Plasticity Theory

PLASTICITY
THEORY
Revised Edition (PDF)
Jacob Lubliner
University of California at Berkeley
Copyright 1990, 2006 by Jacob Lubliner
This book was previously published by Pearson Education, Inc.
Preface
When I first began to plan this book, I thought that I would begin the preface
with the words “The purpose of this little book is...” While I never lost my
belief that small is beautiful, I discovered that it is impossible to put together
a treatment of a field as vast as plasticity theory between the covers of a
truly “little” book and still hope that it will be reasonably comprehensive.
I have long felt that a modern book on the subject — one that would be
useful as a primary reference and, more importantly, as a textbook in a grad-
uate course (such as the one that my colleague Jim Kelly and I have been
teaching) — should incorporate modern treatments of constitutive theory
(including thermodynamics and internal variables), large-deformation plas-
ticity, and dynamic plasticity. By no coincidence, it is precisely these topics
— rather than the traditional study of elastic-plastic boundary-value prob-
lems, slip-line theory and limit analysis — that have been the subject of
my own research in plasticity theory. I also feel that a basic treatment of
plasticity theory should contain at least introductions to the physical foun-
dations of plasticity (and not only that of metals) and to numerical methods
— subjects in which I am not an expert.
I found it quite frustrating that no book in print came even close to
adequately covering all these topics. Out of necessity, I began to prepare
class notes to supplement readings from various available sources. With
the aid of contemporary word-processing technology, the class notes came
to resemble book chapters, prompting some students and colleagues to ask,
“Why don’t you write a book?” It was these queries that gave me the
idea of composing a “little” book that would discuss both the topics that
are omitted from most extant books and, for the sake of completeness, the
conventional topics as well.
Almost two years have passed, and some 1.2 megabytes of disk space have
been filled, resulting in over 400 pages of print. Naively perhaps, I still hope
that the reader approaches this overgrown volume as though it were a little
book: it must not be expected, despite my efforts to make it comprehensive,
to be exhaustive, especially in the sections dealing with applications; I have
preferred to discuss just enough problems to highlight various facets of any
topic. Some oft-treated topics, such as rotating disks, are not touched at
iii
iv Preface
all, nor are such general areas of current interest as micromechanics (except
on the elementary, qualitative level of dislocation theory), damage mechan-
ics (except for a presentation of the general framework of internal-variable
modeling), or fracture mechanics. I had to stop somewhere, didn’t I?
The book is organized in eight chapters, covering major subject areas;
the chapters are divided into sections, and the sections into topical subsec-
tions. Almost every section is followed by a number of exercises. The order
of presentation of the areas is somewhat arbitrary. It is based on the order in
which I have chosen to teach the field, and may easily be criticized by those
partial to a different order. It may seem awkward, for example, that consti-
tutive theory, both elastic and inelastic, is introduced in Chapter 1 (which
is a general introduction to continuum thermomechanics), interrupted for a
survey of the physics of plasticity as given in Chapter 2, and returned to with
specific attention to viscoplasticity and (finally!) rate-independent plasticity
in Chapter 3; this chapter contains the theory of yield criteria, flow rules,
and hardening rules, as well as uniqueness theorems, extremum and varia-
tional principles, and limit-analysis and shakedown theorems. I believe that
the book’s structure and style are sufficiently loose to permit some juggling
of the material; to continue the example, the material of Chapter 2 may be
taken up at some other point, if at all.
The book may also be criticized for devoting too many pages to con-
cepts of physics and constitutive theory that are far more general than the
conventional constitutive models that are actually used in the chapters pre-
senting applications. My defense against such criticisms is this: I believe
that the physics of plasticity and constitutive modeling are in themselves
highly interesting topics on which a great deal of contemporary research is
done, and which deserve to be introduced for their own sake even if their
applicability to the solution of problems (except by means of high-powered
numerical methods) is limited by their complexity.
Another criticism that may, with some justification, be leveled is that
the general formulation of continuum mechanics, valid for large as well as
small deformations and rotations, is presented as a separate topic in Chapter
8, at the end of the book rather than at the beginning. It would indeed
be more elegant to begin with the most general presentation and then to
specialize. The choice I finally made was motivated by two factors. One is
that most of the theory and applications that form the bulk of the book can
be expressed quite adequately within the small-deformation framework. The
other factor is pedagogical: it appears to me, on the basis of long experience,
that most students feel overwhelmed if the new concepts appearing in large-
deformation continuum mechanics were thrown at them too soon.
Much of the material of Chapter 1 — including the mathematical fun-
damentals, in particular tensor algebra and analysis — would normally be
covered in a basic course in continuum mechanics at the advanced under-
Preface v
graduate or first-year graduate level of a North American university. I have

included it in order to make the book more or less self-contained, and while
I might have relegated this material to an appendix (as many authors have
done), I chose to put it at the beginning, if only in order to establish a con-
sistent set of notations at the outset. For more sophisticated students, this
material may serve the purpose of review, and they may well study Section
8.1 along with Sections 1.2 and 1.3, and Section 8.2 along with Sections 1.4
and 1.5.
The core of the book, consisting of Chapters 4, 5, and 6, is devoted to
classical quasi-static problems of rate-independent plasticity theory. Chapter
4 contains a selection of problems in contained plastic deformation (or elastic-
plastic problems) for which analytical solutions have been found: some ele-
mentary problems, and those of torsion, the thick-walled sphere and cylinder,
and bending. The last section, 4.5, is an introduction to numerical methods
(although the underlying concepts of discretization are already introduced
in Chapter 1). For the sake of completeness, numerical methods for both
viscoplastic and (rate-independent) plastic solids are discussed, since nu-
merical schemes based on viscoplasticity have been found effective in solving
elastic-plastic problems. Those who are already familiar with the material
of Sections 8.1 and 8.2 may study Section 8.3, which deals with numerical
methods in large-deformation plasticity, immediately following Section 4.5.
Chapters 5 and 6 deal with problems in plastic flow and collapse. Chap-
ter 5 contains some theory and some “exact” solutions: Section 5.1 covers
the general theory of plane plastic flow and some of its applications, and
Section 5.2 the general theory of plates and the collapse of axisymmetrically
loaded circular plates. Section 5.3 deals with plastic buckling; its placement
in this chapter may well be considered arbitrary, but it seems appropriate,
since buckling may be regarded as another form of collapse. Chapter 6 con-
tains applications of limit analysis to plane problems (including those of soil
mechanics), beams and framed structures, and plates and shells.
Chapter 7 is an introduction to dynamic plasticity. It deals both with
problems in the dynamic loading of elastic–perfectly plastic structures treated
by an extension of limit analysis, and with wave-propagation problems, one-
dimensional (with the significance of rate dependence explicitly discussed)
and three-dimensional. The content of Chapter 8 has already been men-
tioned.
As the knowledgeable reader may see from the foregoing survey, a coher-
ent course may be built in various ways by putting together selected portions
of the book. Any recommendation on my part would only betray my own
prejudices, and therefore I will refrain from making one. My hope is that
those whose orientation and interests are different from mine will nonetheless
find this would-be “little book” useful.
In shaping the book I was greatly helped by comments from some out-
vi Preface
standing mechanicians who took the trouble to read the book in draft form,
and to whom I owe a debt of thanks: Lallit Anand (M. I. T.), Satya Atluri
(Georgia Tech), Maciej Bieniek (Columbia), Michael Ortiz (Brown), and
Gerald Wempner (Georgia Tech).
An immeasurable amount of help, as well as most of the inspiration to
write the book, came from my students, current and past. There are too
many to cite by name — may they forgive me — but I cannot leave out Vas-
silis Panoskaltsis, who was especially helpful in the writing of the sections
on numerical methods (including some sample computations) and who sug-
gested useful improvements throughout the book, even the correct spelling
of the classical Greek verb from which the word “plasticity” is derived.
Finally, I wish to acknowledge Barbara Zeiders, whose thoroughly pro-
fessional copy editing helped unify the book’s style, and Rachel Lerner
and Harry Sices, whose meticulous proofreading found some needles in the
haystack that might have stung the unwary. Needless to say, the ultimate
responsibility for any remaining lapses is no one’s but mine.
A note on cross-referencing: any reference to a number such as 3.2.1,
without parentheses, is to a subsection; with parentheses, such as (4.3.4), it
is to an equation.
Addendum: Revised Edition

Despite the proofreaders’ efforts and mine, the printed edition remained
plagued with numerous errors. In the fifteen years that have passed I have
managed to find lots of them, perhaps most if not all. I have also found it
necessary to redo all the figures. The result is this revised edition.
Contents
Chapter 1: Introduction to Continuum Thermome-

chanics
Section 1.1 Mathematical Fundamentals 1

1.1.1 Notation 1
1.1.2 Cartesian Tensors 3
1.1.3 Vector and Tensor Calculus 6
1.1.4 Curvilinear Coordinates 9
Section 1.2 Continuum Deformation 14

1.2.1 Displacement 14
1.2.2 Strain 15
1.2.3 Principal Strains 21
1.2.4 Compatibility Conditions 23
Section 1.3 Mechanics of Continuous Bodies 26

1.3.1 Introduction 26
1.3.2 Stress 29
1.3.3 Mohr’s Circle 33
1.3.4 Plane Stress 35
1.3.5 Boundary-Value Problems 36
Section 1.4 Constitutive Relations: Elastic 44

1.4.1 Energy and Thermoelasticity 44
1.4.2 Linear Elasticity 49
1.4.3 Energy Principles 54
vii
viii Contents
Section 1.5 Constitutive Relations: Inelastic 59

1.5.1 Inelasticity 59
1.5.2 Linear Viscoelasticity 61
1.5.3 Internal Variables: General Theory 65
1.5.4 Flow Law and Flow Potential 69
Chapter 2: The Physics of Plasticity

Section 2.1 Phenomenology of Plastic Deformation 75
2.1.1 Experimental Stress-Strain Relations 76
2.1.2 Plastic Deformation 80
2.1.3 Temperature and Rate Dependence 85
Section 2.2 Crystal Plasticity 89

2.2.1 Crystals and Slip 89
2.2.2 Dislocations and Crystal Plasticity 94
2.2.3 Dislocation Models of Plastic Phenomena 100
Section 2.3 Plasticity of Soils, Rocks and Concrete 103

2.3.1 Plasticity of Soil 104
2.3.2 “Plasticity” of Rock and Concrete 108
Chapter 3: Constitutive Theory

Section 3.1 Viscoplasticity 111
3.1.1 Internal-Variable Theory of Viscoplasticity 111
3.1.2 Transition to Rate-Independent Plasticity 116
3.1.3 Viscoplasticity Without a Yield Surface 118
Section 3.2 Rate-Independent Plasticity 122

3.2.1 Flow Rule and Work-Hardening 122
3.2.2 Maximum-Dissipation Postulate and Normality 127
3.2.3 Strain-Space Plasticity 130
Section 3.3 Yield Criteria, Flow Rules and Hardening Rules 135
Contents ix
3.3.2 Yield Criteria Independent of the Mean Stress 137

3.3.3 Yield Criteria Dependent on the Mean Stress 141
3.3.4 Yield Criteria Under Special States of Stress or Deformation 144
3.3.5 Hardening Rules 146
Section 3.4 Uniqueness and Extremum Theorems 152

3.4.1 Uniqueness Theorems 152
3.4.2 Extremum and Variational Principles 154
3.4.3 Rigid–Plastic Materials 159
Section 3.5 Limit-Analysis and Shakedown Theorems 162

3.5.1 Standard Limit-Analysis Theorems 162
3.5.2 Nonstandard Limit-Analysis Theorems 168
3.5.3 Shakedown Theorems 170
Chapter 4: Problems in Contained Plastic Defor-

mation
Section 4.1 Elementary Problems 177
4.1.1 Introduction: Statically Determinate Problems 177
4.1.2 Thin-Walled Circular Tube in Torsion and Extension 178
4.1.3 Thin-Walled Cylinder Under Pressure and Axial Force 181
4.1.4 Statically Indeterminate Problems 184
Section 4.2 Elastic–Plastic Torsion 189

4.2.1 The Torsion Problem 189
4.2.2 Elastic Torsion 191
4.2.3 Plastic Torsion 194
Section 4.3 The Thick-Walled Hollow Sphere and Cylinder 205

4.3.1 Elastic Hollow Sphere Under Internal and External Pressure 206
4.3.2 Elastic–Plastic Hollow Sphere Under Internal Pressure 208
4.3.3 Thermal Stresses in an Elastic–Plastic Hollow Sphere 213
4.3.4 Hollow Cylinder: Elastic Solution and Initial Yield Pressure 216
4.3.5 Elastic–Plastic Hollow Cylinder 220
x Contents
Section 4.4 Elastic–Plastic Bending 229

4.4.1 Pure Bending of Prismatic Beams 229
4.4.2 Rectangular Beams Under Transverse Loads 239
4.4.3 Plane-Strain Pure Bending of Wide Beams or Plates 245
Section 4.5 Numerical Methods 250

4.5.1 Integration of Rate Equations 251
4.5.2 The Finite-Element Method 256
4.5.3 Finite-Element Methods for Nonlinear Continua 262
Chapter 5: Problems in Plastic Flow and Collapse

I: Theories and “Exact” Solutions
Introduction 275
Section 5.1 Plane Problems 276

5.1.1 Slip-Line Theory 277
5.1.2 Simple Slip-Line Fields 287
5.1.3 Metal-Forming Problems 291
Section 5.2 Collapse of Circular Plates 298

5.2.1 Introduction to Plate Theory 299
5.2.2 Elastic Plates 303
5.2.3 Yielding of Plates 308
Section 5.3 Plastic Buckling 313

5.3.1 Introduction to Stability Theory 314
5.3.2 Theories of the Effective Modulus 319
5.3.3 Plastic Buckling of Plates and Shells 326
Chapter 6: Problems in Plastic Flow and Collapse

II: Applications of Limit Analysis
Introduction 337
Contents xi
Section 6.1 Limit Analysis of Plane Problems 338

6.1.1 Blocks and Slabs with Grooves or Cutouts 338
6.1.2 Problems in Bending 341
6.1.3 Problems in Soil Mechanics 347
Section 6.2 Beams Under Combined Stresses 355

6.2.1 Generalized Stress 355
6.2.2 Extension and Bending 358
6.2.3 Combined Extension, Bending and Torsion 364
6.2.4 Bending and Shear 369
Section 6.3 Limit Analysis of Trusses, Beams and Frames 374

6.3.1 Trusses 374
6.3.2 Beams 380
6.3.3 Limit Analysis of Frames 385
6.3.4 Limit Design of Frames 390
Section 6.4 Limit Analysis of Plates and Shells 398

6.4.1 Limit Analysis of Plates 398
6.4.2 Limit Analysis of Shells: Theory 404
6.4.3 Limit Analysis of Shells: Examples 407
Chapter 7: Dynamic Problems

Section 7.1 Dynamic Loading of Structures 417
7.1.2 Dynamic Loading of Beams 419
7.1.3 Dynamic Loading of Plates and Shells 425
Section 7.2 One-Dimensional Plastic Waves 434

7.2.1 Theory of One-Dimensional Waves 434
7.2.2 Waves in Elastic–Plastic bars 438
7.2.3 Rate Dependence 446
7.2.4 Application of the Method of Characteristics 448
xii Contents
Section 7.3 Three-Dimensional Waves 452

7.3.1 Theory of Acceleration Waves 453
7.3.2 Special Cases 459
Chapter 8: Large-Deformation Plasticity

Section 8.1 Large-Deformation Continuum Mechanics 465
8.1.1 Continuum Deformation 465
8.1.2 Continuum Mechanics and Objectivity 473
Section 8.2 Large-Deformation Constitutive Theory 478

8.2.1 Thermoelasticity 478
8.2.2 Inelasticity: Kinematics 480
8.2.3 Inelasticity: Thermomechanics 485
8.2.4 Yield Condition and Flow Rule 487
Section 8.3 Numerical Methods in Large-Deformation Plastic-

ity 491
8.3.1 Rate-Based Formulations 492
8.3.2 “Hyperelastic” Numerical Methods 496
References 501
Index 517
Chapter 1
Introduction to Continuum
Thermomechanics
Section 1.1 Mathematical Fundamentals
1.1.1. Notation
Solid mechanics, which includes the theories of elasticity and plasticity, is

a broad discipline, with experimental, theoretical, and computational as-
pects, and with a twofold aim: on the one hand, it seeks to describe the
mechanical behavior of solids under conditions as general as possible, re-
gardless of shape, interaction with other bodies, field of application, or the
like; on the other hand, it attempts to provide solutions to specific problems
involving stressed solid bodies that arise in civil and mechanical engineering,
geophysics, physiology, and other applied disciplines. These aims are not in
conflict, but complementary: some important results in the general theory
have been obtained in the course of solving specific problems, and practical
solution methods have resulted from fundamental theoretical work. There
are, however, differences in approach between workers who focus on one or
the other of the two goals, and one of the most readily apparent differences
is in the notation used.
Most of the physical concepts used in solid mechanics are modeled by
mathematical entities known as tensors. Tensors have representations
through components with respect to specific frames or coordinate systems
(a vector is a kind of tensor), but a great deal can be said about them with-
out reference to any particular frame. Workers who are chiefly interested in
the solution of specific problems — including, notably, engineers — generally
use a system of notation in which the various components of tensors appear
explicitly. This system, which will here be called “engineering” notation,
has as one of its advantages familiarity, since it is the one that is gener-
1
2 Chapter 1 / Introduction to Continuum Thermomechanics
ally used in undergraduate “strength of materials” courses, but it is often

cumbersome, requiring several lines of long equations where other notations
permit one short line, and it sometimes obscures the mathematical nature of
the objects and processes involved. Workers in constitutive theory tend to
use either one of several systems of “direct” notation that in general use no
indices (subscripts and superscripts), such as Gibbs’ dyadic notation, matrix
notation, and a combination of the two, or the so-called indicial notation in
which the use of indices is basic. The indices are used to label components
of tensors, but with respect to an arbitrary rather than a specific frame.
Indicial notation is the principal system used in this book, although
other systems are used occasionally as seems appropriate. In particular,
“engineering” notation is used when the solutions to certain specific problems
are discussed, and the matrix-based direct notation is used in connection
with the study of large deformation, in which matrix multiplication plays an
important part.
Assuming the reader to be familiar with vectors as commonly taught in
undergraduate engineering schools, we introduce indicial notation as follows:
for Cartesian coordinates (x, y, z) we write (x1 , x2 , x3 ); for unit vectors
(i, j, k) we write (e1 , e2 , e3 ); for the components (ux , uy , uz ) of a vector u
we write (u1 , u2 , u3 ).
X
The summation convention is defined as follows: the symbol may
i
be omitted (i.e., it is implied) if the summation (dummy) index (say i)
appears exactly twice in each term of a sum. Example:
ai bi = a1 b1 + a2 b2 + a3 b3 .
The Kronecker delta is defined as

( )
1 if i = j
δij = = δji .
0 if i 6= j
The Levi-Civita “e” tensor or permutation tensor is defined as



 1 if ijk = 123, 231, 312
eijk = −1 if ijk = 321, 213, 132

 0 otherwise.
There is a relation between the “e” tensor and the Kronecker delta known
as the e-delta identity:
eijk elmk = δil δjm − δim δjl .
The fundamental operations of three-dimensional vector algebra, pre-

sented in indicial and, where appropriate, in direct notation, are as follows.
Section 1.1 / Mathematical Preliminaries 3
Decomposition: u = ui ei . (The position vector in x1 x2 x3 -space is denoted

x = xi ei .)
Scalar (dot) product between unit vectors: ei · ej = δij (orthonormality).
Projection: ei · u = ei · ej uj = δij uj = ui .
Scalar (dot) product between any two vectors: u · v = ui ei · ej vj = ui vi .
Vector (cross) product between unit vectors: ei × ej = eijk ek .
Vector (cross) product between any two vectors: u × v = ei eijk uj vk .
Scalar triple product: u · (v × w) = (u × v) · w = eijk ui vj wk .
Note that the parentheses in the direct notation for the scalar triple product
can be omitted without ambiguity, since a product of the form (u · v) × w
has no meaning.
The notation for matrices is as follows. A matrix with entries αij , where
i is the row index and j is the column index, is denoted [αij ] or α. The
transpose of α is the matrix [αji ], also denoted αT . The determinant of α is
denoted det α, and the inverse of α is α−1 , so that α α−1 = α−1 α = I, where
I = [δij ] is the unit matrix or identity matrix .
1.1.2. Cartesian Tensors

Coordinate Transformation
Since our aim is to be able to make statements about physical behavior
independently of any choice of coordinate axes, let us see what the relation
is between two sets of axes. Limiting ourselves to Cartesian coordinate
systems, let us consider a set of axes (xi ), with the corresponding set of unit
vectors (ei ) (also known as the basis of the coordinate system), and another
set (x∗i ) with the basis (e∗i ). If βij is the cosine of the angle between the
x∗i -axis and the xj -axis, then
e∗i · ej = βij .
According to this equation, βij is both the xj -component of e∗i and the
x∗i -component of ej , so that
e∗i = βij ej
and
ei = βji e∗j .
For any vector u = ui ei = u∗i e∗i ,
u∗i = βik uk , ui = βji u∗j .

If the free index i in the second equation is replaced by k and its right-hand
side is substituted for uk in the first equation, then
u∗i = βik βjk u∗j .
Similarly,
ui = βki βkj uj .
Since u∗i = δij u∗j and ui = δij uj , and since the vector u is arbitrary, it follows
that
βik βjk = βki βkj = δij ,
that is, the matrix β = [βij ] is orthogonal . In matrix notation, β β T = β T β =
I. The determinant of a matrix equals the determinant of its transpose,
that is, det α = det αT , and the determinant of a product of matrices equals
the product of the determinants, so that det(αβ) = det α det β. For an
orthogonal matrix β, therefore, (det β)2 = det I = 1, or det β = ±1. If the
basis (e∗i ) is obtained from (ei ) by a pure rotation, then β is called proper
orthogonal , and det β = 1.
An example of a proper orthogonal matrix is the matrix describing coun-
terclockwise rotation by an angle θ about the x3 -axis, as shown in Figure
1.1.1.
x2
x∗2 6
KA
A
A
∗
* x1
A θ
A

  A
cos θ sin θ 0 A
θ
@ K
β =  − sin θ cos θ 0  @A
 
@ - x1
A
0 0 1
A@
@
@
@
R x3 , x∗3
@
@
Figure 1.1.1. Example of a rotation represented by a proper orthogonal matrix.
Linear Operators
An operator λ on the space of three-dimensional vectors is simply a
vector-valued function of a vector variable: given a vector u, λ(u) uniquely
defines another vector. λ is a linear operator if λ(au + bv) = aλ(u) + bλ(v),
where a and b are any real numbers, and u and v are any vectors.
The preceding definitions are independent of any decomposition of the

vectors involved. If the vectors are to be represented with respect to a basis
(ei ), then the linear operator must also be so represented. The Cartesian
components of a linear operator are defined as follows: If v = λ(u), then it
follows from the definition of linearity that
vi = ei · λ(ej uj ) = ei · λ(ej )uj = λij uj ,
def
where λij uj = ei · λ(ej ). Thus λ = [λij ] is the component matrix of λ with
respect to the basis (ei ). In matrix notation we may write v = λ u, where
u (v) is the column matrix whose entries are the components ui (vi ) of the
vector u (v) with respect to the basis (ei ). In direct tensor notations it
is also customary to omit parentheses: in Gibbs’ notation we would write
v = λ · u, and in the matrix-based direct notation, v = λu.
In a different basis (e∗i ), where e∗i = βij ej , the component matrix of λ is
defined by
λ∗ij = βik βjl λkl ,
or, in matrix notation,
λ∗ = β λ β T .
If λ∗ = λ, then λ is an isotropic or spherical operator. An example is the
identity operator I, whose component matrix is I = [δij ]. The most general
isotropic operator is cI, where c is any scalar.
If the component matrix of a linear operator has a property which is
not changed by transformation to a different basis, that is, if the property
is shared by λ and λ∗ (for any β), then the property is called invariant. An
invariant property may be said to be a property of the linear operator λ
itself rather than of its component matrix in a particular basis. An example
is transposition: if λ∗ = β λ β T , then λ∗T = β λT β T . Consequently we may
speak of the transpose λT of the linear operator λ, and we may define its
symmetric and antisymmetric parts:
λS = 12 (λ + λT ) ⇔ λSij = λ(ij) = 12 (λij + λji ),
λA = 12 (λ − λT ) ⇔ λA
ij = λ[ij] = 2 (λij − λji ).
1
If λA = 0 (i.e., λij = λji ), then λ is a symmetric operator. If λS = 0 (i.e.,

λij = −λji ), then λ is an antisymmetric operator.
Tensors
A linear operator, as just defined, is also called a tensor . More generally,
a tensor of rank n is a quantity T represented in a basis (ei ) by a component
array Ti1 ...in (i1 , ..., in = 1, 2, 3) and in another basis (e∗i ) by the component
array Ti∗1 ...in , where
Ti∗1 ...in = βi1 k1 ...βin kn Tk1 ...kn .
Thus a scalar quantity is a tensor of rank 0, a vector is a tensor of rank 1,

and a linear operator is a tensor of rank 2. “Tensor” with rank unspecified
is often used to mean a tensor or rank 2. The tensor whose component array
is eijk is an isotropic tensor of rank 3. Tensors of rank 4 are found first in
Section 1.4.
An array whose elements are products of tensor components of rank
m and n, respectively, represents a tensor of rank m + n. An important
example is furnished by the tensor product of two vectors u and v (a dyad in
the terminology of Gibbs), the tensor of rank 2 represented by u v T = [ui vj ]
and denoted u ⊗ v, or more simply uv in the Gibbs notation. Thus an
arbitrary tensor λ of rank two, whose components with respect to a basis
(ei ) are λij , satisfies the equation
λ = λij ei ⊗ ej .
Clearly, (u⊗v)w = u(v·w); in the Gibbs notation both sides of this equation
may be written as uv · w.
An operation known as a contraction may be performed on a tensor of
rank n ≥ 2. It consists of setting any two indices in its component array
equal to each other (with summation implied). The resulting array, indexed
by the remaining indices, if any (the “free indices”), represents a tensor of
rank n − 2. For a tensor λ of rank 2, λii = tr λ is a scalar known as the trace
of λ. A standard example is u · v = ui vi = tr (u ⊗ v). Note that if n > 2
then more than one contraction of the same tensor is possible, resulting
in different contracted tensors; and, if n ≥ 4, then we can have multiple
contractions. For example, if n = 4 then we can have a double contraction
resulting in a scalar, and three different scalars are possible: Tiijj , Tijij , and
Tijji .
If u and v are vectors that are related by the equation
ui = αij vj ,
then α necessarily represents a tensor α of rank 2. Similarly, if α and β are

tensors of rank 2 related by
αij = ρijkl βkl ,
then the array ρijkl represents a tensor ρ of rank four. The generalization
of these results is known as the quotient rule.
1.1.3. Vector and Tensor Calculus

A tensor field of rank n is a function (usually assumed continuously differ-
entiable) whose values are tensors of rank n and whose domain is a region
R in x1 x2 x3 space. The boundary of R is a closed surface denoted ∂R, and
the unit outward normal vector on ∂R will be denoted n. The partial deriv-
ative operator ∂/∂xi will be written more simply as ∂i . A very common
alternative notation, which is used extensively here, is ∂i φ = φ,i .
If φ is a tensor field of rank n, then the array of the partial derivatives
of its components represents a tensor field of rank n + 1.
The del operator is defined as ∇ = ei ∂i , and the Laplacian operator as
2
∇ = ∂i ∂i = i ∂ 2 /∂x2i . For a scalar field φ, the gradient of φ is the vector
P
field defined by
∇φ = gradφ = ei φ,i .
For a vector field v, we use ∇v to denote (∇ ⊗ v)T , that is,
∇v = vi ,j ei ⊗ ej ,
but this notation is not universal: many writers would call this (∇v)T . There
is no ambiguity, however, when only the symmetric part of ∇v is used, or
when the divergence of v is defined as the trace of ∇v:
div v = ∇ · v = vi ,i .
Similarly, the curl of v is defined unambiguously as
curl v = ∇ × v = ei eijk vk ,j .
For a tensor field φ of rank 2, we define ∇φ as represented by φjk ,i , and
∇ · φ = div φ = φjk ,j ek .
These definitions are, again, not universal.

The three-dimensional equivalent of the fundamental theorem of calculus
is Gauss’s theorem:
Z Z
φ,i dV = ni φ dS, (1.1.1)
R ∂R
where φ is any differentiable field. The particular case where φ is replaced

(on both sides of the equation, of course) by vi (the ith component of a
vector field v), with summation under the integral signs implied, is known
as the divergence theorem. This is the case we use most often.
The two-dimensional Gauss’s theorem refers to fields defined in an area
A in the x1 x2 -plane, bounded by a closed curve C on which an infinitesimal
element of arc length is ds (positive when it is oriented counterclockwise).
It is conventional to use Greek letters for indices whose range is 1, 2; thus
the theorem reads Z I
φ,α dA = nα φ ds. (1.1.2)
A C
x2
J n 3 n2

dx2 Jn1
dsJ
J
−dx1
x1
Figure 1.1.2. Normal vector to a plane curve
Now suppose that the curve C is described parametrically by x1 = x1 (s),

x2 = x2 (s). Then, as can be seen from Figure 1.1.2,
dx2 dx1
n1 = , n2 = − .
ds ds
Thus, for any two functions uα (x1 , x2 ) (α = 1, 2),

I I Z
u1 dx1 = − n2 u1 ds = − u1 ,2 dA
C C A
and I I Z
u2 dx2 = n1 u2 ds = u2 ,1 dA.
C C A
Combining these two equations, we obtain Green’s lemma:
I Z
uα dxα = (u2 ,1 −u1 ,2 ) dA. (1.1.3)
C A
If there exists a continuously

H
differentiable
H
function φ(x1 , x2 ), defined in A,
such that uα = φ,α , then uα dxα = dφ = 0 around any closed contour, so
that u2 ,1 = u1 ,2 . The converse is also true (i.e., the last equality implies the
existence of φ), provided that the area A is simply connected (i.e., contains
no holes); otherwise additional conditions are required.
The preceding result is known as the two-dimensional integrability
theorem and will be used repeatedly.
There exists an extension of Green’s lemma to a curved surface S in
x1 x2 x3 space, bounded by a (not necessarily plane) closed curve C para-
metrized by xi = xi (s), i = 1, 2, 3. This extension (derived from Gauss’s
theorem) is known as Stokes’ theorem and takes the form

I Z
ui dxi = ni eijk uk ,j dS. (1.1.4)
C S
Clearly, Green’s lemma represents the special case of Stokes’ theorem when
n = e3 . From Stokes’ theorem follows the three-dimensional integrability
theorem: a field φ(x) such that u = ∇φ in a region R exists only if ∇×u = 0,
or, equivalently, uj ,i = ui ,j . As in the two-dimensional version, this last
condition is also sufficient if R is simply connected.
1.1.4. Curvilinear Coordinates

The study of tensor fields in curvilinear coordinates is intimately tied to
differential geometry, and in many books and courses of study dealing with
continuum mechanics it is undertaken at the outset. The traditional method-
ology is as follows: with a set of curvilinear coordinates ξ i (i = 1, 2, 3)
such that the position of a point in three-dimensional space is defined by
x(ξ 1 , ξ 2 , ξ 3 ), the natural basis is defined as the ordered triple of vectors
gi = ∂x/∂ξ i , so that dx = gi dξ i ; the summation convention here applies
whenever the pair of repeated indices consists of one subscript and one su-
perscript. The basis vectors are not, in general, unit vectors, nor are they
necessarily mutually perpendicular, although it is usual for them to have
the latter property. One can find, however, the dual basis (gi ) such that
gi · gj = δij . A vector v may be represented as vi gi or as v i gi , where
vi = v · gi and v i = v · gi are respectively the covariant and contravariant
components of v. For tensors of higher rank one can similarly define covari-
ant, contravariant, and several kinds of mixed components. The gradient of
a tensor field is defined in terms of the so-called covariant derivatives of its
components, which, except in the case of a scalar field, differ from the partial
derivatives with respect to the ξ i because the basis vectors themselves vary.
A central role is played by the metric tensor with components gij = gi · gj ,
having the property that dx · dx = gij dξ i dξ j .
An alternative approach is based on the theory of differentiable manifolds
(see, e.g., Marsden and Hughes [1983]).
Curvilinear tensor analysis is especially useful for studying the mechanics
of curved surfaces, such as shells; when this topic does not play an impor-
tant part, a simpler approach is available, based on the so-called “physical”
components of the tensors involved. In this approach mutually perpendic-
ular unit vectors (forming an orthonormal basis) are used, rather than the
natural and dual bases. We conclude this section by examining cylindrical
and spherical coordinates in the light of this methodology.
Cylindrical Coordinates
q
In the cylindrical coordinates (r, θ, z), where r = x21 + x22 , θ = tan−1 (x2 /x1 ),
and z = x3 , the unit vectors are
er = e1 cos θ + e2 sin θ, eθ = −e1 sin θ + e2 cos θ, ez = e3 ,
so that
d d
er = eθ , eθ = −er .
dθ dθ
Using the chain rule for partial derivatives, we may show the ∇ operator to
be given by
∂ 1 ∂ ∂
∇ = er + eθ + ez . (1.1.5)
∂r r ∂θ ∂z
When this operator is applied to a vector field v, represented as
v = vr er + vθ eθ + vz ez ,
the result may be written as

∂ 1 ∂ ∂

∇⊗v = + eθ
er + ez ⊗ (vr er + vθ eθ + vz ez )
∂r r ∂θ ∂z
∂vr ∂vθ ∂vz

= er ⊗ er + eθ + ez
∂r ∂r ∂r
1

∂vr ∂vθ ∂vz
(1.1.6)
+eθ ⊗ er + eθ vr + eθ − er vθ + ez
r ∂θ ∂θ ∂θ
∂vr ∂vθ ∂vz

+ez ⊗ er + eθ + ez .
∂z ∂z ∂z
The trace of this second-rank tensor is the divergence of v:
∂vr 1 ∂vθ ∂vz

∇·v = + vr + + ;
∂r r ∂θ ∂z
and when v = ∇u, the gradient of a scalar field, then this is
!
2 ∂2 1 ∂ 1 ∂2 ∂2
∇ u= + + + u.
∂r2 r ∂r r2 ∂θ2 ∂z 2
Lastly, let us consider a symmetric second-rank tensor field
λ = er ⊗ er λrr + (er ⊗ eθ + eθ ⊗ er )λrθ + eθ ⊗ eθ λθθ

+ (er ⊗ ez + ez ⊗ er )λrz + (eθ ⊗ ez + ez ⊗ eθ )λθz + ez ⊗ ez λzz ;
its divergence — which we have occasion to use in Section 1.3 — is

∂λrr λrr − λθθ 1 ∂λrθ ∂λrz

∇ · λ = er + + +
∂r r r ∂θ ∂z
∂λrθ 2λrθ 1 ∂λθθ ∂λθz

+ eθ + + + (1.1.7)
∂r r r ∂θ ∂z
∂λrz λrz 1 ∂λθz ∂λzz

+ ez + + + .
∂r r r ∂θ ∂z
Spherical Coordinates
q
The spherical coordinates (r, θ, φ) are defined by r = x21 + x22 + x23 ,
q
θ = tan−1 (x2 /x1 ), and φ = cot−1 x3 / x21 + x22 . The unit vectors are
er = (e1 cos θ+e2 sin θ) sin φ+e3 cos φ, eφ = (e1 cos θ+e2 sin θ) cos φ−e3 sin φ,
eθ = −e1 sin θ + e2 cos θ,

so that
∂ ∂ ∂
er = eφ , eφ = −er , eθ = 0,
∂φ ∂φ ∂φ
∂ ∂ ∂
er = eθ sin φ, eφ = eθ cos φ, eθ = −er sin φ − eφ cos φ.
∂θ ∂θ ∂θ
The ∇ operator is given by
∂ 1 ∂ 1 ∂
∇ = er + eφ + eθ .
∂r r ∂φ r sin φ ∂θ
For a vector field v we accordingly have
∂vr ∂vφ ∂vθ

∇ ⊗ v = er ⊗ er + eφ + eθ
∂r ∂r ∂r
1 ∂vr vφ 1 ∂vφ vr 1 ∂vθ

+ eφ ⊗ er ∂φ − + eφ + + eθ
r r r ∂φ r r ∂φ
1 ∂vr vθ 1 ∂vφ cot φ

+ eθ ⊗ er − + eφ − vθ
r sin φ ∂θ r r sin φ ∂θ r
1 ∂vθ vr cot φ

+ eθ + + vφ ,
r sin φ ∂θ r r
(1.1.8)
so that
∂vr vr 1 ∂vφ cot φ 1 ∂vθ
∇·v = +2 + + vφ + ,
∂r r r ∂φ r r sin φ ∂θ
∂ 2 u 2 ∂u 1 ∂ 2 u cot φ ∂u 1 ∂2u
∇2 u = + + + + ,
∂r2 r ∂r r2 ∂φ2 r2 ∂φ r2 sin2 φ ∂θ2
and, for a symmetric second-rank tensor field λ,
∂λrr 1 ∂λrφ 1 ∂λrθ 2λrr − λφφ − λθθ + λrφ cot φ

∇ · λ = er + + +
∂r r ∂φ r sin φ ∂θ r
∂λrφ 1 ∂λφφ 1 ∂λφθ

+ eφ + +
∂r r ∂φ r sin φ ∂θ
λφφ cot φ − λθθ cot φ + 3λrφ

+
r
∂λrθ 1 ∂λφθ 1 ∂λθθ 3λrθ + 2λφθ cot φ

+ eθ + + + .
(1.1.9)
Exercises: Section 1.1

1. Show that
(a) δii = 3
(b) δij δij = 3
(c) eijk ejki = 6
(d) eijk Aj Ak = 0
(e) δij δjk = δik
(f) δij eijk = 0
2. Using indicial notation and the summation convention, prove that
(s × t) · (u × v) = (s · u)(t · v) − (s · v)(t · u).
3. For the matrix  

1 1 0
[aij ] =  1 2 2  ,
 
0 2 3
calculate the values of
(a) aii ,
(b) aij aij ,
(c) aij ajk when i = 1, k = 1 and when i = 1, k = 2.
4. Show that the matrix

12 9 4
 
−

 25 25 5 

 3 4 
β= − 0 

 5 5 

 16 12 3 
25 25 5
is proper orthogonal, that is, β β T = β T β = I, and det β = 1.
5. Find the rotation matrix β describing the transformation composed

of, first, a 90◦ rotation about the x1 -axis, and second, a 45◦ rotation
about the rotated x3 -axis.
6. Two Cartesian bases, (ei ),√and (e∗i ) are given, with e∗1 = (2e1 + 2e2 +
e3 )/3 and e∗2 = (e1 − e2 )/ 2.
(a) Express e∗3 in terms of the ei .

(b) Express the ei in terms of the e∗i .
(c) If v = 6e1 − 6e2 + 12e3 , find the vi∗ .
7. The following table shows the angles between the original axes xi and
the transformed axes x∗i .
x1 x2 x3
x∗1 135◦ 60◦ 120◦
x∗2 90◦ 45◦ 45◦
x∗3 45◦ 60◦ 120◦
(a) Find the transformation matrix β, and verify that it describes a

rotation.
(b) If a second-rank tensor λ has the following component matrix
with the respect to the original axes,
 
3 −4 2
λ =  −4 0 1 ,
 
2 1 3
find its component matrix λ∗ with respect to the rotated axes.
8. (a) Use the chain rule of calculus to prove that if φ is a scalar field,
then ∇φ is a vector field.
(b) Use the quotient rule to prove the same result.
9. Using indicial notation, prove that (a) ∇×∇φ = 0 and (b) ∇·∇×v = 0.
10. If x = xi ei and r = |x|, prove that
∇2 (rn ) = n(n + 1)rn−2 .
11. If φ(x1 , x2 , x3 ) = aij xi xj , with aij constant, show that φ,i = (aij +
aji )xj and φ,ij = aij + aji .
12. Show that ∇2 (φψ) = φ∇2 ψ + 2(∇φ) · (∇ψ) + ψ∇2 φ.
13. Use Gauss’s theorem to prove that,

R
if V is the volume of a three-
dimensional region R, then V = 31 ∂R xi ni dS.
14. Verify Green’s lemma for the area A bounded by the square with cor-
ners at (0, 0), (a, 0), (a, a), (0, a), of u1 (x1 , x2 ) = 0 and u2 (x1 , x2 ) =
bx1 , where b is a constant.
15. Find the natural basis (gi ) and the dual basis (gi ) (a) for cylindrical
coordinates, with ξ 1 = r, ξ 2 = θ, and ξ 3 = z, and (b) for spherical
coordinates, with ξ 1 = r, ξ 2 = φ, and ξ 3 = θ.
16. Starting with the expression in Cartesian coordinates for the gradi-
ent operator ∇ and using the chain rule for partial derivatives, derive
Equation (1.1.5).
Section 1.2 Continuum Deformation

1.2.1. Displacement
The first application of the mathematical concepts introduced in Section
1.1 will now be to the description of the deformation of bodies that can be
modeled as continua. A body is said to be modeled as a continuum if to any
configuration of the body there corresponds a region R in three-dimensional
space such that every point of the region is occupied by a particle (material
point) of the body.
Any one configuration may be taken as the reference configuration. Con-
sider a particle that in this configuration occupies the point defined by the
vector r = xi ei . When the body is displaced, the same particle will occupy
the point r∗ = x∗i ei . (Note that here the x∗i no longer mean the coordinates
of the same point with respect to a rotated basis, as in the Section 1.1, but
the coordinates of a different point with respect to the same basis.) The dif-
ference r∗ − r is called the displacement of the particle and will be denoted
u. The reference position vector r will be used to label the given particle;
the coordinates xi are then called Lagrangian coordinates. Consequently the
displacement may be given as a function of r, u(r), and it forms a vector field
defined in the region occupied by the body in the reference configuration.
Now consider a neighboring particle labeled by r + ∆r. In the displaced
configuration, the position of this point will be
r∗ + ∆r∗ = r + ∆r + u(r + ∆r)
(see Figure 1.2.1), so that
∆r∗ = ∆r + u(r + ∆r) − u(r),
or, in indicial notation,
∆x∗i = ∆xi + ui (r + ∆r) − ui (r).
.
But if ∆r is sufficiently small, then ui (r + ∆r) − ui (r) = ui ,j (r)∆xj , the
error in the approximation being such that it tends to zero faster than |∆r|.
It is conventional to replace ∆r by the infinitesimal dr, and to write the
approximation as an equality. Defining the displacement-gradient matrix α
def
by αij = ui ,j , we may write in matrix notation
dx∗ = (I + α)dx.
Section 1.2 / Continuum Deformation 15
x2
BM u(r + ∆r)

B
B ∆r j
CO
BP ∗
7 PPP C ∆r

r + ∆r PPqC
AK
>
A
r
A
A u(r)
6 r∗

r∗ + ∆r∗

x1

x3
Figure 1.2.1. Displacement
1.2.2. Strain
A body is said to undergo a rigid-body displacement if the distances between

all particles remain unchanged; otherwise the body is said to be deformed .
Let us limit ourselves, for the moment, to an infinitesimal neighborhood
of the particle labeled by r; the deformation of the neighborhood may be
measured by the extent to which the lengths of the infinitesimal vectors dr
emanating from r change in the course of the displacement. The square of
the length of dr∗ is
|dr∗ |2 = dr∗ · dr∗ = dx∗T dx∗ = dxT (I + αT )(I + α) x = dxT (I + 2E) x,
where E = 12 (αT + α + αT α), or, in indicial notation,
1
Eij = (uj ,i +ui ,j +uk ,i uk ,j ),
2
which defines the symmetric second-rank tensor E, known as the Green–

Saint-Venant strain tensor and sometimes called the Lagrangian strain ten-
sor .1 Clearly, E(r) describes the deformation of the infinitesimal neighbor-

hood of r, and the tensor field E that of the whole body; E(r) = 0 for all r
in R if and only if the displacement is a rigid-body one.
The deformation of a region R is called homogeneous if E is constant. It
is obvious that a necessary and sufficient condition for the deformation to
be homogeneous is that the ui ,j are constant, or equivalently, that u varies
linearly with r,
Infinitesimal Strain and Rotation

We further define the tensor ε and ω, respectively symmetric and anti-
symmetric, by
1 1
εij = (uj ,i +ui ,j ), ωij = (ui ,j −uj ,i ),
2 2
so that ui ,j = εij + ωij , and
1
Eij = εij + (εik εjk − εik ωjk − ωik εjk + ωik ωjk ).
2
If |εij | 1 and |ωij | 1 for all i, j, then ε is an approximation to E
and is known as the infinitesimal strain tensor . The displacement field is
then called small or infinitesimal . Moreover, ω can then be defined as the
infinitesimal rotation tensor : if ε = 0, then α = ω, and therefore dx∗ =
(I + ω) x. Now
.
(I + ω)T (I + ω) = I + ω T + ω + ω T ω = I + ω T + ω = I;
In other words, a matrix of the form I + ω, where ω is any antisymmetric

matrix whose elements are small, is approximately orthogonal .
The tensor ω, because of its antisymmetry, has only three independent
components: ω32 = −ω23 , ω13 = −ω31 , and ω21 = −ω12 . Let these compo-
nents be denoted θ1 , θ2 , and θ3 , respectively. Then it is easy to show the
two reciprocal relations
1
θi = eijk ωkj , ωik = eijk θj .
2
Since, moreover, eijk εjk = 0 because of the symmetry of ε, the first relation
implies that
1
θi = eijk uk ,j ,
2
or, in vector notation,
1
θ = ∇ × u.
2
1
As will be seen in Chapter 8, E is only one of several tensors describing finite defor-
mation.
In a rigid-body displacement, then, dui = eijk θj dxk , or du = θ ×dr. That is,

θ is the infinitesimal rotation vector : its magnitude is the angle of rotation
and its direction gives the axis of rotation. Note that ∇ · θ = 0; a vector
field with this property is called solenoidal .
It must be remembered that a finite rotation is described by an orthog-
onal, not an antisymmetric, matrix. Since the orthogonality conditions are
six in number, such a matrix is likewise determined by only three indepen-
dent numbers, but it is not equivalent to a vector, since the relations among
the matrix elements are not linear.
Significance of Infinitesimal Strain Components

The study of finite deformation is postponed until Chapter 8. For now,
let us explore the meaning of the components of the infinitesimal strain
tensor ε. Consider, first, the unit vector n such that dr = ndr, where
dr = |dr|; we find that
.
dr∗2 = dr∗ · dr∗ = (1 + 2Eij ni nj )dr2 = (1 + 2εij ni nj ) r2 .
√ . .
But for α small, 1 + 2α = 1 + α, so that dr∗ = (1 + εij ni nj ) r. Hence
dr∗ − dr .
= εij ni nj
dr
is the longitudinal strain along the direction n (note that the left-hand side
is just the “engineering” definition of strain).
Next, consider two infinitesimal vectors, dr(1) = e1 dr(1) and dr(2) =
(1) (2)
e2 dr(2) . In indicial notation, we have dxi = δi1 dr(1) and dxi = δi2 dr(2) .
Obviously, dr(1) · dr(2) = 0. The displacement changes dr(1) to dr(1)∗ and
dr(2) to dr(2)∗ , where
(1)∗ (1)
dxi = (δij + ui ,j ) xj = (δi1 + ui ,1 ) r(1) ,
(2)∗ (2)
dxi = (δij + ui ,j ) xj = (δi2 + ui ,2 ) r(2) ,
so that
(1)∗ (2)∗ .
dr(1)∗ · dr(2)∗ = dxi dxi = (u2 ,1 +u1 ,2 ) r(1) dr(2) = 2ε12 dr(1) dr(2) .
But
π

(1)∗ (2)∗ (1)∗ (2)∗
dr · dr = dr dr cos − γ12 ,
2
where γ12 is the shear angle in the x1 x2 -plane (Figure 1.2.2). Consequently,
.
2ε12 = (1 + ε11 )(1 + ε22 )γ12 = γ12
for infinitesimal strains.

x2

dr(2)∗
π
− γ12
y + 2 :
6 W
dr(1)∗

*
dr(2) u

-
dr(1)
x1
Figure 1.2.2. Shear angle
Since the labeling of the axes is arbitrary, we may say in general that,
for i 6= j, εij = 21 γij . Both the εij and γij , for i 6= j, are referred to as shear
strains; more specifically, the former are the tensorial and the latter are the
conventional shear strains. A state of strain that can, with respect to some
axes, be represented by the matrix
 
0 γ 0
1
2
∗
ε =  2γ 0 0 
 1 
0 0 0
is called a state of simple shear with respect to those axes, and pure shear
in general.
It cannot be emphasized strongly enough that the tensor ε is an ap-
proximation to E, and therefore deserves to be called the infinitesimal strain
tensor, only if both the deformation and rotation are infinitesimal, that is,
if both ε and ω (or θ) are small compared to unity. If the rotation is finite,
then the strain must be described by E (or by some equivalent finite defor-
mation tensor, discussed further in Chapter 8) even if the deformation per
se is infinitesimal.
As an illustration, we consider a homogeneous deformation in which the
x1 x3 -plane is rotated counterclockwise about the x3 -axis by a finite angle θ,
while the the x2 x3 -plane is rotated counterclockwise about the x3 -axis by the
slightly different angle θ − γ, with |γ| 1. Since all planes perpendicular to
the x3 axis deform in the same way, it is sufficient to study the deformation
of the x1 x2 -plane, as shown in Figure 1.2.3. It is clear that, with respect
to axes rotated by the angle θ, the deformation is just one of simple shear,
x2
C
C
C θ−γ
C C
C9 CCW-
C
C
C
C

C
M
C θ
C ? x1
Figure 1.2.3. Infinitesimal shear strain with finite rotation
and can be described by the infinitesimal strain matrix given above. With
respect to the reference axes, we determine first the displacement of points
originally on the x1 -axis,
u1 (x1 , 0, x3 ) = −(1 − cos θ)x1 , u2 (x1 , 0, x3 ) = sin θx1 ,
and that of points originally on the x2 -axis,
u1 (0, x2 , x3 ) = − sin(θ − γ)x2 , u2 (0, x2 , x3 ) = −[1 − cos(θ − γ)]x2 .
The latter can be linearized with respect to γ:
u1 (0, x2 , x3 ) = −(sin θ −γ cos θ)x2 , u2 (0, x2 , x3 ) = −(1−cos θ −γ sin θ)x2 .
Since the deformation is homogeneous, the displacement must be linear in

x1 and x2 and therefore can be obtained by superposition:
u1 (x1 , x2 , x3 ) = −(1 − cos θ)x1 − (sin θ − γ cos θ)x2 ,
u2 (x1 , x2 , x3 ) = sin θx1 − (1 − cos θ − γ sin θ)x2 .

Knowing that u3 = 0, we can now determine the Green–Saint-Venant strain
tensor, and find, with terms of order γ 2 neglected,
 
0 γ 0
1
2
E =  2γ 0 0 ,
 1 
0 0 0
that is, precisely the same form as obtained for the infinitesimal strain with
respect to rotated axes. Moreover, the result is independent of θ. The reason
is that E measures strain with respect to axes that are, in effect, fixed in the
body.
Further discussion of the description of finite deformation is postponed
until Chapter 8.
Alternative Notations and Coordinate Systems

The “engineering” notations for the Cartesian strain components are εx ,
εy , and εz for ε11 , ε22 , and ε33 , respectively, and γxy for γ12 , and so on.
In cylindrical coordinates we find the strain components by taking the
symmetric part of ∇ ⊗ u as given by Equation (1.1.6):
∂ur ur 1 ∂uθ ∂uz

εr = , εθ = + , εz = ,
∂r r r ∂θ ∂z
∂uθ 1 ∂ur uθ ∂uz ∂ur 1 ∂uz ∂uθ
γrθ = + − , γrz = + , γθz = + .
∂r r ∂θ r ∂r ∂z r ∂θ ∂z
(1.2.1)
In spherical coordinates we similarly find, from Equation (1.1.8),
∂ur 1 ∂uφ ur 1 ∂uθ ur cot φ

εr = , εφ = + , εθ = + + uφ ,
∂r r ∂φ r r sin φ ∂θ r r
∂uφ 1 ∂ur uφ ∂uθ 1 ∂ur uθ

γrφ = + − , γrθ = + − , (1.2.2)
∂r r ∂φ r ∂r r sin φ ∂θ r
1 ∂uθ 1 ∂uφ cot φ
γφθ = + − uθ .
r ∂φ r sin φ ∂θ r
Volumetric and Deviatoric Strain

The trace of the strain tensor, εkk = ∇ · u, has a special geometric signif-
icance: it is the (infinitesimal) volumetric strain, defined as ∆V /V0 , where
∆V is the volume change and V0 the initial volume of a small neighborhood.
An easy way to show this is to look at a unit cube (V0 = 1) whose edges
parallel the coordinate axes. When the cube is infinitesimally deformed, the
lengths of the edges change to 1 + ε11 , 1 + ε22 , and 1 + ε33 , respectively,
.
making the volume (1 + ε11 )(1 + ε22 )(1 + ε33 ) = 1 + εkk . The volumetric
strain is also known as the dilatation.
The total strain tensor may now be decomposed as
εij = 31 εkk δij + eij .
The deviatoric strain or strain deviator tensor e is defined by this equation.

Its significance is that it describes distortion, that is, deformation without
volume change. A state of strain with e = 0 is called spherical or hydrostatic.
1.2.3. Principal Strains

It is possible to describe any state of infinitesimal strain as a superposition
of three uniaxial extensions or contractions along mutually perpendicular
axes, that is, to find a set of axes x∗i such that, with respect to these axes,
the strain tensor is described by
 
ε1 0 0
∗
ε =  0 ε2 0  .
 
0 0 ε3
Let n be unit vector parallel to such an axis; then the longitudinal strain
along n is εij ni nj . By hypothesis, the shear strains εij ni n̄j , where n̄ is any
unit vector perpendicular to n, are zero. Consequently the vector whose
components are εij nj is parallel to n, and if its magnitude is ε, then εij nj =
εni , or
εij nj − εni = (εij − εδij )nj = 0.
Then, in order that n 6= 0, it is necessary that
det(ε − εI) = −ε3 + K1 ε2 + K2 ε + K3 = 0, (1.2.3)
where K1 = εkk , K2 = 12 (εij εij − εii εkk ), and K3 = det ε are the so-called
principal invariants of the tensor ε. Since Equation (1.2.3) is a cubic equa-
tion, it has three roots, which are the values of ε for which the assumption
holds, namely ε1 , ε2 , and ε3 . Such roots are known in general as the eigen-
values of the matrix ε, and in the particular case of strain as the principal
strains.
The principal invariants have simple expressions in terms of the principal
strains:
K1 = ε1 + ε2 + ε3 ,
K2 = −(ε1 ε2 + ε2 ε3 + ε3 ε1 ),
K3 = ε1 ε2 ε3 .
To each εI (I = 1, 2, 3) there corresponds an eigenvector n(I) ; an axis

directed along an eigenvector is called a principal axis of strain.
We must remember, however, that a cubic equation with real coefficients
need not have roots that are all real: it may be that one root is real and the
other two are complex conjugates. It is important to show that the eigen-
values of a symmetric second-rank tensor — and hence the principal strains
— are real (if they were not real, their physical meaning would be dubious).
We can also show that the principal axes are mutually perpendicular.
Theorem 1 . If ε is symmetric then the εI are real.
Proof . Let ε1 = ε, ε2 = ε̄, where the bar denotes the complex conjugate.
(1) (2)
Now, if ni = ni , then ni = n̄i . Since εij nj = εni , we have n̄i εij nj =
εn̄i ni ; similarly, εij n̄j = ε̄n̄i , so that ni εij n̄j = ε̄n̄i ni . However, εij = εji ;
consequently ni εij n̄j = n̄i εij nj , so that (ε − ε̄)n̄i ni = 0. Since n̄i ni is, for
any nonzero vector n, a positive real number, it follows that ε = ε̄ (i.e., ε is
real).
Theorem 2 . If ε is symmetric then the n(I) are mutually perpendicular.
(1) (1) (2) (2)
Proof . Assume that ε1 6= ε2 ; then εij nj = ε 1 ni and εij nj = ε 2 ni .
(2) (1) (1) (2) (1) (2)
But ni εij nj − ni εij nj = 0 = (ε1 − ε2 )ni ni . Hence n(1) · n(2) = 0.
If ε1 = ε2 6= ε3 , then any vector perpendicular to n(3) is an eigenvector, so
that we can choose two that are perpendicular to each other. If ε1 = ε2 = ε3
(hydrostatic strain), then every nonzero vector is an eigenvector; hence we
can always find three mutually perpendicular eigenvectors. Q.E.D.
If the eigenvectors n(i) are normalized (i.e., if their magnitudes are de-
fined as unity), then we can always choose from among them or their nega-
tives a right-handed triad, say l(1) , l(2) , l(3) , and we define Cartesian coor-
dinates x∗i (i = 1, 2, 3) along them, then the direction cosines βij are given
by l(i) · ej , so that the strain components with respect to the new axes are
given by
(i) (j)
εij ∗ = lk ll εkl .
But from the definition of the n(i) , we have
(i) (i) (i)
ll εkl = εi δkl ll = εi lk (no sum on i),
so that
(i) (j)
ε∗ij = εi lk lk = εi δij (no sum on i),
or  
ε1 0 0
 def
ε∗ =  0 ε2 0  = Λ.

0 0 ε3
The same manipulations can be carried out in direct matrix notation.

(j)
With the matrix Λ as just defined, and with L defined by Lij = li (so that
T
L = β ), the equations defining the eigenvectors can be written as
εL = LΛ,
and therefore
ε∗ = βεβ T = LT εL = LT LΛ = Λ.
If one of the basis vectors (ei ), say e3 , is already an eigenvector of ε,

then ε13 = ε23 = 0, and ε33 is a principal strain, say ε3 . The remaining
principal strains, ε1 and ε2 , are governed by the quadratic equation
ε2 − (ε11 + ε22 )ε + ε11 ε22 − ε212 .

This equation can be solved explicitly, yielding

1 1q
ε1,2 = (ε11 + ε22 ) ± (ε11 − ε22 )2 + 4ε122 .
2 2
In the special case of simple shear, we have ε11 = ε22 = 0 and ε12 = 12 γ.
Consequently ε1,2 = ± 21 γ. With respect to principal axes, the strain tensor
is represented by  
1
γ 0 0
 2
ε∗ =  0 − 12 γ 0  ,

0 0 0
that is, the strain can be regarded as the superposition of a uniaxial extension
and a uniaxial contraction of equal magnitudes and along mutually perpen-
dicular directions. Conversely, any strain state that can be so represented is
one of pure shear.
1.2.4. Compatibility Conditions

If a second-rank tensor field ε(x1 , x2 , x3 ) is given, it does not automatically
follow that such a field is indeed a strain field, that is, that there exists a
displacement field u(x1 , x2 , x3 ) such that εij = 21 (uj ,i +ui ,j ); if it does, then
the strain field is said to be compatible.
The determination of a necessary condition for the compatibility of a
presumed strain field is closely related to the integrability theorem. Indeed,
if there were given a second-rank tensor field α such that αji = uj ,i , then
the condition would be just eikm αji ,k = 0. Note, however, that if there
exists a displacement field u, then there also exists a rotation field θ such
that εij + eijl θl = uj ,i . Consequently, the condition may also be written as
eikm (εij + eijl θl ),k = 0. But
eikm eijl θl ,k = (δjk δlm − δjm δkl )θl ,k = θm ,j ,
since θk ,k = 0. Therefore the condition reduces to
eikm εij ,k = −θm ,j .
The condition for a θ field to exist such that the last equation is satisfied
may be found by again invoking the integrability theorem, namely,
eikm ejln εij ,kl = 0. (1.2.4)
The left-hand side of Equation (1.2.4) represents a symmetric second-

rank tensor, called the incompatibility tensor , and therefore the equation
represents six distinct component equations, known as the compatibility
conditions. If the region R is simply connected, then the compatibility
conditions are also sufficient for the existence of a displacement field from
which the strain field can be derived. In a multiply connected region (i.e., a
region with holes), additional conditions along the boundaries of the holes
are required.
Other methods of derivation of the compatibility conditions lead to the
fourth-rank tensor equation
εij ,kl +εkl ,ij −εik ,jl −εjl ,ik = 0;
the sufficiency proof due to Cesaro (see, e.g., Sokolnikoff [1956]) is based on
this form. It can easily be shown, however, that only six of the 81 equations
are algebraically independent, and that these six are equivalent to (1.2.4).
A sufficiency proof based directly on (1.2.4) is due to Tran-Cong [1985].
The algebraic independence of the six equations does not imply that they
represent six independent conditions. Let the incompatibility tensor, whose
components are defined by the left-hand side of (1.2.4), be denoted R. Then
1
Rmn ,n = eikm ejln εij ,kln = eikm ejln (ui ,jkln +uj ,ikln ) = 0,
2
regardless of whether (1.2.4) is satisfied, because eikm uj ,ikln = ejln ui ,jkln = 0.
The identity Rmn ,n = 0 is known as the Bianchi formula (see Washizu
[1958] for a discussion).
Compatibility in Plane Strain
Plane strain in the x1 x2 -plane is defined by the conditions εi3 = 0 and
εij ,3 = 0 for all i, j. The strain tensor is thus determined by the two-
dimensional components εαβ (x1 , x2 ) (α, β = 1, 2), and the only nontrivial
compatibility condition is the one corresponding to m = n = 3 in Equation
(1.2.4), namely,
eαγ3 eβδ3 εαβ ,γδ = 0.
In terms of strain components, this equation reads
ε11 ,22 +ε22 ,11 −2ε12 ,12 = 0,
or, in engineering notation,
∂ 2 εx ∂ 2 εy ∂2γ
+ = , (1.2.5)
∂y 2 ∂x2 ∂x∂y
where γ = γxy . Lastly, an alternative form in indicial notation is
εαα ,ββ −εαβ ,αβ = 0. (1.2.6)
It can easily be shown that, to within a rigid-body displacement, the

only displacement field that is consistent with a compatible field of plane
strain is one of plane displacement, in which u1 and u2 are functions of x1

and x2 only, and u3 vanishes identically. The conditions εi3 = 0 are, in terms
of displacement components,
u3 ,3 = 0, u1 ,3 +u3 ,1 = 0, u2 ,3 +u3 ,2 = 0,
leading to
u3 = w(x1 , x2 ), uα = u0α (x1 , x2 ) − x3 w,α .
The strain components are now
εαβ = ε0αβ (x1 , x2 ) − x3 w,αβ ,
where ε0 is the strain derived from the plane displacement field u0 = u0α eα .
The conditions εαβ ,3 = 0 require that w,αβ = 0, that is, w(x1 , x2 ) = ax1 +
bx2 + c, where a, b and c are constants. The displacement field is thus the
superposition of u0 and of −ax3 e1 − bx3 e2 + (ax1 + bx2 + c)e3 , the latter
being obviously a rigid-body displacement. In practice, “plane strain” is
synonymous with plane displacement.

1. For each of the following displacement fields, with γ 1, sketch the
displaced positions in the x1 x2 -plane of the points initially on the sides
of the square bounded by x1 = 0, x1 = 1, x2 = 0, x2 = 1.
(a) u = 21 γx2 e1 + 12 γx1 e2

(b) u = − 12 γx2 e1 + 21 γx1 e2
(c) u = γx1 e2
2. For each of the displacement fields in the preceding exercise, determine

the matrices representing the finite (Green–Saint-Venant) and infini-
tesimal strain tensors and the infinitesimal rotation tensor, as well as
the infinitesimal rotation vector.
3. For the displacement field (a) of Exercise

√ 1, determine the longitudinal
strain along the direction (e1 + e2 )/ 2.
4. For the displacement field given in cylindrical coordinates by
u = ar er + brz eθ + c sin θ ez ,
where a, b and c are constants, determine the infinitesimal strain com-

ponents as functions of position in cylindrical coordinates.
5. Determine the infinitesimal strain and rotation fields for the displace-
ment field u = −w0 (x1 )x3 e1 + w(x1 )e3 , where w is an arbitrary contin-
uously differentiable function. If w(x) = kx2 , find a condition on k in
order that the deformation be infinitesimal in the region −h < x3 < h,
0 < x1 < l.
6. For the displacement-gradient matrix

 
4 −1 0
α =  1 −4 2  × 10−3 ,
 
4 0 6
determine
(a) the strain and rotation matrices,

(b) the volume strain and the deviatoric strain matrix,
(c) the principal strain invariants K1 , K2 , K3 ,
(d) the principal strains and their directions.
7. For the displacement field u = α(−x2 x3 e1 +x1 x3 e2 ), determine (a) the

strain and rotation fields, (b) the principal strains and their directions
as functions of position.
8. For the plane strain field
εx = Bxy, εy = −νBxy, γxy = (1 + ν)B(h2 − y 2 ),
where B, ν and h are constants,
(a) check if the compatibility condition is satisfied;

(b) if it is, determine the displacement field u(x, y), v(x, y) in 0 <
< L, −h < y < h such that u(L, 0) = 0, v(L, 0) = 0, and
x
∂v ∂u
− (L, 0) = 0.
∂x ∂y
Section 1.3 Mechanics of Continuous Bodies

1.3.1. Introduction
Global Equations of Motion

Mechanics has been defined as the study of forces and motions. It is
easy enough to define motion as the change in position of a body, in time,
with respect to some frame of reference. The definition of force is more
Section 1.3 / Mechanics of Continuous Bodies 27
elusive, and has been the subject of much controversy among theoreticians,
especially with regard to whether force can be defined independently of New-
ton’s second law of motion. An interesting method of definition is based on a
thought experiment due to Mach, in which two particles, A and B, are close
to each other but so far away from all other bodies that the motion of each
one can be influenced only by the other. It is then found that there exist
numbers mA , mB (the masses of the particles) such that the motions of the
particles obey the relation mA aA = −mB aB , where a denotes acceleration.
The force exerted by A on B can now be defined as FAB = mB aB , and FBA
is defined analogously. If B, rather than being a single particle, is a set of
several particles, then FAB is the sum of the forces exerted by A on all the
particles contained in B, and if A is also a set of particles, then FAB is the
sum of the forces exerted on B by all the particles in A.
The total force F on a body B is thus the vector sum of all the forces ex-
erted on it by all the other bodies in the universe. In reality these forces are
of two kinds: long-range and short-range. If B is modeled as a continuum
occupying a region R, then the effect of the long-range forces is felt through-
out R, while the short-range forces act as contact forces on the boundary
surface ∂R. Any volume element dV experiences a long-range force ρb dV ,
where ρ is the density (mass per unit volume) and b is a vector field (with
dimensions of force per unit mass) called the body force. Any oriented sur-
face element dS = n dS experiences a contact force t(n) dS, where t(n) is
called the surface traction; it is not a vector field because it depends not
only on position but also on the local orientation of the surface element as
defined by the local value (direction) of n.
If a denotes the acceleration field, then the global force equation of mo-
tion (balance of linear momentum) is
Z Z Z
ρb dV + t(n) dS = ρa dV. (1.3.1)
R ∂R R
When all moments are due to forces (i.e. when there are no distrib-
uted couples, as there might be in an electromagnetic field), then the global
moment equation of motion (balance of angular momentum) is
Z Z Z
ρx × b dV + x × t(n) dS = ρx × a dV, (1.3.2)
R ∂R R
where x is the position vector.

Equations (1.3.1)–(1.3.2) are known as Euler’s equations of motion,
applied by him to the study of the motion of rigid bodies. If a body is
represented as an assemblage of discrete particles, each governed by Newton’s
laws of motion, then Euler’s equations can be derived from Newton’s laws.
Euler’s equations can, however, be taken as axioms describing the laws of
motion for extended bodies, independently of any particle structure. They
are therefore the natural starting point for the mechanics of bodies modeled
as continua.
Lagrangian and Eulerian Approaches
The existence of an acceleration field means, of course, that the displace-
ment field is time-dependent. If we write u = u(x1 , x2 , x3 , t) and interpret
the xi as Lagrangian coordinates, as defined in Section 1.2, then we have
simply a = v̇ = ü; here v = u̇ is the velocity field, and the superposed dot
denotes partial differentiation with respect to time at constant xi (called
material time differentiation). With this interpretation, however, it must be
agreed that R is the region occupied by B in the reference configuration,
and similarly that dV and dS denote volume and surface elements measured
in the reference configuration, ρ is the mass per unit reference volume, and t
is force per unit reference surface. This convention constitutes the so-called
Lagrangian approach (though Lagrange did not have much to do with it) to
continuum mechanics, and the quantities associated with it are called La-
grangian, referential, or material (since a point (x1 , x2 , x3 ) denotes a fixed
particle or material point). It is, by and large, the preferred approach in solid
mechanics. In problems of flow, however — not only fluid flow, but also plas-
tic flow of solids — it is usually more instructive to describe the motion of
particles with respect to coordinates that are fixed in space — Eulerian or
spatial coordinates. In this Eulerian approach the motion is described not
by the displacement field u but by the velocity field v. If the xi are spatial
coordinates, then the material time derivative of a function φ(x1 , x2 , x3 , t),
defined as its time derivative with the Lagrangian coordinates held fixed,
can be found by applying the chain rule to be
∂φ
φ̇ = + vi φ,i .
∂t
The material time derivative of φ is also known as its Eulerian derivative
D
and denoted φ.
Dt
If the displacement field is infinitesimal, as defined in the preceding sec-
tion, then the distinction between Lagrangian and Eulerian coordinates can
usually be neglected, and this will generally be done here until finite defor-
mations are studied in Chapter 8. The fundamental approach is Lagrangian,
except when problems of plastic flow are studied; but many of the equations
derived are not exact for the Lagrangian formulation. Note, however, one
point: because of the postulated constancy of mass of any fixed part of B,
the product ρ dV is time-independent regardless of whether ρ and dV are
given Lagrangian or Eulerian readings; thus the relation
d
Z Z
ρψ dV = ρψ̇ dV
dt R R
is exact.
1.3.2. Stress
To determine how t depends on n, we employ the Cauchy tetrahedron illus-
trated in Figure 1.3.1.Assuming b, ρ, a and t(n) to depend continuously on
x, we have, if the tetrahedron is sufficiently small,
Z x2
.
ρb dV = ρb ∆V ;
R t(−e1 ) ∆A1 p
@ t(−e3 ) ∆A3
Z
. MBB
ρa dV = ρa ∆V ; @
R B n:
b@ t(n) ∆A
r
Z *

. B
b p
@
t(n) dS = t(n) ∆A

∂R
@@p x1
b
X
+ t(−ej ) ∆Aj
j p

= [t(n) − tj nj ] ∆A, t(−e2 ) ∆A2
x3
Figure 1.3.1. Cauchy tetrahedron
def
where tj = − t(−ej ). Thus we have, approximately,
∆V .
t(n) − tj nj + ρ(b − a) = 0.
∆A
This becomes exact in the limit as the tetrahedron shrinks to a point, i.e.
∆V /∆A → 0, so that
t(n) = tj nj ,
def
that is, t(·) is a linear function of its argument. If we define σij = ei · tj ,
then
ti (n) = σij nj ,
so that σ = [σij ] represents a second-rank tensor field called the stress tensor.
Denoting this tensor by σ, the preceding equation may be rewritten in direct
tensor notation as
t(n) = σn.
The force equation of motion (1.3.1) can now be written in indicial no-
tation as Z Z Z
ρbi dV + σij nj dS = ρai dV.
R ∂R R
By Gauss’s theorem we have

Z Z
σij nj dS = σij ,j dV ;
∂R R
therefore, Z
(σij ,j +ρbi − ρai ) dV = 0.
R
This equation, since it embodies a fundamental physical law, must be inde-
pendent of how we define a given body and therefore it must be valid for any
region R, including very small regions. Consequently, the integrand must
be zero, and thus we obtain the local force equations of motion (due to
Cauchy):
σij ,j +ρbi = ρai . (1.3.3)
When the relation between traction and stress is introduced into Equa-
tion (1.3.2), this equation becomes, in indicial notation,
Z Z Z
ρeijk xj bk dV + eijk xj σkl nl dS = ρeijk xj ak dV.
R ∂R R
By Gauss’s theorem,
Z Z
xj σkl nl dS = (xj σkl ),l dV
∂R R
Z
= (δij σkl + xj σkl ,l ) dV
R
Z
= (σkj + xj σkl ,l ) dV ;
R
therefore Z
eijk [xj (ρbk + σkl ,l −ρak ) + σkj ]dV = 0,
R
which, as a result of (1.3.3), reduces to
Z
eijk σkj dV = 0.
R
Since this result, again, must be valid for any region R, it follows that
eijk σkj = 0,
or equivalently,
σij = σji . (1.3.4)
In words: the stress tensor is symmetric.
In the usual “engineering” notation, the normal stresses σ11 , σ22 , and
σ33 are designated σx , σy , and σz , respectively, while the shear stresses are
written as τxy in place of σ12 , and so on. This notation is invariably used
in conjunction with the use of x, y, z for the Cartesian coordinates and of
ux , uy , and uz for the Cartesian components of a vector u, except that the
components of the displacement vector are usually written u, v, w, and the
body-force vector components are commonly written X, Y, Z rather than
ρbx , and so on. Thus the local force equations of motion are written in
engineering notation as
∂σx ∂τxy ∂τxz
+ + + X = ρax ,
∂x ∂y ∂z
and two similar equations.
The equations in cylindrical coordinates are obtained from Equation
(1.1.7), with the changes of notation self-explanatory:
∂σr σr − σθ 1 ∂τrθ ∂τrz
+ + + + R = ρar ,
∂r r r ∂θ ∂z
∂τrθ 2τrθ 1 ∂σθ ∂τθz
+ + + + Θ = ρaθ ,
∂r r r ∂θ ∂z
∂τrz τrz 1 ∂τθz ∂σz
+ + + + Z = ρaz .
∂r r r ∂θ ∂z
The corresponding equations in spherical coordinates are obtained from
Equation (1.1.9):
∂σr 1 ∂τrφ 1 ∂τrθ 2σr − σφ − σθ + τrφ cot φ
+ + + + R = ρar ,
∂σφ 1 ∂τφθ σφ cot φ − σθ cot φ + 3τrφ
+ + + Φ = ρaφ ,
∂φ r sin φ ∂θ r
∂τrθ 1 ∂τφθ 1 ∂σθ 3τrθ + 2τφθ cot φ
+ + + + Θ = ρaθ .
Projected Stresses
If n is an arbitrary unit vector, the traction t = σn has, in general, a
component parallel to n and one perpendicular to n. These are the projected
stresses, namely,
normal stress : σ(n) = n · t(n) = σij ni nj ,

p
shear stress : τ (n) = |t(n)|2 − [σ(n)]2 ;
note that this is the magnitude of the shear-stress vector
τ (n) = t(n) − nσ(n) = n × (t × n).
Principal Stresses
As in the case of strains, it is possible to find directions n such that
τ (n) = 0, so that t(n) = σ(n)n, or
σij nj − σni = (σij − σδij )nj = 0.

Then, for n 6= 0, we need
det(σ − σI) = −σ 3 + I1 σ 2 + I2 σ + I3 = 0,
where I1 = σkk , I2 = 12 (σij σij − σii σkk ), and I3 = det σ are the principal
invariants of σ. The principal stresses can now be defined exactly like the
principal strains; by Theorem 1 of 1.2.31 they are real, and by Theorem 2 the
principal axes of stress are mutually perpendicular. The principal invariants
of stress can be expressed in the form
I1 = σ1 + σ2 + σ3 ,
I2 = −(σ1 σ2 + σ2 σ3 + σ3 σ1 ),
I3 = σ1 σ2 σ3 .
The mean stress or hydrostatic stress is defined as σm = 31 I1 .

Stress Deviator
def
The stress deviator or deviatoric stress tensor s is defined by sij = σij −
σm δij . The principal invariants of the stress deviator are denoted J1 = skk ,
which vanishes identically, J2 (= 21 sij sij ), and J3 . The principal axes of s are
the same as those of σ, and the principal deviatoric stresses are sI = σI − 13 I1 .
J2 and J3 may be expressed in terms of the principal stresses through the
principal-stress differences σ1 − σ2 etc., namely,
1
J2 = [(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 ],
6
1
J3 = [(σ1 − σ2 )2 (σ1 − σ3 + σ2 − σ3 ) + (σ2 − σ3 )2 (σ2 − σ1 + σ3 − σ1 )
27
+(σ3 − σ1 )2 (σ3 − σ2 + σ1 − σ2 )].
Octahedral Stresses
Let the basis vectors (ei ) be directed along the principal axes, and sup-
pose that n is one of the eight vectors
1
n = √ (±e1 ± e2 ± e3 );
3
a regular octahedron can be formed with planes perpendicular to these vec-
tors. The traction on such a plane (called an octahedral plane) is
1
t(n) = √ (±σ1 e1 ± σ2 e2 ± σ3 e3 ).
3
1
Any cross-reference such as 1.2.3, not enclosed in parentheses, refers to a subsection,
unless it is specified as a figure or a table. With parentheses, for example (1.2.3), the
reference is to an equation.
The normal stress is

1 1
σ(n) = (σ1 + σ2 + σ3 ) = I1 = σm (mean stress),
3 3
and the shear stress is given by
1 2 1
[τ (n)]2 = (σ + σ22 + σ32 ) − (σ1 + σ2 + σ3 )2
3 1 9
2 2
= (σ + σ22 + σ32 − σ1 σ2 − σ2 σ3 − σ3 σ1 )
9 1
def 2
= τoct ,
where τoct is called the octahedral shear stress. By comparing the just-
derived result with the previously obtained expression for J2 in terms of the
principal stresses, it can be shown that
2 2
τoct = J2 . (1.3.5)
3
1.3.3. Mohr’s Circle

Let the x3 -axis coincide with the principal axis defined by σ3 , that is, let
e3 = n(3) . Now any unit vector n that is perpendicular to this axis may be
written as n = e1 cos θ + e2 sin θ. It follows that
t(n) = e1 (σ11 cos θ + σ12 sin θ) + e2 (σ12 cos θ + σ22 sin θ),
since σ13 = σ23 = 0, and
σ(n) = σ11 cos2 θ + σ22 sin2 θ + 2σ12 sin θ cos θ,
which will be designated σθ . By means of the trigonometric identities

cos2 θ = 21 (1 + cos 2θ), sin2 θ = 12 (1 − cos 2θ), and 2 sin θ cos θ = sin 2θ,
σθ may be rewritten as
1 1
σθ = (σ11 + σ22 ) + (σ11 − σ22 ) cos 2θ + σ12 sin 2θ. (1.3.6)
2 2
The projected shear-stress vector is
τ (n) = [σ12 (cos2 θ − sin2 θ) + (σ22 − σ11 ) sin 2θ cos 2θ](−e1 sin θ + e2 cos θ);
the quantity in brackets will be designated τθ , and clearly |τθ | = τ (n), since
the vector in parentheses is a unit vector. With the help of the trigonometric
identities cos 2θ = cos2 θ − sin2 θ, sin 2θ = 2 sin 2θ cos 2θ, we may write
1
τθ = σ12 cos 2θ + (σ22 − σ11 ) sin 2θ. (1.3.7)
2
From Equation (1.3.7) we may obtain the principal directions n(1) and
n(2) directly by finding the values of θ for which τθ vanishes, namely, those
that satisfy
2σ12
tan 2θ = , (1.3.8)
σ11 − σ22
unless σ12 and σ11 − σ22 are both zero, in which case τθ = 0 for all θ.
In the general case, if θ1 is a solution of Equation (1.3.8) (so that n(1) =
e1 cos θ1 + e2 sin θ1 ), then so is θ1 ± 21 π, showing the perpendicularity of
nondegenerate principal directions. As is readily seen, however, dσθ /dθ =
2τθ . It follows that the principal stresses σ1 and σ2 are the extrema of σθ , one
being the maximum and the other the minimum, again with the exception
of the degenerate case in which σθ is constant. The principal stresses, with
the numbering convention σ1 ≥ σ2 , are given by
1 q
2 ;
σ1 , 2 = (σ11 + σ22 ) ± 1
4
(σ11 − σ22 )2 + σ12
2
this convention is consistent with defining θ1 in such a way that
1
2
(σ11 − σ22 ) σ12
cos 2θ1 = q , sin 2θ1 = q ,
1 2
(σ11 − σ22 )2 + σ12 1 2
(σ11 − σ22 )2 + σ12
4 4
so that σθ1 = σ1 .
Using the just-derived expressions for cos 2θ1 and sin 2θ1 and the trigono-
metric identities
cos 2(θ − θ1 ) = cos 2θ cos 2θ1 + sin 2θ sin 2θ1 ,
sin 2(θ − θ1 ) = sin 2θ cos 2θ1 − sin 2θ1 cos 2θ,
we may rewrite Equations (1.3.6)–(1.3.7) as
1 1
σθ = (σ1 + σ2 ) + (σ1 − σ2 ) cos 2(θ − θ1 ), (1.3.9)
2 2
1
τθ = − (σ1 − σ2 ) sin 2(θ − θ1 ). (1.3.10)
2
Equations (1.3.9)–(1.3.10) are easily seen to be the parametric representation
of a circle, known as Mohr’s circle, in the σθ -τθ plane, with its center at
( 21 (σ1 + σ2 ), 0) and with radius 21 (σ1 − σ2 ), a value (necessarily positive in
view of the numbering convention) equal to the maximum of |τθ |. Note that
this maximum occurs when sin 2(θ − θ1 ) = ±1, that is, when θ = θ1 ± 41 π.
The significance of the angle θ1 , and other aspects of Mohr’s circle, can be
seen from Figure 1.3.2.
It can be shown that the maximum over all n (in three dimensions) of
the projected shear stress τ (n) is just the largest of the three maxima of τθ
found in the planes perpendicular to each of the principal directions. With
no regard for any numbering convention for the principal stresses, we have
1
τmax = max τ (n) = max{|σ1 − σ2 |, |σ2 − σ3 |, |σ1 − σ3 |}. (1.3.11)
n 2
τθ
1
2
(σ1 −σ2 )
s 11 , σ12 )
(σ

s(σ , τ )
2θ

2θ1
U θ θ

σ2
x2 K θ1
?
K
A

σθ
σ1

A A *

1
2 (σ1 +σ2 )
A
σ2 σ1
A A
A A
A
σ1

s(σ22 , −σ12 )

A
A
− 21 (σ1 −σ2 )
AU
σ2
x1
Figure 1.3.2. Mohr’s circle.
1.3.4. Plane Stress

Plane stress is defined by the conditions σi 3 = 0 (i = 1, 2, 3) and σij ,3 = 0,
so that the stress field is given by σαβ (x1 , x2 ) (α, β = 1, 2). The summation
convention applies to Greek indices ranging over 1, 2. Consequently, the
equilibrium equations without body force are
σαβ ,β = 0,
or (1) σ11 ,1 +σ12 ,2 = 0 and (2) σ12 ,1 +σ22 ,2 = 0, so that there exist functions
φα (x1 , x2 ) (α = 1, 2) such that (1) σ11 = φ1 ,2 , σ12 = −φ1 ,1 , and (2) σ22 =
φ2 ,1 , σ12 = −φ2 ,2 . Hence φ1 ,1 = φ2 ,2 , and therefore there exists a function
Φ(x1 , x2 ) such that φ1 = Φ,2 , φ2 = Φ,1 . Thus
σ11 = Φ,22 , σ22 = Φ,11 , σ12 = −Φ,12 ,
or, in two-dimensional indicial notation,
σαβ = δαβ Φ,γγ −Φ,αβ .
The function Φ is known as the Airy stress function.

In plane-stress problems described in “engineering” notation, the shear
stress τxy = τyx is sometimes written simply as τ , and the relation between
the stress components and the Airy stress function is accordingly written as
∂2Φ ∂2Φ ∂2Φ
σx = , σ y = , τ = − .
∂y 2 ∂x2 ∂x∂y
1.3.5. Boundary-Value Problems

The standard boundary-value problem of solid mechanics is the following:
find the fields u and σ throughout R if the given information consists of,
first, the body-force field b throughout R, and second, boundary conditions
on ∂R, namely, that at every point of ∂R there is a local basis (ei ) such that
either ti (n) or ui is prescribed for each i, i = 1, 2, 3.
It may be that ∂R consists of two parts, ∂Ru and ∂Rt , such that u is
prescribed on the former and t on the latter. This is not the most general
case, but it is often cited for convenience. The prescribed ti and ui will be
denoted tai and uai , respectively. The boundary conditions may be written
as
ui = uai on ∂Ru , nj σij = tai on ∂Rt , (1.3.12)
even if the boundary is not strictly divided into two parts; points at which
both displacement and traction components are prescribed may be regarded
as belonging to both ∂Ru and ∂Rt — in other words, ∂Ru and ∂Rt may be
thought of as overlapping.
The prescribed body-force field b and surface tractions tai are together
known as the loads, while the generally unknown surface tractions ti at the
points where the displacements ui are prescribed are called reactions. When-
ever displacements are prescribed, the body is said to be subject to external
constraints.1 There may, in addition, exist internal constraints which restrict
the displacement field in the interior of R; an example is incompressibility,
that is, the inability of a body or any part thereof to change its volume,
expressed by εkk = ∇ · u = 0.
A displacement field is called kinematically admissible if it is mathe-
matically well-behaved (for example, continuous and piecewise continuously
differentiable) and obeys the external and internal constraints, if any.
The boundary-value problem is called static if the data are independent
of time and the acceleration is assumed to be zero. It is called quasi-static if
the acceleration is neglected even though the data depend on time. In static
or quasi-static boundary-value problems, the equations of motion (1.3.3)
may be replaced by the equilibrium equations
σij ,j +ρbi = 0. (1.3.13)
A stress field σ that obeys the equilibrium equations (1.3.13) and the trac-
tion boundary conditions (1.3.12)2 is called statically admissible. If the ac-
celeration is not assumed to vanish, then the problem is dynamic, and then
1
More specifically, these are holonomic external constraints. A constraint is nonholo-
nomic if, for example, it is given by an inequality — a displacement component may be
required to have less than (or greater than) a specified value. Such a constraint is called
unilateral.
additional data are required, namely, the initial conditions consisting of the
displacement and velocity fields throughout R at the initial time.
Virtual Displacements
A virtual displacement field is defined as the difference between two
neighboring kinematically admissible displacement fields. In other words,
it is a vector field δu which is such that, if u is a kinematically admissible
displacement field, then so is u + δu. It is furthermore assumed that the
virtual displacement field is infinitesimal, that is, |δui ,j | 1.
Corresponding to a virtual displacement field δu we may define the vir-
tual strain field δε by δεij = 21 (δuj ,i +δui ,j ). Note that the operator δ
when applied to a field represents taking the difference between two possible
fields and is therefore a linear operator which commutes with partial differ-
entiation. Note further that if any displacement components are prescribed
on any part of the boundary then the corresponding virtual displacement
components vanish there, that is, δu = 0 on ∂Ru .
Virtual Work
Given a set of loads and a virtual displacement field δu, we define the
external virtual work as1
Z Z
δW ext = ρbi δui dV + tai δui dS.
R ∂Rt
The internal virtual work is defined as

Z
δW int = σij δεij dV.
R
Since σij = σji , we have σij δεij = σij δui ,j , and therefore
σij δεij = (σij δui ),j −σij ,j δui .
Using Gauss’s theorem, we obtain

Z Z
δW int = nj σij δui dS − σij ,j δui dV.
∂R R
Since, however, δui = 0 on ∂Ru , the surface integral may be restricted to

∂Rt . It follows that
Z Z
δW ext − δW int = (σij ,j +ρbi ) δui dV − (nj σij − tai ) δui dS.
R ∂Rt
The right-hand side vanishes for all virtual displacement fields δu if and
only if the quantities multiplying δui in both integrals vanish identically, that
1
A note on notation: we write δW instead of the more usual δW in order to indicate
that this is not a case of an operator δ applied to a quantity W .
is, if and only if the equilibrium equations (1.3.13) and the traction boundary
conditions (1.3.12)2 are satisfied. Thus the body is in equilibrium under the
applied loads if and only if the principle of virtual work, namely,
δW ext = δW int , (1.3.14)
is obeyed.
The principle of virtual work, also known as the principle of virtual
displacements, may be interpreted as an application of the method of
weighted residuals, whose essential idea is as follows. Suppose that a
certain stress field σ is not exactly statically admissible, and therefore the
equations obeyed by it have the form
σij ,j +ρbi + ρ ∆bi = 0 in R, nj σij = tai + ∆tai on ∂Rt ;
we may think of ∆b and ∆ta as being the residuals of the body force and ap-
plied surface traction, respectively. If we cannot make these residuals vanish
everywhere (which would make the stress field obey the equations exactly),
then we can try to make them vanish in some average sense, namely, by
multiplying them with a vector-valued weighting function, say w, belonging
to a suitable family (say W ) of such functions, such that
Z Z
ρ ∆b · w dV + ∆tai wi dS = 0
R ∂Rt
for every w belonging to W . If we identify W with the set of all virtual dis-
placement fields, the principle of virtual work results. An advantage of this
point of view is that it permits the application of the principle to dynamic
problems as well: since the weighting functions w are not a priori identified
with the virtual displacement fields, the inertia force −ρa may be added to
b, and ∆b may be interpreted as the residual of ρ(b − a).
Viewed in the light of the method of weighted residuals, the principle of
virtual work may be represented by the equation
Z Z
(σij ,j +ρbi )wi dV − (nj σij − tai )wi dS = 0 (1.3.15)
R ∂Rt
for any w that obeys the same conditions as a virtual displacement field,
that is, wi = 0 on ∂Ru , in addition to any internal constraints. Equation
(1.3.15) is also known as the weak form of the equilibrium equations with
the traction boundary conditions, and forms the foundation for many ap-
proximate methods of solution, based on different choices of the family W
to which the weighting functions w belong. If W contains only a finite num-
ber of linearly independent functions, then the body is said to be discretized .
Discretization is discussed below. When w is interpreted as a virtual velocity
field , then Equation (1.3.15) is called the principle of virtual velocities,
and, when, in addition, bi is replaced by bi − ai , as the dynamic principle

of virtual velocities.
A principle of the virtual-work type but different from the one just dis-
cussed is the principle of virtual forces (also called the principle of
complementary virtual work). This principle is based on the notion of
a virtual stress field δσ, defined (by analogy with the definition of a virtual
displacement field) as the difference between two statically admissible stress
fields. A virtual stress field therefore obeys
δσij ,j = 0 in R and nj δσij = 0 on ∂Rt .
The external and internal complementary virtual work are defined respec-
tively by
Z Z
c c
δW ext = uai nj δσij dS and δW int = εij δσij dV.
∂Ru R
An analysis similar to the one for virtual displacements leads to

Z Z
c c
δW int − δW ext = (ui − uai )nj δσij dS + [εij − 12 (ui ,j +uj ,i )]δσij dV.
∂Ru R
Consequently
c c
δW int − δW ext = 0 (1.3.16)
for all virtual stress fields δσ if and only if the strain field ε is compatible
with a kinematically admissible displacement field u.
Discretization
Approximate treatments of continuum mechanics are often based on a
procedure known as discretization, and virtual work provides a consistent
framework for the procedure. A displacement-based discretized model of
the body may be formulated as follows. Let q1 , ..., qN , qN +1 , ..., qN +K de-
note an ordered set of N + K scalars, called generalized coordinates. The
displacement field u(x) is assumed to be given by
NX
+K
u(x) = φn (x)qn , (1.3.17)
n=1
where φn (n = 1, ..., N + K) is a given set of vector-valued functions of

position. For given values of the qn , n = N + 1, ..., N + K, the displacement
field given by (1.3.17) is kinematically admissible for all values of the qn , n =
1, ..., N . The qn for n = 1, ..., N are called the free generalized coordinates,
and those for n = N + 1, ..., N + K, to be denoted qna (n = 1, ..., K), are
called the constrained generalized coordinates. The former will be assembled
in the 1 × N matrix q, and the latter in the 1 × K matrix q a . The integer N
is called the number of degrees of freedom, and K the number of constraints.

In particular, any or all of the qna may be zero.
The strain-displacement relation for infinitesimal deformation may be
written in direct vector notation as ε = (∇u)S , and consequently the strain
tensor at x is
NX
+K
ε(x) = (∇φn )S (x)qn . (1.3.18)
n=1
Before inserting the discretized displacement and strain fields into the
principle of virtual work, it is convenient to express them in matrix nota-
tion. Matrix notation for vector-valued quantities such as displacement and
force is obvious. The column matrices representing stress and strain are,
respectively,    

 σ11  
 ε11 

σ22  ε22 

  
 

 
 
 


 σ   
 ε 

33 33
σ= , ε= , (1.3.19)

 σ23  
 2ε23 

   



 σ13 





 2ε13 



σ12 2ε12
   
so that ε4 , ε5 , and ε6 represent conventional shear strains, and, most impor-

tant,
σ T ε = σij εij .
In certain simpler problems, the dimension of the column matrices σ, ε
may be less than six. In problems of plane stress or plane strain in the xy-
plane, with the component εz or σz ignored (because it does not participate
in any virtual work), the matrices are
   
 σx 
   εx 
 
σ= σy , ε= yε . (1.3.20)

 τ   
 γ  
xy xy
In problems with only one component each of normal stress and shear stress
(or longitudinal strain and shear strain) they are
( ) ( )
σ ε
σ= , ε= . (1.3.21)
τ γ
In every case, the stress and strain matrices are conjugate in the sense that
the internal virtual work is given by
Z
δW int = σ T δε dV. (1.3.22)
R
Equations (1.3.17) and (1.3.18) may now be rewritten in matrix notation

as
u(x) = Φ(x)q + Φa (x)q a , ε(x) = B(x)q + B a (x)q a , (1.3.23)
where x and u are the column-matrix representations of position and dis-

placement, respectively, that is, for three-dimensional continua,
   
 x1 
   u1 
 
x= 2 x , u= 2 u .

 x  
 u 
3 3
Since q a is prescribed, virtual displacement and strain fields are defined

only in terms of the variations of q:
δu(x) = Φ(x) δq, δε(x) = B(x) δq. (1.3.24)
We may now apply the principle of virtual work to these fields. Since
σ T δε = σ T Φ δq = (ΦT σ)T δq,
we obtain, using (1.3.24), the internal virtual work in discrete form:
δW int = QT δq,
where Z
Q= B T σ dV
R
is the internal generalized force matrix conjugate to q. Similarly, with the
body force per unit volume f = ρb and the prescribed surface traction ta
represented by matrices f and ta , the external virtual work is
δW ext = F T δq,
where Z Z
T
F = Φ f dV + ΦT ta dS
R ∂Rt
is the (known) external generalized force matrix conjugate to q. If the resid-
ual force matrix is defined by R = Q − F , then the equilibrium condition
may be written as
R = 0, (1.3.25)
a matrix equation representing N independent scalar equations.

1. If the acceleration field a is the material derivative v̇ of the velocity
field v, find the acceleration at the point (1, 1, 0) at the time t = 0 if
the velocity field in the Eulerian description is
v = C[(x31 + x1 x22 )e1 − (x21 x2 + x32 )e2 ]e−at ,
where C and a are constants.

2. With respect to a basis (ei ), a stress tensor is represented by the matrix

 
0.1 0.6 0.0
σ =  0.6 1.2 0.0  MPa
 
0.0 0.0 0.3
(a) Find the traction vector on an element of the plane 2x1 − 2x2 +
x3 = 1.
(b) Find the magnitude of the traction vector in (a), and the normal
stress and the shear stress on the plane given there.
(c) Find the matrix representing the stress tensor with respect to a
basis e∗i , where e∗i = βij ej with β given in Exercise 4 of Section
1.1.
3. For the stress tensor in Exercise 2, find
(a) the principal invariants I1 , I2 , I3 ,

(b) the principal stresses and principal directions,
(c) the octahedral shear stress,
(d) the component matrix of the stress deviator s with respect to the
basis (ei ),
(e) the principal deviatoric invariants J2 and J3 .
4. Draw the three Mohr’s circles for each of the following states of stress
(units are MPa; stress components not given are zero).
(a) Uniaxial tension, σ11 = 150

(b) Uniaxial compression, σ22 = −100
(c) Biaxial stress, σ11 = 50, σ22 = 100
(d) Biaxial stress, σ11 = 50, σ22 = −50
(e) Triaxial stress, σ11 = 80, σ22 = σ33 = −40
(f) σ11 = 50, σ22 = −10, σ12 = σ21 = 40, σ33 = 30
5. Find the maximum shear stress τmax for each of the stress states of
the preceding exercise.
6. Derive Equation (1.3.11) for τmax as follows: fix a Cartesian basis (ei )
coinciding with the principal stress directions, and form [τ (n)]2 for all
possible directions n by identifying n with the spherical unit vector er ,
so that [τ (n)]2 is a function of the spherical surface coordinates φ and
θ. Lastly, show that this function can become stationary only in the
planes formed by the principal stress directions.
7. For the plane stress field given in engineering notation by
A 2
σx = Axy, τ= (h − y 2 ), σy = 0,
2
where A and h are constants,
(a) show that it is in equilibrium under a zero body force,

(b) find an Airy stress function Φ(x, y) corresponding to it.
8. Show that the equilibrium equations in the absence of body force

(i.e. Equation (1.3.13) with bi = 0) are satisfied if σij = eikm ejln φkl ,mn ,
where φ is a symmetric second-rank tensor field.
9. If a stress field is given by the matrix

 
x21 x2 (a2 − x22 )x1 0
A  (a2 − x22 )x1 1 3 2
(x2 − 3a x2 ) 0 ,
 
3
0 0 2ax23
where A and a are constants, find the body-force field necessary for
the stress field to be in equilibrium.
10. The following displacement field is assumed in a prismatic bar with the
x1 axis being the longitudinal axis, and the cross-section defined by a
closed curve C enclosing an area A in the x2 x3 plane, such that
R
the x1 -
axis intersects the cross-sections at their centroids, so that A x2 dA =
R
A x3 dA = 0.
u1 = u(x1 ) − v 0 (x1 )x2 , u2 = v(x1 ), u3 = 0.
Show that the internal virtual work is

Z L
δW int = (P δu0 + M δv 00 ) dx1 ,
0
R R
where P = A σ11 dA and M = − A x2 σ11 dA.
11. If the bar described in Exercise 10 is subject to a body force and a

surface traction distributed along its cylindrical surface, show that the
external virtual work can be written as
Z L
δW ext = (p δu + q δv + m δv 0 ) dx1 .
0
Find expressions for p, q, and m.

Section 1.4 Constitutive Relations: Elastic

1.4.1. Energy and Thermoelasticity
Energy
As discussed in Section 1.3, the solution of a boundary-value problem in
solid mechanics requires finding the displacement field u and the stress field
σ, that is, in the most general case, nine component fields. Thus far, the
only field equations we have available are the three equations of motion (or
of equilibrium, in static and quasi-static problems). In certain particularly
simple problems, the number of unknown stress components equals that of
nontrivial equilibrium equations, and these may be solved to give the stress
field, though not the displacement field; such problems are called statically
determinate, and are discussed further in Section 4.1. In general, however,
constitutive relations that relate stress to displacement (more particularly, to
the strain that is derived from the displacement) are needed. Such relations
are characteristic of the material or materials of which the body is made, and
are therefore also called simply material properties. In this section a simple
class of constitutive relations, in which the current value of the strain at a
point depends only on the stress at that point, is studied; a body described
by such relations is called elastic. The influence of the thermal state (as
given, for example, by the temperature) on the stress-strain relations cannot,
however, be ignored. In order to maintain generality, instead of introducing
elasticity directly a more basic concept is first presented: that of energy.
The concept of energy is fundamental in all physical science. It makes
it possible to relate different physical phenomena to one another, as well as
to evaluate their relative significance in a given process. Here we focus only
on those forms of energy that are relevant to solid mechanics.
The kinetic energy of a body occupying a region R is
1
Z
K= ρv · v dV
2 R
(v = u̇ = velocity field), so that
d
Z
K̇ = K= ρv · a dV
dt R
(a = v̇ = acceleration field), because, even though the mass density ρ and

the volume element dV may vary in time, their product, which is a mass
element, does not (conservation of mass).
The external power acting on the body is
Z Z Z Z
P = ρb · v dV + t(n) · v dS = ρbi vi dV + ti (n)vi dS.
R ∂R R ∂R
Section 1.4 / Constitutive Relations: Elastic 45
But ti (n) = σij nj , and the divergence theorem leads to

Z
P = [(σij ,j +ρbi )vi + σij vi ,j ] dV.
R
In infinitesimal deformations, the difference between Eulerian and Lagrangian

. .
coordinates can be ignored, so that 12 (vi ,j +vj ,i ) = ε̇ij , and hence σij vi ,j =
σij ε̇ij . In addition, σij ,j +ρbi = ρai , so that
Z
def
P − K̇ = σij ε̇ij dV = Pd (deformation power).
R
If ε̇ ≡ 0 (rigid-body motion), then P = K̇. More generally, if Pd can be

neglected in comparison with P and K̇, then the body may approximately
be treated as rigid. If, on the other hand, it is K̇ that is negligible, then
the problem is approximately quasi-static. A problem of free vibrations may
arise if the external power P is neglected.
The heat f low into the body is
Z Z
Q= ρr dV − h(n) dS
R ∂R
(r = body heating or radiation, h(n) = heat outflow per unit time per unit
surface area with orientation n).
The first law of thermodynamics or principle of energy balance as-
serts that there exists a state variable u (internal-energy density) such that
d
Z
ρu dV = Q + Pd .
dt R
If we apply this law to the Cauchy tetrahedron, we obtain h(n) = h · n

(h = heat flux vector), so that
Z Z Z
Q= ρr dV − h · n dS = (ρr − div h) dV
R ∂R R
by Gauss’s theorem, and we obtain the local energy-balance equation,
ρu̇ = σij ε̇ij + ρr − div h. (1.4.1)
Thermoelasticity
The local deformation, as defined by the strain tensor ε, may be as-
sumed to depend on the stress tensor σ, on the internal-energy density u,
and on additional variables ξ1 , . . . , ξN (internal variables). By definition, a
body is called thermoelastic if ε everywhere depends only on σ and u. The
dependence is not arbitrary; it must be consistent with the second law of

thermodynamics.
We assume for the time being that at a fixed value of u, the relation
between strain and stress is invertible, so that σ may be regarded as a
function of ε and u. Possible restrictions on this assumption, resulting from
internal constraints, are discussed later.
The second law of thermodynamics for a thermoelastic body can be stated
as follows: there exists a state function η = η̄(u, ε) (entropy density) such
that η̇ = 0 in an adiabatic process, that is, in a process in which ρr −div h =
0.
As a consequence, the two equations
∂ η̄ ∂ η̄
u̇ + ε̇ij = 0 and ρu̇ − σij ε̇ij = 0
∂u ∂εij
must be satisfied simultaneously. If we now define the absolute temperature
T by
∂ η̄
T −1 = , (1.4.2)
∂u
then, upon eliminating u̇ between the two equations,
!
∂ η̄
σij + T ρ ε̇ij = 0.
∂εij
If, moreover, the ε̇ij are independent (i.e. if there are no internal constraints
as discussed in Section 1.3), then their coefficients must vanish, since any
five of the six independent components ε̇ij may be arbitrarily taken as zero.
The vanishing of the coefficients yields
∂ η̄
σij = −T ρ ,
∂εij
an equation that gives an explicit form to the assumed dependence of σ on
u and ε. The fact that this relation was derived on the basis of an adiabatic
process is irrelevant, since, by hypothesis, the relation is only among the
current values of σ, ε, and u and is independent of process.
It follows further that
T ρη̇ = ρu̇ − σij ε̇ij = ρr − div h. (1.4.3)
def R
If the total entropy of the body is defined by S = R ρη dV , then
Z Z
Ṡ = ρη̇ dV = T −1 (ρr − div h) dV
R R
r h
Z
= ρ − div + h · ∇T −1 dV
R T T
r h(n)
Z Z Z
= ρ dV − dS + h · ∇T −1 dV.
R T ∂R T R
We can now bring in the experimental fact that heat flows from the
hotter to the colder part of a body. The mathematical expression of this
fact is h · ∇T −1 ≥ 0, so that
r h(n)
Z Z
def
Ṡ − ρ dV − dS = Γ ≥ 0. (1.4.4)
R T ∂R T
Inequality (1.4.4), known as the global Clausius–Duhem inequality,
is usually taken to be the general form of the second law of thermodynamics
for continua, whether thermoelastic or not, though its physical foundation in
the general case is not so firm as in the thermoelastic case (see, e.g., Woods
[1981]). The quantity in parentheses is called the external entropy supply,
and Γ is the internal entropy production. In the thermoelastic case, clearly,
Z
Γ= h · ∇T −1 dV.
R
R
More generally, Γ may be assumed to be given by R ργ dV , where γ is
the internal entropy production per unit mass, and ργ contains, besides the
term h · ∇T −1 , additional terms representing energy dissipation. These are
discussed in Section 1.5. With the help of the divergence theorem, (1.4.4)
may be transformed into the equation
r h
Z
ρη̇ − ρ +∇· − ργ dV = 0.
R T T
The local Clausius–Duhem inequality is obtained when it is assumed, as
usual, that the equation must apply to any region of integration R, however
small, and, in addition, that γ is nonnegative:
r h

ρη̇ − ρ +∇· = ργ ≥ 0. (1.4.5)
T T
The assumptions underlying the derivation of (1.4.5) have been severely
criticized by Woods [1981]. The criticisms do not, however, apply to ther-
moelastic bodies, to which we now return.
Given the definition (1.4.2) of the absolute temperature T and the fact
that T is always positive, it follows from the implicit function theorem of
advanced calculus that we can solve for u as a function of η and ε [i.e.
u = ũ(η, ε)], and we can use η as a state variable. Since, by Equation
(1.4.3), u̇ = T η̇ + ρ−1 σij ε̇ij , we have T = ∂ ũ/∂η and
!
∂ ũ
σij − ρ ε̇ij = 0. (1.4.6)
∂εij
Again, in the absence of internal constraints, the ε̇ij can be specified inde-
pendently. The coefficients in Equation (1.4.6) must therefore all vanish,
yielding the relation
∂ ũ
σij = ρ ,
∂εij
in which the specific entropy η is the other variable. The equation thus
gives the stress-strain relation at constant specific entropy and is accordingly
known as the isentropic stress-strain relation. It is also called the adiabatic
stress-strain relation, since in a thermoelastic body isentropic processes are
also adiabatic. We must remember, however, that the relation as such is
independent of process.
We now drop the assumption that the body is free of internal constraints.
In the presence of an internal constraint of the form cij ε̇ij = 0, a term
pcij , with p an arbitrary scalar, may be added to the the stress without
invalidating Equation (1.4.6). This equation is therefore satisfied whenever
∂ ũ
σij = ρ − pcij .
∂εij
The stress, as can be seen, is not completely determined by the strain and
the entropy density.
As noted in 1.3.5, a commonly encountered internal constraint is incom-
pressibility. In an incompressible body the volume does not change, so that
δij ε̇ij = 0 and thus cij may be identified with δij . The stress is, accordingly,
determinate only to within a term given by pδij , where p is, in this case, a
pressure.
The use of the entropy density η as an independent state variable is
not convenient. A far more convenient thermal variable is, of course, the
temperature, since it is fairly easy to measure and to control. If the Helmholtz
free energy per unit mass is defined as
def
ψ = u − T η = ψ(T, ε),
then ψ̇ = −η Ṫ + ρ−1 σij ε̇ij , so that η = −∂ψ/∂T and, in the absence of

internal constraints,
∂ψ
σij = ρ . (1.4.7)
∂εij
Equation (1.4.7) embodies the isothermal stress-strain relation.
In addition to the stress-strain relations, the properties of a thermoelas-
tic continuum include the thermal stress coefficients and the specific heat.
The former are the increases in the stress components per unit decrease in
temperature with no change in the strain, that is,
∂2ψ

∂σij ∂η
βij = − = −ρ =ρ .
∂T ε=const
∂T ∂εij ∂εij
Clearly, β is a second-rank tensor. The specific heat (per unit mass) at
constant strain is
∂2ψ

∂u ∂ ∂ψ ∂η ∂η
C= = (ψ + ηT ) = + η + T = T = −T .
∂T ε=const ∂T ∂T ∂T ∂T ∂T 2
Linearization
For sufficiently small deviations in strain and temperature from a given
reference state, the stress-strain relation, if it is smooth, can be approximated
by a linear one. Let us consider a reference state “0” at zero strain and
temperature T0 , and let us expand ψ(T, ε) in a Taylor series about this
state:

1 ∂ 2 ψ

∂ψ ∂ψ
ψ(T, ε) = ψ0 + (T − T 0 ) + ε ij + 2
(T − T0 )2
∂T 0 ∂εij 0
2 ∂T 0
∂ 2 ψ 1 ∂ 2 ψ
+ (T − T0 )εij + εij εkl + . . .
∂T ∂εij 0 2 ∂εij ∂εkl 0
Now, ∂ψ/∂T |0 = −η0 (η0 = specific entropy in the reference state) and
∂ψ
= ρ−1 σij
0
(σ 0 = initial stress). Furthermore,
∂εij 0

∂ 2 ψ 1 ∂ 2 ψ
2
= − C0 , = −ρ−1 βij
0
,
∂T 0 T0 ∂T ∂εij 0
and
∂ 2 ψ

∂σij 0 0
ρ = = Cijkl = Cklij ,
∂εij ∂εkl 0 ∂εkl 0
the isothermal elastic modulus tensor (of rank 4) at the temperature T0 .
Thus
1
η = η0 + C0 (T − T0 ) + ρ−1 βij
0
εij
T0
and
0 0 0
σij = σij − βij (T − T0 ) + Cijkl εkl
are the constitutive relations of linear thermoelasticity.
0ε .
In an isentropic process, η ≡ η0 , so that T − T0 = −(T0 /ρC0 )βij ij
Consequently,
T0 0 0

0 0
σij = σij + Cijkl + βij βkl εkl ,
ρC0
which defines the isentropic or adiabatic elastic modulus tensor.
1.4.2. Linear Elasticity
Elasticity
The dependence of the stress-strain relation on the thermal state is often
ignored, and the simple relation σ = σ(ε) is assumed. It is then that a body
is called simply, in the traditional sense, elastic. The internal-energy density
or the free-energy density, as the case may be, may be replaced by the strain-
energy function W (ε) (per unit volume), so that the stress-strain relation,
again in the absence of internal constraints, is
∂W
σij = .
∂εij
This relation is exact in an isentropic process, with η at a fixed value, if

W (ε) is identified with ũ(η, ε), and in an isothermal process, with T at a
fixed value, if W (ε) is identified with ψ(T, ε).
The introduction of the strain-energy function into elasticity is due to
Green, and elastic solids for which such a function is assumed to exist are
called Green-elastic or hyperelastic. Elasticity without an underlying strain-
energy function is called Cauchy elasticity. Throughout this book, “elastic-
ity” means hyperelasticity as a matter of course.
Generalized Hooke’s Law

Linearization for an elastic continuum will be carried out with respect
to a reference configuration which is stress-free at the reference temperature
T0 , so that σ 0 = 0. We may now let C denote either the isothermal or the
isentropic modulus tensor, as appropriate, and under isothermal or isentropic
conditions we obtain the generalized Hooke’s law
σij = Cijkl εkl .
The Cijkl , called the elastic constants (recall that they depend on the tem-
perature), are components of a tensor of rank 4, likewise symmetric with
respect to the index pairs ij and kl; this symmetry reduces the number of
independent components from 81 to 36. But there is an additional symme-
try: since
∂ 2 W ∂ 2 W
Cijkl = = , (1.4.8)
∂εij ∂εkl ε=0 ∂εkl ∂εij ε=0
we also have Cijkl = Cklij , and thus the number of independent components
is further reduced to 21. This number may be reduced even more by material
symmetries. Of these, only isotropy is considered here.
If stress and strain are represented in matrix notation as given by Equa-
tion (1.3.19), then the stress-strain relation may be written as
σ = C ε, (1.4.9)
where the symmetric square matrix C = [CIJ ] (we use capital letters as in-
dices in the six-dimensional space of stress and strain components) is defined
as follows:
C11 = C1111 , C12 = C1212 , C14 = C1123 , C44 = C2323 , etc.

Assuming the matrix C to be invertible, we also have the strain-stress rela-

tions
ε = C −1 σ.
Reverting to tensor component notation, we may write these relations as
−1
εij = Cijkl σkl ,
where the compliance tensor C−1 is given as follows:

−1 −1 −1 −1 −1 −1 −1 −1
C1111 = C11 , C1122 = C12 , C1123 = 12 C14 , C1212 = 14 C66 , etc.
The complementary-energy function is defined for a general elastic ma-

terial as
W c = σij εij − W.
Note that if we assume W c = W c (ε, σ), then
∂W c ∂W
= σij − = 0,
∂εij ∂εij
so that W c is in fact a function of σ only, and
∂W c
εij = .
∂σij
For the linear material we have
1 −1 1
W c = Cijkl σij σkl = σ T C −1 σ,
2 2
and therefore
∂W c
εI = .
∂σI
Isotropic Linear Elasticity

The most general isotropic tensor of rank 4 has the representation
λδij δkl + µδik δjl + νδil δjk .
If Cijkl has this form, then, in order to satisfy the symmetry condition
Cijkl = Cjikl (symmetry of the stress tensor) we must have µ = ν. The
symmetry condition Cijkl = Cklij (existence of strain-energy function) is
then automatically satisfied. Thus
Cijkl = λδij δkl + µ(δik δjl + δil δjk ), (1.4.10)
so that the isotropic linear elastic stress-strain relation is
σij = λδij εkk + 2µεij .

λ and µ are known as the Lamé coefficients. In particular, µ = G, the shear

modulus. The matrix C takes the form
 
λ + 2µ λ λ 0 0 0

 λ λ + 2µ λ 0 0 0 

 λ λ λ + 2µ 0 0 0 
C= .
 
 0 0 0 µ 0 0 
 
 0 0 0 0 µ 0 
0 0 0 0 0 µ
Inverting this matrix, we obtain
λ + µ −λ/2 −λ/2
 
0 0 0

 −λ/2 λ + µ −λ/2 0 0 0 

1  −λ/2 −λ/2 λ + µ 0 0 0 
C −1 = .
 
0 0 0 3λ + 2µ 0 0

µ(3λ + 2µ) 



 0 0 0 0 3λ + 2µ 0 
0 0 0 0 0 3λ + 2µ
If we define the Y oung 0 s modulus as E = µ(3λ + 2µ)/(λ + µ) and the

Poisson’s ratio as ν = 21 λ/(λ + µ), then
1 −ν −ν
 
0 0 0

 −ν 1 −ν 0 0 0 

1  −ν −ν 1 0 0 0 
C −1 =  .
 
E 0 0 0 2(1 + ν) 0 0 
 
 0 0 0 0 2(1 + ν) 0 
0 0 0 0 0 2(1 + ν)
Note that the determinant of C −1 is 8(1 + ν)5 (1 − 2ν)E −6 , so that when

ν = 12 (incompressible material), the compliance matrix is singular. In that
case there exists no matrix C, that is, stress cannot be given as a function
of strain, as we already know.
From the nonzero elements of C −1 we can obtain the corresponding
components of the compliance tensor C−1 :
−1 1 −1
C11 = = C1111 , etc.,
E
−1 ν −1
C12 = − = C1122 , etc.,
E
−1 2(1 + ν) 1 1 −1
C44 = = = = 4C1212 , etc.
E µ G
Consequently the isotropic linearly elastic strain-stress relation in indicial
notation is
1
εij = [(1 + ν)σij − νσkk δij ]. (1.4.11)
E
If the only nonzero stress component is σ11 = σ, and if we denote ε11 by ε,

then Equation (1.4.11) gives the uniaxial stress-strain relation
σ = Eε. (1.4.12)
When σ33 = 0, then Equation (1.4.11), with the indices limited to the values
1 and 2, reads
1
εαβ = [(1 + ν)σαβ − νσγγ δαβ ],
E
which may be inverted to yield
E
σαβ = [(1 − ν)εαβ + νεγγ δαβ ]. (1.4.13)
1 − ν2
In plane problems, with σ and ε defined by Equations (1.3.20), C is given

by  
1 ν 0
E 
C= ν 1 0

1−ν 2
 
0 0 1
2
(1−ν)
for plane stress and by
 
1−ν ν 0
E
C= ν 1−ν 0
 
(1+ν)(1−2ν)
 
0 0 1
2
(1−2ν)
for plane strain. In a stress state composed of uniaxial stress and simple
shear, with stress and strain matrices given by (1.3.21), we have
" #
E 0
C= .
0 G
If E, ν are interpreted as representing the isothermal elastic stiffness,

then for small changes in temperature from the reference temperature T0
we may add to the elastic strain given by Equation (1.4.11) the thermal
strain α(T − T0 )δij , where α is the familiar coefficient of thermal expansion,
obtaining
1
εij = [(1 + ν)σij − νσkk δij ] + α(T − T0 )δij . (1.4.14)
E
The thermal stress coefficients βij are accordingly given by 3Kαδij , where
2 1 E
K =λ+ µ=
3 3 1 − 2ν
is the bulk modulus, which appears in the volumetric constitutive relation
σkk = 3K[εkk − 3α(T − T0 )].

We may also derive the deviatoric or distortional constitutive relation
sij = 2µeij ,
which includes, in particular, Hooke’s law in shear:
τ = Gγ. (1.4.15)
It is characteristic of an isotropic material that the volumetric and deviatoric

stress-strain relations are uncoupled . The uncoupling can also be seen from
the strain-energy function,
1 1
W = λ(εkk )2 + µεij εij = K(εkk )2 + µeij eij , (1.4.16)
2 2
and of the complementary-energy function,
1 1
Wc = (σkk )2 + sij sij . (1.4.17)
18K 4G
1.4.3. Energy Principles
Internal Potential Energy, Variations

R
Let Πint = R W dV be the total strain energy or internal potential energy
of the body. Given a virtual displacement field δu and the corresponding
virtual strain field δε, the first variation of Πint , denoted δΠint , is defined as
follows. We let Πint denote, more specifically, the internal potential energy
evaluated at the displacement field u, while Πint +∆Πint denotes the internal
potential energy evaluated at the varied displacement field u+δu. Assuming
the dependence of Πint on u to be smooth, we can write
1
∆Πint = δΠint + δ 2 Πint + . . . ,
2
where δΠint is linear in δu (and/or in its derivatives, and therefore also in
δε), δ 2 Πint (the second variation of Πint ) is quadratic, and so on. From the
definition of Πint ,
Z
∆Πint = [W (ε + δε) − W (ε)]dV
R
∂W 1 ∂2W
Z Z
= δεij dV + δεij δεkl dV + . . . .
R ∂εij 2 R ∂εij ∂εkl
According to our definition, the first integral in the last expression is δΠint ,
while the second integral is δ 2 Πint .
Even in the presence of internal constraints,
∂W
σij δεij = δεij .
∂εij
In view of the definition of internal virtual work as given in 1.3.5, therefore,
δW int = δΠint .
Total Potential Energy, Minimum Principle

For a fixed set of loads b, ta , let the external potential energy be defined
as Z Z
Πext = − ρbi ui dV − tai ui dS;
R ∂Rt
then the external virtual work over a virtual displacement field δu is
δW ext = −δΠext .
If we now define the total potential energy as Π = Πint + Πext , then we have
δW int − δW ext = δΠ, the first variation of Π. The principle of virtual work,
(1.3.15), then tells us that an elastic body is in equilibrium if and only if
δΠ = 0; (1.4.18)
that is, at equilibrium the displacement field makes the total potential energy
stationary with respect to virtual displacements.
It can further be shown that for the equilibrium to be stable, the total
potential energy must be a minimum, a result known as the principle of
minimum potential energy. If Π is a local minimum, then, for any
nonzero virtual displacement field δu, ∆Π must be positive. Since the first
variation δΠ vanishes, the second variation δ 2 Π must be positive. Πext is,
by definition, linear in u, and therefore, in the absence of significant changes
in geometry,1 δ 2 Πext = 0. Hence
Z
2 2
δ Π = δ Πint = Cijkl δεij δεkl dV.
R
An elastic body is thus stable under fixed loads if Cijkl εij εkl > 0, or, in
matrix notation, εT C ε > 0, for all ε 6= 0. A matrix having this property is
known as positive definite, and the definition can be naturally extended to
the fourth-rank tensor C. The positive-definiteness of C is assumed hence-
forth.
If the loads b and ta are not fixed but depend on the displacement, then
the just-derived principle is still valid, provided that the loads are derivable
from potentials, that is, that there exist functions φ(u, x) and ψ(u, x) defined
on R and ∂R, respectively, such that ρbi = −∂φ/∂ui and tai = −∂ψ/∂ui
(examples: spring support, elastic foundation). In that case,
Z Z
Πext = φ dV + ψ dS,
R ∂R
1
It is the changes in geometry that are responsible for unstable phenomena such as
buckling, discussed in Section 5.3.
and the second variation δ 2 Πext does not, in general, vanish, even in the
absence of significant geometry changes.
Complementary Potential Energy

From the principle of virtual forces we may derive a complementary
energy principle for elastic bodies. If the prescribed displacements are uai
(assumed independent of traction), and if the external and internal comple-
mentary potential energies for a given stress field σ are defined as
Z
Πcext =− uai nj σij dS
∂R
and Z
Πcint = W c (σ) dV,
∂R
respectively, then, analogously,
c c
δW int − δW ext = δΠc ,
where Πc = Πcext + Πcint is the total complementary potential energy. Ac-

cording to the virtual-force principle, then, Πc is stationary (δΠc = 0) if and
only if the stress field σ is related by the stress-strain relations to a strain
field that is compatible — both with internal constraints (if any) and with
the prescribed displacements or external constraints. Since Πc can also be
shown to be a minimum at stable equilibrium, this principle is the principle
of minimum complementary energy.
Discretized Elastic Body

If an elastic continuum is discretized as in 1.3.5, then the stress-strain
relation (1.4.9) implies
σ(x) = C[B(x)q + B a (x)q a ],
and therefore
Q = Kq + K a q a ,
where Z
K= B T C B dV (1.4.19)
R
and Z
Ka = B T C B a dV.
R
The symmetric matrix K is generally referred to as the stiffness matrix of
the discretized model. The equilibrium equation Q = F can now be solved
for the qn in terms of prescribed data as
q = K −1 (F − K a q a ),
provided that the stiffness matrix is invertible.

It is easy to extend the preceding result to include the effects of initial
and thermal stresses. The new result is
q = K −1 (F − K a q a − Q0 + F T ),
where Z Z
0 T 0
Q = B σ dV, FT = B T β(T − T 0 ) dV,
R R
σ 0 denoting the initial stress field, T 0 and T respectively the initial and
current temperature fields, and β the 6 × 1 matrix form of the the thermal-
stress coefficient tensor β defined in 1.4.1.
The invertibility requirement on K is equivalent to the nonvanishing of
its eigenvalues. Note that
Z
T
δq K δq = δεT C δε dV,
R
so that since the elastic strain energy cannot be negative, the left-hand side
of the preceding equation also cannot be negative — that is, K must be
positive semidefinite. The invertibility requirement therefore translates into
one of the positive definiteness of the stiffness matrix. This condition means
that any variation in q implies deformation, that is, that the model has no
rigid-body degrees of freedom.

1. Assume that the internal-energy density can be given as u = ū(ε, T ),
that the heat flux is governed by the Fourier law h = −k(T )∇T ,
and that r = 0. Defining the specific heat C = ∂ ū/∂T , write the
equation resulting from combining these assumptions with the local
energy-balance equation (1.4.1).
2. Let the complementary internal-energy density be defined by κ =

σij εij /ρ − u. Neglecting density changes, assume that the entropy
density can be given as η = η̂(σ, κ).
(a) Show that T −1 = −∂ η̂/∂κ.

(b) Assuming ε = ε(σ, η) and κ = κ̂(σ, η), show that T = −∂κ̂/∂η
and εij = ρ∂κ̂/∂σij .
(c) Defining the complementary free-energy density χ(σ, T ) = κ +
T η, show that η = ∂χ/∂T and the isothermal strain-stress rela-
tion is εij = ρ∂χ/∂σij .
3. Expand the complementary free energy χ (Exercise 2) in powers of σ

and T − T0 , and find the linearized expressions for η and ε.
4. A solid is called inextensible in a direction n if the longitudinal strain

along that direction is always zero. Find an isothermal relation for the
stress in an inextensible thermoelastic solid, and explain the meaning
of any undetermined quantity that may appear in it.
5. Show that, in an isotropic linearly elastic solid, the principal stress and
principal strain directions coincide.
6. Write the elastic modulus matrix C for an isotropic linearly elastic

solid in terms of the Young’s modulus E and the Poisson’s ratio ν.
7. Combine the stress-strain relations for an isotropic linearly elastic solid

with the equations of motion and the strain-displacement relations in
order to derive the equations of motion for such a solid entirely in
terms of displacement, using (a) λ and µ, (b) G and ν.
8. Derive the forms given in the text for C in plane stress and plane
strain.
9. Find the relation between the isothermal and adiabatic values of the
Young’s modulus E in an isotropic linearly elastic solid, in terms of
the linear-expansion coefficient α and any other quantities that may
be necessary.
10. Derive an explicit expression for the strain-energy function W (ε) in

terms of the “engineering” components of strain εx , . . . , γxy , ..., using
E and ν.
11. Show that, if the bar described in Exercise 10 of Section 1.3 is made
of a homogeneous, isotropic, linearly elastic material, then its internal
potential energy is
1
Z L
2 2
Πint = Ē (Au0 + Iv 00 ) dx1 ,
2 0
2
R
where A is the cross-sectional area, I = A x2 dA, and
E(1 − ν)
Ē = .
(1 − 2ν)(1 + ν)
12. Give a justification for replacing Ē in Exercise 11 by E.
13. Combining the results of Exercise 11, as modified by Exercise 12, with
those of Exercise 10 of Section 1.3, show that P = EAu0 and M =
EIv 00 .
Section 1.5 / Constitutive Relations: Inelastic 59
14. For the quantity Ξ defined by

Z
Ξ= {W (ε) − ρbi ui − σij [εij − 12 (ui ,j +uj ,i )]} dV
R
Z Z
− tai ui dS − ti (ui − uai ) dS,
∂Rt ∂Ru
show that, if u, ε, σ and (on ∂Ru ) t can be varied independently of

one another, then δΞ = 0 leads to six sets of equations describing a
static boundary-value problem in linear elasticity. The result is known
as the Hu–Washizu principle.
Section 1.5 Constitutive Relations: Inelastic

In this section a theoretical framework for the description of inelastic ma-
terials is presented, building on the discussion of elasticity in the preceding
section. The concepts introduced here are applied to the development of the
constitutive theory of plasticity in Chapter 3, following a brief survey of the
physics of plasticity in Chapter 2.
1.5.1. Inelasticity
Introduction
If we return to the weak (Cauchy) definition of an elastic body as one
in which the strain at any point of the body is completely determined by
the current stress and temperature there, then an obvious definition of an
inelastic body is as one in which there is something else, besides the current
stress and temperature, that determines the strain. That “something else”
may be thought of, for example, as the past history of the stress and tem-
perature at the point. While the term “past history” seems vague, it can
be defined quite precisely by means of concepts from functional analysis,
and a highly mathematical theory, known as the theory of materials with
memory, has been created since about 1960 in order to deal with it. The
dependence of the current strain on the history of the stress (and its converse
dependence of stress on strain history, whenever that can be justified) can
be expressed explicitly when the behavior is linear. The relevant theory is
known as the theory of linear viscoelasticity and is briefly reviewed later
in this section.
One way in which history affects the relation between strain and stress is
through rate sensitivity: the deformation produced by slow stressing is dif-
ferent — almost invariably greater — than that produced by rapid stressing.
A particular manifestation of rate sensitivity is the fact that deformation will

in general increase in time at constant stress, except possibly under hydro-
static stress; this phenomenon is called flow in fluids and creep in solids.
Rate sensitivity, as a rule, increases with temperature, so that materials
which appear to behave elastically over a typical range of times at ordinary
temperatures (at least within a certain range of stresses) become markedly
inelastic at higher temperatures. For this reason creep is an important de-
sign factor in metals used at elevated temperatures, while it may be ignored
at ordinary temperatures.
If strain and stress can be interchanged in the preceding discussion, then,
since a slower rate implies a greater deformation at a given stress, it accord-
ingly implies a lower stress at a given strain. Consequently, the stress will in
general decrease in time at a fixed strain, a phenomenon known as relaxation.
The rate sensitivity of many materials, including polymers, asphalt, and
concrete, can often be adequately described, within limits, by means of the
linear theory. The inelasticity of metals, however, tends to be highly non-
linear in that their behavior is very nearly elastic within a certain range
of stresses (the elastic range) but strongly history-dependent outside that
range. When the limit of the elastic range (the elastic limit) is attained as
the stress is increased, the metal is said to yield . When the elastic range
forms a region in the space of the stress components, then it is usually called
the elastic region and its boundary is called the yield surface.
Internal Variables
An alternative way of representing the “something else” is through an
array of variables, ξ1 , ..., ξn , such that the strain depends on these variables
in addition to the stress and the temperature. These variables are called
internal (or hidden) variables, and are usually assumed to take on scalar
or second-rank tensor values. The array of the internal variables, when the
tensorial ones, if any, are expressed in invariant form, will be denoted ξ. The
strain is accordingly assumed as given by
ε = ε(σ, T, ξ).
The presence of additional variables in the constitutive relations requires

additional constitutive equations. The additional equations that are postu-
lated for a rate-sensitive or rate-dependent inelastic body reflect the hypoth-
esis that, if the local state that determines the strain is defined by σ, T, ξ,
then the rate of evolution of the internal variables is also determined by the
local state:
ξ˙α = gα (σ, T, ξ). (1.5.1)
Equations (1.5.1) are known as the equations of evolution or rate equa-
tions for the internal variables ξα .
The relation ε = ε(σ, T, ξ) cannot always be inverted to give σ =

σ(ε, T, ξ), even when there are no internal constraints governing the strain.
As an illustration of this principle of nonduality, as it was called by Man-
del [1967], let us consider the classical model known as Newtonian viscosity.
When limited to infinitesimal deformation, the model can be described by
the equation
1
εij = σkk δij + εvij ,
9K
where εv is a tensor-valued internal variable, called the viscous strain, whose
rate equation is
1
ε̇v = s,
2η
where s is the stress deviator. The elastic bulk modulus K and the viscosity
η are functions of temperature; if necessary, a thermal strain α(T − T0 )δij
may be added to the expression for εij . It is clearly not possible to express
σ as a function of ε, T , and a set of internal variables governed by rate
equations. Instead, the stress is given by
σij = Kεkk δij + 2η ėij .
If it is possible to express the stress as σ(ε, T, ξ), then this expression

may be substituted in the right-hand side of (1.5.1), resulting in an alterna-
tive form of the rate equations:
ξ˙α = gα (σ(ε, T, ξ), T, ξ) = ḡα (ε, T, ξ).
Inelastic Strain
For inelastic bodies undergoing infinitesimal deformation, it is almost
universally assumed that the strain tensor can be decomposed additively
into an elastic strain εe and an inelastic strain εi :
εij = εeij + εiij ,

−1
where εeij = Cijkl σkl (with thermal strain added if needed). Newtonian
viscosity, as discussed above, is an example of this decomposition, with εi =
εv .
1.5.2. Linear Viscoelasticity

The aforementioned theory of linear viscoelasticity provides additional ex-
amples of the additive decomposition. For simplicity, we limit ourselves to
states that can be described by a single stress component σ with the conju-
gate strain ε; the extension to arbitrary states of multiaxial stress and strain
is made later. The temperature will be assumed constant and will not be
explicitly shown.
“Standard Solid” Model

Suppose that the behavior of a material element can be represented by
the mechanical model of Figure 1.5.1(a) (page 63), with force representing
stress and displacement representing strain. Each of the two springs models
elastic response (with moduli E0 and E1 ), and the dashpot models viscosity.
The displacement of the spring on the left represents the elastic strain εe , and
the displacement of the spring-dashpot combination on the right represents
the inelastic strain εi . Equilibrium requires that the force in the left-hand
spring be the same as the sum of the forces in the other two elements, and
therefore we have two equations for the stress σ:
σ = E0 ε e , σ = E1 εi + η ε̇i ,
where η is the viscosity of the dashpot element. For the total strain ε we
may write
σ
ε= + εi ,
E0
1 E1 i
ε̇i = σ− ε.
η η
The inelastic strain may consequently be regarded as an internal variable,
the last equation being its rate equation (E1 and η are, of course, functions
of the temperature).
Given an input of stress as a function of time, the rate equation for εi is
a differential equation that can be solved for εi (t):
1
Z t 0
i
ε (t) = e−(t−t )/τ σ(t0 ) dt0 ,
η −∞
where the reference time (at which εi = 0) is chosen as −∞ for convenience,

and τ = η/E1 is a material property having the dimension of time. In
particular, if σ(t) = 0 for t < 0 and σ(t) = σ0 (constant) for t > 0, then
εi (t) = (1/E1 )(1 − e−t/τ )σ0 ; this result demonstrates a form of creep known
as delayed elasticity.
The limiting case represented by E1 = 0 is known as the Maxwell model
of viscoelasticity. Note that in this case τ is infinite, so that the factor
exp[−(t − t0 )/τ ] inside the integral becomes unity, and the creep solution is
just εi (t) = (t/η)σ0 , displaying steady creep.
Generalized Kelvin Model

If models of viscoelasticity with more than one dashpot per strain com-
ponent are used — say a number of parallel spring-dashpot combinations in
E1 E1 E2 En
E0 E0
- pppp -
η η1 η2 ηn
(a) (b)
E1 E2 En
p p p p p p p E∞
η1 η2 ηn
?
(c)
6 6
E0 E∞ η
η
? ?
(d) (e)
Figure 1.5.1. Models of linear viscoelasticity: (a) “standard solid” model; (b)
generalized Kelvin model; (c) generalized Maxwell model; (d)
Maxwell model; (e) Kelvin model.
series with a spring, the so-called generalized Kelvin model shown in Figure
1.5.1(b) — then the inelastic strain is represented by the sum of the dash-
pot displacements, and every dashpot displacement constitutes an internal
variable. Designating these by ξα , we have
n
X
εi = ξα .
α=1
By analogy with the previous derivation, the rate equations for the ξα are,
if τα = ηα /Eα ,
σ ξα
ξ˙α = − .
ηα τα
As before, for a given stress history σ(t0 ), −∞ < t0 < t (where t is the current
time), the rate equations are ordinary linear differential equations for the ξα
that can be integrated explicitly:
Z t 1 −(t−t0 )/τα 0 0
ξα (t) = e σ(t ) dt .
−∞ ηα
The current strain ε(t) can therefore be expressed as

n
!
1
Z t 1 −(t−t0 )/τα
σ(t0 ) dt0 .
X
ε(t) = σ(t) + e
E0 −∞ η
α=1 α
With the uniaxial creep function J(t) defined by

n
1 1
1 − e−t/τα ,
X
J(t) = +
E0 α=1 Eα
the current strain can, with the help of integration by parts [and the as-
sumption σ(−∞) = 0], also be expressed as
Z t dσ 0
ε(t) = J(t − t0 ) dt .
−∞ dt0
If the stress history is given by σ(t0 ) = 0 for t0 < 0 and σ(t0 ) = σ

(constant) for t0 > 0, then, for t > 0, the strain is ε(t) = σJ(t). The creep
function can therefore be determined experimentally from a single creep
test, independently of an internal-variable model (unless the material is one
whose properties change in time, such as concrete). Similarly, in a relaxation
test in which ε(t0 ) = 0 for t0 < 0 and ε(t0 ) = ε (constant) for t0 > 0, the
measured stress for t > 0 has the form σ(t) = εR(t), where R(t) is the
uniaxial relaxation function. Under an arbitrary strain history ε(t0 ), the
stress at time t is then
Z t dε 0
σ(t) = R(t − t0 ) dt .
−∞ dt0
An explicit form of the relaxation function in terms of internal variables can
be obtained by means of the generalized Maxwell model [Figure 1.5.1(c)]. It
can be shown that, in general, R(0) = 1/J(0) = E0 (the instantaneous elastic
modulus), and R(∞) = 1/J(∞) = E∞ (the asymptotic elastic modulus),
with E0 > E∞ except in the case of an elastic material. In particular, E∞
may be zero, as in the Maxwell model [Figure 1.5.1(d)], while E0 may be
infinite, as in the Kelvin model [Figure 1.5.1(e)]. The relaxation function of
the Kelvin model is a singular function given by R(t) = E∞ + ηδ(t), where
δ(t) is the Dirac delta function.
A generalization of the preceding description from uniaxial to multiaxial
behavior is readily accomplished by treating J(t) and R(t) as functions whose
values are fourth-rank tensors. For isotropic materials, the tensorial form of
these functions is analogous to that of the elastic moduli as given in 1.4.2:
Jijkl (t) = J0 (t)δij δkl + J1 (t)(δik δjl + δik δjl ),
Rijkl (t) = R0 (t)δij δkl + R1 (t)(δik δjl + δik δjl ).

By analogy with Newtonian viscosity, it is frequently assumed (not al-

ways with physical justification) that the volumetric strain is purely elastic,
that is, εkk = σkk /3E. It follows that the creep and relaxation functions
obeying this assumption must satisfy the relations
1
3J0 (t) + 2J1 (t) = , 3R0 (t) + 2R1 (t) = 3K.
3K
The use of creep and relaxation functions makes it possible to represent

the strain explicitly in terms of the history of stress, and vice versa, with no
reference to internal variables. Indeed, the concept of internal variables is not
necessary in linear viscoelasticity theory: once the creep function is known,
the assumption of linear response is by itself sufficient to determine the strain
for any stress history (the Boltzmann superposition principle).
As was mentioned before, a mathematical theory of materials with mem-
ory, without internal variables, exists for nonlinear behavior as well. The
theory, however, has proved too abstract for application to the description
of real materials. Virtually all constitutive models that are used for nonlinear
inelastic materials rely on internal variables.
1.5.3. Internal Variables: General Theory
Internal Variables and Thermomechanics

An equilibrium state of a system is a state that has no tendency to
change with no change in the external controls. In the thermomechanics of
an inelastic continuum with internal variables, a local state (σ, T, ξ) may
be called a local equilibrium state if the internal variables remain constant
at constant stress and temperature, or, in view of Equation (1.5.1), if
gα (σ, T, ξ) = 0, α = 1, ..., n.
In an elastic continuum, every local state is an equilibrium state, though the

continuum need not be globally in equilibrium: a nonuniform temperature
field will cause heat conduction and hence changes in the temperature. The
existence of nonequilibrium states is an essential feature of rate-dependent in-
elastic continua; such states evolve in time by means of irreversible processes,
of which creep and relaxation are examples.1
The thermomechanics of inelastic continua consequently belongs to the
domain of the thermodynamics of irreversible processes (also known as non-
equilibrium thermomechanics). The fundamental laws of thermodynamics
1
A process is reversible if the equations governing it are unaffected when the time t
is replaced by −t; otherwise it is irreversible. In the rate-independent plastic continuum,
which forms the main subject of this book, irreversible processes occur even in the absence
of nonequilibrium states, as is seen in Chapter 3.
discussed in 1.4.1 are assumed to be valid in this domain as well, but there
is no full agreement in the scientific community about the meaning of such
variables as the entropy and the temperature, which appear in the second
law, or about the range of validity of this law (there is no comparable con-
troversy about the first law).
Entropy and temperature were defined in 1.4.1 for thermoelastic con-
tinua only; in other words, they are intrinsically associated with equilibrium
states. The assumption that these variables are uniquely defined at nonequi-
librium states as well, and obey the Clausius–Duhem inequality at all states,
represents the point of view of the “rational thermodynamics” school, most
forcefully expounded by Truesdell [1984] and severely criticized by Woods
[1981].
For continua with internal variables, another point of view, articulated
by Kestin and Rice [1970] and Bataille and Kestin [1975] (see also Germain,
Nguyen and Suquet [1983]), may be taken. According to this school of
thought, entropy and temperature may be defined at a nonequilibrium state
if one can associate with this state a fictitious “accompanying equilibrium
state” at which the internal variables are somehow “frozen” so as to have
the same values as at the actual (nonequilibrium) state. If we follow this
point of view, then we are allowed to assume the existence of a free-energy
density given by ψ(ε, T, ξ) such that the entropy density and the stress may
be derived from it in the same way as for elastic continua, with the internal
variables as parameters. Consequently the stress is given by Equation (1.4.7)
if it is entirely determined by ε, T, ξ, and the entropy density is given by
η = −∂ψ/∂T .
In any statement of the second law of thermodynamics, however, the
internal variables must be “unfrozen,” since this law governs irreversible
processes. Let us assume that the second law is expressed by the local
Clausius–Duhem inequality (1.4.5), and let us rewrite this inequality as
ρη̇ − T −1 (ρr − ∇ · h) + h · ∇T −1 = ργ ≥ 0.
With the help of the local energy-balance equation (1.4.1), the expression in
parentheses may be replaced by ρu̇ − σij ε̇ij , and, since the definition of the
free-energy density ψ leads to T η̇ − u̇ = −(η Ṫ + ψ̇), the left-hand side of the
inequality becomes
1
h · ∇T −1 − ρT −1 (ψ̇ + η Ṫ − σij ε̇ij ).
ρ
Furthermore,
∂ψ ∂ψ X ∂ψ
ψ̇ = ε̇ij + Ṫ + ξ˙α .
∂εij ∂T α ∂ξα
With η and σ expressed in terms of ψ, the left-hand side of (1.4.5) — that

is, the internal entropy production — becomes

X ∂ψ
h · ∇T −1 − ρT −1 ξ˙α . (1.5.2)
α ∂ξα
This quantity must be nonnegative in any process and at any state, and in
particular when the temperature gradient vanishes. The Clausius–Duhem
inequality is therefore obeyed if and only if, in addition to the already men-
tioned heat-conduction inequality h · ∇T −1 ≥ 0, the material also obeys the
dissipation inequality (Kelvin inequality)
pα ξ˙α ≥ 0,
X
D= (1.5.3)
α
where
∂ψ
pα = −ρ (1.5.4)
∂ξα
is the “thermodynamic force” conjugate to ξα .
The preceding results must be generalized somewhat if the material pos-
sesses viscosity in the sense that generalizes the Newtonian model: the stress
depends continuously1 on the strain-rate tensor ε̇ in addition to the thermo-
dynamic state variables ε, T, ξ. The definition of a local equilibrium state
now requires the additional condition ε̇ = 0. The stress in the accompany-
ing equilibrium state is still given (by definition) by ρ ∂ψ/∂εij , but it is not
equal to the actual stress σ(ε, T, ξ; ε̇). Instead, it equals
def
σ(ε, T, ξ; 0) = σ e (ε, T, ξ)
(equilibrium stress or elastic stress). If the viscous stress is defined as σ v =

σ − σ e , then the additional term T −1 σij v ε̇ must be added to the internal
ij
entropy production as expressed by (1.5.2). The quantity σij v ε̇ is the viscous
ij
dissipation, and must also be nonnegative.
If the decomposition of the strain into elastic and inelastic parts is as-
sumed to take the form
ε = εe (σ, T ) + εi (ξ),
then, as was shown by Lubliner [1972], such a decomposition is compatible

with the existence of a free-energy density ψ(ε, ξ, T ) if and only if ψ can be
decomposed as
ψ(ε, T, ξ) = ψ e (ε − εi (ξ), T ) + ψ i (ξ, T ). (1.5.5)

1
The continuity of the dependence of the stress on the strain rate is what distinguishes
viscosity from, say, dry friction.
The Kelvin inequality then takes the special form
X ∂ψ i
D = Di − ρ ξ˙α ≥ 0,
α ∂ξα
def
where Di = σij ε̇iij is the inelastic work rate per unit volume. Note that this
rate may be negative, without violating the second law of thermodynamics,
if ψ i decreases fast enough, that is, if enough stored inelastic energy is lib-
erated. The contraction of muscle under a tensile force, driven by chemical
energy, is an example of such a process.
The Nature of Internal Variables

What are internal variables in general? In principle, they may be any
variables which, in addition to the strain (or stress) and temperature, define
the local state in a small2 neighborhood of a continuum. As we have seen,
the components of εi themselves may or may not be included among the
internal variables.
As a general rule, internal variables may be said to be of two types. On
the one hand, they may be “physical” variables describing aspects of the local
physico-chemical structure which may change spontaneously; for example, if
the material can undergo a chemical reaction or a change of phase, then a
quantity describing locally the extent of the reaction or the relative density
of the two phases may serve as an internal variable. Other internal variables
of this type include densities of structural defects, as discussed in Chapter 2.
On the other hand, internal variables may be mathematical constructs; they
are then called phenomenological variables. The inelastic strain εi itself is of
this type, as are the dashpot displacements in viscoelastic models. Here the
form of the functional dependence of the stress (or strain) on the internal
variables, and of their rate equations, is assumed a priori.
In the simplest constitutive models describing nonlinear inelastic mate-
rials, the internal variables are often assumed to consist of the εiij and an
additional variable κ, called a hardening variable. The rate equation for κ
is further assumed to be related to the rate equations for the εiij in such a
way κ̇ = 0 whenever ε̇i = 0, but, in a cyclic process at the end of which
the εiij return to their original values (see Figure 1.5.2), κ will have changed.
Usually, κ is defined so that κ̇ > 0 whenever ε̇i 6= 0. Two commonly used
definitions of κ are, first, the inelastic work , defined as
Z
def
κ= Di dt = Wi , (1.5.6)
2
“Small” means small enough so that the state may be regarded as uniform, but large
enough for the continuum viewpoint to be valid.
σ
Figure 1.5.2. Closed stress-strain cycle with inelastic deformation: the inelastic
strain returns to zero at the end of the cycle, but the internal
state may be different, so that internal variables other than the
inelastic strain may be necessary.
and second, the equivalent (or effective) inelastic strain,

Z r
2 i i def
κ= ε̇ij ε̇ij dt = ε̄i . (1.5.7)
3
The reason for the traditional factor of 23 in the latter definition (which is
due to Odqvist [1933]) is the following: if a specimen of a material that
is (a) isotropic and (b) characterized by inelastic incompressibility (so that
ε̇ikk = 0) is subjected to a uniaxial tensile or compressive stress, then the
inelastic strain-rate tensor must be given by
 
ε̇i 0 0
i
ε̇ =  0 − 12 ε̇i 0 ,
 
i
0 0 − 12 ε̇
q
so that 23 ε̇iij ε̇iij is just equal to |ε̇i |.
In practice there is little difference in the way the two types of internal
variables are used.1 Whether the functions involved are provided by physical
theory or by hypothesis, they contain parameters that must be evaluated
by comparison of theoretical predictions with experimental results. It is
seen in Chapter 2 that in the case of metal plasticity, physical theory has
been remarkably successful in providing a qualitative understanding of the
phenomena, but attempts to generate constitutive equations in terms of
physical variables have not met with success.
1.5.4. Flow Law and Flow Potential

Regardless of whether the inelastic strain components are directly included
among the internal variables, it is always possible to define a flow law , that
1
As is seen in Section 3.1, phenomenological variables are sometimes given physical-
sounding names.
is, a rate equation for εi , by applying the chain rule to the basic assumption
εi = εi (ξ). The result is
ε̇iij = gij (σ, T, ξ),
where
X ∂εiij
gij = gα ,
α ∂ξα
the gα being the right-hand sides of Equation (1.5.1).
Mainly for convenience, it is often assumed that the gij can be derived
from a scalar function g(σ, T, ξ), called a flow potential , by means of
∂g
gij = φ ,
∂σij
φ(σ, T, ξ) being a positive scalar function.
The flow potential g is commonly assumed to be a function of the stress
alone, the most frequently used form being g(σ, T, ξ) = J2 , where J2 is the
second stress-deviator invariant defined in 1.3.2. Since
∂ ∂skl ∂ 1 1

J2 = smn smn = δik δjl − δij δkl skl = sij ,
∂σij ∂σij ∂skl 2 3
it follows that the flow law has the form
ε̇iij = φ(σ, T, ξ)sij . (1.5.8)
One consequence of this flow law is that inelastic deformation is volume-

preserving, or, equivalently, that volume deformation is purely elastic — a
result that is frequently observed in real materials.
Generalized Potential and Generalized Normality
A stronger concept of the flow potential is due to Moreau [1970] and
Rice [1970, 1971] (see also Halphen and Nguyen [1975]). A function Ω of
σ, T, ξ is assumed to depend on stress only through the thermodynamic
forces pα , defined by (1.5.4), conjugate to the internal variables ξα , that is,
def
Ω = Ω(p, T, ξ)r, where p = {p1 , ..., pn }. It is further assumed that the rate
equations are
∂Ω
ξ˙α = . (1.5.9)
∂pα
Equations (1.5.9) represent the hypothesis of generalized normality, and
Ω is called a generalized potential .
The thermodynamic forces pα can be obtained as functions of σ by means
of the complementary free-energy density χ(σ, T, ξ) (also called the free-
enthalpy density or Gibbs function), defined by
χ = ρ−1 σij εij − ψ,

where ψ is the Helmholtz free-energy density. It can easily be shown that

∂χ
pα = ρ
∂ξα
and
∂χ
εij = ρ .
∂σij
It follows that
X ∂εij ∂2χ ˙ X ∂pα
ξ˙α = ρ ξ˙α .
X
ε̇iij = ξα =
α ∂ξα α ∂σij ∂ξα α ∂σij
Combining with (1.5.9), we find

X ∂Ω ∂pα
ε̇iij = ,
α ∂pα ∂σij
or
∂Ω
ε̇iij = . (1.5.10)
∂σij
A sufficient condition for the existence of a generalized potential was
found by Rice [1971]. The condition is that each of the rate functions
gα (σ, T, ξ) depends on the stress only through its own conjugate thermo-
dynamic force pα :
∂Ωα
ξ˙α = ĝα (pα , T, ξ) = ,
∂pα
where, by definition,
Z pα
Ωα (pα , T, ξ) = ĝα (pα , T, ξ) dpα .
0
If we now define X
Ω(p, T, ξ) = Ωα (pα , T, ξ),
α
then Equations (1.5.9)–(1.5.10) follow.

The preceding results are independent of whether the inelastic strain
components εiij are themselves included among the internal variables.
For mathematical reasons, the generalized potential Ω is usually assumed
to be a convex function of p, that is, with the remaining variables not shown,
Ω(tp + (1 − t)p∗ ) ≤ tΩ(p) + (1 − t)Ω(p∗ )
for any admissible p, p∗ and any t such that 0 ≤ t ≤ 1. It follows from this
definition that
(∂Ω/∂p) · (p − p∗ ) ≥ Ω(p) − Ω(p∗ ), (1.5.11)
where ∂Ω/∂p is evaluated at p, and the dot defines the scalar product in n-
dimensional space. Thus, for any p∗ such that Ω(p∗ ) ≤ Ω(p), (p−p∗ )· ξ̇ ≥ 0.
The hypothesis of generalized normality has often been invoked in con-
stitutive models formulated by French researchers, though not always consis-
tently. An example is presented when constitutive theories of plasticity and
viscoplasticity are discussed in detail. Before such a discussion, however, it
is worth our while to devote a chapter to the physical bases underlying the
theories. We return to theory in Chapter 3.

1. Consider a model of linear viscoelasticity made up of a spring of mod-
ulus E∞ in parallel with a Maxwell model consisting of a spring of
modulus E 0 and a dashpot of viscosity η 0 .
(a) If the dashpot displacement constitutes the internal variable, find

the relation among the stress σ, the strain ε, and ξ.
(b) Find the rate equation for ξ (i ) in terms of ε and ξ and (ii ) in
terms of σ and ξ.
(c) Show that a model of the type shown in Figure 1.5.1(a) can be
found that is fully equivalent to the present one, and find the
relations between the parameters E0 , E1 , η of that model and
those of the present one.
2. For a standard solid model with creep function given by
1 1 1

J(t) = − − e−t/τ ,
E∞ E∞ E0
find an expression for the strain as a function of stress when the stress
history is such that σ = 0 for t < 0 and σ = ασ0 t/τ , α being a
dimensionless constant and σ0 a reference stress. Assuming E0 /E∞ =
1.5, sketch plots of σ/σ0 against E∞ ε/σ0 for α = 0.1, 1.0, and 10.0.
3. Show that the relation between the uniaxial creep function J(t) and
relaxation function R(t) of a linearly viscoelastic material is
Z t
J(t0 )R(t − t0 ) dt0 = t.
0
4. Show that if the free-energy density is as given by Equation (1.5.5), and

if the inelastic strain components εiij are themselves used as internal
variables, then the conjugate thermodynamic forces are just the stress
components σij .
5. Show that if the free-energy density is as given by Equation (1.5.5),

and if in addition the specific heat C is independent of the internal
variables, then the free-energy density reduces to the form
ψ(ε, T, ξ) = ψ e (ε − εi (ξ)) + ui (ξ) − T η i (ξ).
Discuss the possible meaning of ui and η i .
6. Find the flow law derived from a flow potential given by g(σ, T, ξ) =
f (I1 , J2 , J3 ).
7. Show that if χ(σ, T, ξ) is the complementary free-energy density and

pα is the thermodynamic force conjugate to ξα , then pα = ρ ∂χ/∂ξα .
Chapter 2
The Physics of Plasticity
Section 2.1 Phenomenology of Plastic Deformation
The adjective “plastic” comes from the classical Greek verb πλ άσσιν,
meaning “to shape”; it thus describes materials, such as ductile metals,
clay, or putty, which have the property that bodies made from them can
have their shape easily changed by the application of appropriately directed
forces, and retain their new shape upon removal of such forces. The shaping
forces must, of course, be of sufficient intensity — otherwise a mere breath
could deform the object — but often such intensity is quite easy to attain,
and for the object to have a useful value it may need to be hardened, for
example through exposure to air or the application of heat, as is done with
ceramics and thermosetting polymers. Other materials — above all metals
— are quite hard at ordinary temperatures and may need to be softened by
heating in order to be worked.
It is generally observed that the considerable deformations which occur
in the plastic shaping process are often accompanied by very slight, if any,
volume changes. Consequently plastic deformation is primarily a distortion,
and of the stresses produced in the interior of the object by the shaping
forces applied to the surface, it is their deviators that do most of the work.
A direct test of the plasticity of the material could thus be provided by
producing a state of simple shearing deformation in a specimen through the
application of forces that result in a state of shear stress. In a soft, semi-fluid
material such as clay, or soil in general, this may be accomplished by means
of a direct shear test such as the shear-box test, which is discussed in Section
2.3. In hard solids such as metals, the only experiment in which uniform
simple shear is produced is the twisting of a thin-walled tube, and this is not
always a simple experiment to perform. A much simpler test is the tension
test.
75
76 Chapter 2 / The Physics of Plasticity
2.1.1 Experimental Stress-Strain Relations
Tension Tests
Of all mechanical tests for structural materials, the tension test is the
most common. This is true primarily because it is a relatively rapid test and
requires simple apparatus. It is not as simple to interpret the data it gives,
however, as might appear at first sight. J. J. Gilman [1969]
The tensile test [is] very easily and quickly performed but it is not possible
to do much with its results, because one does not know what they really mean.
They are the outcome of a number of very complicated physical processes. .
. . The extension of a piece of metal [is] in a sense more complicated than
the working of a pocket watch and to hope to derive information about its
mechanism from two or three data derived from measurement during the
tensile test [is] perhaps as optimistic as would be an attempt to learn about
the working of a pocket watch by determining its compressive strength. E.
Orowan [1944]
Despite these caveats, the tension test remains the preferred method of
determining the material properties of metals and other solids on which it is
easily performed: glass, hard plastics, textile fibers, biological tissues, and
many others.
Stress-Strain Diagrams
The immediate result of a tension test is a relation between the axial force
and either the change in length (elongation) of a gage portion of the specimen
or a representative value of longitudinal strain as measured by one or more
strain gages. This relation is usually changed to one between the stress σ
(force F divided by cross-sectional area) and the strain ε (elongation divided
by gage length or strain-gage output), and is plotted as the stress-strain
diagram. Parameters that remain constant in the course of a test include the
temperature and the rate of loading or of elongation. If significant changes
in length and area are attained, then it is important to specify whether
the area used in calculating the stress is the original area A0 (nominal or
“engineering” stress, here to be denoted simply σe ) or the current area A
(true or Cauchy stress, σt ) — in other words, whether the Lagrangian or the
Eulerian definition is used — and whether the strain plotted is the change in
length ∆l divided by the original length l0 (conventional or “engineering”
strain, εe ) or the natural logarithm of the ratio of the current length l (=
l0 + ∆l) to l0 (logarithmic or natural strain, εl ).
Examples of stress-strain diagrams, both as σe versus εe and as σt versus
εl , are shown in Figure 2.1.1. It is clear that the Cauchy stress, since it
does not depend on the initial configuration, reflects the actual state in the
specimen better than the nominal stress, and while both definitions of strain
involve the initial length, the rates (time derivatives) of conventional and
Section 2.1 / Phenomenology of Plastic Deformation 77
˙ 0 and l/l,
logarithmic strain are respectively l/l ˙ so that it is the latter that is
independent of initial configuration. In particular, in materials in which it is
possible to perform a compression test analogous to a tension test, it is often
found that the stress-strain diagrams in tension and compression coincide to
a remarkable degree when they are plots of Cauchy stress against logarithmic
strain [see Figure 2.1.1(b)].
The rate of work done by the force is F l˙ = σe A0 l0 ε̇e = σt Alε̇l , so that
σe ε̇e and σt ε̇l are the rates of work per unit original and current volume,
respectively. While the calculation of Cauchy stress requires, strictly speak-
ing, measurement of cross-sectional area in the course of the test, in practice
this is not necessary if the material is a typical one in which the volume does
not change significantly, since the current area may be computed from the
volume constancy relation Al = A0 l0 .
As is shown in Chapter 8, the logarithmic strain rate ε̇l has a natural
and easily determined extension to general states of deformation, but the
logarithmic strain itself does not, except in situations (such as the tension
test) in which the principal strain axes are known and remain fixed. The
use of the logarithmic strain in large-deformation problems with rotating
principal strain axes may lead to erroneous results.
Compression Tests
As seen in Figure 2.1.1(b), the results of a simple compression test on
a specimen of ductile metal are virtually identical with those of a tensile
test if Cauchy stress is plotted against logarithmic strain. The problem is
that a “simple compression test” is difficult to achieve, because of the friction
that is necessarily present between the ends of the specimen and the pressure
plates. Compression of the material causes an increase in area, and therefore
a tendency to slide outward over the plates, against the shear stress on the
interfaces due to the frictional resistance. Thus the state of stress cannot be
one of simple compression. Lubrication of the interface helps the problem,
as does the use of specimens that are reasonably slender — though not so
slender as to cause buckling — so that, at least in the middle portion, a
state of simple compressive stress is approached.
Unlike ductile metals, brittle solids behave quite differently in tension
and compression, the highest attainable stress in compression being many
times that in tension. Classically brittle solids, such as cast iron or glass,
fracture almost immediately after the proportional limit is attained, as in
Figure 2.1.1(c). Others, however, such as concrete and many rocks, produce
stress-strain diagrams that are qualitatively similar to those of many ductile
materials, as in Figure 2.1.1(d). Of course, the strain scale is quite differ-
ent: in brittle materials the largest strains attained rarely exceed 1%. The
stress peak represents the onset of fracture, while the decrease in slope of
the stress-strain curve represents a loss in stiffness due to progressive crack-
σe (MPa) σ (MPa)
1500 Nickel-chrome steel 1500

Medium-carbon steel,
(b0 )
1200 heat-treated 1200

900 Cold-rolled steel 900 Points having
Medium-carbon steel, equal true strain Compression, σe vs. εe
annealed
ACj r

H
Hard bronze
600
+ Low-carbon
600
CAUCW rr Tension and compression,
σt vs. εl

steel
Tension, σe vs. εe
300 Soft brass
300 r
Fracture
H Annealed copper
H
Y
εe (%) ε (%)
10 20 30 40 50 10 20 30 40 50
(a) (b)
σ (MPa) σ (MPa)
1200
r Glass, compression
30
900
r
600 Gray cast iron,
compression 15
Compression
300 r Gray cast iron, tension
r Glass, tension r Tension
ε (%) ε (%)
1 2 3 4 0.25 0.5 0.75
(c) (d)
σ1 − σ3
σ1 − σ3 (MPa)
Insensitive
900 clay Dense sand, undisturbed
sensitive clay
1000 )

600 800
A
K
A 6
r400 A Confining pressure Loose sand,
300 r 200 remolded sensitive clay
(MPa)
r 100
9
600
ε (%) ε
10 20
(e) (f)
Figure 2.1.1. Stress-strain diagrams: (a) ductile metals, simple tension; (b)
ductile metal (low-carbon steel), simple tension and compres-
sion; (b’) yield-point phenomenon; (c) cast iron and glass, sim-
ple compression and tension; (d) typical concrete or rock, simple
compression and tension; (e) rock (limestone), triaxial compres-
sion; (f) soils, triaxial compression.
ing. The post-peak portion of the curve is highly sensitive to test conditions
and specimen dimensions, and therefore it cannot be regarded as a material
property. Moreover, it is not compression per se that brings about fracture,
but the accompanying shear stresses and secondary tensile stresses. Never-
theless, the superficial resemblance between the curves makes it possible to
apply some concepts of plasticity to these materials, as discussed further in
Section 2.3.
Unless the test is performed very quickly, soils are usually too soft to al-
low the use of a compression specimen without the application of a confining
pressure to its sides through air or water pressure. This confined compres-
sion test or triaxial shear test is frequently applied to rock and concrete as
well, for reasons discussed in Section 2.3. The specimen in this test is in
an axisymmetric, three-dimensional stress state, the principal stresses being
the longitudinal stress σ1 and the confining pressure σ2 = σ3 , both taken
conventionally as positive in compression, in contrast to the usual con-
vention of solid mechanics. The results are usually plotted as σ1 −σ3 (which,
when positive — as it usually is — equals 2τmax ) against the compressive
longitudinal strain ε1 ; typical curves are shown in Figure 2.1.1(e) and (f).
Elastic and Proportional Limits, Yield Strength

Some of the characteristic features of tensile stress-strain diagrams for
ductile solids when rate sensitivity may be neglected will now be described.
Such diagrams are characterized by a range of stress, extending from zero
to a limiting stress (say σo ) in which the stress is proportional to strain (the
corresponding strains are normally so small that it does not matter which
definitions of stress and strain are used); σo is called the proportional limit.
Also, it is found that the same proportionality obtains when the stress is
decreased, so that the material in this range is linearly elastic, described by
the uniaxial Hooke’s law given by Equation (1.4.12), that is, σ = Eε. The
range of stress-strain proportionality is thus also essentially the elastic range,
and the proportional limit is also the elastic limit as defined in Section 1.5.
When the specimen is stressed slightly past the elastic limit and the
stress is then reduced to zero, the strain attained at the end of the process
is, as a rule, different from zero. The material has thus acquired a permanent
strain.
Rate effects, which are more or less present in all solids, can distort the
results. The “standard solid” model of viscoelasticity discussed in 1.5.1, for
example, predicts that in a test carried out at a constant rate of stressing
or of straining, the stress-strain diagram will be curved, but no permanent
strain will be present after stress removal; the complete loading-unloading
diagram presents a hysteresis loop. The curvature depends on the test rate;
it is negligible if the time taken for the test is either very long or very short
compared with the characteristic time τ of the model.
Even in the absence of significant rate effects, it is not always easy to

determine an accurate value for the elastic or proportional limit. Some
materials, such as soft copper, present stress-strain curves that contain no
discernible straight portions. In others, such as aluminum, the transition
from the straight to the curved portion of the diagram is so gradual that the
determination of the limit depends on the sensitivity of the strain-measuring
apparatus. For design purposes, it has become conventional to define as the
“yield strength” of such materials the value of the stress that produces a
specified value of the “offset” or conventional permanent strain, obtained
by drawing through a given point on the stress-strain curve a straight line
whose slope is the elastic modulus E (in a material such as soft copper, this
would be the slope of the stress-strain curve at the origin). Typically used
values of the offset are 0.1%, 0.2% and 0.5%. When this definition is used,
it is necessary to specify the offset, and thus we would speak of “0.2% offset
yield strength.”
2.1.2 Plastic Deformation
Plastic Strain, Work-Hardening

The strain defined by the offset may be identified with the inelastic strain
as defined in 1.5.1. In the context in which rate sensitivity is neglected, this
strain is usually called the plastic strain, and therefore, if it is denoted εp ,
it is given by
σ
εp = ε − . (2.1.1)
E
The plastic strain at a given value of the stress is often somewhat different
from the permanent strain observed when the specimen is unloaded from this
stress, because the stress-strain relation in unloading is not always ideally
elastic, whether as a result of rate effects or other phenomena (one of which,
the Bauschinger effect, is discussed below).
Additional plastic deformation results as the stress is increased. The
stress-strain curve resulting from the initial loading into the plastic range is
called the virgin curve or flow curve. If the specimen is unloaded after some
plastic deformation has taken place and then reloaded, the reloading portion
of the stress-strain diagram is, like the unloading portion, approximately a
straight line of slope E, more or less up to the highest previously attained
stress (that is, the stress at which unloading began). The reloading then
follows the virgin curve. Similar results occur with additional unloadings and
reloadings. The highest stress attained before unloading is therefore a new
yield stress, and the material may be regarded as having been strengthened or
hardened by the plastic deformation (or cold-working). The increase of stress
with plastic deformation is consequently called work-hardening or strain-
hardening.
The virgin curve of work-hardening solids, especially ones without a

sharply defined yield stress, is frequently approximated by the Ramberg–
Osgood formula
σ σR σ m

ε= +α , (2.1.2)
E E σR
where α and m are dimensionless constants,1 and σR is a reference stress. If
m is very large, then εp remains small until σ approaches σR , and increases
rapidly when σ exceeds σR , so that σR may be regarded as an approximate
yield stress. In the limit as m becomes infinite, the plastic strain is zero
when σ < σR , and is indeterminate when σ = σR , while σ > σR would
produce an infinite plastic strain and is therefore impossible. This limiting
case accordingly describes a perfectly plastic solid with yield stress σR .
If the deformation is sufficiently large for the elastic strain to be ne-
glected, then Equation (2.1.2) can be solved for σ in terms of ε:
σ = Cεn , (2.1.3)
where C = σR (E/ασR )n , and n = 1/m is often called the work-hardening
exponent. Equation (2.1.3), proposed by Ludwik [1909], is frequently used in
applications where an explicit expression for stress as a function of strain is
needed. Note that the stress-strain curve representing (2.1.3) has an infinite
initial slope. In order to accommodate an elastic range with an initial yield
stress σE , Equation (2.1.3) is sometimes modified to read
 σE
 Eε,

 ε≤ ,
E
σ= n (2.1.4)
 σE Eε σE

, ε≥ .


σE E
Ultimate Tensile Strength

It must be emphasized that when the strain is greater than a few percent,
the distinction between the different types of stress and strain must be taken
into account. The decomposition (2.1.1) applies, strictly speaking, to the
logarithmic strain. The nature of the stress-strain curve at larger strains is,
as discussed above, also highly dependent on whether the stress plotted is
nominal or true stress [see Figure 2.1.1(b)]. True stress is, in general, an
increasing function of strain until fracture occurs. Since the cross-sectional
area of the specimen decreases with elongation, the nominal stress increases
more slowly, and at a certain point in the test it begins to decrease. Since,
very nearly, σe = σt exp(−εl ), it follows that
dσe = (dσt − σt dεl )exp(−εl ),
1
If m is a number other than an odd integer, then size − 2(σ/σR )m may be replaced
by size − 2|σ/σR |m−1 (σ/σR ) if the curve is the same for negative as for positive stress and
strain.
and therefore the nominal stress (and hence the load) is maximum when
dσt
= σt .
dεl
If Equation (2.1.3) is assumed to describe the flow curve in terms of Cauchy

stress and logarithmic strain, then the maximum nominal stress can easily
be seen to occur when εl = n.
The maximum value of nominal stress attained in a tensile test is called
the ultimate tensile strength or simply the tensile strength. When the spec-
imen is extended beyond the strain corresponding to this stress, its weakest
portion begins to elongate — and therefore also to thin — faster than the
remainder, and so a neck will form. Further elongation and thinning of the
neck — or necking — proceeds at decreasing load, until fracture.
Discontinuous Yielding
The stress-strain curves of certain impurity-containing metals, such as
mild steel and nitrogen-containing brass, present a phenomenon known as
discontinuous yielding. When the initial elastic limit is reached, suddenly
a significant amount of stretching (on the order of 1 or 2%, and thus con-
siderably larger than the elastic strain achieved up to that point) occurs at
essentially constant stress, of a value equal to or somewhat lower than the
initial elastic limit. If the value is the same, then it is called the yield point
of the material. If it is lower, then it is called the lower yield point, while
the higher value is called the upper yield point. The portion of the stress-
strain diagram represented by the constant stress value is called the yield
plateau, and the drop in stress, if any, that precedes it is called the yield
drop. Following the plateau, work-hardening sets in, as described above.
Figure 2.1.1(b’) shows a typical stress-strain diagram for a material with a
yield point.
As shown in the figure, the stress on the plateau is not really constant
but shows small, irregular fluctuations. They are due to the fact that plastic
deformation in this stage is not a homogeneous process but concentrated
in discrete narrow zones known as Lüders bands, which propagate along
the specimen as it is stretched, giving rise to surface marks called stretcher
strains.
When a specimen of a material with a yield point is loaded into the work-
hardening range, unloaded, and reloaded soon after unloading, the virgin
curve is regained smoothly as described previously. If, however, some time
— of the order of hours — is allowed to elapse before reloading, the yield
point recurs at a higher stress level (see Figure 2.1.2). This phenomenon is
called strain aging.
Bauschinger Effect, Anisotropy

σ

E
E

First Immediate Reloading
loading reloading after aging

0 ε
Figure 2.1.2. Strain aging
specimen of a ductile material that has been subjected to increasing

tensile stress and then unloaded (“cold-worked”) is different from a virgin
specimen. We already know that it has a higher tensile yield stress. If,
however, it is now subjected to increasing compressive stress, it is found
that the yield stress in compression is lower than before. This observation
is known as the Bauschinger effect [see Figure 2.1.3(a)].
The Bauschinger effect can be observed whenever the direction of strain-
ing is reversed, as, for example, compression followed by tension, or shearing
(as in a torsion test on a thin-walled tube) followed by shearing in the op-
posite direction. More generally, the term “Bauschinger effect” can be used
to describe the lowering of the yield stress upon reloading that follows un-
loading, even if the reloading is in the same direction as the original loading
(Lubahn and Felgar [1961]) [see Figure 2.1.3(b)]. Note the hysteresis loop
which appears with large strains, even at very slow rates of straining at
which the viscoelastic effects mentioned above may be neglected.
Another result of plastic deformation is the loss of isotropy. Following
cold-working in a given direction, differences appear between the values of
the tensile yield strength in that direction and in a direction normal to it.
These differences may be of the order of 10%, but are usually neglected in
practice.
Annealing, Recovery
The term “cold-working” used in the foregoing discussions refers to plas-
tic deformation carried out at temperatures below the so-called recrystalliza-
tion temperature of the metal, typically equal, in terms of absolute temper-
ature, to some 35 to 50% of the melting point (although, unlike the melting
point, it is not sharply defined); the reason for the name is explained in the
next section. The effects of cold-working, such as work-hardening, the Bau-
schinger effect, and induced anisotropy, can largely be removed by a process
called annealing, consisting of heating the metal to a relatively high tem-
σ

1
6
Tension
?
ε
Compression
)
σ (psi) (a)
4000
qq
(103
a a a qa q q q
σ psi)
3000 a
q a a a
qaq q q q
30
28
2000
q
26 ε (%) q
0.04 0.07
24
22
q q q q q
20
a a a
18
q
16 q q
14 qqa a Copper, tension
12
q
10 q a a Initial loading
8 a a a a qq qq q a and unloading
qq q
6 a a aaqa q q qq a qa Reloading
qqqqqqqqq a a a
aaaaaaaqaaq l 0.22aqa 0.30a
4
5.04 5.12 5.20
a aq
aaq
2
a a a εe (%)
0
0.02 0.10 0.84 0.96 13.50 13.58 13.66 13.74
(b)
Figure 2.1.3. Bauschinger effect: (a) classical; (b) generalized (from Lubahn
and Felgar [1961]).
perature (above the recrystallization temperature) and holding it there for

a certain length of time before slowly cooling it. The length of time neces-
sary for the process decreases with the annealing temperature and with the
amount of cold work.
Plastic deformation that takes place at temperatures in the annealing
range (i.e., above the recrystallization temperature) is known as hot-working,
and does not produce work-hardening, anisotropy, or the Bauschinger effect.
For metals with low melting points, such as lead and tin, the recrystallization
temperature is about 0◦ C and therefore deformation at room temperature
must be regarded as hot-working. Conversely, metals with very high melting
points, such as molybdenum and tungsten (with recrystallization tempera-
tures of 1100 to 1200◦ C can be “cold-worked” at temperatures at which the

metal is red-hot.
The recrystallization temperature provides a qualitative demarcation be-
tween stress-strain diagrams that show work-hardening and those that do
not. Within each of the two ranges, however, the stress needed to achieve
a given plastic deformation at a given strain rate also depends on the tem-
perature. In particular, it decreases with increasing temperature (see Figure
2.1.4).
σt
T
@
R

(((
εl
Figure 2.1.4. Temperature dependence of flow stress
A characteristic of some metals (including mild steel), with important

implications for design, is a change of behavior from ductile to brittle when
the temperature falls below the so-called transition temperature.
Softening (that is, a spontaneous decrease in yield strength) of work-
hardened metals also occurs at temperatures below recrystallization. This
process, whose rate is considerably slower than that of annealing, is called
recovery. The rate of recovery decreases with decreasing temperature, and
is negligible at room temperature for such metals as aluminum, copper and
steel. These metals may accordingly be regarded for practical purposes as
work-hardening permanently.
2.1.3 Temperature and Rate Dependence

The preceding discussion of the rates of annealing and recovery shows the
close relationship between temperature and rate. A great many physico-
chemical rate processes — specifically, those that are thermally activated —
are governed by the Arrhenius equation, which has the general form
rate ∝ e−∆E/kT , (2.1.5)

where k is Boltzmann’s constant (1.38 × 10−23 J/K), T is the absolute tem-

perature, and ∆E is the activation energy of the process. The rate sensitivity
of the work-hardening stress-strain curve itself increases with increasing tem-
perature. In a good many metals, the dependence on the plastic strain rate
of the stress required to achieve a given plastic strain can be approximated
quite well by ε̇r , where the exponent r (sometimes called simply the rate
sensitivity) depends on the plastic strain and the temperature, increasing
with both. Some typical results for r, obtained from tests at strain rates
between 1 and 40 per second, are shown in Table 2.1.1.
Table 2.1.1
Metal Temperature Value of r for a compression of

(◦ C) 10% 30% 50%
Aluminum 18 0.013 0.018 0.020
350 0.055 0.073 0.088
550 0.130 0.141 0.155
Copper 18 0.001 0.002 0.010
450 0.001 0.008 0.031
900 0.134 0.154 0.190
Mild steel 930 0.088 0.094 0.105
1200 0.116 0.141 0.196
Source: Johnson and Mellor [1973].
The Arrhenius equation (2.1.5) permits, in principle, the simultaneous

representation of the rate sensitivity and temperature sensitivity of the
stress-strain relation by means of the parameter ε̇exp(∆E/RT ), or, more
generally, ε̇f (T ), where f (T ) is an experimentally determined function, since
the activation energy ∆E may itself be a function of the temperature.
Creep
The preceding results were obtained from tests carried out at constant
strain rate (since the strains are large, total and plastic strain need not
be distinguished). Following Ludwik [1909], it is frequently assumed that
at a given temperature, a relation exists among stress, plastic (or total)
strain, and plastic (or total) strain rate, independently of the process, and
therefore this relation also describes creep, that is, continuing deformation at
constant stress. Such a relation is reminiscent of the “standard solid” model
of viscoelasticity, in which this relation is linear. It will be recalled that
this model describes both the rate dependence of the stress-strain relation
(discussed above in this section) and the increasing deformation at constant
stress known as creep, which in this case asymptotically attains a finite value
(bounded creep), though in the limiting case of the Maxwell model it becomes
steady creep. In fact, all linear spring-dashpot models of viscoelasticity lead
εc
Creep curve at high

stress and temperature
Tertiary
creep
Secondary
creep
Standard creep curve
Primary
creep
Creep curve at low
stress and temperature
Figure 2.1.5. Typical creep curves for metals.
to creep that is either bounded or steady.

For metals, the relation, if it exists, is nonlinear — many different forms
have been proposed — and therefore the resulting creep need not belong to
one of the two types predicted by the linear models. Typical creep curves for
a metal, showing the creep strain εc (equal to the total strain less the initial
strain) as a function of time at constant stress and temperature, are shown
in Figure 2.1.5. The standard curve is conventionally regarded as consisting
of three stages, known respectively as primary (or transient), secondary (or
steady), and tertiary (or accelerating) creep, though not all creep curves need
contain all three stages. At low stresses and temperatures, the primary creep
resembles the bounded creep of linear viscoelasticity, with a limiting value
attained asymptotically, and secondary and tertiary creep never appear. At
higher stress or temperature, however, the primary creep shows a logarithmic
or a power dependence on time:
εc ∝ ln t or εc ∝ tα ,
where α is between 0 and 1, a frequently observed value being 31 (Andrade’s

creep law). The logarithmic form is usually found to prevail below, and
the power form above, the recrystallization temperature.
Creep described by the power law can be derived from a formula relating
stress, creep strain and creep-strain rate that has the form (due to Nadai
[1950])
σ = C(εc )n (ε̇c )r , (2.1.6)
where C, n, and r depend on the temperature; this formula reduces to the

Ludwik equation (2.1.3) at constant strain rate, and implies a rate sensitivity
that is independent of the strain. At constant stress, the equation can be

integrated, resulting in a power law with α = r/(n + r).
Tertiary (accelerating) creep is generally regarded as resulting from struc-
tural changes leading to a loss of strength and, eventually, fracture. Whether
secondary (steady) creep really takes place over a significant time interval,
or is merely an approximate description of creep behavior near an inflec-
tion point of the creep curve, is not certain (see Lubahn and Felgar [1961],
pp. 136–141). In either case, however, one may speak of a minimum creep
rate characteristic of the metal at a given stress and temperature, if these
are sufficiently high. At a given stress, the temperature dependence of this
minimum creep rate is usually found to be given fairly closely by the Ar-
rhenius equation. Its dependence on stress at a given temperature can be
approximated by an exponential function at higher stresses, and by a power
function of the form ε̇cmin ∝ σ q , where q is an exponent greater than 1 (the
frequently used Bailey–Norton–Nadai law), at lower stresses. (Note that
Equation (2.1.6) describes the Bailey–Norton law if n = 0 and r = 1/q.) A
commonly used approximation for the creep strain as a function of time, at
a given stress and temperature, is
εc (t) = εc0 + ε̇cmin t,
where ε̇cmin is the minimum creep rate, and εc0 is a fictitious initial value
defined by the εc -intercept of the straight line tangent to the actual creep
curve at the point of inflection or in the steady-creep portion.
In many materials at ordinary temperatures, rate-dependent inelastic
deformation is insignificant when the stress is below a yield stress. A simple
model describing this effect is the Bingham model:
|σ| < σY ,

 0,

i
ε̇ =
σ

σ (2.1.7)
 1− Y
 , |σ| ≥ σY ,
|σ| η
where η is a viscosity, and the yield stress σY may depend on strain. The
Bingham model is the simplest model of viscoplasticity. Its generalizations
are discussed in Section 3.1.

1. Show that the relation between the conventional strain εe and the
logarithmic strain εl is εl = ln(1 + εe ).
2. It is assumed that the stress-strain relations of isotropic linear elastic-

ity, with Young’s modulus E and Poisson’s ratio ν, are exact in terms
of true stress and logarithmic strain. For uniaxial stress, find the rela-
tion (parametric if necessary) between the conventional stress and the
Section 2.2 / Crystal Plasticity 89
conventional strain. Show that the second-order approximation to the

relation is σe = E[εe − ( 12 + 2ν)ε2e ].
3. A uniaxial tension test produces a curve of true stress against loga-

rithmic strain that is fitted by σt = 2 × 105 εl in the elastic range and
1/6
σt = 635εl in the plastic range, with stresses in MPa. Determine (a)
the elastic-limit stress, (b) the logarithmic and conventional strains at
maximum load, and (c) the true and conventional stresses at maximum
load, assuming negligible volume change.
4. If the reference stress σR in the Ramberg–Osgood formula (2.1.2) is

the offset yield strength for a given permanent strain εR , find α in
terms of σR , εR , and E.
5. Find a formula describing a stress-strain relation that (a) is linear for

σ < σE , (b) asymptotically tends to ε ∝ σ m , and (c) is smooth at
σ = σE .
6. Suppose that in Equation (2.1.6) only C depends on the temperature.

Show that, for a given stress, the creep curves corresponding to differ-
ent temperatures are parallel if they are plotted as creep strain against
the logarithm of time.
7. Determine the form of the creep law resulting from Equation (2.1.6).
8. Assuming ε = σ/E + εc , and letting n = 0 in Equation (2.1.6), deter-

mine the resulting relaxation law, i. e. σ as a function of t when a
strain ε is suddenly imposed at t = 0 and maintained thereafter.
9. To what does the Bingham model described by Equation (2.1.7) reduce

when σY = 0? When η = 0?
Section 2.2 Crystal Plasticity

2.2.1 Crystals and Slip
Crystal Structure
Plasticity theory was developed primarily in order to describe the behav-
ior of ductile metals. Metals in their usual form are polycrystalline aggre-
gates, that is, they are composed of large numbers of grains, each of which
has the structure of a simple crystal.
A crystal is a three-dimensional array of atoms forming a regular lattice;
it may be regarded as a molecule of indefinite extent. The atoms vibrate
about fixed points in the lattice but, by and large, do not move away from
them, being held more or less in place by the forces exerted by neighboring
atoms. The forces may be due to ionic, covalent, or metallic bonding. Ionic
bonds result from electron transfer from electropositive to electronegative
atoms, and therefore can occur only in compounds of unlike elements. Ionic
crystal structures range from very simple, such as the sodium chloride struc-
ture in which Na+ and Cl− alternate in a simple cubic array, to the very
complex structures found in ceramics. Covalent bonds are due to the sharing
of electrons, and are found in diamond and, to some extent, in crystalline
polymers.
In a metallic crystal, the outer or valence electrons move fairly freely
through the lattice, while the “cores” (consisting of the nucleus and the filled
shells of electrons) vibrate about the equilibrium positions. The metallic
bond is the result of a rather complex interaction among the cores and the
“free” electrons. It is the free electrons that are responsible for the electrical
and thermal conductivity of metals.
s
s
X
sX
XX X s s s s
s
X
X X
XX

Xs s
X , \ , , ,
s
@ ,
, , ,
s
X
XXX , \@ ,
s s@c s s
, ,
c
, \\, , ,
cX c l@s
c
@l l

X X c @l l
@ls c ls c s
c @
X X X s
c @\c ,,
@\
, , , ,

X X Xc
, , ,
X X X s s s s, s
,
sX
, @@\, ,

s
XXX
(a) (b) (c)
Figure 2.2.1. Crystal structures: (a) hexagonal close-packed (hcp); (b) face-
centered cubic (fcc); (c) body-centered cubic (bcc).
The most common crystal structures in metals are the hexagonal close-
packed (hcp), face-centered cubic (fcc) and body-centered cubic (bcc), shown
in Figure 2.2.1. Because of the random orientation of individual grains
in a typical metallic body, the overall behavior of the aggregate is largely
isotropic, but such phenomena as the Bauschinger effect and preferred orien-
tation, which occur as a result of different plastic deformation of grains with
different orientations, demonstrate the effect of crystal structure on plastic
behavior. It is possible, however, to produce specimens of crystalline solids
— not only metals — in the form of single crystals of sufficiently large size
to permit mechanical testing.
Crystal Elasticity
The linear elastic behavior of a solid is described by the elastic modulus
matrix C defined in 1.4.2. The most general anisotropic solid has 21 inde-
pendent elements of C. For the isotropic solid, on the other hand, the only
nonzero elements of C are (a) C11 = C22 = C33 , (b) C44 = C55 = C66 , and
(c) C12 = C13 = C23 (the symmetry CIJ = CJI is not explicitly shown). But
only two of the three values are independent, since C11 = λ + 2µ, C44 = µ,
and C12 = λ, so that
1
C44 = (C11 − C12 ).
2
In a crystal with cubic symmetry (such as simple cubic, fcc or bcc), with
the Cartesian axes oriented along the cube edges, the nonzero elements of
C are the same ones as for the isotropic solid, but the three values C11 ,
C12 and C44 are independent. It may, of course, happen fortuitously that
the isotropy condition expressed by the preceding equation is satisfied for a
given cubic crystal; this is the case for tungsten.
A crystal with hexagonal symmetry is isotropic in the basal plane. Thus,
if the basal planes are parallel to the x1 x2 -plane, C66 = 21 (C11 − C12 ). The
following elements of C are independent: (a) C11 = C22 , (b) C33 , (c) C12 ,
(d) C13 = C23 , and (e) C44 = C55 .
The anisotropy of crystals is often studied by performing tension tests
on specimens with different orientations, resulting in orientation-dependent
values of the Young’s modulus E. If the maximum and minimum values are
denoted Emax and Emin , respectively, while Eave denotes the polycrystalline
average, the anisotropy index may be defined as (Emax − Emin )/Eave . Values
range widely: 1.13 for copper, 0.73 for α-iron, 0.2 for aluminum, and, as
indicated above, zero for tungsten.
Crystal Plasticity
Experiments show that plastic deformation is the result of relative mo-
tion, or slip, on specific crystallographic planes, in response to shear stress
along these planes. It is found that the slip planes are most often those
that are parallel to the planes of closest packing; a simple explanation for
this is that the separation between such planes is the greatest, and therefore
slip between them is the easiest, since the resistance to slip as a result of
interatomic forces decreases rapidly with interatomic distance. Within each
slip plane there are in turn preferred slip directions, which once more are
those of the atomic rows with the greatest density, for the same reason. A
slip plane and a slip direction together are said to form a slip system.
In hcp crystals, which include zinc and magnesium, the planes of closest
packing are those containing the hexagons, and the slip directions in those
planes are parallel to the diagonals. Hexagonal close-packed crystals there-
fore have three primary slip systems, although at higher temperatures other,
secondary, slip systems may become operative.
Face-centered cubic crystals, by contrast, have twelve primary slip sys-
tems: the close-packed planes are the four octahedral planes, and each con-
tains three face diagonals as the closest-packed lines. As a result, fcc metals,
such as aluminum, copper, and gold, exhibit considerably more ductility
than do hcp metals.
In body-centered cubic crystals there are six planes of closest packing
and two slip directions in each, for a total of twelve primary slip systems.
However, the difference in packing density between the closest-packed planes
and certain other planes is not great, so that additional slip systems become
available even at ordinary temperatures. Consequently, metals having a bcc
structure, such as α-iron (the form of iron found at ordinary temperatures),
tungsten, and molybdenum, have a ductility similar to that of fcc metals.
The preceding correlation between ductility and lattice type is valid in
very broad terms. Real metal crystals almost never form perfect lattices con-
taining one type of atom only; they contain imperfections such as geometric
lattice defects and impurity atoms, besides the grain boundaries contained
in polycrystals. In fact, these imperfections are the primary determinants
of crystal plasticity. Ductility must therefore be regarded as a structure-
sensitive property, as are other inelastic properties. It is only the ther-
moelastic properties discussed in 1.4.1 — the elastic moduli, thermal stress
(or strain) coefficients, and specific heat — that are primarily influenced by
the ideal lattice structure, and are therefore called structure-insensitive.
Slip Bands
In principle, slip in a single crystal can occur on every potential slip plane
when the necessary shear stress is acting. Observations, however, show slip
to be confined to discrete planes.1 When a slip plane intersects the outer
surface, an observable slip line is formed, and slip lines form clusters called
slip bands. In a given slip band, typically, a new slip line forms at a distance
of the order of 100 interatomic spacings from the preceding one when the
amount of slip on the latter has reached something of the order of 1,000
interatomic spacings. It follows from these observations that slip does not
take place by a uniform relative displacement of adjacent atomic planes.
Critical Resolved Shear Stress
It was said above that slip along a slip plane occurs in response to shear
stress on that plane. In particular, in a tensile specimen of monocrystalline
metal in which the tensile stress σ acts along an axis forming an angle φ with
the normal to the slip plane and an angle λ with the slip direction, then the
relation between σ and the resolved shear stress on the slip plane and in the
slip direction, τ , is
σ = (cos φ cos λ)−1 τ. (2.2.1)
It was found by Schmid [1924], and has been confirmed by many experiments,
1
Or, more generally, surfaces (slip surfaces), since slip may transfer from one slip plane
to another which intersects it in the interior of the crystal, especially in bcc metals.
that slip in a single crystal is initiated when the resolved shear stress on
some slip system reaches a critical value τc , which is a constant for a given
material at a given temperature and is known as the critical resolved shear
stress. This result is called Schmid’s law. The critical resolved shear stress
is, as a rule, very much higher for bcc metals (iron, tungsten) than for fcc
metals (aluminum, copper) or hcp metals (zinc, magnesium).
Theoretical Shear Strength
A value of the shear stress necessary to produce slip may be calculated
by assuming that slip takes place by the uniform displacement of adjacent
atomic planes. Consider the two-dimensional picture in Figure 2.2.2: two
d- * τ -x - d
2
h h h h h h h h h
h6 - -
?h h h h h h h h h
)τ
Figure 2.2.2. Slip between two neighboring rows of atoms
neighboring rows of atoms, the distance between the centers of adjacent

atoms in each row being d, and the distance between the center lines of
the two rows being h. Suppose the two rows to be in a stable equilibrium
configuration under zero stress. If one row is displaced by a distance d
relative to the other, a new configuration is achieved that is indistinguishable
from the first. A displacement of d/2, on the other hand, would lead to an
unstable equilibrium configuration at zero stress. As a first approximation,
then, the shear stress necessary to produce a relative displacement x may
be assumed to be given by
2πx
τ = τmax sin , (2.2.2)
d
and slip would proceed when τ = τmax . When the displacement x is small,
the stress-displacement relation is approximately linear: τ = 2πτmax x/d.
But a small displacement x between rows a distance h apart corresponds to
a lattice shear of γ = x/h, and Hooke’s law in shear reads τ = Gγ [Equation
(1.4.15)]. Consequently,
Gd
τmax = .
2πh
Since h ≡ d, the value G/6 is a first, structure-insensitive approximation to
the so-called theoretical shear strength of a crystal.
More refined calculations that take actual crystal structures into account
reduce the value of the theoretical shear strength to about G/30. In reality,
however, the shear strength of single crystals is less than this by one to
three orders of magnitude, that is, it is of order 10−3 G to 10−5 G. Only in
so-called whiskers, virtually perfect crystals about 1 µm in diameter, is a

shear strength of the theoretical order of magnitude observed.
2.2.2. Dislocations and Crystal Plasticity

The discrepancy between theoretical and observed shear strength, as well
as the observation of slip bands, have led to the inevitable conclusion that
slip in ordinary crystals must take place by some mechanism other than
the movement of whole planes of atoms past one another, and that it is
somehow associated with lattice defects. A mechanism based on a specific
defect called a dislocation was proposed independently by G. I. Taylor [1934]
and E. Orowan [1934].
Defects in Crystals
All real crystals contain defects, that is, deviations from the ideal crystal
structure. A defect concentrated about a single lattice point and involving
only a few atoms is called a point defect; if it extends along a row of many
atoms, it is called a line defect; and if it covers a whole plane of atoms, a
planar defect.
Point defects are shown in Figure 2.2.3. They may be purely structural,
such as (a) a vacancy or (b) an interstitial atom, or they may involve foreign
atoms (impurities): (c) a substitutional impurity, (d) an interstitial impu-
rity. As shown in the figure, point defects distort the crystal lattice locally,
hhhhhhhhhhh hhhhhhhh
h h hhh h h h h h h h h h h hh hh
h w
h h h h h hh hh h h h zh h h h h
h h hhh
hh h h h h h h hhhhhhhh
hhhhhhhhhhh hhhhhhhh
(a) (b) (c) (d)
Figure 2.2.3. Point defects: (a) vacancy; (b) interstitial atom; (c) substitu-
tional impurity; (d) interstitial impurity.
the distortion being significant over a few atomic distances but negligible
farther away. Planar defects, illustrated in Figure 2.2.4, include (a) grain
boundaries in polycrystals, and within single crystals, (b) twin boundaries
and (c) stacking faults.
Dislocations
The most important line defects in crystals are dislocations. The con-
cept of a dislocation has its origin in continuum mechanics, where it was
introduced by V. Volterra. Consider a hollow thick-walled circular cylin-
der in which a radial cut, extending through the wall, is made [see Figure
d d
Ad P d d C
d dP
P C
d
d A d d 4 4
Ad P
P
d P Pd
4B
d P
B
d d
P
d d d 4 4
d d d P d P

4A
AAA P
d d
A P P P A
d Ad
A Ad P d dP P d d P (i) C
4
5B
4
d d d d d Pd
P
4 4
A
d A d P P P
A B
d d Ad P d d d Pd
P 4A
A A P
P 4 4
d Ad d P d dP P dPd
A
4 4C 4
d P
A P
d Ad d dP
A C
d

Pd
P
d P
A
d A P
A
dPd
P
A
B
(a) B
4 4A 4
A 4C
4 4
C 5A
d (ii)
B
4 5B 4
Zd
d
Z 4 4
Zd
d A 4A
dZ Zd
d Zd 4 4C 4

d Zd

C
d Z dd d
Z
d

d

d Z
Z Z Z
d
Z
d
Z d Z dZ d
Z d d
ZdZ d Z Stacking faults in a face-centered cubic lattice.
d d d d d
Z
dZ
The normal stacking sequence of (111) planes is
Zdd d d d d
denoted by ABCA... Planes in normal relation
to one another are separated by 4, those with
d d d d d d

a stacking error by 5; (i) intrinsic stacking fault,
(ii) extrinsic stacking fault.
(b) (c) (from Hull and Bacon [1984])
Figure 2.2.4. Planar defects: (a) grain boundary; (b) twin boundary; (c) stack-
ing fault.
2.2.5(a)]. The two faces of the cut may be displaced relative to each other
by a distance b, either in the (b) radial or (c) axial direction, and then reat-
tached. The result is a Volterra dislocation, with Figures 2.2.5(b) and (c)

b

X
zX
y

.
...
....
.
.. .
.
..
.
.......

..
...
.

.....

....

.
.......

..
.

..
...
....

...
.
.......

...

...
...
. ..
......
b

X
X XX ..
.
.. X XX ........
X

XX
X=

X
>
(a) (b) (c)
Figure 2.2.5. Volterra dislocation: (a) Volterra cut; (b) edge dislocation; (c)
screw dislocation.
representing respectively an edge and a screw dislocation. When the rough

edges are smoothed, the result is a cylinder looking much as it did before
the operation, but containing a self-equilibrating internal stress field. If the
material is isotropic and linearly elastic, then the stress and displacement
fields can be calculated by means of the theory of elasticity. In particular,
the strain energy per unit length of cylinder is found to be

Gb2 R

W0 = ln − 1 (2.2.3a)
4π(1 − ν) a
for an edge dislocation and
Gb2 R

W0 = ln − 1 (2.2.3b)
4π a
for a screw dislocation, where G is the shear modulus, ν is the Poisson’s

ratio, and R and a are respectively the outer and inner radii of the cylinder.
An edge dislocation in a crystal can be visualized as a line on one side
of which an extra half-plane of atoms has been introduced, as illustrated in
Figure 2.2.6(a) for a simple cubic lattice. At a sufficient number of atomic
distances away from the dislocation line, the lattice is virtually undisturbed.
Consider, now, a path through this “good” crystal which would be closed
if the lattice were perfect. If such a path, consisting of the same number
of atom-to-atom steps in each direction, encloses a dislocation, then, as
shown in the figure, it is not closed; the vector b needed to close it is called
the Burgers vector of the dislocation, and the path defining it is called the
Burgers circuit.
b b b b b b b b b
q Atoms above
b b b b b b b b Burgers cq cq cq cq cq cq

b the slip plane
b b b b b b
circuit
b b b cAtoms below
qEc qEc qEc qEc qEc qEc
9

the slip plane
b b b b b b b 6b b
r r q c q c q c q c q c q c
E E E E E E
b
b b b b b b b b E E E E E E
CO
b b b b b b b b q E cq E cq E cq E cq E cq E c Dislocation line
E E E E E - b
b b ?b b b b b 6b q cq cq cq cq cq c
b b b b b b b b
(a) (b)
Figure 2.2.6. Dislocation in a crystal: (a) edge dislocation; (b) screw disloca-
tion.
Note that, for an edge dislocation, the Burgers vector is necessarily per-
pendicular to the dislocation line. Indeed, this can be used as the defining
property of an edge dislocation. Similarly, a screw dislocation can be de-
fined as one whose Burgers vector is parallel to the dislocation line [see
Figure 2.2.6(b)].
A dislocation in a crystal need not be a straight line. However, the Burg-
ers vector must remain constant. Thus, a dislocation can change from edge
to screw, or vice versa, if it makes a right-angle turn. It cannot, moreover,

terminate inside the crystal, but only at the surface of a crystal or at a grain
boundary. It can form a closed loop, or branch into other dislocations (at
points called nodes), subject to the conservation of the Burgers vectors:
the sum of the Burgers vectors of the dislocations meeting at a node must
vanish if each dislocation is considered to go into the node (Frank [1951]).
Dislocations and Slip

It is now universally accepted that plastic deformation in crystals results
from the movement of dislocations. As can be seen from Figure 2.2.7, in order
s s s s s s s s s sAtoms before motion
cAtoms after motion
s s s cs cs s s s s
s s s cs cs s s s s
s s s cs cs s s s s b
s s s sc s s s s s
s s s sc s s s s s
s s s sc s s s s s
s s s s s s s s s
Figure 2.2.7. Slip by means of an edge dislocation.
for an edge dislocation to move one atomic distance in the plane containing
it and its Burgers vector (the slip plane), each atom need move only a small
fraction of an atomic distance. Consequently, the stress required to move the
dislocation is only a small fraction of the theoretical shear strength discussed
in 2.2.1. An approximate value of this stress is given by the Peierls–Nabarro
stress,
2G 2πh

τPN = exp − ,
1−ν d(1 − ν)
where h and d denote, as before, the distances between adjacent planes of
atoms and between atoms in each plane, respectively. The Peierls–Nabarro
stress is clearly much smaller than the theoretical shear strength. Its value,
moreover, depends on h/d, and the smallest value is achieved when h/d is
largest, that is, for close-packed planes that are as far apart as possible;√this
result explains why such planes are the likeliest slip planes. When h = 2d,
τPN is of the order 10−5 G, consistent with the observed shear strength of
pure single crystals.
If the stress is maintained, the dislocation can move to the next posi-
tion, and the next, and so on. As the dislocation moves in its slip plane, the
portion of the plane that it leaves behind can be regarded as having expe-
rienced slip of the amount of one Burgers-vector magnitude b = |b|. When
the dislocation reaches the crystal boundary, slip will have occurred on the
entire slip plane. Suppose that the length of the slip plane is s, and that an
edge dislocation moves a distance x in the slip plane; then it contributes a
displacement bx/s, so that n dislocations moving an average distance x̄ pro-
duce a displacement u = nbx̄/s. If the average spacing between slip planes
is l, then the plastic shear strain is
u nbx̄
γp = = .
l ls
However, n/ls is just the average number of dislocation lines per unit per-
pendicular area, or, equivalently, the total length of dislocation lines of the
given family per unit crystal volume — a quantity known as the density
of dislocations, usually denoted ρ. Since only the mobile dislocations con-
tribute to plastic strain, it is their density, denoted ρm , that must appear in
the equation for the plastic strain, that is,
γ p = ρm bx̄,
and the plastic shear-strain rate is
γ̇ p = ρm bv̄,
where v̄ is the average dislocation velocity.
Forces on and Between Dislocations

A shear stress τ acting on the slip plane and in the direction of the Burg-
ers vector produces a force per unit length of dislocation that is perpendicular
to the dislocation line and equal to τ b. To prove this result, we consider an
infinitesimal dislocation segment of length dl; as this segment moves by a
distance ds, slip of an amount b occurs over an area dl ds, and therefore the
work done by the shear stress is (τ dl ds)b = (τ b) dl ds, equivalent to that
done by a force (τ b)dl, or τ b per unit length of dislocation.
Equations (2.2.3) for the strain energy per unit length of a dislocation
in an isotropic elastic continuum may be used to give an order-of-magnitude
estimate for the strain energy per unit length of a dislocation in a crystal,
namely,
W 0 = αGb2 , (2.2.4)
where α is a numerical factor between 0.5 and 1.
Two parallel edge dislocations having the same slip plane have, when
they are far apart, a combined energy equal to the sum of their individual
energies, that is, 2αGb2 per unit length, since any interaction between them
is negligible. When they are very close together, then, if they are unlike
(that is, if their Burgers vectors are equal and opposite), they will annihilate
each other and the resulting energy will be zero; thus they attract each other
in order to minimize the total energy. Like dislocations, on the other hand,
when close together are equivalent to a single dislocation of Burgers vector
2b, so that the energy per unit length is αG(2b)2 , and therefore they repel
each other in order to reduce the energy.
Frank–Read Source
The number of dislocations typically present in an unstressed, annealed
crystal is not sufficient to produce plastic strains greater than a few percent.
In order to account for the large plastic strains that are actually produced,
it is necessary for large numbers of dislocations to be created, and on a
relatively small number of slip planes, in order to account for slip bands.
The Frank–Read source is a mechanism whereby a single segment of an edge
dislocation, anchored at two interior points of its slip plane, can produce a
large number of dislocation loops. The anchor points can be point defects,
or points at which the dislocation joins other dislocations in unfavorable
planes.
If α in Equation (2.2.4) is constant along the dislocation, independently
of its orientation, then an increase ∆L in dislocation length requires an en-
ergy increment W 0 ∆L, that is, work in that amount must be done on it.
This is equivalent to assuming that a line tension T equal to W 0 is acting
along the dislocation. In order to deform an initially straight dislocation
segment into a circular arc subtending an angle 2θ, equilibrium requires a
restoring force F = 2T sin θ perpendicular to the original dislocation seg-
ment. If the length of the segment is L, then the force per unit length is
F/L and can be produced by a shear stress τ = F/bL, or
2αGb
τ= r sin θ.
L
When θ = π/2, that is, when the dislocation segment forms a semicircle, the
shear stress is maximum and equal to
Gb
τmax =
L
if α = 0.5, as it is frequently taken.
If the maximum necessary shear stress is acting on a dislocation seg-
ment pinned at two points, as in Figure 2.2.8, the semicircular form is soon
attained, whereupon the dislocation becomes unstable and expands indefi-
nitely. The expanding loop doubles back on itself, as in (c) and (d), until two
sections meet and annihilate each other, since they have the same Burgers
vector but opposite line sense, forming a closed outer loop that continues to
expand and a new dislocation segment that will repeat the process.
6 @
I
@
@
I
@
r 6 6 6 r r r r r
@
R
@
(a) (b) (c)
6 6
@
I
@ @
I
@
- -
r r r MB r

JJ
] B
S S
w /
S w /
S
(d) (e)
Figure 2.2.8. Frank–Read source (after Read [1953]).
Other mechanisms for the multiplication of dislocations that are similar

to the Frank–Read source involve screw dislocations and include cross-slip
and the Bardeen–Herring source (see, e.g., Hull and Bacon [1984]).
2.2.3. Dislocation Models of Plastic Phenomena
W. T. Read, Jr., in his classic Dislocations in Crystals (Read [1953]), of-

fered the following caution: “Little is gained by trying to explain any and
all experimental results by dislocation theory; the number of possible expla-
nations is limited only by the ingenuity, energy, and personal preference of
the theorist.”
Indeed, much theoretical work has been expended in the past half-
century in attempts to explain the phenomena of metal plasticity, discussed
in Section 2.1, by means of dislocation theory. No comprehensive theory has
been achieved, but numerous qualitative or semi-quantitative explanations
have been offered, and some of these are now generally accepted. A few are
described in what follows.
Yield Stress
If the loops generated by Frank–Read sources or similar mechanisms
could all pass out of the crystal, then an indefinite amount of slip could be
produced under constant stress. In reality, obstacles to dislocation move-
ment are present. These may be scattered obstacles such as impurity atoms
or precipitates, extended barriers such as grain boundaries, or other dislo-
cations that a moving dislocation has to intersect (“forest dislocations”). In
addition, if a dislocation is stopped at a barrier, then successive dislocations
emanating from the same Frank–Read source pile up behind it, stopped from
further movement by the repulsive forces that like dislocations exert on one
another.
The yield stress is essentially the applied shear stress necessary to pro-
vide the dislocations with enough energy to overcome the short-range forces
exerted by the obstacles as well as the long-range forces due to other dis-
locations. The mechanisms are many and complex, and therefore there is
no single dislocation theory of the yield strength but numerous theories at-
tempting to explain specific phenomena of metal plasticity. This is especially
true for alloys, in which the impurity atoms may present various kinds of
obstacles, depending on the form they take in the host lattice — for example,
whether as solutes or precipitates (for a general review, see, e.g., Nabarro
[1975]).
Yield Point
Under some conditions, solute atoms tend to segregate in the vicinity of
a dislocation at a much greater density than elsewhere in the lattice, forming
so-called Cottrell atmospheres. In order to move the dislocation, an extra
stress is required to overcome the anchoring force exerted on it by the solutes.
Once the dislocation is dislodged from the atmosphere, however, the extra
stress is no longer necessary, and the dislocation can move under a stress that
is lower than that required to initiate the motion. This is the explanation,
due to Cottrell and Bilby [1949], of the yield-point phenomenon discussed
in 2.1.2 [see Figure 2.1.1(b’)]. Strain-aging (Figure 2.1.2) is explained by
the fact that the formation of atmospheres takes place by diffusion and is
therefore a rate process. Thus if a specimen is unloaded and immediately
reloaded, not enough time will have passed for the atmospheres to form anew.
After a sufficient time, whose length decreases with increasing temperature,
the solutes segregate once more and the upper yield point returns.
Work-Hardening
As plastic deformation proceeds, dislocations multiply and eventually get
stuck. The stress field of these dislocations acts as a back stress on mobile
dislocations, whose movement accordingly becomes progressively more dif-
ficult, and an ever greater applied stress is necessary to produce additional
plastic deformation. This is the phenomenon of work-hardening.

In a first stage, when only the most favorably oriented slip systems are
active, the back stress is primarily due to interaction between dislocations
on parallel slip planes and to the pile-up mechanism. In this stage work-
hardening is usually slight, and the stage is therefore often called easy glide.
Later, as other slip systems become activated, the intersection mechanism
becomes predominant, resulting in much greater work-hardening. In a final
stage, screw dislocations may come into play.
Since the number of possible mechanisms producing forces on disloca-
tions is great, there is as yet no comprehensive theory of work-hardening that
would permit the formulation of a stress-strain relation from dislocation the-
ory. For reviews of work-hardening models, see Basinski and Basinski [1979]
or Hirsch [1975].
Yield Strength of Polycrystals
The plastic deformation of polycrystals differs from that of single crys-
tals in that, in the former, individual crystals have different orientations and
therefore, under a given applied stress, the resolved shear stress varies from
grain to grain. The critical value of this stress is therefore attained in the dif-
ferent grains at different values of the applied stress, so that the grains yield
progressively. Furthermore, the grain boundaries present strong barriers to
dislocation motion, and therefore the yield stress is in general a decreasing
function of grain size, other factors being the same; the dependence is often
found to be described by the Hall–Petch relation,
kY
σY = σY ∞ + √ ,
d
where d is the grain diameter, and σY ∞ and kY are temperature-dependent
material constants.
The stress σY ∞ , corresponding (theoretically) to infinite grain size, may
be interpreted as the yield stress when the effects of grain boundaries can be
neglected. As such it should be determinable, in principle, from the single-
crystal yield stress by a suitable averaging process, on the assumption of
random orientation of the grains. Such a determination was made by Taylor
[1938], who obtained the result that, if the stress-strain curve for a single
crystal in shear on an active slip system is given by τ = f (γ p ), then for the
polycrystal it is given by
σ = m̄f (m̄εp ),
where m̄ is the average value of the factor (cos φ cos λ)−1 in Equation (2.2.1),
a value that Taylor calculated to be about 3.1 for fcc metals.
Bauschinger Effect
A fairly simple explanation of the Bauschinger effect is due to Nabarro
[1950]. The dislocations in a pile-up are in equilibrium under the applied
Section 2.3 / Plasticity of Soils, Rocks and Concrete 103
stress σ, the internal stress σi due to various obstacles, and the back stress
σb due to the other dislocations in the pile-up; σi may be identified with
the elastic limit. When the applied stress is reduced, the dislocations back
off somewhat, with very little plastic deformation, in order to reduce the
internal stress acting on them. They can do so until they are in positions in
which the internal stress on them is −σi . When this occurs, they can move
freely backward, resulting in reverse plastic flow when the applied stress has
been reduced by 2σi .

1. For a crystal with cubic symmetry, find the Young’s modulus E in
terms of C11 , C12 , and C44 for tension (a) parallel to a cube edge, (b)
perpendicular to a cube edge and at 45◦ to the other two edges.
2. Show the close-packed planes and slip directions in a face-centered

cubic crystals.
3. Derive Equation (2.2.1).
4. For what range of R/a do Equations (2.2.3) give Equation (2.2.4) with
the values of α given in the text?
Section 2.3 Plasticity of Soils, Rocks, and Con-

crete
In recent years the term “geomaterials” has become current as one encom-
passing soils, rocks, and concrete. What these materials have in common,
and in contrast to metals, is the great sensitivity of their mechanical behavior
to pressure, resulting in very different strengths in tension and compression.
Beyond this common trait, however, the differences between soils on the one
hand and rocks and concrete on the other are striking. Soils can usually
undergo very large shearing deformations, and thus can be regarded as plas-
tic in the usual sense, although soil mechanicians usually label as “plastic”
only cohesive, claylike soils that can be easily molded without crumbling.
Rock and concrete, on the other hand, are brittle, except under high triaxial
compression. Nevertheless, unlike classically brittle solids, which fracture
shortly after the elastic limit is attained, concrete and many rocks can un-
dergo inelastic deformations that may be significantly greater than the elastic
strains, and their stress-strain curves superficially resemble those of plastic
solids.
2.3.1. Plasticity of Soil
The Nature of Soil

The essential property of soils is that they are particulate, that is, they
are composed of many small solid particles, ranging in size from less than
0.001 mm (in clays) to a few millimeters (in coarse sand and gravel). Perma-
nent shearing deformation of a soil mass occurs when particles slide over one
another. Beyond this defining feature, however, there are fundamental dif-
ferences among various types of soils, differences that are strongly reflected
in their mechanical behavior.
The voids between the particles are filled with air and water; the ratio
of the void (air and water) volume to the solid volume is known as the void
ratio of the soil. While much of the water may be in the usual liquid form
(free water ), and will evaporate on drying, some of the water is attached to
the particle surfaces in the form of adsorbed layers, and does not evaporate
unless the solid is heated to a temperature well above the boiling point of
water. A soil is called saturated if all the voids are filled with water. If both
water and air are present, the soil is called partially saturated , and if no free
water is present, the soil is called dry.
Clay was mentioned at the beginning of this chapter as a prototype of
a plastic material. Clays are fine-grained soils whose particles contain a
significant proportion of minerals known as clay minerals. The chemistry of
these minerals permits the formation of an adsorbed water film that is many
times thicker than the grain size. This film permits the grains to move past
one another, with no disintegration of the matrix, when stress is applied. It
is this property that soil mechanicians label as plasticity. Claylike soils are
also generally known as cohesive soils.
In cohesionless soils, such as gravels, sands, and silts, the movement of
grains past one another is resisted by dry friction, resulting in shear stresses
that depend strongly on the compression. Materials of this type are some-
times called frictional materials.
Soil Compressibility
If soil that is prevented from expanding laterally is loaded in compression
between layers, at least one of which is permeable to water, an irreversible
decrease in void ratio occurs, a result of the seepage of water from the voids.
The process, known as consolidation, takes time, and sometimes goes on
indefinitely, though at an ever-diminishing rate, much like creep. As a rule,
though, something very near the ultimate compression is attained in a finite
time which depends on the properties of the soil layer. A typical compression
curve is shown in Figure 2.3.1(a). The figure shows both the virgin curve and
the hysteresis loop resulting from decompression followed by recompression.
A soil that has been decompressed is called overconsolidated . The curves
Void ratio, e σ̄ (MPa)
0.50

1.2 Virgin curve

Recompression
1.0
0.25
0.8
HH
Decompression HH
e0 − e
0.6 σ̄ (MPa) ε=
0 0.25 0.50 0.1 0.2 1 + e0
(a) (b)
Figure 2.3.1. Compression curve for soil: (a) consolidation curve; (b) compres-
sive stress-strain diagram [(b) is (a) replotted].
are replotted in Figure 2.3.1(b) as a compressive stress-strain diagram. It is

seen that except for the upward convexity of the virgin curve, the diagram
resembles that of work-hardening metals.
Shearing Behavior
As in ductile metals, failure in soils occurs primarily in shear. Unlike met-
als, the shear strength of soils is, in most circumstances, strongly influenced
by the compressive normal stress acting on the shear plane and therefore by
the hydrostatic pressure. Since soils have little or no tensile strength, the
tension test cannot be applied to them. Other means are necessary in order
to determine their shear strength.
Direct Shear Test. A traditional test of the shear strength of soft clays
and of dry sands and gravels is the direct shear test or shear-box test. A
sample of soil is placed in a rectangular box whose top half can slide over the
bottom half and whose lid can move vertically, as shown in Figure 2.3.2(a).
A normal load is applied to the lid, and a shear force is applied to the top
half of the box, shearing the soil sample.
Simple Shear Test. In this test, developed by Roscoe [1953], it is the
strain that is maintained as one of simple shear [see Figure 2.3.2(b)].
The two tests just described, along with others like them, provide simple
means of estimating the shear strength. However, the stress distribution in
the sample is far from uniform, so that these tests do not actually measure
stress, and no stress-strain diagrams can result from them.
Triaxial Test. This is generally regarded as the most reliable test of
the shearing behavior of soils. As we shall see, it is used to test rock and
concrete as well. This test was discussed in 2.2.1; a normal compressive stress
σ3 (= σ2 ) is applied to the sides of a cylindrical sample by means of air or
N = normal load N
S = shear load
N ?
..............................................................................S

......................................................................................................................
................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
transducers Stress
...............................................................................................................................................................................................................................................................................................................................................................................................................................................................

...............................................................................................................................................................................................................................................................................................................................................................................................................................................................

? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
........ ...............................................................................................................................................................................................................................................................................................................................................................................................................................................................
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .....
q q q ................ ....... ...................................................................................................... . .......................................
!
......................................................................
!
q !! L .......

S ................................................
....... !! L .............................................
q q q
-
.........................................
.
q q q L L
. ..............................
........................................
q
L .............................................
LL q
.........................................
L L
q q q
L .........................................
......................................
q Soil sample
..........................................
..........................................
q q ........................................
........................................ q q q q LL
LL L .......................................
..........................................
q
LL
q
L ......................................... .....................
...............................................
.........................................................
......................................................... ..................................................
L .
.........................................................
.........................................................
..................................................................
.........................................................
.........................................................
L ............................. ....... ....... .
.
...
.
..
. ..
...
.. .
...
.
L !L .............
.......
.......
..........................................................................
...........................................................................
..........................................................................
...........................................................................
.......................................................................... ............... .
..........................................................................
...........................................................................
L !!L ... . . . .
L!!
7 ...........................................................................
..........................................................................
...........................................................................
..........................................................................
...........................................................................
..........................................................................
...........................................................................
..........................................................................
...........................................................................
..........................................................................
L!
...........................................................................
..........................................................................
...........................................................................
..........................................................................
...........................................................................
(a) ..........................................................................
...........................................................................
..........................................................................
YH
H
Hinge ...........................................................................
..........................................................................
...........................................................................
..........................................................................
...........................................................................
..........................................................................
........................................................................... H Sliding
contact
(b)
Figure 2.3.2. Shear tests: (a) direct shear test; (b) simple shear test (after
Roscoe [1953]).
water pressure, and an axial compressive stress σ1 , numerically greater than

σ3 , is applied at the ends (Figure 2.3.3). The results are commonly plotted
σ1
) Piston
?

Soil sample P Confining

fluid
....................
PP
Membrane P )

..... ..
P
-................................. σ=2
P -q
P
>

Cell P
q Pq
P
-.............................................

-
σ3
XX -....................................
Porous stone
z. . .
XX

Ol
Pore pressure Cell pressure
Figure 2.3.3. Triaxial test apparatus.
as graphs of σ1 − σ3 against the axial shortening strain ε1 , with σ3 as a

parameter. (Alternatively, the mean stress (σ1 + 2σ3 )/3 or the normal stress
on the maximum-shear plane (σ1 + σ3 )/2 may be used as a parameter.)
Note that σ1 − σ3 is a measure both of the maximum shear stress given by
Equation (1.3.11), namely, τmax = 12 |σ1 − σ3 |, and of the octahedral
√ shear
stress, given in accordance with Equation (1.3.5) as τoct = ( 2/3)|σ1 − σ3 |.
If σ3 = 0 then the test is called an unconfined compression test, used most
commonly on hard materials such as rock and concrete, but occasionally on
clay if it is performed fast enough (“quick test”). Some typical stress-strain
curves for soils are shown in Figure 2.1.1(f) (page 78).
The dependence of the shear strength of soils on the normal stress acting
on the shearing plane varies with the type and condition of the soil. It is sim-
plest in dry cohesionless soils (gravels, sands, and silts), in which resistance
to shear is essentially due to dry friction between the grains, and therefore
is governed by the Coulomb law of friction:
τ = σ tan φ, (2.3.1)
where τ and σ are respectively the shear and normal stresses on the shearing
plane, and φ is the angle of internal friction, a material property.
In wet cohesionless soils, the applied stress is the sum of the effective
stress in the grains and the neutral stress due to water pressure and possi-
bly capillary tension. If the latter stress is denoted σw (like σ, positive in
compression), then the Coulomb law is expressed by
τ = (σ − σw ) tan φ, (2.3.2)
since the water pressure provides a counterthrust on potential sliding sur-

faces, and therefore it is only the effective stress that governs frictional re-
sistance. The concept of effective stress is due to Terzaghi.
Cohesionless soils also undergo significant volume changes when sheared.
They tend to swell if they are dense, since closely packed grains must climb
over one another in the course of shearing, and shrink if they are loose, since
grains fall into the initially large voids and thus reduce the void volume. A
granular soil thus has a critical density which remains essentially constant
as shearing proceeds, and the soil is termed dense or loose, respectively, if
its density is above or below critical.
In a sample of fine sand or silt that is dense and saturated, and which
has no source of additional water, the swelling that accompanies shearing
produces surface tension on the water which acts as a negative neutral stress.
Consequently, in accord with Equation (2.3.2), such a sample has shear
strength under zero applied stress.
In clays, the stresses in the adsorbed water layers play an important
role in determining strength, and in partially saturated clays this role is
predominant. The shear strength of such clays is given approximately by
τ = c + σ tan φ, (2.3.3)
where φ is the angle of internal friction and c is a material constant called

the cohesion, representing the shear strength under zero normal stress.
The shear response of a saturated clay depends on whether it is in a
drained or undrained condition. The former condition is achieved in a slow
application of the stresses, so that the neutral stresses are not changed during
the loading and therefore play little part in determining the shear strength.
Equation (2.3.1) is consequently a good approximation to the relation be-
tween shear stress and normal stress in this condition. In the undrained
condition, on the other hand, the loading is quick and the applied stress is
carried by the neutral stress. In this condition the shear strength is indepen-
dent of the applied normal stress, and is therefore given by Equation (2.3.3)
with φ = 0; the cohesion c is then called the undrained strength and denoted
cu . Volume changes accompanying shearing are negligible in saturated clays.
The shear-strength response of undrained clays thus resembles that of met-
als. Much of soil engineering practice is based on this model, though it is
not universally accepted; see Bolton [1979], Section 5.1, for a survey of the
criticisms.
2.3.2. “Plasticity” of Rock and Concrete
Unlike soils, materials such as rock, mortar and concrete are generally not
plastic in the sense of being capable of considerable deformation before fail-
ure. Instead, in most tests they fracture through crack propagation when
fairly small strains (on the order of 1% or less) are attained, and must there-
fore be regarded as brittle. However, concrete, mortar, and many rocks
(such as marble and sandstone) are also unlike such characteristically brittle
solids as glass and cast iron, which fracture shortly after the elastic limit is
attained. Instead, they attain their ultimate strength after developing per-
manent strains that, while small in absolute terms, are significantly greater
than the elastic strains. The permanent deformation is due to several mech-
anisms, the foremost of which is the opening and closing of cracks.
Strain-Softening
Following the attainment of the ultimate strength, concrete and many
rocks exhibit strain-softening, that is, a gradual decrease in strength with
additional deformation. The nature of this decrease, however, depends on
factors associated with the testing procedure, including sample dimensions
and the stiffness of the testing machine.
The effect of machine stiffness can be described as follows. Let P denote
the load applied by the machine to the sample, and u the sample displace-
ment. In the course of a small change ∆u in the displacement, the sample
absorbs energy in the amount P ∆u. If the machine acts like an elastic spring
with stiffness k, then a change ∆P in the load implies a change P ∆P/k in
the energy stored in the machine. This change represents release of energy
if P ∆P < 0, that is, once softening takes place. The energy released by
the machine is greater than that which can be absorbed by the sample if
k < |∆P/∆u|, resulting in an unstable machine-sample system in the case
of a “soft” machine; the sample breaks violently shortly after the ultimate
strength is passed. A “stiff” machine, on the other hand, makes for a system
that is stable under displacement control. It is only with a stiff machine,
therefore, that a complete load-displacement (or stress-displacement) curve
can be traced.
It is not certain, however, whether the stress-displacement curve may le-
gitimately be converted into a stress-strain curve, such as is shown in Figure
2.1.1(d) (page 78), that reflects material properties, since specimen defor-
mation is often far from homogeneous. Experiments by Hudson, Brown and
Fairhurst [1971] show a considerable effect of both the size and the shape of
the specimens on the compressive stress-strain curve of marble, including as
a particular result the virtual disappearance of strain-softening in squat spec-
imens. Read and Hegemier [1984] conclude that no strain-softening occurs
in specimens of soil, rock and concrete that are homogeneously deformed. A
similar conclusion was reached by Kotsovos and Cheong [1984] for concrete.
It should be remarked that some rocks, such as limestone, exhibit classi-
cally brittle behavior in unconfined compression tests even with stiff testing
machines — that is, they fracture shortly after the elastic limit is reached.
The Effect of Pressure

An important feature of the triaxial behavior of concrete, mortar and
rocks (including those which are classically brittle in unconfined tests) is
that, if the confining pressure σ3 is sufficiently great, then crack propagation
is prevented, so that brittle behavior disappears altogether and is replaced
by ductility with work-hardening. Extensive tests were performed on marble
and limestone by von Kármán [1911] and by Griggs [1936]; some results are
shown in Figure 2.1.1(e). Note that the strains attained in these tests can
become quite large.
The relation between hydrostatic pressure and volumetric strain also
exhibits ductility with work-hardening; the curves resemble those of Figure
2.3.1(b). It can be said, in general, that rocks and concrete behave in a
ductile manner if all three principal stresses are compressive and close to
one another.
Dilatancy
If the transverse strain ε2 = ε3 is measured in uniaxial compression tests
of rock and concrete specimens in addition to the axial strain ε1 , then, as
discussed in 1.2.2, the volumetric strain εV equals ε1 +ε2 +ε3 . If the stress σ1
is plotted against εV (positive in compression), it is found that εV begins to
decrease from its elastic value at stresses greater than about half the ultimate
strength, reaches zero at a stress near the ultimate strength, and becomes
negative (signifying an increase in volume) in the strain-softening range (see
Figure 2.3.4, showing both a σ1 -ε1 and a σ1 -εV diagram). Similar results
are obtained in triaxial tests under low confining pressures. This volume
increase, which results from the formation and growth of cracks parallel
to the direction of the greatest compressive stress, is known as dilatancy.
This term is sometimes also applied to the swelling of dense granular soils,
σ
Volume strain
Longitudinal strain

ε
Figure 2.3.4. Compression tests on concrete or rock: stress against longitudinal
strain and volume strain.
although the mechanism causing it is unrelated.
Tensile Behavior
Uniaxial tension tests are difficult to perform on rock and concrete, and
the results of such tests vary considerably. The most reliable direct tension
tests are those in which the ends of the specimen are cemented with epoxy
resin to steel plates having the same cross-section as the specimen, with the
tensile force applied through cables in order to minimize bending effects.
The uniaxial tensile strength of rock and concrete is typically between 6 and
12% the uniaxial compressive strength. Strain-softening, associated with the
opening of cracks perpendicular to the direction of tension, is observed in
tests performed in stiff machines.
Chapter 3
Constitutive Theory
Section 3.1 Viscoplasticity

We saw in the preceding chapter that while “yielding” is the most striking
feature of plastic behavior, the existence of a well-defined yield stress is the
exception rather than the rule. It so happens, however, that mild steel,
which belongs to this exceptional class, is one of the most commonly used
of metals, and attempts at a theoretical description of its behavior preceded
those for other metals; such attempts naturally incorporated a criterion as
an essential feature of what came to be known as plasticity theory, as well
as of a later development, known as viscoplasticity theory, which takes rate
sensitivity into account.
It should be pointed out that while most workers in solid mechanics
use “viscoplasticity” in its classical meaning (see Prager [1961]), that is, to
denote the description of rate-dependent behavior with a well-defined yield
criterion, this usage is not universal. Others, following Bodner [1968], use the
term for models of highly nonlinear viscoelastic behavior, without any elastic
range, that is characteristic of metals, especially at higher temperatures.
Such models are discussed in 3.1.3. 3.1.1 is limited to models of classical
viscoplasticity. Both classes of models are subclasses of the internal-variable
models presented in Section 1.5. In 3.1.2, rate-independent plasticity, the
foundation for most of the remainder of this book, is derived as a limiting
case of classical viscoplasticity.
3.1.1. Internal-Variable Theory of Viscoplasticity
Yield Surface
As in Section 1.5, let ξ denote the array of internal variables ξ1 , ..., ξn .
If there is a continuous function f (σ, T, ξ) such that there exists a region
in the space of the stress components in which (at given values of T, ξ)
111
112 Chapter 3 / Constitutive Theory
f (σ, T, ξ) < 0, and such that the inelastic strain-rate tensor ε̇i vanishes in
that region but not outside it, then this region constitutes the aforemen-
tioned elastic range, and f (σ, T, ξ) = 0 defines the yield surface in stress
space; the orientation of the yield surface is defined in such a way that the
elastic range forms its interior. A material having such a yield function
f (·) is viscoplastic in the stricter sense. This definition, it should be noted,
does not entail the simultaneous vanishing of all the internal-variable rates
ξ˙α in the elastic region; if such were the case, strain-aging as described in
the preceding chapter would not be possible, since it requires an evolution
of the local structure while the material is stress-free. However, this pro-
viso is of significance only for processes whose time scale is of the order of
magnitude of the relaxation time for strain-aging, which for mild steel at
ordinary temperatures is of the order of hours. Thus, for a process lasting a
few minutes or less, the internal variables governing strain-aging are essen-
tially constant and their rates may be ignored. For the sake of simplicity, we
adopt a somewhat more restricted definition of viscoplasticity, according to
which all the internal-variable rates vanish in the elastic region, that is, the
functions gα (σ, T, ξ) constituting the right-hand sides of the rate equations
(1.5.1) are assumed to vanish whenever f (σ, T, ξ) ≤ 0. In particular, this
definition includes all those models (such as that of Perzyna [1971]) in which
the rates of the internal variables depend linearly on ε̇i .
In view of this definition it now becomes convenient to redefine the gα as
gα = φhα , where φ is a scalar function that embodies the rate and yielding
characteristics of the material, with the property that φ = 0 when f ≤
0 and φ > 0 when f > 0. Such a function was introduced by Perzyna
[1963] in the form γ(T )<Φ(f )>, where γ(T ) is a temperature-dependent
“viscosity coefficient” (actually an inverse viscosity, or fluidity), and the
notation <Φ(f )> is defined — somewhat misleadingly — as
(
0 for f ≤ 0
<Φ(f )> =
Φ(f ) for f > 0
(the more usual definition of the operator < · > is given below). Note that
our definition of φ is determinate only to within a multiplicative scalar; that
is, if λ is a positive continuous function of the state variables, then φ may
be replaced by φ/λ and the hα by λhα without changing the rate equations.
Hardening
The dependence of the yield function f on the internal variables ξα de-
scribes what are usually called the hardening properties of the material. The
relationship between this dependence and the behavior of the material can
be understood by considering a stress σ that is close to the yield surface
but outside it, that is, f (σ, T, ξ) > 0. In particular, let us look at a case of
Section 3.1 / Viscoplasticity 113
uniaxial stress in a specimen of a material whose static stress-strain curve is

given by the solid curve of Figure 3.1.1, which shows both rising (“harden-
ing”) and falling (“softening”) portions. If the material is viscoplastic, then
its behavior is elastic at points below the curve, and viscoelastic at points
above the curve — that is, the curve represents the yield surface.
σ
Hardening - Softening
Creep A - B - Creep
3

Static curve
Figure 3.1.1. Hardening and softening in viscoplasticity: relation to creep and

static curve.
If the stress is held constant at a value above the static curve, creep
occurs, resulting in increasing strain as shown by the dashed horizontal lines.
If the initial point is, like A, above the rising portion of the static stress-
strain curve, then the creep tends toward the static curve and is bounded,
while if it is, like B, above the falling portion, then the creep tends away
from the static curve and is unbounded. Since the points on the static stress-
strain curve are in effect those on the yield surface, we may generalize from
the uniaxial case as follows: creep toward the yield surface, characterizing
hardening, means that at constant stress and temperature, the yield function
f decreases from a positive value toward zero, that is, f˙ < 0. Similarly,
softening is characterized by f˙ > 0. But
X ∂f X ∂f
f˙ = ξ˙α = φ hα

σ =const,T =const α ∂ξα α ∂ξα
= −φH,
X ∂f
H=− hα . (3.1.1)
α ∂ξα
Thus H > 0 and H < 0 for hardening and softening materials (or hardening
and softening ranges of the same material), respectively. The limiting case
H = 0, which in particular occurs when f is independent of the ξα , describes
a perfectly plastic material.
Viscoplastic Potential
If a viscoplastic material has a flow potential in the sense of 1.5.3 (not
necessarily in the stricter sense of Rice or Moreau), then it may also be
shown to have a viscoplastic potential in the following sense. Let hij be
defined by
X ∂εiij
hij = hα .
α ∂ξα
The flow equations then are
ε̇iij = φhij . (3.1.2)
If there exists a function g(σ, T, ξ), continuously differentiable with respect

to σ wherever f (σ, T, ξ) > 0, such that
∂g
hij = ,
∂σij
then g is called a viscoplastic potential . (The relation hij = λ∂g/∂σij is not

more general, since the factor λ can be absorbed in φ.) Perzyna [1963] and
many others have assumed the existence of a viscoplastic potential identical
with the yield function f , or at least such that ∂g/σij ∝ ∂f /σij ; this identity
is of no great significance in viscoplasticity, but becomes highly important
after the transition to rate-independent plasticity.
Specific Models Based on J2 Flow Potential

In 1.5.3 a flow potential was discussed that depends on the stress only
through J2 , leading to the flow equation ε̇iij = φsij . A yield criterion having
the same stress dependence, that is, one that can be represented by the
equation p
J2 − k = 0
(where k depends on T and ξ and equals the yield stress in shear) is known as
the Mises (sometimes Huber–Mises) yield criterion. A model of viscoplas-
ticity incorporating this yield criterion and a J2 flow potential was first
proposed by Hohenemser and Prager [1932] as a generalization to three-
dimensional behavior of the Bingham model described in 2.1.3. The flow
equation is
1 k
ε̇iij = <1 − √ >sij , (3.1.3)
2η J2
where η is a temperature-dependent viscosity, and the Macauley bracket <·>
is defined by <x> = xH(x), where H(·) is the Heaviside step function:
0, x ≤ 0,

H(x) =
1, x > 0.
In other words,
0, x ≤ 0,

<x> =
x, x > 0.
The previously discussed model of Perzyna [1963] is a generalization
of the Hohenemser–Prager model in which <f > is replaced by H(f )Φ(f ),
or <Φ(f )> in Perzyna’s notation. It will be noted that as k → 0, the
Hohenemser–Prager and Perzyna models reduce to the Maxwell model of
linear viscoelasticity discussed in 1.5.1.
A generalized potential Ω, as discussed in 1.5.4, may be associated with
the Hohenemser–Prager
√ model if it takes the form Ω(σ) = <f >2 /(2η), where
f = J2 − k, and with the Perzyna model if it is Ω(σ) = H(f )Ω0 (f ).
Hardening can be included in a simple manner by letting k be a variable. If
the generalized potential is viewed as a function Ω(σ, k), then the effective
inelastic strain εi defined by (1.5.7) can easily be shown to obey the rate
equation
1 ∂Ω
ε̇i = − √ .
3 ∂k
√
It is convenient to let k = k0 + R/ 3, where k0 is the initial value of k, and
to treat Ω as a function of (σ, R). Then ε̇i = −∂Ω/∂R, and −R may be
regarded as the thermodynamic force conjugate to the internal variable εi .
A more sophisticated model developed by Chaboche [1977] uses as in-
ternal variables εi and a strain-like symmetric second-rank tensor α. The
thermodynamic forces conjugate to these variables are the stress-like vari-
ables −R and −ρ, respectively, and the yield surface is assumed to be given
by
R
q
f (σ, ρ, R) = J¯2 − √ − k0 = 0,
3
where
1
J¯2 = (sij − ρ0ij )(sij − ρ0ij ),
2
ρ0 being the deviator of ρ. The yield surface is thus again of the Mises type,
but capable not only of expansion (as measured by R) but also of transla-
tion (as shown by ρ0 , which locates the center of the elastic region). The
hardening described by the expansion of the yield surface is called isotropic,
while that described by the translation is called kinematic. The significance
of the terms is discussed in Section 3.2.
If a generalized potential is again assumed in the Perzyna form, Ω(σ, R, ρ)
= H(f )Ω0 (f ), then ε̇iij = ∂Ω/∂σij and ε̇i = −∂Ω/∂R as before, the flow
equations being
∂Ω sij − ρ0ij
ε̇iij = = H(f )Ω00 (f ) p .
∂σij 2 J¯2
In addition, α̇ = ε̇i , so that the kinematic-hardening variable α, though
it must be treated as a distinct variable, coincides with the inelastic strain.
Chaboche, however, assumes the generalized potential in the form Ω(σ, R, ρ)

= H(f )Ω0 (f )+Ωr (ρ), where the second term represents recovery (see 2.1.2).
Moreover, Chaboche abandons the generalized normality hypothesis for α
by introducing an additional term representing a concept called fading strain
memory, due to Il’iushin [1954], the better to describe the Bauschinger effect.
The rate equation for α is therefore taken as
∂Ω ∂Ωr
α̇ij = − − F (εi )ε̇i ρij = ε̇iij − F (εi )ε̇i ρij − ,
∂ρij ∂ρij
where F (εi ) is a function to be specified, along with Ω0 (f ), Ωr (ρ), and the
free-energy density ψ(T, ε, εi , α), from which R and ρ can be derived in
accordance with Equation (1.5.4): R = ρ∂ψ/∂εi , ρij = ρ∂ψ/∂αij .
3.1.2. Transition to Rate-Independent Plasticity

Aside from the previously discussed limit of the Hohenemser–Prager model
as the yield stress goes to zero, another limiting case is of great interest,
namely, as the viscosity η goes to zero. Obviously,
√ if s 6= 0 then the inelastic
strain rate would become infinite, unless √ J2 simultaneously tends to k, in
which case the quantity (1/η)<1 − k/ J2 > becomes indeterminate but may
remain finite and positive.
Supposing for simplicity that k in Equation (3.1.3) is constant, for a
given input of stress we can solve this equation for εi as a function of time,
and the dependence on time is through the variable t/η. In other words,
decreasing the viscosity is equivalent to slowing down the process of inelastic
deformation, and the limit of zero viscosity is equivalent to the limit of
“infinitely slow” processes. Thus a slow process can take place if J2 is slightly
larger than k 2 . We can also see this result by forming the scalar product
ε̇iij ε̇iij from Equation (3.1.3), from which we obtain
q
ε̇i 6= 0,
p
J2 = k + η 2ε̇iij ε̇iij ,
an equation that is sometimes interpreted as a rate-dependent yield criterion.
Let us return to the more general model of viscoplasticity considered
above, and particularly one in which φ increases with f . The rate equations
(3.1.2) indicate that, in the same sense as in the Hohenemser–Prager model,
the rate of a process in which inelastic deformation takes place increases with
distance from the yield surface. If such a process is very slow , then it takes
place very near but just outside the yield surface, so that φ is very small. In
the limit as f → 0+ we can eliminate φ (and thus no longer need to concern
ourselves with the actual rate at which the process takes place) as follows:
if f remains equal to zero (or a very small positive constant), then
∂f X ∂f
f˙ = σ̇ij + φhα = 0.
∂σij α ∂ξα
We define
◦ ∂f
f= σ̇ij (3.1.4)
∂σij
and assume H > 0 (i.e., hardening), with H as defined by Equation (3.1.1);
◦
then the condition f˙ = f − φH = 0 is possible together with φ > 0 only if
◦
f > 0; this last condition is called loading. Thus we have the result
1 ◦
φ= <f>,
H
and therefore
1 ◦
ξ˙α = <f>hα . (3.1.5)
H
Note that both sides of Equation (3.1.5) are derivatives with respect to time,
so that a change in the time scale does not affect the equation. Such an equa-
tion is called rate-independent. If it is assumed that this equation describes
material behavior over a sufficiently wide range of loading rates, then the
behavior is called rate-independent plasticity, also called inviscid plasticity
(since it corresponds to the zero-viscosity limit of the Hohenemser–Prager
model), or just plain plasticity. Rate-independent plasticity constitutes the
principal topic of the remainder of this book. The inelastic strain occurring
in rate-independent plasticity is usually denoted εp rather than εi , and is
called the plastic strain. The flow equation for the plastic strain may be
written as
1 ◦
ε̇pij = <f>hij . (3.1.6)
H
For purposes of computation, however, it is sometimes advantageous to
remain within the framework of viscoplasticity without making the full tran-
sition, even when the problem to be solved is regarded as rate-independent.
In other words, a fictitious viscoplastic material of very low viscosity is “as-
sociated” with a given rate-independent plastic material, with rate equations
given, for example (Nguyen and Bui [1974]), by
<f >
ξ˙α = hα , (3.1.7)
η
with the viscosity η taken as constant. Computations are then performed
under time-independent loads and boundary conditions until all strain rates
vanish. It was shown by Zienkiewicz and Cormeau [1974], among others,
that the results are equivalent to those of rate-independent plasticity.
Combined Viscoplasticity and Rate-Independent Plasticity

At extremely high rates of deformation or loading, the internal variables
do not have enough time to change and consequently the deformation can
be only elastic. However, the various rate processes responsible for plastic
deformation, corresponding to the generation of dislocations and the many

different kinds of obstacles that dislocations must overcome, may have very
different characteristic times. This means that not only do metals differ
greatly among one another in their rate-sensitivity, but different mecha-
nisms in the same metal may respond with very different speeds. Thus,
those mechanisms whose characteristic times are very short compared with
a typical loading time produce what appears to be instantaneous inelastic
deformation, while the others produce rate-dependent deformation as dis-
cussed so far in this section. If both phenomena occur in a metal over a
certain range of loading times, then the total inelastic strain εi may be
decomposed as
εi = εvp + εp , (3.1.8)
where εvp is the viscoplastic strain equivalent to that governed by Equation
(3.1.2), and εp is the apparently rate-independent plastic strain, governed
by Equation (3.1.6). It is important to note that the yield functions f and
flow tensors hij are, in general, different for the two inelastic strain tensors.
In particular, the viscoplastic yield surface is always assumed to be inside
the rate-independent plastic (or “dynamic”) yield surface.
3.1.3. Viscoplasticity Without a Yield Surface

As we have seen, in both classical viscoplasticity and rate-independent plas-
ticity the yield surface is a central ingredient; in the latter it is indispensable.
The significance of the yield surface has, however, repeatedly been ques-
tioned. Consider the following remarks by Bell [1973]: “Among the many
matters pertaining to the plastic deformation of crystalline solids, yield sur-
faces and failure criteria early became subjects of overemphasis... Indeed
most of the outstanding 19th century experimentists doubted that such a
phenomenon as an elastic limit, let alone a yield surface, existed... well over
a half-century of experiment, and the study of restricted plasticity theories
for the ‘ideal solid,’ have not disposed of most of the original questions.”
“Unified” Viscoplasticity Models

According to Bodner [1968], “yielding is not a separate and independent
criterion but is a consequence of a general constitutive law of the material be-
havior.” Since the 1970s several constitutive models for the rate-dependent
inelastic behavior of metals have been formulated without a formal hypoth-
esis of a yield surface, but with the feature that at sufficiently low rates the
resulting stress-strain curves may resemble those of materials with fairly well
defined yield stresses. In fact, with yield based on the offset definition (see
2.1.1), these models can predict yield surfaces in accordance with Bodner’s
dictum, particularly if offset strains of the order of 10−6 to 10−5 are used,
in contrast to the conventional 10−3 to 10−2 .
In addition to describing the behavior traditionally called plasticity, in

both monotonic and cyclic loading, these models also aim to describe creep,
especially at higher temperatures, without a decomposition such as (3.1.8).
They have consequently come to be known as “unified” viscoplasticity mod-
els, and are particularly useful for the description of bodies undergoing signif-
icant temperature changes — for example, spacecraft. Perhaps the simplest
such model is due to Bodner and Partom [1972, 1975], in which the flow
equations are given by Equation (3.1.2) with hij = sij and φ a function of
J2 and (in order to describe hardening) of the inelastic work Wi defined by
Equation (1.5.6) as the only internal variable. The rate equation is obviously
Ẇi = 2J2 φ(Wi , J2 ).

√
The hardening in this case is purely isotropic, since 3J2 is the value of the
effective stress necessary to maintain a given inelastic work rate Ẇi .1
More sophisticated “unified” viscoplasticity models, that describe many
features of the behavior of metals at elevated temperatures, have been devel-
oped since 1975 by, among others, Miller [1976], Hart [1976], Krieg, Swearen-
gen, and Jones [1978], Walker [1981], and Krieg, Swearengen, and Rohde
[1987] (see reviews by Chan, Bodner, Walker, and Lindholm [1984], Krempl
[1987], and Bammann and Krieg [1987]). The essential internal variables in
these models are the equilibrium stress tensor ρ, and the scalar drag stress or
friction stress σD ;2 the terminology is loosely related to that of dislocation
theory, and is an example of “physical” nomenclature for phenomenological
internal variables.
In the “unified” models the stress-like variables σD and ρ are used di-
rectly as internal variables, rather than as conjugate thermodynamic forces.
The equilibrium stress ρ, like its counterpart in the Chaboche model, de-
scribes kinematic hardening. Some writers, following Kochendörfer [1938],
relate it to the back stress due to stuck dislocations (see 2.2.3), and conse-
quently the equilibrium stress is also termed back stress; see Krempl [1987]
for a discussion (the relationship between constitutive theory and crystal
behavior has also been discussed by Kocks [1987]). For isotropic behavior
ρ is assumed as purely deviatoric, and the rate equation for inelastic strain
takes the form
3 φ(Γ/σD )
ε̇iij = ėiij = (sij − ρij ), (3.1.9)
2 Γ
1
√
Or, equivalently, a given effective inelastic strain rate, since in this model Ẇi = 3J2 ε̇i
is an increasing function of Wi (or of εi ).
2
A model developed by Krempl and coworkers, known as viscoplasticity based on
overstress, dispenses with drag stress as a variable (see Yao and Krempl [1985] and
Krempl, McMahon, and Yao [1986]). In another model called viscoplasticity based
on total strain (Cernocky and Krempl [1979]), the equilibrium stress is not an internal
variable but a function of total strain; this model is therefore a nonlinear version of the
“standard solid” model of linear viscoelasticity (see 1.5.1).
where Γ = 3J¯2 , with J¯2 as defined in 3.1.1, and φ is a function (whose

p
values have the dimensions of inverse time) that increases rapidly with its
argument. The evolution of the equivalent inelastic strain is given by
ε̇i = φ(Γ/σD ), (3.1.10)
and, in uniaxial stress,

ε̇i = φ(|σ − ρ|/σD ).
Typical forms of φ(x) are Axn , A(ex − 1), and A[sinh(xm )]n , where A, m,
and n are constants, n in particular being a large exponent. For an extension
to initially anisotropic behavior, see, for example, Helling and Miller [1987].
A variety of forms has been proposed for the rate equations for ρ and
σD ; a typical set is due to Walker [1981]:
ρ̇ij = a1 ε̇iij − [a2 ε̇i + a3 (2ρkl ρkl /3)(m−1)/2 ]ρij ,
σ̇D = [a4 − a5 (σD − σD0 )]ε̇i − a6 (σD − σD0 )p ,
where ε̇iij and ε̇i are substituted from (3.1.9)–(3.1.10), and a1 , ...., a6 , m, p
and σD0 are constants.
Endochronic Theory
A different “theory of viscoplasticity without a yield surface” is the en-
dochronic theory of Valanis [1971], originally formulated by him (Valanis
[1971]) for “application to the mechanical behavior of metals,” though its
range of application has recently been extended to other materials, such
as concrete (Bažant [1978]). The basic concept in the theory is that of an
intrinsic time (hence the name) that is related to the deformation history of
the material point, the relation itself being a material property. An intrinsic
time measure ζ is defined, for example, by
dζ 2 = Aijkl dεiij dεikl + B 2 dt2 ,
where the tensor A and scalar B may depend on temperature. (In the
original theory of Valanis [1971], the total strain ε rather than the inelastic
strain εi appeared in the definition.) A model in which B = 0 describes
rate-independent behavior and thus defines the endochronic theory of
plasticity.
An intrinsic time scale is next defined as z(ζ), a monotonically increas-
ing function, and the behavior of the material is assumed to be governed by
constitutive relations having the same structure as those of linear viscoelas-
ticity, as described in 1.5.2, but with z replacing the real time t. As in linear
viscoelasticity, the internal variables can be eliminated, and the stress can
be related to the strain history by means of a pseudo-relaxation function.

The uniaxial relation is
Z z dε 0
σ= R(z − z 0 ) dz , (3.1.11)
0 dz 0
while the multiaxial relation describing isotropic behavior is
Z z
dεkk dεij

0
σij = R1 (z − z )δij 0
+ 2R2 (z − z 0 ) 0 dz 0 .
0 dz dz
With a pseudo-relaxation function analogous to that of the “standard

solid,” that is, R(z) = E1 + E2 e−αz , and with z(ζ) given by
1
z= ln(1 + βζ),
β
where α and β are positive constants, Valanis [1971] was able to fit many
experimental data on repetitive uniaxial loading-unloading cycles and on
coupling between tension and shear.
More recently, Valanis [1980] showed that Equation (3.1.11) can be re-
placed by
dεi dεi
Z z
σ = σ0 + ρ(z − z 0 ) 0 dz 0 . (3.1.12)
dz 0 dz
For rate-independent uniaxial behavior, dζ = |dεi | with no loss in generality.
If the last integral is called α, then the stress must satisfy
|σ − α| = σ0 h(z), (3.1.13)
where h(z) = dζ/dz. Equation (3.1.12) can be used to construct stress-

strain curves showing hardening depending both on the effective inelastic
strain [through h(z)] and on the strain path (through α). The equation has
a natural extension to multiaxial stress states, which for isotropic materials
is
(sij − αij )(sij − αij ) = [s0 h(z)]2 .
This equation represents a yield surface capable of both expansion and trans-
lation in stress space, thus exhibiting both isotropic and kinematic harden-
ing. An endochronic model unifying viscoplasticity and plasticity was pre-
sented by Watanabe and Atluri [1986].

1. Suppose that the yield function has the form f (σ, T, ξ) = F (σ) −
k(T, κ), where κ is the hardening variable defined by either (1.5.6)
or (1.5.7), and the flow equations are assumed as in the form (3.1.2).
What is the “hardening modulus” H, defined by Equation (3.1.1)?
2. If the only stress components are σ12 = σ21 = τ , with τ > 0, write the
equation for the shear rate γ̇ = 2ε̇12 given by the Hohenemser–Prager
model (3.1.3). Discuss the special case k = 0.
3. Generalize the Hohenemser–Prager model to include isotropic and kine-

matic hardening. Compare with the Chaboche model.
4. Find the flow equation for a viscoplastic solid with a rate-dependent

yield criterion given by
p 1
J2 = k + η(2ε̇ij ε̇ij ) 2m .
5. Construct a simple model for combined viscoplasticity and plasticity,

with a perfectly plastic Mises yield criterion and associated flow rule
in both.
6. Derive (3.1.12) from (3.1.11).
Section 3.2 Rate-Independent Plasticity

3.2.1. Flow Rule and Work-Hardening
Flow Rule
In keeping with the formulation of rate-independent plasticity as the
limit of classical viscoplasticity for infinitely slow processes, we henceforth
consider all processes to be “infinitely” slow (compared with the material
relaxation time τ ), and correspondingly regard the material as “inviscid
plastic,” “rate-independent plastic,” or simply plastic. The inelastic strain
εi will from now on be called the plastic strain and denoted εp instead of εi .
The flow equations (3.1.6) may be written as
ε̇pij = λ̇hij , (3.2.1)
where 
 1 ◦
λ̇ = <f >, f = 0, (3.2.2)
H
0, f < 0,
with H as defined by Equation (3.1.1). The rate equations (3.1.5) analo-
gously become
ξ˙α = λ̇hα .
If ∂f /∂ξα ≡ 0, then, as mentioned before, the material is called perfectly

◦ ◦
plastic. In this case H = 0, but f = f˙, and therefore the condition f > 0
Section 3.2 / Rate-Independent Plasticity 123
is impossible. Plastic deformation then occurs only if (∂f /∂σij )σ̇ij = 0

(neutral loading), and the definition (3.2.2) of λ̇ cannot be used. Instead, λ̇
is an indeterminate positive quantity when f = 0 and (∂f /∂σij )σ̇ij = 0, and
is zero otherwise.
In either case, λ̇ and f can easily be seen to obey the Kuhn–Tucker
conditions of optimization theory:
λ̇f = 0, λ̇ ≥ 0, f ≤ 0.
The specification of the tensor function h in Equation (3.2.1), at least to

within a multiplicative scalar, is known as the flow rule, and if there exists
a function g (analogous to a viscoplastic potential) such that hij = ∂g/∂σij ,
then such a function is called a plastic potential .
Deformation Theory
The plasticity theory in which the plastic strain is governed by rate equa-
tions such as (3.2.1) is known as the incremental or flow theory of plasticity.
A deformation or total-strain theory was proposed by Hencky [1924]. In
this theory the plastic strain tensor itself is assumed to be determined by
the stress tensor, provided that the yield criterion is met. Elastic unloading
from and reloading to the yield surface are in principle provided for, al-
though a contradiction is seen as soon as one considers reloading to a stress
other than the one from which unloading took place, but located on the
same yield surface; clearly, no plastic deformation could have occurred dur-
ing the unloading-reloading process, yet the theory requires different values
of the plastic strain at the two stress states. The deformation theory, which
is mathematically much simpler than the flow theory, gives results that co-
incide with those of the latter only under highly restricted circumstances.
An obvious example is the uniaxial case, provided that no reverse plastic
deformation occurs; the equivalence is implicit in the use of relations such
as (2.1.2).
A more general case is that of a material element subject to proportional
or radial loading, that is, loading in which the ratios among the stress com-
ponents remain constant, provided the yield criterion and flow rule are suffi-
ciently simple (for example, the Mises yield criterion and the flow rule with
hij = sij ). A rough definition of “nearly proportional” loading, for which
the deformation theory gives satisfactory results, is discussed by Rabotnov
[1969]. It was shown by Kachanov [1954] (see also Kachanov [1971]) that
the stress states derived from the two theories converge if the deformation
develops in a definite direction.
Another example of a range of validity of the deformation theory, dis-
cussed in separate developments by Budiansky [1959] and by Kliushnikov
[1959], concerns a material whose yield surface has a singular point or cor-
ner, with the stress point remaining at the corner in the course of loading.
For a simplified discussion, see Chakrabarty [1987], pp. 91–94.

The deformation theory has recently been the subject of far-reaching
mathematical developments (Temam [1985]). It has also been found to give
better results than the incremental theory in the study of the plastic buckling
of elements under multiaxial stress, as is shown in Section 5.3.
Work-Hardening
The hardening criterion H > 0, and the corresponding criteria H = 0
for perfect plasticity and H < 0 for softening, were formulated in 3.1.1
for viscoplastic materials on the basis of rate-dependent behavior at states
outside the yield surface. An alternative derivation can be given entirely in
the context of rate-independent plasticity.
For given ξ, f (σ, ξ) = 0 is the equation describing the yield surface in
stress space. If f (σ, ξ) = 0 and f˙|σ=const < 0 (i.e. H > 0) at a given time
t, then at a slightly later time t + ∆t we have f (σ, ξ + ξ̇∆t) < 0; the yield
surface is seen to have moved so that σ is now inside it. In other words,
H > 0 implies that, at least locally, the yield surface is expanding in stress
space. The expansion of the yield surface is equivalent, in uniaxial stress, to
a rising stress-strain curve (see Figure 3.2.1).

σ-space

r
X f (σ, ξ + ξ̇∆t) = 0 if H > 0
y
X
PP f (σ, ξ) = 0
i
Q
k
Q
Q f (σ, ξ + ξ̇∆t) = 0 if H < 0
Figure 3.2.1. Hardening and softening in rate-independent plasticity: motion

of yield surface in stress space
Conversely, a contracting yield surface denotes work-softening, and a

stationary yield surface perfect plasticity. The description of work-softening
materials is best achieved in strain space rather than stress space, and is
discussed later. For now, we treat work-hardening materials only, with per-
fectly plastic materials as a limiting case.
In the simplest models of plasticity the internal variables are taken as (1)
the plastic strain components εpij themselves, and (2) the hardening variable
κ, defined by either Equation (1.5.6) or (1.5.7) (in rate-independent plastic-
ity, εp is written in place of εi ). When the yield function is taken to have
the form
f (σ, εp , κ) = F (σ − ρ(εp )) − k(κ),
both isotropic and kinematic hardening, as discussed in Section 3.1, can be
described; the hardening is isotropic if ρ ≡ 0 and dk/dκ > 0, and purely
kinematic if dk/dκ ≡ 0 and ρ 6= 0. The condition dk/dκ ≡ 0 and ρ ≡ 0 rep-
resents perfect plasticity. The simplest model of kinematic hardening — that
of Melan [1938] — has ρ(εp ) = cεp , with c a constant. More sophisticated
hardening models are discussed in Section 3.3.
Drucker’s Postulate
A more restricted definition of work-hardening was formulated by Drucker
[1950, 1951] by generalizing the characteristics of uniaxial stress-strain curves.
With a single stress component σ, the conjugate plastic strain rate ε̇p clearly
satisfies [see Figure 3.2.2(a)]
 ≥ 0, hardening material,

σ̇ ε̇p = 0, perfectly plastic material,

≤ 0, softening material.

The inequalities are unchanged if the stress and plastic-strain rates are mul-
tiplied by the infinitesimal time increment dt, so that they hold equally
well for dσ dεp . This product has the dimensions of work per unit volume,
and was given by Drucker the following interpretation: if a unit volume
of an elastic-plastic specimen under uniaxial stress is initially at stress σ
and plastic strain εp , and if an “external agency” (one that is independent
of whatever has produced the current loads) slowly applies an incremental
load resulting in a stress increment dσ (which causes the elastic and plastic
strain increments dεe and dεp , respectively) and subsequently slowly removes
it, then dσ dε = dσ (dεe + dεp ) is the work1 performed by the external agency
in the course of incremental loading, and dσ dεp is the work performed in
the course of the cycle consisting of the application and removal of the in-
cremental stress. (Note that for dεp 6= 0, σ must be the current yield stress.)
Since dσ dεe is always positive, and for a work-hardening material dσ dεp ≥
0, it follows that for such a material dσdε > 0. Drucker accordingly defines a
work-hardening (or “stable”) plastic material as one in which the work done
during incremental loading is positive, and the work done in the loading-
unloading cycle is nonnegative; this definition is generally known in the
literature as Drucker’s postulate (see also Drucker [1959]).
Having defined hardening in terms of work, Drucker naturally extends
the definition to general three-dimensional states of stress and strain, such
that
dσij dεij > 0 and dσij dεpij ≥ 0,
1
Actually it is twice the work.
σ
σ-space
ε̇p > 0
-
σ̇ = 0 σr
σ̇ > 0
s
σ̇ < 0 z rσ + dσ
=
7
ε
r
σ∗
σ̇ > 0 σ̇ < 0 Elastic region
k σ̇ =0
ε̇p < 0
(a) (b)
Figure 3.2.2. Drucker’s postulate: (a) illustration in the uniaxial stress-strain

plane; (b) illustration in stress space.
the equality holding only if dεp = 0. For perfectly plastic (“neutrally stable”)
materials Drucker’s inequalities are dσij dεij ≥ 0 and dσij dεpij = 0. It can
be seen that the inequality
σ̇ij ε̇pij ≥ 0, (3.2.3)
sometimes known simply as Drucker’s inequality, is valid for both work-
hardening and perfectly plastic materials.
Because it uses the concept of work, Drucker’s postulate is often referred
to as a quasi-thermodynamic postulate, although it is quite independent of
the basic laws of thermodynamics. Drucker’s inequality (3.2.3) may also be
given an interpretation that is free of any considerations of incremental work:
the left-hand side represents the scalar product σ̇ · ε̇p , and the inequality
therefore expresses the hypothesis that the plastic strain rate cannot oppose
the stress rate.
We should note, lastly, that Drucker’s definition of work-hardening is in
a sense circular. The definition assumes an external agency that is capable of
applying arbitrary stress increments. But as can readily be seen from stress-
strain diagrams, this assumption is not valid for softening or perfectly plastic
materials; for example, in a tension test no increase in stress is possible. In
other words, such materials are unstable under stress control . On the other
hand, they are stable under strain control (or displacement control1 ), since
arbitrary strain increments that do not violate internal constraints may, in
principle, be applied. This fact points to the applicability of strain-space
plasticity, to be discussed later, to a wider class of materials.
Drucker’s statement of his work-hardening postulate is broader than
summarized above, in that the additional stress produced by the external
1
Stability under strain control and displacement control are equivalent when deforma-
tions are infinitesimal, but not when they are finite.
agency need not be a small increment. In particular, the initial stress, say
σ ∗ , may be inside the elastic region, or at a point on the yield surface far
away from σ, and the process followed by the external agency may consist
of elastic loading to a stress σ on the current yield surface, a small stress
increment dσ producing an incremental plastic strain dε, and finally, elastic
unloading back to σ ∗ ; the path is illustrated in Figure 3.2.2(b). With dσ
neglected alongside σ − σ ∗ , the work per unit volume done by the external
∗ ) dεp . Drucker’s postulate, consequently, implies
agency is (σij − σij ij
∗
(σij − σij ) ε̇pij ≥ 0. (3.2.4)
3.2.2. Maximum-Plastic-Dissipation Postulate and Normal-

ity
Maximum-Plastic-Dissipation Postulate
Inequality (3.2.4) is, as we have just seen, a necessary condition for
Drucker’s postulate, but it is not a sufficient one. In other words, its validity
is not limited to materials that are work-hardening in Drucker’s sense. Its
significance may best be understood when we consider its uniaxial counter-
part,
(σ − σ ∗ )ε̇p ≥ 0.
As is seen in Figure 3.2.3, the inequality expresses the property that the
σ
rσ
ε̇p ≥ 0

r
σ∗

ε

rσ∗

ε̇p ≤0 σ r
Figure 3.2.3. Maximum-plastic-dissipation postulate: illustration in the uniax-

ial stress-strain plane.
plastic strain rate is positive (negative) only if the current stress σ is not
less than (not greater than) any stress σ ∗ in the current elastic range —
in other words, if σ equals the current tensile (compressive) yield stress.
Clearly, work-softening and perfectly plastic materials have this property

as well. Inequality (3.2.4) thus constitutes a postulate in its own right,
called the postulate of maximum plastic dissipation. It was proposed
independently by Mises [1928], Taylor [1947] and Hill [1948a]; it was derived
from considerations of crystal plasticity by Bishop and Hill [1951], and is
shown later to follow also from Il’iushin’s postulate of plasticity in strain
space.
Consequences of Maximum-Plastic-Dissipation Postulate

Inequality (3.2.4) has consequences of the highest importance in plastic-
ity theory. To examine them, we represent symmetric second-rank tensors
as vectors in a six-dimensional space, as in 1.3.5, but using boldface rather
than underline notation, and using the dot-product notation for the scalar
product. Our inequality may thus be written as
(σ − σ ∗ ) · ε̇p ≥ 0.
We suppose at first that the yield surface is everywhere smooth, so that

a well-defined tangent hyperplane and normal direction exist at every point.
It is clear from the two-dimensional representation in Figure 3.2.4(a) that if
(3.2.4) is to be valid for all σ ∗ to the inward side of the tangent to the yield
surface at σ, then ε̇p must be directed along the outward normal there; this
consequence is known as the normality rule. But as can be seen in Figure
3.2.4(b), if there are any σ ∗ lying to the outward side of the tangent, the
inequality is violated. In other words, the entire elastic region must lie to
one side of the tangent. As a result, the yield surface is convex.
S
Q
S Q ε̇p ε̇p
ε̇p
rXX S >
Q

X

r rX r
XXXS Q
3
:
:σ
z
X
QX
y
σ Q XXr σ∗

σ C
6
r S
σ−σ ∗ r
:
S Q
Q

r
S Q
Q
σ∗ S
(a) (b) (c)
Figure 3.2.4. Properties of yield surface with associated flow rule: (a) normal-
ity; (b) convexity; (c) corner.
Let us define Dp (ε̇p ; ξ) by

∗ p
Dp (ε̇p ; ξ) = max
∗
σij ε̇ij ,
σ
the maximum being taken over all σ ∗ such that f (σ ∗ , ξ) ≤ 0. It follows

from (3.2.4) that
σij ε̇pij = Dp (ε̇p ; ξ). (3.2.5)
To make it clear that Dp (ε̇p ; ξ) depends only on ε̇p and ξ and not on σ, we
note that, if the yield surface is strictly convex at σ (whether this point is
regular or singular), then this is the only stress that corresponds to a given
normal direction in stress space and hence to a given ε̇p . If the yield surface
has a flat portion, then all points on this portion have the same normal,
that is, different stresses correspond to the same ε̇p , but the scalar product
σ · ε̇p = σij ε̇pij is the same for all of them. Dp (ε̇p ; ξ) will be called simply
the plastic dissipation. Inequality (3.2.4) may now be rewritten as
∗ p
Dp (ε̇p ; ξ) ≥ σij ε̇ij , (3.2.6)
giving explicit meaning to the name “principle of maximum plastic dissipa-

tion.”
Normality
The normality rule is now discussed in more detail. At any point of the
yield surface f (σ, ξ) = 0 where the surface is smooth, the outward normal
vector is proportional to the gradient of f (in stress space), and therefore,
reverting to indicial notation, we may express the normality rule as
∂f
hij = , (3.2.7)
∂σij
where hij is the tensor function appearing in the flow equation (3.2.1). Equa-
tion (3.2.7) expresses the result that the function f defining the yield surface
is itself a plastic potential, and therefore the normality rule is also called a
flow rule that is associated with the yield criterion, or, briefly, an asso-
ciated (sometimes associative) flow rule. A flow rule derivable from a
plastic potential g that is distinct from f (more precisely, such that ∂g/∂σij
is not proportional to ∂f /∂σij ) is accordingly called a nonassociated flow
rule. In the French literature, materials obeying an associated flow rule are
usually called standard materials, and this term will often be used here.
We are now in a position to say that Drucker’s postulate applies to
standard work-hardening (or, in the limit, perfectly plastic) materials. The
frequently expressed notion that Drucker’s postulate is required for the con-
vexity of the yield surface and for the normality rule is clearly erroneous, as
is the idea that work-hardening materials are necessarily standard.
If the yield surface is not everywhere smooth but has one or more singular
points (corners) at which the normal direction is not unique, then at such a
point ε̇p must lie in the cone formed by the normal vectors meeting there [see
Figure 3.2.4(c)]. The argument leading to the convexity of the yield surface
is not affected by this generalization. As will be seen, Equation (3.2.7) can

still be formally used in this case, provided that the partial derivatives are
properly interpreted. In a rigorous treatment, the concept of gradient must
be replaced by that of subgradient, due to Moreau [1963]; its application in
plasticity theory was formulated by Moreau [1976].
Another treatment of singular yield surfaces was proposed by Koiter
[1953a], who supposed the yield surface to be made of a number — say n —
of smooth surfaces, each defined by an equation fk (σ, ξ) = 0 (k = 1, ..., n);
the elastic region is the intersection of the regions defined by fk (σ, ξ) < 0,
and σ is on the yield surface if at least one of the fk vanishes there, it being
a singular point only if two or more of the fk vanish. Equation (3.2.7) is
replaced by
X ∂fk
hij = αk ,
k
∂σ ij
the summation being over those k for which fk (σ, ξ) = 0, and the αk are
nonnegative numbers that may, with no loss of generality, be constrained so
P
that k αk = 1.
3.2.3. Strain-Space Plasticity

As we noted above, it is only in work-hardening materials, which are stable
under stress control, that we may consider processes with arbitrary stress in-
crements, and therefore it is only for such materials — with perfect plasticity
as a limiting case — that a theory in which stress is an independent variable
may be expected to work. No such limitation applies to theories using strain
as an independent variable. Surprisingly, such theories were not proposed
until the 1960s, beginning with the pioneering work of Il’iushin [1961], fol-
lowed by papers by Pipkin and Rivlin [1965], Owen [1968], Lubliner [1974],
Nguyen and Bui [1974], Naghdi and Trapp [1975], and others.
To see that strain-space yield surfaces have the same character whether
the material is work-hardening or work-softening, let us consider one whose
stress-strain diagram in tension/compression or in shear is as shown in Figure
3.2.5(a); such a material exemplifies Melan’s linear kinematic hardening if
E 0 > 0, and is perfectly plastic if E 0 = 0. The stress σ is on the yield surface
if σ = E 0 ε ± (1 − E 0 /E)σE , and therefore the condition on the strain ε is
E(ε − εp ) = E 0 ε ± (1 − E 0 /E)σE , or, equivalently, ε = [E/(E − E 0 )]εp ± εE ,
where εE = σE /E [see Figure 3.2.5(b)]. A yield surface in ε-space is thus
given by the pair of points corresponding to a given value of εp , and the ε-εp
diagram has a “stable” form (i.e., a positive slope) for all E 0 < E, even if
negative.
Yield Criterion and Flow Rule in Strain Space

To formulate the three-dimensional yield criterion in strain space, let C
σ ε
0

E E/(E−E 0 )
1 1
1
?
σE
E εE

1 ε εp

) −εE

−σE
)
0
1

E

(a) (b)
Figure 3.2.5. Material with linear hardening: (a) stress-strain diagram; (b) ε-εp
diagram.
denote the elastic modulus tensor, so that the σ-ε-εp relation may be written
def
σ = C · (ε − εp ) [i.e., σij = Cijkl (εkl − εpkl )]. If fˆ(ε, ξ) = f (C · (ε − εp ), ξ),
then the strain-space yield criterion is just
fˆ(ε, ξ) = 0.
Since
∂ fˆ

∂f
= Cijkl ,
∂εij ∂σkl σ=C·(ε−εp )
the same logic that led to (3.2.1)–(3.2.2) produces the flow equations

 1 ∂f
hij <Cijkl ε̇kl >, f = 0,
ε̇pij = (3.2.8)
L ∂σij
0, f < 0,
where
X ∂ fˆ ∂f
L=− hα = H + Cijkl hkl . (3.2.9)
α ∂ξα ∂σij
The normality rule (3.2.7), when translated into the strain-space formula-
tion, takes the form
ˆ
−1 ∂ f
hij = Cijkl .
∂εkl
Note that L may very well be, and normally may be assumed to be, posi-
tive even when H is zero or negative, that is, for perfectly plastic or work-
softening materials. It is thus not necessary to distinguish between these
material types, the only restriction being L > 0. This condition describes
stability under strain control in the same sense that the work-hardening cri-
terion H > 0 describes stability under stress control; it will here be called
kinematic stability.
The flow equation given by (3.2.8)–(3.2.9), when combined with the re-
lation σ̇ij = Cijkl (ε̇kl − ε̇pkl ), yields

 1 ∂f
Cijmn hmn <Cpqkl ε̇kl >, f = 0,
σ̇ij = Cijkl ε̇kl − (3.2.10)
L ∂σpq
0, f < 0.
This is an explicit expression for σ̇ in terms of ε̇, which may be regarded
−1
as an inversion of ε̇ij = Cijkl σ̇kl + ε̇pij with ε̇p given by Equation (3.2.1).
In this sense the result, which was first derived by Hill [1958] [for standard
materials, i.e. with h given by (3.2.7)], is not necessarily based on strain-
space plasticity.
Work-hardening (Plastic) Modulus
It is easy to show that when ε̇p 6= 0,
fij σ̇ij
H = (Cijkl fij hkl ) ,
Cijkl fij ε̇kl
where fij = ∂f /σij . Thus H may be related to the so-called work-hardening
modulus or plastic modulus dσ/dεp obtained in a simple tension test. If
the material has (a) elastic isotropy, (b) plastic incompressibility, and (c)
sufficient plastic symmetry so that σij = σδi1 δj1 implies that ε̇p22 = ε̇p33 =
− 12 ε̇p11 and ε̇pij = 0 for i 6= j, then (with εp = εp11 )
dσ
H = h11 f11 .
dεp
Il’iushin’s Postulate
It can also be shown that the normality rule follows from a “postulate
of plasticity” in strain space first proposed by Il’iushin [1961], namely, that
in any cycle that is closed in strain space,
I
σij dεij ≥ 0, (3.2.11)
where the equality holds only if the process is elastic; we show this by proving
that (3.2.11) implies the maximum-plastic-dissipation postulate (3.2.4).
Consider a state (ε1 , ξ 1 ) with ε1 on the yield surface, and any strain ε∗
on or inside both the current yield surface and the subsequent yield surface
obtained after a brief plastic process of duration ∆t from (ε1 , ξ 1 ) to (ε1 +
ε̇∆t, ξ 1 + ξ̇∆t), that is,
fˆ(ε1 , ξ 1 ) = 0, fˆ(ε1 + ε̇∆t, ξ 1 + ξ̇∆t) = 0,
and
fˆ(ε∗ , ξ 1 ) ≤ 0, fˆ(ε∗ , ξ 1 + ξ̇∆t) ≤ 0.
In the cycle
1 2 3
(ε∗ , ξ 1 ) → (ε1 , ξ 1 ) → (ε1 + ε̇∆t, ξ 1 + ξ̇∆t) → (ε∗ , ξ 1 + ξ̇∆t),
segments 1 and 3 are elastic, so that, if the process is isothermal,

ρψ̇ in 1 and 3
σij ε̇ij =
ρψ̇ + D in 2,
where ψ(ε, ξ) is X
the free energy per unit mass at the given temperature and
D(ε, ξ, ξ̇) = −ρ (∂ψ/∂ξ α ) ξ˙α is the dissipation per unit volume. Now
α
I I Z
σij dεij = ρ dψ + D dt;
2
but
I
.
ρ dψ = ρψ(ε∗ , ξ 1 + ξ̇∆t) − ρψ(ε∗ , ξ 1 ) = −D(ε∗ , ξ 1 , ξ̇)∆t,
.
and size − 1int2 D dt = D(ε1 , ξ 1 , ξ̇)∆t, the approximations being to within
o(∆t). It follows that
D(ε, ξ, ξ̇) ≥ D(ε∗ , ξ, ξ̇) (3.2.12)
(the superscripts 1 can now be dropped) if (ε, ξ) is a state with ε on the

current yield surface with corresponding ξ̇, and ε∗ is any strain on or inside
the yield surface.
With the free-energy density decomposed as in Equation (1.5.5) (and εp
written in place of εi ), D = σij ε̇pij − ρψ̇ i and therefore, if σ ∗ = C · (ε∗ − εp )
is any stress on or inside the yield surface, inequality (3.2.12) is equivalent
to (3.2.4), which is thus seen to be a consequence of Il’iushin’s postulate.
It remains to be investigated whether the converse holds. Consider an ar-
bitrary process that is closed in strain space, going from (ε∗ , ξ 1 ) to (ε∗ , ξ 2 ),
with ξ 2 not necessarily infinitesimally close to ξ 1 . At any state of the process,
σij ε̇ij = ρψ̇ + D(ε, ξ, ξ̇),
with D = 0 whenever the process is instantaneously elastic, and therefore

I I
σij dεij = [D(ε, ξ, ξ̇) − D(ε∗ , ξ, ξ̇)] dt.
According to the principle of maximum plastic dissipation, the integrand is

nonnegative whenever ε∗ is on or inside the strain-space yield surface at the
current value of ξ, and therefore Il’iushin’s postulate is satisfied for processes

in which the original yield surface is inside all subsequent yield surfaces. The
last condition is satisfied in materials with isotropic hardening, but not in
general. Consequently Il’iushin’s postulate is a stronger (i.e. less general)
hypothesis than the principle of maximum plastic dissipation.
Nguyen–Bui Inequality
On the other hand, an inequality first explicitly stated by Nguyen and
Bui [1974] may be shown to be weaker than the maximum-plastic-dissipation
principle. It is readily seen that this principle, as expressed in the form
(3.2.12), is equivalent to
Cijkl (εkl − ε∗kl )ε̇pij ≥ 0 (3.2.13)
for any strain ε∗ that is on or inside the current yield surface in strain space.
Suppose, in particular, that ε∗ is close to ε and is given by ε∗ = ε ± ε̇ dt,
with dt > 0 and ε̇ the strain-rate tensor in a possible process. With the plus
sign chosen, the process goes from ε to ε∗ and is necessarily elastic, so that
ε̇p = 0 and therefore Inequality (3.2.13) is satisfied as an equality. Thus
ε̇ 6= 0 only if the minus sign is taken, and therefore (3.2.13) takes the local
form
Cijkl ε̇pij ε̇kl ≥ 0, (3.2.14)
or the equivalent form given by Nguyen and Bui,
σ̇ij ε̇pij ≥ Cijkl ε̇pij ε̇pkl .
Inequality (3.2.14), like Drucker’s inequality (3.2.3), may be interpreted as a

stability postulate, this time in strain space: if we take Cijkl aij bkl as defining
a scalar product between two tensors a and b in strain-increment space, then
(3.2.14) expresses the notion that the plastic strain rate cannot oppose the
total strain rate. Inequality (3.2.14) is by itself sufficient for the associated
flow rule to follow, and consequently describes standard kinematically stable
materials.

1. A work-hardening plastic solid is assumed to obey√the Mises yield
criterion with isotropic hardening, that is, f (σ, ξ) = J2 − k(εp ), and
the flow rule hij = sij . Show that
√
p 3sij <skl ṡkl >
ε̇ij = .
4k 2 k 0 (εp )
2. Show that the solid described in Exercise 1 obeys Drucker’s inequality

(3.2.3) if and only if k 0 (εp ) > 0.
Section 3.3 / Yield Criteria, Flow Rules and Hardening Rules 135
3. If σ and σ ∗ are stresses such that J2 = k 2 and J2∗ ≤ k 2 , show that

(sij − s∗ij )sij ≥ 0, and hence that the solid of Exercise 1 obeys the
maximum-plastic-dissipation postulate (3.2.4) independently of k 0 (εp ),
that is, whether the solid hardens or softens.
4. For a work-hardening solid with the yield criterion of Exercise 1, but

with the nonassociated flow rule hij = sij + tij , where sij tij = 0, show
that for some σ̇ Drucker’s inequality (3.2.3) is violated.
5. For the standard isotropically hardening Mises solid of Exercise 1, show

that
sij <skl ε̇kl >
ε̇pij = .
k 0 (εp )

2
2k 1 + √
3G
6. Show that the standard Mises solid obeys the Nguyen–Bui inequality
(3.2.14) whether it hardens or softens.
Section 3.3 Yield Criteria, Flow Rules and Hard-

ening Rules
3.3.1. Introduction
The yield function f in stress space may be written with no loss of gener-
ality in terms of the stress deviator and the first invariant of stress, that is,
f (σ, ξ) = f (s, I1 , ξ), where I1 = σkk = δij σij , so that ∂I1 /∂σij = δij . Since
1 1

skl = σkl − I1 δkl = δik δjl − δij δkl σij ,
3 3
it follows that ∂skl /∂σij = δik δjl − 13 δij δkl . Consequently,
∂f ∂f ∂skl ∂f ∂I1 ∂f
= + = (f ij − 13 δij f kk ) + δij ,
∂σij ∂skl ∂σij ∂I1 ∂σij ∂I1
where f ij = ∂f /∂sij . Accordingly, in a standard material plastic volume

change (“dilatancy”) occurs if and only if the yield criterion depends on I1 ,
i.e. on the mean stress, and, conversely, plastic incompressibility obtains
if and only if the yield criterion depends on s but not on I1 . If the yield
criterion of a plastically incompressible material is significantly affected by
mean stress, then the material is necessarily nonstandard.
Isotropic Yield Criteria

If the yield criterion is initially isotropic, then the dependence of f on σ

must be through the stress invariants I1 , I2 , and I3 , or, equivalently, on the
principal stresses σI (I = 1, 2, 3), provided this dependence is symmetric,
that is, invariant under any change of the index I. Similarly, the dependence
of f on s must be through the stress-deviator invariants J2 and J3 ; the
equivalent dependence on the principal stresses may be exhibited in the
so-called principal stress-deviator plane or π-plane, namely, the plane in
σ1 σ2 σ3 -space given by σ1 + σ2 + σ3 = 0, shown in Figure 3.3.1.
J
J σ3 σ2
6 ...........................................
. . . .
...........................................................................
J
........... .... .... .... .... .
J ..................................................................................................................................................................................
PH ... ... ... ... ... ... ...
..........................................................................................................................
........................J
Y H . . ................... .. .. .. .. ..
.......................................
...........................................................................................................................
. . . .
J . .
.s..O
. .
σ1 ...........................J
H
.............................H ..........J
..... ......................................
.
. .. ... .... .... ...........................
................................................................................................................
..........................J ....... ... ... ...............
.......................................................J ...........0.............................J
........
...............................P
. ......................
.........................................................................................................
.
π-plane ......................................................................... J
....................................................
.........................
J σ1 =σ2 =σ3
J
J
Figure 3.3.1. π-plane.
Indeed, if a point P (σ1 , σ2 , σ3 ) in σ1 σ2 σ3 -space (the Haigh–Westergaard

→ →
space) is represented by the vector OP , then its projection OP 0 onto the
π-plane is the vector whose components are the principal stress deviators
s1 , s2 , s3 . The magnitude of this projection
√ — that is, the distance from P
to the axis σ1 = σ2 = σ3 — is just 2J2 . A yield surface that is independent
of I1 has, in this space, the form of a cylinder perpendicular to the π-plane,
and therefore may be specified by a single curve in this plane. A yield surface
that depends on I1 may be described by a family of curves in the π-plane,
each corresponding to a different value of I1 and forming the intersection of
the yield surface in σ1 σ2 σ3 -space with a plane I1 = constant, that is, a plane
parallel to the π-plane.
A curve√in the π-plane can also be described in terms of the polar co-
ordinates ( 2J2 , θ), where the polar angle θ may defined as that measured
from the projection of the σ1 -axis toward the projection of the σ2 -axis, and
can be shown to be given by
√
3(σ2 − σ3 ) s2 − s3
tan θ = = √ .
2σ1 − σ2 − σ3 3s1
Using some trigonometric identities and the fact that s1 + s2 + s3 = 0, it is

also possible to define θ in terms of the deviatoric stress invariants J2 and
J3 : √
3 3J3
cos 3θ = 3/2
.
2J2
A point with θ = 0 corresponds to σ1 > σ2 = σ3 ; the locus of such points
on the yield surface is said to represent one of the three tensile meridians of
the surface. A point with θ = π/3 corresponds to σ1 = σ2 > σ3 , and lies on
a compressive meridian.
3.3.2. Yield Criteria Independent of the Mean Stress

Since the concept of plasticity was first applied to metals, in which the
influence of mean stress on yielding is generally negligible (Bridgman [1923,
1950]), the oldest and most commonly used yield criteria are those that
are independent of I1 . Such criteria have an alternative two-dimensional
representation: since their dependence on the principal stresses must be
through the differences σ1 − σ2 , σ1 − σ3 and σ2 − σ3 , and since σ1 − σ2 =
(σ1 −σ3 )−(σ2 −σ3 ), the yield criterion can be plotted in a plane with σ1 −σ3
and σ2 − σ3 as coordinate axes.
Tresca Criterion
The Tresca yield criterion is historically the oldest; it embodies the
assumption that plastic deformation occurs when the maximum shear stress
over all planes attains a critical value, namely, the value of the current yield
stress in shear, denoted k(ξ). Because of Equation (1.3.11), this criterion
may be represented by the yield function
1
f (σ, ξ) = max (|σ1 − σ2 |, |σ2 − σ3 |, |σ3 − σ1 |) − k(ξ), (3.3.1)
2
or, equivalently,
1
f (σ, ξ) = (|σ1 − σ2 | + |σ2 − σ3 | + |σ3 − σ1 |) − k(ξ). (3.3.2)
4
The projection of the Tresca yield surface in the π-plane is a regular
hexagon, shown in Figure 3.3.2(a), whose vertices lie on the projections of
the positive and negative σ1 , σ2 and σ3 -axes, while in the (σ1 − σ3 )(σ2 − σ3 )-
plane it takes the form of the irregular hexagon shown in Figure 3.3.2(b).
Of course, the forms (3.3.1) and (3.3.2) for the Tresca yield function are
not unique. The form
f (σ) = [(σ1 − σ2 )2 − 4k 2 ][(σ2 − σ3 )2 − 4k 2 ][(σ1 − σ3 )2 − 4k 2 ]
(with the dependence on ξ not indicated) has the advantage of being analytic
and, moreover, expressible in terms of the principal stress-deviator invariants
J2 and J3 :
f (σ) = 4J2 3 − 27J32 − 36k 2 J2 2 + 96k 4 J2 − 64k 6 .
σ2 −σ3
Tresca
σ3
2k ?
Mises
Tresca

Mises

√
0 3k
σ1 0

σ1 −σ3
k 2k

σ1 +σ2 +σ3
=0

σ2
(a) (b)
Figure 3.3.2. Projections of Tresca and Mises yield surfaces: (a) π-plane; (b)
σ1 −σ3 -σ2 −σ3 plane.
Tresca Criterion: Associated Flow Rule

Although the Tresca yield surface is singular, we can nonetheless derive
its associated flow rule by means of a formal application of Equation (3.2.7)
to the second form, Equation (3.3.2). We write
d
|x| = sgn x, (3.3.3)
dx
where
+1, x > 0
sgn x = 2H(x) − 1 =
−1, x < 0
defines the signum function for x 6= 0. It is conventional to define sgn 0 =
0, but in the present context it is more convenient if sgn x does not have
a unique value at 0 but can have any value between −1 and 1. Strictly
speaking, then, it is not a function in the usual sense but a set-valued function
or multifunction, a concept with considerable use in convex analysis.1 In this
way we obtain
1
ε̇p1 = λ̇[sgn (σ1 − σ2 ) + sgn (σ1 − σ3 )],
4
where, in accordance with Equations (3.2.1)–(3.2.2), for work-hardening ma-
◦
terials λ̇ = <f >/H, with
X ∂k
H= hα ,
α ∂ξα
1
In fact, our use of Equation (3.3.3) comes rather close to the subdifferential calculus
of Moreau.
while for the perfectly plastic material λ̇ is indeterminate. Similar expres-

sions are obtained for ε̇p2 and ε̇p3 . Thus, if the principal stresses are all distinct,
and ordered such that σ1 > σ2 > σ3 , then ε̇p1 = 12 λ̇, ε̇p2 = 0, ε̇p3 = − 12 λ̇. If, on
the other hand, σ1 = σ2 > σ3 , then ε̇p1 = 41 λ̇(1+β), ε̇p2 = 14 λ̇(1−β), ε̇p3 = − 12 λ̇,
where β is any real number between −1 and 1. Analogous expressions can
be obtained for all other combinations of principal stresses.
The Tresca flow rule can also be obtained by the method due to Koiter,
discussed in 3.2.2.
It can be seen that for every combination of principal stresses, |ε̇p1 | +
|ε̇p2 | + |ε̇p3 | = λ̇, and therefore the plastic dissipation is given by
Dp (ε̇p ; ξ) = λ̇k(ξ) = k(ξ)(|ε̇p1 | + |ε̇p2 | + |ε̇p3 |).
If it is desired to use an effective plastic strain εp as an internal variable in

conjunction with the Tresca criterion and its associated flow rule, then the
definition
1
Z
p
ε = (|ε̇p1 | + |ε̇p2 | + |ε̇p3 |) dt
2
is more appropriate than (1.5.7). In Rfact, if it is assumed that k is a function
of εp as thus defined, then Wp = 2 k dεp , so that a one-to-one correspon-
dence can be established between Wp and εp .
Lévy Flow Rule and Mises Yield Criterion
In the nineteenth century Saint-Venant and others used the Tresca yield
criterion together with the (nonassociated) flow rule derived from the J2
potential (see 1.5.4) whose general form was first proposed (for total rather
than plastic strain) by Lévy, namely,
ε̇pij = λ̇sij ,
with λ̇ defined as above. As seen in Section 3.1 in connection with viscoplas-

ticity, and as first pointed out by Mises [1913], the yield criterion with which
this flow rule is associated is the Mises criterion, represented by the yield
function p
f (σ, ξ) = J2 − k(ξ),
where k(ξ) is again the yield stress in shear at the current values of ξ. In view
of the relation (1.3.5) between J2 and the octahedral shear stress, the Mises
criterion is also known as the maximum-octahedral-shear-stress criterion, and
as a result of Equation (1.4.17), which shows the complementary energy of
an isotropic, linearly elastic material to be uncoupled into volumetric and
distortional parts, it is also called the maximum-distortional-energy criterion.
An alternative — and analytic — form of the Mises yield function (with
the dependence on ξ not shown explicitly) is
f (σ) = J2 − k 2 .
Expressing J2 in terms of the principal stresses (see Section 1.3), we may

formulate the Mises yield criterion in the form
(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ1 − σ3 )2 = 6k 2
or
σ12 + σ22 + σ32 − σ2 σ3 − σ3 σ1 − σ1 σ2 = 3k 2 .
The form √taken by the Mises yield surface in the π-plane is that of a circle
of radius 2k, and in the (σ1 − σ3 )(σ2 − σ3 )-plane that of an ellipse. Both
forms are shown, along with those for the Tresca criterion, in Figure 3.3.2
(page 138).
The plastic dissipation for the Mises criterion and associated flow rule is
given by
q
Dp (ε̇p ; ξ) = σij ε̇pij = λ̇sij sij = 2J2 ε̇pij ε̇pij
p
q
= k(ξ) 2ε̇pij ε̇pij .
If k is a function of εp as defined by (1.5.7), then Wp is in one-to-one corre-

spondence with εp .
As mentioned above, Lévy and Mises formulated the flow rule bearing
their name for the total, rather than merely the plastic, strain rate; in this
form it is valid as an approximation for problems in which elastic strains
are vanishingly small, or, equivalently, for materials whose elastic moduli
are infinite — the so-called rigid-plastic materials (see Section 3.4). The
generalization allowing for nonvanishing elastic strains is due to Prandtl
[1924] and Reuss [1930]; expressed in terms of total strain rate, with isotropic
linear elasticity, the result is known as the Prandtl–Reuss equations:
1
ε̇kk = σ̇kk ,
3K (3.3.4)
1
ėij = ṡij + λ̇sij .
2G
Some generalizations of the Mises yield function have been proposed so

that dependence on J3 is included. A typical form is
!α
J2
f (σ) = 1 − c 33 J2 − k 2 .
J2
The exponent α has variously been taken as 13 and 1, k is as usual the yield
stress in simple shear, and c is a parameter that is to be determined so as
to optimize the fit with experimental data.
Anisotropic Yield Criteria
Anisotropy in yielding may be of two types: initial anisotropy and

induced anisotropy. The former exists in materials that are structurally
anisotropic, even before any plastic deformation has taken place; the latter
appears, even in initially isotropic materials, as a result of work-hardening
(Section 2.1). An example of an initially anisotropic yield criterion is Schmid’s
law (Section 2.2), according to which yielding in single crystals occurs when
the shear stress on certain preferred planes (the slip planes) reaches a criti-
cal value; in the special case when every plane is a slip plane, Schmid’s law
reduces to the Tresca criterion.
An anisotropic generalization of the Mises criterion is due to Hill [1950];
it replaces J2 with a general quadratic function of σ and therefore has the
form
1
2
Aijkl σij σkl = k 2 ,
where A is a fourth-rank tensor which has the same symmetries as the elas-
ticity tensors (Aijkl = Ajikl = Aklij ). If the yield criterion is independent of
mean stress, then A also obeys Aijkk = 0, so that it has at most fifteen inde-
pendent components (like a symmetric 5 × 5 matrix); the isotropic (Mises)
case corresponds to Aijkl = δik δjl − 31 δij δkl . A special case considered by
Hill [1948b] refers to a material with three mutually perpendicular planes of
symmetry; if a Cartesian basis is chosen so that the coordinate planes are
parallel to the planes of symmetry, then in this basis the components of A
coupling normal stresses with shear stresses (e.g. A1112 , A1123 , etc. — nine
independent components altogether) are zero, and A is given by
Aijkl σij σkl = A(σ22 − σ33 )2 + B(σ11 − σ33 )2 + C(σ11 − σ22 )2

2 2 2
+ 4Dσ23 + 4Eσ13 + 4F σ12 ,
where A, . . . , F are constants; clearly A1111 = B + C, A1122 = −C, A1133 =

−B, A1212 = F , and so on.
3.3.3. Yield Criteria Dependent on the Mean Stress

A yield criterion depending on the mean stress becomes necessary when it is
desired to apply plasticity theory to soils, rocks, and concrete, as discussed in
Section 2.3. One such criterion has its origin in the Mohr theory of rup-
ture, according to which failure (rupture) occurs on a plane in a body if the
shear stress and normal stress on that plane achieve a critical combination.
Since the strength properties of an isotropic material are unchanged when
the direction of the shear stress is reversed, the critical combination may be
expressed by the functional equation τ = ±g(σ). This equation represents
a pair of curves (each being the other’s reflection through the σ-axis) in the
Mohr plane, and a state of stress, as determined by the three Mohr’s circles,
is safe if all three circles lie between the curves, while it is a critical state if
one of the three is tangent to the curves. These curves are thus the envelopes
of the Mohr’s circles representing failure and are therefore called the Mohr
failure (rupture) envelopes. The point (σ, τ ) is a point of tangency — say
the upper one — if it obeys (1) the equation τ = g(σ), (2) the equation of
the Mohr’s circle [centered at (σm , 0) and of radius τm ], and (3) the tangency
condition. If σ and τ are eliminated between these three equations, there
remains an equation in terms of σm and τm , which constitutes the failure
criterion.1 Concretely, if a point in principal-stress space is located in the
sextant σ1 > σ2 > σ3 , then σm = 12 (σ1 + σ3 ) and τm = 12 |σ1 − σ3 |; the equa-
tion consequently represents a cylindrical surface parallel to the σ2 -axis, and
the failure surface is formed by six such surfaces.
Mohr–Coulomb Criterion
The equations can be reduced explicitly if the Mohr envelopes are straight
lines, that is, if
g(σ) = c − µσ.
This is just Equation (2.3.3), with the sign of σ changed to the usual con-
vention whereby it is positive in tension; c is the cohesion, and µ = tan φ is
the coefficient of internal friction in the sense of the Coulomb model of fric-
tion. The resulting criterion is consequently known as the Mohr–Coulomb
criterion. It is convenient to represent the Mohr’s circle parametrically:
σ = σm + τm cos 2α, τ = τm sin 2α,
where α is the angle between the failure plane and the axis of the least tensile
(greatest compressive) stress. The tangency condition is then µ = cot 2α, so
that α = 14 π − 12 φ, sin 2α = cos φ, and cos 2α = sin φ. The equation in terms
of σm and τm becomes
τm + σm sin φ = c cos φ,
from which it is seen that the failure stress in simple shear is c cos φ. (Need-
less to say, when φ = 0 the Mohr–Coulomb criterion reduces to that of
Tresca.) In terms of the principal stresses the criterion takes the form
max[|σi − σj | + (σi + σj ) sin φ] = 2c cos φ,

i6=j
so that the yield stresses in tension and compression are respectively 2c cos φ/(1+
sin φ) = 2c tan α and 2c cos φ/(1 − sin φ) = 2c cot α. The associated plastic
dissipation was shown by Drucker [1953] to be
Dp (ε̇p ; ξ) = c cot φ(ε̇p1 + ε̇p2 + ε̇p3 ).

1
It is pointed out by Hill [1950] that tangency between the Mohr envelope and Mohr’s
circle does not necessarily occur at real (σ, τ ), and that it is the failure criterion and not
the envelope that is fundamental.
The failure surfaces in σ1 σ2 σ3 -space are obviously planes that intersect

to form a hexagonal pyramid; the plane in the sextant σ1 > σ2 > σ3 , for
example, is described by
σ1 − σ3 + (σ1 + σ3 ) sin φ = 2c cos φ.
A form valid in all six sextants is
σmax − σmin + (σmax + σmin ) sin φ = 2c cos φ,
where σmax and σmin denote respectively the (algebraically) largest and
smallest principal stresses.
The last equation may be rewritten as
1 2
σmax − σmin + [(σmax − σint ) − (σint − σmin )] sin φ = 2c cos φ − I1 sin φ,
3 3
where σint denotes the intermediate principal stress. The left-hand side,
being an isotropic function of the stress deviator, is therefore a function of
J2 and J3 . The Mohr–Coulomb criterion is therefore seen to be a special
case of the family of criteria based on Coulomb friction and described by
equations of the form
F̄ (J2 , J3 ) = c − λI1 ,
where c and λ are constants.
Drucker–Prager Criterion
Another yield criterion of this family, combining Coulomb friction with
the Mises yield criterion, was proposed by Drucker and Prager [1952] and has
become known as the Drucker–Prager criterion. With the Mises criterion
interpreted in terms of the octahedral shear stress, it may be postulated that
p
yielding occurs on the octahedral planes when τoct = 23 k − 13 µI1 , so that, in
view of Equation √ (1.3.5), the
√ criterion may be represented by the yield func-
tion f (s, I1 ) = J2 + µI1 / 6 − k. The yield surface in Haigh–Westergaard
space is a right
√ circular cone about the mean-stress axis, subtending the
−1
angle tan ( 3µ). The yield √ stresses in√ simple shear,
√ tension,
√and compres-
sion are respectively k, 3k/(1 + µ/ 2) and 3k/(1 − µ/ 2); note √ that
for this criterion to be physically meaningful, µ must be less than 2. The
associated plastic dissipation is
q
k 2ε̇pij ε̇pij
Dp (ε̇p ; ξ) = p .
1 + µ2
Projections of the yield surfaces corresponding to the Mohr–Coulomb

and Drucker–Prager criteria onto a plane parallel to the π-plane (i.e. one
with σ1 + σ2 + σ3 = constant) are shown in Figure 3.3.3(a).
σ2
σ3

σ1
Mohr–Coulomb
@
@+ Drucker–Prager
@

σ1 σ3 =0

@
Y
H H Drucker–Prager
σ1 +σ2 +σ3 Mohr–Coulomb
= const.

σ2
(a) (b)
Figure 3.3.3. Mohr–Coulomb and Drucker–Prager criteria: (a) plane parallel

to π-plane; (b) plane stress.
Mises–Schleicher Criterion
A yield criterion that takes into account the difference between the yield
strengths in tension and compression was discussed by Mises [1926] and
Schleicher [1926]. If σT and σC denote, respectively, the tensile and com-
pressive yield stresses, then the criterion may be expressed in the form
3J2 + (σC − σT )I1 − σT σC = 0.
The associated plastic dissipation is

p p
σC − σT ėij ėij σC σT
p
Dp (ε̇ ; ξ) = p + ε̇p .
2 ε̇kk 3(σC − σT ) kk
3.3.4. Yield Criteria Under Special States of Stress or De-

formation
Equibiaxial Stress
A state of stress is called equibiaxial if two of the principal stresses are
equal, as, for example, in the triaxial soil test described in Section 2.3. If
σ2 = σ3 , then q q
p
J2 = 13 |σ1 − σ3 | = 2 13 τm ,
so that the Mises and Tresca yield criteria are formally equivalent, as are the
Mohr–Coulomb criterionMohr–Coulomb and Drucker–Prager criteria.
Plane Stress
The criteria for plane stress are obtained simply by setting σ3 = 0. Thus
the Tresca and Mises criteria are just as they appear in Figure 3.3.2(b) (page
138). In a state of plane stress in the x1 x2 -plane with σ22 = 0 (i.e., a state
of stress that may be represented as a superposition of simple tension or
compression and shear), it can further be shown that both the Mises and
the Tresca yield criteria can be expressed in the form
2 2
σ τ

+ = 1, (3.3.5)
σY τY
where σ = σ11 , τ = σ12 , and σY and τY are respectively the yield stresses
√ in
simple tension or compression and in shear, that is, τY = k, and σY = 3k
or 2k, depending on the criterion. The Mohr–Coulomb criterionMohr–
Coulomb and Drucker–Prager criteria in plane stress are shown in Figure
3.3.3(b).
In general, an isotropic yield criterion with σ3 = 0 may be written (with
dependence on ξ not indicated explicitly) as
f0 (σ1 , σ2 ) = 0,
or equivalently, upon transforming the independent variables, as
f1 [n, 21 (σ1 − σ2 )] = 0,
def
where n = 12 (σ1 + σ2 ). Because of isotropy, the dependence of f1 on its
second argument must be through the absolute value r = 12 |σ1 − σ2 |. The
preceding equation can then be solved for r as a function of n:
r = h(n). (3.3.6)
In particular, h(n) takes the form

q
h(n) = k 2 − 31 n2
for the Mises criterion and

|n| < k

k,
h(n) =
2k − |n|, |n| > k
for the Tresca criterion.
Plane Strain
In plane strain, as defined, for example, by ε̇3 = 0, the situation is
more complicated, since the plane-strain condition requires ε̇p3 = −ε̇e3 , in
turn involving the stress rates. If, however, the elastic strain rates may
be equated to zero (the condition for this is discussed later), then we have
ε̇3 = ε̇p3 = 0. Assuming a plastic potential g(σ1 , σ2 , σ3 ), we obtain the

equation
∂
g(σ1 , σ2 , σ3 ) = 0,
∂σ3
which when combined with the yield criterion in terms of the principal
stresses, permits the elimination of σ3 and hence the formulation of a yield
criterion in terms of σ1 and σ2 , leading once more to Equation (3.3.6).
Consider, for example, the Mises criterion with its associated flow rule
ε̇p = λ̇s, which requires s3 = 32 [σ3 − 21 (σ1 + σ2 )] = 0, that is, σ3 = 12 (σ1 + σ2 ).
Substituting this into the Mises yield criterion yields 34 (σ1 − σ2 )2 = 3k 2 , or
|σ1 − σ2 | = 2k. According to the Tresca flow rule, on the other hand, for
ε̇p3 to be zero, σ3 must be the intermediate principal stress, that is, either
σ1 > σ3 > σ2 or σ1 < σ3 < σ2 , so that
max (|σ1 − σ3 |, |σ2 − σ3 |, |σ1 − σ2 |) = |σ1 − σ2 | = 2k.
Consequently the two criteria coincide, and may be expressed by Equation

(3.3.6) with h(n) = k.
Consider next the Mohr–Coulomb criterionMohr–Coulomb criterion,
with a nonassociated flow rule that is governed by a plastic potential having
the same form as the yield function, that is,
g(σ) = σmax − σmin + (σmax + σmin ) sin ψ,
where ψ is known as the angle of dilatancy, since ψ = 0 describes a plasti-

cally incompressible solid; the special case ψ = φ represents the associated
flow rule. The plane-strain condition ε̇p3 = 0 again requires that σ3 be the
intermediate principal stress. The criterion therefore may be described by
Equation (3.3.6) with
h(n) = c cos φ − n sin φ.
3.3.5. Hardening Rules

A specification of the dependence of the yield criterion on the internal vari-
ables, along with the rate equations for these variables, is called a hardening
rule. In this subsection we first review in more detail the significance of the
two models of hardening — isotropic and kinematic — previously discussed
for viscoplasticity in Section 3.1. Afterwards we look at some more general
hardening rules.
Isotropic Hardening
The yield functions that we have studied so far in this section are all
reducible to the form
f (σ, ξ) = F (σ) − k(ξ).
Since it is only the yield stress that is affected by the internal variables, no
generality is lost if it is assumed to depend on only one internal variable, say
ξ1 , and this is invariably identified with the hardening variable κ, defined
as either the plastic work Wp by Equation (1.5.6) or as the effective plastic
strain εp by Equation (1.5.7). The function
p
h1 corresponding to ξ1 [see
Equation (3.1.5)] is given by σij hij or 23 hij hij , respectively, for each of the
two definitions of κ, so that the work-hardening modulus H is
k 0 (Wq
(
p )σij hij ,
H=
k 0 (εp ) 2
3
hij hij .
As was pointed out in Section 3.2, work-hardening in rate-independent

plasticity corresponds to a local expansion of the yield surface. The present
behavior model (which, as we said in Section 3.1, is called isotropic hard-
ening) represents a global expansion, with no change in shape. Thus for a
given yield criterion and flow rule, hardening behavior in any process can be
predicted from the knowledge of the function k(κ), and this function may,
in principle, be determined from a single test (such as a tension test).
The most attractive feature of the isotropic hardening model, which was
introduced by Odqvist [1933], is its simplicity. However, its usefulness in
approximating real behavior is limited. In uniaxial stressing it predicts that
when a certain yield stress σ has been attained as a result of work-hardening,
the yield stress encountered on stress reversal is just −σ, a result clearly at
odds with the Bauschinger effect (Section 2.1). Furthermore, if F (σ) is an
isotropic function, then the yield criterion remains isotropic even after plastic
deformation has taken place, so that the model cannot describe induced
anisotropy.
Kinematic Hardening
In Sections 3.1 and 3.2 we saw, however, that if f can be written in the
form
f (σ, ξ) = F (σ − ρ) − k(ξ), (3.3.7)
then more general hardening behavior can be described. Isotropic hardening
is a special case of (3.3.7) if ρ ≡ 0 and if k depends only on κ, while purely
kinematic hardening corresponds to constant k but nonvanishing variable ρ.
Kinematic hardening represents a translation of the yield surface in stress
space by shifting its reference point from the origin to ρ, and with uniaxial
stressing this means that the the length of the stress interval representing
the elastic region (i.e., the difference between the current yield stress and the
one found on reversal) remains constant. This is in fairly good agreement
with the Bauschinger effect for those materials whose stress-strain curve in
the work-hardening range can be approximated by a straight line (“linear
hardening”), and it is for such materials that Melan [1938] proposed the
model in which ρ = cεp , with c a constant. A similar idea was also proposed
by Ishlinskii [1954], and a generalization of it is due to Prager [1955a, 1956a],
who coined the term “kinematic hardening” on the basis of his use of a me-
chanical model in explaining the hardening rule (Figure 3.3.4). A kinematic
H
HH H
HH

H
HH
.. ..
.. ..
.. ..
.. ..
..
..
.. ....H
.. ............
.. ..
..
..
.. ..

.......... HH
......
H H...H
H . ........ .
...... ........ H ..... .....
......
.
...
.
... ......... H
.
Pin .
....
.
.......
H......
......
. ..
. . ........ ........
..
..
............ .
....... .....
.
........
HH ........
..H
..............
H ..H ............ .
..
...
H H ..
...... ......
.H................. H..H ............ H
.H................. H..H ............ ...
... ........
........
H H ..
H ..H ................... ..H ............ ........ .
.........
H..H ...................
H H..H ............ ....
......
....... ....
..H ................... H..H ............ ...... ......
........
H .
H ..H ................... H .H...H
............ . ........ .. ...
.
H.H . . .
................. H... . ....
..H ................... ......
H ...
...
H ..H ................... π-plane ............
..H ............. .. ..
......
H H.....H .............
............... ..........
H H ...........
. ..
H
Figure 3.3.4. Prager’s mechanical model of kinematic hardening.
hardening model is also capable of representing induced anisotropy, since a

function F (σ − ρ) that depends only on the invariants of its argument stops
being an isotropic function of the stress tensor as soon as ρ differs from zero.
It should be pointed out that, since ρ is a tensor in stress space (some-
times called the back stress, as discussed in 3.1.3), the equation ρij = cεpij
does not imply proportionality between the vectors representing ρ and εp
in any space other than the nine-dimensional space of second-rank tensors,
and particularly not in the six-dimensional space in which symmetric ten-
sors are represented, since the mappings of stress and strain into this space
must be different [see Equations (1.4.9)] in order to preserve the scalar prod-
uct σ · ε = σij εij ; consequently, as was pointed out by Hodge [1957a], the
translation of the yield surface for a material with an associated flow rule
is not necessarily in the direction of the normal to the yield surface, as was
assumed by Prager in constructing his model.
In more sophisticated kinematic hardening models, internal variables
other than εp and κ are included; in particular, the back stress ρ may be
treated as a tensorial internal variable with its own rate equation. Indeed, the
Melan–Prager model falls into this category when its equation is rewritten
as
ρ̇ij = cε̇pij ; (3.3.8)
here c need not be a constant but may itself depend on other internal vari-
ables. In the model described by Backhaus [1968], for example, c depends on
the effective plastic strain εp . Lehmann [1972] replaces the isotropic relation
(3.3.8) between ρ̇ and ε̇p by a more general one,
ρ̇ij = cijkl (σ, ρ)ε̇pkl .
Another example of a kinematic hardening model is that due to Ziegler

[1959],
ρ̇ij = µ̇(σij − ρij ),
where
∂f σ̇ >
< ∂σ kl
kl
µ̇ =
∂f /∂σmn (σmn − ρmn )
in order to satisfy the consistency condition f˙ = 0.
A modification of Equation (3.3.8) that better reproduces the real Bauschinger
effect consists of including in its right-hand side a term representing “fading
strain memory,” so that the rate equation takes the form
ρ̇ij = cε̇pij − aρij ε̇p .
The more general kinematic hardening models can be similarly modified.
Generalized Hardening Rules

The hardening represented by Equation (3.3.7) with both ρ and k vari-
able seems to have been first studied by Kadashevich and Novozhilov [1952];
it is called combined hardening by Hodge [1957a]. The combined hardening
model proposed for viscoplasticity by Chaboche [1977], presented in Section
3.1, has been applied by Chaboche and his collaborators to rate-independent
plasticity as well.
A model with a family of back stresses ρ(l) (l = 1, 2, ..., n) is due to Mróz
[1967]; a similar model is due to Iwan [1967]. Both models describe materials
whose stress-strain curves are piecewise linear. For materials whose stress-
strain curves in the work-hardening range are smooth with straight-line as-
ymptotes, a class of models known as two-surface models have been proposed
by Dafalias [1975] (see also Dafalias and Popov [1975]), Krieg [1975], and
others. In these models the yield surface in stress space is constrained to
move inside an outer surface, known variously as bounding surface, loading
surface, or memory surface, given by, say, f (σ, ξ) = 0. The work-hardening
modulus H at a given state is assumed to be an increasing function of a
suitably defined distance, in stress space, between the current stress σ and
a stress σ on the outer surface, called the image stress of σ. When this
distance vanishes, the work-hardening modulus attains its minimum value,
and further hardening proceeds linearly, with the two surfaces remaining in
contact at σ = σ.
The various two-surface models differ from one another in the definition
of the bounding surface, in the way the image stress depends on the current
state, and in the variation of work-hardening modulus. In the model of

Dafalias and Popov, both surfaces are given similar combined-hardening
structures, with a “back stress” β playing the same role for the outer surface
that ρ plays for the yield surface, and σp = c(σ − ρ) + β, where c is a
constant. H is assumed to depend on δ = (σ − σ) : (σ − σ) in such a way
that H = ∞ at initial yield, producing a smooth hardening curve.
Experiments by Phillips and Moon [1977] showed that when yield sur-
faces are defined on the basis of a very small offset strain, they undergo
considerable distortion, in addition to the expansion and translation con-
sidered thus far. In order to describe such distortion in initially isotropic
materials, Equation (3.3.7) must be modified to
f (σ, ξ) = F (σ − ρ, ξ) − k(ξ),
where F is initially an isotropic function of its first argument but becomes

anisotropic as plastic deformation takes place. An example of such a function
is that proposed by Baltov and Sawczuk [1965] for a Mises-type yield surface:
1
F (σ − ρ, ξ) = Aijkl (ξ)(σij − ρij )(σkl − ρkl ),
2
where
1
Aijkl (ξ) = δik δij − δij δkl + Aεpij εpkl ,
3
A being a constant. Other proposals are reviewed by Bergander [1980].
Extensive experimental investigation of the hardening of metals were carried
out by Phillips and co-workers; their work, along with that of others, is
reviewed by Phillips [1986].

1. Show that the forms (3.3.1) and (3.3.2) for the Tresca yield function
are equivalent.
2. Derive the associated flow rule for the Tresca yield criterion by means
of Koiter’s method (see 3.2.2).
3. Show that for any combination of principal stresses, the associated flow
rule for the Tresca yield criterion gives |ε̇p1 | + |ε̇p2 | + |ε̇p3 | = φ.
4. An elastic–perfectly plastic solid with a uniaxial yield stress of 300

MPa is assumed to obey the Tresca yield criterion and its associated
flow rule. If the rate of plastic work per unit volume is 1.2 MW/m3 ,
find the principal plastic strain-rate components when
(a) σ1 = 300 MPa, σ2 = 100 MPa, σ3 = 0,

(b) σ1 = 200 MPa, σ2 = −100 MPa, σ3 = 0,

(c) σ1 = 200 MPa, σ2 = −100 MPa, σ3 = −100 MPa.
5. Derive the associated flow rule for the general isotropic yield criterion
given by F (J2 , J3 ) − k 2 = 0, and in particular (a) for the one given
by the equation following (3.3.4) and (b) for the analytic form of the
Tresca criterion.
6. A work-hardening elastic-plastic solid is assumed to obey the Mises cri-

terion with the associated flow rule and isotropic hardening. If the vir-
gin curve in uniaxial tension can be described in the small-deformation
range by σ = F (εp ), state the rate equations (3.2.1)–(3.2.2) explicitly
when k is assumed to depend (a) on εp and (b) on Wp .
7. Derive the Mohr-Coulomb criterion as follows.
(a) Using the theory of Mohr’s circles in plane stress, in particular

Equations (1.3.9)–(1.3.10), find the direction θ such that τθ −
µ(−σθ ) is maximum.
p
(b) Show that this maximum value is 1 + µ2 |σ1 − σ2 |/2 + µ(σ1 +
σ2 )/2, and that the Mohr-Coulomb criterion results when this
value is equated to the cohesion c, with µ = tan φ.
(c) Show that the Mohr circles whose parameters σ1 , σ2 are governed
by this criterion are bounded by the lines ±τθ = σθ tan φ − c.
8. Derive the associated flow rule and plastic dissipation for the Drucker-
Prager yield criterion.
9. Given the yield stresses σT and σC in uniaxial tension and compression,

respectively, find the yield stress in shear resulting from the follow-
ing yield criteria: (a) Mohr–Coulomb, (b) Drucker–Prager, (c) Mises–
Schleicher.
10. Show that in a state of plane stress with σ11 = σ, σ12 = τ and σ22 = 0,
both the Tresca and the Mises yield criteria can be expressed in the
form (3.3.5).
11. Derive the form of Equation (3.3.6) for the Mohr-Coulomb criterion in
plane stress.
12. If the function F in Equation (3.3.7) equals J¯2 , with J¯2 defined as
p
in 3.1.1, while k is constant, and if the evolution of ρ is governed by

(3.3.8), show that the rate equation of ρ is
(skl − ρkl )ṡkl

ρ̇ij = (sij − ρij ) .
2k 2
13. Generalize the preceding result to the case where k depends on εp ,

obtaining rate equations for both ρ and εp .
Section 3.4 Uniqueness and Extremum Theorems

3.4.1. Uniqueness Theorems
Uniqueness Theorems in Elastic Bodies

Consider a body made of a linearly elastic material with no internal
constraints, occupying a region R and subject to prescribed tractions ta
on ∂Rt , prescribed displacements ua on ∂Ru , and a prescribed body-force
field b in R. For convenience, the body force per unit volume is defined as
f = ρb. We suppose that a stress field σ and a displacement field u in R
have been found such that σij = Cijkl εkl (where ε is the strain field derived
from u) and σij ,j +fi = 0 in R, nj σij = tai on ∂Rt , and u = ua on ∂Ru .
In other words, (σ, u) constitutes a solution of the static boundary-value
problem. Is this solution unique? As a result of the classical uniqueness
theorem due to Kirchhoff, the answer is “yes” as regards the stress field and
“almost” as regards the displacement field. For, if (σ (1) , u(1) ) and (σ (2) ,
u(2) ) are two different solutions, and if we write φ = φ(1) − φ(2) for any φ,
then σ ij = Cijkl εkl , σ ij ,j = 0 in R, and nj σ ij ūi = 0 on ∂R. Consequently,
by the divergence theorem,
Z
0= (σ ij ūi ),j dV
ZR
= (σ ij ,j ūi + σ ij ūi ,j ) dV
ZR
= σ ij εij dV
ZR
= Cijkl εij εkl dV.
R
It follows from the positive-definiteness of C (see 1.4.3) that ε must vanish

throughout R. Consequently, σ must vanish as well, while ū may have at
most the form of a rigid-body displacement; full uniqueness of the displace-
ment field depends on having sufficient external constraints.
If the material were nonlinearly elastic, the same method could be ap-
plied, but incrementally. Suppose that the stress field σ and displacement
field u have been found under the current f , ta and ua . We may then prove
the uniqueness of infinitesimal increments dσ resulting from increments df ,
dta and dua , provided that C is interpreted as the tangent elastic modulus
Section 3.4 / Uniqueness and Extremum Theorems 153
tensor as defined by Equation (1.4.8), so that
dσij = Cijkl dεkl .
Incremental uniqueness implies global uniqueness, since any state of loading

can be attained by the successive imposition of small incremental loads.
Consequently, the stress and strain fields are uniquely determined (and the
displacement field determined to within a rigid-body displacement) so long
as C is positive definite.
Uniqueness of Stress Field in an Elastic–Plastic Body

The positive-definiteness of C means that, for an elastic material, dσij dεij >
0 whenever dσ 6= 0. The last inequality is equivalent to Drucker’s first in-
equality for a work-hardening plastic material (see Section 3.2). Indeed,
we know that as long as no unloading occurs, no distinction can be made
between plastic and nonlinearly elastic materials. It was shown by Melan
[1938] that incremental uniqueness of stress and strain can be established
for work-hardening standard materials when unloading has taken place, pro-
vided that the hypothesis of infinitesimal strains is valid. The reason for the
proviso is that, with finite deformations, a distinction must be made between
increments at a fixed material point and those at a fixed point in space (see
Hill [1950] and Chapter 8 of the present book).
By analogy with the proof for elastic bodies, it can be shown that a
sufficient condition for the uniqueness of the stress field in a plastic body is
that if dσ (1) and dσ (2) are two possible incremental stress fields and dε(1)
and dε(2) are the associated incremental strain fields, then
(1) (2) (1) (2)
(dσij − dσij )(dεij − dεij ) > 0 (3.4.1)
unless dσ (1) = dσ (2) . It was shown by Valanis [1985] that condition (3.4.1)
applies in dynamic as well as in quasi-static problems.
Consider next an elastic-plastic body made of standard material, occu-
pying the region R. Let Re and Rp denote the parts of R where f < 0 and
f = 0, respectively. With linear elasticity assumed, the inequality is clearly
satisfied in Re , while in Rp we may use the general flow equation (3.2.1)
together with (3.2.2) and the normality rule (3.2.7). Converting rates into
increments by multiplying them by the infinitesimal time increment dt, we
obtain, at any point in Rp ,
(α) −1 (α) <df (α) > ∂f

dεij = Cijkl dσkl + , α = 1, 2, (3.4.2)
H ∂σij
where
∂f (α)
df (α) = dσ .
∂σkl kl
Hence
(1) (2) (1) (2)−1 (1) (2) (1) (2)
(dσij − dσij )(dεij − dεij ) = Cijkl (dσij − dσij )(dσkl − dσkl )
1
+ (df (1) − df (2) )(<df (1) > − <df (2) >).
H
It is easy to see that any two real numbers a, b satisfy <a> − =
β(a − b) for some β, 0 ≤ β ≤ 1, and that therefore the second term on
the right-hand side is never negative. Since the first term is positive unless
dσ (1) = dσ (2) , the uniqueness of dσ (and therefore of dε) is proved, and
hence the uniqueness of the stress and strain fields under a given history of
f , ta , and ua .
If the material is perfectly plastic, then Equation (3.4.2) for the strain
increments at points where f = 0 must be replaced by
(α) −1 (α) ∂f
dεij = Cijkl dσkl + dλ(α) , α = 1, 2, (3.4.3)
∂σij
where dλ(α) = 0 if df (α) < 0, and dλ(α) > 0 only if df (α) = 0 but is otherwise
undetermined. Thus
(1) (2) (1) (2)
−1 (1) (2) (1) (2)
(dσij − dσij )(dεij − dεij ) = Cijkl (dσij − dσij )(dσkl − dσkl )
+(dλ(1) − dλ(2) )(df (1) − df (2) ).

The second term evidently vanishes either if both df (1) and df (2) vanish, or if
both are negative (leading to dλ(1) = dλ(2) = 0). If df (1) = 0 and df (2) < 0,
then the term equals −dλ(1) df (2) and is positive, as it is when (1) and (2)
are interchanged. The uniqueness of dσ — and hence that of the stress field
— follows, but not that of dε.
If normality is not obeyed, then work-hardening (i.e. the positiveness of
H) is not sufficient for uniqueness. The corresponding sufficient condition
is, instead,
H > Hcr ,
where Hcr is a critical value of the hardening modulus given by Raniecki
[1979] as
s !
1 ∂f ∂f q ∂f
Hcr = Cijkl Cijkl hij hkl − Cijkl hij .
2 ∂σij ∂σkl ∂σkl
Clearly, Hcr = 0 when the normality rule (3.2.7) is obeyed.
3.4.2. Extremum and Variational Principles

In Section 1.4 we derived the two fundamental variational principles of elas-
tostatics, which form a dual pair. The principle of minimum potential energy
teaches that in a body in stable equilibrium the correct displacement field

— that is, the one that, with its associated stress field, forms the solution
of the boundary-value problem — is the one which, among all the kinemat-
ically admissible displacement fields that are close to it, minimizes the total
potential energy Π. Similarly, the principle of minimum complementary en-
ergy asserts that the correct stress field is the one which, among all the
neighboring stress fields that are statically admissible, minimizes the total
complementary energy Πc .
If the material is linear, then unless large displacements come into play,
the respective energies depend at most quadratically on their variables, and
it is easy to see that the restriction to admissible fields that are close to
the correct one may be removed. Indeed, it can be shown directly that,
if Π∗ is the total potential energy evaluated at the arbitrary kinematically
admissible displacement field u∗ , and if Π = Π∗ |u∗ =u , where u is the correct
displacement field, then
Π∗ > Π unless u∗ = u.
For
1
Z Z Z
∗
Π −Π = Cijkl (ε∗ij ε∗kl −εij εkl ) dV − fi (u∗i −ui ) dV − tai (u∗i −ui ) dS;
2 R R ∂Rt
but the surface integral may be changed into one over all of ∂R, since u∗ = u
on ∂Ru , and consequently, by the divergence theorem, into
Z
[σij ,j (u∗i − ui ) + σij (ε∗ij − εij )] dV,
R
∗ = C
where σij = Cijkl εkl is the correct stress field. If we define σij ∗
ijkl εkl
then with the help of the equilibrium equations, we obtain
1
Z
Π∗ − Π = ∗ ∗
(σij εij + σij εij − 2σij ε∗ij ) dV.
2 R
The integrand, however, is
Cijkl (εij εkl + ε∗ij ε∗kl − 2ε∗ij εkl ) = Cijkl (ε∗ij − εij )(ε∗kl − εkl ),
and is positive except when ε∗ = ε.

Similarly,
1
Z Z
−1 ∗ ∗ ∗
Πc∗ − Πc = Cijkl (σij σkl − σij σkl ) − nj (σij − σij )uai dS,
2 R ∂Ru
and with the help of analogous transformations, we obtain

1
Z
−1 ∗ ∗ ∗
Πc∗ − Πc = Cijkl [σij σkl − σij σkl − 2(σij − σij )εij ] dV,
2 R
−1
where εij = Cijkl σkl , so that the integrand is
−1 ∗ ∗
Cijkl (σij − σij )(σkl − σkl ),
also positive except when σ ∗ = σ.

In view of the resemblance between the uniqueness proof and the proofs
of the extremum principles, it appears reasonable that such principles may
be derived for displacement and stress increments (or, equivalently, velocities
and stress rates) in elastic-plastic bodies that are either work-hardening or
perfectly plastic. Such is indeed the case. The theorems to be presented
have been derived by Handelman [1944], Markov [1947], Greenberg [1949],
and Hill [1950];1 the proofs are Hill’s.
Extremum Principle for Displacement

Let du∗ denote a kinematically admissible displacement increment, that
is, one which obeys the internal constraints, if any, and which satisfies du∗ =
dua on ∂Ru . The corresponding incremental strain and stress field are dε∗
and dσ ∗ , where dσ ∗ is not in general statically admissible, but is related
to dε∗ through the associated flow rule; we thus use Equation (3.2.10) with
hij = ∂f /∂σij , to obtain Hill’s [1958] result


 Cijkl ε̇kl , f < 0,

σ̇ij =
!
1 ∂f ∂f
 Cijkl

 ε̇kl − <Cpqmn ε̇mn > , f = 0.
L ∂σpq ∂σkl
Here L is given by Equation (3.2.9) for a standard material, namely,
∂f ∂f
L = H + Cijkl ,
∂σij ∂σkl
and the perfectly plastic case corresponds to H = 0.

Let the functional Λ∗ be defined by
1
Z Z Z
Λ∗ = ∗
dσij dε∗ij dV − dfi du∗i dV − dtai du∗i dS,
2 R R ∂Rt
with
Λ = Λ∗ |du∗ =du .
Note that the form of Λ is essentially that of the potential energy Π of the
linear elastic body, with incremental rather than total displacements and
with the total (elastic-plastic) tangent modulus tensor in place of C. By
1
Some extensions have been proposed by, among others, Ceradini [1966], Maier [1969,
1970] and Martin [1975].
means of the transformations used in the elastic case, we can therefore show
that
1
Z
Λ∗ − Λ = ∗
(dσij dε∗ij + dσij dεij − 2dσij dε∗ij ) dV,
2 R
and it remains to be shown that the integrand is positive unless dσ ∗ = dσ.
With the decomposition dε = C−1 dσ + dεp , the integrand becomes
−1
Cijkl ∗
(dσij ∗
− dσij )(dσkl ∗
− dσkl ) + dσij dεp∗ p∗
ij + dσij dεij − 2dσij dεij .
The term in C−1 is clearly nonnegative, and vanishes only if dσ ∗ = dσ.

As for the remaining terms, in the case of a work-hardening material
their sum becomes
1
φ= (df ∗ <df ∗ > + df <df > − 2df <df ∗ >),
H
where df = (∂f /∂σij ) dσij and df ∗ = (∂f /∂σij ) dσij∗ . Clearly, φ vanishes
if df ∗ and df are both nonpositive, and equals H −1 (df ∗ − df )2 if both are

positive. If df ∗ > 0 and df ≤ 0, φ = H −1 df ∗ (df ∗ − 2df ) > 0, while in the
opposite case φ = H −1 (df )2 . For the perfectly plastic material, we replace
H −1 <df > by dλ, with dλ related to df as in the uniqueness proof, and
similarly H −1 <df ∗ > by dλ∗ , so that
φ = df ∗ dλ∗ + df dλ − 2df dλ∗ .
The first two terms always vanish, as does the last term except when df < 0
and df ∗ = 0, in which case dλ∗ > 0, so that φ = −2df dλ∗ > 0. Consequently
φ is never negative, so that
Λ∗ > Λ (3.4.4)
except when dσ ∗ = dσ. For the work-hardening (but not the perfectly
plastic) material this also means that dε∗ = dε and therefore du∗ = du if
sufficient constraints exist.
Extremum Principle for Stress
The complementary extremum principle concerns a statically admissi-
ble incremental stress field dσ ∗ . In the case of a work-hardening material
this determines an incremental strain field dε∗ , related to dσ ∗ through the
associated flow rule — Equation (3.4.2) with the superscript (α) replaced
by ∗ — but not, in general, derivable from a continuous displacement field.
Clearly,
∗ −1 1
dσij dε∗ij = Cijkl ∗
dσij ∗
dσkl + <df ∗ >2 .
H
−1
If the material is perfectly plastic, on the other hand, then dε∗ij = Cijkl ∗ +
dσkl
dεp∗ p∗ ∗ ∗
ij , with dε not determined by dσ . However, in such a material dσij dεij =
p∗
0, and therefore
∗ −1
dσij dε∗ij = Cijkl ∗
dσij ∗
dσkl .
We now define the functional

1
Z Z
∗ ∗
Ω = dσij dε∗ij dV − ∗
nj dσij duai dS,
2 R ∂Ru
∗ dε∗ defined by the appropriate formula above. We also define, as
with dσij ij
before,
Ω = Ω∗ |dσ∗ =dσ .
Note that the form of Ω is, mutatis mutandis, that of Πc , and also that
Ω = −Λ.
The surface integral in Ω∗ is transformed into

Z
∗
dσij dεij dV,
R
and therefore
1
Z
Ω∗ − Ω = ∗
(dσij dε∗ij + dσij dεij − 2dσij
∗
dεij ) dV.
2 R
The integrand differs from the one in Λ∗ − Λ only in the interchange of

starred and unstarred quantities, and may by the same method be shown to
be nonnegative and to vanish only if dσ ∗ = dσ, so that
Ω∗ > Ω unless dσ ∗ = dσ. (3.4.5)
We furthermore have the double inclusion
−Λ∗ ≤ −Λ = Ω ≤ Ω∗ .
Variational Principles
As mentioned earlier, an extremum principle is stronger than a varia-
tional principle because it asserts an extremum over all admissible functions
of a certain class, not only over those that are infinitesimally close to the ex-
tremal. Going further, we see that a variational principle need not assert an
extremum at all, even a local one, but only the condition that the functional
obeying it is stationary. For example, the function f (x) = x3 is stationary
at x = 0, since f 0 (0) = 0, but has neither a minimum nor a maximum there.
Likewise, f (x, y) = x2 − y 2 is stationary at (0, 0); the point is a saddle
point — a maximum when viewed along the x-axis and a minimum along
the y-axis.
Extremum principles are useful for many reasons, one of them being that
they allow us to evaluate approximate solutions when the exact solution is
unknown: between two incremental displacement fields du∗ , in the absence
of other information we choose the one that produces the smaller value of
Λ∗ . But variational principles are useful even when they are not extremum
principles: they permit compact statements of boundary-value problems,
and they are useful in formulating the “weak form” of such problems, which
is necessary for consistent discretization, as shown in 1.3.5 and as further
shown in Section 4.5 which deals with numerical methods.
The extremum principles derived above clearly imply the corresponding
variational principles:
δΛ = 0, δΩ = 0.
We should remember, however, that the extremum principles rely on the
positive-definiteness of certain local quantities, whose proof requires that
the material be nonsoftening. Consider, on the other hand, the following
functional:
1 1 1 ∂f
Z Z
Θ= Cijkl ε̇ij ε̇kl dV − <Cijkl ε̇ij >2 dV
2 R 2 Rp L ∂σkl
Z Z
− ḃi u̇i dV − ṫai u̇i dV.
R ∂Rt
It is easy to see that Θ is just Λ with rates in place of increments; the

change is made to avoid mixing the differential operator d with the variation
operator δ. Clearly, with σ̇ given in terms of ε̇ by the strain-space flow rule,
Z Z Z
δΘ = σ̇ij δ ε̇ij dV − ḃi δ u̇i dV − ṫai δ u̇i dV,
R R ∂Rt
and δΘ = 0 for an arbitrary kinematically admissible δ u̇ only if σ̇ is statically

admissible.
3.4.3. Rigid–Plastic Materials

In discussing the Saint-Venant–Lévy–Mises flow rule in the preceding sec-
tion, we mentioned in passing that these authors equated ε̇ with ε̇p , in effect
neglecting the elastic strain rate — a treatment tantamount to treating the
nonvanishing elastic moduli as infinite. Any solutions obtained on this basis
are, theoretically, valid for idealized materials called rigid-plastic.1 Practi-
cally, however, they are useful approximations for two classes, not mutually
exclusive, of problems: (a) those in which the elastic strain rates may be
neglected, and (b) those in which the elastic strains are significantly smaller
than the plastic ones. Problems of class (a) include those of impending col-
lapse or incipient plastic flow of elastic–perfectly plastic bodies, for which it
1
In many references, the theory of rigid-plastic materials obeying the Mises yield cri-
terion and the associated flow rule is called the Mises theory, while the corresponding
theory of elastic-plastic materials is called the Prandtl–Reuss theory.
is shown later in this section that, when deformation proceeds at constant

loads, the elastic strain rates vanish identically. In order to accommodate
problems of class (b), the hypothesis of infinitesimal deformations should be
abandoned in formulating any general theorems for this theory. Since the
problems are frequently those of flow, it appears natural to use an Eulerian
formulation, in which R is the region currently occupied by the body and
σ is the true (Cauchy) stress tensor. No reference configuration is used in
this approach, and therefore no displacement field u appears; in its place
we have the velocity field v. The boundary ∂R is accordingly partitioned
into ∂Rt and ∂Rv , with v = va on ∂Rv . No strain tensor is introduced, but
rather the Eulerian deformation-rate tensor (also called stretching tensor )
d, defined by
dij = 21 (vi ,j +vj ,i ).
With infinitesimal deformations, of course, d approximates ε̇.
Uniqueness of Stress Field

In general plasticity theory, the decomposition of the deformation rate
into elastic and plastic parts is far from unequivocal, as is shown in Chapter
8. In a rigid-plastic material, however, d may be identified with the plastic
strain rate, and thus the plastic dissipation per unit volume, Dp , defined in
Section 3.2, satisfies
Dp (d) = σij dij . (3.4.6)
The maximum-plastic-dissipation principle may therefore be written
∗
(σij − σij )dij ≥ 0. (3.4.7)
If d 6= 0, then the equality holds only if σ and σ ∗ are plastically equivalent,

that is, if d is related to both of them through the associated flow rule. In a
Mises material, plastically equivalent stresses differ at most by a hydrostatic
pressure, but in a Tresca material two stresses are plastically equivalent if
they lie on the same facet of the hexagonal cylinder in principal-stress space.
If, now, (σ (α) , v(α) ), α = 1, 2, represent two admissible states of a rigid-
plastic body, corresponding to the same body force and boundary conditions,
then
(1) (2) (1) (2)
nj (σij − σij )(vi − vi ) = 0 on ∂R,
and therefore
Z Z
(1) (2) (1) (2) (1) (2) (1) (2)
nj (σij − σij )(vi − vi ) dS = (σij − σij )(dij − dij ) dV = 0.
∂R R
But the last integrand may be written as

(1) (2) (1) (2) (1) (2)
(σij − σij )dij + (σij − σij )dij ,
and this is positive unless either d(1) and d(2) both vanish, or σ (1) and σ (2)
are plastically equivalent. The general conclusion is thus that two admissible
stress fields σ (1) and σ (2) must be plastically equivalent everywhere except
in their common rigid region, that is, at points where f (σ (α) ) < 0, α = 1, 2.
If the body is made of a Mises material (or any other material whose yield
surface in stress-deviator space is strictly convex) and deforms plastically in
its entirety, then the two stress fields can differ at most by a hydrostatic
pressure field, which must be uniform in order to satisfy equilibrium, and
must vanish if a surface traction is prescribed anywhere on ∂R. Thus there
is not more than one admissible stress field for which the whole body is
plastic (Hill [1948a]), unless ∂R = ∂Rv , in which case the stress field is
determined only to within a uniform hydrostatic pressure. On the other
hand, uniqueness of the velocity field is not established.
Extremum Principle for Velocity

The degree to which the velocity field is determined may be learned
from the kinematic extremum principle to be shown next, first proposed
by Markov [1947] for a Mises material. Given a kinematically admissible
velocity field v∗ , we define a functional Γ∗ by
Z Z Z
Γ∗ = Dp (d∗ ) dV − fi vi∗ dV − tai vi∗ dS,
R R ∂Rt
and, as usual,
Γ = Γ∗ |v∗ =v .
Using the standard transformations we can show that
Z
Γ= nj σij via dS (3.4.8)
∂Rv
and that Z
Γ∗ − Γ = [Dp (d∗ ) − Dp (d) − σij (d∗ij − dij )] dV.
R
Because of (3.4.6), the integrand is just Dp (d∗ ) − σij d∗ij , and this is non-
negative as a result of the maximum-plastic-dissipation principle, Equation
(3.4.7), since the actual stress σ necessarily obeys the yield criterion. Con-
sequently,
Γ∗ ≥ Γ. (3.4.9)
It is not possible, in general, to strengthen the inequality by asserting that
the equality holds only when v∗ = v. The most we can say is that if Γ∗ = Γ,
then v∗ is kinematically admissible, and d∗ is associated with a stress field σ ∗
that is statically admissible and obeys the yield criterion everywhere. More
particularly, however, if the body is one for which the stress field is unique
(see above) and if the stress determines the deformation rate to within a
scale factor (as is true of the Mises material, or any other material with a
smooth yield surface), then the entire deformation-rate field is determined

to within a scale factor. The indeterminacy may be eliminated if a nonzero
velocity is prescribed anywhere on ∂Rv .
Principle of Maximum Plastic Work
Given a statically admissible stress field σ ∗ which nowhere violates the
yield criterion, then, if σ and v are the actual stress and velocity fields,
respectively, we clearly have
Z Z
∗ ∗
(σij − σij )dij dV = nj (σij − σij )via dS.
R ∂Rv
From the principle of maximum plastic dissipation there immediately follows

the result Z Z
∗ a
nj σij via dS ≥ nj σij vi dS,
∂Rv ∂Rv
due to Hill [1948a] and dubbed by him the principle of maximum plastic
work. Here, again, the equality holds only if σ ∗ and σ are plastically
equivalent, and therefore only if σ ∗ = σ whenever σ is unique. Note that
the left-hand side of the inequality is equal to Γ, Equation (3.4.8), and is
therefore bounded both above and below.
Many results relating to variational principles in both elastic-plastic and
rigid-plastic solids are contained in Washizu [1975].

1. Show that for any two real numbers a, b,
<a> − = β(a − b)
for some β, 0 ≤ β ≤ 1.
2. Show that the Hu–Washizu principle (Exercise 14 of Section 1.4) may
be extended to elastic-plastic materials if W (ε) is replaced by W (ε −
εp ).
3. Derive Equation (3.4.8) and the one following it.
Section 3.5 Limit-Analysis and Shakedown Theo-

rems
3.5.1. Standard Limit-Analysis Theorems
The extremum principles for standard rigid-plastic materials that were dis-
cussed in the preceding section can be reformulated as the theorems of
Section 3.5 / Limit-Analysis and Shakedown Theorems 163
limit analysis, which give upper and lower bounds on the loads under
which a body that may be approximately modeled as elastic–perfectly plas-
tic reaches a critical state. By a critical state we mean one in which large
increases in plastic deformation — considerably greater than the elastic de-
formation — become possible with little if any increase in load. In the case of
perfectly plastic bodies this state is called unrestricted plastic flow ,1 and the
loading state at which it becomes possible is called ultimate or limit loading.
It will be shown that, in a state of unrestricted plastic flow, elasticity may
be ignored, and therefore a theory based on rigid-plastic behavior is valid
for elastic-plastic bodies.
The proof of the limit-analysis theorems is based on the principle of max-
imum plastic dissipation, and consequently they are valid only for standard
materials; a limited extension to nonstandard materials is discussed in 3.5.2.
It should be noted that the “loads” in the present context include not
only the prescribed surface tractions ta but all the surface tractions t op-
erating at points at which the displacement (or velocity) is not constrained
to be zero. In other words, the loads include reactions that do work ; the
definition of ∂Rt is accordingly extended. The reason for the extension is
that the velocity fields used in the limit-analysis theorems are kinematically
admissible velocity fields, not virtual velocity fields. The latter is, as we
recall, the difference between two kinematically admissible velocity fields,
and must therefore vanish wherever the velocity is prescribed, whereas a
kinematically admissible velocity field takes on the values of the prescribed
velocity.
We begin by defining a state of impending plastic collapse or incipient
plastic flow as one in which a nonvanishing strain rate (ε̇ 6= 0) occurs under
constant loads (ḟ = 0, ṫ = 0). The qualification “impending” or “incipient”
is important: we are looking at the very beginning of such a state, which
means that (1) all prior deformation has been of the same order of magnitude
as elastic deformation, so that changes of geometry can be neglected, and (2)
acceleration can be neglected and the problem can be treated as quasi-static.
Vanishing of Elastic Strain Rates

In addition to the preceding assumptions, it is assumed that the equa-
tions of equilibrium and the traction boundary conditions can be differenti-
ated with respect to time with no change in form; consequently, the principle
of virtual work is valid with σ̇, ḟ , and ṫ replacing σ, f , and t. For a virtual
1
A rigorous formulation of the theorems in the context of convex analysis is due to
Frémond and Friaâ [1982], who show that the concept of unrestricted plastic flow is too
general for materials whose elastic range is unbounded in the appropriate stress space, and
must be replaced by the weaker concept of “almost unrestricted” plastic flow (écoulement
presque libre).
displacement field we take v δt, where v is the actual velocity field and δt is
a small time increment. The virtual strain field is, accordingly, ε̇ij δt, where
ε̇ij = 12 (vi ,j +vj ,i ). At impending collapse, then,
Z Z Z Z
0= ḟ · v dV + ṫ · v dS = σ̇ij ε̇ij dV = σ̇ij (ε̇pij + Cijkl
−1
σ̇kl ) dV.
R ∂Rt R R
The positive definiteness of the elastic complementary energy implies

−1
Cijkl σ̇ij σ̇kl ≥ 0 unless σ̇ = 0. This fact, combined with Drucker’s inequality
(3.2.3), implies that at impending collapse or incipient plastic flow the stress
rates vanish, so that ε̇e = 0 and ε̇ = ε̇p . In other words, a body experiencing
plastic collapse or flow behaves as though it were rigid-plastic rather than
elastic-plastic. This result, first noted by Drucker, Greenberg, and Prager
[1951], makes possible the rigorous application to elastic-plastic bodies of
the theorems of limit analysis that had previously been formulated for rigid-
plastic bodies. The following presentation of the theorems follows Drucker,
Greenberg and Prager.1
Lower-Bound Theorem
Suppose that at collapse the actual loads are f , t and the actual stress,
velocity and strain-rate fields (in general unknown) are σ, v and ε̇. Suppose
further that we have somehow determined a stress field σ ∗ which does not
violate the yield criterion anywhere and which is in equilibrium with the
loads f ∗ = (1/s)f , t∗ = (1/s)t, where s is a numerical factor. By virtual
work, we have
1
Z Z Z
∗
σij ε̇ij dV = t · v dS + f · v dV
R s ∂R R
1 1
Z Z
= σij ε̇ij dV = Dp (ε̇) dV.
s R s R
But, by the principle of maximum plastic dissipation, Dp (ε̇) ≥ σij ∗ ε̇ , so

ij
that s ≥ 1. In other words, the factor s (the so-called “static multiplier”) is
in fact a safety factor.
Upper-Bound Theorem
Let us suppose next that instead of σ ∗ , we somehow determine a velocity
field v∗ (a collapse mechanism), with the corresponding strain-rate field ε̇∗ ,
and loads f ∗ = cf , t∗ = ct that satisfy
Z Z Z
t∗ · v∗ dS + f ∗ · v∗ dV = Dp (ε̇∗ ) dV,
∂Rt R R
1
See also Drucker, Prager and Greenberg [1952], Hill [1951, 1952], and Lee [1952].
provided the right-hand side (the total plastic dissipation) is positive;2 then,
again by virtual work,
Z Z
Dp (ε̇∗ ) dV = c σij ε̇∗ij dV,
R R
where σ is, as before, the actual stress field at collapse. The principle of
maximum plastic dissipation, however, also implies that Dp (ε̇∗ ) ≥ σij ε̇∗ij .
Consequently c ≥ 1, that is, c (the “kinematic multiplier”) is an overload
factor.
Alternative Formulation for One-Parameter Loading

Rather than using multipliers, the theorems can also be expressed in
terms of a single reference load to which all the loads on the body are propor-
tional. Let this reference load be denoted P , and let the loading (consisting
of applied loads and working reactions) be expressed as
f = P f̃ , t = P t̃,
where f̃ and t̃ are known functions of position in R and on ∂Rt , respectively.

Let, further, PU denote the value of P at collapse. If a plastically admissible
stress field σ ∗ is in equilibrium with P f̃ and P t̃ for some value of P , say
PLB ∗ , then this value is a lower bound, that is, PLB ∗ ≤ PU . An upper bound
PU B ∗ can be found explicitly for a kinematically admissible velocity field v∗ :
∗
R
∗ R Dp (ε̇ )dV
PU ≤ PU B = R . (3.5.1)
· v∗ dV + t̃ · v∗ dS
R
R f̃ ∂Rt
For loading governed by a single parameter, therefore, the two theorems may
also be expressed as follows: The loads that are in equilibrium with a stress
field that nowhere violates the yield criterion do not exceed the collapse loads,
while the loads that do positive work on a kinematically admissible velocity
field at a rate equal to the total plastic dissipation are at least equal to the
collapse loads. If the loads produced by the application of the two theorems
are equal to each other, then they equal the collapse loads.
In particular, if one has succeeded in finding both (a) a statically and
plastically admissible stress field, and (b) a kinematically admissible velocity
field such that the strain rate produced by it is everywhere1 associated to
the stress, then a complete solution is said to have been found. This solution
is not necessarily unique and hence cannot be called an exact solution, but,
as a result of the theorems of limit analysis, it predicts the correct collapse
load. Some applications of this concept are given in Chapter 5. In Chapter
2
If the total plastic dissipation is negative, v∗ can be replaced by −v∗ ; if it is zero, v∗
does not represent a collapse mechanism.
1
If the velocity field involves regions that move as rigid bodies (rigid regions), then the
strain rate there is of course zero and the question of association does not arise.
6 the theorems are used to obtain estimates of collapse loads in problems for
which no complete solutions have been found.
Multiparameter Loadings
We now consider loadings that are governed by several parameters that
can vary independently. These parameters will be called generalized loads;
some of them may be applied loads, and others may be reactions that do
work. Let them be denoted PI (I = 1, ..., N ), so that
N
X N
X
f= PI f̃ (I) , t= PI t̃(I) ,
I=1 I=1
the f̃ (I) and t̃(I) again being known functions. For a kinematically admissible
velocity field v, generalized velocities ṗI can be defined by
Z Z
ṗI = f̃ (I) · vdR + t̃(I) · vdS,
R ∂Rt
so that, by virtual work,

Z N
X def
σij ε̇ij dV = PI ṗI = P · ṗ, (3.5.2)
R I=1
where P and ṗ are the N -dimensional vectors representing the PI and ṗI .
Any combination of generalized loads PI thus represents a point in P-space
(a load point), while a fixed proportion among the PI represents a direction
in this space (a loading direction).
A combination of PI at which unrestricted plastic flow occurs represents
a limit point (or flow point or yield point — the last designation makes sense
only in terms of a rigid-plastic material), and the set of all such points is the
limit locus (or flow locus or yield locus), given by, say,
Φ(P) = 0.
It follows from Equation (3.5.2) and the principle of maximum plastic dis-
sipation that if P and P∗ are on or inside the limit locus, and if ṗ is a
generalized velocity vector that is possible under a load vector P, then
(P − P∗ ) · ṗ ≥ 0. (3.5.3)
By arguments identical with those of 3.2.2 for the maximum-plastic-

dissipation principle in terms of stresses and strain rates, it follows from
inequality (3.5.3) that the limit locus is convex, and that the generalized
velocity vector is normal to the limit locus, that is,
∂Φ
ṗI = λ̇
∂PI
wherever Φ(P) is regular. The appropriate generalization, following either

the Koiter or the Moreau formalism, may be formed for singular limit loci.
A fixed loading direction is described by N − 1 parameters, and keeping
these constant is equivalent to one-parameter loading. Consequently, along
any direction lower-bound and upper-bound load points can be found, and
since they depend on the N − 1 parameters, they form parametric represen-
tations of upper-bound and lower-bound loci.
A simple illustration of multi-parameter loading is provided by an “ideal
sandwich beam,” composed of two equal, very thin flanges of cross-sectional
area A, separated by a distance h that is spanned by a web of negligible area
[see Figure 3.5.1(a)]. The beam is subject to an axial force P , whose line of
M/MU
1
6 SS
o @
@
h A @ P/PU
@ 1
@
?
/ @
(a)
(b)
Figure 3.5.1. Ideal sandwich beam: (a) cross-section; (b) flow locus.
action is defined as midway between the flanges, and a bending moment M ;

P and M may be treated as generalized loads, and the conjugate generalized
˙ and the rotation
velocities are respectively the average elongation rate ∆
rate θ̇. The stresses in the flanges may be assumed to be purely axial, and
equilibrium requires that they be
P M
σ= ± .
2A Ah
If the yield stress is σY , then the yield inequality |σ| ≤ σY in each flange is
equivalent to
P M

P ± ≤ 1,
U MU
where PU = 2σY A, MU = σY Ah. The limit locus, shown in Figure 3.5.1(b),
is thus given by

P M P M

Φ(M, P ) = max
+ , − − 1 = 0.
PU M U PU MU
Since this locus was found on the basis of a stress distribution, it is

strictly speaking a lower-bound locus. The stress distribution is, of course,
unique, and therefore the lower bound must in fact give the true limit locus.
It can easily be shown that the same limit locus results from an application
of the upper-bound theorem. Consider a velocity field in which only one of
the flanges elongates while the other remains rigid. In order for the mean
elongation rate of the beam to be ∆, ˙ that of the deforming flange must be
˙
2∆, and the rotation rate, to within a sign, is θ̇ = 2∆/h. ˙ For the sake
˙
of definiteness, let us take both ∆ and θ̇ as positive. The strain rate in
˙
the deforming flange is 2∆/L, so that the total plastic dissipation — the
˙
numerator on the right-hand side of (3.5.1) — is σY (2∆/L)(AL) ˙
= PU ∆.
We take P as the reference load, and pick a loading direction by letting M =
αP h/2. The denominator in (3.5.1) is thus ∆ ˙ + (αh/2)(2∆/h)
˙ ˙
= (1 + α)∆,
and the upper bound for P is PU /(1 + α). Since MU = PU h/2, the upper
bound for M is αMU /(1 + α). The upper-bound values satisfy M/MU +
P/PU = 1, an equation describing the first quadrant of the previously found
limit locus. The remaining quadrants are found by varying the signs of ∆ ˙
and θ̇.
A velocity field with both flanges deforming leads to an upper-bound
load point lying outside the limit locus just found, with two exceptions: one
where the elongation rates of the flanges are the same, and one where they
are equal and opposite. Details are left to an exercise.
3.5.2. Nonstandard Limit-Analysis Theorems

The theorems of limit analysis can be stated in a form that does not directly
refer to any concepts from plasticity theory:
A body will not collapse under a given loading if a possible stress field can
be found that is in equilibrium with a loading greater than the given loading.
A body will collapse under a given loading if a velocity field obeying the
constraints (or a mechanism) can be found that so that the internal dissipa-
tion is less than the rate of work of the given loading.
In this form, the theorems appear intuitively obvious. In fact, the con-
cepts underlying the theorems were used long before the development of
plasticity theory. Use of what is essentially the upper-bound theorem goes
back to the eighteenth century: it was used in 1741 by a group of Italian
mathematicians to design a reinforcement method for the crumbling dome
of Saint Peter’s Church, and in 1773 by Coulomb to investigate the collapse
strength of soil. The latter problem was also studied by Rankine in the mid-
nineteenth century by means of a technique equivalent to the lower-bound
theorem.
The simple form of the theorems given above hides the fact that the
postulate of maximum plastic dissipation (and therefore the normality of the
flow rule) is an essential ingredient of the proof. It was therefore necessary
to find a counterexample showing that the theorems are not universally
applicable to nonstandard materials. One such counterexample, in which
plasticity is combined with Coulomb friction at an interface, was presented

by Drucker [1954a]. Another was shown by Salençon [1973].
Radenkovic’s Theorems
A theory of limit analysis for nonstandard materials, with a view toward
its application to soils, was formulated by Radenkovic [1961, 1962], with
modifications by Josselin de Jong [1965, 1974], Palmer [1966], Sacchi and
Save [1968], Collins [1969], and Salençon [1972, 1977]. Radenkovic’s first
theorem may be stated simply as follows: The limit loading for a body made
of a nonstandard material is bounded from above by the limit loading for the
standard material obeying the same yield criterion.
The proof is straightforward. Let v∗ denote any kinematically admissible
velocity field, and P∗ the upper-bound load point obtained for the standard
material on the basis of this velocity field. If σ is the actual stress field at
collapse in the real material, then, since this stress field is also statically and
plastically admissible in the standard material,
Dp (ε̇∗ ) ≥ σij ε̇ij ∗ ,
and therefore, by virtual work,
P∗ · ṗ∗ ≥ P · ṗ∗ .
Since v∗ may, as a special case, coincide with the correct collapse velocity
field in the fictitious material, P∗ may be the correct collapse loading in this
material, and the theorem follows.
Radenkovic’s second theorem, as modified by Josselin de Jong [1965],
is based on the existence of a function g(σ) with the following properties:
1. g(σ) is a convex function (so that any surface g(σ) = constant is

convex);
2. g(σ) = 0 implies f (σ) ≤ 0 (so that the surface g(σ) = 0 lies entirely
within the yield surface f (σ) = 0);
3. to any σ with f (σ) = 0 there corresponds a σ 0 such that (a) ε̇p is

normal to the surface g(σ) = 0 at σ 0 , and (b)
0
(σij − σij )ε̇ij ≥ 0. (3.5.4)
The theorem may then be stated thus: The limit loading for a body made
of a nonstandard material is bounded from below by the limit loading for the
standard material obeying the yield criterion g(σ) = 0.
The proof is as follows. Let σ denote the actual stress field at collapse,
P the limit loading, v the actual velocity field at collapse, ε̇ the strain-rate
field, and ṗ the generalized velocity vector conjugate to P. Thus, by virtual

work, Z
P · ṗ = σij ε̇ij dV.
R
Now, the velocity field v is kinematically admissible in the fictitious standard
material. If σ 0 is the stress field corresponding to σ in accordance with the
definition of g(σ), then it is the stress field in the fictitious material that is
plastically associated with ε̇, and, if P0 is the loading that is in equilibrium
with σ 0 , then Z
P0 · ṗ = 0
σij ε̇ij dV.
R
It follows from inequality (3.5.4) that
P0 · ṗ ≤ P · ṗ.
Again, σ 0 may, as a special case, coincide with the correct stress field at
collapse in the standard material, and therefore P0 may be the correct limit
loading in this material. The theorem is thus proved.
In the case of a Mohr–Coulomb material, the function g(σ) may be
identified with the plastic potential if this is of the same form as the yield
function, but with an angle of dilatation that is less than the angle of internal
friction (in fact, the original statement of the theorem by Radenkovic [1962]
referred to the plastic potential only). The same is true of the Drucker–
Prager material.
It should be noted that neither the function g, nor the assignment of σ 0 to
σ, is unique. In order to achieve the best possible lower bound, g should be
chosen so that the surface g(σ) = 0 is as close as possible to the yield surface
f (σ) = 0, at least in the range of stresses that are expected to be encountered
in the problem studied. Since the two surfaces do not coincide, however, it
follows that the lower and upper bounds on the limit loading, being based on
two different standard materials, cannot be made to coincide. The correct
limit loading in the nonstandard material cannot, therefore, be determined
in general. This result is consistent with the absence of a uniqueness proof
for the stress field in a body made of a nonstandard perfectly plastic material
(see 3.4.1).
3.5.3. Shakedown Theorems

The collapse discussed thus far in the present section is known as static
collapse, since it represents unlimited plastic deformation while the loads
remain constant in time. If the loads are applied in a cyclic manner, without
ever reaching the static collapse condition, other forms of collapse may occur.
If the strain increments change sign in every cycle, with yielding on both
sides of the cycle, then alternating plasticity is said to occur; the net plastic
deformation may remain small, but weakening of the material may occur
nevertheless — a phenomenon called low-cycle fatigue — leading to breaking
of the most highly stressed points after a certain number of cycles.
It may also happen that plastic deformation in each cycle accumulates so
that after enough cycles, the displacements are large enough to be equivalent
to collapse; this is called incremental collapse. On the other hand, it may
happen that no further plastic deformation occurs after one or a few cycles
— that is, all subsequent unloading-reloading cycles are elastic. In that case
the body is said to have experienced shakedown or adaptation. It is obvious
that for bodies subject to repeated loading, shakedown is more relevant than
static collapse, and that criteria for shakedown are of great importance.
Residual Stresses: Example

If an initially stress-free body has been loaded into the plastic range, but
short of collapse, and the loads are then reduced to zero, then the stress field
in the unloaded body does not in general vanish. As an example, consider
the four-flange sandwich beam shown in Figure 3.5.2, subject to a bending
moment M only. If the flanges behave similarly in tension and compression,
2
6 J
]
J
h J
J
J
?
YH
H 1 J
6 H JJ
HH
O h O
JA
H

?
10
? 20

Figure 3.5.2. Ideal four-flange beam: geometry.
then the stresses in the flanges satisfy σ10 = −σ1 and σ20 = −σ2 , and the
moment is
M = −σ1 Ah − 3σ2 Ah.
In accordance with elementary beam theory (which is discussed further in

Section 4.4), the longitudinal strain in the flanges varies linearly with dis-
tance from the so-called neutral plane, which in the present case may be
shown to be the plane OO. Thus ε2 = 3ε1 , with ε10 = −ε1 and ε20 = −ε2 .
As long as all flanges are elastic, stress is proportional to strain, and therefore
σ2 = 3σ1 . It follows that

M def e 3M def e
σ1 = − = σ1 , σ2 = − = σ2 .
10Ah 10Ah
By definition, the notation σ e will be used for stresses calculated on the
basis of assumed elastic behavior of a body, or elastic stresses.
def
The outer flanges will yield when |M | = (10/3)σY Ah = ME . Provided
that the web holds, however, the deformation will remain contained so long
as the inner flanges remain elastic. Collapse occurs when all four flanges
yield, that is, when σ1 = σ2 = −σY , so that |M | = 4σY Ah = MU . The
range ME ≤ |M | ≤ MU is called the range of contained plastic deformation.
Suppose, for example, that M = (11/3)σY Ah = M ∗ . The actual stresses
are σ1 = −(2/3)σY , σ2 = −σY , while the elastic stresses are σ1e = −(11/30)σY ,
σ2e = −(11/10)σY . When the moment is removed, the flanges will unload
elastically, a process equivalent to subtracting the elastic stresses from the
actual stresses. The resulting residual stresses are σ1r = −(3/10)σY and
σ2r = (1/10)σY . Their resultant moment is, of course, zero. If the beam is
now repeatedly reloaded with a bending moment of the same sign as before,
the response will be elastic provided that M ≤ M ∗ . Consequently, for the
given load amplitude, shakedown takes place in the first cycle.
Residual Stress and Displacement

A stress field ρ that is in equilibrium with zero body force and zero
prescribed surface tractions is called self-equilibrated , or, more simply, a
field of self-stress. Such a field must clearly satisfy
ρij ,j = 0 in R, ρij nj = 0 on ∂Rt .
In an elastic-plastic body under given loads f , ta , the stress field can

always be written as
σ = σ e + ρ, (3.5.5)
where σ e is the elastic stress field corresponding to the given loads, and ρ is
the self-equilibrated field of residual stresses. The strain field can accordingly
be written as
−1 e
εij = Cijkl −1
σkl + Cijkl ρkl + εpij . (3.5.6)
The first term, which will be written ε0ij , represents the strain field in the
hypothetical elastic body under the prescribed loads. This strain field is
compatible with a displacement field that will be denoted ue , that is,
1
ε0ij = (uei ,j +uej ,i ).
2
Since the total strain field ε is also compatible, the remaining terms of (3.5.6)
are also derivable from a displacement field ur (the residual displacement
field):
1
−1
Cijkl ρkl + εpij = (uri ,j +urj ,i ), (3.5.7)
2
and u = ue + ur .
It may be shown that the plastic strain field εp uniquely determines the
residual stress field ρ and, with sufficient constraints to prevent rigid-body
displacement, also the residual displacement field ur . The method of proof is
analogous to that of the elastic uniqueness theorem of 3.4.1. If ρ and ρ + ρ
are two different residual stress fields, then, from the principle of virtual
work,
Z Z
−1
ρij (Cijkl ρkl + εpij ) dV = −1
(ρij + ρij )(Cijkl ρkl + εpij ) dV
R R
Z
= −1
(ρij + ρij )[Cijkl (ρkl + ρkl ) + εpij ] dV = 0.
R
Rearrangement of terms leads to
Z
−1
ρij Cijkl ρkl dV = 0,
R
and hence to ρ = 0 in view of the positive-definiteness of C−1 .

With ρ uniquely determined by εp , it follows from (3.5.7) that ur is
determined to within a rigid-body displacement.
Quasi-Static Shakedown Theorem

Suppose that an elastic-plastic body has already shaken down under a
loading that is varying in time (but sufficiently slowly so that inertia may be
neglected) within a certain range of the generalized loads. It follows from the
definition of shakedown that the plastic strain field εp remains constant in
time and defines a time-independent residual stress field ρ such that the total
stress field σ, given by (3.5.5), does not violate the yield criterion anywhere:
f (σ e + ρ) ≤ 0.
Clearly, the existence of such a residual stress field is a necessary condition

for shakedown.
It was shown by Melan [1938], however, that this is also a sufficient
condition: shakedown will occur in the given load range if a time-independent
self-stress field ρ∗ , not necessary equal to the actual residual stress field ρ,
can be found such that
f (σ e + ρ∗ ) < 0
for all elastic stress fields σ e corresponding to loadings within the given
range.
To prove the theorem, we consider the nonnegative quantity

1
Z
−1
Y = Cijkl (ρij − ρij ∗ )(ρkl − ρkl ∗ ) dV.
2 R
Since the body may not yet have shaken down, the actual residual stress
field ρ, and hence Y , may be time-dependent, with
Z
−1
Ẏ = Cijkl (ρij − ρij ∗ )ρ̇kl dV.
R
Since ρ and ρ∗ are both self-equilibrated, and since the left-hand side of
(3.5.7) forms a compatible strain field, it follows from the principle of virtual
work that Z
−1
(ρij − ρij ∗ )(Cijkl ρ̇kl + ε̇pij )dV = 0.
R
Consequently
Z Z
Ẏ = − (ρij − ρij ∗ )ε̇pij dV = − (σij − σij ∗ )ε̇pij dV,
R R
where σ∗ = σe + ρ∗ .
By hypothesis, σ∗
does not violate the yield crite-
rion, and therefore, as a result of the maximum-plastic-dissipation postulate
(3.2.4), Ẏ ≤ 0, where the equality holds only in the absence of plastic flow.
Since Y ≥ 0, the condition Ẏ = 0 must eventually be reached, and this
condition corresponds to shakedown.
An extension of the theorem to work-hardening materials is due to Man-
del [1976] (see also Mandel, Zarka, and Halphen [1977]). The yield surface
is assumed to be of the form
f (σ, ξ) = F (σ − cεp ) − k 2 (κ),
where F is a homogeneous quadratic function of its argument, and k is a

nondecreasing function of κ. A body made of such a material will shake
down if, in addition to a time-independent residual stress field ρ∗ , there
exist time-independent internal-variable fields ξ ∗ = (εp∗ , κ∗ ) such that
f (σ e + ρ∗ , ξ ∗ ) < 0
everywhere.
Kinematic Shakedown Theorem
A kinematic criterion for shakedown was derived by Koiter [1956, 1960].
A strong version of Koiter’s theorem states that shakedown has not taken
place if a kinematically admissible velocity field v∗ , satisfying v∗ = 0 on
∂Rv , can be found so that
Z Z Z
∗ ∗
f · v dV + a
t · v dS > Dp (ε̇∗ ) dV.
R ∂Rt R
This inequality, with the principle of virtual work applied to its left-hand
side, can be transformed into
Z Z
e
σij ε̇ij ∗ dV > Dp (ε̇∗ ) dV. (3.5.8)
R R
Suppose, now, that shakedown has taken place, with a time-independent

residual stress field ρ. From the maximum-plastic-dissipation postulate,
D(ε̇∗ ) ≥ (σij
e
+ ρij )ε̇ij ∗ ,
and therefore
Z Z Z
D(ε̇∗ )dV ≥ e
σij ε̇ij ∗ dV + ρij ε̇ij ∗ dV. (3.5.9)
R R R
An application of the virtual-work principle to the last integral shows that it

vanishes, since ρ is self-equilibrated and v∗ = 0 on ∂Rv . Inequalities (3.5.8)
and (3.5.9) are therefore in contradiction, that is, shakedown cannot have
taken place.
A weaker version of the theorem requires only that a strain rate ε̇∗ be
found during a time interval (0, T ) such that the strain field
Z T
∗
ε = ε̇∗ dt
0
is compatible with a displacement field u∗ that satisfies u∗ = 0 on ∂Ru , and

Z T Z Z Z T Z
∗ ∗
f · u̇ dV + a
t · u̇ dS dt > Dp (ε̇∗ ) dV dt.
0 R ∂Rt 0 R
With these conditions met, the body will not shake down during the interval.
Recent developments in shakedown theory have included taking into ac-
count the effects of temperature changes, creep, inertia, and geometric non-
linearities. For a review, see the book by König [1987]

1. Find the ultimate load FU for the structure shown in Figure 4.1.2(a)
(page 185), assuming that all the bars have the same cross-sectional
area, are made of the same elastic–perfectly plastic material with uni-
axial yield stress σY , and act in simple tension.
2. Using both the lower-bound and the upper-bound theorems, find the
limit locus for the beam having the idealized section shown in Fig-
ure 3.5.2 (page 171) subject to combined axial force P and bending
moment M .
3. In a body made of a standard Mohr–Coulomb material with cohesion

c and internal-friction angle φ under a load P , lower and upper bounds
to the ultimate load PU have been found as PU− = ch− (φ) and PU+ =
ch+ (φ). Show how the results can be used to find the best bounds on
PU if the material is nonstandard but has a plastic potential of the
same form as the yield function, with a dilatation angle ψ 6= φ.
4. For the beam of Figure 3.5.2 subject to a bending moment M only,

find the range of M within which shakedown occurs on the basis of the
following assumed self-stress distributions.
(a) ρ1 ∗ = 21 σY = −ρ01 ∗ , ρ2 ∗ = − 16 σY = −ρ02 ∗

(b) ρ1 ∗ = − 12 σY = −ρ01 ∗ , ρ2 ∗ = 16 σY = −ρ02 ∗
(c) ρ1 ∗ = 34 σY = −ρ01 ∗ , ρ2 ∗ = − 14 σY = −ρ02 ∗
5. For the beam of Figure 3.5.2 under combined axial force P and bending
moment M , (a) find the elastic stresses σ e in each flange; (b) given the
self-stress distribution ρ1 ∗ = − 35 σY , ρ2 ∗ = 54 σY , ρ01 ∗ = 15 σY , ρ02 ∗ =
− 52 σY find the range of P and M moments under which shakedown
occurs by ensuring that |ρ∗ + σ e | ≤ σY in each flange.
6. Using the result of Exercise 2 for the beam of Figure 3.5.2 under com-
bined axial force and bending moment, find, if possible, loading cycles
between pairs of points on the limit locus such that there occurs (a)
incremental plastic deformation, (b) alternating plastic deformation,
and (c) shakedown.
Chapter 4
Problems in Contained
Plastic Deformation
Section 4.1 Elementary Problems
4.1.1. Introduction: Statically Determinate Problems
There are a few static or quasi-static boundary-value problems in solid me-

chanics that are approximately statically determinate. That is, much of the
stress field may be determined from the applied loading (whether prescribed
tractions or initially unknown reactions), independently of the material prop-
erties. These problems include the following:
(a) Uniaxial tension or compression of a straight rod or tube with an axial
force P .
(b) Torsion of a thin-walled tube with a torque T .
(c) Simultaneous tension or compression and torsion of a thin-walled tube.
(d) Axially symmetric loading of a thin-walled shell of revolution.
The solution of these problems, as regards stress, is the stuff of elemen-
tary solid mechanics.
In problem (a), away from the ends the only significant stress compo-
nent is the axial stress σ = P/A, where A is the cross-sectional area, and
this stress is uniform if A is constant, that is, if the bar is prismatic. This
solution constitutes the basis of the simple tension or compression test dis-
cussed in Section 2.1; the load-elongation diagram essentially reproduces the
conventional stress-strain diagram.
Similarly, in problem (b) the only significant stress component is the
shear stress τ = T /2Ah, where A is the area enclosed by the curve tracing
the midpoints of the wall, and h is the wall thickness. This stress is constant
if the wall thickness is uniform. If the tube, moreover, is circular, with radius
177
178 Chapter 4 / Problems in Contained Plastic Deformation
c, then τ = T /2πc2 h, while the shear strain is given by γ = cφ/L, where

φ is the relative angle of twist between the ends of the tube, and L is its
length. The T -φ diagram of the tube thus reproduces the τ -γ diagram of the
material. The torsion of a thin-walled circular tube is consequently a simple
means of determining the pure shearing behavior of solids that are sufficiently
strong in tension, such as metals and hard plastics. As discussed in Section
2.3, geomaterials require different tests of shear behavior. A complication of
the torsion test on thin-walled tubes is that if the thickness is too small in
relation to the radius, the induced compressive stress may provoke buckling.
It is characteristic of these two problems that if the material is taken
as perfectly plastic, then the applied load (force or torque) cannot exceed
the value at which yielding first occurs. When this value is reached, the
body continues to deform under constant load — a process that, depending
on the field of application, is called plastic flow or plastic collapse — until
significant changes in geometry take place, or until inertial effects become
important.
The solution of problem (c) for stresses is the superposition of those of
(a) and (b). The resulting stress field forms, locally, a state of plane stress
determined by σ and τ , so that if the material yields according to either
the Mises or the Tresca criterion, the yielding is described by Equation
(3.3.5). Measurements of changes in elongation and twist angle provide a
test of the flow rule, and the movement of the yield curve in the σ-τ plane
may elucidate the hardening rule. Tests based on this problem were first
performed by Taylor and Quinney [1931].
The solution of problem (d) is particularly simple when the shell is a
sphere (or a portion of a sphere), of radius c and wall thickness h, un-
der uniform internal and external pressure. This loading produces a state
of uniform, essentially plane stress, with both principal stresses equal to
σ = pc/2h, p being the pressure difference (internal less external). Another
important case is that of a cylindrical tube under a combination of uniform
pressure and an axial force P ; in this case the principal stresses are σ1 = pc/h
(circumferential) and σ2 = pc/2h+P/2πch (axial). Since arbitrary combina-
tions of principal stresses can be produced by varying the axial force and the
pressure, this case has formed the basis of numerous experiments, beginning
with those of Lode [1925], to study yield criteria, flow rules, and hardening.
4.1.2. Thin-Walled Circular Tube in Torsion and Extension
We consider, as a special case of problem (c), a circular tube made of a rate-

independent plastic material governed by either the Mises or the Tresca yield
criterion, given in terms of σ and τ by Equation (3.3.5). We furthermore
assume isotropic hardening based on a hardening variable κ. The criterion
Section 4.1 / Elementary Problems 179
may therefore be written in the form
f (σ, τ, κ) = σ̄ − σY (κ) = 0, (4.1.1)

√
where σ̄ = σ 2 + ατ 2 , and α = (σY /τY )2 is a constant equal to 3 for Mises
˙
criterion and 4 for the Tresca criterion. Note that f inte = σ̄.
The flow rule will be assumed to be associated and hence derivable from
a plastic potential which may be identified with σ̄. The associated flow
rule is equivalent to the postulate of maximum plastic dissipation, Equation
(3.2.4), which in the present state of stress reduces to
(σ − σ ∗ )ε̇p + (τ − τ ∗ )γ̇ p >= 0
for any σ ∗ and τ ∗ such that f (σ ∗ , τ ∗ , κ) ≤ 0. Thus the plastic elongation

rate ε̇p and shearing rate γ̇ p are the strain rates conjugate to the stresses σ
and τ , respectively, and are given by
σ ατ
ε̇p = λ̇ , γ̇ p = λ̇ .
σ̄ σ̄
If the hardening variable κ is defined by
r
1 p 2
κ̇ = (ε̇p )2 + (γ̇ ) ,
α
then κ̇ = λ̇, and therefore the hardening modulus H, as defined by Equation

(3.1.1), is just σ 0 Y (κ). In view of Equation (4.1.1), H is a function of σ̄
in any plastic loading process. In particular, H(σ̄) is the slope of the σ-εp
curve obtained from a simple tension or compression test at a point where
|σ| = σ̄; it is also α1 times the slope of the τ -γ p curve obtained from the
√
simple torsion test, problem (b), at a point where |τ | = σ̄/ α.
Equations (3.2.1)–(3.2.2) for this problem thus reduce to
σ̇ σ
ε̇ = + < σ̄˙ >,
E H σ̄ (4.1.2)
τ̇ ατ
γ̇ = + < σ̄˙ > .
G H σ̄
When a stress path σ(t), τ (t) is prescribed, Equations (4.1.2) can be used
to find the corresponding strain path ε(t), γ(t) by integration, in general
numerical. An explicit form may be obtained when the stress path is radial,
that is, when σ = aσ̄ and τ = bσ̄, a and b being constants constrained by
the requirement a2 + αb2 = 1. With ε̄(σ̄) defined by
dσ̄
Z
ε̄(σ̄) =
H(σ̄)
(note that this describes the σ-εp curve in simple tension and compression),
the strains are given by
a
ε= σ̄ + aε̄(σ̄),
E
b
γ = σ̄ + αbε̄(σ̄).
G
When the strain path is given, Equations (4.1.2) form a pair of cou-
pled nonlinear differential equations for the stresses; they must in general be
solved numerically. If, however, the work-hardening is very slight (this in-
cludes the limit of perfect plasticity), then σY may be regarded, at least over
a certain range of strain, as a constant, and σ and τ are no longer indepen-
dent but are coupled through the yield criterion. The quantity σ̄/H ˙ σ̄ may
be eliminated between the two equations, resulting in the single equation
ατ σ̇ σ τ̇
− = ατ ε̇ − σ γ̇.
E G
Suppose, for example, that the tube is stretched until it just yields
(σ = σY , τ = 0); thereafter, it is twisted, with further axial deformation
prevented, that is, ε̇ = 0. After σ is eliminated, the equation becomes
dγ 1 ατ 2
= + ,
dτ G E(τY2 − τ 2 )
and may be integrated to yield

1 α τY τY + τ

γ= τ+ ln −τ .
G 2(1 + ν) 2 τY − τ
This result can also be in terms of the ratio of plastic to elastic shear strain:
γp α τY 1 + τ /τY

e
= ln −1 .
γ 2(1 + ν) 2τ 1 − τ /τY
It is clear that the shear strain, and therefore the angle of twist, grows
indefinitely as the torque approaches its ultimate value 2πc2 hτY . Unlike the
simple problems (a) and (b), however, in this case the growth is asymptotic.
Indeed, so long as the shear stress is less than, say, 99its yield value, the
plastic strain remains of the same order of magnitude as the elastic strain.
The tube can then be said to be in a state of contained plastic deformation.
In the present chapter we concentrate on problems of this type; problems in
plastic flow and collapse are considered in Chapter 5.
As the shear stress approaches the yield value, the axial stress approaches
zero, and a state of virtually pure torsion is attained. The constant axial
strain σY /E changes, in the process, from purely elastic to plastic. Since
these changes take place over a rather narrow range of strain, the validity of
the assumption of negligible work-hardening, which was necessary to obtain

the solution, may be limited to this range. It therefore follows that the τ -γ
curve for this problem will asymptotically attain the curve for simple shear.
If desired, the remaining strain-rate components ε̇r and ε̇θ may be found
through the flow rule. Since σr = σθ = 0, and therefore sr = sθ = − 12 sz , the
Mises flow rule (α = 3) implies
ε̇r = ε̇θ = − 21 ε̇z .
Since, however, σr = 0 is the intermediate principal stress, the Tresca flow

rule (α = 4) implies
ε̇r = 0, ε̇θ = −ε̇z .
A completely analogous result is produced if the twist is held constant

while extension proceeds. For this and related problems, see Hill [1950],
pp. 71–75, or Chakrabarty [1987], Section 3.3(i). Hill’s results relating to
large deformation must be taken with caution, because, as was pointed out
in 2.1.1, the use of logarithmic strains is inappropriate when the principal
strain axes rotate. Moreover, the results are based on the use of nonobjective
stress rates; this matter is discussed further in Chapter 8.
4.1.3. Thin-Walled Cylinder Under Pressure and Axial Force

In a cylindrical shell under a uniform internal pressure p and a tensile axial
force P , the principal stresses are σ1 = pc/h (circumferential) and σ2 =
1
σ + σa (axial), where σa = P/2πch. If the material is governed by the
2 1
Mises criterion with isotropic hardening, this criterion
q can once more be
written in the form (4.1.1), but this time with σ̄ = σa + 43 σ12 .
2
It is convenient to use σa and σ1 as the variables defining the stress

state, since the yield function is defined in terms of these stresses. The
associated flow rule will be assumed, and its form may again be derived
from the postulate of maximum plastic dissipation. Let εz denote the axial
and εθ the circumferential strain. Since the plastic dissipation takes the form
Dp = σ1 ε̇p1 + σ2 ε̇p2 = σ1 (ε̇pθ + 21 ε̇pz ) + σa ε̇pz ,
it follows that
σa
ε̇pz =
< σ̄˙ >,
H σ̄
where H is again the uniaxial hardening modulus, and
1 3 σ1
ε̇pθ + ε̇pz = < σ̄˙ >,
2 4 H σ̄
from which 3
σ1 − 12 σa
ε̇pθ = 4
< σ̄˙ > .
H σ̄
The equations for the total strain rates are therefore
σ̇a + ( 12 − ν)σ̇1 σa
ε̇z = + < σ̄˙ >,
E H σ̄ (4.1.3)
(1 − 12 ν)σ̇1 − ν σ̇a 3
σ1 − 12 σa
ε̇θ = + 4 < σ̄˙ > .
E H σ̄
The calculations may now proceed as in the previous example. However,

since in the present case the principal axes of strain do not rotate, there
is no difficulty in extending the results into the large-deformation range by
interpreting the strains as logarithmic. The strain rates are given by
l˙ ċ
ε̇z = , ε̇θ = .
l c
The volume enclosed by the shell is πc2 l (if h is neglected next to c), and its
change is given by
V̇ l˙ 2ċ 1 5 3σ1 σ̄˙

= + = ε̇z + 2ε̇θ = (1 − 2ν)σ̇a + − 2ν σ̇1 + .
V l c E 2 2H σ̄
When the deformations are sufficiently great, the elastic strains may be
neglected. The shell material may then be regarded as incompressible, so
that the solid volume 2πchl remains constant, and the thickness h may be
determined from this condition. Consider, for example, a closed cylindrical
tank containing gas under pressure and undergoing an axial extension at a
constant absolute temperature T . It follows from the ideal-gas law,
pV = nRT,
where n is the number of moles of gas and R is the universal gas constant
(Avogadro’s number × Boltzmann’s constant), that plc2 is constant. But
pc
plc2 = lch = σ1 lch,
h
and therefore σ1 is also a constant, given by the initial value of pc/h. We
may now write
V̇ 1 − 2ν 3σ1 σ̄˙
= σ̇a + .
V E 2H σ̄
We assume, first, linear work-hardening (H = constant) after an initial
yield stress σE . We further suppose that the √ gas pressure alone is not suf-
ficient to cause yielding, that is, σ1 < 2σE / 3. We may then integrate the
preceding equation to obtain
s
V 1 − 2ν 3σ1 σa2 + 34 σ12
ln = σa + < ln >.
V0 E 2H σE
If, on the other hand, the work-hardening is described by the Ramberg–

Osgood equation (2.1.2), then
 !(m−1)/2 √ !m 
2 3 22
V 1 − 2ν 3mσ1  σa + 4 σ1 3σ1
ln = σa + − − 1 .
V0 E 2(m − 1)E σ R 2σR
Note that the elastic part of the volumetric strain has the same sign as σa ,
but the plastic part is never negative, whether the tank undergoes extension
or compression. Note also that the expression for the volume becomes infinite
in the limit as H → 0 for the linear work-hardening case and as m → ∞ for
the Ramberg–Osgood case; both of these limits represent perfect plasticity.
In this example, then, contained plastic deformation cannot occur if the
material is perfectly plastic. The difference between the present problem
and the preceding one does not lie in the fact that one involves torsion and
extension, and the other extension and gas pressure. Rather, the crucial
difference is that the twisted tube was subject to the kinematic constraint
of no axial extension; it is this constraint that prevented unrestricted plastic
deformation immediately upon yielding. Had the axial force, rather than
the length, been held constant, contained plastic deformation would not
have taken place.
In the case of linear work-hardening an explicit expression can also be
found for the axial extension and the radial expansion. The former is gov-
erned by
l˙ σ̇a σ 2 σ̇a
= + 2 a3 2 , (4.1.4)
l E (σa + 4 σ1 )H
which may be integrated to give
r " √ #
l σa 1 q 3 2σ a 3σ 1
ln = + <σa − σY2 − 43 σ12 − σ1 tan−1 √ − cos−1 >.
l0 E H 2 3σ1 2σY
The axial force at a given value of l may then be found from
l0
P = 2πc0 h0 σa . (4.1.5)
l
A similar integration for the radius change leads to
s
c νσa 3σ1 σa2 + 34 σ12
ln = − + < ln >
c0 E 4H σE r " √ #
1 q 3 −1 2σ a −1 3 σ 1
− <σa − σY2 − 34 σ12 − σ1 tan √ − cos >.
2H 2 3 σ1 2σY
The force P given by Equation (4.1.5) becomes stationary (Ṗ = 0),

indicating an instability, when
l˙ σ̇a
= .
l σa
Equating the right-hand side of this equation to that of (4.1.4), and assuming
that H E so that the elastic strain rate can be neglected, we find a
critical value of the hardening modulus above which instability will not occur,
namely
σ3
Hc r = a2 .
σ̄
Note that this is positive only when σa > 0, that is, when the axial force
is tensile; when this force is compressive, any nonnegative value of H is
sufficient for stability. Note further that, in the absence of pressure, Hc r =
σa ; this is just the necking condition discussed in 2.1.2. The pressure tends
to lower the critical value of the hardening modulus, so that it is a stabilizing
factor.
Different values of Hc r are found for problems in which the ratio σa /σ1
or the ratio P/p is held fixed (see, e.g., Chakrabarty [1987], Section 3.3(iii)].
4.1.4. Statically Indeterminate Problems

In contrast to the statically determinate problems just discussed, in statically
indeterminate problems contained plastic deformation occurs even without
kinematic constraints. Consider the simple pin-connected three-bar assem-
blage shown in Figure 4.1.1(a). Since a pin joint cannot transfer moments,
the bars carry axial forces only and each bar can be assumed to be in a state
of uniaxial tension or compression, the stress being σ = P/A, where P is the
axial force in the bar. If all three bars have the same cross-sectional area
A and are made of the same elastic–perfectly plastic material, with Young’s
modulus E and yield stress σY , the elastic solution for the bar forces P0 and
P1 ( = P10 ) is
F F cos2 θ
P0 = , P1 = ,
1 + 2 cos3 θ 1 + 2 cos3 θ
def
and is valid while P0 ≤ PU (where PU = σY A), or F ≤ (1+2 cos3 θ)PU = FE ;
here FE denotes, as usual, the elastic-limit value of the load F . When
def
F = FE , the downward displacement ∆ of the joint equals σY L/E = ∆E .
The range 0 < F < FE , or 0 < ∆ < ∆E , represents the elastic range.
When F > FE , P0 remains constant at PU , and P1 may be determined by
statics alone to be 21 (F − PU ) sec θ, provided that this value does not reach
def
PU . The latter condition occurs when F = (1 + 2 cos θ)PU = FU . In the
range FE < F < FU , the displacement ∆ is determined by the deformation
of bars 1 and 10 alone, a deformation that remains purely elastic. It is this
elastic deformation that also determines the plastic deformation of bar 0.
The range in question is thus the range of contained plastic deformation.
Contained plastic deformation ends when ∆ = ∆E sec2 θ. Beyond this
value, the displacement grows indefinitely while F remains at the value FU ,
sf s
f s
f
FU
TT 10 0 1
TT FE

TT +
θ s
- θ

TT
TT
TT f
s

∆
?
F ∆E ∆E sec2 θ
(a) (b)
Figure 4.1.1. Three-bar assemblage: (a) structure; (b) load-deflection dia-

gram.
precisely the definition of plastic collapse given in Section 3.4, and the load
FU is the ultimate load of the structure. The complete F -∆ diagram is
shown in Figure 4.1.1(b).
It is not difficult to see that if the number of bars in the assemblage is
multiplied, as in Figure 4.1.2(a), the load-deflection diagram has the form
shown in Figure 4.1.2(b). The range of contained plastic deformation is now
represented by numerous segments, whose initial and final points must be
calculated step by step.
f
s.......................................... f s f s f s sf sf ...........................................
f
s FU (
ll.................n........0.................................................A...A.. 20 CC 10 0 1 2............................................................................., , n
ll...................................................A. CC z ...................................................., .
,
l.......................................A........... C Y
HjH ...........................
. . . . . . . ,
l
l...........................A...A....... C C Q
k ....................H ..................
.... .,
,
...............,
.... H θ1
l ,
......w
Q
l l...............A......A.... C C ..@ I .

.. . .... . Q , FE
l......A...A...A.C C ..........,
l , .
@ Q θ2
l l l.A.C f .
s,
,, @θ
l , n

∆
?
F ∆E
(a) (b)
Figure 4.1.2. Multi-bar assemblage: (a) structure; (b) load-deflection diagram.
It can be shown that in the elastic range, the bar forces (if all bar areas
are equal) are given by
F F cos2 θk
P0 = n , Pk = n . (4.1.6)
X X
3 3
1+2 cos θk 1+2 cos θk
k=1 k=1
Note that the bar forces decrease with distance from the central bar. It can
thus be assumed that the order of yielding of the bars is outward from bar
0. If bar 0 and m − 1 bars on either side of it have yielded (so that Pk = PU ,
k = 0, ..., m − 1), then it can be shown that for k = m, ..., n,
m−1
" !#
cos2 θk X
Pk = n F − PU 1+2 cos θl , (4.1.7)
X
3 l=1
2 cos θl
l=m
provided that Pm ≤ PU , that is,

m−1 n
!
X 2 X
F ≤ PU 1+2 cos θk + 2θ
cos3 θk .
k=1
cos m k=m
When all but the last pair of bars have yielded, that is, when m = n, the
upper bound on F is just the ultimate load FU , given by
n
!
X
FU = PU 1+2 cos θk .
k=1
Of course, as we learned in 3.4.1, this value could easily have been determined
without going through the step-by-step process.
If, on the other hand, the structure were reduced to bar 0 only, then it
would be statically determinate and naturally there would be no contained
plastic deformation; the elastic range would be followed immediately by
unrestricted plastic flow. Contained plastic deformation, therefore, is due
either to constraints or to the introduction of redundant elements that make
the structure statically indeterminate.
A thin-walled tube under torsion or pressure forms, as we have seen, a
statically determinate problem. A thick-walled tube may be thought of as
the limit of an assemblage of a large number of concentric thin-walled tubes,
all but one of which are redundant. The range of contained plastic defor-
mation is therefore characterized by a torque-twist or pressure-expansion
diagram which is the limit of that of Figure 4.1.2(b) as the segments become
infinitesimal — that is, a smooth curve in which the ultimate load may be
attained asymptotically, as in the problem of 4.1.2 [see Figure 4.1.3(a)] or
at a finite displacement, as in Figure 4.1.3(b).
Elastic-Plastic Boundary
At any stage of contained plastic deformation in the problem of Figure
4.1.2, the spaces between the plastic and elastic groups of bars may be said to
form the elastic-plastic boundary. As the load is increased and an additional
pair of bars yields, the boundary moves to the next pair of spaces.
In a relatively simple problem such as this one, the elastic-plastic bound-
ary is defined by a single parameter, for example the integer m. Problems
T or p T or p

φ or u φ or u
(a) (b)
Figure 4.1.3. General load-displacement diagrams: (a) ultimate load reached

asymptotically; (b) ultimate load reached at a finite displace-
ment.
with radial symmetry, as shown in the next two sections, are similarly sim-
ple: the elastic and plastic zones can be separated only by a spherical or a
cylindrical surface, and the boundary is determined by its radius. It is the
simplicity of the elastic-plastic boundary that makes problems of this nature
tractable analytically.
In general, the elastic-plastic boundary is one of the unknown quantities
of a problem, and when it is not defined by a small number of parameters
then it can only be determined by trial and error, step by step, in the course
of an incremental solution. Typically, the given loading is scaled down until it
is small enough so that the whole body is elastic. Since the elastic problem
is linear, the stresses are directly proportional to the load (in the case of
one-parameter loading). With the help of the elastic solution it is therefore
possible to find the value of the load at which the body first yields, such as
FE in the preceding example, as well as the point or points where yielding
first occurs. The elastic-plastic boundary originates at these points.
Following the initiation of yielding, the loading must be increased by
small increments, and a trial solution is found for a statically admissible
stress-increment field dσ. The resulting stress f atσ must then be tested to
determine whether it is plastically admissible for the current yield surface.
If it is not, then the trial solution must be corrected so as to bring all stress
values inside or on the yield surface; the procedure is usually an iterative
one. The points at which σ is inside the yield surface constitute the current
elastic domain, while those where σ is on the yield surface form the plastic
domain. The elastic-plastic boundary is thus a part of the solution at each
step. The solution must usually be carried out numerically, and methods
are discussed in Section 4.5. Sections 4.2, 4.3, and 4.4 deal with problems
that can be treated analytically.
Another feature of the problem of Figure 4.1.2, as well as of thick-walled

tubes under pressure or torsion, is that when the ultimate load is reached,
the entire body is plastic; in other words, the elastic-plastic boundary passes
out of the body. In the case of our bar assemblage, this is a result of the
symmetry of the problem, with respect to both geometry and loading. In the
absence of such symmetry it is not necessary for all the bars of an assemblage
of this type to be plastic at collapse: one bar may remain elastic, since it
can rotate rigidly as the other bars deform. There are other problems as
well in which an elastic-plastic boundary is present at collapse. An example
is the cantilever beam studied in Section 4.4, and other examples are given
in Section 5.1.
1. Solve Equations (4.1.2) for the stress-strain relation in tension when

the tube is twisted until it just yields, and thereafter it is stretched
with further twisting prevented. Assume negligible work-hardening.
2. If the load on the assemblage of Figure 4.1.1(a) is not vertical but has
both a vertical component V directed downward and a horizontal load
H directed to the right, the problem is no longer symmetric and the
three bar forces must be assumed to be distinct. Discuss the behavior
of the assemblage when (a) V = cH, with c constant and H increasing,
and (b) V is held constant at a value at which the assemblage is elastic,
and then H is gradually increased from zero.
3. Suppose the assemblage of Figure 4.1.1(a) to carry no load but to be

subject to a monotonically increasing temperature change, the temper-
ature being the same in all bars. If the Young’s modulus E, the thermal
expansion coefficient α, and the yield stress σY (equal in tension and
compression) are independent of temperature, find the temperature
change ∆T at which yielding begins.
4. Do Exercise 3 when σY is a linearly decreasing function of temperature;

that is, σY = σY 0 − a(T − T0 ), where a is a constant and T0 is the
initial temperature.
5. Show that Equations (4.1.6) describe the elastic state of the assemblage
of Figure 4.1.2(a).
6. Show that Equations (4.1.7) give the forces in the bars of the assem-
blage of Figure 4.1.2(a) that have remained elastic when the central
bar and m − 1 bars on either side of it have yielded. Find the value
attained by the load F so that bars m are just about to yield.
Section 4.2 / Elastic–Plastic Torsion 189
7. When the load F has attained the value found in Exercise 6 and is
then removed, find the residual bar forces. Which of the bars go into
compression?
Section 4.2 Elastic-Plastic Torsion

4.2.1. The Torsion Problem
The twisting of prismatic bars under the action of equal and opposite torques
applied at the ends is one of the classical topics of solid mechanics. For elastic
bars of circular cross-section, the relation between torque and twist was first
derived by Coulomb in 1784. A general theory for bars of arbitrary cross-
section was published by Saint-Venant in 1855; this theory makes it possible
to predict where the stress is maximum, and therefore where, and at what
values of the torque and twist, yielding will first occur.
Saint-Venant’s theory is based on the semi-inverse method originated
by him. An inverse method is one in which a solution (a stress field or a dis-
placement field) is assumed, while the boundary conditions that it satisfies,
and in particular the resulting surface tractions, are determined afterwards.
In a semi-inverse method, a part of the solution is assumed, and the rest is
left to be determined by solving appropriate differential equations. In the
torsion problem, assumptions can be made on both the stress and displace-
ment fields. It is important to note that the value of the torque is not a part
of the boundary data; instead, it is the angle of twist that is prescribed, and
the torque results as a reaction to the imposed twist, not a load.
The Boundary-Value Problem

We consider a prismatic bar whose longitudinal axis (e.g., the centroidal
axis) is the x3 -axis and whose cross-section (assumed simply connected) is
defined by a closed curve C in the x1 x2 -plane (see Figure 4.2.1). The bar is
defined as being in a state of torsion or twist if one end section, viewed in
its plane (i.e., with any warping ignored), rotates rigidly with respect to the
other. The end sections are assumed to be free to warp, so that there is no
normal traction on them. The lateral surface is traction-free. Consequently,
if the rotation of the section at x3 = 0 is defined as zero and if the relative
angle of rotation between the two end sections is θL, then the boundary
conditions are
at x3 = 0 : u1 = u2 = 0, σ33 = 0;
at x3 = L : u1 = −θLx2 , u2 = θLx1 , σ33 = 0;
dx2 dx1
on C, 0 < x3 < L : Ti = σij nj = σiα nα = 0, where n1 = , n2 = − .
ds ds
x2
6
H
HH
C H
H
HH - x1
H H
HH
HH H H HH
HH H
Y
H H
HH HH H HH
HH H H
H
L HH HH
H HH
HH H j x3
H
H
j
Figure 4.2.1. Prismatic shaft: geometry.
Stress Assumptions
If we assume that σ11 = σ22 = σ33 = σ12 = 0, then the equilibrium
equations (with zero body force) reduce to
τα ,3 = 0 (α = 1, 2), τα ,α = 0,
where τα = σα3 . The first two equations imply that τ1 and τ2 are func-
tions of x1 and x2 only, so that the problem is mathematically (though not
physically) two-dimensional. The third equation implies that there exists a
function φ(x1 , x2 ) such that τ1 = φ,2 and τ2 = −φ,1 . To satisfy the traction
boundary conditions, we need

dxα dφ
τα nα = φ,α = = 0,
ds C ds C
so that φ must be constant on C. With no loss of generality we may let the
constant be 0. If the section is multiply connected, with interior contours C1 ,
C2 ,..., then we can set φ = 0 on the exterior boundary curve C, but on each
of the interior curves we can only set it equal to some unknown constant.
The torque T is given by
Z Z
T = (x1 τ2 − x2 τ1 ) dA = − xα φ,α dA
AZ AI Z
= − [(xα φ),α −2φ] dA = − nα xα φ dA + 2 φ dA
Z A C A
= 2 φ dA,
A
since φ vanishes on C. We may think of the surface formed by the stress

function φ(x1 , x2 ) as a “tent” pitched over the base formed by the boundary
curve C; then the torque is twice the volume enclosed by the tent.
The contour lines of the tent (curves along which φ has a constant value)
are also significant: if at a given point of the cross-section, a local Cartesian
coordinate system is established so that the x1 -axis is parallel to the contour,

then locally τ1 = φ,2 and τ2 = 0. Thus every point of the shaft is stresswise
in a state of simple shear, and the shear stress is everywhere directed along
the contour, while its magnitude is just the slope of the tent surface. If the
shaft is made of an elastic-plastic material, then the yield criterion may be
expressed in terms of the stress function φ as |∇φ| = k, where k = τY is the
yield stress in shear.
Displacement Assumptions
If the material is isotropic, then regardless of whether it is elastic, plastic,
or whatever, the only nonvanishing strain components (under the assumption
of infinitesimal deformation) consistent with the stress assumptions can be
def def
the shear strains γ1 = 2ε13 and γ2 = 2ε23 . A displacement field consistent
with this condition and with the displacement boundary conditions is
u1 = −θx2 x3 , u2 = θx1 x3 , u3 = θψ(x1 , x2 ),
where ψ is known as the warping function; the factor θ is placed in front of

it for convenience. Note that θ, which is simply called the twist, is the angle
of twist per unit length, assumed uniform. The dependence of u3 on θ is not
linear in all situations, but it is so in many of the ones that we are going to
study. The strain components are
γ1 = θ(ψ,1 −x2 ), γ2 = θ(ψ,2 +x1 ).
4.2.2. Elastic Torsion

The isotropic linear elastic stress-strain relations τα = Gγα translate to
τ1 = φ,2 = Gθ(ψ,1 −x2 ),

(4.2.1)
τ2 = −φ,1 = Gθ(ψ,2 +x1 ).
These are two equations in the two unknown functions φ and ψ. We may
eliminate either one and be left with one differential equation in one unknown
function.
The Stress Method

If we eliminate ψ, so that φ is the unknown of the problem, we obtain
∇2 φ = −2Gθ, (4.2.2)
which, together with the boundary condition φ = 0 on C, constitutes a stan-

dard mathematical problem. The conventional method of solution consists
of, first, finding a particular solution φ1 which satisfies the differential equa-
tion but not necessarily the boundary conditions, and next, assuming that
φ = φ1 + φ2 , so that φ2 is the solution of the differential equation ∇2 φ2 = 0

(the Laplace equation) with the boundary condition φ2 = −φ1 on C — a
classical problem known as the Dirichlet problem.
Equation (4.2.2) together with the boundary condition is mathematically
identical with the problem of the deflection of a membrane, tightly stretched
in a frame occupying the contour C by an isotropic tension H and subject
to a transverse pressure p; if this deflection is denoted w(x1 , x2 ), then the
equation governing it is
p
∇2 w = − ,
H
and the correspondence, first discussed by Prandtl [1903], is known as the
membrane or soap-film analogy. With appropriate numerical scaling, it is
possible to use the results of experiments on soap films in order to calculate
the torsional properties of shafts (see, e.g., Timoshenko and Goodier [1970],
pp. 324–325]. Even without actual experiments, the analogy is useful in
helping to visualize the nature of the solution. In particular, it is often
intuitively obvious where the slope of the membrane deflection, and hence
the stress, is maximum. Typically, this maximum occurs at the points on
the exterior boundary nearest the axis.
For some cross-sections, such as an ellipse and an equilateral triangle,
a complete solution φ can be found in closed form. Using the conventional
notation x, y in place of x1 , x2 , we may write the solution for an ellipse
described by (x/a)2 + (y/b)2 = 1 as
!
a2 b2 x2 y 2
φ = Gθ 2 1− 2 − 2 .
a + b2 a b
The torque on a shaft made of an isotropic, linearly elastic material with
shear modulus G always has the form T = GJθ, where J has dimensions of
length4 and depends on the geometry of the cross-section alone; the product
GJ is known as the torsional stiffness.
For the ellipse, we have J = πa3 b3 /(a2 + b2 ). The maximum stress is
2T
τmax = max(a, b).
πa2 b2
The circle is, of course, a special case of the ellipse with a = b.
For the equilateral triangle, a stress function having the form
φ(x, y) = Cf1 (x, y)f2 (x, y)f3 (x, y),
where
f1 (x, y) = 0, f2 (x, y) = 0, f3 (x, y) = 0
are the equations of the three lines forming the triangle, obviously satisfies
the boundary condition φ = 0, and can easily be shown to satisfy the differ-
ential equation (4.2.2) with a suitable value of the constant C. If c denotes
√
the side of the triangle, then the torsional stiffness is 3Gc4 /80, and the
maximum shear stress is τmax = 20T /c3 .
Rectangular Shaft
A closed-form solution does not exist for any other polygonal shaft. A
Fourier series solution can be obtained for a rectangular shaft. Consider the
rectangle bounded by x = + − a/2, y = + − b/2, with b >= a. If we assume
that !
a2
φ1 = Gθ − x2 ,
4
then this satisfies the boundary condition on the longer sides but not on the
shorter sides; it is reasonable to expect that the larger b/a, the closer φ1 will
be to φ. The “correction” φ2 may be assumed in the form of a Fourier series
in x that is even in x and vanishes on x = + − a/2, and whose coefficients
depend on y:
∞
X (2m + 1)πx
φ2 = Gθ fm (y) cos .
m=0
a
Now
∞
" #
2
X
00 (2m + 1)2 π 2 (2m + 1)πx
∇ φ2 = Gθ fm (y) − 2
fm (y) cos ,
m=0
a a
00 = [(2m+1)π/a]2 f ; the solution
and if this is to vanish for all (x, y), then fm m
that is even in y is fm (y) = Am cosh(2m + 1)πy/a, where the coefficients Am
must be such that φ2 = −φ1 on y = + − b/2. To satisfy this condition we
need the appropriate Fourier series expansion for φ1 , that is, we need
∞
a2 X (2m + 1)πx
− x2 = Bm cos ;
4 m=0
a
then the solution is complete if Am cosh [(2m + 1)πb/2a] = −Bm . In accor-

dance with the theory of Fourier series,
!
4 a/2 a2 (2m + 1)πx 8(−1)m a2
Z
Bm = − x2 cos dx = .
a 0 4 a (2m + 1)3 π 3
Hence, finally,
∞
8a2 X (−1)m cosh (2m + 1)πy/a (2m + 1)πx
φ2 = −Gθ cos .
π m=0 (2m + 1)3 cosh (2m + 1)πb/2a
3 a
Integration leads to
∞
" #
31 64a X 1 (2m + 1)πb
J =a b − 5 5
tanh .
3 π b m=0 (2m + 1) 2a
Since the hyperbolic tangent function approaches unity as its argument in-
creases, the series converges very rapidly.
The maximum shear stress occurs at (± 12 a, 0) and is given by
∞
" #
8 X 1 (2m + 1)πb
τmax = Gaθ 1 − 2 2
sech .
π m=0 (2m + 1) 2a
It is seen from the formula for J that as b/a → ∞, J → a3 b/3, which

is the result based on φ1 alone. For such infinitely narrow rectangles we
also obtain τmax = Gθa = T a/J. These results are often applied to the
torsion of shafts of open thin-walled section made up of several rectangles
(possibly even curved) of dimensions ai , bi (i = 1, ..., n) with bi ai . For
such sections we have the approximations
n
. 1X 3 . T amax
J= a bi , τmax = .
3 i=1 i J
The Displacement Method

If we choose the warping function ψ rather than the stress function φ as
the unknown function of the problem, the governing differential equation is
the Laplace equation
∇2 ψ = 0.
The traction boundary conditions in terms of ψ are
∂ψ dy dx
n1 (ψ,1 −y) + n2 (ψ,2 +x) = −y −x = 0,
∂ν ds ds
or
∂ψ 1 d 2
= r (r2 = x2 + y 2 ).
∂ν 2 ds
This result makes it clear that the warping vanishes if and only if the cross-
section is circular. For an ellipse, we find that ψ = [(b2 −a2 )/(b2 +a2 )]xy is an
exact solution of the problem, but in general this problem is more difficult to
work with than the one with φ as the unknown. Moreover, as is shown next,
it is only with φ that a boundary-value problem for elastic-plastic shafts can
be formulated.
4.2.3. Plastic Torsion

It is clear that in the elastic range the stress function is of the form φ(x1 , x2 ) =
Gθ × f (x1 , x2 ), where f (x1 , x2 ) is determined by the cross-section geometry
alone. As the twist θ is increased, the stresses grow proportionately, until at
one or more points of the cross-section the yield criterion is met; these are
the points — necessarily located on C — where |∇f | attains its maximum
value, and the value of θ when this occurs is θE = τE /(G|∇f |max ), where
τE is the elastic-limit shear stress. As the twist is further increased beyond
θE , one or more plastic regions form and expand, while the elastic regions
shrink.
If the shaft material is elastic–perfectly plastic, with τE = τY = k,
then the equation satisfied by the stress function in the plastic regions is
|∇φ| = k. Eventually this equation becomes valid everywhere, a condition
known as fully plastic torsion. In a solid (simply connected) shaft, this state
is attained in the limit as θ → ∞, while in a hollow (multiply connected)
shaft it may occur at a finite value of θ.
The torsion of elastic-plastic shafts with work-hardening has generally
been studied by means of deformation theory, it being argued that under
monotonic loading such a treatment would provide a good approximation to
the solutions of the incremental theory,1 although the two coincide exactly
only when the shaft is circular. A formulation according to Prandtl–Reuss
theory is due to Prager [1947]. In either case, the governing equations must
be solved numerically.
Fully Plastic Torsion

The mathematical problem given by
|∇φ| = k in A, φ = 0 on C, (4.2.3)
has a unique solution (Ting [1966a]), which will be denoted φp . The solution
describes a roof of constant slope, and φp (x1 , x2 ) is simply k times the
distance from (x1 , x2 ) to the nearest point on C. A point (x1 , x2 ) is a ridge
point whenever there is more than one such nearest point. The ridge points
represent the remnants of the elastic zone as it shrinks to zero area, so that
the plastic strain rates vanish there. At a ridge point, therefore, γ̇1 = γ̇2 = 0
even in the limit as θ → ∞.
A line consisting of ridge points is called a ridge line. Since a ridge point
is the meeting place of contours having different directions, a ridge line is the
locus of a discontinuity in ∇φ, and therefore the stress is discontinuous across
the plane formed by the x3 -axis and the ridge line. A stress discontinuity
does not, by itself, violate equilibrium: consider an element ∆S of a surface
perpendicular to the page and a very thin volume element containing it,
as shown in Figure 4.2.2. In the limit as the thickness goes to zero, the
element is in equilibrium if the traction is continuous, that is, if the stress
components σn (the normal stress) and τ (the shear stress) are continuous;
the lateral stress σt , on the other hand, may be discontinuous. Now, if n is
a unit vector normal to a ridge line, then the traction on the plane normal
1
See, e.g., Mendelson [1968], Section 11-6; Kachanov [1971], Section 30; Chakrabarty
[1987], Section 3.6(vi).
t(n)
6
@
τ @ σn
@
@ n
@@
I
@ @@
@ @@ ∆S

@ @@@
@@
@ @
@
R@R
? σt (may be discontinuous)
t(−n)
Figure 4.2.2. Stress discontinuity
to n is e3 (n1 τ1 + n2 τ2 ), but n1 τ1 + n2 τ2 is just the component of ∇φ parallel

to the ridge line, and this is continuous if the ridge line bisects the angle
between the two φ contours that meet there.
An experimental method for finding the solution is furnished by the
sand-heap analogy: if sand is piled onto a horizontal table having the shape
of the cross-section, the slope of the resulting heap cannot exceed the angle
of internal friction of the sand, and the maximal heap is formed when the
critical slope obtains everywhere except at a vertex or on ridge lines; any
additional sand poured will just slide off.
If C is a regular polygon, then the φp surface takes the form of a pyramid.
The limit of a regular polygon as the number of sides becomes infinite is a
circle; the corresponding surface is a cone. For any other convex polygon,
it is formed by ridge lines which bisect the angles between nearest sides, as
seen in Figure 4.2.3.

A
A A A
A A A
A A A
A A
A
Figure 4.2.3. Ridge lines and contours in a convex polygon.
R
The torque given by 2 A φp dA is the ultimate torque TU . Thus, for a
circle of radius a, TU = 32 πka3 . For a rectangle of sides a and b, with b > a,
TU = 61 ka2 (3b − a).
In a nonconvex polygon, that is, one with re-entrant corners, the ridge
lines and contours are not all straight. Consider the cross-section shown in
Figure 4.2.4. For a point (x, y) with x > y > 0, the nearest point on the
@
@ Ridge lines
@
@

r(x,y) r(a,y)

XXX
y
X Contour lines
x
@
@
@
@
a -
Figure 4.2.4. Ridge lines and contours in a nonconvex polygon.
side with the re-entrant corner is just the corner, while the pnearest point on
the other side is (a, y); (x, y) is therefore a ridge point if x2 + y 2 = a − x,
that is, if it lies on the parabola described by
y 2 = a2 − 2ax.
To the left of this parabola, the φp contours are circular arcs centered about
the corner, and to the right they are straight lines parallel to the y-axis. The
structure is, of course, symmetric about the line y = x.
Contained Plastic Torsion

For values of the twist θ such that θE < θ < ∞, the cross-sectional area
A consists of elastic and plastic regions, with Equation (4.2.2) governing
the stress function φ in the former and Equation (4.2.3) in the latter. The
boundary of each region consists of those parts of C (if any) that belong to it,
and of the elastic-plastic boundary Γ, which is in general unknown and whose
determination, as discussed in Section 4.1, constitutes a part of the problem.
Moreover, values of φ on Γ are unknown. However, Equation (4.2.2) is
an elliptic partial differential equation, and such an equation requires the
specification of boundary data on a known, closed boundary. Consequently,
the solution in the elastic region cannot be constructed. Equation (4.2.3),
on the other hand, is parabolic, and its solution can be obtained by working
inward from the plastic part of C. This solution is therefore the same as the
solution for the fully plastic cross-section, provided that the plastic region
lies entirely between the normals to C at the points where Γ meets C (see
Prager and Hodge [1951], Section 3.2); this result was proved by T. W. Ting
[1966a]. We are thus led to the membrane-roof analogy due to Nadai [1923]:
we imagine a rigid roof having the shape of the surface of the ultimate sand
heap, and a membrane stretched underneath it and subjected to a uniform
upward pressure. Under sufficiently low pressures the membrane does not
touch the roof, a state corresponding to elastic torsion. As the pressure is
increased, parts of the membrane become tangent to the roof and remain
there, representing the plastic regions, while those parts of the membrane
that are still free represent the elastic region. The membrane-roof analogy
was used by Nadai [1950] for the experimental solution of elastic-plastic
torsion problems.
It is clear from the analogy that the elastic region consists of those points
where φ < φp and |∇φ| < k. It was shown by Ting [1966b] that the correct
solution is the one that minimizes the functional
Z
def
K[φ∗ ] = (|∇φ∗ |2 − 4Gθφ∗ ) dA,
A
the minimum being taken either over all those functions φ∗ that satisfy
φ∗ ≤ φp or those that satisfy |∇φ∗ | ≤ k, in addition to satisfying φ∗ = 0
on C. The properties of the solution for various cross-sections were further
investigated in a series of papers by T. W. Ting [1967, 1969a,b, 1971]. Ting’s
minimum principle also permits the application of numerical methods based
on discretization, similar to that discussed for elastic bodies in 1.4.3, except
for the presence of inequality constraints.
Elastic–Plastic Warping
An important feature of the elastic-plastic solution is that once a point of
the cross-section has become plastic, the stress state there remains constant.
Any strain increment in the plastic region is therefore purely plastic. The
flow rule γ̇αp = λ̇τα , which follows from isotropy, implies that in the plastic
zone,
γ̇1 ψ,1 −x2 φ,2
= =− .
γ̇2 ψ,2 +x1 φ,1
Let the normal to the boundary subtend an angle χ with the x1 -axis; then,
in the plastic zone,
φ,1 = −k cos χ, φ,2 = −k sin χ.
Consequently,
ψ,1 cos χ + ψ,2 sin χ = x2 cos χ − x1 sin χ,
or
∂ψ
= d,
∂n
where n denotes the directed distance, measured outward, along a line nor-
mal to the boundary, and d denotes the distance from this line to the origin
(see Figure 4.2.5). If n is measured from the intersection point R of the
x2

χ
*Ar K?

n
r P
r

R0 A C
R
Γ

K
A
d 0 x1
AU
Figure 4.2.5. Elastic-plastic warping: illustration of the notation used in the

theory.
given normal line with the elastic-plastic boundary, then at a point P on the
line such that n is the distance from R to P ,
ψ(P ) = ψ(R) + nd.
To obtain ψ(R), we solve Equations (4.2.1) for the ψ,α . Along Γ,
1
dψ = ψ,α dxα = x2 dx1 − x1 dx2 + τα dxα ,
Gθ
so that Z R Z R
ψ(R) = (x2 dx1 − x1 dx2 ) + τα dxα ,
R0 R0
where R0 is a point on Γ such that ψ(R0 ) = 0 — for example, the intersection

of Γ with a line of symmetry, if such a line exists, or else an arbitrary point.
Warping of a Fully Plastic Rectangular Shaft

The warping of a fully plastic rectangular shaft can be obtained more
simply. Because of symmetry, it is sufficient to consider only the quadrant
0 < x < 21 a, 0 < y < 12 b, shown in Figure 4.2.6, which we divide into the
regions I and II, respectively above and below the ridge line whose equation
is y = [(b − a)/2] + x. In region I, τ2 = 0 and therefore
γ̇2 = θ̇(ψ,2 +x) = 0.
The general solution of the partial differential equation ψ,2 +x = 0 is

y
b
2
I
Ridge line
II
b−a
2
a x
0
2
Figure 4.2.6. Fully plastic rectangular shaft: first quadrant, regions separated
by ridge line.
ψ(x, y) = −xy + f (x), f being an arbitrary continuously differentiable func-

tion. Symmetry requires zero warping on the y-axis, so that f (0) = 0. The
nonvanishing shearing-rate component is γ̇1 = θ̇[−2y + f 0 (x)], but since the
shearing rate must vanish on the ridge line, it follows that f 0 (x) = b − a + 2x
(so that γ̇1 = θ̇[b − a + 2(x − y)], and, consequently, f (x) = (b − a)x + x2 .
In region II the warping function is governed by ψ,1 −y = 0, so that
ψ = xy + g(y) with γ̇2 = θ̇[2x + g 0 (y)]. For this to vanish on the ridge
line, we must have g 0 (y) = b − a − 2y for y > 21 (b − a), and if we integrate
g 0 in the form g(y) = −[y − 21 (b − a)]2 , then the warping function is also
continuous on the ridge line. For y < 12 (b − a), however, the vanishing of
ψ on the y-axis requires g(y) = 0, and therefore the complete solution for
g is g(y) = −<y − 21 (b − a)>2 , < · > being the Macauley bracket, and the
shearing rate is γ̇2 = 2θ̇[(x − <y − 12 (b − a)>].
The results may be used to obtain the value of the ultimate torque by
means of the upper-bound theorem (Section 3.4). The plastic dissipation is
 a
 2k|θ̇|x, 0<y <b− ,
2



b−a a a


Dp (ε̇) = 2k|θ̇| x − y + , b − < y < b − + x,
 2 2 2
b−a a b


2k|θ̇| y − x − , b− +x<y < .



2 2 2
The relation Z
TU θ̇ = Dp (ε̇) dA
A
yields the previously obtained ultimate torque 61 ka2 (3b − a).
Circular Shaft
For a circular shaft, the derivation of the elastic-plastic solution is simple.
By symmetry, the elastic-plastic boundary must be a circle of radius, say,
r∗ . The elastic region is thus r < r∗ and the plastic region r > r∗ . In the
latter region we have the fully plastic solution, φ = k(c − r) (where c is the
shaft radius), while in the former,
1 d dφ

2
∇ φ= r = −2Gθ.
r dr dr
Integrating once, we obtain
dφ
− = τzθ = τ = Gθr
dr
in the elastic region; continuity of stress at the elastic-plastic boundary,
where τ = k, requires r∗ = k/Gθ; thus θE = k/Gc. Integrating again, we
find that for r < r∗ , !
r∗ r2
φ(r) = k c − − ∗ ,
2 2r
and the torque is
" 3 #
2π 1 θE

T = k(c3 − r∗ 3 /4) = TU 1 − , (4.2.4)
3 4 θ
where TU = 2πkc3 /3 as before. The torque-twist diagram is therefore one

in which the ultimate torque is approached asymptotically as θ → ∞ [Fig-
ure 4.2.7(a)]. As usual in problems of contained plastic deformation, the
approach is quite rapid: T /TU = 0.99 when θ/θE = 3.
T T
TU TU
TE
TE
c = 3
b 2

θ θ
θE θE
(c/b)θE
(a) (b)
Figure 4.2.7. Torque-twist diagrams for circular shaft: (a) solid; (b) hollow.
(Compare with Figure 4.1.3.)
If the shaft is hollow, with inner radius b, then the expressions derived
above for φ is the elastic and plastic regions are valid as long as b < r∗ < c.
The onset of plasticity occurs when T = π(c4 − b4 )k/2c = TE , while the
ultimate torque is TU = 2π(c3 − b3 )k/3. The torque in contained plastic
deformation is given by
!
c3 b4 r∗ 3
T = 2πk − ∗− .
3 4r 12
As can be seen in Figure 4.2.7(b), the ultimate torque is attained when

r∗ = b, that is, when θ = k/Gb = (c/b)θE . In this case the approach is not
asymptotic.
The elastic-plastic torsion of a circular shaft of variable diameter was
studied by Sokolovskii [1945], extending the elastic theory due to Michell
[1900]. In such a shaft the shear strain γrθ and shear stress τrθ are present
in addition to γzθ and τzθ . Equilibrium requires that the stresses be derived
from a stress function φ̄(r, z) by means of
1 ∂ φ̄ 1 ∂ φ̄
τzθ = , τrθ = − ,
r2 ∂r r2 ∂z
where φ̄ obeys
∂ 2 φ̄ 3 ∂ φ̄ ∂ 2 φ̄
− + 2 =0
∂r2 r ∂r ∂z
in the elastic region and
∇φ̄ = kr 2

in the plastic region. The equations must in general be solved numerically.

A problem mathematically similar to the preceding is that of the torsion
of a segment of a torus by means of equal and opposite forces acting over the
end sections and perpendicular to the plane of the torus. Both the elastic
and the elastic-plastic problems were treated by Freiberger [1949, 1956a].
Sokolovskii’s Oval Shaft

While no known solution exists for the contained plastic torsion of an
elliptic shaft, an inverse method has been applied by Sokolovskii [1942] to
the torsion of a shaft of an oval cross-section that differs only slightly from
an elliptic one, and whose form is determined from the solution. A form of
φ is assumed in the elastic region, namely,
!
k x2 y 2
φ(x, y) = c− − . (4.2.5)
2 ā b̄
The angle of twist is given in accord with Equation (4.2.2) as
k(ā + b̄)
θ= . (4.2.6)
2Gāb̄
Furthermore, s
2 2
x y
|∇φ| = k + ,
ā b̄
so that the elastic-plastic boundary Γ is given by the ellipse
2 2
x y
+ = 1,
ā b̄
or, parametrically, by
x = ā cos χ, y = b̄ sin χ.
In terms of this representation,
∇φ = −k(e1 cos χ + e2 sin χ).
The angle χ thus defines the direction of steepest descent of the φ surface
at the elastic-plastic boundary; note that this direction is not, in general,
normal to this boundary, except where it coincides with the principal axes
of the ellipse.
The form of φ in the plastic zone may now be obtained by means of
the previously discussed general solution for fully plastic torsion. Thus, if
straight lines are drawn outward from Γ in the direction of steepest descent,
then along these lines φ has the constant slope −k, and these lines are
normal to the external boundary curve C. Consider one such line, drawn
from the point (ξ, η) on Γ to a point (x, y) on the external boundary C.
Since φ(x, y) = 0, the distance between the two points is φ(ξ, η)/k. If
ξ = ā cos χ and η = b̄ sin χ, then
1
x = ā cos χ + (c − ā cos2 χ − b̄ sin2 χ) cos χ,
2
1
y = b̄ sin χ + (c − ā cos2 χ − b̄ sin2 χ) sin χ.
2
From these expressions we can determine the semiaxes of C, namely,
1 1
a = x|χ=0 = (ā + c), b = y|χ=π/2 = (b̄ + c);
2 2
note that
ā − b̄ = 2(a − b). (4.2.7)
The boundary curve may now be represented in terms of a and b as
x = (2a − b) cos χ − (a − b) cos3 χ,
y = (2b − a) sin χ + (a − b) sin3 χ.

These equations describe a closed oval shape if 12 a < b < 2a. For most ratios
a/b within this range, the oval is almost indistinguishable from the ellipse
[see Figure 4.2.8(a)].
Equations (4.2.6) and (4.2.7) can now be solved simultaneously to give
the the semiaxes ā, b̄ of the elastic-plastic boundary as functions of the twist
θ: s
k k 2

ā = a − b + + (a − b) +2 ,
2Gθ 2Gθ
Ellipse

Sokolovskii 2G(a−b)θ/k = 0.75

oval
1.0
2.0

a = 4b/3 ∞ 0 x

(a) (b)
Figure 4.2.8. Sokolovskii oval: (a) comparison with ellipse; (b) elastic-plastic
boundary at various values of the twist.
s
k k 2

b̄ = b − a + + (a − b)2 + .
2Gθ 2Gθ
For the elastic-plastic boundary to lie entirely within C, it is necessary that
ā ≤ a and b̄ ≤ b, or
s
2
k k

+ (a − b)2 + ≤ min(a, b).
2Gθ 2Gθ
Supposing that a > b, we find the minimum value of θ for which the solution
is valid to be
kb
θmin = .
Ga(2b − a)
As θ increases above this value, the ellipse shrinks; in the limit as θ → ∞,
b̄ goes to zero, and ā goes to 2(a − b), so that in the fully plastic state a
ridge line extending over −2(a − b) ≤ x ≤ 2(a − b) is formed. Elastic-plastic
boundaries corresponding to different values of θ are shown in Figure 4.2.8(b)
for a = 4b/3.
By means of a different inverse method, Galin [1949] was able to solve a
series of elastic-plastic torsion problems for shafts of nearly polygonal cross-
section.

1. Show that for the torsion problem, the compatibility conditions (1.2.4)
are equivalent to the single equation
γ2 ,1 −γ1 ,2 = constant,
and that the constant is 2θ.

2. If the stress field in the plastic region is expressed as
σ13 = k cos α, σ23 = k sin α,

Section 4.3 / The Thick-Walled Hollow Sphere and Cylinder 205
find the forms taken by the equilibrium equation and the boundary
conditions on C in terms of α. Interpret α in relation to the plastic
stress function φp .
3. Formulate the torsion problem in terms of the deformation theory of

plasticity (see 3.2.1). Show that for a circular shaft the solution coin-
cides with that given by the incremental theory.
4. Find the ultimate torque and the fully plastic warping for a shaft whose
cross-section is an equilateral triangle.
5. If φ̄(r, z) is the stress function for the torsion of a circular shaft of

variable diameter,
(a) (a) find the boundary conditions that must be satisfied by φ̄ on

the outer surface (say r = r̄(z)) and on the axis r = 0, and
(b) (b) show that the torque at a section z is
T = 2π[φ̄(r̄(z), z) − φ̄(0, z)].
6. Discuss the contour lines of the fully plastic stress function in the
Sokolovskii oval shaft. Plot for a/b = 4/3.
7. Use the formulation of Exercise 2 to show that the fully plastic stress
distribution for a shaft of elliptic cross-section, bounded by (x/a)2 +
(y/b)2 = 1, is given implicitly by
(a2 − b2 ) sin 2α
x cos α + y sin α = p .
2 a2 sin2 α + b2 cos2 α
Plot the contours of the stress function for a/b = 4/3, and compare
the result with that of Exercise 6.
Section 4.3 The Thick-Walled Hollow Sphere and

Cylinder
Like the problem of a tube under torsion, that of an axisymmetrically loaded
shell of revolution is statically determinate when the shell is thin-walled but
ceases to be so when the shell is thick-walled. In the present section we
present problems of pressurized thick-walled elastic-plastic shells that are
spherical or cylindrical. Problems of the hollow sphere and cylinder are
sufficiently different to warrant presentation of both, but sufficiently similar
so that not all topics need be treated for both. Treatment of the sphere
includes the solution for finite strain and thermal stresses, while treatment
of the cylinder includes the effects of unloading and reloading and of work-
hardening. More details can be found in books such as Johnson and Mellor
[1973] and Chakrabarty [1987].
4.3.1. Elastic Hollow Sphere Under Internal and External

Pressure
Basic Equations
In a hollow sphere of inner radius a and outer radius b, subject to normal
pressures on its inner and outer surfaces, and made of an isotropic mater-
ial, the displacement and stress fields must be spherically symmetric. The
only nonvanishing displacement component is the radial displacement u, a
function of the radial coordinate r only. The only nonvanishing strains are,
by Equations (1.2.2), the radial strain εr = du/dr and the circumferential
strains εθ = εφ = u/r. The strains obviously satisfy the compatibility con-
dition
d
εr = (rεθ ). (4.3.1)
dr
The only nonvanishing stress components are the radial stress σr and
the circumferential stresses σθ = σφ , which satisfy the equilibrium equation
dσr σr − σθ
+2 = 0. (4.3.2)
dr r
Elastic Solution
The strain-stress relations for an isotropic linearly elastic solid, Equa-
tions (1.4.11), reduce in the present case to
1
εr = (σr − 2νσθ ),
E
1
εθ = [(1 − ν)σθ − νσr ].
E
The compatibility equation (4.3.1) may now be rewritten in terms of the
stresses to read
d 1+ν
[(1 − ν)σθ − νσr ] + (σθ − σr ) = 0,
dr r
and with the help of (4.3.2) reduces to
d
(σr + 2σθ ) = 0.
dr
The quantity σr + 2σθ is accordingly equal to a constant, say 3A. Further-

more, dσθ /dr = − 21 dσr /dr, so that 32 d(σθ − σr )/dr = dσr /dr, and Equation
(4.3.2) may be rewritten as
d 3
(σθ − σr ) + (σθ − σr ) = 0,
dr r
leading to the solution
3B
σθ − σr = ,
r3
where B is another constant. The stress field is therefore given by
2B B
σr = A − , σθ = A + .
r3 r3
With the boundary conditions
σr |r=a = −pi , σr |r=b = −pe ,
where pi and pe are the interior and exterior pressures, respectively, the
constants A and B can be solved for, and the stress components σr , σθ can
be expressed as
" 3 3 #
1 pi − pe a a
σr = − (pi + pe ) + 1+ −2 ,
2 2[1 − (a/b)3 ] b r
" 3 3 #
1 pi − pe a a
σθ = − (pi + pe ) + 1+ + ,
2 2[1 − (a/b)3 ] b r
that is, the stress field is the superposition of (1) a uniform stress field equal
to the negative of the average of the external and internal pressures, and (2)
a variable stress field proportional to the pressure difference.
Sphere Under Internal Pressure Only

The preceding solution is due to Lamé. It becomes somewhat simpler if
the sphere is subject to an internal pressure only, with pi = p and pe = 0.
The stresses are then
!
p b3
σr = − −1 ,
(b/a)3 − 1 r3
!
p b3
σθ = +1 .
(b/a)3 − 1 2r3
If the sphere material is elastic-plastic, then the largest pressure for which
the preceding solution is valid is that at which the stresses at some r first
satisfy the yield criterion; this limiting pressure will be denoted pE . Since
two of the principal stresses are equal, the stress state is equibiaxial (see
3.3.4), and both the Tresca and Mises criteria reduce to
σθ − σr = σY , (4.3.3)
where σY is the tensile yield stress, since σθ > σr everywhere. The value of
σθ − σr is maximum at r = a, where it attains 3p/2[1 − (a/b)3 ]. The largest
pressure at which the sphere is wholly elastic is therefore
!
2 a3
pE = σ Y 1− 3 . (4.3.4)
3 b
4.3.2. Elastic-Plastic Hollow Sphere Under Internal Pressure
Stress Field
When the pressure in the hollow sphere exceeds pE , a spherical domain
of inner radius a and outer radius, say, c becomes plastic. The elastic domain
c < r < b behaves like an elastic shell of inner radius c that is just yielding
at r = c, so that σr and σθ are given by
!
pc b3
σr = − −1 ,
(b/c)3 − 1 r3
!
pc b3
σθ = +1 .
(b/c)3 − 1 2r3
where pc = −σr (c) is such that the yield criterion is met at r = c, that is, it
is given by the right-hand side of (4.3.4) with a replaced by c:
!
2 c3
pc = σ Y 1− 3 .
3 b
Therefore, !
2 c3 c3
σr = − σY − ,
3 r3 b3!
c<r<b (4.3.5)
2 c3 c3
σθ = σY + .
3 2r3 b3
In particular, σθ (b) = σY (c/b)3 .
In the plastic domain, the yield criterion (4.3.3) holds everywhere, so
that the equilibrium equation (4.3.2) may be integrated for σr , subject to
continuity with the elastic solution at r = c, to yield
!
2 c3 c3
σr = − σY 1 − 3 + ln 3 .
3 b r
We immediately obtain σθ = σr + σY as well, namely,

!
1 c3 c3
σθ = σY 1 + 2 3 − 2 ln 3 .
3 b r
The radius c marking the extent of the plastic domain is obtained, at a given
pressure p, from the condition that σr (a) = −p, or
!
2 c3 c3
p = σY 1 − 3 − ln 3 . (4.3.6)
3 b a
When c = b, the shell is completely plastic. The corresponding pressure

is the ultimate pressure, given by
b
pU = 2σY ln .
a
Displacement
The displacement u in the elastic domain is given by u(r) = (r/E)[(1 −
ν)σθ − νσr ]. In particular,
" #
σY c c3
u(c) = (1 + ν) + 2(1 − 2ν) 3 .
3E b
In the plastic domain the displacement is determined from the fact that
the volume strain is purely elastic, that is,
!
du u 1 − 2ν 2(1 − 2ν)σY c3 c3
+2 = (σr + 2σθ ) = 2 + 3 − ln 3 .
dr r E 3E b r
The left-hand member is equal to r−2 d(r2 u)/dr, and therefore the equation
can be integrated subject to continuity at r = c. In particular
" !#
σY a c3 2 c3 c3
u(a) = (1 − ν) 3 − (1 − 2ν) 1 − 3 + ln 3 .
E a 3 b a
When p = pU , c = b, and the expression simplifies to

" #
σY a b3 b
u(a) = (1 − ν) 3 − 2(1 − 2ν) ln .
E a a
Solution for Finite Strains

If b/a is sufficiently large, the strains become of order (b/a)3 σY /E and
may be too large to be regarded as infinitesimal. A solution for finite strains
may be obtained by regarding the strains as logarithmic, a procedure that is
permissible in the present case, since the principal strain directions do not
change (see Section 2.1).
If r is the current radius of a spherical surface whose initial radius is r0 ,
then the current volume of the infinitesimal shell contained between r and
r + dr is 4πr2 dr, while its initial volume is 4πr02 dr0 . The volume strain
is consequently ln(r2 dr/r02 dr0 ). Assuming plastic incompressibility, we can
equate this strain to (1 − 2ν)(σr + 2σθ )/E, or
r02 dr0 1 − 2ν 1 − 2ν

.
2
= exp − (σr + 2σθ ) = 1 − (σr + 2σθ ), (4.3.7)
r dr E E
since the stresses are small compared to E. The equilibrium equation (4.3.2)
is exact if it is interpreted as Eulerian, that is, if r denotes the current radius
and σr , σθ are the true stresses. Eliminating σθ between (4.3.2) and (4.3.7),
we obtain
dr0 1 − 2ν d 3
r02 = r2 − (r σr ).
dr E dr
Integration leads to
r03 σr 3A
3
= 1 − 3(1 − 2ν) − 3 ,
r E r
where A is a constant of integration. Note that this relation holds in both the
elastic and plastic domains. At r = b we have r0 = b0 , and since σr (b) = 0,
1/3
b0 3A A

.
= 1− 3 = 1 − 3.
b b b
But b0 = b−u(b), where u(b) may be obtained from the previous solution for
the elastic domain, namely, u(b) = (1 − ν)bσθ (b)/E = (1 − ν)b(c/b)3 σY /E.
Consequently, A = (1 − ν)c3 σY /E. The change in the internal radius can be
expressed as
3 " #
a0 3 c3

=1− (1 − ν)σY 3 − (1 − 2ν)p . (4.3.8)
a E a
If the material is perfectly plastic, then the previously derived stress

field in the plastic zone is still valid, and the relation between c and p is
still given by (4.3.6). However, the preceding solution for the radius change
is not limited to a perfectly plastic material. The yield stress σY enters
only because it is contained in the elastic solution, and reflects the fact that
yielding first occurs at r = c. If the material work-hardens, then σY need
only be replaced by the initial yield stress σE .
At at any point of the shell, the state of stress is a superposition of the
hydrostatic stress σθ and of the uniaxial compressive stress σr − σθ acting in
the radial direction. Since it is only the latter stress that produces plastic
deformation, in the plastic domain this stress is related to εpr in the same
way that σ is related to εp in a uniaxial compression test. Let this relation
be denoted σ = −σY (−εp ); then (4.3.3) may be written as
σθ − σr = σY (−εpr ).
Since εθ = ln(r/r0 ),
1 − 2ν
εr + 2εθ = (σr + 2σθ ),
E
and
1
εr = εpr + (σr − 2νσθ ),
E
it follows that
2 r
εpr = [(1 − ν)σθ − νσr ] − 2 ln ,
E r0
so that
r (1 − ν)σθ − νσr

σθ − σr = σY 2 ln − 2 . (4.3.9)
r0 E
This equation, along with the equilibrium equation (4.3.2) and the elastic
compressibility condition (4.3.7), constitutes the system of three simultane-
ous equations that govern the three variables σr , σθ , and r0 as functions of r
in the plastic domain. The equations must, in general, be solved numerically.
A simplification is achieved if the material is elastically as well as plas-
tically incompressible, that is, if ν = 21 . In this case (4.3.9) becomes
r σθ − σr

σθ − σr = σY 2 ln − ,
r0 E
and therefore may be solved for σθ − σr as
r

σθ − σr = f 2 ln .
r0
Here f is the same function that gives the relation between the true stress
and the total logarithmic strain in uniaxial compression, σ = f (ε). Inserting
the last equation into (4.3.2) and integrating leads to
!
r f (2 ln(r/r0 )) 2 c3
Z
σr = 2 dr − 1− 3 σE ,
c r 3 b
with continuity at r = c taken into account. The internal pressure is there-

fore !
c3
Z c
2 f (2 ln(r/r0 ))
p= 1 − 3 σE + 2 dr, (4.3.10)
3 b a r
a result that agrees with (4.3.6) when f (ε) ≡ σY (perfect plasticity). Since
the material is assumed incompressible,
r3 − r03 = a3 − a30 .
.
In the elastic region, however, r3 − r03 = 3r2 u(r), while u(r) = σE c3 /2Er2 .
Hence
c3 = α(a3 − a30 ), (4.3.11)
where
2E
α= .
3σE
Equation (4.3.10) may now be rewritten with the help of the variable s =
(a3 − a30 )/r3 to read
!
2 c3 2
Z 1−(a0 /a)3 f (− 32 ln(1 − s))
p= 1− 3 σE + ds. (4.3.12)
3 b 3 α s
Equations (4.3.11)–(4.3.12) may be used to calculate the pressure as a func-

tion of the expansion ratio ρ = a/a0 .
When large plastic strains are taken into account, it is found that the
pressure attains a maximum at a finite value of the expansion, representing
plastic instability analogous to necking of tensile specimens. Differentiating
(4.3.12) with respect to a we obtain
dp c2 dc 2 1
= −2σE 3 + f (2 ln ρ) 3 .
da b da a ρ −1
From (4.3.11),
dc 2E 2
c2 = a ,
da 3σE
so that dp/da = 0 when ρ is given by
3
2Eρ3 (ρ3 − 1) b

= ,
3f (2 ln ρ) a0
an equation that can be easily solved graphically for a given stress-strain
curve. In particular, for the perfectly plastic material the equation is quadratic
in ρ3 and can be solved explicitly.
If the material is work-hardening and the shell is sufficiently thin, how-
ever, the instability does not occur before the shell has become completely
plastic. The limiting wall ratio, for which the pressure is maximum just
when the shell becomes fully plastic, is obtained by setting c = b, so that
ρ3 = 1 + α(b/a0 )3 . Defining
!
2 b3
ε0 = ln 1 + α 3 ,
3 a0
we find that the limiting value of ε0 (and hence of b/a0 ) is given by
f (ε0 ) = σE e3ε0 /2 .
Spherical Cavity in an Infinite Solid

In all the foregoing, the limit as b → ∞ describes the expansion of a
spherical cavity in an infinite solid. When a/a0 is very large, the left-hand
side of (4.3.8) is negligible next to unity, as is the last term in the brackets on
the right-hand side (since p E). Consequently c/a becomes approximately
constant, with the value
1/3
c E

= ,
a 3(1 − ν)σY
which holds in a work-hardening solid as well if σY is replaced by σE . If the

material is incompressible then the pressure is given by (4.3.12), with c3 /b3 =
0 and the upper limit of the integral equal to unity. At this upper limit,
corresponding to r = a, the argument of f becomes infinite. Nonetheless,
the integral converges if the derivative of f approaches a constant value for
large arguments. In particular, if f 0 (ε) = H1 (constant), then the pressure
is given by
2 2E 2

p= 1 + ln σ E + π 3 H1 .
3 3σE 27
Bishop, Hill and Mott [1945], who derived this result, applied it to obtain
a theoretical estimate of the maximum steady-state pressure in the deep
penetration of a smooth punch into a quasi-infinite medium, and found close
agreement with an experiment using cold-worked copper.
4.3.3. Thermal Stresses in an Elastic–Perfectly Plastic Hol-

low Sphere
Thermoelastic Stresses
If the inner and outer surfaces of the sphere are at different temperatures,
say Ta and Tb , respectively, then the temperature inside the shell varies with
r. The thermoelastic stress-strain-temperature relations (1.4.13) are, in the
present case,
1
εr = (σr − 2νσθ ) + αT,
E (4.3.13)
1
εθ = [(1 − ν)σθ − νσr ] + αT.
E
The compatibility equation (4.3.1) now reads
d 1+ν dT
[(1 − ν)σθ − νσr ] + (σθ − σr ) + Eα = 0,
dr r dr
and with the help of (4.3.2), reduces to
d Eα dT
(σr + 2σθ ) − 2 = 0,
dr 1 − ν dr
or, upon integrating,
Eα
σr + 2σθ = 3A − 2 T.
1−ν
Combining this again with (4.3.2) leads to
d 3 Eα 2
(r σr ) = 3Ar2 − 2 r T,
dr 1−ν
which can be integrated to yield σr (r) when the temperature distribution

is known. Equation (4.3.2) can then be used to obtain σθ (r), and the dis-
placement field u(r) can be obtained from Equation (4.3.13)2 through the
relation u = rεθ .
If the temperature difference Ta − Tb is not too great, then the steady-
state temperature field T (r) obeys the spherically symmetric Laplace equa-
tion,
d2 T 2 dT
2
+ = 0.
dr r dr
The solution is
(b/a)Tb + Ta Ta − Tb b
T (r) = + . (4.3.14)
(b/a) − 1 (b/a) − 1 r
If the constant A is redefined so as to include the contribution of the constant
term in (4.3.14), then the solution for the stresses is
2B b
σr = A + −λ ,
r3 r (4.3.15)
B λb
σθ = A − 3 − ,
r 2r
where
Eα(Ta − Tb )
λ= .
(1 − ν)[(b/a) − 1]
If the stresses are due to the temperature difference only, then the constants
A, B may be determined from the conditions σr (a) = σr (b) = 0. The results
are " !#
b 1 b3 b2 b
σr = −λ − + + ,
r (b/a)2 + (b/a) + 1 r3 a2 a
" !#
b 1 b3 b2 b
σθ = −λ + 2 3
− 2 2
−2 ,
2r (b/a) + (b/a) + 1 2r a a
so that " #
λ b 3 b3
σθ − σr = − .
2 r (b/a)2 + (b/a) + 1 r3
The greatest numerical value of the right-hand side occurs at r = a, where
it equals
|λ| (b/a)(b/a − 1)(2b/a + 1)
|σθ − σr |max = .
2 (b/a)2 + (b/a) + 1
Yielding therefore occurs when this quantity first equals σY , so that the
temperature difference required for initial yielding is
(1 − ν)σY 1 + (a/b) + (a/b)2
|Ta − Tb |E = . (4.3.16)
Eα 1 + (a/2b)
The combined effect of internal pressure and temperature gradient was

analyzed by Derrington and Johnson [1958]. The stress field is given by
the superposition of the previously obtained stress fields in the shell under
pressure alone and temperature gradient alone. However, in the shell under
combined pressure and temperature gradient the maximum of |σθ − σr | does
not always occur at r = a. It does so when p/λ (with λ positive) is sufficiently
large or sufficiently small, the upper and lower limiting values depending on
the wall ratio b/a. At intermediate values of p/λ, yielding may begin in the
interior of the shell or (if b/a ≤ 2 and λ ≥ 3σY ) at r = b. At the lower
limiting value of p/λ, initial yielding occurs simultaneously at r = a and at
r = min(b, 2a).
Thermoplastic Stresses
We return to the shell under temperature gradient alone. When |Ta −Tb |
exceeds |Ta − Tb |E as given by (4.3.16), a plastic region a < r < c is formed,
in which the equilibrium equation (4.3.2), the yield criterion σθ − σr = σY
and the boundary condition σr (a) = 0 together produce the stress field
r
σr (r) = −2σY ln ,
a

r
a ≤ r ≤ c. (4.3.17)
σθ = −σY 1 + ln r ,
a
In the elastic region c < r < b, the stresses are given by Equations (4.3.15).
The boundary condition σr (b) = 0 leads to
2B
A+ = λ,
b3
and the continuity of the stresses at r = c yields two additional equations
for B and λ in terms of c, which may be solved to give
1 − c/b + ln(c/a)
B = c3 σY ,
(2 + c/b)(1 − c/b)2
c 2(1 − c3 /b3 ) + ln(c/a)3

λ = σY .
b (2 + c/b)(1 − c/b)2
Note that c/b attains the value 1 only as λ becomes infinite.
It was observed by Cowper [1960] that the elastic domain c < r < b
is equivalent to a shell that is just yielding at the inner radius c under
the combined action of a temperature difference and an internal pressure
2σY ln(c/a). It follows from the previous discussion that if λ, and hence c, is
sufficiently large, a second plastic region is formed, starting either at r = b
(if c ≥ b/2) or at r = 2c (if c ≤ b/2). The former case occurs if b/a ≤ 2e1/3
and the latter if b/a ≥ 2e1/3 . Even when the second plastic domain is taken
into account, the sphere can become completely plastic only at an infinite
temperature difference. For more discussion and further details on thermal
stresses in elastic-plastic hollow spheres, see Johnson and Mellor [1973], pp.
204–214, and Chakrabarty [1987], pp. 343–350.
4.3.4. Hollow Cylinder: Elastic Solution and Initial Yield

Pressure
Basic Equations
The problem of a thick-walled cylindrical tube of inner radius a and
outer radius b, subject to normal pressures on its inner and outer surfaces,
is not as simple as the corresponding problem of the hollow sphere, because
the mechanical state in general varies not only with the radial coordinate r
but also the axial coordinate z. If, however, the cylinder is of sufficiently
great length compared to b, then some simplifying assumptions may be made.
First, the stresses and strains on sections far enough away from the ends may
be regarded as independent of z. In addition, plane sections perpendicular
to the tube axis may be assumed to remain plane, so that the axial strain
εz is constant.
As a result of symmetry about the z-axis, the shear-stress components
(in cylindrical coordinates) τrθ and τzθ vanish, and all remaining stress com-
ponents are independent of θ as well, and hence functions of r only. The
equilibrium equations are therefore
dσr σr − σθ
+ =0 (4.3.18)
dr r
and
dτrz τrz
+ = 0.
dr r
The latter equation has the solution rτrz = constant, and since the traction
on the tube boundaries consists of pressures only, τrz must vanish. Equation
(4.3.18) is therefore the only equilibrium equation of the problem.
If u(r) is the radial displacement, then the radial and circumferential

strain components are, from Equations (1.2.1), εr = du/dr and εθ = u/r, so
that the compatibility condition is once more Equation (4.3.1).
Elastic Solution
The elastic stress-strain relations reduce to
1
εr = [σr − ν(σθ + σz )],
E
1
εθ = [σθ − ν(σr + σz )],
E
1
εz = [σz − ν(σr + σθ )].
E
If the third equation is solved for σz and the result is substituted in the first
two, these equations become
1+ν
εr = [(1 − ν)σr − νσθ ] − νεz ,
E
1+ν
εθ = [(1 − ν)σθ − νσr ] − νεz .
E
Substitution in the compatibility relation (4.3.1) results in
d σr − σθ
[(1 − ν)σθ − νσr ] = ,
dr r
which, when combined with (4.3.18), becomes
d
(σθ + σr ) = 0.
dr
Consequently σθ + σr is constant, as is σz . Equation (4.3.18) can now be
rewritten as
d σθ − σr
(σθ − σr ) + 2 = 0.
dr r
The solution of this equation is σθ − σr = 2B/r2 , where B is a constant. If
the constant value of σθ + σr is 2A, then
B B
σr = A − , σθ = A + .
r2 r2
With the boundary conditions
σr |r=a = −pi , σr |r=b = −pe ,
where pi and pe have the same meaning as for the sphere, the constants A
and B can be solved for, and the stress components σr , σθ can be expressed
as " 2 2 #
1 pi − pe a a
σr = − (pi + pe ) + 2
1+ −2 ,
2 2[1 − (a/b) ] b r
" 2 2 #
1 pi − p e a a
σθ = − (pi + pe ) + 1+ +2 .
2 2[1 − (a/b)2 ] b r
Under internal pressure only, these equations simplify to
!
p b2
σr = − −1 ,
(b/a)2 − 1 r2 !
(4.3.19)
p b2
σθ = +1 .
(b/a)2 − 1 r2
The axial stress σz is given by
2νp
σz = + Eεz ,
(b/a)2 − 1
and the radial displacement u is
" #
(1 + ν)p b2
u(r) = (1 − 2ν)r + − νεz r.
E[(b/a)2 − 1] r
Since the axial stress σz is constant, it is equal to P/π(b2 − a2 ), where P

is the resultant axial force. If P is prescribed, then εz may be determined
accordingly. Alternatively, the value of εz may be prescribed; for example,
a condition of plane strain (εz = 0) may be assumed.
If the tube is open-ended, then P = 0 and therefore σz = 0, so that this
condition is equivalent to plane stress (as is shown later, the equivalence
no longer holds when the tube is partially plastic). If the tube is closed-
ended, then P must balance the resultant of the interior pressure over the
interior cross-sectional area πa2 , so that σz = p/[(b/a)2 − 1] = 12 (σr + σθ ).
The tube may also be assumed to be in a state of plane strain, that is,
εz = 0, and thus σz = ν(σr + σθ ) = 2νp/[(b/a)2 − 1]. Note that this result
coincides with the preceding one if and only if ν = 21 , that is, if the material
is elastically incompressible. The axial strain is given, for the three cases,
by
αp
Eεz = 2 2 (4.3.20)
b /a − 1
where 
 1 − 2ν, closed end,

α= 0, plane strain, (4.3.21)
 −2ν,

open end.
The axial stress is
(2ν + α)p
σz = . (4.3.22)
b2 /a2 − 1
Initial Yield Pressure
The largest pressure for which the preceding solution is valid is that at
which the stresses at some r first satisfy the yield criterion; this limiting
pressure will again denoted pE . It can be readily seen that in all three
conditions — open end, closed end, and plane strain — the axial stress σz
is the intermediate principal stress. Consequently, if the Tresca criterion
is assumed, then it takes the simple form σθ − σr = 2k, since σθ > 0 and
σr < 0. Since σθ −σr = pa2 /[1−(a/b)2 ]r2 , the maximum is reached at r = a,
and therefore !
a2
pE = k 1 − 2 . (4.3.23)
b
The Mises criterion in terms of σr , σθ , σz is
σr2 + σθ2 + σz2 − σr σθ − σr σz − σθ σz = 3k 2 ,
and may be rewritten as

2 2
σθ − σr 1 σr + σθ

2
=k − σz − .
2 3 2
Since σz is explicitly involved, the form taken by the criterion in the present
case depends on the end conditions. In each case, however, the right-hand
side is constant, so that as in the Tresca case, yielding first occurs at r = a,
where the left-hand side takes the value p2E /[1 − (a/b)2 ]2 .
In the closed-end condition, substitution of σz = 12 (σr + σθ ) leads to pE
given by (4.3.23). In the open-end condition, the criterion is the one that
applies in plane stress, namely, σr2 − σr σθ + σθ2 = 3k 2 . Hence
1 − (a/b)2
pE = k q . (4.3.24)
1 + 13 (a/b)4
In plane strain, as we have found, σz = ν(σθ + σr ), and the criterion

takes the form
(1 − ν + ν 2 )(σr2 + σθ2 ) − (1 + 2ν − 2ν 2 )σr σθ = 3k 2 ,
producing the critical pressure
1 − (a/b)2
pE = k q . (4.3.25)
1 + 13 (1 − 2ν)2 (a/b)4
The three results may be combined in the form
1 − (a/b)2
pE = k q ,
1 + 31 (1 − 2ν − α)2 (a/b)4
where α is defined by (4.3.21).

4.3.5. Elastic-Plastic Hollow Cylinder

When the pressure exceeds pE , a plastic zone extends from r = a to, say,
r = c. The elastic zone c < r < b behaves like an elastic tube of inner radius
c that is just yielding at r = c, so that σr and σθ are given by
!
pc b2
σr = − −1 ,
(b/c)2 − 1 r2 !
c<r<b (4.3.26)
pc b2
σθ = +1 ,
(b/c)2 − 1 r2
where pc = −σr (c) is such that the yield criterion is met at r = c.
Tresca Criterion: Stress Field

If it is again assumed that σθ ≥ σz ≥ σr (an assumption that must be
verified a posteriori ) then the Tresca criterion yields
!
c2
pc = k 1 − 2 .
b
The associated flow rule for the Tresca criterion implies that with σz
as the intermediate principal stress, εz is purely elastic and given by [σz −
ν(σr + σθ )]/E, so that
2νpc
σz = + Eεz .
(b/c)2 − 1
In the plane-strain condition, εz = 0, the assumption that σz is the interme-
diate principal stress at r = c can be immediately verified, since
b2 b2
+ 1 ≥ 2ν ≥ − + 1.
c2 c2
For the other two conditions, σz cannot be determined without studying the
plastic zone.
The assumption σθ ≥ σz ≥ σr will be retained as valid in the plastic
zone as well. The yield criterion σθ − σr = 2k must then be satisfied for
a ≤ r ≤ c, and therefore the equilibrium equation (4.3.18) may be integrated
immediately for σr . The solution
r2
σr = −p + k ln
a2
satisfies the boundary condition at r = a. Continuity of σr at r = c leads to
the result !
c2 c2
p = k 1 − 2 + ln 2 , (4.3.27)
b a
first derived by Turner [1909]. Equation (4.3.27) can be used to determine

c for a given pressure p.
The ultimate pressure pU is attained when the whole tube is plastic, that
is, when c = b; it is given by
b2
pU = k ln . (4.3.28)
a2
If εz , as assumed, is purely elastic, then σz = Eεz + ν(σr + σθ ) in the

plastic zone as well. The resultant axial force is
Z b Z b
P = 2π rσz dr = πEεz (b2 − a2 ) + 2πν (σr + σθ )r dr.
a a
But r(σθ + σr ) = r(σθ − σr + 2σr ) = r2 dσr /dr + 2rσr = d(r2 σr )/dr by the
equilibrium equation (4.3.18), so that the integral on the right-hand side is
2πνr2 σr |ba = 2πνpa2 . Hence
P = π[Eεz (b2 − a2 ) + 2νpa2 ],
independently of c. It follows that εz has the same form (4.3.20) as in the

purely elastic tube, with α given by (4.3.21), and p by (4.3.27).
The stress field in the elastic zone can now be written as
!
c2 c2
σr = −k − ,
r2 b! 2
c2 c2
σθ = k 2
+ 2
, c<r<b (4.3.29)
"
r b !#
c2 α c2 c2
σz = k 2ν 2 + 1 − 2 + ln 2 .
b (b/a)2 − 1 b a
In the plastic zone, the stress field is

!
c2 c2
σr = −k 1 − 2 + ln 2 ,
b r!
c 2 c 2
σθ = k 1 + 2 − ln 2 , a<r<c
"
b r ! !#
c2 c2 α c2 c2
σz = k 2ν 2 − ln 2 + 1 − 2 + ln 2 .
b r (b/a)2 − 1 b a
(4.3.30)
Limit of Validity
It is clear from the last equation that σz ≥ 0 everywhere if α ≥ 0, that
is, in the closed-end and plane-strain conditions. If α = −2ν (open-end
condition), then σz must be negative in some parts and positive in others
in order that P = 0; note that in the elastic-plastic tube this condition

is not equivalent to plane stress, as was remarked before. Nonetheless, the
inequality σz ≥ σr can be verified for this condition as well. The last equation
can in this case be rewritten as
!
2νk b2 c2 c2
σz = 2νσr + + − ln −2 .
(b/a)2 − 1 a2 b2 a2
The quantity in parentheses is easily seen to be nonnegative whenever a ≤

c ≤ b, and therefore, since σr < 0 throughout the plastic region, the inequal-
ity σz ≥ σr is always satisfied.
To ensure that σθ − σz ≥ 0 is satisfied everywhere, it is sufficient that it
be satisfied where this quantity is smallest, namely, at r = a, where it equals
" ! !#
c2 c2 α c2 c2
k (1 − 2ν) 1 + 2 − ln 2 + 2ν − 2
1 − 2 + ln 2 .
b a (b/a) − 1 b a
For the results to be valid for all pressures up to ultimate, this quantity
must be nonnegative for all c up to and including b, and therefore the limit
of validity is given by the equation
b2 α b2
2(1 − ν) − (1 − 2ν) ln − ln = 0,
a2 (b/a)2 − 1 a2
a relation between b/a and ν that depends on the end condition. For ν = 0.3,
the limiting values of b/a are respectively 5.43, 6.19 and 5.75 for the closed-
end, open-end and plane-strain conditions.
The preceding result is due to Koiter [1953b], who pointed out that while
an analysis may also be carried out for higher values of b/a, the hypothesis
of infinitesimal deformations is no longer tenable.
The Plane-Stress Problem

A condition of plane stress (σz = 0) is descriptive of a thin circular disk
with a concentric hole around whose edge a radial pressure p is applied. In
this case the preceding solution for σr and σθ , given by Equations (4.3.29)1,2
and Equations (4.3.30)1,2 , is valid as long as σθ ≥ 0, or
c2 c2
1+ − ln ≥ 0.
b2 a2
The entire disk can become plastic (c = b) only if b/a ≤ e. For larger values
of b/a, the plastic zone cannot expand beyond a limiting radius c whose
value is furnished by the preceding inequality, turned into an equation. The
maximum c/a turns out to be a decreasing function of b/a, with the limit
e1/2 = 1.649 as b/a → ∞ (a circular hole in an infinite plate). In other
words, the larger the disk in relation to the hole, the less the plastic zone
can expand. Note that the pressure can never exceed 2k.
While it may be thought that the preceding limitation is due to the dis-
continuous nature of the Tresca criterion, this is not the case. Qualitatively
similar results are given by the Mises criterion. The equation representing
the Mises criterion in plane stress,
σr2 − σr σθ + σθ2 = 3k 2 ,
can be solved for σθ :

r q
1 3
σθ = σr + 4k 2 − σr2 .
2 2
When this expression is substituted in (4.3.18) then the differential equation
is separable. Defining s = −σr /2k, we obtain
dr ds
= −q √ .
r 3
1 − s2 + s
2 2
The integral may be evaluated by means of the substitution s = sin(θ +

π/6), which turnsqthe denominator of the right-hand side into cos θ and the
3
numerator into ( 2 cos θ − 21 sin θ) dθ. The solution is
√
3 1
ln r = A − θ − ln cos θ,
2 2
where A is a constant of integration. Consider, for example, the question
of the largest value of b/a for which a fully plastic plane-stress solution is
possible. In this case p = 2k, so that r = a corresponds to s = 1 or θ = π/3,
while r = c = b corresponds to s = 0, or θ = −π/6. It follows that b/a is
given by
b 1 √
ln = ( 3π − ln 3) = 1.086.
a 4
The fully plastic solution is therefore possible only if b/a ≤ e1.086 , a result
not too different from the one furnished by the Tresca criterion.
To determine the greatest extent of the plastic domain in the limit as
b/a → ∞, we again set θ = π/3 at r = a, and θ = 0 at r = c, since σr = −k
there. We obtain
c 1 π

ln = √ − ln 2 = 0.560.
a 2 3
Determination of Displacement
The displacement field in a partly plastic tube governed by the Tresca
yield criterion and its associated flow rule was also determined by Koiter
[1953b]. Since the volume strain εr + εθ + εz is elastic throughout the tube,

it follows that
du u 1 − 2ν (1 − 2ν)(1 + ν)
+ = (σr + σθ + σz ) − εz = (σr + σθ ) − 2νεz .
r r E E
Substituting σθ from Equation (4.3.18), we obtain
d (1 − 2ν)(1 + ν) d 2
(ru) = (r σr ) − 2νεz r,
dr E dr
or, upon integration,
(1 − 2ν)(1 + ν) A
u(r) = rσr − νεz r + ,
E r
where A is a constant of integration. This solution is valid throughout the
tube. In the elastic zone, it must coincide with that given by
r 1+ν
u = rεθ = [σθ − ν(σr + σz )] = [(1 − ν)σθ − νσr ]r − νεz r.
E E
Substituting for σr , σθ from (4.3.29)1,2 leads to
2kc2 (1 − ν 2 )
A= .
E
In the case of an incompressible solid (ν = 21 ) in plane strain (εz = 0), the
displacement field reduces to
kc2
u(r) = . (4.3.31)
2Gr
An analogous procedure for the annular disk (plane stress) leads to
1−ν 2kc2
u(r) = rσr + .
E Er
Mises Criterion (Plane Strain)

If the tube material obeys the Mises yield criterion, then in the plane-
strain condition a plastic zone a ≤ r ≤ c forms when the pressure exceeds
pE as given by Equation (4.3.25). In the elastic zone the stresses σr and σθ
are given by (4.3.26), and
pc
σz = ν(σr + σθ ) = 2ν ,
(b/c)2 − 1
where pc is given by the right-hand side of (4.3.25) with c replacing a. Defin-
ing
k
p̄ = q ,
(b/c)4 + 13 (1 − 2ν)2
we may write the stress components as

! !
b2 b2
σr = p̄ 1 − 2 , σθ = p̄ 1 + 2 , σz = 2ν p̄.
r r
In the plastic region, the stress components cannot be determined with-

out considering the displacement. Since the Mises flow rule implies that
ε̇pkk = 0, the volume strain εkk = εr + εθ is purely elastic, and hence the
stresses can be written as
σr = sr + K(εr + εθ ), σθ = sθ + K(εr + εθ ), σz = −(sr + sθ ) + K(εr + εθ ),
K being the bulk modulus. The quantities sr , sθ , εr , and εθ form the basic
un-knowns of the problem. It is convenient to regard them as functions of
r and of c, with the latter as the time-like variable. sθ , however, may be
eliminated through the yield criterion, which takes the form
s2r + sr sθ + s2θ = k 2 ,
yielding
1 q 2

sθ = 4k − 3s2r − sr ;
2
the positive root is chosen because 2sθ + sr is positive in the elastic region.
The equilibrium equation (4.3.18) now becomes
∂sr ∂ 1 q 2

+ K (εr + εθ ) = 2
4k − 3sr − sr . (4.3.32)
∂r ∂r 2r
The strains εr and εθ furthermore satisfy the compatibility condition
∂
εr = (rεθ ). (4.3.33)
∂r
Finally, the flow rule must be invoked, in the form of the Prandtl–Reuss
equations (3.3.4), which in the present case become
2G ∂ ∂sr
(2εr − εθ ) = + 2Gλ̇sr ,
3 ∂c ∂c
2G ∂ ∂sθ
(2εθ − εr ) = + 2Gλ̇sθ ;
3 ∂c ∂c
recall that c is used to denote time. Eliminating λ̇ between the two equations
and substituting for sθ produces
∂sr G 2 2 ∂εr ∂εr
q
2 2 2
= 2 4k − 3sr − 2k + 3sr 4k − 3sr . (4.3.34)
∂c 4k ∂c ∂c
Equations (4.3.32)–(4.3.34) are three nonlinear partial differential equa-

tions in the unknown variables sr , εr and εθ in a ≤ r ≤ c for a ≤ c ≤ b.
Their solution can be effected numerically, as was done by Hodge and White
[1950].1 Hodge and White’s results are reviewed in Section 4.5, where they
are compared with a numerical solution based on the finite-element method.
A considerable simplification is achieved if the material is assumed to be
elastically as well as plastically incompressible, that is, if ν = 12 . In this case
the relation sz = 0 implies that both εez = 0 and εpz = 0, and as we know
from 3.3.4, when the latter condition holds, the Mises criterion coincides with
the Tresca criterion based on k. Hence the stresses are given by Equations
(4.3.29) and (4.3.30) in the elastic and plastic regions, respectively, with
α = 0 and ν = 21 .
A comparison of the results of this simplification with those obtained
numerically for ν < 21 shows that the values of σr and σθ are practically
indistinguishable. For the axial stress and the displacement, on the other
hand, the following corrections provide a good approximation:2
. .
σzcomp = 2νσzinc , ucomp = 2(1 − ν)uinc .
Unloading and Reloading

If a pressure p > pY is applied to the interior of the tube and then reduced
back to zero, the unloading process is elastic if p is not too great. The result-
ing field of residual stresses is obtained by subtracting the elastic stresses
given by (4.3.19) and (4.3.22), with p given by (4.3.27), from the elastic-
plastic stresses given, in the case of the Tresca criterion, by (4.3.29)–(4.3.30).
Using, furthermore, Equation (4.3.23), we obtain the residual stresses in the
form " ! #
p a2 r2
σr = −k 1 − 2 − ln 2 ,
p
" E
r ! a #
p a 2 r2
σθ = −k 1 + 2 − ln 2 − 2 , a<r<c
pE r !
a
p r2
σz = −2νk − 1 − ln 2 ,
pE a
and ! !
c2 p a2 a2
σr = −k 2
− − 2 ,
a p!
E r2 b!
c2 p a2 a 2
σθ = k 2
− 2
+ 2 , c<r<b
a pE !r b
c2 p a2
σz = 2νk − .
a2 pE b2
1
A similar analysis of the tube under closed-end and open-end conditions was performed
by Marcal [1965].
2
Hodge and White [1950] (see also Prager and Hodge [1951], Section 16).
Note that the residual axial stress is independent of the end condition; the
reason is that the axial strain εz , being purely elastic, is completely removed.
The largest value of |σθ − σr | can be seen to occur at r = a, where
it equals 2k(p/pE − 1). Also, at that location σr = 0 and σz = νσθ , so
that renewed yielding takes place if p/pE = 2, provided that 2k remains
the yield stress on reversed loading. The unloading is accordingly elastic
if p < 2pE = 2k(1 − a2 /b2 ), and this occurs for all pressures p up to the
ultimate pressure pU , given by (4.3.28), if the wall ratio b/a is such that
pU < 2pE , that is, if it is less than that which satisfies the equation
!
b2 a2
ln 2 = 2 1 − 2 .
a b
The largest wall ratio for which the unloading is elastic at all pressures is
found to be about 2.22; the corresponding ultimate pressure is about 1.59k.
When b/a > 2.22 and p > 2pE , unloading produces a new plastic zone,
a < r < c0 (with c0 < c), in which σθ − σr = −2k.
If the shakedown pressure is defined as pS = min(pU , 2pE ), then for
p ≤ pS not only is unloading elastic, but so is any subsequent reloading with
a pressure no greater than p. In other words, shakedown (as discussed in
3.5.3) takes place: the initial loading extends the elastic range of the tube,
with the limiting pressure for purely elastic expansion increased from pE to
p. This strengthening can be attributed to the development of compressive
hoop stresses σθ in the inner portion of the tube — an effect similar to that
of hoops around a barrel. The process is commonly known as autofrettage,
a French term meaning “self-hooping.”
Effect of Work-Hardening
A solution for a work-hardening material will be obtained following
Bland [1956]. We retain the assumption σθ > σz > σr and the Tresca
yield criterion and flow rule, but assume that the yield stress k is a function
of a hardening variable κ, defined, say, by (1.5.6) (“plastic work”); in the
present case, this definition reduces to
κ̇ = k(κ)(|ε̇pθ | + |ε̇pr |) = 2k(κ)|ε̇pθ |,
since ε̇pr = −ε̇pθ . As long as ε̇pθ ≥ 0 (no reverse yielding), the preceding
relation can be integrated to give
κ dκ0
Z
= 2εpθ ,
0 k(κ0 )
so that κ is in a one-to-one relation with εpθ , and k can also be viewed as a

function of εpθ .
A similar relation in uniaxial tension, with σ1 = σ, σ2 = σ3 = 0, gives

κ̇ = k(κ)(|ε̇p1 | + |ε̇p2 | + |ε̇p3 |) = 2k(κ)(|ε̇p1 |),
since ε̇p2 = ε̇p3 = − 12 ε̇p1 . The relation between κ and εpθ in the tube is thus
the same as between κ and εp1 in uniaxial tension. If σ1 = σY (εp1 ) describes
the uniaxial relation between stress and plastic strain, then, in the tube,
k = 21 σY (εpθ ).
In the elastic zone the stress field is given by Equations (4.3.29), with k
replaced by 12 σE , where σE = σY (0) is the initial yield stress. In the plastic
zone the equilibrium equation (4.3.18) becomes
dσr 1
= σY (εpθ ), (4.3.35)
dr r
with the boundary conditions σr = −p at r = a and σr = 21 σE (1 − c2 /b2 ) at
r = c. The elastic relations for εeθ = εθ − εpθ and εer = εr + εpθ , together with
the strain-displacement relations εθ = u/r and εr = du/dr, lead to
u 1
= εpθ − νεz + [(1 − 2ν)(1 + ν)σr + (1 − ν 2 )σY (εpθ )]
r E
and
du 1
= −εpθ − νεz + [(1 − 2ν)(1 + ν)σr − ν(1 + ν)σY (εpθ )].
dr E
Eliminating u between these equations, with the substitution from (4.3.35),
leads further to
dεp dεp
" #
1 − ν2
2εpθ +r θ =− 2σY (εpθ ) + rH(εpθ ) θ ,
dr E dr
where H = dσY /dεpθ is the hardening modulus. The last equation can be
integrated, subject to the boundary condition that εpθ = 0 at r = c, to yield
" #
1 − ν2 c2
εpθ = σE 2 − σY (εpθ ) ,
E r
With the help of the last equation, r can be eliminated as a variable in favor
of εpθ , so that (4.3.32) can be integrated into a relation between σr and εpθ .
If εp0 denotes the value of εpθ at r = a, then the pressure is given by
εp0 E + (1 − ν 2 )H(εpθ )
!
1 c2 1
Z
p= 1− 2 σE + σY (εpθ ) dεpθ , (4.3.36)
2 b 2 0 Eεpθ + (1 − ν 2 )σY (εpθ )
with c given by
c2 1 E

= σY (εp0 ) + εp .
a2 σE 1 − ν2 0
The two last equations provide the relation between p and c through the
parameter εp0 . The integration is in general carried out numerically, but an
explicit relation is easily obtained if the hardening is linear, that is, if H is
constant. Details are left to an exercise.
Section 4.4 / Elastic-Plastic Bending 229

1. Discuss the behavior of an elastic-plastic spherical shell under both
external and internal pressure in comparison with its behavior under
internal pressure only.
2. Evaluate Equation (4.3.10) for the case of linear work-hardening, the

uniaxial relation being σY = Hεp , where the plastic modulus H is
constant.
3. A spherical shell of initial inner and outer radii a0 , b0 is made of a

material whose stress-strain curve is given by σ = f (ε) with no elastic
range.
(a) Find an equation to replace (4.3.12).

(b) Find the condition for plastic instability.
4. A hollow cylinder of initial inner and outer radii a0 , b0 is assumed to be

in plane strain and made of an isotropically hardening Mises material
that is both plastically and elastically incompressible.
(a) Show that if the uniaxial

√ stress-strain
√ curve is given by σ = f (ε),
then σθ − σr = (2/ 3)f (2εθ / 3).
(b) Assuming large strains, with εθ = ln(r/r0 ), find the analogues of
Equations (4.3.10) and (4.3.12).
(c) Determine the criterion for plastic instability.
5. Discuss how Equation (4.3.36) would be integrated for a cylinder made

of a material with no elastic range.
6. Integrate Equation (4.3.36) and thus find an explicit p-c relation for
the case of linear work-hardening [i. e., σY (εp ) = σE + Hεp , with H
constant].
Section 4.4 Elastic-Plastic Bending

4.4.1. Pure Bending of Prismatic Beams
General Concepts
A straight bar is said to be in a state of pure bending if the end sections
are subject to normal tractions whose resultants are equal and opposite mo-
ments about an axis perpendicular to the longitudinal axis of the bar. It
follows from equilibrium requirements that the stresses on every transverse

section add up to the same moment (constant bending moment). In practice
a condition of pure bending can be achieved by placing a bar on symmet-
rically located supports, with equal overhangs of length, say, a; if equal
transverse forces F are applied at the ends of the bar, then the portion be-
tween the supports is in a state of constant bending moment F a [see Figure
4.4.1(a)].
y

E
E
(((
a - a - zX ( E
( r(y
(
,z
(
)
] Neutral plane
J
y (
XXX 0 0
X EX
(0,0)
4 g M XX
E XXXXX
? ? E
F F EX
y
X X Plane of bending
(a)
E
E

Center of E
curvature C
C
(b)
Q/\\/\ P/\/P/\C
1/κ C
C

9
@9D CC Neutral plane
@
:D εx
:D@
(c)
Figure 4.4.1. Pure bending: (a) possible beam geometry and loading; (b) local
geometry in the cross-section plane; (c) local geometry in the
plane of bending.
If the bar is prismatic (of uniform cross-section, in terms of both geome-

try and material properties) and sufficiently long in comparison with cross-
sectional dimensions, then it can be assumed, as with the cylinder discussed
in Section 4.3, that the stresses are independent of the longitudinal coor-
dinate, except in the vicinity of the end sections. Let this coordinate be
denoted x; then the preceding assumption, as well as the equilibrium equa-
tions and the boundary conditions of zero traction on the lateral surface, is
satisfied if
σx = σ(y, z),
with all other stress components equal to zero. If the moment components
about the y- and z-axes are My and Mz , respectively, then equilibrium is
satisfied if
Z Z Z
σ(y, z) dA = 0, zσ(y, z) dA = My , − yσ(y, z) dA = Mz .
A A A
(4.4.1)
Since every cross-section of the bar has the same stress distribution, it
can be expected to have the same strain distribution as well. When the strain
components are infinitesimal and independent of x, three of the compatibility
conditions (1.2.4) reduce to
∂ 2 εx ∂ 2 εx ∂ 2 εx
= = = 0,
∂y 2 ∂z 2 ∂y∂z
so that the longitudinal strain εx must be of the form a + by + cz, with a, b
and c constant. It will be written as
εx = −κ[(y − y0 ) cos α + (z − z0 ) sin α]. (4.4.2)
Here (y0 , z0 ) are the coordinates of the intersection of the cross-section with
a fiber that does not elongate (a neutral fiber ), α is the angle between the
plane of bending and the xy-plane, and κ is the curvature (the reciprocal of
the radius of curvature) of the neutral fibers in the plane of bending. For an
illustration of the geometry, see Figure 4.4.1(b) and (c).
The identical vanishing of the shear stresses τxy and τxz implies, if the
material properties are isotropic, the vanishing of the corresponding strains
γxy and γxz , so that the displacements satisfy
∂v ∂u ∂w ∂u
+ = 0, + = 0.
∂x ∂y ∂x ∂z
Consequently,
∂2v ∂2u ∂εx
2
= − =− = κ cos α,
∂x ∂x∂y ∂y
∂2w ∂2u ∂εx
2
= − =− = −κ sin α.
∂x ∂x∂z ∂z
Combining, we find that
∂2
κ= (v cos α − w sin α).
∂x2
The quantity in parentheses is the displacement of a fiber from its reference
position in the plane of bending. Supposing this to be the xy-plane (α = 0),
we may write κ = ∂ 2 v/∂x2 . Strictly speaking, the curvature of the bent
fiber is given by
∂ 2 v/∂x2
,
[1 + (∂v/∂x)2 ]3/2
and consequently the infinitesimal-deformation theory is valid as long as

the slope of bent fibers with respect to the longitudinal axis is small, or,
equivalently, as long as the radius of curvature remains large compared with
the length of the bar.
Elastic Bending
If the bar is made of a single linearly elastic material with Young’s modu-
lus E, then Equation (4.4.2) can be combined with the stress-strain relation
σx = Eεx to give
σ(y, z) = −Eκ[(y − y0 ) cos α + (z − z0 ) sin α]. (4.4.3)
It is convenient to place the origin of the yz-plane at the centroid of the

cross-section, that is, Z Z
y dA = z dA = 0.
A A
Inserting (4.4.3) into the first of Equations (4.4.1) yields y0 cos α+z0 sin α = 0
as the equation governing the coordinates of the neutral fibers — that is,
they lie in the plane (the neutral plane) that is perpendicular to the plane
of bending and that contains the centroidal fiber. Equation (4.4.3) can
accordingly be simplified to
σ(y, z) = −Eκ(y cos α + z sin α). (4.4.4)
It is also convenient to make the y- and z-axes the principal axes of area,
that is, Z
yz dA = 0.
A
Inserting (4.4.4) into the remaining Equations (4.4.1) leads to
My = −EIy κ sin α, Mz = EIz κ cos α, (4.4.5)
where Z Z
2
Iy = z dA, Iz = y 2 dA
A A
are the principal second moments of area (often called the principal moments
of inertia) of the cross-section.
The inclination of the bending plane is given, from (4.4.5), by tan α =
My Iz /Mz Iz . The bending plane is perpendicular to the moment vector,
whatever that may be, if Iy = Iz ; this equality occurs whenever the cross-
section has two or more nonperpendicular axes of symmetry, as in the case of
a circle or a regular polygon (equilateral triangle, square, etc.). Otherwise,
the moment vector is normal to the bending plane only if it is itself directed
along a principal axis, that is, if My = 0 or Mz = 0.
Combining Equations (4.4.4) and (4.4.5) gives the stress distribution
M z y My z
σx = − + . (4.4.6)
Iz Iy
All the preceding results are valid as long as the extreme values of the
stress given by (4.4.6) remain in the elastic range. More generally, we may
assume a nonlinear stress-strain relation σ = f (ε). The function f , with
the right-hand side of (4.4.2) as its argument, defines σ(y, z), which when
inserted in (4.4.1) furnishes three coupled, nonlinear equations for the para-
meters κ, α, and y0 cos α + z0 sin α, the last of which is the displacement be-
tween the neutral plane and the centroidal axis. Coupled nonlinear equations
are generally difficult to solve. A simplification is achieved if the bending is
symmetric.
Symmetric Bending
An axis of symmetry is necessarily a principal axis of the cross-section.
Let the section be symmetric about the y-axis, and let My = 0, Mz = M .
Then α = 0, and Equations (4.4.5)–(4.4.6) simplify to M = EIκ, σx =
−M y/I, where I = Iz , in the elastic range. However, symmetry alone
dictates α = 0, independently of material properties, and this condition
may be assumed to hold under nonlinear material behavior as well. With
the stress-strain relation σ = f (ε) as above, the parameters κ and y0 are
governed by
Z Z
f (−κ(y − y0 )) dA = 0, f (−κ(y − y0 ))y dA = −M. (4.4.7)
A A
If y0 can be eliminated between these equations, then the result is the

moment-curvature relation
M = Φ(κ). (4.4.8)
Doubly Symmetric Sections

An even greater simplification is achieved if the z-axis, also, is an axis
of symmetry and if the stress-strain relation is the same in tension and
compression, that is, if f is an odd function. In that case y0 = 0 by symmetry,
and the function Φ is given explicitly by
Z
Φ(κ) = f (κy)y dA.
A
Consider, for example, a bar of rectangular section of width b and depth

2c (so that I = 2bc3 /3), made of an elastic–perfectly plastic material. The
elastic moment-curvature relation and stress distribution are valid as long
as |σ|max = 3|M |/2bc2 ≤ σY , or |M | ≤ ME , where ME = 2σY bc2 /3 is the

initial yield moment.
When M > ME , the elastic relation σ = −Eκy holds at values of
y at which this quantity does not exceed σY in magnitude, that is, for
def
|y| ≤ y ∗ = σY /E|κ|; this range forms the elastic core. The plastic zones
are −c < y < −y ∗ and y ∗ < y < c, where σ = σY sgn κ and σ = −σY sgn κ,
respectively; the stress distribution is shown graphically in Figure 4.4.2(a).
Because of symmetry, the moment is given by
y∗
" #
Z c Z Z c
2
M = −2b σy dy = 2b Eκ y dy + σY sgn κ y dy .
0 0 y∗
Upon integration and substitution for y ∗ , this becomes

" 2 #
1 κE

M = MU sgn κ 1 − , |κ| > κE , (4.4.9)
3 κ
where κE = σY /Ec is the absolute value of the curvature corresponding

to |M | = ME , and MU = σY bc2 is the ultimate moment which is attained
asymptotically as |κ| → ∞. The moment-curvature relation given by (4.4.9)
is shown in Figure 4.4.2(b), and is quite similar to the torque-twist relation
(4.2.4) for an elastic–perfectly plastic shaft of solid circular cross-section.
The difference is that while the ratio TU /TE between the ultimate and initial
yield torques is 4/3 in the torsion of a circular shaft, in the bending of
a rectangular beam the corresponding ratio MU /ME , known as the shape
factor , is 3/2.
M
y y
c c
MU
y∗
@
@ ME
σY σY
@ σ
σ
−σY @
−σY
@
−y ∗ @
−c
κ −c
κE
(a) (b) (c)
Figure 4.4.2. Elastic-plastic beam: (a) stress distribution; (b) moment-

curvature relation; (c) ultimate stress distribution.
Suppose, now, that the material work-hardens according to the law

(2.1.4), with 0 < n < 1; the limiting cases n = 0 and n = 1 represent per-
fect plasticity and unlimited elasticity, respectively. The moment-curvature
relation can easily be shown to be

" 2 #
κ n

ME κE

M= sgn κ 3 − (1 − n) , |κ| > κE , (4.4.10)
2+n κE κ
where κE = σE /Ec, ME = 2σE bc2 /3. Note that there is no ultimate moment
when n > 0.
Elastic-plastic moment-curvature relations analogous to (4.4.9) can be
found for other doubly symmetric sections, although the integration often
gets cumbersome. It is fairly easy, however, to determine the ultimate mo-
ment and hence the shape factor. In the limit as κ → ∞, the stress distrib-
ution becomes as shown in Figure 4.4.2(c), that is, consisting of two blocks
of constant stress of value σY and −σY , respectively, each distributed over
one-half the area, and statically equivalent to equal and opposite forces of
magnitude σY A/2, each acting at the centroid of the half-area. The ultimate
moment is therefore
1
MU = σY Ad, (4.4.11)
2
where d is the distance between the centroids of the half-areas. Since ME =
σY I/c, where c is one-half the depth, the shape factor is Acd/2I. The
quantity 21 Ad is often called the plastic modulus of the section and denoted
Z.
For a bar of solid circular cross-section with radius c, we have I = πc4 /4,
A = πc2 , and d/2 = (4/3π)c; hence the shape factor is 16/3π ≈ 1.7. For
a thin-walled circular tube of wall thickness t, I = πc3 t, A = 2πct, and
d/2 = (2/π)c, so that MU /ME = 4/π ≈ 1.27. As a rule, the shape factor is
greater the more material is concentrated near the center, and the smaller
(closer to 1) the more material is concentrated near the extreme fibers. For
rolled structural shapes, typical values are near 1.2 for I-beams and between
1.1 and 1.15 for wide-flange beams. The ideal sandwich beam (or ideal I-
beam) discussed in 3.5.1 has a shape factor of 1. The moment-curvature
relation for such a beam is
(
EIκ, |κ| ≤ MU /EI,
M= (4.4.12)
MU sgn κ, |κ| ≥ MU /EI,
analogous to the torque-twist relation for a thin-walled circular tube, dis-

cussed in 4.1.1.
Sections Without Double Symmetry
For sections without symmetry about the z-axis the determination of a
moment-curvature relation is more complicated, since y0 as a function of κ
must first be found from the first of Equations (4.4.7) and then inserted in
the second. For an elastic–perfectly plastic material it is easy, however, to
find the asymptotic value of y0 in the limit of infinite curvature, since the
6 A 6
A
A √
A a = h/ 2
A
h A
A
A ?
A
A
? A
b -
Figure 4.4.3. Isosceles triangle: ultimate neutral plane.
neutral plane must then divide the cross-section area into equal halves, as
in the case of sections with double symmetry. Unlike this special case, in
general this plane does not go through the centroid. The ultimate moment
is still given by (4.4.11), with d = d1 + d2 , where d1 and d2 are the distances
from the centroids of each of the half-areas to the ultimate neutral plane.
The relations κE = σY /Ec, ME = σY I/c, and the expression 12 Acd/I for the
shape factor, are valid if c is interpreted as the distance from the centroid to
the farthest fiber in the xy-plane. When |κ| exceeds κE somewhat, a plastic
zone forms on the side of the farthest fiber, but the rest of the section remains
elastic as long as the magnitude of the stress at the opposite extreme fiber
does not attain σY . It is only when this last condition is reached that the
second plastic zone forms. The calculations are often involved.
Consider, for example, an isosceles triangle of base b and height h. For a
section of this shape, A = bh/2, c = 2h/3, and I = bh3 /36. To determine d,
we must first find the location of the ultimate neutral plane. If the distance
from this plane to the apex is a (see Figure 4.4.3), then the area of the
isosceles triangle of height a and base ba/h must equal one-half √ the area
of the whole 2
√ triangle, that is, 2 ba /h = 4 bh; hence a = h/ √ 2, while d1 =
1 1
a/3 = h/3 2. For a trapezoid of√height h − a = (1 − 1/ 2)h and major

and minor bases b and ba/h = b/ 2, respectively, it √ can be shown that the
√ the centroid to the minor base is (8 − 5 2)h/6. Consequently
distance from √
d = 2(2− 2)h/3, and the shape factor is easily calculated to be 4(2− 2) ≈
2.34. The elastic-plastic solution for the isosceles triangle can be found, for
example, in Nadai [1950], Volume I, page 358.
Unloading and Residual Stresses

Consider a symmetrically bending elastic-plastic beam whose moment-
curvature relation has the general form M = Φ(κ), of which (4.4.9), (4.4.10)
and (4.4.12) are special cases. Suppose, in particular, that initial loading
has produced a curvature κ0 . If the moment M = Φ(κ0 ) is removed, then
the residual stress distribution can be calculated on the assumption that the
unloading is elastic, that is, stresses equal to Φ(κ0 )y/I are added to those
produced by the initial loading. The assumption is justified if the resulting
stresses are in the elastic range.
The springback, described by the change in curvature, is likewise elas-
tic, and given by −Φ(κ0 )/EI. The residual curvature, which defines the
permanent deformation of the beam, is
Φ(κ0 )
κres = κ0 − .
EI
If, for example, the moment-curvature relation is (4.4.10), the ratio of resid-
ual to initial curvature is (if κ0 > 0)
" 1−n 3 #
κres 1 κE κE

=1− 3 − . (4.4.13)
κ0 2+n κ0 κ0
In the perfectly plastic case (n = 0), half of the curvature is recovered when
κ0 = 2.88κE .
The calculation of residual stresses is particularly simple for doubly sym-
metric sections, since there is then no need to find y0 , and the residual
stresses are given by
y
σres = Φ(κ0 ) − f (κ0 y).
I
A typical residual-stress distribution is shown in Figure 4.4.4. Note that the
y (mm) y (mm) y (mm)

60 60 60
+ =

-- --
- % %
40 40 -- % %
40
@ -
- % E

@ -- % E
@ 240 σ (MPa)
- % σ (MPa) E σ (MPa)
−240 --@ 204.5 306.7
%
−35.5 E 66.7
--
@- % -E
-
−40 @
-
- %
−40-E
-
% %
−60 - %
−60 %
−60
Figure 4.4.4. Residual-stress distribution in an elastic–perfectly plastic rectan-

gular beam.
stresses near the center of the section keep their sign, while near the extreme
fibers the loading is reversed. If the material is perfectly plastic, then the
residual stresses at the extreme fibers are ±(1 − |M/ME |)σY . Reverse yield-
ing can occur only if the shape factor is greater than 2, a condition unlikely
to occur in doubly symmetric sections.
Asymmetric Plastic Bending

When the moment is not perpendicular to an axis of symmetry, the
elastic-plastic solution becomes exceedingly difficult even for a simple cross-
section. The initial yield condition can be determined by finding, for given
My and Mz , the maximum absolute value of the right-hand side of (4.4.6)
and equating it to σY , a procedure that gives a relation between My and
Mz . For a rectangle, this relation is
2c
|Mz | + |My | = ME
b
if ME is defined for symmetric bending in the xy-plane as before; the relation
is represented by the rhombus shown in Figure 4.4.5(a). The fully plastic
condition can be studied according to the following procedure, illustrated in
Figure 4.4.5(b).
y
My
Ultimate
0 =(b/h)M
MU U r neutral
B β plane
(b/h)ME Ultimate BMB B)
"b
" b yield locus

"
" b
b B B
B Oα
"
" b
b Mz z B B
b "ME MU Bd B
b "
b
b "Initial
" B Br
b "yield BN *

b" locus
Centroids

of half-areas
(a)
(b)
Figure 4.4.5. Asymmetric plastic bending: (a) initial and ultimate yield loci for
a rectangular beam; (b) ultimate neutral place and centroids of
half-areas.
The ultimate neutral plane is assumed to form an angle α with the z-axis.
Since this plane divides the cross-section into two equal areas, its location
can be determined by geometry, as can the locations of the centroids of the
half-areas. If the distance between these centroids is once more denoted d,
then the magnitude of the fully plastic moment is again σY Ad/2. It must be
kept in mind, however, that d depends on α. The line joining the centroids
of the half-areas forms an angle β with the y-axis, which depends on α. The
moment vector is perpendicular to this line, and its components are, if the
half-area below the neutral plane is in tension,
1 1
My = − σY Ad sin β, Mz = σY Ad cos β.
2 2
Since both d and β depend on α, the preceding equation forms a parametric

representation, in terms of α, of a closed curve in the My Mz -plane called the
interaction diagram.
For a rectangle of width b and depth h, it can be shown that
!
h b2 b2 h
d cos β = 1 − 2 tan2 α , d sin β = tan α, tan α ≤
2 3h 3h b
and
!
h2 b h2 h
d cos β = cot α, d sin β = 1 − 2 cot2 α , tan α ≥ .
3b 2 3b b
It is easy to eliminate α between the expressions for My and Mz to obtain

the interaction relation directly. If we define MU = σY bh2 /4 as before,
MU0 = σY b2 h/4 = (b/h)MU , my = My /MU0 , and mz = Mz /MU , then the
relation is given by the two equations

3 my 3 my
|mz | + m2y = 1, ≤ 1, and |my | + m2z = 1, ≥ 1.
4 mz 4 mz
The relation is represented by the outer curve of Figure 4.4.5(a). The inter-
action curves for many other sections with double symmetry are found to lie
fairly close to this curve.
4.4.2. Rectangular Beams Under Transverse Loads
End-Loaded Cantilever: Elastic Solution

The elastic theory of beams under transverse loads has its origin in Saint-
Venant’s research on the bending of a prismatic cantilever carrying a concen-
trated force F = jFy + kFz at its free end. Saint-Venant began by assuming
that Equation (4.4.6) for the longitudinal stress σx remains valid, along with
σy = σz = τyz = 0. If the beam spans 0 < x < L, with x = 0 denoting the
built-in end, the moment acting on the cross-section at x is (L − x)i × F,
or Mz = (L − x)Fy , My = −(L − x)Fz . Since σx depends on x, equilibrium
requires the presence of shear stresses τxy , τxz , independent of x, that satisfy
Z Z
τxy dA = Fy , τxz dA = Fz .
A A
Since the problem is linear, it may be treated as a superposition of the

two problems corresponding to Fz = 0 and Fy = 0, and since the label-
ing of the principal axes as y or z is arbitrary, only one of these problems
need be treated — for example, Fz = 0, Fy = F . It must be kept in mind
that τxz cannot be assumed to vanish. The application of the one nontrivial
equilibrium equation to the stresses, and of the compatibility conditions to

the strains resulting from them, reduces the problem to one of solving the
Laplace equation in the yz-plane subject to certain boundary conditions, in
addition to that of finding the stress function of the elastic torsion problem
(see 4.2.2); the latter is unnecessary if the y-axis is a symmetry axis of the
cross-section and is the line of action of the force, as will be assumed. De-
tails of the theory, along with solutions for circular, elliptic, rectangular and
other cross-sections can be found in most books on elasticity.1 In particular,
for any cross-section the deflection satisfies ∂ 2 v/∂x2 = F (L − x)/EIz , so
that within the limitations of small-deformation theory, the local moment-
curvature relation is the same as in pure bending.
The results for the rectangular beam show that the z-average of τxz
vanishes, while that of τxy is given by
1
Z b/2 3F 2
def
τ̄xy = τxy dz = (c − y 2 ),
b −b/2 4bc2
where b and c are the width and half-depth of the cross-section, respectively,
as before. Moreover, the shear stresses τxy become independent of z in the
limit of both infinitely narrow (b/c → 0) and infinitely wide (b/c → ∞)
rectangles, corresponding respectively to solutions for plane stress and plane
strain. For such beams, then, the stress distribution is
F (L − x)y F 2
σ= , τ= (c − y 2 ), (4.4.14)
I 2I
where σ = σx , τ = τxy , and I = Iz = 2bc3 /3.
End-Loaded Cantilever: Elastic-Plastic Solution
For the state of stress just derived, both the Mises and the Tresca yield
criteria are given by Equation (3.3.5). It is not difficult to see that in the
cantilever under study, the left-hand side of this equation attains its maxi-
mum wherever |σ| is greatest, that is, at the corners (0, ±c). Consequently,
def
plasticity begins when F = FE = 2σY bc2 /3L. If the load F is increased
beyond this value, plastic zones develop at the top and bottom faces of the
beam, spreading from the corners, as shown in Figure 4.4.6.
In the plastic zones the yield criterion requires that the stresses obey
σ ∂σ τ ∂τ σ ∂σ τ ∂τ
+ = 0, + 2 =0
σY2 ∂x τY2 ∂x σY2 ∂y τY ∂y
in addition to the equilibrium equations
∂σ ∂τ ∂τ
+ = 0, = 0.
∂x ∂y ∂x
1
For example, Love [1927], Chapter 15; Timoshenko and Goodier [1970], Chapter 11;
or Boresi and Chong [1987], Chapter 7.
y F
6
c
ξ -
3

H y = y∗ (x)
H
Y

Plastic x
zones Q 0 L
Q
y = −y ∗ (x)
Q
s
Q

−c
Figure 4.4.6. Elastic and plastic zones in an end-loaded cantilever.
These equations can be satisfied simultaneously only if σ and τ are both

constant in each plastic zone, and since τ must vanish on the top and bottom
faces, it must vanish everywhere in the plastic zones, and therefore σ = ±σY .
In particular, if the elastic-plastic boundary at x is located at y = ±y ∗ (x),
then σ = σY in −c ≤ y ≤ −y ∗ (x) and σ = −σY in y ∗ (x) ≤ y ≤ c. In
the elastic core −y ∗ (x) ≤ y ≤ y ∗ (x) the y-distribution of σ is linear as in
the elastic case, that is, σ = −σY y/y ∗ (x). The resultant moment about the
z-axis at x is !
y ∗2
Z c
2
M = −b yσ dy = σY b c − ,
−c 3
and since for equilibrium this must equal F (L−x), it follows that the elastic-
plastic boundary is given by
s
F (L − x)

∗
y (x) = 3 c2 − b . (4.4.15)
σY
def
In particular, y ∗ (x) = c at x = L − 2σY bc2 /3F = L − ξ; the length ξ repre-
sents the extent of the elastic zone at the given value of F . We can accord-
ingly write, more simply,
s
2(L − x)
y ∗ (x) = c 3 − .
ξ
The elastic-plastic boundary shown in Figure 4.4.6 is thus given by two sym-
metrically located segments of a parabola whose vertex is on the extension
of the beam axis beyond the built-in section.
As in the case of the twisted shaft, the deformation in the plastic zone
is contained by the elastic deformation as long as the elastic core is of non-
vanishing thickness. The regime of contained plastic deformation ends when
y ∗ (0) = 0, that is, when the vertex of the parabola reaches the built-in sec-
def
tion of the beam. The corresponding value of the load is F = σY bc2 /L = FU ,
the ultimate load . When it is attained, unrestrained plastic flow may begin,
taking the form of a rigid rotation of the remaining elastic region about the
point (0, 0), with an attendant stretching and shrinking of the x-fibers in
the upper and lower plastic zones, respectively. The resulting mechanism is
known as a plastic hinge. Plastic hinges are used extensively in the plastic
analysis of beams and framed structures under various loadings, as is shown
later in this book.
The preceding derivation of the ultimate load neglects the growth of the
shear stress in the elastic core. In fact, since τ = 0 in the plastic zones,
the transverse force must be balanced by the shear stress in the elastic core
only. This shear stress has a parabolic distribution in −y ∗ (x) ≤ y ≤ y ∗ (x),
with a maximum value, at (x, 0), of 3F/4by ∗ (x), so that the maximum shear
stress in the beam is τmax = 3F/4by ∗ (0). As the elastic core shrinks, this
maximum shear stress grows until it attains τY , when a secondary plastic
zone forms. With the help of Equation (4.4.15), the value of the force F at
which this occurs is found to be given by
2
F 16 F 16

+ 2
− 2 = 0,
FU 3β FU 3β
where β = σY c/τY L. The only positive root of the equation is
r !
F 8 3
= 1 + β2 − 1 .
FU 3β 2 4
For span-to-depth ratios greater than 3, we have √ c/L < 1/6, so that in
a beam made of a Mises material (σY /τY = 3), β 2 will be no greater
than 1/12. The secondary plastic zone will therefore not form before F =
0.985FU , and the effect of the shear stress in the elastic core on the formation
of the plastic hinge may accordingly be neglected.
As a further refinement we note that since the shear stress in the elastic
core varies with x, a transverse normal stress σy must develop there as well.
However, the effect of this stress on the criterion for the initiation of the
inner plastic zone, and consequently on the ultimate load for reasonably
long beams, is negligible for sufficiently great span-to-depth ratios.1
In order to determine the deflection, it will be assumed that the moment-
curvature relation derived for pure bending holds locally in the elastic-plastic
def
portion of the beam as well. Letting κ = v̄ 00 (x), where v̄(x) = v(x, 0), and
using Equation (4.4.9) with the relevant substitutions, we obtain

1
, x < L − ξ,


σ Y
 p
3 − 2(L − x)/ξ
v̄ 00 (x) = x
Ec  L− ,
 x > L − ξ.
ξ
1
For a “stubby” cantilever whose span is comparable to its depth, beam theory cannot
be used; the problem of the collapse load for such a beam is studied as a limit-analysis
problem in 6.1.2.
Note that v̄ 00 (0) → ∞ as F → FU ; this is consistent with the notion of a

plastic hinge, since infinite curvature at a point on a curve corresponds to a
slope discontinuity.
Rather than determine the whole function v̄(x), we calculate only the
tip deflection ∆ = v̄(L) − v̄(0) by means of the fundamental relation
Z L
v̄(L) − v̄(0) = v̄ 0 (x) dx,
0
which by integration by parts becomes

Z L
∆= (L − x)v̄ 00 (x) dx, (4.4.16)
0
since v̄ 0 (0) = 0 and v̄ 0 (L) = 0L v̄ 00 (x) dx. Hence, for 2L/3 <= ξ <= L, a
R
change of variable from x to L − x leads to

"Z #
σY ξ x2 x L
Z
∆ = dx +
p dx
Ec 0 ξ ξ 3 − 2x/ξ
" s #
FE 2 F F

= ∆E 5− 3+ 3−2 ,
F FE FE
where ∆E = σY L2 /3Ec is the deflection when F = FE . Since FU = 3FE /2,

we see that the tip deflection increases by a factor of 20/9 as F goes from
FE (beginning of plastic deformation) to FU (incipient plastic flow).
Uniformly Loaded Cantilever
A theory of elastic beams carrying a transverse load, comprising both
body force and surface traction, that is uniformly distributed along the
length is due to Michell. For such a beam, proportionality between moment
and curvature no longer holds exactly, although it is a good approximation
when the beam is long compared to its cross-sectional dimensions. For a
cantilever carrying a downward load q per unit length, the resultant shear
force at x (positive upward) is Q = −q(L − x), and the bending moment is
M = − 21 q(L − x)2 . If the beam is of narrow rectangular cross-section and
carries the load on its upper surface y = c, then the plane-stress solution
gives the following stress distribution:
!
My q 3y y3 Q 2 2 q 3
σx = − + − 3 , τxy = (c −y ), σy = (y −3c2 y−2c3 ).
I b 10c 2c 2I 6I
Note that σy equals −q/b at y = c and vanishes at y = −c, as it should.

The shear stress has the same parabolic distribution as in the end-loaded
cantilever, but the longitudinal stress no longer varies linearly over the cross-
section. The correction, however, becomes negligible at all sections suffi-
ciently far away from the free end, that is, where (L − x)2 c2 . At such
sections, similarly, the magnitude of σy becomes negligible in comparison

with that of σx , except near y = 0.
With the correction to σx and the effects of τxy and σy neglected, it is
easy to determine that plasticity begins at (0, ±c) when q = 2ME /L2 =
def
4σY bc2 /3L2 = qE . With ξ again denoting the length of the elastic portion
of the beam and y ∗ (x)pthe half-depth of the elastic core in the elastic-plastic
portion, we have ξ = qE /qL and
2
y∗
2
x

+2 L− = 3.
c ξ
The elastic-plastic boundary therefore consists of two arcs of an ellipse cen-

tered at (L,√0), with semimajor and semiminor axes respectively
√ √ equal to
∗
3/2ξ and 3c. Collapse occurs when y (0) = 0, or ξ = 2/3L. Thus
the ultimate load intensity is qU = 3qE /2. Indeed, the ratio between the
ultimate and initial-yield values of the load is necessarily equal to the shape
factor MU /ME in any beam that is statically determinate.
The deflection is derived as for the end-loaded cantilever. The curvature
distribution is
 1

 p
 , x < L − ξ,
00 σ Y 3 − 2(L − x)2 /ξ 2
v̄ (x) = −
Ec  (L − x)2

 , x > L − ξ.
ξ2
The tip deflection is therefore

 s 
σY ξ2 L2
∆=− 3 − 2 3 − 2  .
Ec 4 ξ2
√
As ξ decreases from L (q = qE ) to 2/3L (q = qU ), the deflection may be
seen to double.
Uniformly Loaded Simply Supported Beam

A uniformly loaded beam of span 2L that is simply supported at its
ends is equivalent to a uniformly loaded cantilever of length L carrying, in
addition, an end load that is opposed to the distributed load and equal in
magnitude to the resultant of the latter (see Figure 4.4.7). The moment
distribution at any state, as well as all other quantities while the beam is
completely elastic, can therefore be obtained by superposing the solutions
of the two previously treated problems, with F = qL. The moment at x is
1 1
M (x) = qL(L − x) − q(L − x)2 = q(L2 − x2 ).
2 2
y y F=qL
⇐⇒
6
??????????????????????? ????????????
x x
4 g
Figure 4.4.7. Uniformly loaded simply supported beam, equivalence with can-
tilever.
The maximum moment occurs at x = 0 and equals qL2 /2. Consequently,

qE = 2ME /L2 and, as expected, qU = 3qE /2. Whenpq > qE , the elastic-
plastic portion extends over 0 < x < η, where η = 1 − q/qE L, and the
boundary is described by
2
y∗
2
x

− 3ρ = 3(1 − ρ),
c L
where ρ = q/qU . The boundary√ is thus given by portions of hyperbolas

whose asymptotes are y = ± 3(c/L)x. At collapse, the boundary attains
the asymptotes.
The curvature distribution is
1


 p , x < η,
ρx2 /L2 )
3(1 − ρ + !

00 σY 
v̄ (x) = 3 x2
Ec 
 ρ 1 − , x > η.
2


2 L
The deflection is once more given by Equation (4.4.16), namely,

"Z ! #
σY η L−x 3 L x2
Z
∆= p dx + ρ (L − x) 1 − 2 dx .
Ec 0 3(1 − ρ + ρx2 /L2 ) 2 η L
p
Carrying out the integration and inserting η = 1 − 2ρ/3L, we obtain
" s s s #
σY 2 1 3ρ − 2 ρ 1 3ρ + 1 2
∆= L √ sinh−1 + 1− √ ρ+ρ− − 1− .
Ec 3ρ 3(1 − ρ) 3 2ρ 3 3ρ
In this case the deflection tends to infinity as ρ → 1, but remains of the

order of ∆E = (5/12)σY L2 /Ec until q gets quite close to qU ; for example,
∆ ≈ 2.0∆E when ρ = 0.95.
4.4.3. Plane-Strain Pure Bending of Wide Beams or Plates
Initially Flat Plates

If the width b of a rectangular beam is considerably greater than its

depth 2c, then it is better to regard is as a plate, and, in the problem of
bending, replace the assumption of plane stress with that of plane strain;
the assumption can be justified everywhere except near the sides z = ±b/2.
The solutions to elastic problems in plane stress are formally identical to
those in plane strain if the Young’s modulus E and the Poisson’s ratio ν in
the plane-stress solution is replaced by E 0 = E/(1 − ν 2 ) and ν 0 = ν/(1 − ν),
respectively; in addition, a stress σz = ν(σx + σy ) must be present in order
to maintain the state of plane strain. In fact, the plane-strain solutions
are “exact” in the sense that all the compatibility conditions are satisfied,
while in all but the simplest plane-stress solutions some of the conditions are
violated, except in the limit of infinitely thin sheets (see, e.g., Timoshenko
and Goodier [1970], Article 98).
The elastic moment-curvature relation in the pure bending of a wide
plate is accordingly M = E 0 κ, while the distribution of bending stress is
again given by σx = −M y/I. If the stress σz = νσx is taken into account,
yielding occurs wherever |σx | = σY 0 , where

 σY , Tresca,
σY 0 = σY
 √ , Mises.
1 − ν + ν2
A similar relation relates the elastic-limit stress in plane strain, σE 0 , to σE

for a work-hardening material. With E replaced by E 0 and σY (or σE ) by σY 0
(or σE 0 ), the results obtained for narrow rectangular beams in 4.4.1 apply to
wide plates. For example, Equation (4.4.13) for the residual curvature upon
unloading holds if κE = σY 0 /E 0 c.
Initially Curved Plates: Elastic State

A section of a wide plate of constant curvature, subject to a constant
bending moment M , is shown in Figure 4.4.8(a). The state of stress at
each point may be assumed to be independent of θ and to consist of the
components σr , σθ , and σz , with σz = ν(σr + σθ ) to maintain plane strain.
The only equilibrium equation is therefore (4.3.18):
dσr σr − σθ
+ = 0.
dr r
Unlike the expansion of a cylindrical tube, however, the bending of a plate

requires a tangential displacement uθ = v(r, θ) in addition to the radial dis-
placement ur = u(r, θ). In view of the strain-displacement relations (1.2.1),
the strain-stress relations may be written as
∂u 1 u 1 ∂v 1 , 1 ∂u ∂v v
= 0 (σr − ν 0 σθ ), + = 0 (σθ − ν 0 σr ) + − = 0.
∂r E r r ∂θ E r ∂θ ∂r r
σθ a2 /M σr a2 /M
8.0 7.755 1.2
1.070
6.0 0.9
3JJ

Mk σr
M 4.0 σθ 0.6
J OCC
2.0 1.360 0.3

J b
C
1.443
0 1.0 r/a
J C
a

1.2 1.4 1.6 1.8 2.0

J
−2.0
b = 2a
J CC
−4.0

JC
−4.917

JC
−6.0
(a) (b)
Figure 4.4.8. Curved wide plate or bar in pure bending: (a) geometry; (b)
elastic stress distribution for b = 2a.
Eliminating u, v and σθ between these equations and (4.3.18) results in
d3 σr 5 d 2 σr 3 dσr
3
+ 2
+ 2 = 0.
dr r dr r dr
The general solution is
C
σr = A ln r + B + ,
r2
and (4.3.18) immediately gives
C
σθ = A(ln r + 1) + B − .
r2
The condition that the curved boundary, r = a and r = b, is traction-free
requires that σr vanish there, that is,
Ab2 ln b + Bb2 + C = 0, Aa2 ln a + Ba2 + C = 0.
The couple M is given by

Z b A B b
M =− σθ r dr = − [b2 (1 + 2 ln b) − a2 (1 + 2 ln a)] − (b2 − a2 ) − C ln .
a 4 2 a
Solving the three equations for A, B, and C in terms of M , we obtain the
stresses
!
M b r a2 b2 b
σr = 4 b2 ln + a2 ln − 2 ln
∆ r a r a ! (4.4.17)
M a2 b2 b 2 b 2 r 2 2
σθ = 4 ln + b ln + a ln − b + a ,
∆ r2 a r a
where 2
b

2 2 2 2 2
∆ = (b − a ) − 4a b ln .
a
The variation of the stresses with r is shown in Figure 4.4.8(b).
The maximum value of |σr − σθ | occurs at r = a, where it equals σθ (a).
Initial yielding according to the Tresca criterion takes place when M = ME ,
where
k∆ b

2 2 2
ME = 2b ln − b + a ,
2 a
k being the yield stress in shear. For b = 2a, ME = 0.258ka2 .
Initially Curved Plates: Elastic-Plastic State

The state of stress in an initially curved plate or wide beam in plane
strain under a bending moment M > ME was studied by Shaffer and House
[1955]. At first the plastic zone spans a ≤ r ≤ c1 , while c1 ≤ r ≤ b
is elastic. In the plastic zone the equilibrium equation (4.3.18), together
with the Tresca yield condition σθ − σr = 2k and the boundary condition
σr (a) = 0, gives the stresses
r r

σr = 2k ln , σθ = 2k ln + 1 .
a a
In the elastic region the stresses are still given by (4.4.17), but the constants
A, B, and C must be determined to satisfy continuity with the plastic stresses
at r = c1 , in addition to σr (b) = 0.
As the moment is increased further, at a certain value, say M1 , σθ (b)
attains the value −2k. For b = 2a, M1 = 0.387ka2 . When M > M1 , a
second plastic zone forms in c2 ≤ r ≤ b, in which stresses are given by
b b

σr = 2k ln , σθ = −2k 1 − ln .
r r
The stresses in the elastic core c1 ≤ r ≤ c2 are still given by (4.4.17). The
four continuity conditions for σr and σθ at r = c1 and r = c2 furnish the
three constants A, B, and C, and in addition, a relation between c1 and c2 ,
namely, !
c22 ab

2 2
c1 1 + ln = c2 1 − ln 2 .
ab c1
√
The fully plastic state is attained when c1 = c2 = ab. The ultimate
moment is
" √ #
Z ab r
Z b
b

k
MU = 2k − ln + 1 r dr + √ 1 − ln r dr = (b − a)2 .
a a ab r 2
For b = 2a, MU = 0.5κ2 .

As in the thick-walled tube in plane strain, both the elastic and plastic
parts of εz vanish as long as σz is the intermediate principal stress, so that
σz = ν(σr + σθ ) in the plastic regions as well. In the inner plastic zone, the
condition that σr < σz < σθ is (1 − 2ν) ln(c1 /a) < ν, while in the outer
plastic zone the corresponding condition σθ < σz < σr is (1 − 2ν) ln(b/c2 ) <
1 − ν. Clearly, these
√ conditions are satisfied at all stages if they are satisfied
when c1 = c2 = ab, and the second condition is satisfied whenever the
first one is, since ν ≤ 1 − ν. The results are consequently valid provided
ln(b/a) ≤ 2ν/(1 − 2ν). For ν = 0.3, the limiting value of b/a is e1.5 = 4.48.
The displacement field in the elastic-plastic pure bending of wide bars
was derived by Shaffer and House [1957] for an incompressible material and
by Eason [1960a] for compressible materials. The overall change in geometry
can be measured by the fractional change in the angle subtended by the bar,
∆α/α, and this is found to remain of the elastic order of magnitude until
M reaches about 0.95MU , similar to the case of initially straight bars. The
corresponding plane-stress problem was treated by Eason [1960b].

1. Find the shape factor MU /ME for the elastic–perfectly plastic I-beam
having the cross-section shown in Figure 6.2.2.
2. Find the moment-curvature relation of an elastic–perfectly plastic beam

of square cross-section that is bent about a diagonal.
4. Find the moment-curvature relation analogous to (4.4.10) for a rec-

tangular beam made of a linearly work-hardening material with initial
yield stress σE and work-hardening modulus H.
5. Find the shape factor for an elastic–perfectly plastic T-beam made of

two equal rectangles of length d and thickness t. Calculate for d/t = 8
and for the limiting case as d/t → ∞.
6. Find the equations of the interaction diagram for pure bending of an

elastic–perfectly plastic beam of ideal angle section composed of two
narrow rectangles of length a and b, respectively, having the same
thickness t. Sketch the diagram for b = 0.6a.
7. Derive an equation analogous to (4.4.13) for the residual curvature in

the beam of Exercise 4.
8. Discuss the behavior of an elastic–perfectly plastic rectangular beam

that is bent by means of a uniform bending moment to a curvature
κ0 > κE , and then unloaded and reloaded by a bending moment in the
opposite direction until plastic deformation again takes place. Plot the
moment-curvature diagram.
9. Discuss the development of a secondary plastic zone due to shear

stresses in a uniformly loaded cantilever beam and a uniformly loaded
simply supported beam of rectangular cross-section.
10. Discuss the plane-stress elastic-plastic solution, including the secondary

plastic zone, for a simply supported beam of rectangular cross-section
carrying a concentrated load at x = αL, with 0 < α < 21 .
Section 4.5 Numerical Methods

The problems discussed in the preceding sections of this chapter, for which
solutions were often presented in closed form, have been extremely simple:
highly regular shapes, simple boundary conditions, and idealized material
behavior. Real bodies whose mechanical behavior we wish to study are
rarely characterized by such simplicity. But models that approximate the
behavior of real bodies involve complex computations which have become
feasible only with the development of high-speed digital computers over the
past thirty years.
A computer can perform only a finite number of calculations in a finite
time. The description of the mechanical state of a body, if it is to be achieved
numerically, must therefore be defined by a finite number of variables. In
the simple problems discussed in Section 4.1, stress and strain were assumed
to be essentially constant throughout the body, and therefore each such
problem is governed by the rate equations alone, constituting a set of one
or more ordinary differential equations. For example, Equations (4.1.2) and
(4.1.3) each represent a set of two coupled nonlinear ordinary differential
equations.
In the more general problems studied in Sections 4.2 to 4.4, stress and
displacement are unknown functions of position, with the former required
to satisfy the equilibrium equations and the latter related to them through
constitutive relations. Knowledge of the values of these functions at all
points of the body amounts to an infinity of data. For the problem to
be numerically tractable, this infinity must be reduced to a finite amount.
That is, the stress and displacement fields must be determined, at least
approximately, by a finite number of parameters. The process whereby this
reduction is accomplished is the discretization of the problem, which was
discussed in general terms in 1.3.5. The discretized problem is again governed
by a coupled set of nonlinear ordinary differential equations. Their number,
Section 4.5 / Numerical Methods 251
however, can be quite large, and therefore special techniques are required in
order to economize on computer time.
Numerical techniques of time integration of the differential equations
arising in elastic-plastic problems are discussed in 4.5.1. In 4.5.2 we present
an overview of the most usual discretization scheme, namely, the finite-
element method (other methods are mentioned in passing). Finally, in
4.5.3 we discuss, in summary form, the combination of the two topics, that
is, the formulation of finite-element methods for elastic-plastic and elastic-
viscoplastic continua.
Matrix notation is used throughout this section, but with boldface
rather than underline notation. For a column-matrix-valued function of
a column-matrix-valued variable — say φ(ξ) — the rectangular matrix
[∂φi /∂ξj ] is denoted ∂φ/∂ξ, ∂ξ φ or φξ . In particular, if φ(ξ) is scalar-
valued, then ∂φ/∂ξ = φξ = ∂ξ φ is a row matrix.
4.5.1. Integration of Rate Equations
Viscoplasticity
For viscoplasticity, including viscoplasticity without a yield surface as
given by the “unified” models (i.e., rate-dependent behavior with internal
variables), the governing equations are
σ = C(ε − εi ) (4.5.1)
and
ξ̇ = φh, (4.5.2)
where φ and h are functions of (σ, ξ, T ) or of (ε, ξ, T ), and the internal-
variable matrix ξ includes εi .
When the temperature and the stress or strain are prescribed functions
of time, the viscoplastic rate equations (4.5.2) are just a set of coupled first-
order ordinary differential equations, which may be written in matrix nota-
tion as
ξ̇ = φ(ξ, t). (4.5.3)
A rather general method of integrating Equation (4.5.3) is the gener-

alized Euler method, according to which, if ξ(t) is known and ∆ξ =
ξ(t + ∆t) − ξ(t), then, approximately,
∆ξ = ∆t [(1 − β)φ(ξ, t) + βφ(ξ + ∆ξ, t + ∆t)], (4.5.4)
where ξ = ξ(t), and β is a parameter between 0 and 1 to be chosen in

accordance with the solution algorithm followed. The choice β = 0 represents
the forward Euler method (sometimes called simply Euler’s method),
one that is fully explicit in the sense that the unknown quantity ∆ξ appears
only on the left-hand side of the equation. While this method is simple, the
error accumulates rather rapidly unless very small time increments are used.
With 0 < β ≤ 1, the procedure is implicit. In particular, the choice β = 1
represents the backward Euler method. A common choice is β = 21 ,
representing the Crank–Nicholson method.
With β > 0, Equation (4.5.4) constitutes a set of coupled nonlinear
equations for the ∆ξα . An iterative scheme is required, as a rule, in or-
der to achieve a solution. In the direct iteration method or method
of successive approximations, an initial guess ∆ξ (0) (e.g., ∆ξ (0) = 0) is
made and substituted for ∆ξ in the right-hand side of (4.5.4). The result of
computing the right-hand side is then called ∆ξ (1) , is substituted again, and
so forth. The iteration stops when two successive approximations are suffi-
ciently close, that is, when an appropriately defined magnitude (or norm) of
∆ξ (k+1) − ∆ξ (k) , denoted ||∆ξ (k+1) − ∆ξ (k) ||, is less than some prescribed
error tolerance. The magnitudes||ξ|| may be defined in a variety of ways, for
X X
example, max |ξα |, |ξα |, or ξα2 .
α
α α
An iteration method that usually produces faster convergence than the
direct iteration method is the Newton–Raphson method. The right-hand
side of Equation (4.5.4) is subtracted from its left-hand side, and the result
is rewritten as
ψ(∆ξ) = 0,
it being understood that ψ also depends on ξ, t and ∆t. The initial guess
∆ξ (0) (which may again be zero, or which may be calculated from an explicit
scheme) is introduced; this is the predictor phase of the solution. Next, ψ
at ∆ξ (1) is evaluated by the approximation
.
ψ(∆ξ (1) ) = ψ(∆ξ (0) ) + J(∆ξ (1) − ∆ξ (0) ),
where
J = (∂ψ/∂∆ξ)|∆ξ(0) = I − β ∆t (∂φ/∂ξ)|ξ+∆ξ(0) .
By treating the approximation as an equality and setting ψ(∆ξ (1) ) = 0, we
obtain
∆ξ (1) = ∆ξ (0) + J−1 ψ(∆ξ (0) );
this is the corrector phase. The process may be continued with ∆ξ (1) re-
placing ∆ξ (0) in order to produce ∆ξ (2) , and so on, until ||ψ(∆ξ (k) )|| is
sufficiently small. With a reasonably good guess for ∆ξ (0) , it is usually not
necessary to recalculate J at each iteration
Rate-Independent Plasticity
The incremental stress-strain relations of rate-independent plasticity are
σ̇ = C(ε̇ − ε̇p ). (4.5.5)
The flow equations (3.2.1) are written in matrix notation as
ε̇p = λ̇g, (4.5.6)
where g is the column-matrix representation of the flow tensor with compo-

nents hij , defined in 3.1.1. If f (σ, ξ) = 0 is the equation of the yield surface,
then, in accordance with Equation (3.2.8),
 1

< fσ C ε̇ >, f = 0,
λ̇ = L (4.5.7)
 0, f < 0,
where L is given by Equation (3.2.9), or, in matrix notation,
L = H + fσ Cg,
with H denoting the work-hardening modulus as before.

Equations (4.5.5)–4.5.7) can be combined symbolically in the form
σ̇ = Cep ε̇, (4.5.8)
where it must be understood that in view of (4.5.7), the elastic-plastic mod-

ulus matrix Cep is a nonlinear operator, since it depends on the direction of
ε̇ in addition to depending on the current state as specified by (σ, ξ, T ) or
by (ε, ξ, T ) (recall that εp is included in ξ). The rate equations for ξ,
ξ̇ = λ̇h, (4.5.9)
when combined with (4.5.7) will symbolically be written in the form
ξ̇ = Λε̇, (4.5.10)
where Λ is a nonlinear operator like Cep .

Note that the governing equations, (4.5.5)–4.5.7) and (4.5.9), have here
been chosen in their strain-space form, while the statically determinate prob-
lems studied in Section 4.1, in particular those described by Equations (4.1.2)
and (4.1.3), were formulated under stress control. The reason for the choice
of strain control as the basis for the development of numerical methods is
twofold: first, because strain control, not being limited to work-hardening
materials, is more general than stress control; and second, because the strain-
space formulation is naturally associated with a displacement-based spatial
discretization scheme, such as the most usual versions of the finite-element
method.
The predictor/corrector scheme described for the viscoplastic rate equa-
tions can be applied to rate-independent plasticity as well. An effective
method consists of an elastic predictor followed by a plastic corrector. It is
supposed that σ and ξ (which includes εp ) have been determined with suf-
ficient accuracy, with f (σ, ξ) ≤ 0 satisfied, at a certain value of the strain
tensor ε. The elastic predictor/plastic corrector scheme is used to determine
the effect of a small strain increment ∆ε: it is initially guessed that the
effect is purely elastic, so that the stress will change to σ (1) = σ + C ∆ε,
and the internal-variable matrix will remain as ξ = ξ (1) . For simplicity, the
superscript (k) is applied to quantities evaluated at σ (k) , ξ (k) . If f (1) ≤ 0,
then the elastic prediction is correct, and the process is repeated for the
next strain increment. If, however, f (1) > 0, then the strain increment must
include some plastic strain, and a correction must be applied to σ (1) and
ξ (1) .
The stress correction is ∆σ (2) = −C ∆εp(2) , and ∆εp(2) is given, in prin-
ciple, by integrating 4.5.6) over the increment. In practice, it will be defined
by an approximation:
∆εp(2) = ∆λ g(1) .
Likewise,
∆ξ (2) = ∆λ h(1) ,
and ∆λ is defined so that the state defined by σ (2) = σ (1) + ∆σ (2) and
ξ (2) = ξ (1) +∆ξ (2) lies on the yield surface, at least in the first approximation.
To within an error that is of an order higher than the first in ∆λ,
.
f (2) = f (1) + (∂f /∂σ)(1) ∆σ (2) + (∂f /∂ξ)(1) ∆ξ (2) = f (1) − L(1) ∆λ,
where L(1) is the value at σ (1) , ξ (1) of the quantity L defined by (3.2.9).
The satisfaction of the yield criterion at σ (2) , ξ (2) , to this approximation,
requires that
f (1)
∆λ = (1) .
L
In fact, it can be shown that the state at (2) is on or outside the yield
surface if the yield function f is a convex function of its arguments,1 since
convexity implies [see Equation (1.5.11)] that
f (2) − f (1) − (∂f /∂σ)(1) ∆σ (2) − (∂f /∂ξ)(1) ∆ξ (2) ≥ 0,
and the definitions of ∆σ (2) , ∆ξ (2) , ∆λ and L reduce the left-hand side of
the inequality to its first term. Usually, f (2) is close enough to zero so that
this correction is sufficient. Otherwise, the plastic correction process can
be repeated until the state obtained is on the yield surface to a sufficient
degree of accuracy (for fairly simple yield surfaces and flow rules, the final
1
A yield function f is convex if it is, for example, of the combined-hardening type
(3.3.7), with F and −k convex functions of their arguments. The convexity of F is neces-
sary for the yield surface to be convex, while the convexity of −k means essentially that
the isotropic hardening rate is at most linear.
state may be obtained by scaling). The next strain increment can then be
applied.
Under stress control, the effect of a stress increment ∆σ on the yield
function is examined. If f (1) = f (σ + ∆σ, ξ) ≤ 0, then ∆ε = C−1 ∆σ,
and the next increment can be applied. If f (1) > 0, then ξ is corrected by
.
∆ξ (2) = ∆λ h(1) , where ∆λ = f (1) /H (1) in order that f (2) = 0. As under
strain control, the correction is repeated as needed.
The preceding scheme, called the return-mapping algorithm, is a gen-
eralization of the radial-return algorithm, so called because, according to
the Mises flow rule, Cg is proportional to the stress deviator s, and therefore
∆σ (2) ∝ −s(1) , that is, the stress correction is directed toward the origin of
the stress-deviator space. Other predictor/corrector algorithms have been
proposed, for example the initial-stress algorithm of Zienkiewicz, Valli-
appan, and King [1969].
More generally, the elastic predictor/plastic corrector method may be
thought of as a split of the elastic-plastic problem into an elastic problem
governed by
1
∆εij = (∆ui ,j +∆uj ,i ), ∆εpij = 0, ∆ξα = 0,
2
and a plastic problem governed by the differential equations
d d
σ(λ) = −Cg(σ(λ), ξ(λ)), ξ(λ) = h(σ(λ), ξ(λ)), (4.5.11)
dλ dλ
constrained by the yield criterion
f (σ(λ), ξ(λ)) = 0, (4.5.12)
and subject to the initial condition
{σ(λ), ξ(λ)}|λ=0 = {σ (1) , ξ},
where σ (1) is, as defined above, the stress tensor in the trial elastic state and
ξ is the internal-variable array at the end of the preceding time step.
The problem formed by Equations (4.5.11)–4.5.12) may be solved ei-
ther by direct iteration or by the Newton–Raphson method. Some general
methods, applicable to any yield criterion and hardening rule, have been
developed, including the general closest-point projection method and the
cutting-plane algorithm (see Simo and Hughes [1988]). For example, in
the closest-point projection method, the equations of the backward Euler
method for perfect plasticity with an associated flow rule, which may be
written as
f (σ + ∆σ) = 0,
def
ψ(∆σ, ∆ξ) = ∆εp − ∆λ ∂σ f (σ + ∆σ)T = 0,
are linearized by means of a Newton–Raphson scheme with ∆εp = −C−1 ∆σ

(since ∆ε = 0 in the corrector phase), yielding
ψ (0) − [C−1 + ∆λ(0) ∂σσ f (σ + ∆σ (0) )](∆σ (1) − ∆σ (0) )

− (∆λ(1) − ∆λ(0) ) ∂σ f (σ + ∆σ (0) )T = 0
and
f (σ + ∆σ (0) ) + ∂σ f (σ + ∆σ (0) )(∆σ (1) − ∆σ (0) ) = 0.
These equations may be solved for ∆λ(1) and ∆σ (1) , and hence for ∆εp(1) .
The accuracy and stability of integration algorithms for rate equations
are discussed by Ortiz and Popov [1985].
4.5.2. The Finite-Element Method

In statically indeterminate problems, the equilibrium equations must be
solved simultaneously with the constitutive relations. When the latter are
represented by nonlinear rate equations, the resulting problem is one of non-
linear partial differential equations, and must be solved numerically, that is,
the differential equations must be replaced by algebraic equations, and this
is accomplished by means of a spatially discretized model of the continuum,
as discussed in 1.3.5.
Generalities
The most commonly used spatial discretization method nowadays is the
finite-element method, originally developed as a method of analysis of
complex framed structures (such as aircraft structures) in which the “ele-
ments” are simple elastic members such as truss bars, straight and curved
beams, and the like. The method was soon extended to inelastic behavior
and to bodies modeled by one-, two-, and three-dimensional domains of arbi-
trary shape, the elements being small subdomains having a relatively simple
geometry. In the case of one-dimensional domain (which may be straight or
curved), finite elements are necessarily line segments. In two-dimensional
and three-dimensional domains, a particular division into subdomains is
known as a finite-element mesh, and the elements are typically polygons
and polyhedra, respectively. Some meshes are shown in Figure 4.5.1. In or-
der to maintain the simplicity of the element geometry, it may be necessary
to approximate the boundary of the domain — typically, a curved boundary
by one that is piecewise straight, as in Figure 4.5.1(b). Occasionally, ele-
ments with curved boundaries may be used advantageously, as for example
in a domain with circular geometry, in which the elements may be partly
bounded by arcs of circles [see Figure 4.5.1(c)].
Associated with each element are a number of points known as nodes,
usually located on the boundary (but occasionally in the interior) of the
Q
Q
Q H
Q @HH
Q
@ H

Q @ @
Q l
DD @
Q l @
@ D l
@ D ll
@ T
(a) (b) (c)
Figure 4.5.1. Finite-element meshes: (a) general mesh; (b) approximation of

curved boundary; (c) domain with circular geometry using curvi-
linear elements.
element. The nodes usually include, at the very least, the vertices of the
polygon. It is generally most convenient to use, as the generalized coordi-
nates of the model, the values at the nodes of the functions describing the
displacement field; these may simply be the nodal displacements, although
in finite-element models for beams, plates and shells the nodal rotations may
be included as well. It is often convenient to include, among the nodal dis-
placements, initially even those that are prescribed, and to eliminate them
after the global formulation. A finite element is characterized, then, by its
shape, by the nodes associated with it, and by the assumed variation of
displacement within the element, which is described by means of shape func-
tions or interpolation functions.1 These functions are ordinarily taken as
polynomial in terms of a local set of rectilinear coordinates.
The simplest finite element is one having the shape of a triangle in two
dimensions, or of a tetrahedron in three dimensions, and in which the nodes
are the vertices only [see Figure 4.5.2(a-b)]. The variation of displacement is
r
""T r r
" L
" T
" T L
" L
r"X
" T
X T L
XXX L
r Lr
XXX T
Tr
XXXT
X
(a) (b)
Figure 4.5.2. Constant-strain elements: (a) two-dimensional; (b) three-
dimensional.
taken as linear. It is easy to see that the number of parameters necessary to

1
If the shape functions describing the variation of displacement in terms of the nodal
displacements are the same as those describing the geometry of the element in terms of
the global coordinates, then the element is called isoparametric.
describe the variation of each displacement component in such an element

is just equal to the number of vertices. Since the partial derivatives of the
displacement field are constant within the element, the strain is likewise
constant, and the element is called a constant-strain element. Since the
displacement in a constant-strain element varies linearly along the edges (or
sides), it is continuous between adjacent elements. Furthermore, a rigid-
body displacement of all the nodes produces a rigid-body displacement field
in each element and therefore in the whole domain. The linear variation of
the displacement also means that the strain within each element is constant,
so that a combination of nodal displacements representing a global state of
constant strain does in fact produce such a state. An element that meets
these criteria is known as a conforming element. It has been shown that
the approximate solutions obtained by successive refinements of conforming
finite-element meshes converge monotonically to the exact solution of the
corresponding elasticity problem.
The constant-strain element has, as its principal advantage, its simplicity.
On the other hand, since the displacement field is represented by a piecewise-
linear approximation, it is clear that a rather fine mesh is required in order
to obtain reasonably accurate results. More refined elements permit the use
of coarser meshes and thus reduce computational time and storage. Some
two-dimensional examples are shown in Figure 4.5.3 (while they are shown
as rectilinear, they may also be curvilinear):
(a) The six-node triangular element (T6 ), with a general quadratic variation
of the displacement.
(b) The four-node quadrilateral element (Q4 ), with a bilinear variation of
the displacement; that is, when one local coordinate is held fixed, then the
variation with respect to the other coordinate is linear.
(c) The nine-node quadrilateral element (Q9 ), with a biquadratic variation
of the displacement.
r((r
r((((
r r``
E
""T ``` E
`r
" C ``` E
r " T C Er
r r
"
r
" T C E
r X
" T C
"
X E
XrXX r X
XXX T C E
T CX
XXX E
Tr Xr r r Er
XXXT
X
(a) (b) (c)
Figure 4.5.3. Higher-order two-dimensional elements: (a) six-node triangle
(T 6); (b) four-node quadrilateral (Q4); (c) nine-node quadri-
lateral (Q9).
Other elements in use include the seven-node triangular element (T7 ),

like T6 but with an additional node at the center, and the eight-node quad-
rangular element (Q8 ), like Q9 but without the central node.
Each element of Figure 4.5.3 has a three-dimensional counterpart: (a) the
ten-node tetrahedron, (b) the eight-node hexahedron, and (c) the twenty-
seven-node hexahedron. In addition, wedge-type elements are sometimes
used, as in Figure 4.5.4. Elements used for the analysis of plates and shells
generally have the same shapes as the two-dimensional elements of Figure
4.5.2 (flat for plates, curved for shells), but as mentioned above, the general-
ized coordinates may include nodal rotations in addition to nodal displace-
ments.
r
, XXXr
X
, ,l
, , l
,
r X r
, l
,
X XXr,, l
lr
Figure 4.5.4. Wedge element.
The finite-element method is not the only technique for the numerical
solution of boundary-value problems in solid mechanics. An older technique
is the finite-difference method; see, for example, Mendelson [1968] for
its application to plasticity problems. A relatively new technique is the
boundary-element method, based on certain fundamental solutions in
elasticity theory; for its application to elastic-plastic bodies, see Telles and
Brebbia [1979] and Maier [1983]. For torsion problems, a “cellular analogy,”
in which solid shafts are idealized as multi-celled structures, was developed
by Johnson [1988].
Displacement Formulation
The most common formulation of the finite-element method is as a
displacement-based discretization of the type discussed in 1.3.5. In order
to simplify the discussion, it will be assumed that all constrained general-
ized coordinates are zero. Let the number of unknown nodal displacements
associated with a given element be N e (in general in this section, the super-
script e denotes a reference to the given element). The column matrix of
these displacements will be denoted qe , and may be derived from the global
nodal-displacement matrix q by means of the relation
qe = Ae q, (4.5.13)
where Ae is an N e × N matrix. If the local description of the nodal dis-
placements is in the same frame as the global one, then Ae is what is known
as a Boolean matrix: Aemn = 1 if the mth elemental degree of freedom cor-
responds to the nth global degree of freedom, and Aemn = 0 otherwise.
For a given element occupying the subdomain Ωe , the interpolation func-

tions φn of Equation (1.3.17), when assembled in a 3 × N e matrix, will be
denoted Ne . Similarly, the functions (∇φn )S of Equation (1.3.18), when as-
sembled in a 6 × N e matrix, will be denoted Be ; this matrix is constant in a
constant-strain element. The finite-element analogues of Equations (1.3.17)–
(1.3.18) accordingly are

uh (x) = Ne (x)qe , εh (x) = Be (x)qe .

Ωe Ωe
Here the superscript h designates the discretization, and |Ωe denotes the
restriction of a function to Ωe .
The element internal-force matrix Qe is analogously obtained by the
virtual-work method discussed in 1.3.5 as
Z
Qe = BeT σ dΩ,
Ωe
where dΩ denotes an infinitesimal element of area or volume as appropriate.

In a constant-strain element, with constant material properties assumed,
the stress and internal variables are also constant, so that the preceding
integral reduces to a product. In a higher-order element the values of the
state variables must be specified at certain points within the element known
as Gauss points, and numerical integration must be carried out in terms of
these values.
By the principle of virtual work,
X
QT δq = QeT δqe ,
e
P
where e denotes summation over all the elements. In view of (4.5.13), the
global internal-force matrix is immediately obtained as
X
Q= AeT Qe .
e
Let F denote the column matrix representing the global discretized ex-
ternal loads, that is,
"Z Z #
X
eT eT eT a
F= A N f dΩ + N t dΓ ,
e Ωe ∂Ωet
where ∂Ωet is that portion (if there is any) of the boundary of Ωe that forms
a part of ∂Rt , and dΓ is an infinitesimal (surface or line) element of such a
boundary. The equilibrium equation is then
Q = F.
For an elastic body, this equation becomes
Kq = F,
where X
K= AeT Ke Ae ,
e
is the global stiffness matrix, with

Z
e
K = BeT CBe dΩ,
Ωe
being the element stiffness matrix. The corresponding equations for non-
linear continua with rate-type constitutive equations are discussed in 4.5.3.
Mixed Formulation
In inelastic (and some elastic) problems it is sometimes advantageous
to base finite-element methods on a mixed formulation, in which the ap-
proximations for displacement, strain and stress are by means of distinct
functions; one instance of the advantage is the treatment of incompress-
ibility. According to this point of view, the strain-displacement relations
and the constitutive relations are constraints, with the stress and strain, re-
spectively, as the corresponding Lagrange multipliers. Consequently, these
relations are to be satisfied in a weak way, rather than pointwise, and the
corresponding weak forms may be derived from the Hu–Washizu principle
(see Exercise 14 of Section 1.4 and Exercise 2 of Section 3.4). Since the vari-
ations δu, δε, and δσ are independent of one another, the principle furnishes
three separate variational equations, written (for consistency with the rest
of the present section) in matrix notation:
Z Z
T S T
(σ ∇ δu − f δu) dV − taT δu dS = 0, (4.5.14)
R ∂Rt
Z
[σ − C(ε − εi )]T δε dV = 0, (4.5.15)
R
Z
(∇S u− ε)T δσ dV = 0, (4.5.16)
R
where ∇S u is the appropriate column-matrix representation of (∇u)S , and

Equation 4.5.1) is assumed as the constitutive relation (εi may be replaced
by εp ).
Equations (4.5.14)–4.5.16) are the weak forms, respectively, of the equi-
librium equations, of the stress-strain relations and of the strain-displacement
relations. When the strain-displacement and stress-strain relations are en-
forced pointwise, the displacement formulation is recovered.
In the mixed finite-element formulation, stress and strain have distinct

discretizations:
σ h (x) e = Se (x)ce , (4.5.17)

Ω

h
ε (x) = Ee (x)ae , (4.5.18)

Ωe
where in general Se ! = Ee . Since the interpolations of stress and strain
are in general assumed discontinuous between the elements, the variational
equations (4.5.15)–4.5.16) hold for each element Ωe . Substituting (4.5.17)–
4.5.18) into (4.5.15) and 4.5.16), respectively, gives
Z
h e eT −1
σ (x) = S (x)H EeT C(ε − εi ) dΩ, (4.5.19)

Ωe Ωe
Z
def ¯ S
εh (x) = Ee (x)He−1 SeT ∇S u dΩ = ∇ u, (4.5.20)

Ωe Ωe
where Z
e
H = SeT Ee dΩ,
Ωe
and ∇¯ S is a discrete approximation to the operator ∇S . Equation (4.5.14),

with the aid of Equations 4.5.19)–4.5.20), now yields
def X Z
G(u, δu) = ¯ S u − εi )T C ∇
(∇ ¯ S δu dΩ − Geext = 0, (4.5.21)
e Ωe
where Z Z
Geext = f T δu dΩ + taT δu dΓ.
Ωe ∂Ωet
Equation (4.5.21) is a variational equation in the displacements, to be used

in place of the virtual-work equation in the discretization process discussed
in 1.3.5.
4.5.3. Finite-Element Methods for Nonlinear Continua
Nonlinear Problems: Incremental Solution

As long as inertia effects can be ignored and the deformation remains
infinitesimal, the discretization procedure described above can be applied
in nonlinear problems when the displacement, strain, and stress fields are
replaced by their respective time derivatives, that is, the velocity, strain-rate,
and stress-rate fields v, ε̇, and σ̇.
We consider a problem in which the discretized external-load matrix
F is a given function of time. One approach is to insert F in the time-
differentiated equilibrium equation
Q̇ = Ḟ.
This equation can be integrated over a time interval ∆t to yield
∆R = ∆Q − ∆F = 0.
Viscoplastic Problems
As has already been said, as the temperature goes up, the behavior
of metals becomes significantly rate-dependent and is therefore much better
described by viscoplasticity theory (whether with or without a yield surface)
than by rate-independent plasticity. In addition, it was pointed out in 3.1.2
that the solutions of problems in classical viscoplasticity under constant loads
attain asymptotically the equivalent rate-independent plasticity solutions,
and are often easier to achieve.
To simplify the writing, the factor φ in Equation (4.5.2) is taken as unity,
and the yield criterion, if any, is assumed to be embodied directly in h. The
rate equations are thus
ξ̇ = h(σ, ξ); (4.5.22)
these equations include the flow equation
ε̇i = g(σ, ξ). (4.5.23)
When Equation (4.5.1) is differentiated with respect to time and combined

with (4.5.23), the displacement-based discretization leads to
Q̇ = Kq̇ − G, (4.5.24)
where X Z
eT
G= A BeT Cg dΩ,
e Ωe
and K is the previously defined elastic global stiffness matrix. The problem,
then, is to integrate Equations
(4.5.22) and (4.5.24) for ξ and q, with σ given
by (4.5.1) together with ε e = Be qe , subject to the equilibrium equation
h
Ω
Q = F.
As with the previously considered integration of the rate equation alone,
the simplest technique is the Euler method, expressed by
∆ξ = ∆t h (4.5.25)
and
∆σ = C(∆ε − ∆t g), (4.5.26)
where h (which includes g) is evaluated at the beginning of the time step.
Applying the discretization process to Equation (4.5.26) gives
∆q = K−1 (∆F + ∆t G). (4.5.27)

The method is effective only with small time increments, not only because
of the accumulation of discretization error, but because roundoff error may
produce unstable results — that is, small changes in the initial conditions
may produce large differences in the solution after a certain time. The Euler
method is conditionally stable in the sense that stability holds only if the
time increment ∆t is less than some critical value ∆tcr .
The generalized Euler method discussed previously has been shown to
be unconditionally stable when β ≥ 12 . The initial values of h and g in
Equations (4.5.25–26) are replaced by linear approximations to their values
at t + β ∆t, where t denotes the beginning of the time step. These approxi-
mations are
h + β(hσ ∆σ + hξ ∆ξ) and g + β(gσ ∆σ + gξ ∆ξ),
respectively. Hence Equations (4.5.25–26) are replaced by
(I − β ∆t hξ ) ∆ξ = DT (h + βhσ ∆σ)
and
(I + β ∆t Cgσ ) ∆σ = C[∆ε − DT (g + βgξ ) ∆ξ]
(note that the dimension of the identity matrices I in these equations is not
in general the same). Eliminating ∆ξ between the two equations, we obtain
∆σ = C̄(∆ε − ∆t ḡ),
where
C̄ = [C−1 + β ∆t gσ + β 2 DT gξ (I − β ∆t hξ )−1 hσ ]−1
and
ḡ = g + βgξ (I − β ∆t hξ )−1 h.
By analogy with (4.5.27), we obtain the discretized equation
∆q = K̄−1 (∆F + ∆t Ḡ),
where Z
X
eT eT
K̄ = A B C̄B dΩ Ae
e
e Ωe
and Z
X
eT
Ḡ = A BeT Cḡ dΩ.
e Ωe
The improved accuracy and stability of the generalized Euler method is

achieved at the cost of computing and inverting a new stiffness matrix at
each time step. If desired, several iterations may be performed in each time
step in order to improve the results.
Elastic-Plastic Problems
The finite-element solution of elastic-plastic problems has been the sub-

ject of intense development since the 1960s, and a number of different ap-
proaches have been proposed. Differences persist over such issues as the use
of many constant-strain elements against fewer higher-order elements, the
use of local constitutive equations in rate form against that of variational
inequalities in the derivation of the discrete equations, and others. In this
section a brief outline of some of the most common approaches is presented.
Much of the theory can be found in textbooks such as Oden [1972] and
Zienkiewicz [1977] (see also Zienkiewicz and Taylor [1989]), while computer
implementation is treated by Owen and Hinton [1980]. More recent devel-
opments based on a mixed formulation are discussed by Simo and Hughes
[1988].
A finite-element model may be regarded as successful if it achieves con-
vergence of the solutions obtained under successive refinements of the mesh.
It is especially so if, in those cases where a closed-form solution (or another
reliable solution) is available, the convergence is to this solution. The strain-
space formulation of the rate equations of plasticity permits, in principle,
the application of any solution method to problems involving bodies that
may be work-hardening, perfectly plastic, or strain-softening. However, the
finite-element method has been successful only for work-hardening and per-
fectly plastic bodies. Attempts to solve problems involving strain-softening
behavior have almost invariably led to mesh-sensitivity, that is, lack of con-
vergence under mesh refinement. This result should not be too surprising.
Our study of uniqueness criteria (Section 3.4) showed that no such crite-
ria exist for softening bodies (if a body is nonstandard, then it must even
have some hardening). Furthermore, our discussion of the physical nature
of softening behavior in 2.3.2 showed that such behavior may be caused by
a localization of strain rather than by pointwise strain-softening. Several
schemes that account for localization (or a size effect) have been formulated
in order to permit successful finite-element solutions of problems involving
strain-softening bodies. Such schemes are beyond the scope of the present
introductory treatment, which is consequently confined to work-hardening
and perfectly plastic bodies.
We begin with an introduction to displacement-based methods. We sup-
pose that the current global displacement matrix q and the current local
state (σ, ξ) at each point are known.1 The constitutive relation can be
taken in the form (4.5.8), and, if the sign of λ̇ is assumed to be known at all
1
Points in the present context means elements if these are constant-strain, and Gauss
points in the case of higher-order elements.
points, then the tangent stiffness Cε is


 C, f < 0 or λ̇ ≤ 0,
Cε = 1
 C − Cgfσ C, f = 0 and λ̇ > 0.
L
A possible initial assumption is that λ̇ has the same sign at the current state
as it did at the preceding state. The global tangent stiffness Kt may be
defined in terms of Cε in the same way as K in terms of C, that is,
X Z
eT eT
Kt = A B Cep B dΩ Ae .
e
e Ωe
The problem of determining the state at successive discrete points in

time when the external-load history is prescribed2 may be broken down,
at each time point, into two phases. The first phase constitutes the rate
problem: the generalized velocities are obtained from
q̇ = K−1
t Ḟ, (4.5.28)

and, with q̇ known, ε̇ may be determined from the interpolation ε̇h =

Ωe
Be q̇e , and the assumption on λ̇ may be checked. If it is verified, then the
rate problem has been solved, since q̇ and ε̇ are now known, as are ξ̇ through
Equation (4.5.9) or (4.5.10), and σ̇ through (4.5.5) or (4.5.8). Otherwise, an
(0)
iteration scheme may be used: If Kt denotes Kt evaluated on the basis of
the initial assumption on λ̇, then the solution
(k−1) −1
q̇(k) = Kt Ḟ, k = 1, 2, ...,
permitting the evaluation of a new distribution of λ̇ and hence the calculation

(k)
of a new tangent stiffness matrix Kt . The process normally converges after
a small number of iterations.
Once the rate problem is solved, the second phase of the solution is the
determination of the state resulting from the imposition of an incremental
load ∆F. An explicit method begins with
∆q(0) = K−1
t ∆F,
from which ∆ε(0) is computed, and hence ∆ξ (1) and ∆σ (1) as in 4.5.1, with
a single implementation of the return-mapping algorithm. In general, the
incremental nodal forces ∆Q(1) resulting from ∆σ (1) do not equal ∆F, cre-
ating residual forces ∆R(1) that may be added to ∆F in the next step of the
iteration, which yields ∆q(1) from the general algorithm
∆q(k) = ∆q(k−1) + K−1 (k)

t ∆R , (4.5.29)
2
It is assumed, for simplicity, that any prescribed boundary velocities are zero.
where the stiffness matrix Kt may or may not be modified at each iteration
according to the new values of σ and ξ. In the tangent-stiffness method,
which is equivalent to the Newton–Raphson method, the stiffness matrix Kt
is recomputed at each iteration, providing faster convergence of the itera-
tion process at the expense of more computation at each iteration. These
computation costs are reduced in the so-called modified tangent-stiffness
methods, in which Kt is computed only once for each load increment, for ex-
ample in the first or in the second iteration. In the initial-stiffness method
the elastic stiffness K is used throughout the process, greatly decreasing the
need for matrix inversion but increasing the number of iterations. Conver-
gence occurs when the residual forces are sufficiently small, based on some
appropriate norm. The next load increment can then be applied, and the
iteration procedure can begin again.
Algorithmic Tangent Moduli
When the return-mapping algorithm is used in conjunction with the tan-
gent stiffness defined by Cε , the result is a loss of the quadratic rate of as-
ymptotic convergence, particularly important for large time steps (Nagtegaal
[1982], Simo and Taylor [1985]). A procedure that preserves the quadratic
rate of asymptotic convergence replaces Cep in the expression for Kt by
the so-called algorithmic (or consistent) tangent moduli, defined as follows:
given an algorithm that produces ∆σ when σ, ε, ξ and ∆ε are given, or
formally, if
∆σ = ∆σ(σ, ε, ξ, ∆ε),
then
Calg = ∂∆σ/∂∆ε|t+∆t .
While Calg coincides with the elastic-plastic tangent modulus tensor Cε in
problems with only one independent stress component, it does not do so in
general. For more details, see Simo and Taylor [1985] and Simo and Hughes
[1988].
The definition of Calg given above is consistent with the incremental form
of the equilibrium equations, ∆Q = ∆F. Other forms may be associated
with the total form, Q = F, and with the mixed formulation, given by
Equation (4.5.21). The derivation for the case of the mixed formulation is
shown next.
Application of Mixed Formulation
In the mixed formulation, the finite-element discretization may be writ-
ten as
uh (x) e = Ne (x)qe ,

Ω
S h
∇ u (x) = Be (x)qe , (4.5.30)

Ωe
εh (x)

= ¯ S uh (x)
∇ = B̄e (x)qe ,
Ωe Ωe
where1 Z
B̄e (x) = Ee (x)He−1 SeT Be dΩ.
Ωe
Substitution of (4.5.30) into Equation (4.5.21), the reduced weak form of the
equilibrium equation, gives
Z
¯ S u − εp ) dΩ
X
Q= AeT B̄eT C(∇
e Ωe
for the global internal-force matrix, where εp is evaluated by integration of

the rate equations. An application of the Newton–Raphson method again
yield Equation (4.5.29), with the tangent stiffness matrix now given by
Kt = ∂Q/∂q
Z
¯ S u − εp )]/∂(∇
¯ S u)} [∂(∇
¯ S u)/∂q] dΩ
X
= AeT B̄eT {∂[C(∇
e Ωe
Z
X
eT
= A B̄ Calg B̄ dΩ Ae ,
eT e
e Ωe
since
¯ S u)/∂q = [∂(∇
∂(∇ ¯ S u)/∂qe ] (∂qe /∂q) = B̄e Ae ,
and
¯ S u − εp )]/∂(∇
Calg = ∂[C(∇ ¯ S u)
t+∆t
is the tensor of algorithmic (consistent) tangent moduli obtained by differ-

entiation of the algorithmic (discrete) flow law.
Examples
As one example of the application of the finite-element method to the
solution of elastic-plastic problems, we examine the thick-walled elastic–
perfectly plastic cylindrical tube in plane strain, with the material assumed
to obey the Mises criterion and its associated flow rules; the corresponding
problem for the Tresca criterion was treated analytically in 4.3.5. If the
material is plastically and elastically incompressible (Poisson’s ratio equal to
one-half), then the Mises and Tresca results coincide.
Two cases are presented. In both, the ratio b/a of the outer to the inner
radius is 2, and k/G = 0.003. The finite elements used are concentric rings
(toroids) of rectangular cross-section, so that the mesh in any rz-plane is
a rectangular one; four-noded (Q4) elements are chosen for their simplicity.
It was shown by Nagtegaal, Parks and Rice [1974] that in a nearly incom-
pressible body in plane strain represented by Q4 elements, an unrealistically
stiff response (“locking”) is obtained unless special precautions are taken.
1
The use of the matrix function B̄e has given the procedure the name “B-bar”.
e3 e5 e7 e9 e
oo .1 11
r0 26
oo . 2 e
4 e
6 e
8 e
10 12 e ?
40 - 8 - 8 - 8 - 8 - 8 -
(a)
e4 e7 e10 e13 e16 e19 e22 e25 e28 e31

oo .1
r0 .2 5 8 11 14 17 20 23 26 29 32 6
?
oo .3 e
6 e
9 e
12 e
15 e
18 e
21 e
24 e
27 e
30 e
33 6
?
40 - - - - - - - - - - -
10 @ 4 each
(b)
oo -e1 e4 e7 e10 e13 e16 e19 e22 e25

0.25 6
r 0.50
-2 5 8 11 14 17 20 23 26 ?
6
0.25
oo -3 e e
6 e
9 e
12 e
15 e
18 e
21 e
24 e
27 ?
- - - - - - - -
8 @ 1 each
8 - 8 -
(c)
Figure 4.5.5. Meshes used in analysis of elastic-plastic cylinder: (a) coarse

mesh, displacement control, equal radial displacements applied at
nodes 1 and 2; (b) fine mesh, displacement control, equal radial
displacements applied at nodes 1, 2, and 3; (c) fine mesh, force
control, forces applied at nodes 1, 2, and 3 in the proportions
shown.
The previously discussed B-bar procedure, in which the volume strain is in-
dependently approximated by a discontinuous interpolation, is one way out,
and such a procedure is used in the present analysis. The program used is
FEAP (Finite Element Analysis Program), developed by R. L. Taylor (for
an introductory treatment, see Zienkiewicz [1977], Chapter 24).
In the first case the cylinder is virtually incompressible (ν = 0.4999)
and is analyzed under displacement control, with the radial displacement
of the inner surface, u(a), monotonically increased up to 4ka/3G and then
decreased until the pressure turns negative. Two meshes are used, a coarser
and a finer one, shown respectively in Figure 4.5.5(a) and (b). The mesh
represents a 1-radian sector of the cylinder, and therefore, at each value of
P P
u(a) the pressure is calculated according to p = ( R)/ah, where R is
the sum of the radial reactions at nodes 1 and 2 in mesh (a) and at nodes
1, 2, and 3 in mesh (b). The results of the computations are shown in Table
4.5.1 as p(a) and p(b) , respectively, and they are compared with the exact
values pex obtained according to the theory of Section 4.3. The convergence
is evident.
Table 4.5.1. Pressure-Displacement Relation for Incompressible Cylinder:

Computed and Exact Values
3G u(a)/ka p(a) /k p(b) /k pex /k

1 0.5010 0.5000 0.4996
2 0.9578 0.9562 0.9539
3 1.1973 1.1937 1.1928
4 1.3192 1.3156 1.3139
3 0.8184 0.8162 0.8143
2 0.3177 0.3160 0.3147
In the second case, the cylinder is analyzed under pressure control, with
the mesh shown in Figure 4.5.5(c), and both compressible (ν = 0.3) and
incompressible behavior are examined. For the former, the computed results
for the stress distributions σr , σθ , and σz , and for the displacements u(a)
and u(b), are shown for various values of c/a (where c is the radius of the
elastic-plastic boundary) in Figure 4.5.6, where they are compared with the
numerical solution of Hodge and White [1950];1 the graph of p/k against c/a
in Figure 4.5.6(d) was used to obtain c/a for the imposed values of p/k.
Finally, the distributions of σz for both compressible and incompressible
behavior when c/a = 1.5, and a plot of p/k against 2Gu(b)/ka for the
incompressible case, are shown in Figure 4.5.7.
1
The graphs are taken from Prager and Hodge [1951].
σr /2k σz /2k
0 rr 0.30
r rr
r r r
–0.1 r r rr h c/a = 0.25
r r
r r r hh r hhhh 1.8 r @ c/a =
hrrhhhhhh 1.6 r r hhr h 1.6
–0.2
h @ 1.8
r rh r
h
hhh
0.20 rr
hhhh r hhhhhh 1.4
hh
–0.3
r r r r hhhhhh 1.2
h h 0.15 rrr r r r ((r ( 1.4
h 1.0
r 1.2
r r r r r r r 1.0
r r
r r
–0.4 0.10
r
r r r r r r r r
r r
–0.5
r r
0.05 r
r
r
–0.6 r 0
r
–0.7 –0.05
r
–0.8 r/a –0.10 r
1.0 1.2 1.4 1.6 1.8 2.0
(a)
–0.15
σθ /2k –0.20 r/a

1.0 1.0 1.2 1.4 1.6 1.8 2.0
(b)
0.9
r c/a = 2Gu/ka p/k
r r 1.8
5 2.5
0.8 r r
r 1.6
rr r
0.7
r r r 1.4 4 r 2.0
r r 1.2 2Gu(a)/ka
r
rr r

0.6 r r 1.0
r r r
3 1.5
r r r
p/k
0.5 rr r
r
r
r r r
r

r 2 1.0
0.4 r r
r r
r r 2Gu(b)/ka
r r

0.3 r r
r 1 r 0.5
r
0.2 r/a
1.0 1.2 1.4 1.6 1.8 2.0
(c) 0 0.0 c/a
1.0 1.2 1.4 1.6 1.8 2.0
(d)
Figure 4.5.6. Compressible cylinder (ν = 0.3), results of finite-element calcu-

lations: (a)–(c) stress distributions at various values of c/a; (d)
displacements of inner and outer curfaces, and internal pressure,
against c/a.
σz /k
0.6
r r r r
0.4
r r r r p/k
r r 1.5
r
r r
0.2
r 1.0 r
r
0 r

r r
r 0.5

r

–0.2

r/a 0 2Gu(b)/ka
1.0 1.5 2.0 0 1.0 2.0 3.0
(a) (b)
Figure 4.5.7. Compressible and incompressible cylinder, results of finite-

element analysis (dots) and of Hodge and White [1950] or ex-
act analysis (curves): (a) axial stress distributions at c/a = 1.5
(dashed line, incompressible [exact]; solid line, compressible
[Hodge and White]; (b) pressure against displacement of outer
surface (incompressible only).
Another example is the end-loaded elastic–perfectly plastic cantilever

beam in plane stress, discussed in 4.4.2. Computations, based on the return-
mapping algorithm for plane stress formulated by Simo and Taylor [1986],
were performed with a 4 × 40 and an 8 × 80 mesh of Q4 elements, covering
half the beam (above or below the middle line). Numerical and analytical
results for the load–deflection relation are shown in Figure 4.5.8.
Lastly, an example is given of a problem for which no analytical solution
is known. Figure 4.5.9 shows the torque-twist relation for an elastic–perfectly
plastic I-beam having the cross-section shown in the figure. The computed
results, due to Baba and Kajita [1982], are shown along with the ultimate
torque calculated by means of the sand-heap analogy (see 4.2.3). As we saw
in 4.2.1, in a displacement formulation of the torsion problem the warping
function ψ(x1 , x2 ) is the only unknown; four-noded rectangular elements are
used, with a cubic interpolation for ψ, and with the values of ψ, ψ,1 , and
ψ,2 at the nodes as the nodal variables. A comparison of the results with
those obtained by the “cellular analogy” was made by Johnson [1988].
F/FE
+ ++
+ ××
1.5 ×
+
×
+
×
+
×
1.0 +
×
+ Coarser mesh
× Finer mesh
Analytical
40 -F
0.5 ×
+ ? σY
4 = 2 × 10−3
E
?6
×
+
0.0 ∆/∆E
0.0 0.5 1.0 1.5 2.0 2.5
Figure 4.5.8. End-loaded rectangular cantilever beam in plane stress: finite-

element and analytical results for the force-deflection diagram.

1. Find the elastic-plastic tangent-modulus tensor Cep for the solid obey-
ing the associated flow rule with the yield function
q √
f (σ, ρ, ε̄p ) = 2J¯2 − 2/3 (σE + βH ε̄p ),
where σE , β, and H are constants,

1
J¯2 = (sij − ρij )(sij − ρij ),
2
and ρ and ε̄p are governed respectively by
2 p
q
ρ̇ij = (1 − β)H ε̇pij , ε̄˙ = 2
3
ε̇pij ε̇pij .
3
2. (a) Formulate the radial-return algorithm for the elastic-plastic solid

of Exercise 1.
(b) By differentiating the algorithm, find the consistent tangent-mod-
ulus tensor, and compare with the result of Exercise 1.
3. Provide the details of the derivation of Equation (4.5.21) from Equa-

tion (4.5.14) by using the relations (4.5.19–20).
T /TE
4
Sand-heap analogy
b b
b
3
b b b
bb
b
b
2 b
b ?
4
T --
1 - 3 12
6
?
20 -
0 θ/θE
0 1 2 3 4 5
Figure 4.5.9. Torsion of an I-beam: finite-element results for the torque-twist
diagram (Baba and Kajita [1982]). The ultimate load obtained
by the sand-heap analogy is shown for comparison.
Chapter 5
Problems in Plastic Flow and

Collapse I
Theories and “Exact” Solutions
Introduction
In Chapter 4 the concepts of plastic flow and plastic collapse were re-
garded as essentially equivalent, representing a state in which a body con-
tinues to deform under constant applied forces. In practical applications,
however, the two concepts have quite different meanings. Plastic collapse
describes undesirably large deformations of an already formed body (a struc-
ture) that result from excessive forces; the calculation of collapse loads of
simple structures was studied in Section 4.1. The concept of plastic flow,
on the other hand, is usually applied to the deliberate forming of a mass of
solid (such as metal or clay) into a desired shape through the application of
appropriate forces.1 It is remarkable that these two large classes of prob-
lems, of fundamental importance in mechanical and civil engineering, can be
attacked by the same methodology — the theory of rigid–perfectly plastic
bodies, with the help of the theorems of limit analysis. A particularly exten-
sive body of theory, filling entire books, exists for problems of plane strain; a
summary of the theory, with some applications, is presented in Section 5.1.
In Section 5.2 we deal with the plastic collapse of circular plates.
Apart from plastic collapse, collapse of a elastic-plastic body may also be
due to structural instability. Such collapse (e.g., the buckling of a column)
may begin when the body is still fully elastic, and plastic deformation occurs
as a part of post-buckling behavior . Buckling that follows yield is covered by
the theory of plastic instability, which is treated in Section 5.3.
1
The plastic flow of soil constitutes an exception, since substantial movement of a soil
mass supporting a building or forming an earth dam is generally regarded as failure.
275
276 Chapter 5 / Problems in Plastic Flow and Collapse I
Section 5.1 Plane Problems

In Section 4.3 we found the stress field in a pressurized elastic–perfectly
plastic hollow cylinder in a state of plane strain by solving the equilibrium
equations together with, on the one hand, the compatibility condition in the
elastic region and, on the other hand, the yield criterion in the plastic region,
and by satisfying the boundary conditions on the outer and inner surfaces
and continuity conditions at the elastic-plastic boundary. The solution is
valid for all pressures up to the ultimate pressure pU .
When the ultimate pressure is reached, the tube becomes fully plastic.
The compatibility and continuity conditions then become irrelevant. The
equilibrium equation (4.3.18) and the yield criterion now constitute equa-
tions for the two unknown stress components σr and σθ , which may be solved
so that the traction boundary conditions are satisfied. For both the Mises
and Tresca yield criteria (which are equivalent in plane plastic flow), the
solution produces the stresses given by Equations (4.3.30)1,2 with c = b.
Since the fully plastic tube problem is formulated entirely in terms of
the stresses, it is often said to be statically determinate, though in a looser
sense than that of Section 4.1, since it is not equilibrium (“statics”) alone
that determines the stress field, but the yield criterion as well. The fully
plastic torsion problem of 4.2.3 can similarly be characterized as statically
determinate, since once again there is one nontrivial equilibrium equation
which, together with the yield criterion, can be used to determine the two
unknown stress components subject to the traction boundary conditions.
The same notion of static determinacy can be applied, in principle, to
a body of arbitrary shape that is assumed to be undergoing plane plastic
deformation, or to be in a state of plane stress: there are, in general, three
unknown stress components (σ11 , σ22 , σ12 in Cartesian coordinates) and
two plane equilibrium equations, plus the yield criterion. However, static
determinacy in this sense is effective in producing a unique stress field only
in special cases, namely those in which (a) only traction boundary conditions
are relevant, and (b) the entire body must become plastic for unrestricted
plastic flow to occur. As we saw in the beam problems studied in Section 4.5,
it is in general possible to have unrestricted flow in a plastic region that
occupies only a part of the body, the rest of the body remaining in the
elastic range and hence behaving as though it were rigid (recall, from 3.5.1,
the vanishing of the elastic strain rates at incipient plastic flow).
Even when the aforementioned conditions (a) and (b) are satisfied, there
may occur situations in which the traction boundary are not by themselves
sufficient to choose between two possible stress tensors at a point (see Figure
5.1.4, which is discussed later).
Since, in the general case, some of the boundary conditions may be
kinematic (velocity boundary conditions), it becomes necessary to find a
Section 5.1 / Plane Problems 277
kinematically admissible velocity field such that the strain rates derived from
it obey the flow rule. Consequently, the stress and velocity problems are
coupled. In the special cases where a unique stress field can be found directly,
then the velocity field can be found afterwards, as was done with the warping
of a fully plastic rectangular shaft (see 4.2.3). In the axisymmetric plane-
strain problem, the only velocity component is the radial velocity v, and the
flow rule is equivalent to the incompressibility constraint
dv v
+ = 0,
dr r
which may be solved to give v(r) = v(a)(a/r).
If, in addition, elastic regions remain in the course of plastic flow, then
a solution of the problem requires the determination of the elastic-plastic
boundary and of a stress field in the elastic regions which is continuous with
the plastic stress field at the boundary. While this was accomplished for the
beam problems of Section 4.5, it is in general an exceedingly difficult task. In
many problems, the main objective is to find the load that produces plastic
flow or collapse, and a complete elastic-plastic solution is not necessary to
achieve this objective: we know from the theorems of limit analysis (Sec-
tion 3.5) that the correct critical load is obtained from a plastically and
statically admissible stress field and a kinematically admissible velocity field
that is associated with it in the plastic region. Thus a rigid-plastic bound-
ary may be established on a purely kinematic basis, and the plastic stress
field needs to be extended into the rigid region only so that is statically and
plastically admissible. With this extension, the solution becomes a complete
rigid–plastic solution, usually known simply as a complete solution (Bishop
[1953]).
A systematic method of determining stress fields and associated velocity
fields in perfectly plastic bodies obeying the Mises (or Tresca) yield crite-
rion in plane strain was developed in the 1920s by Prandtl, Hencky, Mises
and others, and generalized by Mandel [1962] to include other yield criteria
and plane stress. This method, generally known as slip-line theory, is dis-
cussed in the next subsection. Some applications are presented in succeeding
subsections.
5.1.1. Slip-Line Theory
Shear Directions
A convenient way to establish the necessary relations for the stress field
in a plastic region is with the help of the definitions n = 12 (σ1 + σ2 ) and
r = 12 |σ1 − σ2 |, as given in 3.3.4, in conjunction with the yield condition
(3.3.6). We introduce the Mohr’s circle relations
σ11 = n + r sin 2θ, σ22 = n − r sin 2θ, σ12 = −r cos 2θ, (5.1.1)
τθ
x2
6
6
II
σ22 -
Ib nP
i A
Yb
AA
r A
6
σ12
S σ22
2θ i θ
?b b
S - σ12
6
- σθ P Y
S PP n
r σ
6
12
PP P
q
Pr
S
@ θ

P
Sb I
b X
? @ σ12
n - II @6
@ - σ11
σ11 X
- - x1
(a) (b)
Figure 5.1.1. First (I) and second (II) shear-line directions in the (a) Mohr’s-
circle and (b) physical planes.
where θ is the angle from the x1 axis to one of the principal shear directions,
namely the one along which the maximum shear stress r is directed to the left
when one is facing the outer normal (see Figure 5.1.1). This direction will
be called the second shear direction, and a line having this direction locally
everywhere will be called a second shear line. The other shear direction
(shear line) will be called the first.
When equations (5.1.1) are substituted into the equilibrium equations
σ11 ,1 +σ12 ,2 = 0, σ12 ,1 +σ22 ,2 = 0,
the resulting equations are, upon substitution of r = h(n),
[1 + h0 (n) sin 2θ]n,1 −h0 (n) cos 2θ n,2 +2r(θ,1 cos 2θ + θ,2 sin 2θ) = 0,
−h0 (n) cos 2θ n,1 +[1 − h0 (n) sin 2θ]n,2 +2r(θ,1 sin 2θ − θ,2 cos 2θ) = 0.
(5.1.2)
Equations (5.1.2) constitute a pair of nonlinear partial differential equations
for n, θ. A useful method of numerical solution for these equations is the
method of characteristics.
Method of Characteristics
In order to understand how this method can be used to solve the system,
it is simpler to consider first a single first-order partial differential equation
of the form
Av,1 + Bv,2 = C,
where A, B and C are functions of x1 , x2 and v. The equation can be

multiplied by an infinitesimal increment dx1 and rewritten as

B

A( v,1 dx1 + v,2 dx1 = C dx1 .
A
The quantity in parentheses becomes a perfect differential dv = (v,1 ) dx1 +

(v,2 ) dx2 if dx2 = m dx1 , where m = B/A. The direction defined by
dx2 /dx1 = m is called a characteristic direction, and a curve that is every-
where tangent to a characteristic direction is known as a characteristic curve
or simply a characteristic. Along such a curve, then,
dx1 dx2 dv
= = .
A B C
If v is known at one point of a characteristic curve, then dv can be calculated
for a neighboring point on this curve, and continuing this operation allows v
to be determined at all points on the curve. If v is known at all points of a
curve Σ that is nowhere tangent to a characteristic, then its values may be
calculated along all the characteristics that intersect Σ.
If two characteristics emanating from different points of Σ should inter-
sect at some point Q of the x1 x2 -plane, then in general two different values of
v will be obtained there. The point Q is therefore the locus of a discontinuity
in v.
Consider, now, a curve Σ that is nowhere tangent to a characteristic, and
suppose that a discontinuity (jump) in v,1 or v,2 occurs at a point P of Σ.
Since the only information necessary to determine the characteristic curve
through P and the values of v along this curve is the value of v at P , the
directional derivative of v along the characteristic will have the same value
on either side of it. Consequently, if any jump in v,2 or v,1 is propagated
through the x1 x2 -plane, it must occur across the characteristic through P .
In order to apply the method of characteristics to a system of several
first-order partial differential equations, the number of real characteristic
directions at each point must equal the number of unknown variables. The
system is then called hyperbolic. The system (5.1.2) is consequently hyper-
bolic if it has two real characteristic directions. It is called parabolic if there
is one such direction, and elliptic if there are none.
Characteristics of Equations (5.1.2)

In order to determine whether Equations (5.1.2) constitute a hyperbolic
system, we begin by solving them for n,1 , n,2 in terms of θ,1 , θ,2 :
−2r
n,1 = (θ,1 cos 2θ + θ,2 sin 2θ − h0 θ,2 ),
1 − h02
−2r
n,2 = (θ,1 sin 2θ − θ,2 cos 2θ + h0 θ,1 ).
1 − h02
Note that the solutions break down when |h0 (n)| = 1. Assuming that this
breakdown does not occur, we attempt to determine the characteristics by
assuming that, along a characteristic, dn = λ dθ. Now
dn = n,1 dx1 + n,2 dx2

−2r
= {θ,1 [cos 2θ dx1 + (sin 2θ + h0 ) dx2 ] + θ,2 [(sin 2θ − h0 ) dx1 − cos 2θ dx2 ]}
1 − h02
= λ dθ
= θ,1 (λ dx1 ) + θ,2 (λ dx2 ).
Equating the coefficients of θ,1 and θ,2 , respectively, in the second and fourth
lines leads to the system of equations
[(1 − h02 )λ + 2r cos 2θ] dx1 + 2r(sin 2θ + h0 ) dx2 = 0,
2r(sin 2θ − h0 ) dx1 + [(1 − h02 )λ − 2r cos 2θ] dx2 = 0,

which constitute a second-order eigenvalue problem. The characteristic
equation is
(1 − h02 )2 λ2 − 4r2 cos2 2θ − 4r2 sin2 2θ + 4r2 h02 ,
yielding the eigenvalues

2r
λ = ±√ .
1 − h02
The roots are real if and only if |h0 | ≤ 1; the problem is hyperbolic if |h0 | < 1,
parabolic when |h0 | = 1, and elliptic when |h0 | > 1.
Referring to the examples of the yield condition (3.3.6) discussed fol-
lowing its formulation, we note that the plane-stress problem for the Mises
criterion is hyperbolic only when |n| < 3k/2, parabolic when |n| = 3k/2,
and elliptic when |n| > 3k/2; this last condition occurs when
1 σ1
< < 2.
2 σ2
For the Tresca criterion, the problem is hyperbolic when |n| < k and par-
abolic when |n| ≥ k; the elliptic case does not arise.
In plane strain, on the other hand, h0 ≡ 0 for both criteria with the
associated flow rule when the elastic strains can be neglected, and there-
fore the problem is hyperbolic throughout the plastic domain. The same is
true for the Mohr–Coulomb criterion with the appropriate (not necessarily
associated) flow rule, since h0 = − sin φ, where φ is the angle of internal
friction. More generally, for any yield criterion given by Equation (3.3.6)
with |h0 | < 1, it is convenient to define a variable angle of internal friction,
φ(n), by sin φ(n) = −h0 (n); this is just the inclination of the Mohr envelope
with respect to the σ-axis at the point of tangency. The eigenvalues λ are
correspondingly given by ±2r sec φ.
From the trigonometric identity

sin 2a + sin 2b
tan(a + b) =
cos 2a + cos 2b
we may derive the directions of the eigenvectors. The eigenvector corre-
sponding to λ = 2r sec φ is given by dx2 /dx1 = − cot(θ − 12 φ), and will be
called the first or α characteristic, while the eigenvector corresponding to
λ = −2r sec φ is given by dx2 /dx1 = tan(θ + 21 φ), and will be called the
second or β characteristic. In the case of the Mises and Tresca criteria in
plane strain (the classical case), the characteristic directions coincide with
the shear directions and are orthogonal. In general the characteristics of the
two families intersect at an angle 21 π ± φ; they coalesce into a single family
when φ = 12 π, that is, in the parabolic case.
Defining the dimensionless variable ω by
cos φ(n)
Z
ω= dn,
2h(n)
we may write the characteristic relations, following Mandel [1962], as
dω = dθ along a first characteristic,

(5.1.3)
dω = −dθ along a second characteristic.
Note that ω = n/2k in the classical case, and
cot φ n

ω=− ln 1 − tan φ
2 c
for the Mohr–Coulomb material, yielding the preceding value in the limit as
φ → 0, with c = k.
If we introduce a curvilinear coordinate system α, β such that first and
second characteristics are given respectively by β = constant and α = con-
stant, then we can write the canonical equations
∂ ∂
(ω − θ) = 0, (ω + θ) = 0,
∂α ∂β
which have the general solution
ω = ξ(α) + η(β), θ = ξ(α) − η(β),
ξ and η being arbitrary functions.

In the following discussion of the geometric properties of the characteris-
tic network we shall limit ourselves to the case with φ constant — that is, the
Mohr–Coulomb criterion — so that a change in θ is also a change in the di-
rection of the characteristic. This case includes the classical case, φ = 0, and
the characteristic network is then called a Hencky–Prandtl network . Among

its properties are the following:
1. Suppose that one α characteristic is straight, that is, for a given β,
∂θ/∂α = 0. It follows that ξ 0 (α) = 0, and consequently all the α character-
istics are straight. Obviously, the same result holds for β characteristics.
2. Consider a pair of α characteristics, defined by β and β 0 , respectively,
and a pair of β characteristics defined by α and α0 , the points of intersection
being labeled A = (α, β), B = (α0 , β), C = (α, β 0 ), and D = (α0 , β 0 ), as in
Figure 5.1.2. It follows from the solution above that
χAC = θ(α, β) − θ(α, β 0 ) = η(β 0 ) − η(β),

χBD = θ(α0 , β) − θ(α0 , β 0 ) = η(β 0 ) − η(β),
and consequently the two angles χAC and χBD are equal. It can similarly
be shown that χAB equals χCD . This result is due to Hencky [1923] and is
known as Hencky’s theorem. In words: the angle formed by the tangents
of two given characteristics of one family at their points of intersection with
a characteristic of the other family does not depend on the choice of the
intersecting characteristic of the other family.
D r(α0 , β 0 )
rC hh
(α, β 0 ) h
hhhh χBD
((((
(
hhhh ((((((
0 , β) (h(
hhhh
B (α ((((
rA r((( h
χAC
hhhh
(α, β) h
Figure 5.1.2. Characteristics of Equations (5.1.2).
3. Now take (α0 , β 0 ) in Figure 5.1.2 infinitesimally close to (α, β). If

Rβ (α, β) denotes the radius of curvature of the β characteristics at (α, β)
and if dsα and dsβ denote infinitesimal arc lengths along the α and β char-
acteristics, respectively, then
dsβ(AC) = Rβ (α, β)χAC = [Rβ (α0 , β) + dsα ]χBD .
Since Rβ (α0 , β) = Rβ (α, β) + dRβ , it follows that dRβ = −dsα along an α

characteristic. Similarly, dRα = dsβ along a β characteristic. This result is
due to Prandtl [1923].
Traction Boundary Conditions

Traction boundary-value problems may be of three types, with the con-
struction of characteristics corresponding to each type shown in Figure 5.1.3.
Problem 1. The boundary is nowhere tangent to a characteristic.

Problem 2. The boundary is composed of characteristics of both families.
Problem 3. The boundary is of mixed type.
@X `` ```
@ `` P AA
P
XX@ P
X@ A
X `
` ``` PP A
P
@ `` PP
XX PPP A
XX X@ `
`` ` P P A
X@
X
X `
`` PP
@ ` A

(a) (b) (c)
Figure 5.1.3. Traction boundary-value problems: (a) Problem 1; (b) Problem

2; (c) Problem 3.
If the relation between ω and n is invertible, then the state of stress at a

point is determined by (ω, θ). Referring to Figure 5.1.4, we note that, along
an arc whose normal forms an angle χ with the x1 -axis, the normal stress,
shear stress and transverse (interior) normal stress are respectively given by
σ = n + r sin 2(θ − χ), τ = r cos 2(θ − χ), σ 0 = n − r sin 2(θ − χ), (5.1.4)
where n and r are determined by ω. If the arc forms a part of the boundary,
however, then at most σ and τ will be given there (traction
√ boundary con-
ditions); σ 0 can then have either of the values σ ± 2 r2 − τ 2 . Usually the
right value of σ 0 can be chosen by physical intuition.
HH
h σ
h Hh
h
h H τ

hhh σH 0 YH
H
hh Yh
H Hh H H
hh h
hhhh H
H
h
h
hhh h h Hh
hhh Hh
H HjH
hH
h
y hhh H
h Hh
θ
h H
Yχ

Figure 5.1.4. Stresses at a boundary.
In the classical case, the two choices for σ 0 give the respective explicit
expressions for θ and ω
s
1 τ σ 1 τ2
θ = χ ± cos−1 , ω= ± 1− .
2 k 2k 2 k2
Stress Discontinuities
An arc such as that of Figure 5.1.4 may also be located in the interior of
the plastic domain and be part of a line of stress discontinuity. To satisfy
equilibrium, σ and τ must be continuous across such a line, but ω (and there-
fore n) and θ are discontinuous, so that the directions of the characteristics
of each family change abruptly. In the general case r is also discontinuous,
but in the classical case r equals k and 0
√ is therefore continuous; σ then takes
2 2
each of the two possible values ±2 k − τ on either side of the disconti-
nuity line. Note that there is no discontinuity if and only if |τ | = k, that
is, if the arc is along a characteristic (which in the classical case is a shear
line). When |τ | < k, θ changes by cos−1 (τ /k). It can be seen from a Mohr’s-
circle construction that the line of stress discontinuity must bisect the angles
formed by the characteristics of each family on either side of it. It can also
be seen that if the discontinuity line is thought of as the limit of a narrow
zone of continuous but rapid change in σ 0 while σ and τ remain constant, all
the intermediate Mohr’s circles must be of radius less than k, showing that
this zone is elastic and that the discontinuity line is therefore the remnant of
an elastic zone (just like the ridge lines in the torsion problem). As a result
of this property, it was shown by Lee [1950] that a line of stress discontinuity
acts like an inextensible but perfectly flexible filament.
If the two regions separated by a line of stress discontinuity are denoted
1 and 2, and if the inclination of the line is χ, then σ and τ as given by the
first two Equations (5.1.4) are continuous across this line. Specializing to
the classical case, with n = 2ωk and r = k, leads to
2ω1 + sin 2(θ1 − χ) = 2ω2 + sin 2(θ2 − χ),
cos 2(θ1 − χ) = cos 2(θ2 − χ).
These equations may be solved for θ2 and ω2 in terms of θ1 and ω1 , yielding
the jump conditions due to Prager [1948]:
θ2 = 2χ − θ1 ± nπ, ω2 = ω1 ± sin 2(θ1 − χ), (5.1.5)
n being an integer, and the appropriate sign being taken as indicated by the
problem.
It was shown by Winzer and Carrier [1948] that if several straight lines
of stress discontinuity separating domains of constant stress meet at a point,
then these lines must number at least four. Winzer and Carrier [1949] also
discussed stress discontinuities between fields of variable stress.
The preceding arguments can be carried over from the classical to the
general case (see Salençon [1977]). In particular, the jump conditions may
be written directly in terms of the variables in Equations (5.1.4),
n1 + r1 sin 2(θ1 − χ) = n2 + r2 sin 2(θ2 − χ),
(5.1.6)
r1 cos 2(θ1 − χ) = r2 cos 2(θ2 − χ).
Velocity Fields
If a traction boundary-value problem is solved by constructing a charac-
teristic network, as described above, over a part of the region representing
the body, then the loading forms an upper bound to that under which plas-
tic flow becomes possible, since, as will be shown below, a kinematically
admissible velocity field can then be found. As mentioned before, the exact
flow load is found when the stress field can be extended in a statically and
plastically admissible manner into the rigid region. The following discussion
will be limited to the classical case; for a more general discussion, see, for
example, Salençon [1977].
The equations governing the velocity components v1 , v2 are found by
combining the associated flow rule,
ε̇11 ε̇22 ε̇12
= = ,
σ11 − σ22 σ22 − σ11 2σ12
with the strain-rate–velocity relations
1
ε̇11 = v1 ,1 , ε̇22 = v2 ,2 , ε̇12 = (v1 ,2 +v2 ,1 )
2
and with Equations (5.1.1) to obtain
v1 ,1 +v2 ,2 = 0, v1 ,2 +v2 ,1 −2 cot 2θ v2 ,2 = 0. (5.1.7)
The characteristics of Equations (5.1.7) are found by assuming dv1 = λ dv2 .

Thus
dv1 = v1 ,1 dx1 + v1 ,2 dx2 = v2 ,1 (−dx2 ) + v2 ,2 (2 cot 2θ dx2 − dx1 )

= λ dv2 = v2 ,1 (λ dx1 ) + v2 ,2 (λ dx2 ),
so that the characteristic directions are the eigenvectors of the system
λ dx1 + dx2 = 0,
dx1 + (λ − 2 cot 2θ) dx2 = 0.

The eigenvalues λ are the roots of
λ2 − 2λ cot 2θ − 1 = 0,
namely, (
cot θ
λ = cot 2θ ± csc 2θ =
− tan θ.
For λ = cot θ we have dx2 /dx1 = − cot θ (i.e., the first shear direction) while
for λ = − tan θ we have dx2 /dx1 = tan θ (i.e., the second shear direction).
We see therefore that the characteristics of the velocity equations are the
same as those of the stress equations. The characteristic relations are thus
dv1 = cot θ dv2 along an α characteristic,
dv1 = − tan θ dv2 along a β characteristic.
A plane whose coordinates are v1 and v2 is known as the hodograph

plane, and a diagram in this plane showing the velocity distribution is called
a hodograph. If P and Q are two neighboring points in the x1 x2 -plane
(the physical plane) lying on the same shear line, and if P 0 and Q0 are the
points in the hodograph plane representing the respective velocities, then as
shown by Geiringer [1951] and Green [1951], it follows from the characteristic
relations that the line element P 0¯Q0 is perpendicular to P¯Q. Consequently
the Hencky–Prandtl properties apply to the hodograph as well. A rigid
region, if it does not rotate, is represented by a single point in the hodograph
plane.
It is also instructive to express the characteristic relations in terms of the
velocity components along the characteristic directions, given respectively by
vα = v1 sin θ −v2 cos θ and vβ = v1 cos θ +v2 sin θ. The relations were derived
by Geiringer [1931] and are known as the Geiringer equations:
dvα = vβ dθ along an α characteristic,
(5.1.8)
dvβ = −vα dθ along a β characteristic.
It can be seen that these relations express the condition that the lon-
gitudinal strain rates ε̇11 and ε̇22 vanish when the x1 and x2 axes coincide
locally with the characteristic directions, in other words, that the shear lines
are inextensible. This result also follows directly from the flow rule, since
with respect to such axes we have σ11 − σ22 = 0.
It must be remembered, however, that the flow rule actually gives only
the ratios among the strain rates, and the preceding result must strictly be
written as ε̇11 /ε̇12 = ε̇22 /ε̇12 = 0. Another interpretation of this result is
that ε̇12 is infinite, meaning that either v1 ,2 or v2 ,1 is infinite. This happens
if the tangential velocity component is discontinuous across a characteristic
(the normal component must be continuous for material continuity), that
is, if slip occurs. The characteristics are thus the potential loci of slip and
are therefore also called slip lines. Kinematically admissible velocity fields
with discontinuities across slip lines are often used in the construction of
solutions. In particular, slip may occur along a characteristic forming the
boundary between the plastic and rigid regions.
If slip occurs along an α characteristic, with the tangential velocity hav-
ing the values vα and vα∗ on either side of it, then, since vβ has the same
value on both sides, an application of Equation (5.1.8)1 along the two sides
of the slip line gives dvα = dvα∗ , or d(vα∗ − vα ) = 0. An analogous result ap-
plies to a β characteristic. Thus the discontinuity in the tangential velocity
remains constant along a slip line. It follows further that the curves in the
hodograph plane that form the images of the two sides of a line of velocity
discontinuity are parallel.
5.1.2. Simple Slip-Line Fields
A great many practical problems can be solved by means of slip-line fields

containing straight slip lines. As we have seen, if one slip line (characteristic)
of a given family is straight, then all the slip lines of that family must be
straight. Families of straight slip lines, as can be seen in Figure 5.1.5, may
be of three types: (a) parallel, (b) meeting at a point, and (c) forming an
envelope.
b
b
EA lb
A EA b
l
A b
A AA
E A el b
A A A A Le l bb
AA A AA
E A b
A L e ll
A
A A A A E AA L e
A AA A A E L e
(a) (b) (c)
Figure 5.1.5. Families of straight slip lines: (a) constant-state field; (b) cen-
tered fan; (c) noncentered fan with envelope.
(a) If all the slip lines of one family are straight and parallel, then those
of the other family must be likewise. Since θ is constant, it follows from
Equations (5.1.3) that ω is constant as well, and therefore that the state of
stress is uniform. A region in which the slip lines are of this type is called
a region of constant state; this term is taken from wave-propagation theory
and is not strictly applicable here, because, while the stress components are
constant, the velocity components are not necessarily so; Equations (5.1.8),
with dθ = 0, yield the solutions vα = f (β), vβ = g(α). The functions f and
g are arbitrary except as constrained by boundary conditions.
(b) If the slip lines of one family are straight and meet at a point, then
those of the other family must be concentric circular arcs. Such a system of
slip lines is called a centered fan. A number of problems may be solved by
inserting a centered fan between two regions of constant state, in such a way
that the bounding radial lines of the fan are also the bounding parallel lines
of the constant-state regions. ω is constant along all the straight lines, while
along the circular arcs of the fan, dω = ±dθ, depending on whether the arcs
are α or β characteristics. If the θ difference between the two constant-state
regions is ∆θ, then this is just the angle subtended by the bounding lines of
the fan, and ∆ω = ±∆θ.
(c) An envelope of slip lines is also called a limiting line, and a family
of straight slip lines forming an envelope is called a noncentered fan; the

envelope is called the base curve of the fan. A limiting line cannot be in the
interior of the plastic region, and therefore must form a part of either the
boundary of the body or of the rigid-plastic boundary. Other properties of
limiting lines are discussed by Prager and Hodge [1951], Section 25.
Some Applications
We now consider some simple applications of slip-line fields consisting of
constant-state regions and centered fans, illustrated in Figure 5.1.6. These
results are of considerable importance in soil mechanics, where they are used
to study the stability of slopes and the carrying capacity of foundations made
of clays for which the hypothesis of constant shear strength (the undrained
strength discussed in 2.3.1) can be justified.
(a) Consider, first, a wedge of angle 2γ with a uniform pressure on one
side and no traction on the other. A possible shear-line net consists of two
regions of constant state, separated by a line of stress discontinuity bisecting
the wedge. With the principal stresses 2k − p, −p on one side and 0, −2k
on the other, continuity of the normal stress across the discontinuity line is
possible only if p = 4k sin2 γ. This value is therefore a lower bound to the
pressure causing incipient plastic flow. As can be seen in Figure 5.1.6(a),
when the wedge is acute (γ < 14 π) a velocity field may be constructed such
that regions ABC and AEF slip along the slip lines AB and AF , respec-
tively, while ACDE flows perpendicular to the stress-discontinuity line AD,
so that slip also occurs along AC and AE. The material below BAF may
be rigid.
(b) When the wedge is obtuse (γ > 41 π) no such velocity field is possible.
On the other hand, it is now possible to insert a centered fan of angle 2γ − 12 π
between two constant-state regions, producing the pressure p = 2k(1 + 2γ −
1
2
π), which exceeds the previously obtained lower bound for all γ > 14 π.
(c) Now consider a truncated wedge with pressure on the top face. At
each corner we can construct a plastic zone consisting of a centered fan
between two triangular regions of constant state, and plastic flow can occur
when the two plastic regions meet, so that the top face can have a downward
velocity. The angle subtended by each fan is just γ, so that p = 2k(1 + γ).
(d) The limit of the preceding case as γ → 21 π represents a half-plane
carrying a rigid block, and therefore the limiting pressure on the interface
between the half-plane and the block is k(2 + π). This result was obtained
by Prandtl [1920] by assuming a single triangular region of constant state
under the block, with a centered fan on either side and another constant-
state region outside each fan [see Figure 5.1.6(f)]. Prandtl’s solution was
criticized by Hill, who pointed out that, since the elastic solution of the
problem leads to infinite stresses at the corners of the block, plastic zones
must be there from the outset. As the pressure is increased, these zones will
D

A

B A
r B
P

PAr

BHH p
E B C P
P H

B B P
P
PA p
P P B B H
B BP A
P PP P B BB HH
P U P P B
B
BP P
P A
γ PP
P
BBBBB
B HH
B
B B
P A
P
B BP P

B B A PP A

Rigid
PPA

F Rigid B
jγ AA
(b)
(a)
p
???????????
A
@ @ @ @ B
A@
γ@ @@ HHB
AA *A B @ @ @ @ @ @ @ @

Rigid H
H @ @ @ @
HBB
Rigid
(c)
(d)
p
?????A B
A S
A S
@ @ @
@ L@ @ @@ L@ @ @
O
AH S @
S @ @
@ L@ @ @ L@ @ @
H @
HH S L @ L @
D L L
HHSC
66666666q 666 66 6666 Rigid
(e) (f)
Figure 5.1.6. Simple slip-line fields: (a) acute wedge; (b) obtuse wedge; (c)
truncated wedge loaded on the top edge; (d) rigid block on a half-
plane, Hill solution; (e) finite truncated wedge, discontinuous
stress field; (f) rigid block on a half-plane, Prandtl solution.
grow until they meet, as in case (c).

In the last two examples, no stress field outside the plastic regions has
been presented, so that the resulting pressures must be regarded as upper
bounds only. For example (c), Drucker and Chen [1968] have shown how to
construct a statically admissible stress field, leading to a lower bound equal
to the upper bound on the pressure. The following example shows that for
a finite truncated wedge, a lower pressure than that obtained in (c) above
can be found.
(e) A solution for a finite truncated wedge, uniformly loaded on its top
and bottom faces, is based on a fully plastic stress distribution, with the
four triangular constant-state regions separated by the stress-discontinuity
lines OA, . . . , OD, which bisect the angles at the corners. In order for these
bisectors to meet at one point, the ratio of the bottom to the top face of the
trapezoid must be (1+sin γ)/(1−sin γ). Because of symmetry, only the right
half of the wedge need be considered. Since the flank BC is traction-free,
region 2 is in a state of simple compression parallel to BC, and the values
of θ and ω there are θ2 = γ + π/4 and ω2 = − 12 , respectively. In region 1, by
symmetry, θ = θ1 = π/4. By Equation (5.1.5)2 , ω1 = − 21 − sin γ. Equation
(5.1.1)2 then gives, in region 1,
σy = −2k(1 + sin γ) = −p,
where p is the pressure on the top face; note that this value is less than that
obtained in (c) above. The pressure on the bottom face can similarly be
found as q = 2k(1 − sin γ). It can be seen that the aforementioned geometric
restriction is necessary for equilibrium.
Problems with Circular Symmetry

In classical problems with axial symmetry, the slip lines are along the
shear directions and therefore at 45◦ to the radial and tangential directions.
They are therefore given by logarithmic spirals, r ∝ e±θ . These slip lines
can be used to construct velocity fields in hollow prisms with a circular
bore under internal pressure. In Figure 5.1.7, for example, an axisymmetric
stress field — the same as in the hollow cylinder under internal pressure
— is assumed in the region inside the largest circle, with a vanishing stress
field outside this circle. Plastic flow is assumed to occur only in the curved
triangular regions bounded by the slip lines that meet the sides of the square
at their midpoints. The remaining regions move diagonally outward as rigid
bodies. A statically admissible stress field and an associated kinematically
admissible velocity field are thus found, and the pressure must be that for
the cylinder, namely, p = 2k ln b/a.
b - b -
P
i
PP
Slip lines
B
a
BN
Figure 5.1.7. Hollow prism with a circular cutout.
5.1.3. Metal-Forming Problems
A number of problems representing metal-forming processes can be solved

approximately by means of the theory of plane plastic strain, if the metal
is idealized as a material that is rate-independent and rigid–perfectly plas-
tic, and if the thermal stresses that result from the temperature gradients
induced by the forming process can be neglected. In a metal-forming prob-
lem — unlike a structural problem — unrestricted plastic flow is the desired
condition. The solution is intended to furnish the smallest applied force un-
der which the metal will flow, rather than the largest load under which the
structure will not collapse. For this reason an upper bound to the force is a
“safe” answer.
Metal-forming processes that closely approximate plane-strain conditions
(and for which slip-line theory can be used to generate upper bounds on
the forming forces) include forging, indentation, and cutting of wide strips,
as well as continuous processes such as extrusion, drawing and rolling. In
problems representing the latter category of processes, it is not the initiation
of the process that is studied, but a state of steady plastic flow in which a
large amount of plastic deformation has already taken place, and the stress
and velocity fields are taken as constant in time in an Eulerian sense, much
as in steady-flow problems of fluid mechanics.
An extensive bibliography of slip-line fields for metal-forming processes
can be found in Johnson, Sowerby and Venter [1982]. Many examples are
also to be found in Johnson and Mellor [1973], Chapters 11 and 14, and
Chakrabarty [1987], Chapters 7 and 8. Only a few selected problems will be
treated here.
Indentation
The problem of the half-plane carrying a rigid block, illustrated in Figure
5.1.6(d), can also be interpreted as describing the beginning of indentation of
a half-plane by a flat punch. The Hill solution, which requires slip between
the plastic zones and the punch, implies smooth contact. In the Prandtl
solution, on the other hand, the triangular region directly under the punch
moves rigidly downward with it, corresponding to rough contact. This ma-
terial in this region is often called dead metal .
The solution shown in Figure 5.1.6(c) for the truncated wedge of half-
angle γ with uniform pressure on its top face can be adapted to the problem
of indentation by a flat indenter at the bottom of a flat trench (Figure 5.1.8)
if γ is replaced by π − γ. The indentation pressure is thus p = 2k(1 + π − γ).
Tγ- ?
T
T C
X
X C
X X
T

X X
C@
@ @
@ C
@ @
C C
π−γ @

Figure 5.1.8. Flat indenter at the bottom of a flat trench.
A solution for the frictionless indentation of a half-plane by an acute

wedge-shaped indenter of half-angle α (α < π/4) was proposed by Hill, Lee
and Tupper [1947]. This problem is one of pseudo-steady plastic flow , in
which the geometry of the slip-line field (and therefore the stress and ve-
locity fields) changes as penetration proceeds, but in a geometrically similar
manner — that is, the angles remain the same, and only the scale changes.
As shown in Figure 5.1.9, the slip-line field covers Zone 1. Zone 3 is that
in which the material is elastic and therefore treated as rigid, while the in-
termediate Zone 2 contains material that has yielded but is restrained from
moving. As can be seen, the solution allows for the piling up of mater-
ial (the formation of a “coronet”) about the indenter, although in practice
such piling up is observed only in work-hardening materials. The angle θ
subtended by the centered fan between the two constant-state triangles can
be determined from the wedge half-angle α by means of the condition that
the volume of piled-up material (shown as shaded in the figure) equals the
volume of that portion of the indenter that has penetrated the work (shown
crosshatched). The slip-line field is equivalent to that in Figure 5.1.6(b),
with θ = 2γ − π/2, and therefore the contact pressure q is
q = 2k(1 + θ).
pppp pJ ppppppppppppppp α- A B ppp`

pppppppp
ppppppppppppp
ppppppppppppppppppp
J
p p p p p p p pppppppppQ
pppppppppppppppppppppppppp
ppppppppppppppppppppppppp

p p pQ
Q
k pp ppJ ppppp+
ppppppppppppppppppppppppppppppp
ppppppppppppppppppppppppppppppppppppp
D
J
```
J 3 `
Q Q
Q Q ` J
Q
`J
D J
J
D Zone
Q Q
JJ1
Q 3 `
DD J
H `` ` D
θ P ` hZone J
2
(
Zone 3
Figure 5.1.9. Indentation of a half-plane by an acute wedge-shaped indenter.
The indenter force, at a given state of indentation, is thus
F = 4k(1 + θ)AB.
Because of geometric similarity, the indenter force is consequently propor-

tional to indentation depth.
Forging and Cutting
The simultaneous application of identical flat punches to a strip of finite
thickness, Figure 5.1.10(a), may be used to model forging, with the bottom
punch representing the anvil and the top punch the forging tool. Similarly,
the cutting of a strip of metal with a wirecutter-like tool can be described as
the simultaneous indentation by a pair of identical wedge-shaped indenters
located opposite each other [Figure 5.1.10(b)]. Because of symmetry, in each
case only the top half of the strip needs to be considered, and the middle
plane may be regarded as a frictionless foundation. The solution of both
problems was studied by Hill [1953].
? ?
TT`
`TT
6 6
(a) (b)
Figure 5.1.10. Forging and cutting: (a) forging; (b) cutting.
For the cutting problem Hill showed that when the plastic region has
not yet reached the foundation, the slip-line field is the same as for the
semi-infinite domain. When, however, the plastic region extends through

the thickness of the strip, a different mode of deformation takes over: piling
up ceases, and the material on either side of the plastic region moves rigidly
outward; the slip-line field is shown in Figure 5.1.11(a).
? ?
...............
..............
.............
............
...........
..........
TT` .........
........
.......
......
.....
....
...
..
6 @@.
AA
- 2h -
AA
...
...
....
.....
......
.......
........
.........
..........
...........
`TT ? ............
.............
..............
...............
@@
6 6
2a-
(a) (b)
Figure 5.1.11. Slip-line fields: (a) forging; (b) cutting.
For the forging problem, the slip-line field is shown in Figure 5.1.11(b).
A triangular dead-metal region attaches itself to the punch, and indentation
proceeds as in the cutting problem, provided that h < 8.74a, where h is
the half-thickness of the strip and a the half-width of the punch. When
h = 8.74a, it can be shown that the punch pressure is p = 2k(1 + π/2), as
for the semi-infinite domain. It follows that for h > 8.74a the zone of plastic
deformation does not go through the strip and the pressure remains at this
value.
Drawing and Extrusion
Drawing and extrusion are processes in which a billet of material is forced
to flow through a die shaped to produce the required cross-section. In draw-
ing, as the name suggests, the material is pulled. Extrusion involves pushing.
In direct extrusion the die is stationary with respect to the container holding
the billet, and a ram moves in the container, pushing the billet outward with
the help of a pressure pad. In reverse extrusion the container is closed at
one end, and the die is pushed inside the container. The three processes are
shown in Figure 5.1.12.
The technologically important applications of these processes are pre-
dominantly three-dimensional — drawing produces wire, and extrusion is
used to make lightweight structural shapes, trim and the like. In the absence
of three-dimensional solutions, however, the solutions of the corresponding
plane problems provide qualitative information on the nature of the plastic
regions and hence allow estimates for the required forces.
Figure 5.1.13(a) illustrates a solution due to Hill [1948c] describing fric-
tionless extrusion through a square die with 50% reduction. The slip-line
field consists the two centered fans OAB and OA0 B 0 ; because of symmetry,
Die holder
@ Container

+
-@
R
............................. ..
.............................
..............................
....
...........................................
...................................................
..........................................................
................................................................................ J@..............................
JJJJJJJ .............................
..............................
................................................................................
.................................................................................
.................................................................................
................................................................................ .............................
..............................
........................................................
........................................................
.................................................................................
................................................................................
.................................................................................
................................................................................
.................................................................................
................................................................................
................................................................................
................................................................................
- ........................................................
........................................................
........................................................
........................................................
........................................................
........................................................
..............................
.............................
.............................. .... @
..AK@
..............................................................
........................................................
.................................................... .............................
..............................
.............................
@@@@@
@@
..............................
.............................
.............................. @@@
@@@@@
@
-JJJ
I
@
JJJ
A @ Ram
XX Die
y Extruding
A
metal Die A
Pressure pad
(a) (b)
Container

+
@ Closure plate
- ..
.. .........................
.........................
.........................
.........................
.........................
......................... @
.........................
......................... @@@
...............................................................
...............................................................
...............................................................
...............................................................
...............................................................
...............................................................
............................................................... @@+

.. .........................
.........................
.........................
@
- ...........................
.........................
.........................
......................... @
.........................@ @@

@@

@@
Die @@
(c)
Figure 5.1.12. Drawing and extrusion: (a) drawing; (b) direct extrusion; (c)
reverse extrusion.
only OAB need be considered. Since the exit slip line OA is a line of con-
stant stress, σ11 must vanish identically on it in order for the extruded metal
to its left to be in equilibrium, and σ12 = 0 because the line forms an angle
of 45◦ with the x-axis. The yield criterion requires |σ22 | = 2k, and, since
the sheet is being compressed, it follows that σ22 = −2k. Equations (5.1.1),
with n = 2kω and r = k, accordingly require that ω = − 12 and θ = π/4 on
OA.
The characteristic relations can now be used to determine the state along
AB. Since the fan subtends 90◦ , θ = 3π/4, and Equation (5.1.3)2 shows that
ω = 21 (1 + π) there. By Equations (5.1.1), then, we have σ11 = −(2 + π)k,
σ22 = −πk, and σ12 = 0 on AB. The average value of −σ11 along OAB is
thus equal to the extrusion pressure,
π

p= 1+ k.
2
A statically admissible extension of the stress field into the rigid region due
to Alexander [1961] shows this to be the exact pressure, not merely an upper
bound. The stress field on OB is extended into the dead-metal region ABC,
while that on OA is extended into OAP ; the extruded metal to the left of
AP is stress-free, so that AP is a line of stress discontinuity. The extension
to the right of the arc OB is achieved analytically.
A simple slip-line field for a reduction of 32 is shown in Figure 5.1.13(b),
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
.......................................................................................................................... ...........................................................................
...........................................................................
...........................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
.......................................................................................................................... ...........................................................................
...........................................................................
...........................................................................
...........................................................................
...........................................................................
...........................................................................................................................
..........................................................................................................................
...........................................................................................................................
..........................................................................................................................
.............................................................
.......................................................................................................................... ...................................
...........................................................................
...........................................................................
............................................................................
............................................................................
...........................................................................
......
...........................................................................
...........................................................................
.......................................... C .......................................... .............................................................. ...............................................................
.......................................... B .......................................... .......................................... ..........................................
....... .......................................... .......
............................................................... ...............................................................
..........................................
@
-? .......................................... (2+π)k ..........................................
..........................................
@
................................... ..........................................
.......................................... . .................... ................................... . ....................
............................................................... ...............................................................
.......................................... πk
@
6 .......................................... .......
..........................................
.......................................... .......
.......................................... ............................................................... .......................................... @ ...............................................................
(((( ........................................................ .......................................... ........................................................ ..........................................
. ....... .......
(
( ....................... .............................................................. ........................................................ ..............................................................
.
Ph
(
.......................................... .......................................... ..........................................
Hh A . .................... ................................. ..........................................
@Hhhhhh ............................................................... .....................
......... ..........................................
.......................................... ..........................................
@HHH ...............................................................
....... @ .......
...............................................................
.......................................... @ ..........................................
@ H .......................................... ..........................................
? @ H 0
.......
............................................................... @ .......
...............................................................
.......................................... ..........................................
6 @ P 2k O .......
..............................................................
.
@
@ ..........................................
.......
..........................................
..........................................
..........................................
..........................................
........................................................
(b)
........................................................

..........................................
..........................................
(((( ..........................................
.......
(
( A.0.. ...............................................................
(
. . . . ..........................................
.............................................................
h
Hhhhh ..........................................
@HH hh .......................................... ........................................................
.......................................... ..........................................
................................... ..........................................
@ HH ...................................
.......................................... ..........................................
@ H .......................................... ..........................................
H ................................... ..........................................
................................... ..........................................
@ .......................................... 0 ..........................................
.......................................... C 0
. . . . . .
B
...........................................................................................................................
.......
...................................
@ . . . . . .
...........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
..........................................................................................................................
(a)
Figure 5.1.13. Extrusion: (a) frictionless extrusion through a square die with
50% reduction; (b) slip-line field for two-thirds reduction.
and leads to an extrusion pressure of
4 π

p= 1+ k.
3 2
Figures 5.1.14(a) and (b) illustrate both drawing and extrusion through a
tapered die. If the container walls in the extrusion problem are smooth, then
the slip-line fields are identical if the die angle α and the reduction ratio r
are the same; the stress fields in the regions covered by the slip-line field
differ only by a hydrostatic stress.
A particularly simple slip-line field due to Hill and Tupper [1948], valid
for a smooth die when r = 2 sin α/(1 + 2 sin α), is shown in Figure 5.1.14(c).
In the extrusion problem, ω and θ on the exit slip line OA are the same as
in the preceding problem, and therefore their values in the constant-state
region ABC are ω = − 21 (1 + 2α), θ = π/4 + α. The normal pressure on AC
is
q = (1 − 2ω)k = 2(1 + α)k,
and the tangential stress there is zero. For equilibrium, the extrusion pres-
sure p must be
4(1 + α) sin 2α
p = rq = k.
1 + 2 sin 2α
.............................................. ..............................................
....................................................
.............................................
.............................................
............................................. ....................................................
.............................................
.............................................
.............................................
...............................................................
...." ..............................................
............................................. ...............................................................
...." ..............................................
.............................................
....................................
..............." ..............."
"
. .".
"
. .".
. . . . . . . . ....
................ ................ ....................................
. . .
" . . .
"
..............." ..............." ........................
..." .."
" .. ..
." ........................
........................
....
....................................
........................
........................
....
....................................
......... ......... ........................
b ................ b................ . ...........
bb ..............
............ bb ..............
............ ....................................
............
.......................................................
............
....................................................... ......................
.
.
bbb .............................................
..............................................
.............................................
..................................................
bb
b .............................................
..............................................
.............................................
.................................................. ....
b
b .............................................
.............................................
............................................. b
b .............................................
.............................................
.............................................
(a) (b)
............................................................................
...........................................................................
.......................................................................................
............................................................................
...........................................................................
...........................................................................
...........................................................................
" ...................
"" .
. ....................... C
6 6 "" α ................... ◦ . 6
r .................. 45
...................
w""
.............
.......... 45◦
""
? "X 1
A@XXX B
α

@
45◦@ ?
O

CC
.........
b................
..............
bbb ..........
b ...
b
(c)
Figure 5.1.14. Drawing and extrusion through a tapered die: (a) drawing; (b)
extrusion; (c) slip-line field for drawing and extrusion through
a smooth tapered die with r = 2 sin α/(1 + 2 sin α) (Hill and
Tupper [1948].
The corresponding drawing problem is solved by superposing a hydrostatic

tension equal in magnitude to this pressure. The drawing stress therefore
has the same value as the extrusion pressure.
For further solutions of metal-forming problems using slip-line theory,
see the aforementioned references by Johnson and Mellor [1973], Johnson,
Sowerby and Venter [1982], and Chakrabarty [1987].1 Additional plane
problems, in which complete solutions are not available, are discussed in
Section 6.1 in the context of limit analysis.
1. For a Hencky–Prandtl network in which the values of ω − θ on any

two neighboring first characteristics and the values of ω + θ on any two
neighboring second characteristics differ by the same small constant,
show that the diagonals of the network are lines of constant ω or of
constant θ (use Figure 5.1.2).
1
Finite-element methods for metal-forming problems are treated by Kobayashi, Oh,
and Altan [1989].
2. Show that when four straight lines of stress discontinuity meet on an

axis of symmetry of stress field and separate four regions of constant
stress, as in Figure 5.1.6(e), the angles AOD and BOC must be right
angles.
3. Show that any velocity field that is associated with the stress field of
the preceding exercise represents rigid-body motion.
4. Derive the Geiringer equations (5.1.8).
5. Find the differential equations for the velocity field in plane plastic flow
in a standard Mohr–Coulomb material. Determine the characteristics
of the velocity field.
6. Show that in classical problems of plane plastic flow with axial sym-
metry, the slip lines are given in polar coordinates by r ∝ e±θ .
7. Show that the relation between θ and α in the slip-line field of Figure
5.1.9 is
1 π θ

−1
α= θ + cos tan − .
2 4 2
8. The slip-line field of Figure 5.1.6(c) may be regarded as representing
a stage in the squashing of an originally pointed wedge of half-angle α
(greater than γ) by a lubricated flat plate.
(a) Using geometry and volume constancy, find the relation between
α and γ.
(b) Find the relation between the applied force and the distance
moved by the plate.
(c) Determine the smallest value of α for which the solution is valid.
9. Discuss how the slip-line field of Figure 5.1.6(c) can be used to study
the necking of a symmetrically notched tension specimen.
10. Discuss the velocity fields associated with the slip-line fields of Figures
5.1.13(a) and (b).
Section 5.2 Collapse of Circular Plates

The goal of this section is derivation of collapse loads for axisymmetrically
loaded circular plates made of a perfectly plastic material obeying the Tresca
criterion. For such plates, complete solutions exist in closed form, and they
are treated in 5.2.3. For the collapse loads of plates without circular sym-
metry, limit analysis must be used to obtain estimates, and this is done in
Section 5.2 / Collapse of Circular Plates 299
Chapter 6. The introduction to plate theory given in 5.2.1 is general, and

not limited to circular plates. Similarly, the presentation of elastic relations
and yield criteria at the beginning of 5.2.2 and 5.2.3, respectively, is general,
but solutions will be given for axisymmetric problems only.
5.2.1. Introduction to Plate Theory
Derivation of Plate Equilibrium Equations

A plate may be defined as a solid body occupying in the undeformed con-
figuration the region A × [−h/2, h/2], that is, the set of points {(x1 , x2 , x3 )
| (x1 , x2 ) ∈ A, −h/2 ≤ x3 ≤ h/2}, where A is a closed domain in the x1 x2 -
plane bounded by a simple closed curve C (we are assuming no holes in the
plate), with h considerably smaller than the typical dimension of A. The
plane x3 = 0 is called the middle plane of the plate. The outward normal
unit vector to C in the x1 x2 -plane has components να (α = 1, 2), and the
counterclockwise tangential unit vector has components tα .
We shall approach the study of the mechanics of plates by a combined
use of the three-dimensional equilibrium equations and of virtual work. The
plate equilibrium equations will be derived directly from the former. First,
we define the stress resultants as follows:
Z h/2
Nαβ = σαβ dx3 (membrane forces),
−h/2
Z h/2
Qα = σα3 dx3 (shear forces),
−h/2
Z h/2
Mαβ = − x3 σαβ dx3 (moments).
−h/2
The surface loads are

Z h/2
h/2
pα = σα3 |−h/2 + fα dx3 ,
−h/2
Z h/2
h/2
q = σ33 |−h/2 + f3 dx3 ,
−h/2
Z h/2
h/2
mα = −x3 σα3 |−h/2 − x3 fα dx3 ,
−h/2
where f is the body force per unit volume.

Distinguishing the x3 -coordinate from the xα (α = 1, 2), we write the
local equilibrium equations
σαβ ,β +σα3 ,3 +fα = 0, (5.2.1)
σα3 ,α +σ33 ,3 +f3 = 0. (5.2.2)

Integrating these equations through the thickness and performing integration

by parts where necessary yields
Nαβ ,β +pα = 0 (5.2.3)
and
Qα ,α +q = 0. (5.2.4)
When Equations (5.2.1) are multiplied by x3 and then integrated through
the thickness, the result is
Mαβ ,β +Qα + mα = 0. (5.2.5)
We can eliminate Qα between Equations (5.2.4) and (5.2.5), and obtain
Mαβ ,αβ = q − mα ,α . (5.2.6)
Equations (5.2.3) and (5.2.4–5) or (5.2.6) are the plate equilibrium equa-
tions, the former for in-plane or membrane forces and the other for bending
forces. Note that the two modes of behavior — in-plane deformation and
bending — are statically uncoupled. In the elementary theory they are also
kinematically uncoupled, and therefore can be studied separately.
Displacement Assumptions and Virtual Work

The elementary displacement model for plate behavior is described by
the following displacement field:
uα (x1 , x2 , x3 ) = ūα (x1 , x2 ) − x3 w,α (x1 , x2 ), u3 (x1 , x2 , x3 ) = w(x1 , x2 );
here the ūα are the in-plane displacements and w is the deflection. It follows
that εi3 = 0, i = 1, 2, 3, so that σij δεij = σαβ δεαβ , and εαβ = ε̄αβ −x3 w,αβ ,
where ε̄αβ = 21 (ūα ,β +ūβ ,α ).
It is important to check whether the assumption of infinitesimal strain is
valid. The Green–Saint-Venant strain tensor (Section 1.2) has the in-plane
components
1 1
Eαβ = εαβ + uγ ,α uγ ,β + w,α w,β .
2 2
If we neglect the contributions of the in-plane displacements ūα , then the
right-hand side reduces to
1 1
−x3 w,αβ + x23 w,αγ w,βγ + w,α w,β .
2 2
If δ is a typical deflection and l a typical dimension of A, then the first term
is of order hδ/l2 , the second of order (hδ/l2 )2 , and the third of order (δ/l)2 .
While the second term is negligible in comparison to the third whenever h/l
is sufficiently small, as is normal in plate theory, for the third term to be
negligible in comparison to the first term it is necessary for the deflection to

be small compared to the plate thickness. Otherwise, the Green–Saint-Venant
strain tensor must be used, given in general by
Eαβ = Ēαβ − x3 w,αβ ,
with
1
Ēαβ = ε̄αβ + w,α w,β . (5.2.7)
2
Equation (5.2.7) will be used in the next section when the buckling of plates
is studied.
Under the hypothesis of infinitesimal strain, the internal virtual work
becomes Z
δW int = (Nαβ δ ε̄αβ + Mαβ δw,αβ ) dA.
A
This equation may be rewritten as
(1) (2)
δW int − δW ext = δW ext ,
(1)
where δW ext denotes the part of the external virtual work due to the body
(2)
force and the surface tractions on the planes x3 = +−h/2, and δW ext is that
due to applied forces and moments along the edge. The first part accordingly
is given by
Z "Z h/2
#
(1) h/2
δW ext = fi δui dx3 + (σi3 δui ) |−h/2 dA.
A −h/2
Now
fi δui = fα δ ūα − x3 fα δw,α +f3 δw
and
h/2 h/2 h/2 h/2
(σi3 δui ) |−h/2 = σα3 |−h/2 δ ūα − (x3 σα3 ) |−h/2 δw,α +σ33 |−h/2 δw,
so that Z
(1)
δW ext = (pα δ ūα + q δw + mα δw,α ) dA.
A
Since Nαβ is symmetric, it follows that Nαβ δ ε̄αβ = Nαβ δ ūα ,β , and
Z
(1)
δW int −δW ext = [(Nαβ δ ūα ,β −pα δ ūα )+(Mαβ δw,αβ −mα δw,α −q δw)] dA.
A
Now
Nαβ δ ūα ,β = (Nαβ δ ūα ),β −Nαβ ,β δ ūα ,
and
Mαβ δw,αβ − mα δw,α

= (Mαβ δw,α ),β −[(Mαβ ,β +mα ) δw],α +(Mαβ ,αβ +mα ,α ) δw,
so that upon applying the two-dimensional divergence theorem we obtain

I
(1)
δW int − δW ext = [νβ Nαβ δ ūα + νβ Mαβ δw,α +να (Mαβ ,β +mα ) δw]ds
CZ
− [(Nαβ ,β +pα ) δ ūα + (Mαβ ,αβ +mα ,α −q) δw] dA.
A
(5.2.8)
However, by the equilibrium equations (5.2.3) and (5.2.6) the area integral
vanishes. In the remaining contour integral, the expression in parentheses
may be replaced by −Qα as a result of Equation (5.2.5). Moreover, the three
functions w, w,1 and w,2 are not independent on C, because, if w,α is de-
composed as w,α = να ∂w/∂n + tα ∂w/∂s, then the normal derivative ∂w/∂n
(which represents the local rotation of the plate) can be prescribed indepen-
dently of w, but the tangential derivative ∂w/∂s is entirely determined by w.
Performing the aforementioned decomposition and defining Mn = να νβ Mαβ
(the normal bending moment), Mnt = να tβ Mαβ (the twisting moment), and
Qn = να Qα , we obtain
∂δw ∂δw
I
(1)
δW int − δW ext = νβ Nαβ δ ūα + Mn + Mnt + Qn δw ds.
C ∂n ∂s
By integration by parts,
∂ δw ∂Mnt
I I
Mnt ds = − δw ds,
C ∂s C ∂s
and therefore
∂w
I
(1)
δW int − δW ext = νβ Nαβ δ ūα + Mn δ + Vn δw ds,
C ∂n
where Vn = Qn − ∂Mnt /∂s is the effective shear force along the edge. A
graphic illustration of the equivalence between a varying twisting moment
and a distributed transverse force may be seen in Figure 5.2.1.
? ? ? ?
z }| { z }| { z }| { z }| {
6
6
6
6
6?
?
?
?
?
Figure 5.2.1. Effective shear force along a plate edge.
Finally, let the applied in-plane forces along the edge be Fαa , the ap-
plied bending moment (acting about the tangent to C) Mna , and the applied
transverse force Vna , all per unit length. Then
∂w
I
(2)
δW ext = Fαa δ ūα + Mna δ + Vna δw ds,
C ∂n
so that the boundary conditions are
either νβ Nαβ = Fαa or ūα prescribed,

∂w
either Mn = Mna or prescribed,
∂n
either Vn = Vna or w prescribed.
Leaving aside the in-plane forces and displacements, we see that at every
point of the edge two conditions must be specified. For example, along a
clamped edge the conditions are w = 0 and ∂w/∂n = 0; along a simply
supported edge, w = 0 and Mn = 0; and along a free edge, Vn = 0 and Mn =
0. The condition Vn = 0 was first derived by Kirchhoff, and consequently the
theory of plates that has thus far been outlined is known as Kirchhoff plate
theory. In the original plate theory formulated by Sophie Germain, a free
edge was assumed to be subject to the three boundary conditions Qn = 0,
Mnt = 0, and Mn = 0, resulting in an improperly posed boundary-value
problem for elastic plates.
In the present treatment the principle of virtual work was used to derive
the boundary conditions that are consistent with the displacement model
adopted, while the equilibrium equations (5.2.3)–(5.2.6) were derived from
the three-dimensional ones — that is, they were shown to be necessary, but
not sufficient. However, it can easily be seen that the principle of virtual
work also implies the equilibrium equations (5.2.3) and (5.2.8), and as the
only necessary ones: since the displacement components ūα , w can vary
independently in A, and since the area integral in Equation (5.2.8) must
vanish, it follows that the coefficients of δ ūα and δw must vanish. Equation
(5.2.5) may then be used as the definition of Qα .
Before introducing constitutive equations, it must be noted that although
the displacement model is one in which ε33 = 0, this constraint is not re-
alistic. Actually, it is the stress σ33 which is very nearly zero, or at least,
its maximum value is very small in comparison to those of the stresses σαβ
(α, β = 1, 2). Similarly, the shear stresses σ3α , though important in the
equilibrium equations, are generally of small magnitude. Consequently, most
points of the plate are nearly in a state of plane stress. The elastic behavior
of isotropic plates should therefore be described by Equations (1.4.13), and
plasticity by a plane-stress yield criterion.
5.2.2. Elastic Plates
Elastic Relations
Equations (1.4.13),
E
σαβ = [(1 − ν)εαβ + νεγγ δαβ ].
1 − ν2
lead to
Eh
Nαβ = [(1 − ν)ε̄αβ + ν ε̄γγ δαβ ]
1 − ν2
and
Mαβ = D[(1 − ν)καβ + νκγγ δαβ ],
where D = Eh3 /[12(1 − ν 2 )] is the plate bending modulus, and καβ = w,αβ
is the curvature tensor .
It can be seen that the problem of the in-plane forces is identical with the
plane-stress problem, with Nαβ , Fαa , ūα and ε̄αβ corresponding to σαβ , Tαa ,
uα , and εαβ , respectively. For the flexure problem, the equilibrium equa-
tion, when combined with the moment-curvature and curvature-deflection
relations, becomes
q̄
∇4 w = ,
D
where q̄ = q − mα ,α is the effective transverse load per unit area. In what
follows we shall assume, as is the case in most problems, that mα = 0 and
therefore q̄ will be replaced by q.
Axisymmetrically Loaded Circular Plates

Given a circular plate of radius a, if the load q is a function (in polar
coordinates) of r only and if the edge conditions are uniform, then the deflec-
tion w can likewise be assumed to be a function of r only, the only nonzero
shear force is Qr = Q, and the only moments are Mr and Mθ . Equation
(5.2.4) then reduces to
1 d
(rQ) + q = 0,
r dr
which can be integrated to yield
1
Z r
Q=− rq dr,
r 0
and Equation (5.2.5) becomes
dMr Mr − M θ
+ + Q = 0.
dr r
The curvature tensor components are
d2 w 1 dw
κr = , κθ = ,
dr2 r dr
and therefore the elastic relations take the form
! !
d2 w 1 dw 1 dw d2 w
Mr = D 2
+ν , Mθ = D +ν 2 .
dr r dr r dr dr
Substituting this result in the moment-shear equations results in

!
d3 w 1 d2 w 1 dw
D 3
+ 2
− 2 = −Q.
dr r dr r dr
But the left-hand side is just
d 1 d dw

D r ,
dr r dr dr
so that the differential equation can be solved by integration.
The simplest problem is the one where the load is uniform, that is, q =
constant. Then Q = −qr/2, and the integration for w results in
qr4
w(r) = + Ar2 + B ln r + C,
64D
where A, B, C are constants of integration. For the deflection to be finite
at the center we must have B = 0, and it is convenient to set C = 0, that is,
measure the deflection relative to the center rather than relative to the edge.
The remaining constant, A, is then determined from the edge condition.
Clamped Edge. Here the edge condition is w0 (a) = 0, leading to A =
−qa2 /32D. The deflection of the edge relative to the center is thus qa4 /64D,
or equivalently, the center deflection relative to the edge is qa4 /64D. The
moments are
q q
Mr (r) = [(3 + ν)r2 − (1 + ν)a2 ], Mθ (r) = [(1 + 3ν)r2 − (1 + ν)a2 ].
16 16
Simply Supported Edge. We have
(3 + ν)qr2
Mr (r) = + 2(1 + ν)DA,
16
and therefore the condition Mr (a) = 0 leads to A = −(3 + ν)qa2 /32(1 + ν)D.
The deflection is therefore
q
w(r) = [(1 + ν)r4 − 2(3 + ν)a2 r2 ],
64(1 + ν)D
the maximum deflection being (5+ν)qa4 /64(1+ν)D, or, with ν = 0.3, about
four times as large as for the clamped plate. The moments are
(3 + ν)q 2 q
Mr (r) = (r − a2 ), Mθ (r) = [(1 + 3ν)r2 − (3 + ν)a2 ].
16 16
As a preliminary step to determining the deflection due to an arbitrary

axisymmetric load q(r), we consider the case of a force F concentrated on a
circle of radius b. This may be viewed as the limit as c → b of the annular

loading 

 0, 0 < r < b,

F

q(r) = , b < r < c,


 π(c − b2 )
2
0, c < r < a.

It follows that rQ is constant for r c (with Q = 0 in the
former region), and that
Z c F
rQ|cb =− q(r)r dr = − .
b 2π
Consequently, in the limit as c → b we have


d 1 d

dw
 0,
 r < b,
r = F
dr r dr dr 
 , r > b.
2πDr
Denoting the deflections in r b by w1 and w2 , respectively, and
setting w1 (0) = 0 for convenience, we have
w1 (r) = A1 r2 ,
r F 2 r
w2 (r) = A2 r2 + B2 b2 ln + C2 b2 + r ln .
b 8πD b
The continuity conditions w1 (b) = w2 (b), w10 (b) = w20 (b), and w100 (b) = w200 (b)
yield
F F
A2 = A1 − , B2 = C2 = .
8πD 8πD
Thus
F r

2 2 2 2 2
w2 (r) = A1 r + (r + b ) ln + b − r ,
8πD b
!
F r b2
w20 (r) = 2A1 r + 2r ln + −r ,
8πD b r
and in r > b,
" !#
F r b2
Mr (r) = 2(1 + ν)DA1 + 2(1 + ν) ln + (1 − ν) 1 − 2 .
8π b r
If the edge r = a is clamped, then w20 (a) = 0 and therefore

" !#
F a 1 b2
A1 = − ln − 1− 2 ,
8πD b 2 a
while if the edge is simply supported, then Mr (a) = 0 and

" !#
F a 1−ν b2
A1 = − ln + 1− 2 .
8πD b 2(1 + ν) a
These results may be combined by writing

" !#
F a b2
A1 = − ln + λ 1 − 2 ,
8πD b a
where λ equals − 21 for the clamped plate and (1 − ν)/2(1 + ν) for the simply
supported plate. Since we have set w(0) = 0, the center deflection relative
to the edge is
F a

−w(a) = b2 ln − (1 + λ)(a2 − b2 ) .
8πD b
Let the solution derived above be written as

F
w(r) = − g(r, b; λ),
8πD
 " !#
a ρ2
r2 ,

 ln ρ + λ 1 − a2 r < ρ,



g(r, ρ; λ) = !
2 a 2 r 2 r2
ρ2 , r > ρ.

 r ln r − ρ ln ρ + (1 + λ)r − 1+λ


a2

Now consider an arbitrary load distribution q(r). The total load contained in
the infinitesimal annulus ρ < r < ρ + dρ is 2πq(ρ)ρ dρ, and consequently the
deflection due to this load alone is −(1/4D)q(ρ)g(r, ρ; λ)ρ dρ. The deflection
due to the entire load is obtained by superposition:
1
Z a
w(r) = − q(ρ)g(r, ρ; λ)ρ dρ,
4D 0
and for the center deflection, in particular, we have

1
Z a
a

2 2 2
−w(a) = q(ρ) (1 + λ)(a − ρ ) − ρ ln ρ dρ.
4D 0 ρ
Consider, for example, a downward load F uniformly distributed over an

inner circle of radius b; then
F b Z
a

w(a) = 2
(1 + λ)(a2 − ρ2 ) − ρ2 ln ρ dρ
4πb D" 0 ! ρ #
F a2 b2 b2 a b2
= (1 + λ) − − ln + .
4πD 2 4 4 b 16
When b = a, we obtain
F a2
w(a) = [4(1 + λ) − 1] ,
64πD
from which we obtain the previous results for the clamped and simply sup-
ported cases (with q = −F/πa2 ) by inserting the appropriate values of λ.
With b = 0 we obtain the solution for a load concentrated at the center.
The deflection at any r is obtained by evaluating w(r) by superposition, with
r > b, and then taking the limit as b → 0:
" ! #
1 F b a r r2
Z
w(r) = lim · 2 2
r ln − ρ2 ln + (1 + λ)r2 − 1 + λ 2 ρ 2
ρ dρ
b→0 4D πb 0 r ρ a
F a

= r2 ln + (1 + λ)r2 .
8πD r
The maximum deflection is

1, clamped,
F a2


w(a) = 3+ν
16πD 
 , simply supported.
1+ν
The bending moments are
F a

Mr = 2(1 + ν) ln − (1 − ν) + 2(1 + ν)λ ,
8π r
F a

Mθ = 2(1 + ν) ln + (1 − ν) + 2(1 + ν)λ .
8π r
The logarithmic singularity at the center indicates that the assumptions of
elementary plate theory do not hold near the point of application of a con-
centrated load. For more details, see Timoshenko and Woinowsky-Krieger
[1959], Section 5.1.
5.2.3. Yielding of Plates
Plate Yield Criteria

A plate will be said to yield in bending at a point (x1 , x2 ) if the stress
tensor there obeys the yield criterion at every x3 except x3 = 0; points in
the middle plane are regarded as remnants of the elastic core. If the stresses
σi3 are assumed negligible in magnitude next to the σαβ (this does not mean
that they can be neglected in the equilibrium equations, because derivatives
occur there), then we may apply a plane-stress yield criterion, say
!
σαβ
f = 0,
σy
in every plane x3 6= 0. Equilibrium is satisfied if

4
σαβ = − Mαβ sgn x3 . (5.2.9)
h2
If the ultimate moment is defined as MU = σy h2 /4, then the plate yield
criterion is given by
f (mαβ ) = 0,
where mαβ = Mαβ /MU ; the Mises and Tresca criteria become, respectively,
m211 − m11 m22 + m222 + 3m212 = 1 (Mises),

max(|m1 |, |m2 |, |m1 − m2 |) = 1 (Tresca).
A yield criterion that approximates the behavior of doubly reinforced

concrete slabs is the Johansen criterion,
max(|m1 |, |m2 |) = 1.
Problems in contained plastic bending of plates may be studied by means

of numerical methods analogous to those of Section 4.5.1 The plastic collapse
of perfectly plastic plates may in general be investigated by means of limit
analysis, as is done in Section 6.4. If such plates are, however, circular
and axisymmetrically loaded and supported, then exact solutions may be
obtained for the collapse state.
Fully Plastic Axisymmetrically Loaded Circular Plates

If the loading and support are axisymmetric, then the only nonvanishing
moments are Mr and Mθ , and the equilibrium equation is
Z r
(rMr )0 − Mθ = qr dr. (5.2.10)
0
This equation and the yield condition constitute two equations for Mr and
Mθ . Equivalently, if the yield condition is solved for Mθ in terms of Mr
and the resulting expression for Mθ is substituted in (5.2.10), the result
is a nonlinear first-order differential equation for Mr (r). At the center of
the plate, Mr = Mθ and consequently, if the curvature there is positive
(concave upward), Mr (0) = MU constitutes an initial condition with which
the differential equation may be solved. In addition, a boundary condition
at the edge r = a must be satisfied; this yields the ultimate load. Let us
recall that for a simply supported plate, the edge conditions are w = Mr = 0
; thus Mr = 0 is a boundary condition with which the differential equation
may be solved. For a clamped plate, the edge must form a hinge circle,
1
For an introduction to finite-element methods for plates, see Zienkiewicz [1977], Chap-
ter 10.
that is, a locus of slope discontinuity (a special case of the hinge curve
discussed in Section 6.2). As we shall √see, the edge condition there becomes
Mr (a) = −MU or Mr (a) = −2MU / 3 for the Tresca or Mises material,
respectively.
Figure 5.2.2 shows the Mises and Tresca yield criteria for axisymmet-
rically loaded circular plates. It follows from the preceding discussion that
the center of the plate is in the moment state corresponding to point B, and
that a simply supported edge corresponds to point C. A simply supported
plate may thus be assumed to be entirely in the regime BC. For the the
Tresca material, this means that Mθ = MU everywhere, and the problem to
be solved is therefore linear.
Mθ Mises

= B
C
Tresca
9
A
D O MU Mr
D0 s
Figure 5.2.2. Mises and Tresca yield criteria for axisymmetrically loaded circu-
lar plates
Solution for Tresca Plate

Let us consider, for example, a downward load F uniformly distributed
over a circle of radius b, the plate being unloaded outside this circle. The
equilibrium equation is then
F r2

 − , r < b,


2
2πb
(rMr )0 − MU =
 F
 − ,
 r > b.
2π
The solution for r b satisfying the condition at r = a is
F a

Mr = − MU −1 .
2π r
Continuity at r = b requires that
MU
F = 2π .
1 − 2b/3a
This result includes the extreme cases F = 6πMU for the uniformly loaded
plate (b = a) and F = 2πMU for a plate with concentrated load. This last
case could not have been treated directly because the moments would have
to go to infinity at the center —a condition incompatible with plasticity.
On segment BC of the Tresca hexagon, the flow rule yields κ̇r /κ̇θ = 0,
that is, ẇ00 /(ẇ0 /r) = 0. Consequently the velocity field obeying the edge
condition must be ẇ(r) = (1 − r/a)v0 , where v0 is the center velocity (i.e.,
the plate deforms in the shape of a cone). It is shown in Chapter 6 that the
upper-bound load obtained with this velocity field equals the one obtained
here, as indeed it must, since the solution is complete.
In order to study the clamped plate, we must know the velocity fields
associated with the other sides of the Tresca hexagon. On CD we have
κ̇r + κ̇θ = 0, that is, ẇ00 + ẇ0 /r = 0, a differential equation whose solution is
ẇ = C +D ln r. Neither this velocity field nor the preceding (conical) one can
possibly meet a condition of zero slope at a clamped edge, and this is why
a hinge circle is necessary there. Finally, on AB and DE the flow rule gives
κ̇θ /κ̇r = 0. If this is interpreted as κ̇θ = ẇ0 /r = 0, then the velocity field is
ẇ = constant (i.e., rigid-body motion). Alternatively, we may interpret it as
κ̇r = ∞; this would be the state at a hinge circle, and from this follows the
clamped-edge condition Mr = −MU . On the Mises√ellipse, the only relevant
point where κ̇θ /κ̇r = 0 is D0 , where Mr = −2MU / 3.
It follows from these considerations that a clamped plate must be in
regime BC near the center and in CD near the edge; point C gives the state
at r = c for some c such that 0 < c < a. With the same loading as assumed
for the simply supported plate above, we have to solve the problem both for
c > b and c < b.
Case 1 : c > b. The equilibrium equation is
F r2
(rMr )0 = MU − , 0 < r < b,
2πb2
F
(rMr )0 = MU − , b < r < c,
2π
F
rMr0 = MU − , c < r < a.
2π
Let m = Mr /MU and p = F/2πMU ; then the solution satisfying the condi-
tions at r = 0 and r = c is
pr2
m(r) = 1 − 0 < r < b,
3b2
c

m(r) = (p − 1) − 1 , b < r < c,
r
r
m(r) = −(p − 1) ln , c < r < a.
c
The additional conditions to be met are continuity at r = b and m = −1 at
r = a, producing the two equations
1 c

1 − p = (p − 1) −1 ,
3 b
a
(p − 1) ln = 1.
c
The assumption c > b requires 1 < p < 3, and the limiting case c = b, p = 3
corresponds to b/a = e−1/2 . Consequently, the present case represents the
range 0 < b/a < e−1/2 . Eliminating c between the two equations, we obtain
b 3(p − 1) − p−1
1
= e , p < 3.
a 2p
For example, p = 2 corresponds to b/a = 0.276, p = 1.5 to b/a = 0.068, and

p = 1.286 to b/a = 0.01. The ultimate concentrated load is 2πMU (p = 1,
b = 0), the same as for the simply supported plate. This limit is approached,
however, only at extremely small values of b/a.
Case 2 : b > c. The equilibrium equation is
F r2
(rMr )0 = MU − 0 < r < c,
2πb2
F r2
rMr0 = MU − c < r < b,
2πb2
F
rMr0 = MU − b < r < a.
2π
An analysis similar to that in Case 1 leads to
b 5−p+ln(p/3)
− 2(p−1)
=e , p > 3.
a
The extreme case b = a (uniformly loaded plate) corresponds to p = 5.63,
an increase of 88% over the ultimate load of the simply supported case.
Much of the preceding theory is due to H. G. Hopkins and various collab-
orators. For more details and other solutions, see, for example, Hopkins and
Prager [1953], Hopkins and Wang [1954], and Drucker and Hopkins [1955].
Section 5.3 / Plastic Buckling 313
For plates without circular symmetry, complete solutions are not avail-
able. Estimates of collapse loads will be found with the help of the upper-
bound theorem of limit analysis — and exceptionally of the lower-bound
theorem — in 6.4.1.
The effect of large deflections on the collapse load of a simply supported
circular plate was studied by Onat and Haythornthwaite [1956]; the analysis
is based on the deformation of the plate into a conical shell (see also Hodge
[1959], Section 11-7). Approximate methods for determining collapse loads
of shells are studied in 6.4.3.

1. Derive the plate equilibrium equations (5.2.6) by means of virtual work.
Discuss the significance of the shear forces Qα in this formulation.
2. A uniformly loaded circular plate made of an elastic–perfectly plastic

material with yield stress σY is assumed to obey the Tresca criterion.
If the total load is F , determine the value FE at which yielding begins
when the plate is (a) simply supported and (b) clamped. Compare
with the corresponding ultimate loads.
3. Determine the velocity field in a fully plastic clamped circular Tresca

plate of radius a carrying a load F that is uniformly distributed over
an inner circle of radius b. Use the upper-bound theorem to determine
the relation between the load and the ratio b/a. Compare with the
results in the text.
4. A simply supported circular Tresca plate of radius a carries a load F

that is uniformly distributed over the perimeter of the circle r = b.
Find the value of F and the moment distribution when the plate is
fully plastic.
5. A simply supported circular Tresca plate of radius a carries a load

F that is uniformly distributed over the entire plate. The ultimate
moment is MU inside the circle b = a/2 and 21 MU outside this circle.
Find the value of F and the moment distribution when the plate is
fully plastic.
Section 5.3 Plastic Buckling

While Sections 5.1 and 5.2 dealt with the plastic collapse of bodies whose
material behavior may be idealized as perfectly plastic, the present section
is devoted to the study of the buckling collapse of bodies made of work-

hardening material. The elementary theory of elastic column buckling is part
of the knowledge of all students of mechanics. It is worthwhile, however, to
begin this section, in 5.3.1, with a general introduction to stability theory
and to use the buckling of bars as an illustration of the theory. In 5.3.2 we
discuss theories of the modulus that must be used in the determination of
critical loads. Finally, in 5.3.3 the plastic buckling of plates and shells is
studied.
5.3.1. Introduction to Stability Theory
Elastic Stability
In 1.4.3 it was shown that an elastic body under conservative loads is
in equilibrium if and only if the total potential energy Π is stationary with
respect to virtual displacements, a condition expressed by Equation (1.4.18):
δΠ = 0.
It was further stated without proof that the equilibrium is stable only if Π
is a minimum. An intuitive, though not strictly rigorous proof can be based
on the following observation: if the configuration of the body is to change
slightly from the one at equilibrium, and if the potential energy at the altered
configuration is greater than at equilibrium, then additional work must be
done on the body in order to effect the change, and therefore the change
cannot take place spontaneously. The proof can be easily made rigorous
for discrete systems (those with a finite number of degrees of freedom), and
the result is known as the Lagrange–Dirichlet theorem. The proof for
continua runs into technical difficulties, but these will be ignored here, and
the result will be accepted.
Mathematically, the condition that Π is a minimum at equilibrium can
be expressed as follows: let Π denote the potential energy evaluated at the
displacement field u, and Π + ∆Π the potential energy evaluated at the
varied displacement field u + δu. Assuming the dependence of Π on u to be
smooth, we can write
1
∆Π = δΠ + δ 2 Π + . . . ,
2
where δΠ (the first variation defined in 1.4.3) is linear in δu (and/or in
its derivatives, and therefore also in δε), δ 2 Π is quadratic, and so on. We
may limit ourselves to virtual displacements that are small enough so that
terms beyond the quadratic can be neglected. Since δΠ vanishes if Π is
stationary at u, clearly Π is a minimum only if δ 2 Π is nonnegative for all δu
that are compatible with the constraints (i.e., for all virtual displacements).
We may therefore say that the equilibrium is stable only if δ 2 Π > 0 for all
virtual displacements. The criterion for the onset of instability, known as
the Trefftz criterion, is thus
δ 2 Π = 0.
Since Π = Πint + Πext , we have δ 2 Π = δ 2 Πint + δ 2 Πext . An equivalent

statement of the Trefftz criterion is therefore
δ 2 Πint + δ 2 Πext = 0.
In a linearly elastic body,

1
Z
Πint = Cijkl εij εkl dV,
2 R
and therefore
1
Z
∆Πint = Cijkl [(εij + δεij )(εkl + δεkl ) − εij εkl ] dV
2Z R Z
= Cijkl εij δεkl dV + 1
2
Cijkl δεij δεkl dV.
R R
As a result of the definitions of the first and second variations, the first
integral in the last expression is δΠint , and the second integral is δ 2 Πint ,
that is, Z
δ 2 Πint = Cijkl δεij δεkl dV. (5.3.1)
R
Generalization to Quasi-Elastic Materials

A general theory of stability in elastic–plastic solids capable of large
deformations was formulated by Hill [1958]. Here, we shall limit our consid-
eration to infinitesimal deformations. We can then immediately generalize
the preceding result for linearly elastic bodies to nonlinearly elastic ones if
we interpret C as the tangent modulus tensor defined by Cijkl = ∂σij /∂εkl .
Since δσij = Cijkl δεkl , the integrand of (5.3.1) can be rewritten as δσij δεij .
Since it is a given equilibrium state that is examined as to its stability, the
past history of the body is irrelevant to this examination, and therefore the
result can be further extended to materials that are quasi-elastic in the sense
that, at a given state, a small stress variation δσ can be uniquely associated
with a small strain variation δε. As we have seen, rate-independent plastic
materials have this property. We may therefore state, as a generalization to
quasi-elastic bodies of the Trefftz criterion, the energy criterion for the
stability of such bodies subject to conservative loads:
Z
δσij δεij dV + δ 2 Πext = 0. (5.3.2)
R
Furthermore, the first term on the left-hand side will be written as δ 2 Πint
without thereby implying the existence of an internal potential energy Πint .
In other words, we shall define
Z
def
δ 2 Πint = δσij δεij dV.
R
In the absence of internal constraints, the quantity δσij δεij is positive

for all nonvanishing δε if the material is linearly elastic. It is so likewise for
rate-independent plastic materials that are stable in the sense of Drucker
(see 3.2.1), that is, work-hardening and obeying an associated flow rule. For
bodies made of such materials, then, instability can occur only if it is possible
for δ 2 Πext to be negative.
If the loads f and ta are independent of displacement, the external po-
tential energy is Z Z
Πext = − fi ui dV − tai ui dS,
R ∂R
as given in 1.4.3. We see, then, that ∆Πext = δΠext (i.e. δ 2 Πext = 0) unless
the virtual displacement significantly alters the region occupied by the body,
thereby introducing terms that are quadratic in δu into ∆Π.
Buckling of Quasi-Elastic Bars
The classic example of this occurrence is the buckling of a bar (column,
strut) under a compressive axial load P acting through the neutral axis. If
the bar is initially straight, with length L, and then bends, with the deflection
of the neutral axis given by p v(x) (0 < x < L), then the chord spanned by
RL
the neutral axis becomes 0 1 − v 0 2 dx, so that the work done by the load
P is Z Lq ! Z L
. 1 2
P L− 0 2
1 − v dx = P v 0 dx.
0 2 0
If the only other load is a distributed transverse load q, then the external
potential energy is
Z L 1
Z L
2
Πext = − qv dx − P v 0 dx, (5.3.3)
0 2 0
and the first and second variations are, respectively,

Z L Z L
δΠe = − q δv dx − P v 0 δv 0 dx
0 0
Z L L
= (P v − q) δv dx − P v 0 δv 0
00
0
(the last form having been obtained by integration by parts), and

Z L
δ 2 Πext = −P (δv 0 )2 dx.
0
If the bar is subject to a distributed axial compressive load of intensity

p per unit length in addition to end loads P0 and PL at x = 0 and x = L,
respectively, then the external potential energy can be written as
Z L Z L
Πext = − qv dx − ps dx + P0 s(0) − PL s(L),
0 0
where s(x) is the shortening of the bar due to bending at point x, given by
1
Z x
2
s(x) = s(0) + v 0 dx.
2 0
If P (x) now denotes the internal axial force at x (positive in compression),

then equilibrium requires
P 0 + p = 0,
and therefore
Z L Z L 1
Z L
2
− ps dx = 0
P s dx = P s|L
0 + P v 0 dx.
0 0 2 0
Equation (5.3.3) for Πext needs to be changed only by placing P under the
integral sign, and
Z L
δ 2 Πext = − P (δv 0 )2 dx.
0
In accordance with elementary beam theory, the state of stress at each

point will be approximated as uniaxial, so that δσij δεij = δσx δεx . Further-
more, δεx = −y δv 00 , so that
Z L Z Z L
δ 2 Πint = (−y) δσx dA δv 00 dx = δM δv 00 dx,
0 A 0
where M is the bending moment. The energy criterion for bars therefore
takes the form Z Z L L
δM δv 00 dx − P (δv 0 )2 dx = 0. (5.3.4)
0 0
Integration by parts of the first integral leads to
Z L L
(δM 0 + P δv 0 ) δv 0 dx = δM δv 0 0 .
0
Since the end conditions on beams are usually such that either the rotation
v 0 or the bending moment M cannot be varied, the right-hand side of this
equation vanishes.
Finally, we assume an effective modulus Ē such that δM = ĒI δv 00 ; in
a linearly elastic material, of course, this is just the Young’s modulus E.
In nonlinear materials Ē may be assumed to be determined by the average
stress in the bar, σ = P/A. Theories of the effective modulus are discussed
in 5.3.2.
In problems in which constraints on the deflection v itself can be dis-
regarded, the virtual rotation δv 0 may be taken as the unknown variable.
Writing this variable as θ, we obtain, upon observing that the rotation is
not constrained in the interior of the bar, the differential equation
(ĒIθ0 )0 + P θ = 0. (5.3.5)
More generally, the first integral in (5.3.4) must be integrated by parts

twice, so that the energy criterion becomes
Z L L
(δM 0 + P δv 0 )0 δv dx = (δM 0 + P δv 0 ) δv 0 . (5.3.6)
0
The quantity in parentheses may be interpreted as the virtual shear force,

and therefore either it or the virtual displacement can be expected to vanish
at each end. The differential equation expressing the energy criterion is
therefore
(ĒIθ0 )00 + (P θ)0 = 0. (5.3.7)
End-Loaded Prismatic Bars

If the bar is prismatic then A and I are constant. If, moreover, it is
subject to an axial end load only, then P is constant. In such a bar, then,
the average stress P/A and therefore, by hypothesis, the effective modulus
are constant. Equation (5.3.7) thus becomes
ĒIθ000 + P θ0 = 0, (5.3.8)
and can be immediately solved as
θ = B cos λx + C sin λx + D,
where B, C, and D are constants, and λ is defined by

P
λ2 = .
ĒI
If the bar is pinned at both ends, then θ0 (0) = 0 and θ0 (L) = 0. The first
condition requires C = 0, and the second
λB sin λL = 0.
For a nontrivial solution,1 λ and B must both be different from zero, and
instability occurs only if
sin λL = 0,
1
Note that D is irrelevant since it represents a rigid rotation.
that is, if λ = nπ/L, where n is a positive integer. The fundamental mode

of buckling corresponds to n = 1, that is, λ = π/L, and the load producing
it (the critical load ) obeys
P I
= π2 2 .
Ē L
If the bar is elastic, Ē = E, and
EI def
P = π2 = PE ,
L2
where PE is known as the Euler load .
For other end conditions, the equation governing the critical load can be
written as
P I
= π2 2 , (5.3.9)
Ē Le
where Le is known as the effective length of the bar, given by Le = π/λ1 , λ1
being the smallest nonzero value of λ. For a cantilever column, for example,
the characteristic equation is cos λL = 0, leading to a fundamental mode
descibed by λ1 = π/2L, and hence Le = 2L. For a bar that is clamped
at x = 0 and pinned at x = RL, the end conditions are θ(0) = θ0 (L) = 0,
in addition to the constraint 0L θ dx = 0, describing zero deflection of the
pinned end relative to the clamped end. The three conditions lead to the
characteristic equation λL = tan λL, whose lowest root is λ1 L = 4.4934.
Hence Le /L = π/4.4934 = 0.699.
The solution of Equation (5.3.9) for the critical load for inelastic bars
will be postponed until after the discussion of the effective modulus.
5.3.2. Theories of the Effective Modulus
Tangent-Modulus (Engesser–Shanley) Theory

The first analysis of inelastic column buckling is due to Engesser [1889],
who based the calculation of the critical load on the incremental relation
δσx = Et δεx , where Et is the tangent modulus, defined as the slope dσ/dε
of the uniaxial compression curve at the current value of σ = P/A. The
essential assumption is that, while a bending moment requires a nonuni-
form stress distribution, the deviation of the stress from the average is, at
least initially, sufficiently small so that the stress-strain curve can be locally
approximated by straight line with slope Et . Consequently, the effective
modulus Ē is just Et . Since this is a function of σ, it is convenient to
rewrite Equation (5.3.9) in terms of σ rather than P , and to designate the
solution as the critical stress σcr = Pcr /A. It is conventional to define the
def p
“radius of gyration” r = I/A, and to designate Le /r as the slenderness
ratio. Equation (5.3.9) now reads
σ π2
= . (5.3.10)
Et (σ) (Le /r)2
A plot of the solution of this equation, σcr against Le /r, is often called a
column curve. When σcr is given explicitly as a function of Le /r, the relation
is called a column formula.
If the material has a definite elastic-limit stress σE such that Et (σ) = E
for σ < σE , then the slenderness ratio at which σ = σE is called the critical
slenderness ratio, defined by
s
L E

=π .
r cr σE
The critical slenderness ratio is clearly a material property (for mild steel it
is about 90), and for supercritically slender bars the critical load is given by
the elastic solution. The portion of the column curve for Le /r > (L/r)cr is
thus the Euler hyperbola given by
π2E
σcr = .
(Le /r)2
If the material is perfectly plastic, so that σY = σE , then a stress greater

than σE is not possible, and therefore σcr = σE for Le /r ≤ (L/r)cr . For
materials without a definite elastic limit, the column curve approaches the
Euler hyperbola asymptotically as Le /r → ∞. For a material whose uniaxial
stress-strain relation is described by the Ramberg–Osgood equation (2.1.2),
the tangent modulus is easily obtained as
E
Et = , (5.3.11)
1 + αm(σ/σR )m−1
and the column curve is obtained from

m
σ σ π 2 E/σR

+ αm = .
σR σR (Le /r)2
Since such a material hardens indefinitely, no cutoff stress exists for short
bars.
A formula describing a stress-strain relation with no definite elastic limit
that approaches perfect plasticity asymptotically, with an ultimate stress
σ∞ , was proposed by Prager [1942] in the form
Eε
σ = σ∞ tanh .
σ∞
The asymptote is approached quite fast: when the total strain equals twice
the elastic strain (ε = 2σ/E), the stress is already given by σ = 0.9575σ∞ .
The tangent modulus can be readily obtained as
" 2 #
σ

Et = E 1 − .
σ∞
The preceding formula can be easily generalized to

n
σ

Et = E 1 − , (5.3.12)
σ∞
where n = 2 corresponds to the Prager formula. The case n = 1 describes an
exponential stress-strain curve, σ = σ∞ [1 − exp(−Eε/σ∞ )]. The greater the
value of n, the more rapid the approach to perfect plasticity. The column
curve for the generalized Prager formula may be plotted from
σ/σ∞ π 2 E/σ∞
= .
1 − (σ/σ∞ )n (Le /r)2
Explicit column formulas can be obtained from this equation for n = 1 and
n = 2.
The Reduced-Modulus (Kármán) Theory

Soon after the publication of Engesser’s theory, it was recognized by
engineers, beginning with Considère [1891] and eventually including En-
gesser himself, that the tangent-modulus theory was in contradiction with
the elastic–plastic behavior of metals; a formal theory was proposed by von
Kármán [1910], based on the following reasoning.
When an initially straight bar under a compressive axial load begins to
buckle, the fibers on the concave side undergo additional compression, but
in those on the convex side the compressive strain, and hence the stress,
is reduced. The stress change in the latter fibers is consequently elastic.
Writing, for convenience, stress and strain as positive in compression, the
incremental stress-strain relation is accordingly
( )
Et δε, δε > 0,
δσ =
E δε, δε < 0,
where Et is, as before, the elastic–plastic tangent modulus at the stress σ.

Let y = y0 give the location of the neutral fibers, with y > y0 and y < y0
being the areas of additional and reduced compression, respectively. Since
δε = (y − y0 ) δv 00 , the additional bending moment is
Z
δM = (y − y0 ) δσ dA
A
or
δM = Er I δv 00 ,
where
1
Z Z
2 2
Er = E (y − y0 ) dA + Et (y − y0 ) dA (5.3.13)
I y<y0 y>y0
is the reduced modulus. The value of y0 is determined, as usual, by the

constancy of the axial load:
Z
0 = δP = δσ dA,
A
or Z Z
E (y − y0 ) dA + Et (y − y0 ) dA = 0. (5.3.14)
y<y0 y>y0
Elimination of y0 permits the expression of Er as a function of σ and of bar

geometry.
If the bar is rectangular, with width b and depth h, then Equation
(5.3.13), with the help of I = bh3 /12, gives
4
Er = (Et h31 + Eh32 ).
h3
where
h h
− y0 , h 2 = + y0 .
h1 =
2 2
Equation (5.3.14), with the factor b omitted, then becomes
Z y0 Z h/2 1
E (y − y0 ) dy + Et (y − y0 ) dy = (Et h21 − Eh22 ) = 0,
−h/2 y0 2
p
so that h1 /h2 = E/Et , and
√ √
h1 E h2 Et
=√ √ , = √ √ .
h E + Et h E + Et
Finally, then,
4EEt
Er = √ √ 2 .
E + Et
The reduced modulus is seen to be greater than the tangent modulus

by a factor between 1 and 4, and therefore gives a correspondingly greater
critical load.
Figure 5.3.1 shows column curves based on both the tangent-modulus
and reduced-modulus theories, for the Ramberg–Osgood formula with α =
0.1 and m = 10, and for the Prager formula.
σcr
σ0
1.4 Ramberg–Osgood
TM RM
Elastic
1.2
n
σR , Ramberg−Osgood
Prager σ0 =
1.0 σ∞ , Prager
RM
TM
0.8
0.6
0.4
0.2
q
σ0 Le
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 E r
Figure 5.3.1. Column curves based on tangent-modulus (TM) and reduced-
modulus (RM) theories for the Ramberg–Osgood formula with
α = 0.1 and m = 10, and for the Prager formula. The elastic
column curve is included for comparison.
Comparison of the Two Theories

The reduced-modulus theory dominated engineering practice for most
of the first half of the twentieth century, since it was rigorously based on
elastic–plastic theory. However, experiments in which bars are subjected to
increasing axial loads until they buckle have consistently shown the results
to be in much better agreement with the tangent-modulus theory. The first
explanation of this discrepancy was given by Shanley [1947].
The reduced-modulus theory is based on the assumption that buckling
occurs with no first-order change in the axial load, δP = 0. But, as Shanley
wrote, “upon reaching the critical tangent-modulus load, there is nothing
to prevent the column from bending simultaneously with increasing axial
load.” It is thus possible for the neutral fibers to coincide initially with the
extreme fibers on the concave side, all the other fibers undergoing additional
compression. Any other possible location of the neutral fibers would corre-
spond to a higher axial load, and therefore the tangent-modulus load Ptm is
a lower bound to the elastic–plastic buckling load.
The tangent-modulus load is thus not a load at which instability occurs,
but bifurcation: when this load is exceeded the bar may be in one of several
configurations — it may remain straight, or it may be bent in either direction
(buckling in one plane only is assumed). If the bar buckles at Ptm , then, as
the load is increased, the neutral axis moves inward, tending asymptotically
to the position corresponding to the reduced-modulus load. The latter load
is therefore an upper bound to the buckling load.
Effect of Imperfections
In tests on real bars, imperfections such as initial curvature or eccen-
tricity of the load are inevitable. Some bending moment, however slight,
must therefore be present as soon as any load is applied, and consequently
bending proceeds from the beginning of loading. Consider a pinned elastic
column with the load applied at a small distance e from the centroidal axis.
The deflection v can easily be shown to be governed by
EIv 00 + P v = P e,
and the maximum deflection is found to be

" s ! #
π P
vmax = e sec −1 .
2 PE
The deflection remains of the order of e until the load gets close to the Euler
load, when it begins to grow large.
A similar conclusion holds for a column with an initial deflection v0 of
amplitude e. The equation governing the total deflection v under an axial
load P is
EI(v 00 − v000 ) + P v = 0,
and if the column is pinned while v0 is assumed as v0 (x) = e sin(πx/L), then
the maximum deflection is
e
vmax = .
1 − P/PE
For an imperfect column, then, the buckling load may be interpreted as

the load in the vicinity of which the imperfections become significantly am-
plified. It is for this reason that the tangent-modulus load must necessarily
appear as the buckling load of work-hardening elastic–plastic columns, since
the possibility of remaining straight when this load is exceeded is open only
to perfect columns.
In columns made of an elastic–perfectly plastic material, both the re-
duced and tangent moduli are zero, and therefore a perfect column, as dis-
cussed previously, will buckle elastically or yield in direct compression at
supercritical and subcritical slenderness ratios, respectively. Such a sharp
transition is, in fact, hardly ever observed in real columns made of structural
steel, a material that is fairly well represented as elastic–perfectly plastic.
The deviation from theoretical behavior can also be ascribed to imperfec-

tions. An approximate theory, developed by several nineteenth-century au-
thors and recently reviewed by Mortelhand [1987], results in the formula
(1 − σ/σY )σ/σY π2E
= ,
1 − (1 + η)σ/σY σY (Le /r)2
where η = Ahe/2I, h being the beam depth and e, as before, the amplitude
of the initial deflection.
The effect of imperfections is of even greater significance in the analysis
of post-bifurcation behavior. This topic has been extensively reviewed by
Calladine [1973] and Hutchinson [1974].
Other Uniaxial Buckling Problems: Rings and Arches
From the preceding arguments we can infer that in all buckling problems
in which uniaxial stress is assumed, the elastic solution can be used to give
the buckling load with E replaced by Et , provided that Et can be taken as
constant. Take, for example, a circular ring under an external pressure q
per unit length of center line, the radius of the center line being R. The
compressive force in the ring is thus qR, and the stress is σ = qR/A, where
A is the cross-sectional area. The well-known solution is due to Bresse (see
Timoshenko and Gere [1961]). The fundamental buckling mode is shown
in Figure 5.3.2(a), with the ring deformed into an ellipse, and the critical
pressure is
3EI
qcr = 3 ,
R
where I is the second moment of area of the cross-section for bending in the
plane of the ring.
q
C CCW ?
? q
U CW
AA
AAU
es s
@
R BMB
%
Hj
H 3
e B %
X R
B
z
X 9
e R %
- e 2α %
BB R
:

y
X
X e B %
*
H
Y eB %
H
eB%
@
I (b)

K
A
6 OCC A
(a)
Figure 5.3.2. Buckling of a ring or arch: (a) complete ring; (b) hinge-ended
arch.
The four points on the buckled ring whose radial displacement is zero are
the nodes of the buckling modes. Any half of the ring between two opposite
nodes is equivalent to a semicircular arch that is hinged at both ends, and

therefore Bresse’s formula furnishes the critical buckling pressure for such
an arch. More generally, for a hinged-ended arch subtending an angle 2α
the fundamental buckling mode can be expected to be as shown in Figure
5.3.2(b), and the result for the critical pressure is
!
π2 EI
qcr = 2
−1 .
α R3
These results may be immediately converted to the inelastic case. Lim-

iting ourselves to the ring, we obtain
σ 3
= ,
Et (σ) (R/r)2
p
where r = I/A as before. If the stress-strain relation is given by the
generalized Prager formula with n = 1, an explicit formula for the stress,
and hence for the critical pressure, is obtained:
σ∞ A/R
qcr = .
1 + (σ∞ /3E)(R/r)2
For a ring of rectangular cross-section, with depth h (in the radial direction)
and width b (in the axial direction), we have A = bh and r2 = h2 /12.
Defining p = q/b as the pressure in the usual sense (per unit area), we
obtain
σ∞ h/R
pcr = ,
1 + 4(σ∞ /E)(R/h)2
a formula that coincides with that derived by Southwell [1915] when σ∞ is
identified with the yield stress. Southwell’s result was actually intended for
cylindrical tubes rather than rings, but based on an assumption of uniaxial
stress — a highly questionable assumption for a shell, as shown in the next
subsection.
5.3.3. Plastic Buckling of Plates and Shells
Introduction
Consider a flat plate subject to an applied in-plane force per unit length
Fαa around its edge and a distributed in-plane force per unit area pa . The
membrane-force field Nαβ obeys the equilibrium equation (5.2.3) and the
boundary condition νβ Nαβ = Fαa . The equilibrium becomes unstable if it is
possible for the plate to undergo a deflection w(x1 , x2 ) with no additional
forces applied. The membrane forces Nαβ can be assumed to be related to
the average Green–Saint-Venant strains Ēαβ given by Equation (5.2.7), and
if the former do not change, then neither do the latter. The middle plane
will therefore undergo a second-order displacement ūα such that
2ε̄αβ = ūα ,β +ūβ ,α = −w,α w,β .
The work done by the applied forces on this displacement is

I Z I Z
Fαa ūα ds + pα ūα dA = νβ Nαβ ūα ds + pα ūα dA
C A ZC A
= [(Nαβ ,β +pα )ūα + Nαβ ūα ,β ] dA

A
with the help of the two-dimensional divergence theorem. The quantity

in parentheses vanishes as a result of (5.2.3). Since Nαβ is symmetric,
Nαβ ūα ,β = Nαβ ε̄αβ = − 21 Nαβ w,α w,β , and the second variation of the ex-
ternal potential energy can finally be obtained as
Z
δ 2 Πext = Nαβ δw,α δw,β dA. (5.3.15)
A
For δ 2 Πint we have

Z
δ 2 Πint = δMαβ δw,αβ dA. (5.3.16)
A
Since
δMαβ δw,αβ = (δMαβ δw,α ),β −δMαβ ,β δw,α ,
and since the edge conditions are usually such that νβ δMαβ δw,α = 0 on C,
the equation expressing the energy criterion becomes
Z
(δMαβ ,β −Nαβ δw,β ) δw,α dA = 0.
A
Support conditions on a plate are rarely such that the deflection itself is
unconstrained; another integration by parts is usually necessary, leading to
the differential equation
δMαβ ,αβ −(Nαβ δw,β ),α = 0. (5.3.17)
Let the incremental relation between strain and stress for an isotropic
material in a state of plane stress be written in the form
1
ε̇αβ = [(1 + ν̄)σ̇αβ − ν̄ σ̇γγ δαβ ],
Ē
where ν̄ is the instantaneous contraction ratio, and Ē is the instantaneous

modulus, in general not equal to the uniaxial tangent modulus discussed
before. If the tangent-modulus theory is applied to the plate problem, then

an incremental moment-curvature relation can be written in the form
δMαβ = D̄[(1 − ν̄) δw,αβ +ν̄ δw,γγ δαβ ], (5.3.18)
where D̄ = Ēh3 /12(1 − ν̄ 2 ) is the effective plate modulus. The parameters

Ē and ν̄, and hence also D̄, are functions of Nαβ , and the general form
of (5.3.17) when the relations (5.3.18) are substituted is complicated. In
what follows, only examples with a uniform membrane-force field will be
considered.
Circular Plate Under Radial Load
A simple example is that of a circular plate under a uniformly distributed
compressive radial load applied around its edge. Let a be the radius of the
plate, h its thickness, and N the magnitude of the applied load per unit
length of circumference. It is easy to see that a uniform state of plane stress,
σr = σθ = −N/h, is in equilibrium, satisfies the compatibility conditions if
the plate is elastic, and obeys the yield criterion everywhere if it obeys it
anywhere when the plate is plastic.
If the buckling is assumed axisymmetric, the deflection of the middle
plane being w(r), then the second integration by parts leading to (5.3.17)
may be dispensed with, and the energy criterion can be expressed by the
differential equation
1
[(r δMr )0 − δMθ ] + N φ = 0,
r
where φ = δw0 is the virtual rotation of the radial lines, with the prime
denoting differentiation with respect to r. The axisymmetric form of (5.3.18)
is
φ φ

δMr = D̄ φ0 + ν̄ , δMθ = D̄ + ν̄φ0 ,
r r
and the equation governing φ is thus
1 N 1

φ00 + φ0 + − 2 φ = 0, (5.3.19)
r D̄ r
a Bessel equation of order 1. Theq general solution that is regular at r = 0
is φ(r) = J1 (λr), where λ = N/D̄ and J1 is the Bessel function of the
first kind of order 1. Let λ = k/a be the smallest nonzero root for which φ
satisfies the boundary condition. The critical load is then Ncr = σcr h, where
σcr is, in view of the definition of D̄, the solution of the nonlinear equation
2
k 2 Ē h
σ= , (5.3.20)
12 1 − ν̄ 2 a
Ē and ν̄ being, as noted above, functions of σ.
If the edge of the plate is clamped, then the edge condition φ(a) = 0
leads to the characteristic equation J1 (k) = 0, whose smallest nonzero root
is k = 3.832. If the edge of the plate is free to rotate, then Mr must vanish
there, so that the edge condition is φ0 (a) + ν̄φ(a)/a = 0. The characteristic
equation then becomes
1 − ν̄
J0 (k) − J1 (k) = 0.
k
Except in the case of the elastic plate, for which ν̄ = ν (the Poisson’s ratio),
this equation must be solved simultaneously with (5.3.20) in order to find
σcr .
It is now necessary to evaluate Ē and ν̄ as functions of σ. We assume
the plate material to be work-hardening and governed by the Mises criterion
and its associated flow rule. The plastic strain rate can then be written as
9 skl ṡkl
ε̇pij = sij ,
4 HσY2
where σY is the current value of the uniaxial yield stress and H is the uniaxial
plastic modulus (related to the tangent modulus by Hinv = Et inv − Einv),
as can easily be verified by the substitutions s11 = 23 σ, s22 = s33 = − 31 σ,
and |σ| = σY . In a state of plane stress,
1
skl ṡkl = σαβ σ̇αβ − σαα σ̇ββ ,
3
and therefore, if currently σ1 = σ2 = σ,
1
ε̇p1 = ε̇p2 = (σ̇1 + σ̇2 ).
4H
The complete incremental stress-strain relations are therefore
1 1 ν 1

ε̇1 = + σ̇1 − − σ̇2 ,
E 4H E 4H
1 1 ν 1

ε̇2 = + σ̇2 − − σ̇1 .
E 4H E 4H
Thus
1 1 1 1 3
= + = +
Ē E 4H 4Et 4E
and
1 + 4ν − E/Et
ν̄ = .
3 + E/Et
.
It should be noticed that for small values of Et /E, ν̄ = −1 + 4(1 + ν)Et /E
.
and Ē = 4Et , so that the factor Ē/(1 − ν̄ 2 ) in Equation (5.3.20) tends to a
constant fraction, 21 (1−ν), of the elastic value. This result, which is similar to
what occurs in other plate and shell buckling problems, is quite unreasonable
when compared with the uniaxial case, and indeed with experimental data.
Furthermore, the result is not limited to the Mises criterion but would also be
produced by any isotropic yield criterion that is smooth at σ1 = σ2 = ±σY ,
since all such yield loci must be tangent there. We are left with the conclusion
that incremental plasticity with a smooth yield surface may not be applicable
to the analysis of multiaxial instability problems.
Considerable improvement is obtained when the deformation theory of
plasticity discussed in 3.2.1 is used. The Hencky theory, in particular, is
based on the Mises criterion, and gives the plastic strain as
3ε̄p
εpij = sij , (5.3.21)
2σ̄
where r r
3 2 p p
σ̄ = σij σij , ε̄p = ε ε
2 3 ij ij
are the equivalent stress and plastic strain, related to each other by the
uniaxial relation. The incremental form of (5.3.21) is
3 ε̄p

dεpij = dε̄ − dσ̄ sij + ε̄p dsij .
p
2σ̄ σ̄
Upon introducing the uniaxial secant and tangent moduli Es and Et , defined
by
1 1 ε̄p 1 1 dε̄p
= + , = + ,
Es E σ̄ Et E dσ̄
the incremental relation may be written as
3 1 1 dσ̄ 1 1

dεpij = − sij + − dsij .
2 Et Es σ̄ Es E
Applying the relation to the plane-stress case with σ1 = σ2 leads to the
instantaneous modulus and contraction ratio
1 1 3
= +
Ē 4Et 4Es
and
E/Et + 2(1 − 2ν) − 3E/Es
ν̄ = − .
E/Et + 3E/Es
For a gradually flattening uniaxial stress-strain curve, when Et Es E
. .
we have Ē = 4Et as in the incremental theory, but ν̄ = −(1 − 6Et /Es ), and
.
therefore Ē/(1 − ν̄ 2 ) = Es /3. To the first approximation, then, it is the
secant modulus, rather than the elastic modulus, that governs the buckling,
resulting in a much smaller critical load than that given by the incremental
theory.
An explanation of the failure of incremental theory based on an isotropic

yield criterion to predict a reasonable buckling load is due to Phillips [1972],
who points out that the yield criteria of work-hardening materials become
anisotropic almost immediately upon plastic loading.
Torsional Buckling of a Cruciform Column
If a column is made up of thin plate sections that do not form a closed
tube and is sufficiently short, then under the action of a compressive axial
load it will buckle by twisting rather than bending. Consider, for example,
the cross-shaped column shown in Figure 5.3.3, and in particular the flange
whose middle plane is the xy-plane with y positive. If the virtual twist angle
per unit length of the cross-section at x is φ(x), then the virtual deflection
at (x, y) is δw(x, y) = yφ(x). The axial load P may be assumed to be
uniformly distributed with intensity P/4b per unit width of flange. The
second-order external potential energy on a fiber of width dy is therefore
P
Z L
2
d(δ Πext ) = − dy y 2 φ02 dx,
4b 0
and, for the whole flange,

y=b P b2 L
Z Z
2
δ Πext = 2
d(δ Πext ) = − φ02 dx.
y=0 12 0
The energy criterion may be expressed by adding this quantity to δ 2 Πint

as given by (5.3.16) and equating the result to zero. Now
δw,11 = yφ00 , δw,12 = φ0 , δw,22 = 0,
so that
δMαβ δw,αβ = D̄[y 2 φ002 + (1 − ν̄)φ02 ].
However, φ00 is of order φ0 /L, and, if L b, then the first term in brackets
may be neglected in comparison with the second. Consequently, indepen-
dently of the form taken by φ0 ,
12D̄(1 − ν̄)
Pcr = ,
b
or, because of the definition of D̄, we can define σcr = Pcr /4bh as the solution
of 2
h
σ = Ḡ ,
b
where Ḡ = Ē/2(1 + ν̄) is the instantaneous shear modulus, dτ /dγ. Note
that the result is independent of the column length L.
In any incremental theory with a yield criterion of the form (3.3.5) and
an associated flow rule, when τ = 0 the normal to the yield locus is directed
x
"??
???
"
????? "?????
??
"
???
" 6
"
?
b -
L
#
"#
"
"#
"#
"
" ##
"
kφ#
q
0 ? z
66 "" 666666
"" 66
" 666
"
y
""
Figure 5.3.3. Cruciform (cross-shaped) column: initial and buckled geometry.
in the σ-direction, and therefore dγ p = 0. Consequently Ḡ = G, so that the

buckling load is unaffected by plasticity — again an untenable result.
On the other hand, in the Hencky theory we have
1 1

dγ p = 3 − dτ,
Es E
so that
Es
Ḡ = .
3 + (1 − 2ν)Es /E
Once again, it is primarily the secant modulus that determines the critical
load. Experiments by Gerard and Becker [1957] on aluminum columns show
very good agreement with the prediction of deformation theory for ν = 21
(see Figure 5.3.4).
Shell Under External Pressure

The behavior of shells is in general much more complicated than that of
plates. However, the buckling behavior of certain thin-walled shells may be
studied by means of a simplified theory, known as the Donnell–Mushtari–
Vlasov (DMV) theory, whose structure is essentially the same as that of
plate theory (see Niordson [1985], Chapter 15). It must be kept in mind,
1.2
r r r
rr rrr r
rrrr rrr
(a)
1.0 rrrrr rr
rr rr
rr rr r
.8 rrr
(b)
σcr
G(h/b)2.6 rr
r
r rr
.4 r
.2
0
0 10 20 30 40 50 60
σcr (ksi)
Figure 5.3.4. Torsional buckling of a cross-shaped column: predictions of (a)

incremental and (b) deformation theories with ν = 1/2. Dots
represent experimental results (Gerard and Becker [1957]).
though, that shells are far more imperfection-sensitive than bars or plates;
even the slightest imperfections can greatly reduce the buckling load.
For an elastic thin-walled cylindrical tube of mean radius R and thickness
h, subject to an external pressure p, the critical pressure can be found from
the Bresse formula for the ring by substituting D for EI, resulting in
3
E h

pcr = .
4(1 − ν 2 ) R
The substitution is equivalent to assuming that the axial strain is zero. In

other words the tube, when viewed axially, is in a state of plane strain as
opposed to the plane-stress state of the ring. The axial stress is thus σz =
νσθ , where σθ = −pR/h is the circumferential stress, and the circumferential
stress-strain relation is σθ = Eεθ /(1 − ν 2 ).
In an elastic–plastic tube governed by incremental theory, the plane-
strain condition can only be enforced incrementally, that is, dσz = ν̄ dσθ . If
buckling occurs after yielding, then the state of stress just before buckling is
not known a priori , but must be determined by integrating the incremental
relation, and checking at each step whether the buckling criterion is met.
The situation is somewhat simpler with deformation theory. The condi-
tion εz = 0 results in
1 Es

σz = 1 − (1 − 2ν) σθ .
2 E
The incremental relation is

1 Es 1 − 2ν dEs

dσz = 1 − (1 − 2ν) dσθ − dσ̄.
2 E 2E dσ̄
But
dEs Es Es

=− −1
dσ̄ σ̄ Et
and
q (2σθ − σz ) dσθ + (2σz − σθ ) dσz
dσ̄ = d σθ2 − σθ σz + σz2 = .
2σ̄
Thus both dσ̄ and dσz can be expressed as multiples of dσθ , and the expres-
sions can be substituted in
1 1 1 1 2σθ − σz

dεθ = (dσθ − ν dσz ) + − dσ̄,
E 2 Es E σ̄
resulting in
1
dε̄θ = dσθ ,
Ẽ
where Ẽ is in general a complicated function of σθ ; it is this Ẽ that replaces
E in the Bresse formula. When ν = 21 , however, Ẽ turns out to be just
4Et /3, where Et is the uniaxial tangent modulus. In this case the same
result is given by incremental plasticity theory.
The problem of the spherical shell is easier, since the state of stress before
buckling is σθ = σφ = −σ, where σ = pR/2h. The critical value of σ for
elastic buckling is given by
E h
σcr = p 2
.
3(1 − ν ) R
The value for plastic buckling can be obtained by solving
Ē h
σ=p ,
3(1 − ν̄ 2 ) R
where Ē and ν̄ are the same functions of σ as are used for the circular plate
under radial force. Figure 5.3.5 shows the critical stress for the buckling of a
spherical shell, based on both the incremental and deformation theories, us-
ing the Ramberg–Osgood formula with α = 0.1 and m = 6. For comparison,
the stress-strain curve is also shown.

1. Find the reduced modulus Er in terms of the elastic modulus E and the
tangent modulus Et for (a) an ideal sandwich section and (b) for a thin-
walled square tube section. Compare with the result for a rectangular
section when Et /E = 0.1 and when Et /E = 0.02.
2.0
1.8
(a)
1.6 (c)
(b)
1.4
σcr
σR
1.2
or
σ
σR 1.0
0.8 (a) Incremental theory

(b) Deformation theory
0.6 (c) Stress-strain curve
0.4
0.2
0
0 1 2 3 4 5 6 7

p 1 Eh
σR R
Eε
or σ
R
3(1 − ν 2 )
Figure 5.3.5. Buckling of a spherical shell under internal pressure: critical stress
based on (a) incremental and (b) deformation theories, with (c)
stress-strain curve included (from Hutchinson [1972]).
2. Plot column curves based on both the tangent-modulus and the re-
duced-modulus theories (for a rectangular section) for (a) the gener-
alized Prager formula with n = 1, (b) the generalized Prager formula
with n = 4, (c) the Ramberg–Osgood formula with α = 0.3 and m = 4.
3. For a ring under radial pressure, plot suitably nondimensionalized

buckling curves (qcr against R/r) on the basis of (a) the Prager formula
and (b) the Ramberg–Osgood formula with α = 0.2 and m = 6.
4. Is it possible to find an instantaneous plate modulus D̄ and a con-

traction ratio ν̄ for the buckling problem of a circular plate under a
compressive radial load when the plate material obeys the incremental
theory of plasticity with the Tresca criterion and its associated flow
rule? Explain.
5. Perform the analysis leading to the results shown in Figure 5.3.5.

Chapter 6
Problems in Plastic Flow and

Collapse II
Applications of Limit Analysis
Introduction
The theorems of limit analysis for standard elastic–perfectly plastic three-
dimensional continua in arbitrary states of deformation were proved in 3.5.1.
The proof, due to Drucker, Prager, and Greenberg [1952],1 was the final link
in a chain of development of the theory, which began with proofs for beams
and frames by Gvozdev [1938], Horne [1950], and Greenberg and Prager
[1951], followed by a proof for bodies in plane strain by Drucker, Greenberg,
and Prager [1951].
In this chapter applications of the theorems are presented. Section 6.1
deals with plane problems in both plane strain and plane stress. Section
6.2 deals with beams under combined loading (including arches), Section 6.3
with trusses and frames, and Section 6.4 with plates and shells.
It is shown that as a rule, plausible velocity fields are easier to guess than
stress fields, and therefore in many cases only upper-bound estimates are
available. Of particular importance are velocity fields called mechanisms,
in which deformation is concentrated at points, lines, or planes, with the
remaining parts of the system moving as rigid bodies. The use of mechanisms
for estimating collapse loads antedates the development of plasticity theory.
Examples include Coulomb’s method of slip planes for studying the collapse
strength of soil, the plastic-hinge mechanism due to Kazinczy [1914] for steel
frames, and the yield-line theory of Johansen [1932] for reinforced-concrete
slabs, later extended to plates in general.
1
But already given in the book by Prager and Hodge [1951].
337
338 Chapter 6 / Problems in Plastic Flow and Collapse II
Section 6.1 Limit Analysis of Plane Problems

As pointed out in Section 5.1, slip-line theory as a rule gives only upper
bounds, unless an admissible extension of the stress field from the region
covered by the slip-line field to the rest of the body is found. A convenient
method for finding lower bounds is by means of discontinuous stress fields,
some examples of which were studied in Section 5.1.
6.1.1. Blocks and Slabs with Grooves or Cutouts
Among the earliest applications of limit analysis to plane problems are stud-
ies of the effects of cutouts on the yielding of rectangular slabs or blocks
subject to tension perpendicular to a side. In the absence of cutouts, the
slab may be assumed to collapse under a uniform state of tensile stress
σ = σY = 2k, the Tresca criterion being assumed in plane stress. The
cutouts may be expected to reduce the average or nominal tensile stress σ
required for collapse to 2ρk (0 < ρ < 1), where ρ is called the cutout factor .
Limit analysis may be used to find bounds on the cutout factor.
Tension of a Grooved Rectangular Block

The following example, illustrated in Figure 6.1.1, is discussed by Prager
and Hodge [1951]. A statically admissible stress field is shown in Figure
6.1.1(a). The trapezoidal regions on either side of the line of symmetry
OO correspond to the truncated wedge of Figure 5.1.6(e) (page 289), except
that the stresses are tensile rather than compressive. The rectangular regions
beyond the trapezoids are in simple tension with stress σ = 2k(1 − sin γ),
where γ is the wedge semi-angle. In order to produce the best lower bound, γ
should be as small as possible, subject to the geometric restriction discussed
in connection with Figure 5.1.6(e) and the evident additional restriction
that the flanks of the wedge be tangent to the cutout circles. It can be
shown, the details being left to an exercise, that these restrictions require
that 2 sin 2γ = 1 + sin γ. The smallest angle obeying this equation is 0.379
radian, so that sin γ = 0.370, and the lower bound to the cutout factor is
0.630.
A simple kinematically admissible velocity field is shown in Figure 6.1.1(b).
The portion above the line AB slides rigidly along this line with respect to
the portion below. If the sliding speed is v, then the plastic√ dissipation per
unit area along AB is kv, while the length of √ the line is 3a 2. The external
rate of work per unit thickness is σ · 4a · v/ √2. Equating this rate to the
total internal dissipation per unit thickness, 3 2kva, yields the upper bound
σ = 3k/2, or ρ = 3/4.
Figure 6.1.1(c) shows a slip-line field in which the region bounded by
the slip lines AB and A0 B (as well as its mirror image) is in a state of
Section 6.1 / Limit Analysis of Plane Problems 339
B
Q
L Q

L

Q

Q JJJJJ J
JJ
45◦
L

QJJJJ J J I -

JJ J
L

J JJJJJ JJ
J B A0

a L

J J Ia
J
b b b b b AAU q bO

C
L

J
J A
O J
J
L O
JJJ
JJ J
L A

γ J
j JJJJ JJ
J
L
J JJJJQ
J
JJJJ J Q

L
J
J
JJ Q
L
Q

QL
(c)
2a - 2a -
(a) (b)
Figure 6.1.1. Tension of a grooved rectangular block: (a) statically admissible

stress field; (b) kinematically admissible velocity field; (c) slip-
line field.
axisymmetric stress as studied in Section 4.4. The stress σθ in this region is

r

σθ = 2k 1 + ln ,
a
where r is measured from the center O of the circle. The resultant axial
force per unit thickness is therefore
Z 2a Z 2a r

F =2 σθ dr = 4k 1 + ln dr
a a a
= 8ka ln 2.
The nominal stress is thus σ = F/4a = 2k ln 2, corresponding to a cutout
factor of ln 2 = 0.693. Since the stress field has not been extended to the
regions outside the slip-line field, this value cannot be a lower bound. On
the other hand, the slip-line field implies a solution for the velocity field,
which, together with the rigid axial motion of the regions outside it, forms a
kinematically admissible velocity field for the body as a whole, and therefore
the result gives an improved upper bound for the cutout factor. The bounds
on the cutout factor are therefore
0.630 ≤ ρ ≤ 0.693.
Tension of a Square Slab With a Slit

A thin square slab with a narrow slit perpendicular to the direction of
the load is shown in Figure 6.1.2(a). The slit width is assumed to be of
the same order of magnitude as the slab thickness, both being much smaller
than the sides of the square. A kinematically admissible velocity field may
be based on a shearing failure mode as shown in Figure 6.1.2(b). If the
shearing plane makes an angle α with the load direction then the total area
of the surface of sliding is (1 − β)ah csc α, where h is the slab thickness; and
if the relative velocity of motion of the two parts of the slab is v, then the
plastic dissipation per unit area is kv sec α. The total internal dissipation is
therefore 2k(1 − β)ahv csc 2α. The external work rate is σahv, and therefore
an upper bound to the cutout factor is (1 − β) csc 2α. The best upper bound
is obtained for α = π/4, and gives ρ = 1 − β.
666666666666666666 666
c
T #
Tcc #
1 #
T c #
T 2c #2
? T c
c##
α ξa H
T 4 HH 3
H 4
- 2βa - - 6 3 H H T
(1 − β)a (1 − β)a 2 #c 2
#c T
# c T
1
## c T
c
# cT

# cT
?????????????????? ???
(c)
2a - - h
(a) (b)
Figure 6.1.2. Tension of a square slab with a narrow slit: (a) geometry and
loading; (b) kinematically admissible velocity field; (c) statically
admissible stress field.
A discontinuous stress field for the determination of a lower bound was

constructed by Hodge [1953] (see Hodge [1959], Section 12-2), and is shown
in Figure 6.1.2(c). Since the problem is one of plane stress and not plane
strain, it is not the Prager jump conditions (5.1.5) but the more general
jump conditions (5.1.6), together with the yield criterion, that must be used.
There being four distinct regions, the total number of unknown stress vari-
ables (ni , ri , θi ) (i = 1, 2, 3, 4) is 12. The equations for these variables are
furnished by, first, two traction boundary conditions each on two sides of
the square and on the face of the slit, and second, by two jump conditions
each on the boundaries between regions 1 and 2, 2 and 3, and 2 and 4. In
particular, the fact that τxy vanishes on the external boundaries of regions
1, 3 and 4 means that it vanishes throughout these regions. Regions 3 and 4
can reasonably be assumed to be in a state of simple tension and compres-
sion, respectively, so that θ3 = θ4 = 3π/4. In region 1, θ is either 3π/4 or
π/4; it will be assumed that σy ≥ σx there, so that θ1 = π/4. The remaining

boundary conditions accordingly give
n1 − r1 = σ, n3 − r3 = 0, n4 + r4 = 0,
so that n1 , n3 and n4 can be eliminated. The jump conditions are
n2 + r2 sin 2(θ2 − χ12 ) = σ + r1 cos 2χ12 ,

n2 + r2 sin 2(θ2 − χ23 ) = r3 (1 − cos 2χ23 ),
n2 + r2 sin 2(θ2 − χ24 ) = −r4 (1 + cos 2χ24 ),
r2 cos 2(θ2 − χ12 ) = r1 sin 2χ12 ,
r2 cos 2(θ2 − χ23 ) = −r3 sin 2χ23 ,
r2 cos 2(θ2 − χ24 ) = −r4 sin 2χ24 ,
where the angles χ12 , χ23 and χ24 , giving the inclinations of the normals to
the boundary lines with the x-axis, can be expressed in terms of β and ξ.
Once the equations are solved, satisfaction of the yield inequality
2r + |n − r| + |n − r| ≤ 4k
in each region produces four inequalities on the cutout factor ρ = σ/2k:
ρβ ≤ 1 − ξ, ρ ≤ 1 − β, ρβ ≤ ξ,
q
ρ [ξ + β(1 − β)]2 + 4β 2 ≤ β + ξ(1 − β).
If the second inequality were satisfied as an equality, then the result β = 1−ρ
would coincide with the upper bound. It turns out that if the equality is
assumed, then the remaining inequalities are obeyed if ξ = 1 − β + β 2 . In
this problem, then, the exact cutout factor has been found. Other problems
involving slabs with cutouts were treated by Hodge and various coworkers
(see Hodge [1959], Chapter 12, for a survey).
6.1.2. Problems in Bending
Pure Bending of a Notched Bar

The notched bar shown in Figure 6.1.3(a) is of rectangular cross-section
and subject to equal and opposite couples M applied at its ends. The discon-
tinuous stress field shown is statically admissible, and plastically admissible
in classical plane strain as well as for the Tresca criterion in plane stress. It
is, of course, equivalent to the limiting stress distribution in a perfect beam,
limited to the material below the notch, and gives a lower bound of 12 kba2
for the ultimate moment, where b is the width of the beam.
Stress-free
e % Stress-free
e % O
e%
M - 2k 6 M
a
2k -
?
(a)
Rotating rigid
regions

e %

ω e A % Iω O
II
e ...%
r......
..
...
.... I
M .....
......
.....
....... M
R α
, ......
.......
........
C
.......
........ D
.......
........
.......
,........
.......
......
.......
......
.......
......
.....
....
.....
....
...
ll6 ....
B I
@
Thin zones in which @ Static rigid
shearing occurs “hinge”
(b)
Figure 6.1.3. Pure bending of a notched bar: (a) statically admissible stress
field; (b) mechanism.
An upper bound can be found on the basis of a mechanism, shown in

Figure 6.1.3(b), in which the outer portions of the bar rotate rigidly by
sliding along the arcs ACB and ADB, the inner region ACBD remaining
stationary; the arcs must accordingly be circular. The mechanism resembles
the plastic hinge discussed in 4.4.2.
If the angular velocity of rotation is ω and the radius of the arcs is
r, then the plastic dissipation per unit area of sliding surface is krω. If
the angle subtended by the arcs is 2α, then the total area of the sliding
surfaces is 4brα. But r = 21 a csc α, so that the total internal dissipation
is kωba2 α csc2 α. Equating this to the external work rate 2M ω gives the
upper bound M = 21 kba2 α csc2 α. The smallest value of this occurs when
tan α = 2α, and gives the upper bound of M = 0.69kba2 . For a bar that is
wide enough to be regarded as bending in plane strain, A. P. Green [1953]
found a slip-line field that gives the improved upper bound of M = 0.63kba2
End-Loaded Cantilever in Plane Strain and Plane Stress

Another problem studied by Green [1954a] by means of slip-line theory
concerns the impending collapse of an end-loaded wide tapered cantilever,
shown in Figure 6.1.4.1 If the taper is not extreme and if the ratio L/h of
the length to the least depth is sufficiently great, it is reasonable to suppose
that the collapse mechanism is of the plastic-hinge type. The slip-line fields
shown in Figures 6.1.4(a) and (b) produce deformations similar to those
1
For the prismatic beam, the problem was also studied by Onat and Shield [1955].
corresponding to a hinge mechanism. In (a), the rigid portion to the right of

the slip-line field rotates about point Y ; in (b) it slides over the circular arc
P Y Q, as in the preceding problem, and AP Y QB is a continuous slip line
that is a locus of velocity discontinuity.
L - A D
A
D -
?θ
6 P
Y O h Y
C C O
Q
?
E 6θ
-
B
B E
? ?
F F
(a) (b)
XX
XXX
XXX
0 ◦
F
XXX 75-
1 L -
y XXX
X
z 2k Tension ?
k @
@ 6
Unstressed 2 I h
4
Rk
O : @
@ ?
Compression
–2k9

3 W O-◦ 75
0 ? (d)
F
(c)
Figure 6.1.4. End-loaded tapered cantilever: (a)–(b) slip-line fields; (c) sta-
tically admissible stress field. (d) Weakly supported prismatic
cantilever.
A complete solution is found, with the help of the discontinuous but

statically and plastically admissible stress field shown in Figure 6.1.4(c), for
the beam with 15◦ taper (θ = 75◦ ), in which case FU = kbh for all values
of L/h. For greater values of θ, no extension of the stress field into the
rigid regions is given and therefore the calculated value of F is an upper
bound to FU , though, as Green argues, it is likely to be very close, since
the proposed slip-line field shows remarkable similarity with the plastically
deformed region observed in experiments. For θ ≥ 75◦ the load is given, in
accordance with the slip-line field in (a), by
L 1
F = 2kbh sin 2θ for ≥ tan θ(sin 2θ − cos 2θ).
h 2
For a given θ, as L/h is decreased below the limiting value, F increases.

For prismatic beams, Green [1954a,b] also constructed solutions for uni-
formly distributed loading, and for other boundary conditions, including
“weakly” supported cantilevers (see Figure 6.1.4(d)) as well as beams fixed
at both ends.
A much simpler velocity field for a prismatic beam is shown in Figure
6.1.5(a). In this picture region 1 undergoes simple shearing, regions 2 and
3 rotate rigidly, and regions 4 and 5 undergo tension or compression. In
the last-named regions the vertical velocity does not vanish, so that the
condition of no motion at the wall is not satisfied. A modification of the
mechanism that satisfies this condition, shown in Figure 6.1.5(b), was pro-
posed by Drucker [1956a]; here region 6 does not displace. The mechanism of
Figure 6.1.5(a) is equivalent (with the direction of the force reversed) to that
of Figure 6.1.5(c) for a center-loaded simply supported beam. Calculations
are done here only for mechanism (a).
y
4
π/4 π/4

2 6 @4 2 @4 2
I? @ @ I?
6
x 1 d h 6 1 1
5 5
?

@
3 5@ 3
@ 3
@ ? @ @
L - L - 6
L - L -6
? ? ?
F F F 2F F
(a) (b) (c)
Figure 6.1.5. Velocity fields for a prismatic beam: (a) simple velocity field for
a cantilever; (b) Drucker’s modification; (c) center-loaded simply
supported beam.
The velocity components vx and vy will be denoted u̇ and v̇, respectively,

with the subscripts referring to the regions. The mechanism implies the
following velocity field:
˙ 1− x ,

u̇1 = 0, v̇1 = ∆
L
∆˙ d

x

u̇2 = y− ˙
, v̇2 = ∆ 1 − ,
L 2 L
˙
˙ 1 − x , v̇4 = ∆ y − d .

u̇4 = ∆
L L 2
The velocities in regions 3 and 5 are analogous to those in 2 and 4, re-
spectively. Note that the velocity field is continuous. Note further that
ε̇x + ε̇y = 0 in every region, so that the mechanism applies to plane strain
as well as to plain stress. For either the Mises or the Tresca criterion, the
˙
dissipation Dp (ε̇) equals τY ∆/L ˙
in region 1 and 2τY ∆/L in regions 4 and 5.
Equating the external work rate to the total internal dissipation,
˙
∆ ˙
∆

h−d 2

˙
F ∆ = τY · bLd + 2τY · b ,
L L 2
leads to the upper bound
" 2 #
d h d

F = τY bh + 1− . (6.1.1)
h 2L h
The upper bound can be optimized by minimizing with respect to d. The

minimum occurs at d = h − L and leads to F = τY bh(1 − L/2h); but this
result is valid only for L < h. For L ≥ h, d = 0, that is, the mechanism is a
plastic hinge consisting of two deforming triangles, and F = τY bh2 /2L — a
result that, in plane stress, agrees with that of elementary beam theory for
the Tresca criterion.
End-Loaded Cantilever in Plane Stress: Lower Bound

Drucker [1956a] also constructed a lower bound for the problem by means
of the following statically admissible stress field:
F x 2α2 2|y|

σx = cos α 1 − sgn y,
bh h 1 − cos α h
F α 2|y|

τxy = sin α 1 − , σy = 0,
bh 1 − cos α h
with α ≤ π/2. The yield criterion (3.3.5)1 is met at x = L if
hσY 1 − cos α
α= and F = τY bh . (6.1.2)
2LτY α
In accordance with the limitation on α, the result is limited to h/L ≤

πτY /σY . It can easily be ascertained that for small values of h/L, the result
for F approaches σY bh2 /4L as in the beam-theory approach.
For h/L > πτY /σY , the stress field is
d
σx = 0, σy = 0, τxy = τY for |y| < ,
2
and
x π 2|y| − d

σx = σY cos 1− sgn y, σy = 0,
L 2 h−d
1
Drucker [1956a] considered a Tresca material only (i.e., σY = 2τY ). The extension to
the more general yield criterion (3.3.5) is straightforward.
π 2|y| − d d

τxy = τY sin 1− for |y| > .
2 h−d 2
The stress field is in equilibrium if d = h − πτY L/σY . The load is then
2 d π τY L

F = τY bh 1 + −1 = τY bh 1 − (π − 2) . (6.1.3)
π h 2 σY h
The upper and lower bounds are compared in Figure 6.1.6.
2F L
τY bh2
1.0
0.8
Upper
σY Lower
1 bound
=2 bound
)
0.6 τY
(Tresca)
0.4
0.2
h
0
0.02 0.04 0.1 0.2 0.4 1.0 2.0 4.0 10.0 20.0 L
Figure 6.1.6. Upper and lower bounds for an end-loaded prismatic cantilever
beam
I-Beams
Green [1954b] extended his results for prismatic beams to I-beams by
assuming that the slip-line fields derived for rectangular beams in plane stress
prevail in the web, while the flanges are in pure tension or compression. The
shear force is thus carried entirely by the web, while the bending moment
is the sum of that provided by the web and the couple formed by the flange
forces.
The simple velocity fields shown in Figure 6.1.5 were proposed for short
I-beams by Leth [1954] with d as the actual web depth. The upper-bound
load based on the field in (a) or (c) is
h−d

F = τY Aw + 2Af ,
L
where Aw = tw d is the web area and Af = bf (h − d) is the flange area,

tw and bf being respectively the web thickness and flange width. Since Af
and Aw are typically of the same order of magnitude, the second term in
parentheses may usually be neglected, and the upper-bound loaded may be
approximated as τY Aw .
For sufficiently long I-beams, however, it is reasonable to assume that

.
failure is by a plastic-hinge mechanism, with F = MU /L as given by ele-
mentary beam theory. For an I-beam,
σY . σY h
MU = [Aw d + 2Af (h + d)] = (Aw + 4Af ),
4 4
the approximate expression being based on the assumption that h − d
h. Comparing the two upper bounds, then, we find that the plastic-hinge
mechanism furnishes the lower one for
L σY Af

≥ 1+4 ,
h 4τY Aw
approximately, or about 5/2 for a beam made of Tresca material with Af =
Aw . Additional results in the limit analysis of beams, derived on the basis
of local behavior, are discussed in Section 6.2 (see 6.2.4).
6.1.3. Problems in Soil Mechanics
Yield Criterion and Flow Rule

The most commonly used yield criterion for soils is the Mohr–Coulomb
criterion discussed in 3.3.3,
σmax − σmin + (σmax + σmin ) sin φ = 2c cos φ,
where c is the cohesion and φ is the angle of internal friction. The Mohr–
Coulomb criterion includes, as limiting cases, (1) the Tresca criterion (with
φ = 0 and c = k), used to describe, for example, saturated clays, and (2)
the cohesionless friction model (c = 0) for cohesionless soils (dry sands and
gravels).
If the material is taken as standard, then the flow rule at a regular point
of the yield surface is
ε̇pmax = λ̇(1 + sin φ), ε̇pmin = −λ̇(1 − sin φ), ε̇pint = 0.
The flow rule implies a constant dilatancy ratio, defined as (ε̇p1 +ε̇p2 +ε̇p3 )/γ̇max
p ,
p p p p
which at a regular point is given by (ε̇max + ε̇min )/(ε̇max − ε̇min ) and is
therefore equal to sin φ. The measured dilatancy of most soils (as well as
rocks and concrete) is usually significantly less than this,1 except in the
case of undrained clays in which both internal friction and dilatancy are
negligible. Most such materials, therefore, cannot be modeled as standard.
In a nonstandard model, the flow rule may be taken in the same form as
above, but with the dilatancy angle ψ replacing φ.
1
The dilatancy ratio rarely exceeds 0.1, while the angle of internal friction may be as
high as 45◦ in dense, well-graded soils with angular particles.
The plastic dissipation in a standard Mohr–Coulomb material was shown

by Drucker [1953] to be
Dp (ε̇p ) = c cot φ(ε̇p1 + ε̇p2 + ε̇p3 )
at any point of the yield surface, including the corners.

In plane plastic flow with ε̇3 = 0, as was pointed out in 3.3.4, σ3 is the
intermediate principal stress, even if ψ 6= φ. The criterion therefore takes
the form
|σ1 − σ2 | + (σ1 + σ2 ) sin φ = 2c cos φ.
Mechanisms
A Coulomb mechanism in plane strain is one in which polygonal blocks
of material move rigidly relative to one another. The interfaces between the
blocks may be regarded as very narrow zones in which the strain rates are
very large. In a nondilatant material, only shearing takes place in such a
zone, so that the movement is one of sliding, and the interfaces are just slip
lines, as discussed in Section 5.1. In the presence of dilatancy, dilatation
as well as shearing takes place, and the movement involves separation in
addition to sliding.1 In fact, it can easily be shown (the details are left to
an exercise) that the velocity discontinuity forms an angle equal to ψ with
the interface – that is, if the magnitude of the discontinuity is v then the
sliding speed is v cos ψ and the separation speed is v sin ψ. If the thickness of
the zone is h, then the average longitudinal strain rate perpendicular to the
interface is (v sin ψ)/h and the average shearing rate is (v cos ψ)/h, so that
the principal strain rates are 21 (±1 + sin ψ)v/h. In the standard material,
then, the plastic dissipation per unit area of interface is cv cos φ.
In another type of mechanism, introduced by Petterson and developed
by Fellenius, the velocity discontinuities are along circular arcs, with the
material inside an arc rotating rigidly about the center of the circle and
thus sliding past the remaining material. This mechanism, known as a slip
circle, is clearly appropriate only for a nondilatant (e.g., Tresca) material.
Nevertheless, like the Coulomb mechanism with pure sliding, it is often used
regardless of material properties. Both the location of the center of the
circle and its radius can be chosen so as to minimize the upper-bound load
predicted by the mechanism.
In a standard Mohr–Coulomb material, a velocity discontinuity such that
the mass on one side rotates rigidly while that on the other side remains
stationary takes the form of a logarithmic spiral rather than a circle. As can
be seen from Figure 6.1.7, dr/(r dθ) = tan φ, which can be integrated to give
1
In practice, soil mechanicians often use the Coulomb mechanism with pure sliding
regardless of the yield criterion or flow rule assumed. Also, the mechanism is usually
analyzed by means of statics rather than kinematics. For the Tresca material, the results
are equivalent.
r = r0 exp[tan φ(θ − θ0 )]. Clearly, for φ = 0, the curve becomes the circle
r = r0 .
AA
dr*
φ A

Q
QA
r dθ
QA

r

j dθ
H
H

Z }

Figure 6.1.7. Velocity discontinuity in a standard Mohr–Coulomb material.
In accordance with Radenkovic’s theorems (see 3.5.1), any upper bound

found for a standard Mohr–Coulomb material is also an upper bound for a
nonstandard material with the same yield criterion, while any lower bound
for a standard Mohr–Coulomb material whose angle of internal friction
equals the dilatancy angle of the nonstandard material is also a lower bound
for the latter. The following examples illustrate the procedure. The exam-
ples are limited to soils that can be modeled as homogeneous, a condition
rarely encountered in real soil masses. For an extensive survey of appli-
cations of limit analysis to soil mechanics, including numerical results, see
Chen [1975].
Stability of a Vertical Bank

A vertical bank of height h occupies the half-strip 0 ≤ y ≤ h, x ≥ 0.
We wish to determine the maximum height so that the bank does not col-
lapse. Since the weight of the bank per unit horizontal area is wh, where
w is the specific weight, we may regard h as the equivalent of a load, and
denote its greatest safe value by hU . A lower bound to hU may be found
by assuming, with Drucker and Prager [1952], the stress field σx = τxy = 0,
σy = −w(h − y), which satisfies the equilibrium equations and leaves the
vertical and horizontal surfaces of the bank traction-free. The greatest nu-
merical value of σy , equal to wh, must not exceed the yield stress in uniaxial
compression for the standard Mohr–Coulomb material with internal-friction
angle ψ, namely
2c cos ψ π ψ

0
σC = = 2c tan + .
1 − sin ψ 4 2
If necessary — that is, if the bank is not situated atop a hard stratum —
the stress field may be extended into the half-space y ≤ 0 without violating
the yield criterion as follows:
( )
−w(h − y), x > 0,
σx = αwy, τxy = 0, σy =
wy, x < 0,
where α = (1 − sin ψ)/(1 + sin ψ). The stress field thus contains admissible
discontinuities that separate it into three zones.
For the standard Tresca √ material, the lower bound may be written as
2k/w. An improvement to 2 2k/w is achieved by means of an admissible
stress field consisting of seven zones, proposed by Heyman [1973], who also
discusses incomplete stress fields leading to somewhat higher lower bounds.
A numerical solution by Pastor [1976] yields a lower bound of 3.1k/w.
However, the stress fields producing these improvements include tensile
stresses in some regions, while the Drucker–Prager stress field does not.
Indeed, for a material that cannot take tension it was shown by Drucker
[1953], on the basis of a mechanism including a tension crack , that 2k/w is
an upper bound as well.
r
6
)N
v φ

k
h
-β

k
?
x
Figure 6.1.8. Coulomb mechanism in a vertical bank.
An upper bound for a Mohr–Coulomb material may be found with the

help of the Coulomb mechanism shown in Figure 6.1.8, in which a wedge
of angle β separates from the remainder of the bank. The weight (per unit
thickness perpendicular to the page) of the wedge is W = 12 wh2 tan β, and if
the magnitude of the velocity of the wedge is v, then its downward component
is v cos(β + φ). Equating the external work rate to the internal dissipation,
1 2
wh tan β cos(β + φ)v = chv cos φ sec β, (6.1.4)
2
produces the upper bound

2c cos φ
h= .
w sin β cos(β + φ)
The least upper bound is obtained by maximizing the denominator with
respect to β:
d
[sin β cos(β + φ)] = cos β cos(β + φ) − sin β sin(β + φ) = 0,
dβ
or
π

tan(β + φ) = cot β = tan −β .
2
The equation is obeyed if β = 14 π − 12 φ. But
π φ π φ 1

sin − cos + = (1 − sin φ),
4 2 4 2 2
and therefore hU ≤ 2σC /w (= 4k/w for a Tresca material). We are thus left
with the wide limits,
2c π ψ 4c π φ

tan + ≤ hU ≤ tan + .
w 4 2 w 4 2
The mechanism based on the logarithmic spiral produces a slight improve-
ment over that of Coulomb: the optimal spiral reduces the factor in the
upper bound from 4 to 3.83.
Stability of a Simple Slope
The improvement in the upper bound obtained by using a curved rather
than a planar slip surface is significantly greater for inclined banks, known
as simple slopes. It is conventional in soil mechanics to define the stability
factor 1 Ns = whU /c, a function of the slope α as well as of the internal-
friction angle φ. The optimal Coulomb wedge gives
4 sin α cos φ
Ns = .
1 − cos(α − φ)
It was shown by Taylor [1937] that limit-equilibrium calculations based

on the logarithmic spiral are so close to those based on the slip circle as to be
undistinguishable. Chen [1975] showed further that there is no significant
difference between the results of the limit-equilibrium method and upper-
bound limit analysis. Some examples are shown in Table 6.1.1.
Very little work has to date been done on lower bounds in the limit
analysis of slope stability.
1
Some authorities, including D. W. Taylor [1948], use the stability number m = 1/Ns
instead.
Table 6.1.1. Stability Factors of Homogeneous Simple Slopes

Failure Surface
α φ Plane Curved
90◦ 0◦ 4.00 3.83
25◦ 6.28 6.03
75◦ 0◦ 5.21 4.56

25◦ 9.80 6.03
60◦ 0◦ 6.93 5.24

25◦ 17.36 12.74
Thrust on Retaining Walls

If a vertical bank of soil is too high for stability — and in a cohesionless
soil, any height is too great — then it must be held back by a retaining
wall. The soil and the wall exert on each other a mutual thrust, equal to
the resultant of the horizontal stress σx in the soil. Failure of the soil may
occur when it is in one of two states, active and passive.
In a passive failure, the wall moves into the soil, increasing the horizontal
pressure until the yield criterion is reached in the soil. The thrust P is thus
an increasing load on the soil mass, doing positive work, and its limiting
value, known as the passive thrust and denoted Pp , is an ultimate load in
the usual sense.
In an active failure, the wall is pushed outward as a result of the horizon-
tal pressure, and this pressure is reduced until the soil yields. In the process
the thrust decreases and does negative work. If the limiting thrust (the
active thrust) is denoted Pa , then the upper and lower bounds are bounds
on −Pa , and the usual nomenclature of limit analysis must be reversed:
the upper-bound theorem gives a lower bound to Pa , and the lower-bound
theorem gives an upper bound.
The static analysis is due to Rankine. The geometry is taken as the same
as in the preceding problem, Figure 6.1.8, and the wall is assumed smooth.
Equilibrium is satisfied if σy = −w(h − y), τxy = 0, and σx depends on y
only. The Mohr–Coulomb criterion is assumed to be met everywhere, that
is,
|σx − σy | + sin φ (σx + σy ) = 2c cos φ.
Two solutions exist for σx :
1 − sin φ cos φ
σx = σy − 2c
1 + sin
φ 1 + sin φ
2 π φ π φ

= σy tan − − 2c tan − ,
4 2 4 2
representing active failure, and
1 + sin φ cos φ
σx = σy + 2c
1 − sin φ 1 − sin φ
π φ π φ

2
= σy tan + + 2c tan + ,
4 2 4 2
representing passive failure. Introducing σy = −w(h − y) and integrating

over 0 ≤≤ h gives the limiting thrusts,
1 π φ π φ

Pa = wh2 tan2 − − 2ch tan −
2 4 2 4 2
and
1 π φ π φ

Pp = wh2 tan2 + + 2ch tan + .
2 4 2 4 2
The preceding formulas, known as Rankine’s formulas, are widely used
in soil mechanics. In view of Radenkovic’s second theorem, however, it must
be recognized that they are not true lower bounds (in the usual sense for
Pp , in the reverse sense for Pa ) unless the friction angle φ is replaced by the
dilatancy angle ψ. Upper bounds can be obtained by means of the Coulomb
mechanism, following Coulomb’s own analysis of 1776, but making sure that
a kinematic approach with an associated flow rule is taken; Coulomb assumed
pure sliding, and analyzed the wedge statically.
For active failure, the mechanism is the same as for the free-standing
vertical bank. With the wall again assumed to be smooth, the external
rate of work is given by the left-hand side of (6.1.4) with the additional
term −P v sin(β + φ), and the internal dissipation equals the right-hand side.
Consequently,
1
P = wh2 tan β cot(β + φ) − ch cos φ sec β csc(β + φ).
2
Both terms on the right-hand side can be shown to be stationary at β =
1
4
π − 12 φ, so that the largest P (corresponding to the smallest upper bound
on −P ) is given by
1 π φ π φ

Pa = wh2 tan2 − − ch cos φ sec 2
− .
2 4 2 4 2
But
π π φ π φ π φ

cos φ = sin − φ = sin 2 − = 2 sin − cos − ,
2 4 2 4 2 4 2
so that
π φ π φ

2
cos φ sec − = 2 tan − ,
4 2 4 2
and the Rankine formula for the active thrust is recovered.

An analogous result is obtained for the passive thrust. Here the wedge
moves upward and to the right, the velocity forming an angle β − φ with
the vertical (details are left to an exercise). We thus see that the Rankine
formulas give the correct limiting thrusts on a smooth wall for the standard
Mohr–Coulomb material. For the nonstandard material, they furnish the
Radenkovic bounds.
The kinematic approach may be extended to obtain upper bounds in
the presence of friction between the soil and the wall, by adding the term
µP |vy | to the internal dissipation, µ being an average coefficient of friction
and vy the vertical component of velocity; it is assumed that the wall moves
horizontally. While no analytical solutions exist, the wedge angle β giving
the lowest upper bound can easily be found numerically for given µ, φ, and
wh/c.

1. Find the best value of the wedge angle γ for the stress field of Figure
6.1.1(a).
2. Assume that the velocity discontinuity in Figure 6.1.1(b) is inclined at

an arbitrary angle α. Find the upper bound to the cutout factor, and
show that α = 45◦ gives the best upper bound.
3. Show that for a rectangular slab of sides 2a and 2b (b > a) with a slit
of length 2βa parallel to the shorter side, the cutout factor for simple
tension perpendicular to the slit is still 1 − β.
4. Find lower and upper bounds on the cutout factor for the slab in
Exercise 1 when b < a.
5. Find the value of the load F corresponding to the slip-line fields of

Figure 6.1.4(a) for θ ≥ 75◦ .
6. Determine the velocity fields corresponding to the mechanisms of Fig-

ure 6.1.5(b) and (c) and the corresponding upper bounds to the col-
lapse load F .
7. Derive Equation (6.1.2) for beams with h/L ≤ πτY /σY .
8. Derive Equation (6.1.3) for beams with h/L ≥ πτY /σY .
9. Find an upper bound to the critical height h of the bank of Figure

6.1.8 when the straight velocity-discontinuity line is replaced by a log-
arithmic spiral (Figure 6.1.7).
Section 6.2 / Beams Under Combined Stresses 355
10. Derive the equations governing the upper bound to the passive thrust
and the lower bound to the active thrust on a retaining wall in the
presence of friction between the soil and the wall.
11. Assuming plane strain, use a stress field like that of Figure 5.1.5(e) to
find a lower bound to the ultimate tensile force carried by the symmet-
rically notched tension specimen of Exercise 9, Section 5.1. Compare
with the result of that exercise.
Section 6.2 Beams Under Combined Stresses

6.2.1. Generalized Stress
Introduction
A concept of great usefulness in the limit analysis of beams, arches,
frames, plates and shells was introduced by Prager [1955b, 1956b, 1959]. It
is that of generalized stress and strain.
Consider the ideal sandwich beam as shown in Figure 3.5.1(a) (page
155), but subject to distributed loading so that the axial force P and bending
moment M vary along its length. While M and P can no longer be regarded
as generalized loads, they may be regarded as generalized stresses in the
following sense. At any section of the beam, the stresses in the flanges are
σ = P/2A ± M/Ah. The local values of the elongation and rotation obey
the relations
d∆ dθ
= ε, = κ,
dx dx
where ε is the mean longitudinal strain and κ is the curvature. The strains
in the flanges are ε ± hκ/2, and the internal virtual work can easily be shown
to be Z L
δW int = (P δε + M δκ)dx,
0
the span of the beam being 0 < x < L. The local axial force P and bending
moment M may now be regarded as the generalized stresses, with ε and κ,
respectively, as their conjugate generalized strains.
Figure 3.5.1(a), in addition to representing the limit-load locus for the
beam under external axial force and moment, thus also represents the yield
locus for the ideal sandwich beam. Such yield loci are also called interaction
diagrams; an example was already studied in 4.4.1 (see Figure 4.4.5(a), page
224).
The ideal sandwich beam is statically determinate in the sense of 4.1.1,
since it has no range of contained plastic deformation: if the material is
perfectly plastic, then the beam can undergo unlimited deformation as soon
as either flange yields. In any real beam, as we already know, the ultimate
moment MU is greater than the elastic-limit moment ME , and therefore
two distinct yield loci exist: the elastic-limit (or initial yield) locus and the
ultimate yield locus [again, see Figure 4.4.5(a)]. Under the hypothesis of
rigid–plastic behavior, however, only the ultimate yield locus is relevant.
Generalized Stress and Strain: Definitions

Generalized stresses may coincide with the actual stresses, or they may
be local stress resultants integrated over one or (as in the present example)
two dimensions, or even over a whole finite element of the body (such as a
bar in a truss). If the generalized stresses are denoted Qj (j = 1, ..., n),
then the conjugate generalized strains qj are in general defined by
n
Z X
δW int = Qj δqj dΩ,
Ω j=1
R
where Ω (·) dΩ describes integration over the entire body with respect to vol-
ume, area, or length, as appropriate, or summation over all finite elements.1
Let Q and q denote the generalized stress and strain vectors, respec-
tively. As illustrated by the ideal sandwich beam, a yield locus in terms of
generalized stresses, say Φ(Q) = 0, may be derived in exactly the same way
as the limit locus in terms of generalized loads was derived in 3.5.1. For
rigid–plastic materials, the generalized plastic dissipation is thus
D̄p = Q · q̇, (6.2.1)
and the principle of maximum plastic dissipation may be written as
(Q − Q∗ ) · q̇ ≥ 0
or
D̄p (q̇) ≥ Q∗ · q̇ (6.2.2)
for any Q∗ such that Φ(Q∗ ) ≤ 0.

Finally, the theorems of limit analysis may be restated as follows.
Lower-Bound Theorem. A load point P is on or inside the limit locus if
a generalized stress field Q∗ can be found that is in equilibrium with P and
obeys Φ(Q∗ ) ≤ 0 everywhere.
Upper-Bound Theorem. A load point P is on or outside the limit locus
if a kinematically admissible velocity field, yielding the generalized velocity
1
In technical mathematical language, dΩ is a measure in a space of three, two, one or
zero dimensions.
vector ṗ∗ conjugate to P and the generalized strain-rate field q̇∗ , can be
found so that Z
P · ṗ∗ = D̄p (q̇∗ ) dΩ. (6.2.3)
Ω
Elastic and Plastic Generalized Strain

When it is desired to describe elastic–plastic behavior in terms of gen-
eralized stress and strain, then it is necessary to decompose the generalized
strain into elastic and plastic parts:
q = qe + qp .
But with the exception of some simple cases, there is in general no one-to-
one correspondence between qe and εe or between qp and εp . Consider, for
example, a real (as distinct from ideal) beam subject to symmetric bending
only; the moment M is the only generalized stress, and the curvature κ is
the only generalized strain. The actual strain at a point is given by
ε = −κy.
In the elastic range, the moment-curvature relation is M = EIκ, and

therefore the elastic part of the curvature is
M
κe = .
EI
The plastic strain is
σ
εp = ε − εe = −(κe + κp )y −
E
1 My

p
= −κ y − σ+ .
E I
Thus, while
1
Z
κp = − yεp dA,
I A
there is no inverse relation by which κp determines εp . The quantity σ +
M y/I does not vanish in the range of contained plastic deformation, and
neither does its time derivative. Consequently,
Z
σ ε̇p dA 6= M κ̇p ,
A
except when σ̇ = 0, a condition that implies that Ṁ = 0 and hence κ̇e = 0,

and therefore holds only on the ultimate yield locus.
It follows that the principle of maximum plastic dissipation in terms of
generalized stress and generalized plastic strain is in general valid only under
unrestricted plastic flow, with q̇p = q̇. The exceptional cases are those in
which no contained plastic deformation occurs locally, as at a section of an
ideal beam, or in a bar carrying axial force only (truss member).
6.2.2. Extension and Bending

Introduction
The theory of symmetric pure bending of elastic–plastic beams, discussed
in 4.4.1, can easily be extended to beams subject to an axial force P in
addition to the bending moment M , provided the deflection is so small that
the additional bending moment resulting from the axial force acting over
the deflection (the so-called P-∆ effect) is negligible. With this proviso
satisfied, and with the uniaxial stress-strain relation given by σ = f (ε), the
first of Equations (4.4.7) needs only to have its right-hand side changed from
0 to P .
As seen in 4.4.1, the problem of asymmetric bending is difficult even
in the absence of axial force. An analysis of rectangular beams subject to
axial force and bending moments about both axes was carried out by Shakir-
Khalil and Tadros [1973]. In the present subsection, only symmetric bending
is considered.
Consider a beam whose centroidal fiber coincides with the z-axis, whose
cross-section is symmetric about the y-axis, and which is made of an elastic–
perfectly plastic material. In bending in the yz-plane, the strain is given as
by Equation (4.4.2) with α = 0 (but with the axes renamed): ε = −κ(y−y0 ),
so that κy0 = ε0 is the strain of the centroidal fibers or, equivalently, the
average strain over the cross-section.
In the elastic range, this strain is elastically related to the average stress
P/A, that is, κy0 = P/AE. Since κ = M/EI regardless of the value of P , it
follows that the coordinate y0 of the neutral axis is given by y0 = P I/M A,
and the stress distribution is
P My
σ= − .
A I
Let the y-coordinates of of the extreme bottom and top fibers be y = −c1 and
y = c2 , respectively. The magnitudes of the extreme values of the stresses
are
P M c1 P M c2
σ1 = + , σ 2 = − ,
A I A I
and the elastic limit corresponds to max(σ1 , σ2 ) = σY , that is,

P M M P

max
+ , − = 1,
PU ME1 ME2 PU
where
I I
PU = σY A, ME1 = σY , ME2 = σY .
c1 c2
If the section is doubly symmetric, then c1 = c2 = h/2 and ME1 =
ME2 = ME = 2σY I/h. The elastic-limit locus is then simply |P/PU | +
M/ME1
1

ME2 /ME1 @
@
@

@
@

@

–1 @
@ P/PU
@ 1

@
@

@
@

@
–ME2 /ME1
@
@
–1
Figure 6.2.1. Elastic-limit locus for an asymmetric beam under combined bend-
ing and extension.
|M/ME | = 1, that is, it has exactly the same form as the yield locus of
the ideal sandwich beam [Figure 3.5.1(b)] but with the elastic-limit moment
ME replacing the ultimate moment MU .1 For a section without double
symmetry, the elastic-limit locus is as shown in Figure 6.2.1 if c2 > c1 (so
that ME1 > ME2 ).
Note that for a certain ratio of P to M , namely P/M = (c2 − c1 )A/2I,
yielding occurs simultaneously at the top and bottom fibers, and two plastic
zones form as the generalized stress (M, P ) moves outside the elastic-limit
locus. Otherwise only one plastic zone forms at first. As the point (M, P )
moves farther from the origin, the stress at the extreme fiber opposite the
plastic zone increases until it, too, reaches the yield-stress value, creating a
second plastic zone. Further loading results in the shrinking of the elastic
core, exactly as in pure bending, until it becomes negligibly thin, indicating
that the ultimate yield locus has been reached.
Yield Locus for Symmetric Bending

In order to determine the ultimate yield locus, let us assume for the sake
of definiteness that M > 0, so that the bottom fibers are in tension, with
σ = σY , and the top fibers are in compression, with σ = −σY . If y0 is the
1
For the ideal sandwich beam, of course, ME = MU .
y-coordinate of the vanishing elastic core, then

Z y=y0 Z y=c2 Z y=c2
M = −σY y dA − y dA = 2σY y dA (6.2.4)
y=−c1 y=y0 y=y0
and Z y=y0 Z y=c2

P = σY dA − dA (6.2.5)
y=−c1 y=y0
R y=c2
since y=−c 1
y dA = 0. For a given cross-section, the integrals in Equations
(6.2.4)–(6.2.5) can be evaluated in terms of y0 , so that these equations furnish
a parametric representation of the ultimate yield locus in the M P -plane in
terms of the parameter y0 , the range of y0 being −c1 ≤ y0 ≤ c2 .
Associated Flow Rule

The principle of maximum plastic dissipation holds in states of uniaxial
stress regardless of the yield criterion and flow rule obeyed by the material
under multiaxial stress. Consequently the generalized strain rate (ε̇0 , κ̇)
must be perpendicular to the yield locus in the M P -plane, that is,
ε̇0 dP + θ̇ dM = 0.
Now consider Equations (6.2.4)–(6.2.5). If b(y) denotes the width of the

beam at the level y, then dA = b(y)dy, and therefore
dM = −2σY b(y0 )y0 dy0 , dP = 2σY b(y0 )dy0 ,
so that normality is equivalent to ε̇0 − y0 κ̇ = 0 — precisely the definition of

ε̇0 .
Examples of Yield Loci

In order to compare the behavior of different cross-sections, it is conve-
nient to describe the yield locus in terms of the dimensionless generalized
stresses m = M/MU and p = P/PU , where PU = σY A (A being the total
cross-sectional area) and MU is given by Equation (4.4.11). The parameter
can also be made dimensionless by defining, for example, η = 2y0 /h, where
h = c1 + c2 is the depth of the beam; the range of η is thus −η1 ≤ η ≤ η2 ,
where ηi = 2ci /h, i = 1, 2. Equations (6.2.4)–(6.2.5) can therefore be writ-
ten symbolically as
m = m̄(η), p = p̄(η). (6.2.6)
As an example, consider a rectangular beam of width b and depth h. It

is easy to see that Equations (6.2.4)–(6.2.5) become
!
h2
M = σY b − y02
4
and
P = 2σY by0
In dimensionless form,
m = 1 − η2, p = η,
so that in this case the yield locus can be described in explicit form,
m = 1 − p2 ,
and forms a parabola.

For a circular bar of radius a, it is more convenient to define η by η =
−1
sin (y0 /a), so that its range is − 12 π ≤ η ≤ 12 π. The yield locus is given by
1
p= (2η + sin 2η), m = cos3 η.
π
If M is negative, it need only be replaced by −M (and m by −m) in

all the results for doubly symmetric sections. Without double symmetry, P
must also be replaced by −P (and p by −p).
The yield loci for the rectangular and circular bars are shown as curves 1
and 2 in Figure 6.2.2. It is seen that they differ only slightly over the entire
range. The yield locus for an I-beam, on the other hand, can be expected
to lie between the loci for the rectangular beam and for the ideal sandwich
beam, and closer to the latter. Figure 6.2.2 also shows, as curve 3, the locus
for the I-beam shown in the adjacent picture.
In practice, curved yield loci such as the one for the rectangular bar
are often replaced, following Onat and Prager [1954], by piecewise linear
approximations such as the one shown with dashed lines in Figure 6.2.2.1
Such approximations may also be regarded as the exact yield loci for certain
idealized sections, which are in turn approximations to the true sections.
The advantage of this point of view (see Hodge [1959], Section 7-3) is that
a velocity field that gives generalized strain rates associated with the ap-
proximate yield locus may be easily visualized in the context of the idealized
section. A frequently used yield locus for wide-flange steel sections is given
by
0.85|m| + |p| = 1, |p| ≥ 0.15,
(6.2.7)
|m| = 1, |p| ≤ 0.15.
Application: Collapse of a Semicircular Arch

As an illustration of the use of the yield locus in terms of moment and ax-
ial force, we consider the pinned-ended semicircular arch of radius a, loaded
by a concentrated vertical force 2F at midspan, shown in Figure 6.2.3(a).
1
A general theory of plasticity with piecewise linear yield loci is due to Hodge [1957a].
m
1.0 H ?
H 6
H
H 16
H
H
H
0.8 H - 1 20
H
J
J
0.6 J1 ?
10 -
1 J
31J
2

)
J
0.4
J
J
J
0.2 J
J
J
0
J p
0.2 0.4 0.8 0.8 1.0
Figure 6.2.2. Yield loci under combined bending and extension for rectangular
beams (curve 1), circular beams (curve 2), and the I-beam shown
(curve 3). Dashed lines represent a piecewise linear approxima-
tion for rectangular beams.
The vertical reactions at the supports are each equal to F by symmetry, but
the horizontal reactions ±H are unknown. Let χ denote the angle between
the resultant reaction and the vertical, so that tan χ = H/F . A free-body
diagram of a segment of the arch is shown in Figure 6.2.3(b), and equilibrium
shows that the axial force and moment are
P (φ) = −F sin φ − H cos φ = −F (sin φ + tan χ sin φ)
F
=− sin(χ + φ),
cos χ
M (φ) = F a(1 − sin φ) − Ha cos φ
Fa
= [cos χ − sin(χ + φ)].
cos χ
With the moment and axial force varying along the arch, it must be
assumed that the yield criterion (whichever is chosen) is met only at certain
critical sections. As in the case of transversely loaded beams discussed in
4.4.2, plastic hinges form at those sections. If a single hinge were to form in
the arch, it would necessarily, because of symmetry, be at midspan. A three-
hinged arch, however, is a stable structure. Consequently, plastic collapse
of the arch requires the formation of two additional plastic hinges, located
symmetrically about the midpoint of the arch. The collapse mechanism is

shown by the dashed curves in Figure 6.2.3(a); the arch segments between
the hinges move as rigid bodies.
2F
b
? P (φ)
S - M (φ)
o
b b

7 S

φ
a w

H
e q
e }
Z
Z -χ 6
4 4 R Z/ P
(a) (b)
Z
Figure 6.2.3. Pin-ended circular arch: (a) initial geometry and loading (solid
line) and collapse mechanism (dashed line); (b) free-body dia-
gram of a segment.
The location of the off-center hinges can easily be determined by noting

that both moment and axial force have local maxima only at φ = π/2 − χ,
and therefore any convex combination of them also has a local maximum
there. Collapse is thus determined by the requirement that the yield criterion
Φ(M, P ) = 0 is met at φ = 0 and φ = π/2 − χ. Assuming a section with
double symmetry so that Φ(M, P ) = Φ(|M |, |P |), we can find χ and the
value of F at collapse by solving simultaneously
Φ(F tan χ, F a|1 − tan χ|) = 0 and Φ(F sec χ, F a(sec χ − 1)) = 0.
The choice of a nonlinear yield criterion necessitates the simultaneous

solution of two nonlinear equations, a task that may be unpleasant. With
a piecewise linear yield criterion, on the other hand, once the side of the
polygon is chosen that is appropriate for each equation, the equations are
linear in F . Eliminating F , a quadratic equation in cos χ or sin χ is obtained.
For simplicity, the square yield locus corresponding to the ideal sand-
wich section will be used. With the help of the dimensionless quantity
η = MU /PU a, the equations to be solved are
[|1 − tan χ| + η tan χ]F a = MU ,
[(1 + η) sec χ − 1]F a = MU .

Consequently χ satisfies
|1 − tan χ| + η tan χ = (1 + η) sec χ − 1.
It can immediately be seen that if tan χ > 1, then tan χ = secχ, or

χ = π/2, so that the additional plastic hinges coincide with the central
hinge, and no collapse is possible. It is necessary, then, that tan χ < 1. For
.
η negligibly small, tan χ is very nearly 3/4 and therefore FU = 4MU /a —
a result equivalent to neglecting the influence of axial forces. For η small
but not negligible, a first-order approximation in η may be effected, leading
. .
to tan θ = 3/4 − 5η/16 and FU = (4MU /a)/(1 + 4.25η). Other examples of
arch collapse based on piecewise linear yield loci were studied by Onat and
Prager [1953, 1954]; see also Hodge [1959], Section 7-4.
6.2.3. Combined Extension, Bending and Torsion

In a bar under a combination of axial force, bending moment, and torque,
the nonvanishing stress components are σz , τxz , and τyz , with τxz = ∂φ/∂y
and τyz = −∂φ/∂x, φ being the stress function. Any point of the bar is
in a state of plane stress in the plane that "is perpendicular
# to ∇φ; the
σ τ
state of stress in that plane can be given by , where σ = σz and
τ 0
q
τ = |∇φ| = τxz 2 + τ 2 . Both the Mises and the Tresca yield criteria are
yz
given by Equation (3.3.5):
2 2
σ τ

+ = 1.
σY τY
Since σij ε̇ij = σ ε̇ + τ γ̇, the associated flow rule may be obtained from the
maximum-plastic-dissipation principle as
σ τ
ε̇p = λ̇ , γ̇ p = λ̇ , (6.2.8)
σY2 τY2
and the plastic dissipation is
q
Dp (ε̇) = (σY ε̇)2 + (τY γ̇)2 . (6.2.9)
Lower Bound
A lower-bound yield locus for combined extension, bending, and torsion
can be found by generalizing an approach proposed by Hill and Siebel [1951]
for combined bending and torsion only. The approach is based on assuming,
on the one hand, the same distribution of normal stress as in 6.2.2, but with
|σ| = ασY , where 0 ≤ α ≤ 1; and on the other hand, √ the same distribution of
shear stress as in fully plastic torsion, but with τ = 1 − α2 τY . The second
assumption leads immediately to
def T p
t= = 1 − α2 .
TU
The first assumption means that the m-p relation would be given by (6.2.6)
if m and p were defined as M/αMU and P/αPU , respectively. With the
standard definitions, the relation is therefore

p m
= p̄(η), = m̄(η).
α α
Eliminating α, we can describe the yield surface in mpt-space in terms of
the single parameter η:
p p
p= 1 − t2 p̄(η), m= 1 − t2 m̄(η).
For the rectangular beam, we have the explicit description

p p2
m= 1 − t2 − √ .
1 − t2
The projections of this surface on both the mt- and pt-planes are unit circles.
Upper Bound
An upper-bound yield curve for combined extension, bending and torsion
can be found following the method of Hill and Siebel [1953]. The assumed
velocity field is taken so that it results only in an extension rate ε̇0 of the
centroidal fiber, a pure curvature rate κ̇ about the centroidal axis, and a
rate of twist θ̇ about the centroid; warping is neglected.1 With symmetric
bending in the yz-plane assumed, the strain rates are
ε̇z = ε̇0 − κ̇y, γ̇θz = rθ̇,

p
where r = x2 + y 2 . The plastic dissipation is therefore
q
Dp (ε̇∗ ) = τY |θ̇| r2 + (αy − β)2 ,
where α = (σY /τY )(κ̇/θ̇) and β = (σY /τY )(ε̇0 /θ̇).

A simple way of finding the upper-bound generalized stresses correspond-
ing to the assumed velocity field is to determine the stresses that are related
to it by the associated flow rule and that obey the yield criterion — but
that do not, in general, form a statically admissible stress field — and then
to calculate their resultants. The associated flow rule (6.2.8) produces the
stresses
σ2
σ = Y (ε̇0 − κ̇y) = −µσY (αy − β)
λ̇
and
τ = µτY r,
1
A correction for warping, resulting in better upper bounds for noncircular sections, is
due to Gaydon and Nuttall [1957].
where τ = τzθ (so that τrz ≡ 0),1 and µ = τY θ̇/λ̇ is a function of position
that can be determined by requiring that the yield criterion (3.3.5) be obeyed
everywhere, resulting in
1
µ= p .
r2 + (αy − β)2
The stress resultants M , P and T are therefore given by
y2
Z Z
M =− yσ dA = ασY p dA,
A A r2 + (αy − β)2
1
Z Z
P = σ dA = βσY p dA, (6.2.10)
A A r2 + (αy − β)2
r2
Z Z
T = rτ dA = τY p dA.
A A r2 + (αy − β)2
The integrations in Equations (6.2.10) may be performed, numerically if
necessary, to yield a parametric representation of the upper-bound yield
surface in mpt-space in terms of the parameters α and β. Computed yield
curves representing the projections of the lower-bound and upper-bound
yield surfaces in the mt- and pt-planes, are shown in Figure 6.2.4(a) and (b)
for a circular and a square bar, respectively.
Extension and Torsion of a Circular Bar

For a circular bar, the upper-bound solution presented above is, in fact,
a complete solution, since an axisymmetric shear-stress distribution τθz =
τ (r), τrz = 0 is statically admissible. A closed-form result can be obtained
for extension and torsion alone, that is, for α = 0.
We define the dimensionless parameter ζ = β/a, where a is the radius
of the bar. Equations (6.2.10)2,3 give
2π
Z a r dr
Z a dx
p= ζaσY =ζ
σY πa2
p p
0 ζ 2 a2 + r2 0 x + ζ2
and
2π a r3 dr 3 1 x dx
Z Z
t= τY = ,
τ πa3
p p
2
0 2 2
ζ a +r 2 2 0 x + ζ2
3 Y
or q q
p = 2ζ ζ2 +1−ζ , t = 2ζ 3 − (2ζ 2 − 1) ζ 2 + 1. (6.2.11)
The first of Equations (6.2.11) can be rewritten as

q
p + 2ζ 2 = 2ζ ζ 2 + 1.
1
Thus the traction boundary conditions are not satisfied for any but a circular bar.
m,p
6
Upper bound (exact) Upper bound
1.0
@
p-t R
@ m-t
0.8
@
I
@
Lower bound Lower bound
0.6
0.4
0.2
t 0 -t
1.0 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1.0
(a)
m,p
6
Upper bound
1.0
A
p-t U
A m-t
0.8
@
I
@
Lower bound
0.6
0.4
0.2
t 0 -t
1.2 1.0 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1.0 1.2
(b)
Figure 6.2.4. Lower-bound and upper-bound yield loci under bending, torsion
and extension: (a) circular bar; (b) square bar.
Squaring both sides, we find
4ζ 4 + 4ζ 2 = 4ζ 4 + 4ζ 2 p + p2 ,
or 4ζ 2 (1 − p) = p2 , which can immediately be solved for ζ:

p
ζ= √ .
2 1−p
Substituting in the equation for t, we obtain the explicit relation
1 p
t = (2 + p) 1 − p.
2
Squaring both sides of this equation, we may rewrite it as
p2 − p3
t2 + p 2 = 1 + ,
4
a form that is convenient for determining the extent to which the present
upper-bound yield curve differs from the previously found lower bound, de-
scribed by the unit circle. The right-hand side of the last equation has its
maximum at p = 2/3, its value there being 28/27. Consequently the distance
of the
ppoints on the dimensionless yield curve from the origin lies between 1
and 28/27 = 1.018, and the curve differs only slightly from the circle.
The stress distribution giving rise to the resultants just obtained can also
be determined as the solution of a constrained extremum problem: find the
stresses satisfying (3.3.5) everywhere such that the torque,
Z
T = rτ dA,
A
is maximum for a given axial force,

Z
P = σ dA.
A
Substituting for τ by solving (3.3.5), the problem may be written as

Z q Z
τY δ r 1 − (σ/σY )2 dA +νδ σ dA = 0,
A A
where ν is a Lagrangian multiplier, or

Z " #
(σ/σY )r
β−p δσ dA = 0,
A 1 − (σ/σY )2
where β = νσY /τY . There being no further constraint on the distribution

of σ, the quantity in brackets under the integral sign must be zero, that is,
σ β
=p 2 ,
σY β + r2
and the yield criterion is satisfied everywhere if and only if
τ r
=p 2 .
τY β + r2
Integration leads immediately to p and t given by Equations (6.2.11), with

ζ = β/a as before.
Bending and Torsion of a Bar of Arbitrary Cross-Section

A complete solution for combined bending and torsion of a fully plastic
bar of arbitrary cross-section may be obtained by integrating numerically
a nonlinear partial differential equation first derived by Handelman [1944].
The stress function φ(x, y) is now regarded as unknown, and it is to be found
so that it maximizes the bending moment for a given torque (or vice versa).
For doubly symmetric bending in the yz-plane, we have
σY
Z q
M= |y| τY2 − (∂φ/∂x)2 − (∂φ/∂y)2 dA.
τY A
The constrained extremum problem may therefore be written as

Z q
|y| δ τY2 − (∂φ/∂x)2 − (∂φ/∂y)2 − ν δφ dA = 0,
A
where ν is again a Lagrangian multiplier. Since φ = 0 and hence δφ = 0 on

the boundary, integration by parts leads to
  
∂  y∂φ/∂x
Z 
q 
A  ∂x 2 2
τY − (∂φ/∂x) − (∂φ/∂y) 2
  
∂  y∂φ/∂y 
+ q  + ν δφ dA = 0.
∂y 2 2 2
τY − (∂φ/∂x) − (∂φ/∂y) 
Since φ is unconstrained in the interior of A, the contents of the curly brack-

ets must vanish everywhere, and this furnishes the required partial differen-
tial equation for φ, an elliptic equation subject to the boundary condition
φ = 0. Different values of ν give different ratios of M to T . The equation was
solved numerically for a square cross-section by Steele [1954], and for other
cross-sections by Imegwu [1960], who showed that the interaction diagram
in the mt-plane is remarkably insensitive to cross-sectional shape.
6.2.4. Bending and Shear

In the problems involving combinations of bending moments, axial force and
torque that have been considered so far in this section, these stress resul-
tants are independent of one another as far as the equilibrium of the beam
is concerned, and therefore their interaction may be rigorously studied on a
purely local basis. In particular, for the purpose of lower-bound analysis, if a
statically admissible stress distribution is found for each such resultant sepa-
rately, then a linear combination of such stress distributions is also statically
admissible.
If a beam is subject to transverse loading, then this loading determines
both the shear force V and the bending moment M , which are therefore
related to each other by the equilibrium equation
dM
= V.
dz
Strictly speaking, then, M and V are not generalized stresses that can be
specified independently of each other.
Indeed, each case of a transversely loaded beam presents a distinct prob-
lem of limit analysis. As discussed in 6.1.2, even such closely related cases as
the end-loaded cantilever and the center-loaded simply supported beam are
different. If the span-to-depth ratio is sufficiently high, however, the limit
loads tend to those predicted by the elementary plastic-hinge mechanism —

that is, the overall collapse of the beam is almost completely determined by
local behavior at a critical section. It thus becomes reasonable to formulate
an approximate local yield locus, in terms of M and V , that governs local
behavior, analogous to the local yield loci found above. The problem of limit
analysis of an arbitrarily loaded beam then reduces to the determination of
the critical section or sections; this problem is studied in 6.3.2.
The particular problem of the end-loaded cantilever may be used as the
test problem for studying the local yield locus, as discussed by Drucker
[1956]. Here, the problem has already been investigated from two points of
view: in 4.5.2 as the limit of elastic–plastic bending, and in 6.1.2 as a problem
in plane limit analysis. The latter approach, as was indicated there, gives
both upper and lower bounds for the collapse load of the beam. The former
approach, as is shown next, furnishes a lower bound.
Lower Bound as Limit of Elastic-Plastic Solution
We saw in 4.5.2 that as the elastic core shrinks under the influence of
an increasing moment, the maximum shear stress there grows until it also
reaches the yield value, after which a secondary plastic zone forms in which
|τ | = τY , σ = 0. Between this central plastic zone and each of the two outer
plastic zones is an elastic zone in which, going from inside to outside, the
normal stress increases linearly in magnitude from 0 to σY , and the shear
stress decreases (parabolically in a rectangular beam or in the web of an
I-beam) from τY to zero. The limiting state is, of course, one in which the
elastic zones shrink to vanishing thickness. This limit does not represent,
strictly speaking, a statically admissible stress field, since it includes a dis-
continuity in τ . Since, however, the discontinuity occurs at two isolated
points, equilibrium of any finite element is not violated and the stress distri-
bution is acceptable. If the width of the plastic zone is c, then V = τY bc and
M = σY b(h2 − c2 )/4. Defining v = V /VU , where VU = τY A, we immediately
obtain the dimensionless interaction curve given by
m = 1 − v2.
This curve has clearly the same form as the one previously found for the
interaction between moment and axial force, with v replacing p. In fact, it
can be shown that for any doubly symmetric section the yield locus in the
mv-plane is given parametrically by
m = m̄(η), v = p̄(η),
where m̄ and p̄ are the same functions as in Equation (6.2.6). The proof is
based on the observation that the stress distribution for combined bending
moment and axial force, shown in Figure 6.2.5(a), is the superposition of
those shown in Figure 6.2.5(b) and (c). However, the stress distribution
(b) alone produces the same moment as (a), and the stress distribution (c)
alone produces the same axial force as (a). If, now, (c) represents a block
of shear stress rather than normal stress, then it produces a shear force,
rather than an axial force, by exactly the same formula. When suitably
nondimensionalized, therefore, the shear force is exactly the same function
of the depth of the central zone as is the axial force.

- -
- -
-
- = + -
-
- -
- -
-
- -
- -
- -
(a) (b) (c)
Figure 6.2.5. Fully plastic stress distribution at a beam cross-section under

combined bending moment and axial force.
In a rectangular cantilever of length L carrying a concentrated transverse

force F at its tip, collapse occurs when a plastic hinge forms at the built-in
end. At this point M = F L and V = F . Defining f = F L/MU , we find that
the lower bound based on Figure 6.2.5(b-c) predicts
2
σY h

f =1− f 2,
2τY 2L
a quadratic equation for f that can be solved explicitly for f . Assuming
the Tresca criterion (σY = 2τY ) and defining δ = h/2L, we can write the
solution as
1 p
f= 2 1 + 4δ 2 − 1 . (6.2.12)
2δ
Upper Bound: Hodge Approach
The generalized strain rates conjugate to M and V are the curvature rate
κ̇ and the shear rate γ̇, respectively. The strain-rate components conjugate
to the stresses σ and τ are ε̇ = −κ̇y and γ̇. Given κ̇ and γ̇, the associated
flow rule (6.2.8) produces the stresses
σY2 τY2
σ=− κ̇y, τ= γ̇.
λ̇ λ̇
Satisfaction of the yield criterion (3.3.5) requires
q q
λ̇ = (τY γ̇)2 + (σy κ̇y)2 = τY γ̇ 1 + νy 2 ,
where ν = σY κ̇/τY γ̇. The stresses are therefore

νσY y τY
σ = −p , τ=p . (6.2.13)
1 + ν 2y2 1 + ν 2y2
The stress resultants are accordingly given in terms of the parameter ν as
y2
Z
M = νσY p dA,
A 1 + ν 2y2
1
Z
V = τY p dA.
A 1 + ν 2y2
b -
?
6 a
6
h - c
Figure 6.2.6. I-beam: geometry.
Hodge [1957b] evaluated the integrals in the preceding equations for the
I-beam shown in Figure 6.2.6. If c/b and a/h are small, then the parametric
form of the yield locus in dimensionless form is, after neglecting terms of
order higher than the first in these quantities,
2 tanh ω + j(coth ω − ωcsch 2 ω)

m= ,
2+j
sech ω + jωcsch ω
v= ,
1+j
where ω = sinh−1 (νh/2), while j = ch/2ab is a dimensionless shape para-
meter. The limiting cases j = 0 and j = ∞ correspond, respectively, to
the ideal I-beam (with finite flange thickness but negligible web thickness)
and the rectangular beam. The calculated interaction curves are shown in
Figure 6.2.7.
Hodge also derived the same stress distribution by solving the con-
strained extremum problem, as in 6.2.3. On this basis Hodge regards the
yield locus as also providing a lower bound. However, the shear stress given
v
1.0
Rectangular beam
(j = ∞)

0.8
Ideal I-beam -
(j = 0)
0.6
I-beam *

(j = 1)
0.4
0.2
0 m
0 0.2 0.4 0.6 0.8 1.0
Figure 6.2.7. Interaction curves for I-beam (Hodge [1959]).
by (6.2.13)2 does not vanish at the extreme fibers, and it therefore appears
difficult to accept the stress distribution as statically admissible.
Plots of the lower-bound curve defined by Equation (6.2.12), the upper-
bound curve due to Hodge [1957b], and the lower-bound and upper-bound
curves due to Drucker [1956] (see 6.1.2) are shown in Figure 6.2.8. It can
be seen that all the curves have the limit f = 1 as δ → 0, so that the effect
of shear is negligible for beams that are not overly deep, and that all tend
asymptotically to the hyperbola f = 1/δ for δ large. It must be remem-
bered, however, that values of δ greater than about 0.5 cannot reasonably
be regarded as describing beams.

1. Using the assumptions of elementary beam theory, derive an expression
for the internal virtual work in a beam of arbitrary cross-section subject
to a variable bending moment M and axial force P .
2. Find the ultimate yield locus for a beam having the idealized section
of Figure 3.5.2.
3. For a beam whose cross-section is an isosceles triangle, find the ul-

timate yield loci for combined axial force and bending moment (a)
perpendicular and (b) parallel to the axis of symmetry.
4. Analyze the collapse of a semicircular arch of ideal sandwich section

f
1.0
0.8
0.6
0.4
0.2
0 δ
0.01 0.02 0.05 0.1 0.2 0.5 1.0 2.0 5.0 10.0
Upper bound, Hodge
Lower bound, Eq. (6.2.12)
Two-dimensional theory (Fig. 6.1.6)
Figure 6.2.8. Lower-bound and upper-bound curves under combined bending
and shear of a rectangular beam, load against depth-span ratio.
that is built in at the supports and carries a concentrated load 2F at

the vertex.
5. Analyze the collapse of a symmetric simply supported arch of ideal

sandwich section, forming a circular segment of angle 2α, and loaded
by (a) a concentrated load 2F at the vertex, (b) a uniform vertical
load of intensity q, and (c) a uniform radial load of intensity q.
6. Show that for combined torsion and axial force or bending moment
of a circular bar, the unit circles are lower bounds to the yield loci in
both the pt- and mt-planes.
7. Find a result analogous to (6.2.12) for an end-loaded cantilever of cir-

cular cross-section with radius a, letting δ = a/L.
8. Evaluate the integrals following Equation (6.2.13) for a rectangular

cross-section. Compare with the cited result of Hodge [1957b] for j =
∞.
Section 6.3 Limit Analysis of Trusses, Beams and

Frames
6.3.1. Trusses
A truss is an assemblage of stiff bars that are more or less flexibly connected
to one another at their ends. In an ideal truss, the connection is through
frictionless pins, with the center of each pin coinciding with the intersection
Section 6.3 / Limit Analysis of Trusses, Beams and Frames 375
r = −1 r=0 r=0
>Ar rB >Ar rB >Ar@ rB
@
@
>Dr rC >Dr rC fr @rC
@
D
(a) (b) (c)
r=1 r=1
r
A
>@ rB r
A
>@ rB
@ @
@ @
>Dr @rC >Dr @rC
@ @
(d) (e)
Figure 6.3.1. Some simple trusses.
of the centroidal lines of the bars meeting there. Moreover, all the loads are
applied at the joints. Consequently, all bars carry axial forces only, and if
n is the number of bars, then the bar forces P1 , . . . , Pn in effect constitute
the stress field.
In order for each pin to be in equilibrium, the vector sum of all the
bar forces in the members connected through that pin must be zero. If the
number of joints is j, then the total number of joint equilibrium equations is
3j, unless all the bars and loads lie in the same plane, in which case the truss
is a plane (or planar ) truss and the number of equations is 2j. A nonplane
truss is usually called a space truss.
The unknowns of the problem, in addition to the bar forces, are the
reaction components, s in number. If n + s = kj (where k = 2 or 3 in
the plane truss or space truss, respectively), then the equilibrium equations
are precisely enough to determine the unknowns, and the truss is statically
determinate (or isostatic). If r = n + s − kj is positive, then the truss is
statically indeterminate (or hyperstatic or redundant), and r is called the
degree of static indeterminacy (or, more simply, the indeterminacy number
or redundancy number ). If this number is negative, then the truss is unstable
(or hypostatic) and is, in fact, a mechanism. In a stable truss, the number s
of reaction components must be at least equal to the number of equilibrium
equations for the truss as a whole, three for the plane truss and six for the
space truss. If s is greater than this number, then the truss is externally
indeterminate.
Some simple trusses, with r equal to 0, −1 and 1 are shown in Figure
6.3.1. In particular, the truss in (e) is statically indeterminate of degree one
even though, apparently, r = 6 + 4 − 2·4 = 2. In fact, bar AD cannot deform
and therefore cannot carry any force. This bar may, as a result, be ignored
in any analysis of the truss.
Limit Analysis of Trusses
A truss member will be said to fail if it can undergo significant lengthen-
ing or shortening with no significant change in the bar force. Failure in this
sense can result from yielding, if the material is perfectly plastic or nearly
so, or, in the case of a compression member, from buckling. Since the bar
force in a failed member is no longer determined by equilibrium but by the
failure criterion, it can be presumed as known if the properties of the bar
are known, and the number n of unknown bar forces drops by one, as does
the indeterminacy number r. The truss therefore becomes unstable if r + 1
bars fail. In particular, a statically determinate truss collapses as soon as
one bar fails.
Any choice of r+1 bars that fail provides a mechanism with one degree of
freedom that can be used with the upper-bound theorem. Given a reference
velocity v, the elongation rates ∆ ˙ i of the ith failed bar may be determined
by geometry. Let PU i denote the ultimate bar force in tension, and PU−i the
+
magnitude of that in compression; the latter is the lesser of the yield force
and the buckling force. The internal dissipation in a failed bar is PU±i |∆ ˙ i |,
˙
the superscript sign being that of ∆i .
The truss of Figure 4.1.1 has three bars, six reaction components, and
four joints, so that r = 1 and two bars must fail for collapse to occur.
In 4.1.4 we read, however, that all three bars must fail for this truss to
collapse. With PU+ the same in all three bars, the collapse load in this case
is FU = PU+ (1 + 2 cos α).
If we were to assume a mechanism in which bars AD and CD fail, while
BD remains rigid, the external work rate would be zero and the upper-
bound theorem would fail to give a finite upper bound to FU . Consider,
now, an asymmetric mechanism with bars BD and CD only failing. Bar
AD then rotates rigidly about its support, and if the downward component
of the velocity of pin D is v, then its leftward component is v cot α. The
velocity component in the direction of bar CD, equal to its elongation rate,
is v(cos α + cot α sin α) = 2v cos α. Equating the external work rate to the
total internal dissipation,
F v = PU+ v(1 + 2 cos α),
leads to the an upper bound equal to FU . The equality can be explained

as follows: suppose that the symmetry of the system is disturbed ever so
slightly (e.g., by giving the load a small leftward component or by making
bar CD just a little bit weaker than AD). The correct collapse mechanism
would then, indeed, be the asymmetric one just considered. But a minimal
asymmetry should not significantly affect the collapse load calculated on
e
@ e
@ @
e
@ e
@ @
e@ e@ @
e@ e@ @
@
e @
e hhhh@
h
(i) (ii) (iii)
hh hhhh
@ @ hhh @ h
@ @ @

@ @ @
@ @ @
hhhh@ @ hhhh@
h h
(iv) (v) (vi)
Figure 6.3.2. Collapse mechanisms for the truss of Figure 6.3.1(d).
the basis of assumed symmetry, and hence the collapse loads given by the
symmetric and asymmetric mechanisms should be the same.
Let us now look at the possible collapse mechanisms of the truss of Figure
6.3.1(d) under a downward force F applied at joint C. First, it is apparent
that any mechanism in which bars AC and CD do not deform does not
permit any translation of C, so that the force F cannot do any work; such
a mechanism is not admissible. A mechanism in which both AC and CD
fail but the other bars remain rigid allows rotation of BC about B, leading
to a zero external work rate for the given loading. The only mechanisms to
be considered, therefore, are those in which either AC or CD fails, but not
both; these mechanisms number six, and are shown in Figure 6.3.2. Dashed
lines represent bars that fail.
We assume that the bars are of identical cross-section and sufficiently
stiff so that PU+ = PU− = PU in every bar. The mechanism (i) in which
CD and BC fail presents a rigid rotation of AC about A; if the downward
velocity of C is v, then the angular velocity of rotation of AC is v/a, where
a is the length of the nondiagonal bars, so that the shortening rate of CD
and the lengthening rate of BC are also v. The total internal dissipation is
thus 2PU v, and the resulting upper bound on F is 2PU . If (ii) CD and AB
fail, then AC rotates about A, BD rotates about D, and BC rotates about
its midpoint. The shortening rate of CD is again v, as is the lengthening
rate of AB, and the upper bound is once more 2PU .
The mechanisms in which AC fails along with either (iii) BC or (iv) AB,
and the one √ (v) in which CD and BD fail, all give the same upper bound
of (1 + 1/ 2)PU . Finally, in the shear mechanism (vi) in which AC and √
BD fail, these bars lengthen and shorten, respectively, at the rate v/ 2,
√
leading to the upper bound 2PU . This upper bound, being the smallest,
must equal the ultimate load.
To confirm this equality, we analyze the truss statically under this value
of F , assuming that PAC = PU√and PBD = −PU . We find that all the other
bar forces are of magnitude P/ 2, so that the “stress field” (the distribution
of bar forces) is plastically admissible.
It is clear that if the cross-sectional
√ areas of all three nondiagonal bars
were √ to be reduced by a factor of 2, their ultimate bar forces would fall
to 1/ 2 times those of the diagonals, and the stress field would still be
— just barely — plastically admissible under the just-calculated collapse
load. Collapse would then occur with all bars yielding, indicating the most
efficient use of material; such a truss would be an example of minimum-
weight design.
A kinematic analysis of the minimum-weight truss shows that all six
mechanism based on failure in two bars produce the same upper bound. The
most general collapse mechanism, then, is a linear combination of the six. It
is easy to see that this mechanism has four degrees of freedom, represented
by the horizontal and vertical displacements of joints B and C.
Limit Design of Trusses

The design of structures on the basis of limit analysis is known as limit
design. Limit design for minimum weight is particularly simple for trusses
in which no bars buckle, because the weight of a bar is directly proportional
to its area, and hence to its strength. Let the ultimate bar force of the ith
bar be PU i , and its length Li . In minimum-weight design, the object is then
to minimize n
X
PU i Li
i=1
subject to inequality constraints on the PU i , ensuring that the truss is strong
enough to carry the prescribed load. The problem is one of linear pro-
gramming; it is discussed in its general form after a study of the present
example.
Suppose, for simplicity, that PU (AB) = PU (BC) = PU (CD) = PU 1 and
PU (AC) = PU (BD) = PU 2 . We are thus led to minimize
√
3PU 1 + 2 2PU 2
subject to the following inequalities:

PU 2 √
2PU 1 ≥ F, PU 1 + √ ≥ F, 2PU 2 ≥ F ;
2
in this elementary case the six mechanisms furnish only three independent
inequalities, and the second of the three is redundant, since it is satisfied
whenever the other two are. The solution is simple, as √we already know: all
the inequalities are satisfied as equalities when PU 2 = 2PU 1 = F/2.
Linear-Programming Formulation of Limit Design

If a truss has n bars, then the number of different combinations of r + 1
bars — that is, the number of different mechanisms — is
n!
m= .
(n − r − 1)! (r + 1)!
Let v denote a positive reference velocity in the kth mechanism (k =

1, . . . , m), such that the elongation rate of the ith bar in this mechanism
is αik v, and the velocity conjugate to the applied force Fj (j = 1, . . . , l) is
βjk v. According to the upper-bound theorem, then,
n
X l
X
αik PU i ≥ βjk Fj , k = 1, . . . , m. (6.3.1)
i=1 j=1
The problem of minimum-weight design of a truss is then one of minimizing

the linear function
n
X
G(PU 1 , . . . , PU n ) = Li PU i
i=1
subject to the linear constraints (6.3.1); this problem is one of linear pro-
gramming. The connection between limit design and linear programming
was apparently first noted by Heyman [1951].
The standard linear-programming problem is one of maximizing, rather
than minimizing, the objective function cT x, where c and x are 1 × n column
matrices, subject to the constraints
aTk x ≤ bk , k = 1, . . . , m, (6.3.2)
where the ak are 1 × n column matrices, and the bk are real numbers. Often
the additional constraint that the xi be nonnegative is imposed. The problem
of minimum-weight design of a truss becomes a standard problem if we
identify xi with PU i , ci with −Li , ak with the kth column of the matrix
−[αik ], and −bk with the right-hand side of (6.3.1). By defining the m × n
matrix A through AT = [a1 , . . . , am ] and the m × 1 matrix b through
b = (b1 , . . . , bm ), inequality (6.3.2) may be rewritten in the short form
Ax ≤ b, (6.3.3)
and the additional constraint as
x ≥ 0. (6.3.4)
In an alternative formulation of the linear-programming problem, the

constraint conditions (6.3.3)–(6.3.4) are replaced by
Ax = b, x ≥ 0. (6.3.5)
An efficient numerical solution method due to Dantzig, called the simplex

algorithm, exists for the problem in this form (see Dantzig [1963]).
The constraints (6.3.3)–(6.3.4) can be converted to the form (6.3.5) by
introducing the slack variables yk (k = 1, . . . , m), which form the m × 1
matrix y, and rewriting (6.3.3) as
Ax + y = b, y ≥ 0. (6.3.6)
The (n + m) × 1 matrix x̄ is now defined by x̄i = xi (i = 1, . . . , n) and

x̄n+k = yk (k = 1, . . . , m). The m×(n+m) matrix Ā is similarly defined by
Āki = Aki (i = 1, . . . , n) and Āk,n+j = δjk (j = 1, . . . , m). The constraint
inequalities (6.3.4) and (6.3.6) together can now be written as
Āx̄ = b, x̄ ≥ 0,
a form identical with (6.3.5).
6.3.2. Beams
Any transversely loaded beam, except an ideal sandwich beam, is statically
indeterminate in the sense of Section 4.1 — that is, the stress field cannot
be deduced from the loading independently of unknown properties: at any
section, an infinity of stress distributions can be found that give the same
resultant moment M and shear force V . It is conventional, however, to call
a beam statically determinate (or indeterminate) if it is externally determi-
nate (or indeterminate) in the same sense as a truss, that is, if the number
of independent reaction components is the same as (or greater than) the
number of equilibrium equations available to determine them. In the ab-
sence of internal hinges, this number is three for plane bending. Any hinge,
whether frictionless or a plastic hinge, provides an additional equilibrium
equation: at a frictionless hinge, M = 0, since such a hinge cannot transmit
moment, while at a plastic hinge M = MU+ or M = −MU− . The indetermi-
nacy number of a beam is accordingly r = s − h − 3, where s is the number
of reaction components1 and h is the number of hinges. Like a plane truss,
the beam collapses when r is reduced to −1, so that if h0 hinges are present
initially, the number of plastic hinges required for collapse is s − h0 − 2, and
specifically, one if the beam is statically determinate, and two or more if it is
statically indeterminate. A plastic hinge may form at any point of the beam
at which the condition |M | = MU is possible, that is, in the interior of a
1
Provided that these components do not include three collinear forces.
span, at a built-in end, or at an intermediate support. A collapse mechanism

is admissible if it does not violate any support condition (possibly relaxed
by the formation of a plastic hinge) and if it produces a positive external
work rate.
If the effect of shear on the formation of a plastic hinge can be ne-
glected, as will be assumed, then a hinge that has rotated by an angle ∆θ
can be thought of as the limit of a small segment, of length, say, ∆x, in
which the curvature is ∆θ/∆x and the plastic dissipation per unit length is
MU |∆θ̇|/∆x. The total internal dissipation in the hinge is therefore MU |∆θ̇|.
A moment distribution is statically and plastically admissible if it is in
equilibrium with the applied loads, is consistent with all force and moment
end conditions and frictionless hinge conditions (if any), and is such that
|M | ≤ MU everywhere. In the moment distribution at collapse, the points
where M = ±MU are precisely the ones where plastic hinges form.
Example: Beam with Point Loads

If the beam carries point loads only, then the moment can vary only in
straight-line fashion between points where concentrated forces (loads or reac-
tions) act, and therefore the actual collapse mechanism must be one in which
hinges form at load points, built-in ends, or intermediate supports. Consider
the beam shown in Figure 6.3.3(a), indeterminate to the first degree, and
requiring two plastic hinges for collapse. The possible mechanisms are (1)
with hinges at A and B, (2) with hinges at A and C, and (3) with hinges at
B and C. They are shown as Figure 6.3.3(b), (c), and (d), respectively.
Let us look first at mechanism 1. If the downward displacement of point
˙ +
B is ∆, then that of point C is 21 ∆, so that the external work rate is αF ∆
1 ˙ ˙
(1 − α)F ∆ = 2 (1 + α)F ∆. The angles of rotation of the hinges at A
1
2
and B, assumed small, are respectively 3∆/L and 9∆/2L; the total internal
˙
dissipation is therefore 15MU ∆/2L, and if this is equated to the external
work rate, the resulting upper bound on F L/MU is 15/(1 + α).
A similar analysis of mechanisms 2 and 3 leads to the respective upper
bounds of 12/(2 − α) and 6/(1 − α); but the second of these is greater than
the first for any α between 0 and 1, so that mechanism 3 may be discarded.
It can easily be seen that mechanism 1 gives the lesser upper bound when
α > 2/3, and mechanism 2 when α < 2/3. The ultimate load is therefore
given by
12 2

FU L

 , α≤ ,
2 − α 3

=
MU 15 2
, α≤ .



1+α 3
The preceding result may also be cast in the form of two inequalities,
αF (1–α)F
L L L
- - - ∆
3 3 3 fh
fD
? ?
f f
hh? h?
h ? ?
A B C hhh
3∆ 6 3∆ 6
(a) L 2L
(b)
fh fPP
?
f f
hhhh
hhh?
h hf
hhh ?
PP f
PP PP ?
hh
h
(c) (d)
F2 L/MU
10
HH
5
HH
Hq(6, 3)
H
A
A
A F1 L/MU
5 10
(e)
Figure 6.3.3. Single-span beam simply supported at one end and built in at the
other, with concentrated loads at the third points: (a) geometry
and loading; (b)–(d) collapse mechanisms; (e) interaction dia-
gram.
parametrically dependent on α, that the total load F must obey:
(2 − α)F L (1 + α)F L
≤ 12, ≤ 15.
MU MU
Yet another way of presenting the result would be to regard the loads at B
and C as two independent loads F1 and F2 . The inequalities are accordingly
rewritten as
MU
F1 + 2F2 ≤ 12 ,
L (6.3.7)
MU
2F1 + F2 ≤ 15 ,
L
represented graphically by the interaction diagram shown in Figure 6.3.3(e).
The last pair of inequalities, Equations (6.3.7), can also be derived by
means of an equilibrium analysis. Suppose that a plastic hinge has already
formed at A, with MA = −MU ; the beam is then statically determinate, and
the bending moments at B and C can easily be calculated to be, respectively,
(2F1 + F2 )L/9 − 2MU /3 and (F1 + 2F2 )L/9 − MU /3. The requirement that
these moments not exceed MU gives precisely the inequalities (6.3.7).
The inequalities are also design criteria: they give the minimum value of
MU that the section must have in order to carry a given set of loads F1 , F2 .
Example: Beam with Distributed Load

Suppose, now, that the beam just examined carries a uniformly distrib-
uted load of intensity F/L rather than the point loads. It can still be assumed
with some certainty that one of the plastic hinges necessary for collapse will
form at the built-in end, but the other hinge can be, in principle, anywhere
along the span of the beam. In fact it will form, of course, at the section
where the bending moment has a local extremum.
If a hinge has formed at the built-in end (x = 0), then the bending
moment at any x, 0 ≤ x ≤ L, is
x fx x

M (x) = MU − 1 − + 1− ,
L 2L L
where f = F L/MU . The maximum of the quantity in brackets occurs at

def
x/L = 12 + 1/f , and equals (f − 4 + 4/f )/8 = φ(f ). Any value of f for
which φ(f ) does not exceed unity is a lower bound for fU = FU L/MU . For
example, φ(10) = 0.8, so that 10 is a lower bound. On the other hand,
φ(12) = 1.04, so that fU must be somewhat less than 12.
The actual value of fU is obtained by setting φ(f ) equal to unity, giving
the quadratic equation
f 2 − 12f + 4 = 0.
√
This equation has the two roots 6 ± 4 2. Clearly, since what is sought is
the greatest lower bound, the larger root must be chosen. We thus obtain
fU = 11.657
In a kinematic solution, a mechanism with a plastic hinge at x = 0
and another at x = αL, with α to be determined, is assumed, as in Figure
6.3.4. If the downward displacement of the hinge is ∆, then the average
αL -
fX x
(((( f
?X??? (? (
(??
XX ??
( (
XX?
XXXX?f(
?
(
((
?
?((
?
(
L -
Figure 6.3.4. Uniformly loaded single-span beam, simply supported at one end
and built in at the other: collapse mechanisms.
displacement of both rigid portions of the beam is ∆/2, so that the external
˙
work rate is F ∆/2. The hinge at the built-in end rotates by an angle ∆/αL,
and the other hinge by the angle ∆/αL + ∆/(1 − α)L. The total internal
dissipation is therefore
2−α ∆˙
MU .
α(1 − α) L
It follows that an upper bound to fU is
2(2 − α)
f= .
α(1 − α)
The least upper bound is found by minimizing f with respect to α:
1 df 2−α 2−α 1
= − −
2 dα α(1 − α)2 α2 (1 − α) α(1 − α)
1
= [α(2 − α) − (1 − α)(2 − α) − α(1 − α)] = 0,
[α(1 − α)]2
leading to the quadratic equation
α2 − 4α + 2 = 0.
√
√ relevant root is α = 2 − 2 and gives
Since α must be less than 1, the only
the least upper bound f = 6 + 4 2 = 11.657, which, of course, coincides
with the previously found greatest lower bound.
Without the analytical solution, assumed values of α give upper bounds
that may be satisfactory; for example, α = 0.5 leads to f = 12, and α = 0.6
leads to f = 11.667. Moreover, assumed mechanisms can also be used to
give lower bounds without resorting to an analytical solution. Consider,
for example, the mechanism with α = 0.5, corresponding to f = 12. The
moment distribution corresponding to this mechanism is given by
" 2 #
x x
M (x) = −MU 1−7 +6 ,
L L
and is not plastically admissible because |M |max = M (7L/12) = (25/24)MU .

If, however, all the moments are multiplied by 24/25, then the distribution
becomes plastically admissible, and in equilibrium with a load for which
f = (24/25)12 = 11.52, which is thus a lower bound.
Example: Continuous Beam

A beam with two or more spans, separated by intermediate supports
that exert transverse force reactions, is called a continuous beam. The rule
governing the degree of static indeterminacy is the same for continuous as
for simple beams. However, collapse of a continuous beam may occur in one
span only, and does not in general require r + 1 plastic hinges. The collapse
of a span between two intermediate supports, or between an intermediate
support and a built-in end support, requires three hinges. A span between
an intermediate support and a simple end support will collapse with only
two hinges. It is thus possible for a continuous beam to remain statically
indeterminate at collapse. For this reason the kinematic method is preferable
by far for the limit analysis of continuous beams.
F1 F2
L L L L
- - - -
4 4 4 4
A ? ?
f fB C
Figure 6.3.5. Continuous beam, simply supported at one end, built in at the
other, and with an intermediate support, carrying a concentrated
load at the midpoint of each span.
Looking at the beam shown in Figure 6.3.5, we see immediately that the
feasible collapse mechanisms are (a) the one in which span AB collapses like
the beam of Figure 6.3.4, and (b) the one in which span BC collapses like
a beam with both ends built in. In mechanism (b) hinges develop at B (or
rather, just to the right of B), at C, and at midspan. The loads F1 , F2 are
governed by the uncoupled inequalities,
√ MU MU
F1 ≤ (6 + 2 2) , F2 ≤ 16 .
L L
These inequalities also specify the minimum value of MU .
6.3.3. Limit Analysis of Frames

A rigid frame (or simply a frame) is an assemblage of bars that are joined
together rigidly, so that they cannot rotate with respect to one another. The
joints transmit bending moment, and the members resist the applied loads
primarily through bending; axial force and shear are considered secondary
effects. Collapse is assumed to occur when sufficient plastic hinges have
formed to produce a mechanism. In a multistory frame, collapse may be
limited to a single story, and therefore the overall degree of static indeter-
minacy is not a relevant parameter for the determination of the necessary
number of hinges.
Simple Frame
A one-story, one-bay frame such as shown in Figure 6.3.6 is statically
indeterminate of degree three, and the collapse of the frame as a whole indeed
requires four hinges, as shown in Figures 6.3.6(a) and (c). Consider, however,
Figure 6.3.6(d), which illustrates the beam mechanism. This mechanism does
not entail collapse in the sense of unlimited displacements; the deflection of
the beam is limited by that of the columns. In practice, however, a structure
may be said to collapse when its displacements can become significantly

greater than those in the elastic range. If the axial elongation of the beam is
neglected, a deflection ∆ of the central hinge requires
q that the beam-column
.
joints move laterally inward by a distance 2 L − 14 L2 − ∆2 = ∆2 /L. This
1
distance represents the elastic deflection of the columns, ∆e , which is of the

same order of magnitude as the beam deflection when the whole frame is
. p
elastic. Now ∆/∆e = L/∆e is a large number. We are therefore justified
in regarding the beam mechanism as a collapse mechanism.
F2
F1 - sB
hhh? s ((
D - hhh( s ((s
? -s ? s - shhh?
s ((s
(
E 6C( E
E E
E H E
E E
A Es ? EsE s s s s
L -
(a) (b) (c) (d)
Figure 6.3.6. One-story, one-bay frame: (a)–(c) four-hinge mechanisms; (d)

beam mechanism.
The only pertinent collapse mechanisms for the frame of Figure 6.3.6
are the beam mechanism (d), the panel or sidesway mechanism (c), and the
composite mechanism (b), which is a superposition of (c) and (d) in which
the hinge at B is eliminated. The composite mechanism (a) — the mirror
image of (b) — in which joint D is rigid, entails negative work done by the
horizontal force and therefore is viable only when this force is zero, in which
case it is equivalent to (b). When a hinge is assumed to be at a joint such as
B or D, it will actually be in the weaker of the two members meeting there
— that is, it forms when the bending moment (which, for equilibrium, must
be the same in both members as the joint is approached) reaches the smaller
of the two values of MU .
Let MU 1 , MU 2 , and MU 3 denote the values of MU in AB, BD, and DE,
respectively. If it is assumed that the horizontal load F1 is, for example, a
wind load which is just as likely to act to the left at D as to the right at B
(so that the loading of Figure 6.3.6 represents only one of two mirror-image
cases), then the frame design should be symmetric, and MU 3 = MU 1 .
The upper-bound theorem applied to the three mechanisms (b)-(d) gives
the following inequalities:
F2 L ≤ 4MU 2 + 4 min(MU 1 , MU 2 ),
F1 H ≤ 2MU 1 + 2 min(MU 1 , MU 2 ), (6.3.8)

2F1 H + F2 L ≤ 4MU 1 + 4MU 2 + 4 min(MU 1 , MU 2 ).
As before, these inequalities serve both analysis and design. For the
purposes of analysis, let us assume equal values of MU for all three members.
The interaction diagram between F1 and F2 is then as shown in Figure
6.3.7(a). The design implications of inequalities (6.3.8) are discussed in the
following subsection.
MU 2 /M̄
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
F2 L/MU .....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
3 .....
.....
.....
.....
..... Safe region
....
.....
.....
.....
....
.....
.....
.....
....
.....
...........
8............
...........
...........
............
............
A .....
.....
....
.....
.....
............
............
.............
.............
.............. .....
....
.....
.....
.....
..............
.....
.....
.....
..... ....
.....
.....
.....
....
.....
.....
A
.....
.....
.....
A .....
.....
.....
....
.....
.....
.....
.....
.....
..... 2 .....
.....
....
.....
.....
.....
.....
.....
.....
..... .....
....
.........................................
........................................
........................................
.....
A
.....
.....
.....
.....
AA .......................................
.......................................
......................................
.....................................
.....................................
.....
.....
.....
.....
..... ....................................
A.....
.....
.....
.....
......
4 .....
......
.....
......
.....
......
..... Safe boundary
......
.....
......
.....
...... 1
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
.....
......
..... F1 H/MU MU 1 /M̄
4 8 1 2 3
(a) (b)
Figure 6.3.7. Interaction diagrams for the frame of Figure 6.3.6: (a) load plane;
(b) design plane (see page 391).
Complex Frames
In a frame comprising several stories and bays, the number of possi-
ble collapse mechanisms can become quite large. Every transversely loaded
member may form a beam mechanism, and each story may produce a panel
mechanism. Furthermore, at any joint at which three or more members
come together, a plastic hinge may form independently in each member near
the joint (if only two members meet, the hinge can form only in the weaker
member).
It is convenient to establish a basis of independent mechanisms, called
elementary mechanisms, such that all mechanisms may be regarded as su-
perpositions of the elementary ones. These elementary mechanisms, as first
discussed by Neal and Symonds [1952], consist of all the beam and panel
mechanisms, and in addition, of the joint mechanisms constituted by the
formation of plastic hinges, at a joint, in every one of the members that
come together there, resulting in a rotation of the joint [see Figure 6.3.8(e)].
The joint mechanisms are not in themselves collapse mechanisms, since the
external work rate associated with them is zero (unless an external moment
acts at the joint), but they are used in combination with beam and/or panel
mechanisms in order to cancel superfluous hinges.
Let r denote, as before, the degree of redundancy of the frame. A simple
method of determining r is to cut the frame at a sufficient number of sections
so that it just becomes statically determinate, that is, equivalent to a set of
simply supported beams and/or cantilevers; r is then the number of stress
resultants (moments, axial forces and shear forces) that can arbitrarily be
specified at the cuts. Equivalently, r is the number of sections at which the
moment can be arbitrarily prescribed. Suppose, now, that the number of
critical sections — that is, sections at which a plastic hinge can form — is n.
It follows that there are n − r independent relations among the n moments
at the critical sections, and these relations are equilibrium equations. Each
such equation can be associated, by means of the principle of virtual work,
with a mechanism. Consequently, there are n − r independent mechanisms.
As an example, consider the two-bay frame shown in Figure 6.3.8(a).1
By means of two cuts, the frame can be transformed into three disconnected
cantilevers, and therefore r = 2×3 = 6. The critical sections, as shown, num-
ber 10. Consequently the frame has four independent mechanisms. In terms
of elementary mechanisms, these are (b)–(c) the two beam mechanisms, (d)
the panel mechanism, and (e) the joint mechanism.
In the method of superposition of mechanisms due to Neal and
Symonds [1952], the analysis begins by determining the upper bounds pre-
dicted by the elementary beam and panel mechanisms. Because of the
symmetry of the structure, the two beam mechanisms give the same up-
per bounds. We thus obtain the two inequalities
(2F )Lθ̇ ≤ M0 θ̇ + 2M0 (2θ̇) + 2M0 θ̇ = 7M0 θ̇, (b, c)
F (2L)θ̇ ≤ (4·M0 + 2·1.5M0 )θ̇ = 7M0 θ̇, (d)

both of which yield the upper bound F = 3.5M0 /L.
In order to improve the upper bound, we proceed to study composite
mechanisms. Mechanism (f) is a superposition of (c), (d), and (e) in which
the hinges at sections 5 and 6 are eliminated, while a hinge is created at
4. The internal dissipation of the mechanisms can therefore be obtained by
subracting from the sum of the right-hand members of the two preceding
inequalities the quantity (1.5M0 + 2M0 − 2M0 )θ̇ = 1.5M0 θ̇, or (7 + 7 −
1.5)M0 θ̇ = 12.5M0 θ̇. The external work rate is just the sum of the left-
hand sides, or 4F Lθ̇. Mechanism (f) therefore gives the upper bound F =
3.125M0 /L.
Mechanism (g) is a further superposition of (f) and (b) in which the hinge
at 2 is eliminated. The internal dissipation is (12.5 + 7 − 1)M0 θ̇ = 18.5M0 θ̇,
while the external work rate is 6F Lθ̇. We thus obtain the even smaller upper
bound of F = 3.083M0 /L.
In the present example, it appears that we have run out of reasonable
mechanisms, and the result should give us the collapse load. In more complex
cases, it may be quite difficult to make sure that all the possible collapse
1
The circled numbers next to the members mean that the value of MU for a member
is the given number times a reference moment M0 .
2F 2F
4 6
F x? xxx x? i x 8
2i 3
-
2x 5 7 2 6
1i i
1.5 1i 2L
1x x10 x9 ?
2L - 2L -
(a)
rPPP r rP
PPr r
r
(b) (c)
r r
r r r r

r r r
(d) (e)
r r`` r r`` r
`r
```r
`r

r r r r r r
(f) (g)
Figure 6.3.8. Two-bay frame: (a) geometry and loading; (b)–(c) beam mech-
anisms; (d) panel mechanism; (e) joint mechanism; (f)–(g) com-
posite mechanisms.
mechanisms have been explored. The only way to check whether the best
upper bound that has been found indeed gives the collapse load is to see
if it is also a lower bound, that is, to find a statically admissible moment
distribution such that |M | = MU at all sections corresponding to hinges in
the mechanism, and |M | ≤ MU elsewhere. In the present example this is
easy, since the optimal mechanism represents total collapse and thus involves
r+1 = 7 hinges, leaving the structure statically determinate at collapse. The
four independent equilibrium equations (which can be formed by applying
the principle of virtual work to the elementary mechanisms) and the seven
hinge conditions give eleven equations for the ten critical-section moments
and the load F . It turns out that the moments at 2, 5, and 6 do not exceed
the local values of MU , and the load is in fact equal to the best upper-bound
value derived above.

The situation is more difficult in multistory frames, in which the best
mechanism found by the method of superposition of mechanisms often rep-
resents partial collapse, with fewer than r+1 plastic hinges and therefore not
enough equations for a rigorous moment check. If the degree of redundancy
remaining at failure is small (one or two), a trial-and-error procedure is
usually applied: guesses are made for a sufficient number of critical-section
moments so that the equilibrium equations can be solved. In more com-
plicated cases, the moment-distribution method due to Horne [1954] and
English [1954] may be used; for examples of application, see Hodge [1959],
Chapter 3.
A method of analysis based on the lower-bound theorem is the method
of inequalities, due to Neal and Symonds [1951]. The critical-section mo-
ments Mi (i = 1, . . . , n) are governed by the 2n inequalities
−MU i ≤ Mi ≤ MU i ,
and by the n − r equilibrium equations, at least one of which contains the

load F . The equilibrium equations can be transformed by means of linear
combinations so that only one of them contains F . The problem now is again
one of linear programming: the equation containing F serves as the defining
equation for F as the function of the Mi that is to be maximized, with the
remaining equations, as well as the inequalities, serving as constraints.
6.3.4. Limit Design of Frames

Limit Design of a Simple Frame
The achievement of a minimum-weight design is not so simple for frames
as it is for trusses, because there is no simple proportionality, or indeed any
one-to-one relation, between weight and strength. In simple frames, such as
the one analyzed in the preceding subsection, a trial-and-error approach is
usually the easiest (see Heyman [1953]). In complex frames a more system-
atic approach, based on some simplifying assumptions, is necessary in order
that the problem may be converted into one of mathematical (not necessarily
linear) programming.
As a first example of limit design of a frame, we consider the fixed-
base rectangular frame of Figure 6.3.6 (page 386), for which we derived
inequalities (6.3.8). Let the design loads (working loads times appropriate
def
safety factors) be such that F2 L = 3F1 H = 12M̄ , where M̄ is a reference
quantity having the dimensions of a moment. The sections chosen for the
columns and the beam must then satisfy the following inequalities:
MU 2 + min(MU 1 , MU 2 ) ≥ 3M̄ , (a)

Table 6.3.1. Plastic Moduli of Selected Wide-Flange Sections
Z (in.3 ) Shape Z (in.3 ) Shape Z (in.3 ) Shape

287 W14×159 212 W14×120 150 W16×77
260 W14×145 198 W16×100 130 W16×67
234 W14×132 175 W16×89 105 W16×57
MU 1 + min(MU 1 , MU 2 ) ≥ 2M̄ , (b)

MU 1 + MU 2 + min(MU 1 , MU 2 ) ≥ 5M̄ . (c)
The solution of these inequalities is shown in Figure 6.3.7(b) (page 387) as
the “safe region” whose polygonal boundary will be called the safe boundary.
Any point in the safe region or on the safe boundary will be said to represent
a safe design.
We note that MU 1 must be at least equal to M̄ and MU 2 to 1.5M̄ ; but
if these criteria are met, there is no need for MU 1 to be greater than 2M̄ ,
or for MU 2 to be greater than 3M̄ . Thus the choice of sections can be made
from a rather restricted range. Let the frame dimensions be L = 24 ft.
and H = 12 ft., so that the total weight of the frame is proportional to
w1 + w2 , where w1 and w2 denote the weight per unit length of the column
and beam, respectively. The design loads will be taken as F1 = 1.0 × 105 lb
and F2 = 1.5 × 105 lb, corresponding to M̄ = 3.6 × 106 lb-in. With the usual
value of σY = 36 × 103 lb/in.2 for A36 structural steel, it follows that that
M̄ /σY = 100 in.3 . Then the ranges of the plastic modulus Z = MU /σY for
the columns and the beam are
100 in.3 ≤ Z1 ≤ 200 in.3 , 150 in.3 ≤ Z2 ≤ 300 in.3 ,
and they must obey the inequalities
2Z1 + Z2 ≥ 500 in.3 , Z1 + 2Z2 ≥ 500 in.3 .
It will be assumed that for architectural reasons, the section depth is

to be limited to 16 in. The choice will be made from standard wide-flange
sections, where the designation Wd × w refers to a section whose depth is d
(in inches) and whose weight per unit length is w (in pounds per foot). The
strength of a wide-flange beam, for a given weight, increases sharply with
the depth, and therefore the deepest available sections should be chosen for
economy. A listing of section properties for selected wide-flange sections is
shown in Table 6.3.1.
Four designs will be tried.
1. In a design with the lightest possible columns, a W16×57 section
presents Z1 = 105 in.3 , requiring ZU 2 ≥ 290 in.3 . Unfortunately, no standard
W16 section has a plastic modulus close to this value. We consequently

choose, for the beam, a W14×159 section, giving Z2 = 287 in.3 . If this is
assumed to be close enough, we obtain a design with w1 + w2 = 216 lb/ft.
2. The lightest possible beam section is a W16×77, with Z2 = 150 in.3 ,
.
requiring columns with Z1 = 200 in.3 , which is provided by a W16×100
(Z = 198 in.3 ); this choice yields w1 + w2 = 177 lb/ft., a considerable
improvement over the first trial.
3. The beam and column sections of design 2 can be reversed, yielding
the same weight.
4. A beam section intermediate between those in designs 2 and 3 is
a W16×89 and gives Z2 = 175 in.3 ; this requires Z1 ≥ 162.5 in.3 , and the
lightest section satisfying this criterion is again W16×89; thus w1 +w2 = 178
lb/ft., virtually the same as designs 2 and 3, and superior to them by virtue
of the greater ease of connections resulting from having the same section
throughout. The minimum required value of Z1 = Z2 is 167 in.3 , so that
this design carries an overdesign factor of 1.05. The additional margin of
safety provides an allowance for axial force. If, for example, Equation (6.2.7)
is used for the interaction, then each member can carry an axial force up to
(1 − 0.85/1.05)PU = 0.19PU with no loss in moment-carrying capacity.
On the basis of design 4, MU 1 = MU 2 = 6.3 × 106 lb-in. With the
overdesign factor included, the collapse forces are F1 = 1.05 × 105 lb and
F2 = 1.575 × 105 lb The weight of the beam is about 2000 lb, and we are
therefore justified, in retrospect, in having neglected it.
Since the structure is statically determinate at collapse, an equilibrium
analysis can easily be performed. The moment at the only other critical
section, namely B, is found to be of magnitude 3.78 × 106 lb-in., so that the
yield criterion is nowhere violated, and the correct mechanism was chosen.
While this last result is obvious in the present example, in frames in which a
large number of possible mechanisms exists, the moment check is a necessity,
as in analysis.
The same equilibrium analysis shows that the beam and column DE
carry compressive axial forces of 0.875 × 105 lb, and that the maximum
shear force (in CD and DE) has the same magnitude. We consider, first,
the effect of axial force. The slenderness ratio of the columns, even if they
are taken as doubly pinned, is about 20, so that they are not expected to
buckle elastically. The cross-sectional area of a W16 × 89 section is 26.2 in.2 ,
so that PU = 9.432 × 105 lb, and |P/PU | = 0.093 < 0.19.
We consider, finally, the effect of shear. The web area of the section is
7.9 in.2 , and therefore the average shear stress in the web is some 11 × 103
lb/in.2 , well below the shear yield stress of about 20 × 103 lb/in.2 for A36
steel. It follows that the frame can be safely designed on the basis of bending
alone.
Foulkes’ Theorems
For a complex frame, the time required for a trial-and-error method
of limit design would be prohibitively long. Any systematic approach is
based on the assumption of a functional relation between the weight per
unit length and the flexural strength (as measured by MU or Z) of a beam.
Clearly, no such relation exists in general, but an approximate relation can
be established for a limited range of sections that is used in the design of a
frame. If a relation of the form
w = a + bMU (6.3.9)
can be found, then the problem of minimum-weight design of a frame can

also be transformed into one of linear programming: the total weight is
X X
W =a Li + b Li MU i ,
i i
the summation being over all members, and therefore the objective function
is
n
X
G(MU 1 , . . .) = Li MU i ,
i=1
independently of the parameters a and b in the approximate representation

(6.3.9); this approach was presented by Foulkes [1953].
In the just-studied simple frame, for the three W16 sections considered,
the unit weight (in lb/ft.) is very nearly given by
.
w = 5 + 0.48Z,
with Z in cubic inches. The total weight of the frame, to within an additive
constant, is therefore proportional to Z1 + Z2 . The theoretical minimum-
weight design is thus achieved by finding the point (Z1 , Z2 ) on the safe
boundary where the constant-weight line Z1 + Z2 = const. is tangent to the
boundary. This point is Z1 = Z2 = 167 in.3 , and would represent the actual
minimum-weight design if a section with this value of Z could be found.
Note that the point of tangency between the constant-weight line and
the safe boundary is a vertex of the boundary, that is, an intersection of
two of the lines forming this boundary.1 Since each such line represents a
particular collapse mechanism, the theoretical minimum-weight frame can
collapse in either of two mechanisms, or in a linear combination of the two.
In particular, a linear combination of the two mechanisms, with nonnegative
coefficients, can be found so that the inequality produced by the combined
1
If the constant-weight lines are parallel to one of the boundary lines, then any point on
the boundary segment of this line, including the two vertices, represents a minimum-weight
design,
mechanism is represented by a line that is parallel to the constant-weight

lines. Such a mechanism is known as a Foulkes mechanism or a weight-
compatible mechanism.
If n independent values of MU may be used in a design, then the design
space is n-dimensional. The theoretical minimum-weight design is repre-
sented by the point of tangency of a constant-weight hyperplane with the
safe boundary, which is made up of intersecting hyperplanes. If the point is
unique, then it is a vertex of the boundary and is therefore an intersection
of at least n of the hyperplanes making up the boundary, each represent-
ing a distinct collapse mechanism. A Foulkes mechanism can be formed by
combining these mechanisms linearly so that the combined mechanism is
represented by a hyperplane that is parallel to the constant-weight hyper-
planes.
Consider a frame design for which a Foulkes mechanism can be assumed.
Let Li denote the combined length of all the members whose ultimate mo-
ment is MU i . In a Foulkes mechanism, the plastic hinges forming in these
members have the property that the sum of the absolute values of their an-
gular velocities is proportional to Li . If the proportionality factor is c, then
the total dissipation in the mechanism is
n
X
c MU i Li = cG.
i=1
If, moreover, a statically and plastically admissible distribution of bending

moments compatible with the design can be found, then the design loads Fj
are the collapse loads, and therefore, by virtual work
X
Fj vj = cG,
j
where the vj are the velocities conjugate to the Fj in the Foulkes mechanism.
Consider, now, any other safe frame design, described by values MU∗ i
of the ultimate moments. Since the Foulkes mechanism is a kinematically
admissible mechanism, it follows from upper-bound theorem that
n
MU∗ i Li = cG∗ ,
X X
Fj vj ≤ c
j i=1
and therefore G∗ ≥ G. This result, due to Foulkes [1954], may be stated in

words as follows: A frame design that admits a Foulkes mechanism and a
compatible, statically and plastically admissible distribution of bending mo-
ments is a minimum-weight design.
Upper-bound and lower-bound theorems for minimum-weight design were
also established by Foulkes. The upper-bound theorem is obvious, since any
safe design provides an upper bound to the minimum weight. Conversely,

any design based on a Foulkes mechanism, without necessarily satisfying the
condition of admissible moments, provides a lower bound to the minimum
weight. A general method, based on Foulkes’ theorems, for the mimimum-
weight design of highly redundant frames was first developed by Heyman
and Prager [1958].
The concept of a Foulkes mechanism can be extended to structures other
than frames, including those modeled as continua. If, for example, a beam
can be designed with an arbitrarily varying cross-section, then a minimum-
weight design, based on the relation (6.3.9), is one in which the moment
distribution at collapse is such that |M | = MU everywhere. Plastic flow, in
this case, is not localized in hinges but occurs throughout the beam, with
a curvature-rate distribution κ̇. Since the plastic dissipation in a Foulkes
mechanism is proportional to G, it follows that
Z L Z L
MU |κ̇| dx = c MU dx,
0 0
and the Foulkes mechanism is one in which |κ̇| = c.

Similarly, a Foulkes mechanism for a plate obeying the Tresca criterion
is one in which
|κ̇1 | + |κ̇2 | + |κ̇1 + κ̇2 | = 2c,
where κ̇1 and κ̇2 are the principal curvature rates.
The general criterion for minimum-weight design of continua is due to
Drucker and Shield [1957]. Applications to plate design were discussed by
Hopkins and Prager [1955], Prager [1955b], and Freiberger and Tekinalp
[1956], and to shells by Onat and Prager [1955], Freiberger [1956b] and Onat,
Schumann, and Shield [1957]. The results for variable-section beams were
applied to frame design by Heyman [1959, 1960] and by Save and Prager
[1963]. Further contributions are due to Chan [1969], Maier, Srinivasam,
and Save [1972], and Munro [1979]. Many of these results are reviewed in
the books by Neal [1963], Massonnet and Save [1965], Heyman [1971], Save
and Massonnet [1972], Rozvany [1976], Horne [1979], and Borkowski [1988].
Additional Remarks
1. In both the analysis and the design of frames, all loads were assumed
to be concentrated, thus fixing in advance the locations of the critical sec-
tions. If any span carries a distributed load, then the critical section in that
span must be assumed, and therefore any mechanism gives an upper bound
to the collapse load; improvements to the upper bound can be achieved by
changing the hinge locations. On the other hand, if any distributed load is
replaced by a statically equivalent (equipollent) set of concentrated loads,
then the collapse load calculated on the basis of the concentrated loads is
a lower bound on the collapse load for the distributed load. This result,
derived by Symonds and Neal [1951], is known as the load-replacement
theorem.
2. Only rectangular frames were studied as examples. Frames with
inclined members, such as the gable frame of Figure 6.3.9, can be studied
analogously. The frame shown has three independent mechanisms; these can
r
Q Q Q
? QQ Q QQ
Q
!r
Q QQ
r
Q Q
r
QQ
- Q ! ! Q QQr

r r
(a) (b) (c)
r
Q
r
Q Q
Qr #Q
Z
Q
Q !!Q
HH # Z
Q
r
! # Z
Q r r#
Q QH Q
!! QH Hr Z
Q

r r r r r
(d) (e) (f)
r
Q
Q Q
Q
Q r
HQH
r Qr
Q !
! QH
! QHHr

r r
(g) (h)
Figure 6.3.9. Gable frame: (a) geometry and loading; (b) beam mechanism;
(c) panel mechanism; (d) gable mechanism; (e)–(f) other panel
mechanisms; (h) composite mechanism.
be taken as the beam mechanism (b), the panel mechanism (c), and the gable
mechanism (d), which may also be regarded as a kind of panel mechanism.
Other panel mechanisms, each of which is a combination of (c) and (d) with
one hinge eliminated, are shown as (e)–(g), and a composite mechanism as
(h).
3. A relation between weight and strength given by (6.3.9) can be a rea-
sonable approximation to the properties of actual sections if their number
is small. For a larger range of sections, a better approximation is usually
obtained with a nonlinear relation such as w ∝ MUα , where α = 2/3 for
.
geometrically similar sections, and α = 0.6 gives a good approximation for
many standard I-beam sections. The problem of minimum-weight design
then becomes one of nonlinear programming. The computational implemen-
tation of nonlinear programming is not yet at a point where it can be readily

applied to the plastic design of complex structures.
1. The truss of Figure 6.3.1(d) is subject to equal horizontal forces F

acting to the right at B and C, and a downward vertical force 2F
acting at B. If the ultimate bar force PU is the same in all bars and
is equal in tension and compression, find the smallest upper bound for
FU by analyzing mechanisms, and verify that it is also a lower bound.
2. Find a minimum-weight design for the truss of the Exercise 1, assuming

that PU = ±σY A in every bar.
3. A beam of span L, built in at the left end and simply supported at

the right end, carries a load αF uniformly distributed over the left half
and a concentrated load (1 − α)F at a distance L/4 from the right end.
(a) Find FU as a function of α by means of both a kinematic and a

static analysis.
(b) Show how some reasonable upper and lower bounds can be found
by means of assumed mechanisms.
(c) Find the minimum value of MU as a function of α.
4. In a continuous beam like that of Figure 6.3.5, the loads F1 and F2

are uniformly distributed over the respective spans rather than concen-
trated. Plot an interaction diagram between F1 L/MU and F2 L/MU .
5. If the rate of deflection of a beam is denoted v(x) and the curvature

rate is κ̇ = v 00 (x), show how a Foulkes mechanism for a beam built in
at both ends can be generated by means of a deflection curve made up
of parabolic arcs. Find a minimum-weight design for an ideal sandwich
beam of uniform depth but variable flange area.
6. An asymmetric frame is shown in the adjacent figure. 2F

- L/2
(a) Assuming MU to be the same in all members, find F 6
- ?
L/2
6
FU . L ?
(b) Assuming only that the value of MU in the short ?

column is one-half of that in the long column, find
a minimum-weight design for the frame.
7. A two-story one-bay frame has the geometry and load-

ing shown in the adjacent figure, and is assumed to 3F
collapse under the loading. - L/2
2F - ?
6
(a) Find the value of MU if this value is the same in 2F L
all members. F- ? ?
(b) Find the value of MU if this is the value of the
6
L- L
ultimate moment in the lower columns and beam, ?
while in the upper columns and beam this value is
0.5MU .
(c) Assuming only that MU has one value (say MU 1 )
in the lower columns and beam and another value
(say MU 2 ) in the upper columns and beam, find
the minimum-weight design, and plot the design
diagram relating MU 1 /F L and MU 2 /F L.
Section 6.4 Limit Analysis of Plates and Shells

The concept of plate yielding was already developed in Article 5.2.3, with the
yield criterion given in terms of the moments Mαβ . As shown in 6.4.1, these
moments serve as the generalized stresses in the limit analysis of plates.
In shells (as in arches), the equilibrium of moments is coupled with mem-
brane forces (analogous to axial forces), and therefore both moments and
membrane forces appear as generalized stresses. With the greater number
of generalized stresses, the yield loci become more difficult to represent and
approximations leading to piecewise linear yield criteria often become nec-
essary. The theory is developed in 6.4.2, and examples are studied in 6.4.3.
Thorough coverage of the material in this section can be found in the book
by Save and Massonnet [1972].
6.4.1. Limit Analysis of Plates
Plastic Flow of Plates

With the stresses given by Equation (5.2.9), the plastic dissipation per
unit volume at plastic collapse (or, equivalently, under the hypothesis of
rigid–plastic behavior) is
4 4
Dp = σαβ ε̇αβ = − 2
Mαβ sgn x3 (−x3 κ̇αβ ) = 2 |x3 |Mαβ κ̇αβ ,
h h
where h is the plate thickness. Integrating through the thickness, we obtain
the plastic dissipation per unit area,
D̄p = Mαβ κ̇αβ . (6.4.1)
Section 6.4 / Limit Analysis of Plates and Shells 399
∗ that does not violate the yield

Now consider any stress distribution σαβ
∗
criterion and gives zero membrane forces. If the moments resulting from σαβ
∗
are Mαβ , then the inequality (3.2.4) expressing the principle of maximum
plastic dissipation can be integrated through the thickness, giving
∗
(Mαβ − Mαβ )κ̇αβ ≥ 0. (6.4.2)
The moments and the curvature rates are thus respectively the generalized
stresses and generalized strain rates for the analysis of the bending collapse of
plates. Since yield criteria in terms of moments have already been formulated
(see 5.2.2), the associated flow rule can be deduced from (6.4.2) to have the
form
∂f
καβ = λ̇ .
∂Mαβ
For the Mises, Tresca, and Johansen criteria, respectively, we obtain the
plastic dissipations per unit area as
q
D̄p (κ̇) = √2 MU κ̇2 + κ̇1 κ̇2 + κ̇2 (Mises),
3 1 2
1
D̄p (κ̇) = 2 MU (|κ̇1 | + |κ̇2 | + |κ̇1 + κ̇2 |) (Tresca),
D̄p (κ̇) = MU (|κ̇1 | + |κ̇2 |) (Johansen).
With moments replacing stresses and curvature rates replacing strain

rates, the theorems of limit analysis can be applied to the estimation of
collapse loads of plates as in any other problems. As with the plane problems
studied in the preceding section, much more use can be made of the upper-
bound theorem than of the lower-bound theorem.
Hinge Curves
A hinge curve or yield curve (hinge line or yield line if straight) is a curve
in the region A occupied by the middle plane, across which the slope of the
deflection w is discontinuous. Let ∆θ denote the change in slope encountered
along a line normal to the curve, and suppose that this change takes place
uniformly over a narrow zone of width δ. Within this zone, the numerically
larger principal curvature rate, say κ̇1 , is given by |κ̇1 | = ∆θ̇/δ, with |κ̇1 |
|κ̇2 |. If we integrate D̄p (κ̇) through the width of the zone, we√obtain, for the
plastic dissipation per unit length of the hinge curve, (2MU / 3)|∆θ̇| for the
Mises criterion and MU |∆θ̇| for the Tresca and Johansen criteria. The total
internal dissipation needed in limit analysis,
Z
Dint = Dp (ε̇)dV,
R
is therefore
2
Z q Z
Dint = √ MU 2 2
κ̇1 + κ̇1 κ̇2 + κ̇2 dA + |∆θ̇| ds (Mises),
3 Z A HCZ
1

Dint = MU (|κ̇1 | + |κ̇2 | + |κ̇1 + κ̇2 |)dA + |∆θ̇| ds (Tresca),
ZA 2 Z HC
Dint = MU (|κ̇1 | + |κ̇2 |) dA + |∆θ̇|ds (Johansen),
A HC
where HC denotes the hinge curve.
Comparing the Mises and Tresca results, we note that the expressions
inside the square brackets coincide if
1 q
(|κ̇1 | + |κ̇2 | + |κ̇1 + κ̇2 |) = κ̇21 + κ̇1 κ̇2 + κ̇22 ,
2
and this occurs if and only if one of three conditions κ̇1 = 0, κ̇2 = 0, or
κ̇1 + κ̇2 = 0 is met. Consequently, an upper-bound load obtained for a
Tresca plate on the basis of a velocity field obeying one of these conditions √
almost everywhere1 serves for a Mises plate as well if multiplied by 2/ 3.
Yield-line theory was developed by Johansen [1932] for the ultimate-
load design of reinforced-concrete slabs. In a polygonal plate, the yield
curves are yield lines and divide the plate into portions that move as rigid
bodies, so that all the dissipation takes place on the yield lines. If any edge
is clamped, then either the portion of the plate adjacent to it does not move,
or the edge itself becomes a yield line. In accordance with the upper-bound
theorem, R
the yield-line pattern must be such as to minimize Dint /Dext , where
Dext = A q ẇ dA is the external work rate.
Applications of Yield-Line Theory
As an example, consider a simply supported rectangular plate of dimen-
sions 2a×2b under a uniform load F/4ab. “Simply supported” is interpreted
in the traditional sense, that is, the plate deflection vanishes at the edges;
such a plate is called “position-fixed” by Johnson and Mellor [1973]. Accord-
ing to a weaker definition of simple support, used by Johnson and Mellor,
only downward deflection is prevented, and the plate is free to lift off.
For the traditional definition, the yield-line pattern may be one of the
two shown in Figure 6.4.1.
Actually, pattern (a) is a special case of pattern (b), with c = a, but
because of its relative simplicity it is instructive to study it separately. It can
easily be shown that, if v0pis the center velocity, then the slope-discontinuity
rate on the yield lines is (v0 /a)2 + (v0 /b)2 , so that, for a Tresca material,
r
1 1 a b
p
Dint = 4MU v0 a2 + b2 2
+ 2 = 4MU v0 + ,
a b b a
1
“Almost everywhere” means everywhere except on a set of points whose total area is
zero.
H
HH @
H @
HH @
6 HH @
b HH
H @
? H @
a - c-
(a) (b)
Figure 6.4.1. Yield-line patterns for a simply supported rectangular plate. Pat-
tern (a) is a special case of pattern (b), with c = a.
while Dext = 1thirdF v0 , so that the upper bound on F is 12MU (a/b + b/a).
For pattern (b), we have
c b v0 a b

Dint = 4MU v0 + + 2(a − c)MU ·2 = 4MU v0 +
b c b b c
and
F 1 1 1 c

Dext = v0 ·4bc + ·4b(a − c) = F v0 3 − .
4ab 3 2 6 a
The ratio Dint /Dext is a minimum at c given by
s 
2
c b b b
=  3+ − ,
a a a a
which can be seen to be less than one for all b/a < 1. Consequently,
s 2
2
8 a b b
F ≤ MU  3 + + 
3 b a a
which is less than the upper bound given by pattern (a) except in the case
of the square, when the two upper bounds are necessarily equal. The pre-
ceding results
√ hold for a plate made of a Mises material if MU is replaced
by 2MU / 3.
If the plate is clamped, then the same yield-line pattern as above can be
assumed, provided that the edges become yield lines. It can easily be verified
that the total dissipation along the edges is just equal to that on the interior
yield lines, so that the upper-bound load obtained by this method is twice
what it is for the simply supported plate. It is shown later, however, that a
better upper-bound load can be found for clamped plates in general.
A lower bound for the simply supported rectangular plate can be ob-
tained by assuming the statically admissible moment field
2 2
x1 x2 x1 x2

m11 = 1 − , m22 = 1 − , m12 = −λ ,
a b ab
where mαβ = Mαβ /MU , and λ is a constant to be determined. It can readily

be verified that the center and the midpoints of the edges yield according to
all three criteria. The equilibrium condition
F
mαβ ,αβ = −
4abMU
gives
a b

F = 8MU + +λ .
b a
The best lower bound is therefore obtained with the largest λ such that the
yield criterion is not violated anywhere. This occurs, for each of the three
yield criteria, if the criterion is satisfied at the corners, where m11 = m22 = 0
and |m12 | = λ. Thus the largest admissible √ value of λ turns out to be 1 for
the Johansen, 21 for the Tresca, and 1/ 3 for the Mises criterion. For a
square plate (a = b), we obtain upper-to-lower-bound ratios of 1, 1.2, and
1.344, respectively, for the three criteria. For the square Johansen plate, we
thus have the exact collapse load FU = 24MU . Upper and lower bounds
for uniformly loaded rectangular Johansen plates with various combinations
of simply supported and clamped edges have been calculated by Manolakos
and Mamalis [1986].
A plate in the shape of a regular polygon will, according to yield-line
theory, deform plastically into a pyramid, with yield lines on all diagonals.
If we think of a circle as the limit as n → ∞ of an n-sided polygon, then
it becomes clear that for a circular plate to undergo plastic flow (collapse),
the entire plate must become plastic — that is, the yield condition must be
met everywhere, as was assumed in Section 5.3.
Applications of circular-plate results
The results for the clamped circular plate derived in 5.2.3 may be ap-
plied to finding an upper-bound load for a clamped plate of arbitrary shape.
Consider the largest circle that can fit into the area A occupied by the plate;
we can then assume a velocity field such that this circle is a hinge circle
and the material inside it collapses like a clamped circular plate, while the
material outside it remains rigid. If, for example, the plate is uniformly
loaded and square, then the total load that would make the plate collapse
in this mode is 4/π times the load carried by the largest inscribed circle, or
(4/π) × 5.63 × 2πMU = 45.04MU ; this is less, and therefore a better upper
bound, than the one of 48MU given by yield-line theory.
If a clamped plate is carrying a single concentrated load F , then the
upper-bound collapse load of 2πMU is obtained for any inscribed circle cen-
tered at the load. It was shown by Haythornthwaite and Shield [1958] that
the moment field Mr = −MU , Mθ = 0 inside the circle can be extended
outside it without violating equilibrium or the yield criterion, so that 2πMU
is in fact the collapse load.
The same hinge-circle mechanism may be used to obtain 2πMU as the

upper bound for a concentrated load carried by a simply supported plate of
arbitrary shape. Since the moment field is Mr = 0, Mθ = MU inside the
circle, it can be continued outside it in a discontinuous but statically and
plastically admissible manner as Mr = Mθ = 0, so that 2πMU is the collapse
load in this case as well (see Zaid [1959]).
All the collapse loads for the axisymmetric Tresca plate may be used
as bounds on the corresponding loads for the Mises plate, by virtue of the
following reasoning: (1) Any moment field that is in equilibrium with the
load and that obeys the Tresca criterion represented by the largest hexagon
inscribed in the Mises ellipse (as in Figure 5.2.2, page 292), obviously does
not violate the Mises criterion, and therefore the load is a lower bound for
the Mises collapse load. (2) All the velocity fields associated with the sides of
the Tresca hexagon satisfy one of the conditions κ̇1 = 0, √κ̇2 = 0, κ̇1 + κ̇2 = 0;
consequently, the Tresca collapse load multiplied by 2/ 3 is an upper bound
for the Mises collapse load. Therefore, if fM and fT are the ultimate values
of f = F/2πMU for the Mises and Tresca plates, respectively, then
2
fT ≤ fM ≤ √ fT .
3
In order to obtain fM exactly, we must solve the quadratic equation
Mr2 − Mr Mθ + Mθ2 − MU2 = 0
for Mθ , making sure that the correct root is chosen, and substitute it in the
equilibrium equation. Let ρ = r/a, and let φ(ρ) be a function such that
the distributed load (assumed to be acting downward) is given by q(ρ) =
−(f MU /a2 )φ0 (ρ)/ρ; note that φ(0) = 0 and φ(1) = 1. The differential
equation for m = Mr /MU ,
r
dm 1 3
ρ + m− 1 − m2 = −f φ(ρ),
dρ 2 4
must be solved subject to the initial condition m(0) = 1, and fM is the value
of f for which the solution satisfies m(1) = 0 for a simply supported plate
and m(1) = −1 for a clamped plate.
If the plate is uniformly loaded, then φ(ρ) = ρ2 . A numerical solution of
the differential equation leads to fM = 3.26 for the simply supported plate
and fM = 5.92 for the clamped plate. These values may be compared with
the respective
√ lower bounds of 3 and 5.63, and the upper bounds of 3.46
(= 2 3) and 6.50. Other results relating to plate collapse may be found in
the books by Hodge [1959], Chapter 10; Hodge [1963]; Johnson and Mellor
[1973], Chapter 15; and Save and Massonnet [1972].
6.4.2. Limit Analysis of Shells: Theory

A shell is to a plate essentially as an arch is to a beam. As we saw in 6.2.2, the
use of a nonlinear P -M interaction locus makes the collapse analysis of arches
difficult, and in practice a piecewise linear locus is needed. This simplifica-
tion is all the more necessary for shells, in which more than one component
of both membrane force and moment must in general be accounted for. As
with one-dimensional members, a piecewise linear yield locus is generated
by giving the shell an ideal sandwich structure. Alternatively, the piecewise
linear locus may be viewed as an approximation to the “exact” nonlinear
one for a solid (or nonideal sandwich) shell.
The geometry of a shell is usually described by its middle surface, anal-
ogous to the middle plane of a plate, so that the two free surfaces are at a
distance h/2 from it. At any point of the middle surface, a local Cartesian
basis (ei ) may be formed such that e3 is perpendicular to the middle sur-
face, while e1 and e2 define its tangent plane. With respect to this basis,
the stress resultants Nαβ , Qα , and Mαβ may be defined in the same way as
for plates. The equilibrium equations they satisfy are, of course, different;
they require a global curvilinear coordinate system. The general theory of
shells is not presented here; only some special cases are studied.
The deformation of the middle surface can be described by means of the
strain tensor with components ε̄αβ , describing stretching, and the curvature
tensor with components καβ , describing bending. With shearing deformation
neglected, the internal virtual work per unit surface area is
Mαβ δκαβ + Nαβ δ ε̄αβ ,
and consequently, the Mαβ and Nαβ constitute the generalized stresses for
the most general shearless theory of limit analysis of shells; the shear forces
Qα are reactions and do not enter the yield locus.
If the mechanical behavior of the shell is isotropic in the tangent plane,
then the yield locus is expressible in terms of the principal moments M1 ,
M2 , and the principal membrane forces N1 , N2 . If, in addition, symmetry or
another constraint require one of the principal strains ε̄1 , ε̄2 or the principal
curvatures κ1 , κ2 to be zero, then the conjugate force or moment ceases
to be a generalized force and becomes an internal reaction instead; it can
therefore be eliminated from the yield locus, further reducing the number of
dimensions of the space in which the yield locus must be described.
Piecewise Linear Yield Locus

Following Hodge [1959], we derive the piecewise linear yield locus on the
basis of the ideal sandwich shell, made up of two thin sheets of thickness t
separated by a core of thickness h. The sheets are elastic–perfectly plastic
and obey the Tresca yield criterion with a uniaxial yield stress σY . The
principal stresses in the two sheets will be denoted σ1+ , σ2+ and σ1− , σ2− ,
respectively, the sign convention being chosen so that
1
Mα = (σα− − σα+ )ht, α = 1, 2,
2
and
Nα = (σα− + σα+ )t, α = 1, 2.
Consequently,
σα± = N 2t
α ∓ Mα .
ht
If the principal stresses σα± are not to violate the Tresca yield criterion,
they must satisfy the six inequalities
|σ1± | ≤ σY , |σ2± | ≤ σY , |σ1± − σ2± | ≤ σY .
In order to express these inequalities in terms of the Mα and Nα , we define

MU = σY ht and NU = 2σY t, as well as the dimensionless quantities mα =
Mα /MU , nα = Nα /NU . We thus obtain
|m1 − n1 | ≤ 1, |m1 + n1 | ≤ 1,
|m2 − n2 | ≤ 1, |m2 + n2 | ≤ 1, (6.4.3)

|m1 − m2 + n1 − n2 | ≤ 1, |m1 − m2 − n1 + n2 | ≤ 1.
The six absolute-value inequalities (6.4.3) are equivalent to twelve algebraic
inequalities, so that the yield locus is bounded by twelve hyperplanes in the
four-dimensional m1 m2 n1 n2 -space.
If one of the quantities mα , nα represents an internal reaction rather
than a generalized stress, then it can be eliminated from the yield locus.
Suppose this quantity to be m2 (i.e., suppose the shell to be constrained so
that κ2 = 0); then the inequalities involving m2 may be rewritten as
−1 + n2 ≤ m2 ≤ 1 + n2 , − 1 − n2 ≤ m2 ≤ 1 − n2 ,
−1 + m1 + n1 − n2 ≤ m2 ≤ 1 + m1 + n1 − n2 , (6.4.4)
−1 + m1 − n1 + n2 ≤ m2 ≤ 1 + m1 − n1 + n2 .
The actual values of m2 do not matter, as long as some m2 can be found so
that all the inequalities (6.4.4) can be satisfied, and this is the case whenever
the first member of each inequality is no greater than the third member of
every inequality. The first two inequalities give
|n2 | ≤ 1, (6.4.5a)
and the second two give

|n1 − n2 | ≤ 1. (6.4.5b)
Combining inequalities from the first and second pairs leads to the following
additional nontrivial inequalities:
|2n2 − n1 + m1 | ≤ 2, |2n2 − n1 − m1 | ≤ 2. (6.4.5c-d)
Together with (6.4.3)1,2 , we thus have six absolute-value inequalities involv-
ing m1 , n1 and n2 , or a total of twelve algebraic inequalities. The yield
locus is therefore a dodecahedron in the three-dimensional m1 n1 n2 -space, il-
lustrated in Figure 6.4.2. The derivation of this locus is due to Hodge [1954],
n2
6
@
@ n1
@ B
@ B
B B
B
B B

@
@ B
B
B - m1
B
B
B
B
B
B

@
@
Figure 6.4.2. Piecewise linear yield locus for a cylindrical shell (from Prager
[1959]).
who also derived the exact nonlinear yield locus for a solid shell made of uni-
form material. As discussed previously, the piecewise linear locus may be
thought of as an approximation to the exact one when the appropriate values
of MU and NU are used, namely MU = σY h2 /4 and NU = σY h, with σY the
uniaxial yield stress of the solid-shell material.
The plastic dissipation per unit area of the ideal sandwich shell, given
rigid–plastic behavior, is
σY t + − − − −

D̄p = |ε̇1 | + |ε̇+
2 | + |ε̇ +
1 + ε̇ +
2 | + |ε̇ 1 | + |ε̇ 2 | + |ε̇ 1 + ε̇ 2 | .
2
But
h
ε̇± ˙
α = ε̄α ∓ κ̇α , α = 1, 2.
2
Consequently,
1
D̄p = [|NU ε̄˙ 1 + MU κ̇1 | + |NU ε̄˙ 1 − MU κ̇1 |
4
+ |NU ε̄˙ 2 + MU κ̇2 | + |NU ε̄˙ 2 − MU κ̇2 |
+ |NU (ε̄˙ 1 + ε̄˙ 2 ) + MU (κ̇1 + κ̇2 )| + |NU (ε̄˙ 1 + ε̄˙ 2 ) − MU (κ̇1 + κ̇2 )|] .
(6.4.6)
((
E E
E E ((
((( E E 6 iP u0 (z)
P

E E E
u(z) E - E((
EE
(( (E
E E w(z)-
E E
E
E
E E
E
E
a E dθ
E E
E E
E E
E E
E E
(
E(( E r z
Figure 6.4.3. Displacements in a radially loaded cylindrical shell.
Yield criteria that are appropriate for the approximate treatment of

reinforced-concrete shells are discussed by Save and Massonnet [1972], Chap-
ter 9.
6.4.3. Limit Analysis of Shells: Examples
Radially Loaded Cylindrical Shell: Basic Equations

As a first example of the application of the theory discussed above, we
consider a circular cylindrical shell of uniform mean radius a and thickness h,
subject to an axisymmetric radial pressure distribution that may vary along
the axial coordinate z. Because of axial symmetry the displacement of the
middle surface has only the radial component u(z) and the axial component
w(z). The displacements throughout the shell are assumed to be governed
by the “plane sections remain plane” hypothesis applied to a longitudinal
strip subtending a small angle dθ (see Figure 6.4.3):
ur (r, z) = u(z), uθ (r, z) = 0,
uz (r, z) = w(z) − (r − a)u0 (z).

In accordance with Equations (1.2.1), the only nonvanishing strain compo-
nents are
u . u
εθ = =
r a
and
εz = w0 − yu00 ,
where y = r − a. The internal virtual work per unit area is

Z h/2 δu
(σθ δεθ + σz δεz ) dy = Nθ + Nz δw0 + Mz δu00 ,
−h/2 a
where
Z h/2 Z h/2 Z h/2
Nθ = σθ dy, Nz = σz dy, Mz = − yσz dy.
−h/2 −h/2 −h/2
The generalized stresses are thus three in number. The conjugate generalized
strains are ε̄θ = u/a, ε̄z = w0 , and κz = u00 . The fact that κθ = 0 (i.e., circles
remain circles) removes Mθ from the rank of generalized stresses. The yield
locus is therefore given by Equations (6.4.3)1,2 and (6.4.5a)–(6.4.5d), with
M1 = Mz , N1 = Nz , and N2 = Nθ .
The plastic dissipation per unit area is
|u̇| ẇ0 ẇ0

1

+ MU u̇00 + NU − MU u̇00

D̄p = 2NU + N U
4 a a a (6.4.7)
u̇ u̇

0 00
+ ẇ0 − MU u̇00 .

+ NU + ẇ + MU u̇ + NU
a a
The equilibrium equations can be obtained by applying the principle of

virtual work. If the shell axis occupies the interval −L ≤ z ≤ L, then the
internal virtual work is
Z L δu
δW int = 2πa (Nθ + Nz δw0 + Mz δu00 ) dz.
−L a
Through integration by parts this becomes
n L
δW int = 2πa (Mz δu0 − Mz0 δu + Nz δw)−L
Z L
Nθ

+ Mz00 + δu − Nz0 δw dz .
−L a
In addition to the radial pressure p (positive outward), let the shell be loaded
by a bending moment Mz+ , an axial force Nz+ , and a shear force Q+ r , all per
unit length of circumference, at z = L, and similarly, Mz , Nz , and Q−
− −
r at
z = −L. The external virtual work is therefore
"Z
L
δW ext = 2πa p δu dz + Nz+ δw(L) + Mz+ δu0 (L)
−L
− − 0 −
+ Q+

r δu(L) −Nz δw(−L) − Mz δu (−L) − Qr δu(−L) .
Equating the internal and external virtual work leads to the equilibrium
equations
Nθ
Nz0 = 0, Mz00 + =p (6.4.8)
a
and the boundary conditions
(Mz − Mz± ) δu0 = 0, (Mz0 + Q±

r ) δu = 0, (Nz − Nz± ) δw = 0, z = ±L.
Note that the axial end forces must be equal and opposite for equilibrium.
If these are zero, then Nz = 0 everywhere.
For the shell without end load, the piecewise linear yield locus reduces
to a hexagon in the Mz Nθ -plane, formed by the three pairs of parallel lines
described by
|m| = 1, |2n + m| = 2, |2n − m| = 2, (6.4.9)
where m = Mz /MU and n = Nθ /NU (see Figure 6.4.4, which also shows the
“exact” nonlinear yield locus as well as the simplified square locus given by
|m| ≤ 1, |n| ≤ 1).
n
B
G H
A C
F D
J E I
Exact
Approximate
Figure 6.4.4. Yield loci for a cylindrical shell without end load.
The plastic dissipation for the yield criterion described by (6.4.9) may
be obtained from (6.4.7) by substituting, as in uniaxial stress, ε̄˙ 1 = − 12 ε̄˙ 2 ,
or ẇ0 = −u̇/2a. The result is
|u̇|

1 u̇ u̇

+ MU u̇00 + NU − MU u̇00 .

D̄p = NU + N U (6.4.10)
2 a 2a 2a
When the state of generalized stress can be assumed to be represented by
points on one of the inclined lines AB, BC, DE, and EF of Figure 6.4.4,
then Nθ can be expressed in terms of Mz , and the result substituted in the
equilibrium equation (6.4.8)2 . The equation then becomes a linear differen-
tial equation for Mz , which can be solved subject to appropriate boundary
conditions. Since the normality rule gives 2MU u̇00 = ±NU u̇/a along the
aforementioned lines, the dissipation is just NU |u̇|/a. The same result ob-
tains at the vertices B and E. Along the lines AF and CD, normality
requires u̇/u̇00 = 0. A nonvanishing velocity thus implies u̇00 = ±∞, that is,
the formation of a plastic hinge circle.
A simple case occurs when the pressure p is constant and the ends of the
tube are free. Then Mz = 0, and the yield criterion reduces to |Nθ | = NU .
The ultimate pressure is therefore given by |pU | = NU /a.
A solution for a pressurized tube with clamped ends was derived by
Hodge [1954]. Here the central portion of the tube is in regime DE, and
the outer portions in EF ; the boundary is at z = ±ηL. The parameter η
and the dimensionless ultimate pressure p̄ = pU a/NU are given implicitly as
functions of the dimensionless shell parameter ω, defined by
NU L2
ω2 = .
2MU a
The results are

sin ω(1 − η)
sinh ωη = √ ,
2 cos ω(1 − η) + 1
2 − cos ω(1 − η)
p̄ =
2[1 − cos ω(1 − η)]
for ω ≤ 1.65, and
tan ω(1 − η) = coth ωη,
1
p̄ = 1 +
2(2 cosh ωη − 1)
for ω > 1.65. For more details, see Hodge [1959], Section 11-2.
Cylindrical Shell with a Ring Load

As another example, suppose that a free-ended shell is subject to a ra-
dially inward ring loading at z = 0, its intensity being F per unit length of
circumference; that is,
p(z) = −F δ(z),
where δ(·) is the Dirac delta function. Note that
1 d 1 d2
δ(z) = sgn z = |z|.
2 dz 2 dz 2
It is reasonable to assume initially that the hoop stress is compressive,

that is, Nθ ≤ 0. If the shell is extremely short, then its collapse should not
depend very much on whether the force is applied around a ring or uniformly
distributed over the surface in a statically equivalent manner. For very short
.
shells, then, we should expect FU = 2NU L/a, with bending having little or
no effect.
When bending is taken into account, it can be seen that at least in a
central portion of the tube, Mz ≥ 0. It will be assumed, to begin with, that
the entire shell is plastic and in regime AB of Figure 6.4.4. Eliminating Nθ ,
we obtain the following dimensionless differential equation for m(ζ), using
the dimensionless variable ζ = z/L, and the parameters f = F a/(2NU L)
and ω as previously defined:
m00 (ζ) + ω 2 m(ζ) = 2ω 2 − 4ω 2 f δ(ζ).
The general solution of this equation that is even in ζ is
m(ζ) = 2 − 2ωf sin(ω|ζ|) + C cos ωζ,
with C an arbitrary constant. The free-end condition m(1) = m0 (1) = 0
yields C = −2 cos ω and
sin ω
f= . (6.4.11)
ω
The solution may accordingly be written as
m(ζ) = 2[1 − cos ω(1 − |ζ|)].
The requirement that 0 ≤ m ≤ 1 limits the validity of this solution to
ω ≤ π/3. For sufficiently short shells, then, Equation (6.4.11) gives a lower
bound to the collapse load; in particular, the limit as ω → 0 is f = 1, as
previously discussed. An associated kinematically admissible velocity field,
however, can easily be found, namely,
u̇(ζ) = −v0 cos ωζ. (6.4.12)
It can readily be checked that the generalized strain rates derived from this
velocity field satisfy the normality condition for regime AB, and hence Equa-
tion (6.4.11) gives the collapse load for ω ≤ π/3.
Alternatively, the upper-bound theorem can be applied directly to the
velocity field (6.4.12), the dissipation per unit length being 2πNU |u̇|, and
the external work rate 2πaF v0 . The result is Equation (6.4.11) as an upper
bound, as long as u̇(ζ) does not change sign; this condition is met for ω ≤
π/2. Equation (6.4.11) is therefore an upper bound for π/3 ≤ ω ≤ π/2.
However, a better upper bound can be obtained for this range by assuming
a velocity field with a plastic hinge circle at ζ = 0, namely,
u̇(ζ) = −v0 (cos ωζ + β sin ω|ζ|). (6.4.13)
The additional dissipation at the hinge circle is 4πaMU βv0 /L = 2πNU Lβv0 /ω 2 .
Equating dissipation and external work rate yields
β
f = sin ω − (1 − 2 cos ω)
2ω 2
if u̇ does not change sign, that is, if β ≤ ω cot ω. Choosing this limiting
value for β gives the smallest value of f , namely,
2 − cos ω
f= ,
2ω sin ω
which is less than the right-hand side of (6.4.11) for π/3 < ω < π/2.
A lower bound for ω > π/3 can be obtained by assuming the solution
m(ζ) = 2 − cos ωζ − 2ωf sin(ω|ζ|)
valid for |ζ| < η, where η is such that m(η) = m0 (η) = 0, and continuing
= 0, Nθ = 0 for |ζ| > η. The conditions at η lead to
it statically as Mz √
ωη = π/3 and f = 3/(2ω), or, in dimensional form,
s
6MU NU
F = ,
a
a result that can be seen to be independent of the length. As can be seen,

however, as the length increases, so does the discrepancy between the upper
and lower bounds.
For longer shells the solution based on the hexagonal yield locus becomes
difficult. For an infinitely long shell, Drucker [1954b] found the collapse load
s
3MU NU
F =2 (6.4.14)
a
p
or f = 3/2/ω. This load is based on the following moment distribution:
√
6
m(ζ) = 2 − cos ωζ − sin ω|ζ|, 0 ≤ |ζ| ≤ ζ1 (AB),
ω
= −2 + cosh ω(|ζ| − ζ2 ), ζ1 ≤ |ζ| ≤ ζ3 (BC),
= −2 + 2 cos ω(|ζ| − ζ4 ), ζ3 ≤ |ζ| ≤ ζ4 (DE),
= 0, ζ4 ≤ |ζ| (rigid),
where
√
−1 2+3 2
ωζ1 = cos = 0.469, ω(ζ2 − ζ1 ) = cosh−1 2 = 1.317,
7
1 1 1
ω(ζ3 − ζ2 ) = cosh−1 4 = 1.032, ω(ζ4 − ζ3 ) = cos−1 = 0.659.
2 2 4
It follows that ωζ4 = 3.477, and therefore the result is valid for ω ≥ 3.477.
Note that m(0) = 1 and m(ζ2 ) = −1, so that plastic hinges develop at those
points. At ζ3 the hoop stress Nθ changes abruptly from a negative to a
positive value, and at ζ4 back to zero.
As pointed out by Drucker [1954b], the mathematics is greatly simplified

if the the yield locus is replaced by the rectangle |Mz | ≤ MU , |Nθ | ≤ NU .
With this yield criterion, Eason and Shield [1955] found complete solutions
for shells of all lengths, and with the load not necessarily at the center. Since
the rectangle circumscribes the hexagon, the collapse loads found by Eason
and Shield are upper bounds on those that would be found for the hexagonal
yield locus. Furthermore, a rectangle with vertices at (±MU0 , ±NU0 ) may be
inscribed in the hexagon; then the collapse load for the rectangle is a lower
bound for the hexagon when MU and NU are respectively replaced by MU0
and NU0 . The values of MU0 and NU0 may be chosen to as to maximize the
lower bound.
With the rectangular yield locus, both kinematic and static solutions are
quite easy. The sides |Mz | = MU of the rectangle (like those of the hexagon)
correspond to hinge circles, while the sides |Nθ | = NU describe velocities u̇
varying linearly with position. For each side, Mz and Nθ are polynomial
functions of z.
For the symmetric problem under consideration, Eason and Shield’s re-
sults are
1 1

√
f= 2
+ 1 , ω ≤ 1 + 2,
2 ω
√
2 √
f= , ω ≥ 1 + 2.
ω
The latter result is equivalent to
s
MU N U
F =4 , (6.4.15)
a
independent of the length. Equation (6.4.15) also gives the lower bound
s
MU0 NU0
F =4 ,
a
where
MU0 N0
+ 2 U ≤ 2, MU0 ≤ MU .
MU NU
The 0 0
p lower bound is maximized for MU = MU , NU = 2 NU , and equals
1
2 2MU NU /a. The bounds thus bracket Drucker’s collapse load (6.4.14) for
the long shell.
Spherical Cap Under Pressure
In shells of revolution, as a rule, all four generalized stresses are active.
An approximate theory, in which the effects of one of them are ignored, was
proposed by Drucker and Shield [1959], but its applicability is limited.
It was shown by Hodge [1959] that for a shell of revolution, a point
on the four-dimensional piecewise linear yield locus (6.4.3) corresponds to
plastic deformation only if it lies on the intersection of two of the twelve

hyperplanes, one of which represents yielding of the top sheet and the other
represents yielding of the bottom sheet. The generalized stresses thus satisfy
two yield equations and two equilibrium equations, making the problem
“statically determinate.” Furthermore, there are two normality conditions
on the generalized strain rates, and since the latter are derived from two
velocity components, the problem is “kinematically determinate” as well.
We consider here a spherical cap of radius a subtending a half-angle
φ0 , clamped around its edge, and carrying a uniform radial pressure p (see
Figure 6.4.5). The problem was treated by Onat and Prager [1954] on the
basis of a nonlinear yield criterion and by Hodge [1959, Section 11-6] on the
basis of the piecewise linear yield criterion (6.4.3).
p
C ?
AUA N CW

JJ
^ BMB

J B
J B - φ
J Ba φ0
J s
J B
J B
JB
JB
JB
Figure 6.4.5. Spherical cap under uniform external pressure: geometry and
loading.
If the radial displacement is ur = u and the meridional displacement is

uφ = v, then the generalized strains are
u + v cot φ u + v0
ε̄θ = , ε̄φ = ,
a a
u0 − v u00 − v 0
κθ = cot φ , κφ = ,
a2 a2
where (·)0 = d(·)/dφ.
To find an upper bound on the collapse pressure, we assume the velocity
field
u̇ = −v0 (cos φ − cos φ0 ), v̇ = 0.
The generalized strain rates are accordingly
v0 v0
ε̄˙ θ = ε̄˙ φ = − (cos φ − cos φ0 ), κ̇θ = κ̇φ = cos φ.
a a2
The plastic dissipation per unit area is then, from Equation (6.4.6),
D̄p = NU v0 [| cos φ0 − (1 − k) cos φ| + |(1 + k) cos φ − cos φ0 |]

(
2NU v0 (cos φ − cos φ0 ), 0 ≤ φ ≤ φ∗ ,
=
2kNU v0 cos φ, φ∗ ≤ φ ≤ φ0 ,
where
MU cos φ0
k= , cos φ∗ = .
NU a 1−k
The expression for φ > φ∗ is necessary, of course, only if φ0 > φ∗ .
In addition, since u̇0 does not vanish at the edge φ = φ0 , a plastic hinge
circle forms there, necessitating the additional dissipation (per unit length)
given by MU v0 sin φ0 /a = kNU v0 sin φ0 . The total internal dissipation is
thus Z φ0
Dint = 2πa2 D̄p sin φ dφ + 2πakNU v0 sin2 φ0 .
0
The external work rate is

Z φ0
2
Dext = 2πa pu̇ sin φ dφ = πa2 pv0 (1 − cos φ0 )2 .
0
Setting Dext = Dint yields the upper bound

" 2 #
NU 1 + cos φ0 k2 cos φ0

p=2 1+k + , cos φ0 ≤ 1 − k,
a 1 − cos φ0 1 − k 1 − cos φ0
NU 1 + cos φ0
p = 4k , cos φ0 ≥ 1 − k.
a 1 − cos φ0
A lower bound may be obtained by assuming that the stress field in

the shell is one of simple membrane compression, Nθ = Nφ = −NU , with
Mθ = Mφ = 0. The corresponding lower-bound pressure is p = 2NU /a.
A lower bound that is better for sufficiently small cap angles was found by
Hodge [1959] by assuming Nθ , Nφ , Mθ and Mφ to be sinusoidally varying
functions of φ, substituting into the equilibrium equations, and choosing the
free coefficients so as to maximize the pressure subject to the yield inequal-
ities. The resulting best lower bound is given by
 s 
2 2
pa 1 1−k 1 − cos φ0

=2+ 1 − +4 .
NU 1 − cos φ0 1+k 1 + cos φ0
Clearly, this result is an improvement over the previous one if and only if
the quantity under the square-root sign is less than unity.

1. Using yield-line theory, find upper bounds for a uniformly loaded, sim-
ply supported Tresca plate having the shape of (a) an equilateral tri-
angle, (b) a right isosceles triangle, and (c) a regular hexagon.
2. Repeat Exercise 1 for clamped plates.
3. Find lower bounds for some of the plates in Exercises 1 and 2.
4. Using yield-line theory, find upper bounds for the plates of Exercises
1 and 2 when they carry a single concentrated load at the center or
centroid. Compare with the result FU = 2πMU .
5. A square Tresca plate carrying a uniform downward load is simply

supported along its edges against downward deflection but is free to
lift off. Using yield-line theory, find an upper bound to the collapse
load.
6. Using some of the methods of Section 6.2, find lower and upper bounds
for the yield locus of an ideal sandwich shell obeying the Mises criterion
and its associated flow rule (a) when M2 is not a generalized stress and
(b) when, in addition, N1 = 0.
7. Find the complete solution for the clamped pressurized cylindrical shell
with clamped ends obeying the “square” yield locus of Figure 6.4.4.
Plot the nondimensional ultimate pressure p̄ = pU a/NU against the
shell parameter ω.
8. Repeat Exercise 7 for a shell that is (a) clamped at one end and free
at the other and (b)simply supported at both ends.
9. Repeat Exercise 8 for the hexagonal yield locus.
10. For a short free-ended cylindrical shell subject to a noncentered ring

load, find the collapse load based on the “square” yield locus.
Chapter 7
Dynamic Problems
Section 7.1 Dynamic Loading of Structures
7.1.1 General Concepts
All the problems studied so far have been static or quasi-static. In any
complete solution of such a problem, the stress field satisfies the equilibrium
equation with the prescribed body force and the static boundary conditions
with the prescribed surface tractions. The effects of inertia are neglected.
In a body made of a standard perfectly plastic material, the ultimate
loading is the greatest loading under which a solution to the static problem
can be found such that the yield criterion is nowhere violated. If a loading in
excess of the limit loading is applied, then, obviously, the static problem (in
which, by definition, inertia is ignored) has no solution, and inertia effects
must be taken into account. If the time of loading is short, enough of the
external work may be transformed into kinetic energy so that excessive defor-
mation is prevented; for example, when a nail is struck by a hammer, it may
experience a force in excess of its static ultimate load without permanent
deformation.
The problem of impact or impulsive loading of structural elements such
as beams, plates and shells has been most often treated within the con-
stitutive framework of limit analysis: rigid–perfectly plastic behavior, with
the yield criterion in terms of generalized stresses. This approach, which
has been reasonably successful with regard to the determination of perma-
nent deformations, is presented in this section; it is generally regarded as
justified when the energy imparted to the body greatly exceeds the elastic
energy that can be stored. If local constitutive equations are used, waves
propagate through the body. The propagation of one-dimensional and mul-
tidimensional waves in elastic-plastic bodies is studied in Sections 7.2 and
7.3, respectively.
417
418 Chapter 7 / Dynamic Problems
It must be pointed out at the outset that the solution of dynamic prob-
lems on the basis of rate-independent plasticity is in conflict with the def-
inition of rate-independent plastic behavior as the limiting behavior of vis-
coplastic bodies at very slow rates (see 3.1.2). The use of rate-independent
plasticity for such problems is based on the assumption of a “dynamic” yield
stress that is independent of rate in the range of rates encountered in the
dynamic problem, but not, in general, identical with the static yield stress.
The adequacy of this procedure, as opposed to the use of viscoplasticity the-
ory, has long been the subject of debate and is discussed in 7.2.3. Until then,
the rate-independent model will be tacitly assumed.
Dynamic Behavior of Rigid–Perfectly Plastic Bodies:
General Results
Most of the structures for which static collapse loads have been deter-
mined on the basis of the rigid–perfectly plastic model have also been subject
to dynamic analysis; the solutions have been surveyed by Krajcinovic [1973]
and Jones [1975]. A common feature of the solutions is that if the body is
restrained against rigid-body motion, then the velocity field eventually takes
on a mode form in the sense that its spatial variation becomes independent of
time. In other words, the velocity becomes the product of a time-dependent
amplitude and a function of position:
˙
v(x, t) = ∆(t)w(x).
A proof that all solutions must converge to mode form is due to Martin
[1980]; it is based on an extremum principle,
√ according to which the final
mode is the one that minimizes Dint / K, where Dint is the total plastic
dissipation and K is the kinetic energy.
The fact that at least the final phase of the motion is in mode form
means that all points of the body come to rest at once. If the time at which
this occurs is ts , then the permanent displacement field up (x) can be found
by integrating the velocity over time from t = 0 to t = ts .
A theorem due to Martin [1975] gives a lower bound on ts for a body on
which all prescribed loads and prescribed surface velocities are zero for t >
t0 . Let v∗ be any kinematically admissible velocity field, with a strain-rate
field ε̇∗ derived from it. When the body force, prescribed surface traction
and prescribed surface velocity are all zero, the dynamic principle of virtual
velocities reduces to
Z Z
∗
− ρv̇ · v dV = σij ε̇∗ij dV,
R R
where v is the actual velocity field and σ is the actual stress field. Combining
this result with the maximum-plastic-dissipation postulate, we obtain
Z Z
− ρv̇ · v∗ dV ≤ Dp (ε̇∗ ) dV.
R R
Section 7.1 / Dynamic Loading of Structures 419
The right-hand side of this inequality is independent of time, while on the

left-hand side only v̇ depends on time. If v0 (x) = v(x, t0 ), then, since
v(x, ts ) = 0, integration of both sides from t0 to ts leads to
Z Z
∗
ρv0 · v dV ≤ (ts − t0 ) Dp (ε̇∗ ) dV.
R R
Consequently,
v0 · v∗ dV
R
ts − t0 ≥ R
R ∗ . (7.1.1)
R Dp (ε̇ ) dV
Another result due to Martin [1975] governs the permanent displacement
field up (x) = u(x, ts ). If σ ∗ is any statically and plastically admissible stress
field, then the static principle of virtual work implies that
Z Z
∗ ∗
σij vi nj dS = σij ε̇ij dV,
∂R R
while the dynamic principle of virtual velocities gives

Z Z
− ρv̇ · v dV = σij ε̇ij dV.
R R
Again using the maximum-plastic-dissipation postulate, we obtain

d
Z Z
∗
σij vi nj dS ≤ − ρv̇ · v dV = − K,
∂R R dt
where
1
Z
K= ρv · v dV.
2 R
Since K(ts ) = 0, integration of both sides of the inequality leads to
Z
∗
σij (upi − u0i )nj dS ≤ K0 , (7.1.2)
∂R
where u0 is the displacement field at t0 . In the case of impulsive loading

lasting a very short time, u0 can be taken as zero.
The preceding results can easily be formulated in terms of generalized
stresses and strains.
7.1.2 Dynamic Loading of Beams
Equations of Motion for Elastic–Plastic Beams

The equations of motion of a beam according to the elementary (Euler–
Bernoulli) theory can be obtained by adding inertial forces to the distributed
load. We consider only a beam of doubly symmetric cross-section, with the
centroidal axis along the x-axis, and with bending confined to the xy-plane.
If the deflection in the positive y-direction is u(x, t), then the inertial force
per unit length is −ρ̄∂ 2 u/∂t2 , where ρ̄ = ρA is the beam mass per unit
length, ρ being the mass density and A the cross-sectional area. With q
denoting the distributed load per unit length, the equation of motion is
∂2M ∂2u
= q − ρ̄ . (7.1.3)
∂x2 ∂t2
When q = 0, and a functional relation is assumed between the moment
M and the curvature κ = ∂ 2 u/∂x2 , it can be shown that the deflection can
be expressed as
u(x, t) = tf (η),
where η = x2 /t, and that κ (and hence M ) as well as the velocity v = ∂u/∂t
are functions of the single variable η alone. If, say, κ = Ψ(M ), then the
moment is governed by the differential equation
d2 √ dM ρ̄ 0 √ dM

η + Ψ (M ) η = 0. (7.1.4)
dη 2 dη 16 dη
For a linearly elastic beam, Ψ0 (M ) = 1/EI, so that the equation is lin-
ear; a general solution is due to Boussinesq. For an elastic–plastic beam,
Equation (7.1.4) is in general nonlinear, and must be solved numerically.
It is, however, piecewise linear if a moment-curvature relation with linear
work-hardening is assumed, and an analytic solution can then be obtained.
The first such solution was found by Bohnenblust (see Duwez, Clark, and
Bohnenblust [1950]), who treated an infinitely long beam in which one point
is suddenly given a velocity that remains constant in time — equivalent to
being struck by a very heavy concentrated mass traveling at that velocity
(so heavy that the resistance of the beam is insufficient to decelerate it), or
as in the experiments that were performed by Duwez et al., to the beam
itself moving transversely as a rigid body before it impinges on a concen-
trated rigid obstacle. The deformation
p remains elastic if the impact velocity
v0 does not exceed vE = κE EI/ρ̄, where κE is the elastic-limit curvature
as defined in 4.4.1. When v0 > vE and the hardening is linear, different
deformation patterns develop for vE < v0 < 2.087vE and for v0 > 2.087vE .
The analogous problem for rigid–plastic beams, with various types of
work-hardening and with extensions to variable velocity and to semi-infinite
beams, was treated by Conroy [1952, 1955, 1956, 1963]. Equation (7.1.4) for
such a beam is valid only in the plastic portions of the beam. In the rigid
portions, the moment is governed by Equation (7.1.3) with q = 0 and with
the deflection given by a straight-line function of x, say u(x, t) = f (t)+g(t)x.
The boundary between the rigid and plastic portions is itself an unknown
function of time. It was shown by Conroy [1952] how the plastic por-
tions shrink to points, representing plastic hinges, in the limit as the work-
hardening decreases to zero. For linear hardening, Conroy [1955] showed
that certain problems can be solved by means of linear elastic analysis.
Impact Loading of a Perfectly Plastic Free-Ended Beam

The impact loading of a free-ended beam by a suddenly applied force
that is a prescribed function of position and time was studied by Lee and
Symonds [1952] for a rigid–perfectly plastic beam loaded at the center and by
Bleich and Salvadori [1953, 1955] for an elastic–perfectly plastic beam under
a loading that is symmetric about the midpoint but is otherwise arbitrary.
We consider, first, the problem treated by Lee and Symonds [1952]. A
free-ended beam of length 2L is suddenly loaded at its midpoint by a con-
centrated force F (t), which rises to a peak value Fm and declines to zero in a
short time. Let ∆(t) denote the midpoint displacement. It can be assumed
that at first the beam is accelerated as a rigid body, so that the displacement
is ∆ everywhere, and the load F is balanced by a uniformly distributed iner-
¨ with 2ρ̄L∆
tial force ρ̄∆, ¨ = F [see Figure 7.1.1(a)]. The bending moment at
a distance x from the midpoint may be obtained from the free-body diagram
of Figure 7.1.1(b), and is
F (t)
M (x, t) = (L − x)2 ;
4L
the maximum moment is M (0, t) = F (t)L/4. The assumption of rigid-
body motion is therefore valid as long as this moment is less than MU , or
F (t) < 4MU /L.
F L–x -
? M
x
6666666666666666666 ?6666666
¨ = F
ρ̄∆
F
2L 2L
2L -
(a) (b)
F F/2
hhhh ((( (( (
((( (((
w w
h
hhh hhh(? (
( ? ( (
hhh (( (( ( θ (
(( ( ( ( x
? E
E E E

EE E
EE E
¨
ρ̄(∆–xθ̈)
(c)
(d)
Figure 7.1.1. Impact loading of a perfectly plastic free-ended beam: (a) geom-
etry and loading; (b) free-body diagram; (c) hinge rotation; (d)
motion of one-half of the beam.
When F (t) ≥ 4MU /L, a plastic hinge must form at x = 0. Let the
half-angle of rotation of the hinge be θ(t), as in Figure 7.1.1(c). Because of
symmetry, it is sufficient to study the motion of half the beam, say 0 < x < L
[see Figure 7.1.1(d)]. Since the deflection is
u(x, t) = ∆(t) − xθ(t)
as long as θ(t) 1, the global force equation of motion may be obtained by

balancing the force F/2 acting on half the beam with the total inertial force:
F L 2
¨ − ρ̄ L θ̈.
Z
= ρ̄ ¨ − xθ̈)dx = ρ̄L∆
(∆
2 0 2
Similarly, the moment equation of motion is obtained by balancing the mo-
ment MU with the moment about x = 0 of the inertial force:
L L2 ¨ L3
Z
MU = ρ̄ ¨ − xθ̈)x dx = ρ̄
(∆ ∆ − ρ̄ θ̈.
0 2 6
The linear and angular accelerations are, respectively,
¨ = 2MU (φ − 3),
∆
ρ̄L2
3MU
θ̈ = (φ − 4),
ρ̄L3
where φ(t) = F (t)L/MU . The distribution of bending moment is now
2
x 1 x

M (x, t) = MU 1 − 1 − [φ(t) − 4] .
L 2 L
It can easily be shown that this distribution has a local extremum in the
interior of the beam when φ(t) > 6, and that this extremum is located at
x/L = φ(t)/3[φ(t) − 4] and has the value −2MU [φ(t) − 6]3 /27[φ(t) − 4]2 .
This value equals −MU when φ reaches 22.9; an additional plastic hinge
then forms in each half of the beam. For F (t) ≥ 22.9MU /L, therefore, the
equations of motion must reflect the additional hinges, which move in toward
the middle as the load increases. Further results relative to this problem
were obtained by Symonds and Leth [1954] and by Cotter and Symonds
[1955]. The effect of distributed rather than concentrated loads was studied
by Salvadori and DiMaggio [1953] and by Seiler and Symonds [1954] (see
also Seiler, Cotter and Symonds [1956]). An extension of the method to
elastic–perfectly plastic beams is due to Alverson [1956].
Bleich and Salvadori [1953, 1955] examined the motion of an impulsively
loaded elastic–perfectly plastic beam by using the natural modes of vibration
of elastic beams — for the entire beam during the initial elastic phase, and for
the portions of the beam separated by plastic hinges after the beam yields,
provided that the hinges are stationary. Salvadori and DiMaggio [1953]
studied the development of hinges for various degrees of concentration of the
load, ranging from uniformly distributed to concentrated. For a comparison
between the results of Lee and Symonds and those of Bleich and Salvadori,
see the discussion by Symonds and by Bleich and Salvadori following the
paper by Bleich and Salvadori [1955]. The impact loading of a rigid–perfectly
plastic beam that is built in at both ends was studied by Symonds and Mentel
[1958], who took into account the axial forces that develop when such a beam
deflects.
Cantilever Struck by a Mass at Its Tip

The method of Lee and Symonds was applied by Parkes [1955] to study
the impact on a cantilever beam of length L by an object of mass m traveling
with a velocity v0 . The beam is again assumed to be rigid–perfectly plastic,
so that the kinetic energy of the striker can be absorbed only in a plastic
hinge. Initially the hinge is at the tip, but it moves in time toward the
built-in end.
Let x denote distance along the beam measured from the tip and ∆(t)
the tip deflection, and let x = x̄(t) give the position of the hinge at t. The
portion x > x̄(t) remains undisturbed. In the portion x < x̄(t), the curvature
rate κ̇ = ∂ 2 v/∂x2 (where v = ∂u/∂t) vanishes, so that the velocity is given
by
x

˙
v(x, t) = ∆(t) 1− .
x̄(t)
Since the plastic hinge occurs at a maximum of the bending moment, the
shear force there is zero, and therefore the net force on the moving portion
of the beam is zero. The translational equation of motion for this portion of
the beam is therefore
m∆ ¨ + ρ̄ d (x̄∆)
˙ = 0, (7.1.5)
2 dt
while the rotational motion can be obtained by taking moments about the
tip, yielding
ρ̄ d 2 ˙
(x̄ ∆) = MU . (7.1.6)
6 dt
Equations (7.1.5)–(7.1.6) can be integrated to give
ρ̄ ρ̄

˙
m + x̄(t) ∆(t) = A, ˙
[x̄(t)]2 ∆(t) − MU t = B,
2 6
˙
where A and B are constant. Since x̄(0) = 0 and ∆(0) = v0 , it follows that
A = mv0 and B = 0. Hence, if we define β = ρ̄L/2m and ξ = x̄(t)/L, then
˙
∆(t) 1
=
v0 1 + βξ
and
mv0 L βξ 2
t= , (7.1.7)
3MU 1 + βξ
provided that ξ < 1. The plastic hinge reaches the built-in end at time
˙ 1 ) = v0 /(1 + β).
t = t1 , given by Equation (7.1.7) when ξ = 1. Note that ∆(t
For t > t1 , the motion is in mode form:
x

˙
v(x, t) = ∆(t) 1− .
L
The equation for ∆(t) can be obtained by taking moments about x = L,
namely, !
ρ̄L2 ¨
mL + ∆ = −MU .
3
Integration, with continuity at t = t1 , leads to
˙
∆(t) 1 3MU (t − t1 ) 3MU
= − = (ts − t),
v0 1 + β mv0 L(3 + 2β) mv0 L(3 + 2β)
where
mv0 L 3 + 2β mv0 L
ts = t1 + =
3MU 1 + β MU
is the time at which the motion stops. Consequently, the permanent deflec-
tion up may be obtained by integrating the velocity over time up to ts .
If the mass of the striker is much greater than that of the beam, that is,
if β 1, then the permanent deflection is approximately
mv02 L x

up (x) ≡ 1− .
2MU L
Thus the beam remains straight, and the deflection is that which is necessary
so that all the kinetic energy is absorbed by the hinge at the built-in end.
If, on the other hand, the mass of the striker is much less than that of the
beam, the final shape is a superposition of a rotation and a local deformation
near the tip. The form is, approximately, given by
2m2 v02 L
up (x) = ln if βx/L 1,
3MU ρ̄ x
with
2m2 v02
up (0) = ln β.
3MU ρ̄
Parkes [1955] also carried out experiments in which mild-steel cantilever
beams were subjected to impact by relatively heavy weights dropped on
their tips, as well as by bullets. Reasonably good agreement with the theory
was obtained when MU was given a “dynamic” value based on the data of
Manjoine [1944] for the rate sensitivity of the yield stress of mild steel.
A numerical study of the Parkes problem by Symonds and Fleming [1984]
shows that when elasticity is taken into account, then the initial phase of
the deflection is not even approximately represented by the traveling-hinge

solution — that is, plastic deformation is not concentrated in a narrow zone
in the interior. On the other hand, convergence to mode form is a prominent
feature of the solution.
7.1.3 Dynamic Loading of Plates And Shells
Dynamic Loading of Rigid–Plastic Circular Plates

The dynamic problem of a simply supported rigid–plastic circular plate
that is suddenly loaded by a uniformly distributed pressure — a blast load-
ing — was studied by Hopkins and Prager [1954], as a sequel to the same
authors’ static solution (Hopkins and Prager [1953]) discussed in 5.2.3. The
equilibrium equation (5.2.10) is replaced by the equation of motion,
Z r
0
(rMr ) − Mθ = − (p + µẅ)r dr. (7.1.8)
0
where we write u0 for ∂u/∂r and u̇ for ∂u/∂t, while µ is the mass per unit
area and p is the pressure (assumed acting in the negative z-direction).
The Tresca yield criterion, as represented by the hexagon of Figure 5.2.2
(page 310), is assumed. As in the static case, the plate is assumed to be in
regime BC, with point B corresponding to the center of the plate. A conical
deformation can therefore be assumed, given by
r

w(r, t) = −∆(t) 1 − .
a
Inserting this expression in the equation of motion (7.1.8) and substituting

Mθ = MU , we can integrate the equation to obtain
!
r2 ¨ r2 r3
Mr (r, t) = MU − p + µ∆(t) − .
6 6 12a
The boundary condition Mr (a, t) = 0 yields µ∆¨ = 2(p − pU ), where pU =

2 ¨ we may write the
6MU /a is the static ultimate pressure. Eliminating µ∆,
following expression for Mr :
r2 r

Mr = MU − p − (p − pU ) 2 − .
6 a
This result is valid, however, only if the right-hand side does not exceed MU ,
that is, if the quantity in brackets is nonnegative for all r. The necessary and
sufficient condition for this is p ≤ 2pU . Pressures satisfying this condition
may be called moderate pressures, while pressures greater than 2pU will be
called high pressures.
At high pressures, the deflection may be assumed in the shape of a

truncated cone, with a hinge circle of radius r0 separating the outer conical
region from the inner region, which moves rigidly with a velocity ẇ = −∆. ˙
The latter region is at point B of the Tresca hexagon, so that Mr = Mθ =
MU , and the equation of motion for r < r0 reduces to
¨ = p.
µ∆
For r > r0 , the velocity is
˙ a−r
ẇ(r, t) = −∆(t) ,
a − r0
and the acceleration is therefore given by
a−r
µẅ = −p .
a − r0
The equation of motion (7.1.8) now becomes
p
(rMr )0 = MU − (2r3 − 3r0 r2 + r03 ),
6(a − r0 )
which when integrated subject to the initial condition Mr = MU at r = r0 ,
leads to
(r − r0 )3 (r + r0 )
rMr = rMU − p .
12(a − r0 )
Finally, the condition Mr = 0 at r = a gives r0 as the solution of the cubic
equation
r0 2 r0 2pU

1− 1+ = . (7.1.9)
a a p
Suppose, now, that the pressure is suddenly removed at time t = t0 . For
p ≤ 2pU , the form of the preceding solution is still valid for t > t0 , with p
replaced by zero. The midpoint acceleration is now given by µ∆ ¨ = −2pU .
The time history of the midpoint deflection is therefore
1
∆(t) = [p(2t0 t − t20 ) − pU t2 ], 0 < t < t0 ,
µ
1
= (p − pU )t2 , t0 < t < ts ,
µ
˙ s ) = 0, so that
where ts is the time at which the motion stops, that is, ∆(t
ts = (p/pU )t0 . The permanent midpoint deflection is thus
p(p − pU ) 2
∆p = ∆(ts ) = t0 .
µpU
The result for ts coincides with the lower bound given by Martin’s the-
orem, Equation (7.1.1), when the conical velocity field is used, say ẇ∗ =
∆˙ ∗ (1 − r/a). The denominator of the right-hand side of (7.1.1) is then

2πMU ∆ ˙ ∗ , while the numerator is
2
a
˙∗ 1− r a2 ˙ 2
˙ ∗ = 2π a (p − pU )t0 ∆
Z
2π ˙ 0 )∆
µ∆(t r dr = 2π µ∆(t0 )∆ ˙ ∗.
0 a 12 6
Since MU = pU a2 /6, it follows that
p

ts − t 0 ≥ − 1 t0 ,
pU
and the right-hand side coincides with the exact value.
When p > 2pU , the removal of the pressure means that in the central
¨ = 0, and therefore
region the acceleration becomes zero, that is, ∆
˙
∆(t) ˙ 0 ) = pt0 .
= ∆(t
µ
˙ a nonvanishing
Since the velocity in the conical region is proportional to ∆,
acceleration there requires the radius of the hinge circle to be a function of
time, say r̄(t), with r̄(t0 ) = r0 . For r > r̄(t), then, the velocity and the
acceleration are, respectively,
pt0 a − r pt0 r̄˙ (t) a − r
ẇ(r, t) = − and ẅ(r, t) = − ,
µ a − r̄(t) µ [a − r̄(t)]2
so that the equation of motion is
pt0 r̄˙
(rMr )0 = MU + [3a(r2 − r̄2 ) − 2(r3 − r̄3 )].
6(a − r̄2 )2
Integrating, with Mr = MU at r = r̄, we obtain
pt0 r̄˙ (r − r̄)2
rMr = rMU + [2a(r + 2r̄) − (r2 + 2rr̄ + 3r̄2 )].
12(a − r̄)2
Setting Mr = 0 at r = a, we obtain for r̄ the differential equation
dr̄ 2pU a3
(a2 + 2ar̄ − 3r̄2 ) =− .
dt pt0
The factor on the left-hand side is positive, and therefore the radius of the
hinge circle will decrease. The equation can be integrated to give
t p
=1+ [a2 (r0 − r̄) + a(r02 − r̄2 ) − (r03 − r̄3 )].
t0 2pU a3
Let t1 denote the time at which the radius of the hinge circle goes to zero.
Then " 2 3 #
t1 p r0 r0 r0
=1+ + − ,
t0 2pU a a a
where r0 is given by Equation (7.1.9). For t > t1 , the velocity field (though
not the deflection) is fully conical, and is consequently in mode form.
Perzyna [1958] studied the effect of different time profiles of the pressure
pulse. The corresponding problem for a clamped plate was treated by Flo-
rence [1966]. The problem of impact loading of a circular plate, in which a
uniform velocity is suddenly imparted to the plate (except the edge), was
treated in a similar manner by Wang [1955] for a simply supported plate
and by Wang and Hopkins [1954] for a clamped plate. Non-axisymmetric
problems were treated by Hopkins [1957].
For simply supported plates, the results based on the Tresca criterion
are immediately transferable to the Johansen criterion, since regime BC of
Figure 5.2.2 is common to both. For clamped plates the criteria differ. The
blast loading of square clamped plates obeying the Johansen criterion was
studied by Cox and Morland [1959].
Pressure-Pulse Loading of a Cylindrical Shell

Relatively few dynamic problems have been solved for elastic–plastic or
rigid–plastic shells. We consider here a solution due to Hodge [1955] for
a circular cylindrical shell, made of a rigid–perfectly plastic material and
clamped at both ends, that is suddenly loaded by a uniform radial pressure
which does not decrease in time, the initial value of the pressure being greater
than the static ultimate pressure pU . For mathematical simplicity, the square
yield criterion of Figure 6.4.4 (page 409) is adopted; an analysis by Hodge
and Paul [1957] based on the hexagonal locus shows that the solution is not
greatly affected by changes in shape of the yield locus.
The static ultimate pressure may be obtained by assuming that Nθ = NU
throughout. The equilibrium equation (6.4.8)2 is then
NU
Mz00 = p − .
a
A collapse mechanism may be assumed to consist of hinge circles at the
center (z = 0) and at the built-in ends (z = ±L). For generalized-stress
points on side GH of the square of Figure 6.4.4, the associated flow rule
requires that u̇00 = 0. The radial velocity field is accordingly
|z|

˙ 1−
u̇(z) = ∆ . (7.1.10)
L
A moment distribution consistent with this velocity field has Mz (0) = −MU
and Mz (±L) = MU , and is given by
!
z2
Mz = M U 2 2 −1 ,
L
so that
NU 4MU NU 2

pU = + 2
= 1+ ,
a L a ω2
p
where ω = NU L2 /2MU a is the shell parameter defined in 6.4.3.
In the dynamic problem it can still be assumed that Nθ = NU , so that
the velocity field is
|z|

˙
u̇(z, t) = ∆(t) 1− .
L
The equilibrium equation is replaced by the equation of motion,
NU
Mz00 = p − − µü, (7.1.11)
a
where µ is again the mass per unit area. Integrating the equation of motion
subject to the condition Mz = −MU at z = 0 leads to
!
NU z2 2
¨ z − |z|
3

Mz = −MU + p − − µ∆ .
a 2 2 6L
The end condition Mz = MU at z = ±L then gives
¨ = 3 (p − pU ).
µ∆
2
¨ from the expression for Mz , we obtain
Eliminating µ∆
! !
z2 p − pU 2|z|3
M z = MU 2 2 −1 − z − .
L 4 L
This solution is valid only if |Mz | ≤ MU everywhere, requiring
Mz00 (0) ≥ 0, Mz0 (L) ≥ 0, Mz0 (−L) ≤ 0.
The first condition is satisfied if

MU NU 6

def
p ≤ pU + 8 2 = 1+ 2 = p1 .
L a ω
The last two conditions are fulfilled if
MU NU 6

p ≥ pU − 16 2 = 1− 2 . (7.1.12)
L a ω
If the pressure has a constant value p0 for 0 < t < t0 and then suddenly
drops to zero, then inequality (7.1.12) is obeyed after unloading only if ω 2 ≤
6; a shell satisfying this criterion may be called, following Hodge [1955], a
short shell. For a short shell with pU < p0 < p1 (a moderate pressure), the
solution thus far derived is valid both for 0 < t < t0 and for t > t0 , with p
given by p0 and 0, respectively. The midpoint velocity is
˙ 3
∆(t) = (p0 − pU )t, 0 < t < t0 ,
2µ
3
= (p0 t0 − pU t), t0 < t < ts ,
2µ
where ts = (p0 /pU )t0 is the time at which the motion stops. The permanent
radial expansion at the midsection is
3p0 t2 p0

∆(ts ) = −1 .
4µ pU
In a long shell (ω 2 > 6) under moderate pressure, the solution is valid

during the first phase but not after unloading. In order to obtain a valid
solution, we must abandon the hypothesis that the entire shell is in regime
GH of Figure 6.4.4. Instead, only a central portion of the shell, say |z| < z̄(t),
will be assumed to be in this regime, with the sections z = ±z̄(t) at point
G, and the remainder of the shell in regime GJ. In the latter regime the
flow rule gives u̇ = 0, and therefore the outer portions of the shell undergo
no further motion. The velocity field is thus
|z|

˙
u̇(z, t) = ∆(t) 1− , |z| < z̄(t),
z̄(t)
= 0, |z| > z̄(t).
The equation of motion for |z| < z̄ is

¨ ˙˙
!
NU ¨ + µ ∆ − ∆z̄

Mz00 =− + µ∆ |z|.
a z̄ zbar2
Since Mz0 (0, t) = 0, a first integration gives

˙˙
!
NU ¨ − ∆z̄ |z|z

Mz0 =− ¨ z+µ ∆
+ µ∆ .
a z̄ 2z̄
Furthermore, Mz0 = 0 at |z| = z̄, so that

˙ z̄˙
!
∆ ¨ NU
µ +∆ = −2 . (7.1.13)
z̄ a
Integrating again yields, upon elimination of NU by means of Equation

(7.1.13) and with Mz (0, t) = −MU ,
˙ z̄˙
! !
∆ ¨ z 2 |z|3
Mz = −MU + µ −∆ − .
z̄ 4 6z̄
Finally, since Mz = MU at |z| = z̄, we obtain

˙ z̄˙
!
∆ ¨ MU
µ −∆ = 24 . (7.1.14)
z̄ z̄ 2
Combining Equations (7.1.13) and (7.1.14) produces
¨ =− NU MU
µ∆ − 12 2 , (7.1.15)
a z̄
˙ z̄˙ = − NU MU
µ∆ z̄ + 12 . (7.1.16)
a z̄
¨ from
Differentiating (7.1.16) with respect to time and substituting for µ∆
(7.1.15), we find that
µ∆˙ z̄¨ = 0.
Since ∆ ˙ cannot be zero while the shell is in motion, it follows that z̄¨ = 0,
˙
so that z̄ is a constant. We obtain its value from Equation (7.1.16) at t = t0 ,
when z̄ = L and ∆ ˙ = 3(p0 − pU )t0 /2µ. Hence
2L NU 1 − ω62
z̄˙ = − ,
3t0 a p0 − pU
which is always negative for a long shell (ω 2 > 6); consequently, the hinge
circles move toward the center, and their location is given by
z̄(t) = L + (t − t0 )z̄˙ . (7.1.17)
The √
motion stops when the right-hand side of (7.1.16) vanishes, that is, when
z̄ = 6L/ω. The time when this occurs is
√ ! " #
L 6 3 p0 − pU
ts = t0 + 1− = t0 1 + √ .
|z̄˙ | ω 2 (1 + 6/ω)(NU /a)
The permanent radial expansion at the midsection can be obtained by inte-

˙
grating ∆(t), as given by (7.1.16) and with z̄ given by (7.1.17), from t0 to
ts , and adding the result to ∆(t0 ) = 3(p0 − pU )t20 /4µ.
Under high pressures, that is, p0 > p1 , during the initial phase hinge
circles may be assumed at z = ±z0 and at z = ±L, with M = −MU and
N = NU for |z| ≤ z0 , and the velocity given by
˙
u̇(z, t) = ∆(t), |z| < z0 ,
˙ L−z
= ∆(t) , |z| > z0 .
L − z0
Equation (7.1.11) for |z| < z0 gives
¨ =p− NU
µ∆ .
a
For |z| > z0 , therefore, the equation of motion may therefore be written as
NU |z| − z0

Mz00 = p− ,
a L − z0
and may be integrated twice, subject to the initial conditions Mz0 = 0 and
Mz = −MU at |z| = z0 , to give
NU (|z| − z0 )3

Mz = p − − MU .
a 6(L − z0 )
The conditions Mz = MU at the built-in ends |z| = z0 lead to

s
12MU
z0 = L − ,
p0 − (NU /a)
and z0 > 0 if p0 > p1 .

After unloading, the inner hinge circles must be assumed to move, their
location being, say, |z| = z̄(t). The equation of motion for |z| < z̄(t) reduces
to µ∆¨ = −NU /a. Continuity of the velocity at t = t0 gives the midsection
radial velocity for t0 < t < t1 as
1 NU

˙
∆(t) = p0 t 0 − t ,
µ a
where t1 is the time at which the hinge circles coalesce at the midsection.
For |z| > z̄(t) the equation of motion is
NU |z| − z̄ ˙ z̄˙ L − z .
Mz00 = − − µ∆
a L − z̄ (L − z̄)2
Integrating subject to Mz0 = 0 and Mz = −MU at |z| = z̄ and substituting

Mz = MU at |z| = L gives
˙ z̄˙ = − NU MU
µ∆ (L − z̄) − 12 .
a L − z̄
Substituting for ∆ ˙ as a function of time, we find the the differential equation

for z̄(t) is separable and can be written as
dt (L − z̄)dz̄
=− ,
t̄ − t (L − z̄)2 + 3L2 /ω 2
where t̄ = (p0 a/NU )t0 . We can integrate the equation subject to z̄(t0 ) = z0 ,
obtaining
t̄ − t0 2 (L − z̄)2 + 3L2 /ω 2

= .
t̄ − t (L − z0 )2 + 3L2 /ω 2
We finally find t1 as
  s 
p0 a 3 + ω 2 (1 − z0 /L)2 

t1 = t0 1 + − 1 1 − .
NU 3 + ω2
For t > t1 , the velocity field may again be assumed to be of the form (7.1.10)
(mode form). Calculation of the permanent deformation may be carried out
by analogy with the preceding examples. Details may be found in the paper
by Hodge [1955] (see also Hodge [1956]).
As the reader can see, there are a great many similarities between the
dynamic shell and plate problems formulated according to the rigid–perfectly
plastic model, and these in turn resemble the beam problems. The comments
expressed previously concerning the validity of the traveling-hinge solution
may therefore be expected to be applicable to structural impact problems
in general.

1. Formulate Martin’s inequality (7.1.2) in terms of generalized stresses
to give a bound on the permanent deflection for (a) beams and (b)
plates.
2. Determine the motion due to impact loading of a free-ended rigid–
perfectly plastic beam by a concentrated force at the center if (a) the
force has the constant value Fm for 0 < t < t0 and is removed at t = t0 ,
and (b) the force rises linearly from zero to Fm at t = t0 /2 and then
declines linearly to zero at t = t0 .
3. Analyze the motion of a rigid–perfectly plastic cantilever struck by a
mass m that is uniformly distributed over a length a at the free end of
the beam. In particular, find the stopping time ts and compare with
the bound (7.1.1). Show that the solution reduces to that of Parkes as
a/L → 0.
4. Analyze the motion of a rigid–perfectly plastic beam struck by a con-
centrated mass m at midspan when the beam is (a) simply supported
and (b) clamped.
5. Analyze the motion of a rigid–perfectly plastic circular Tresca plate of
radius a that is suddenly loaded by a pressure p uniformly distributed
over the circle r < b (where b < a) when the plate is (a) simply
supported and (b) clamped.
6. Analyze the motion of a simply supported rigid–perfectly plastic cir-
cular Tresca plate that is struck by a concentrated mass m at the
center.
7. Compare the stopping time ts found in the text for a long clamped
cylindrical shell under a moderate impulsive pressure with the bound
(7.1.1).
8. Analyze the motion of a long clamped cylindrical shell under a high

pressure for t > t1 . Find the stopping time ts and compare with (7.1.1).
9. Analyze the motion of a clamped cylindrical shell under a pressure-

pulse loading. Study various cases.
10. Analyze the motion of an infinitely long cylindrical shell suddenly

loaded by a ring load.
Section 7.2 One-Dimensional Plastic Waves

7.2.1 Theory of One-Dimensional Waves
In studying the propagation of longitudinal stress and strain waves in a thin
bar, it is common to represent the problem by a “one-dimensional” approx-
imation in which the only nonvanishing stress component is assumed to be
the longitudinal one, and the contribution of the transverse displacement to
the deformation and the inertia is ignored. It is known from elastic bar the-
ory that this approximation yields good results at points of the bar whose
distance from the bar ends is more than a few diameters; near the ends,
three-dimensional corrections are necessary.
We let x denote the Lagrangian coordinate along the bar axis and u(x, t)
the corresponding displacement. The conventional (“engineering”) strain
ε(x, t) and velocity v(x, t) are then given by
∂u ∂u
ε= , v= ,
∂x ∂t
and satisfy the kinematic compatibility relation
∂ε ∂v
= . (7.2.1)
∂t ∂x
The equations of motion, in the absence of body force, reduce to
∂σ ∂v
=ρ , (7.2.2)
∂x ∂t
where σ is the nominal (“engineering”) stress and ρ is the mass density in
the undeformed state.
Section 7.2 / One-Dimensional Plastic Waves 435
The equations governing the problem of torsional motion of a thin-walled

circular tube are exactly the same as the preceding ones if u and v are in-
terpreted as the circumferential displacement and velocity (uθ and vθ ), re-
spectively, and ε and σ as the conventional shear strain and shear stress (γzθ
and τzθ ). Except for some comments concerning finite strain, the following
theory applies to the torsional problem as well.
Shock Fronts
A shock front is said to occur at a point x = α(t) of the bar if the
velocity v is discontinuous there. We suppose that the shock front is moving
at a finite speed c in the positive x-direction (say from left to right), that
is, c = α0 (t) > 0, and we designate the values of v just to the right (in
front) of the shock and just to the left of (behind) the shock by v + and v − ,
respectively. The jump in v is defined as
[|v]| = v − − v + .
Similar definitions apply to ε and σ. It is readily apparent that with v a
discontinuous function of x and of t, its partial derivatives with respect to
both variables are in effect infinite at the shock front. By Equation (7.2.2),
the same is true of ε and σ.
Relations among the jumps in velocity, strain, and stress may be derived
by treating the shock front as a thin zone in which these quantities change
very rapidly, their partial derivatives having large constant values. If the
shock thickness is h, then, approximately,
. ∂v
[|v]| = ±h ,
∂x
for a front moving to the left and to the right, respectively, with similar
relations for ε and σ.
Another approximation for the jump may be obtained by following it
along the time axis. Since v − is the earlier and v + the later value, and since
the duration of the shock passage at a given point is h/c, it follows that
. h ∂v
[|v]| = .
c ∂t
Applying these approximations to Equations (7.2.1)–(7.2.2), we obtain the
shock relations
1
[|ε]| = ± [|v]| , [|σ]| = ±ρc[|v]| , (7.2.3)
c
where the + and − signs apply to fronts moving to the left and to the right,
respectively.
Eliminating [|v]| , we are left with the shock-speed equation
[|σ]|
ρc2 = . (7.2.4)
[|ε]|
Impact of a Rigid–Plastic Bar

As an application of the shock relations (7.2.3), we study the problem
of longitudinal impact of a bar treated as rigid–plastic. Suppose that the
bar, of length L, is moving rigidly at a speed v0 in the direction of its axis
until, at time t = 0, it squarely strikes a rigid target. The situation at t > 0
is shown in Figure 7.2.1; it is assumed that plastic deformation takes place
only in an infinitely narrow zone at x = ξ(t), where x is measured from the
free end [so that ξ(0) = L], with the remaining portions of the bar rigid. In
ξ(t) -
At rest
- x
v̄(t)
Shock
front
Figure 7.2.1. Impact of a rigid-plastic bar on a rigid wall.
particular, the portion near the target, x > ξ(t), is assumed to have come to
rest, while the portion near the free end is moving with a uniform velocity:
v(x, t) = v̄(t), x < ξ(t),

v(x, t) = 0, x > ξ(t).
The section x = ξ(t), since the velocity is discontinuous there, is the location
of a shock front moving in the negative x-direction with a speed c = −ξ˙ with
respect to the undeformed material. For convenience, stress and strain will
be taken as positive in compression, so that the signs in Equations (7.2.3)
are reversed.
We begin by assuming the bar as rigid–perfectly plastic, so that it is at
yield at the shock front. The true stress, then, equals σY at x = ξ(t) and
is continuous there. The nominal stress is also σY ahead of the front, but it
equals σY /(1− ε̄) behind the front if ε̄ is the strain there. The discontinuities
at the front are therefore
σY σY ε̄
[|σ]| = − σY = , [|ε]| = ε̄, [|v]| = −v̄,
1 − ε̄ 1 − ε̄
and the shock relations are

v̄ σY ε̄ ˙
ε̄ = − , = −ρξv̄,
ξ˙ 1 − ε̄
from which, on the one hand,

σY
ρξ˙2 = (7.2.5)
1 − ε̄
while, on the other hand,

σY ε̄2
ρv̄ 2 = , (7.2.6)
1 − ε̄
and upon differentiating,
!
σY ε̄2
ρv̄ dv̄ = d . (7.2.7)
2 1 − ε̄
In addition, the equation of motion of the undeformed portion is
ρξ(t)v̄˙ (t) = −σY ,
so that
σY v̄ ξ˙
ρv̄ v̄˙ = − = σY ε̄ . (7.2.8)
ξ ξ
Combining (7.2.7) and (7.2.8) gives
!
dξ 1 ε̄2
2 = d . (7.2.9)
ξ ε̄ 1 − ε̄
If the initial impact velocity is v̄(0) = v0 , then the initial strain at the impact
end is obtained in terms of v0 through (7.2.6), namely,
ρv02 ε20
= .
σY 1 − ε0
Equation (7.2.9) may now be integrated, giving
2
ξ 1 1 1 − ε̄
ln = − − ln .
L 1 − ε̄ 1 − ε0 1 − ε0
If ξ0 is the value of ξ where ε̄ = 0, then
√ ε0

ξ0 = L 1 − ε0 exp − .
2(1 − ε0 )
and the shock front stops when ξ(t) = ξ0 , so that the portion 0 < x < ξ0
remains undeformed.
The preceding analysis is equivalent to that of G. I. Taylor [1948], who
based it on a different approach. By means of some approximating assump-
tions Taylor derived a formula for the dynamic yield stress in terms of the
impact speed and the specimen dimensions before and after impact.
For a bar made of work-hardening material, the problem was treated by
Lee and Tupper [1954]. The geometry and notation are the same as in the
preceding treatment. If the conventional stress-strain relation is given by
σ = F (ε) and the initial yield stress is σE , then the material just ahead of
the shock front may be assumed to be about to yield, so that σ = σE there,

while immediately behind the front the stress is σ̄ = F (ε̄). The stress jump
is therefore [|σ]| = σ̄ − σE . From the shock relations we obtain, first,
σ̄ − σE
ρξ˙2 = ,
ε̄
which may be contrasted with (7.2.5), and second,
ρv̄ 2 = (σ̄ − σE )ε̄.
Proceeding as before, we obtain
dξ d[(σ̄ − σE )ε̄]
2 = .
ξ σE ε̄
The initial strain at the impact end is related to the impact speed v0 by
means of
ρv02 = (σ0 − σE )ε0 ,
where σ0 = F (ε0 ). The relation between ε̄ and ξ, which provides the distri-
bution of permanent strain along the bar, is obtained as
2 ε̄
ξ d{[F (ε) − σE ]ε}
Z
ln = ,
L ε0 σE ε
provided that ε̄ is positive; setting ε̄ = 0 gives ξ0 , which defines, as before, the

extent of the undeformed portion. The result was used by Lee and Tupper
[1954] in order to predict the permanent deformation of steel specimens, and
experiments performed by them showed fairly good agreement.
7.2.2 Waves in Elastic-Plastic Bars

Taking the elasticity of the bar into account changes the nature of the prob-
lem drastically. In an elastic solid, disturbances cannot be propagated at a
speed faster than the elastic wave speed. It would therefore be impossible
for the entire portion of the bar ahead of the plastic shock front to move as
a rigid body from the moment of impact; an elastic front has to intervene.
Moreover, when an elastic wave of compression reaches the free end, it is
reflected as a tension wave and will bring about unloading when it reaches
the plastically deforming material.
At a section of the bar where the material is elastic, with Young’s mod-
ulus E, we have [|σ]| /[p
|ε]| = E, so that the shock speed is just the elastic
bar-wave speed ce = E/ρ, independent of the local state variables. The
same is true when the behavior is elastic-plastic, provided that the stress
change across the shock represents unloading.
Donnell Theory
To understand the nature of a front of loading into a plastic state, it is
simplest to begin, following an approach due to Donnell [1930], by consid-
ering a material with a “bilinear” stress-strain relation (Figure 7.2.2). We
E0
σE 1

E

1
ε
Figure 7.2.2. Bilinear stress-strain diagram.
return to the convention of stress and strain as positive in tension. If the

initial state is stress-free, then any disturbance involving a stress increase
up to the yield stress σE will be propagated at the aforementioned elastic
speed ce . On the other hand, p stress increases above σE will travel at the
slower “plastic” speed cp = E 0 /ρ. Thus in a semi-infinite bar occupying
the half-line x > 0 in which at t = 0 a stress σ0 > σE is applied at the left
end, we distinguish three regions, whose time dependence is shown in Figure
7.2.3:
0 : ce t < x σ = 0,
I : cp t < x < ce t σ = σE ,
II : 0 < x < cp t σ = σ0 .
The values of strain and velocity in each of the regions can be obtained
from the shock relations (7.2.3); they are shown in Figure 7.2.3.
Finite Bars
If the bar is of finite length, say L, and is supposed to occupy the interval
0 < x < L, then a front propagating to the right from x = 0 will, upon
reaching the end x = L, be reflected as one propagating to the left. The
nature of the reflected front depends on the end condition at x = L, and if the
bar is elastic-plastic, on the intensity of the incident front. End conditions
are assumed as being of two kinds: (a) free, characterized by σ = 0, and
(b) fixed , characterized by v = 0 if the frame of reference is stationary with
respect to the fixed end.
Consider, now, an elastic shock front, with a stress −σ0 and zero velocity
behind it, moving into a region of zero stress and velocity −v0 (so that from
the shock relations, σ0 = ρce v0 ) until it reaches a free right end. This is the
problem of an elastic bar traveling freely to the left, at a speed v0 , until it
t
σ = σ0
σE σ0 − σE
ε= +
E E0
σE σ 0 − σE
v=− −
ρce ρcp

σ = σE
σE
ε=
E
σE
v=−
ρce

!
!
!!
II !!
!
I !!
!
! σ=ε=v=0
! !!
!! 0
!
! x
Figure 7.2.3. Semi-infinite bar of bilinear material: regions in xt-plane.
impinges upon a fixed rigid obstacle.1 Behind the reflected front the stress
must be zero, so that it is a front of unloading, and therefore, by the shock
relations, the velocity behind it will be +v0 .
When the reflected front reaches x = 0, at the time t = 2L/ce after the
initial impact, it will in turn be reflected to the right. If this end is regarded
as a fixed end, then, in order to have v = 0 behind the second reflected
front, the stress there must be +σ0 , that is, a tensile stress. Unless the bar
somehow fuses with the obstacle, there is nothing to transmit a tensile stress
between them, and therefore the assumption that the left end is fixed for
t > 2L/ce must be abandoned in favor of the one that the end is free. In
this case, the state behind the second reflected front has σ = 0 and v = +v0 ;
that is, the bar rebounds from the obstacle. The rebound time is consequently
2L/ce .
If the bar is elastic-plastic, with a compressive yield stress σE , then the
def
preceding results are valid as long as σ0 ≤ σE , that is, if v0 ≤ σE /ρce = vE .
If v0 > vE , and if the material is modeled by a bilinear stress-strain relation
as on the previous page, then an elastic and a plastic shock front propagate
from the impact end at the respective speeds ce and cp , and the results for
the semi-infinite bar hold until the initial elastic front, with σ = −σE and
v = vE − v0 behind it, reaches the free end. The reflected front, being one of
unloading, is elastic, the stress behind it being zero and the velocity 2vE −v0 .
def
Let α = cp /ce . The reflected elastic and initial plastic fronts meet at
point A (Figure 7.2.4), with xA = 2αl/(1 + α) and tA = 2L/(1 + α)ce .
1
This problem was studied by Lenskii [1948] and De Juhasz [1949].
From this point, fronts may propagate both to the left and to the right.
We assume, first, that only elastic fronts propagate, leaving behind them
a region with σ = −σ1 and v = v1 . The strain cannot be uniform in this
region, since plastic deformation has taken place to the left of A but not
to the right of A. Thus a stationary discontinuity front, at which only the
strain is discontinuous, forms at A.
hhh
Strain ((((
discontinuity ((

( ( ((((
((( (
(
(((( σ=0
hhhh( A ((((
σ = −σ0 hhhhh v = 2vE − v0
v=0 h hhh
h
h hhh
hhhh
hhh
σ = −σ0 hhhh
v = vE − v0 ( ( (((
( (( ((
(((
( ( ((( σ=0
(( (( ( v = −v0
( (
(((( x
0 L
Figure 7.2.4. Finite bar of bilinear material: wave reflection and intersection.
The shock relations along the rightward and leftward fronts, respectively,
are
σ1 = ρce (v1 − 2vE + v0 ), σ1 − σ0 = −ρce v1 .
Noting that σ0 = σE + ρcp (v0 − vE ), we obtain from these relations the

solutions σ1 = 21 ρce (1 + α)(v0 − vE ) and v1 = 21 [(3 − α)vE − (1 − α)v0 ]. In
order to justify the assumption that only elastic fronts propagate from A (in
other words, that the plastic front is absorbed at A), we must have σ1 < σE ,
the condition for which is v0 < [(3 + α)/(1 + α)]vE .
Let us assume that this last condition is met, and that the leftward elastic
front is reflected from the impact end as though rebound occurred, as with
the elastic bar, at t = 2L/c. The rebound velocity is found to be 2vE −v0 ; but
this is positive only if v0 < 2vE . This, then, is the condition for rebound at
t = 2L/ce . It can be ascertained that no further plastic deformation occurs,
and therefore the bar is left with a permanent deformation of ε0 − σ0 /E to
the left of point A.
If v0 > 2vE , then the bar remains in contact with the obstacle at least
until the next front reaches the left end. If 2vE < v0 < [(3 + α)/(1 + α)]vE ,
then the compressive stress holding them together is ρce (v0 − 2vE ).
If v0 > [(3 + α)/(1 + α)]vE , then the plastic front continues to the right
beyond A. Ahead of it, we have σ = −σE and v = 3vE − v0 . If the stress
and velocity behind it are again denoted −σ1 , v1 , then the shock relations
are
σ1 − σE = ρcp (v1 − 3vE + v0 ), σ1 − σ0 = −ρce v1 ,
yielding v1 = 2αvE /(1 + α).
It is clear that the numerical problem of determining the rebound time
becomes more complex as the impact velocity increases. Some solutions are
shown in Figure 7.2.5.
Kármán–Taylor–Rakhmatulin Theory
It is not difficult to extend the Donnell theory to materials whose stress-
strain curve consists of more than two line segments, provided that the
overall curve is convex downward. Such a curve may in turn be regarded
as an approximation to one that is smooth beyond the yield stress (Figure
7.2.6). It is clear that as the segments become shorter, the jumps in the field
quantities at the plastic shock fronts become smaller. If the smooth curve
is seen as the limit of the segmented one, then it follows that in a bar made
of a material described by the smooth curve, no plastic shock fronts can be
propagated.
The theory that was developed independently by Taylor [1946, 1958],
von Kármán (see Kármán and Duwez [1950]) and Rakhmatulin [1945] during
World War II is based on complementing the basic field equations (7.2.1)–
(7.2.2) with the constitutive equation
σ = F (ε). (7.2.10)
While its form is that of a nonlinear elastic stress-strain relation, Equation

(7.2.10) is assumed in this theory to hold only during an initial loading
process, unloading being linearly elastic. An initial yield stress σE may be
incorporated if F (ε) = Eε for |ε| < ε0 , where ε0 = σE /E (here it is assumed
that the initial loading curves are the same in tension and compression). The
function F (·) is assumed to be continuously differentiable, with F 0 (ε) > 0,
and convex in the sense that εF 00 (ε) ≤ 0.
A solution to the system of equations (7.2.1), (7.2.2), and (7.2.10) that
may be valid for some problems, at least in a certain region of the xt-plane,
can be obtained by assuming that the three variables σ, v, ε depend on x
and t only through the combination c = x/t, that is, that they are constant
along lines of slope c emanating from (0, 0). (Such a solution is sometimes
called a “similarity” solution.) Since ∂c/∂x = 1/t and ∂c/∂t = −c/t,
Equations (7.2.1)–(7.2.2) become
dv = −c dε, dσ = −ρc dv. (7.2.11)

t t
ce + cp
σ1 = ρ (v0 − vE )
1 2 0, vr
v1 = [(3 − α)vE − (1 − α)v0 ] X X vr = (1 − α)vE + αv0
2 tr
+ X
XX XXXX
XX
XXXXXXX
XXX X X
0, 2vE −v0 ρce v0 −2σE , 0 X XX 0, vr

XX1 0, 2v −v 1
t0
t0 X
E 0
X 0, 2vE −v0
σ0 , 0 XXX σ0 , 0 XXX
XX
XXX XX
XXX
XXX XXX
1 σE , vE −v0 σE , vE −v0
cp ce 1

0, −v0 0, −v0
x x
0 v0 L 0 v0 3+α L
1< <2 2< <
vE vE 1+α
(a) (b)
t t
3+α 1 + α k 2L

σ1 = σE + ρcp v0 − vE tk = , k = 0, 1, . . . , n
2α 1+α 1−α ce
v1 = vE tn ≤ tr
1+α 5+α

σ2 = σE + ρcp v0 − vE σ2k+1 = σE + ρcp (v0 − 3vE +
v2k+1 )
1+α 2α 1−α k
v2 = 0 v2k+1 = (−1)k vE
ce + cp 2α(1 − α)
nh i o
σ3 = σ E − ρ 3+ v E − v 0 1+α 1+α
2 (1+ α)2 σ2k+2 = σ2k+1 − ρce v2k+1
3+α v2k+2 = 0
v3 = −(1 − α) v0 − vE
1+α
0, 4vE −v0
X X 0, 4vE −v0

tr XX XXX
X
tr XX

XXX XXX C 8 X X
0, 4vE −v0
XXX X X
X
XXX 7 XXX
ρce v0 −2σE , 0
XXX X X C 6 XX
XXX
XX CW
XXX 5 X
4

t1 3 i
P t1 3
XXX
2 PPP X
XX
2

t0
XX 1 PP t0 X 1
σ0 , 0 XXX 0, 2vE −v0 σE , 3vE −v0 X
σ0 , 0 XXX 0, 2vE −v0
XXX XXX
X XXX XXX
XX
σE , vE −v0 X
σE , vE −v0

0, −v0 0, −v0
x x
0 L 0 L
3+α v0 2α(1 − α) 2α(1 − α) v0
< <3+ 3+ < < 4 (provided α < 0.28)
1+α vE (1 + α)2 (1 + α)2 vE
(c) (d)
Figure 7.2.5. Solutions for impact of a finite bar of bilinear material at various
impact velocities.
σ
(

((((

ε
Figure 7.2.6. Piecewise linear and smooth stress-strain curves.
These equations are clearly of the same form as the shock relations (7.2.3),
with the jumps replaced by differentials. Indeed, the limit of the shock
relations for “infinitesimal shocks” propagating to the right takes precisely
the form of Equations (7.2.11).
Eliminating dv between the two equations (7.2.11) yields
dσ
ρc2 = .
dε
Combining this result with (7.2.10) leads to a relation between c and ε:
s
F 0 (ε)
c(ε) = . (7.2.12)
ρ
Provided that the sign of ε is given, this relation will be assumed invertible
if |ε| > εE . Thus to any given c > ce there corresponds a value of ε and, by
(7.2.10), of σ. From the first of Equations (7.2.11) we deduce, in addition, a
relation for the velocity: Z
v=− c(ε) dε, (7.2.13)
the lower limit of the integral depending on the initial conditions.

The preceding solution can be applied to the problem of impact on a
semi-infinite bar, treated previously according to the Donnell model. The
notation will be simplified by treating stress and strain as positive in com-
pression, and consequently the minus signs in Equations (7.2.11) and (7.2.13)
will be eliminated. The relations between the impact velocity v0 , the maxi-
mum stress σ0 and the maximum strain ε0 are accordingly given by
Z ε0
v0 = c(ε) dε, σ0 = F (ε0 ). (7.2.14)
0
As in the Donnell theory, an elastic shock front propagates into the

undeformed region, with σ = σE immediately behind it. In the region
c(ε0 )t < x < ce t, however, the state varies continuously, the solution be-
ing the “similarity” solution described previously. For x < c(ε0 )t, the state
is again uniform, as in the Donnell theory.
In a finite bar, with the right end x = L free, the incident elastic front
is again reflected as an unloading front, with a speed that is, at least at
first, equal to ce .2 However, this front propagates, from the outset, into a
region of plastically deformed material, with strains that are not necessarily
infinitesimal. At finite strain, the elastic relation ∆σ = E∆ε between stress
and strain changes, with E the Young’s modulus measured at small strains,
is not valid if σ and ε are the nominal stress and conventional strain, respec-
tively. Experiments show, on the other hand, that the slope of the unloading
line is very nearly E if the stress and strain are taken as the true (Cauchy)
stress and the logarithmic strain, respectively. As we noted before, when the
strain is positive in compression then, with incompressibility assumed, the
true stress σt is related to the nominal stress σ by σ = σt /(1 − ε). Since the
change of strain in an elastic process is infinitesimal, we accordingly have
.
∆σ = ∆σt /(1 − ε), and ∆ ln(1 − ε) = ∆ε/(1 − ε). The unloading relation is,
consequently,
∂σ E
= . (7.2.15)
∂ε (1 − ε)2
Assuming, first, that ε0 is sufficiently small for the right-hand side of

(7.2.15) to be closely approximated by E, we can show that as in the Donnell
theory, rebound occurs at t = 2L/ce if v0 < 2vE . After rebound, the entire
bar is permanently deformed. As before, a plateau of permanent strain,
of value ε0 − σ0 /E, extends to the left of point A, while the permanent
strain decreases from this value to zero between A and the right end. The
distribution of permanent compressive strain is shown as a thickening of the
bar, since volume remains essentially constant. Thus the final shape of the
bar is as shown in Figure 7.2.7, and this shape has frequently been found in
tests.
Figure 7.2.7. Final shape of a bar with a smooth stress-strain curve.
If the strains are too large to be considered infinitesimal, then it follows

from Equation (7.2.15) that the elastic unloading front does not propagate
at the constant speed ce . Instead, its trace in the xt-plane is curved in the
2
The problem of the finite bar was studied by Lee [1953] and others (see Cristescu
[1967], Section II.9).
region c(ε0 )t < x < ce t, the slope being −ce /(1 − ε+ ) where x/t = c(ε+ ).
The determination of the state behind this front must be accomplished nu-
merically, for example, by the method of characteristics discussed in Section
5.1.
7.2.3 Rate Dependence
Dynamic Stress-Strain Relations

It was already pointed out by Kármán and Duwez [1950] that an assumed
stress-strain relation such as σ = F (ε) used in solving dynamic problems is
not necessarily the same as the stress-strain relation obtained from static or
quasi-static tests. Most metals are rate-dependent, or viscoplastic, and rate-
independent plasticity, as was pointed out in Chapter 3, is an approximation
valid for sufficiently slow processes. The work of Kármán, Taylor, Rakhmat-
ulin, and others who have followed the “rate-independent” approach is based
on the assumption of a “dynamic” stress-strain relation valid in the range
of strain rates that is encountered in dynamic tests, but not necessarily
identical with the static one. Kolsky and Douch [1962], among other ex-
perimenters, performed tests in which they deduced dynamic stress-strain
relations for several metals from wave-propagation experiments, and com-
pared them with the static relations. The results showed that the relations
were markedly different, with the dynamic curve well above the static one,
for commercially pure copper and aluminum, but that they were identical for
a precipitation-hardened aluminum alloy. Many other experimenters have
obtained results, reviewed by Clifton [1983], in which the dynamic stresses
are well above the static ones. Bell [1968] is relatively alone among major
investigators in maintaining that the static stress-strain relation governs the
dynamic problem.
The appearance of a plateau of permanent strain, predicted by the KTR
theory, has been seen by many as justifying the “rate-independent” ap-
proach. Many others, however, have criticized this approach. In particular,
it has been pointed out that in most tests the plateau occurs in a region that
is too close to the impact end for the one-dimensional approximation, which
neglects shear deformation and transverse inertia, to be valid.
More generally, the very assumption of a dynamic stress-strain relation
that is more or less rate-independent in the dynamic range of rates has
been questioned. For example, it was found by Sternglass and Stuart [1953],
Bianchi [1964], and other investigators that in wires or bars that have been
statically prestressed into the plastic range (and not unloaded), additional
pulses travel at the elastic wave speed, in clear contradiction with the rate-
independent model (see Clifton [1983] for a review of the relevant evidence).
Viscoplastic Theory
An alternative theory, treating the material as viscoplastic, was proposed
by Sokolovskii [1948] and with somewhat greater generality by Malvern
[1951a,b]; the viscoplasticity model of Perzyna [1963], discussed in Section
3.1, is essentially a three-dimensional generalization of the model underlying
the Sokolovskii–Malvern theory. The constitutive equation used by Malvern
is
∂ε 1 ∂σ
= + g(σ, ε), (7.2.16)
∂t E ∂t
where the function g(σ, ε) vanishes for stresses whose magnitudes are be-
low those given by the static stress-strain curve. For example, if the static
relation is given by σ = F (ε), then a possible form of g is
g(σ, ε) = G(|σ| − |F (ε)|)sgn σ,
where G 1s a function that is positive for positive argument and zero oth-
erwise. In the Sokolovskii model, |F (ε)| = σE for |ε| ≥ εE , that is, the
static behavior of the material is taken as perfectly plastic. With G given
by G(x) = A < x >, A being constant, the resulting equation for g(σ, ε) is
just the one-dimensional version of the Hohenemser–Prager model [Equation
(3.1.3)].
The use of constitutive equations of the form (7.2.16) is not limited to
materials that are viscoplastic in the classical sense, that is, those that are
characterized by a static yield stress below which the behavior is elastic. It
is used in creep models such as the Bailey–Norton–Nadai law [see Equation
(2.1.6)], with g(σ, ε) = B|σ|m |ε − σ/E|−n sgn σ.
A generalized constitutive equation that includes both the rate-indepen-
dent and the Malvern models as special cases was proposed in the early
1960s by Simmons, Hauser, and Dorn [1962], Cristescu [1964], and Lubliner
[1964]. This equation has the form
∂ε ∂σ
= f (σ, ε) + g(σ, ε) (7.2.17)
∂t ∂t
for loading (σ∂σ/∂t > 0); for unloading, f (σ, ε) is replaced by 1/E, or if
finite strains must be taken into account, (1 − ε)2 /E. The Malvern theory
corresponds to the special case f (σ, ε) ≡ 1/E, and the KTR theory is recov-
ered if f (σ, ε) = 1/F 0 (ε) and g(σ, ε) ≡ 0. Explicit calculations by Lubliner
[1965] and by Lubliner and Valathur [1969] show that solutions of impact
problems according to the generalized theory tend to those of the KTR the-
ory if the bar is short, the impact is strong, and if the viscoplastic relaxation
time is long, while under the converse conditions the results of the Malvern
theory are approached.
7.2.4 Application of the Method of Characteristics
Characteristic Relations
The three equations (7.2.1), (7.2.2), and (7.2.17) form a system of partial
differential equations for the unknown variables σ, v, and ε. (Note that
in using these equations we are returning to treating stress and strain as
positive in tension.) As in Section 5.1, it is simpler to consider first a single
first-order partial differential equation of the form
∂v ∂v
A +B = C,
∂x ∂t
where A, B, and C are functions of x, t, and v. Rewriting the equation as
∂v A ∂v

B dt + dt = C dt,
∂x B ∂t
we find that the left-hand side is proportional to dv = (∂v/∂x) dx+(∂v/∂t) dt
if dx = c dt, where c = A/B; the characteristic direction is thus defined by
dx/dt = c. Along a characteristic, then,
dx dt dv
= = .
A B C
If a discontinuity in v, resulting from the intersection of two character-
istics of the same family, should propagate, then it forms a shock front, as
discussed in 7.2.1. Any discontinuity in ∂v/∂t or ∂v/∂x that is propagated
through the xt-plane must (as discussed in 5.1.1) be across a characteristic,
and dx/dt = c is just the speed with which the front carrying the jump is
propagated. If v is the velocity, then a front across which the acceleration
∂v/∂t is discontinuous is called an acceleration wave.
The system formed by Equations (7.2.1)–(7.2.2) and (7.2.17) is equiva-
lent to
∂v ∂σ ∂ε ∂v ∂ε ∂σ

λ1 ρ − + λ2 − + λ3 −f − g = 0,
∂t ∂x ∂t ∂x ∂t ∂t
where λ1 , λ2 , λ3 are arbitrary multipliers. Let us regroup the terms of this
equation so that it takes the form
∂σ ∂σ ∂v ∂v ∂ε

λ1 + λ3 f + λ2 − λ1 ρ − (λ2 + λ3 ) + λ3 g = 0.
∂x ∂t ∂x ∂t ∂t
The determination of the characteristics of the system is tantamount to
finding the values of dx/dt for which the bracketed terms in the last equation
are proportional, respectively, to dσ, dv, and dε, where
∂σ ∂σ
dσ = dx + dt
∂x ∂t
and likewise for dv and dε. The possible values of dx/dt must satisfy
dx λ1 λ2 0
= =− = .
dt λ3 f λ1 ρ λ 2 + λ3
One way that all these equations can be obeyed is if

(a) λ1 = λ2 = 0, λ3 arbitrary (let λ3 = 1).
Two other possibilities are given by
(b) λ2 = −λ3
and s
λ1 f
(c) =± .
λ2 ρ
In case (a) we have
dx
= 0.
dt
In cases (b) and (c),
dx λ1
=− = ±c,
dt λ2 f
√
where c = 1/ ρf .
If f is positive, then at each point there exist three distinct characteristic
directions, which accordingly define three families of characteristic curves in
the xt-plane. Since the number of characteristic directions equals the number
of unknown variables, the system of equations is hyperbolic, as in 5.1.1.
Along an infinitesimal characteristic segment of family (a) we have dx =
0, and therefore changes in the unknown variables are given by Equation
(7.2.17), that is,
f dσ − dε + g dt = 0 along dx = 0. (7.2.18a)
Along characteristic segments of families (b) and (c), respectively, we obtain

the relations
g
dσ = ρc dv − dt along dx = c dt, (7.2.18b)
f
g
dσ = −ρc dv − dt along dx = −c dt. (7.2.18c)
f
Viscoplastic Bars
For a bar made of an elastic-viscoplastic material governed by (7.2.16)
(with infinitesimal strains), the characteristic relations (7.2.18) become
1
dσ − dε + g dt = 0 along dx = 0,
E
dσ = ρce dv − Eg dt along dx = ce dt,
dσ = −ρce dv − Eg dt along dx = −ce dt.
Since the characteristics are straight, the numerical integration of these

equations is quite straightforward. For certain simple forms of the func-
tion g, the equations may be solved analytically. Examples include so-
lutions by Sokolovskii [1948] and by Kaliski and Wlodarczyk [1967] for
g(σ, ε) =< |σ| − σY > (sgn σ)/η (the Hohenemser–Prager model), and by
Cristescu and Predeleanu [1965] for g given by
1 E0

g(σ, ε) = 1+ σ − E 0 ε − σE sgn σ , |σ| > σE ,
η E
with g = 0 for |σ| < σE .
Rate-Independent Bars; Unloading Wave

Since the KTR theory corresponds to g == 0, it is clear that Equa-
tions (7.2.11) representing the “similarity” solution are just the character-
istic relations along characteristics of family (c), these characteristics being
curved. The lines x = ct along which the state variables are constant are the
characteristics of family (b). The characteristic network for the problem of
constant-velocity impact on a semi-infinite bar made of a rate-independent
material is shown in Figure 7.2.8(a).
t t
Unloading wave
BBN
AA A
A
X
A A X

A A

A % tm
A % (c)

A %

A
%
A % (b)

X
A %
X X
% (a) X XXX
% X
X XX X
X
X
v
%
-x v -x
v0 v m vE
(a) (b)
Figure 7.2.8. Characteristic networks for impact on a semi-infinite bar made of

a rate-independent material: (a) constant-velocity impact; (b)
impact velocity rising to a peak and declining.
Suppose, now, that the impact is not sudden, but that the boundary
condition at x = 0 has the form v(0, t) = v0 (t), where v0 (0) = 0 and, for
t > 0, v0 is a continuous function that increases with t until it reaches a
maximum value vm at t = tm , after which it decreases.
As before, we may suppose the characteristics of family (a) to be straight
lines along which σ, v, and ε are constant. Rather than all of them emanating
from (0, 0), however, each one originates from a point on the t-axis such that
v = v0 (t). We may integrate the characteristic relations along families (b)
and (c) to obtain Z ε
v= c(ε)dε, σ = F (ε).
0
The second of these equations, however, is valid only in a loading process,
that is, only as long as ∂σ/∂t > 0. Since σ and v are monotonically related,
at x = 0 this condition prevails only for t < tm .
What happens then? As first demonstrated by Rakhmatulin [1945], a
front behind which ∂σ/∂t ≤ 0, called an unloading wave [see Figure 7.2.8(b)],
begins to propagate into the plastically deforming region of the bar. Since in
this region ∂σ/∂t > 0, across this front the stress rate ∂σ/∂t, and therefore
also the strain rate ∂ε/∂t and the acceleration ∂v/∂t, are discontinuous. An
unloading wave is therefore an acceleration wave.
The calculation of the unloading wave in the xt-plane must be performed
step by step. Its speed, say cu , is initially c(εm ), where εm is the strain at
(0, tm ). An example of a graphic construction, due to Rakhmatulin and
Shapiro [1948], is shown in Figure 7.2.9 (taken from Cristescu [1967]) for a
bar in which the stress at the end of the bar first increases monotonically,
then remains constant for a while, and finally decreases monotonically to
zero. As can be seen, the speed of the unloading wave increases as the wave
t t
!!S

!!
4

P
3 ``` Y
H
HH
. . . .
. .
`
..
.......
........ 2 . . Unloading wave
......................................................................
1...............................
. . . . . . .
.. . Constant-state region
...................... . .....
.............
...
.
0

Y
−σ x
σ m σE 0
Figure 7.2.9. Approximate construction of unloading wave (Cristescu [1967]).
propagates. It was shown by Lee [1953] that the unloading-wave speed is

bounded by
c(ε) ≤ cu ≤ ce .
A proof of this result is given in Section 7.3, in which the general theory of
elastic-plastic acceleration waves is studied. In his first paper, Rakhmatulin
[1945] erroneously showed the unloading wave as attaining speeds faster
than that of the elastic wave (see the discussion by Bodner and Clifton
[1966]). More information on wave propagation in elastic-plastic and elastic-
viscoplastic bars can be found in the books by Cristescu [1967], Chapters II
and III, and Nowacki [1978], Chapter III.

1. Discuss how a shock front might form in an elastic-plastic bar whose
stress-strain curve beyond the initial yield stress σE is given by σ =
σE {1 + [(ε − σE /E)/α]2 }. Consider the limit of a rigid–plastic bar
(E = ∞).
2. For a given impact velocity v0 , find the distribution of permanent strain

following impact in a rigid–plastic bar of length L when the stress-
strain relation beyond the initial yield stress is given by F (ε) = σE [1 +
(ε/α)1/m ]. Letting αp= 1 and m = 5, plot the strain distributions for
various values of v0 / σE /ρ.
3. Repeat Exercise 2 for the stress-strain relation of Exercise 1 with α = 1.
4. An elastic-plastic bar with linear work-hardening, of initial length L,

strikes a rigid obstacle at t = 0. Find the rebound time in terms of
L/ce and α = cp /ce when v0 /vE is 3, 4, and 5.
5. A semi-infinite elastic-plastic bar with linear work-hardening (with tan-

gent modulus E 0 in the plastic range) is subjected at time t = 0 to a
stress σ0 > σE which remains constant until t = t0 , whereupon it is
removed. Discuss the motion of the bar for various ranges of σ0 .
6. Repeat Exercise 5 for bars with the stress-strain relations of Exercises

1 and 2.
7. A semi-infinite viscoplastic bar with the constitutive equation (7.2.16)

is subjected to a suddenly applied stress or velocity. Find the equation
governing the decay of the stress discontinuity along the initial shock
front. Show that the decay is exponential if g(σ, ε) is a linear function
of its arguments for σ > σE .
Section 7.3 Three-Dimensional Waves

Three-dimensional waves in rate-independent elastic-plastic solids appear
to have been studied first by Craggs [1957], who found relations for the
speeds of plastic waves on the basis of the characteristics of the govern-
ing differential equations. As in the one-dimensional case, the character-
istic speeds coincide with those of acceleration waves, and in this section
only the acceleration-wave approach is followed. Furthermore, only rate-
independent elastic-plastic solids are discussed. In elastic-viscoplastic solids
Section 7.3 / Three-Dimensional Waves 453
the wave speeds are those of elastic waves, while the study of specific dy-
namic boundary-value problems must, as a rule, be pursued numerically,
except when the constitutive model is so simple that analytical methods,
analogous to those used for linearly viscoelastic solids, may be used; exam-
ples of both approaches are given in the book by Nowacki [1978]. Accelera-
tion waves in elastic-plastic solids were first investigated by Thomas [1958,
1961] for perfect plasticity, and with greater generality by Hill [1962] and
Mandel [1962].
7.3.1 Theory of Acceleration Waves
General Theory
The general theory of acceleration waves is due to Hadamard [1903]; a
modern exposition may be found in the monograph by Truesdell and Toupin
[1960]. It is presented here in a simplified form suitable for infinitesimal
deformations. We consider an acceleration wave moving in x1 x2 x3 -space at
the speed c. The unit normal vector to the wave surface, pointing in the
direction of wave motion, will be denoted n. The jump operator [|·]| will be
used as previously defined, and the superposed dot will denote ∂/∂t. A time-
dependent field φ is assumed to be continuous across the wave but to have
possibly discontinuous derivatives. At time t, φ has the value φ(x1 , x2 , x3 , t)
at a point of the wave front. At an infinitesimally later time t + dt, the given
point will have moved to xi + ni c dt, i = 1, 2, 3, and φ will accordingly have
the value φ(x1 , x2 , x3 , t) + dφ, where
dφ = φ,i ni c dt + φ̇dt.
The continuity of φ requires [|dφ]| = 0, or
cni [|φ,i |] = −[|φ̇]| .
Since, however, it is only the component of delφ along n that undergoes a

jump, [|φ,i |] must be proportional to ni , and therefore, since nj nj = 1,
c[|φ,j |] = −[|φ̇]| nj .
The last result is known as the Hadamard compatibility condition.

When applied to the velocity field v, the condition yields
c[|vi ,j |] = −[|v̇i|] nj .
When applied to the stress field σ and combined with the momentum-
balance equation (1.3.1), it gives
[|σ̇ij|] nj = −ρc[|v̇i|] .
If, finally, the material is characterized by an effective tangent modulus ten-

sor C̄ such that
[|σ̇ij|] = C̄ijkl [|ε̇ij|] ,
then combining all the results of this paragraph yields the eigenvalue problem
(Qik − ρc2 δik )[|v̇k|] = 0,
where
def
Qik = C̄ijkl nj nl (7.3.1)
defines the acoustic tensor Q. The eigenvalues of Q, if positive, equal ρc2
with c the wave speed, while the eigenvectors are the directions of [|v̇]| or
polarization directions.
Elastic Waves
If the material is linearly elastic, then the tangent modulus tensor C̄ is
just the elastic modulus tensor C, and the corresponding acoustic tensor will
be denoted Qe :
Qeik = Cijkl nj nl . (7.3.2)
If the material is also isotropic, then C is given by Equation (1.4.10), and
Qe is given by
Qeik = (λ + µ)ni nk + µδik .
One eigenvector is obviously n, with the corresponding eigenvalue being
λ + 2µ, so that the wave speed is
s s
λ + 2µ 1−ν
ce1 = = ce ,
ρ (1 + ν)(1 − 2ν)
p
where ce = E/ρ is the elastic bar-wave speed defined in the preceding
section. Since the polarization direction coincides with that of propagation,
a wave of this nature is called longitudinal ; it is also known a dilatational
wave or a P-wave.
Any vector perpendicular to n is also an eigenvector of Q and corre-
sponds to the degenerate eigenvalue µ. The polarization being perpendicular
to the wave normal, the wave is called transverse (also a shear wave or an
S-wave), and its speed is
µ
r
e
c2 = .
ρ
As in the bar studied previously, the elastic acceleration-wave speeds in
three-dimensional solids are independent of the local field variables, so that
all disturbances, including sufficiently weak shocks, travel at these speeds. A
shock is “sufficiently weak” if it does not entail temperature and/or density
changes large enough to affect the elastic moduli.
Elastic-Plastic Waves
We now consider an acceleration wave propagating in a standard elastic-
plastic solid as described in Section 3.2, with a yield function f . If f < 0
at the wave front, then the solid is behaving elastically, and the wave is an
elastic one. It will therefore be assumed that f = 0 at the wave front. Then
the relation (3.2.10) between stress rate and strain rate, with the normality
rule (3.2.7) incorporated, may be written, following Hill [1959], in the form
1
σ̇ij = Cijkl ε̇kl − Cijmn mmn <g>, (7.3.3)
l
where
∂f
∂σij
mij = s , g = Cpqkl mpq ε̇kl , l = h + Cklmn mkl mmn ,
∂f ∂f
∂σkl ∂σkl
h being a plastic modulus, not exactly the same as in Section 3.2, but to
be specified more precisely later. Note that m is a unit tensor, that is,
mij mij = 1.
Applying the jump operator to Equation (7.3.3) yields
1
[|σ̇ij|] = Cijkl [|ε̇kl|] − Cijmn mmn [|<g>]| .
l
We may now write
[|<g>]| = η[|g]| ,
where η is a real number that depends on the ranges of g + and g − .
Several cases may be distinguished:
(a) g+ > 0, g− >0: η = 1,

(b) g+ > 0, g− ≤0: 0 < η < 1,
(c) g+ ≤ 0, g− >0: 0 < η < 1,
(d) g+ ≤ 0, g− ≤0: η = 0.
In addition, we may consider the situation where the material ahead of

the wave is about to yield (i.e., g + > 0), but is still elastic (i.e., σ̇ + = C · ε̇+ ).
Since g − > 0 also, we have
g−
η= ,
g− − g+
and thus obtain two additional cases:
(e) g − > g + : η > 1,
(f) g − < g + : η < 0.
The waves represented by these cases can be categorized as follows, ac-

cording to Mandel [1962]: case (a) represents a plastic wave, with the ma-
terial deforming plastically on both sides; case (b) an unloading wave; and
case (d) an elastic wave. Case (c), which was not considered by Mandel, can
be called a reloading wave. Cases (e) and (f) represent loading waves; the
former will here be called strong and the latter weak .
The effective tangent modulus C̄ may now be expressed as
C̄ijkl = Cijkl − ηNij Nkl ,

√
where Nij = Cijkl mkl / l. The acoustic tensor Q accordingly may be written
in the form
Qij = Qeij − ηdi dj ,
def
where Qe is given by (7.3.2), and di = Nik nk defines a vector d that Hill
[1962] calls the “traction” exerted by N on the wave front.
Analysis of Wave Speeds
The analysis to be presented is based on that of Mandel [1962]. Since Q is
symmetric, it has real eigenvalues and mutually perpendicular eigenvectors.
Whenever such an eigenvalue —psay A — is positive, it represents a a wave
propagated at the speed c = A/ρ. The symmetry of Q is due to the
assumption of an associated flow rule; the propagation of acceleration waves
in nonstandard elastic-plastic materials was also studied by Mandel [1964],
whose conclusions are summarized later.
In what follows, the eigenvalues of Q and Qe will be denoted Aα and
Aeα (α = 1, 2, 3), respectively, with the ordering A1 ≥ A2 ≥ A3 and Ae1 ≥
Ae2 ≥ Ae3 . If a Cartesian basis is chosen along the eigenvectors of Qe , then
the characteristic equation of Q is
F (A) = 0 (7.3.4)
where
F (A) = det(Q − IA)

Ae − ηd2 − A −ηd1 d2 −ηd1 d3
1 1
= −ηd1 d2 A2 − ηd22 − A
e −ηd2 d3

−ηd1 d3 −ηd2 d3 Ae1 − ηd23 − A

= (Ae1 − A)(Ae2 − A)(Ae3 − A) − η[d21 (Ae2 − A)(Ae3 − A)

+ d22 (Ae1 − A)(Ae3 − A) + d23 (Ae1 − A)(Ae2 − A)].
The character of the roots of Equation (7.3.4) depends on η. The simplest

case is η = 0, corresponding to an elastic wave. In this case, of course,
Aα = Aeα (α = 1, 2, 3).
If η > 0, as in cases (a)–(c) and in case (e), then F (Ae1 ) = −ηd21 (Ae1 −
Ae2 )(Ae1 − Ae3 ) ≤ 0. Similarly, F (Ae2 ) ≥ 0, F (Ae3 ) ≤ 0, and F (−∞) = +∞.
Consequently,1
A3 ≤ Ae3 ≤ A2 ≤ Ae2 ≤ A1 ≤ Ae1 .
Furthermore, the Aα decrease monotonically with η. If Apα denotes the value
of Aα for η = 1 (plastic wave), and Auα a value of Aα for 0 < η < 1 (unloading
and reloading), then we have the inclusion
Apα ≤ Auα ≤ Aeα , α = 1, 2, 3.
Here we see the bounding of the wave speed of an unloading wave, previously
obtained by Lee [1953] for waves in a thin bar.2
The speeds of strong loading waves, case (e), are even slower than those
of plastic waves. The inclusions are
p p p
Asl
3 ≤ A3 , Ae3 ≤ Asl
2 ≤ A2 , Ae2 ≤ Asl
1 ≤ A1 .
In fact, Asl
3 may be zero, in which case the surface does not propagate.
In a weak loading wave, case (f), η is negative and therefore the nature
of the roots changes drastically. The ordering of the wave speeds is
p p
Ae3 ≤ Awl
3 ≤ A2 , Ae2 ≤ Awl
2 ≤ A1 , Ae1 ≤ Awl
1 .
The existence of waves propagating faster than elastic waves (in a bar) was
apparently first noted by Lee [1953]. Mandel [1962] gave, as an example of
such a wave moving at an infinite speed, the loading wave in a finite bar
following reflection from a fixed end.
An analysis of plastic waves in a nonstandard material, also due to Man-
del [1964], shows that (a) the eigenvectors of Q are not mutually perpendic-
ular, (b) Ap3 may be negative (representing a slip or rupture surface rather
than a wave), and (c) Ap1 may be greater than Ae1 . While soils are usu-
ally modeled as nonstandard in static problems, theories of waves in soils
have typically been based on standard elastic-plastic models, for example by
Grigorian [1960] and by Chadwick, Cox, and Hopkins [1964].
Isotropically Elastic Materials

If the material is elastically isotropic, with the modulus tensor given by
(1.4.10), and if in addition it is plastically incompressible (tr ε̇p = 0), so that
mkk = 0, then
Cijkl mij mkl = 2µmij mij = 2µ,
1
It was also shown by Mandel [1962] that A3 > 0 for η ≤ 1, except that Ap3 may be zero
if the material is perfectly plastic.
2
As pointed out in the preceding section, Rakhmatulin’s [1945] finding of an unloading
wave propagating faster than the elastic wave is erroneous.
l = h + 2µ, g = 2µmkl ε̇kl ,

and
2µ
Nij = √ mij .
h + 2µ
If plastic isotropy is also assumed, then the hardening modulus h can be
related very simply to the uniaxial tangent modulus as well as to the tangent
modulus in shear. Recalling the definition of g and multiplying every term
of Equation (7.3.3) by mij , with contraction implied, we obtain
1 2µh

mij σ̇ij = 1 − Cijkl mij mkl g = mkl ε̇kl .
h + 2µ h + 2µ
In a quasi-static uniaxial test, the left-hand side is just m11p
σ̇11 . Moreover,
mij = 0 for i! = j and m22 = m33 = − 2 m11 , so that m11 = 2/3, and
1
r
2
mkl ε̇kl = (ε̇11 − ε̇22 ).
3
Consequently,
2µh
σ̇11 = (ε̇11 − ε̇22 ).
2µ + h
Plastic incompressibility further implies that
σ̇11 = (3λ + 2µ)(ε̇11 + 2ε̇22 ).
Eliminating ε̇22 , we obtain the tangent compliance

ε̇11 λ + µ + 2µ(3λ + 2µ)/3h 1 2
= = + ,
σ̇11 (3λ + 2µ)µ E 3h
so that h = 23 H, where H = dσY /dεp is the uniaxial tangent plastic modulus.

It can also be easily shown that h is twice the shearing tangent plastic
modulus dτY /dγ p .
The eigenvector of Qe corresponding to the fastest elastic wave (the
longitudinal wave) is of course n. If this is identified with e1 , then d is given
by
2µ
di = δik Nk1 = √ mi1 .
h + 2µ
Since any vector perpendicular to n is also an eigenvector of Qe , we are free
to choose e2 and e3 in the tangent plane of the wave; let us choose them so
that d3 = 0. The eigenvalue problem now reads
 
λ + 2µ − ηd21 − A −ηd1 d2 0
0 = (Q − AI)[|v̇]| =  −ηd1 d2 µ − ηd22 − A 0  [|v̇]| .
 
0 0 µ−A
(7.3.5)
Obviously, one eigenvector is e3 , with the corresponding eigenvalue A2 =

µ = Ae2 = Ae3 (necessary because Ae3 ≤ A2 ≤ Ae2 ). Hence, at any η, an elastic
transverse wave can propagate. If η > 0, then we also have A3 ≤ Ae2 ≤ A1 ≤
Ae1 , a result first derived by Craggs [1957]. The waves corresponding to A1
and A3 are in general neither longitudinal nor transverse, unless d1 d2 = 0.
When neither d1 nor d2 is zero, the waves are called longitudinal–transverse.
7.3.2 Special Cases
Purely Longitudinal and Transverse Waves

(a) If d2 = 0, then A3 = µ and the polarization is along e3 . In this case
both transverse waves are elastic, while for the longitudinal wave we obtain
4µ2
A1 = λ + 2µ − ηd21 = λ + 2µ − η m2 .
h + 2µ 11
The tensor m can be shown to have the same form as in the quasi-static
uniaxial test: m13 = 0 because d3 = 0 by choice of basis, m12 = 0 because
d2 = 0 by hypothesis, m23 = 0 because [|ε̇23|]p= 0, and m22 = m33 = − 12 m11
from the resulting symmetry; hence m11 = 2/3, and
4ηµ
A1 = λ + 2µ − . (7.3.6)
3(1 + h/2µ)
Note that for h = 0 (perfect plasticity) and η = 1 (plastic wave) we have
A1 = λ + 23 µ = K, the elastic bulk modulus. In this case the longitudinal
wave may be said to propagate as though the material were a fluid.
(b) If d1 = 0, then, by contrast, it is the longitudinal wave that is elastic
(A1 = λ + 2µ), and in addition to the elastic transverse wave with A2 = µ,
there is a transverse wave with
4µ2
A3 = µ − ηd22 = µ − η m2 .
h + 2µ 12
It can easily be shown √that this case is equivalent to quasi-static simple
shear, so that m12 = 1/ 2, and
ηµ
A3 = µ − . (7.3.7)
1 + h/2µ
For η = 1 (plastic wave), A3 reduces to the tangent modulus in shear,
µh/(2µ + h). If h = 0 (perfect plasticity), then A3 = (1 − η)µ, so that a
transverse plastic wave (η = 1) or a transverse strong loading wave (η > 1)
cannot be propagated.
Longitudinal–Transverse Waves
In an isotropic and homogeneous half-space, say x1 > 0, whose bound-
ary x1 = 0 is subject to time-varying tractions (or displacements) that are
uniform over the plane (i.e., independent of x2 and x3 ), all the field variables
may be assumed to be functions of x1 and t only, and the wave is said to be
plane. A plane wave is longitudinal if the traction or displacement is purely
normal, and transverse if the traction or displacement is purely tangential. If
the boundary tractions or displacements include both normal and tangential
components, then the wave is longitudinal–transverse.
The propagation of longitudinal–transverse plane waves in an elastic-
perfectly plastic half-space was studied by Bleich and Nelson [1966], and in
an elastic-plastic half-space with work-hardening by Ting and Nan [1969]; for
further references, see Nowacki [1978], Section 22. Here, we limit ourselves
to finding expressions for the wave speeds through a direct application of
Equation (7.3.5).
If the tangential traction is in the x2 -direction, then the half-space can be
assumed to be in a state of plane strain, so that, if the material is isotropic,
σ13 = σ23 = 0. Furthermore, all field quantities, including the velocity, can
be assumed to be independent of x2 , so that ε̇22 = 0. The Lévy–Mises flow
rule requires that σ22 = σ33 , and the stress-deviator tensor is given by
 
s τ 0
s =  τ − 12 s 0 ,
 
0 0 − 21 s
where s = 2
3
(σ11 − σ22 ) and τ = σ12 . The Mises yield criterion may be
written as
3 2
s + τ 2 = k2 . (7.3.8)
4
The “traction” vector d has the components
s s
2 µ 2 µ
d1 = s, d2 = τ, d3 = 0.
h + 2µ k h + 2µ k
The characteristic equation is, accordingly,

! !
2µ2 s2 2µ2 τ 2 2 4µ4 s2 τ 2
λ + 2µ − A − η µ−A−η −η = 0.
h + 2µ k 2 h + 2µ k 2 (h + 2µ)2 k 4
Upon multiplying the equation by 2k(h + 2µ)/µ, we may rewrite it as
3s2 (A − µ){A − [K + 34 (1 − η)µ]}

h
+4τ 2 (A − λ − 2µ)[A − (1 − η)µ] + 2k 2 (A − λ − 2µ)(A − µ) = 0.
µ
It can easily be ascertained that the special cases τ = 0 and s = 0 lead to

Equations (7.3.6) and (7.3.7), respectively. For a plastic wave (η = 1) in an
elastic-perfectly plastic medium (h = 0) the equation reduces to the result

of Bleich and Nelson [1966]:
3s2 (A − µ)(A − K) + 4τ 2 A(A − λ − 2µ) = 0.
A problem closely related to that of longitudinal–transverse waves plane

waves in a half-space is that of longitudinal–torsional waves in a long thin-
walled cylindrical tube. This problem was studied by Clifton [1966, 1968]
and T. C. T. Ting [1969].
Spherical Radial Wave

Purely longitudinal waves arise when the loading is spherically or cylin-
drically radial, such as that produced by the explosion of a point charge or
a line charge in an infinite medium. The cylindrical problem is relatively
complicated, since the condition of plane strain must be enforced, and the
problem depends on the yield criterion chosen. The spherical problem is
fairly simple; it was first treated by Lunts [1949].
In an isotropic and plastically incompressible elastic-plastic solid, a spher-
ically symmetric state of deformation requires εpθ = − 21 εpr , and is equivalent
to a state of uniaxial stress equal to σr − σθ . The yield criterion may be
written as
|σr − σθ | = σY (εpr ),
where σ = σY (εp ) is the uniaxial relation between yield stress and plastic
strain. If, now, H is again the uniaxial plastic modulus, then
1
ε̇pr = <sgn (σr − σθ )(σ̇r − σ̇θ )>
H
when the yield criterion is obeyed. It will be assumed that the loading is
compressive, that is, that σr − σθ = −σY . If v denotes the radial velocity,
and if (·)0 = ∂(·)/∂r, then the radial and tangential strain rates are given by
σ̇r − 2ν σ̇θ <σ̇θ − σ̇r >

v0 = − ,
E H (7.3.9)
v (1 − ν)σ̇θ − ν σ̇r <σ̇θ − σ̇r >
= + .
r E 2H
The equation of motion is
σr − σθ
σr0 + 2 = ρv̇.
r
Applying the jump operator to this equation leads to [|σr0 |] = ρ[|v̇]| . If, as
before, A = ρc2 , where c is the wave speed, then the Hadamard compatibility
conditions give [|v 0|] = ρ[|σ̇r|] /A. Since the velocity itself is continuous at
an acceleration wave front, application of the jump operator to Equations

(7.3.9) produces the two homogeneous linear equations for [|σ̇r|] , [|σ̇θ|] :
1 1 1 1 2ν

+ − [|σ̇r|] − + [|σ̇θ|] = 0,
H E A H E
1 ν 1 1−ν

− + [|σ̇r|] + + [|σ̇θ|] = 0,
2H E 2H E
if it is assumed that the wave is a compressive plastic wave, that is, σθ −σr =
σY and σ̇θ − σ̇r > 0. Setting the determinant of the two equations equal to
zero gives
E + 2(1 − ν)H E
A= .
3E + 2(1 + ν)H 1 − 2ν
Since h = 23 H, it can easily be seen that this equation is equivalent to (7.3.6)
with η = 1.
Cylindrical Shear Wave

A purely transverse wave will be propagated cylindrically if, for example,
a rigid axle, embedded in an infinite medium, suddenly has a torque applied
to it. If the medium is isotropic, then it may be assumed that the only
velocity component is the circumferential velocity vθ = v(r, t) and the only
stress component is σrθ = σθr = τ (r, t). The equation of motion is
τ
τ0 + 2 = ρv̇,
r
from which [|v 0|] = [|τ̇ |] /A. If the shear strain is γ, then
v τ̇ 3<τ̇ sgn τ >

γ̇ = v 0 − = + ,
r µ H
so that
1 3

[|v 0|] = + [|τ̇ |] ,
µ H
and thus
µH
A= ,
H + 3µ
a result equivalent to (7.3.7) with η = 1. The propagation of the wave was
studied by Rakhmatulin [1948].

1. Show that in an elastically isotropic material, the plastic modulus h is
twice the tangent plastic modulus in shear.
2. Find the eigenvalues A1 and A3 of Equation (7.3.5) in the general case.

3. Show that the state of stress in a purely transverse wave is that of

simple shear.
4. Using suitable assumptions, establish the governing equations of com-

bined longitudinal–torsional wave propagation in a long thin-walled
elastic-plastic cylindrical tube.
5. For a Mises material with linear work-hardening, find explicit expres-

sions for the wave speeds of longitudinal–transverse plastic waves in
terms of the ratio of normal stress to shear stress at the wave.
6. Find the speed of a spherically radial elastic-plastic wave as a function

of η.
Chapter 8
Large-Deformation Plasticity
Section 8.1 Large-Deformation Continuum Mechan-

ics
8.1.1 Continuum Deformation
As a model of mechanical behavior, plasticity theory is applicable primar-

ily to those solids that can experience inelastic deformations considerably
greater than elastic ones. But the resulting total deformations, and the rota-
tions accompanying them, may still be small enough so that many problems
can be solved with small-displacement kinematics, and this is the situation
that has, with a few exceptions, thus far been addressed in this book. How-
ever, when strains or rotations become so large that they cannot be neglected
next to unity, the mechanician must resort to the theory of large or finite
deformations.
Since the reference and displaced configurations of a body, as discussed
in Section 1.2, may be quite different, it is appropriate to use a notation that
makes the difference apparent, a notation based on that introduced by Noll
[1955] and made current by the monograph of Truesdell and Toupin [1960].
A material point is denoted simply X, and its Lagrangian coordinates are
denoted XI (I = 1, 2, 3); a point in the current or displaced configuration is
denoted x, and its Cartesian coordinates (sometimes called Eulerian coordi-
nates) are xi (i = 1, 2, 3). Different indices are used for the two coordinate
systems because the corresponding bases, the Lagrangian basis (eI ) and the
Eulerian basis (ei ), are, in principle, independent of each other.
Whenever direct notation is used for tensors in this chapter, vectors are
treated as column matrices and second-rank tensors as square matrices; thus
uT v is used for the scalar product u · v, while AB denotes the second-rank
tensor with components Aik Bkj . Furthermore, the scalar product of two
second-rank tensors is A : B = tr (AT B). If Γ is a fourth-rank tensor,
465
466 Chapter 8 / Large-Deformation Plasticity
then Γ : A denotes the second-rank tensor with components Γijkl Akl . In

particular, a fourth-rank tensor with components ∂Uij /∂Vkl will be denoted
∂U/∂V, and the second-rank tensor with components ∂Uij /∂Vkl Akl will be
denoted (∂U/∂V) : A.
Deformation Gradient
The motion of the body is described by the functional relation
x = χ(X, t).
When χ is continuously differentiable with respect to X, then the defor-

mation gradient at X is the second-rank tensor F(X, t) whose Cartesian
components are
∂χi
FiI = .
∂XI
Note that these components are evaluated with respect to both bases simul-
taneously; in the terminology of Truesdell and Toupin [1960], the deforma-
tion gradient is a two-point tensor .
If, in a neighborhood of the material point X, the function χ(X, t) is
invertible — in other words, if the material points in the neighborhood are in
one-to-one correspondence with their displaced positions — then, by the im-
plicit function theorem of advanced calculus, the matrix of components
of F(X, t) (the Jacobian matrix) must be nonsingular, that is, J(X, t) 6= 0,
def
where J(X, t) = det F(X, t) is the Jacobian determinant. If we consider
only displaced configurations that can evolve continuously from one another,
then since J = 1 when the displaced and reference configurations coincide,
we obtain the stronger condition J(X, t) > 0.
−1
The inverse of F(X, t), denoted F−1 (X, t), has the components FIi (X, t)
−1
= ∂XI /∂xi |x=χ(X, t) . Note that F is also a two-point tensor, but of a dif-
ferent kind from F: while the components of the latter (FiI ) are such that
the first index refers to an Eulerian and the second to a Lagrangian basis,
−1
in the former it is the reverse (FIi ).
Local Deformation
Consider two neighboring material points X and X0 = X + u, where u is
a “small” Lagrangian vector emanating from X. If the displaced positions
of X and X0 are respectively given by x and x0 , then, F being continuous,
we have (in the matrix-based direct notation that is used throughout this
chapter)
x0 = x + F(X, t)u + o(|u|) as |u| → 0.
Since a rigid-body displacement does not change distances between points,

Section 8.1 / Large-Deformation Continuum Mechanics 467
let us compare the distance between x and x0 with that between X and X0 :
q
|x0 − x| = (x0 − x)T (x0 − x)
q
= uT FT (X, t)F(X, t)u + o(|u|2 )
q
= uT C(X, t)u + o(|u|),
def
where C = FT F is known, in the Noll–Truesdell terminology, as the right
Cauchy–Green tensor . The notation will henceforth be simplified by writing
F for F(X, t) and so on. The components of C are given by
CIJ = χi ,I χi ,J ,
where (·),I = ∂(·)/∂XI . C is a Lagrangian tensor field which, moreover, is

symmetric (C = CT ) and positive definite (uT Cu > 0 for any u 6= 0).
Stretch and Strain
The stretch at (X, t) along a direction u is defined by
s
|χ(X + hu, t) − χ(X, t)| uT Cu
λu = lim = .
h→0 h|u| uT u
We have λu = 1 for every u if and only if C = I; then the displacement is
locally a rigid-body displacement. If u(α) (α = 1, 2, 3) are the eigenvectors
def
of C and if λα = λu(α) , then the λ2α are the eigenvalues of C, and the λα
def 1
(the principal stretches) are the eigenvalues of U = C 2 .
A strain is a measure of how much a given displacement differs locally
from a rigid-body displacement. In particular, a Lagrangian strain is a mea-
sure of how much C differs from I. The following Lagrangian strain ten-
sors are commonly used: (a) the Green–Saint-Venant strain tensor1 already
mentioned in Section 1.2, E = 12 (C − I), with eigenvalues 21 (λ2α − 1); (b) the
conventional strain tensor Ee = U − I, with eigenvalues λα − 1 (the princi-
pal conventional strains); and (c) the logarithmic strain tensor El = ln U,
with eigenvalues ln λα (the principal logarithmic strains). Note that all three
strains may be regarded as special cases of (1/k)(Uk − I), with (a), (b), and
(c) corresponding respectively to k = 2, k = 1 and the limit as k → 0.
Furthermore,
1 k 1
(λ − 1) = [(1 + λ − 1)k − 1] = λ − 1 + o(|λ − 1|),
k k
so that )
Ee
= E + o(||E||),
El
1
Often called simply the Lagrangian strain tensor.
where ||E|| denotes some measure of the magnitude (a norm) of E. In other

words, the different Lagrangian strain tensors are approximately equal when
they are sufficiently small. For large strains the Green–Saint-Venant strain
tensor is analytically the most convenient, except in cases where the principal
directions u(α) are known a priori.
Since U is symmetric and positive definite, we can form the two-point
tensor R = FU−1 . Note that
RT R = U−1 FT FU−1 = U−1 U2 U−1 = I,
that is, R is orthogonal. Also, det U = J, so that det R = J/J = 1, and

consequently, R is a proper orthogonal tensor, or a rotation. The decompo-
sition F = RU is the right polar decomposition of F, and U is called the
right stretch tensor ; this is why C is called the right Cauchy–Green tensor.
R is usually called simply the rotation tensor .
If the displacement is locally a rigid-body one, then we simply have
F = R. To study the general case, let us consider points near X given by
X(α) = X + hu(α) ,
where u(α) is, as above, an eigenvector of C. The displaced images of these

points are
x(α) = x + hv(α) + o(h),
where
v(α) = Fu(α) = RUu(α) .
Since, however, u(α) is also an eigenvector of U, it follows that Uu(α) =
λα u(α) , and therefore
v(α) = λα Ru(α) ,
so that R represents the rotation of the eigenvectors of C.
For an arbitrary vector u, Uu is not in general parallel to u. In fact, if
we consider the ellipsoid centered at X, given by
uT Cu = r2 ,
with principal semiaxes r/λα , we see that its displaced image is approxi-
mately the sphere of radius r centered at x. The ellipsoid given by the
preceding equation is called the reciprocal strain ellipsoid .
We may also ask what is the effect of the displacement on the sphere
of radius r, centered at X, in the reference configuration. If x + v is the
displaced image of X + u, then v = dotFu, so that uT u = dotvT B−1 v
(where B = FFT ), and vT B−1 v = r2 defines an ellipsoid centered at x with
principal semiaxes λα r; this ellipsoid is called simply the strain ellipsoid .
The tensor B−1 is called the Finger deformation tensor , while B is called
(again in the terminology of Truesdell et al.) the left Cauchy–Green tensor .

It is an “Eulerian” tensor, and its eigenvalues are also λ2α . Its name derives,
as may be surmised, from the left polar decomposition of F (see Exercise 1).
It can easily be shown that B = RCRT .
The most commonly used “Eulerian” strain tensor is the Almansi strain
tensor Ea = 21 (I − B−1 ). It is easy to show that E = FT Ea F, so that
Ea = RU−1 EU−1 RT . Hence, in order for Ea to be approximately equal to
E (i.e. to have Ea = E + o(||E||)), it is necessary not only for ||E|| but also
for ||R − I|| to be small.
In many treatments the right Cauchy–Green tensor C is given another
definition, namely, as a metric tensor . If the Lagrangian coordinates XI
are used to describe points in the displaced body, then they are no longer
Cartesian coordinates, since the surfaces XI = constant are not necessarily
planes. An infinitesimal vector dx in the displaced body has the square of
its magnitude given by
dx · dx = CIJ dXI dXJ .
Hence C is often called the metric tensor in the displaced (strained) body.
The identity I is accordingly regarded as the metric tensor in the unstrained
body. The analysis of continuum deformation based on metric tensors is
usually carried out by using curvilinear coordinates to begin with (see, e.g.,
Green and Zerna [1968], Eringen [1962], Marsden and Hughes [1983]). For
our purposes, its main usefulness is in the derivation of compatibility con-
ditions, that is, the finite-deformation analogue of Equation (1.2.4), and to
that end Cartesian coordinates are adequate. The result is usually given as
the vanishing of a fourth-rank tensor called the Riemann–Christoffel tensor
or the curvature tensor , being the necessary (locally also sufficient) condition
for a symmetric positive definite tensor field C to be derivable from a con-
figuration χ. It can be shown that the result is equivalent to the symmetric
second-rank tensor equation
eIKM eJLN [EKL ,M N

− 12 CP−1Q (ELP ,M +EM P ,L −ELM ,P )(EKQ ,N +EN Q ,K −EKN ,Q )] = 0,
(8.1.1)
which can be seen to reduce to (1.2.4) for sufficiently small strains.
A fundamental difference between the Lagrangian and the Eulerian de-
formation and strain tensors must be emphasized. The Lagrangian tensors
transform according to the usual rules (Section 1.1) under a rotation of the
Lagrangian basis, while they are invariant under a rotation of the Eulerian
basis. To see the invariance, let (e∗i ) denote the rotated Eulerian basis, with
e∗i = Qij ej , Q being orthogonal. Since x∗i = Qij xj , it follows that χ∗i (X, t)
= Qij χj (X, t), and hence FiI∗ = Qij FjI . These relations may be written
as x∗ = Qx, χ∗ (X, t) = Qχ(X, t), and F∗ = QF. Consequently, C∗ =
FstarT F∗ = FT QT QF = C. Since the Eulerian basis is, in a sense, fixed in

space, invariance under its rotation is related to the isotropy of space itself,
and is invoked in the formulation of fundamental physical principles.
Conversely, Eulerian tensors are invariant under a rotation of the La-
grangian basis; such a rotation may represent, for example, a rotation of the
given body in space. This kind of invariance may, accordingly, be used to es-
tablish material isotropy — the fact that the mechanical response of a body
does not depend on its orientation. Eulerian tensors transform according to
the usual rules under a time-independent rotation of the Eulerian basis; but
only the ones called objective do so when the rotation is time-dependent.
The objectivity of tensors is discussed in 8.1.2.
Area and Volume Deformation

Consider again a material point X and two neighboring points X0 =
X + hu0 , X00 = X + hu00 , defined by two vectors u0 , u00 and a small real
number h. The vector area of the triangle formed by the three points is
1
A0 = h2 u0 × u00 ,
2
or, in components
1
A0I = h2 eIJK u0J u00K .
2
The displaced positions of X, X and X00 are x, x0 = x + hv0 + o(h) and
0
x00 = x + hv00 + o(h), respectively, where v0 = Fu0 and v00 = Fu00 . The
displaced area is
1
A = h2 v0 × v00 + o(h2 )
2
or
.
Ai = 21 h2 eijk vj0 vk00 = 12 h2 eijk FjJ FkK u0J u00K
= 41 h2 eijk (FjJ FkK − FjK FkJ )u0J u00K
= 14 h2 eijk FjJ FkK (u0J u00K − u0K u00J )
= 41 h2 eijk FjJ FkK eIJK eILM u0L u00M

1 1 2 0 00 −1
= 2 eijk eIJK FjJ FkK 2 h eILM uL uM = (det F)FIi A0I .
In direct notation,
A = JFT −1 A0 + o(|A0 |).
If, now, X000 = X + hu000 is a fourth point, then the volume of the tetra-
hedron in the reference configuration is V0 = 31 hA0 · u000 = 13 hAT0 u000 . Since
the displaced position of X000 is x000 = X + hFu000 + o(h), the displaced volume
is
1
V = hAT Fu000 + o(h3 ) = JV0 + o(V0 ).
3
The Jacobian determinant is thus seen as having the important property of

measuring the ratio between displaced and referential infinitesimal volumes,
that is,
dV = J dV0 . (8.1.2)
If ρ and ρ0 denote the mass density in the displaced and reference config-
urations, respectively, then conservation of mass in an infinitesimal volume
requires that ρ dV = ρ0 dV0 . Therefore,
ρ0
J= .
ρ
Deformation-Rate Measures
The velocity field at time t is defined by
∂χ(X, t)
χ̇(X, t) = .
∂t
In the Eulerian description, the velocity field is given by v(x, t), and the
velocity gradient L, an Eulerian tensor, is defined by Lij = ∂vi /∂xj . In view
of the definition of F, Ḟ is given by
∂ ∂χi ∂ χ̇i ∂vi ∂χj

ḞiJ = = = = Lij FjJ ,
∂t ∂XJ ∂XJ ∂xj ∂XJ
or
Ḟ = LF. (8.1.3)
A rigid-body motion is characterized by the right Cauchy-Green tensor
field C being constant in time (not necessarily equal to I, since the initial
configuration of the motion may be deformed with respect to the reference
configuration), or, equivalently, Ċ = 0. Using differential calculus, we obtain
Ċ = FT Ḟ + ḞT F = FT (L + LT )F = 2FT DF,
def
where D = 21 (L + LT ) is the (Eulerian) deformation-rate tensor or (in Trues-
dell’s terminology) stretching tensor; it was denoted d in Chapter 3. An
immediate corollary of the last result is
Ė = FT DF. (8.1.4)
def
If W = 21 (L − LT ), then obviously L = D + W. The antisymmetric tensor
W is called the spin, vorticity or (Truesdell) spinning tensor.
The deformation-rate tensor D is of fundamental significance in the
study of deformational motions of continua. Let us consider two neighbor-
ing points1 X, X + dX, with their displaced positions at time t designated
1
The “differentials” dX have exactly the same meaning as the small vectors hu used
in the preceding discussion, but their use makes terms of the type o(h) unnecessary.
respectively by x and x + dx, so that dx = F(X, t)dX. Because of this

last relation, it is easy to obtain the material (Eulerian) time derivative, as
defined in Section 1.3, of dx:
D
dx = ḞdX = Ldx.
Dt
The preceding equation shows us that L is the operator that transforms
dx
√ into its material time derivative. Now if the magnitude of dx is ds =
dxT dx, then
D 1 D 1
ds = (dxT dx) = dxT D dx.
Dt 2ds Dt ds
If n is a unit vector, so that dx = n ds, then
D
ln ds = nT Dn,
Dt
so that with respect to a Cartesian basis (ei ), the diagonal components D11 ,
D22 , and D33 are just the logarithmic strain rates (see Section 2.1) along e1 ,
e2 and e3 , respectively.
What is the meaning of the off-diagonal components? We may answer
the question by considering two small Lagrangian vectors dX(1) , dX(2) , such
that χ(X + dX(α) , t) = x + dx(α) + o(|dx(α) |), α = 1, 2, with dxα = FdXα .
With, furthermore, dxα = nα dsα , we have dx(1) ·dx(2) = ds1 ds2 cos θ, where
cos θ = n(1) · n(2) . But we also have dx(1) · dx(2) = dX(1)T CdX(2) , so that
D
(dx(1) · dx(2) ) = dX(1)T ĊdX(2) = 2dX(1)T FT DFdX(2)
Dt
= 2dx(1)T D dx(2) = 2ds1 ds2 n(1)T Dn(2) .
On the other hand,

D D D
(dx(1) · dx(2) ) = cos θ (ds1 ds2 ) − ds1 ds2 sin θ θ.
Dt Dt Dt
Now, if n(1) and n(2) are instantaneously perpendicular, with sin θ = 1, then
1D
n(1)T Dn(2) = − θ.
2 Dt
But −(D/Dt)θ is the rate at which an instantaneously right angle becomes
acute — precisely what we know as the shearing rate.
We can easily determine the rate of change of infinitesimal volumes by
performing the material time differentiation of Equation (8.1.2):
D D

dV = J dV0 = J˙ dV0 = (∂J/∂F) : Ḟ dV0 ;
Dt Dt
but ∂J/∂F = JFT −1 , so that

1 D D
dV = tr L = tr D = ∇ · v = ln dV.
dV Dt Dt
A motion in which the volume remains constant (called an isochoric motion)
is therefore characterized by ∇ · v = 0 throughout. The velocity field v is
then called solenoidal .
8.1.2 Continuum Mechanics and Objectivity
Stress
The stress tensor σ was introduced in Section 1.3 in connection with in-
finitesimal-displacement theory. However, the procedure used in defining it
is valid under finite deformation, provided that the oriented surface element
n dS is in the current configuration. The stress tensor is then called the
Cauchy stress, and following Truesdell et al. it will be denoted T. The
equations of motion (1.3.3) are exact if all quantities occurring in them are
Eulerian, that is, if ρ denotes mass per current volume and (·),j = ∂(·)/∂xj ,
the xj being Eulerian coordinates. It can accordingly be shown that the
deformation power Pd defined in Section 1.4 is given exactly by
Z
Pd = T : D dV.
R
Note that as T is symmetric, T : D = T : L, since T : W = 0. Now,

from Equation (8.1.3), L = ḞF−1 , and therefore
T : L = tr (TḞF−1 ) = tr (F−1 TḞ) = (TFT −1 ) : Ḟ,
so that an alternative expression for Pd is

Z
Pd = T̄ : Ḟ dV0 ,
R0
where R0 is the region occupied by the body in the reference configura-

def
tion, and T̄ = JTFT −1 is a two-point tensor known as the first Piola–
Kirchhoff stress or, more simply, as the Piola stress. In indicial notation,
−1
T̄iJ = JTik FJk .
Since the area deformation discussed in the preceding subsection can be
written, for infinitesimal areas, as
dA = JFT −1 dA0 ,
it follows that
T̄ dA0 = T dA.
Consequently, the Piola stress is the multiaxial generalization of the uniaxial

nominal (engineering) stress.
It can be shown in several ways that the Lagrangian equations of motion
can be written as
T̄iJ ,J +ρ0 bi = ρ0 ai , (8.1.5)
where a and b denote, as before, the acceleration and the body force per unit
mass, respectively. The derivation of Equation (8.1.5) is left to an exercise.
Yet another Lagrangian form for the deformation power can be derived
by using Equation (8.1.4). Since D = FT −1 ĖF−1 , it follows that
T : D = tr (TFT −1 ĖF−1 ) = tr (F−1 TFT −1 Ė),
and therefore Z
Pd = S : Ė dV0 ,
R0
where S = F−1 T̄ = JF−1 TFT −1 is a symmetric Lagrangian tensor known

as the second Piola–Kirchhoff stress or the Kirchhoff–Trefftz stress, with
−1 −1 −1
components SIJ = FIk T̄kJ = JFIk FJl Tkl .
Virtual Work
The principle of virtual work introduced in 1.3.5 can readily be extended
to finite deformations (there is, however, no large-deformation counterpart to
the principle of virtual stresses). Consider two possible configurations χ and
χ + δu, where δu is a virtual displacement field as before, and suppose that
the acceleration field vanishes identically. Multiplying the left-hand side of
Equation (8.1.5) by δui , with summation over i implied as usual, integrating
over the reference volume R0 , and applying the divergence theorem leads to
Z Z Z
T̄iJ δui ,J dV0 = ρ0 bi δui dV0 + t̄ai δui dS0 , (8.1.6)
R0 R0 ∂R0t
where t̄a is the prescribed surface traction per unit reference area, with
t̄i = n0J T̄iJ , n0 being the unit normal vector in the reference configuration.
The left-hand side of the above equation can be cast in another form by
introducing the virtual strain field δE, which, from the definition of E, can
easily be shown to be given by
1
δEIJ = (FkI δuk ,J +FkJ δuk ,I ).
2
The principle of virtual work can now be written as
Z Z Z
S : δE dV0 = ρ0 b · δu dV0 + t̄a · δu dS0 ,
R0 R0 ∂R0t
A rate form of the principle of virtual work can be obtained by differen-

tiating the left-hand side of (8.1.5) with respect to time (the right-hand side
is again assumed to be zero) and then multiplying it by δvi , where δv is a

virtual velocity field. The result is
Z Z Z
a
T̄˙ iJ δvi ,J dV0 = ρ0 ḃi δvi dV0 + t̄˙ i δvi dS0 , (8.1.7)
R0 R0 ∂R0t
Energy Balance
Like the local equations of motion, the local energy-balance equation
(1.4.1) is exact as an Eulerian equation if the term σij ε̇ij is replaced by
T : D. It can easily be recast into the Lagrangian form
ρ0 u̇ = S : Ė + ρ0 r − Divh̄, (8.1.8)
where Div denotes the divergence operator with respect to Lagrangian coor-
dinates, and h̄ = JF−1 h is the Lagrangian heat-flux vector. The term S : Ė
may, of course, be replaced by T̄ : Ḟ.
Objective Rates
Consider a possible motion χ(X, t) of a body, and another possible mo-
tion χ∗ (X, t) of the same body that differs from the first motion by a super-
posed rigid-body motion. The relation between the two motions must have
the form
χ∗ (X, t) = Q(t)χ(X, t) + c(t),
where Q(t) is a proper orthogonal tensor and c(t) a vector, both continuous
functions of time that are also, for simplicity, assumed to be continuously
differentiable. It is clear that according to a basic principle of relativity,
the relation between the two motions is equivalent to that between the de-
scriptions of one and the same motion as seen by two observers who are
moving relative to each other, with Q(t) describing the relative rotation of
their respective Eulerian bases. If g is an Eulerian vector generated by the
motion χ, then it is called objective if its counterpart g∗ , generated by χ∗ , is
related to g by g∗ = Qg. Similarly, if G is an Eulerian second-rank tensor
generated by χ, then it is objective if G∗ is related to it by G∗ = QGQT .
Since F∗ = QF, it follows that dx = F dX is an objective vector and B is
an objective tensor. The area element dA is also an objective vector (since
dV = dxT dA). The objectivity of the surface tractions t, and therefore of
the Cauchy stress T, is argued on physical grounds, in that the tractions
represent contact forces within the body. The assertion that if a motion
χ generates a stress T then the transformed motion χ∗ generates the stress
T∗ = QTQT is known as the principle of objectivity or principle of
material frame indifference.1
1
The principle is usually stated as an axiom. For a critique, see Woods [1981].
On the other hand, vectors and tensors derived by means of material

time differentiation are not, in general, objective. Thus, if g∗ = Qg, then
(D/Dt)g∗ = Q(Dg/Dt + Ωg), where Ω = QT Q̇. It can be readily shown
that Ω is an antisymmetric tensor: since QT Q = I, it follows that
d
(QT Q) = Q̇T Q + QT Q̇ = ΩT + Ω = 0.
dt
It can also be shown that the vector ω derived from Ω by ωj = 12 eijk Ωik is
the angular velocity of the rotation described by Q.
Similarly, if G∗ = QGQT , then
D ∗ D

G =Q G + ΩG − GΩ QT .
Dt Dt
Since Ḟ∗ = Q(Ḟ + ΩF), the velocity gradient L = ḞF−1 transforms as
L∗ = Q(L + Ω)QT = Q(D + W + Ω)QT .
Decomposing L∗ into its symmetric and antisymmetric parts, we find that
D∗ = QDQT , W∗ = Q(W + Ω)QT .
Thus the Eulerian deformation-rate tensor D is an objective tensor, a result

that could also have been found from Equation (8.1.4), since the Lagrangian
tensor E, and hence also Ė, is invariant under the transformation.
Note that for an objective tensor G, combinations such as
D D D
G − WG + GW, G − LG − GLT , G + LT G + GL
Dt Dt Dt
are objective tensors known as objective rates of G; they are called, respec-
tively, the Jaumann, Oldroyd , and Cotter–Rivlin rates. Note further that
they differ from one another by terms that are bilinear in D and G and
therefore themselves objective. Clearly, an infinity of objective rates can
be constructed. Adding the term (tr D)G to the Oldroyd rate yields the
Truesdell rate, related to the material time derivative of the second Piola–
Kirchhoff stress:
D

−1
Ṡ = JF T + (tr D)T − LT − TLT FT −1 ,
Dt
and the quantity in brackets is just the Truesdell rate of the Cauchy stress.
∗
Let G be a symmetric tensor, and let its Jaumann rate be denoted G.
An important property of this rate, not shared by the other rates given
above, is the following:
D ∗
(G : G) = 2G : G, (8.1.9)
Dt
since
G : (GW) − G : (WG) = tr (GGW − GWG) = 0.
Another important property is that if G is a deviatoric tensor (i.e., if
∗
tr G = 0), then G is also deviatoric. Both properties, however, are shared
by objective rates having the Jaumann form but with W replaced by some
other antisymmetric tensor representing a spin.

1. Show that there exists a symmetric positive definite tensor V (the
leftstretch tensor ) such that F = VR, where R is the same as in the
right polar decomposition of F, and B = V2 .
2. Show that E = FT Ea F, where E is the Green–Saint-Venant strain and
Ea is the Almansi strain.
3. A state of deformation known as simple shear occurs when F is given
by the component matrix (with respect to coincident Eulerian and
Lagrangian bases)  
1 γ 0
 0 1 0 .
 
0 0 1
Find C, B, R, the principal stretches, and the eigenvectors of C and
B.
4. With respect to the finite-strain compatibility condition (8.1.1), con-
sult one of the references given in the text and show that the vanishing
of the Riemann–Christoffel tensor is equivalent to Equation (8.1.1).
5. Find the rate of change (D/Dt) dA of an infinitesimal Eulerian area
element dA.
6. Show that the invariants of C are the same as those of B, and then
show that if ψ is a function of these invariants, then F(∂ψ/∂C)FT =
B(∂ψ/∂B). (Hint: Show that FCk FT = Bk+1 .)
7. Derive the Lagrangian equations of motion (8.1.5) by at least two meth-
ods.
8. Show that the Jaumann, Oldroyd, Cotter–Rivlin, and Truesdell rates
of an objective Eulerian second-rank tensor field G are objective.
9. Show that if Ḡ is a Lagrangian tensor field and G is an Eulerian tensor
field related to it by G = FT −1 ḠF−1 , then the Cotter–Rivlin rate of G
˙ while if G = (1/J)FḠFT , then the Truesdell
is linearly related to Ḡ,
˙
rate is linearly related to Ḡ.
10. Show that (D/Dt)f + f ∇ · v is an objective rate if f is an objective

scalar field.
Section 8.2 Large-Deformation Constitutive The-

ory
8.2.1 Thermoelasticity
A thermoelastic body was defined in 1.4.1, under small deformations, as
one in which the strain is determined by the stress and the internal-energy
density, and in the absence of internal constraints, “strain” and “stress” in
the definition may be interchanged. For large deformations, and assuming
no internal constraints, we may define a thermoelastic body as one in which
the Cauchy stress T depends only on the deformation gradient F and the
internal-energy density u. The second law of thermodynamics assures the
existence of an entropy density η = η̄(u, F) and of an absolute temperature
T such that T −1 = ∂ η̄/∂u, as in Section 1.4.
The “isotropy of space” mentioned previously requires, however, that
the scalar η be invariant under a rotation of the Eulerian basis. As we
know, under such a rotation (by an orthogonal matrix Q) F becomes QF;
its polar decomposition factors R and U become, respectively, QR and U.
Consequently, η (and hence T ) must be independent of R and can depend
on F only through U or, equivalently, through C or E. The reduced local
form of the second law becomes
(S + T ρ0 ∂ η̄/∂E) : Ė = 0,
from which follows, in the absence of internal constraints (i.e., with the
several components of Ė independent),
S = −T ρ0 ∂ η̄/∂E.
With the absolute temperature used as an independent variable, the reduced
local form of the second law can be expressed in terms of the Helmholtz free-
energy density ψ(T, E) in the form
(S − ρ0 ∂ψ/∂E) : Ė = 0, (8.2.1)
and yields the Lagrangian finite-deformation analogue of Equation (1.4.3),
namely
S = ρ0 ∂ψ/∂E. (8.2.2)
From this we can in turn obtain the relation for the Cauchy stress:
T = ρF(∂ψ/∂E)FT . (8.2.3)
Section 8.2 / Large-Deformation Constitutive Theory 479
Incompressible Elastic Continuum

If the continuum is subject to an internal constraint given by G : Ė = 0,
which may or may not be derivable from a holonomic constraint of the form
h(T, E) = 0, then Equation (8.2.1) may be changed to
(S − ρ0 ∂ψ/∂E + pG) : Ė = 0,
where p is an undetermined Lagrangian multiplier. Equation (8.2.2) is ac-

cordingly replaced by
S = −pG + ρ0 ∂ψ/∂E.
In particular, the incompressibility constraint J = 1 is equivalent to
J˙ = (∂J/∂E) : Ė = C−1 : Ė = 0.
The stress-deformation relations now become
S = −pC−1 + ρ∂ψ/∂E,
T = −pI + ρF(∂ψ/∂E)FT ,
so that p is an indeterminate pressure (note that ρ = ρ0 in this case).
The constitutive relations take a special form when the continuum is
isotropic in the reference configuration, that is, when ψ(T, E) is invariant
under a rotation of the Lagrangian basis. In such a case ψ can depend on E
only through a set of invariants, for example, the principal invariants of C,
defined in the same way as for stress in Section 1.3, namely I1 = tr C, I2 =
1
2
(C : C−I12 ), I3 = det C. In the incompressible case the last of these is unity,
and therefore I1 and I2 are the only independent deformation variables;
moreover, it can be shown that in this case, I2 = −tr C−1 . Consequently,
∂ψ ∂ψ −2
∂ψ/∂E = 2∂ψ/∂C = 2 I+2 C .
∂I1 ∂I2
Since FIFT = FFT = B and FC−2 FT = B−1 , we obtain

∂ψ ∂ψ −1

T = −pI + 2ρ B+ B .
∂I1 ∂I2
Finally, noting that the invariants of C are the same as those of B, the last
relation may be written more concisely as
T = −pI + 2ρ(∂ψ/∂B)B.
Compressible Elastic Continuum

For the compressible continuum, rather than dealing directly with the
tensors E or C, it is more convenient to separate volume deformation from
distortion. Let us recall that the infinitesimal strain tensor ε can be decom-
posed additively as ε = εv I+e, where εv = 13 tr ε is the volume strain and e is
the strain deviator (see, for example, Equation (1.4.16) for the strain-energy
function of an isotropic, linearly elastic continuum). The corresponding de-
composition for finite deformation is not√ additive but multiplicative, since
volume deformation is defined by J = det C. In particular, let us define
1
F̄ = J −1/3 F, B̄ = J −2/3 B, C̄ = J −2/3 C, Ē = (C̄ − I);
2
note that det C̄ = 1. If we now set the Helmholtz free energy as ψ =
ψ(T, J, Ē), then we may write
∂ψ

S = ρ0 JC−1 + J −2/3 Dev (∂ψ/∂ Ē) ,
∂v
∂ψ

T=ρ J I + dev (F̄∂ψ/∂ ĒF̄T ) ;
∂J
here dev is the ordinary deviator operator, that is, dev H = H− 13 tr H, while
Dev is a “Lagrangian” deviator operator defined by Dev H = H − 13 (H :
C)C−1 . Note that for the Cauchy stress, the deviator is determined by
the dependence on the distortion while the pressure is determined by the
dependence on the volume deformation. If the free energy depends only on
J then the continuum is a fluid.
Finally, for the isotropic continuum, with ψ = ψ(T, J, B̄), the stress is
given by
∂ψ

T=ρ J I + 2 dev (B̄∂ψ/∂ B̄) ,
∂J
and if the invariants I¯1 and I¯2 are introduced, by
∂ψ ∂ψ ∂ψ

T=ρ J I + 2 dev ¯ B + ¯ B−1 .
∂v ∂ I1 ∂ I2
8.2.2 Inelasticity: Kinematics
Multiplicative Decomposition
In Section 1.5 we introduced the additive decomposition of the infini-
tesimal strain tensor into elastic and inelastic (later renamed plastic) parts,
which is basic to virtually all the subsequently developed theory. For strains
that are too large to be treated as infinitesimal, we found in 2.1.2 that the
decomposition still works, at least for longitudinal strain, when this is taken
as the logarithmic strain. In order to formulate a theory of plasticity for
large deformation, we have to establish the appropriate decomposition for
arbitrary states of deformation.
An additive decomposition of the logarithmic strain is equivalent to a

multiplicative decomposition of the stretch λ, that is, λ = λe λp . Let us con-
sider, first, problems of three-dimensional deformation in which the principal
directions of strain are known and remain constant and in which, in addi-
tion, the principal directions of elastic and plastic strain may be assumed
to coincide. In such problems we may, with no loss of generality, let R = I,
where R is the rotation tensor defined in 8.1.1, and apply the multiplicative
decomposition to each of the principal stretches. The result can be written
as U = Ue Up = Up Ue or as V = Ve Vp = Vp Ve ; here U and V are re-
spectively the right and left stretch tensors defined in 8.1.1 and Exercise 1
of Section 8.1, and in the class of problems under discussion they coincide.
These formulas, of course, cannot be valid in the general case: if Ue
and Up are somehow defined but are not coaxial, then Ue Up 6= Up Ue .
Furthermore, neither product is in general symmetric and therefore cannot
equal U. The same reasoning applies to V. What form, then, should the
multiplicative decomposition take in general?
An answer to this question may be based on a theory first explicitly
formulated by Kröner [1960], and further developed by Lee and Liu [1967],
Fox [1968], Lee [1969], Mandel [1972], and others on the basis of consider-
ations of the behavior of crystals. The reasoning behind it goes like this:
if an infinitesimal neighborhood (x, x + dx) (by which we mean the points
contained between xi and xi + dxi , i = 1, 2, 3) in a plastically deformed
crystal could be cut out and instantaneously relieved of all stresses, it would
be mapped into (x̂, x̂ + dx̂), the transformation being composed of a rigid-
body displacement and a purely elastic deformation (“unloading”).1 The
position x̂ is arbitrary, but we may assume a linear relation between dx and
dx̂, that is, dx̂ = F−1 −1
e dx. Here Fe (or Fe ) is not to be interpreted as
a deformation gradient, since the transformation is not a displacement of
the whole body; that is, it is not assumed that there exists a differentiable
function x̂(x) such that F−1 e = ∇x̂. Nevertheless, since Fe is nonsingular,
it may be subjected to the polar decompositions Fe = Re Ue = Ve Re . The
tensors Ue and Ve represent the elastic deformation that is removed along
with the removal of the stresses. We may accordingly define the elastic right
Cauchy–Green tensor Ce = U2e = FTe Fe and the elastic Green–Saint-Venant
strain Ee = 12 (Ce − I).
Now consider a stress-free reference configuration of the crystal, such that
def −1
dx = F dX; then dx̂ = F−1 e F dX. We define Fp = Fe F, and accordingly
write
F = Fe Fp . (8.2.4)
1
As pointed out by Mandel [1972], the elastic unloading operation has to be regarded
as a fictitious one in a material with a strong Bauschinger effect, in which real unloading
produces reverse plastic deformation,
The right Cauchy–Green tensor C is then given by
C = FTp Ce Fp . (8.2.5)
The transformations involved in the decomposition are shown schemati-

cally in Figure 8.2.1; the unloaded element (x̂, x̂ + dx̂) is usually said to be
in an “intermediate configuration.”
- Intermediate “configuration”
Fp
dx̂
3

dX

F−1

e
6
X Fe
?

r h
h h
jF dx
Reference configuration

x

rhh h
Current configuration
Figure 8.2.1. Multiplicative decomposition of deformation.
If we define Je = det Fe and Jp = det Fp , then
J = Je Jp
and
Je = ρp /ρ, Jp = ρ0 /ρp ,
where ρp is the density in the intermediate configuration. In metals, plastic

deformation is in general virtually isochoric, so that Jp may be taken as
equal to unity.
The Invariance Question

The process that moves the body from the reference configuration to the
current one is responsible, in general, for both elastic and plastic deforma-
tions. By hypothesis, the elastic deformation is removed in the fictitious local
unloading process producing the intermediate configuration. Consequently,
the intermediate configuration may be said to be only plastically deformed,
with the effects of the internal processes leading to plastic deformation em-
bodied in Fp . However, Fp represents more than just plastic deformation: if
the polar decomposition Fp = Rp Up is performed, then the deformation is
represented by Up , or equivalently by Cp = U2p = FTp Fp or Ep = 21 (Cp − I),
while Rp represents plastic rotation.
Since the intermediate configuration is a fictitious one, its orientation is
in principle arbitrary. Consequently it may be argued that if the current
state is described by tensors based in the intermediate configuration, then
the description should be invariant under an arbitrary rotation of this con-
figuration, dx̂∗ = Q dx̂. If this is so, then Fp can influence the state only
through Up (or, equivalently, Cp or Ep ). Note that Ee is not invariant under
such a rotation (the proof is left to an exercise), and therefore, under the
present hypothesis, it cannot serve as a Lagrangian measure of elastic strain.
The description of the state by the total strain E, a plastic strain Ep (not
necessarily defined by means of the multiplicative decomposition), and pos-
sibly other (scalar) internal variables, in addition to the temperature, has its
origin in the work of Green and Naghdi [1965, 1971], and has more recently
been discussed by Casey and Naghdi [1980, 1983], Simo and Ortiz [1985],
Casey [1985], and Rubin [1986].
On the other hand, it has been argued by many other workers1 that it
is precisely the tensor Ee which, along with the hardening variables and the
temperature, determines the current thermomechanical state. The argument
may be based on the physics of crystal plasticity, with Ee taken as repre-
senting the distortion of the lattice, while the hardening variables represent
dislocation densities and such. A consequence of this argument is that Fp
cannot be a state variable if Ee is one (if E or C is used as a state variable,
then Fp enters through the definition of Ee ). The physical aspect of the ar-
gument is as follows: a perfectly plastic crystal specimen that is loaded until
it yields, plastically deformed, and finally unloaded is mechanically undis-
tinguishable from its original state — in other words, plastic deformation
alone (in the absence of hardening) does not produce a change of state.
1
For example Lee [1969], Mandel [1972], Kratochvı́l [1973], Dashner [1986], Stickforth
[1986].
If Ee is an essential Lagrangian state variable, then the orientation of the

intermediate configuration cannot be arbitrary (Mandel [1972]), but must be
related to a “director frame” attached to the lattice. An intermediate con-
figuration that rotates along with a director frame is called isoclinic (Mandel
[1973]). If Fp is defined with respect to an isoclinic intermediate configura-
tion, then it is a Lagrangian tensor in the sense of being invariant under a
rotation of the Eulerian frame.
The two arguments can be reconciled under the following circumstances:
1. If the dependence of the thermomechanical state functions on Ee
is isotropic, then it is equivalent to an isotropic dependence on Be , since
Be = Re Ce RTe . But it is also equivalent to an isotropic dependence on
UC−1 p U, since
Be = FF−1p Fp
T −1 T
F = RUC−1 p UR ,
T
and this is a special case of a dependence on C and Cp , and therefore on

(E, Ep ).
2. If it is assumed that Rp = I (Haupt [1985]), then Ce = U−1 −1
p CUp ,
another special case.
Plastic Deformation Rates
The velocity gradient L is given by
L = Le + Fe Lp F−1
e ,
where
Le = Ḟe F−1 −1
e , Lp = Ḟp Fp .
Consequently,
Ċ = FTp [Ċe + 2(Ce Lp )S ]Fp , (8.2.6)
with Ċe = 2FTe De Fe , De = LSe . We see that Ċ depends on Le only through
its symmetric part, and on Lp through the symmetric part of Ce Lp . The
Eulerian deformation-rate tensor D is
D = De + (Fe Lp F−1 S
e ) .
The tensor Lp is referred to by Mandel [1972] as the “plastic distortion

rate” and is related to the dislocation motion as follows: if the ith slip system
consists of dislocations moving in the glide plane whose unit normal is ni ,
in the direction of the unit vector mi , and producing the shear rate γ̇i , then
X
Lp = γ̇i mi ⊗ ni , (8.2.7)
i
the summation being over all slip systems.

As regards a “plastic deformation rate” Dp , however, there is no un-
equivocal definition; among the ones that have been proposed are D − De =
(Fe Lp F−1 S S S
e ) , Lp , and (Ce Lp ) . The first of these tensors is Eulerian. The
other two are defined in the intermediate configuration, and are therefore
Lagrangian tensors if this configuration is isoclinic.
8.2.3 Inelasticity: Thermomechanics
Internal Variables
Let α denote the internal-variable vector describing dislocation densities,
point-defect densities and other structural properties. In accordance with
the argument of Mandel and others, the local thermomechanical state is then
determined by (Ee , α, T ), while in the hypothesis of Green and Naghdi it is
determined by (E, Ep , α, T ). Mathematically, both postulates are special
cases of of (E, Fp , α, T ), or of (E, ξ, T ) if we define ξ = (Fp , α) as the
apparent internal-variable vector. Furthermore, if equations for the shear
rates γ̇i are given, then Equation (8.2.7) constitutes a rate equation for
Fp . The description of the state by (E, ξ, T ) is the basic one used in the
remainder of this section, with the special cases invoked as appropriate.
Stress
The deformation power per unit mass is, as a result of Equation (8.2.6),
1 1
S : Ė = (Se : Ėe + Σ : Lp ), (8.2.8)
ρ0 ρ0
where
Se = Fp SFTp , Σ = Ce Se = JFTe TFTe −1
are stress tensors defined in the intermediate configuration. Both Se and Σ
differ from the Kirchhoff stress (weighted Cauchy stress) tensor P = JT by
quantities of order |Fe − I|. Furthermore, Se is a symmetric tensor that is
conjugate to the elastic strain tensor Ee , and Jp−1 Se may be regarded as the
second Piola–Kirchhoff stress with respect to the intermediate configuration.
The tensor Σ, which was introduced by Mandel [1972], is not in general
symmetric.
Free-Energy Density
In accordance with the two arguments cited above, the free-energy den-
sity is given either by the form favored by Mandel and others, namely,
ψ = ψ̄(Ee , α, T ), (8.2.9)
or by the Green–Naghdi form,1
ψ = ψ̃(E, Ep , α, T ), (8.2.10)
1
In the work of Green and Naghdi and their followers, α is usually taken to consist of
a single scalar variable κ.
both forms being special cases of ψ(E, Fp , α, T ) and therefore of ψ(E, ξ, T ).

If Equation (8.2.2) is assumed to be valid for a rate-independent plastic
material, then, with the free-energy density given by (8.2.9), it can be shown
to be equivalent to
Se = ρ0 ∂ ψ̄/∂Ee ; (8.2.11)
the proof is left to an exercise. By analogy with the uncoupled form (1.5.4)
for the free-energy density under infinitesimal deformation, the function ψ̄
in Equation (8.2.9) is often restricted even further to be given by
ψ̄(Ee , α, T ) = ψe (Ee , T ) − T ηp (α) + up (α). (8.2.12)
Here ψe is the thermoelastic free energy, usually identified with the lattice
energy, while up and ηp are respectively the stored energy and the config-
urational entropy due to dislocations and other relevant lattice defects. It
follows that Se = ρ0 ∂ψe /∂Ee ; this means that at a given temperature, there
is a one-to-one correspondence between Se and Ee , as well as between Σ and
Ee (or Ce ) in view of the relation Σ = Ce Se . As was remarked above, Σ is
not in general a symmetric tensor; however, C−1 e Σ = Se is symmetric, as is
ΣCe = Ce Se Ce . Consequently Σ must obey a constraint, for example
(ΣCe )T = ΣCe .
With Ce a function of Σ, such a constraint is equivalent to three nonlinear

equations involving the components of Σ, limiting Σ to a six-dimensional
manifold in the nine-dimensional space of second-rank tensors. If Ce = I,
then the equations are linear, and the manifold becomes “flat” — namely,
the space of symmetric second-rank tensors. If Ee is small, then the manifold
is “slightly curved” or “almost flat.”
The Green–Naghdi form (8.2.10) for the free-energy density is often
rewritten, with no loss in generality, as
ψ = ψ̂(E − Ep , Ep , κ, T ), (8.2.13)
with further specializations for “special materials” (Casey and Naghdi [1981]).
While E − Ep was identified with elastic strain in the original work of Green
and Naghdi [1965], in later work (Green and Naghdi [1971]) this identifica-
tion was dropped. Nevertheless, the decomposition
E = Ee + Ep
is not infrequently used, for example by Simo and Ortiz [1985]. Casey [1985]
justifies it as an approximation that is valid when (i) small plastic deforma-
tions are accompanied by moderate elastic strains, (ii) small elastic strains
are accompanied by moderate plastic deformations, or (iii) small strains are
accompanied by moderate rotations.
Elastic Tangent Stiffness

Let Γ = ∂S/∂E denote the usual “Lagrangian” elastic tangent stiffness
tensor. With the free-energy density given by (8.2.9), the elastic tangent
stiffness tensor in the intermediate configuration may be defined as Γe =
∂Se /∂Ee = ρ0 ∂ 2 ψ̄/∂Ee ∂Ee . The relation between Γ and Γe is
Γe : A = Fp [Γ : (FTp AFp )]FTp , (8.2.14)
where A is any symmetric second-rank tensor defined in the intermediate

configuration, and Γe : A denotes the tensor with components ΓeIJKL AKL ;
the derivation of the relation (8.2.14) is left to an exercise. With the free-
energy density given by Equation (8.2.12), we have Γe = ρ0 ∂ 2 ψe /∂Ee ∂Ee .
Plastic Dissipation
With the identification ξ = (Fp , α), the Kelvin inequality (1.5.3) be-
comes
D = −ρ0 [(∂ψ/∂Fp ) : Ḟp + (∂ψ/∂α) · α̇] ≥ 0.
With ψ given by Equation (8.2.9), this becomes
D = −ρ0 [(∂ ψ̄/∂Ce ) : (∂Ce /∂Fp ) : Ḟp + (∂ ψ̄/∂α) · α̇] ≥ 0.
But (∂Ce /∂Fp ) : Ḟp = −2(Ce Lp )S and ∂ ψ̄/∂Ce = Fp (∂ψ/∂C)FTp = (1/2ρ0 )Se .
Consequently,
D = Dp − ρ0 (∂ ψ̄/∂α) · α̇ ≥ 0,
where Dp is the plastic dissipation per unit mass, variously given by
Dp = Σ : Lp = Se : (Ce Lp )S = P : (Fe Lp F−1 S

e ) .
With ψ̄ given by Equation (8.2.12), the total dissipation per unit mass is
D = Dp + ρ0 T η̇p − ρ0 u̇p ≥ 0.
When the free-energy density is given by Equation (8.2.13), the dissipa-

tion is " #
∂ ψ̂
D = S : Ėp − ρ0 (∂ ψ̂/∂Ep ) : Ėp + κ̇ .
∂κ
8.2.4 Yield Condition and Flow Rule

In the theory of Green and Naghdi, the yield function is given as f (S, T, Ep ,
κ). The flow rule is given as a rate equation for Ep , and the development of
the theory is very much like that of infinitesimal-deformation plasticity.
Mandel [1972] assumed the yield function in stress space as f˜(Σ, α, T ).

He further established the following version of the maximum-plastic-dissipation
postulate (3.2.4):
(Σ − Σ∗ ) : Lp ≥ 0 for any Σ∗ such that f˜(Σ∗ , α, T ) ≤ 0, (8.2.15)
from which he derived the nine-dimensional associated flow rule
Lp = φ∂ f˜/∂Σ, φ ≥ 0.
This result implies that not only the plastic deformation but also the plastic
rotation Rp are determined by the same flow rule.
When it is remembered that Σ, although not in general symmetric, is
nonetheless limited to a six-dimensional manifold, then the validity of this
result may be questioned (Lubliner [1986]). Let the yield function in strain
space be defined as
f¯(E, ξ, T ) = f˜(Σ(E, ξ, T ), α, T ),
and the total specific dissipation as
D(E, ξ, ξ̇, T ) = −(∂ψ/∂ξ) · ξ̇.
With the free-energy density given by (8.2.8), it can easily be shown that
inequality (8.2.15) is equivalent to the large-deformation analogue of (3.2.12),
namely
D(E, ξ, ξ̇, T ) − D(E∗ , ξ, ξ̇, T ) ≥ 0 (8.2.16)
for any E∗ such that f¯(E∗ , ξ, T ) ≤ 0. Suppose, in particular, that the
process is isothermal, that it is elastic from (E∗ , ξ, T ) to (E, ξ, T ), and that
its continuation from (E, ξ, T ) is plastic; the time derivatives of (E, ξ, T )
at (E, ξ, T ) are (Ė, ξ̇, 0). If E∗ is close to E, then E∗ = E − hĖ + o(h) for
some small positive h. For the left-hand side of (8.2.16) we therefore have
D(E, ξ, ξ̇, T ) − D(E∗ , ξ, ξ̇, T ) = h(∂D/∂E) : Ė + o(h),
where ∂D/∂E is evaluated at (E, ξ, ξ̇, T ). Hence
(∂D/∂E) : Ė ≥ 0. (8.2.17)
Furthermore,
ρ0 ∂D/∂E = −ρ0 (∂ 2 ψ/∂E∂ξ)·ξ̇ = −(∂S/∂ξ)·ξ̇ = (∂S/∂E) : Ė−Ṡ = Γ : Ė(p) .
The last equality defines the apparent Lagrangian plastic strain rate
def
Ė(p) = Ė|Ṡ=0, Ṫ =0 = Ė − Γ−1 : Ṡ.
The rate Ė(p) , not to be confused with Green and Naghdi’s Ėp , is related to
Lp by
Ė(p) = FTp [(Ce Lp )S + 2Γe−1 : (Lp Se )S ]Fp (8.2.18)
(Lubliner [1986]). Inequality (8.2.17) may consequently be written as
Ė : Γ : Ė(p) ≥ 0, (8.2.19)
the large-deformation counterpart to (3.2.15).

From the isothermal consistency condition in strain space,
f˙¯ = (∂ f¯/∂E) : Ė + (∂ f¯/∂ξ) · ξ̇ = 0,
it follows, by analogy with the corresponding arguments in Section 3.2, that

ξ̇ 6= 0 only if (∂ f¯/∂E) : Ė > 0, and this inequality is compatible with (8.2.17)
only if
Ė(p) = φΓ−1 : (∂ f¯/∂E), φ ≥ 0. (8.2.20)
The last result may be regarded as an associated flow rule in strain space.
Let us now return to stress space. Suppose, first, that the yield surface is
given in S-space by fˆ(S, ξ, T ) = 0. With S = S(E, ξ, T ), we have
f¯(E, ξ, T ) = fˆ(S(E, ξ, T ), ξ, T ),
so that ∂f /∂E = ∂ fˆ/∂S : ∂S/∂E = Γ : ∂ fˆ/∂S, since Γ is symmetric.

Consequently, (8.2.20) is equivalent to
Ė(p) = φ∂ fˆ/∂S, φ ≥ 0. (8.2.21)
Next, suppose the yield surface to be given in Se space by f (Se , ξ, T ) =

0; then fˆ(S, ξ, T ) = f (Fp SFTp , ξ, T ), and therefore
∂ fˆ/∂S = FTp (∂f /∂Se )Fp . (8.2.22)
Combining Equations (8.2.18), (8.2.21) and (8.2.22), we obtain
(Ce Lp )S + 2Γe−1 : (Lp Se )S = φ∂f /∂Se . (8.2.23)
Finally, let Σ be the stress variable, as in Mandel’s theory, with the

previously considered yield function f˜. Since f (Se , ξ, T ) = f˜(Ce Se , ξ, T ),
it can easily be shown by means of the chain rule that
∂f /∂Se = (Ce ∂ f˜/∂Σ)S + 2Γe−1 : [(∂ f˜/∂Σ)Se ]S . (8.2.24)
Now if the left-hand side of (8.2.23) is written as ∆ : Lp , ∆ being a fourth-

rank tensor, then the right-hand side of (8.2.24) is just ∆ : (∂ f˜/∂Σ). Com-
bining the two equations we obtain ∆ : (Lp − φ∂ f˜/∂Σ) = 0, or
Lp = φ∂ f˜/∂Σ + L00p , (8.2.25)

where L00p may be any tensor satisfying ∆ : L00p = 0. Within the nine-
dimensional space of second-rank tensors, the tensors obeying this equation
form a three-dimensional subspace. Therefore, the flow rule (8.2.25) deter-
mines only the projection of Lp into the complement of this subspace, itself
a six-dimensional subspace. This last subspace is the so-called cotangent
space of the aforementioned manifold containing Σ; it is just the space that
contains all possible values of ∂φ/∂Σ for continuously differentiable scalar-
valued functions φ(Σ).
Negligible Elastic Strains
If the elastic strains are negligible next to unity, then the preceding
theory can be considerably simplified. (Ce Lp )S is approximately equal to
.
LSp , and since Ee = Γe−1 : Se , the term 2Γe−1 : (Lp Se )S in (8.2.18) may be
neglected. Consequently,
.
Ė(p) = FTp LSp Fp = Ėp ,
and the “Lagrangian” flow rule (8.2.21) may be replaced by
Ėp = φ ∂ fˆ/∂S,
equivalent to the associated flow rule of the Green–Naghdi theory.
To formulate an approximate form in the intermediate configuration, we
define the rotated Kirchhoff stress as P̃ = JRTe TRe ; then Se and Σ are both
approximately equal to P̃, and either (8.2.22) or (8.2.24) may be replaced
by
LSp = φ ∂f /∂ P̃.
An Eulerian version can be written if the dependence of f on P̃ is

isotropic, since this is then equivalent to an isotropic dependence on the
Kirchhoff stress P, and ∂f /∂ P̃ = RTe (∂f /∂P)Re . Moreover, D − De =
.
(Fe Lp F−1e )
S = Re LSp RTe , and therefore the approximate associated flow
rule reads
D − De = φ ∂f /∂P. (8.2.26)
. .
In addition, Je = 1, so that J = Jp . If, as usual, plastic volume change is
.
neglected, then J = 1, and P can be replaced by T. The resulting form of
the associated flow rule is then the same as in small-deformation theory.
A similar result is obtained if the principal axes of strain do not rotate
(i.e., if they are the same in the reference, intermediate, and current con-
.
figurations). In this case R = Re = I, so that P̃ = P = T, and the result
is valid even if the yield function does not depend isotropically on P̃. It is
on this basis that large-deformation problems have been treated in previous
chapters of this book.
While this subsection has been concerned with the possible forms taken
by the associated flow rule, analogous forms may be used for a nonassociated
flow rule if the various forms of the yield function f (f˜, f¯, fˆ) are replaced
by the corresponding forms of a plastic potential, say g. Finally, the results
may be applied to viscoplastic flow laws if φ is replaced by a function of the
state variables, g (or any of its variants) is a viscoplastic potential, and Fp
is written as Fi , Lp as Li , and so forth.

1. Show that if both sides of Equation (8.2.2) are differentiated with
respect to time, then the result is equivalent to
?
P = Λ : D,
where the left-hand side is the Jaumann rate of the Kirchhoff stress
P = JT and the Eulerian tangent elastic modulus tensor Λ is defined
by
∂2ψ
Λijkl = ρ0 FiI FjJ FkK FlL + 1 J(δik Tjl +δil Tjk +δjk Til +δjl Tik ).
∂EIJ ∂EKL 2
3. Show that if Ee is defined by means of the multiplicative decompo-

sition, then it is not invariant under a rotation of the intermediate
configuration.
5. Derive the relation (8.2.14) between Γ and Γe .
6. Show that with the given assumptions on the free energy, inequality
(8.2.16) is equivalent to (8.2.15).
7. Derive (8.2.18).
8. Show that if |Ee | 1, then |∆ : Lp − LSp | |Lp |, where ∆ : Lp is the

left-hand side of (8.2.23).
Section 8.3 Numerical Methods

As we saw in Section 4.5, numerical methods for rate-independent elastic–
plastic solids are based on the rate form of the constitutive equations, Equa-
tions (4.5.5)–(4.5.7). It was natural that the extension of such methods to
large-deformation problems, first undertaken in the late 1960s, would be
based on some analogous form. While these rate-based formulations have

been extensively used, they have been shown to possess some severe flaws,
which are discussed in 8.3.1. An alternative “hyperelastic” formulation man-
ages to avoid the difficulties posed by the rate-based formulations, though
the versions available as of this writing rely on some simplifying assumptions
on the constitutive equations. It is likely, however, that in future versions
such assumptions will become unnecessary. The “hyperelastic” formulation
is discussed in 8.3.2.
8.3.1 Rate-Based Formulations
Introduction
The earliest extensions to large deformations of the rate-based numeri-
cal methods for elastic–plastic solids were in Lagrangian form;1 among the
first was one by Hibbit, Marcal, and Rice [1970], who, on the basis of the
large-deformation principle of virtual work, Equation (8.1.6), derived an in-
cremental stiffness containing three parts in addition to the “small-strain”
stiffness, namely the initial-load, initial-stress, and initial-strain stiffnesses.
Since, however, the problems to be solved have typically been ones of
large-scale plastic flow, it was felt by many that an Eulerian form of the
equations would be preferable. Most solution schemes have followed a for-
mulation similar to that of McMeeking and Rice [1975], with the analogue
of (4.5.5), in particular, usually taken as
?
P = C : (D − Dp ), (8.3.1)
where the superposed asterisk denotes the Jaumann rate; this rate is gener-
ally preferred, in view of the properties discussed at the end of Section 8.1,
though it was pointed out by Truesdell and Noll [1965, p. 404] that “despite
claims and whole papers to the contrary, any advantage claimed for one such
rate over another is pure illusion.”
The Kirchhoff stress appears in Equation (8.3.1) because, unlike the
Cauchy stress, its use leads to a symmetric global tangent stiffness matrix
upon discretization. With isotropic elasticity assumed, as is usual, C is taken
as given by (1.4.10), with the Lamé coefficients λ and µ constant. The plastic
deformation rate Dp is almost invariably assumed to be governed by a flow
rule that is associated (in Kirchoff-stress space) with a generalized Mises
yield criterion that may incorporate kinematic hardening. Such a criterion
takes the form
1
f (P, H, κ) = (P − H)0 : (P − H)0 − [k(κ)]2 = 0, (8.3.2)
2
1
Formulations based on convected coordinates, such as that due to Needleman [1972],
are equivalent to Lagrangian ones.
where (·)0 = dev (·), H (replacing ρ in the small-deformation theory) is the

tensor-valued internal variable denoting the center of the elastic region (the
“back stress”), and κ is, as before, an internal variable representing isotropic
hardening. The associated flow rule is then
Dp = φ(P − H)0 , (8.3.3)
where φ is determined by means of the consistency condition f˙ = 0. With

appropriate rate equations for H and κ (unnecessary, of course, in the case
of perfect plasticity), φ is found as a linear form in D, and the subsequent
substitution of (8.3.3) in (8.3.1) leads to
?
P = Λ : D, (8.3.4)
where Λ is the Eulerian tangent modulus tensor analogous to Cep .

The finite-element scheme is based on the rate form of the principle of
virtual work, Equation (8.1.7), translated into Eulerian form. With the same
notation as in Section 4.5, the discretization is

bv h (x) = Ne (x)q̇e , Dh (x) = Be (x)q̇e .

Ωe Ωe
The global rate equation then takes the form (4.5.28), with the load rate Ḟ
(not to be confused with the derivative of the deformation gradient appearing
elsewhere in this chapter) defined by
"Z Z #
X
eT eT eT a
Ḟ = A N ḟ dΩ + N ṫ dΓ ,
e Ωe ∂Ωet
while the tangent stiffness Kt is given by

X Z
eT eT e
Kt = A B ΛB dΩ + Kes Ae .
e Ωe
Here Kes is the additional element stiffness (the initial-stress stiffness) due to
the various terms resulting from the conversion of the equations to Eulerian
form.
The formulation given above runs into a number of difficulties. When
the rate problem has been solved, the imposition of a load increment requires
an integration in time that maintains objectivity, necessitating incrementally
objective integration algorithms (Rubinstein and Atluri [1983], Pinsky, Ortiz,
and Pister [1983]) that may be computationally expensive. The rate-based
formulation itself, moreover, has flaws that makes its results suspect. As we
saw in Section 8.2, there is no uniquely defined plastic strain rate, and a
flow rule of the form (8.3.3) can be given only as an approximation under
restricted circumstances. Other difficulties arise from the use of Equation
(8.3.1) and from the choice of the Jaumann rate.
Hypoelasticity
In the absence of plastic deformation, Equation (8.3.1), with isotropic
elasticity assumed, becomes
?
P = λ(tr D)I + 2µD.
This equation represents a model of behavior known in the literature as

hypoelasticity (Truesdell and Noll [1965]). However, as shown by Simo and
Pister [1984], such an equation with constant coefficients1 cannot be derived
from an energy-based elastic (also called hyperelastic) stress-strain relation
such as (8.2.2) (see also Exercise 1 of Section 8.2), and the result holds
when the Jaumann rate is replaced by any other objective rate. Elasticity
without a strain energy is difficult to motivate physically, since it may entail
nonvanishing dissipation in a closed cycle of deformation.
Kinematic Hardening
Consider, next, the generalization to large deformation of the Melan–
Prager kinematic hardening model, Equation (3.3.8), using the Jaumann
rate:
Ḣ − WH + HW = cDp . (8.3.5)
Note that this equation implies tr Ḣ = 0, so that if H = 0 initially then
H = H0 , that is, H is purely deviatoric.
Let the model described by Equations (8.3.2)–(8.3.4) now be applied
to a problem in which the displacement is equivalent to simple shear. For
simplicity, incompressible rigid–plastic behavior may be assumed, so that
Dp = D, J = 1, and P = T. The deformation gradient is
 
1 γ 0
F =  0 1 0 ,
 
0 0 1
yielding
     
0 γ̇ 0 0 γ̇ 0 0 γ̇ 0
1 1
L =  0 0 0 , D =  γ̇ 0 0  , W =  −γ̇ 0 0  .
   
2 2
0 0 0 0 0 0 0 0 0
The stress-deviator and back-stress tensors can be assumed to be given,

respectively, by
   
σ τ 0 α β 0
T0 =  τ −σ 0  , H = H0 =  β −α 0  .
   
0 0 0 0 0 0
1
In the general theory of hypoelasticity the coefficients are assumed to depend on stress;
they must do so isotropically.
For compatibility with the yield criterion (8.3.2) and the flow rule (8.3.3), it
is necessary that
σ = α, τ = k + β.
The rate equation (8.3.5) now reduces to the two simultaneous equations
1
α̇ − β γ̇ = 0, β̇ + αγ̇ = cγ̇.
2
Assuming γ = γ(t) with γ̇ > 0, these equations can be solved as
1 1
α = c(1 − cos γ), β = c sin γ.
2 2
This solution, first discussed by Lehmann [1972], implies that the shear
stress is an oscillating function of the shear strain, clearly in contradiction
with the notion of hardening. Numerous proposals have been put forward
to deal with this contradiction. Some (e.g., Dafalias [1985a,b], Loret [1983])
involve a generalization of the flow rule into the nine-dimensional one en-
visaged by Mandel. Others simply suggest replacing the Jaumann rate in
Equation (8.3.5) by some other objective rate. For example, Haupt and
Tsakmakis [1986] propose the Truesdell rate, which, with the incompress-
ibility condition tr L = 0 taken into account, becomes
D −1
Ḣ − LH − HLT = F (F HFT −1 )FT .
Dt
Since
1 D −1

D=− F C FT ,
2 Dt
the rate equation can be integrated to yield
1 1
H = cF(I − C−1 )FT = c(B − I);
2 2
in other words, this proposal corresponds to the integrated form ρ = cεp of
the small-deformation kinematic hardening model. Note, however, that H
now is not purely deviatoric.
With F as given above, B is
 
1 + γ2 γ 0
B= γ 1 0 ,
 
0 0 1
and therefore
   
γ2 γ 0 γ2 γ 1
0
1  0 1 1 2
2
H = c γ 0 0 , H = c  γ −2γ 0  ,
  
2 2
0 0 0 0 0 0
that is,
1 1
α = cγ 2 , β = cγ,
4 2
reproducing straight-line hardening as for infinitesimal deformations.
Non-oscillatory behavior is also found when the rate used is of the Jau-
mann form — that is, a corotational rate — but with W replaced by ṘRT ,
where R is the rotation tensor defined in 8.1.1; see Dafalias [1983].
However, as pointed by Atluri [1984] (see also Reed and Atluri [1985]),
non-oscillatory behavior per se does not necessarily represent a solution that
agrees with experiment; for example, the just-discussed solution produces a
normal stress σ = α = 12 (τ − k)γ that is considerably larger than experimen-
tally observed in metals. Instead, Atluri suggests that, in view of the fact
that any objective rate of H differs from its Jaumann rate by terms depend-
ing on H and D, these terms be determined on the basis of experimental
data.
8.3.2 “Hyperelastic” Numerical Methods

An alternative methodology, which avoids the difficulties posed by the rate
formulation, was proposed by Simo and Ortiz [1985]. It is based on the
multiplicative decomposition of the deformation gradient, as discussed in
8.2.2, and the stress-deformation relations given by Equation (8.2.2), with
the free-energy density, however, assumed to be given in general by (8.2.10)
rather than by (8.2.9) (which is generally regarded as more consistent with
the multiplicative decomposition); the circumstances under which the two
representations coincide were discussed in 8.2.2. The formulation may, in
principle, be given equivalently with respect to the reference, current, or
intermediate configuration. What follows is a summary of the presentation
given in Simo and Hughes [1988] (see also Simo [1988a,b]).
With the stress-deformation relations in direct form rather than in rate
form, objectivity is enforced automatically when these relations are, as is
usual, invariant under a rigid-body rotation of the current configuration —
that is, when the free-energy density ψ is determined by tensors that are in-
variant under such a rotation. Plastic incompressibility may be conveniently
incorporated by assuming, as in the infinitesimal-deformation theory, that
the mean stress is determined only by the volume deformation, which is
elastic. With the help of the multiplicative decomposition of the deforma-
tion gradient as in 8.2.1, namely, F = J 1/3 F̄ (see 8.2.1), with J = Je and
F̄ = F̄e Fp , the free-energy function may be assumed as a special (uncoupled)
case of (8.2.12), with
ψe (Ee , T ) = ψeV (Je , T ) + ψeD (Ēe , T ),
where Ēe = 12 (F̄Te F̄e − I); then 13 tr T = ρ∂ψV /∂Je .

An efficient method for solving the governing equations turns out to be

the elastic predictor—plastic corrector algorithm, similar to the one pre-
sented for infinitesimal deformation in Section 4.5. The problem is assumed
to be displacement-driven, so that F is assumed to be known as a function
of time. During the predictor phase, Fp is assumed to remain constant. If
the trial elastic state does not violate the yield condition, then it is the cor-
rect state at the next time point. Otherwise, the plastic corrector must be
activated.
In the corrector phase it is the total deformation gradient F that is
assumed to remain constant,1 and the problem is governed by the equations
of evolution, that is, the flow equation [which is assumed, in Lagrangian
form, as a rate equation for Cp (see 8.2.4 for a discussion of the limitations
on this assumption)] and the rate equations for the 986]. that it is precisely
the tensor Ee which, along with the hardening variables, if any.
An integration algorithm that automatically incorporates the equivalent
of incremental objectivity is based on performing the time discretization of
the rate equations (that is, their transformation into incremental equations
by means of an appropriate integration scheme) in their Lagrangian form.
The time-discretized equations are then transformed (“pushed forward”) into
Eulerian form for an implementation of the return-mapping algorithm, anal-
ogous to the one given in Section 4.5. As in the infinitesimal-deformation
case, the use of consistent (algorithmic) tangent moduli is essential in or-
der to achieve a quadratic rate of asymptotic convergence with Newton’s
method.
The finite-element discretization may be performed with respect to the
last computed configuration as a reference configuration — the so-called
updated Lagrangian approach — but with a mesh in which the nodes are
fixed in the initial configuration, permitting the plotting of the deformed
mesh at each computed configuration.
Several examples of the application of the method discussed here are
given by Simo [1988b]; the one shown here describes the necking of a cir-
cular bar made of a nonlinearly work-hardening material. Isoparametric
constant-volume Q4 elements are used. Figures 8.3.1(a)–(c) show three dif-
ferent meshes in the initial configuration — of 50, 200, and 400 elements,
respectively — and Figures 8.3.1(d)–(f) show the corresponding deformed
meshes in the deformed configuration; it is seen that there are no significant
differences among the three. Figure 8.3.2 shows the ratio of the neck radius
to the initial radius plotted against the relative elongation of the bar; the
centered symbols represent experimental results, the solid curve represents
results calculated by the “hyperelastic” method in 15 time steps, and the
1
Recall that in the infinitesimal-deformation case only the plastic strain εp and the
total strain ε, respectively, remain constant in the two phases, while the rotation may be
ignored.
(a) (b) (c)
(d) (e) (f)
Figure 8.3.1. Necking of a circular bar, numerical results: (a)–(c) initial

meshes; (d)–(f) deformed meshes after 14% elongation (from
Simo [1988b]).
other two curves represent results calculated by means of other numerical

methods at considerably greater computational cost: one (the dashed curve)
in 100 load steps and the other (the dotted curve) in 29,000 (!) load steps.
Moreover, the hyperelastic method is found to be fairly insensitive to both
mesh refinement and step size.
1.00u
t
•u
t
• • .
4
•
• ut .
4
t
•u
•u
t
. . 4.
0.75 t• .. 4
u
t
u ... .
...... 4.
a/a0 • .... 4.
.... 4.
....
• .... (c)
(a) This work ...
0.50 ........... (b) HEMP (b) ...
(c) NIKE2D
(a)
Experimental:
.
4 2499R
0.25
t
u 2515ST
2501R
• 2502R
0.00 0.1 0.2

∆L/L0
Figure 8.3.2. Necking of a circular bar, numerical and experimental results:

ratio of neck radius to initial bar radius against relative elongation
(from Simo [1988b]).
References
Alexander, J. M. [1961], Q. J. Appl. Math. 19, 31.

Alverson, R. C. [1956], J. Appl. Mech. 23, 411.
Atluri, S. N. [1984], Computer Methods Appl. Mech. Engrg. 43, 137.
Baba, S., and T. Kajita [1982], Int. J. Num. Methods Engrg. 18, 927.
Backhaus, G. [1968], Z. angew. math. Mech. 48, 99.
Baltov, A., and A. Sawczuk [1965], Acta Mech. 1, 81.
Bammann, D., and R. D. Krieg [1987], in Unified Constitutive Equations for
Creep and Plasticity (ed. A. K. Miller), Elsevier Applied Science, London,
p. 303.
Basinski, S. J., and Z. S. Basinski [1979], in Dislocations in Solids (ed. F. R.
N. Nabarro), Vol. 4, North-Holland, Amsterdam, p. 261.
Bataille, J., and J. Kestin [1975], J. Méc. 14, 365.
Bazant, Z. P. [1978], Int. J. Solids Struct. 14, 691.
Bell, J. F. [1968], The Physics of Large Deformation of Crystalline Solids,
Springer Tracts in Natural Philosophy, Vol. 14, Springer-Verlag, Berlin.
Bell, J. F. [1973], The Experimental Foundations of Solid Mechanics, in
Handbuch der Physik (ed. S. Flügge), Vol. VIa/1, Springer-Verlag, Berlin.
Bergander, H. [1980], Z. angew. Math. Mech. 60, 509.
Bianchi, G. [1964], in Stress Waves in Anelastic Solids(IUTAM Symp., Prov-
idence, 1963, ed. H. Kolsky and W. Prager), Springer-Verlag, Berlin, p. 101.
Bishop, J. F. W. [1953], J. Mech. Phys. Solids 2, 44.
Bishop, J. F. W., and R. Hill [1951], Philos. Mag. 42, 414.
Bishop, J. F. W., R. Hill, and N. F. Mott [1945], Proc. Phys. Soc. 57, 147.
Bland, D. R. [1956], J. Mech. Phys. Solids 4, 209.
Bleich, H. H., and I. Nelson [1966], J. Appl. Mech. 33, 149.
Bleich, H. H., and M. G. Salvadori, [1953], Proc. ASCE 79, Sep. 287.
Bleich, H. H., and M. G. Salvadori [1955], Trans. ASCE 120, 499.
Bodner, S. R. [1968], in Mechanical Behavior of Materials Under Dynamic
501
502 References
Loads (Symp. San Antonio, 1967, ed. U. S. Lindholm), Springer-Verlag,

New York, p. 176.
Bodner, S. R., and R. J. Clifton [1966], J. Appl. Mech. 34, 91.
Bodner, S. R., and Y. Partom [1972], J. Appl. Mech. 39, 751.
Bodner, S. R., and Y. Partom [1975], J. Appl. Mech. 42, 385.
Bolton, M. [1979], A Guide to Soil Mechanics, Macmillan, London.
Boresi, A. P., and K. P. Chong [1987], Elasticity in Engineering Mechanics,
Elsevier, New York.
Borkowski, A. [1988], Analysis of Skeletal Structural Systems in the Elastic
and Elastic-Plastic Range, Elsevier, Amsterdam.
Bridgman, P. W. [1923], Proc. Am. Acad. Arts Sci. 58, 163.
Bridgman, P. W. [1952], Studies in Large Plastic Flow and Fracture, McGraw-
Hill, New York.
Budiansky, B. [1959], J. Appl. Mech. 26, 259.
Calladine, C. R. [1973], Int. J. Mech. Sci. 15, 593.
Casey, J. [1985], Int. J. Solids Struct. 21, 671.
Casey, J., and P. M. Naghdi [1980], J. Appl. Mech. 47, 672.
Casey, J., and P. M. Naghdi [1981], J. Appl. Mech. 48, 285.
Casey, J., and P. M. Naghdi [1983], Q. J. Mech. Appl. Math. 37, 231.
Ceradini, G. [1966], Meccanica 4, 77.
Cernocky, E. P., and E. Krempl [1979], Int. J. Non-Linear Mech. 14, 183.
Chaboche, J. L. [1977], Bull. Acad. Polon. Sci. Sér. Sci. Tech. 25, 33.
Chadwick, P., A. D. Cox, and H. G. Hopkins [1964], Philos. Trans. Roy.
Soc. London A256, 235.
Chakrabarty, J. [1987], Theory of Plasticity, McGraw-Hill, New York.
Chan, H. S. Y. [1969], J. Appl. Mech. 36, 73.
Chan, K. S., S. R. Bodner, K. P. Walker, and U. S. Lindholm [1984], in Proc.
2nd Symp. Nonlinear Constitutive Equations High-Temp. Appl. (Cleveland,
1984).
Chen, W. F. [1975], Limit Analysis and Soil Mechanics, Elsevier, Amster-
dam.
Clifton, R. J. [1966], in Proc. 5th U.S. Natl. Cong. Appl. Mech. (Min-
neapolis, 1966), American Society of Mechanical Engineers, New York, p.
465.
Clifton, R. J. [1968], J. Appl. Mech. 35, 782.
Clifton, R. J. [1983], J. Appl. Mech. 50, 941.
Collins, I. F. [1969], J. Mech. Phys. Solids 17, 323.
Conroy, M. F. [1952], J. Appl. Mech. 19, 465.
References 503

Conroy, M. F. [1963], J. Méc. 2, 455.
Considère, F. [1891], Cong. Int. Procédés de Constr. (Paris, 1889), Vol. 3,
p. 371.
Cotter, B. A., and P. S. Symonds [1955], Proc. ASCE 81, 21.
Cottrell, A. H., and B. A. Bilby [1949], Proc. Phys. Soc. A62, 49.
Cowper, G. R. [1960], J. Appl. Mech. 47, 496.
Cox, A. D., and L. W. Morland [1959], J. Mech. Phys. Solids 7, 229.
Craggs, J. W. [1957], J. Mech. Phys. Solids 5, 115.
Cristescu, N. [1964], in Stress Waves in Anelastic Solids (IUTAM Symp.,
Providence, 1963, ed. H. Kolsky and W. Prager), Springer-Verlag, Berlin,
p. 118.
Cristescu, N. [1967], Dynamic Plasticity, North-Holland, Amsterdam.
Cristescu, N., and M. Predeleanu [1965], in Proc. 4th Int. Cong. Rheol.
(Providence, 1963, ed. E. H. Lee), Vol. 3, Wiley, New York, p. 79.
Dafalias, Y. F. [1975], On Cyclic and Anisotropic Plasticity, Ph.D. thesis,
Department of Civil Engineering, University of California, Berkeley.
Dafalias, Y. F. [1983], J. Appl. Mech. 50, 561.
Dafalias, Y. F. [1985a], in Plasticity Today (IUTAM Symp., Udine, 1983, ed.
A. Sawczuk and G. Bianchi), Elsevier Applied Science, London, p. 135.
Dafalias, Y. F. [1985b], J. Appl. Mech. 52, 865.
Dafalias, Y. F., and E. P. Popov [1975], Acta Mech. 21, 173.
Dantzig, G. B. [1963], Linear Programming and Extensions, Princeton Uni-
versity Press, Princeton, N.J.
Dashner, P. A. [1986], J. Appl. Mech. 53, 55.
De Juhasz, K. J. [1949], J. Franklin Inst. 248, 15, 113.
Derrington, M. G., and W. Johnson [1958], Appl. Sci. Res. A7, 408.
Donnell, L. H. [1930], Trans. ASME 52, 153.
Drucker, D. C. [1950], Q. Appl. Math. 7, 411.
Drucker, D. C. [1951], in Proc. 1st U.S. Natl. Cong. Appl. Mech., American
Society of Mechanical Engineers, New York, p. 487.
Drucker, D. C. [1953], J. Mech. Phys. Solids 1, 217.
Drucker, D. C. [1954a], J. Appl. Mech. 21, 71.
Drucker, D. C. [1954b], in Proc. 1st Midwestern Conf. Solid Mech. (Urbana,
1953), p. 158.
Drucker, D. C. [1956], J. Appl. Mech. 23, 509.
Drucker, D. C. [1959], J. Appl. Mech. 26, 101.
504 References
Drucker, D. C., and W. F. Chen [1968], in Engineering Plasticity(ed. J. Hey-

man and F. A. Leckie), Cambridge University Press, Cambridge, England,
p. 119.
Drucker, D. C., H. J. Greenberg, and W. Prager [1951], J. Appl. Mech. 18,
371.
Drucker, D. C., and H. G. Hopkins [1955], in Proc. 2nd U.S. Natl. Cong.
Appl. Mech. (Ann Arbor, 1954), American Society of Mechanical Engineers,
New York, p. 517.
Drucker, D. C., and W. Prager [1952], Q. Appl. Math. 10, 157.
Drucker, D. C., W. Prager, and H. J. Greenberg [1952], Q. Appl. Math. 9,
381.
Drucker, D. C., and R. T. Shield [1957], Q. Appl. Math. 15, 269.
Drucker, D. C., and R. T. Shield [1959], J. Appl. Mech. 26, 61.
Duwez, P. E., D. S. Clark, and H. F. Bohnenblust [1950], J. Appl. Mech. 17,
27.
Eason, G. [1960a], Appl. Sci. Res. A8, 53.
Eason, G. [1960b], Q. J. Mech. Appl. Math. 12, 334.
Eason, G., and R. T. Shield [1955], J. Mech. Phys. Solids 4, 53.
Engesser, F. [1889], Z. Archit. Ing.wesen 35, 455.
English, J. M. [1954], Trans. ASCE 119, 1143.
Eringen, A. C. [1962], Nonlinear Theory of Continuous Media, McGraw-Hill,
New York.
Florence, A. L. [1966], J. Appl. Mech. 33, 256.
Foulkes, J. [1953], Q. Appl. Math. 10, 347.
Foulkes, J. [1954], Proc. Roy. Soc. London A223, 482.
Fox, N. [1968], Q. J. Mech. Appl. Math. 21, 67.
Frank, F. C. [1951], Philos. Mag. 42, 809.
Freiberger, W. [1949], Austral. J. Sci. Res. A2, 354.
Freiberger, W. [1956a], Q. Appl. Math. 14, 259.
Freiberger, W. [1956b], J. Appl. Mech. 23, 576.
Freiberger, W., and B. Tekinalp [1956], J. Mech. Phys. Solids 4, 294.
Frémond, M., and A. Friaâ [1982], J. Méc. Théor. Appl. 1, 881.
Galin, L. A. [1949], Prikl. Mat. Mekh. 13, 285.
Gaydon, F. A., and H. Nuttall [1957], J. Mech. Phys. Solids 6, 17.
Geiringer, H. [1931], in Proc. 3rd Int. Cong. Appl. Mech. (Stockholm,
1930), Vol. 2, p. 185.
Geiringer, H. [1951], Proc. Natl. Acad. Sci. U.S.A. 37, 214.
Gerard, G., and H. Becker [1957], Handbook of Structural Stability, Part I,
References 505
NACA Tech. Note 3781.

Germain, P., Q. S. Nguyen, and P. Suquet [1983], J. Appl. Mech. 50, 1010.
Gilman, J. J. [1969], Micromechanics of Flow in Solids, McGraw-Hill, New
York.
Green, A. E., and P. M. Naghdi [1965], Arch. Ration. Mech. Anal. 18, 251.
Green, A. E., and P. M. Naghdi [1971], Int. J. Engrg. Sci. 9, 1219.
Green, A. E., and W. Zerna [1968], Theoretical Elasticity, Oxford University
Press, London.
Green, A. P. [1951], Philos. Mag. 42, 900.
Green, A. P. [1953], Q. J. Mech. Appl. Math. 6, 223.
Green, A. P. [1954a], J. Mech. Phys. Solids 3, 1.
Green, A. P. [1954b], J. Mech. Phys. Solids 3, 143.
Greenberg, H. J. [1949], Q. Appl. Math. 7, 85.
Greenberg, H. J., and W. Prager [1951], Proc. ASCE 77, Sep. 59.
Griggs, D. T. [1936], J. Geol. 44, 541.
Grigorian, S. S. [1960], Prikl. Mat. Mekh. 24, 1057.
Gvozdev, A. A. [1938], in Proc. Conf. Plast. Def.(in Russian), Akademiia
Nauk SSSR, Moscow, p. 19.
Hadamard, J. [1903], Leçons sur la propagation des ondes et les équations de
l’hydrodynamique, Hermann, Paris.
Halphen, B., and Q. S. Nguyen [1975], J. Méc. 14, 39.
Handelman, G. H. [1944], Q. Appl. Math. 1, 351.
Hart, E. W. [1976], J. Engrg. Matls. Technol. 98, 193.
Haupt, P. [1985], Int. J. Plast. 1, 303.
Haupt, P., and C. Tsakmakis [1986], Int. J. Plast. 2, 279.
Haythornthwaite, R. M., and R. T. Shield [1958], J. Mech. Phys. Solids 6,
127.
Helling, D. E., and A. K. Miller [1987], Acta Mech. 69, 9.
Hencky, H. [1923], Z. angew. Math. Mech. 3, 241.
Hencky, H. [1924], Z. angew. Math. Phys. 4, 323. 353.
Heyman, J. [1951], Q. Appl. Math. 8, 373.
Heyman, J. [1953], Struct. Engr. 31, 125.
Heyman, J. [1959], Proc. Instn. Civ. Engrs. 12, 39; Q. J. Mech. Appl.
Math. 12, 314.
Heyman, J. [1960], Int. J. Mech. Sci. 1, 121.
Heyman, J. [1971], Plastic Design of Frames, Cambridge University Press,
Cambridge, England.
Heyman, J. [1973], Int. J. Mech. Sci. 15, 845.
506 References
Heyman, J., and W. Prager [1958], J. Franklin Inst. 266, 339.

Hibbit, H. D., P. V. Marcal, and J. R. Rice [1970], Int. J. Solids Struct. 6,
1069.
Hill, R. [1948a], Q. J. Mech. Appl. Math. 1, 18.
Hill, R. [1948b], Proc. Roy. Soc. London A193, 281.
Hill, R. [1948c], J. Iron Steel Inst. 158, 177.
Hill, R. [1950], The Mathematical Theory of Plasticity, Oxford University
Press, London.
Hill, R. [1951], Philos. Mag. 42, 868.
Hill, R. [1952], Philos. Mag. 43, 353.
Hill, R. [1953], J. Mech. Phys. Solids 1, 265.
Hill, R., E. H. Lee, and S. J. Tupper [1947], Proc. Roy. Soc. LondonA, 188,
273.
Hill, R., and M. P. L. Siebel [1951], Philos. Mag. 42, 721.
Hill, R., and M. P. L. Siebel [1953], J. Mech. Phys. Solids 1, 207.
Hill, R., and S. J. Tupper [1948], J. Iron Steel Inst. 159, 353.
Hirsch, P. B. [1975], The Physics of Metals, Vol. 2: Defects(ed. P. B. Hirsch),
Cambridge University Press, Cambridge, England, p. 189.
Hodge, P. G., Jr. [1953], Div. Appl. Math. Report B11-22, Brown University,
Providence, R.I.
Hodge, P. G., Jr. [1954], J. Appl. Mech. 21, 336.
Hodge, P. G., Jr. [1955], J. Mech. Phys. Solids 3, 176.
Hodge, P. G., Jr. [1956], in Proc. 2nd Midwestern Conf. Solid Mech.
(Lafayette, 1955), p. 150.
Hodge, P. G., Jr. [1957a], in Proc. 9th Int. Cong. Appl. Mech. (Brussels,
1956), Vol. 8, p. 65.
Hodge, P. G., Jr. [1957b], J. Appl. Mech. 24, 453.
Hodge, P. G., Jr. [1959], Plastic Analysis of Structures, McGraw-Hill, New
York.
Hodge, P. G., Jr. [1963], Limit Analysis of Rotationally Symmetric Plates
and Shells, Prentice-Hall, Englewood Cliffs, N.J.
Hodge, P. G., Jr., and B. Paul [1957], in Proc. 3rd Midwestern Conf. Solid
Mech. (Ann Arbor, 1957), p. 29.
Hodge, P. G. Jr., and G. N. White, Jr. [1950], J. Appl. Mech. 17, 180.
Hohenemser, K., and W. Prager [1932], Z. angew. Math. Mech. 12, 1.
References 507
Hopkins, H. G. [1957], Proc. Roy. Soc. London A241, 153.

Hopkins, H. G., and W. Prager [1953], J. Mech. Phys. Solids 2, 1.
Hopkins, H. G., and W. Prager [1954], Z. angew. Math. Phys. 5, 317.
Hopkins, H. G., and W. Prager [1955], J. Appl. Mech. 22, 372.
Hopkins, H. G., and A. J. Wang [1954], J. Mech. Phys. Solids 3, 117.
Horne, M. R. [1950], J. Instn. Civ. Engrs. 34, 174.
Horne, M. R. [1954], Proc. Instn. Civ. Engrs. 3, 51.
Horne, M. R. [1979], Plastic Theory of Structures, 2nd ed., Pergamon Press,
Oxford.
Hudson, J. A., E. T. Brown, and C. Fairhurst [1971], in Proc. 13th Symp.
Rock Mech. (Urbana, 1970), p. 180.
Hull, D., and D. J. Bacon [1984], Introduction to Dislocations, 3rd ed., Perg-
amon Press, Oxford.
Hutchinson, J. W. [1972], J. Appl. Mech. 39, 155.
Hutchinson, J. W. [1974], in Advances in Applied Mechanics(ed. C.-S. Yih),
Vol. 14, Academic Press, New York, p. 67.
Il’iushin A. A., [1954], Prikl. Mat. Mekh. 18, 641.
Il’iushin A. A., [1961], Prikl. Mat. Mekh. 25, 503.
Imegwu, E. O. [1960], J. Mech. Phys. Solids 8, 141.
Ishlinskii, I. U. [1954], Ukr. Mat. Zh. 6, 314.
Iwan, W. D. [1967], J. Appl. Mech. 34, 612.
Johansen, K. W. [1932], Proc. Int. Assn. Bridge and Struc. Engrg. 1, 277.
Johnson, D. [1988], Int. J. Solids Struct. 24, 321.
Johnson, W., and P. B. Mellor [1973], Engineering Plasticity, Van Nostrand
Reinhold, London.
Johnson, W., R. Sowerby, and R. Venter [1982], Plane Strain Slip-Line The-
ory and Bibilography, Pergamon Press, Oxford.
Jones, N. [1975], Shock Vib. Digest 8, 89.
Josselin de Jong, G. [1965], in Proc. IUTAM Symp. on Rheology and Soil
Mechanics (Grenoble, 1964), p. 69.
Josselin de Jong, G. [1974], in Proc. Symp. on the Role of Plasticity in Soil
Mechanics (Cambridge, England, 1973), p. 12.
Kachanov, L. M. [1954], Prikl. Mat. Mekh. 19,
Kachanov, L. M. [1971], Foundations of Theory of Plasticity, North-Holland,
Amsterdam.
Kadashevich, Yu. I., and V. V. Novozhilov [1958], Prikl. Mat. Mekh. 22,
104.
Kaliski, S., and E. Wlodarczyk [1967], Arch. Mech. Stos. 19, 433.
508 References
Kármán, T. v. [1910], VDI (Ver. deut. Ing.) Forschungsh. 81.

Kármán, T. v. [1911], Z. Ver. deut. Ing. 55, 1749.
Kármán, T. v., and P. Duwez [1950], J. Appl. Phys. 21, 987.
Kazinczy, G. [1914], Betonszemle 2, 68.
Kestin, J., and Rice, J. R. [1970], in A Critical Review of Thermodynamics
(ed. E. B. Stuart, B. Gal-Or, and A. J. Brainard), Mono Book Corp.,
Baltimore, p. 275.
Kliushnikov, V. D. [1959], Prikl. Mat. Mekh. 23, 405.
Kobayashi, S., S.-I. Oh, and T. Altan [1989], Metal Forming and the Finite-
Element Method, Oxford University Press, London.
Kochendörfer, A. [1938], Z. Phys. 108, 244.
Kocks, U. F. [1987], in Unified Constitutive Equations for Creep and Plas-
ticity (ed. A. K. Miller), Elsevier Applied Science, London, p. 1.
Koiter, W. T. [1953a], Q. Appl. Math. 11, 350.
Koiter, W. [1953b], Anniversary Volume on Applied Mechanics Dedicated to
B. Biezeno, H. Stam, Haarlem, p. 232.
Koiter, W. T. [1956], Proc. Kon. Ned. Akad. Wet. B59, 24.
Koiter, W. T. [1960], Progress in Solid Mechanics (ed. I. N. Sneddon and R.
Hill), Vol. 1, North-Holland, Amsterdam, p. 206.
Kolsky, H., and L. S. Douch [1962], J. Mech. Phys. Solids 10195.
König, J. A. [1987], Shakedown of Elastic-Plastic Structures, Elsevier, Ams-
terdam.
Kotsovos, M. D., and H. K. Cheong [1984], ACI J. 81, 358.
Krajcinovic, D. [1973], Shock Vibr. Digest 5, 1.
Kratochvı́l, J. [1973], Acta Mech. 16, 127.
Krempl, E. [1987], Acta Mech. 69, 25.
Krempl, E., J. J. McMahon, and D. Yao [1986], Mech. Matls. 5, 35.
Krieg, R. D. [1975], J. Appl. Mech. 42, 641.
Krieg, R. D., J. C. Swearengen, and W. B. Jones [1987], in Unified Constitu-
tive Equations for Creep and Plasticity (ed. A. K. Miller), Elsevier Applied
Science, London, p. 245.
Krieg, R. D., J. C. Swearengen and R. W. Rohde [1978], in Inelastic Be-
havior of Pressure Vessel and Piping Components (ed. T. Y. Chang and E.
Krempl), American Society of Mechanical Engineers, New York, p. 15.
Kröner, E. [1960], Arch. Ration. Mech. Anal. 4, 273.
Lee, E. H. [1950], Proc. 3rd Symp. Appl. Math.(Ann Arbor, 1949), McGraw-
Hill, New York, p. 213.
Lee, E. H. [1952], Philos. Mag. 43, 549.
References 509
Lee, E. H. [1953], Q. Appl. Math. 10, 335.

Lee, E. H. [1969], J. Appl. Mech. 36, 1.
Lee, E. H., and D. T. Liu [1967], J. Appl. Phys. 38, 19.
Lee, E. H., and P. S. Symonds [1952], J. Appl. Mech. 19, 308.
Lee, E. H., and S. J. Tupper [1954], J. Appl. Mech. 21, 63.
Lehmann, Th. [1972], Ing.-Arch. 41, 297.
Lenskii, V. S. [1948], Prikl. Mat. Mekh. 12, 165.
Leth, C.-F. A. [1954], Div. Appl. Math. Report A11-107, Brown University,
Providence, R.I.
Lode, W. [1925], Z. angew. Math. Mech. 5, 142.
Loret, B. [1983], Mech. Matls. 2, 287.
Love, A. E. H. [1927], Mathematical Theory of Elasticity, 4th ed., Cambrdige
University Press, Cambridge, England; reprinted by Dover, New York.
Lubahn, J. D., and R. P. Felgar [1961], Plasticity and Creep of Metals, Wiley,
New York.
Lubliner, J. [1964], J. Mech. Phys. Solids 12, 59.
Lubliner, J. [1965], J. Méc. 4, 111.
Lubliner, J. [1972], Int. J. Non-Linear Mech. 7, 237.
Lubliner, J. [1974], Int. J. Solids Struct. 10, 310.
Lubliner, J. [1986], Mech. Matls. 5, 29.
Lubliner, J., and M. Valathur [1969], Int. J. Solids Struct. 5, 1275.
Ludwik, P. [1909], Elemente der Technologischen Mechanik, Springer-Verlag,
Berlin.
Lunts, Ya. L. [1949], Prikl. Mat. Mekh. 13, 55.
Maier, G. [1969], Int. J. Solids Struct. 5, 261.
Maier, G. [1970], J. Mech. Phys. Solids 18, 319.
Maier, G. [1983], Mech. Res. Comm. 10, 45.
Maier, G., R. Srinivasam, and M. A. Save [1972], in Proc. Int. Symp.
Computer-Aided Structural Design (Coventry, 1972), Vol. 1, p. 32.
Malvern, L. E. [1951a], Q. Appl. Math. 8, 405.
Malvern, L. E. [1951b], J. Appl. Mech. 18, 203.
Mandel, J. [1962], J. Méc. 1, 30.
Mandel, J. [1964], in Stress Waves in Anelastic Solids(IUTAM Symp., Prov-
idence, 1963, ed. H. Kolsky and W. Prager), Springer-Verlag, Berlin, p.
331.
Mandel, J. [1967], C. R. Acad. Sci. Paris 264A, 133.
Mandel, J. [1972], Plasticité Classique et Viscoplasticité (CISM, Udine, 1971),
Springer-Verlag, Vienna–New York.
510 References
Mandel, J. [1973], Int. J. Solids Struct. 9, 725.

Mandel, J. [1976], Mech. Res. Comm. 3, 251, 483.
Mandel, J., J. Zarka, and B. Halphen [1977], Mech. Res. Comm. 4, 309.
Manjoine, M. J. [1944], J. Appl. Mech. 11, 211.
Manolakos, D. E., and A. G. Mamalis [1986], Int J. Mech. Sci. 28, 815.
Marcal, P. V. [1965], Int. J. Mech. Sci. 7, 229, 841.
Markov, A. A. [1947], Prikl. Mat. Mekh. 11, 339.
Marsden, J. E., and T. J. R. Hughes [1983], Mathematical Foundations of
Elasticity, Prentice-Hall, Englewood Cliffs, N.J.
Martin, J. B. [1975], J. Mech. Phys. Solids 23, 123.
Martin, J. B. [1980], in Variational Methods in the Mechanics of Solids (ed.
S. Nemat-Nasser), Pergamon Press, Oxford, p. 278.
Massonnet, C. E., and M. A. Save [1965], Plastic Analysis and Design, Blais-
dell, Waltham, Mass.
McMeeking, R. M., and J. R. Rice [1975], Int. J. Solids Struct. 11, 601.
Melan, E. [1938], Ing.-Arch. 9, 116.
Mendelson, A. [1968], Plasticity: Theory and Applications, Macmillan, New
York.
Michell, J. H. [1900], Proc. London Math. Soc. 31, 141.
Miller, A. K. [1976], J. Engrg. Matls. Technol. 98, 97.
Mises, R. v. [1913], Göttinger Nachr. Math.-Phys. Kl., 582.
Mises, R. v. [1926], Z. angew. Math. Mech. 6, 199 (footnote to paper by F.
Schleicher).
Mises, R. v. [1928], Z. angew. Math. Mech. 8, 161.
Moreau, J. J. [1963], C. R. Acad. Sci. Paris 257A, 4117.
Moreau, J. J. [1970], C. R. Acad. Sci. Paris 271A, 608.
Moreau, J. J. [1976], in Applications of Methods of Functional Analysis to
Problems in Mechanics (ed. P. Germain and B. Nayroles), Springer-Verlag,
Vienna–New York, p. 56.
Mortelhand, I. M. [1987], Heron 32, 115.
Mróz, Z. [1967], J. Mech. Phys. Solids 15, 163.
Munro, J. [1979], Engineering Plasticity by Mathematical Programming (Proc.
NATO Adv. Study Inst., Waterloo, Canada, 1977, ed. by M. Z. Cohn and
G. Maier), Pergamon Press, New York, Ch. 7.
Nabarro, F. R. N. [1950], Some Recent Developments in Rheology, British
Rheologists’ Club, London.
Nabarro, F. R. N. [1975], in The Physics of Metals, Vol. 2: Defects (ed. P.
B. Hirsch), Cambridge University Press, Cambridge, England, p. 152.
References 511
Nadai, A. [1923], Z. angew. Math. Mech. 5, 1.

Nadai, A. [1950], The Theory of Flow and Fracture of Solids, McGraw-Hill,
New York.
Naghdi, P. M., and J. A. Trapp [1975], Int. J. Eng. Sci. 13, 785.
Nagtegaal, J. C. [1982], Computer Methods Appl. Mech. Engrg. 33, 469.
Nagtegaal, J. C., D. M. Parks, and J. R. Rice [1974], Computer Methods
Appl. Mech. Engrg. 4, 153.
Neal, B. G. [1963], The Plastic Methods of Structural Analysis, Chapman &
Hall, London.
Neal, B. G., and P. S. Symonds [1951], J. Instn. Civ. Engrs. 35, 21.
Neal, B. G., and P. S. Symonds [1952], Proc. Instn. Civ. Engrs. 1, 58.
Needleman, A. [1972], J. Mech. Phys. Solids 20, 111.
Nguyen, Q. S., and H. D. Bui [1974], J. Méc. 13, 321.
Niordson, F. I. [1985], Shell Theory, North-Holland, Amsterdam.
Noll, W. [1955], J. Ration. Mech. Anal. 4, 3.
Nowacki, W. K. [1978], Stress Waves in Non-Elastic Solids, Pergamon Press,
Oxford.
Oden, J. T. [1972], Finite Elements of Nonlinear Continua, McGraw-Hill,
New York.
Odqvist, F. K. G. [1933], Z. angew. Math. Phys. 13, 360.
Onat, E. T., and R. M. Haythornthwaite [1956], J. Appl. Mech. 23, 49.
Onat, E. T., and W. Prager [1953], J. Mech. Phys. Solids 1, 77.
Onat, E. T., and W. Prager [1954], in Proc. 1st Midwestern Conf. Solid.
Mech. (Urbana, 1953), p. 40.
Onat, E. T., and W. Prager [1955], Ingenieur 67, 0.46.
Onat, E. T., and R. T. Shield [1955], in Proc. 2nd U.S. Natl. Cong. Appl.
Mech. (Ann Arbor, 1954), American Society of Mechanical Engineers, New
York, p. 535.
Onat, E. T., W. Schumann, and R. T. Shield [1957], Z. angew. Math. Phys.
8, 485.
Orowan, E. [1934], Z. Phys. 89, 605.
Orowan, E. [1944], Proc. Instn. Mech. Engrs. 151, 133 (discussion of paper
by H. O’Neill).
Ortiz, M., and E. P. Popov [1985], Int. J. Num. Methods Engrg. 21, 1561.
Owen, D. R. [1968], Arch. Ration. Mech. Anal. 31, 91.
Owen, D. R. J., and E. J. Hinton [1980], Finite Elements in Plasticity: The-
ory and Practice, Pineridge Press, Swansea.
Pastor, J. [1976], Application de l’analyse limite à l’étude de la stabilité des
512 References
pentes et des talus, Thèse, Université de Grenoble (France).

Parkes, E. W. [1955], Proc. Roy. Soc. London A228, 462.
Perzyna, P. [1958], Arch. Mech. Stos. 10, 635.
Perzyna, P. [1963], Q. Appl. Math. 20, 321.
Perzyna, P. [1971], in Advances in Applied Mechanics(ed. by C.-S. Yih), Vol.
11, Academic Press, New York, p. 313.
Phillips, A. [1972], AIAA J. 10, 951.
Phillips, A. [1986], Int. J. Plast. 2, 315.
Phillips, A., and H. Moon [1977], Acta Mech. 27, 91.
Pinsky, P. M., M. Ortiz, and K. S. Pister [1983], Computer Methods Appl.
Mech. Engrg. 40, 137.
Pipkin, A. C., and R. S. Rivlin [1965], Z. angew. Math. Mech. 16, 313.
Prager, W. [1942], Duke Math. J. 9, 228.
Prager, W. [1947], J. Appl. Phys. 18, 375.
Prager, W. [1948], in Courant Anniversary Volume, Interscience, New York,
p. 289.
Prager, W. [1955a], Proc. Instn. Mech. Engrs. 169, 41.
Prager, W. [1955b], Ingenieur 67, 0.141.
Prager, W. [1956a], J. Appl. Mech. 23, 493.
Prager, W. [1956b], in Proc. 8th Int. Cong. Appl. Mech. (Istanbul, 1952),
Vol. 2, p. 65.
Prager, W. [1959], An Introduction to Plasticity, Addison-Wesley, Reading,
Mass.
Prager, W. [1961], An Introduction to Mechanics of Continua, Ginn, Boston.
Prager, W., and P. G. Hodge, Jr. [1951], Theory of Perfectly Plastic Solids,
Wiley, New York.
Prandtl, L. [1903], Phys. Z. 4, 758.
Prandtl, L. [1920], Göttinger Nachr. Math.-Phys. Kl., 74.
Prandtl, L. [1923], Z. angew. Math. Mech. 3, 401.
Prandtl, L. [1924], in Proc. 1st Int. Cong. Appl. Mech.(Delft, 1924), p. 43.
Rabotnov, Yu. N., [1969], Creep Problems in Structural Members, North-
Holland, Amsterdam.
Radenkovic, D. [1961], C. R. Acad. Sci. Paris 252, 4103.
Radenkovic, D. [1962], in Sém. de Plasticité (Ecole Polytechnique, Paris,
1961), P.S.T. Ministère de l’Air No. N.T. 116.
Rakhmatulin, Kh. A. [1945], Prikl. Mat. Mekh. 9, 91, 449.
Rakhmatulin, Kh. A. [1948], Prikl. Mat. Mekh. 12, 39.
Rakhmatulin, Kh. A., and G. S. Shapiro [1948], Prikl. Mat. Mekh. 12, 369.
References 513
Raniecki, B. [1979], Bull. Acad. Pol. Sci. Sér. Sci. Tech. 27, 391 [721].
Read, H. E., and G. A. Hegemier [1984], Mech. of Matls. 3, 271.
Read, W. T., Jr. [1953], Dislocations in Crystals, McGraw-Hill, New York.
Reed, K. W., and S. N. Atluri [1985], Int. J. Plast. 1, 63.
Reuss, A. [1930], Z. angew. Math. Mech. 10, 266.
Rice, J. R. [1970], J. Appl. Mech. 37, 728.
Rice, J. R. [1971], J. Mech. Phys. Solids 19, 433.
Roscoe, K. H. [1953], in Proc. 3rd Int. Conf. Soil Mech. Found. Engrg.
(Switzerland, 1953), Vol. 1, p. 186.
Rozvany, G. I. N. [1976], Optimal Design of Flexural Systems, Pergamon
Press, Oxford.
Rubin, M. B. [1986], Int. J. Engrg. Sci. 24, 1083.
Rubinstein, R., and Atluri, S. N. [1983], Computer Methods Appl. Mech.
Engrg. 36, 277.
Sacchi, G., and M. A. Save [1968], Meccanica 3, 43.
Salençon, J. [1972], Ann. I.T.B.T.P, No. 295-296, 90.
Salençon, J. [1973], in Proc. Int. Symp. Foundations of Plasticity (Warsaw,
1972, ed. A. Sawczuk), Noordhoff, Leyden.
Salençon, J. [1977], Applications of the Theory of Plasticity in Soil Mechan-
ics, Wiley, Chichester, England.
Salvadori, M. G., and F. L. DiMaggio [1953], Q. Appl. Math. 11, 223.
Save, M. A., and C. E. Massonnet [1972], Plastic Analysis of Plates, Shells
and Disks, North-Holland, Amsterdam.
Save, M. A., and W. Prager [1963], J. Mech. Phys. Solids 11, 255.
Schleicher, F. [1926], Z. angew. Math. Mech. 6, 199.
Schmid, E. [1924], in Proc. 1st Int. Cong. Appl. Mech. (Delft, 1924), p.
342.
Seiler, J. A., B. A. Cotter, and P. S. Symonds [1956], J. Appl. Mech. 23,
515.
Seiler, J. A., and P. S. Symonds [1954], J. Appl. Phys. 25, 556.
Shaffer, B. W., and R. N. House [1955], J. Appl. Mech. 22, 305.
Shaffer, B. W., and R. N. House [1957], J. Appl. Mech. 24, 447.
Shakir-Khalil, H., and G. S. Tadros [1973], Struct. Engrg. 51, 239.
Shanley, F. S. [1947], J. Aeronaut. Sci. 14, 261.
Simmons, J. A., F. Hauser, and J. E. Dorn [1962], Univ. Calif. Pubs. Engrg.
5, 177.
Simo, J. C. [1988a], Computer Methods Appl. Mech. Engrg. 66, 199.
Simo, J. C. [1988b], Computer Methods Appl. Mech. Engrg. 68, 1.
514 References
Simo, J. C., and T. J. R. Hughes [1988], Elastoplasticity and Viscoplasticity,

Computational Aspects, Stanford Univ., Dept. of Appl. Mech.
Simo, J. C., and M. Ortiz [1985], Computer Methods Appl. Mech. Engrg.
49, 221.
Simo, J. C., and K. S. Pister [1984], Computer Methods Appl. Mech. Engrg.
46, 201.
Simo, J. C., and R. L. Taylor [1985], Computer Methods Appl. Mech. Engrg.
48, 101.
Simo, J. C., and R. L. Taylor [1986], Int. J. Num. Methods Engrg. 22, 649.
Sokolnikoff, I. S. [1956], Mathematical Theory of Elasticity, 2nd ed., McGraw-
Hill, New York.
Sokolovskii, V. V. [1942], Prikl. Mat. Mekh. 6, 241.
Southwell, R. V. [1915], Philos. Mag. 29, 67.
Steele, M. C. [1954], J. Mech. Phys. Solids 3, 156.
Sternglass, E. J., and D. A. Stuart [1953], J. Appl. Mech. 20, 1953.
Stickforth, J. [1986], Int. J. Plast. 2, 347.
Symonds, P. S., and W. T. Fleming, Jr. [1984], Int. J. Impact Engrg. 2, 1.
Symonds, P. S., and C.-F. A. Leth [1954], J. Mech. Phys. Solids 2, 92.
Symonds, P. S., and T. J. Mentel [1958], J. Mech. Phys. Solids 6, 186.
Symonds, P. S., and B. G. Neal [1951], J. Franklin Inst. 252, 383, 469.
Taylor, D. W. [1937], J. Boston. Soc. Civ. Engrs. 24, 197.
Taylor, D. W. [1948], Fundamentals of Soil Mechanics, Wiley, New York.
Taylor, G. I. [1934], Proc. Roy. Soc. London A145, 362, 388.
Taylor, G. I. [1938], J. Inst. Metals 62, 307.
Taylor, G. I. [1946], J. Instn. Civ. Engrs. 26, 486.
Taylor, G. I. [1947], Proc. Roy. Soc. London A191, 441.
Taylor, G. I. [1948], Proc. Roy. Soc. London A194, 289.
Taylor, G. I. [1958], in Scientific Papers (ed. G. K. Batchelor), Vol. 2,
Cambridge University Press, Cambridge, England, p. 467.
Taylor, G. I., and H. Quinney [1931], Philos. Trans. Roy. Soc. London
A230, 323.
Telles, J. C. F., and C. A. Brebbia [1979], Appl. Math. Modelling 3, 466.
Temam, R. [1985], Mathematical Problems in Plasticity, Gauthier-Villars,
Paris.
Thomas, T. Y. [1958], J. Ration. Mech. Anal. 5, 251.
Thomas, T. Y. [1961], Plastic Flow and Fracture in Solids, Academic Press,
References 515
New York.
Timoshenko, S., and J. M. Gere [1961], Theory of Elastic Stability, 2nd ed.,
McGraw-Hill, New York.
Timoshenko, S. P., and J. N. Goodier [1970], Theory of Elasticity, 3rd ed.,
McGraw-Hill, New York.
Timoshenko, S., and S. Woinowsky-Krieger [1959], Theory of Plates and
Shells, 2nd ed., McGraw-Hill, New York.
Ting, T. C. T. [1969], J. Appl. Mech. 36, 203.
Ting, T. C. T., and N. Nan [1969], J. Appl. Mech. 36, 189.
Ting, T. W. [1966a], J. Math. Mech. 15, 15.
Ting, T. W. [1966b], Trans. Am. Math. Soc. 123, 1966.
Ting, T. W. [1967], Arch. Ration. Mech. Anal. 25, 342.
Ting, T. W. [1969a], Arch. Ration. Mech. Anal. 34, 228.
Ting, T. W. [1969b], J. Math. Mech. 19, 531.
Ting, T. W. [1971], Indiana Univ. Math. J. 20, 1047.
Tran-Cong, T. [1985], Ing.-Arch. 55, 13.
Truesdell, C. [1984], Rational Thermodynamics, Springer-Verlag, New York.
Truesdell, C., and W. Noll [1965], The Nonlinear Field Theories, in Handbuch
der Physik (ed. S. Flügge) Vol. III/3, Springer-Verlag, Berlin.
Truesdell, C., and R. A. Toupin [1960], The Classical Field Theories, in
Handbuch der Physik (ed. S. Flügge) Vol. III/1, Springer-Verlag, Berlin.
Turner, L. B. [1909], Trans. Cambridge Philos. Soc. 21, 377.
Valanis, K. C. [1971], Arch. Mech. 23, 517.
Valanis, K. C. [1980], Arch. Mech. 32, 171.
Valanis, K. C. [1985], J. Appl. Mech. 52, 649.
Walker, K. P. [1981], NASA Report CR-165533.
Wang, A. J. [1955], J. Appl. Mech. 22, 375.
Wang, A. J., and H. G. Hopkins [1954], J. Mech. Phys. Solids 3, 22.
Washizu, K. [1958], J. Math. Phys. 36, 306.
Washizu, K. [1975], Variational Methods in Elasticity and Plasticity, 2nd ed.,
Pergamon Press, Oxford.
Watanabe, O., and S. N. Atluri [1986], Int. J. Plast. 2, 37, 107.
Wilkins, M. L. [1964], in Methods of Computational Physics, Vol. 3 (ed. B.
Alder et al.), Academic Press, New York.
Winzer, A., and G. F. Carrier [1948], J. Appl. Mech. 15, 261.
Winzer, A., and G. F. Carrier [1949], J. Appl. Mech. 16, 346.
Woods, L. C. [1981], Q. Appl. Math. 39, 119; Bull. Inst. Math. Appl. 17,
98.
516 References
Yao, D., and E. Krempl [1985], Int. J. Plast. 1, 259.

Zaid, M. [1959]. J. Appl. Mech. 26, 462.
Ziegler, H. [1959], Q. Appl. Math. 17, 55.
Zienkiewicz, O. C. [1977], The Finite Element Method, 3rd ed., McGraw-Hill,
London.
Zienkiewicz, O. C., and I. Cormeau [1974], Int. J. Num. Methods Engrg. 8,
821.
Zienkiewicz, O. C., and R. L. Taylor [1989], The Finite Element Method, 4th
ed., Vol. 1, McGraw-Hill, London.
Zienkiewicz, O. C., S. Valliappan, and I. P. King [1969], Int. J. Num. Meth-
ods Engrg. 1, 75.
Index
Absolute temperature, 46, 478 continuous, 384

Acceleration wave ideal sandwich, 167, 235, 355
one-dimensional, 448, 451 under combined stresses, 355–
three-dimensional, 453–462 373
Acoustic tensor, 454 wide, 246–248
Active failure, 352 with distributed load, 383
Airy stress function, 35 with point loads, 381
Algorithmic tangent moduli, 267 Beam mechanism, 385
Almansi strain tensor, 469 Bending
Anisotropy, 82–84, 120 and extension, combined, 358–
Arch 364
buckling of, 326 and shear, combined, 369–373
semicircular, collapse of, 361– and torsion, combined, 364–366,
364 369
Associated flow rule, 129–134, 138– asymmetric, 238–239
148, 179 elastic-plastic, 233–249
in a plate, 399 of plates, 299–308
in beams under combined stresses, of rectangular beams, 233–235
360–361 plane of, 231–232
in large-deformation plasticity, pure, 229–239
488–490, 492 in plane strain, 246–249
in soil, 353 of a notched bar, 341–342
Lévy–Mises, 139 of a wide plate, 245–249
Tresca, 138 Bingham model, 88, 114
Autofrettage, 227 Blocks and slabs with grooves or
cutouts, 338–341
B-bar procedure, 268, 270 Boundary conditions, 36
Back stress, 101, 103, 119, 148 Boundary-value problem, 36
Bailey–Norton–Nadai law, 88, 447 Bounding surface, 149
Bank, vertical, stability of, 349– Buckling, 313–334
351 Bulk modulus, 53
Bauschinger effect, 83–84, 147, 149, Burgers circuit, 96
481 Burgers vector, 96–99
dislocation model of, 102
Beam, 229–244, 380–385 Cantilever, end-loaded
517
518 Index
beam-theory solutions, 239–243 Deformation, 14–20

numerical solution, 272 finite, 16–17, 465–471
plane-strain and plane-stress so- inelastic, 116–118
lutions, 342–347 infinitesimal, 16–21, 40
weakly supported, 344 plastic, see Plastic deformation
Cauchy–Green tensor(s), 467–468, Deformation gradient, 466
481, 482 Deformation power, 45, 473
Cauchy stress, 76–82, 473, 478 Deformation-rate tensor, 160, 471
Cauchy tetrahedron, 29 Deformation theory of plasticity,
Centered fan, 287 see Plasticity, deformation
Characteristics, 278–287, 448–450 theory of
Clausius–Duhem inequality, 47 Deviatoric strain, 20
Clay, 75, 104–108 Deviatoric stress, 32
Column curve, 320 Dilatancy, 109, 135, 347
Column formula, 320 Dilatation, 20
Compatibility conditions, 23 Discontinuous yielding, 82
Complementary-energy function, 51– Discretization, 38–41, 159, 250–268,
54 492
Complete solution, 165, 277 Dislocations, 94–103
Composite mechanism, 386 Dissipation
Compression test, 77–79 internal, 168
unconfined, 106 plastic, see Plastic dissipation
Concrete, 103, 105, 106, 108–110 viscous, 67
Constraints, internal, 36, 478 Dissipation inequality, 67
Contained plastic deformation, 180 Donnell theory, 439
Conventional strain, 76 Donnell–Mushtari–Vlasov (DMV)
Coronet, 292 theory, 332
Corotational rate, 496 Drag stress, 119
Cotter–Rivlin rate, 476 Drawing, 294–297
Coulomb mechanism, 348–353 Drucker’s inequality, 126, 134
Creep, 60–65 Drucker’s postulate, 125–129
Creep strain, 87–89 Drucker–Prager criterion, 143–145
Critical resolved shear stress, 93 Dynamic problems, 417–462
Crystals, 89–102
defects in, 92 e-delta identity, 2
Cutout factor, 338 Effective inelastic strain, 69
Cutting, 293–294 Effective stress, 107
Cylinder, thick-walled hollow, 216– Elastic body, 55–56
228 Elastic limit, 60
Cylindrical shell with a ring load, Elastic-limit locus, 358
410–413 Elastic modulus tensor, 49
Elastic-plastic boundary, 186–188,
Dead metal, 292 241, 244, 276
Defects in crystals, 92 Elastic-plastic modulus matrix, 253
Index 519
Elastic range, 60, 79–89 Finger deformation tensor, 468

Elastic strain, 81, 82 Finite-element method, 251–272, 493–
Green–Saint-Venant, 481 499
Elements, finite, 256–272, 497 First law of thermodynamics, 45
Endochronic theory, 120–121 Flow equation(s), 122–132
Energy, 44–57 Flow locus, 166
free (Helmholtz), 48, 486 Flow potential, 70, 114
internal potential, 54 Flow rule, 123–134, 138–146
strain, 54, 57 associated, see Associated flow
Energy balance, 45, 475 rule
Energy criterion (buckling), 315, nonassociated, 129, 490
317, 318, 327, 328, 331 Forging, 293–294
Engesser–Shanley theory, 319–324 Foulkes mechanism, 394–395
Entropy, 46 Foulkes’ theorems, 393–395
configurational, 486 Frames
specific, 48, 49 limit analysis, 385–390
Entropy density, 46, 48, 478 limit design, 390–397
Entropy production, internal, 47 Frank–Read source, 99–101
Entropy supply, external, 47 Free energy, 48, 486
Equibiaxial stress, 144
Free-energy density, 50, 478, 485–
Equilibrium equations, 36–38
488
cylindrical shell, 408
Friction
discretized, 256, 260, 263
angle of internal, 107
frame, 388
dry, 104, 107
hollow cylinder, 216, 220, 221,
Friction stress, 119
225, 228
Frictional materials, 104
hollow sphere, 206, 208, 210,
211, 215
incremental, 267 Gable frame, 396
plate, 300 Gable mechanism, 396
torsion, 190 Gauss points, 260
truss, 375 Gauss’s theorem, 7
weak form, 261, 268 Geiringer equations, 286
Equilibrium state, 65 Generalized coordinates, 39
Equivalent inelastic strain, 69 Generalized loads, 166, 355
Eulerian coordinates, 28, 465 Generalized stress and strain, 355–
Eulerian tensor, 470 357, 399, 404, 408, 409,
Extension and torsion 414
of a circular bar, combined, 366– Generalized velocity, 356
368 Grain boundaries, 92–102
of a thin-walled circular tube, Green–Saint-Venant strain tensor,
178 15, 467
Extremum principles, 154–161 elastic, 481
Extrusion, 294–297 Green’s lemma, 8
520 Index
Hadamard compatibility condition, Incompressibility, 48, 277, 479, 495,

453 496
Haigh–Westergaard space, 136, 143 Indentation, 292–294
Hall–Petch relation, 102 Indicial notation, 2
Hardening modulus, 154 Inelastic strain, see Strain, inelas-
Hardening properties, 112 tic
Hardening rules, 146–150 Inelastic work, 119
Hardening variable(s), 68, 121, 124, Inelastic work rate, 68, 119
147, 483, 497 Inelasticity, 59–60, 480–487
Heaviside step function, 114 Inequalities, method of, 390
Helmholtz free energy, 48 Infinitesimal deformation, 40
Hencky theory, see Plasticity, de- Infinitesimal strain, 16–21
formation theory of Initial-stiffness method, 267
Hencky’s theorem, 282 Initial-stress algorithm, 255
Hencky–Prandtl network, 282 Initially curved plates, 246–249
Hencky–Prandtl properties, 286 Instability, 183
Hinge circle Interaction diagram (or curve), 239,
in a circular plate, 309, 402, 355
426 Intermediate configuration, 482–490
in a cylindrical shell, 410, 411, Internal constraints, 36, 478
428 Internal dissipation, 168
Hohenemser–Prager model, 114, 447 Internal-energy density, 45
Hooke’s law Internal entropy production, 47
generalized, 50 Internal friction
in shear, 54 angle of, 107
Hu–Washizu principle, 59, 162, 261 Internal potential energy, 54
Internal variables, 60–71, 111–120,
Huber–Mises criterion, 114
485
Hypoelasticity, 494
Interpolation functions, 257, 260
Intrinsic time, 120
I-beam, 346, 361–373, 396 Irreversible processes, 65
Ideal sandwich beam, 167, 235, 355 Isoclinic intermediate configuration,
Ideal sandwich shell, 404 484
Il’iushin’s postulate, 128, 132, 134 Isotropic elasticity, 51–54
Impact or impulsive loading, 417– Isotropic hardening, 115, 121, 125,
433 134, 146–147
of a bar, longitudinal, 436–442 Iteration scheme, 266
of a cylindrical shell, 428
of beams, 419–425 Jacobian determninant, 466
of circular plates, 425–428 Jacobian matrix, 466
of of a cylindrical shell, 433 Jaumann rate, 476, 492–496
Imperfections (column), effect of, Johansen criterion, 309, 399–402
324 Joint mechanism, 387
Incipient plastic flow, 159 Jump conditions, 284, 340
Index 521
Kármán–Taylor–Rakhmatulin the- Limit locus, 166–168

ory, 442–446 Limit-load locus, 355
Kármán theory (buckling), 321–324 Limiting line, 287
Kelvin inequality, 67 Linear programming, 378–380
Kinematic hardening, 115, 119, 121, Linear viscoelasticity, 59–65
125, 130 Load, 36
Kinematic multiplier, 165 critical (buckling), 319
Kinematic shakedown theorem, 174 Euler, 319
Kinematic stability, 132 in limit analysis, 163
Kinematically admissible displace- ultimate, 185, 186, 188
ment field, 36–39, 155 Load-replacement theorem, 396
Kinematically admissible velocity Loading, 117, 165–173
field, 161, 163, 165, 166, impact or impulsive, 417–433
169, 170, 174, 277, 285, limit, 163, 169–170
286, 290, 338–340, 411 neutral, 123
Kirchhoff plate theory, 303 proportional or radial, 123
Kirchhoff stress, 485 ultimate, 163
Kirchhoff–Trefftz stress, 474 Loading direction, 166–168
Kronecker delta, 2 Loading surface, 149
Kuhn–Tucker conditions, 123 Locking, 268
Logarithmic strain, 76–82
Lagrange (Lagrangian) multipliers, Lower-bound theorem, 164, 356
261 Lüders bands, 82
Lagrange–Dirichlet theorem, 314
Lagrangian coordinates, 14, 434, Macauley bracket, 114
465 Malvern theory, 447
Lagrangian tensor, 469, 474, 476 Martin’s theorems, 418–419
Lamé coefficients, 52 Material time derivative, 28, 472–
Levi-Civita “e” tensor, 2 476
Lévy–Mises flow rule, 139 Maximum-plastic-dissipation pos-
Limit analysis tulate, 127, 160–166, 168,
loads in, 163 174, 175, 356, 357, 360,
of beams, 355–373, 380–385 399, 418, 419, 488
of frames, 385–390 consequences of, 128
of plane problems, 338–354 Maximum plastic dissipation, pos-
of plates, 398–403 tulate or principle of, see
of shells, 404–415 Maximum-plastic-dissipation
of slope stability, 351 postulate
of trusses, 374–378 Maximum plastic work, principle
theorems of, 162–170, 356 of, 162
Limit design Mechanism
of frames, 390–397 beam, 385
of trusses, 378–380 composite, 386
Limit loading, 163, 169–170 Coulomb, 348–353
522 Index
gable, 396 Noncentered fan, 288

joint, 387 Nonduality, principle of, 61
panel or sidesway, 386 Nonlinear programming, 396
Melan–Prager model (kinematic hard- Nonstandard material, 135, 163, 168–
ening), 147 170, 349, 354
Membrane analogy, 192 Normal stress, 30–40, 195
Membrane force, 300, 326 continuity of, across disconti-
Membrane-roof analogy, 198 nuity line, 288
Memory surface, 149 Normality, 128–132
Meridians, compressive and tensile, generalized, 70, 116
137 Notched bar, pure bending of, 341–
Mesh-sensitivity, 265 342
Metal-forming problems, 291–297
Metric tensor, 9, 469 Objectivity, 470, 475, 493, 496
Middle surface, 404 Octahedral shear stress, 33
Minimum-weight design Offset yield strength, 80
of continua, 395 Oldroyd rate, 476
of frames, 390–395 Overload factor, 165
of trusses, 378–379
Mises criterion, 114, 139–146 P-∆ effect, 358
associated flow rule, see Lévy– Panel mechanism, 386
Mises flow rule Passive failure, 352
in plane plastic flow, 276, 277, Peierls–Nabarro stress, 97
280, 281 Perfectly plastic, 81, 113, 122, 124–
plate, 309, 399–403 126, 128–131, 139, 154, 156,
Mises–Schleicher criterion, 144 157, 159
Mixed formulation, 261 Permanent strain, 79
Mohr rupture theory, 141 plateau of, 445
Mohr’s circle, 33–34, 141, 277 Permutation tensor, 2
Mohr–Coulomb criterion, 142–146, Perzyna model (viscoplasticity), 112,
280, 347 447
Moment-curvature relation, 233–237, π-plane, 136
240, 242, 246 Piola–Kirchhoff stress, 473–474
Piola stress, 473
Necking, 82, 184, 212, 497 Piola–Kirchhoff stress, 485
Neutral stress, 107 Plane plastic flow
Newton–Raphson method, 252 limit analysis, 338–354
Nguyen–Bui inequality, 134 slip-line theory, 277–290
Nodal displacements, 257 Plane strain, 24–25, 40
Nodal rotations, 257 pure bending in, 246–249
Nodes yield criteria in, 145–146
in crystals, 97 Plane stress, 35, 40
in finite elements, 256 yield criteria in, 144–145
Nominal stress, 76–82 Plastic collapse, 178
Index 523
Plastic deformation, 75–85 rate-independent, 111, 114, 116–

contained, 180 118
multiplicative decomposition of, Plates
481–484 buckling of, 326–331
Plastic dissipation, 139, 160, 165, circular, collapse of, 298–313
168 elastic, 303–308
associated, 140, 142–144 initially curved, 246–249
at a plastic hinge, 381 limit analysis of, 398–403
in a Foulkes mechanism, 395 theory of, 299–303
in a plate, 399 yielding of, 308–313
in a standard Mohr–Coulomb Poisson’s ratio, 52
material, 348 Polar decomposition, 468, 469
in beams under combined stresses, Postulate of maximum plastic dis-
356, 364, 365 sipation, see Maximum-plastic-
in large-deformation plasticity, dissipation postulate
487 Prager formula (stress-strain), 320
maximum, postulate or princi- Prandtl–Reuss equations, 140
ple of, see Maximum-plastic- Pressire
dissipation postulate on a wedge, 288–290
Plastic flow, 178 Pressure
in a cylindrical shell, 181, 407–
plane, see Plane plastic flow
410
pseudo-steady, 292
in a spherical shell, 178
steady, 291
in a thick-walled sphere, 206–
unrestricted, 163, 186
216
Plastic hinge, 342, 362, 371, 420–
in a thick-walled tube, 216–228
424
on a circular ring, 325–326
Plastic modulus
on a spherical cap, 414
of a beam section, 235 on a thin-walled tube, critical,
uniaxial (tensile-compressive), 333
132, 455, 458, 461 Principal axes
Plastic potential, 123 of area, 232
Plastic rotation, 483 of strain, 21
Plastic strain, 80–86, 117, 118 of stress, 32
Plastic work, 147 Principal invariants
maximum, principle of, 162 of Cauchy–Green tensors, 479
Plastically equivalent stresses, 160 of strain, 21
Plasticity of stress, 32
deformation theory of, 123, 195, of stress deviator, 136
330–332 Principal strains, 21
endochronic theory of, 120 Principal stress-deviator plane(π-
flow or incremental theory of, plane), 136
123 Principal stresses, 31
of soil, 104–109 Principle
524 Index
of energy balance, 45 Ring, buckling of, 325–326

of maximum plastic dissipation, Rocks, 103, 108–109
see Maximum-plastic-dissipation Rolling, 291
postulate Rotation, 16–23
of maximum plastic work, 162 Rotation tensor, 468, 481
of minimum complementary en-
ergy, 56, 155 Safety factor, 164
of minimum potential energy, Schmid’s law, 93
55, 154 Second law of thermodynamics, 46
of nonduality, 61 Shakedown, 170–175
of objectivity, 475 Shape factor, 234–237, 244
of virtual displacement, 38 Shear and bending, combined, 369–
of virtual forces, 39 373
of virtual work, 38 Shear angle, 17
Pure bending, see Bending, pure Shear directions, 278, 281, 285, 290
Pure shear, 18, 23 Shear force
in plates, 299, 302
Quasi-elastic material, 315 Shear lines, 278, 284, 286
Quasi-static shakedown theorem, 173 inextensibility of, 286
Quotient rule, 6 Shear modulus, 52, 192
Shear strain, 18, 191, 202
Radenkovic’s theorems, 169–170, 349, Shear strength, 93–97
353 theoretical, 93
Radial-return algorithm, 255 Shear stress, 30, 91–102
Ramberg–Osgood formula, 81 critical resolved, 93
Rankine’s formulas, 353 in torsion, 191, 193–195, 202
Rate equations, 60–70 octahedral, 33
Rate sensitivity, 59 Shear test
Recrystallization temperature, 83– direct, 105
87 simple, 105
Reduced-modulus theory (buckling), Shear-box test, 75, 105
321–324 Shearing rate, 200
Relaxation, 65 Shell(s)
Residual stress, 172–175 cylindrical, with a ring load,
in a beam, 237 410–413
in a thick-walled tube, 226 limit analysis of, 404–415
Retaining walls, 352–354 under external pressure, buck-
Return-mapping algorithm, 255, 266, ling of, 332–334
267, 272 Shock front, 435
Ridge line, 195–204 Shock relations, 435
Ridge point, 195, 197 Shock-speed equation, 435
Riemann–Christoffel tensor, 469 Sidesway mechanism, 386
Rigid-plastic material, 140, 159– Signum function, 138
162 Simple shear, 18, 191
Index 525
Simplex algorithm, 380 tangent, 266, 268

Slenderness ratio (column), 319 Strain, 15–25
Slip bands, 92–99 Almansi, 469
Slip directions, 91 conventional, 76, 467
Slip in crystals, 91–102 creep, 87–89
Slip-line theory, 277–290 deviatoric, 20
Slip lines elastic, 61, 81, 82
envelope of (limiting line), 287 Green–Saint-Venant, 15, 467
in crystals, 92 inelastic, 61–63, 80
in plane plastic flow, 286–290 infinitesimal, 16–21
Slip planes, 91–102 logarithmic, 76–82
Slip system, 91–102 permanent, 79
Slope stability, 351 plastic, 80–86, 117, 118
Soap-film analogy, 192 thermal, 53
Softening, 108–110, 113, 124–126, Strain aging, 82
128, 130, 131, 265 Strain deviator, 20
Soil mechanics, problems in, 347– Strain-displacement relation, 40
354 Strain energy, 54, 57
Soil, plasticity of, 104–109 Strain-energy function, 50
Sokolovskii–Malvern theory, 447 Strain tensor, 15–24
Sphere, thick-walled hollow, 206– Almansi, 469
216 Green–Saint-Venant, 15, 467
thermal stresses in, 213–216 Strain-hardening, 80
Spherical cap under pressure, 413– Strain-softening, see Softening
415 Stress, 29–40
Stability factor, 351 Cauchy, 76–82, 473, 478
Stability of a vertical bank, 349– deviatoric, 32
351 discontinuity of, 284
Stability theory, 314–319 effective, 107
Standard material, see also Asso- Kirchhoff–Trefftz, 474
ciated flow rule, 129, 132, neutral, 107
134, 135, 153, 156, 163 nominal, 76–82
Static multiplier, 164 normal, 30–40
Statically admissible moment dis- octahedral shear, 33
tribution, 389, 401 Piola, 473
Statically admissible stress field, 36, Piola–Kirchhoff, 473–474, 485
39, 155, 157, 161, 162, 277, shear, see Shear stress
290 thermal, 57
Statically determinate, 177, 276, in a thick-walled hollow sphere,
375, 380 213–216
Statically indeterminate, 184, 375, true, 76
380, 385 yield, see Yield stress
Stiffness matrix, 261, 263, 264, 266– Stress deviator, 32
268 Stress function
526 Index
Airy, 35 Cauchy–Green, 467–468, 481,

in torsion, 190–192, 194, 195, 482
197, 202 Eulerian, 470
Stress-strain diagram (or curve), 76– Lagrangian, 469
86, 113–121 Levi-Civita “e” or permutation,
bilinear, 439 2
piecewise linear, 442 metric, 9, 469
static, 113 rotation, 468, 481
uniaxial, 124, 125, 127 strain, 15–24
Stress-strain relation(s), 44–56, 80– stress, 29
86 stretch, 468
adiabatic, 48 tangent modulus, 315
bilinear, 439, 440 two-point, 466
dynamic, 437 Tensor field, 6–10
incremental, 252 Tensor product, 6
isentropic, 48 Theoretical shear strength, 93
isothermal, 48 Thermal strain, 53
linear, 49 Thermodynamic forces, conjugate,
nonlinear elastic, 442 115–119
rate dependence of, 85 Thermoelastic body, 45–48, 478
temperature dependence of, 85 Thermoelasticity, 45–49, 478–480
Stress tensor, 29, 473 Thrust on retaining walls, 352–354
Stretch, 467, 481 Torque, 177, 189–202
Stretch tensor, 468 ultimate, 180, 196, 200–202
Stretch tensors, 481 Torque-twist diagram, 201
Stretching tensor, 160 Torque-twist relation, 272
Superposition of mechanisms, 388 Torsion, 189–204
Surface traction, 27–41 and bending, combined, 364–
369
Tangent modulus tensor, 315 and extension of a thin-walled
effective, 454, 456 circular tube, 178
Tangent-modulus theory (buckling), and extension, combined, 366–
319–324 368
Tangent stiffness, 266–268 fully plastic, 195
elastic (in finite deformation, of an I-beam, numerical solu-
487 tion, 272
Temperature Traction, 27–41
absolute, 46, 478 Transition temperature, 85
recrystallization, 83–87 Trefftz criterion, 315
transition, 85 Tresca criterion, 137–145, 298
Tensile strength, 82 associated flow rule, 138, 139,
Tension test, 75–77 146
Tensor, 1–13 in plane plastic flow, 276, 277,
acoustic, 454 280, 281
Index 527
plate, 309, 399, 403 Virtual work, 37–41

True stress, 76 principle of, 38
Truss, 374–380 Viscoelasticity, 59–65
Tube, thick-walled, 216–228 Viscoplasticity, 88, 111–121, 251–
Twist, 191 252, 447
Twisting moment (plate), 302
Two-point tensor, 466 Warping function, 191, 194, 200
Two-surface models, 149 Waves in elastic-plastic solids
one-dimensional, 434–452
Ultimate load, 185, 186, 188 three-dimensional, 452–462
Ultimate loading, 163 Weakly supported cantilever, 344
Ultimate moment, 234–236, 248 Wedge, 288–290
Ultimate tensile strength, 81 Whiskers, 94
Ultimate torque, 180, 196, 200–202 Work-hardening, 80–86, 124–134,
Unconfined compression test, 106 138, 141, 147, 149, 153,
“Unified” viscoplasticity models, 118– 154, 156, 157
120 dislocation models of, 101–102
Uniqueness theorems, 152–154, 160– in a thick-walled sphere, effect
161 of, 212–213
Unloading, 123, 127, 481, 483 in a thick-walled tube, effect
Unloading wave, 451, 456 of, 227–228
Unrestricted plastic flow, 163, 186 in torsion, 195
Upper-bound theorem, 164, 356 isotropic, see Isotropic hard-
ening
Variational principles, 158–159 kinematic, see Kinematic hard-
Vector field, 7 ening
Velocity discontinuity, 287, 343, 348, Work-hardening modulus, 132
435, 436, 448 in isotropic hardening, 147
Velocity field in two-surface models, 149
conical (in a plate), 426 Work-softening, see Softening
extremum principle for, 161
in plane problems, 277–292, 337– Yield criterion, 111–116, 123–131,
346 135–147, 254, 255, 263
kinematically admissible, see Kine- anisotropic, 140
matically admissible veloc- dependent on the mean stress,
ity field 135
virtual, 475 Drucker–Prager, 143–145
Virtual displacement field, 37 in strain space, 131
Virtual displacements, principle of, independent of the mean stress,
38 137
Virtual forces, principle of, 39 isotropic, 135–147
Virtual strain field, 37 Johansen, 309, 399–402
Virtual stress field, 39 Mises, see Mises criterion
Virtual velocity field, 475 Mises–Schleicher, 144
528 Index
Mohr–Coulomb, 142–146, 347

piecewise linear, 363
plate, 309
Tresca, see Tresca criterion
Yield curve (for combined exten-
sion, bending and torsion),
365
Yield curve (in stress plane), 178
Yield drop, 82
Yield function, 488–491
Yield line, 400
Yield-line theory, 400–402
Yield locus, 166
piecewise linear, 361
Yield point, 101
Yield strength, 80
of polycrystals, 102
offset, 80
Yield stress, 80–88, 101
dynamic, 418, 437
initial, 437, 442
rate sensitivity of, 424
static, 418, 447
Yield surface, 60, 111–121, 123–
134, 136–150, 253, 254, 489
convexity of, 129, 254
dynamic, 118
in π-plane, 137, 140
in Haigh–Westergaard space,
136, 143
in strain space, 132
viscoplastic, 118
Young’s modulus, 52

Plasticity Theory

Uploaded by

Copyright:

Available Formats

Plasticity Theory

Uploaded by

Document Information

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

Plasticity Theory

Uploaded by

Copyright:

Available Formats

PLASTICITY

Revised Edition (PDF)

graduate or first-year graduate level of a North American university. I have

Addendum: Revised Edition

Chapter 1: Introduction to Continuum Thermome-

Section 1.1 Mathematical Fundamentals 1

Section 1.2 Continuum Deformation 14

Section 1.3 Mechanics of Continuous Bodies 26

Section 1.4 Constitutive Relations: Elastic 44

Section 1.5 Constitutive Relations: Inelastic 59

Chapter 2: The Physics of Plasticity

Section 2.2 Crystal Plasticity 89

Section 2.3 Plasticity of Soils, Rocks and Concrete 103

Chapter 3: Constitutive Theory

Section 3.2 Rate-Independent Plasticity 122

3.3.2 Yield Criteria Independent of the Mean Stress 137

Section 3.4 Uniqueness and Extremum Theorems 152

Section 3.5 Limit-Analysis and Shakedown Theorems 162

Chapter 4: Problems in Contained Plastic Defor-

Section 4.2 Elastic–Plastic Torsion 189

Section 4.3 The Thick-Walled Hollow Sphere and Cylinder 205

Section 4.4 Elastic–Plastic Bending 229

Section 4.5 Numerical Methods 250

Chapter 5: Problems in Plastic Flow and Collapse

Section 5.1 Plane Problems 276

Section 5.2 Collapse of Circular Plates 298

Section 5.3 Plastic Buckling 313

Chapter 6: Problems in Plastic Flow and Collapse

Section 6.1 Limit Analysis of Plane Problems 338

Section 6.2 Beams Under Combined Stresses 355

Section 6.3 Limit Analysis of Trusses, Beams and Frames 374

Section 6.4 Limit Analysis of Plates and Shells 398

Chapter 7: Dynamic Problems

Section 7.2 One-Dimensional Plastic Waves 434

Section 7.3 Three-Dimensional Waves 452

Chapter 8: Large-Deformation Plasticity

Section 8.2 Large-Deformation Constitutive Theory 478

Section 8.3 Numerical Methods in Large-Deformation Plastic-

Section 1.1 Mathematical Fundamentals

Solid mechanics, which includes the theories of elasticity and plasticity, is

ally used in undergraduate “strength of materials” courses, but it is often

The Kronecker delta is defined as

The Levi-Civita “e” tensor or permutation tensor is defined as

eijk elmk = δil δjm − δim δjl .

The fundamental operations of three-dimensional vector algebra, pre-

Decomposition: u = ui ei . (The position vector in x1 x2 x3 -space is denoted

Scalar (dot) product between unit vectors: ei · ej = δij (orthonormality).

Scalar (dot) product between any two vectors: u · v = ui ei · ej vj = ui vi .

Vector (cross) product between unit vectors: ei × ej = eijk ek .

Vector (cross) product between any two vectors: u × v = ei eijk uj vk .

Scalar triple product: u · (v × w) = (u × v) · w = eijk ui vj wk .

1.1.2. Cartesian Tensors

u∗i = βik uk , ui = βji u∗j .

u∗i = βik βjk u∗j .

Figure 1.1.1. Example of a rotation represented by a proper orthogonal matrix.

The preceding definitions are independent of any decomposition of the

If λA = 0 (i.e., λij = λji ), then λ is a symmetric operator. If λS = 0 (i.e.,

Thus a scalar quantity is a tensor of rank 0, a vector is a tensor of rank 1,

then α necessarily represents a tensor α of rank 2. Similarly, if α and β are

αij = ρijkl βkl ,

BM u(r + ∆r)