OXFORD SERIES ON MATERIALS MODELLING

Series Editors

Adrian P. Sutton, FRS

Department of Physics, Imperial College London

Robert E. Rudd
Lawrence Livermore National Laboratory
 OXFORD SERIES ON MATERIALS MODELLING

Materials modelling is one of the fastest growing areas in the science and engi-
neering of materials, both in academe and in industry. It is a very wide field
covering materials phenomena and processes that span ten orders of magni-
tude in length and more than twenty in time. A broad range of models and
computational techniques has been developed to model separately atomistic,
microstructural, and continuum processes. A new field of multiscale modeling
has also emerged in which two or more length scales are modeled sequentially
or concurrently. The aim of this series is to provide a pedagogical set of texts
spanning the atomistic and microstructural scales of materials modeling, written
by acknowledged experts. Each book will assume at most a rudimentary knowl-
edge of the field it covers and it will bring the reader to the frontiers of current
research. It is hoped that the series will be useful for teaching materials modeling
at the postgraduate level.

APS, London
RER, Livermore, California

1. M. W. Finnis: Interatomic Forces in Condensed Matter

2. K. Bhattacharya: Microstructure of Martensite—Why It Forms and How It Gives
Rise to the Shape-Memory Effects
3. V. V. Bulatov, W. Cai: Computer Simulations of Dislocations
4. A. S. Argon: Strengthening Mechanisms in Crystal Plasticity
5. L. P. Kubin: Dislocations, Mesoscale Simulations and Plastic Flow
6. A. P. Sutton: Physics of Elasticity and Crystal Defects
7. D. Steigmann: A Course on Plasticity Theory

Forthcoming:
D. N. Theodorou, V. Mavrantzas: Multiscale Modelling of Polymers
 A Course on Plasticity Theory

David J. Steigmann
Department of Mechanical Engineering, University of California, Berkeley
 Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© David J. Steigmann 2022
The moral rights of the author have been asserted
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2022942494
ISBN 978–0–19–288315–5
DOI: 10.1093/oso/9780192883155.001.0001
Printed and bound by
CPI Group (UK) Ltd, Croydon, CR0 4YY
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
To my family
Preface

The theory of plasticity has a long and interesting history dating back about two and a
half centuries. Activity in the field expanded rapidly over the course of the past century
in particular, giving rise to a rapid pace of advancement. During much of the latter phase
of its modern development, the field was beset by ambiguity and controversy concerning
some of its conceptual foundations. Unsurprisingly, this led to the emergence of differ-
ent, often incompatible, schools of thought on the subject. A comprehensive survey of
the state of plasticity theory during this period may be found in the review article by
Naghdi. Meanwhile, great strides were being made by applied mathematicians in lay-
ing the foundations of modern continuum mechanics. Their emphasis on permanence
and rigor meant that the unsettled subject of plasticity theory was largely avoided, how-
ever, with the result that this lacuna in the panoply of continuum theories began to be
filled quite recently, around the turn of the millennium, after the antagonism of the older
schools had begun to fade.
Practically everything known about plasticity through the middle of the past century
is documented in the superb treatises by Prager and Hodge, Nadai, Hill, and Kachanov,
which should be carefully read by any serious student of our subject. At around the same
time, new developments were taking place in the application of differential geometry to
the continuum theory of defects associated with plasticity. This has become a large and
active discipline in its own right, and a substantial part of this book is devoted to it.
The works of Bloom and Wang, and the volume edited by Kröner, are recommended to
those interested in learning about its foundations, while those by Clayton, Epstein and
El anowski, Epstein, and Steinmann cover many of the more recent developments. The
modern engineering theory, as distinct from the geometrical theory, is ably summarized
in the books by Lubliner, Besseling and van der Giessen, and Bigoni. The books by Han
and Reddy and by Gurtin et al. are recommended for mathematical developments and
some of the more recent thinking on the subject.
While writing this book I have been guided by the belief that one can always learn
something from any thoughtful person. Accordingly the contents reflect my understand-
ing of the work of researchers and scholars spanning a large and diverse range of views
on the subject of plasticity. In the course of surveying the modern literature, I have been
struck by the continuing isolation of the various schools from one another, with scant
evidence of cross-fertilization. Particularly glaring, from my perspective, is the lack of
acknowledgment of the efforts of Noll in laying the foundations of the modern theory.
This has been rectified to a great degree by Epstein and El anowski, and I follow their
lead in giving primacy to Noll’s perspective. In fairness, Noll is not an easy read, and
much study is needed to grasp the full import of his work.
 Preface vii

The book is certainly not self-contained. Readers are presumed to have had prior
exposure to a good introductory course on basic continuum mechanics at the level of
the excellent books by Chadwick and Gurtin, for example. Aspects of this basic back-
ground are summarized as needed, but not developed in any detail. The emphasis here
is on conceptual issues concerning the foundations of plasticity theory that have proved
challenging, to me at least. These have led me to the view that the time has come to seek
a measure of consolidation and unification in the field. I do not ignore the classical the-
ory, but rather develop it from the perspective of the modern theory. For example, the
classical theory of perfectly plastic solids was presented historically in a way that led to
its natural interpretation, from the vantage point of modern continuum mechanics, as a
theory of non-Newtonian fluids rather than as a model of the behavior of certain solids.
The resolution of this dilemma is a prime example of the clarity that can be achieved
once a secure logical foundation for the general theory has been established.
Some explicit solutions to the equations of plasticity theory are covered in this book,
but not nearly to the extent found in the older books. The reason for this omission
is, firstly, that the small collection of explicit solutions that are known is ably covered
elsewhere, so that duplication is hardly justified, and secondly, that due to the advent
of modern computing, they are not nearly as relevant as they once were. I devote the
remainder of the book to the theoretical foundations of the subject, in accordance with
my own predilections, rather than to matters having to do with computation. The reason
for this emphasis is my belief that students are typically not as well versed in the concep-
tual foundations as they should be if they are to realize the full potential of computational
mechanics. A number of exercises of varying degrees of difficulty appear throughout.
These serve to reinforce understanding and to encourage the reader to fill in any gaps in
the development. Comprehensive solutions to selected exercises are included at the end
of the book.
Those who might have read my previous book, Finite Elasticity Theory, will find the
style and presentation of this one to be quite familiar. The present book is perhaps a bit
more demanding, however, insofar as various concepts from non-Euclidean differential
geometry are covered in detail. I gratefully acknowledge the small group of dedicated
graduate students at the University of California, Berkeley, whose interest and persis-
tence provided the impetus for the development of a graduate course on which the book
is based. I am especially grateful to one of them, Milad Shirani, for his critical reading
of the manuscript and for preparing the figures.
David Steigmann
Berkeley, 2021

References
Besseling, J. F., and van der Giessen, E. (1994). Mathematical Modelling of Inelastic Deformation.
Chapman and Hall, London.
Bigoni, D. (2012). Nonlinear Solid Mechanics: Bifurcation Theory and Material Instability. Cam-
bridge University Press, Cambridge, UK.
viii Preface

Bloom, F. (1979). Modern Differential Geometric Techniques in the Theory of Continuous Distributions
of Dislocations. Lecture Notes in Mathematics, Vol. 733. Springer, Berlin.
Chadwick, P. (1976). Continuum Mechanics: Concise Theory and Problems. Dover, New York.
Clayton, J. D. (2011). Nonlinear Mechanics of Crystals. Springer, Dordrecht.
Epstein, M. (2010). The Geometrical Language of Continuum Mechanics. Cambridge University
Press, Cambridge, UK.
Epstein, M., and Elżanowski, M. (2007). Material Inhomogeneities and Their Evolution. Springer,
Berlin.
Gurtin, M. E. (1981). An Introduction to Continuum Mechanics. Academic Press, Orlando.
Gurtin, M. E., Fried, E., and Anand, L. (2010). The Mechanics and Thermodynamics of Continua.
Cambridge University Press, Cambridge, UK.
Han,W., and Reddy, B. D. (2013). Plasticity: Mathematical Theory and Numerical Analysis.
Springer, N.Y.
Hill, R. (1950). The Mathematical Theory of Plasticity. Clarendon Press, Oxford.
Kachanov, L. M. (1974). Fundamentals of the Theory of Plasticity. MIR Publishers, Moscow.
Kröner, E. (Ed) (1968). Proc. IUTAM Symposium on Mechanics of Generalized Continua. Springer,
N.Y.
Lubliner, J. (2008). Plasticity Theory. Dover, N.Y.
Nadai, A. (1950). Theory of Flow and Fracture of Solids. McGraw-Hill, N.Y.
Naghdi, P. M. (1990). A critical review of the state of finite plasticity. J. Appl. Math. Phys. (ZAMP)
41, 315–394.
Noll, W. (1967). Materially uniform simple bodies with inhomogeneities. Arch. Ration. Mech.
Anal. 27, 1–32.
Prager, W., and Hodge, P. G. (1951). Theory of Perfectly Plastic Solids. John Wiley & Sons, N.Y.
Steigmann, D. J. (2017). Finite Elasticity Theory. Oxford University Press, Oxford.
Steinmann, P. (2015). Geometrical Foundations of Continuum Mechanics: An Application to First
and Second-Order Elasticity and Elasto-Plasticity. Lecture Notes in Applied Mathematics and
Mechanics, Vol. 2. Springer, Berlin.
Wang, C. -C. (1979). Mathematical Principles of Mechanics and Electromagnetism. Part A: Analytical
and Continuum Mechanics. Plenum Press, N.Y.
Contents

1 Preliminaries 1
1.1 Phenomenology 1
1.2 Elements of continuum mechanics 4

2 Brief resumé of nonlinear elasticity theory 15

2.1 Stress and strain energy 15
2.2 Conservative problems and potential energy 21
2.3 The Legendre–Hadamard inequality 25
2.4 Material symmetry 29

3 A primer on tensor analysis in three-dimensional space 34

3.1 Coordinates, bases, vectors, and metrics 34
3.2 Second-order tensors 41
3.3 Derivatives and connections 44
3.4 The Levi-Civita connection 50
3.5 The curl, Stokes’ theorem, and curvature 54
3.6 Torsion and the Weitzenböck connection 59

4 Deformation and stress in convected coordinates 63

4.1 Deformation gradient and strain 63
4.2 Strain compatibility 66
4.3 Stress, equations of motion 71

5 Elastic and plastic deformations 74

5.1 Elastic–plastic deformation, dislocation density 74
5.2 Differential-geometric considerations 79
5.3 Incompatibility of the elastic strain 82

6 Energy, stress, dissipation, and plastic evolution 87

6.1 Materially uniform elastic bodies 87
6.2 Surface dislocations and stress relaxation 93
6.3 Dissipation due to plastic evolution 97
6.4 Superposed rigid-body motions 100
x Contents

6.5 Yielding and plastic flow 102

6.6 Crystallinity versus isotropy 109
6.7 Discontinuous fields 111

7 Isotropy 119
7.1 The flow rule 119
7.2 Von Mises’ yield function 121
7.3 The classical theory for isotropic rigid-plastic materials 123
7.4 Bingham’s model of viscoplasticity 126
7.4.1 Example: Steady channel flow 127
7.5 Plane strain of rigid-perfectly plastic materials: Slip-line theory 129
7.5.1 Stress, equilibrium 130
7.5.2 Velocity field 132
7.5.3 Cartesian form of the equations 133
7.5.4 Further theory for plane strain 136
7.5.5 Axisymmetric state exterior to a traction-free circular hole 139
7.6 Anti-plane shear 143

8 Small-deformation theory 147

8.1 The displacement field 147
8.2 Approximations for small displacement gradients 151
8.3 The frame-invariant Prandtl–Reuss theory 153

9 Strain hardening, rate sensitivity, and gradient plasticity 156

9.1 “Isotropic” hardening 156
9.2 Rate sensitivity: Viscoplasticity 159
9.3 Scale effects 160
9.3.1 Crystalline symmetry 162
9.3.2 Isotropy 163
9.4 Scale-dependent yielding 166
9.5 Gradient plasticity 168
9.5.1 Energetic response functions 169
9.5.2 Stress power, balance laws, and dissipation 170
9.5.3 Example 174

Solutions to selected problems 177

Index 192
1
Preliminaries

We begin with a fairly descriptive discussion of the main observations about plastic
behavior and the basic mechanisms responsible for it. This is followed by a brief resumé
of the standard continuum theory that underpins our subsequent development of a
theoretical framework for the description of elastic-plastic response.

1.1 Phenomenology
Much of the basic phenomenology of plasticity can be understood in terms of a simple
tension-compression test on a uniform metallic bar. Suppose the bar has length l 0 in its
unloaded state, and let T = F/A be the uniaxial Cauchy stress in the direction of the bar
axis, where F is the axial force and A is the cross-sectional area of the deformed bar.
The stretch of the bar, presumed to be strained homogeneously, is λ = l/l0 , where l is
the bar’s length when deformed. If the stress is not too large, the response of the bar is
typically well described by the linear relation

T = E ln λ (1.1)

between the stress and the logarithmic strain ln λ, in which the proportionality constant
E is Young’s modulus—a property of the material of which the bar is made. This relation
presumes the state of the bar, as determined by the stretch and the Cauchy stress, to be
uniform. The bar is then in equilibrium insofar as the effects of body forces (e.g., the
weight of the bar) can be neglected.
The range of stresses for which this relation holds is limited. It fails when the stress
reaches certain limits, called the yield stresses in uniaxial tension or compression. Often
these limits coincide in magnitude, so that (1.1) is valid provided that

|T| < TY , (1.2)

where TY , another property of the material, is the initial yield stress, the qualifier reflect-
ing the fact that the yield stress usually evolves with the state of the material under
continued deformation, its current value typically exceeding the initial value. This phe-
nomenon, called strain hardening, is depicted schematically in Figure 1.1. If the bar is

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0001
2 Preliminaries

F
F T=
A

F=0
F=0
Ty

E E E

l0 l li

ln λ

–Ty ln λP ln λ

Figure 1.1 Uniaxial stress–strain response of a bar.

unloaded to zero stress from a state in which the value of |ln λ| exceeds that associ-
ated with initial yield, then the bar does not return to its initial length l 0 , but rather
to an intermediate length li . We say, rather loosely, that the bar has been permanently
deformed. The plastic stretch associated with l i is λp = li /l0 . Further, the slope of the
unloading curve is approximately constant and equal to that of the loading curve,
namely E.
The stretch λ of the bar just prior to unloading is thus given by

λ = λe λp , (1.3)

where λe = l/li is the elastic stretch, so named because, according to the graph,

T = E(ln λ − ln λp ) = E ln λe . (1.4)

Thus, the elastic stretch bears the same relation to the stress as that associated with the
initial elastic response of the bar.
The phenomenology just described leads immediately to the important observation
that the elastic properties of the material, as reflected in the uniaxial case by Young’s
modulus, are roughly insensitive to plastic deformation. This observation carries over to
other elastic properties of crystalline materials, as documented in the extensive experi-
mental work of G. I. Taylor and associates and summarized in the introductory chapter
of Hill’s classic treatise. In particular, the basic lattice structure of a metallic crystal, the
seat of its elastic properties, remains largely undisturbed by the relative plastic slip of
 Phenomenology 3

crystallographic planes. This observation will be incorporated as a cornerstone of the

theory to be developed.
Beyond this it is invariably true that TY /E ≪ 1 in metals, implying that |T| /E ≪ 1
and hence that |ln λe | ≪ 1. Accordingly, T ≃ Eεe , where εe = λe − 1; that is, the elastic
strain εe is invariably small in magnitude. A further observation about the uniaxial bar
test is that λp remains unchanged as long as |T| <TY , the current value of the yield stress
just prior to unloading. Indeed the primary purpose of plasticity theory is to describe
how λp , or, more accurately, its three-dimensional counterpart, evolves when the yield
limit is reached. In connection with this it is necessary to have an effective model of strain
hardening, this arguably constituting the main open problem of the phenomenological
theory. Indeed this aspect of the subject is a principal focus of much contemporary
research. Later in the book, we will endeavor to summarize some of the current thinking
in this area.
We have mentioned the role of slip along crystallographic planes as a basic mechanism
of plastic deformation, giving rise to an overall shear deformation on the macro-scale.
This is essentially a frictional process and thus entails the dissipation of energy. The
sliding does not take place all at once, but is instead the product of the progressive move-
ment of dislocations through the crystal lattice. Clear illustrations of this phenomenon
are given in Figure 102.2 in the book by Gurtin et al. and Chapter 1 of the book by
Kovács and Zsoldos. Roughly, the movement of a dislocation is initiated by the break-
ing of an atomic bond between two atoms occupying adjacent layers as they displace
relative to each other in response to an applied shear stress, say. A displaced atom then
forms a bond with its new nearest neighbor. This process continues in a sequential man-
ner until all the atoms in a given layer are displaced by one lattice spacing relative to
those occupying the adjacent layer. The reason why this process occurs via the pas-
sage of a dislocation rather than all at once is that the dislocation mechanism requires
substantially less effort. This can be readily understood in terms of the famous carpet
analogy: Thus, imagine being tasked with the job of displacing a carpet across a floor,
all the while maintaining a substantial amount of contact between the two. This is the
analog of the relative plastic slip of adjacent planes of the lattice. One can drag the car-
pet wholly, of course, but it is much easier to create a narrow wrinkle at one end and
simply push it across the remaining part of the carpet. The net effect of this proce-
dure is that the entire carpet has been displaced en masse once it has been traversed by
the wrinkle.
A dislocation engenders a local distortion of the lattice in the course of its movement
along a crystallographic plane. Recalling that the lattice is the seat of the elastic response
of the material, it follows that dislocations indirectly induce a local stress field in their
vicinity. Thus, to the extent that dislocations are present in an unloaded crystal, they
generate a field of residual stress in the material. This too is something that a good theory
should be able to predict.
Naturally dislocations in real crystals are, like atoms, discrete features, but, like atoms,
they are usually so densely distributed in a typical sample as to render meaningful their
description in terms of a continuous distribution. We then speak of a dislocation density in
much the same way as mass density is used to model densely distributed matter. In turn,
4 Preliminaries

the notion of a dislocation density has a fascinating connection with certain concepts in
non-Euclidean differential geometry, to be explored later.
Most metallic parts used in engineering applications are polycrystalline, consisting
of small grains of pure crystal within which the mechanism of dislocation motion is
operative. These grains join at grain boundaries, where their interactions contribute to
the overall plastic response of the aggregate. Often these grains are more or less ran-
domly oriented, so that at a mesoscopic scale the aggregate responds in the manner of
an isotropic continuum. For this reason the classical theory of plasticity is concerned
almost exclusively with the response of isotropic materials, whereas theories for crys-
talline materials are largely confined to the research literature. The volume edited by
Teodosiu and the books by Havner and Gurtin et al. are exceptions to this rule and con-
stitute essential reading in the field of crystal plasticity. This is not to say that the theory
for isotropic materials is passé. On the contrary, the difficulties encountered in recon-
ciling classical plasticity theory with modern continuum mechanics are readily resolved
in the framework of the modern theory. Accordingly we devote substantial space to the
isotropic theory in this book.
We confine attention to the purely mechanical theory because this is where the main
conceptual challenges lie. Treatments of the thermodynamical theory may be found in
the books by Epstein and El anowski and by Maugin.

1.2 Elements of continuum mechanics

For the most part, our development is based on the standard framework of continuum
mechanics as conceived by Cauchy. Thus, we do not take couple stresses or higher order
stresses into account. This is very much in accord with the vast majority of work in plas-
ticity theory. Much of the modern literature also seeks to describe length-scale effects
associated with plastic response. This is typically modeled by including gradients of plas-
tic deformation among the variables appearing in constitutive equations, while keeping
much of Cauchy’s framework intact. We will devote some effort later to a discussion of
these developments. For now, however, we shall be content with a brief survey of the
basic elements of continuum mechanics that are needed for our work. Detailed discus-
sions of everything said here may be found in the textbooks by Gurtin, Chadwick, and
Liu, for example.
Concerning notation, we adopt the standard symbols At , A−1 , A∗ , SymA, SkwA,
DevA, and JA . These are, respectively, the transpose, inverse, cofactor, symmetric part,
skew part, deviatoric part, and determinant of a second-order tensor A. If A is invertible,
then A∗ = JA A−t . We also use Sym to identify the linear space of symmetric tensors.
The tensor product of 3-vectors is indicated by interposing the symbol ⊗, i.e., a ⊗ b,
and is defined by (a ⊗ b)v = (b · v)a for any vector v. The Euclidean inner product
√ and defined by A · B = tr(AB ), where tr(·) is the trace; the
t
of tensors A,B is denoted
induced norm is |A| = A · A. For a fourth-order tensor A, the notation A[B] stands
for the second-order tensor resulting from the linear action of A on B. Its transpose
At is defined by B · At [A] = A · A[B], and A is said to possess major symmetry if
 Elements of continuum mechanics 5

At = A. If A · A[B] = At · A[B] and A · A[B] = A · A[Bt ], then A is said to possess

minor symmetry. The notation (·)A , with a bold subscript, stands for the derivative of a
function with respect to tensor A.
Suppose a body B, consisting of a fixed set of material points, occupies a config-
uration κt at time t, a region in a three-dimensional Euclidean space. The restriction
to Euclidean space is not sufficiently general to accommodate all conditions. Rather,
it reflects a prejudice derived from our terrestrial, non-relativistic, experience, which is
nevertheless sufficient to cover most problems that arise at the level of our present tech-
nological development. Let y be the position in κt , relative to a specified origin, of a
material point p ∈ B. To convey the notion that this position is occupied by p, and only
by p, we conceive of an invertible map χ from B to κt such that

y = χ(p, t). (1.5)

Rather than deal with the ethereal body B directly, to facilitate analysis we pick some
fixed region of Euclidean space, labeled κ, that stands in one-to-one relation to it. We
call this a reference configuration. For example, it is usually convenient to choose a region
that could, in principle, be occupied by the body, even if it is never actually occupied in
the course of its motion. Quite often analysts choose κ = κt0 , the actual configuration at
time t0 , which of course automatically fulfills the occupiability condition. Whatever the
choice of the reference configuration, we stipulate that there exists a one-to-one map κ
from B to κ such that

x = κ(p), (1.6)

where x is the position of p in κ relative to some fixed origin. In this way we effectively
identify p with the position x that it occupies in our chosen κ. We then have a one-to-one
relation

y = χκ (x, t), (1.7)

called the deformation of p from κ to κt , where

χκ (x, t) = χ(κ−1 (x) , t), (1.8)

in which the subscript is intended to identify our choice of κ. At the risk of being
imprecise, we usually suppress it when there is no risk of confusion, and simply write

y = χ(x, t), (1.9)

with the caveat, of course, that this is not the same function as that appearing in (1.5).
We are typically interested in deformations that are continuous and differentiable,
meaning that for material points p1 and p2 occupying positions x1 = κ(p1 ) and x2 = κ(p2 ),
6 Preliminaries

respectively, that are near to each other in κ, there exists a tensor field F(x, t), called the
deformation gradient, such that

χ(x2 , t) − χ(x1 , t) = F(x1 , t)(x2 − x1 ) + r(x2 , x1 , t), (1.10)

where

|r(x2 ,x1 , t)| = o(|x2 − x1 |), (1.11)

in which the Landau symbol o(ϵ) identifies terms that are smaller than ε for small ϵ;
that is, o(ϵ)/ϵ → 0 as ϵ → 0. It then follows that p1 and p2 are near to each other
in κt as well. Equation (1.10) defines the deformation gradient and effectively fur-
nishes the definition of differentiability in this context. In deference to this we often
write

F = ∇χ, (1.12)

to denote the gradient of χ with respect to x. The invertibility of (1.9) implies that F is
an invertible tensor. This is a consequence of the Inverse Function Theorem. See the
book by Fleming.
Unfortunately, (1.10) does not afford a useful way to compute the deformation gra-
dient in terms of the function χ(x, t). To rectify this, suppose the points p1 and p2 are
connected by a smooth curve c ⊂ κ with arclength parametrization x(s), such that
x1 = x(s1 ) and x2 = x(s2 ). Assuming again that these points are near to each other,
we then have

x2 − x1 = x′ (s1 )(s2 − s1 ) + o(s2 − s1 ), (1.13)

where |o(s2 − s1 )| = o(s2 − s1 ). Combining this with (1.10) gives

s2 −s1 { χ(x(s2 ), t)
1
− χ(x(s1 ), t)} = F(x1 , t)x′ (s1 ) + 1
s2 −s1 o(s2 − s1 ), (1.14)

and passage to the limit s2 → s1 yields

χ ′ = Fx′ (1.15)

at x = x1 , where, with t fixed,

χ ′ (s) = d
ds χ(x(s), t). (1.16)

Of course this is just the chain rule. In view of (1.9) it is meaningful to write it in the
form

dy = Fdx. (1.17)
 Elements of continuum mechanics 7

As we will see in Chapter 4, this formula affords a direct way to obtain expressions for
F when the positions y and x are specified in terms of coordinate systems. See the book
by Steigmann for some explicit examples.
From what has been said it should be evident that, at the material point p, dx(= x′ ds)
and dy(= y′ ds) are tangential to the curves c ⊂ κ and ct ⊂ κt , respectively, the latter
having the parametric representation y(s, t) = χ(x(s), t). We call c a material curve , to
convey the meaning that it is convected by the deformation to a curve ct consisting of
the same material points. Consider two material curves that intersect at x, with tangents
dx and du. These are transported to dy and dz, respectively, where dz = Fdu, with
F = F(x, t), as in (1.17). The local state of distortion of these material curves existing at
the material point p, occupying position x in κ, is characterized by

dy · dz = Fdx · Fdu = dx · Ft (Fdu) = dx · Cdu, (1.18)

where

C = Ft F (1.19)

is the right Cauchy–Green deformation tensor. Choosing the material curves to coincide,
i.e., du = dx, yields the squared stretch of a curve. Equation (1.18) then furnishes the
local angle made by the tangents to two material curves after deformation.
For our purposes it will prove convenient to work with the Lagrange strain

E = 12 (C − I), (1.20)

which is in one-to-one relation to the Cauchy–Green tensor, where I is the referential

unit tensor, defined by Iv = v for all vectors v belonging to the vector space T κ associated
with κ. The latter is often called the translation space of κ, to convey the notion that it
coincides with the set of all position differences that can be formed within it. Because
κ resides in Euclidean space by assumption, this translation space is identical to the
tangent space of the underlying (Euclidean) manifold. Similarly, we denote the translation
(tangent) space associated with κt by Tκt . Thus, Euclidean spaces are effectively flat
in the sense that they coincide with their tangent spaces. More will be said about this
in Chapter 3.
Having discussed the bare essentials of the kinematics of deformation, we move on to
the basic balance laws concerning mass and momentum.
Let ρ(y, t) be the (positive) mass density of the body in the configuration κt . The mass
of a subregion πt ⊂ κt is simply
Z Z
M(πt ) = ρdv = ρκ dV, (1.21)
πt π
8 Preliminaries

where π ⊂ κ is the image of πt under the inverse deformation, i.e., πt = χ(π, t), meaning
that the two regions are related by the deformation map and consist of the same set of
material points; and

ρκ = JF ρ, (1.22)

where JF = |det F| , is the referential mass density. This is simply the familiar change-of-
variable formula from calculus. In this book we assume κ is occupiable, so that det F > 0,
but this is by no means essential.
The principle of conservation of mass is the assertion that the mass of a fixed set
of material points remains invariant in time. Therefore, the time derivative of M(πt )
vanishes. Because the domain π is fixed for the material points that occupy πt , we can
pass the derivative through the right-most integral—assuming sufficient regularity of the
integrand—to obtain
Z
ρ̇κ dV = 0, (1.23)
π

where the superposed dot is the material derivative, the partial time derivative holding p,
and hence x, fixed. Because π is an arbitrary subvolume of κ, assuming the integrand to
be continuous we can invoke the localization theorem—basically the mean-value theorem
for integrals—to conclude that the integrand vanishes pointwise, i.e., that ρ̇κ = 0 at
every x ∈ κ. In other words, the function ρκ (x, t) = det F(x, t)ρ(χ(x, t), t), expressed as
a function of x and t, is independent of t and hence a fixed function of x. This result
does not apply in the presence of diffusion, however. In this case our reasoning must
be adjusted to account for the flux of mass through the boundary ∂πt . See the book by
Gurtin et al.
The balance of linear momentum is the assertion that the net force acting on the
material occupying πt is balanced by the rate of change of its momentum. Thus,
Z Z Z
d
tda + ρbdv = dt ρvdv, (1.24)
∂πt πt πt

where t, the traction, is the areal density of contact force, b is the body force per unit mass,
and v = ẏ = ∂ χ(x, t) is the material velocity. To reduce the right-hand side we proceed
∂t
as in the reduction of the mass conservation principle. Thus, invoking conservation of
mass in the form ρ̇κ = 0, we have
Z Z Z Z
d d d
dt ρvdv = dt ρJF vdV = dt ρκ vdV = ρκ v̇dV
πt π π π
Z Z
= ρJF v̇dV = ρv̇ dv, (1.25)
π πt
 Elements of continuum mechanics 9

which reduces (1.24) to

Z Z
tda + ρ(b − v̇)dv = 0. (1.26)
∂πt πt
Assuming the integrands to be bounded in magnitude, this may be used to establish
that t is a function of the tangent plane to ∂πt at the point y ∈ ∂πt ; equivalently, a function
of the unit normal n to ∂πt at y. See the important paper by Noll. For definiteness we
take this to be the exterior unit normal. With this result in hand we may proceed via
Cauchy’s theorem to show that the dependence is linear, and hence that there exists a
tensor field T(y, t), the Cauchy stress, such that

t = Tn. (1.27)

Substituting into (1.26) and invoking the divergence theorem, we obtain

Z
[divT + ρ(b − v̇)]dv = 0, (1.28)
πt
where divA is the vector field defined by

c · divA = div(At c), (1.29)

for any tensor field A(y, t) and any fixed vector c. Here divw is the scalar field
defined by

divw = tr(gradw), (1.30)

where tr is the trace and gradw, the gradient of a vector field w(y, t) with respect to y, is
the tensor field defined, as in (1.17), by

dw = (gradw)dy. (1.31)

Assuming the integrand in (1.28) to be a continuous function of y, we can localize

and arrive at Cauchy’s equation of motion,

divT + ρb = ρv̇, (1.32)

holding at each y ∈ κt . Generalizations to discontinuous fields will be considered in

Chapter 6.
The balance of moment of momentum is the assertion that
Z Z Z
y × tda + y × ρbdv = dt
d
y × ρvdv, (1.33)
∂πt πt πt
where y is the position field relative to a fixed origin. Invoking conservation of mass, the
traction formula (1.27), the linear momentum balance (1.32), and localizing as before,
10 Preliminaries

we arrive ultimately at the local algebraic restriction

T = Tt , (1.34)

again at every y ∈ κt . Equations (1.32) and (1.34) are often referred to as the spatial
equations of motion.

Problem 1.1 Prove (1.34) by carrying out the steps indicated.

Equivalent referential forms of the equations, due essentially to Piola, may be derived
with the aid of the Piola–Nanson formula

nda = F∗ νdA, (1.35)

connecting the oriented area measure nda on ∂πt to its counterpart νdA on ∂π. Here

F∗ = JF F−t (1.36)

is the cofactor of F. Thus, from (1.22), (1.26), and (1.27),

Z Z Z Z
ρκ (v̇ − b)dV = ρ(v̇ − b)dv = Tnda = PνdA, (1.37)
π πt ∂πt ∂π

where

P = TF∗ (1.38)

is the Piola stress. Clearly, this provides a measure of force per unit reference area,
whereas the Cauchy stress furnishes a resolution of the same force per unit area of sur-
face after deformation. Applying the divergence theorem again, this time in the reference
configuration, we have

Z
[DivP + ρκ (b − v̇)]dV = 0, (1.39)
π
 Elements of continuum mechanics 11

where Div, the divergence with respect to x, is defined, with obvious adjustments, in the
same way that div was defined. Localizing as usual, we find that (1.39) is equivalent to

DivP + ρκ b = ρκ v̇, (1.40)

holding at each x ∈ κ, this formulation having the convenient feature that the function
ρκ (x) is known a priori, whereas the symmetry condition (1.34) is equivalent to

PFt = FPt . (1.41)

To express the latter condition in a more convenient form we introduce the Piola–
Kirchhoff stress S, defined by

P = FS. (1.42)

Then PFt = FSFt and the invertibility of F implies that (1.41) is equivalent to the
symmetry

S = St . (1.43)

Before concluding these preliminaries we pause to state the mechanical energy

balance,

P(π, t) = S(π, t) + dt K(π, t),

d
(1.44)

where
Z
K(π, t) = 1
2 ρκ v · vdV (1.45)
π

is the kinetic energy of the material occupying π,

Z
S(π, t) = P · ḞdV (1.46)
π

is the stress power, in which A · B = tr(ABt ) is the inner product of tensors A and B, and
Z Z
P(π, t) = p · vdA + ρκ b · vdV (1.47)
∂π π

is the power of the forces acting on the material in π, in which

p = Pν (1.48)
12 Preliminaries

is the Piola traction, related to the Cauchy traction t by

Z Z
pdA = tda. (1.49)
∂π ∂πt

The mechanical energy balance will play a central role in our development.

Problem 1.2 Derive (1.44) from the momentum balance (1.40). Hint: Dot multiply
(1.40) by the material velocity v. Show that v · DivP = Div(Pt v) − P · ∇v, where ∇
is the gradient with respect to x, and that ∇v = Ḟ. Integrate over π ⊂ κ and invoke
the divergence theorem.

With reference to the problem, note that if we dot multiply (1.40) by an arbitrary
vector field u instead of the material velocity v, we arrive at the integral statement
Z Z Z
p · udA + ρκ (b − v̇) · udV = P · ∇udV (1.50)
∂π π π

in place of (1.44). It is usual, though no doubt unwise from the pedagogical point of view,
to call u a virtual velocity field, or worse, a virtual displacement, to emphasize the fact
that it has nothing whatever to do with the actual material velocity. This is the weak form
of the equation of motion, so named because it requires a weaker degree of regularity
than the local, or strong, form.
Although we have obtained it as a necessary condition for (1.40), it is also sufficient.
To see this we simply start with (1.50), write P · ∇u = Div(Pt u) − u · DivP, and apply
the divergence theorem, reaching
Z Z
(p − Pν) · udA + {DivP + ρκ (b − v̇)} · udV = 0. (1.51)
∂π π

As u is arbitrary, we choose

u = f(x)2 {DivP + ρκ (b − v̇)}, (1.52)

where f(x) is any function that vanishes on ∂π, thereby reducing (1.51) to
Z
2
f 2 |DivP + ρκ (b − v̇)| dV = 0, (1.53)
π

and hence requiring that (1.40) hold locally in π, leaving

Z
(p − Pν) · udA = 0 (1.54)
∂π
 References 13

as the remaining content of (1.51). Choosing u = p − PN on ∂π then yields

Z
2
|p − Pν| dA = 0, (1.55)
∂π

which in turn requires that (1.48) hold locally, at each point of ∂π.
More often this procedure is invoked with π replaced by κ. In this case position
y = χ(x, t) is typically assigned as a fixed function ϕ(x), say, on a part ∂κy of the bound-
ary ∂κ , implying that the actual velocity v vanishes there. We then stipulate that u should
also vanish on ∂κy , as a condition of a so-called kinematically admissible virtual velocity
field. The argument leading to (1.53), with π replaced by κ, remains valid, but in place
of (1.54) we now have
Z
(p − Pν) · udA = 0, (1.56)
∂κp

where ∂κp = ∂κ \ ∂κy . Choosing

u = g(x)2 (p − Pν), (1.57)

where g(x) is any function that vanishes on the curve(s) ∂(∂κp ) = ∂(∂κy ) in accordance
with kinematic admissibility, we then have
Z
2
g2 |p − Pν| dA = 0, (1.58)
∂κp

which requires that (1.48) hold pointwise on ∂κp .

References
Batchelor, G. K. (Ed.) (1958). The Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 1: Mechanics
of Solids. Cambridge University Press, Cambridge, UK.
Chadwick, P. (1976). Continuum Mechanics: Concise Theory and Problems. Dover, New York.
Epstein, M., and El anowski, M (2007). Material Inhomogeneities and Their Evolution. Springer,
Berlin.
Fleming, W. (1977). Functions of Several Variables. Springer, Berlin.
Gurtin, M. E. (1981). An Introduction to Continuum Mechanics. Academic Press, Orlando.
Gurtin, M. E., Fried, E., and Anand, L. (2010). The Mechanics and Thermodynamics of Continua.
Cambridge University Press, Cambridge, UK.
Havner, K. S. (1992). Finite Plastic Deformation of Crystalline Solids. Cambridge University Press,
Cambridge, UK.
Hill, R. (1950). The Mathematical Theory of Plasticity. Oxford University Press, Oxford.
Liu, I-Shih. (2002). Continuum Mechanics. Springer, Berlin.
Kovács, I., and Zsoldos, L. (1973). Dislocations and Plastic Deformation. Pergamon Press, Oxford.
14 Preliminaries

Maugin, G. A. (1992). The Thermomechanics of Plasticity and Fracture. Cambridge University

Press, Cambridge, UK.
Noll, W. (1974). The foundations of classical mechanics in the light of recent advances in
continuum mechanics. Reprinted in Truesdell, C. (Ed.), The Foundations of Mechanics and
Thermodynamics, pp. 32–47 Springer, Berlin.
Steigmann, D. J. (2017). Finite Elasticity Theory. Oxford University Press, Oxford.
Teodosiu, C. (Ed) (1997). Large Plastic Deformation of Crystalline Aggregates. CISM Courses and
Lectures, No. 376. Springer, Vienna.
2
Brief resumé of nonlinear elasticity
theory

Familiarity with the basic elements of nonlinear elasticity theory is essential to a proper
understanding of virtually the entire range of topics comprising solid mechanics in gen-
eral, and plasticity theory in particular. Accordingly we devote the present chapter to
a brief survey of those aspects of nonlinear elasticity that will prove central to our later
work.

2.1 Stress and strain energy

Following Noll’s landmark 1958 paper, we define elasticity to mean that the value of the
Cauchy stress T(p, t) existing at the material point p ∈ B at time t is determined by the
present value of the deformation function χ(x′ , t) for x′ ∈ Nκ (x), an arbitrary, and hence
arbitrarily small, neighborhood of the place x occupied by p in reference configuration
κ. Our assumption of differentiability of the deformation—see (1.10)—implies that the
deformations influencing this stress are approximated by

χ(x′ , t) = χ(x, t) + F(x, t)(x′ − x) + o(|x′ − x|), (2.1)

and are therefore determined, at linear-order accuracy, by χ(x, t) and F(x, t).
The requirement of frame invariance satisfied by all sensible constitutive functions
implies that the constitutive equation giving the stress is, among other things, transla-
tion invariant in the sense that it remains invariant under constant time translations and
perturbations of the position y currently occupied by a material point. It is therefore not
explicitly dependent on t or χ(x, t). Accordingly, at leading order the stress is determined
by F(x, t). Termination at this order leads to Noll’s simple material model of elasticity,
according to which

T(p, t) = Tκ (F(x, t); x) (2.2)

for some constitutive function Tκ , the subscript identifying the reference configuration
relative to which the deformation gradient is computed. It is best to make this explicit to

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0002
16 Brief resumé of nonlinear elasticity theory

avoid confusion when choosing alternative reference configurations, as we shall do later.

Of course it is possible to keep further terms in the expansion (2.1) and to contemplate
their influence on constitutive equations for so-called materials of higher grade. While
these are important and useful, they are of limited relevance to the current state of the
art in plasticity theory. Given the constitutive function Tκ for the Cauchy stress, that for
the Piola stress, for example, relative to the same reference configuration, follows easily
from (1.38):

P(p, t) = Pκ (F(x, t); x) = Tκ (F(x, t); x)F∗ (x, t). (2.3)

Frame invariance imposes further restrictions on these constitutive functions, but to

save time we defer these and proceed instead to a further specialization of the purely
mechanical theory. This is the so-called work inequality, which posits that it is necessary
to perform non-negative work on a sample of material to cause it to undergo a hypothet-
ical cyclic process; that is, a process in which the deformation and velocity fields at the
start and end of a process, occurring in a time interval [t1 , t2 ], coincide at every material
point, i.e.,

χ(x, t1 ) = χ(x, t2 ) and χ̇(x, t1 ) = χ̇(x, t2 ). (2.4)

Taking gradients, these imply that

F(x, t1 ) = F(x, t2 ) and Ḟ(x, t1 ) = Ḟ(x, t2 ). (2.5)

This assumption is related to the hypothesis known more popularly as the non-
existence of perpetual motion machines. However, given that all material points of the
sample are involved, the creation of a cyclic process is no small feat from the experimen-
tal point of view. Moreover, while this assumption has a thermodynamical flavor, it is
not in fact a consequence of any principle of thermodynamics. It is nevertheless realistic,
and in accord with everyday experience.
Proceeding, we impose the work inequality on an arbitrary subvolume π ⊂ κ, invoke
(1.44), and, noting that the kinetic energies at the start and end of the cycle coincide,
conclude that
Z t2
S(π, t)dt ≥ 0, (2.6)
t1

where S is the stress power defined in (1.46). Assuming the integrand therein to be
continuous in [t1 , t2 ] × π, Fubini’s theorem (see Fleming’s book) ensures that we can
interchange the order of integration, and hence that
Z Z t2
P · Ḟdt ≥ 0 for all π ⊂ κ. (2.7)
π t1
 Stress and strain energy 17

Localizing as usual, we have

Z t2
P · Ḟdt ≥ 0, at each x ∈ κ, (2.8)
t1

and hence, in the case of elasticity,

I
Pκ (F) · dF ≥ 0, (2.9)

in which the integral is taken round a smooth curve in F-space, in accordance with the
definition of a cyclic process, and the parametric dependence on x, having no bear-
ing on the discussion, has been suppressed. By considering the reversal of this cyclic
process—itself cyclic—it is possible to show—see the books by Gurtin and Steigmann,
for example—that this statement is satisfied as an equality for all smooth closed curves.
This, in turn, is necessary and sufficient for the existence of a strain-energy function
Ψ(F; x), say, such that

P = ΨF , (2.10)

the gradient of Ψ with respect to F. Given the function Ψ, this gradient is easily computed
via the chain rule by writing dΨ as a linear form in dF and reading off the coefficient
tensor, i.e.,

dΨ = ΨF · dF. (2.11)

We thus have

Pκ (F; x) = ΨF (F; x), (2.12)

and hence the rather remarkable conclusion that the stress is determined by a single
scalar-valued function. We then refer, somewhat dramatically, to a hyperelastic material.
The strain-energy function has as arguments variables defined on κ, which should there-
fore be appended as a subscript. In an effort to minimize clutter we normally refrain from
doing this when the context is clear.
On combining (2.2), (2.3), and (2.10), the symmetry condition (1.34) satisfied by
the Cauchy stress may be expressed in terms of the strain-energy function as

(ΨF )Ft = F(ΨF )t , (2.13)

which in turn is equivalent to the statement

(ΨF )Ft · Ω = 0 (2.14)

for arbitrary skew Ω; that is, for all Ω such that Ωt = −Ω. The identity

AB · D = A · DBt (2.15)
18 Brief resumé of nonlinear elasticity theory

allows us to rewrite this as

ΨF · ΩF = 0. (2.16)

Our intention is to characterize all functions Ψ that satisfy this requirement. If we

succeed, then we can replace the symmetry of the Cauchy stress once and for all by the
general form of Ψ to be derived. Before undertaking this task, however, a brief digression
is in order.
Consider the initial-value problem

Q̇ = ΩQ; Q(0) = i, (2.17)

where Q(u) is a one-parameter family of tensors, the superposed dot stands for the
derivative with respect to u, Ω is a fixed, but arbitrary, skew tensor, and i is the spa-
tial identity tensor. We want to show that Q(u) is a rotation tensor for all values of the
parameter u.
To this end we define

Z(u) = Q(u)Q(u)t , (2.18)

and find, with (2.17), that this satisfies the initial-value problem

Ż = ΩZ − ZΩ; Z(0) = i. (2.19)

Clearly Z(u) = i is a solution, and the uniqueness theorem for solutions to ordinary
differential equations implies that there is no other. Thus, Q(u) is orthogonal and hence
invertible. To establish that it is a rotation, we need to show that its determinant equals
unity. This follows from the the fact that the cofactor of a tensor is the derivative of its
determinant, i.e.,

J̇Q = Q∗ · Q̇ = JQ Q−t · Q̇ = JQ tr(Q̇Q−1 ) = JQ trΩ, (2.20)

which vanishes because Ω is skew, implying that JQ(u) = JQ(0) = 1, as claimed.

Returning to the task at hand, consider the one-parameter family

F(u) = Q(u)F0 ; F0 = F(0), (2.21)

with Q(u) as in the foregoing. Because this pertains to a single (but arbitrary) material
point, the fact that F is the gradient of a deformation does not impose any restriction on
its values beyond JF0 > 0, assuming the reference configuration to be occupiable. We
then have

Ḟ = Q̇F0 = Q̇Qt F = ΩF. (2.22)

 Stress and strain energy 19

Accordingly, for this choice (2.16) reduces to

0 = ΨF · Ḟ = Ψ̇, (2.23)

or simply that Ψ(F(u)) = Ψ(F0 ), where we have again suppressed the passive argument
x. Dropping the subscript and reinstating this argument, we conclude that the symmetry
of the Cauchy stress implies the invariance of the energy under superposed rotations,
i.e.,

Ψ(QF; x) = Ψ(F; x) (2.24)

for any rotation Q. For Q spatially uniform, this is precisely the condition of frame invari-
ance of the energy, which implies that it is insensitive to arbitrary rigid-body motions

χ(x, t) → Q(t)χ(x, t) + c(t) (2.25)

superposed on the deformation χ(x, t), where c is an arbitrary vector. See the treatment
of frame invariance in the important paper by Murdoch and its adaptation to elasticity
in the book by Steigmann.
We have shown that (2.24) is necessary for (1.34). It is also sufficient. To demonstrate
this, recall the definition (1.19) of the Cauchy–Green deformation tensor. Consider two
−1
deformation gradients F and F̃ and let Q̃ = F̃F . Let C and C̃ respectively be the
Cauchy–Green tensors formed from F and F̃.
t
Problem 2.1 If C̃ = C, show that Q̃ Q̃ = i, so that Q̃ is orthogonal. Conversely, show
that if Q̃ is orthogonal, then C̃ = C.

Suppose that Ψ(F̃; x) = Ψ(F; x) whenever F̃ = Q̃F, with Q̃ a rotation and hence
orthogonal. This is simply a restatement of (2.24). Then, from the problem, it follows
that Ψ(F̃; x) = Ψ(F; x) whenever C̃ = C. This means that Ψ depends on F through the
induced Cauchy–Green tensor, and hence that

Ψ(F; x) = Ψ̂(C; x) (2.26)

for some function Ψ̂. Equivalently,

Ψ(F; x) = U(E; x), (2.27)

where, from (1.20), U(E; x) = Ψ̂(2E + I; x).

20 Brief resumé of nonlinear elasticity theory

Problem 2.2 We have shown that (2.24) implies (2.26). Show that (2.26) implies
(2.24), and hence that the two statements are equivalent.

Consider a one-parameter F(u) of deformation gradients. This induces a one-

parameter family of strains, given by

E(u) = 21 [F(u)t F(u) − I]. (2.28)

Substituting these into (2.27) and differentiating, we have

ΨF · Ḟ = UE · Ė, (2.29)

where UE is the gradient of the function U(·; x), and

t
Ė = 12 (Ḟ F + Ft Ḟ) = Sym(Ft Ḟ), (2.30)

in which Sym(·) is the symmetric part of its tensor argument. Because UE occurs in
an inner product with a symmetric tensor, and because the set of symmetric tensors
constitutes a linear space (i.e., a 6-dimensional vector space), we may assume, without
loss of generality, that UE is a symmetric tensor. For, the skew part, if non-zero, would
be annihilated by the inner product. We thus have

ΨF · Ḟ = UE · Sym(Ft Ḟ) = UE · Ft Ḟ = F(UE ) · Ḟ, (2.31)

after making use of the identity A · BD = Bt A · D.

Problem 2.3 Prove this identity and the identity (2.15).

We thus conclude that

ΨF = F(UE ). (2.32)

Because the left-hand side is just the Piola stress, and in view of the definition (1.42), it
follows that the Piola–Kirchhoff stress is given by

S = UE . (2.33)

This is automatically symmetric, of course, and (1.38) then implies that the Cauchy
stress is too. This concludes our proof that (2.24) is sufficient for the symmetry of the
Cauchy stress. With the previous proof of necessity, and the fact that (2.24) is necessary
and sufficient for (2.27), we conclude that (2.27) is equivalent to the symmetry of the
Cauchy stress for elastic materials that satisfy the work inequality.
 Conservative problems and potential energy 21

Problem 2.4 Show that Pκ (QF; x) = QPκ (F; x) and Tκ (QF; x) = QTκ (F; x)Qt .

2.2 Conservative problems and potential energy

Hyperelasticity affords a rich framework for the analysis of motion, which will be familiar
to those versed in analytical dynamics. From (1.46) and (2.10), it yields the stress power
in π as a time derivative,

S(π, t) = dt U(π, t),

d
(2.34)

where
Z
U(π, t) = ΨdV (2.35)
π
is the total strain energy in π. We may write this alternatively as
Z
U= ψdv, (2.36)
πt
where ψ(F; x), given by

Ψ = JF ψ, (2.37)

is the strain energy per unit current volume.

If the forces acting on the material in κ are such that the power they generate is
derivable from a load potential L, i.e., if there exists L(κ, t) such that

P(κ, t) = dt L(κ, t),

d
(2.38)

then the mechanical energy balance (1.44), applied with π = κ, becomes

d
dt E(κ, t) = 0, (2.39)

where

E(κ, t) = K(κ, t) + E[χ], (2.40)

in which

E = U(κ, t) − L(κ, t) (2.41)

is the total mechanical energy. Here E is the total potential energy of the configuration
χ(κ, t), i.e., the energy of elasticity and the forces together. Thus, the total energy is
conserved, and such loads are said to be conservative. Of course elasticity theory, and
hyperelasticity in particular, is perfectly sensible for non-conservative loads too, but these
22 Brief resumé of nonlinear elasticity theory

do not give rise to any conservation law of this kind. Nor, in fact, does the response of
any real material, even if the loading can be associated with a load potential. In reality,
(2.39) is replaced by the inequality

d
dt E(κ, t) ≤ 0, (2.42)

due to effects such as heat conduction, viscosity, friction, and indeed plasticity, as we
shall see later, that dissipate energy and which are not taken into account in pure elasticity
theory.
Nevertheless, the notion of energy is extremely useful if interpreted appropriately.
For example, consider a motion χ(x, t) and an instant t 0 such that

χ(x, t0 ) = χ0 (x) and χ̇(x, t)|t0 = 0. (2.43)

Suppose the motion to be such that

lim χ(x, t) = χ∞ (x) and lim χ̇(x, t) = 0. (2.44)

t→∞ t→∞

This limiting state, if it exists, is said to be asymptotically stable relative to the state existing
at time t0 . It then follows from (2.40) and (2.42) that asymptotically stable deformations
minimize the potential energy, i.e.,

E[χ∞ ] ≤ E[χ0 ]. (2.45)

Thus, the problem of finding stable deformations is reduced to the problem of minimiz-
ing the potential energy, presuming, of course, that such an energy exists, i.e., that the
material is hyperelastic and the loading is conservative.

Problem 2.5 (a) Consider the case of dead loading in which the Piola traction and body
force are assigned as fixed functions of x, independent of time, on ∂κp = ∂κ \ ∂κy
and in κ, respectively. Show that this loading is conservative, and that the potential
energy is
Z Z
E[χ] = (Ψ − ρκ b · χ)dV − p · χdA,
κ ∂κp

apart from a constant. The constant, which has not been made explicit, is
unimportant. Why is that?
(b) State conditions under which a pressure load t = −pn is conservative, where
t is the Cauchy traction, p is the pressure, and n is the exterior unit normal to the
boundary ∂κt . Obtain expressions for the associated load potentials.
 Conservative problems and potential energy 23

Problem 2.6 Consider a viscoelastic material model in which the Piola stress is
given by

P = ΨF + μFĖ,

where μ is the viscosity of the material and Ė is the material derivative of the
Lagrange strain. This model is too simple to furnish a quantitative description of the
actual viscoelastic response of most materials but nevertheless exhibits the correct
qualitative behavior.
Define the dissipation D as the difference between the power of the applied loads
and the rate of change of the sum of the strain and kinetic energies, i.e.,

D = P− dtd (U + K).

Thus, from (1.44) and (2.34), D vanishes in the case of purely elastic response
(μ = 0).
(a) Show that D ≥ 0 if the viscosity is positive and that D then vanishes if and
only if Ė = 0 at all material points.
(b) In the case of conservative loading show that the total energy is dissipated,
i.e.,
d
dt E ≤ 0,

if the viscosity is positive.

Let χ(x; u) be a one-parameter family of deformations, and denote χ(x; 0) simply by

χ(x). We define kinematic admissibility of this family to mean that χ(x; u) is as smooth
as required by the analysis to follow and that it equals the fixed, u-independent function
ϕ(x), say, on ∂κy . We are interested in characterizing energy-minimizing deformations
χ(x) ; namely, those functions such that

E[χ(x)] ≤ E[χ(x; u)]. (2.46)

To this end we define

f(u) = E[χ(x; u)]. (2.47)

Then, for (2.46) it is necessary that

f˙ = 0 and f¨≥ 0, (2.48)

where

f˙= f ′ (u)|u=0 and f¨= f ′′ (u)|u=0 (2.49)

are the first and second variations of the potential energy, respectively.
24 Brief resumé of nonlinear elasticity theory

In the case of dead loading, for example,

Z Z
f(u) = {Ψ(F(x; u); x) − ρκ b · χ(x; u)}dV − p · χ(x; u)dA, (2.50)
κ ∂κp

where F(x; u) = ∇χ(x; u), and the first and second variations are
Z Z
f˙= (P · ∇u − ρκ b · u)dV − p · udA (2.51)
κ ∂κp

and
Z
f¨ = M[∇u] · ∇udV
κ
Z Z
+ (P · ∇v − ρκ b · v)dV − p · vdA, (2.52)
κ ∂κp

respectively, where
2
u(x) = ∂ χ(x; u)|u = 0 and v(x) = ∂ 2 χ(x; u)|u = 0 (2.53)
∂u ∂u
are the virtual velocity and acceleration. These vanish identically on ∂κy . Further,

P = ΨF (∇χ(x); x) (2.54)

is the Piola stress associated with the minimizing deformation χ(x), and

M = ΨFF (∇χ(x); x) (2.55)

is the fourth-order tensor representing the second derivative of the strain-energy func-
tion, again evaluated at the minimizing deformation. This presumes, of course, that the
strain energy is twice differentiable with respect to the deformation gradient, this in turn
implying the major symmetry condition

M = Mt , (2.56)

where the transpose Mt is defined by

A · Mt [B] = B · M[A] (2.57)

for all second-order tensors A, B ; and, here and in (2.52), the square bracket notation
is used to denote the second-order tensor resulting from the linear action of a fourth-
order tensor on a second-order tensor. In terms of Cartesian components, for example,
(M[A])ij = Mijkl Akl , in which a double sum on k and l from one to three is implied.
 The Legendre–Hadamard inequality 25

Recalling that u vanishes on ∂κp , we use the divergence theorem in (2.51) to reduce
(2.48)1 , with (2.51), to
Z Z
(Pν − p) · udA − (DivP + ρκ b) · udV = 0, (2.58)
∂κp κ
and proceed as in Chapter 1 to conclude that

DivP + ρκ b = 0 in κ, and p = Pν on ∂κp , (2.59)

and hence that energy minimizers are equilibrium deformations. Further, as u is an arbi-
trary vector field that vanishes on ∂κy , and as v is such a field, the second line in (2.52)
vanishes by (2.58), and (2.48)2 reduces to
Z
M(F(x); x)[∇u] · ∇udV ≥ 0, (2.60)
κ
for all such u.

2.3 The Legendre–Hadamard inequality

We proceed to show that the deformation χ(x), with gradient F(x) = ∇χ, must then be
such that the Legendre–Hadamard inequality

a ⊗ n · M(F; x)[a ⊗ n] ≥ 0 (2.61)

is satisfied at every point x ∈ κ and for all vectors a and n.

Proof. Consider

u(x) = ϵξ(w) with w = ϵ−1 (x − x0 ), (2.62)

where x0 is an interior point of κ, ϵ is a positive constant, and ξ is compactly sup-

ported in a region D, the image of a strictly interior neighborhood κ′ ⊂ κ of x0
under the map w(·). Accordingly, u vanishes on all of ∂κ, and on ∂κy in particular,
and is therefore admissible. For this choice (2.60) reduces, after dividing by ϵ3 and
passing to the limit ϵ → 0, to
Z
∇ξ · A[∇ξ]dV ≥ 0, (2.63)
D

where A = M|x0 = At . Here and henceforth ∇ is the gradient with respect to w.

We extend ξ to complex-valued vector fields as

ξ = ξ1 + iξ2 , (2.64)
26 Brief resumé of nonlinear elasticity theory

where ξ1,2 (w) are real-valued, and use these to derive

∇ξ · A[∇ξ̄] = ∇ξ1 · A[∇ξ1 ] + ∇ξ2 · A[∇ξ2 ] + i(. . .), (2.65)

in which an overbar denotes complex conjugate.

Problem 2.7 Fill in the steps leading from (2.62) to (2.63). Show that the imaginary
part (. . .) in (2.65) vanishes.

Thus, if (2.63) holds for real-valued ξ, then

Z
∇ξ · A[∇ξ̄]dV ≥ 0 (2.66)
D

for complex-valued ξ.
Consider

ξ(w) = a exp(ikn · w)f(w), (2.67)

where a and n are real fixed vectors, k is a non-zero real number, and f is a real-
valued differentiable function compactly supported in D. This yields

∇ξ = exp(ikn · w)(ikfa ⊗ n + a ⊗ ∇f). (2.68)

Substitution into (2.66) and division by k2 results in

Z Z
0 ≤ a ⊗ n · A[a ⊗ n] f 2 dV + k−2 a ⊗ ∇f · A[a ⊗ ∇f]dV. (2.69)
D D

Finally, in the limit k → ∞ we obtain

a ⊗ n · A[a ⊗ n] ≥ 0, (2.70)

which is just (2.61) on account of the arbitrariness of x0 . We have thus shown that
(2.61) is necessary for (2.60). The converse is not true, however, except in special cir-
cumstances. Our proof of the Legendre–Hadamard inequality is patterned after that
found in the book by Giaquinta and Hildebrandt, which contains a wealth of valuable
information about the calculus of variations.
Thus, if χ(x) is an energy minimizer, then it necessarily satisfies (2.61), at every point
x ∈ κ. If there is any point where (2.61) is violated, for some a ⊗ n, then χ(x) is not an
energy minimizer and is therefore unstable. Accordingly, there is considerable interest
among elasticians in establishing conditions on the function Ψ which are such as to
ensure that (2.61) is satisfied. See the books by Antman, Ciarlet, and Steigmann for
discussions of those that have been proposed and studied. For our purposes, however,
simpler restrictions will prove to be more relevant.
 The Legendre–Hadamard inequality 27

Consider the relation (1.42) between the Piola and Piola–Kirchhoff stresses. For
hyperelastic materials this is equivalent to (2.32). We evaluate this on the one-parameter
family χ(x; u) and differentiate, obtaining

ΨFF [Ḟ] = ḞS + FUEE [Ė], (2.71)

where Ė is given by (2.30). Here UEE is the fourth-order tensor representing the second
derivative of U with respect to the strain E. This possesses major symmetry, as in (2.56).
However, because E is symmetric, it also possesses the additional minor symmetries

A · UEE [B] = At · UEE [B] = A · UEE [Bt ]. (2.72)

Accordingly,

UEE [Ė] = UEE [Sym(Ft Ḟ)] = UEE [Ft Ḟ]; (2.73)

that is, the skew part of Ft Ḟ is annihilated when operated on by UEE . Because Ḟ is
arbitrary, we may then use (2.71), with (2.55), to conclude that

M[A] = AS + FUEE [Ft A], for all A. (2.74)

Then,

M[a ⊗ n] = a ⊗ Sn + FUEE [Ft a ⊗ n], (2.75)

and the Legendre–Hadamard inequality (2.61) reduces, with the aid of an identity
proved in Problem 2.3, to
2
|a| n · Sn + Ft a ⊗ n · UEE [Ft a ⊗ n] ≥ 0. (2.76)

Suppose the reference configuration itself to be both stable and stress free. We will
see later that plasticity generally precludes the latter possibility, but it is nevertheless a
basic premise of essentially all work on the subject of linear elasticity in which plasticity
plays no role. In this case the classical elastic modulus tensor,

C(x) = UEE|E=0 , (2.77)

is such that

b ⊗ n · C[b ⊗ n] ≥ 0, (2.78)

where a = 1b in which 1 is the so-called shifter. This is like the identity, but not quite.
Instead of mapping vectors in T κ or Tκt to themselves, it simply converts vectors in the
former into vectors in the latter. That is, it converts vectors belonging to T κ into vectors
28 Brief resumé of nonlinear elasticity theory

belonging to Tκt , without otherwise altering them. Accordingly, F(x, t) = 1 when the
body is undeformed, i.e., when κt momentarily coincides with κ. Because C = I in this
case, it follows that 1t 1 = I, and hence that 1 is orthogonal. In fact, it is a rotation, since
it is a possible value of F; accordingly, it has a positive determinant, assuming, as we
do, that κ is occupiable. Moreover, because F is the gradient of a field defined over the
body, the rotation 1 is necessarily uniformly distributed and thus independent of x. See
the elegant proof given in Gurtin’s book.
The notion of shifter was once more widely embraced but has since fallen out of favor.
This has had the unfortunate and unintended consequence that fundamental deficiencies
have been mistakenly attributed to linear elasticity theory, most notably the allegation that
it fails to satisfy the requirement of frame invariance. This incorrect view can be traced
to a misinterpretation of the shifter as a conventional identity. One wonders how teachers
of the subject could sleep at night, with the unquestioned success of linear elasticity as a
predictive theory on the one hand, and its supposed failure to be frame invariant—and
thus wrong—on the other. Unfortunately, typical pedagogical practice is to sweep this
puzzle under the rug by simply omitting any mention of it.
We shall have occasion to make use of the shifter concept in due course. In the mean-
time, the interested reader may find excellent treatments of it in the article by Ericksen,
published as an appendix to the authoritative treatise by Truesdell and Toupin, and in
the book by Marsden and Hughes.
Returning to the matter at hand, it is conventional, and also realistic for essentially all
solids of interest, to ensure that (2.78) is satisfied by requiring C to be such that

A · C[A] > 0 for all non-zero symmetric A (2.79)

that map T κ to itself. This implies that A · C[A] ≥ 0 for all A, symmetric or not (why?),
and this in turn guarantees that (2.60) is satisfied when the body is undeformed, i.e.,
Z Z
∇u · M(1; x)[∇u]dV = 1t ∇u · C(x)[1t ∇u]dV ≥ 0. (2.80)
κ κ
If |E| is small, and if the stress vanishes when E vanishes, then the Piola–Kirchhoff
stress is given to leading order by the classical linear stress–strain relation

S = C[E]. (2.81)

Inequality (2.79) is the statement that C is positive definite, and hence invertible, on the
linear space of symmetric tensors. Then (2.81) is invertible, i.e.,

E = L[S], (2.82)

where L = C −1 , not to be confused with the load potential, is the fourth-order elastic
compliance tensor. This possesses major symmetry and the two minor symmetries and is
also positive definite on the space of symmetric tensors.

Problem 2.8 Prove these statements.

 Material symmetry 29

2.4 Material symmetry

The concept of material symmetry plays a crucial role in elasticity theory and, as we
shall see later, in plasticity theory too. This has to do with the idea that there may be
certain non-trivial transformations of a neighborhood Nκ (p) of a material point p that
are undetectable by experiment. These are called material symmetry transformations.
They are of fundamental importance to an understanding of material response.
To set the stage for this discussion, consider a change of reference configuration.
Instead of using κ as reference, let us use the fixed configuration μ, say. As usual the
position of a material point p in κ is denoted by x; let u be its position in μ. We suppose,
as usual, that both configurations stand in one-to-one correspondence with the body B,
and hence that they are in such correspondence with each other as well. Thus, there
exists an invertible map λ, say, such that

u = λ(x), (2.83)

which, if smooth, has an invertible gradient

R = ∇λ. (2.84)

To avoid ambiguity we must reinstate subscripts on constitutive functions that involve

the deformation gradient, as these are computed on the basis of a specified reference
configuration. Thus, for example, the constitutive function giving the value of the strain
energy per unit volume of κt , based on κ as reference, is ψκ (Fκ ; x), while that giving
the same value, but based on μ, is ψμ (Fμ ; u). The actual configuration κt has not been
affected. Thus,

Fκ dx = dy = Fμ du = Fμ Rdx, (2.85)

and therefore

Fκ = Fμ R. (2.86)

The energy content has not been affected either. Accordingly,

ψκ (Fκ ; x) = ψμ (Fμ ; u), (2.87)

where

ψμ (Fμ ; u) = ψκ (Fμ R(λ−1 (u)); λ−1 (u)). (2.88)

This indicates how the constitutive function changes when we choose a different refer-
ence configuration. Note that the definition of a hyperelastic material, while involving
a reference configuration, is independent of the particular reference configuration
adopted.
30 Brief resumé of nonlinear elasticity theory

Having deduced the formula for the change of constitutive function attending a
change of reference, consider next a local change of reference in which an open neigh-
borhood Nκ (p) is transformed into another open neighborhood Nμ (p) without affecting
the remainder of the body. We can think of surgically removing Nκ (p), transforming it
to Nμ (p), and reinserting it in the body. Taking these neighborhoods to be open means
we don’t need to worry about what happens at their boundaries in this thought experi-
ment, as there are none. Again let λ be the smooth invertible map from the former to the
latter; that is, the position u of a material point in Nμ (p) is given by (2.83), where x is
the position of the same point in Nκ (p). We arrange λ such that it preserves the location
of p, so that if x0 is the position of p in Nκ (p), then it is also its position in Nμ (p), i.e.,
x0 = λ(x0 ). Then (2.87), evaluated at p, yields

ψκ (Fκ ; x0 ) = ψμ (Fμ ; x0 ), where Fκ = Fμ R0 with R0 = ∇λ|x0 . (2.89)

We say that λ generates a symmetry transformation at the point p if it is such that

the mechanical responses of Nμ (p) and Nκ (p) to a given deformation coincide at p;
that is, if

ψκ (F; x0 ) = ψμ (F; x0 ) (2.90)

for any possible deformation gradient F. This means that if Nκ (p) is occupiable, and
hence if JF > 0, then we should arrange λ such that Nμ (p) is too; that is, such that
JR > 0. The italics are used here to emphasize the fact that this discussion pertains to a
single point p. Combining (2.89) and (2.90), we arrive at the restriction

ψκ (F; x) = ψκ (FR; x) (2.91)

on the single function ψκ , in which the subscript 0 has been dropped to emphasize the
fact that p, though fixed, is an arbitrary point of the body.
Let gκ(p) be the set of tensors such that (2.91) is satisfied:

gκ(p) = {R(x) | ψκ (F; x) = ψκ (FR(x); x)}. (2.92)

This set again pertains to the particular point p. Accordingly, the notion of material
symmetry is local in the sense that the set of R’s satisfying (2.91), for a given F, generally
varies from one material point to another. For this reason it is necessary to identify the
material point p to which this set pertains.
Before proceeding we address a subtle point that often leads to confusion. In (2.92)
R(x) need not be a gradient field; that is, it should not be interpreted as the gradient
of a single map defined over the entire body. This is because the map λ used in the
definition of material symmetry has been defined on a neighborhood of p. The maps
themselves may vary from one neighborhood to another and hence from one material
point to another. In this way we allow symmetry groups at different material points to
be unrelated, and thus accommodate non-uniform composites, for example, in which
 Material symmetry 31

the materials constituting a body may have entirely different properties. As we shall see,
this issue has major consequences for plasticity theory in particular.
It is easy to show that the set defined by (2.92) is a group (see Steigmann, for example).
This means, among other things, that the product of two members of gκ(p) is also a
member. It then follows that Rn is a member for any positive integer n, i.e., ψκ (F; x) =
ψκ (FRn ; x). Noting that det(FRn ) = JF det(Rn ) = JF JnR , we conclude that the energy
remains unaffected by unbounded dilation (JR > 1 and n → ∞) or compaction (JR < 1
and n → ∞). This plainly nonsensical conclusion can only be avoided by imposing
JR = 1. We thus require that gκ(p) be contained in the unimodular group u, i.e.,

gκ(p) ⊂ u = {R | JR = 1}. (2.93)

Following Noll, we define our elastic material point p to be a solid point if there exists
a local configuration κ∗ (p) ; that is, a neighborhood of p, such that

gκ∗ (p) ⊂ Orth+ , (2.94)

the group of proper-orthogonal tensors, or rotations, defined by

Orth+ = {R ∈ u | Rt R = I}. (2.95)

Note that κ∗ (p) need have no relation to the reference configuration κ used for the pur-
poses of analysis. Rather, the definition is motivated by the fact that, in the case of
crystalline solids, for example, there exists an undistorted configuration of the crystal
which is such that a unit cell of the lattice is mapped to itself by certain rotations char-
acterizing its symmetry. In the general case the same definition may be motivated by
the observation that a change of shape, i.e., a strain induced by the map λ, is mechan-
ically detectable in solids, whereas symmetry transformations should be undetectable.
Accordingly, the Cauchy–Green deformation tensor induced by λ, namely Rt R, should
be fixed at the value I. Isotropic solids are those for which there exists κ∗ (p) such that
gκ∗ (p) = Orth+ , meaning that the constitutive function relative to κ∗ (p) is insensitive to
any rotation preceding the application of the deformation.
It stands to reason that the symmetry group relative to a distorted configuration does
not generally belong to Orth+ . A simple illustration is afforded by the unit cell of a cubic
crystal. The cell is in the shape of a cube when the lattice is undistorted, and is then
mapped to itself by certain discrete rotations that do not affect its mechanical response.
This is no longer the case if the lattice is first sheared, for example. Rather, the symmetry
group relative to the sheared configuration is given by Noll’s rule.
To derive this rule consider two references κ1 and κ2 , and let π be a smooth invertible
map, with gradient K, from the first to the second. We regard these references as local
neighborhoods of the material point p in two fixed global configurations. Let p have
position x1 in κ1 and x2 in κ2 , so that x2 = π(x1 ). Then from the change-of-reference
32 Brief resumé of nonlinear elasticity theory

formula (2.87) we have ψκ2 (F; x2 ) = ψκ1 (FK(x1 ); x1 ); equivalently,

ψκ1 (F;π−1 (x2 )) = ψκ2 (FK−1 ; x2 ). (2.96)

Suppose that R ∈ gκ1 (p) , i.e.,

ψκ1 (F;π−1 (x2 )) = ψκ1 (FR;π−1 (x2 )), (2.97)

where, from (2.96), ψκ1 (FR;π−1 (x2 )) = ψκ2 (FRK−1 ; x2 ). Then (2.96) and (2.97)
combine to yield

ψκ2 (FRK−1 ; x2 ) = ψκ1 (F;π−1 (x2 )) = ψκ2 (FK−1 ; x2 ), (2.98)

which is the same as

ψκ2 (F̂; x2 ) = ψκ2 (F̂KRK−1 ; x2 ), where F̂ = FK−1 . (2.99)

We thus arrive at Noll’s rule:

gκ2 (p) = {KRK−1 | R ∈ gκ1 (p) }. (2.100)

Note once again that this discussion pertains to a single point p. This means that
K is not a gradient field, but is instead specific to p. Again this is because the map π
underpinning this discussion is defined on a neighborhood of p, and the maps themselves
may thus vary from one material point to another, each with its own neighborhood. In
other words, (2.100) is meaningful whether or not a single global map π exists for the
entire body such that K = ∇π|x1 . Note also that gκ2 (p) ⊂ u, but if gκ1 (p) is contained in
Orth+ then gκ2 (p) is not, unless K is suitably restricted.

Problem 2.9 Suppose an elastic material is isotropic, at a particular material point

p, relative to reference configuration κ1 . Thus, gκ1 (p) = Orth+ . Consider a map
from configuration κ1 to another configuration κ2 given by π(x) = λk(k · x) +
λ−1/2 [x − k(k · x)] for all x ∈ κ1 , where λ is a positive constant and k is a constant
unit vector.
(a) Consider a rotation R ∈ gκ1 (p) . What is the corresponding element of gκ2 (p) ? Is
the material isotropic relative to configuration κ2 ?
(b) Suppose the material is transversely isotropic relative to κ1 , i.e., gκ1 (p) = {R ∈
Orth+ and Re = e} for some unit vector e. The response of the material is then
insensitive to all rotations about the axis e. Show that the material is also trans-
versely isotropic relative to κ2 if e = k. (Hint: Use the Rodrigues representation
formula for rotations, discussed, for example, in the book by Chadwick.)
 References 33

References
Antman, S. S. (2005). Nonlinear Problems of Elasticity. Springer, Berlin.
Chadwick, P. (1976). Continuum Mechanics: Concise Theory and Problems. Dover, New York.
Ciarlet, P .G. (1988). Mathematical Elasticity, Vol 1: Three Dimensional Elasticity. North-Holland,
Amsterdam.
Ericksen, J. L. (1960). Tensor fields, appendix to: Truesdell, C., and Toupin, R. A., The classical
field theories, in Flügge, S. (Ed.), Handbuch der Physik, Vol. III/1, pp. 794–858. Springer,
Berlin.
Fleming, W. (1977). Functions of Several Variables. Springer, Berlin.
Giaquinta, M., and Hildebrandt, S. (2004). Calculus of Variations I. Springer, Berlin.
Gurtin, M. E. (1981). An Introduction to Continuum Mechanics. Academic Press, Orlando.
Marsden, J. E., and Hughes, T. J. R. (1994). Mathematical Foundations of Elasticity. Dover, New
York.
Murdoch. A. I. (2003). Objectivity in classical continuum physics: A rationale for discarding the
“principle of invariance under superposed rigid-body motions” in favor of purely objective
considerations. Continuum Mech. Thermodyn. 15, 309–20.
Noll, W. (1958). A mathematical theory of the mechanical behavior of continuous media. Arch.
Ration. Mech. Anal. 2, 197–226.
Steigmann, D. J. (2017). Finite Elasticity Theory. Oxford University Press, Oxford.
3
A primer on tensor analysis in
three-dimensional space

Certain concepts that we will encounter in our study of the kinematics of elastic–plastic
deformation, such as the notion of a dislocation density, may be framed in terms of
differential geometry, a proper understanding of which requires a facility with tensor
analysis. The geometric interpretations we will encounter are profound and beautiful,
and promote an understanding of our subject from a perspective that unifies it with other
branches of the physical sciences. To prepare for this, we devote the present chapter to
a survey of tensor analysis and differential geometry in three-dimensional space. We
emphasize the Euclidean space of our common experience; on the one hand, to develop
intuition, and on the other, because this affords a relatively easy segue to concepts from
non-Euclidean geometry that are of immediate relevance to the phenomena with which
we are concerned.
Most of this chapter is normally part and parcel of the background acquired by the
beginning graduate student. Indeed the literature on this subject is vast, and shows no
sign of abating. Accordingly, we will be brief. Thorough discussions of any concept
encountered here that may be unfamiliar may be found in the excellent texts by Sokol-
nikoff, Lichnerowicz, Green and Zerna, Flügge, and Simmonds, among many others. A
lucid account of advanced topics pertaining to affine geometry, including those geome-
tries of relevance here, may be found in the books by Schouten, Lovelock and Rund,
and Szerkeres, the last of these coming from a more modern perspective.

3.1 Coordinates, bases, vectors, and metrics

As a practical matter, we may think of Euclidean 3-space as consisting of positions y,
say, measured from a specified origin, that locate the points of the space. Each point is
i
identified by a triplet {ξ }; i = 1, 2, 3. The elements of the triplet are the coordinates of
the point, and the position, varying as it does from one point to another, is accordingly
i
a function of the coordinates; we write y(ξ ), the intended meaning being that y is a
function of all three coordinates. We shall assume this position field to be as smooth as
required by the ensuing development.

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0003
 Coordinates, bases, vectors, and metrics 35

1
Let {ci } be a collection of three constants. Then, for example, y(ξ , c2 , c3 ) is the
parametric representation of a particular ξ1 - coordinate curve. The derivative

1 1
g1 = ∂y/∂ξ = lim [y(c1 + Δξ, c2 , c3 ) − y(c1 , c2 , c3 )] (3.1)
Δξ→0 Δξ

is the limit of the scaled chord connecting the points on this curve having coordinates
c1 and c1 + Δξ and is therefore tangential to the curve at the point with coordinate c1 . If
1
g1 is a continuous function of ξ , then the curve is smooth. You can convince yourself
that g1 is oriented in the direction of increasing ξ1 . In the same way we define the three
vector fields
i
gi = y, i ≡ ∂y/∂ξ , (3.2)

in which the comma notation affords a convenient way to denote the partial derivative
with respect to the ith coordinate. The three vectors g1 , g2 , and g3 are then tangential to
1 2 3
the three coordinate curves y(ξ , c2 , c3 ), y(c1 , ξ , c3 ), and y(c1 , c2 , ξ ), respectively, at the
i
point with coordinates ξ = ci .
Henceforth we assume the three coordinates to be independent in the sense that the
induced tangent vectors gi are linearly independent. This is the case if and only if the
scalar triple product g1 · g2 × g3 is non-zero. For the sake of convenience we restrict
ourselves to right-handed coordinates for which g1 · g2 × g3 > 0. In general, for a given
coordinate system, linear independence cannot be guaranteed to hold globally. There
may be particular points, curves, or surfaces where this condition breaks down. Exam-
ples are furnished by the origin in a system of plane polar coordinates, or the axis of
a system of cylindrical polar coordinates. These singularities occur where the relation
between the position y and the coordinates ξ i ceases to be one-to-one. To characterize
them, we introduce global right-handed Cartesian coordinates yi , in terms of which posi-
tion is represented simply as y = yk ik , where the ij are constant orthonormal unit vectors
aligned with the three Cartesian coordinate lines, and a sum over diagonally repeated
indices from 1 to 3 is implied. This is Einstein’s famous summation convention, invented
to dramatically declutter formulae encountered in relativity theory. Because the coordi-
i i
nates yk are uniquely determined by y(ξ ), they are also functions ŷk of the ξ , and we
have the useful result
i
gi = yk,i ik , where yk,i = ∂yk /∂ξ , (3.3)

in which the superposed caret is suppressed to avoid clutter, yielding the gi when the
i
functions yk = ŷk (ξ ) are known. Again we have a sum on k for each fixed i. Introducing
the permutation symbol
 
 1, if {ijk} is an even permutation of {123} 
eijk = ii · ij × ik = −1, if {ijk} is an odd permutation of {123} , (3.4)
 
0, otherwise
36 A primer on tensor analysis in three-dimensional space

we then have

g1 · g2 × g3 = yi,1 ii · y,2j ij × yk,3 ik = eijk yi,1 y,2j yk,3 = det(yi, j ), (3.5)

where (yi, j ) is the 3 × 3 matrix with (i, j) entry equal to yi, j . The inverse function theorem
guarantees that the relation between the y’s and the ξ ’s, and hence that between the
latter and y, is locally one-to-one wherever this determinant is non-zero. Accordingly,
the singularities just mentioned occur wherever this determinant vanishes. At regular
i
points where the determinant is non-zero, we can thus invert the function y(ξ ) to obtain
i i
ξ = ξ̂ (y). Moreover, the coordinates ξ i are right-handed if and only if

det(yi, j ) > 0, (3.6)

a restriction that we assume henceforth to hold almost everywhere in 3-space, i.e., on

sets of positive volume measure.
At this stage we recall that the gradient of a scalar-valued function f(y) is the unique
vector gradf such that

df = gradf · dy, (3.7)

which is shorthand for the formula

f ′ = gradf · y′ , (3.8)

wherein

d d
y′ = ŷ(t), f′ = f(ŷ(t)) (3.9)
dt dt

i
and ŷ(t) = y(ξ (t)) is the parametric representation of an arbitrary curve. By the chain
rule and (3.7), applied to the coordinate functions,

i i
df = f, i dξ = f, i gradξ̂ · dy, (3.10)

i
where, as usual, f, i = ∂f/∂ξ . Substituting this in place of the left-hand side of (3.7) and
invoking the arbitrariness of the curve, i.e., the arbitrariness of dy, we find that

i
gradf = f, i gi , where gi = gradξ , (3.11)

in which the superposed caret has again been dropped. Evidently g1 is orthogonal to the
1
surface defined by ξ = c1 , etc.
 Coordinates, bases, vectors, and metrics 37

Cartesian coordinates have the interesting property that

ik = ∂y/∂yk and ik = gradyk . (3.12)

Regarding the ξi as functions of the yk , we can use this fact together with the chain rule
to obtain
i i
∂ξ ∂ξ
gi = k
gradyk = k ik , (3.13)
∂y ∂y

which may be compared to (3.3) and in which a sum on k is again implied. Here
and henceforth we write out the partial derivative explicitly when differentiating with
respect to Cartesian coordinates. From this we can conclude that the set {gi } is linearly
independent wherever {gi } is, and with the same orientation. Thus,
1 2 3
∂ξ ∂ξ ∂ξ
ii · j ij × k ik = det(∂ξ /y j ) = [det(yi, j )]−1 ,
i
g1 · g2 × g3 = (3.14)
∂yi ∂y ∂y

which is non-zero if and only if g1 · g2 × g3 is non-zero, and with the same sign. Accord-
ingly, {gi } constitutes a basis for 3-vectors wherever {gi } does, i.e., wherever det(yi, j ) ̸= 0.
Invoking (3.3), (3.13), and the chain rule, we also find that
i i  
∂ξ ∂ξ i 1, if i = j
g · gj = k ik · yl, j il = k yk, j = ξ, j = δij ≡
i
. (3.15)
∂y ∂y 0, otherwise

Here δji , the components of the Kronecker delta, are the elements of the 3 × 3 unit
matrix, and we have used the fact that ik · il equals unity if k = l, and zero otherwise.
Because of this result the basis {gi } is said to be reciprocal to {gi }, and vice versa. It is
clear from (3.12) that the induced tangent vectors coincide with their reciprocals when
the coordinates are Cartesian.
We will have occasion to make use of tensorial generalizations of the permutation
symbol. We define these, using Cartesian coordinates as intermediaries, by
r s t
∂ξ ∂ξ ∂ξ ijk
ϵrst = y,ri y,sj y,tk eijk and ϵrst = e , where e ijk = eijk . (3.16)
∂yi ∂y j ∂yk

Thus,

ϵrst = det(yi, j )erst and ϵrst = [det(yi, j )]−1 erst . (3.17)

These enter in a natural way in cross-product operations. For example, using (3.3) and
the chain rule gives

gi × gj = y,ik yl, j ik × il = yk,i yl, j eklm im

r
= yk,i yl, j eklm gradym = yk,i yl, j ym
,r eklm gradξ , (3.18)
38 A primer on tensor analysis in three-dimensional space

which is just

gi × gj = ϵijr gr . (3.19)

In the same way we can establish that

gi × g j = ϵijr gr . (3.20)

These are the extensions of the Cartesian formula ii × ij = eijr ir to arbitrary right-handed
coordinate systems.
Our next order of business is to define the metric components

gij = gi · gj . (3.21)

Using the reciprocal basis, we may also define reciprocal metric components

g ij = gi · g j . (3.22)

These obviously possess the symmetries gij = gji and g ij = g ji . Further, we can exploit
the fact that {g j } is a basis for 3-vectors under our hypotheses to write gi in the form
gi = aij g j with a unique set {aij } of coefficients. Then,

aik = aij δkj = aij g j · gk = gi · gk = gik , (3.23)

implying that

gi = gij g j . (3.24)

In the same way we have

gi = g ij gj . (3.25)

We may combine these results as

0 = δik gk − gi = δik gk − g ij gj = (δik − g ij gjk )gk , (3.26)

and conclude, by the presumed linear independence of the gi , that

g ij gjk = δik , (3.27)

and hence that the matrices (g ij ) and (gij ) are mutual inverses.
 Coordinates, bases, vectors, and metrics 39

This observation begs an obvious question concerning the invertibility of these

matrices. To address this, let

g = det(gij ). (3.28)

From (3.3) and (3.21) we have

X
gij = yk,i yl,j ik · il = yk,i yk, j (3.29)
k

which implies that g = [det(y,ij )]2 and hence, with (3.6), that (gij ) is indeed invertible;
and that

g−1 = det(g ij ). (3.30)

i
The name metric derives from dy = gi dξ (see (3.2)) and ds = |dy| , yielding

i j
ds2 = gij dξ dξ . (3.31)

i
Integrating ds along a curve with parametric equations ξ (t), from t 1 to t2 , thus yields
the arclength of the part of the curve connecting the latter two points. Further, if dy ̸= 0
i
then ds2 > 0, and by the linear independence of {gi }, at least one element of {dξ } is
non-zero. Equation (3.31) then implies that (gij ) is a positive definite matrix and hence
that g > 0.
As a side note we conclude from (3.17) that
√ √
ϵijk = geijk and ϵijk = eijk / g (3.32)

in our right-handed coordinate system.

We may exploit the status of {gi } and {gi } as bases for 3-vectors to write any such
vector, v say, in either of the forms

v = vi gi = vi gi , (3.33)

with unique, and generally distinct, coefficients vi and vi . These are classically referred
to as the co- and contravariant components of v, respectively, and we shall adhere to this
convention. For example, using the second representation,

v · g j = vi gi · g j = vi δij = v j , (3.34)

so that contravariant components are simply the projections of the vector in ques-
tion onto the corresponding reciprocal basis elements. The first representation similarly
yields
40 A primer on tensor analysis in three-dimensional space

v · gj = vj (3.35)

and hence the conclusion that the covariant components are its projections onto the
tangent basis elements.
Beyond this we can use the two representations in (3.33) to infer that

vj g j = vi gi = vi gij g j , (3.36)

and hence that

vj = gji vi , (3.37)

where we have used gji = gij . Alternatively, we can again proceed from (3.33), or directly
from the invertibility of the matrix of metric components, to derive the companion
formula

v j = g ji vi . (3.38)

These rules go by the names lowering the index and raising the index, respectively.
i
Problem 3.1 Compute the gi , gij , gij , and gi in spherical coordinates {ξ } = {r, θ, ϕ},
using

y1 = r cos ϕ cos θ, y2 = r cos ϕ sin θ, y3 = r sin ϕ.

At what points, if any, are the relations between the Cartesian and spherical
coordinates non-invertible?

Problem 3.2 Establish the four representations

ui v j gij = ui vi = uj vi g ij = uj v j

for the scalar product u · v of two 3-vectors.

Problem 3.3 Establish the two representations

ϵkij ui v j gk = ϵkij ui vj gk

for the cross product u × v of two 3-vectors.

i
Problem 3.4 Find {gi }, {gi }, gij , and gij in terms of {ξ } = {ξ, η, ϕ} for the follow-
ing coordinate systems: (a) Parabolic coordinates: y1 = ξη cos ϕ, y2 = ξη sin ϕ,
2
y3 = 12 (ξ −η2 ). (b) Elliptic-cylindrical coordinates: y1 = cosh ξ cos η, y2 = sinh ξ sin η,
y3 = ϕ.
 Second-order tensors 41

Problem 3.5 Consider the complex-valued analytic function g(z) of the complex vari-
able z = x + iy, where x(= y1 ) and y(= y2 ) are Cartesian coordinates in the plane.
i
Let {ξ } = {ϕ, ψ, y3 }, where ϕ = Re g and ψ = Im g. Consider the example g(z) = z2
and compute {gi }, {gi }, gij , gij in terms of the yi . Sketch the coordinate surfaces,
i.e., the surfaces on which each coordinate is constant.

i
Problem 3.6 Consider an oblique coordinate system {ξ } for which g1 = i1 ,
g2 = cos ϕi2 + sin ϕi1 , g3 = i3 with ϕ a constant angle.

(a) Find the yi as functions of the ξi . (Hint: construct the position vector and
suppose the ξi vanish at the origin.) Invert these relations.
(b) Obtain gij , gij and find {gi } in terms of {ii }.
(c)
(c) If vi are the Cartesian components of a vector, what are the co- and
i
contravariant components of the vector in the {ξ } system?

3.2 Second-order tensors

Vectors are essentially first-order tensors in the sense that they are fully specified by
components, relative to a basis, bearing a single index (which, of course, ranges over
the dimension of the space). That is, they have 31 components, where the exponent
is the order and the base is the dimension of the space. Second-order tensors, with 32
components, are effectively 9-dimensional vectors. They are linear combinations of the
elements of {gi ⊗ gj }, {gi ⊗ g j }, {gi ⊗ g j }, or {gi ⊗ g j }, and their components accord-
ingly bear two indices. To demonstrate the linear independence of the elements of the
first set, we equate to the zero tensor (i.e., the 9-dimensional zero vector) the linear com-
bination αij gi ⊗gj . Now operate the result on gk to obtain αik gi = 0 and thus conclude that
all αij vanish. Because the set {gi ⊗ gj } contains nine linearly independent elements, it
constitutes a basis for the linear space of second-order tensors. That is, any such tensor,
A say, is expressible as

A = Aij gi ⊗ gj , (3.39)

with unique coefficients Aij . Reasoning in precisely the same way, we have the alternative
representations

A = Aij gi ⊗ g j = Ai·j gi ⊗ g j = A·i j gi ⊗ gj , (3.40)

again with unique coefficients. The Aij and Aij respectively are the co- and contravari-
ant components, and the Ai·j and A·i j respectively are the left-mixed and right-mixed
components; the dot is inserted as a device to distinguish one type from the other.
42 A primer on tensor analysis in three-dimensional space

With (3.40) in hand we have

Ag j = (Akl gk ⊗ gl )g j = Akl gk (gl · g j ) = Akl δlj gk = Akj gk , (3.41)

and dotting with gi gives

Aij = gi · Ag j = A · gi ⊗ g j . (3.42)

Thus, the contravariant components are simply the projections of the tensor onto the
corresponding reciprocal basis elements. In the same way we can easily establish the
similar formulas

Aij = A · gi ⊗ gj , Ai·j = A · gi ⊗ gj , and A·i j = A · gi ⊗ g j . (3.43)

Beyond this, we can use the metric and reciprocal metric to connect the various sets
of components via index raising and lowering. For example,

glm Akl gk = Akl gk (g l · g m ) = Agm = Aij gi (gj · gm ) = Aim gi , (3.44)

and a further dotting with gn gives

Anm = gnk gml Akl . (3.45)

In the same way,

Aij = gik A·kj = gjk Ai·k , Aij = gik gjl Akl , etc. (3.46)

Problem 3.7 Show that At = Ai·j g j ⊗ g i , and hence that Ai·j = A·j i if A is symmetric. For
symmetric A, show that Aij = Aji and Aij = Aji . What if A is skew-symmetric, i.e.,
At = −A?

The unit tensor i is defined by iv = v for any vector v. Accordingly, it has the various
representations

i = (gi · igj )gi ⊗ g j

= (gi · gj )gi ⊗ g j = gij gi ⊗ g j = gj ⊗ g j
= δij gi ⊗ g j = g ij gi ⊗ gj = g j ⊗ gj = δij gi ⊗ gj . (3.47)

The metric and reciprocal metric, respectively, are therefore the co- and contravariant
components of the unit tensor, whereas the left- and right-mixed components coincide
and are equal to the Kronecker delta. Note that the mixed components are the same in
 Second-order tensors 43

every coordinate system, whereas the co- and contravariant components vary from one
system to another. Obviously the unit tensor is symmetric.
This brings us to the central property of tensors, namely their coordinate invariance.
We can think in terms of imposing such invariance as part of the definition of a ten-
sor. Given this, it is obvious that since the bases used to represent tensors depend on
the coordinates, the components of these tensors must compensate to ensure that the
tensors themselves do not. To explore this issue, consider a coordinate transformation
i i i j
from {ξ } to {ξ̄ }, say, given by the functions ξ̄ = f i (ξ ). We suppose these functions
to be differentiable and invertible, at least locally. For example, we can assume a locally
i
invertible relation between the ξ̄ and Cartesian coordinates yi in the same way that we
i
previously assumed such a relation between the latter and the ξ . This, in turn, ensures
i
the local invertibility of the functions f i . Using the ξ̄ we can construct the elements ḡi of
the induced tangent basis and the elements ḡi of the associated reciprocal basis. Natu-
i
rally, this is accomplished by using (3.2) and (3.11), but with the ξi replaced by ξ̄ . The
chain rule then furnishes
j i
∂ξ ∂ξ̄
ḡi = g
i j
and ḡi = j
g j, (3.48)
∂ξ̄ ∂ξ

where, in a harmless abuse of notation, we have replaced the functions f i by their values.
These are the change-of-basis formulas referred to previously. Then for any vector v,
i i
∂ξ ∂ξ
vi gi = v = v̄ j ḡj = v̄ j g,
j i
and thus vi = j
v̄ j , (3.49)
∂ξ̄ ∂ξ̄

and, in the same way,

j
∂ξ̄
vi = v̄ .
i j
(3.50)
∂ξ

The invertibility of the coordinate transformation implies that the same results hold with
the barred and unbarred variables interchanged, i.e.,
i j
∂ξ̄ ∂ξ
v̄i = j
vj and v̄i = v.
i j
(3.51)
∂ξ ∂ξ̄

Exactly the same line of reasoning leads to the transformation rules

k l i j
∂ξ ∂ξ ij ∂ξ̄ ∂ξ̄
A¯ij = i j Akl , A¯ = k l Akl , etc., (3.52)
∂ξ̄ ∂ξ̄ ∂ξ ∂ξ

for the components of second-order tensors.

44 A primer on tensor analysis in three-dimensional space

3.3 Derivatives and connections

We’ve already introduced and used the gradient operation on scalars defined by (3.7).
We extend this to vector-valued functions v(y) and define the gradient, gradv, to be the
second-order tensor such that

dv = (gradv)dy. (3.53)

i i
Regarding v as a function of the coordinates via v(y(ξ )), we have dv = v, i dξ = v, i (gi ·
dy). Inserting this on the left-hand side of the definition and demanding that equality
hold for all dy gives

gradv = v, i ⊗ gi . (3.54)

To extract a representation in terms of components, we write

v, i = (v j gj ), i = v,ji gj + v j gj,i , (3.55)

where we account for the fact that the gj , unlike their Cartesian counterparts, generally
depend on the coordinates. Now gj,i , like any other vector, may be represented in terms
of a basis. We choose the basis {gk } at the particular point in question, with coordinates
ξ i , and use (3.33) and (3.34) to conclude that

gj,i = Γkji gk , where Γkji = gk · gj,i . (3.56)

The gammas are called the connection coefficients; they effectively connect the tangent
i i
space Span{gk } at the point ξ i to the tangent space at the neighboring point ξ + dξ .
We’ve been quite cavalier about the choices of bases used in our various representa-
tions. In Euclidean space we could just as well have used any other basis; for example,
the basis {gk } existing at some remote point of the space. The reason is that Euclidean
space is effectively flat in the sense that it coincides with its own tangent space. Accord-
ingly, vectors attached to points remote from each other can be added with impunity. In
a more general space, or manifold, vectors at a particular point are elements of the space
tangent to the manifold at that point. Vectors attached to distinct points of the manifold
belong to distinct tangent spaces, and their addition would thus have no vectorial mean-
ing. The choice we made in the course of arriving at (3.56) thus makes sense in spaces
more general than Euclidean, as we shall see later.
However, in Euclidean space we can invoke (3.2) and, assuming the position field
i
y(ξ ) to be twice differentiable, conclude that

Γkji = gk · y,ji = gk · y,ij = gk · gi,j = Γkij , (3.57)

where the comma followed by two subscripts is a shorthand for second partial deriva-
tives. Accordingly, the gammas are symmetric in the subscripts.
 Derivatives and connections 45

Problem 3.8 Compute the gammas for spherical coordinates.

It turns out that in a certain non-Euclidean space of relevance to plasticity theory the
gammas do not possess this symmetry. To cover this possibility we define the torsion

T·ijk = Γk[ij] , (3.58)

where the square bracket is used to denote skew symmetrization; that is,

1 k
Γk[ij] = (Γ − Γkji ). (3.59)
2 ij
Clearly, then, the torsion vanishes in Euclidean geometry. Later, we will encounter
another non-Euclidean geometry wherein the torsion also vanishes. This is Riemannian
geometry, which, in the extension to four dimensions, furnishes the setting for Einstein’s
astounding theory of gravitation.
Indeed in a more general geometry one does not have (3.56), but rather
i
dgj = Γkji gk dξ , (3.60)

by which we mean
i
ġj = Γkji gk ξ̇ , (3.61)

i
where ξ (t) are the parametric equations of a curve in the manifold, superposed dots
are derivatives with respect to t, and all quantities are defined on the curve. See Section
9.7 of the book by Stoker, or the delightfully lucid discussion in Chapter 5 of Brillouin’s
book. Given the Γkji , we have a linear system of ordinary differential equations, and a
standard existence-uniqueness theorem ensures that these can be integrated, granted the
availability of initial conditions gj (t0 ), giving gj (t). Such geometries, which subsume all
i
those of interest to us, are said to be affinely connected because the ξ̇ enter (3.61) linearly.
We can think of (3.61) as defining the directional derivatives of the gj in the direction of
i
the tangent, with components ξ̇ , to the curve. The gammas appearing in the expressions
for these derivatives go by the name Koszul connection in the recent literature. See the
book by Szerkeres, for example.
On the other hand, (3.56) makes sense only if the gj and Γkji are defined on the man-
ifold in a neighborhood of the point on the curve in question. Indeed if this is the case,
i
then by the chain rule the left-hand side of (3.61) becomes gj,i ξ̇ , provided that the gj are
differentiable, and (3.56) follows if the resulting equation is to hold for all curves; that is,
i
for arbitrary ξ̇ . Following Einstein, we then have a space with distant parallelism. These
are exemplified by Euclidean space, naturally, and also by a certain space with torsion
alluded to previously that will prove to be of interest to us. Clearly, then, the passage
from (3.61) to (3.56) is not automatic, but requires that certain integrability conditions
46 A primer on tensor analysis in three-dimensional space

be satisfied. For example, if the gj are twice differentiable, so that gj,[kl] vanishes, then
for (3.56) to hold it is necessary—and in a simply connected region, also sufficient (see
Ciarlet’s book)—that

(Γijk gi ),l = (Γijl gi ),k . (3.62)

It turns out that this imposes a serious restriction on the gammas and their derivatives.
This restriction is automatically satisfied in Euclidean space provided that the position
field therein is thrice differentiable. For we then have gj,kl = y,jkl , which is completely
symmetric in the subscripts.
We have digressed, perhaps too far, to discuss non-Euclidean spaces. Further dis-
cussion will be forthcoming as the need arises. In the meantime we return to the lowly
gradient of a vector field in Euclidean space. Thus, we combine (3.55) and (3.56) and
shuffle the indices to obtain

v, i = vk;i gk , where vk;i = vk, i + Γkji v j (3.63)

is the so-called covariant derivative. We then have

gradv = vk;i gk ⊗ gi , (3.64)

and hence the interpretation of the vk;i as the left-mixed components of gradv. Given
these, we can raise and lower indices in the usual way to obtain the various other kinds
of components.
To arrive at (3.64) we started with the tangent-basis decomposition of v in terms of its
contravariant components. Of course we might have used the alternative reciprocal-basis
decomposition, in terms of covariant components, instead. Then, in place of (3.55) we
would have had

v, i = vj,i g j + vj g,ji . (3.65)

This time we use the first of (3.33) with (3.35) to express the g,ji as

g,ji = (gk · g,ji )gk , (3.66)

in terms of the reciprocal basis existing at the point in question, in which the coefficients
follow by differentiating (3.15) and substituting (3.56), giving

gk · g,ji = −Γkij . (3.67)

Thus,

v,i = vk;i gk , where vk;i = vk,i − Γkij vj . (3.68)

This is also called a covariant derivative. Note that in both types the second of the lower
indices on the gammas corresponds to the index used to label the derivative. Be aware
 Derivatives and connections 47

that some workers use a definition of the gammas in which the lower indices are switched
relative to our definition. In any case we now have

gradv = vk;i gk ⊗ gi , (3.69)

and hence the interpretation of vk;i as the covariant components of the (same) tensor
gradv.
We note in passing that the covariant derivative operations, applied directly to the
tangent and reciprocal basis elements, yield

gi;j = 0 and gi;j = 0, (3.70)

i.e., the basis elements are covariantly constant. These extend their Cartesian counter-
parts; namely, ∂ii /∂y j = 0, to arbitrary coordinates.
With the gradient in hand we can compute the divergence, divv, of a vector field. This
is the scalar field defined intrinsically, i.e., without reference to coordinates, by

divv = tr(gradv). (3.71)

Using (3.54) and (3.63) we have

divv = v, i · gi = vk;i gk · gi = vi;i = (gki vk );i . (3.72)

Alternatively, using (3.69),

divv = vk;i gk · gi = gki vk;i . (3.73)

If these are to agree, as they must, then it seems the gki can be taken outside the covariant
derivative. We return to this point below, after we introduce covariant derivatives of
second-order tensors.
Another extremely important differential operator is the Laplacian, Δϕ, of a scalar-
valued function ϕ, defined intrinsically by

Δϕ = div(gradϕ), (3.74)

where grad is the gradient. Combining this with (3.54) and (3.11) results in

Δϕ = (gradϕ), i · gi = gi · (ϕ, j g j ), i
= gi · (ϕ,ji g j + ϕ, j g,ji ) = gki ϕ;ki , (3.75)

where

ϕ;ki = ϕ,ki − Γkij ϕ, j . (3.76)

Note that ϕ;ki = ϕ;ik in a torsion-free space, provided that ϕ is twice differentiable.
48 A primer on tensor analysis in three-dimensional space

Moving on to second-order tensors, we proceed, as in Chapter 1, to define the diver-

gence of a tensor field A(y) intrinsically, at a fixed point of space, to be the vector field
divA such that

c · divA = div(At c), (3.77)

for any fixed vector c. Writing v for At c and invoking (3.72), we then have

c · divA = gi · At, i c = (At, i )t gi · c. (3.78)

We have rather nonchalantly failed to specify which comes first: the transpose or the
derivative. In fact, the order doesn’t matter.

Problem 3.9 Prove this.

Then,

c · [divA − (A, i )gi ] = 0, (3.79)

and as this purports to hold for all fixed c, we may take it to coincide with the square
bracket, fixed at the point in question, to conclude that the norm of the bracket, and
hence the bracket itself, vanishes. Thus,

divA = (A, j )g j . (3.80)

This exceedingly useful formula conceals fairly complicated component expressions.

For example, starting with the representation (3.39), differentiating everything in sight
and applying the product rule to the tensor products, we get

;j gi ⊗ gk ,
A, j = Aik (3.81)

where

Aik ik il k lk i
;j = A, j + A Γlj + A Γlj (3.82)

is one among four kinds of covariant derivative. Another kind follows by using the first
of Eqs. (3.40): Thus,

A, j = Aik;j gi ⊗ gk , (3.83)

where

Aik;j = Aik,j − Alk Γlij − Ail Γlkj . (3.84)

 Derivatives and connections 49

Finally,

divA = Aij;j gi = Aik;j gkj gi , etc. (3.85)

Problem 3.10 Show that for two differentiable vector fields u and v,

(u ⊗ v), i = u, i ⊗ v+u ⊗ v, i .

Problem 3.11 Verify (3.82) and (3.84), and write out expressions for the remaining
two kinds of covariant derivative, namely Ai·k;j and A·k;ji .

The covariant derivative provides a covariant (i.e., tensorially meaningful) extension

of the partial derivative. Unsurprisingly, it possesses all of the useful properties of the
latter, including the product rule. In fact, modern treatments include the product rule
among the defining properties of the covariant derivative. Here we content ourselves
with a verification of the product rule in the example

v = Au = (Aij gi ⊗ gj )u = Aij uj gi ; that is, vi = Aij uj . (3.86)

Then,

vi;k = vi,k + v j Γijk = Aij,k uj + Aij uj,k + Ajl ul Γijk

= Aij,k uj + Aij uj,k + Alj Γilk uj + (Ail Γlkj uj − Aij Γljk ul )
= (Aij,k + Ail Γlkj + Alj Γilk )uj + Aij (uj,k − Γljk ul )
= Aij;k uj + Aij uj;k , (3.87)

in which zero has been added in the second line, in the guise of the parenthetical term.
In the same way we can establish that

(Aij Bkl );m = Aij ;m Bkl + Aij Bkl;m (3.88)

and

(ϕvk );l = ϕ;l vk + ϕvk;l , (3.89)

etc., with the proviso that the covariant derivative of a scalar field is the same as its partial
derivative, i.e.,

ϕ;l = ϕ,l . (3.90)

We extend this convention to any invariant. Thus, for vectors,

v;k = v,k (3.91)

50 A primer on tensor analysis in three-dimensional space

and for second-order tensors,

A;k = A,k . (3.92)

For example,

A,k = Aij;k gi ⊗ gj + Aij (gi;k ⊗ gj + gi ⊗ gj;k ), (3.93)

and (3.81) follows from (3.70).

Concerning the metric, we combine (3.21) and (3.56), obtaining

gij,k = (gi · gj ),k = glj Γlik + gli Γljk ; that is, gij;k = 0. (3.94)

This establishes that the unit tensor (3.47) is spatially uniform, i.e., independent of the
coordinates. A more clever derivation follows by invoking the product rule for covariant
derivatives together with (3.70). Thus,

gij;k = (gi · gj );k = gi;k · gj + gi · gj;k , (3.95)

which of course vanishes identically. In the same way we easily establish that

gij;k = 0. (3.96)

The metric and reciprocal metric are therefore covariantly constant, and the connection
is said to be metric-compatible.

3.4 The Levi-Civita connection

This important result, known as Ricci’s lemma, explains why it is indeed correct to take the
metric outside the covariant derivative, as in the comparison of (3.72) and (3.73). It leads
straightforwardly to an important relation giving the gammas in terms of the metric and
its partial derivatives. To derive this we first introduce the auxiliary connection symbols

Γjik = glj Γlik , (3.97)

allowing (3.94) to be expressed more tidily as

gij,k = Γjik + Γijk . (3.98)

Note that Γj[ik] = glj T·lik , so that Γjik = Γjki in a torsionless (e.g., Euclidean or Riemannian)
geometry. It follows from (3.98) that

1
(gki,j + gkj,i − gij,k ) = Γi[kj] + Γj[ki] + Γk(ij) , (3.99)
2
 The Levi-Civita connection 51

where the round braces are used to denote symmetrization, i.e.,

1
Γk(ij) = (Γkij + Γkji ). (3.100)
2

Accordingly, in the torsionless Euclidean space with which we are currently concerned,

1
Γkij = (gki,j + gkj,i − gij,k ). (3.101)
2

To recover the connection symbols we use

m l
Γm mk l mk
ij = δl Γij = g glk Γij = g Γkij , (3.102)

arriving finally at

1 mk
Γm
ij = g (gki,j + gkj,i − gij,k ). (3.103)
2

This is called the Levi-Civita connection. Note that its derivation, from (3.98), made
no use of (3.56), but instead relied only on the notion of a metric and a metric-compatible
connection. Accordingly, it applies equally to Riemannian geometry, defined simply as
a space endowed with a (symmetric) metric and a Levi-Civita connection. In particular,
then, Euclidean geometry is a special case.
In fact, we are following the program of Cartan’s intuitive approach to Riemannian
geometry, whereby a Euclidean tangent space is associated with a point of the Rieman-
nian manifold. In this tangent space the dot product is well defined, and may be used to
generate a metric as in (3.21). The metric of the Riemannian space is then defined to be
that of the Euclidean tangent space at the point in question. In fact, we can go a step fur-
ther and take the Levi-Civita connection of the Riemannian space to coincide with that of
the Euclidean tangent space, again at the point in question. We then speak of an osculat-
ing Euclidean space. Thus, the partial derivatives of the metrics are also matched and the
Euclidean rules for covariant differentiation are preserved in the Riemannian geometry.
Concise descriptions of this procedure are given in the books by Willmore, Lichnerow-
icz, Stoker, and Cartan. In this way much of Euclidean geometry is neatly carried over
to Riemannian geometry. However, the curvature of Riemannian space, a concept to be
introduced shortly, intervenes to prevent the matching of higher-order derivatives of the
metric. In a space with torsion we can also construct a Euclidean tangent space but this
cannot be osculating because in the latter the connection is symmetric.
By the way, the so-called non-metric geometry conceived by Weyl, in which the con-
nection is not metric compatible, is also of interest. In this geometry Ricci’s lemma is
replaced by gij;k = qijk (with q[ij]k = 0), where qijk , the source of non-metricity, may be
interpreted as a continuous distribution of point defects, such as interstitial atoms or
vacancies, in a crystal. See the paper by Povstenko, for example. We are not concerned
with such geometries here, however.
52 A primer on tensor analysis in three-dimensional space

A most useful property of the Levi-Civita connection follows by contracting (3.103)

on the upper index. Using the symmetry of the metric, we find that

2Γiij = gik gik,j + (gik gkj,i − gki gij,k ), (3.104)

in which the second parenthetical term is seen, on relabelling the summation indices, to
be the same as the first. Accordingly, the parenthesis vanishes, leaving

1 ik
Γiij = g gik,j . (3.105)
2
This is simplified by recalling that the derivative of the determinant of a matrix with
respect to one of its components is the corresponding component of the cofactor matrix.
Thus,

∂g
= ggik , (3.106)
∂gik

so that
1 ∂g 1 √ √
Γiij = gik,j = g, j = ( g), j / g. (3.107)
2g ∂gik 2g

With this result in hand let us revisit the divergence and Laplacian operators. For the
first of these, (3.63) and (3.72) give
√
( g), j j 1 √
divv = vi;i = vi, i + √ v = √ ( gvi ), i , (3.108)
g g

which does not require the explicit (and rather tedious) computation of the connection
coefficients. For the vector field v = gradϕ we have vi = g ij ϕ, j ; Eqs. (3.74) and (3.108)
then furnish the Laplacian

1 √
Δϕ = √ ( gg ij ϕ, j ), i . (3.109)
g

Problem 3.12 Write out the Laplacian Δf(ξ, η, ϕ) explicitly in terms of f and its partial
derivatives with respect to ξ, η, ϕ for the coordinate systems of Problem 3.4.

Problem 3.13 In cylindrical polar coordinates {r, θ, z} the position function is given by
y= rer (θ) + zi3 , where r is the radius from the z-axis, z measures distance along this
axis, θ is the azimuthal angle, and er (θ) = cos θi1 + sin θi2 . In domains with cone-
shaped boundaries, it may be more convenient to use a conical-polar coordinate
system. For this purpose we introduce coordinates {r̄, θ, z̄}, where r = r̄ cos ϕ,
 The Levi-Civita connection 53

z̄ = z − r̄ sin ϕ, and ϕ is the constant cone angle. The position function may then
be written y = r̄ēr (θ) + z̄i3 , where ēr (θ) = cos ϕer (θ) + sin ϕi3 (draw a figure). Let
i
{ξ } = {r̄, θ, z̄} and write out the divergence Aij;j explicitly in terms of Aij and their
partial derivatives.

Problem 3.14 Show that the Laplacian of a function f(r, θ, ϕ) in spherical coordinates
is

1 2
Δf = [(r fr )r + (sec2 ϕ)fθθ + sec ϕ(fϕ cos ϕ)ϕ ],
r2

where the subscripts stand for partial derivatives.

i
Problem 3.15 Position in a helical coordinate system {ξ } = {r, θ, s} is described by

i
y(ξ ) = r(s) + r cos θn(s) + r sin θb(s),

where r(s) is the equation of the center-line of a helical tube, s is arclength along the
center-line, n(s) is the principal normal to the center-line, and b(s) is the binormal.
The coordinates r and θ, respectively, are the radius from the center-line and the
counterclockwise azimuth measured from the principal normal in a plane cross
section of the tube. The curvature of the center-line is κ(s) = |t′ (s)| , where t = r′ (s)
is the unit tangent to the center-line. Wherever κ is non-zero, we stipulate that
n = κ−1 t′ (s) and b = t × n so that {t,n,b} is an orthonormal basis at every point
of the curve.
The Serret–Frenet equations for smooth space curves are (see any elementary
geometry or dynamics text).

n′ (s) = τb − κt and b′ (s) = −τn,

where τ(s) is the torsion of the center-line (not to be confused with the torsion, i.e.,
the skew part of the connection, referred to in the text).
i
Let {ξ } = {r, θ, s} and obtain {gi }, {gi }, gij , gij in terms of t,n,b,κ, τ and the
coordinates. Show that this coordinate system is non-orthogonal; i.e., the matrix
(gij ) is not diagonal.
Helices of uniform radius and pitch are distinguished by the property that the
curvature and torsion are constants, independent of arclength. Obtain an explicit
expression for the Laplacian Δf(r, θ) in this special case, for a function f that is
independent of arclength. This problem is of obvious importance in diverse appli-
cations including, for example, heat conduction in helical wires and fluid flow in
helical tubes.
54 A primer on tensor analysis in three-dimensional space

Problem 3.16 Work out the transformation formula

k m n
∂ξ̄ ∂ξ ∂ξ
Ā·kij = l i j
A·lmn
∂ξ ∂ξ̄ ∂ξ̄

for components Ak·ij of a third-order tensor.

(a) Thus, show that the connection coefficients Γkij are not the components of a
tensor, firstly by proceeding from the formula (3.56) for the connection coeffi-
cients, and secondly by proceeding directly from the Levi-Civita form (3.103)
without using (3.56). In fact this result follows from the tensorial property of
the covariant derivatives of a tensor, whether or not the connection has either of
the forms (3.56) or (3.103). See the book by Lovelock and Rund or Szekeres’
book for fuller discussions.
(b) Suppose the connection is asymmetric, with a non-zero torsion T·kij = Γk[ij] .
Show that the T·kij are the components of a tensor, provided that the coordinates
i i
ξ̄ are twice-differentiable functions of the ξ .
(c) Using (3.16), argue that ϵijk;l = 0 and ϵijk
;l = 0 without invoking the formulae for
the covariant derivatives of third-order tensor components. Use a similar argu-
ment to explain why the metric and reciprocal metric are covariantly constant
without resorting to Ricci’s lemma.

3.5 The curl, Stokes’ theorem, and curvature

The last of the main differential operators in Euclidean space is the curl, defined, for a
vector field v, to be the vector curlv such that

c · curlv = div(v × c), (3.110)

for any fixed c. Thus, with grad(v × c) = (v, i × c) ⊗ gi we have

c · curlv = gi · v, i × c = c · gi × v, i , (3.111)

yielding

curlv = gi × v, i = vj;i gi × g j = ϵijk vj;i gk . (3.112)

Note that in this expression the covariant derivatives may be replaced by partial
derivatives without affecting the result (why?).
 The curl, Stokes’ theorem, and curvature 55

We extend the curl operation to a tensor field A by requiring its curl, curlA, to be the
tensor field such that

(curlA)c = curl(At c), (3.113)

again for any fixed c. Then,

(curlA)c = gi × (At, i c). (3.114)

From (3.81) we have

At, i = Ajk
;i gk ⊗ gj and At, i c = Ajk
;i gk (gj · c), (3.115)

so that

gi × (At, i c) = Ajk jk
;i g × gk (gj · c) = A;i gkl g × g (gj · c)
i i l

= A·jl;i ϵilm gm (gj · c) = (ϵilm A·jl;i gm ⊗ gj )c, (3.116)

whence it follows, from (3.113), that

curlA = ϵilm A·jl;i gm ⊗ gj . (3.117)

Of central importance in later chapters is Stokes’ theorem, stated here for Euclidean
geometry. Thus, consider a simply connected patch of surface ω bounded by a smooth
curve γ. Let ω be oriented by a unit-normal field n, and let y(s) be the arclength
parametrization of γ. We stipulate that γ be traversed in the direction dy such that ν ,
defined by νds = dy × n|γ , is the exterior unit normal to ω on γ. Then for any smooth
vector field v(y), Stokes’ theorem is the assertion that
Z Z
v · dy = n · curlvda. (3.118)
γ ω

This too can be extended to a smooth tensor field A(y) by taking v = At c with c an
arbitrary fixed vector. Then,
Z Z Z
c · Ady = At c · dy = n · curl(At c)da
γ γ ω
Z Z
= n · (curlA)cda = c · (curlA)t nda, (3.119)
ω ω
yielding
Z Z
Ady = (curlA)t nda. (3.120)
γ ω
56 A primer on tensor analysis in three-dimensional space

As a caution we note that this expression really has only a symbolic significance. In
practice, to compute the integrals we replace the integrands by their component forms;
in particular, by their Cartesian components. This restriction arises from the fact that the
associated basis {ik } is uniform and thus effectively “factors out” of the integrals. The
resulting expressions have an invariant meaning, but this is lost if we replace the Carte-
sian components by those relative to a general system of coordinates. For we would
not then recover an invariant statement on re-attaching basis elements. Stated differ-
ently, if the integrands are replaced by components relative to arbitrary coordinates,
then the integrations entail the addition of vectors belonging to different tangent spaces,
an operation having no invariant meaning. This observation also applies to the diver-
gence theorem when applied to tensors, as stated in Chapter 1. This caution does not
apply to (3.118) because the integrands therein are scalar fields, for which the notion of
tangent space is irrelevant.
Consider now a parametrization ξ (uα ) of the surface ω in terms of a pair uα ; α = 1, 2,
i

of surface coordinates. These could be the longitudinal and latitudinal angles on the
surface of a sphere, for example. Consider a pair of intersecting u1 - and u2 -coordinate
i i
curves passing through a point on the surface. On these curves we have dξ = dζ and
i
dξ = dηi , respectively, where

i i i
dζ = ξ,1 du1 and dηi = ξ,2 du2 . (3.121)

The oriented area measure of the surface is

i i
nda = gi dζ × gj dη j = ϵijk dζ dη j gk , (3.122)

where the basis elements are evaluated at the point in question. Using (3.112) we then
have
i i
curlv · nda = ϵijk dζ dη j gk × gl · v,l = ϵkij ϵklm dζ dη j gm · v,l . (3.123)

The well-known e − δ identity satisfied by the permutation symbols (see Flügge’s book,
for example) may be used with (3.32) to conclude that

ϵkij ϵklm = δli δm l m

j − δj δi . (3.124)

Then,
l m
curlv · nda = (dζ dηm − dζ dηl )gm · v,l , (3.125)

l m m l
in which the parenthetical term, namely (ξ,1 ξ,2 − ξ,1 ξ,2 )du1 du2 , is the wedge product or
l m
exterior product dξ ∧ dξ . Then,

l m l m l m
curlv · nda = gm · v,l dξ ∧ dξ = (vm,l − vk Γkml )dξ ∧ dξ = vm,l dξ ∧ dξ , (3.126)
 The curl, Stokes’ theorem, and curvature 57

l m m l
where the final equality comes from dξ ∧ dξ = −dξ ∧ dξ , whereas Γkml = Γklm in
Euclidean space. Accordingly, (3.118) may be expressed in the simple form
Z Z
i l m
vi dξ = vm,l dξ ∧ dξ . (3.127)
γ ω

Remarkably, this formula does not involve any geometry whatsoever, Euclidean or
otherwise. It is, in fact, the fundamental statement of Stokes’ theorem, valid for any
smooth functions vi defined on any manifold. Rather, the Euclidean form is prop-
erly interpreted as a special case, obtained by introducing the geometric quantities that
appear in the course of reversing the steps in the derivation. The fundamental form is,
however, more convenient for the considerations that follow.
To wit, consider the so-called parallel transport of a vector field u(t) defined on a curve
i
with parametrization ξ (t). This means that u is constant on the curve and hence that u̇,
the derivative of u(t), vanishes. Writing u = ui gi and invoking (3.61), we have

j
(u̇k + Γkij ui ξ̇ )gk = 0, (3.128)

implying that the parenthesis vanishes on the curve:

j
duk = −Γkij ui dξ . (3.129)

Granted initial conditions uk (t0 ), the theory of ordinary differential equations guarantees
the existence of unique functions uk (t) on the curve, but this by no means ensures that
i i i
a vector field uk (ξ ) exists in space such that uk (t) = uk (ξ ) on the curve ξ (t). If such
j
a field did exist, then we would have duk = uk, j dξ , and if parallel transport occurred on
arbitrary curves, then the field would satisfy

uk, j = −Γkij ui ; (3.130)

i.e., it would be covariantly constant. Further, if this field were twice differentiable, such
that uk,jl = uk,lj , then it would be necessary that

(Γkij ui ),l = (Γkil ui ), j . (3.131)

Putting this another way, if the parallel field did exist, then the duk would be exact
differentials, and their closed-contour integrals would vanish:
Z Z
j
0=− duk = Γkij ui dξ . (3.132)
γ γ
58 A primer on tensor analysis in three-dimensional space

Taking the contour to bound a simply connected surface patch ω and applying Stokes’
theorem, with vj = Γkij ui and k ∈ {1, 2, 3} fixed, we would then have
Z Z
l j l j
0 = (Γkij ui ),l dξ ∧ dξ = (Γkij,l ui + Γkij ui,l )dξ ∧ dξ
ω ω
Z Z
l j l j
= (Γkij,l ui − Γkij Γiml um )dξ ∧ dξ = (Γkmj,l − Γkij Γiml )um dξ ∧ dξ
ω ω
Z
1 l j
= Rk um dξ ∧ dξ , (3.133)
2 ω ·mlj

with

Rk·mlj = 2(Γkm[j,l] − Γki[j Γi|m|l] ), (3.134)

where we have observed the skew symmetry of the wedge product and adopted the
convention that an index between vertical bars is exempted from skew symmetrization.
For example,

1
A[j|m|l] = (Ajml − Almj ), (3.135)
2
so that

Rk·mlj = Γkmj,l − Γkml,j + Γkil Γimj − Γkij Γiml . (3.136)

Thus, unless the Rk·mlj vanish, a parallel field does not exist, whereas their vanishing
ensures the existence of such a field in a neighborhood of the point in question. Section
3.7 of the book by Lovelock and Rund contains a particularly clear explanation of this
phenomenon. Working in two dimensions, we find that parallel fields do not exist on
the surface of a sphere (see Flügge’s book). Because of examples like this, the Rk·mlj are
referred to as the components of the curvature tensor, although at this stage we have not
shown that they actually satisfy the appropriate (fourth-order) tensor transformation
formula. But we can derive some small comfort from the fact that

Rk·m(lj) = 0, (3.137)

which of course is an immediate consequence of the definition (3.134).

Problem 3.17 Combine (3.130) and (3.131) to obtain Rk·mlj um = 0.

Problem 3.18 Show that (3.62) is true, with the gammas given by (3.56), if and only
if Rk·mlj = 0.

Problem 3.19 Consider a vector field with covariant components vi , and let Aij = vi;j .
Suppose the vi are twice differentiable, so that vi,jk = vi,kj .
 Torsion and the Weitzenböck connection 59

(a) Show that, in a general affine space with both torsion and curvature,

1 m
Ai[j;k] + T·ljk Ail = R vm .
2 ·ijk
(b) Note that the left-hand side of this equation is tensorial; i.e., it represents the
covariant components of a third-order tensor. Use this fact to establish the trans-
formation formula giving R̄m n
·ijk in terms of the R·pqr and hence conclude that R·ijk
m

are indeed the components of a fourth-order tensor. (You could proceed from
(3.136) to reach the same conclusion, but the present method is considerably
simpler.)
(c) It follows from the above formula that, unlike partial differentiation, covariant
differentiation generally does not commute, i.e., vi;jk ̸= vi;kj , where vi;jk is short-
hand for (vi;j );k . Exceptionally, vi;jk = vi;kj if the space is Euclidean (Tl·jk = 0 and
Rm·ijk = 0). Prove this fact directly, using only the tensorial property of vi;jk and
the fact that it is possible to parametrize Euclidean space in terms of Cartesian
coordinates.

3.6 Torsion and the Weitzenböck connection

Consider now a space with connection coefficients of the form (see (3.56))

Γkij = gk · gi,j . (3.138)

Of course this subsumes Euclidean space, with the g’s constructed, per our recipe, from
i
a position field y(ξ ). However, let us not confine ourselves to this case. Rather, let us
consider this connection, called the Weitzenböck connection, on its own merits. That is,
for the moment let us not require that the gi be the derivatives of any position field. In
this case the gi are said to be anholonomic. In particular, this space may possess torsion
because there is no a priori reason why the g[i,j] should vanish.
To compute the curvature, we begin with

Γkmj,l = gk,l · gm,j + gk · gm,jl , (3.139)

provided, of course, that the g′ s are twice differentiable. We also have

Γkil Γimj = Γksl Γimj δsi = Γksl Γimj gs · gi

= Γksl gs · Γimj gi = Γksl gs · gm,j
= −gk,l · gm,j . (3.140)

Then,

Γkmj,l + Γkil Γimj = gk · gm,jl , (3.141)

60 A primer on tensor analysis in three-dimensional space

and (3.134) yields

Rk·mlj = 2gk · gm,[jl] , (3.142)

which, of course, vanishes identically because of the assumed degree of differentiability.

Accordingly, any geometry with a Weitzenböck connection is flat. From this last formula
we also have
1 k
gm,[jl] = (gk · gm,[jl] )gk = R g , (3.143)
2 ·mlj k
which implies that that the vanishing of the curvature components constitute the inte-
grability conditions for the system (3.60), ensuring the existence of twice-differentiable
fields gi in a simply-connected region. This situation naturally subsumes Euclidean
space, but, because torsion is allowed, flat spaces need not be Euclidean.

Problem 3.20 Assuming the differentials (3.60) to be exact, apply Stokes’ theorem to
arrive at an integral constraint involving the curvature tensor.

If the curvature does not vanish, then all we can say is that (3.60) is integrable on
curves, but associated fields gi do not exist in the space. In this case the geometry
is not characterized by a connection of the form (3.138). An example is Rieman-
nian space, characterized by a Levi-Civita connection (3.103) in which the metric is
the basic variable. That is, whereas the Levi-Civita connection was derived from the
Euclidean connection (3.56), it has an independent meaning as a connection in its
own right. Taking it as the definition of the connection in terms of an arbitrary sym-
metric metric—tantamount to assuming a Riemannian geometry—and substituting into
(3.136), we do not conclude that the curvature vanishes. Indeed gravity is essentially
the same thing as curvature in the four-dimensional Riemannian space(-time) of gen-
eral relativity (Schrödinger, 1950; Wald, 1984). In a general Riemannian geometry we
naturally speak of the Riemann curvature tensor.
Before leaving this topic we note that we can define the “velocity” of a point on a
i i
curve with parametrization ξ (t) by gi ξ̇ . We denote this by ẏ, i.e.,
i
dy = gi dξ . (3.144)

Naturally we are curious to know whether this is integrable, that is, whether an asso-
i
ciated position field y(ξ ) exists. From all that has been said, we would expect this to
be answerable in the affirmative if the connection were of Weitzenböck type with zero
torsion, for we would then be back in Euclidean space. To confirm this we apply Stokes’
theorem, keeping in mind the skew symmetry of the wedge product:
Z Z Z Z
i j i j i
dy = gi dξ = gi,j dξ ∧ dξ = Γk[ij] gk dξ ∧ dξ . (3.145)
γ γ ω ω
 References 61

Therefore, dy is not exact, and a position field does not exist, if the connection (3.138)
possesses torsion. If the torsion vanishes, then dy is exact, and gi = y,i (see (3.2)).
We will see that this scenario describes a body containing a continuous distribution of
dislocations, the evolution of which in space and time is intimately connected to plastic-
ity in crystalline materials. In this context the left-most integral is the Burgers vector of
the distribution, and the torsion may then be interpreted as a dislocation density. This
observation effectively inaugurated the geometric school of material defects. The seminal
papers by Bilby et al., Eckart, Kondo, Kröner, Noll, and Wang, and the monograph by
Epstein, are essential reading for anyone interested in learning more about the founda-
tions of this beautiful subject. A comprehensive survey, written at the level of this book,
is given in the review article by Gairola.

References
Bilby, B. A., Bullough, R., and Smith, E. (1955) Continuous distributions of dislocations: A
new application of the methods of non-Riemannian geometry. Proc. Roy. Soc. Lond. A231,
263–73.
Brillouin, L. (1964). Tensors in Mechanics and Elasticity. Academic Press, New York.
Cartan, E. (2001). Riemannian Geometry in an Orthogonal Frame (transl. by V. V. Goldberg).
World Scientific, Singapore.
Ciarlet, P. G. (2005). An Introduction to Differential Geometry with Applications to Elasticity.
Springer, Dordrecht.
Eckart, C. (1948). The thermodynamics of irreversible processes. IV: The theory of elasticity and
anelasticity. Phys. Rev. 73, 373–82.
Epstein, M. (2010). The Geometrical Language of Continuum Mechanics. Cambridge University
Press, Cambridge, UK.
Flügge, W. (1972) Tensor Analysis and Continuum Mechanics. Springer, Berlin.
Gairola, B. K. D. (1979). Non-linear elastic problems. In: Nabarro, F. R. N. (Ed.), Dislocations in
Solids, Vol. 1. North-Holland, Amsterdam.
Green, A. E., and Zerna, W. (1968). Theoretical Elasticity. Oxford University Press, Oxford.
Kondo, K. (1955) Non-Riemannian geometry of imperfect crystals from a macroscopic view-
point. In: Kondo, K. (Ed.), RAAG Memoirs of the Unifying Study of Basic Problems in Engineering
Sciences by Means of Geometry, Vol. 1. Gakujutsu Bunken Fukyu-Kai, Tokyo.
Kröner, E. (1960) Allgemeine kontinuumstheorie der versetzungen und eigenspannungen. Arch.
Ration. Mech. Anal. 4, 273–334.
Lichnerowicz, A. (1962). Tensor Calculus. Methuen, London.
Lovelock, D., and Rund, H. (1989). Tensors, Differential Forms and Variational Principles. Dover,
New York.
Noll, W. (1967). Materially uniform simple bodies with inhomogeneities. Arch. Ration. Mech.
Anal. 27, 1–32.
Povstenko, Y. Z. (1991). Connection between non-metric differential geometry and the mathe-
matical theory of imperfections. Int. J. Engng. Sci. 29, 37–46.
Schouten, J. A. (1954). Ricci-Calculus. Springer, Berlin.
Schrödinger, E. (1950). Space-Time Structure. Cambridge University Press, Cambridge, UK.
Simmonds, J. G. (1994). A Brief on Tensor Analysis. Springer, New York.
62 A primer on tensor analysis in three-dimensional space

Sokolnikoff, I. S. (1964). Tensor Analysis: Theory and Applications to Geometry and Mechanics of
Continua. Wiley, New York.
Stoker, J. J. (1989). Differential Geometry. Wiley-Interscience, New York.
Szerkeres, P. (2004). A Course in Modern Mathematical Physics. Cambridge University Press,
Cambridge, UK.
Wald, R. M. (1984). General Relativity. University of Chicago Press, Chicago.
Wang, C.-C. (1967). On the geometric structure of simple bodies, a mathematical foundation for
the theory of continuous distributions of dislocations. Arch. Ration. Mech. Anal. 27, 33–94.
Willmore, T. J. (1959). Differential Geometry. Oxford University Press, Oxford.
4
Deformation and stress in convected
coordinates

Having surveyed the rudiments of tensor analysis and differential geometry in curvilinear
coordinates, it is appropriate and instructive to apply what we’ve learned to the analysis
of deformation and stress.

4.1 Deformation gradient and strain

To this end we will use the notation of the previous chapter to describe the geometry
of the Euclidean space in which the current configuration, κt , of a body resides. To
distinguish this from the Euclidean geometry of the reference configuration κ, we adopt a
different notation. Thus, in κ we denote the position field by x rather than y, as in Chapter
i i
1, and take it to be a function x(ξ ), where {ξ } is a set of right-handed curvilinear
coordinates. Inserting this into the deformation function (1.9), we have a parametrization
of κt in terms of the same coordinates:

i i
y(ξ , t) = χ(x(ξ ), t). (4.1)

We can now bring all the apparatus of Chapter 3 to bear on this position field at each
fixed time t. By using a single coordinate system ξ i to parametrize both κ and κt , we are
in effect taking the coordinates to convect with the deformation of the body. That is,
as in Chapter 3 we suppose the coordinates to be in one-to-one correspondence with
the position x, at least locally, and hence in one-to-one correspondence with a material
point of the body. In this way each material point is identified with a unique triplet
i
{ξ }, the elements of which then serve as its permanent labels. In particular, the material
i
derivatives ξ̇ vanish, and the material velocity—the time derivative of y holding x fixed—
is simply the partial time derivative in the convected coordinate formalism. A good way
to visualize this situation is to imagine drawing a curvilinear coordinate grid on a balloon.
Suppose a particular material point on the balloon has the coordinates (1, 2) in this (two-
dimensional) grid. As the balloon is inflated the grid deforms with it, and the label (1, 2)
continues to identify the same material point. The coordinate curves are thus embedded

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0004
64 Deformation and stress in convected coordinates

in the material; that is, they are material curves. The use of ξ i and t as independent
variables is therefore tantamount to a material, or Lagrangian, description of the motion.
Convected coordinates are not as widely used in continuum mechanics as they once
were. This is unfortunate, because they furnish a particularly useful tool for analysts who
have made the modest investment of effort, as we have, needed to learn the prerequi-
site tensor analysis. They are an indispensable tool in the study of rods and shells, for
example, regarded as one- and two-dimensional continua respectively. The books by
Sedov, Green and Zerna, and Green and Adkins are excellent sources of information on
the practical use of convected coordinates.
i
Using x(ξ ) we define the elements

ei = x,i (4.2)

of the referential tangent basis, and their reciprocals

i
ei = ∇ξ , (4.3)
i
where ∇ is the gradient with respect to x, i.e., ei = ∇ξ̂ (x). We then compute the
associated metric and reciprocal metric components

eij = ei · ej and eij = ei · e j , (4.4)

and the connection coefficients

m
Γ̄ij = em · ei,j = 12 emk (eki,j + ekj,i − eij,k ), (4.5)

etc., together with the counterparts of the various concepts discussed in Chapter 3. Here,
however, we use an overbar on the gammas and a vertical bar to denote the associated
covariant derivatives. For example, the gradient of v = vi ei with respect to x is

∇v = vk|i ek ⊗ ei , (4.6)

where
k
vk|i = vk,i + Γ̄ji vj . (4.7)

A convected-coordinate representation of the deformation gradient F defined in

Chapter 1 is easily derived. Fixing the time t, from (3.2) and (4.3) we have
i
Fdx = dy = gi dξ = gi (ei · dx) = (gi ⊗ ei )dx, (4.8)

and as this purports to hold for arbitrary dx, it follows that

F = gi ⊗ ei . (4.9)

This maps vectors in Tκ , the tangent space to κ, to vectors in Tκt , the tangent space
to κt , and is accordingly referred to as a two-point tensor. Using Ft = ei ⊗ gi , the right
 Deformation gradient and strain 65

Cauchy–Green deformation tensor is then given by

C = (Ft gi ) ⊗ ei = gij ei ⊗ e j . (4.10)

The covariant components of this tensor are thus none other than the components of
the metric induced by the coordinates in κt .

Problem 4.1 The contravariant components of C are not equal to g ij . What are they?

Problem 4.2 Show that eij are the contravariant components of the left Cauchy–Green
deformation tensor B = FFt .

It is conceptually helpful to distinguish the Euclidean tangent spaces Tκt and T κ .

Thus, we identify the unit tensor i, introduced in Chapter 3, as the spatial unit tensor that
maps any vector in Tκt to itself. In the same way, pursuant to the discussion in Chapter
1, let I be the referential unit tensor, mapping any vector in T κ to itself. Proceeding as in
(3.47), we have

I = ei ⊗ ei = eij ei ⊗ e j , (4.11)

together with the counterparts of the various other representations found there. From
(1.20) and (4.10) we then derive the Lagrange strain

E = Eij ei ⊗ e j , (4.12)

where

Eij = 21 (gij − eij ). (4.13)

This rather elegant formula indicates that the covariant components of the strain are just
one-half the difference of the metrics in κt and κ at the material point with coordinates
i
ξ . However, the contravariant components are not proportional to the difference of the
reciprocal metrics. Be aware that this situation is typical of the convected-coordinate
formalism: Results that hold for components of one kind—covariant, contravariant, or
mixed—are not valid for other kinds.
The determinant J F of the deformation gradient is given by the ratio

JF = Fe1 · Fe2 × Fe3 /e1 · e2 × e3 (4.14)

of scalar triple products. See the discussion in Chadwick excellent pocketbook, for
example. Now (4.9) implies that the numerator is simply g1 · g2 × g3 . In Chapter 3
66 Deformation and stress in convected coordinates

√ √
we showed that this is given by g, where g = det(gij ). In the same way, e1 · e2 × e3 = e,
where

e = det(eij ). (4.15)

Then,
p
JF = g/e. (4.16)

Thus, the relation (1.22) between the referential and spatial mass densities may be
expressed, in terms of convected coordinates, as
√ √
ρκ e = ρ g. (4.17)

In Chapter 3 we also discussed the oriented surface area measure given by (3.122),
i √
i.e., nda = ϵijk dζ dη j gk , where ϵijk = geijk . We interpret this as pertaining to a surface
ω ⊂ κt . Suppose this to be a material surface, so that its image Ω ⊂ κ consists of the
same material points. Recalling that the coordinates are convected with these points, we
conclude that the area measure on Ω is
i √
νdA = ϵ̄ijk dζ dηj ek , where ϵ̄ijk = eeijk , (4.18)

and ν is a unit-normal field. Accordingly, the covariant components of these area

measures, relative to their respective reciprocal bases, are related by

nk da = JF νk dA. (4.19)

Reattaching the relevant bases, we arrive at the Piola–Nanson formula (1.35),

nda = JF νk g k dA = JF g k (ek · ν)dA = F∗ νdA, (4.20)

where

F ∗ = JF g k ⊗ ek (4.21)

is the cofactor of the deformation gradient.

Problem 4.3 Show that F−t = gk ⊗ ek and hence that (4.21) is consistent with (1.36).

4.2 Strain compatibility

Later, in Chapter 5, we will devote considerable space to a discussion of strain com-
patibility. This entails a set of restrictions on the strain components arising from the
premise—a prejudice rooted in our terrestrial experience—that bodies always occupy
 Strain compatibility 67

k
regions of Euclidean space as they deform. The connections Γkij = gk ·gi, j and Γ̄ij = ek ·ei, j
are therefore torsionless and curvature-free. The strain compatibility conditions may
then be obtained by solving (4.13) for the metric gij in terms of the strain E ij , sub-
stituting the result into the Levi-Civita formula (3.103) for Γkij , and requiring that
the Riemannian curvature components (3.136) vanish, together with their referential
counterparts R̄k·mlj .
Before undertaking this arduous task it is worthwhile to first look for possible symme-
tries of the curvature arising from the Levi-Civita connection. To this end it is convenient
to introduce the fully covariant curvature

Rpmlj = gpk Rk·mlj . (4.22)

Then from (3.137) we have

Rpm(lj) = 0, (4.23)

whether or not the connection is Levi-Civita. We intend to show that if it is Levi-Civita,

then the major symmetry

Rpmlj = Rljpm (4.24)

obtains, so that, in addition,

R(pm)lj = 0. (4.25)

Accordingly, in this case we have the significant simplification

Rpmlj = R[pm][lj] . (4.26)

The demonstration of this important result is somewhat tedious. We’ll endeavor to

make it as brief and straightfoward as possible. To begin, we combine (3.136) and (4.22)
to obtain

Rpmlj = gpk (Γkmj,l + Γkil Γimj ) − gpk (Γkml,j + Γkij Γiml ), (4.27)

in which (3.97) is used to write

gpk Γkmj,l = (gpk Γkmj ),l − gpk,l Γkmj = Γpmj,l − gpk,l Γkmj . (4.28)

On combining this with (3.98), the first set of parentheses in (4.27) becomes

gpk (Γkmj,l + Γkil Γimj ) = Γpmj,l − Γimj Γipl , (4.29)

and application of a similar reduction to the second set of parentheses then yields

Rpmlj = Γpmj,l − Γljp,m + Γiml Γipj − Γimj Γipl . (4.30)

68 Deformation and stress in convected coordinates

We can now form the difference of the left- and right-hand sides of (4.24). Noting
that Γijp Γilm = Γijp Γilm = Γiml Γipj and that Γijm Γilp = Γimj Γipl , we find that in this difference
the products of the gammas cancel out, leaving only derivative terms

Rpmlj − Rljpm = Γpmj,l + Γljp,m − (Γpml,j + Γljm,p ). (4.31)

Substituting (3.101), we then have

1
2 (Rpmlj − Rljpm ) = (gpm,jl − gpm,lj ) + (glj,pm − glj,mp ) + (gjm,lp − gmj,pl )
+(gjp,ml − gjp,lm ) + (gml,pj − gml,jp ) + (gpl,jm − glp,mj ), (4.32)

in which each set of parentheses vanishes separately, provided, as we assume, that

the metric is twice differentiable. Thus, the symmetry condition (4.24) holds in
Riemannian—hence also in Euclidean—space, wherein the connection is Levi-Civita.
This major simplification leads to (4.26), and hence to the conclusion that the fourth-
order Riemann tensor stands in one-to-one relation to the second-order Einstein tensor
with contravariant components defined by

Πij = 14 ϵ ikl ϵ jmn Rklmn . (4.33)

To prove this claim we need only invert this relation to obtain

Rpqrs = ϵipq ϵjrs Πij . (4.34)

Observe that (4.26) is indeed satisfied.

Problem 4.4 Show that (4.33) implies (4.34). Hint: Make repeated use of (3.124).

Accordingly the components of the Riemann curvature vanish if and only if the Πij
vanish. Beyond this, from (4.24) and (4.33), and with a relabeling of summation indices,
we have

Π ji = 1 jkl imn
4ϵ ϵ Rklmn = 14 ϵ jkl ϵimn Rmnkl
1 jmn ikl
= 4ϵ ϵ Rklmn = Πij . (4.35)

Thus, the three independent components Π[ij] vanish identically, leaving six non-trivial
compatibility conditions. We can put these into another, arguably more convenient,
form in terms of the Ricci tensor, the second-order tensor with covariant components
defined by

Rij = Rk·jik . (4.36)

 Strain compatibility 69

The relationship between the Einstein and Ricci tensors may be established with the aid
of the identity (see Eq. 3.19 in Flügge’s book)

ϵ ijk ϵ lmn = g il (g jm g kn − g jn g km ) + g in (g jl g km − g jm g kl ) + g im (g jn g kl − g jl g kn ), (4.37)

yielding

mn − R··mn ) + g (R··mn − R··mn ) + g (R··mn − R··mn ), (4.38)

4Π il = ϵ ijk ϵ lmn Rjkmn = g il (R··mn nm in lm ml im nl ln

where R··ijkl = g jm R·imkl . Using R··ijkl = −R··jikl , which follows from (4.26), we reduce this
to

··mn + g R··mn − Rg ,
2Π il = g in Rlm im nl il
(4.39)

where

R = R··nm n
mn = R·n , (4.40)

in which R·ij = g ik Rkj , is called the scalar curvature. Raising indices in the usual way, we
kji lm·i
use R ij = R··· im nl nli il in lm mli
k with g R··mn = R···n (= R ) and g R··mn = R··m = R···m (= R ), which
il

follows from (4.26), finally reaching

Π il = R il − 12 Rg il . (4.41)

It is customary to take this as the definition of the Einstein tensor in n dimensions;

for example, in relativity theory (n = 4), whereas the formula (4.33) is valid in three
dimensions only.

Problem 4.5 Invert this relation to obtain Rij = Π ij − Πg ij , where Π = Π·nn and Π·ij =
gjk Π ik .

Problem 4.6 Show that the Einstein tensor is divergence-free, i.e., Π;ijj = 0. This is
a consequence of the Bianchi identities (see, for example, Section 86 of Lich-
nerowicz’ book, cited in Chapter 3). Incidentally, this result plays a decisive role in
Einstein’s theory—hence the name.

Thus, the relation between the Einstein and Ricci tensors is one-to-one. In three
dimensions the Riemann, Einstein, and Ricci tensors are therefore equivalent. Note that
we have established the symmetry of the Ricci tensor in three dimensions, so that the six
strain compatibility conditions may be expressed simply as

Rij = 0. (4.42)
70 Deformation and stress in convected coordinates

Symmetry in n dimensions is typically proved by using (3.136) and (4.36) to write

k
Rij = Γjk,i − Γji,k
k k m
+ Γmi Γjk − Γmk
k
Γjim , (4.43)

in which the i, j -symmetry of the second and fourth terms on the right-hand side is
k m
obvious. The symmetry of the third term follows from Γmi Γjk = Γkim Γjm
k
= Γkmj Γikm , and
k √
that of the first term from Γjk,i = (ln g),ji , which in turn follows from (3.107).
For the purpose of exhibiting the role of the strain in (4.42), it is convenient to express
the components Rij in terms of covariant derivatives with respect to the referential
k
connection Γ̄ij . To this end we introduce

k
S·kij = Γijk − Γ̄ij . (4.44)

This is the difference of two connections induced by a given system of (convected)

coordinates. Having solved Problem 3.16(a), we know that these are the components of
a third-order tensor.

Problem 4.7 Prove this.

Problem 4.8 Demonstrate that the symmetry of the Ricci tensor follows directly from
(4.23)–(4.25).

Equation (4.43) then furnishes

Rij = S·kjk,i − S·kji,k + S·kmi S·mjk − S·kmk S·mji

m m k k
+S·kmi Γ̄jk − S·kmk Γ̄ji + S·mjk Γ̄mi − S·mji Γ̄mk + R̄ij, (4.45)

where R̄ij —the components of the referential Ricci tensor—are obtained by substituting
k
Γ̄ij in place of Γijk . They vanish, of course, because the geometry of κ is Euclidean, but
the equation remaining is still rather formidable. To simplify it we use an old geometer’s
k
trick: We choose the coordinates ξ i to be Cartesian in κ, so that the Γ̄ij vanish, and with
them the second line in (4.45). The first two terms, involving partial derivatives of the
S’s, are then the Cartesian representations of the corresponding referential covariant
derivatives. Because the first line involves tensors exclusively, we can then use the tensor
transformation rules established in Chapter 3 to conclude that

Rij = S·kjk|i − S·kji|k + S·kmi S·mkj − S·kmk S·mji (4.46)

in arbitrary coordinates.
To achieve our aim we must represent the Sk·ij in terms of the strain components. Using
the geometer’s trick, we can again select Cartesian coordinates in κ to reduce (4.44) to
 Stress, equations of motion 71

S·kij = Γijk , with Γijk given by (3.103). Reverting to general coordinates and invoking Ricci’s
lemma in the form eij|k = 0, we obtain

S·kij = gkl (Eli|j + Elj|i − Eij|l ). (4.47)

Problem 4.9 It is not immediately obvious that the right-hand side of (4.46) possesses
the required symmetry with respect to interchange of i and j. The symmetry of the
second and fourth terms is obvious, while that of the third follows from S·kmi S·mkj =
S·mki S·kmj = S·kmj S·mki . As for the first term, establish that

S·kjk = (JF ),j /JF ,

and use this to show that S·kjk|i = S·kik|j .

Problem 4.10 Obtain the strain compatibility conditions by substituting (4.47) into
(4.46). Along the way you will encounter terms like g|jm i . Show that g|i
jm
=
−g g gkl|i = −2g g Ekl|i and use this in Rij = 0 to derive the long-awaited strain
jk ml jk ml

compatibility equations. These indicate that the strain field cannot be arbitrary, but
instead that it satisfies certain differential constraints.

4.3 Stress, equations of motion

Moving on to stress, recall that the Cauchy stress T maps spatial vectors—the orienta-
tion n of a surface in κt , for example—to other spatial vectors; for example, the traction
t acting on the tangent plane to the surface. Accordingly we decompose T in bases
constructed from the gi . One such decomposition is

T = T ij gi ⊗ gj , (4.48)

in terms of contravariant components. We then find that (1.32) and (1.34) are equivalent
to

T;jij + ρb i = ρa i and T ij = T ji , (4.49)

respectively, where b i = b · g i are the body-force components, Tij;j is the covariant

divergence based on the Levi-Civita connection (3.103), and of course a i = v̇ · g i are
the components of the acceleration in which v = ẏ is the material velocity. In particular,

ai = g i · (v j gj )· = g i · (v̇ j gj + v j ẏ,j )
= g i · (v̇ j gj + v j v,j ) = v̇i + vi;j v j , (4.50)
72 Deformation and stress in convected coordinates

where we have used the independence of ξ i and t to commute the coordinate and time
derivatives (ẏ),j = (y,j )· , this being permissible if the velocity is twice differentiable with
respect to the coordinates and time jointly. The last of these equalities is the well-known
formula for the acceleration in terms of the convective derivative of the velocity. This is
often a difficult concept for the beginning student, whereas the result is immediate in
the convected-coordinate framework.
Combining (4.21) and (4.48), we find that the Piola stress P, defined by (1.38), admits
the representation

P = P i ⊗ ei , where P i = JF T ji gj . (4.51)

This, like the deformation gradient, is a two-point tensor. Therefore the Piola–Kirchhoff
stress S, defined in (1.42), has the representation

S = S ij ei ⊗ ej , where S ij = JF T ij , (4.52)

the latter holding for contravariant components only. Further, the symmetry condition
(4.49)2 —equivalent to T ij gi × gj = 0 —imposes the restriction

gi × P i = 0 (4.53)

on the Piola stress vectors P i .

Problem 4.11 Prove that S ij = JF T ij , as claimed.

Problem 4.12 Derive the Doyle–Ericksen formula

∂W
JF T ij = 2
∂gij

for hyperelastic materials, where W is the strain-energy function.

Problem 4.13 Show that P i = Pe i and hence that Pi is proportional to the traction
i
exerted on a surface where ξ , with i ∈ {1, 2, 3} fixed, is constant.

Using the referential counterparts of (3.80) and (3.107), the divergence of the Piola
stress occurring in (1.40) is found to be

DivP = (P),j e j = (P i ⊗ ei ),j e j = (P,ji ⊗ ei + P i ⊗ ei,j )e j

j √
= P,ii + Γ̄ij P i = √1e ( ePi ),i . (4.54)

In the same way, from

T = T i ⊗ gi , where T i = T ji gj (4.55)
 References 73

are the Cauchy stress vectors, we have

1 √
√ i
divT = g ( gT ),i . (4.56)

Substituting this into (1.32) and multiplying the result by JF , with the aid of (4.16) and
(4.51)2 we immediately arrive at (1.40): The convected-coordinate formalism facilitates
a trivial derivation of the Piola equations of motion from the Cauchy equations, and vice
versa. The conventional derivation—see Chadwick’s book, for example—is based on the
formula

Div(TF∗ ) = JF divT, (4.57)

which in turn is a consequence of the Piola identity DivF∗ = 0.

Problem 4.14 Prove this assertion.

It’s particularly easy to establish the Piola identity by using convected coordinates.
Thus, we combine (4.16), (4.21), (4.51)1 and the formula (4.54) for the divergence,
with Pi replaced by JF gi :
√ JF √
DivF∗ = 1
√
e
( eJF g i ),i = √ i
g ( gg ),i
√
( g),k
= JF { √
g − Γkii }g k . (4.58)

This vanishes by (3.107).

References
Chadwick, P. (1976). Continuum Mechanics: Consise Theory and Problems. Dover, New York.
Doyle, T. C. and Ericksen, J. L. (1956). Nonlinear elasticity, in: Advances in Applied Mechanics IV.
Academic Press, New York.
Flügge, W. (1972) Tensor Analysis and Continuum Mechanics. Springer, Berlin.
Green, A. E. and Adkins, J. E. (1970). Large Elastic Deformations. Oxford University Press,
Oxford.
Green, A. E. and Zerna, W. (1968). Theoretical Elasticity. Oxford University Press, Oxford.
Sedov, L. I. (1966). Foundations of the Non-linear Mechanics of Continua, trans. by Schoenfeld-
Reiner, R., ed. by Adkins, J. E., and Spencer, A. J. M. Pergamon Press, Oxford.
5
Elastic and plastic deformations

Recall the observation made in Chapter 1 to the effect that the elastic properties of a
crystalline solid are largely insensitive to plastic deformation. This is a manifestation
of the fact that plastic deformation is accommodated by the relative slip of adjacent
crystallographic planes while the underlying crystalline lattice remains intact. This
means that there exists an undistorted state of the material—the seat of the crystalline
symmetry—that evolves with plastic deformation. We attempt to codify this notion in
the present chapter, and thereby to construct the rudiments of a theory of elastic-plastic
deformation.

5.1 Elastic–plastic deformation, dislocation density

With reference to the discussion in Section 2.4, let gκi (p) be the symmetry group relative
to a local undistorted state κi (p) of the material point p; thus, gκi (p) ⊂ Orth+ . This is
meaningful whether or not the solid is crystalline. We regard κi (p) as an evolving local
configuration in accordance with the foregoing comments. To acknowledge conventional
terminology, the subscript on κi (p) stands for “local intermediate configuration.”
Noll’s rule—see (2.100)—then yields the symmetry group at the material point p in
a global reference configuration κ,

gκ(p) = {KRK−1 | R ∈ gκi (p) }, (5.1)

where K(x, t), in which x = κ(p) the position of p in κ, is the map from the local configu-
ration κi (p) to the tangent space Tκ (p) at p. Recalling the discussion surrounding Noll’s
rule in Section 2.4, K(x, t) is in general not a gradient field. The inverse map,

G = K−1 , (5.2)

taking Tκ (p) to κi (p), is commonly called the plastic deformation. It is the generalization to
three dimensions of the plastic stretch λp discussed in Chapter 1, and is typically denoted
by Fp in most of the literature.

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0005
 Elastic–plastic deformation, dislocation density 75

Note that we are tacitly interpreting κi (p) as a vector space rather than a configuration
in the usual sense. Later, we will interpret κi (p) as the tangent space to a certain non-
Euclidean manifold associated with the material point p.
In the same way we can contemplate a map H from κi (p) to the tangent space Tκt (p),
at p, to the current global configuration κt of the body B. This map is the generalization
of the elastic stretch λe mentioned in Chapter 1. It represents the elastic distortion of
the lattice, at the point p, required to place it into κt , and is more often denoted by Fe .
Identifying p with x = κ(p) as before, we denote this elastic deformation by H(x, t). Being
defined on κi (p) rather than on a neighborhood of p in an actual configuration, this too
is generally not a gradient field.
Of course there is another map, F(x, t), from Tκ (p) to Tκt (p). This is the deformation
gradient ∇χ(x, t), the gradient at x of the deformation field χ(·, t) mapping the global
configuration κ to the global configuration κt . Because Tκ (p) is the result of mapping
κi (p) by K, whereas Tκt (p) is the map of κi (p) by H, it follows that

H = FK, where F = ∇χ. (5.3)

This, of course, is equivalent to F = HG —just another notation for the famous decom-
position F = Fe Fp —generalizing the formula λ = λe λp of Chapter 1. Note, however, that
whereas it is true that λ = λp λe , it is not true that F = GH, for the simple reason that tensor
multiplication does not commute. An alternative explanation is that Tκt (p) is the range of
H, but not the domain of G, implying that the proposed product is not well defined. As
usual we suppose that κ is occupiable, i.e., that JF > 0. Assuming κi (p) to be occupiable,
too, we also have JK > 0 and JH > 0, the former implying that JG (= J− 1
K ) > 0.
To better understand what’s going on in this decomposition we note that the undis-
torted state to which we have repeatedly referred is the state used by crystallographers to
characterize crystal symmetry. This is normally regarded as a state of minimum energy
of the atomic arrangement constituting the crystal in which the lattice is unstressed. How-
ever, a local undistorted state κi (p) defined by the condition gκi (p) ⊂ Orth+ may support
a state of stress whose form is dictated by its symmetry. For example, cubic crystals and
isotropic materials may support a pure pressure while occupying an undistorted state.
It is common, however, to assume that κi (p) is stress free, to promote the analogy to the
stress-free configuration of the uniaxial bar of Figure 1.1, achieved by unloading from a
state that has undergone prior plastic deformation. However, it is not generally possible
to have a state of vanishing stress at all points of the body. Usually there is a distribution
of residual stress due to the presence of various defects such as the dislocations accompa-
nying plastic slip. These induce local lattice distortions in the case of crystalline metals,
for example, which in turn generate elastic strain and a consequent distribution of stress
(Figure 5.1). This is typically the case even when the body is entirely unloaded, i.e.,
when no body forces are applied and the boundary tractions vanish.
Nevertheless it is possible, in principle, to remove the mean stress via an equilibrium
unloading process. In particular, in equilibrium the mean value, T̄, of the Cauchy stress
76 Elastic and plastic deformations

Figure 5.1 Local lattice distortion in the vicinity of an edge

dislocation in a cubic crystal.

T is given by
Z Z
vol(κt )T̄ = Sym{ t ⊗ yda + ρb ⊗ ydv}, (5.4)
∂κt κt

where t and b, respectively, are the boundary traction and body force and y is the position
of a material point in the current configuration κt of the body. See Chadwick’s book for
a simple derivation based on (1.32). Accordingly, T̄ = 0 if the entire body is unloaded.
The mean-value theorem for continuous functions guarantees the existence of ȳ ∈ κt
such that T(ȳ, t) = T̄. Therefore, T(ȳ, t) = 0 for some ȳ ∈ κt if the body is unloaded and
in equilibrium. Let

d(κt ) = sup |y − z| . (5.5)

y,z∈κt

This is the diameter of κt . Then for every y ∈ κt we have T(y, t) → T(ȳ, t) as d(κt ) → 0.
Accordingly, the local value of the stress is made arbitrarily close to zero as the diameter
of the body shrinks to zero.
Of course, it is not possible to reduce the diameter of a given body to zero. However,
we may regard any body as the union of an arbitrary number of arbitrarily small disjoint
sub-bodies πt , i.e., κt = ∪∞
(n) (n) (n)
n=1 πt , with d(πt ) → 0. Imagine separating these sub-
(n)
bodies and unloading them individually. We then have T(y, t) → 0 for every y ∈ πt , for
(n)
every n. Because every y in κt belongs to some πt , this process results in a state in which
(n)
the material is destressed pointwise. We expect that each piece πt will experience some
distortion in this process, and therefore that the unstressed sub-bodies cannot be made
congruent to fit together into a connected region of three-dimensional Euclidean space.
Thus, there is no global stress-free configuration of the body, and hence no position
field r, say, such that dr = Gdx (or H−1 dy); that is, there is no neighborhood in the
vanishingly small unloaded sub-bodies that can be used to define a gradient of a position
 Elastic–plastic deformation, dislocation density 77

field. Accordingly, as we have already noted, neither G nor H is a gradient. It follows that
for any closed curve Γ ⊂ κ the vector
Z
BΓ = Gdx (5.6)
Γ

does not vanish. This is called the Burgers vector associated with Γ. In the same way we
can define
Z
bγ = H−1 dy, (5.7)
γ

not to be confused with the body force, where γ = χ(Γ, t) is the image in κt of Γ under
the deformation map.
(n)
Problem 5.1 Show that if the tractions acting on the disjoint sub-bodies πt vanish,
(n)
then they become stress free in the limit d(πt ) → 0 whether or not they are in
equilibrium, provided that the norm of ρ(b − v̇), where b is the body force, is
bounded.

Problem 5.2 Show that the Burgers vectors bγ and BΓ are one and the same.

Consider a simply connected surface Ω ⊂ κ. On combining (5.6) with Stokes’ formula

(3.120), applied in κ, we have the associated Burgers vector
Z
B(Ω, t) = αtκ νdA, (5.8)
Ω
where ν is a unit normal to Ω and

ακ = Curl K−1 (5.9)

is the referential dislocation density tensor in which Curl is the curl based on x ∈ κ. This
presumes the plastic deformation to be smooth and hence continuous. We will revisit
this formula for discontinuous plastic deformation fields in Chapter 6.
In the same way we have
Z
B(Ω, t) = b(ω, t) = αtκt nda, (5.10)
ω

where ω = χ(Ω, t) is the image of Ω in κt with unit-normal field n, related to ν by the

Piola–Nanson formula (1.35), and

ακt = curl H−1 (5.11)

is the spatial dislocation density tensor in which curl is based on y ∈ κt .

78 Elastic and plastic deformations

The first equality in (5.10) suggests that these dislocation densities are not indepen-
dent. We establish their relationship via a somewhat circuitous argument. Thus consider
two global reference configurations κ1 and κ2 , say. Let λ be an invertible differentiable
map from the first to the second, so that the position of a material point p in κ2 is
x2 = λ(x1 ), where x1 is its position in κ1 . Let R = ∇1 λ be its gradient with respect to
x1 . We hold κt and κi (p) fixed in this discussion, so that H is unaffected. Then, because
K−1 dx = H−1 dy in which the right-hand side is fixed, it follows, using obvious notation,
that

K−1 −1
1 dx1 = K2 dx2 . (5.12)

From dx2 = Rdx1 we then have

K2 = RK1 . (5.13)

If Ω1 is a simply connected surface in κ1 , then, with Ω2 = λ(Ω1 ),

Z Z Z Z
(Curl2 K−
2
1 t
) ν 2 dA 2 = K −1
2 dx 2 = K −1
1 dx 1 = (Curl1 K−1 t
1 ) ν1 dA1 .
Ω2 ∂Ω2 ∂Ω1 Ω1
(5.14)
The outer equality may be reduced, with the aid of the Piola–Nanson formula

ν2 dA2 = R∗ ν1 dA1 , (5.15)

to
Z Z
(Curl2 K−1 t ∗
2 ) R ν1 dA1 = (Curl1 K−1 t
1 ) ν1 dA1 . (5.16)
Ω1 Ω1
Because Ω1 is arbitrary—and hence also ν1 —it follows that

(Curl2 K−1 t ∗ −1 t
2 ) R = (Curl1 K1 ) , (5.17)

which may be rearranged as

JR Curl2 K−1 −1
2 = R(Curl1 K1 ); (5.18)

equivalently,

JK2 K−1 −1 −1 −1
2 Curl2 K2 = JK1 K1 Curl1 K1 . (5.19)

It follows that

α = JK K−1 Curl K−1 (5.20)

furnishes a measure of dislocation density that is intrinsic in the sense of being insensitive
to the reference configuration. For this reason it is often called the true dislocation density.
 Differential-geometric considerations 79

Alternatively, we may proceed from (5.10), or use a formula like (5.18), with obvious
notational adjustments, to establish that

α = JH H−1 curlH−1 . (5.21)

Following essentially the same reasoning as that leading to (5.19), we conclude that α
is insensitive to arbitrary invertible differentiable variations of κt , and hence that it is
invariant under superposed rigid-body deformations in particular. Finally, using (5.9),
(5.11), (5.20), and (5.21) yields the relation

JF ακt = Fακ (5.22)

connecting the spatial and referential dislocation densities. It follows from this that ακt
is actually a two-point tensor.
We have noted that κi (p) should be interpreted as a vector space. It may be regarded as
the tangent space to a certain body manifold, but this manifold is not Euclidean as it does
not support a position field. This interpretation is the basis of an elegant differential-
geometric theory of plastically deformed bodies, which we pause to outline in some
detail.

5.2 Differential-geometric considerations

As in Chapter 4 we use convected coordinates ξi to identify the material point p. Recall
that these induce the linearly independent and positively oriented coordinate tangent
basis vectors ei = x,i in Tκ . We use these with the plastic deformation tensor G to define
vectors mk ∈ κi (p) by

mi = Gei . (5.23)

These too are linearly independent and positively oriented because J G is positive; thus,
{mi } is a basis for κi (p). Then, as in (4.9),

G = m i ⊗ ei , (5.24)

and the plastic Cauchy–Green deformation tensor is

Gt G = mij ei ⊗ ej , where mij = mi · mj (5.25)

is the metric induced in κi (p) by the coordinates. As this is positive definite, so too is
the matrix (mij ). Accordingly m > 0, where m = det(mij ), and we have well-defined
reciprocal basis elements mi = mij mj , where (mij ) = (mij )−1 .
80 Elastic and plastic deformations

With these in hand we define the Weitzenböck connection

k
Γ̂ij = mk · mi,j , (5.26)

as in (3.138). Proceeding exactly as in the passage from the latter to (3.142), we conclude
that

R̂k·mlj = 0, (5.27)

where

k k k
R̂k·mlj = 2(Γ̂m[ j,l ] − Γ̂i[ j Γ̂|m|l ] ) (5.28)

is the associated curvature. Thus, the geometry with connection Γ̂ is flat. The vanishing
of the curvature tensor constitutes the integrability condition for the system

k
mi,j = Γ̂ij mk , (5.29)

which follows from (5.26), ensuring the existence of the fields mi in any path-connected
domain, this of course being an immediate consequence of (5.23). Further, (5.25)2 and
k
(5.29) imply that Γ̂ij is compatible with the metric mij in the sense that

l
mij,k = Γ̂jik + Γ̂ijk , where Γ̂jik = mlj Γ̂ik . (5.30)

This is Ricci’s lemma. That is, the covariant derivative of the metric with respect to the
connection vanishes—see (3.94).
The torsion induced by this connection is

k
T̂·kij = Γ̂[ij] = mk · m[i,j] , (5.31)

where

k
mi,j = G,j ei + Gei,j = G ,j ei + Γ̄ij Gek
k
= G,j ei + Γ̄ij mk , (5.32)

in which Γ̄ is the (symmetric) Levi-Civita connection induced by the coordinates in κ.

Thus, m[i,j] = G,[j ei] , and as this is generally non-zero, it follows that the torsion does not
vanish; the geometry based on Γ̂ is not Euclidean.
 Differential-geometric considerations 81

Exceptionally, if G were a smooth gradient field, then we would have mi = r,i for
some twice differentiable field r, say. In this event we would have

l
G,j = (r,kj − Γ̄kj r,l ) ⊗ ek , (5.33)

giving m[i,j] = r,[ij] , which vanishes identically, together with the torsion. The underlying
geometry would then be Euclidean. This subsumes homogeneous plastic deformations
for which G is spatially uniform, in which case (5.29) and (5.32) imply that Γ̂ = Γ̄.
k
In view of (5.30) we can apply (3.99) to the metric mij and connection Γ̂ij to obtain

l l l
1
2 (mki,j + mkj,i − mij,k ) = mli Γ̂[kj] + mlj Γ̂[ki] + mlk Γ̂(ij) . (5.34)

Let {m
ij }(= {ji }) be the Levi-Civita connection based on the same metric, i.e.,
m

{m
ij } = 2 m (mki,j + mkj,i − mij,k ).
1 mk
(5.35)

The curly bracket is a rather old-fashioned notation for the connection. We resurrect it
here to avoid introducing too many Γ′s. Equation (5.34) is then equivalent to
m
{m
ij } = (T̂·kj mli + T̂·ki mlj )m
l l mk
+ Γ̂(ij) . (5.36)

m m m m m
Solving for Γ̂(ij) and substituting into the identity Γ̂ij = Γ̂(ij) + Γ̂[ij] = Γ̂(ij) + T̂·ijm , we
conclude that
k
Γ̂ij = {kij } − K·ijk , (5.37)

where

K·ijk = (T̂·njl mli + T̂·nil mlj )mkn − T̂·ijk (5.38)

is the contortion tensor. Clearly K·ijk vanishes if T̂·ijk vanishes. Conversely, if K·ijk vanishes,
k
then T̂·jik (= K·[ij] ) vanishes.
The torsion T̂·ijk is equivalent to the true dislocation density, which may be written in
the form

α = J−1
G GCurlG. (5.39)

To establish this recall that the curl is defined by

(CurlG)c = Curl(Gt c), (5.40)

82 Elastic and plastic deformations

for any fixed vector c. Writing c = ci mi and using (5.24) in the form Gt = ei ⊗ mi , we
have Gt c = ci ei , and the referential form of (3.117) yields

Curl(Gt c) = ϵ̄ijk cj|i ek , (5.41)

l √
where cj|i = cj,i −cl Γ̄ji and ε̄ijk (= eijk / e) are the components of the referential permutation
l l
tensor. Because ε̄ijk = ε̄[ij]k , whereas Γ̄ji = Γ̄ij , this simplifies to

Curl(Gt c) = ϵ̄ijk c[j,i] ek = ϵ̄ijk ek (m[j,i] · c) = (ϵ̄ijk ek ⊗ m[j,i] )c, (5.42)

and (5.40) furnishes

CurlG = ϵ̄ijk ek ⊗ m[j,i] . (5.43)

Then (5.29) and (5.39) combine to give

α = αij mi ⊗ mj , (5.44)

where

αij = ϵ̂kli T̂·lkj , (5.45)

√ p
in which ε̂ijk = eijk / m, with m = det(mij ), and use has been made of JG = m/e.

Problem 5.3. Use the relevant version of (3.124) to invert (5.45), and thus show that

l
T̂·nm = 12 ϵ̂mnk αkl .

These results make clear the fact that α is equivalent to the torsion and reinforce our
earlier conclusion that it is intrinsic; that is, independent of κt or κ.

5.3 Incompatibility of the elastic strain

The elastic strain is

Ê = 12 (Ht H − I), (5.46)

where, from (4.9), (5.2), and (5.24),

H = FK = (gi ⊗ ei )(ej ⊗ mj ) = δij gi ⊗ mj = gi ⊗ mi . (5.47)

Thus,

Ê = Êij mi ⊗ mj , (5.48)
 Incompatibility of the elastic strain 83

where

Êij = 12 (gij − mij ). (5.49)

We intend to show that this strain is incompatible if, as is typically the case, the torsion
T̂k·ij is non-zero. This means that elastic strain cannot vanish in the presence of a density
of dislocations. Because elastic strain gives rise to stress, as we shall see in Chapter 6,
it follows that dislocations indirectly generate a stress field. Thus, a residual stress field
persists even in the absence of any loading applied to the body. To demonstrate this we
solve (5.49) for the metric mij , substitute into (5.37), and impose the flatness conditions
R̂k·mlj = 0 and Rk·mlj = 0. Here R is the Riemann curvature of Euclidean space based on
the Levi-Civita connection Γ, computed from the metric gij and induced in κt by the
coordinates. However, before proceeding we dispose of some preliminaries.
First, we recall the condition (3.137) satisfied by all curvature tensors. In particular,

R̂pm(lj) = 0, (5.50)

where—see (3.97) and (4.30)—

i i
R̂pmlj = mpk R̂k·mlj = Γ̂pmj,l − Γ̂pml,j + Γ̂ml Γ̂ipj − Γ̂mj Γ̂ipl . (5.51)

In fact, we also have

R̂(pm)lj = 0. (5.52)

Proof : It follows from (5.51) that

R̂pmlj + R̂mplj = Γ̂pmj,l + Γ̂mpj,l − (Γ̂pml,j + Γ̂mpl,j )

i i i i
+Γ̂ml Γ̂ipj + Γ̂pl Γ̂imj − (Γ̂mj Γ̂ipl + Γ̂pj Γ̂iml ). (5.53)

i n i n i
But Γ̂pl Γ̂imj = min Γ̂mj Γ̂pl = Γ̂npl Γ̂mj = Γ̂mj Γ̂ipl , implying that the second line vanishes. Using
(5.30) we then reduce this to

R̂(pm)lj = mmp,[jl] , (5.54)

and thus establish (5.52), assuming the metric to be a twice-differentiable function.

84 Elastic and plastic deformations

Taken together with (5.50), this implies that

R̂pmlj = R̂[pm][lj] . (5.55)

We can then proceed as in the derivation of (4.33) and (4.34) to conclude that R̂ stands
in one-to-one relation to Π̂, defined by
ij
Π̂ = 14 ϵ̂ikl ϵ̂ jmn R̂klmn , (5.56)

with inverse
ij
R̂pqrs = ϵ̂ipq ϵ̂jrs Π̂ , (5.57)
√
where ϵ̂ijk = meijk . Accordingly, (5.27) is equivalent to

ij
Π̂ = 0. (5.58)

However, recalling that the derivation of the major symmetry condition (4.32) relied on
i
a Levi-Civita connection, it would appear that we do not have R̂pmlj = R̂ljpm because Γ̂jk
ij ji [ij]
is not Levi-Civita. Apparently, then, Π̂ ̸= Π̂ . But, in fact, it turns out that Π̂ = 0,
implying that (5.58) represents only six restrictions, as in the discussion (in Section 4.2)
about compatibility of the overall strain.

Problem 5.4 Prove the identity Div[(CurlA)t ] = 0, and establish the equivalence of the
following statements:

(a) Div[(CurlG)t ] = 0, where G is the plastic part of the deformation gradient.

n
(b) ϵ̂ijk (T̂.ji,k
n
+ T̂.jil Γ̂lk ) = 0, where ϵ̂, T̂, and Γ̂, respectively, are the permutation tensor,
the torsion tensor, and the connection in the intermediate state.
(c) ϵ̂mlj R̂kmlj = 0, where R̂ is the curvature tensor based on Γ̂.
[ij]
(d) Π̂ = 0.
Thus, the equations in (b)–(d) are identities.
Returning to the matter at hand, we follow a procedure similar to that used in Section
4.2 and define a tensor with components

Ŝ·kij = Γkij − {kij }. (5.59)

Then (5.37) may be expressed in the form

k
Γ̂ij = Γkij − Lk·ij , where Lk·ij = Ŝ·kij + K·kij , (5.60)
 Incompatibility of the elastic strain 85

and (3.136) yields

R̂k·mlj = Rk·mlj + Γkij Li·ml − Γkil Li·mj + Lk·ij Γiml − Lk·il Γimj
+Lk·ml,j − Lk·mj,l + Lk·il Li·mj − Lk·ij L·i ml , (5.61)

where R is the Riemann tensor in κt . Recalling the geometer’s trick described in Section
4.2, we choose the convected coordinates to be Cartesian in κt . This is always possible
at any fixed instant t, though, of course, the coordinates, insofar as they convect with
the deformation, will generally not be Cartesian at earlier and later instants. This choice
means that the Γ′s vanish instantaneously. Because R̂ and R vanish at all times, and hence
at the instant in question, (5.61) reduces to

Lk·ml,j − Lk·mj,l + Lk·il Li·mj − Lk·ij Li·ml = 0. (5.62)

Reverting to general coordinates we then have the tensor equation

Lk·ml;j − Lk·mj;l + Lk·il Li·mj − Lk·ij Li·ml = 0, (5.63)

where the semi-colon is the covariant derivative based on the connection Γ. This is valid
at all times because the coordinates are no longer required to be Cartesian.

Problem 5.5 Establish the formula

Ŝ·kij = mkm (Êmi;j + Êmj;i − Êij;m ).

This result may be combined with (5.60)2 and (5.63) to arrive at a set of differential
constraints involving the elastic strain Ê, the metric m, and the contortion K, which gen-
erally preclude the possibility that the elastic strain vanishes identically in the body. To
see this we investigate the implications of the assumption that the Êij vanish together
with their derivatives. Then the Ŝ·kij vanish and the Lk·ij coincide with K·kij , reducing
(5.63) to
k
Km[j,l] − Ki[k j K|im|l ] = 0, (5.64)

where the coordinates have again been taken to be Cartesian in κt , with Γkij = 0, for the
sake of simplicity. This, in turn, yields the three restrictions

ϵ̂ jml (Kmj,l
k
− Kijk Kml
i
) = 0. (5.65)

Compare this to the identity proved in Problem 5.4(b), which implies, for vanishing
Ŝk·ij and again with Γkij = 0, that

ϵ̂ jml (T̂mj,l
k
− Kijk T̂ml
i
) = 0. (5.66)
86 Elastic and plastic deformations

i i
If we add the last two equations, and note, from (5.38), that Kml + T̂ml is symmetric in the
subscripts, we find that the resulting equation is identically satisfied. Thus, (5.65) is also
an identity. However, there are further restrictions contained in (5.64)—equivalent to
(ij)
Π̂ = 0 —that are not accounted for by (5.65), and these impose additional constraints
on the manner in which the torsion tensor can depend on the coordinates. In the general
case there is no reason why these extra constraints should be satisfied. Thus, the premise
is false: the elastic strain field cannot be identically zero in the presence of a general
non-zero torsion field.

References
Cermelli, P., and Gurtin, M. E. (2001). On the characterization of geometrically necessary
dislocations in finite plasticity. J. Mech. Phys. Solids 49, 1539–68.
Chadwick, P. (1976). Continuum Mechanics: Concise Theory and Problems. Dover, New York.
Gairola, B. K. D. (1979). Non-linear elastic problems. In: Nabarro, F. R. N. (Ed.), Dislocations in
Solids, Vol. 1. North-Holland, Amsterdam.
Rajagopal, K. R., and Srinivasa, A. R. (1998). Mechanics of the inelastic behavior of
materials—part 1: Theoretical underpinnings. Int. J. Plasticity 14, 945–67.
6
Energy, stress, dissipation, and plastic
evolution

In this chapter we outline the basic constitutive theory for energy, stress, and the dis-
sipation accompanying plastic evolution. We extend the notion of dislocation density
to accommodate surface dislocation and discuss its role in relaxing the stress in adjoin-
ing crystal grains and at phase interfaces, and further results concerning the dissipation
associated with the evolution of the latter are given. Also discussed is the notion of yield
and the role it plays in the flow rule for plastic evolution, and the essential distinction
between crystallinity and isotropy in the basic formulation of the theory.

6.1 Materially uniform elastic bodies

Our basic conceptual framework is the theory of materially uniform simple materials
pioneered by Noll, which posits the existence of a strain-energy function, and associ-
ated stress, given by constitutive functions that are the same for all material points. This
furnishes an appropriate model for a uniform single crystal, for example. Crystalline
aggregates may then be viewed as composites consisting of individual materially uni-
form crystal grains. This is adequate for most applications involving metal plasticity,
and also furnishes a foundation for theories of non-crystalline solids.
For example, with reference to (2.36) the strain energy contained in a part πt ⊂ κt is
Z
U= ψ(H)dv, (6.1)
πt

in which the constitutive function ψ(H) for the energy per unit current volume is
assumed to be the same function of H at all p, implying that any dependence on the latter
occurs implicitly, via the field H(x, t). While the restriction to uniform functions is not
essential, what is essential is the assumption that energy arises in response to distortion
of the crystal lattice, as represented by H. This assumption effectively embodies G. I.
Taylor’s observations to the effect that elastic properties, encoded in the undistorted lat-
tice, are insensitive to plastic deformation. We extend the same constitutive assumption
to non-crystalline materials.

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0006
88 Energy, stress, dissipation, and plastic evolution

Proceeding as in (2.37), the strain energy per unit volume of a reference configuration
κ is

Ψ = JF ψ. (6.2)

It proves advantageous to introduce an energy W defined by

Ψ = JG W. (6.3)

This may be viewed as the strain energy “per unit volume of κi (p)”, although κi (p) is
a vector space rather than a configuration. That is, even though we cannot associate a
volume with κi (p), J G is nevertheless well defined. Eliminating Ψ and using J−1
G JF = JH ,
we conclude that W is the function of H given by

W(H) = JH ψ(H). (6.4)

Note that W plays the same role for H as that played by Ψ for F in Chapter 2, and in
fact the two functions coincide when K = I, i.e., when F = H. Accordingly it will serve
as our basic constitutive function. From (5.2) and (5.3), the energy per unit volume of
κ is

Ψ(F,K) = J−1
K W(FK). (6.5)

Evidently, then, at fixed t a plastic deformation field K(x) gives rise to a strain-
energy function for a conventional non-uniform elastic solid of the kind encountered
in Chapter 2. Recalling the requirement (2.24) that such functions be invariant under
the replacement F → Q̄F for any rotation Q̄, we thus have

J−1 −1
K W(FK) = Ψ(F,K) = Ψ(Q̄F,K) = JK W(Q̄FK), (6.6)

or simply

W(H) = W(Q̄H), (6.7)

and this holds at all t as W does not depend explicitly on t. Then, exactly as in the
derivation of (2.27),

W(H) = U(Ê), where Ê = 12 (Ht H−I) (6.8)

is, of course, the elastic strain, and U is again an energy “per unit volume of κi (p).”
We have put an overbar on the rotation Q̄ to distinguish it from the spatially uniform
rotation Q(t) of (2.25) associated with a superposed rigid-body motion. This is because
the argument leading to (2.24) is purely local, implying that Q̄ may depend on both x
and t. This distinction will be discussed further in Section 6.4.
 Materially uniform elastic bodies 89

With these results in hand we may proceed as in Chapter 2 to derive the Piola stress

WH = HŜ (6.9)

in terms of the Piola–Kirchhoff stress

Ŝ = UÊ , (6.10)

both referred to κi (p). With reference to (1.38), the first of these is related to the Cauchy
stress T by

WH = TH∗ = T(FK)∗ = TF∗ K∗ = PK∗ , (6.11)

where P is the Piola stress referred to κ, given, as in (2.10), by the partial derivative

P = ΨF , (6.12)

at fixed K. Thus,

WH = (ΨF )K∗ . (6.13)

Problem 6.1. Assume A and B to be invertible and show that (AB)∗ = A∗ B∗ .

Recall the Legendre–Hadamard condition (2.61):

a ⊗ n · ΨFF [a ⊗ n] ≥ 0 for all a ⊗ n. (6.14)

To interpret this in terms of W, we differentiate (6.13), at fixed K, on a one-parameter

family H(u) = F(u)K, reaching

WHH [Ḟ K] = (ΨFF [Ḟ])K∗ , (6.15)

where the superposed dot is the derivative with respect to u, evaluated at u = 0, say.
Scalar multiplication by Ḟ K and use of the identity A · BC = Bt A · C, with B = Ḟ and
C = K, yields

ḞK · WHH [Ḟ K] = JK Ḟ · ΨFF [Ḟ], (6.16)

which we apply with Ḟ = a ⊗ n to express (6.14) in the form

a ⊗ m · WHH (H)[a ⊗ m] ≥ 0 for all a ⊗ m, (6.17)

with m = Kt n. It follows that Ψ satisfies the Legendre–Hadamard condition at F if and

only if W satisfies it at H = FK.
90 Energy, stress, dissipation, and plastic evolution

Proceeding exactly as in the derivation of (2.73), we have

WHH (H)[A] = AŜ + HUÊÊ (Ê)[Ht A] (6.18)

for any tensor A.

We impose the normalization condition W(1) = 0, where 1 is the shifter discussed in
Section 2.3. Equation (6.7) then implies that W(H) = 0 if H ∈ Orth+ . We assume the
converse to also be true, and hence that W(H) = 0 if and only if H ∈ Orth+ . In view
of our assumption that κi (p) is stress free, we further assume that Ŝ = 0 if and only if
Ê = 0; i.e., if and only if H ∈ Orth+ , and, as in Section 2.3, that the classical fourth-order
tensor of elastic moduli,

C = UÊÊ|Ê=0 , (6.19)

is positive definite in the sense that A · C[A] > 0 for all non-zero symmetric A. This is
precisely the same tensor that we defined in (2.77). Our assumptions imply that

WHH (Q)[A] = QC[Qt A] (6.20)

for any rotation Q, and hence, as shown in Section 2.3, that the Legendre–Hadamard
condition is automatically satisfied
  at zero elastic strain.
 
The presumed smallness of Ê justifies the low-order Taylor expansion

 2
 
U(Ê) = U(0) + Ê · UÊ (0) + 12 Ê · C[Ê] + o(Ê ) (6.21)

of the strain-energy function. Our assumptions then imply that the strain energy is
approximated at leading order by the homogeneous quadratic function

U ≃ 12 Ê · C[Ê], (6.22)

yielding

Ŝ ≃ C[Ê] (6.23)
 
 
at leading order in the elastic strain. Because the norm Ê of the elastic strain is invariably
small in metals—recall the discussion in Section 1.1—this approximation suffices for
essentially all situations of practical interest. Moreover, the positive definiteness of C
ensures that (6.23) is invertible, and hence that

Ê ≃ L[Ŝ], (6.24)

where L = C −1 is the fourth-order compliance tensor. The modulus and compliance

tensors possess major symmetry and both minor symmetries, as discussed in Section 2.3.
 Materially uniform elastic bodies 91

Writing (6.22) in the form U ≃ 21 Ŝ · Ê, we then have

U ≃ 12 Ŝ · L[Ŝ]. (6.25)

Our assumptions thus imply that the energy is approximated by a positive definite
function of Ê or Ŝ.
Recall that the solid is undistorted in κi (p) ; therefore,

W(H) = W(HR), (6.26)

where R ∈ gκi (p) is a rotation, and hence

U(Ê) = U( 12 (Ht H-I)) = U( 12 (Rt Ht HR-I))

= U( 21 Rt (Ht H-I)R) = U(Rt ÊR). (6.27)

This implies that the transformation Ê → Rt ÊR induces the transformation

Ŝ → Rt ŜR. (6.28)

To see this consider a one-parameter family Ê(u) of elastic strains, with u ∈ (−u0 , u0 ).
Following a procedure detailed in the book by Gurtin et al., we define

f(u) = U(Ê(u)). (6.29)

Then,

ḟ = UÊ (Ê0 ) · A, (6.30)

′
where ḟ = f ′ (u)|u=0 , Ê0 = Ê(0) and A = Ê (u)|u=0 . According to (6.27) it is also true that

f(u) = U(Rt Ê(u)R), (6.31)

and hence that

ḟ = UÊ (Rt Ê0 R) · Rt AR. (6.32)

Problem 6.2. Prove the identity A · BCD = Bt ADt · C.

Comparing (6.30) and (6.32) gives

[RUÊ (Rt Ê0 R)Rt − UÊ (Ê0 )] · A = 0, (6.33)

92 Energy, stress, dissipation, and plastic evolution

implying, as the term in braces is symmetric, A is an arbitrary symmetric tensor, R is

orthogonal, and dropping the subscript 0 , that

UÊ (Rt ÊR) = Rt UÊ (Ê)R, (6.34)

which is just (6.28). That is, the stress evaluated at Rt ÊR equals the stress evaluated at
Ê, pre-multiplied by Rt and post-multiplied by R.

Problem 6.3. The classical stress–strain relation for isotropic materials is

Ŝ = λ(trÊ)I + 2μÊ,

where λ and μ, the Lamé moduli, are constants (see the Appendix to this section).
Verify (6.28) for arbitrary orthogonal R.

Note that we have extended R from the group of rotations to the group of orthogonal
tensors. This is permissible because, if R ∈ Orth+ satisfies (6.27), then −R ∈ Orth
does too. Because the group Orth is obtained by adding −I to the elements of the
group Orth+ , it follows that the symmetry group is effectively a subgroup of Orth.
In the case of isotropy it coincides with Orth. For example, if the 180◦ rotation R =
2e ⊗ e − I about a unit vector e belongs to the symmetry group, then so does the
reflection −R = I − 2e ⊗ e through the plane with normal e, and vice versa. In this
way we accommodate the reflection symmetries observed in crystals, a kind of sym-
metry that cannot be associated with an orientation-preserving transformation having
JR > 0.

Appendix: Isotropy
The strain-energy function U(Ê) for isotropic materials is such that U(Ê) = U(Rt ÊR)
2 3
for any orthogonal R. It therefore depends on Ê via the scalars trÊ, tr(Ê ), and tr(Ê ).
Sufficiency is immediate because these scalars are invariant under Ê → Rt ÊR for every
orthogonal R. To prove necessity, we first note that the principal invariants of Ê are
2
I1 = trÊ, I2 = 21 [(trÊ)2 − tr(Ê )], and I3 = det Ê; these are the coefficients in the cubic
characteristic equation for the eigenvalues of Ê, and also in the Cayley–Hamilton formula
for Ê. See, for example, the book by Gurtin et al. The latter may be used to express
2 3
det Ê in terms of the list {trÊ, tr(Ê ), tr(Ê )}, and so this list determines the principal
invariants.
We need to show that if Ik (A) = Ik (B); k = 1, 2, 3 for any symmetric tensors A,B,
then U(A) = U(B). For U(Ê) is then determined by the Ik (Ê), and hence by {trÊ,
2 3
tr(Ê ), tr(Ê )}. Thus, suppose Ik (A) = Ik (B); k = 1, 2, 3. The characteristic equation
 Surface dislocations and stress relaxation 93

for eigenvalues then implies that A,B have the same (real) eigenvalues λi . By the spectral
representation for symmetric tensors, we have
X X
A= λi ai ⊗ ai and B= λi bi ⊗ bi ,

where {ai } and {bi } are orthornormal sets. Then Q = ai ⊗bi is orthogonal, and bi = Qt ai ,
yielding B = Qt AQ and U(B) = U(Qt AQ). But U(Qt AQ) = U(A) by hypothesis; thus,
U(A) = U(B), and necessity is proved.
In the case of isotropy the most general homogeneous quadratic strain-energy func-
2
tion is thus of the form U(Ê) = 12 λ(trÊ)2 + μtr(Ê ), where λ and μ are constants, and, as
is well known, for this to be positive definite it is necessary and sufficient that 3λ+2μ > 0
and μ > 0. To obtain the stress we use
′ ′ ′
UÊ · Ê = λ(trÊ)I · Ê + μ(Ê · Ê)′ = [λ(trÊ)I + 2μÊ] · Ê ,

where the primes are the derivatives, with respect to the parameter, of a one-parameter
family of strains, and thus conclude, from (2.33), that Ŝ is as stated in Problem 6.3.

6.2 Surface dislocations and stress relaxation

We have seen that if the plastic deformation tensor K−1 is a smooth function of x in κ,
then there exists a dislocation density ακ = CurlK−1 such that
Z
B(Ω⊥ , t) = αtκ ν⊥ dA, (6.35)
Ω⊥

where Ω⊥ ⊂ κ is any simply connected surface with unit-normal field ν⊥ , with B(Ω⊥ , t)
the associated Burgers vector. Consider a surface Ω across which K−1 is discontinuous,
and suppose Ω⊥ intersects Ω orthogonally. Let Γ be the curve of intersection. If K−1 is
smooth in the regions on either side of Ω, then Stokes’ theorem (3.120) may be applied
to the individual parts Ω±
⊥ of Ω⊥ separated by Ω. Adding the two expressions and using
Γ = ∂Ω+⊥ ∩ ∂Ω− ⊥ , we then have
Z Z Z Z
αtκ ν⊥ dA = (CurlK−1 )t ν⊥ dA = K−1 dx + [K−1 ]dx, (6.36)
Ω⊥ Ω⊥ ∂Ω⊥ Γ

where

[K−1 ] = (K−1 )+ − (K−1 )− (6.37)

is the discontinuity of K−1 on Ω. Here we use subscripts ± , respectively, to denote limits

as Ω, with unit-normal field ν, is approached from the regions into which ν and −ν are
directed; and, dx = ν × TdS, where T = ν⊥|Γ and S measures arclength on Γ. Note that
94 Energy, stress, dissipation, and plastic evolution

Ω ⊥)
⊥ to
ν⊥(
Ω
S
ν (⊥ t
o Ω)
* Ω⊥

Figure 6.1 A surface Ω across which K is dis-

continuous, intersecting a material suface Ω⊥
orthogonally along the curve Γ.

T ∈ TΩ(x,t) , the tangent plane to Ω at the point x at time t (Figure 6.1), and that ν × T is
tangential to Γ as ∂Ω+⊥ is traversed in the sense of Stokes’ theorem. Thus, on Γ we have
[K−1 ]dx = [K−1 ](ν × T)dS.
For a given surface Ω, with unit-normal field ν, the foregoing construction allows for
infinitely many surfaces Ω⊥ with corresponding intersection curves Γ and unit-normal
fields ν⊥|Γ . Thus, T is an arbitrary unit tangent to TΩ(x,t) , and [K−1 ](ν × T) may be
regarded as a linear function of the variable T ∈ TΩ(x,t) . Accordingly, there is a tensor
field βκ defined on Ω—the referential surface dislocation density—such that

[K−1 ](T × ν) = βtκ T. (6.38)

Because the action of βtκ on ν is undefined in this expression, we can extend its domain
to three-space and impose βtκ ν = 0 without loss of generality. From (6.36) and noting
that T × ν = −ν × T, we have
Z Z Z
B(Ω⊥ , t) = K−1 dx = αtκ ν⊥ dA + βtκ T dS. (6.39)
∂Ω⊥ Ω⊥ Γ

Equation (6.38) can be solved for βtκ by introducing an orthonormal basis {T1 , T2 }
for TΩ(x,t) and writing the three-dimensional identity as

I = ν ⊗ ν + Tα ⊗ Tα (6.40)

in which α is summed from one to two. Then,

[K−1 ] = [K−1 ]I = k ⊗ ν + [K −1 ]Tα ⊗ Tα , where k = [K−1 ]ν. (6.41)

 Surface dislocations and stress relaxation 95

Further, if we choose Tα such that {T1 , T2 , ν} is positively oriented, then (6.38) furnishes

[K−1 ]T1 = [K−1 ](T2 × ν) = βtκ T2 , (6.42)

and, similarly,

[K−1 ]T2 = −[K−1 ](T1 × ν) = −βtκ T1 . (6.43)

Thus,

[K−1 ] = k ⊗ ν − βtκ ϵ(ν) , (6.44)

where

ϵ(ν) = T1 ⊗ T2 − T2 ⊗ T1 (6.45)

is the two-dimensional permutation tensor on TΩ(x,t) . The name is justified by the fact
that it is independent of the particular choice of orthonormal vectors Tα in TΩ(x,t) and
thus has an invariant meaning.

Problem 6.4. Prove this.

To solve (6.44) we post-multiply both sides by ϵ(ν) and use (k ⊗ ν)ϵ(ν) = k ⊗ ϵt(ν) ν with
= 0, together with ϵ2(ν) = −Tα ⊗ Tα = ν ⊗ ν − I, finally reaching
ϵt(ν) ν

βtκ = [K−1 ]ϵ(ν) . (6.46)

As might be expected we can repeat the entire exercise with a simply connected sur-
face ω in κt and a surface ω⊥ that intersects it orthogonally. We would then establish the
representation
Z Z
b(ω⊥ , t) = αtκt n⊥ da + βtκt tds (6.47)
ω⊥ γ

for the relevant Burgers vector, where γ = ∂ω+⊥ ∩ ∂ω− ⊥ ; s is arclength on γ; t ∈ Tω(y,t) ,
with unit normal n; n × t is the unit tangent to γ; and βκt —the spatial surface dislocation
density—is given by

βtκt = [H−1 ]ϵ(n) , (6.48)

96 Energy, stress, dissipation, and plastic evolution

in which [H−1 ] is the jump of H−1 across ω and ϵ(n) is the two-dimensional permutation
tensor on Tω(y,t) . The counterpart of (6.44) is

[H−1 ] = h ⊗ n − βtκt ϵ(n) , where h = [H−1 ]n. (6.49)

Note that in the foregoing discussion there is no requirement that Ω be a material

surface. That is, it may evolve relative to the material and thus be associated with a mov-
ing surface in the reference configuration κ. Therefore, the notion of surface dislocation
applies not only to grain boundaries but also to evolving phase interfaces, to be discussed
in Section 6.7.

Problem 6.5. If Ω is a material surface, then the referential and spatial surface dislo-
cation densities are not independent. Derive the relationship between them, and
show that if one vanishes, then both vanish.

It is instructive to revisit the foregoing procedure with K−1 replaced by the deforma-
tion gradient F = ∇χ. Assuming χ to be a continuous function of x, and that its gradient
suffers a jump across Ω, we have
Z Z Z Z
t
0= dχ = (∇χ)dx = (CurlF) ν⊥ dA + [F]dx. (6.50)
∂Ω⊥ ∂Ω⊥ Ω⊥ Γ

Problem 6.6. Show that CurlF vanishes in the regions separated by the surface Ω,
assuming χ to be a twice-differentiable function of x therein.

Using dx = T̂dS on Γ, where T̂ = ν × T is the unit tangent, it follows that

Z
[F]T̂dS = 0. (6.51)
Γ

As Ω⊥ is arbitrary, and hence so too Γ = Ω ∩ Ω⊥ , and as T̂ · ν = 0, we may localize and

conclude that [F]T̂ = 0 for all T̂ ∈ TΩ(x,t) . Then, with

[F] = [F]I = [F]ν ⊗ ν + [F]Tα ⊗ Tα , (6.52)

where {Tα } is any orthonormal basis for TΩ(x,t) , with unit normal ν, we arrive at
Hadamard’s lemma

[F] = f ⊗ ν on Ω, (6.53)

where f = [F]ν. See Chapter C in the treatise by Truesdell and Toupin for a comprehen-
sive discussion of discontinuity relations of this kind. We conclude that [F] is a rank-one
 Dissipation due to plastic evolution 97

tensor at a coherent interface Ω; that is, at an interface across which χ is continuous

but F may be discontinuous. This is a severe constraint on the limiting values F± at
the surface Ω.
Here, the rank of a tensor is the dimension of its image space:

RankA = dim(Span{Av}) (6.54)

for any three-vector v. In our example, [F]v = (ν · v)f and Span{[F]v} = Spanf is one-
dimensional.
In contrast, (6.49) shows that Rank[H−1 ] = 3, and hence that [H−1 ] is a full-rank
tensor if surface dislocation is present. In this case there is no constraint on the limiting
values H± . To better understand the significance of this result, consider adjoining stress-
free crystal grains separated by the interface Ω. Because the associated modulus tensors
are positive definite, it follows that they are also free of elastic strain and their respec-
tive strain energies vanish. In particular, H± ∈ Orth+ and (6.49) yields a (generally
non-zero) surface dislocation density βκt . The so-called tilt and twist boundaries furnish
illustrative examples—see Section 4.4 in Lardner’s book and Section 8.2.2.3 in the book
by Nabarro. However, if βκt should happen to vanish, then from (6.49) we would have
(H−1 )+ = (H−1 )− + h ⊗ n. If one of the adjoining grains is stress free, so that H− ,
say, is a rotation, then H+ is not a rotation and the adjacent grain is necessarily stressed.
Because of the positive definiteness of (6.22), its energy is also positive and hence so too
the total energy of the two grains together. Accordingly, surface dislocation induced by
plastic deformation—see (6.46) and Problem 6.5—is an additional interfacial degree of
freedom which is available to minimize the energy in adjacent grains. Presumably this
is related to the fact that metallic crystals most often occur as aggregates of adjoining
grains rather than singly.

6.3 Dissipation due to plastic evolution

A central feature of plastic evolution is its irreversibility. This is intimately connected with
the notion of energy dissipation. To introduce the concept of dissipation attending plastic
evolution in as simple a manner as possible, we confine attention to smooth deformations
in the present section. Deformations with discontinuous gradients will be taken up in
Section 6.7.
We begin with the energy balance (1.44) and the strain energy
Z
U(π, t) = Ψ(F,K)dV; where Ψ(F,K) = J−1
K W(H) with H = FK. (6.55)
π

The total mechanical energy is U(π, t) + K(π, t), where K is the kinetic energy, and we
define the dissipation D, as in Problem 2.6, by

D(π, t) = P(π, t) − d
dt [U(π, t) + K(π, t)], (6.56)
98 Energy, stress, dissipation, and plastic evolution

where P(π, t) is the power supplied to the arbitrary part π of κ. We suppose that

D(π, t) ≥ 0 for all π ⊂ κ, (6.57)

this effectively serving as a surrogate for the second law of thermodynamics in the
present, purely mechanical, setting.
It follows from (1.44) that D = S − dtd U, where S is the stress power. Combining with
(2.34), it follows immediately that
Z
D(π, t) = DdV, (6.58)
π

where

D = P · Ḟ − Ψ̇, (6.59)

and localizing (6.57) yields

D≥0 at all x ∈ κ. (6.60)

To obtain a useful expression for the dissipation we differentiate (6.55)2 :

Ψ̇ = J−
K [Ẇ − (J̇K /JK )W ].
1
(6.61)

Here we use the identity J̇K /JK = K−t · K̇ together with

Ẇ = WH · Ḣ = WH · (ḞK + FK̇), (6.62)

which is reduced, with the aid of the identities A · BC = ACt · B = Bt A · C, to

Ẇ = WH Kt · Ḟ + Ft WH · K̇. (6.63)

Recalling (6.13), which yields WH Kt = JK P and Ft WH = JK Ft PK−t , Eq. (6.61) is finally

reduced to

Ψ̇ = P · Ḟ − E · K̇K−1 , (6.64)

where

E = ΨI − Ft P (6.65)

is the Eshelby tensor. Accordingly, the local dissipation is given in terms of the rate of
plastic deformation K̇ by

D = E · K̇K−1 . (6.66)

This important result, first announced by Epstein and Maugin, highlights the role of
the Eshelby tensor as the driving force for dissipation. We will use it in Section 6.5, in
 Dissipation due to plastic evolution 99

conjunction with inequality (6.60), to derive restrictions on constitutive equations for

the plastic evolution K̇. Observe, from (6.64) and the chain rule, that

ΨF (F,K) = P and ΨK (F,K) = −EK−t , (6.67)

the first of these being simply a restatement of (6.12).

The expression (6.66) for D makes clear the fact that the dissipation vanishes in the
absence of plastic evolution; i.e., D = 0 if K̇ = 0. On the basis of empirical observation, we
introduce the hypothesis that plastic evolution is inherently dissipative; thus, we suppose
that D ̸= 0 if and only if K̇ ̸= 0. In view of (6.60), this means that

K̇ ̸= 0 if and only if D > 0. (6.68)

It may be observed, from the definition (6.65), that the Eshelby tensor is purely ref-
erential, mapping the translation space T κ to itself. It proves convenient to introduce a
version of the Eshelby tensor, E′ , that maps κi (p) to itself. This is given by the relation

1 −t ′ t
E = J−
K K EK. (6.69)

To establish this we may proceed from the observation that if E∗ is the spatial Eshelby ten-
sor, derived by taking the current configuration as reference, i.e., by taking κ to coincide
with κt at a particular instant t, then

E∗ = ψi−T, with ψ = J−1

F Ψ, (6.70)

where i is the spatial identity, so that

E = JF Ft E∗ F−t . (6.71)

Thus, E is the pullback of E∗ from κt to κ. It follows that E is the pullback of E′ from

κi (p) to Tκ (p), and E′ is the pullback of E∗ from Tκt (p) to κi (p). In particular,

E′ = JH Ht E∗ H−t = WI − Ht WH , (6.72)

which implies that E′ is determined solely by H and hence purely elastic in origin. This
is equivalent to

E′ = UI − Ht HŜ, (6.73)

which further implies that E′ remains invariant if H is replaced by Q̄H for any
rotation Q̄.
100 Energy, stress, dissipation, and plastic evolution

In the practically important case of small elastic strain, (6.22) and (6.23) yield the
estimate
 
 
E′ = −Ŝ + o(Ê). (6.74)

Thus, the Eshelby tensor, referred to κi (p), is given, to leading order and apart from
sign, by the Piola–Kirchhoff stress referred to the same state.
From (6.69) and the identity A · BC = ACt · B we have that

E · K̇K−1 = EK−t · K̇ = J−1 −t ′ −1 ′

K K E · K̇ = JK E · K
−1
K̇. (6.75)

Thus,

JK D = E′ · K−1 K̇, (6.76)

and the assumption of inherent dissipativity is equivalent to the statement:

K̇ ̸= 0 if and only if E′ · K−1 K̇ > 0. (6.77)

It is interesting to observe that if Ê = 0, then, according to our constitutive hypotheses,

U = 0 and Ŝ = 0. Equations (6.73) and (6.76) then give D = 0 and (6.77) implies that
there can be no plastic evolution. That is, without stress, there can be no change in the
plastic deformation. This comports with the observed phenomenology.

Problem 6.7. Show that the dissipation D is invariant under material symmetry
transformations.

6.4 Superposed rigid-body motions

We know that the symmetry of the Cauchy stress T is equivalent to the statement
W(H) = W(Q̄H) for all rotations Q̄. This follows from WH = TH∗ , implying that
WH ·ΩH vanishes for any skew tensor Ω. The result then follows exactly as in Section 2.1.
Because the argument leading to this conclusion is purely local, the rotation Q̄ may con-
ceivably vary from one material point to another. This stands in contrast to the spatially
uniform rotation Q (t) associated with a superposed rigid-body motion; our notation is
intended to distinguish these cases explicitly. In a superposed rigid-body motion, the
deformation χ(x, t) is changed to Q(t)χ(x, t) + c(t), as in (2.25), and this implies that
F(= ∇χ) changes to Q(t)F. This simple deduction does not apply to H or G, how-
ever. We therefore encounter the question of how the elastic and plastic deformations
transform under a superposed rigid-body motion.
To address this we must keep in mind the fact that, at this stage in our analysis, Q̄ bears
no a priori relation to a superposed rigid-body motion. On the contrary, our purpose
 Superposed rigid-body motions 101

is to establish such a relation. This requires that we introduce additional hypotheses. A

natural one is that the strain energy W should be invariant; that is,

W(H+ ) = W(H), (6.78)

where H+ is the value of the elastic deformation associated with the superposed rigid
motion. Because of the symmetry of the Cauchy stress, (6.8) is applicable and implies
that W(H) = Ŵ(H t H), where Ŵ(Ht H) = U( 12 (Ht H-I)). Then, Ŵ((H+ )t H+ ) =
Ŵ(Ht H), i.e.,

Ŵ(Ht Zt ZH) = Ŵ(Ht H), where Z = H+ H−1 . (6.79)

This must hold for all admissible H, i.e., for all H with JH > 0. It therefore holds with
H=1, the shifter, giving

Ŵ(1t Zt Z1) = Ŵ(I). (6.80)

The constitutive hypotheses introduced in Section 6.1 imply that the right-hand side
vanishes, and that Ŵ(1t Zt Z1) > 0 if 1t Zt Z1 ̸= I. The two statements are reconciled only
if 1t Zt Z1 = I, and this in turn yields Zt Z = i. Because JZ > 0, from (6.79)2 , we then
have Z ∈ Orth+ . We label this as Q̄. Conversely, this ensures that (6.79)1 is satisfied for
any admissible H. We conclude that

H+ = Q̄H, (6.81)

where Q̄(x, t) is an arbitrary rotation field, and thus recover (6.7). Then, H+ = Q̄FK,
where H+ = F+ K+ = QFK+ in which K+ is the inverse plastic deformation associated
with the superposed rigid-body motion. Hence,

Q̄FK = QFK+ , (6.82)

implying that

JK+ = JK . (6.83)

We would like to use this result to arrive at some definite conclusion about the rela-
tionship between K+ and K, but this requires a further hypothesis. A natural one is
that the dissipation is insensitive to superposed rigid-body motions. As justification we
note that in continuum mechanics both the energy balance and the entropy produc-
tion inequality have this property, the latter specializing to the dissipation inequality in
a purely mechanical setting. See the discussion in the book by Liu, for example. Thus,
we impose D+ = D, where
+
JK+ D+ = (E′ )+ · (K+ )−1 K̇ , (6.84)

in which (E′ )+ is the Eshelby tensor based on H+ . From (6.73) and (6.81) it fol-
lows that (E′ )+ = E′ and hence that E′ is invariant under superposed rigid-body
102 Energy, stress, dissipation, and plastic evolution

motions. The presumed invariance of the dissipation, combined with (6.83), then leads
to
+
E′ · (K+ )−1 K̇ = E′ · K−1 K̇. (6.85)

To explore the implications we define Y = K+ K−1 and note, from (6.83), that JY = 1.
Suppose Y(x, t0 ) = I, so that the superposed rigid motion commences at time t0 . We
thus recast (6.85) in the form

E′ · K−1 Y−1 ẎK = 0. (6.86)

Recall that K requires the specification of a reference configuration, and that (5.13)
furnishes the relationship between the values of K associated with two reference config-
urations κ1 and κ2 , say. Thus, K2 = RK1 , where R is the gradient of the map from κ1
to κ2 . We exploit the substantial freedom in the choice of reference configuration and
choose this map such that R = K− 1
1 at the material point p in question, at a particu-
lar instant in time. The fact that R is a gradient and independent of time, whereas K1
is time-dependent and not a gradient, is thereby rendered irrelevant. Thus, our choice
yields K2 = I at the material point and time in question. Applying (6.86) with κ = κ2 ,
we then have

E′ · Y−1 Ẏ = 0, where Y = K+2 , (6.87)

with the time derivative evaluated at the considered instant. From (6.77) we then have
that Ẏ = 0 at the instant in question. As this instant is arbitrary, Ẏ vanishes at all instants,
yielding Y(x2 , t) = Y(x2 , t0 ) = I, where x2 is the position of p in κ2 . Then, K+2 = K2 and
the transformation formula yields RK+1 = K+2 = K2 = RK1 , implying that K+1 = K1 and
hence that K+ = K for all choices of reference configuration. Combining this result with
(6.82), we conclude that Q̄(x, t) = Q(t). Finally, the transformation formulas for F,H,
and K under superposed rigid-body motions are

F → QF, H → QH, and K → K. (6.88)

6.5 Yielding and plastic flow

In this section we extend the concept of yield, introduced in Section 1.1 in a one-
dimensional setting, to three dimensions. Thus, we assume the onset of yield, and hence
the possibility that K̇ ̸= 0, occurs when the elastic strain is such that G(Ê) = 0, where G
is an appropriate threshold function, or yield function, pertaining to the material at hand.
In a crystalline material, for example, this implies that yield occurs when the lattice is
sufficiently distorted. Of course we may derive this from the more basic assumption
that the yield function is dependent on H, and that yield is insensitive to superposed
 Yielding and plastic flow 103

rigid motions. This extends to three dimensions the observation, in Section 1.1, to the
effect that the uniaxial bar yields when the stress, and hence the elastic stretch, reaches a
critical value. We expand the notion of material uniformity by taking the yield function
to be the same function at every point p of the materially-uniform body. Thus, yield
occurs when the elastic distortion lies on a certain manifold in six-dimensional space.
The material is said to respond in the elastic range if Ê is such that G(Ê) < 0, whereas, in
rate-independent plasticity theory, the region of strain space where G(Ê) > 0 is deemed
to be inaccessible. These statements are modified in the case of rate-dependent, or vis-
coplastic, behavior, to be discussed in Chapters 7 and 9. For metals, the diameter of the
elastic range is typically such as to severely limit the norm of the elastic strain, so that
as a practical matter we may assume the elastic strain—at least in the rate-independent
theory—to be small. This provides post facto justification for the small-elastic-strain
assumption invoked in the foregoing.
In these circumstances our constitutive hypotheses imply that the relation between Ê
and Ŝ is approximately linear and one-to-one, so that we may equally well characterize
yield in terms of the statement F(Ŝ) = 0, where

F(Ŝ) = G(L[Ŝ]) (6.89)

is the yield function, expressed in terms of the stress. In keeping with the foregoing
definition we suppose elastic response to be operative in the elastic range, defined by
F(Ŝ) < 0, and, in the case of rate-independent response, that no state of stress existing
in the material can be such that F(Ŝ) > 0.
We further suppose that F(0) < 0, and hence that the stress-free undistorted state
belongs to the elastic range. In this way we partition six-dimensional stress space into
the regions defined by positive, negative, and null values of F, with the first of these being
inaccessible in any physically possible situation. This appears to disallow behavior of the
kind associated with the Bauschinger effect, in which yield can occur upon load reversal
before the unloaded state is attained. However, empirical facts support the view that this
effect is accompanied by the emergence of dislocations. As we have seen in Section 5.3,
these give rise to non-uniform distributions elastic strain, and hence stress, which cannot
be directly correlated with the overall global response represented in the test data. From
this point of view the Bauschinger effect is thus an artifact of the test being performed,
not directly connected with constitutive properties per se. Conventionally, however, the
Bauschinger effect is modeled at the constitutive level on the basis of the ad hoc notion
of kinematic hardening, in which the yield surface translates in stress space according to
certain prescribed rules. More recent efforts aimed at modeling the Bauschinger effect
take account of the energetic influence of defects, such as the true dislocation density, at
the constitutive level. These give rise to space-time partial differential equations for the
plastic deformation. Tentative efforts along these lines are described in Chapter 9 in the
context of a more expansive discussion of strain hardening.
Returning to the simpler theory, consider a cyclic process in which the deforma-
tion and velocity fields start and end at the same values. We assume, as in Chapter 2,
104 Energy, stress, dissipation, and plastic evolution

that non-negative work must be performed to effect such a process, and thus impose
(2.8),
Z t2
P · Ḟdt ≥ 0, (6.90)
t1

where t1,2 respectively are the times when the cycle begins and ends. We henceforth
suppress the material point p in the notation—equivalently, its reference position x—as
the ensuing discussion pertains to a fixed material point. Suppose the times t1,2 are such
that the associated stresses satisfy F < 0; the cycle begins and ends in the elastic range.
If the cycle lies entirely within the elastic range, then K is fixed at the material point in
question and we can proceed exactly as in Section 2.1 to conclude, assuming P to be a
function of F, that there exists a family of energy densities Ψ(F,K), say, parametrized by
K, such that P = ΨF (F,K), as in (6.67)1 .

Problem 6.8. Carry out the details.

Now consider a cycle such that there exists a subinterval of time [ta , tb ] ⊂ [t1 , t2 ] during
which F = 0, and that F < 0 outside this subinterval. Then we may have plastic flow, i.e.,
K̇ ̸= 0 during this subinterval, while K̇ = 0 outside it, implying that K(t1 ) = K(ta ) and
K(t2 ) = K(tb ). Substituting (6.59) and noting that the process is cyclic in the sense that
F(t2 ) = F(t1 ), we arrive at the statement
Z tb
Ψ(F(t1 ), K(tb )) − Ψ(F(t1 ), K(ta )) + Ddt ≥ 0. (6.91)
ta

Equivalently,
Z tb
[ΨK (F(t1 ), K(t)) · K̇(t) + D(t)]dt ≥ 0. (6.92)
ta

To ensure that a cycle beginning in the elastic range (F(Ŝ(t1 )) < 0) also ends there,
we pass to the limit tb − ta → 0. This yields H(t2 ) = F(t2 )K(tb ) = F (t1 )K(tb ) →
F(t1 )K(ta ) = H(t1 ), implying that Ê(t2 ) → Ê(t1 ) and hence, via the elastic stress–strain
relation, that Ŝ(t2 ) → Ŝ(t1 ).
Dividing (6.92) by tb − ta (> 0), passing to the limit and invoking the mean-value
theorem, we conclude that

ΨK (F(t1 ), K(ta )) · K̇(ta ) + D(ta ) ≥ 0, (6.93)

which may be written, using (6.66) and (6.67)2 , as

[E(F(ta ), K(ta )) − E(F(t1 ), K(ta ))] · K̇(ta )K(ta )−1 ≥ 0. (6.94)

 Yielding and plastic flow 105

Alternatively, from (6.69) this inequality may be restated in the form

[E′ (Ê(ta )) − E′ (Ê(t1 ))] · K(ta )−1 K̇(ta ) ≥ 0, (6.95)

where E′ (Ê) is the function defined by (6.8)2 and (6.73):

E′ (Ê) = UI − (I + 2Ê)UÊ . (6.96)

This means that the dissipation is maximized by strains Ê that lie on the yield surface
G(Ê) = 0.
As we have seen, E′ is approximated by −Ŝ in the case of small elastic strain. In this
−1 −1
  together with K K̇ = −
case we substitute (6.74),  ĠG
 (which follows from GK = I)
   
into (6.95), divide by Ê , and pass to the limit Ê → 0 to derive the leading-order
restriction
−1 ∗ −1 ∗
Ŝ · SymĠG ≥ Ŝ · SymĠG ; F(Ŝ ) ≤ 0, F(Ŝ) = 0, (6.97)

in which the qualifier Sym has been inserted to account for the fact that the Piola–
Kirchhoff stress is symmetric; the inner product then picks up only the symmetric part
−1
of ĠG .
−1
To characterize those ĠG that satisfy (6.97), consider an arbitrary smooth path
Ŝ(u), with u > 0, such that limu→0 Ŝ(u) = Ŝ and

F(Ŝ(u)) ≤ 0. (6.98)

Then, if the yield function F is differentiable,

 
  ′
0 ≥ F(Ŝ(u)) − F(Ŝ) = FŜ (Ŝ) · [Ŝ(u) − Ŝ] + o(Ŝ(u) − Ŝ) = uFŜ (Ŝ) · Ŝ + o(u), (6.99)

′ d
where Ŝ = du Ŝ(u)|u=0 . We divide by u and pass to the limit, concluding that

′
FŜ (Ŝ) · Ŝ ≤ 0. (6.100)

In the same way, it follows from (6.97) that

−1 −1 ′
0 ≥ SymĠG · [Ŝ(u) − Ŝ] = u(SymĠG ) · Ŝ + o(u), (6.101)

and hence that

−1 ′
SymĠG · Ŝ ≤ 0. (6.102)
106 Energy, stress, dissipation, and plastic evolution

In the case of equality in (6.100) and (6.102) we have that SymĠG−1 is orthogonal—
′
in the vector space of symmetric tensors—to arbitrary tensors Ŝ that are, in turn,
orthogonal to FŜ (Ŝ). Thus,
−1
SymĠG = λFŜ , (6.103)
−1 ′
for some scalar field λ(x, t), with the derivative evaluated at Ŝ. Then, SymĠG · Ŝ =
′
λFŜ · Ŝ , and inequalities (6.100) and (6.102) are consistent provided that

λ ≥ 0. (6.104)

Taken together, these are the Kuhn–Tucker necessary conditions for the optimization
problem (6.97). The book by Zangwill may be consulted for further discussion in the
context of general optimization theory.
The flow rule for the evolution of plastic deformation is thus given by
−1
ĠG = λFŜ + Ω, (6.105)

where Ω(x, t) is a skew tensor field, called the plastic spin.

With these results in hand (6.97) becomes
∗ ∗
(Ŝ − Ŝ ) · FŜ (Ŝ) ≥ 0; F(Ŝ ) ≤ 0, F(Ŝ) = 0, (6.106)
∗
in which Ŝ is an arbitrary stress in the closure of the elastic range. Thus, the vector
∗
Ŝ − Ŝ must either be orthogonal to FŜ (Ŝ) or form an acute angle with it. Because
FŜ (Ŝ) is an exterior normal to the elastic range at the boundary point Ŝ, this means that
its interior, defined by F < 0, must lie to one side of the tangent plane T F at Ŝ. This
tangent plane is unique by virtue of the assumed differentiability of F and the consequent
continuity of FŜ . (We will return to the matter of the differentiability of F later.) This,
in turn, means that the elastic range is a convex set, i.e., that if Ŝ1 and Ŝ2 belong to the
elastic range, then so do all points on the straight-line path Ŝ(u) = uŜ1 + (1 − u)Ŝ2
with u ∈ [0, 1] (suggestion: draw a figure). We conclude that the elastic range is always
a convex set in a formulation of plasticity theory based on the work inequality. It is
important to emphasize, however, that the work inequality itself is an assumption rather
than a fundamental law of mechanics, and accordingly that convexity of the elastic range
is not a general requirement. It is, however, a basic feature of the classical theory as well
as the great majority of its modern generalizations.
Being constitutive in nature, the yield function is naturally subject to material sym-
metry requirements. Thus, as in (6.27), the yield function G(Ê) must be such that

G(Ê) = G(Rt ÊR), (6.107)

 Yielding and plastic flow 107

for all orthogonal R ∈ gκi (p) ∪ {−I}. We have shown that for such R the stress transforms
as indicated in (6.28). Then (6.23) implies that

C[Rt ÊR] = Rt (C[Ê])R. (6.108)

Because C and L are mutual inverses, we then have

L[Rt ŜR] = L[Rt (C[Ê])R] = L[C[Rt ÊR]] = Rt ÊR. (6.109)

Combining this with (6.24) we conclude that G(L[Ŝ]) = G(L[Rt ŜR]), and therefore
that (6.107) furnishes the restriction

F(Ŝ) = F(Rt ŜR) (6.110)

on the yield function.

Further, invariance of material response under the symmetry transformation H →
HR is equivalent to invariance under the transformation FK → FKR, which is tan-
tamount to invariance under K → KR at fixed F, or, equivalently, under G → Rt G.
−1 −1 −1
Accordingly, ĠG → Rt (Ġ G−1 )R. This implies that SymĠG → Rt (SymĠG )R
and hence that the plastic multiplier λ is invariant under material symmetry transforma-
−1 −1
tions. We also have SkwĠG → Rt (SkwĠG )R and thus conclude that the plastic
spin transforms as

Ω → Rt ΩR. (6.111)

From (6.74), (6.76), and (6.105) the dissipation is given, to leading order in the small
elastic strain, by

JK D = λŜ · FŜ ; F(Ŝ) = 0. (6.112)

Because λ ≥ 0, the dissipation is positive only if λ > 0 and hence only if

Ŝ · FŜ > 0, (6.113)

for all Ŝ such that F(Ŝ) = 0. This should be imposed as an a priori restriction on the
yield function.
Empirical facts indicate that yield in metals is insensitive to pressure over a very large
range of pressures, and certainly for the pressures normally encountered in practice.
Thus, yield is insensitive to the value of trT, where T is the Cauchy stress. From (6.9)
and (6.11) we have that, in the case of small elastic strain,
 
 
trT = trŜ + o(Ê). (6.114)
108 Energy, stress, dissipation, and plastic evolution

Problem 6.9. Prove this.

As the model we are pursuing purports to be valid to leading order in elastic strain, it
follows that the yield function should be insensitive to trŜ. Writing

Ŝ = DevŜ + 13 (trŜ)I, (6.115)

where DevŜ is the deviatoric part of Ŝ, we conclude that the function F̂(DevŜ, trŜ) =
F(DevŜ+ 31 (trŜ)I) is insensitive to its second argument, and hence that the yield function
depends on Ŝ entirely through its deviatoric part. Accordingly, we write

F(Ŝ) = F̃(DevŜ). (6.116)

This assumption imposes a restriction on plastic evolution. To see this we differentiate

on the one-parameter family Ŝ(u), obtaining

′ ′ ′
FŜ · Ŝ = F̃DevŜ · (DevŜ)′ = F̃DevŜ · Dev(Ŝ ) = Dev(F̃DevŜ ) · Ŝ , (6.117)

′ ′
where Ŝ = du d
Ŝ(u), and use has been made of (DevŜ)′ = Dev(Ŝ ) together with the
orthogonality of the deviatoric and spherical tensors, i.e., DevŜ · I = 0. It follows that
FŜ = Dev(F̃DevŜ ) and therefore that FŜ is deviatoric. Then (6.105) yields the conclusion
that plastic flow is isochoric, i.e.,

−1
tr(ĠG ) = 0, (6.118)

and hence that (JG )· vanishes. Further, as Dev(Rt ŜR) = Rt (DevŜ)R for any orthogonal
R, it follows from (6.110) and (6.116) that

F̃(DevŜ) = F̃(Rt (DevŜ)R). (6.119)

We have made essential use of the assumption that the yield function F(Ŝ) is differen-
tiable. As justification we recall that F, like the strain energy U, is a constitutive function.
Having approximated the latter by a quadratic function of Ŝ (see (6.25)), logical consis-
tency demands that we also approximate F(Ŝ) by a quadratic function. Such functions
are, of course, continuously differentiable. Because DevŜ is a linear function of Ŝ, it fol-
lows that F̃(DevŜ) should likewise be approximated by a quadratic function. We take up
this issue again in Chapter 7.
The foregoing considerations about yield and flow are quite general and apply to both
crystalline and non-crystalline materials.
 Crystallinity versus isotropy 109

6.6 Crystallinity versus isotropy

The conventional theory of crystal plasticity rests on a kinematical interpretation of plas-
tic deformation according to which the rate of plastic deformation in a single crystal is
presumed to be expressible as a superposition

−1 X
ĠG = δi si ⊗ n i (6.120)

of simple shear rates, in which G is the plastic part of the deformation gradient, δi are
the slips, and the si and ni are orthonormal vectors specifying the ith slip system. See
the books by Havner and Gurtin et al. for a fuller discussion. The sum ranges over the
currently active slip systems. Here the δi are determined by suitable flow rules, arranged
to ensure that the response is dissipative, and the skew part of (6.120), in which the slip-
system vectors are specified, furnishes the plastic spin. This decomposition is virtually
ubiquitous in the literature on crystal plasticity. However, it has been criticized on the
grounds that for finite deformations it cannot be associated with a sequence of simple
shears unless these are suitably restricted—see the paper by Rengarajan and Rajagopal.
In particular, the order of the sequence generally affects the overall plastic deformation,
a fact which is not reflected in (6.120). In the work of Deseri and Owen, conditions
are derived under which (6.120) yields an approximation to the deformation associated
with a sequence of slips. They find that such deformations are well approximated by
(6.120) in face-centered cubic crystals, but the issue remains unresolved for other crystal
classes. Further, sequential slip is typical in experiments but simultaneous multi-slip is
not usually observed.
This state of affairs provides impetus for alternative phenomenological models based
purely on considerations of material symmetry. Indeed, the decoupling of the symmetric
−1
and skew parts of ĠG in the flow rule (6.105) affords considerably more latitude in
the fitting of theory to experiment than is possible using (6.120). Accordingly, this is the
viewpoint that will be adopted in this book. A drawback, however, relative to a formula-
tion based on (6.120), is that it is then necessary to provide a constitutive specification
for the plastic spin.
In crystal-elasticity theory the stress arises in response to lattice distortion. The theory
is based on the idea that linearly independent, undistorted lattice vectors li (i ∈ {1, 2, 3})
are mapped to their images ti in κt in accordance with the Cauchy–Born hypothesis;
that is, the li are convected as material vectors. Chapter 4 of Weiner’s book contains an
extensive discussion. To accommodate plasticity, this hypothesis is assumed to apply to
the elastic deformation. Thus, ti = Hli where lj are the lattice vectors in κi (p). The lattice
set {li } associated with κi (p) is assumed to be an intrinsic property of the crystal. It is
therefore regarded as a uniform field (i.e., independent of x) in a single crystal, regarded
as a materially uniform body.
The ti are observable in principle. In practice they are computed from their recip-
rocals ti , the normals to lattice planes, which are measured in X-ray diffraction experi-
ments. The decomposition (5.3) yields ti = Fri , where r i = Kli are the lattice vectors in
110 Energy, stress, dissipation, and plastic evolution

κ. The plastic deformation is then given by K = ri ⊗ li , where the lj are the reciprocals of
the lj . The elastic deformation is H = ti ⊗ li , and the deformation gradient is F = ti ⊗ ri .
The material derivatives of the referential lattice vectors are ṙi = K̇li + Kl̇i . These
imply that if l̇i ̸= 0, then the lattice vectors are non-material (ṙi ̸= 0) in the absence of
plastic flow (K̇ = 0). However, we adopt the prevalent view that plastic flow is solely
responsible for the non-materiality of the lattice, i.e., that plastic flow alone accounts for
the evolution of material vectors relative to the lattice. We thus impose l̇i = 0 and regard
the set {li } of lattice vectors as assigned data. This, in turn, yields the materiality of the
set {r i } in the absence of plastic flow, in accordance with the conventional statement
of the Cauchy–Born hypothesis for purely elastic deformations. Note the contrast with
the representations (4.9), (5.24), and (5.47), for the deformation gradient and its elastic
and plastic factors, in which the ei are material vectors, i.e., ėi = 0.
The present interpretation comports with (6.120) in which the slip-system vectors are
considered to be fixed; these, in turn, may be determined once the set {li } is specified.
Further, for the purpose of integrating the flow rule it is necessary to assign an initial
value of plastic deformation at time t0 , say, and this is another reason why it is necessary
to specify the lattice {li }. If we choose κ = κt0 , for example, then because the referential
lattice coincides with the actual lattice at time t 0 and is therefore measurable in principle,
the specification of {li } provides the initial value of K, and hence that of G.
The fact that the lattice {li } is fixed, independent of x and t, implies that the elements
R of the (discrete) material symmetry group gκi (p) are also fixed, i.e., invariant in time
and the same at all material points of the crystal. This has important consequences for
the further development of the theory.
Naturally, none of the foregoing considerations concerning lattice vectors apply to
isotropic solids, for which gκi (p) ∪ {−I} = Orth. Thus, for these there is no requirement
that R ∈ gκi (p) be uniformly distributed, or even that it be independent of time. We exploit
this observation, in Chapter 7, to simplify the theory for isotropic solids accordingly.

Problem 6.10. Consider a crystalline solid having cubic symmetry relative to κi (p). Let
the edges of the cube be aligned with an orthonormal basis {li }. It is known—see,
for example, the book by Green and Adkins—that the most general homogeneous
quadratic strain-energy function for such a solid is a linear combination of

(E11 + E22 + E33 )2 , E11 E22 + E11 E33 + E22 E33 and E212 + E213 + E223 ,

where Eij = Ê · Sym(li ⊗ lj ) are the Cartesian components of the strain Ê.
(a) Show that the general quadratic strain-energy function for a uniform cubic
solid is thus expressible in the form

U(Ê) = 12 [C1 (E11 + E22 + E33 )2 + C2 (Ē211 + Ē222 + Ē233 ) + C3 (E212 + E213 + E223 )],

where C i are constants (the elastic moduli) and Ēij are the components of DevÊ.
(b) Show that U(Ê) is positive definite if and only if all Ci > 0.
(c) Derive expressions for the Cartesian components S ij of the Piola-Kirchhoff
stress Ŝ = UÊ relative to the basis {li ⊗ lj }.
 Discontinuous fields 111

6.7 Discontinuous fields

It is of interest to extend the notion of dissipation to accommodate discontinuities in
the deformation gradient, to account for the dissipation attending the evolution of phase
boundaries. To this end we first extend the balance laws to allow for such discontinuities.
A basic tool in this endeavor is the extension of Reynolds’ transport theorem to non-
material subregions of the body. Thus, consider a region Rt ⊂ κt and let vn (y, t) be the
normal speed of its boundary ∂Rt in the direction of its exterior unit normal. Then for
any smooth scalar field φ(y, t),

Z Z Z
d
dt φdv = φt dv + φvn da, (6.121)
Rt Rt ∂Rt

where φt = ∂φ(y, t)/∂t. See Chapter 2 of Liu’s book for an elementary proof. If Rt should
happen to be a material region πt , say, i.e., a region that evolves in such a way as to always
contain the same set of material points, then vn = v · n, where v(y, t) = ẏ is the material
velocity and n(y, t) is the exterior unit normal to ∂πt . We thus recover the conventional
statement of the transport theorem:

Z Z Z
d
dt φdv = φt dv + φv · nda. (6.122)
πt πt ∂π t

Similarly, if R(t) ⊂ κ is an evolving, and hence non-material, subvolume of the chosen

reference configuration, with outward normal speed Vν (x, t), then

Z Z Z
d
dt ΦdV = Φ̇dV + ΦVν dA, (6.123)
R R ∂R

for smooth fields Φ(x, t); and, if R = π, a material region, then, because ẋ vanishes by
definition,

Z Z
d
dt ΦdV = Φ̇dV. (6.124)
π π

Consider now a field Φ that suffers a discontinuity across a non-material surface S ⊂

κ. Consider two disjoint material subvolumes π1 , π2 in κ, and suppose S traverses π2 ,
dividing it into two non-material regions R± , so that π2 = R+ ∪ R− . Let (∂π2 )± =
∂R± ∩ ∂π2 . Then, ∂π2 = (∂π2 )+ ∪ (∂π2 )− . It may be helpful to draw a figure. Finally, let
s = π2 ∩ S ; this is the part of S contained in π2 . If Φ is smooth in the separate regions
112 Energy, stress, dissipation, and plastic evolution

R± , and if V ν is the normal speed of S in the direction of the unit normal νS to s directed
from R− to R+ , then it follows from (6.123) that
Z Z Z
d
dt ΦdV = Φ̇dV ∓ Φ± Vν dA, (6.125)
R± R± s

where Φ± , respectively, are the limits of Φ as s is approached from R± . Adding these two
equations then yields
Z Z Z
d
dt ΦdV = Φ̇dV − [Φ]Vν dA, where [Φ] = Φ+ − Φ− (6.126)
π π s

is the jump of Φ across s and π = π2 . On the other hand, (6.124) holds as it stands if
π = π1 .
In the same way, if the Piola stress P(x,t) is smooth in R± , then the divergence
theorem, applied to each of these subregions, yields, upon adding the two statements,
Z Z Z
PνdA = DivPdV + [P]νS dA. (6.127)
∂π π s

This holds if π = π2 , whereas the same statement, but with the integral over s omit-
ted, holds if π = π1 . The conventional statement of the divergence theorem is similarly
generalized as
Z Z Z
a · νdA = DivadV + [a] · νS dA, (6.128)
∂π π s

where a(x, t) is any piecewise smooth vector field; and, for any such scalar field Φ(x, t),
as
Z Z Z
ΦνdA = ∇ΦdV + [Φ]νS dA. (6.129)
∂π π s

Applying the latter to the components of a on a fixed basis and adding the results yields
Z Z Z
a ⊗ νdA = ∇adV + [a] ⊗ νS dA. (6.130)
∂π π s

In particular, then, if the deformation χ(x, t) is continuous,

Z Z
FdV = χ ⊗ νdA, (6.131)
π ∂π

where, of course, F = ∇χ is the deformation gradient.

 Discontinuous fields 113

As an example consider the partwise linear momentum balance (1.24), expressed in

referential form, i.e.,
Z Z Z
d
dt ρ κ vdV = ρ κ bdV + PνdA. (6.132)
π π ∂π
This makes sense even for discontinuous fields and so should be regarded as the funda-
mental form of the balance statement. Applying (6.126) to the left-hand side, we reduce
this to
Z Z Z
PνdA + ρκ (b − v̇)dV = − [ρκ v]Vν dA. (6.133)
∂π π s

For π = π1 the right-hand side of this expression vanishes and we can proceed as in
Chapter 1 to conclude that this is equivalent to the local equation of motion (1.40). If
π = π2 then the latter applies in the separate regions R± .
Consider a sequence of regions π2 that collapse onto s. Assuming |ρκ (b − v̇)| to be
bounded in R± , (6.133) reduces, in the limit, to
Z Z
{P+ νS + P− (−νS )}dA + [ρκ v]Vν dA = 0. (6.134)
s s

As π2 is arbitrary and hence so too its intersection s with S, we may localize and conclude
that

[P]νS + [ρκ v]Vν = 0 on S. (6.135)

As a second example consider the global mass conservation law

Z
d
dt ρκ dV = 0. (6.136)
π

If π = π1 , then, as we have seen in Chapter 1, we may localize and conclude that ρ̇κ = 0
pointwise, which of course also holds in R± in the case when π = π2 . Applying (6.126)
and (6.136) directly to π2 , we have
Z Z
ρ̇κ dV − [ρκ ]Vν dA = 0. (6.137)
π s

Because ρ̇κ vanishes in π \ s this reduces to

Z
[ρκ ]Vν dA = 0, (6.138)
s

which we again localize, concluding that

[ρκ ]Vν = 0 on S. (6.139)

114 Energy, stress, dissipation, and plastic evolution

Accordingly, if S is not a material surface, i.e., if Vν ̸= 0, then [ρκ ] = 0 and (6.135)

becomes

[P]νS + ρκ Vν [v] = 0 on S. (6.140)

This agrees with (6.135) when S is a material surface, and is thus valid in general.
Next, we apply (6.126) with Φ replaced by the deformation gradient:
Z Z Z
d
dt FdV = ḞdV − [F]Vν dA. (6.141)
π π s

To reduce this we differentiate (6.131) and substitute into the left-hand side to obtain
Z Z Z
v ⊗ νdA = ḞdV − [F]Vν dA, (6.142)
∂π π s

 
 
where v = χ̇ is the material velocity. Collapsing π onto s and assuming Ḟ to be bounded
then yields
Z Z
{v+ ⊗ νS + v− ⊗ (−νS )}dA + [F]Vν dA, (6.143)
s s

which we localize as usual to conclude that

[v] ⊗ νS = −[F]Vν on S. (6.144)

Combining this with (6.53), i.e., [F] = f ⊗ νS , then yields

[v] = −Vν f on S. (6.145)

We are finally in a position to examine the dissipation. Recall that this is subject to
inequality (6.57), where D is defined in (6.56). This statement makes sense for fields
that are smooth or not, provided the associated integrals exist. For our present purposes
it is convenient to write the total energy U + K in the form
Z
2
U(π, t) + K(π, t) = ΦdV, where Φ = Ψ + 12 ρκ |v| (6.146)
π

is the energy density. Then,

Z
D(π, t) = P(π, t) − d
dt ΦdV, (6.147)
π
 Discontinuous fields 115

where P(π, t), given by (1.47), is the power supply. Using (6.128), we reduce the latter
to
Z Z Z Z
P(π, t) − ρκ b · vdV = Pt v · νdA = Div(Pt v)dV + [Pt v] · νS dA, (6.148)
π ∂π π s

in which the integrals over π are to be regarded as integrals over π \ s, the two sets having
the same volume measure. In the latter (1.40) is valid, and may be scalar-multiplied by
v to obtain

Div(Pt v) + ρκ b · v = Φ̇ + D, (6.149)

where D is the volumetric dissipation density given by (6.59) and (6.66). Thus,
Z Z
P(π, t) = (Φ̇ + D)dV + [Pt v] · νS dA, (6.150)
π s

which combines with (6.126) and (6.147) to give

Z Z
D(π, t) = DdV + DS dA, (6.151)
π s

where

DS = [Pt v] · νS + Vν [Φ] (6.152)

is the surface density of dissipation.

Problem 6.11. Prove these statements.

To further reduce (6.152) we introduce the arithmetic mean ⟨·⟩ of the limiting values
of a function on either side of a surface of discontinuity. Then for any functions f and g,

[fg] = [f]⟨g⟩ + ⟨f⟩[g]. (6.153)

Accordingly,

[Pt v] · νS = ([Pt ]⟨v⟩ + ⟨Pt ⟩[v]) · νS = [P]νS · ⟨v⟩ + [v] · ⟨P⟩νS

= [P]νS · ⟨v⟩ + ⟨P⟩ · [v] ⊗ νS , (6.154)
2
and (6.140), (6.145), together with [v] · ⟨v⟩ = 12 [|v| ], result in
2
[Pt v] · νS = − 12 Vν [ρκ |v| ] − Vν ⟨P⟩ · [F]. (6.155)
116 Energy, stress, dissipation, and plastic evolution

Using [F] = f ⊗ νS with f = [F]νS , i.e., [F] = [F]νS ⊗ νS , together with (6.140), (6.145),
and the easily derived rule A · a ⊗ b = a · Ab, we have

⟨P⟩ · [F] = [F]νS · ⟨P⟩νS = νS · [Ft ]⟨P⟩νS = νS · ([Ft P] − ⟨Ft ⟩[P])νS , (6.156)

in which, again by (6.140) and (6.145),

[P]νS = ρκ V2ν [F]νS . (6.157)

Now, (6.153) gives ⟨Ft ⟩[F] = [Ft F] − [Ft ]⟨F⟩, so that νS · ⟨Ft ⟩[F]νS = νS · [Ft F]νS − νS ·
[F]t ⟨F⟩νS , where νS · [F]t ⟨F⟩νS = [F]νS · ⟨F⟩νS = νS · ⟨Ft ⟩[F]νS . This gives νS · ⟨Ft ⟩[F]νS =
2 νS · [F F]νS , and (6.152) is finally reduced to
1 t

DS = Vν νS · ([E] + 12 ρκ V2ν [Ft F])νS , (6.158)

where E is the Eshelby tensor, defined by (6.65). Note that DS vanishes if S is a material
surface (Vν = 0).
The global dissipation inequality thus requires that
Z Z
DdV + DS dA ≥ 0 for all π ⊂ κ. (6.159)
π s

If π = π1 , then s = ∅ and localization yields the pointwise inequality D ≥ 0, as before.

Naturally this also holds in R± if π = π2 . Collapsing π onto s results in
Z
DS dA ≥ 0 for all s ⊂ S, (6.160)
s

assuming, of course, that D is bounded. It follows, by localizing as usual, that

DS ≥ 0 on S, (6.161)

and (6.158) once again highlights the central role played by the Eshelby tensor in con-
trolling the dissipation. The latter equation indicates that the function F(Vν ), obtained
by fixing all variables other than V ν therein, then satisfies F(Vν ) ≥ 0 and F(0) = 0.
Thus, F is minimized at Vν = 0, yielding the phase equilibrium condition at a stationary
phase boundary:

νS · [E]νS = 0 on S, when Vν = 0. (6.162)

See the paper by Abeyaratne and Knowles and the book by Liu for derivations of (6.158)
and (6.162) in a more general thermodynamical setting.
 References 117

Problem 6.12. Fill in the steps leading to these two equations.

With these results in hand we may conclude, with reference to Section 2.2 and in the
case of conservative problems, that plastic evolution furnishes the dissipation needed to
ensure that asymptotically stable equilibria minimize the total potential energy. Thus,
suppose there exists a trajectory {χ(x, t), K(x, t)} with

{χ(x, t0 ), K(x, t0 )} = {χ0 (x), K0 (x)} (6.163)

and

lim {χ(x, t), K(x, t)} = {χ∞ (x), K∞ (x)}, (6.164)

t→∞

with

χ̇(x, t)|t0 = 0 and lim χ̇(x, t) = 0. (6.165)

t→∞

Then, because D(κ, t) ≥ 0 in the presence of plastic evolution, (2.42) obtains and yields
(2.45) in the form

E[χ∞ , K∞ ] ≤ E[χ0 , K0 ], (6.166)

where
Z
E[χ, K] = Ψ(∇χ,K)dV − L (6.167)
κ

is the total potential energy in which L is the relevant load potential.

References
Abeyaratne, R., and Knowles, J. K. (1990). On the driving traction acting on a surface of strain
discontinuity in a continuum. J. Mech. Phys. Solids 38, 345–60.
Batchelor, G. K. (Ed) (1958). The Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 1: Mechanics
of Solids. Cambridge University Press, Cambridge, UK.
Deseri, L., and Owen, D. R. (2002). Invertible structured deformations and the geometry of
multiple slip in single crystals. Int. J. Plasticity 18, 833–49.
Epstein, M., and Maugin, G. A. (1990). The energy-momentum tensor and material uniformity
in finite elasticity. Acta Mechanica 83, 127–33.
Green, A. E., and Adkins, J. E. (1970). Large Elastic Deformations. Oxford University Press,
Oxford.
Gurtin, M. E., Fried, E., and Anand, L. (2010). The Mechanics and Thermodynamics of Continua.
Cambridge University Press, Cambridge, UK.
118 Energy, stress, dissipation, and plastic evolution

Havner, K. S. (1992). Finite Plastic Deformation of Crystalline Solids. Cambridge University Press,
.

Cambridge, UK.
Lardner, R. W. (1974). Mathematical Theory of Dislocations and Fracture. University of Toronto
Press, Toronto.
Liu, I-Shih. (2002). Continuum Mechanics. Springer, Berlin.
Nabarro, F. R. N. (1987). Theory of Dislocations. Dover, New York.
Noll, W. (1967). Materially uniform simple bodies with inhomogeneities. Arch. Ration. Mech.
Anal. 27, 1–32.
Rengarajan, G., and Rajagopal, K. R. (2001). On the form for the plastic velocity gradient L p in
crystal plasticity. Math. Mech. Solids 6, 471–80.
Truesdell, C., and Toupin, R. A. (1960). The classical field theories. In: Flügge, S. (Ed.),
Handbuch der Physik, Vol. III/1, pp. 226–793. Springer, Berlin.
Weiner, J. H. (2002). Statistical Mechanics of Elasticity. Dover, New York.
Zangwill, W. I. (1969). Nonlinear Programming. Prentice-Hall, Englewood Cliffs, NJ.
7
Isotropy

The theory for isotropic solids is by far the most highly developed and most widely
applied branch of plasticity. Its dominant position is justified by the fact that most engi-
neering metals are polycrystalline, with the undistorted crystal grains oriented more or
less randomly on the mesoscale, so that, at this scale, the solid responds in the manner
of an isotropic material. Thus, the theory for isotropic solids covers the large major-
ity of engineering applications, and for this reason we devote considerable space to
it here.
We can imagine that as the solid is deformed, the individual grains reorient them-
selves accordingly, and hence that a texture is developed in the material in response to
plastic flow. We might thus conceive of a more general framework than that which we
have considered thus far, one in which the basic kinematic descriptors are independent
deformation and rotation fields, both assigned to the material points of the continuum,
with the latter representing the mesoscopic-level grain orientation. This would interact
with the conventional deformation in accordance with appropriate balance laws and con-
stitutive equations. The resulting model would fall under the umbrella of the theory of
Cosserat continua, a well-established framework conceived precisely to account for rota-
tional degrees of freedom not taken into account by the conventional Cauchy continuum.
Section 98 of the treatise by Truesdell and Noll provides a good survey of the theory of
Cosserat continua. Though the extension of such a model to plasticity would encounter
substantial obstacles, some progress along these lines has recently been made. We refer
the interested reader to the paper by Neff and the book by Epstein and Elżanowski for
further discussion. We shall not delve into this model here, however, as it has yet to
reach a level of maturity that would justify its inclusion in a course. Instead, we will
focus on the standard theory, which remains the primary model to this day, despite its
limitations.

7.1 The flow rule

To begin, we recall that the yield function satisfies the material symmetry condi-
tion (6.110), in which R is an orthogonal tensor belonging to the symmetry group.

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0007
120 Isotropy

This restriction is entirely similar to the condition (6.27) satisfied by the strain-energy
function. We may thus proceed, as in the passage from the latter to (6.34), to conclude
that

FŜ (Rt ŜR) = Rt FŜ (Ŝ)R. (7.1)

This, of course, is valid in general, not only in the case of isotropy. However, in the case
of isotropy it may be used to effect a major simplification of the flow rule.
To this end, recall that invariance of the strain energy or yield function under mate-
rial symmetry transformations is tantamount to their invariance under the replacement
of H by H̄ = HR, where, in the case of isotropy, R is any rotation whatsoever. From
(5.3), this is equivalent to invariance under replacement of K by K̄ = KR—equivalently,
replacement of G by Ḡ = Rt G —with F remaining fixed. To see how this replacement
affects plastic flow, we note that
· −1 −1 t
(Ḡ) Ḡ = Rt ĠG R + Ṙ R, (7.2)

where, recalling the discussion in Section 6.6, we have allowed for the possibility that R
may be time-dependent. In particular, the argument used there to justify the conclusion
that R is fixed in the case of crystalline response is not applicable in the case of isotropy.
Substituting (6.105), we conclude that
· −1 t
(Ḡ) Ḡ = λRt FŜ (Ŝ)R + Rt (Ω + RṘ )R, (7.3)

where Ω is the plastic spin.

Because Ω is skew we can always find a rotation R(t) to nullify the parenthetical term.
As usual, we suppress the passive argument x in the notation as we are concerned with
a fixed material point. Thus, suppose B(t) satisfies the initial-value problem

Ḃ = ΩB with B(0) = B0 , (7.4)

where B0 is a rotation. Let Z = BBt ; then,

Ż = ΩZ − ZΩ, with Z(0) = I. (7.5)

The unique solution is Z(t) = I, implying that B is orthogonal. Further,

−1
J̇B = B∗ · Ḃ = JB tr(ḂB ) = JB trΩ, (7.6)

and this vanishes because Ω is skew. Then, JB(t) = JB(0) = 1 and B is a rotation. Because
the rotation R in (7.3) is arbitrary, we are free to pick R = B and thus obtain
· −1 t
(Ḡ) Ḡ = λ̄FŜ (S̄), with λ̄ = λ and S̄ = B ŜB. (7.7)
 Von Mises’ yield function 121

Problem 7.1. For isotropic materials, show that the Cauchy stress T is invariant under
the replacement H → H̄ = HB. Because F is invariant, this means that the Piola
stress P is likewise invariant.

Accordingly, use of the flow rule (7.7), with initial condition Ḡ(0) = Bt0 G(0), yields
the same deformation and stress fields as would be obtained by using the original flow
rule (6.105). In effect, then, for isotropic materials we can exploit the degree of free-
dom afforded by the material symmetry group to suppress plastic spin in the flow rule
altogether. This major simplification is not possible in crystalline materials.

Problem 7.2. Why not?

Because Ḡ and G are mechanically indistinguishable in the case of isotropy, we may,

without loss of generality, identify the former with the plastic deformation and regard H̄
as the elastic deformation. Then, to ease the notation, we simply drop the overbars and
write (7.7) as
−1
ĠG = λFŜ . (7.8)

7.2 Von Mises’ yield function

Recall the material symmetry condition (6.119),

F̃(DevŜ) = F̃(Rt (DevŜ)R), (7.9)

where, in the case of isotropy, R is an arbitrary orthogonal tensor.

We have argued, in Section 6.5, that the yield function should be approximated by a
quadratic function of DevŜ, for consistency with other assumptions made in the course
of setting up the theory. With reference to the Appendix to Section 6.1, the most general
such function in the case of isotropy is a linear combination of tr(DevŜ), (tr(DevŜ))2 ,
 2
 
and tr((DevŜ)2 ) = DevŜ , the first two of which vanish identically. The most general
yield function of the required kind such that the yield surface F = 0 partitions stress space
into regions defined by F > 0 and F < 0, in which the latter contains the stress-free state,
is then of the form
 2
 
F̃(DevŜ) = 1
2 DevŜ − k2 . (7.10)

This is the famous yield function proposed by von Mises and discussed in the books
by Hill and Prager, among many others. The present derivation, based on material sym-
metry arguments, promotes understanding of its position in the context of the overall
theory. We have derived von Mises’ function as the most general yield function compat-
ible with isotropic material symmetry, with pressure insensitivity, and with our further
122 Isotropy

requirement of a quadratic dependence on Ŝ, the latter to ensure consistency with our
similar approximation of the strain-energy function. Aficionados will object that our
formulation therefore excludes the equally famous Tresca yield function from consider-
ation. This is also compatible with isotropy and pressure insensitivity, but not with our
further consistency requirement. However, on p. 21 of Hill’s treatise, we find the state-
ment “For most metals von Mises’ criterion fits the data more closely than Tresca’s.”
This conclusion, in turn, is based on the results of careful experiments conducted by
Tayor and Quinney—see their classic paper, entitled “The Plastic Distortion of Metals,”
reproduced in G. I. Taylor’s Collected Works. More recently, a critique of Tresca’s crite-
rion, and a strong preference for von Mises’, has been offered in the book by Christensen.
Thus, we consider our restriction to von Mises’ function to be entirely appropriate.
Because the linear space of symmetric tensors may be decomposed as the direct sum
of the five-dimensional space of deviatoric tensors and the one-dimensional space of
spherical tensors, it follows that the yield surface,
√ defined by F = 0, is a cylinder in six-
dimensional Ŝ -space, with axis I and radius 2k. The elastic range, defined by F < 0, is
trivially convex. Here k is the yield stress in shear. That is, if the state of stress is a pure
shear of the form

Ŝ = S(i ⊗ j + j ⊗ i), (7.11)

 2
 
with i and j orthonormal, then DevŜ = 2S2 and the onset of yield occurs when |S| = k.
Here k may be a fixed constant, corresponding to perfect plasticity, or may depend on
appropriate variables that characterize the manner in which the state of the material
evolves with plastic flow. The latter pertains to work hardening, the modeling of which is
the central open problem of the phenomenological theory of plasticity. Some tentative
ideas concerning such modeling are discussed in Chapter 9.
To generate the flow rule for the plastic deformation, we proceed as in (6.117),
obtaining
′
FŜ · Ŝ = F ′ = F̃ ′ = 12 (DevŜ · DevŜ)′
′ ′
= DevŜ · (DevŜ)′ = DevŜ · DevŜ = DevŜ · Ŝ , (7.12)

and hence

FŜ = DevŜ. (7.13)

 2
 
This yields Ŝ · FŜ = DevŜ , which conforms to the dissipation inequality (6.113)
whenever the yield condition F = 0 is satisfied. Finally, (7.8) delivers the flow rule

ĠG−1 = λDevŜ. (7.14)

 The classical theory for isotropic rigid-plastic materials 123

Problem 7.3. With reference to Problem 6.9, derive an expression for the yield func-
tion F(Ŝ) in the case of cubic symmetry. Assume this to be a quadratic function
of DevŜ, and derive an expression for FŜ . State the form of the flow rule (6.105)
in this case, together with appropriate restrictions on the plastic spin arising from
considerations of material symmetry. Note that it is necessary to provide a con-
stitutive prescription for the plastic spin. You need not do so here, but it would
be worthwhile to research this issue independently. See, for example, the paper by
Edmiston et al.

7.3 The classical theory for isotropic rigid-plastic

materials
The elastic strain is invariably small in the case of rate-independent response or under
low strain-rate conditions because it is then bounded by the diameter of the elastic range.
If the overall strain is nevertheless large, then the main contribution to the strain comes
from plastic deformation. In this case it is appropriate to consider the idealization of zero
elastic strain, which entails the restriction Ht H = I. The elastic deformation is therefore
a rotation field, which we denote by Q. The stress Ŝ is then effectively a symmetric
Lagrange-multiplier tensor associated with the constraint of vanishing elastic strain. It
is constitutively indeterminate, as is the stress in a rigid body subject to the constraint
Ft F = I. In the present context it is subject only to the yield condition F = 0 and the
balance laws, but is otherwise arbitrary. Further, JH = 1 and the relation JH T = HŜHt
between the Cauchy stress T and Piola–Kirchhoff stress Ŝ, which may be inferred from
(6.9) and (6.11), becomes

Ŝ = Qt TQ. (7.15)

Therefore,

DevŜ = Dev(Qt TQ) = Qt TQ − 13 tr(Qt TQ)I = Qt τQ, (7.16)

where

τ = devT ≡ T − 31 (trT)i (7.17)

is the deviatoric part of the Cauchy stress in which i is the spatial identity. This differs
from the deviatoric operator Dev defined in (6.115), based on the referential identity.
The distinction is necessitated by the fact that the plastic deformation G is insensitive to
superposed rigid-body motions, implying that κi (p), like Tκ , is similarly insensitive. It is
appropriate, then, to adopt I as the identity for both vector spaces. On the other hand,
T preserves Tκt , so that (7.17) makes sense with the identity interpreted as the spatial
identity. In contrast, Q behaves like the shifter 1, mapping κi (p) to Tκt . Thus, Qt maps
the latter to the former. This implies that Qt Q = I and QQt = i. To make this explicit
124 Isotropy

we introduce a rotation Q̄, mapping κi (p) to itself, such that Q = 1Q̄. This makes sense
because the composition of rotations is a rotation, and, as we have seen in Section 2.3,
the shifter is a rotation. We then have
t
F̃(DevŜ) = F̃(Qt τQ) = F̃(Q̄ 1t τ1Q̄). (7.18)

Now, Q̄ is a rotation of the same kind as the rotation R appearing in the symmetry
condition (7.9); that is, R , like Q̄, rotates vectors in κi (p) to vectors in κi (p). Accordingly,
because R is otherwise arbitrary in the case of isotropy, we may invoke (7.9), with R = Q̄,
t
to conclude that F̃(Q̄ 1t τ1Q̄) = F̃(1t τ1), and hence that

F̃(DevŜ) = F̃(1t τ1). (7.19)

This is meaningful insofar as 1t τ1 is a tensor of the same kind as DevŜ; i.e., both preserve
κi (p). Substituting the von Mises form and using 11t = i, we have

F̃(DevŜ) = 1 t t
2 tr(1 τ11 τ1) − k2 = 12 tr(1t τ2 1) − k2
= 1 2 t
2 tr(τ 11 ) − k2 = 12 tr(τ2 ) − k2 . (7.20)

Thus, F(Ŝ) = F̃(τ), where

2
F̃(τ) = 1
2 |τ| − k2 . (7.21)

This is the conventional form of von Mises’ yield function found in the literature. It
emerges from the formalism presented here when the elastic strain vanishes.
Using (7.16), we recast the flow rule (7.14) as

Q(ĠG−1 )Qt = λτ. (7.22)

This may be reduced to the classical form via the decomposition

L=D+W (7.23)

of the spatial velocity gradient L = gradv into the sum of the symmetric straining tensor
D and the skew vorticity tensor W. Then with L = ḞF−1 and (5.3) we find, in the present
specialization to H = Q, that

L = Q̇Qt + Q(ĠG−1 )Qt , (7.24)

in which the first term is skew whereas the second, according to (7.22), is symmetric. The
uniqueness of the decomposition then yields W = Q̇Qt , together with D = Q(ĠG−1 )Qt ,
and hence the classical flow rule

D = λτ; λ ≥ 0, (7.25)

proposed, albeit in a different form, of course, by St. Venant, in 1870, and Lévy, in 1871.
 The classical theory for isotropic rigid-plastic materials 125

The Cauchy stress is given by

T = τ − pi, (7.26)

where p is the pressure field. This pressure field is associated with the incompressibility
of the material, i.e., with the constraint trD = 0, which follows from (7.25). It is not
related to the constitutive response of the material, but is instead an additional unknown
to be obtained by solving the equations of motion, augmented by the constraint equation
together with any subsidiary conditions. Chadwick’s book includes a discussion of the
pressure associated with incompressibility, and of similar modifications to the expression
for the stress when alternative constraints are operative.
Equation (7.25) is the fundamental equation of the classical theory, which predates
the modern theory for finite elastic–plastic deformations, as discussed here, by at least a
century. Its derivation in the framework of the modern theory, based on material sym-
metry restrictions pertaining to the concept of an undistorted state, is significant insofar
as it promotes confidence in both the classical and modern theories. It is remarkable
that the original form, deduced without the benefit of the tools of modern continuum
mechanics, emerges in such a straightforward manner from the modern theory.
Indeed, prior to the advent of the modern theory, the classical theory was regarded
as a separate model, seemingly unrelated to the theory for elastic–plastic response and
without a clear connection to the notion of isotropic material symmetry as espoused in
continuum mechanics. To elaborate, we take the inner product of (7.25) with itself to
find λ, and, recalling that this equation is operative when F̃(τ) = 0, obtain τ = τ̂(D),
where
√
2k
τ̂(D) = |D| D. (7.27)

This is an isotropic function; that is, τ̂(QDQt ) = Qτ̂(D)Qt for any (spatial) rotation Q.
However, this feature is a consequence of frame invariance, or invariance of τ̂(D) under
superposed rigid-body motions. It is not related to the notion of isotropic material sym-
metry, except to the extent indicated in our derivation of (7.25). The claim, prevalent in
the classical literature, to the effect that (7.25) is a model of isotropic solids, is thus dif-
ficult to reconcile with the modern notion of isotropy as discussed in the foregoing, and
hence a potential source of confusion which we have endeavored here to clarify. It seems
reasonable to view this difficulty as one of the reasons why plasticity theory received so
little attention from the main proponents of the twentieth-century renaissance of con-
tinuum mechanics. In fact, from the modern perspective
√ (7.27) is more appropriately
interpreted as a model, operative when |τ| = 2k, of a non-Newtonian fluid, with a
variable viscosity η given by
√
2η = 2k/ |D| . (7.28)
√
Otherwise, i.e., when |τ| < 2k, we have D = 0, because Ġ vanishes when F < 0. It is
well known—see the books by Chadwick and Gurtin, for example—that in this case the
deformation χ is rigid; hence, the name rigid-plastic material attached to this model.
126 Isotropy

Problem 7.4. A popular yield function for isotropic soils is the Drucker–Prager
function  2
 
F(Ŝ) = 12 DevŜ − [α(trŜ) + k]2 ,

where α and k are material constants. What is the corresponding flow rule for
ĠG−1 ? Reduce this to an equation for the straining tensor D in the case of rigid-
plastic response. Is the material compressible or incompressible? (For a thorough
discussion of soil plasticity, consult the book by Yu.)

7.4 Bingham’s model of viscoplasticity

We will study the rigid-plastic model in some detail in Section 7.5, but, to expand on
our remarks concerning the interpretation of (7.27) as a model of viscous fluids, we
first pause to consider a model of viscoplastic materials proposed, in 1922, by Bingham.
Prager’s book contains a thorough discussion. This model is obtained simply by adding
a conventional viscous term to the right-hand side of (7.27),
√
2k
τ = (2η + |D| )D, (7.29)

where η is now a fixed positive constant, and the Cauchy stress is again of the form
(7.26). If we were to include only 2η within the parentheses, then we would have the
classical Newtonian fluid in which η is the constant shear viscosity. Accordingly, (7.29)
combines the effects of viscosity and plasticity, and thus furnishes a prototypical model
of viscoplastic materials. Taking its norm, we have
√
|τ| = 2η |D| + 2k. (7.30)

Thus, if the motion is non-rigid,

√ is, if D ̸= 0, then the stress necessarily lies outside the
that√
yield surface, i.e., |τ| ≥ 2k. If |τ| < 2k, then this model is not applicable; we assume in
this case that D vanishes, as before, and hence that the motion is rigid. Equation (7.29)
thus models rigid-viscoplastic materials. √
√ observe that |τ| is an increasing function of |D| whenever |τ| ≥ 2k, whereas
Further,
|τ| → 2k as η and/or |D| approach zero. Thus, (7.29) exhibits rate-sensitivity—a fea-
ture of the high-temperature response of real metals—and reverts to the conventional
rate-independent model in the slow-deformation limit. It also furnishes a fairly realistic
rheological model of wet paint.
The viscoplastic theory plays an important role in the solution of dynamical problems,
extensive discussions of which may be found in the books by Cristescu, Kachanov, and
Lubliner.
 Bingham’s model of viscoplasticity 127

7.4.1 Example: Steady channel flow

To illustrate the main features of Bingham’s model we apply it to the simple problem of
steady rectilinear channel flow. Our solution is adapted from that given in Prager’s book.
Thus, consider the steady flow of a Bingham material in a channel: a region defined,
in terms of Cartesian coordinates x, y, z, by −∞< x <∞, 0< y <h, and −∞< z <∞. We
identify these coordinates with the projections of the position y ∈ κt of a material point
(4.1) onto the elements of a fixed orthonormal basis {i, j, k}; thus, y = xi + yj + zk.
We use the theory to investigate the possibility of the existence of a velocity field of
the form

v = v(y)i. (7.31)

The spatial velocity gradient L follows from

Ldy = dv = v′ (y)idy = v′ (y)i(j · dy) = {v′ (y)i ⊗ j}dy, (7.32)

giving L = v′ (y)i ⊗ j and

D = 12 v′ (y)(i ⊗ j + j ⊗ i), (7.33)

and hence trD = 0, so that the assumed flow is kinematically possible in our idealized
viscoplastic material. Further, as Lv = 0 and because vt vanishes, that is, because v is
steady in the sense that it is independent of t when expressed as a function of y and t,
we have v̇ = 0, where v̇ = vt + Lv is the material acceleration. Assuming negligible body
force, (1.32) then requires that

divT = 0, (7.34)
√
with T given by (7.26) and (7.29), provided that |τ| ≥ 2k. Later, we will need to
interpret this inequality in terms of the function v(y). Preliminary to this, we compute

D2 = 14 (v′ )2 (i ⊗ i + j ⊗ j) (7.35)
√
and |D| = tr(D2 ) = 12 (v′ )2 . Thus, |D| = |v′ (y)| / 2, and (7.29) yields
2

τ = τ(y)(i ⊗ j + j ⊗ i), where τ(y) = ηv′ (y) + kv′ (y)/ |v′ (y)| . (7.36)

Substituting this into (7.26), we find that (7.34) reduces to

gradp = τ′ (y)i, (7.37)

which implies that ∂p/∂y = ∂p/∂z = 0, and gradp = p′ (x)i, where p′ (x) = τ′ (y). Thus,
p(x) is the linear function of x given by p(x) = xτ′ (y) + p0 , where p0 is a constant, and
since this is independent of y, we must have xτ′′ (y) = 0; hence,
128 Isotropy

τ′ (y) = −C and p(x) = p0 − Cx, (7.38)

where C is another constant. We take C > 0, corresponding to a negative pressure

gradient along the x-axis.
We assume the flow to be attached to stationary rigid walls at y = 0, h, where the
conventional no-slip conditions hold, and we seek a velocity profile that is symmetric
about the plane y = h/2. This means that v′ (y) vanishes at y = h/2. Because of symmetry
we further suppose that τ(h/2) = 0, this following automatically in the case of purely
viscous response (k = 0). Thus,

τ(y) = 12 C(h − 2y). (7.39)

For the purely viscous fluid this gives v′ (y) = τ(y)/η, and hence the classical parabolic
profile

v(y) = 21η Cy(h − 2y), (7.40)

satisfying the√
no-slip conditions v(0) = v(h) =
√ 0. However, in the viscoplastic case we
require |τ| ≥ 2k, where, from (7.36)1 , |τ| = 2 |τ| . Thus, viscoplastic flow is possible
only if

|τ(y)| ≥ k. (7.41)

Consider the region where τ(y) > 0. Then we have flow wherever

τ(y) ≥ k, (7.42)

i.e., in the region

y ≤ ȳ, where ȳ = h
2 − Ck , (7.43)

which makes sense provided that ȳ > 0, i.e., provided that the pressure gradient C
satisfies
2k
C> h . (7.44)

Evidently, then, these conditions require that y < h/2. In view of the no-slip condition
v(0) = 0, we assume that v′ (y) ≥ 0 in this region, and conclude, from (7.36)2 , that

v′ (y) = η1 (τ − k). (7.45)

The assumption about v′ (y) is then consistent with the requirement (7.42). Substituting
(7.39) and imposing the no-slip condition, we arrive at the velocity field

v(y) = 2Cη [y(h − y) − 2k

C y]; 0 ≤ y ≤ ȳ. (7.46)
 Plane strain of rigid-perfectly plastic materials: Slip-line theory 129

In the region ȳ < y ≤ h/2 we still have τ(y) ≥ 0, but |τ(y)| < k. Thus, D = 0 and
hence v′ (y) = 0, implying that that the velocity is uniform in this region:

v(y) = v(ȳ); ȳ<y ≤ h/2. (7.47)

This region moves as a rigid plug, provided, of course, that (7.44) is satisfied. Otherwise,
the no-slip condition implies that there is no flow at all. Accordingly, flow initiates when
the pressure gradient attains the critical value C = 2k/h. Naturally the solution in the
region h/2 < y < h follows from that already derived by invoking symmetry.

Problem 7.5. Position in a cylindrical polar-coordinate system {r, φ, z} is given by

y = rer (φ) + zk,

p
where r = x2 + y2 , tan φ = y/x, er (φ) = cos φi1 + sin φi2 , and k = i3 . Using the
basis {er , eφ , k}, where eφ (φ) = k × er , we can write the Cauchy stress tensor in
the form

T = Trr er ⊗ er + Tφφ eφ ⊗ eφ + Tzz k ⊗ k + Trφ (er ⊗ eφ + eφ ⊗ er )

+Trz (er ⊗ k + k ⊗ er ) + Tφz (eφ ⊗ k + k ⊗ eφ ).

Assuming all components Trr , Trφ , etc., to depend only on r and φ, use the methods
of Chapter 3, or otherwise, to show that
h  i
divT = ∂T + 1 T −T + ∂ T er
∂r rr r rr φφ ∂φ r φ
h  i
+ ∂ Trφ + 1r 2Trφ + ∂ Tφφ eφ
∂r ∂φ
h  i
+ ∂ Trz + 1r Trz + ∂ Tφz k.
∂r ∂φ

Problem 7.6. Use Bingham’s model to solve the problem of anti-plane viscoplastic
flow in a circular pipep aligned with the z-axis. Assume a velocity field of the form
v = w(r)k, where r = x2 + y2 is the radius in a system of cylindrical polar coor-
dinates. Suppose the material occupies the annular region (0 <)a ≤ r ≤ b and that
w(b) = 0 and w(a) = W. Assume that the axial pressure gradient vanishes and find
the velocity field w(r).

7.5 Plane strain of rigid-perfectly plastic materials:

Slip-line theory
The rate-independent rigid-perfectly plastic model of Section 7.3 takes a particularly
interesting form when specialized to plane-strain conditions. The theory, thus special-
ized, lends itself to the solution of many practically important problems and is thus the
subject of an extensive literature. For this reason we devote a fair amount of attention to
the plane-strain theory here.
130 Isotropy

Thus, we confine attention to velocity fields of the form v = vα (y1 , y2 )iα , where
Greek indices are summed, here and henceforth, from 1 to 2, {y1 , y2 } is a system of
plane Cartesian coordinates, and the iα are fixed unit vectors defined by (3.12). The
latter equation implies, as is well known, that there is no distinction between co- and
contravariance when using such coordinates. Accordingly, we will adopt the common
convention of simply lowering all indices. Then, D = Dαβ iα ⊗ iβ , where 2Dαβ = vα,β +
vβ,α , with vα,β = ∂vα /∂yβ , and Eq. (7.25) yields τ = ταβ iα ⊗iβ , implying that p = −T33 .
The pressure field is equal to the confining stress required to maintain the plane-strain
condition. Using 3p = −trT, we conclude that p = − 21 Tαα .

7.5.1 Stress, equilibrium

The yield criterion F = 0 reduces, with (7.21), to

2k2 = ταβ ταβ = (Tαβ + pδαβ )(Tαβ + pδαβ )

= Tαβ Tαβ + 2pTαα + p2 δαα

= Tαβ Tαβ − 2p2 = Tαβ Tαβ − 12 (Tαα )2 , (7.48)

where δαβ is the Kronecker delta, or, more simply, to

(T11 − T22 )2 + 4T212 = 4k2 . (7.49)

Problem 7.7. Show that the principal stresses are

T1 = −p + k, T2 = −p − k and T3 = −p.

It follows that

T = −pi + k(u1 ⊗ u1 − u2 ⊗ u2 ), (7.50)

where {ui }, with u3 = i3 , are the orthonormal principal stress axes. Let t and s be plane
orthonormal vector fields such that
√ √
u1 = 2
2 (s + t), u2 = 2
2 (s − t). (7.51)

Then,

T = −pi + τ, where τ = k(t ⊗ s + s ⊗ t), (7.52)

and this implies that k is the shear stress resolved on the s, t -axes.
 Plane strain of rigid-perfectly plastic materials: Slip-line theory 131

Problem 7.8. If a(y) and b(y) are vector fields, show that div(a ⊗ b) = (grada)b +
(divb)a, where grad and div are the gradient and divergence operators based on
position y.

Accordingly, in a perfectly plastic material (k = const.), equilibrium without body

force, i.e., divT = 0, requires that

grad(p/k) = (gradt)s + (divs)t + (grads)t + (divt)s. (7.53)

If we introduce a two-dimensional scalar field, θ(y), such that

t = cos θi1 + sin θi2 , s = − sin θi1 + cos θi2 , (7.54)

then

dt = sdθ and ds = −tdθ, (7.55)

and thus, with dθ = gradθ · dy,

gradt = s ⊗ gradθ, grads = −t ⊗ gradθ, (7.56)

and therefore, with the divergence defined by (3.71),

divt = s · gradθ and divs = −t · gradθ. (7.57)

We thus reduce (7.53) to

grad(p/2k) = (s · gradθ)s − (t · gradθ)t, (7.58)

which is equivalent to the two equations

t · grad(p/2k + θ) = 0 and s · grad(p/2k − θ) = 0, (7.59)

due to Prandtl and Hencky.

These require that p/2k ± θ take constant values on the trajectories defined by
dy2 /dy1 = tan θ, − cot θ, respectively. The latter are the characteristic curves of a hyperbolic
system of partial differential equations for the stress fields p and θ. Remarkably, the stress
is statically determinate ; that is, granted suitable boundary conditions, it can be deter-
mined directly from the equilibrium equations alone without knowledge of the motion.
These striking features of the theory of perfectly plastic materials contrast sharply with
situation encountered in the theory of elasticity, for example, where the equilibrium
equations for the displacement field are typically of elliptic type.
132 Isotropy

7.5.2 Velocity field

Naturally, there must be a velocity field to accompany the stress field, be it statically
determinate or not. To obtain this, it proves advantageous to decompose the velocity
field in the variable {s, t} basis. Thus,

v = vt t + vs s. (7.60)

To compute the velocity gradient L = gradv, we use (7.55), obtaining

Ldy = dv = dvt t + vt sdθ + dvs s − vs tdθ

= (gradvt · dy)t + vt s(gradθ · dy) + (gradvs · dy)s − vs t(gradθ · dy). (7.61)

Thus,

L = t ⊗ gradvt + vt s ⊗ gradθ + s ⊗ gradvs − vs t ⊗ gradθ, (7.62)

and

2D = t ⊗ gradvt + gradvt ⊗ t + s ⊗ gradvs + gradvs ⊗ s

+ vt (s ⊗ gradθ + gradθ ⊗ s) − vs (t ⊗ gradθ + gradθ ⊗ t). (7.63)

Substituting into (7.25) and (7.52)2 , we derive

t · Dt = 0 and s · Ds = 0, (7.64)

which together imply that the motion is isochoric, as required, with vanishing extension
rates in the directions of t and s.

Problem 7.9. Show that Eqs. (7.64) are equivalent to the pair

t · (gradvt − vs gradθ) = 0 and s · (gradvs + vt gradθ) = 0.

These are the celebrated Geiringer equations. If the stress field is known, they furnish
linear partial differential equations for the velocity components vt and vs .
Remarkably, this theory allows for the existence of tangential discontinuities of the
velocity field along the trajectories with unit tangents s and t. To see that only tangential
discontinuities are allowed, consider the local condition of conservation of mass at a
surface of discontinuity. We derived the referential form of this condition, holding in
κ, in Chapter 6—see (6.139). Here, however, we require its counterpart in κt . Rather
than derive this condition anew, we refer to its derivation in Liu’s book. The relevant
condition, given by Liu’s Eq. (2.32), is

v[ρ] = [ρv] · n, (7.65)

where [·] is the jump at the discontinuity surface with unit normal n, and v is the normal
speed of the surface in the direction of n. Recalling that ρκ = ρJF and assuming ρκ to be
 Plane strain of rigid-perfectly plastic materials: Slip-line theory 133

continuous, we conclude, in the case of the isochoric motions (JF = 1) that are possible
in incompressible materials, that [ρ] vanishes, and hence that [v] · n = 0. Any velocity
discontinuity must then be purely tangential to the discontinuity surface.
This implies that the normal velocity vs (resp., vt ) is continuous across the curve with
unit-tangent field t (resp., s). The tangential velocity vt (resp., vs ) may assume unequal
values on either side of such a curve, and hence so too the tangential derivative t · gradvt
(resp., s · gradvs ). Subtracting the limits of the relevant Geiringer equation on either side
of the curve, we conclude that t · grad[vt ] (resp., s · grad[vs ]) vanishes on the curve with
normal s (resp., t), so that the slip [vt ] (resp., [vs ]), if non-zero, is uniform along it. This
explains why the present model is referred to as slip-line theory.
The literature on this subject, relating in particular to the solution of problems involv-
ing metal forming and extrusion, is enormous. The books by Nadai, Hill, Kachanov,
Johnson, and Haddow, Johnson and Mellor, and Lubliner, and the article by Geiringer
describe further theory and many worked-out solutions. Numerical solutions are dis-
cussed in the article by Collins.
There is also a plane-strain slip-line theory for crystalline materials. This is not as
nearly as well known as the theory for isotropic materials. It is substantially more com-
plicated than the isotropic theory, and therefore not discussed here. However, an account
of this subject, accessible to readers of this book, may by found in the paper by Gupta
and Steigmann and the references cited therein.
We limit ourselves to a discussion of a few explicit solutions, both to illustrate some
of the more unusual features of the theory and to whet the reader’s appetite for the vast
literature on the subject.

7.5.3 Cartesian form of the equations

Most of the literature is based on equations cast in terms of Cartesian coordinates.

Problem 7.10. Using (7.54), show that T = Tαβ iα ⊗ iβ − pi3 ⊗ i3 , where

T11 = −p − k sin 2θ, T22 = −p + k sin 2θ, and T12 = T21 = k cos 2θ.

Equilibrium without body force, i.e., Tij,j = 0, thus requires that p,3 = 0 and
Tαβ,β = 0, the latter yielding

p,1 + 2k(θ,1 cos 2θ + θ,2 sin 2θ) = 0 and p,2 + 2k(θ,1 sin 2θ − θ,2 cos 2θ) = 0. (7.66)

Note that if, at a particular point, we orient the basis {iα } such that i1 = t and i2 = s,
then θ vanishes at that point, and these equations reduce to

0 = p,1 + 2kθ,1 = i1 · grad(p + 2kθ) = t · grad(p + 2kθ), and, similarly,

0 = p,2 − 2kθ,2 = i2 · grad(p − 2kθ) = s · grad(p − 2kθ), (7.67)

which, of course, are just the Prandtl–Hencky equations.

134 Isotropy

With the Cartesian form of the equations in hand, the obvious thing to do when
searching for simple solutions is to assume that p or θ depends on just one of the coor-
dinates. Let’s consider the possibility that θ depends only on y2 ; Eqs. (7.66) reduce to

p,1 = k(cos 2θ)′ and p,2 = k(sin 2θ)′ , (7.68)

where (·)′ = d(·)/dy2 . Assuming p to be twice differentiable, we can eliminate it as

follows:

k(cos 2θ)′′ = p,12 = p,21 = k(sin 2θ),21 = 0. (7.69)

Then cos 2θ is a linear function of y2 . One such solution, due to Prandtl, is

p
cos 2θ = −y2 /h, sin 2θ = − 1 − (y2 /h)2 , (7.70)

where h is a positive constant. This is meaningful provided that |y2 | ≤ h, and from (7.68)
we have

p = −k(y1 /h) + f(y2 ), where f ′ = k(sin 2θ)′ . (7.71)

Thus,
p
p = p0 − k[y1 /h + 1 − (y2 /h)2 ], (7.72)

where p0 is a constant, and the equilibrium stress field follows from the solution to
Problem 7.10. Our solution is adapted from Prager’s version of Prandtl’s solution.
To obtain the velocity field, we invoke (7.25) with

τ = k sin 2θ(i2 ⊗ i2 − i1 ⊗ i1 ) + k cos 2θ(i1 ⊗ i2 + i2 ⊗ i1 ). (7.73)

Then,

D = α(i1 ⊗ i1 − i2 ⊗ i2 ) + β(i1 ⊗ i2 + i2 ⊗ i1 ), (7.74)

where

α = −λk sin 2θ and β = λk cos 2θ. (7.75)

q
Because λ ≥ 0 we then have λk = α2 + β2 and
q q
2
α/ α2 + β = − sin 2θ, β/ α2 + β2 = cos 2θ, (7.76)
 Plane strain of rigid-perfectly plastic materials: Slip-line theory 135

specializing, in the present solution, to

p q q
α = 1 − (y2 /h) α2 + β2
2 and β = −(y2 /h) α2 + β2 . (7.77)

It follows from the expression for D that the components of the velocity field satisfy

v1,1 = −v2,2 = α and 1

2 (v1,2 + v2,1 ) = β. (7.78)

Accordingly, α and β are not arbitrary. Assuming the velocity components to be thrice
differentiable, we derive the compatibility condition

α,11 − α,22 + 2β,12 = 0, (7.79)

which, with reference to (7.77), is easily seen to be solved if

q p
α2 + β2 = C
h( 1 − (y2 /h)2 )−1 ; C = const. (7.80)

For we then have

p
α= C
h and β = − Ch yh2 ( 1 − (y2 /h)2 )−1 , (7.81)

a constant and a function of y2 alone, respectively.

The first and second of Eqs. (7.78) integrate to give

v1 = C
h y1 + f(y2 ) and v2 = − Ch y2 + g(y1 ), (7.82)

where f and g are functions to be determined. These yield v1,2 = f ′ (y2 ) and
v2,1 = g ′ (y1 ), which combine with the third equality in (7.78) and the second in (7.81)
to give g ′′ (y1 ) = 0 and

v2 = − Ch y2 + Dy1 + E, (7.83)

where D and E are constants. Then, v1,2 = 2β − D, yielding

p
v1 = C[y1 /h + 2 1 − (y2 /h)2 ] − Dy2 + F, (7.84)

where F is another constant.

Clearly E and F are rigid-body translational velocities. Further, the terms involving D
contribute a fixed tensor to the vorticity W, and therefore represent a rigid-body rotation.
Suppressing the rigid-body terms, we thus have the velocity field
p
v1 = C[y1 /h + 2 1 − (y2 /h)2 ], v2 = −Cy2 /h, (7.85)

valid in the region defined by |y2 | ≤ h.

136 Isotropy

Evidently v2 = ∓C at y2 = ±h, and so this solution simulates the squeezing, assuming

C > 0, of a layer of thickness 2 h between rigid walls approaching each other at the relative
speed 2C. The solution predicts that T12 = ∓k at y2 = ±h but does not satisfy the no-
slip conditions v1 = 0 there. We also note that |D12 | (= |β|) → ∞ as |y2 | → h. Thus,
if the material is viscoplastic, then the viscosity cannot be neglected when y2 is close to
±h. In this case the present solution is valid outside a thin boundary layer adjoining the
walls in which Bingham’s model may be used to meet a no-slip condition. See Prager’s
book for a discussion of the relevant boundary-layer theory. Further discussion of the
present solution is given in the books by Nadai and Hill, the latter including corrections
to accommodate finite-width strips.
Corresponding slip lines may be determined, if desired, by using (7.70)1 to obtain

dy2 /dy1 = 2h(sin 2θ)dθ/dy1 . (7.86)

Then, on the trajectory with unit tangent t, where dy2 /dy1 = tan θ, and on that with unit
tangent s, where dy2 /dy1 = − cot θ, we have

y1 = h(sin 2θ + 2θ) + const. and y1 = h(sin 2θ − 2θ) + const., (7.87)

respectively. These, together with (7.70)1 , are the parametric equations of a pair of
orthogonal cycloids.

7.5.4 Further theory for plane strain

To investigate the possibility of finding further solutions, we return to the Prandtl–
Hencky form of the equations. It proves convenient for this purpose to retain Cartesian
coordinates, but to ease the notation we label them y1 = x and y2 = y and reserve
subscripts to identify partial derivatives. It is also convenient to introduce orthogonal
curvilinear coordinates (α, β) to parametrize the slip lines (not to be confused with the
use of the same symbols in the previous subsection). We arrange these such that β is
constant on a trajectory with unit tangent t, and α is constant on one having unit tan-
gent s. We refer to these trajectories as α-lines and β-lines, respectively. They satisfy the
equations

( dy
dx )|β = tan θ and dx )|α = − cot θ.
( dy (7.88)

The Prandtl–Hencky equations (7.59) imply that ±p/2k + θ, respectively, assume

constant values on these curves, and hence that

p/2k + θ = ξ(β) and − p/2k + θ = η(α), (7.89)

for some functions ξ and η. Similarly, the Geiringer equations require that

dvt − vs dθ = 0 on α-lines, and dvs + vt dθ = 0 on β-lines. (7.90)

 Plane strain of rigid-perfectly plastic materials: Slip-line theory 137

Using the chain rule, we have dξ = ξx dx + ξy dy = (ξx + ξy dy/dx)dx. Then, because

dξ = 0 on an α-line,

ξx + (tan θ)ξy = 0. (7.91)

In the same way, because dη = 0 on a β-line,

ηx − (cot θ)ηy = 0. (7.92)

Noting, from (7.89), that

θ = 12 (ξ + η), (7.93)

these are seen to furnish a pair of quasilinear partial differential equations for the fields
ξ = ξ̂(x, y) and η = η̂(x, y), solutions to which may be substituted into (7.89) to furnish
the associated stress fields.
A standard approach to systems such as this is the hodograph transformation, whereby
the roles of the dependent and independent variables are interchanged to arrive at a linear
system for the latter, regarded as functions of the former. That is, we seek a system to be
solved for the functions x = x̂(ξ, η) and y = ŷ(ξ, η). To derive this, we invert the system
    
dξ ξx ξy dx
= , (7.94)
dη ηx ηy dy

obtaining
    
dx −1 ηy −ξy dξ
=j , where j = ξx η y − η x ξy , (7.95)
dy −ηx ξx dη

which implies, if j ̸= 0, that

  !
ξx ξy yη −xη
= j −y xξ . (7.96)
ηx ηy ξ

With this proviso, Eqs. (7.91) and (7.92) reduce to the linear, variable-coefficient system

yη − xη tan θ(ξ, η) = 0, yξ + xξ cot θ(ξ, η) = 0 (7.97)

for the functions x = x̂(ξ, η) and y = ŷ(ξ, η), where θ(ξ, η) is given by (7.93). As explained
in the books by Hill and Kachanov, a further transformation reduces these to a pair of
linear decoupled telegraph equations, solutions to which are developed in Chapter 6,
Section 7 of the former.
138 Isotropy

The central difficulty with this procedure, of course, is that to determine the stress
field in the x, y -plane, the solutions x̂(ξ, η) and ŷ(ξ, η) thus derived must be inverted
to obtain the functions ξ̂(x, y) and η̂(x, y), the possibility of doing so hinging on the
condition j ̸= 0. Fortunately, the case j = 0 excluded by this procedure leads to a number
of useful simple solutions, and so we confine ourselves to this case in the remainder of
this subsection. Naturally, then, the system (7.97) no longer applies.
To investigate this case we multiply (7.91) by ηy and invoke the condition j = 0,
together with (7.92), obtaining ξy ηy / sin 2θ = 0. Similarly, multiplication of (7.92) by ξy
and use of (7.91) yields ξx ηx / sin 2θ = 0. There are three possibilities: (a) ξ and η are
both constant. Then Eqs. (7.89) imply that p and θ are also constant. The slip lines of
this constant state are orthogonal straight lines. (b) ηx and ξy both vanish. Then (7.91)
implies that ξ is constant. (c) ηy and ξx both vanish, and (7.92) implies that η is constant.
In case (c) we use (7.93) to infer that ξx = 2θx and ξy = 2θy , and (7.91) reduces to
θx + (tan θ)θy = 0. With dθ = (θx + θy dy/dx)dx, this, in turn, implies that θ is constant
on α-lines. Accordingly, the latter are straight. From (7.89) we have that p/k = ξ(β) − η,
so that the stresses are constant along a given α-line (β = const.), but vary from one such
line to another. This field is called a simple state. Case (b) is entirely similar and also gives
rise to a simple state.
An example of a constant state is afforded by a uniform state of stress in a region
adjacent to a straight traction-free boundary. If the boundary has unit normal i1 , say,
then T 11 and T 12 vanish, and if T22 > 0, then the larger and smaller in-plane principal
stresses are T1 = T22 and T2 = 0. With reference to Problem 7.6 we then have T22 = 2k
and p = −k. The principal stress axes are u1 = i2 and u2 = −i1 , and (7.51) may be
solved to obtain
√ √
t= 2
2 (i1 + i2 ), s= 2
2 (i2 − i1 ); thus, θ = π/4. (7.98)

Problem 7.11. Use (7.52) to verify that the zero-traction condition, Ti1 = 0, is satisfied.

To illustrate a simple state, consider the problem of a straight, semi-infinite traction-

free crack and the tentative slip-line field sketched in Figure 7.1. This is partitioned into
the three regions labeled A, B, and C, in which A and C are constant states and B is a
simple state. In region A, T 22 and T 12 vanish, and, assuming T 11 to be positive, T11 = 2k.
We also have θ = 3π/4 and p = −k.
If we trace a β-line—a curve on which α is constant—from region A into region C,
where θ = π/4, and make use of (7.89)2 , we obtain −p/2k + π/4 = k/2k + 3π/4. Thus,
p = −(1 + π)k in region C.

Problem 7.12. Use (7.52) to derive the Cartesian stress components T11 = πk, T22 =
(π + 2)k, and T12 = 0 in region C.

The state in region B is similarly obtained by tracing a β-line from region A. Thus,
−p/2k + φ = k/2k + 3π/4, where φ is the azimuthal angle in a system of plane polar
 Plane strain of rigid-perfectly plastic materials: Slip-line theory 139

B β
rigid

A C

crack

Figure 7.1 Slip-line field near the tip of a crack. The

solution exhibits reflection symmetry with respect to
the horizontal dashed line.

coordinates with origin at the crack tip. Thus,

t = er (φ) and s = eφ (φ), (7.99)

where er and eφ are defined in Problem 7.5. Thus, in region B we have

θ = φ, p = −k(1 + 3π/2 − 2φ); π/4 ≤ φ ≤ 3π/4. (7.100)

Evidently the stress field in the vicinity of the crack tip is bounded. Note that, as no length
scale is specified in this problem, the extent of the plastic regions cannot be determined.

Problem 7.13. Use (7.52) to write the stress in the polar form T = Trr er ⊗er +Tφφ eφ ⊗
eφ + Trφ (er ⊗ eφ + eφ ⊗ er ) + Tzz k ⊗ k. Show that Trr = Tφφ = Tzz = −p and
Trφ = k in region B.

Problem 7.14. If we reflect this solution through the y-axis passing through the right-
most point of the constant state labelled C, we obtain a solution for a body
containing two cracks lying on the x-axis with crack tips equidistant from the y-axis
and with the slip-line fields meeting at the origin. Assuming the rigidly deforming
parts of the body in the region of positive (resp., negative) y to move vertically with
velocity V (resp., −V), use the Geiringer equations (7.90) to show that a tangential
velocity discontinuity occurs on the curves separating the rigid and plastic portions
of the body.

7.5.5 Axisymmetric state exterior to a traction-free circular hole

An interesting example of a state that is neither constant nor simple is furnished by the
practically important problem of yielded material occupying the region exterior to a
traction-free hole. Suppose the hole is circular, with radius a. The exterior unit normal
140 Isotropy

to the boundary is n = −er (φ), and, with the aid of Problem 7.5, the traction condition
reduces to

Trr = Trφ = 0 at r = a. (7.101)

Assuming Tφφ|r=a to be positive, the ordered principal stresses at r = a are {T1 , T2 } =

{Tφφ , 0}, and therefore, with reference to Problem 7.7, we have Tφφ|r=a = 2k, provided,
of course, that the material is in a state of yield at the boundary. Alternatively, Tφφ|r=a =
−2k if Tφφ|r=a is negative. We consider only the first case, i.e.,

Tφφ − Trr = 2k and Trφ = 0 at r = a. (7.102)

To obtain a solution, we suppose the first of these conditions to hold in the exterior
region r > a and assume that Trr , Tφφ , and Trφ depend only on r in this region. Because
of the simple structure of the assumed solution it is convenient to bypass the Prandtl–
Hencky equations and instead appeal directly to the equation of equilibrium, divT = 0,
with divT as given in Problem 7.5. Thus,

Tr′φ + 2r Trφ = 0 and Trr′ + 1r (Trr − Tφφ ) = 0; r > a, (7.103)

where (·)′ = d(·)/dr. The first equation has the solution Trφ = C/r2 , where C is a
constant which vanishes by virtue of the boundary condition. Thus, Trφ (r) vanishes for
r ≥ a. Invoking our assumption that (7.102)1 holds for r > a, we reduce the second
equilibrium equation to T′rr = 2k/r, which, together with the boundary condition, gives

Trr (r) = 2k ln(r/a) and Tφφ (r) = 2k[1 + ln(r/a)]; r ≥ a. (7.104)

This solution meets all conditions stipulated in the problem and furnishes a post facto
verification of our assumptions. Observe that Trr > 0 for all r>a. Thus, the material in
the region outside a circle of radius r exerts a tensile traction on the material occupying
the region [a, r). The alternative case mentioned earlier corresponds to a compressive
traction.
To find a velocity field consistent with this solution, we use p = − 12 (Trr + Tφφ ) (Why
is this true?) to obtain the pressure field

p = −k[1 + 2 ln(r/a)]. (7.105)

We combine this with our solution, and, making use of (7.52)1 , substitute into (7.25) to
conclude that the straining tensor is

D = λτ = λk(eφ ⊗ eφ − er ⊗ er ). (7.106)
 Plane strain of rigid-perfectly plastic materials: Slip-line theory 141

We show that this is compatible with a purely radial velocity field of the form
v = u(r)er (φ). Thus,

Ldy = dv = u′ (r)er dr + ue′r (φ)dφ

= u′ er dr + ueφ dφ = (u′ er ⊗ er + ur eφ ⊗ eφ )dy, (7.107)

where we have used dr = er · dy and rdφ = eφ · dy, these following from the formula
for y given in Problem 7.5. The velocity gradient is thus given by the expression in
parentheses. Because this is symmetric, it is also the straining tensor, i.e.,

D = u′ er ⊗ er + ur eφ ⊗ eφ , (7.108)

and comparison with (7.106) furnishes u/r = λk and u′ = −λk. It follows that u satisfies
the differential equation u′ + u/r = 0, the solution to which is

u(r) = C/r, (7.109)

with C a constant - positive in accordance with the requirement λ > 0. The value of this
constant is determined by specifying a (positive) value of the velocity at a particular
value of the radius. The foregoing stress field is entirely independent of this value, this
circumstance being an artifact of the perfect plasticity property k = const.
To interpret this solution in term of slip-line theory, we recall that T1 = Tφφ and
T2 = Trr ; then, u1 = eφ and u2 = −er . Equations (7.51) and (7.54) imply that

θ = φ + π/4, (7.110)

and the equation of an α-line, Eq. (7.88)1 , is

sin φ+cos φ
( dy
dx )|β = tan(φ + π/4) = cos φ−sin φ = x+y
x−y . (7.111)

dx )|β = −βx /βy .

Recalling that β is constant on such a line, i.e., βx dx + βy dy = 0, we have ( dy
Thus, βx = βy ( yx+y
−x ), and the α-lines are such that

βy
y−x [(x + y)dx + (y − x)dy] = 0. (7.112)

The possibility that βy vanishes must be rejected as it implies that α-lines are the
vertical lines on which x assumes constant values; this is inconsistent with (7.110). Thus,
the α-lines satisfy the differential equation

(x + y)dx + (y − x)dy = 0, (7.113)

which, however, is not exact; that is, there is no function β such that βx = x + y and βy =
y − x. We render it such by introducing a non-zero integrating factor F(x, y), replacing
(7.113) with
142 Isotropy

F(x + y)dx + F(y − x)dy = 0, (7.114)

and redefining β such that

βx = F(x + y) and βy = F(y − x). (7.115)

This may be done without loss of generality; for (7.114) then implies that dβ = 0 on
α-lines, which is all that the definition of β requires. The requirement βxy = βyx then
furnishes the partial differential equation

(x − y)Fx + (x + y)Fy = −2F, (7.116)

the solution to which proceeds by the method of characteristics, whereby

{x(t), y(t), F(t)} is sought, with
dF dy
dt = Fx dx
dt + Fy dt , (7.117)

such that
dF
dt = −2F, dx
dt =x−y and dy
dt = x + y. (7.118)

The first of these has the solution F = F0 e−2t , with F 0 a constant, and the second pair
combine to give dtd (x2 + y2 ) = 2(x2 + y2 ), yielding 2t = ln r2 + const. These furnish the
integrating factor F = C/r2 , where C is a constant. We observe that this constant cancels
out in (7.116), and so it can be specified arbitrarily while fulfilling all restrictions that
have been stipulated for the function β. We choose it such that
y−x
βx = − xx+y
2 +y2 and βy = x2 +y2 . (7.119)

To find β we use rx = x/r, which implies that x/r2 = (ln r)x ; similarly, y/r2 = (ln r)y .
We also use tan φ = y/x to obtain (tan φ)x = −y/x2 and (tan φ)y = 1/x; i.e., φx sec2 φ =
−y/x2 and φy sec2 φ = 1/x. These give φx = −y/r2 and φy = x/r2 , reducing (7.119)1 to
βx = φx − (ln r)x . Then, β = φ − ln r + f(y), where, to satisfy the second of Eqs. (7.119),
it is necessary that f′ (y) = 0. The value of the constant f may be specified by stipulating
that the α-line in question intersects the circle r = a at the azimuthal angle φ0 . Because β
is constant on this curve we may take β = φ0 , finally reaching

φ = β + ln(r/a), (7.120)

in which β = φ0 on the considered α-line. The entire family of α-lines is generated by

varying the constant β. Proceeding in the same way we find the equations of the β-lines
to be

φ = α − ln(r/a), (7.121)
 Anti-plane shear 143

in which α is the azimuthal angle at which the curve intersects the circle r = a. Thus, the
slip lines are orthogonal logarithmic spirals. Experimental observations of these spirals
are discussed in the books by Nadai and Kachanov.

Problem 7.15. Use the stress field to show that ±p/2k + θ are constants on the
appropriate logarithmic spirals that define the slip lines.

Problem 7.16. Generalize this problem to accommodate a shear traction τeφ applied at
the hole boundary (r = a), with |τ| <k. Assuming an axisymmetric stress field, with
Trr , Tφφ , and Trφ depending only on r, obtain the equilibrium stress distribution
assuming the material to be completely yielded. Can you find a velocity field that
is consistent with this solution?

Problem 7.17. Investigate the equilibrium state of stress under the assumptions that
the stress Trφ depends only on φ and the material is in a state of yield.

7.6 Anti-plane shear

This is a much simpler class of problems, exemplified, in the axisymmetric case, by
Problem 7.6. More generally, anti-plane shear is associated with velocity fields of the
form

v = w(y1 , y2 )k, (7.122)

with gradient

L = k ⊗ gradw, where gradw = w,α i α . (7.123)

The straining tensor is

D = 12 (k ⊗ gradw + gradw ⊗ k), (7.124)

and (7.25) furnishes the deviatoric Cauchy stress

τ = σ ⊗ k + k ⊗ σ, with λσ = 21 gradw. (7.125)

√
The yield condition |τ| = 2k is equivalent to

|σ| = k. (7.126)

Then, λ = 1
2k |gradw| and

−1
σ = kt, where t = |gradw| gradw, (7.127)
144 Isotropy

which is meaningful if the motion is non-rigid, i.e., if gradw ̸= 0. In this case we introduce
a two-dimensional scalar field θ, as in (7.54)1 , such that

t = cos θi1 + sin θi2 , (7.128)

and use (7.26) to reduce the stress to

T = −pi + k(t ⊗ k + k ⊗ t). (7.129)

Assuming equilibrium without body force, and noting that (gradt)k = 0, we arrive at
the equation to be solved:

grad(p/k) = (divt)k. (7.130)

Problem 7.18. Fill in the steps leading to (7.130).

This yields p,α = 0 and p = P(y3 ), with P ′ (y3 )/k = divt. Then, P ′′ (y3 ) vanishes and
(7.130) reduces to

C = kdivt, (7.131)

where C is the constant axial pressure gradient in the y3 -direction. This may be cast as
an equation for θ. Thus,

(gradt)dy = dt = sdθ = s(gradθ · dy) = (s ⊗ gradθ)dy, where s = − sin θi1 + cos θi2 ,
(7.132)
implying that gradt = s ⊗ gradθ and hence that divt = s · gradθ.
Adopting the terminology of plane-strain slip-line theory, we conclude, in the case of
a vanishing axial pressure gradient, that dθ = 0 on β-lines, and hence that these lines are
straight. Thus,

θ = η(α) (7.133)

for some function η. Moreover, from (7.127)2 and (7.128) we have that s · gradw = 0.
It follows that

w = ω(α), (7.134)

for some function ω, and hence that the velocity is constant on a given β-line. Naturally,
this constant varies from one such line to another.
 References 145

To illustrate this model we again consider the field exterior to a traction-free circular
hole of radius a. Using (7.129), the zero-traction condition is easily seen to require that

p=0 and er (φ) · t = 0 at r = a, (7.135)

the first of these implying that P(y3 ) = 0 and thus that p vanishes indentically in the
material region; and the second yielding t = ±eφ (φ) and s = ∓er (φ) at r = a. We
consider the first pair of alternatives for the sake of definiteness.
In fact we obtain a solution simply by extending these conditions into the region r > a,
i.e., by taking t = eφ (φ) and s = −er (φ) in the material region. This follows easily from

dt = deφ = − 1r er (eφ · dy), (7.136)

yielding gradt = − 1r er ⊗ eφ and divt = − 1r er · eφ = 0, as required. The corresponding

stress field is given simply by

T = k(eφ ⊗ k + k ⊗ eφ ), (7.137)

and from (7.128) we have θ = φ + π/2. We may therefore take α = φ and η(α) = α + π/2.
The condition s · gradw = 0 implies that w is independent of r and hence a function of
φ alone, i.e., that w = ω(φ), with

gradw = 1r ω′ (φ)eφ . (7.138)

In the absence of further conditions, this is as far as we can go in solving the problem as
stated.

Problem 7.19. Generalize this problem to accommodate a shear traction τk applied

at the hole boundary (r = a), with |τ| <k. Find an axisymmetric equilibrium stress
field in the region r > a and show that it approaches the solution for the traction-free
hole as r/a → ∞. What restrictions does the velocity field w satisfy?

This concludes our introduction to the classical theory of the rigid, perfectly plastic
response of isotropic materials. We have barely scratched the surface of this large field.
The references cited contain many further solutions to a large variety of problems, not
limited to conditions of plane or anti-plane strain. Applications to three-dimensional
problems, including extensions to account for hardening of the material, may be found
in the papers by Durban and Fleck.

References
Batchelor, G. K. (Ed) (1958). The Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 1: Mechanics
of Solids. Cambridge University Press, Cambridge, UK.
146 Isotropy

Chadwick, P. (1976). Continuum Mechanics: Concise Theory and Problems. Dover, New York.
Christensen, R. M. (2016). Theory of Materials Failure. Oxford University Press, Oxford.
Collins, I. F. (1982). Boundary value problems in plane strain plasticity. In H. G. Hopkins and
M. J. Sewell (Eds), Mechanics of Solids: The Rodney Hill 60th Anniversary Volume, pp. 135–84.
Pergamon Press, Oxford.
Cristescu, N. D. (2007). Dynamic Plasticity. World Scientific, Singapore.
Durban, D. (1979). Axially symmetric radial flow of rigid/linearly hardening materials. ASME J.
Appl. Mech. 46, 322–8.
Durban, D. (1980). Drawing of tubes. ASME J. Appl. Mech. 47, 736–740.
Durban, D., and Fleck, N. A. (1992). Singular plastic fields in steady penetration of a rigid cone.
ASME J. Appl. Mech. 59, 706–10.
Edmiston, J., Steigmann, D. J., Johnson, G. J. and Barton, N. (2013). A model for elastic-
viscoplastic deformations of crystalline solids based on material symmetry: theory and
plane-strain simulations. Int. J. Eng. Sci. 63, 10–22.
Epstein, M., and Elżanowski, M. (2007). Material Inhomogeneities and Their Evolution. Springer,
Berlin.
Geiringer, H. (1973). Ideal plasticity. In C. Truesdell (Ed.) Mechanics of Solids, Vol. III, pp. 403–
533. Springer, Berlin.
Gupta, A., and Steigmann, D. J. (2014). Plane strain problem in elastically rigid finite plasticity.
Quart. J. Mech. Appl. Math. 67, 287–310.
Gurtin, M. E. (1981). An Introduction to Continuum Mechanics. Academic Press, Orlando, FL.
Hill, R. (1950). The Mathematical Theory of Plasticity. Clarendon Press, Oxford.
Johnson, W., and Mellor, P. B. (1980). Engineering Plasticity. Van Nostrand Reinhold, London.
Johnson, W., Sowerby, R., and Haddow, J. B. (1970). Plane-Strain Slip-Line Fields: Theory and
Bibliography. Edward Arnold, London.
Kachanov, L. M. (1974). Fundamentals of the Theory of Plasticity. MIR Publishers, Moscow.
Lubliner, J. (2008). Plasticity Theory. Dover, New York.
Nadai, A. (1950). Theory of Flow and Fracture of Solids. McGraw-Hill, New York.
Neff, P. (2006). A finite-strain elastic-plastic Cosserat theory for polycrystals with grain rotations.
Int. J. Engng. Sci. 44, 574–94.
Prager, W. (1961). Introduction to the Mechanics of Continua. Ginn & Co., Boston.
Truesdell, C., and Noll, W. (1965). The nonlinear field theories of mechanics. In: S. Flügge (Ed.),
Handbuch der Physik III/1. Springer, Berlin.
Yu, H.-S. (2006). Plasticity and Geotechnics. Springer, Berlin.
8
Small-deformation theory

Most of the literature concerned with explicit solutions to problems involving simultane-
ous elastic–plastic deformations is cast in a framework that presumes small deformations
at the outset. The basic theory, due to Prandtl and Reuss, facilitates the appropriation of
results from linear elasticity theory, this being by far the most highly developed branch
of solid mechanics. To make contact with this substantial literature, it is thus incum-
bent upon us to develop the small-deformation theory in some detail. Unfortunately, we
immediately encounter the difficulty that this theory, as traditionally formulated, fails
to be frame invariant, as do traditional formulations of linear elasticity theory. This
means that a valid calculation based on the conventional formulation, performed in
a frame of reference appropriate to Tokyo, for example, will be invalid in a rotated
frame one might use in San Francisco, and that the calculation in Tokyo, if valid at
one instant in time, will cease to be valid at other times as the local frame rotates with
the Earth. In view of the well-documented success of the small-deformation theories as
predictive models, it is thus clear that there must be some conceptual detail in the con-
ventional expositions that is not quite right. The resolution of this difficulty is provided
by the shifter concept, briefly introduced in Section 2.3. Accordingly we confine our-
selves in the present chapter to the development of a frame-invariant formulation of the
Prandtl–Reuss theory.

8.1 The displacement field

Students of solid mechanics usually learn about the displacement field prior to the study
of more advanced concepts, such as the deformation gradient. It is when encountering
the latter in graduate courses that the confusion starts. Typically, the displacement field
u(x, t) is defined by

u(x, t) = y − x, with y = χ(x, t), (8.1)

in which y and x are interpreted, not as positions in point spaces, but rather as position
vectors in the associated translation, or tangent spaces. For example, y = yk ik in terms of
Cartesian coordinates {yk }. We then have y ∈ Tκt , and for consistency it is necessary to
interpret χ as a vector in Tκt . However, x ∈ Tκ and (8.1) makes no sense as it stands

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0008
148 Small-deformation theory

because it entails the subtraction of vectors belonging to distinct vector spaces. This
inconsistency is easily addressed by using the shifter, 1, to inject x into Tκt , the effect of
which is to convert x into an element of Tκt without otherwise altering it. Its image there
is

x∗ = 1x, (8.2)

and is used to redefine the displacement as

u(x, t) = χ(x, t) − x∗ . (8.3)

This definition is meaningful because the right-hand side involves the difference of
vectors belonging to the same space.
From (8.2) we have that x∗ is a linear function of x. Its gradient with respect to x is
therefore given by

∇x∗ = 1. (8.4)

Recalling the discussion in Section 2.3 to the effect that 1 is a rotation, the fact that it
is also a gradient implies that it is spatially uniform. The proof of this assertion pro-
ceeds in exactly the same way as the proof of the fact that the rotation accompanying a
superposed rigid-body deformation is necessarily uniform. See the proofs in the books
i
by Chadwick and Gurtin, for example. Using convected coordinates {ξ } to illustrate,
we then differentiate (8.2) with respect to the coordinates, obtaining

e∗i = 1ei , (8.5)

where ei = x,i and e∗i = x∗,i . Thus,

1 = 1I = 1ei ⊗ ei = e∗i ⊗ ei . (8.6)

Using ei = δij ej = e j (e∗j · e∗i ) = (ej ⊗ e∗j )e∗i = 1t e∗i , we also have

1 = e∗i ⊗ 1t e∗i = (e∗i ⊗ e∗i )1 = i1, (8.7)

and thus

1I = i1. (8.8)

From (8.6) and the metric e∗ij = e∗i · e∗j = ei · ej = eij , which follows from (8.5) and the
fact that 1 is a rotation, we also have that

1t 1 = I and 11t = i, (8.9)

both of which have been used on previous occasions, in Sections 2.3 and 7.3, respec-
tively.
 The displacement field 149

Taking the gradient of (8.3), we arrive at

F = 1 + ∇u, (8.10)

where F = ∇χ is the usual deformation gradient. Again this makes sense because each
term in this equation is a tensor of the same type; that is, each is a two-point tensor that
maps T κ to Tκt . The traditional expression, F = I + ∇u, is not tensorially meaningful
because I maps T κ to itself and is therefore not of the same type as F.
The point is reinforced by the example of the superposed rigid-body deformation
described in Eq. (2.25). The associated deformation gradient is QF in which the rotation
Q is spatially uniform. To ensure that (8.10) continues to apply after the rotation we
should require that 1 be transformed to Q1 and that ∇u be transformed to Q(∇u). From
(8.5), the first of these is tantamount to the statement that e∗i is transformed to Qe∗i , and
hence that the vector space Tκt is itself rotated by Q, whereas the second implies that
du = (∇u)dx is transformed to Q(∇u)dx = Qdu = d(Qu), yielding the transformation

u → Qu + c, (8.11)

where the vector c is a function of t. In the same way, from (8.4) we find that x∗ trans-
forms to Qx∗ + e, where the vector e is a function of t. This is consistent with the
conventional notion that x remains invariant under superposed rigid-body motions pro-
vided that e = 0. Taken together with (8.3), these in turn imply that χ(x, t) transforms
to Q(t)χ(x, t) + c(t), which is just (2.25).
To interpret (8.11) visually, imagine a displacement vector that is stationary with
respect to a fixed frame of reference attached to the ground. Suppose we observe this
vector from the vantage point of a passing airplane. From our viewpoint the entire vector
undergoes a time-dependent rotation and translation in accordance with (8.11). How-
ever, in the traditional interpretation, based on (8.1), we start with (2.25) and, assuming
x to remain invariant, conclude, contrary to the airborne observer, that u transforms
to Qy − x + c, and hence that ∇u transforms, rather bizarrely, to Q(I + ∇u) − I. This
tensorially meaningless statement lies at the heart of the widespread and persistent mis-
conception that linear elasticity theory is deficient in that it fails to be frame invariant. In
contrast, no such difficulty arises when the displacement and its gradient are properly
interpreted in terms of (8.3) and (8.10).
This thought experiment suggests that we’ve been too cavalier about the relation-
ship between frame invariance and invariance under superposed rotations. The former
pertains to invariance with respect to a change of observer, whereas the latter pertains
to a single observer. In Murdoch’s paper it is shown that invariance under superposed
rigid-body motions follows from frame invariance. Accordingly the rotation Q is more
properly interpreted as a rotation from one frame of reference to another.
It proves convenient to introduce a purely referential displacement gradient, Du,
defined by

∇u = 1(Du), Du = 1t (∇u) = ∇(1t u), (8.12)

150 Small-deformation theory

and to write (8.10) in the form

F = 1(I + Du). (8.13)

Assuming that the various observers adopt the same reference configuration (so that the
operator ∇ remains invariant), from the foregoing discussion we then have that Du is
invariant under a change of frame, this being a trivial consequence of the fact, which fol-
lows from (8.11), that 1t u → 1t u+d(t), where d(t) = (Q1)t c. Combining this with (8.9)1
and the definition (1.20) of the Lagrange strain E, we derive the strain-displacement
relation

E = ϵ + 12 (Du)t Du, where ϵ = 12 {Du + (Du)t }. (8.14)

Note that ϵ is invariant under a change of frame.

Small-deformation theories are based on the premise that |Du| ≪ 1 at all material
points. In these circumstances the leading-order approximation E ≃ ϵ furnishes a lin-
ear strain–displacement relation that is again insensitive to changes of frame; here and
henceforth, we use the symbol ≃ to identify expressions that are valid to leading order. In
contrast, in conventional treatments the strain–displacement relation is given by (8.14)
with Du replaced by ∇u. In this case, although E is invariant under a change of frame, ϵ is
not, and the linear approximation E ≃ ϵ yields a strain measure that fails to be invariant.
This is the reason why linear elasticity theory, as traditionally formulated, is not frame
invariant. Of course no such difficulty arises if proper account is taken of the true nature
of the displacement gradient as defined by (8.10).
Naturally the foregoing discussion pertains entirely to kinematics and therefore
applies whether or not the material is elastic. However, see the paper by Steigmann for
a fuller discussion of this issue as it relates to linear elasticity theory, and the book by
Marsden and Hughes for a more general exposition of the shifter concept.

Problem 8.1. Using the convected-coordinate formalism we have ∇u = u,i ⊗ ei in

which u ∈ Tκt in accordance with (8.3). We may thus write u = u j e ∗j . Then,
Du = 1t u,i ⊗ ei = (1t u),i ⊗ ei , where 1t u = uj ej . Show that the components ui|j ,
k
where (·)|i is the covariant derivative with respect to the referential connection Γ̄ij ,
are invariant under a change of frame. Derive the representation (4.12) for the
strain, in which

Eij = ϵij + 21 uk|i uk|j , with ϵij = 21 (ui|j + uj|i ),

∗k k
where ui = eij u j in which eij = ei · ej is the referential metric. Show that Γ̄ij = Γ̄ij .

Problem 8.2. Use (5.2) and (5.3) with (1.20) to obtain

E = Gt ÊG + P̂,
 Approximations for small displacement gradients 151

where Ê is the elastic strain, given by (5.46), and P̂, the plastic strain (not to be
confused with the Piola stress), is defined by

P̂ = 12 (Gt G − I).

Show that Ê = Êij mi ⊗ mj and P̂ = P̂ij ei ⊗ ej , where Êij = 12 (gij − mij ) and
P̂ij = 12 (mij − eij ), and thus establish the relation Eij = Êij + P̂ij , valid for
covariant components only.

8.2 Approximations for small displacement

gradients
We have already mentioned the leading-order approximation E ≃ ϵ, valid if |Du| ≪ 1,
where ϵ is defined by (8.14)2 . This implies that |E| ≪ 1, but the converse is not valid.
For, if the strain is small, then the polar decomposition of the deformation gradient
implies that F is approximated by a rotation that may differ markedly from 1, and this in
turn yields a finite displacement gradient. This situation is encountered in thin flexible
structures undergoing large deformations while the strains remain small.

Problem 8.3. Use (8.13) and (8.14)2 together with the polar factorization F = RU, in
which R is a (two-point) rotation tensor and U is the symmetric, positive definite
right-stretch tensor, to show that

U = I + ϵ + O(ε2 ), U−1 = I − ϵ + O(ε2 ) and R = 1{I + ω + O(ε2 )},

 
where ε = |Du| , O(ε2 ) = O(ε2 ) and

ω = 12 {Du − (Du)t }.

Accordingly, U − I ≃ ϵ and R − 1 ≃ 1ω if ε ≪ 1. Note that the skew tensor ω,

which characterizes the linear-order rotation due to a displacement with a small
gradient, is invariant under superposed rigid-body motions.

Problem 8.4. Use (8.13) and (8.14) with J2F = det(Ft F) to obtain

J2F = det(A − λI)|λ=−1 , where A = 2ϵ + (Du)t Du.

Use the cubic characteristic equation for the eigenvalues of A to show that

JF = 1 + trϵ + O(ε2 ),

and hence that JF ≃ 1 if ε ≪ 1.

152 Small-deformation theory

Problem 8.5. Show that

F∗ − 1 ≃ 1{(trϵ)I − (Du)t }

if ε ≪ 1, where F∗ , given by (1.36), is the cofactor of the deformation gradient.

Thus, F∗ ≃ 1.

Problem 8.6. Use (5.3) to conclude, again if ε ≪ 1, that H ≃ 1K. Note that this
approximation preserves the transformation formulas (6.88) for changes of frame.

Having disposed of the kinematical approximations, we proceed to estimate the rela-

tions among the various stress measures and to obtain the approximate equation of
motion. For example, if ε ≪ 1, which we assume henceforth, then (1.42) and (8.13)
imply that

P ≃ 1S, (8.15)

where P and S, respectively, are the Piola and Piola–Kirchhoff stresses. It then follows
from (1.38) and the result of Problem 8.5 that the relation between the Cauchy and
Piola–Kirchhoff stresses is given, to leading order, by

T ≃ 1S1t . (8.16)

According to (1.27), (1.35), and (1.48), these imply that the Cauchy and Piola tractions,
t and p, respectively, coincide at leading order, i.e.,

t ≃ p ≃ 1(Sν), (8.17)

where ν is the exterior unit normal to the boundary ∂π of a material region π ⊂ κ on

which the traction is acting.
To derive the leading-order approximation to the equation of motion (1.40), we
assume, in addition to (8.15), that

DivP ≃ Div(1S). (8.18)

This requires, beyond the small-displacement-gradient assumption ε ≪ 1 already made,

that the norm of the second gradient of the displacement, D(Du), be small, of the same
order as ε.

Problem 8.7. Prove this claim.

These assumptions, however, can only be verified, if true, after a solution to the the-
ory, yet to be formulated, has been obtained. This is exactly the same situation one
encounters when using linear elasticity theory. In view of the success of that theory, this
 The frame-invariant Prandtl–Reuss theory 153

issue need hardly deter us from proceeding with the derivation of a parallel formulation
of elasto-plasticity theory.
Using convected coordinates, together with the referential counterpart of the formula
(3.80) for the divergence, we have DivP ≃ Div(1S) = {(1S),i }ei = 1(S,i )ei = 1(DivS),
and the relevant equation of motion, approximating (1.40) at leading order, is

DivS + ρκ 1t b ≃ ρκ 1t v̇. (8.19)

Problem 8.8. A fixed frame of reference used to describe the motion of a body may
be regarded as vector space spanned, at a particular point, by {e∗i } in which the e∗i
are independent of time. This implies that 1 is also independent of time. Use this
with v = χ̇ to show that 1t v̇ = üi ei . In effect, we then regard the vector space Tκt as
being fixed while κt itself evolves in the associated point space.

Problem 8.9. It follows easily, from (1.22), (1.32), (1.38), and (1.40), that JF divT =
Div(TF∗ ). Use the divergence and localization theorems with the Piola–Nanson
formula (1.35), or direct calculation, to establish the identity

JF divw = Div{(F∗ )t w},

where w is a spatial vector field. Show, under our assumptions, that

divT ≃ 1Div(1t T1) and divw ≃ Div(1t w).

8.3 The frame-invariant Prandtl–Reuss theory

Entirely independently of any assumption about the displacement gradient, in Chapter
6 we formulated the theory of elastic–plastic response on the premise that the norm of
the elastic strain Ê is small. Applying the triangle inequality to the result of Problem 8.2,
we have

|E| ≤ |Gt ÊG| + |P̂|. (8.20)

Thus, the restriction to small |Ê| does not imply that |E| is small. Indeed, this obser-
vation underpins the finite-deformation theory of rigid-plastic materials, developed in
the previous chapter, in which |Ê| = 0. Moreover, the assumption of a small displace-
ment gradient, while implying that |E| is small, imposes no restriction on |Ê| or |P̂|.
Accordingly, one may contemplate a model for small displacement gradients having no
restrictions on the elastic and plastic deformations beyond those implied by the result of
Problem 8.6. Here, however, we proceed, as in the literature, to assume that the norms
of both the elastic and plastic strains are small, on the order of that of the displacement
gradient. The assumption about the plastic strain implies, as in Problem 8.3, that the
right-stretch factor UG in the polar factorization G = RG UG , in which RG is a rotation,
154 Small-deformation theory

is approximated, to linear order in P̂, by I + P̂. We further suppose that RG is approxi-

mated by I + Ω̂, where |Ω̂| ≪ 1. It then follows that Ω̂ is skew to linear order, and that
G is approximated, again to linear order, by I + P̂ + Ω̂. Taking this approximation to be
exact amounts to assuming that Ġ = (P̂)· + (Ω̂)· . We then have that ĠG−1 is approxi-
mated by (P̂)· + (Ω̂)· , and the flow rule (6.105) is approximated to consistent order by

(P̂)· = λFŜ + Ω, with Ω = −(Ω̂)· . (8.21)

These assumptions are also seen to imply that Gt ÊG is approximated by Ê and hence,
with reference to Problem 8.2, that E is approximated by Ê + P̂. With the added
assumption of small displacement gradients, we then have

ϵ ≃ Ê + P̂, with Ê = L[Ŝ], (8.22)

where L is the compliance tensor.

Problem 8.10. With reference to Problem 6.3 and the Appendix to Section 6.1, show
that, in the case of isotropy,

L[Ŝ] = 91κ (trŜ)I + 21μ DevŜ,

where μ and κ(= λ + 23 μ) are the (positive) shear and bulk moduli, respectively.

Problem 8.11. Use (6.9) and (6.11) to show, granted our assumptions, that Ŝ ≃ S.

Accordingly, assuming, on the basis of (8.22), that ϵ̇ ≃ (Ê)· + (P̂)· , and that (Ŝ)· ≃ Ṡ,
the relevant system is given by (8.17), (8.19) together with
 
L[Ṡ] + λFS + Ω; λ ≥ 0, if F(S) = 0,
Sym(Du)· = . (8.23)
L[Ṡ]; if F(S) < 0

This is the frame-invariant extension of the classical Prandtl–Reuss equations, originally

stated for isotropic materials, to accommodate anisotropic response. Recalling that the
plastic spin may be suppressed in the case of isotropy, the system reduces in this case to
( √ )
·
1
9κ (trṠ)I + 21μ DevṠ + λDevS; λ ≥ 0, if |DevS| = 2k,
Sym(Du) = √ . (8.24)
1 1
9κ (trṠ)I + 2μ DevṠ; if |DevS| < 2k

The literature on the Prandtl–Reuss theory for isotropic materials is vast. Many
explicit solutions are known, a large number of which are worked out in the books
by Hill, Kachanov, and Lubliner, for example, cited in Chapter 7. The book by
 References 155

Unger documents a number of explicit solutions having immediate relevance to frac-

ture mechanics, and a modern account of the basic theory may be found in Chapter 8
of the book by Bigoni.

Problem 8.12. Use convected coordinates, as in Problem 8.1, to show that

(Du)· = u̇i|j ei ⊗ e j .

References
Bigoni, D. (2012). Nonlinear Solid Mechanics: Bifurcation Theory and Material Instability.
Cambridge University Press, Cambridge, UK.
Chadwick, P. (1976). Continuum Mechanics: Concise Theory and Problems. Dover, New York.
Gurtin, M. E. (1981). An Introduction to Continuum Mechanics. Academic Press, Orlando, Fl.
Marsden, J. E., and Hughes, T. J. R. (1994). Mathematical Foundations of Elasticity. Dover,
New York.
Murdoch. A. I. (2003). Objectivity in classical continuum physics: A rationale for discarding the
“principle of invariance under superposed rigid-body motions” in favor of purely objective
considerations. Continuum Mech. Thermodyn. 15, 309–20.
Steigmann, D. J. (2007). On the frame invariance of linear elasticity theory. ZAMP 58, 121–36.
Unger, D. J. (1995). Analytical Fracture Mechanics. Dover, New York.
9
Strain hardening, rate sensitivity, and
gradient plasticity

In this final chapter we extend the theory developed thus far to accommodate strain hard-
ening and also extrapolate Bingham’s model to more general models of viscoplasticity.
The modeling of strain hardening, in particular, is arguably the main open problem in
the phenomenological theory of plasticity. We survey some classical models of harden-
ing and, following this, discuss more recent efforts aimed at modeling observations on
the correlation of hardening with sample size, these indicating that material strength is
enhanced in sufficiently small metallic bodies relative to larger bodies consisting of the
same material. Such scale effects are well documented in the papers by Hutchinson,
Stölken and Evans, and Fleck et al. and have given rise to a substantial literature on
gradient plasticity.

9.1 “Isotropic” hardening

The simplest model of strain hardening is so-called “isotropic” hardening in which the
yield stress - the parameter k in von Mises’ yield function, for example—is not a constant
but instead varies with the state of the material. We say “so-called” because this model
is not confined to isotropic material symmetry. Rather than describing a property of
isotropic materials, the term instead refers to an overall expansion of the yield surface
F = 0 that does not alter its shape.
The idea is most easily introduced via von Mises’ function (7.10), generalized to
 2
 
F(Ŝ, σ) = 1
2 DevŜ − k2 (σ), (9.1)

where k(σ) is the evolving yield stress in shear and σ is a new scalar variable that evolves
with plastic deformation. The models most often considered are associated with one of
two evolution equations for σ. These are
 
 
σ̇ = ĠG−1  or σ̇ = Ŝ · ĠG−1 , (9.2)

A Course on Plasticity Theory. David J. Steigmann, Oxford University Press. © David J. Steigmann (2022).
DOI: 10.1093/oso/9780192883155.003.0009
 “Isotropic” hardening 157

together with the initial condition σ(x, t0 ) = σ0 (x). Both models ensure that σ̇ > 0
whenever Ġ ̸= 0.

Problem 9.1. Verify this claim for (9.2)2 . Hint: Use (6.74) and (6.76) together with
(6.68) to confirm that Ŝ · ĠG−1 > 0, to leading order in the small elastic strain,
whenever Ġ ̸= 0.

Problem 9.2. Show that both models are invariant under G → Rt G and Ŝ → Rt ŜR for
any fixed R ∈ gκi (p) ⊂ Orth+ . Thus, they are meaningful for any type of material
symmetry.

Note that both models are also invariant under the time rescaling t → ct for any
positive constant c. They are therefore rate-independent. Further, in both cases σ̇ is inde-
pendent of the reference configuration κ and invariant under superposed rigid-body
motions and time translations t → t + a, for any constant a. Thus, both models are
intrinsic to κi (p) and also frame-invariant. The first alternative is the one most often
used in practice. Accordingly, we will confine attention to (9.2)1 henceforth. See Section
2.3 of Hill’s book for a fuller discussion of these models, including techniques for the
experimental determination of the function k(σ).
The variable σ(x, t) reflects the history of plastic flow, and the function k(σ) reflects
the influence of this history on the current value of the yield stress. Empirical obser-
vations typically indicate that k′ (σ) > 0, such that k, like σ, increases with plastic
deformation. Hence, the term “strain hardening”. Further, this generalization has no
effect on the discussion leading to inequality (6.97), apart from the adjustment that F
now depends on the current value of σ.

Problem 9.3. Prove this.

Accordingly, the flow rule (6.103) remains in effect with this adjustment, with FŜ now
interpreted as a partial derivative, i.e., as the derivative at fixed σ. This, of course, special-
izes to (7.14) in the case when F is given by (9.1). Recall that in this flow rule the scalar
multiplier field λ(x, t) must be non-negative in accordance with the maximum dissipa-
tion inequality (6.97). In the case of perfect plasticity this multiplier is to be regarded as
an additional field to be determined in the course of solving the problem at hand. In the
case of isotropic hardening, however, we can relate λ to the notions of loading, neutral
loading, and unloading as follows.
Thus, suppose there exists an interval of time during which plastic deformation
evolves, this being necessary for the existence of a continuous derivative Ġ ̸= 0. Then
F(Ŝ, σ) vanishes identically in this interval at the material point in question, and hence
so too its material derivative:

FŜ · (Ŝ)· + Fσ σ̇ = 0. (9.3)

158 Strain hardening, rate sensitivity, and gradient plasticity

This is called the consistency condition for continuing plastic flow. In the case of (9.1) it
reduces to

DevŜ · (Ŝ)· − 2kk′ (σ)σ̇ = 0, (9.4)

which specializes, with (9.2)1 , to

   
   
0 = DevŜ · (DevŜ)· − 2kk′ (σ) ĠG−1  = DevŜ · (DevŜ)· − 2λkk′ (σ) DevŜ , (9.5)

where we have used λ ≥ 0. Solving for λ, we obtain

λ= 1
2kk′ (σ)|DevŜ|
DevŜ · (DevŜ)· , (9.6)

which is meaningful, assuming k′ (σ) > 0, provided that

  √
 
DevŜ = 2k and DevŜ · (DevŜ)· ≥ 0, (9.7)

the latter corresponding, in the case of strict inequality, to plastic loading, and, in the
case of equality, to neutral loading. The remaining possibility, DevŜ · (DevŜ)· < 0, is
referred to as unloading. In this case we stipulate that λ = 0 and hence that Ġ vanishes.
Our terminology refers to a state of yield existing at the onset of the particular loading
or unloading condition in question. Thus in the √ case of loading we have that (DevŜ)·
makes an acute angle with DevŜ, with |DevŜ| = 2k. This √ implies that λ > 0, Ġ ̸= 0
and hence σ̇ > 0. With√k′ (σ) > 0 it follows that the radius 2k(σ) of the hypercylinder
defined by |DevŜ| = 2k(σ) increases monotonically with plastic flow. In the case of
neutral loading or unloading we have λ = 0, Ġ = 0, σ̇ = 0, and k(σ) remains fixed. This
model thus furnishes a faithful representation of the basic phenomenology described in
Section 1.1.
The extension to crystalline symmetry is straightforward. Substituting the flow rule
(6.105), together with (9.2) 1 , into the consistency condition (9.3), we obtain

 
0 = FŜ · (Ŝ)· + λFσ FŜ + Ω̄ , (9.8)

where Ω̄ is the skew tensor defined by λΩ̄ = Ω. This yields Ω = 0 when Ġ = 0, a con-
dition that requires λ = 0 by virtue of the orthogonal decomposition (6.105). Moreover,
 Rate sensitivity: Viscoplasticity 159

with reference to (6.77) and (6.112), it furnishes a unique Ω̄ when Ġ ̸= 0; for we then
require that λ > 0. In other words, we lose no generality if we replace (6.105) by

ĠG−1 = λ(FŜ + Ω̄); λ ≥ 0. (9.9)

Further, because the linear space of tensors is the direct sum of the orthogonal subspaces
of symmetric and skew tensors, we conclude, from Pythagoras’ theorem, that
  q   
F + Ω̄ = F 2 + Ω̄2 . (9.10)
Ŝ Ŝ

Assuming Fσ < 0, as in the extended von Mises yield function, we then have

λ= 1
FŜ · (Ŝ)· , (9.11)
|Fσ ||FŜ +Ω̄|

provided that F(Ŝ, σ) = 0 and FŜ · (Ŝ)· ≥ 0, the latter encompassing plastic loading
and neutral loading, whereas λ = 0 and Ġ = 0 in the case FŜ · (Ŝ)· < 0 associated with
unloading.
Isotropic hardening does not account for the Bauschinger effect, described briefly in
Section 6.5. Some tentative ideas concerning the modeling of this effect are discussed in
Section 9.5, in connection with the theory of gradient plasticity.

9.2 Rate sensitivity: Viscoplasticity

Another of our aims is to extend Bingham’s model of viscoplasticity to accommodate
elastic deformations and crystalline symmetry. The important paper by Perzyna, cited
in this chapter’s References section, should be consulted for further information on
viscoplastic response.
−1
We seek a flow rule in which ĠG is given explicitly. In the course of adapting
Bingham’s model (7.29), our first order of business is thus to invert it to obtain an explicit
expression for D. To this end we write (7.30) in the form

|D| = 21η G(τ) |τ| , where G(τ) = 1 − k√

|τ|/ 2
, (9.12)

this presuming the value of the yield function F̃(τ), defined by (7.21), to be non-negative
and hence that G(τ) ≥ 0. Substituting into (7.29) we then obtain
|τ|
2ηD = ( G(τ)G(τ) √ )τ;
|τ|+ 2k
G(τ) ≥ 0, (9.13)
√
which may be simplified on noting that G(τ) |τ| + 2k = |τ| ; thus, recalling that D
vanishes if F̃(τ) < 0, we have
 
G(τ)τ; G(τ) ≥ 0,
2ηD = , where G(τ) = 1 − √ k 2 . (9.14)
0; G(τ) < 0 F̃(τ)+k
160 Strain hardening, rate sensitivity, and gradient plasticity

This may be generalized immediately to accommodate elastic–plastic deformations

of isotropic materials thus,
( )
−1 (1 − √ k
)DevŜ; F̃(DevŜ) ≥ 0,
2ηĠG = F̃(DevŜ)+k2 , (9.15)
0; F̃(DevŜ) < 0

where F̃(DevŜ) is given by (7.10). See the paper by Atai and Steigmann for an
application of this model to the simulation of the transient dynamics of thin metallic
sheets.
Extensions of Bingham’s model to accommodate crystalline symmetry are much more
varied. For one thing, there can be more than one viscosity. Further, most of the models
in use are based on the slip-system decomposition (6.20). See the volume edited by
Teodosiu, for example. A simple alternative model, merely one among many consistent
with the framework discussed in this book, is
 
−1
1
2η Φ(F)FŜ + Ω; F ≥ 0,
ĠG = , (9.16)
0; F<0

where η is again a positive viscosity, F is the yield function, and Φ is a positive definite
function of F for F ≥ 0, vanishing at F = 0. To ensure continuity at F = 0 we write this
in the form
 
−1 Φ(F)(FŜ + Ω̄); F ≥ 0,
2ηĠG = , (9.17)
0; F<0

where Ω = (Φ/2η)Ω̄. This model is easily seen to be dissipative, as required, provided

that FŜ satisfies inequality (6.113).
The essential feature of all such models is that Ġ is non-zero when F ≥ 0. In con-
trast to rate-independent hardening, there is no consistency condition. Nor is there
the requirement—a restriction that must be respected in rate-independent plasticity—
to ensure that F ≤ 0 in the course of integrating the flow rule. From the viewpoint
of numerical analysis, these are major advantages of viscoplasticity theory over the
rate-independent theory.

9.3 Scale effects

A natural way to model length-scale effects in the plastic deformation of metals is to
include the gradient ∇G of the plastic deformation as an argument of the relevant consti-
tutive functions. This has dimensions of reciprocal length, meaning that a material length
scale must be present in the non-dimensionalized constitutive functions. Because the
constitutive functions pertain to material points, this scale is necessarily representative
of the fine structure of the material. Accordingly, ∇G exerts influence on the response
 Scale effects 161

when the plastic deformation varies significantly over a length scale comparable to the
local material scale.
In our work we have discussed only two constitutive functions, the strain-energy func-
tion and the yield function, both of which are, of course, scalar-valued. Recall that these
have been defined intrinsically, to reflect the material properties existing in κi (p), and are
therefore entirely independent of a reference configuration. This codifies the view that
the choice of reference is largely arbitrary, more a matter of convenience than anything
else, and so should not influence physical properties as encoded in the undistorted state
of the material. While we naturally intend to retain this feature of constitutive functions
in the extension to scale-dependent response, we must now contend with the fact that
both G and ∇G, its gradient with respect to x ∈ κ, are very much dependent on the
choice of reference configuration.
We will prove the remarkable fact that any scalar-valued constitutive function,
F(G, ∇G) say, that is insensitive to the choice of reference configuration, is necessarily
a function of the true dislocation density α defined by (5.20), i.e.,

F(G, ∇G) = f(α), (9.18)

for some function f. Sufficiency follows immediately from the fact, proved in Section 5.1,
that α is invariant under a change of reference configuration κ → κ̄ defined by x̄ = λ(x)
for any smooth function λ, where x is the position of a material point p in κ, and x̄ its
position in κ̄ (in Chapter 5 we used the notation x1 = x and x2 = x̄). The proof of
necessity, which is more difficult, is facilitated by using convected coordinates.
Proceeding, we obtain a useful expression for the third-order tensor ∇G on noting
that
i
(∇G)dx = dG = G,i dξ = G ,i (ei · dx) = (G,i ⊗ ei )dx, (9.19)
i i
where we have abused the notation, albeit harmlessly, by writing G(ξ , t) = G(x(ξ ), t),
and, of course, the commas are partial derivatives with respect to the coordinates. Thus,

∇G = G,i ⊗ ei , where G = mj ⊗ e j , (9.20)

the second of these being simply a restatement of (5.24). We use this with (5.29) and
j
(3.66)—the latter in the form e,ij = −Γj(κ)ki ek (where Γj(κ)ki are the same as the Γ̄ki of
Chapter 4)—to arrive at
k
∇G = (mj,i ⊗ e j + mj ⊗ e,ij ) ⊗ ei = (Γ̂ji − Γk(κ)ji )mk ⊗ e j ⊗ ei . (9.21)

This, of course, pertains to the use of κ as reference. Using κ̄ instead gives

Ḡ = mj ⊗ ē j and ¯ Ḡ = (Γ̂kji − Γk )mk ⊗ ē j ⊗ ēi ,

∇ (9.22)
(κ̄)ji

where ēi are the reciprocals of ēi = x̄,i . The presumed insensitivity of our generic consti-
tutive function to the choice of reference configuration is thus equivalent to the constraint
162 Strain hardening, rate sensitivity, and gradient plasticity

F(G, ∇G) = F(Ḡ, ∇

¯ Ḡ). (9.23)

It follows, exactly as in the derivation of the representation (4.9) of the deformation

gradient, that

∇λ = ēi ⊗ ei , (9.24)

yielding (∇λ)−1 = ei ⊗ ēi , (∇λ)−t = ēi ⊗ ei , and, thus,

ē j = (∇λ)−t ej . (9.25)

Further, again with a slight abuse of notation, and assuming that λ is twice differentiable
with respect to the coordinates (so that the connection Γk(κ̄)ji is well defined), we have

Γk(κ̄)ji = ēk · x̄,ji = (∇λ)−t ek · λ,ji , (9.26)

which, of course, is symmetric in the subscripts.

In Section 6.4 we made essential use of the arbitrariness in the choice of reference,
i.e., the arbitrariness of the map λ. We do so again here, choosing λ(x) such that ∇λ = G
at the material point p in question, at a fixed instant in time; then, ēk = mk and Ḡ = I
(see (5.13)). In addition, noting that the first and second gradients of any map may be
k
specified independently at any one point, we choose λ(x) such that Γk(κ̄)ji (= mk ·λ,ji ) = Γ̂(ji)
at the point and instant in question, this being possible by virtue of the symmetry of Γ(κ̄)
in the subscripts. (See the paper by Cermelli and Gurtin for an explicit example of such
a map.) This results in

¯ Ḡ = T̂,
∇ where T̂ = T̂·kji mk ⊗ m j ⊗ mi , (9.27)

k k
in which T̂·kji = Γ̂[ji] is the torsion of the Weitzenböck connection Γ̂ji . We thus reduce
(9.23) to

F(G, ∇G) = F(I, T̂) = F̃(T̂), say. (9.28)

We have seen, in Section 5.2, that T̂ stands in one-to-one relation to the dislocation den-
sity α, and (9.18) thus follows from (9.28). Because of the arbitrariness of the material
point and time instant in question, this result holds at all such points and at all times.

9.3.1 Crystalline symmetry

Naturally, our constitutive function must conform to requirements imposed by mate-
rial symmetry. It is thus invariant under K → KR; equivalently, G → Rt G, for all
R ∈ gκi (p) ⊂ Orth+ . Recalling the discussion of Section 6.6, in the case of crystalline
symmetry the elements of gκi (p) are spatially uniform.
 Scale effects 163

Problem 9.4. If A is uniform and invertible, show that Curl(A−1 K−1 ) =

(CurlK−1 )A−t .

Thus, on replacing K by KR in (5.20), we find that

α → Rt αR, (9.29)

and hence that the constitutive function f(α) must be such that

f(α) = f(Rt αR); R ∈ gκi (p) . (9.30)

For example, any function of the form f(α) = g(|α|) automatically meets this require-
ment, for any orthogonal R.
Another way to arrive at this result is to note, from (9.22)2 , that mi → Rt mi and,
from K → KR with K = ei ⊗ mi , that mi → Rt mi for all rotations R belonging to gκi (p) .
k
This implies, for uniform R, that the Weitzenböck connection Γ̂ij is preserved under
symmetry transformations, i.e.

k k
Γ̂ij → Rt mk · (R t mi ),j = Rt mk · Rt mi,j = mk · RRt mi,j = mk · mi,j = Γ̂ij . (9.31)

k
It follows that the torsion T̂k·ij = Γ̂[ij] is invariant under symmetry transformations, and
therefore that it is uniquely defined for any kind of crystalline symmetry once a coordi-
nate system has been selected. The torsion therefore furnishes an invariant measure of
defect content, or inhomogeneity, in crystalline solids.
Note that length-scale effects in crystalline materials, represented by functions of the
form F(G, ∇G), are operative if and only if the plastic deformation is incompatible, i.e.,
CurlG ̸= 0.

9.3.2 Isotropy
The situation is quite different for isotropic materials. Again with reference to Section
6.6, recall that for these there is no requirement that the rotations R be uniformly dis-
tributed. For example, the scalars I1,2,3 comprising the arguments of the strain-energy
function for an isotropic solid, defined in the Appendix to Section 6.1, are invariant
under any rotation R, uniform or not. For non-uniform R, in place of (9.31) we find that

k k
Γ̂ij → Γ̂ij + Rt mk · Rt,j mi . (9.32)

The torsion thus transforms as T̂·kij → T̂·kij + Rt mk · Rt,[ j mi] , and the components of the
dislocation density as αij → αij + βij , where βij = ε̂kli Rt mj · Rt,[l mk] . In place of (9.30) we
have
164 Strain hardening, rate sensitivity, and gradient plasticity

f(α) = f(Rt (α + β)R), (9.33)

for any rotation R, which, for R = I, yields f(α) = f(α + β), and hence—because β can
assume arbitrary values—the conclusion that f is necessarily independent of the dislo-
cation density. Of course a dislocation density can be computed, but it cannot enter the
constitutive functions as an independent variable in the case of isotropy. This conclu-
sion is natural in view of the fact that dislocations are associated intimately with crystal
lattices, whereas isotropic materials are not crystalline. It follows from this line of rea-
soning that constitutive functions of the form F(G, ∇G) do not characterize isotropic
materials.
To explore this issue further, we recall that because of material symmetry, the field K
is non-unique to the extent that no distinction can be made between K and KR as far as
mechanical response is concerned, where, in the case of isotropy, R(x, t) is any rotation
field whatsoever. Equivalently, no distinction can be made between G and Rt G . We can
therefore select R to be the rotation RG in the polar factorization G = RG UG , in which
UG is the symmetric, positive definite plastic right-stretch tensor, to conclude that G
and UG are mechanically indistinguishable. Further, UG is the unique root of U2G = M,
where

M = Gt G = mij ei ⊗ e j (9.34)

is the plastic Cauchy–Green tensor (see (5.25)). Conversely, if G and Rt G yield the same
value of M, then R is an arbitrary rotation (see Problem 2.1) and hence an element of
the material symmetry group in the case of isotropy. Accordingly, for isotropic materials
G and M are equivalent descriptors of plastic deformation.

Problem 9.5. Use (9.34) to determine Ṁ in terms of Ġ. We have shown that G and M
are mechanically equivalent in isotropic materials. It should therefore be possible
−1
to determine Ġ in terms of Ṁ. Show that this is indeed the case if we take ĠG
to be symmetric, as stipulated, for isotropic materials, in Chapter 7. Write the flow
rule (7.8) in the form Ṁ = . . . .

In view of this observation, a natural course to follow when contemplating scale-

dependent constitutive functions for isotropic materials would be to adapt the devel-
opment for crystalline materials by considering constitutive functions of the form
F(M, ∇M) . However, as any function of M and ∇M is expressible as a function of
G and ∇G, and thus ultimately—if also insensitive to the reference configuration—as a
function of α, this option must be dismissed for the reasons already given. Another way
to arrive at this conclusion is to note that

∇M = mij|k ei ⊗ e j ⊗ ek (9.35)

may be reduced to the (third-order) zero tensor by a map λ from κ to κ̄, constructed such
that Γk(κ̄)ij = {kij } at the material point in question, where {kij } is the Levi-Civita connection
 Scale effects 165

based on the metric mij (see (5.35)). This is again permissible because Γ(κ̄) is symmetric
in the subscripts. The claim then follows from the fact that every Levi-Civita connection
is compatible with its own metric, so that Ricci’s lemma is then operative. The gradient
¯ M̄, computed using κ̄ as reference, is thereby reduced to zero at the point in question.
∇
Combining this with the fact that M̄ may be reduced to I at the same point, the presumed
invariance of the constitutive function under an arbitrary change of reference then yields
F(M, ∇M) = F(I, 0), and the conclusion follows.
The obvious next step in the search for a model of scale-dependent plasticity in
isotropic materials is to consider constitutive functions that depend on M, ∇M, and
∇∇M. However, rather than pursue this option in full generality, we follow the example
of the model for crystalline materials in which length-scale effects are due entirely to the
incompatibility of the plastic deformation. Extending this notion to isotropic materials,
we thus seek a model that is sensitive to the incompatibility of M, or, equivalently, to
that of the plastic strain 21 (M − I). We have encountered essentially the same issue in
Section 4.2, concluding there that strain compatibility is associated with the vanishing
of the relevant Riemann tensor. The appropriate descriptor of incompatibility in the
present instance is therefore the (generally non-vanishing) Riemann tensor constructed
from the Levi-Civita connection {kij }, with components

Rk.mlj = {kmj },l − {kml }, j + {kil }{imj } − {kij }{iml }. (9.36)

This involves the metric and its partial coordinate derivatives through the second order,
and accordingly accounts for plastic length-scale effects. The geometry of the mate-
rial manifold is thus Riemannian in the case of isotropic materials, in contrast to the
Weitzenböckian geometry of crystalline materials.

Problem 9.6. The connection {kij } also incorporates length-scale effects because it
involves the metric and its first-order coordinate derivatives. Would it be reasonable
to propose a constitutive function having {kij }mk ⊗ mi ⊗ m j as its only argument?

Problem 9.7. What is the form of the function F(M, ∇M, ∇∇M) which is such as
remain invariant under an arbitrary thrice-differentiable map x̄ = λ(x)?

We have seen, in Section 4.2, that the Riemann tensor associated with any Levi-Civita
connection possesses certain symmetries that, in three dimensions, render it equivalent
to the associated (symmetric) Ricci tensor. Accordingly, any function of the components
(9.36) is a (different) function of the Ricci tensor components (see (4.36))

ρij = Rk.jik . (9.37)

Using these we construct the symmetric tensor

ρ = ρij mi ⊗ mj , (9.38)
166 Strain hardening, rate sensitivity, and gradient plasticity

which is manifestly independent of any reference configuration. Accordingly, a consti-

tutive function of the form

F = f(ρ) (9.39)

is automatically insensitive to the choice of reference configuration.

To explore the extent to which such a function is consistent with material symmetry,
we note that the metric mij (= mi · mj ) remains globally invariant under symmetry trans-
formations; that is, it is invariant under mi → Rt mi with R(x, t) an arbitrary rotation
field, for all x ∈ κ. This means that its partial derivatives with respect to the coordi-
nates ξi are likewise invariant, and hence so too the connection {kij } and its derivatives.
Thus, the components ρij are invariant; then, because mi → Rt mi , material symmetry
considerations require that

f(ρ) = f(Rt ρR) for all rotations R. (9.40)

9.4 Scale-dependent yielding

The simplest way to incorporate scale effects into plasticity theory is to modify the yield
function accordingly. Thus, for crystalline materials we would propose that F(Ŝ) be
replaced by a function F(Ŝ, α), subject to
t
F(Ŝ, α) = F(Rt ŜR,R αR); R ∈ gκi (p) . (9.41)

For isotropic materials we could replace F(Ŝ) by a function F(Ŝ, ρ), subject to
t
F(Ŝ, ρ) = F(Rt ŜR,R ρR), for all rotations R. (9.42)

For example, we could consider yield functions of the form F = H(Ŝ)−k2 , where k is the
scale-dependent yield stress, a function of α or ρ, as appropriate. Indeed, in the case of
crystalline materials, such a proposal may be motivated by G. I. Taylor’s work in which
the flow stress is found to be related to dislocation content.
We retain the assumption—appropriate for rate-independent response—that plastic
evolution, i.e., K̇ ̸= 0, is possible only when F = 0, and further assume the stress Ŝ to
be always confined to the current elastic range defined by F(·, α) ≤ 0 or F(·, ρ) ≤ 0,
according as the material is crystalline or isotropic, respectively. In view of our restriction
to materially uniform bodies, we require, as assumed previously, that F be the same
function at all material points. The considerations about cyclic processes pertaining to
the work inequality (6.90), culminating in (6.103) and (6.104), are seen to carry over
unchanged, provided that the function F is continuous and α (or ρ, as appropriate)
is a continuous function of time during the cycle. Because α involves CurlG, which
may be specified independently of G at a given material point, such continuity is an
additional assumption beyond the continuity of K assumed previously. Exceptionally, the
 Scale-dependent yielding 167

continuity of α follows from that of K for cycles in which CurlG remains fixed. Invoking
(6.90) for such cycles, we recover (6.103) and (6.104), as before, but with F replaced
by F(S,α). Accordingly, the flow rule (6.105) remains in effect with this adjustment.
This flow rule generally predicts a non-zero value of CurlĠ in a given initial-
boundary-value problem, and therefore appears, at first glance, to be inconsistent with
the assumption CurlĠ = 0 just invoked in the course of deriving it. A little reflection is
sufficient to overcome this objection, however. For the work inequality pertains to a mate-
rial point, whereas the post facto evaluation of CurlĠ from the flow rule, at the same point,
requires the function Ġ(x, t) for x in a neighborhood of the point in question. Moreover,
the work inequality (6.90) purports to apply to all cyclic processes, and hence to those
restricted as indicated. Beyond this, the cyclic process associated with the work inequality
need have no relation to the actual process that the flow rule purports to describe.
In the case of isotropy we can similarly arrange a cyclic process in which the variation
of ρ is controlled entirely by the variation of K at a given material point, so that continuity
of the latter implies that of the former. Invoking (6.90) for such cycles, we arrive at
(6.105), with F replaced by F(Ŝ, ρ) and with vanishing plastic spin. A simple model of
this kind is discussed in the paper by Krishnan and Steigmann.

Problem 9.8. Prove these assertions.

To determine the plastic multiplier λ appearing in the flow rule, we need to invoke the
relevant consistency condition. In the case of crystalline materials, for example, this is

FŜ · (Ŝ)· + Fα · α̇ = 0, (9.43)

where, from (5.20),

−1 −1
α̇ = J−1
G G(CurlĠ) + (ĠG )α − (J̇G /JG )α, with J̇G /JG = tr(ĠG ). (9.44)

Substitution of (6.105) in (9.43) results in a linear constraint on λ and ∇λ jointly, fur-

nishing a first-order partial differential equation for λ whose solutions are subject to the
restriction λ > 0. This is a significant complication relative to scale-independent isotropic
hardening in which λ is determined algebraically.
The situation is even more complicated for isotropic materials owing to the fact that
in the operative consistency condition, namely

FŜ · (Ŝ)· + Fρ · ρ̇ = 0, (9.45)

the derivative ρ̇ involves Ṁ and its spatial derivatives through the second order,
culminating in a second-order partial differential equation for λ.
The formidable obstacles to analysis posed by the differential equations for the plastic
multiplier do not arise in the case of rate-sensitive viscoplastic response. For example,
(9.17) may be used with F equal to F(Ŝ, α) or F(Ŝ, ρ), as appropriate, with vanishing
plastic spin in the case of the second alternative. An example of this model, specialized
168 Strain hardening, rate sensitivity, and gradient plasticity

to cubic crystal symmetry with plastic spin incorporated, was proposed and studied by
Edmiston et al.
The proposed modifications to the yield function thus model scale effects while retain-
ing intact the structure of conventional viscoplasticity theory. They do not, however,
furnish models of the Bauschinger effect. We turn our attention to this problem in the
next, and final, section of this book.

9.5 Gradient plasticity

Naturally, there is no requirement that scale effects be confined to the yield function. It
is equally permissible to generalize the strain-energy function to take them into account.
The resulting models constitute a substantial branch of modern plasticity theory, known
collectively as gradient plasticity, that has dominated the recent research literature. The
book by Gurtin et al. is essential reading in this area.
Gradient plasticity is substantially more complicated than standard plasticity the-
ory. For example, the theory for isotropic materials is complicated by the fact that the
Ricci tensor involves spatial derivatives of the plastic strain through the second order.
In contrast, the relevant descriptor of scale effects in crystalline materials, the disloca-
tion density, involves the derivatives of the plastic deformation through the first order.
Accordingly, we confine ourselves to an outline of a basic framework for crystalline mate-
rials and refer to the paper by Steigmann for a discussion of the isotropic theory. To
capture the main ideas as simply as possible, we restrict attention to smooth fields.
Thus, we consider strain energies of the form W(H,α) in place of functions of the
form W(H) discussed previously. The definition (5.20) then yields the strain energy per
unit volume of κ as the function

Ψ(F,G, CurlG) = JG W(H, J−1

G GCurlG), (9.46)

and the strain energy of an arbitrary part π ⊂ κ of the body is

Z
U(π, t) = ΨdV. (9.47)
π

Imposing invariance of W under rotations, as in Chapter 6, we have

W(H,α) = U(Ê, α), (9.48)

where Ê is the elastic strain. This is subject to the material symmetry restriction
t
U(Ê, α) = U(Rt ÊR,R αR); R ∈ gκi (p) . (9.49)

As in the standard theory, the Piola and Piola–Kirchhoff stresses relative to κi (p) are WH
and Ŝ = UÊ , respectively, and are related by WH = HŜ.
 Gradient plasticity 169

9.5.1 Energetic response functions

Consider a one-parameter family of states {H(u), G(u), ∇G(u)} associated with a fixed
material point, and let superposed dots denote derivatives with respect to the parameter.
Then,

Ẇ = WH · Ḣ + Wα · α̇. (9.50)

Using (5.3) with the identity A · BC = ACt · B = Bt A · C, the first term on the right-hand
side may be written as

WH · Ḣ = WH Kt · Ḟ + Ft WH · K̇, (9.51)

where

K̇ = −KĠK. (9.52)

Also, from (9.44) it follows that

Wα · α̇ = (Wα )αt Kt · Ġ − (J̇G /JG )Wα · α + J−

G G Wα · CurlĠ.
1 t
(9.53)

Combining these results with

J̇G /JG = Kt · Ġ, (9.54)

we find the derivative of the referential energy density Ψ to be

Ψ̇ = JG [Ẇ + (J̇G /JG )W], (9.55)

where

Ẇ + (J̇G /JG )W = WH Kt · Ḟ + J−
G G Wα · CurlĠ
1 t

+[(W − Wα · α)I + (Wα αt − Ht WH )]Kt · Ġ. (9.56)

Altogether,

Ψ̇ = P · Ḟ + JG E′ Kt · Ġ + N · CurlĠ, (9.57)

where

P = ΨF = JG WH Kt , N = ΨCurlG = Gt Wα (9.58)

and

JG E′ Kt = ΨG , where E′ = (W − Wα · α)I + (Wα αt − Ht WH ) (9.59)

is Eshelby’s tensor, referred to κi (p) and extended to account for the dependence of W
on α. The tensors P,N, and E′ , in which P is the Piola stress relative to κ, are the relevant
170 Strain hardening, rate sensitivity, and gradient plasticity

response functions in the present theory, and the rate of change of the energy (9.47) is
given by
Z
U̇(π, t) = (P · Ḟ + JG E′ Kt · Ġ + N · CurlĠ)dV. (9.60)
π

Problem 9.9. Show that E′ → Rt E′ R under material symmetry transformations.

9.5.2 Stress power, balance laws, and dissipation

The lesson of Problem 1.2 is that the balance of mechanical energy, Eq. (1.44), follows
from the equation of motion, Eq. (1.40). We subsequently showed, conversely, that the
equation of motion follows from the virtual balance of energy, Eq. (1.50), in which the
actual velocity v = χ̇ is replaced by a kinematically admissible virtual velocity field u(x).
In the case of equilibrium (v̇ = 0 ) the latter is equivalent to the statement

P=S (9.61)

for any material region π ⊂ κ, where

Z Z
P= p · udA + ρκ b · udV (9.62)
∂π π

is the virtual power imparted to π and

Z
S= P · ∇udV (9.63)
π

is the virtual stress power. As explained in the book by Gurtin et al. and as is clear
from (1.50), Eq. (9.61) covers the non-equilibrium situation, too, if the inertia ρκ v̇ is
absorbed into the body force. However, for the sake of discussion we confine attention
here to equilibria.
We extend these ideas to gradient plasticity by replacing the “stress power” (1.46)
with
Z
S(π, t) = ˙
{P · Ḟ + JG QKt · Ġ + M · ∇G}dV, (9.64)
π

where Q and M, respectively, are second- and third-order tensors to be specified. The
virtual stress power S is given by the same expression, but with Ḟ = ∇u and with Ġ
regarded as a virtual plastic velocity field. We intend to integrate the terms involving
∇u and ∇Ġ by parts, using the divergence theorem. The first of these terms is easily
 Gradient plasticity 171

handled, as in the passage from (1.50) to (1.51). To treat the second, we write (9.21) in
the form
k
∇G = mj|i ⊗ e j ⊗ ei , where mj|i = mj,i − Γ̄ji mk . (9.65)

Thus,

∇Ġ = ṁj|i ⊗ ej ⊗ ei , (9.66)

and, writing M in the form

M = μkl ⊗ ek ⊗ el , (9.67)

we then have

M · ∇Ġ = μij · ṁi|j . (9.68)

Problem 9.10. Show that

j i
μij · ṁi|j = (μij · ṁi )|j − μij|j · ṁi , where μij|j = μij,j + μik Γ̄kj + μkj Γ̄kj .

The divergence theorem for smooth vector fields w = wi ei is

Z Z
DivwdV = w · νdA, (9.69)
π ∂π

or, in terms of components,

Z Z
wj|j dV = w j νj dA, (9.70)
π ∂π

where νj = ν · ej . Applying this with wj = μij · ṁi we then obtain

Z Z Z
M · ∇ĠdV = μij νj · ṁi dA − μij|j · ṁi dV. (9.71)
π ∂π π
˙ i , together with the identity a · Ab = a ⊗ b · A, we finally derive
Using ṁi = Ge
Z Z
S(π, t) = (Pν · χ̇ + Mν · Ġ)dA + {(JG QKt − DivM) · Ġ − DivP · χ̇}dV, (9.72)
∂π π

where

Mν = (μij ⊗ ei )νj and DivM = μij|j ⊗ ei . (9.73)

172 Strain hardening, rate sensitivity, and gradient plasticity

Invoking the virtual power statement (9.61) with S given by (9.72), in which χ̇ is
replaced by u and Ġ is a virtual field, we conclude that the virtual power has the form
Z Z
P= (p · u + Φ · Ġ)dA + (ρκ b · u + B · Ġ)dV. (9.74)
∂π π

The fields Φ and B are force-like quantities that generate power against Ġ . Following
Di Carlo and Quiligotti, we refer to these as remodeling forces.
Because u and Ġ are independent and arbitrary, we may invoke (9.61) with Ġ = 0.
This yields (1.50) (with v̇ = 0 ), which, as we have seen, implies that

p = Pν on ∂π and ρκ b = −DivP in π. (9.75)

The remaining content of (9.61) is

Z Z Z Z
Φ · ĠdA + B · ĠdV = Mν · ĠdA + (JG QKt − DivM) · ĠdV, (9.76)
∂π π ∂π π

which, because Ġ(x) is arbitrary, implies that

Φ = Mν on ∂π and B = JG QKt − DivM in π. (9.77)

These results must be modified if there is any a priori restriction on admissible vir-
tual fields Ġ. For example, in the case of plastic incompressibility Ġ is restricted by the
−1
constraint tr(ĠG ) = 0. Here, however, for simplicity’s sake we do not impose any
constraints of this kind.
Conversely, if (9.75) and (9.77) are satisfied, then

P(π, t) = S(π, t), (9.78)

where P(π, t) is the actual power, given by the right-hand side of (9.74) in which u is
replaced by the actual velocity χ̇ and Ġ is the material time derivative of G.
The dissipation is defined, as in the specialization of (6.56) to equilibrium, by

D(π, t) = P(π, t) − U̇(π, t), (9.79)

which combines with (9.60), (9.64), and (9.78) to give

Z
D(π, t) = DdV, (9.80)
π

where

D = JG (Q − E′ )Kt · Ġ + M · ∇Ġ − N · CurlĠ. (9.81)

 Gradient plasticity 173

As usual we assume that D(π, t) ≥ 0 for all π ⊂ κ and hence that

D≥0 at all x ∈ κ. (9.82)

The expression for D may be reduced by using (5.43) to write N · CurlĠ as a linear
form in ∇Ġ. Thus,

CurlĠ = ϵ̄ ijk ek ⊗ ṁ[ j,i] = ϵ̄ ijk ek ⊗ ṁ[ j|i] = ϵ̄ ijk ek ⊗ ṁj|i , (9.83)

k k
where the second equality follows from the symmetry Γ̄ji = Γ̄ij and the third from
ϵ̄ijk ṁ(j|i) = 0. We then have

N · CurlĠ = ϵ̄ ijk ek · Nṁj|i = ϵ̄ ijk Nt ek · ṁj|i = ηij · ṁi|j , where ηij = ϵ̄ jik Nt ek , (9.84)

and (9.81) may thus be written in the form

D = D · ĠG−1 + X · ∇Ġ, (9.85)

with

D = JG (Q − E′ ) and X = M − ηij ⊗ ei ⊗ ej . (9.86)

Naturally, we require, as in Problem 6.7, that the dissipation be invariant under mate-
rial symmetry transformations G → Rt G in which R ∈ gκi (p) is a fixed uniform rotation.
To explore the implications, let D̃ and X̃ be the transformed values of D and X. Then,
−1 −1
with ĠG → Rt ĠG R and ∇Ġ → Rt ∇Ġ, invariance of the dissipation is equivalent
to
−1
(RD̃Rt − D) · ĠG + (RX̃ − X) · ∇Ġ = 0. (9.87)

As this purports to hold at all material points and for all Ġ and ∇Ġ, and as the latter may
be specified independently at any such point, we conclude that

D → Rt DR and X → Rt X (9.88)

under symmetry transformations.

Problem 9.11. Show that Q → Rt QR under symmetry transformations.

j
Problem 9.12. Derive (9.87). Hint: Write X̃ in the form ξ ⊗ej and show that X̃·Rt ∇Ġ =
j i i j
ξ ⊗ ej · Rt Ġ,i ⊗ ei = ξ · Rt Ġ,i = Rξ · Ġ,i = Rξ ⊗ ej · Ġ,i ⊗ ei = RX̃ · ∇Ġ.

Problem 9.13. Show that N → NR under symmetry transformations and hence that
N · CurlĠ is invariant. Show that ηij → Rt ηij and hence that M → Rt M.
174 Strain hardening, rate sensitivity, and gradient plasticity

9.5.3 Example
The foregoing framework is quite general, the only a priori restrictions being that the
fields D and X must be such as to satisfy (9.82) and (9.88). To construct an explicit
model within the general framework we must therefore choose particular forms for D
and X, and rely, for the assessment of the resulting model, on a posteriori comparisons
of its predictions against empirical data. Given the paucity of such data we are content,
at present, to illustrate the main ideas as simply as possible. Accordingly, we consider
the choices

D = μĠG−1 and X = μl2 ∇Ġ, (9.89)

where μ is a scalar field and l is a material length scale, that is, a material constant
having the dimension of length. Assuming μ to be invariant under material symmetry
transformations, these trivially satisfy (9.88), reduce (9.85) to
 2  2
   
D = μ(ĠG−1  + l2 ∇Ġ ), (9.90)

and conform to (9.82) provided that μ ≥ 0.

Parallel developments of gradient plasticity discussed in the book by Gurtin et al. are
based on models that are “always on,” in the sense that no threshold need be met for
the onset of plastic flow. Indeed plasticity models of this kind, sans yield function, are
commonly used to model viscoplastic response. To simplify matters we adopt the same
assumption here, and, for the sake of definiteness, choose

μ = 2η/P(Ê, α), (9.91)

where η, a positive constant, is a material viscosity and P is a positively valued

dimensionless function satisfying the usual material symmetry restriction P(Ê, α) =
P(Rt ÊR, Rt αR). An example of such a function is
2
P = (1 + pl2 |α| )q , (9.92)

where p and q are positive dimensionless constants.

According to (9.86) and (9.89), we have

JG Q = μĠG−1 + JG E′ and M = μl2 ∇Ġ + ηij ⊗ ei ⊗ ej , (9.93)

the latter yielding

μij = μl2 eik e jl ṁk|l + ηij , (9.94)

and the balance law (9.77)2 reduces to

ĠG−1 Kt = μ−1 (B − JG E′ Kt + DivM). (9.95)

 Gradient plasticity 175

We observe, from (9.73), (9.91), and (9.93), that the highest-order spatial derivative
of Ġ occurring in this equation is l2 eik e jl ṁk|lj ⊗ ei . Using ṁk = Ġek , and the consequent
fact that ṁk|lj = Ġ|lj ek , we reduce this to

l2 eik e jl ṁk|lj ⊗ ei = l2 (e jl Ġ|lj )ei ⊗ ei = l2 e jl Ġ|lj = l2 ΔĠ, (9.96)

where Δ(·) = eij (·)|ij is the referential Laplacian. Accordingly, (9.95) is a quasilinear
second-order elliptic partial differential equation for Ġ. It becomes a linear equation if
we take P = const. in (9.91).
If no external agency is available to supply a volumetric remodeling force, i.e., if B
vanishes, then (9.95) furnishes the flow rule

ĠG−1 = μ−1 {(DivM)Gt − JG E′ }, (9.97)

which, together with (9.91), may be compared to the first branch of (9.17). It may also
be compared to conventional scale-independent models of kinematic hardening, in which
the bracket on the right-hand side of (9.97) is replaced by a difference between the
stress Ŝ and a tensor-valued variable that purports to describe the Bauschinger effect.
See Eq. (3.1.9) in Lubliner’s book, for example. With reference to (9.59), written in
the form

E′ = (U − Uα · α)I + (Uα αt − Ht HŜ), (9.98)

the tensor variable in question is seen to be represented, in the case of small elastic strain
(Ht H ≃ I), by the energetic effect of the dislocation density together with the diver-
gence term in (9.97). Solutions to (9.97) also exhibit plastic spin. The Bauschinger
effect, associated with the emergence of dislocations and hence an inherently scale-
dependent phenomenon, is thus a natural outcome of gradient plasticity theory for
crystalline materials.
Equation (9.97) requires the specification of boundary conditions on ∂κ. With ref-
erence to (9.76) these are seen to be consistent with the specification of Φ or G. For
example, taking G to be fixed (Ġ = 0) on a part of ∂κ and Φ = 0 on the remainder
corresponds to the “hard” and “free” conditions specified by Gurtin et al., the former
simulating a boundary, abutting a hard material, that blocks the passage of dislocations
via plastic slip, and the latter a boundary across which they can flow freely. Given an
initial field G(x, t0 ) compatible with the condition imposed on a hard part of the bound-
ary, the solution to the boundary-value problem for Ġ(x, t) can then be used to update
G(x, t).
Of course the tentative model offered in this subsection is merely one among an
infinite number of possibilities encompassed by the general framework. Ultimately,
the usefulness of this or any similar model must be judged by the community of
plasticians.
176 Strain hardening, rate sensitivity, and gradient plasticity

References
Atai, A. A., and Steigmann, D. J. (2014). Transient elastic-viscoplastic dynamics of thin sheets. J.
Mech. Mat. Struct. 9, 557–74.
Batchelor, G. K. (Ed.) (1958). The Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 1: Mechanics
of Solids. Cambridge University Press, Cambridge, UK.
Cermelli, P., and Gurtin, M. E. (2001). On the characterization of geometrically necessary
dislocations in finite plasticity. J. Mech. Phys. Solids 49, 1539–68.
Di Carlo, A., and Quiligotti, S. (2002). Growth and balance. Mech. Res. Comm. 29, 449–456.
Edmiston, J., Steigmann, D. J., Johnson, G., and Barton, N. (2013). A model for elastic-
viscoplastic deformations of crystalline solids based on material symmetry: Theory and
plane-strain simulations. Int. J. Eng. Sci. 63, 10–22.
Fleck, N. A., Muller, G. N., Ashby, M. F., et al. (1994). Strain gradient plasticity: Theory and
experiment. Acta Metall. et Mater. 42, 475–87.
Gurtin, M. E., Fried, E., and Anand, L. (2010). Mechanics and Thermodynamics of Continua.
Cambridge University Press, Cambridge, UK.
Hill, R. (1950). The Mathematical Theory of Plasticity. Clarendon Press, Oxford.
Hutchinson, J.W. (2000). Plasticity at the micron scale. Int. J. Solids Structures 37, 225–38.
Krishnan, J., and Steigmann, D. J. (2014). A polyconvex formulation of isotropic elastoplasticity
theory. IMA J. Appl. Maths. 79, 722–38.
Lubliner, J. (2008). Plasticity Theory. Dover, New York.
Perzyna, P. (1962/3). The constitutive equations for rate-sensitive plastic materials. Quart. Appl.
Math. 20, 321–32.
Steigmann, D. J. (2022). Gradient plasticity in isotropic solids. Math. Mech. Solids. 27,
1896–1912.
Stölken, J. S., and Evans, A. G. (1998). A microbend test method for measuring the plasticity
length scale. Acta Mater. 46, 5109–15.
Taylor, G.I. (1934). The mechanism of plastic deformation of crystals. Part 1: Theoretical. Proc.
R. Soc. Lond. A 145, 326–87.
Teodosiu, C. (Ed.) (1997). Large Plastic Deformation of Crystalline Aggregates. CISM Courses and
Lectures, Vol. 376. Springer, Vienna.
Solutions to selected problems

t t
2.1. If C̄ = C then Ft Q̄ Q̄F = F F. Pre-multiply by F−t and post-multiply by F−1 to conclude
t t
that Q̄ Q̄ = i. Conversely, if Q̄ Q̄ = i then it follows immediately that C̄ = C.

2.3. AB · D = tr(ABDt ) = tr((ABDt )t ) = tr(DBt At ) = DBt · A. Also, A · BD = tr(A(BD)t ) =

tr(ADt Bt ) = tr(Bt ADt ) = Bt A · D.

2.5(b). Suppose a uniform pressure p is assigned on the entire boundary. We have

Z Z
P = −p n · χ̇da = −p F∗ ν · χ̇dA
∂κt ∂κ
Z Z Z
= −p Div[(F∗ )t χ̇]dV = −p (χ̇ · DivF∗ + F∗ · Ḟ)dV = −p J̇F dV,
κ κ κ

where we’ve used the Piola identity DivF∗ = 0 together with ∇χ̇ = Ḟ and (JF )F = F∗ . If p is fixed,
independent of time, then

P= dt L,
d

where
Z
L = −p JF dV = −pvol(κt ).
κ

A simpler way to arrive at the same conclusion is to use

Z Z Z Z
n · χ̇da = divχ̇dv = (J̇F /JF )dv = J̇F dV.
∂κt κt κt κ

This is not the only kind of pressure loading for which a load potential exists. Further exam-
ples include a uniform volume-dependent pressure and a hydrostatic pressure in a uniform
gravitational field. Can you derive the forms of the associated load potentials?

2.6. The mechanical power identity for the entire body is P = S + dt K,

d
where
Z
S= P · ḞdV.
κ

Using

P = ΨF + μFĖ,
178 Solutions to selected problems

where μ is the viscosity and Ė is the material derivative of the Lagrange strain, we have

P · Ḟ = Ψ̇ + μFĖ · Ḟ,

where (noting that A · BD = Bt A · D)

t t
FĖ · Ḟ = Ḟ F · Ė = Sym(Ḟ F) · Ė = Ė · Ė.

We have used E = 12 (Ft F − I) together with the fact that Ė is symmetric. Thus,
 2
 
P · Ḟ = Ψ̇ + μ Ė

and
Z  2 Z
 
S= dt U
d
+ μ Ė dV, where U= ΨdV.
κ κ

The dissipation D is
Z  2
 
D=P− d
dt (U + K) = S − dt U
d
= μ Ė dV,
κ

where we have invoked the mechanical energy balance in the second equality. Thus, D vanishes
in the case of purely elastic response (μ = 0).

(a) Assuming μ > 0 we immediately conclude that D ≥ 0. If Ė = 0 at all material points, then
obviously D = 0. Conversely, if D = 0, then Ė = 0 at all points. For if there is a point where
Ė ̸= 0, then by continuity there is a neighborhood of that point where |Ė| > 0, yielding a
positive value of the integral and hence D > 0, a contradiction.
(b) In the case of conservative loading we have P = dt L and E
d
= (U −L)+K. Then, D = − dtd E
and the inequality D ≥ 0 is equivalent to

d
dt E ≤ 0.

2.7. For the first part, write x = x0 + ϵw in (2.60), with dV(x) = ϵ3 dV(w). Divide by ϵ3 (> 0)
and let ϵ → 0 to get (2.63). In the second part, the coefficient of i is found to be

∇ξ2 · A[∇ξ1 ] − ∇ξ1 · A[∇ξ2 ] = ∇ξ1 · At [∇ξ2 ] − ∇ξ1 · A[∇ξ2 ],

which vanishes by the major symmetry of A (At = A).

2.9. (a) If gκ1 (p) = Orth+ , then R ∈ gκ1 (p) is an arbitrary rotation. The corresponding element
of gκ2 (p) is KRK−1 , where

K = λk ⊗ k + λ−1/2 (I − k ⊗ k) and K−1 = λ−1 k ⊗ k + λ1/2 (I − k ⊗ k).

 Solutions to selected problems 179

Assume the material to be isotropic relative to κ2 , so that gκ2 (p) = Orth+ . Then (KRK−1 )−1 =
(KRK−1 )t , which is equivalent to K2 R = RK2 because K is symmetric. We obtain

K2 R − RK2 = (λ2 − λ−1 )(k ⊗ Rt k − Rk ⊗ k).

There are infinitely many rotations R for which this is not zero, unless λ = 1. For such R we have
KRK−1 ∈ / Orth+ and so our assumption is false: gκ2 (p) ̸= Orth+ and the material is not isotropic
relative to κ2 unless λ = 1, i.e., unless K = I.
(b) If R ∈ Orth+ and Rk = k, then Rt k = k and K2 R − RK2 = 0; thus, KRK−1 ∈ Orth+ from
part (a). Using Rodrigues’ formula we have

R = k ⊗ k + S, where S = cos θ(i ⊗ i + j ⊗ j) + sin θ(j ⊗ i − i ⊗ j),

in which θ is an arbitrary real number and {i, j, k} is a right-handed orthonormal basis. We obtain

KRK−1 = λ−1 Kk ⊗ k + λ1/2 KS = k ⊗ k + S = R,

and the material is also transversely isotropic relative to κ2 .

3.6. dy = gi dξ i = i1 dξ1 + (sin ϕi 1 + cos ϕi2 )dξ2 + i3 dξ3 = ii dy i , yielding dy1 = dξ1 + (sin ϕ)dξ2 ,
dy = (cos ϕ)dξ2 and dy3 = dξ3 . Integrate with ϕ = constant and set y i (0, 0, 0) = 0 to get y1 =
2

ξ1 + (sin ϕ)ξ2 , y2 = (cos ϕ)ξ2 and y3 = ξ3 . The rest is straightforward.

3.7. Use A = A·ij gi ⊗ g j with b · At a = a · Ab for all a, b to get gk · At g l = g l · (A·ij gi ⊗ g j )gk =

· gi = A·ij δkj δil = A·lk . Then, At = (gk · At g l )g k ⊗ gl = A·lk g k ⊗ gl = A·j i gi ⊗ g j where, in
A·ij δkj g l
the last step, we have raised and lowed indices using the metric and reciprocal metric. Comparing
with A we have that A·j i = A·ij if At = A and A·j i = −A·ij if At = −A. The rest is straightforward.

3.13. We need dy = gi dξ i = ēr dr̄ + r̄dēr + i3 dz̄, where dēr = cos ϕeθ (θ)dθ and {ξ i } = {r̄, θ, z̄},
yielding g1 = ēr , g2 = r̄ cos ϕeθ and g3 = i3 . The rest is straightforward.

3.15. We have g1 = y,r = cos θn+sin θb, g2 = y,θ = −r sin θn+r cos θb, g3 = y,s = r′ +r cos θn′ +
r sin θb′ = (1−κr cos θ)t+τr cos θb−τr sin θn. Then, g11 = 1, g22 = r2 , g23 = g32 = τr2 , and g33 = H,
all other gij being zero, where H = (1 − κr cos θ)2 + τ2 r2 . This yields g = det(gij ) = r2 (1 − κr cos θ).
√
Also, |κr| < 1 and |cos θ| ≤ 1 so g = r(1 − κr cos θ). We also get g11 = 1, g22 = H/g, g33 = r2 /g,
g23 = g32 = −τr2 /g, all others zero. For a function f(r, θ) we use Δf = g−1/2 (g1/2 gij f,j ),i to get
(eventually!)
   h i
Δf = ∂2 f + 1 − κr cos θ 1 ∂f κ sin θ 1− τ2 r2 1 ∂f
∂r2 1−κr cos θ r ∂r + 1−κr cos θ (1−κr cos θ)2 r ∂θ
h i 2
τ2 r2 1 ∂ f
+ 1+ 2.
(1−κr cos θ)2 r ∂θ
2

3.16. The transformation formula

∂ξ¯ ∂ξ m ∂ξ n l
k

Ā·kij = A ,
∂ξ l ∂ξ¯ i ∂ξ¯ j ·mn
180 Solutions to selected problems

for the components A·kij of a third-order tensor, follows directly from A·kij gk ⊗ g i ⊗ g j = A =
Ā·kij ḡk ⊗ ḡ i ⊗ ḡ j .
(a) Using (3.56) we have

∂ξ k l ∂gi ∂ξ¯
m

Γijk = g k · gi,j = l ḡ ·
∂ξ¯ ∂ξ
m
∂ξ¯
j

 
∂ξ k ∂ξ¯ l ∂ ∂ξ¯
m n

= j ḡ · ¯m ḡ
¯l
∂ξ ∂ξ ∂ξ ∂ξ i n
 
∂ξ k ∂ξ¯ ∂ξ¯ l ∂ξ k ∂ξ¯ ∂ ∂ξ¯
m n m l

= ¯l i Γ̄nm + ¯m ,
∂ξ¯ ∂ξ ∂ξ
j l j
∂ξ ∂ξ ∂ξ ∂ξ i
   
∂ξ¯ ∂ξ¯ ∂ξ s ∂2 ξ¯ ∂ξ s ∂ξ¯
l l l m
∂ξ s
where ∂¯ m = ∂s ¯m = ¯m i . Use this with
s
j = δj to get
∂ξ¯ ∂ξ
m
∂ξ ∂ξ i ∂ξ ∂ξ i ∂ξ
s
∂ξ ∂ξ ∂ξ

∂ξ k ∂ξ¯ ∂ξ¯ l ∂ξ k ∂2 ξ¯
m n l

Γijk = ¯l i Γ̄nm + j.
∂ξ¯ ∂ξ ∂ξ
j l i
∂ξ ∂ξ ∂ξ

The presence of the second term on the right implies that Γijk are not the components of a tensor.
(b) We have

∂ξ k ∂ξ¯ ∂ξ¯ l ∂ξ k ∂2 ξ¯
m n l

Γjik = ¯l j Γ̄nm +
∂ξ¯ ∂ξ ∂ξ
i l j i
∂ξ ∂ξ ∂ξ
∂ξ k ∂ξ¯ ∂ξ¯ l ∂ξ k ∂2 ξ¯
n m l

= ¯l j Γ̄mn + j,
∂ξ¯ ∂ξ ∂ξ
i l i
∂ξ ∂ξ ∂ξ

implying that

∂ξ k ∂ξ¯ ∂ξ¯
n m
l l
Γijk − Γjik = j (Γ̄nm − Γ̄mn ).
∂ξ¯ ∂ξ ∂ξ
l i

k
Thus, Γ[ij] is a third-order tensor.

3.19. (a) Let Aij = vi;j = vi,j − vm Γijm . Then,

Aij;k = Aij,k − Alj Γikl − Ail Γjkl

= vi,jk − vm,j Γikm − vm Γik,j
m
− vl,k Γijl + vm Γlkm Γijl − vi,l Γkjl + vm Γilm Γkjl
+vl,k Γijl − vm,k Γijm − vl,j Γikl + vm,j Γikm ,

in which the last line is seen to vanish identically. Thus,

m m
vi;jk − vi;kj = (Γik,j − Γij,k + Γljm Γikl − Γlkm Γijl )vm − (vi,l − vm Γilm )(Γjkl − Γkjl ),

which is the stated formula.

 Solutions to selected problems 181

i
(c) Let ξi be general coordinates and let ξ̄ = y i be Cartesian coordinates. The latter are admis-
sible coordinates in Euclidean space. Because vi;jk are the covariant components of a third-order
tensor, we have

vi;jk = ∂y i ∂y j ∂y k v̄l,mn ,
l m n

∂ξ ∂ξ ∂ξ

where, on the right-hand side, the partial derivatives of v̄l are with respect to the y′ s and we have
l
used the fact that the Γ̄ik are identically zero. Then,

vi;kj = ∂y i ∂y k ∂y j v̄l,mn = ∂y i ∂y j ∂y k v̄l,nm ,

l m n l m n

∂ξ ∂ξ ∂ξ ∂ξ ∂ξ ∂ξ

and

vi;jk − vi;kj = ∂y i ∂y j ∂y k (v̄l,mn − v̄l,nm ) = 0

l m n

∂ξ ∂ξ ∂ξ

if v̄l is twice continuously differentiable.

4.8. Symmetry of the Ricci tensor follows directly by using the symmetry of the metric
together with the major symmetry and minor skew symmetries of the Riemann tensor. Thus,
Rij = mkl Rkjil = mkl Rilkj = mlk Rlijk = Rji .

4.10. We present the conventional derivation based on the Riemann tensor. We have

i i
Rpmlj = Γpmj,l − Γpml,j + Γml Γipj − Γmj Γipl ,

where

Γkij = 21 (gki,j + gjk,i − gij,k ) and Γijl = glk Γkij .

Using

l l
(eij + 2Eij )|k = gij|k = gij,k − glj Γ̄ik − gli Γ̄jk ,

together with eij|k = 0 and Γijl = Γjil , we get

Γkij = Γ̄kij + 2γkij ,

where

l
γkij = Ekl Γ̄ij + Ekij ,

with

Ekij = Eki|j + Ejk|i − Eij|k . (*)

182 Solutions to selected problems

Choosing the convected coordinates to be Cartesian in the reference configuration we have

l
Γ̄ij = 0, identically, and hence the simplification

Γkij = 2Ekij , with Ekij = Eki,j + Ejk,i − Eij,k ,

and

Γijl = 2glk Ekij .

Substituting into the expression for R, we get

Rpmlj = 2[(Epmj,l − Epml,j ) + 2gik (Ekml Eipj − Ekmj Eipl )],

where

(gik ) = (eij + 2Eij )−1 .

Using Rpmlj = 0 and transforming to arbitrary coordinates, the compatibility conditions reduce to

Epmj|l − Epml|j + 2gik (Ekml Eipj − Ekmj Eipl ) = 0,

where E kml is given by (*), and where

m m m
Eijk|l = Eijk,l − Emjk Γ̄il − Eimk Γ̄jl − Eijm Γ̄kl .

Using (4.36) with the expression for Rpmlj yields the six non-trivial compatibility conditions.

4.12. We have Ė = 21 ġij e i ⊗e j and Ẇ = U̇ = UE · Ė = S· Ė = 12 S · e i ⊗e j ġij = 12 Sij ġij = 21 JF Tij ġij .

The result follows by equating this to Ẇ = ∂W ġij and demanding that the two expressions agree for
∂gij
all symmetric ġij . Note that for the partial-derivative notation to be meaningful, we must replace
gij by 21 (gij + gji ) in the expression for W and regard gij as being non-symmetric, so that ∂W = ∂W .
∂gij ∂gji

5.2. H−1 dy = H−1 Fdx = (FK)−1 Fdx = K−1 F−1 Fdx = Gdx.

5.4. From the referential counterparts of (3.85) and (3.117), we have (CurlA)t = Bjm ej ⊗
em , where Bjm = ϵ̄ilm A·jl|i ; and, Div[(CurlA)t ] = Bjm jm mil j
|m ej , where B|m = ϵ̄ A·l|im in which use has
been made of the fact that the permutation tensor is covariantly constant. Bjm
|m vanishes because,
in Euclidean space, A·jl|im is symmetric in the last pair of subscripts. This follows from the fact
that A·jl|im are the components of a fourth-order tensor, the covariant derivatives reducing, in a
Cartesian system, to symmetric partial derivatives. It is a simple matter to show that this symmetry
is preserved by the relevant tensor transformation formula.
 Solutions to selected problems 183

(b) From (5.43) it follows that

(CurlG)t = L k ⊗ ek ,

where

L k = ϵ̄ijk T̂jil ml .
√
Then, from (4.51), (4.54), (5.29), (5.31), and ϵ̄ijk = eijk / e (see (3.32)),
√ n
0 = Div[(CurlG)t ] = 1
√
e
( eL k ),k = ϵ̄ijk (T̂ji,k
n
+ T̂jil Γ̂lk )mn .

This is equivalent to
n
ϵ̂ijk (T̂ji,k
n
+ T̂jil Γ̂lk ) = 0. (*)

(c) Next, note that

h k k
 k i k i
i
ϵ̂mlj R̂·kmlj = ϵ̂mlj Γ̂mj,l − Γ̂ml,j − Γ̂ij Γ̂ml − Γ̂il Γ̂mj
h k k
 k i k i
i
= ϵ̂mlj Γ̂[mj],l − Γ̂[ml],j − Γ̂ij Γ̂[ml] − Γ̂il Γ̂[mj]
h  k k i
i
= ϵ̂mlj T̂mj,l
k k
− T̂ml,j i
− Γ̂ij T̂ml − Γ̂il T̂mj
 k
  k

= k
ϵ̂jml T̂mj,l i
+ T̂mj k
Γ̂il + ϵ̂lmj T̂ml,j + T̂mli
Γ̂ij
 k

= 2ϵ̂jml T̂mj,l
k
+ T̂mj i
Γ̂il ,

and therefore (*) yields ϵ̂mlj R̂·kmlj = 0, which is equivalent to

ϵ̂mlj R̂kmlj = 0.

Conversely, we prove (*) by reversing the steps and invoking R̂·km(lj) = 0.

(d) From (5.57) we have
pq
R̂kmlj = ϵ̂pkm ϵ̂qlj Π̂ .

Then,
pq
0 = ϵ̂mlj R̂kmlj = ϵ̂mlj ϵ̂mpk ϵ̂qlj Π̂
  pq
= δpl δkj − δpj δkl ϵ̂qlj Π̂
pq
= (ϵ̂qpk − ϵ̂qkp ) Π̂
pq
= 2ϵ̂qpk Π̂ .
184 Solutions to selected problems

[ij]
Accordingly, the equations Π̂ = 0 are identities.

6.6. Replace G by F and mj by gj = y,j in (5.43).

 
t   t
6.9. From JH T = HŜH and Ŝ = C[Ê] + o(Ê) we have JH trT = tr(HŜH ) = tr(Ht HŜ) =
     
     
tr[(I + 2Ê)Ŝ] = trŜ + o(Ê). Use J2H = det(I + 2Ê) = 1 + O(Ê) to get JH = 1 + O(Ê). Then,
   
   
J− 1
H = 1 + O(Ê) also, and trT = trŜ + o(Ê).

6.10. First, note that

 2
1
E11 E22 + E11 E33 + E22 E33 = Ē11 Ē22 + Ē11 Ē33 + Ē22 Ē33 + 3 trÊ
 

+ 23 trÊ Ē11 + Ē22 + Ē33
 2

1 1
= 3 trÊ + 2 2Ē11 Ē22 + 2Ē11 Ē33 + 2Ē22 Ē33
 2 


1 1
= 3 trÊ + 2 Ē11 Ē22 + Ē33 + Ē22 Ē11 + Ē33


+Ē33 Ē11 + Ē22
 2  
= 1
3 trÊ − 12 Ē211 + Ē222 + Ē233 .

Thus, the strain energy is expressible in the form

U(Ê) = 12 [C1 (E11 + E22 + E33 )2 + C2 (Ē211 + Ē222 + Ē233 ) + C3 (E212 + E213 + E223 )].

If this is to be positive-definite, i.e., U(Ê) > 0 for all non-zero Ê, it must be positive for non-
2
zero strains with DevÊ = 0. Then 21 C1 (trÊ) > 0 with trÊ ̸= 0, implying that C1 > 0. Similarly, we
can choose Ê = DevÊ, with Ēij = 0 for i ̸= j, to conclude that C2 > 0; and finally, choose Ê = DevÊ
with Ēij = 0 for i = j, to conclude that C3 > 0. Thus, C1,2,3 > 0 are necessary conditions for the
positive definiteness of U(Ê). From the above expression for U(Ê), they are clearly also sufficient.
·
To obtain the stress, we use Ŝ · (Ê) = U̇, where
·
U̇ = C1 (trÊ)I · (Ê) + C2 [Ē11 (Ē11 )· + Ē22 (Ē22 )· + Ē33 (Ē33 )· ] + C3 [E12 (E12 )·
+E13 (E13 )· + E23 (E23 )· ].
· · ·
Use Ėij = li · (Ê) lj = li ⊗ lj · (Ê) = Sym(li ⊗ lj ) · (Ê) , and, similarly,
·
(Ēij )· = Sym(li ⊗ lj ) · (DevÊ)
· ·
= Sym(li ⊗ lj ) · Dev(Ê) = Dev[Sym(li ⊗ lj )] · Dev(Ê)
·
= Dev[Sym(li ⊗ lj )] · (Ê) ,

to conclude that
· ·
(Ē11 )· = Dev(l1 ⊗ l1 ) · (Ê) = (l1 ⊗ l1 − 13 I) · (Ê) ,
 Solutions to selected problems 185

and, similarly, that

· ·
(Ē22 )· = (l2 ⊗ l2 − 13 I) · (Ê) and (Ē33 )· = (l3 ⊗ l3 − 31 I) · (Ê) .

Then,

·
Ē11 (Ē11 )· + Ē22 (Ē22 )· + Ē33 (Ē33 )· = (Ē11 l1 ⊗ l1 + Ē22 l2 ⊗ l2 + Ē33 l 3 ⊗ l3 ) · (Ê)
·
− 13 (Ē11 + Ē22 + Ē33 )I · (Ê)
·
= (Ē11 l1 ⊗ l1 + Ē22 l2 ⊗ l2 + Ē33 l3 ⊗ l3 ) · (Ê) .

Thus,

· ·
U̇ = C1 (trÊ)I · (Ê) + C2 (Ē11 l1 ⊗ l1 + Ē22 l2 ⊗ l2 + Ē33 l3 ⊗ l3 ) · (Ê)
·
+C3 [E12 Sym(l1 ⊗ l2 ) + E13 Sym(l1 ⊗ l3 ) + E23 Sym(l2 ⊗ l3 )] · (Ê) ,

and therefore

Ŝ = C1 (trÊ)I + C2 (Ē11 l1 ⊗ l1 + Ē22 l2 ⊗ l2 + Ē33 l3 ⊗ l3 )

+ 12 C3 [E12 (l1 ⊗ l2 + l2 ⊗ l1 ) + E13 (l1 ⊗ l3 + l3 ⊗ l1 ) + E23 (l2 ⊗ l3 + l3 ⊗ l2 )].

7.3. From Problem 6.10 we have

F̃(DevŜ) = 12 A1 (S̄211 + S̄222 + S̄233 ) + A2 (S212 + S213 + S223 ) + const.,

where A1,2 are constants and the stress components are with respect to the normalized lattice basis
{li }. As in that problem, we also have

FŜ = A1 (S̄11 l1 ⊗ l1 + S̄22 l2 ⊗ l2 + S̄33 l3 ⊗ l3 )

+A2 [S12 (l1 ⊗ l2 + l2 ⊗ l1 ) + S13 (l1 ⊗ l3 + l3 ⊗ l1 ) + S23 (l2 ⊗ l3 + l3 ⊗ l2 )].

Note that FŜ ∈ Dev. Then,

Ŝ · FŜ = DevŜ · FŜ = A1 (S̄211 + S̄222 + S̄233 ) + 2A2 (S212 + S213 + S223 ),

where we’ve used S̄ij = Sij for i ̸= j. Using

 2
  2 2 2 2 2 2
DevŜ = S̄11 + S̄22 + S̄33 + 2(S12 + S13 + S23 ),

we see that Ŝ·FŜ vanishes if DevŜ vanishes (because, then, S̄211 + S̄222 + S̄233 = 0 and S212 +S213 +S223 =
0). Further, to ensure that Ŝ · FŜ > 0 when DevŜ ̸= 0 it is necessary and sufficient that A1 > 0
(pick Sij = 0 for i ̸= j) and A2 > 0 (pick S̄11 = S̄22 = 0; thus, S̄33 = 0 too).
186 Solutions to selected problems

Because A2 > 0, we can rewrite the yield condition F̃(DevŜ) = 0 with F̃ given by

F̃(DevŜ) = 12 A(S̄211 + S̄222 + S̄233 ) + S212 + S213 + S223 − k2 ,

where A is a positive constant and k is the yield stress for stress states of the form S12 (l1 ⊗l2 +l2 ⊗l1 ),
S13 (l1 ⊗ l3 + l3 ⊗ l1 ) or S23 (l2 ⊗ l3 + l3 ⊗ l2 ), i.e., for shearing stresses on the crystallographic axes.
As expected, the symmetry of the lattice implies that the yield stresses on these axes are all equal.
Note that this yield function reduces to von Mises’ function when A = 1.

7.4. We have
−1
ĠG = λFŜ ,

where

F(Ŝ) = 12 DevŜ · DevŜ − [α(trŜ) + k]2 .

Then,
· · ·
FŜ · (Ŝ) = DevŜ · Dev(Ŝ) − 2α[α(trŜ) + k]I · (Ŝ)
· ·
= DevŜ · (Ŝ) − 2α[α(trŜ) + k]I · (Ŝ) ,

yielding

FŜ = DevŜ − 2α[α(trŜ) + k]I.

t t
In the rigid-plastic case we have H = Q ∈ Orth+ , Ŝ = Q TQ, DevŜ = Q τQ, where τ = devT.
Then trŜ = trT, and

−1
D = Q(ĠG )Qt = λQFŜ Qt
= λ{τ − 2α[α(trT) + k]i},

which implies that

trD = −6αλ[α(trT) + k].

This is generally non-zero, and so the material is compressible.

7.6. We have v = w(r)k and L = k ⊗ gradw = w′ (r)k ⊗ er . Then, v̇ = vt + Lv = 0, and we solve

divT = 0, assuming zero body force, i.e.,

gradp = divτ,

with
√
τ = (2η + 2k
|D |
)D, if |τ| ≥ 2k,
 Solutions to selected problems 187

and D = 0 if |τ| < 2k, where

D = 12 w′ (r)(k ⊗ er + er ⊗ k).

Thus, |D| = 21 (w′ )2 and

τ = τ(k ⊗ er + er ⊗ k),

where
 
τ(r) = ηw′ + kw′ / w′  .

We thus have

gradp = (τ′ + τ/r)k,

implying that ∂p/∂r = 0 = ∂p/∂φ and

p′ (z) = τ′ + τ/r.

Thus, p′′ (z) = 0 and p′ (z), the axial pressure gradient, is constant. If this is zero, then

τ(r) = A/r,
√
where A is constant. We have |τ| = 2τ and so yield occurs when |τ(r)| ≥ k, i.e., for values of
radius satisfying

r ≤ |A| /k.

From the relation between τ and w we have

 
ηw′ + kw′ / w′  = τ = A/r.

This is valid for r ≤ |A| /k. Given the geometry of the problem, when W > 0 it is reasonable to
expect that w′ < 0. Then,

ηw′ − k = A/r,

which gives

ηw(r) = kr + A ln r + B,

where B is another constant. This is valid in the region (a,r̄), where

r̄ = |A| /k.
188 Solutions to selected problems

In the region (r̄, b) we have |τ(r)| < k and |D| = 0; thus, w′ = 0 in this region, implying that
w(r̄) = w(b) = 0. Using this together with w(a) = W(> 0), we obtain

B = −kr̄ − A lnr̄ and A ln(r̄/a) = −ηW − k(r̄ − a).

The latter gives

A/k = [−(η/k)W − (r̄ − a)]/ ln(r̄/a),

which implies that A < 0. Using r̄ = −A/k, we can reduce this to

x ln x = (η/ka)W + x − 1,

where x = r̄/a. This determines x (hence r̄) in terms of W, or W in terms of x. Writing W (x) in
the latter case, we find that

(η/ka)W′ (x) = ln x > 0.

Thus, x increases with increasing W, as expected, until W = W∗ , corresponding to x = b/a. For

W ≥ W∗ (corresponding to yielding of the entire annulus) we put r̄ = b in the above expressions
to determine A and B in terms of W.

7.17. We have

Trr,r + 1r (Trr − Tφφ + Trφ,φ ) = 0,

Trφ,r + 1r (2Trφ + Tφφ,φ ) = 0

and

(Trr − Tφφ )2 + 4T2rφ = 4k2 .

With a little effort you can rewrite the first pair of equations as

(rTrr ),r + Trφ,φ = Tφφ ,

Tφφ,φ + 1r (r2 Trφ ),r = 0. (*)

Substitute the first of these into the second to get

(rTrr ),rφ + Trφ + (rTrφ ),r + Trφ,φφ = 0.

If Trφ,r = 0 this simplifies to

(rTrr ),r φ + 2Trφ + Trφ,φφ = 0.

 Solutions to selected problems 189

Use the yield condition to write

q
rTrr = r(Trr − Tφφ ) + rTφφ = ±2r k2 − T2rφ + rTφφ .

We choose the positive root for definiteness. The solution for the negative root proceeds similarly.
With Trφ,r = 0 we get
q
(rTrr ),rφ = 2( k2 − T2rφ ),φ + (rTφφ,φ ),r ,

where, from (*), rTφφ,φ = −(r2 Trφ ),r = −2rTrφ . Then, (rTφφ,φ ),r = −2Trφ and the first of the
last three equations becomes
q
Trφ,φφ = −2( k2 − T2rφ ),φ .

Integrate once to get

q
Trφ,φ = −2 k2 − T2rφ + const.

One family of solutions can be obtained by taking the constant to be zero. The resulting
equation has the solution

Trφ = −k sin(c + 2φ),

where c is another constant. Using Tφφ,φ = −2Trφ (from (*)), we can integrate again to get Tφφ ,
and then use the yield condition to get Trr .
Note that another class of solutions, consistent with the assumption Trφ,r = 0, is Trφ = k (or
−k). In this case the slip lines are aligned with the radial and azimuthal axes. The yield condition
then requires Tφφ = Trr , and (*) gives

Tφφ = Trr = −2kφ + const.

This solution plays an important role in understanding the stress field near a corner of a rect-
angular block. See the interesting discussion on pp. 541 and 542 of Nadai’s book Theory of Flow
and Fracture of Solids.

7.19. The boundary conditions give

τk = Tn = per − k(t · er )k,

so that

τ/k = −er · t|r=a .

This makes sense because |τ/k| ≤ 1 and t is a unit vector.

190 Solutions to selected problems

To get an axisymmetric stress field, assume that

t = tr (r)er + tφ (r)eφ ,

with

t2r + t2φ = 1.

With a little work you can show that

gradt = t′r er ⊗ er + 1r tr eφ ⊗ eφ + t′φ eφ ⊗ er − 1r tφ er ⊗ eφ ,

and

divt = tr(gradt) = t′r + 1r tr .

Because divt = 0 and in view of the boundary condition, we then have

tr = −(a/r)τ/k

and
p
tφ = ± 1 − (a/r)2 (τ/k)2

for r ≥ a. Evidently tr → 0 and tφ → ±1 as r/a → ∞, so that t → ±eφ far away from the hole
boundary.
To characterize a velocity field consistent with this solution, recall that

s · gradw = 0,

where

s = k × t = tr e φ − tφ e r and gradw = (∂w/∂r)er + 1r (∂w/∂φ)eφ .

Then w(r, φ) satisfies the linear differential equation

t′r ∂w/∂φ + tφ ∂w/∂r = 0.

To be admissible, any solution to this equation must be such that λ ≥ 0, where

λτ̂ = 12 gradw.

With τ̂ = kt, this requires that t · gradw ≥ 0, i.e.,

rtr ∂w/∂r + tφ ∂w/∂φ ≥ 0.

 Solutions to selected problems 191

8.1. This follows immediately from the invariance of Du = (u j ej ),i ⊗ e i = u|ji ej ⊗ e i . We also
∗k k
have Γ̄ij = Γ̄ij because both connections are Levi-Civita and the starred and unstarred metrics
coincide.

8.12. Differentiate the result of Problem 8.1.

t −1 −1
9.5. We have Ṁ = Ġ G + Gt Ġ. Use Ġ = (ĠG )G to write this as Ṁ = 2Gt [Sym(ĠG )]G.
−1 −1 −1
If ĠG ∈ Sym, then Sym(ĠG ) = ĠG and the inverse relation is 2Ġ = G−t Ṁ. The flow
rule (7.8) for isotropic materials is thus easily seen to be equivalent to Ṁ = 2λGt (FŜ )G.
t
For von Mises’ yield function, Eq. (7.13) may be combined with JG Ŝ = GSG , where S is the
−1 √
Piola–Kirchhoff stress relative to κ, to reduce this to ṀM = μDev(MS), where μ = 2λ/ JM .
−1
Using J̇M = M∗ · Ṁ = JM tr(ṀM ), we conclude that J̇M = 0 and hence that this form of the
flow rule preserves plastic incompressibility.

9.6. No, the {ijk } are not tensorial and so the object {ijk }mk ⊗ m i ⊗ m j can be made to assume
arbitrary values simply by changing the coordinate system.

9.7. To the author’s knowledge this is an open problem.

9.8. With reference to Problem 9.5, allow the metric mij , and hence G, to vary with time while
holding its partial coordinate derivatives fixed at the point in question.

9.10. With w j = μij · ṁi we have

k k
μij · ṁi|j = μij · (ṁi,j − ṁk Γ̄ij ) = w,jj − μij,j · ṁi − μij · ṁk Γ̄ij
j k
= w|jj − w k Γ̄kj − μij,j · ṁi − μij · ṁk Γ̄ij
j i
= (μij · ṁi )|j − (μij,j + μik Γ̄kj + μkj Γ̄kj ) · ṁi .

9.13. We have N = Gt Wα . Because α → Rt αR, it follows, exactly as in the derivation of (6.28),

that Wα → Rt Wα R. From G → Rt G we then have that N → NR, and the definition of ηij in (9.84)
t
yields ηij → Rt ηij . From (5.43) we have CurlĠ = ϵ̄ijk ek ⊗ ṁ[j,i] . Then Ġ → R Ġ yields ṁj → Rt ṁj ,
ek → ek , and CurlĠ → ϵ̄ ek ⊗ R ṁ[j,i] = (CurlĠ)R. Finally, N · CurlĠ → NR · (CurlĠ)R =
ijk t

N·CurlĠ, this following from the easily derived identity AB · CD = ABDt ·C and the orthogonality
of R.
Index

affine geometry 34 Cartesian components 24 conical–polar coordinate

affine space 59 Piola-Kirchhoff stress 110 system 52
anholonomic g 59 replacing 56 connection see Levi-Civita
anti-plane shear 143–5 Cartesian coordinates 35–7, connection (Ricci’s
arclength parametrization 6 41, 43, 70, 127, 130, lemma); Weitzenböck
areal density of contact force 133, 147, 181 connection
t 8 geometer’s trick 70, 85 connection coefficients 64
asymptotically stable state 22 Cauchy stress 1, 9, 10, 15, conservative loads 21
axial pressure gradient 144 17–20, 73, 75, 101, constant state 138
azimuthal angle/axis 52, 138, 125–6 constitutive functions
142–3, 189 Cosserat continua, rotational non-dimensionalized 160
degrees of freedom 119 scalar-valued 161
balance laws, equation of deviatoric 143 see also strain-energy
motion 7, 111, 119, T 15, 71, 89, 100, 121, 123, function; yield function
123, 170 129, 152 constitutive hypotheses 100,
balance of linear momentum 8, vectors 73 101, 103
9 Cauchy theorem/equation of continuum mechanics 4–13
balance statement 113 motion 9 contravariant components,
Cauchy traction t 12, 22, 152 Einstein tensor 68
balloon, curvilinear coordinate
Cauchy–Born hypothesis, convected coordinates 64, 85,
grid 63
elastic deformation 109 148
Bauschinger effect 103, 159,
Cauchy–Green deformation formalism 65, 71, 150
168, 175
tensor 7, 19, 31, 65, 79, coordinates, bases, vectors, and
gradient plasticity 175
164 metrics 34–41
beta-line 138
Cayley–Hamilton formula 92 Cosserat continua, rotational
Bianchi identities 69 chain rule 6, 37, 43, 45, 99, degrees of freedom 119
Bingham’s model of viscoplas- 137 crack, straight, semi-infinite
ticity 126–9, 136, 156, change of reference 30, 161, traction-free 138
159 165 crystal/s
circular pipe flow 129 change-of-basis formulas 43 constitutive specification for
isotropic materials 126–9 channel flow, steady plastic spin 109
body rectilinear 127–8 crystal-elasticity theory 109
diameter of 76 circular pipe flow, Bingham’s cubic crystals 31, 109, 168
global stress-free model 129 fixed lattice 110
configuration of 76 classical theory for lattice structure of a metallic
body force 77 isotropic rigid-plastic crystal 2
boundary conditions 131, 140, materials 123–6 relative plastic slip
175, 189, 190 closed-contour integrals 57 of crystallographic
boundary point 106 co- and contravariant planes 2–3
boundary traction 75–6 components of v 39, 68 lattice vectors 109–10
boundary-value problems 13, coherent interface, rank-one local lattice distortions 75,
22, 111, 131, 136, 138, tensor 97 87, 109–10
140, 167, 175 compatibility condition 135, metallic 97
Burgers vector 77, 95, 96 165, 182 model for a uniform single
dislocation densities 93 metric-compatible 50–1, 80 crystal 87
of distribution 61 non-trivial 68, 69, 71 plasticity
 Index 193

conventional theory 109 convected-coordinate small,

rate of deformation in representation 64 approximations 151–3
single crystal 109 discontinuous fields 111–17 dissipation
reflection symmetries 92 polar decomposition 151 defined 23, 172, 178
simple shear rates/slips 109 and strain 63–6 discontinuities 111
slip-system vectors 109 deformations 5–6 of energy 3, 23
crystalline materials 74, 87 differentiability 15 Eshelby tensor 98, 116
Bauschinger effect 175 kinematic admissibility 7, 23 evolution of phase
crystal-elasticity theory 109 stable 22 boundaries 111
crystallinity vs and stress, in convected global dissipation
isotropy 109–10 coordinates 63–73 inequality 116
flow rule 109 see also Cauchy–Green inequality 101, 122
plane-strain slip-line deformation tensor; smooth/non-smooth
theory 133 small-deformation fields 114
scale-dependent theory insensitive to superposed
yielding 166–8 deviatoric Cauchy stress 143 rigid-body motions 101
symmetry 74–5, 158, deviatoric and spherical maximum dissipation
162–3, 168 tensors 108, 122 inequality 157
undistorted differentiable variations, plastic evolution and 87,
configuration 31 arbitrary invertible 79 97–100, 117
yield and flow 102–8 differentiable yield subject to inequality 114
see also gradient plasticity function 108 surface density of 115
theory differential equations volumetric dissipation
cubic crystal symmetry 168 density 115
partial 103, 131, 132, 137,
cubic solid, general yield function and 107
175
quadratic strain-energy distribution, Burgers vector
theory 57, 167
function 110 of 61
differential geometry 4, 34,
divergence theorem 9–10, 12,
curl 63, 79–82
25, 56, 112, 170–1
curlve 54–6 diffusion 8 conventional statement 112
defined 81 discontinuities for smooth vector fields 171
referential dislocation arithmetic mean of limiting Drucker–Prager function
density tensor 77 values 115 isotropic soils 126
CurlF 96 dissipation 111 Prandtl’s solution 134
current elastic range 166 discontinuous fields,
curvature, major symmetry 67 deformation eigenvalues 92–3
curvature tensor 58, 60, 83 gradient 111–17 Einstein, summation
integrability condition 80 dislocation densities 3, 34, 61, convention 35
vanishing 80 74–9, 93, 103, 162–4, Einstein tensor
curvilinear coordinate grid 63 168 contravariant
cyclic plastic flow 104 associated Burgers vector 93 components 68
cycloids, orthogonal elastic–plastic divergence-free 69
cycloids 136 deformation 74–9 Ricci tensor and 69
cylinder, in 6-D space 122 energetic effect of 175 elastic bodies, materially
cylindrical polar-coordinate referential surface 79, 94 uniform 87–93
system 52, 129 spatial surface 79, 95 elastic compliance tensor 28
surface dislocation and 87, elastic deformations 75, 101,
dead loading 22, 24 97 109, 110, 121, 123, 159
decomposition F 75 true 78, 81, 103, 161 Cauchy–Born
defects, energetic influence dislocations, Burgers vector of hypothesis 109
of 103 distribution 61 see also elastic–plastic
deformation fields, and velocity displacement field 147–51 deformations
fields 103 elliptic equilibrium elastic modulus tensors 27,
deformation gradient 6, 20, equations 131 90, 97
29, 30, 75 displacement elastic range 103
cofactor 66 gradients 149–51 convex set 106
194 Index

elastic strain 88, 90, 151 exterior unit normal 9 helices, helical forms 53
field 86 exterior/wedge product 56–7 hodograph transformation 137
incompatibility of 82–6 hole, traction-free circular
limited norm 103 flow rule 119–21, 123, 157–8, hole, axisymmetric state
rate-independent 175, 191 exterior to 139–40, 145
response 123 approximated to consistent hyperbolic system,
small elastic strain 100, 103 order 154 characteristic
stress–strain relation 104 Bingham’s model 159 curves 131
vanishing 123 evolution of plastic deforma- hyperelasticity 17, 27, 29, 72
yield function 108 tion 106, 121–4, 126, analysis of motion 21–2
elastic stretch 2, 15, 103 154
elastic–plastic deforma- integration 110, 160
image space dimension 97
tions 34, 74–86, 125, major simplification 120
147, 153, 160 non-zero value of CurlG 167 integrals, localization
theorem 8
differential-geometric plastic multiplier 167
considerations 79–82 theory of crystal invariance, strain energy 101,
dislocation density 74–9 plasticity 109–10 120
fundamental equation fourth-order compliance inverse function theorem 6, 36
of 125 tensor 58, 68, 90 isochoric motion 132
incompatibility of elastic frame-invariance 19, 28, 149, isochoric plastic flow 108
strain 82–6 150 ‘isotropic’ hardening 156–7,
elastic–plastic response Prandtl–Reuss theory 153–5 167
theory 153 full-rank tensors 97 isotropic soils, Drucker–Prager
elliptic equilibrium equations, function 126
displacement field 131 Geiringer equations 132–3, isotropy/isotropic materials 4,
energetic influence of 136, 139 31–2, 75, 92, 119–46,
defects 103 Cartesian form of the 154, 156, 160, 163–8,
energetic response functions equations 133–4 191
169–70close up geometer’s trick, Cartesian anti-plane shear 143–5
energy coordinates 70, 85 Bingham’s model of
stress, dissipation, and plastic geometry, differential– viscoplasticity 126–9
evolution 87–118 geometric classical theory for
total mechanical energy 97 considerations 79–82 isotropic rigid-plastic
total potential energy 21 global dissipation materials 123–6
energy density 114 inequality 116 crystallinity vs
entropy production global mass conservation isotropy 109–10
inequality 101 law 113 elastic–plastic
equation of motion gradf, scalar-valued function deformation 160
balance laws 7, 111, 119, f(y) 36 flow rule 119–21
123, 170 gradient plasticity theory 156, lattice vectors 110
leading-order 159, 168–75
plane strain of rigid-perfectly
approximation 151–3 Bauschinger effect 175
plastic materials:
weak/strong forms 12 see also crystalline materials
slip-line theory 129–43
equilibrium stress 130–2 gradw, anti-plane shear 143
scale-dependent
Eshelby tensor 98, 99, 100, grain boundaries 4
yielding 166–8
101, 116, 169 surface dislocation 96
strain-energy function 92–3,
Euclidean space 5, 7, 34, 54,
121
57, 59–60, 182 Hadamard’s lemma 96
geometry 81 see also Legendre–Hadamard stress–strain relation 92
reference configuration inequality von Mises’ yield
kappa 63 hardening function 121–3
occupation by deforming ‘isotropic’ hardening 156–7, vs crystallinity 109–10
bodies 67 167
Riemann curvature 83 rate-independent 160 kinematic admissibility 13,
tangent spaces 65 see also kinematic hardening; 170
vector field v, curlv 54 work hardening kinematic hardening 103
 Index 195

Kronecker delta 37, 42, 130 metals orthogonal logarithmic

Kuhn–Tucker optimization rate-sensitivity 126 spirals 143
problem 106 von Mises’ yield orthogonal straight lines 138
function 121–3 orthogonal subspaces,
Lagrange strain 23, 65, 150 see also crystalline materials symmetric and skew
Cauchy–Green deformation metric-compatibility tensors 159
tensor 7, 31 condition 50–1 orthogonal tensors 92, 108,
Lagrange-multiplier metric/reciprocal metric 119, 121, 159
tensor 123 components 37–9, orthonormal basis 127
Lagrangian description, 42–3, 46, 50, 54, 64–5, principal stress axes 130
motion 64 179 orthonormal sets 93
Lamé moduli 92 lowering/raising the index 40
Laplacian operators 47, 52, twice-differentiable
53, 175 function 83 paint, rheological model 126
lattice structure of a metallic modulus tensors 90, 97 parabolic profile 128
crystal, dislocations 3–4 motion parallel transport (of a vector
leading-order approxima- equation of motion 12, 113, field) 57
tion, equation of 152–3, 170 permutation symbol 35, 56
motion 151–3 fixed frame of reference 153 permutation tensor 84, 95–6,
Legendre–Hadamard Lagrangian description 64 182
inequality 25–8, 89, 90 phase equilibrium condition,
Levi-Civita connection (Ricci’s neighborhood stationary phase
lemma) 50–4, 60, 67–8, of a point 31, 32, 45, 58, 75, boundary 116
71, 80–1, 84, 164–5 76, 167, 178 phenomenology
major symmetry 67–8, 84 transformations 30 basic 1–4
linear elasticity theory 28, 147, neutral loading 158 theory of plasticity 122
150, 152 Newtonian fluid, constant Piola equations of motion,
frame invariance 149, 150 shear viscosity 126 convected-coordinate
load potential 21 Noll, theory of materi- formalism 65, 71
local algebraic restriction 10 ally uniform elastic Piola identity 71
‘local intermediate bodies 87–93 Piola traction/stress, and
configuration’ 74 Noll’s rule 31–2, 74 Cauchy traction t 12,
localization theorem 8, 153 symmetry group 74 151, 152, 168
non-crystalline materials 87 Piola–Kirchhoff stress 89, 100,
mass and momentum yield and flow 102–8 152, 168
balance of linear non-Newtonian fluid Cartesian components 110
momentum 8, 9 model 125
S 10, 11, 20, 24, 27, 123,
balance of moment of nonlinear elasticity 152
momentum 9 theory 15–33
symmetry 105
basic laws 7–8 normalization condition
W (1) = 0 90 Piola–Nanson formula 10, 66,
body force per unit mass 8
77, 78, 153
principle of conservation of
mass 8, 113 occupiability condition 5 plane strain
referential mass density 8 Orth and Orth+ 31, 92 further solutions 136
material curve 7 orthogonal rigid-perfectly plastic
material derivative 8 B 120 materials, slip-line
material surface 114 proper-orthogonal theory 129–43
material symmetry 29–32 tensors 31 slip-line theory 129–33, 144
restriction 168 Q(u) 18–19, 93–4f plastic deformation 2–4, 74–5,
transformations 29, 100 R 92, 107, 108, 119, 121, 77, 79, 81, 87–8, 93,
material velocity 8, 114 163, 191 97–8, 100–1, 103, 106,
mean-value theorem 104 S 106 109–10, 121–3, 156–7,
mechanical energy orthogonal curvilinear 160–1, 163–5, 168
balance 11, 12, 21 coordinates 136 evolution, flow rule 106
total 97 orthogonal cycloids 136 flow rule 122
metallic crystals 97 orthogonal decomposition 158 homogeneous 81
196 Index

plastic deformation (Continued) see also isotropy/isotropic rigid-plastic materials 123–6

inducing surface materials finite deformation
dislocation 97 pointwise inequality 116 theory 153
length–scale effects 160 polar factorization 153, 164 rigid-viscoplastic
local dissipation rate 98 potential energy, total 21 materials 126
tensor 79, 93 Prager see Drucker–Prager Rodrigues representation
see also flow rule; plastic flow function formula 32
plastic evolution Prandtl–Hencky rods and shells, 1-D and
asymptotically stable equations 131, 2-D 64
equilibria minimize total 133, 136 rotation field 101
potential energy 117 Prager’s version 134 rotations 91
causing dissipation 97–100 Prandtl–Reuss theory 153–5 Q(t) 88
K, constitutive equations 99 pullback 99 R 164
restriction 108 Pythagoras’ theorem 159
total potential energy 117
scalar-valued constitutive
plastic flow quasilinear partial differential
functions 161
absence 110 equations 137
consistency condition 158 scalar-valued function f(y) 36
cyclic 104 scalars
rank-one tensor, coherent
history 157 interface 97 derivatives and
isochoric 108 reciprocal-basis decom- connections 44–50
non-materiality of the position 46, 47, invariant 92
lattice 110 79 scalar fields 56, 105, 111
onset 174 reference configuration 5, 78, two-dimensional 144
texture and 119–20 102, 111 scale, material length scale 160
work hardening 122 arbitrariness 162 scale effects 160–1
yield stress 156, 157 choice 161 scale-dependent yielding
yielding and 102–8 kappa 63 crystalline materials 161,
see also plastic deformation; referential dislocation density 166–8
plasticity theory; tensor 77 isotropic materials 166–8
plastic incompressibility 172, referential and spatial mass second law of
191 densities 66 thermodynamics 98
plastic loading, inequality 158, referential surface, dislocation Serret–Frenet equations,
178 density 94 smooth space curves 53
plastic multiplier lambda 107 referential tangent basis, and shear
plastic spin reciprocals 64 anti-plane shear 143–5
material symmetry 123 relativity, Einstein’s summation constant shear viscosity 126
skew tensor field 106 convention 35 shear stress 130
suppress in flow rule 120, remodeling forces 172, 175 simple shear rates/slips 109
121 Reynolds’ transport yield stress 122, 156
transformation 107 theorem 111
shifter concept 27, 28, 90,
plastic stretch 2, 74 Ricci tensor 101, 123, 124, 147
plasticity theory 103 Einstein tensor and 69
misinterpretation 28
based on work second-order 68, 70, 165
simple state 138
inequality 106 symmetry 181
deformed bodies, Ricci’s lemma (Levi-Civita skew tensor 151
differential–geometric connection) 50–4, 60, skew tensor field, plastic
considerations 79–82 67–8, 71, 80–1, 84 spin 106
phenomenological theory Riemann curvature, Euclidean slip, no-slip condition 128
of 122 space 50–1, 83 slip-line field 139f
smooth fields 168 Riemann tensor 60, 67, 68, slip-line theory 129–43, 133
soils 126 85, 165, 181 slip-system decomposition 160
standard plasticity rigid-body motions, slip-system vectors 110
theory 168 superposed 100–2, 123 small-deformation
work hardening in plastic rigid-perfectly plastic materials, theory 147–55
flow 122 slip-line theory 129–43 displacement field 147–52
 Index 197

approximations for total strain energy 21 grain boundaries 96

small displacement W 101 induced by plastic
gradients 151–3 see also constitutive func- deformation 97
frame-invariant tions; stress and strain and stress relaxation 93–7
Prandtl–Reuss energy symmetric tensors 20, 92
theory 153–5 straining tensor 140, 143 spectral representation 93
smooth fields 55, 168 stress symmetry, in n dimensions 70
smooth space curves, dissipation, and plastic symmetry group 30, 74
Serret–Frenet evolution 87–118 transformations 30
equations 53 equation 93
smooth vector fields, equations of motion 71–3 tangent spaces 56, 64, 75, 147
divergence theorem 171 equilibrium 130–2 tangential velocity 133
soil plasticity, Drucker–Prager evaluation 92 Taylor expansion, strain-energy
function 126 residual stress 75, 83 function 90
solutions to selected statically determinate 131 telegraph equations 137
problems 177–91 stress power 11, 21, 170–5 tensor field 9, 48, 55
spatial dislocation density stress relaxation, and surface skew 100
tensor 77 dislocations 93–7 smooth 55
spatial equations of motion 10 stress field 137, 143, 189 tensors
spatial identity 99 axisymmetric 143, 190 9-dimensional
spatial and referential disloca- bounded 139, 141 second-order 41–3
tion densities 79, 94, dislocations and 83 analysis in 3-D space 34–44
95 equilibrium 134 Cauchy–Green deformation
spectral representation, flow rule 121 tensor 7, 19, 31, 65, 79,
symmetric tensors 93 Geiringer equations 132 164
stable deformations 22 partial differential contortion 81
stable state, asymptotic 22 equations 131 coordinate invariance 43
stationary phase boundary, residual stress field 75, 83 curvature 58, 60, 83
phase equilibrium straight, semi-infinite derivatives and
condition 116 traction-free crack 138 connections 44–50
steady rectilinear channel and velocity field 132 deviatoric and spherical 108,
flow 127–8 stress and strain energy 15–20 122
Stokes’ formula 77 classical linear stress–strain Einstein tensor 68, 69
Stokes’ theorem 54–9, 60, 93 relation 28 elastic compliance 28
strain, Lagrange strain 65 conservative problems and elastic modulus 27, 90, 97
strain compatibility 66–71 potential energy 21–5 Eshelby tensor 98, 99, 100,
conditions 67 isotropic materials 92 101, 116, 169
strain hardening 1–2f, 103 major symmetry Euclidean inner product 4
rate sensitivity, and gradient condition 24 first-order 41
plasticity 156–76 response of a bar 2f fourth-order compliance 58,
strain-displacement strain energy per unit current 68, 90
relation 150 volume 21 full-rank 97
strain-energy function 87, 90, see also strain-energy image space dimension 97
97, 110, 120 function; yield stress inner product 11
conventional non-uniform summation convention, Lagrange-multiplier
elastic solid 88 Einstein 35 tensor 123
invariance 101, 120 superposed rigid-body Levi-Civita connection
isotropic materials 92–3, motions 100–2, 123, (Ricci’s lemma) 50–4,
121, 163 151 60, 67–8, 71, 80–1, 84
low-order Taylor deformation 148–9 modulus 90, 97
expansion 90 superposition, rate of plastic orthogonal 92, 108, 119,
per unit current volume 21 deformation in single 121, 159
per unit volume of reference crystal 109 permutation 84, 95–6, 182
configuration kappa 88, surface area, orientation plastic deformation 79, 93
168 and 66 plastic spin skew 106
Taylor expansion 90 surface dislocations 96 product of 3-vectors 4
198 Index

tensors (Continued) triangle inequality 153 Bingham’s model 126–9,

rank-one tensor at a coherent true dislocation density 78 136
interface 97 two-point tensor 64 rigid-viscoplastic
referential dislocation materials 126
density 77 unloading 158 viscous fluids model 126
Ricci tensor von Mises’ yield func-
Einstein tensor and 69 vector field 9, 12, 25, 46, 49, tion 121–3, 124,
second-order 68, 70, 165 52, 58, 130 159
Riemann curvature 60, 67, defining 35
68, 85, 165, 181 divergence theorem 171 wedge product 56–7
second-order 41–3, 68, 70, parallel transport of 57 skew symmetry 58
165 smooth 55 Weitzenböck connec-
v tion 59–61,
skew 106, 151
co- and contravariant 80
spatial dislocation density 77 symmetry
components 39
straining 140, 143 curlv 54 transformations 163
symmetric 20, 92, 93 vector space 75, 79 torsion 162
third-order 54, 59, 70, 161, symmetric tensors 106 work hardening 122
170, 180 vectors work inequality 16, 20, 166,
torsion tensor constraints 86 9-dimensional, as 167
transformation rules 43, 70 second-order tensors 41 plasticity theory 106
two-dimensional complex-valued vector
permutation 95–6 fields 25 X-ray diffraction 109–10
two-point 64 as first-order tensors 41
theory of ordinary differential slip-system vectors 110 yield condition 186
equations 57 ‘velocity’, point on a curve with yield criterion F = 0 130
third-order tensors 54, 59, 70, parametrization 60 yield function 102
161, 170, 180 velocity discontinuities 133 differentiability 108
tilt and twist boundaries 97 velocity field/s elastic strain 108
torsion anti-plane shear 143 and flow, crystalline/non-
covariant derivative of and deformation fields 103 crystalline
metric 80 and stress field 132 materials 102–8
non-zero 83, 86 virtual velocity field, material symmetry
displacement 12, 13, condition 119–20
and Weitzenböck con-
170 restriction 107
nection 59–61,
velocity gradient 124, 127, subject to material symmetry
162–3
132, 141 requirements 106
torsion tensor, constraints 86 see also constitutive functions;
L = gradv 124, 132
traction 8 virtual field 172 flow rule
traction formula 9 virtual power statement 172 yield stress
traction-free circular hole, virtual stress power 170 k 186
axisymmetric state virtual velocity plastic flow 156, 157
exterior to 139–40, 145 and acceleration 24 in shear 122, 156
transformation formulas, equilibrium 170 stress and strain energy 1–2
F H K, superposed virtual velocity field yield surface
rigid-body motions 102 displacement 12, 13, 170 cylinder in 6-D space 122
translation space 7 kinematically admissible 13, kinematic hardening 103
transport theorem see 170 yielding and plastic flow 102–8
Reynolds’ transport viscoelasticity, model, Piola Young’s modulus 1, 2
theorem stress 23
Tresca yield function 122 viscoplasticity 103 zero-traction condition 138