx {w} (A.53)
in which
Ox =LN3j{Ox}
2 2 2
= L(l - 3g2 + 2e)(1 - 2g + g )x(3g - 2e)-(g - g )xJLOx10;10x20;2Y
(A.54)
8Neither Equation A.32 nor Equation A.34 contains integrals representing internal virtual work attributable
to warping resistance, the birnoment effect, and higher-order axial force-flexure interaction effects. Hence
these effects are implicitly neglected in the development of Equation A.56. They are included in a geometric
stiffness matrix developed in Reference A.8. Comparison of the results of typical problems indicates that
the influence of the neglected terms on the response of bisymmetrical members of practical proportions is
small.
and the other shape functions are as in Equation A.36. Integrating and combining
terms, treating the integrals in which v and 8, are the real displacement components
in the same way, adding the unmodified terms of the basic 12-degree-of-freedom geo
metric stiffness matrix (Eq. 9.19), and including the finite rotation terms of Equation
A.48 leads to the fourteen degree of freedom geometric stiffness matrix of Equation
A.56. Preceding this is the corresponding elastic stiffness matrix in which the torsional
terms are those of Equation 7.28 and the remainder those of Equation 4.34. (See pages
442 and 443 for Equations A.55 and A.56.) ,
Example A.2 illustrates the lateral buckling anct nonlinear analysis of two simply
supported beams subjected to pure bending and, in particular, the effect of pre-buck
ling deflection on the critical load. Warping at the ends is unrestrained in both cases.
From Reference A.11, the critical bending moment is
; E1yc1( 7TZw) 1 +
(A.57)
Mer = ( =t) [ = ( = ) ]
---;==1=- = =;;=l =_ = = = 1 +=7T= = = =w= =
L
in which the numerator is the value determined by the classical theory which neglects
pre-critical displacements (Ref. A.12) and the denominator a term that accounts for
them. In the example it is shown that this effect is significant in the shallow stocky
member whereas in the deeper thinner one it is almost imperceptible. It is also dem
onstrated that, whereas critical loads determined by conventional linear stability anal
ysis (Equation 8.7) do not account for precritical displacements, they are included in
the incremental nonlinear analysis of a model subjected to a small minor axis per
turbing moment.
1. Use Equation A.57 to calculate the critical lateral buckling load for two
sections: aWlOXlOO and a W27X102. E = 29,000 ksi, 11 = 0.3.
2. Compare with critical loads determined from Equation 8.7 and the stiff x
ness matrices of Equations A.55 and A.56.

3. For each section perform a nonlinear analysis with a minor axis moment
of 0.001 M applied at each end.
L WlOXlOO: A = 29.4 in.2, I, = 623.0 in.4, Iy = 207.0 in.4, J = 10.9 in.4, Cw= 5150 in.6
:0 \!7.298 x 1011(1
2
+ 0.2105)
12,300 _ . .
M - - - 15100 kips
_ _
m.
V(0.667)(0.992) 0.813
er
W27X102: A = 30.0 in.2, I, = 3620 in.4, Iy = 139.0 in.4, J = 5.29 in.4, Cw= 24,000 in.6
2io \!2.378 x 1011(1 2.021)

Mer =
V(0.9616)(0.999)
+
=
ll,loo
0.980
= ll,300
.
m.
.
kips
2. Use a four-element idealization. From Equation 8.7:

WlOxlOO Mer = 12,300 in. kips
W27X102 Mer = 11,100 in. kips
3. Comparative results, cfitical loads are from Equation 8.7 and incremen Mx 10-3 (in kips)
tal analysis by Equation 8.3.
15
12.3
} WlOx 100
11.1 } W27x 102

10
0
0 0.5 1.0 1.5 w, (in)
Example A.3 illustrates the phenomenon of elastic torsional-flexural instability of a

beam-column. For the bisymmetrical section member subjected to equal major axis
bending moments applied at the two ends (Case a of the example) it is shown in
Reference A.12 that the critical axial force must be the smaller of the two forces that
satisfy the equation
A
(Pcf - P)(Pct - P) = - M2 (A.58)
Ip
in which Per = 1T2ElylL2 and P01 = A(GJ + 7T2ECwlL2)1IP are the minor axis and
torsional critical loads, respectively. A comparable closed-form solution for unequal
end moments does not exist, but it has been found that satisfactory approximate so
lutions can be obtained by replacing M in Equation A.58 by an equivalent uniform
moment, Meq A simple commonly used expression is
Meq 0.6M1 + 0.4M 0.4M 1 (A.59)

2
=
in which M1 is the larger of the two end moments in absolute value, and moments
causing bending in the same direction have the same signs (Refs. A.13, A.14). In the
example, it is seen that the results obtained by the application of Equations A.589 and
A.59 are in good agreement with those from Eq. 8.7 for the two cases considered:
equal end moments and a moment applied at one end only.
The member shown is a W24X76 section:

r
A = 22.4 in.2, I, = 2100 in.4, ly = 82.5 in.4, p p
M= .05PL M= .05 PL
J = 2.68 in.4, Cw = 11,100 in.6, E = 29,000 ksi, " = 0.3
r
For each loading case:
1. Calculate the torsional-flexural critical load using Eqs. A.58 and A.59.
2. Compare with critical loads determined from Eq. 8.7 and the stiffness matrices L=24'
L
of Equations A.55 and A.56.
M=.05PL
(a ) (b)
9 Although not the purpose of the example, it may be noted that the negative sign of the second critical load
in Case a demonstrates that buckling under a tensile force can occur when it is accompanied by bending of
sufficient magnitude.
References 441
Case a
-rr2(29,000)(82.5) 2.68 -rr2(11,100) 22.4

Per
=
(288)2
=
84.7
2 k Pct - 29,000
( +
2.6 (288)2
)( ) 2182
=
699.7 k
Meq
=
Equation A.59: M
Equation A.58: (284.7 - P)(699.7 - P)

=
G:) (o.05 x 288P)2 = 2.129P2
:. Pc1r
= { :zk
k =
Equations 8.7, A.55, A.56, using four-elemental idealization Pc1r 170 k

Case b
Equation A.59: M,q 0.6M

Equation A.58: (284.7 - P)(699.7 - P) =
(0.36)(2.129 P)2
{+213 k
Pc1r =
+.4000 k
Equation 8.7: Pc1r =
225 k
REFERENCES
A.1 L. E.Malvern, Introduction to the Mechanics of a Continuous Medium, Prentice
Hall, Englewood Cliffs, 1969.
A.2 K.-J. Bathe, Finite Element Procedures, Prentice-Hall, Englewood Cliffs, 1996.
A.3 Y.-B. Yang and S.-R. Kuo, Theory and Analysis of Nonlinear Framed Structures,
Prentice-Hall, Englewood Cliffs, 1994.
A.4 A. E. H. Love, A Treatise on the Mathematical Theory of Elasticity, 4th edition,
1927 (1944 printing), Dover Press, New York.
A.5 Y. C. Fung, A First Course in Continuum Mechanics, Prentice-Hall, Englewood
Cliffs, 1977.
A.6 J. H. Argyris, 0. Hilpert, G. A.Malejannakis, and D. W. Scharf, "On the Geo
metric Stiffness of a Beam in Space-A consistent V. W. Approach," Comp.
Meth. in Appl. Mech. and Engr., Vol. 20, 1979.
A.7 Y.-B. Yang and W.McGuire, "StiffnessMatrix For Geometric Nonlinear Anal
ysis," JI. Struct. Engr., ASCE, Vol. 112, Apr., 1986.
A.8 A. Conci andM. Gattass, "Natural Approach for Geometric Nonlinear Analysis
of Thin-Walled Beams," Intl. JI. Num. Meth. in Engr., Vol. 30, 1990.
A.9 C.-S. Chen, "Geometric Nonlinear Analysis of Three-Dimensional Structures,"
Master of Science Thesis, Cornell University, Ithaca, NY, 1994.
A.10 L. Euler, "Nova Methodus Corporum Rigidorum Determinandi," Novi Com
mentarii Acad. Scient. Imper. Petropolitanae (Russia). Vol. 20, 1775.
A.11 P. Vacharajittiphan, S. T. Woolcock, and N. S. Trahair, "Effect of In-plane De
formation on Lateral Buckling," JI. Struct. Mech., Vol. 3, No. 1, 1974.
A.12 S. P. Timoshenko and J.M. Gere, The Theory of Elastic Stability, 2nd edition,
A.13 W.McGuire, Steel Structures, Prentice-Hall, Englewood Cliffs, N.J., 1968.
A.14 T. V. Galambos, editor, Guide to Stability Design Criteria for Metal Structures,
5th edition, John Wiley and Sons, New York, 1998.
s
U1 Vi W1 Bxi B y1 Bz1 U2 V2 W2 Bx2 By2 8,2 0;1 0.:.2

EA- 0 0 0 0 0
EA
-
- L 0 0 0 0 0 0 0
L
6EI, _ 12E/, 6EI,
12E/, 0 0 0
u
0 0 0 0
u
0 0
L3 L3
12Ely 0
6Ely 0 0 0
12Ely
-L3 0
6Ely
-u 0 0 0
u -u
+
(6GJ 12ECw) +
-(6GJ 12ECw) +
(GJ 6ECw) (GJ 6ECw) +
0 0 0 0 0 0 0 2 2
SL L3 SL L3 10 L 10 L
4Ely 0 .0 0
6Ely 0
2Ely 0 0 0
L u L
4EI, 0
6EI, 0 0 0
2EI, 0 0
L -u L
EA
- 0 0 0 0 0 0 0
L
[k.J = I 12E/, _
0 0 0 0 0
6EI,2
L3 L
12Ely 0
6Ely 0 0 0
L3 u
+
(6GJ 12ECw) +
-(GJ 6ECw) +
Sym 0 0
SL L3 10 L2 -(: 6w)
4Ely 0 \) 0
I
L
4EI, 0 0
L
+
(2GJL 4ECw) -(GIL - 2ECw)
lS L 30 L
(2GJL 4ECw) +
lS L
(A.55)
u, u, w, 8..i 8y 1 8,, u, u, w, 8,2 8,2 8,2 o;I 0;2
F,.2
0 F,,
0 0 0 0 - 1T
L 0 0 0 0 0 0
/,
6F,2 11My1 - M,2 M.,2 F,, 6F,2 M,1 - llM,2 M,, M,2
0 0
M.,2 F,2
SL 10L 0
L 10 SL lOL L 10 10 10
6F ,2 llM,, - M,2 F,2 M.,2 -
6F,, M,1 llM,2 F,2 M.,2 -
0 0 M,, M,2
SL lOL 10 L SL lOL 10 L 10 10
6F,2/p 2M,1 - M,2 2M1, - M,.2
+ +
llM,1 - M12 llM,1 - M,2 6F,2/p 2M,,
2Mz1 M,2 M_,2 F,2/p F,2/p
0
SAL s s lOL lOL SA 10 10 lOA lOA
2 F , 2L M.,2 F,2 M,, + 2M,2 F,2L M.,2 (3Mz1 - M,i)L M,,L
0 0 -
lS L -
10 10 30 2 30 30
2 F,2L F,2 My1 + 2M_,2
M.-2 M., 2 F.iL (3My1 - Myi)L M,1L
0
IS 10 L 10 2 30 30 30
F ,2
0 0 0 0 0 0 0
L
fk.J = I
6F,2 M,1 - llM,2 M,, M,2
M.,2 F,2
0
SL 10L L 10 10 10
6F,2 M,1 - llM,2 F,2 M. .i M,, M, 2

SL 10L 10 L 10 10
6F,2/P M,1 - 2M,2 M,.1 - 2M ,2
--
F ,2/p F,2/ p
-
SAL s s lOA lOA
2F,2L _M,2L (Mz1 - 3M,,)L

0
lS 30 30
2 F ,2L M,2L (M,1 - 3M,,)L

lS 30 30
2F, 2/p F,,Ip
lSA 30A
2F,2/p
lSA
--'
(A.56)
Appendix B
On Rigid Body Motion and

Natural Deformation
Throughout the text we have been concerned with both rigid body motion-displace
ment without change of size or shape of a body-and displacement resulting from its
deformat1on. For example, in Chapter 3 it was shown that both are included in Equa
tion 3.7, the basic global stiffness equation. In Example 8:1, the effect of nondefor
mative rotation of a rigid member on the stability of a system was demonstrated. And
in Section 9.1 a method for incorporating both deformation and rigid body rotation in
nonlinear analysis was developed.
The need to distinguish between the two effects is critical in general finite element
analysis and, in particular, to the efficient solution of large displacement problems. In
the moderate displacement framework analysis problems with which we are concerned,
the rigid body effect is generally small, and the simple force recovery procedure of
Section 12.5.2 is adequate. But to assist in the appreciation of its limitations as well as
to provide further insight into the processes of nonlinear analysis this review of a
method for separating rigid body and deformation effects is offered. It is based on
further analysis of the element considered in Section 12.5.2. Several equations of that
section will be used, either literally or in modified form. For the coherence of the
present discussion they will be repeated here with numbers associated with this ap
pendix and any modifications noted.
B.1 THE NATURE OF THE PROBLEM
In linear analysis, rigid body motion is of concern only to the extent of ensuring that
the internal forces or support reactions are sufficient to provide kinematic stability, that
is, to prevent incipient rigid body motion of all or a part of the structure. This may be
addressed by testing the [Kff] matrix of the global stiffness equation for singularity as
in Problems 3.7 and 3.12, by the application of a Gauss-Jordan elimination procedure
to the system equilibrium equations (Section 6.4 of the first edition of this text), or by
a test such as that described in Section 11.2.2. Once kinematic stability has been as
sured, displacements are determined, and member forces calculated in the routine way
by substitution of the relevant displacements in the element stiffness equation
{F} = [k]{} (B.1)
The vector {F} is a vector of forces in equilibrium on the undeformed configuration.

It is therefore free of any rigid body motion effect. Its validity is only a function of the
underlying theory used in the development of the element stiffness equation.
444
B.2 Distinguishing Between Rigid Body Motion and Natural Deformation 445
Kinematic stability is also a precondition for nonlinear analysis. 1 And in each linear
step of the typical nonlinear analysis, displacement increments are calculated by the
solution of global stiffness equations using methods that are in principle the same as
those of linear analysis. In this case the equation for calculating internal forces in
nonlinear elastic analysis may be written as
J
{2F) = {1F) + [ke + kg]{dnl (B.2)
in which {1F) is a vector of element forces and [ke + kg] the element tangent stiffness
matrix at the start of the step. The vector {dnl consists of displacements determined
by extracting the rigid body motion effects from the total incremental set and {2F} is
the resulting vector of member forces at the end of the step. Equation B.2 is similar
to Equation 12.3lc but it differs in two respects. The first is that whereas the equation
of Section 12.5.2 anticipated a two-step process in which the end-of-step forces were
first referred to the initial configuration and then transformed to the final local axes,
here a direct determination of {2F) is implied. The second is the difference in the
definition of the displacement increments.
We will examine Equation B.2 and, in doing so, we'll assume that equilibrium on
the correct deformed configuration at the start of the load step, as represented by the
forces of {1F), has been achieved. At the end of the step the forces of {2F} are in
equilibrium on the deformed configuration-to the order of the terms included in the
geometric stiffness matrix (Equation 9.19). But the validity of that equilibrium state is
affected not only by any finite element approximation made in the analysis but by the
way in which the incremental displacements are used in the force recovery process.
Those calculated in the global analysis include the effects of both deformation and
rigid body motion. However, any change in the distribution of the components of force
on the member ends is solely a function of its deformation. Thus if, in force recovery,
the rigid body motion effect is not extracted from the displacements calculated in the
global analysis-or some equivalent device employed-the deformed equilibrium con
figuration will be invalid in principle.
B.2 DISTINGUISHING BETWEEN RIGID BODY MOTION

A ND NATURAL DEFORMATION
In framework analysis, the main problem is to distinguish between the natural defor
mation of an element due to stretching, flexure, and twisting, and its rigid body rotation.
There are various ways to do this. We will illustrate features that are common to most
by considering the application of Equation B.2 to the case of a prismatic element of
bisymmetrical cross section subjected to an axial force and bending about its .z axis.
Shearing deformation will be neglected. The approach followed will be that of Ref-
erence B.6. In the summary presented here there are some steps that should be intu
itively correct but are not fully substantiated. Formal justification is contained in
Reference B.6 as well as in parts of References B.1-B.5. Examples of the extension
of the natural deformation approach to three-dimensional elements, twisting, and in
elastic deformation may also be found in these references.
Forces on the element considered are shown in Figure B.1 and deformation and
displacement information in Figure B.2. As in Section 12.5.2, to simplify the notation
we have omitted all subscripts and superscripts except those needed to identify the
'There are cases, as in some suspension systems, in which there is a useful, kinematically stable configuration
remote from an oriinally unstable one. If a method for locating and defining the stable state exists or can
be devised, that state can be taken as the initial configuration for structural analysis.
446 Appendix B On Rigid Body Motion and Natural Deformation
x
Figure B.1 Element forces.
ends of the member and its orientation at the beginning and end of the load step. In
applying Equation B.2 we will use the elastic and geometric stiffness matrices of Equa
tions 4.34 and 9.18 and the following vectors:
1. The forces at the start of the load step:
(B.3a)
2. The increment of forces:
{dF} =LdF0 dV0 dM 0 dFb dVb dMbjT (B.3b)
3. The forces at the end of the load step referred to the final configuration:
{2F} =L2 F0 2V0 2M0 2 Fb 2Vb 2 MbJT (B.3c)
4. The displacement increments transformed to the initial configuration:
{dA} =Lua Va Ba Ub Vb 8bJT (B.3d)
5. Natural displacement increments referred to the final configuration:
(B.3e)
x
Figure B.2 Element displacements.
As may be seen from Figure B.2, the relationships between the components of the two
displacement vectors are
(vh - Va)
ea,, ea - e, ea - tan _1 (B.4a)
(L - Ua)
=
+ Uh
=
(Vb - Va)
(B.4b)
ebn eb - e, eb - tan - I
(L + ub - Ua)
= =
and
u,, 2L - L 7' V(L + Ub - u0)2 + (vb - Va)2 - L
=
(B.4c)
in which, for incremental steps sufficiently small that 2L L =
(ub - Ua)2 + (vb - Va)2

u,, (ub - Ua) +
= (B.4d)
2L
The rigid body motion consists of a translation and a rotation. The magnitude of
the translation is irrelevant but the rotation, en is a function of the initial length of the
element and the displacement increments determined in the global analysis (Eqs. B4a
and B.4b). No work is required in this process. Therefore the 'components of {'F),
which are in equilibrium at the start, remain so. Their magnitude and their distribution
on the ends of the rigidly displaced element are unchanged. But they rotate with the
element and thus the directions of the axial and transverse components follow the
rotation.
The deformation that takes place is a function of the stretch of the member and the
rotation of the two ends with respect to the chord drawn between them. The changes
in end force that result from this natural straining are the components of {dF}. They
are in the same direction as the rotated components of {'F} and, if we use the stiffness
method approach, they are equal to the tangent stiffness matrix at the start of the step
multiplied by the vector of natural displacements, that is, {dF} [kc + kg]{da,,} in
=
which the elements of {da,,} are determined from the relationships of Equation B.4.
Thus, since {2F) {'F} + {dF}, we have Equation B.2.
=
With the updated coordinates, element length, and {2F) in hand, the next step or
iteration can be undertaken with these quantities used in determining the starting
point.
B.3 CRITIQUE OF FORCE RECOVERY METHODS

B.3.1 The Natural Deformation Approach
To illustrate the application of Equation B.2 and test its validity, it is useful to expand
the matrices and cjleck the equilibrium of the elements of {2F). Consider {dF} firs!.
From Equations 4.34 and 9.18:
+ -
dF,, (EA Fb) 0 0 (EA+ Fb) 0 0 0
dVa
(12E/ + 6F6) (6El + FbL)
L2 5 L 10
0
-(12E/ + 6Fh (6E/ + Fh L
L2 5 L
)
10
0
) ----
( + 2bL2) -(6E/ + F6L) (

dMa 1 4E/ 5
0
L 10
2EI - F;2)
(},"'
L +
dF1, (EA Fh) 0 0 u,,
dV h Sym.
(12E/ + 6F"
2
) -(6EI )
+ F"L
0
L 5 L 10
dM"
+
( 4/ 2;L 2)
J eh,,
J
448 Appendix B On Rigid Body Motion and Natural Deformation
or
-(EA+ Fb)un
( 6EJ FbL
)
+ lQ (Ban + Bbn)
L
1
( 4EJ +
2FbL2
) Ban + ( 2EI -
FbL2
) Bbn
{dF} 15 30
(B.5)
L
=
(EA+ Fb )un
( 6EJ
) FbL .
- L + lQ (Ban + Bbn)
( - F;2) Ban + (
2EI
2 bL
5
2 Bbn 4EI + )
It is obvious the {dF} is a vector of forces in equilibrium in the 2x and 2y directions
and since {1F) is also, it is clear that total equilibrium in those directions is satisfied.
To check moment equilibrium, we take moments about point a of the deformed ele
ment (see Fig. B.1)
2-Ma = (Ma + dMa) + (Mb + dMb) + (Vb + dVb) 2L

Noting that (Ma + Mb + VbL) = 0 and (dMa + dMb + dVbL) = 0, this reduces to
2-Ma = Vbun + dVbun -:f. 0
The first unbalanced moment term results from neglect of a degree of flexure-axial
force interaction we believe to be insignificant in the analysis of small strain, moderate
displacement problems (see the development of Equation A.34 in Appendix A). The
effect is included in References B.3, B.6, and B.7 with the result that the equilibrant
of this term appears in the geometric stiffness matrices of each. Since the second un
balanced moment term is normally smaller than the first, it follows that we believe that
it, too, is insignificant and that all of the equilibrium conditions are substantially
satisfied.
It may be noted that, except for a minor difference in the length used for the com
putation of shear force increments, Equation B.5 is identical to the force recovery
procedure proposed in reference B.3.
B.3.2 The Elementary Approach
The elementary or conventional approach to force recovery described in Section 12.5.2

I
may be symbolized as
(B.6)
in which the terms are those of Equations 12.32. The difference between this and
Equation B.2 is in the definition of the incremental displacement vector and the em
ployment of a transformation matrix in Equation B.6. Since it is convenient to use the
elementary approach in small strain, moderate displacement framework analysis, it is
desirable to compare it with the more rigorous natural deformation method. We will
do this by considering the beam element of Figure B.1 situated as in Figure B.3: It is
located along the global X axis at the start of the load step, rigid body translation is
neglected, and it is constrained to rotate about the left end during the step (for sim
plicity, forces are not shown). An informative qualitative comparison can be obtained
by casting parts of Equations B.2 and B.6 in similar terms.
ub
1+--- L --
Figure 8.3 Constrained element.
(}an (}bn
First, using Equations B.4a and B.4b to eliminate and from Equation B.5:
( A Fb)un E +
(-
6EI FbL ) ( a b) ( 12E/ -
-
+- (}
FbL )+ (} - -- + (}
L 10 L 5 r
2 L2
+
2+
(
1 ( 4EI s ) oa (2EI - F; ) ob - 6EI F ) 0,
2 +
( B.7 )
L (EA Fb)un +
(- ( 12E/ - FbL)
L 10 ) ( a b) --
6EI -
-
FbL + (} + (} + + (}
L 5 r
dMb (2E/ - F;2 ) oa ( 4EI F2 ) ob - ( 6EI F2 ) 0,

+ + +
Similarly, using the global displacement components to determine the force incre
ments of Equation B.6:
(B.8)
Also, for the element considered, [Tr] = [ ; J,

[T ]
[T ]
in which
[f'y] =
[ cos 8,
-sin 8,
sin 8,
cos 0,
OJ
0 (B.9)
0 0 1
450 Appendix B On Rigid Body Motion ani' Natural Deformation
Thorough comparison of Equations B.2 and B.6 would require systematic analysis
of the influence of the independent variables in these three equations, but the following
practically useful conclusions can be drawn from their visual inspection:
1. The total rotation ( and thus the rigid body rotation ) or members of practical
structures will invariably be small, even at the limit of resistance. Therefore, for
.
most cases [fr] will approach a unit matrix.
2. The only difference in the incremental end moment terms of Equations B.7 and
B.8 is the presence of Br in one and vb/L in the other. Even for Br equal to
10 degrees-a very large value-the difference is insignificant.
3. For small stretching, Un = ub ( see Eq. B.4d ) and the difference in the computed
axial force increments of Equations B.7 and B.8 will be small.
4. Aside from the negligible difference in the representation of Bn the incremental
end shear recovery terms of Equations B.7 and B.8 differ in an axial force con
tribution: FbL/5 vs. 6FbL/5. This is the result of the rigid body rotational effect
of Fb x vb which has been eliminated from Equation B.7 but is present in Equa
tion B.8 ( see discussion of the external stiffness matrix in Reference B.6 ) . It may
be seen, however, that letting sin Br= vb/L in the transformation matrix fir] of
Equation B.9 when transforming initial forces to the final configuration by the
elementary approach results in a shearing component Fb Vb/L that essentially
nullifies the difference.
REFERENCES
B.1 T. Belytschko and B. J. Hsieh, "Non-Linear Transient Finite Element Analyses

with Convected Coordinates," Intl. fl. Num. Meth. in Engr., Vol. 7, No. 3, 1973.
B.2 J. H. Argyris et al., "Finite Element Method-The Natural Approach," Comp.
Meth. in Appl. Mech. and Engr., Vol. 17/18, 1979.

B.3 M. Gattass and J. F. Abel, "Equilibrium Considerations of the Updated Lagran
gian Formulation of Beam-Columns with Natural Concepts, Intl. fl. Num. Meth.
in Engr., Vol. 24, 1987, pp. 2119-2141.
B.4 A. Conci and M. Gattass, "Natural Approach for Thin-Walled Beam-Columns
with Elastic-Plasticity," Intl. fl. Num. Meth. in Engr., Vol. 29, 1990, pp. 1653-1679.
B.5 A. Conci and M. Gattass, "Natural Approach for Geometric Non-Linear Analysis
of Thin-Walled Frames," Intl. fl. Meth. in Engr, Vol. 30, 1990, pp. 207-231.
B.6 Y.-B. Yang and S.-R. Kuo, Theory and Analysis of Nonlinear Framed Structures,
B.7 Y.-B. Yang and W. McGuire, "Stiffness Matrix for Geometric Nonlinear Analy
sis," fl. Struct. Div., ASCE, Vol. 112, No. 4/April, 1986.
Author Index
Pages in italics indicate an end-of-chapter reference to a Ellis, C. A., 223,241

published work. Euler, L.,223,241,432,441
Abel,J. F., 278,299,358,374,445,448,450 Forsythe, G. , 331,333,338

Adams,P. F., 286,299 Fox, L.,302,337
Almroth, B. 0., 231,241 Frazer, R. A., 3,8
Argyris,J. H., 3,8,137,173,413,418,432,433,436,441, Fung, Y. C. , 268,424,441
445,450
Arnold, P., 286,299 Galambos, T. V., 223,241,278,280,299,440,441
Atsuta, T., 277,299 Gallagher, R. H., 2,4,8,181,191,214
Attalla, M. R., 290,291,292,299 Gattass, M., 317,338,432,438,441,445,448,450
George, A., 317,337
Baker,J. F., 3,8 Gere,J. M., 187,214,231,233,239,241,253,256,259,260,
Barsoum, R. S., 191,214,215 265,267,268,439,440,441
Bathe, K. J., 2,4,8,46,55,236,241,244,268,272,299, Glockner,P. G. , 403,405,418
307,317,337,338,351,369,373,374,420,423,424, Gloudeman,J. F., 326,338
425,426,427,441 Golub, G. H. , 2,8,304,314,317,326,329,337
Beer, F. P., 95,136 Goodier, J. N., 57,75,92
Belytschko, T., 445,450 Goodman,J. , 3,8
Bergan,P. G., 350,351,359,373,374 Goto, Y., 394,418
Bjorhovde, R., 394,418
Boley, B. A., 252,265,268 Hancock, G. J., 346,351,352,373
Brown, D. M., 223,241 Heyman,J., 3,8
Brozzetti, J., 394,418 Hilpert, 0., 432,433,436,441
Brush, D. 0., 231,241 Hodge,P. G., 276,299
Buchanan,J. L.,313,328,329,337 Hoff,H. J. , 252,265,268
Horne, M. R., 3,8
Carter, C. J., 187,191,193,214 Horrigmoe, G., 351,373
Castigliano, A., 2,8, Hsieh, B. J., 445,450
Cayley, A., 2,8 Hsu, C. T., 294,300
Chen, C. S., 432,433,436,441
Chen,P. F. S., 293,300 Irons, B. M., 321,338
Chen, W. F., 277,299,403,418
Clarke, M. J., 346,351,352,373 Johnson, E. R., 95,136
Clough, R. W., 3,8,350,373
Collar, A. R., 3,8 Kaufman, S., 252,265,268
Colson, A., 394,418 Kavlie, D., 413,418
Conci, A. , 432,438,441,445,450 Kelsey, S., 3,8,137,173
Crisfield, M. A., 349,351,359,373,374 Kirsch, U., 413,418
Croll, J. G., 231,241 Krakeland, B., 351,373
Cross,H., 3,8 Krieg, D. B., 358,374
Krieg, R. D. , 358,374
Deierlein, G. G., 216,241,290,291,292,299 Kuo, S. R., 349,351,373,420,441,445,448,450
DenHartog,J.P., 77, 92
Dill, E. H., 3, 8 Lee, S. L., 371,374
. Duncan, W. J., 3,8 Levy, S., 326,338
451
452 Author Index
Liu,J. W. H., 317,337 Ripperger, E. A.,137,164,173,244,268

Love,A. E. H., 121,136,422,424,441 Roark,R. J., 205,215
Lowder,H. E., 413,418 Rockey, K. C.,191,215
Lu,L. W., 286,299 Rosanoff,R.A.,326,338
Lui,E., 403,418 Rossow,E. C.,371,374
Luik,R. 413,418 Roy,J.R.,413,418
Rubenstein,M. E, 413,418
MacGregor,J. G., 294,300
Malejannakis, G.A.,432,433,436,441 Scarborough,J.B., 310,337
Malvern,L. E., 420,421,423,424,432,441 Scharf,D. W., 432,433,436, 441
Maney,G.B., 3,8 Seaburg,P.A.,187,191,193,214
Manuel,F. S.,371,374 Sobieszczanski,J., 413,418
Martin,H. C.,3,8 Sl!!reide,T. H.,351,373
Mattiasson, K., 370,374
Maxwell,J. C.,2,8 Timoshenko, S. P.,2,8,57,75,77,92,187,205,214,215,
McGuire,W., 191,215,216,241,253,268,278,299,290, 226,231,233,239,241,253,256,259,260,265,267,
291,292,299,358,374,432,440, 441,448,450 268,439,440,441
.
Melan,J., 3,8 Topp,L.,3,8
Melosh,R. J., 3,8,413,418 Trahair,N. S.,439,441
Mendelson,A.,270,299 Turner,M. J., 3,8
Menezes, I. F. M.,317,338 Turner,P.R.,313,328,329,337
Miyashita, S.,394,418
Mohr, 0., 2,8 Utku,S.,78,92
Moler, C.B., 331,333,338
Mroz, Z.,292,299 Vacharajittiphan,P.,439,441
Mukherjee, S.,317,338 VanLoan,C.F.,2,8,304,314,317,326,329,337
Munoz,P.R.,294,300 V lasov,V. Z.,187,214
Nardo, S. V.,252,265,268 Walker,A. C.,231,241

Nethercot,D. A.,191,215 Wilbur,J.B., 78,92
Noor,A. K., 413,418 Wilkinson,J. H.,323,338,
Norris,C. H., 78,92 Wilson,E. L.,317,338,369,374
Woolcock, S. T.,439,441
Oden,J. T.,137,164,173,244,268
Oldfather,C.A.,223,241 Yang,Y.B., 191,215,349,351,373,420,432,441,445,448,
Orbison,J. G., 278,299,358,374 450
Ostenfeld,A.,3,8 Young,W. C.,205,215
Paulino, G. H., 317,338 Ziegler,H., 231,241

Pesquera,C. I., 354,374 Ziemian,R. D., 216,241
Porter,L.,275-276,299 Zienkiewicz,0. C.,349,373
Powell, G. H., 275-276,293,299,300,413,418 Zyckowski,M., 270,299
Przemieniecki,J. S.,137,173
Subject Index
A.ccuracy,301,310,322-326 Beam element,66,67

nonlinear solution,341,351,359 bending and axial force,245,427-430,442
A.dmissibility,143,144,159,161 bent about y axis, 72,73
A.lgorithm, bent about z axis, 68-72
backward substitution,303 curved,207
Cholesky banded,316 flexible end-connections,394
Cholesky decomposition,308 flexibility matrix,70
forward substitution,303 nonprismatic,177,185,202-204
Gauss decomposition,305 offset ends,399
invert lower triangular matrix,363 shearing deformation,74,204
root-free Cholesky,308 sign convention,66,72
A.nalysis levels,216 stiffness matrix,70,73,246,394,397,398,401
elastic critical load,218,234,247,256,259, 260, temperature gradient,126,197
439-441 Beam on elastic foundation,76,87
examples,247-254,256,259-263 Betti's law,62
elastic first-order (linear elastic),218,234 Bifurcation,218,219
(see also examples in Chapters 2-7) Bifurcation points,237,351,359,360
elastic second-order,218,235,247,248 Bimoment,188,437
examples,247-254 Boundary conditions,224
inelastic critical load,218,235,278,280-282 displacement,40,147
examples,279-282 plane of symmetry,404,405
inelastic first-order,219,234,274-277,282,284,285,287 Boundary element,392
examples,282,284,285,287,288 Bracing
inelastic second-order,219,235,274-277,279,282, column,254
284-287,290 frame,262
examples,279,282,284,285,286,287,290 lateral,262
plastic analysis,219,228,274 Buckled configuration,p5
A.ntisymmetry,113,119,375,403-408
A.pproximate solution methods,153,167,409,412,413 Characteristic.equation, 329, 363
A.pproximation of geometry,185,393,398 Characteristic values,329
A.re length methods,349 Cholesky decomposition, 307-309, 316, 361, 363
A.rches,208,222,248 root-free,308,359
A.rtificial nodes,108 Cofactors,302
A.rtificial skewness,330 Compatibility,33,144,159,164,217
A.ssembly of element matrices,31-39 Complementary strain energy,59, 63, 76,77,200,203
A.utomatic load incrementation,351,352 Computational efficiency,301,304,344,349
A.xes,see Coordinate systems Computer core storage,316, 317, 321, 376
A.xial force element,16-26,67,125,142,146,150,153,159, Computer programs,3-7
165,176,180,181,183,185,200,202,243 automatic numbering schemes,317
general-purpose programs,4,7,406
Backward substitution,process of,303-305 input-phase,4,323,392
Band solver,316,317 interactive graphics,6
Banded storage,316,317 MA.STA.N2,6
Bandwidth,47,301,316,317,368 MA.TLA.B,7
minimization of,47, 317 output-phase,6,323
Basis vectors,312-315 postprocessing,7
Beam-columns,223,225,228,229,249-252,261,282,442 preprocessing,6
453
454 Subjectlndex
Computerization, 3 Deformations,
Condensation of matrices, 307, 376, 377 axial, 66, 100, 243
connections, 394, 396, 397 axial and bending, 245, 426
Condition number, 326-329, 360 axial and torsional, 255
approximate, 329 bending and torsion, 434
bound,338 flexural, 66, 69, 243
eigenvalues, 329 natural, 445-450
estimating, 327 neglected, 74, 403
Congruent transformation, 64 plastic, 274-276
Conjugate gradient method, 312-315 torsional, 66, 68, 74
with preconditioning, 314, 331 St. Venant, 255, 438
Conjugate vector, 99 warping, 255, 438
Connections, transverse shearing, 66, 74, 204, 403
classification, 393, 394 Deformed geometry, updating, 352
flexible (modeling), 393-398 Degrees-of-freedom,9-11
geometry, 398 active variables, 321
pinned (unrestrained), 393, 397, 398 element numbering system, 32, 318
semi-rigid (partially restrained), 229, 394 global numbering system, 32, 317
stiffness, 394 juncture, 379
Constitutive relations rate of twist, 68, 188, 437
elastic moduli, 56, 57, i48, 427 rigid-body-motion, 32, 324, 355
tangent modulus, 223 rotational (angular), 75
Constraint, Determinant, 302, 329, 363
displacement,385 Direct stiffness method, 9
equations, 386 basic equations, 31-39
force release and, 397, 398 general procedure, 39-46
Continuity of displacement, Direction angles, 96, 106, 107, 432
at node points, 31-33 Direction cosines, 96, 106, 107, 390, 432
see also Compatibility Discretization of structures, see Idealization of
Continuous beam, 108-112 structures
Continuum structures, 2 Displacement compatibility,see Compatibility
Contragredient transformation, 99, 400 Displacement control, 348, 349
Convergence, Displacement method, 9
criterion, 310, 314, 350, 351, 365, 368 Displacement variables-, 75, 76
current stiffness parameter,351 Displacements,
improving, 311, 331, 366 actual, 144
iterative, linear equations, 310-315 angular, 10, 12, 75, 431
iterative, nonlinear equations, 350 generalized, 158
number of iterations, 352 incremental, 234, 339-349, 422
Coordinate systems, initial, 120-122
global, 13 joint, 10
joint, 388-393 large, 220, 222, 248
local, 13 magnitude, 220
right-hand orthogonal, 11-14 moderate,220
Cost of computations, 301, 304, 344, 349, 351, 375, 413 physical, 158
Cramer's rule, 302 real, 144
Critical load analysis relative, 393
classical examples, 220, 223, 225, 233 rotational, 243, 431
elastic, 218-219, 235, 360-369 translational, 9, 11, 420, 422
inelastic, 218-219, 235, 278, 361 virtual, see Virtual displacements
solution schemes, 339, 360-369 Distributed loads, 93, 108-120, 148, 160, 194, 287
Current stiffness parameter, 351 Drift-off error, 236, 344
Duality, forces and displacements, 137, 158
Decomposition theorem, matrix, 302 Dynamic analysis, 10
Deflection, see Displacements
Deflection stability (shakedown), 288 Eigenproblems, 339, 360-369
Deformation gradient, 422 deflation, 366
Subject Index 455
general form, 360 node point, 33

Hotelling's method, 366 solving nonlinear, 339-369
inverse iteration, 367 Equilibrium matrix, 63
polynomial expansion, 363 Equivalent nodal loads, 110, 112
reduction, 361 Error analysis, 322-335
solutions, 369 Errors,
standard form, 360 accuracy and, 322-335, 344
Sturm sequence, 366, 367 amplification, 327
Eigenvalues, 235, 307, 329, 360-369 corrective procedures, 333-335
maximum, 364 detection of, 326, 333
minimum, 368 drift-off, 236, 344
Eigenvectors, 235, 360-369 estimates, 329,330
normalizing, 364 idealization, 322,323
Elastic behavior, 1, 11, 14, 56, 138 iterative improvement, 333, 334
Elastic constants, 14, 56, 69, 175, 427 major sources of, 322-323
Elastic supports, 76, 87 manipulation, 323-326
Elastic unloading, 277, 358 precision, 334
Element end releases, 397, 398 process, 323, 344
Element forces, see Internal forces propagation, 334
Element matrix, relative, 327
flexibility, 1&_, 62-66 round-off, 323-326
internal force, 41 tolerance, 310, 314
stiffness, 16, 62-73, 394, 397, 398, 401 truncation, 323, 326, 330
Element relative displacement, 393 Euclidean norm, 326, 350
Energy, see Strain energy Euler method, 341
Equation solving (linear), 301 External complementary virtual work, 163
accuracy, 301, 322-335
backward substitution, 303, 304 False position method (regula falsi), 356
band solver, 316, 317 Finite element method, 1
Cholesky thod, 307-309 framework analysis and, 3
Cramer's rule, 302 types of element, 5
direct inversion, 302 Finite joint size, 393, 403
elimination methods, 302-309 effect of, 402
equation condition, 323-326 Finite rotations, 431-437
error analysis, 322-335 effect on internal virtual work, 435
forward substitution, 303 effect on joint equilibrium, 431, 436, 437
frontal solver, 317-322 Euler equation, 432
Gauss elimination, 303-307 rotation component vector, 433
Gauss-Seidel iteration, 310-312 Fixed-end forces, 93, 110, 112, 122, 125, 127, 148, 382
Iterative improvement, 333, 334 Flexibility equations, 16, 18, 19
Jacobi iteration, 310, 311 Flexibility matrix,
Equation solving (nonlinear), 235-238 axial force element, 19
single-step, 236, 339-345 beam element, 70, 203
iterative, 236, 339, 345-350 element, 18, 201, 202
modified iterative, 350 ring beam element, 207
Equations, linearization of, 425, 426 tapered axial member, 202
Equilibration of matrix, 330-332 tapered beam, 203
Equilibrium, 32, 139, 144, 153, 154, 159 torsional member, 68
on deformed state, 219, 237-239, 420 Flexibility method, 9
on undeformed state, 217, 237 Flexibility-stiffness transformation, 62-65
stability of, 231 connections, 394, 395
Equilibrium equations, Flexible end-connections. see Connections
handedness, 316 Floating point operations (flops), 304
conjugate gradient, 313 Floor systems, 398
differential, 160 Force points, 229, 275
element, 63 Force recovery,
linearization of, 425 conventional (elementary), nonlinear, 352-355
456 Subject Index
Force recovery (Continued) connections, 393

elementary vs natural approach, 445-450 joint size, 393, 398
Force redistribution, 229, 248, 252, 262, 283, 285, 286, 288-289 Ill-conditioning, 323-326, 359
Force release, 148, 398 artificial, 330
Force-displacement equations, 16 framed structures, 324
element, 33, 199 measure of, 326, 327
Forces, Imperfections, 217, 279, 359
element, 16, 40, 41 Incremental solution schemes, 234-237, 339-359
equivalent, see Equivalent nodal loads external, automatic load incrementation, 351, 352
external, 33 iterative, 236, 237, 339, 345-350
fixed-end, see Fixed-end forces number of iterations, 352
generalized, 12 modified iterative, 350
internal, 16, 40, 41 single-step, 339-345
offset element end, 399-401 step size, .140, 346, 351, 352
redundant, 48 Indeterminacy,
unbalanced, 236, 346, 351 static and kinematic, 47-49
Forward elimination, process of, 304, structure-cutting concept, 48
Framework element, 66-73, 97, 426-443 Indeterminate structures, 47, 121
flexibility matrix, 18, 63, 67-70 Inelastic analysis (see Analysis levels)
sign convention, 16, 66 elementary members, 226-229
stiffness matrix, 73, 257, L.)8, 442, 443 Inelastic properties
Frobenius norm, 326 plastic hinge, 226, 228
Frontal solver, 317-322 plastic moment, 226, 228, 272
active variable, 321 squash load, 272
advantages, 321 tangent modulus, 223, 278
Infinity ( uniform ) norm, 326
Gauss elimination, 304-307, 359, 377 Influence coefficients,
nai:Ve, 307 flexibility, 18
Gauss-Seidel iteration, 310-312 stiffness, 16
Geometric stiffness matrix Initial strain, 120, 122, 196, 236, 425
axial force member, 243-245 Instability (see also Stability)
combined bending and axial force, 245, 246, 426-430, buckled shapes, 223, 225
437-443 equation solving, 308
combined bending and torsion, 258, 429-430, 431-437, flexural buckling, 223, 225
437-443 lateral buckling, 259, 260, 439
combined torsion and axial force, 254-256, 426-430, limit points, 219, 358
437-443 sidesway and no sidesway buckling, 282
three dimensional, 257, 258, 426-443 torsional flexural buckling , 261, 440
Global analysis, framework, 39-41 Interaction equations, 273
Global axes, 14 Internal complementary virtual work, 163, 200, 203
Global flexibility coefficients, 41 Internal forces, 16, 40, 41
Global stiffness coefficients, 33 relation to joint forces, 201, 399-401
Global stiffness matrix, 9, 13, 33, 40 Internal virtual work, see Virtual work, internal
equation solving, 305, 308, 313 Interpolation, 180, 181
Green-Lagrange strains, 420, 423, 424 Inverse, 302
Grid structures, 116 lower triangular matrix, 363
Inverse iteration, 329, 367
Half-bandwidth, 316, 317 Iterative improvement, 333, 334
Hinges, Iterative, incremental-, 236, 237, 339, 345-350
element ends, at, 397, 398
internal, 86 Jacobi iteration, 310, 311
Hooke's law, 56, 57 Joint coordinates, 388-393
Hotelling's method, 366 Joint displacements, see Degrees-of-freedom
Joint equilibrium, 32-35
Idealization of structures, 4, 14-16, 322, 323, 333, 375, 376, equation solving, 333, 346, 351
431 Joint loads, 33
Subject Index 457
Joint size, see Finite joint size norm,326,350

Joints,9,393,see also Node points orthogonal,96, 313,324
positive definite,307,359
positive semidefinite,307
Kinematic indeterminacy,47-49
rotation,95,96,433
Kinematic instability,
scaled,333
equation solving, 308
singular,323,359,360,363
symmetrical,47,301,305, 307,308,316,361
Large displacement theory,220,222 transformation,95,98
Lagrangian formulations,420,424-426 transpose,13
Limit point,219,237,349,358 triangular,302,303,307,361-363
post-limit,358,359 unit triangular,303
snapback,349 Maxwell's reciprocal theorem,61
snapthrough,221,349 Mechanism behavior,228
Limit states Member forces,see Internal forces
nominal strength,280 Member,see Elemerit
plastic,219 Minor,of determinant,302
serviceability ,216 Mixed force-displacement equations,16
strength,216 Models,analytical,6
Linear algebraic equations, 301-335, 340 Modified iterative,350
Linear dependence,323 Multi-step methods,see Iterative, incremental
Load control method,347
Loads, Natural deformation,444-450
antisymmetric,404 Neutral equilibrium,231,360
between nodal points,93,108,194,287 Node points,12
distributed,93,108,148,160,194,287 artificial, 108
equivalent nodal,110,112,195 interface,379
fictitious nodal,379,382 interior,379
lack-of-fit,120,121 Nonlinear analysis,
reference,235 aims of,216,217,419
self-straining,120-130
classical examples,220-230
statically\equivalent,379
connections,397,398,403
symmetri. 404
equation linearization,425
thermal,93
geometric nonlinear,242,257-258,419
types,93
in design,216
work-equivalent, 195
levels of,218,219
Local axes,13
material nonlinear,269,274--278
matrix method examples,246-254,256,259-263,
Material elastic stiffness matrix, 148,199 279-290,437,439-441
Matrices, solution methods,235,236,339-369
congruent transformation of,64 stiffness method equations,234-235
decomposition of,302-309,361 see also Analysis levels, Geometric stiffness matrix,
equilibration of,330,331 Plastic reduction matrix
factored,302,303,305,306,361 Nonlinearity
inversion of,41,302,327,328,363 geometric,217
numerical condition of,314,323-326 geometric and material,217
sparse (weakly populated),47,301,316,317 material,217
Matrix of elastic constants,148,199 sources of,
Matrix structural analysis,finite element analysis and,3 combined effects,218,398,403
Matrix, geometric effects,217
adjoint,302 material effects,218
banded,301,316,317,368 Nonuniform elements, l84-186,202
consistent,326 Normality criterion, 270,274,358
diagonal, 302, 331 Normalizing, 364
equilibrium, 63 Norms,326-328,350
indefinite, 307,308 maximum,350
458 Subject Index

Norms (Continued) Principles,symmetry and antisymmetry,113,119
modified absolute,350 of antisymmetry,375,403-408
modified Euclidean,350 of symmetry,375,403-408
Numerical condition,314,323-327
ill-conditioned,323-326,359 Quadratic form,59
well-conditioned,323
Rayleigh-Ritz method,157
Offset element ends,see Finite joint size Reaction elements,87,392
Orthogonal matrix,96,313
Reactions,40,41
Orthogonality,324 Reanalysis techniques,376, 408-413
Reciprocity,60-62
Panel zone,398,403 Redundancy,static and kinematic,48
Partially restrained connections,see Connections Redundant forces,48
P-delta effect,218,220,224,251,252,284,285,288 Reference states,237,238,420
Pinned connection,393,397,398 deformed configuration,219,420,427
Piola-Kirchoff stresses,420,423 undeformed configuration,217,219,420
Pivoting,307,333 Relaxation schemes,311,366
Plastic analysis Release,13,148,393-398
design by plastic analysis,273 Residual stress,278,288
distributed plasticity,226,273,290-292 Residual vector,314
interaction diagrams,229,272,273 Rigid body displacement,18
limit load,219,228,273 Rigid body motion,18,35,148,180,243,324,355,379,
mechanisms,228,273 444-450
multiple yield surfaces,293 rotation,243,355
plastic hinge,220,272 translation,243
plastic reduction matrix,274-277 Rigidframes,251,252,282,284,285,286,287,290,406
plastic strain,271 Rigid links,399-403
reinforced concrete,293 Rootfinding,356
stress resultant yield surface,272,277,290-294 Rotation matrices,95,96,433
Plastic hinge constraints,355-358 Rotational springs,394
Plastic reduction matrix,274-277,283-284 nonlinear,397

Plasticity theory,270-273 Round-off error,323-326
flow rule,270 Runge-Kutta methods,340-345
hardening rules,271 advantages,344
normality criterion,270,274 first-order,341
von Mises criterion,270 limitations,341
yield locus or function,271 mid-point, 342
yield surface,271,272,274,358 second-order, 341-344
gradient to,275
Polynomial expansion, 363,364 Scaled matrix,333
Post-limit state,349 Self-straining,120-130,194,236,425
Potential energy, Shape functions,
stability of equilibrium,and,231-234 in linear analysis,175-181
total,231 in nonlinear analysis,245,246,256,430,438
variation of,231 Shear center,66,398
virtual work,and,231 Shear deformations,398
Pre-conditioning of matrix,330-332 Sign convention,9,12,13,16
conjugate gradient method,and,314 Simultaneous linear equations,301-335
Predictor-corrector method, 342 Single-step methods,236,237,339,340-345
Principal _cross-sectional axes, 66 Singular matrices, 323,359,360. 363
Principle(s), Skewness, 330
of virtual displacements, Slope-deflection equations, 76, 78
for deformable bodies, 142. 181. 183, 244, 421,425,427 Small displacement theory, 223, 224
for rigid bodies, 137,139 Snap t hrough, 221,2 22,349
of virtual forces, 137, 158. 162.164. 199 Solution accuracy. 301, 322-326
Subject Index 459
Solution methods,approximate,153,167 substructure, 378, 379

Solution methods (linear equations), tangent,234
direct,302-309 tapered axial member,185,186
iterative,310-315 torsional member,68,190, 257, 443

Solution methods (eigenproblems), 360-369 see also Geometric stiffness matrix, Plastic reduction
inverse iteration,367 matrix
polynomial expansion, 363 Stiffness method,9,

power method,364 in nonlinear analysis, 216
Solution methods (nonlinear), Strain,9, 57, 69, 421
advantages,344 finite,243, 245,255,422-426
arc length,349 Green-Lagrange,420,422-426
displacement control,348, 349 infinitesimal, 57,421
Euler,341 initial,120,122,196,425
incremental-iterative, 236,237, 345-350 plastic,271,275,276
limitations,344 magnitude of, 220
load control,347 Strain-displacement equations,146,181,185
mid-point Runge-Kutta,342 Strain energy,
Newton-Raphson,347 element,59,99
predictor-corrector,342 quadratic form,59
simple step,236,237,341 virtual,143,145
single step,339-345 see also Complementary strain energy
work control,349 Stress,9, 57,69
Space frames and trusses,104,263 Cauchy (physical),421,424
Sparseness,301,316,317 Piola-Kirchoff,420,422,424-426
Stability,230-233 residual,278, 288
bifurcation of equilibrium,218,359 Stress-strnin relationships,56-58,69,148
critical loads,218,220,247 Structural analysis,
deflection stability,288 brief history,-2
eigenvalues,218 computerized,3
eigenvectors,218 conditions of, 11-13
elastic stability limit,219,248,249 matrix theory,3
inelastic stability limit,219 Structural joints,see Finite joint size
kinematic stability,308 Structures,
linear stability analysis,235 analytical model,6,12
potential -;:nergy criteria,231 determinate,48
stable, neutral,and unstable equilibrium,231 framed,1,--15,56
Static condensation,307, 376, 377, 394, 396 grid,116
Static indeterminacy,47 idealization,error in,322, 323
Statically equivalent loads, 379 indeterminate,48, 121
Step size,351, 352 modifications to,409
tiffness equations, rigid frames,10,102,406
element,16-18,183 self-strained,120
global,19,33,40, 313 space truss,104
nonlinear solution,339 symmetrical,403
Stiffness matrix, Substructuring,5, 307, 375, 377-385
axial force member,67, 245 Superposition,principle of,11,404
beam element,70, 73, 204,205,246 Support conditions,40,62,151,224, 408
connection,effect of, 397, 398 Support settlement,76, 82, 124
element, 16-18, 183 Supports,elastic,76, 87
framework element,73, 257,258, 442, 443 Suspension systems,221
geometric,234, 235, 257, 258, 443 Symmetry,principle of,113, 119, 375, 403-408
global,19,33, 40, 47
joint size,effect of,401 Tangent modulus,223, 278
plastic reduction, 234, 276 Tangent stiffness matrix. 234. 236. 238
shear deformation effect,204, 205 Tapered members, 153, 161. 179, 184, 185, 202, 203
460 Subject Index
Temperature effects, 125,197 conjugate, 99

Torsion, conjugate direction, 313
bimoment, 188,437,438 norm, 326
degrees-of-freedom, 68,188,437 residual, 314
nonuniform, 66,187-190,437-441 row, 13
simple (St. Venant) torsion, 67,188,254,426-430, Virtual displacements, 137,138,140,144,244,255,420,
431-437 421,425
warping torsion, 74,187,437-441 and total potential energy, 231
Torsional member, 67,68,74,186-194,254-257,426-439 Virtual forces, 137,159,161.
Total potential, 313 Virtual strain energy, 143
Transformation matrices, 94-98,100,390 Virtual work,
force, 400 external, 142,148
Transformation(s) internal, 142,143,145-147,244,245,255,421,424
congruent, 64 principle, 137,169,425
contragredient, 99,400 see also Complementary virtual work
coordinate, 93,94,390
degrees of freedom, 93,98 Wave front solution, 318-322
energy approach, 98-100 Work
flexibility-stiffness, 62-65 and energy, 58-60
force, 93,96,399 external, 99,142
stiffness matrices, 98,390 internal, 142
Truncation, 323,326,330 virtual, see Virtual work
Work control, 349
Unbalanced load, 346,351 Work-equivalent loads, 110-112
Unit load method, 166
Updating deformed geometry, 352 Yield surface, 271-272,274-278
breaching, 356
Vectors, elastic unloading, 358
basis, 312-315 isotropically expanding, 35
column, 13 moving tangent to, 357
consistent, 326 tolerable, 358

Matrix Structural Analysis 2nd Edition PDF

Uploaded by

Copyright:

Available Formats

Matrix Structural Analysis 2nd Edition PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Matrix Structural Analysis 2nd Edition PDF

Uploaded by

Copyright:

Available Formats

2%/? &* =? ? 1/7? /?

Preface to the Second Edition vii

student become attuned to the overall approach to computerized structural analysis in

1.1 A Brief History of Structural Analysis 2

Chapter 2 Definitions and Concepts 9

2.1 Degrees of Freedom 9

Chapter 3 Formation of the Global Analysis Equations 31

3.1 Direct Stiffness Method-The Basic Equations 31

Chapter 4 Stiffness Analysis of Frames-I 56

4.1 Stress-Strain Relationships 56

4.5 The Framework Element Stiffness Matrix 66

Chapter 5 Stiffness Analysis of Frame-II 93

'i l Coordinate Transformations 93

Chapter 6 Virtual Work Principles 137

6.1 Principle of Virtual Displacements-Rigid Bodies 138

6.5.3 Formulation of the Virtual Forces Principle 162

Chapter 7 Virtual Work Principles in Framework Analysis 174

7.1 Description of the Displaced State of Elements 175

Chapter 8 Nonlinear Analysis of Frames-An Introduction 216

8.1 Nonlinear Behavior, Analysis, and Design 216

Chapter 9 Geometric Nonlinear and Elastic Critical Load Analysis 242

9.1 Geometric Stiffness Matrices for Planar Elements 243

Chapter 10 Material Nonlinear Analysis 269

10.1 Nonlinear Material Behavior 269

Chapter 11 Solution of Linear Algebraic Equations 301

11.1 The Basic Choice-Direct Inversion versus Elimination or Iteration 302

11.6.3 Error Estimates and Preconditioning 329

Chapter 12 Solution of Nonlinear Equilibrium Equations 339

12.1 Incremental Analysis 339

Chapter 13 Special Analysis Procedures 375

13.l Condensation 376

13.7 Reanalysis Techniques 408

Appendix A Nonlinear Analysis-A Further Look 419

A.1 Virtual Displacement Principles in Lagrangian Formulations 420

Appendix B On Rigid Body Motion and Natural Deformation 444

B.1 The Nature of the Problem 444

Author Index 451

In matrix structural analysis, many physical quantities and mathematical operations

{F) Vector of element nodal forces

[G] Constraint equation coefficients

IP Polar moment of inertia

[I] Identity matrix

[k] Element stiffness matrix

[L] Lower triangular matrix

[OJ, \OJ Null matrix and vector

PY Axial yield or squash load

p Ratio of axial force to squash load, number of elements

{Q) Vector of element force distribution functions

q Distributed load intensity

IR) Vector of reaction forces, unbalanced load vector

lrl Residual vector

[T] Force transformation matrix

[U] Upper triangular matrix

x,y,z Cartesian coordinates

a, {3, 8 Direction angles

/3 Rate of twist, relaxation factor

[r], [ 'Y] Transformation matrix

'Y Shear strain

{&} Vector of nodal point displacements

2%/? &* =? ? 1/7? /?