(9780198567783) David J. Steigmann - Finite Elasticity Theory

PREFACE
This book is based on notes developed for a one-semester course offered at Berkeley.
Typically, this serves graduate engineering students studying Mechanics, but also occasion-
ally attracts interest on the part of students studying Mathematics and Physics. For this
reason, and to suit my own predilections, the level of mathematical rigor is appropriate for
readers possessing a relatively modest background. This has the pedagogical advantage of
allowing time to make contact with physical phenomena, while providing context for such
mathematical concepts as are needed to support their modeling and analysis. Advanced
readers seeking more than this should consult the books by Antman (2005), Ciarlet (1998),
and Silhavy (1997), for example. My expectation, and part of the motivation for this work,
is that books and treatises of the latter kind may be more fully appreciated by students after
reading an introductory course.
Throughout the book, we focus on the purely mechanical theory. However, extensive
reference will be made of the notions of work, energy, and, in the final chapter, dissipation.
The emphasis here is on developing a framework for the phenomenological theory.
Despite what contemporary students are often taught, such theories remain the best hope
for the quantitative study of physical phenomena occurring on human (macroscopic) scales
of length and time. This is perhaps best illustrated by our own subject, which developed
rapidly after the introduction ofa clear and concise framework for phenomenological mod-
eling. Thus, researchers began to exploit the predictive potential of the theory of nonlinear
elasticity only after constitutive relations derived from statistical mechanics were largely
abandoned in favor of those of phenomenological origin, which could be fitted to actual
data. In turn, nonlinear elasticity, because of its secure logical, physical, and mathemat-
ical foundations, has served as a template for the development of theories of inelasticity,
continuum electrodynamics, structural mechanics, thermodynamics, diffusion, rheology,
biophysics, growth mechanics, and so on. The final chapter, consisting of a brief introduc-
tion to plasticity theory, illustrates how elasticity interacts with and informs other branches
of solid mechanics. In short, the study of nonlinear elasticity is fundamental to the under-
standing of those aspects of modern mechanics research that are of greatest interest and
relevance.
These notes are mainly about the conceptual foundations of nonlinear elasticity and the
formulation of problems, occasionally including a worked-out solution. The latter are quite
rare, due the nonlinearity of the equations to be solved, and so recourse must usually be
made to numerical methods, which, however, lie outside the scope of this book. Explicit
solutions are of great importance, however, because they offer a means of establishing a dir-
ect correlation between theory and experiment, and thus extracting definitive information
about the constitutive equations underpinning the theory for use in computations.
viii | PREFACE
Although elasticity theory is inherently nonlinear, courses on the purely linear theory,
treating the equations obtained by formally linearizing the general theory, are quite preva-
lent. This is due to the great utility of the linear theory in solving problems that arise in
engineering and physics. To a large degree, and mainly for historical reasons, such courses
are delivered independently of courses of the present kind. The explanation for this schism
is that the nonlinear theory did not come into its own until the latter half of the last cen-
tury and, by then, the linear theory had matured into a major discipline in its own right,
on par with classical fluid dynamics, heat transfer, and other branches of the applied and
engineering sciences. This fueled research on applications of the theory relying on and,
in turn, advancing techniques for treating elliptic linear partial differential equations. The
word Finite in the title refers to the possibly large deformations covered by the nonlin-
ear theory, as distinct from the infinitesimal deformations to which the linear theory is
limited. Elasticity theory is, nevertheless, nonlinear and the use of the linear approximation
to it should always be justified, in the circumstances at hand, by checking its predictions
against the assumptions made in the course of obtaining the equations. However, this is
inconvenient and, thus, almost never done in practice.
Unfortunately, all this is somewhat disquieting from the standpoint of contemporary stu-
dents, who must grapple not only with the question of whether or not a problem may be
modeled using elasticity theory, but may also feel obliged to categorize it as either linear or
nonlinear at the outset. Those more interested in concepts and in the formulation of new
theories of the kind mentioned above will derive much value from an understanding of non-
linear elasticity, whereas my view is that linear elasticity has virtually nothing to offer in this
regard, due to the severe restrictions underpinning its foundations.
The book collects what I think students should know about the subject before embarking
on research, including my interpretations of modern works that have aided me in refin-
ing my own understanding. Those seeking to grasp how and why materials work the way
they do may be disappointed. For them I recommend Gordon (1968, 1978) as an en-
gaging source of knowledge that should ideally be acquired, but which rarely, if ever, is,
before reading any textbook on the mechanics of materials. In particular, these may be read
in lieu of an undergraduate course on Strength of Materials, which is to be avoided at all
costs. If the present book comes to be regarded as a worthy supplement to, say, Ogden’s
modern classic Nonlinear Elastic Deformations (1997), then I will regard the writing of it
to have been worthwhile. Readers having a grasp of continuum mechanics, say at the level
of Chadwick's pocketbook Continuum Mechanics: Concise Theory and Problems (1976), will
have no trouble getting started. Reference should be made to that excellent text for any con-
cept encountered here that may be unfamiliar. The reader is cautioned that current fashion
in continuum mechanics is to rely largely on direct notation. Indeed, while this invariably
serves the interests of clarity when discussing the conceptual foundations of the subject,
there are circumstances that call for the use of Cartesian index notation, and we shall avail
ourselves of it when doing so proves to be helpful. We adopt the usual summation con-
vention for repeated subscripts together with the rule that subscripts preceded by commas
always indicate partial differentiation with respect to the Cartesian coordinates. Direct no-
tation is really only useful to the extent that it so closely resembles Cartesian index notation,
PREFACE | ix
while the latter, being operational in nature, is invariably the setting of choice for carrying
out the more involved calculations.
Some topics are given more attention than others, in accordance with my personal views
about their relative importance and the extent to which they are adequately covered, or
more often not, in the textbook and monograph literatures. My intention to use these notes
in my future teaching of the material leads me to buck the current trend and not include
answers to the exercises. The latter are sprinkled throughout the text, and an honest at-
tempt to solve them constitutes an integral part of the course. The book is definitely not
self-contained. Readers are presumed to have been exposed to a first course on continuum
mechanics, and the standard results that are always taught in such a course are frequently in-
voked without derivation. In particular, readers are expected to have a working knowledge
of tensor analysis in Euclidean three-space and the reason why tensors are used in the for-
mulation of physical theories—roughly, to ensure that the predictions of such theories are
not dependent on the manner in which we coordinatize space for our own convenience.
The contemporary books by Liu (2002), and by Gurtin, Fried and Anand (2010) can
be heartily recommended as a point of departure for those wishing to understand the
foundations for modern applications of continuum mechanics. A vast amount of important
material is also contained in Truesdell and Noll’s Nonlinear Field Theories of Mechanics
(1965) and Rivlin’s Collected Works (Barenblatt and Joseph, 1997), which should be read
by anyone seeking a firm understanding of nonlinear elasticity and continuum theory in
general.
David Steigmann, Berkeley 2016
REFERENCES
Antman, S.S. (2005). Nonlinear Problems of Elasticity. Springer, Berlin.

Barenblatt, G.I. and Joseph, D.D. (Eds) (1997). Collected Papers of R.S. Rivlin, Vols. 1 and 2. Springer,
NY.
Chadwick, P. (1976). Continuum Mechanics: Concise Theory and Problems. Dover, NY.
Ciarlet, P.G. (1988). Mathematical Elasticity. North-Holland, Amsterdam.
Gordon, J.E. (1968), The New Science of Strong Materials, or Why You Don't Fall Through the Floor.
Princeton University Press, Princeton, NJ.
Gordon, J.E. (1978). Structures, or Why Things Don’t Fall Down. Penguin, London.
Gurtin, M.E., Fried, E. and Anand, L. (2010). The Mechanics and Thermodynamics of Continua.
Cambridge University Press, Cambridge.
I-Shih Liu (2002). Continuum Mechanics. Springer, Berlin.
Ogden, R.W. (1997). Non-Linear Elastic Deformations. Dover, NY.
Silhavy, M. (1997). The Mechanics and Thermodynamics of Continuous Media. Springer, Berlin.
Truesdell, C. and Noll, W. (1965). The nonlinear field theories of mechanics. In: S. Fligge (Ed.),
Handbuch der Physik HI/1. Springer, Berlin.
CONTENTS
Concept of an elastic material 6.1... 6... cece eee ee ees

Observers and invariance .........
cc cece ee cveccenervcvees
Mechanical power and hyperelasticity... 0.0.0.0...

0.0 e eee
3.1 Elasticity and energy
3.2. Work inequality
Material symmetry... 0... ee eee eee ee eee eee
4.1 Stress response
4.2. Strain energy
43 Isotropy
Fiber symmetry 20... 0. ccc ccc eee eee eee eee tenn eees
Stress response in the presence of local constraints
onthe deformation...... 0.0.00. cence cnecacseeessevess
6.1 Local constraints
6.2 Constraint manifolds and the Lagrange multiplier rule
6.3 Material symmetry in the presence of constraints
Some boundary-value problems for uniform isotropic
incompressible materials... 0.0... 0... c ccc eee ene eens
7.1 Problems exhibiting radial symmetry with respect to a fixed axis
7.1.1 Pressurized cylinder
7.1.2 Azimuthal shear
7.1.3 Torsion ofa solid circular cylinder
7.14 Combined extension and torsion
7.2. Problems exhibiting radial symmetry with respect to a fixed poin
7.2.1 Integration of the equation
7.2.2 Pressurized shells, cavitation
Some examples involving uniform, compressible isotropic
Materials... 0. cc cc ce eee eee ete teen ences
8.1 Spherical symmetry, revisited
8.2. Plane strain
8.3 Radial expansion/compaction
Material stability, strong ellipticity and smoothness of
equilibria... 0... ccc eee ee ee eens
9.1 Small motions superposed on finitely deformed equilibrium states
9.2 Smoothness of equilibria
9.3 Incompressibility
xii | CONTENTS
10 Membrane theory... . 0... ccc eee eee eee eee e eens

10.1 General theory
10.2 Pressurized membranes
10.3 Uniqueness of the director 100
10.4 Isotropic materials 102
10.8 Axially symmetric deformations of a cylindrical membrane 104
10.6 Bulging of a cylinder 108
11 Stability and the energy criterion ©... . 0... cee cee eee eens 113
11.1 The energy norm 113
11.2 Instability 116
11.3 Quasiconvexity 121
11.4 Ordinary convexity 123
11.4.1 Objections to ordinary convexity 123
11.5 Polyconvexity 126
11.6 Rank-one convexity 128
11.7 Equilibria with discontinuous deformation gradients 131
11.8 The Maxwell—Eshelby relation 132
11.8.1 Example: alternating simple shear 133
12 Linearized theory, the second variation and bifurcation of
equilibria... ce ec eee eee ee ee eee eee ee enes 137
13 Elements of plasticity theory .. 1.0... cece eee eee eee eens 142
13.1 Elastic and plastic deformations 143
13.2 Constitutive response 146
13.3 Energy and dissipation 148
13.4 Invariance 152
13.$ Yielding, the work inequality and plastic flow 183
13.6 Isotropy 156
13.7 Rigid-plastic materials 160
13.8 Plane strain of rigid-perfectly plastic materials: slip-line theory 161
13.8.1 State of stress, equilibrium 161
13.8.2 Velocity field 163
Supplemental notes 0.0... 0... eee ee eee ee eee eee eee e eee
1 The cofactor 167
Gradients of scalar-valued functions of tensors 168
AWN
Chain rule 169

Gradients of the principal invariants of a symmetric tensor 169
Relations among gradients 171
NAA
Extensions 172
Korn’s inequality 174
ON
Poincaré’s inequality 174
Index 177
Concept of an elastic material
One would think this would be the easiest chapter to write, but alas such is not the case.
Thus, we will have to settle for the present, rather superficial substitute, which may be
skipped over by anyone—and thus presumably everyone reading this book—who has some
passing acquaintance with the concept of elasticity. When attempting to define the property
we call elasticity, and how to recognize it when we see it, we encounter certain non-trivial
obstacles, not least among these being the fact that elastic materials per se do not actually
exist. That is, there are no known examples of materials whose responses to stimuli conform
to conventional notions of elasticity in all circumstances. In fact, even the concept of ideal
elastic response is open to a wide range of interpretations. Rather than delve into the under-
lying philosophical questions, for which I am not qualified, I defer to the thought-provoking
account contained in a contemporary article by Rajagopal (2011).
For our purposes, the idea of elasticity may by abstracted from the simplest observations
concerning the extension of a rubber band, say, to a certain length. Naturally, one finds that
a force is required to do so and, if the band is left alone for a period of time, that this force
typically settles to a more-or-less fixed value that depends on the length. This is not to say
that the force remains at that value indefinitely, but often there is a substantial interval of
time, encompassing the typical human attention span, during which it does. More often,
one fixes the force,f say, by hanging a weight of known amount from one end; the length
of the band adjusts accordingly, reaching a corresponding value that is sensibly fixed over
some time interval. If one has a graph of force vs. length, then usually one can read off the
force corresponding to a given length and vice versa. The situation for a typical rubber band
is shown in Figure 1.1, where the abscissa is scaled by the original (unforced) length of the
band. This scaling, denoted by A, is called the stretch.
If one looks closely one may observe a slight hysteresis on this graph. This is due to small-
scale defects or irregularities among the long-chain molecules of which the rubber is made.
They have the effect of impeding attainment of the optimal or energy minimizing state of
the material under load, and are usually reduced to the point of being negligible by sub-
jecting the rubber to a cyclic strain, which effectively “works the kinks out.” This is known
as the Mullins effect. Studies of it in the mechanics literature are confined mostly to its
description and prediction, based on phenomenological theory (see Ogden’s paper, 2004),
rather than its explanation. A notable exception is the book by Miiller and Strehlow (2010),
Finite Elasticity Theory. DavidJ. Steigmann.

© DavidJ. Steigmann, 2017. Published 2017 by Oxford University Press.
2 | CONCEPT OF AN ELASTIC MATERIAL
fA
a
y al
a= 1/1,
Figure 1.1 Uniaxial force-extension relation for rubber. Stiffening is due to straightening of long-
chain molecules
which offers an interesting explanation in terms of microstructural instabilities and associ-

ated thermodynamics. For the most part these subjects will not be covered, although we will
devote considerable attention later to the notion of stability and its connection to energy
minimization.
Ignoring hysteresis, then, we can expect to extract a relation of the form f = F(A) from
a graph of the data. Here, F is a constitutive function; i-e., a function that codifies the nature
of the material in terms of its response to deformation. We are justified in attributing the
function to material properties—and not just the nature of the experiment—provided that
the material is uniform and no other forces are acting, In this case, equilibrium consider-
ations yield the conclusion that the forces acting at the ends of an arbitrary segment of the
band are opposed in direction, but equal in magnitude, the common value of the latter be-
ing given by the force f. If the stretch, which is really a function defined pointwise, is also
uniform, then it can be correlated with the present value of the length of the band. That
is, the stretch is really a local property of the deformation function describing the config-
urations of the band, and may be correlated with the end-to-end length provided that it is
uniformly distributed.
Because the length of a segment is arbitrary in principle, we may pass to the limit and
associate the response with a point of the material, defined as the limit of a sequence of
intervals whose lengths tend to zero. In this way, we associate the global force-extension
response with properties of the material per se, presumed to be operative on an arbitrarily
small length scale. This is one of the premises of continuum theory; namely, that the prop-
erties of the material are assigned to points of the continuum. These days, thanks to the rise
of computing, it is often augmented by the notion of a hierarchy of continua that operate at
length scales smaller than the unaided eye would associate with a point. In some cases, the
smaller scale continua are replaced by discrete or finite-dimensional, systems. Some form of
CONCEPT OF AN ELASTIC MATERIAL | 3
communication among the length scales is then required, furnishing material properties on
the larger scales in terms of system response on the smaller scales. The basic idea is known
collectively as the multi-scale method. While it is currently an active field producing inter-
esting and sometimes useful work, it is not a panacea for the limitations of conventional
continuum theory, but rather a way of exploiting computing power to avoid the empirical
work that standard continuum theory requires to realize its full potential. Recalling the du-
bious effectiveness of early formulations of rubber elasticity based on statistical mechanics,
it is perhaps not surprising that multiscale methods are typically no more reliable than basic
continuum theory, while often requiring the use of ever more models operating at ever finer
length scales. All is well and good if this process converges in some sense, but whether or
not it does depends as much on the problem being addressed as anything else, and in any
case the issue is almost never explored carefully. This brings to mind Truesdell’s (1984)
amusing observation to the effect that continuum theory is immunized by its very nature
against the next great discovery in atomic physics, remaining indifferent to the parade of
sub- and sub-sub-atomic particles that blink in and out of existence while we labor over our
engineering calculations, oblivious to their comings and goings. We digress, however.
If the material is non-uniform, little can be concluded from the simple rubber-band ex-
periment about the nature of the material, in contrast to the situation for uniform materials
that are uniformly stretched. This is due to the fact that the stretch will now be non-uniform,
despite the uniformity of the force intensity (by virtue of equilibrium), due to the variations
in the way the material responds locally to that force. In this case, we perform a sequence
of experiments on ever shorter segments. If the force-deflection curves thus obtained con-
verge, then the limit response may be said to characterize the material at the length scale
associated with the last segment in the sequence. This generates the responsef = F(A; x),
where x is the location from one end, say, of the band to a point contained in the inter-
section of the sequence of the segments prior to deformation. Here the stretch is now a
function of x, and the force required to maintain it at the value A(x) will reflect the non-
uniformity of the material; hence, the explicit dependence of F on x. Because the value of
this function—the force—is uniform in the equilibrated band, it is the stretch that must ad-
just to the non-uniformity of the material, producing a field .(x). If the sequence of tests
does not converge, then we assign the response F(A; x) to the one point x that remains as the
segment lengths diminish to zero. In principle, if not in practice, this is the sort of thing one
does to test for material non-uniformity and to quantify the associated response function.
The attendant difficulties may indeed give some impetus to multiscale methods, but it
bears repeating that these entail the use of models in lieu of actual data. Such practices are
fraught with their own difficulties, not least among them being the need for empirical testing
of the models purportedly operating on the smaller scales. In any case, as for uniform bands,
we have a relation for the force that depends on the present values of the stretch function,
but not on the history of the deformation, or on how quickly or slowly it occurs. This is due
to our restriction to time scales on which the force and stretch are sensibly constant, and is
what most people mean by the word elasticity.
Interestingly, a relation of much the same kind is found when the band is deformed very
rapidly, as when a wave is caused to propagate through it. This is due to the fact that the
deformation then occurs on a time scale that is too short for effects like viscosity to be
effective in relaxing the force, whereas in the slow experiment in which data are recorded
on long time scales, such effects have already run their course. Behavior on intermediate
time scales, which is the province of viscoelasticity theory, is beyond the scope of this book.
An up-to-date account, accessible to students of nonlinear elasticity, may be found in the
article by Wineman (2009).
A principal lesson of the rubber-band experiment is that material response appears to be
local. That is, the force acting on a short segment of the band, and hence (by equilibrium)
the force at any cross section within it, is determined by the deformation of that segment
and not by the deformations of other segments comprising the original band. This idea has
been codified by Noll in his principle of local action. Current pedagogy tends to discour-
age the use of this kind of language, as it seems to confer special status on simple ideas.
Nevertheless, this principle furnishes a logical point of departure for the abstraction of sim-
ple experiments of the foregoing kind. It exemplifies the kind of fundamental reasoning that
is needed to move from observations to a conception of how Nature works, and from there
to the formulation of a predictive theory, which is surely among the noblest aspirations of
Man. In our case, this takes the form of an assumption to the effect that the Cauchy stress
at a material point, labelled p, say, is sensitive only to the deformation of material points in
its vicinity (see Figure 1.2).
More precisely, the Cauchy stress T(p, t) is determined by the deformation x (x, t) for
those reference positions x satisfyingx € N,(p), a neighborhood of the material pointp of
the body occupying position x in reference configuration x. Here y = x (x, £) is the position
in three-space at time ¢ associated with the same material point. In the older literature one
often sees the word “particles” used in place of our “points.” However, the former connotes
a collection of discrete objects, which is not what is intended when using continuum theory.
Figure 1.2. Neighborhood of a material point, p, in the reference configuration

CONCEPT OF AN ELASTIC MATERIAL | 5
It is appropriate to append a subscript to the deformation function and write y = x,(x, 6),
to acknowledge the choice of reference. The fact that current position is unrelated to the
reference implies that the function taking x to y necessarily depends on this choice. This
issue will be revisited later. The Cauchy stress is measurable without reference to «, both in
principle and in practice, and so a subscript would not be appropriate.
In summary, we suppose that T(p, t) is determined by x, (x,t) for x’ € N,(p), which
contains the position x ofp ink.
If the deformation function is smooth in its first argument, and if the diameter of N, (p)
is suitably small, then the deformations that determine T(p, t) may be approximated by
Xe (x, t) = Xe (x, t) + E, (x, t)(x’ - x) + o( |x’ -x|), (1.1)
where F,, called the deformation gradient, is the derivative of the function x,(x, ¢) with re-
spect to its first argument. This should carry the subscript in principle, but in practice it is
cumbersome to do so and for the most part we shall not. The small “oh” identifies terms
that tend to zero faster than the argument does, as the latter tends to zero. For points suffi-
ciently close to the place x occupied by p in the reference, the response is then determined
primarily by x,(x,£) and F,(x,¢). If one is concerned with leading-order effects, then it
would be sensible to retain only the first term in (eqn 1.1) and consider a model in which
the stress at p is sensitive only to x, (x,t). However, we will see that such dependence is
precluded by invariance arguments and so the actual leading term is the deformation gra-
dient. Retention of this term alone leads to a famous model for materials named the simple
material by Noll, who advanced the idea not only for elasticity, but for other theories in
continuum mechanics as well.
One can, of course, envisage applications in which retention of further terms is appro-
priate, the next one being the gradient of F, (x, f). The model thus derived turns out to be
rather useful for describing localized effects such as surface tension in solids. More recently,
it has been used to model materials reinforced by a dense distribution of fibers in which the
fibers are presumed to offer elastic resistance to flexure. Flexure is nothing more than the
curvature induced by deformation, while curvature is determined by the second derivative
of the position function on a fiber with respect to arc length; this in turn is determined by
both the deformation gradient and its gradient. Having simpler applications in mind, we
do not study this relatively complex model here. The interested reader will find excellent
treatments of it in papers by Toupin (1962, 1964) and by Spencer and Soldatos (2007).
The model we intend to study is thus of the form
Tip, t) = Ge Xu (x, t), F, (x, t);x), (1.2)
in which the last argument is intended to indicate a parametric dependence on the mater-
ial point and, hence, on its position in «. This is needed if the properties of the material,
codified in the constitutive function G, (-, -;x), vary from point to point; that is, if the mater-
ial is non-uniform. In principle, this function is determined by experiments, but these are
cumbersome and expensive, and so before going to the laboratory we should try to sim-
plify it as far as possible. The manner of doing just that comprises much of the theoretical
underpinnings of the subject. We might have included a dependence on time, but as we

shall see presently this too is disallowed by invariance arguments. We note, however, that
the subscript « on the constitutive function is required, because this function depends on
variables associated with a reference placement whereas its values do not; the function itself
must then depend on the reference. The reference is not prescribed for us, but instead is
specified by us, subject to the requirement that positions within it be in one-to-one cor-
respondence with points of the material. Almost always workers in the subject confine
attention to references that are, or could be, occupied by the material in the course e of its
motion, this carrying the mild restriction
J >0, where J = detF,. (1.3)
Changing the reference means changing the constitutive function in such a way as to
leave the Cauchy stress invariant. After all, experiments designed to measure the Cauchy
stress do not require knowledge of our idiosyncratic choice of reference. In this way, given
the constitutive function based on a particular choice of reference, we can compute that
which applies to any other admissible choice. The reader is cautioned that long-standing
practice is to associate the reference with a stress-free configuration of the material. Not
only does this promote the erroneous view that the reference needs to have some special
physical status, it also demands that we accept the fiction that the existence of global stress-
free states is the norm, rather than the exception. We will take up this issue in Chapter 13.
REFERENCES
Miller, I, and Strehlow, P. (2010). Rubber and Rubber Balloons: Paradigms of Thermodynamics.
Springer, Berlin.
Noll, W. (1958). A mathematical theory of the mechanical behavior of continuous media. Arch.
Ration. Mech. Anal. 2, 197-226.
Ogden, R.W. (2004). Elasticity and inelasticity of rubber. In: G. Saccomandi and R. W. Ogden (Eds)
Mechanics and Thermomechanics of Rubberlike Solids, pp. 135-185. CISM Courses and Lectures
No. 452. Springer, Wien.
Rajagopal, K. (2011). Conspectus of concepts of elasticity. Math. Mech. Solids 16, 536-562.
Spencer, AJ.M., and Soldatos, K.P. (2007). Finite deformations of fibre-reinforced elastic solids with
fibre bending stiffness. Int. J. Non-linear Mech. 42, 355-368.
Toupin, R.A. (1962). Elastic materials with couple-stresses. Arch. Ration. Mech. Anal. 11, 385-414.
Toupin, R.A. (1964). Theories of elasticity with couple-stress. Arch. Ration. Mech, Anal. 17, 85-112.
Truesdell, C. (1984). An Idiot’s Fugitive Essays on Science: Methods, Criticism, Training, Circumstances.
Springer, Berlin.
Wineman, A.S. (2009). Nonlinear viscoelastic solids—a review. Math. Mech. Solids 14, 300-366.
Observers and invariance
The historical development of Physics has been guided by one overarching idea: that the
laws of Nature have nothing to do with us and, hence, that the mathematical descrip-
tion of these laws should satisfy invariance requirements representing such indifference
in mathematical terms. This egalitarian, as distinct from egocentric philosophy marks the
development of Physics just as surely as it characterizes the healthy psychological develop-
ment of human beings. Alas, as with all that seems obvious, it must face certain challenges; in
this instance that offered by the uncertainty principle of quantum mechanics, which teaches
that the act of making an observation has a non-negligible effect on that which is observed.
(See the book by Murdoch (2012) for an interesting discussion.) We shall not, however, de-
velop elasticity theory from the quantum mechanical point of view here, despite promising
developments emerging from current research.
To understand the consequences of the idea of material indifference for elasticity the-
ory, it is necessary to admit different points of view, or observers, into contention so that
we may know what it is about them that should not influence a sound physical theory. For
example, in Relativity Theory an observer is identified with a frame of reference. Observers
have little in common except for their agreement on one thing—the speed of light in vac-
uum. Accordingly, the speed of light in vacuum is said to be frame invariant. This seemingly
innocuous constraint on the laws of physics has the most profound mathematical con-
sequences, known collectively as the Theory of Relativity. Classical Mechanics, to which
attention is confined here and in most treatments of continuum mechanics, is based on a
similar idea, except that classical observers are presumed to agree on two things—the dis-
tance between any pair of material points and the time lapse between events. A penetrating
discussion may be found in a paper by Noll (1973).
This is not all, however. Following an important paper by Murdoch (2003), we suppose
that observers also agree on the nature of the material. In our case, that it is elastic, and
hence on the list of variables (e.g., the Cauchy stress, the deformation, and the deformation
gradient in the case of elasticity) that are related by the constitutive equations pertaining
to any observer. After all, the manner in which a sample of material responds to stimuli is
presumably unaffected by the observer of such response; and so, if a particular set of vari-
ables is found by one observer conducting an experiment to be relevant then it should be so
for all. We are belaboring this matter perhaps more than we should, because as reasonable

8 | OBSERVERS AND INVARIANCE
as the concept may appear to the uninitiated, it has been the cause of considerable confu-
sion and suffering among the educated. My own not inconsiderable reading leads me to the
view that the interpretation offered by Noll and Murdoch is superior to the alternatives as
far as classical mechanics is concerned. To this day, workers are often divided over this issue
along the party lines that have emerged during the modern development of our subject.
The relationship between a pair of classical observers, O and O*, say, may be expressed
in the form (see Figure 2.1)
xi, 6) = Q(t)x,(%t)+e(t) and ff =tta, (2.1)
wherein ¢ is the time on the watch, « is the reference configuration used, and x, (x, f) is
the deformation, all pertaining to O; whereas the same variables, carrying the superscript
*, pertain to O*. Here Q(t) is a time-dependent orthogonal tensor (Q'Q= QQ! = I, the
identity tensor), c(t) is a vector-valued function, and a is a constant. The ideas under-
pinning this relation are explained in Noll’s paper. We note that it is quite similar to the
relation existing between a deformation as perceived by one observer and a second deform-
ation, perceived by the same observer, obtained by superposing a rigid-body deformation
on the first. However, in the latter the orthogonal transformation is required to be proper-
orthogonal, whereas in eqn (2.1) it is not. We shall return to this point presently. Basically,
eqn (2.1), part 1, ensures that the distance between material points p, and p, ata particular
instant is the same for both observers, whereas eqn (2.1), part 2, ensures that the time lapse
between successive events is likewise the same. Indeed, eqn (2.1) is necessary and suffi-
cient for such agreement. We note that the two observers are free not only to wear different
watches, but also to choose different references. Before Murdoch, the literature was marred
by the frequent repetition of the unnatural view that these references may be assumed to
coincide.
In concert with eqn (2.1), we suppose, this time truly without loss of generality (see the
Problems), that the configuration R,,, say, occupied by the body at time f), is chosen by
xX, t)
Oo
Figure 2.1 Configuration of a body, as perceived by observers O and O*

OBSERVERS AND INVARIANCE | 9
observer (© to serve as reference; in short, « = R,,. We further suppose that observer O*

takes up the same suggestion and selects «* = Rie. Then, det F*, > 0, while (2.1) requires
x’ = Kx+c, (2.2)
wherein K = Q(t,), etc. The Chain Rule (see Supplemental Notes, Part 3) yields the chain
of equalities:
dy’ = Qdy = QF ,dx= QE, K'dx’, andthus Fi, =QEK’, (2.3)
the latter implying that
(det Q)(det K) = 1. (24)
In view of our proposal regarding the two observers’ perceptions of material response,
the relevant constitutive relation for O* is necessarily of the form (cf. eqn (1.2))
T*(p, f°) = Gx @, #), Fe (x*, #);x*). (2.5)

In every course on continuum mechanics, and in Murdoch’s paper, it is explained that
T*(p, f*) = Q(t)T(p, t) Q(t)‘; therefore,
Gx t), F, (x, t);x)Q(t), (26)

ee (Xe (x*, t*), Ft, (x, t*);x*) = Q(t)G.
and this holds for all orthogonal Q(t), for all c(t) and for all a.
To obtain necessary conditions, consider a situation in which x; and F*, are observed
by O* to persist at fixed values during the interval [¢*, t{]. This observer perceives a static
configuration of the body, while the other observer is flying past in an airplane, say, all the
while observing the same body. Evaluating (2.6) at times t, = t* — a and ¢, = tf — a and
eliminating G+, (x7, Ft.; x"), we derive
Qc (y,, F,; x)Q, = Q,G (y,, F,; x)Q',, (2.7)
where y, = xX. (x,t), Fi = F, (x,t), Q, = Q(t,), etc. This is a restriction on the constitu-
tive function used by ©. Furthermore,
Quy, te =Q.,y,+&, (2.8)
where c, = c(#,), etc., and from (2.3), we have Q ,F,K' = Q.,F,K’, which furnishes
F,=QF,, where Q=Q‘,Q). (29)
Here, of course, F, and F, are the values of F, at different times and so (1.3) requires that:
Qe Orth’, (2.10)
the set of proper-orthogonal (rotation) tensors; i.e., the orthogonal tensors with determin-
ant +1. Consequently eqn (2.7) may be re-written as
G.(Qy +d, QF;x) = QG,(y,F;x)Q, where d=Qic with c=e~c. (211)
This is precisely the requirement that the constitutive function G, pertaining to O be

invariant under superposed rigid-body motions.
The reader is cautioned that such invariance, rather than observer consensus regarding
material response, is used by many workers as a basic premise regarding constitutive behav-
ior. Murdoch was the first to show that the latter implies the former, and the demonstration
of this implication given here closely follows his work. Indeed, it is difficult to under-
stand why one would impose invariance under superposed rigid motions as a primitive
hypothesis. For, the response of materials is constrained by the equations of motion
dvT+pb=py and T=T', (2.12)
where 9 is the spatial mass density, b is the body force per unit mass, superposed dots
refer to material time derivatives (8/d¢ at fixed x) and div is the spatial divergence; that
is, the divergence based on position y. These imply that an inertial force is imposed on
the material when the body is subjected to a rigid-body motion superposed on any given
motion; indeed, one would generally need to supply a rather strange distribution of body
force to maintain rigidity of the superposed motion. While such might be produced at a
given point of the body, it is extremely unlikely that it could be generated globally. Even
if it could, there seems to be no reason to suppose a priori that the material responds to
such forces in accordance with (2.11). In contrast, the alternative view, based on observer
consensus, imposes no restrictions on material behavior apart from agreement on the kind
of response (here, elastic) that is elicited. Beyond this conceptual advantage, this point of
view is in harmony with ideas underlying Relativistic Physics, which in principle should
subsume Classical Mechanics.
As a further caution we point out that occasional critics of Murdoch’s reasoning object
that eqn (2.6) yields the conclusion that the constitutive equation for O*, say, necessarily
changes as his/her motion evolves relative to ©. In other words, if O is entitled to the use
of some fixed constitutive function, then O* is not and must therefore be expected to keep
close track of O. On the contrary, eqn (2.6) merely imposes a restriction on the constitutive
equations used by the two observers so as to ensure their agreement, if indeed they are ever
consulted, about the nature of material response. We return to this point below.
Continuing, we have arrived at eqn (2.11) as a logical consequence of eqn (2.6). To
explore the potential for further consequences, consider the special caseeQ, =Q, = +1.
Then, Q = Iand (2.11) reduces to
Gi. (y + d, F; x) = G, (y, F; x), (2.13)
implying that the constitutive function is unaffected by arbitrary variations in its first argu-
ment; i.e., that itis translation invariant. It is thus independent of that argument, and so we
arrive at the major simplification
OBSERVERS AND INVARIANCE | ll
G.(y, F; x) = G(x), (2.14)

for some function G,, while (2.11) reduces to
G,(QF;x) = QG,(F;x)Q’ forall Qe Orth’. (2.15)
If we use the polar decomposition F = RU, where R is a rotation and U is the unique
positive-definite symmetric tensor satisfying U* = F'F, and if we choose Q = Ry) it follows
that
G,(F;x) = RG, (R'E; x)R' = RG, (U;x)R' = FG, (U;x)F,,

where G,(U;x) = U"G,(U;x)U". (2.16)
This choice of Q yields a rotation that depends on £ alone, and is therefore admissible in
(2.15). In practice, it is usually inconvenient to compute U from F, whereas it is trivial to
evaluate the Cauchy-Green deformation tensor C = F'F. Using the fact that U is uniquely
determined by C, we then have:
G, (F; x) = FH,(C;x)F', (2.17)
where H,(C;x) = G,(./C;x). Thus, we have reached the remarkable conclusion that
observers agree on the nature of material response only if the constitutive equation per-
taining to any one of them is determined entirely by the symmetric right Cauchy—Green
deformation tensor.
We have shown, by special choice of Q, that (2.17) follows from (2.15); ie., that (2.17)
is necessary for (2.15). To show that it is sufficient, we use it to obtain
G, (QE; x) = QFH, [(QF)'QF;x](QF)! = QFH, (FF; x)F'Q = QG,(F}x)Q,, (2.18)

which is valid for any rotation Q. Therefore, eqns (2.15) and (2.17) are equivalent. Hence,
we conclude that the Cauchy stress at material pointp is given by a function of the form
T(p,t) = FH, (C;x)F’. (2.19)
Euler’s laws of motion require that T be symmetric, and H, therefore takes values
in the set of symmetric tensors; the Cauchy stress is completely determined by a sym-
metric tensor-valued function of a symmetric tensor. From the experimental point of
view, this affords a major simplification over the original hypothesis embodied in (1.2).
Indeed, reasoning of this kind is one of the hallmarks of modern continuum mechanics
and should always be applied before attempting any laboratory assessment of material
behavior.
If desired, the constitutive equation used by O* may be determined in terms of that used
by O. Combining (2.6), (2.14) and (2.17) furnishes
Gi. = FYHY, (C*jx")(F*)', (2.20)

where
H?,(C*;x') = KH, (K'C'K; K'(x* — ¢))K' (2.21)
in which K and cy are fixed parameters. This constitutive function is fixed once and for all,
and depends on the same list of variables, as interpreted by ©*, as those involved in the
relation used by O.
Other stress measures are of use in the formulation of elasticity theory. They may be
defined in terms of their connections to the Cauchy stress. For example, the popular Piola
stress, P, is given by
P=TF’, (2.22)
where F* is the cofactor ofF defined by
F*(a x b) = Fa x Fb (223)
for all three-vectors a and b; this may be used with eqn (1.3) to show that
F* = JF", (2.24)
provided that F is invertible, as we have assumed. Whether or not this is the case, it is
possible to show that the Cartesian components satisfy
1
Fi, = 5 untascaFics (2.25)
where ¢ is the permutation symbol (e,,, = +], etc.). See Part 1 of the Supplemental Notes.
In addition, the second Piola—Kirchhoff stress, S, is defined by
P = FS. (2.26)
These stresses should carry the subscript « in principle, as is clear from their definitions,
but to avoid cluttered notation we shall invoke our policy regarding F and, thus, usually
refrain from doing so. Using the definitions, it is easy to show that the symmetry of T is
equivalent to that ofS. Using eqns (2.19), (2.22), and (2.26), we also have
S=§,(C;x), where §,(C;x) = /detCH,(C;x), (2.27)
and
P =P,(F;x), where P,(F;x) = F§,(F'F;x). (228)
The Piola stress is useful because the equation of motion may be expressed concisely in
terms of it as
PROBLEMS | 13
DivP + p,b = px, (2.29)
where Div is now the referential divergence (based on x) and p, = Jp is the referential mass
density. Conservation of mass - the notion that the mass of a set of material points remains
always invariant - is expressible concisely as 6, = 0.
In the old days some workers were seemingly put off by the fact that the Piola stress,
by virtue of eqns (2.22) and (2.26), is not symmetric. They tended to prefer the second
Piola—Kirchhoff stress for this reason. Of course there is no free lunch and the equation of
motion based on the second Piola~Kirchhoff stress, given by substituting eqns (2.26) into
(2.29), is seen, unlike eqns (2.12) or (2.29), to involve the deformation explicitly. This is of
no consequence whatsoever, either to the theory or to its implementation, and we shall not
belabor it further.
Problems
1. Given the (Cauchy) stress-response function G, (F,; x,), anda differentiable map
X, = A(x,) from reference configuration k, to reference configuration K,, derive
the constitutive function G,(F,;x,).
2. Repeat the argument about observer consensus, this time without requiring the
observers to choose some initial configurations as reference, to derive the appro-
priate restriction on G, (x, F;x). Clearly point out any changes in the argument,
and whether or not the final result is different from eqn (2.19). Note that the ref-
erences are only required to be in one-to-one correspondence with those adopted
in the text.
3. How does the argument change if an observer decides to switch to the use ofa
mirror to observe the body at some instant in a specified time interval? Of course,
this happens every day in many branches of science.
4, Write the balance law (2.12) in global form and use Nanson’s formula:
an = F*N, (2.30)
where N and n respectively are the unit normals to a material surface in the ref-
erence and current configurations, and @ is the ratio of the area measures of
the surface in the current placement to that in the reference, to derive a global
form of the equation involving integration over the reference. Localize this to ob-
tain eqn (2.29). (A proof of the so-called localization theorem, which is one of
the main tools of continuum mechanics, may be found in the book by Gurtin,
1981,).
5. Prove the Piola identity DivF* = 0, Hint: use the result of the previous prob-
lem together with the divergence and localization theorems. Alternatively,
with reference to eqn (2.25), use the fact that FY = Wasa, where Yuss =
1
3 ec CaacXiXe,c = — Vina.
REFERENCES
Gurtin, M.E, (1981). An Introduction to Continuum Mechanics. Academic Press, Orlando.

Murdoch, Al. (2003). Objectivity in classical continuum physics: a rationale for discarding the
‘principle of invariance under superposed rigid-body motions’ in favor of purely objective con-
siderations. Continuum Mech. Thermodyn. 15, 309-320.
Murdoch, A.L. (2012). Physical Foundations of Continuum Mechanics. Cambridge University Press,
Cambridge.
Noll, W. (1973). Lectures on the foundations of continuum mechanics and thermodynamics. Arch.
Ration. Mech. Anal. $2, 62-92.
FURTHER READING
Lidstrém, P. (2015). A note on frame invariance and material isomorphism for simple materials.
Math. Mech. Solids 20, 461-479.
Mechanical power and hyperelasticity
3.1 Elasticity and energy

The well-known mechanical power identity of conventional continuum mechanics follows
on scalar-multiplying (2.12) by the velocity y, integrating over an arbitrary set S of material
points, and applying the divergence theorem. Thus,
d
P(S,t) = S(S,t) + aes t), 31)
where
P(S,t) -| toda fb jae S(S,t) = [rts 3.2)

op P P
and
1 .
K(S,t) = - [os -ydv (3.3)
2 Jp
are respectively the power supplied to, the stress power in, and the kinetic energy of S, which
occupies the volume P at time t. Here, P is a subset of the region of space R occupied by the
entire body at time f, L = FF" is the spatial velocity gradient, and
t= Tn, (3.4)
where n is the exterior unit-normal field on the boundary @P, is the traction, or contact force
per unit area, transmitted to S by the environment. The dot between vectors is, of course,
the usual Euclidean inner product, while that between tensors is defined by A - B = tr(AB‘),
where fr(-) is the usual trace operation. This extends the definition of the Euclidean inner
product to (2nd order) tensors; in fact, it is seen to be identical to the usual vector definition
when written out in terms of components on an orthonormal basis.

16 | MECHANICAL POWER AND HYPERELASTICITY
The balance (3.1) presumes conservation of mass; that is,
d
qs) =0, where M(S) = [oa- [eae (3.5)
P n
and 7 is the subset of the fixed region x associated with S.

The balance given in eqn (3.1) differs in form from that associated with a discrete par-
ticle. The latter is P = dK’/dt, where P and K are the power supplied to the particle, by the
net force acting on it, and its kinetic energy, respectively; there being, of course, no analog
of stress power. For example, if the particle is a mass tethered to an elastic spring and acted
upon by an applied force, undergoing a one-dimensional motion y(t) while maintaining
frictionless contact with a horizontal plane, then the energy balance takes the special form
P = d€/dt, where € = K + U in which U is the spring energy, obtained by integrating the
spring force F(y) = -U/‘(y), leaving unspecified an irrelevant constant of integration. Given
F(y), such integration is always possible and yields, in the case of unforced motion, the con-
servation law d€ /dt = 0, expressing the fact that the total mechanical energy remains fixed
in the course of the motion.
Proceeding by analogy we suppose that elastic bodies are like elastic springs and that a
similar conservation law holds for them in the case of unforced motion. Thus we assume the
existence of an energy U/ such that the stress power is expressible as S = di//dt, yielding
the conservation law dE /dt = 0 with € = K +U; this time, of course, for the continuum
instead of the particle. Forced motions are then such as to satisfy P = d&/dt. We assume U
to be an absolutely continuous function; here, of volume, and thus suppose that
U(S,t) = / Wd», (3.6)
where W is the (referential) strain-energy density.

We know, from eqn (3.2), that the stress power is expressible in terms of the stress and
the rate of deformation. Using the connection eqn (2.22) between the Cauchy and Piola
stresses, and the formula given in eqn (2.24), we derive
S(S,t) = / JT. FF'dv = / tr(JTE“E' )dv = / P - Fdy, 37)
and, therefore, according to the analogy,
/ P. Fdy = / Wd». (38)
Because 7 C « is arbitrary, we may localize and use eqn (2.28) to conclude that
P,(F;x) -F = W, (39)
pointwise in «. For this to make sense it must be possible to integrate along a path F(t),
between specified limits, to obtain the difference of the function W determined by those
ELASTICITY AND ENERGY | 17
limits and thus depending on the associated values of F. Fixing the lower limit and allowing
the upper to be arbitrary, we thereby construct a function W(F; x), to within a function of
x only, such that
[P, (F; x) — W.(F;x)] - F = 0, (3.10)

where W, is the tensor-valued derivative of the scalar W with respect to the tensor F (see
Supplemental Notes, Part 2).
The first factor in the inner product is an element of the set of second-order tensors. This
isa linear space, just like the space of conventional vectors. We follow common practice and
denote it by Lin. The second factor is the limit ofa difference quotient involving elements
of Lin*, the subset of Lin consisting of tensors with positive determinant (see eqn (1.3)).
While this is not a linear space, the set of differences between its elements is and in fact is
just Lin. Choosing an arbitrary path F(t) in Lin* containing the point F, we conclude from
eqn (3.10) that the term in brackets is orthogonal to any, hence, every element of Lin and,
therefore, that it vanishes. To see this explicitly we exploit the arbitrariness of F and, after
an appropriate scaling of physical units, select F to be the square bracket itself, concluding
that its norm, defined by |A| = /A- A, vanishes and hence that the bracket vanishes, at last
yielding
P,(F;x) = W,(F)x). (11)
Thus, the stress is determined by a scalar-valued function of the deformation gradient,

which, like the constitutive equation for the stress, codifies the properties of the particular
material at hand. This, too, is therefore a constitutive function, furnishing that for the stress
via eqn (3.11). This model is known as hyperelasticity. Its antecedent, given by eqn (1.2),
is known as Cauchy elasticity or simply elasticity. Because we have obtained it as a special
case, by restricting the theory such that the stress power is expressible as a time derivative, it
would appear that hyperelasticity is special. However, we shall see that any elastic material
is necessarily hyperelastic when a further condition is imposed that reflects a widespread
view about how real materials behave.
Before embarking on this demonstration we digress to consider restrictions on the strain-
energy function W following from eqns (2.15), (2.22) and (2.27), which combine to yield
P, (QF) = QP, (E) (3.12)

for all rotations Q, Here, because we are concerned only with properties pertaining to a
material point, we drop explicit reference to x, which plays only a passive role, to help lighten
the notation. This will be done henceforth when discussing local properties.
Following Gurtin (1983), consider a path Q(t) in the set of rotations such that Q(0) =I
and let F(t) = Q(t)F be an associated path of deformation gradients in which F is fixed.
We are confining attention to a particular material point; the fact that F is the gra-
dient of a position field does not impose any restriction on its values thereat, and so
our choice is not subject to any qualifications beyond detF > 0. In this case (3.12)
yields
18 MECHANICAL POWER AND HYPERELASTICITY
P,(F) = QP, (B), (3.13)

and (3.9) implies that
W(F) = B,(F) - F = QP, (F) - QF = #[QPb, (DF Q] (3.14)
where @' = Q'Qis skew and we have used the rule tr(AB) = tr(BA). But P, (E)F is just the
value of JT associated with F. Because this is symmetric, the inner product with Q vanishes
and it follows that W = 0. Integrating from ft = 0 to f = T, say, we find that
W(QF) = W(F), (3.15)

where Q = Q(r). Because the path in the set of rotations is arbitrary, so is Q and we con-
clude that the strain—energy function is insensitive to arbitrary rotations superposed on the
given deformation. This invariance is therefore a necessary condition for the symmetry of
the Cauchy stress. We drop the tildes and, following on our earlier success, pick Q = R,,,
obtaining
W(F) = W(U) = W(C), (3.16)
where W(C) = W(/C), and this, in turn, yields eqn (3.15) for any rotation; eqns (3.15)
and (3.16) are, therefore, equivalent. Thus, eqn (3.16) follows from the symmetry of the
Cauchy stress.
Substituting eqns (3.11) and (3.16) into eqn (3.9) we find
W;-F=(W) =We-C= SymWe -C, (3.17)
where the prefix Sym identifies the symmetric part and C(t) = F(t)'F(t). This belongs
to the set of positive-definite symmetric tensors, while C belongs to the linear space of
symmetric tensors. The inner product thus involves only SymWe and our notation makes
this explicit. Alternatively, we may follow common practice and simply define We to be
symmetric. Using C = F'F + F'F with the rules A. BD = B‘A- D = DA’ - B’, which follow
easily from the properties of the trace operator, we have
SymWe .FF= SymWe FE = F(SymW) - F, (3.18)
yielding
[W, -2F(SymW,)]-F = 0. (3.19)
Reasoning as before we conclude that
W; = 2F(SymW,), (320)
WORK INEQUALITY | 19
and eqns (3.11) and (2.27), part 2, combine to give
§,(C) = 2SymWe. (321)
This, of course, is symmetric and therefore so, too, is the Cauchy stress. Thus, we have
shown that eqn (3.15) implies the symmetry of the Cauchy stress. Taken together with our
previous result, it follows that such symmetry is equivalent to the invariance of the strain
energy under superposed rotations.
3.2 Work inequality

Returning to the basis of hyperelasticity, while most of us may be content with the motiv-
ation provided by the analogy with springs, we should not ignore objections raised by the
skeptics. For them we recount an idea that has become folklore not only in elasticity theory,
but in other branches of continuum mechanics as well. Thus, consider the work done on
a collection S of material points during a time interval [t,, t, ]. According to eqns (3.1) and
(3.7) this is given by
W, = K(S,t) -K(S,t,) + / ( P- Par) dt. (3.22)

tL
Suppose the process is cyclic in the sense that the deformation and velocity fields are the
same at the start and end of the time interval; that is,
x(@t)=x(%b) and X¥@%t)=xX(%4), forall xex. (3.23)
Taking gradients, we infer that
F(x, t,) = F(x,t,) and F(x,t,) = F(x,t,). (3.24)
Considering that all points of the body are involved, cyclic processes are no small feat from
the experimental point of view, and so our skeptics may not be assuaged after all. We shall
therefore resort to regarding such a process as a thought experiment. In general these should
be taken with a rather large pinch of salt.
Continuing, we evidently have K(S, t,) = KC(S, t;) ina cyclic process, leaving
2
Wy = / ( / Pp. eat) dv, (3.25)
Fe ty
where we have interchanged the order of integration, which may be done with impunity if
the process is sufficiently smooth (see, for example, Fleming’s (1977) book). Intuitively,
we expect that it should be necessary to perform non-negative work on a body to cause it
to undergo such a process; that is, YW. > 0. This hypothesis is called the work inequality.
Not accepting it means having to explain how it is that work can be extracted from a body
undergoing a cyclic process. Experience suggests that this is futile, and so the hypothesis
is widely regarded by the community as being sacrosanct, even though it is really just a
thought experiment. In practice, one must contend with instabilities or oscillations that may
intervene when one attempts to create a cyclic process from a sequence of homogeneous
deformations, these typically causing the deformation to become non-uniform and thus
unrelated to the boundary displacements that we detect or control in a typical experiment.
From the experimental point of view, we do not know the local state of deformation in
such circumstances and thus cannot be sure that the process is indeed cyclic. Of course,
homogeneous deformations are directly related to boundary displacements, as discussed
previously in the context of rubber bands. Again we digress.
Localize and we obtain the pointwise restriction
/ B,(F) - Fdt > 0 (326)

f
in the case of elasticity.

To explore the consequences of this inequality, let F(t) be the deformation gradient at
the material point considered, associated with a cyclic process, Consider another process
with gradient F*(t) (not the cofactor), defined by F*(t) = F(t) with t = t, + t, —t. This
is the simply the reversal of the original process; that is, F*(f,2) = F(t,:), F(t) = -F(r)
and F’(t,,) = —F(t,,,). It is, therefore, a cyclic process, and, hence, subject to the work
inequality:
0< / ” BBY) «Hat = - / ” (R(t) « F(r)dr, (327)

ty 4
which is just eqn (3.26) with the inequality reversed. Therefore, for elasticity,
t2
/ P,(F) - Fdt = 0 (3.28)
fh
in a cyclic process.
Now, as ¢ traverses the interval [t,, f,], the deformation gradient traces out a curve in
the nine-dimensional space Lin’. Suppose C is such a curve, and suppose it is closed and
smooth, so that it meets the conditions associated with a cyclic process. Then, eqn (3.28)
is equivalent to
§ B.(F) dF =0. (329)

Cc
Let Fy and F be distinct points on C, and let I’, and I, be the two disjoint parts of C
connecting F, to F, Then, eqn (3.29) may be expressed as
[ P,(F) - dE -| PB, (EF) - dF, (3.30)

ry 2
PROBLEMS | 21
Figure 3.1 A cyclic process in deformation-gradient space
implying that the path integral f_ P,(E) - dF, where I is any smooth curve connecting F,
and F, is in fact the same for all paths having the same endpoints and is thus dependent only
on the latter (see Figure 3.1).
Fixing F, we thus have a function
W(B) = [e@ . dE, (331)

r
modulo a constant. Let F(z) be a parametrization of I’, arranged such that Fy = F(0) and
F = F(u). Then, by elementary calculus,
W,-F(u) = W =P, (F)-F(u), (3.32)
where the dash is an ordinary derivative with respect to u. This is the same as (3.10) and
carries the same consequence; namely, the connection eqn (3.11).
Conversely, if eqn (3.11) holds then P,(F) - dF = dW(B), ensuring that eqn (3.29) is
satisfied. Thus, the work inequality for cyclic processes is satisfied by elastic materials ifand
only if they are hyperelastic.
Problems
1. If one observer concludes that an elastic material is hyperelastic, does every

observer conclude the same? If so, how are their strain-energy functions related?
2. Prove the virtual-work theorem; i.e., show that a body is equilibrated if and only if
iE - Vvdu = [ow -vdv+ / p-vda, (3.33)

K K ap
for all v that vanish on dK \ dk,

3. We showed that if an elastic material is hyperelastic; ie. if Pp = W,, then the
mechanical power theorem for the entire body may be expressed in the form
d€/dt = P, where
E(k, t) = U(k,t) + K(x, 2), (3.34)
in which K is the kinetic energy, P is the power of the applied loads, and U is
the strain energy. (Actually, we showed this for a sub-volume 7 C «; the present
special case is recovered on choosing 7 = x.). Thus, the total mechanical energy
€ is conserved; i.e,, it is independent of time, if there are no loads acting on the
body. It is possible for non-zero applied forces to generate a conservation law of
the same kind. These forces should be such such that P = d£/dt for some func-
tion L. In this case, the motion satisfies the conservation law d&’/dt = 0, where
€' = (U -L) +K. The term in parentheses is called the potential energy of the
body and loads, in combination. Because of this conservation law, such forces are
called conservative.
(a) Show that dead loading, in which b and p respectively are assigned as func-
tions of x in « and on d«,, furnishes an example of conservative loading.
What is the load potential £?
(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 surface of the body after deformation. Give the corresponding load
potentials.
REFERENCES
Fleming, W. (1977). Functions of several variables, Springer, Berlin.

Gurtin, MLE. (1983). Topics in finite elasticity. SLAM, Philadelphia, PA.
FURTHER READING
Barenblatt, G.I. and Joseph, D.D. (eds) (1997). Collected papers of R.S. Rivlin, Vols 1 and 2. Springer,
NY.
Carroll, M.M. (2009). Must elastic materials be hyperelastic? Math. Mech. Solids 14, 369-376.
Material symmetry
4.1 Stress response
Consider what happens if we surgically remove a small neighborhood of a material point in

a reference configuration, subject it to some sort of transformation such as a deformation,
and then re-insert it. If the material response to a given deformation, represented by the
stress or the strain energy, should happen to be the same as it was before the operation,
then the latter is mechanically undetectable and the two local neighborhoods—that before
the operation and the one after—are effectively indistinguishable as far as the properties of
the material are concerned. Such an operation is called a material symmetry transformation.
Our purpose in this chapter is to outline the general theory of such transformations and to
apply it to some practical examples.
Before doing this, it is necessary to understand how a change of reference manifests it-
self in the theory. This is the lesson of Problem 1 in Chapter 2. Thus, let « and ju be two
reference configurations and suppose, as before, that R is the configuration occupied by the
body at time t. Then,
Xe(% t) = y = x, (2,6), (4.1)

where x and z, respectively, are the positions of material pointp in « and jz. Let A(-) be the
fixed map that transforms « to 4; that is, z = A(x). Because the two references are in one-
to-one correspondence with points of the body, they are in such correspondence with each
other as well. This means that A is invertible. By the inverse function theorem, its gradient
R = VA(x) is then an invertible tensor. Applying the chain rule to eqn (4.1) and reinstating
the appropriate subscripts, we derive
F,dx = dy = F,dz =F,Rdx, andhence F, =F,R (42)
For the time being we confine attention to Cauchy elasticity, returning to hyperelasticity
later. Since the change of reference is merely a change in the way we record information, it
has nothing to do with the actual state of the material at time t, which is thus unaffected by
the change. Accordingly,

© DavidJ. Steigmann, 2017. Published 2017 by Oxford University Pres
24 | MATERIAL SYMMETRY
T(p,£) = G.(F,;x) = G,(F,;2), where G,(F,;z) = G,(F,R(z);A7(z)). (43)
This is how the constitutive function for the Cauchy stress is obtained when the reference
configuration is changed.
Let us focus attention on a particular material point po. Because the stress at this point
is sensitive only to deformations in some neighborhood of it, we need only consider local
changes of reference N,(po) > N,,(po), say, where N,, (po) is the image of N, (po) under
the map A. This allows us to effectively marginalize the parametric dependence of the con-
stitutive function on reference position, in the case of non-uniform materials, simply by
arranging A such that A(x) = X, where x, is the reference position of po; thus,
T (po, t) = G,(Fx3%o) = G,(F,j;%), where F, = F,R(x). (44)
Consider now an experiment in which N, (po) is subjected to a deformation x (x, £),

for x € N, (po); the response at po is G,(F,; Xo), where F, is the gradient of x (x,t) at x.
Let N,,(po) be subjected to the same deformation; that is, to x (z,t), for z € N,,(po), in
which x (-,£) is the same function. The response is G,,(F,,; X)), where F,, is the gradient of
x (z, t) at xo. Note that F, = F,,(= F, say) in this case, because they are the gradients of the
same function, evaluated at the same point. So, the responses elicited by the experiment
on N, (po) and N, (po) are G, (F; %)) and G, (F; Xo), respectively. These need not have any
relation to each other, except in the case when z = A(x) is a symmetry transformation, in
which case they coincide. That is, symmetry transformations at po are such as to satisfy
G, (F; %)) = G,(F; x). (45)
For, no experiment involving measurement of the Cauchy stress can then distinguish
between N, (p)) and N,,(po}. Combining this with eqn (4.4), part 2, we derive
G, (F; X)) = G,(FR;%)), where R= VA(x%) (4.6)
is the gradient at py of the symmetry transformation, in which F € Lin’ is arbitrary. Given

the set of all such transformations, this amounts to a restriction on the function G,(-; Xo).
Since it requires FR to be the pointwise value ofa deformation gradient whenever F is, the
restriction makes sense only if det R > 0. This, in turn, implies that symmetry transform-
ations correspond to possible deformations of the material. We shall have reason to return
to this point later.
The following observation is fundamental: Let G,q,) be the set of all R such that
eqn (4.6) is satisfied (not to be confused with the response function (1.2)); that is,
Geo) = {Ri Ge (F}X0) = G, (FR; xo) } . (47)

STRESS RESPONSE | 25
Then this set is a group, in the sense that
(i) Ie Gaiyo)s
(ii) If RyR,€ Gig then RR, € Gp),

(ii) If Re Geto) then R’e€ Grip): (48)
The first of these is obvious from the definition Of Geiyg) To prove the second, we observe
(suppressing the passive argument x,) that G,(F(R\R,)) = G,((FR,)R,) = G,(FR,) =
G, (F), and the third follows from G,(FR™) = G,((FR’)R) = G,(FR'R) = G,(F).
Note that R € G,i.) implies that R' € G,;,.) for any integer n > 0. Thus, G,(F) =
G,(FR"), where det(FR") = (det F)(detR") = (det F)(detR)". Let n — oo. Then, if
detR > 1 we have det(FR") — 00, corresponding to unbounded dilation; whereas, if
detR < 1 we have det(FR") — 0, corresponding to unbounded compaction. Material
symmetry then requires that the stress remain unaffected by unlimited dilation or compac-
tion of the material. This is plainly unphysical, and so we impose the requirement
Gaiyo) CuUe= {R: detR = 1} . (49)
U is called the unimodular group.

Noll introduced the elegant idea of characterizing elastic materials as either fluid or solid,
depending on the nature of the symmetry group. For example, in an inviscid compressible
fluid the Cauchy stress is given by
T = -p(p)1, (4.10)
where p(p) is the pressure-density relation. In this case, we have G,(F}x) =
-p(p,(x)/ det F)I, yielding G, (FR; x) = —p(p, (x)/ det(FR))I. It follows immediately that
Gey) = U and, so in view of eqn (4.9), we are justified in saying that fluids have maximal
symmetry.
For solids we assume the existence of N, (p) such that
Gey & Orth’. (4.11)
Such N, (p) is called a local undistorted configuration. The idea is motivated by the structure
of a unit cell of an undistorted crystal lattice; these are mapped to themselves by discrete
rotations. Furthermore, we have in mind the fact that, for solids, a change in shape is detect-
able by experiment. Accordingly, the map A is detectable if it generates a strain. Symmetry
transformations should, therefore, be strain-free, and this, in turn, implies that R'R = I. The
restriction eqn (4.9) then yields eqn (4.11), even for non-crystalline solids. Isotropic solids
are those for which N, (p) exists such that
Geip) = Orth’. (4.12)
Note that we have not invoked frame invariance. For constitutive functions that are
admissible from this standpoint we use eqn (2.19) to conclude that
G,(FR) = F[RH,(R'CR)R']F', (4.13)

so that ifR € Gi, then
H,(C) = RH, (R'CR)R’, (4.14)
and if Geip) Cc Orth’,
H, (R‘CR) = R'H,(C)R. (4.18)
Suppose, for example, that a particular crystal lattice is such that the 180° rotation
R=2n@n-I (4.16)
about the unit vector n belongs to G,.). Then, both R and -R satisfy eqn (4.15), imply-
ing that the reflection through the plane with normal n is mechanically undetectable. In
this way, the symmetry group may be extended to accommodate reflection symmetry with
respect to crystallographic planes, despite the fact that such transformations cannot be
associated with an actual deformation.
We have indicated that the symmetry group depends not only on the material, but also
on the reference. To see precisely how this occurs consider a general invertible map 7, with
gradient P (not to be confused with the Piola stress) that takes reference k,, say, to K,. From
eqn (4.3), part 2, we have the connection
G,, (F) %) = Gy, (FP(%); 7" (m2), (4.17)

where x, = 7 (x,), implying that
G,, (Fj 7'(%))) = G,, (FP(%)

7; x). (4.18)
Suppose now that R € G,,q). Then,
G,, (Fj 07'(x,)) = G,, (FR; 77'(x))). (419)
Eqn (4.18), however, implies that
G,, (FR; 7” (x)) = G,, (FRPGn) 5%), (420)

and therefore (4.18) and (4.19) give
G,, (FRP(%))"';2) = Gy, (Fj 7") = Gy, (FP(m) 5%). (421)

Defining F = FP", we recast this in the form
G,, (F; X)) = G,,(FPRP"; x), (4.22)

PROBLEM | 27
and conclude that
Gey) = {PRP*: REG}. (423)
This result is known as Noll’s Rule.

It follows that if G,, 4) satisfies eqn (4.11), then in general G,,.,) does not. Therefore, the
existence of a (local) reference configuration relative to which eqn (4.11) holds is an essen-
tial aspect of the definition of a solid. However, eqn (4.9) is always satisfied. In particular,
fluids have the property that G,) = U for all choices of x. This fact gives meaning to impre-
cise, but often-heard remarks to the effect that “fluids have no reference configuration.” It
would be better in this case to say that the structure of the constitutive function is insensitive
to the reference configuration.
4.2 Strain energy
We have outlined the theory of material symmetry in terms of restrictions on the stress re-
sponse. We may just as easily do so for the strain—-energy response. Repeating the foregoing
essentially verbatim, we arrive at the definition
&y) = {R: W,(FR;x) = W,(F;x)} (4.24)

of the associated symmetry group, which as before, is restricted by
&x(p) Cc U~. (4.25)
The obvious question is: How are g,q) and G,) related? To explore this, suppose
REG). Then, using eqns (2.22) and (4.6) and suppressing the passive variable x, it
follows that
P,(FR) = G,(FR)(ER)* = P,(F)R’, (426)
where we have used the general rule (AB)* = A*B*.
Problem
Prove this rule.
Thus, P,(F) = P, (FR)R‘, or
W,(F) = W;(E)R' = W,(F), where F = ER, (427)
in which the chain rule has been used to derive the second equality and the subscript « has
been suppressed for clarity. Integration at fixed R then furnishes
W,(F) = W,(FR) + C(R), (4.28)
in which C is independent of F. Evaluating at F = I leads to the implication
W,(F) = W,(FR) + W,(D) - W,(R): (429)
Conversely, if this is satisfied then by reversing the steps, we see that R € G,q) and, thus,
that the two statements are equivalent.
Next, suppose R € g,(,), so that W, (I) = W,(R) in particular. Then eqn (4.29) is sat-
isfied and so R € G,,,); that is, 9.4) © G,i). Recall that admissible strain-energy functions
meet the invariance requirement, eqn (3.15); in particular, W,(Q) = W, (I) for all rota-
tions Q. This means that for solids, i.e., for Gap) € Orth’, any R € Gy) satisfies W,(R) =
W,(I), but such R also satisfies eqn (4.29), so that W,(F) = W,(FR) and R € g,q). We
have thus shown, for solids, that Gq) © geq and, hence, that
&«v) = Ge) . (4.30)
The symmetry groups for stress and energy are thus one and the same. Beyond this, we
may proceed exactly as in the case of the stress—response function to extend the symmetry
group to include improper orthogonal transformations as needed to incorporate reflection
symmetry, using the fact that admissible strain-energy functions are expressible as W,(F) =
W,(C), and the consequent fact that W,(C) = W,(R'CR) for all R € Buiy» Of course, an
explicit dependence on x is allowed, to cover non-homogeneous materials.
4.3 Isotropy
In view of the foregoing result, it is enough to characterize symmetry in terms of the strain-
energy function. In the case of isotropy, then, there is presumed to exist «(p) such that
Bey) = Orth"; that is,
W,(C;x) = W,(R'CR;x) forall R€ Orth. (431)
In virtually every text on continuum mechanics, it is established that this restriction is

equivalent to the statement
W,(C; x) = Uh, 1,4; x), (432)
for some function U, where
1
I,=trC, I,=tC*=—-[f -tr(C’)] and I, = detC (4.33)
2
ISOTROPY | 29
are the principal invariants of C. The proofis a model for extensions to other kinds of sym-
metry, such as that described in the next chapter, and so, at the risk of being repetitive, we
pause to outline it explicitly.
Suppose, then, that A,B € Sym", the set of positive-definite symmetric tensors, and that
these are such that their invariants coincide: I,(A) = I,(B). Then, because the invariants
define the characteristic equation
w-lhw+hu-t,=0 (4.34)
for the eigenvalues 11, it follows that A and B also share the same (real-valued and positive)
eigenvalues. From the spectral representation for symmetric tensors it is concluded that
A= \ Ba; @a, and B= y Lb; ® b;, (4.35)
where the sets {a,} and {b,} are orthonormal. The latter property means that the tensor Q,
defined by Q = a; @ b,, is orthogonal. Thus,
B= > HiQ'a, ® Va, = VY (o> La, ® a.) Q= VAQ, (4.36)
and eqn (4.31) implies that W,(A;x) = W,(Q'AQ;x) = W,(B;x), meaning that every
W.(5 x) satisfying eqn (4.31) is determined by the principal invariants of its argument
and, hence, that eqn (4.32) is valid. Conversely, if the latter is true, then since I,(R'CR) =
I,(C) for all orthogonal R, eqn (4.31) follows, and is thus necessary and sufficient for
eqn (4.32).
To obtain the stress we use the chain rule in the form
3 3
SymWe -C = (WY = D> Uk = D> USym(L)c- C, (437)

k=] k=1
where U, = dU/dI,. Using (Supplemental Notes, Part 4)
Sym(jc =I, Sym(L)c=h1-C and Sym(I,)¢ =C* =,C", (4.38)
together with eqns (2.22), (3.11), and (3.20), we thus derive
JT = 2[(U, +1,U,)B - U,B’ + 1,U,1], (439)
whereJ = ./I, and B = FF' is the left Cauchy—Green deformation tensor.

In the next chapter, the foregoing argument is extended to obtain the general form
of the response function for transversely isotropic materials. For these, there exists «
such that
W,(F;x) = W,(FR;x) forall Re Orth* withaxis m; (4.40)
that is, for all rotations R such that Rm = m, the axis of transverse isotropy.
However, it is time for some exercises.
Problems 1
1, Prove that Uis, indeed, a group.

2. Prove the second equality in eqn (4.27).
3. Let W,(F;x) = w(J;x), where -dw/dJ = p(p), the pressure-density relation for
fluids. Show that g,) = U and, hence, that eqn (4.30) is satisfied for fluids.
4, We established, for solids, that the symmetry group for admissible stress—
response functions may be extended to include elements of Orth, the set of
orthogonal tensors. Show that the same conclusion applies to the symmetry
group based on energy response.
5. Given eqn (4.39) for some choice of reference, compute the stress—response
function relative to any other choice.
6. Show that, for an isotropic elastic material, the strain-energy function is express-
ible as a function of the principal stretches, ie, U = @(A,, Aa, As; X), in which the
stretches can be permuted arbitrarily (i.e., they can appear in this function in any
order). Use W = 3S: C and the spectral representation C = )~A?u; @ u, (note
that |u,| = 1, but u, ¥ 0!) to obtain the representation
0
P= > a Qu, (441)
where v; = Ru, and R is the rotation in the polar decomposition of F. Thus, U =

>> Aju; ® u,, R = v; @ u, and F = }- A,v, @ u,. Show that the cofactor ofF may
be represented in the form F* = )~ u,v, ® u, where 2, = J/A,. All sums range
over {1, 2, 3}.
7. Foran isotropic (relative to « ) material we have W(F(u)) = W(F) where F(u) =
FQ (u), Fis fixed and Q(u) isa one-parameter family of rotations with Q(0) = L
Thus, W’ = 0, where W’ = dW/du. Prove that F‘P must then be symmetric and
that this, in turn, is equivalent to the statement TB = BT, granted the symmetry
of the Cauchy stress. Reverse the argument and show that TB = BT implies isot-
ropy. This condition is therefore necessary and sufficient for isotropy, granted
the symmetry of the Cauchy stress. Show that the condition is equivalent to three
universal relations that apply to all isotropic materials:
Br(Tiu - Tx) = Tr2(By - By) + TsBys - Ty3By

By3(T. — T33) = To3(Bry — B33) + T13Bay - T1Bi3 (442)
Bs, (T33 - Ti) = Ts: (Bs3 - Bi) + Tu By - T:Bu,
where the components are referred to {e; ® e;}.

PROBLEMS 1 | 31
Consider a unit cube subjected to the homogeneous deformation
Yi = ax + kayxy, Yo = Arh, Vy = Ags, (4.43)
where 4), 4, a, and k are positive constants. Note that for homogeneous ma-
terials this deformation is automatically in equilibrium in the absence of body
forces.
Show that the universal relations yield a single relation of the form
Tu - Tx = Ty2F(a,,4),k). (4.44)
Find expressions for the traction, t, acting on the material planes x, = const. and
x, = const., and suppose n-t vanishes on these planes in the deformed con-
figuration, so that only shear tractions are acting. Obtain the purely geometric
relation
ay = (1+k)aj. (4.45)
This furnishes a simple necessary condition for isotropy that can be tested
experimentally, Indicate the meaning of this equation on a figure.
Before leaving this discussion we pause to develop a formulation of isotropic elasticity

that has proved to be convenient in applications. To this end, we use the result of Problem
no. 6 above, together with
Le +aee, L=Warenevn and =A Verne 3

(4.46)
where the A, are the principal stretches. These, of course, are the principal invariants of C.
The principal invariants of U, namely
i, =trU, i, =trU* and i, =detU, (447)
are
ip =A, tAQtAs, & =A AZ +A A, +AIA, and i, = A,AZA;, (4.48)
which may be used with (4.46) to verify that
L={-2b, L=i-2ii, and =i. (4.49)
It then follows from (4.32) that the strain-energy function, in the case of isotropy, is
expressible in the form
W,(C;x) = wits inyis3x), where w(i,,

iz, i3;x) = U(G - 21, § - Qi, 25x). (4.50)
Asa useful corollary, the strain—energy is a completely symmetric function w of the prin-
cipal stretches in this case, remaining invariant with respect to interchange of any two of
them, In fact, the nonlinear system eqn (4.49) is uniquely invertible. Its inverse may be
expressed in the form eqn (4.48), in which (see Rivlin’s (2004) paper)
Leo {1 + 4A
Fe yi
r= J3 2A cos) 5Tp - 2m
2 |} j i= 1,2,3,
$= 1,2,3 (451)
451
|" 3
where
1
A=(-3h)'? and @=cos” pen ~ 911, +271) . (452)
A simple application of the chain rule yields
Om/OA, = wy + (A, + Ag)w + AQAWy, (4.53)
where w, = dw/di,, together with two similar relations obtained by permuting the principal
stretches. Using these relations in the solution to Problem no. 6 above, we derive
We =, Soy, @ u, + wy[(Az + As)vi @ uy + (Ay +A3)v. @ wy
+ (A, + Az)v; @ uy] + w3(A,Asv, @ uy, + A,Asv, @ u, + A,A,V; @ u,). (454)
The first sum on the right-hand side is recognizable as R, the rotation factor in the polar
decomposition F = RU of F; the third is just RU*(= R*U* = F*); and the second is
(Ar + As) Ou, += (Ay tarts) Dov, @u- > Av @uy=4R-F. (455)
Thus,
P=W,=Ro, (4.56)
where
o = (w, + iw,)I - w,U + w,U* (457)
is the Biot stress tensor.

We caution the reader that the factorization eqn (4.56) of the Piola stress into the rota-
tion and the Biot stress is appropriate only in the case of isotropy, whereas a tensor usually
referred to as the Biot stress, which yields eqns (4.56) and (4.57) in the case of isotropy, has
a wider significance. We shall not need the general form here, however, and so suggest that
reference be made to Ogden (1997) for further discussion.
PROBLEMS | 33
Problems
1. Use the relations between the invariants J, ofC (or B) and the invariants i, ofU
(or V) discussed previously to establish the three-dimensional formula
WR =i,F*-FC+(i-i,)F, where i= iyi, - iy = (Ay +A.) (A, + Ag) (Ar +3).

(4.58)
In practice, one has direct access to the I, from, say, a finite-element analysis
in which the deformation gradient is obtained from nodal displacement data.
Discuss the problem of obtaining the i, in terms of the I, and, thus, obtaining R
in terms of the deformation gradient directly. You should appreciate the conveni-
ence afforded by this method because it means one does not have to go through
all the steps needed to carry out a computationally intensive polar decomposition
in applications calling for the evaluation of R.
2. Establish the formulas
(i). =R, (i,)p = i: R-F, (is)e =F* and (i,)e =R. (4.59)
At this stage, it is instructive to revisit eqn (2.15). Recall that this is formally equivalent
to the statement that the constitutive function adopted by observer () is insensitive to ar-
bitrary rigid-body motions superposed on a given motion. In fact, it is widespread practice
to impose this requirement in place of frame invariance. However, this interpretation of
eqn (2.15) is flawed, if only because it is not possible to subject a deformable body to an ar-
bitrary rigid-body motion. To see this, imagine a uniform, isotropic elastic body undergoing
the rigid-body motion
y = Q(t)x + c(t), (460)

with Q € Orth*. Then, F = Q, yielding R = Qand U =I, and the Cauchy stress T, defined
by P = TF” reduces, with the aid of eqn (4.39), to
Te=dl, (4.61)
where c is a constant. The divergence of the Cauchy stress vanishes. If no body forces are
acting, the equation of motion (2.12) reduces to
0 = A(t)y + d(é), (4.62)
where A = QQ! and d = ¢ - Ac. Evaluating the gradient with respect to y at an arbitrary
point of the body yields A = 0; thus, Q = 0, and é = 0. If we identify the reference config-
uration with the initial configuration of the body and assume the initial velocity to vanish
pointwise, then the initial value of Q is I, and the initial values of Q, cand éall vanish, yield-
ing y = x forall t. The only rigid motion is then the trivial motion in which the body remains
stationary.
REFERENCES
Ogden, R:W. (1997). Non-linear elastic deformations. Dover, NY.

Rivlin, R.S. (2004). A note on the constitutive equation for an isotropic elastic material. Math. Mech.
Solids 9, 121-129.
FURTHER READING
Barenblatt, G.I. and Joseph, D.D. (Eds) (1997). Collected papers of R.S. Rivlin, Vols | and 2. Springer,
NY.
Rajagopal, K. and Wineman, A.S. (1987). New universal relations for nonlinear isotropic elastic
materials. J. Elasticity 17, 75-83.
Fiber symmetry
The term fiber symmetry refers to a symmetry group consisting of rotations about an axis.
Typically, this axis is identified with the unit tangent to a fiber embedded in the material, as
ina fiber-reinforced composite or a fibrous biological tissue (Figure 5.1). Such materials are
said to be transversely isotropic; they are effectively isotropic in the plane orthogonal to the
fiber direction. Our objective here is to solve the representation problem for transversely
isotropic strain—energy functions, i.e., to find the maximal list of variables upon which these
functions may depend. We have already solved the representation problem for isotropy,
concluding, in that case, that the energy is a general function of the principal invariants
Ti3 of C.
Recall that the general restriction imposed by material symmetry is
W(C) = W(R'CR) (5.1)
for all positive definite, symmetric C, and for all R € g,q), the symmetry group relative to
configuration « at the material point p. This (local) configuration is undistorted if g.q) C
Orth, the group of orthogonal tensors.
Transverse isotropy is associated with the symmetry group
Spy ={Q: Qe Orth and Qm(x)=+m(x) with |m(x)| = 1} » (82)
where m(x) is the fiber axis at the material point p. As all arguments presented here are
purely local, henceforth, we suppress this material point in the notation. The strain energy
is thus invariant under all rotations about the fiber axis, and under reflection through the
plane—the isotropic plane—perpendicular to this axis.
As a prelude to our main theorem, note that if Qm = +m, then, as Q € Orth, we have
Q'm = +m; this follows simply on multiplying by Q'’. Thus, R' € g, ifR € g (we drop the
subscript on g,,)). Moreover,
g={Q; Qe Orth and QMQ' =M}, (53)
Finite Elasticity Theory. DavidJ, Steigmann,

© David J. Steigmann, 2017. Published 2017 by Oxford University Press.
36 | FIBER SYMMETRY
Figure 5.1 A material with a continuous distribution of embedded fibers
where
M=m®@m. (5.4)
This is called the structural tensor for transverse isotropy.
Proof First, we note that for either choice of sign,
m @m = tm @ +m = Qm ®@ Qm = Q(m @ m)Q’, (5.5)
and so M = QMQ‘. This shows that Qm = +m > QMQ! = M. To show the con-
verse, suppose QMQ' = M with Q & Orth and M as defined in eqn (5.4). Then,
QM = MQ, or Q(m @ m) = (mM @ m)Q, i.e. Qm @ m = m @ Q'‘m. Then,
Qm = (Qm @ m)m = (m @ Qm)m = (m- Qm)m. (56)

This, and the orthogonality of Q, imply thatm-m = Qm- Qm = (m: Qm)’m-m
and, hence, thatm - Qm = +1. Thus, QMQ' = M & Qm = +m.
Our strategy is to replace the representation problem eqn (5.1), eqn (5.2) by an
equivalent representation problem for isotropic functions. We make use of the following:
Theorem W is invariant under g, i.e, W(C) = W(RCR’) for all R € g, if, and| only if, the
function W, defined by W(C) = W(C,M), is invariant under Orth, ie, W(C,M) =
W(QCQ;, QMQ‘) for all Q. € Orth; that is, ifand only if W i is a jointly isotropic function
of its arguments.
FIBER SYMMETRY | 37
Proof of sufficiency: Suppose W(C, M) is invariant under Orth. Then, because g C

Orth, it is also invariant under g. Pick Q € g. Then,
W(QCQ) = W(QCQ,M) = W(QCA, QA(QAMQ)Q)

= W(C,Q'MQ) (invariance under g)
=W(C,M) (because Q' € @) (57)
and thus W(QCQ’) = W(C), ie., W is invariant underg.
Proof of necessity: Before proceeding, we define a function
W(C,P) = W(RCR’), (58)
where R € Orth satisfies RPR’ = M, ie, P = R'MR. Note that ifR € g, then P=M
and W(C,P) reduces to W(C) = W(C,M). Thus, W defines an extension of W
from g to Orth. We now show that W is invariant under Orth.
For any Q € Orth, by the definition of W,
W(QCQ’, QPQ'’) = W(R(QCQ')R)), (59)

where R € Orth satisfies RQP(RQ)' = M. Let R = RQ, Then RPR' = M, and the
definition ofW yields
W(C,P) = W(RCR’). (5.10)
Thus, W(QCQ', QPQ') = W(C, P), and so W is invariant under Orth.
To summarize, we have shown that w(C) is invariant under g © W is an isotropic

function. We turn now to our main result, the:
Representation theorem for transverse isotropy: W(C) is a function of I,(C); k = 1,2,3,

and of
L,(C)=m-Cm=C-M and 1,(C)=m-C’m=C’-M, (5.11)
Proof of sufficiency: We know that I.(QCQ‘) = I,(C);k = 1,2, 3, for all Q € Orth.
Furthermore,
trQCQ’(QMQ’)] = tr(QCMQ’) = tr(Q.QCM) = tr(CM), (s.12)

and the same is true if C is replaced by C’; accordingly, I,(QCQ‘, QMQ’) =
1,.(C,M); k = 4, S. It follows that if
W(C,M) = U(h,...Js), (5.13)

then W(C, M) = W(QCQ’, QMQ’).
38 | FIBER SYMMETRY
Proof of necessity: Suppose W(C, M) = W(QCQ’, QMQ’) for all Q € Orth, We want
to show that W(C,M) = W(C,M) whenever I,(C,M) = I,(C,M); k = 1,...,5.
For, W(C, M) is then determined by the list I,(C, M); k = 1,..., 5.
Proceeding, consider any symmetric tensors A and A, and any unit vectors n
and n. Let N = n @ nandN = n @ i, and suppose
1,(A) =1,(A); k= 1,2,3,

n-n=n-n (trN=¢rN),
n-An=f-An_ (tr(AN) = tr(AN)),
n-A’n=n-An (tr(A’N) = tr(A'N)). (5.14)
From the first of these we conclude that A and A have the same eigenvalues;
therefore,
3 3
A= y Au, @u, and A= y 4,u,; ® u, (5.15)
isl ict
where {u,} and {u,} are orthonormal triads. Thus,
Q=4u, @ yu; € Orth, (5.16)
and
A= QAQ'. (5.17)
From eqn (5.14).3,4 it follows that for any scalars @, B, y,
n-(al+BA+yA )n=n-(al+ BA+YA’)n, (5.18)
whereI = QIQ! = Lice.,
n: (a+ BA+ yA’)n = Q'n- (al + BA+ yA’) Q'n, (5.19)
We will prove the theorem under the restriction that the eigenvalues are distinct,
leaving the general case to the interested reader. Before proceeding, we pause to
verify a:
Lemma: I, Aand A’ are linearly independent, and
S = Span{I, A, A’}, where S = Span {u, ® uy, u, @ uy, u; @ us}. (5.20)
The proof is standard and may be found, for example, in Gurtin (1981). We sketch
it here. Assume that
aol + 4A +a,A’ = 0, (521)
FIBER SYMMETRY | 39
with {ao, a1, 4} # {0, 0, 0}. This is equivalent to

3
IC + a,A, + a,A7)u; @ u, = 0, (5.22)

i=l
and hence to
AataA,+a,rd*=0; i= 1,2,3. (5.23)

This means that the three distinct A, all satisfy the same quadratic equation, which has
at most two distinct roots. This contradiction leads to the conclusion that eqn (5.21)
is true if and only if {a, a,,a,} = {0, 0,0}, and hence that the set {I, A, A’} is linearly
independent. Furthermore, by the spectral representation (5.15), first part, we have
I, A, A’ € S,athree-dimensional vector space. As any set of of n linearly-independent
elements of an n-dimensional vector space constitutes a basis for that vector space,
eqn (5.20) follows, and we conclude that every B € S satisfies
B=al+BA+yA’ (5.24)
for some (unique) a, B, 7.
Returning to the theorem, we see that eqn (5.19) is equivalent to the statement
n- Bn = Q‘n- B(Q‘n) (5.25)
forall B € S. Let R € Orth be such that n = Rn. Then, N = RNR’ and eqn (5.25) reduces
to
D-N=0 forall N=n@®n, (5.26)
where
D = B- (Q‘R)'B(Q‘R) € Sym. (527)
Let n, be the eigenvectors of an arbitrary symmetric tensor S, ie, S = an S;N,, where S;

are the corresponding eigenvalues andN, = n, @ n;. From eqn (5.26) we conclude that D €
Sym satisfies D - S = 0 for all S € Sym and, hence, that D vanishes, i.e., that
(Q'R)'B(Q'R) =B, forall BES. (5.28)
Now, every B € S is expressible in the form B = a Bu; ® u; for some scalars B,.
Then, since eqn (5.28) holds for any B € S, it is necessary and sufficient that
R'Q(u @ u) = (U@u)R'Q, (529)
for all u@u€ {u, @u,,u, @u,,u, @u,}. Thus, (R'Qu) @u=u® (Q'Ru), which
yields
40 | FIBER SYMMETRY
R'Qu = (u: R'Qu)u = (R‘Qu: u)u. (5.30)
However, |R'Qu| = |u| because R'Q€ Orth. Thus, |u|’ = R'Qu- R'Qu = (R'Qu-
u)? |u|’, so thatu - R‘Qu = +1. Then, R‘'Qu, = +u,, or Qu, = +Ru,; i = 1,2, 3. Finally,
3 3
RAR' = )°A,Ru, ® Ru, = )) A,Qu, ® Qu, = QAQ' =A. (531)

i=l isl
To summarize, we have shown that if eqn (5.14), parts 1-4, hold, then
N=RNR' and A=RAR' (5.32)
for some R € Orth. Then,
W(A,N) = W(RAR', RNR’) = W(A,N), (533)
and we conclude that W(A, N) is determined by I,(A); k = 1, 2,3, and by n- An, n- A’n
and n - n, the last of these being redundant if n is a unit vector.
Our outline of this representation theorem follows the proof given in the paper by
Liu (1982). The papers by Boehler (1979) and Zheng (1994) should also be consulted.
The method of the theorem may be extended to cover any type of symmetry that can be
characterized by a set of structural tensors, i.e., by tensors § such that RSR' = § for all
R € g. In fact, the latter restriction may be relaxed, as shown in the paper by Man and
Goddard (2017). The general issue of material symmetry and attendant representation the-
orems is discussed in a series of fundamental papers by Rivlin and associates (Barenblatt
and Joseph (1997)).
To use the present representation theorem in the context of the theory of elasticity for
transversely isotropic materials, we impose
AC) = U(N, Js), (534)
and find, for any parametrized path of deformations, that
5 5
SymWe C= (WY = S> Uh, = D5 USym(I)e + C, (538)

kel kel
with U, = dU/dI,. Here, we have
I,=M-C (5.36)
and
I, =M-(C’) = (CC+CC)-M=(MC+CM)-C, (537)

REFERENCES | 41
and, therefore,
Sym(Iy)e =M and Sym(Is)c = MC + CM. (5.38)
Combining these results with eqns (3.11) and (3.20), we derive the constitutive represen-
tations
P = 2F[(U, + 1,U,)1- U,C + U,C* + Usm @m+U,(Cm@m+m@Cm))] (539)
and
JT =2[(U, + 1,U,)B - U,B’ + 1,U,1 + U,Fm @ Fm + U,(BFm @ Fm + Fm ® BFm)],

(5.40)
the second of which may be compared to eqn (4.39).
The paper by Horgan and Murphy (2016) describes an interesting application of this
model to a specific boundary-value problem, which highlights the unusual features of trans-
verse isotropy. Transverse isotropy plays a large role in the study of bioelasticity. The
collection edited by Dorfmann and Ogden (2015) provides a thorough account.
REFERENCES
Barenblatt, G.I. and Joseph, D.D. (Eds) (1997). Collected papers of R.S. Rivlin, Vols 1 and 2. Springer,
NY.
Boehler, J.-P. (1979). A simple derivation of representations for non-polynomial constitutive
equations in some simple cases of anisotropy. ZAMM §9, 157-167.
Dorfmann, L. and Ogden, R.W. (Eds) (2015). Nonlinear mechanics of soft fibrous materials. CISM,
Vol. 559. Springer, Wien.
Gurtin, MLE. (1981), An Introduction to Continuum Mechanics. Academic Press, Orlando.
Horgan, C.O. and Murphy, J.G. (2016). Extension or compression induced twisting in fiber-
reinforced nonlinearly elastic circular cylinders. J. Elasticity 125, 73-85.
L-Shih Liu (1982). On representations of anisotropic invariants. Int. J. Eng. Sci. 20, 1099-1109.
Man, C.-S. and Goddard, J. (2017). Remarks on isotropic extensions of anisotropic constitutive
functions via structural tensors. Math. Mech. Solids (in press).
Zheng, Q.-S. (1994). Theory of representations for tensor functions—a unified invariant approach
to constitutive equations. Appl. Mech. Rev. 47, 545-587.
Stress response in the presence
of local constraints
on the deformation
Quite often the conditions of the problem at hand and the nature of the material are such
that the deformation conforms very nearly to one or more constraints on its gradient. Thus,
for example, rubber-like solids are nearly incompressible and so deform isochorically pro-
vided that, in doing so, no boundary data are violated. From the phenomenological point of
view, such behavior is due to the significant energetic cost associated with deviations from
a locally isochoric mode of deformation. Indeed, this cost is often so high that if position is
assigned on the boundary in such a way as to require an overall volume change, the material
will rupture locally, rather than maintain a smooth, and necessarily non-isochoric, deform-
ation. Roughly, isochoric deformations are energetically optimal in rubber-like solids. One
may imagine that the application of an arbitrary pressure to such a solid would not affect its
deformation to any significant degree, and not at all in the limit of perfect incompressibility.
Conversely, the pressure acting on the material is not determined by its deformation. In the
same way, deformations of directionally reinforced solids, such as fiber composites, may
be idealized as being inextensible along the local direction of reinforcement, the uniaxial
stress along this direction being unrelated constitutively to deformations of the material
that are consistent with the constraint. These are examples of useful constraints in which
the deformation gradient is restricted a priori. Because they narrow the class of admissible
deformations, they invariably aid in the analytical treatment of problems. This is illustrated
in Chapter 7.
For example, if the material is incompressible during a time interval Z then we have
J(t) = const. for t € I. Differentiating and using J; = F* yields F* - F = 0. This imposes a
restriction on F, and so the argument leading from eqn (3.10) to (3.11) no longer holds.
Evidently, the manner in which the stress is related to the deformation is thus modified.
Our purpose, here, is to determine how it is modified. The subject is not especially well
treated in the text and monograph literatures, and so we present a systematic discussion of
it here, based on the Lagrange-multiplier theorem.

CONSTRAINT MANIFOLDS AND THE LAGRANGE MULTIPLIER RULE | 43
6.1 Local constraints

We consider local constraints at material point p, as perceived by observer O, of the form
¢« (F) = 0. (6.1)
It is natural to assume that all observers, O* included, agree that a constraint is in force and
thus to require that an expression of the kind
Qe (F*) =0 (6.2)
hold whenever eqn (6.1) does. Because constraints reflect the nature of the material under
certain conditions, on which O and O* are presumed to agree, we may follow the example
of the strain—-energy function in Problem 1 of Chapter 3 and assert that @*,(F*) = , (F).
If you have worked through that simple exercise then you know that this implies $,(F) =
(QE) for all rotations Q and that the latter is equivalent to
¢.(F) = ¥.(C); (63)

for some function y,. The symmetry of C implies that there can be no more than six inde-
pendent constraints at any material point. For, otherwise the constraints would overspecify
the components of C.
Problem
Show that there can be no non-trivial constraints of the form A - F = B with A and
B fixed.
For example, incompressibility requires that the value of J at a material point be the
same in all configurations of the body. This is equivalent to the requirement that detC
be independent of the deformation. Then, since C = I, when the body is undeformed, the
constraint function is given by
w,(C) = detC - 1. (64)
In the case of inextensibility, |FE| is unaffected by deformation, where E is the unit-

tangent vector to an inextensible material curve in «. Its value is unity in that configuration,
and the constraint function is thus given by
¥.(C) = VE@E-C-1. (65)
6.2 Constraint manifolds and the Lagrange multiplier rule

Evidently eqn (6.1) defines a manifold M in the nine-dimensional space Lin’, just as an
equation of the form F(x) = 0 defines a surface in three-space. If there are n constraints
44 | STRESS RESPONSE IN THE PRESENCE OF LOCAL CONSTRAINTS
OF
Figure 6.1 Local geometry of the constraint manifold
@(F) then F € M, where M = '_,M, and M, = {F : 6(F) = 0}. M is called the

constraint manifold. Since J(= det F) is non-zero in any deformation we require that 0 ¢ M.
Therefore, M is nota linear space.
On any curve F(u) € M, the stress and strain energy are related by
(W(F)) =P-F, (66)
as in eqn (3.9), where F’ € Ty, the vector space tangent to the constraint manifold at
the point F(u). We assume that each point F on M is the center of an open ball B in
Lin*. Further, for any possible process we have @’(u) = O and therefore ob. - F = 0 forall
F’ € Ty, where the gradients are evaluated at the point F(u) and we have suppressed the
subscript « for clarity. This implies that each of the gradients ¢? is orthogonal to T,y (see
Figure 6.1).
By definition, the constraints are independent if and only if the set {#{} is linearly
independent; that is, ifand only if the linear equation
da? = (67)
holds with all a, = 0. In this case, {¢°} is a basis for the orthogonal complement to Ty.
The tangent space and its orthogonal complement together comprise the nine-dimensional
translation space of Lin*, the linear (vector) space consisting of all differences that can be
formed from the elements of Lin*. We have already seen that this is just Lin, and so
Lin = Ty ® Span{p}. (68)
The notation @ identifies Lin as the direct sum of the vector subspaces Ty, and Span{o-’},
meaning that every element of Lin is expressible as the sum of elements of the two vector
spaces comprising the direct sum. Of course, direct-sum decompositions are not unique.
Two that come immediately to mind in the case of Lin are Sym ® Skw, the direct sum of
CONSTRAINT MANIFOLDS AND THE LAGRANGE MULTIPLIER RULE | 45
the linear spaces of symmetric and skew tensors, and Sph @ Dey, in which the factors are,
respectively, the linear spaces consisting of the spherical and deviatoric tensors.
Thus, anyA € Lin such that A- F’ = 0 forall F’ € T,y satisfies
A= y 1,2 (69)
for some scalars A;. Equation (6.6) and the chain rule imply that P — Wy furnishes an
example of such tensors, and hence that
P=W,t+ - 10°, (6.10)

i=)
where the Lagrange multipliers A, may depend on the material pointp and the time t, which
have played passive roles in course of the derivation. Because the domain of W is M, the
derivative W,, defined by W’ = W, - F’,, is to be interpreted as an element of the dual space
of T 44, which may be assumed to coincide with T,, itself. We do this all the time, for ex-
ample, as when we ignore the distinction between three-dimensional Euclidean space and
its dual. In contrast, the functions @(F) are defined on Lin* and their gradients belong
to Lin.
Typically, one wishes to compute the gradient W, explicitly via the chain rule, as in Part
3 of the Supplement. When doing this, the fact that the associated F’ is not an arbitrary
element of Lin may prove to be an inconvenience. We may effectively sidestep this issue
by using a smooth extension W of W instead. The extended function has the ball B as its
domain, is differentiable there, and by definition, agrees with W on M. Differentiating the
consequent equation yields (W)’ = W’ at all points F(u) € M, so that
(W, - We) -F =0, (6.11)
and therefore We- Wp € Span{o.?}. The use of W in place of W in the formula (6.10)
for the stress thus amounts to an adjustment to the (as yet unknown) Lagrange multipliers.
Moreover, if W is another extension, then, because it agrees with W on M, it follows that
eqn (6.11) remains valid with W substituted in place of W, and eqn (6.10) continues to
hold with possibly different multipliers. Therefore, any smooth extension may be used
without loss of generality. The obvious choice, and the one tacitly made in all treatments
of constrained elasticity, is
W(F)=W(F), for Fé Lin’. (6.12)
That is, the extended function may be taken to be the original function, but now with
domain Lin* rather than M.
Since W is a constitutive function, it is subject to invariance requirements. Proceeding as

before, we conclude that
W(F) = G(C), andtherefore W(F) = G(C) (6.13)
for some G defined on the image of M in Sym’*, with
G(C) = G(C) forall Ce Sym’, (6.14)
in accordance with eqn (6.12). We compute P = FS where
1 uy
5S = SymGe + So asymy? (6.15)

i=l
in which ¥(C) = 6(F). Here, we have used the formula eqn (5.4) of the Supplement.
This is justified because the extended strain—-energy function is defined for F € Lin* and
the induced F’ is an arbitrary element of Lin. Accordingly,
P =2F smc. + \; asm . (6.16)

irl
In the example of incompressibility we find, from eqn (6.4), that
Syme = (det C)C* (6.17)
and for inextensibility we use |FE| = /E @ E- Cwith the chain rule to derive
Symic = : |FE[E@E. (6.18)
The gradients of the associated functions of F are obtained by using the formula
dr = 2E(Symc), (6.19)
which is derived just as eqn (3.20) was derived. Then, in the case of incompressibility,
= 2(det C)F", (620)
whereas, for inextensibility,
dy = |FE|' FE@ E, (621)
It is easy to verify that these are linearly independent elements of Lin and, thus, that the two
constraints are independent.
MATERIAL SYMMETRY IN THE PRESENCE OF CONSTRAINTS | 47
When using eqn (6.10) to compute the stress, all gradients are evaluated at F ¢ M.
Accordingly, for incompressibility and inextensibility we have
P=W,p-pF' and P=W,+Ae@E, (6.22)
respectively, where p and A are the Lagrange multipliers and e = FE is the unit tangent to
the inextensible curve after deformation. The associated Cauchy stresses are
T=(W,)F'-pl and T=J'(W,)F'+Te@e, (623)
where T = J”’A. This yields the interpretation of the Lagrange multipliers as a pure pressure
in the first instance, and a uniaxial stress in the second. Recalling our earlier discussion, the
fact that these are unrelated to the deformation is only to be expected. If both constraints
are operative, then of course, the stress is obtained by simply adding the constraint terms in
accordance with eqn (6.10).
To evaluate the Lagrange multipliers, which at this stage are arbitrary scalar functions of
x and tf, we append the n constraint equations to the system consisting of the equations of
motion and the boundary and initial conditions. This yields a formally determinate problem
consisting of3 + 1 equations for the three components of the deformation function x (x, f)
and the n Lagrange multipliers. In this way the multipliers are found to be influenced by
material constitution only indirectly via the initial-boundary-value problem at hand.
6.3 Material symmetry in the presence of constraints

Recall that R € g,(), if and only if, W, (F; x) = W, (FR; x). For constrained materials, this
statement makes sense only ifF € M implies that FR € M. Then,
Regu ={R: FRe M wheneverF € M}, (6.24)
and as the statement R € g,j) makes sense only ifR € gj, it follows that
Sup) C 8m- (6.25)
Following Podio-Guidugli (2000), we say that the material symmetry is compatible with
the constraint.
For example, in the case of inextensibility we have M = {F: |FE| = 1} in which E(x)
is a field of unit vectors in x. Then,
&Miinet) = (R: |(FR)E| = 1 whenever |FE| = 1}. (6.26)
For incompressibility, MM = {F: det F = 1} and
SM(incomp) = {Ri det(FR) = 1 whenever detF = 1} = U. (627)

In the case of isotropy, we have g, = Orth* and so R € g,(,.) implies det(FR) = det F;
thus, R € gadcincomp) and isotropy is compatible with incompressibility. That is, an isotropic
material could be incompressible, although, of course, not every incompressible material
is isotropic. In fact, since g, C U is always true, it follows that any kind of symmetry is
compatible with incompressibility. On the other hand, for arbitrary R € Orth* we have
|(FR)E| = VE: (R'CR)E = /RE- C(RE) 4 VE- CE = |FE|, (6.28)

and SO Be (ise. g ZM(wexi)} isotropy is not compatible with inextensibility and so an isotropic
material cannot be inextensible in a fixed direction.
In the case of transverse isotropy we have
Bettas) = {Ri Re Orth’ and RE = E}. (6.29)
Then, |(FR)E| = |FE| whenever R € (ras), implying that gy (iraus) C @a¢inet)} transverse
isotropy is compatible with inextensibility.
As an example, we cite the case of incompressibility and isotropy. In this case, the natural
extension of the strain—energy function is
U* (Li, hh; x) = Uh, L, 1;x), (6.30)
in which [,, are the usual invariants of C, defined for all C € Sym* in accordance with
eqn (6.14). The Cauchy stress is then given by (see eqns (4.39) and (6.23), part 1)
T = 2(U* + 1,U3)B
- 2U;B’ - pl, (631)
where p is the Lagrange multiplier. Alternatively, using eqn (4.50) with the extension
w" (i, 43x) = w(iy i, 1; x), (632)
in which j,, are the invariants ofU € Sym*, we have eqn (4.56) in which
o = (wi +iwy)I-wiU - pu". (6.33)
Problems
1. Howis the argument leading from the work inequality to the existence ofa strain-
energy function affected by a constraint of the form ¢(F) = 0? Are there any
restrictions on the extended function W’(F) (the extension of W(F) from M
to Lin*) arising from the requirement T = T‘?
2. Find the form of the constitutively indeterminate Cauchy stress for the following
local constraints:
PROBLEMS | 49
(a) A laminated material formed from sheets of stiff paper interspersed in a

soft matrix material. Take the sheets to be continuously distributed paral-
lel planes in some reference configuration. Let the planes of the sheets be
spanned by an orthonormal set {E,,E,}. The constraints are then given
by E, . CE, = 0 and E, - CE, = 1 = E, - CE,. Show that these imply there
can be no extensional or shear strain in the plane of the sheets, but that
transverse normal and shear strains are permitted. Show that a mater-
ial constrained in this way cannot be isotropic. Could it be transversely
isotropic?
(b) The body is laminated as in (a) but the constraint is now that the planes
experience no change in local surface area in any deformation of the body.
Show that the projection of the Cauchy stress onto the tangent plane of the
deformed image of a typical lamina is constitutively indeterminate. Can you
describe this in physical terms? What kind of material symmetry is consistent
with this constraint?
. Consider an incompressible elastic material that is homogeneous and isotropic
relative to the chosen reference configuration. Take this configuration to be a unit
cube with edges parallel to E,, and let the deformation be homogeneous and iso-
choric with gradient F = 1,e, @ E, + A,e, @ E, + (A,A,)'e; @ Es, where A, A,
are positive constants and {e,} = {E,}. Let the tractions vanish on planes with
normals -te, and ey, and let the forces on planes with normals te, be +fe,,
respectively.
(a) Find the constraint pressure in equilibrium in the absence of body force.
Show that A, = 4;'” furnishes a solution, no matter what the strain-energy
function may be.
(b) Obtainf as a function of A, using the so-called neo-Hookean strain-energy
function defined by W = 4 (I, ~ 3), where u is a positive constant (which
can be shown to be the shear modulus in the case of small strains). This sim-
ple function is quantitatively accurate for rubber if the principal stretches lie
in the approximate range 1/2 < A, < 2.
4. Generalize the result of Problem 6 in Chapter 4 to the case of incompressibility.
5. Howis Problem 7 of Chapter 4 affected by the constraint of incompressibility?
6. Show that all observers agree on the values of the Lagrange multipliers, i.e., that
they are absolute scalars.
. Consider an incompressible elastic solid that is homogeneous and isotropic rela-
tive to the chosen reference configuration. Suppose that a material described by
this relation is stressed in its undeformed state, i.e, T is non-zero when F =I.
In this case, we say that the material is residually stressed. Suppose the residually
stressed undeformed configuration of the body to be in equilibrium without body
force. Also, suppose the traction acting on the undeformed body vanishes on a por-
tion of the boundary. Show that the residual stress must then vanish identically
everywhere in the body.
$0 | STRESS RESPONSE IN THE PRESENCE OF LOCAL CONSTRAINTS
REFERENCE
Podio-Guidugli, P. (2000). A primer in elasticity. J. Elasticity §8, 1-104.
FURTHER READING
Fleming, W. (1977). Functions of several variables. Springer, Berlin.

Some boundary-value problems
for uniform isotropic incompressible
materials
We have already made mention of the fact that analytical solutions to the equations and
boundary/initial conditions of nonlinear elasticity theory are as rare as hen’s teeth. The
youthful student might feel some justification in believing that they are, thus, unworthy
of serious study and certainly unworthy, in the digital age, of the often substantial effort
required to find them. While it is true that the quest for analytical solutions often requires
the investigator to limit attention to rather contrived problems of limited relevance, it is
also true that, once secured, they prove to be of the greatest benefit to those seeking to test
constitutive equations (for the strain—energy function, say) against empirical data. This is
our main justification for considering some simple equilibrium deformations that can be
reproduced with relative ease in the laboratory. The best source for analyses of this kind is
Ogden (1997), which goes well beyond the present treatment.
7.1 Problems exhibiting radial symmetry

with respect to a fixed axis
7.1.1 Pressurized cylinder
Take the reference configuration of the elastic material to be the right circular cylinder de-
scribed in terms of cylindrical polar coordinates by A < R < Band 0 < @ < 2m. If the
cylinder is subjected to uniform pressures at its cylindrical boundaries, and if the material
constituting the cylinder is uniform and isotropic, then one has the intuition, based on the
cylindrical symmetry of the problem, that cylindrically symmetric deformations should be
possible in equilibrium. These are described by a map of the form y = x (x) (dropping the
subscript « for convenience), where
x= Re(0)+Zk and y=re,(@) + zk, (7.1)

© David J. Steigmann, 2017, Published 2017 by Oxford University Press.
$2 | SOME BOUNDARY-VALUE PROBLEMS FOR UNIFORM ISOTROPIC
with
e,(@) =cosOe,+sinfe, and k=e,, (7.2)
where {e;} is a fixed orthonormal basis, and
r=r(R) and z=Z. (73)
This deformation is completely specified by the single function r(R). To visualize it, we
observe that it maps a circle R = C, say, to the circle r = c, where c = r(C) (see Figure 7.1).
To set up the problem of determining r(R), we obtain the deformation gradient and
substitute into the relevant constitutive equation. The result is then substituted into the
equation of equilibrium and an attempt is made to integrate it. To this end, we use the chain
rule dy = Fdx, where
dx = dRe,(0) + Reg(@)d0 +dZk and dy =dre,(0)+re,(6)d0+dzk, (74)
where
eo(9) = e'(0) = —sinGe, + cosPe, =k x e,(0). (7.5)
These expressions are entirely general. For the present rather simple class of deformations
we use (7.3) to re-write the second of them as
dy = r'(R)dRe,(0) + reg(0)d0 + dZk. (7.6)
Figure 7.1 Cross section of a cylinder

PROBLEMS EXHIBITING RADIAL SYMMETRY WITH RESPECT TO A FIXED | 53
We want to write this as a tensor operating on the vector dx; that tensor may then be identi-
fied with the desired deformation gradient. To achieve this we note, from eqn (7.4) part 1,
that
dR=e,(0)-dx, RdO=e,(0)-dx and dZ=k- dx, (77)
and, hence, that
dy = r'(R)e,(0)[e,(0) - dx] + (r/R)eo()[eo (8) + dx] + k(k + dx), (78)

from which we simply read off
F=r(R)e,(0) @e,(0) + (r/R)eo(0) @e (0) +k@k. (79)
The restriction given in eqn (1.3) reduces to
J=r(R)(r/R) > 0, (7.10)
which implies that r(R) is an increasing function and, hence, that concentric circles R = C,,,
with C, > C,,are mapped to concentric circles r = c,2, respectively, with c, > c,. With this
it is trivial to obtain the polar decomposition
U=F, ReI. (7.11)
At this stage it is apparent from eqn (7.10) that considerable simplification is achieved if
the deformation is isochoric, as it must be if the material is incompressible. Accordingly, we
consider incompressibility and integrate eqn (7.10), withJ = 1, to obtain
r -@ =R’-A’, where a=r(A), (7.12)
which could have been guessed at the outset. Furthermore, eqn (7.9) furnishes
F = (R/r)e,(6) @ e,(0) + (r/R)ee(8) @ e,(6) +k @k,

and B=(R/r)’e,(0) @ e,(@) + (r/R)*e,(0) @e,(80) +k @k, (7.13)
where B = FF‘ is the left Cauchy—Green deformation tensor.

The representation given in eqn (6.31) for the Cauchy stress in an incompressible,
isotropic material leads to
T=T-pl, (7.14)
where the constitutively determined part of the stress, T, is of the form
T = T,(r;a)e,(8) ® e,(8) + Too(r; a)eo(8) ® e9(8) + T.(ra)k@k, (7.15)

54 | SOME BOUNDARY-VALUE PROBLEMS FOR UNIFORM ISOTROPIC
in which T,,(1; 4), etc, are obtained by inserting eqn (7.13) into eqn (6.31), while use
has been made of eqn (7.12) to convert functions of R into functions of r, depending
parametrically on the unknown constant a.
In the absence of body forces, the equation to be solved is divT = 0, which is equiva-
lent to
gradp = divT. (7.16)
Problem
Prove the rule u- divA = div(A'u) - A: gradu, where grad is the gradient with
respect to position y, and use it to work out the coefficients in the expression
divA = (e, - divA)e, + (e, - divA)ep + (k- divA)k, where A= A,e, @e, + Ae, @
@g t....
Accordingly, we have
~ d~ Il .~ ~~
divT = |<7. +-(T,, - iw)| e,(9), (7.17)
dr r
which, in conjunction with
gradp = Op/dre,(@) + F'dp/dOe,(0) + dp/dzk, (7.18)
leads us to conclude that 0p/00 = 0 = Op/dzand

d d~ 1. «
GP) = FT + (Tn — Too) = fri a); (7.19)
and, therefore, that
p(r) = p(a) + / f(x a)dx. (720)

The boundary conditions at the cylindrical generating surfaces r = a, b, with exterior unit
normals n = Fe,, respectively, are
-Te,=P,e, at r=a, and Te,=-P,e, at r=b, (7.21)
where P,, are the pressures acting there (not to be confused with the boundary values of p).
From eqns (7.14) and (7.15), these are seen to be equivalent to the two relations
P, =p(a)-T,(a;a) and P, = p(b) - T,.(b; a), (7.22)
where
ba = B- A’. (723)
PROBLEM | $5
Combining these with eqn (7.20) finally delivers
b
AP = T,,(b; a) - T,,(a; a) - / f(y a)dr, (7.24)
where AP = P, - P, is the net inflation pressure. For a given strain—energy function, this
generates the inflation pressure corresponding to any given radius a.
Problem
Complete the analysis using the so-called neo-Hookean strain—energy function de-
fined by U = > (I, - 3), where 4, a positive constant, is the shear modulus of the
material. This is normalized so as to vanish in the absence of strain, at I, = 3. Show
that the Cauchy stress in a neo-Hookean material is given simply by
T = —pl + wB. (7.25)
The neo-Hookean model has an interesting history. It actually has a basis in statistical
mechanics (see Treloar (1975) and Weiner (2002)), and its relative simplicity makes it
attractive to those interested in analytical work. In particular, it is completely specified by
the single parameter ju. It is also rather well behaved from the mathematical point of view,
as we shall see later in Chapter 9. However, while it furnishes a good quantitative model of
rubber for moderate principal stretches lying in the approximate range (;, 2), its behavior
deviates substantially from that of rubber outside this range. If you have done the preceding
exercise about the response of cylinders, you will have observed that it yields a reasonable
relationship between inflation pressure and deformed inner radius only when the latter is
small-to-moderate. It is, therefore, predictive only for small to moderate strains. In fact,
from the empirical point of view, it is no better than the purely phenomenological Varga
strain—-energy function (see Varga, 1966) defined by
w(is,i,) = 2u(i, - 3), (7.26)
with the same A.

To justify the interpretation of the parameter jz as the shear modulus, we digress to
consider the simple-shear deformation defined by
y=xt y(E,+x)e, (7.27)
where y, the amount of shear, may be any real number, but is assumed here to be inde-
pendent of x. The effect of this deformation on a unit block of material is illustrated in
Figure 7.2.
It is a special case of the deformation analyzed in Problem 7 of Chapter 4.
Figure 7.2. Simple shear of a block
The deformation gradient and Cauchy—Green deformation tensor are easily seen to be
given by
F=I+ye,@E, and C=I+y(E, @E,+E,@E,)+y’E, @E,. (7.28)
This is an example of a homogeneous deformation. Homogeneous deformations are char-

acterized by the property that the deformation gradient is uniform, i.e., independent of x,
Here, the invariants I,, are determined by the number y, whereas J = 1. In particular,
I, = 3 + y’, and the neo-Hookean material yields the strain energy W = W(y), where
A 1
W(y) = shy". (729)
This model thus behaves like a linear spring, with stiffness 4, in simple shear. The energy
change associated with a change in the shear is
(W) =P-F=(P-e, @E,)y = ty, (730)

where T = e, - PE, is the shear stress, the projection of the (Piola) traction onto the plane
with normal e,. Accordingly, t = zy and the ratio of shear stress to the amount of shear—
the shear modulus—is just 4, as claimed. For a general isotropic material, this ratio depends
on y, in the manner of a nonlinear spring. However, for the neo-Hookean material, the
shear response is characterized by a constant modulus and is thus linear; hence, the name
neo-Hookean.
A strain—-energy function that is good over virtually the entire range of feasible deform-
ations of rubber has been given by Ogden (1997). It is rather unwieldy for analytical work
and, thus, not discussed in the present chapter, but has emerged as the formulation of choice
for numerical simulations. The relevant details can be found in Ogden (1997).
PROBLEM | $7
For general isotropic elasticity, including incompressibility, there is a relationship among

the stress components in simple shear, which is universal in the sense that it does not in-
volve the properties of the material at hand. This is given simply by specializing the result
of Problem 7 of Chapter 4. It predicts that a non-zero normal stress difference always ac-
companies simple shear. Furthermore, using it one can easily show that the normal stresses
vanish faster than the shear stress as the amount of shear vanishes; the normal stress effect is
thus inherently nonlinear, which is why one never hears about it in linear elasticity theory.
This prediction conforms to empirical observation and is one of the major successes of non-
linear elasticity theory. However, the reader is cautioned that true simple shear is practically
unattainable in the laboratory and thus mainly of theoretical interest. Its important features
may, however, be replicated in other deformations that are experimentally feasible.
Problem
Show that the Varga material has a nonlinear simple shear response, and that the
parameter yZ is the slope of the t vs y curve at y = 0. Thus, it characterizes the
linear part of the shear response of this material at the unstressed state.
Simple shears, and homogeneous deformations in general, are simpler than the cylin-
drical deformation considered thus far in that, for uniform materials (W is not explicitly de-
pendent on x), they deliver uniform constitutively determined stresses whose divergences
vanish identically. This yields divT = -gradp in the case of incompressibility; therefore, p
is uniform if the body is in equilibrium without body forces. The complete stress is then
uniform and, thus, determined entirely by boundary data, which must, of course, be such
as to admit homogeneous deformations in the interior. Otherwise, the premise is false and
the (non-homogeneous) deformation must be found by solving the nonlinear differential
equations.
Before leaving deformations of cylinders, we discuss a special case for which the deform-
ation is homogeneous. Thus, consider the case ofa solid circular cross section described by
A = Oand suppose the deformation is such that r(0) = 0. Then eqn (7.12) yields r = Rand
eqn (7.13) part 1 reduces to F = e,(@) ® e,(@) + e9(9) ® e9(9) +k @k =I. There is no
deformation, no matter what the external pressure may be. To make the problem a bit more
interesting, we relax the assumption z = Z (cf. eqn (7.3) part 2) and replace it by
z=AZ, (731)
where A is a constant. An easy calculation yields
F=r'(R)e,(0) @ e,(6) + (r/R)es(0) @ eo(0) +AK@k (732)

and
J =Ar'(R)(r/R) > 0 (7.33)

in lieu of eqns (7.9) and (7.10). Incompressibility, taken together with the condition on
r(0), now furnishes
r(R) = R/VA, (7.34)
provided that A > 0. This, in turn, ensures, as before, that r(R) is an increasing function.
Furthermore, eqn (7.32) becomes
F=Ak@k+A'“"(1-k @h), (735)
which is independent of x. The deformation is, therefore, homogeneous and, hence so too,
the constitutive part of the stress if the strain—-energy function is uniform.
For the neo-Hookean material, the stress is (see eqn (7.25))
T = (ud? - p)k @k+ (UAT -p)(I-k@k). (736)
To find p, which is uniform if the cylinder is in equilibrium with vanishing body force, we
need a boundary condition. Suppose, for example, that the lateral surface R = A(r = a) is
traction free. Then, Te, vanishes at r = a, and hence, in this case, everywhere in the body,
yieldingp = wA“ and
T=uV -AYK@kK (737)
The stress in the bar is uniform and uniaxial, and varies with the axial extension.
Problems
1. Obtain the uniaxial force-extension relationship for the neo-Hookean bar, and
obtain an expression for Young’s modulus—the slope of this relationship at A =
1—in terms of ju.
2. Show that the foregoing solution is valid in all neo-Hookean cylinders, regardless
of section connectedness or shape, if the lateral surface is traction-free.
3. Biological tissues are characterized by a load-bearing microstructure consisting of
collagen fibers that are “crimped” in the form of wavy curves in their relaxed state.
As the tissue extends, the collagen fibers straighten, or “decrimp,” by unbending
until they are more-or-less straight; the load required to achieve this is fairly small.
Once the decrimp phase is complete, further extension of the tissue requires ac-
tual stretching of the collagen fibers. This requires relatively large force compared
to that required for decrimping. To model this behavior on the macroscale, we
require a strain-energy function which is such that the uniaxial force-extension
curve is nearly horizontal for small-to-moderate stretches, while growing rapidly
for larger stretches.
Consider the candidate strain—energy function
U= ;, lest ch -3)}-1], (738)

PROBLEMS | 59
where yt and y are positive material constants (with 4 having dimensions of

force/area, while y is dimensionless). Obtain the force—extension response for an
incompressible cylindrical bar subjected to zero traction on its lateral surface, and
show that its qualitative properties match the foregoing description of bio-tissue.
4. Consider a cylindrical body occupying the reference configuration defined by
A<R<B,-L/2 < Z < L/2,0 <@ < 2m. Suppose the cylinder is turned in-
side out (everted) so that, after deformation, it occupies a new cylindrical region.
Thus, the deformation maps the material point with reference position
x = Re,(@) + Zk (739)
to its final position
y =r(R)e,(0) + z(Z)k, (7.40)
where a < r < band z(Z) = —Z (ie. the cross sectional plane Z = L/2 in the
reference configuration is mapped to the plane z = -L/2 in the current configur-
ation, etc.). Also, the inside of the reference cylinder is mapped to the outside of
the deformed cylinder, and the outside is mapped to the inside. Thus, r(A) = b
and r(B) = a.
(a) Find the function r(R) meeting the stated boundary conditions if the de-
formation is isochoric.
(b) Compute C = F'F and obtain U by inspection. Using your result, compute
the rotation factor Rin the polar decomposition F = RU.
(c) Can this deformation be maintained in equilibrium in an incompressible
isotropic material with zero tractions on the lateral surfaces?
7.1.2 Azimuthal shear
Imagine a hollow cylinder welded to a rigid shaft at its inner radius, R = A, and a rigid cy-
lindrical sleeve at its outer radius, R = B. Fix the sleeve and rotate the shaft about its axis k,
through the angle ®. For uniform isotropic incompressible materials, in equilibrium with
zero body force, one may feel justified in assuming that an interior circle R = C, say, merely
rotates uniformly about the shaft without a change in radius and that different concentric
circles rotate by different amounts. That is, the azimuth changes by an amount, @ say, that
depends only on R. To test the hypothesis, we proceed as before to construct the rele-
vant deformation and stress tensors, and then investigate the possibility of satisfying the
equation of equilibrium. If © is the azimuthal angle prior to deformation we then have
x= Re,(@)+Zk and y=Re(0)+Zk, where 0=©@+@(R). (7.41)

Then
dy = dRe,(0) + Re,(@)d0 +dZk, where d@? =d@+¢'(R)dR. (7.42)
Using eqn (7.4) part 1 with the relevant azimuth yields
dy = e,(0)[e,(O) - dx] + e6(O){[eo(O) + RP'(R)e,(O)] - dx} + k[k- dx], (743)
and, hence,
F=e,(0) @e,(O) + e6(0) @ e,(O) + rP'(res(P) @e(O) +k @k, (7.44)
in which r = R.
Notice that this may be factored as
F=FQ, (745)
where
Q=e,(8) Be,() + eo(8) @ e(O) +k@k (7.46)

is a rotation and
F=1+y(r)e(0) @e,(0) (7.47)
is a simple shear on the e,(@), e,(@) axes of amount y(r) = ré’(r) (compare eqn (7.28)
part 1). This is an inhomogeneous simple shear. Furthermore,J = det F det Q = 1, imply-
ing that the deformation is, indeed, isochoric.
Proceeding with the neo-Hookean material for the sake of illustration, we use eqn (7.25)
together with B = FF’ = EF’; ie.,
B =e,(9) @e,(@) + (1+ y7)es(9) @ eo(8) +k @k+ y[e,(4) @ eg(4)

+e,(0) ®e,(9)], (748)
yielding the stress
T =(u-p)e,(0) @e,(@) + [u(1 + y*) -ples(9) @ eo(0) + (u-p)k@k

+ py [e(8) @ eo(8) + e6(9) ® e,()]. (749)
Among the three scalar equations of equilibrium, the projection e9(0) - divT = 0 proves
to be immediately useful and yields
d 2 id,
0= al? + 78 = age Tie) (7.50)
PROBLEMS | 61
where T.)(r) = zy (r). Thus,
urd’ (r) = t/r (751)
witht = const. This gives
'd Ae 2
oiy=ose [ Suo+ 5 ( ), (7.52)

BJ, ©
The constant t is determined by imposing the condition 6(B) = 0; thus,
T A’B’
(7.53)
uBR A
on 0
and
The deformation is now completely determined.

We have not used the remaining components of the equilibrium equation. We do so now,
writing the latter in the form
d 1 d 2
gradp = divB = w {| 52. + —(B,, - 2a )| e,(0) + (<2. + “By =(0)| » (755)
dr r dr r
where
B,=1, B,-Bs =-y* and Be, =y. (7.56)
Combing the last of these with eqn (7.51), we find that
45 +28 -if(p y-.4 “\=0 ()

dr” pe ag ” Pdr bh 7
Consequently, eqn (7.55) furnishes 0p/00 = 0 = dp/dzand
d
5) = -py?/r. (758)
This determines the constraint pressure distribution apart from a constant.

The solution may be used to generate the overall torque-twist relation of the annular
cylinder. To see this we compute the traction transmitted by the material to the central
shaft. This is
Te,(9),,-. = ~p(A)e,(0) + wBe,(@) = [4 - p(A)]e,(@) + (t/A*)eo(9), (7.59)

62. | SOME BOUNDARY-VALUE PROBLEMS FOR UNIFORM ISOTROPIC
and so the torque, per unit axial length, transmitted by the shaft to the material is
1 an
m=- [ / y, x {Lu — plA)le,(8) + (t/A*)e9(8)}.Ad0dz, (760)

0 0
where
y, = Ae,(O) + Zk (761)
is position on the interface between shaft and material. Expanding the cross product, and
using the periodicity of e,(@) and e, (0), we finally derive m = m(®)k, where
A’B
m(®) =-2NT = 4x u® (3) , (7.62)
The linearity of this relationship, which is atypical, is an artifact of the linearity of the
neo-Hookean response in simple shear. Importantly, this prediction is insensitive to the
pressure field, which, as we have seen, is determined apart from a constant. That is, the
boundary-value problem, as stated, determines the stress apart from a constant pressure
field and, thus, yields a non-unique stress field. To obtain a unique stress, it is necessary to
impose one additional scalar condition. One choice is the net axial force transmitted across
a cross section.
Problems
1. Consider equilibrium without body force and assume a deformation of the form
y=xtw(nk (7.63)
where r € [a, b] is the radius from an axis of symmetry prior to deformation.

(a) Show that the deformation gradient is
F=l+w(nk@e,. (7.64)
(b) Find w(r) for a neo-Hookean material, assuming the boundary conditions
w(a) = W, w(b) = 0.
(c) Compute the traction on the inner boundary and determine the allowable
range of values of W, if the bond at r = a fails at a critical value of the shear
stress.
2. Consider the deformation
y=x+w(@)k, (7.65)
PROBLEMS | 63
where @ is the azimuthal angle in a cylindrical polar coordinate system in the

reference configuration.
(a) Show that the deformation gradient is
F=l+r'w(@)k@ eg. (7.66)
(b) Show that the most general function w(@) for which the principal invariants
1,33 of C are independent of @ is of the form
w(9) = AO +B, (7.67)
where A and B are constants. Can you interpret this deformation in physical
terms?
(c) Using the expression for the stress in an incompressible isotropic material,
discuss the problem of maintaining this deformation in equilibrium without
body force, using a reference configuration in the form of a right circular cy-
linder with annular cross section ofinner and outer radii, a and b. Specifically,
is this deformation possible if the tractions are zero at the inner and outer
surfaces? Are there any restrictions on the strain—-energy function in this
case?
7.1.3 Torsion of a solid circular cylinder

In this deformation an entire cross section Z = const. is rotated about the axis of the cylin-
der, without expansion or contraction, by an amount that depends on the value of Z. We
assume this dependence to be linear. Thus,
y = Re,(@)+Zk, with 9=0+4TZ, (7.68)
where T is the constant twist, i.e., the rate of rotation with respect to the axial coordinate.
Using eqn (7.4), part 1, with the appropriate azimuth and proceeding as before, we derive
F=e,(0) @e,(O) + e(0) @e,(@) +rte,(9) @k+k@k, (7.69)
in which r = R.,
Once again, this may be factored, this time as
F= FQ, (7.70)
where Qis the rotation encountered earlier, and
F=l+y(re(0) @k (771)
is now an inhomogeneous simple shear on the e,(0), k axes of amount y(r) = rt. Again,
we haveJ = det F det Q = 1 and the deformation is isochoric. We have
B=e,(0) @e,(0) + (1+ y’)eo(P) @ eo(9) +k @k+ y[k® eo(A) + en(6) @ kl],
(7:72)
which we use together with the neo-Hookean model to complete the solution subject to the
condition that the lateral surface of the cylinder be free of traction.
Thus, we solve
gradp = jdivB, (7.73)

subject to
Te,(0)=0 at r=A. (774)
Notice that for eqn (7.73) to have a solution, it is necessary that the right-hand side
have zero curl, i.e., curl(divB) = 0. For general incompressible materials, this condition is
replaced by curl(divT) = 0, where T is the constitutive part of the stress. This, in turn, im-
poses restrictions on the deformation without regard to the reactive constraint pressure
field. The pressure field may then be determined post facto, in principle, by path integration.
The zero curl condition ensures that the result obtained is independent of the path and,
hence, a well-defined function of x.
Returning to the problem at hand, we have
d 1
divB = Fe + -(B, - aw )| e,(9) = -rt’e,(@), (7.78)
r r
yielding <p = —rt’ and, hence,
P(r) = po - SHUr’, 1 2.2

(7:76)
where py = p(0). We have succeeded here, as well as in the previous examples, in generat-
ing the pressure field (apart from a constant) because the curl condition is automatically
satisfied. The Cauchy stress is
T= set r — py, | 1+ uB, (7.77)

1 242
yielding
1
Te,(0) = (Gurr - pot u) e,(@). (7.78)
The constant py is obtained by imposing eqn (7.74), yielding the unique stress field
1
T=p seer -~A’)- 1 1+ uB. (7.79)
PROBLEMS | 65
In linear elasticity, terms that are nonlinear in t are neglected. Doing so here, we find
that T ~ T,,,, where
Ti = Het [kK @ e9(9) + e6(0) @k], (7.80)

which, of course, generates the classical linear shear stress distribution over a cross sec-
tion. This distribution persists in the nonlinear case, but now normal stresses also arise in
response to the twist. This, of course, is just the usual normal stress effect in disguise.
The overall response of the cylinder may be determined by computing the net force on
a cross section and the net twisting moment required to effect the twist. These, in turn,
require the traction
1
Tk = geei(r —A’)k + prtes(6) (781)
acting on a cross section. The resultant force is
2n A
f= [ / Tkrdrd@ = f(t)k, (7.82)

0 0
where
1
f(t) = tur’. (7.83)
Evidently the force is a manifestation of the normal stress effect, vanishing in the linear
theory.
Finally, the twisting moment is
an a
m= / / y X Tkrdrd@ = m(t)k, (7.84)

0 0
where
1
m(t) = 5 fut. (7.85)
This is precisely the same result predicted by linear elasticity, the coincidence again being
due to the peculiar (i-e., linear) behavior of the neo-Hookean material in simple shear.
Problems
1. Verify the formulas for the net force and twisting moment.
2. Show that a straight generator of the lateral surface of the cylinder is deformed
into a helix. Find the ratio of its final length to its initial length.
7.1.4 Combined extension and torsion
The one surprising aspect of the torsion problem, at least for those not previously aware of
the normal stress effect, is the prediction of a compressive axial force accompanying twist.
This is to be regarded as the reaction force supplied by plates welded to the cross sections
at the ends of the bar, arising in response to the restriction that the perpendicular distance
between the plates (the end-to-end length of the bar) remains fixed. This suggests that if
the reaction force is relaxed, then the end-to-end length should adjust accordingly. To in-
vestigate this possibility, we propose a simple adjustment of the foregoing kinematics to
accommodate axial extension. Thus, in place of eqn (7.68), we consider the deformation
y=r(R)e(0)+zk, with @=O+p~z and z=AZ (7.86)
where 7 is the constant twist. Here, the rate of rotation with respect to axial length on the
deformed cylinder, and A is a positive constant. We allow r to be unequal to R to accommo-
date incompressibility; the cross section must adjust to the axial stretch so as to preserve
volume.
The usual procedure generates
F=r(R)e,(0) @e,(O) + (r/R)eg(O) @ eg(@) + rAyes(O) @k+AkK@k. (787)
Incompressibility is not automatic this time; to enforce it, we compute the determinant of
F in terms of the given parameters and set it to unity. The easiest way to proceed is to use
the scalar triple product, or box product, definition of the determinant. Readers unfamiliar
with this should consult the excellent discussion in Chadwick (1976). Thus,
J = [Fe,(@), Fep(@), Fk] = [r’(R)e,(@), (r/R)es(@), rAwes(O) + Ak] = A(r/R)r'(R),

(7.88)
in which the square brackets are used to denote the box product. This is the same as
eqn (7.33), and carries the same result; namely, eqn (7.34) in the case of an isochoric
deformation and a solid section with r(0) = 0. Inserting the latter into eqn (7.87), we
arrive at
B=) 'e,(0) @e(O) + (A +r W?r*)es(O) ® e6 (0)

+VKOk+ rWr[k
@ e9(9) + e9(8) @k]. (7.89)
For the neo-Hookean material we use eqn (7.73), obtaining

nm
gradp = — Br — Beo )e,(@), (790)
implying once again that p depends only on r, with derivative
d
oP = ere. (791)
We then have
p(r) = pat Sve -r), where p,=p(a) and a= A/V). (7.92)
Suppose, again, that the traction vanishes at the lateral surface. Then, Te,(@) vanishes at
r= a,yielding p, = 4/2 and, hence, the unique pressure field
ptr) =u t+ Syne —P)| (798

and with this the stress field is completely specified.
The traction ona cross section is now given by
Tk = (A? - p)k + uryd7eo(8) (7.94)

and 8 generates the twisting & moment m = mk, where (compare
pare eqn
eq (7.85))
1 4742 1 4
m= 3m a Yr = THA y (7.95)
whereas the net force is f = fk, where
2 2 -1 1 2452/2
f=mpa |r’ -2r ~qvha . (7.96)
Relaxing 8 the force has the effect of coupling

pang the extension to the twist, resulting g in
1
V-A = gue, (7.97)
which, in turn, requires that A > 1 whenever y # 0. Thus, the bar extends as it is twisted.
This is, again, just the normal stress effect, variously referred to as the Swift effect or the
Poynting effect, depending on the context. The prediction that extension of a bar is induced
by a twisting moment is corroborated by experiments.
It bears mentioning that we have said nothing about the stability of these equilibria.
In practice, torsional buckling ensues if the twist is large, yielding the possibility of an
alternative deformation ofa slender bar into a helical shape.
7.2 Problems exhibiting radial symmetry

with respect to a fixed point
In this class of problems, the distance of material points from a specified origin changes, but
nothing else, while all points lying on a sphere centered at the origin move radially by the
same amount. Thus,
y=A(R)x, where R= |x| (7.98)

is the distance from the origin prior to deformation. The distance after is simply the radius
r(R) = RA(R), (799)
yielding A as the ratio of the radii, also known as the hoop stretch.
To obtain the deformation gradient, consider
dy = A'(R)xdR + A(R)dkx. (7.100)
As usual, we want this as a linear function of dx, so that we can read off the desired result. To
this end, we differentiate R? = x - x, obtaining dR = u - dx, whereu = R''xis the normalized
radius vector; hence,
F= RX (R)x@x+A(R)I =r (Rou Q@ut+AaA(R\(1-u Quy), (7.101)
where r’(R) = RA‘(R) + A(R). We then have
Jaw (7.102)
and thus require, as before, that r(R) be an increasing function; i.e., /(R) > 0. The polar
decomposition is trivial in this case, yielding R = I, U = F, and the principal stretches
{A} = {r, A, A}. (7.103)
Suppose the material is incompressible and the deformation, therefore, isochoric.

Putting J = 1 in eqn (7.102) yields a simple differential equation, having the unsurprising
solution
p-@=R-A’, (7.104)
where a = r(A). Then, the hoop stretch distribution is

e-wa\'3
MR) = (14 @ ) , (7.105)
and the radial stretch is A, = A™.

For isotropic materials we combine eqns (4.41) and (6.23), withJ = 1, obtaining
T= > Tv, ® v; — pl, (7.106)
in which
T, =A,0@/dA,,__ etc., (7.107)
and, for the present class of deformations, A, = A;(= A); consequently, t, = t;. Using v, =
u thus yields
T = Tu @u+t,(1-u@u)
- pl, (7.108)
which, in the present circumstances, may be written, for uniform materials, in the form
T=h(r)y @y-[p-g(r)IL (7.109)
with
A(r)=r7f(r), f(r)=.-t% and g(r)=t. (7.110)
Then, for equilibrium in the absence of body forces,
0 = divT = div(hy @ y) - grad(p - g). (7.111)
Of course, no one can remember the formula for the divergence in spherical coordinates,
and so we will use Cartesians instead, i.e., divT = T,,e,, where T,, = 0T,/dy,, yielding
(hyiy;),; — pi + Bi = 0. (7.112)
Expanding this using y,; = 6, (the Kronecker delta), y,, = 3 and (-), = 17(-)’y, for any
function of r alone, yields
gradp = [4h + rh'(r) + r'e'(r) ly. (7.113)
This is enough to conclude that p also depends on r alone, with derivative
pr) =r[4h4 rh'(r) + r'e'(r)] (7.114)
Integration and application of suitable boundary conditions thus determines the solution.
Now that we know the constraint pressure depends only on radius, we may re-write
eqn (7.109) in the form
T=F(r)u@ut Gir)I = H(ry @y+ G(r)L (7.115)
where
F(r)=ti-ty G(r) =ty H(r) = r°F(r) (7.116)

and
h=t-p, bh=m-pl=h) (7.117)
are the (principal) radial and hoop components of the Cauchy stress. Noting that
eqn (7.115) resembles eqn (7.109), we proceed immediately to obtain the equilibrium
equation
4H +rH'(r) + 7'G'(r) = 0. (7.118)
This is easily seen to be equivalent to
(F + G)' +2r'F = 0, (7.119)
which may be converted to the more recognizable form
~d 44+ —(t,-t)
2
=0. (7.120)
dr r
There is just one non-trivial equilibrium equation to be solved for the single unknown
function A(R).
7.2.1 Integration of the equation

Actually, it proves convenient to use stretch as the independent variable. First, we define
@(A) = w(A7,A,A). (7.121)
Using the chain rule,
(A) = (8w/9A,)ddy/dd + (Be/IAq)dda/dd + (BW/9Ag)dA5/dd

= 20w/dA, -2470@/8A,, (7.122)
we find that
ty — t, = A\Ow/BA, — A,9e/BA,
1
=27dw/9A, -2 500) + 1 80/8%
= -A-A'(A).
1 Ar
2
(7.123)
Equation (7.120) then yields
d = -@'(A)
—t, A, A
= R'O'(A), (7.124)
dr r
which may be reduced, using
d 1d d
dr’
—t=——t,=N—t,
A, dR’ aR" .
5)
to
d
RM ahh =@ (A). (7.126)
To convert the derivative on the left, we use eqn (7.105), reaching
, (A? ~ 1)
RA‘(R) = - a (7.127)
We then use this with (7.126) to obtain
d
ah w'(r
= 2) 7 (7.128)
and, thus, reduce the problem to a quadrature.
7.2.2 Pressurized shells, cavitation
Consider, for example, a spherical shell A < R < B, traction-free at the inner radius, and
subject to a negative pressure or suction at the outer radius. This simulates a triaxial state of
stress in a region of material surrounding a spherical hole. The boundary conditions are of
the form
thu = Tu = Pu, (7.129)
in which —P is the assigned pressure, and therefore,
t;=P at A=A, and t,=0 at A=A,, (7.130)
where A, = a/A and i, = b/B, with b? — a’ = B’ — A’, are the hoop stretches at the inner
and outer radii. Taken together with eqn (7.128), this furnishes
p- |
* (A)
DL haa (7.131)
b
yielding P vs a (or b) once the strain-energy function is specified.

This seemingly innocuous result may be used to furnish a graphic illustration of the
power of nonlinear elasticity to predict dramatic phenomena. Consider the case of a solid
sphere, A = 0, and suppose the sphere remains solid, no matter the suction, (a = 0). Then
A(R) = 1,A,, = 1, F = Land T is of the form T = —pl, in whichp is uniform in equilibrium
without body force. Then, unsurprisingly, P(= —p) is indeterminate; the rubber remains
undeformed no matter the suction. This state, therefore, furnishes a solution for all values
of suction.
Experimental evidence (see Gent and Lindley, 1958) suggests that a hole forms spon-
taneously at the center of the sphere when the suction is sufficiently strong. This cavitation
solution corresponds to a > 0, where a = r(0). In this case A, is unbounded and the critical
suction for its sudden onset is
co Ar x
Pat -| ult in (7.132)

The lower limit is explained by observing that
A, = b/B = (1 + @°/B’)"”?, (7.133)
and thus that A, — 1 as a > 0. Of course, all this is sensible only if P.,, is finite, or in
other words, if the integral in eqn (7.132) exists. This may or may not be the case, depend-
ing on the strain—energy function at hand. In the latter case, we conclude that cavitation
is not feasible and, hence, that the trivial solution is the only one available in the class of
deformations considered. Ironically, strain—energy functions are deduced, traditionally, on
the basis of experiments involving finite stretches, and so the theoretical study of cavitation
using the present solution requires knowledge of material response over a far wider range
of deformation than is normally encountered in experiments designed to quantify material
response. Indeed, a rubber band breaks at a fairly moderate value of uniform overall stretch.
However, having said this it must also be noted that rupture is invariably accompanied
by strongly inhomogeneous deformations that may include cavitation on the micro-scale!
All of this is food for thought as one contemplates theory and supporting experiments for
failure mechanisms in rubber.
The post-cavitation response is given simply by
P(a) -| o) dd, (7.134)

»o B-1
where A, is given by eqn (7.133). This bifurcates off the trivial solution at P = P.,,.
Problems
1. Repeat the foregoing for the simpler case of plane strain, ie. for the two-
dimensional radial expansion ofa cylinder.
In problems 2-5 assume the material to be incompressible, isotropic, and neo-
Hookean.
2. Consider the eversion of an incompressible hemispherical shell. Assume the de-
formation is such that the final radius depends only on the initial radius, and that
the elevation angle above the equator is mapped to its opposite value, below the
equator. Show that equilibrium cannot be maintained with vanishing tractions
at the inner and outer constant-radius surfaces. The actual deformation entails a
flaring of the shell as required to meet the zero-traction conditions.
3. Find the critical negative pressure P,,, for the onset of cavitation of a solid sphere
(A = 0). What is the relation between the negative pressure and the cavity radius
a=r(0)?
4. A solid circular bar has initial radius A and length L. Suppose the bar has density
p and let it spin about its own axis at the constant rate w. This spin causes a con-
traction of the bar along its axis. Let u(x) = cos xe, + sin xe,. The corresponding
PROBLEMS | 73
deformation is described by x = Ru(@) + Zk and y = r(R)u(@) + zk, where @ =

© + wt, z = AZ, and Jis the constant stretch along the axis. Suppose the traction
is zero on the lateral surface of the cylinder.
(a) Calculate the resultant forces -Efk on the two ends of the cylinder. Find the
value of 4. corresponding tof = 0.
(b) Obtain a relation between the deformed length of the cylinder and .
. The kinematics of pure flexure ofa block are described by
x=x,E,, y=re,(@) + zk. (7.135)
Here,r = f(x) and@ = g(x,), for some functionsf andg to be determined. Thus,
straight lines x, = const. and x, = const. are mapped to concentric circular arcs
and rays through the origin, respectively (drawa figure). Furthermore, k = e; and
Z = x, so the deformation is a plane strain (take {e,} = {E,}).
The reference configuration is the region defined byA, < x, < A,,-B < x, <
B,-H < x, < H. Suppose there are no tractions applied to the edges x, = Ay, Ay.
The neutral axis is defined to be the vertical line x, = x’ that neither lengthens
nor shortens in the course of deformation. Find r, = f(x), the radius of curvature
of the neutral axis in the deformed configuration. Let a, = f(A,), a. = f(A) and
find a relationship involving a,, a, and r,.
Show that the resultant forces on the edges x, = +B vanish. Calculate the re-
sultant moments of the traction distributions on the edges. What is the relation
between the moment and the curvature x, = 1/r, of the neutral axis?
. Consider the homogeneously deforming unit cube of Problem 3 in Chapter 6,
but now suppose that it is subjected to equi-biaxial loading. Thus, the traction
vanishes on the faces with unit normals +e, while the forces on the faces with
unit normals te, and te, are tfe, and +fe,, respectively, where f > 0.
(a) Show thata solution with A, = A, exists for all suchf.
(b) Using the so-called Mooney-Rivlin strain-energy function defined by W =
C\(, - 3) + C,(h - 3), where C, and C, are given material constants, show
that another branch of solutions, with A, 4 A,, becomes possible when f
reaches a critical value. Thus, there is a bifurcation of equilibria at this value,
at which the solution with equi-biaxial stretch bifurcates to one with unequal
stretches. This behavior has been observed experimentally and has come to
be known as the Treloar—Kearsely instability. We will study the stability of
these solutions later in the course.
(c) Our block is isotropic, by assumption, with respect to its initial configuration
(the unit cube), which we have chosen as reference. Consider a deformation
characterized by unequal biaxial stretch (i.e., 4, 4 Az, A3 = 1/A,A,). What is
the symmetry group relative to this deformed configuration? Is the material
isotropic relative to this configuration?
REFERENCES
Barenblatt, G.I. and Joseph, D.D. (Eds) (1997). Collected papers of R.S. Rivlin, Vols 1 and 2. Springer,
NY.
Gent, A.N. and Lindley, P.B, (1958). Internal rupture of bonded rubber cylinders in tension. Proc. R.
Soc. Lond. A249, 195-205.
Ogden, R.W. (1997). Non-linear Elastic Deformations. Dover, NY.
Treloar, L.R.G. (1975). The Physics of Rubber Elasticity, 3rd edn. Oxford University Press, Oxford.
Varga, O.H. (1966). Stress-strain Behavior of Elastic Materials. Wiley Interscience, NY.
Weiner, J.H. (2002). Statistical Mechanics of Elasticity. Dover, NY.
FURTHER READING
Ball, J.M. (1982). Discontinuous equilibrium solutions and cavitation in nonlinear elasticity. Phil.
Trans, R. Soc. Lond. A306, 5S7-611.
Some examples involving uniform,
compressible isotropic materials
We study some examples of deformation in unconstrained isotropic materials. As the sim-

plification afforded by an a priori constraint on the deformation is not available, we confine
attention to strain-energy functions that facilitate analytical treatment.
8.1 Spherical symmetry, revisited

Recall the kinematical development in eqns (7.98)-(7.102) for deformations having a cen-
ter of symmetry, but this time do not impose incompressibility. We adopt the constitutive
formulation developed in eqns (4.56) and (4.57), which yields
P = (w, + iw,)I — w,F + w,F* (8.1)
in the present circumstances, and hence,
DivP = V(w, + iw.) + F*(Vw;) — Div(w,F), (8.2)
where use has been made of the Piola identity DivF* = 0.

Using egn (7.101), we find that
FY =Vu@utAr(1-u@u) (3.3)
and, for uniform materials,
F*(Vw;) = (w3)'F*u = A’(w)'u, (84)
where we have made use of VR = u. Furthermore,
Vim, + iw.) = (Ww, + nw2)‘u. (8.5)

76 | UNIFORM, COMPRESSIBLE ISOTROPIC MATERIALS
Consider a strain-energy function having w, = 0. In this case the equilibrium equation

reduces to
[(wi)’ + A*(ws)'Ju = 0, (86)

yielding the ordinary differential equation:
(w,) + A?(w3)’ = 0. (87)
As an example, consider the class of compressible Varga-type materials defined by
w(iy,
is, 3) = Qui, + F(is)], (88)
where {4 is a positive material constant. This is simply the linear shear modulus, as in
the case of conventional Varga materials. We then have w, = 244, a constant, and, w; =
2uF'(i;). The differential eqn (8.7) simplifies to F’(i,)i,(R) = 0. Assuming F’(i,) # 0,
recalling that i, = J and using eqn (7.102), we obtain
rr (R) =JR’, with J = const. (89)
Integrating and imposing r(A) = 4A, where A, is an assigned positive constant, we finally
obtain the deformation
(RP =JR +03 -JA’. (8.10)
Here A can be identified with the initial radius ofa sphere, and A, and the ratio of final to
initial sphere radii.
It remains to determine the constantJ. For example, in the case of a solid sphere, it would
be natural to require that r(0) = 0, corresponding to another solid sphere. In this case, we
find J = A3 and r(R) = A,R. The deformation gradient is F = A,I, a uniform equi-triaxial
stretch. We refer to this as the trivial solution.
To explore conditions under which cavitation is possible, we consider the case r(0) =
a > 0, for some constant a. Evidently, this requires J < A}. Furthermore, if the newly-
created hole is traction free, then we must impose Tu = Oat r = a, where
Tu =A’ dw/dA,u. (8.11)
Accordingly, we impose
A? +F(J) =0 (8.12)
at r = a. Because 1 > oo asr —> a(R — 0) in this case, we requireJ to be such that
FG) =0. (8.13)

SPHERICAL SYMMETRY, REVISITED | 77
FQ)
Le
on
ion]
Figure 8.1 Constitutive response of a Varga-type material capable of supporting cavitation
Accordingly, ifa solution exhibiting cavitation is to exist, the function F must have at least
one stationary point, J), say. Such a function, adjusted to ensure that the strain energy
and Cauchy stress vanish when the material is undeformed, is sketched in Figure 8.1. The
cavitated solution is then given by
(RP =foR? + (VV -Ja)A’, (8.14)
and is available provided that the boundary displacement is such that A, > J)’.
In this solution and in the trivial solution, the deformation is controlled entirely by A,.
To choose between them, we compare the total energies required to maintain the two
solutions. In the case of any spherically symmetric deformation this is given by
E(A,) = 40 [ wR’ dR, (8.15)

0
where
w/2p = i, + FY)
= r(R) +2r/R+F()
= R?(R’ry + FC). (8.16)
We have
A
/ R?(R’r)'R'dR = A’r(A) = A,4', (8.17)

0
which is fixed by the data, and hence the energy comparison
E(A4) — Exw(Aa) = 430 | , R’[F(J) - F(o) JAR, (8.18)

0
Jl/3
da
Figure 8.2 Cavitated solution bifurcates off trivial solution at A4 = IV 3
where E,,, is the energy of the cavitated equilibrium solution. Of course, this is meaningful
only ifA, > Jj’. For the kind of material sketched in Figure 8.1, Jy furnishes the minimum
of the function F, ensuring that E,.,(A,) < E(A,); the cavitated equilibrium deformation
thus requires less energy than any alternative spherically symmetric deformation, includ-
ing the trivial equilibrium deformation. However, we have not proved that it minimizes
the energy relative to any kinematically possible (non-spherically symmetric) deformation.
Nevertheless, the analysis provides support for the conclusion that cavitation emerges when
the boundary radius exceeds the critical value J}’*, The cavity radius isa = (A - Jo)'A (see
Figure 8.2).
8.2 Plane strain

The term plane strain is used in reference to the two-dimensional situation:
x=x,+zk, y=y,+zk with y, = x,(x)), (8.19)
where k is a unit normal to a fixed plane &2 in which the deformation occurs, containing x,
and y,. The associated deformation gradient is of the form
F=F,+k@k, (8.20)
wherein F, maps {2 to itself. This may be written
F, = > AaVo ®@ Uns (821)

a=]
where the A, are the principal stretches (A, = 1) and {v,}, {u,} are orthonormal princi-
pal strain axes in Q (u, = v; = k). The former are the roots of the quadratic characteristic
equation
PROBLEMS | 79
V-IA+] =0, (8.22)
where
I= At + Ax = trU,, J = AyA2 = det U,, (8.23)
and U, is the right stretch factor in the polar decomposition of F, i.e, Uj = Ye Aate ®
u,. The rotation factor is Ry = Ve @u, and the cofactor is F) = 4.4, @u, tA,
V, ® Uy.
The stretches are determined by J and J, implying that the strain energy for isotropic
materials in a plane~strain deformation is
@(A,,A, 1) = w(L)), (8.24)
for some function w. This furnishes
00/0), = w, + Aw, O@/0A, = wy, + Aywy (8.25)
whereas 0w/0A3, evaluated at A, = 1, is a function of the A, and, hence, a function of x;,.

The Piola stress reduces to
P =P, + 0w/dA,;k Ok, (8.26)
where
P, = wi(v, @ wy + Vv. @ u,) + w,(A.v, ® u; + Arv, @ up)
For general applications, it is useful to observe that
Fy + Fy = (A, + A.)(v1 @ u, + v. @ wu) = IR (8.28)
and, hence, that
P, = I'w,(F, + F}) + wyF; (8.29)
in any plane-strain deformation.
Problems
1, (a) In three dimensions, establish the polar decomposition
F=RU (8.30)
in which R € Orth* and
U= > Au, @ u;
= (AI)(su, ® u, + su, @ wy + u; @ u)(t'?(u, @ u, + u, @ w,)

+ tu, @ u,|, (831)
where A,(> 0) are the principal stretches, {u,} are the orthonormal principal
axes of U and the factors correspond to a pure equi-triaxial stretch ofamount
(> 0), a pure shear of amount s(> 0), and an isochoric uniaxial extension
of amount ¢t(> 0) with accompanying lateral contraction. These are coaxial
and so may be composed in any order. [Hint: the problem is solved if you
can establish an invertible relation between the {A,} and {A, s, t}. This would
imply that the two expressions above for U are equivalent. ] Show that the
pure shear factor may be identified as the spectral decomposition ofa simple-
shear deformation on a fixed set of axes.
(b) In two dimensions, show that F may, without loss of generality, be decom-
posed in the form (8.30), where
U = (AI)(su, @ u, + su, @ u,) (832)
is the composition of an areal dilation of amount A and a pure shear of

amount s(> 0).
2. In two dimensions, use the spectral decomposition of U to derive
U=TIr'GI+C), where C=FF=U’, (8.33)
and, thus, obtain I directly in terms of the invariants of C. Use this to obtain an
explicit formula for U"', and use it to confirm that IR = F + F*.
8.3 Radial expansion/compaction

Henceforth, we drop the subscript (-), and consider deformations of the form
x=Re,(@), y=r(R)e,(@). (8.34)
We derive
F=re,@e,+(r/R)ep Qe, F* =(r/R)e, We, + res @ ey (8.35)
and
F+F*=11, where I=R'(Rr)’. (8.36)

PROBLEM | 81
Furthermore,
J = (r/R)r (837)
and the requirementJ > O implies that r(R) is an increasing function: r > 0. Accordingly,
in this case we have
P= w+ w F*. (8.38)
Problem
Show that equilibrium without body force for is equivalent to the ordinary differen-
tial equation (compare with eqn (8.7))
(wi)' + (r/R)(wy)' = 0. (8.39)
For uniform materials the trivial solution is
r(R) = AgR, (8.40)
where A, = r(A)/A and r(A) is the (assigned) radius after deformation of the disc of initial
radius A. To find a more interesting, yet tractable, alternative, consider again the special
class of Varga-type materials
w = 2u[1+ F(U)]. (841)
Before proceeding, consider the response of such a material to a uniform equi-biaxial

stretch, in which A, = A, = J’, as exemplified by the trivial solution. In this mode of
deformation the strain energy reduces to
w = 2u[2J'? + F()]. (8.42)
The Piola stress components are w/0A, = Ow/0A, = 24P(J), where
PU) =14+)'?F(). (8.43)
We would expect, on physical grounds, that P > -Loo as J — 00,0, respectively, and
hence that F'(J) + -00 asJ > O and F(J) > Oat large values of J. If, in addition, the
energy and stress vanish in the undeformed state, then F(1) = -2 and F’(1) = -1. A func-
tion with all these properties, similar to that sketched in Figure 8.1, is depicted in Figure 8.3.
This has an isolated minimum at some J, > 1.
FQ)
2 ay
Figure 8.3 Constitutive response of a Varga-type material capable of supporting plane-strain

cavitation
Problem
(a) Use this material to derive the general solution
r(R) = J(R?- A’) +A’. (8.4)
Thus, J = 42 for the trivial solution. Show that a cavity forms if J = Jo and A, >
Ji, with radius a = A(A2 - Jo)”. Plot this as a function of A, and show that it
branches off the trivial solution at A, = J)”.
(b) Carry out an energy comparison and show that the cavitated equilibrium deform-
ation minimizes the energy in the class of purely radial deformations, provided
thatA, > Ji”.
(c) Plot the Piola traction at the outer edge of the disc as a function of A, and show
that it increases without bound for the trivial solution, but saturates at a fixed value
in the cavitated solution ifA, > J}.
Also of interest are the so-called harmonic materials defined by
w(I,J) = 2u[FU) - J] (845)
for some function F. These have the remarkable property that they yield explicit solutions to
the general plane~strain equilibrium problem in terms of analytic functions of the complex
variable x, + ix,. However, they yield somewhat unrealistic predictions in deformations that
induce severe compression. This is borne out by eqn (8.45), which furnishes the question-
able prediction that a degenerate deformation with J — 0 can be attained at a finite value
of the energy. For this reason, the harmonic material is useful mainly in problems involving
small-to-moderate strains with possibly finite rotations.
PROBLEMS | 83
Problems
1. Show that all purely radial equilibrium deformations of harmonic materials are of
the form
r(R) = IR/2 + C/R, (8.46)
where I and C are constants. Show that cavitation is not possible in a harmonic
material.
2. Consider a spherical shell of uniform, isotropic material, occupying the annu-
lar region A < R < B. Solve the equilibrium problem (no body force) in the
class of radial deformations x > y = A(R)x, where A(R) = r(R)/Rand R = |x|.
Consider the following cases:
(a) The material is compressible with strain energy given by w/2u = f(i,) — iy,
where j(> 0) is a material constant. Assume f”(i,) > 0. State restrictions
onf ensuring that the energy and stress vanish in the reference configuration.
Assume the surface R = B to be traction free and the surface R = A to be
subjected to pressure P.
(b) The material is compressible with strain energy given by w/2 = i, + g(ia).
Assume g’(i;) > 0. State restrictions ong ensuring that the energy and stress
vanish in the reference configuration. Same loading conditions as in (a).
(c) Show how the addition of a term linear in i, to the strain—-energy function
affects the analyses of problems (a) and (b).
3. Recall that for plane strain of compressible isotropic materials the Piola stress may
be written in the convenient form
P=I'w(F+F*)+wF; [=A tay, J=Aiay, (8.47)
wherein all tensors are two-dimensional. Consider two-dimensional deform-

ationsx — y defined by
x=Ru(®), y=r(R)u(d), (8.48)
where u =e, and 6 = © + ['(R). This combines radial expansion/contraction

with azimuthal shear. It simplifies matters to write
y = u(R)u(O) + v(R)v(O), (8.49)
where v(x) = k x u(x) and
u(R) =r(R)cosP(R), v(R) =r(R)sinT(R). (8.50)

Find a pair of coupled ODEs for u(R) and v(R). Solve them for the special case
of harmonic materials with strain energies of the form w/2 = F(I) - J. Assume
F'(I) > 0. This furnishes a good model if the strains are moderate while the
rotations are large. Consider the BCs
r(B)=B, r(A)=AA; TB) =0, T(A)=t. (851)
Is there a limit, according to this model, on the amount of rotation t for

prescribed A > 0? Explain.
REFERENCES
Abeyaratne, R. and Horgan, C.O. (1984). The pressurized hollow sphere problem in finite elastostat-
ics for a class of compressible materials. Int. J. Solids Struct. 20, 715-723.
Carroll, M.M. (1988). Finite strain solutions in compressible isotropic elasticity. J. Elasticity 20,
65-92.
Horgan, C.O. (1992). Void nucleation and growth for compressible nonlinearly elastic materials. Int.
J. Solids Struct. 29, 279-291.
John, F. (1960). Plane strain problems for a perfectly elastic material of harmonic type. Comm. Pure
Appl. Math. 13, 239-296.
Ru, C.Q, (2002). On complex-variable formulation of finite plane elastostatics of harmonic materials.
Acta Mech. 156, 219-234.
Ru, C.Q,, Schiavone, P., Sudak, L.J. and Mioduchowski, A. (2005). Uniformity of stresses inside an
elliptic inclusion in finite plane elastostatics. Int. J. Non-linear Mech. 40, 281-287.
Steigmann, DJ. (1992). Cavitation in elastic membranes. J. Elasticity 28, 277-287.
Material stability, strong ellipticity
and smoothness of equilibria
9.1 Small motions superposed on finitely

deformed equilibrium states
Consider a small amplitude wave propagating through the elastic material. Suppose the ma-
terial has been predeformed to some equilibrium state, with position field y,(x), prior to
the passage of the wave. The wave causes the material point x to undergo a further dis-
placement to position x (x,t), say. Supposing the material to be incompressible and using
obvious notation, the Piola stresses in these configurations are
P, = W,(F,;x) -p.FT and P= W,(F;x)-pF. (9.1)
Here, we suppose the displacement from x, to x to be small in the sense that
X (xt €) = Xe(x) + €x’(u £) + of€), (92)

with |e| < 1, uniformly in x and ¢. We assume a concomitant change in the constraint
pressure; i.e.,
p(x tj €) = p.(x) + ep’(% t) + of), (9.3)
and we seek a system valid to linear order in € for the perturbation fields x’ and p’. We
have made excessive use of the notation (-)’, relying on the context to convey the intended
meaning; here, derivatives with respect to €, evaluated at € = 0.
We may use eqns (9.1)-(9.3) to deduce that
P(x, t; €) = P.(x) + €P’(x, £) + o(e), (94)

86 | MATERIAL STABILITY, STRONG ELLIPTICITY AND SMOOTHNESS
where
P= M[F]-p'F - p.Fi[F'], (9.5)
wherein the derivatives
M = Wer (9.6)
and F¥ are evaluated at F,. In terms of components,
Mauys = PW/dF 40 5. (9.7)
Problem
Show that M possesses major symmetry, ie, M = M', where, for 4th order
tensors, the transpose is defined by A- M'[B] = B- M[A]. Thus, Miyy = Myaia-
Assuming zero body force for simplicity, we now substitute eqn (9.4) into the equation of
motion
DivP = p,X", (98)
divide the result by €, and let € — 0 to arrive at the linear differential equation
DivP’ = 1,1, (99)
where
u= x’ (9.10)
is the linear approximation to the small displacement, and
F’ = Vu. (9.11)
The constraint of incompressibility imposes a restriction on Vu. To see this, we write
J=}.+6J +0(€), (9.12)
withJ = J, = 1, divide by €, and let € — 0 to obtain J’ = 0. However, J' = J; F, yielding

the restriction
FY. Vu =0. (9.13)

PROBLEM | 87
This and eqn (9.9) provide the system to determine u and p’, subject to appropriate
boundary and initial conditions.
If the underlying equilibrium deformation x,(x) is homogeneous and if the material is
uniform—this situation being the simplest—then E, and p, are uniform. In particular, M
is then uniform and equation (9.9) simplifies to
Pei; = (Miss — p.OF*,/OF 5) u;ps - Ps (9.14)
Problem
Show that JOF/,/F, = Fi,F, — FF, and, hence, that (OF*,/OF.g)tj,n4 = 0. The
latter is the linearized form of the Piola identity Fi, = 0.
Thus,
Px ii; = Masjpuna - Fi ply with Feuias = 0. (9.15)
The compressible case is recovered by omitting the Lagrange multiplier p’ and suppressing
the second equation.
Consider a plane harmonic wave of the form
u(x,t) = aexp[i(k-x-—qt)], p’(x,t) = qexp[i(k-x-t)], (9.16)
wherein the constant vectors a and k are the polarization and wave vectors, respectively.
The constant q is the amplitude of the perturbed constraint pressure; the constant w is
the frequency, and i is the complex unit (i = —1). We show that this furnishes a solution
to eqn (9.15). Naturally, these simple functions are not able to satisfy initial or bound-
ary conditions, and so we suppose that the wave has been propagating for some time in
an unbounded medium. Equivalently, attention is confined to an interval of time prior to
impingement of the wave on the boundaries of the body.
A convenient alternative representation of the waveform is obtained by introducing the
wave number k = |k| and wavespeed c = w/k. Then,
u(x, f) = aexp[ik(N-x-ct)] and p’(x,t) = qexp[ik(N-x-ct)], (9.17)
where N = kk. These make clear the fact that the wave form is preserved on the plane
defined by
N-x=ct+d, (9.18)
with unit normal N. Here, d is the perpendicular distance from the plane to the origin at
time zero. Thus, the distance from the plane to the origin changes with velocity c.
Evidently, this perturbation of the underlying equilibrium solution remains bounded in

amplitude provided that c or w is a real number. This case is referred to as material stability,
to highlight the fact that no boundary or initial conditions are involved and, hence, that
the stability or otherwise of the underlying solution depends entirely on the properties of
the material per se. Of course, it remains to verify that eqn (9.17) furnishes a solution to
eqn (9.15). Simple calculations give
tia = ijk, Ujan = —Uykaks, uj =-w@°u, and p’, = ip’ky (9.19)
and eqn (9.13) yields the restriction
u- Fik = 0, (920)
which, by virtue of Nanson’s formula, requires that the displacement be polarized in the im-
age of the plane defining the plane wave in the deformed equilibrium configuration. Using
these results, eqn (9.15), is reduced to the algebraic equation
A(E,, k)u + ip’F*k = p,@°u, (921)
where A(F, k) is the so-called acoustic tensor, having components
Ay = Mias(E)kaks. (922)
Problem
Use the major symmetry of M to demonstrate that A is symmetric.
From egn (9.21) it follows that
u- A(E,,k)u = 9,” |u|” ; (923)
and the remaining content of eqn (9.21) is
F*k- A(F,, k)u + ip’ |F*k|’ = 0, (924)
which determines p’ in terms of u. Our procedure can yield complex values because of the
assumption eqn (9.16). To rectify this, we can qualify the latter by taking real or imaginary
parts a priori. We conclude that p’ is bounded if, and only if, u is bounded. The first result
implies that w* > 0; hence, w € R and material stability, whenever A is positive definite.
Whether or not this is the case evidently depends only on F, and the strain-energy function.
In the case ofa compressible material the constraint eqn (9.13) is not relevant and the
foregoing results remain valid with the Lagrange multiplier suppressed. Equation (9.21) is
replaced by the eigen-problem
A(E,, ku = 9,@7u. (925)

SMOOTHNESS OF EQUILIBRIA | 89
Because A is symmetric, it has three real eigenvalues w’ and three mutually orthog-
onal eigenvectors—the polarization vectors. Material stability obtains if, and only if, all
eigenvalues are positive and, hence, if and only if A is positive definite.
Problems
1. According to the foregoing analysis the equilibrium state is stable with respect to
perturbations of the form
u(x,t) = aexp[i(k- x - wt) ] (9.26)
if the associated acoustic tensor is positive definite. Show that an unstable solution
exists if the acoustic tensor has a negative eigenvalue. Furthermore, show that an
unstable solution of the form
u(x, t) = af exp(ik - x) (9.27)
exists when the acoustic tensor is positive semi-definite. Conclude that strict
positive definiteness of the acoustic tensor is a necessary condition for stability.
2. Consider the propagation of infinitesimal waves superposed on a static finite de-
formation (in equilibrium without body force) of a homogeneous incompressible
elastic solid. Suppose the underlying finite deformation is a homogeneous tri-
axial stretch with deformation gradient F = A,e, @ E, + A,e, @ E, + A3e, @ E;,
where A, are positive constants and {e,} = {E,}. Consider plane harmonic waves
superposed on the static solution. Obtain an expression for the acoustic tensor us-
ing the neo-Hookean strain-energy function and analyze the associated eigenvalue
problem for the wavespeeds. Are there conditions under which the wavespeeds
can be imaginary numbers?
9.2 Smoothness of equilibria

Suppose a given equilibrium deformation x (x) is C’, i.e., twice differentiable in the sense
that x,, F,, and F,,, are continuous functions of x. It satisfies
Pia t+ 0b; = 0 (928)
everywhere in the body, with
P= OW/dF,. (9.29)
We consider only the unconstrained case for now.

The chain rule furnishes
Pisa = Mision, + Ri, (930)
where
R, = 0 W/dx,0F4, (9.31)
and eqn (9.28) reduces to
MiasFaa + Pxb, + R, = 0. (9.32)
Suppose, instead, that x(x) is C’ and piecewise C’, ie. x, and F,, are continuous, but
F,,3 May jump across one or more surfaces in the body. We want to derive conditions that
allow for this possibility, but first we need some preliminary discussion.
Consider a patch of surface, described in parametric form by the position field &(1,, u2),
and let N(u,, 4) be a unit-normal field on this patch. The surface divides the reference
configuration into two parts, denoted by + and -. A point off the surface may be located by
specifying the value of ¢ in the normal-coordinate parametrization (Figure 9.1)
(uy, ta, ¢) = k(u4, uy) + CN(m4, up) (9.33)
of the surrounding 3-space. It is easy to demonstrate that the relationship between the
coordinates {t,, #, ¢ } and x is invertible in any sufficiently small three-dimensional neigh-
borhood of a point on the surface. In particular, there is a one-to-one relationship among
the Cartesian coordinates x, and {u,,¢} in this neighborhood; we use Greek subscripts,
ranging over {1,2}, to identify surface coordinates.
Furthermore, we assume these relations to be as smooth as required by the analysis.
Confining attention to such a neighborhood, we are then justified in writing
do = Voc -dx, (934)
where
dx = du, + Ndg + cdN. (9.35)
Because &,, and dN are tangential to the surface, it then follows that
dg =N- dx, (9.36)
and comparison with eqn (9.34) yields
N=Vco. (9.37)
SMOOTHNESS OF EQUILIBRIA | 91
Figure 9.1 Normal-coordinate parametrization of three-space in the vicinity of a surface
For any functionf of position, and hence of the coordinates, we define the upper and
lower surface limits
flu, u) = jim flu, ua,¢) and f(a, u,) = lim flu, Ur, 6), (9.38)
and the associated discontinuity
if] = f(y, uy) -fi(u, uy). (9.39)
If the deformation x(x) is continuous across the surface, i.e., if the material does
not fracture, then the x, are continuous, ie., X,"(4, 42) = X/ (ui, u). Assuming sufficient
smoothness of these limits on the surface, we can differentiate and conclude that
[Xie] = [Xian] = 0, ete. (940)

However, there is no reason to conclude that there is any relationship between the limits
Xiu, 2) = lim (8x,/85) and X,.(m,%) = lim (8x:/85). (941)
Ina small neighborhood of the surface, we may apply the chain rule to derive
Fp = Xjatlas + XicSip- (9.42)
The jump across the surface is then found to be
[Fis] =a,Nz, where a; = [X,<]. (9.43)

The discontinuity in the deformation gradient is thus necessarily of the form
[F]=a@N (94)
for some vector a. Tensors of this type are said to be rank-1. Here, as in the theory
of matrices, the rank of a tensor A, say, is equal to the dimension of its image space:
RankA = dim{Av}, where v is any vector. In the present example the image space is the
one-dimensional space spanned by a.
Suppose now that x (x) is C’. Then [F] vanishes and so a vanishes: [x;, ] = 0. We have
Xi (a1, 2) = X(t, v2), and assuming these surfacial limits to be smooth, it follows that
[Xica] = 0. Proceeding from eqn (9.42), we compute
Fipa = Xjatlapa + Xic San + Xjap Mappa
+Xjca Ua ASB + Kis 4a, SA + Xiss CaS B (9.45)
Taking jumps then yields
[Fn,] = a)NgNa, where a, = [Xj cc]. (9.46)
Next, we take limits of eqn (9.32) as the surface is approached from above and below,
obtaining
Musk 5tk a ten = 0, (947)
wherein the missing terms are continuous across the surface. Subtracting the two equations
and invoking eqn (9.46), we find that
MausyaNaNoa; =0, (9.48)
or
A(F, N)a = 0, (9.49)
where A(F, N) is the acoustic tensor based on N. It follows that a discontinuity is possible,
ie., a # 0, if and only if det A(F, N) = 0. This is an equation for the local orientation N
of the discontinuity surface. On the other hand, if the strain-energy function is such that
A(F, N) is non-singular for any deformation, i.e., if the equations of equilibrium are always
of elliptic type, then eqn (9.49) requires that the discontinuity vanish and the underlying
deformation is C’. It is possible to continue in a recursive manner to show that if an equilib-
rium deformation is piecewise C”, then it is, in fact, C" for any n, provided that the acoustic
tensor is non-singular, granted sufficient regularity of the function W(F).
INCOMPRESSIBILITY | 93
Problem
Verify this claim. Hint: the ¢ —derivatives of the deformation are the only ones
having potential discontinuities.
Note that material stability, which is tantamount to the positive definiteness of the
acoustic tensor and which confers its nonsingular character, is enough to ensure that the
smoothness conditions to which we have referred are fulfilled. However, this falls short of
proving that equilibria are arbitrarily smooth in the presence of strong ellipticity. For, there
is no known proof of the piecewise C* continuity that was presumed at the outset, although
partial results of this kind are known fora restricted class of boundary data (see the paper
by Healey and Rosakis, 1997).
9.3 Incompressibility
We have seen that if the material is incompressible then the stress is given by
Pis = OW/OF,4 — pF’. (9.50)
Suppose that F,, and p are continuous functions of x, but that their gradients may be
discontinuous across some surface. Then, as before,
[Fix] =a)NgN, and [py] = qNa, (9.51)
for some a, and q. The second of these is derived as eqn (9.44) was derived, on replacing x,
by p. On either side of the discontinuity surface, eqn (9.28) applies and yields
MayjEpa - Papa t= 9, (9.52)
wherein the missing terms are continuous. Taking limits from above and below this surface,
and subtracting the resulting equations, as before, we arrive at
A(F, N)a = qF*N. (9.53)
Here, however, a is subject to the restriction J(F(x)) = 1, identically, at all points removed
from the discontinuity surface. This implies that VJ vanishes identically in the body, minus
the surface. Using the chain rule, this is found to be equivalent to
PF’ Fa = 0. (9.54)
Taking the jump, we arrive at
a: F*N = 0. (9.55)
Accordingly, eqn (9.53) requires that a- A(F, N)a vanishes if a discontinuity is to exist.
It follows that if the strong ellipticity—or material stability—condition is satisfied, then
the only resolution is a = 0; then, eqn (9.53) yields q = 0, and there is no discontinuity.
Proceeding by recursion, it is possible to show that both the deformation and pressure fields
are arbitrarily smooth, granted the degree of continuity assumed at the outset.
Problem
Prove this.
REFERENCES
Healey, TJ. and Rosakis, P. (1997). Unbounded branches of classical injective solutions to the forced
displacement problem in nonlinear elastostatics. J. Elasticity 49, 65-78.
Knowles, J.K. and Sternberg, E. (1975). On the ellipticity of the equations of finite elastostatics for a
special material.J. Elasticity 5, 341-361.
Membrane theory
A membrane is a thin sheet of thickness h, which is much smaller than any spanwise dimen-
sion of the sheet, such as its overall diameter or the diameter of an interior hole. Membranes
provide a particularly useful setting for the empirical testing of formulations for the strain-
energy function. Their relatively easy deformability affords empirical access to large regions
of strain space. In this respect they are similar to rubber bands in furnishing archetypal ex-
amples through which elasticity can be understood, both qualitatively and quantitatively.
Indeed, the empirical work of Treloar (1975) on rubber elasticity—arguably the most
important collection of work of its kind—was conducted on thin membranes.
10.1 General theory

Our intention, here, is to exploit the thinness of the membrane to derive an approximate
two-dimensional theory that captures the most important aspects of the behavior of thin
sheets. We concentrate on equilibria, although extensions to accommodate dynamics are
straightforward. To this end, let / be the next smallest length scale in the problem at hand,
such as a spanwise dimension or the length scale for the spatial variation ofa distribution of
load. We suppose that h/1 < 1, and proceed to derive the leading-order two-dimensional
approximation to the three-dimensional equations. This leading-order approximate model
is what we mean by membrane theory. To highlight its important features, we consider
the simplest case in which the reference configuration « is a thin prismatic plate-like re-
gion of three-space. In this simplest case, we can decompose the reference configuration
into the Cartesian product ofa midplane Q and a through-thickness fiber C: « = Q* C
(Figure 10.1). Correspondingly, we write position in « as
x=x(u,¢), where k(u,¢)=ut+ck. (10.1)
Here, u is (two-dimensional) position on Q, ¢ € C isa linear through-thickness coord-

inate and k is the (fixed) unit normal to . Thus, C may be identified with the interval
[-h/2, h/2] containing ¢. We suppose that all length scales have been non-dimensionalized
by J; equivalently, we adopt! as the unit of length (J = 1) and assume thath < 1.

96 | MEMBRANE THEORY
Figure 10.1 Reference configuration ofa thin sheet
In a deformation x (x) of the body, the plane Q is carried to a surface w, described

parametrically by
r(u) = x (x(u, 0). (102)

As we shall see, the determination of this function is the main objective of membrane
theory. As usual, the problem to be solved is
DivP = 0 (103)
in K, subject to some set of boundary conditions. We exclude body forces for the sake of
brevity and convenience; their inclusion presents little difficulty. Equivalently,
O= Piss = Pre + P., (10.4)
where Greek indices range over {1,2} and the prime stands for d(-)/@¢. In other words,
we have identified ¢ with x;. We thus identify x, with u,, the Cartesian coordinates of u;
and E, with k.
This suggests the decomposition
P=P1+Pk@k, (10.5)
where
1=I1-k@k (10.6)
is the projection onto the plane &2. Here, we have used this to expand the identity P = PI.
Furthermore, 1 = E, ® E, and so
GENERAL THEORY | 97
P1 = P,e, @ E,. (10.7)
Equation (10.4) is then seen to be the component form of
Div|(P1) + P’k = 0, (108)
where Div,(-) is the two-dimensional divergence with respect to position u. This is just
another way of writing eqn (10.3); accordingly, it holds at all points of « and, hence, on the
plane &2 in particular, ie., at ¢ = 0, where it reduces to
Div(Pol) + Pik = 0. (109)
The subscript (-)» refers to function values on the plane, and (-)) refers to a ¢ - derivative,
evaluated at ¢ = 0.
We will focus attention on uniform incompressible materials, for which
Py = W,(F) - qoF}- (10.10)
We use q instead ofp to denote the constraint pressure, for reasons that will become clear
as we proceed. Let ¥(u,¢) = x (x(u,¢)). Then,
Fdx = dx =(VxX)dut x'dc, (10.11)
where, for the purposes of this chapter, V is the (two-dimensional) gradient with respect to
u, whereas, from eqn (10.1),
dx = du+kdc. (10.12)
As du = 1du, we have
Fdx = Fldu + Fkdg, (10.13)
and comparison with (10.11), together with F = Fl + Fk @ k, yields
F=Vx+x' Ok (10.14)
Evaluating this at the midplane, we obtain
F, = Vr(u) + d(u) @k, (10.15)
for use in eqn (10.10), where d = Xj. Observe that r(u) and d(u) are independent vector
fields on (2. In the literature, d is often referred to as the director field. It represents the tan-
gent to a material curve after deformation, evaluated at ¢ = 0, and oriented perpendicular
to Q in«. The first term on the right in eqn (10.15) has the representation
Figure 10.2 Position and director fields on the deformed surface
Vr=ry @E, (10.16)
in obvious notation. Because the r, lie tangential to the deformed midsurface wv, this tensor
maps 2 to the tangent plane to w at the material point associated with u (see Figure 10.2).
The areal stretch w and orientation n of the material surface w may be inferred from
Nanson’s formula. Thus,
an = F*k = FE, x FyE;

= (Vr)E, x (Wr)E, = 4) X 1, (10.17)
implying that
o = |r, X rp]. (10.18)
Furthermore,
Jo = det Fy = [F,E,, FoE,, FoE;|

= F,E, x FE, - Fok =r, Xr, +d, (10.19)
and so
Jo=an-d. (10.20)
Accordingly, if the deformation is isochoric, as it must be for incompressible materials,

then
an-d=1. (10.21)
PRESSURIZED MEMBRANES | 99
This yields the conclusion that
d=ar'n+(Vrje, (1022)
where e is a two-vector lying in &2.

Thus far, we have merely recast the equations without invoking any approximations. We
do so now, by estimating the lateral-traction boundary conditions. For example, if p* are
the tractions acting at the major surfaces ¢ = +h/2 with unit normals N = +k, then p* =
+P*k. Accordingly, for small h, we have p* = 4Pok+ (h/2)P)k + o(h). The net-force
density and the force-difference density on these surfaces are thus approximated by
p’+p =hPik+o(h) and p*-—p =2P,k+o(h). (10.23)
It may be noted that the degree of differentiability required by these estimates is consist-
ent with our earlier discussion about smoothness of equilibria in the presence of strong
ellipticity.
Substituting the first estimate into the exact equation eqn (10.9), we derive
Div, (Pol) + h'(p* + p’) + ’o(h) = 0. (10.24)
An attempt to balance the terms reveals that p* + p” can be of order h, at most (including
the possibility that it vanishes) and, hence, that
p’ +p =hp+o(h), (10.25)
where p is a vector field of order unity. In the same way, eqn (10.23), part 2, indicates that
p’-p =2q+0(1), (10.26)
where q is likewise of order unity. Inserting these into eqns (10.24) and (10.23), part 2, and
passing to the limit h — 0, furnishes the leading-order differential-algebraic problem
Div, (Pol) +p=0, Pok=q (10.27)
for the determination of the fields r(u) and d(u). In the case of incompressibility we have a
system for the determination of {r(u), e(u), qo(u)}.
10.2 Pressurized membranes

An important example is furnished by lateral pressure loading. Suppose the upper surface
¢ = h/2is traction free, while the lower surface ¢ = —h/2is loaded bya pressure of intensity
p- Then, of course, p* = 0.
Problem
Show that p’ = p(F*)"k.
Thus, p- = p[F*k + O(h)], yielding
plFsk + O(h)] =hp+o(h) and -p[F*k+ O(h)] = 2q+ o(1). (10.28)
These are reconciled by taking
p = hP + o(h),
with P of order unity. It follows that p = PF3k and q = 0, and the equations to be solved are
Div(Pol)+a@Pn=0 and P,k=0. (10.29)
The pressure is seen to contribute a force that is distributed over the membrane surface,
in the same way that a conventional body force is distributed over a body’s volume. The sec-
ond equation implies that the membrane is in a state of plane stress, at leading order. In the
older literature, conditions of the latter type were typically imposed, rather than derived, as
we have done. This is unnatural, however, and obscures the logical structure of the theory.
10.3 Uniqueness of the director

Observe that the plane-stress condition, or alternatively eqn (10.27), part 2, amounts,
via eqns (10.15), (10.22), and (10.10), to an algebraic relationship among the entries of
{r(u), e(u), qo(u)}- In the next subsection—on isotropic materials—we will use it to evalu-
ate e and qp in terms of the midplane deformation r(u). Before doing so, we would like to
know whether or not such solutions are unique. We proceed to answer this question in the
affirmative, with the proviso that the strong-ellipticity condition is satisfied.
First, observe that eqn (10.29), part 2, is equivalent to the statement F‘(W;)k = qk; here
and, henceforth, we drop the subscript (-), for convenience. This, in turn, is equivalent to
0 = 1F'(W,)k = (Vr)'(W,)k and q=k-F(W,)k=d-(W,)k. (1030)
Next, let us fix r(u) in the function W(F), where F is given by (10.15) and (10.22). This
results in a function of e, which we denote by G(e). Consider a path e(u) in the space of
two-vectors, and let o(u) = G(e(u)). This has the derivative
& = We-d@k= W,- (Vr)e @k = (Vr)é- (W,)k = é- (Vr)'(We)k, (1031)
from which it follows that
G, = (Vr)'(W.)k. (10.32)
PROBLEM | 101
We see, from eqn (10.30), part 1, that G is stationary at any solution to our problem, i.e.,
G, =0.
The second derivative is
o=G,-é+(G,) -¢, (1033)
where
(G.) = (Vr)'(Wap[(Vr)é ® k])k. (1034)
Problem
Reduce this tod = G,-@ + (Vr)e - {A(F, k)}(Vr)e, where A(F, k) is the acoustic
tensor based on the unit vector k.
The linear space of two-vectors is a convex set. As such, it contains the straight-line path
e(u) = (1 - u)e, + ue,, with u € [0,1], for any pair e, and e, of distinct two-vectors. That
is, every vector can be expressed a convex combination of two given vectors. On this path
we have é = 0 and
ao =(Vrje- {A(F,k)}(Vre, (1035)
withe = e, - e,(7 0).

Recall that strong ellipticity or material stability is the requirement, for incompressible
materials, that a - A(F, b)a > 0 forall non-zero a and any unit vector b, such that a - F*b =
0. Picking a = (Vr)é and b = k, and invoking eqn (10.17), we find that a- F*b = (Vr)e-
(an), which vanishes identically. Accordingly, strong ellipticity implies that ¢ > 0 for all
u € [0,1]. Integration then yields
é(u) =6(0) + / " &(x)dx > 6(0) (1036)

ifu > Oand, hence,
o(t)=a(0)+ [ a(u)du > o (0) + a(0), (10.37)
G(e,) - G(e,) > (e.-e1) - G,(e:). (10.38)
This means that G(e) is a convex function. Of course, we can interchange e, and e,, and
repeat the foregoing argument, obtaining
G(e,) - Ge.) > (e, - e2) - G.(e2). (10.39)

Adding these inequalities, we conclude that
[G,(e.) - G.(e,)]+(e,-e,) > 0, forall e, #e,. (10.40)
Suppose now that there are two solutions, e, and e,, say, to the stationarity problem.
Then G,(e,) - G.(e,) is manifestly zero, and the only possibility consistent with (10.40)
is e, = e,; the solution e is unique. Equation (10.30), part 2, then furnishes a unique
constraint pressure.
Beyond this, if e,, say, is the solution, then eqn (10.38) yields G(e,) > G(e,) for any e,
not equal to e,. We conclude that the solution to eqn (10.30), part 1, minimizes the energy
relative to any alternative value of e.
10.4 Isotropic materials

By far the majority of applications of membrane theory concern isotropic, incompressible
materials, and so we confine our further attention to this important case. With reference to
eqn (4.41) and Chapter 6, we may write P = FS, with
S = J A/(80/82, - g/d,)u, ® w, (1041)

i=l
where @(A,, A», A3) is the (extended) strain-energy function, written in terms of the princi-
pal stretches, and u, are the associated (orthonormal) principal axes. As F is invertible, the
plane stress condition eqn (10.29), part 2, is equivalent to the statement
Sk =0. (10.42)
This implies that k is an eigenvector of S, with eigenvalue zero. We may, therefore, identify
k with us, say, and conclude that
g = 38/93 = (A,A,)18w/AAy, (1043)

which is seen to be equivalent to eqn (10.30), part 2, and where isochoricity has been im-
posed in the form A,,A3 = 1. Because the u, are orthogonal, the u, lie in the plane (2;
accordingly, from eqn (10.15),
Ay, =(Vr)u, and A,v, = (Vr)u, (10.44)
implying that the principal vectors v, are tangential to w. This is enough to conclude that v,
is perpendicular to the deformed surface and, thus, aligned with its unit normal n. Here, of
course, v, are the eigenvectors of the left stretch tensor V. We suppose, without loss of gen-
erality, that {u,} and {v;} are right-handed triads. The positivity of the scalar triple product
J = [Fu,, Fu,, Fu,] then furnishes v, = n. Finally, eqn (10.15) delivers
ISOTROPIC MATERIALS | 103
d =A,n, (10.45)
implying that w = A;' ande = 0 in eqn (10.22); this is the unique director field.
Let
@(Any Az) = @(Ai, Ad, (Ara) "). (10.46)
Then,
d@/0A, = Ow/dA, + (8w/8A;)9A3/9A,

= 0w/8A, —Aj'A39w/9A,
= dw/dA, -Aj'q. (10.47)
Likewise,
80/9A, = 8w/dA, - A5'q. (10.48)
It follows that
2
S= )°A,180/8A,u, ® uy, (1049)

a=)
and
2
P= y- 8@/0A,Vq ® Uy. (10.50)

a=l
Note that P = P1 because we have already solved Pk = 0. Also, we have used the trac-
tion conditions at the major surfaces to evaluate the constraint pressure a priori, and so
the membrane problem does not involve a Lagrange multiplier.
Often the h-multiplied version of the membrane problem is preferred. This is
Div, (hP) + apn = 0, (10.51)
wherep is the actual pressure, apart from an error of order o(h), @ = 4,A, and
2
hP = ) > U/9A4V @ Va, (1052)

a=]
where
U(Ay Az) = h@(A,, An) (10.53)
is the strain energy per unit area of Q2.

Before proceeding to an example, we pause to re-write the equations in yet another,

arguably more convenient, form. To this end, note that
hP = hP1 = h(PE,) @E, =p, ® Eu, (10.54)
where
Pp, = APE, = hPixe;. (1085)
Accordingly,
Div, (hP) = hPiaa@s = Paes (10.56)
and the equation to be solved is
Pi + pan = 0. (10.57)
Physically, the stress vectors p, are the force resultants (forces per unit length) transmitted
across the material lines on which the u, are constant.
10.5 Axially symmetric deformations

of a cylindrical membrane
Consider a reference configuration ofa membrane in the shape of a right circular cylinder.
We can regard this configuration as a mapping from an initial plane, as in the previous sub-
section, by writing u = u,E, with u, = zand u, = RO, where z is the axial coordinate along
the axis of the tube, @ € [0,277) is the azimuthal angle, assuming constant values on the
generators of the tube, and R is the (constant) tube radius. As before, we identify E, with k,
the unit normal to the plane (Figures 10.3a—d).
We suppose the deformed membrane to be a surface of revolution, parametrized in the
form
r(u) = r(z)e,(6) + &(z)e,, (10.58)
where e,(@) = cose, + sin@e, (note carefully that the subscript labels do not conform
to the common convention). The axis of symmetry is directed along e,, and we impose
f{e;} = {E,}. Here, r(z) and &(z) are the radius and axial coordinate, after deformation, of
the material circle defined by z = const. on the reference cylinder. The latter is recovered by
putting r(z) = Rand &(z) = z (Figure 10.3).
To construct the membrane deformation gradient, we use
(Vr)du = dr = [r'(z)e,(0) + E'(z)e, ]dz + r(z)e,(0)d0, (10.59)

AXIALLY SYMMETRIC DEFORMATIONS OF A CYLINDRICAL MEMBRANE | 105
9 n
NS! 4 €: p
“|
(c) (d)
Figure 10.3 Geometry of reference and deformed surface. (a) Plane isometric to reference cy-
linder. (b) Geometry of reference cylinder. (c) Section of deformed surface of revolution. (d)
Meridian of deformed surface.
where e,(9) = e'(9) =e, x e,(@). Noting that dz = du, = E,-du and dO = R"du, =
R”E, - du, we conclude, in accordance with eqn (10.16), that
Vr = (re, + £'e,) @ E, + (r/R)es ® E,, (10.60)
which immediately delivers
Vr= hw ® uy, (1061)

a=)
with u, = E,, u, = E,, and
Aiv, =re,+&'e,, Av, = (r/R)es. (1062)
Hence, the principal stretches:
A, = V(r)? +(&’)? and A, =r/R, (10.63)
We also have the stress vectors:
Py, = (» OU/OA pv, (ce) ») E,, (10.64)

Bel
so that
p, = 9U/dA,v, and p, = 0U/0A,%, (1065)
yielding
a 0
Paa = ay atm) + Rag rive): (10.66)
where
f, =A;'0U/8A, and tf, =A;'dU/dA,. (1067)
Problem
Show that the ¢, are the principal Cauchy stress resultants, i.e., the eigenvalues of
h*T, where T is the value of the Cauchy stress at the midsurface and h* is the
thickness of the membrane in its deformed configuration.
Using eqns (10.62) and (10.65) we derive

a
py atin) = (A2A7")'Arvs + dA; (r"e, +&e,) and
5g Aitave) = d,t,e,(0) = -A, he, (1068)

a i
and eqn (10.57) reduces to
— R'Ayhe, + (A,Aj"f)/Ayy, + 2A (re, + Ee.) + pA Aon = 0. (1069)
Due to of our unorthodox labelling of axes, the exterior unit normal to the deformed mem-
brane is obtained using A,A,n = —A,v, X A,v,. Projecting eqn (10.69) onto e,, we then
find that
(A,A7't&')' — pr’(r/R) = 0, (10.70)
which may be integrated, in the case of a uniform inflation pressure, to yield
A7'0U/0A,8' = spriR+ C, (1071)
where C is a constant.
Problem
Prove that this constant is proportional to the axial force acting on a cross section
of the cylinder. The result is, therefore, an elementary consequence of axial force
equilibrium.
PROBLEM | 107
Projection of eqn (10.69) onto the tangent, v,, to the meridian of the deformed mem-
brane yields
~ (7 /R)t + (AATH)AL + AAT HAT (r'r” + &E") = 0. (10.72)
The second term is the same as (A,¢,)! - A2Aj't,A; and the final term in parentheses is 1.,A'.
The equation thus reduces to (A,£,)' = jt, or
(rt,)’ = tr’. (10.73)
Together with eqn (10.71), this provides a system for the determination of r(z) and &(z).
Problem
Consider the problem of an unpressurized membrane of length 2L mounted on par-

allel rings of radius R at z = +L. An axial force F is applied to these rings. The
membrane is composed ofa neo-Hookean material (In Problem no, 2 of Section 9.1
you proved that this material satisfies strong ellipticity). Compute the relation
between this force and the axial half-length, |, of the membrane, i.e., ! = E(L).
To proceed, use symmetry to justify the assumption that £ (z) is an odd function;
ie., €(-z) = -€(z). Then, (0) = 0. Next, observe that eqn (10.63), part 1, implies
the existence of an angle @(z) such that
r=A,sing, §& =A, cos¢. (1074)
Thus, tan d = dr/d&, implying that @ is the angle made by the tangent to the me-
ridian with the symmetry axis. From Figure 10.4, we infer that the tangent to the
meridian vanishes at the throat of the membrane and, hence, that (0) = 0.
From eqn (10.71) we have A;'0U/0A,&' = F/27R, or
F/2nR = 0U/8d, cos @, (10.75)
whereas eqn (10.73) furnishes (8U/0A,)’ = 4,A;'0U/A,. Expanding the deriva-

tive on the left and solving for 4), we derive
a ie
ze-LO”S”””C SL z=0
(a) (b)
Figure 10.4 Reference and deformed configurations of the membrane

M = (0°U/ANY'R™ sin $(AU/IA, — A,0°U/9A,9A2), (10.76)
where use has been made of eqn (10.74), part 1, in the form
A, = RTA sing. (1077)
Equations eqns (10.75)-(10.77) provide a differential-algebraic system to be solved

for the functions A, (z), A,(z) and $(z), subject to the boundary conditions (0) =0
andA,(L) = 1.
To solve this system use a shooting scheme, with Euler backward differ-
encing commencing at z/L=1. This entails guessing the missing boundary
value ¢, = $(L) and adjusting it until a solution having (0) = 0 is achieved.
Nondimensionalize the problem using the length L and the shear modulus of the
material. Finally, compute
L
i= | A, cos bdz (1078)

0
to obtain the value of I corresponding to the assigned value of the force. Can you
plot the shape of the deformed membrane, i.e., rvs. &?
Re-analyze this problem for the bio-elastic material defined by eqn (7.38). Does
this material satisfy strong ellipticity? Consider various (positive) values of y.
A useful result follows from eqn (10.73), which we write in the form
M9U/OA, = A4(9U/9A,)’ = (AyOU/9Ag) - NOUBA. (1079)

Noting that 4{0U/@A, + A,0U/dA, = U'(A,,A,) for uniform materials, we find (see
Pipkin, 1968) that this integrates to H = const., where
H(Ay,A2) = U(Ay, A) — A,OU/4A,. (10.80)
This furnishes a simple check on the accuracy of solutions.
Problem
Using the results of the previous exercise, plot H as a function ofz and confirm that
it is indeed constant.
10.6 Bulging of a cylinder

A party balloon may be idealized as a long cylindrical tube. This is sealed at its ends and pres-
surized, but we assume the tube to be long enough that the details of the deformation near
these ends can be safely ignored. One typically observes that the cylinder deforms, roughly,
BULGING OF A CYLINDER | 109
a r
r_ +
—————
Figure 10.5 Bulging of a cylinder
into another cylinder, with radius depending on the pressure, until a certain threshold
pressure is reached, at which point a bulge starts to form at one end. This bulge propa-
gates down the length of the cylinder at a more-or-less fixed value of the pressure p’, say,
until it consumes the entire balloon. Thereafter, an increase in pressure again produces a
roughly cylindrical membrane, with radius again depending on the pressure. During the
bulge-propagation phase at pressure p*, the membrane is deformed into two coexistent
cylinders, separated by a transition region in which the radius varies with the axial coord-
inate (Figure 10.5). We use membrane theory to derive a simple model of this interesting
phenomenon.
With reference to the figure, we seek a solution in which the membrane has deformed
radii r* on either side of a transition region. We refer to these uniform states as phases. In
either of the phases we have r’ = Oand, therefore, A, = &’, assuming & (z) to be an increasing
function. An elementary balance of axial forces yields 27 ROU/0A, = prr’, or
1
aU/IA, = abr /R, (1081)
and comparison with eqn (10.71) indicates that C = 0. We recast the latter equation, which
holds throughout the membrane (including the transition region) in the form
8U/0A,&' = SPRAM, (10.82)
In any deformation that maps the cylinder to a cylinder (r’ = 0) this reduces to
dU/dA, = SRP) (1083)
which is just eqn (10.81). In this case elementary statics also provides t, = pr, or
0U/dA, = SR(phi A). (10.84)
The total strain energy stored in a purely cylindrical deformation of a membrane of ini-
tial length L is 27 RLU. As the initial volume is 2 R’L, the strain energy per unit of initial
volume is
E = (2/R)U. (10.85)
The volume contained by the deformed cylinder, per unit of initial volume, is v =
WrA,L/1R'L, or
vaa,Qi. (10.86)
Consider a uniform equilibrium deformation carrying the membrane from a cylinder of

radius r to a cylinder of radius r*. The induced change of strain energy is
E’-E = ze | dU = 2 OU/dA\dA, + OU/OAdA,, (10.87)
where the integration limits refer to the two states, Using eqns (10.83) and (10.84), we
reduce this to
E’-E = / 2priArdA, + parda,. (10.88)
Thus, from eqn (10.86),

+
E' -E -/ p(v)dv. (10.89)
To determine p(v) for a given membrane, we select the function U(A,, A.) and solve
eqns (10.83) and (10.84) for A,, say. We then solve the same system forp in terms of A,
and plotp against v = A,A2. Alternatively, fixp and solve eqns (10.83) and (10.84) for A,
and A,, and then plotp against v.
Problem
Carry out the details in the case of Ogden’s (1997) strain-energy function
3
U(Ay, a2) = So mila, dai a), (1090)

isd
where
(Ay, A232) = oc [AT + AS + (Aya) - 3], (1091)
with
@,=13, a@,=5.0, a, = -2.0 (10.92)
and
by/p, = 2.01 x 10%, s/f, = -1.89 x 10°. (1093)
Plot Rp(v)/,, against v.

PROBLEM | 111
Note: Ball (1977) has shown that this strain-energy function satisfies a condition
known as polyconvexity, which we shall discuss later. Furthermore, polyconvex-
ity is sufficient for strong ellipticity, and so Ogden’s (1997) function satisfies the
conditions we have assumed in the course of deriving membrane theory.
Returning to the coexistent phase problem, recall that eqn (10.82) holds throughout the
membrane. If p* is the pressure associated with the two-phase solution (Figure 10.6), then
in the uniform phases we have
(A,8U/9A,)* = (R/2)p*v*, (10.94)
whereas eqn (10.80) furnishes
(2/R)(A,0U/4A, - U)* = (2/R)(A,9U/9A, - UY. (10.95)
These combine to give
E'-E =p*(v'-v). (10.96)
Comparison with eqn (10.89) then furnishes the means to calculate p*. Thus,
piv-v)= / plv)dv. (1097)
This is the famous Maxwell equal-area rule for phase coexistence: The left-hand side
is the area of a rectangle of base v* -v and height p* and, of course, the right-hand
side is the area under the graph of p(v) between v and v*. One simply adjusts the
value of p* accordingly. In practice, one finds a unique pair (v*,v”) for which this con-
struction is possible, and simply reads off the associated value of p* from the graph
(Figure 10.6).
py)
Uv U,
Figure 10.6 Pressure-volume characteristic for purely cylindrical deformations. Shaded lobes
have equal areas
Problem
Complete the detailed analysis using results obtained in the previous exercise.
Further discussion of this problem, and other interesting examples of phase coexistence
in membranes, may be found in the references cited. We will take up Maxwell’s rule again,
in another context, in the next chapter.
REFERENCES
Ball, J.M. (1977). Convexity conditions and existence theorems in nonlinear elasticity. Arch, Ration.
Mech, Anal. 63, 337-403.
Ogden, R-W. (1997). Non-linear Elastic Deformations. Dover, NY.
Pipkin, A.C. (1968). Integration of an equation in membrane theory. ZAMP 19, 818-819,
Treloar, L.R.G. (1975). The Physics of Rubber Elasticity, 3rd edn. Oxford University Press, Oxford.
FURTHER READING
Chater, E. and Hutchinson, J.W. (1984). On the propagation of bulges and buckles.J. Appl. Mech. 51,
269-277,
Miller, I. and Strehlow, P. (2010). Rubber and Rubber Balloons: Paradigms of Thermodynamics.
Springer, Berlin.
Stamenovic, D, (1988). Equilibria of a pressurized membrane with a non-monotonic pressure-
volume characteristic. Q.J. Mech. Appl. Math. 41, 71-81.
Il
Stability and the energy criterion
In this chapter we elaborate on the notion of stability of equilibria with particular emphasis
on the energy criterion for conservative problems, i.e., problems in which the loading may
be associated with a potential energy, as in Problem no. 2 of Section 3.2. We show that for
an equilibrium state of a conservative system to be stable, it is necessary that it furnish a
minimum of the potential energy consisting of the strain energy and load potential. This
subject has a foundation in thermodynamics (see the paper by Ericksen (1966)). However,
in keeping with the theme of this book, we rely on a purely mechanical argument.
11.1 The energy norm

An immediate issue we must face is the matter of how to quantify stability and instability.
Intuitively we imagine that a state of equilibrium, say, is stable or unstable according as the
size of any disturbance to this state remains bounded, or otherwise, as time evolves. This
notion of size indicates that we must incorporate an appropriate norm into the definition of
stability/instability. In this respect, the study of stability for continuous systems is far more
intricate than it is for discrete or finite-dimensional systems. For, in the latter case all norms
are equivalent, whereas in the former there is no such equivalence. One can construct ex-
amples where a given state can be stable as measured by a given norm, but unstable in terms
of another. See the book by Como and Grimaldi (1995) and the treatise by Knops and
Wilkes (1973). Thus, it is immediately clear that no state can be stable without qualifica-
tion, ie., the judgment as to stability or otherwise is inherently norm dependent. Therefore,
it becomes necessary to choose a norm judiciously, to ensure the greatest degree of contact
with the phenomena at hand.
A choice of norm that presents itself almost immediately is the so-called energy norm
advocated by Mikhlin (1965) and subsequently studied by Como and Grimaldi (1995).
This meets the formal definition ofa norm while conferring a meaning that is intrinsic to the
problems considered. We develop its description in stages, starting with some elementary
observations. First, recall the definition eqn (9.6) of the tensor of elastic moduli; namely,
M = Wee, (1.1)

114 | STABILITY AND THE ENERGY CRITERION
from which it follows that M = P,. In other words, on any path F(), we have P = M[F].
In the same way we can define the strain-dependent moduli
C = Wer, (112)
where E is the Lagrange strain, i.e.,
E= 5l (FF-D,
it
(113)
and W(E) = W(2E +I) is the strain-energy function expressed as a function of strain.
Accordingly, C = S,, the derivative of the 2nd Piola-Kirchhoff stress § = W, with respect
to E. Here we adopt the convention that the derivative W, is an element of Sym. Like M, the
tensor of strain-dependent moduli possesses the major symmetry C' = C. Unlike M,, it also
possesses the minor symmetries A’ -C[B] = A- C[B] and A-C[B‘] = A- C[B]. Further,
S = C[E]. Applying the chain rule to the relation P = FS thus furnishes
. . 1 . .
M{E] = FS + SFCIFF + FF]. (114)
Due to the minor symmetry of C and because F is arbitrary, we then have
M([A] = AS + FC[F'A] (11.5)
for all A.
Consider a deformation path with F(0) = I. Then for small yz,
. 1 . .
W(F()) = W(D) + Po: Fy + 5h Mol Fo] “Fy + o(’), (11.6)
where Py, Mg, and F,, respectively, are the values of P, M, and F at F = I. Without loss of
generality we can impose W(I) = 0, as the value of W(I) may be adjusted arbitrarily at any
material point without affecting the values of measurable quantities. Suppose that Py(= S,)
vanishes. Then,
1 . .
W(F(u)) = 5 4°Co[Fo] «Fo + o(u"), (117)
where C, is the value ofC at E = 0.
Let x (2) be the deformation associated with gradient F(z), and let v(x) be the value of
X at w = 0. Then, F, = Vvand
1
W(F(u)) = 5h Cole] -€+o(u’), (118)
where
é = Sym(Vv) (119)
PROBLEM | 115
is the infinitesimal strain tensor. This is the strain measure used in linear elasticity theory.
We have invoked the minor symmetries of Cy to replace Vv by its symmetric part. It is
customary in that theory to take C, to be positive definite; that is,
C,[A]-A> 0 (11.10)
for all A with non-zero symmetric part (SymA # 0). We observe in passing that M,[A] -
A=C,[A] - Aunder our hypotheses, and so our assumption implies material stability of the
reference configuration, i.e., strong ellipticity at F = I.
Problem
Prove this claim; that is, show that ifA = a @ b 7 0, then SymA # 0.
Crucially, eqn (11.10) implies that C, is bounded below in the sense that
Clel-e >A lel’, (ul)
where A(x) is a positively valued scalar field. To see this we use the spectral theorem for
symmetric matrices. Let § be a 6-vector consisting of the components of € on any ortho-
normal basis. Let S be the symmetric 6 x 6 matrix consisting of the components of Cy on
the same basis. Then, from the spectral theorem,
BSE = SO ACE's)? > AD O(E's)? = AE, (11.12)
where s; are the (orthonormal) eigenvectors of S, A; are the associated eigenvalues and A. =
min{A,}. The latter is strictly positive because S is positive definite, and of course this is just
(11.11),
We define ||v|| by
IIvl? = [ostel -edv with e = Sym(Vv). (11.13)
This is called the energy norm. It is clearly intrinsic to the material at hand. To verify that it
is in fact a norm, we first use eqn (11.11) to infer that
IIvil? = i [lel ay, (11.14)
where A = min,<, A(x). This exists, and is strictly positive, because A(x) is positive and
continuous—assuming that Co(x) is likewise continuous, and because « is compact (i.e.,
closed and bounded) in three-space (see Palis and deMelo, 1982).
Korn’s inequality asserts the existence of a positive constant, k say, such that
k | lel av > [iveta, (11.15)
provided that v(x) vanishes on some non-empty subset of the boundary 8x. Poincaré’s
inequality is the assertion that
2 2
[iv dv > cf dv (11.16)
for some positive constant c, under the same restriction on v(x). Accordingly, there is a
positive constant a such that
IIv|l* > a [ we dv. (11.17)

«
Proofs of the Korn and Poincaré inequalities are sketched in Parts 7 and 8 of the
Supplement.
From this it follows that ||v|| vanishes only if the integral of Iv|? vanishes and hence only if
v vanishes, provided that v(x) is continuous. Conversely, it is immediate from the definition
of ||v|| that it vanishes if v(x) vanishes. It follows that ||v|{ is a positive-definite function
of v(x).
Problem
Show that ||-|| satisfies the triangle inequality.
Thus, ||-|| satisfies all the conditions required to qualify as a norm
11.2 Instability
Consider a motion x (x,t) and let x,(x) be an equilibrium deformation. Let u(x,t) =
x (x,t) - x-(x) be the displacement from equilibrium, and consider the associated function
G(t) = / p(x) ux, t)[ dv, Cus)

Suppose u(x, f) vanishes on some non-empty subset of the boundary x. Assuming p, (x)
to be continuous, we have, by the compactness of x,
G(t) < (maxp,) f JuCs t)| dy, (11.19)

INSTABILITY | 117
which implies, by eqn (11.17), that there is a positive constant, c, say, such that
G(t) < ¢ |lull’. (11.20)
This simple result suggests a strategy whereby we seek sufficient conditions for the un-
bounded growth of G(t). These would then ensure unbounded growth of ||ul| and, hence,
instability of the state x,(x) as judged by the energy norm. The negation of this result would
then yield necessary conditions for stability of equilibrium. Our procedure has an heuristic
aspect, on which we comment later. We begin by evaluating the derivatives
G= 2 f pou id (1121)
and
G = 4K(t) + 2 f ou - udy, (11.22)

K
where A(t) is the kinetic energy of the body. Ignoring body forces, which play a secondary
role in the argument, and assuming for simplicity that there are no local constraints, we have
fp,ii = DivP(F;x) = Div(AP), (1123)
where
AP = P(F;x) — P,(x) (11.24)
in which P,(x) = P(V x,(x);x) is the equilibrium stress. Accordingly,
G =4K(t) + 2 fu - Div(AP)dv. (11.25)
At this stage, we simplify matters by confining attention to mixed displacement/dead-load

problems in which u vanishes on a part of the boundary and the Piola traction is fixed, ie.,
(AP)N vanishes, on the complementary part. We integrate the second term on the right-
hand side by parts using the divergence theorem, obtaining
G=4K(t)-2 / AP - AFdy, (11.26)

K
where AF = Vu.
Next, assume there exists u(x, ¢) such that F(x,t) € B(F,(x)), where B(F,) is the open
ball of radius 6 centered at F, defined by
B(E,) ={F: |Vu] < 5}. (1127)

We note that B(F,) is a convex set; that is, if F, and F, are in B(F,), then all points on the
line F(u) = uF, + (1 - u)F,, with u € [0,1], are also in B(F,). To see this we use
F(u) - F, = u(F, - F,) + (1 - u)(F, - F,) (11.28)
and conclude, from the triangle inequality, that
|F(u) - F,| < u|F, -F.|+ (1-4) |F,-F,| < ud + (1-u)d = 6. (11.29)
Thus, F(u) € B(F,) and so B(F,) is convex by definition. Note, however, that the domain
of the constitutive function P, namely, the set of F’s with detF positive, is not convex.
Therefore, our assumption on u(x, t) is restrictive.
Problem
Prove the nonconvexity of the set Lint = {F: det F > 0}. Hint: Consider
F,=-3i1@itj@j-k@k, F, =i@i-3j@j-k@k, (11.30)
where {i,j,k} is an orthonormal set; both have positive determinant. Show that
F(2)2 = -1.
If the strain-energy function is twice continuously differentiable, then Taylor’s theorem

with remainder (see Fleming, 1977), which is valid in any convex region and, hence, in
B(F,), furnishes
P(F) = P(F,) + M*[AF], (1131)
where M* = M(F, + @* AF) with 0 < a* < 1. The same theorem yields
W(E) = W(E,) + P(E.) AF + 5M" AF] . AF, (1132)
where M*™* = M(E, + a** AF) withO < a** < 1. Then,
AW =P,- AF+ ~AP » AF + “£{AF] - AF, (11.33)
where £ = M*™ — M’*, and substituting into eqn (11.26) yields
G = 4K(t) -4 raw —P,- AF)dv- 2 f clarl - AFdv. (11.34)
Recalling Problem no. 2 of Section 3.2, the potential energy of a dead-loaded body is
e= | wav- [ p- Xda, (11.35)

x a.
”
PROBLEM | 119
where p(x) is the fixed Piola traction on the part 9x, of the boundary. This part is com-
plementary to the part on which u vanishes. Conservation of energy (Problem no. 2,
Section 3.2) yields
<¢ +K)=0. (11.36)
The change of potential energy accompanying the displacement u is
aé= [ awér- | p-uda, (1137)

axp
where
/ p:uda = / Plu - Nda = [rceman = pe. - AFdv, (11.38)

oKp OK K kK
and we have used the fact that DivP, vanishes. We have
G = 8K(t) - 4H(t) - 2 f clar) - AFdv, (11.39)
where
H(t) = K(t) + AE(E) (11.40)
is the total change in mechanical energy induced by the displacement. This is fixed in time,
ie, H(t) = 71(0), by virtue of eqn (11.36).
Consider a motion with vanishing initial velocity: u(x,0) = 0. Then, 7((0) = A&,,
the potential energy difference induced by the initial displacement u)(x) = u(x, 0).
Combining, we obtain
G = 8K(t) - 4A, - 2 f cive) - Vudv. (141)

K
We have
[ive - Vudv < max (x, ) fiver dv, (11.42)
where {4(x, t) is the largest absolute value of the members of the set of eigenvalues of £. This
is a real number because £ possesses major symmetry and can, therefore, be associated with
a9 x 9 symmetric matrix. To see this we use an argument like that used in eqn (11.12). This
time let S be the symmetric 9 x 9 matrix consisting of the components of C, and let s, and
i, be its eigenvectors and eigenvalues. Let & be the 9-vector consisting of the components
of Vu. With = max{|A,|} we have
ESE = DAEs) < Sola E's)? sm E's = HEP, (a)
and, hence, eqn (11.42). Our assumption that £ is continuous, together with F € B(F,),
implies that 2 is bounded (see Proposition 2.18 in the book by Palis and deMelo, 1982).
Accordingly, the Korn inequality yields the existence ofa non-negative constant c,, say, such
that
[cove] -Vudv < ¢ |lull’. (11.44)
Because KC(t) > 0, we conclude that
G>-4A&+R, with R= O(|lull’). (11.45)
Weare finally ready to establish our main result. Suppose the equilibrium state is stable in
the sense that |lul| < € for some assigned e, for all t > 0. Let up(x) be such that AE, < 0,
ie, ELx, + uo] < E(x,], and suppose that x,(x) + uo(x) is not equilibrated, i.e., G(0) 7 0
(cf. eqn (11.22)). Then, we can choose € small enough to ensure that
-4A&,+R> A, (11.46)
a positive constant. It follows that G(t) > A and, hence, from eqns (11.20) and (11.45),
that
1
IJul|? > zor + C4 (11.47)
for some positive constant c, and some constant c,. We have used the fact that G(0) =0
(cf. eqn (11.21)). Thus there is a time, t, say, such that | u(x, t,)| > &, contrary to the
stability hypothesis. We have thus shown that if xo(x) = x(x) + u(x) isanon-equilibrium,
kinematically admissible displacement field (uy = 0 on the complement of dx,), then for
the equilibrium configuration x,(x) to be unstable with respect to the energy norm, it is
sufficient that up(x) be such as to furnish AE, < 0.
The negation of this statement furnishes a necessary condition for stability: If x,(x) is
stable with respect to the energy norm, then it is necessary that
E[x.] < Elxo] (11.48)
for all non-equilibrium deformations x (x) that meet assigned position data on the com-
plement of dx,. This is the famous energy criterion of elastic stability theory. It effectively
reduces the equilibrium problem to the central problem of the Calculus of Variations:
Find a vector-valued function that minimizes a scalar-valued functional. Here, of course,
the potential energy. Needless to say, the criterion is meaningful only for conservative
problems.
QUASICONVEXITY | 121
The energy criterion furnishes the point of departure for a vast amount of modern re-
search on finite elasticity. Indeed, the question of the existence of energy minimizers was
settled in the landmark paper byJ. Ball (1977). This constitutes a major milestone in 20th
century research on Finite Elasticity and the Calculus of Variations more broadly.
Of course, we have made a number of questionable assumptions in the course of deriving
the energy criterion, not least among these being the existence ofa motion having gradient
in the ball B(F,(x)) at each x € x. Beyond this, we have assumed this motion to be suffi-
ciently smooth in space and time as to justify our various formal manipulations. To this day,
conditions ensuring such global regularity are not known. Nevertheless, our development
offers a formal justification of the energy criterion and thus provides a degree of confidence
in its validity. The work of Koiter (1966), in particular, offers arguments in support of this
criterion as being both necessary and sufficient for stability.
Problem
Consider the problem of the extension of a unit cube of Mooney-Rivlin material
under equibiaxial forcef (cf. Problem no. 6(b) in Section 7.2). (a) Show that the
potential energy of the deformed material is
E(A,,A2) = w(A, A.) -fhy - fas, (11.49)
where w(A,, A) is the three-dimensional strain energy obtained by imposing the

incompressibility condition A,A.A3 = 1. (b) Show that equilibrium corresponds to
stationarity of the energy; i.e., JE/dA, = 0 for a = 1,2. (c) Let A* be the critical
value of equibiaxial stretch at which bifurcation from equibiaxial to unequal biaxial
stretch is possible. Show that the solution with equibiaxial stretch (A, =A, =A,
say) is stable if A < A* and unstable if 1 > A*. Show that the solution with un-
equal biaxial stretch is stable. Thus, in practice the cube undergoes a transition from
equibiaxial to unequal biaxial stretch at A > A*.
Hint: The energy is minimized at an equilibrium state if and only if the matrix
0°E/0A, 8A, is positive definite. Thus, failure of positive definiteness marks the
transition from stability to instability.
Remark By restricting attention to deformations that are homogeneous, we have
provided only a partial analysis of stability, because we have not allowed non-
homogeneous deformations into the competition for the minimum of the
energy. In other words, it is conceivable that a non-homogeneous deform-
ation might be able to produce a lower-energy state than the equilibrium states
considered here.
11.3, Quasiconvexity
We inquire into some implications of the energy criterion eqn (11.48). Staying with mixed
displacement/dead-load problems for illustrative purposes, this criterion may be written in
the form
[we + AF) - weld | p: Axda > 0, (11.50)

K Oxp
for all Ax that vanish on the complement of dx,. Among these, we consider
Ax (x) = e@(z);_ z= 8" (x-%) (1151)
with € a positive constant, Xo an interior point of « and @ #0 only in some region D

contained in the interior of «. Note that the the nature of the boundary load potential is
irrelevant for this choice of Ax and hence that our further considerations are valid in more
general cases of (conservative) loading.
We have AF = V@(z) in this case, where the gradient is taken with respect to z.
Accordingly, eqn (11.50) reduces to
[twee + VO@sx) - W(EG);x) Ja) = 0 (159

where x’ is the image of D under the inverse of the map eqn (11.51),. Note that the map
defined by eqn (11.51). has gradient eI, with determinant e°. Then, dv(z) = e*dv(x)
and, after division by €°, we derive
[wet + €2z) + Vb(z)jXo + €z) — W(F, (Xp + €z)} Xp + Ez) |dv(z) > 0. (11.53)
We now let € — 0 and use the dominated convergence theorem (see Fleming, 1977) to
conclude that
/ [W(F.(x) + Vb(2); x0) ~ W(F.(%); x) Jdv(z) > 0 (uss)

for all x9 and for any @ with @ = 0 on OD. This is the quasiconvexity condition. It was
discovered by Morrey, and plays a major role in Ball’s existence theorem for energy
minimizers. A particularly good account may be found in Ciarlet (1988).
We have derived quasiconvexity as a necessary condition for energy minimization. It
has an interesting physical interpretation: Consider a uniform material with strain energy
W*(E) = W(E; x). Then, quasiconvexity means that the energy
F[@] = | wees + V0)de, with @=0 on OD and F,=F,(x%), (1155)

D
is minimized absolutely by @ = 0, i.e., by the homogeneous deformation Fyz. This is pre-

cisely the potential energy for a pure displacement boundary-value problem in which
perturbations of the deformation vanish on the entire boundary.
Note that the quasiconvexity condition imposes a restriction on the deformation F,(x)),
for each x) € «; if there are any points where it is violated, for some ¢, then x,(x) cannot
ORDINARY CONVEXITY | 123
be an energy minimizing deformation. However, the quasiconvexity condition is manifestly

non-local and this poses an obstacle to its direct verification. This fact has given impetus to
the search for purely local conditions that imply quasiconvexity.
11.4 Ordinary convexity

One of these local conditions is ordinary convexity, ie.,
W(F, + Vj X%) - W(Fo; %) = We(Fo; %) - VO. (11.56)
We confine attention to unconstrained materials. Then,
/ [W(E + V0; x0) — W(Eoj x0) dv > BCP; x) / Vo(2)dv

= P(F,; My) - [ od © Nda, (11.87)
and this vanishes if @ vanishes on 0D. Accordingly, convexity is sufficient for quasicon-
vexity. However, this condition suffers from major drawbacks and, therefore, cannot be
regarded as realistic. We elaborate on two of these here.
11.4.1 Objections to ordinary convexity

(a) Consistency with the symmetry of the Cauchy stress implies that compressive states of
stress are impossible.
To see this we write eqn (11.56) in the form
W(E) - W(E) > B(F) - (F- F), (11.58)

which purports to hold for all F and F with positive determinant. Therefore, it holds if
F = QF with Qa rotation. We have seen that W(F) = W(EF) in this case if and only if the
Cauchy stress T is symmetric. Thus,
0 > P(F) - [((Q-DF] = tr{P[(Q-1F]}

= [PF'(Q-1)'] =JT-(Q-D), (11.59)
where T is the Cauchy stress associated with F. Consider a one-parameter family of ro-
tations Q(e) with € € (-&,&)) and Q(0) =I, and define f(e) = T - [(Q{e) - 1]. Then,
f(0) = 0,f'(e) = T- Q(e) andf’(e) = T - Q’(e). Note that 2(e) = Q'(e)Q(e)' is skew
symmetric. We have Q’(e) = Q(e)Q(e) and
Q"(e) = M(e)Qle) + Q(€)’QE), (11.60)

where &'(€) is also skew symmetric, as are w = Q(0) and a = 2'(0). Accordingly,
Q‘(0) = w and Q’(0) = a + w’. These imply that f’(0) = 0 and f’(0) = T - w’, by virtue
of the symmetry of T.
Inequality eqn (11.59) then implies, for small ¢, that
1
0>f(e) = 5° Yf"(0) + o(e’)/e"]. (11.61)
Dividing by € and passing to the limit, we conclude that f”(0) < 0, ile,
T-@ <0, (11.62)
for all skew w.
Problems
1. Show that eqn (11.62) is equivalent to
w(t, +h) + @i,(f +) + @,(b +b) = 0, (11.63)
where t, are the principal Cauchy stresses and w, are the components of w on the
principal stress axes. Therefore, convexity and the symmetry of the Cauchy stress
together imply that the state of stress is tensile in the sense that
{+t >0, t+f>0 and 4A+f,>0 (11.64)
pointwise, at all deformations. This is an unrealistically severe restriction to

impose on a general boundary-value problem.
2. Consider a homogeneous, compressible elastic material (W not explicitly de-
pendent on x) subjected to a prescribed dead-load traction distribution p(x) on
its entire boundary.
(i) Show that, in the absence of body forces, a necessary condition for the
existence of an equilibrium deformation x (x) is that f,, pda = 0.
(ii) Suppose the equilibrium deformation x is uniform in the sense that its gradi-
ent F is independent of x. As we have shown, the change in potential energy
associated with a kinematically admissible deformation x — x is
AE = [owen _W(E)-B(F)-(Vx-F)]dv. (1165)

Show that AE > 0 for arbitrary kinematically admissible x if, and only if, the
integrand is non-negative at every point of the body. Thus, conclude that in
this case, contrary to the situation in mixed position/traction problems, F
must satisfy the condition of ordinary convexity.
PROBLEMS | 125
(iii) Use this result to prove that a homogeneously deformed bar in equilibrium un-
der equal and opposite compressive tractions at its ends (and zero tractions
on the remainder of its boundary) is unstable.
(b) Strict convexity precludes buckling in the mixed dead-load problem.
Using p = PN on dx, and Ax = 0 on the complement of 0«,, we can write
[owe + AF) - W(F,) ]dv- / p:- Axda

P
= [twee.+ aR) -wee)lav- [ P'Ay + Nda

ax
= [owe + AF) - W(E.) - P(F,) - AF]dy, (11.66)
where we have invoked DivP, = 0 in the course of integrating by parts. From this is it ob-
vious that strict convexity; ie., strict inequality in eqn (11.56), implies that AE > 0 for
any non-zero Ax that vanishes on the complement of d«,. Therefore, strict convexity is
sufficient for an equilibrium state to furnish a strict minimum of the potential energy.
Let F, = F, and F, = F, + AF be the gradients of two equilibrium deformations x, and
X2, respectively, and suppose x, is a strict minimizer. Then,
/ [W(E,) -W(E,) -B(F,) - ( -F,)]dv > 0, (1167)
provided that F, 7 F,. In the same way, if x, is a strict minimizer, then
/ [w(P,) - W(F,) - B(F,) - (F, - F,)]dv > 0,

K
(1168)
again, provided that F, 7 F,. We re-write this as
- ftw0e,)- we) - 2) - (B,-B)av > 0 (16)

and add it to (11.67), concluding that
[ee - B(F,)- (F,-F,)]dv > 0, providedthat F, 7 F,. (11.70)

K
However, since Div(AP) = Oin« and (AP)N = 0on 6x,, where AP = P(F,) - P(F,);
and, since Ax (= x2 ~ X:) vanishes on the complement of 3x,, we have
o= f (apyar Nda= f AP. APay (11.71)

OK K
that is,
[ee - P(F,) - (F, - F,)]dv = 0, (11.72)

K
which is reconciled with eqn (11.70) if and only if F, = F,. Then, x. - x1 =¢, a rigid
translation of the entire body. If the complement of dx, is non-empty, c vanishes and
the equilibrium deformations coincide. We conclude that if an equilibrium deformation
minimizes the energy strictly, then it is unique. In particular, then, strict convexity of
the energy implies unqualified uniqueness of equilibria in the mixed position/dead-load
problem. Thus, strict convexity precludes buckling—the phenomenon of bifurcation of
equilibrium—under all dead loads, and is, therefore, unrealistic.
Nevertheless, the negation of this result furnishes the well known Euler criterion for po-
tential instability: Non-uniqueness of equilibria (e.g., buckling) implies that they are not
strict minimizers of the energy, i.e. that AE < 0 relative to the considered equilibrium
deformation, for some Ax that vanishes on the complement of 0x,. Accordingly, such equi-
libria fail to satisfy the strict form of the necessary condition eqn (11.48) for stability and
might, therefore, be unstable. Said differently, the onset of non-uniqueness of equilibria
signals a potential instability. We cannot assert definitively that non-uniqueness implies in-
stability of equilibrium because the strict form of eqn (11.48) is not known to be necessary
for stability. The Euler criterion is studied in more detail in Chapter 12.
11.5 Polyconvexity
This is the statement that there exists a function P(F, F*, det F), jointly convex in its
arguments, such that W(F) = ®(F, F*, det F), ie,
W(F) - W(Fo) > ACE) - (F - Fy) + B(F,) - (F* — F*) + C(Fy)(det F - det Fy), (11.73)
with
A(F)=®,, B(F)=@, and C(F) = gap. (11,74)
Consider a deformation with gradient F = F, + V@, where F, is uniform. Then,
[wees + 90) - we)

D
> A(F,) : / Vodv + B(F,) - fe - F})dv + C(F) [(ooce -detF,)dv. (11.75)

D D D
We assign @ = 0 on dD. We show that the right-hand side vanishes, and thus that polycon-
vexity is sufficient for quasiconvexity. To this end we derive three identities:
POLYCONVEXITY | 127
(a) to Vodv = Sop ¢ ® Nav. This of course is a variant of the divergence theorem.
Obviously, the integral vanishes because # vanishes on the boundary.
(b) Recall that Ft, = Wisgs with Wiss = 5 ¢je€ascX;Xu,c = —Wasa, where e is the permuta-
tion symbol (e,.; = +1, etc.). Thus, by the divergence theorem,
[ (FY - Fi)dv =, @E, | (thas - W°,)Noda. (1176)

D aD
Let v(x) bea scalar field and consider the vector ésac¥,cN = (N x Vv),.Wehave
N x Vv =N x P(Vv) on 2D, where P = I- N @ Nis the projection onto the
local tangent plane of 0D. Thus, N x Vv involves only the tangential derivatives
of v in the surface, which in turn are determined by the values of v on the sur-
face. Choosing v = Xziap = Xiiaps we find that WissNs = W,Nzs on OD because
vanishes there. Thus, f, (F* — F)dv vanishes.
(c) It is elementary to show that detF = eesscFiaFiaFec = 5 FiaFi, = SCGFR) a» a
divergence. We have, of course, invoked the Piola identity. Then,
1 1
[ sccear = ;/ x -F*Nda = >| x nda, (1177)
D 3 Jap 3 J x(a)
where x (D) is the image of 0D under the deformation map and we have used
Nanson’s formula. Because @ = 0 on 0D the same result follows on replacing F
by F,. Accordingly, J, (det F — det F,)dv vanishes.
We have shown that polyconvexity implies quasiconvexity. Indeed, polyconvexity is cen-

tral to the hypotheses underpinning Ball’s existence theorem. It does not suffer from the
drawbacks of ordinary convexity. For example, Ball (1977) has shown that Ogden’s strain-
energy function satisfies polyconvexity. Further examples of polyconvex strain energies are
discussed in the papers by Steigmann (2003a,b).
Thus far, we have shown that both ordinary convexity and polyconvexity imply quasicon-
vexity. However, polyconvexity does not imply convexity, as shown by the following
counter-example. In general, we have
W(E + AF) - W(B) - P(F)- AF = 5 M(F)LAF] . AF + o(|AF|’). (11.78)
Set AF = OAwith
A fixed. Then if W(F) is convex,
SP(M@PIAl -A+0(6°)/6} > 0. (1179)

Dividing by 0? and passing to the limit, we conclude that W(F) is convex only if
M(F)[A]-A>0 (11.80)
for all A; in other words, only if the function G(@) = +6°M(F)[A] - A satisfies G"(@) > 0.
Consider the deformation gradient
F(O) = O(21@it+j@j+rk@k)+(1-A)G@ir+2j@j+k@k) (11.81)
with 6 € [0,1]. This has det F(@) = (1+0)(2-6) > 0, and is, therefore, an admissible
deformation gradient.
Consider W(F) = det F, which is trivially polyconvex. Picking 9A = F(@) — F(0), we
find that G(@) = -36?. Then, G’(@) = -1, proving that det F is not a convex function of F.
11.6 Rank-one convexity

Consider the region D C x in the definition of quasiconvexity. We follow a construction
due to Graves (1939), but confine attention to two dimensions for the sake of simplicity.
Graves’ treatment is valid in » dimensions. Let the origin of z be located at x, and attach
an orthonormal basis {M, N} there. Consider a lens-shaped region R C D as shown in
Figure 11.1, and letx = M:zandy =N-z.
Then,
R= {(x,y): |x] <h, O<y <Bix)}=R, UR, (11.82)
where
R,={(xyy): |x] <h, O<y <OB(x)}, R,=R\R,. (11.83)
with 0 < @ < land B(x) = h? — x’. Consider the function
(1-@)y in R,
F(z) = -6[y- B(x)] in R, (11.84)
0, outside R.
| y = B(x) =h? - x?
he y = OB(x)
Bd
-h h
Figure 11.1 Lens-shaped region

RANK-ONE CONVEXITY | 129
Note thal
6(1-@)B at y=6B
F(z) = -6B(0-1) at =6B’ (11.85)
so thatF is continuous.
We define
$(z) = aF(z), (11.86)
with a fixed. Its gradient is
Vo =a® VE, (11.87)
where
VF = F,Vx+F,Vy =F.M+5E,N
_ (1-@)N in R, (1-9@)N in R,
= _9N-6(2x)M in R, -ON in R, ’ (11.88)
ash — 0. Wealso have F > Oash — 0. Thus,
(1-8@)a@N in R,
$2) > 0 VO>" gv@N in R (11.89)
Let
V = measR = / B(x) dx,

ah
h
V, = measR, = / OB(x)dx = OV,

-h
V, = measR, = (1-0)V. (11.90)
Then, the quasiconvexity condition yields
o<v" [we +V¢) -W(R,) dv

- —8 W(F,
+ V¢)dv + 1 =-@ [ W(Ro+ VO)de— =1 | w(E.)av (1191)
V, Ri
Using the mean-value theorem in each integral, letting h —> 0 and dividing by 0, we obtain
the pointwise condition
1-0 1
W[EF,
+ (1-@)a@N]+ pW (Fo - 6a @ N) - 9 WF) >0; O<6@ <1. (1192)
For small 0,
W(Fy
- 9a @ N) = W(F,) - O9P(F,) -a@N
+ 0(8), (1193)
and, therefore,
1-0 a 0
Wl - 6a @N)= wr) - (1-6)P(F,)-a@N+(1- a) ) (11.94)
It follows from eqn (11.92) that
WI[F) + (1 -@)a @N] —- W(F,)- (1 -6)P(F,) « sen (1-0) oo, (11.95)
and passage to the limit yields
W(F, +a@N) - W(F)) - P(E) -a@N>0 (11:96)

in which a @ Nis arbitrary.
This is the condition of rank-one convexity, so named because it requires the convexity
of W with respect to rank-one perturbations of the deformation gradient. We have derived
it as a consequence of quasiconvexity. Accordingly, it is necessary for quasiconvexity, and
therefore necessary for minimum energy. In principle, it constitutes a restriction on the
value of the deformation gradient F, = F,(x9), at each x» € «. If there is any a@ N such
that inequality eqn (11.96) is violated at some Xq, then the field x,(x) cannot be an energy
minimizer and, hence, cannot be stable. In particular, eqn (11.96) does not constitute a
restriction on the function W. That is, it could be violated at some deformation gradients in
the domain of the strain-energy function. Such gradients must then be relegated to sets of
measure zero in « if the deformation is to minimize the potential energy.
In the case of incompressibility eqn (11.96) remains valid with the restriction a - F7N =0.
The derivation in this case may be found in the paper by Fosdick and MacSithigh (1986).
Problem
Show that det(F + a @ N) = detF+a-F'N.
As N is a unit vector we have
a 1
W(F, + a@N) - W(E,) - P(E.) -a@N = 5M(Fo)[a @ NI -a@N+o((al).
(11.97)
Dividing eqn (11.96) through by lal” and passing to the limit yields the Legendre-
Hadamard inequality
a: A(Fy, N)a > 0, (11.98)
where A(F,; N) is the acoustic tensor defined by eqn (9.22). If the deformation is such that
Legendre-Hadamard condition is violated at any point in the body, then it is not an energy
EQUILIBRIA WITH DISCONTINUOUS DEFORMATION GRADIENTS | 131
minimizer. The strict form of this inequality coincides with the material stability condition
of Chapter 9.
11.7 Equilibria with discontinuous deformation gradients

Phase transformations in so-called shape-memory alloys are characterized by deformations
with discontinuous gradients. The characterization of such deformations and the condi-
tions under which they arise constitute a major branch of research in nonlinear elasticity.
The paper by Ball and James (1987) is essential reading in this area.
We know from our earlier discussion that if a deformation x (x) is continuous with
a gradient F(x) that jumps across one or more surfaces in the body, then discontinu-
ities in the deformation gradient have the structure [F] = a ® N (cf. eqn (9.44)), where
N is a unit normal to such a surface. If the deformation is equilibrated, then it is ne-
cessary that Sos pda = 0, where S is any region in the configuration «. We omit body
forces, but they may easily be included without affecting our conclusions. We apply this
to a pill-box of thickness h containing a surface of discontinuity with normal N (see
Figure 11.2).
Because of the linearity ofp with respect to the normal, the limit ofp from the side of the
surface with normal -N is p_(-N) = -p_(N). Letting h tend to zero, we then find that
/ [p]da = 0, (11.99)
where
[p] = p,(N) - p.(N) (11.100)
is the jump in traction and where z is an arbitrary subsurface of the discontinuity surface.
The localization theorem yields the conclusion that
Figure 11.2 Pill-box surrounding a surface of discontinuity

[p] =0 (11.101)
at all points of the discontinuity surface. In the case of an unconstrained elastic material
this is
P(F,)N = P(F_)N. (11,102)
We seek conditions on the strain-energy function ensuring that this equality can be satisfied
with [F] 40.
Fix a point on the surface of discontinuity and define
F(u) = uF,
+ (1-u)F. =F. +uaQ@N (11.103)
withu € [0,1]. Let
f(u) = det F(u) = det F_ det(I + uF"'a @ N)

= det F_(1 + uF 'a-N). (11.104)
Observe that f(0) = det F. > O and f(1) = det F, > 0. Then, because the graph of f(u) is
a straight line between these endpoints, we have f(u) > 0; therefore, F(u) belongs to the
domain of W.
Next, define
g(u) =aQ@N-
P(E. +ua@N) =a- P(F.
+ ua@N)N. (11.105)
Then,
g(0) = a: P(F_)N =a- P(F,)N = g(1). (11.106)
Because g(x) is differentiable, by Rolle’s theorem there is 4) € [0,1] such that g’(u,) = 0,
ie.,
0 = M(F(u))[a@N] -a@N=a- A(F(m%), Na, (11.107)
and so there is a deformation gradient in the domain of W where the strong ellipticity condi-
tion is violated. Thus, ifsuch a discontinuity is to exist, the strain energy cannot bea strongly
elliptic function, i.e., it cannot be strongly elliptic at all points in its domain. Because strong
ellipticity is necessary for the strict form of the rank-one convexity condition, it also cannot
be a strictly rank-one convex function.
11.8 The Maxwell-Eshelby relation

Any deformation gradient occurring in an energy-minimizing deformation field must be
such as to satisfy the rank-one convexity condition pointwise. Deformation gradients at
THE MAXWELL-ESHELBY RELATION | 133
which this condition is violated are relegated to sets of zero volume measure, suchas discon-
tinuity surfaces, where they can make no contribution to the overall energy. In particular,
the limiting values F, on either side of a discontinuity surface must satisfy rank-one con-
vexity if the deformation field is such as to furnish a global minimum of the energy. For
example,
W(F_+a@N)-W(E.) >a-P(F_.)N (11.108)
for all a @ N. Choosing a @ N = [F], we infer that
W(F,) - W(F_) > a- P(E)N. (11.109)
In the same way,
W(F, -a@N) - W(F,) > -a- B(F,)N, (11.110)

or
W(E,) - W(E_) < a- P(F,)N. (ith)
Invoking eqn (11.102), we arrive at the Maxwell—Eshelby relation
W(F,) - W(E_) = P(F,) - (F, -F.). (11.112)
11.8.1 Example: alternating simple shear

Consider a deformation with piecewise uniform gradient
F, =1+ Yi @j (11.113)
with y. = -y,. These are simple shears of alternating sign. Here, j is the normal to the
plane of shear and i is the direction of shear. We can imagine a laminate consisting of such
alternating shears, extending over a volume of the body (Figure 11.3).
Ys
Y-
Ve
Figure 11.3 Alternating simple shear

We have
[F] = [y]i @j, (11.114)
and so N =j and a = [y ]i, whereas the traction-continuity condition eqn (11.102) fur-
nishes [Pj] = 0 and, hence
[rt] =0, (11.115)
where T = i- Pj is the shear stress on the plane of shearing.

The strain energy in simple shear is w(y) = W(I + yi @ j). Thus,
tT=P-i@j=W,-F(y) =w(y), (11.116)
and eqn (11.115) furnishes
t(y.) = t(y.). (1L117)
The Maxwell-Eshelby relation reduces to
w(y.) -w(y.) = Ps - [F] = [y]i- Pj. (11.118)

This is Maxwell’s equal area rule, in the form
V+
/ t(y)dy = t*(%- ¥-), (11.119)
y-
with t* = t(y,), requiring that the area under the shear stress vs amount-of-shear curve
equal that of the rectangle with base y, — y_ and height t*. If the material properties possess
reflection symmetry with respect to the discontinuity surface, then w(y) is an even function
and t(y) is odd.
The rank-one convexity condition implies that in each separate phase of uniform shear,
WI + (vy + Ay)i@®j]-W(I+ yi@j) = Ayi- BU + yi@j)j, (11.120)
for all Ay, or
wy + Ay)-w(y) = Ayw'(y), (11.121)

so that any value of y appearing in an energy-minimizing state belongs to a domain of
convexity of the function w(-). The situation is then as depicted in Figure 11.4,
Shears y € (y., y,) are unstable and are thus excluded from the deformation field per se.
Note that the response of the material in this case cannot distinguish between w(-) and its
convexification, the lower convex envelope of w(-). A fuller discussion of this and similar
problems may be found in Ericksen (1991).
PROBLEM | 135
(b)
Figure 11.4 (a) Stress and (b) energy as functions of amount of shear.
Problem
We have seen that a compressible inviscid fluid is an elastic material with a strain—
energy function of the form W(F) = w(J), with J = det F. Suppose the fluid is
uniform in the sense that the same function pertains to every material point. The
fluid fills a rigid container completely and no body forces are acting. Then, the
potential energy of any configuration x (x) of the fluid is
E[x]= [wea (11.122)
where « is the region enclosed by the rigid container.

(a) Prove that an equilibrium deformation x (x) is a minimizer of the potential
energyif, and only if, ](x) = det[V
x (x)] satisfies w(J) - wJ) > (J -J)w’(J)
for any J > Oand forallx € x.
(b) Let v = 1/p be the specific volume (volume per unit mass). Suppose that at
a certain fixed temperature the constitutive function for the pressure, p(v), is
as depicted in Figure 11.5, wherein the shaded lobes have equal areas. Give a
complete analysis of the stable equilibrium states of the fluid at the pressures
Pw Pv, and p,. Identify those aspects of the state of the fluid that are uniquely
determined in each case.
Py
P,
—_—_ uo v
liquid vapor
Figure 11.5 Non-monotone pressure-volume characteristic

REFERENCES
Ball, J.M. (1977). Convexity conditions and existence theorems in nonlinear elasticity. Arch. Ration.
Mech. Anal. 63, 337-403.
Ball, J.M. and James, R.D. (1987). Fine phase mixtures as minimizers of energy. Arch. Ration, Mech.
Anal. 100, 13.
Ciarlet, P.G. (1988). Mathematical Elasticity. North-Holland, Amsterdam.
Como, M. and Grimaldi, A. (1995). Theory of Stability of Continuous Elastic Structures. CRC Press,
Boca Raton.
Ericksen, J.L. (1966). A thermo-kinetic view of elastic stability theory. Int. J. Solids Struct. 2, 573-580.
Ericksen, J.L (1991). Introduction to the Thermodynamics of Solids. Chapman and Hall, London,
Fleming, W. (1977). Functions of Several Variables. Springer, Berlin.
Fosdick, R.L. and MacSithigh, G.P. (1986). Minimization in incompressible nonlinear elasticity
theory. J. Elasticity 16, 267-301.
Graves, L.M. (1939). The Weierstrass condition for multiple integral variation problems. Duke Math.
J. 5,556.
Koiter, W.T. (1966). Purpose and achievements of research in elastic stability. In: Recent Advances in
Engineering Science, pp. 197-218. Gordon and Breach, London.
Knops, RJ. and Wilkes, E.W. (1973). Theory of elastic stability. In C. Truesdell (Ed.): Solid
Mechanics, Vol. II], pp. 125-302. Springer, Berlin.
Mikhlin, S.G. (1965). The Problem of the Minimum ofa Quadratic Functional. Holden-Day, San
Francisco.
Palis,J. and deMelo, W. (1982). Geometric Theory of Dynamical Systems. Springer, Berlin.
Steigmann, D,J. (2003a). On isotropic, frame-invariant polyconvex strain-energyy functions. Q, J.
Mech. Appl. Math. 56, 483-491.
Steigmann, D,J. (2003b). Frame-invariant polyconvex strain-energy functions for some anisotropic
solids. Math. Mech. Solids 8, 497-506.
FURTHER READING
Hilgers, M.G. and Pipkin, A.C. (1992). The Graves condition for variational problems of arbitrary
order. IMA J. Appl. Math. 48, 265-269,
12
Linearized theory, the second variation

and bifurcation of equilibria
We have already studied the linearized theory in some detail in Chapter 9. Here we elab-
orate on this theory in the case of equilibrium. That is, we study the linear theory of small
equilibrium deformations superposed ona (finitely deformed) equilibrium state. This topic
is often referred to as the theory of small deformations superposed on large. Ogden (1997)
is the main source for this theory and should be consulted for further developments. Our
purpose in discussing it is to outline a practical implementation of Euler’s non-uniqueness
criterion for potential instability, also known as buckling. This criterion is a cornerstone of
engineering analysis.
In Chapter 9 we outlined the theory for incompressible materials. Here, we generalize by
imposing a constraint of the form $(F) = 0 onall admissible deformations. Let y,(x) be an
equilibrium deformation, and consider the static perturbation
1
x (x €) = y,(x) + eu(x) + 56 vex) + o(’) (12.1)
with € € (-€, €9) and |é)| « 1. Here,u = 2y;..,v = EF Keos etc. Then,
1
F(x; €) = Fy(x) + € Vu(x) + 56 Vv(x) + o(e’), (12.2)
and this must be such that @(F(x; €)) = 0 forall € € (—€p, €,). Accordingly,
0= Pie-0 = br(Fo) -Vu, O= Pico = Grr (Fo) [Vu] »~Vut+ dr(Fo) - Vv, ete. (12.3)
which constitute constraints on Vu and Vv. Note that there is no solution if x (x; €) is
linear in €, i.e., if v(x) vanishes. For, frame-invariant constraints are inherently nonlinear
(i.e; Par 7% O—see the Problem in Section 6.1) and a purely linear perturbation Vu is thus
overdetermined by eqn (12.3).
The stress is
P = P(E) + A¢;(F), (124)

138 | LINEARIZED THEORY, THE SECOND VARIATION AND BIFURCATION
where P = W, is the constitutive contribution and A is a Lagrange multiplier. From

eqn (12.2),
P(F) = P(F,) + eM(F,)[Vu] + o(€) (12.5)
and
be(F) = be(Fo) + €dre(Fy)[Vu] + o(€). (126)

We assume that
AsAg ter + sen + o(€’), (12.7)
and conclude that

P =P, +€P’ + o(€), (12.8)
where
Py = P(Fy) + AoPe(Fo), (129)

and
P’ = H(Foj A) [Vu] + Abe (Fo), (12.10)
with
H(Foj Ag) = M(Fy) + AoGer(Fo) = (W + dob) erizg: (12.11)
Equilibrium without body force requires, of course, that DivP = 0. Dividing by € and
passing to the limit, we derive
DivP’ = 0, (12.12)
subject to
br (Fo) - Vu = 0. (12.13)
This is a linear system for the fields u(x) and A’(x). In a mixed position/traction boundary-
value problem, we assign p’ = P’N on 0x, and u on its complement.
Problems
1. Suppose ax, is loaded by a pressure of fixed intensity p. Show that
p’ = -p[(divu)I - (gradu)']F5N, (12.14)

PROBLEM | 139
where grad and div are the gradient and divergence operators based on pos-
ition y,.
2. Complete the differential equation -gradq’ = ... for the incremental constraint
pressure q’ in an incompressible material.
Consider the potential energy of the deformation x (x; €). We again confine attention
to the mixed dead-load problem for the sake of illustration. The potential energy is the
function of € defined by
Fle) = [wear | p- x(x €)da, (12.15)
where p is fixed, independent of €. We assume that x(x;¢) is likewise fixed on the

complement of dx,, and hence that u and v vanish there. Then,
F(e) = [™ . F(a [ P.N: x'(e)da. (1216)

K aKxp
Recalling that @(F(€)) =0, we have 0 = $'(e) = ¢; - F’(e). Evaluating at € = 0 then

yields
F(0) = / [We (Fs) + g¢e(Fo)]

« Vado — [ P.N - uda, (1217)
ax
for any scalar field g(x). Identifying this with the equilibrium Lagrange-multiplier field
Ao(x) and integrating by parts then furnishes
F(0) = [> - Vudy - [ rivceynrar, (12.18)
which reduces to
F'(0) = fo - DivPydv. (12.19)
Because P, is an equilibrium stress field, it nullifies the first variation F’ (0) and hence renders
the potential energy stationary (cf. Problem no. 3 in Chapter 3).
Problem
Clearly, the first variation vanishes at an equilibrium state for all variations u. In par-
ticular, the latter need not satisfy any equilibrium equations or boundary conditions
beyond u = 0 on the complement of d«,. Prove the converse, i.e, that if the first
variation vanishes for ail such variations, then the underlying state is in equilibrium.
140 | LINEARIZED THEORY, THE SECOND VARIATION AND BIFURCATION
Accordingly, the energy comparison reduces to
ELe(as €)] ~Elyg(2)] = Se°LF"(0) + o(€*)/e*) (120)

Dividing by e€? and passing to the limit, we conclude that y, is a stable deformation only if
the second variation F’(0) is non-negative; i.e.,
F’(0) > 0. (1221)
To make this explicit, we differentiate eqn (12.16), reaching
F(e) = [nF -F(e)+W,- FO} [ P.N - x" (€)da, (12.22)
and
F'(0) = [omentva - Vu + P(E,) « Vv}dv -| P,N - vda. (12.23)

ax
Using
/ 6(B,) - Vvdv = / {Div(B(F,)'v) ~ v - DivB(F,) }dy, (1224)

together with DivP(F,) = -Div{Ag@p(F,)} (from DivP, = 0), integrating the first term by
parts, and invoking v = 0 on the complement of d«,, we deduce that
F’(0) = | {M(F))[Vu] » Vu — Aobe(Eo) » Vv }du. (1228)

Finally, we use eqns (12.3), part 2, and (12.11) to reduce this to
F’(0) = [Hes do)) [Vu] - Vudv. (1226)
Using this expression it is possible to show that the Legendre-Hadamard inequality is a

pointwise necessary condition for eqn (12.21). A simple proof may be found in the book by
Fraejis de Veubeke (1979).
Problem
An elastic body is in frictionless contact with a rigid body on a part of its boundary.
Give a direct proof (not relying on the 2nd variation) that a deformation minim-
izes the energy only if it exerts a pointwise compressive pressure distribution on
the rigid body. Hint: To ensure that the elastic body and the rigid body do not
inter-penetrate, variations u should be such that u - n > 0 on the contacting surface,
where n is the exterior unit normal field to the boundary of the rigid body.
REFERENCES | 141
Consider the linearized equilibrium problem with null incremental data, i.e,
DivP’=0 and (Fy): Vu=0 in x, P'N=0 on 9Ok,,

u=0 on dK\dk,. (12.27)
Clearly, this admits u = Oand A’ = 0asa solution no matter the values of the underlying de-
formation y, and Lagrange multiplier Ag. A bifurcation is a non-trivial solution {u, A’} to the
same problem. Its existence or otherwise depends on the underlying state. It corresponds
to non-uniqueness of equilibrium in the linear approximation. For any bifurcation we have
O= fo - DivP'do = fours —P’- Vu)dv
= [ P’N - uda- fr - Vud. (12.28)

ax K
Accordingly,
[Hes Ay)) [Vu] - Vudv = 0, (12.29)
and so bifurcations nullify the second variation of the energy. Taken together with
eqn (12.21), this motivates the Euler—Hill_Trefftz criterion. Given {y,, Ao}, if there is a non-
zero {u, A’} that furnishes a minimum value, namely zero, to the second variation of the
energy, then the underlying state {y,, Ao} is potentially unstable. This may be cast as a
variational problem subject only to eqn (12.27), part 2, and the requirement that u = 0
on d«\dk,. Equations (12.27), parts 1 and 3, emerge as the Euler equation and natural
boundary condition in this approach. This problem is of course linear, and thus far more
tractable than the actual (nonlinear) problem. Ogden (1997) discusses a number of explicit
applications of this criterion.
Note that the case of several simultaneous constraints is covered, rather obviously, by
using eqn (6.10) in place of eqn (12.4) and repeating the argument leading to eqn (12.26),
for all constraints acting simultaneously. The unconstrained case is recovered by suppress-
ing eqn (12.13) and ignoring the Lagrange multipliers.
REFERENCES
Ogden, R-W. (1997). Non-linear Elastic Deformations. Dover, NY.

Fraejis de Veubeke, B.M. (1979).A Course in Elasticity. Springer, NY.
FURTHER READING
Hill, R. (1957). On uniqueness and stability in the theory of finite elastic strain. J. Mech. Phys. Solids
§, 229-241,
13
Elements of plasticity theory
Plasticity theory furnishes a foremost example of the utility of the concept of Elasticity in the
formulation of more general models of material behavior. Roughly, plasticity theory seeks
to describe the response of materials in which a strain persists after removal of load. This
can occur when the load has reached a certain threshold. Various materials, such as metals,
snow, plasticine, polymers, and paint come to mind. Existing theory pertains mainly to
metals, for which the underlying mechanisms are reasonably well understood. If the metal
is crystalline, with rows of lattice cells stacked one upon the other, and if a shear stress is
applied in the axes of the lattice, then one typically observes a shear strain on these axes
developing in response to the stress, If the shear stress meets or exceeds a critical value,
then relative slipping of the stack ensues, producing a permanent macroscopic shear de-
formation. This is essentially a frictional effect, and hence invariably dissipative in nature,
in contrast to pure elasticity. To describe it a suitable notion of energy dissipation will prove
necessary.
The picture is similar in the case of simple tension of a bar (Figure 13.1). If we plot the
mean cross-sectional axial stress (the axial force divided by the current cross-sectional area)
against the current length of the bar, we typically see a response like that depicted in the fig-
ure. Upon initial loading, the length of the bar increases roughly in proportion to the stress.
Their ratio is denoted by E. Further load or extension results in the onset of a nonlinear
response, with the slope changing sharply and dropping significantly below E. If the load
is reduced after the onset of this nonlinear regime, then the resulting deformation is quite
different from that achieved by initial loading to the same stress level; the unloading typ-
ically is again linear, but somewhat offset relative to the initial loading curve. Reasoning as
in Chapter 1, we are justified in attributing these observations to the material per se if the
deformation and stress fields are uniform. In this case, the constant E is a material property,
the famous Young’s modulus. The mean stress is then equal to the local stress, and its value
at the upper end of the linear part of the loading regime is the axial yield stress. All the while,
the bar may shear or twist, while being extended or compressed, but here we focus attention
on the relationship between uniaxial stress and length, /. As in the case of pure elasticity, the
latter is normalized by initial length, /), yielding the usual stretch A(= 1/1).
Finite Elasticity Theory, DavidJ. Steigmann.

ELASTIC AND PLASTIC DEFORMATIONS | 143
Figure 13.1 Idealized uniaxial response of an elastic-plastic material
13.1 Elastic and plastic deformations

If the load is removed after the onset of yield, then the length of the bar will recover, linearly,
to a value |, as shown in the figure, in which the subscript is used to identify an intermediate
configuration. This generates a so-called permanent stretch, or plastic stretch, given by A, =
[,/ly, corresponding to a vanishing axial stress. The stretch A just prior to the unloading is
then given by
A= Ay (13.1)
where A, = I/l;. This is called the elastic stretch, because it is that part of the total stretch
required to restore the bar to length / under the application of the stress existing prior to
the unloading. Thus, A, = 1 and A = A, when the stress vanishes. One thing worth noting is
that the elasticity of the material—here characterized by the modulus E—is insensitive to
the plastic deformation. Indeed, this conclusion extends to various other aspects of elastic
response, as observed in famous experiments conducted by G.L Taylor. We shall elaborate
in due course. Attention is confined to the rate-independent theory, in which the response,
as depicted in Figure 13.1, is either insensitive to the rate of deformation or the deformation
proceeds so slowly that any rate dependence is not relevant.
One slightly unsatisfying aspect of this picture is that it mixes notions of stress and
deformation together in describing the different types of stretch. In modern continuum
mechanics we are accustomed to separating these notions for as long as possible so as to
better understand the distinctions between kinematics and kinetics, deferring their inter-
mingling to a separate class of constitutive relations, of which elasticity is, of course, a primary
example. This issue has in fact been the source of much confusion over the course of the his-
torical development of the subject of plasticity theory. Nevertheless, the different notions
of stretch embodied in eqn (13.1) furnish a useful description of the underlying phenom-
ena and, therefore, remain central to the subject. Our purpose in this chapter is to extend
these ideas to general deformations and states of stress. We aim for a formulation of this
important subject that is as clear and unambiguous as the modern theory of finite elasticity.
Indeed, the motivation for this chapter stems from the conviction that such a development
remains elusive to the present day.
144 | ELEMENTS OF PLASTICITY THEORY
Figure 13.2 Elastic and plastic deformations
To begin, we introduce an intermediate configuration, «;, in which the material is pre-

sumed to be free of stress (see Figure 13.2). Let x, be a reference configuration, selected for
convenience as per usual practice, and let x, be the configuration at the present time, t. The
deformation from the reference configuration to the current has gradient F at time t and
material point x, as usual. Let H stand for the corresponding variable based on the use of K;
as reference, and let K be that obtained when «; is used as reference and x, is replaced by x,.
Then,
H = FK, (13.2)
in which F = Vx and x (x, ¢) is the usual deformation. Because H is the value ofF in the
absence of plastic deformation (K = I), we assume that J, > 0 and hence conclude that
Jx > 0. Throughout this chapter we use the notation J, to denote the determinant of a
generic tensor A. Comparing with eqn (13.1), we see that H corresponds to A, and K to A.
In much of the literature eqn (13.2) is written as F = F,F,, in which F,(= H) and F,(= K"')
respectively are the elastic and plastic parts of the deformation. Here we use
G=KkK' (133)
to denote the latter. It is important to note that while the factors in eqn (13.1) may be
interchanged without loss of generality, this is not the case in eqn (13.2) for the simple
reason that tensor multiplication does not commute.
In the course of extrapolating eqns (13.1) to (13.2) we have, of course assumed that x;
is stress free. In the one-dimensional situation, the associated length J, is achieved simply
by removing the load. This corresponds to the removal of the stress pointwise in the case
of uniform stress. However, pointwise removal of the stress is generally not feasible in the
three-dimensional context. That is, it is not generally possible to have T(x, t) = 0 for all x
in «,. In reality, there is a distribution of residual stress due to the presence of various defects
in the body. These induce local lattice distortions in the case of crystalline metals, for ex-
ample, which in turn manifest themselves as elastic strain and a consequent distribution of
stress. 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.
ELASTIC AND PLASTIC DEFORMATIONS | 148
Nevertheless, it is possible, in principle, to remove the mean stress via an equilib-

rium unloading process. In particular, in equilibrium the mean stress, T, is given by (see
Chadwick, 1976)
- 1 1
vol(x,)T = sf (t@ysy@ddas = [ p(b@®yt+ty@b)dy, (13.4)
any Kt
where t and b, respectively, are the boundary traction and body force. Accordingly, T =Oif
the entire body is unloaded.
In view of the Mean-Value Theorem for continuous functions (see Fleming, 1977), there
exists y € , such that T(y, f) = T. Therefore, T(y,t) = 0 for some y € x, if the body is
unloaded and in equilibrium. Let
d(k,) = sup ly -z]. (13.5)

yiZEXt
This is the diameter of x, Then, for every y € K, we have T(y,t) — T(y,#) as d(x,) > 0.
Accordingly, the local value of the stress can be made arbitrarily small as the diameter of the
body is made to shrink to zero.
Of course, it is not possible to reduce the diameter ofa given body to zero. However,
we may regard any body as the union of an arbitrary number of arbitrarily small disjoint
sub-bodies Pp, ie, Ky = Ue Pm, with d(P™) —> 0. Imagine separating these sub-bodies
and unloading them individually. We then have T(y, t) > 0 for every y € P“, for every n.
Because everyy in k, belongs to some P™, this process results in a state in which the material
is pointwise unstressed. Of course, each piece P has in general experienced some (elastic)
distortion in this process, and so the unstressed sub-bodies cannot be made congruent to
fit together into a connected region of 3-space. Thus, there is no global stress-free configur-
ation of the body and, hence, no position field x;, say, such that dy, = Gdx (or H''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 field. Accordingly, unlike F, neither G nor H has the prop-
erty of being a gradient. It follows that for any closed curve C x,, with image y = x (I’,¢)
in x, under the deformation map, the vector
B= [sax= [ wiay (136)

r Y
does not vanish. This is called the Burgers vector associated with the specified curve, induced
by the plastic deformation.
In view of the foregoing, we regard x; as being associated with a material point x, rather
than as a configuration per se. It has the properties ofa vector space. In fact it may be re-
garded 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 (see the paper by Noll, 1967, and the book
by Epstein and Elzanowski, 2007), which, however, is not emphasized here as it is largely
superfluous as far as the formulation of problems is concerned.
13.2 Constitutive response
We have mentioned that the elastic properties of the material are essentially independent
of the plastic deformation. This idea is codified in the expression
U(P,) = / W(H)dv, (137)

where wp is the strain energy per unit volume of «,. This is determined entirely by the elastic
deformation, in accordance with our hypothesis. In particular, this function is entirely un-
affected by the relative slipping between adjacent lattice planes in a crystalline metal due to
plastic deformation. This reflects the observation (see Batchelor, 1958) that relative slip-
ping achieved without stress does not alter the structure of the lattice cells and so leaves the
elastic constitutive response unchanged. In reality, slip is accomplished in steps, much as
the overall displacement of a carpet is achieved by bunching it up locally and displacing the
resulting bulges from one end of the carpet to the other. These stepwise displacements are
called dislocations. They invariably generate localized lattice distortions and hence stresses.
These are relieved, in principle, but cutting the body into small pieces, yielding the disjoint
local intermediate configurations that we have identified with «,; accordingly, the latter are
not realized in practice, but rather serve as a conceptual aid.
Here and henceforth, we shall confine attention to uniform materials for which the func-
tion ¥ is not explicitly dependent on the material point; the strain-energy density is then
given by one and the same function at all points of the body. This is the notion underlying
Noll’s theory of materially uniform simple bodies, which has had the most profound influence
on the development of modern plasticity.
It proves convenient to base the theory on the strain energy “per unit volume” of x;. This
may be defined unambiguously despite the fact that there is no global intermediate config-
uration and hence no associated volume per se; we simply multiply w by the local volume
ratio induced by the deformation from «; to «,. This ratio is of course just J}, which is well
defined. The desired function is
W(H) = Juw(H). (138)
The use of this function, rather than w, affords a simple extension of what we have already
learned in the case of pure elasticity. This follows from the fact that in the absence of plas-
tic deformation—a situation we intend to encompass in the theory to be developed—H
reduces to F and the energy W is then just the conventional strain energy per unit refer-
ence volume. In this specialization we have seen that the symmetry of the Cauchy stress—a
restriction we impose a priori—is equivalent to the invariance of the energy under super-
posed rotations, ie., W(F) = W(QEF) forall rotations Q, Because W is independent ofK
by construction, it follows that
W(H) = W(QH) (139)

CONSTITUTIVE RESPONSE | 147
for all rotations Q, This carries the same implications as in the case of pure elasticity,
namely,
W(H) = W(Cy), (13.10)
where Cy = H'H is the right Cauchy-Green tensor derived from H. Equivalently, we

may use
W(H) = W(Ex), (13.11)
where Ey = >(Cy ~ I) is the elastic Lagrange strain and W(Ey) = W(2E, +1). As before,
the Piola stress based on x; as reference is Wy, and this is related to the 2nd Piola—Kirchhoff
stress §,—also based on k,—by Wy, = HS,, with
S, = Wz, (13.12)
The usual Cauchy stress T is given by
Wy = TH’. (13.13)
Normally metals can undergo only small elastic strains before yielding, at least if the rate
of strain is sufficiently small. We simplify the model accordingly by supposing that |E,| is
always small enough that the use of the quadratic-order approximation
(Ey) = WO) + Ey « Wey (0) + By CLBq] + o(Eaf) (1314)

is justified, where
C = Weys,,(0) (13.18)
is the 4th-order tensor of elastic moduli, evaluated at zero strain. This possesses the ma-
jor and minor symmetries discussed in Chapter 11. Because x; is associated with vanishing
stress by assumption, the coefficient We,, (0) of the linear part of the expansion eqn (13.14)
vanishes. Accordingly, the leading-order strain energy is purely quadratic:
W(Eu) ~ sn - C(Ex]. (13.16)
This, of course, is just the usual elastic energy for small strains, yielding
S, = C[Ey]. (13.17)
Relying on an observation made in Chapter 11, we take C to be positive definite (A-

C[A] > 0 for all A with non-zero symmetric part), and conclude that W is a convex
function, i.e.,
W(E,) - W(E,) > W,(E,) - (E, - E:) (13.18)
for all E, # E,, in which the subscript ;, has been suppressed for the sake of clarity.
Because ; is free from elastic distortion, in the case of a crystalline metal the lattice is
perfect and undistorted in x,. This has the consequence that
W(H) = W(HR), (13.19)
for all rotations characterizing the symmetry of the lattice and, hence, that
W(E,) = W(R‘E,R). (1320)
We have seen that the collection of all such rotations is a group, the symmetry group of the
lattice. For crystalline solids this group is always discrete, whereas for isotropic or trans-
versely isotropic solids it is connected. In particular, isotropic materials satisfy eqn (13.20)
for all rotations.
In the purely quadratic case, this has the well-known consequence that W(En) is of the
form
~ 1
W(Ex) = 5M HEn)’ + LE, + Eq, (1321)
in which A and ware the classical Lamé moduli. These are subject to the restrictions 4 > 0
and 3A + 242 > 0, which are necessary and sufficient for the positive definiteness ofC in the
present context. Using eqn (13.12), this in turn generates the classical stress-strain relation
S, = Atr(Ey)I + 2uEy (13.22)
for isotropic materials.
13.3 Energy and dissipation

It is convenient to adopt a referential description of the strain energy. Proceeding from
eqns (13.7) and (13.8) we have
ur) = f wonar= f jevar= f sagen (13.23)
Thus,
uP.) = [ WOR, Kar (1304)

Py
PROBLEM | 149
where, from eqn (13.2),
W(F, K) = J; W(FK). (13.25)
The total mechanical energy in P, C x, is then given by
1
E(P,) = / Ddy, where ®=W+ 5h lvl, (13.26)
Pr
where , is the mass density in x,.
Problem
For fixed K, prove that W is strongly elliptic at F if and only if W is strongly elliptic
at H.
The power of the forces acting on P, is
P(P,) = / p-ydat+ / p,b + ydv. (13.27)

oPr Py
Using the equation of motion eqn (2.28) we derive
. 1 .n\ fate .
Peb- y= 5h ly| } - [Div(P'y) - P- Vy]. (13.28)
Substituting into eqn (13.27), applying the divergence theorem and using p - y = P'y - N,
where N is the exterior unit normal to OP,, we arrive at the Mechanical Energy Balance (cf.
eqn (3.1))
P(P,) = “KP, +S(P,), (13.29)
where
K(P,) = of De ly|’ dv (13.30)
is the kinetic energy and
S(P,) = [ P-Fdv (1331)

Pr
is the stress power.
Next, we define the dissipation D to be the difference between the power supplied and
the rate of change of the total energy; thus,
D(P,) = P(P,) - SEP). (1330)

Using eqn (13.26) in the form €& = K’ +’ and combining with eqn (13.32), it follows
immediately that
D(P,) = / Dd», (13.33)

Pr
where
D=P.F-W, (1334)
In the purely elastic context we see from eqn (3.9) that D vanishes identically. Here,
we impose the requirement D(P,) > 0 for all P, C «, and conclude, from the localization
theorem, that
D>0 (13.38)
pointwise. This assumption serves as a surrogate for the 2nd law of thermodynamics in the
present, purely mechanical, setting.
Problem
Suppose the state {x,.(x), K,,(x)} is asymptotically stable relative to the static state
{Xo(x), Ko(x)} in the sense that any dynamical trajectory {x (x, t), K(x, t)} initiating
at the latter tends to the former, pointwise, as f > 00. Show, for conservative prob-
lems, that the potential energy of the asymptotically stable state is no larger than that
of the initial state.
To obtain a useful expression for the dissipation we proceed from eqn (13.25), obtaining
W = Je (W - (ie/Ix) WI. (1336)

Here, we use the identity Jx/Jx =K'-K together with
W = Wy: H= W,K'- F+FW,-K. (1337)
Recalling that Wy, = T(FK)* = PK* and hence that WyK' = J,P and F'Wy = J,xF'PK",
eqn (13.36) is reduced to
W=P.F-E-RK"', (13.38)
where
E=WI-F'P (13.39)
is Eshelby’s Energy-Momentum Tensor. Accordingly, the local dissipation may be written in

the form
PROBLEMS | 151
D=E- KK". (13.40)
This result, due in the present context to Epstein and Maugin (1995), highlights the role of
the Eshelby tensor as the driving force for dissipation. We use it here, in conjunction with
eqn (13.35), to derive restrictions on constitutive equations for the plastic evolution K. We
note in passing, relying on eqn (13.38) and the chain rule, that
P=W,(F,K) and E=-,(F,K)K’. (13.41)
The expression eqn (13.40) for D makes clear the fact that the dissipation vanishes in
the absence of plastic evolution, i.e, D = 0 ifK = 0. On the basis of empirical observation,
we introduce the hypothesis that plastic evolution is inherently dissipative; thus, we sup-
pose that D # 0 if, and only if, K # 0. In view of our previous assumption eqn (13.35), this
means that
K 0 ifandonlyif D> 0. (13.42)
It may be observed from the definition eqn (13.39) that the Eshelby tensor is purely
referential in nature, mapping the translation space of x, to itself. For reasons that will be
explained later, it proves convenient to introduce a version of the Eshelby tensor, E,, that
maps K; to itself. This is defined by the relation
E=)/K‘EK'. (13.43)
Problems
1. Use eqn (13.39) to show that if E’ is the Eshelby tensor derived by taking the
current configuration as reference; i.e, E’ = wl — T, then E = J-F'E’F". Thus, E
is the pullback of EF’ from x, to x,. Show that E is the pullback of E; from x; to x,,
and that E; is the pullback of E’ from x, to K;.
2. Prove that
E, = WI- H'Wy. (13.44)
This implies that E, is determined entirely by H and, hence, purely elastic in ori-
gin. Show that E, = WI - C,,S, and, hence, that E, is insensitive to superposed
rigid-body motions.
3. Prove, in the case of small elastic strain, that
E, = -S, + o(|Eu|), (13.45)

where §; is given by eqn (13.17) and, hence, that the Eshelby tensor based on the
intermediate configuration is given, to leading order and apart from sign, by the
2nd Piola—Kirchhoff stress referred to the same configuration.
4. Prove that
J.D =E,.- K'K (13.46)
and hence that the assumption of inherent dissipativity is equivalent to the

statement:
K #0 ifandonlyif E,-K’K> 0. (13.47)
It is interesting to observe that if E, = 0, then W = 0, S, = 0 and, hence, D = 0; then,

eqn (13.47) implies that there can be no plastic evolution. That is, without stress, there
can be no change in the plastic deformation. This is in accord with common observation.
13.4 Invariance
We have observed that the symmetry of the Cauchy stress is equivalent to the statement
W(H) = W(QH) for all rotations Q, Because the argument leading to this conclusion
is purely local, the rotation Q can conceivably vary from one material point to another.
This stands in contrast to the rotation Q(f) associated with a superposed rigid-body mo-
tion, which must be spatially uniform and, hence, the same at all material points; here, we
distinguish these cases explicitly in the notation.
Ina superposed rigid-body motion, the deformation x (x, f) is changed to
x* (ut) = Q(t)x (x £) + c(t) (13.48)
for some spatially uniform vector function c. It follows immediately—as we have seen—
that F(= Vx) goes into F*(= Vx*), with F* = QF. The argument cannot be adapted to
H, however, because it is not the gradient of any position field.
Nevertheless, it follows that H* = QFK, whereas H* = F*K* = QFK’. Consequently,
QFK = QFK’. (13.49)
We would like to use this to arrive at some conclusion about the relationship between K*
and K, but this requires a further hypothesis. A natural one is that the dissipation is in-
sensitive to superposed rotations. To explore the implications we define Z = K'K'' and
note, from (13.49), that Jz = 1. Suppose Z(t.) =I, so that the superposed rigid motion
commences at time fy. Using eqn (13.46) we find that the dissipation transforms to
JxD* = Et - (K*)'K" = E, - (K'Z"'ZK + K'R) = J,D+E,-K'Z"'ZK, (13.50)

YIELDING, THE WORK INEQUALITY AND PLASTIC FLOW | 153
wherein we have invoked the invariance of the Eshelby tensor E,. Accordingly, if D* = Das
assumed, then
E, -K'Z'ZK = 0, (1351)
and this purports to hold for K with J, > 0. It, therefore, holds for K = I, yielding Z = K*.
This amounts to selecting x; as the reference configuration for the superposed rigid motion,
this entailing no loss of generality as the argument is purely local. This Z is a bona fide plastic
flow, and therefore subject to our strong dissipation hypothesis (13.47). This requires that
7, vanish, and hence, given the initial condition, that K* = K; thus, G* = G. From (13.49)
it then follows that Q = Q(t). Altogether, then,
F*=QF, H’=QH and K'=K. (13.82)
13.5 Yielding, the work inequality and plastic flow

The situation depicted in Figure 13.1 suggests that the onset of yield may be characterized
by the statement |T| = Ty, where Ty is the yield stress in uniaxial tension, a material prop-
erty that may evolve with continued plastic flow. The inequality |T| < Ty is associated with
elastic response, whereas |T| > Ty is impossible. Because of the one-one relation between
stress and elastic stretch existing under our hypotheses, we could equally well describe yield
using a relation of the form f(A,) = 0.
In the three-dimensional setting, we assume that the onset of yield may be described us-
ing a relation of the form G(E,) = 0, where G is an appropriate yield function pertaining to
the material at hand. 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 rigid motions.
Thus yield occurs when the elastic distortion lies on a certain manifold in 6-dimensional
space. Again, because of the one-to-one relation between E,, and S, under our hypoth-
eses, we could equally well characterize yield in terms of the statement F(S;) = 0, where
F(S,) = G(C"'[S, ]) is the yield function, expressed in terms of the stress. We suppose elas-
tic response to be operative when the stress satisfies F(S,) < 0, in which case the stress is
said to belong to the elastic range, and that no state of stress existing in the material can
be such that F(S,) > 0. We further suppose that F(0) < 0 and, hence, that the stress-free
state belongs to the elastic range. In this way we partition 6-dimensional stress space into
the regions defined by positive, negative, and null values of F, with the first of these being in-
accessible 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, causing nonuniform distributions of stress
and elastic strain in the material, 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.
F>0
Figure 13.3 A cyclic process
Consider now a cyclic process, as described in Section 3.2 of Chapter 3. Reasoning as we

did there, we have
2
/ P - Fdt > 0, (13.53)
4
where f,,, respectively, are the times when the cycle begins and ends. Suppose these times
are such that the associated stresses satisfy F < 0; the cycle begins and ends in the elastic
range (Figure 13.3).
Suppose the cycle is such that there exists a sub-interval of time [t,, t,] C [t1,t] dur-
ing which F = 0, and that F < 0 outside this sub-interval. Then, we may have plastic flow,
ie. K ¥ 0, during this sub-interval, while K = 0 outside it, implying that K(t,) = K(t,) and
K(t,) = K(¢,). Substituting eqn (13.34) and noting that the process is cyclic in the sense
that F(t,) = F(t,), we arrive at the statement
YERCE) KC) = WORG). KC) + ff ” Dat > 0. (1354)

Equivalently,
| ” Duq(FC4,), K(t)) R(t) + D(A)]de > 0. (1355)

Dividing by #, — t,(> 0) and passing to the limit, it follows from the mean value theorem
that
Wy(F(t,), K(t,)) K(f.) + D(ta) > 0, (1356)

PROBLEM | 155
which may be written, using eqns (13.40) and (13.41), part 2, as
(E(F(t,), K(t,)) - E(F(t,),K(t,))] - K(t,)K(t,)7 > 0. (1357)
Problem
Prove that this is equivalent to the inequality
[E,(Bu(t.)) - E,(En(t))] - K(t.)"K(f.) > 0, (1358)

where E,(E,,) is the function of elastic strain obtained by recasting eqn (13.44). This
means that the dissipation is maximized by states E,, (equivalently, S,) that lie on the
yield surface.
In the case of small elastic strain, we substitute eqn (13.45) together with K'K = -GG",
which follows by differentiating GK = I, divide by |Ey|, and pass to the limit in eqn (13.58)
to derive the restriction
[S;(t,) - S,(t,)] - SymG(t,)G(t,)"' > 0, (13.59)

in which we have inserted the qualifier Sym to reflect the fact that the term in the square
brackets is symmetric; the inner product then picks up only the symmetric part of GG".
We summarize this result in the statement:
(S-—S*)-SymGG"' > 0; F(S*) <0, F(S) =0, (13.60)
where the subscript ; has been suppressed to promote clarity.

This inequality has a simple geometric interpretation having important implications for
the structure of constitutive equations specifying the evolution of G (Figure 13.4). First,
SymGG" must be perpendicular to the tangent plane T; to the yield surface at S. Secondly,
the entire elastic range, defined by F < 0, must lie to one side of T; at S. Thus, the elastic
range is a convex set; that is, if S,, belong to the elastic range, then so does every point
u € [0, 1] on the straight line S(u) = uS, + (1 - u)S,.
If F(S) is a differentiable function, then T; depends continuously on S and the sur-
face F(S) = 0 has exterior normal in the direction of the derivative F, at point S. We
conclude that
SymGG" = AFs, (13.61)
for some scalar Lagrange multiplier field A(x, t) > 0, to be determined from the particular
initial-boundary-value problem at hand. Precisely the same result is implied by the Kuhn-
Tucker necessary conditions of optimization theory (see Zangwill, 1969). In the present
setting, this pertains to the optimization problem:
max($-GG"!) subjectto S=S' and F(S) <0. (13.62)

Sym GG"!
Sym GG"!
Ts)
(a) (b)
Figure 13.4 (a) Sym GG"! is perpendicular to Tp at S. (b) The elastic range, F < 0, lies to one
side of Tp at S
Plastic evolution is, therefore, such as to satisfy
GG" =AB, +2 (13.63)
for some skew tensor §2(x, t), called the plastic spin.
The foregoing considerations about yield and flow are quite general, and apply to both
crystalline and non-crystalline materials. Modern theory for crystalline media is still in a
state of active development (see Gurtin, Fried and Anand, 2010), particularly with re-
spect to issues such as strain hardening—the evolution of the yield function with plastic
flow—and plastic spin. In contrast, the classical theory, which purports to apply to isotropic
materials, is well established and much simpler. However, although the associated literature
is vast, it is seriously marred by the lack of any clear exposition of the explicit role played by
(isotropic) material symmetry in the logical development of the subject. One of our main
objectives here is to provide this missing link and, thus, to firmly establish the classical the-
ory on the basis of the modern theory for finite elastic-plastic deformations. For all these
reasons, attention is hereafter confined to the case of isotropy.
13.6 Isotropy
As we have seen, for isotropy the constitutive functions—exemplified above by the strain-
energy function—must be insensitive to the replacement of H by HQ, where Q is any
rotation. We have seen that such invariance implies, in particular, that
W(E,) = W(Ey), where Ey = Q'E,Q (13.64)
is the rotated strain. In general, this yields
S = Q'SQ, (13.65)
PROBLEM | 157
where § = Win» which can easily be confirmed in the special case eqn (13.22) on replacing
Ey by Ey.
Problem
Prove that this holds for isotropy in general.
With this result in hand, we are justified in requiring that the yield function, being a
reflection of material properties, should satisfy the material symmetry restriction
F(S) = F(S) (13.66)
with S given by eqn (13.65), for any rotation Q, Accordingly, as in the preceding Problem,
Fy = Q'F,Q, (13.67)
Because of eqn (13.2), invariance statements of this kind are equivalent to the statement
that scalar-valued constitutive functions should remain invariant if K is replaced by K =
KQ- equivalently, if G is replaced by G = Q'G, with F remaining fixed. To see how this
replacement affects plastic flow, we compute
(G)G" = QA'GG'Q+QQ, (13.68)
where we have allowed for the possibility that Q may be time-dependent. Substituting
eqns (13.63) and (13.67) we conclude that
(G)G" = AF + Q(2Q + QQIQ (13.69)

Now, for any skew @ we can always find a rotation Q(t) to nullify the parenthetical term
in eqn (13.69). To see this, suppose B(t) satisfies the initial-value problem
B=WB with B(0) = By, (13.70)
where W is skew and B, is a rotation, Let Z = BB’. Then,
Z=WZ-ZW, with Z(0)=1 (1371)
Clearly, a solution is furnished by Z(t) =I. A theorem on ordinary differential equa-

tions ensures that this is the only solution and, therefore, that B is necessarily orthogonal.
Furthermore,
J, = BY -B = J,tr(BB"), (13.72)
and this vanishes because W is skew. Accordingly, Jaa) = Jao) = 1, and B is a rotation.
Because the rotation in eqn (13.69) is arbitrary, we are free to pick Q = B (with W = Q,
of course), to conclude that
(G)G' =AF, with AH=A. (13.73)
Thus, by exploiting the degree of freedom afforded by the material symmetry group in the
case of isotropy, we can effectively suppress plastic spin in the flow rule and use
GG" = AE. (13.74)

This is a major simplification that is not available in the case of crystalline materials.
Problem
Why not?
Before proceeding we pause to take note of an important empirical fact that applies with
a high degree of accuracy to metals; namely, that yield is almost entirely insensitive to
pressure. This is true in essentially all metals for pressures over a very large range that en-
compasses most applications. Thus, yield is insensitive to the value of trT, where T is the
Cauchy stress.
Problem
Show that in the case of small elastic strain, trT = érS, + o(|Eu|).
Thus, 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 trS;. It should, therefore, depend on
S, entirely through its deviatoric part, DevS,, Again, omitting the subscript, we write
F(S) = F(DevS). (13.75)
Problem
Show that DevS = Q'(DevS)Q and, hence, that
F(DevS) = F(Q'(DevS)Q). (13.76)
Recall that in the theory for small elastic strains, we agreed to expand the strain—energy
function up to quadratic order in the elastic strain. Moreover, the stress is approximated
by an invertible, linear function of elastic strain. Accordingly, the strain energy may be
regarded as a quadratic function of the stress S$. For consistency we also approximate the
PROBLEM | 159
yield function by a quadratic function of the same stress. Because DevS is a linear function
of S, this means that F should be approximated by a quadratic function. The most gen-
eral such function in the case of isotropy is a linear combination of tr(DevS), (trDevS)”
and tr(DevS)? = |DevS|’, of which the first two vanish identically. The most general yield
function of the required kind such that the yield surface F = 0 separates regions defined by
F < Qand F > Oin stress space is then of the form
~ 1
F(DevS) = 5 |DevS|’ - k. (13.77)
This is the famous yield function proposed by von Mises. The present derivation, based on
material symmetry arguments in respect ofan intermediate configuration, together with the
assumption of differentiability of the yield function, promotes understanding ofits position
in the overall theory.
Because the set of symmetric tensors can be regarded as the direct sum of the S-
dimensional linear space of deviatoric tensors and the one-dimensional space of spherical
tensors, it follows that the yield surface defined byF = Ois a cylinder in 6-dimensional stress
space of radius J 2k. Here, k is the yield stress in shear. That is, ifthe state of stress is a pure
shear of the form
S = S(i@j+j@i), (13.78)
with i andj orthonormal, then |DevS|” = 2S? and the onset of yield occurs when || = k.
Here, k may be a fixed constant, corresponding to perfect plasticity, or may depend on ap-
propriate variables that characterize the manner in which the state of the material evolves
with plastic flow. The latter case refers to so-called strain hardening, the understanding of
which is the central open problem of the phenomenological theory of plasticity.
The reader is likely aware that alternative yield functions, such as that associated with
the name Tresca, are frequently used in the theory of plasticity for isotropic materials. This
function, which we do not record here, is in fact nondifferentiable and, hence, inaccessible
by the present line of reasoning. However, experiments conducted by Taylor and Quinney
indicate that it is less accurate from the empirical point of view that the von Mises function,
despite the seeming generality gained by relaxing the assumption of differentiability (see
the paper by Taylor and Quinney in Taylor's Collected Works, 1958).
The theory is completed by substituting eqn (13.77) into eqn (13.74), to generate the
flow rule for the plastic deformation. To this end we use eqns (13.75) and (13.77) with the
chain rule, obtaining
F,-S=F=(F) = 5 (Devs - DevS)’

= DevS - (DevS) = DevS - DevS = DevS -S, (13.79)
and, hence, Fs = DevS. Finally, eqn (13.74) provides von Mises’ flow rule
GG" = ADevS,. (13.80)
This implies that J, is fixed, and hence that no volume change is induced by plastic flow.
13.7 Rigid-plastic materials

The elastic strain is invariably small in metals under low strain-rate conditions because it
is 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 restric-
tion H'H = I. The elastic deformation is, therefore, a rotation field, which we denote by R.
Because the elastic strain vanishes identically, the strain energy is fixed in value and the
stress is arbitrary, i.e.,
O=W=S,-E,, with E,=0. (13.81)
Accordingly, at this level of the discussion S; is an arbitrary symmetric tensor, constitutively

unrelated to the deformation as in a rigid body, granted that it satisfies the yield criterion.
Furthermore, J,; = 1 and the relation between the Cauchy stress and S, reduces to
S, = R'TR, (1382)
and so
DevS, = Dev(R'TR) = R'(DevT)R. (13.83)
The yield function reduces to
F(DevS) = F(R'(DevT)R) = F(DevT), er)
the second equality being a consequence of isotropy, and is, therefore, expressible in terms
of the Cauchy stress alone, as in the more conventional expositions of the classical theory.
Using eqn (13.83), we may cast the flow rule eqn (13.80) in the form
R(GG")R' = ADevT. (13.85)
We can express this in a more convenient and conventional form by using the well-known
decomposition
L=D+W (13.86)
of the spatial velocity gradient L into the sum of the straining tensor D = SymL and the
vorticity tensorW = SkwL. Using L = FF together with eqn (13.2), we find in the present
specialization to H = R that
PLANE STRAIN OF RIGID-PERFECTLY PLASTIC MATERIALS: SLIP-LINE THEORY | 161
L = RR'+R(GG")R' (13.87)
in which the first term is skew while the second, according to eqn (13.85), is symmetric.
The uniqueness of the decomposition then yields D = R(GG"')R' and, hence, the classical
flow rule
D =ADevT, (13.88)
due to St. Venant. The Cauchy stress is
T = DevT - pl, (13.89)
where the pressure p is constitutively indeterminate. Equation (13.88) is the central

equation of the classical theory and predates the modern theory for finite elastic-plastic
deformations by at least a century. Its straightforward derivation via the modern theory,
relying on simple ideas about material symmetry, brings unity and perspective to this most
important branch of solid mechanics.
Problem
Show that the dissipation is D = 2A? and is, therefore, positive ifand only ifA > 0.
13.8 Plane strain of rigid-perfectly plastic materials:

slip-line theory
We consider deformations in the y,, y,—plane and, thus, confine attention to velocity fields
of the formv = v,.(y,, y2)e,- Then eqn (13.88) furnishes DevT = T = T,ge, ® e,, 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 thatp = -} Twa
13.8.1 State of stress, equilibrium

The yield criterion eqn (13.77), with eqn (13.84), reduces to
2k = Tap lap = (Tap + pdap) (Tap + pba)
= Tag Tag + 2pToa + p'Saa

1
= Tap Tap 7 2p’ = Tap Top 7 5 (Toa) (13,90)
or
(Ty - Tr)? + 4Ty, = 4k’. (13.91)

Problem
Show that the principal stresses are
T,=-ptk, T,=-p-k and T,=-p. (1392)
We conclude that the stress state is
T = -pl + k(u, Qu, -u, @u,), (13.93)
where {u;}, with u; = es, are the principal stress axes. Let t and s be orthonormal vector
fields, such that
u, = Z's
2 +t), Wea
V2
z's —t). (13.94)
Then,
T=-pl+k(t@st+s@t), (13.95)
which implies that k is the shear stress on the s, t axes.
Problem
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

requires that
grad(p/k) = (gradt)s + (divs)t + (grads)t + (divt)s. (1396)
We define a field 6 (x) such that
t=cos@e,+sinde,, s=-sin@e, + cos@e,. (13.97)
Then,
dt=sd@ and ds = -tdé, (13.98)
yielding
gradt=s@®grad0, grads = -t ® gradO (13.99)

PROBLEM | 163
and, therefore,
divt=s-grad0, divs = -t - grad@. (13.100)
Substituting into eqn (13.96), we derive
grad(p/2k) = (s- gradO)s - (t - grad )t, (13.101)
which is equivalent to the two equations
t- grad(p/2k+6@)=0 and s-grad(p/2k-6) =0, (13.102)
due to Prandtl and Hencky. These require that p/2k + @ take constant values on the trajec-
tories defined by dy,/dy, = tan 0, - cot @, respectively. The latter are the characteristic curves
of the hyperbolic system of PDEs for the fields p and 9. Remarkably, the stress is statically
determinate, i.e., granted suitable boundary conditions, it can be determined without know-
ledge of the deformation. These striking features of the theory of perfectly plastic materials
contrast sharply with the mathematical setting of the theory of elasticity.
13.8.2 Velocity field

It proves advantageous to decompose the velocity field in the (variable) basis {s, t}. Thus,
v=vttus. (13.103)
To compute the velocity gradient, we combine the chain rule with eqn (13.98) to obtain
dv = dv,t + v,sd@ + dv,s — v,td

= (gradv, - dy)t + v.s(grad@ - dy) + (gradu, - dy)s - v,t(grad0 - dy)
= Lady, (13.104)
and conclude that
L =t © grady, + v,s @ gradO + s @ gradv, — vt @ grado. (13.105)
Then,
2D = ¢ @ grady, + gradv, @ t +s ® gradv, + gradv,@s

v,(s ® gradO + grad@ @ s) — v,(t @ gradO + grad @ t). (13.106)
Using eqns (13.88) and (13.95) leads to
t-Dt=0 and s-Ds=0O, (13.107)

which together imply that the deformation is isochoric and the extension rates vanish along
the directions t ands.
Problem
Show that eqn (13.107) are equivalent to the pair
t+ (gradv, —v.grad@) =0 and s- (gradv, + v,grad@) = 0. (13.108)
These are the celebrated Geiringer equations. They are linear PDEs for the components
v, and v, if the stress field is known.
Suppose the normal velocity v, (resp., v,) is continuous across the trajectory with unit-
tangent field t (respectively, s). This means that fissures do not form in the material. Taking
jumps, the first (resp. second) equation implies thatt - grad[v,] (respectively, s - grad[v,])
vanishes on this trajectory, so that the slip [v,] (respectively, [v,]), if non-zero, is uniform
along it. Hence, the name slip-line fields given to this subject.
The literature on this topic is vast. The books by Hill (1950), Kachanov (1974), and
Johnson, Sowerby and Haddow (1970) and the article by Geiringer (1973) describe further
theory and many worked-out solutions. Numerical solutions are discussed in the article by
Collins (1982).
REFERENCES
Batchelor, G.K. (ed.) (1958). The Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 1: Mechanics of
Solids. Cambridge University Press, Cambridge.
Collins, IF. (1982). Boundary value problems in plane strain plasticity. In: H.G. Hopkins and
MJ. Sewell (Eds) Mechanics of Solids: The Rodney Hill 60th Anniversary Volume, pp. 135-184.
Pergamon Press, Oxford.
Epstein, M. and Elzanowski, M. (2007). Material Inhomogeneities and Their Evolution. Springer,
Berlin.
Epstein, M. and Maugin, G.A. (1995). On the geometrical material structure of anelasticity. Acta
Mechanica 115, 119-131.
Fleming, W. (1977). Functions of Several Variables. Springer, Berlin.
Geiringer, H. (1973), Ideal plasticity. In: C. Truesdell (Ed.) Mechanics of Solids, Vol. II (C. Truesdell,
ed.), pp. 403-533. Springer, Berlin.
Gurtin, M.E., Fried, E. and Anand, L. (2010). The Mechanics and Thermodynamics of Continua.
Cambridge University Press, Cambridge.
Hill, R. (1950). The Mathematical Theory of Plasticity. Clarendon Press, Oxford.
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.
REFERENCES | 165
Noll, W. (1967). Materially uniform simple bodies with inhomogeneities. Arc. Ration. Mech. Anal. 27,
1-32.
Taylor, G.I. and Quinney, H. (1958). The plastic distortion of metals. In: G.K. Batchelor (Ed.) The
Scientific Papers of Sir Geoffrey Ingram Taylor, Vol. 1: Mechanics of Solids, pp. 252-290. Cambridge
University Press, Cambridge.
Zangwill, WI. (1969). Nonlinear Programming. Prentice-Hall, Englewood Cliffs, NJ.
SUPPLEMENTAL NOTES
1. The cofactor
Consider a vector-valued map that takes a x b into Aa x Ab. If we can construct a linear
map of this kind, then there is a tensor A* such that A*(a x b) = Aa x Ab. A‘ is called
the cofactor of A. This would qualify as a working definition of the cofactor, provided it
could be shown that such a linear map exists. An elegant proof is given in the appendix of
Chadwick (1976).
Our approach will be to simply assume linearity, construct a representation Ave, ® e; for
A*, and use it to confirm linearity after the fact. We have
Aa X Ab = ey AvAjn@ Dn €xs (1.1)
where e,, is the permutation symbol. Write A*(a x b) = A*(€,a:b,e) = Ai etm AD @-

Using the fact that {e,} is a basis, we find that the original equation is equivalent to
Ay, ei 1B = Cp Ap Andi Dy,. (1.2)
However, the a; and b,, are arbitrary real numbers, so this must be satisfied no matter how
we choose them. Pick b,, = 81s Sans 53, in succession, where 4,, is the Kronecker delta. We
get Area) = ey AyA;,a). Now pick a; = 5), 53), 53, in succession. This yields Aj gg = e,AigAj-
Note that the left-hand side is skew in the subscripts q, r. For our result to make sense, the
right-hand side had better be also (check: eA;Ajq = eyAjAiq = CyiAigAj = —ryAigAj,). Now
multiply through by e,,, and sum on q, r. Use the fact that €jq€q = 26, to get:
1
Ayip = 5 Hvar Aig. (1.3)
This formula was derived by making convenient choices of the vector components 4), b,,
ie. we have shown that it is a necessary condition for the definition to be true. To show that
it is also sufficient, we must substitute into the left-hand side of eqn (1.2) and show that we
get the right-hand side, for any a,b,,.
Problem
Do so.
We have constructed A* = Aje, @ e; such that A*(a x b) = Aa x Ab for all a,b and A*
does not depend on a,b. The function f(v) = A*v is linear and vector valued, and so the
cofactor is a tensor.
168 | SUPPLEMENTAL NOTES
REFERENCE
2. Gradients of scalar-valued functions of tensors
The gradient of a scalar-valued function of tensors is defined in exactly the same way as for
functions of vectors or positions. Let g(A) be such a function and suppose it is differentiable
at A,. This means that for each A, in an open set in Lin containing Aj, there is a linear
function f(B), depending parametrically on the fixed tensor Aj, such that
g(A.) = g(A,) + f(A,

— Aj) + 0([A,
— Aj]). (2.1)
Because f(B) is linear, it is expressible as the inner product ofa unique tensor with B; we
call this tensor Vg(A,). Often the notation g, (A,) is used to make the independent variable
explicit. Thus,
g(A,) = g(A) + Ve(A,) : (A, - Aj) + o(|A, - A,|). (2.2)
Using a mixed basis for illustrative purposes, let A= Aje;@E 3. Then Aj, =
Ao e, @ Ep. Let g(A,c) = g(Age; ® Eg); then, eqn (2.2) may be written
(AD) = (A?) + (AR - AP)e: @ Ey» Vg(Ai) + 0(|A2 - Ail). (23)

This must hold for all A,,. Imposing it for A, - A, = Ae, @ E,, for example (A? - A =
A6b,,55)), yields
B(Aj? + Ad: dea) - B(Ai?) = Aer @ E, - Vg(A) + 0A). (24)

Dividing by A and passing to the limit, we get
93
e,- [Vg(A.) JE, = ~ lay (2.5)
2
wherein we hold fixed all components other than Ay. In general we then have
95
Va(A) = Se, @ Es
iB
(26)
provided that all the derivatives are independent. This would not be the case if there were
any a priori relation among the components, as is the case for symmetric or skew tensors,
for example.
SUPPLEMENTAL NOTES | 169
3. Chain rule
Consider a curve in Lin described by a differentiable function A(t) where ¢ is a parameter in

some open interval (a, b). Let g(t) = g(A(t)). Suppose thatg is differentiable with respect
to A and that A is differentiable with respect to ¢. Furthermore, let A, = A(t,,). Then g(t)
is differentiable and, from eqn (2.2) above,
a(b) = (tf) + Vg(A,) - (A, — Ai) + o([A, - A,|). G1)
Wealso have
A, - A, = (t,-,)A(h) + 0(f - 4), (32)
and, therefore,
|A. - A,| = O(#, — ty). (33)
Thus,
a(t) - (t) = (fh -))Vg(A,) - A(t.) + 0(f, - 4). (34)
Dividing by £, — t, and passing to the limit, we obtain the chain rule:
g = Vg(A)-A (35)
In the text we use the notation
dg = Vg(A) - dA. (3.6)
4. Gradients of the principal invariants of a symmetric tensor

We need formulas for the gradients of the invariants I,(A) with respect to A. We assume
the tensor A to be symmetric so that its off-diagonal components are not independent.
This means that a formula like eqn (2.6) above is not applicable; we resort to an alternative
method based on the chain rule.
Let A(t) describe a curve in Sym, and consider
(A) = trA=1-A. (4.1)
Then,
VI,(A)-A=1,=1-A (42)
The symmetry of A(t) implies that Ais symmetric too (the proofis immediate). If we de-
compose the tensor VJ,(A) into the sum of symmetric and skew parts, and then form the
inner product with A, we find that
VI,(A) -A = (SymVI,(A)) - A, (43)
where
2SymT = T+T' (44)
for any tensor T. Then eqn (4.2) yields
[(SymV1,(A)) -I]-A=0 (45)
for all symmetric A. Now the term in brackets is a symmetric tensor, and the condition
says that it is orthogonal to every element in the set of symmetric tensors. That this set is a
linear space follows from the fact that an arbitrary linear combination of symmetric tensors
is symmetric and the set also contains the zero tensor. Therefore, the term in brackets must
be the zero tensor, yielding
SymV1,(A) = 1. (46)
Note that the derivation yields no information about the skew part of VI, (A), which may
be arbitrary. It is very common to simply set the skew part to zero, and to equate VI, (A)
to SymVI, (A). In particular, it is impossible to determine the skew part from the analysis;
however, it is quite unnecessary to do so. This convention extends to any scalar field defined
on the linear space of symmetric tensors.
Next, consider
L(A) = tra’. (47)
Problem
Prove that trA* = 11 (tra)? - tr(A?)].
Then,
1,(A) = Cn _1-A%), (48)

and
- og de, ,
VI(A) A= h =i, ~51- (AA
+ AA). (49)
Using the trace definition of the inner product we can show that
I. (AA) =I. (AA) =A-A. (4.10)
Thus,
[SymV1I,(A) ~ (,1- A)]-A=0 (4.11)
for all symmetric A, yielding
SymVI(A) =L1-A. (4.12)
Finally, recall the standard result
J = F*. Ff, (4.13)
where F* is the cofactor ofF andJ = det F. By the same reasoning, with
1,(A) = detA (4.14)
we get
[SymVI,(A) - A*] -A=0, (4.15)
and so
SymVI,(A) = A’, (4.16)
where
A*=1,A" (4.17)
if Ais invertible.
5. Relations among gradients

In Elasticity we encounter the need to relate the gradients of the two sides of the equality
W(E) = G(C) where C = F'F and F € Lin* and C € Sym’. A path F(t) in the former set
induces a path C(t) in the latter and the equality may be differentiated to obtain
W,-F=Gc-C=Gec- (FF + FE). (3.1)

Using the symmetries of the inner product operation we write
Gc: FF =F(G,.)'-F and Gc-F'F = F(G) - F, (52)
so that
{Wr -F[Gc + (Gc)']} - F = 0. (53)

Since the path is arbitrary, F is an arbitrary element of Lin. The collection of terms in braces
also belongs to Lin, therefore,
Wy = E[Gc + (Gc)'] = 2F(SymGe). (54)
The symmetry ofC means that only the symmetric part of Ge is determinate and it is only
this part which appears in the result. Indeed, we may use the fact that C is symmetric to
replace Ge by SymGg in eqn (5.1) at the outset.
If G is an isotropic function of C then it depends on the principal invariants I,(C) and
the chain rule provides
3
(SymGe) - > G,[Sym(i)c] -C= 0, (5.5)
kel
where G, = 0G/dI,. The term in braces, an element of Sym, is thus orthogonal to every
other element of Sym. Therefore, it vanishes, yielding
SymGc = (G, + 1,G,)I1 - G,C + G;C*. (5.6)
6. Extensions
In the literature one often encounters component formulas like
OW/dF.4 = 'a(9G/ACa, + 9G/dCaa) (6.1)
in place of eqn (5.4) above. However, we have seen that the representation of the gradient
in terms of partial derivatives is possible only if the components of the tensor argument are
all independent. This is not the case here because C4; = Cy, and so eqn (6.1) cannot be
valid as it stands.
In practice, the issue is moot because the scalar-valued function Gis usually given and a
procedure like that demonstrated in the previous section may be used to compute the gra-
dient. Nevertheless, eqn (6.1) arises frequently in theoretical studies and the question of its
validity is thus of independent interest. First, we note that this formula follows immediately
from the chain rule
OW/OFx = (8G/OCgc)OCgc/OFia, with Cac = FaFicy (6.2)
provided that the partial derivatives dG/0Czgc are interpreted in the usual sense of holding
fixed all components other than the one with respect to which the derivative is taken. This
suggests that we introduce an extension H of G from Sym* to Lin. Thus, H(C) is defined
for C in Lin and satisfies H(C) = G(C) for C in Sym* C Lin. We assume the extension to
be differentiable in Lin and conclude, for any path C(¢) in Sym’, that
(He - Ge) -C = 0 (63)

for any C in Sym. Because the first factor belongs to Lin = Sym ® Skw it follows that
Ge = He + W, (64)
where W € Skw. Furthermore, since He = (@H/@C,3)E, ® Ez it follows that eqn (6.1)

holds with G replaced by H.
Given G, an obvious choice for H, which automatically satisfies the requirements of a
smooth extension, is
H(C) = G(SymC), C € Lin. (65)

Normally, this is the only extension discussed, either implicitly or explicitly. An exception
is the paper by Cohen and Wang (1984), where the general issue is treated in detail. An
obvious question arises as to whether or not any generality is lost if eqn (6.5) is adopted. To
examine this, let H’ be another extension of G to Lin; then, for C € Sym*, H’(C) = H(C)
identically and
(Hi, -H-)-C=0 (66)
for any C in Sym. As before, we conclude that
Ho =He +W, (67)
for some skew W,, and it follows that any smooth extension ofG may be used in eqn (6.1)
without affecting the result. It follows that no generality is lost by adopting the obvious
extension given by eqn (6.5).
REFERENCE
Cohen, H. and Wang, C.-C. (1984). A note on hyperelasticity. Arch. Ration. Mech. Anal. 85, 213.
7. Korn’s inequality
An easy calculation yields 2u;45)4(4,8) = “apap + Uaptaa, Where round braces are used
to denote symmetrization, i.e., 24(4.3) = Uap + Ug. Write the second term as (tp), -
U,sUp, integrate over K, and use the divergence theorem to obtain
2 f urneaasdr [iver avs f N- [(Vu)alda- fw: V(Divu)dv. = (7.1)

K K OK x
Write the third integrand on the right as Div(uDivu) - (Divu)? and apply the divergence
theorem again. For the special case in which u = 0 on Ox, all the boundary integrals vanish
and we get
2 fas mand = [iver dv + [ona (72)
This furnishes an example of Korn’s inequality
kf antaard > [iver a, (73)
where k is a positive constant depending only on the shape of the region «. In the present
example (u = 0 on 0x) we have k = 2 forall regions, and this is the optimum value because
eqn (7.2) implies that eqn (7.3) is satisfied as an equality when u(x) has zero divergence,
whereas strict inequality obtains in the general case. Furthermore, in this case the optimum
Korn constant happens to be the same for all «. Ifu vanishes on a portion of 8x, eqn (7.3)
remains valid, but the constant k then depends on the region. The optimal constant is an
eigenvalue of a variational problem associated with eqn (7.3).
Horgan’s paper (1995) contains an accessible account of various applications of Korn’s
inequality to Mechanics.
REFERENCE
Horgan, C.O. (1995). Korn’s inequalities and their applications in continuum mechanics. SIAM
Rev. 4, 491-511.
8. Poincaré’s inequality
Poincaré’s inequality is the assertion that there is a positive constant c such that
fiver dy > cf Ist dv. (8.1)

As in the case of Korn’s inequality this is most easily proved for the case in which u = 0
on dx. Let f(x,) be a single component of the vector field u(x). Then, f vanishes on the
boundary.
Consider a cross section ofx defined by x, = z(x,, x,), and let z,,(x,, x.) be the minimum
value of this function on 8x for a given point (x, x) of the cross section. Then, z,, is the
x3-coordinate ofa point on the boundary (draw a figure). For the fixed values of («,, #,) in
question we have f(x, *2,2,,) = 0, and, therefore,
Ff (%, 2,2) =f hoax = [oe Vf)dx; < fis dx, (8.2)
<([[4)" (es) "-vee( ft)

zm zm zm
Z—Z, ( / . |vé|’ ins) _

IA
where Z4(, x,) is the maximum value of the function z over « at the same values of (x, x,)
and we have made use of the Cauchy-Schwartz inequality in the 2nd line. We square and
integrate with respect to x, to obtain
fidx; < 5 hm) / | Vf] das < 5H | [Vf] des, (8.3)

zm zm 2m
where h(x), #2) = Zw — Z,, and H is the maximum of h (the maximum thickness of« in the
x3-direction). We now integrate eqn (8.3) over the x,, «,-plane to get
1
[re < sit f ivsl av (8.4)
Recalling that f is a component of u, we apply eqn (8.4) three times to find that
3
¥ ftv dv > cf usuady, (8.5)
K
Az1
where c is a positive constant and Vu, = ua,E3. However, >> |Vual’ = >0(Vua+ Vua) =
Hasta = |Vul’. Thus, eqn (8.5) is just eqn (8.1).
Poincaré’s inequality is a special case of the Sobolev inequalities. One can find a deriv-
ation for general u(x) in any book about Sobolev spaces (e.g. Sobolev 1963). The same
inequality remains valid (with a different c, naturally) for functions u(x) that vanish on
a part of 0x. See, for example, the book by Morrey (1966), which has had an enormous
influence on the mathematical development of nonlinear elasticity theory.
REFERENCES
Morrey, C.B. (1966). Multiple integrals in the calculus of variations. Springer, Berlin.
Sobolev, S.L. (1963). Applications of functional analysis in mathematical physics. American
Mathematical Society, Providence.
INDEX
Note: Figures are indicated by an italicf following the page number.
A exhibiting radial symmetry chain of equalities 9

acoustic tensor 88, 89, 92, 93, with respect to a fixed to compute gradients via 45
101, 130 point 67-73 to compute the velocity
axial force 62, 66, 106, 107, 109, pressurized cylinder 51-9 gradient 163
142 pressurized shells and to obtain stress 29
axial stress 142, 143 cavitation 71-3 Ciarlet, P.G. 122
axis ofsymmetry 62, 104 torsion of a solid circular classical flow rule 161
cylinder 63-5
azimuthal angle 59, 63, 104 classical Lamé moduli 148
for uniform isotropic
azimuthal shear $9-63, 84 classical mechanics 7
incompressible
classical theory 156, 160
materials 51-74
B buckling 67, 125, 126, 137
central equation of 161
Ball, J.M. 111, 121, 127, 131 Burgers vector 145 coexistent cylinders 109
Ball’s existence theorem 122, 127 coexistent phase problem 111
bars 58, 66, 67, 143 Cc cofactor 12, 30, 79, 167, 171
cylindrical 59 Cauchy elasticity 17,23 Cohen, H. 173
deformed 125 Cauchy-Green deformation collagen fibers 58
neo-Hookean 58 tensor 11,29, $3, 56, 147 Como, M. 113
simple tension of 142 Cauchy-Schwartz inequality 175 compatibility 47, 48
solid circular 72 Cauchy stress 5, 6, 12, 16, 47, 49, compressible inviscid fluid 25, 135
Bauschinger effect 153 64, 77, 147, 158, 160, 161 compressible isotropic
bifurcation 72,78, 141 constitutive function 24 materials 83
from equibiaxial to unequal divergence 33 configuration ofa body 8f
biaxial stretch 121 in an incompressible isotropic conservation of energy 119
of equilibria 73, 126 material 53 conservation of mass 13, 16
Biot stress 32 indeterminate 48 constitutive contribution 138
blocks
in an inviscid compressible constitutive functions 2, 5-6, 10,
fluid 25 12, 13, 17, 25, 27, 33, 46, 135
isotropic 73
ata material point 4,11 for Cauchy stress 24
kinematics of pure flexure of 73
in aneo-Hookean material 55 domain of 118
simple shear of S6f
principal 106, 124
body force 10, 54, 58, 59, 69, 100, isotropy for 156
radial and hoop components
145 restriction on 9
of 69
Boehler, J.P. 40 scalar-valued 157
symmetry of 18, 19, 30, 123,
boundary conditions 54, 58, 59, 146, 152 constitutive response 146-8
62, 69, 71, 87, 96, 99, 108, see also stress of a Varga-type material capable
139, 141, 163 Cauchy traction 22 of supporting cavitation 77f,
boundary traction 144, 145 cavitation 71-3, 77, 78, 82, 83 82
boundary-value problems 41 plane-strain 82f constraint manifolds, and
azimuthal shear 59-63 shellsand 71-3 the Lagrange multiplier
combined extension and Chadwick, P. 66, 167 tule 43-7
torsion 66-7 chain rule 23, 27, 32, 46, $2, 70, constraints 137, 14]
exhibiting radial symmetry with 90, 91, 93, 114, 151, 159, 169, frame invariant 137
respect to a fixed axis $1-67 172, 173 local 42-50, 117
178 | INDEX
constraints (continued) uniqueness of 161 deformed surface 55

material symmetry in the deformation gradient 5, 7, 20, 33, position and director fields 98f
presence of 47-9 42, 52, 53, $6, 62, 63, 68, 76, degree of freedom 158
non-linear frame-invariant 137 78, 89, 128 deviatoric tensors 45, 159
pressure 49, 61, 64, 69, 85, 87, cyclic process in diameter 5,95
97, 102, 103, 139 deformation-gradient ofabody 145
simultaneous 141 space 21f of elastic range 160
continua 2 discontinuity in 92 differentiability 99, 159
continuum mechanics §, 7, 9, 11, equilibria with direct sums 44-5, 159
13, 15, 19, 28, 143 discontinuous 131-2 director fields 97, 98f, 103
continuum theory 2,3, 4 membrane 104 director, uniqueness of 100-2
convex functions 101, 128, 132, pointwise value of 24 discontinuities 91, 92,94
147 rank-one perturbations of 130 discontinuity surfaces 92, 93, 131,
convexity 130 scalar-valued function of 17 133, 134
ordinary 123-6, 127 deformations 3, 4, 5, 8, 13, 18, 22, dislocations 146, 153
rank-one 128-31, 132, 133, 23, 24, 26, 77, 130, 161 dissipation 153, 155, 161
134 axially symmetric 104-8 energy and 142, 148-52
crystalline materials 142, 144, cyclic process in local 150-1
148, 156, 158 deformation-gradient superposed rigid-body
cubes 31,49, 73, 121 space 21f motion 152
curves 20, 43, 44, 57,97, 134, 142, cylindrically symmetric $1 distortions, elastic 145, 148, 153
145, 163, 169 of directionally reinforced divergence 10, 13, 33, 57, 69, 139,
closed 145 solids 42 174
described by a differentiable elastic 143-5, 146, 160 operator 139, 162
function 169 elastic-plastic 156, 161 referential 13
force-deflection 3 equilibrium $1, 78, 82, 83, 87, spatial 10
force-extension 58 89, 92, 110, 116, 124, 125, theorem 15, 117, 127, 149, 174
inextensible 43, 47 126, 131, 135, 137 two-dimensional 97
smooth 21 field 19 dominated convergence
wavy 58 function 2, 5, 47 theorem 122
cyclic processes 19, 20, 21, 154 homogeneous 20, 31, 56, 57,
cylinders 58, 73, 87, 89, 121, 122, 125 E
axially symmetric isochoric 42, 49, $3, 59, 60, 64, elastic bodies 16, 140
deformations of cylindrical 66, 68, 98, 164 elastic deformations 143-5, 146,
membranes 104-8 non-isochoric 42 160
bulging of 108-12 nonequilibrium 120 elastic distortions 145, 148, 153
coexistent 109 path 40, 114 elastic energy for small strains 147
cross section $2f with piecewise uniform elastic Lagrange strain 147
deformed 66 gradient 133 elastic-plastic deformations 156,
pressure-volume characteristic plastic 143-5 161
for purely cylindrical pressurized cylinder 52 elastic range 153, 154, 155, 156f,
deformations 111f rate of 16, 143 160
pressurized $1~9, 108 response to 2 elastic strain 144, 147, 151, 153,
torsion ofa solid circular 63-5 rigid-body 8 155, 158, 160
shear 55, 80, 142 elastic stretch 143, 153
D simple-shear $5 elasticity 3, 143
dead loading 22, 117, 118, 121, spherically symmetric 77,78 Cauchy 17,23
124, 125, 126, 139 stable 140 definition of 1
decomposition 96, 160, 161 stress response in the presence and energy 15-19
direct-sum 44 oflocal constraints on 42-50 general isotropic 57
polar 11, 30, 32, 33, 59, 68, 79 in unconstrained isotropic ellipticity 93, 94, 99, 100, 101,
spectral 80 materials 75-84 107, 108, 111, 115, 132
INDEX | 179
energy Euler criterion for potential relations among 171~2

comparison 140 instability 126, 137 of scalar-valued functions of
conservation of 119 Euler equation 141 tensors 168
and dissipation 142, 148-52 Euler—Hill-Trefftz criterion 141 spatial velocity 15, 160
elasticity and 15-19 Euler’s laws of motion 11 two-dimensional 97
asa function ofshear 135f extended strain—energy velocity 15, 160, 163
kinetic 15, 16, 22, 117, 149 function 46, 102 see also deformation gradient
mechanical 16, 22, 119, 149 extension 172-3 Grimaldi, A. 113
minimizers 2, 121, 122, 123, combined torsion and 66-7 groups 25
130, 132, 134 rates 164 Gurtin, ME. 13, 17, 38
norm 113-16, 117, 120
potential 22, 113, 118, 119, F H
120, 121, 122, 124, 125, 130, harmonic materials 82, 83, 84
fiber symmetry 35-41
135, 139, 150 homogeneous deformation 20,
first variation 139
symmetry groups 28 31, 56, 57, $8, 73, 87, 89, 121,
flexure 5,73
energy balance, mechanical 149 122, 125
energy criterion, stability and flow rule 158, 159, 160, 161
Horgan, C.O. 41,174
fluids 25, 27, 30, 135
the 113-36 hyperelasticity, mechanical power
Epstein, M. 1$1 force 1, 3,67 and 15-22
equation of motion 12, 13, 33, 86, axial 62, 66, 106, 107, 109, 142
149 body 10, $4, $8, 59, 69, 100, I
equations of equilibrium 145
incompressibility 42, 43, 46,
elliptic 92 -difference density 99 47, 48, 49, 53, $7, 66,
scalar 60 inertial 10 75, 86, 93-4, 99, 121,
equilibrium 2,95, 126 net-force density 99 130
bifurcation 73, 141 power of 149 incompressible materials 42, 48,
with discontinuous deformation formulation of elasticity theory 12 97, 98, 101, 102, 139
gradients 131-2 Fosdick, R.L. 130 boundary-value problems for
equations of 60, 92 Fraejis de Veubeke, B.M. 140 uniform isotropic 51-74
instability of 126 frame invariance 7, 137 inertial force 10
non-uniqueness of 126 frictional effects 142 inextensibility 43, 46,47, 48
plane strain 82 infinitesimal strain tensor 115
smoothness of 89-93, 99 G instability 73, 113, 116-21
stability 67, 112 of equilibrium 126
Geiringer equations 164
stress 117,139 potential 126, 137
Geiringer, H. 164
unloading process 145 intermediate configuration 143,
general isotropic elasticity 57
without body force 49, 57, 59, 144, 146, 152, 159
Goddard,J. 40
62, 63, 71, 81, 138, 162 invariance 5, 6, 7, 10, 18, 19, 25,
equilibrium deformations 87, 89, gradients 5, 19, 20, 23, 24, 26, 33, 28, 33, 46, 146, 152-3, 156,
92, 116, 124, 125, 126, 131, 44, 45, 46, 47, 54, 93, 129 157
135, 137 computing 45, 172 invariants 6, 10, 13, 29, 32, 33, 36,
cavitated 78, 82 constraintson 42 37, 48, 56, 80, 169
gradients of 125 deformations 49 frame 7, 137
radial 83 equilibrium deformation 124, principal 29, 31,35, 63, 172
simple $1 125 translation 10
uniform 110 membrane deformation 104 inverse function theorem 23
Ericksen, J.L. 134 operator 162 isochoric deformations 42, 49, 53,
Eshelby’s Energy-Momentum ofa position field 17, 145 59, 60, 64, 66, 68, 98, 164
Tensor 150, 151, 152, 153 of the principal invariants of a isochoricity 102
Euclidean inner product 15 symmetric tensor 169-71 isotropic elasticity $7
Euler backward differencing 108 quasiconvexity 121, 122 isotropic functions 36, 37, 172
180 | INDEX
isotropic materials 2.5, 30, 48, 56, linear space 17, 18, 44,45, 170 mean stress 142,145
68, 100, 102-4, 148, 156, 159 deviatoric tensors 159 mean value theorem 129, 145,
block 73 symmetric tensors 170 154
compressible 75-84 oftwo-vectors 101 mechanical energy 16, 22, 119,
incompressible $1-74 linear theory 65 149
plane strain of compressible 83 of small equilibrium deform- mechanical power, and
stress-strain relation 148 ations superposed on finitely hyperelasticity 15-22
symmetry groups 148 deformed equilibrium membrane deformation
transverse 29, 40 state 137 gradient 104
isotropic solids see isotropic linearity 62, 131, 167 membrane problem, h-multiplied
materials local constraints 117 version of 103
isotropy 28-33, 48, 156-9, 160 stress response in presence membrane theory 95-112
transverse 30, 35, 36, 37, 41, 48 of local constraints on bulging of cylinders 108-12
deformation 42-50 general theory 95-9
J local stress 142 pressurized 99-100
James, R.D. 131 local undistorted configuration 25 reference and deformed
localization theorem 13, 131,150 configurations 107f
K unpressurized 107
kinetic energy 15, 16, 22, 117, 149 M metals 142, 158
Knops, RJ. 113 MacSithigh, G.P, 130 crystalline 144
Korn’s inequality 116, 120, 174, Man, C.-S. 40 elastic strains 147, 160
175 manifolds 145, 153 Mikhlin, $.G. 113
Kronecker delta 69, 167 constraint 43-7 modern theory
Kuhn-Tucker necessary mass, conservation of 13, 16 of crystalline media 156
conditions of optimization mass density 10, 13, 149 for finite elastic-plastic
theory 155 material points 5, 7, 13, 15, 17, 23, deformations 156,161
24, 45, 59, 98, 114, 135, 144,
of finite elasticity 143
L 145, 146
Mooney-Rivlin material 121
Lagrange-multiplier rule, Cauchy stress at 4, 11
Mooney-Rivlin strain-energy
constraint manifolds constraints at 43
function 73
and 43-7 deformation gradient at 20
Morrey, C.B. 122,175
Lagrange-multiplier theorem 42 deformations 4
motion 6, 22, 33, 116, 119,
Lagrange multipliers 45, 47, 48, displacements of 85
121
49, 87, 88, 103, 138, 139, 141, distance between 8
equations of 10, 12, 13, 33, 47,
155 distance from a specified
86, 149
Lagrange strain 114, 147 origin 67
Euler's laws of 11
lateral pressure loading 99 fiber axis at 35
forced 16
lattice 25, 26, 142, 144, 146, 148 invariance 152
having gradient in a ball 121
leading-order during a time interval 19
material stability 88, 89, 93, 94,
one-dimensional 16
approximate model 95
differential algebraic 101, 115, 131 tigid-body 33
small motions superposed on
problem 99 material symmetry 23-34, 35, 40,
strain energy 147 156, 187, 158, 159, 161 finitely deformed equilibrium
left Cauchy—Green deformation in the presence of states 85-9
tensor 29, $3 constraints 47-9 superposed rigid-body 10, 151,
Legendre-Hadamard transformation 23 152, 153
inequality 130, 140 matrices, spectral theorem for unforced 16
lens-shaped region 128 symmetric 115 Miller, 1. 1
light, speed of light in vacuum 7 Maugin, G.A, 151 Mullins effect 1
linear map 167 Maxwell equal-area rule for phase multi-scale method 3
linear shear modulus 76 coexistence 111, 134 Murdoch, Al. 7, 8,9, 10
linear shear stress distribution 65 Maxwell-Eshelby relation 132-5 Murphy, J.G. 41
INDEX | 181
N perfectly plastic materials 162, pressure 22, 42, 54,71, 83, 94,
Nanson’s formula 13, 88, 98, 127 163 100, 103, 109, 111, 135, 138,
natural boundary condition 141 permanent stretch (plastic) 143 158, 161
neighborhood, of a material permutation 12, 127, 167 compressive 140
point 4, 23,24 phase transformations, in constant 62
neo-Hookean material 60, 62, 65, shape-memory alloys 131 constraint 49, 61, 64, 69, 85,
66, 72, 107 phenomenological theory 1, 159 87, 97, 102, 103, 139
Cauchy stress $5 physics 3 -density relation 25,30
shear response 56 development of 7 external $7
strain energy $6 relativistic 10 field 62, 67
stress 58 pill-box 131 inflation $5, 106
neo-Hookean model 55, 64 Piola identity 13, 75, 87, 127 lateral loading 99
neo-Hookean strain—-energy Piola—Kirchhoff stress 12, 13, load 22
function 49, 55, 89 114, 147, 182 negative 71,72
net-force density 99 Piola stress 12, 13, 16, 32, 79, 81, pressurized cylinder 51
Noll, W. 4, 5, 7, 8, 25, 146 83, 85, 147 pure 47
Noll’s Rule 27 Piola traction 56, 82, 117,119 -volume characteristics 111f,
Noll’s theory of materially uniform plane harmonic wave 87, 89 135f
simple bodies 146 pressurized cylinders $1-9, 108
plane strain 72, 73, 78-80, 82f
non-crystalline materials 25, 156 pressurized membranes 99-100
compressible isotropic
non-uniform materials 3, 5, 24 pressurized shells 71-3
materials 83
nonconvexity 118 principal stress axes 124, 162
condition 161
nonlinear elasticity theory 4, 51,
equilibrium 82 principle oflocal action 4
$7, 71, 131, 175
of rigid-perfectly plastic proof
nonlinear frame-invariant
materials 161-4 of necessity 37, 38
constraints 137
see also strain of sufficiency 37
nonlinear simple shear
response $7
plane stress 100
nonlinear spring 56
plane-stress condition 100, 102 Q
normal stress $7, 65, 66, 67 plastic deformations 143-5 quasiconvexity 121-3, 126, 127,
plastic evolution 151, 152, 156 128, 129, 130
oO plastic flow 153, 154, 156, 157, Quinney, H. 159
159
observers 21,33, 43
and invariance 7-14
plastic materials, perfectly 162 R
relationship between a pair of plastic spin 156, 158 radial compaction 80-4
classical 8 plastic stretch (permanent) 143 radial expansion 80-4
Ogden, RW. 32, 43, 51, 56, 110, plasticine 142 radial symmetry
111, 127, 137, 141 plasticity, perfect 159 with respect to a fixed
optimization problem 155 plasticity theory 142-65 axis 51-67
optimization theory 155 Podio-Guidugli, P. 47 with respect to a fixed
Kuhn-Tucker necessary Poincaré’s inequality 116, 174-5 point 67-73
conditions of 155 pointwise condition 129, 132, 140 Rajagopal, K. 1
optimum Korn constant 174 polar decomposition 11, 30, 32, rank-one convexity 128-31, 132,
ordinary convexity 123-6, 127 33, $3, 59, 68, 79 133, 134
orthogonality 36 polarization vectors 87, 89 rate-independent theory 143
orthonormal vector fields 162 polyconvexity 111, 126-8 relativity theory 7
polymers 142 residual stress 49, 144
P positive-definite symmetric restriction 13, 24, 25, 28, 40, 42,
paint 142 tensors 11, 18,29 48, 53, 66, 83, 88, 116, 146,
partial derivatives 172, 173 potential energy 22, 113, 118, 148, 155
perfect incompressibility 42 119, 120, 121, 122, 124, 125, constitutive equations for
perfect plasticity 159 130, 135, 139, 150 plastic evolution 151
182 | INDEX
restriction (continued) shear deformation 55, 80, 142 stable deformation 140
on constitutive equations used shear modulus 55, 76, 108 static perturbation 137
by observers 10 shear response stationarity problem 102
on constitutive function 9 neo-Hookean material 56 Steigmann, DJ. 127
on deformation 64, 122 nonlinear simple 57 stiffening 2f
distinct eigenvalues 38 nonlinear simple shear 56 strain
general boundary-value shear strain 49, 142 elastic 144, 147, 151, 153, 155,
problem 124 shear stress 56, 57, 62, 65, 134, 158, 160
idealization of zero elastic 142, 162 elastic strain function 155
strain 160 shear traction 31 Lagrange 114, 147
imposed by material shells and cavitation 71-3 rotated 156
symmetry 35 shooting scheme 108 shear 49, 142
imposed by the constraint of simple shear 56, 57, 60, 62, 64, 6S, see also plane strain
incompressibility 86 133-5 strain energy 22, 23, 27-8, 32, 77,
incompressible material 93 alternating 133-4 81, 84, 103, 109, 110, 113,
material symmetry 157 ofablock 56f 127, 132, 146, 158
mild 6 deformation 55 density 16, 146
pointwise 20 strain energy 134 fixed value 160
on strain-energy function skew 18, 123, 124, 156, 157, 158, incompressible material 83
17,63 161, 167, 170, 173 invariance under superposed
on stress response 27 skewtensor 45, 156, 168 rotations 19
on value of deformation slip 146, 164 invariant 35
gradient 130 slip-line fields 164 for isotropic materials in a
tight Cauchy-Green tensor 11, slip-line theory 161-4 plane-strain deformation 79
147 small elastic strain 147, 151, 155, leading-order 147
rigid body 140, 160 158 neo-Hookean material 56
deformation 8 small motions, superposed on referential description of 148
motions 10, 33, 151, 152 finitely deformed equilibrium Tesponse 27
rigid-plastic materials 160-1 states 85-9 in simple shear 134
Rolle’s theorem 132 smoothness of equilibria stress and 44
rotated strain 156 89-93, 99 three-dimensional 121
rotation 11, 17, 28, 30, 32, 35, snow 142 uniform material 122
43, 60, 63, 84, 123, 147, 148, Sobolev inequalities 175 strain energy function 18, 21,71,
156, 157 Sobolev spaces 175 72,75, 76, 83, 88, 92, 111,
arbitrary 18, 158 Soldatos, KP. 5 114, 118, 127, 132, 156
crystal lattice 26 solids 5, 25, 28, 30, 42 admissible 28
discrete 25 crystalline 148 compressible inviscid fluid 135
factor 32, $9, 79 isotropic 148 deformations of rubber 56
field 160 spatial domain of 130
finite 82 divergence 10 expanding to quadratic
rate of 66 mass density 10 order 158
superposed 19, 146, 152 velocity gradient 15, 160 extended 46, 48, 102
rubber band 1, 2f, 3, 4, 20, 72, 95 spectral theorem for symmetric isotropic elastic material 30,31
rubber elasticity 3, 95 matrices 115 local constraints 43
speed of light in vacuum 7 Mooney-Rivlin 73
S Spencer, AJ.M. 5 natural extension of 48
second variation 140 spherical symmetry 75-8 neo-Hookean 49, $5, 89
of the energy 141 spherical tensors 159 Ogden 110
shear 57, 133, 134 spin, plastic 158 restrictions 63
azimuthal $9-63, 84 spring, nonlinear $6 transversely isotropic 35S
energy as a function of 135f stability 88, 89, 93, 94, 101, 115, uniform $8
see also simple shear 131 Varga $5
INDEX | 183
strain hardening 156, 189 function 4 gradients of the

straining tensor 160 hoop 68,71 principal invarients of
strains, elastic energy for left stretch tensor 102 symmetric 169-71
small 147 permanent (plastic) 143 infinitesimal strain 115
Strehlow, P. 1 plastic 143 invertible 23
stress 62, 64, 137, 144, 146, 154, principal 30, 31, 32, 49, 55, 68, orthogonal 8, 10, 30, 35
159, 162 78, 80, 105 positive-definite symmetric 11,
axial 142, 143 radial 68 18, 29
Cauchy 123, 124, 147,158, 161 right stretch factor 79 rank-1 92
computing 47 triaxial 76, 80, 89 skew 45, 156, 168
and elastic stretch 153 strict convexity 125, 126 spherical 159
equilibrium 117, 139 strict minimizers 125, 126 strain-dependent moduli 114
field 164 structural tensors 36, 40 straining tensor 160
as a function of shear 135f superposed rigid-body motion 10, stress 59
incompressible 63,93 151, 1$2, 153 stretch 102
local 142 surface limits, upper and lower 91 structural 36, 40
mean 142,145 symmetric 11, 18, 29, 38, 39,
symmetric tensor 11, 18, 29, 38,
neo-Hookean material 58, 60 159, 160, 168, 169-71
39, 159, 160, 168, 169-71
normal $7, 65, 66, 67 vorticity tensor 160
symmetry
Piola 12, 13, 16, 32, 79, 81, 83, axis of 62, 104
zero 170
85, 147 theory of elasticity, for transversely
of Cauchy stress 18, 19, 30,
Piola-Kirchhoff 12, 13, 114, isotropic materials 40
146, 152
147, 152 theory of relativity 7
fiber 35-41
plane 100, 102 theory of small deformations
general restriction imposed by
power 15, 16, 17,149 superposed on large 137
material 35
removal of 144 thin sheets 95, 96f
spherical 75-8
residual 49, 144 torsion
see also material symmetry;
rigid-plastic materials 160 combined extension and 66-7
radial symmetry
shear 56, 57, 62, 65, 134, 142, ofa solid circular cylinder 63-5
symmetry groups 25, 26, 27, 28,
162 torsional buckling 67
30, 35, 73, 148, 158
space 153,159 Toupin, RA. $
symmetry transformations 23,
state 162 traction 15,31, 49, 56, 61, 63, 64,
24, 25
statically determinate 163 65, 73, 99, 131, 134
symmetry groups 28 boundary 144, 145
T Cauchy 22
tensor 59
triaxial 71 Taylor, G.I. 143, 189 compressive 125
uniaxial 142, 143 Taylor's theorem 118 conditions 103
using chain rule to obtain 29 tensors 15, 17,29, 32, 40, 45, 53, dead-load 124
vectors 104, 105 83, 86, 92, 98, 144, 167, 169 on inner boundary 62
see also Cauchy stress acoustic 88, 89, 92, 93, 101, Piola $6, 82, 117, 119
stress response 23-7, 28, 30 130 shear 31
in the presence of local argument 172 vanishing 67,72
constraints on the Biot stress 32 zero 59
deformation 42-50 Cauchy-Green deformation traction-continuity condition
stretch 1, 3,30, 58, 70, 72, 73, 79, tensors 11,29, 53, 56, 147 equation 134
102, 142, 143 deformation 59 translation invariant 10
axial 66 deviatoric 45, 159 transverse isotropy 30, 35, 36, 37,
biaxial 73, 81, 121 elastic moduli 113, 147 41, 48
collagen fibers $8 Eshelby’s Energy-Momentum transversely isotropic materials
constant 73 Tensor 150, 151, 152,153 29, 35, 40, 148
elastic 143,153 gradients of scalar-valued Treloar—-Kearsely instability 73
finite 72 functions of 168 Treloar, L.R.G. 95
184 | INDEX
Tresca 159 polarization 87,89 Y

triangle inequality 116, 118 stress 104, 10S yield
Truesdell, C. 3 wave 87 criterion equation 160, 161
twisting moment 65, 67 vector fields, orthonormal 162 function 153, 156, 157, 158,
vector-valued function 8, 120 159, 160
U vector-valued map 167 insensitive to pressure 158
uncertainty principle of quantum velocity field 19, 161, 163-4
stress 142
mechanics 7 velocity gradient 15, 160, 163
surface 155,159
unimodular groups 25 virtual-work theorem 21
yield stress 1$3
unique director field 103 von Mises flow rule 159
axial 142
unpressurized membranes 107 von Mises function 159
inshear 159
vorticity tensor 160
in uniaxial tension 153
Vv
WwW Young’s modulus $8, 142
vanishing traction 72
Varga strain-energy function $5 Wang, C.-C. 173
Varga-type material 57, 76, 77f, wave vectors 87 Zz
81, 82f Wilkes, E.W. 113 zero tensor 170
vector Wineman, A.S. 4 zero volume measure 133
Burgers 145 work inequality 19-22, 48, 153-6 Zheng, Q.-S. 40

(9780198567783) David J. Steigmann - Finite Elasticity Theory

Uploaded by

Copyright:

Available Formats

(9780198567783) David J. Steigmann - Finite Elasticity Theory

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(9780198567783) David J. Steigmann - Finite Elasticity Theory

Uploaded by

Copyright:

Available Formats

PREFACE

David Steigmann, Berkeley 2016

Antman, S.S. (2005). Nonlinear Problems of Elasticity. Springer, Berlin.

Concept of an elastic material 6.1... 6... cece eee ee ees

Mechanical power and hyperelasticity... 0.0.0.0...

10 Membrane theory... . 0... ccc eee eee eee eee e eens

Chain rule 169

Poincaré’s inequality 174

Finite Elasticity Theory. DavidJ. Steigmann.

which offers an interesting explanation in terms of microstructural instabilities and associ-

Figure 1.2. Neighborhood of a material point, p, in the reference configuration

Xe (x, t) = Xe (x, t) + E, (x, t)(x’ - x) + o( |x’ -x|), (1.1)

Tip, t) = Ge Xu (x, t), F, (x, t);x), (1.2)

underpinnings of the subject. We might have included a dependence on time, but as we

J >0, where J = detF,. (1.3)

Finite Elasticity Theory. DavidJ. Steigmann.

xi, 6) = Q(t)x,(%t)+e(t) and ff =tta, (2.1)

Figure 2.1 Configuration of a body, as perceived by observers O and O*

observer (© to serve as reference; in short, « = R,,. We further suppose that observer O*

dy’ = Qdy = QF ,dx= QE, K'dx’, andthus Fi, =QEK’, (2.3)

the latter implying that

(det Q)(det K) = 1. (24)

T*(p, f°) = Gx @, #), Fe (x*, #);x*). (2.5)

Gx t), F, (x, t);x)Q(t), (26)

Qc (y,, F,; x)Q, = Q,G (y,, F,; x)Q',, (2.7)

Quy, te =Q.,y,+&, (2.8)

F,=QF,, where Q=Q‘,Q). (29)

G.(Qy +d, QF;x) = QG,(y,F;x)Q, where d=Qic with c=e~c. (211)

This is precisely the requirement that the constitutive function G, pertaining to O be

dvT+pb=py and T=T', (2.12)

Gi. (y + d, F; x) = G, (y, F; x), (2.13)

G.(y, F; x) = G(x), (2.14)

G,(QF;x) = QG,(F;x)Q’ forall Qe Orth’. (2.15)

G,(F;x) = RG, (R'E; x)R' = RG, (U;x)R' = FG, (U;x)F,,

G, (F; x) = FH,(C;x)F', (2.17)

G, (QE; x) = QFH, [(QF)'QF;x](QF)! = QFH, (FF; x)F'Q = QG,(F}x)Q,, (2.18)

T(p,t) = FH, (C;x)F’. (2.19)

Gi. = FYHY, (C*jx")(F*)', (2.20)

H?,(C*;x') = KH, (K'C'K; K'(x* — ¢))K' (2.21)

where F* is the cofactor ofF defined by

S=§,(C;x), where §,(C;x) = /detCH,(C;x), (2.27)

P =P,(F;x), where P,(F;x) = F§,(F'F;x). (228)

DivP + p,b = px, (2.29)

Gurtin, M.E, (1981). An Introduction to Continuum Mechanics. Academic Press, Orlando.

3.1 Elasticity and energy

P(S,t) -| toda fb jae S(S,t) = [rts 3.2)

Finite Elasticity Theory. DavidJ. Steigmann.

The balance (3.1) presumes conservation of mass; that is,

and 7 is the subset of the fixed region x associated with S.

U(S,t) = / Wd», (3.6)

where W is the (referential) strain-energy density.

S(S,t) = / JT. FF'dv = / tr(JTE“E' )dv = / P - Fdy, 37)

and, therefore, according to the analogy,

/ P. Fdy = / Wd». (38)

[P, (F; x) — W.(F;x)] - F = 0, (3.10)

P,(F;x) = W,(F)x). (11)

Thus, the stress is determined by a scalar-valued function of the deformation gradient,

P, (QF) = QP, (E) (3.12)

P,(F) = QP, (B), (3.13)

W(F) = B,(F) - F = QP, (F) - QF = #[QPb, (DF Q] (3.14)

T(p, f°) = Gx @, #), Fe (x, #);x*). (2.5)

Gi. = FYHY, (Cjx")(F)', (2.20)

H?,(C;x') = KH, (K'C'K; K'(x — ¢))K' (2.21)

where we have used the general rule (AB)* = AB.