AAE 553 Text PDF

. .
Theory of Elasticity
a rst course on
fundamental principles
and methods of analysis
James F. Doyle
and
C-T. Sun
Purdue University
jf
d
. .
iv
These notes were prepared for use in the course AAE553: Elasticity in Aerospace Ap-
plications at Purdue University. All notice of errors, corrections, or recommendations
for improvement, will be gracefully received by the authors.
The manuscript itself was prepared using a combination of LATEX and PostScript.
Copyright(c) 1986{2003, 2005, 2006, 2007, 2008

Contents
Table of Contents iv
Notation vii
1 Cartesian Tensors 1
1.1 Indicial Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Coordinate Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Scalar, Vector and Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Properties of Second Order Tensors . . . . . . . . . . . . . . . . . . . . . . 8
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Deformation 15
2.1 General Description of Deformations . . . . . . . . . . . . . . . . . . . . . 15
2.2 Deformation of Lines, Areas, and Volumes . . . . . . . . . . . . . . . . . . 20
2.3 Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5 Deformation in Terms of Displacement . . . . . . . . . . . . . . . . . . . . 37
2.6 Special Deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3 Stress 51
3.1 Cauchy Stress Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Properties of the Cauchy Stress Tensor . . . . . . . . . . . . . . . . . . . . 58
3.3 Stresses Referred to the Undeformed State . . . . . . . . . . . . . . . . . . 62
3.4 Stress in Special Deformations . . . . . . . . . . . . . . . . . . . . . . . . . 67
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4 Elastic Materials 81
4.1 Work and Strain Energy Concepts . . . . . . . . . . . . . . . . . . . . . . 81
4.2 Elastic Constitutive Relations . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3 Finite Strain Isotropic Elastic Materials . . . . . . . . . . . . . . . . . . . 93
4.4 Elastic Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
v
vi Contents
5 Linear Elasticity Problems 111
5.1 Reduction to the Linear Theory of Elasticity . . . . . . . . . . . . . . . . . 111
5.2 Plane Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.3 Airy Stress Function Formulation . . . . . . . . . . . . . . . . . . . . . . . 132
5.4 A Guide to Selecting Stress Functions . . . . . . . . . . . . . . . . . . . . 143
5.5 Applications of Complex Variables . . . . . . . . . . . . . . . . . . . . . . 155
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6 Nonlinear Elasticity Problems 175
6.1 Discretizing Continuous Media . . . . . . . . . . . . . . . . . . . . . . . . 175
6.2 Equilibrium of Discretized Systems . . . . . . . . . . . . . . . . . . . . . . 184
6.3 Stiness Properties of Discretized Systems . . . . . . . . . . . . . . . . . . 195
6.4 Total Lagrangian Incremental Formulation . . . . . . . . . . . . . . . . . . 200
6.5 Stability of Discrete Systems . . . . . . . . . . . . . . . . . . . . . . . . . 213
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
7 Variational Methods 231
7.1 Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
7.2 Direct Methods of Solution . . . . . . . . . . . . . . . . . . . . . . . . . . 238
7.3 Semi-Direct Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
References 273
Index 275
vii
Notation
a radius
A cross-sectional area
b thickness, depth, radius
^b bi body forces per unit mass
B bulk modulus
D plate stiness, Eh3 =12(1 ; 2 )
eî unit or base vectors
eij Eulerian strain tensor
E E^ Young's modulus, viscoelastic modulus
EI beam exural stiness
Eij Lagrangian strain tensor
G shear modulus
h beam or rod height, plate thickness
hi (r st) interpolation functions
I second moment of area, I = bh3 =12 for rectangle
I1 I2 I3 invariants of strain
J J general functional
Jo J Jacobians
Je Jacobian of the element transformation
K stiness, thermal conduction
k ] K] stiness matrix
L length of element, distance to boundary
M Mx moment
n^ ni direction vector
P (t) P^ applied force history
q distributed load, heat ux
r radial coordinate
R radius
S So arc length
t time
t^ ti tractions
T time window, temperature
u(t) response velocity, strain, etc.
u v w displacements
U U strain energy, strain energy density
V potential of loads
cW work
xo yo z o original rectilinear coordinates
x y z deformed rectilinear coordinates
Greek letters:
coecient of thermal expansion
ij matrix of direction cosines
ijk permutation symbol
. August 2007
viii Notation
small quantity, variation
ij Kronecker delta
determinant, increment
viscosity, damping
plate curvature
material property: (3 ; 4 ) or (3 ; )=(1 + )
angular coordinate
Poisson's ratio

Shear modulus
Lamê constant, eigenvalue
o mass density
stress, strain
space transform variable
x y rotation
Hz Helmholtz functions
lateral contraction
Special Symbols:
r2 dierential operator, @x@ 22 + @y@ 22
vector cross-product
] square matrix
f g vector
Subscripts:
comma, partial dierentiation
Superscripts:
K Kirchho stress
o original conguration
complex conjugate
bar, local coordinates
_ dot, time derivative
0 prime, derivative with respect to argument
Chapter 1
Cartesian Tensors
The theory of elasticity deals with quantities known as tensors. The purpose of
this chapter is to introduce the basic concepts of tensor analysis and the associate
notations. The reward of the eort is that concepts and derivations (especially in
the three-dimensional theory) are handled conveniently and neatly. Since theorems
proven in one coordinate system are proven for all coordinate systems, then it is
sucient (and convenient) to deal only with Cartesian (or rectilinear) coordinate
systems.
1.1 Indicial Notation

In the theory of elasticity, we deal with groups of things such as u v w represent-
ing displacements, x y z representing coordinates, or xx yy xy representing
stresses. We would like a notation that handles such groups conveniently we will use
the subscript or indicial notation to achieve this.
Let the symbol xi with the range i = 1 n be used to denote any one of the
variables in the set fx1 x2 xng. The symbol i is called an index. Similar notations
with multiple indices such as tij i j = 1 n are also used to represent individual
components in the set of n n] elements ft11 t12 tnng. We will restrict ourselves
to having the subscript range always go from 1 to 3.
Summation Convention
Consider the equation
x1 y1 + x2 y2 + x3 y3 = a
which can be interpreted as the scalar product of two vectors whose components are
x^ = x1 x2 x3 and y^ = y1 y2 y3 in the 3-dimensional space. This equation can
f g f g
be written as
X
3
xi yi = a
i=1
1
2 Chapter 1. Cartesian Tensors
With the aid of the summation convention, the equation above is written in the
simple, shorthand, form
xi yi = a
The summation convention states that the repetition of an index in a product term
denotes a summation with respect to that index over its range. The repeated index is
called a dummy index as opposed to one that is not summed, which is referred to as
a free index. Since a dummy index indicates a summation operation, any index can
be used without changing the result. For example, xi yi and xj yj have the identical
meaning.
When there are more than two summation operations to be performed, caution
must be exercised in the use of the summation convention. The following rules can
therefore help:
If a subscript occurs twice in a term in an equation, then it must be summed
over its range. These are the repeated or dummy indices, e.g.,
Cikk = Ci11 + Ci22 + Ci33
If a subscript occurs once in a term then it must occur once in every other term
in the equation. These are the free or live indices, e.g.,
F1 = ma1
Fi = mai ) F2 = ma2
F3 = ma3
If a subscript occurs more than twice in a term, then it is a mistake, e.g.,
Aiij Bij
An example of a proper equation is
ij = Sij + HijklEkl
where the free indices are i and j and the dummy indices are k and l.
A shared repeated index is a contraction. Contraction is a special summing op-
eration performed on quantities with indices. It is done by equating two indices and
summing over the range of that index. For example,
Aijkl (contraction over i) Aiikl
After contraction, the original two free indices become a pair of dummy indices and
the summation convention applies.
When contraction is applied to a pair of indices of a tensor of order n, a new
tensor results. The order of this new tensor is n ; 2.
Sometimes the summation convention can P be ambiguous in those cases we will
make the summation explicit by using the sign.
1.1. Indicial Notation 3
Special Symbols
A number of special symbols have been introduced as a convenience in using the tensor
notation. Two especially useful symbols are the Kronecker delta and the Permutation
symbol. Some of their properties are summarized here.
A base vector is a unit vector parallel to a coordinate axis. Let e^1 e^2 e^3 be the base
vectors for the (x1 x2 x3 ) coordinate system then, a typical vector can be written as
V^ = V1 e^1 + V2e^2 + V3e^3 = Vieî
The Vi are referred to as the components of the vector V^ .
The Kronecker delta is denoted by ij and is dened as follows
( 2 3
1 0 0
ij = 10 if ii =
=j
j or ij ] = 64 0 1 0 75 = d I c
6
0 0 1
Some of its properties are:
Symmetrical ps = sp
Trace pp = 11 + 22 + 33 = 3
Contraction Aik ij = A1k 1j + A2k 2j + A3k 3j = Ajk
Dot Product eî e^j = ij e.g., e^1 e^2 = 0 e^3 e^3 = 1
Note that written in matrix form, it is the same as the identity matrix.
The permutation symbol is dened as:
8
< 1 ijk form an even permutation of 123, e.g., 312
>
ijk = > ;1 ijk form an odd permutation of 123, e.g., 321
: 0 otherwise, e.g., 122
Some of its properties are:
Anti-symmetric ijk = ;ikj = kij
Trace iik = 11k + 22k + 33k = 0
Contraction ijspqs = ipjq iq jp
;
Determinant detaij ] = ijk a1i a2j a3k

Cross Product eî e^j = ijk e^k
e.g., e^2 e^3 = 23k e^k = e^1
. August 2007
Note that there is no simple representation for it using matrix notation since it is of
size 3 3 3].
Example 1.1: Use the tensor notation to perform both the dot product and cross
product of two vectors A^ and B^ .
This is accomplished by expanding each vector in terms of its components plus
unit vectors and realizing that it is only the unit vectors that participate in the
vector operations. That is,
A^ B^ = (Ai eî) (Bj e^j ) = Ai Bj (êi e^j ) = Ai Bj ij = Ai Bi
= A1 B1 + A2 B2 + A3 B3
^ ^
A B = (Ai eî) (Bj e^j ) = Ai Bj (êi e^j ) = Ai Bj ijk e^k
= (A2 B3 ; A3 B2 )ê1 + (A3 B1 ; A1 B3 )ê2 + (A1 B2 ; A2 B1 )ê3
It is thus clear from this approach which vector operations result in vectors and
which result in scalars.
1.2 Coordinate Transformation

The concept of component is fundamental to tensor analysis. A given tensor quantity
will have dierent components in dierent coordinate systems, thus we wish to know
how they change under a coordinate transformations.
Consider two Cartesian coordinates systems (x1 x2 x3) and (x1 x2 x3) as shown 0 0 0
in Figure 1.1. Let e^1 e^2 e^3 be the base vectors for the (x1 x2 x3) coordinate system,
and e^1 e^2 e^3 the base vectors for the (x1 x2 x3 ) system as also shown in the gure.
0 0 0 0 0 0
x2
x2
0
B
B
B 3

V^
B
B6e^2 x1
0
B - x1
e^3 ;
; e^1
;
;
x3 ;; x03
;
Figure 1.1: Base vectors and rotated coordinate system.
Since the coordinate axes are mutually orthogonal, we have
eî e^j = ij eî e^j = ij
0 0
A vector V^ can be projected onto the two coordinate systems with the result:
V^ = Vj e^j = Vj e^j0 0
1.2. Coordinate Transformation 5
Taking the scalar product of this equation with eî, we obtain 0
Vj e^j eî = Vj e^j eî

0 0 0 0
which yields, on using the properties of the Kronecker delta,

Vj ij = Vj e^j eî or Vi = Vj e^j eî = Vj eî e^j
0 0 0 0 0
This can be further rewritten by introducing the matrix of direction cosines

ij eî e^j 0
(1.1)
and substituting to get
Vi = ij Vj0
This gives the relation for transformation of components in one coordinate system
into components in another.
A similar procedure (but taking a scalar product with eî ) leads to the inverse
relation
Vi = jiVj 0
These two equations taken together yields the circular result

Vi = ij kj Vk
0 0
from which we conclude that

ij kj = ik
This can be written in matrix form as
] ]T = d I c or ]T ] = d I c
Thus, ij are orthogonal and the relation is known as the orthogonality relation. Also,
it is easy to see that det ] = 1.
Example 1.2: Show that if two directions of an orthogonal triad are known, then
the third direction can be found by using the orthogonality conditions.
Let the given vectors be
e^1 = 12 e^1 ; 21 e^2 + 12 e^3
0
p
e^2 = 12 e^1 ; 21 e^2 ; 12 e^3

0
p
and it is desired to obtain e^ . The given vectors are orthogonal since e^1 e^2 =
0 0 0
1 1 1
4 + 4 ; 2 = 0. We can obtain the third vector from knowledge that
0
e^3 = e^1 e^2 = 12 e^1 + 12 e^2 + 0ê3
0 0
p p
The rst row of the direction cosines is 1j = e^1 e^j = e^1j so that the direction 0 0
cosines are easily established as

2 1 3
6 2 ; 12 12 7 p
ij ] = 64 1
2 ; 12 ; 12 75 p
0 1 1
2 2
p p
That is, each row consists of the components of the primed unit vectors.
. August 2007
1.3 Scalar, Vector and Tensor Fields
A quantity is a eld quantity when it is a function of position. For example =
(x1 x2 x3 t). Let fx1 x2 x3 g and fx1 x2 x3g be two xed sets of rectangular Carte-
0 0 0
sian coordinates related by the transformation law

xi = ij xj
0
where ij are the direction cosines. A system of quantities is called by dierent names
depending on how the components of the system are dened in the variables x1 x2 x3
and how they are transformed when the variables x1 x2 x3 are changed to x1 x2 x3 . 0 0 0
A system is called a scalar if it has only a single component in the variables xi

and a single component in the variables xi and if and are numerically equal
0 0 0
at the corresponding points. That is,

(x1 x2 x3 ) = (x1 x2 x3 ) 0 0 0 0
A system is called a vector eld or a tensor eld of order one if it has three components
vi in the variables xi and the components are related by the transformation law
0
v^ = vi eî = vk e^ k = vikie^k
0 0 0
Dropping the unit vectors and rearranging the subscripts, these become
vi = ik vk
0
or f v = ] v
0
g f g
The tensor eld of order two is a system which has nine components tij in the variables
x1 x2 x3 , and nine components tij in the variables x1 x2 x3 , and the components are
0 0 0 0
related by the transformation law

tij = imjntmn
0
or t ] = ] t ] ]T
0
The pattern is, each order of tensor requires n3 components and n ] matrices in
the transformation.
We obtain from a generalization to a tensor of order n:
tijk
0

= imjnkp tmnp
tijk = minj pk tmnp0

(1.2)
If all components of a tensor vanish in one coordinate system, then they vanish in all
other coordinate systems.
A second order tensor tij is often represented as a square matrix tij ]. In this
form matrix algebra applies. Higher order tensors, however, do not have a simple 2-D
matrix representation.
1.3. Scalar, Vector and Tensor Fields 7
Isotropic Tensors
An interesting type of tensor frequently occurs in the study of elasticity, namely, that
of an isotropic tensor. An isotropic tensor is one whose components remain unchanged
under a coordinate transformation. For a second order isotropic tensor, for example,
tij = tij = ik jl tkl
0
A collection of particular isotropic tensors of dierent order is:

order isotropic tensor
0 1
1
2 ij
3 ijk
4 ij pq ipjq pj iq
5 ijk pq and permutations
6 ij pq rs ipjq rs pj iq rs iprsjs and permutations
It is noted that there cannot be a vector (tensor of order one) that is isotropic.
Calculus of Tensor Fields

A tensor eld occurs when the tensor is a function of position. Examples of scalar,
vector and second order tensor elds are, respectively,
= (x1 x2 x3 t)
vi = Vi(x1 x2 x3 t)
tij = Tij (x1 x2 x3 t)
Since these are continuous functions of position then they are amenable to calculus
operations such as dierentiation and integration.
If xi = ij xj then for a vector vj we have
0
vj (^x ) = jk vk (^x)

0 0
Dierentiating both sides of the equation, we obtain

@vj0
@vk = @vk @xm = @vk
@xi0
= jk
@xi jk @xm @xi
0
jk im
0
@xm
This says that partial derivatives of any tensor eld behave like the components of a
Cartesian tensor. (It should be noted that this is not true in curvilinear coordinate
systems.) From this it is apparent that the term
@tij
@xk
. August 2007
is a third order tensor. In general dierentiation with respect to x increases the order
of a term by 1.
Consider a tensor eld Tjkm (x) in a volume V bounded by a surface S . Then,

the Integral theorem states

Z Z
ni Tjkm dS = @x@ Tjkm dV (1.3)
S V

i
where ni are components of the unit vector n^ along the exterior normal of S . Special
cases of this are
Z Z @
ni dS = dV
S V @xi
Z Z @vi
nivi dS = dV
S V @xi
Z Z @vk dV
ijk nj vk dS = ijk @x (1.4)
S V j
The second of these is usually referred to as Gauss's theorem.
Consider the case when the surface integral is of the product of a scalar and vector
vi, then Z Z Z @vi @ i
ni vi dS = vini dS = @x + vi dV
S S V i @xi
Furthermore, let the vector be represented as the gradient of a scalar vi = @=@xi so
that
@vi = @ 2 + @ 2 + @ 2 = @ 2 = r2
@xi @x21 @x22 @x23 @xi @xi
then the integral theorem becomes
Z @ Z @ @ i dV
@x ni dS = r2 + @x
S i V i @xi
Now interchange the role of and and subtract the two integral relations to get
Z h @ @ n i dS = Z r2 ; r2i dV
@x ni ; @x i (1.5)
S i i V
This is usually referred to as Green's theorem and we will nd a very useful application
of it in Chapter 7.
1.4 Properties of Second Order Tensors

Since second order tensors are so prevalent in the theory of elasticity (they are used
to represent stress and strain, for example), it is of value now to summarize some of
their major properties. Because second order tensors can be represented conveniently
by matrices, many of the following results can also be established simply from matrix
theory.
1.4. Properties of Second Order Tensors 9
Symmetry and Antisymmetry
A tensor Sij is symmetric if
Sji = Sij or S ]T = S ]
A tensor Aij is antisymmetric if
Aji = Aij; or A ]T = ; A ]
It follows that an anti-symmetric tensor must have zeros on the diagonal.
Any second order tensor can be decomposed into the sum of a symmetric and an
antisymmetric part by
Bij = 21 Bij + 12 Bji + 12 Bij 12 Bji
;
= Sij + Aij
Note that the contraction of a symmetric and antisymmetric tensor is zero. That is,
Sij Aij = 0.
Principal Values of Symmetric Tensors

Consider the relation
Vi = Tij nj Vi = vector ni = unit vector
and let Tij be a symmetric tensor. That is, a vector Vi is produced by contracting the
unit vector with the second order symmetric tensor Tij . We will give an interesting
geometrical interpretation of this relation that will aid in the understanding of the
properties of stress and strain.
^ :

:
V Vi = Tij nj n^
Figure 1.2: The ellipsoid associated with the transformation of second order tensors.
As the vector n^ traces a sphere, the vector V^ traces an ellipsoid.
Consider dierent initial vectors n^ each of the same size but dierent orientations
they will correspond to dierent vectors V^ . Figure 1.2 shows a collection of such
vectors where n^ traces out the coordinate circles of a sphere. Note that V^ traces an
ellipse but not necessarily in the coordinate planes many such traces would form an
. August 2007
ellipsoid. That is, the sphere traced by n^ is transformed into an ellipsoid traced by V^
and the principal axes of the ellipsoid do not necessarily coincide with the coordinate
directions.
In general, the vectors n^ and V^ are not parallel. An interesting question to ask,
however, is: Are there any initial vectors that have the same orientation before and
after transformation? The answer will give an insight into the properties of all second
order tensors, not just strain but also stress and moments of inertia to name two
more.
Assume that there is an n^ such that it is parallel to V^ , that is,
Vi = ni = Tij nj or Tij ; ij ] nj = 0
where is some scalar multiplier. This is given in expanded matrix form as
2 38 9
T11 ; T12 T13 > n >
64 T12 T22 ; T23 75 < n12 = = 0
T13 T23 T33 ; > : n3 >
This has a non-trivial solution for ni only if the determinant of the coecient matrix
vanishes. Expanding the determinantal equation we obtain the characteristic equation
3 ; I1 2 + I2 ; I3 = 0 (1.6)
where the invariants I1 I2 I3 are dened as
I1 = Tii = T11 + T22 + T33
I2 = 21 Tii Tjj ; Tij Tji] = T11 T22 + T22 T33 + T33 T11 ; T122 ; T232 ; T132
I3 = detTij ] = T11 T22 T33 + 2T12T23 T13 ; T11 T232 ; T22 T132 ; T33 T122 (1.7)
The characteristic equation yields three roots or possible values for
(1) (2) (3)
These are called eigenvalues or principal values. For each principal value, there is a
corresponding solution for vector n^. The three n^ 's
n^(1) n^ (2) n^(3)
are called eigenvectors or principal directions. The principal values and directions are
the same as those of the principal axes of the ellipsoid in Figure 1.2.
The principal values can be computed from the invariants 25] by rst dening
q
Q = 91 I12 ; 3I2] R = 541 ; 2I13 + 9I1 I2 ; 27I3]
= cos 1 R= Q3 ]
;
then
q q
1 = 31 I1 ; 2 Q cos ( 13
) 2 3 = 31 I1 ; 2 Q cos ( 13 (
2)) (1.8)
1.4. Properties of Second Order Tensors 11
For two-dimensional problems where T13 = 0 and T23 = 0, this simplies to
q
1 2 = 21 T11 + T22 ] 21 T11 ; T22 ]2 + 4T122 3 = T33 (1.9)
This coincides with the construction known as Mohr's circle.
Example 1.3: Prove that the principal directions are orthogonal.

Consider two particular directions (designated a and b), then they satisfy
Tij naj = a nai Tij nbj = b nbi
Contract the rst with nbi and the second with nai to give the scalar relations
Tij nbinaj = a nbinai Tij nainbj = b nai nbi
Subtract these two and noting that because of the symmetry of Tij , the left hand
terms are equal and cancel, to then give
0 = (a ; b )nai nbi
We conclude that as long as the two principal values are dierent then
nainbi = 0 or n^ a n^ b = 0
which proves the orthogonality.
Example 1.4: Transform the components of Tij to the coordinate system dened
by the principal directions.
Let the original coordinate system be dened by the base vectors
e^1 = f1 0 0g e^2 = f0 1 0g e^3 = f0 0 1g
and the transformed system by
e^1 = n^ (1)
0
e^2 = n^ (2)
0
e^3 = n^ (3)
0
Then the transformation matrix is given by

ij = eî e^j = n^ (i) e^j = n(ki) e^k e^j = n(ji)
0
We note that since n(ik) are the direction cosines of an orthogonal triad then
n(ik) = ki e.g., n(2)
i = f21 22 23 g
From the eigenvalue problem, we have
Tij n(jk) = (k) n(ik)
. August 2007
(there is no sum on k). Use of the direction cosines leads to
Tij kj = (k) ki
Multiply both sides by li and recognizing the left hand side as the transform of Tij
and the right hand side as the Kronecker delta gives
T lk = (k) kl
0
or in expanded matrix form

2 T T T 3 2 (1) 0 0 3
4 11 T12 13
0 0 0
6 7
22 23 5 = 4 0 (2) 0 5
T
0 0
Sym T33 0
0 0 (3)
That is, with respect to the new coordinate system, Tij has a diagonal form.
We can show that (i) 's are the maximums and minimums of the associated
ellipsoid surface.
Cayley-Hamilton Theorem
An interesting aspect of symmetric second order tensors is that higher order products
of it can always be written in terms of the invariants. This is intimately related to
the principal values.
For example, the principal values of Tij ] are obtained from the characteristic
equation for the eigenvalue problem. The Cayley-Hamilton theorem states that the
matrix Tij ] satises the same characteristic equation, that is,
Tij ]3 ; I1 Tij ]2 + I2 Tij ]1 ; I3 ij ] = 0
From this, we have
Tij ]3 = I1 Tij ]2 ; I2 Tij ]1 + I3 ij ] (1.10)
allowing the next higher order power to be written as
Tij ]4 = I1 Tij ]3 ; I2 Tij ]2 + I3 Tij ]1
But the third power term is already known, hence this can be reduced to
Tij ]4 = (I12 ; I2) Tij ]2 + (I3 ; I1I2 ) Tij ]1 + I1 I2 ij ]
In this way, it can be easily shown that terms with higher powers can be expressed
in terms of Tij ]2 , Tij ]1 , and ij ], and the invariants.
Exercises 13
Exercises
1.1 Write out explicitly the components of
ij = Sij + Hijpq Epq
1.2 Show by expansion
(a) A^ (B^ C^ ] = (A^ C^ )B^ ; (A^ B^ )C^
(b) r^ (r^ V^ ) = 0
(c) ijk kpq = ip jq ; iq jp
1.3 Prove, by expansion into component form, that
r^ (A^ B^ ) = A^(r^ B^ ) ; B^ (r^ A^) + (B^ r^ )A^ ; (A^ r^ )B^
1.4 Prove that the principal values of a symmetric second order tensor are extremal
values.
1.5 Show that if eî = ik e^k then eî = kie^k , when eî and eî are orthogonal triads.
0 0 0
1.6 Show that det ij ] = 1.

1.7 Find the components of the second order tensor
21 2 03
Tij ] = 4 2 3 0 5
0 0 4
with respect to axes rotated about the 3-direction by an angle . For a number
of angles, evaluate the invariants I1 I2 I3 .
1.8 Find the components of the second order tensor
21 2 03
Tij ] = 4 2 3 0 5
0 0 4
with respect to a set of axes dened by
e^1 = 21 e^1 ; 12 e^2 +
0 1 e^3
p
2 e^2 = 21 e^1 ; 12 e^2 ;
0 1 e^3
p
2 e^3 =
0 1 e^1 + 1 e^2
p
2
p
2
1.9 A second order tensor is given as
2 21 15 12 3
Tij ] = 4 15 ;6 15 5
12 15 ;6
What are its principal values and their directions?
. August 2007
1.10 The principal directions for a tensor are
e^1 = 12 e^1 ; 12 e^2 + 12 e^3
0
p e^2 = 12 e^1 ; 21 e^2 ; 12 e^3
0
p e^3 =
0 1 e^1 + 1 e^2
p
2
p
2
If the principal values are f1 2 3g, what are the component values of the
tensor?
1.11 If Sij ] is symmetric and Aij ] is anti-symmetric, show that Sij Aij = 0.
Chapter 2
Deformation
This chapter considers how lengths, areas and volumes change during a deformation.
These are important basic considerations because changes in lengths are related to
strains, changes in areas are associated with stresses, and a change of volume is
related to conservation of mass. The deformation gradient is shown to be the essential
quantity necessary to describe the changes of these quantities. The duality of the
Lagrangian and Eulerian formulations is maintained throughout.
2.1 General Description of Deformations

We set up a common global coordinate system and associate xoi with the undeformed
conguration and xi with the deformed conguration. That is,
Initial position: xô = xoi eî Final position: x^ = xi eî
where both vectors are referred to the common set of unit vectors eî .
common global axes Po PP u

^
PP
PP
q
o
1 P
x
^
x
^

e^2 6
-

e^3;
; e^1

;
;
Figure 2.1: Undeformed and deformed congurations.
Motion
A motion is expressed in either of the following equivalent forms
xi = xi (xo1 xo2 xo3 t) or xoi = xoi (x1 x2 x3 t)
15
16 Chapter 2. Deformation
The variables xoi and xi are called the Lagrangian and Eulerian variables, respectively.
In the Lagrangian system, all quantities are expressed in terms of the initial position
coordinates and time in the Eulerian system, the independent variables are xi and
t where xi are the position coordinates at the time of interest. Realizing that the
description of deformation is essentially geometric, then the dierence between the
Lagrangian and Eulerian descriptions can be stated as:
Lagrangian: Put a rectangular grid on the original (undeformed) body | deter-
mine what it will look like during the motion.
Eulerian: Put a rectangular grid on the current (deformed) body | determine
what it looked like in the original state.
In other words, the Lagrangian grid is always superposed on the same material points
and therefore deforms the Eulerian grid is always rectangular (in the deformed state)
and is superposed on a constantly changing set of material points. For this reason,
the two descriptions are sometimes referred to as material and spatial descriptions,
respectively. The signicant dierence in the two descriptions is in the designation of
the neighboring points.
Lagrangian Eulerian
Figure 2.2: Grids illustrating Lagrangian and Eulerian descriptions.
The choice of description also has an eect on the time derivatives that must be
used. That is, consider any quantity (xo1 xo2 xo3 t), the material derivative of is
d = @
dt @t
If is expressed in Eulerian form as (x1 x2 x3 t) = (xo1 xo2 xo3 t), then
d = @ + @ dxj = @ + v @
dt @t @xj dt @t j @xj
where vj is the convective velocity. Thus the velocity in the Eulerian description must
) but
take into account the fact that not only is the quantity itself changing (@ =@t
the particles being considered are also changing.
It must be reiterated that there is no essential dierence between the two de-
scriptions the preference of one over the other is usually based on convenience for
particular problems.
2.1. General Description of Deformations 17
A deformation is a comparison of two states | the intermediate ones are not im-
portant and typically time does not appear explicitly. The deformation of a material
point is expressed as
xi = xi (xo1 xo2 xo3 ) or xoi = xoi (x1 x2 x3 )
A displacement is the shortest distance traveled when a particle moves from one
location to another, that is,
u^ = r^ ; rô = xi eî ; xoi eî or ui = (xi ; xoi )
Since displacement is a comparison of two states then it is the same in both the
Lagrangian and Eulerian descriptions.
Example 2.1: Illustrate the duality between the Lagrangian and Eulerian de-
scriptions of motion.
In the two-dimensional motion given by the Lagrangian description
x1 = 21 (xo1 + xo2 )et + 12 (xo1 ; xo2 )e t
;
x2 = 21 (xo1 + xo2 )et ; 12 (xo1 ; xo2 )e t

;
the displacement vector is

u1 = x1 ; xo1 = 21 (xo1 + xo2 )et + 12 (xo1 ; xo2 )e t ] ; xo1
;
u2 = x2 ; xo2 = 21 (xo1 + xo2 )et ; 12 (xo1 ; xo2 )e t ] ; xo2

;
Here, the displacements at time t are given in terms of the position (xo1 xo2 ) occupied
by the particle at time t = 0. Alternatively, the same displacement could be specied
in terms of the position (x1 x2 ) occupied by the particle at time t. Inverting the
above relations gives the Eulerian descriptions
xo1 = 21 (x1 + x2 )e t + 12 (x1 ; x2 )et
;
xo2 = 21 (x1 + x2 )e t ; 12 (x1 ; x2 )et

;
and
u1 = x1 ; xo1 = x1 ; 21 (x1 + x2 )e t + 21 (x1 ; x2 )et ]
;
u2 = x2 ; xo2 = x2 ; 12 (x1 + x2 )e t ; 21 (x1 ; x2 )et ]

;
Thus, the displacement can be written either as u = u(xoi ) or u = u(xi ) and we

expect the relationship to be invertible.
Deformation Gradients
In the mechanics of deformable bodies, we are particularly interested in the defor-
mation of neighboring points that they are dierent is in the nature of deformable
bodies.
. August 2007
common global axes Po
@ P o 0
@
dx ^Ro 1P

e^2 6
dx^ ?P 0

-
e^3 ;

; e^1
;
;
Figure 2.3: Deformation of neighboring points.
Consider a deformation in the vicinity of the point P that is, consider two points
P and P separated by dxoi in the undeformed conguration and by dxi in the deformed
0
conguration. The positions of the two points are related through

P : xi = xi (xoi)
P : xi + dxi = xi(xoi + dxoi)
0
@xi dxo + 1 @ 2 xi dxodxo + : : :

xi(xoi ) + @x

o j 2 @xo @xo j k
j j k
Hence if dxoi is small, that is, the neighboring points are very close to each other, then
@xi dxo
dxi = @x (2.1)
o j
j
This describes how the separation in the deformed conguration is related to the
separation in the undeformed conguration. It is expressed in matrix form as
2 @x1 @x1 @x1 3
8 9 66 @xo1 @xo2 @xo3 77 8 o 9
< dx1 >
> = 66 @x2 @x2 @x2 77 > < dx1o >=
> dx 2 = 6 7 dx
: dx3 > 66 @x1 @x2 @x3 77 >
o o o : o> 2
4 @x3 @x3 @x3 5 dx3
@xo1 @xo2 @xo3
The inverse (which should exist for continuous deformations) can also be written as
2 @xo @xo @xo 3
8 o 9 66 @x11 @x21 @x31 77 8 9
< dx1o >
> = 66 @xo2 @xo2 @xo2 77 > < dx1 >
= o
o = @xi dx
> dx = 6 7 dx2 or dx @xj j
: dx2o > 66 @xo1 @x2o @x3o 77 > : dx3 >
i
3 4 @x3 @x3 @x3 5
@x1 @x2 @x3
@x o
i @xi
The quantities @xo @x are called the deformation gradients and form the basis
j j
of the description of any deformation. Deformation gradients relate the behavior of
2.1. General Description of Deformations 19
neighboring particles and this is essential for forming derivatives. Derivatives in the
Lagrangian and Eulerian coordinates are dierent because fundamentally dierent
particle pairs are considered.
The relation
@xi dxo
dxi = @x o j
j
uniquely species dx1 dx2 dx3 in terms of dxo1 dxo2 dxo3. On the assumption that the
deformation is continuous then we should be able to write
o
fdx g =
@xi ] 1fdxg
;
@xoj
which is true only if
@xi ] 6= 0
det @x o
j
Dene the Jacobians referenced to the undeformed and deformed congurations as,
respectively,
J o det @x@xi ] = @x1 @x2 @x3

ijk o o o

o
j @xi @xj @xk
J det @x i ] = @x1 @x2 @x3
o o o o

@xj ijk
@xi @xj @xk (2.2)
Some other useful forms for the Jacobian are
@xi @xj @xk
J o = ijk @x @xi @xj @xk
J o = 61 pqr ijk @x
o @xo @xo o @xo @xo
1 2 3 p q r
@x o @xo @xo @x o @xo @xo
J = ijk @xi @xj @xk J = 61 pqr ijk @xi @xj @xk
1 2 3 p q r
Note that the Jacobian is a scalar quantity.
We will impose the restriction on any deformation that no region of nite volume
is deformed into a region of zero or innite volume. That is, we restrict the values of
the Jacobian as
0 < Jo < 1 0<J < 1
It is necessary to always check this to see if the deformation is physically possible.

Earlier, we said that a deformation is a comparison of two states | the interme-
diate ones do not matter. The Jacobian, therefore, plays a valuable role in restricting
the path between the two states.
Example 2.2: A particular deformation is described by

x1 = xo1 + k xo2 x2 = 2xo1 + 3xo2 x3 = xo3
. August 2007
where k is a parameter. What are the restrictions on the allowable deformation?
The deformation gradient is
21 k 03
@x
@xoi ] = 4 2 3 0 5
j 0 0 1
The Jacobian is the determinant of this and is
J o = 3 ; 2k
Therefore, in order for this to be a real deformation, we must have the restriction
that k < 1:5. Suppose, however, k = 2 so that
x1 = xo1 + 2 xo2 x2 = 2xo1 + 3xo2 x3 = xo3
The new points could be plotted (drawn) and the deformed shape would `look reason-
able'. However, if the deformation is followed through its history (as the parameter
k is changed continuously) then at one stage (k = 1:5) the volume is zero and after
that the body will have turned itself inside out.
2.2 Deformation of Lines, Areas, and Volumes

The deformation gradient forms the basis for the description of any deformation. This
section demonstrates its use in the description of the deformation of lines, areas, and
volumes.
; dx
^
;
ô
; dx
;
Figure 2.4: Deformation of lines.
Lines
The descriptions of a line segment before and after deformation are
dxô = dxoieî dx^ = dxi eî
A straight forward application of the deformation gradient gives
dx^ = @x@xi dxoe^
o j i
j
2.2. Deformation of Lines, Areas, and Volumes 21
Consider the special case of a line dxô originally oriented only along e^1 , that is,
dxô = dxo1e^1 + dxo2 e^2 + dxo3 e^3 = dxo1e^1
After deformation this line segment becomes
dx^ = dx1 e^1 + dx2 e^2 + dx3e^3
= @x 1 dxo e^ + @x2 dxo e^ + @x3 dxo e^
@xo 1 1 @xo 1 2 @xo 1 3
1 1 1
Even though dxô is only horizontal, dx^ has all three components.
Areas
The area of a parallelogram region can be calculated by considering the vector cross
product of lines that bound it. That is, if the region is dened by two vectors dxâ
and dx^b, we have
Area = dxâ dx^b = vector
Note that we consider the area to be a vector it has a direction as well as a magnitude
and the direction is given by a normal to the surface.
:

1

dx^b

dxôb o

dA
BM dA CO
dAôBB 1
^
dA C
C
:

oa a
B d x
^ C dx ^
B
C
Figure 2.5: Deformation of areas.
The initial area is
dAô = dxôa dxôb

or, on substituting for the vectors,

dAok e^k = dxoa ob oa ob
i eî dxj e^j = ijk dxi dxj e^k

In component form, this is

dAok = ijk dxoa ob
i dxj
Multiply both sides by pqk and replacing the contraction of two permutation symbols
by the delta functions leads to the following alternative form
pqk dAok = dxoa
p dxq
ob dxoa dxob
; q p
Consider the situation of a rectangle so that the only non-zero components of length
are dxoa
1 and dx2 . That is, with p = 1 and q = 2 the area is dA3 = dx1 dx2 , as
ob o oa ob
expected.
. August 2007
After deformation, the area becomes
dA^ = dxâ dx^b or dAk = ijk dxai dxbj
and by use of the deformation gradient
dAk = dA nk = ijk @x @xi @xj dxoa dxob
o @xo p q
p q
This can be rearranged (by noting that i j p q are dummy indices) to also give
dAk = ;ijk @x @xi @xj dxoa dxob
o @xo q p
p q
Adding this to the previous expression for dAk results in
@xi @xj (dxoa dxob ; dxoa dxob ) = @xi @xj dAo
2 dAk = ijk @x o @xo p q q p ijk o o pqr r
p q @xp @xq
As a nal step to simplify the relation between the deformed and undeformed areas,
multiply both sides by the deformation gradient @xk =@xo1 and expand on r. The only
non-zero situation is for r = 1, hence we get
2dAk @x
@x1
k
o = @xk @xi @xj @xk @xj @xi
ijk ( o o o ; o o o )dA1 = 2ijk o o o dA1
@x1 @x2 @x3 @x1 @x2 @x3
@xi @xj @xk
@x1 @x2 @x3
By introducing the Jacobian we get the simpler expression
dAk @xk o o
@xo1 = J dA1
Similar expressions are obtained by multiplying by the deformation gradients @xk =@xo2
and @xk =@xo3 to nally get
dAk @x o @xk dAo
k o
= J o dAo or dA = J (2.3)
i i
@xoi @xi k
This elegant form for the deformation of areas is somewhat similar to the corre-
sponding one for line segments note, however, that it is the Eulerian form of the
deformation gradient that is used.
Volume
Consider the parallelepiped of sides dxôa dxôb dxôc which deforms into dxâ dx^c dx^c:
The volume before deformation is dV o = (dxôa dxôb ) dxôc or in expanded form
dV o = (dxoa o ob oc
p e^p dxq e^q ) dxr e^r
= dxoa ob oc
p dxq pqk e^k dxr e^r
= pqr dxoa ob oc
p dxq dxr
2.2. Deformation of Lines, Areas, and Volumes 23
Similarly, the volume after deformation is
dV = ijk dxai dxbj dxck
Expand dV using the deformation gradient and recognizing the collection of gradients
as J o, rearrange to get
dV = ijk J odxoa ob oc
i dxj dxk or dV = J odV o (2.4)

6
6

dxôc dx ^ ob 1

dx^c dx
^ b :

dx ôa a
dx^
Figure 2.6: Three edges and three faces of a deforming volume.
An alternative statement of continuity is that mass is constant and since mass =
density volume = constant, then
o dV o = dV = J o dV o
or simply
o = J o J o = o =
In a similar manner, we have
= o J J = =o
Note that J o and J can change from point to point depending on the particular
deformation gradient but an interesting result that can be obtained is
@ (J o) = @ (o =) = 0
@xi @xi
@ (J ) = @ (=o ) = 0 (2.5)
@xoi @xoi
These somewhat surprising relationships will be used in the next chapter when we
consider stress.
. August 2007
Example 2.3: Reconsider the deformation described by
x1 = xo1 + k xo2 x2 = 2xo1 + 3xo2 x3 = xo3
Determine how lines and areas are deformed.
The deformation gradients are
21 k 03 2 3 ;k 0 3
@x @xo
@xo ] = 4 2 3 0 5
i @xi ] = 4 ;2 1 0 5 (3 ;1 2k)
j 0 0 1 j 0 0 1
A line initially in the 2-direction (dxoi = f0 1 0gdSo ) after deformation has the
components
dx1 = 1 0 + k dSo + 0 0 = k dSo
dx2 = 2 0 + 3 dSo + 0 0 = 3 dSo
dx3 = 0 0 + 0 dSo + 1 0 = 0
Hence the new length and its direction are given by
p
dS = k2 + 9 dSo n^ = kpe^1 + 3ê22
9+k
Now consider the area formed by this line projected in the 3-direction, that is,
dAoi = f1 0 0gdAo after deformation it has the components
dA1 = (3 ; 2k)3 dAo ; 2 0 + 0 0]=(3 ; 2k) = 3dAo
dA2 = (3 ; 2k);k dAo + 1 0 + 0 0]=(3 ; 2k) = ;k dAo
dA3 = (3 ; 2k)0 dAo + 0 0 + (3 ; 2) 0]=(3 ; 2k) = 0
From this, it is apparent that there is also a component of area generated in the
2-direction. The magnitude and direction of the deformed area are, respectively,
p
dA = 9 + k2 dAo pe1 ; ke^22
n^ = 3^
9+k
The vector directions of the line and area are perpendicular to each other.
Example 2.4: Consider the following plane inhomogeneous deformation

x1 = R ; xo2 ] sin(xo1 =R) x2 = R ; R ; xo2] cos(xo1 =R) x3 = xo3 (2.6)
where R is a positive parameter. What (if any) are the restrictions for this to be
a valid deformation? Draw the deformed shape of material initially lying between
;h < xo2 < h, and calculate the orientation and magnitude of the deformed areas.
This deformation is shown in Figure 2.7. Note that initially horizontal lines
become arcs of concentric circles, while initially vertical lines become radial lines
emanating from a common point. The location of the common point changes as the
deformation changes. This deformation resembles that of bending.
pp p p p p pp p ppp ppp pp pbpp ppp pp ppp pp p pp pp p p p p
2.3. Strain 25
ppp p pp p pp p p x pxp pp p pp a ppp

0
after
a : xo1 = R=2
c o2
6
pp p p p p p
2
p
b : xo1 = R
pp p p p p p pp p p
0
before
ppp ppp a b
xo1 -
x1
Figure 2.7: Shape before and after deformation.

The deformation gradient is given by
" # 2 (R ; xo2) cos(xo1 =R)=R ; sin(xo1 =R) 0 3
@xi = 4 (R ; xo ) sin(xo =R)=R cos(xo =R) 0 5
@xoj 2 1 1
0 0 1
The Jacobian is the determinant of the deformation gradient matrix and on multi-
plication simplies to o
J o = 1 ; xR2
We note that as long as xo2 < R that the volume remains positive.
The areas are related through the inverse of the above deformation gradient.
Since it is a 2-D problem, we get
dA cos(xo =R) sin(xo1 =R)
dAo
1 = 1 1
dA2 ;J o sin(xo1=R) J o cos(xo1 =R) dAo2
Areas that were initially vertical and facing the 1-direction are preserved this can
be checked by comparing dA2 for xo1 = R=2 to dA1 for xo1 = R. Areas that were
initially horizontal either contract (xo2 > 0) or expand (xo2 < 0). This is the hallmark
of a beam or plate in bending.
In the limit as xo2 =R << 1, then J o 1 and areas are preserved. This is
the situation that is be prevalent when the bending of thin-walled structures is
considered.
2.3 Strain
Strain is a measure of the `stretching' of the material particles within a body. That
is, it is a measure of the relative displacement without rigid body motion and is an
essential ingredient for the description of the constitutive behavior of materials.
Elementary Strain Measures

There are many measures of strain in existence, so it is worthwhile to review some
of the more common ones so as to put into perspective the measures we wish to
. August 2007
introduce. Assume that a line segment of original length of Lo is changed to L: Some
of the common measures of strain are:
Engineering strain:
e = change in length = L
original length L o
True strain:
change in length = L = L
eT = nal (current) length L Lo + L
Logarithmic strain:
ZL Z L dl
eN = true strain = = ln( L )
Lo Lo l Lo
An essential requirement of a strain measure is to allow the nal length to be calcu-
lated. This is true of each of the above since
Engineering: L = Lo + L = Lo + Lo e = Lo (1 + e)
True: L = Lo + L = Lo + e LoT = Lo T
T
(1 ; e ) (1 ; e )
N
Logarithmic: L = Lo exp(e )
All these measures are equivalent since they allow L (or L) to be calculated knowing
Lo . The relation among the measures are
eT = 1 +e e eN = ln(1 + e)
Because the measures are equivalent, then it is a matter of convenience as to which
measure is to be chosen in an analysis.
The diculty with these strain measures is that they do not have proper transfor-
mation properties. This would poses a problem in developing our three dimensional
theory because the quantities involved should transform as tensors of the appropriate
order.
Tensorial Strain Measures

As a medium deforms, various positions of the medium will translate and rotate. The
easiest way to distinguish between deformation and the local rigid-body motion is to
consider the change in distance between two neighboring material particles. We will
use this to establish our strain measures.
2.3. Strain 27
Suppose that two material particles, before the motion, have coordinates (xoi ) and
(xoi + dxoi )
and after the motion, (xi ) and (xi + dxi). The initial and nal distances
between these neighboring particles are given by
dSo2 = dxoi dxoi = (dxo1 )2 + (dxo2 )2 + (dxo3 )2
and
dS 2 = dxi dxi = @xm @xm o o
@xo @xo dxi dxj
i j
respectively. Only in the event of stretching or straining is dS 2 dierent from dSo2.
That is, @x @x
2 2 2 m m
dS ; dSo = dS ; dxi dxi = @xo @xo ; ij dxoi dxoj
o o
i j
is a measure of the relative displacements. It is insensitive to rotation as can be easily
demonstrated by considering a rigid body motion. On the other hand, if the Eulerian
variables are employed, the same change of length can be expressed as
@x o @xo
2 2 2
dS ; dSo = dxidxi ; dSo = ij ; @xm @xm dxidxj
i j
Consequently, these equations can be written as
dS 2 dSo2 = 2Eij dxoidxoj = 2eij dxi dxj
; (2.7)
respectively, by introducing the strain measures
@x @x @x o @xo
m m
1
Eij 2 @xo @xo ; ij eij 2 ij ; @xm @xm
1 (2.8)
i j i j
It is easy to observe that both Eij and eij are symmetric tensors of the second or-
der and are called the Lagrangian and Eulerian strains, respectively. The matrix
expressions for both strains are
2Eij ] @xmo ] @xmo ]T ; d I @xm ] @xm ]T

o o
2 eij ] d I
@xi @xj c c ;
@xi @xj
where d I c is the unit matrix.
The two strain measures are related by
Eij = epq @x p @xq
@xo @xo
i j
Thus only knowing eij , for example, it is not possible to determine Eij without knowl-
edge of the full deformation gradient @xi =@xoj . This is because both strain measures
are missing the information about the rigid body rotation.
. August 2007
Sometimes the straining is described in terms of the Cauchy-Green deformation
tensor dened as
X
Cij k @xko @xko or C ] @xo ] @xo ]T
@xi @xj @x @x
Then
Eij = 12 Cij
The principal values of Cij ] have the meaning of principal stretches and we will nd
a useful application in discussing rubber elasticity in Chapter 4.
Example 2.5: Show that the Lagrangian strain measure is a second order tensor.
Consider two coordinate systems the change of lengths must be the same in the
two coordinate systems
dS 2 ; dSo2 = 2Eij dxoidxoj = 2Eij dx oidx oj
0 0 0
Since dxoi is a rst order tensor, its components in the primed system are
dxoi = ij dxoj dxoi = jidx oj
0 0
0
Substituting the latter of these gives

dS 2 ; dSo2 = 2Eij piqj dx op dx oq = 2Eij dx oi dx oj
0 0
0 0 0 0 0
Interchanging subscripts we conclude that

Eij = ip jq Epq
0
which is the transformation law for second order tensors.
Example 2.6: Determine the strain for the deforming body of Figure 2.7.
The displacement gradient is
" # " # 2 o xo1 =R) ; 1 ; sin(xo1 =R) 0 3
@ui = @xi ; = 4 J Jcos( o sin(xo =R) cos(xo1 =R) ; 1 0 5
@xoj @xoj ij 0
1
0 1
with J o = 1 ; xo2 =R. The strains are
o o
E11 = J o C ; 1 + 21 (J o C ; 1)2 + (J o S )2 + (0)2 ] = ; xR2 + 12 ( xR2 )2
E22 = C ; 1 + 12 (C ; 1)2 + (;S )2 + (0)2 ] = 0
E12 = 21 ;S + J o S ] + 21 (J o C ; 1)(;S ) + (J o S )(C ; 1) + (0)2 ] = 0
where we used C cos(xo1 =R) and S sin(xo1 =R). Only line segments initially in
the x1 -direction are strained. There is a line, xo2 = 0, which is not strained other
2.3. Strain 29
lines are strained in proportion to their distance from this line. In the limit of a
very thin body (xo2 =R << 1), the strain distribution is approximately linear
o
E11 ; xR2 ;xo2 dx
dv
o
1
where v is the u2 displacement of the xo2 = 0 line. These are the strain characteristics
of a beam or plate in bending.
Principal Strains and Invariants

Because Eij and eij are symmetric second order tensors, they enjoy all the properties
as discussed in Section 1.4. In particular, they have principal values which have an
interesting physical interpretation in terms of the ellipsoid of Figure 1.2.
Consider the line segment before deformation as a vector dSô = dxoieî. We can
then interpret Equation (2.7) as a vector product relation
dS 2 dSo2 = 2Vj dxoj = 2V^ dSô
; Vj Eij dxoi
(2.9)
Consider dierent initial vectors dSô each of the same size but dierent orientations
they will correspond to dierent vectors V^ . Figure 1.2 shows a collection of such
vectors where dSô traces out the coordinate circles of a sphere. Note that V^ traces
an ellipse and many such traces would form an ellipsoid. That is, the sphere traced
by dSô is transformed into an ellipsoid traced by V^ .
When V^ and dSô are parallel, the dot product is extremal and consequently, so
also is the strain. Thus, the axes of the ellipsoid will correspond to the principal
strains. With this in mind, we have the interpretation of the above equation as
1 2 2 = noi Eij noj
2 dS =dSo ; 1]
= E11 no1 no1 + E22 no2no2 + E33 no3 no3 + 2E12no1 no2 + 2E13 no1no3 + 2E23 no2no3
where noi = dxoi=dSo. This is the equation of the ellipsoid with respect to the co-
ordinate directions. The extremum values can be established by dierentiation with
respect to nop when doing this, however, we must also take into account the constraint
that nok nok = 1. We implement this by way of a Lagrange multiplier and extremize
= Eij noj noi nok nok
;
where is the (as yet unknown) Lagrange multiplier. Dierentiating this leads to
Eij noj noi = 0
; or Eij ; ij ]T noj = 0
We now recognize this as the eigenvalue problem discussed in Section 1.4 and as
the principal value of strain.
. August 2007
Physical Interpretation of Normal Strains
To relate these strain tensors to the strain quantities with which we are familiar,
consider the line element
dxo1 = dSo dxo2 = dxo3 = 0
at the initial state. After deformation, the line element is given by dxi with magnitude
dS .
f0 dSo 0g 6 dS

12
:

- dS
fdS o 0 0g
Figure 2.8: Deformation of two initially perpendicular line elements.
Let E1 be the extension per unit original length of the element, that is,
E1 = dS dS
; dSo
or dS = (1 + E1 )dSo
o
For this line element we also have
dS 2 dSo2 = 2E11dSo2
;
and combining with the above yields

(1 + E1)2 dSo2 dSo2 = 2E11dSo2
;
Thus, the extension E1 is related to the Lagrangian strain component E11 as

q
E11 = E1 + 21 E12 or E1 = 1 + 2E11 1 ;
Similar relations for line elements originally in the xo2 and xo3 directions can be obtained
as
q
E22 = E2 + 12 E22 or E2 = 1 + 2E22 1 ;
q
E33 = E3 + 21 E32 or E3 = 1 + 2E33 1 ;
The components E11 , E22 and E33 are called the normal components of strain.
Notice that there is a certain asymmetry (as regards the stretching direction) in
the meaning of the normal components. For example, as the line is stretched, E1
increases possibly without limit resulting in E11 doing the same. If, however, the line
is shrunk so that E1 is negative, then there is a denite limit given by E1 = 1 which
;
2.3. Strain 31
corresponds to L = ;Lo meaning that the line length has shrunk to zero. That is,
we have the limits
;1 < E1 < 1 ;0:5 < E11 < 1
This asymmetry between the stretching and shrinking directions is important when
considering the constitutive behavior.
Applying the binomial expansion to the expression for E1 in terms of E11 gives
E1 = (1 + E11 12 E112 + ) 1
; ;
= E11 12 E112 +
;
When E11 is small, that is, when E11

1, the above equation reduces to
E1 E11

This simply says that E11 can be interpreted as an extension per unit length only
when it is small.
Consider a line segment initially at an arbitrary angle
in the xo1 ; xo2 plane in
the undeformed conguration, then its length after deformation is obtained from
dS 2 dSo2 = 2Eij dxoidxoj = 2 E11dxo1 dxo1 + E21 dxo1dxo2 + E12 dxo2dxo1 + E22 dxo2 dxo2]
;
Realizing that dxo1 = dSo cos

and so on, and that the strain of this arbitrary line is
dS 2 dSo2 2E dSo2
;
leads to
E = E11 cos2
+ E21 cos
sin
+ E12 cos
sin
+ E22 sin2
This gives us the transformation rule for the components of strain and so that they
transform as second order tensors.
Physical Interpretation of Shear Strains

By rearrangement, the o-diagonal components of the strain tensor can be written
in terms of only measurements of change of lengths. That is, the shear strain can be
interpreted in terms of normal strains this is the strain gage rosette interpretation of
the components of the strain tensor. An alternative interpretation in terms of changes
of angles follows.
A deformation can exhibit distortion in the conguration, that is, exhibit a change
in the angle between two line elements. Consider, in the initial state, two line elements
parallel to xo1 and xo2 , respectively. The two line elements are denoted by dxoi and dxoi,
respectively, with
. August 2007
These two elements are perpendicular to each other initially. After deformation, dxoi
is deformed into dxi , and dxoi into dxi.
Denoting the angle between dxi and dxi by 12 and taking the dot product of
these two vectors, we obtain
@xi @xi dxo dxo = @xi @xi dxo dxo = 2E dxo dxo
dx^ dx^ = dS dS cos 12 = dxi dxi = @x o @xo k m @xo @xo 1 2 12 1 2
k m 1 2
which can readily be rewritten as
dS dS cos 12 = 2E12dSodSo
By substituting for dS and dS in terms of the extensions, we get
dS = (1 + E1)dSo dS = (1 + E2 )dSo
thus leading to
cos 12 = (1 + E2E)(112 + E )
1 2
Denoting the change in angle by
12 12 12
;
and using the expressions for the extensions in terms of the strain components, we
nally obtain
sin 12 = 2E12
1 + 2E11 1 + 2E22
p p
Thus, all the Lagrangian strain components E11 E22 and E12 contribute to the change
of angle. However, it is only when E12 = 0, that the angle between the two elements
would be preserved. The component E12 therefore seems a good measure of the
`shearing' of perpendicular line segments.
Since the term sin 12 must lie in the range 1, we then have the limits on E12 of

q q q q
1 < 12 < 21 or
;
2 ; 1 + 2E11 1 + 2E22 < 2E12 < + 1 + 2E11 1 + 2E22
The limits on E12 are a combination of those on 12 and on the stretches.
Consider two perpendicular lines segments initially oriented at an arbitrary angle
from the 1- and 2-axis, respectively, in the undeformed conguration. The change
of angle 12 after deformation is obtained from
0
@xi @xi dxo dxo = 2E + ]dxo dxo

dS dS sin 12 = dxi dxi = @x km km k m
0
o @xo k m
k m
Expanding this gives
dS dS sin 12 = (2E11 + 1)dxo1 dxo1 + 2E21 dxo1 dxo2 + 2E12 dxo2 dxo1 + (2E22 + 1)dxo2dxo2
0
2.3. Strain 33
Realizing that the undeformed segment lengths are given by
dxoi = dSofcos
sin
0g dxoi = dSof; sin
cos
0g
we obtain
dS dS sin 12 = 2E12 dSodSo
0 0
where
E12 = ;(E11 ; E22 ) cos
sin
+ E12(cos2
; sin2
)
0
The quantity E12 can be shown to be the Lagrangian strain component E12 in the
0
new coordinate system obtained by a rotation

in the x1 ; x2 plane (rotation about
the x3 axis).
Physical Interpretation of Eulerian Strains

A similar procedure leads to the geometrical interpretation of the Eulerian strain
components. Consider a line element dxi in the deformed state which is parallel to
the x1 -axis
dx1 = dS dx2 = dx3 = 0
Let the extension per unit deformed length for this element after deformation be
e1 = dS dS
; dSo
For this particular line element we have

dS 2 ; dSo2 = 2e11 dS 2
Comparing the above two equations, we obtain
e1 = 1 ; 1 ; 2e11
p
Again, if e11 is small, then e1 = e11 :

The Eulerian strains also exhibit limits. For example, as the line is stretched
e1 increases but only up to the denite limit given by e1 = 1 which corresponds to
L = 1. This results in the limit e11 = 0:5. The full limits are
;1 < e1 < 1 ;1 < e11 < 0:5
Notice the complementarity between these limits and the corresponding ones for the
Lagrangian strains.
Similarly, if we consider two line elements dxi and dxi which are parallel to x1 and
x2 axes in the deformed state, then the angle change of these two elements from the
initial state to the deformed state can be easily derived as
sin 12 = p 2ep
12
1 ; 2e11 1 ; 2e22
where 12 = 12 ; 21 in which 12 is the angle between these two elements before
deformation. Again, it is apparent that ; 21 < 12 < 12 .
. August 2007
2.4 Rotation
A general deformation can be conceived as a straining action plus a rotation. While
rotations of rigid bodies are fairly easy to comprehend, the description of the rotations
of deformable bodies requires more development. The conceptual diculty arises
primarily because dierent lines through a given point on the body can have a dierent
rotation. We will nd it necessary to introduce the idea of an àverage' or mean
rotation.
Rotation of Single lines

We wish to consider the rotation of a general line element OP which deforms to O P . 0 0
For convenience, let the two lines occupy the same position at the origin so that O
and O coincide.
0
common global axes 3 P

1 P 0
ô
dx
dx^

6

= + 3
0
O;
P -
P 2
A PPP
; A 3
;
PP
; A PP
PP
1 ;
; A P 0
A P
Figure 2.9: Rotation of a line element.
Consider a line segment that is initially lying in the 1-2 plane that is, consider
the lines OP and OP (the latter is the projection of the deformed line onto the 1-2
0
plane) rotating about the x3-axis. The rotation is obtained as a change in orientation
as follows:
orientation of OP : tan
= dxo2

dx o
1
@x2 dxo + @x2 dxo
o 1 @xo 2
orientation of OP : tan
= dx2 = @x 1 2
dx1 @x1 dxo + @x1 dxo
0 0
@xo1 1 @xo2 2
The rotation of the line projection is 3 =
;
or 0
tan 3 = tan(
;
) = 1tan
0

; tan
+ tan
tan
0
( 2o ; o1 ) + ( @xo2 + @x1o ) cos 2

+ ( @x2o ; @xo1 ) sin 2
@x @x
= @x 1 @x2 @x1 @x2 @x2 @x1
( @x1o + @xo2 ) + ( @xo1 ; @xo2 ) cos 2
+ ( @x1o + @xo2 ) sin 2
@x1 @x2 @x1 @x2 @x2 @x1

2.4. Rotation 35
where
q q
cos
= dxo1 = (dxo1)2 + (dxo2)2 sin
= dxo2= (dxo1 )2 + (dxo2 )2
was used. From this, it is clear that dierent line elements will have dierent amounts
of rotation (since
will be dierent). In fact, some will be positive, some negative. To
characterize the rotation at a point, it is necessary to remove the angle dependence.
This will be done by averaging.
Averaged Measure of Rotation

A measure of average rotation is given as
1 Z 2 1 Z A + B cos 2
; C sin 2

tan 3 2 tan 3 d
= 2 D + C cos 2
+ B sin 2
d
o
where the coecients
A = @x 2 ; @x1
@x1 @x2
o o B = @x2o + @xo1 C = @x1o ; @xo2 D = @x1o + @xo2
@x1 @x2 @x1 @x2 @x1 @x2
are independent of
. Let the denominator be written as
f = D + C cos 2
+ B sin 2
df = (0 ; 2C sin 2
+ 2B cos 2
)d
Therefore, the integral can be rewritten as

1 Z 2 1 df A Z 2 d

tan 3 = 2 2 +
o f 2 o D + C cos 2
+ B sin 2
= I1 + I2
The rst integral is simply
I1 = 41 ln f ]2o = 0
The denominator of the second integral can be rearranged as (using the sine of the
sum of two angles)
p
D + C cos 2
+ B sin 2
= D + pC 2 + B 2 (sin cos 2
+ cos sin 2
)
= D + C 2 + B 2 sin( + 2
)
with tan C=B . Hence the integral becomes using + 2
A Z 2 d
A Z +4 d
I2 = 2 p = 4 p
o D + C + B sin( + 2
)
2 2 D + C + B 2 sin
2
Providing that D2 > C 2 + B 2 , this gives

2 A 1 D tan(=2) + pC 2 + B 2 +4
I2 = 4 q 2 ;
tan 1 q
D ; (C + B )2 2 D ; (C + B )
2 2 2
. August 2007
The tan 1 ( ) term is zero or multiples of 2. Since for small deformations it is required
;
that
A = @x 2 ; @x1 = @u2 ; @u1
@xo1 @xo2 @xo1 @xo2
be a measure of rotation, then let tan 1( ) be 2. Hence
;
1A
I2 = q 2
D ; (C 2 + B 2)
2
Consequently, the average rotation becomes (after substituting for the coecients)
1 ( @x2 ; @x1 )
2 @xo @xo
tan 3 = s 1 2
( 1o )( 2o ) ; 41 ( 1o + @xo2 )2
@x @x @x
@x1 @x2 @x2 @x1
Similarly, for lines initially in the other planes, we have
1 ( @x1 ; @x3 )
2 @xo @xo
tan 2 = s 3 1
( @x1o )( @x3o ) ; 14 ( @xo1 + @x
@x @x @x 3 2
@xo )
1 3 3 1
1 ( @x3 ; @x2 )
2 @xo @xo
tan 1 = s 2 3
@x @x @x3 )2
@x
( 2o )( 3o ) ; 14 ( o2 +
@x2 @x3 @x3
@xo2
In the limit of small deformations, these three angles are simply related to the anti-
symmetric component of the deformation gradient.
Example 2.7: Consider a simple shear deformation parallel to the xo1 ; xo2 plane
and given mathematically by
x1 = xo1 + k xo2 x2 = xo2 x3 = xo3
Determine the average rotation of a point.
Substituting for xi (xoi ) into the formula for the average rotation, we get
1 1
tan 1 = 0 tan 2 = 0 tan 3 = q 2 (0 ; k) = q ; 2 k1
(1)(1) ; 14 (k + 0)2 1 ; 4 k2
If the deformation is small, this gives approximately
3 ; 21 k
2.5. Deformation in Terms of Displacement 37
2 6
kxo
x2 2
o -

-1
Figure 2.10: Simple shear deformation.

In looking at Figure 2.10, we see that the vertical and horizontal lines rotate angles
of
tan 90 = ;k tan 0 = 0
respectively. For small deformations, the average rotation is the average of the
rotations of these two mutually perpendicular lines.
The constraint D2 > C 2 + B 2 for this rotation about the 3-axis becomes
@x1 )( @x2 ) > ( @x1 + @x2 )2
4( @x o @xo @xo @xo
1 2 2 1
Specically, for the simple shear problem, this is equivalent to
4 > k2 or k < 2
When D2 < C 2 + B 2 , that is when the deformation is larger, then the integration
must be performed dierently.
2.5 Deformation in Terms of Displacement

Sometimes it is convenient to deal with displacements and displacement gradients
instead of the deformation gradient. These are obtained by using the relations
xm = xom + um xom = xm ; um
giving the derivatives as
@xm = @um + @xom = ; @um
im
@xoi @xoi @xi im @xi
The previous results will now be summarized in terms of these displacement gradients.
Strains and Rotations in Terms of Displacement

The Lagrangian strain tensor Eij can be written in terms of the displacement by
h m @um i
Eij = 21 @u @xoi + im
@xoj + jm ; ij
h i
= 12 @umo @umo + im @umo + jm @umo + im jm ; ij
@xi @xj @xj @xi
h i
= 21 @uoi + @uoj + @umo @umo
@xj @xi @xi @xj
. August 2007
Similarly, the Eulerian strain tensor eij becomes
h @ui @uj @um @um i
eij = 21 @x +
j @xi @xi @xj
;
Typical expressions for Eij and eij in unabridged notations are given in the following:
E11 = @x @u1 + 1 h @u1 2 + @u2 2 + @u3 2i
o 2 @xo @xo1 @xo1
1 1
1 @u2 1 h @u1 @u1 @u2 @u2 @u3 @u3 i
E12 = 21 @u @xo + @xo + 2 @xo @xo + @xo @xo + @xo @xo
2 1 1 2 1 2 1 2
1 1 h @u1 2 @u2 2 @u3 2 i
e11 = @u
@x1 ; 2 @x1 + @x1 + @x1
h i
e12 = 12 @u1 + @u2 ; 21 @u1 @u1 + @u2 @u2 + @u3 @u3
@x2 @x1 @x1 @x2 @x1 @x2 @x1 @x2
These strain components cannot be interpreted as the normal strains and shear strains
used in introductory courses on solid mechanics. The nonlinear terms in these strain
components make their geometrical meaning less obvious.
A typical rotation term can be written in terms of displacement as
1 @u2 @u1
2 ( @xo @xo )
;
tan 3 = s 1 2
@u2 1 @u1 @u2 2
@u 1
1+ o 1+ o 4 o + o
@x1 @x2 ;
@x2 @x1
Thus the antisymmetric tensor dened as
!
1 @u
!ij = 2 @xo @xo j
;
@ui
i j
characterizes the rotation since is always zero when !ij is zero.
Alternative Expressions for Strains

From
xi = xoi + ui or dxi = dxoi + dui
we have for the deformed length
dS 2 = dxidxi = (dxoi + dui)(dxoi + dui)
Expanding this and moving the rst term to the left-hand-side, we obtain
dS 2 dSo2 = 2duidxoi + duidui
;
By using the chain rule,
@ui dxo
dui = @xo j
j
and decomposing the second order tensor @ui =@xoj into its symmetric and anti-symmetric
parts, we have
dui = (ji + !ji)dxoj = (ij ! ij )dxoj
;
where
! !
ij 1 @uj + @ui !ij 12 @x@uj @ui

2 @xo @xo
o @xo
;
i j i j
The symmetric tensor ij is recognized as the linear part of Eij and is often referred to
as the small or innitesimal strain tensor. The tensor !ij is called the (Lagrangian)
rotation tensor. The change in length now becomes
dS 2 dSo2 = 2(ij !ij )dxoidxoj + (ij !ij )(ik !ik )dxojdxok
; ; ; ;
Noting that
! ij dxoidxoj = 0
and interchanging the dummy indices i and k, we have
dS 2 dSo2 = 2ij + (kj !kj )(ki ! ki)]dxoidxoj
; ; ;
Comparing the above equation with Equation (2.7), we arrive at

Eij = ij + 12 (kj !kj )(ki !ki)
; ;
Similarly, we can show that

eij = ij 12 (kj !kj )(ki !ki)
; ; ;
where
! !
ij 1 @u j @ui 1 @u
!ij 2 @x @x j @ui

2 @xi + @xj ;
i j
These expressions make it clear that the nonlinear part of the strain measure involves
the rotation.
From Equation (2.7), it is easy to show that the vanishing of the strain tensors
is a necessary and sucient condition for all small material elements at a point to
displace without change in length. In other words, Eij = 0 (or equivalently eij = 0)
indicates, at most, a rigid body displacement. However, the condition ij = 0 does
not lead to Eij = 0 if !ij = 0. On the other hand, it is possible that Eij = 0 while
6
ij = 0 and !ij = 0.

6 6
. August 2007
; *

dxo2 2; 6@u2 dxo
@xo1 1
; 1 -?
;

dxo1 @u
1 dxo
6
- @x1 1
o
Figure 2.11: Rotation of line elements

Interpretation of ! ij
In the case of nite deformation, the geometrical interpretation of !ij is not easy to
come by. An attempt will be made using the two-dimensional case as follows.
Consider two line elements dxoi and dxoi with
dxo2 = dSo dxo 1 = dxo3 = 0
which are parallel to the x1 - and x2 -axes, respectively, in the initial state. These line
elements are deformed into dxi and dxi, respectively. The rotation angles that dxoi
and dxoi experience during the deformation are given by
@uo2 @uo1
@x1 @x2
tan 1 = tan 2 = (2.10)
1 + @u 1
@xo1 1 + @u 2
@xo2
Thus, the rotation involves the gradients @u @u1

@xo1 and @xo2 , and the \projected extensions"
2
11 and 22 . In the two-dimensional case, there is only one non-vanishing component
in !ij , which is !
@u 2
!12 = 21 @xo ; @xo @u 1
1 2
Using Equations (2.10), we rewrite the above as
h 1 tan ; 1 + @u2 tan i
!12 = 12 1 + @u
@xo1 1
@xo2 2
Thus, !12 is a weighted average of tan 1 and tan(;2). This should not be confused
with 3. For small 11 22 1 2 we obtain
!12 1

2 (1 ; 2 )
meaning that ! 12 is the average of these two rotations.
Example 2.8: A cubic element is rotated about the x3-axis by =2 as shown
2
6
B' C' A B
-1
A' D' D C
Figure 2.12: Rigid body rotation of a cube.
in Figure 2.12 where points A, B, C, D are displaced to A B C D , respectively.
0 0 0 0
Obtain the symmetric and anti-symmetric parts of the displacement gradient.

The deformed coordinates are given as
x1 = ;xo2 x2 = xo1 x3 = xo3
The displacement components are readily obtained as
u1 = ;xo2 ; xo1 u2 = xo1 ; xo2 u3 = 0
Thus the symmetric and anti-symmetric parts of the displacement gradient are
2 ;1 0 0 3 2 0 +1 0 3
ij ] = 4 0 ;1 0 5 ! ij ] = 4 ;1 0 0 5
0 0 0 0 0 0
It can be easily veried that in this case Eij = 0 and tan 3 = 1 (thus 3 = =2.)
Consequently, in large displacements, existence of non-zero components of ij cannot
be interpreted as the presence of strain.
Innitesimal Strain and Rotation

The full nonlinear deformation analysis of problems is quite dicult and so simpli-
cations are often sought. Three situations for the straining of a block are shown in
Figure 2.13. The general case is that of large displacements, large rotations, and large
strains. In the chapters dealing with the linear theory, the displacements, rotations,
and strains are all small, and Case (c) prevails. Nonlinear analysis of thin-walled
structures such as shells is usually restricted to Case (b), where the de"ections and
rotations can be large but the strains are small. This is a reasonable approximation
because structural materials do not exhibit large strains without yielding or fracture
and structures are designed to operate without this occurring.
If the displacement gradients are small in the solid, that is,

@ui
1 and @ui
1
@xj
o @xj
. August 2007
pp p p pp p
p p
(a)
p p p p p p
pppp ppp ppp pp ppp ppp p p p pp p pp pp p p p
(b) (c)
Figure 2.13: Combinations of displacements and strains. (a) Large displacements,

rotations, and strains. (b) Large displacements and rotations but small strains.
(c) Small displacements, rotations, and strains.
then the product terms in the Lagrangian strain tensor Eij and the Eulerian strain
tensor eij can be neglected. The results are
!
1 @ui @uj
Eij ij = 2 @xo + @xo
j i!
eij ij = 12 @x @ui + @uj
@x j i
where ij and ij are the innitesimal strain tensors. This assumption also leads to the
conclusion that the components Eij and eij are small. Thus, the innitesimal strain
components have direct interpretations as extensions or change of angles. Further,
the magnitudes of the strain components are small as compared with unity, indicating
that the deformation (extension of a material element and the change of angle between
two material elements) is small. Consequently, we have
E1 e1 E2 e2 E 3 e3
Similarly,
23 23
13 13 12 12

which leads to the conclusion

ij ij
meaning that the innitesimal Lagrangian strain components ij and the innitesimal
Eulerian strain components ij are equal in value.
If, in addition to the above, the following condition exists
u
i
1
L
where L is the smallest dimension of the body, then
xi xoi
and the distinction between the Lagrangian and Eulerian variables vanishes. As a
result, the functional forms of the displacement components ui in these two variables
become identical, and the strain components ij and ij have identical functional
forms with no distinction between them necessary. Henceforth, when we use the small
strain approximations, ij will be used to denote both the innitesimal Lagrangian
and Eulerian strain tensors, and xi to denote both Lagrangian and Eulerian variables.
Example 2.9: Establish the relation between the incremental components of the
Lagrangian strain tensor and the incremental components of the displacements.
Let the change of the new positions be rewritten as
xi ;! xi + xi dxi ;! dxi + dxi
The change in length of a line element can be written in terms of the Lagrangian
strain as X X X
2 ij Eij dxoi dxoj = dS 2 ; dSo2 = i dxi dxi ; i dxoi dxoi
The increments in strains due to the changes dxi are obtained from this as
X X X X
2 ij Eij dxoi dxoj = 2 i dxi dxi ; 0 or o o
ij Eij dxi dxj = pdxp dxp
We will consider two representations for dxi .
First, noting that dxoi is not changed,
X @ up o
dxp = d(xop + up ) = dup = i @xoi dxi
Then, with the usual representation for dxp ,

X o o X @ up o @xp o X @xp @ up X @xp @up
ij Eij dxi dxj = ij @xoi dxi @xoj dxj or Eij = p @xoi @xoj = p @xoi @xoj ]
The increments in strain are not solely related to the increments in displacement
gradients. Further, if we use the decomposition of the displacement gradient,
@ui = + ! @ui + @uj
2ij = @x @ui ; @uj
2!ij = @x
@xoj ij ij o @xo
j i
o @xo
j i
then X @xp
Eij = p @xoi pj
The increments in strain are not solely related to the increments in the small strain
tensor.
As a second choice, let
X X X
dx = d(xo + u ) = du = @ ui dx = @ui ]dx = + ! ]dx
i i i i k @xk k k @xk k k ik ik k
. August 2007
where the symmetric and antisymmetric decompositions are given by, respectively,
@ui + @uj
2ij = @x @ui ; @uj
2!ij = @x
j @xi j @xi
Therefore,
X o o X X
ij Eij dxi dxj = ij ij + !ij ]dxi dxj = ij ij dxi dxj
where the anti-symmetry of !ij and symmetry of dxi dxj was used to set the product
to zero. Substituting for dxi in terms of dxoi now gives
X @xi @xj X @xoi @xoj
Emn = ij @xom @xon ij and mn = ij @xm @xn Eij (2.11)
These surprising results shows that although Eij , ij and ij are small, they
are not equal. The main reason for this is because they are referred to dierent
congurations. We will utilize this relation when we consider small variations of the
strain eld in Chapter 6.
2.6 Special Deformations

Some special deformations are considered here that will be useful to our discussions
in the later chapters.
I: Homogeneous Deformation
If the nal position of each particle is a linear function of its initial position, that is,
xi = Cik xok + Bi or dxi = Cik dxok
then the deformation gradient does not depend on xoi,
@xi = C = constants
ij
@xoj
This deformation is said to be homogeneous. Some special characteristics of it are:
The strain is the same irrespective of the point (xoi) considered.
Straight lines deform into straight lines (but angles may change).
All deformations can be treated as locally homogeneous.
2.6. Special Deformations 45
II: Rigid Body Rotation
In a rigid body rotation all the points are given a displacement but the relative
distance between points is unchanged. Consider the two dimensional case described
by
x1 = xo1 cos xo2 sin
;
x2 = xo1 sin + xo2 cos

x3 = xo3
where is the angle of rotation. The corresponding displacements are
u1 = xo1 (cos 1) x2o sin
; ;
u2 = xo1 sin + xo2 (cos 1) ;
u3 = 0
from which the displacement gradients are determined to be
" # 2 (cos 1) sin 0
3
@ui = 6 sin (cos 1) 0 7
; ;
Eij = 0
@xoj 4 0 0
;
0
5
It is now easy to show that all strain components are zero. However, the innitesimal
strain tensor given by !
1 @uj @ui
ij = 2 @xo + @xo
i j
has the components
2 3 2 1 2 3
(cos ; 1) 0 0 ;
2 0 0
ij ] = 64 0 (cos ; 1) 0 75 64 0 ; 12 2 0 75
0 0 0 0 0 0
It is only when is very small is this strain tensor nearly zero.
III: Uniform Extension
A special case of homogeneous deformation is the uniform extension as shown in
Figure 2.14. The motion is given explicitly by
x1 = 1 xo1 + b1
x2 = 2 xo2 + b2
x3 = 3 xo3 + b3
where the 's are stretches and the b's are constants. The special case of simple
extension is described by 2 = 3 = o and is analogous to uniaxial stress.
. August 2007
6
2
6
- -1
b1 L1 1 L1 + b1
Figure 2.14: Simple extension in two dimensions.
The corresponding displacements are
u1 = ( 1 ; 1)xo1 + b1
u2 = ( 2 ; 1)xo2 + b2
u3 = ( 3 ; 1)xo3 + b3
from which the displacement gradients are determined to be
" # 2 1 ; 1 0 0
3
@ui = 64 0 ; 1 0 75
2
@xoj 0 0 3 ; 1
The Lagrangian strain tensor is
21 2 3
2 ( 1 ; 1) 0 0
Eij ] = 64 0 1 ( 2 ; 1)
2 2 0 75
1 2
0 0 2 ( 3 ; 1)
showing it to be diagonal. That is, the coordinate directions are the principal strain
directions. The extensions of line elements arranged in the xo1 xo2 and xo3 directions
are q
E1 = 1 + 2E11 ; 1 = 1 ; 1 E2 = 2 ; 1 E3 = 3 ; 1
and the rotations are given simply as
1 = 2 = 3 = 0
The i have meaning of \new length over original length".
IV: Simple Shear
A simple shear deformation parallel to the xo1 ; xo2 plane is shown in Figure 2.15 and
given mathematically by
x1 = xo1 + k xo2
x2 = xo2
x3 = xo3
2.6. Special Deformations 47
6
2
kxo
x2 2
o -

-1
Figure 2.15: Simple shear.

u1 = k xo2 u2 = 0 u3 = 0
from which the Lagrangian strain tensor is calculated with the result
2 3
0 12 k 0
Eij ] = 64 12 k 21 k2 0 75
0 0 0
Note the presence of a non-zero E22 normal component. It can easily be veried
that there is no extension in the xo1 -direction (E1 = 0), and the extension in the
xo2 -direction is p
E2 = 1 + k 2 ; 1
The rotations are
q
1 = 0 2 = 0 tan 3 = ; 12 k= 1 ; 14 k2
Also, as k approaches 2, 3 approaches ;=2.
V: Generalized Plane Homogeneous Deformation

Consider the deformation of Figure 2.16 where E1 and E2 are extensions and 1 and
2 are rotations and described by
x1 = (1 + E1) cos 1 xo1 + (1 + E2) sin 2 xo2
x2 = (1 + E1) sin 1 xo1 + (1 + E2 ) cos 2 xo2
x3 = xo3
The displacement gradient is
" # 2 (1 + E1) cos 1 ; 1 (1 + E2) sin 2 0
3
@ui = 6 (1 + E ) sin (1 + E2) cos 2 ; 1 0 75
@xj
o 4 1 1
0 0 0
. August 2007

(1 + E2 )dS2o
2

:
(1 + E1 )dS1o
6

1
-
Figure 2.16: Generalized plane homogeneous deformation.

The strain components are obtained as
2 3
E1 + 12 E12 1
2 (1 + E1 )(1 + E12 ) sin( 1 + 2 ) 0
6 1
Eij ] = 4 2 (1 + E1 )(1 + E2 ) sin(1 + 2) E2 + 2 E2 2 0 75
0 0 0
The small strain approximation gives
2 1 (1 + E ) sin + 1 (1 + E ) sin 0 3
(1 + E1 ) cos 1 ; 1 2 1 1 2 2 2
ij ] = 64 21 (1 + E1 ) sin 1 + 12 (1 + E2 ) sin 2 (1 + E2) cos 2 ; 1 0 75
0 0 1
It is only when E1 , E2, 1, and 2 are separately very small that we get the approxi-
mation 2 1 ( + ) 0 3
E1 2 1 2
ij ] 64 12 (1 + 2 ) E2 0 75
0 0 0
Example 2.10: A block experiences a rigid body motion. Estimate the allowable
rotation angle such that the error in strain is not to exceed 10
when the small
strain formulas are used.
The small strain estimate is
11 = cos ; 1 1 ; 21 2 + ; 1 ; 12 2
Therefore we want
1 2 6 < 4:5 10 3 radians] < 0:25o
2 < 10 10 or
; ;
This is a very small angle. In other words, it does not take much of a rigid body
rotation in order to invalidate the use of the small strain approximations.
Exercises 49
Exercises
2.1 A block rotates an angle about the x3 axis. Write down its deformation and
obtain the deformation gradient. Show that the volume change is zero.
2.2 Consider the following deformation with a large value of k
x1 = xo1 + k xo2 x2 = xo2 x3 = xo3
Draw the deformed shape of a volume that was initially a cube. Show that
the formulas describing the deformation of areas are in agreement with the
geometric construction.
2.3 Consider the following deformation
x1 = 3xo1 + k xo2 x2 = 2xo1 + 4xo2 x3 = xo3
What (if any) are the restrictions on k for this to be a valid deformation? Draw
the deformed shape. Show by measurement the consistency of the physical in-
terpretation of the Lagrangian strains with their connection to the deformation
gradient.
2.4 For the previous deformation, determine the principal strains. Draw the be-
fore and after positions of the principal element. Draw the deformed shape
of a volume that was initially a cube. Show that the formulas describing the
deformation of areas are in agreement with the geometric construction.
2.5 Consider the following inhomogeneous deformation
x1 = R ; xo2 ] sin(xo1 =R) x2 = R ; R ; xo2] cos(xo1 =R) x3 = xo3
What (if any) are the restrictions for this to be a valid deformation? Draw the
deformed shape of material initially lying between ;h < xo2 < h. Calculate the
orientation and magnitude of an area that was initially vertical and facing the
1-direction.
2.6 For the previous deformation, determine the Lagrangian and Eulerian strain
tensors. Under what circumstance(s) are they the same?
2.7 Show that the Lagrangian and Eulerian strains are related by
Eij = epq @x p @xq
@xo @xoi j
2.8 Give the physical meaning to the components of the Eulerian strain eij .
2.9 Show that if
J o = ijk @x 1 @x2 @x3
@xo @xo @xo then @xi @xj @xk
J o = ijk @x o @xo @xo
i j k 1 2 3
. August 2007
2.10 Show that @ (J ) = 0 @ (J o ) = 0
@xoi @xi
Physically, what are these saying?
2.11 The Lagrangian strain tensor at a point is
2 p 3
2 ;1 p2 7
6
Eij ] = 4 p ;1 p 3 ; 25
2 ; 2 4
What is the engineering strain of a line element initially oriented as n^ = 12 e^1 ;
1 1
2 e^2 + 2 e^3 . What is the angle change between two line elements initially
p
oriented as n^ a = 12 e^1 ; 12 e^2 + 12 e^3 and n^ b = ; 12 e^1 + 12 e^2 + 12 e^3 .

p p
2.12 Consider the homogeneous deformation of a square in a two-dimensional body

such that the corners move as
(0 0) ) (0 0) (1 0) ) (1 1:5) (0 1) ) (;1 2)
Describe the deformation mathematically. Determine the Lagrangian strain
tensor. What can you say about the deformation given by
(0 0) ) (0 0) (0 1) ) (1 1:5) (1 0) ) (;1 2)
2.13 For the previous deformation, nd the principal strain values and directions.
Determine and draw the nal position of lines initially oriented along the prin-
cipal directions.
2.14 Show that the two expressions for a deforming area
dAok = ijk dxoa ob
i dxj
and
pqk dAok = dxoa ob
p dxq ; dxq dxp
oa ob
are equivalent forms.
2.15 Consider a 2-D square body ABCD which deforms homogeneously into A B C D . 0 0 0 0
(a) Write the displacement components in terms of the Lagrangian and Eule-
rian variables.
(b) Compute Eij and eij .
(c) Find the extension of element AC after deformation by use of Eij .
(d) Find the extension of element AB after deformation by use of Eij .
(e) Find the engineering strain of a vertical element in the deformed body.
x2
6
1:0
B C
0:5 B
0
C 0
-x1
AA 0
D D 0
0 1 2 2:5
Chapter 3
Stress
The kinetics of rigid bodies are described in terms of forces the equivalent concept
for continuous media is stress (loosely dened as force over unit area). The deformed
state is the natural conguration in which to consider stress and this gives rise to the
Cauchy stress tensor. However, to maintain the duality of the Lagrangian and Eule-
rian descriptions, we also introduce stresses dened with respect to the undeformed
conguration.
3.1 Cauchy Stress Principle

Actions can be exerted on a continuum through either contact forces or forces con-
tained in the mass. There are three types of forces and moments worth distinguishing:
Extrinsic: arise (in part) from outside the body, e.g., gravity, magnetic, elec-
trostatic.
Mutual: arise within the body and act upon pairs of particles, e.g., Newtonian
gravitation.
Contact: act upon bounding surfaces and are equipollent (same mechanical
eect) to the loading of one portion of the material, e.g., stress, pressure.
The contact force is often referred to as a surface force or traction as its action occurs
on a surface. We are primarily concerned with contact forces.
More general actions can take place in the form of surface moments and body
moments. Theories that allow existence of these moments are called couple stress
theories, or microstructural theories, and such solids are sometimes called Cosserat
continua. These will not be considered in the following presentation.
Traction Vector
Consider a small surface element of area A on our imagined exposed surface A in the
deformed conguration as shown in Figure 3.1. There must be forces and moments
acting on A to make it equipollent to the eect of the rest of the material. That is,
when the pieces are put back together these forces cancel each other. Let these forces
51
52 Chapter 3. Stress
be thought of as contact forces and so give rise to contact stresses (even though they
are inside the body). Cauchy formalized this by introducing his concept of traction
vector. pp p p p p pp p p p p p pp p p p
pp p p f^ p p p p p p
p AK ppp
common global axes ppppp A ; n
^ pp
ppp A; pp
ppp pp
p p p A p p pp
p pp p p p
e^2 6 p pp p p pp p p p p p p p p pp
-
e^3;
; e
^1
;
;
Figure 3.1: Exposed forces on an arbitrary section cut.
Let n^ be the unit vector which is perpendicular to the surface element A and
let f^ be the resultant force exerted from the other part of the surface element with
the negative normal vector. We assume that as A becomes vanishingly small, the
ratio f=^ A approaches a denite limit df=dA ^ . The vector obtained in the limiting
process is
lim f^ = df^ t^(^n)
A 0 A dA
!
which is called the traction vector. This vector represents the force per unit area
acting on the surface and its limit exists because the material is assumed continuous.
The superscript n^ is a reminder that the traction is dependent on the orientation of
the area.
Also note that, in general, there are moments or torques acting on A and in the
limit
m^ (^n) = lim m^
A 0 A!
We make the assumption that m^ is zero for all n^. Retaining m^ gives couple stress
theory which is needed, for example, when doing notch problems where the notch
geometry is on the order of the grain size. We will neglect couple stresses.
To give explicit representation of the traction vector, consider its components on
the three faces of a cube as shown in Figure 3.2. The traction on the 1-face is
n^ = e^1 : t^(^n) = t(îe1 ) eî = t(^1e1 )e^1 + t(^2e1 ) e^2 + t(^3e1 ) e^3
while on the 2-face
n^ = e^2 : t^(^n) = t(îe2 ) eî = t(^1e2 )e^1 + t(^2e2 ) e^2 + t(^3e2 ) e^3
Note that, in general, t(^1e2 ) 6= t(^1e1 ) : Since this description is somewhat cumbersome,
we simplify the notation by introducing
ij t(^jei )
3.1. Cauchy Stress Principle 53
t^(ê2) @
e^2
I
; ;
;@ 6 ; ; 22 ;
- 21 ;
@ 6
; ; ;
; ; ; ;
; ; *t^(ê1 ) ;

; 23 ;; 12

t^(ê3 )
-
6
e^1 32 -
; 11
@
I
@ 6
- 31 13

;
@ ;
; ; ; ;

;
e^3 ; 33

;
;
; ;
; ;
Figure 3.2: Stressed cube.

where ì' refers to the face and `j ' to the component. More specically,
11 t(^1e1 )
13 t(^3e1 )
31 t(^1e3 )
:::
The normal projections of t^(^n) on these special faces are the normal stress components
11 22 33 , while projections perpendicular to n^ are shear stress components 12 13
21 23 31 ,32 .
It is important to realize that while t^ resembles the elementary idea of stress (force
over area) it is not stress t^ transforms as a vector and has only three components.
The tensor ij is our denition of stress it has nine components with units of force
over area, but at this stage we do not know how these components transform.
Relation between t and n i j
We know that the traction vector t^(^n) acting on an area dAn^ depends on the normal
n^ of the area. The particular relation can be obtained by considering a traction on
an arbitrary surface of the tetrahedron shown in Figure 3.3. On the three faces per-
pendicular to the coordinate directions the components of the three traction vectors
are denoted by ij . The vector acting on the inclined surface ABC is t^ and the unit
normal vector n^ . The equilibrium of the tetrahedron requires that the resultant force
acting on it must vanish.
The equation for the balance of forces in the x1 -direction for the tetrahedron is
given by
t1dA ; 11 dA1 ; 21 dA2 ; 31 dA3 + b1 dV = 0
where b1 is the x1 -component of the body force ^b (which may also contain inertia
terms), t1 is the x1 -component of the traction vector, dAi is the area of the face
perpendicular to the xi axis, dA is the area of the inclined surface, and
dV = 31 hdA
. August 2007
common global axes C
@
@ t^
@

@
6; n ^ @

P
- @

;; B
;

;
; A
;
Figure 3.3: Tetrahedron.
is the volume of the tetrahedron. In this, h is the smallest distance from point P to
the inclined surface ABC. Noting that the normal to the area has the components
n^ = n1 e^1 + n2e^2 + n3 e^3
we conclude that the components of area are
area of face 1: dA1 = n1 dA
Now divide through by dA in the equilibrium relation, and letting h ! 0, we obtain
t1 = 11 n1 + 21 n2 + 31 n3 = j1 nj
Similar equations can be derived by considering the balance of forces in the x2 - and
x3 -directions. The results are, respectively,
t2 = 12 n1 + 22 n2 + 32 n3 = j2nj
t3 = 13 n1 + 23 n2 + 33 n3 = j3nj
These three equations can be written in the indicial notation as
ti = jinj (3.1)
This compact relation says that we need only know nine numbers ij ] to be able to
determine the traction vector on any area passing through a point. These elements
are called the Cauchy stress components and form the Cauchy stress tensor. It is a
second order tensor (since ti and nj transform as rst order tensors) but not necessarily
symmetric (at this stage). The tensorial property indicates that the above relation is
true in any Cartesian coordinate system.
Example 3.1: Consider a block of material with a uniformly distributed force

acting on the 1-face. Determine the tractions on an interior plane.
2 ^
6 A
A
t
;
;
*

;n
A ^ -
P A
A
P
A

-
1
Figure 3.4: Block with end forces.
First consider a vertical cut to expose an interior 1-face, i.e., n^ = f1 0 0g.
Represent the components of the applied force as Pi = fP 0 0g, then equilibrium
gives
ti A ; P i = 0 or t1 = PA = 11 t2 = 0 = 12 t3 = 0 = 13
Similarly, for the 2-face and 3-face we obtain
21 = 22 = 23 = 0 31 = 32 = 33 = 0
Thus the state of stress in the entire body is given by the stress tensor
2 P=A 0 0 3
ij ] = 4 0 0 0 5
0 0 0
Now make a cut at an angle of , that is, the unit normal vector n^ to the inclined
face has the components ni = fcos sin 0g. The traction ti acting on this surface
is given by ti = jinj from which we have
t1 = 11 n1 + 21 n2 + 31n3 = 11 cos = PA cos

t2 = 0 t3 = 0
Finally, consider the normal and tangential components of the traction acting
on this area. Let s^ be the tangential unit vector, we have
tn = t^ n^ = tk nk = t1 n1 = PA cos2 ts = t^ s^ = tk sk = t1 s1 = ; PA cos sin

These two components are in fact the transformed components of the stress tensor
ij , obtained by rotating the x1 ; x2 axes about x3 for an angle . That is
0
tn = 11
0
ts = 12
0
Thus the transformation for t1 and t2 contain sin at most whereas the transforma-
tion for tn and tt contain cos 2 and cos sin .
. August 2007
Equations of Motion in Terms of Stress
Recall that Newton's laws for the equation of motion of a rigid body can be written
as
X^
F = ma^
X^
M = m x^ a^
where a^ is the acceleration and m the mass. These equations will now be used to
establish the equations of motion of a deformable body.
common global axes A;
;
;

; t^
;
;;;
^b

1 V
e^2 6
-
e^3 ;

; e^1
;
;
;
Figure 3.5: Arbitrary small volume.
Consider an arbitrary volume V taken from the deformed body, then the Newton's
laws of motion become, respectively,
Z Z Z
t^dA + ^b dV = u^# dV
A V V
and Z Z Z
^
x^ t dA + x^ b dV = x^ u^# dV
^
A V V
where t^ is the traction on the boundary surface A, and ^b is the body force per unit
mass. In indicial notation, these are rewritten as
Z Z Z
tidA + bi dV = u#idV
Z A Z V ZV
ijk xj tk dA + ijk xj bk dV = ijk xj u#k dV
A V V
These are the equations of motion in terms of ti . We can now obtain the equations of
motion in terms of the stress ij by using ti = pinp and noting that, by the integral
theorem of Chapter 1,
Z Z Z pi
ti dA = pi np dA = @ dV
A A V @xp
The equations of motion become
Z @pi
+ bi ; u#i dV = 0
V @xp
Z @
(ijk xj pk ) + ijk xj bk ; ijk xj u#k dV = 0
V @xp
Noting that
@ (x ) = + x @pk
@xp j pk jk j
@xp
we can expand the second equation and simplify it using the rst equation. Further-
more, since the volume V is arbitrary, we conclude that the integrands must vanish
and therefore
@pi + b = u#
i i
@xp
ijk jk = 0
The second equation shows that ij is a symmetric tensor since the contraction of a
symmetric tensor with an anti-symmetric tensor is zero. Hence the two equations of
motion become
@ij + b = u#
i i
@xj
ij = ji (3.2)
These equations of motion are written in expanded notation as
@11 + @21 + @31 + b = u#
1 1
@x1 @x2 @x3
@12 + @22 + @32 + b = u#
2 2
@x1 @x2 @x3
@13 + @23 + @33 + b = u#
3 3
@x1 @x2 @x3
It is worth repeating that due to the symmetry property of the stress tensor, only
six components are independent. As a result, the number of independent stress
components in the above are reduced since 12 = 21 13 = 31 and 23 = 32 .
It is also worth noting that the equations of motion in terms of the Cauchy stress
measure is applicable to large deformations.
Example 3.2: Under what circumstances (if any) is the following symmetric
stress eld in static equilibrium?
11 = 3x1 + k1 x22 22 = 2x1 + 4x2 12 = 21 = a + bx1 + cx21 + dx2 + ex22 + fx1 x2
all others being zero.
We will consider the case when the body forces (including inertia) are zero. The
rst two static equilibrium equations become
3 + d + 2ex2 + fx1 = 0 4 + b + 2cx1 + fx2 = 0
These must be true for any x1 , x2 which leads to
3+d=0 e=0 f =0
4+b=0 c=0 f =0
It is interesting that the k1 term does not aect the equilibrium.
. August 2007
3.2 Properties of the Cauchy Stress Tensor
Since the Cauchy stress tensor is symmetric and second order, then it inherits many
properties as already outlined in Chapter 1 for second order tensors. It is worthwhile,
however, to review these here in the physical context of stress.
The Cauchy Stress Tensor

The concept of stress introduced in the last section is dierent from the elementary
idea of force over area. In preparation for the more subtle concepts of Lagrange
and Kirchho stress to be introduced later, it is worth our while to recapitulate the
developments of the Cauchy stress tensor.
On any surface of a deformed body, there exists a traction vector t^. The magnitude
and the direction of this vector at a point changes on dierent surfaces with dierent
unit normal vectors n^ .
The three components of the traction vector on the x1 -face (the one that is per-
pendicular to x1 -axis ) are denoted by
11 12 13
Similarly, on the x2 -face and x3 -face we have the components of the traction vectors
as
21 22 23 and 31 32 33
respectively. Although these nine scalars ij ] have the size of a second order tensor,
they cannot be assumed automatically to have formed a tensor, as they have not yet
been shown to have satised the coordinate transformation law.
By considering the equilibrium of a small arbitrary volume, the relation ti = jinj
was established for tractions acting on arbitrary areas. The quotient rule then showed
that ij is a second order tensor. From the above, we conclude that the state of
stress in a body is completely given by the stress tensor ij . Therefore, given any
surface with the unit normal vector n^ , one is able to determine the stress vector (force
intensity) acting on that surface if the stress tensor is known.
The nal step considered the equations of motion of an arbitrary volume from
which we concluded the symmetry of the Cauchy stress tensor.
In this development, the physical entity is the traction vector the stress tensor
is viewed as a derived (or dened) quantity that connects the components of the
traction vector acting on any plane.
Principal Stresses
It is noted the traction vector t^ acting on a surface depends on the direction n^ and is
usually not parallel to n^ . We now attempt to nd an n^ such that the stress vector is
acting in the direction of n^, that is,
ti = ni
3.2. Properties of the Cauchy Stress Tensor 59
where is a scalar representing the magnitude of the stress vector. A direction which
satises this is called a principal direction of stress, and is called the corresponding
principal stress.
By using the relation ti = jinj , the above yields
jinj = ni
which can be rewritten as
ji ; ij ]T fnj g = 0
Thus, three equations result for the three unknowns n1 n2 and n3. It is well known
that for the three homogeneous equations to yield a non-trivial solution, the deter-
minant of the coecient matrix must vanish, that is,
detji ; ij ] = 0
This is recognized as a standard eigenvalue problem. Viewing ij as a real, symmetric
matrix the eigenvalues i are guaranteed to be real, and the corresponding eigenvec-
tors ni are mutually orthogonal. The proof of this theorem can be found in many
books on matrix theory and is also discussed in Chapter 1.
Example 3.3: Consider a state of stress at a point given by the stress tensor
22 2 03
ij ] = 4 2 2 0 5
0 0 1
The stress invariants can be computed as I1 = 5 I2 = 4 and I3 = 0. The charac-
teristic equation is, therefore,
3 ; 5 2 + 4 = 0
The principal values are readily obtained as follows
1 = 4 2 = 1 3 = 0
The unit normal n^ (1) corresponding to 1 = 4 can be obtained by substituting
this value back into the system of equations to obtain
;2n(1) (1)
1 + 2n2 = 0
2n(1) (1)
1 ; 2n2 = 0
;3n(1)
3 = 0
This indicates that the three equations are not independent. Thus, only two equa-
tions are available to determine the solution. Since there are three unknowns, two
equations can determine only the ratios among the three quantities n(1) (1)
1 n2 and
n(1)
3 . However, with the specication that n^ is a unit vector then
n21 + n22 + n23 = 1
. August 2007
and the solution is uniquely obtained as
n(1) 1 n(1) 1 n(1)
1 = p2 2 = p2 3 =0
Following similar manipulations, the unit vectors n^ (2) and n^ (3) corresponding to
2 and 3, respectively, can be determined. We have
n(2)
1 =0 n(2)
2 =0 n(2)
3 =1
and
n(3) 1 n(3) ;1 n(3)
1 =p 2 2 =p 2 3 =0
It is straightforward to show that the n^ (i) are orthogonal.
Normal Stress
The normal stress n is the component of the traction vector in the direction of n^ ,
the unit normal to the surface of interest. This component of stress can be obtained
by taking the scalar product of t^ and n^ as
N = tini = ij ninj
If the coordinate axes are chosen to coincide with the principal directions of stress,
then ij has the form of a diagonal matrix, that is,
ij = 0 if i = j 6
and
11 = 1 22 = 2 33 = 3
In that case, the normal stress N can be expressed as
N = 1 n21 + 2 n22 + 3 n23
If the principal stresses are ordered such that
1 2 3

then we obtain
1 > N > 3
by using n21 + n22 + n23 = 1. This indicates that 1 and 3 are the maximum and
minimum normal stresses, respectively, at the point.
The extremum values of the normal stress can be established more formally by
considering the behavior of the normal component of traction. We know that N =
3.2. Properties of the Cauchy Stress Tensor 61
jinj ni but when nding the extremum we must also take into account the constraint
that nk nk = 1. We implement this by way of a Lagrange multiplier and extremize
= jinj ni nk nk
;
where is the (as yet unknown) Lagrange multiplier. Dierentiating this with respect
to np leads to
jinj ni = 0 or ji ij ]T nj = 0
; ; f g
We now recognize as the principal value of stress.
Shear Stress
The shearing stress is the projection of the stress vector on the surface of interest.
That is,
t^ = N n^ + n^ or n^ = t^ ; N n^
0 0
where n^ is perpendicular to n^. The magnitude of the shearing stress, , is given by

0
2 = ( n^ ) ( n^ ) = (t^ ; N n^ ) (t^ ; N n^) = jt^j2 ; N2 = ti ti ; N2

0 0
where jt^j is the magnitude of the traction vector. If the principal directions are chosen
as the coordinate axes, then
t1 = 1j nj = 1 n1
t2 = 2j nj = 2 n2
t3 = 3j nj = 3 n3
Thus,
t^ 2 = titi = t2i + t22 + t23 = (1 n1 )2 + (2 n2 )2 + (3 n3 )2
j j
These allow the magnitude of the shear stress to be obtained as

2 = (1 n1 )2 + (2 n2 )2 + (3 n3 )2 (1 n21 + 2 n22 + 3 n23 )2
;
= n21 (1 n21 )12 + n22(1 n22)22 + n23 (1 n23 )32 212 n21 n22
; ; ; ;
;22 3 n22 n23 21 3 n21n23

;
Using ni ni = 1 to replace the terms in parenthesis, as for example, 1 n21 = n22 + n23 ,
;
we obtain nally
2 = n21n22 (1 2)2 + n22 n23(2 3 )2 + n23 n21 (3 1 )2
; ; ;
It is obvious from this that = 0 on the surfaces with

n1 = 1 n2 = n3 = 0
n2 = 1 n1 = n3 = 0
n3 = 1 n1 = n2 = 0
. August 2007
while the normal stress N either reaches the maximum values or the minimum value.
Consider all the surfaces that contain the x2 -axis with
n1 = 0
6 n3 = 0
6 n2 = 0
The shear stress then is
2 = n23 n21(3 1 )2 = (1 n21 )n21(3 1 )2
; ; ;
since n21 + n23 = 1. The extremum value of occurs at

@ ( 2 ) = 0 = (2n 4n3)( )2
1 1 3 1
@n1 ; ;
Solving for n1 gives us the directions

n1 = 1p n2 = 0 n3 = 1 p
2 2
That is, the maximum value of the shearing stress occurs on the surfaces bisecting
the angle between the x1 - and x3 -axes. The corresponding value of the shearing stress
is
2
max = 41 (3 1 )2 or max = 12 3 1
; j j j ; j
This conclusion can be reached for the general case if the principal stresses are ordered
as 1 > N > 3 .
3.3 Stresses Referred to the Undeformed State

To complete our duality of treatments of stress and deformation, we need to consider
the equations of motion with respect to the undeformed conguration. Specically,
the Cauchy equations of motion are in terms of the spatial partial derivatives and
these must be changed to derivatives with respect to the undeformed state. In doing
so, we introduce two new measures of stress, the Lagrangian and Kirchho stresses,
respectively.
Equations of Motion in the Undeformed State

The equations of motion were derived in terms of the Cauchy stress tensor as
@ji + b = u#
i i
@xj
ij = ji
which are in reference to the deformed state.
3.3. Stresses Referred to the Undeformed State 63
There are advantages to expressing the equations of motion in terms of variables
in the undeformed state. We begin with the body force ^b, which is the body force per
unit mass in the deformed conguration. Dene the body force per unit mass in the
undeformed state as ^bo such that the resultant force is the same. That is,
boi o dV o = bi dV
In view of the mass conservation law, o dV o = dV , we obtain
boi = bi
This body force relation is also valid for inertial forces.
We change the spatial derivatives to material derivatives as follows:
! !
@ji = @ji @xop = @ @xop ; @ 2 xop = @ @xop
@x @xo @x @xo ji @x
j p j p j
ji o
p@x @x @xo ji @x
j p j
Thus, the equations of motion become
!
@ @xop + b = u#
@xop ji @xj i i
Noting that =o = J and @J=@xop = 0 (see Equation(2.5)), we obtain

!
@ o @xop + o b = ou#
@xop ji @xj i i
It remains now to replace the term in parenthesis with a quantity that has meaning
in the undeformed conguration.
I: Lagrangian Stress
The simplest approach is to dene a new stress as
o @xo
pi ji @xp
L
j
and substitution of it into the equation of motion yields
@piL o o o
@xop + bi = u#i
The symmetry condition ij = ji is now given by
@xj = L @xi
piL @xo pj @xo
p p
indicating that piL is non-symmetric. Because of the role played by piL we can in-
terpret it as a stress tensor, and it is called the Lagrange stress tensor. Using this
new stress, the equations of motion are relatively simple (indeed they resemble the
Cauchy equations) but the non-symmetric components will complicate the constitu-
tive relation.
. August 2007
II: Kirchho Stress
We would prefer to have a symmetric stress tensor, to that end let us replace the
term in parenthesis with
@xi K o @xop
@xo kj ji @x
j j
from which we have o @xo @xo
pqK = ji @xp @xq
i j
ijK jiK ijK
It is evident that = making a symmetric tensor. Substitute this into the
equations of motion to get
" #
@ @xi K + obo = o u#
i
@xok @xoj kj i
Again, because of the role played by piK we can interpret it as a stress tensor and it is
called the Kirchho
stress tensor or second Piola stress tensor. The Kirchho stress
tensor is symmetric which simplies the constitutive relation, but the equations of
motion are slightly more complicated.
Interpretation of Lagrangian and Kirchho Stresses

In the analysis of stresses in a body with nite deformation, it may be convenient to
use the Lagrangian variables. Since true loading exists only in the deformed state,
the corresponding loading and stress in the body at the initial (undeformed) state
could be considered as ctitious however, we showed that even the Cauchy stress is
a derived (or dened) quantity and therefore we are free to dene other stresses. To
appreciate the motivation in introducing the new denitions of stress, it is worthwhile
to keep the following in mind:
The traction vector is rst dened in terms of a force divided by area.
The stress tensor is dened according to a transformation relation for the trac-

tion and area normal.

To refer to the surface before deformation, a traction vector tô acting on an area
dAo must be dened. The introduction of such a vector is arbitrary and two usual
denitions will be discussed. But rst the Cauchy stress will be reconsidered so as to
motivate the developments.
I: Cauchy Stress
In the deformed state, on every plane surface passing through a point, there is a
traction vector ti dened in terms of the deformed surface area. That is, letting the
3.3. Stresses Referred to the Undeformed State 65
t^
common global axes tô *n
;^

;
; n^ o
;

dA
dAo
e^2 6
undeformed deformed
-
e^3 ;

; e^1
;
;
Figure 3.6: Traction vectors in the undeformed and deformed congurations.
traction vector be t^ and the total resultant force acting on dA be df^ then
dfi
ti dA
Let all the traction vectors and unit normals in the deformed body form two respective
vector spaces. Then the Cauchy stress tensor ij was shown to be the transformation
between these two vector spaces, that is,
ti = jinj
The tensorial property of the Cauchy stress tensor can be established from the quo-
tient rule. Dened in this manner, the Cauchy stress tensor is an abstract quantity
however, on special plane surfaces such as the ones with unit normals parallel to e^1 e^2
and e^3, respectively, the nine components of ij can be related to the stress vector
and thus have physical meaning.
Thus the meaning of ij are the components of stress derived from the force vector
dfi divided by the deformed area. This, in elementary terms, is called true stress.
II: Lagrangian Stress
Let the resultant force dfô referred to the undeformed conguration be identical to
the force df^ acting on the deformed area. That is,
dfio = dfi
The Lagrangian traction vector toi is dened as
toi dA dfio = dfi
o dAo
The Lagrangian stress tensor ijL is dened as the second order tensor which relates
the two vector spaces toi and noi in the following manner:
toi = jiL noj
The meaning of piL are the components of stress derived from the force dfi divided
by the original area. In elementary terms, this is called engineering stress.
. August 2007
III: Kirchho Stress
Let the resultant force dfô referred to the undeformed conguration be given by a
transformation of the force df^ acting on the deformed area. In particular, let
@x o
dfi = @xi dfj
o
j
which follows the analogous rule as for the deformation of line segments
dxoi = @x
o
i
@xj dxj
It is important to realize that this is not a rotation transformation but that the force
components are being `deformed'.
The Kirchho traction vector is dened as
dfio = @xoi dfj
toi dA o @xj dAo
This leads to the denition of the Kirchho stress tensor ijK :
toi = jiK noj
Formally, this is the same as for the Lagrangian stress, but recall that the components
of force are dierent.
The meaning of pqK are components of stress derived from the transformed com-
ponents of the force vector, divided by the original area. There is no elementary
equivalent to this stress.
Relations Among the Stress Tensors

From the denition of the Lagrangian traction vector, we have
dfi = toidAo = tidA
Using the Lagrangian and Cauchy stress tensors, the above equation becomes
jiL nojdAo = pi npdA
Since, from Chapter 2, we have the relation between the areas
@xo
np dA = J o @xj noj dAo
p
we obtain
L o o o @xoj o o
jinj dA = J pi @x nj dA
p
3.4. Stress in Special Deformations 67
Thus
@xoj
jiL o
= J @x pi
p
We have J o = o =, thus the relation between Lagrangian and Cauchy stress tensors
can also be written as
o @xo
jiL = @xj pi ji = o @x
@xj L
o pi (3.3)
p p
Note that, in general, ijL is not a symmetric tensor.
The relation between the Kirchho stress tensor and the Cauchy stress tensor is
derived in a similar manner. We get
o @xo @xo
ji = @x i @xj mn
K ji = o @x
@xi @xj K
o @xo mn (3.4)
m n m n
The matrix form of the relation between the Cauchy and Kirchho stress is
@xo
ijK ] = J @xoi ] 1 mn ] oj ] 1T
; ;
ij ] = 1 @xoi ]mn
K ] @xj ]T (3.5)
@xm @xn J @xm @xon
where the meaning of @xi =@xoj ] 1 is the ij th component of the inverse of the matrix
;
@xp =@xoj ].
Note that the Kirchho stress tensor is a symmetric tensor. Since the strain
tensors we introduced are symmetric, it is more convenient for us to use the Kirchho
stress tensor in the formulation of the stress-strain laws.
3.4 Stress in Special Deformations

The description of some special deformations was given in the previous chapter. It
is the purpose of this section to investigate the corresponding stresses. The three
fundamental cases considered are that of simple extension, simple shear, and a state
similar to bending. In the rst, the magnitude of the areas change but not their
orientation. In the second and third, both the magnitude and orientation of the areas
change.
Simple Extension
The deformation corresponding to a uniform extension is given by
x1 = 1xo1 x2 = 2 xo2 x3 = 3xo3
The unit cubic solid in Figure 3.7 is subjected to simple extension where the applied
load is acting in only one direction. In this particular instance, the stretches are
1 = 2 2 = 3 = 0:5
. August 2007
For this problem, the basic information is given in terms of the forces and so the
stresses will be established by using the connection between them, the tractions, and
the stresses.
1
2
6 1
- df^ = 100
-
1
Figure 3.7: Cube with uniaxial load.
The deformation gradients are given by
2 3 2 3
2 0 0
@xi ] = 6 0 :5 0 7 @xo : 5 0 0
@xi ] 1 = 6 0 2 0 7
@x o 4 5 @xp ] = @xo
;
4 5
p 0 0 :5 i p 0 0 2
The Jacobian is therefore
J o = 2 :5 :5 = :5

which shows that there is a volume change.

I: Cauchy Stress
The Cauchy stress tensor is obtained from information about the traction vectors
ti = jinj df i
dA = 1i n1 + 2i n2 + 3in3
On the x1 -face n^ = e^1 , giving ti = 1i , and the components of force and area are
therefore
dfi = 100 0 0 :
f g dA1 = 2 3 = 0:5 0:5 = 0:25
dA2 = 0 dA3 = 0
On the x2 -face and x3 -face, we have ti = 2i , and ti = 3i , respectively, and in both
cases
dfi = f0 0 0g
Thus, for the respective faces, we have
n^(1) = f1 0 0g : df1 = 100 = 400 = = 0
11 = dA :25 12 13
1
n^(2) = f0 1 0g : 21 = 23 = 22 = 0
(3)
n^ = f0 0 1g : 31 = 32 = 33 = 0
In summary, the components of the stress tensor are
2 3
400 0 0
ij ] = 64 0 0 0 75
0 0 0
The components of the other stress denitions will be obtained by using this and the
deformation gradients.
II: Lagrange Stress
Convert the Cauchy stress to Lagrangian stress by
o @xo o " @xo @xop @xop
#
L p p
pi = ij @x = i1 @x + i2 @x + i3 @x
j 1 2 3
Since J o = o= = 12 then the components of Lagrange stress are now obtained by
substitution in the above expression. For example,
11L = 21 400 21 + 0 0 + 0 0] = 100

12L = 12 0 12 + 0 0 + 0 0] = 0

In summary, 2 3
100 0 0
piL ] = 64 0 0 0 75
0 0 0
Note that its magnitude is signicantly dierent from that of the Cauchy stress.
III: Kirchho Stress
Convert the Cauchy stress to Kirchho stress by
o @xo @xo
pq = ij @xp @xq
K
i j
@xo @xo @xo @xo @xo @xo
= J o 11 @xp @xq + 12 @xp @xq + 13 @xp @xq
1 1 1 2 1 3
@xop @xoq @xop @xoq @xop @xoq
+ 21
@x2 @x1 + 22 @x2 @x2 + 23 @x2 @x3
@xop @xoq @xop @xoq @xop @xoq
+ 31
@x3 @x1 + 32 @x3 @x2 + 33 @x3 @x3
Since only 11 6= 0, we have simply
@xo @xo
pqK = J o 11 @xp @xq
1 1
. August 2007
and this gives, for instance,
11K = 21 (400)( 12 )2 = 50
22K = 21 (400)(0)2 = 0
33K = 12 (400)(0)2 = 0
In summary, 2 3
50 0 0
pqK ] = 64 0 0 0 75
0 0 0
Since the original area is dAo = 1 and the deformed area is dA = 0:25 then
the Cauchy stress has the interpretation of force divided by deformed area while the
Lagrange stress has the interpretation of force divided by original area. The Kirchho
stress is the distorted force dfio divided by the original area dAo = 1.
Simple Shear
Consider the simple shear deformation given by
x1 = xo1 + kxo2 x2 = xo2 x3 = xo3
and let the material have the following simple constitutive behavior
ij = 2eij
where is a modulus. We now wish to obtain the Lagrange and Kirchho stresses.
I: Cauchy Stress
The Cauchy stress will be obtained by use of the constitutive relation, and the others
will then be obtained by transformations of it. The deformation gradients are
2 3 2 3
1 k 0 @x o 1 ;k 0
@x
@xi
p
o ] = 64 0 1 0 75 and p ] = @xp ] 1 = 64 0 1 0 75
@xi @xoi
;
0 0 1 0 0 1
Note that there is no volume change since J = Jo = 1: The Eulerian and Lagrangian
strain tensors are, respectively,
2 3 2 3
@xop @xop 6 0 k 2 0 7 0 k 0
2eij = ij ; = 4 k ;k 0 5 2Eij = @xpo @xpo ; ij = 64 k k2 0 75
@xi @xj 0 0 0 @xi @xj 0 0 0
The Cauchy stress tensor, therefore, is
2 3 2 3
0 k 0 0 1 0
ij = 64 k ;k2 0 75 = k 64 1 ;k 0 75
0 0 0 0 0 0
The negative 22 component may seem counter-intuitive but note that it is inherited
directly from the e22 component of strain.
II: Lagrange Stress
The Lagrangian stresses are obtained from
o @xo @xo @xo
piL = ij @xp = i1 @xp + i2 @xp
j 1 2
giving, for example,
@xo1 + @xo1 = 0 + (k)(;k) = ;k2
11L = 11 @x 12 @x
1 2
@xo @x o
12L = 21 @x1 + 22 @x1 = (k)(1) + (;k2)(;k) = k + k3
1 2
0
21L = 11 @x 2 + @x2 = 0 + (k)(1) = k
o
@x1 12 @x2
The complete stress tensor is
2 3
;k 1 + k2 0
piL = k 64 1 ;k 0 75
0 0 0
Note that this stress tensor is not symmetric. It is also worth noting that in contrast
to the Cauchy stress, this tensor has a non-zero component 11L .
III: Kirchho Stress
The Kirchho stresses are obtained from
o @xo @xo
pq = ij @xp @xq
K
i j
@xo @xo @xo @xo @xo @xo
= 12 @x @x + @x @x + 22 @xp @xq
p q p q
1 2 2 1 2 2
Some particular evaluations are
11K = k (1)(;k) + (;k)(1)] + (;k2)(k2 ) = ;2k2 ; k4
12K = k (1)(1) + (;k)(0)] + (;k2 )(1)(;k) = k + k3
The complete stress tensor is
2 3 1 + k2 0 3
;2k ; k
pqK = k 64 1 + k2 ;k 0 75
0 0 0
Note that the Kirchho stress tensor is symmetric in contrast to the Lagrangian stress
tensor.
The magnitude of the shear deformation is governed by the parameter k. It is
worth noting that when it is small, all three stress tensors approach the same values.
Another point worth noting is that the simple constitutive relation ij = 2eij in the
Eulerian variables does not lead to an analogous simple relation between ijK and Eij .
. August 2007
Meaning of Forces in Simple Shear
Continuing the discussion of the simple shear deformation problem, it is of interest to
investigate the nature of the forces present and the areas they act on during a simple
shear. The approach taken is to use the denition of the traction vector in terms of
forces and areas, and the relation between it and the various stress components.
df2
6 = ;
k2 dA
df-1 =
kdA
2
6 2n
6df2 =
kdA
1n -
df1 = 0
-
1
Figure 3.8: Cauchy forces acting on the deformed areas.
For the Cauchy stress, in general,
ti = jinj = df
dA
i
On face 1, n^ = f1 0 0g so that
dfi = n + n + n =
dA 1i 1 2i 2 3i 3 1i
giving the various tractions as
df1 = = 0 df2 = = k df3 = = 0
11 12
dA dA dA 13
The forces themselves are
df1 = 0 df2 = k dA df3 = 0
On the other face, n^ = f0 1 0g, giving the tractions as
df1 = k df2 = k2 df3 = 0
dA dA ;
dA
and the forces as
df1 = k dA df2 = k2 dA
; df3 = 0
We interpret df2 as a compressive force needed to keep the 2-face from de"ecting in
the 2-direction. The forces acting on the faces of the deformed body is shown in
Figure 3.8.
We have for the Lagrange traction vector,
tLi = piL nop = dAdfi
o
To use this, it is rst necessary to determine the orientation of the faces and this is
done easily by considering the deformation of the areas. Recall that the deformed
and undeformed areas are related by
nopdAo = o @x @xi n dA
o i
p
On face 1, n^ = f1 0 0g so that
nopdAo = @x @x1 dA = f1 k 0gdA
o
p
Consequently, the direction and magnitude of the undeformed area are, respectively,
1 k
o
np = p p 0 and dA o = p1 + k2 dA
1 + k2 1 + k2
On face 2, n^ = f0 1 0g so that, as above,
nopdAo = @x @x2 dA = f0 1 0gdA
o
p
giving the direction and magnitude as
nop = f0 1 0g dAo = dA
Now that we know the areas, the forces can be obtained. For example, on face 1
dfi = L no + L no + L no
dAo 1i 1 2i 2 3i 3
giving
df1 = ;k2 p 1 ; k p k + 0 = 0
dAo 1 + k2 1 + k2
df2 = (1 + k2) p k ; k2 p k + 0 = p k
dAo 1 + k2 1 + k2 1 + k2
Therefore, the forces themselves are
df1 = 0
df2 = p k 2 dAo = p k 2 1 + k2 dA = k dA
p
1+k 1+k
. August 2007
6df2 = ;
k2 dAo = ;
k2 dA
df
-1
=
kdAo =
kdA
2
k dAo =
kdA
2n B
6 B df2 = p
B B 6 1 + k2
B 1n B -
B B
B B df1 = 0
B B
-
1
Figure 3.9: Lagrange forces acting on the undeformed areas.
Similarly, on face 2
dfi = L
dAo 2i
giving for the tractions and forces
df1 = k or df = kdA o = kdA
dA o 1
df2 = ;k2 or df = ;k2dAo = ;k2dA
dAo 2
The forces on both faces are the same as for the Cauchy stress this is expected from
the denition of Lagrangian stress. Note that the components dfi on the 1-face are
with respect to the coordinate directions and not with respect to the rotated face.
The traction vector associated with the Kirchho stress is
dfio
tKi = piK nop = dA o
The orientation of the faces in the undeformed conguration is the same as obtained
above in the Lagrange analysis. Hence on face 1
dfio = K no + K no + K no
dAo 1i 1 2i 2 3i 3
giving
df1o = ;k2 (2 + k2 ) p 1 + k(1 + k2) p k + 0 = ; p k2
dAo 1 + k2 1 + k2 1 + k2
o
df2 = k(1 + k2 ) p 1 ; k2 p k + 0 = p k
dAo 1 + k2 1 + k2 1 + k2
Therefore, the forces can be written as
2
df1o = p;k 2 dAo = ;k2 dA
1+k
df2o = p k 2 dAo = kdA
1+k
On face 2,
dfio = K
dAo 2i
giving
df1o = k(1 + k2) or df o = k(1 + k2 )dAo = k(1 + k2)dA
dAoo 1
df2 = k2 or df2o = k2 dAo = k2 dA
dAo ; ; ;
It is seen that the 1 and 2 face forces (as a system) are dierent from both the Cauchy
and Lagrange systems.
6df2 = ;
k2 dAo = ;
k2 dA
df 2 o 2
2 -1 =
k (1 + k )dA =
k (1 + k )dA
2n B df = p
k 2 dAo =
kdA
6 B
B B 6 2 1+k
B 1n B -
B B
df = p;
k2 dAo = ;
k2dA
B B 1
B B 1 + k2
-
1
Figure 3.10: Kirchho forces acting on the undeformed areas.
Note that the components dfi on the 1-face are with respect to the coordinate
directions and not with respect to the rotated face. The normal and tangential forces
on the rotated face are
dfno = df1ono1 + df2ono2 = 0 p
dfto = df1ono2 + df2ono1 = kdAo = k 1 + k2dA
;
This is analogous to the Cauchy system in that there is a zero normal component and
a tangential component that is k area.
Stresses in Bending
Consider the following plane inhomogeneous deformation
x1 = R xo2 ] sin(xo1=R)
; x2 = R R xo2] cos(xo1 =R)
; ; x3 = xo3
where R is a positive parameter. Let the constitutive behavior be given by the linear
relation
ijK = 2G Eij + ij Ekk
. August 2007
pp p p pp pp p p p ppp ppp pp bppppp ppp pp pp p pp p pp pp p p p p
ppp p pp p p p p p p x p pxp p pp p ppp a p ppp

0
after
a : xo1 = R=2
b : xo1 = R
6o2
d 2
pp p p p p p
p
pppp p pp p p p p p p
0
before
ppp pppp p a b
xo1 x-1
Figure 3.11: Shape before and after deformation.

where G and are moduli. We are interested in determining the Cauchy stress.
This deformation is shown in Figure 3.11. Note that initially horizontal lines
become arcs of concentric circles, while initially vertical lines become radial lines
emanating from a common point whose location changes with the deformation. The
deformation gradient is given by
" # 2 (R ; xo2 ) cos(xo1 =R)=R ; sin(xo1 =R) 0 3
@xi = 64 (R ; xo ) sin(xo =R)=R cos(xo =R) 0 75
@xoj 2 1 1
0 0 0
The Jacobian is the determinant of the deformation gradient matrix and on multipli-
cation simplies to
J o = 1 ; xR2
o
We note that as long as xo2 < R that the volume remains positive.
The Lagrangian strains are
E11 = ; xR2 + 12 ( xR2 )2
o o
others: Eij = 0
The Kirchho stresses are therefore
11K = (2G + )E11 22K = E11 others: ijK = 0
We can think of these stresses as acting on the undeformed conguration. The trac-
tions on a surface xo1 = constant are parabolic and independent of the position xo1 in
the limit of small xo2 , however, they have the familiar linear distribution of a beam in
bending. This situation therefore resembles a beam in pure bending. The stress 22K
would give rise to a normal traction on the lateral surface what this implies is that
the given deformation could be achieved only with the aid of additional tractions on
the lateral surfaces.
Since 11K and 22K are the only non-zero stresses, we have for the Cauchy stresses
@x @x
i j K @xi @xj K
ji = @xo @xo 11 + @xo @xo 22
o 1 1 2 2
This leads to three non-zero components of the Cauchy stress
1
11 = R2 (R ; x2) C 11 + S 22 = J oC 2 11K + J1o S 2 22K
o 2 2 K 2 K
o

22 = R12 (R ; xo2)2 S 211K + C 2 22K = J oS 2 11K + J1o C 2 22K
o

12 = R2 (R ; x2) SC11 ; CS22 = J o CS11K ; J1o CS22K
1 o 2 K K
o
where C cos(xo1=R) and S sin(xo1=R). These stresses exhibit a rather complex
dependence on both xo1 and xo2 .
The presence of the non-zero shear stress is, perhaps, a bit surprising. Keep in
mind, however, as xo1 is changed, that the components ij are not necessarily oriented
with respect to the deformed cross-section. It is instructive, therefore, to consider the
components of the Cauchy stress with respect to the deformed cross-section. Consider
a n area whose normal is initially horizontal, then after deformation the normal has
the orientation
n1 = cos(xo1=R) = C n2 = sin(xo1 =R) = S
We now transform the stress components to get
nn = 11 C 2 + 22 S 2 + 212 CS = J o11K
tt = 11 S 2 + 22 C 2 212CS = J1o 22K
;
tn = (11 22 )CS + 12 (C 2 S 2 ) = 0

; ; ;
Thus, the Cauchy stress components with respect to line preserving orientations show
a close connection to the Kirchho stress. Indeed, if we consider the case when xo2
R
(that is, it is like a very slender beam) but we still allow the large de"ections, then
we get
J o = = R1 (R ; xo2 ) 1
o
leading to
nn 11K tt 22K
There are many practical problems where the de"ections are large but the strains
small, under those circumstances it is found useful to invoke the above approximate
relation between the Kirchho and Cauchy stresses.
. August 2007
Exercises
3.1 Given that ti = jinj where ti and nj are rst order tensors, prove that ij is
a second order tensor.
3.2 The concept of `complementary shear' says that the shear stresses on two per-
pendicular faces are equal. Given that ti = ji nj with ij being symmetric,
show that the tractions on two arbitrary faces passing through a common point
are related by
t(1) (2) (2) (1)
i ni = ti ni
3.3 The normal component of the traction vector is given by
N = tini = jinj ni
Find the directions that make this component an extremum. Note that the
normal vector must satisfy n21 + n22 + n23 = 1.]
3.4 Determine the principal values and their orientations for the stress tensor
22 1 13
ij ] = 4 1 2 0 5
1 0 2
3.5 Consider the following components of a stress tensor
21 2 03
ij ] = 4 2 3 0 5
0 0 0
Determine the components of the traction vector with respect to an area rotated
about the x3 axis. Determine the components of stress transformed an angle
about the same axis. How do the above compare or are they related ?
3.6 Under what circumstances (if any) is the following symmetric stress eld in
static equilibrium ?
11 = 3x1 +g2(x2 ) 22 = 4x2 +g1 (x1 ) 12 = 21 = a+bx1 +cx21 +dx2+ex22 +fx1x2
3.7 Under what circumstances (if any) is the following symmetric stress eld in
static equilibrium with no body forces ?
11 = 3x1 22 = 4x2 12 = Ax1 + cx2
Establish the tractions on the three sides of the triangular body (0 0) (1 0) (0 1).
Is it in overall equilibrium ?
3.8 A circular bar in torsion has a symmetric stress eld given by
13 = ;Gx2 23 = Gx1
with all others being zero. Are the stresses in equilibrium? Determine the
tractions on a boundary surface whose normal is
n^ = fcos sin 0g
Exercises 79
3.9 A stress eld is given by
= ;2F x1 x2 22 = ;h 12 = ;h
2 2F x23 2
2F x1 x2 =
11 h r4 r4 r4 21
with all others being zero and r2 = x21 + x22 . Are the stresses in equilibrium?
Determine the tractions along a line x2 = x1 , where is a parameter. What
are the tractions along a curve x21 + x22 = R2 = constant.
3.10 The stress in a continuum is given as
2 3x x 5x2 0 3
1 2 2
ij ] = 4 5x22 0 2x3 5
0 2x3 0
What form must the body forces take if equilibrium is to be satised every-
where?
3.11 Consider the simple shear deformation
x1 = xo1 + kxo2 x2 = xo2 x3 = xo3
and the constitutive behavior
Determine the Lagrangian stress and Cauchy stress. Investigate the forces and
the areas they act on.
3.12 Consider the deformation
x1 = R ; xo2 ] sin(xo1 =R) x2 = R ; R ; xo2] cos(xo1 =R) x3 = xo3
and the constitutive behavior
Determine the Lagrangian stress and Cauchy stress. Investigate the forces and
the areas they act on.
3.13 A rigid block has a Cauchy stress 11 only acting on it. If the block is given a
rigid body rotation about the x3 axis (such that 11 moves with it), what are
the new Cauchy stresses? Determine the components of the Kirchho stress
before and after the rotation. Show that the Lagrangian strain tensor is also
invariant to the same rigid body rotation.
. August 2007
Chapter 4
Elastic Materials
Elasticity deals with a particular set of assumptions about the material behavior.
Some of the considerations are the following:
The stress at a point depends on geometric changes that take place only in its
immediate vicinity. This is really a question of scale.

There are no history eects | the present state of strain gives the present stress.
This rules out, for example, plasticity eects.

There is an instant recovery of the original shape when the forces are removed.
This rules out creep, viscoelasticity, and other rate dependent eects.
Temperature changes only cause a change in volume but otherwise do not di-
rectly aect the material parameters.

Two other material properties of relevance are Inhomogeneity and Anisotropy. A
material is inhomogeneous when the properties vary with location, whereas a material
is anisotropic when the properties vary with the coordinate reference frame. A special
case of the latter of considerable interest is that of isotropy (no directional dependence)
because many engineering materials of interest are approximated quite adequately by
it.
4.1 Work and Strain Energy Concepts

The concepts of Cauchy stress and Lagrangian strain can be developed independently
of each other. A constitutive relation involves both of them simultaneously. The
concepts of work and energy also involve stress and strain simultaneously so we nd
it useful to begin our discussion of constitutive theories with a discussion of some work
and energy concepts these will provide a relevant setting for the rational construction
of constitutive theories.
Work
Consider a typical force-de"ection or stress-strain curve as shown in Figure 6.9. The
elasticity requirement is that both the loading and unloading paths coincide.
81
82 Chapter 4. Elastic Materials

; Pi ui P
; 6 xxx
B
A
xx
xx
-
(a) (b) u - y
x
Figure 4.1: Typical elastic behavior. (a) Discretized arbitrary body. (b) Force-
displacement behavior. (c) Stressed innitesimal element showing strain.
The work done by a force at a point is the vector dot product of the force and the
displacement at the point. For example, in terms of our global coordinate system,
d W = P^ du^ = Pxdu + Py dv + Pz dw
where the `hat' indicates a vector. The force is understood to be constant during the
innitesimal displacement du^. When the force moves along a path from State A to
State B , the work done is
ZB ZB ZB
W = d W = P^ du^ = (Pxdu + Py dv + Pz dw)
A A A
The loading curves of Figure 4.1 can be interpreted as a sequence of possible equilib-
rium states as the load (or de"ection) is changed. Thus, A and B are two possible
equilibrium states with dierent load conditions. We use the symbol to signify
that an increment of work is performed in moving from one state to the other | the
change in conguration may be small or large. In the special case when the initial
conguration is the unstressed, unstrained, virgin state we will simply use W without
the for the work done in reaching a certain state.
The systems we are interested in have multiple forces and moments, so we will
generalize the above expression for work to
XZ B ZB
W = Pidui = fP gT fdug (4.1)
i A A
It is understood that Pi and ui are common (pointing in the same direction) com-
ponents of generalized forces and displacements, respectively. Thus the individual
contribution to the work could refer to forces and displacements P3 du3 or torques
and twists T9 d9, and so on.
Strain Energy
Consider an arbitrary sub-volume V of a body, and let it have displacement increments
dui, then the external work increment due to the tractions ti and body force bi is
Z Z
d We = ti duidA + bi duidV
A V
4.2. Elastic Constitutive Relations 83
But ti = ij nj , so substitute and use the integral theorem and equilibrium conditions
to get Z @ui Z
d We = ij d @x dV = ij d (ij + !ij ) dV
V j V
Recalling that the contraction of a symmetric and antisymmetric tensor is zero, then
the term with !ij disappears giving
Z
d We = ij dij dV = d U
V
It is interesting that the work is the product of the Cauchy stress and the small strain
increment and not the Eulerian strain increment as perhaps anticipated. The term
U is the strain energy so that the above relation is the formal equivalence between
the external work done and the internal stored energy.

We would also like to write the work expression in terms of the undeformed con-
guration. Following developments similar to the example in Section 2.5 leading to
Equation (2.11), and using the relation between Cauchy and Kirchho stress,
@xoi @xoj dE
dmn = @x @xm @xn K
ij mn = o o
m @xn @xi @xoj ij
and recalling that the deformed and undeformed volumes are related by dV = dV o o=,
then the internal work term becomes
Z
d U = o pqK dEpq dV o
V
Hence the Cauchy stress / Eulerian (small) strain combination is energetically equiv-
alent to the Kirchho stress / Lagrangian strain combination.
Example 4.1: Show the explicit relation between the strain energy and the com-
ponents of stress and strain.
Using the Lagrangian variables, we have
Z h i
K dExx + K dEyy + K 2 dExy + dV o
d U = o xx
V yy xy
K dEyy or K dExy and that the shear
Note that there are no products such as xx xx
stress multiplies twice the shear strain.
4.2 Elastic Constitutive Relations

The mathematical description of material behavior and its reactions to applied loads
is called a constitutive relation. There are a number of restrictions on the allowable
form of constitutive relations and Reference 22] gives a clear explanation. A primary
concept is that of material objectivity and we begin with a discussion of this concept.
. August 2007
The deformation situations that arise in practice are usually of the type of cases (b)
and (c) of Figure 2.13 (large rotations but small strains) and we utilize this to make
some useful approximations. A second class of elastic materials is represented by the
rubber-like materials these generally exhibit very large strains but are essentially
isotropic. We consider these in the next section.
Material Objectivity
It is generally accepted that material properties should be independent of the coor-
dinate frame of the observer. Hence we would like to use entities in the constitutive
relation so that the frame independence is guaranteed.
As shown in Chapter 1, rst and second order tensors transform according to
f v^ = Q^ (t)] v^

g f g T^ ] = Q^ (t)] T^ ] Q^ (t)]T

(4.2)
where Q ] is an orthogonal tensor that rotates the frame of reference. Requiring
this to be true for time dependent rigid motions make the tensors objective. Not all
tensors are objective for example, consider the velocity and acceleration obtained
from the displacement
f u^
g = Q^ ] f u^ g
f v^_

g = Q^ ] f v^_ g + Q^_ ] f u^ g
f v#^

g = Q^ ] f v#^ g + 2 Q^_ ] f u^_ g + Q#^ ] f u^ g
Accordingly, velocity and accelerations should not be used in constitutive relations.
Parenthetically, the acceleration is not objective is well known and is the reason for
the Coriolis force in mechanics.
To summarize, in essence, material objectivity says that functions and elds whose
values are scalars, vectors, or tensors are called frame indi
erent (or objective) if both
the dependent and independent vector and tensor variables transform according to
Equation (4.2) 22]. Since our concept of elasticity does not include rate eects, then
practically it means that the use of Lagrangian strain and Kirchho stress in the
constitutive relations will automatically satisfy material objectivity.
Hyperelastic Materials
Consider a small volume of material under the action of applied loads on its surface.
Then a straight forward assumption about elastic behavior is that:
For an elastic body, the stress depends only on deformation and not on
the history of the deformation.
This is expressed as
ijK = fij (Ekl ) (4.3)
which means that the nine components of stress are given by nine separate functions
of all the strains. Using a Taylor series expansion in terms of strains will then give a
rather complicated collection of functions and associated material parameters such
relations are not practical to use. This approach has the further disadvantage that
it does not indicate, in a rational way, the terms that can be removed in formulating
relations for special materials.
An alternative assumption about elastic behavior is that:
The work done by the applied forces is transformed completely into strain
(potential) energy, and this strain energy is completely recoverable.
That the work is transformed into potential energy that is completely recoverable
means the material system is conservative. Using material variables (Lagrangian
strain and Kirchho stress), the increment of work done on the small volume is
Z hX i
d We = o ij ijK dEij dV o
V
The potential is comprised entirely of the strain energy U the increment of strain
energy is
Z hX @ U i
d U = d U (Eij ) = o ij @E dEij dV o
V ij
where is the strain energy density. From the hypothesis, we can equate d e and
U W
d , and because the volume is arbitrary, the integrands must be equal, hence we
U
have
@
ijK = @E U
(4.4)
ij
A material described by this relation is called hyperelastic. Note that it is valid for
large deformations and for anisotropic materials however, rather than develop this
general case, we will look at each of these cases separately.
We can recast the above in terms of the deformation tensor instead of the strain
tensor | this is the standard formulation for large strains. That is, the constitutive
relation is written as
@U = 2 @U X @xm @xm
ijK = @E @Cij C ij = m o
@xi @xoj = 2Eij + ij (4.5)
ij
where Cij is called the right Cauchy-Green deformation tensor. This form is com-
pletely equivalent to the one written in terms of the Lagrangian strain but is slightly
more convenient because analytical treatments of rubber elasticity, for example, tend
to use principal stretches rather than strains.
. August 2007
While Relations (4.4) and (4.5) are simple and elegant, they too would lead to
impractical expressions when expanded. As will be demonstrated, however, they
aord a rational way to construct special elastic relations.
Example 4.2: The following quantities were recorded during the large deforma-
tion testing of a uniaxial specimen:
P , the applied force
L=Lo , the unit change of axial length
t W=Wo, the unit change of transverse width.
Establish the relationships necessary to convert this information to stress and strain.
1.5
1.0
o =E K =E
11 11 =E
.
.5
E11 e11
.0
.0 .5 1.0 .0 .5 1.0 1.5 .0 .2 .4
Figure 4.2: The forms of constitutive behavior for the experimental observed be-
. . .
havior. (a) Basic recorded data. (b) Kirchho stress against Lagrangian strain.
(c) Cauchy stress against Eulerian strain.
Following from the examples of Section 2.3, we have that the stretches are
2 = 1 + t = 1 ;
1 = 1 + 3 = 1 + t = 1 ; = 2
where we have introduced ;t = as the ratio of the axial straining to the
transverse straining. The Lagrangian and Eulerian strains in the axial direction are
E11 = + 12 2 e11 = e ; 12 e2 = (1 + )2 1 + 21 ]
The stresses are
K = 1 dP = o
11 = 1 dP = o dP
o
1 dA1 (1 + ) 11 2 3 dA1 (1 ; )
o 2 o dAo1
The stress o can be thought of as the \force over original area," although here it is
introduced solely as a normalizing factor.
As shown in Figure 4.2(a), there are three possibilities for the behavior of o =
P=Ao against L=Lo : it can be concave up indicating hardening, be concave
down indicating softening, or be linear. The corresponding stress-strain curves are
also shown in Figure 4.2. Note that for the range of nonlinear behaviors shown, all
the Kirchho stress/Lagrangian strain relations show softening, whereas the Cauchy
stress/Eulerian strain show hardening. Therefore, whether a material is physically
linear or nonlinear, softening or hardening, is not a denite concept but depends on
the measures used for the stress and strain. Of course, the mechanical problem can
be objectively nonlinear even though the constitutive relation is linear because the
description of the geometry can be nonlinear.
Example 4.3: Contrast the physical response of materials described by linear

constitutive relations.
.20
o =E
.10
.00
.
(a) (b)
-.10
-.20

-1.00 -.75 -.50 -.25 .00 .25 -.25 .00 .25 .50 .75 1.00 1.25 1.50
Figure 4.3: Physical responses for linear constitutive relations. (a) Material.
, .
(b) Spatial.
Consider the uniaxial constitutive relations
material: 11 K = EE spatial: 11 = Ee11
11
where, for simplicity, we let the modulus of both materials be the same. Substituting
the respective expressions for stress and strain leads to
material: o = E(1 + )(1 + 21 ) spatial: o = E( 11;+ )2 1 + 21 ]
These are shown plotted in Figure 4.3.
There are two obvious implications from Figure 4.3: linear constitutive relations
imply highly nonlinear physical behaviors, and the two descriptions are completely
dierent. In addition, both descriptions exhibit instabilities. Consider the spatial
description, for example: as the load is increased, a point is reached where further
load increments cannot be sustained and large deformations ensue. This is an ex-
ample of a limit point instability. It is worth noting that the instability occurs while
the cross-sectional area is still of signicant size. The material description exhibits
an instability in compression.
In subsequent chapters, we will deal with large displacements and rotations, but
relatively small strains. In those cases, we will use a linear constitutive relation and
restrict ourselves to strain levels such that < 0:20 this avoids both instabilities,
and all relations can be reasonably approximated as linear. For structural materials
such as aluminum and steel, these strain levels would have been associated with
gross plastic yielding. The case of rubber elasticity will be given special treatment.
. August 2007
Small Strain Elastic Materials
Reinforced materials are likely to have directional properties and are therefore aniso-
tropic. They are also more likely to have small operational strains. Thin-walled
structures such as frames and shells are likely to have large displacements and rota-
tions but rather small strains so as to operate without plasticity occurring. We take
advantage of this small strain situation to eect a set of material approximations.
Because the strains are assumed small, we can take the Taylor series expansion of
the strain energy density function
X @ U X @ 2 U
U (Eij ) = U (0) + 0Eij + 12 Eij Epq +
ij @Eij ijpq @Eij @Epq 0
By using the Lagrangian strain tensor, the expansion is valid for large de"ections and
rotations but for small strains. Noting that U (0) = 0 and
@Eij =
@Epq ip jq
then get
@ U @ U + X @ 2 U E + = o + X D E +
ijK = @E
ij @Eij 0 rs @Eij @Ers 0 rs ij rs ijrs rs
where ijo corresponds to an initial stress. Because of symmetry in ijK and Ers,
Dijrs reduces to 36 coecients. But because of the explicit form of Dijrs in terms of
derivatives, we have the further restrictions
2U
Dijrs = @E@ @E = @ 2 U = Drsij
ij rs 0 @Ers @Eij 0
This additional symmetry reduces the elastic tensor to 21 constants. This is usu-
ally considered to be the most general small strain elastic material. Note that the
corresponding result starting with Equation (4.3) gives 36 constants.
We can write this relation in the matrix form
K
f g = D ]fE g
K K K K K
f g f11 22 33 12 g
T fE g fE11 E22 E33 2E12 g
T
where D ] is of size 6 6]. Because of the symmetry of both the stress and strain,
we have D ]T = D ]. Special materials are reduced forms of this relation.
A later section considers material symmetries in detail, here we discuss just a
couple of examples. An orthotropic material has three planes of symmetry and this
reduces the number of material coecients to nine and the elastic matrix is given by
2 3
D11 D12 D13 0 0 0
66 D12 D22 D23 0 0 0 77
66 D D D 77
D ] = 66 013 023 033 1 (D 0; D ) 0
0
0
0
77
66 2 11 12
1 (D ; D )
77
4 0 0 0 0 2 11 12 0 5
0 0 0 0 0 1 (D ; D )
2 11 12
For a transversely isotropic material this reduces to ve coecients because D55 = D44
and D66 = (D11 ;D12 )=2. A thin ber-reinforced composite sheet is usually considered
to be transversely isotropic 17]. A point worth noting is that the material coecients
are given with respect to a particular coordinate system. Hence, we must transform
the coecients into the new coordinate system when the axes are changed.
For the isotropic case, every plane is a plane of symmetry and every axis is an
axis of symmetry. It turns out that there are only two independent elastic constants,
and the elastic matrix is given as above but with
D11 = D22 = D33 = + 2 D12 = D23 = D13 =
The constants and are called the Lam&e constants. The stress-strain relation for
isotropic materials (with no initial stress) are usually expressed in the form
X X K
ijK = 2Eij + ij k Ekk 2Eij = ijK ; 3 + 2 ij k kk (4.6)
The small deformation version of this relation is called Hooke's law and this ver-
sion using Lagrangian strain and Kirchho stress is sometimes referred to as the
St. Venant-Kirchho
law 8] we will refer to both of them as simply Hooke's law.
The expanded form of the Hooke's law for strains in terms of stresses is
Exx = E1 xx K ; ( K + K )]
yy zz
Eyy = E1 yyK ; (zzK + xx

K )]
Ezz = E1 zzK ; (xxK + K )]

yy (4.7)
2Exy = 2(1E+ ) xy K 2Eyz = 2(1E+ ) yzK 2Exz = 2(1E+ ) xz K
and for stresses in terms of strains

K =
xx E
(1 + )(1 ; 2 ) (1 ; )Exx + (Eyy + Ezz )]
yyK = (1 + )(1 E (1 ; )Eyy + (Ezz + Exx)]
; 2 )
zzK = (1 + )(1 E (1 ; )Ezz + (Exx + Eyy )] (4.8)

; 2 )
K =
xy E 2E K = E 2E K = E 2E
xy yz
2(1 + ) yz 2(1 + ) xz 2(1 + ) xz
where E is the Young's modulus and is the Poisson's ratio related to the Lam&e
coecients by
E = (3 ++2) = 2( + ) = (1 ; 2E )(1 + )
=G= E
2(1 + )
. August 2007
The coecient = G is called the shear modulus.
Viewing the relation between the normal components of stress and strain as form-
ing a 3 3] matrix of the material parameters, then it can be inverted only if the
determinant remains positive. The determinant is
det = (1 ; 2 )(1 + )2 > 0
hence we conclude that
;1 < < 0:5
The same conclusion can be drawn by considering the strain energy function 22]. A
negative Poisson's ratio would indicate a material that, under uniaxial tension, would
expand in the transverse direction this is possible for some of the structured materials.
A Poisson's ratio of 0:5 would indicate an innite bulk modulus or very little volume
change for a given stress level. This is sometimes referred to as incompressibility note,
however, that = 0:5 is not the incompressibility condition under large deformations
as we discuss shortly for rubber-like materials.
Example 4.4: Particularize Hooke's law to the case of plane stress.

A special case that arises in the analysis of thin-walled structures is that of plane
stress. Here, the stress through the thickness of the plate or shell is approximately
K 0, K 0, and K 0. This leads to
zero such that zz xz yz
Ezz = ;E xx
K + K ] = ; E + E ]
yy 1 ; xx yy
Substituting this into the 3-D Hooke's law then gives
Exx = E1 xx
K ; K ] K =
xx E E + E ]
yy (1 ; 2 ) xx yy
Eyy = E1 yy
K ; K ] K=
yy E E + E ] (4.9)
xx (1 ; 2 ) yy xx
The shear relation is unaected.
Example 4.5: Show how temperature can aect the constitutive relation.
A temperature change can aect the constitutive behavior in two ways: rst, it
can change the values of the material coecients and second, it causes a volumetric
expansion. We are only concerned here with the latter eect | this is called thermoe-
lasticity. Because the temperature change only causes a volume change, then only
the normal strain components are aected and the Hooke's law of Equation (4.7) is
modied to
E = 1 K ; ( K + K )] + T
xx E xx yy zz
Eyy = E1 yy
K ; ( K + K )] + T
zz xx
Ezz = E1 zzK ; ( xx
K + K )] + T
yy (4.10)
where is the coecient of thermal expansion and T is the temperature change.
Example 4.6: Recover the bulk behavior from Hooke's law.

The sum of the normal components of stress and strain are related by
X K = E XE
k kk (1 ; 2 ) k kk
Under small deformation situations, the rst sum is three times the hydrostatic
pressure while the second sum is the volume change, hence we can write this in the
form of a bulk pressure-volume relation
;p = 3(1 ;E 2 ) VV = K VV K = 3(1 ;E = 3 + 2
2 ) 3
where K is called the bulk modulus.
Example 4.7: Obtain the isotropic Hooke's law on the assumption that Dijpq is
an isotropic tensor.
Section 1.3 gives a collection of isotropic tensors. Since Dijpq is fourth order, it
must have the form
Dijpq = ij pq + ip jq + iq jp
where the coecients are constants. Because of the symmetry of the stress and
strain tensors, Dijpq must be symmetric in ij and pq which leads to = .
The stress-strain relation then becomes
ijK = Dijpq Epq = ij pq + ip jq + iq jp]]Epq
Perform the contractions using the delta functions and compare with the usual form
of Hooke's law using the Lam%e parameters
ijK = Eppij + 2Eij = 2
Eij + Ekk ij
We therefore conclude that the representation of Dijpq is given by
Dijpq = ij pq +
ipjq + iq jp] (4.11)
This is the form that originally motivated the use of the Lam%e parameters.
Example 4.8: Show how an initial stress state aects the current relation be-
tween an increment of stress and an increment of strain for an isotropic material.
Let the initial stress state ijo be associated with the displacement eld uoi . Fur-
thermore, let the current displacement ui be represented as
ui = uoi + i
. August 2007
where i is the (small) increment of displacement from the current value of uoi . Using
this in the strain-displacement relation allows the total strain to be decomposed as
@ui + @uj + X @uk @uk
2Eij = @xo @xo k @xoi @xoj
j i
oi @uoj X @uok @uok
= @u
@xoj + @xoi + k @xoi @xoj
@i @j X @k @k X @uok @k X @k @uok
+ @x o + @xo + k @xo @xo + k @xoi @xoj + o o
j i i j k @xi @xj
The various collections of terms in parentheses are labeled as follows
Eij = Eijo + ij + ij
Note that ij is an increment of strain from the current conguration but referenced
to the zero conguration. The interaction term ij contains components of both uoi
and i this is the term we are especially interested in.
Let the constitutive relation be
X X
ijK = 2
Eij + ij k Ekk 2
Eij = ijK ; 2
+ 3 ij K
k kk
Then, after substituting for the strains, the stresses are
X X
ijK = ijo + 2
ij + ij k kk + 2
ij + ij k kk
We are interested in taking derivatives of this stress with respect to pq . Since
@k
@xoj ij + !ij
then, for the purpose of dierentiation, we can replace the gradient of i with ij .
K stress, for example,
We now get for the 11
K = o + 2
+ + + ] + 2
X @uok X @uok
11 11 11 11 22 33
k @xo1 k1 + kp @xop kp
with similar expressions for the other components. Let us dene the current tangent
moduli as
K o
ET 11 @ 11 = (2
+ ) + (2
+ ) @u1 (2
+ )1 + E o ]
@11 @xo1 11
h ( o + o + o )g i
= (2
+ ) 1 + 21
f 11
o ;
2
+ 3 11 22 33
The derivatives are taken such that the other strains are kept constant. The eect
of the initial stress is to change the tangent modulus | an increase in stress causes
an increase in modulus. This is the same phenomenon as observed when tuning a
4.3. Finite Strain Isotropic Elastic Materials 93
o 6= 0, then
violin string, say. Suppose the initial stress is uniaxial such that only 11
two of the moduli are
h i
ET 11 = (2
+ ) 1 + 21
f1 ; 2
+ 3 g 11 o
h i
ET 22 = (2
+ ) 1 + 21
f ; 2
+ 3 g 11 o
We see that the material becomes anisotropic due to the stress 11 o.

A way to visualize the above results is to consider a block of material that is
under a quasi-static load state. Now superpose a stress wave disturbance of small
amplitude. The current tangent modulus relates the small increments (due to the
stress wave) of strain to the small increments of stress. Although the material is
isotropic, the stress wave experiences the material as being anisotropic. As an aside,
residual stresses can be detected by monitoring the small changes in wave speed
caused by the small changes in tangent moduli 13].
4.3 Finite Strain Isotropic Elastic Materials

Many structural materials (steel and aluminum, for example) and rubber-like ma-
terials are essentially isotropic in that the stiness of a sheet is about the same in
all directions. We will develop constitutive relations for this special case. We will
emphasize, however, the large strain case of the rubber materials.
Mooney-Rivlin Materials
A way to specify isotropy is to specify that the strain energy is a function of the strain
invariants only that is, U = U (I1 I2 I3), where the invariants are computed by
X X
I1 = k Ekk I2 = 21 I12 ; 21 ik Eik Eik I3 = detEij ]
The stress-strain relation becomes, using the chain rule for dierentiation,
U @I1 + @ @I2 + @ @I3
ijK = @@I @E U U
1 ij @I2 @Eij @I3 @Eij

The derivatives of the invariants with respect to strain are
@I1 = @I2 = I E @I3 = I I E + X E E
@Eij ij @Eij 1 ij ij ;
@Eij 2 ij 1 ij
;
k ik kj
On substituting these into the above constitutive relation, and rearranging, we get
X
ijK = oij + 1 Eij + 2 pEipEpj (4.12)
. August 2007
which is a nice compact relation. The coecients i are functions of only the invari-
ants 11] and have the explicit representation
o = @@I + @@I I1 + @@I I2

U U U
1 = @@I @@I I1
;
U
;
U
2 = @@I U
1 2 3 2 3 3
Although the tensor form of the relation in Equation (4.12) is quadratic, this does
not imply that the stress-strain relation can only be quadratic in the components
since the coecients are arbitrary function of the invariants. Indeed, if (I1 I2 I3) U
is considered to be expanded as a polynomial in the invariants, it is seen how this

form gives an elasticity description with many material coecients.
Before proceeding, we recast the above in terms of the deformation tensor instead
of the strain tensor | this is the standard formulation for large strains. The invariants
of the deformation tensor are computed by
X
I1 = j Cii = C11 + C22 + C33
= X21 + 22 + 23
I2 = 1 2 2 2
ij 2 Cii Cjj ; Cij Cji ] = C11 C22 + C22 C33 + C33 C11 ; C12 ; C23 ; C13
= 21 22 + 22 23 + 23 21
I3 = detCij ] = C11 C22C33 + 2C12C23 C13 ; C11 C232 ; C22C132 ; C33C122
= 21 22 23 (4.13)
When the body is unloaded, the principal stretches i are each unity giving the
invariants as 3, 3, and 1, respectively. Also, in the following we will use the designation
J I3 for the jacobian.
p
Following as above, the stress-deformation relation becomes
ijK = 2 @ (I@C
U 1 I2 I3 ) = 2 @ @I1 + 2 @ @I2 + 2 @ @I3
U U
@I1 @Cij @I2 @Cij @I3 @Cij

U
ij
The derivatives of the invariants with respect to Cij are
@I1 = @I2 = I C @I3 = I C ] 1 (4.14)
1 ij
ij ij
@Cij 3 ij
;
@Cij @Cij ;
where the meaning of Cij ] 1 is the ij th component of the inverse of Cij ]. The stress-
;
deformation relation can be written as

ijK = 0ij + 1 Cij + 2 Cij ] ;1 (4.15)
where the coecients are now given by
0 = 2 @@I + 2 @@I I1
U U
1 = 2 @@I
;
U
2 = 2 @@I I3
U
1 2 2 3
There is no essential dierence between this form and the one using strains.
Let U (I1 I2 I3) be expanded as a polynomial in the invariants as
X p q r
U =
pqr pqr (I1 ; 3) (I2 ; 3) (I3 ; 1)
where the pqr are constants. A reduced form for elastomers can be taken to be
U = 100 (I1 ; 3) + 010 (I2 ; 3) + 200 (I1 ; 3)2 + 300 (I1 ; 3)3 + 001 (I3 ; 1)2
The number of terms retained will depend on the quality and type of experimental
data available for the characterization. Reference 8] suggests not including any I2
dependent terms because the resulting strain energy density function is not polycon-
vex.
Many materials that are capable of sustaining large deformations (such as elas-
tomers and biological materials) usually exhibit incompressibility this can be imposed
by setting I3 = 1 in the energy density expansion. Under these conditions, the rst
term alone in the expansion gives what is called a neo-Hookean material, while the
rst two terms alone is often referred to as a Mooney-Rivlin material. However, as
shown in an example to follow, the energy balance is unaected by the addition of a
pressure, and hence the constitutive relation must be amended to give
ijK = 2 @ U@C
(I1 I2) ; pJ C ] 1
ij
;
ij
where the pressure p is treated as an unknown to be specied by the conditions of
the problem. An example problem to follow shows how it is determined in a plane
stress case.
Example 4.9: Consider a state of hydrostatic stress where the pressure is related
to the Cauchy stress by ij = ;pij . Determine the corresponding Kirchho stresses.
The Kirchho stresses are related to the Cauchy stresses by
X o @xo
@xi ];p ] j ]T = ;pJ X @xoi @xoj T 1
ijK ] = J
mn @x mn @x m @xm ] @xm ] = ;pJ Cij ]
;
m n
The deformation tensors reduce to
@xi ] = ]
@x Cij ] = 2 ij ] Cij ] 1 = ij ]=2
ij
;
o
j
Consequently, the constitutive relation reduces to
ijK = 0 + 1 + 2 ]ij = 2 @@IU + 2 @@IU I1 ; @@IU + @@IU I3 ]ij = ;pJ ij ]
1 2 2 3
This shows that all invariants contribute to the pressure as a result, even if I3 = 1 is
imposed for incompressible materials (so that I3 terms do not appear in the energy
. August 2007
expansion) the constitutive relation still gives a hydrostatic pressure. To be explicit,
when the condition I3 = 1 is imposed, the energy balance is unaected by the
addition of a pressure, and hence the constitutive relation must be amended to give
ijK = 2 @ U@C
(I1 I2 ) ; pJ C ]
ij ;1
ij
where p is treated as an unknown to be specied by the conditions of the problem.
Example 4.10: Particularize the Mooney-Rivlin material constitutive relations

for plane stress under nearly incompressible conditions.
We will work with the principal stretches, hence begin with the deformation
written as
x1 = 1xo1 x2 = 2 xo2 x3 = 3 xo3
where the 's are stretches. The special case of simple extension is described by
2 = 3 = o and is analogous to uniaxial stress.
The deformation gradients are
" # 2 3 " # 2 1 0 0 3
@xi = 4 01 0 00 5 @xoi = 4 1=
0 1=2 0 5
@xoj 2 @xj
0 0 3 0 0 1=3
The Cauchy-Green deformation tensor is
2 2 0 0 3
1
Cij ] = 4 0 22 0 5
0 0 23
Because of near incompressibility, we can set
J = 1 2 3 1 I3 = 21 22 23 1
The deviatoric Mooney-Rivlin material energy function is
U d = 10 (21 + 22 + 23 ; 3) + 01 ( 12 + 12 + 12 ; 3)
1 2 3
The Kirchho stresses are given by (all stress and strain values are principal values)
@ U @i ; 1 p = 1 h2 ; 2 1 i ; 1 p
iiK = @
i @Eii 2i i 10 i 01 3
i 2i
since Eii = 12 2i ; 1]. The Cauchy stresses are related to the Kirchho stresses by
@x ] K ] @x ]T
] = @x ) ii = 2i iiK = i 210 i ; 201 13 ] ; p
o @xo i
Both constitutive relations show a strong nonlinear dependence on the stretches i .
Let the membrane stretching be in the 1 ; 2 plane so that the plane stress
condition species that 33 = 0 from this we conclude that the pressure is given by
p = 2 ; 2 1 ] = 2 1 ; 2 2 2
3 10 3 01 3 10 2 2 01 1 2
3 1 2
where again the incompressibility condition is used. The two Kirchho stresses can
now be written as
K = 2 1 ; 1 ] ; 2 1 ; 2]
11 10 K = 2 1 ; 1 ] ; 2 1 ; 2 ]
41 22 01 4 2 22 10 42 21 01 4 1
1 2
The corresponding Cauchy stresses are
11 = 210 21 ; 212 ] ; 201 12 ; 21 22] 22 = 210 22 ; 212 ] ; 201 12 ; 21 22 ]
1 2 1 2 1 2
since ii = 2i iiK in principal coordinates. An interesting feature of these relations
is that, although the expansion of the strain energy density function is relatively
simple, it nonetheless gives rise to highly nonlinear stress-deformation relations.
Example 4.11: Use the neo-Hookean material model to discuss the ination of
a spherical balloon.
1.5
R = 1 m (39 in)
T = 10 mm (:39 in)
pp pp p p p p p p p p p p p p p p tp pp p
10 = 210 kPa 1.0 Mooney-Rivlin
6
P/Po
p p r p p
;
;

;

;
.5
Neo-Hookean
; - Stretch
? 22 22 ? .0
1.0 2.0 3.0 4.0 5.0
Figure 4.4: In"ation of a spherical balloon. (a) Geometry and free body diagram.
(b) Pressure-stretch response
Let the initial radius and thickness be R and T , respectively then during ina-
tion their current values are
r = 1 R t = 3 T
It is assumed that the balloon is stretching in the local 1 ; 2 plane so that the
thickness direction is 3. The spherical symmetry gives 1 = 2 = and 3 =
1=(1 2 ) = 1=2 because of near incompressibility. Under these conditions, the
plane stress neo-Hookean relation becomes
K = K = 2 1 ; 1 ]
11 = = 2 K = 2 2 ; 1 ]
22 10 6 11 22 11 10 4
. August 2007
Consider a free body cut so that the sphere is split in two then equilibrium
between the membrane stress and the pressure gives
r2po = 2rt 11 or po = 11 2t=r
This is the same relation as given in elementary treatments of thin-walled spherical
pressure vessels. Here, however, we allow for the change of thickness and radius
under pressure and the relation becomes
p = 2 2 ; 1 ] 23 T = 2 1 ; 1 ] 2T
o 10 4 R 10 7 R
This relation is shown plotted in Figure 4.4 and exhibits the interesting behavior
that the pressure peaks at = 71=6 1:38 and thereafter decreases. What this
means is that once the pressure reaches the peak value there is a sudden expansion
of the balloon with a consequent decrease of the internal pressure. This is behavior
regularly observed for toy balloons.
Example 4.12: Discuss the ination of a spherical balloon governed by Hooke's

law.
K = K , the three principal strains
From Equation (4.9), under the conditions 11 22
reduce to
E = E = 1 ; K
11 22 E E = ;2 K
11 33 E 11
The principal stretches are
s r
p p
1 = 2 = 2E11 + 1 = 1 + 2(1E; ) 11
K 3 = 2E33 + 1 = 1 ; 4E 11
K
Solve these to give

s
K= E 2 3 = 1 ; (1 2; ) 21 ; 1]
11 2(1 ; ) 1 ; 1]
and the Jacobian
s q
J = 1 2 3 = 21 1 ; (1 2; ) 21 ; 1] J=0:5 = 21 3 ; 221
The Jacobian relation shows that the local material volume can become zero at
q
1 = (1 + )=(2 ) 1:22
where the numerical value is for = 0:5. This value of maximum stretch is sig-
nicantly smaller than the stretch achieved using the neo-Hookean material of the
previous example. Furthermore, the material is not nearly incompressible even if
0:5.
We conclude that Equation (4.8) is not a good material description to use when
large strains ( > 20%) are expected. Indeed, we will reserve that relation for the
case of large displacement and rotations but small strains, the situations that prevail
for thin-walled structures.
Nearly Incompressible Materials: rubber-like behavior
The imposition of pure incompressibility can cause numerical diculties. Addition-
ally, it seems that for rubber-like materials it is a better approximation, anyway, to
assume that they are only nearly incompressible materials (the bulk modulus is sev-
eral orders of magnitude larger than the shear modulus). For these materials, it is
usual then to split the stored energy function into an isochoric or deviatoric part (no
volume/density change) and a volumetric part. We demonstrate this modication
here, the material discussed is given deeper coverage in References 4, 7, 8].
In the previous developments, the Jacobian of the deformation is given by
J = I31=2
The incompressibility condition under large deformations is that J = 1 and not
= 0:5 as shown earlier. As shown in a previous example, all invariants contribute
to the pressure as a result, even if I3 = 1 is imposed then the constitutive relation
gives a hydrostatic pressure.
Since the invariants I1 and I2 contribute to the pressure, we rst introduce mod-
ied stretches given by
i = i=I31=6

J = 1 2 3 = 1

leading to reduced invariants dened as

J1 I1 =I31=3 = I1=J 2=3
J2 I2 =I32=3 = I2=J 4=3
J3 I31=2 = J

The strain energy function then takes the form

1
U = U d + K (J3 ; 1) =
2 X (J ; 3)m (J ; 3)n + 1 K (J ; 1)2
2 mn mn 1 2 2 3
where K plays the role of a bulk modulus and the associated term accounts for the
energy due to bulk compression.
The required derivatives are
@J1 = @I1 1 1 @I3 I1 @J2 = @I2 1 2 @I3 I1
@Cpq @Cpq I31=3 3 @Cpq I34=3
;
@Cpq @Cpq I32=3 ;
3 @Cpq I 5=3
3
@J3 = 1 @I3 J3
@Cpq 2 @Cpq J3I31=2
and the derivatives of the invariants with respect to Cij are as given in Equation (4.14).
The stress-deformation relation for the two term Mooney-Rivlin material, for example,
can now be written as
ijK = 0 ij + 1Cij + 3 Cij ] 1 + K J3 1]J3Cij ]
;
;
;1 (4.16)
. August 2007
where the coecients are given by
0 = 210 11=3 + 201 I21=3 1 = ;201 21=3 3 = ; 32 10 I11=3 ; 34 01 I22=3
I3 I3 I3 I3 I3
Equation (4.16) resembles the relations developed earlier, the dierence is that the
compressibility condition J 6= 1 is retained. As a consequence, Mooney-Rivlin mate-
rial, for example, has three material parameters.
As shown in an example to follow, the small deformation modulus is
Dijkl = ; 34 (10 + 01 ) + K ]ij kl + 2(10 + 01)ik jl + il jk ]
The equivalent Lam&e parameters are
eq = ; 43 (10 + 01) + K eq = 2(10 + 01 )
Consequently, this material has three parameters that can be utilized to model actual
material behavior. The pure incompressibility condition is achieved by setting K
very large which eectively reduces the modeling to two parameters. Reference 8]
recommends choosing K so that the eective Poisson's ratio (for small deformations)
is 0:475. That is,
K = 3 +3 2 = 231 1 + ] 20
; 2 ]
= 2(10 + 01 )
This is the form implemented in many FEM codes.
The neo-Hookean material has the relatively simple form
h i
ijK = ij ; 13 I1Cij ] 1 =I31=3 + K J3 ; 1]J3Cij ] 1
; ;
where = 210 is like a shear modulus. If K is chosen as above, then this has
a single parameter. Keep in mind, however, that many authors 4, 8] choose the
Mooney-Rivlin parameters such that
210 = 1 ; ] 201 = ]
where is a parameter varying in the range 0 < < 0:10.
Example 4.13: Determine the tangent modulus for the Mooney-Rivlin material.
The tangent modulus, in general, is given by
@ K @ K
Dijkl = @Eij = 2 @Cij
kl kl
Performing the indicated derivatives then leads to 7]
Dijkl = 1Cij 1Ckl1 + 2 ij Ckl1 + Cij 1 kl ] + 3Cik1 Cjl 1 + Cil 1 Cjk1 ]
; ; ; ; ; ; ; ;
+4 ij kl + 5 ik jl + il jk ] + 6 Cij Ckl1 + Cij 1 Ckl ] + KJ 2 Cij 1 Ckl1
; ; ; ;
4.4. Elastic Symmetries 101
where
1 = 49 10 I11=3 + 169 01 I22=3 + K J ; 1]J 3 = 32 10 I11=3 + 34 01 I22=3 ; K J ; 1]J
I3 I3 I3 I3
2 = ; 43 10 11=3 ; 83 01 I21=3 4 = 401 21=3 5 = ;201 21=3 6 = 83 01 11=3
I3 I3 I3 I3 I3
We can write this relation in the matrix form
f K g = D ]fE g f g f 11K 22K 33K 12K gT fE g fE11 E22 E33 2E12 gT
where D ] is of size 6 6] and populated as
2D 3
1111 D1122 D1123
66 D2211 D2222 D2223 77
D ] = 64 .. .. ... .. 75
. . .
D2311 D2322 D2323
Because of the symmetry of both the stress and strain, we have D ]T = D ].
However, although the material is isotropic in the sense that it is a function of only
the invariants, the moduli are anisotropic under large deformation. This is a severe
example of the initial load interaction case considered earlier.
The small deformation modulus is recovered by setting
I1 = 3 I2 = 3 I3 = 1 Cij = ij Cij 1 = ij
;
to get
Dijkl = ; 34 (10 + 01 ) + K ]ij kl + 2(10 + 01 )ik jl + il jk ]
The equivalent Lam%e parameters are obtained by utilizing the result of Equation (4.11)
to get
eq = ; 34 (10 + 01 ) + K
eq = 2(10 + 01 )
The linear behavior is valid only for very small strains.
4.4 Elastic Symmetries

If the internal composition of a material possesses symmetry of any kind, then symme-
try can be observed in its elastic properties. Structured materials such as composites
and crystals exhibit these special symmetries. We discuss them within the context of
small strain theories.
For the purpose of this discussion, we take the generalized Hooke's law in the form
f g = c ]f g
where c ] is of size 6 6] and is symmetric. The presence of material symmetries

reduces the number of independent constants still further. Such simplications in
. August 2007
the generalized Hooke's law can be obtained as follows. Let x y z be a coordinate
system and x y z be the second system which is symmetric to the rst in accor-
0 0 0
dance with the form of its elastic symmetry. Since the directions of similar axes of
both systems are equivalent with respect to elastic properties, the equations of the
generalized Hooke's law will have the same form in both coordinate systems, and the
corresponding elastic constants should also be identical.
Monoclinic System: one elastic symmetry plane

Let the symmetric coordinate system be x y z with the base vectors
0 0 0
e^1 = f1 0 0g
0 0
e^2 = f0 1 0g 0
e^3 = f0 0 ;1g
The transformation matrix is given by
2 3
1 0 0
ij ] = 64 0 1 0 75
0 0 ;1
The new stress components referred to the primed system are
2 3
xx xy ;xz
ij ] = ] ij ] ]T = 64 yx yy ;yz 75
0
;zx ;zy zz

The strain components in the new coordinate system can similarly be obtained as
2 3
xx xy ;xz
ij ] = 64 yx yy ;yz 75
0
;zx ;zy zz

The elastic symmetry requires that
8 9 8 9
> xx >
0 0
> xx >
>
>
0
yy > > 0 >
> yy > >
>
< zz = > >
< zz > =
c
0 0
> = ]
ij > 2 >
> yz > > > yz >
0 0
>
>
: >
0
xz >

>
> 2 >
: 2xz >
0

xy xy
0 0
By using the above relations this can be rewritten as

8 9 2 38 9
>
> xx > c 11 c 12 c 13 ;c14 ;c15 c16 > xx >
>
> yy > 66 c12 c22 c23 ;c24 ;c25 c26 77 > yy >
>
> 6 7 >
> >
< zz = 6 c13 c23 c33 ;c34 ;c35 c36 7 < zz > =
= 66 77
>
> yz > > 66 ;c14 ;c24 ;c34 c44 c45 ;c46 77 > > 2yz >
>
>
> >
xz > 4 ;c15 ;c25 ;c35 c45 c55 ;c56 5 > 2xz >
> >
: xy c16 c26 c36 ;c46 ;c56 c66 : 2xy
Comparison of this with the general matrix leads to the conclusion
c14 = c15 = c24 = c25 = c34 = c35 = c46 = c56 = 0
The matrix of elastic constants simplies to the form
2
c11 c12 c13 0 0 c16 3
66 c12 c22 c23 0 0 c26 77
66 c c c 0 0 c 77
66 13 23 33 36 7
66 0 0 0 c44 c45 0 777
4 0 0 0 c45 c55 0 5
c16 c26 c36 0 0 c66
Note that the number of independent elastic constants reduces to 13.
Orthotropic System: three orthogonal planes of symmetry

Let x y z be perpendicular to the three symmetry planes, respectively. The or-
thotropy assures that no change in mechanical behavior will be incurred when the
x y z directions are reversed. Following the procedure described previously, the ma-
trix of elastic constants for orthotropic materials assumes the form
2
c11 c12 c13 0 0 0 3
66 c12 c22 c23 0 0 0 77
66 c c c 0 0 0 77
66 13 23 33 7
66 0 0 0 c44 0 0 777
4 0 0 0 0 c55 0 5
0 0 0 0 0 c66
The number of independent elastic constants reduces to 9.
Hexagonal System: transversely isotropic

This system has a plane of symmetry in addition to an axis of symmetry perpendicular
to the plane. Assume that the plane of symmetry coincides with the x ; y plane,
and the axis of symmetry is the z-axis. Thus, any pair of orthogonal axes (x y ) 0 0
lying on the x ; y plane are similar to (x y). Hence, the stress-strain relations with
respect to (x y z ) where z = ;z should remain identical to those with respect to
0 0 0 0
the (x y z) system. In order to satisfy this invariant property, the cij matrix assumes
the following form
2 3
c11 c12 c13 0 0 0
66 c12 c11 c13 0 0 0 77
66 c c c 0 0 0 77
66 13 13 33 77
66 0 0 0 c44 0 0 77
4 0 0 0 0 c44 0 5
0 0 0 0 0 21 (c11 ; c12 )
. August 2007
It is noted that for transversely isotropic solids, there are ve independent elastic
constants.
Isotropic System
For the isotropic case every plane is a plane of symmetry and every axis is an axis
of symmetry. It turns out, there are only two independent elastic constants, and the
elastic constant matrix is given by
2 3
c11 c12 c13 0 0 0
66 c12 c22 c23 0 0 0 77
66 c c c 77
66 13 23 33 1 0 0 0 77
66 0 0 0 2 (c11 ; c12 ) 1 0 0 77
4 0 0 0 0 2 ( c11 ; c12 ) 0 5
0 0 0 0 0 1 (c ; c )
2 11 12
in which
c11 = c22 = c33 = + 2 c12 = c23 = c13 =
The constants and are called the Lam&e constants. The stress-strain relations for
isotropic materials are usually expressed in the form
ij = 2ij + kk ij 2ij = ij (4.17)
3 + 2 kk ij
;
Note that except for the isotropic material, the material coecients are given
with respect to a particular coordinate system. Generally, we must transform the
constants into the new coordinate system when the system is changed.
Linear Elastic Isotropic Materials

Most structural materials are adequately represented by their linear elastic behav-
ior. We therefore summarize this behavior and in the process introduce some of the
common material constants used.
In practice, the elastic constants commonly used for an isotropic material are
K E G and . These four constants are called the bulk modulus, the Young's mod-
ulus, the shear modulus, and the Poisson's ratio, respectively. We shall study some
special states of stress in order to reveal the physical signicance of these engineering
constants and their relation to the Lam&e constants.
I. Hydrostatic Pressure
A state of hydrostatic pressure is given by
ij = pij
;
where p is the pressure. In view of the traction relation
ti = ij nj = ;pij nj = ;pni
we note that on any plane passing through a point, the traction vector is always
perpendicular to the plane. Taking the contraction of the stress, we have
ii = ;3p
From the stress-strain relations, we also have
ii = 2ii + iikk = (3 + 2)kk
Comparison of these leads to
p = ;(3 + 2) 13 kk = ;Kkk = ;K VV
where
K 13 (3 + 2) = 13 kk
kk
is called the bulk modulus. Since kk denotes the volume change, it is obvious that
K measures the rigidity in the dilatational deformation.
II. Simple Tension
A state of simple tension applied in the x1 -direction is characterized by the state of
stress 11 = o , and all other ij vanish. The quantity o is the uniaxial tensile stress.
The ratio 11 =11 is dened to be the Young's modulus E.
The condition that ij = 0 except 11 = o leads to the strain conditions ij = 0
if i 6= j by use of the stress-strain relations. The normal strain 11 is
211 = 11 ; 3 + 2 11 = 2( + )
3 + 2 11
The expression for the Young's modulus is then obtained as
E 11 = (3 ++2)
11
It is noted that in simple tension 22 6= 0 and 33 6= 0. In fact,
h i
22 = 33 = 21 0 ; 3 + 2 11 = ; 2( + ) 11
The ratio
; 22 = 2( + )
11
is called the Poisson's ratio.
. August 2007
III. Simple Shear
Consider a state of simple shear in the x1 ; x2 plane where the only non-vanishing
stress component is 12 = 21 , and the corresponding non-vanishing strain component
is 12 = 21 . From the stress-strain relations, we have
12 = 212
The shear modulus is dened as the ratio of the shear stress to the total angle change
212 , that is,
212 = G
12
In this context, the quantity 212 is often referred to as the engineering shear strain.
Summary
The expanded form of the Hooke's law for strains in terms of stresses is
xx = E1 xx ; (yy + zz )]
yy = E1 yy ; (zz + xx)]
zz = E1 zz ; (xx + yy )]
2xy = 2(1E+ ) xy 2yz = 2(1E+ ) yz 2xz = 2(1E+ ) xz
and for stress in terms of strain
xx = (1 + )(1 E (1 ; )xx + (yy + zz )]
; 2 )
yy = (1 + )(1 E (1 ; )yy + (zz + xx)]

; 2 )
zz = (1 + )(1 E (1 ; )zz + (xx + yy )]

; 2 )
xy = 2(1E+ ) 2xy yz = 2(1E+ ) 2yz xz = 2(1E+ ) 2xz
Strain Energy for Linear Elastic Materials

When the strains are small, we need not distinguish between the undeformed and
deformed congurations. Under this circumstance, let the material obey Hooke's law
and be summarized as
Z Z
U =
1 + + + ] dV = 1 f gT f g dV
2 xx xx yy yy xy xy 2
V V
For a thin plate under plane stress conditions where zz = 0, xz = 0, yz = 0,
this reduces to Z
U =
1 + + ] dV
2 V xx xx yy yy xy xy
Substituting Hooke's law gives the alternative forms
Z
U = 12 E1 xx 2 + 2 ; 2 + 2(1 + ) 2 ] dV
yy xx yy xy
V
Z E
= 12 1 ; 2 2xx + 2yy + 2xxyy + 21 (1 ; )xy 2 ] dV
V
Energy considerations can put a limit on the allowable values for the material prop-
erties as we now discuss.
The strain energy density can be written in terms of the principle strains as
E h2 2 2 i
U =
(1 + )(1 ; 2 ) 1 + 2 + 3 + 2 (1 2 + 2 3 + 3 1 )
If either < ;1 or > 0:5 then the coecient is negative and it would be easy to
identify normal strain states that have negative strain energies. Hence we conclude
that
(1 ; 2 )(1 + ) > 0 or ; 1 < < 0:5
for normal materials. A negative Poisson ratio would indicate a material that, under
uniaxial tension, would expand in the transverse direction. This is possible for some
of the so-called structured materials.
. August 2007
The relation among the elastic constants are presented as follows in tabular form:
in
terms =
=G= E= = K=
of
G G(3 + 2G) (3 + 2G)

+G 2( + G) 3
E C + (E ; 3) C ; (E + ) C + (3 + E )
4 4 6
(1 ; 2 ) (1 + )(1 ; 2 ) (1 + )
2 3
K 3(K ; ) 9K (K ; )
2 3K ; 3K ;
GE G(2G ; E ) E ; 2G GE
E ; 3G 2G 3(3G ; E )
G 2G 2G(1 + G) 2G(1 + G)

(1 ; 2 ) 3(1 ; 2G)
GK 3K ; 2G 9KG 3K ; 2G
3 3K + G 2(3K + G)
E E E E
(1 + )(1 ; 2 ) 2(1 + ) 3(1 ; 2 )
KE 3K (3K ; E ) 3EK 3K ; E
9K ; E 9K ; E 6K
K 3K 3K (1 ; 2 ) 3K (1 ; 2 )
1+ 2(1 + )
p
where C = E 2 + 2 E + 9 2
Exercises 109
Exercises
4.1 Consider the 2-D stress-strain relation
f 11 22 12 gT = C ]f11 22 212 gT
Reduce C ] by imposing isotropy in the x1 ; x2 plane.
4.2 Find the components cijpq in terms of the Lame constants for 3-D isotropic
solids.
4.3 A cube of steel of side 250 mm is loaded with a uniformly distributed pressure
of 200 MPa on the four faces having normals in the x and y directions. Rigid
constraints limit the total deformation of the cube in the z direction to 0:05 mm.
Determine the normal stress, if any, which develops in the z direction.
4.4 A block of aluminum 4 1 0:25 in3 is loaded in the long direction by 1000 lb.
What loads must be added to the narrow side faces in order to prevent their
motion? What is the resultant motion of the other faces?
4.5 Show that the principal directions of stress and strain coincide for an isotropic
material.
4.6 Show, using the Cayley-Hamilton theorem, that the third invariant of strain is
given by
I3 = 13 Eip Epq Eqi ; 13 I13 + I1I2
4.7 Use the previous result to show that
@I3 = I ; I E + E E
@Eij 2 ij 1 ij ik kj
4.8 During the large deformation testing of a uniaxial specimen, the following quan-
tities are recorded:
P : Applied Force
L : Unit change of axial Length
L
Wo : Unit change of transverse Length
W o
Establish the relationships necessary to convert these measured quantities to
Kirchho stress and Lagrangian strain.
. August 2007
Chapter 5
Linear Elasticity Problems
The developments of the last four chapters form the basis of the eld equations of
the theory of elasticity. This chapter shows how these are put in a form for solving
boundary value problems. In the general case, the equations are non-linear and
therefore can only be solved approximately. Therefore, this chapter emphasizes the
formulation of the linear theory.
The stress function approach is a very powerful method for solving plane elasticity
problems. As will be seen, satisfying all the eld equations reduce to nding a single
bi-harmonic function. As a consequence, the main diculty in solving boundary
value problems is in satisfying the boundary conditions and not so much the eld
equations. This problem is further exacerbated if the functional form of the tractions
are not `similar' to the functional form of the boundary geometry. We therefore
have a need to be able to solve our elasticity equations in general coordinate systems
dictated by the geometry of the boundaries. We will illustrate the approach using the
transformation of coordinates to cylindrical coordinates as well as the use of complex
variables approach.
5.1 Reduction to the Linear Theory of Elasticity

In the previous developments of the theory of elasticity, two main sources of non-
linearity arose. They were associated with large deformations and nonlinear material
behavior. We will now show the process of reduction to the linear theory by simpli-
fying both of these.
Non-linear Elasticity Problems

All the essential equations are collected here and put in a form that constitutes the
complete set necessary for a solution. It is emphasized that, at this stage, they are
applicable to large deformations and to general non-linear elastic material behavior.
Equilibrium: " #
@ @xi K + o bo = ou# ijK = jiK
i
@xok @xoj kj i
111
112 Chapter 5. Linear Elasticity Problems
Strain-displacement:
@ui + @uj + @uk @uk
2Eij = @xo @xo @xo @xo
j i i j
Constitutive relation:
@
ijK = @E U
ij
Boundary conditions:
specify ui on Aou
toi = ijK noj on Aot
There are fteen unknowns (three displacements, six strains, and six stresses) and f-
teen eld equations (three equilibrium, six strain-displacement, and six constitutive).
The boundary conditions are necessary since the eld equations, being dierential
equations, will give rise to additional unknown functions after integration. In gen-
eral, closed-form analytical solutions cannot be obtained and approximate numerical
solutions must be resorted to.
Classical (Linear) Theory of Elasticity

We will simplify the equations in an eort to obtain analytical solutions. The primary
assumptions of the linear theory of elasticity adopted here are:
All deformations are small.
The constitutive relation is linear and isotropic.
There are no inertial eects.
The basic equations then reduce to
@ij + b = 0
i
@xj
ij = ji
ij = 1 ( @ui + @uj )
2 @xj @xi
ij = 2ij + kk ij
specify: ui or ti = ij nj for BC
Note that linearizing the equations does not reduce the number of unknowns.
While it is possible to attempt to solve all the eld equations simultaneously, it is
more common to rst reduce the total number of unknowns. Historically two major
formulations have emerged: Displacement and Stress, respectively.
5.1. Reduction to the Linear Theory of Elasticity 113
I: Displacement Formulation | Navier's Equation
The displacements are taken as the basic unknowns, that is, at each point there
are three unknown functions u1 u2 u3. These must be determined subjected to the
constraint that the stresses derived from them are equilibrated.
Using Hooke's law, we write the stress in terms of displacement
h @ui @uj i @uk
ij = @x + + ij
j @xi @xk
and substitute into the equilibrium equations to get
2 ui 2 uk
@x@ @x + ( + ) @x@ @x + bi = 0
k k i k
These are the Navier's equations and are three equations with three unknowns. (While
we have reduced the number of unknowns, we have increased the order of the deriva-
tives.) The boundary conditions must be specied in terms of ui and are
on Au : ui are specied
@u h @u @u i
on At : @xk ni + @x i + @xj nj = ti are specied
k j i
Notice that the second boundary condition is actually a set of inhomogeneous dier-
ential equations.
These equations are dicult to solve directly and so inverse methods are usually
used. A formalization of this process uses the idea of displacement potentials. That
is, using Helmholz's theorem, the displacement vector eld can be decomposed into
a scalar potential and a vector potential in the form
@ + @k
ui = @x @k = 0
ijk
i @xj @xk
If the body force is absent, then the Navier's equations can be expressed as
@ r2 + @ r2 = 0
( + 2) @x ijk
i @xj k
This equation is satised if
2
r = constant
2
r k = constant
Thus, the problem reduces to solving a set of Poisson's equations in the region to
nd particular solutions for the displacement potentials these particular solutions
automatically satisfy the governing eld equations. The complete solution to the
boundary value problem is synthesized from a collection of particular solutions written
as X X
= ann k = bknn
n n
where an and bkn are undetermined coecients. From these, the displacements them-
selves are obtained by dierentiation and the coecients obtained by satisfying the
boundary conditions.
. August 2007
II: Stress Formulation | Beltrami-Mitchell Equations
As an alternative formulation, the stresses are assumed as the basic unknowns. That
is, at each point in the body there are six unknown functions 11 22 33 12 23 31 .
These obviously must satisfy equilibrium. However, there are only three equilibrium
equations, hence, further restrictions must be imposed. These restrictions come from
the requirement that the strains associated with the stresses must be compatible.
Suppose a stress eld is proposed and it is equilibrated. The use of Hooke's law
converts it to a strain eld. Suppose now it is desired to obtain the displacements.
This can be done by integrating the strain-displacement relation
@uj + @ui = 2
ij
@xi @xj
This can be viewed as a system of six independent partial dierential equations for
three unknown ui. Theoretically, only three equations are needed to determine the
displacement elds. If the six strain components ij are arbitrarily assigned, then
multiple values of the displacements would result. For a unique solution in ui, some
restrictions must be placed on the strains ij . By dierentiating the above, we obtain,
for instance 2 3 3
2 @ ij = @ ui + @ uj
@xk @xl @xj @xk @xl @xi @xk @xl
Interchanging subscripts in this relation leads to
@ 2 ij + @ 2 kl @ 2 ik @ 2 jl = 0
@xk @xl @xi @xj ;
@xj @xl ;
@xi @xk
Among the 81 equations given here, some of them are identically satised, and some
of them are repetitions. Only six equations are nontrivial and independent, and in
unabridged notation, these equations are
@ 2 xx = @ h ; @yz + @zx + @xy i
@y@z @x @x @y @z
@ yy = @ ; @xz + @xy + @yz i
2 h
@z@x @y @y @z @x
@ zz = @ ; @xy + @yz + @zx i
2 h
@x@y @z @z @x @y
@ xy = @ xx + @ 2 yy
2
2 @x@y
2
@y2 @x2
2
@ yz = @ 2 yy + @ 2 zz
2 @y@z @z2 @y2
@ zx = @ 2 zz + @ 2 xx
2
2 @z@x @x2 @z2
simply connected multiply connected

g

COMMON GLOBAL AXES

6

-
;

;
;
;
;
Figure 5.1: Simply and multiply connected bodies.
These six equations are known, collectively, as the equations of compatibility, rst
obtained by St. Venant in 1860.
A body is said to be simply connected if every closed curve drawn in the body can
be shrunk to a point without passing out of the body. For example, a hollow sphere is
simply-connected while an open-ended hollow cylinder is a multiply-connected body.
The equations of compatibility are necessary and sucient for a simply-connected
body. For a multiply-connected body, they are necessary but no longer sucient ad-
ditional conditions must be imposed to ensure the single-valuedness of displacement.
To obtain compatibility in terms of stress, use Hooke's law in the form
ij = 1 + ;
E ij E kk ij
to replace the strains in the compatibility equations with stresses and simplify this
by utilizing the equilibrium equations to get
@ 2 ij + 1 @ 2 kk + @bk + @bi + @bj = 0
@xk @xk 1 + @xi @xj 1 ; @xk ij @xj @xi
The stress eld must satisfy this equation and the equilibrium equations
@ij + b = 0
i
@xj
in order to be admissible. The boundary conditions to be satised are
on At : ij nj = ti = given
on Au : ui = given
Note that the second set of boundary conditions are obtained by integrating the
strain-displacement relations in conjunction with the stress-strain relations.
. August 2007
Example 5.1: Analyze Levy's Problem (1898).
As an example of solving an elasticity problem, we will nd the stresses in a
semi-innite wedge of mass density , due to uid pressure of specic weight . We
will use the inverse approach that is, we assume the stresses to be of a particular
form and then determine the coecients from the eld equations and boundary
conditions.
y
6
-x
;@
water ; @
; @
; @
; @
;@R @
;@
; @
R
@
concrete @
@
Figure 5.2: Dam with water and own weight loading.
I. Stress Fields
Since the pressure exerted by the water varies linearly with depth, assume that a
linear expansion for the stresses is sucient, that is,
xx = a1 x + b1 y + c1
yy = a2 x + b2 y + c2
xy = a12 x + b12 y + c12
There are a total of 9 coecients to be determined from
Equilibrium
Compatibility
Boundary Conditions
Let the origin be at the apex of the dam, then
c1 = c2 = c12 = 0
Equilibrium gives
@ xx + @ xy + b = 0 or a1 + b12 + 0 = 0
@x @y x
@ xy + @ yy + b = 0 or a12 + b2 ; g = 0
@x @y y
Hence the stresses reduce to
xx = a1 x + b1 y
yy = a2 x + b2 y
xy = ;b2 x ; a1 y + gx
That is, any values substituted for a1 a2 b1 b2 will give a system of stresses that
satisfy the equilibrium equations.
Compatibility is automatically satised since the stresses are linear functions of
position (hence, so are the strains) and the compatibility equation has only double
derivatives.
II. Boundary Conditions
We now must choose values for the coecients that satisfy our particular boundary
value problem. In general, the tractions on any boundary are related to the stresses
by
tx = xx nx + yxny + zx nz
ty = xy nx + yy ny + zy nz
y
6
-x
@
n^ a
@ n^ b
QQ @
k
;
normal pressure @;
@
@
R@
R @ traction free
a-face @
@ b-face
@
Figure 5.3: Tractions on the dam faces.

For the b-face, we have fnx ny nz g = fcos sin 0g and substituting into the
traction equations gives
tx = 0 = (a1 x + b1y) cos + (;b2x ; a1 y + gx) sin + 0
ty = 0 = (;b2 x ; a1 y + gx) cos + (a2 x + b2y) sin + 0
Along the b-face, x and y are related by
tan = ;xy or x = ;y tan
Substituting for x and canceling the y's gives
;2a1 tan + 0 + b1 + b2 tan2 = 0 + g tan2
a1 + a2 tan2 + 0 + 2b2 tan = 0 ; g tan
Notice that there is no x or y dependence.
For the a-face, we have fnx ny nz g = f; cos sin 0g and x = y tan , giving
the tractions as (noting that the pressure is p = ;y)
tx = p cos = ;y cos ty = ;p sin = y sin
. August 2007
These two equations become
2a1 tan + 0 + b1 + b2 tan2 = ; + g tan2
a1 + a2 tan2 + 0 + 2b2 tan = ; tan + g tan
Again, notice that we do not have x or y appearing in these equations.
Solving all four of these equations simultaneously gives
tan2 + 3 tan tan ; 2 tan ; tan
a2 = (tan + tan )3 ; g (tan + tan )2

a1 = tan tan tan + tan ; a2

b2 = 1 ; tan + a2 (tan ; tan ) ; g
2 tan + tan

b1 = 1 tan2 ; 3 tan + a2 (3 tan + tan ) ; g
2 tan + tan
These coecients can be substituted back to give the stress distributions.
The critical stage of this problem was in satisfying the boundary conditions
we were successful because the functional form of the stresses at the boundaries
coincided with the functional form of the tractions. For example, on the a-face the
relevant functions are linear in x and y. Note, for example, that if the dam surface
had a relation of the form x = 3y2 , say, then we would not have been able to satisfy
the boundary conditions.
Example 5.2: A block rests on a horizontal plane under the action of gravity.
Investigate the stresses and displacements.
a = L=4 b = L=8)
L = 102 mm (4:0 in)
= 0:3
aluminum 10 3 ;
Figure 5.4: Deformation of a block under gravity loading. (a) Geometry and node
positions. (b) Front and side view of deformed shape. (deformations are exaggerated
5000).
Let the coordinate system be as shown in Figure 5.4 with gravity acting vertically.
The body force per unit volume is fzb = ;g. Assume all the stresses are zero except
the vertical stress, then from the equilibrium equations we get
zz = gz ; L] ij = 0
We will discuss the consequences of this assumed stress state on the displacement
eld.
Using Hooke's law and the strain-displacement relations gives
xx = @u
@x = ; g z ; L] ;! u = ; g xz ; L] + u (y z )
E E o
yy = @v
@y = ; g z ; L] ;! v = ; g yz ; L] + v (x z )
E E o
zz = @w
@z = + g z ; L] ;! w = g z2 ; 2zL] + w (x y)
E 2E o
Substitute these into the shear strain relations to get
xy = @u @v @uo @vo
@y + @x = 0 = @y + @x ;! uo = u(z) + y + 1 vo = v(z) ; x + 2
The other shear strain relations give
xz = @u + @w = 0 = ; gx + @ u + @wo ;! w = gx2 ; @ u x + f (y)
@z @x E @z @x o 2E @z
@v @w gy @ v
@w o gy 2 @ v
yz = @z + @y = 0 = ; E + @z + @y ;! wo = 2E ; @z y + g(y)
Since both expressions for wo must be the same and since they are independent of
z, we conclude that
2 gx2 + c x + c
u = c1 z + c2 v = c3 z + c4 f (y) = gy
2E + c5 y + c6 g ( x ) = 2E 7 8
Putting all these together, we get the displacement elds as
u = ; g
E xz ; L] + y + c1 z + 1
v = ; g
E yz ; L] ; x + c3 z + 2
w = + 2gE z2 ; 2zL] + g 2 2
2E x + y ] ; c1 x ; c3 y + 3
The constants are obtained from the boundary conditions.
Let the bottom center of the block be at the origin and let it have zero displace-
ments then i = 0. Furthermore, because of symmetry, the vertical line x = 0, y = 0
remains straight and vertical hence the slopes @u=@z and @v=@z at the origin are
zero. Finally, to suppress rotation about the z -axis set @u=@y ; @v=@x to zero. The
resulting displacement elds are
u = ; g
E xz ; L]
v = ; g
E yz ; L]
w = 2gE z2 ; zL] + g 2 2
2E x + y ]
. August 2007
The most interesting aspect of this solution is that the plane z = 0 does not remain
horizontal in fact, it has the vertical displacement
w = g 2 2
2E x + y ]
To suppress this would require a complicated set of tractions localized to the plane,
in other words, stress concentrations would be induced at the edges.
Figure 5.4(a) shows the nite element mesh using Hex20 elements. The gravity
loads for each node are computed by
Z Z
Pzi = ghi dV o = g hi jJ j dV
vo v
This is computed numerically using full integration. The boundary conditions im-
posed are that there is no vertical displacement at the bottom, the center is xed
and the nodes along x = 0, z = 0 are restrained to move only along the x-direction.
The exaggerated deformed shapes are shown in Figure 5.4(b). Note that there
is no lateral contraction at the top. The normalized displacements are
wmax = 1:00200 vmax = 0:98667
;gL2 =2E gbL=E
These are quite close to the analytical solution. When the base is fully constrained,
the corresponding results are
wmax = 0:9662 vmax = 0:00
;gL2 =2E gbL=E
The constraining eect also aects the stresses with the unconstrained being
xx = ;0:0011 yy = ;0:0028 zz = ;0:9483 4 = ;1:0061
gL gL gL 3:77
compared to
xx = +0:2940 yy = +0:2989 zz = ;0:9901 4 = ;1:0505
gL gL gL 3:77
The stresses are at the integration point closest to the bottom center and the zz is
given an approximate correction to estimate the stress at the base. It is clear that
the constraint generates a stress that is on the order of Poisson's ratio times the
axial stress.
Uniqueness of Solutions
For an elasticity problem, the body force bi and the boundary conditions over the
boundary surface A, are usually prescribed. There are two types of boundary condi-
tions: displacement-prescribed and traction-prescribed. In general, the total bound-
ary surface can be divided into two parts, Au and At over which the displacements
and tractions are prescribed, respectively. A solution to an elasticity problem is one
for which the stresses satisfy the equilibrium equations
@ij + b = 0
i
@xj
the strains are compatible, and the boundary conditions in terms of the tractions
and displacements are satised. This solution is unique in the sense that the state of
stress (and strain) is determinate without ambiguity. By contrast, the nonlinear eld
equations do not lead to a unique solution.
To prove the uniqueness of solution, we start with the assumption that there are
two possible solutions ui and ui to the same problem. Denote the dierence of the
0 00
solution by
ui = ui ui 0
;
00
ij = ij ij 0

;
00
ij = ij ij 0

;
00
Since ij and ij satisfy the same equilibrium equations and boundary conditions, it
0 00
is easy to see that the dierence solution satises

@ij = 0 in V
@xj
and that the boundary conditions are
ui = 0 on Au
ti = 0 on At
Since the strain energy density function is given as
= 21 ij ij (=
U ) 6 U
0
; U
00
we have Z Z Z
1 1
dV = 2 ij ij dV = 2 ij @x
U
@ui dV
V V V j
Integrating by parts, we obtain
Z Z Z
1 @ui dV = 1 @ 1 @ij
2 V ij @xj 2 V @xj (ij ui )dV ; 2 V @xj uidV
Z
= 1 u n dA
2 A ij i j
Z
= 1 t u dA
2 A i i
Z Z
1 t u dA 1
= 2 Au i i + 2 At ti ui dA
. August 2007
Since ui vanishes on Au, and ti vanish on At, we have
Z
U dV = 0
This is possible only if U = 0, which in turn requires ij and ij to be zero. Thus, we
conclude that the only dierence between the two solutions is a rigid body motion
(since this does not contribute to the stresses or strains). Moreover, if the displace-
ments are known on any part of the boundary, then they are determined uniquely in
V and on A.
5.2 Plane Problems

It is very dicult to solve general 3-D problems, and therefore various schemes of
simplication have arisen. One such scheme, which has a lot of practical use, is to
reduce the dimensionality of the problem. In this section, we consider those problems
where the essential behaviors occur in a plane. Chapter 7 considers other types of
reductions of the eld equations in forming approximate structural theories.
Reduction to 2-D Equations

The reduction of a 3-D problem to an equivalent 2-D one involves approximation and
therefore is not valid for general cases. To make explicit the situations of interest,
consider a 3-D body such that:
the body is bounded by two "at surfaces lying in the 1 ; 2 plane,
the cross-section of the body is uniform in the x3 direction,
the load is uniformly distributed in the x3 direction,
there are no shears on the "at faces.
Note that this body does not have to be thin. The second of these restrictions rules
out bending while the third rules out torsion.
Plane Strain Formulation

The plane strain assumption is that u3 = 0. This leads to the zero strains
33 = 0 13 = 0 23 = 0
In summary, the strains and strain-displacement relation become
0 " #

ij = 0 0 1 @u @u
= 2 @x + @x =1 2

5.2. Plane Problems 123
COMMON GLOBAL AXES x2
6
;;
;
x
-1
; ;;
;
;
;
;
;

;x3 ;

; h
Figure 5.5: Solid body bounded by two planes.
The only non-trivial compatibility condition is
@ 2 11 + @ 2 22 = 2 @ 2 12
@x22 @x21 @x1 @x2
Based on the linear, isotropic Hooke's law, the stresses under plane strain condi-
tions must be
ij = 2ij + ij kk
In expanded form, this is
11 = 211 + (11 + 22 + 0)
22 = 222 + (11 + 22 + 0)
33 = 20 + (11 + 22 + 0)
12 = 212
13 = 0
23 = 0
This can be rewritten so as to group the 1 2 terms together as
= 2 + 33 =
with ranging from 1 to 2. Thus, in plane strain problems, the 33 normal strain
is zero, but the corresponding 33 normal stress is not.
The reduced stress tensor is given by
0
ij = 0 =1 2
33
Provided that the body force b3 is zero, the equilibrium equations reduce to
@ + b = 0 = 1 2

@x
Thus we are left with two equilibrium equations.
. August 2007
Approximate Nature of Plane Stress
The plane stress assumption is that if the plate is very thin then the stresses 13 , 23 ,
33 , are zero. If we can impose these conditions then we can simplify our equations
as done for the plane strain case. However, unlike the plane strain case were the
imposition of u3 = 0 leads to an exact two-dimensional formulation, it is not clear
a priori that the above stress state can actually be imposed. This we will rst
investigate.
Let us seek an exact solution under the restriction
13 = 0 23 = 0 33 = 0
This solution must satisfy the equilibrium and compatibility equations, which under
the assumption of no body forces reduce to, respectively,
@ij = 0 @ 2 ij + 1 @ 2 kk = 0
@xj @xk @xk 1 + @xi @xj
In what follows, roman subscripts range from 1 to 3, while greek subscripts range
from 1 to 2.
The compatibility equation with ij = 13 23 33 become
@2' = 0 @2' = 0 @2' = 0 ' 11 + 22
@x1 @x3 @x2 @x3 @x3 @x3
We therefore conclude that the distribution of the sum of the normal stresses is of
the form
' = cx3 + f (x1 x2)
Let us now restrict the stress distributions to be symmetric about the middle plane,
then c = 0 and we have
' = '(x1 x2 )
The normal stresses are only a function of the in-plane coordinates. By setting ij =
11 22 in the compatibility equation and adding, we get
2
r2 ' = 0
2
r2
@2 + @2
@x21 @x22
showing that ' is a harmonic function.
The explicit form for the two equilibrium equations is
@11 + @12 = 0 @12 + @22 = 0
@x1 @x2 @x1 @x2
We can satisfy these equations by introducing a function such that
2 2 2
11 = @@x2 22 = @@x2 12 = ; @x@ @x
2 1 1 2
The function is called a stress function which we will develop in more detail later
in this chapter. Adding the normal stresses, we get
' = r22 2 2 2
r2 r2 = r2 ' = 0
showing that is a bi-harmonic function.

The compatibility equation with ij = 11 is
2
r2 11 +
@ 2 11 + 1 @ 2 ' = 0
@x23 1 + @x21
Noting that
2
r2 11 =
@ 2 r2 = @ 2 ' @2' = ; @2'
@x2 2 @x2
2 2 @x2 1 @x2 2
The compatibility equation can be arranged as
( )
@2 ' + @2 = 0
@x22 1 + @x23
The other two compatibility equations (with ij = 22 and 12, respectively) give
( ) ( )
@2 ' + @2 = 0 @2 ' + @2 = 0
@x21 1 + @x23 @x1 @ 2 1 + @x23
By integrating these we conclude that
' + @ 2 = a + bx + cx
1 2
1+ @x23
where a, b and c are arbitrary constants. Integrating the stress function then gives
(x1 x2 x3 ) = ; 1 + '(x1 x2 ) 12 x23 + a + bx1 + cx2 ] 21 x23 + 1(x1 x2 )x3 + 0(x1 x2)
The coecients a b c do not participate in the stress solution and hence may be
taken as zero. We are imposing that the stresses are symmetric in x3 hence we also
take 1 (x1 x2 ) as zero. We thus have the stress function representation
(x1 x2 x3 ) = 0(x1 x2 ) ; 1 + '(x1 x2 ) 12 x23
where ' is a harmonic function and 0 is a bi-harmonic function.
From the above solution representation, we see that the stresses in our 3-D problem
is comprised of two parts the rst depends on the bi-harmonic function 0 and the
latter on the harmonic function '. This latter solution, being proportional to x23 ,
may be made as small as we please compared to the rst if we restrict the plate to
be suciently thin.
In conclusion, if we impose the plane stress conditions, we can get a solution that
satises equilibrium but the compatibility conditions will be approximated. The error
can be made as minimal as necessary if the plate thickness is restricted to being very
thin. Chapter 7 provides an alternative justication for plane stress theories.
. August 2007
Plane Stress Formulation
Based on the foregoing discussion, we will assume
13 = 0 23 = 0 33 = 0
That is, the stress tensor reduces to
0
ij ) ij = 0 =1 2
33
The equilibrium equations become
@ + b = 0 = 1 2

@x
Thus we are left with two equilibrium equations.
Assume the material obeys the linear, isotropic Hooke's law such that
ij = 2ij + ij kk
In expanded form, this is
11 = 211 + (11 + 22 + 33 )
22 = 222 + (11 + 22 + 33 )
33 = 233 + (11 + 22 + 33 ) = 0
12 = 212
13 = 213 = 0
23 = 223 = 0
In contrast to the plane strain case, we have a non-zero 33 strain component. This
is related to the other normal strains by

33 = 2;+ 11 + 22 )
This can be rewritten so as to group the 1 2 terms together as
!
2
= 2 + 2 +
In summary, the strains and strain-displacement relation become

" #
ij ) ij = 0 1 @u
= 2 @x + @x @u
=1 2
33 33
We assume in our approximation that the only strains participating in the compati-
bility equations are the in-plane strains. That is,
@ 2 11 + @ 2 22 = 2 @ 2 12
@x22 @x21 @x1 @x2
When the body is only somewhat thin, the resulting case is referred to as Generalized
Plane Stress.
Summary of 2-D Plane Elasticity Problems
The basic unknowns are
2 Displacements : u1 u2 or u =1 2
3 Strains : 11 22 12 or =1 2
3 Stresses: 11 22 12 or
Note that 33 in plane strain and 33 in plane stress, are not part of the basic unknowns
but are obtained after the solution. The appropriate eld and material relations are
h @u i
Strain-Displacement: = 12 @u @x + @x
2 2 2 12
Compatibility: @@x112 + @@x222 = 2 @x@ @x
2 1 1 2
@
Equilibrium:
@x + b = 0
Stress-Strain: = 2 + 23;;2
] = 2 +
]
= 21 ; 3 ;4
] = 21 ;
]
Plane Strain: = 3 ; 4 = 1 ; 2 =
33 = 0
33 = (11 + 22 ) = (11 + 22 )
Plane Stress: = 31 ;
= =
+ 1+
33 = 0
33 = 2;+ (11 + 22 ) = ; E (11 + 22 )
Substituting the strains in terms of stresses into the compatibility equations, using
the equilibrium equations and rearranging gives
!
2
r (11 + 22 ) = ;
4 @b 1 + @b2
(1 + ) @x1 @x2
This is called the Beltrami-Mitchell equation. Note that if no body forces are present
then
2
r (11 + 22 ) = 0
That is, the rst invariant of stress is an harmonic function.

. August 2007
Cylindrical Coordinates
Some, but not all, entities follow the usual transformation law with a change of co-
ordinate system, For example, in the cylindrical coordinates (r
z), the strain com-
ponents rr zz , rz r z are related to the rectangular components xx yy zz ,
xy yz zx by the usual tensor transformation law. That is, the stress and strain
components can be referred to a local rectangular frame of reference oriented in the
direction of the curvilinear coordinates.
r
I
@
rr
;
H
YH

@
;
@
COMMON GLOBAL AXES x 2
6
;
;
;

;
;
r
;
;
x
-1
; ;
;
;
;
;
;
;

;x3 ;

; h
Figure 5.6: Cylindrical coordinates.
However, if displacement vectors are resolved into components in the directions of
the curvilinear coordinates, the strain-displacement relationship involves derivatives
of the displacement components and, therefore, is in"uenced by the curvature of the
coordinate system. The strain-displacement relations may appear quite dierent from
the corresponding formulas in rectangular coordinates.
I: Transformation of Derivatives
We start with the relations between the cylindrical coordinates (r
z) and the rect-
angular coordinates (x y z) given by
x = r cos
y = r sin
z=z
and
r2 = x2 + y2
= tan 1 xy
;
z=z
The derivatives are
@r = x = cos
@x r
@
= y = sin
@x r2 ;
r ;
@r = y = sin
@y r
@
= x = cos
@y r2 r
By using the chain rule for dierentiation, it follows that any derivatives with respect
to x and y in the Cartesian equations may be transformed into derivatives with respect
to r and
as
@ = @r @ + @
@ = cos
@ sin
@
@x @x @r @x @
@r ;
r @
@ = @r @ + @
@ = sin
@ + cos
@
@y @y @r @y @
@r r @
II: Displacements
In the cylindrical coordinate system, the components of the displacement vector are
denoted by ur u uz . Components of the same vector resolved in the directions of the
rectangular coordinates are ux uu uz . These components of displacement are related
according to
ux = ur cos
u sin
uy = ur sin
+ u cos
uz = uz
The transformation can be written alternatively in matrix form as
( ) " #( ) ( ) " #( )
ur = cos
+ sin
ux ux = cos
; sin
ur
u ; sin
cos
uy uy + sin
cos
u
III: Strain-Displacement
Set a local Cartesian system (êr e^ e^z ) at point (r
z) in which e^r e^ and e^z are
the unit base vectors in the r
and z direction, respectively. Denoting the strain
components with respect to this coordinate system by ij , that is,
0
2 3
rr r rz
ij ] = 64 r z 75
0
zr z zz

and using the transformation law
2 3
cos
sin
0
ij = im jnmn
0
ij ] = 64 ; sin
cos
0 75
0 0 1
. August 2007
we obtain the relation between the two sets of strain as
xx = rr cos2
+ sin2
r sin 2
yy = rr sin2

+ cos2
+ r sin 2
xy = ( rr ) cos

sin
+ r (cos2
sin2
)
; ;
zx = zr cos

+ z sin
zy = zr sin

+ z cos
zz = zz
Substituting the strain-displacement relation in Cartesian coordinates into the
above, and recognizing such terms, for example, as
@ux = (cos
@ sin
@ )(u cos
u sin
)
@x @r ;
r @
r ;
2 2
= cos2
@ur + sin
ur cos
sin
@ur + sin
@u + cos
sin
u cos
sin
@ur
@r r ;
r @
r @
r ;
r @
we obtain
rr = @u @r
r
= rr + 1r @u
u
@
zz = @z@u z
2r = 1 @ur + @u u

r @
@r r ;
2rz = @u r @uz
@z + @r
2z = 1 @uz + @u
r @
@z
IV: Stress and Equilibrium
Using the local Cartesian system (êr e^ e^z ) the components of the stress tensor at a
point (r
z) are denoted by
2 3
rr r rz
ij ] = 64 r z 75
0
zr z zz

From the coordinate transformation law we obtain
xx = rr cos2
+ sin2
r sin 2
yy = rr sin2

+ cos2
+ r sin 2

zz = zz
xy = ( rr ) sin
cos
+ r (cos2
sin2
)
; ;
xz = rz cos

z sin
yz = rz sin

+ z cos
The derivation of the equilibrium equations in the cylindrical coordinate system

is a straightforward exercise following closely to that of the strain-displacement. We
get
@rr + 1 @r + @rz + rr + b = 0
;
r
@r r @
@z r
@r + 1 @ + @z + 2 + b = 0
@r r @
@z r r
@rz + 1 @z + @zz + 1 + b = 0
@r r @
@z r rz z
where br b and bz are the components of the body force vector ^b in the r
and z
directions, respectively.
V: Summary of Plane Elasticity in Cylindrical Coordinates
The foregoing equations apply to 3-D bodies, we now restrict the equations to the
case of plane elasticity. The summary is similar to that for rectangular coordinates
for plane problems in that we assume we have removed the z dependence.
The basic unknowns are
2 Displacements: ur u
3 Strains: rr r
3 Stresses: rr r
and the corresponding eld equations in terms of these are
Strain-Displacement: rr = @u @r
r
= 1r ur + 1r @u
@

2r = 1 @ur + r @ u
r @
@r r ! !
@rr @ 2 rr @ @ @ @r
Compatibility: r @r 2
;
@
2 @r r @r + 2 @r r @
= 0
;
Equilibrium: @ rr rr 1 @r

@r + r + r @
+ br = 0
;
1 @ + @r + 2 + b = 0
r @
@r r r
. August 2007
Stress-Strain: rr = 2Grr + 23 2 (rr + )]

;
= 2G + 23 2 (rr + )]

;
;
r = 2Gr
rr = 21G rr 3 4 (rr + )]
;
;
= 21G 3 4 (rr + )]

;
;
r = 21G r

Plane Strain: = 3 4
;
zz = 0
zz = (rr + ) = (rr + )
Plane Stress: = 31 +
;
zz = (2G + ) (rr + ) = E (rr + )

; ;
zz = 0
Substituting the strains in terms of stresses into the compatibility equations, using
the equilibrium equations and rearranging gives
!
2
r (rr + ) = ;
4 @br 1 @b
+
(1 + ) @r r @
This is the Beltrami-Mitchell equation in cylindrical coordinates.
5.3 Airy Stress Function Formulation

The plane theory of elasticity has eight unknown functions described by a set of eight
coupled partial dierential equations. A traditional approach to solving a system of
equations is to reduce the number of unknowns at the expense of increasing the order
of the governing dierential equations in the theory of elasticity we reduce the system
to a single unknown function called the Airy stress function.
Cartesian and Cylindrical Coordinates

The stress formulation reduces to nding a set of stresses that satisfy equilibrium
@xx + @xy + b = 0 @xy + @yy + b = 0
x y
@x @y @x @y
5.3. Airy Stress Function Formulation 133
and compatibility
" #
2
r (
4 @bx @bx 4 r2 V
xx + yy ) = + =
;
(1 + ) @x @y (1 + )
We now further reduce this formulation to the determination of a single function.
Suppose the body forces can be derived from a potential V (x y) as
bx = ; @@xV by = ; @@yV
For example, gravity loading in the y- direction is described by V = gy, then
bx = 0 by = ;g
(Note that by is a force per volume since g = W=M = W=V = W=V .) Further,
let the stresses be obtained from a stress function (x y) as
2
xx = @@y2 + V
2
yy = @@x2 + V
xy = ; @x@y @2 (5.1)
It can be easily veried by substitution that stresses obtained in this manner will
pp p p p p p p p p p p p p p pp pp p pp pp pp pp p p p p p pp pp p p p p p p p p p p p p p p p p p p
automatically satisfy equilibrium.
pp p p p p p p p p p p p
p p p p p p p p p p p p p p p p p p pp pp pp p p p p p p p p p p p p p p p p p
equilibrated true compatible
p
Figure 5.7: A true stress eld satises equilibrium and compatibility.
But the stresses must also satisfy compatibility that is, on substituting for the
stresses in terms of the stress function, compatibility becomes
2 2
r r = ;
2( ; 1) r2 V
(1 + )
The function is called the Airy Stress Function. Note that
2 2 ! @2 2! 4 4 4!
2 2
r r =
@ @ @ @ @ @
@x2 + @y2 @x2 + @y2 = @x4 + 2 @x2 @y2 + @y4
. August 2007
The general solution to the above equation can be put in the form
= c + p
where the functions c p are the complementary and particular solutions, respec-
tively. They satisfy
2 2 = 0
r r c
2 2
r r p = ; 2((1+;1)) r2 V
Thus, c is a bi-harmonic function, while p depends on the body force eld and is
not necessarily bi-harmonic.
The Airy stress function is a scalar function, hence most of the results for cylin-
drical coordinates follow directly from the corresponding Cartesian results.
The stresses are related to the stress function by
rr = 1r @ + 1 @2 +
@r r2 @
2 V
2
= @@r2 + V
!
@
r = @r r @
;
1 @
The radial and hoop components of the body force are given by
br = @@r
;
V
b = 1r @@
;
V
The Airy stress function still satises the bi-harmonic equation but written as
4 = r2 r2 = ;2 ; 1 r2 V @2 + 1 @ + 1 @2
2
r
+1 @r2 r @r r2 @
2
r
That is, the only dierence is that the Laplace operator is written in cylindrical
coordinates.
Axisymmetric Problems
We begin the use of the Airy stress function by looking at a couple of problems that
are axisymmetric in the stresses. Note that this does not necessarily mean that the
displacements are also axisymmetric. To make the discussion explicit, we will consider
a curved beam problem that is referred to as Golovin's Curved Beam Problem (1881).
5.3. Airy Stress Function Formulation
135
a pp
M
qqq
bp p p
o
q
XXX
tp p p p p p
@ z
X
qq q t
@
^ = 0@
q
q q q q q q q q q
@
. @
R
@ ^= 0
M o
Figure 5.8: Curved beam with resultant moments.

I: Stresses
A curved beam is subjected to end moments Mo as shown in Figure 5.8. From the
moment balance condition, it is evident that the moment on any radial cross-section
along the beam is constant. In addition, the surface tractions are independent of
. Hence, this is an axisymmetric problem in stress (although not necessarily in

displacements).
For axisymmetric stress problems, the potential functions and V , are indepen-
dent of
. We get for the bi-harmonic equation
2 !
r
2 d 1 d 1
= dr2 + r dr = r r dr d
( " !#)
r
4 1 d d 1 d
= r dr r dr r dr r dr d
By direct integration of r2 c = 0, we obtain the general solution for c as

c = A logn r + Br2 logn r + Cr2 + D
where A B C D are constants of integration. Since D will not contribute to the
stress eld it can be dropped.
The body force is absent in our problem, therefore the Airy stress function is given
by entirely by c
(r
) = (r) = A logn r + Br2 logn r + Cr2
This gives the stresses as
rr = 1r @ =
@r r2
A + B (1 + 2 log r) + 2C
n
2
= @ =
@r r2 !
2
A + B (3 + 2 log r) + 2C
; n
r = @r @ 1 @ = 0
;
r @
. August 2007
There are three coecients to be solved for by satisfying the boundary conditions.
The boundary conditions to be imposed are
at r = a : tr = rr nr + r n = ;rr = 0
t = r nr + n = ;r = 0
at r = b : tr = rr = 0
t = r = 0
These become
rr jr=a = 0 = aA2 + B (1 + 2 logn a) + 2C
rr jr=b = 0 = bA2 + B (1 + 2 logn b) + 2C
since the shear traction condition is automatically satised. One more equation, in
addition to the above, is needed to determine the constants A B and C . We cannot
impose tractions on the ends as the boundary conditions simply because we do not
know them. So we impose conditions on the resultants instead. That is,
Zb
F = dr = 0
a
Zb
Fr = r dr = 0
a
Zb
M = rdr = Mo
a
These become, on substituting for the stresses,
Z @2
@ @
F = @r2 dr = @r ; @r = b rr (b) ; a rr (a) = 0

b a
Fr = 0 b Z
Z @2
Mo = @r2 rdr = r @r ; @
@
@r dr = b2rr (b) ; a2rr (a) ; (b) + (a)
a
b
= ;A logn( ) ; B b2 logn b ; a2 logn a] ; C b2 ; a2 ]
a
Solving these equations for the coecients in terms of Mo gives
A = 4N Mo b2 a2 log b
n a
B = 2N Mo b2 ; a2 ]
C = ; 2N Mo hb2 ; a2 + 2(b2 log b ; a2 log a)i

n n
h i2
N = (b2 ; a2 )2 ; 4b2a2 logn ab
and we nally have for the stresses
b r a2 a
4 M o b2 a2
rr = N r 2 logn a + logn b + b2 logn r
2 2 2 2
= 4MNo b ; ar2 logn ab + logn rb + ab2 logn ar + 1 ; ab2
r = 0
The stress distribution is sketched in Figure 5.9. Notice the very large increase in
hoop stress at the inner radius.
rr
pp pp
pp p ppp p
p p p p p ppp pp
ppppppppp pp p
p p p ppp
p
p p p p p p ppppp
ppppppppp p
Figure 5.9: Stress distributions in curved beam.
A signicant aspect of this solution is that the hoop stress is not linearly
distributed on the cross-section (as is found in elementary beam theory). However,
with b = a + h and we consider the limit of small h then indeed we do recover a linear
distribution of stress.
II: Displacements from Stresses
We will now obtain the displacements for the curved beam problem. This will be
done by rst obtaining the strains from the stresses and then integrating the strain-
displacement relations.
The derivatives of displacement can be obtained from Hooke's law and the strain-
displacement relation as
rr = @u
@r
r
= 1 (1 ; ) ; ]
2G rr

3;
4
1 A
= 2G + r2 + B f1 ; 4 + 2(1 ; 2 ) logn rg + 2(1 ; 2 )C
Note that the displacements u ur could be functions of
, hence the full expression
for the hoop and shear strains must be used. That is,
= urr + 1r @u
@

= 1 (1 ; ) ; ]
2G rr
1 A
= 2G ; r2 + B f3 ; 4 + 2(1 ; 2 ) logn rg + (1 ; 2 )2C
2r = 1r @u r @u u
+
@
@r r ; = 1 =0
G r
. August 2007
Integrate the radial strain equation to get
2Gur = ; A + Br ;1 + 2 logn r ; 4 logn r] + 2(1 ; 2 )Cr + g1(
) 0
r
where g1(
) is a function of integration. Combine this with the hoop strain relation
to get
2G 1r @u
@

= 4 B (1 ;
) ;
1 g (
)
r 1
0
Now integrate this with respect to
2Gu = 4B (1 ; )r
; g1(
) + g2(r)
where g2(r) is another function of integration.
Both displacements are known to within two arbitrary functions g1 (
) and g2(r).
At this stage we have not used the shear strain-displacement relation imposition
of it will remove the arbitrariness in the integration functions. Substitute for the
displacements into the shear strain relation to get
0 = 1r @u r @u u
+
@
@r r ; = 1 g (
)] + 4B (1 ; )
+ g (r)]
r 1
00
2
0
1 4B (1 ; )r
; g (
) + g (r)]
;
r 1 2
which simplies to
g1 (
) + g1 (
) + rg2(r) g2(r) = 0
00 0
;
Collect like functions of

and r and since they separately must, at most, be constant,
then
g1 (
) + g1(
) =
00
rg2(r) g2(r) =
0
; ;
where is the constant. Solving these dierential equations gives

g1(
) = C1 cos
+ C2 sin
+
g2(r) = C3r +
The functions g1(
) and g2 (r) are seen to be rigid body motions. In other words, the
stress function can be used to obtain the displacements within a rigid body motion.
Note also that u is linear in r and therefore \plane sections remain plane" even
though the strains (and stresses) are not linearly distributed on the cross-section.
Example 5.3: Derive the Lam%e solution for a pressurized cylinder.

Consider a hollow cylinder subjected to uniform external and internal pressures
po and pi , respectively. This is an axisymmetric problem and therefore we can take
the stress function as
= A logn r + Br2 logn r + Cr2
pppp p p p p p p p p p p p p p p p pppp ;
pp p p p p p p p p p p p p p p p p p p p p
;

6
6 pp p p p p p ppp ; p p p p p p pp
pp pp po9
ppp ppp i H z pp pp
*

2b 2a ppp p p p p X
;

:

X
j
H
-
p p
?
@
ppp p pp p p p p p p p p pp p p p H
R
@ p p yX
X
pp p pp
pppp p p p p p p p pp p p p p p p pppppp p p H
Y
?
@ I
@
Figure 5.10: Pressurized cylinder.

which gives the stresses (same as for Golovin's problem)
rr = rA2 + B 1 + 2 logn r] + 2C
= ; rA2 + B 3 + 2 logn r] + 2C
r = 0
The boundary conditions to be satised are
at r = a : tr = ; rr = pi
t = ; r = 0
at r = b : tr = rr = ;po
t = r = 0
Again, the two shear conditions are satised automatically. This gives two equations,
but there are three unknowns A B and C .
Our diculty here is that we have a solution that appears to satisfy all the re-
quirements: equilibrium and compatibility are satised since the stresses are derived
from a bi-harmonic stress function, and any pair of the three coecients are capable
of satisfying the traction boundary conditions. Lam%e's problem is an example of a
multiply connected region that is, a body with two or more independent boundary
contours. In these cases, the compatibility equations are an incomplete statement
of compatibility for the body. To see this, look at the displacements in particular,
note that
2Gu = B (1 + )r
This cannot be allowed since, for continuity, we have
at = 0 : u = 0
at = 2 : u = 2BG (1 + )r2
which gives both a zero and a non-zero value at the same point. That is, the
presence of B gives rise to multi-valued displacements, and only B = 0 can satisfy the
continuity condition. Bi-harmonic functions are guaranteed to satisfy compatibility
only for simply connected regions for multiply connected regions, we impose the
additional constraint that the displacement be single valued.
. August 2007
The boundary conditions now become
at r = a : ;pi = aA2 + 2C
r=b: ;po = bA2 + 2C
which gives
2 2
A = (po ; pi ) (b2a;b a2 )
2 ; b po
2C = a(bp2i ;
2
a2 )
and the stresses as
1 a2 a 2
rr = (1 ; a2 =b2 ) b2 pi ; po + r2 (po ; pi )
1 a2 a 2
= (1 ; a2 =b2 ) b2 pi ; po ; r2 (po ; pi )
This is the Lam%e solution.
Consider the special case when a = 0, that is, it is a solid cylinder. Then the
stresses reduce to
rr = ;po
= ;po
There is no dependence on r and the stress is a state of hydrostatic stress. Consider
the companion problem of an innite sheet with a circular hole and let the remote
stress in the radial direction be ;po . The stresses are
2
rr = ;1 + ar2 ]po
2
= ;1 ; ar2 ]po
Note that at r = a the radial stress goes to zero but the hoop stress increases
to twice the applied pressure. That is, if there is no hole in the sheet then the
maximum stress is po the presence of the hole causes a redistribution of stress and
in so doing the maximum stress rises to 2po . This is an example of a stress riser or
stress concentration.
Now consider the case b ! 1 and po = 0 that is, it is a pressurized hole in an
innite sheet. The stresses are
2
rr = ; ar2 pi
2
= + ar2 pi
These stresses are the same at r = a and decay as 1=r2 . As we impose a nite
boundary at r = b, rr still has the same boundary conditions but is forced to zero
more rapidly. Hence, to compensate, increases giving a maximum stress of
2 a2
max = bb2 +
; a2 pi
on the inner boundary.
A nal special case of interest is that of a thin walled pressure vessel. Consider
the approximation a ! b, that is, b = a + t, then
a2 = (b ; t)2 = b2 ; 2bt + t2 = 1 ; 2t + t2 ' 1 ; 2t
b2 b2 b2 b b2 b
Substituting into the Lam%e solution with po = 0 gives
rr 0
pi bt
The hoop stress is the dominant stress. If the cylinder is thin-walled with the radius
being twenty times that of the thickness, then the generated hoop stress is twenty
times that of the applied pressure.
Example 5.4: A disk of radius b rotates at a constant angular velocity '. De-
termine the resulting stresses.
pp p p p p p p p p p p p p p p p p p p
ppp pp
p
b
'
p p p p p p p p pbp p p p p p
@
@
R
@
Figure 5.11: Rotating disk.

This is an example where the body force terms are important | the centrifugal
force constitutes the body force eld. These body forces are given by
br = r'2
b = 0
It can be easily veried that a suitable body force potential is
V = ; 12 r2'2
Thus, the stress function satises the nonhomogeneous equation
r4 = ; 2(+;11) r2 V = 4((+;1)1) '2
Since the body force eld and the boundary conditions are axisymmetrical, the
resulting stress eld must be also axisymmetrical. Thus, the solution is given by
= A logn r + Br2 logn r + Cr2 + p
. August 2007
where p is a particular solution. Note, however, that both logn r and r2 logn r
produce singular stresses at r = 0 (see the tables at the end of this chapter) and
therefore are not admissible. Accordingly, we set A = B = 0. It is easy to show that
p = 16(;+11) r4 '2
is a particular solution and thus reduces to
= Cr2 + 16(;+11) '2 r4
The stresses corresponding to this function are
rr = 1r @
@r + V = 2 C ; + 3 '2r2
4( + 1)
= @@r2 + V = 2C + 4(;+51) '2r2
2
r = ; @r @ 1 @ = 0
r @
The boundary conditions at r = b are
tr = rr = 0
t = r = 0
the second equation is satised automatically. From the rst equation, we obtain
C = 8(++31) '2 b2
Thus, the complete solution for the rotating disk is given by
= 8(++31) '2 b2 r2 + 16(;+11) '2 r4
and the stresses are
rr = 4(++31) '2 b2 ; r2]
= 4(++31) '2 b2 + (( ; 5) r2 ]
+ 3)
The maximum stress 1s
rr jmax = jmax = 4(++31) '2 b2
and occurs at the center of the disk (r = 0).
The problem of a hollow disk is solved in the analogous manner, it is just a
matter of retaining the A logn r and Br2 logn r terms in the stress function.
5.4. A Guide to Selecting Stress Functions 143
5.4 A Guide to Selecting Stress Functions
The key in using the Airy stress function to solve plane elasticity problems lies in
the selection of candidate stress functions. Since stress functions satisfy the equi-
librium and compatibility equations, they are required only to satisfy the boundary
conditions. Thus, the boundary conditions give us the clue about the nature of the
functions to be selected. We establish a very general collection of stress functions and
then show how they are synthesized to solve particular boundary value problems.
Collection of Particular Solutions of r4 = 0

The general solution for the bi-harmonic equation in cylindrical coordinates was ob-
tained by J.H. Michell (1899) by direct substitution of = f (r)e . The solutions
are summarized as
(r
) = Ao + Bo
+ A logn r + Br2 logn r + Cr2 + Dr2
h 3 1 i sin
+ A1 r + B1r + C1 r + D1 r logn r + Er
cos
Xh m 2 i sin m
Am r + Bm rm+2 + Cm r1m + Dm rrm cos

1
+ m
m=2
X mh
1 i
+ r Am cos m
+ Bm sin m
+ Cm cos(m ; 2)
+ Dm sin(m ; 2)
m=2
The brace indicates that either term can be used. The constant term Ao does not yield
any non-trivial stresses and is therefore usually omitted. Stresses and displacements
obtained from these can be found in the charts of Tables 5.1 & 5.2 at the end of this
chapter.
A quick way to obtain harmonic functions in Cartesian coordinates is to extract
separately the real and imaginary parts of an analytic function. For example, if
= R + iI = (x + iy)n
p
i ;1
then
n R I
1 x y
2 2
x ;y 2 2xy
3 x3 ; 3xy2 3x2y ; y3
4 x4 ; 6x2y2 + y4 4x3y ; 4xy3
5 x ; 10x y + 5xy 5x y ; 10x2y3 + y5
5 3 2 4 4
Each of these is a harmonic function. If (x y) is harmonic, then the product functions
x and y are bi-harmonic because
@2 @2 @ + x @ 2 + x @ 2 = 2 @ + xr2 = 2 @
+
@x2 @y2 x ] = 2 @x @x@x @y@y @x @x
. August 2007
Therefore,
2 2 x] = 2@ 2 ] = 0
r r
@x r
Similarly for the y product. This gives a quick scheme for obtaining bi-harmonic
functions. For example,
x y
x 2 : xy xy : y2
3
x xy
;
2 : 2
2x y 2
xy y : 3 2xy2 ;
x4 3x2 y2 : 3x3 y xy3

; ; xy 3xy3 : 3x2 y2 y4 ; ;
is a collection of bi-harmonic functions obtained from the table of harmonic functions

above. This can be generalized to the statement: Let o 1, and 2 be any harmonic
functions, then a representation of a bi-harmonic function can be formed by the linear
combination
(x y) = o(x y) + x1 (x y) + y2(x y)
Generating bi-harmonic functions is thus a straightforward procedure.
A bi-harmonic stress function is always the exact solution to some problem | the
art of solving practical problems is nding the right combination of these functions
to satisfy the given boundary conditions.
Example 5.5: A hollow cylinder is subjected to a non-uniform outer pressure

given by p = po sin2 . Determine the stress distribution.
pppppp? p p?
p p?p p p p p p p p p p p p( ) = po sin2
pp p p p p p p p p p pp
ppp p p p p p p p p p p p p p p p p p p p p p p p ppp
6
2b
6
2a pppp ppppp pp ppp
ppp p p p p p p p p p p ppp p
? pp pp pp p p p p p p p pp pp p
p p ppppp p
? p p p p p p p pp p p p p p p p ppp pp
6
66
Figure 5.12: Hollow cylinder with non-uniform pressure.

This is a variation on the Lam%e problem and we will use it to illustrate how the
stress functions can be selected. We begin by rewriting the pressure distribution in
the equivalent form
p = 21 po (1 ; cos 2 )
This is done because the charts of Tables 5.1 & 5.2 use trigonometric functions with
multiple arguments and not multiple powers. The boundary conditions are
at r = a : rr = r = 0
at r = b : rr = ; 12 po + 12 po cos 2
r = 0
From the dependence nature of rr at the boundaries, we select the stress functions
that produce rr either independent of or dependent on cos 2 . From the stress
function table, we nd the following stress functions which satisfy these requirements
r2 log r r2 log r r2 cos 2 r4 cos 2 1 cos 2 cos 2
n n r2
Further examination of these functions reveal that r2 logn r produces multiple{valued
displacement and is therefore not suitable for the present problem. Thus, a proper
Airy stress function is given by
= Ar2 + B log r + C + Dr2 + Er4 + F 1 + G] cos 2
n r2
The constants A :: G are to be determined by the boundary conditions.
Example 5.6: Determine the state of stress in a large plate, with a small hole,
uniformly loaded in the y direction remote from the hole.
6
6 1
6 6
6 1
6 =0
pp pp ppapp pp pp pp p ppp r pp pp pp pp pp pp pp p ppp

y
6
pp p p p p p p p pp p
pp p
*

-x

= p +
; p p p p p p pp p

; ppppp
?
?
1
? ?
?
?
1
=0
Figure 5.13: Hole in an innite sheet.
This problem is usually referred to as the Kirsch's Hole in an Innite Sheet
Problem. The basic strategy we will apply is to add two stress systems together:
the rst gives the correct applied tractions at innity while the second enforces the
zero tractions around the edge of the hole without aecting the stresses at innity.
I: Remote Stress Function
Initially, neglect the hole and obtain a stress function for the remote stress. That
is, knowing
2
xx = @@y2o = 0
2
yy = @@x2o = 1
@ 2 o = 0
xy = ; @x@y
. August 2007
leads us to choose the stress function as
o = 2 x2
1
In the vicinity of the hole, we will need to use cylindrical coordinates when satisfying
the boundary conditions, hence rewrite o as
o = 2 r2 cos2 = 2 r2 ( 21 + 21 cos 2 ) = 41 r2 + 14 r2 cos 2
1 1
1 1
Our plan is to add to this a stress function that will satisfy the boundary conditions
at r = a. Whatever form it takes, the stresses must be consistent with this at r ! 1
and therefore they must go to zero at r ! 1.
Although the o obtained above satises the stress condition at r ! 1, it does
not satisfy the boundary condition at r = a of
0 = tr = rr nr + r n = ; rr
0 = t = ; r
The stress function o yields the following stresses at r = a
rr = 12 ; 21 cos 2
1 1
r = 2 sin 2
1
Additional bi-harmonic functions must be added to o in order to clear these trac-

tions without disturbing the stress condition at r ! 1 which are already satised
by o
II: Complete Stress Function
Using the above mentioned boundary conditions as a guide, the added bi-harmonic
functions must produce stresses which are either independent of or dependent on
cos 2 (for rr ) and sin 2 (for r ). Meanwhile, the additional stresses must vanish
as r ! 1. From the bi-harmonic function table, the suitable candidate stress
functions are
log r 1 cos 2 cos 2
n r2
The general stress function that satises the remote conditions is therefore
= A logn r + 4 r2 + 4 r2 + C2 r12 + D2 ] cos 2
1 1
giving the stresses

A 6 C 4 D
rr = r2 + 2 ; 2 + r4 + r2 cos 2
1 1
2 2
A 6 C
= ; r2 + 2 + 2 + r4 cos 2
1 1
2
; 6 C 2 D
r = 0 ; 2 + r4 + r2 sin 2
1
2 2
There are three constants A C2 D2 to be determined by the boundary conditions

at r = a. Note that as r becomes very large that the additional terms do indeed
vanish.
y
6

pp p p p p p p p p p ppp rr
ppppp pp
x -
Figure 5.14: Stress distributions near the hole.

III: Hole Boundary Conditions
The boundary conditions at the edge of the hole are that the tractions are zero, that
is,

tr = ; rr = 0 = aA2 + 2 ; 2 + 6aC42 + 4aD22 sin 2
1 1

tr = ; r = 0 = ; ; 2 + 6aC42 + 2aD22 cos 2
1
Since this must be true for any then

A + = 0 1
a2 2
1
+ 6aC42 + 4aD22 = 0
2
; + 6aC42 + 2aD22 = 0
1
2
Solving these simultaneously gives the coecients as
A = ; 2 a2 C2 = 4 a4 D2 = ; 2 a2
1 1 1
The stresses are, nally,

a2 4a2 3a4
rr = 2 1 ; r2 ; 1 ; r2 + r4 cos 2
1
2 4
= 2 1 + ar2 + 1 + 3ra4 cos 2
1
2a2 3a4
r = 2 1 + r2 ; r4 sin 2
1
Figure 5.14 shows the distribution of the stress along the = 0 axis.
The hoop stress around the edge of the hole is
= f1 + 2 cos 2 g
1
showing that at = 0, the maximum stress is three times the remote stress. Also
note that at = =2, = ; . 1
. August 2007
Problems with Concentrated Forces
These examples demonstrates a common approach to solving problems using stress
functions. After the function is chosen, it is rst necessary to verify that it is a solution
of the bi-harmonic equation. This will guarantee that the stresses are equilibrated and
the strains compatible, everywhere in the body. Then the function is investigated for
the types of tractions it gives rise to. The behavior of these tractions is what species
the type of boundary value problem being solved. For example, zero tractions are
associated with a free surface, a linear traction distribution with an applied load that
has a simple distribution.
I: Flamant's Problem (1892)

The stress function we will investigate is
!
= Ax tan ; 1 x
y
This has the representation in cylindrical coordinates as
(r
) = Ar
+ 12 ] sin(
+ 21 )
Since this can be expanded into forms that are in Tables 5.1 & 5.2 we conclude that
it is biharmonic.
ty
..........
...........
.......... ..........
...........
..........
...........
.......... y 6= 0 ...........
..........
tx 6y
1
.......... y 0 ...........
..........
...........
.......... ...........
..........
.......... x
-
...........
.......... ...........
Figure 5.15: Tractions on two surfaces at constant y.
We will work primarily in Cartesian coordinates. The stresses are
2 2 2 3 @ 2 = 2A xy2
xx = @@y2 = 2A xr4y yy = @@x2 = 2A yr4 xy = @x@y
;
r4
Keep in mind that the tractions can be evaluated for any (x y) but recognizing a
boundary is really a question of identifying some obvious feature of the tractions for
example, zero traction for a free surface. Note that for the present case, the stresses
are singular at r = 0, that is
! 1 as r ! 0
Also note that the stresses are zero at innity
! 0 as x or y ! 1
To consider a particular boundary value problem we must look at the tractions on
particular lines. Consider a horizontal plane y 6= 0 such that nx = 0 ny = 1, then
the tractions are
2
tx = xxnx + xy ny = xy = 2A xy r4
y 3
ty = xy nx + yy ny = yy = 2A r4
Note that along y = 0, the traction vanishes except for r = 0.
6y
ub t^
6

;
;
n^
dy ds
;dx
ua -
x
Figure 5.16: Traction resultants on an arbitrary curved surface.
We will look the resultants. Consider a general boundary S with positive direction
as shown in the Figure 5.16. The components of the unit normal vector to S are given
by
nx = dS dy ny = ; dSdx
The traction has two components given by
2 ! ! 2 ! ! !
tx = xxnx + xy ny = @y2 dS + ; @x@y ; dS = dS @
@ dy @ dx d
@y
2 ! ! 2 ! ! !
ty = xy nx + yy ny = ; @x@y @ dy + @ ; dx = ; d @
dS @x2 dS dS @x
The resultant forces on the contour length between points a and b are
Zb " #b Zb " #b
Fx = txdS = + @y @ Fy = ty dS = ; @
a a a @x a
. August 2007
which are dependent only on the values of the stress function at the end points. The
resultant forces on the entire horizontal section y 6= 0 are
" #b " 2 #x=
@
Fx = + @y = ;A r2 x ;1
=0
a x=+
" #b " !
1
# x=
@
Fy = ; @x = ; +A tan y + A r2 ;1 x xy ;1
= ;A
a x=+ 1
Let the resultant force be equal to the applied force P (force per unit length), that
is,
;A = ;P
from which the constant is determined.

The stress eld is thus obtained as
2 3 2
xx = 2P xr4y yy = 2P yr4 xy = 2P xyr4
These are the stresses in a half-plane subjected to a downward vertical force applied
at the origin of the coordinate system. This is referred to as Flamant's solution for
the half-plane.
There is a subtlety in the solution that is worth discussing. Recall that a bi-
harmonic stress function always gives rise to a stress eld that is the solution of a well
posed elasticity problem. This problem solved, however, may not be the problem of
immediate interest. For example, if the half-plane problem is posed as \concentrated
normal traction and zero shear traction" then the Flamant result is not the solution.
The reason is that the Flamant solution also has a singular behavior in the shear
traction. To eect a practical solution, it is often necessary to relax the statement of
the problem. For example, if the half-plane problem is posed as \concentrated normal
traction and concentrated shear traction with a zero resultant" then the Flamant
result is the solution. Note that the shear distribution in not known in advance.
II: Partial Load on a Half Plane
Consider the stress function
(r
) = Ar2
; sin
cos
] = Ar2
; 12 sin 2
]
This is in Tables 5.1 & 5.2 hence it is bi-harmonic. We will now discuss what boundary
problem it is associated with.
Either by dierentiation, or from the tables, we get that the stresses are
rr = A2
+ sin 2
] = A2
; sin 2
] r = A;1 + cos 2
]
Note that these stresses do not depend on r. Thus on lines r = a = constant the
tractions are
tr = A2
+ sin 2
] t = A;1 + cos 2
]
y
6

; r
;
;
;
x
- ???????? x
-
66666666666 66666666666
Figure 5.17: Loading on half of a half plane.

which also do not depend on r. These tractions, however, do not have an easily
recognizable distribution.
Along the lines
= 0 and
= , the tractions are

= 0 : tr = 0 t = 0
= : tr = 0 t = A2
This is a constant normal traction on half of the surface. We therefore recognize this
as the loading shown in Figure 5.17. The coecient A would be chosen to give the
appropriate intensity of the surface traction.
An interesting solution is obtained by combining two of the partial load solutions
but shifted a distance apart. With opposite magnitudes, the result in a normal applied
load over a nite segment of the surface. Thus,
h
= A (x ; a)2 + y2] tan 1 ( x ;y a ) ; (x ; a)y
;
2 2 1 ( y ) + (x + a)y
i
;(x + a) + y ] tan
;
x+a
Let the contact area 2a shrink but also let the intensity correspondingly increase so
that A2a = constant, then we approach, in the limit of small a, the solution to a
concentrated point load.
Note that the solution has a more complicated expression than the Flamant solu-
tion this is because there are no surface shear stresses.
Example 5.7: Show by dierentiation that the function = Ax tan 1(x=y) is

;
bi-harmonic.
Using the relations
@ 1 x y @ 1 x ;x
@x tan ( y ) = x2 + y2 @y tan ( y ) = x2 + y2
; ;
we have for some of the derivatives

@ = A tan 1 x + A xy r2 x2 + y2
;
@x y r2
2
@ = 2A y 3 2
@ = 2A x2 y
@x 2 r 4 @y2 r4
. August 2007
Therefore,
2 2
r2 = @@x2 + @@y2 = 2rA2 y
This shows that is not harmonic. Further dierentiation gives
@ 2 r2 = 16A x2y ; 4A y @ 2 r2 = ; 12A y + 16A y3
@x2 r6 r4 @y2 r4 r6
Therefore,
r2r2 = ; 16r4A y + 16r6A y(x2 + y2) = 0
Hence, is indeed bi-harmonic.
Example 5.8: Determine the stresses in a disk under diametral compression.
pp p p p p p p p p p p p p p
?
pp p p p Hp p Hp p p p
?
pp p p p pppp
pp p p p p p p p p p p p p p
ppp p D p p pp ppp p p p pp
pp

Bp p p r2
;
p 1 HH p p BH r1

pp p pp *

pp H pp +
:

ppp
j
= +
pppppppp pppp pppppppp pppp
- pp -
pp pp p pp 2 p p

ppp p pp p p ppp
ppppppp p p p p p
6 6
Figure 5.18: A disk under diametral compression is solved as the superposition of

two Flamant solutions plus a uniform traction.
Let the center of the disk be the origin of the coordinate system, then the two
Flamant solutions have the stress functions
" #
= x tan 1 1 x ;
2D y
At a point B on the circumference there is a compression in the radial directions of
r1 and r2 of amount
2P cos 1 2P cos 2
r1 r2
respectively. But these radial lines are perpendicular to each other and therefore
cos 1 = cos 2 = 1
r1 r2 D
We conclude that the two principal stresses at B are 2P=D. Since this conclusion
is true for any point B on the circumference then a uniform traction of amount
2P=D applied around the circumference will then give a zero traction boundary.
The stress function to be added is
P x2 + y2]
= D
The stresses on the horizontal line of symmetry are given by
2 P D2 ; 4x2 2 ;2 P 4D 4
xx = hD D2 + 4x2 yy = hD D2 + 4x2 ]2 ; 1 xy = 0
where h is the thickness of the disk.
Approximate Boundary Conditions

A well posed elasticity problem has boundary conditions in the form of imposed
tractions or imposed displacements. However, in the solution approach where we
synthesize the solution as a collection of particular solutions, we do not always have
the ability to exactly match the boundary conditions. We illustrate this with the
problem of a deep cantilever beam.
6y
......
..... tx (y) = 0
......
.....
......
..... ty (y) = o h2 ; y2 ]
......
.....
...... 2h
6 6
......
.....
...... -
.....
......
.....
...... ?
6 x
.....
......
.....
...... -
6
L
Figure 5.19: Cantilever beam with end shear traction.

Consider a deep cantilever beam with a parabolic shear traction distribution on
the end. Based on the traction distribution, it seems that the stress function
(x y) = Axy + Bxy3 + Cy3
is sucient to solve the problem because it leads to a parabolic shear distribution.
This is what we want to investigate.
First, it is clear that is bi-harmonic because the highest power in the polynomial
is three. The stresses are obtained as
2 2 @ 2 = ;A ; 3By2
xx = @@y2 = 6Bxy + 6Cy yy = @@x2 = 0 xy = ; @x@y
Consider the tractions on the horizontal planes y = h such that nx = 0 ny = 1.
That is,
tx = 0 = xy = ;A ; 3Bh2 ty = 0 = yy = 0
Note that the normal traction condition is automatically satised. In fact, we only
get one equation from the four traction conditions and this leads to A = ;3Bh2 .
. August 2007
Now look at the face at x = L where nx = 1 ny = 0. The tractions are
tx = 0 = xx = 6BLy + 6Cy ty = oh2 y2] = xy = A 3By2
; ; ;
These two conditions leads to three equations

6BL + 6C = 0 o h2 = A ; o = 3B ;
Solving gives the coecients

2A = ;o h2 B = o =3 C = BL
;
Thus the stress solution is

xx = 2o x L]y
; yy = 0 xy = o h2 y2] ;
At this stage, we have a stress eld that satises the tractions on three sides of the
body. In order to guarantee that this is indeed the solution we must also satisfy the
boundary conditions along the face at x = 0. But what are the traction conditions?
These were not specied as part of the problem.
This is an example of a `mixed boundary value problem', that is, some of the
boundary conditions are traction specied while others are displacement specied. To
obtain the displacements we must integrate the strain-displacement relations. Thus,
from the normal strains (after using Hooke's law with = 0 for simplicity) get
Eux(x y) = 2ox2 y=2 xyL] + f1(y)
; uy (x y) = f2 (x)
where f1 and f2 are functions of integrations. The displacements must also satisfy the
shear strain-displacement relation, hence substitute and regroup in terms of only x
and y. The separate groups must be equal to a constant ( , say), therefore integration
gives the separate functions f1(y) and f2(x). We nally get for the displacements
Eux(x y) = 2ox2 =2 xL+h2 y2 =3]y y+c1
; ; ; Euy (x y) = o L x=3]x2 + x+c2
;
where c1 c2 are unknowns. These contribute a rigid body motion.

Look at the displacements at x = 0
Eux(0 y) = 2oh2 y2=3]y y + c1
; ; Euy (0 y) = c2
The horizontal displacement is non-zero, not what we wanted for the xed end con-
dition. The above solution is not the exact solution for the xed cantilever beam
problem the simple polynomial stress function is not capable of representing the sin-
gular stress behavior at the xed end where y = h. The solution, however, is the
exact solution if the tractions at x = 0 were specied as
tx = +2oLy ty = o h2 y2]
; ;
5.5. Applications of Complex Variables 155
Note that if these tractions were arbitrarily specied then global equilibrium is prob-
ably violated.
The above solution gives a good approximation to the cantilever beam problem
because it satises the exact traction conditions top and bottom, and as can be
veried, satises an approximate version of the tractions in the form of resultants
on the ends. In fact, this is a very useful approach to obtaining practical solutions:
satisfy some of the traction conditions exactly, and the others approximately in the
form of resultants. If the region of interest is remote from these latter boundaries,
then the solution will be quite insensitive to the specic distributions of the applied
tractions. This is known as St. Venant's principle.
5.5 Applications of Complex Variables

The previous sections demonstrated the fundamental role played by harmonic and
bi-harmonic functions in the plane theory of elasticity. These functions have a very
simple representation in terms of complex variables as a consequence, many of the
results of the plane theory can be summarized very elegantly and conveniently using
these harmonic functions. In addition, the use of complex variables allows the in-
troduction of conformal mapping techniques which facilitate the solution of problems
with a larger range of boundary geometries.
Representation of Harmonic Functions

The essence of the complex variable approach is the introduction of two new vari-
ables which can be viewed simply as a linear coordinate transformation. By suitably
choosing the transformation we can simplify many of the relations.
Consider two new variables dened as
z x + iy
z x iy
;
where i = ;1. These can be viewed simply as a linear coordinate transformation

p
in which i is just a constant. A quantity with a bar is referred to as the complex

conjugate that is, in an expression replace i with ;i. The notation g (z ) means that
wherever z is seen in the function g(z), replace it with z. Also g(z) = g(z) as seen
from the following
g(z) = ez g(z) = ez g(z) = ez = ez = g(z)
We also have that
z = x + iy = r cos
+ ir sin
= rei
This indicates that it will be easy to alternate between Cartesian and cylindrical
coordinates.
. August 2007
We will now transform all functions and their derivatives according to
(x y) ;! (z z)
The derivatives can be transformed as
@ = @ @z + @ @ z = @ (1) + @ (1) = @ + @
@x @z @x @ z @x @z @ z @z @ z
@ = i @ ;i @ or i = ; + @
@ @
@y @z @ z @y @z @ z
Consequently, we have
@ = 1 @ ; i @ @ = 1 @ + i @
@z 2 @x @y @ z 2 @x @y
As a simple illustration of the use of complex variables, consider the expression for
the volumetric strain
e = @u + @v = @u + @u + i @v ; i @v = @ (u + iv) + @ (u ; iv) = 2Real @ (u + iv)
@x @y @z @ z @z @ z @z @ z @z
Notice how natural the complex variable u + iv arose.
Laplace's equation can now be rewritten as
@2 @2 @ @ @ + i @ = 4 @ @
2
r = + = ; i
@x2 @y2 @x @y @x @y @z @ z
Hence, harmonic functions must satisfy the equation
@2 = 0
@z@ z
This is easily integrated to give
@ = f (z) or = Z f (z)dz + g(z) = F (z) + g(z)
@z
Since the harmonic function is real, then
2 = F (z) + g(z) + F (z) + g(z) = F + g] + F + g]
That is, a solution of r2 = 0 can be written as
= 21 1(z) + 1(z)] = Real1(z)]
which is simply the real part of an analytic function.
The bi-harmonic equation can similarly be transformed to
2 2
r r ) = 0
@4) = 0
)
@z2 @ z2
On integration get
@ 3 ) = f (z )
@z@z@ z
@ 2 ) = zf (z) + g(z)
@z@z
@ ) = z Z f (z)dz + Z g(z)dx + h (z)
1
@z ZZ ZZ
) = z f (z)dz g(z)dz + zh1 + h2
Redening some terms, we have
) = zF (z) + G(z) + zh1 (z) + h2 (z)
Since ) is real, then
2) = zF + G + zh1 + h2 + zF + G + zh1 + h2
or, after regrouping,
2) = zF + h1] + G + h2] + zF + h1 ] + G + h2 ]
= z1 + 2 + z1 + 2
Hence, a general representation of a bi-harmonic function is
) = Realz 1 + 2]
where 1 = 1(z) 2 = 2(z). Compare this representation to ) = x1 + 2 used
earlier in the chapter.
Application to Navier's Equations

Recall that the eld equations of the linear theory of elasticity can be reduced to a
single set called the Navier's equations, written for plane problems as
@e + b = 0
r2 u + ( + ) @x x
@e + b = 0
r2v + ( + ) @y y
A solution of these equations is a solution to the eld equations. We will transform

this set using complex variables.
. August 2007
Multiply the second equation by i and add to the rst to get
@ + i @ ]e = 0
r2u + iv] + ( + ) @x @y
(We neglect body forces for the present.) This becomes, on replacing (x y) with (z z)
2
4 @ u + iv] + 2( + ) @ e ] = 0
@z@ z @ z
Integrate with respect to z and get
@ (u + iv) + 2( + )e = g(z)
4 @z
Add this equation to the conjugate of itself and obtain
@
4 @z (u + iv) + @@z (u ; iv) + 4( + )e = g + g
Since the bracketed term is e as shown earlier, then
4( + 2)e = g + g
That is, e is the real part of an analytic function and is therefore harmonic. Hence
the displacements can be written as
hZ i Z
2u + iv] = ;( + ) g dz + zg + 21 gdz + h
4( + 2)
Dene the new functions
( + ) Z
4( + 2) g dz h(z) ;(z)
giving the nal expression for the displacements as
+ 3
2u + iv] = + (z) ; z (z) ; (z)
0
This says that for any two analytic functions and , the displacements, strains,
stresses and so on, obtained from them will automatically satisfy all the necessary
eld equations.
These results are easily rewritten in cylindrical coordinates. Utilizing the trans-
formation equations
ur = ux cos
+ uy sin
u = ux sin
+ uy cos
;
we can introduce a complex displacement function as
ur + iu = uxcos
; i sin
] + uy sin
+ i cos
] = ux + iuy ]e i ;
These become, in terms of the stress function,

2ur + iu ] = ++3 (z) ; z (z) ; (z) e i 0 ;
The only dierence in comparison to the Cartesian form is the presence of the expo-
nential term.
Stresses and Boundary Tractions

By using the strain-displacement relation and Hooke's law, the stresses can be ob-
tained as
xx + yy = 2 + ] = 4Re ]
0 0 0
yy ; xx + 2ixy = 2z + ] 00

() ddz()
0 0
In Cartesian coordinates, the boundary tractions are usually given in the combinations
xx xy ] or yy xy ] hence a convenient form for the above is
tx ; ity = xx ; ixy = + ; z ; "
0 0 00
tx + ity = yy + ixy = + + z + "

00 00 00
These are special cases of the traction on an arbitrary plane given by

tn ; itt = tx ; ity ]ei
= 12 xx + yy ] ; 12 yy ; xx ; i2xy ]ei2
= + ; z + ]ei2
u
0 00 00 0
6y
b
t^
6

;
;
n^
dy ds
;dx
a -
x
u
Figure 5.20: Tractions on an arbitrary boundary.
Similarly for cylindrical coordinates, the stresses are
rr + = xx + yy
; rr + 2ir = yy ; xx + 2ixy ]e2i
. August 2007
In terms of the stress function, these become
rr + = 4Re (z)] 0
; rr + 2ir = 2z (z) + (z)]e2i

00 0
We also have the relation

rr ir = (z) + (z)] z (z) + (z)]e2i
;
0 0
;
00 0
When performing evaluations of the functions, it may be necessary to replace z with

z = rei .
Recall that the resultant forces on any arc-length is given (in terms of the Airy
stress function) by
@ ) B @ ) B
Fx = @y Fy = @x
A A
This is written in terms of the complex potential functions as
@ ) @ ) @ @ B @ ) B
Fy + iFx = ; @x + i @y = ; @x ; i @y ) = ;2 @z

A A
Further, since the stress function has the general representation

) = z1 + 2 + z1 + 2
then the resultant forces are represented by
h i
1 F + iF = z + +
y x 1
0 0
;
2 1 2
Hence, on any boundary where the traction is specied

z1(z) + 1(z) + 2 (z) = f (z)
0 0
z = zboundary
Associate the following ! 1 2 ! , then the traction boundary condition
0
becomes
z (z) + (z) + (z) = f (z) z = zboundary
0
A special case of importance is that of a traction free boundary, there we have

z (z) + (z) + (z) = 0
0
z = zboundary
The application of this useful formula will be demonstrated on the elliptical hole
problem.
Structure of (z) and (z)
In some cases it may be possible to obtain (z) and (z) by integration, but typically
they are determined by synthesizing particular solutions.
In a nite, simply connected region, the functions (z) and (z) are single valued
analytic functions of z and therefore can be represented by
X 1
X 1
(z) = anzn (z) = bn zn

o o
Note that each term in the series is a harmonic function.
If the region is not simply connected then and need not be single valued and
other solutions are possible. For the displacements and stresses to be single valued in
a nite multiply connected region, the functions are represented by
Xm
(z) = ; 2(11+ ) (F1k + iF2k ) log(z ; zk )] + o(z)
k=1
Xm
(z) = ; 2(1 + ) (F1k ; iF2k ) log(z ; zk )] + o(z)
k=1
X n 1
X 1
o(z) = anz o (z) = bn zn

o o
where F1k F2k
are force resultants on each contour, zk is an arbitrary point in each
`hole', and m is the number of contours.
g

;
1
;

nite nite innite
simply connected multiply connected multiply connected
Figure 5.21: Types of connected regions.
In an innite region with several internal contours
(z) = ; (2F1(1++iF2)) log(z) + 41 (11 + 22 )]z + o
1 1
(z) = ; 2(F(1 1 + iF2 ) log(z ) + 1

+ ) 2 (11 ; 22 ) + i12 ]z + o
1 1 1
X 1
X
1
o(z) = an z n ;
o(z) = bnz n ;
o o
. August 2007
Note that the added series give stresses that decay at innity.
Example 5.9: Express the case of a uniform stress distribution using complex
variables.
The appropriate stress functions and their derivatives are
(z ) = A1 z (z ) = B1 z
(z ) = A1
0
= B1
0
= 0
00
The stresses are

xx + yy = 4Re A1 ] = 4Re A11 + iA12 ] = 4A11
xx ; yy + 2i xy = 2z 0 + B1 ] = 2B11 + iB12 ]
These can be separated out to give
xx = 2A11 ; B11 yy = 2A11 + B11 xy = B12
This corresponds to a uniform stress state. These equations may also be rearranged
as
B12 = xy
1
B11 = 21 yy ; xx]
1 1
A11 = 14 yy + xx]
1 1
This form allows us to determine the coecients from knowledge of the remote
stresses. Note that we can always set A12 = 0.
Example 5.10: An innite sheet has a circular hole with arbitrary tractions
applied around the edge of the hole and a constant remote stress. Determine the
stress distribution.
This problem is a generalization of the Kirsch solution obtained earlier in the
chapter. This example also serves as an application of Fourier series.
Choose the stress functions as
X X X
(z ) = A 1 (z ) = B 1 (z ) = ; A n
1 1
n zn n zn n z n+1
0 0 00
o o
since this gives constant stresses at innity. The tractions around the hole are related
to rr and r , hence obtain
rr ; i r = (z) + (z) ; z (z ) + (z)]e2i
0 0 00 0
X 1 X 1 h X nz X 1 i 2i
= An zn + An zn ; ; An zn+1 + Bn zn e
Let z = rei and regroup to get
Xh i in X in
rr ; i r = An (n + 1) ; r12 B(n+2) e rn + An err ; Bo ei2 ; B1 1r ei
1 ; 1
o o
yy
1
6 1
- yx
xy
1
ppp p p p p p p p pp r
y 6
- xx
1
p p p ap p p p p p
*

x
-
;

;
Figure 5.22: Generalized hole problem.

These stresses are recognized as a Fourier series in for xed r. Hence, let the
applied tractions at r = a be represented by
X
tr ; itr ] = rr ; i r ]r=a
1
Cnein
n= ;1
That is, the arbitrary traction distribution is specied through the Fourier coe-
cients Cn . The stress coecients are related to these by
Xh i in X in i X
A (n + 1) ; 1 B e + A e ; B ei2 ; B e =
1 1 1
C ein
;
n a2 n+2 an n an o 1 a n
o o n=
;1
Equate the corresponding powers of to give the simultaneous equations necessary

to solve for the coecients:
An a1n = Cn n 3
A 1 ; B = C
2 a2 o 2
A1 a1 ; B1 a1 = C1
Ao ; a12 B2 + Ao = Co
Ao(n + 1) a12 ; Bn+2 an1+2 = C ; n n1
Solving these gives the recurrence relations for the coecients as
An = anCn n3
Bn = ;a C n+2 + An 2 (n ; 1)a2 n 3
n ; ;
Since Ao and Bo are associated with the stresses at innity, let these be known also,
hence the starter values for the recurrence relations are
A2 = a2Bo + a2 C2
B2 = (Ao + Ao )a2 ; a2Co ; B1 + A1 = aC1
. August 2007
We need an additional relation to obtain B1 and A1 , separately.
Look at the contributions B1 and A1 to the displacement. In particular, consider
=A 1
0
1z =B 1 0
1z
= A1 ln z = B1 ln z
The displacements are
+ 3
1
2
ur + iu ] =
A1 ln z ; zA1 zz ; B1 ln z e i ;
+ 3

2 i i
= +
A1 ln r + i ] ; A1 e ; B1 ln r ; i ] e
;
Looking at these displacements at = 0 and = 2 gives, respectively,

2
ur + iu ] = ( )A1 ln r + 0] ; A1 ; B1 ln r ; 0]

2
ur + iu ] = ( )A1 ln r + i2] ; A1 ; B1 ln r ; i2]
The dierence is + 3

+
A1 i2 + B1 i2
Therefore, to have single valued displacements requires that this dierence be zero.
That is,
B1 = ; ++3
A1
This completes the set of equations. Some of the interesting special cases are dis-
cussed next.
If the applied tractions around the edge of the hole are zero, then Cn = 0 giving
A2 = a2 Bo
An = 0 n 3
B1 = A1 = 0
B2 = (Ao + Ao )a2
B3 = A1 2a2 = 0
B4 = A2 3a2 = 3a4 Bo
Bn = 0 n 5
The stress functions for this case are therefore
2 2 4
= Ao + Bo az2
0
= Bo + (Ao + Ao) az2 + Bo 3 az4
0
Let the stresses at innity be

xx = 0 xy = 0 yy = 1
Hence the coecients are

Ao = 41 1
Bo = 21 1
and the stress functions become
2 2 4
= 0 1 1 + 2 a ]
1
= 21 1 + az 2 + 3 az4 ]
0 1
4 z2
2
00
= 14 4;4 az 3 ]
1
(5.2)
These recover the solution to the Kirsch problem as obtained earlier in the chapter.
The case of a hole with a uniform pressure is easily obtained by setting
Cn = 0 except Co 6= 0
Ao = 0
Bo = 0
The elegance of the Fourier series approach is that any traction distribution can
be represented simply by changing the coecients Cn . In particular, for the point
loaded hole set Cn = 1.
The above solutions are represented in terms of innite series however, they
can be truncated depending on the accuracy required. With this in mind, the fast
Fourier transform (FFT) algorithm can be used to do the summation very eciently
on a computer.
Example 5.11: Consider an elliptical hole in an innite sheet under remote

tension. Determine the stress distribution.
This, of course, is analogous to the Kirsch problem what makes it dierent is
that the hole is not circular and therefore it is necessary to introduce a scheme for
treating such boundaries. One approach is to introduce elliptical coordinates and
transform all the governing equations similar to what was done earlier for cylindri-
cal coordinates. Instead, we introduce a much more powerful technique that uses
conformal mappings.
yy
1
6 1
- yx
xy
1
y 6
- xx
1 6
2b z = x + iy = rei
6
r
?
-
*

x 2a
ppp p p p p p p p pp
-
;1

; pppppp pp = + i = ei#
Figure 5.23: Geometry of an elliptical hole.

. August 2007
The conformal mapping idea is quite general but to concretize the method we
will consider only the case of an ellipse. Let z = x + iy be the physical plane and
= + i be the transform plane. The mapping that relates the two is then
b
z = f () = c + m 1 ] c = a +2 b m = aa ; +b
In cylindrical coordinate form with z = rei and = ei# , we have
h i
rei = c e+i# + m
e i# ;
A special cases of this is, for example, at # = 0

h i
rei = c + m

Hence if = 1 then x = c1 + m] y = 0 which corresponds to the end of the ellipse.
Another special case is when # = 90o , then
rei = ci ; im ]
Now if = 1 then x = 0 y = +c1 ; m] which corresponds to the top of the ellipse.
The traction free boundary conditions, expressed in terms of the conformal map-
ping, becomes
f () d d + + d d = 0 =
d dz d dz boundary
We wish to nd a set of stress functions that satises this condition.
Based on the behavior of a circular hole, let the stress functions be
(z ) = () = A0 ; A2 1 (z) = () = Bo ; B2 1 ; B4 313
giving for their derivatives
(z) = d (z ) = fA + A 1 g d
0
dz o 2 2 (z) = fB + B 1 + B 1 g d
dz
0
o 2 2 4 4 dz
Substituting these into the traction free boundary condition, requires that
C + m ]fAo + A2 12 g d
dz + f Ao ; A2
1 g + fB ; B 1 ; B 1 g d = 0
o 2 4 33 dz
with z = zboundary . On the boundary = ei# = ei# therefore, the mapping is given
as
dz = c1 ; m ] or d = 2
d 2 dz c(2 ; m)
Hence the loading condition now becomes
i#2
ce i# + me+i# ]fA + A e i#2 g e + (ei2# ; m)fAo e i# ; A2 ei# g
;
o 2 ;
c
;
i2#
+ e c fBo ei# ; B2 e i# ; B4 e i3# g = 0
; ;
Equating equal powers of # results in
ei3# : mAo ; A2 + 1c Bo = 0
ei# : Ao + mA2 + Ao + mA2 ; 1c B2 = 0
e i# :
;
A2 ; mAo ; 1c B4 13 = 0
giving the solution
A2 = mAo + 1c Bo
1 B = A + A + m(A + A ) = A + A + m2 (A + A )
c 2 o o 2 2 o o o o
1 B = ;mA + mA + 1 B = 1 B
3c 4 o 0 c o c o
The stress functions can now be written as
2
(z ) = fAo + mAo + 1c Bo] 12 g c(2; m)
0
2
(z) = fBo + (1 + m2 )(Ao + A0) + m
0
C (B o + o )] c2 + 3Bo 14 g 2
B c( ; m)
Imposing the remote stress limits of xx = 0, xy = 0, and yy = , gives
1
Ao = 41 c 1
Bo = 12 c 1
The stress functions are then

2
0
(z ) = 21 f1 + (m + 2) 12 g (2 ; m)
1
2
(z ) = 21 f1 + ((1 + m2 ) + 2m) 12 + 314 g (2 ; m)
0 1
(5.3)
These can now be used to obtain the stresses at any point.
Compare this solution with the stress function for the circular hole. Obviously
if a = b then m = 0 and we recover the Kirsch solution. But more importantly, we
note that both have the same structure with the associations $ z and that the
geometry appears in the 2 =(2 ; m) term.
We now consider some special cases. The stress invariant is given by
2
x + y = Re f1 + (m + 2) 12 g (2 ; m)
1
z = c + m ]
A special case of considerable interest is the stress along the inner boundary. Here
one of the normal stresses is zero allowing the other to be determined directly from
the invariant as
e 2i#
t = 1 ;2 i#
Re f1 + (m + 2)e g (e2i# ; m)
. August 2007
The maximum stress occurs at # = 0 and is
1 (3 + m)
t = Re f1 + (m + 2)g 1 ; m = (1
1 1
; m)
Note that if m = 0 the boundary is circular and then t = 3 . At the other
1
extreme, if m = 1 the boundary corresponds to that of a crack and then t = 1.

Consider this second case in more detail. In particular, consider when the radius is
sharp but not zero, then
2 a ; b a ; paR !
R = ba m= a+b = p
a + aR
This gives the maximum stress as
" p p # " p #
t = 1
3 aR + a ;p aR =
3a + p 1
4a +p2 aR = 1 + 2r a
1
a + aR ; a + aR 2 aR R
Consequently, the sharper the radius the higher the stress.
Exercises 169
Exercises
5.1 Use the Fourier series representation to solve the problem of a circular hole
with diametrically opposite point loads.
5.2 Solve the problem of a circular hole with a traction distribution given by
po sin2 .
5.3 Solve the problem of a crack loaded by a uniform pressure.
5.4 Use the Fourier series representation to solve the problem of a crack with op-
posite point loads at its center.
5.5 Determine if the following strain eld is compatible.
xx = 2x2 + 3y2 + z + 1 yy = x2 + 2y2 + 3z + 2 zz = 3x + 2y + z2 + 1
xy = 4xy yz = xz = 0
5.6 The stresses in a 3-D stressed body are
xx = ;A(L ; x)y xy = 18 A(h2 ; 4y2 ) = yx
with ij = 0 otherwise, and A L h being constants. Is this stress system in
equilibrium? If the problem is one of plane stress, are the strains compatible?
What are the displacements? Show that if u v dv=dx are zero at x = 0 y = 0
then the vertical deection of the line y = 0 is
v(x) = 6AE 3L ; x]x2
5.7 Let it be required to have a displacement eld of the form

v(x y) = Ax2 u(x y) =?
What should the functional form of u be so that all the eld equations are
satised?
5.8 Show that while the non-trivial stresses
2
xx = Py1 ; 3yb2 ]
(other stresses being zero) satises equilibrium, the corresponding strains are
not compatible.
5.9 What must be the relationship between the coecients A and B in order for
the following strains to be compatible?
xx = Ax2 + y2 ] yy = Ax2 + y2 ] 2xy = Bxy = yx others = 0
. August 2007
5.10 The tractions on the upper and lower surfaces of a rectangular block are
ty = x2 tx = 0
while on the ends they are zero. Prove that the stress system
xx = 0 yy = x2 xy = 0 = yx
is not a solution to the problem.
5.11 Show that while the stress function
(x y) = 12 Py2 1 ; y2 =(6b2 )]
gives stress that are in equilibrium, the corresponding strains are not compat-
ible.
5.12 Consider the Airy stress function
q
(x y) = A logn ( x2 + y2)
Sketch the stress distributions along a few coordinate lines. What are the
tractions along the surface x2 + y2 = a2 .
5.13 What class of problems is solved by the following Airy stress function?
(x y) = Ax2 + Bxy + Cy2
5.14 Consider the following polynomial stress function
(x y) = Ax2 + Bx2 y + Cxy2 + Dy3
Under what circumstance(s) is it bi-harmonic? Use it to solve the problem of
pure bending of a prismatic bar.
5.15 Show that the following polynomial stress function
(x y) = Axy + Bx3 + Cx3y + Dxy3 + Ex3 y3 + Fxy5
can be used to solve the rectangular dam problem. Note that since this poly-
nomial is not symmetric in x and y that the orientation of the axes must be
chosen appropriately.
5.16 Motivated by the desire to use Fourier series to represent the applied tractions,
it is proposed to use the following stress function
(x y) = cos(nx=L)f (y)
where n = 0 1 : : : and L is a constant. Determine the allowable form for f (y)
for this to be an acceptable Airy stress function. If the applied tractions are
represented as X
P (x) an cos(nx=L)
n
determine the stress function in terms of an .
Exercises 171
5.17 From the previous problem, show that the stress xx at a point on the surface
of a half-plane is a compression equal to the applied pressure at that point.
5.18 If the body forces are absent, show that the displacements can be given by the
Airy stress function as
2
ux = ; @@x + 1 + x @
4 @y 2
u y = ; @ + 1 + x @
@y 4 @x
@. 2
where r2 = 0 and r2 = @x@y
5.19 An innite plate containing a circular hole is subjected to a pure shear stress
at innity. Given the Airy stress function
(r ) = Ar2 + Br 2 + C ] sin
;
Find A, B , and C . Plot around the hole.

5.20 Show that the problem of a curved cantilever beam with an end shear force can
be solved with the following stress function.
(r ) = Ar3 + Br 1 + Cr logn r] sin
;
5.21 Show that Flamant's problem of a point load on a half-plane can be solved with
the following stress function.
(r ) = Ar sin
5.22 A ring, inner radius b and outer radius a, is split and the two ends moved
radially apart an amount . Show that the following stress function solves the
problem.

(r ) = ; (2
+1) r logn r sin + 2(1 +1 2 ) 2 a2 sinr ; a12 r3 sin = ab
5.23 A rigid disk has a resultant moment applied to it. What is the simplest distri-
bution of traction on its edge that will keep it in equilibrium?
5.24 A rigid disk is solidly embedded in an innite sheet. Determine the stress
distribution in the sheet due to an applied moment acting on the disk.
5.25 An elastic disk is bonded to a rigid ring. The composite disk rotates with
angular constant speed '. Find the stress and displacement elds.
. August 2007
rr r
r2 2 0 2
log r 1=r2 0 ;1=r2
0 1=r2 0
r2 log r 2 log r + 1 0 2 log r + 3
r2 2 ;1 2
r3 cos 2r cos 2r sin 6r cos
r3 sin 2r sin ;2r cos 6r sin
r sin 2 cos =r 0 0
r cos ;2 sin =r 0 0
r log r cos cos =r sin =r cos =r
r log r sin sin =r ; cos =r sin =r
cos =r ;2 cos =r3 ;2 sin =r3 2 cos =r3
sin =r ;2 sin =r3 2 cos =r3 2 sin =r3
r4 cos 2 0 6r2 sin 2 12r2 cos 2
r4 sin 2 0 ;6r2 cos 2 12r2 sin 2
r2 cos 2 ;2 cos 2 2 sin 2 2 cos 2
r2 sin 2 ;2 sin 2 ;2 cos 2 2 sin 2
cos 2 ;4 cos 2 =r2 ;2 sin 2 =r2 0
sin 2 ;4 sin 2 =r2 2 cos 2 =r2 0
cos 2 =r2 ;6 cos 2 =r4 ;6 sin 2 =r4 6 cos 2 =r4
sin 2 =r2 ;6 sin 2 =r4 6 cos 2 =r4 6 sin 2 =r4
rn cos n ;n(n ; 1)rn 2 cos n
;
n(n ; 1)rn 2 sin n
;
n(n ; 1)rn 2 cos n
;
rn sin n ;n(n ; 1)rn 2 sin n

;
;n(n ; 1)rn 2 cos n
;
n(n ; 1)rn 2 sin n
;
rn+2 cos n ;(n + 1)(n ; 2)rn cos n (n + 1)nrn sin n (n + 2)(n + 1)rn cos n
rn+2 sin n ;(n + 1)(n ; 2)rn sin n ;(n + 1)nrn cos n (n + 2)(n + 1)rn sin n
cos n =rn ;(n + 1)n cos n =rn+2 ;(n + 1)n sin n =rn+2 (n + 1)n cos n =rn+2
sin n =rn ;(n + 1)n sin n =rn+2 (n + 1)n cos n =rn+2 (n + 1)n sin n =rn+2
cos n =rn 2 ;
;(n + 2)(n ; 1) cos n =rn ;n(n ; 1) sin n =rn (n ; 1)(n ; 2) cos n =rn
sin n =rn 2 ;
;(n + 2)(n ; 1) sin n =rn n(n ; 1) cos n =rn (n ; 1)(n ; 2) sin n =rn
Table 5.1: Stresses

Exercises 173
2
ur 2
u
r2 ( ; 1)r 0
log r ;1=r 0
0 ;1=r
r2 log r ( ; 1)r log r ; r ( + 1)r
r2 ( ; 1)r ;( + 1)r log r
r3 cos ( ; 2)r2 cos ( + 2)r2 sin
r3 sin ( ; 2)r2 sin ;( + 2)r2 cos
2r sin ( ; 1) sin + ( + 1) log r cos ; cos ( ; 1) cos ; ( ; 1) log r sin ; sin
2r cos ( ; 1) cos + ( ; 1) log r sin ; sin ;( ; 1) sin ; ( + 1) log r cos ; cos
2r log r cos ( + 1) sin + ( ; 1) log r cos ; cos ( + 1) cos ; ( ; 1) log r sin ; sin
2r log r sin ;( + 1) cos + ( ; 1) log r sin ; sin ( + 1) sin + ( ; 1) log r cos + cos
cos =r cos =r2 sin =r2
sin =r sin =r2 ; cos =r2
r4 cos 2 ;(3 ; )r3 cos 2 (3 + )r3 sin 2
r4 sin 2 ;(3 ; )r3 sin 2 ;(3 + )r3 cos 2
r2 cos 2 ;2r cos 2 2r sin 2
r2 sin 2 ;2r sin 2 ;2r cos 2
cos 2 ( + 1) cos 2 =r ;( ; 1) sin 2 =r
sin 2 ( + 1) sin 2 =r ( ; 1) cos 2 =r
cos 2 =r2 2 cos 2 =r3 2 sin 2 =r3
sin 2 =r2 2 sin 2 =r3 ;2 cos 2 =r3
rn cos n ;nrn 1 cos n
;
nrn 1 sin n
;
rn sin n ;nrn 1 sin n

;
;nrn 1 cos n
;
rn+2 cos n ;(n + 1 ; )rn+1 cos n (n + 1 + )rn+1 sin n

rn+2 sin n ;(n + 1 ; )rn+1 sin n ;(n + 1 + )rn+1 cos n
cos n =rn n cos n =rn+1 n sin n =rn+1
sin n =rn n sin n =rn+1 ;n cos n =rn+1
cos n =rn 2 ;
(n ; 1 + ) cos n =rn ;1 ;(n ; 1 ; ) sin n =rn 1 ;
sin n =rn 2 ;
(n ; 1 + ) sin n =rn ;1 ;(n ; 1 ; ) cos n =rn 1 ;
Table 5.2: Displacements.
. August 2007
Chapter 6
Nonlinear Elasticity Problems
In the general case of deformable bodies, we can have large displacements, large ro-
tations, and large strains. This renders the governing equations highly nonlinear
and therefore they can only be solved using computational methods. Furthermore,
because of the complicated load history dependence, this suggests a time or load
incremental type solution. We combine these two requirements into an incremen-
tal/iterative solution algorithm. The total Lagrangian scheme is introduced as a
specic example of a solution scheme and this is combined with Newton-Raphson
equilibrium iterations to give accurate solutions of the nonlinear equations.
An essential step is formulating the problem as a set of discrete unknowns this
is demonstrated using a nite element formulation. Furthermore, the eld equations
must be recast in terms of discrete systems and this is done via the principle of virtual
work.
6.1 Discretizing Continuous Media

An eective strategy for solving general nonlinear problems for continuous media is
to divide the body into many subregions each of which have a relatively simple defor-
mation distribution. We illustrate this process here while the process is approximate,
we can establish conditions under which the exact result can be obtained.
1-D Spatial Interpolations

Consider the 1-D space shown in Figure 6.1 and discretized as indicated by the black
dots. The values of the functions, ui, at the points are the discretized representa-
tion. Consider a typical line segment between two dots of length L divided into two
segments where the common point is at (x) as shown in Figure 6.2. Dene
h1 = L1 =L h2 = L2 =L hi = hi(x)
We have the obvious constraint that h1 + h2 = 1. The lengths of these segments
uniquely dene the position of the common point.
175
176 Chapter 6. Nonlinear Elasticity Problems
ui uj function
t t t t t t t t t tspace
ik jk
Figure 6.1: A discretized 1-D medium.

The position of a point (x) along the line can be written as
( ) " #( )
1 = 1 1 h1 (6.1)
x x1 x2 h2
where the subscripts 1 2 refer to the node. We can invert this to get the expressions
for the coordinates
( ) " #( ) " #( )
h1 = 1 x2 ;1 1 = 1 a1 b1 1
h2 L ; x1 1 x L a2 b2 x
with L x2 ; x1 . From this, it is apparent that functions of (x) can equally well be
written as functions of (h1 h2 ). That is, any function of interest can be written as
X
u(x) = 2i hi(x)ui = L1 x2 ; x]u1 + L1 ;x1 + x]u2 = h1(x) u1 + h2 (x) u2
where ui are the nodal values. The functions hi (x) are called interpolation functions.
PP PP
L1 L2 h1 (P
)P
Ph2 ( )
P
h1 (rP
)P
P
h (r )
PP 2
u c u s c PPPs - s

c PPs -r
1h (x) 2h
0 1 ;1 0 1
1h 2h
Figure 6.2: Coordinates for 1-D spaces. (a) Physical. (b) Natural. (c) Isoparametric.
In order for the two coordinates fh1 h2g to describe the single coordinate x, it
must be supplemented by the constraint h1 + h2 = 1. We can invoke this constraint
explicitly by introducing natural coordinates given as
h1 = 1 ; h2 = (6.2)
These are shown in Figure 6.2(b) where ranges from 0 to 1. We have for a typical
function X
u(x) = 2i hi( )ui = 1 ]u1 + ]u2
;
6.1. Discretizing Continuous Media 177
We can introduce yet a third set of coordinates which we will call isoparametric
coordinates given as
h1 = 12 1 ; r] h2 = 21 1 + r] (6.3)
These are shown in Figure 6.2(c) where r ranges from ;1 to +1. We have for a typical
function X
u(x) = 2i hi(r)ui = 12 1 ; r]u1 + 12 1 + r]u2
This form better generalizes for higher dimensions and is especially useful when nu-
merical methods are used for integration purposes.
;
s
h1 (r)
1
s
h3 (r)
0 +1
sr
h2 (r )
-
;1
s s
; 13
s
+ 13 +1
s -r
1h 3h 2h 1h 3h 4h 2h
Figure 6.3: Higher order interpolations. (a) 3-Node. (b) 4-Node.

For any segment length, we can increase the order of function by increasing the
number of nodes. Thus Figure 6.3(a) has 3 nodes and therefore allows for the inter-
polations
h1 = 21 1 r] 1 1 ; r2 ] h2 = 12 1 + r] 1 2 h3 = 1 r2 ]
; ;
2 ;
2 1 ; r ] ;
Note that each function has the value 1 at its corresponding node. These expressions
could be simplied but the current form best expresses the hierarchical form of the
interpolations. In a similar way, the addition of a fourth node allows the interpolations
h1 = 12 1 r] 12 1 r2 ]
; ; ; ;
1 2 3
16 1 + 9r ; r ; 9r ]
h2 = 21 1 + r] 21 1 r2]
; ; ;
1 2 3
16 1 + 9r ; r ; 9r ]
h3 = + 1 r2] ; ;
1 2 3
16 7 + 27r ; 7r ; 27r ]
h4 = 1 2 3
;
16 9 + 27r ; 9r ; 27r ]
where the hierarchical form is clear. What the higher order interpolations aord,
quite obviously, is the use of higher order functions.
In the form presented, the interpolations also apply to the general curved line in
space that deforms into another general line in space. That is, we can think of the
coordinates themselves as being interpolated. To clarify this point, let the original
coordinates be designated with a superscript ò', then any point along the line is given
by
XN
xo = hi (r)xoi
i=1
. August 2007
where (xoi) are the original coordinates of the nodes. Thus each (r) coordinate has a
corresponding (xo ) location.
A typical variable is therefore represented by
X X
xo = Ni hi (r) xoi u = Ni hi(r) ui
where ui and xi are the nodal values and N is the total number of nodes. The element
strains, for example, are obtained in terms of derivatives of element displacements.
Using the isoparametric coordinate system, we get, for example,
@ = @r @
@xo @xo @r
The above interpolation has xo as a function of r, but to evaluate the derivatives of
r with respect to xo we need to have the inverse relation between the two sets of
variables. This is obtained as
@ = @xo @ or @ g = J ]f @ g
e
@r @r @x o f
@r @xo
The notation we use anticipates the generalizations needed when we consider multiple
dimensions. The function Je ] is called the Jacobian operator relating the isoparamet-
ric coordinates to the physical coordinates. The relation for the derivatives requires
@ g = J 1 ]f @ g ;
f
@xo e @r
which requires that Je 1 ] exists. In most cases, the existence is clear however, in
;
cases where the segment is much distorted or folds back on itself the Jacobian trans-
formation can become singular.
Example 6.1: Determine the relation between the strain and the degrees of
freedom of a three-noded line segment.
s
xo1
1i
s
xo3
3i 2i
s s
xo2 -xo ;1
1i
0s
3i
s
+1-r
2i
Figure 6.4: Three noded 1-D element.

Let the ends of the segment have positions xo1 and xo2 and let the middle node
be at xo3 as shown in Figure 6.4. The position of an arbitrary point is given by
X3
xo = o o o
i hi (r)xi = h1 (r)x1 + h2 (r)x2 + h3 (r)x3
o
= 21 1 ; r] ; 12 1 ; r2 ]]xo1 + 21 1 + r] ; 21 1 ; r2 ]]xo2 + 1 ; r2 ]xo3
= 1 r;1 + r]xo + 1 r1 + r]xo + 1 ; r2 ]xo
2 1 2 2 3
The derivative of the coordinate is
@xo = 1 ;1 + 2r]xo + 1 1 + 2r]xo + 0 ; 2r]xo = 1 xo ; xo ] + xo ; 2xo + xo ]r = J
@r 2 1 2 2 3 2 2 1 1 3 2 e
Typically, Je is a function of the coordinate r but in this case where
xo2 = xo1 + L xo3 = xo1 + L=2
then
Je = L=2 Je 1 = 2=L
;
and is constant.
The displacement at an arbitrary interpolated point is given by
X3
u(xo ) = u(r) = i hi (r)ui
and the strain is therefore
@u = @r @u = J 1 ] @u = J 1 ]X @hi u
xx = @x e @r e i @r i
; ;
o @xo @r
h i
= L2 12 ;1 + 2r]u1 + 12 1 + 2r]u2 + 0 ; 2r]u3
= L1 u2 ; u1 ] + Lr u1 ; 2u3 + u2 ]
This gives a linear distribution of strain along the segment. Furthermore, in the
limit of small segment size, we would get u3 12 (u1 + u2 ) causing the second term
to be zero leaving a constant strain state given by xx = u2 ; u1 ]=L. As will be
seen, this is an essential requirement for convergence.
Hexahedral Discretization for 3-D Bodies

Consider the general blocks of material in Figures 6.5(a) & 6.6(a) undergoing defor-
mation. We will describe the deformation just in terms of the edge behaviors in turn,
these edges are discretized using nodal values as just described. Both 8-noded and
20-noded hexahedrals are considered.
Consider the 8-noded hexahedral solid shown in Figure 6.5(a), in the isoparametric
coordinates of Figure 6.5(b) it is a cube with sides of length 2. The coordinates are
given by
X X X
xo = 8i hi(r s t)xoi yo = 8i hi(r s t)yio zo = 8i hi (r s t)zio
where the interpolation functions are
h1 = 81 (1 ; r)(1 ; s)(1 ; t) h5 = 81 (1 ; r)(1 ; s)(1 + t)
h2 = 18 (1 + r)(1 ; s)(1 ; t) h6 = 81 (1 + r)(1 ; s)(1 + t)
h3 = 18 (1 + r)(1 + s)(1 ; t) h7 = 18 (1 + r)(1 + s)(1 + t)
h4 = 81 (1 ; r)(1 + s)(1 ; t) h8 = 81 (1 ; r)(1 + s)(1 + t)
. August 2007
t t t
4i 6
zo s 6 4i
@

t t
@
8i
8i
t
t
@
a @t 3i
1h 3i
@
t 1i
t t
@ @

5i
@
@ t 7i@
i @ a@
@a

x o
7 it
i 5 ti
t
@ @
@ 2 @ 2

t @ @

y@
Ro
@
@
r@
R
6i @
6it

Figure 6.5: Deformation of a straight edged 3-D block using the Hex8 discretization.
(a) Physical coordinates. (b) Isoparametric coordinates.
These functions have the value unity at each of the indicated nodes which additionally
coincides with the subscripted number. For future reference, we can write these in
the shorthand notation
hi = 81 (1 + rir)(1 + sis)(1 + tit)
where (ri si ti) are the nodal coordinates and are specied in Table 6.1.
t t
s 6u4i
4i 6 zo t @
t12 @11t@

8i
t t 8iu
16
a
t u3i
@ @
1h 3i t u 1i
t t t
@
t20 19@13
@@ 9 10 t
15
5i t u

@
i @ t
i
7 @ iu a 7 @ a

xo
5 @ u2i
t
@ @ 17 18 t
@2 i
t @t @
yR
@o @
t
14

r@
i i
@ u
R
@
6 6
Figure 6.6: Deformation of a curved 3-D block using the Hex20 discretization.
(a) Physical coordinates. (b) Isoparametric coordinates.
The power of the isoparametric approach to discretization is that the procedure
is easily generalized to higher order interpolation functions. We illustrate the idea
with the curved block shown in Figure 6.6(a). The block is discretized in terms of 20
points, the rst eight being the same as for the linear hexahedron and the additional
twelve corresponding to mid-edge nodes | the numberings for these nodes are shown
in Figure 6.6(b). The coordinates of an arbitrary point are then given by
X X X
xo = 20i hi (r s t)xoi yo = 20i hi(r s t)yio zo = 20i hi (r s t)zio
i 1 2 3 4 5 6 7 8
ri ;1 +1 +1 ;1 ;1 +1 +1 ;1
si ;1 ;1 +1 +1 ;1 ;1 +1 +1
ti ;1 ;1 ;1 ;1 +1 +1 +1 +1
i 9 10 11 12 13 14 15 16 17 18 19 20
ri 0 +1 0 ;1 ;1 +1 +1 ;1 0 +1 0 ;1
si ;1 0 +1 0 ;1 ;1 +1 +1 ;1 0 +1 0
ti ;1 ;1 ;1 ;1 0 0 0 0 +1 +1 +1 +1
Table 6.1: Coecients for interpolation functions.
where the interpolation functions are
i=1 8: hi = 18 (1 + rir)(1 + sis)(1 + tit)(rir + sis + ti t ; 2)
i = 9 11 17 19 : hi = 14 (1 ; r2)(1 + sis)(1 + ti t)
i = 10 12 18 20 : hi = 41 (1 + rir)(1 ; s2 )(1 + ti t)
i = 13 14 15 16 : hi = 41 (1 + rir)(1 + sis)(1 ; t2 )
As before, (ri si ti) are the nodal coordinates and are specied in Table 6.1.
The element strains are obtained in terms of derivatives of element displacements.
Using the isoparametric coordinate system, we get, for example,
@ = @ @r + @ @s + @ @t
@xo @r @xo @s @xo @t @xo
But to evaluate the derivatives of (r s t) with respect to (xo yo zo ), we need to
have the explicit relation between the two sets of variables. As illustrated earlier, we
rst form the inverse relation according to
8 @ 9 2 @xo @yo @zo 3 8 @ 9
>
> >
> > >
> 66 @r @r @r 77 > @x o>
< @ = 6 @xo @yo @zo 7 < @ >
@r > > = @ g = J ]f @ g
= 66 77 or e
>
> >
@s > 64 @so @so @so 75 > @y > > o > f
@r @xo
:@
> > @x @y @z : @ > >
@t @t @t @t @zo
where Je ] is called the Jacobian operator relating the isoparametric coordinates to
the global coordinates. The relation for the derivatives requires
@ g = J 1 ]f @ g
;
f
@xo e @r
which requires that Je 1] exists. The structure of Je ] is
;
2 3
XN hir xoi hir yio hir zio
Je ] = 64 hisxoi hisyio hiszio 75
i hit xoi hit yio hit zio
. August 2007
where the subscript `comma' indicates partial dierentiation. This can be expressed
in matrix form as
2 3
2 3 xo1 y1o z1o
h1r h2r h3r h4r hNr 66 xo2 y2o z2o 77
6
Je ] = 4 h1s h2s h3s h4s hNs 75 66 .. .. .. 77 = @1 h] X ]
h1t h2t h3t h4t hNt 4 .o .o .o 5
xN yN zN
The derivatives for the linear hexahedron Hex8, for example, are given by
@xo = ; 1 (1 ; s)(1 ; t)xo + 1 (1 ; s)(1 ; t)xo + 1 (1 + s)(1 ; t)xo
@r 8 1 8 2 8 3
1 o 1
; (1 + s)(1 ; t)x4 ; (1 ; s)(1 + t)x5 +
o
8 8
o
@x = ; 1 (1 ; r)(1 ; t)xo ; 1 (1 + r)(1 ; t)xo + 1 (1 + r)(1 ; t)xo
@s 8 1 8 2 8 3
+ 18 (1 ; r)(1 ; t)xo4 ; 81 (1 ; r)(1 + t)xo5 +
with analogous expressions for the other derivatives. It is therefore clear that both
Je ] and Je ] 1 depend on (r s t) as well as the nodal coordinates however, it is
;
independent of the absolute global position of the element. If the block is rectangular,
the Jacobian matrix is constant with diagonal only terms. Additionally, the product
of the diagonal terms (which is the determinant) is related to the volume as Vcube o =
detjJej 8.
In general, integrals involving the deformation will need to be performed numeri-
cally this is demonstrated later in the chapter after the form of the system matrices
are established.
A nal point of crucial importance concerns the compatibility of many elemen-
tal blocks after assemblage. Each block uses the same interpolation functions, and
shared faces use the same nodes consequently, the complete shared face has the same
displacement shape and no gaps are formed. (This is true even for very large deforma-
tions.) The blocks are therefore compatible. While the displacements are continuous
across blocks, derivatives of displacement are not necessarily so that is, the strain
distribution may be discontinuous. This raises the issue of convergence to the exact
(continuous) solution | we leave that to later when the eld equations have been
reformulated.
Example 6.2: Establish the explicit form for determining the displacement gra-
dients and the strains for the Hex20 element.
We have for a typical displacement function
X
u(xo yo z o ) = 20
i hi (r s t)ui
where ui are the nodal values of the function. The displacement gradients can
therefore be computed as
@u = X @hi u @v = X @hi v @v = X @hi v
@xo i @xo i @xo i @xo i @yo i @yo i
and so on. By use of the Jacobian, we have
f @x@ o g = Je 1 ]f @r
;
@g
so that we can express the displacement gradients in the matrix form
8 9
> u1 >
8u 9 2 h h h h 3>> u2 >
>
< x= 1r 2r 3r 20r < =
u y = J e
1 ] 4 h1s h2s h3s h20s 5 u3
:u > >
;
h1t h2t h3t h20t > .. >

z > . >
: u20
where as before the subscript `comma' indicates partial dierentiation. Introduce
the matrix of derivatives as
8 o 9 2h 3 2A 3
< x = ir x
y o ;! Je 1 ] 4 his 5 4 Ay 5 (6.4)
: o
;
z hit i=1N Az i=1N

The same matrix A ]i gets applied to each of the displacements. The subscript i is
associated with the shape function (and hence also the node). Arrange the degrees
of freedom as
f u g = fu1 v1 w1 u2 v2 w20 gT
and dene the expanded nodal matrix of derivatives BD ]i
8u 9 2A 0 0 3
>
> x >
> x
> u y > 66 Ay 0 0 77
>
> uz > > 66 Az 0 0 77
>
> > 66 0 Ax 0 77
< vx > =
fuxg = > vy > BD ]i = 666 0 Ay 0 777
>
> vz >> 66 0 Az 0 77
>
> w x >
> 66 0 0 Ax 77
>
> w > 4 0 0 Ay 5
: wy > 0 0 Az i=1N
z
The vector of displacement gradients can then be evaluated from
fu x g = BD ]f u g BD ] = 9 60] = BD1 BD2 BD20 ]
The interpretation is that BD ] is expanded left to right for each node in the element.
As formulated, neither Je ] nor the A ]i matrices changes during the deformation
and hence neither does the displacement gradient operator BD ] however, since this
is dierent for each element and for each integration point in the element, then it is
usually computed òn-the-y' during a deformation.
Typical terms for the strains are obtained from fu x g according to
Exx = u x + 21 u 2x +v 2x +w 2x ]
Eyy = v y + 21 u 2y +v 2y +w 2y ]
2Exy = u y +v x +u x u y +v x v y +w x w y ]
. August 2007
The vector of displacement gradients is thus considered the prime quantity to be
evaluated.
Example 6.3: Evaluate the ability of the Hex20 interpolation functions to de-
termine strains.
b = 2:54 mm (1:0 in)
h = 2:54 mm (1:0 in)
L = 254 mm (10:0 in)
aluminum 10 4 ;
Figure 6.7: Bent block of square cross-section. 20 2] mesh.
The block in Figure 6.7 is given the deformation corresponding to Figure 2.7.
That is, each node is given the exact displacements corresponding to Equation (2.6)
and the interpolation functions are used to compute the strains. The middle plane
of the block shown should not experience any stretching.
.4
.3
.2
Strain
.1
.0
-.1
-.2
Increment Increment
-.3
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Figure 6.8: Strain comparisons. (a) 10 1] mesh. (b) 20 2] mesh.

Figure 6.8 shows the nodal strain for two levels of discretization circles corre-
spond to corner nodes while triangles correspond to mid-edge nodes. It is clear that
two elements through the thickness enables an accurate evaluation of the strains
from the given displacements.
6.2 Equilibrium of Discretized Systems

The formulation of a problem in terms of a set of governing dierential equations plus
a set of boundary conditions is known as the strong formulation of the problem. This
6.2. Equilibrium of Discretized Systems 185
is not always the most convenient form for solving actual problems and is especially
true for nonlinear problems where the displacement elds are approximated as just
discussed. We now recast the equilibrium developments of Chapter 3 in the form
suitable for discretized systems. The underlying principle invoked is that of virtual
work.
Virtual Work Formulation of Equilibrium

Consider a typical force-de"ection or stress-strain curve as shown in Figure 6.9. The
elasticity requirement, as discussed in depth in Chapter 4, is that both the loading
and unloading paths coincide. That chapter also introduced some work and energy
concepts which we utilize here.

; Pi ui P
; 6 xxx
B
A
xx
xx
-
(a) (b) u - y
x
Figure 6.9: Typical elastic behavior. (a) Discretized arbitrary body. (b) Force-
displacement behavior. (c) Stressed innitesimal element showing strain.
Let ui(xoi) be the displacement eld which satises the equilibrium equations in
V . On the surface A, the surface traction ti is prescribed on At and the displace-
ment on Au. Consider a variation of displacement ui (we will call this the virtual
displacement), then
ui = ui + ui
where ui satisfy the equilibrium equations and the given boundary conditions. Thus,
ui must vanish over Au but be arbitrary over At that is, ui must satisfy the
geometric constraints of the problem.
Let We be the external virtual work done by the body force bi (which could
include inertia eects) and the traction ti that is,
Z Z Z
We = bi uiV + ti ui dA + ti ui dA (6.5)
V At Au
The last term is zero hence we can replace At in the second integral with A on the
understanding that ui can not be varied over the portion Au. We can then express
this virtual work as
Z Z
We = bi uidV + jiuinj dA
V A
. August 2007
Z Z
= bi uidV + @x@ (jiui)dV
V V j
Z h @ ji i Z @ui )dV
= bi + @x uidV + ji ( @x
V j V j
Z h @ ji i Z
= bi + @x uidV + jijidV
V j V
where the last term was reduced using the decomposition of the deformation gradient
into ij + !ij and noting that the contraction of the antisymmetric rotation with the
symmetric stress is zero. We also used the integral theorem of Equation (1.3).
Dene the total virtual work as
Z Z h ji i
W = We ; jijidV = bi + @ @xj uidV
V V
These developments actually paralleled what was done in deriving the Cauchy stress
equations of motion in Chapter 3, therefore, we can look at it in one of two ways.
First, because the term in brackets is zero due to equilibrium, then we conclude that
the total virtual work is zero. That is,
Z
W = We ; jijidV = 0
V
On the other hand, if we say that the total virtual work is zero for every independent
virtual displacement ui, then we conclude that the term in brackets is zero. That
is, we obtain the equilibrium equations in terms of the Cauchy stress. Phrased more
formally, the principle of virtual work states that a deformable body is in equilibrium
if the total virtual work is zero for every independent kinematically admissible virtual
displacement. We will interpret the symbol as meaning a variation and the above
equation as a variational principle. A nal point worth mentioning is that the product
in the integral involves the Cauchy stress with the small Eulerian strain components
and not the Eulerian strain itself.
We would also like to write the virtual work expression in terms of the undeformed
conguration. Following developments similar to the example in Section 2.5 leading
to Equation (2.11), we get the relation
@x o @xo
mn = @x i @x j Eij
m n
The relation between Cauchy and Kirchho stress is
mn = o @x m @xn K
@xoi @xoj ij
and recalling that the deformed and undeformed volumes are related by dV = dV o o=,
the internal virtual work term becomes
mn mn dV = pqK Epq dV o
Hence the Cauchy stress / Eulerian (small) strain combination is virtual work equiv-
alent to the Kirchho stress / Lagrangian strain combination. The equivalent result
was established in Chapter 4 for the strain energy.
We are now in a position to write the virtual work form of equilibrium as
Z Z
W = We ; mn mn dV = We ; o pqK Epq dV o = 0 (6.6)
V V
In contrast to the dierential equations of motion, there are no added complications
using the undeformed state as the reference state. It is useful to realize that, during
a deformation, the reference state t = 0 could be any one of the previous equilibrium
positions and not necessarily the original stress-free state. We will make use of this
in our incremental formulation for the computer.
Some Particular Stationary Principles

The virtual work form is completely general, but there are further developments that
are more convenient to use in some circumstances. We now look at some of these
developments.
I: Stationary Potential Energy
A system is conservative if the work done in moving the system around a closed path
is zero. We say that the external force system is conservative if it can be obtained
from a potential function. For example, for a set of discrete forces, we have
X
Pi = ; @@uV or V = ;
i Pi ui
i
where ui is the displacement associated with the load Pi. The negative sign in the
denition of V is arbitrary, but choosing it so gives the interpretation of V as the
capacity (or potential) to do work. The external virtual work term now becomes
X X @V
We = iPiui = ; i @u ui = ; V
i
We get almost identical representations for conservative body forces and conservative
traction distributions. Follower forces, as well as dissipative forces such as friction,
are examples of nonconservative forces.
The internal virtual work is associated with the straining of the body and therefore
we will use the representation
Z Z
U = ij ij dV = o ijK Eij dV o
V V
and call U the virtual strain energy of the body. When straining is conservative,
we can view U as a potential function: examples of such materials were developed in
Chapter 4 and generally are referred to as hyperelastic materials.
. August 2007
The principle of virtual work for conservative systems can now be rewritten as
U + V = 0 or * U + V ] = 0 (6.7)
The term inside the brackets is called the total potential energy and this relation is
called the principle of stationary potential energy. We may now restate the principle
of virtual work as: For a conservative system to be in equilibrium, the rst-order vari-
ation in the total potential energy must vanish for every independent admissible virtual
displacement. Another way of stating this is that among all the displacement states of
a conservative system that satisfy compatibility and the boundary constraints, those
that also satisfy equilibrium make the total potential energy stationary. In compari-
son to the conservation of energy theorem, this is much richer, because instead of one
equation it leads to as many equations are there are degrees of freedom (independent
displacements).
If there are nonconservative forces (those for which we cannot write a potential)
the principle is extended to
X
U + V ] ; j Qj uj = 0 (6.8)
where Qj are the individual nonconservative forces. This is sometimes referred to
as the extended principle of stationary potential energy. Thus the main dierence
between the principle of virtual work and the (unextended) principle of stationary
potential energy is that the former is more general because it also applies to noncon-
servative systems which the latter does not.
Example 6.4: Determine the equilibrium conditions for the nonlinear system
shown in Figure 6.10.
F
F = Ku1 + u] U
...
....
...r LLLL r -
...
....
....
w P
7;! u u
*

(a) (b) V
Figure 6.10: Equilibrium of a nonlinear spring. (a) Force-de"ection relation. (b) Po-
tential energy.
Identify u, the resulting displacement at the point of application of the load,
as the independent admissible displacement. The response of the nonlinear spring
is shown in Figure 6.10(a): under tension it stiens, under compression it shows
softening. The virtual work for this spring is
W = Fu = K 1 + u]uu

This is also the virtual strain energy U . Integrating then gives
U = 21 Ku21 + 23 u]
The potential of the applied force is
V = ;Pu
The total potential energy of the system is, therefore,
* = 21 Ku2 1 + 23 u] ; Pu
These terms are shown plotted in Figure 6.10(b) for dierent values of displacement
u. It is apparent that * can achieve two stationary values | a valley and a peak.
The principle indicates that both occur at equilibrium positions.
A stationary potential energy requires that
@* = 0 ) Ku1 + u] ; P = 0
@u
We recognize this as the equilibrium balance between the external applied load P
and the internal force F of the spring. If the spring were linear ( = 0), it would
reduce to the single equilibrium equation
Ku = P or u = P=K
In the nonlinear case, however, we have two possible equilibrium positions
p
u = ;1 12+

4P=K P ;1
K
(The approximation is for slight nonlinearity when is small.) The rst is close
to the linear equilibrium position, but what is the meaning of the second position?
Furthermore, this second position corresponds to a negative displacement, which
surely cannot happen because the load is positive. A hallmark of nonlinear systems
is the possibility of multiple equilibrium positions for a given load state. Indeed,
looking at Figure 6.10(a), we see that even at zero load (F = 0) there is the equal
possibility of two deections. A crucial discussion is the distinction between the
two equilibrium points in terms of the stability of their equilibrium | we leave that
discussion to later.
It is hard to imagine an \ordinary" material or spring behaving in this way.
However, engineering structures and many of the structured materials (an example
is corrugated cardboard) do behave this way.
II: Discrete Systems

For computer solution of nonlinear problems, the governing equations must be re-
duced somehow to equations using discrete unknowns. That is, we introduce some
generalized coordinates (or degrees of freedom with the constrained degrees removed).
At present, we need not be explicit about which coordinates we are considering but
. August 2007
the situations discussed in Section 6.1 where the nodal displacements associated with
the discretization of a continuum are what we have in mind.
Accept that we can write a function as
u = u(u1 u2 : : : uN )
where ui are the generalized coordinates. We get these generalized coordinates by
the imposition of holonomic constraints | the constraints are geometric of the form
fi(u1 u2 : : : uN t) = 0 and do not depend on the velocities.
The total potential of the conservative system can be written in the form
U + V = * = *(u1 u2 : : : uN )
and its variation is given by

X
N @*
* = @u uj
j =1 j
Couple this with the virtual work of the nonconservative forces to get
XN @* X XN h @* X i
u j Q u
j j = 0 or Qj uj = 0
j =1 @uj j =1 @uj
; ;
j j
Since the ui can be varied arbitrarily, then the principle of stationary potential
energy in terms of generalized coordinates leads to
Fi =
@ h U + V i ; Q = 0 for i = 1 2 : : : N (6.9)
i
@ui
The expression, Fi = 0, is our statement of static equilibrium. It is important to
realize that the single equation * = 0 (for a conservative system) is actually a system
of N equations if there are N degrees of freedom in the system. Thus, in comparison
to the conservation of energy theorem, this is much richer, because instead of just
one equation, it leads to as many equations are there are degrees of freedom.
Example 6.5: Consider a long slender rod, xed at one end, with an applied
load at the other. Determine the equilibrium conditions for the long rod assuming
a linear elastic constitutive relation.
Let the unknown displacement elds be discretized as
u(xo ) = c1 xo v(xo ) = 0 w(xo ) = 0
where c1 is the single discrete unknown. This gives the strains
@u + 1 h( @u )2 + ( @v )2 + ( @w )2 i = c + 1 c2
E11 = @xo 2 @xo @xo @xo 1 2 1
with all other strains being zero. Assume the simple constitutive relation ijK = EEij
so that
K = EE = c + 1 c2 ]
11 11 1 2 1
with all other stresses being zero.
The virtual work of the applied load is
We = P uL = c1 L
The virtual strain energy is
Z Z
U = o
K E11 dV o =
11 o
E c1 + 12 c21 ]c1 + c1 c21 ]dV o
V V
= EV o c1 + 12 c21 ]1 + c1 ]c1 = EV o c1 + 23 c21 + 12 c31 ]c1
The principal of virtual work becomes
h i
We ; U = 0 ) PL ; EV o c1 + 32 c21 + 21 c31] c1 = 0
Since c1 is arbitrary then so also is c1 and we conclude that the bracketed term
must be zero. That is, h 3 i PL
1 + 2 c1 + 12 c21 c1 = EV o
This simple problem has lead to a nonlinear relation between the unknown defor-
mation (as represented by c1 ) and the applied loads. The origin of the nonlinearity
is in the strain-displacement relation.
The typical problem in solid mechanics is where the applied loads are specied
and the displacements are unknown, therefore we can anticipate that these problems
usually require some nonlinear solver routine. Most of them are iterative and a
popular one is discussed in Section 6.4 here we mention a scheme called Picard
iteration. The basic idea is to arrange the equation so that the nonlinearities are on
the left hand side and are estimated from a previous guess. A possibility is
ci1+1 = PL=EV o
1 + 23 c1 + 12 c21 ]i
where i is the iteration counter. The rst few iterations are
c01 = 0
c11 = PL=EV o
c21 = PL=EV o
1 + 32 (PL=EV o ) + 12 (PL=EV o )2 ]
This solution illustrates a characteristic of many nonlinear systems namely, that
they can exhibit instabilities. For the approximation shown, when PL=EV o = ;1
the deection is indicated as being innite.
III: Finite Element Formulation

Rather than use the result of Equation (6.9), we nd it more convenient to begin with
the equations of motion themselves, Equation (6.6), written at the current time t
Z
K Eij dV o = We
o ijV
. August 2007
where inertia eects are neglected. In anticipation of using matrix notation, introduce
Ekl ! f6 1g = fE g = fE11 E22 E33 2E12 2E23 2E13 gT
klK ! f6 1g = fK g = f11K 22K 33K 12K 23K 13K gT
so that the integrand of the internal virtual work can be written as
Z
K T
f g fE g dV
o
V o
There is nothing special about the sequence of components in the vectors, the above
sequence is one of the standard forms 4].
The variation of the Lagrangian strain is given by
Xh i
2Eij = @uoi + @uoj + k @uko @uok + @uok @uko
@xj @xi @xi @xj @xi @xj
The basic idea of the discretization is that, since the actual distribution of displace-
ments is quite complicated, it was approximated as a collection of piecewise simple
distributions over many regions and these distributions were characterized by the dis-
crete nodal values. That is, let the displacements in a small region of volume Vmo be
represented by
X
ui(xo1 xo2 xo3 ) = k hk (xo1 xo2 xo3)uik = h ]fuig or f u g = H ]f u g
where hk (xo1 xo2 xo3 ) are known shape functions and uik are the unknown nodal values.
All relevant quantities can now be written in terms of both of these. For example,
the derivatives and variations are given by
@ui = X @hk u = @h ]f u g @ui = X @hk u = @h ]fug
ik k @xoj ik
@xoj k @xoj @xoj
Hence the variation of strain can be written symbolically as
fE g = BE ]fug
where BE ] contains various spatial derivatives of hk as well as the nodal displacement
values f u g its content will be made explicit later.
The internal virtual work becomes
Z
T K
fE g f gdV
o
Vo Z
X T T f K g dV o g = X fugT fF g = fugT fF g
) fugm f BE ] m m m
m m Vo m
where fF g is the assemblage of all the element forces fF gm given by

X Z T K o
fF g =
m o BE ] f g dVm
Vm
(6.10)
The force vector fF g is interpreted as the set of nodal forces that are (virtual) work
equivalent to the actual distributed stresses and strains. The integral is done for each
element and the assemblage process forms the vector fF g of size fN 1g.
The virtual work of the applied body forces (without inertia) and surface tractions
leads to
XZ o+X
Z
o
f
i Vo i iu dV i Ao ti ui dA
Z Z
) fP g = f
o
H ] ff gdV g + f o H ]T f t gdAog
T o
V A
Assemblage is done as in the linear case, and these give the equations of motion as
fF g = fP g ; fF g = 0 (6.11)
This is our governing equation for a general nonlinear system discretized in the form
of nite elements. It is completely equivalent to Equation (6.9), but because we began
with the principle of virtual work we do not need to discuss the dierence between
conservative and nonconservative systems. That is, the present form of the equations
is applicable to situations involving plasticity as well as those with follower forces.
Note that the the load distribution is determined initially (although its history
will change) only the body stress term fF g needs to be updated.
Example 6.6: Establish the explicit form for the strain operator matrix BE ] for
the Hex20 discretization.
Typical terms for the variation of strains are obtained from fu x g according to
Exx = u x + u x u x +v x v x +w x w x ]
Eyy = v y + u y u y +v y v y +w y w y ]
2Exy = u y +v x +u x u y +u x u y +v x v y +v x v y +w x w y +w x w y ]
We can replace all functions using the interpolation functions and then express these
in matrix form as
8 9 8 9
>
> Exx >
> >
> u1 >
>
>
< Eyy >
= h i>
< v1 >
=
.. = B ] + B ] .. or fE g = BE ]fug
> . > L N > . >
>
> 2Eyz >
> >
> v20 >
>
: 2Exz : w20
The matrices are dened as
2 3
Ax 0 0
66 0 Ay 0 77
6 0 0 Az 777
BL ]i = 666 Ay Ax 0 77
64 0 Az Ay 5
Az 0 Ax i=1N
. August 2007
2 3
uxAx vxAx wxAx
66 uy Ay vy Ay wy Ay 77
6 uz Az vz Az wz Az 77
BN ]i = 666 uxAy + uy Ax vxAy + vy Ax wx Ay + wy Ax 777 (6.12)
64 uy Az + uz Ay vy Az + vz Ay wy Az + wz Ay 5
uxAz + uz Ax vx Az + vz Ax wxAz + wz Ax i=1N
where the matrices A ]i have already been established in Equation (6.4). The
strain operator matrix is
BE ] = 6 60] = BL1 + BN 1 BL2 + BN 2 BL120 + BN 120 ]
The nonlinear matrix BN ] contains the displacement gradients and therefore changes
during the deformation.
Example 6.7: Determine the equivalent nodal loads for a distributed loading.
The basic idea is to say that the virtual work of the equivalent nodal loads is
equal to the virtual work of the actual loading. Consider, for example, a body force
acting in the x-direction the virtual work is
XN Z hZ i
Pxiui = f b (xo yo z o )u dV o = XN fxb (r s t)hi (r s t) jJe jdV ui
i Vo x i V
We conclude that Z
Pxi = f b (r s t)hi (r s t) jJe jdV
V x
In a similar manner, the nodal forces due to surface tractions and line loads are,
respectively,
Z Z
Pxi = tx(r s t)hAi (r s t) jJA jdA Pxi = qx (r s t)hSi (r s t) jJS jdS
A S
The notations hAi (r s t) and hSi (r s t) mean that the functions are evaluated on the
appropriate surface and line, respectively. Also note that the jJe j to be used is the
one for areas or lines as appropriate.
In general, each of these integrals need to be evaluated numerically. It is useful
to keep in mind, however, that lower order interpolations can be used for loads
than is used for the element nodal forces and in this way the applied loads can
be determined separately from the continuum modeling. For example, if the linear
hexahedral interpolations are used then the nodal loads due to a line load are
uniform: P1 = 21 qo L P2 = 12 qo L linear: P1 = 13 qm L P2 = 32 qm L
with similar expressions for the tractions and body forces. (In both cases P3 = 0).
When the expectation is that small elements are needed then these representations
of the load are suciently accurate.
6.3. Stiness Properties of Discretized Systems 195
6.3 Sti ness Properties of Discretized Systems
Earlier in the chapter, we introduced the elastic modulus as a measure of the stiness
of a linear material this is a material property. When we deal with linear structures
or structured materials we introduce the concept of structural stiness which depends
on both geometry and material properties. When we deal with nonlinear problems, we
must further introduce the very important concept of the tangent stiness. Unlike the
elastic stiness of linear structures, the tangent stiness changes as the load changes
giving rise to some surprising consequences such as its eect on the stability of the
equilibrium.
General Stiness Relations

The assembled element nodal forces are computed from Equation (6.10) as
X Z T K o
fF g =
m o BE ] f g dVm
Vm
The variation of the nodal forces leads to
fF g =
@F ]fug = K ]fug
T
@u
where KT ] is the tangent (or total) stiness of the system. This is the matrix we
wish to establish in explicit form.
Substituting for fF g in terms of the stress leads to
X Z h T K i
T f K g dV o
fF g = B E ] + B E ] m
m o
Vm
f g
The stresses are a function of the strains so that the rst term becomes
BE ]T fK g = BE ]T @i ]fE g = BE ]T D ]BE ]fug
K
@Ej
The matrix D ] is the tangent modulus for the material.
The strain operator is considered a function of the displacement gradients so that
the second term becomes
BE ]T fK g = fu xgT @B E T K T T K
@u x ] f g = fug BD ] ]fBD g
The nal form will be demonstrated for the Hex20 element in the example problem
to follow.
We therefore have for the total stiness
X Z h T i
KT ] = m o BE ] D ]BE ] + BD ]T K ]fBD g dVmo = KE ] + KG]
Vm
. August 2007
The integral is over the element volume Vmo and the summation is associated with
the assemblage of the collection of elements. We see that the total stiness relation
is comprised of two parts: one is related to the tangent modulus properties of the
material, the other is related to the current value of stress. The rst matrix is often
called the elastic stiness because in the linear case it is primarily a function of
the elastic material properties. The second matrix is called the initial stress matrix
because in the linear case it depends on the stress and not on the material properties.
The combination of both matrices is sometimes referred as the tangent sti
ness. For
nonlinear problems, both matrices depend on the stress and/or current deformation
with the rst distinctly related to the material tangent modulus @ijK =@Epq and the
second distinctly related to the changing geometry of the element (through BE ). It
would seem appropriate to call the rst matrix the tangent stiness and the second
the geometric sti
ness we will refer to the combination as the total sti
ness. All
matrices are symmetric.
The stiness relation is quite general in that it is not restricted to any particular
constitutive relation. Thus, for example, for elastic-plastic materials it is necessary
only to replace
Dep] ;! D ]
The geometric stiness is unaected by the constitutive relation.
Both stinesses change during a general deformation. Indeed, both are compli-
cated functions of space and this requires an ecient scheme for the computation of
the volume integrals we discuss this next.
Example 6.8: Establish the geometric stiness matrix for the Hex20 element.
From Equation (6.12), only the BN ] part of BE ] is a function of the displace-
ment gradients. Consequently,
2 3
uxAx vx Ax wx Ax
66 uy Ay vy Ay wy Ay 77
66 77
BE ]i = 66 u A + u A v A + v A w Aw+
u z A z v z Az z Az 77
64 ux Ay + uy Ax vx Ay + vy Ax wx Ay + w
w
y Ax 75
y z z y y z z y y z z Ay
uxAz + uz Ax vx Az + vz Ax wx Az + wz Ax i=1N
Each column can be expanded to the form
2 3
Ax 0 0
66 0 Ay 0 77 8 9
66 0 0 Az 777 < x =
66 Ay Ax 0 77 : y
64 0 Az Ay 5 z
Az 0 A x
Note that the matrix is actually BL ]i . The product f K gT fBE g has the compo-
nent products
2 3
Ax 0 0
66 0 Ay 0 77
K K K K K K g 6
6 0 0 Az 777
f xx yy zz xy yz xz 6 66 Ay Ax 0 77
4 0 Az Ay 5
Az 0 Ax
This can be re-arranged to
2 Ax 0 0 3
2 K K K
xx 0 0 36 Ay 0 0 77
xy xz 6
66 K K K
yx 0 0 77 66 Az 0 0 77
66 yy yz
K K K
zx 0 0 77 66 0 Ax 0 77
66 0
zy zz
0 K K
0 xx 77 66 0 Ay 0 77 = K ]BD ]i
66 xy 77 66 0 Az 0 77
4 0 0 K K
0 xy yy 5 66 0 0 Ax 777
.. .. .. .. .. . . . 64
. . . . . 0 0 Ay 5
0 0 Az i=1N
Since fu x g = BD ]fug then
BE ]T f K g = fu x gT @BE T K T T K

@u ] f g = fug BD ] ]fBD g
x
from which it follows that
X Z h i
KG ] = m Vmo BD ]T K ]fBD g dVmo
This matrix plays a signicant role in the buckling instability of structures.
Numerical Quadrature
The foregoing stiness relations (and some of the earlier results) show that we have a
need to evaluate integrals over lengths, areas and volumes. This type of integration
(where the integrand is specied) is called quadrature. Since the integration occurs
repeatedly in a nonlinear problem, it is essential that it be performed eciently and
accurately. Clearly, it must be done numerically for general elements.
An advantage of the isoparametric formulation is that general volume integrals,
for example, get reduced to integrals on a 2 2 2] cube irrespective of the physical
size of the block. That is,
Z Z Z Z
Q( xo y o z o ) dV o = +1 +1 +1 Q(r s t) det jJ (r s t)j dr ds dt
o e
V ; 1 ;1 ; 1
where the physical dimensions enter through the Jacobian. This signicantly facili-
tates standardizing the numerical process.
. August 2007
I: Lagrange Interpolation
The basic idea is to replace the integral with a summation. Thus, for example, the
line and volume integrations become, respectively,
Z X Z X
F (r) dr = iiF (ri) F (r s t) drdsdt = ijkijk F (ri sj tk )
The key is the proper choice of ijk at the appropriate locations (ri sj tk ) so as to
maximize the accuracy.
To motivate the procedure, consider the 1-D function F (r) evaluated at the (n +1)
distinct points r0 r1 . Represent this with the polynomial
F (r) aa + a1 r + a2r2 +

and set up the (n + 1) simultaneous equations to evaluate the ai coecients. The

integral is then given by
Z +1 +1
I= F (r) dr a0 r + a1 r2 + a2 r3 + 1
; 1 ;
Needless to say that if F (r) is a polynomial of degree equal to or less than the
representation, then the integral is exact. This approach, however, is a relatively
costly procedure.
An ecient way to produce a polynomial representation is through Lagrangian
interpolation. That is, we use the representation
F (r) F0 L0 (r) + F1 L1 (r) + F2 L2 (r) +

where
Lj (r) = (r(r r r)(0)(r r rr1)) rr rrj ]] ((rr rnr 1 )()(r r rnr) )
; ; ; ; ; ;
Lj (ri ) = ij
j ; 0 j ; 1 j j
; j ; n 1 j; ; n
In the functions Lj (r) shown, the center terms in square brackets are missing and
are only shown to help in establishing the permutations for evaluating Lj (r). For
example, if we have a quadratic polynomial, n = 2, then
L0 (r) = ((rr rr1)(

; r r2 )
;
L1(r) = ((rr rr0 )(
; )(r r2)
;
L2 (r) = ((rr rr0)(
; r r1)
;
0 ;1 )(r0 r2 )
; 1; 0 r1 r2 )
; 2
; 0 )(r2 r1 )
;
It is clear that when r = r0 r1 r2 we get F (r) = F0 F1 F2 , respectively. The

accomplishment of Lagrangian interpolation is that it avoids the necessity of having
to solve the simultaneous equations indicated above.
n ri Wi
1 +0:0 2: 0 p
2 0:57735 02691 89626 = 1= 3 1:0
; ;
p
+0:57735 02691 89626 = +1= 3 1:0 p
3 0:77459 66692 41483 = 0:6 0:55555 55555 55555 = 5=9
; ; ;
+0:0 0:88888 88888 88888 = +8=9

p
+0:77459 66692 41483 = + 0:6 0:55555 55555 55555 = +5=9
4 0:86113 63115 94053
; 0:34785 48451 37454
; 0:33998 10435 84856 0:65214 51548 62546
+0:33998 10435 84856 0:65214 51548 62546
+0:86113 63115 94053 0:34785 48451 37454
Table 6.2: Sampling points and weights for Gauss-Lagrange Quadrature
II: Gaussian Integration Points

There is one further renement that can be implemented. That is, rather than choose
equi-spaced points, we preselect the discretized points so as to minimize the error for
a given level of polynomial. These preselected locations are shown in Table 6.2.
That is, the integration variable r is chosen so that the integration limits are from
r = ;1 to r = +1 and the integral is
Z +1
I= F (r) dr W0 F (r0) + W1 F (r1) + W2F (r2) +
; 1
where the weights Wi are also shown in Table 6.2. The equispaced approximation
integrates a polynomial of order n exactly, but the Gauss-Lagrange unequal spaced
approximation integrates exactly a polynomial of degree (2n ; 1) 4]. A further point
is that the approximation for n and n + 1 have the same accuracy and so the even
formulas (n = 0 2 4 :::) are typically used.
These formulas are directly extended to multiple dimensions by
Z +1 X
F (r) dr = iWiF (ri)
Z +1 Z +1 1 ;
XX
F (r s) drds = i j Wi Wj F (ri sj )
Z +1 Z +1 Z 1+1 1
; ;
XXX
F (r s t) drdsdt = i j k WiWj Wk F (ri sj tk )
;1 ;1 ; 1
That the same weighting factors and coordinates are used irrespective of the di-
mension of the problem makes this integration scheme very easy to implement in a
computer code.
. August 2007
Example 6.9: Use numerical quadrature to evaluate the integral
Z 3 dx
I=
1 x
We will use the order n = 2 (that is, three-point) rule. First introduce r = x ; 2
so that the integral becomes Z +1 dr
I= r+2
1
;
The approximation is
I W0 F (r0 ) + W1 F (r1 ) + W2F (r2 )
95 1 ; 1p0:6 + 89 1 ;1 0 + 59 1 + 1p0:6 1:09803
The exact result is I = ln(3) = 1:09861. This is surprisingly accurate for a three
term polynomial approximation.
6.4 Total Lagrangian Incremental Formulation

All nonlinear problems are solved in an incremental/iterative manner with some sort
of linearization done at each time or load step. In this section, we develop the total
Lagrangian incremental formulation. The incremental idea is shown in Figure 6.11
where everything is assumed known at time step t and it is desired to obtain the
solution at time t + t. The fundamental equation we deal with is the static version
of the equation of motion, Equation (6.11), written at the next time
t+t ; fF gt+t X Z T K
fF g = 0 = fP g fF g =
m V o BE ] f g
m
Since fF gt+t is unknown the developed scheme is implicit.

Vt u -
*

V t+t

;
;
ut;
ut+t
Vo ;
;

;

6
-

;
;
Figure 6.11: Decomposition of displacement.
This scheme seems quite suitable for the 3-D continuum elements where the de-
grees of freedom are the nodal displacements. There are many other formulations,
indeed for 3-D thin-walled structures, a corotational scheme is quite popular.
6.4. Total Lagrangian Incremental Formulation 201
Simple Truss Example
A truss is composed of slender members that support only axial load consequently,
these members must be triangulated for equilibrium under normal loads. We use
the truss as an introductory example to illustrate the eect of axial loads on the
stiness properties of a structure. This will also serve as a test case for our computer
formulation. Some geometric approximations are made but these will not be part of
our nal procedure for general continuous solids.
Consider the simple truss whose geometry is shown in Figure 6.12. The members
are of original length Lo and the unloaded condition has the apex at a height of h.
The two ends are on pinned rollers. For simplicity, assume the truss is elastic with
the axial force related to the axial strain by
Fo = EAo xx
More general constitutive relations is left to later.
y P-1x * Fo
6P2y
v v-
r
6
6P2
-x
6 Fo

?Ks v2
H H t
ddd rd ddd

L !
o ! HH
a H H L 6h + v2

a HH
HH u
P-1 ! 2Ks
!
a HHH P3 = P1
H -1 t

...........
..........
........... ...........
.......... ...........
..........
...........
..........
........... ...........
..........
........... ..........
........... 1l 2l
Figure 6.12: Simple pinned truss with a grounded spring.

Let the height h be small relative to the member length, and let the de"ections
be somewhat small then we have the geometric approximations
Lx = L cos Lo ; u1 Ly = L sin h + v2 u1
v2
The deformed length of the member is
q
L = (Lo ; u1)2 + (h + v2 )2 Lo ; u1 + Lh v2 + 12 Lo ( Lv2 )2
o o
The axial force is computed from the strain as
Fo = EAo = EAo L ; Lo = EA h ; u1 + h v2 + 1 ( v2 )2i
o
Lo Lo Lo Lo 2 Lo
Note that we consider the parameters of the constitutive relation to be unchanged
during the deformation.
Look at equilibrium in the deformed conguration specically, consider the re-
sultant horizontal force at Node 1 and vertical force at Node 2 giving
0 = P1x + Fo cos P1x + Fo
0 = P2y ; Fo sin ; Ksv2 P2y ; Fo ; Ksv2 h+ Lo
v2
. August 2007
Rewrite these in vector form as
0 P ;F 0
1x o
0 = P2y ; Fo ; Ksv2 or fF g = fP g ; fF g = f 0 g
We refer to the last form of the equation as the loading equation fP g is the vector
of applied loads, fF g is the vector of element nodal forces, and fF g is the vector of
out-of-balance forces. For equilibrium, we must have that fF g = f 0 g, but, as we will
see, this is not necessarily (numerically) true during an incremental approximation of
the solution.
Example 6.10: Determine the deections when the loads are P1x = P , P2y = 0.

1.50
1.25
1.00
Load [P/Pcr]
.75 decreasing h
decreasing h
.50
Simple stepping
.25 N-R iterations
(a) (b)
.00
0. 2. 4. 6. 8. 10. -200. -100. 0. 100. 200.
Deflection [u/uc] Deflection [(h+v)/uc]
Figure 6.13: Load-de"ection behavior for the simple truss. (a) Horizontal displace-
ment u1. (b) Vertical displacement v2.
For this special case, we get Fo = ;P and the two deections are
P
u1 = EA + ( Lh )2 K LP ; P + 21 ( Lh )2 ( K LP ; P )2 Lo
h o s o o s o
v2 = L K L ; P Lo P
o s o
The load-deection relations are nonlinear even though the deections are assumed
to be somewhat small. Furthermore, when the applied load is close to Ks Lo , we
get very large deections. (This is inconsistent with our above stipulation that the
deections are \somewhat small," let us ignore that issue for now and accept the
results as indicated.) That is, at P = Pcr = KsLo , we get very large deections
meaning that the structure has become unstable. We say P has reached a critical
value.
The full solutions are shown plotted in Figure 6.13 for dierent values of h. The
eect of a decreasing h is to cause the transition to be more abrupt. Also shown are
the behaviors for P > Ks Lo . These solutions could not be reached using monotonic
loading, but they do in fact represent equilibrium states that can cause diculties
for a numerical scheme that seeks the equilibrium path approximately. That is, it
is possible to accidentally (begin to) converge on these spurious equilibrium states.
Incremental Solution Scheme

We formulate the solution in an incremental fashion. That is, we view the deformation
as occurring in a sequence of steps associated with time increments t, and at each
step it is the increment of displacements that are considered to be the unknowns.
To help x ideas, look again at the truss in Figure 6.12. We have already shown
that the equilibrium equation is
0 P ;F 0
1x o
0 = P2y ; Fo ; Ksv2 or fF g = fP g ; fF g = f 0 g (6.13)
with = (h + v2)=Lo and the axial force is the nonlinear function of the displacements
h i
Fo = EA ; Lu1 + Lh Lv2 + 12 ( Lv2 )2
o o o o
Consider the equilibrium equation at time step tn+1
fF g n+1 = fP gn+1 ; fF (u)gn+1 = f 0 g
We do not know the displacements f u gn+1, hence we cannot compute the axial force
Fo nor the nodal forces fF gn+1. As is usual in such nonlinear problems, we linearize
about a known state. That is, assume we know everything at time step tn, then write
the Taylor series approximation (for small displacement increment) for the element
nodal forces
fF (u)gn+1 fF (u)gn +
@F ] fug + = fF (u)g + K ] fug + (6.14)
@u n n T n
The square matrix KT ] is sometimes called the tangent sti

ness matrix but as dis-
cussed in the previous section we will call it the total stiness. The explicit form it
takes for our truss problem is
2 @F1x @F1x 3 2 @ Fo @ Fo 3
6 @u1 @v2 77 66 ; @u1 77
KT ]n = @F ]n = 64 @F @v2
;
2y @F2y 5
= 6
4 @ Fo 1 @ o
F 75
@u @u L Fo + @v + Ks
@u1 @v2 n 1 o 2 n
Performing the dierentiations
@ Fo = EA ; 1 @ Fo = EA h + v2 = EA
@u1 Lo @v2 L2o L2o Lo
. August 2007
then leads to the stiness
2 3
KT ] = @F ] = EA 4 1
K L 5
;
+ Fo 0 0 = K ] + K ]
s o E G
@u Lo 2 + EA
; Lo 0 1
Note that both matrices are symmetric. The total stiness has the same decompo-
sition as discussed earlier. The rst matrix is the elastic stiness of a truss member
oriented slightly o the horizontal by the angle = (h + v2 )=Lo. The second matrix
shows the eect of the axial loading.
We are now in a position to solve for the increments of displacement re-arrange
the approximate equilibrium equation into a loading equation as
fP n+1 ; fF gn ; KT ]n fug 0
g =) KT ]nfug = fP gn+1 ; fF gn
Again, consider the special case when P1x = P , P2y = 0 then the system of equations
to be solved is
" #
EA 1 ; + Fo 0 0 u1 = P ; ;Fo
Lo ; 2 + Lo 0 1 n v2 0 n+1 ; Fo + Ksv2 n
with = KsLo=EA. Note that the right-hand-side has a load term associate with the
current known state. A simple solution scheme involves computing the increments at
each step and updating the displacements as
u1(n+1) = u1(n) + u1 v2(n+1) = v2(n) + v2
The axial force and orientation also need to be updated as
h i
Fo jn+1 = EA ; Lu1 + Lh Lv2 + 12 ( Lv2 )2 n+1 n+1 = h + v2 j
Lo n+1
o o o o
Table 6.3 and Figure 6.13 show the results using this simple stepping scheme, where
Pcr = KsLo and ucr = Pcr =EA.
Table 6.3 also shows the out-of-balance force
fF g n+1 = fP gn+1 ; fF gn
computed at the end of each step. Clearly, nodal equilibrium is not being satised and
it deteriorates as the load increases. In order for this simple scheme to give reasonable
results, it is necessary that the increments be small. This can be computationally
prohibitive for large systems because, at each step, the total stiness must be formed
and decomposed. A more-rened incremental version that uses an iterative scheme
to enforce nodal equilibrium will now be developed.
P=Pcr u1=ucr (h + v2 )=ucr P F 1x F2y
.0500 .051000 10.5000 100.00 .050003 .989497E-02
.1500 .153401 11.6315 300.00 .256073 .446776E-01
.2500 .257022 13.1329 500.00 .450836 .589465E-01
.3500 .362372 15.0838 700.00 .761169 .759158E-01
.4500 .470656 17.7037 900.00 1.37274 .100333
.5500 .584416 21.3961 1100.0 2.72668 .137041
.6500 .709589 26.9600 1300.0 6.19165 .192208
.7500 .862467 36.1926 1500.0 17.0481 .257334
.8500 1.09898 53.8742 1700.0 62.5275 .938249E-01
.9500 1.66455 94.1817 1900.0 324.939 -4.00430
1.050 3.23104 163.411 2100.0 958.535 -24.0588
Table 6.3: Incremental results using simple stepping.

6
incremental
Load

true

Deection
-
Figure 6.14: Illustration of the simple incremental scheme and its deviation from
the true equilibrium path. The increments are load controlled.
Newton-Raphson Equilibrium Iterations
As was just pointed out, if the estimates for the new displacements during an incre-
mental solution are substituted into Equation (6.13), then this equation will not be
satised because the displacements were obtained using only an approximation of the
nodal forces given by Equation (6.14). What we can do, however, is repeat the above
process at the same applied load level until we get convergence. We illustrate this
here.
We begin by writing the loading equation at time step tn+1 as
i
fF gn+1 = fP gn+1 ; fF gn+1
where i is an iteration counter. The current displacement is estimated according to

i i 1
f u gn+1 f u gn+1 + fug
i ;
Note that the increment is from the previous estimate of the current displacement
f u gn+1 and not f u gn as done previously. The Taylor series approximation for the
. August 2007
i u1 v2 u1 u1ex
; v2 v2ex
; F 1x F 2y
1 -3.02304 76.0009 -3.12184 72.2009 .3207E+7 -.244E+7
2 29.0500 75.9986 28.9512 72.1986 .21923 -72.3657
3 -25.8441 3.95793 -25.9429 .157933 .2594E+7 -107896.
4 .105242 3.95793 .64416E-2 .157926 .6835E-2 -.158213
5 .09867 3.80001 -.12436E-3 .8344E-5 12.4694 -.498780
6 .09880 3.80000 .15646E-6 .4053E-5 .000000 -.3948E-5
7 .09880 3.80000 .000000 .000000 -.4882E-3 .1878E-4
exact .09880 3.80000
Table 6.4: Newton-Raphson iterations for a load step 0:95 Pcr .
element nodal forces is

F (u) i F (u) i 1 + @F ]i 1 fugi +
; ;
= fF (u)gin 1 + KT ]in 1fugi + ; ;
f g n+1 f g n @u n
(6.15)
The new estimates for the displacements are obtained by solving
i 1 i
fKT gn fug = fP gn+1 ; fF gn
; i 1 i i 1 + fugi
f u gn+1 f u gn
; ;
This is repeated until fugi are suciently small.

The algorithm during each time step increment may thus be stated as
form: i 1
fKT gn+1
; i 1
fF gn+1
;
solve: i 1 ; i i 1
fKT gn+1 fug = fP gn+1 ; fF gn+1
;
update: i i 1
f u gn+1 = f u gn+1 + fug
i;
update: i
fKT gn+1
i
fF gn+1
and repeat until fugi becomes less than some tolerance value. The iteration process
is started (at each increment) using the starter values
fu 0n+1 = u n
g f g KT 0n+1 = KT n
f g f F 0n+1 = F n
g f g f g
This basic algorithm is known as the full Newton-Raphson method.

Combined incremental and iterative results are given in Figure 6.13. We see that it
essentially gives the exact solution. Iteration results for a load level equal 0:95 Pcr are
given in Table 6.4 where the initial guesses correspond to the linear elastic solution.
We see that convergence is quite rapid and the out-of-balance forces go to zero.
It is worth pointing out the converged value above Pcr in Figure 6.13 this corre-
sponds to a vertical de"ection where the truss has \"ipped" over to the negative side.
Such a situation would not occur physically, but does occur here due to a combination
of linearizing the problem (i.e., the small angle approximation) and the nature of the
iteration process (i.e., no restriction is placed on the size of the iterative increments).
General Nonlinear Algorithm
The previous development are applicable to general continua, it is necessary only to
utilize the appropriate discretizations. As shown in the previous section, the total
stiness has the representation
X Z h T i
KT ] = m o BE ] D ]BE ] + BD ]T K ]fBD g dVmo = KE ] + KG]
Vm
The corresponding element nodal forces are compted as
X Z T K
fF g =
m o BE ] f g
Vm
Otherwise, the scheme is identical to that just described for the simple truss.
In the following, we concentrate on the basic algorithm for the full Newton-
Raphson method because it best illustrates the essential ingredients. This algorithm,
for monotonically increasing loads, can be stated as:
Step 1: Specify parameters of the algorithm such as tolerances, and maximum
iterations.
Step 2: Read the initial geometry and material properties.
Step 3: Specify load increments, number of steps.
Step 4: Begin loop over time (load) increments:
Step T.1: Increment the load vector fP gt+t .
Step T.2: Initialize for equilibrium iterations
f u g0t+t = f u gt KT ]0t+t = KT ]t fF g0t+t = fF gt
Step T.3: Begin loop over equilibrium iterations:
Step I.1: ITERATE:
Step I.2: Assemble nodal force vector fF gi.
Step I.3: Form the eective load vector
fPeff git+t fP gt+t ; fF git+t
Step I.4: Assemble the elastic stiness matrix KE ].
Step I.5: Assemble the geometric stiness matrix KG ].
Step I.6: Form the total stiness matrix as
KT ] = KE ] + KG ]
Step I.7: Decompose the total stiness to
KT ] = U ]T d D c U ]
. August 2007
Step I.8: Solve for the new displacement increments from
U ]T d D c U ]fugi = fPeff git+t
Step I.9: Update the displacements
f u git+t = f u git+1 t + fugi
;
Step I.10: Test for convergence.

if: jfugi j=jf u gi j < tol converged, goto UPDATE
if: jfugi j=jf u gi j > tol not converged, goto ITERATE
Step T.4: End loop over iterations.
Step T.5: UPDATE:
ut+t = uit+t xyzt+t = xyzti+t
Step T.6: Store results for this time step.
Step T.7: If maximum load not exceeded continue looping over loads.
Step 5: End loop over time (load) increments.
Step 6: END
It is possible to enhance this algorithm by including automatic step changes, au-
tomatic testing for appropriate time step size, and monitoring the spectral properties
of the total stiness. The parameters and can also be adjusted automatically.
Load "" Load

6 "
" 6

i ;!
i ;!
full modied

Deection
-
Deection
-
Figure 6.15: Full and modied Newton-Raphson methods.

The full Newton-Raphson method has the disadvantage that, during each itera-
tion, the tangent stiness matrix must be formed and decomposed. The cost of this
can be quite prohibitive for large systems. Thus, eectively, the computational cost
is like that of the incremental solution with many steps. It must be realized, however,
that because of the quadratic convergence, six Newton-Raphson iterations, say, are
much more eective than six load sub-increments.
The modied Newton-Raphson method is basically as above except that the total
stiness is not updated during the iterations but only after each load increment.
This generally requires more iterations and sometimes is less stable but it is less
computationally costly when it succeeds.
Both schemes are illustrated in Figure 6.15 where the starting point is from the
zero load state. It is clear why the modied method will take more iterations. The
plot for the modied method has the surprising implication that we do not need
to know the actual total stiness in order to compute correct results | what must
be realized in the incremental/iterative scheme is that we are imposing equilibrium
(iteratively) in terms of the applied loads and resultant nodal forces we need good
quality element stiness matrices in order to get the good quality element nodal
forces, but the assembled total stiness matrix is used only to suggest a direction for
the iterative increments. To get the correct converged results we need to have good
element stiness relations, but not necessarily a good assembled total stiness matrix.
Clearly, however, a good quality tangent stiness will give more rapid convergence as
well as increase the radius of convergence.
Example 6.11: Use Hex20 elements to model the large deection, large strain
of the block shown in Figure 6.16. Compare the results to those obtained using the
Tet4 tetrahedron elements.
(a) L = 127 mm (5:0 in)

b = 25:4 mm (1:0 in)
h = 25:4 mm (1:04 in) (b)
aluminum 10 ;
Figure 6.16: Large deformation of a thick beam. (a) Initial and deformed shapes
using Hex20 elements. (b) Initial and deformed shapes using Tet4 elements.
One end of the block is xed while the end has vertical loads which do not change
their direction during the loading history. The z axis is along the length.
In this type of problem, the initial linear displacement is vertical only and con-
sequently the axial straining is exaggerated. This in turn leads to very large unbal-
anced forces which require the Newton-Raphson iterations to bring them back to
zero. It is possible, however, that the initial forces are so large (because they are
associated with axial stretching) that convergence is impossible to achieve without
. August 2007
using a small time step.
1.0
iter=50
v=h v=h
.5
u=h
.
.0
u=h
w=h
-.5 (a) Load
w=h (b) Load
0. 5. 10. 15. 20. 25. 0. 5. 10. 15. 20. 25.
Figure 6.17: Large deformation of a thick beam. (a) Displacements against load us-
ing Hex20 elements. (b) Displacements against load showing the anisotropic behavior
of the tetrahedron arrangement.
While it is possible to devise sophisticated step sizing schemes to help cope with
these situations, a simple but eective scheme is described here. First, the modied
Newton-Raphson iteration scheme (where the stinesses are updated only at the
beginning of a load step) is used because these assure the displacement increments
will be estimated based on a converged total stiness. Second, the full increment of
load is applied but the displacements are updated only partially by
f u gi = f u gi + ifugi
where i
i = 2i2max o i < imax i = o i > imax
Typically, imax = 5 which gives increments of 161 , 18 , 14 , 12 , 11 , Clearly, this increases
the number of iterations, but the computational cost is not exorbitant since only the
element nodal forces need be updated at each iteration. The signicant advantage
is that it gives the out-of-balance forces time to adjust before the full increment of
load is eectively applied.
Figure 6.17(b) shows the displacements at all four end nodes. It is expected that
the u displacements be nonzero because of the Poisson's ratio eect but that all four
be centered about zero. It is clear from the gure that there is a denite sideways
drift for the Tet4 results indicating that the tetrahedron arrangement exhibits some
anisotropy.
Interestingly for the Tet4 element, the initial iterations reach the maximum limit
set at 50 but after about six or so increments of load decrease signicantly. Actually,
the out-of-balance forces during the initial increments are reasonably well converged
but seem to oscillate about small deviations.
Example 6.12: Treat the simple shear deformation of a block as a load control
problem and show the relation between the Cauchy and Kirchho stresses.
2y t
6y t
6
kxo
-2 -x
xo2

1-x
Figure 6.18: A block in simple shear.
A simple shear deformation parallel to the xo1 ; xo2 plane is shown in Figure 6.18
and given mathematically by
x1 = xo1 + k xo2 x2 = xo2 x3 = xo3
u1 = k xo2 u2 = 0 u3 = 0
indicating that horizontal lines move horizontally only. The deformation gradients
are 21 k 03 2 1 ;k 0 3
@xo
@x p 4
@xoi ] = 00 10 01
5 and p ] = 4 0 1 0 5
@xi 0 0 1
Note that there is no volume change because J = Jo = 1: The Lagrangian strain
tensor is 20 k 03
X @xp @xp
2Eij = p @xo @xo ; ij = 4 k k2 0 5
i j 0 0 0
There is no straining of horizontal lines. There is also no straining in the third
direction indicating that this is a state of plane strain.
Let the material have the following linear constitutive behavior:
X
ijK = 2
Eij + ij k Ekk
where
and are the Lam%e constants. The Kirchho stress tensor, therefore, is
21 2 3 2 k
2 k
k 0 7 1 0 3
ijK = 64
k
k2 + 12 k2 0 5 =
k 4 1 (1 + )k 0 5
0 0 1 k2 0 0 k
2
where = =2
. The tensile 22 K component arises from the fact that lines originally
in the 2-direction are being stretched. The Cauchy stresses are obtained from
X K @xp @xq
pq = ij o ij @xoi @xoj
@x @x @x @x @x @x @x @x @x @x
K p
= 11 @xo @xo + 12 @xo @xo + @xo @xo + 22 @xo @xo + 33 @xpo @xoq
q K p q p q K p q K
1 1 1 2 2 1 2 2 3 3
. August 2007
Substituting for the deformation gradients leads to the complete stress tensor as
2 (2 + )k + (1 + )k3 1 + (1 + )k2 0 3
pq =
k 4 1 + (1 + )k2 (1 + )k 0 5
0 0
The Cauchy stress tensor, as expected, is symmetric.
The magnitude of the shear deformation is governed by the parameter k. It is
worth noting that when it is small, both stress tensors approach the same values.
K
yy
5. Stress 15. Stress
4. xx
K
xy 10.
Infinite sheet
3. FEM
xy
.1
2.
1.
K
xx 5. yy
0.
(a) Load 0.
(b) Load
0. 10. 20. 30. 40. 0. 10. 20. 30. 40.
Figure 6.19: Stresses for a block under shear load control. (a) Kirchho stress.
(b) Cauchy stress.
Imagine a free body cut parallel to the x-axis this will expose two tractions
related to the Cauchy stress by
tx = xy ty = yy
The tx traction, when multiplied by the area, gives a resultant horizontal force that
we will consider to be the applied load. The resulting deformation is then related
to the traction (and hence load) as
xy =
k1 + (1 + )k2 ] = tx = P=hL
where hL is the area over which the resulting force P acts. Let us consider the
problem to be load driven: then the deformation parameter k is a nonlinear function
of the load. However, we can easily solve for it using a Newton-Raphson iterative
scheme as
ki+1 = ki +
1 +P=hL ; fo fo =
ki 1 + (1 + )(ki )2 ]
(1 + )3(ki )2 ]
where i is the iteration counter. This converges very rapidly.
Once we know k we can then determine the Kirchho stresses. The results are
shown plotted in Figure 6.19 as the continuous lines. It is interesting to note that
the Cauchy xx is the largest of the stresses.
Example 6.13: Use the CST element to obtain a numerical solution of the shear
problem.
6.5. Stability of Discrete Systems 213
;! ;! ;! ;! ;!
L = 203 mm (8:0 in)
b = 101 mm (4:0 in)
h = 12:7 mm (0:5 in)
aluminum
Figure 6.20: Undeformed and deformed shape of a block under shear.

The analytical solution just developed was for a very large sheet under homoge-
neous deformation. This is impractical to achieve here so we will model the block
as shown in Figure 6.20.
The top row of elements have a stiness 1000 times that of the other elements,
it is also constrained to move only horizontally. In the innite sheet case the lateral
sides have shear components but clearly in the present case the normal tractions are
zero.
Figure 6.21 shows the contours of Cauchy stress at the maximum load drawn
on the deformed block. What they all have in common is that they show a nearly
uniform region of stress in the middle portion. We therefore expect to have a rea-
sonable comparison with the innite sheet solution in this region. Figure 6.19 shows
a comparison of the stress histories with that for the innite sheet | all the trends
are in agreement.
xx yy xy
Figure 6.21: Contours of Cauchy stresses on the deformed block.

We do not expect the Cauchy xx stress to go to zero at the boundaries because
these boundaries are inclined in the deformed conguration
6.5 Stability of Discrete Systems

Up to now we have concentrated on establishing the equilibrium of deformable bodies.
This section discusses a very important aspect of equilibrium namely, its stability.
This is a quintessentially nonlinear phenomenon and therefore could not be part of the
equilibrium analyses of Chapter 5. Many of the ideas to be presented are developed
in more depth in the paper by Allman 2] and the book by Criseld 7].
Some Notions of Stability

There are a number of denitions of stability and the one chosen should be appro-
priate for the phenomena being investigated. Our basic notion is that if a system is
. August 2007
slightly disturbed from its equilibrium position and eventually returns to the original
position after removal of the disturbance, then it is stable. Conversely, if it gives
a disproportionate response, then it is unstable. This basic notion is shown in Fig-
ure 6.22. There are other concepts of stability such as the original as proposed by
Euler, which was phrased in terms of alternate equilibrium positions not \too far"
from the conguration under discussion. We look at both of these concepts.
Po (t) u A uo

Q(t) u_
Tim
e
Tim
e
Figure 6.22: Dynamic view of a static instability. (a) Small ping loads applied in
addition to the slowly increasing primary load. (b) Response to the two loads, the
second ping causes a change in primary loading path indicating in instability.
I: Potential Energy View of Stability

Consider the equilibrium situations shown in Figure 6.23 where the black dot repre-
sents a ball and the line represents a surface it is resting on. In each case, the black
ball is in equilibrium as can be easily veried by drawing a free body diagram. Now
consider displacing the ball a very small amount away from the equilibrium position
| this is indicated by the white ball. What will happen?
In Figure 6.23(a), the ball will roll back toward the original position if there is
dissipation in the system, it will oscillate and eventually settle down to the original
position. We therefore say that the original (and not the disturbed) equilibrium
is stable. In Figure 6.23(c), the ball will roll away from the original position we
therefore say that the original equilibrium is unstable. The ball in Figure 6.23(b) will
stay put and therefore we say that the original equilibrium is neutral.
In simple terms, we can think of the position of the ball as its total potential
energy. The change in this potential energy as the ball is displaced can be expressed
as
1 2 X @* 1 X X @2*
* * + 2 * + = i ui + 2 i j
@ui @ui@uj uiuj +
x h xh xh
(a) (b) (c)
@* = 0 @* = 0 @* = 0
@u @u @u
@2* > 0 @2* = 0 @2* < 0
@u2 @u2 @u2
Figure 6.23: Disturbed equilibrium state of a ball. (a) Stable equilibrium. (b) Neu-
tral equilibrium. (c) Unstable equilibrium.
We have already shown in Section 6.2 that the condition * = 0 governs equilibrium
a study of the second order term 2 * therefore governs the nature of the equilibrium,
that is,
2* > 0 :stable equilibrium
2
*=0 :neutral equilibrium
2* < 0 :unstable equilibrium
Actually, if 2 * = 0 then we must check the next higher order terms also. We conclude
that for stable equilibrium, the potential energy is a relative minimum as depicted in
Figure 6.23(a).
To clarify the meaning of the second variation, let the equation of the surface be
quadratic near the equilibrium point that is,
y = x2 or v = u2
where is a parameter. The three cases shown in Figure 6.23 correspond to > 0,
= 0, and < 0, respectively. The total potential when a small horizontal force Q
is applied is
* = U + Mgv ; Qu = U + Mgu2 ; Qu
(In this case, the strain energy U is zero.) The dierent orders of variation of the
potential are
* = @@u* u = @@uU + 2Mgu ; Q] u = 2Mgu ; Q] u
2 2
2* = @@u*2 u2 = @@uU2 + 2Mg] u2 = Mg] u2
@ 3* @ 3U
* = @u3 u = @u3 + 0] u3 = 0 ] u3
3 3
The rst equation gives the equilibrium condition when the system is in equilibrium
this term is zero, therefore it is the second equation that determines the sign of *.
. August 2007
Depending on the value of , this second term can be either positive or negative. One
of the important points to note is that the relation for the second variation does not
contain the applied load Q but does contain the gravity load Mg.
II: Instability as Loss of Structural Stiness

To clarify our stability idea in an actual situation, consider the simple pinned structure
shown in Figure 6.24(a). The initially vertical bar is assisted in remaining vertical by
the action of the horizontal spring the spring is unstretched when the bar is vertical.
An equilibrium analysis in the undeformed state shows that the displacements are
related to the forces through the stiness relations
Ku = Q Kb v = P; (6.16)
where Kb is the axial stiness of the bar. For the purpose of the later discussion, let
the bar be very sti, from which we conclude that the displacement is only horizontal,
that is, v = 0.
.......
......
.......
dd
P = const r
Q- (t) ?rf ....
...
....
!
a K2
...r
... !
7;! u LLLKL ........ a
!
a
! r K1 d.... ...
- f ....
LLLL r d....
...
Q(t) ...
....

? P = const

L y

y
L
rg rg
6 6
-x -x
.......
......

.......
...... (a) .......
......

....... (b)
Figure 6.24: Disturbed equilibrium state of a pinned bar. (a) Small de"ection
model. (b) Large de"ection model.
A key point in stability analysis is to look at equilibrium of the structure in its
(slightly) deformed state. In this rst linear analysis, we will consider all displace-
ments to be small. Consider the situation when the bar has already displaced by
an amount u as shown in Figure 6.24(a). Since it is in equilibrium, summing the
moments about the base gives
; QL Pu + KuL = 0
; or K ; P=L]u = Q
It is worth noting at this stage the dierent roles played by the two forces: Q appears
as a regular applied load on the right-hand side, but P appears on the left-hand side
almost like a stiness term. We refer to this as an initial stress term and is the same
as the geometric stiness as discussed earlier.
Clearly, we can solve for the displacements for any combinations of loads, thus
u = K ;QP=L
These are shown plotted in Figure 6.25(a) for dierent values of Q. The obvious point
of note is that in the vicinity of P = KL the displacements are very large even for
negligible Q. (Note that it is a consequence of the linear approximations that the
innity occurs | de"ections in a real nonlinear situation would be nite as we see
later.) Keep in mind that every point shown in the gure is an equilibrium position,
and based on our discussion above we would declare that the structure is unstable
near P = KL. One could imagine, therefore, a loading sequence that need only \skip
over" this special value of load to land on one of the other equilibrium paths with

nite displacement we would then not have to worry about stability. The situation
is much subtler than this, and we need to rene our concepts of equilibrium and
stability in order to address it.
2.0
(a) (b) *

1.8
1.6
decreasing Q
1.4
Load P/Po
1.2 P=KL
unstable
1.0

.8 stable
.6
.4 decreasing Q equil points
P<KL
.2 P>KL
.0 Displacement Displacement
-1.0 -.8 -.6 -.4 -.2 .0 .2 .4 .6 .8 1.0 -1.0 -.8 -.6 -.4 -.2 .0 .2 .4 .6 .8 1.0
Figure 6.25: Behavior of the linear bar model. (a) Displacements for various values
of loads. (b) Potential energy for dierent P values and constant Q.
The strain energy and potential of the loads for our simple system are, respectively,
U = 21 Ku2 V = ;Pv ; Qu ; 12 P=L]u2 ; Qu
The approximation is based on small de"ections. The total potential is

*= +V= 1 Ku2 ; 1 P=L]u2 ; Qu

U
2 2
This is shown plotted in Figure 6.25(b) for dierent values of P but the same value
of Q. Note that most points on these plots are not equilibrium points | equilibrium
is where
F =
@ * = Ku ; P=L]u ; Q = 0
@u
. August 2007
and are shown as dots in the gure.
For P < KL, the plots show a series of valleys the bottoms of which correspond
to the equilibrium positions. For P > KL, the plots show a series of peaks the top
of which correspond to the equilibrium positions. We can draw a direct comparison
between these plots and the hill and valley of Figure 6.23. We thus go beyond declar-
ing just the point P = KL unstable and now see that all points beyond P = KL
are unstable. The special point P = KL is called the critical value or sometimes a
bifurcation point. In those cases when Q is small, the de"ections in the region close
by are small we will take advantage of this to do a linearized eigenvalue analysis to
determine these bifurcation points. In the literature (see Reference 30] for a very
large collection of solutions) this is generally known as a buckling analysis. It is em-
phasized, however, that such an analysis determines only the critical points and says
nothing of the post-buckling behavior.
The second variation in strain energy is related to the elastic stiness as
X X 2U
Kij = i j @u@ @u or KT ] = KE ] + KG]
i j
Hence the second variation of the total potential energy must be related to a stiness
like term actually, it is called the total stiness as discussed in Section 6.3. Hence
the conditions for stability of equilibrium can be applied to the total stiness matrix
that is, we inspect whether this matrix is positive denite, positive semi-denite, or
negative denite. A necessary and sucient condition that 2 * be positive denite
is that the determinant of the matrix KT ] and all its principal minors be positive.
Usually, it is unnecessary to examine a sequence of minors since the determinant itself
vanishes before any of its minors. Thus,
detKT ] = 0
is the criterion for the onset of instability.
III: Euler's Bifurcation View
Consider the loaded column shown in Figure 6.24(b) this is similar to that of Fig-
ure 6.24(a) except for the additional spring and that the de"ection can be large. We
use the angle o the vertical,
, as the independent variable and allow it to vary in
the range from zero to .
Since the frame member is rigid, we easily establish the relationship between the
vertical and horizontal displacements as
u = L sin
v = L cos
; L
The total potential energy is given by
* = 21 K1u2 + 12 K2v2 ; ;P ]v ; Qu
= 21 KL2 sin2
+ 12 K2L2 (1 ; cos
)2 ; PL(1 ; cos
) ; QL sin

The equilibrium congurations are found by setting
F =
@ * = hK cos
+ K (1 ; cos
) ; P sin
i sin
; Q cos
= 0
1 2
@
L L
Rearranging gives

P = K1 cos
+ K2(1 ; cos
) ; QL sin
cos
where we view Q simply as a parameter.

(a) * (b)
1.0 limit point
.8
P = 0.6 Pc
.6
.4
Load P/Pc
.2
decreasing Q
.0

-.2
-.4
-.6 unstable equil points unstable equil points
-.8
Angle Angle
-1.0
0. 30. 60. 90. 120. 150. 180. 0. 30. 60. 90. 120. 150. 180.
Figure 6.26: Behavior of the nonlinear bar model with K2 = 0. (a) Rotation for
various values of loads. (b) Potential energy for dierent P values and small constant
Q.
Figure 6.26(a) shows a plot of the force P against angle for dierent values of Q
when K2 = 0. In each case, the force cannot exceed a particular maximum value
and in the limit of small Q this corresponds to the critical value Pc = K1 L. One
interpretation of this gure is to view it as imposing the displacement and the force
is the resulting reaction. Suppose, however, we wish to impose the force, then what
is the interpretation of the decreasing force? More specically, suppose we are at the
peak value and we increment the load a small amount P , what happens?
To answer this question, we need to look at the potential energy plotted in Fig-
ure 6.26(b). The total potential energy (when K2 = 0) is given by
* = 21 K1L2 sin2
; PL1 ; cos
] ; QL sin
Again, it is emphasized that most points on the potential plots are nonequilibrium
points. Consider the line corresponding to the load values P = 0:6Pc in Figure 6.26(a).
It intersects the equilibrium curve at three points, but only two of the points (near 0
and 180) are stable. In other words, all points immediately past the peak are unstable
and therefore a load/angle combination in this range would cause a large displace-
ment. The member would rotate until it found the second stable equilibrium point
at a large angle.
. August 2007
The maximum load point is called a limit point, because the load cannot exceed
this limiting value. The phenomenon of quickly jumping from one equilibrium con-
guration to another distant one is called snap-through.
In the previous developments we saw that Q ;! 0 is a special case. Let us now
deal directly with this situation. The equilibrium relation is
F =
@ * = hK cos
+ K (1 ; cos
) ; P sin
i sin
= 0
1 2
@
L
This has two solutions. The rst is
sin
= 0 or
= 0 2 : : :

These correspond to when the bar is in a vertical alignment. The other solution gives
P = K1 cos
+ K21 cos
]
;
Both solutions are shown plotted in Figure 6.27 for dierent values of K2 =K1. We see
that the solutions intersect at P=K1L = 1, which is the critical value we previously
identied. The presence of K2 does not aect this critical value but it does change
the shape of the equilibrium curve. We now investigate its stability.
The rst variation of the potential energy is related to the equilibrium solution,
while the second variation determines its stability. This is given by
2
KT = @@
= @@
*2 = K1L2 (2 cos2
1) + K2L2 (1 2 cos2
+ cos
) PL cos
F
; ; ;
For the solution

= 0, this becomes
2
KT = @@
*2 = K1L2 PL ;
Hence, when P > K1L this solution path is unstable. That is, the intersection with
the second solution has caused this solution path to become unstable. On the other
hand, for
= this becomes
@ 2*
KT = @
2 = K1 L2 K2 L2 + PL
;
When K2 = 0, say, then the solution is stable as long as P is not reversed. If K2 > K1 ,
then a minimum value of P is required to maintain stability otherwise the bar will
snap back to the original upright position.
For the second solution with P = K1 cos
+ K2 1 cos
], we get
;
2
KT = @@
*2 = K2 K1]L2 sin2
;
3.0
stable
unstable
K2 > K1
2.5
Load [P/KL]
2.0
1.5
1.0
K2 = K1

;
;
.5
;
.0 P = K1 LAngle K2 < K1
-180 -135 -90 -45 0 45 90 135 180
Figure 6.27: Bifurcation view, two equilibrium paths intersect.

Thus, so long as K2 > K1 the solution paths are stable. In particular, if K2 = 0, the
solution is always unstable, conrming what we saw in Figure 6.26(a). The complete
picture is shown in Figure 6.27.
We now have the following bifurcation view of the loading process: as P is in-
creased, the solution (
= 0) is stable until the critical point P = K1L is reached a
further increment of load along this path makes the solution unstable | any distur-
bance will cause it to snap. Where it goes depends on the relative values of K1 and
K2: for K2 > K1 there is a small increase in
onto the stable path and the load can
then continue to be increased. For K2 < K1, there is a very large increase in
where
the bar snaps to the inverted position. This snap, in reality, would involve a dynamic
process.
IV: Dynamic View of Static Instabilities
It was dicult to discuss the previous notions of stability without simultaneously
mentioning dynamics. After all, even our simple notion asks the question of what
happens to the ensuing dynamics once the disturbance is applied.
A schematic of our dynamic view is illustrated in Figure 6.22. A stable loading
state (resulting from the slowly applied load P (t)) is illustrated by the segment A a
disturbance (in the form of a short duration ping load Q(t)) causes oscillations about
the equilibrium path these are temporary and the structure eventually comes back to
the equilibrium path. At, or beyond, a critical point, however, a small disturbance will
cause a signicant dynamic process to ensue. Depending on the particular problem,
a nearby equilibrium path may or may not be found. It is worth noting that the
new equilibrium path may not be statically connected to the original one that is, we
could not devise a proportional loading sequence (where the ratio of all the loads is
kept constant) to connect the two equilibrium states.
The nonlinear governing system of Equation (6.11) is easily modied to account
for inertia eects and leads to
M ]f u# g + C ]f u_ g = fP g ; fF (u)g
. August 2007
where M ] and C ] are the mass and damping matrix respectively. Let us conceive
of the total applied load and response as made up of two parts
f P (t) = Po(t) + Q(t)
g f g f g u(t) = uo(t) + (t)
f g f g f g
That is, there is the primary response u o, which is due to Po , and the smaller
f g f g
perturbation response , which is due to the ping load Q . Expanding F (uo + )

f g f g
in a Taylor series for small

f F (uo + )
g f F (uo) + @F
g
@u ]o + = F (uo) + KT ]o +
f g f g f g
and setting the corresponding powers of to zero leads to the two equations
0 : M ] u#o + C ] u_ o = Po
f g f F (uo) 0
g f g; f g
1 : M ] # + C ] _ + KT ]o = Q
f g f g f g f g
The rst equation is for quasi-static primary loading. The second equation, which
is often referred to as the variational equation 20], shows that the response due to
the ping is that of a linear system with a constant stiness KT ]o how the stiness
changes during quasi-static loading is governed by the rst equation.
As we showed for the total Lagrangian scheme, the incremental form of the quasi-
static loading relation essentially becomes
KT ] u t+t = P t+t F t
f g f g ; f g
and so long as KT ] is nonsingular we can uniquely determine the deformation. As

the deformation unfolds, however, the total stiness can change, and in particular it
can become singular | this is precisely the situation of interest here. After the ping
is applied, the system is in free vibration governed by
M ] # + C ] _ + KT ]o = 0
f g f g f g f g
We look for solutions of the form f (t)g = fgeit . Substitute into the equation of
motion to get h i
KT ] + i C ] ; 2 M ] fgeit = 0
There can be nontrivial solutions only if the determinant is zero, which leads to a
characteristic equation to determine the eigenvalues i and eigenvectors fgi. The
general solution is written as a combination of
f g1 e
i1 t f g2 e
i2 t f gN e
iN t
We state our stability criterion in terms of the properties of the eigenvalues i. For
the system to be asymptotically stable we want
Im ] > 0 since f geit = fgei(R +iI )t = fge I t eiR t ;
Thus, a negative imaginary component of i would give an exponentially increasing
function of time. If the criterion is not true for any one of the roots, then the system
is unstable.
The static instability criterion of Euler is essentially the case of 1 = 0 that
is, both the real and imaginary parts are zero simultaneously. There are structural
problems, however, where this criterion is insucient. For example, for follower-force
type problems (and related problems such as aeroelastic "utter) instability occurs
when the real part of 1 is still positive. Such situations are usually referred to as
dynamic (or kinetic) instabilities 35].
Thus, a key ingredient of the dynamic approach is to monitor the spectral behavior
of KT ]. Since the total stiness is available (when using the total Lagrangian scheme)
an expedient method is to do an undamped vibration eigenanalysis | we will then
refer to the eigenvalues as ! = !2, which are real only.
Agents and Imperfections

To summarize, motivated by dynamic heuristics, our view of static instability is that
the structure is in a state of unstable equilibrium when the second derivative of the
total potential energy is negative. The agent for the change in the above analysis was
an applied load Q quite separate from the primary loading P , but note that we get
a critical value even in the limit of Q = 0.
The agent Q is also referred to as an imperfection since it can be thought of as a
slightly mis-applied P , and since slight geometric imperfections have the same result.
The structure (and the solution) is called perfect when Q = 0. Imperfect structures
exhibit limit type instabilities while the perfect structure can also exhibit bifurcations.
In the remainder of this chapter, we will concentrate on static instabilities and
generally adopt the Euler approach of looking at equilibrium in the deformed con-
guration. We focus on monitoring the eigenvalues of the total stiness matrix, our
stability criterion being that if one of them goes to zero then the load is at a singular
point. This rules out static problems that have a non-symmetric total stiness | we
leave some of these cases until the next chapter where a fully dynamic view is devel-
oped. The discussion of stability involves an analysis of the post-buckling behavior
| this clearly needs our full nonlinear analysis techniques developed in the earlier
chapters. In many situations it is sucient just to know the rst critical load and
in the next few sections we develop methods for obtaining this information without
doing a complete nonlinear post-buckling analysis. In the nal sections, however,
we navigate through a complete load/unload cycle for a structure undergoing large
deformations and buckling.
Proportional Loading Along the Primary Path

With reference to Figure 6.22, we concentrate on the quasi-static primary loading
path, and assume all inertia eects are negligible. The total potential energy of our
. August 2007
general nonlinear system is
*(u ) = U (u) ; f u gT fP g
where U (u) is the strain energy that is a function only of the discrete displacement
vector f u g, fP g is a xed load vector, and is a scalar load multiplier. Since the
load vector never changes, we say that this loading is proportional, being controlled
by the single parameter . This is shown schematically in Figure 6.28
6 bfundamental
path
r
stable
B unstable
A; b limit point
u2 r bifurcation point
;
;
; - u1
;
Figure 6.28: Examples of limit and bifurcation singular points occurring along a
fundamental loading path.
First consider small changes in the total potential due to small changes in dis-
placement with xed
f gf
@ 2 * ] u +
*(u ) = @@u* u + 12 u T @u@u
g f g f g
We have that
@* = @ ( u ) @ 2 * ] = @ 2 ] = @ ] K ]
P
U U F
T
@u
f g f
@u g; f g fF g
@u@u @u@u @u
where KT ] is the total stiness. The variation of the potential is therefore given by
*(u ) = T
fF g f u + 21 u T KT ] u +
g f g f g
The energy changes should be stationary for an equilibrium loading path that is, the
rst variation should be zero irrespective of fug. Hence, we have that
@ * = (u ) = @
@u P = F P = 0
U
f
@u g fF g f g; f g f g; f g
This is the equation that denes the equilibrium path this path can be viewed as
a continuous curve in ( u T ) space. In this, only the nodal force vector F is a
f g f g
function of the displacements.

Now consider a loading history along a sequence of equilibrium states. In particu-
lar, consider two equilibrium states, A and B , a small fug and apart, as shown
in Figure 6.28. We have
fF g (u )jB = (uA + u A + )
fF g
= fF g(u )jA + @@u

F
]jAfug + f @@
F
gjA +
Since A and B are equilibrium states, then fF gjA and fF gjB are both zero giving
@ @ ] = @F ] u = K ] u
A u + P P 0
F F
@u
f gj f
@ A
g j
@u A j f g; T f g f g; f g
The approximation is because we are neglecting the higher-order terms. We will refer
to this as the loading equation. Provided that det KT ] = 0, we getj j 6
f ug = KT ] ; 1 fP g
This is the standard tangential solution used in a nonlinear analysis and described
earlier as part of the total Lagrangian scheme.
For stable equilibrium, the small changes of energy should be positive for any
small perturbation fug about the equilibrium point, hence we require that
*(u + u ) ; *(u ) > 0 or fu T KT ] u > 0
g f g for all fug
since fF gT fug = 0 through equilibrium. For this to be true, we require that KT ]
be positive denite. There are two situations of interest to us here. First, when
f u T KT ] u < 0
g f g for some fug
then KT ] is not positive denite and it will have at least one negative eigenvalue. It
is therefore unstable. The other case is when
fu T KT ] u = 0
g f g for some fug
which is a neutral equilibrium state and KT ] has a zero eigenvalue. Consequently,
det jKT ]j = 0 and a full investigation of the nature of its equilibrium requires the use
of higher-order terms in the expansion of the potential.
For the neutral equilibrium case, we cannot nd a unique fug using the loading
equation and we have a singular point. This singular point can be either a limit point
or a bifurcation point. To see the distinction between these two, we must look at the
spectral properties of the total stiness and this is done in the examples to follow.
Figure 6.28 shows examples of the two singular points. As shown in the gure, a 2-D
plot of u2 against load shows little or no motion until the singular point is reached,
then it has two possible paths to take. In the simple example illustrated, the paths
. August 2007
(a) 6 q
r q
6
(b)
q
-u -u
6 6
stable
(c) r unstable
r bifurcation point
limit point
q q (d)
-u -u
Figure 6.29: Classication of singular points and the eect of initial imperfections.
(a) Limit point. (b) Asymmetric bifurcation. (c) Stable symmetric bifurcation.
(d) Unstable symmetric bifurcation.
can be concave up (stable), concave down (unstable), or asymmetric (one stable, one
unstable).
We are now in a position to classify each of the singular points. For a limit point,
= 0, giving two solutions symmetrically placed about the limit point. In either
case, there is a direction in which negative energy results and hence a limit point is
unstable.
For a bifurcation point, there are two cases that can arise. For an asymmetric
bifurcation there are two solutions, one corresponding to the fundamental path and
is unstable. For the other solution, the energy change is the same as for the limit
point and hence an asymmetric bifurcation point is unstable.
For a symmetric bifurcation, again there are two solutions, The primary path is
unstable. The bifurcated path may be stable or unstable.
Path Following Methods

Path-following schemes 7] are the computational implementation of the static meth-
ods Reference 12] gives an excellent application and discussion (with many refer-
ences) of current generalized path-following procedures. Severe diculties can be
encountered with limit points where the load-de"ection curve becomes horizontal.
The arc-length methods were introduced to overcome these diculties. The essence
of the arc-length method is that the load parameter becomes a variable just like the
displacement variables these N + 1 unknowns are solved by using N equilibrium
equations and a constraint equation. Various forms of constraint equations can be
used, a good discussion of some of the simpler ones is given in Reference 6] and a
comprehensive survey is given in Reference 33].
Implementing the arc-length method requires a level of programming sophistica-
tion beyond the level directed by this book. As an alternative, the dynamic approach
accepts that the structural behavior is dynamic in the vicinity of a critical point and
follows the ensuing motion. The next section discusses this approach.
Example 6.14: Show the connection between a buckling eigenanalysis and a

vibration eigenanalysis at the elevated load.
The vibration eigenvalue problem is
KT ]f gi ; !i2 M ]f gi = 0
At the static singular point, !1 = 0, we have
KT ]f g1 = 0
Now consider the case when the structural response is only slightly nonlinear, then
we can represent the total stiness in the form
KT ] KE ] + KG ]
where KE ] is the elastic stiness, KG ] is the geometric stiness, and is a loading
factor. At the singular point, we therefore have
KT ]f g1 = KE + KG ]f g1 = 0
This is the eigenvalue problem for the buckling of the structure, and we conclude
that the vibration mode shape is the same as the rst buckling mode shape. This
result will be demonstrated later for plates.
Example 6.15: Use a vibration analysis to distinguish between limit and bifur-
cation singular points.
Consider the free undamped vibration of the system when loaded near a critical
point that is, let fug = f geit leading to
KT ]f g ;
2 M ]f g = 0
It is therefore sucient for us to introduce the eigenvalues m and eigenvectors f gm
of the total stiness such that
KT ]f gm = m M ]f gm m =
2m
and let the eigenvectors be normalized such that
f gTi M ]f gj = ij f gTm KT ]f gm = m
When the solution follows the path from a stable state, the lowest eigenvalue, 1 , is
zero at the singular point. Hence, we have that
KT ]f g1 = 0
. August 2007
Now multiply the transpose of the loading equation by f g1 to get
n o
fugT KT ] ; fP gT f g1 = 0 giving fP gT f g1 = 0
This relation is used to distinguish between limit and bifurcation points as follows:
limit point: = 0 fP gT f g1 6= 0
bifurcation point: 6= 0 fP gT f g1 = 0 (6.17)
Figure 6.28 shows examples of the two singular points. As shown in the gure, a 2-D
plot of u2 against load shows little or no motion until the singular point is reached,
then it has two possible paths to take. In the simple example illustrated, the paths
can be concave up (stable), concave down (unstable), or asymmetric (one stable,
one unstable).
Example 6.16: Use a modal analysis to discuss the displacement increment

shape at singular points.
We can get further insight into these singular points by using a modal represen-
tation of the displacement increment in terms of the eigenvectors
X
fug = 1f g1 + 2 f g2 + = m f gm
Now substitute this into the loading equation, multiply the resulting equation by
f gm , then making use of the orthogonality properties leads to
; f gT f g = 0 or = 1 f gT f g
m m P m m m P m
We therefore have the general modal representation for the displacement increment
X 1h T i
fug = f g +
1 1 f g f g f g = f g + f v g
P m m 1 1
m=2 m
Note that f gT1 f v g = 0. For the limit point where = 0 and assuming m 6= 0,
we get
fug = 1 f g1
The displacement increment has the shape of the rst eigenmode. The displacement
increment for the bifurcation point, on the other hand, depends on all the modes.
Exercises 229
Exercises
6.1 The total potential of a certain system is
* = 15 x5 ; 41 ax4 ; 23 ax3 + a2
where a is a parameter and x is the generalized coordinate. Determine all of
the equilibrium congurations and indicate which ones are stable and unstable.
6.2 If in the derivation of the beam geometric stiness matrix, we use the rod shape
functions instead of the beam shape functions, i.e.,
v(x) = (1 ; Lx )v1 + ( Lx )v2 = f1 (x)v1 + f2 (x)v2
Show that the derived inconsistent geometric stiness matrix is given by
2 1 0 ;1 0 3
6 7
kG ] = FLo 64 ;01 00 01 00 75
0 0 0 0
Where might such an element be more useful than its consistent counterpart?
6.3 Determine a constraint element by adding
1 2
2 ki ui + j uj ; o ]
to the potential energy term. Generalize the results for more than two con-
straints.
6.4 A particle in the (x y) plane moves in the force eld Fx = ;ky, Fy = kx (k
is a constant). Prove that when the particle describes any closed path in the
counterclockwise sense, the work performed on the particle is 2kA where A is
the area enclosed by the path.
6.5 Suppose the rotation distribution in a circular shaft can be written in terms of
the nodal rotations as
(x) = (1 ; Lx )1 + ( Lx )2 f1(x)1 + f2 (x)2
Use minimum potential energy to derive the torsion element.
6.6 For a uniform column with clamped ends, assume v = ax2 (L ; x)2 and deter-
mine the critical load. Pcr = 42EI=L2 ]
6.7 For a uniform column with clamped at one end and free at the other, assume
v = ax2 and determine the critical load. Pcr = 2:5EI=L2 ]
6.8 For a uniform column with clamped at one end and free at the other, assume
the mode shape is the static deection shape due to a point load at the tip and
determine the critical load. Pcr = 42EI=17L2 ]
. August 2007
Chapter 7
Variational Methods
Variational methods can be used as an alternative statement of the basic equations of

mechanics. Consequently, they can be used as a means of deriving the basic governing
equations of the theory of elasticity and this will be demonstrated by way of the
Calculus of Variations. Perhaps more importantly, however, they can also be used
to obtain approximate solutions in particular they form the basis for constructing
rational approximate structural theories.
The concepts to be discussed have an intimate relationship to the principle of
virtual derived in Chapter 6. Here the emphasis is on the calculus of variations
as a mathematical tool to formalize the calculus for the variations associated with
virtual work and the stationary principles. While the principles are not necessarily
mechanics based, it must be kept in mind, however, that their application to solids
and structures can only be justied by the same mechanics reasoning behind virtual
work.
7.1 Calculus of Variations

We are familiar with nding the extreme values of a function u(x). We rst set
du = 0
dx
and then solve the equation to obtain x = x1 x2 at which the function u(x) as-
sumes extrema. We shall be concerned with the calculation of the extreme values of
functions dened by certain integrals whose integrands contain one or several func-
tions assuming the roles of arguments. We shall use the term functional to refer to
functions dened by integrals whose arguments themselves are functions.
Single Integrals
Consider the integral
Z x1
J u ] = F (x u u )dx
0
e.g., F = 21 u ]2 + u ; q(x)
0
xo
231
232 Chapter 7. Variational Methods
where F is a known, real function of the real arguments x u, u du=dx, and
0
are parameters of the problem. The value of the integral depends on the choice of
u = u(x), hence, the notation J u ]. To make the symbol J u ] meaningful, it is
clearly necessary to impose some restrictions on the choice of the argument u(x) and
on the prescribed function F appearing in the integrand. We shall suppose that the
admissible arguments belong to a class C 2 (smooth in the second derivatives) and
assume at the end of the interval (xo x1) the specied values are uo and u1. Thus,
u(xo) = uo u(x1) = u1
where uo and u1 are prescribed in advance. The entire set of admissible arguments
u(x) can thus be viewed as a family of smooth curves passing through (xo uo) and
(x1 u1).
For a given curve u = u(x) of the set, the integral yields a denite numerical
value J u ], and we pose a problem of determining that particular curve u(x) in the
competing set which makes the integral a minimum. If u(x) minimizes this integral,
then every function u(x) in the neighborhood of u(x) can be represented in the form
u = u(x) + (x) = ddu
where is a small real parameter. We note that the function u(x) is determined with
= 0. We shall call the dierence u(x) ; u(x) = (x) the variation of u(x) and write
u (x) = ddu
Moreover, every function in the set fu(x)g satises the end conditions and we must
have (xo) = 0 and (x1 ) = 0 at the ends.
Since u(x) minimizes the integral, then
J u ] = J u + ] J u ]
The left-hand member in this inequality is a continuously dierentiable function of ,
and therefore, a necessary condition that u(x) minimize the integral is
dJ u + ] j = 0
d =0
That is,
dJ u + ] j = d Z x1 F (x u + u + )dx = Z x1 ( @F + @F )dx = 0
d =0 d xo
0 0 0
xo @u @u 0
Denote the variation of J as

J dJ ud+ ] j=0
7.1. Calculus of Variations 233
then the above condition can be denoted as J = 0. In anticipation of later develop-
ments, we now introduce a subscript notation to indicate the partial derivatives. We
have, for example,
F u @F
@u F u
@F
@u
0
0
With this notation, we rewrite the derivative as

dJ u + ] j = Z x1 (F + F )dx = 0
d =0 xo u u
0
0
Integrate the second term in this by parts to get

Z x1 x1 Z x1 dF
F u dx = F u ; u
dx
0
xo dx
0 0
xo xo
and since (xo) = (x1 ) = 0, we can write it in the form
Z x1 dF u
!
F u ; dx (x) dx = 0
0
xo
This integral must vanish for every (x), and we conclude therefore that
F u dFdxu = 0
0
;
is a necessary condition that the integral be minimized by u = u(x). On expanding

it we get the second order dierential equation
2
F u u ddxu2 + F u u du
0 0
dx + F u x F u = 0
0 0 ;
This is usually
R called the Euler equation associated with the minimizing of the integral
J u ] = xxo1 F (x u u )dx.
0
Similar calculations performed on the functional

Z x1
J u ] = F (x u u u 0 00
u(n))dx
xo
yield the Euler equation
F u dx d F + d2 F ( 1) n d F u(n) = 0
n
u u
;
dx2 dxn ;
0 00
It is noted that a functional involving n derivatives of u results in an ordinary dier-

ential equation involving n derivatives of the function F .
. August 2007
Double Integrals
We now consider the problem of minimizing the double integral
Z Z
J u ] = F (x y u u x u y )dxdy u x @u
@x u y
@u
@y
R
on the set fu(x y)g of functions of class C 2 where each u(x y) in the set takes specied
continuous values, u = (s), on the boundary S of the region R.
Let us suppose that a certain function u(x y) in this set is included in the formula
u(x y) = u(x y) + (x y) u
where is a small parameter. Since u = (s) on the boundary of R, (x y) = 0 on

S . We form the integral J u + ] and observe that
dJ u + ] = 0 or J = 0
d =0 j
since u(x y) minimizes the functional. But,

Z Z
J u + ] = F (x y u + u x + x u y + y )dxdy
R
so that
Z Z Z Z
J = (F u +F ux x +F uy y )dxdy = (F u u+F ux u x +F uy u y )dxdy
R R
Notice that we do not take variations with respect to x or y. We rewrite this as
Z Z @F ux @F uy
! Z Z "@ @
#
J = F u ; @x ; @y dxdy + @x (F ux ) + @y (F uy ) dxdy
R R
Recall the Green's Theorem in a plane from Section 1.3,
Z Z @Q @P ! Z
@x @y
; dxdy = (Pdx + Qdy)
R S
Apply this to the second integral to obtain
Z Z @F ux @F uy
! Z
J = F u ; @x ; @y u dxdy + (F ux dy ; F uy dx)
R S
But = 0 on S , and since J vanishes for an arbitrary choice of in R, we conclude
that
F u ; @F@xux ; @F@yuy = 0 in R
for the minimizing function u(x y).
A calculation similar in every respect to the foregoing for the functional
Z Z
J u ] = F (x y u u x u y u xx u yy ) dxdy
R
in which the admissible u assume specied continuous values on the boundary S of
R, leads to the Euler Equation:
@ (F ) ; @ (F ) + @ 2 (F ) + @ 2 (F ) + @ 2 (F ) = 0
F u ; @x ux
@y uy @x2 uxx @x@y uxy @y2 uyy
These examples show that we can associate a dierential formulation with a vari-
ational formulation and vice versa.
Example 7.1: Consider the following functional and boundary conditions

Z Zh i
J u ] = (u x )2 + (u y )2 + 2f (x y)u dxdy u = (s) on S
R
Determine the corresponding dierential equation.
The Euler equation corresponding to J u ]u) = min is
2f ; 2u xx ;2u yy = 0
This can be written more familiarly as
r2u = f (x y) in R
This is Laplace's equation.
Example 7.2: Consider the inhomogeneous bi-harmonic equation given by

r4 = c = constant
with the homogeneous boundary conditions on the rectangle
@2 = @2 = 0 on x = a
@x@y @y2
@2 = @2 = 0 on y = b
@x@y @x2
What is the corresponding variational statement of the problem?
The dierential equation for can be identied as the Euler equation for the
variational problem
ZbZah i
J ] = (r2 )2 ; 2c dxdy = min
; b ; a
These two examples show the connection between satisfying dierential equations
and minimizing a functional.
. August 2007
The Variational Operator
We now introduce the notation of `variations' in order to show the analogy between
the calculus of variations and dierential calculus.
We drop the notation and replace it with the formal operator where
u

We say that u is the variation of u. Corresponding to the variation in u, we have

the rst variation of the functional F (x u u )
0
F = @F
@u u + @F u +
@u 0
0
In the more general case of a functional F (x y u v u x v y ), we have

F = @F@u u + @F v + @F u x + @F v y +
@v @u x @v y
Note that we do not vary x or y, that is, x = 0 and y = 0. With this in mind, the
analogy with the denition of the dierential is complete. That is, the dierential
of a function is a rst order approximation to the change in that function along a
particular line, while the variation of a functional is a rst order approximation to
the change from curve to curve.
The laws of variations of sums, products, powers and so forth, follow those of the
dierential. Thus,
(uv) = u v + u v ( uv ) = u v v2 u v
;
The operators , @ , and d are commutative, that is

d u = du @ F = @F
dx dx @u @u
We also have that Z2 Z2
J = F dx = F dx
1 1
since the limits of the integral do not vary.
Boundary Conditions
Up to now, we have considered the boundary conditions where u is specied and thus
u = 0. We now generalize this to encompass a wider range of problems.
Consider the 1-D functional equation
Z x2
J u ] = F (x u ::) dx ; (u2 ; u1) = min
x1
where is a parameter, and u1 u2 are the end values of u(x). To minimize this, we
apply the variational operator to get
Z x2
J = F dx ; (u2 ; u1) = 0
x1
We get for terms inside the integral
F = @F@u u + @F u + @F u +
@u @u 0
0
00
00
Integrating the second term by parts gives

Z x2 @F
u dx =
Z x2 @F du
dx = @F u]x2 ; Z x2 d ( @F )u dx
@u x1 x1 dx @u
0
x1 @u 0
x1 @u dx 0 0 0
We must integrate the third term by parts twice. This results in

Z x2 @F @F d @F x2 Z x2 d2 @F
u dx = @u u ]x1 ; dx ( @u )u] + dx2 ( @u )u dx
00 x 2 0
x1 @u00 00
x1 x1
00 0
The variational terms can be collected as

Z x2 h @F d @F d2 @F i h @F d @F i x2 h @F i x2
J = + u dx + ; u + u x1= 0 0
x1 @u dx @u dx2 @u @u dx @u x1 @u
; ;
0 0 0 00 00
All three groups of terms must vanish. The rst group gives rise to the governing
dierential equation, or the Euler equation
@F d @F + d2 @F = 0
@u dx @u dx2 @u ;
0 0
The remaining terms give rise to the boundary conditions. Since they are associated
with dierent variations of u, then we have separately
h @F d @F i x2 h @F i x2
@u dx @u 0
; ux1= 0
00
;
@u u x1= 0 00
0
Each of these represent two distinct types of boundary conditions. The rst of these,
for example, says at x = x1 that
u = 0 u = specied or @u @F d @F = 0
dx @u 0
;
00
;
The rst is called a geometric boundary condition, while the second is called a natural
boundary condition, respectively.
When we apply the stationary principles, we need to identify two classes of bound-
ary conditions, called essential and natural boundary conditions. The essential
. August 2007
boundary conditions are also called geometric boundary conditions because they cor-
respond to prescribed displacements. The natural boundary conditions are sometimes
called the force boundary conditions because they correspond to prescribed boundary
forces in structural mechanics problems.
Example 7.3: Consider the function

F = 21 EAu ]2 ; q(x) 0
where EA is possibly a function of x but not u. Determine the governing dierential

equation and the corresponding boundary conditions.
The various derivatives are
@F = ;q @F = EAu 0
@F = 0
@u @u 0
@u00
The Euler equation is

d EA du ] ; q = 0
; dx dx
with the associated boundary conditions
u = specied or du ; = 0
EA dx
This corresponds to the axial loading of a rod of variable stiness EA, and we
recognize as the applied load.
The variational formulation can be written as
Z x2 h i x
J u ] = 1 2
2 EAu ] ; q(x)
0
dx ; ux21 = min
x1
R
As we will see next, the terms xx12 q(x) dx + ujxx21 will have the interpretation of the
potential of applied loads.
7.2 Direct Methods of Solution

Classical methods in the calculus of variations reduce the fundamental question of
the existence of a solution for an arbitrary problem to the question of the existence of
solutions of dierential equations (the Euler equations). This approach is not always
eective, and is greatly complicated by the fact that what is needed to solve a given
variational problem is not a solution of the corresponding dierential equation in a
small neighborhood of some point, but rather a solution in some xed region, which
satises prescribed boundary conditions on the boundary of R. The diculties in-
herent in this approach (especially when several independent variables are involved)
have led to a search for variational methods of a dierent kind, known as direct meth-
ods, which do not entail the reduction of variational problems to problems involving
dierential equations.
7.2. Direct Methods of Solution 239
Once direct variational methods have been developed, they can be used to solve
dierential equations, and this technique, the inverse of the one we have discussed,
plays an important role in the modern theory of the subject. The basic idea is
the following: Suppose it can be shown that a given dierential equation is the
Euler equation of some functional, and suppose it has been proven somehow that this
functional has an extremum for suciently smooth admissible functions, then this
very fact proves that the dierential equation has a solution satisfying the boundary
conditions corresponding to the given variational problem. Moreover, variational
methods can be used not only to prove the existence of a solution of the original
dierential equation, but also to calculate a solution to any desired accuracy.
Ritz Method
The Ritz (or Rayleigh-Ritz) method provides a powerful way to obtain approximate
solutions directly from the variational problem. That is, we will seek
J u ] = min or J = 0
without relying on solving the corresponding Euler equations. In elasticity problems,
the energy principles conveniently set up the variational problems ready for applica-
tion of the Ritz method. However, this method can be also used to solve dierential
equations if they can be identied as the Euler equations of a variational problem.
In general, a continuously distributed deformable body consists of an innity
of material points and therefore has innitely many degrees of freedom. The Ritz
method is an approximate procedure by which continuous systems are reduced to
systems with nite degrees of freedom. The fundamental characteristic of the method
is that we operate on the functional corresponding to the problem, either J or the
total potential energy *. Suppose we are looking for the solution for J = 0 with
prescribed boundary conditions on u. Let
X
u(x y z) = i=1aii(x y z)
1
where i are independent expansion or trial functions, and the ai are multipliers
to be determined in the solution. The trial functions satisfy the essential (geomet-
ric) boundary conditions but not necessarily the natural boundary conditions. The
variational problem states that
J u ] = J a1 1 + a2 2 + ] = min
Thus J a1 1 + a22 + ] can be regarded as a function of the variables a1 a2 . To
satisfy J = min., we require that
@J = 0 @J = 0
@a1 @a2
. August 2007
These equations are then used to determine the coecients ai. Normally, we only
include a nite number of terms in the expansion.
An important consideration is the selection of the trial functions i. Selecting
ecient admissible functions may not be easy fortunately, many problems closely
resemble other problems that have been solved before, and the literature is full of
examples that can serve as a guide. It must also be kept in mind that these func-
tions need only satisfy the essential boundary conditions and not (necessarily) the
natural boundary conditions. For practical analyses, this is a signicant point and
largely accounts for the eectiveness of the displacement-based nite element analysis
procedure as will be shown later in this chapter.
For convenience in satisfying the boundary conditions on u, we usually set
X
u = uo + nan n
where uo conforms to the non-homogeneous boundary conditions. For homogeneous
displacement boundary conditions, we set uo = 0.
Example 7.4: Consider a bar xed at one end and subjected to an axial con-
centrated force at the other end, as shown in Figure 7.1. The variation of Young's
modulus is E (x) = Eo (1 + x=L)2 . Obtain a Ritz approximate solution.
6
y u(x)
6
- x - x
......
.....
.....
......
.....
...... P
-
......
.....
......
..... 6
F (x)
E = Eo (1 + x=L)2
- x
Figure 7.1: Bar with variable modulus.

The boundary conditions for this problem are:
essential: ujx=0 = 0 du j = P
natural: EA dx x=L
The exact solution is easily calculated to give
Zx
u(x) = (x)dx = EPLA (1 +x=L
x=L) F (x) = EA du
dx = P = constant
0 o
Both the displacement distribution and force distribution are shown plotted in Fig-
ure 7.1. We will use these results to evaluate the quality of the Ritz approximate
solutions. Specically, we wish to investigate the use of dierent trial functions.
Since the deformation is one-dimensional, then the strain is
xx = @u =
@x dx
du
and the total potential energy of the body is

ZL du 2
* = 12 EA dx dx ; PuL
0
The integration over the cross-sectional area has already been performed. We will
calculate the displacement and force distributions using the following assumed form
for the displacement:
u(x) = a0 + a1 x + a2 x2
This must satisfy the essential boundary condition, hence a0 = 0. Note that the
remaining polynomial does not necessarily satisfy the natural boundary condition.
Substituting the assumed displacements into the total potential energy expression,
we obtain
ZL
* = 21 EoA(1 + x=L)2 (a1 + 2a2 x)2 dx ; P (a1 L + a2 L2 )
0
Invoking the stationarity of * with respect to the coecients an , we obtain the
following equations for a1 and a2
@ * = Z L E A(1 + x=L)2 (a + 2a x)dx ; PL = 0
@a1 o 1 2
0
@ * = Z L E A(1 + x=L)2 (a + 2a x)2x dx ; PL2 = 0
@a2 o 1 2
0
Performing the required integrations gives
Eo A 70L 85L2 a1 = PL
30 85L2 124L3 a2 PL2
Note that this is symmetric. Solving this system gives for the two coecients
a1 = 78 P
97 Eo A a2 = ;9730 E PAL
o
This Ritz analysis, therefore, yields the approximate solution
u(x) = 9778EPA x ; 2610L x2 ]
o
and the force distribution is
F (x) = EA du =
dx 97
78P 1 ; 10 x](1 + x=L)2
13L
These results are shown in Table 7.1 as the 2-term columns. The most striking
fact of these results is the accuracy of the displacements and yet the axial force is
. August 2007
displ force
x=L exact 1 term 2 term bi-linear exact 1 term 2 term bi-linear
0.0 0.0 0.0 0.0 0.0 1.0 0.428 0.804 0.6316
0.5 0.3333 .2143 .3247 .3158 1.0 0.96 1.113 1.421
0.5 0.3333 .2143 .3247 .3158 1.0 0.96 1.113 0.729
1.0 0.5 .4285 .4948 .4779 1.0 1.714 0.740 1.297
Table 7.1: Displacement and force results for the non-uniform rod.
not constant and equal to P . This reiterates the fact that the Ritz approach only
approximates equilibrium.
An interesting result is obtained if we use only a linear expansion for the dis-
placements. In this circumstance, after imposing the essential boundary condition,
we have the one term expansion
u(x) = a1 x
Substituting this into the potential energy expression and minimizing, we obtain
@ * = Z L E A(1 + x=L)2 (a )dx ; PL = 0
@a1 o 1
0
Performing the required integration gives
Eo A 70L]a = PL
30 1
We recognize this as precisely the rst term in the above 2 2] matrix form. That
is, as we increase the expansion of u(x), then each additional term adds a row and
column to the matrices but otherwise the existing matrices are unaected.
Solving for a1 gives
a1 = 73 EPA
o
The approximate solution for the displacement and force are, respectively,
u(x) = 7E3PA x] F (x) = 37P 1](1 + x=L)2
o
These results are also shown in Table 7.1 as the 1-term columns. Note that this
force distribution does not satisfy the dierential equation of equilibrium.
Example 7.5: Use the Ritz method to nd an approximate solution for a rect-
angular body xed at one end and loaded with a concentrated force P at x = L as
shown in Figure 7.2.
The solution given by the elementary beam theory is
3h x i 2h x x )2 iy
v(x) = PL
6EI L3( )2 ; ( x )3
L u(x) = ;y(x) = ; PL
2EI L2( ) ; ( L
......
.....
......
.....
......
.....
......
.....
a
......
.....
...... y 6
.....
......
..... 6
......
..... - x 6P 2h
......
.....
......
.....
......
.....
......
.....
......
.....
......
..... ?
......
..... -
L
Figure 7.2: Cantilevered rectangular body.
The tip deection is
PL3 3 ; 1] = PL3
vtip = 6Eb(2 h)3 =12 2Ebh3
We will use this to motivate the Ritz functions as well as for comparison.
Consider the problem as a plane problem such that either zz = 0 (plane strain)
or zz = 0 (plane stress). In either case, the strain energy expression reduces to
Z
U = 21 (2
(1 + )(2xx + 2yy ) + 2
xxyy + 4
2xy ] dV
where is chosen appropriately as given in Section 5.2. Motivated by the elementary
solution, let the displacements be approximated as
u(x y) a1 2Lx ; x2 ]y v(x y) a2 3Lx2 ; x3 ]
This expansion satises the geometric boundary conditions of all displacements being
zero at x = 0. There are of course many displacement elds that can satisfy these
conditions.
The strains associated with this displacement eld are
@u = 2a L ; x]y @v = 0 2xy = @u @v 2
xx = @x 1 yy = @y @y + @x = (a1 + a2 3)2Lx ; x ]
Note that these strains are automatically compatible, but the stresses associated
with them are not necessarily in equilibrium. That is, we must determine an so as
to obtain a `good' solution. The strain energy is calculated as
Z +h Z L h i
U = 21 2
(1 + )4a21 L2 ; 2Lx + x2 ]y2 + 0 +
(a1 + 3a2 )2 4L2 x2 ; 4Lx3 + x4 ] dx dy dz
h
; o
1 8 2 3 3 2
= 2 2
(1 + ) 9 a1 L h b +
(a1 + 3a2 ) 15 L hb 16 5
The potential of the external load is given by

V = ;Pv(x = L y = 0) = ;2Pa2 L3
The total potential energy is therefore
h i
* = U + V ] = 21 2
(1 + ) 98 a21 L3 h3 b +
(a1 + 3a2 )2 16
15 L 5 hb ; 2Pa2 L3
. August 2007
This is the function to be minimized.
We consider * to be a function of a1 a2 , and dierentiate with respect to an to
get the stationary values. That is,
@ * = 0 = 2
(1 + ) 8 a L3h3 b +
(a + 3a ) 16 L5 hb
@a1 9 1 1 2 15
@ * = 0 =
(a + a 3) 16 L5 hb ; 2PL3
@a 1 2 5
2
which gives the solution
h 2i
a1 = 2
(1;+3P)4bh3 a2 = 2
(1 +P)4bh3 1 + 5(1 9+L2)h
The displacement eld is given by
u(x y) = 2
(1;+3P)4bh3 2Lx;x2]y
2
v(x y) = 2
(1 +P)4bh3 1+ 5(1 9+L2)h ]3Lx2 ;x3]
The tip deection is
3 5(1 + )h2 ]
vtip = 2
(1 PL
+ )2bh 3 1 + 9L2
The dierence with the elementary theory occurs in two places. First, the Ritz
solution has an extra term that depends on the ratio h=L this disappears for slender
beams | it is the shear deformation contribution. The leading terms are compared
as
PL3 ! PL3 or 1 ! 1
2
(1 + )2bh 3 2Ebh 3 (1 + ) (1 + )
where E = 2
(1 + ) was used. The elementary theory is based on uniaxial stress
which is dierent than both plane problems.
Weighted Residual Methods

When the dierential equation is available, we can work directly with it to form an
approximate solution. Let it be required to solve a linear dierential equation
L(u) = 0 in R
subjected to some linear homogeneous boundary conditions. Suppose we have an
assumed solution u~ then L(~u) will give some error or residual. In the weighted residual
method, this error is minimized over the domain by
ZZ
W (x y)L(~u(x y)) dxdy = min
R
where W is some weighting function. There are a number of methods that fall into
this category | they dier only in their specication of the weighting function. In
any event, we see that the dierential formulation has been recast as a variational
formulation.
I. Galerkin Method
In 1915, Galerkin proposed a method of approximate solution of the boundary-value
problems in mathematical physics that is of much wider scope than the method of
Ritz. The Galerkin method when applied to variational problems with quadratic
functionals, reduces to the Ritz method. It is important to note that in Galerkin's
formulation there is no reference to any connection of a potential. Indeed, the Galerkin
method can be applied to a broad class of problems phrased in terms of integral and
other types of functional equations.
Assume, for simplicity of exposition, that the domain R is two-dimensional, we
seek an approximate solution of the problem in the form
X
U N (x y) = Ni=1 ii(x y)
where U N satisfy the same boundary conditions as u(x y). The nite sum ordinarily
will not satisfy the dierential equation, and the substitution of U N will yield an error
L(U N ) = N (x y) N (x y) 6= 0 in R
If max N (x y) is small, U N (x y) can be considered a satisfactory approximation of
U (x y). Thus, N (x y) can be viewed as an error function, and the task is then to
select the i to minimize N (x y).
A reasonable minimization technique is suggested by the following: if we represents
UP(x y) by the series U (x y) = Pi=1 ii and consider the N th partial sum U N =
1
N then the orthogonality condition

i=1 i i
ZZ
L(U N )j (x y)dxdy = 0 as N ! 1
R
is equivalent to the statement that L(u) = 0. This led Galerkin to impose on the
error function L(U N ) a set of orthogonality conditions
ZZ
L(U N )j (x y)dxdy = 0 (j = 1 2 : : : N )
R
yielding the set of N equations
ZZ X N
L( ii)j dxdy = 0 (j = 1 2 : : : N )
R i=1
for the determination of the constants i in the approximate solution.
The resulting equations are always symmetric and positive denite if the operator
L is symmetric and positive denite. Note that the addition of natural boundary
conditions also destroys the symmetry. In comparison to the Ritz method, this ap-
proach can be applied to problems for which the governing equations are known but
for which a potential does not exist. However, it has the disadvantage that higher
order continuity conditions must be imposed when applied to a discretized domain.
When the operator is symmetric and positive denite, integration by parts recover
the Ritz formulation.
. August 2007
II. Least Squares
We can also form the squared error as
ZZ
L(U N )L(U N )dxdy = (j = 1 2 : : : N )
R
This becomes
ZZ X
N
L( ii)L(j )dxdy = 0 (j = 1 2 : : : N )
R i=1
The resulting equations are always symmetric and positive denite. This also has the
disadvantage that higher order continuity conditions must be imposed when applied
to a discretized domain. Note that integration by parts will not reduce the order of
derivatives occurring in the functional.
Example 7.6: Consider the rectangular panel shown in Figure 7.3 with end
parabolic tractions. Regarding the panel being in a state of plane stress, deter-
mine the distribution of stresses.
y
6 2
tx = po(1 ; yb2 )
ty = 0
b -
- x-
a
-
Figure 7.3: Parabolic traction distribution.

As shown in Chapter 5, for plane problems if we take the stresses derived from
a stress function as
@2
xx = @y@y @2
yy = @x@x @2
xy = ; @x@y
then the stresses automatically satisfy the equilibrium equations. Furthermore, if
the stress function is restricted to being bi-harmonic
r4 = 0
then the compatibility conditions are also satised. In the following, we will seek an
approximate solution for (x y).
The boundary conditions in terms of (x y) are given by
@2 = 0 2 h 2i
on x = a ty = xy = ; @x@y tx = xx = @@y2 = po 1 ; yb2
on x = b @2 = 0
tx = xy = ; @x@y
2
ty = yy = @@x2 = 0
Let the stress function be composed of two parts
h 2i
= o +
where o = 12 po y2 1 ; 6yb2
This will satisfy the traction boundary conditions. Then the bi-harmonic equation
becomes
r4 = 2bp2o
and the boundary conditions reduces to

@2 = @2 = 0

on x = a
@x@y @y2
@2 = @2 = 0

on y = b
@x@y @x2
The dierential equation for can be identied as the Euler equation for the

variational problem
ZbZa
J ] =

(r2 )2 ; 4bp2o

dxdy = min
; b ; a
Thus, the problem can be solved by the direct method. Assume
= (x2 ; a2 )2(y2 ; b2 )2(a1 + a2 x2 + a3 y2 + )

Such an expansion satises the boundary conditions and the expansion functions
are independent. If one term is taken, then
!
@J = 0 ) a 64 + 256 b2 + 64 b4 = po
@a1 1 7 49 a2 7 a4 a4 b2
For a square plate (a = b), we nd
a1 = 0:04253 apo6
Thus the approximate stress function is given by
h 2i
= 21 poy2 1 ; 6ya2 + 0:04253 ap6o (x2 ; a2 )2 (y2 ; b2 )2
The corresponding stress components are obtained from as
2 2 2
xx = po(1 ; ya2 ) ; 0:1702 po (1 ; 3ay2 )(1 ; xa2 )2
2 2
yy = ; 0:1702 po (1 ; 3ax2 )(1 ; ay2 )2
2 2
xy = ; 0:6805 po (1 ; xa2 )(1 ; ya2 ) xy
a2
. August 2007
If three terms are taken, then (again for a square plate (a = b))
a1 = 0:0404 apo6 a2 = a3 = 0:01174 ap8o
and the stress along the axis x = 0 is
2 2 2 4
xx = po(1 ; ya2 ) ; 0:1616 po (1 ; 3 ay2 ) + 0:0235 po (1 ; 12 ya2 + 15 ay4 )
These stresses satisfy equilibrium and boundary conditions, but not compatibility.
Relation between the Ritz and Finite Element Methods

At this stage, it is worthwhile to summarize some of the characteristics of the Ritz
and Galerkin methods. The most important ones are:
Usually, the accuracy of the assessed displacement is increased with an increase
in the number of trial functions.

While fairly accurate expressions for the displacements are obtained, the corre-
sponding forces may dier signicantly from the exact values.

Equilibrium is satised in an average sense through minimization of the total
potential energy. Therefore, forces (computed on the basis of the displacements)

do not, in general, satisfy the equilibrium equations.
The approximate system is stier than the actual system so that, for example,
buckling loads and vibration resonances are overestimated while displacements

are underestimated.
The question that arises is that of the appropriate additional terms to be used if
more terms are to be included so as to achieved a converged accurate solution the
main attributes required is that it complete and compatible. For example, for a 1-D
problem the simple polynomial
1 x x2 x3 : : :
is complete. The trigonometric sequence
1 sin x cos x sin 2x cos 2x sin 3x cos 3x : : :
is also complete. Note, however, that the cosines on their own could not represent an
asymmetric distribution.
As we go to higher dimensions, the question of completeness gets a little more
involved. In the 2-D case, there is the Pascal triangle
1
x y
x 2 xy y2
x3 x2 y xy2 y3
An analogous sequence can be written in terms of the trigonometric functions.
A number of observations can now be made about the use of stationary principles.
First, if the Ritz functional contains derivatives up to order m, then there must
be continuity of displacement derivative up to m ; 1, and the order of the highest
derivative that is present in the governing dierential equation is then 2m. For
example, in a beam bending problem where the strain is d2v=dx2, m = 2 because the
highest derivative in the functional is of order 2, and there must be continuity of v
and dv=dx. The reason for obtaining a derivative of order 2m = 4 in the governing
dierential equation is that integration by parts is employed m = 2 times. The
Galerkin functional has higher order derivatives.
A second observation is that through the stationarity condition we obtain the
governing dierential equations and the proper boundary conditions. Hence, the
eect of the natural boundary conditions are implicitly contained in the expression
for the potential *. (Note that the essential boundary conditions must be stated
separately.)
The classical or conventional Ritz method is the discretized implementation of the
stationary potential principle. Some of its most important characteristics are:
The trial functions usually span the entire domain.
Usually, the accuracy of the assessed displacement is increased with an increase
in the number of trial functions.

While fairly accurate expressions for the displacements are obtained, the corre-
sponding forces may dier signicantly from the exact values.

Equilibrium is satised in an average sense through minimization of the total
potential energy. Therefore, forces (computed on the basis of the displacements)

do not, in general, satisfy the equilibrium equations.
The approximate system is stier than the actual system.
One disadvantage of the classical Ritz analysis is that the trial functions are
dened over the whole region. This causes a particular diculty in the selection of
appropriate functions in order to solve accurately for large stress gradients, say, we
may need many functions. However, these functions are dened over the regions in
which the stresses vary rather slowly and where not many functions are required.
Another diculty arises when the total region is made up of subregions with dierent
kinds of strain distributions. As an example, consider a building modeled by plates
for the "oors and beams for the vertical frame. In this situation, the trial functions
used for one region (e.g., the "oor) are not appropriate for the other region (e.g., the
frame), and special displacement continuity conditions and boundary relations must
be introduced. We conclude that the conventional Ritz analysis is, in general, not
particularly computer-oriented.
We can view the nite element method as an application of the Ritz method, where
instead of the trial functions spanning the complete domain, the individual functions
span only subdomains (the nite elements) of the complete region. Figure 7.4 shows
. August 2007
Figure 7.4: Continuous domain discretized as nite elements.
an example of a bar with a hole modeled as a collection of many triangular regions.

The use of relatively many functions in regions of high strain gradients is made pos-
sible simply by using many elements as shown around the hole in the gure. The
combination of domains with dierent kinds of strain distributions may be achieved
by using dierent kinds of elements to idealize the domains.
In order that a nite element solution be a Ritz analysis, it must satisfy the
essential boundary conditions. This refers to the actual boundaries of the problems
and to inter-element compatibility.
However, in the selection of the displacement functions, no special attention need
be given to the natural boundary conditions, because these conditions are imposed
with the load vector and are satised approximately in the Ritz solution. The ac-
curacy with which these natural boundary conditions are satised depends on the
specic trial functions employed, and on the number of elements used to model the
problem.
7.3 Semi-Direct Methods

Another application of the variational approach is the conversion of one set of govern-
ing dierential equations into another, usually lower order, set. This is known as the
semi-direct method because it still yields a strong formulation of the problem. The
approach will be demonstrated with the construction of structural theories from the
3-D theory of elasticity.
Bending of Deep Beams

Consider a rectangular beam of length L, thickness h, and width b, as shown in
Figure 7.5. If b is small, then the beam can be regarded as in a state of plane stress.
We begin by expanding the displacements in the beam by a Taylor series about
the mid-plane displacements u(x 0) and v(x 0) however, since we are interested in
"exural deformations, we set u(x 0) = 0 and retain only odd powers of y for u(x y)
7.3. Semi-Direct Methods 251
y v0 0 6 6q (x)
6
vL L
V M
6
5
6
6
5
h
- x
L 6 b
M0 V0 ?
5 5
L L
Figure 7.5: Timoshenko beam with distributed and end loads.

and even powers of y for v(x y). Thus
u(x y) u(x 0) + y @@yu jy=0 + = ;y(x) + O(y3)
v(x y) v(x 0) + y @y @ v j + = v(x) + y2(x) + O(y4)
y=0
where we have used the notations

v(x) = v(x 0) (x) = ; @@yu jy=0 (x) = 12 @y @ 2 v j
2 y=0
This approximation says that the deformation is governed by three independent func-
tions, v(x), (x), and (x), that depend only on the position along the centerline.
We could, of course, retain more terms in the expansion and develop an even more
rened theory. In fact, we will use the lateral traction boundary conditions to reduce
the representation down to eectively only two functions.
The strains corresponding to the above deformation are
xx = @@xu = ;y @ @x +O(y3) yy = @ v = y2+O(y3) xy = @ u + @ v = ;+ @v +O(y2)
@y @y @x @x
The normal strains are predominantly linear on the cross-section, whereas the shear
strain is predominantly constant. Substitute these strains into the Hooke's law for
plane stress to get, for example,
h i h i
yy = 1 ;E 2 yy + xx = (1 ;E 2 ) 2 ; @
@x + O ( y 2) y
We expect the normal tractions to be zero on the lateral surfaces when there is no
distributed load, so choose
2 = @
@x or yy = ;xx
This gives the other normal stress as
h i h i
xx = 1 ;E 2 xx + yy = (1 ;E 2 ) ; @ @x + 2 @ + O(y2) y = ;Ey @ + O(y3)
@x @x
. August 2007
To the indicated order, the normal stresses form a uniaxial system of stresses. We
assume the presence of a distributed lateral load does not greatly aect this conclu-
sion.
The shear stress is h
xy = G ; + @x @v + O(y2)i
This is predominantly constant on the cross-section and nonzero on the boundary.
The lateral boundaries are also shear traction free so if we take
@v
= @x
this would impose that condition. It would also impose that there is no shear at all
in the beam. Instead we will interpret the constant shear as actually representing the
average shear as indicated in Figure 7.6.
6y
@@@@
z
xx xy
@
@
(a) @@ R
@ x
(b) (c)
Figure 7.6: Distributions of stress an the cross-section. (a) Arbitrary cross-section.
(b) Normal stress. (c) Shear stress.
We therefore conclude that the stresses are essentially
h @v i
xx = ;yE @
@x xy = G ; + @x others = 0
The strain energy now becomes
Z Z
U =
1 + ] dV = 1 E2 + G 2 ] dV
2 xx xx xy xy 2 xx xy
V V
Substitute for the strains to get the total strain energy as
Z Z Z
U =
1 L h=2 Ey 2( @ )2 + G( ; @v )2 b dydx = 1 L EI ( @ )2 + GA( ; @v )2 dx
2 o ; h=2 @x @x 2 o @x @x
where the cross-sectional properties
Z h=2 Z h=2
A bdy = bh I by2dy = 121 bh3
; h=2 ;h=2
were introduced. If the applied surface tractions and end loads on the beam are as
shown in Figure 7.6, then the potential of these loads is
ZL ZL L L
V = ; q(x)v dx ; ML L + M0 0 ; VLvL + V0v0 = ; q(x)v dx ; M0 ; V v0
o o
Our variational principle for the beam may now be stated as
(Z L L L )
1 @ 2 1 @v 2
EI ( @x ) + 2 GA(; + @x ) + qv dx + M0 +V v0 = 0
0 2
There are two entities, v(x) and (x), which are subject to variation.
Taking the variation inside the integrals and using integration by parts, we get
Z L @v @ h @ i Z L @ h @v i
GA( ; + @x ) + @x EI @x dx + GA(; + @x ) + q v dx
0 0 @x
h i L h @v ] ; V ivL= 0
+ EI @
@x ; M + GA ; +
0 @x 0
(7.1)
The fact that the variations v and can be varied separately and arbitrarily, and
that the limits on the integrals are also arbitrary, lead us to conclude from the terms
in square brackets that
@ hGA( @v ; )i = ;q
@x @x
@ EI @ + GA @v ; i = 0
h i h
(7.2)
@x @x @x
The associated boundary conditions (at each end of the beam) are specied in terms
of any pair of conditions selected from the following groups:
@v ; )
v or V = GA( @x or M = EI @
@x (7.3)
Thus a free boundary is specied as
V =0 M =0
An inadmissible set of boundary conditions are
v=0 V =0 or =0 M =0
If either of these are imposed then there is no guarantee that the remaining term in
Equation (7.1) is zero.
These are the Timoshenko equations for a deep beam. In comparison to the ele-
mentary Bernoulli-Euler beam theory, this theory accounts for the shear deformation
associated with deep beams.
. August 2007
If the surface tractions on the body are as shown, then
ZL ZL
V = bq+vdx ; bq (x)vdx
o o
;
Z h=2 Z h=2
+ bf1 (y)u(L y)dy ; bfo(y)u(o y)dy
h=2 h=2
Z h=2 Z h=2
; ;
+ bg1 (y)v(L y)dy ; bgo (y)v(o y)dy

;h=2 ; h=2
Substituting for u and v and integrating gives
ZL
V = p(x)vdx + Me(L) ; Mo (o) + Q1 v(L) ; qov(o)
o
where the various resultants are given by
ZL
p(x) = bq+ (x) ; q (x)] dx
o
;
Z h=2 Z h=2
Mo = bfo (y)ydy M1 = bf1 (y)ydy
h=2 h=2
Z h=2 Z h=2
; ;
Qo = bgo dy Q1 = bg1 dy
; h=2 ; h=2
In order to assure pure "exure we require that
Z h=2 Z h=2
bf1 (y)ydy = 0 and bfo(y)ydy = 0
; h=2 ; h=2
that is, the resultant forces in the x-direction must vanish.
In order to account for the truncation error of the expansion u ;y, a correction
coecient is often added to the expansion that is, u = ;y. The coecient
can be evaluated many ways: for a rectangular cross-section, it is usually taken to be
2/3 in static problems and 2 =12 for dynamic problems.
Example 7.7: Recover the Bernoulli-Euler beam equations from the previous
formulation.
In the above developments, make the further assumption that
@v
= @x or @v
u = ;y @x
then the shear strain is zero and the potential becomes
Z Lh @ 2v 2 i
* = = 1 EI ; q ( x )v ( x ) dx
o 2 @x2
The Euler equation becomes
@ 2 EI @ 2 v ; q(x) = 0
@x2 @x2
This is known as the Bernoulli-Euler beam equation. The expansion u ;y assures
that plane sections remain plane after deformation. The assumption = @v=@x
further requires that the plane section remains normal to the neutral axis. It is clear
that can be regarded as the rotation of the cross-section.
Twisting of Long Structural Members

Consider a long straight bar of constant cross-section in equilibrium under the action
of end torques. The lateral boundaries are therefore traction free and the predominant
response is an axial twisting. There are two cases of interest: one is when the cross-
section is solid, the other is when it is in the form of a thin-walled tube.
6z
@ ;
;
? @
t tT
@
y- 7!
;! 0 7!
;! L
@ --
@ T0 GJ (x) L L
@
R
@ x
Figure 7.7: Torsion member with arbitrary cross-section. The x-axis is along the
length.
I: Torsion of Solid Bars

Because the bar is slender, we begin by expanding the displacements in a Taylor
series (in terms of y and z) about the mid-point values. While the dominant action
is a rotation of the cross-section about the origin, there is also a warping of the
cross-section. This leads to the approximate displacements
u(x y z) u(y z) + v(x y z) ;xxz + w(x y z) xxy +
where x is the angle of twist per unit length and u(y z) is the warping.
The three normal strains are zero, the shear strains are
@ v = @u ; z
xy = @@yu + @x @y x xz =
@ u + @ w = @u + y
@z @x @z x yz =
@ v + @ w = 0
@z @y
We conclude that there are only two nonzero stresses given by
xy = xy = G @u
@y ; x z ] @u
xz = xz = G + x y ]
@z
Since there are only two non-zero stresses, we can represent them in the form of
a potential function. That is, let
xy = @ @z xz = ;
@
@y
. August 2007
We will refer to (y z) as the torsion stress function. The tractions are zero on the
boundary. Consider a small boundary wedge as shown in Figure 7.7, the resultant
force in the normal direction should be zero
Fn = 0 = Axy sin
+ Axz cos
= A @ dz ; A @ (; dy ) = A d
@z ds @y ds ds
where ds is a segment of the circumference and A is the area of the triangle. We
conclude from this that along the lateral boundary
d = 0 or (s) = constant
ds
For simply connected regions (no cut-outs) we can take the constant as zero.
The resultant torque on the cross-section is
Z Z Z @ @
Tx = dTx = ;xy z + xz y] dydz = ; @z z + @y y] dydz
A A A
Noting that
@ (z) = + z @ @ (y) = + y @
@z @z @y @y
we can divide the integral into two parts
Z Z @ (z) @ (y) Z Z
Tx = 2 dA ; @z + @y ] dydz = 2 dA ; d(z) dy + d(y) dz]
A A A A
The d(z) and d(y) terms integrate to give boundary values, but as already noted,
can be taken as zero on the boundary, and hence we conclude that the second
integral terms evaluates to zero. Thus the moment is given simply by
Z
Tx = 2 dA
A
We have thus reduced the torsion problem to nding a function (y z) that van-
ishes along the boundary. We will use our stationary principle to eect solutions for
arbitrary cross-sections.
The strain energy expression reduces to
Z Z h @ i Z h @ i
U =
1 + ] dV = 1 ( ) 2 +( @ )2 dA dx = L ( ) 2 +( @ )2 dA
2 V xy xy xz xz 2G V @z @y 2G A @z @y
The only applied loads are the end torques consider one end as xed then the other
end rotates an amount xL, consequently, the total potential of the problem is
L Z h @ @ i Z
2 2
* = 2G ( @z ) + ( @y ) dA ; xL 2 dA
A A
Our variational principle now becomes
Z h @ i
( @y )2 + ( @ @z )2 ; 4Gx dA = 0
A
There is just the single entity (y z) subject to variation. The strong formulation
reduces to the dierential equation
@ 2 + @ 2 = ;2G
x
@y2 @z2
subject to being zero on the boundary.
Suppose we nd a function (x y) that satises the zero boundary conditions,
then c1 (x y) also satises the boundary conditions. Substitute into the stationary
principle and minimize with respect to c1 to get
h Z @ 2 @ 2 Z i
2c1 ( @z ) + ( @y ) ] dA ; 4Gx] dA c1 = 0
A A
Solving for c1 and substituting into the strain energy gives
1 2
h Z i2 Z h @ 2 @ 2 i
U = GJx J 4 dA = ( @z ) + ( @y ) dA
2 A A
where GJ is called the torsional sti
ness per unit length. We will now establish this
quantity for some typical cross-sections.
Example 7.8: Evaluate the torsional stiness of a bar with a rectangular cross-
section.
6
z
-y
Figure 7.8: Rectangular cross-section.

We take a Ritz approach and expand in terms of bases functions that satisfy
the boundary condition. For example, we could take
X X
= y2 ; b2=4]z2 ; h2=4] m n cmn y
mzn
Because of symmetry, m and n, must be even. For illustrative purposes, we will just
do a single term expansion. Thus let
= c1 y2 ; b2 =4]z2 ; h2 =4]
. August 2007
Substituting into the potential energy and evaluating the integrals leads to
h 3 3 3 3i
c21 b90h b2 + h2 ] ; 4Gx c1 b36h = 0
Performing the variation on c1 leads to c1 = 5Gx =b2 + h2 ] and the function as
= b52 G x 2 2 2 2
+ h2 ] y ; b =4]z ; h =4]
At this stage everything is determined about the solution.
For example, the stresses are given by
xy = @ = 10Gx z y2 ; b2 =4]

@z b2 + h2 ] xz = ; @
@y = ; 10Gx y z 2 ; h2 =4]
b2 + h2 ]
The maximum stresses occur at the middle of the sides and not at the extremities
indeed, the stresses are zero at the extremities. The torque rotation relation is
Tx = GJx 5 b3 h3
J 18 b2 + h2
When cross-sections is square of side a, the torsional stiness is
5 a4 = 0:1389 a4
GJ = G 36 GJexact = 0:1406 a4
The comparison with the exact value 31] is remarkably good being dierent by only
-1%. The accuracy depends on the aspect ratio being -3% for h=b = 2, -14% for
h=b = 10, and -20% in the limit of h=b ! 1. The next example problem will get a
better estimate for large h=b.
Example 7.9: Estimate the stresses and torsional stiness for a cross-section
that is narrow.
...........
..........
..........
........... ...........
..........
..........
.......... ...........
........... ..........
...........
z@
6
@@@
@
@
- y@@ @ L
@@ @
Rx
@
@ @@ @
@@@ @
@
@ h
b
Figure 7.9: Bar with narrow rectangular cross-section.
Choose the torsion function as
= c1 z 2 ; h2 =4]
This gives zero values on the top and bottom but not the sides. However, in the
limit of large b and/or small h, the eects of the sides will be negligible. Substituting
into the potential energy and evaluating the integrals leads to
h 3 3 i
43 bh4 c21 + 4Gx 23 bh4 c1 = 0
Performing the variation on c1 leads to c1 = ;Gx and the function as
= ;xz2 ; h2 =4]
The stresses are given by
xy = ;2Gx z xz = 0
and the torque rotation relation is
J 31 bh3 Tx = GJx
The torsional stiness is exact in the limit of b=h ! 1 and only o by +6% when
b=h = 10.
An estimate of the torsional stiness properties of rolled sections such as angles,
channels and I-beams can be obtained simply by summing the stiness of each
segment. Thus, X X
GJ = i GJ = i G 31 bh3
If this is not adequate, then the section can be treated as a folded plate as discussed
in the next subsection.
II. Torsion of Tubes

The general problem of the torsion of a cross-section with multiple cut-out is very dif-
cult to solve. Even approximate methods such as that of Ritz are also dicult to use
because of the requirement to satisfy multiple boundary conditions simultaneously.
There is one situation that is relatively easy to solve and that is when the interior
boundaries coincide with a stress line. A special case of this is that of thin-walled
tubes, this is tractable which we now illustrate.
Sp p pop p p p p p p p pppppp p6 z z @
p p p p p p p p p
p p
p pp p p p p p p p p p p p p p p p p p s
p p n p p p p p p p p p p p p p p p
6
p p p p p p p p p p;p p;p p p p p p p@ s n;
pp p pppp pp pp pppp p p p p p p Sp p p pip p p p p p p p p p p p p p p p p p p p p p p p p p p@ I p p pp@pp pp pppppp;pp p p;

pp p p p p p p p p p p p p pp
;a(
p p
I
@ p p p@p p p p p;p
ppp ppp pp p pp -y p
p ;( (
( (
KA pp -
p p y
p p p p pp p p p p p p p p p p p p p p p

( A
p
p p p p p p p p pp pp pp p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p pp pp pp pp pp pppppp pp p p p p pppp p p p p pp p
p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p pAp A dA
ppppppppp pp p p p p p p p p p p pp pp pppppppp
Figure 7.10: Torsion of thin-walled tubes.

Consider the general thin-walled tube shown in Figure 7.10. Introduce a set of
local axes n and s also as shown. Previously it was established that on a free surface
. August 2007
the torsion function must satisfy @=@s = 0, from which we conclude that the function
is constant on the inner and outer surfaces. For solid sections we set the constant as
zero, for multiply connected sections we can set only one as zero, and the other is a
constant to be determined as part of the solution. Here we choose
= c1 on Si = 0 on So
Since the wall is thin, then expand the torsion function as
(n s) 0 (s) + 1 (s)n + 3 (s) 12 n2 +

Retaining just the linear terms and imposing the boundary values leads to
(n s) = c21 ch1 n
; 0 = c21 1 = ch1
;
where h is the local thickness of the tube.

The local (n s) axes can be thought of as a point transformation from the (y z)
axes from which we immediately get the stress relations
xn = @
@s =0 xs =
;
@ = c1
@n h
If the wall thickness is constant, we conclude that the shear stress is constant. The
strain energy for a tube segment of length L reduces to
Z I I
= L ( @ )2 dA = L h( @ )2 ds = L c21 ds
2G A @n 2G @n 2G h
U
The torque caused by the stress xs on an arc segment of length ds is

dTx = axsh ds = a ch1 h ds = ac1 ds
where a is the perpendicular distance from the origin to the line of action of xs. A
simple geometric construction based on triangles with base ds shows that a ds = 2 dA
from which we get that Z
Tx = 2c1 dA = 2c1A
A
where A is the area enclosed the centerline of the tube.
The total potential of our problem is
L I ds
*= U ;
2
TxL = 2G c1 h 2c1A ; xL
Minimizing with respect to c1 gives
I ds n I ds

c1 = 2GAx= h
(n s) = GAxL1 h=2 ]= h
;
We can now summarize the various relations
q I ds I ds
x = 2GA h Tx = GJx 2
J = 4A = h
where q xsh = c1 = constant, is called the shear ow.
Example 7.10: Compare the stiness properties of a thin-walled circular tube
with and without a longitudinal slit.
ppp p p p p p p p p p p p p pppp pppp p p p p p p p p p p p pppp
pp p pp pp p p p p p p p p p p p p p p p p p p p pp pp p p pp pp p pp pp p p p p p p p p p p p p p p p p p p p pp pp p p p p
ppppp ppppp pp pp ppppp ppppp ppp ppp
pp pp
pp p p p p p p p p pp pp p p p p p p pp p ppp
p
p p p p pppp p p p p p p p p p p p p p pppp p p p p
p p p p pppp p p p p p p p p p p p p p pppp p p p
ppppppppppp pp ppppppppppp pp
Figure 7.11: Thin-walled tubes.
Let the diameter be D. For the tube without a slit
I ds

J1 = 4A = h = 4(D2 =4)2 =(D=h) = 41 D3h
2
The stiness of the split tube can be estimated by thinking of it as a narrow rectangle.
Then,
J2 = 13 bh3 = 13 Dh3
The ratio of the two is
J1 = 3 D2
J 4 h2 2
A tube with D=h of 20 gives a ratio of 300, that is a signicant dierence.
Flat Plate Theory

Fundamentally, as discussed in Section 5.2 for plane stress, thin plate theory is an
approximate structural theory and therefore it is best to approach it by way of a
variational principle. We will begin by developing a plate theory (called Mindlin plate
theory) that takes the shear deformation into account | this is the plate equivalent
of the Timoshenko beam. The transition to achieve the classical or (very) thin-plate
theory is then more transparent.
I: General Governing Equations

Consider a rectangular plate of thickness h as shown in Figure 7.12. The plate lies in
the x-y plane and is subjected to both in-plane and transverse loads. The mid-plane
of the plate is taken at z = 0.
. August 2007
6z 6z
;
;
; ;
; -
; - y
;
;

* q(x y) ; ;
;
;
-
y
/ M
; ;
M.. Q V M MQ

; -;
5
; ; ; 6; ;
; ; ; ; 6; xy
;Nyy
; /
;
;
- ;
;
; yz
;
- ; -
- 66 5 ;
;
x
; Nxx
; ; Nxy
;
x
; ; yy
Nxy xx xz ( xz ) xy
Figure 7.12: Element of stressed plate.

Because the plate is thin, we begin by expanding the displacements in a Taylor
series (in terms of z) about the mid-plane values as
u(x y z) u(x y) zx (x y)
;
v(x y z) v(x y) zy (x y)

;
w(x y z) w(x y) (7.4)

where x and y are rotations of the subscripted faces in the directions of the cur-
vatures. These say that the deformation is governed by ve independent functions:
u(x y), v(x y) are the in-plane displacements w(x y) is the out-of-plane displace-
ment and x (x y), y (x y) are the rotations of the mid-plane. The normal and shear
strains corresponding to these deformations are
xx = @@xu = @x
@u ; z @x
@x
yy = @ v @v
= ;z y @
@y @y @y
@ u @ v
@u @v @x @y
xy = @y + @x = @y + @x ; z @y + @x

xz = @@zu + @@xw = ; x + @w
@x
@ v @ w
@w
yz = @z + @y = ; y + @y
Because the plate is thin, the stress in the z direction cannot be very large. A special
case that arises in the analysis of thin-walled structures is that of plane stress. Here,
the stress through the thickness of the plate is approximately zero such that zz 0,
xz 0, and yz 0. This leads to
zz = E xx + yy ] = 1 xx + yy ]

; ;
;
Substituting this into the 3-D Hooke's law then gives
xx = E1 xx ; yy ] xx = (1 ;E 2 ) xx + yy ]
yy = E1 yy ; xx] yy = (1 ;E 2 ) yy + xx ] (7.5)
The shear relation is unaected. Substituting for the strains in the Hooke's law then
leads to
h @u @v @x @y i
xx = 1 ;E 2 xx + yy ] = 1 ;E 2 @x + @y ; z @x + @y
h @v @u @y i
yy = 1 ;E 2 yy + xx] = 1 ;E 2 @y +
@x ; z
@y + @x
@x
zz = 0 (7.6)
h @u @v @x @y i
xy = Gxy = G @y + @x ; z @y + @x
h i h @w i
xz = Gxz = G ; x + @w @x
yz = G yz = G ; y + @y
Although the plate is treated as being in plane stress, we still retain the xz and yz
shear stresses.
The strain energy for the plate is
Z
U =
1 xxxx + yy yy + xy xy + xz xz + yz yz ] dV
2 V
Substitute for the stresses and strains and integrate with respect to the thickness to
get the total strain energy as
Z h i
U =
1 D ( @x + @y )2; 1 (1 ; )4 @x @y ; ( @y + @x )2 ] dxdy
2 A @x @y 2 @x @y @x @y
Z h i
+ 12 Gh (x ; @w @x ) 2 + ( ; @w )2 dx dy
y
@y (7.7)
A
Z h h
+ 12 E h ( @u

) 2 + ( @v )2 + 2 @u @v i + Gh( @u + @v )2 idx dy
A @x @y @x @y @y @x
where D Eh3 =12(1 ; 2 ) is called the plate bending sti
ness and E E=(1 ; 2 ).

If the applied surface tractions and loads on the plate are as shown in Figure 7.12,
then the potential of these loads is
Z
V = ; qw (x y)w dx dy ; Nxxu ; Nxy v ; Mxxx ; Mxy y ; Vxz w +
where the edge loads can be on each face. The energies de-couple into in-plane (u
and v) and out-of-plane (w, x and y ) sets hence, we now nd it convenient to treat
them separately.
. August 2007
II: In-Plane Membrane Behavior
The energy and potential for the in-plane behavior are
Z " h @u 2 @v 2 @u @v i @u @v 2 #
1
U = 2 E h ( @x + ( @y + 2 @x @y + Gh @y + @x dx dy

A
V = ;Nxxu ; Nxy v + (7.8)
Application of our variational principle with variations in u and v leads to two
dierential equations
Eh r2 u ; 1 (1 + ) @ 2 u ; @ 2 v = 0 (7.9)
1 ; 2 2 @y2 @x@y
Eh r2 v ; 1 (1 + ) @ 2 v ; @ 2 u = 0 2 @2 + @2
1 ; 2 2 @x2 @x@y r
@x2 @y2
These are the Navier's equations. as used in Section 5.5 the derivation here, however,
shows that they are not restricted to very thin plates.
For the associated boundary conditions, we specify one condition from either set:
h @u @v i h @u @v i
u or Nxx = 1 Eh ;
2 @x + @y v or N xy = Gh @y + @x
We can give interpretation of the boundary conditions in terms of resultants on the
cross-section. For example, the resultants of the normal and shear stresses are dened
on the cross-section as
Z Z
Nxx(x y) xx(x y z) dz Nxy (x y) xy (x y z) dz
and leads to
h i
Nxx = (1 Eh 2 ) @u
;
+ @v = xxh
@x @y
h i
Nyy = (1 Eh 2 ) @v
;
+ @u = yy h
@y @x
h @u + @v i = h
Nxy = 2(1Eh + ) @y @x xy (7.10)
That is, Nxx and so on, are the resultant forces per unit length due to the stresses
acting on the edge faces.
Example 7.11: Specialize the Navier's equation to the case where there is only
an x dependence.
There are no derivatives with respect to y. The rst of the two Navier's equations
becomes
Eh @ 2 u = 0 or E h @ 2 u = 0

1 ; 2 @x2 @x2
which is the one-dimensional rod equation. The second of the Navier's equations
becomes
Eh @ 2 v ; 1 (1 + ) @ 2 v = 0 or @2v = 0
Gh @x
1 ; 2 @x2 2 @x2 2
This is also a one-dimensional equation but it is for the transverse shear behavior.
This is not the transverse exural shear behavior.
For the associated boundary conditions, we specify one condition from either
set:
Eh @u @v
u or Nxx = 1 ; 2 @x v or Nxy = Gh @x
These mimic the boundary conditions for the elementary theory.
III: Out-of-Plane Flexural Behavior
The energy and potential associated with the out-of-plane behavior are
Z " @x @y 2 h @x @y @y @x 2 i#
1 1
U =
2 AD @x + @y ; 2 (1 ; ) 4 @x @y ; @x + @y dxdy
Z h 2 @w 2idx dy
+ 21 Gh (x ; @w @x + ( y;
@y
ZA
V = ; qw (x y)w dx dy ; Mxxx ; Mxy y ; Vxz w + (7.11)
An application of our variational principle with the variations of w x and y ,
leads to, respectively,
qw + Gh @x @ h @w ; i + Gh @ h @w ; i = 0
@x x @y @y y
h i h
1 D (1 ; )r2 + (1 + ) @ ( @x + @y ) + Gh @w ; = 0
i
x
2 @x @x @y @x x
h i h
1 D (1 ; )r2 + (1 + ) @ ( @x + @y ) + Gh @w ; = 0
i
2 y y
@y @x @y @y
(7.12)
These are the governing equations for the Mindlin plate this theory accounts for
the shear deformation. The associated boundary conditions (on each edge face of
the plate) are specied in terms of any three conditions selected from the following
groups:
h i
w or Vxz = Gh @w ; x
"@x #
x or Mxx = D @x + @y @ x @ y
" #
1
y or Mxy = 2 (1 ; )D @y + @x @x @y
. August 2007
These are specied for an x-face, the other faces are similar.
We can give interpretation to the boundary conditions in terms of resultants of
the stresses on the cross-section. For example, taking resultants for the shear stress
dened as
Z Z h i
Qxz (x y) xz (x y z) dz = G ; x + @w @x dz
leads to
h i h @w i = V
Qxz = Gh ; x + @w @x = V xz Q yz = Gh ; y +
@y yz (7.13)
We can also take a moment due to the stresses acting on the edge faces. For example,
Z 3 h x @y i
Mxx ; xxz dz = 12(1Eh; 2 ) @ @x + @y
and all resultants can be written as
h x @y i
Mxx = D @ @x + @y
h y @x i
Myy = D @ @y + @x
h @y @x i
2Mxy = D @x + @y (1 ; ) (7.14)
These moment resultants are related only to the rotations.
In order to account for the truncation error of the expansions u and v, we could
add correction coecients to the energies as was done with the Timoshenko beam
theory.
IV: Flexural Behavior of Very Thin Plates
The plate theory derived here (called classical plate theory) is the 2-D equivalent of
the Bernoulli-Euler beam theory. Rather than go directly to the governing equations,
we will retrace the developments of the Mindlin plate, but with the assumptions of
the classical theory.
We modify the Mindlin equations to the thin-plate theory in two steps. First, we
assume that the transverse shear deformation is negligible this is equivalent to saying
that the shear stiness in the transverse direction is innite. This leads to
@w ; = 0 @w ; = 0
x
@x @y y
It is important to realize that while these combinations are zero, their product with Gh
is nonzero (because it is related to the transverse shear resultant). The displacements
for the "exural motion are approximated as
u(x y z) ;z @w
@x (x y ) v
( x y z ) ;z
@w (x y) w(x y z) w(x y)
@y
The normal and shear strains corresponding to these deformations are
xx = @x @ u = ;z @ 2 w
= @ v = ;z @ 2 w
yy
@x2 @y @y2
xy = @@yu + @x @ v = ;2z @ 2 w
@x@y
xz = @@zu + @@xw = 0 yz = @@zv + @@yw = 0
We reiterate that, although the transverse shear strains are zero, the transverse shear
forces are nonzero. Also note that there is an in-plane shear that depends on the
distance from the midplane | there is no comparable quantity in beam theories.
Substituting the above strains into the Hooke's law for plane stress gives
h 2 2 i
xx = 1;;Ez 2 @@xw2 + @@yw2
h 2 2 i
yy = 1;;Ez 2 @@yw2 + @@xw2
h @2w i
xy = ;2Gz @x@y (7.15)
The strain energy for a plate in plane stress is
Z
U =
1 xxxx + yy yy + xy xy ] dV
2
V
Substitute for the stresses and strains and integrate with respect to the thickness to
get the total strain energy as
Z h @ 2 w )2 ; @ 2 w @ 2 w ]idx dy
U = 2 D (r2 w)2 + 2(1 ; )( @x@y
1
@x2 @y2
The potential of the applied loads is
Z
V = ; qw (x y)w dx dy ; Mxx @w @x ; Vxz w +
where the edge loads are on each face. Using variational principle with the variation
of only w then leads to the governing equation
Dr2 r2w = q (7.16)
Performing the integration by parts required to get the boundary conditions is
rather involved for an arbitrary boundary | a detailed description is given in Refer-
ence 27]. The associated boundary conditions are found to be
@3w @ 3 w
w or Vxz = ;D @x3 + (2 ; ) @x@y2
@w @ 2 w @ 2 w
@x or Mxx = D @x2 + @y2 (7.17)
. August 2007
The shear to be specied is called the Kirchho
shear. This shear is not the resultant
Qxz but is actually given by
Kirchho shear: Vxz = Qxz ; @M @y
xy
This can be understood physically by realizing that the shear moment Mxy can be
interpreted as a couple comprised of vertical forces a small distance apart. Then,
because the moment is distributed, so too are the vertical forces, which consequently
at any given location will have an imbalance in the vertical forces. Alternatively, the
classical plate theory has restrictive degrees of freedom, where the shear strains xz
and yz are zero. That is, the shear resultants Qxz and Qyz do not have a relationship
to the corresponding deformation. While this can be rationalized in the constitutive
relation by saying that the shear modulus in the transverse direction is very large,
it means that the resultant force is associated with higher-order derivatives of the
deformation.
The resultants can be written as
h 2 2 i
Mxx = D xx + yy ] = D @@xw2 + @@yw2
h 2 2 i
Myy = D yy + xx] = D @@yw2 + @@xw2
Mxy = Myx = D(1 ; )xy = D(1 ; ) @x@y @2w (7.18)
These resultants are related only to the out-of-plane de"ection. The stresses are
obtained from equations such as
xx = ; MIxxz
p
with Ip h3=12.
Example 7.12: Specialize the thin-plate exural equations when there is no y

dependence.
There are no derivatives with respect to y, and the summary of plate equations
becomes
Displacement : w = w(x t)
Slope : x = @w @x
2
Moment : Mxx = +D @@xw2
3
Shear : Vxz = ;D @@xw3
4
Loading : q = D @@xw4 (7.19)
These are the equations for a beam if we make the associations
D () EI or E=(1 ; 2 ) () Eb
A plate deforming as assumed here is called cylindrical bending.
. August 2007
Exercises
7.1 Show that the variational problem
Z 10 du
J u ] = 12 ( dx )2 ; 100u] dx = min
0
corresponds to the following dierential equation
d2 u + 100 = 0 0 x 10
dx2
subject to the boundary conditions u(0) = u(10) = 0.
7.2 Obtain a Ritz solution to the previous exercise using the trial function f (x) =
x(10 ; x). Compare this approximate solution with the exact solution.
7.3 In general, it is not possible to obtain a functional for problems whose governing
dierential equation contains odd-power derivatives. A special exception is the
following case
d2 u + a du + bu = 0
dx2 dx
where a and b are constants. Show that the variational problem
Z
J (u) = 12 eax ( du 2
dx ) ; bu] dx = min
corresponds to the dierential equation.
7.4 Suppose we wish to derive a \higher-order" rod element and to that end we
take the deection function in the cubic form
u(x) = a0 + a1 x + a2 x2 + a3 x3
Introducing the nodal degree of freedoms u1 = u(0), 1 = du(0)=dx, u2 = u(L),
2 = du(L)=dx, show that the displacement can be represented in terms of the
nodal values as
u(x) = g1 (x)u1 + g2 (x)L1 + g3 (x)u2 + g4 (x)L2
where the functions gn (x) are identical to those for the beam shape functions.
Show that the higher-order rod element stiness matrix is
2 36 3L ;36 3L 3
EA 66 3L 4L2 ;3L ;L2
k ] = 30
77
L 4 ;36 ;3L2 36 ;3L2 5
3L ;L ;3L 4L
Note that this is almost identical to the geometric stiness matrix for the beam
Why?
Exercises 271
7.5 Suppose that only the axial forces F1 F2 are taken as the nodal loads in the
previous exercise, show that the elementary rod stiness relation is recovered.
Propose some nodal loads that do not give the trivial result.
7.6 With reference to Figure 7.1, use the Ritz method to nd an approximate
solution using
u(x) = a0 + a1 x
7.7 Consider a cantilever beam, xed at the end x = 0 and subjected to a given
displacement vL = c at the other. Show that the following is a set of admissible
displacements and obtain the corresponding Ritz solution.
2 X
v(x) = cx
L2 + an 1 ; cos(2nx=L)]
7.8 Consider a cantilever beam, xed at the end x = 0 and subjected to a concen-
trated lateral applied force at the other. Using the Ritz method, show that the
following displacements
v(x) = a0 + a1 x + a2 x2 + a3 x3
leads to the exact solution. Show that the addition of extra terms have zero
contributions.
7.9 A uniformly loaded beam is simply supported at both ends. Find the deection
and bending moment at the center using the Ritz method. First use a quadratic
function in x and then use a sine function in x. Compare the results with the
exact solution and say why the second solution is better than the rst.
. August 2007
References
1] Abramowitz, M. and Stegun, I. A., Handbook of Mathematical Functions, Dover,
New York, 1965.
2] Allman, D.J., \On the General Theory of the Stability of Equilibrium of Discrete
Conservative Systems," Aeronautical Journal, 27, pp. 29{35, 1989.
3] Argyris, J.H. and Kelsey, S., Energy Theorems and Structural Analysis, Butter-
worths, London, 1960.
4] Bathe, K.-J., Finite Element Procedures in Engineering Analysis, Prentice-Hall, En-
glewood Clis, NJ, 1982, 2/E 1996.
5] Bathe, K.-J., Finite Element Procedures in Engineering Analysis, Prentice-Hall, En-
glewood Clis, NJ, 1982.
6] Criseld, M.A., Nonlinear Finite Element Analysis of Solids and Structures, Vol 1:
Essentials, Wiley & Sons, New York, 1991.
7] Criseld, M.A., Nonlinear Finite Element Analysis of Solids and Structures, Vol 2:
Advanced Topics, Wiley & Sons, New York, 1997.
8] Dhondt, G., The Finite Element Method for Three-Dimensional Thermomechanical
Applications, Wiley & Sons, Chichester, England, 2004.
9] Doyle, J.F., Wave Propagation in Structures, Springer-Verlag, New York, 1989, 2/E
1997.
10] Doyle, J.F., Static and Dynamic Analysis of Structures, Kluwer, The Netherlands,
1991.
11] Doyle, J.F., Nonlinear Analysis of Thin-walled Structures: Statics, Dynamics, and
Stability, Springer-Verlag, New York, 2001.
12] Eriksson,A., Pacoste, C. and Zdunek, A., \Numerical Analysis of Complex Insta-
bility Behavior using Incremental-Iterative Strategies," Computer Methods in Applied
Mechanics and Engineering, 179, pp. 265{305, 1999.
13] Guz, A.N., Makhort, F.G., Gushcha, O.I. and Lebedev, V.K., \Theory of
Wave Propagation in an Elastic Isotropic Body with Initial Deformations," Soviet
Applied Mechanics, 6, pp. 1308{1313, 1970.
14] Hamilton, W.R., The Mathematical Papers of Sir W.R. Hamilton, Cambridge Uni-
versity Press, Cambridge, 1940.
15] Hildebrand, F.B., Methods of Applied Mathematics, Prentice-Hall, Englewood Clis,
NJ, 1965.
16] Ince, E.L., Ordinary Dierential Equations, Dover, New York, 1956.
17] Jones, R.M., Mechanics of Composite Materials, McGraw-Hill, New York, 1975.
18] Lanczos, C., The Variational Principles of Mechanics, University of Toronto Press,
Toronto, 1966.
19] Langhaar, H.L., Energy Methods in Applied Mechanics, Wiley & Sons, New York,
1962.
273
274 References
20] Leipholz, H., Stability Theory, an Introduction to the Stability of Dynamic Systems
and Rigid Bodies, 2/E, John Wiley & Sons and B.G. Teubner, Stuttgart, 1987.
21] Love, A.E., A Treatise on the Mathematical Theory of Elasticity, Dover, New York,
1944.
22] Malvern, L.E., Introduction to the Mechanics of a Continuous Medium, Prentice-
Hall, New Jersey, 1969.
23] Megson, T.H.G., Aircraft Structures, Halsted Press, New York, 1990.
24] Oden, J.T., Mechanics of Elastic Structures, McGraw-Hill, New York, 1967.
25] Press, W.H., Flannery, B.P., Teukolsky, S.A. and Vetterling, W.T., Numer-
ical Recipes, Cambridge University Press, Cambridge, 1986, 2/E 1992.
26] Rooke, D.P. and Cartwright, D.J., Compendium of Stress Intensity Factors, Her
Majesty's Stationary Oce, London, 1976.
27] Shames, I.H. and Dym, C.L., Energy and Finite Element Methods in Structural
Analysis, Hemisphere, Washington, 1985.
28] Thompson, J.M.T. and Hunt, G.W., A General Theory for Elastic Stability, Wiley
& Sons, London, 1973.
29] Thompson, J.M.T. and Hunt, G.W., Elastic Stability, Wiley & Sons, London,
1993.
30] Timoshenko, S.P. and Gere, J.M., Theory of Elastic Stability, McGraw-Hill, New
York, 1963.
31] Timoshenko, S.P. and Goodier, J.N., Theory of Elasticity, McGraw-Hill, New
York, 1970.
32] Timoshenko, S.P. and Woinowsky-Krieger, S., Theory of Plates and Shells,
McGraw-Hill, New York, 1968.
33] Ulrich, K., State of the Art in Numerical Methods for Continuation and Bifurcation
Problems with Applications in Continuum Mechanics | A Survey and Comparative
Study, Laboratorio Nacional de Comutacao, Brazil, 1988.
34] Whittaker, E.T. and Watson, G.N., A Course of Modern Analysis, Cambridge
University Press, Cambridge, 1927.
35] Ziegler, H., Principles of Structural Stability, Ginn and Company, Massachusetts,
1968.
Index
Engineering shear strain 106
Agent 223 Engineering stress 65
Antisymmetric tensor 9 Equilibrium 215
Arc-length method 226 Essential boundary condition 237
Assemblage 193 Eulerian variable 16
Extension 30, 31
Base vector 3
Beam, Timoshenko 253 Flexure 265, 266, 268
Bifurcation 218, 221, 223, 225 Force, generalized 82
Boundary condition 264, 265, 267 Frame indierent 84
Boundary conditions, natural 237 Functional 231
Buckling 218, 227 Fundamental path 226
Bulk modulus 91
Galerkin method 248
Cauchy-Green tensor 28, 85 Gauss's theorem 8
Cauchy stress 83, 87, 187, 213 Generalized plane stress 126
Cauchy 211, 212 Geometric boundary condition 237
Compatibility 114, 182, 188, 248 Geometric stiness 196
Completeness 248
Complex conjugate 155 Homogeneous deformation 44
Conservation of energy 190 Hooke's law 89, 251, 263
Conservative system 85, 187 Hyperelastic material 85
Constitutive relation 83
Contraction of indices 2 Imperfection 223, 226
Coordinates 89, 176, 177, 190 Incompressibility condition 90
Cylindrical bending 269 Index, dummy 2
Index, free 2
Deformation 17 Index, subscript 1
Degree of freedom 188, 189, 268 Instability 87, 221, 223
Deviatoric deformation 99 Integral theorem 8
Direction cosines 5 Interpolation functions 176
Discrete systems 189 Invariant 10, 93
Isochoric deformation 99
Eigenanalysis 222, 223 Isotropic tensor 7
Eigenvalue 10
Eigenvector 10 Jacobian 19, 25, 178, 181
Elastic material 87, 89, 251, 263
Elastic stiness 204 Kirchho shear 268
Energy, kinetic 267 Kirchho stress 64, 83, 187, 211
Energy, potential 188, 189, 217, 223 Kronecker delta 3
Energy, strain 93, 106, 187, 189, 217, 263, 267
Energy 265 Lagrange stress 63
275
276 Index
Lagrangian strain 43, 83, 87, 187 Stationary potential energy 188
Lagrangian variable 16 Strain 83, 90, 106, 187, 262, 267
Lame constants 89, 91, 100, 101, 211 Stress concentration 140
Lame solution 140 Stress function 125
Laplace's equation 156, 235 Stress 83, 88, 92, 187, 212, 268
Limit point 87, 220, 225, 226 Stretch 45
Loading equation 202, 204, 225 Strong formulation 184
Symmetric tensor 9
Mode shape 227
Mooney-Rivlin material 95 Tangent stiness 92, 196, 203, 209, 218, 222,
Motion 15 223, 227
Multiply connected 139, 161, 260 Taylor series 88, 203, 205
Tensor eld 6
Navier's equations 157, 264 Thermoelasticity 90
Neo-Hookean material 95 Thin-walled structure 25, 41, 90, 262
Newton-Raphson iteration 206, 209, 212 Torsion stress function 256
Nonconservative system 187 Torsional stiness 257
Total stiness 196, 225
Objective tensors 84 Traction 52, 212, 263
Transversely isotropic 89
Objectivity 84
Orthogonality 5, 228 Trial functions 239
Orthotropic material 88, 89, 93 True stress 65
Truss 201, 203
Pascal triangle 248 Variation 232
Permutation symbol 3
Picard iteration 191 Variational principle 186
Plane strain 122 Vector eld 6
Plane stress 90, 124, 262 Vibration 227
Plate bending stiness 263 Virtual displacement 185
Plate 25, 29, 90, 262, 263, 265, 266 Virtual work 185, 186, 187, 193
Poisson's ratio 89
Positive denite 218, 225 Wave 93
Principal direction 10, 59
Principal value 9, 10, 59 Young's modulus 89
Proportional loading 221, 224
Quadrature 197
Ritz method 239, 248, 249
Scalar 6
Shape function 192
Shear ow 261
Shear 90, 210, 212, 263, 265, 266
Simple extension 45, 96
Simply connected 161
Smoothness 232
Snap-through 220
Spectral analysis 223
St. Venant's principle 155
St. Venant-Kirchho law 89

AAE 553 Text PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

AAE 553 Text PDF

Uploaded by

Copyright:

Available Formats

. .

Copyright(c) 1986{2003, 2005, 2006, 2007, 2008

1.1 Indicial Notation

Determinant detaij ] = ijk a1i a2j a3k

1.2 Coordinate Transformation

Vj e^j e^i = Vj e^j e^i

which yields, on using the properties of the Kronecker delta,

This can be further rewritten by introducing the matrix of direction cosines

These two equations taken together yields the circular result

from which we conclude that

e^2 = 12 e^1 ; 21 e^2 ; 12 e^3

cosines are easily established as

sian coordinates related by the transformation law

A system is called a scalar if it has only a single component  in the variables xi

at the corresponding points. That is,

related by the transformation law

tijk = minj pk tmnp0

A collection of particular isotropic tensors of dierent order is:

Calculus of Tensor Fields

vj (^x ) = jk vk (^x)

Dierentiating both sides of the equation, we obtain

the Integral theorem states

1.4 Properties of Second Order Tensors

Principal Values of Symmetric Tensors

Example 1.3: Prove that the principal directions are orthogonal.

Then the transformation matrix is given by

or in expanded matrix form

1.6 Show that det ij ] = 1.

2.1 General Description of Deformations

common global axes Po PP u

x2 = 21 (xo1 + xo2 )et ; 12 (xo1 ; xo2 )e t

the displacement vector is

u2 = x2 ; xo2 =  21 (xo1 + xo2 )et ; 12 (xo1 ; xo2 )e t ] ; xo2

xo2 = 21 (x1 + x2 )e t ; 12 (x1 ; x2 )et

u2 = x2 ; xo2 = x2 ;  12 (x1 + x2 )e t ; 21 (x1 ; x2 )et ]

Thus, the displacement can be written either as u = u(xoi ) or u = u(xi ) and we

conguration. The positions of the two points are related through

@xi dxo + 1 @ 2 xi dxodxo + : : :

J o det @x@xi ] = @x1 @x2 @x3

It is necessary to always check this to see if the deformation is physically possible.

Example 2.2: A particular deformation is described by

2.2 Deformation of Lines, Areas, and Volumes

Figure 2.4: Deformation of lines.

or, on substituting for the vectors,

In component form, this is

Example 2.4: Consider the following plane inhomogeneous deformation

ppp p pp p pp p p x pxp pp p pp a ppp

Figure 2.7: Shape before and after deformation.

Elementary Strain Measures

Tensorial Strain Measures

2Eij ]   @xmo ] @xmo ]T ; d I  @xm ] @xm ]T

Substituting the latter of these gives

Interchanging subscripts we conclude that

which is the transformation law for second order tensors.

Principal Strains and Invariants

and combining with the above yields

Thus, the extension E1 is related to the Lagrangian strain component E11 as

When E11 is small, that is, when E11

Realizing that dxo1 = dSo cos

Physical Interpretation of Shear Strains

@xi @xi dxo dxo = 2E + ]dxo dxo

new coordinate system obtained by a rotation

Physical Interpretation of Eulerian Strains

Determinant detaij ] = ijk a1i a2j a3k

A system is called a scalar if it has only a single component in the variables xi

A collection of particular isotropic tensors of dierent order is:

Dierentiating both sides of the equation, we obtain

1.6 Show that det ij ] = 1.

u2 = x2 ; xo2 = 21 (xo1 + xo2 )et ; 12 (xo1 ; xo2 )e t ] ; xo2

u2 = x2 ; xo2 = x2 ; 12 (x1 + x2 )e t ; 21 (x1 ; x2 )et ]

conguration. The positions of the two points are related through

J o det @x@xi ] = @x1 @x2 @x3

2Eij ] @xmo ] @xmo ]T ; d I @xm ] @xm ]T

@xi @xi dxo dxo = 2E + ]dxo dxo

common global axes 3 P

ij = 0 and !ij = 0.

x2 = xo1 sin + xo2 cos

u2 = xo1 sin + xo2 (cos 1) ;

where n^ is perpendicular to n^. The magnitude of the shearing stress, , is given by

2 = ( n^ ) ( n^ ) = (t^ ; N n^ ) (t^ ; N n^) = jt^j2 ; N2 = ti ti ; N2

It is obvious from this that = 0 on the surfaces with

since n21 + n23 = 1. The extremum value of occurs at

Noting that =o = J and @J=@xop = 0 (see Equation(2.5)), we obtain

This rules out, for example, plasticity eects.

rectly aect the material parameters.

Eyy = E1 yyK ; (zzK + xx

Ezz = E1 zzK ; (xxK + K )]

zzK = (1 + )(1 E (1 ; )Ezz + (Exx + Eyy )] (4.8)