Much of the world around us, both natural and manmade, is built from and held
together by solid materials. Understanding how they behave is the task of solid
mechanics, which can in turn be applied to a wide range of areas from earthquake
mechanics and the construction industry to biomechanics. The variety of materials
(such as metals, rocks, glasses, sand, flesh and bone) and their properties (such as
porosity, viscosity, elasticity, plasticity) are reflected by the concepts and techniques
needed to understand them, which are a rich mixture of mathematics, physics, experiment and intuition. These are all brought to bear in this distinctive book, which
is based on years of experience in research and teaching. Theory is related to
practical applications, where surprising phenomena occur and where innovative
mathematical methods are needed to understand features such as fracture. Starting
from the very simplest situations, based on elementary observations in engineering and physics, models of increasing sophistication are derived and applied. The
emphasis is on problem solving and on building an intuitive understanding, rather
than on a technical presentation of theoretical topics. The text is complemented by
over 100 carefully chosen exercises, and the minimal prerequisites make it an ideal
companion for mathematics students taking advanced courses, for those undertaking research in the area or for those working in other disciplines in which solid
mechanics plays a crucial role.
GREGORY KOZYREFF
Fonds de la Recherche ScientifiqueFNRS
and Universite Libre de Bruxelles
JOHN OCKENDON
University of Oxford
ISBN13
9780511506390
eBook (EBL)
ISBN13
9780521854894
hardback
ISBN13
9780521671095
paperback
Contents
List of illustrations
Prologue
page viii
xiii
Modelling solids
1.1 Introduction
1.2 Hookes law
1.3 Lagrangian and Eulerian coordinates
1.4 Strain
1.5 Stress
1.6 Conservation of momentum
1.7 Linear elasticity
1.8 The incompressibility approximation
1.9 Energy
1.10 Boundary conditions and wellposedness
1.11 Coordinate systems
Exercises
1
1
2
3
4
7
10
11
13
14
16
19
24
Linear elastostatics
2.1 Introduction
2.2 Linear displacements
2.3 Antiplane strain
2.4 Torsion
2.5 Multiplyconnected domains
2.6 Plane strain
2.7 Compatibility
2.8 Generalised stress functions
2.9 Singular solutions in elastostatics
2.10 Concluding remark
Exercises
28
28
29
37
39
42
47
68
70
82
93
93
vi
Contents
Linear elastodynamics
3.1 Introduction
3.2 Normal modes and plane waves
3.3 Dynamic stress functions
3.4 Waves in cylinders and spheres
3.5 Initialvalue problems
3.6 Moving singularities
3.7 Concluding remarks
Exercises
103
103
104
121
124
132
138
143
143
Approximate theories
4.1 Introduction
4.2 Longitudinal displacement of a bar
4.3 Transverse displacements of a string
4.4 Transverse displacements of a beam
4.5 Linear rod theory
4.6 Linear plate theory
4.7 Von K
arman plate theory
4.8 Weakly curved shell theory
4.9 Nonlinear beam theory
4.10 Nonlinear rod theory
4.11 Geometrically nonlinear wave propagation
4.12 Concluding remarks
Exercises
150
150
151
152
153
158
162
172
177
187
195
198
204
205
Nonlinear elasticity
5.1 Introduction
5.2 Stress and strain revisited
5.3 The constitutive relation
5.4 Examples
5.5 Concluding remarks
Exercises
215
215
216
221
233
239
239
Asymptotic analysis
6.1 Introduction
6.2 Antiplane strain in a thin plate
6.3 The linear plate equation
6.4 Boundary conditions and SaintVenants principle
6.5 The von K
arman plate equations
6.6 The EulerBernoulli plate equations
6.7 The linear rod equations
6.8 Linear shell theory
245
245
246
248
253
261
267
273
278
Contents
vii
282
283
287
287
288
309
320
321
Plasticity
8.1 Introduction
8.2 Models for granular material
8.3 Dislocation theory
8.4 Perfect plasticity theory for metals
8.5 Kinematics
8.6 Conservation of momentum
8.7 Conservation of energy
8.8 The ow rule
8.9 Simultaneous elasticity and plasticity
8.10 Examples
8.11 Concluding remarks
Exercises
328
328
330
337
344
358
360
360
362
364
365
370
372
378
378
379
388
391
408
413
417
417
426
Appendix
References
Index
428
440
442
Illustrations
1.1
1.2
1.3
1.4
1.5
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15
2.16
3.1
3.2
3.3
3.4
3.5
3.6
3.7
A reference tetrahedron.
page 8
The forces acting on a small twodimensional element.
9
A small pillboxshaped region at the boundary between two elastic
solids.
18
Forces acting on a polar element of solid.
22
A system of masses connected by springs.
25
A unit cube undergoing (a) uniform expansion, (b) onedimensional
shear, (c) uniaxial stretching.
30
A uniform bar being stretched under a tensile force.
32
A paper model with negative Poissons ratio.
33
A strained plate.
34
A bar in a state of antiplane strain.
38
A twisted bar.
39
A uniform tubular torsion bar.
43
The crosssection of (a) a circular cylindrical tube; (b) a cut tube.
44
The unit normal and tangent to the boundary of a plane region.
49
A plane annulus being inated by an internal pressure.
53
A plane rectangular region subject to tangential tractions on its faces. 57
The tractions applied to the edge of a semiinnite strip.
59
The surface displacement of a halfspace and corresponding surface
pressure.
65
83
A family of functions (x) that approach a deltafunction as 0.
Contours of the maximum shear stress created by a point force acting
at the origin.
85
Four point forces.
91
Plots of the rst three Bessel functions.
108
A P wave reecting from a rigid boundary.
116
A layered elastic medium.
117
Dispersion relation for symmetric and antisymmetric Love waves.
120
Illustration of exural waves.
128
The onedimensional fundamental solution.
134
The twodimensional fundamental solution.
135
viii
List of illustrations
3.8
3.9
3.10
3.11
3.12
3.13
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12
4.13
4.14
4.15
4.16
4.17
4.18
4.19
4.20
4.21
4.22
4.23
4.24
4.25
4.26
4.27
ix
x
5.1
5.2
5.3
5.4
5.5
5.6
6.1
6.2
7.1
7.2
7.3
7.4
7.5
7.6
7.7
7.8
7.9
7.10
7.11
7.12
7.13
7.14
7.15
7.16
7.17
7.18
8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
8.9
8.10
8.11
8.12
List of illustrations
The deformation of a small scalene cylinder.
Typical forcestrain graphs for uniaxial tests on various materials.
A square membrane subject to an isotropic tensile force.
Response diagrams for a biaxiallyloaded incompressible sheet of
MooneyRivlin material.
Scaled pressure inside a balloon as a function of the stretch for various
values of the MooneyRivlin parameter.
Gas pressure inside a cavity as a function of ination coecient for
various values of the MooneyRivlin parameter.
The edge of a plate subject to tractions.
The geometry of a deformed twodimensional plate.
Denition sketch of a thin crack.
Denition sketch for contact between two solids.
(a) A Mode III crack. (b) A crosssection
in the (x, y)plane.
218
233
234
235
236
238
254
269
288
288
290
292
293
297
301
303
304
306
307
310
311
314
317
319
326
326
329
330
331
332
333
336
339
340
341
342
347
348
List of illustrations
8.13
8.14
xi
Prologue
Although solid mechanics is a vitally important branch of applied mechanics, it is often less popular, at least among students, than its close relative,
uid mechanics. Several reasons can be advanced for this disparity, such as
the prevalence of tensors in models for solids or the especial diculty of handling nonlinearity. Perhaps the most daunting prospect for the student is the
multitude of dierent behaviours that can occur and cause elementary theories of elasticity to become irrelevant in practice. Examples include fracture,
buckling and plasticity, and these pose intellectual challenges in solid mechanics that are every bit as fascinating as concepts like ight, shock waves
and turbulence in uid dynamics. Our principal objective in this book is to
demonstrate this fact to undergraduate and beginning graduate students.
We aim to give the subject as wide an accessibility as possible to mathematicallyminded students and to emphasise the interesting mathematical
issues that it raises. We do this by relating the theory to practical applications where surprising phenomena occur and where innovative mathematical
methods are needed.
Our layout is essentially pragmatic. Although more advanced texts in solid
mechanics often begin with quite general theories founded on basic mechanical and thermodynamic principles, we start from the very simplest models,
based on elementary observations in engineering and physics, and build our
way towards models that are the basis for current applied research in solid
mechanics. Hence, we begin by deriving the basic Navier equations of linear
elasticity, before illustrating the mathematical techniques that allow these
equations to be solved in many dierent practically relevant situations, both
static and dynamic. We then proceed to describe some approximate theories
for the elastic deformation of thin solids, namely bars, strings, beams, rods,
plates and shells. We soon discover that many everyday phenomena, such as
the buckling of a beam under a compressive load, cannot be fully described
xiii
xiv
Prologue
using linear theories. We therefore give a brief exposition of the general theory of nonlinear elasticity, and then show how formal asymptotic methods
allow simplied linear and weakly nonlinear models to be systematically deduced. Although we regard such asymptotic techniques as invaluable to any
applied mathematician, these last two topics may both be omitted on a rst
reading without loss of continuity. We go on to present simple models for
fracture and contact, comparing and contrasting these apparently similar
phenomena. Next, we show how plasticity theory can be used to describe
situations where a solid yields under a suciently high stress. Finally, we
show how elasticity theory may be generalised to include further physical effects, such as thermal stresses, viscoelasticity and porosity. These combined
elds of solid mechanics are increasingly nding applications in industrial
and medical processes, and pose ever more elaborate modelling questions.
Despite the breadth of the models and relevant techniques that will emerge
in this book, we will usually try to present the theoretical developments
ab initio. Nonetheless, the book is very far from being selfcontained. Any
student who aspires to becoming a solid mechanics specialist will have to
delve further into the literature, and we will provide references to help with
this.
We assume only that the reader has a reasonable familiarity with the
calculus of several variables. Fluency with the more advanced techniques
required for Chapters 6 and 7, in particular, will readily be acquired by
a student who works through the exercises in the early Chapters, especially those cited in the text. Indeed, we rmly believe that solid mechanics
provides a wonderful arena in which to build an understanding of such important mathematical areas as linear algebra, partial dierential equations,
complex variable theory, dierential geometry and the calculus of variations.
Our hope is that, having read this book, a student should be able to confront
any practical problem that may be encountered in everyday solid mechanics
with at least some idea of the basic mathematical modelling that will be
required.
During the writing of this book, we received a great deal of help and inspiration as a result of discussions with David Allwright, Jon Chapman, Sam
Howison, L. Mahadevan, Roman Novokshanov and Domingo Salazar, as well
as many other colleagues and students too numerous to thank individually.
We would like to express our particular gratitude to Gareth Jones, Hilary
Ockendon and Tom Witelski who gave invaluable advice on draft Chapters.
We are also indebted to David Tranah and his colleagues at Cambridge
University Press for helping to make this book a reality.
1
Modelling solids
1.1 Introduction
In everyday life we regularly encounter physical phenomena that apparently
vary continuously in space and time. Examples are the bending of a paper
clip, the ow of water or the propagation of sound or light waves. Such phenomena can be described mathematically, to lowest order, by a continuum
model, and this book will be concerned with that class of continuum models
that describes solids. Hence, at least to begin with, we will avoid all consideration of the atomistic structure of solids, even though these ideas lead
to great practical insight and also to some beautiful mathematics. When
we refer to a solid particle, we will be thinking of a very small region of
matter but one whose dimension is nonetheless much greater than an atomic
spacing.
For our purposes, the diagnostic feature of a solid is the way in which it
responds to an applied system of forces and moments. There is no hardandfast rule about this but, for most of this book, we will say that a continuum is
a solid when the response consists of displacements distributed through the
material. In other words, the material starts at some reference state, from
which it is displaced by a distance that depends on the applied forces. This is
in contrast with a uid, which has no special rest state and responds to forces
via a velocity distribution. Our modelling philosophy is straightforward. We
take the most fundamental pieces of experimental evidence, for example
Hookes law, and use mathematical ideas to combine this evidence with
the basic laws of mechanics to construct a model that describes the elastic
deformation of a continuous solid. Following this simple approach, we will
nd that we can construct solid mechanics theories for phenomena as diverse
as earthquakes, ultrasonic testing and the buckling of railway tracks.
1
Modelling solids
Hookes observation is exemplied by a simple highschool physics experiment in which a tensile force T is applied to a spring whose natural length
is L. Hookes law states that the resulting extension of the spring is proportional to T : if the new length of the spring is , then
T = k( L),
(1.2.1)
Hooke devised his law while designing clock springs, but noted that it
appears to apply to all springy bodies whatsoever, whether metal, wood,
stones, baked earths, hair, horns, silk, bones, sinews, glass and the like. In
practice, it is commonly observed that k scales with 1/L; that is, everything
else being equal, a sample that is initially twice as long will stretch twice
as far under the same force. It is therefore sensible to write (1.2.1) in the
form
L
,
(1.2.2)
T = k
L
where k is the elastic modulus of the spring, which will be dened more
rigorously in Chapter 2. The dimensionless quantity ( L)/L, measuring
the extension relative to the initial length, is called the strain.
Equation (1.2.2) is the simplest example of the allimportant constitutive
law relating the force to displacement. As shown in Exercise 1.3, it is possible
to construct a onedimensional continuum model for an elastic solid from
this law, but, to generalise it to a threedimensional continuum, we rst need
to generalise the concepts of strain and tension.
(1.3.1)
Many basic problems in solid mechanics amount to determining the displacement eld u corresponding to a given system of applied forces.
The mathematical consequence of our statement that the solid is a continuum is that there must be a smooth onetoone relationship between X
and x, i.e. between any particles initial position and its current position.
Modelling solids
This will be the case provided the Jacobian of the transformation from X
to x is bounded away from zero:
xi
.
(1.3.2)
0 < J < ,
where J = det
Xj
The physical signicance of J is that it measures the change in a small
volume compared with its initial volume:
dx1 dx2 dx3 = J dX1 dX2 dX3 ,
or
dx = J dX
(1.3.3)
V (0)
Since V (t) designates a xed set of material points, M (t) must be a constant,
namely its initial value M (0):
J dX = M (t) = M (0) =
0 dX,
(1.3.5)
V (0)
V (0)
where 0 is the density in the rest state. Since V is arbitrary, we deduce that
J = 0 .
(1.3.6)
Hence, we can calculate the density at any time t in terms of 0 and the
displacement eld. The initial density 0 is usually taken as constant, but
(1.3.6) also applies if 0 = 0 (X).
1.4 Strain
To generalise the concept of strain introduced in Section 1.2, we consider the
deformation of a small line segment joining two neighbouring particles with
initial positions X and X + X. At some later time, the solid deforms such
that the particles are displaced to X + u(X, t) and X + X + u(X + X, t)
respectively. Thus we can use Taylors theorem to show that the line element
1.4 Strain
+ X2
+ X3
.
X1
X2
X3
(1.4.2)
Let L = X and = x denote the initial and current lengths respectively
of the line segment; the dierence L is known as the stretch. Then, to
lowest order in L,
2 = X + (X )u(X, t)2 .
(1.4.3)
where
1
Eij =
2
uk uk
uj
ui
+
+
Xj
Xi
Xi Xj
3
.
(1.4.5)
k=1
(1.4.7)
It is clear from (1.4.4) that the stretch is measured by the quantities Eij ;
in particular, the stretch is zero for all line elements if and only if Eij 0. It
is thus natural to identify Eij with the strain. Now let us ask: what happens
when we perform the same calculation in a coordinate system rotated by an
Modelling solids
u = P u.
(1.4.8)
Since P is orthogonal, (1.4.8) may be inverted to give X = P T X . Alternatively, using sux notation, we have
X = pj Xj ,
ui = pi u .
(1.4.9)
(1.4.10)
which may be manipulated using the chain rule, as shown in Exercise 1.4,
to give
Eij = pi pj E .
(1.4.11)
(1.4.12)
so the 3 3 symmetric array (Eij ) transforms exactly like a matrix representing a linear transformation of the vector space R3 . Arrays that obey the
transformation law (1.4.11) are called secondrank Cartesian tensors, and
E = (Eij ) is therefore called the strain tensor.
Almost as important as the fact that E is a tensor is the fact that it
can vanish without u vanishing. More precisely, if we consider a rigidbody
translation and rotation
u = c + (Q I)X,
(1.4.13)
where I is the identity matrix while the vector c and orthogonal matrix
Q are constant, then E is identically zero. This result follows directly from
substituting (1.4.13) into (1.4.6) and using the fact that QQT = I, and
conrms our intuition that a rigidbody motion induces no deformation.
The word tensor as used here is eectively synonymous with matrix, but it is easy to
generalise (1.4.11) to a tensor with any number of indices. A vector, for example, is a tensor
with just one index.
1.5 Stress
1.5 Stress
In the absence of any volumetric (e.g. gravitational or electromagnetic) effects, a force can only be transmitted to a solid by being applied to its
boundary. It is, therefore, natural to consider the force per unit area or
stress applied at that boundary. To do so, we now analyse an innitesimal
surface element, whose area and unit normal are da and n respectively. If it
is contained within a stressed medium, then the material on (say) the side
into which n points will exert a force df on the element. (By Newtons third
law, the material on the other side will also exert a force equal to df .) In
the expectation that the force should be proportional to the area da, we
write
df = da,
(1.5.1)
(1.5.2)
This expression implies that (i) the stress acts only in a direction normal
to the surface element, (ii) the magnitude of the stress (i.e. p) is independent of the direction of n. In an elastic solid, neither of these simplifying
assumptions holds; we must allow for stress which acts in both tangential
and normal directions and whose magnitude depends on the orientation of
the surface element.
First consider a surface element whose normal points in the x1 direction,
and denote the stress acting on such an element by 1 = (11 , 21 , 31 )T . By
doing the same for elements with normals in the x2  and x3 directions, we
generate three vectors j (j = 1, 2, 3), each representing the stress acting
on an element normal to the xj direction. In total, therefore, we obtain nine
scalars ij (i, j = 1, 2, 3), where ij is the icomponent of j , that is
j = ij ei ,
(1.5.3)
Modelling solids
x3
a1
n
a2
x2
a3
x1
Fig. 1.1 A reference tetrahedron; ai is the area of the face orthogonal to the xi axis.
The outward normal to the face with area a1 is in the negative x1 direction
and the force on this face is thus a1 1 . Similar expressions hold for the
faces with areas a2 and a3 . Hence, if the stress on the fourth face is denoted
by , then the total force on the tetrahedron is
f = a aj j .
(1.5.4)
(1.5.5)
Now we shrink the tetrahedron to zero volume. Since the area a scales
with 2 , where is a typical edge length, while the volume is proportional
to 3 , if we apply Newtons second law and insist that the acceleration be
nite, we see that f /a must tend to zero as 0. Hence we deduce an
Readers of a sensitive disposition may be slightly perturbed by our glibly letting the dimensional
variable tend to zero: if is reduced indenitely then we will eventually reach an atomic scale
on which the solid can no longer be treated as a continuum. We reassure such readers that
(1.5.6) can be more rigorously justied provided the macroscopic dimensions of the solid are
large compared to any atomistic lengthscale.
1.5 Stress
22
12
21
11
x2
11
x2
x1
21
12
x1
22
expression for :
i = ij nj ,
or = n.
(1.5.6)
(1.5.7)
n = P n.
(1.5.8)
or ij = pi pj .
(1.5.9)
Thus ij , like Eij , is a secondrank tensor, called the Cauchy stress tensor.
We can make one further observation about ij by considering the angular
momentum of the small twodimensional solid element shown in Figure 1.2.
The net anticlockwise moment acting about the centre of mass G is (per
unit length in the x3 direction)
2 (21 x2 )
x1
x2
2 (12 x1 )
,
2
2
10
Modelling solids
(1.5.10)
ij
2 ui
= gi +
.
2
t
xj
(1.6.3)
11
12
Modelling solids
(1.7.2)
(1.7.3)
(1.7.4)
for all orthogonal matrices P = (pij ). It can be shown (see, for example,
Ockendon & Ockendon, 1995, pp. 79) that this is sucient to reduce the
specication of Cijk to just two scalar quantities and , such that
Cijk = ij k + 2ik j ,
(1.7.5)
where ij is the usual Kronecker delta, which represents the identity matrix;
consequently,
ij = (ekk ) ij + 2eij .
(1.7.6)
This relation can also be inverted to give the strain corresponding to a given
stress, that is
1
(kk )
(1.7.7)
eij =
ij
ij .
2
(3 + 2)
In Chapter 9 we will consider solids, such as wood or brereinforced materials, that are not isotropic, and for which (1.7.6) must be generalised.
The material parameters and are known as the Lame constants, and
is called the shear modulus.As we shall see in Chapter 2, and measure a
materials ability to resist elastic deformation. They have the units of pressure; typical values for a few familiar solid materials are given in Table 1.1.
It will be observed that these values may be very large for relatively hard
Cartilage
Rubber
Polystyrene
Granite
Glass
Copper
Steel
Diamond
(GPa)
(GPa)
3 105
0.04
2.3
10
28
86
100
270
9 105
0.003
1.2
30
28
37
78
400
13
Table 1.1 Typical values of the Lame constants and for some everyday
materials (1 GPa = 109 N m2 = 104 atmospheres; a typical car tyre
pressure is two atmospheres).
materials, the signicance being that tractions much less than these values
will result in small deformations, so that linear elasticity is valid.
Now we substitute our linear constitutive relation (1.7.6) into the momentum equation (1.6.3) and replace X with x to obtain the Navier equation,
also known as the Lame equation,
2u
= g + ( + ) grad div u + 2 u.
(1.7.8)
t2
Recall that does not vary to leading order, so (1.7.8) comprises three
equations for the three components of u. It may alternatively be written in
component form
2 uj
2 ui
2 ui
=
g
+
(
+
)
+
,
i
t2
xi xj
x2j
(1.7.9)
(1.7.10)
(1.7.11)
14
Modelling solids
The values given in Table 1.1 show that rubber has this property, as do many
biomaterials such as muscle.
If a material is almost incompressible, we can set
1
= ,
(1.8.1)
(1.8.2)
p = div u,
div u = 0.
(1.8.3a)
(1.8.3b)
The condition (1.8.3b) means that each material volume is conserved during
the deformation, and it imposes an extra constraint on the Navier equation.
The extra unknown p, representing the isotropic pressure in the medium,
gives us the extra freedom we need to satisfy this constraint.
1.9 Energy
We can obtain an energy equation from (1.6.3) by taking the dot product
with u/t and integrating over an arbitrary volume V :
ij ui
ui
2 ui ui
dx =
dx +
dx.
(1.9.1)
2
gi
t t
t
V
V
V xj t
The nal term may be rearranged, using the divergence theorem, to
ij ui
eij
ui
ij
dx =
ij nj da
dx.
(1.9.2)
t
V xj t
V t
V
1.9 Energy
15
V
1 u 2
dx +
W dx
2 t
V
ui
ui
dx +
ij nj da, (1.9.3)
gi
=
t
V
V t
(1.9.4)
(1.9.5)
Here the summation convention is invoked such that (ekk )2 is the square of
the trace of E, while (eij eij ) is the sum of the squares of the components
of E.
The rst term in braces in (1.9.3) is the net kinetic energy in V , while the
terms on the righthand side represent the rate of working of the external
body force g and the tractions on V respectively. Hence, in the absence of
other energy sources resulting from, say, chemical or thermal eects, we can
interpret equation (1.9.3) as a statement of conservation of energy. The difference between the rate of working and the rate of change of kinetic energy
is the rate at which elastic energy is stored in the material as it deforms; W
is therefore called the strain energy density. This is analogous to the energy
stored in a stretched spring (see Exercise 1.1) and, at a fundamental scale,
is a manifestation of the energy stored in the bonds between the atoms. If
, > 0, we can easily see from (1.9.5) that W is a nonnegative function
of the strain components, whose unique global minimum is attained when
eij = 0. In fact, Exercise 1.7 demonstrates that it is only necessary to have
, ( + 2/3) > 0.
The net conservation of energy implied by (1.9.3) reects the fact that the
Navier equation is not dissipative. Furthermore, even without the constitutive relation (1.7.6), the steady Navier equation is a necessary condition for
the net gravitational and strain energy in an elastic body D, namely
U=
W g u dx,
(1.9.6)
D
16
Modelling solids
on D.
(1.10.1)
(1.10.2)
and we will now show that, if a solution u of (1.7.6), (1.10.2) with the
boundary condition (1.10.1) exists, then it is unique.
Suppose that two solutions u(1) and u(2) exist and let u = u(1) u(2) . Thus
u satises the homogeneous problem, with ub = g = 0. Now, by multiplying
(1.10.2) by ui , integrating over D and using the divergence theorem, we
obtain
ui ij nj da =
eij ij dx = 2
W dx,
(1.10.3)
D
17
is identically zero in D, and the displacement can therefore only be a rigidbody motion (i.e. a uniform translation and rotation; see Exercise 1.6). Since
u is zero on D, we deduce that it must be zero everywhere and, hence, that
u(1) u(2) .
Now we attempt the same calculation when the surface traction, rather
than the displacement, is specied:
n = (x)
on D.
(1.10.4)
Like the Neumann problem for a scalar elliptic partial dierential equation
(Ockendon et al., 2003, p. 154), the Navier equation only admits solutions
satisfying (1.10.4) if socalled solvability conditions are satised. If we integrate (1.10.2) over D and use the divergence theorem, we nd that
ij nj da +
gi dx = 0
(1.10.5)
D
da +
g dx = 0.
(1.10.6)
This represents a net balance between the forces, namely surface traction
and gravity, acting on D. An analogous balance between the moments acting
on D may also be obtained by taking the cross product of x with (1.10.2)
before integrating, to give
x da +
xg dx = 0,
(1.10.7)
D
18
Modelling solids
n
solid 2
solid 1
Fig. 1.3 A small pillboxshaped region at the boundary between two elastic solids.
and the displacement on others, for example in contact problems and in fracture, as described in Chapter 7. Another common generalisation of (1.10.1)
and (1.10.4) occurs when two solids with dierent elastic moduli are bonded
together across a common boundary D, as shown in Figure 1.3. Then the
displacement vectors are the same on either side of D and, by balancing
the stresses on the small pillboxshaped region shown in Figure 1.3, we see
that
(1) n = (2) n,
(1.10.8)
where (1) and (2) are the values of on either side of the boundary. Thus
there are six continuity conditions across such a boundary.
On the other hand, if two unbonded solids are in smooth contact, only
the normal displacement is continuous across D. However, this loss of information is compensated by the fact that the four tangential components
of (1) n and (2) n are zero and the normal components of these tractions
are continuous. Frictional contact between rough unbonded surfaces poses
serious modelling challenges, as we will see in Chapter 7.
For elastodynamic problems, we may anticipate that (1.7.8) admits wavelike solutions. It may, therefore, be viewed as a generalisation of a scalar
wave equation, such as the familiar equation
2w
2w
=
T
t2
x2
(1.10.9)
which describes small transverse waves on a string with tension T and line
density (see Section 4.3). We will examine elastic waves in more detail
in Chapter 3 but, in the meantime, we expect to prescribe Cauchy initial
conditions for u and u/t at t = 0, as well as elliptic boundary conditions
such as (1.10.1) or (1.10.4).
19
xy = 2exy ,
yy
xz = 2exz ,
zz
yz = 2eyz ,
(1.11.1)
where
u
,
x
v
,
eyy =
y
w
ezz =
,
z
exx =
u v
+
,
y
x
v w
2eyz =
+
,
z
x
u w
2exz =
+
,
z
x
2exy =
(1.11.2)
20
Modelling solids
(1.11.3)
2u
( u) + 2 u,
= gx + ( + )
2
t
x
2v
2 = gy + ( + ) ( u) + 2 v,
t
y
2
w
2 = gz + ( + ) ( u) + 2 w.
t
z
(1.11.4)
r = 2er ,
rz = 2erz ,
zz = err + e + ( + 2)ezz ,
z = 2ez ,
(1.11.5)
u
u
1 ur
+
,
r
r
r
uz
ur
+
,
=
z
r
1 uz
u
+
.
(1.11.6)
=
z
r
err =
2er =
2erz
ezz
2ez
21
1
1 r
(rrr ) +
+
,
= gr +
2
t
r r
r
z
r
2 u
z
r
1
1
2 = g +
(rr ) +
+
+
,
t
r r
r
z
r
2 uz
zz
1
1 z
2 = gz +
(rrz ) +
+
,
t
r r
r
z
(1.11.7)
ur
2 u
2
2 = gr + ( + ) ( u) + ur 2 2
,
t
r
r
r
2 u
( + )
u
2 ur
2
( u) + u 2 + 2
,
2 = g +
t
r
r
r
2 uz
2 = gz + ( + ) ( u) + 2 uz ,
(1.11.8)
t
z
where
1 u
uz
1
(rur ) +
+
,
r r
r
z
1
ui
1 2 ui 2 ui
2 u i =
+
r
+ 2
r r
r
r 2
z 2
u =
(1.11.9)
+ ,
r
+ ,
+
= (r + r, ) = + r
= (r, + ) =
(1.11.10)
(1.11.11)
22
Modelling solids
rr
rr
r
er
r sin cos
r(r, , ) = r sin sin .
(1.11.13)
r cos
Again, we can apply the constitutive relation (1.7.6) literally, to obtain
rr = ( + 2)err + e + e ,
r = 2er ,
= err + ( + 2)e + e ,
r = 2er ,
= err + e + ( + 2)e ,
= 2e .
(1.11.14)
23
,
=
r
r sin
r
r
u cot
ur
u cot
1 u
1 u
1 u
+
+
, 2e =
+
.
=
r sin
r
r
r sin
r
r
(1.11.15)
err =
e
e
+
1 r
,
+
r sin
r
1 (sin )
2 u
1 (r2 r )
+
2 = g + 2
t
r
r
r sin
r cot
1
+
,
+
r sin
r
2 u
1 (sin )
1 (r2 r )
+
2 = g + 2
t
r
r
r sin
r + cot
1
+
,
+
r sin
r
(1.11.16)
2 ur
= gr + ( + ) ( u)
2
t
r
2ur
2
2 u
2
(u sin ) 2
,
+ ur 2 2
r
r sin
r sin
( + )
2 u
( u)
= g +
t2
r
u
2 ur
2 cos u
2
2 2 2 2
,
+ u + 2
r
r sin r sin
24
Modelling solids
2u
t2
( + )
( u)
= g +
r sin
u
2 cos u
2 ur
2
+ u + 2
,
+ 2 2
2 2
r sin
r sin
r sin
(1.11.17)
where
1 u
1 2
1
(sin u ) +
,
r ur +
2
r r
r sin
r sin
1
ui
1
ui
1
2 ui
2 u i = 2
.
r2
+ 2
sin
+ 2 2
r r
r
r sin
r sin 2
(1.11.18)
u=
Exercises
1.1
1.2
1.3
d
dt
2
d2
+ k( L) mg = 0.
dt2
1
+ k( L)2 + mg(L ) = const.
2
b
w w b
1 w 2
d 1 b w 2
dx +
T dx = T
.
dt 2 a
t
x
x t a
a 2
Interpret this result in terms of conservation of energy.
A system of masses m along the xaxis at positions Xn = nL
(n = 0, 1, 2, . . .) are linked by springs satisfying Hookes law (1.2.1),
as shown in Figure 1.5. If each mass is displaced by a distance un (t),
show that the tension Tn joining Xn to Xn+1 satises
Tn Tn1 = m
d2 un
,
dt2
where
Tn = k (un+1 un ).
Exercises
Tn1
Tn1
Tn
25
Tn
un1
un
m
un+1
Deduce that
m
d2 un
= k (un+1 2un + un1 )
dt2
1.4
1.5
1.6
26
1.7
1.8
Modelling solids
By writing the linearised strain tensor eij as the sum of a zerotrace contribution eij (1/3)ij ekk and a purely diagonal contribution (1/3)ij ekk , show that W can be rewritten as
1
1
2
eij ij ekk .
+
(ekk ) + eij ij ekk
W=
2
3
3
3
Deduce that, for W to have a single global minimum at eij = 0, it is
sucient for and +2/3 to be positive. By considering particular
values of eij , show that it is also necessary.
Suppose a solid body occupies the region D and the displacement u
is prescribed on D. Let
W g u dx,
U=
D
1.9
1.10
and deduce that the minimisation of U with respect to all displacements satisfying the given boundary condition leads to the steady
Navier equation. Deduce also that, if no boundary condition is imposed, the natural boundary condition is the vanishing of the traction
ij nj on D.
Show also that the minimisation of U subject to the constraint
div u 0 leads to the steady incompressible Navier equation (1.8.3a),
where p is a Lagrange multiplier.
An elastic body D at rest is subject to a traction n = (x) on its
boundary D. By taking the cross product of x with (1.10.2) before
integrating over D, derive (1.10.7) and deduce that the net moment
acting on D must be zero.
Suppose that u satises the steady Navier equation (1.10.2) in a
region D and the mixed boundary condition (x)u+(x) n = f (x)
on D. Show that, if a solution exists, it is unique provided > 0
and 0.
Exercises
1.11
27
u
u
r +
+ ur e .
+
r
(r, r)
err
er
er
e
r
r
u
u
ur
1 ur
1
, 2er =
+
, e =
=
r
r
r
r
r
u
+ ur .
2
Linear elastostatics
2.1 Introduction
This chapter concerns steady state problems in linear elasticity. This topic
may appear to be the simplest in the whole of solid mechanics, but we will
nd that it oers many interesting mathematical challenges. Moreover, the
material presented in this chapter will provide crucial underpinning to the
more general theories of later chapters.
We will begin by listing some very simple explicit solutions which give
valuable intuition concerning the role of the elastic moduli introduced in
Chapter 1. Our rst application of practical importance is elastic torsion,
which concerns the twisting of an elastic bar. This leads to a class of exact
solutions of the Navier equation in terms of solutions of Laplaces equation
in two dimensions. However, as distinct from the use of Laplaces equation
in, say, hydrodynamics or electromagnetism, the dependent variable is the
displacement, which has a direct physical interpretation, rather than a potential, which does not. This means we have to be especially careful to ensure
that the solution is singlevalued in situations involving multiplyconnected
bars.
These remarks remain important when we move on to another class of
twodimensional problems called plane strain problems. These have even
more general practical relevance but involve the biharmonic equation. This
equation, which will be seen to be ubiquitous in linear elastostatics, poses
signicant extra diculties as compared to Laplaces equation. In particular,
we will nd that it is much more dicult to construct explicit solutions using,
for example, the method of separation of variables.
An interesting technique to emerge from both these classes of problems is
the use of stress potentials, which are the elastic analogues of electrostatic or
gravitational potentials, say, or the stream function in hydrodynamics. We
28
29
will nd that a large class of elastostatic problems with some symmetry, for
example twodimensional or axisymmetric, can be described using a single
scalar potential that satises the biharmonic equation.
Fully threedimensional problems are mostly too dicult to be suitable for
this chapter. Nonetheless, we will be able to provide a conceptual framework
within which to represent the solution by generalising the idea of Greens
functions for scalar ordinary and partial dierential equations. The necessary
Greens tensor describes the response of an elastic body to a localised point
force applied at some arbitrary position in the body. This idea opens up one
of the most distinctive and fascinating aspects of linear elasticity: because
of the intricacy of (1.7.8), many dierent kinds of singular solutions can
be constructed using stress functions and Greens tensors, each being the
response to a dierent kind of localised forcing, and the catalogue of these
dierent responses is a very helpful toolkit for thinking about solid mechanics
more generally.
(2.2.1)
This vector partial dierential equation for u is the starting point for all
we will say in this chapter. As discussed in Section 1.10, it needs to be
supplemented with suitable boundary conditions, which will vary depending
on the situation being modelled.
If the displacement u is a linear function of position x, then the strain
tensor E is spatially uniform. It follows that the stress tensor is also uniform
and, therefore, trivially satises (2.2.1). Such solutions provide considerable
insight into the predictions of (2.2.1) and also give a feel for the signicance
of the parameters and . To avoid suces, we will write u = (u, v, w)T
and x = (x, y, z)T .
x,
3
(2.2.2)
30
Linear elastostatics
1
1
z
(a)
(b)
(c)
1 + /3
1 + /3
1 + /3
1+
1
1
Fig. 2.1 A unit cube undergoing (a) uniform expansion (2.2.2), (b) onedimensional
shear (2.2.6), (c) uniaxial stretching (2.2.8).
2
(2.2.4)
eij = ij and ij = + ij .
3
3
This is a socalled hydrostatic situation, in which the stress is characterised
by a scalar isotropic pressure p, and ij = pij . The pressure is related to
the relative volume change by p = K, where
2
K =+
3
(2.2.5)
31
u = v = 0 ,
w
0
(2.2.6)
0 1 0
0 1 0
E=
= 1 0 0 .
(2.2.7)
1 0 0 ,
2
0 0 0
0 0 0
Note that does not aect the stress, so the response to shear is accounted
for entirely by , which is therefore called the shear modulus.
x
1 0
0
u = y ,
E = 0 0 ,
(2.2.8)
z
0 0
(1 2) + 2
0
0
.
=
0
(1 2) 2
0
0
0
(1 2) 2
(2.2.9)
This simple solution may be used to describe a uniform elastic bar that is
stretched in the xdirection under a tensile force T , as shown in Figure 2.2.
Notice that, since the bar is assumed not to vary in the xdirection, the
outward normal n to the lateral boundary always lies in the (y, z)plane.
If the curved surface of the bar is stressfree, then the resulting boundary
condition n = 0 may be satised identically by ensuring that yy = zz = 0,
which occurs if
.
(2.2.10)
=
2 ( + )
32
Linear elastostatics
y
n = 0
T
z
Fig. 2.2 A uniform bar being stretched under a tensile force T .
Hence the bar, while stretching by a factor in the xdirection, must shrink
by a factor in the two transverse directions; if happened to be negative, this would correspond to an expansion. The ratio between lateral
contraction and longitudinal extension is called Poissons ratio.
With given by (2.2.10), the stress tensor has just one nonzero element,
namely
xx = E,
(2.2.11)
where
E=
(3 + 2)
+
(2.2.12)
is called Youngs modulus. If the crosssection of the bar has area A, then
the tensile force T applied to the bar is related to the stress by
T = Axx = AE;
(2.2.13)
E
,
(1 + )(1 2)
E
.
2(1 + )
(2.2.14)
(2.2.15)
While E is a positive constant with the dimensions of pressure, is dimensionless and constrained on physical grounds to lie in the range
1 < < 1/2. The lower bound for comes from the condition +2/3 > 0
33
(b)
(a)
Fig. 2.3 A paper model with negative Poissons ratio. Each line segment is a strip
of paper viewed endon.
required for convexity of the strain energy, as shown in Exercise 1.7. The upper bound 1/2 is approached as , in other words as the material
becomes incompressible, as discussed in Section 1.8.
For most solids, is positive, but it is possible (see Figure 2.3) to construct
simple hinged paper models with negative Poissons ratio; try extending a
crumpled piece of paper! Socalled auxetic materials, in which < 0, have
been developed whose microscopic structure mimics such paper models so
they also display negative values of and they expand in all directions when
pulled in only one (see Lakes, 1987).
u
ax + by
u = v = cx + dy ,
w
z
(2.2.16)
and, as in Section 2.2.3, the stress and strain tensors are both constant. Here
we choose to satisfy the condition zz = 0 on the tractionfree upper and
34
Linear elastostatics
Tyy
Tyx
Txy
Txx
z
Txx
Txy
Tyx
Tyy
Fig. 2.4 A plate being strained under tensions Txx , Ty y and shear forces Txy , Ty x .
(a + d) =
(a + d),
=
+ 2
1
(2.2.17)
where again denotes Poissons ratio. With this choice, and with E again
denoting the Youngs modulus, the only nonzero stress components are
xx =
E(a + d)
,
1 2
xy =
E(b + c)
,
2(1 + )
yy =
E(a + d)
.
1 2
(2.2.18)
We denote the net inplane tensions and shear stresses applied to the plate
by Tij = hij , as illustrated in Figure 2.4. We can use (2.2.18) to relate these
to the inplane strain components by
Eh
(exx + eyy ),
1 2
Eh
exy ,
=
1+
Eh
=
(exx + eyy ).
1 2
Txx =
Txy = Tyx
Tyy
(2.2.19a)
(2.2.19b)
(2.2.19c)
These will provide useful evidence when constructing more general models
for the deformation of plates in Chapter 4.
If no force is applied in the ydirection, that is Txy = Tyy = 0, then (2.2.19)
reproduces the results of uniaxial stretching, with d = a and Txx = Eha.
On the other hand, it is possible for the displacement to be purely in the
(x, z)plane, with
b = c = d = 0,
yy =
Ea
,
1 2
xx =
Ea
.
1 2
(2.2.20)
35
xi
2 xj
xi
2 xi
xj
We recognise the symmetric part of u as the linear strain tensor E, while
the skewsymmetric part may be written in the form
0
3 2
1
u uT = 3
(2.2.23)
0
1 ,
2
2 1
0
so that, with = (1 , 2 , 3 )T , (2.2.21) becomes
u(x) = u0 + (x x0 ) + E(x x0 ) + .
(2.2.24)
(2.2.25)
(2.2.26)
36
Linear elastostatics
where
E = P EP T
1 0 0
= 0 2 0 .
0 0 3
(2.2.27)
We can hence think of the strain at any point as comprising three orthogonal expansions or contractions, depending on the signs of {1 , 2 , 3 }, followed
by a small rigidbody rotation. Generalising Section 2.2.1, we note that the
net relative volume change associated with this expansion/contraction is
(1 + 1 )(1 + 2 )(1 + 3 ) 1 = 1 + 2 + 3 + .
(2.2.28)
The sum of the eigenvalues is the invariant trace of the matrix E so that the
relative volume change is
Tr (E) = ekk = u.
(2.2.29)
With respect to the principal axes, the stress tensor is also diagonal, namely
1 0 0
1 = (1 + 2 + 3 ) + 21 ,
= 0 2 0 , where
= (1 + 2 + 3 ) + 22 ,
2
(2.2.30)
0 0 3
3 = (1 + 2 + 3 ) + 23 ,
and the eigenvalues {1 , 2 , 3 } are called the principal stresses.
As an illustration, let us reexamine the simple shear of Section 2.2.2. The
eigenvalues and corresponding eigenvectors v i of E are readily found to be
,
2
1
v 1 = 1 ,
0
1 =
2 = ,
2
1
v 2 = 1 ,
0
3 = 0,
0
v 3 = 0 .
1
(2.2.31a)
(2.2.31b)
Hence the shear is equivalent to an expansion of magnitude /2 in the direction (1, 1, 0)T plus an equal and opposite contraction in the direction
(1, 1, 0)T . In the original nonprincipal axes, we can decompose the displacement as follows:
y
x+y
xy
y
(2.2.32)
u = 0 = x + x + y y x .
2
4
4
0
0
0
0
The terms on the righthand side represent a rigidbody rotation, an expansion and an orthogonal contraction.
37
In general, the strain tensor E and, therefore, the principal strains i and
principal stresses i are functions of position x. At any such position, the
strain and stress tensors are diagonal only with respect to the principal axes;
if these axes are rotated by another orthogonal matrix Q, then the strain
and stress tensors become E and respectively, where
E = QE QT ,
= Q QT .
(2.2.33)
It is easiest to understand the consequence of such a rotation in two dimensions, when we can take
cos sin
(2.2.34)
Q=
sin cos
and (2.2.33b) becomes
1 cos2 + 2 sin2 (1 2 ) sin cos
,
=
(1 2 ) sin cos 2 cos2 + 1 sin2
(2.2.35)
i j 
.
2
(2.2.36)
This is often used as a diagnostic to test for failure, since many elastic
materials, particularly metals, are susceptible to plastic deformation when
subject to excessive shear stress. The socalled Tresca criterion proposes
that a material will yield and become plastic when S exceeds some critical
yield stress Y. We will show below in Section 2.6.5 how this criterion can
be applied to predict the failure of a gun barrel, and a fuller account of
plasticity will be given in Chapter 8.
38
Linear elastostatics
surface
traction
D
x
Fig. 2.5 A bar in a state of antiplane strain.
= 0
0
0
w
y
w
x
w
.
(2.3.1)
At any point of the bar, the traction on an element normal to (cos , sin , 0)T
is (cos w/x + sin w/y), whose maximum value as varies is w.
If we neglect gravity, the Navier equation reduces to the twodimensional
Laplaces equation
2w 2w
+
= 2 w = 0,
x2
y 2
(2.3.2)
w
n
on D.
(2.3.3)
2.4 Torsion
39
D
x
Fig. 2.6 A bar being twisted under a moment M .
on D
(2.3.5)
u
yz
u = v = xz ,
(2.4.1)
w
(x, y)
40
Linear elastostatics
where is a constant representing the twist of the bar about its axis. As in
Section 2.3, the stress tensor is of the form
0
0 xz
= 0
(2.4.2)
0 yz
xz yz 0
where, now,
xz =
y ,
x
yz =
+x .
y
(2.4.3)
in D,
(2.4.4)
1 d 2
=
X +Y2
n
2 ds
on D.
(2.4.5)
(2.4.6)
Notice that the solvability condition for the Neumann problem (2.4.4),
(2.4.6) is satised identically. Hence, given the shape of the crosssection
D, the displacement w = is uniquely determined up to an arbitrary
constant, corresponding to an arbitrary uniform translation. Once has
been found, the moment applied at each end of the bar is given by
(xyz yxz ) dxdy = R,
(2.4.7)
M=
D
say, where
2
2
y
+ x +y
R=
x
dxdy.
y
x
D
(2.4.8)
2.4 Torsion
41
The boundaryvalue problem (2.4.4), (2.4.6) for may be usefully reformulated as follows. Whenever the stress tensor is of the form (2.4.2) and
there is no body force, the steady Navier equation reduces to
yz
xz
+
= 0.
x
y
(2.4.9)
,
y
yz =
,
x
(2.4.10)
=
+ y,
x
y
=
x.
y
x
(2.4.11)
Elimination of retrieves (2.4.4), while elimination of reveals that satises Poissons equation
2 = 2
in D.
(2.4.12)
Indeed, it is easy to see from (2.4.11) that and +(x2 +y 2 )/2 are harmonic
conjugates.
The zerostress boundary condition (2.4.5) implies that is constant on
D and, without loss of generality, we may take
=0
on D.
(2.4.13)
y
dxdy,
(2.4.14)
x
dxdy = 2
M =
x
y
D
D
42
Linear elastostatics
= 0,
r = a.
(2.4.16b)
r
It follows that is a constant and, hence, that the torsional rigidity is
2 a
2
a4
2
.
(2.4.17)
R=
r3 drd =
x + y dxdy =
2
D
0
0
Note that a circular cylinder is the only case in which the righthand side of
(2.4.6) is zero. For any other crosssectional shape, twisting a bar inevitably
results in a nonzero axial displacement w = .
Alternatively, we may solve the Dirichlet problem (2.4.12), (2.4.13) for .
Assuming that is a function only of r, we have
d
1 d
2
r
= 2,
r < a,
(2.4.18a)
=
r dr
dr
= 0,
r = a,
(2.4.18b)
a2 r2
.
2
(2.4.19)
(2.4.20)
43
Do
D Di
z
y
x
Fig. 2.7 A uniform tubular torsion bar, with inner and outer free surfaces given by
Di and Do respectively.
on Do ,
=k
on
Di ,
(2.5.1)
+y
dxdy
x
R =
x
y
D
=
(x) +
(y) 2 dxdy
x
y
D
= 2
dxdy + 2kAi ,
(2.5.2)
D
after using Greens theorem, where Ai is the area of the hole inside Di .
To obtain and thus R uniquely, we still have to evaluate the unknown
constant k.
By working with the stress function , we have reduced the torsional
rigidity problem to a seemingly innocuous Poisson equation (2.4.12) with
44
Linear elastostatics
y
(b)
(a)
b
Fig. 2.8 The crosssection of (a) a circular cylindrical tube; (b) a cut tube.
Dirichlet boundary conditions (2.5.1) on the two boundaries of the annular tube. However, we have temporarily lost sight of the displacement eld
w = , which has yet to be determined from (2.4.11). To be physically
acceptable, must be a singlevalued function of (x, y), which implies that
dx +
dy 0
(2.5.3)
y
C x
for all simple closed paths C contained in D. In fact, it is readily shown that
(2.5.3) holds for all such paths if it holds when C is the inner boundary Di
(see Exercise 2.5).
By substituting for in favour of , we therefore obtain
dy
dx =
ds = 2Ai
(2.5.4)
y
D i x
D i n
where, as before, Ai is the area of the void in the tube crosssection. This
constraint gives the extra information needed to determine the constant k
and, hence, and R. If there are m holes in the crosssection, with taking
the constant value ki (i = 1, . . . , m) on each hole, then there are m relations
of the form (2.5.4) to determine the ki .
As an illustration, suppose that D is the circular annulus a < r < b in
plane polar coordinates (r, ), as illustrated in Figure 2.8(a). Assuming that
is a function of r alone, it must satisfy the boundaryvalue problem
d
1 d
2
r
= 2,
a < r < b,
(2.5.5a)
=
r dr
dr
= k,
r = a,
(2.5.5b)
= 0,
r = b,
(2.5.5c)
45
(2.5.6)
d
2
a
d
=
2a
+
ds =
k
,
dr r=a
log (a/b)
2
D i n
0
(2.5.7)
while Ai = a2 , so (2.5.4) tells us that
k=
b2 a2
.
2
(2.5.8)
This value of k eliminates the logarithmic term from (2.5.6), and we might
have anticipated that this would be necessary by recalling that and +r2 /2
are harmonic conjugates. Since
b2 a2
r2
= const. + k
log r
(2.5.9)
+
2
2
and log r is the harmonic conjugate of , we deduce that
b2 a2
,
= const. k
2
(2.5.10)
b2 r 2
,
2
(2.5.11)
and it is straightforward to substitute this into (2.5.2) and obtain the torsional rigidity
4
R=
(2.5.12)
b a4 .
2
Had we chosen to use instead of , we would have discovered that is
constant and then quickly reproduced (2.5.12) using (2.4.8). Even in this
simple radially symmetric geometry, the bother of nding the arbitrary constant k has outweighed the convenience of introducing a stress function.
When considering multiplyconnected domains it is therefore often a better
idea to return to the physical variable .
46
Linear elastostatics
Notice that (2.5.12) reproduces the result (2.4.17) for a solid circular bar
as a tends to zero. On the other hand, if the tube is thin, so a and b are
nearly equal, then
(2.5.13)
R = 2a4 1 + O () ,
where = b/a 1 1.
We can now compare this with the torsional rigidity of a tube, such as part
of an old bicycle frame, with a thin axial cut, as illustrated in Figure 2.8(b).
Here the crosssection is simply connected, so that must satisfy
1 2
1
2
r
+ 2 2 = 2,
a < r < b,
(2.5.14a)
=
r r
r
r
with
=0
on r = a, b,
(2.5.14b)
on = 0, 2.
(2.5.14c)
and
=0
(2.5.15)
0 < < 1,
(2.5.16a)
with
=0
on = 0, 1,
(2.5.16b)
whose solution is
= 2 a2 (1 ).
(2.5.17)
47
Comparing (2.5.18) with (2.5.13), we see that the change in topology caused
by introducing the cut has a dramatic inuence on the strength of the tube,
reducing its torsional rigidity by two orders of magnitude!
2.6 Plane strain
2.6.1 Denition
A more common conguration than that of antiplane strain is plane strain,
in which a solid is displaced in the (x, y)plane only, with the displacement
being independent of z. Writing
u (x, y)
u = v (x, y) ,
(2.6.1)
0
we nd that the stress tensor takes the form
xx xy 0
= xy yy 0 ,
0
0 zz
where
xx
yy
u
u v
+
+ 2 ,
=
x y
x
v
u v
+
+ 2 ,
=
x y
y
u v
+
,
=
y
x
u v
+
.
=
x y
(2.6.2)
xy
zz
(2.6.3a)
(2.6.3b)
xy
yy
+
= 0,
x
y
(2.6.4)
(2.6.5a)
(2.6.5b)
48
Linear elastostatics
(2.6.6)
2
2
4
4
4
=
+
+
+
2
+
.
=
x2
y 2
x2
y 2
x4
x2 y 2
y 4
(2.6.7)
2A
,
y 2
xy =
2A
,
xy
yy =
2A
.
x2
(2.6.8)
In the same way that the stress function in (2.4.10) is only dened to
within a constant, A is only dened to within a function that is linear in
x and y; in other words, any such function may be added to A without
contributing to the stress.
By using (2.6.3), we obtain the following expressions for the displacement
gradients
2A
2A
u
= 2 + (1 ) 2 ,
x
x
y
2
2
A
A
v
= (1 )
2,
2
2
y
x
y
2
A
u v
+
=
.
y
x
xy
2
(2.6.9a)
(2.6.9b)
(2.6.9c)
Now, by taking 2 /xy of (2.6.9c) and using (2.6.9a) and (2.6.9b), we can
eliminate u and v and hence nd that A satises the biharmonic equation:
4 A = 0.
(2.6.10)
49
y
t
n
D
x
D
(x, y) = X(s), Y (s)
Fig. 2.9 The unit normal n and tangent t to the boundary D of a plane region D.
(2.6.12)
on
D.
(2.6.13)
As illustrated in Figure 2.9, we parametrise D using (x, y) = X(s), Y (s) ,
where s is arclength, so the unit tangent and outward normal vectors are
given by
X
,
t=
Y
n=
Y
,
X
(2.6.14)
50
Linear elastostatics
where is shorthand for d/ds. Using (2.6.8) to write the stress components
in terms of A, we thus nd that (2.6.13) can be written in the form
d
A/y
= .
(2.6.15)
ds A/x
If no surface traction is applied, that is = 0, then it follows from
(2.6.15) that A is constant on D. Since, as noted above, an arbitrary
linear function of x and y may be added to A without aecting the stresses,
we can, in a simplyconnected region, take this constant to be zero without
loss of generality. Then, by taking the dot product of A with t and n
respectively, we deduce that
A
dA
=
=0
ds
n
(2.6.16)
A = 0,
(2.6.17)
Hence, if A satises the biharmonic equation in D and the boundary conditions (2.6.17) on D, then 2 A = 0 and a second use of (2.6.17) reveals that
A 0. This result conrms that the stresses inside a closed body in plane
strain are uniquely determined by the tractions applied to the boundary.
We can combine (2.6.8) and (2.6.9) to obtain the strain energy density in
the form
1
(xx exx + 2xy exy + yy eyy )
2
2 2
2 2
A
2A 2A
1
.
(1 ) A + 2
2 2
=
4
xy
x y 2
W=
(2.6.19)
51
Then the calculus of variations can be used to show that minimisers of the
net strain energy
W dxdy
U=
D
2A
.
r2
(2.6.20a)
1
4
+
+
A = 0.
(2.6.21)
A=
r2
r r r2 2
rr =
We will only consider here cases where a purely normal pressure P is applied,
so the boundary conditions on r = a are
A
1 2 A 1 A
= P,
=0
on r = a;
(2.6.22)
+
r2 2
r r
r r
52
Linear elastostatics
dA
= P a2
dr
on r = a.
(2.6.24)
(2.6.25)
(2.6.27)
(2.6.28)
(2.6.30)
where the ci are again arbitrary constants. For the special case n = 1, the
only physically acceptable solution is
f (r) = c1 r3 .
(2.6.31)
r=a
53
r=b
(2.6.33)
Notice that the n = 1 term does not appear in the series in (2.6.32), so it is
possible to satisfy the boundary condition (2.6.22) only if
2
cos
P ()
d = 0.
(2.6.34)
sin
0
This is simply a manifestation of the solvability condition (1.10.6) and states
that the net force on the disc must be zero.
54
Linear elastostatics
1 dA
= P
r dr
on r = a.
(2.6.35b)
As in Section 2.5, the vital fourth relation needed to determine the constants ci comes from the realisation that the annulus is multiply connected.
We need A(r) to be such that the displacement ur (r) exists and is singlevalued, where, from the constitutive relations (1.11.5),
dur
ur
d2 A
ur
1 dA
dur
= ( + 2)
+ ,
+ ( + 2) .
(2.6.36)
=
r dr
dr
r
dr2
dr
r
By crossdierentiation, we nd that that the compatibility condition for ur
to exist is
d3 A 1 d2 A
1 dA
= 0,
(2.6.37)
+
2
dr3
r dr2
r dr
which is satised by (2.6.25) only if c3 = 0. The boundary conditions (2.6.35)
may then be used to evaluate the remaining three constants and hence show
that
r
2
P a2 b2
P a2
2
a
log
A=
r
.
(2.6.38)
2 (b2 a2 )
b2 a2
b
In a plane strain problem with multiple tractionfree holes, we would have
A
(2.6.39)
= Ki
n
on the ith hole, where the constants ki and Ki must be determined by ensuring that both displacement components (u, v) are singlevalued around
each hole. As noted in Section 2.5, the utility of introducing a stress function in such cases is negated by the diculty of solving for all the arbitrary constants, and it is often preferable to work with physical variables in
A = ki ,
55
multiplyconnected domains. By way of illustration, had we posed the annular problem above in terms of the radial displacement ur (r) rather than
A, we would have obtained the considerably more tractable problem
d 1 d
(rur ) = 0,
a < r < b,
(2.6.40a)
dr r dr
ur
dur
+ = P,
r = a,
(2.6.40b)
( + 2)
dr
r
ur
dur
+ = 0,
r = b.
(2.6.40c)
( + 2)
dr
r
This quickly yields the solution
r
b2
P a2
+
,
(2.6.41)
ur =
2 (b2 a2 ) + r
and we can then recover the stress components using the constitutive relations (1.11.5). In our radially symmetric geometry, r is zero and the only
non zero stresses are the radial stress rr , the hoop stress and the axial
stress zz , given respectively by
b2
b2
P a2
P a2
2P a2
1
1
+
,
=
=
,
rr = 2
zz
b a2
r2
b2 a2
r2
b2 a2
(2.6.42)
where denotes Poissons ratio, as usual. These results are in accordance
with (2.6.38).
We can now use the above solution to predict the failure of a gun barrel
as P increases. As explained in Section 2.2.5, the Tresca criterion predicts
that the material will fail when
i j 
= Y ,
(2.6.43)
S = max
i,j
2
where Y is the yield stress and i are the principal stresses. Here the stress
tensor is diagonal, so i are just the three stress components given in (2.6.42).
If we assume that the material has positive Poissons ratio then, since
< 1/2 and r < b, we can deduce from (2.6.42) the inequalities
rr < 0 < zz < ,
(2.6.44)
P a2 b2
rr
= 2
.
2
(b a2 ) r2
(2.6.45)
for an auxetic material with < 0, the second inequality in (2.6.44) is reversed and the maximum
shear stress may be z z instead of (2.6.45)
56
Linear elastostatics
(2.6.46)
P = Y 1 2 .
b
Notice in particular how the barrel becomes more susceptible to failure as
its thickness b a is reduced. It turns out that this is a canonical problem
for the theory of plasticity, as will be explained in Chapter 8.
2.6.6 Plane strain in a rectangle
In Section 2.6.4, we showed how plane strain in a disc can be solved by separating the variables in polar coordinates. On the face of it, the procedure
was only slightly more complicated than that for solving Laplaces equation,
the extra complication arising from the application of two boundary conditions rather than one on r = a. In other geometries, however, solving the
biharmonic equation generally involves serious practical diculties that are
not encountered with Laplaces equation.
To illustrate the problems that may occur, we will now consider plane
strain in a cylinder with rectangular crosssection x < a, y < b. For simplicity, we suppose that the faces are subject to purely tangential tractions,
that is
xx = 0,
xy = g (y)
on x = a,
(2.6.47a)
yy = 0,
xy = f (x)
on y = b,
(2.6.47b)
a
a
f+ (x) + f (x) dx a
b
b
g+ (y) + g (y) dy = 0.
(2.6.48b)
The Airy stress function therefore satises the biharmonic equation subject
to
A = 0,
A = 0,
2A
= g (y)
xy
2A
= f (x)
xy
on x = a,
(2.6.49a)
on y = b,
(2.6.49b)
57
y
f+
g+
b
a
g
f
Fig. 2.11 A plane rectangular region subject to tangential tractions on its faces.
(2.6.50)
(2.6.51)
58
Linear elastostatics
(2.6.53)
n odd
where An and Bn are constants and the functions Fn and Gn are dened by
na
nx
na
nx
Fn (x) = a sinh
cosh
x cosh
sinh
, (2.6.54a)
2b
2b
2b
2b
ny
ny
nb
nb
cosh
sinh
y cosh
. (2.6.54b)
Gn (y) = b sinh
2a
2a
2a
2a
In contrast to more elementary Fourier series examples, we now encounter
a serious diculty when we try to determine An and Bn in terms of the
applied tractions. Considering x = a, for example, we nd that
ny
n
n
g+ (y) =
aFn (a)An sin
+ b sin
Bn Gn (y) , (2.6.55)
2ab
2b
2
n odd
but {sin(ny/2b), Gn (y)} do not form an orthogonal set on b < y < b.
Hence, if we were to multiply (2.6.55) by Gm (y) and integrate over y < b in
the usual Fourier series procedure, we would end up with an innite system
of equations for the constants An and Bn . We will discuss this diculty
further below.
2.6.7 Plane strain in a semiinnite strip
We now present an example which is technically easier than the rectangular geometry discussed above and which has wide signicance in the theory of linear elasticity. We consider plane strain in the semiinnite strip
0 < x < , h/2 < y < h/2, with zero stress on y = h/2 and with prescribed tractions
xx = x (y),
xy = y (y)
(2.6.56)
2A
(0, y) = y (y),
xy
(2.6.57)
A
(x, h/2) = 0.
y
(2.6.58)
However, because the end x = 0 is not tractionfree, we cannot use the same
conditions on the other face y = h/2. All we can safely deduce from the
59
y
stress free
xx = x (y)
xy = y (y)
stress free
(2.6.59a)
A
(x, h/2) = c,
y
(2.6.59b)
h/2
h/2
y (y) dy, c =
h/2
h/2
x (y) dy.
(2.6.61)
(h 2y)2
4a(h + y) + 4bx(h + y) + ch(h + 2y)
3
8h
(2.6.62)
60
Linear elastostatics
satises both the biharmonic equation and the boundary conditions (2.6.58),
(2.6.59). Hence, if we set
A = A + A,
(2.6.63)
A
A
(2.6.64)
h
2
h
a1 + sin
2
h
2
a2 + cos
h
2
a3 sin
h
2
a4 = 0,
(2.6.66a)
h
h
h
h
h
h
sin
cos
a1 sin
+
a2
cos
2
2
2
2
2
2
h
h
a3 + cos
a4 = 0,
sin
2
2
(2.6.66b)
which admit nontrivial solutions for the constants ai only if the relevant
determinant is zero. This eventually leads to the transcendental equation
sin (h) = h,
(2.6.67)
and the eigenvalues satisfying (2.6.67) are complex. This occurs because
the ordinary dierential equation (2.6.51), along with the boundary conditions G = G = 0 at y = h/2 is not a selfadjoint eigenvalue problem,
in contrast with the St
urmLiouville problems which would be encountered
were we solving Laplaces equation. This has the additional implication that
the calculation is much simpler if the problem is posed on the strip 0 < y < h
61
the eigenfunctions are not mutually orthogonal, which seriously complicates the task of tting the initial conditions, as in (2.6.55). Had we instead
imposed the conditions A = 2 A/y 2 = 0 on y = h/2, we would have
found that the eigenvalues are real and the eigenfunctions are orthogonal
trigonometric functions. We will see a practically relevant example of this
in Section 4.6 but, unfortunately, specifying 2 A/n2 on the boundary has
no obvious physical signicance in plane strain.
can in fact be
A completeness argument can be given to show that A
expressed as a sum of terms of the form (2.6.65), with Re() > 0, although,
because of the diculties described above, the coecients can only be computed numerically by inverting an innite matrix. Thus, whatever tractions
will decay exponentially as x . The stress far
are imposed on x = 0, A
from the edge of the strip will then be characterised entirely by the stress
function A , which corresponds to a fareld stress, as x ,
xx
xy
3b 2
2
4y
h
,
2h3
yy 0.
(2.6.68)
The net tensile force T , shear force N and bending moment M exerted
on any section x = const. by the stress eld (2.6.68) are given by
h/2
h/2
xx dy = c,
N (x) =
xy dy = b,
(2.6.69a)
T (x) =
h/2
h/2
h/2
M (x) =
h/2
yxx dy = a + bx +
ch
.
2
(2.6.69b)
(2.6.70)
h/2
N (0) =
h/2
h/2
y (y) dy,
(2.6.71a)
h/2
M (0) =
h/2
yx (y) dy.
(2.6.71b)
62
Linear elastostatics
2 u
2 v
=
=
y
1 x
u v
+
=
y
x
T
12M (x)y
+
,
h
h3
3N 2
2
h
.
4y
2h3
(2.6.72a)
(2.6.72b)
xy = (x)
on y = 0.
(2.6.74)
Assuming further that this integral converges, we also recall the inversion
formula
1
ikx dx.
Ae
(2.6.76)
A=
2
63
(2.6.77a)
y = 0,
(2.6.77b)
y = 0,
(2.6.77c)
y .
(2.6.77d)
(2.6.78)
(2.6.79b)
(2.6.79c)
ky
(2.6.79a)
2!
d!
v
! d A = ((2 1) yk) P (k)eyk ,
= (1 )k 2 A
dy
dy 2
(2.6.81)
64
Linear elastostatics
(2.6.82)
x
(s2 + y 2 )2
Alternatively, we can also obtain (2.6.83) from the identity
E
yy
xy
xx
2v
+
,
= 2(1 + )
x2
x
y
y
(2.6.84)
dx
s x
"
where H[P ] is called the Hilbert transform of P , being the principal value
integral dened in Exercise 2.12. As shown
#
$in Exercise 2.13, this transform
may be inverted using the formula H H[f ] f , and hence we can calculate the pressure required to achieve a given surface displacement v0 (x),
namely
# $
P =
H v0 .
(2.6.86)
1
As an illustrative example, the surface displacement
a
1 + x2 /L2
(2.6.87a)
1 x2 /L2
a
,
(1 )L (1 + x2 /L2 )2
(2.6.87b)
v0 (x) =
corresponds to a surface pressure
P (x) =
and both are plotted in Figure 2.13. Notice that in some parts a negative
pressure must be applied to obtain the given bounded surface displacement.
65
0.8
0.6
v0 /a
0.4
(1 )LP/a
0.2
4
2
x/L
Fig. 2.13 The surface displacement v0 (x) of a halfspace and corresponding surface
pressure P (x) dened by (2.6.87).
One can deduce generally from (2.6.85) that it is only possible to achieve a
localised displacement, with v0 0 as x , if
P (x) dx = 0,
(2.6.88)
xy
yy
V
+
=
,
x
y
y
(2.6.89)
2A
,
y 2
xy =
2A
,
xy
yy = V +
2A
.
x2
(2.6.90)
66
Linear elastostatics
satises the
Now when we eliminate the displacements, we discover that A
inhomogeneous biharmonic equation
1 2
4
2 V = 0.
(2.6.91)
A+
1
This result makes it appear that gravity, which is modelled by setting
since 2 V = 0. However, if traction conditions
V = gy, has no eect on A,
for are given, V will enter when (2.6.90) is used to transform these con In other words, the substitution (2.6.90)
ditions into boundary data for A.
eliminates the body force from the problem, but replaces it with boundary
tractions. This substitution is often useful in solving elastostatic problems
numerically.
As an illustration, let us try to nd the stress in the elastic medium
surrounding a tunnel, a problem of importance say to civil engineers involved
with underground railways. Supposing that the tunnel is circular, with radius
a, and at a depth H, we incorporate gravity into the plane polar stress
components (2.6.20) by setting
1 2A
1 A
,
rr = g (r sin H) + 2 2 +
r
r r
1 A
,
r =
r r
= g (r sin H) +
2A
,
r2
(2.6.92a)
(2.6.92b)
(2.6.92c)
where the depth is y = Hr sin . As shown above, the modied Airy stress
function satises the biharmonic equation, and the zerotraction conditions
rr = r = 0 on the tunnel wall lead to
1 2A
1 A
A
= 0,
= g (H r sin )
(2.6.93)
+
r r
r2 2
r r
on r = a.
By eliminating A/r
from the boundary conditions (2.6.93), we obtain
2A
= ga2 (H a sin )
+A
2
on r = a.
(2.6.94)
), whose general
This is eectively an ordinary dierential equation for A(a,
solution is
3
) = ga2 H + ga cos + Aa cos + Ba sin ,
A(a,
2
(2.6.95a)
67
where A and B are arbitrary constants, and (2.6.93) then gives us the second
condition
ga2
A
= gaH +
cos + A cos + B sin
r
2
on r = a.
(2.6.95b)
(2.6.96)
We do not yet have a unique solution, since the constants A and B still
appear to be arbitrary. As in Section 2.6.5, the key to closing the problem in this multiplyconnected domain is to ensure that the displacement
eld is singlevalued. As shown in Exercise 2.15, this condition allows us to
determine both remaining constants, and we nd that
A = 0,
B=
(2.6.98)
(1 2) a2
a2
H + r sin 1 +
,
= g 1 2
r
4(1 ) r2
a2
ga2 cos (1 2)
1 2 ,
=
r
4(1 )
r
2
(1 2) a2 sin
a
.
H +
= gr sin + g 1 + 2
r
4(1 ) r
(2.6.99a)
(2.6.99b)
(2.6.99c)
68
Linear elastostatics
3 4
ga sin
2(1 )
on r = a.
(2.6.100)
Hence a small tunnel suers twice as much compressive stress as would occur
in its absence. Moreover, the minimum compressive stress occurs at = /2,
which is the roof of the tunnel, while the maximum compressive stress is at
the bottom = /2. In fact, it can be shown that this nal result remains
valid for all values of H and a (see Mindlin, 1939).
2.7 Compatibility
In Section 1.6, Newtons second law led us to write down the equation
ij
+ gi = 0
xj
(2.7.1)
2.7 Compatibility
69
T
First consider a plane strain problem, in which u = u(x, y), v(x, y), 0
and
u
1 u v
v
,
exy =
+
,
eyy =
.
(2.7.4)
exx =
x
2 y
x
y
Since the three strain components are functions of just two displacements,
(2.7.4) is overdetermined when viewed as a system of equations for (u, v).
However, whenever u is a twice continuously dierentiable singlevalued
function, we must have 2 u/xy 2 u/yx and similarly for v. It then
follows from crossdierentiation that
2 eyy
2 exy
2 exx
+
2
= 0.
y 2
xy
x2
(2.7.5a)
2
= 0,
(2.7.5b)
z 2
yz
y 2
2 ezx
2 exx
2 ezz
2
= 0,
(2.7.5c)
+
x2
zx
z 2
and it might look as if we have then found all the compatibility conditions
needed for the existence of u, v and w. However, we also have to ensure
that displacements transverse to each coordinate plane exist and are singlevalued, so that, for example, 2 w/xy 2 w/yx. Once we take this
into account, we obtain the additional three compatibility relations
exy
eyz
exz
2 exx
=
+
,
(2.7.5d)
yz
x
y
z
x
2 eyy
eyz
eyx
ezx
=
+
,
(2.7.5e)
zx
y
z
x
y
exy
2 ezz
ezy
ezx
=
+
.
(2.7.5f)
xy
z
x
y
z
We thus have a total of six compatibility conditions, which now seems excessive for a system of six equations in the three unknowns ui . In fact, only
three of the equations are independent, in the sense that crossdierentiation
to eliminate all but three unknowns from (2.7.5) inevitably leads to just one
independent equation. However, all six compatibility conditions are necessary for a singlevalued displacement to exist. This situation is analogous
70
Linear elastostatics
to that which arises when considering a vector eld u which is the gradient of a potential. In this case, it is necessary that all three components of
curl u vanish, although only two components of curl u are independent in
the above sense. Our situation is a generalisation of this statement, since, as
shown in Exercise 2.21, the compatibility conditions (2.7.5) are equivalent
to the components of the tensor equation
0
/z /y
curl (curl E)T = 0, where curl = /z
0
/x . (2.7.6)
/y /x
0
We also remark that any three independent compatibility equations for eij
are just as underdetermined as the three equilibrium equations (2.7.1) are
for ij . Hence, the fact that the compatibility conditions ensure the existence of three physically acceptable displacement components suggests that
(2.7.1) will similarly guarantee the existence of three scalar stress functions.
We have already encountered an application of this idea to plane strain in
Section 2.6, and we will show below in Section 2.8 how it may be generalised
to nonplanar problems.
It is important to note that, if eij does not satisfy (2.7.5), then there is
no physically acceptable displacement eld that gives rise to such a strain
and it is, therefore, incompatible with linear elasticity. In Section 2.9.4 we
will discuss one physical interpretation of incompatibility in detail. Meanwhile, we simply conjecture that, if we are presented with an elastic material
which has been deformed under the action of, say, some nonzero boundary
tractions, then, when the tractions are removed, the necessary and sucient
condition for the material to return to its pristine unstrained state is that
the strain eld in it satises (2.7.5), and we will see specic examples of this
in Section 8.4.
(2.8.1)
(2.8.2)
71
and so, by taking the curl once more and using (2.8.1), we obtain
4 u = 0.
(2.8.3)
with
A = 0.
(2.8.5)
(It may be shown that this decomposition applies to any continuously differentiable vector eld u.) Substitution of (2.8.5) into (2.2.1) reveals that
the potential functions and A must also satisfy the biharmonic equation
4 = 0,
4 A = 0.
(2.8.6)
These results all suggest that we may be able to construct large classes
of solutions of (2.2.1) by considering suitable harmonic or biharmonic scalar
problems. We have already seen some examples of this approach in plane
and antiplane strain (Sections 2.32.6) and we will now show how analogous
ideas may be applied in some other commonlyoccurring congurations.
2.8.2 Plane strain revisited
We begin by reexamining the theory of plane strain in the light of the concept of compatibility developed in Section 2.7. As we showed in Section 2.6,
the two conservation laws
xy
xx
+
= 0,
x
y
xy
yy
+
=0
x
y
(2.8.7)
72
Linear elastostatics
components, is
2 yy
2 xy
2 xx
+
2
2 (xx + yy ) = 0.
y 2
xy
x2
(2.8.8)
Now (2.8.7) and (2.8.8) form a closed system of partial dierential equations
for the stress components (xx , xy , yy ). If, as in Section 2.6, we write
xx =
2A
,
y 2
xy =
2A
,
xy
yy =
2A
,
x2
(2.8.9)
then the stress balance equations (2.8.7) are satised identically, and substitution of (2.8.9) into (2.8.8) quickly retrieves the biharmonic equation
for A.
It is worth noting that an inhomogeneous equation for A of the form
4 A = f
(2.8.10)
does not, as might have been suspected from (2.6.91), correspond to the
imposition of a body force. The mere existence of A implies that (2.8.7) are
satised, so there is no such force. Instead the righthand side f of (2.8.10)
corresponds to a distribution of incompatibility through the material, and
we will discuss in Section 2.9.4 and Chapter 8 what the physical signicance
of such a conguration might be.
Now let us suppose that we do impose a body force in, say, the xdirection
so that the rst Navier equation (2.8.7a) is modied to
xy
xx
+
= f.
x
y
(2.8.11)
Now there is no longer an Airy stress function, but we can instead try
to obtain an integrability condition from the two remaining homogeneous
equations (2.8.7b) and (2.8.8). From (2.8.7b) we deduce the existence of a
function A1 (x, y) such that
xy =
A1
,
y
yy =
A1
,
x
2
2 A1
= 0.
xx
+ (1 )
y 2
y 2
x
(2.8.12)
(2.8.13)
2
2
A2
+ (1 ) A1 = 0.
(2.8.14)
y 2
y 2
73
2
L.
(2.8.15)
A1 =
y 2
The existence of the Love function L, also known as the Love stress function or the Love strain function, guarantees that the strain eld is compatible
and the stress balance in the ydirection is satised. It is related to the stress
components by
2
A2
2
(2.8.16a)
=
+ (1 ) L,
xx =
x
x y 2
2
A1
2
xy =
(2.8.16b)
=
L,
y
y y 2
2
A1
yy =
(2.8.16c)
=
2 L,
x
x y 2
and, when we nally substitute these into the xmomentum equation (2.8.11),
we nd that L is such that
xy
xx
+
= (1 )4 L = f.
x
y
(2.8.17)
Even though the Love function is ideally suited to problems with a unidirectional body force, the derivation given above remains valid if there is
no body force, in which case f 0 and L satises the biharmonic equation.
Hence, since L is also easily related to the stress components by (2.8.16), it
provides an alternative to the Airy stress function for plane strain problems;
indeed it can be shown that L/x diers from A by a harmonic function.
More importantly, when we try to nd u and v using
u
= (1 )xx yy ,
x
u v
+
= xy ,
y
x
v
= (1 )yy xx ,
2
y
2
(2.8.18a)
(2.8.18b)
(2.8.18c)
(1 2) L +
u=
2
y 2
2 xy
74
Linear elastostatics
4
% A=
+ 2
A = 0,
(2.8.21)
x2
y
the tilde denoting that 2 /z 2 has been omitted from the Laplacian. However, this is now insucient to determine the zdependence of A, and so we
75
must ensure that all ve remaining compatibility conditions are also satised. To this end we nd that the nonzero strain components are given
by
2A
2A
,
y 2
x2
2A
,
Eexy = (1 + )xy = (1 + )
xy
2A
2A
Eeyy = yy xx =
,
x2
y 2
% 2 A;
Eezz = (xx + yy ) =
Eexx = xx yy =
(2.8.22a)
(2.8.22b)
(2.8.22c)
(2.8.22d)
(2.8.24)
=
+
x2
z 2
xy
z 2
2
2
A
(1 + ) 2 + = 0. (2.8.26)
=
y 2
z
Since terms linear in x and y do not contribute to the stress, we lose no
generality in setting
(1 + )
2A
+ = 0,
z 2
(2.8.27)
(x, y)z 2
+ z1 (x, y) + 0 (x, y),
2(1 + )
(2.8.28)
76
Linear elastostatics
where
% 2 0 = ,
% 2 = 0,
% 2 1 = = const..
(2.8.29)
yy = xy = 0,
2xz
.
u=
2yz
2
2
2
2
x y + z
(2.8.30a)
(2.8.30b)
This solution represents bending of a beam about the yaxis, and will provide
a useful check on the approximate beam theory to be developed in Chapters 4
and 6.
(2.8.31)
1
1
2
2 uz
+ 2(1 )
+
+ (1 2) 2 ur = 0. (2.8.32)
rz
r2
r r r2
z
77
(2.8.33)
2
r r r
r r2
r r
r r r2 r
implies that
2
2
2
f
1
1
2(1 )
2 + (1 2) 2
+
2
r
r r r
z
rz
2
2
2
1
2(1 )
+ (1 2) 2 f. (2.8.34)
=
+
rz
r2
r r
z
Since the righthand side may be written as
2
1
2
(1 2) +
r
f,
rz
r r
r
we may infer from (2.8.32) the existence of a function L(r, z) such that
L
1 2L
1
1
2
ur =
, uz =
(1 2) L +
r
. (2.8.35)
2 rz
2
r r
r
With this displacement, it is straightforward to evaluate the stress components from (1.11.5) as
2L
2
rr =
L ,
(2.8.36a)
z r2
L
1
rz =
r
2 L ,
(2.8.36b)
r r r
r
L
1
2
r
+ (1 ) L ,
(2.8.36c)
zz =
z r r
r
1 L
(2.8.36d)
+ 2 L .
=
z
r r
Now to obtain an equation for L, we must substitute these into the
ez component of the Navier equation, which reads
1
zz
(rrz ) +
= f,
(2.8.37)
r r
z
where the body force is f ez per unit volume. After a lengthy algebraic
manipulation, we nd that L satises
(1 )4 L = f,
(2.8.38)
78
Linear elastostatics
which is identical to the equation (2.8.17) for the planar Love function. We
will now show that this is no accident and, in Section 2.9, we will further
show how to use Love functions to construct solutions of the Navier equation
corresponding to point forces.
(2.8.39)
where the body force f per unit volume is no longer presumed to be unidirectional. To obtain the Galerkin representation, we make the outoftheblue
hypothesis that the displacement can be written in the form
u = a2 B b grad div B,
(2.8.40)
for some vector eld B(x) and suitably chosen constants a and b.
The advantage of this representation is that, when we substitute (2.8.40)
into (2.8.39), we nd that B satises
( + 2)(a b) (grad div)2 B + a curl4 B = f ,
(2.8.41)
and the dierential operator on the lefthand side may be made proportional
to the biharmonic operator by choosing
+ 2
a
=
= 2(1 ).
b
+
(2.8.42)
(2.8.43)
(2.8.44)
The planar and cylindrical Love functions may now be obtained by supposing that B = L(x, y)i or B = L(r, z)ez respectively. Evidently each will
only satisfy (2.8.43) if the body force f shares the corresponding symmetry.
This is reasonable since we expect the curl of (2.8.40) to determine curl B given u, and, if
B = , (2.8.40) is eectively a Poisson equation for . Note also that, in view of (1.7.11),
(2.8.44) may be viewed as an example of a Helmholtz representation (2.8.5).
79
However, in cases where there is no body force, (2.8.43) opens up the possibility of nding a wealth of scalar stress functions describing problems with
various symmetries, simply by posing suitable forms for B.
(2.8.45)
f
.
1
(2.8.46)
(2.8.47)
(2.8.48)
(2.8.49)
(2.8.50)
80
Linear elastostatics
and =
= z
,z
, (3 4)
x
+y
,
x y
2
x
y
(2.8.52)
whence, after a simple calculation, (2.8.51) gives
T
2u = 0, 0, 4(1 )(1 2) ,
(2.8.53)
(2.8.54)
1
+ (x1 + y2 )
2u = 2(1 )1
x
2
1
2v = 2(1 )2
+ (x1 + y2 )
y
2
and
(2.8.55a)
(2.8.55b)
2
2
(x1 + y2 ).
xy xy
(2.8.56)
Hence (2.6.9c) suggests that we choose 1 /y + 2 /x to be zero and this
in turn suggests that we take 1 and 2 to be harmonic conjugates. Then
(2.6.9c) gives the Goursat representation
2
u v
+
y
x
= 2(1 )
1
2
+
y
x
1
A = + (x1 + y2 ),
2
(2.8.57)
81
2 3
2 2
+
,
y 2
z 2
xy =
2 3
,
xy
yy =
2 1
2 3
+
,
z 2
x2
yz =
2 1
,
yz
zz =
2 2
2 1
+
,
x2
y 2
(2.8.59a)
zx =
2 2
.
zx
(2.8.59b)
It is easy to verify that the steady Navier equation with no body force is
identically satised for any stress eld of this form. Another popular choice
is the Morera stress functions i , for which
1
2
2 1
3
,
xy =
,
(2.8.60a)
xx = 2
yz
z z
x
y
2
3
2 2
1
yy = 2
,
yz =
,
(2.8.60b)
zx
x x
y
z
3
1
2 3
2
zz = 2
,
zy =
.
(2.8.60c)
xy
y y
z
x
At rst glance, little is gained by formulating threedimensional problems in
terms of three stress functions rather than just using the three components
of the displacement u. However, we will now show that these potentials
suggest a useful unication of all the stress functions that we have thus far
encountered.
In general, given any symmetric matrix eld A(x), we can construct a
symmetric stress tensor (x) with zero divergence using the formula
ij =
k,l,m,n
ikl jmn
2 Akm
,
xl xn
(2.8.61)
82
Linear elastostatics
where ijk is the alternating symbol. Using (2.7.6), we can formally write
(2.8.61) in the form
= curl (curl A)T ,
(2.8.62)
which makes it clear that the tensor thus dened is symmetric and identically satises div 0.
All the stress functions we have used in this chapter can now be viewed
simply as particular choices of the matrix A. For example, the Maxwell and
Morera functions are obtained by choosing
1 0 0
0 3 2
A = 0 2 0 ,
or A = 3 0 1 ,
(2.8.63)
0 0 3
2 1 0
respectively. The Airy stress function for plane strain (or plane stress) is obtained by setting 1 = 2 = 0, 3 = A, while the antiplane stress function
from Section 2.4 corresponds to 3 = 0, 1 /x 2 /y = (x, y). The
Love function L does not fall into the same category, since its existence relies
on the compatibility conditions as well as one component of the equilibrium
equation. This is why it is often referred to as a strain function rather than
a stress function.
Finally, we note the strong analogy between the equations (2.8.59), (2.8.60)
satised by the Maxwell and Morera stress potentials and the compatibility conditions that we found for the strain components in Section 2.7. The
formulation (2.8.62) shows that this resemblance is no accident. By using
(2.7.6), we can obtain the compatibility condition for , namely
T
= 0,
(2.8.64)
curl curl (1 + ) kk I
and the equations satised by the Maxwell functions i , for example, can
then be found by substituting for from (2.8.62).
2.9 Singular solutions in elastostatics
2.9.1 The deltafunction
In potential theory, it is often helpful to introduce point singularities such
as point charges or dipoles in electrostatics or point masses in gravitation.
In inviscid uid dynamics it is also common to discuss singular solutions describing sources and vortices, in the latter of which the singularity is localised
on a line rather than a point. As we shall see, such fundamental solutions
are very useful building blocks for constructing more general solutions.
83
(x)
1.2
0.8
= 0.25
0.6
= 0.5
0.4
= 1.0
0.2
4
3
2
1
(2.9.1)
(x) =
2
(x + 2 )
have the property that
(x) dx 1
(2.9.2)
for any suitably smooth test function . We can then dene a multidimensional deltafunction (x) as a product of onedimensional deltafunctions,
for example
(x) = (x)(y)(z),
(2.9.4)
84
Linear elastostatics
(2.9.6)
2
1
2
three dimensions:
= (x)(y)(z). (2.9.7c)
4 x2 + y 2 + z 2
xy
yy
+
= 0.
x
y
(2.9.8)
(2.9.9)
85
Fig. 2.15 Contours of the maximum shear stress S created by a point force acting
at the origin. Here = 1/4 and f /S = 0.2, 0.4, . . . , 2.0.
f
r2 (log r 1).
8(1 )
(2.9.11)
2
x y2
(3
4)
log
r
f
,
2r2
(2.9.12)
u=
xy
8(1 )
r2
and the stress is recovered from (2.8.16) as
f
(3 2)x3 (1 2)xy 2 (3 2)x2 y (1 2)y 3
.
=
4 (1 ) r4 (3 2)x2 y (1 2)y 3 (1 2)x3 (1 + 2)xy 2
(2.9.13)
We can hence nd the maximum shear stress S dened in (2.2.36), by calculating the eigenvalues i of :
S=
f
1 2 
=
4(1 )2 x2 + (1 2)2 y 2 .
2
2
2(1 )r
(2.9.14)
Typical contours of this function are plotted in Figure 2.15 (with = 1/4).
The shear stress becomes unbounded as the point force at the origin is
approached, and we would expect the material to yield plastically on that
surface at which S reaches a critical value.
The response to a point force in three dimensions may be found in an analogous fashion by using the axisymmetric Love function; see Exercise 2.19.
Such problems become very much harder if we attempt to impose any
boundary conditions, and analytical solutions are usually impossible to nd
unless the geometry is very simple. One relatively straightforward example
86
Linear elastostatics
A
2A
y 2
xy
(2.9.15)
= 0 (x, y h) + 0 (x, y + h) +
,
2
2A
A
xy
x2
now satises the twodimensional biharmonic equation in y > 0
where A
and the boundary conditions
#
$
f x (1 + 2)h2 (1 2)x2
2A
A
= 0,
=
(2.9.16)
y
x2
2(1 ) (h2 + x2 )2
on y = 0. This is now precisely the halfspace problem we have already
analysed in Section 2.6.8, and may be solved by substituting the righthand
side of (2.9.16) into (2.6.80) to give
2hxy
f
2L
2 L =
2
y
y
2L
2
L = 0
y 2
on y = 0.
(2.9.18a)
(2.9.18b)
87
R2 (log R 1) + L,
8(1 )
R2 = x2 + (y h)2 ,
(2.9.19)
= (r, z),
(2.9.20)
1
z
,
r
2 r
2uz = 2(1 )
1 (z)
.
z
2 z
(2.9.21)
From the formulae of Section 1.11.2, the vanishing of the shear stress requires
that
ur
uz
+
=0
on z = 0,
(2.9.22)
z
r
so that
2
2
(1 2)
= 0.
rz
r
(2.9.23)
(1 2) = 0
z
(2.9.24)
2
uz
(rur ) + ( + 2)
= 2(1 )
2 2
r r
z
z
z
.
=
z
zz =
(2.9.25)
88
Linear elastostatics
(2.9.27)
(2.9.28)
(2.9.29)
subject to suitable boundary conditions. We can then use the key property
(2.9.5) of the deltafunction and formal dierentiation under the integral to
deduce that
u(x) =
R3
G(x )f () d
(2.9.30)
(2.9.31)
2 Gij
2 Gki
+
= ij (x),
xj xk
xk xk
(2.9.32)
89
From a practical viewpoint, Gij may be interpreted physically as the displacement in the jdirection at x in response to a unit point force in the
idirection at . In other words, (2.9.33) shows that the response to an arbitrary body force f (x) may be written as a superposition of point force
responses like those constructed above in Section 2.9.2.
In the case where the elastic medium occupies the whole space, with
bounded stress at innity, we can infer the Greens tensor directly from the
point force solutions found in Section 2.9.2. In plane strain, the displacement caused by a unit point force in the xdirection is given by (2.9.12)
with f = 1:
2
x y2
1
G11 =
(3 4) log r ,
(2.9.34a)
8(1 )
2r2
xy
1
,
(2.9.34b)
G12 =
8(1 ) r2
and, by symmetry, we deduce that the Greens tensor is given by
2
xy
x y2
1
2r2 (3 4) log r
r2
G(x, y) =
.
xy
y 2 x2
8(1 )
(3 4) log r
r2
2r2
(2.9.35)
The corresponding threedimensional result is derived in Exercise 2.18.
Finding the Greens tensor explicitly in a body with nite boundaries is
usually impossible unless there is sucient symmetry to allow some trick to
be employed. In general, we must now solve (2.9.31) along with boundary
conditions for G corresponding to those that are imposed on u (so that
(2.9.33) satises the required boundary conditions). Since the problem is
not translation invariant, we are not at liberty to move the deltafunction
to the origin, as we did in (2.9.31). Hence we can no longer assume that G
is a function only of x , and we must rewrite (2.9.29) as
LG(x, ) = (x ).
(2.9.36)
90
Linear elastostatics
(2.9.39)
(2.9.40)
where we have neglected the body force for simplicity. Hence, multiplying
(2.9.39) by ui , (2.9.40) by Gij , then subtracting and integrating, we nd that
ui () =
G i (x ) ( n (x)) u(x) i n(x ) dS, (2.9.41)
D
where G i = (Gi1 , Gi2 , Gi3 )T and i is the stress eld associated with the
displacement G i .
Hence, given the traction and displacement on the boundary D, we have
an explicit representation of u everywhere in D. However, we do not expect
both u and n to be given simultaneously on the boundary, so let us assume
for example that the surface traction n is prescribed. In this case, (2.9.41)
becomes an integral equation for u() as tends to the boundary of the
elastic solid. To solve this integral equation rather than the Navier equation
numerically is the basic idea behind the boundary element method. This
approach may prove advantageous compared to the more commonly used
nite element method, since only the boundary of the solid needs to be
discretised, rather than the whole body. On the other hand, it also brings
its own share of diculties. For example, the matrices resulting from the
discretisation are usually full, and care must be taken with the singularities
that occur in G i as approaches D.
91
f
f
f
x
P a2
.
2r
(2.9.42)
92
Exercises
(2.9.43)
for any suitably smooth test function . Hence (2.9.42) satises the compatibility relations everywhere except at the origin, where the localised source
of incompatibility is known as a nucleus of strain (Love, 1944, Article 132).
Note that no traction need be applied away from the origin to maintain the
stress eld associated with (2.9.42), and the material is therefore said to be
in a state of selfstress.
More generally, in Exercise 2.21 it is shown that the compatibility conditions (2.7.5) are the six components of the symmetric tensor equation
ij = ikl jmn
2 emk
0,
xn xl
(2.9.46)
2
=
2
2
y
xy
x
2
(2.9.47)
which reproduces (2.7.5a) and shows explicitly how 4 A is related to compatibility. Thus the righthand side of (2.9.44) can be thought of as a point
singularity in the incompatibility tensor, and this prompts us to wonder what
the physical signicance might be of introducing other point singularities on
the righthand side of (2.9.46). For example, the physical interpretation of
solutions of
2 eyy
2 exy
2 exx
+
2
= (x) (y)
(2.9.48)
y 2
xy
x2
has fundamental implications for the theory of metal plasticity, to which we
will return in Chapter 8.
93
Exercises
2.1
2.2
2.3
Use strips of sti paper and sticky tape to construct the model shown
in Figure 2.3(a). Now when you pull it in the direction shown, it
expands in the transverse direction, as indicated by Figure 2.3(b). It
therefore has negative Poissons ratio.
An elastic membrane is stretched to an isotropic tension T such that
its height is given by z = w(x, y). Assuming that the membrane is
nearly horizontal (so that w 1), show that T must be spatially
uniform and that w must satisfy Laplaces equation.
A uniform cylindrical bar is held with its axis along the zaxis and
its boundary pinned. If the gravitational acceleration g acts in the
z direction, show that the vertical displacement w(x, y) satises
2 w =
2.4
94
2.5
2.6
Linear elastostatics
#
$ #
$
+
Cn cosh kn ( ) sin kn log (b/r) ,
n=1
where
n
kn =
,
log (b/a)
#
$
4 b2 (1)n a2 log2 (b/a)
#
$.
Cn =
n cosh(kn ) n2 2 + 4 log2 (b/a)
.
$
#
2 2 + 4 log2 (b/a) 2
n
n=1
Now suppose that the tube is thin, so that b/a = 1 + , where 1.
By expanding the above expression for small , show that
1 93(5)
2 4 3
a 1
+
+ ,
R=
3
2
6
2.7
Exercises
95
2.8
2.9
2A
1
=
F (z) + F (
z)
z z
2
3
i mi ni ,
i=1
subject to
3
i=1
m2i = 1,
3
n2i = 1,
i=1
3
mi ni = 0.
i=1
Assuming that the principal stresses i are distinct, use the method
of Lagrange multipliers to show that the extreme values occur
when
m3 = n3 = 0,
or m1 = n1 = 0,
or m2 = n3 = 0,
2.10
1
m21 = m22 = n21 = n22 = ,
2
1
m22 = m23 = n22 = n23 = ,
2
1
m23 = m21 = n23 = n21 = ,
2
1 2
,
2
2 3
S=
,
2
3 1
S=
.
2
S=
96
Linear elastostatics
2.11
A=
and
i
f2 (x, y) =
2
eikx sign(k)eky dk =
x
,
+ y2 )
(x2
respectively.
(b) Hence show that (2.6.79) may be written as
!
f!1 (k, y),
!xx = P (k) 1 + y
y
f1
(k, y),
!xy = P!(k)y
x
deduce (2.6.80).
(c) Show from (2.6.82) that
f
v
1
= P!(k) y
(k) 2(1 )f!2 (k) ,
2
x
x
and hence deduce (2.6.83).
Exercises
2.12
97
2 =
(1 + t2 )2
(s2 + y 2 )
((x s)2 + y 2 )2
2,
/y (1 + t2 )
and explain why this integral tends to zero as y tends to zero with
y 1. Finally deduce that, as y 0, (2.6.83) tends to (2.6.85),
where the Hilbert transform is dened by
1
P (s) ds
1 P (s) ds
= lim
.
+
H[P ](x) =
s x
0
sx
2.13
=
f (s) log (s x)2 + y 2 ds.
y
2
2.14
98
2.15
Linear elastostatics
(a) Derive the expression (2.6.97) for the modied Airy stress
function outside a circular underground tunnel.
(b) Prove the identity
e
er
2 ur
+ rerr + 2r
+ ur = r2
2
(E2.1)
2.16
a2
8(1 )A
+ (1 2)r +
cos
r
1 + 2 log a
8(1 )B
2
g(1 2)a sin ,
+
1 + 2 log a
1 2A
1 A
,
+
2
2
r
rr
sin 2
1 A
= g
(r sin H)
,
2
r r
rr = g sin2 (r sin H) +
r
= g cos2 (r sin H) +
2A
,
r2
H + a cos +
cos 2
sin 3
A(a, ) =
2
2
12
+ A cos + B sin ,
A
ga
a
(a, ) =
H + a cos + H cos 2 sin 3
r
2
4
A
B
+ cos + sin ,
a
a
Exercises
99
where A and B are arbitrary constants. Use (E2.1) and the fact that
ur is singlevalued to show that
A = 0,
B = ga3 /4,
so that
r
a2
H 1 + log
+ r cos + H 1 2 cos 2
a
2r
4
2
1
a2
r a
a
+
r(1 2) log
sin +
sin 3 ,
2
a
r
6r3
4r
2
= ga
A
2
2.17
2.18
,
1+
E =
1 + 2
E.
(1 + )2
where k = (k, , m). Show that the Fourier transform of the Greens
tensor in three dimensions is given by
G! = G!ij =
1
+ kkT
I
,
k2
( + 2) k4
(E2.2)
100
2.19
Linear elastostatics
x
.
8 (1 )
4
+
16 x (1 ) x2 x2
x2
2.20
,
x
y
x2
2 x2
2 x2
1
2 x 2 1
2
y 2 2
+
2
= (1 )
,
y
x
y
2 y 2
2 y 2
2
2
x 2 1
y 2 2
1
1
.
=
2
y
x
xy
2 xy 2 xy
xx = (1 )
yy
xy
2.21
2 ui
0
xk xj
for all i and m, where mjk is the usual alternating symbol, and show
that this is equivalent to (ui ) 0. Deduce that
mjk nil
2.22
2 eij
0
xl xk
for all m and n, where eij is the linearised strain tensor. Hence
obtain the compatibility conditions (2.7.5) and show that they are
equivalent to (2.7.6).
Show that the compatibility conditions (2.7.5) can be written in
terms of the stress components as
2
2 yy
2 xy
2
2 xx
+
2
=
+ 2 Tr ( ),
y 2
xy
x2
1 + x2
y
2
xy
yz
xx
xz
2
=
+
+
Tr ( ),
yz
x
y
z
x
1 + yz
the other relations being obtained by permuting x, y and z.
Exercises
2.23
,
dx2
(x2 + 2 )3/2
and
2 dx
2 )3/2
ds
= 2.
(1 + s2 )3/2
Repeat this procedure with log x2 + y 2 + 2 to derive (2.9.7b) and
nd a similar derivation of (2.9.7c).
Show that
2
$
2 # 2
x + y 2 log x2 + y 2 = 8 + 4 log x2 + y 2
+ 2
2
x
y
and deduce that, if A = (1/8) x2 + y 2 log x2 + y 2 , then A satises
(x2
2.24
101
4 A = (x)(y).
2.25
xy
yy
+
=0
x
y
k!
xy + !
yy = 0,
2
2
xy k !yy + k + (!
!xx + 2k!
xx + !yy ) = 0.
2
Hence obtain
!xx
ikf
=
1
!xy =
if
1
!yy =
ikf
1
k 2 + 22
,
k 2 + 2
(k 2 + 2 )2
2
,
k 2 + 2
(k 2 + 2 )2
2
k 2 + 2
(k 2 + 2 )2
102
2.26
Linear elastostatics
Write down the displacement eld for the four point forces depicted
in Figure 2.16. Let a 0 and show that the displacement tends to
u=
1 2 af er
,
1 2 r
3
Linear elastodynamics
3.1 Introduction
This chapter concerns simple unsteady problems in linear elasticity. As noted
in Section 1.10, the unsteady Navier equation (1.7.8) bears some similarity to
the familiar scalar wave equation governing small transverse displacements
of an elastic string or membrane. We therefore start by reviewing the main
properties of this equation and some useful solution techniques. This allows
us to introduce, in a simple context, important concepts such as normal
modes, plane waves and characteristics that underpin most problems in
linear elastodynamics.
In contrast with the classical scalar wave equation, the Navier equation is
a vector wave equation, and this introduces many interesting new properties.
The rst that we will encounter is that the Navier equation in an innite
medium supports two distinct kinds of plane waves which propagate at two
dierent speeds. These are known as P waves and S waves, and correspond
to compressional and shearing oscillations of the medium respectively. Considered individually, both P  and S waves behave very much like waves as
modelled by the scalar wave equation (1.10.9). In practice, though, they
very rarely exist in isolation since any boundaries present inevitably convert
P waves into S waves and vice versa. We will illustrate this phenomenon of
mode conversion in Section 3.2.5 by considering the reection of waves at a
plane boundary.
In twodimensional and axisymmetric problems, we found in Chapter 2
that the steady Navier equation may be transformed into a single biharmonic
equation by introducing a suitable stress function. In Sections 3.3 and 3.4
we will nd that the same approach often works even for unsteady problems,
and pays dividends when we come to analyse normal modes in cylinders and
spheres.
103
104
Linear elastodynamics
2w
= T 2 w,
t2
(3.1.2)
where 2 = 2 /x2 + 2 /y 2 is the twodimensional Laplacian. The threedimensional version of (3.1.2) governs many familiar wave propagation problems including acoustic waves and light waves.
There are two ways in which we can approach (3.1.1) and (3.1.2) mathematically. One possibility is to try to construct a general solution that will
apply whatever mechanism is driving the waves. However, this strategy is
only useful in practice when the general solution can be found explicitly
and also takes a suciently simple form for realistic boundary conditions
to be imposed. Alternatively, we can attempt to describe the solution as
a linear combination of elementary solutions, as in the method of normal
modes in classical mechanics. This second approach can always be applied
in principle and ties in with the rst when we view the general solution as a
superposition of elementary solutions. We therefore begin the next section
by discussing some elementary solutions of (3.1.1) and (3.1.2).
3.2 Normal modes and plane waves
3.2.1 Normal modes
A normal mode of a dynamical system is a motion in which the whole system
oscillates at a single frequency . We can seek normal modes of an elastic
105
(3.2.1a)
(3.2.1b)
where f (x) = f1 (x) + if2 (x). Solutions of the form (3.2.1) are called waves
in the frequency domain.
Substitution of (3.2.1) into (3.1.1) reveals that f (x) must satisfy the ordinary dierential equation
2
d2 f
f = 0,
+
dx2
T
(3.2.2)
where
k=
.
T
(3.2.3)
Thus (3.2.1b), with f (x) given by (3.2.3), is a solution of the wave equation
for any value of the frequency .
Particular values of may be selected by applying suitable boundary
conditions. As a simple example, consider a string xed at its two ends
x = 0 and x = a, so we require w to satisfy w(0, t) = w(a, t) = 0. Clearly f
must likewise satisfy
f (0) = f (a) = 0,
(3.2.4)
106
Linear elastodynamics
(3.2.5), namely that the frequency increases as the tension increases and as
the density decreases.
Now suppose we are additionally given the initial displacement and velocity of the string, say
w(x, 0) = w0 (x),
w
(x, 0) = v(x).
t
(3.2.7)
nx
{An cos(n t) + Bn sin(n t)} ,
sin
(3.2.8a)
w(x, t) =
a
n=1
where the coecients are found by Fourier analysis of the initial data:
a
nx
nx
2 a
2
dx, Bn =
dx.
w0 (x) sin
v(x) sin
An =
a 0
a
n a 0
a
(3.2.8b)
This analysis, although very straightforward, illustrates the following important points that will be useful in analysing more complex models later
in this chapter.
(i) We will always be able to look for timeharmonic solutions of the form
(3.2.1b) whenever the problem we are solving is both linear and autonomous in t: in other words, when there is no explicit dependence on t.
(ii) When looking for normal modes, we might as well have ignored the Re in
(3.2.1b), performed all the calculations, and then taken the real part right
at the end. This approach works because all the usual linear operations,
such as dierentiation with respect to x or t or multiplication by a real
constant, commute with taking the real part. Henceforth, in this chapter,
we will therefore for simplicity follow the convention of assuming the real
part. We note, though, that this would be a very dangerous procedure
were we to be considering a nonlinear model!
(iii) Although normal modes may at rst glance appear to be rather special
solutions of the wave equation (3.1.2), it can be shown that the general
solution, subject to simple homogeneous boundary conditions, may be
written uniquely as a superposition of normal modes.
(iv) When a , it may be shown that (3.2.8) generalises to the halfrange
Fourier transform representation
sin (kx) {A(k) cos(kct) + B(k) sin(kct)} dk,
(3.2.9a)
w(x, t) =
0
where c2 = T / and
2
w0 (x) sin(kx) dx,
A(k) =
0
B(k) =
2
kc
107
(3.2.9b)
We can extend these ideas to the twodimensional wave equation (3.1.2)
by seeking solutions of the form
w(x, y, t) = A(x, y)eit ,
(3.2.10)
2 A + k 2 A = 0,
(3.2.11)
a2
b2
where m and n are arbitrary integers. Thus there is a doubly innite set of
normal modes in this case. In general, there is one innite set of frequencies
for each spatial dimension in the problem.
Next we consider radiallysymmetric vibrations of a circular drum, so that
A is a function of the plane polar distance r and (3.2.11) reads
d2 A 1 dA
+ k 2 A = 0,
+
dr2
r dr
(3.2.13a)
with A(ka) = 0.
(3.2.13b)
This is Bessels equation of order zero, and it is easy to see that A() oscillates for large , because the rst and last terms in (3.2.13) will be approximately in balance. Moreover, for small , we can spot that the rst and
108
Linear elastodynamics
Jn ()
1
0.6
n=0
1
0.4
0.8
0.2
10
15
20
25
30
15
20
25
30
0.2
0.4
Yn ()
n=0
1
2
0.4
0.2
10
0.2
0.4
0.6
0.8
Fig. 3.1 Plots of the rst three Bessel functions Jn () and Yn () for n = 0 (solid),
n = 1 (dashed), n = 2 (dotdashed).
second terms in (3.2.13) both balance and dominate the third term when
A = log . This suggests that the solution is either wellbehaved or logarithmically singular as 0, and this conclusion may easily be conrmed
using Frobenius method. The bounded solution with A(0) = 1 is denoted
by J0 () while the singular solution with A (2/) log is called Y0 (), and
these important functions are plotted in Figure 3.1.
Here, we require the amplitude A to be bounded at the centre of the drum
r = 0 and must therefore choose A = const. J0 (). The condition at r = a is
then satised if
ka = 0,i
(i = 1, 2, . . .),
(3.2.14a)
where 0,1 < 0,2 < are the zeros of J0 (). As indicated in Figure 3.1,
there are an innite number of these zeros and 0,n as n . The
natural frequencies of the drum for radiallysymmetric modes are thus given
by
0,i
i =
a
(n = 1, 2, . . .).
109
(3.2.14b)
Unlike the modes on a string, the natural frequencies here are irrational
multiples of each other. In musical terms, the harmonics are out of tune
with each other, and this gives rise to the characteristic sound of a drum,
which is quite dierent from that of a piano or a violin, for example.
If we drop the assumption of radial symmetry, then the Helmholtz equation becomes
1 2A
2 A 1 A
+
+
+ k 2 A = 0,
(3.2.15)
r2
r r
r2 2
in plane polar coordinates (r, ). Again we impose A = 0 on r = a, and now
we insist also that A must be a 2periodic function of . This restricts us
to seeking separable solutions of the form
(3.2.16)
A(r, ) = f (r) C1 cos(n) + C2 sin(n) ,
where n is an integer, and it follows that f satises
r2
df 2 2
d2 f
+r
+ k r n2 f = 0
2
dr
dr
with f (a) = 0.
(3.2.17a)
df 2
d2 f
+ n2 f = 0
+
2
d
d
with f (ka) = 0.
(3.2.17b)
This is Bessels equation of order n, whose two linearly dependent solutions are denoted Jn () and Yn (); the cases n = 0, 1, 2 are plotted in
Figure 3.1. For any integer n, Yn () is singular as 0, so we must choose
f = const. Jn (). The condition at r = a is then satised if
ka = n,i
(i = 1, 2, . . .),
(3.2.18)
where n,1 < n,2 < are the zeros of Jn (). There are an innite number
of these zeros and n,i as i . The natural frequencies of the drum
are thus given by
2
n,i
=
2 T
n,i
,
a2
(3.2.19)
110
Linear elastodynamics
(3.2.21)
w(x, t) = A exp ik(x ct) ,
where c = ;
k
the quantity c is known as the phase velocity. From (3.2.21) we can easily
see that c represents the speed at which wave crests propagate. We can also
see that the minus sign included by convention in (3.2.20) ensures that the
wave propagates in the positive xdirection when and k are both positive.
We can generalise this approach to the twodimensional wave equation
(3.1.2) by looking for a plane wave solution of the form
w(x, y, t) = A exp i(k1 x + k2 y t) = A exp i k x t , (3.2.22)
where we dene the wavevector k = (k1 , k2 )T . By writing (3.2.22) in the
form
kx
w = A exp i kX t ,
where X =
,
(3.2.23)
k
we observe that it represents a harmonic wave travelling in the direction
of k at speed /k. The phase velocity, at which the wave crests propagate,
is thus given by
k
.
(3.2.24)
c=
k2
When we substitute (3.2.20) into (3.1.1), we nd that the amplitude can
be nonzero only if
2 = k 2 T /.
(3.2.25a)
This dispersion relation tells us how the frequency of any given wave depends
on its wavelength. We can also view (3.2.25a) as an equation for the phase
velocity, namely
(3.2.25b)
c = T /,
111
which tells us that all waves move at the same speed, irrespective of their
wavelength. Waves with this property are called nondispersive, to distinguish them from dispersive waves in which c varies with k.
Similarly, substitution of (3.2.22) into the twodimensional wave equation
(3.1.2) leads to the dispersion relation
2 = k2 T /.
The phase speed
c = c =
T /
(3.2.26)
(3.2.27)
w(x, t) =
R2
A(k) exp i (k x (k)t)
+ B(k) exp i (k x + (k)t) dk, (3.2.28b)
where (k) is given by the dispersion relation (3.2.26) and the amplitude
functions A(k) and B(k) can be found from the Fourier transform of the
initial conditions. By writing the displacement in the form (3.2.28b), we
indicate that this and all the other results in this section apply also to the
wave equation in three space dimensions.
3.2.3 Scattering
Scattering refers generally to the problem of irradiating a target with a
known incoming waveeld and trying to determine the resultant scattered
eld that is the result of the presence of the target. This idea is important,
for example, in tomography, seismology and ultrasonic testing, where we try
to infer, noninvasively, the properties of some inhomogeneities in a bulk
elastic medium by measuring the scattered waveelds that they produce.
112
Linear elastodynamics
w
w
m 2w
(0+, t)
(0, t) =
x
x
T t2
(3.2.29)
ik A 0
as r .
(3.2.31)
r
r
113
k B = 0.
(3.2.34)
2 u = k2 (Ak + Bk) exp i (k x t) ,
grad div u = A k2 k exp i (k x t) ,
2u
2
=
(Ak
+
Bk)
exp
i (k x t) .
t2
Hence the Navier equation (1.7.8) reduces to
2
k2 (Bk) + 2 ( + 2)k2 Ak = 0,
(3.2.35a)
(3.2.35b)
(3.2.35c)
(3.2.36)
and
2 = ( + 2) k2
(3.2.37a)
A=0
and
2 = k2 .
(3.2.37b)
or
114
Linear elastodynamics
The vectorial nature of the Navier equation has thus led to the existence of
two dispersion relations, corresponding to two distinct types of waves.
(i) Pwaves, also known as Primary or Pressure waves, take the form
u = Ak exp i (k x t) ,
(3.2.38a)
where 2 = (+2)k2 /. We recall from Section 1.9 that and +2/3
are both positive, so is real. The phase speed is thus given by
*
+ 2
(3.2.38b)
cp =
.
(3.2.39b)
cs =
(3.2.40)
115
116
Linear elastodynamics
y
P
S
0110 Rigid
1010 boundary
1010 x
1010
1010
1010
this is impossible unless we allow for two reected waves: one P wave and
one S wave. Otherwise, there are not enough degrees of freedom to make
both displacement components zero on x = 0. We therefore seek a reected
wave eld of the form
#
$
cos
exp i kp (x cos + y sin ) t
u r e f = rp
sin
#
$
sin
+ rs
exp i ks (x cos + y sin ) t , (3.2.43)
cos
where ks = /cs . Recall that S waves are transverse so the amplitude is
orthogonal to the wavevector.
From the condition u = 0 on x = 0, we nd that the P wave reection
is specular, meaning that the angle of reection is equal to the angle of
incidence . However, the S wave reection angle satises
sin =
cs
sin
cp
(3.2.44)
and, since cp > cs , it follows that < , as illustrated in Figure 3.2. The
reection coecients are given by
rp =
cos( + )
,
cos( )
rs =
sin(2)
.
cos( )
(3.2.45)
117
rock
2 ,
1 ,
h
coal
rock
2 ,
the case of refracted waves, and the phenomenon of total internal reection
becomes correspondingly more complicated.
We now describe an example that illustrates how P  and S waves can
interact in a practical situation.
3.2.6 Love waves
Let us consider the dynamic version of antiplane strain, the static case of
which
was introduced
T in Section 2.3. If the displacement eld takes the form
u = 0, 0, w(x, y, t) , then w satises
2
2w
w 2w
2
=
+
,
(3.2.46)
w =
x2
y 2
t2
which is just the familiar twodimensional scalar wave equation (3.1.2), with
wave speed cs . This is to be expected, since antiplane strain is volumepreserving, with the displacement depending only on the transverse variables.
The solution strategies presented above for twodimensional waves on a
membrane are all directly applicable to the solution of (3.2.46). Here we
will illustrate them by describing Love waves, which are antiplane strain
waves guided through a particular type of layered medium. As illustrated
in Figure 3.3, the geometry is that of a uniform layer of one material, with
constant thickness 2h, encased inside an innite expanse of a second material. This setup might model, for example, a coal seam in a rock stratum,
with the displacement in the horizontal zdirection. This is in contrast with
118
Linear elastodynamics
the situation depicted in Figure 3.2 where, if we identify the rigid boundary
with a horizontal rock stratum, the displacement is in the vertical direction.
For this reason, the S waves considered in Section 3.2.5 are referred to as
SV waves, while the antiplane S waves described by (3.2.46) are referred to
as SH waves. The density and shear modulus of the coal are denoted by
1 and 1 respectively, while those of the rock are labelled 2 and 2 .
The transverse displacement wi (x, y, t) in either medium satises
2
wi 2 wi
2 wi
2
=
cs i
+
(i = 1, 2),
(3.2.47)
x2
y 2
t2
where csi = i /i (i = 1, 2) are the S wave speeds. On the boundaries of
the seam, the displacements and tractions must be continuous, as shown in
Section 1.10, so that
w1 = w 2 ,
w1
w2
= 2
,
y
y
on y = h.
(3.2.48)
2
k fi = 0
(3.2.50)
fi +
c2si
and, hence, are either exponential or sinusoidal. We suppose that the amplitude of the waves decays at innity so that
f2 = A2 ey
in y > h,
f2 = B2 ey
in y < h,
(3.2.51)
where is real and positive. In such modes, the seam acts as a waveguide,
propagating waves in the xdirection without any energy radiating or leaking out to y as discussed in Section 3.2.3. Substitution of (3.2.51)
into (3.2.50) reveals that k and must be such that
c2s2 k 2 2 = c2s2 2 > 0.
(3.2.52)
(3.2.53)
(3.2.54)
119
(3.2.55)
(3.2.56)
(3.2.57)
For antisymmetric waves, with A1 = 0 and B2 = A2 , the analogous calculation in Exercise 3.7 leads to
1 m cot mh = 2 .
(3.2.58)
120
Linear elastodynamics
= cs2 k
= cs1 k
Fig. 3.4 Dispersion relation between frequency and wavenumber k for symmetric
(solid) and antisymmetric (dotted) Love waves. The phase speed /k is bounded
between cs1 and cs2 .
c2s1
2
2
+k .
4h2
(3.2.60)
Hence, real values of k can only exist if exceeds a critical cuto frequency
cs1 /2h. The existence of a cuto frequency below which waves cannot
propagate without attenuation is a characteristic of all waveguides.
If we do not take the limit 2 /1 , then the dispersion relation
between and k must be obtained by eliminating , m between (3.2.52),
(3.2.54) and either (3.2.57) or (3.2.58), as shown in Exercise 3.7. Crucially,
the values of m satisfying (3.2.57) or (3.2.58) with > 0 are all real, so the
inequalities (3.2.59) continue to hold and there is still a cuto frequency. As
shown in Figure 3.4, there is an innite family of waves, both symmetric and
antisymmetric, and, in each case, the phase speed /k is bounded between
cs1 and cs2 .
121
y
x
y
x y
Now we seek frequencydomain solutions of the form
u = (up ep y + us es y ) exp i (kx t) ,
(3.2.61)
(3.2.62)
where the dispersion relations (3.2.37) for P  and S waves tell us that
2p = k 2
2
,
c2p
2s = k 2
2
.
c2s
(3.2.63)
We assume as usual that k and are real and positive, and that Re(p ) > 0
and Re(s ) > 0, so that the waves propagate without attenuation in the
xdirection while decaying exponentially as y . Evidently such waves
can only exist if the wavenumber k in the xdirection is greater than both
/cs and /cp . The propagation speed c = /k must therefore satisfy
c < cs < cp ,
(3.2.64)
so that Rayleigh waves inevitably travel more slowly than either P waves or
S waves.
After the rather tortuous manipulations of Exercise 3.8, we nd an equation for c in the form
2
1/2
1/2
c2
c2
c2
1 2
=4 1 2
.
(3.2.65)
2 2
cs
cp
cs
It can be shown that this leads to a sextic equation for c but that only one
positive real value for c2 exists and satises (3.2.64). Rayleigh waves are
therefore nondispersive, with a constant propagation speed c.
These results led Rayleigh to suggest that, if a P  or S wave originated at
some source, for example an earthquake, in y < 0, mode conversion would
cause such a wave to propagate indenitely in the xdirection near y = 0.
This hypothesis is born out by seismological observation: the radius of the
earth is large enough for (3.2.61) to be a good approximation and many
earthquake records reveal the arrival of a large Rayleigh wave after rst the
P  and then the S waves have been detected.
3.3 Dynamic stress functions
We now describe some of the ways in which elastic wave propagation can
be reduced to the study of scalar wave equations by the use of stress functions. By taking the divergence and the curl of the dynamic Navier equation
122
Linear elastodynamics
(1.7.8), we obtain
2
2 2
cp (div u) = 0
t2
and
2
2 2
cs (curl u) = 0
t2
(3.3.1)
respectively. This reminds us that irrotational waves (with curl u = 0) propagate at speed cp while equivoluminal waves (with div u = 0) move at speed
cs . It also provides a route for nding solutions of the Navier equation that
often proves handy. We can rst solve the scalar wave equation (3.3.1a) for
div u and then the inhomogeneous wave equation
2
2 2
cs u = c2p c2s (div u)
(3.3.2)
2
t
for the displacement u. Moreover, by crossdierentiating (3.3.1), we obtain
2
2
2 2
2 2
(3.3.3)
cp
cs u = 0,
t2
t2
which we can interpret as a factorisation of the the Navier wave operator
into two scalar operators corresponding to P waves and S waves respectively. This generalises the result that each component of the displacement u
satises the biharmonic equation under static conditions. We recall that in
Chapter 2 we showed that, when there is twodimensional or cylindrical symmetry, the problem can be reduced to a single scalar biharmonic equation
by using a suitable stress function. Now we show briey how the same idea
may be applied to dynamic problems.
First we consider plane strain, in which the second component of the
Navier equation (1.7.8), namely
2
u
2v
2v
+ 2 v,
(3.3.4)
+
2 = ( + )
t
xy y 2
can be rewritten as
2u
2v
1 2v
2
+ (1 2) v 2 2 + 2 = 0.
xy
cs t
y
v=
1 2L
,
2 xy
(3.3.5)
(3.3.6)
for some function L(x, y, t), which is called the dynamic Love function by
analogy with the static Love stress function introduced in Section 2.8.2. It
123
is then readily veried that the rst Navier equation is satised if and only
if L satises the scalar wave equation
2
2
2 2
2 2
(3.3.7)
cp
cs L = 0.
t2
t2
As in the static case, we could have started from the x rather than the
ycomponent of the momentum equation, or indeed any linear combination
of the two. Thus one can dene a oneparameter family of stress functions
analogous to L.
With u and v given by (3.3.6), the stress components may be evaluated
from L using
2L
1 2L
2
+ (1 ) L 2 2
,
(3.3.8a)
xx =
x y 2
cs t
2
1 2L
L
1 2L
2
,
(3.3.8b)
xy =
y y 2
c2s t2
2c2s t2
2L
1 2L
2
yy =
.
(3.3.8c)
L 2 2
x y 2
cs t
For radially symmetric problems, in which the displacement takes the form
u = ur er + uz ez and is dependent only on the cylindrical polar coordinates
(r, z) and t, it is again possible to dene a Love function L(r, z, t) such that
1
L
1 2L
1
1 2L
2
, uz =
(1 2) L 2 2 +
r
,
ur =
2 rz
2
cs t
r r
r
(3.3.9)
and again L satises (3.3.7). In this case, the stress components
by
2L
1 2L
2
,
rr =
L 2 2
z r2
cs t
1 2L
1
1 2L
L
2
rz =
r
L 2 2 2 2 ,
r r r
r
c t
2cs t
s
2
1
1 L
L
zz =
,
r
+ (1 ) 2 L 2 2
z r r
r
cs t
1 L
1 2L
2
=
.
+ L 2 2
z
r r
cs t
are given
(3.3.10a)
(3.3.10b)
(3.3.10c)
(3.3.10d)
We will now use the ideas of Sections 3.23.3 to enumerate some of the rare
congurations for which explicit solutions of the multidimensional elastic
wave equations can be written down.
124
Linear elastodynamics
+
+
=
.
r2
r r
r2
z 2
c2s t2
(3.4.1)
(3.4.2)
(3.4.3)
where
m2 =
2
k2 ,
c2s
(3.4.4)
and we will only consider cases in which m is real. The general solution of
(3.4.3) is
f (r) = A J1 (mr) + B Y1 (mr) ,
(3.4.5)
where J1 and Y1 again denote Bessel functions, while A and B are arbitrary
constants.
To describe waves in a circular cylinder 0 r < a, we must set B = 0
to eliminate the solution Y1 which is unbounded as r 0. If the curved
boundary of the cylinder, r = a, is free of traction, we require
u
u
=0
(3.4.6)
r =
r
r
on r = a, which implies that
m
d 1
J1 (mr)
= J2 (ma) = 0,
dr r
a
r=a
(3.4.7)
the latter equality following from known properties of Bessel functions (Gradshteyn & Ryzhik, 1994, Section 8.4). We deduce that ma must equal one of
125
the zeros 2,n of J2 as dened in Section 3.2.1. There is thus one family of
waves corresponding to each integer value of n.
The dispersion relation
cs + 2
2,n + k 2 a2
(3.4.8)
=
a
tells us that the bar acts as a waveguide, in which the frequency must exceed
the cuto value 2,1 cs /a, while the phase velocity /k is not constant, but
is bounded below by cs . Torsional waves are therefore dispersive, and always
travel faster than S waves. Also, for a cylinder that is nite in the zdirection,
we can use (3.4.8) to nd the normal frequencies.
For cylindrically symmetric longitudinal waves, in which u = 0 and
ur and uz are independent of , the Love stress function L introduced in
Section 3.3 is especially useful. We look for harmonic waves propagating in
the zdirection by setting
L(r, z, t) = f (r)ei(kzt) .
(3.4.9)
2
k2 ,
c2p
m2s =
2
k2 ,
c2s
(3.4.11)
as shorthand.
The general solution of (3.4.10) is
f (r) = A J0 (mp r) + B J0 (ms r) + C Y0 (mp r) + D Y0 (ms r) ,
(3.4.12)
126
Linear elastodynamics
where
f (r) =
1 d
U
(rUr ) + ikUz +
.
r dr
r
(3.4.16)
2 2
cp (div u) = 0,
t2
we nd that f (r) again satises Bessels equation
1
1
f + f + m2p 2 f = 0.
r
r
(3.4.17)
(3.4.18)
(3.4.19)
Having found div u, we can use (3.3.2) to determine the displacement components. To calculate 2 u in polar coordinates, we have to use (1.7.11) and
127
the polar coordinate formulae for div, grad and curl given in Appendix A7.1
to obtain
2
Ur
2U
2
2
Ur +
+ k Ur 2 cos er
u=
r
r2
r
U
2
2Ur
2
+ U +
+ k U 2 sin e
r
r2
r
Uz
1
2
+ Uz +
+ k Uz cos ez ei(kzt) .
(3.4.20)
r
r2
Similarly, by taking the grad of (3.4.16) in polar coordinates, we nd that
f (r)
sin e + ikf (r) cos ez ei(kxt) .
grad div u = f (r) cos er
r
(3.4.21)
Just considering the ez component of (3.3.2), we nd that Uz must satisfy
c2p
1
1
2
(3.4.22)
Uz + Uz + ms 2 Uz = 1 2 ikf,
r
r
cs
with f given by (3.4.19). The bounded solution of this inhomogeneous Bessel
equation is
ikA
Uz = 2
J1 (mp r) + B J1 (ms r),
(3.4.23)
k + m2p
where B is an arbitrary constant. Next we can use (3.4.16) and (3.4.23) to
determine U as
U =
m2p A
r J1 (mp r) ikBr J1 (ms r) rUr Ur .
k 2 + m2p
(3.4.24)
(3.4.25)
128
Linear elastodynamics
t=0
2/5
4/5
6/5
8/5
ikB
J0 (ms r) J2 (ms r)
2ms
mp A
J1 (ms r)
J
, (3.4.27)
(m
r)
J
(m
r)
+C
0
p
2
p
2(k 2 + m2p )
r
(3.4.29)
129
(3.4.30)
One can easily check that the displacement eld (3.4.29) satises u = 0,
which is why (3.4.30) describes only P waves. This in turn implies that we
can write u = for some potential function (r, t), that is
ur =
.
r
1 2 2 2
= 0,
2
r c2p t2
r
r r
(3.4.31)
(3.4.32)
(3.4.33)
c
(r) = 0.
p
t2
r2
(3.4.34)
Thus the combination (r) satises the standard onedimensional wave equation, with wavespeed cp .
If we seek normal modes in which (r, t) = f (r)eit , then (3.4.34) becomes
2
d2
(rf ) = 0,
(3.4.35)
(rf ) +
dr2
c2p
whose general solution is
f=
A sin(r/cp ) + B cos(r/cp )
,
r
(3.4.36)
130
Linear elastodynamics
2 2c2 )
2(c
ur
ur
p
s
+ 2 = eit c2p f +
f
(3.4.37)
rr = ( + 2)
r
r
r
and, with f (r) given by (3.4.36), we nd that the zerostress condition on
r = a reduces to
1
tan(a/cp )
2 a2
= 1
.
(3.4.38)
a/cp
4c2s
By plotting the left and righthand sides versus , one can easily deduce that
this transcendental equation admits a countably innite set of real solutions
for , corresponding to the natural frequencies for radial oscillations of the
sphere.
More complicated formulae for solutions that are not radially symmetric
may be found in Love (1944, Chapter 12). These involve socalled spherical
harmonics, which inevitably arise when one attempts to separate the variables in spherical polar coordinates. We will only discuss torsional waves, in
which the displacement consists of just a rotation about the zaxis, so that
u = u e . Using (1.11.17), one can easily show that this is only possible if
u is independent of , and it follows that div u is identically zero and we
have pure S waves.
The Navier equation (1.11.17) reduces to
u
u
1
1 2 u
1
2 u
r
+ 2
sin
2 2 , (3.4.39)
= 2
c2s t2
r r
r
r sin
r sin
and we seek a separable normal mode in which
u (r, , t) = f (r)g()eit .
(3.4.40)
131
1)
Pn (z) = n
2 n! dz n
and the rst few cases are
P1 (z) = z,
P2 (z) =
3z 2 1
,
2
P3 (z) =
5z 3 3z
.
2
(3.4.44)
f (r) =
,
(3.4.45)
r
where B is an arbitrary constant. Fortunately, the Bessel function Jn+1/2
takes a relatively simple form when n is an integer, namely
)
2 n+1/2 1 d n sin z
n
Jn+1/2 (z) = (1)
z
,
(3.4.46)
z dz
z
and the rst few cases are
)
2 sin z
cos z ,
J3/2 (z) =
z
z
)
2 (3 z 2 ) sin z 3 cos z
,
J5/2 (z) =
z
z2
z
)
2 3(5 2z 2 ) sin z (z 2 15) cos z
.
+
J7/2 (z) =
z
z3
z2
(3.4.47a)
(3.4.47b)
(3.4.47c)
= 0,
(3.4.48)
r
r
and the corresponding condition for f is
f (a)
f (a)
= 0.
a
(3.4.49)
This may be established by writing g() as a power series in cos and then requiring the series
to converge as cos 1.
132
Linear elastodynamics
(3.4.50)
for the natural frequencies . For each integer value of n, (3.4.50) has a
countably innite family of solutions, say n,m (m = 1, 2, . . . ) (see Exercise 3.11). Thus there is a doubly innite set of natural frequencies corresponding to all possible values of m and n, as we would expect for a problem
in two spatial dimensions (r, ).
w
= g(x)
t
at
t = 0,
(3.5.2)
133
c
= (x)(t),
(3.5.4)
t2
x2
where is the Dirac deltafunction dened in Section 2.9 and w is assumed
to be zero for t < 0. The righthand side of (3.5.4) indicates that there
is a jump in w/t at t = 0, and we can therefore restate (3.5.4) as the
initialvalue problem
2
2w
2 w
c
= 0,
in t > 0,
(3.5.5a)
t2
x2
w
= (x)
at t = 0.
(3.5.5b)
w = 0,
t
Substitution of (3.5.5) into the dAlembert solution (3.5.3) leads to the fundamental solution
which is also known as the Riemann function for the onedimensional wave
equation. It may conveniently be recast as
1
R(x, t) =
H(x + ct) H(x ct) ,
(3.5.6b)
2c
134
Linear elastodynamics
R(x, t)
(3.5.7)
w
(x, 0) = 0,
t
(3.5.8)
1
(x + ct) + (x ct) .
2
(3.5.9)
f (s)R(x s, t) ds +
g(s)R(x s, t) ds.
(3.5.10)
w(x, t) =
t
for example noting that d(s)/ds = H(s) + 1 and using Exercise 2.23
135
R(r, t)
Fig.
3.7 The twodimensional fundamental solution R(x, y, t) versus radial distance
r = x2 + y 2 .
outside, and that w/t is zero everywhere except on the the lines x = ct.
Using (3.5.3), it is straightforward to generalise this observation to arbitrary
localised initial data: if f (x) and g(x) are zero outside some interval, say
a < x < b, then for all time w/t is zero except on the two intervals
a ct < x < b ct and a + ct < x < b + ct.
Now let us consider twodimensional elastic waves in a membrane or in
antiplane strain, when there is no general solution like (3.5.1) to guide us.
Nevertheless, the wave equation (3.1.2) can still be solved subject to singular
initial data analogous to (3.5.5), for example by seeking a solution in plane
polar coordinates, as in Exercise 3.13. This reveals that the solution of (3.1.2)
with
w
= (x)(y)
at t = 0,
(3.5.11)
w = 0,
t
is
H (ct r)
,
(3.5.12)
w = R(x, y, t) =
2c c2 t2 r2
where r2 = x2 +y 2 and H again denotes the Heaviside function. The function
R(r, t) dened by (3.5.12) is plotted in Figure 3.7. We see that, like the onedimensional fundamental solution, R(r, t) is zero outside the cone r2 = c2 t2 .
In this case, although the the leading edge of the disturbance at r = ct is
sharp, the retreating edge or tail is not, and R/t is nonzero everywhere
inside the cone.
As in the onedimensional case, we can write the general solution of the
twodimensional wave equation, subject to the initial conditions,
w = f (x, y),
w
= g(x, y)
t
at t = 0,
(3.5.13)
136
Linear elastodynamics
w(x, y, t) =
g(, )R(x , y , t) dd
From this, one can infer that the properties of R noted above also apply to
arbitrary localised disturbances. If the initial data f and g are zero outside
some region, say r < a, then for all time w is zero outside the cone r = a+ct.
However, the smooth tail of R implies that the disturbance persists inside
the cone for all time.
In three dimensions, as typied by (3.4.33), the radiallysymmetric wave
equation takes the form
2 w 2 w
1 2w
,
= 2 w =
+
2
2
c t
r2
r r
(3.5.15)
(3.5.16)
where F and G are two arbitrary scalar functions. This illustrates that, as
in onedimensional waves, a localised radiallysymmetric disturbance propagates outwards with a sharp leading edge and a sharp tail. Huygens principle
states that this happens for radiallysymmetric waves in any odd number of
space dimensions, but not the twodimensional case considered above.
We can use the general radiallysymmetric solution (3.5.16) to construct
the threedimensional Riemann function
R(x, y, z, t) =
(r ct)
.
4cr
(3.5.17)
This enables us to write the general solution to the threedimensional initialvalue problem in a form analogous to (3.5.14), known as the retarded potential solution (Ockendon et al., 2003, p. 121). Remarkably, a retarded potential solution can also be written down for the full threedimensional Navier
equation, as shown in Love (1944, Section 211). However, we emphasise that
these solutions only apply in an innite medium, and the presence of any
boundaries usually makes it impossible to construct a Riemann function.
137
3.5.3 Characteristics
The concept of characteristics for hyperbolic partial dierential equations
gives fundamental insight into wave propagation. There are several ways of
thinking about characteristics mathematically but, for our purposes, it is
most convenient to generalise (3.5.8), (3.5.9) and base our denition on the
solution of the equations that is zero for t < t0 and driven by a localised
forcing at x = x0 , t = t0 . As in (3.2.28b), such a forcing will excite a
superposition of plane waves of the form u = aei(kx(k)t) , where
k x t = k x0 t0 .
(3.5.18)
(3.5.19)
x x0 2 = c2 (t t0 )2
(3.5.20)
has normal (k, ) in (x, t)space and is hence the envelope of (3.5.18) as k
varies. This is illustrated in two dimensions in Figure 3.8 and Exercise 3.14
gives a derivation of this result. The wavefronts are circles or spheres in two
or three dimensions respectively.
For the unsteady Navier
equation,
Section 3.2.4 shows that (3.5.19) generalises to 2 c2s k2 2 c2p k2 = 0 and hence the characteristic surface
through x = y = z = t = 0 is a twosheeted cone
2
(3.5.21)
x + y 2 + z 2 c2p t2 x2 + y 2 + z 2 c2s t2 = 0,
as illustrated in Figure 3.9 for just two spatial dimensions. This is in stark
contrast with, for example, the threedimensional scalar wave equation or
with electromagnetic waves in isotropic media, where Maxwells equations,
although also vectorial, admit only one wavespeed, namely the speed of
light.
138
Linear elastodynamics
2
2 2
x +y =c t
k1 x + k2 y = t
x2 + y 2 = c2s t2
x2 + y 2 = c2p t2
y
Fig. 3.9 The twosheeted characteristic cone for the Navier equation.
139
(a)
(b)
increasing t
x
w
(c)
x
(d)
Fig. 3.10 The response of a string to a point force moving at speed V : (a) V = 0,
(b) 0 < V < c, (c) V = c, (d) V > c.
a dramatic change in behaviour when the source velocity passes through one
of the speeds of sound in the material.
As a rst illustration, consider the onedimensional problem of an elastic
string on which a point force of magnitude P is imposed at the moving point
x = V t. This situation is modelled by the equation
2w
2w
T
= P (x V t),
t2
x2
(3.6.1a)
where, as usual, is the Dirac deltafunction. Assuming that the string starts
from rest with zero displacement, we impose the initial conditions
w=
w
=0
t
at
t = 0.
(3.6.1b)
Pc
(c
+
V
)x
ct
+
(c
V
)x
+
ct
2cx
V
t
, (3.6.2)
4T (c2 V 2 )
140
Linear elastodynamics
how long we wait, the solution never settles down to a travelling wave in
which w is just a function of x V t.
In two or three space dimensions, it is more dicult to obtain solutions to
initialvalue problems analogous to (3.6.1), but it is now sometimes possible
to nd travelling wave solutions. For example, an elastic membrane subject
to a moving point force is modelled by the equation
2w
T 2 w = P (x V t)(y).
t2
(3.6.3)
y
T
where c = T / and the Mach number is dened by
M=
V
.
c
(3.6.5)
When M < 1, so the force is moving subsonically, (3.6.4) is an elliptic partial dierential equation, and w is thus just a multiple of the
Greens function, as in Section 2.9. By changing variables again to X = / 1 M 2 and
y, so that (3.6.4) is transformed to Poissons equation (see Exercise 3.16),
the solution is easily found to be
P
(x V t)2
2
w=
(3.6.6)
log
+ y + const..
1 M2
2T 1 M 2
Notice that the displacement is singular (as expected) as the point force
(V t, 0) is approached, and that the entire membrane is disturbed by the
force. Indeed the displacement grows in amplitude as (x, y) approaches innity, a feature typical of twodimensional elliptic problems. Had we tried to
solve an initialvalue problem, we would have found that the inuence of the
point force spreads out at speed c, eventually invading the whole membrane,
as shown schematically in Figure 3.11(a).
If M > 1, so the force is moving supersonically, then (3.6.4) is a hyperbolic
partial dierential equation. In fact, it is analogous to the onedimensional
wave equation (3.5.4), and we can infer the solution directly from (3.5.6) as
P
H y
H y+
,
< 0,
M2 1
M2 1
w = 2T M 2 1
0,
> 0.
(3.6.7)
(a)
141
increasing t
x
(b)
y
x
P
,
w = T M2 1
0,
xVt
Vtx
<y<
,
2
M 1
M2 1
otherwise,
(3.6.8)
1
2
angle is tan
1/ M 1 sin1 (1/M ). This may again be visualised
by plotting wavefronts caused by the moving force, as in Figure 3.11(b).
Since the force moves supersonically, it overtakes the wavefronts generated
by itself. A similar wave pattern occurs behind a supersonic aircraft and is
responsible for the notorious sonic boom.
As was the case with (3.6.2), the subsonic (3.6.6) and supersonic (3.6.8)
solutions both grow without bound as the Mach number approaches unity.
This clearly violates the assumption of small displacements under which the
model (3.6.3) for an elastic membrane is valid. To describe the behaviour
as M passes through unity it is therefore necessary to incorporate nonlinear
eects into (3.6.3).
142
(a)
Linear elastodynamics
(b)
(c)
Fig. 3.12 P wave (dashed) and S wave (solid) fronts generated by a point force
moving at speed V in plane strain with (a) 0 < V < cs , (b) cs < V < cp , (c) V > cp .
Now let us apply the same ideas to plane strain driven by a moving point
force in the xdirection, which is described by the equations
2
2
u 2u
u
2v
2u
+
+ 2 = P (x V t)(y),
2 ( + )
t
x2
xy
x2
y
(3.6.9a)
2v
2v
2u
2v
2v
= 0.
(3.6.9b)
+ 2
+
2 ( + )
t
xy y
x2
y 2
These may be conveniently cast in terms of the Love stress function L dened
by (3.3.6), which must now satisfy
P
1 2
1 2
2
2
2 2
2 2 L=
(x V t)(y).
(3.6.10)
cp t
cs t
1
We again seek a travelling wave solution in which L is a function of
= x V t and y, in which case L must satisfy
2
2
P
2
2
2
2
()(y),
(1 Ms ) 2 + 2 L =
(1 Mp ) 2 + 2
y
1
(3.6.11)
where we now dene two Mach numbers
Mp =
V
,
cp
Ms =
V
.
cs
(3.6.12)
Exercises
143
If Mp < Ms < 1, then the motion is subsonic with respect to both P and S waves, and (3.6.11) is an elliptic equation similar to the biharmonic
equation. In this case, as shown in Figure 3.12(a), all waves propagate out
to innity.
If Mp < 1 < Ms , then the motion is subsonic with respect to P waves,
which therefore propagate to innity, but supersonic with respect to S waves, which are thus conned to a Mach cone, as illustrated in Figure 3.12(b).
In this transonic case, (3.6.11) is of mixed hyperbolicelliptic type.
If Ms > Mp > 1, then the motion is entirely supersonic, so both P waves
and S waves are conned to Mach cones. Notice the analogy with the twosheeted characteristic cone shown in Figure 3.9.
3.7 Concluding remarks
The ideas expounded in this chapter describe the fundamentals of many
models that are regularly used to simulate elastic waves in situations ranging
from tomography to oil exploration. In particular, nondestructive testing of
solid structures is frequently carried out by examining elastic wave reection,
refraction and diraction. In these studies, it is often the parameters and
(which will usually be functions of position) that have to be determined from
the response of a material to elastic waves, and hence they are called inverse
problems. A classic example is the question, rst posed by Kac (1966), can
one hear the shape of a drum? ; in other words, if we know all the natural
frequencies of a membrane, can we determine its boundary?
The main complication that arises when considering the Navier equation
as opposed to the scalar wave equation is the existence of two distinct wavespeeds cp and cs . Many of the problems analysed in this chapter are simplied signicantly if we consider incompressible materials (cf Section 1.8), in
which cp and div u = 0, so that A = 0 in (3.2.34) and only S waves
propagate at nite speed.
There are many other fascinating and practically important problems in
the theory of elastic waves that fall beyond the scope of this chapter. These
include niteamplitude waves, to be discussed in Chapter 5, and wave propagation through inhomogeneous media, to which we will return briey in
Chapter 9.
Exercises
3.1
Let a mass m be attached to a string with line density and tension T . Using the boundary conditions (3.2.29) at this point, show
144
Linear elastodynamics
3.3
imk/2
,
1 imk/2
cT =
1
.
1 imk/2
Consider a wave with unit amplitude travelling in the positive xdirection on a string under tension T , whose density changes from
to + at x = 0. From continuity of the displacement and the force
at x = 0, show that
jump in density transmits a wave with am this
plitude cT = 2/ 1 + /+ and reects a wave with amplitude
cR = cT 1.
The Helmholtz equation in plane polar coordinates is
1 2A
2 A 1 A
+ 2 2 + k 2 A = 0.
+
2
r
r r
r
To study the far eld scattered from a localised obstacle, write
r = R/, where 0 < 1, and use the ansatz A A0 eiku(R)/ ,
where A0 and u are independent of . Show that
A0
du 2
A0
+
= 0,
= 1,
and
2
dR
R
R
and deduce that outwardpropagating waves take the form
A A+ ()r1/2 eikr
3.4
3.5
as
r .
Show that, for any vector a and nonzero vector k, there exists a
unique vector B and scalar A such that a = Ak+Bk and kB = 0.
[Hint: calculate a k and ak.]
Suppose that an elastic medium occupies the halfspace x < 0 and
that the face x = 0 is held xed. A plane Swave is incident from
x with
sin
uinc =
exp{iks (x cos + y sin ) it},
cos
where ks = /cs . Show that the reected wave takes the form
sin
exp{iks (x cos + y sin ) it}
uref = rs
cos
cos
exp{ikp (x cos + y sin ) it},
+rp
sin
Exercises
145
3.6
cos( + )
,
cos( )
rp =
sin(2)
.
cos( )
= 2
=0
+
+
+
y
x
x
x y
to be satised across x = 0. If an incident S wave as in Exercise 3.5, with wavevector (cos , sin ), generates reected and
transmitted S  and P waves with wavevectors ( cos , sin ),
( cos , sin ) and (cos + , sin + ), (cos + , sin + ) respectively,
show that
sin
sin
sin +
sin +
=
=
=
.
cp
cs
cp+
cs+
3.7
2 tan h
2
2
k
k
.
=
1
2
c2s1
c2s1
c2s2
Write s = h 2 /c2s1 k 2 , = cs1 /cs2 , = 1 /2 to obtain
the parametric representation
s
2 + 2 tan2 s
h
=
s 1 + 2 tan2 s, kh =
,
cs2
1 2
1 2
(E3.1)
where n s < (n + 1/2) and n is an integer.
(b) Show that antisymmetric Love waves satisfy the condition
(3.2.58) and deduce that their dispersion relation is given
parametrically by (E3.1) with tan s replaced by cot s and
(n 1/2) s < n.
146
Linear elastodynamics
cs2
cg
cs1
k
Fig. 3.13 Group velocity cg = d/dk versus wavenumber k for symmetric (solid)
and antisymmetric (dotted) modes of Love waves.
3.8
Exercises
3.9
147
1 2
2U
1 2U
2U
+
,
xy
y 2
2c2s t2
xy 2c2s t2
2U
1 2U
yy =
,
x2
2c2s t2
for two functions U(x, y, t) and (x, y, t). Deduce from the twodimensional Navier equation and constitutive relations that U and
may be chosen to satisfy
2
2
2
2 2
2 2
2 2
cs =
cp
cs U = 0.
t2
t2
t2
3.10
2
2
U=0
c
p
s
t2
x2
t2
x2
is
U = Fp (x cp t) + Gp (x + cp t) + Fs (x cs t) + Gs (x + cs t) ,
3.11
3 2
tan
12 2
=
12 5 2
tan
75 8 2
=
75 33 2 + 4
when n = 1,
when n = 2,
when n = 3,
1,m m
2,m
3,m
as m .
148
3.12
3.13
Linear elastodynamics
Show that, after a suitable change of independent variables, the onedimensional wave equation (3.1.1) can be written as
2w
= 0.
w
=
+
c2 t2
r2
r r
admits similarity solutions of the form
1 r
w(r, t) = f
,
t
t
where f () satises
2
d
2 df
2
c
f = 0.
d
d
(E3.2)
0 < c,
= ct
2
c t2 r 2
0
and deduce that
, 0 r < ct,
2
w(r, t) = 2c c t2 r2
0,
r > ct,
satises (E3.2) subject to
w
= (x)(y)
at t = 0.
t
The characteristic surface through the origin for the twodimensional
wave equation (3.1.2) is dened to be the envelope of the planes
(3.5.18) subject to the dispersion relation (3.5.19). Show that it is
equivalent to the envelope of the planes
w = 0,
3.14
Exercises
149
3.15
cki
(t t0 )
k
2T (c + V )
c+V
Using the fact that, from (2.9.7a),
d2
z = 2(z),
dz 2
deduce that the general solution of (E3.3) is
Pc
(c + V ) + (c V ) + F () + G(),
w=
2
2
4T (c V )
where F and G are arbitrary functions. Transform back to (x, t)
and apply the initial conditions to obtain the solution (3.6.2) when
c>V.
When V > c, write x V t = X and show that (3.6.1a) becomes
3.16
V 2 c2
2w
2w
2w
P c2
+
(X).
2V
=
X 2
Xt
t2
T
P log X 2 + y 2
+ const..
w=
2T 1 M 2
4
Approximate theories
4.1 Introduction
So far in this book we have been considering linear elasticity only for very
simple geometries such as cylinders, spheres and halfspaces. In this chapter,
we will consider more general solids under the restriction that they are thin
and the equations of elasticity can consequently be simplied. A familiar
example that we have already encountered is the wave equation governing
the transverse displacements of a thin elastic string, and we will revisit this
model below in Section 4.3.
A string is characterised by its inability to withstand any appreciable shear
stress, so its only internal force is a tension acting in the tangential direction.
Similarly, a membrane is a thin, nearly twodimensional structure, such as
the skin of a drum, which supports only inplane tensions. A bar, on the other
hand, is a nearly onedimensional solid that can be subject to either tension
or compression. However, many thin elastic bodies also have an appreciable
bending stiness and therefore admit internal shear stress as well as tension.
A familiar example is a exible ruler, which clearly resists bending while
deforming transversely in two dimensions, and is known as a beam. A thin,
nearly onedimensional object which can bend in both transverse directions,
such as a curtain rod or a strand of hair, will be referred to as a rod. On
the other hand, a nearly planar elastic structure with signicant bending
stiness, for example a pane of glass or a sti piece of paper, is called a
plate. Finally, a shell is a thin, nearly twodimensional elastic body which is
not initially planar, for example a pingpong ball or the curved panel of a
car.
In this chapter, we will show how models for all these structures can be
derived using net force and moment balances combined with plausible constitutive relations. We will soon nd, however, that it is dicult to enunciate
150
151
x
T (x, t)
T (x + x, t)
x
Fig. 4.1 The forces acting on a small length x of a uniform bar.
all the physical assumptions that are needed. To do this requires a more systematic asymptotic analysis, and it will be deferred to Chapter 6. Here we
will focus initially on small displacements and hence obtain linear governing equations. However, we will nd that the physicallybased approach also
allows us to look at the eect of geometric nonlinearity, when the displacements are large but the strain is small. The nonlinear models that result
exhibit interesting nonuniqueness properties that help us to understand
important physical phenomena such as buckling.
(4.2.2)
152
Approximate theories
u/x, by analogy with Hookes law. Motivated by the exact static solution
obtained in Section 2.2.3, we postulate the constitutive relation
u
,
(4.2.3)
x
where E again denotes Youngs modulus. By substituting (4.2.3) into (4.2.2),
we nd that u(x, t) satises the wave equation
T = EA
2u
2u
=
,
x2
t2
(4.2.4)
which says that longitudinal waves in the bar travel with speed E/.
Since E < + 2, these waves travel slower than the longitudinal Pwaves
of Section 3.2.4. Typical boundary conditions are that u or the axial force
EAu/x should be prescribed at each end of the bar and, of course, u and
u/t need to be given at t = 0.
We should point out that various implicit assumptions underly the derivation given above. We have assumed, for example, that the longitudinal displacement u is uniform in each crosssection of the bar, and that the constitutive relation (4.2.3) holds, although its derivation in Section 2.2.3 was
performed only under static conditions. Neither of these assumptions is exactly true in practice, reecting the fact that solutions of (4.2.4) are not
exact solutions of the Navier equations. Intuitively, we expect (4.2.4) to be
a good approximate model when the bar is thin and the displacement is
small, specically if u A L, where L is the length of the bar. However, this simple example illustrates the diculty of assessing the validity of
such an ad hoc model. We will show in Chapter 6 how such models can be
derived from the underlying continuum equations in a more rigorous manner
that allows the accuracy to be carefully estimated.
4.3 Transverse displacements of a string
Before deriving a model for the transverse displacements of a beam, let us
remind ourselves briey of the corresponding derivation for an elastic string,
which is characterised by the fact that the only internal force that it can
support is a tension T acting in the tangential direction. Assuming that
gravity g acts transversely to the string, as illustrated in Figure 4.2, the net
force experienced by a small segment of undisturbed length x is
x+x
0
cos
x,
+
T
g
sin x
where denotes the line density.
153
T (x + x, t)
T (x, t)
w
x
x
Fig. 4.2 The forces acting on a small length x of an elastic string.
154
Approximate theories
N (x + x, t)
T (x + x, t)
M (x, t)
g
M (x + x, t)
T (x, t)
N (x, t)
Fig. 4.3 The forces and moments acting on a small segment of an elastic beam.
for convenience N is dened to act perpendicular to the xaxis. Again considering small, purely transverse displacements, we obtain the equations
2 w N
2w
T
= 0,
T 2 +
g = 2 .
(4.4.1)
x
x
x
t
Now our task is to relate N to w, and we start by performing a moment
balance on the segment shown in Figure 4.3. When doing so, we recall the
suggestion from Section 2.6.7 that each section of the beam exerts on its
neighbours a bending moment M about the yaxis, as well as the tangential
and normal force components T and N . Conservation of angular momentum
about the point x leads to the equation
M (x + x, t) + M (x, t) + xN (x + x, t) = g
x2
2
+ I 2 ,
2
t
(4.4.2)
where w/x is once again the small slope of the beam and the moment
of inertia of the segment about its end is given by I = (x)3 /3. When we let
x 0, the righthand side of (4.4.2) is thus negligible and we are left with
M
+ N = 0.
(4.4.3)
x
Finally, we need a constitutive relation for M . It is a practical experience,
for example when bending a exible ruler, that applying an increasing moment to a beam results in an increasing curvature. For small displacements,
this suggests a constitutive relation of the form
M = B
2w
,
x2
(4.4.4)
(a)
(b)
155
(c)
Fig. 4.4 The end of a beam under (a) clamped, (b) simply supported and (c) free
conditions.
2xz
u
,
(4.4.5)
u = v =
xx = Ez.
2yz
2
2
2
2
x y + z
w
We identify the transverse displacement of the beam with x2 /2, neglecting
the small y 2 and z 2 on account of the slenderness of the beam. The net
moment generated by this stress eld on the crosssection A is given by
zxx dydz = E
z 2 dydz,
(4.4.6)
M=
A
where
z 2 dydz
I=
(4.4.7)
2w
2w
4w
+ T 2 g = 2 .
4
x
x
t
(4.4.8)
It is interesting to note that the real and imaginary parts of the solution of the Schr
odinger
equation
2
=
t
2m x2
satisfy (4.4.8) with T = g = 0, /2m = EI/.
i
156
Approximate theories
n=2
0.5
0.5
0.2
0.4
0.6
0.8
0.2
0.4
0.6
n=3
0.5
0.8
0.2
0.5
0.5
0.5
1
1
1
157
0.4
0.6
0.8
Fig. 4.5 The rst three buckling modes of a clamped elastic beam (with a = 1 and
L = 1).
d4 w
d2 w
+
P
= 0,
dx4
dx2
(4.4.13a)
at
x = 0, L.
(4.4.13b)
(4.4.14)
nx
L
(4.4.15)
where A is a constant.
The rst three such eigenfunctions are shown in Figure 4.5; these are
the shapes that a beam can adopt as it buckles under a suciently large
compressive force. The amplitude A appears to be indeterminate, although
we might expect it to be a smoothly increasing function of the applied force.
This is a deciency of the model that will be addressed in Section 4.9.
158
Approximate theories
(4.4.17)
=
1+
.
(4.4.18)
k
T
Evidently this reduces
to the dispersion relation for a string if the waves are
long, with k T /EI. However, the bending stiness becomes increasingly
important, and the waves become increasingly dispersive, as the wavelength
decreases. This observation can be used to explain, for example, why the
harmonics of a piano string with small but nonzero bending stiness are
slightly sharp compared with the fundamental (see Exercise 4.1).
For a beam under compression, with P = T > 0, wavelike solutions of
(4.4.8) exist only if k 2 > P/EI. For smaller values of k, (4.4.17) leads to
complex values of , corresponding to solutions that grow exponentially in
time. It
follows that the beam is unstable to waves with wavelength greater
than 2 EI/P , and we will investigate this further in Section 4.9.3.
159
Nz
Ny
T
x
Fig. 4.6 The internal force components T , Ny and Nz in a thin elastic rod.
y
Mz
My
Mx
Fig. 4.7 Crosssection through a rod showing the bending moment components.
2w
2 w N
+
,
+
f
=
x2
x
t2
(4.5.1)
T
where N = Ny , Nz and f is the body force per unit length.
In general we must now allow for three components of the bending moment
M = (Mx , My , Mz )T . As shown in Figure 4.7, My is the moment about the
yaxis previously referred to simply as M in Section 4.4, while Mz represents
bending in the (x, y)plane. We will begin by neglecting the rst component
Mx , which corresponds to twisting of the rod about its axis. A balance of
moments in the y and zdirections analogous to (4.4.3) then leads to
My
Nz = 0,
x
Mz
+ Ny = 0.
x
(4.5.2)
160
Approximate theories
2xz
2xy
z
+ y x2 + y 2 z 2 ,
u=
2yz
2
2
2yz
x2 y 2 + z 2
xx = z Ez y Ey.
(4.5.3a)
(4.5.3b)
The two constants y and z represent bending in the (x, y)plane and the
(x, z)plane respectively. The moments about the y and zaxes generated
over the crosssection A by this stress eld are given by
zxx dydz,
Mz =
yxx dydz.
(4.5.4)
My =
A
Iyz
Izz
=
A
y 2 yz
yz z 2
(4.5.5)
dydz
(4.5.6)
EI
+
f
=
.
(4.5.7)
x2
x4
t2
Note that we can choose the y and zaxes such that the symmetric positive
denite matrix I is diagonal; these correspond to the directions in which the
resistance to bending is maximised and minimised. It follows that the y and
zcomponents of (4.5.7) are simply two decoupled beam equations for v and
w; note, though, that this decoupling does not occur if the rod crosssection
is nonuniform in x. By analogy with (4.4.8), we therefore expect to have to
impose initial conditions on w and w/t as well as two boundary conditions
each on v and w at either end of the rod. The conditions corresponding
to clamped, simply supported and free ends are obvious generalisations of
(4.4.9), (4.4.10) and (4.4.11).
Examples of values of Iyy and Izz for various crosssection shapes with
respect to principal axes are given in Figure 4.8. These are crucial pieces
of information to bear in mind when designing beams or rods in practice.
By considering (4.5.4), we observe that Iyy measures resistance to bending
about the zaxis, while Izz represents resistance to bending about the yaxis.
For example, if one considers the bending of a long strip of paper with
T
161
(b)
(a)
Izz = a3 b/12
a
Iyy = ab3 /12
b
(c)
Fig. 4.8 Examples of crosssections in the (y, z)plane and their bending stinesses.
(4.5.8)
(4.5.9)
Mx = R
x
162
Approximate theories
between the torque and the twisting gradient, where R again denotes the
torsional
rigidity. We thus nd that satises the wave equation, with wave
speed R/ (Iyy + Izz ). For the simple case of a circular crosssection, it is
easy to show that
a4
a4
,
R=
(4.5.10)
2
2
and hence that the wave speed is equal to cs . This reproduces the wave speed
of the rst family of torsional waves found in Section 3.4.1; that is, (3.4.8)
with n = 0 and 2,0 = 0: the theory presented here assumes each section
deforms uniformly and thus fails to discern the higher modes.
Iyy + Izz =
Txy
Tyy
+
=0
x
y
(4.6.2)
w
w
Nx + Txx
+ Txy
x
y
2w
w
w
+
Ny + Txy
+ Tyy
g = 2 , (4.6.3)
y
x
y
t
For consistency with the notation of this chapter, x should be substituted for z in the results
of Section 2.4.
163
Ny
Tyy
Txy
Txx
y
Tyx
Nx
Tyx
Nx
Txx
y
Txy
x
Tyy
Ny
Myy
Myx
+
Nx = 0.
x
y
(4.6.4)
1
u=
2
2(ax + by)z
2(bx + cy)z
2
2
2
ax + 2bxy + cy + (a + c)z /(1 )
(4.6.5)
164
Approximate theories
z
Mxy
Myx
y
Myy
Mxx
x
E(a + c)z
,
1 2
xy =
Ebz
,
1+
yy =
E(a + c)z
.
1 2
(4.6.6)
Mxy =
h/2
h/2
h/2
zyy dz,
Myy =
h/2
h/2
zxy dz,
(4.6.7b)
where h is the plate thickness. Hence, Mij are the components of the tensor
h/2
0 1
xx xy
z dz.
1 0
h/2 xy yy
By substituting (4.6.5) and (4.6.6) into (4.6.7) we obtain the relations
2w
Eh3
,
12(1 + ) xy
2
w
Eh3
2w
=
+ 2 ,
12 (1 2 ) x2
y
2w 2w
Eh3
.
=
+
12 (1 2 )
x2
y 2
Mxx = Myy =
Myx
Mxy
(4.6.8a)
(4.6.8b)
(4.6.8c)
165
It is reassuring that, in the notation of this section, the plane strain result
(2.6.73) implies that
Myx =
h3 2 w
,
6(1 ) x2
(4.6.9)
.
y 2
x2
(4.6.10)
Thus, without the imposition of a transverse bending moment, the plate will
automatically bend upwards in the ydirection when we attempt to bend it
downwards in the xdirection. Unfortunately, it is dicult to observe this
eect using a piece of paper because Poissons ratio is close to zero. By
substituting (4.6.10) into (4.6.8b), we nd in this case that
Myx =
Eh3 2 w
,
12 x2
(4.6.11)
which is consistent with (4.4.4) and (4.4.7), since h3 /12 is the moment of
inertia per unit length in the ydirection. Hence we see that the constant of
proportionality between curvature and bending moment in one dimension
depends on whether the transverse curvature or the transverse moment are
set to zero. This is analogous to the biaxial strain examples considered in
Section 2.2.4, where the eective onedimensional elastic modulus depends
on whether zero stress or zero strain is imposed in the transverse direction.
When we use the constitutive relations (4.6.8) in (4.6.3), we are left with
the plate equation
Txx
2w
2w
2w
2w
4
+
T
+
2T
D
w
g
=
,
xy
yy
x2
xy
y 2
t2
(4.6.12)
Eh3
12 (1 2 )
(4.6.13)
166
Approximate theories
2w
.
t2
(4.6.14)
w
2w
= 0,
+
2
n
n
(4.6.17)
where is the curvature of the boundary (see Exercise 4.8). Only for a
straight boundary, then, can we apply the obvious generalisation of (4.4.10)
in which w and its second normal derivative are zero.
Finally we consider the boundary conditions to be imposed at a free edge,
where a horizontal force balance evidently requires that T = 0. If, for the
sake of argument, the boundary is at x = 0, then, for there to be zero applied stress, we are apparently led to the conditions Nx = Mxx = Myx = 0.
However, it is to impossible to impose three boundary conditions on the
biharmonic problem (4.6.14)! This conceptual diculty caused great consternation to the pioneers of plate theory (see Love, 1944, Article 297), and
167
Myx
Mxx
2
=0
x
y
(4.6.18)
3w
3w
+
(2
)
= 0.
x3
xy 2
(4.6.19)
= 0,
x2
y 2
By using (4.6.4) and (4.6.8), we may also rewrite the second of (4.6.18) as
Nx =
Mxx
,
y
(4.6.20)
which shows that the shear stress Nx as calculated from plate theory does
not in general vanish at a free edge! Note also that a discontinuity in the
bending moment at a corner in the boundary results in a deltafunction in
the shear stress, corresponding to a point force in the transverse direction
(see Timoshenko & WoinowskyKrieger, 1959). This suggests the existence
of a threedimensional SaintVenant region in the vicinity of the corner.
It is also possible to obtain (4.6.18) as the natural boundary conditions
associated with minimising the strain energy in the plate. Although it is
beyond the scope of this chapter to derive this systematically, we note that,
in the exact solution (4.6.5) where tension is neglected, the nonzero stress
components are given by (4.6.6) and the corresponding strain components
by
exx = az,
exy = bz,
eyy = cz.
(4.6.21)
168
Approximate theories
This suggests that, more generally, the net strain energy due to bending will
take the form
2 2
2 2
2w 2w
w
D
dxdy,
U=
w 2(1 )
2
2
2
x y
xy
(4.6.23)
where the integral is taken over the region in the (x, y)plane occupied by
the plate. The integrand in (4.6.23) shows that the total energy depends on
a combination of the mean curvature
2 w = 1 + 2
and the Gaussian curvature
2w 2w
x2 y 2
2w
xy
(4.6.24)
2
= 1 2 ,
(4.6.25)
where 1 and 2 are the principal curvatures of the surface z = w(x, y).
We will soon discover that these geometric quantities are fundamental to all
theories of plates and shells.
Exercise 4.10 demonstrates that, if we minimise (4.6.23) using the calculus
of variations, then w must satisfy the biharmonic equation, which agrees
with the steady version of (4.6.14), when tension and gravity are neglected.
In addition, Exercise 4.10 shows that (4.6.19) are the natural boundary
conditions for the minimisation.
4.6.3 Simple solutions of the plate equation
Let us start by considering steady solutions of (4.6.14) with T = 0, that is
g
(4.6.26)
4 w = .
D
Solutions of (4.6.26) describe the sagging of a plate under gravity. We will
consider both clamped and simply supported boundary conditions; it is impossible to solve (4.6.26) with entirely free boundaries since some boundary
tractions are needed to balance gravity.
For example, to describe the sagging of an elastic disc r < a, we can
easily solve (4.6.26) in cylindrical polar coordinates. With clamped boundary
conditions w = dw/dr = 0 at r = a, we nd that
2
g 2
w=
(4.6.27)
a r2 .
64D
Alternatively, the simply supported boundary conditions (4.6.17) at r = a
lead to the solution
5+ 2
g 2
2
2
a r .
(4.6.28)
a r
w=
64D
1+
0
0.0002
0.0004 w
0.0006
0.5
0.4
0.3
0.2
0.1 y
0
0.2
0.4
x 0.6
169
0.8
10
Fig. 4.11 The displacement (4.6.29) of a simply supported rectangular plate sagging
under gravity, with a = 1, b = 1/2, g/D = 1.
Note in particular that the reduced rigidity aorded by the simple support
compared with the clamped plate means that (4.6.28) gives a larger displacement at the centre of the disc by a factor of (5 + )/(1 + ) compared
with (4.6.27).
Next let us consider the sag of a rectangle 0 < x < a, 0 < y < b. For
clamped boundary conditions, we encounter exactly the same diculty as
in Section 2.6.6. The nonorthogonality of the functions encountered in separating the variables in the biharmonic equation with boundary conditions
on w and w/n makes any analytical solution impractical. However, the
simply supported boundary conditions w = 2 w/n2 = 0 are now physically realistic and it is straightforward to separate the variables and obtain the solution as a sum of mutually orthogonal trigonometric functions,
namely
16g sin (mx/a) sin (ny/b)
w= 6
.
(4.6.29)
D
mn (m2 /a2 + n2 /b2 )2
m,n odd
170
Approximate theories
)
+
x2
x2
y 2
xy
xy
2
2
2
w
w
+ g = 0. (4.6.30)
+ 2
+ 2 D
y
y 2
x
Given D(x, y), (4.6.30) is a simple generalisation of the biharmonic equation
which is straightforward in principle to solve for w(x, y), at least numerically.
The corresponding inverse problem is, given a desired sag prole w(x, y), to
determine the required distribution of the bending stiness D(x, y). When
viewed as a problem for D, (4.6.30) is a secondorder linear partial dierential equation, and two immediate questions arise: what boundary conditions
should be imposed? and will D be positive?.
Since the terms involving the highest derivatives of D in (4.6.30) are
2
2
w
2w 2D
2w 2D
2w 2D
w
+
+ 2
+ 2(1 )
+ 2
,
x2
y
x2
xy xy
y 2
x
y 2
we can read o the discriminant
2 2 2
2
w
w
w
2w
2w
2
+ 2
+ 2 ,
= (1 )
xy
x2
y
y 2
x
(4.6.31a)
+
. (4.6.31c)
= (1 )2
x2 y 2
xy
x2
y 2
Now, if there is a simply supported edge at x = 0, then we must have
Myx = 0 there and we deduce from (4.6.31b) that > 0 and (4.6.30) is
therefore hyperbolic. On the other hand, the righthand side of (4.6.31c)
shows that < 0 wherever the Gaussian curvature is positive, which we expect to be the case nearly everywhere in a typical windscreen. It follows that
(4.6.30) must change type somewhere between the simply supported boundary and the interior, and it is an interesting open question as to whether
smooth solutions to this mixed type partial dierential equation exist (see
Salazar & Westbrook, 2004).
4.6.5 More general inplane stresses
If we do not make the simplifying assumption that the inplane stress is
isotropic and known, then we need to obtain constitutive relations for the
171
components Txx , Txy and Tyy . We appeal to the exact biaxial strain solution
(2.2.19) to pose the constitutive relations
Eh
(exx + eyy ) ,
1 2
Eh
exy ,
=
1+
Eh
=
(exx + eyy ) ,
1 2
Txx =
Txy = Tyx
Tyy
(4.6.32a)
(4.6.32b)
(4.6.32c)
2A
,
y 2
Txy =
2A
,
xy
Tyy =
2A
.
x2
(4.6.34)
Then elimination of u and v from (4.6.32) reveals that A satises the twodimensional biharmonic equation
4 A = 0.
(4.6.35)
D
w
g
=
,
y 2 x2
xy xy
x2 y 2
t2
(4.6.36)
172
Approximate theories
4.7 Von K
arm
an plate theory
4.7.1 Assumptions underlying the theory
There is an interesting geometrically nonlinear generalisation of plate theory
which can be derived plausibly using the ideas of Section 4.6, although we
will have to wait until Chapter 6 for a systematic exposition on the range
of applicability of the theory.
We follow the derivations used in the previous section to obtain the governing equations
Txy
Txx
+
= 0,
x
y
Txy
Tyy
+
=0
x
y
(4.7.1)
2w
2w
2w
2w
4
+
T
+
2T
D
w
g
=
xy
yy
x2
xy
y 2
t2
(4.7.2)
for the transverse displacement. Any small segment of the plate will be locally approximately planar, so we also assume that the constitutive relations
(4.6.32) still hold. Our focus, then, is on obtaining improved approximations for the strain components eij . We note with concern that in (4.6.32)
we are including the inplane displacement, although we derived (4.7.1) by
neglecting the inplane acceleration. A systematic justication of this step
requires us to render the equations dimensionless and take a careful asymptotic limit, as we will demonstrate in Chapter 6. Nevertheless, it is apparent
that our model (4.6.12) requires both the inplane displacements (u, v) and
the transverse displacement w to be suciently small. The von K
arman theory emerges when we make a specic choice about exactly how small they
must be, and it will transpire that the inplane displacements must be an
order of magnitude smaller than the transverse displacement. This choice
means that all the displacement components contribute to the leadingorder
strain of the plate, as we will discover below, yet it also allows us to retain
the linear constitutive relations (4.6.32).
X
X + u(X, Y )
X = Y ,
x(X) = Y + v(X, Y ) .
(4.7.3)
0
w(X, Y )
4.7 Von K
arm
an plate theory
173
X + X + u(X + X, Y + Y )
x(X + X) = Y + Y + v(X + X, Y + Y )
w(X + X, Y + Y )
1 + u/X
u/Y
(4.7.4)
= x(X) + X v/X + Y 1 + v/Y + O X2
w/X
w/Y
after the deformation.
Now, when calculating the change in length of the small line element X,
we retain only the leading terms in u, v and w to obtain
.
w 2
u
2
2
2
x X = X 2
+
X
X
.
v
w w
w 2
v
u
2
+
+
+ Y 2
+
. (4.7.5)
+ 2XY
Y
X
X Y
Y
Y
Our assumption that w, although small, greatly exceeds u and v, means that
the terms on the righthand side of (4.7.5) are all of the same order. It also
means that the Lagrangian and Eulerian coordinates are identical to leading
order, so we can replace (X, Y ) with (x, y) without incurring any additional
errors. We therefore take the inplane strain components to be
u 1 w 2
+
exx =
,
(4.7.6a)
x 2 x
1 u v
w w
+
+
,
(4.7.6b)
exy =
2 y
x
x y
v 1 w 2
+
eyy =
.
(4.7.6c)
y 2 y
Then (4.7.5) gives us the lowestorder change in length of a line element in
the form
x2 X2 = 2exx X 2 + 4exy XY + 2eyy Y 2 .
(4.7.7)
Notice that (4.7.6) may also be obtained from (1.4.5) by assuming that w is
an order of magnitude larger than u and v.
By including the nonlinear terms in (4.7.6) we are abandoning one of the
principal assumptions of linear elasticity, namely that the strain may be
approximated as a linear function of the displacement gradients. However,
we still plan to use the linear constitutive relations (4.6.32) between stress
and strain. The von K
arman theory is our rst example of a theory that is
174
Approximate theories
.
(4.7.8)
y 2
xy
x2
x2 y 2
xy
Second, let us pose the question: which displacement elds give rise to
zero inplane strain? We see from (4.7.7) that such deformations will preserve the lengths of all line elements in the plate: x and X will be
equal for all X and Y . Such lengthpreserving transformations are known
as isometries, and, in plane strain, the only isometries are rigidbody motions. When we allow the additional freedom of transverse as well as inplane
displacements, there is another class of isometries, corresponding to bending
of the plate. It is, for example, easy to bend this page without signicantly
changing the lengths of any lines printed upon it.
When we have zero inplane strain, the equations exx = exy = eyy = 0
form three coupled partial dierential equations for u, v and w. The elimination of u and v that led to (4.7.8) reduces the system to a single equation
for w, namely
2 2
2w 2w
w
= 0.
(4.7.9)
2
2
x y
xy
Thus a surface z = w(x, y) can be isometrically deformed from a plane
if and only if w satises (4.7.9), which is a version of the MongeAmp`ere
equation (Ockendon et al., 2003, p. 380). A surface possessing this property is
known as developable. Roughly speaking, a developable surface is any shape
into which a piece of paper may be bent without creasing or tearing, and
familiar examples include cylinders and cones (see, for example, Kreyszig,
1959, Section 58).
Equation (4.7.9) also states that the plate has zero Gaussian curvature.
We recall from Section 4.6.2 that the Gaussian curvature is dened to be
K = 1 2 ,
(4.7.10)
where 1 and 2 are the principal curvatures of the surface z = w(x, y).
A developable surface therefore has either 1 = 0 or 2 = 0 everywhere,
4.7 Von K
arm
an plate theory
175
(a)
(b)
(c)
(d)
Fig. 4.12 (a) A cylinder, (b) a cone, (c) another developable surface, (d) a hyperboloid, which is ruled but not developable.
so at each point there is one principal direction in which the surface has
zero curvature. These directions dene a family of straight lines, known as
generators, embedded in the surface. Indeed, we can say that the surface
is generated by moving a straight line through a given trajectory. Simple
examples of developable surfaces include cylinders (not necessarily circular)
and cones, as illustrated in Figure 4.12(a) and (b), where the generators
are shown as black lines. However, there are many less familiar cases, such
as that shown in Figure 4.12(c). We also caution that there are also nondevelopable ruled surfaces, that is, surfaces containing a straight line through
any point. For example, the hyperboloid shown in Figure 4.12(d) contains
two families of straight lines, although its Gaussian curvature is negative
everywhere. To be clear: every developable surface is ruled but most ruled
surfaces are not developable.
176
Approximate theories
an Airy stress function A, dened in the usual way, so that (4.7.2) becomes
2A 2w
2w
2A 2w
2A 2w
4
D
w
g
=
.
+
y 2 x2
xy xy
x2 y 2
t2
(4.7.11)
Next, we use (4.6.32) and (4.7.8) to eliminate the strain components and
hence obtain
2 2
2w 2w
w
.
(4.7.12)
4 A = Eh
2
2
x y
xy
The coupled partial dierential equations (4.7.11) and (4.7.12) for w and A
comprise the von K
arman model. Evidently (4.7.11) reproduces the linear
plate equation (4.6.36), so the improved approximation for the strain has led
to the introduction of just one new term: the righthand side of (4.7.12). We
recognise the expression in braces as the Gaussian curvature, which is zero
only if the deformed plate is developable. Hence, any transverse displacement
that does not consist of a pure bending of the plate inevitably causes inplane
stress, in contrast with linear plate theory, where inplane and transverse
deformations are decoupled. It is the nonlinear coupling provided by the
righthand side of (4.7.12) that makes the von K
arman equations so much
more dicult to solve than the linear plate equations.
In the same way that we were led to the strain energy (4.6.22) for linear bending of a plate, we may use (4.6.23), generalised to include linear
stretching, to suggest that the energy per unit area in a von K
arman plate
is
D
2
2 2
2w 2w
w
x2 y 2
xy
1
2
+
,
(Txx + Tyy )2 2(1 + ) Txx Tyy Txy
2Eh
2 2
w 2(1 )
where the integrated stress components are given by (4.6.32) and the strain
components are again dened by (4.7.6). A lengthy exercise in the calculus
of variations shows that when we consider virtual displacements (u, v, w)
that minimise the net strain energy in the plate, we do retrieve (4.7.11),
(4.7.12).
Concerning boundary conditions, we have to impose two conditions on w,
as described in Section 4.6.2, and two conditions on A, analogous to those
described in Section 2.6.3 for plane strain problems.
177
X=
Y
W (X, Y )
to
X + u(X, Y )
.
x(X) =
Y + v(X, Y )
W (X, Y ) + w(X, Y )
(4.8.1)
X + X
X + X =
Y + Y
W (X + X, Y + Y )
1
0
+ Y 1 + O X2
= X + X
0
W/X
W/Y
(4.8.2)
is deformed to x + x, where
u/Y
1 + u/X
+ Y 1 + v/Y + O X2 . (4.8.3)
x = X
v/X
(W + w)/Y
(W + w)/X
Now when we calculate the change in the square of the length of the small
line element X, we nd to leading order that
2
x X = X 2
+ 2XY
u
W w
2
+2
+
X
X X
w
X
v
W w W
u
+
+
+
Y
X
X Y
y
v
W
+2
+ Y 2 2
Y
Y
2 .
w
w w
+
X
X Y
.
w
w 2
, (4.8.4)
+
Y
Y
178
Approximate theories
(4.8.5a)
(4.8.5b)
(4.8.5c)
Again following Section 4.7.2, we can eliminate the horizontal displacement (u, v) from (4.8.5) to obtain
2 2
2 eyy
2 exy
2w 2w
w
2 exx
+2
=
y 2
xy
x2
x2 y 2
xy
2W 2w 2W 2w
2W 2w
. (4.8.6)
+
+
2
x2 y 2
y 2 x2
xy xy
Notice that the righthand side of (4.8.6) may be written as
2
2
2
2
2
2 (W + w) 2 (W + w)
W 2W
(W + w)
W
X 2
Y 2
XY
X 2 Y 2
XY
which is the change in the Gaussian curvature compared to that of the
initial shape of the plate. The isometries of our initially curved shell are
therefore those deformations that preserve the Gaussian curvature. This
evidently generalises the deformation of a at plate, which can be bent
without stretching only into a surface with zero Gaussian curvature.
If we denote the current shape of the shell by
z = f (x, y) = W (x, y) + w(x, y)
(4.8.7)
and the initial Gaussian curvature by K0 (x, y), then, for the strain to be
zero, f must satisfy the inhomogeneous MongeAmp`ere equation
2 2
f
2f 2f
= K0 (X, Y ).
(4.8.8)
2
2
X Y
XY
This famous nonlinear partial dierential equation has the unusual property
that its type is determined by the righthand side, as shown in Exercise 4.12.
If K0 = 0 then the initial shell shape is developable, so at each point it
resembles a cylinder, as illustrated in Figure 4.13(a), and we recover (4.7.9),
derived previously for an initially at plate. In this case, (4.8.8) is parabolic,
with one repeated set of real characteristics corresponding to the generators
of the initial developable surface. This reects the fact that the shell can be
arbitrarily bent about the generators without causing any inplane stress.
(b)
179
(c)
Fig. 4.13 Typical surface shapes with (a) zero, (b) negative and (c) positive Gaussian curvature.
If K0 < 0 then the initial shell shape is anticlastic, like the saddle point
shown in Figure 4.13(b), and (4.8.8) is hyperbolic. Hence there are two
distinct families of real characteristics and, for there to be no inplane strain,
any localised bending of the shell will cause nonlocal displacements in the
region of inuence spanned by these characteristics. Such a shell is therefore
somewhat stronger than a developable surface.
Finally, if K0 > 0 then the initial shell shape is synclastic, as in Figure 4.13(c), and (4.8.8) is elliptic. Such a shell will be stronger still because
any local bending causes deformations throughout the entire shell.
D
w
g
=
,
y 2 x2
xy xy
x2 y 2
t2
(4.8.9)
xy
1+
2A
Eh
= Tyy =
(exx + eyy ) ,
2
x
1 2
(4.8.10a)
(4.8.10b)
(4.8.10c)
180
Approximate theories
with the strain components now given by (4.8.5). By eliminating the inplane
displacement (u, v) between these, we nd that A now satises
2 2
2 2
2W 2W
W
2f 2f
f
4
. (4.8.11)
+
A = Eh
2
2
2
2
x y
xy
x y
xy
In (4.8.9) and (4.8.11) we have two equations for A and w, generalising the
von K
arman equations, which may be recovered by setting W 0, f w.
To make the model more tractable, let us now suppose that the transverse
displacement is much smaller than the initial deection of the centresurface,
so that w W . Then, if we keep only the lowest order terms in w, (4.8.9)
and (4.8.11) reduce to
2W
y 2
2W
y 2
2W 2A
2W 2A
2A
+
2
= D4 w,
x2
xy xy
x2 y 2
2W 2w
2W 2w
2w
+
2
= (Eh)1 4 A,
x2
xy xy
x2 y 2
(4.8.12)
(4.8.13)
where, for additional simplicity, we have also neglected the body force and
time dependence. Given the initial shell prole W (x, y), we therefore have
a coupled system of linear partial dierential equations governing the small
transverse displacement w and the Airy stress function A.
y = Ly ,
W = W0 W ,
w = w0 w ,
A=
Dw0
A,
W0
(4.8.14)
where L is a typical lateral dimension, and w0 a typical boundary displacement. We obtain, after suppressing primes,
2W
y 2
2W
y 2
2W 2A
2A
2W 2A
+
2
= 4 w,
x2
xy xy
x2 y 2
2W 2w
2W 2w
2w
+
2
= 4 A,
x2
xy xy
x2 y 2
(4.8.15)
(4.8.16)
where
=
h2
.
12(1 2 )W02
(4.8.17)
181
For a suciently thin shell, we may therefore neglect the righthand side of
(4.8.16) so that the equations decouple: given suitable boundary conditions,
our strategy would be to solve
2W 2w
2W 2w
2W 2w
+
2
=0
y 2 x2
xy xy
x2 y 2
(4.8.18)
for w and then substitute the result into (4.8.15) to obtain an equation for A.
We therefore have to solve two linear secondorder partial dierential
equations in succession, each involving the dierential operator on the lefthand side of (4.8.18). It is easy to verify that the discriminant which
determines the type of (4.8.18) is equal to the initial Gaussian curvature of
the shell, that is
2 2
2W 2W
W
.
(4.8.19)
= K0 (x, y) =
2
2
x y
xy
The problem is therefore parabolic, hyperbolic or elliptic, according to
whether the shell is developable, anticlastic or synclastic. This immediately
casts doubt on the model (4.8.18) for, suppose we wished to prescribe w
around the perimeter of an anticlastic shell. Then (4.8.18) would present us
with an illposed Dirichlet problem for a hyperbolic partial dierential equation, and there would probably be no solution at all. On the other hand, to
apply clamped boundary conditions on a synclastic shell, we would apparently need to impose both w and w/n on a secondorder elliptic equation,
which again leads to an illposed problem.
Fortunately we can understand the deciencies in (4.8.18) when we realise from (4.8.6) that, when w is small enough for quadratic terms to be
neglected, (4.8.18) is a necessary condition for there to be no inplane stretching, i.e. only bending is allowed. Hence, if the stresses applied to the shell
are violent enough to produce inplane stretching, then we must reconsider
our use of (4.8.14) to render the problem dimensionless. One possibility is
to rescale A such that
= A = O (1),
A
(4.8.20)
which models deformations other than pure bending which increase the inplane stress by an order of magnitude. Substituting (4.8.20) into (4.8.15)
and (4.8.16) before neglecting , we nd that the equations again decouple,
now determined by
with A
2W 2A
2W 2A
2W 2A
+
2
=0
y 2 x2
xy xy
x2 y 2
(4.8.21)
182
Approximate theories
2
= 4 A.
y 2 x2
xy xy
x2 y 2
(4.8.22)
A = 1/2 A,
(x, y) = 1/4 (, ).
(4.8.23)
Recalling that W is a prescribed function of x and y, we see that this transforms (4.8.15) and (4.8.16) into
2W 2A
2W 2A
2W 2A
4 w,
+
2
=
(4.8.24)
y 2 2
xy
x2 2
2W 2w
2W 2w
2W 2w
4 A,
+
2
=
(4.8.25)
y 2 2
xy
x2 2
2 = 2 / 2 + 2 / 2 . This is eectively the full problem for w, A,
where
except that the coecients are constant to lowest order on the (, )scale.
We have therefore identied the following three alternative classes of deformation that a thin shell may suer.
(i) If relatively small stresses are applied at the boundaries, then the deformation described by (4.8.18) is an isometry, and the inplane stress
that it produces may be determined a posteriori from (4.8.15).
(ii) When large inplane stresses are applied, then the bending stiness of
satises the decouthe shell becomes negligible in comparison, so that A
pled equation (4.8.21). The inplane stretching caused by these stresses
then drives the equation (4.8.22) for the transverse displacement.
(iii) Both eects of bending stiness
and
inplane stretching can balance in
narrow regions of width O 1/4 . These regions may be close to the
boundary of the shell, or they may take the form of creases that are
remote from the boundary.
In cases (i) or (ii), the reduction of order means that it will in general
be impossible to nd solutions that are everywhere compatible with any
183
(4.8.26)
W =
2w
2A
= 4 w,
2 = 4 A.
(4.8.27)
2
x
x
Now, letting 0, we see that w can only be a combination of two forms:
w(x, y) = a(y)
(4.8.28)
Txx =
y 2
,
2
Txy = 0,
Tyy = 0.
(4.8.31)
184
Approximate theories
(a)
(b)
z
y
x
(c)
(d)
Fig. 4.14 Deformations of a cylindrical shell; (a) original shape, (b) bending about
the generators, (c) rotating the generators, (d) bending along the generators.
We can associate this stress in the xdirection with compression of the generators.
Next, let us consider the anticlastic shell
W = xy,
(4.8.32)
2A
= 4 w,
xy
2w
= 4 A.
xy
(4.8.33)
(4.8.34)
185
(b)
z
y
x
(c)
(d)
Fig. 4.15 Deformations of an anticlastic shell; (a) original shape, (b) bending along
the lines x = const., (c) bending along the lines y = const., (d) bending along the
lines x y = const.
(x + y)2
,
2
(4.8.35)
4
x + y 4 + O (1).
24
(4.8.36)
y 2
,
2
Txy = 0,
Tyy =
x2
2
(4.8.37)
shows that both the normal stresses increase quadratically. due to stretching
of the straight lines x = const. and y = const. respectively.
Finally, let us consider the synclastic shell
x2 + y 2
W =
.
(4.8.38)
2
186
Approximate theories
(a)
(b)
z
y
x
(c)
(d)
Fig. 4.16 Deformations of a synclastic shell; (a) original shape, (b) and (c) two
possible deformations, (d) onedimensional bending.
2 w = 4 A,
(4.8.39)
(4.8.41)
187
(b)
(c)
Fig. 4.17 A beam (a) before and (b) after bending; (c) a closeup of the displacement
eld.
188
Approximate theories
N0
(a)
(b)
N (s + s)
T0
T (s + s)
T0
M (s)
M (s + s)
N0
T (s)
N (s)
Fig. 4.18 (a) The forces and moments acting on a small segment of a beam. (b) The
sign convention for the forces at the ends of the beam.
189
and note that N is now dened to be the normal, rather than transverse,
force; since is no longer assumed to be small this is a signicant distinction.
We thus obtain the equations
d
(T cos N sin ) = 0,
ds
d
(N cos + T sin ) = 0,
ds
dM
N = 0,
ds
(4.9.2)
the rst two of which simply show that the internal forces in the x and zdirections are conserved. If a force (T0 , N0 ) is applied at the righthand end,
with an equal and opposite force at the lefthand end, as in Figure 4.18(b),
it follows that
T = T0 cos + N0 sin ,
N = N0 cos T0 sin .
(4.9.3)
d
.
ds
(4.9.4)
d2
+ N0 cos T0 sin = 0.
ds2
(4.9.5)
190
Approximate theories
For small , we can simplify the geometric relations (4.9.1) to lowest order
to give
x = s,
dw
= ,
dx
(4.9.7)
where is shorthand for (L). When taking the square root, we note that
we expect d/ds to be negative and thus obtain the solution in parametric
form as
)
d
2F
=s
.
(4.9.12)
EI
sin sin
0
L
1
191
x
2F/EI
0.2
0.4
0.6
0.8
0.1
0.25
0.2
0.5
0.3
0.75
0.4
0.5
1
0.6
1.25
0.7
1.5
0.8
2F/EI = 4
Fig.
4.19 (a) Final angle of a diving board versus applied force parameter
L 2F/EI. (b) Deection of a diving board for various values
of the force parameter; the dashed lines show the linearised solution when L 2F/EI = 1 and 2.
=L
,
(4.9.13)
EI
sin
sin
0
which can be written in terms of elliptic integrals. Given the applied force F ,
we have to solve (4.9.13) for (numerically), and the deection may then
be reconstructed using (4.9.12) and (4.9.1); see Exercise 4.15.
In Figure
4.19(a), we plot the nal angle as a function of the force
parameter L 2F/EI. As the applied force increases, the angle decreases,
tending towards /2. In Figure 4.19(b), we show the diving board prole
corresponding to dierent applied forces, and we can see clearly how the
force causes the deection to increase.
For small applied forces, we would expect the linear equation (4.9.8) to
be applicable. The linear model that describes our diving board is
dw
d3 w
(0) = 0,
= F,
w(0) =
3
dx
dx
and the solution is readily found to be
EI
d2 w
(L) = 0,
dx2
(4.9.14)
F 2
x (3L x).
(4.9.15)
6EI
In Figure 4.19(b), (4.9.15) is plotted as dashed curves, and we see that there
is excellent agreement with the full nonlinear model provided the deections
are reasonably small.
w=
192
Approximate theories
(4.9.16a)
(4.9.16b)
ns
L2 P
= n2 ,
,
A sin
(4.9.17)
=
L
2 EI
0,
otherwise,
where n is any integer. Thus the amplitude A of the solution is zero, unless
P takes one of a discrete set of critical values, in which case it is arbitrary.
Hence buckling can only occur when
P = n2 2
EI
.
L2
(4.9.18)
(4.9.19)
L2 P
.
2 EI
(4.9.20)
(a)
A
193
(b)
A
1.4
1.2
0.8
0.6
0.4
0.2
10
15
20
2.5
7.5
10 12.5 15 17.5 20
Fig. 4.20 (a) Response diagram of the amplitude of the linearised solution (4.9.17) for a buckling beam versus the force parameter . (b) Corresponding
response of the weakly nonlinear solution; the exact nonlinear solution is shown
using dashed lines.
(4.9.21)
(0) = (1) = 0,
(4.9.22)
(4.9.23)
194
Approximate theories
= 0,
(4.9.24)
+
n
1
d 2
6
and we clearly see that the required balance is obtained by choosing
= .
(4.9.25)
Before doing any further calculation, we can already infer from this that the
amplitude of the solution will vary as the square root of the excess loading.
After making the choice (4.9.25), we write as an asymptotic expansion
of the form
0 () + 1 () + .
(4.9.26)
By substituting (4.9.26) into (4.9.24) and equating successive powers of ,
we nd that
d2 0
+ n2 2 0 = 0,
0 (0) = 0 (1) = 0, (4.9.27a)
d 2
3
d2 1
2 2
2
2 2 0
, 1 (0) = 1 (1) = 0. (4.9.27b)
+
n
+
n
1
1
0
d 2
6
The general solution of (4.9.27a) is
0 = A0 sin (n) ,
(4.9.28)
and we seem to have made very little progress from (4.9.17). However, when
we examine the problem for 1 , we nd that (4.9.27b) has no solutions
unless the righthand side satises a solvability condition, namely
1
3
2 0
sin (n) d = 0.
(4.9.29)
1 0 n
6
0
By substituting for 0 from (4.9.28), we obtain the following equation for
the leadingorder amplitude:
81
(4.9.30)
A0 A20 2 = 0.
n
When 1 is negative, the only real solution of (4.9.30) is A0 = 0 but, for
see Hinch (1991), for example, for the denition of an asymptotic expansion; for the purposes
of this chapter, we can interpret as meaning approximates increasingly well as 0
(a)
(b)
A0
195
A0
stable
stable
unstable
stable
Fig. 4.21 (a) Pitchfork bifurcation diagram of leadingorder amplitude A0 versus
forcing parameter 1 . (b) The corresponding diagram when asymmetry is introduced.
2
1
= 0,
(4.9.31)
A A 8
n2
and this weakly nonlinear response is plotted in Figure 4.20(b). This gives
the physically expected result that the amplitude of each buckled solution
increases with the excess applied force. We can also see how Figure 4.20(a)
approximates this response for very small A. For comparison, we also show
the response of the full nonlinear solution of (4.9.16); see Exercise 4.18.
We can use the argument after (4.4.18) to show that the solution = 0 is
stable for 1 < 0 but loses stability to the nonzero solution as 1 crosses zero.
This justies the labelling in Figure 4.21(a) and can also be explained on
the basis of an energy argument, as in Exercise 4.14. Another key ingredient
of bifurcation pictures like Figure 4.21(a) is the symmetry that we have
assumed in our strut model. Any imperfection or asymmetry in the model
would deform the response diagram into something like Figure 4.21(b), in
which there is a smooth dependence of the solution on the control parameter.
This is exemplied by the reintroduction of gravity in the model as in
Exercise 4.16.
4.10 Nonlinear rod theory
We are now in a position to generalise and combine the linear rod theory of
Section 4.5 with the nonlinear beam theory of Section 4.9 to derive a model
for the equilibrium of a rod that can both twist about its tangent vector
196
Approximate theories
and bend in two perpendicular directions. We let F and M denote the force
and moment exerted on any crosssection s = const. of the rod, s being
arclength along the centreline. As in Section 4.9, we will assume that the
rod is inextensible so that s is eectively a Lagrangian variable. Neglecting
body forces for simplicity, a force balance gives
dF
= 0,
ds
so that
F = F 0 = const..
(4.10.1)
This equation generalises the rst two equations of (4.9.2). Also, if r(s)
parametrises the centreline of the rod, then a moment balance analogous
to the nal equation of (4.9.2) gives
dr
dM
+
F = 0, so that M + rF 0 = M 0 = const.. (4.10.2)
ds
ds
The equations (4.10.1), (4.10.2) for the vectors F , M and r now need
to be closed by a constitutive law for M , and this is what poses the most
interesting modelling challenge. Instead of adopting the usual dierential
geometry denition of curvature, we consider what happens to a Lagrangian
orthonormal system ei (s) (i = 1, 2, 3) as the rod deforms. This system is
xed in the rod and we choose
e1 (s) =
dr
,
ds
(4.10.3)
with e2 and e3 dening the normal plane. At dierent sstations, this system
will have rotated by dierent amounts, and we infer that there exists a vector
(s) such that
dei
= (s)ei (s).
ds
(4.10.4)
(4.10.5)
e2 =
dv
, 1,
dx
197
T
,
e3 =
T
dw
, , 1
,
dx
(4.10.6)
where measures the twist of the rod about the xaxis. It is easy to check
that these are consistent with (4.10.4) and correspond to
T
d
d2 w d2 v
=
, 2 , 2
.
(4.10.7)
dx
dx dx
Now we see that we can reproduce (4.5.5) and (4.5.9) by choosing
R
0
0
B = 0 EIyy EIyz ,
(4.10.8)
0 EIyz EIzz
in the notation of Section 4.5.
T
To retrieve nonlinear beam theory, we write r = x(s), 0, w(s) , so that
cos
0
sin
e1 = 0 ,
e2 = 1 ,
e1 = 0 ,
(4.10.9)
sin
0
cos
and we easily nd that
T
d
.
= 0, , 0
ds
(4.10.10)
198
Approximate theories
about spinning tops, gyroscopes and inertia coupling in classical mechanics can be used to explain the often infuriating coupling between torsion
and bending that is only too familiar to those who have to wind cables and
garden hoses.
This Kirchho analogy disappears when we consider the dynamics of a
rod subject to a body force g, the model for which is precisely as above with
d/ds replaced by /s and (4.10.1) replaced by
F
2r
.
= g +
2
t
s
(4.10.12)
We are grateful to the late Prof. T. Brooke Benjamin for drawing this experiment to our
attention.
199
(a)
b
(b)
z
M
y
y
x
mg
Fig. 4.22 (a) A system of pendulums attached to a twisting rubber band. (b) Transverse view dening the angle (x, t).
+ sin = 0.
(4.11.3)
t2
x2
This is one of the most famous completely integrable nonlinear partial
dierential equations.This means that it can be solved explicitly for any
given initial data on < x < by using inverse scattering theory; see for
instance Drazin & Johnson (1989). Moreover, localised travelling waves exist
that exhibit strong stability properties; even though their speed depends
on their amplitude, they can pass through each other virtually unscathed.
Such localised waves are called solitons and, for (4.11.3), we can nd them
by setting to be a function only of = x vt, where the wave speed v
is constant. Then (4.11.3) is reduced to the nonlinear ordinary dierential
equation
d2
2
+ sin = 0.
(4.11.4)
v 1
d 2
If we seek an isolated soliton solution, in which const. as , then
the only possible solutions of (4.11.4) are
0
1
200
Approximate theories
t=0
z
t=1
t=2
x
y
2x
(T cos N sin ) = 2 ,
s
t
2w
where
N = EI
2
.
s2
(4.11.9)
Although this is a more complicated model than (4.11.3), we can still seek
a travelling wave solution in which x, w, T , N and are functions only of
= s vt. We also postulate the boundary conditions
T T ,
N 0,
0,
x s,
w0
as
s , (4.11.10)
which would apply, for example, to a long taut rope that had been shaken
briey at one end. As in Section 4.11.1, by looking for a travelling wave, we
reduce (4.11.8) to nonlinear ordinary dierential equations, which may be
w
2
1
201
0.5
1
1.5
2
v
Fig. 4.24 Travelling wave solution of the nonlinear beam equations (with k = 1).
integrated to obtain
T cos N sin = v 2 (cos 1) + T ,
(4.11.12)
(4.11.13)
where M = v /T . We recall from Chapter 3 that T / is the phase
velocity of smallamplitude transverse waves on a string, and M is the Mach
number of the wave.
By analogy with (4.11.4), we see that, as long as M < 1, there is a solution
connecting 0 as + to 2 as , namely
)
T
1
k( 0 )
(1 M 2 ),
= 4 tan
(4.11.14)
e
,
where k =
EI
where the constant 0 represents an arbitrary translation. With given by
(4.11.14), we can integrate (4.11.7) to obtain the parametric description
1 2 tanh
x
=
(4.11.15)
w
2 sech
k
for the shape of the rope, where = k( 0 ). As shown in Figure 4.24, this
represents a single loop, in which the rope turns through 2, propagating
at speed v which depends, through (4.11.14), on the amplitude. The multivaluedness of w(x) is only physically realistic if the rope is allowed to deform
slightly out of the (x, z)plane.
202
Approximate theories
t2
t
2
w
2w
.
2c
N cos + (T T ) sin =
t2
t
x
= cos ,
w
= sin ,
N = EI
(4.11.16)
(4.11.17)
(4.11.18)
T = T T ,
N=
EI
2
N,
t=
2c3
EI
,
(4.11.19)
so that (4.11.16)(4.11.18) become
X
= cos ,
W
= sin ,
(4.11.20)
N =
,
2
2 2 X
2X
,
(T 1) cos N sin =
4 2
2 2 W
2W
N cos + (T 1) sin =
,
4 2
203
(4.11.21)
(4.11.22)
(4.11.23)
EI
.
T 2
(4.11.24)
We suppose, as in Section 4.11.2, that the beam is held straight and horizontal, under a uniform tension at innity, so that
T 1,
N 0,
0,
X ,
W 0
as
. (4.11.25)
(4.11.26)
X
= 0,
T1 sin + N cos +
W
= 0.
(4.11.27)
(4.11.28)
2
.
(4.11.29)
204
Approximate theories
3 = 0.
(4.11.31)
3 = 0.
(4.11.32)
EI
c t
s
2 s
s
Now if we seek a travelling wave, in which is again a function only of
= s V t, then we obtain the ordinary dierential equation
d
d3 1 d 3 2T
(1 M )
= 0,
(4.11.33)
+
3
d
2 d
EI
d
where M = V /c again. As shown in Exercise 4.21, the solution of this equation satisfying 0 as + and 2 as is
)
2T (1 M )
1
( 0 )
.
(4.11.34)
e
,
where =
= 4 tan
EI
We thus reproduce the travelling wave solution (4.11.14) found previously,
noting that is the leadingorder approximation to k as M 1.
4.12 Concluding remarks
In this chapter, we have emphasised phenomena that can occur when there
is some strong geometric constraint of thinness, and hence the possibility of geometric nonlinearity. However, what is more remarkable than the
phenomena themselves is the mathematical challenge posed by passing from
a fully threedimensional model to simpler models for nearly one or twodimensional solids. This challenge is exemplied by the fact that, although
(4.3.2) is an apparently physically obvious paradigm for linear wave propagation, we will see in Chapter 6 that it can only be derived systematically
Exercises
205
Exercises
4.1
4.2
4.3
4.4
w(0) =
dw
(0) = 0,
dx
w(L) =
dw
(L) = 0.
dx
206
Approximate theories
111111111
000000000
000000000
111111111
000000000
111111111
000000000
111111111
F
Fig. 4.25 A beam clamped near the edge of a table.
dw
(0) = 0,
dx
d3 w
F
(L) =
,
dx3
EI
w(0) =
4.5
Find the maximum magnitude of the slope dw/dx and deduce that
these results are valid provided F L(L )/EI 1.
Consider a beam of length L that is simply supported at two points
and allowed to sag under its own weight, as shown in Figure 4.26.
The supports are assumed to be a distance < L apart and to
be symmetric about the middle of the beam. Derive the following
dimensionless model for the vertical displacement w(x) of the beam:
d4 w
= R(x ),
dx4
d3 w
dw
d2 w
d3 w
(0) =
(0)
=
(1/2)
=
(1/2) = w(/2) = 0,
dx
dx3
dx2
dx3
where = /L and is the Dirac deltafunction.
Show that the dimensionless reaction force R is equal to 1/2 and
Exercises
207
z
x
L
4.6
explain this result physically. Solve for w(x) and plot the solution
as is varied. Show that, if is small, then the beam sags down at
its ends and goes up in the middle while, if is close to 1, then it
sags down in the middle and its ends move upwards. Explore how
the behaviour varies between these two limits.
Longitudinal waves in a bar of section A, Youngs modulus E and
length L are modelled by (4.2.4). The bar is initially unstressed and
at rest. The ends x = 0 and x = L are stressfree except for an
impulsive compressive force J applied at x = 0 for 0 < t < T , where
T < L/c. Show that
1,
x ct < cT,
u
ct x
=
,
cT < x ct < 0,
U
cT
0,
0 < x ct,
for 0 < t < L/c, where the maximum displacement at x = 0 is given
by U = JcT /EA.
Show that an incident wave u = f (x ct) impinging on the stressfree boundary at x = L gives rise to a reected wave f (2L x ct).
Deduce that, as T 0 with U kept xed, the resulting displacement
after the rst reection from x = L is
u = U H(ct x) + H(x + ct 2L) ,
4.7
208
4.8
Approximate theories
show that the most unstable waves have wavelength 2 2EI/P
and grow over a timescale t EI/P . Suppose a strand of uncooked spaghetti is held vertically while a 1 kg weight is placed on
the upper end. Explain why it will shatter into fragments roughly
1 cm long.
[Typical parameter values for spaghetti are 2 103 kg m1 ,
EI 1.5 104 N m2 .]
Writing (nx , ny ) = (sin , cos ), show that Msn , as given by (4.6.16)
and (4.6.8), may be written as
2w
Eh3
2(1
)
sin
cos
Msn =
12 (1 2 )
xy
2
2
2
w
w 2w
w
2
2
cos .
+ 2 sin 2 +
x2
y
x
y 2
Now use the chain rules
cos
=
+ sin ,
x
1 + n s
n
sin
=
cos ,
y
1 + n s
n
4.9
Hence obtain the conditions (4.6.17) that apply on a simply supported boundary.
Show that the steady sag of a cylindrically symmetric plate with
nonuniform bending stiness D(r) is governed by the equation
2
d w
d d2 w 1 dw
gr2
dw
dD
r 2 +
+ Dr
=
. (E4.1)
+
dr
dr
dr
dr dr2
r dr
2
For the special case in which we wish w to be given by
2
7+ 2 2
12a4
2
2
r
a
,
+
w = a r
1+
(1 + )2
show that the only solution D(r) of (E4.1) that is bounded as r a
is
D=
(1 + )g
.
24(7 ) (5 + )a2 (1 + )r2
Exercises
4.10
209
2 w 2
2
2w 2
w 2 2w 2
+
2
dxdy,
(1 )
x2 y 2
y 2 x2
xy xy
U = D
2 w
2 w 2 w
2 w
(1 ) div
y 2 x xy y x2 y xy x
T
dxdy.
4.11
Deduce that (4.6.19) are the natural boundary conditions for the
minimisation.
Suppose the centresurface of a weaklycurved shell, initially given
by z = W (x, y), undergoes a transverse displacement w(x, y), so
that it is deformed to z = f (x, y) = W (x, y) + w(x, y). Following
the approach adopted in Section 4.6.1, show that the transverse displacement and inplane stress resultants satisfy
Txx
4.12
2w
2f
2f
2f
4
+
T
+
2T
D
w
g
=
,
xy
yy
x2
xy
y 2
t2
Txy
Txy
Tyy
Txx
+
=
+
= 0,
x
y
x
y
210
Approximate theories
d
(T sin + N cos ) = g,
ds
d2
= N.
ds2
L
EI d 2
(1 cos )P ds,
U [] =
2
ds
0
subject to the boundary conditions (0) = (L) = 0, leads to the
Euler strut equation (4.9.16a). Give physical interpretations of the
terms in the integrand.
Exercises
211
4.15
2 EIA2 2
n ,
4L
+ 2 1 + O 4 ,
2
where = s/L, and 1 () is to be found. Hence nd and plot the
leadingorder shape of the board. If is not assumed to be small,
show that is related to by
= 2,
sin sin
0
where is to be determined from
= 2.
sin sin
0
Deduce that the dimensionless shape of the board is given parametrically by x = x(), y = y(), where
)
2
sin sin + sin ,
x() =
1
sin d
,
y() =
sin sin
2 0
4.16
and varies between and 0. Hence plot the shape of the board for
various values of . Compare these exact solutions with the small
approximation.
Consider a beam of length L and mass per unit length, clamped
horizontally at each end and sagging under gravity. If a horizontal
212
Approximate theories
4.17
3 Ai() + Bi() = 0,
Exercises
z
1
213
n=1
0.8
0.6
0.4
0.2
Fig. 4.27 The rst three buckling modes of a vertically clamped beam.
4.18
sin2 (/2)
2s
1
d =
sin(/2)
2
L
sin (/2)
0
and deduce that the nth branch of the response diagram is given
implicitly by
=
4.19
2
4n2 2
K sin (/2) ,
2
EI
w0
as
214
Approximate theories
1
F
(
)
t d.
w(0, t) =
2 (EI3 )1/4 0
If the origin is forced to move at speed V , so w(0, t) = V t, show that
the required force is
)
2
3 1/4
.
F (t) = 2V EI
t
4.20
z
z
Show that the Miura transform
u
z
converts the mKdV equation into the Kortewegde Vries (KdV)
equation
v 3 v
v
+ 6v
+
= 0.
z z 3
v(, ) = u2 i
4.21
= 0,
2
d
4 d
d
where
)
=
2T (1 M )
.
EI
5
Nonlinear elasticity
5.1 Introduction
In Chapters 2 and 3, we have analysed solutions of the steady and unsteady
Navier equations, which were derived in Chapter 1 under the two assumptions that underpin linear elasticity.
First, we assumed that we could discard the nonlinear terms in the relation
(1.4.5) between strain and displacement. Geometrically nonlinear elasticity
concerns large deformations in which these terms are not negligible, so the
strain is a nonlinear function of the displacement gradients. This inevitably
leads to the further complication that the Lagrangian and Eulerian variables
may no longer be approximated as equal.
The second assumption behind linear elasticity is that the stress is a linear
function of the strain. This is a reasonable approximation for small strains,
but it does not work well for materials such as rubber, which can suer large
strain and still remain elastic (see Treloar, 2005). Models for such materials
require mechanically nonlinear elasticity, in which the stress is a nonlinear
function of the strain.
As indicated in Section 1.6, the fundamental diculty to be confronted
is that the balance of stresses is performed in the deformed state, while the
constitutive relation must be imposed relative to the reference conguration.
As a rst step in addressing this diculty, we will revisit the concepts of
stress and strain. We will show how they may both be expressed in a Lagrangian frame of reference, allowing a selfconsistent constitutive law to be
imposed between them. For mechanically nonlinear materials, such laws are
far less easy to specify than (1.7.6), and we will see that great care has to be
taken to avoid models that allow unphysical behaviour. Finally we will show
how the nonlinear theory, be it mechanical or geometric or both, can be
greatly simplied for hyperelastic materials for which the stress is related to
215
216
Nonlinear elasticity
the strain via the partial derivatives of an appropriate strain energy density,
thereby generalising (1.9.4).
(5.2.1)
(5.2.2)
(5.2.3)
C = F T F,
(5.2.4)
where
and C, which will shortly be shown to be a tensor, is called the Green deformation tensor.
Now suppose that a rigidbody motion is superimposed, so that each point
x is further displaced to
x = c + P x,
(5.2.5)
where the vector c and orthogonal matrix P are constant. It is a straightforward matter to substitute for x into (5.2.2) and thus obtain the new
deformation gradient as
Fij =
xi
= Pik Fkj
Xj
(5.2.6)
(5.2.7)
(5.2.8)
217
where
X = c + QX,
(5.2.9)
where the vector c and orthogonal matrix Q are constant. Now the chain
rule gives
xi
(5.2.10)
= F QT
F =
Xj
and the corresponding Green deformation tensor is thus
T
C = F F = QCQT .
(5.2.11)
(5.2.13)
(5.2.14)
This is called the left Green deformation tensor or, more commonly, the
Finger tensor. We will nd that B is more natural than C as a measure of
deformation for an observer using Eulerian coordinates x.
218
Nonlinear elasticity
dX
dx
N
n
da
dA
(5.2.15)
219
This gives the stress in the deformed conguration on an area element which,
in the reference conguration, was given by dA. The quantity in braces,
1
,
(5.2.19)
T = det (F ) F T
is a secondrank tensor called the rst PiolaKirchho stress tensor.
As shown in Section 1.4, the Cauchy stress tensor is symmetric, but
T in general is not. We can, though, use the symmetry of to deduce the
following identity satised by T :
T F T = F T T.
(5.2.20)
ij
2 xi
= gi +
,
2
t
xj
(5.2.22)
which must apply to any continuum for which a Cauchy stress tensor (ij )
can be dened. In (5.2.22), gi are the components of the applied body force
and is the density of the material, related to the initial density 0 by
0
= J = det (F ).
(5.2.23)
220
Nonlinear elasticity
With respect to Lagrangian variables, the integration domain V (0) is timeindependent, so we may dierentiate through the rst integral, using the
fact that J = 0 (X) is independent of t. In the nal integral, we recognise
the term in braces as the rst PiolaKirchho tensor Tij and apply the
divergence theorem to obtain
Tij
2 xi
dX =
gi 0 dX +
dX.
(5.2.26)
2 0
t
X
j
V (0)
V (0)
V (0)
Since this must hold for all reference volumes V (0) we deduce, assuming
that the integrand is continuous, the following Lagrangian form of Cauchys
equation:
Tij
2 xi
.
(5.2.27)
0 2 = 0 gi +
t
Xj
Although it is not immediately obvious, (5.2.27) may also be obtained
directly by using (5.2.19) to replace with T in (5.2.22) and applying the
chain rule (Exercise 5.2).
5.2.4 Example: onedimensional nonlinear elasticity
The simplest example that illustrates both mechanical and geometric nonlinearity is a unidirectional displacement u(X, t) in the Xdirection. In this
very special situation, S and E just involve scalars Sxx and Exx and the most
general constitutive law is simply a functional relationship between them,
that is Sxx = (Exx ). From (5.2.13), we obtain the onedimensional strain
1 u 2
u
+
(5.2.28)
Exx =
X
2 X
and hence
Txx = (F S)xx
x
=
u
1
+
X
2
u
X
2
.
(5.2.29)
+
= Txx .
(5.2.30)
1+
X
X
2 X
If we suppose that the material is mechanically linear, so that is the linear
function
(Exx ) = ( + 2) Exx ,
(5.2.31)
then (5.2.30) is a cubic equation for u/X with only one real root.
221
3
u
+
1+3
X
2
u
X
2 1/2
,
(5.2.33)
and they are generally not straight lines. Even more importantly, they depend on the solution u/X itself, and it can be shown that characteristics
of the same family will almost always eventually cross. When this happens,
the pieces of information carried by each of these two characteristics are
usually incompatible, so that u becomes multivalued. This results in the
formation of a shock, which is a curve in the (X, t)plane across which the
rst derivatives of u are discontinuous (Bland, 1988, Section 10.1).
(5.3.1)
1 0 0
= 0 2 0 .
0 0 3
U = RT R,
where
(5.3.2)
222
Nonlinear elasticity
(5.3.3)
(5.3.4)
between line elements dX and dx in the reference and deformed congurations respectively. Using (5.3.3), we can decompose dX into its components
along each of the eigenvectors ei :
i ei (ei dX) .
(5.3.5)
dx = M T
i
(5.3.6)
I2 (C) = 21 22 + 22 23 + 23 21 ,
I3 (C) = 21 22 23 .
(5.3.9)
223
I2 (C) =
1
Tr (C)2 Tr C 2 ,
2
It is also clear that I1 (C), I2 (C) and I3 (C), like the eigenvalues and characteristic polynomial of C, are invariant under orthogonal transformations
of the axes. Hence the equations given in (5.3.10) are identities, whether or
not C is diagonal. Henceforth we will suppress the argument of Ik to avoid
cluttering the equations.
Now suppose a scalar function (C) of the Green deformation tensor is
known to be invariant under rigidbody motions of the reference state. As
shown in Section 5.2.1, transformation of the axes by an orthogonal matrix
Q converts C to C = QCQT , so the invariance required of may be stated
as
QCQT (C)
(5.3.11)
for all orthogonal matrices Q. In particular, (5.3.11) must hold if Q is equal
to R, the orthogonal matrix that diagonalises C, so
(C) 2 ,
(5.3.12)
(5.3.13)
(5.3.14)
224
Nonlinear elasticity
(5.3.15)
S could also depend on X if the elastic properties were spatially nonuniform, but we will ignore this possibility here.
The key requirement is that the relation (5.3.15) be invariant under rigidbody motions; in other words, the way a material responds to stress should
not depend on the frame in which it is observed. As demonstrated in Section 5.2.1, rotation of the Eulerian coordinate axes through an orthogonal
matrix P transforms F to F = P F , so this frame indierence leads to the
restriction
S (F ) S (P F )
(5.3.16)
for all orthogonal matrices P . Since the Green deformation tensor C is invariant under rigidbody motions, (5.3.16) is certainly satised if the constitutive
relation takes the form
S = S(C),
(5.3.17)
(5.3.18)
Then, if we take the rotated material and apply exactly the same deformation as before, the net deformation is given by
(5.3.19)
X
x(X ) = x X 0 + Q (X X 0 ) .
As shown in Section 5.2.1, the Green deformation tensor corresponding to
(5.3.19) is C = QCQT , and the corresponding stress tensor at the point X 0
is therefore
S (C) = S QCQT .
(5.3.20)
225
If the material is isotropic, then this should be identical to the stress which
we would have obtained without the original rotation, albeit related to the
new coordinates X rather than to X. By applying the usual rule for the
rotation of a tensor, we deduce that
S (C) = QS(C)QT ,
(5.3.21)
which must apply identically for any xed point X 0 and any rotation
matrix Q. The constitutive relation (5.3.17) between S and C must therefore
satisfy the symmetry condition
S(QCQT ) QS(C)QT
(5.3.22)
(5.3.23)
(5.3.26)
226
Nonlinear elasticity
However, we will see in the next section that the requirement (5.3.26) is far
from sucient for the constitutive relation (5.3.23) to lead to a physically
acceptable theory of mechanically nonlinear elasticity.
5.3.4 The energy equation
In Chapter 1, we found that the work done on any region of a linear elastic
solid is balanced by the change in both the kinetic energy and the strain
energy stored by the material as it deforms. Now let us try to repeat this
analysis for a nonlinear elastic material. We start by multiplying the momentum equation (5.2.27) by the velocity xi /t and integrating over a control
material volume V0 :
xi 2 xi
xi
xi Tij
dX +
0
dX =
0 gi
dX. (5.3.27)
2
t t
t
V0
V0
V 0 t Xj
The time derivative may be taken outside the rst integral, while the nal
integral can be manipulated using the divergence theorem to obtain
d
dt
V0
0
2
xi
t
2
Fij
dX +
Tij
dX
t
V0
xi
xi
dX +
Tij Nj dA. (5.3.28)
0 gi
=
t
V0
V 0 t
Recalling that 0 dX dx, we see that the rst term on the lefthand side
of (5.3.28) is the rate of change of kinetic energy, while the rst term on the
righthand side is the rate at which work is done by the body force. Finally,
the boundary integral is the rate at which work is done by stress on V .
We can therefore interpret the second term on the lefthand side of (5.3.28)
as the rate at which elastic energy is stored by the medium. Unfortunately,
this implies that apparently legitimate nonlinear constitutive relations may
allow a net extraction of energy from an elastic material under periodic
loading cycles. Let us consider, for example, a material with the isotropic
constitutive relation
S = 21 22 23 1 I.
(5.3.29)
subject to a deformation in which
0
0
1 (t)
F = 0
2 (t) 0 ,
0
0
1
(5.3.30)
227
0
0
2 2
1 (t)
T = F S = 1 2 1 0
(5.3.31)
2 (t) 0
0
0
1
and hence nd that
2 2
Fij
d1
d2
dX = V0 1 2 1 1
+ 2
Tij
,
t
dt
dt
V0
(5.3.32)
W
,
Fij
(5.3.33a)
W
.
F
Fij
d
dX
t
dt
(5.3.33b)
W dX,
(5.3.34)
V0
and interpret the righthand side as the strain energy stored in the medium;
the change in this energy automatically integrates to zero over a periodic
loading cycle. We will now explore this happy circumstance in more detail.
5.3.5 Hyperelasticity
A material is called hyperelastic if there is a strain energy density W (Fij )
satisfying (5.3.33). The theory of hyperelasticity has proved to be by far
the most commonlyused and convenient way to construct constitutive relations for materials like rubber. By posing a functional form for W, and
inferring from (5.3.33) the corresponding stressstrain relation, one automatically avoids thermodynamically unfeasible situations like that described
above. However, the mere existence of W is not sucient to guarantee wellposedness of the resulting elastic model. At the moment all we can say is
228
Nonlinear elasticity
that our experience with linear elasticity in Chapter 1 suggests that W (Fij )
should have a minimum (taken to be zero without loss of generality) at
Fij = ij .
As in Section 5.3.3, we must insist that the constitutive relation be invariant under rigidbody motions of the reference state. In the hyperelastic
framework, this implies that W can depend only on the six components
of C, and the second PiolaKirchho stress is then given by
S=2
W
,
C
(5.3.35)
(5.3.36)
W Ik
.
Ik C
(5.3.37)
I2
= I1 I C,
C
I3
= I3 C 1 ,
C
(5.3.38)
2 = 2
W
W
W
+ I1
+ I2
I1
I2
I3
W
W
+ I1
I2
I3
W
W
W
, 1 + I1 2 = 2
, 0 + I1 (1 + I1 2 ) I2 2 = 2
,
I3
I2
I1
(5.3.40)
229
0
I3
1
0
+ I1
I2
I2
1
2
2
+
+ I1
= 0,
I3
I2
I3
2
2
1 2
+ I1
+ I1 I2
= 0,
I3
I3
I1
2
1 2
2
+
+ I1 I2
+ I1
= 0.
I1
I2
I1
(5.3.41a)
(5.3.41b)
(5.3.41c)
Any choice of the functions i that does not satisfy (5.3.41) will lead to
behaviour such as that exhibited in (5.3.32); indeed it was these integrability
conditions that suggested the counterexample (5.3.29).
We note from (5.2.21) that the Cauchy stress is given in terms of W as
1/2 W
= I3
FT
(5.3.42)
i = I3
W
i
(5.3.43)
230
Nonlinear elasticity
3
i=1
(i 1)2 +
3
(i 1) (j 1) ,
2
(5.3.46)
i,j=1
where the scalar constants and , which we will shortly identify with the
Lame constants of Chapter 1, must satisfy > 0 and 3+2 > 0 for (5.3.46)
to be positive denite.
Since, in principal axes, W = W(i ) and C = diag 2i , (5.3.35) implies
that
1 W
.
(5.3.47)
Si =
i i
Substitution from (5.3.45) and (5.3.46) thus leads to
Si (1 + 2 + 3 3) + 2 (i 1) (E1 + E2 + E3 ) + 2Ei (5.3.48)
as the strain tends to zero. The stress and strain tensors are thus related by
S Tr (E) I + 2E,
(5.3.49)
and, since the left and righthand sides both transform as tensors, (5.3.49)
holds with respect to any chosen axes.
If the displacement is small, relative to any other lengthscale, then the
deformation gradient F is approximately equal to the identity and hence,
from (5.2.19) and (5.3.48), both the PiolaKirchho stress tensors and the
Cauchy stress tensor are all approximately equal. In addition, the quadratic
231
terms in the strain tensor may be neglected to leading order, and (5.3.49)
thus reproduces the linear elastic constitutive relation (1.7.6).
However, (5.3.49) holds just as long as the strain is small, so the displacement need not be small provided it is close to a rigidbody motion.
In such situations, we can thus justify the use of the mechanically linear
constitutive equation (5.3.49), while retaining the geometrically nonlinear
relation between x and X. Indeed we have already done this several times
in Chapter 4.
5.3.7 Incompressibility
Many rubberlike materials are observed to be almost incompressible, and it
is common to employ constitutive relations that impose this. From (1.3.6),
which holds for nonlinear deformations, the condition of incompressibility is
J = 0 / = 1 or, in terms of the strain invariants,
I3 1.
(5.3.50)
(I3 1)2
(I1 , I2 ) ,
+W
2
(5.3.51)
2
p,
(5.3.52)
232
Nonlinear elasticity
(det F ) Fij = 0.
Fij
(5.3.55)
(5.3.56)
We also note from (5.3.28) that no additional work is done by the increment
provided
Tij Fij = 0.
(5.3.57)
(5.3.58a)
neoHookean:
W = (I1 3) ,
2
(5.3.58b)
MooneyRivlin:
W = c1 (I1 3) + c2 (I2 3) ,
W = 2 (1 + 2 + 3 3) ,
(5.3.58c)
+ 2 + 3 3
,
(5.3.58d)
Ogden:
W = 2 1
where , c1 , c2 and are positive constants. In each case, the material is also
assumed to be incompressible, so the constitutive relation is supplemented
by the constraint I3 1.
Varga:
5.4 Examples
(a)
1
F
(b)
233
(d)
(c)
(e)
(f)
Fig. 5.2 Typical forcestrain graphs for uniaxial tests on various materials: (a) neoHookean, (b) MooneyRivlin, (c) Varga, (d) Ogden ( < 1), (e) Ogden (1 < < 2),
(f) Ogden ( > 2).
In Figure 5.2, we show typical forceversusstrain response curves for uniaxial stress tests on materials described by each of the above constitutive
relations (see Exercise 5.10). The neoHookean and MooneyRivlin materials are characterised by an initial linear response, followed by transition to a
less sti linear behaviour. Varga materials fail at a nite value of the applied
force. For Ogden materials, the behaviour depends on : when < 1, the
force reaches a maximum before decaying towards zero at large strain; thus
once again only a nite force is needed for the material to fail completely.
For between 1 and 2, the response is qualitatively similar to that for a
MooneyRivlin material, while for > 2 the stiness always increases with
increasing strain.
5.4 Examples
5.4.1 Principal stresses and strains
We conclude with some simple examples which have sucient symmetry for
analytical progress to be made. In all cases, we will assume incompressibility
of the medium, which leads to the relations
I3 = 21 22 23 = 1,
I1 = 21 + 22 + 23 ,
2
2
I2 = 2
1 + 2 + 3 ,
(5.4.1)
234
Nonlinear elasticity
F
h
F
Fig. 5.3 A square membrane subject to an isotropic tensile force F .
for the strain invariants. In all the examples considered below, the stress
and strain tensors are diagonal, and the principal stresses are found from
(5.3.54) to be
i = i
W
p.
i
(5.4.2)
2c2
p,
2i
(5.4.3)
F
W
,
p=
1
2 3 h
F
W
,
p=
2
1 3 h
W
p = 0,
3
(5.4.4)
.
h
1
1 3
2
2 3
(5.4.5)
5.4 Examples
(a)
(b)
235
(c)
30
15
25
12.5
20
3
10
15
7.5
c
1
10
2.5
1
1
236
Nonlinear elasticity
P
c = 0.3
0.2
0.1
0
2
Fig. 5.5 Scaled pressure inside a balloon as a function of the stretch = R/r for
various values of the MooneyRivlin parameter c = c2 /c1 .
r
,
R
(5.4.7)
and the normal stretch r is determined by incompressibility. Since the balloon is assumed to be thin, we can use the biaxial result above to calculate
the tension,
W
W
,
(5.4.8)
r
T = T = T = hr = hr
r
noting that the thickness of the balloon is hr following the deformation.
The nal observation is that a normal force balance relates the tension to
the internal pressure P and the curvature of the membrane via
P =
2T
.
r
(5.4.9)
5.4 Examples
237
5.4.4 Cavitation
Suppose a spherical cavity embedded in a rubber continuum expands as
a result of an increasing internal pressure P . Such a situation could be
achieved by dissolving a gas in a rubber matrix, for example. Now we use
spherical polar coordinates (R, , ) in the rest state and again assume the
displacement is purely radial, so that and are preserved while each
radial position R expands to a new radius r(R). We can hence write down
the principal stretches
R =
dr
,
dR
= =
r
,
R
(5.4.11)
and the incompressibility condition R = 1 gives us a dierential equation for r(R), namely
dr
R2
= 2.
dR
r
(5.4.12)
r(a)
a
(5.4.14)
238
Nonlinear elasticity
P/c1
0.2
10
0.1
c = 0
Fig. 5.6 Gas pressure inside a cavity as a function of ination coecient for various
values of the MooneyRivlin parameter c = c2 /c1 .
R2
R3 + 30 a3 a3
2/3 ,
R3 + 30 a3 a3
= =
R
1/3
.
(5.4.15)
(5.4.16)
R
Now solving the rstorder dierential equation (5.4.17) subject to the two
boundary conditions
rr (a) = P,
rr 0
as
R ,
(5.4.18)
Exercises
239
Exercises
5.1
5.2
,
(E5.1)
xj Xk
J Xk
(assuming the summation convention). [Hint: integrate both
240
Nonlinear elasticity
sides over an arbitrary material volume V (t) and use the divergence theorem.]
(b) Substitute the PiolaKirchho stress tensor for the Cauchy
stress tensor in the momentum equation (5.2.22) and use the
chain rule to obtain
1
2 xi
1
J Til Fjl .
(E5.2)
0 2 = 0 gi + JFkj
t
Xk
5.3
(c) Expand out the righthand side of (E5.2) and apply the identity (E5.1) to obtain the Lagrangian momentum equation
(5.2.27).
Suppose the scalar function f of the twodimensional matrix F is
invariant under the transformation F
P F , where P is any orthogonal matrix. Let f (F ) = g(A, B, , ), where F is written in the
form
A cos B cos
F11 F12
=
.
F =
F21 F22
A sin B sin
By expressing P as the rotation matrix
cos sin
P =
,
sin cos
show that g must satisfy
g(A, B, , ) g(A, B, , )
5.4
for all . Explain why g(A, B, , ) h(A, B, ) for some function h of three independent variables. Deduce that f = f (C), where
C = F TF .
[A conceptually equivalent argument works in three dimensions;
see, for example, Noll (1958); Spencer (1970).]
(a) Suppose that two 2 2 symmetric matrices S and C satisfy
a functional relation S = S(C) which enjoys the symmetry
(5.3.22) for all orthogonal matrices Q. By considering
1 0
Q=
,
0 1
or otherwise, show that S must be diagonal whenever C is.
(b) With respect to coordinates in which C (and hence also S) is
diagonal, deduce that S must take the form
0
s1 (21 , 22 )
,
S=
0
s2 (21 , 22 )
Exercises
241
5.5
(E5.3)
S11
S22
2
.
2
1
2 = 1 + sin(t),
5.6
Show that
det (F )
= det (F )Fji1 .
Fij
5.7
[Hint: recall the inversion formula Fij1 = cofactor (Fji ) / det(F ).]
Recall from Section 5.3.1 the existence of two orthogonal matrices
R and M such that F = M T RT R, where = diag (i ). Use the
chain rule to show that
W
W
= M T RT
R.
T =
F
242
Nonlinear elasticity
Now use the relationship (5.2.19) between the Cauchy and Piola
Kirchho stresses to show that
1/2 W
RM M T RT = I3
5.8
1 .
5.9
5.10
W
.
C
I1
= ij ,
Cij
(b)
I2
= I1 ij Cij ,
Cij
(c)
I3
1
= I3 Cij
.
Cij
,
=
A
1
1 2
1/2
5.11
5.12
Exercises
243
where c > 31/6 . By introducing = 1 2 , show that the asymmetric solution satises
1 = ,
where
2 =
5.13
P ()
2 =
+
P () + 4c 2 2
2c
, P () = 1 + 3 + c 4 .
5.14
2/3
34 + 11 11
4.663.
7
(a) By changing independent variables to , show that the differential equation (5.4.17) for the stress outside a spherical
cavity may be written in the form
drr
W
W
= 2
2R
3 1
d
244
Nonlinear elasticity
at = 0 ,
rr 0 as 1,
=
1+
1/3
3
0 1 a3
(a + h)3
6
Asymptotic analysis
6.1 Introduction
In Chapter 4, we derived various approximate models for thin or slender
elastic congurations such as rods and plates. These models were obtained
using net force and moment balances combined with ad hoc constitutive
relations, for example between the bending moments and the curvatures.
In this chapter, we show how such models may be derived systematically
from the underlying continuum equations and boundary conditions. We
concentrate on a few canonical models for plates, beams, rods and shells.
Each of these models is important in its own right, and their derivation
illustrates the tools that are widely useful for analysing more general thin
structures.
The basic idea is to exploit the slenderness of the geometry so as to simplify the equations of elasticity asymptotically. This process is made systematic by rst nondimensionalising the equations, so that all the variables are
dimensionless and of order one. This highlights the small slenderness parameter = h/L, where h is a typical thickness and L a typical length of the
elastic body. A simplied system of equations is then obtained by carefully
taking the limit 0. Typically, the solution is sought as an asymptotic
expansion in powers of the small parameter , and the techniques demonstrated here fall within the general theories of asymptotic expansions and
perturbation methods. Kevorkian & Cole (1981), Hinch (1991) and Bender
& Orszag (1978) provide very good general expositions of these methods.
In fact we have already encountered an example of this approach when
discussing the torsion of a thinwalled tube in Section 2.5, and an even simpler paradigm is the dimensionless model for antiplane strain of a thin plate.
We will use this example in Section 6.2 to introduce the asymptotic arguments that form the basis of this chapter. Their systematic nature oers
245
246
Asymptotic analysis
x = Lx ,
y = hy ,
w = W w ,
(6.2.2)
247
2 w 2 w
+
=0
x 2
y 2
with
w
= 0 on y = 0, 1,
y
(6.2.3)
with
w0
= 0 on y = 0, 1.
y
(6.2.4)
Hence w0 = w0 (x), but this function is not determined at this stage and
we must proceed further in the asymptotic expansion. Collecting all O 2
terms, we nd that
d2 w0
2 w2
=
y 2
dx2
with
w2
= 0 on y = 0, 1.
y
(6.2.5)
(6.2.6)
This solvability condition for w2 gives rise to the thin sheet equation
d2 w0
=0
dx2
(6.2.7)
for w0 .
To complete the model, we need to derive boundary conditions for (6.2.7).
We do this by subjecting the ends of the plate to a mathematical magnifying
glass. We rst focus on the end x = 0 by setting x =
x and letting the
displacement in this boundary layer near the edge be given by the function
w(x, y) = w(
x, y).
(6.2.8)
2w
+
=0
2
2
x
with
= 0 on y = 0, 1
y
(6.2.9)
and w
= f (y) prescribed on x
= 0. Letting now w
w
0 + w
1 + , using
248
Asymptotic analysis
a0
+
an en x cos ny,
w
0 =
2
(6.2.10)
n=1
x0
(6.2.11)
w0 (0) =
f (y) dy,
(6.2.12)
g(y) dy.
(6.2.13)
and similarly
w0 (1) =
These are the crucial boundary conditions for the thin sheet model (6.2.7):
the only information that w0 has about f and g is their average values.
Now we undertake the more challenging task of extending this asymptotic
argument to the bending of elastic plates.
on z = h/2.
(6.3.1)
249
y = Ly ,
z = hz .
(6.3.2)
v = W v ,
w = W w .
(6.3.3)
(6.3.4b)
,
zz = 2 zz
(6.3.4c)
(6.3.4a)
where is to be determined. A suitable choice may then be obtained by balancing the inplane stresses and strains in the constitutive relation (1.11.1),
which leads to
=
2 EW
,
h
(6.3.5)
250
Asymptotic analysis
t .
t=
(6.3.6)
2
E
To summarise, the geometry of the plate has driven our nondimensionalisation (6.3.2) of the spatial variables. Our scalings (6.3.3) of the displacements amount to assumptions about how big the transverse and inplane
deections are; thereafter, stress components are rescaled so as to balance
as many terms as possible in the momentum equations. We will now derive
the leadingorder model that results from these scalings.
(6.3.7a)
(6.3.7b)
(6.3.7c)
where
G=
gh2
4 EW
(6.3.8)
on z = 1/2.
(6.3.9)
Txy
Txx
+
,
x
y
0=
Txy
Tyy
+
x
y
(6.3.10a)
and
Ny
2w
Nx
+
G,
=
2
t
x
y
(6.3.10b)
251
where w denotes the average of w across the plate, and the net inplane and
transverse stresses are dened as in Section 4.6 by
1/2
1/2
Tij =
ij dz,
Nj =
jz dz.
(6.3.11)
1/2
1/2
Myy
Myx
+
Nx ,
x
y
Mxy
Mxx
+
+ Ny ,
x
y
0=
Myy = Mxx =
(6.3.12)
(6.3.13a)
1/2
1/2
1/2
zxy dz.
(6.3.13b)
M
)
G.
yy
xx
t2
x2
xy
y 2
(6.3.14)
Hence the deformation of the plate is governed by the averaged equations (6.3.10a) and (6.3.14). We must now relate the averaged stresses and
moments to the displacements by using the dimensionless constitutive equations. These take the form
2 xx =
xy =
2 yy =
2 xz =
2 yz =
4 zz =
2 (1 )ux + 2 vy + wz
,
(1 + )(1 2)
u y + vx
,
2(1 + )
2 ux + 2 (1 )vy + wz
,
(1 + )(1 2)
u z + wx
,
2(1 + )
vz + w y
,
2(1 + )
2 ux + 2 vy + (1 )wz
,
(1 + )(1 2)
(6.3.15a)
(6.3.15b)
(6.3.15c)
(6.3.15d)
(6.3.15e)
(6.3.15f)
where is Poissons ratio and, in this chapter only, we will occasionally use
subscripts as shorthand for partial derivatives of the displacements. When
we take the limit 0 in (6.3.15), we treat all the other parameters as
252
Asymptotic analysis
ux + vy
2
zz =
,
1
1 2
yy
ux + vy
2
zz =
.
1
1 2
(6.3.16)
This allows us simply to let tend to zero in the above equations without
having to write the dependent variables as explicit asymptotic expansions.
With w independent of z (to leading order), we may integrate (6.3.15d)
and (6.3.15e) to obtain
u(x, y, z, t) = u(x, y, t) zwx (x, y, t),
(6.3.17a)
(6.3.17b)
as 0. The rst terms on the righthand sides give purely inplane displacements, while the terms proportional to z correspond to bending of the
plate.
By substituting (6.3.17) back into (6.3.15) and integrating, we nd that
the averaged stresses and moments satisfy
Txx =
ux + v y
,
1 2
Myx =
Txy =
uy + v x
,
2(1 + )
wxx + wyy
,
12(1 2 )
Myy = Mxx
Tyy =
ux + v y
,
1 2
wxx + wyy
,
12(1 2 )
wxy
.
=
12(1 + )
Mxy =
(6.3.18)
(6.3.19a)
(6.3.19b)
We are reassured by the fact that the expressions (6.3.18) for the inplane
tensions agree with the exact biaxial straining solutions (2.2.19). Equally, the
relations between the bending moments and the second derivatives
of w reproduce those found in (4.6.8) by ad hoc comparison with exact
plane stress solutions.
Recalling that Txx , Txy and Tyy satisfy (6.3.10a), we see that the inplane
displacements satisfy a plane strain problem. As explained in Section 2.6,
one approach to this problem is to introduce an Airy stress function A(x, y)
253
such that
Txx =
2A
,
y 2
Txy =
2A
,
xy
Tyy =
2A
,
x2
(6.3.20)
(6.3.21)
Also, by substituting (6.3.19) into (6.3.14), we nd that the transverse displacement w satises the linear plate equation
wtt =
4 w
G.
12(1 2 )
(6.3.22)
254
Asymptotic analysis
z
y
x
y
z
xy = y ,
xz = z
at
x = 0,
(6.4.1)
on the dimensionless stress components; see Figure 6.1 for the sign convention. The tractions x , y and z may in general be any functions of y
and z.
In the plate solutions constructed in Section 6.3, the inplane stress components are linear functions of z of the form
xx = Txx + 12zMyx ,
xy = Txy 12zMxx ,
yy = Tyy 12zMxy ,
(6.4.2a)
2w
1
2
z .
+G z
t2
4
(6.4.2b)
(6.4.2c)
Hence, unless the applied tractions have exactly the same dependence on
z, it will be impossible to apply the boundary conditions (6.4.1) directly.
Moreover, even if the boundary tractions do by chance have a zdependence
consistent with (6.4.2), we can see that there is still a boundary layer to be
negotiated before the conditions (6.4.1) can be imposed. To illustrate the
255
(6.4.3)
However, only four boundary conditions can be imposed on our model for
the plate, so at least one of these must be neglected.
To address this issue, as in Section 6.2 we now focus our attention on a
thin boundary layer in which x =
x, where x
is O (1). To get a nontrivial
balance in the Navier equations (6.3.7), we also nd it necessary to rescale
some of the stress components as follows:
xz =
xz
,
yz =
yz
,
zz =
zz
.
2
(6.4.4)
In view of (6.3.4), all stress components are now of the same order of magnitude. In particular, the transverse stress is two orders of magnitude larger
near the edge of the plate than it is in the interior. This gives us a strong
clue that the boundary layer may have an unexpected inuence on the rest
of the plate.
y
z
yy
yz
xy
+
+
,
0=
x
y
z
yz
xz
zz
0=
+
+
.
x
y
z
0=
(6.4.5a)
(6.4.5b)
(6.4.5c)
on z = 1/2,
(6.4.6a)
on x
= 0.
(6.4.6b)
256
Asymptotic analysis
v
+ 2
+
,
x
y
z
v
u
2(1 + )
xy =
+
,
y
x
v
2 (1 + )(1 2)
+ 2 (1 )
+
,
yy =
x
y
z
u
w
22 (1 + )
+
,
xz =
z
x
v w
2(1 + )
yz =
+
,
z
y
w
v
2 (1 + )(1 2)
+ 2
+ (1 )
.
zz =
x
y
z
2 (1 + )(1 2)
xx = (1 )
(6.4.7a)
(6.4.7b)
(6.4.7c)
(6.4.7d)
(6.4.7e)
(6.4.7f)
As in the antiplane strain example of Section 6.2, we must match the two
solutions: the asymptotic behaviour of the inner solution as x
must be
the same as that of the outer solution as x 0. In other words, we impose
that
2w
u
(0, y) z 2 (0, y) + , (6.4.8a)
x
x
2w
v
w
(0, y) +
x
(0, y) z
(0, y) + , (6.4.8b)
v v(0, y) z
y
x
xy
w
(0, y) +
x
u
u(0, y) z
x
w
(0, y) + ,
x
Txx (0, y) + 12zMyx (0, y) + ,
w
w(0, y) +
x
xx
(6.4.8c)
(6.4.8d)
(6.4.8e)
(6.4.8f)
(6.4.8g)
(6.4.8h)
257
(6.4.9)
(0)
(0)
(0)
(0)
(0)
xx
xz
xy
yz
xz
zz
+
, 0=
+
, 0=
+
,
x
z
x
z
x
z
and the integrated versions of these equations are, simply,
1/2
1/2
1/2
(0)
(0)
dz =
dz =
(0) dz.
0=
x
1/2 xx
x
1/2 xy
x
1/2 xz
0=
(6.4.10)
(6.4.11)
Hence the average leadingorder stresses are constant over the entire boundary layer and, in view of (6.4.6), (6.4.8d), and (6.4.8e), we immediately
deduce that the two boundary conditions to be applied to (6.3.21) are
1/2
x (y, z) dz,
(6.4.12a)
Txx (0, y) =
Txy (0, y) =
1/2
1/2
1/2
y (y, z) dz.
(6.4.12b)
In the same way, multiplying (6.4.10a) by z and averaging over the thickness
of the plate, we obtain
1/2
z
(0) dz
(6.4.13)
0=
x
1/2 xx
and therefore, combining (6.4.6) with (6.4.8d),
1/2
zx (y, z) dz.
Myx (0, y) =
(6.4.14)
1/2
Equations (6.4.12) and (6.4.14) show that the net inplane tractions and
their moment along the edges are balanced by the corresponding stresses on
the other side of the boundary layer, as we might have expected physically.
Equation (6.4.14), together with (6.3.19a), provides one boundary condition
for the outer plate equation (6.3.22). One further boundary condition is
needed, and to obtain it we must proceed to the next order in , where
(6.4.5c) yields
(1)
0=
(0)
(1)
yz
zz
xz
+
+
.
x
y
z
(6.4.15)
258
Asymptotic analysis
0=
x
1/2
1/2
(1)
xz
1/2
dz +
1/2
(0)
yz
dz,
y
(6.4.16)
but all we can say thus far, using (6.4.6) and (6.4.8f), is that
Nx (0, y)
1/2
1/2
z (y, z) dz =
1/2
1/2
(0)
yz
d
xdz.
y
(6.4.17)
Hence we cannot entirely escape from resolving the details of the displacements and stresses in the boundary layer if we wish to establish this last
boundary condition for the model in the interior of the plate.
Expanding u
, v and w
as in (6.4.9), we nd from the constitutive relations
(6.4.7) that the leadingorder displacements are all independent of x
. The
matching conditions (6.4.8) then give
(0, y) z
u
(0) = u
w
w
(0, y), v(0) = v(0, y) z
(0, y), w
(0) = w(0, y).
x
y
(6.4.18)
u
(0)
w
(1)
+
= 0,
z
x
(6.4.19)
w
(1) = x
w
(0, y).
x
(6.4.20)
Now, from the second Navier equation (6.4.5b), we obtain the antiplane
strain problem
(0)
(0)
yz
xy
+
= 0,
x
z
(6.4.21)
v (1)
u
(0)
+
,
y
x
(0)
2(1 + )
yz
=
w
(1)
v (1)
+
.
z
y
(6.4.22)
x
2
z 2
(6.4.23)
259
v (1)
= 2(1 + )y
x
y
(1)
(1)
w
v
=
z
y
at x
= 0,
(6.4.24a)
1
at z = .
2
(6.4.24b)
v (1)
x
(0, y),
z
xy
(6.4.25)
as x
, while combining (6.4.8a) and (6.4.8e) yields
u
2w
v (1)
(0, y) + z
(0, y) + 2(1 + ) Txy (0, y) 12zMxx (0, y) .
x
y
xy
(6.4.26)
By crossdierentiation, we see that these two conditions are consistent only
if
2w
(0, y) = 12(1 + )Mxx (0, y),
(6.4.27)
xy
and this reassuringly reproduces the outer constitutive relation (6.3.19b).
After using this relation, we can write v(1) in the form
u
(1)
(0, y) x
= 0,
z
0,
(6.4.29a)
1
z= ,
2
(6.4.29b)
x
= 0,
(6.4.29c)
x
.
(6.4.29d)
2 d
ds = Txy (0, y)
xdz =
y (y, z) dz, (6.4.30)
0=
n
1/2
where the integration region is the strip [0, ] [1/2, 1/2], and we have
260
Asymptotic analysis
already established that the righthand side above vanishes. Hence we can
solve for by separating the variables:
an (y) cos n(z + 1/2) en x ,
(6.4.31)
n=1
where
an (y) =
2
n
1/2
1/2
Fy (y, z) cos n(z + 1/2) dz,
(6.4.32a)
and
Fy (y, z) = 12Mxx (0, y)z + y (y, z).
(6.4.32b)
,
x
(0)
yz
=
,
z
(6.4.33)
(0)
= 0 to the
giving the transition of xy from the prescribed function y on x
expected linear zdependence in the outer region. This therefore illustrates
the antiplane strain version of SaintVenants principle: the details of the
traction applied to the edge are lost outside the narrow boundary layer.
Only the net force and moment exerted by the edge traction are transmitted
to the interior of the plate.
At last, we are in a position to evaluate (6.4.17) as
1/2
z (y, z) dz =
Nx (0, y)
y
1/2
(
x, y, 1/2) (
x, y, 1/2) d
x.
(6.4.34)
Substituting for from (6.4.32), we get
(
x, y, 1/2) (
x, y, 1/2) d
x=
=2
1/2
=
1/2
Fy (y, z)
1 (1)n
n=1
1 (1)n
n=1
n2 2
an (y)
cos n(z + 1/2)
dz
1/2
1/2
(6.4.35)
261
Mxx (0, y) +
z (y, z) dz =
y (y, z)z dz .
Nx (0, y)
y
1/2
1/2
(6.4.36)
We easily recognise in the lefthand side of (6.4.36) the dierence between
the vertical shear forces on either side of the boundary layer. The righthand side represents the rate of change of the net bending moment about
the xaxis that is produced by the outer moment and the edge traction. All
that matters for the plate model (6.3.21), (6.3.22) is that both contributions
balance; this is akin to the angular momentum balance (4.4.3) that arises
in linear beam theory. Both sides of (6.4.36) are statically equivalent in the
interior of the plate. Neither the transverse stress nor the twisting moment
is conserved across the boundary layer; instead (6.4.36) represents a playo
between the two eects.
To summarise, we have obtained the four boundary conditions (6.4.12),
(6.4.14) and (6.4.36) that apply at a straight edge of a plate when specied
tractions are imposed. At a stressfree edge, for example, these take the form
Txx = Txy = Myx = Nx
Mxx
= 0,
y
(6.4.37)
262
Asymptotic analysis
6.5.2 Scalings
We consider a uniform plate of thickness h whose centresurface lies in the
plane Z = 0, where now (X, Y, Z) are Lagrangian coordinates which must be
distinguished from Eulerian coordinates (x, y, z). The coordinates are scaled
with their typical values as follows:
X = LX ,
Y = LY ,
Z = hZ .
(6.5.1)
263
(6.5.3c)
(6.5.4a)
(6.5.4b)
(6.5.4c)
where now
G=
0 gh
.
4 E
(6.5.5)
on Z = 1/2.
(6.5.6)
T 12
T 11
+
,
X
Y
0=
T 21
T 22
+
,
X
Y
(6.5.7)
264
Asymptotic analysis
T 31
2w
T 32
=
+
G,
t2
X
Y
(6.5.8)
M22
M12
M11
M21
+
T 13 ,
+
+ T 23 ,
0=
X
Y
X
Y
where the averaged stresses and bending moments are dened by
1/2
T ij =
Tij dZ,
0=
M11 M12
=
M21 M22
(6.5.9)
(6.5.10)
1/2
1/2
T21 T22
Z dZ.
T11
T12
1/2
(6.5.11)
2 u Y
uZ
2 u X
F =I +
(6.5.12)
2 vX
2 vY
vZ .
3
3
2
wX + w
X wY + w
Y w
Z
Recall from Section 5.2.2 that the rst PiolaKirchho stress tensor is not
symmetric, but satises
T F T = F T T.
(6.5.13)
This provides three independent scalar equations which, using (6.5.3) and
(6.5.12), may be reduced to
(6.5.14a)
T21 = T12 + O 2 ,
2
T31 T13 = wX T11 + wY T12 + O ,
(6.5.14b)
2
(6.5.14c)
T32 T23 = wX T21 + wY T22 + O .
The inplane averaged stress tensor is therefore symmetric to leading order,
and, as in Section 6.3, we can deduce from (6.5.7) the existence of an Airy
stress function A(X, Y ) such that
T 11 =
2A
,
Y 2
T 21 = T 12 =
2A
,
XY
T 22 =
2A
X 2
(6.5.15)
+
G
=
+
t2
X 2
XY
XY
Y 2
2
2
2
2w 2A
w A
w 2A
+
+
2
. (6.5.16)
Y 2 X 2
XY XY
X 2 Y 2
265
0
0
u Z + wX
1
E=
F TF I =
0
0
vZ + wY
2
2
u Z + wX vZ + wY
0
2
u Y + vX + wX wY
0
2uX + wX
2
+
u Y + vX + wX wY
2vY + wY2
0
2
2
2
0
0
2w
Z + u Z + vZ
+ O 3 . (6.5.17)
The smallness of justies assuming that the second PiolaKirchho stress
tensor S is linear and, arguing as in Section 5.3.6, we write, in dimensional
variables,
S = Tr (E)I + 2E.
The dimensionless form of (6.5.18) is
Tr (E)I +
E.
2 S12 S22 S23 =
(1 + )(1 2)
1+
2
S13 S23 S33
(6.5.18)
(6.5.19)
v = v(X, Y, t) ZwY .
(6.5.21)
2 + w 2 + 2(u + v ) + 2(1 )w
Z
wX
X
Y
Y
+ O 2
2(1 + )(1 2)
(6.5.22)
and, hence,
w
Z =
2 + w 2 + 2(u + v )
wX
X
Y
Y
.
2(1 )
(6.5.23)
had we used a nonlinear relation of the form (5.3.23), the smallness of would have led to a
similar theory in which some of the coecients in the lowestorder model were modied as in
the example
of Section 5.2.4; the correction to the stressstrain relation would be at most of
O 2
266
Asymptotic analysis
We can now use (6.5.21) and (6.5.23) to evaluate the remaining stress components as
2 + (2v + w 2 )
2uX + wX
Y
Y
,
2(1 2 )
u Y + vX + wX wY
,
=
2(1 + )
T11 =
T21 = T12
T22 =
2 ) + 2v + w 2
(2uX + wX
Y
Y
,
2(1 2 )
(6.5.24a)
(6.5.24b)
(6.5.24c)
up to order 2 .
We now have all the ingredients we need to construct the von K
arman
equations. By averaging (6.5.24), we obtain
T 11 =
T 12 =
2 + (2v + w 2 )
2uX + wX
2A
Y
Y
=
,
2
2
Y
2(1 )
2A
u Y + v X + wX wY
=
,
XY
2(1 + )
T 22 =
2 ) + 2v + w 2
(2uX + wX
2A
Y
Y
,
=
X 2
2(1 2 )
(6.5.25a)
(6.5.25b)
(6.5.25c)
2
wX
2A
2A
=
,
2
Y 2
X 2
vY +
wY2
2A
2A
=
.
2
X 2
Y 2
(6.5.26)
= 0.
(6.5.27)
A+
2
2
X Y
XY
Expressions for the bending moments are found by multiplying (6.5.24)
with respect to Z and then integrating:
2
w
1
2w
M21 =
,
(6.5.28a)
+
12(1 2 ) X 2
Y 2
2w
1
,
12(1 + ) XY
2
w
1
2w
.
=
+
12(1 2 )
X 2
Y 2
M11 = M22 =
M12
(6.5.28b)
(6.5.28c)
267
2
. (6.5.29)
=
+
t2
12(1 2 )
Y 2 X 2
XY XY
X 2 Y 2
With a suitable change of notation, (6.5.27) and (6.5.29), are the von
K
arman equations (4.7.11), (4.7.12) for transverse motions of a thin elastic
plate. In principle the approach of Section 6.4 can be adapted to determine
the boundary conditions to be applied to (6.5.27) and (6.5.29) when specied
edge tractions are given. However, since the strain in any boundary layer
region will be small, we can anticipate that the results of such an analysis
would be analogous to those found in Section 6.4, with just a rigidbody
translation and rotation superimposed.
Z = hZ ,
x = Lx ,
z = Lz .
(6.6.1)
268
Asymptotic analysis
In contrast to (6.5.3), the rst PiolaKirchho stress components are nondimensionalised using
(T11 , T31 ) = E T11 , T31 ,
(T13 , T33 ) = 2 E T13 , T33 ,
(6.6.2)
and the timescale is chosen to make t of O (1), where
)
h 0
t.
t= 2
(6.6.3)
(6.6.4a)
(6.6.4b)
0 gh
= G.
3 E
(6.6.5)
We can relate the stress components in (6.6.4) using the symmetry property
(6.5.13) which, in two dimensions, reduces to
zX T11 + zZ T13 = xX T31 + xZ T33 ,
(6.6.6)
on Z = 1/2,
(6.6.7)
T 11
2x
,
=
t2
X
2z
T 31
G,
=
t2
X
(6.6.8)
(zxt xzt ) dZ =
(zT11 xT31 ) dZ + Gx, (6.6.9)
t 1/2
X 1/2
which represents net conservation of angular momentum.
269
Z
X
z (0)
= sin ,
X
(6.6.11)
where (X, t) is the angle between the centreline of the plate and the xaxis. Finally, we suppose that each transverse section through the plate is
to leading order simply rotated through the angle , as shown in Figure 6.2,
so that
x(1) = Z sin ,
z (1) = Z cos .
(6.6.12)
270
Asymptotic analysis
+
+O 2 , (6.6.13)
F =
(2)
sin cos
Xj
ZX sin zZ
in which the leadingorder term is simply a rotation through the angle .
Hence, the strain tensor is given by
(2)
(2)
1 T
2ZX
xZ cos + zZ sin
E=
F F I
(2)
(2)
(2)
2
2 x(2)
Z cos + zZ sin 2xZ sin + 2zZ cos
+ O 2 . (6.6.14)
Since the strain is small, we can again limit our attention to the mechanically linear dimensional constitutive relation
S = F 1 T = Tr (E)I + 2E.
(6.6.15)
We nondimensionalise (6.6.15) using (6.6.2) and expand the stress components in powers of , writing
(0)
(1)
(0)
(1)
T11 = T11 + T11 + O 2 ,
T31 = T31 + T31 + O 2 ,
(6.6.16)
to obtain
(0)
(0)
(2)
(2)
(1 )ZX + zZ cos xZ sin
(1 + )(1 2)
(6.6.17a)
(2)
0=
(2)
(0)
T31
cos
(0)
T11
(2)
xZ cos + zZ sin
,
2(1 + )
(6.6.17b)
(2)
x cos + zZ sin
,
(6.6.17c)
sin = Z
2(1 + )
(2)
(2)
ZX + (1 ) zZ cos xZ sin
0=
,
(1 + )(1 2)
(6.6.17d)
T11 =
ZX cos
,
1 2
(0)
T31 =
ZX sin
.
1 2
(6.6.18)
271
(0)
(1)
Tx = T 11 ,
Tz = T 31 ,
(6.6.19)
2z
Tz
G.
=
2
t
X
(6.6.20)
Next we use (6.6.10) and (6.6.16) to expand x, z and the stress components
in the angular momentum equation (6.6.9) and simplify the result to obtain
M
+ Tx sin Tz cos = 0,
X
where the bending moment is given by
1/2
(0)
(0)
M=
T11 cos + T31 sin Z dZ =
1/2
X
.
12(1 2 )
(6.6.21)
(6.6.22)
Equations (6.6.20) and (6.6.21) are equivalent to the dynamic beam model
(4.11.8), when we identify the tangential and normal internal stresses with
N = Tz cos Tx sin .
T = Tx cos + Tz sin ,
(6.6.23)
However, since we have modelled the bending of a plate rather than a beam,
we have obtained the constitutive relation (6.6.22) between bending moment and curvature, rather than (4.11.9). One advantage of our systematic
approach is that we can determine the details of the internal stresses and
strains which are not predicted by the ad hoc theory of Section 4.9. As an
illustration, we will now show briey how the longitudinal stretching of the
plate can be found.
Z
,
T
Z
. (6.6.24)
T13 =
33
2(1 2 )
4
2(1 2 )
4
(2)
(2)
Z 2 X sin
,
2(1 )
z (2) = b(x, t) +
Z 2 X cos
,
2(1 )
(6.6.25)
272
Asymptotic analysis
where a and b are as yet arbitrary. Note that the centreline of the plate is
deformed to
(0)
x(X, 0, t)
x (X, t) + 2 a(X, t) +
,
(6.6.26)
xc (X, t) =
z(X, 0, t)
z (0) (X, t) + 2 b(X, t) +
so that
xc
cos
2 aX
+
+ .
sin
bX
(6.6.27)
(6.6.28)
(3)
(3)
+
zZ cos xZ sin ,
(1 + )(1 2)
(6.6.29a)
and
2
1 2 + 4Z 2 (1 3 + 2 ) X
0=
8(1 2 )(1 2)
273
(6.6.31)
Equation (6.6.31) tells us how much the plate stretches along its centreline. Recall that T = (Tx cos + Tz sin ) is the tension in the plate, so we
should not be surprised that the stretch increases linearly with T . Perhaps
less intuitive is the nal term which is quadratic in the curvature and tells
us that bending the plate inevitably causes it to shrink.
A
,
(6.7.1)
=
L
where L is the rod length and A denotes the area of the crosssection. As
with the linear plate equation, we suppose that the rod undergoes transverse
displacements of order W , small enough for the linear Navier equations to
be valid.
The scalings that apply to the present situation are analogous to those
employed in Section 6.3, namely
)
L
t=
t ,
(6.7.2a)
E
(x, y, z) = L(x , y , z ),
(6.7.2b)
(6.7.2c)
(u, v, w) = W (u , v , w ),
EW
, xz
, yy
, yz
, zz
(xx , xy , xz , yy , yz , zz ) =
, (6.7.2d)
xx , xy
L
so that the dimensionless Navier equations (with primes dropped) take the
form
xy
xz
2u
xx
+
+
,
= 2
2
t
x
y
z
yy
yz
xy
2v
+
+
,
2 2 =
t
x
y
z
yz
zz
xz
2w
2 2 =
+
+
2 G,
t
x
y
z
4
(6.7.3a)
(6.7.3b)
(6.7.3c)
274
Asymptotic analysis
The unit normal n to the boundary D lies entirely in the (y, z)plane,
that is n = (0, ny , nz )T . If no surface tractions are applied to the rod, we
therefore have the boundary conditions
ny xy + nz xz = ny yy + nz yz = ny yz + nz zz = 0
on D.
(6.7.5)
By integrating (6.7.3) with respect to y and z, using the divergence theorem and applying the boundary conditions (6.7.5), we obtain the averaged
equations
2
T
2u
,
=
2
t
x
Ny
2v
,
=
2
t
x
Nz
2w
2 G,
=
2
t
x
where bars now denote the average over the crosssection, that is
f=
f dydz
(6.7.6)
(6.7.7)
Ny = xy ,
Nz = xz .
(6.7.8)
2
Mx
,
(yw zv) =
2
t
x
(6.7.9)
where
Mx = yxz zxy
(6.7.10)
2
Mz
Ny ,
(yu) = 2
2
t
x
My
2
Nz ,
(zu) = 2
2
t
x
(6.7.11)
275
where
Mz = yxx
My = zxx ,
(6.7.12)
,
t2
x2
2 My
2w
=
G,
t2
x2
(6.7.13)
(6.7.14)
u y + vx
,
2(1 + )
u z + wx
,
=
2(1 + )
vz + w y
.
=
2(1 + )
2 xx =
xy =
(6.7.15a)
2 yy
xz
(6.7.15b)
2 zz
2 yz
(6.7.15c)
(6.7.16)
At leading order, we deduce from (6.7.15) that the transverse displacements satisfy
w(0)
v (0)
w(0)
v (0)
=
=
+
=0
y
z
z
y
(6.7.17)
276
Asymptotic analysis
= 0,
+
=
y 2
z 2
xy
xz
subject to the Neumann boundary condition
a
c
b
c
u(0)
u(0)
+
z
+ nz
+
+y
=0
ny
y
x
x
z
x
x
(6.7.19a)
on D.
(6.7.19b)
The boundaryvalue problem (6.7.19) may be simplied by writing u(0) in
the form
u(0) (x, y, z, t) = U (x, t) y
b
c
a
z
+ (y, z) ,
x
x
x
(6.7.20)
z + nz
+y =0
ny
y
z
in D,
(6.7.21a)
on D.
(6.7.21b)
c
,
x
where
R=
(1 + )
(6.7.22)
277
(0)
xz
xy
=
= 0,
x
x
(6.7.23)
(0)
(0)
yz
yy
+
= 0,
y
z
(0)
yz
zz
+
= 0.
y
z
(6.7.24)
These are just the steady plane strain equations and, subject to the zerotraction boundary conditions (6.7.5), imply that the leadingorder transverse
stress components are all zero:
(0)
(0)
(0)
= yz
= zz
= 0.
yy
(6.7.25)
(1)
(1)
(0)
=
xx
(1 )ux + vy + wz
,
(1 + )(1 2)
(0)
yy
=0=
ux + (1 )vy + wz
,
(1 + )(1 2)
(0)
(1)
(0)
(0)
zz
(1)
(6.7.26a)
(1)
(6.7.26b)
(1)
ux + vy + (1 )wz
,
=0=
(1 + )(1 2)
(6.7.26c)
U
2a
u(0)
2b
=
y 2 z 2.
x
x
x
x
(6.7.27)
By integrating over D and recalling that y and z are zero, we deduce that
the tension is equal to the axial stretch,
T =
U
,
x
(6.7.28)
2a
2b
I
,
yz
x2
x2
2a
2b
= My = Iyz 2 Izz 2 ,
x
x
yxx = Mz = Iyy
(0)
zxx
(6.7.29a)
(6.7.29b)
278
Asymptotic analysis
(6.7.30)
I
,
yy
yz
t2
x4
x4
2b
4a
4b
=
I
I
G,
yz
zz
t2
x4
x4
(6.7.31)
(6.8.1)
in which 1 and 2 are two spatial parameters. Since we restrict our attention
to linear elasticity, there is no need to distinguish between Lagrangian and
Eulerian coordinates.
We will use curvilinear coordinates (1 , 2 , n) to describe any point in the
plate whose position vector is
r = r c (1 , 2 ) + nn,
(6.8.2)
= 0,
1 2
2 rc
n=0
1 2
(6.8.3)
279
(see for example Kreyszig, 1959). This choice simplies the derivations but
does not involve any loss of generality, since we will nd that the nal equations may readily be expressed in any other convenient coordinate system.
We set
r c
r c
,
,
a2 (1 , 2 ) =
(6.8.4)
a1 (1 , 2 ) =
1
2
and hence dene an orthonormal basis {e1 , e2 , n}, where
e1 =
1 r c
,
a1 1
e2 =
1 r c
,
a2 2
n = e1 e2 .
(6.8.5)
e1
1 a2
=
e2
2
a1 1
e2
1 a1
=
e1 ,
1
a2 2
n
= 2 a2 e2 ,
2
(6.8.6a)
(6.8.6b)
(6.8.6c)
r
= a2 (1 2 n)e2 ,
2
r
= n,
n
(6.8.7)
h2 = a2 (1 2 n),
h3 = 1.
(6.8.8)
(1 , 2 ) = L(1 , 2 ),
(1 , 2 ) =
(6.8.9b)
280
Asymptotic analysis
where W is small enough for linear elasticity to apply. The Cauchy stress
tensor, relative to the curvilinear coordinate system, is now scaled using
(11 , 12 , 22 ) =
EW
2 EW
11 , 12 , 22 , (13 , 23 , 33 ) =
13 , 23 , 33 ,
L
L
(6.8.9c)
(6.8.9d)
gL2
(k e1 ) ,
EW
g2 =
gL2
(k e2 ) ,
EW
g3 =
gL2
(k n) .
EW
(6.8.10)
a1
a2
2 u1
,
g1 =
(a2 11 ) +
(a1 12 ) + 12
22
t2
a1 a2 1
2
2
1
(6.8.11a)
a2
a1
2 u2
,
g2 =
(a2 12 ) +
(a1 22 ) + 12
11
t2
a1 a2 1
2
1
2
(6.8.11b)
2w
g3 = 1 11 + 2 22 ,
(6.8.11c)
t2
at leading order. Once we have used the constitutive relations to express
11 , 12 and 22 in terms of the displacements, (6.8.11) will give us three
equations for u1 , u2 and w.
The constitutive relations imply that the displacement components are
all independent of n to leading order. Furthermore, the dimensionless constitutive relation for 33 reads
e11 + e22 + (1 )e33
= 2 33 ,
(1 + )(1 2)
(6.8.12)
281
which we can use to express e33 in terms of e11 and e22 , and hence obtain
11 =
e11 + e22
,
1 2
22 =
e11 + e22
.
1 2
(6.8.13)
(6.8.14a)
(6.8.14b)
(6.8.14c)
(6.8.15)
2u
,
=
g
2
t
2w
g3 = K : ,
(6.8.16)
t2
is the surface gradient
= u wn is the inplane displacement,
where u
operator, is inplane stress and K is the curvature tensor. In the same
notation, the inplane stress satises the constitutive relation
1
E+
Tr E I,
(6.8.17)
=
2
1+
1
where the surface strain tensor is given by
1
u)T wK.
(6.8.18)
u + (
E =
2
The theory just derived describes deformations in which the inplane tensions dominate and the bending stiness is negligible. However, our model
allows for nontrivial displacements that cause zero leadingorder stress. For
example, if we let the shell tend to a at sheet so that 1 = 2 = 0, then the
normal displacement w decouples entirely from the inplane stresses and is
arbitrary in the absence of bending stiness. Similarly, a cylinder bent about
282
Asymptotic analysis
a2 2 0
u1
0 a1 1
u1
+
0
a2 1 u2
a1
0
2 u2
1 u1 a2 /1 2 u2 a1 /2
. (6.8.19)
=
u1 a1 /2 + u2 a2 /1
It is easy to calculate that the characteristic curves for this system are given
by
)
a1
1
d2
=
.
(6.8.20)
d1
a2
2
Hence (6.8.19) is elliptic, parabolic or hyperbolic, depending on the sign of
the Gaussian curvature, exactly as in Section 4.8, and we can expect the
existence of a zerostress mode compatible with given boundary conditions
to depend strongly on this type.
In cases where the imposed boundary conditions are compatible with
(6.8.19), the displacement eld is not uniquely determined by our model.
The resolution of the indeterminacy would require us to consider a geometrically nonlinear limit similar to that which led to the von K
arman theory in
Section 6.5. This would describe weakly nonlinear deformations of a shell, in
which the inplane stresses and bending stiness balance, but this is beyond
the scope of this book.
Exercises
283
Exercises
6.1
6.2
Repeat the linear rod derivation from Section 6.7 using the fast timescale
)
t .
t = L
E
Show that the decompositions (6.7.18), (6.7.20) still apply, where the
functions a, b, c and U satisfy the equations
2a
= 0,
t2
2b
= 0,
t2
(Iyy + Izz )
2c
2c
=
R
,
t2
x2
2U
2U
=
.
t2
x2
[This shows that longitudinal and torsional waves occur over a much
shorter timescale than the exural waves considered in Section 6.7.]
284
6.3
Asymptotic analysis
H
H
H
H
T11 +
T12 ,
T21 +
T22 ,
T23 =
X
Y
X
Y
H
H
T31 +
T32 ,
T33 =
X
Y
on Z = H 1/2.
Use the asymptotic approach of Section 6.5 to derive the equations
2
2w
XY
2H 2w
2H 2w
2H 2w
+
+
2
Y 2 X 2
XY XY
X 2 Y 2
2w 2w
A=
X 2 Y 2
4
and
2 (H + w) 2 A
2w
4 w
=
+
G
+
t2
12(1 2 )
Y 2
X 2
2 (H + w) 2 A
2 (H + w) 2 A
+
2
.
XY XY
X 2
Y 2
6.4
Show that these reduce to (4.8.9) and (4.8.11) when w H.
Show that the strain tensor in Section 6.6 is given by
2 1
1
x2X + zX
1 (x
X xZ + zX zZ )
.
E=
2 1 (xX xZ + zX zZ ) 2 x2Z + zZ2 1
Hence, show that T is O 2 E unless (6.6.10) is assumed. Further,
show that the components of S must be at most O (E) for the
scaling (6.6.2) to hold and therefore that
2
2
2
2
x(0)
z (0)
x(1)
z (1)
+
= 1,
+
= 1,
X
X
Z
Z
z (0) z (1)
x(0) x(1)
+
= 0.
X Z
X Z
From this, justify (6.6.11) and (6.6.12).
Exercises
6.5
285
u y + vx
,
2(1 + )
u z + wx
=
,
2(1 + )
vz + w y
,
=
2(1 + )
2 xx =
xy =
2 yy
xz
2 zz
2 yz
2 ur
,
gr =
t2
r
2 u
z
1
+
,
g =
t2
r
z
2 uz
zz
1 z
+
,
gz =
t2
r
z
for a circular cylindrical shell of radius r.
(c) Show that displacements of the form u = f (), ur = f (),
uz = 0 generate no strain and hence no stress in the leadingorder shell theory.
(d) Show also that such a displacement exists for a cylindrical
shell with any crosssection.
[This shows that a cylinder can undergo deformations that are
one order of magnitude larger if bent along a generator than
otherwise.]
6.7
286
Asymptotic analysis
radius r are
+
2 ur
,
gr =
2
t
r
2 u
1
(sin ) +
cos ,
g =
t2
r sin
2 u
1
(sin ) +
+ cos .
g =
t2
r sin
(c) Suppose that the strain is identically identically zero. Eliminate ur and u (or use (6.8.19) directly) to show that u
satises the equation
2 u
u
2
sin
= 0,
+ sin
2
sin
while
u
u
= sin2
,
sin
and ur =
u
.
u = sin ,
(ii) u = sin log tan(/2) ,
(iii) u = sin(k) sin tank (/2),
and nd the corresponding expressions for u and ur .
[Hint: show that sin / = / with = log tan(/2).]
(d) Note that (i) is simply a rigid body motion. Show that the
remaining solutions are physically unacceptable if the shell
is a complete sphere. Discuss solutions (ii) and (iii) for hemispherical shells. Show that (ii) gives rise to a multivalued displacement. To which physical situation could it correspond?
7
Fracture and contact
7.1 Introduction
Fracture describes the behaviour of thin cracks, like the one illustrated in
Figure 7.1, in an otherwise elastic material. The crack itself is a thin void,
whose faces are usually assumed to be stressfree. When a solid containing
a crack is stressed, we will nd that the stress is localised near the crack
tip, becoming singular if the tip is sharp. We will see that the strength
of the singularity can be characterised by a stress intensity factor, which
depends on the size of the crack and on the applied stress. This factor determines the likelihood that a crack will grow and, therefore, that the solid will
fail.
On the other hand, contact refers to the class of problems in which two
elastic solid bodies are brought into contact with each other, as illustrated in
Figure 7.2. When the material properties are the same in the two bodies, the
geometrical conguration near the edge of the contact region is apparently
similar to that of fracture, with voids now outside the contact set, that is, the
set of points at which the two solids are in contact. The mathematical setup
of such problems does therefore have some similarity with fracture, but, in
the steady state, there is one crucial dierence. Whereas much of the study
of fracture concerns cracks of prescribed length, in contact problems the
contact set itself is often unknown in advance. They are thus known as free
boundary problems, that is problems whose geometry must be determined
along with the solution. An example of such a problem is to determine how
the contact area between a car tyre and the road depends on the ination
pressure.
The reader may be reassured to know that, although several cracks emanate from many rivets
in an airliner, they are carefully monitored to ensure that none of them is dangerous.
287
288
crack faces
contact set
289
at the surface of the material, which is usually the only place where they
can be discerned by the naked eye.
Brittle fracture covers that class of fractures, such as windscreen shattering, where the initiation and growth of the crack do not aect the material
modelling, which is still governed by the Navier equations. Such a situation
might not be true for a metal, where quite a large region near the crack
tip may have to be modelled using plasticity rather than elasticity equations; then the fracture is called ductile, and we will discuss this further in
Chapter 8.
We can say little more here about the physical background of this vast
and important subject. However, we urge the reader to think intuitively that
fracturing, i.e. crack growth, is basically a local stressrelieving mechanism.
As a crack grows, the strain energy in the evolving fractured conguration
is less than the value it would have taken had the crack not evolved. This
argument suggests that the criterion for a crack to evolve under a small
increase of loading is that the release, W, of strain energy resulting from
a small growth A in the crack face area is such that
W dx > S,
(7.2.1)
where V denotes the region excluding the crack and S is the surface energy
of the area A. In principle, S could be estimated from molecular considerations but, in the theory of Grith cracks (see, for example, Sneddon &
Lowengrub, 1969, Chapter 1; Freund, 1989, Chapter 1) S is thought of as
resulting from an eective surface tension ; in this case, S = A.
It is interesting to compare fracture with the more familiar failure of a
liquid, which cavitates when the pressure is too low, so that the molecular
bonds weaken and allow the molecules to form a vapour. This often results in
the appearance of a spherical bubble, whereas a crack morphology is usually
the one preferred by solids. The dierence in behaviour arises because it is
easier to create new interface between a liquid and a vapour than it is to
create new interface between a solid and a void. Hence a crack can grow
most easily by creating new surface energy just in the vicinity of the crack
tip. It is also instructive to contrast fracture, which often occurs as a result
of some remote loading process, with the drilling of a thin hole through the
solid (for example using a shaped charge; see Section 3.6). It is an everyday
observation that the energy required to drill a hole per unit surface area
generated is greater than that required to cleave the material using an axe.
290
(b)
yz III
y
III
c
III
c
yz = 0
x
y
yz III
Fig. 7.3 (a) A Mode III crack. (b) A crosssection in the (x, y)plane.
(7.2.2a)
as
x2 + y 2 ,
(7.2.2b)
on
y = 0, x < c.
(7.2.2c)
There is a fundamental dierence between (7.2.2) and most of the boundary value problems we have thus far considered for elastostatics, namely
that the boundary on which the Neumann data (7.2.2c) are prescribed is
not smooth at the crack tip. To see the kind of diculty that this can cause,
suppose we were to shift the origin to (c, 0) and concentrate on the region
near the crack tip. Then we would need to nd a displacement eld in which
2 w = 0 everywhere except on y = 0, x > 0, with
w
=0
y
on y = 0, x > 0.
(7.2.3)
291
nA n/21
r
cos(n/2),
2
z =
nA n/21
r
sin(n/2).
2
(7.2.5)
This plethora of solutions gives us the strong hint that we will not be able
to solve (7.2.2) uniquely unless we supply some extra information about the
behaviour of w near (c, 0). We also note that, whenever n is not an even
positive integer, the stress is nonanalytic at the crack tip r = 0, and, if
n < 2, the stress is not even bounded.
From a mathematical point of view it is natural to ask whether the solution
of (7.2.2) must be singular at (c, 0) and, if so, what is the mildest singularity
which we have to endure. To answer these questions, one possibility is to
round o the crack, that is, to replace it with a thin but smooth boundary.
A particularly convenient shape is the ellipse
y2
x2
+
= 1,
2
c2 cosh c2 sinh2
(7.2.6)
which is smooth for all positive but approaches the slit geometry of Figure 7.3(b) in the limit 0. Now we can easily solve Laplaces equation by
introducing elliptic coordinates dened by
z = x + iy = c cosh ,
where
= + i,
(7.2.7a)
so that
x = c cosh cos ,
y = c sinh sin .
(7.2.7b)
in >
(7.2.8a)
with
w
=0
on = ,
(7.2.8b)
292
Im z
r1
r2
1
c
c
Fig. 7.4 Denition sketch for the function
Re z
z 2 c2 .
cIII
e sin
2
as
(7.2.8c)
(7.2.9)
We thus have a unique displacement eld for any positive value of and,
when we let 0, we obtain the solution of the crack problem as
cIII
Im (sinh ) .
(7.2.10)
w=
III
Im
z 2 c2 ,
z 2 c2 r1 r2 ei(1 +2 )/2 ,
(7.2.11)
(7.2.12)
with r1 , r2 , 1 and 2 dened as shown in Figure 7.4. The angles are taken
to lie in the ranges < j < , so that a branch cut lies along the crack
between z = c and z = c (see, for example, Priestley, 2003, Chapter 9). As
shown in Figure 7.5, the displacement w is discontinuous across this branch
cut.
293
y
x
Fig. 7.5 Displacement eld for a Mode III crack.
Despite its simplicity, the formula (7.2.11) lies at the very heart of the theory of brittle fracture. It reveals the famous square root singularity in which
the elastic displacement varies as the square root of the distance from the
crack tip. By rounding o the crack before taking the limit 0, we have
managed to select one of the many possible singular solutions suggested by
(7.2.4). This dependence immediately implies that the stress tensor, whose
nonzero components are
xz =
w
x
and
yz =
w
,
y
(7.2.13)
diverges as the inverse square root of the distance from the tip. This physically unacceptable consequence renders the Navier model invalid suciently
close to the tip, but there are several ways of recovering from this setback,
including the following.
(i) Rounding the tip
This is the approach adopted above when considering the elliptical crack
(7.2.6), and it is easily shown, using this example, that the stress at the
crack tip is given approximately by
)
c
(7.2.14)
yz 2III
r0
when r0 , the tip radius of curvature, is small (Exercise 7.1).
(ii) Plastic zone
Even if the tip is rounded, (7.2.14) shows that the stress may still become
A similarly fundamental square root singularity arises in the mathematically analogous theory
of aerofoils; (see, for example, Batchelor, 2000, p. 467).
294
w
= f (x)
y
on y = 0, x < c,
(7.2.15)
KIII
(7.2.17)
lim yz (x, 0) x c = .
xc
2
295
The stress intensity factor KIII is the parameter that characterises the
propensity of the crack to propagate. For the crack modelled by (7.2.2),
for example, we nd that KIII = cIII . We postulate that the crack will
propagate if KIII exceeds a critical value, which must be determined either from (i)(iii) above or from experiment, and which we expect to be
proportional to the surface energy . Under this postulate, (7.2.16) immediately reveals that a Mode III crack is more likely to grow if it is
longer or subject to greater stress. We emphasise that a consequence of
this postulate is that KIII is the only piece of information concerning the
global stress eld that is relevant for deciding whether or not the crack
propagates.
We will shortly nd that the concept of a stress intensity factor applies
to many other crack congurations, but rst let us review the mathematical
challenges posed by problems like (7.2.2).
296
and then the boundary condition (7.2.2c) would have led to a singular
integral equation for the function f ().
(iii) Since w is an odd function of y, the problem may be formulated in the
halfplane y 0 with the boundary conditions
w(x, 0) = 0,
x > c,
w
(x, 0) = 0,
y
x < c.
(7.2.20)
(a)
(b)
y
II
297
y
II
II
II
Fig. 7.6 (a) A planar Mode II crack. (b) The regularised problem of a thin elliptical
crack.
on .
(7.2.21)
The imposed uniform shear stress far from the crack implies that A has the
asymptotic behaviour
A II xy
as
x2 + y 2 .
(7.2.22)
We recall from Section 2.6 that the general solution of the twodimensional
biharmonic equation may be written in the form
A = Re zf (z) + g(z) ,
(7.2.23)
where f and g are any two analytic functions of z = x + iy. To transform
the boundary conditions for A into conditions on f and g, it is helpful to
write (7.2.23) in the form
A=
1
zf (z) + g(z) + z f(
z ) + g(
z) ,
2
(7.2.24)
298
on .
(7.2.26)
g(z)
b
z = () = a + ,
(7.2.28)
where a and b are real constants with b > a > 0. The principal radii of the
ellipse are (a + b) and (a b), so that
ce
ce
,
b=
,
(7.2.29)
2
2
in the terminology of Section 7.2.2. We also introduce the notation
F () = f () ,
G() = g ()
(7.2.30)
a=
for the transformed versions of f and g. Thus F is analytic in the unit disc
 1, and G is analytic apart from a double pole at the origin, where
G()
iII b2
2 2
as
0.
(7.2.31)
(1/)F () G ()
+
()
()
at  = 1,
(7.2.32)
a 2 b
,
2
(7.2.33)
where
(1/) =
a + b 2
,
() =
299
and we use the fact that = 1/ on the unit circle. Provided b > a > 0, the
map z = () is conformal and hence () is nonzero in the unit disc. The
only singularity in the righthand side of (7.2.32) therefore arises from the
pole in G() at the origin, where we nd that
G ()
iII b
()
as
0.
(7.2.34)
Let us subtract this from both sides of (7.2.32) to remove the singularity
from the righthand side and thus obtain
2 G () + a + b 2 F ()
i
iII b
b
II
F (1/)
=
at  = 1.
a 2 b
(7.2.35)
Now we arrive at the crux of the argument. The lefthand side of (7.2.35)
is analytic in  1, while the righthand side is analytic in  1. We can
therefore use (7.2.35) to analytically continue the lefthand side into  < 1
and hence obtain a function that is analytic on the entire complex plane
and, because of (7.2.27), tends to zero as (see, for example, Carrier,
Krook & Pearson, 1983, Section 2.8). According to Liouvilles theorem, the
only function that has these properties is the zero function, and we deduce
that the left and the righthand sides of (7.2.35) must both be identically
zero. We can therefore evaluate the function F () as
F () = iII b
(7.2.36)
By integrating with respect to and ignoring an irrelevant constant of integration, we thus obtain
iII b2 1
2
(7.2.38)
.
G() =
2
2
Finally, by using the inverse mapping
z z 2 4ab
,
=
2a
(7.2.39)
300
where the square root is again dened by (7.2.12), we evaluate the functions
f and g in terms of the original variable z as
iII b
z z 2 4ab ,
(7.2.40a)
f (z) =
2a
iII
(7.2.40b)
g(z) = 2 (a2 b2 )(z 2 2ab) + (a2 + b2 )z z 4ab .
4a
When substituting these into the formula (7.2.23) for A, we simplify the
resulting expression by letting the crack thickness tend to zero, so that a
and b both tend to c/2, and thus obtain
II
(7.2.41)
Im (
z z) z 2 c2 .
A=
2
By using the chain rules
=
+
,
=i
,
(7.2.42)
x
z z
y
z z
as shown in Exercise 7.4, we can recover the stress components as
z f (z) g (z) + 2f (z)
xx = Re
4z 3 + c2 (
II
z 5z)
Im
,
(7.2.43a)
=
2
(z 2 c2 )3/2
xy = Im zf (z) + g (z)
2z 3 + c2 (
z 3z)
II
Re
,
(7.2.43b)
=
2
(z 2 c2 )3/2
yy = Re zf (z) + g (z) + 2f (z)
II c2
z z
.
(7.2.43c)
=
Im
2
(z 2 c2 )3/2
By approximating these in the neighbourhood of the crack tip z = c, we nd
that
1
51
II c
xx
+ 7 sin
,
(7.2.44a)
sin
2
2
4 2r1
1
51
II c
+ 3 cos
,
(7.2.44b)
cos
xy
2
2
4 2r1
II c
31
yy
,
(7.2.44c)
sin 1 cos
2
2 2r1
as r1 0, where r1 and 1 are dened as in Figure 7.4. Hence, as in Mode III
cracks, the stress diverges like the inverse square root of the distance from
301
S/II
1
1
0.8
0.6
y0
0.4
0.2
1
0
2
2
1
x
Fig. 7.7 Contour plot of the maximum shear stress S around a Mode II crack, with
crack halflength c = 1.
the crack tip. Ahead of the crack, the only nonzero stress component on the
plane y = 0 is found, by setting 1 = 0, to be
II c
as x c.
(7.2.45)
xy (x, 0)
2(x c)
We therefore dene the stress intensity factor for Mode II cracks by
(7.2.46)
KII = lim 2 (x c)xy (x, 0) = II c.
xc
As in Mode III cracks, we postulate that the crack tip will propagate if KII
exceeds some critical value.
We visualise the stress eld in Figure 7.7 by plotting the contours of the
maximum shear stress
*
2 A 2 1 2 A 2 A 2
1 2 
=
+
= zf (z) + g (z) .
S=
2
2
2
xy
4 x
y
(7.2.47)
As expected, the stress is concentrated near the crack tips and minimised
along the crack faces.
This conformal mapping approach can in principle be applied to more
complicated crack geometries. However, for the special case of a planar crack,
an alternative and somewhat simpler way to proceed is to pose the problem in the halfspace y > 0, and then solve it using the PapkovichNeuber
potentials dened in (2.8.51). The boundary conditions on the Airy stress
302
as
y ,
(7.2.48)
on y = 0, x < c
(7.2.49)
on y = 0, x > c,
(7.2.50)
and
A=0
but the second boundary condition on y = 0, x > c is less obvious. We can
either argue on physical grounds that xx = 0 or use symmetry arguments
to say that
2A
=0
y 2
on y = 0, x > c.
(7.2.51)
Once we have solved this problem for A, the displacement is still determined only up to an arbitrary rigidbody motion. To specify the solution
uniquely, we focus on the case of a pure shear parallel to the crack face, so
that
u II y,
v0
as
y .
(7.2.52)
2u = 2II y
+ y ,
(7.2.53)
x
2
1
+ y ,
(7.2.54)
2v = 2 (1 )
y
2
while A satises
1
A
= II x (1 ) +
+
(y) .
y
y
2 y
(7.2.55)
(1 ) .
(7.2.57)
y
303
y
0.2
0.1
2 (1 ) II / = 0
1
0.5
0.5
0.1
0.25
0.2
0.5
Fig. 7.8 The displacement of a Mode II crack of halflength c = 1 under increasing
shear stress II .
We can then integrate (7.2.55) to obtain the Airy stress function in the form
y
.
(7.2.58)
2
The remaining boundary conditions (7.2.49) and (7.2.50) give that
A = II xy +
(x, 0) = 2II ,
x
x < c
(7.2.59)
(x, 0) = 0,
y
x > c,
(7.2.60)
and
304
I
y
x
As depicted in Figure 7.8, the crack is deformed into an ellipse that grows
and rotates as the applied shear stress is increased.
I 2
x
2
as
x2 + y 2 .
(7.2.64)
The details of the calculation, following the steps used in Section 7.2.4, may
be found in Exercise 7.5. This time the Goursat functions f and g take the
form
I
(7.2.65a)
z + 2 z 2 c2 ,
f (z) =
4
c2 I
I
g(z) =
log z + z 2 c2 + z 2 ,
(7.2.65b)
2
4
2z 3 + c2 (
z 3z)
xx = I + I Re
2 (z 2 c2 )3/2
I c
3 cos(1 /2) + cos(51 /2) ,
4 2r1
I c2
z z
Im
xy =
4
(z 2 c2 )3/2
I c
cos(31 /2) sin 1 ,
2 2r1
2z 3 c2 (z + z)
I
Re
yy =
2
(z 2 c2 )3/2
I c
cos3 (1 /2) (3 2 cos 1 ) ,
2r1
305
(7.2.66a)
(7.2.66b)
(7.2.66c)
I c
as r1 0,
(7.2.67)
xx , yy
2r1
so we dene the Mode I stress intensity factor as
(7.2.68)
As shown in Figure 7.10, the stress is again concentrated near the crack and
minimised along the crack faces.
As in Section 7.2.4, we can instead adopt the PapkovichNeuber representation, again subtracting the linear fareld displacement, so that
1
+ y ,
(7.2.69a)
2u = I x
x
2
1
2v = (1 )I y + 2(1 )
+ y ,
(7.2.69b)
y
2
and the Airy stress function satises
A
= (1 ) +
y
y
1
+ y .
2
(7.2.70)
We still have the stressfree boundary conditions (7.2.49) on the crack faces,
while symmetry implies that
A
=v=0
y
on y = 0, x > c.
(7.2.71)
306
S/I
1
1
0.8
0.6
y0
0.4
0.2
1
0
2
2
1
x
Fig. 7.10 Contour plot of the maximum shear stress S around a Mode I crack, with
crack halflength c = 1.
Notice that A/y is equal to zero on the entire xaxis. Thus (7.2.70)
leads to the boundary condition
1
=
(7.2.72)
y
2
on y = 0, and, following the same argument as in Section 7.2.4, we deduce
that (7.2.72) holds everywhere. We can then use this identity to integrate
(7.2.70) with respect to y and thus obtain
1
I x2
A = y
+
.
2
1 2
2
(7.2.73)
= 2I
y
=0
on y = 0, x < c,
(7.2.74a)
on y = 0, x > c,
(7.2.74b)
and, again by analogy with the Mode III solution (7.2.11), we deduce that
z 2 c2 .
= 2I y + 2I Im
(7.2.75)
To complete the solution, we let = Re{h(z)} for some analytic function h
and, using (7.2.72) and (7.2.75), we nd that
(7.2.76)
h (z) = (1 2)I z z 2 c2 .
307
y
0.2
0.1
2 (1 ) I / = 0
1
0.5
0.5
0.25
0.1
0.5
0.2
1
I z 2 z z 2 c2 + c2 log z + z 2 c2 . (7.2.77)
2
(7.2.78a)
(7.2.78b)
Figure 7.11 shows that the crack becomes elliptical, growing wider and
shorter as the applied normal stress increases.
We conclude this discussion with two remarks. First, we note that a general planar crack in two space dimensions will be a linear superposition of
Modes I, II and III, and hence the stress near its tip will be characterised
by three intensity factors. The calculations carried out above suggest that
all three will scale with the square root of the crack length, but unfortunately there are no general rules about the factors to which a crack will be
most vulnerable. Second, the most famous threedimensional crack conguration is the pennyshaped crack which is the radially symmetric version of
our earlier Mode I crack and it can be solved explicitly, with a great deal
of eort, by transforming to spheroidal coordinates and taking a limit as in
Section 7.2.2. Both these congurations, along with many others, are lucidly
described in Sneddon & Lowengrub (1969).
308
w
=0
y
on y = 0, > 0,
(7.2.79b)
where = x + V t and cs is the shear wave speed. Notice that this problem
generalises the static semiinnite Mode III crack problem considered in
(7.2.3).
To close the model (7.2.79), we should impose some conditions at innity,
but this would require us to solve for the stress away from the crack. Even
without precise knowledge of this far eld, we expect the physically relevant
solution to have a square root singularity analogous to (7.2.4) with n = 1.
The corresponding solution of (7.2.79) is
)
V2
+ iy ,
where B 2 = 1 2 ,
(7.2.80)
w = A Re
B
cs
and it is only through the constant A that the global stress eld is felt. Now
the dynamic stress intensity factor is
)
)
1/4
w
V2
B
KIII = 2 lim
(, 0) = A
= A
1 2
.
0
y
2
2
cs
(7.2.81)
7.3 Contact
309
7.3 Contact
7.3.1 Contact of elastic strings
We start by considering the simplest elastic contact problem, namely an elastic string making steady contact under a prescribed body force p(x) against
a smooth, nearly at, rigid obstacle . We use the terminology contact set
to denote the set of points at which the string makes contact with the obstacle, and noncontact set for those points where it does not. If the transverse
displacement w(x) is assumed to be small, then a force balance on a small
element of the string in the noncontact set yields the familiar equation
T
d2 w
= p(x),
dx2
(7.3.1)
where the string tension T is spatially uniform. Provided friction is negligible, T is also constant throughout the contact set, where w is simply equal
to the prescribed obstacle height f (x) above the xaxis, which is zero in
Figure 7.12.
Now we must consider how to join the solutions in the contact and noncontact sets, which leads us to the question of the smoothness of the solution
at points where contact is lost, a question we also had to address in our
discussion of crack tips. Here this problem is resolved immediately by a local
force balance, which reveals that T and T dw/dx must both be continuous.
Thus, our task is to solve (7.3.1) in the noncontact set, subject to specied
boundary conditions at the ends of the string and
w = f,
df
dw
=
dx
dx
(7.3.2)
310
w
1.0
increasing p
0.8
0.6
0.4
0.2
1.0
0.5
0.5
1.0
Fig. 7.12 Solution for the contact between a string and a level surface, with applied
pressure p/T = 0, 1, 2, 3, 4, 5.
(7.3.4)
p
(1 x)(1 + x 2s)
2T
(7.3.5)
w(s) =
to obtain
w =1
in x > s, where
*
s=1
2T
.
p
(7.3.6)
7.3 Contact
(a)
1.0
(b)
0.5
311
(c)
0.30
0.30
0.30
0.25
0.25
0.25
0.20
0.20
0.20
0.15
0.15
0.15
0.10
0.10
0.10
0.05
0.05
0.5
1.0 1.0
0.5
0.05
0.5
1.0 1.0
0.5
0.5
1.0
Fig. 7.13 Three candidate solutions for the contact problem between a stressfree
string and the surface z = 2x2 3x4 .
(7.3.7)
with its ends xed at w(1) = w(1) = 0. With p = 0, (7.3.1) implies that
the string must be straight wherever it is out of contact. The most obvious
linear function satisfying both boundary conditions is simply w 0 but, as
can be seen in Figure 7.13(a), this would imply that the string penetrates
the obstacle, which is impossible. We must therefore allow the string to make
contact with the obstacle, as illustrated in Figure 7.13(b). Recalling that the
string must meet the obstacle tangentially, the solution in this case is easily
constructed as
2
8
(1 + x), 1 x ,
9
3
2
2
(7.3.8)
w = 2x2 3x4 , x ,
3
3
8 (1 x),
x 1.
9
3
We can, however, construct a third solution, shown in Figure 7.13(c), in
which the string loses contact with the middle of the obstacle. Again noting
that the string is linear when out of contact and must always make contact
312
2
8
(1 + x), 1 x ,
9
3
1
2
2x2 3x4 , x ,
1
1
1
w=
,
x ,
3
3
3
2
1
2x2 3x4 , x ,
2
8
(1 x),
x 1.
9
3
(7.3.9)
d2 w
= p R.
dx2
(7.3.10)
The reaction force R must be zero outside the contact set. The solution
shown in Figure 7.13(b) has negative values of R and is therefore unphysical;
the correct solution is therefore (7.3.9).
To summarise, we can select a physically relevant solution of the contact
problem by insisting that
w = f,
d2 w
p<0
dx2
(7.3.11a)
w > f,
d2 w
p=0
dx2
(7.3.11b)
and
with continuity of w and dw/dx at the boundaries between the contact and
noncontact sets. The two conditions (7.3.11) may usefully be combined to
2
d2 w
d w
with w f, T 2 p 0, (7.3.12)
(w f ) T 2 p = 0,
dx
dx
and this socalled linear complementarity problem can be written in another
way which is very convenient computationally.
7.3 Contact
If w satises
1
1
dw
dw dv
dx
p(w v) dx
T
dx
dx
1 dx
1
313
for all v f,
(7.3.13)
with v and w satisfying the same xed boundary conditions, then, as shown
in Elliott & Ockendon (1982, p. 103), w satises (7.3.12). The statement
(7.3.13) is called a variational inequality, and, as shown in Exercise 7.8, w
satises (7.3.13) if and only if it is the minimiser of the functional
1 2
T dv
+ pv dx
2 dx
1
over all v f . This has the obvious physical interpretation of minimising
the elastic energy minus the work done by pressure over the virtual displacements v which prevent interpenetration of the string and the obstacle.
Moreover, this constrained optimisation problem is very much easier to solve
numerically than the free boundary problem (7.3.1), (7.3.2).
d4 w
d2 w
EI
p = 0,
dx2
dx4
(7.3.14)
(7.3.15)
314
(a)
(b)
1.0
(c)
w
1.0
1.0
1.0
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
0.5
1.0
0.5
1.0
0.5
0.5
1.0
1.0
0.5
0.5
1.0
Fig. 7.14 The contact between a beam and a horizontal surface under a uniform
pressure p. (a) The beam sags towards the surface: p/EI = 0, 12/5, 24/5. (b) The
beam makes contact at just one point: p/EI = 24/5, 24. (c) The contact set grows:
p/EI = 24, 2048/27, 384.
d2 w
(1) = 0,
dx2
d2 w
(1) = 0,
dx2
(7.3.16a)
(7.3.16b)
(7.3.17)
As shown in Figure 7.14(a), the beam sags under increasing applied pressure
until it rst makes contact with the surface z = 0 when p/EI = 24/5.
The solution (7.3.17) cannot remain valid for p/EI > 24/5, because it
predicts negative values of w. So, let us instead seek a solution in which
the beam makes contact with z = 0 over a region s x s. We need
only solve the problem in x > 0, since we can then infer the solution for
negative values of x by symmetry. We therefore replace (7.3.16a) with the
free boundary conditions
w(s) = 0,
dw
(s) = 0,
dx
d2 w
(s) = 0.
dx2
(7.3.16c)
Notice again that, compared with (7.3.16a), we have introduced one more
boundary condition to compensate for not knowing the location of the free
boundary a priori. Again, the problem is easy to solve and we nd that
w=
(x s)3 (2 s x)
(1 s)4
(7.3.18a)
7.3 Contact
315
pressure by
24
p
=
EI
(1 s)4
or
s=1
24EI
p
1/4
.
(7.3.18b)
In Figure 7.14(c) we show how the contact set grows as p increases, with
s = 0, 1/4, 1/2 when p/EI = 24, 2048/27, 384 respectively. However, since s
must lie between 0 and 1, the solution (7.3.18) is only valid when p/EI 24.
This begs the question: what happens for values of p/EI between 24/5
and 24? The answer is that the beam makes contact at just one point,
namely the origin. If we replace (7.3.16c) with
w(0) = 0,
dw
(0) = 0,
dx
(7.3.16d)
then we nd
w=
x2 (3 x)
p
x2 (1 x)(3 2x),
2
48EI
(7.3.19)
in x 0, with the even extension of (7.3.19) in x 0. As shown in Figure 7.14(b), the beam initially meets z = 0 at one point, then attens as
p increases, until its curvature matches the curvature of the obstacle. Only
then does the contact set start to grow. It is easy to verify that (7.3.19)
matches (7.3.17) and (7.3.18) when p/EI = 24/5 and 24 respectively.
The nal surprise comes when we consider the reaction force R exerted
by the surface z = 0 on the beam, given by
R = p + EI
d4 w
.
dx4
(7.3.20)
Consider rst the solution (7.3.19) in which the contact set is the point
x = 0. By dierentiating (7.3.19) and using the symmetry about x = 0, we
nd that
d3 w
5p
d3 w
(0+)
=
(0) =
3.
(7.3.21)
3
3
dx
dx
8EI
Hence the third derivative of w is discontinuous across x = 0 and (7.3.20)
then implies that there is a deltafunction in the reaction force:
5p
R=
6EI (x).
(7.3.22)
4
Similarly, dierentiation of (7.3.18) reveals that
1/4
12
3
p 3/4
d3 w
(s+)
=
=
,
dx3
(1 s)3
2
EI
(7.3.23)
316
2
1
T dw 2 EI d2 w
+
+ pw dx.
(7.3.25)
min
wf 1
2 dx
2
dx2
It is easy to show that (7.3.14) is the EulerLagrange equation associated
with (7.3.25) in the absence of any contact.
To generalise the above theories into three dimensions, we now consider
an elastic membrane with transverse displacement w(x, y) making contact
with a smooth obstacle z = f (x, y) under an applied body force p(x, y).
Clearly the twodimensional analogue of (7.3.1) is Poissons equation
T 2 w = p(x, y),
(7.3.26)
7.3 Contact
317
f (x)
punch
(7.3.29)
xy = 0,
yy < 0
on
y = 0, x < c.
(7.3.30a)
Outside the contact set, the surface traction is zero and there must be no
interpenetration, that is
xy = yy = 0,
v > f (x)
on y = 0, x > c.
(7.3.30b)
Our task, then, is to solve plane strain subject to the mixed boundary
conditions (7.3.30) on y = 0 and given behaviour, typically zero stress, at
318
innity. Such problems are almost always so complicated that they must be
solved computationally but, enlightened by (7.3.14), we can see that this is
not as fearsome a task as might be supposed. One must devise a code that
minimises the integral
W dxdy,
y>0
where W is the strain energy density dened in Section 1.9, over all displacement elds satisfying the constraint v(x, 0) f (x) . The proof of
the equivalence of this proposed algorithm to nding a displacement eld
satisfying the Navier equation and the boundary conditions (7.3.30) relies
crucially on the fact that the boundary conditions together with the inequalities in (7.3.30) determine a unique solution with bounded stress everywhere.
One cunning way to construct analytic solutions of the contact problem
described above is to suppose that we know in advance both the size c of
the contact set and the contact pressure, say p(x), exerted by the punch.
We thus replace the boundary conditions (7.3.30) with
xy = 0,
yy = P (x)
on y = 0,
(7.3.31)
where
P (x) =
p(x),
x < c,
0,
x > c.
(7.3.32)
We can now use the solution obtained in (2.6.85) to calculate the displacement v0 (x) of the boundary y = 0, given by
1
v0 (x) =
(7.3.33)
In x < c, this allows us to recover the shape of the punch that corresponds
to the assumed pressure prole p(x). It only remains to check for consistency
that there is no interpenetration in x > c.
As an example (guided by hindsight), let us consider the pressure prole
p(x) = F
c2 x2 ,
(7.3.34)
7.3 Contact
319
v0
1
10
5
10
1
2
3
4
Fig. 7.16 The penetration of a quadratic punch into an elastic halfspace y > 0.
2
(1 )F
c (
c2 '
1 2 log
x2
2
2
0,
x < c,
+
x x2 c2 c2 cosh1 (x/c), x > c.
v0 (x) =
(7.3.35)
(1 )c2 F
log x
2
as
x .
(7.3.36)
This reects the weakness in the theory of plane strain noted in Section 2.6.8,
and the same diculty does not occur for more realistic threedimensional
congurations (see Barber, 1993).
We can observe in Figure 7.16 that both v0 and its rst derivative are
continuous across x = c. Indeed, it is readily veried from (7.3.35) that
the thickness of the void between the punch and the elastic body varies like
c (
c2 '
4 2c
2v0 (x)
2
(x c)3/2
(7.3.37)
1 2 log
+x
(1 )F
2
2
3
as x c. In contrast, we recall that, in the theory of brittle fracture without
cohesion, the crack thickness typically varies as the square root of the distance from the crack tip. Recall also that the stress grows like (x c)1/2
as a
crack tip is approached, while the above example, by construction, has
p 2c c x as x c.
320
These observations prompt the question: why should the stress singularity in the smooth contact problem be weaker than that in the fracture
problem? The answer lies in the fact that the crack geometry is prescribed
a priori; it is perfectly sensible to seek the stress distributions around a
crack with arbitrary faces and an arbitrary crack tip surrounding these faces.
However the boundary of a smooth contact set is free to decide its location,
and it does so in such a way as to prevent the stress intensication that is
characteristic of fracture. Notice also that the introduction of cohesion into
the tip region of a brittle crack, as shown in Exercise 7.3, can also reduce
the strength of the singularity to a 3/2power as in (7.3.37), when we
regard the cohesion length as a free parameter.
Exercises
321
Exercises
7.1
For the elliptical Mode III crack solution (7.2.9), show that the stress
components on y = 0, x > c cosh are given by
x cosh
yz = 20 e
sinh .
xz = 0,
x2 c2
Deduce that the stress at the crack tip is
yz = 20 e cosech
20
as
0.
when r0 is small.
1 x
d
,
f () tan
y c
y
c x
322
7.3
"
c x
is f () = / c2 2 . Are there any others?
Consider a semiinnite planar Mode III crack along the positive
xaxis with a cohesive stress C in a neighbourhood of the crack
tip. The antiplane displacement w(x, y) satises Laplaces equation
outside the crack, subject to the boundary conditions
C, 0 < x < ,
w
(x, 0) =
w(x, 0) = 0, x < 0,
y
0, x > ,
and the fareld condition
)
2 K
Re z
w
as
x2 + y 2 ,
.
y
2 ( 2 + 2 )
2C,  < ,
w
(, 0) =
0,
 > ,
and
)
w
2 K
as
w
2C
=
Im log
,
+
where = + i, and deduce that
)
2C
+
2 K
w
2 +
=
Re log
.
Exercises
323
+ C 1 tan1
(x, 0) =
x
2
in x < 0, and deduce that the stress at the crack tip is nite if
=
7.4
K 2
.
8C 2
By writing the Airy stress function in the form (7.2.24) and using
the chain rules (7.2.42), show that the stress components in plane
strain are given in terms of the Goursat functions by
2A
=
Re
z
f
(z)
+
2f
(z)
+
g
(z)
,
x2
2A
=
Re
z
f
(z)
+
2f
(z)
g
(z)
,
y 2
2A
= Im zf (z) + g (z) .
xy
7.5
+
.
=
a 2 b
4
4
In the limit a, b c/2, deduce that F and G are given up to arbitrary
integration constants by
cI 1
3 ,
F () =
8
1
c2 I
c
c2 I
+
log
.
G() =
16
7.6
324
A
1
+
,
2( + ) x
y
v=
1
A
.
2( + ) y
x
x2
y 2
1 xy
2A
2 2
2
=
.
xy
1 x2
y 2
Deduce that
2
z2
A
i
1
= 0,
i = zf (z) + g(z),
1
1
z ) g (
z) .
(3 4)f (z) z f (
2
7.7
2 (1 )
f (z).
Finally, use the expression (7.2.65a) for f (z) to reproduce the displacement (7.2.78) of a Mode I crack.
Suppose a semiinnite Mode III crack propagates at constant speed
V along the xaxis in a channel y < h, < x < , so that
the displacement w satises (7.2.79). Suppose also that antiplane
shear displacements are applied on the walls of the channel such
that w = b on y = h, respectively. Deduce that
2w
w
= 0,
+
X 2
y
Exercises
325
1/2
where x + V t = = 1 V 2 /c2s
X, and verify that the solution
is
by
b
w=
Im log 1 + 1 ez/h
,
h
(x, 0)
y
2 h X
as X 0, and deduce that the propagation speed is given by
*
K 4 h2
,
V = cs 1 III
4 b4
7.8
U [w] U [v] =
T
dw
dx
p (w v) dx
dv
dw
dx
dx
T
dx
2
dw dv
dx
dx
2
dx
p(w v) dx.
T
0
dx dx
dx
1
7.9
326
w
1
0.8
Increasing
M/EI
0.6
0.4
0.2
0.2
0.4
0.6
0.8
Fig. 7.17 A exible ruler attened against a table with applied bending moment
M/EI = 0, 3, 6, 9, 12, 15.
w
w
0.01
0.035
0.03
0.008
0.025
0.006
0.02
0.004
0.015
0.01
0.002
0.005
1.5
1
0.5
0.5
1.5
1.5
1
0.5
0.5
1.5
2
Fig. 7.18 (a) A wave travelling along a rope on the ground. (b) Some higher travelling modes.
7.10
7.11
Exercises
327
2
d4 w
2d w
+
V
+ g = 0,
d 4
d 2
g4
2EIk 4
d2 w
dw
=
=0
d
d 2
at
= .
k2 2
2 + k 2 2 2 sec k cos
k
for the shape of the hump in the rope, and show that the speed is
given by
*
k EI
,
V =
where k satises the transcendental equation
tan k = k.
[This equation has a countably innite set of solutions; the displacement corresponding to the rst is shown in Figure 7.18(a) and the
next three are shown in Figure 7.18(b).]
Given that the lowest positive root is given by k1 9/2, deduce
that icking one end of the rope up and down over a time will
produce a wave travelling at speed
EI 1/4
.
V 3
2
8
Plasticity
8.1 Introduction
It is unfortunate that what are colloquially called plastics are polymeric
solids that usually behave elastically at room temperature and viscoelastically when hot; hence the word plasticity should not be thought of as
applying just to such materials. Indeed, some of the most important examples of plastic behaviour occur in metals. A convenient way to gain some
intuition about this phenomenon is to apply an increasing bending moment
to a paper clip. There is always a critical bending moment above which
the clip fails to revert to its initial state, which means that it has become
inelastic. This observation is encapsulated in Figure 8.1, which shows the
qualitative stress/strain curve for a metal that yields when some norm of the
stress tensor reaches the yield stress Y . If a stress lower than Y is applied,
the material responds elastically (although possibly nonlinearly), returning
to its original state when the loading is removed. However, when a stress
greater than Y is applied and then removed, a nonzero permanent strain
remains.
Macroscopic plasticity modelling thus poses two key challenges. The rst
is to explain the existence of a critical yield stress, above which a solid ceases
to behave elastically, and the second is to predict the irreversible behaviour
that occurs at a stress equal to or greater than this critical value. The part
of the loading curve in Figure 8.1(b) above the dotted line is where the
material is said to be undergoing work hardening.
A limiting case of the behaviour depicted in Figure 8.1, which has proved
extremely useful in practical models of plastic behaviour, is known as perfect plasticity. In perfectly plastic theories, the stress is never allowed to exceed the yield stress, and the material can thus exist in one of two distinct
states. Below the yield stress, it behaves as an elastic solid and is described
by the models presented previously in this book; at and only at the yield
328
8.1 Introduction
(a)
stress
Y
(b)
stress
loading
329
loading
Y
unloading
unloading
strain
strain
Fig. 8.1 A typical stressstrain relationship for a plastic material: (a) below the
yield stress Y ; (b) when the yield stress is exceeded.
330
Plasticity
(a)
Y
un load
loa
din ing
g
(b)
stress
strain
un load
loa
din ing
g
stress
strain
Fig. 8.2 The stressstrain relationship for a perfectly plastic material: (a) under a
single loading; (b) under periodic loading/unloading.
start with a brief discussion of plastic ow in granular solids, whose smallscale behaviour is more familiar from everyday experience.
331
y
R
F
11111111111111
00000000000000
00000000000000
11111111111111
00000000000000
11111111111111
00000000000000
11111111111111
W
00000000000000
11111111111111
00000000000000
11111111111111
Fig. 8.3 The forces acting on a particle at the surface of a granular material.
which can be sustained without the particles giving way and owing, for
example, via a series of avalanches. It is also characteristic of the perfectly
plastic behaviour described in Section 8.1.
When the material is piled maximally, it follows that the surface slope
must equal tan everywhere, and the pile height z = h(x, y) therefore satises the equation
2
h =
h
x
2
+
h
y
2
= tan2 .
(8.2.1)
This is called the eikonal equation for h and it arises in many other contexts, notably in geometrical optics (see Ockendon et al., 2003, p. 359).
Singlevalued solutions of (8.2.1) predict shapes that are reminiscent of our
everyday experience of piles of granular materials (for example sugar, sand,
etc.); see Exercise 8.1.
8.2.2 Granular ow
The above discussion gives us condence that Coulombs law provides a
good model for a maximally piled granular material, under static conditions
at least. To model a owing granular material, we must apply Coulombs
law to particles in the interior, and it is much easier to implement this
in two dimensions rather than three. Let us then consider the forces acting on a twodimensional surface element inside a granular medium whose
unit normal is n = (cos , sin )T , as depicted in Figure 8.4. Assuming that
the internal stress can be described using a stress tensor , we nd that the
332
Plasticity
y
F
1111111111
0000000000
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
Fig. 8.4 The normal force N and frictional force F acting on a surface element
inside a granular material.
normal traction exerted on the particles below the element by those above
is given by
N=
cos
xx xy
sin
xy yy
1
1
= (xx + yy ) + (xx yy ) cos(2) + xy sin(2). (8.2.2)
2
2
(cos , sin )
As pointed out in Section 8.2.1, we expect the particles to exert a compressive force on each other, but never a tensile one, and N must therefore
be nonpositive, for all choices of the angle . It follows that none of the
principal stress components (i.e. the eigenvalues of ) can be positive.
Similarly, the tangential (frictional) shear stress on our surface element is
given by
F =
( sin , cos )
xx xy
cos
xy yy
sin
1
= (yy xx ) sin(2) + xy cos(2). (8.2.3)
2
Now Coulombs law implies that F  must be bounded by N tan , for all .
In addition, ow can occur only if there is some value of for which F  is
equal to N tan . If so, this direction denes a slip surface along which we
expect ow to occur.
333
F
+
1
4 (xx
2
yy )2 + xy
xx + yy
2
Fig. 8.5 The Mohr circle in the (N, F )plane, and the lines where F  = N  tan .
From (8.2.2) and (8.2.3), we see that, as varies, the tractions lie on the
socalled Mohr circle in the (N, F ) plane, given by
2
(xx yy )2
1
2
2
+ xy
,
(8.2.4)
F + N (xx + yy ) =
2
4
where dierent points on the circle correspond to dierent choices of the
angle . Since we require N to be nonpositive for all , the Mohr circle must
lie in the halfplane N 0, as shown in Figure 8.5. Coulombs law then tells
us that F  N  tan for all , and the Mohr circle must therefore lie
in the sector N tan F N tan . Finally, if the material is owing,
then there must be one value of such that F  = N  tan , and the Mohr
circle must therefore be tangent to the lines F  = N  tan , as shown in
Figure 8.5.
Elementary trigonometry now tells us that the stress components in a
granular material must satisfy
2 1/2
(xx + yy ) cos .
(8.2.5)
2 xx yy xy
When the material is owing and equality holds, (8.2.5) is called the Coulomb
yield criterion. Notice that (8.2.5) is a relation between the two stress invariants Tr ( ) and det ( ). This is reassuring, since it implies that the condition
for the material to yield is independent of our choice of coordinate system.
We can only make signicant analytical progress if we assume that the
ow is slow enough for the inertia of the particles to be negligible in
334
Plasticity
comparison with the frictional forces between them and, perhaps, gravity.
This assumption is often valid in practice and allows us to neglect the acceleration term in Cauchys momentum equation, which thus reduces to
xy
xy
yy
xx
+
= 0,
+
+ g = 0.
(8.2.6)
x
y
x
y
Given our restriction to two dimensions, we see that (8.2.6) and the yield
criterion (8.2.5) give us three equations in the three stress components xx ,
xy and yy . It is therefore possible, in principle, to solve for the stress
tensor in a owing granular material in two dimensions without specifying
any particular constitutive relation. This is in stark contrast with all theories
of elasticity that we have encountered thus far.
The easiest case to analyse occurs when gravity is negligible so we can use
(8.2.6) to introduce an Airy stress function A. Then, when the material is
owing, equality in (8.2.5) implies a nonlinear partial dierential equation
for A, namely
2 2
A
cos2 2 2
2A 2A
A .
=
(8.2.7)
x2 y 2
xy
4
When = /2, this reduces to the MongeAmp`ere equation encountered
previously in Chapter 4. When (0, /2), (8.2.7) is, unexpectedly, a
hyperbolic partial dierential equation.
Even with the body force included, the system (8.2.5), (8.2.6) is likewise
hyperbolic, as shown in Exercise 8.3. This means that the stress eld in the
owing material is conned to certain regions of inuence bounded by the
relevant characteristics of (8.2.7). Outside these regions, the inequality in
(8.2.5) is strict, so the granular material does not ow but behaves like an
elastic solid and hence satises elliptic equations. By combining these two
regimes, we obtain a perfectly plastic theory, as described in Section 8.1, in
which the yield stress is never exceeded. The key to solving such models is to
locate the free boundary that separates the owing and nonowing regions.
The switch in behaviour from hyperbolic to elliptic makes these problems
very dicult in general, but there are a handful of symmetric situations
where we can nd an explicit solution.
8.2.3 Example: a tunnel in granular rock
One such conguration is the problem of a circular tunnel of radius a bored
through granular rock, subject to a uniform isotropic pressure p at innity
and a positive pressure P at r = a to simulate the tunnel bracing. To model
Had we included a depthdependent hydrostatic pressure at innity, as in Section 2.6.9, analytical progress would have been considerably more dicult.
335
(8.2.8)
The boundary conditions on the tunnel wall and in the far eld read
rr = P
on r = a,
rr p ,
as
(8.2.9a)
r .
(8.2.9b)
(8.2.10)
(8.2.11)
(8.2.12)
While the rock remains elastic, we can use the usual constitutive relations
(1.11.5):
rr = ( + 2)
ur
dur
+ ,
dr
r
ur
dur
+ ( + 2) .
dr
r
(8.2.13)
(P p )a2
p r
+
,
2( + )
2r
(8.2.14)
a2
,
r2
= p (p P )
a2
.
r2
(8.2.15)
Notice the analogy between (8.2.14) and the gun barrel solution (2.6.41),
with the outer radius b tending to innity and a superimposed isotropic
pressure.
At the tunnel wall r = a, we have
2p
1,
=
rr
P
(8.2.16)
336
Plasticity
k
20
15
10
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Fig. 8.6 The triaxial stress factor k dened by (8.2.18) versus angle of friction .
2p
.
1+k
(8.2.17)
1 + sin
.
1 sin
(8.2.18)
=
p
p
. (8.2.19)
rr = p + p
k + 1 r2
k + 1 r2
Meanwhile, in the plastic region r < s, the quasistatic balance (8.2.8) is
now supplemented by the yield condition (8.2.11). The resulting equation
1k
drr
+
rr = 0,
dr
r
(8.2.20)
337
338
Plasticity
b
,
2r
(8.3.2)
339
displace one side of the cut by a distance b relative to the other side, and
nally weld the two cut faces together again. Clearly there will be a region
of very large strain close to the axis, in accordance with (8.3.2). We will
thus have created a bar that is in a state of selfstress, so that it is in
equilibrium under the action of no external forces, yet there is a nonzero
stress distribution in the interior.
We can also interpret (8.3.1) in terms of incompatibility. We recall from
Section 2.7 that the assumption of a singlevalued displacement eld gives
rise to compatibility relations between the strain components. Since (8.3.1)
is not singlevalued, we would expect these relations to be violated. Indeed,
in Exercise 8.4 we show that w satises
2w
2w
= b(x)(y),
yx xy
(8.3.3)
w
=
,
x
y
= .
y
x
(8.3.4)
340
Plasticity
(a)
(b)
Fig. 8.8 The displacement eld in an edge dislocation: (a) pristine material; (b) after
the cutandweld operation.
(8.3.5)
and we can easily verify that (8.3.5) is consistent with (8.3.3). The singularity
in (8.3.1) at x = y = 0 is thus analogous to the singularity at the core of a line
vortex in inviscid uid dynamics, where a nite circulation is concentrated
on a line in an otherwise irrotational ow.
Similar behaviour occurs if we displace one cut face radially relative
to the other before welding them back together, as shown in Figure 8.8.
This conguration
can be described using a plain strain displacement eld
T
u = u(x, y), v(x, y), 0 , in which u is discontinuous across the positive
xaxis. We can nd such a displacement by using the PapkovichNeuber
representation used in Section 2.8.6, namely
1
x1 + y2
2u = 2(1 ) 2 grad +
,
(8.3.6)
2
0
where 1 , 2 and are all harmonic functions of x and y. Now the required
discontinuity in u may be obtained by choosing = 0, 1 proportional to
and 2 equal to a constant.
As shown in Exercise 8.5, this approach leads us to the displacement eld
b sin
b
cos
u
+
(8.3.7)
=
v
2 0
2(3 4) sin
which is depicted in Figure 8.8. The associated stress eld is given in
341
(b)
Fig. 8.9 An edge dislocation in a square crystal lattice: (a) pristine crystal; (b) insertion of an extra row of atoms.
rr r rz
r z
rz z zz
sin()
(1 ) cos()
0
2b
(1 ) cos() (1 ) sin()
, (8.3.8)
=
0
(3 4)r
0
0
sin()
so we again have a state of selfstress which is innite on the line r = 0.
Exercise 8.6 shows that (8.3.8) corresponds to an Airy stress function
A=
b
{(1 2)r cos + 2(1 )r log r sin } ,
(3 4)
4b
(3 4)
(x) (y),
(8.3.9)
(8.3.10)
which is eectively (2.9.48). Now we recall from Section 2.8.2 that the existence of A ensures that the plane strain Navier equations are satised, while
the biharmonic equation for A was obtained by assuming compatibility between the strain components. Hence, as in (8.3.3), we can interpret (8.3.10)
as describing a line of incompatibility along the zaxis.
A key insight into the physical mechanism for plastic ow in metals comes
when we imagine executing the cutandweld operation shown in Figure 8.8
at an atomic scale. For example, on the square lattice shown in Figure 8.9(a),
we can achieve this by inserting an extra column of atoms, as depicted in
Figure 8.9(b). This gure is eectively the same as Figure 8.8(b) after the
ledge below the positive xaxis has been removed. Far from the crystal
mist, it simply seems that the atoms below the positive xaxis have been
displaced one atom spacing to the right, as in the displacement eld (8.3.7).
We are thus led to conjecture that, with b equal to a single atomic spacing,
342
(a)
Plasticity
(b)
(c)
(d)
Fig. 8.10 A moving edge dislocation: (a) initial conguration; (b) and (c) the dislocation moves to the left as the atoms realign themselves, eventually (d) leading
to a net displacement of the upper block of atoms relative to the lower.
Of course we are not really justied in using the atomic scale b in a macroscopic model like
(8.3.1). Indeed the key challenge in the theoretical modelling of dislocations is to make the
connection between atomic and macroscopic scales.
343
This can be achieved without forcing large numbers of atoms to slide over
each other, and this explains the huge discrepancy noted above between
predictions of the yield stress based on interatomic forces and experimentally measured values. We also note for future reference that the congurational changes illustrated in Figure 8.10 take place with no appreciable
change in volume of the crystal.
All these observations led theoreticians in the 1930s and earlier to suggest that macroscopic metal plasticity could be explained by the motion
of dislocations, and it was a great day when, two decades later, electron
micrographs of metal crystals that had undergone plastic deformation were
able to conrm the theory. They revealed countless (up to 1012 cm2 ) black
curves in regions which would have been invisible had the crystal been perfect. Each of these curves represented a dislocation bounding a plane region
of slip in the crystal and, the more the metal had been deformed plastically,
the more black lines were observed.
Since any realistic sample contains so many dislocations, it is impractical
to track each one individually. Instead, any macroscopic theory must use
a density, which is in fact a tensor, that characterises the direction and
Burgers vector averaged over a large number of dislocations. Moreover, it
would need a rule for how dislocations move, analogous to that used to
obtain a model for collective motion of vortices in an inviscid uid or of
superconducting vortices in a typeII superconductor. Unfortunately, it has
proved much harder to carry this out in practice for dislocations, and a
denitive macroscopic model does not yet exist.
Rather than delving further into this fascinating open problem, we will
simply use microscopic thoughtexperiments like Figure 8.10 to suggest the
following guidelines that will be employed below to obtain a closed macroscopic model for metal plasticity.
(i) Dislocations may move under the action of an applied shear stress, but
are generally impervious to any isotropic pressure.
(ii) Collective motion of many dislocations can only occur when a suciently high shear stress exists at every point. The critical yield stress
is a constant for any given metal.
(iii) Plastic ow, which results from collective dislocation motion, is essentially incompressible.
These conjectures can be tested experimentally or by performing atomistic
computations on a discrete lattice.
furthermore, such a density tensor would have to vanish in regions in which the incompatibility
tensor (2.9.46) vanishes
344
Plasticity
We conclude by mentioning two further benets that accrue from dislocation theory. One concerns what happens when a metal is repeatedly
deformed so strongly that many new families of dislocations are created on
dierent slip planes. These impede further movement, since dislocations cannot easily pass through each other. This gives a theoretical explanation of
work hardening, whereby a metal becomes more resistant to yield if it has
already gone through several stress cycles.
There is one other interesting macroscopic interpretation of dislocations
which is purely of theoretical interest. This concerns the modelling of brittle
fracture when no plastic eects are taken into consideration. It is a simple
matter (see Exercise 7.2) to see that the displacement eld (7.2.11) near a
Mode III crack can be represented as a distribution of virtual screw dislocations along the crack. Similarly, the displacement elds near a Mode I
or Mode II crack can be thought of as being caused by a distribution of
edge dislocations, all these displacements having closely related Papkovich
Neuber representations. This situation is reminiscent of the theory of ight,
in which a thin aerofoil can be regarded as a distribution of vortices in an
otherwise irrotational uid. Of course, the stress intensication near the tips
of a crack in the theory of brittle fracture implies the existence of plastic
regions near these tips, and these will contain real, not virtual, dislocations.
0
cos
0
0 xz
(8.4.1)
n =
0
0 yz
sin = (xz cos + yz sin ) 0
xz yz 0
0
1
345
on a surface element normal to n = (cos , sin , 0)T . We require the amplitude of the shear stress to be bounded by a critical yield stress Y for all
such surface elements, and this leads to the inequality
+
2 + 2 ,
xz
(8.4.2)
Y
yz
with equality when the material is owing.
To illustrate the mathematical structure in this relatively simple case, let
us return to the elastic torsion bar problem from Section 2.4. We recall that
the displacement eld is given by
u = (yz, xz, (x, y))T ,
(8.4.3)
where represents the twist of the bar about its axis. We also recall the
elastic stress function dened such that
xz =
,
y
yz =
,
x
(8.4.4)
(8.4.5)
 Y ,
(8.4.7)
r Y .
(8.4.8)
The lefthand side is maximised when r = a, and we deduce that the bar
will rst yield at its surface when the twist reaches a critical value
Y
.
(8.4.9)
c =
a
When > c , the condition (8.4.8) is violated, so our solution (8.4.6)
is no longer valid. Instead, as in the tunnel problem of Section 8.2.3, there
will be a plastic region near the boundary of the bar where the metal has
yielded, although we expect the material at the centre still to be elastic. We
therefore again have to introduce a free boundary, say r = s, that separates
the yielded and unyielded material, where s is to be determined as part of
346
Plasticity
the solution, and repeat the key assumption that, even when the material
has yielded, it ows slowly enough for the inertia terms to be neglected. We
can thus employ a stress function throughout the bar, satisfying (8.4.5) in
0 r < s and the yield condition
 = Y
(8.4.10)
s2 + r2
, 0 r < s,
as
(8.4.11)
=
2
s(a r),
s < r < a,
where
s=
Y
.
(8.4.12)
We thus see how the plastic region grows as increases past its critical value
c , and the nonzero stress components are easily found to be
y,
0 r < s,
x,
0 r < s,
xz =
yz =
(8.4.13)
ys/r, s < r < a,
xs/r, s < r < a.
We note that as Y tends to zero, a stress singularity develops at r = 0, which
reects the fact that (8.4.10) is a hyperbolic equation for ; the introduction
of even a small amount of elasticity removes the singularity.
As in Section 2.4, we can use (8.4.13) to calculate the torque M applied
to the bar as a function of the twist
(xyz yxz ) dxdy
M=
crosssection
3
a Y
,
c ,
2
c
(8.4.14)
= 3
a Y
3c
4 3 , > c .
6
Figure 8.11 shows how the torque increases linearly with the twist until
reaches its critical value c . Thereafter, it tails o rapidly, and we observe
that only a nite torque 2a3 Y /3 is required for the bar to fail completely.
Now suppose that, once a maximum twist M has been applied, so the
bar has yielded down to some radius
r = sM =
c a
Y
=
,
M
M
(8.4.15)
347
2M/a3 Y
1.4
1.2
1.0
0.8
0.6
0.4
0.2
/c
the applied torque is then removed suciently slowly that we can still use
a quasistatic model. We would expect the bar to recover and twist back
towards its starting conguration. We denote this subsequent displacement
; in other words, u
is the displacement relative to the state just before
by u
we released the torque. Consistent with our assumption of perfect plasticity
is the further assumption that, as soon as the torque is decreased, all the
onceyielded material instantly returns to being elastic. However, it will now
is zero.
start with the nonzero stress (8.4.13) when u
has the same structure (8.4.3) as the original disLet us suppose that u
placement, that is
(yz, xz, 0)T ,
=
(8.4.16)
u
since we recall from Section 8.2 that 0 for a circular bar. The net twist
so we expect
to start at zero when = M
of the bar is given by M + ,
and then fall to negative values as the bar is unloaded. The total stress
consists of the elastic stress corresponding to (8.4.16) added to the initial
stress (8.4.13) reached at the end of the plastic phase, that is
+ M ,
0 r < sM ,
xz
y
=
(8.4.17)
yz
x
+ M sM /r , sM < r < a.
A calculation analogous to (8.4.14) leads to the following formula for the
applied torque during the recovery phase:
3c
c
2M
(8.4.18)
4 3 .
=+
a4
3
M
The nal resultant twist 0 when the torque has been completely released
348
Plasticity
2M/a3 Y
plastic
M /c
1.2
1.0
0.8
0.6
0.4
0.2
0.0
elastic
0.5
1.0
recovery
1.5
2.0
2.5
3.0
/c
0 /c
Fig. 8.12 The normalised torque M versus twist applied to an elasticplastic
cylindrical bar, showing the recovery phase when the torque is released.
(8.4.19)
349
Plastic
2
2 = 2Y 2
Elastic
2 = 2
# $
=
=0
n
=0
Fig. 8.13 The freeboundary problem for an elasticperfectly plastic torsion bar.
two boundary conditions, namely that both and its normal derivative are
continuous. In principle, this problem determines both and the position
of the free boundary between the elastic and plastic regions.
This switch in behaviour across the free boundary is reminiscent of the
complementarity conditions encountered when modelling contact problems
in Chapter 7. Indeed, the freeboundary problem depicted in Figure 8.13
may be cast as a variational inequality and hence proved to be well posed,
despite the nonlinearity and the change of type from hyperbolic to elliptic.
Moreover, in most practical cases it can be proved to be equivalent to the
problem of minimising
2 4 dxdy
over all suciently smooth functions that vanish on the outer boundary and
are such that  Y /.
We also note that, as in our models for granular materials, although the
yield condition enables us to determine the stress, it tells us nothing about
the displacement in the bar once it has yielded. We will return to this point
in Section 8.10.
350
Plasticity
(8.4.21)
2
2 2
2A
2A 2A
= 4Y2 .
(8.4.22)
A +4
xy
x2 y 2
As always the key aspect of the problem is to locate the boundary where the
switch from (8.4.21) to (8.4.22) occurs. Here, a traction balance shows that
A and its rst and second partial derivatives must all be continuous across
such a boundary.
The nonlinear partial dierential equation (8.4.22) satised by A when the
material has yielded is another generalisation of the MongeAmp`ere equation, and closely resembles the equation (8.2.7) for granular ow in plane
strain. It is not immediately classiable using the standard theory of secondorder partial dierential equations, but the method of Exercise 4.12 shows
that it is hyperbolic, with two families of real characteristics
satisfying
2
/ 2
dy
A
A 2A
=2
Y
.
(8.4.23)
dx
xy
x2
y 2
These correspond to the directions in which the shear stress is maximal (see
Exercise 8.8), and hence the characteristics of (8.4.22) are the slip surfaces,
along which we might expect the material to ow.
351
P a2
,
r2
P a2
r2
(8.4.24)
as long as the material remains fully elastic (recall that P is the pressure
applied to the inner surface r = a).
As in the case of granular ow, the yield condition simplies considerably
when there is radial symmetry, and (8.4.20) reduces to
(rr )2 4Y2 ,
(8.4.25)
with equality when the material has yielded. The choice of square root is
dictated by (8.4.24) and hence the material yields when
rr = 2Y .
(8.4.26)
Y s2
,
r2
Y s2
,
r2
(8.4.27)
(8.4.28)
and the yield condition (8.4.26) simultaneously, with the boundary condition
rr = P on r = a, to obtain
r
r
rr = P + 2Y log
,
= P + 2Y + 2Y log
.
(8.4.29)
a
a
352
Plasticity
( rr ) /Y
PM = 2Y
0.5
0.0
1.5
0.5
2.0
PM = Y
2.5
3.0
r/a
1.0
1.5
2.0
Fig. 8.14 Residual shear stress r r in a gun barrel versus radial distance r for dierent values of the maximum internal pressurisation: PM /Y =
1, 1.25, 1.5, 1.75, 2.
s = a exp
2Y
2
This shows how the plastic region grows rapidly as P increases through Y .
We can now perform an unloading analysis similar to that in Section 8.4.1.
Suppose P increases to a value PM greater than Y , before decreasing to zero.
Our assumption that the material instantaneously becomes elastic when
P < PM means that, as in (8.4.17), the free boundary remains stuck at
its maximum value sM = a exp (PM /2Y 1/2). The subsequent response is
purely elastic so that
rr =
Y s2M
a2 (PM P )
+
,
r2
r2
Y s2M
a2 (PM P )
r2
r2
(8.4.31)
in r > sM , while
r
a2 (PM P )
+
,
(8.4.32a)
a
r2
r a2 (P P )
M
= PM + 2Y + 2Y log
(8.4.32b)
a
r2
in a < r < sM .
As in the torsion bar problem of Section 8.4.1, there remains a residual
stress eld when the loading P has returned to zero. As shown in Figure 8.14,
the residual shear stress rr increases in magnitude as PM is increased
past Y . The boundary r = a ends up in a state of negative shear stress
which, if PM > 2Y , will be large enough to cause the metal to yield again
as it recovers!
rr = PM + 2Y log
353
(8.4.33)
where the yield function f is some appropriate measure of the stress. Now
we ask ourselves what restrictions need to be imposed on the function f to
ensure that (8.4.33) is physically realistic.
If we assume that our metal is isotropic, then f must be invariant under
rotation of the axes. It follows that f can be written as a function of the
three stress invariants (cf Section 5.3.2)
1
Tr ( )2 Tr 2 , I3 ( ) = det ( ), (8.4.34)
I1 ( ) = Tr ( ), I2 ( ) =
2
and we have therefore reduced f from six degrees of freedom (the elements
of the symmetric tensor ij ) to three. Recall, for example, that the Coulomb
yield criterion (8.2.5) for granular material in two dimensions depends only
on I1 ( ) and I3 ( ).
For metals, further simplication can be achieved by recalling the experimental evidence that they do not deform plastically under an isotopic stress,
for which ij = pij , say; indeed it is intuitively reasonable that a pressure could not cause a dislocation like that sketched in Figure 8.9 to move.
Hence we expect that our yield function f should not vary if ij is replaced
by ij + cij , where c is any scalar, that is
f (ij + cij ) f (ij ) .
(8.4.35)
Other materials, for example glass, can, however, ow under sucient hydrostatic pressure.
354
Plasticity
(8.4.38)
Since this holds for all c, we deduce that f is independent of p and therefore
just a function of ij .
Now we can impose invariance under both rigidbody rotation and isotropic
pressure by writing
f ij = f 1 , 2 , 3 ,
(8.4.39)
where k are the eigenvalues of ij . It is easy to see that the principal axes
of ij and ij coincide, and that k are related to the principal stresses k
(k = 1, 2, 3) by
k = k + p,
where
p=
1 + 2 + 3
.
3
(8.4.40)
(8.4.41)
(8.4.42)
is called the Tresca condition. Notice that the function (8.4.41) is symmetric
with respect to permutation of its arguments, so that there are no preferred
directions.
In antiplane strain, with the stress tensor given in (8.4.1), the principal
deviatoric stresses are easily found to be
+
+
2 + 2 ,
2 + 2 ,
xz
=
3 = 0,
(8.4.43)
1 = xz
yz
yz
2
(a)
(b)
2
355
2
2 4
1.0
0.5
3
0
1.0
2
0.5
0.5
1.0
1
0.5
2
4
4
1
1.0
Fig. 8.15 (a) The Tresca yield surface (8.4.42) plotted in the space of principal
stress deviators (1 , 2 , 3 ) (normalised with Y ), showing the intersection with the
plane 1 + 2 + 3 = 0. (b) The projection in the (1 , 2 )plane.
356
Plasticity
(a)
(b)
2
2
0 2 4
1.0
0.5
3
0
1.0
0.5
0.5
1.0
1
2
0.5
2
4
4
1.0
1
Fig. 8.16 (a) The von Mises yield surface plotted in the space of principal stress
deviators (1 , 2 , 3 ) (normalised with Y ), showing the intersection with the plane
1 + 2 + 3 = 0. (b) The projection in the (1 , 2 )plane.
in antiplane strain. However, even in plane strain, the whole hexagon comes
into play in general (see Exercise 8.9), and the situation is even worse in
three dimensions. The presence of the corners in the yield surface is worrying, since it implies nonsmooth dependence of the solution on the stress
components. It also makes the Tresca condition awkward to implement in
general, and this has led to the adoption of the von Mises yield function
f 1 , 2 , 3 =
1/2
.
(8.4.46)
By construction, this is a valid measure of the stress with all the required
properties of invariance although, unlike the Tresca yield function, it has no
immediate interpretation in terms of a maximal shear stress.
In (8.4.46), we have dened f such that Y is still the critical yield stress
under pure shear. However, under a uniaxial stress, with 2 = 3 = 0,
the von Mises criterion predicts that yield will occur when 1 = 3Y , in
contrast with the value 2Y predicted by the Tresca condition.
The yield surface corresponding to (8.4.46), shown in Figure 8.16(a), is
still a cylinder whose axis points in the direction (1, 1, 1)T . However, elementary geometry implies that its crosssection is circular so, in contrast with
Figure 8.15, the von Mises yield surface is smooth. Again we can use the
constraint 1 + 2 + 3 to project the surface onto the (1 , 2 )plane, where
357
3 (1 + 2 )2
(1 2 )2
+
= Y ,
4
4
(8.4.47)
as shown in Figure 8.16(b). In antiplane strain, the Tresca and von Mises
yield criteria coincide, but in higher dimensions (8.4.46) is often easier to
implement, especially when we notice that it may be written in terms of the
nonprincipal stress components as
2
1
1
= Tr ( )2 = ij ij .
f
2
2
(8.4.48)
when
1 2 3 0.
(8.4.49)
358
Plasticity
0
2
4
0
2
0
Fig. 8.17 The Coulomb yield surface plotted in the space of principal stresses
(1 , 2 , 3 ) (with = /6).
8.5 Kinematics
This is the rst time in this book that we have tried to write down a mathematical model for irreversible ow, as opposed to elastic strain. Such models
are inevitably more readily stated in terms of the velocity of the medium,
rather than the displacement. We therefore introduce the velocity vector
v(x, t) dened as
u
x
=
,
(8.5.1)
v=
t
t
X
where the time derivative is taken with the Lagrangian coordinate X held
xed. It is convenient to introduce a shorthand for this convective or material
derivative, namely
D
=
+ (v ) ,
(8.5.2)
Dt
t X
t x
the latter identity following from the chain rule.
By applying the convective derivative to u, we obtain a kinematic relationship,
v=
u
+ (v ) u,
t
(8.5.3)
8.5 Kinematics
359
(8.5.7)
If the strains are small enough for us to neglect the nonlinear terms in
(8.5.3), then Dij is simply the time derivative of the linear strain tensor eij .
Otherwise, the relationship between strain and rateofstrain is complicated,
Note that uid mechanics textbooks often omit the factor of 1/2 in the denition of Di j
360
Plasticity
(8.6.1)
t
Along with the yield condition
f ( ) = Y ,
(8.6.3)
this gives us four scalar equations so far for the nine unknowns ij and vi .
Now our task is to obtain a constitutive relation for the stress eld associated
with a given ow velocity v. Before doing so, we will rst derive an equation
representing conservation of energy.
361
We will assume that the parameter c, called the specic heat capacity (at
constant volume), is constant.
In accordance with the rst law of thermodynamics, the energy inside V
will change because of (i) work done by stress on the boundary V , (ii) work
done by the body force g on the interior of V and (iii) heat conduction
through V . Combining these eects together, we obtain
dU
=
v ( n) dS +
v (g) dx +
kT n dS
dt
V (t)
V (t)
V (t)
=
{div ( v) + v g + div (kT )} dx,
(8.7.2)
V (t)
(8.7.5)
362
Plasticity
so the dissipation is exactly balanced by the change in W, and the net mechanical energy is conserved. The production of heat is thus a consequence
of our assumption that a owing plastic material (like a viscous uid) does
not store any elastic energy, and we will return to this point in Section 8.9.
Dij ij
subject to
f (ij ) = Y ,
(8.8.1)
whose solution is
Dij =
f
,
ij
(8.8.2)
Note that this would not be true if W depended explicitly on T ; see Exercise 9.7.
363
already imposed on the yield criterion. For example, an isotropic yield function will lead to an isotropic ow rule. In addition, our insistence that the
yield criterion be insensitive to isotropic stress means that f must satisfy
f
=0
kk
(8.8.3)
(8.8.4)
D2 = 0,
(8.8.6)
D3 = .
D1 = ,
2
2
However, this is often awkward to implement in practice because of the
corners that occur in the yield surface when the principal stresses cross each
other (see Calladine, 2000). In threedimensional problems, the von Mises
yield function (8.4.48) is therefore usually preferred. In this case, (8.8.2)
simply tells us that the rateofstrain tensor is proportional to the stress
deviator, that is
ij ,
(8.8.7)
Dij =
2Y
as shown in Exercise 8.13. This is called the Levyvon Mises ow rule, and
it again implies that the principal axes of stress and strainrate coincide.
364
Plasticity
The ow rule (8.8.7) closely resembles the constitutive relation for an inole of viscosity
compressible Newtonian viscous uid, with Y / playing the r
(see, for example, Ockendon & Ockendon, 1995, Chapter 1). Unlike the situation in a Newtonian uid, the eective viscosity is unknown in advance and
must be determined as part of the solution, using the fact that the stress
is bounded by Y . However, by using (8.8.7), we can write the dissipation
(8.7.5) in the form
.
(8.8.8)
=
2Y ij ij
We must therefore ensure that is nonnegative, so that the plastic ow
dissipates mechanical energy rather than gaining it. It is straightforward to
reach the same conclusion for the Tresca ow rule (8.8.6).
In summary, when we adopt the von Mises yield function, the mathematical model comprises the three rstorder dierential equations (8.6.2),
relating and v and the seven algebraic relations (8.6.3), (8.8.7), which also
involve the unknown scalar .
(8.9.1)
We assume that these are related to the stress ij by the usual isotropic
linear elastic relation (1.7.6) and the plastic ow rule (8.8.7) respectively.
Thus
epij
e
=
ij ,
(8.9.2)
Eeij = (1 + ) ij kk ij ,
t
2Y
8.10 Examples
ij +
E
ij .
t
t
t
2Y
365
(8.9.3)
Combining (8.9.3) with the momentum equation (8.6.2) and the yield condition (8.6.3), we obtain a closed system of ten equations for the six stress
components ij , the three displacement components ui and the scalar .
To x ideas, let us suppose that inertia and gravity are negligible so that
the momentum equation (8.6.2) reduces to
ij
= 0.
(8.9.4)
xj
This holds in both elastic and plastic regions, as does (8.9.3) provided we set
= 0 when the material is elastic. On the other hand, when the material
is plastic, we know that must be positive and the yield condition (8.6.3)
must be satised. The problem thus boils down to the system (8.9.4) and
(8.9.3), along with the complementarity conditions
f (ij ) Y ,
0.
(8.9.5)
f (ij ) Y = 0,
To solve this problem, we would need to specify the initial displacement
(typically zero) and three boundary conditions on either the displacement or
the stress. In addition, the strain and displacement must both be continuous
across the free boundary separating the elastic and plastic regions. We will
now illustrate the procedure using three simple examples.
8.10 Examples
8.10.1 Torsion revisited
First we return to the torsion bar example from Section 8.4.1. If we limit
our attention to circular bars, we can conveniently use cylindrical polar
coordinates, seeking a solution in which the displacement takes the form
u = (t)rze . This automatically enforces continuity of strain and displacement throughout the bar. According to (1.11.6), the only nonzero strain
component is
r
,
(8.10.1)
ez =
2
so that (8.9.3) gives us the equation
z
d
+
(8.10.2)
z = r
t
Y
dt
for z (r, t). The initial condition is z (r, 0) = 0, but, before we can solve
(8.10.2), we need a further condition to determine .
366
Plasticity
(8.10.3)
d
dt
(8.10.4)
in s < r < a.
Hence we see that is positive as long as is an increasing function of t.
If the twist reaches a maximum and then starts to decrease, then (8.10.4)
gives us a negative value of , which is thermodynamically unacceptable.
Hence it is because of the imposed nonnegativity of that the bar must
revert to being entirely elastic as it unloads, as was assumed in Section 8.4.1.
(8.10.5)
epzz
eprr
= ,
= ,
= 0.
(8.10.6)
t
2
t
2
t
By combining the elastic and plastic strain components as in Section 8.9,
we thus obtain the equations
err
rr
zz
=
t
t
t
t
zz
rr
e
=
+
E
t
t
t
t
zz
rr
ezz
=
.
E
t
t
t
t
E
E
,
2
E
,
2
(8.10.7a)
(8.10.7b)
(8.10.7c)
8.10 Examples
367
zz = (rr + ) .
(8.10.9)
E
= (1 + ) ( rr ) + E.
t r
r
t
(8.10.10)
(8.10.11)
While P Y , the material remains elastic, so that is zero, and integration of (8.10.10) leads to
ur
ur
E
= (1 + ) ( rr ) .
(8.10.12)
r
r
We therefore nd that
ur
ur
= rr
= rr ,
E
E
(8.10.13)
r
r
and the pure elastic stress eld (8.4.24) gives us the displacement
ur =
P a2
,
2r
(8.10.14)
Y s2
,
2r
(8.10.15)
368
Plasticity
(Incidentally, we can use these and (8.10.9) to verify that the inequalities in
(8.10.5) are indeed satised.) Hence (8.10.10) takes the form
r
ur
2(1 2)Y
ur
+
=
log
,
(8.10.17)
r
r
s
which is easily integrated with respect to r to
r
(1 )Y r
log
ur =
give
C(t)
1
+
.
2
r
(8.10.18)
+
.
(8.10.19)
ur =
s
2
r
Finally, substitution of (8.10.19) into (8.10.11) leads to
ds
4(1 )Y
s .
=
2
r
dt
(8.10.20)
8.10.3 L
uders bands
One of the most famous manifestations of metal plasticity is the formation
of L
uders bands in polished bars or sheets placed under a suciently large
tensile stress to cause plastic ow (see, for example, Thomas, 1961, p. 100).
As illustrated in Figure 8.18, surface roughness associated with plastic strain
is observed in bands inclined at a welldened angle to the direction in
which the tension is applied. To describe this situation fully necessitates the
solution of a dicult twodimensional freeboundary problem. We therefore
focus on the simplied model problem of an innite plate lying in the (x, y)plane, with a straight line separating an elastic region in y < mx from a
plastic region in y > mx.
The
elastic region is subject to a uniaxial stress T , which must be equal
to 3Y under the von Mises criterion. In the plastic region, we have a plane
stress eld of the form
xx xy 0
= xy yy 0 ,
(8.10.21)
0
0 0
8.10 Examples
369
x
T
Fig. 8.18 L
uders bands in a thin sheet of metal.
2xx yy
3xy
0
1
.
= 3xy
2yy xx
0
3
0
0
xx yy
(8.10.22)
We assume that these stress components are spatially uniform, which would
clearly be inadequate to describe a nite plate with stressfree edges.
Continuity of traction across y = mx implies that
xx mxy = mT,
xy myy = 0,
(8.10.23)
(8.10.24)
This gives only three equations for the four unknowns xx , xy , yy and m.
The last piece of information comes from the zzcomponent of (8.9.3), namely
E
E
ezz
= (xx + yy )
(xx + yy ) .
t
t
6Y
(8.10.25)
In the elastic region, ezz is constant and given by T /E. Hence, by continuity of strain at the elasticplastic interface, ezz /t = 0 and (8.10.25)
becomes
E
(xx + yy ) +
(xx + yy ) = 0.
(8.10.26)
t
6Y
Although is as yet undetermined, we know that it must be positive
whenever plastic ow is taking place, so that (xx + yy ) necessarily tends
370
Plasticity
1
T
2T
(8.10.27)
,
xy = ,
m = ,
xx = yy =
3
3
2
and the elasticplastic boundary therefore makes an angle of
= cot1
2 35
(8.10.28)
with the xaxis. Notice that only one component of (8.9.3) was required to
get this result; the others may be used subsequently to calculate the plastic
strain and the ow parameter .
371
the iteration, which can therefore only predict the sequence of congurations
and hence any ultimate steady state.
In all the plastic models that we have considered, analytical progress is
almost always impossible if the inertial terms are included on the lefthand
side of the momentum equation (8.6.2). Fortunately they are often negligible
in practice. We can determine when this is true by calculating the appropriate dimensionless parameter as follows. First we write (8.6.2) in terms of
the deviatoric stress as
v
+ (v ) v
t
= p + + g.
(8.11.2)
V 2
.
Y
(8.11.3)
For example, the density and yield stress of lowstrength steel are approximately 8000 kg m3 and 5 108 N m2 respectively. Hence, if we bend a
paper clip at about 10 cm s1 , then the corresponding value of Re is roughly
107 , and inertia is certainly negligible.
We note, however, that there is also a highRe regime in which the plastic
ow is so fast that the inertia term in (8.11.2) is dominant, and the plastic
stress is negligible in comparison. This limit applies, for example, to the
violent metal plasticity caused by a device called a shaped charge, in which
a metal cone is compressed by an explosive to such an extent that its vertex
emits a plastic metal jet moving at enormous speed (up to 14 kilometres
per second). This exerts a pressure on the target that can be orders of
magnitude larger than its yield stress, and the target therefore gives way
plastically as the jet forces its way through. Shaped charges are chillingly
eective, able to penetrate over a metre of steel. When Re is very large,
(8.11.2) and (8.8.4) reduce to the Euler equations of incompressible inviscid
uid dynamics: the target ows like an incompressible inviscid liquid to
lowest order. Thus shaped charge penetration is relatively simple to model
compared with plastic penetration at lower stresses.
372
Plasticity
Exercises
8.1
8.2
F = n N n.
Exercises
373
Fig. 8.19 The Mohr surface in the (N, F )plane for threedimensional granular ow,
and the line F = N  tan . The region where solutions exist is shaded.
8.4
Reach the same conclusion for the system (8.2.5) (with equality),
(8.2.6) including a body force.
(a) For the displacement eld (8.3.1) corresponding to a screw
dislocation, calculate the jump in w as a simple closed contour
C in the (x, y)plane is traversed, and hence show that
# $
b if C encloses the origin,
w
w
dx +
dy =
w C =
x
y
0 otherwise.
C
Show that
2
b if S contains the origin,
w
2w
dxdy =
xy yx
0 otherwise,
S
where S is the interior of C. Deduce that w satises (8.3.3).
(b) Alternatively, observe that, although w has a jump of magnitude b across the positive xaxis, w/x is continuous. Hence
explain why
by
w
=
,
x
2 (x2 + y 2 )
8.5
w
bx
=
b H(x)(y),
y
2 (x2 + y 2 )
where H denotes the Heaviside function (3.5.7), and, by crossdierentiating, obtain (8.3.3).
Verify that the PapkovichNeuber potentials
= 0,
1 =
2b
,
(3 4)
2 =
2b
(3 4)2
374
8.6
Plasticity
Using the relations (2.6.20) for the polar stress components in terms
of the Airy stress function A, show that (8.3.8) is consistent with
A = c1 r cos + c2 r log r sin ,
and determine the appropriate values of the constants c1 and c2 .
Hence show that A satises
Eb
Eb
sin
2
=
,
A=
(1 + )(3 4)
r
(1 + )(3 4) x
8.7
3c
1
2M
+
4
=
,
a3 Y
3
c
3M
and hence evaluate the residual twist in the bar when M = 0.
Suppose we now twist the bar in the opposite direction, so that M
is negative. Use (8.4.17) to show that the bar yields again at r = a
when
= 2c ,
3c
2M
2
> 1.
a3 Y
3 33M
[Thus, after the bar has yielded in one direction, a smaller torque is
required to make it subsequently yield in the opposite direction. This
is known as the Bauschinger eect.]
Show that the stress in the bar is then given by
0 r < sm ,
+ M ,
y
xz
+ M sm /r, sm < r < ,
=
yz
x
< r < a,
Y /r,
where the position of the new yield boundary is
r==
c a
.
3
a3 Y
3 33M
3
Exercises
375
2M/a3 Y
2
1.0
0.5
3
2
1
1
1
1.0
/c
0.5
8.8
8.9
2 (xy Y )
.
xx yy
376
Plasticity
2
1 2
1+
p.
For each xed value of p, this gives us a straight line in the (1 , 2 )plane. By considering the intersection of this line with the polygon
in Figure 8.15(b), show that the plane strain Tresca yield condition
is 1 2  = 2Y provided
p
8.10
2(1 + )Y
.
3(1 2)
[If this condition is violated, then the normal stress zz that must be
applied to prevent outofplane displacement will cause the material
to yield in the zdirection.]
(a) Recall that
xi
J = det
Xj
is the Jacobian of the transformation from Lagrangian to Eulerian variables. Prove Eulers identity
DJ
= J div v,
Dt
where D/Dt is the convective derivative. Deduce that the density satises the continuity equation
+ div(v) = 0.
t
(b) Hence prove Reynolds transport theorem: for any volume V (t)
that is convected with velocity v(x, t) and any dierentiable
function F (x, t),
DF
d
dx.
F dx =
dt
V (t)
V (t) Dt
8.11
Exercises
377
8.12
vj
d
vi
(xi ) =
xi xj =
xi xj .
dt
xj
xi
Hence show that the length of the line element satises (8.5.6).
Include elastic energy in (8.7.1) by writing
1
W
2
v + cT +
dx.
U=
J
V (t) 2
The strain energy density W is a function of the elastic deformation
gradient tensor Fije (see (5.2.2)) satisfying
e 1
W
= Jik Fjk
.
e
Fij
Show that
DW
e
= JDij
ij ,
Dt
e is the elastic rateofstrain tensor. Deduce that the dissiwhere Dij
pation function in (8.7.4) becomes
p
ij ,
= Dij
8.13
p
where Dij
is the plastic rateofstrain, so that the net rateofstrain
e + Dp .
is given by Dij = Dij
ij
(a) Show that the ow rule (8.8.2) is invariant under rotations of
the coordinate axes.
(b) Show that the Levyvon Mises ow rule leads to (8.8.7) when
we choose principal axes in which the stress tensor is diagonal.
(c) Deduce that (8.8.7) is valid with respect to any chosen axes.
9
More general theories
9.1 Introduction
We will now mention a wide range of important solid mechanics phenomena
that warrant discussion even in a mathematicallyoriented book but have
been ignored or scarcely mentioned so far.
A phenomenon of interest in industries ranging from food to glass is that
of viscoelasticity. Here the molecular structure of the material is such that
it ows under any applied stress, no matter how small. As we will see in
Section 9.2, viscoelastic materials are quite unlike plastic materials. Their
behaviour depends critically on the timescale of the observer; for example when a ball of silly putty or suitably dilute custard impacts a wall,
it rebounds almost elastically but, if left on a table, it will slowly spread
horizontally under the action of gravity.
Yet another attribute of solids is wellknown to be of great practical importance in the kitchen, when glass can be observed to break in hot water,
and in the railway industry, where track can distort in high summer. This
is thermoelasticity, which describes the response of elastic solids to temperature variations. To model this response ab initio, even at a macroscopic
scale, requires more thermodynamics than is appropriate for this text, and
in Section 9.3 we will only present the simplest ad hoc model that can give
useful realistic results.
Even the above list of diverse phenomena only relates to solids that are
fairly homogeneous on a macroscopic scale. Hence this book would certainly not be a fair introduction to the mathematics of solid mechanics
without some discussion of the increasingly important properties of composites. These are solids in which the elastic properties vary on a lengthscale
much smaller than any macroscopic lengthscale of practical relevance but
not so small as to render a continuum model inappropriate. They range from
378
9.2 Viscoelasticity
379
380
(a)
(b)
u
T
(d)
(c)
Te
Te
ue
T
Tv
T
uv
Tv
u
Fig. 9.1 (a) A spring; (b) a dashpot; (c) a spring and dashpot connected in parallel;
(d) a spring and dashpot connected in series.
(9.2.1)
du
,
dt
(9.2.2)
9.2 Viscoelasticity
(a)
te
381
(b)
(c)
where
Te = ku,
Tv = Y
du
,
dt
(9.2.3)
where the subscripts e and v stand for the elastic and viscous contributions respectively. These are easily combined to give the Voigt constitutive
equation
Y
du
du
+ ku = k tr
+u ,
where
tr = ,
(9.2.4)
T =Y
dt
dt
k
and tr is known as the relaxation time of the Voigt element.
The qualitative behaviour depends crucially on the size of the relaxation
time compared with the characteristic timescale te of any given experiment, and we therefore dene a dimensionless parameter, called the Deborah
382
(a)
(b)
Fig. 9.3 Displacement of a Voigt element due to the applied tension shown in Figure 9.2(a). (a) Small Deborah number; the corresponding response of an elastic
spring is shown as a dashed curve. (b) Large Deborah number; the corresponding
response of a dashpot is shown as a dashed curve.
number, by
De =
tr
.
te
(9.2.5)
When De is small, the time derivative in (9.2.4) has a small inuence, and the
element therefore behaves somewhat like an elastic spring. In Figure 9.3(a)
we show the response of such an element to the same applied tension as in
Figure 9.2(a). Here we see that the displacement lags slightly behind the
corresponding elastic displacement (shown as a dashed curve), where the
time lag is of order tr . However, when De is large, the time derivative is
the dominant term on the righthand side of (9.2.4), so the Voigt element
responds like a dashpot, as shown in Figure 9.3(b). However, if we wait long
enough, the displacement will always decay to zero, over a timescale tr ,
when the tension is released. The Voigt constitutive relation (9.2.4) therefore
describes the basic building block for a damped solid.
On the other hand, Figure 9.1(d) illustrates what is known as a Maxwell
element, in which the spring and dashpot experience the same tension T
and the net displacement is the sum of the elastic and viscous contributions;
hence
duv
.
(9.2.6)
where
T = kue , T = Y
u = u e + uv ,
dt
Now dierentiation with respect to t leads to the Maxwell constitutive equation
d T
dT
du
=Y
+ T = tr
+ T,
(9.2.7)
Y
dt
dt k
dt
where tr = Y/k is again the relaxation time.
Now, when De 1, the time derivative term is small, so (9.2.7) is approximately that of a dashpot, as shown in Figure 9.4(a). The response when
9.2 Viscoelasticity
(a)
383
(b)
Fig. 9.4 Displacement of a Maxwell element due to the applied tension shown in
Figure 9.2(a). (a) Small Deborah number; the corresponding response of a dashpot
is shown as a dashed curve. (b) Large Deborah number; the corresponding response
of an elastic spring is shown as a dashed curve.
where K(t), known as the creep function, characterises the viscoelastic response of the system. For example, the Voigt and Maxwell constitutive laws
may be cast in the form (9.2.8), with the creep function given by
K(t) =
1 et/tr
k
and
K(t) =
tr + t
,
Y
(9.2.9)
384
where and are the usual Lame constants, while and are called
the dilatational viscosity and shear viscosity respectively. As explained in
Section 8.8, the rateofstrain tensor Dij may be approximated as simply the
time derivative of the strain tensor eij provided the displacement gradients
are small enough for linear elasticity to be valid.
If we make the simplifying assumption that the elastic and viscous components have the same Poissons ratio, then we can dene a single relaxation
time
(9.2.11)
tr = = .
( (ekk ) ij + 2eij ) ,
(9.2.12)
ij = 1 + tr
t
for small displacements of a Voigt solid.
Alternatively, we can base a Maxwell model on the idea that the stress
at any point gives rise to both an elastic strain eeij and a viscous strain evij .
9.2 Viscoelasticity
385
Again we assume that these are both small enough to be described by linear
theory and that the viscous and elastic components have the same Poissons
ratio to obtain
(evkk ) ij + 2evij .
eij = eeij + evij , ij = (eekk ) ij + 2eeij = tr
t
(9.2.13)
Now these are easily combined to give the linear Maxwell model
ij
(9.2.14)
+ ij = tr ( (ekk ) ij + 2eij ) .
t
t
Let us illustrate the equations that result from a viscoelastic constitutive
relation by returning to the model for compressional waves in a bar. We
recall from Section 4.2 that Newtons second law leads to the equation
tr
T
2u
,
(9.2.15)
=
t2
x
where and A are the density and crosssectional area of the bar. To close
the model we need to specify a constitutive relation between the tension T
and the displacement u. The Voigt model (9.2.12) leads to
u
2u
T = EA
+ tr
,
(9.2.16)
x
xt
A
(9.2.17)
2u
T
+ T = EA
,
t
xt
(9.2.18)
3
3u
u 2u
tr 3 + 2 = E 2 .
t
t
x t
(9.2.19)
386
where the creep function is now in general a fourthrank tensor. If the material is isotropic, then (9.2.20) may be characterised using just two scalar
creep functions L(t) and M (t) by writing
t
kk
ij
x, t ij + M t t
x, t
dt .
L tt
eij =
t
t
(9.2.21)
In either case, the integral should really be performed with the Lagrangian
variable X held xed, and we must emphasise once more that all these
linear theories are only valid for small strains. We can expect them to work
well for small displacements of a damped solid or for small creep of an
elastic uid, but for a uid that undergoes a substantial deformation, a
more sophisticated approach is required, as we will now briey explain.
+ (v ) v = p + + g,
(9.2.22)
t
where v is the velocity vector.
Since we would not expect a Voigt solid to undergo particularly large
displacements, we concentrate on the example of a Maxwell uid, in which
the deformation is decomposed into an elastic component and a viscous
component. As described in any book on viscous ow (for example Ockendon
& Ockendon, 1995), the constitutive relation for a Newtonian viscous uid
is simply
ij = 2Dij ,
(9.2.23)
where Dij is the rateofstrain tensor and is again the shear viscosity.
9.2 Viscoelasticity
387
On the other hand, a neoHookean solid, with strain energy density given
by (5.3.58a), obeys the constitutive relation
ij = (Bij ij ) ,
(9.2.24)
(9.2.25)
that is
DB
= v T B + Bv,
(9.2.26)
Dt
where D/Dt = /t + (v ) is the usual convective derivative.
The neoHookean stress tensor (9.2.24) therefore satises the equation
= 2D,
(9.2.27)
A=
tr + = 2D,
(9.2.29)
(9.2.30)
= 2D,
(9.2.31)
A=
DA
+ vA + Av T .
Dt
(9.2.32)
instead of
in
388
A=
DA
A + A a (DA + AD) ,
Dt
2 xj
xi
(9.2.33)
(9.2.34)
The upper and lower convected derivatives are obtained when a = 1 and
when a = 1 respectively, and another common choice is the Jaumann
9.3 Thermoelasticity
Thermal eects in solids can be dramatic, even if we exclude the possibility of
the solid melting into a liquid. For example, the easiest way in which to fracture an ice cube is to put it into even lukewarm water. We note that density
variations are already allowed for in the mass and momentumconservation
equations (1.3.6) and (1.6.5). Hence it only remains for us to incorporate
thermal eects in the constitutive equation (1.7.6) and then to include heat
in the energy calculation from Section 1.9. We will assume throughout that
the displacement gradients are small enough for linear elasticity to be valid.
9.3 Thermoelasticity
389
We begin to construct a linear thermoelastic constitutive relation by considering an unconstrained elastic solid whose absolute temperature is raised
from T0 in the reference state to a new temperature T . There is ample experimental evidence that many common materials undergo a uniform isotropic
volume change, namely an expansion if T > T0 or a contraction if T < T0 .
For moderate temperature variations, the thermal expansion is found to vary
linearly with temperature, so that
(9.3.1)
eij = (T T0 )ij ,
3
where is called the coecient of thermal expansion. By referring to Section 2.2.1, we can interpret as the relative volume expansion caused by a
unit increase in temperature.
We are therefore prompted to propose the constitutive relation
2
(9.3.2)
ij = 2eij + ekk ij + (T T0 )ij .
3
This is the only linear relation between ij , eij and T which reduces to
(1.7.6) when T = T0 and to (9.3.1) when ij = 0. In a linear theory, the
Lame constants are assumed not to vary signicantly over the temperature
ranges of interest. When we use (9.3.2) in the momentum equation (1.6.5),
we nd that the Navier equation generalises to
2u
2
2
(9.3.3)
2 = g + ( + ) grad div u + u + T,
t
3
so that the temperature gradient generates an eective body force in the
Navier equation.
At rst glance, it might appear from (9.3.3) that a uniform temperature
change, with T = 0, could not drive a displacement. However, whenever
boundary conditions are specied on the displacement, the constitutive relation (9.3.2) will lead to nontrivial stress, and vice versa. We illustrate this
by returning to the steady uniaxial loading of a bar, considered previously
in Section 2.2.3. Here we examine the eect of heating the bar uniformly to
some temperature T . If we seek a displacement of the form
u = (x, y, z)T ,
then the stress tensor is given by (9.3.2) as
0
0
2
= 2 0 0 + ( 2) +
(T T0 ) I.
3
0 0
(9.3.4)
(9.3.5)
390
(3 + 2)(T T0 )
1+
=
(T T0 ),
2( + )
6( + )
3
(9.3.6)
EA
(T T0 ),
3
(9.3.8)
where A is the crosssectional area. Note that heating the track will cause a
compressive force and cooling a tensile force, as expected. From Section 4.4.3,
we know that the track will buckle if the compressive stress is too large, and
(9.3.8) tells us that the maximum temperature that can be sustained before
buckling occurs is
T T0 =
3 2 I
,
AL2
(9.3.9)
2
1 u
U=
+ W + cT dx,
(9.3.10)
2 t
V
where, referring to Section 1.9, we recognise the rst two terms in the integrand as the kinetic energy and the strain energy. The nal term is the thermal energy, with c denoting the specic heat capacity, a material parameter
which we will assume to be constant. As in Section 8.7 and Exercise 8.12,
the rst law of thermodynamics tells us that all three forms of energy are
In the days before continuouslywelded rails, gaps had to be left between sections of track to
allow for thermal expansion to take place without stress generation. These gaps caused track
and wheels to wear and led to many accidents.
As usual, this should be dened at constant volume
391
392
E
E1
(a)
(b)
E2
x
Fig. 9.5 (a) The variation of Youngs modulus with position in a bar. (b) The
corresponding longitudinal displacement u(x); the approximation (9.4.6) is shown
as a dashed line.
du
= T,
dx
(9.4.1)
(1
)nL
1
x
1
N
E2
E1
T E1
u(x) =
A
(n + 1)L 1
1
x
,
+
E2
N
E1
E2
(n + )L
nL
<x<
,
N
N
(n + 1)L
(n + )L
<x<
.
N
N
(9.4.3)
In Figure 9.5, we plot a typical piecewise constant Youngs modulus prole
and the corresponding displacement u(x). We can see that u(x) uctuates
rapidly in response to the variations in E. However, on the scale of the whole
bar, these uctuations are barely noticeable, and the displacement appears
to be approximately linear in x. Indeed, from the exact solution (9.4.3), we
393
(1 )
TL
+
u(x + L/n) u(x) =
N A E1
E2
TL
(1/E),
=
NA
where
N
(1/E) =
L
L/N
dx
E(x)
(9.4.4)
(9.4.5)
du
T,
dx
(9.4.7)
where the eective Youngs modulus of the composite bar is the harmonic
average of E, that is
1
.
(9.4.8)
Ee =
(1/E)
This illustrates what the point of homogenisation theory is. When we consider the bar as a whole, the complicated nescale variations in E become
irrelevant, and the bar behaves like a uniform medium whose modulus is
given by (9.4.8). Now we show how a formal asymptotic method, known as
the method of multiple scales (see Kevorkian & Cole, 1981), may be used to
reach the same conclusions in more general congurations. These techniques
come into their own in more complicated problems where explicit solutions
like (9.4.3) do not exist.
As always, before attempting any asymptotic analysis, it is necessary to
nondimensionalise the problem. Here, we scale x with the bar length L, E
with a typical value E1 and u with T L/E1 A to obtain
E
du
= 1.
dx
(9.4.9)
Now x lies between 0 and 1, but E varies over a much smaller lengthscale,
394
(9.4.10)
E(x, )
u
u
+
(9.4.11)
= .
(9.4.12)
dx
(9.4.14)
/
Ee (x) = 1
0
du0
= 1,
dx
d
E(x, )
(9.4.15)
=
1
(1/E)
(9.4.16)
This is a very convenient assumption but, as discussed in Kevorkian & Cole (1981), it is not
necessary for more general problems in which the microgeometry is not known in advance,
and can be replaced by the condition that u not grow as  .
E
E1
(a)
395
(b)
E2
x
Fig. 9.6 (a) The variation of Youngs modulus with position in a bar with = 1/5;
the eective Youngs modulus is shown as a dashed curve. (b) The corresponding
longitudinal displacement u(x); the leadingorder approximation, given by (9.4.15),
is shown as a dashed curve.
)E
1 (x) + E2 (x)
0
which averages out the ne scale variations. We see in Figure 9.6(b) that the
displacement calculated using Ee (x) (dashed curve) provides an excellent
approximation to the exact solution (solid curve), even though = 1/5 is
not particularly small.
(9.4.19)
396
b
1
We suppose that varies both rapidly over a short lengthscale and gradually across the whole crosssection, so that we can write
= (x, y, , ),
(9.4.20)
(x, y, , + b) (x, y, , ).
(9.4.21)
We choose such that the period in is equal to 1, and then dene b to be the
period in the direction. A typical setup is illustrated in Figure 9.7, where
the modulus is piecewise constant on the ne scale, although 1 (x, y) and
2 (x, y) might vary over the long scale.
Now, as in Section 9.4.1, we allow the displacement eld w(x, y, , ) to
vary over both lengthscales, provided it shares the same periodicity (9.4.21)
as . The chain rule thus transforms (9.4.19) into
( w) + { (xw) + x ( w)} + 2 x (xw) = 0,
(9.4.22)
where x = (/x, /y)T , = (/, /)T . The assumed periodicity
means that we need only solve (9.4.22) on a single rectangular cell [0, 1][0, b]
397
(9.4.24)
(9.4.25)
This equation and the boundary conditions (9.4.23) are identically satised
if w0 is independent of and , so that w0 = w0 (x, y), and it may be shown
(see Exercise 9.10) that this is the only solution. As we might have hoped,
the displacement therefore does not vary on the ne lengthscale to leading
order.
Comparing the terms at O (), we obtain
w1
w1
w0
w0
= R1 (x, y, , ), say,
x
y
(9.4.26)
which is a linear elliptic partial dierential equation for w1 driven by xw0 .
As shown in Exercise 9.10, equation (9.4.26) admits periodic solutions
for w1 only if the righthand side has zero mean, so that
1 b
R1 (x, y, , ) dd = 0.
(9.4.27)
0
This requirement is satised identically here, and we can write the solution
in the form
w0
w0
w1 = W (1) (x, y, , )
+ W (2) (x, y, , )
,
(9.4.28)
x
y
where W (1) and W (2) satisfy the equations
W (1)
W (1)
and
W (2)
W (2)
=
(9.4.29)
(9.4.30)
398
It can be shown that (9.4.29) and (9.4.30) have unique periodic solutions
for W (1) and W (2) , up to arbitrary additive functions of (x, y). Although it
is usually not possible to obtain these solutions in closed form, it is straightforward in principle to compute them numerically for any given modulus
function (x, y, , ).
Finally, proceeding to the terms of O 2 in (9.4.22), we nd that w2
satises an equation analogous to (9.4.26), but with the righthand side
replaced by
w1
w1
R2 (x, y, , ) =
y
w1
w1
w0
w0
, (9.4.31)
x
x
y
y
and again we need this to have zero mean over our periodic cell in (, ).
The rst term satises this requirement identically, and, when we substitute
for w1 from (9.4.28), we nd that
1 b
0=
R2 (x, y, , ) dd
0
0

.
1 b
w0
W (2) w0
W (1)
+1
+
=
x
x
y
0
0
.
W (1)
w0
W (1) w0
+
+1
dd. (9.4.32)
+
y
x
y
Thus our homogenised displacement w0 (x, y) satises the anisotropic model
w0
w0
w0
w0
+
+
+
= 0,
11
12
21
22
x
x
x
y
y
x
y
y
(9.4.33)
where
b 1
b 1
W (1)
W (1)
dd,
dd,
11 =
1+
21 =
0
0
0
0
b
22 =
1+
W (2)
b
dd,
12 =
(9.4.34a)
1
W (2)
dd.
(9.4.34b)
399
0
Hence the eective moduli are given by
b
11 =
d,
21 = 12 = 0,
0
(9.4.35)
/
22 = b
d
,
(9.4.36)
(2) = W (2) + ,
(9.4.37)
the equations (9.4.29) and (9.4.30) for W (1) and W (2) are transformed to
(9.4.38)
(1) = (2) = 0,
where
=
1 , (, ) D1 ,
2 , (, ) D2 .
(9.4.39)
Hence (1) and (2) both satisfy Laplaces equation in each of D1 and D2 ,
along with the continuity conditions
.
(
'
(j)
(j)
=0
(9.4.40)
=
n
on the boundary D2 between them.
Crucially, we now assume that the geometry possesses reectional symmetry in both the  and the direction. Using this symmetry and the periodicity of W (j) , we deduce that (j) must satisfy the boundary conditions
(1) (0, ) = 0,
(1) (1, ) = 1,
(2) (, 0) = 0,
(2) (, b) = b,
(1)
(, 0) =
(2)
(0, ) =
(1)
(, b) = 0,
(2)
(1, ) = 0,
(9.4.41a)
(9.4.41b)
400
(a)
(1)
=0
(2) =b
(b)
(2)
=0
(1) =
b
(1)
=0
D1
(1) =1
(2) =0
(2) =
D2
(2)
=0
(2)
=0
2
1
D2
(1)
=0
(2) =0 1
(1)
=0
(2)
=0
(1) =0 1
(1)
=0
Fig. 9.8 A symmetric, piecewise constant shear modulus distribution; the symmetry
axes are shown as dotdashed lines. (a) The boundary conditions satised by (1)
and (2) . (b) The corresponding boundary conditions for the harmonic conjugate
functions (1) and (2) .
Now, since (j) satisfy Laplaces equation in each separate region, we can
introduce the harmonic conjugates (j) which satisfy the CauchyRiemann
equations
(j)
(j)
=
,
(j)
(j)
=
.
(9.4.42)
(j)
(j)
=
n
.
= 0,
(9.4.43)
# $
where
again denotes the jump across D2 . The boundary conditions
(9.4.41) become
(1) (, 0) = 0,
(1) (, b) = ,
(1)
(1)
(0, ) =
(1, ) = 0, (9.4.44a)
(2)
(2)
(, 0) =
(, b) = 0, (9.4.44b)
401
(1)
dd =
b
(1)
dd
1'
(=b
(1)
d = 1 , (9.4.45)
=
=0
21 = 12 = 0,
(9.4.46)
(9.4.47)
then the boundary conditions for %(1) are identical to (9.4.41a), although
the continuity conditions on D2 read
%(1)
%(1)
(1)
(1)
2
(9.4.48)
% = % ,
= 1
.
n
n
D1
D2
D1
D2
so that
2
,
%(1) =
1
(9.4.49)
'
(=b
2 b
.
=
%11 = 2 %(1)
=0
(9.4.50)
%11 22 = b2 1 2
(9.4.51)
%11 in the direction with 1 and 2 switched. This has particularly startling
implications if b = 1 and the geometry is antisymmetric about a diagonal
in the resulting square cell, as in the examples shown in Figure 9.9. In such
congurations, swapping 1 and 2 is equivalent to swapping and , so
402
Fig. 9.9 Some modulus distributions that are antisymmetric about the diagonals
of a square.
that
%11 22 . Hence a homogenised bar consisting of such cells is isotropic
with
11 = 22 = 1 2 .
(9.4.52)
Such results provide an invaluable check on the numerical methods to
which one usually has to resort to calculate the eective elastic constants.
uk
,
x
(9.4.53)
(9.4.54)
but it is more convenient here to persist for the moment with the more
general case (9.4.53), which we can write symbolically as a tensor product:
= Cu.
(9.4.55)
403
that C, u and are all periodic on some representative cell D, for example
a cuboid, in space.
The advantage of working with the rstorder system (9.4.53), rather than
the Navier equations, is that when we expand
u u0 (x, ) + u1 (x, ) + , 0 (x, ) + 1 (x, ) + , (9.4.56)
we only need to consider terms up to O (), in contrast with Section 9.4.2
where we had to proceed to O 2 .
At leading order, we soon nd that
0 = 0,
(9.4.57)
C u0 = 0,
(9.4.58)
(9.4.59)
(9.4.60)
(9.4.61)
(9.4.62)
(9.4.63)
(9.4.64)
As always, the nal step is to demand that be periodic on our unit cell
D. This means that
1 d 0,
(9.4.65)
D
404
(9.4.66)
where
Ce (x) = (C A + C).
(9.4.67)
We remark again that the eective elastic constants can rarely be evaluated explicitly, although, for any given microstructure, they can in principle
be calculated numerically over a representative unit cell. Another crucial
observation is that, even if the matrix is constructed from isotropic material, so that C is given by (9.4.54), the averaging will in general lead to
an anisotropic macroscopic constitutive law. Of course, most natural heterogeneous media do not have a regular periodic microstructure. For such
materials, we must view D as a representative cell that is characteristic of
the medium as a whole. Unless the microstructure has a particular directionality, we might expect a random geometry to lead to isotropic macroscopic
behaviour.
9.4.4 Waves in periodic media
We conclude this section with a short discussion of waves in composite media. Unfortunately, the multiple reections, refractions and diractions that
can occur make the theory much more complicated than in the static cases
considered above. However, it is very similar to the theory of electromagnetic wave propagation through lattices, which underpins much of solid state
physics, so we will illustrate the basic ideas by considering onedimensional
motion of an elastic bar.
The generalisation of (4.2.4) to an inhomogeneous bar is
u
2u
E(x)
= 2.
(9.4.68)
x
x
t
When we seek solutions in the frequency domain and nondimensionalise as
in (9.4.10), we obtain the ordinary dierential equation
du
k2
d
E(x, x/)
+ 2 u = 0,
(9.4.69)
dx
dx
where
)
k = L
,
=
E1
(9.4.70)
405
du1
(x, 0) = 0,
d
u2 (x, 0) = 0,
du2
(0) = 1.
d
(9.4.72)
Using the standard theory of linear ordinary dierential equations (see, for
example, Boyce & DiPrima, 2004, Chapter 3), we easily nd that the Wronskian
du2
du1
u2
(9.4.73)
W = u1
d
d
satises
E()W () const. = 1,
(9.4.74)
(9.4.75)
where A and B are arbitrary constants. The rst key observation is that
u( + 1) also satises (9.4.71), and hence
u1 ( + 1) = Au1 () + Bu2 (),
(9.4.76)
(9.4.77)
where
A = u1 (1),
B=
du1
(1),
d
C = u2 (1),
D=
du2
(1).
d
(9.4.78)
406
(9.4.79)
du2
(1).
d
(9.4.81)
Given the periodic function E(), we can compute T (k) for each value of k by
solving (9.4.71) and (9.4.72) numerically. If T (k) < 2 then the roots (9.4.80)
are complex with unit magnitude, and hence 1,2 are pure imaginary. We
infer that waves can propagate, although they will not in general be periodic,
and the corresponding values of k are called the pass bands. The stop bands,
in which waves cannot be supported, occur where T (k) > 2.
In Figure 9.10 we show the pass and stop bands when the Youngs modulus
takes the form
E() = 1 + sin(2).
(9.4.82)
The shaded region is the stop band in which T  > 2. The parameter
measures the nonuniformity in E: when = 0, we have E 1 and all
waves are transmitted through the medium without attenuation. When
slightly exceeds zero, a narrow stop band appears near k = , and grows as
increases.
We can conrm this behaviour by analysing the problem asymptotically
in the limit 0. As indicated by Figure 9.10, we expect the stop band to
407
k/
(9.4.84)
(9.4.85)
(9.4.86)
408
2
2
dB
+ k1 A + B cos()
2
dx
4
dA
A k1 B sin().
+ 2
dx
4
(9.4.87)
9.5 Poroelasticity
The theory of Section 9.4.3 may be applied, for example, to a solid that
takes the form of a homogeneous elastic matrix in which are embedded many
smallscale voids or pores. For such a material, the elastic constants Cijk
would be constant in the elastic matrix and zero in the voids. The situation
is much more interesting, though, when the pores are suciently connected
for a viscous uid to pass through them, as in a liquidlled sponge, for example. Other important deformable porous media include oil reservoirs and
biological tissue. In principle, one could follow a homogenisation procedure
to derive an averaged macroscopic model governing the deformation of such
a medium. However, the need to couple the Navier equations for the solid
9.5 Poroelasticity
409
matrix to the NavierStokes equations for the uid makes this task considerably more complicated than any of the examples encountered in Section 9.4,
so we will bypass most of the diculties associated with the uid dynamics.
We know from (9.4.67) that we can express the elastic contribution to the
stress in the form
e = Ce u,
(9.5.1)
(9.5.3)
(9.5.4)
u k
P,
t
(9.5.5)
410
(9.5.6)
On the other hand, conservation of mass for the uid leads to the equation
+ 0 v = 0.
(9.5.7)
t
By combining these two equations, we nd that, as also shown in Exercise 9.15,
1 0
( u) ,
(9.5.8)
v =
0 t
which shows how compression of the matrix forces the uid to ow.
For simplicity, we have neglected the compressibility of the solid and uid
phases. We emphasise, though, that the matrix as a whole certainly will be
compressible; for example, it is quite possible to compress a sponge made
out of incompressible rubber. We have also assumed that is always close
to its initial value 0 , which is consistent with our assumption that linear
elasticity is valid, and that the deformation is suciently slow for the inertia
terms in the Navier equations to be ignored. Equations (9.5.2)(9.5.7) thus
comprise the very simplest model that couples elastic deformation to uid
ow.
If we assume that the parameters e , e , k and are all constant, then
a straightforward crossdierentiation reveals that div u satises the linear
diusion equation:
(e + 2e ) 0 k
(div u) =
2 (div u) .
(9.5.9)
t
Hence, the eects of elasticity, viscosity and permeability are all collected in
the eective diusion coecient
(e + 2e ) 0 k
.
(9.5.10)
D=
9.5 Poroelasticity
411
P =0
11111111
00000000
00000000
11111111
00000000
11111111
00000000L
11111111
F
x
u
+ 0 v = q(t),
t
(9.5.12)
where q(t) represents the net volume ux of both solid and uid phases. We
can use (9.5.12) to eliminate v from (9.5.11) and thus nd that u satises a
forced heat equation
2u
u
D 2 = q(t),
t
x
(9.5.13)
x = 0,
u
= v = V
t
at
x = L,
(9.5.14)
noting that, in linear elasticity, the motion of the boundary x = L is neglected. It follows that q = V , and (9.5.13) may then readily be solved by
412
F/(e + 2e )
3
2
1
Pe = 0
2.0
1.5
1.0
0.5
0.2
0.4
0.6
0.8
1.0
V t/L
Fig. 9.12 Dimensionless stress applied to a sponge versus dimensionless time for
dierent values of the Peclet number Pe; the larget limits are shown as dotted
lines.
(L x)(2L x)
2V L2 sin (nx/L) n 2 2 Dt/L 2
t+
+ 3
e
.
6D
D
n3
n=1
(9.5.15)
We can use this to calculate, for example, the net compressive stress F
that we would need to apply to the plunger to force it to move at constant
speed. From (9.5.2)(9.5.4), we nd that
Vx
u(x, t) =
L
u
F = P (e + 2e )
x
x=L
u
= (e + 2e )
,
x x=0
(9.5.16)
and the result of substituting (9.5.15) into this is plotted in Figure 9.12. The
behaviour depends on the key dimensionless parameter known as the Peclet
number Pe = LV /D. When Pe = 0, the eect of the uid is negligible, so
the stress just increases linearly as the elastic matrix contracts. However, for
nonzero Pe, the initial response is dominated by the need to force the uid
through the matrix. After some time, the behaviour becomes linear again,
with
V t Pe
F
e + 2e
L
3
as
t ,
(9.5.17)
as indicated by the dotted lines in Figure 9.12. As the Peclet number increases, the uid has an increasing inuence, and a longer time is taken for
the solution to settle down to (9.5.17).
9.6 Anisotropy
413
9.6 Anisotropy
We conclude the chapter with a brief account of the eects of anisotropy,
which, as we have just seen, is an endemic property of macroscopic models of
heterogeneous elastic media. Moreover, wood and brereinforced materials,
for example, have strongly anisotropic response to stress. In Section 1.7, we
saw that the most general linear relation between elastic stress ij and linear
strain eij is
ij = Cijk ek ,
(9.6.1)
which denes Cijk as a tensor with 81 components. Since there are only six
independent components in the symmetric tensors ij and eij , we could also
write the relation above as
xx
exx
e
yy
yy
zz
ezz
(9.6.2)
= A
,
yz
2eyz
xz
2exz
xy
2exy
where A = (Aij ) is a 6 6 matrix with entries A11 = C1111 , A12 = C1122 ,
A13 = C1133 , . . . and the factor of 2 multiplying the odiagonal strain components accounts for the fact that they appear twice in (9.6.1).
The existence of a strain energy density W such that ij = W/eij
implies, by crossdierentiation, that the matrix (Aij ) is symmetric. We are
thus left with at most 21 independent elastic moduli. In terms of C, this can
be summarised by the relations
Cijk = Cjik = Cijk = Ckij .
(9.6.3)
The only other restriction we can impose in general is that A must be positive denite, to ensure that W is a positive denite function of the strain
components.
Let us now consider a rotation of the coordinate axes by an orthogonal
matrix P . In the rotated frame, (9.6.1) becomes
ek ,
ij = Cijk
where
Cijk
= Pim Pjn Pkr Ps Cmnrs .
(9.6.4)
For the isotropic materials considered in Section 1.7, we asserted that the
relationship between and e is the same as that between and e for
any orthogonal matrix P . Now we can only say that this is true for each
orthogonal matrix P representing a material symmetry. For every such P ,
414
(9.6.5)
Usually, most of these equations are satised trivially, but some of them
may imply nontrivial relationships between the elastic constants Cijk .
The simplest symmetry that we can encounter is reectional symmetry
with respect to the (x, y)plane. Such materials are called homoclinic. To x
ideas, we will focus on the example of a material containing bres oriented
in the zdirection. In this case, (9.6.5) should be satised with
1 0 0
P = 0 1 0 ,
(9.6.6)
0 0 1
or, in a more compact notation, Pij = (1)i 3 ij . We must therefore have
Cijk = (1)i 3 +j 3 +k 3 + 3 Cijk , and hence Cijk vanishes if the index 3 appears an odd number of times. This leaves 13 nonzero entries in Aij , namely
0
0 A16
A11 A12 A13
A
0
0 A26
12 A22 A23
0
0 A36
A13 A23 A33
(9.6.7)
A=
.
0
0 A44 A45
0
0
0
0
0
0 A45 A55
0
0 A66
A16 A26 A36
Continuing with our example of zoriented bres, let us further assume
that they are positioned in the (x, y)plane regularly enough to impose in
addition reectional symmetry with respect to the (x, z) and (y, z)planes.
Such a material with three orthogonal planes of symmetry is called orthotropic. In this case (9.6.5) must hold simultaneously with the three transformations Pijz = (1)i 3 ij , Pijx = (1)i 1 ij and Pijy = (1)i 2 ij . Hence,
following the same reasoning as in the monoclinic case, Cijk vanishes whenever any of the indices 1, 2 or 3 appears an odd number of times. In particular, C1112 = C2212 = C3312 = C2313 = 0, so that (9.6.7) reduces to
0
0
0
A11 A12 A13
A
0
0
0
12 A22 A23
0
0
0
A13 A23 A33
A=
(9.6.8)
.
0
0
0
0 A44
0
0
0
0
0
0 A55
0
0
0
0
0 A66
9.6 Anisotropy
415
If, on the other hand, the bres are randomly positioned in the xyplane,
then, on a macroscopic scale, the material is transversely isotropic. The
resulting matrix (Aij ) will clearly be a special case of (9.6.8) and we rst
note that, in the absence of preferred direction in the (x, y)plane, A11 = A22 ,
A23 = A13 and A44 = A55 . Furthermore if we restrict the displacements to
be in the (x, y)plane only, then we have an isotropic problem in plane strain.
Thus, as in (1.7.6), there must exist two constants c1 and c2 such that
xx = c1 (exx + eyy ) + c2 exx ,
(9.6.9a)
xy = c2 exy ,
(9.6.9b)
(9.6.9c)
0
0
0
A11 A12 A13
A
0
0
0
12 A11 A13
0
0
0
A13 A13 A33
A=
(9.6.11)
.
0
0 A44
0
0
0
0
0
0 A44
0
0
0
0
0
0 (A11 A12 )/ 2
Note that the same conclusion could be reached by a blind use of (9.6.5)
with the family of orthogonal transformations
cos sin 0
B = sin cos 0 ,
(9.6.12)
0
0
1
but with considerably more eort.
As a simple example, let us generalise the analysis of Section 2.4 to model
the torsion of an orthotropic bar. For a displacement eld of the form (2.4.1),
the only nonzero strain components are exz and eyz , and, according to
(9.6.8), the nonzero stress components are
y ,
yz = A44
+x .
(9.6.13)
xz = A55
x
y
Hence, instead of Laplaces equation, now satises
A55
2
2
+
A
= 0,
44
x2
y 2
(9.6.14)
416
where A44 and A55 are both positive, and the nostress boundary condition
is
dy
dx
y
A44
+x
= 0,
(9.6.15)
A55
x
ds
y
ds
on the boundary of the bar. We recall from Section 2.4 that, once we have
solved for (x, y), we can evaluate the torsional rigidity using
1
(xyz yxz ) dxdy
R=
D
+ x A55 y
y
dxdy.
(9.6.16)
A44 x
=
y
x
D
The most striking implication of this model is that the elastic constants
A44 and A55 can be eliminated from the problem by rescaling x and y appropriately. If we dene
A44 1/4
A55 1/4
X,
y=
Y,
(9.6.17)
x=
A44
A55
then, with respect to the new variables, satises
2
2
+
= 0,
X 2
Y 2
(9.6.18)
dX
1 d 2
dY
=
X +Y2
ds X
ds Y
2 ds
(9.6.19)
subject to
Y
+ X +Y2
dxdy.
X
R = A44 A55
Y
X
D
(9.6.20)
ij
2 ui
2 um
=
=
C
.
ijmn
t2
xj
xj xn
(9.6.21)
417
(9.6.22)
and, if nontrivial waves are to exist, they must therefore satisfy the dispersion relation
det Cijmn kj kn 2 im = 0.
(9.6.23)
This generalises the dispersion relations (3.2.37) governing P waves and S waves in an isotropic solid.
One of the main consequences of (9.6.23) is that the waves are dispersive,
so that the phase speed / k of plane waves generally depends on the wave
vector k. This is in contrast with waves in an innite isotropic solid, which
propagate with phase speed equal to either cp or cs . Recall, though, that
the presence of boundaries can give rise to dispersion even in an isotropic
medium; for example, Love waves and waves in beams are both dispersive.
Exercises
9.1
Show that the constitutive law for the linear Jereys viscoelastic
element shown in Figure 9.13 is
dT
du
trT
+ T = k tru
+u .
(E9.1)
dt
dt
Find expressions for the modulus k and the relaxation times trT and
tru , in terms of the spring constants and impedance of the individual
components, and hence show that tru trT .
418
Tn1
Tn
Tn
un1
un
m
un+1
Fig. 9.14 A system of masses connected by springs and dashpots in parallel along
the xaxis.
9.2
Obtain the Voigt and Maxwell creep functions (9.2.9) by taking the
limits trT 0 and tru respectively.
(a) Suppose parallel spring/dashpot elements separate a linear
array of masses m, each displaced a distance un (t) from its
equilibrium position nL, as shown in Figure 9.14.
Show that
Tn = Tne + Tnv ,
dun
dun+1
e
v
,
Tn = k(un+1 un ),
Tn = Y
dt
dt
in the notation of Section 9.2.2. Use the equation of motion
for the nth mass to show that
m
d2 un
d
= Y (un+1 2un + un1 )
2
dt
dt
+ k (un+1 2un + un1 )
Exercises
Tn1
Tn1
419
Tn
Tn
un1
un
un+1
Fig. 9.15 A system of masses connected by springs and dashpots in series along the
xaxis.
9.3
(b) Repeat the above calculation for the series elements shown in
Figure 9.15 and show that the result is a discretisation of the
onedimensional Maxwell model (9.2.19).
A standard rheological test
an oscillatory displacement
is to impose
on a sample, say u = Re aeit , and measure the resulting force
T (t). For a linear material, T (t) will also be proportional to eit ,
and we dene the complex modulus G = T /u. The complex modulus
is then decomposed into its real and imaginary parts: G = G + iG ,
where G is called the storage modulus and G the loss modulus.
Show that the average work done in each period is
2/
1
du
dt = a2 G .
T
2 0
dt
2
For the Jereys model (E9.1), show that
G =
9.4
ktru
k (tru trT )
,
trT
trT 1 + t2rT 2
k (tru trT )
.
trT 1 + t2rT 2
Deduce that the rate at which the Jereys element dissipates energy
is a nonnegative increasing function of .
Recall from Chapter 5 the denitions of deformation gradient and of
the Green and Finger deformation tensors, namely
Fij =
xi
,
Xj
C = F T F,
G =
B = F F T.
DB
= v T B + Bv,
Dt
420
Now consider a vector eld f (x, t) that convects and rotates with
the ow. Following the procedure in Exercise 8.11, show that f must
satisfy
Df
= (f ) v.
Dt
Finally, suppose that the tensor eld A(x, t) is such that, if Af = g
at time t, then Af = g for all t, whenever f (x, t) and g(x, t) are
both convected by the ow in the above sense. Deduce that A must
satisfy
DA
+ Av T v T A = 0.
Dt
9.6
9.7
For the strain energy density calculated in Exercise 9.6, show that
W
= kk .
T
3
We can use (9.3.2) to estimate the stress induced by a temperature
variation T as kk = O (ET ), where E is Youngs modulus.
Deduce that W/T is negligible in comparison with c for temperature variations T such that E2 T /c 1.
[Steel, for example, has 7800 kg m3 , c 500 J kg1 K1 ,
E 200 GPa and 13 106 K1 , so that T must be much less
than 105 K for the estimate to be valid.]
Exercises
9.8
421
,
c2 t2
x2
3 x
t
x2
where c = E/. Suppose the bar starts from rest at zero stress
with a temperature prole T (x, 0) = T0 H(x), where H is the Heaviside function given by (3.5.7). Show that, for x ct
/c, thermal
diusion is negligible and u approximately satises
1 2u
2u 1
=
T0 (x),
c2 t2
x2
3
where is the Dirac deltafunction. Obtain the solution
u(x, t) =
T0
2x x ct x + ct ,
12
and deduce that the maximum tensile stress caused in the bar is
given by ET0 /6.
9.9
1 2u
2 u T
,
=
2 t2
x2
x
governing this situation, where = Lc/. Also derive the initial and
boundary conditions
u
= 0,
x
T
= 0,
u = 0,
x
u
= 0,
T = 1, u =
t
T = 0,
x = 0, t > 0,
x = 1, t > 0,
t = 0, 0 < x < 1,
422
n + 1/2
n=0
cos (n + 1/2)x
2
2 2
u= 2
e(n+1/2) t
2
2
2
2
(n + 1/2) ( + (n + 1/2) )
n=0
+ cos (n + 1/2)t (n + 1/2) sin (n + 1/2)t .
9.10
9.11
Consider the solutions of (9.4.29) and (9.4.30) with periodic boundary conditions on a rectangular cell in which the modulus is piecewise constant, as shown in Figure 9.7. If no symmetry is assumed,
show that the function (1) dened by (9.4.37) satises the boundary
conditions
(1) (1, ) (1) (0, ) = 1,
(1) (, b) (1) (, 0) = 0,
(1)
(1)
(1, )
(0, ) = 0,
(1)
(1)
(, b)
(, 0) = 0,
Exercises
423
%11
%21
%12
%22
b2 1 2
=
=
=
=
.
11
12
21
22
11 22 21 12
9.12
u1k
Cijk
=
.
i
i x
Deduce that the solution may be written as
u1i = Aijk
u0j
,
xk
Amn
Cijk
=
.
j
k
i
Verify that this gives us 18 independent equations for the 18 independent functions Aijk (x, ).
By substituting for u1 in (9.4.61), show that 0 is given by
u0m
Akmn
+ Cijmn
.
0ij = Cijk
xn
9.13
9.14
0 < 1/4,
1,
E() = E2 , 1/4 < 3/4,
1,
3/4 < 1.
424
k/
E2
Fig. 9.16 Dimensionless wavenumber k versus Youngs modulus contrast E2 for a
piecewise uniform bar. The stop bands are shaded.
[Hint: both u and Edu/d must be continuous.] Show that the trace T
of themonodromy matrix is given by
1 + E2
k
k
k
k
.
cos
sin
sin
T = 2 cos
2
2
2 E2
E2
2 E2
The stop bands where T  > 2 are shown as shaded regions in Figure 9.16. Show that, when E2 is close to 1, the boundaries of these
bands are given asymptotically by
(2n + 1) 1
k (2n + 1) +
(E2 1) + ,
4
2
n n
n
(E2 1) +
(E2 1)2 + ,
or k 2n +
2
4
8
where n is an integer.
9.15
1 0
(1 0 ) (1 u) .
1+u
Exercises
425
(f ) + (f v) = 0.
t
Show that this reduces to (9.5.7) when the uid density is
constant and 0 .
Epilogue