# FEAP - - A Finite Element Analysis Program

Version 7.5 Theory Manual
Robert L. Taylor
Department of Civil and Environmental Engineering
University of California at Berkeley
Berkeley, California 94720-1710
E-Mail: rlt@ce.berkeley.edu
November 2003
Contents
1 Introduction 1
2 Introduction to Strong and Weak Forms 3
2.1 Strong form for problems in engineering . . . . . . . . . . . . . . . . . 3
2.2 Construction of a weak form . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Heat conduction problem: Strong form . . . . . . . . . . . . . . . . . . 5
2.4 Heat conduction problem: Weak form . . . . . . . . . . . . . . . . . . . 7
2.5 Approximate solutions: The ﬁnite element method . . . . . . . . . . . . 8
2.6 Implementation of elements into FEAP . . . . . . . . . . . . . . . . . . 12
3 Introduction to Variational Theorems 16
3.1 Derivatives of functionals: The variation . . . . . . . . . . . . . . . . . 16
3.2 Symmetry of inner products . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Variational notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4 Small Deformation: Linear Elasticity 21
4.1 Constitutive Equations for Linear Elasticity . . . . . . . . . . . . . . . 23
5 Variational Theorems: Linear Elasticity 26
5.1 Hu-Washizu Variational Theorem . . . . . . . . . . . . . . . . . . . . . 26
5.2 Hellinger-Reissner Variational Theorem . . . . . . . . . . . . . . . . . . 28
5.3 Minimum Potential Energy Theorem . . . . . . . . . . . . . . . . . . . 29
i
CONTENTS ii
6 Displacement Methods 31
6.1 External Force Computation . . . . . . . . . . . . . . . . . . . . . . . . 32
6.2 Internal Force Computation . . . . . . . . . . . . . . . . . . . . . . . . 32
6.3 Split into Deviatoric and Spherical Parts . . . . . . . . . . . . . . . . . 34
6.4 Internal Force - Deviatoric and Volumetric Parts . . . . . . . . . . . . . 37
6.5 Constitutive Equations for Isotropic Linear Elasticity . . . . . . . . . . 37
6.6 Stiﬀness for Displacement Formulation . . . . . . . . . . . . . . . . . . 40
6.7 Numerical Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7 Mixed Finite Element Methods 45
7.1 Solutions using the Hu-Washizu Variational Theorem . . . . . . . . . . 45
7.2 Finite Element Solution for Mixed Formulation . . . . . . . . . . . . . 50
7.3 Mixed Solutions for Anisotropic Linear Elastic Materials . . . . . . . . 52
7.4 Hu-Washizu Variational Theorem: General Problems . . . . . . . . . . 56
7.4.1 Example: Interpolations linear for u and constant φ . . . . . . . 60
8 Enhanced Strain Mixed Method 62
8.1 Hu-Washizu Variational Theorem for Linear Elasticity . . . . . . . . . 62
8.2 Stresses in the Enhanced Method . . . . . . . . . . . . . . . . . . . . . 66
8.3 Construction of Enhanced Modes . . . . . . . . . . . . . . . . . . . . . 67
8.4 Non-Linear Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
8.5 Solution Strategy: Newton’s Method . . . . . . . . . . . . . . . . . . . 69
9 Linear Viscoelasticity 74
9.1 Isotropic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
10 Plasticity Type Formulations 80
10.1 Plasticity Constitutive Equations . . . . . . . . . . . . . . . . . . . . . 80
10.2 Solution Algorithm for the Constitutive Equations . . . . . . . . . . . . 82
CONTENTS iii
10.3 Isotropic plasticity: J
2
Model . . . . . . . . . . . . . . . . . . . . . . . 85
10.4 Isotropic viscoplasticity: J
2
model . . . . . . . . . . . . . . . . . . . . . 91
11 Augmented Lagrangian Formulations 93
11.1 Constraint Equations - Introduction . . . . . . . . . . . . . . . . . . . . 93
11.2 Mixed Penalty Methods for Constraints . . . . . . . . . . . . . . . . . . 94
11.3 Augmented Lagrangian Method for Constraints . . . . . . . . . . . . . 97
12 Transient Analysis 100
12.1 Adding the transient terms . . . . . . . . . . . . . . . . . . . . . . . . . 100
12.2 Newmark Solution of Momentum Equations . . . . . . . . . . . . . . . 101
12.3 Hilber-Hughes-Taylor (HHT) Algorithm . . . . . . . . . . . . . . . . . 104
13 Finite Deformation 105
13.1 Kinematics and Deformation . . . . . . . . . . . . . . . . . . . . . . . . 105
13.2 Stress and Traction Measures . . . . . . . . . . . . . . . . . . . . . . . 108
13.3 Balance of Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
13.4 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
13.5 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
13.6 Material Constitution - Finite Elasticity . . . . . . . . . . . . . . . . . 112
13.7 Variational Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
13.8 Linearized Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
13.9 Element Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
13.10Consistent and Lumped Mass Matrices . . . . . . . . . . . . . . . . . . 118
13.11Stress Divergence Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 119
13.12Geometric stiﬀness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
13.13Material tangent matrix - standard B matrix formulation . . . . . . . . 120
13.14Loading terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
CONTENTS iv
13.15Basic ﬁnite element formulation . . . . . . . . . . . . . . . . . . . . . . 121
13.16Mixed formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
A Heat Transfer Element 128
B Solid Elements 136
B.1 Displacement elements . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
C Structural Elements 138
C.1 Truss elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
C.2 Frame elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
C.2.1 Small displacement element . . . . . . . . . . . . . . . . . . . . 138
C.3 Plate elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
C.4 Shell elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
D Isoparametric Shape Functions for Elements 140
D.1 Conventional Representation . . . . . . . . . . . . . . . . . . . . . . . . 140
D.2 Alternative Representation in Two Dimensions . . . . . . . . . . . . . . 142
D.3 Derivatives of Alternative Formulation . . . . . . . . . . . . . . . . . . 144
E Properties for J
2
plasticity models 146
E.1 Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
E.2 Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
F Matrix Form for Equations of Solids 149
F.1 Stress and Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
F.2 Split into Deviatoric and Spherical Components . . . . . . . . . . . . . 150
F.3 Linear Elastic Constitutive Equations . . . . . . . . . . . . . . . . . . . 152
F.3.1 Example: Isotropic behavior . . . . . . . . . . . . . . . . . . . . 153
Chapter 1
Introduction
The Finite Element Analysis Program FEAP may be used to solve a wide variety of
problems in linear and non-linear solid continuum mechanics. This report presents the
background necessary to understand the formulations which are employed to develop
the two and three dimensional continuum elements which are provided with the FEAP
system. Companion manuals are available which describe the use of the program [21]
and information for those who wish to modify the program by adding user developed
modules [20].
In this report, Chapters 2 and 3 provide an introduction to problem formulation in
both a strong and a weak form. The strong form of a problem is given as a set of
partial diﬀerential equations; whereas, the weak form of a problem is associated with
either variational equations or variational theorems. Vainberg’s theorem is introduced
to indicate when a variational theorem exists for a given variational equation. A
variational statement provides a convenient basis for constructing the ﬁnite element
model. The linear heat equation is used as an example problem to describe some of
the details concerning use of strong and weak forms.
Chapters 4 and 5 provides a summary of the linear elasticity problem in its strong
and weak forms. Chapter 6 discusses implentation for displacement (irreducible) based
ﬁnite element methods. Chapters 7 and 8 then discuss alternative mixed methods for
treating problems which include constraints leading to near incompressibility. General
mixed and enhanced strain methods are presented as alternatives to develop low order
ﬁnite elements that perform well at the nearly incompressible regime. Special attention
is given to methods which can handle anisotropic elastic models where the elasticity
tangent matrix is fully populated. This is an essential feature required to handle both
inelastic and non-linear constitutive models.
Chapter 9 presents a generalization of the linear elastic constitutive model to that
for linear viscoelasticity. For applications involving an isotropic model and strong
1
CHAPTER 1. INTRODUCTION 2
deviatoric relaxation compared to the spherical problem, a situation can arise at large
times in which the response is nearly incompressible – thus requiring use of elements
that perform well in this regime. Alternative representations for linear viscoelastic
behavior are presented in the form of diﬀerential models and integral equations. The
latter provides a basis for constructing an accurate time integration method which is
employed in the FEAP system.
Chapter 10 presents the general algorithm employed in the FEAP system to model
plasticity type presentations. A discussion is presented for both rate and rate indepen-
dent models, as well as, for a generalized plasticity model. Full details are provided for
the case of isotropic models. The formulation used is based on a return map algorithm
for which analytic tangent matrices for use in a Newton solution algorithm can be
obtained.
Chapter 11 discusses methods used in FEAP to solve constraints included in a ﬁnite
element model. Such constraints are evident in going to the fully incompressible case,
as well as, for the problem of intermittant contact between contiguous bodies. The
simplest approach is use of a penalty approach to embed the constraint without the
introduction of additional parameters in the algebraic problem. An extension using the
Uzawa algorithm for an augmented Lagrangian treatment is then considered and avoids
the need for large penalty parameters – which can lead to numerical ill-conditioning of
the algebraic problem. A ﬁnal option is the use of Lagrange multipliers to include the
constraint. All of these methods are used as part of the FEAP system.
Chapter 12 presents a discussion for extension of problems to the fully transient case.
The Newmark method and some of its variants (e.g., an energy-momentum conserving
method) are discussed as methods to solve the transient algorithm by a discrete time
stepping method.
Finally, Chapter 13 presents a summary for extending the methods discussed in the ﬁrst
twelve chapters to the ﬁnite deformation problem. The chapter presents a summary
for diﬀerent deformation and stress measures used in solid mechanics together with
a discussion on treating hyper-elastic constitutive models. It is shown that general
elements which closely follow the representations used for the small deformation case
can be developed using displacement, mixed, and enhanced strain methods.
Chapter 2
Introduction to Strong and Weak
Forms
2.1 Strong form for problems in engineering
Many problems in engineering are modeled using partial diﬀerential equations (PDE).
The set of partial diﬀerential equations describing such problems is often referred to
as the strong form of the problem. The diﬀerential equations may be either linear or
non-linear. Linear equations are characterized by the appearance of the dependent
variable(s) in linear form only, whereas, non-linear equations include nonlinear terms
also. Very few partial diﬀerential equations may be solved in closed form - one case
being the linear wave equation in one space dimension and time. Some equations
admit use of solutions written as series of products of one dimensional functions for
which exact solutions may be constructed for each function. Again, in general it is not
possible to treat general boundary conditions or problem shapes using this approach.
As an example consider the Poisson equation

2
u
∂x
2
+

2
u
∂y
2
= q(x, y) (2.1)
deﬁned on the region 0 ≤ x ≤ a, 0 ≤ y ≤ b with the boundary condition u = 0 on all
edges. This diﬀerential equation may be solved by writing u as a product form
u =

m

n
sin(
mπx
a
) sin(
nπy
b
)u
mn
(2.2)
which when substituted into the equation yields
3
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 4

m

n
_
_

a
_
2
+
_

b
_
2
_
sin(
mπx
a
) sin(
nπy
b
)u
mn
= q(x, y) (2.3)
The solution may now be completed by expanding the right hand side as a double
sine series (i.e., Fourier series) and matching terms between the left and right sides.
Evaluation of the solution requires the summation of the series for each point (x, y)
of interest. Consequently, it is not possible to get an exact solution in closed form.
Indeed, use of a ﬁnite set of terms leads to an approxiamte solution with the accuracy
depending on the number of terms used.
More general solutions may be constructed using separable solution; however, again,
the solutions are obtained only in series form. In the sequel, we will be concerned
with the construction of approximate solutions based on the ﬁnite element method.
This is similar to a series solution in that each mesh used to construct an FE solution
represents a particular number of terms. Indeed, if sequences of meshes are constructed
by subdivision the concept of a series is also obtained since by constraining the added
nodes to have values deﬁned by a subdivision the results for the previous mesh is
recovered - in essence this is the result for fewer terms in the series. Meshes constructed
by subdivision are sometimes referred to as a Ritz sequence due to their similarity with
solutions constructed in series form from variational equations. It is well established
that the ﬁnite element method is one of the most powerful methods to solve general
problems represented as sets of partial diﬀerential equations. Accordingly, we now
direct our attention to rewriting the set of equations in a form we call the weak form
of the problem. The weak form will be the basis for constructing our ﬁnite element
solutions.
2.2 Construction of a weak form
A weak form of a set of diﬀerential equations to be solved by the ﬁnite element method
is constructed by considering 4 steps:
1. Multiply the diﬀerential equation by an arbitrary function which contracts the
equations to a scalar.
2. Integrate the result of 1. over the domain of consideration, Ω.
3. Integrate by parts using Green’s theorem to reduce derivatives to their minimum
order.
4. Replace the boundary conditions by an appropriate construction.
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 5
2.3 Heat conduction problem: Strong form
The above steps are made more concrete by considering an example. The governing
partial diﬀerential equation set for the transient heat conduction equation is given by

d

i=1
∂q
i
∂x
i
+ Q = ρ c
∂T
∂t
(2.4)
where: d is the spatial dimension of the problem; q
i
is the component of the heat ﬂux
in the x
i
direction; Q is the volumetric heat generation per unit volume per unit time,
T is temperature; ρ is density; c is speciﬁc heat; and t is time. The equations hold for
all points x
i
in the domain of interest, Ω.
The following notation is introduced for use throughout this report. Partial derivatives
in space will be denoted by
( · )
,i
=
∂( · )
∂x
i
(2.5)
and in time by
˙
T =
∂T
∂t
(2.6)
In addition, summation convention is used where
a
i
b
i
=
d

i=1
a
i
b
i
(2.7)
With this notation, the divergence of the ﬂux may be written as
q
i,i
=
d

i=1
∂q
i
∂x
i
(2.8)
Boundary conditions are given by
T(x
j
, t) =
¯
T (2.9)
where
¯
T is a speciﬁed temperature for points x
j
on the boundary, Γ
T
,; and
q
n
= q
i
n
i
= ¯ q
n
(2.10)
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 6
where ¯ qn
n
is a speciﬁed ﬂux for points x
j
on the ﬂux boundary, Γ
q
, and n
i
are direction
cosines of the unit outward pointing normal to the boundary. Initial conditions are
given by
T(x
i
, 0) =
¯
T
0
(x
i
) (2.11)
for points in the domain, Ω, at time zero. The equations are completed by giving a
relationship between the gradient of temperature and the heat ﬂux (called the thermal
constitutive equation). The Fourier law is a linear relationship given as
q
i
= −k
ij
T
,j
(2.12)
where k
ij
is a symmetric, second rank thermal conductivity tensor. For an isotropic
material
k
ij
= kδ
ij
(2.13)
in which δ
ij
is the Kronecker delta function (δ
ij
= 1 for i = j; = 0 for i = j). Hence for
an isotropic material the Fourier law becomes
q
i
= −kT
,i
(2.14)
The diﬀerential equation may be expressed in terms of temperature by substituting
Eq. 2.14 into Eq. 2.4. The result is
(kT
,i
)
,i
+ Q = ρc
˙
T (2.15)
The equation is a second order diﬀerential equation and for isotropic materials with
constant k is expanded for two dimensional plane bodies as
k
_

2
T
∂x
2
1
+

2
T
∂x
2
2
_
+ Q = ρc
∂T
∂t
(2.16)
We note that it is necessary to compute second derivatives of the temperature to com-
pute a solution to the diﬀerential equation. In the following, we show that, expressed
as a weak form, it is only necessary to approximate ﬁrst derivatives of functions to
obtain a solution. Thus, the solution process is simpliﬁed by considering weak (varia-
tional) forms. The partial diﬀerential equation together with the boundary and initial
conditions is called the strong form of the problem.
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 7
2.4 Heat conduction problem: Weak form
In step 1, we multiply Eq. 2.4 by an arbitrary function W(x
i
), which transforms the
set of diﬀerential equations onto a scalar function. The equation is ﬁrst written on one
side of an equal sign. Thus
g(W, q
i
, T) = W(x
i
)
_
ρc
˙
T − Q + q
i,i
_
= 0 (2.17)
In step 2 we integrate over the domain, Ω. Thus,
G(W, q
i
, T) =
_

W(x
i
)
_
ρc
˙
T − Q + q
i,i
_
dΩ = 0 (2.18)
In step 3 we integrate by parts the terms involving the spatial derivatives (i.e., the
thermal ﬂux vector in our case). Green’s theorem is given by
_

φ
,i
dΩ =
_
Γ
φn
i
d Γ (2.19)
Normally, φ is the product of two functions. Thus for
φ = V U (2.20)
we have
_

(U V )
,i
dΩ =
_
Γ
(U V )n
i
dΓ (2.21)
The left hand side expands to give
_

[U V
,i
+ U
,i
V ] dΩ =
_
Γ
(U V )n
i
dΓ (2.22)
which may be rearranged as
_

U V
,i
dΩ = −
_

U
,i
V dΩ +
_
Γ
(U V )n
i
dΓ (2.23)
which we observe is an integration by parts.
Applying the integration by parts to the heat equation gives
G(W, q
i
, T) =
_

W(x
i
)
_
ρc
˙
T − Q
_
dΩ −
_

W
,i
q
i
d Ω
+
_
Γ
Wq
i
n
i
dΓ = 0 (2.24)
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 8
Introducing q
n
, the boundary term may be split into two parts and expressed as
_
Γ
Wq
n
dΓ =
_
Γ
T
Wq
n
dΓ +
_
Γ
q
Wq
n
dΓ (2.25)
Now the boundary condition Eq. 2.10 may be used for the part on Γ
q
and (without
any loss in what we need to do) we can set W to zero on Γ
u
(Note that W is arbitrary,
hence our equation must be valid even if W is zero for some parts of the domain).
Substituting all the above into Eq. 2.24 completes step 4 and we obtain the ﬁnal
expression
G(W, q
i
, T) =
_

W(x
i
)
_
ρc
˙
T − Q
_
dΩ −
_

W
,i
q
i
dΩ
+
_
Γ
q
W¯ q
n
dΓ = 0 (2.26)
If in addition to the use of the boundary condition we assume that the Fourier law is
satisﬁed at each point in Ω the above integral becomes
G =
_

W
_
ρ c
˙
T − Q
_
dΩ +
_

W
,i
k T
,i
dΩ
+
_
Γ
q
W ¯ q
n
dΓ = 0 (2.27)
We note that the above form only involves ﬁrst derivatives of quantities instead of the
second derivatives in the original diﬀerential equation. This leads to weaker conditions
to deﬁne solutions of the problem and thus the notion of a weak form is established.
Furthermore, there are no additional equations that can be used to give any additional
reductions; thus, Eq. 2.27 is said to be irreducible [26, Chapter 9].
2.5 Approximate solutions: The ﬁnite element method
For ﬁnite element approximate solutions, we deﬁne each integral as a sum of integrals
over each element. Accordingly, we let
Ω ≈ Ω
h
=
N
el

e=1

e
(2.28)
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 9
where Ω
h
is the approximation to the domain created by the set of elements, Ω
e
is the
domain of a typical element and N
el
is the number of nodes attached to the element.
Integrals may now be summed over each element and written as
_

(·) dΩ ≈
_

h
(·) dΩ =
N
el

e=1
_

e
(·) dΩ (2.29)
Thus our heat equation integral becomes
G ≈ G
h
=
N
el

e=1
_

e
W
_
ρc
˙
T − Q
_
dΩ −
N
el

e=1
_

e
W
,i
q
i
dΩ
+
N
el

e=1
_
Γ
eq
W ¯ q
n
dΓ = 0 (2.30)
Introducing the Fourier law the above integral becomes
G ≈ G
h
=
N
el

e=1
_

e
W
_
ρc
˙
T − Q
_
dΩ +
N
el

e=1
_

e
W
,i
kT
,i
dΩ
+
N
el

e=1
_
Γ
eq
W ¯ q
n
dΓ = 0 (2.31)
In order for the above integrals to be well deﬁned, surface integrals between adja-
cent elements must vanish. This occurs under the condition that both W and T are
continuous in Ω. With this approximation, the ﬁrst derivatives of W and T may be
discontinuous in Ω. The case where only the function is continuous, but not its ﬁrst
derivatives, deﬁnes a class called a C
0
function. Commonly, the ﬁnite element method
uses isoparametric elements to construct C
0
functions in Ω
h
. Standard element inter-
polation functions which maintain C
0
continuity are discussed in any standard book
on the ﬁnite element method (e.g., See [26, Chapter 7]). Isoparametric elements, which
maintain the C
0
condition, satisfy the conditions
x
i
=
N
el

I=1
N
I
(ξ)x
I
i
(2.32)
for coordinates and
T =
N
el

I=1
N
I
(ξ)T
I
(t) (2.33)
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 10
for temperature. Similar expressions are used for other quantities also. In the above,
I refers to a node number, N
I
is a speciﬁed spatial function called a shape function for
node I, ξ are natural coordinates for the element, x
I
i
are values of the coordinates at
node I, T
I
(t) are time dependent nodal values of temperature, and nel is the number
of nodes connected to an element. Standard shape functions, for which all the nodal
parameters have the value of approximations to the variable, satisfy the condition
N
el

I=1
N
I
(ξ) = 1 (2.34)
This ensures the approximations contain the terms (1, x
i
) and thus lead to convergent
solutions. In summation convention, the above interpolations are written as
x
i
= N
I
(ξ) x
I
i
(2.35)
and
T = N
I
(ξ) T
I
(t) (2.36)
The weight function may also be expressed as
W = N
I
(ξ) W
I
(2.37)
where W
I
are arbitrary parameters. This form of approximation is attributed to
Galerkin (or Bubnov-Galerkin) and the approximate solution process is often called
a Galerkin method. It is also possible to use a diﬀerent approximation for the weight-
ing functions than for the dependent variable, leading to a method called the Petrov-
Galerkin process.
The shape functions for a 4-node quadrilateral element in two-dimensions may be
written as
N
I
(ξ) =
1
4
(1 + ξ
I
1
ξ
1
)(1 + ξ
I
2
ξ
2
) (2.38)
where ξ
I
i
are values of the natural coordinates at node I. Later we also will use an
alternative representation for these shape functions; however, the above suﬃces for
most developments. Derivatives for isoparametric elements may be constructed using
the chain rule. Accordingly, we may write
∂N
I
∂ξ
i
=
∂N
I
∂x
j
∂x
j
∂ξ
i
=
∂N
I
∂x
j
J
ji
(2.39)
where the Jacobian transformation between coordinates is deﬁned by
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 11
J
ji
=
∂x
j
∂ξ
i
(2.40)
The above constitutes a set of linear equations which may be solved at each natural co-
ordinate value (e.g., quadrature point) to specify the derivatives of the shape functions.
Accordingly
∂N
I
∂x
j
=
∂N
I
∂ξ
i
J
−1
ji
(2.41)
Using the derivatives of the shape functions we may write the gradient of the temper-
ature in two dimensions as
_
T
,x
1
T
,x
2
_
=
_
N
I,x
1
N
I,x
2
_
T
I
(t) (2.42)
Similarly, the gradient of the weighting function is expressed as
_
W
,x
1
W
,x
2
_
=
_
N
I,x
1
N
I,x
2
_
W
I
(2.43)
Finally the rate of temperature change in each element is written as
˙
T = N
I
(ξ)
˙
T
I
(t) (2.44)
With the above deﬁnitions available, we can write the terms in the weak form for each
element as
_

e
Wρc
˙
TdΩ = W
I
M
IJ
˙
T
J
(2.45)
where
M
IJ
=
_

e
N
I
ρ c N
J
dΩ (2.46)
deﬁnes the element heat capacity matrix. Similarly, the term
_

e
W
,i
k T
,i
dΩ = W
I
K
IJ
T
J
(2.47)
where
K
IJ
=
_

e
N
I,i
k N
J,i
dΩ (2.48)
deﬁnes the element conductivity matrix. Finally,
_

e
W QdΩ −
_
Γ
eq
W ¯ q
n
dΓ = W
I
F
I
(2.49)
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 12
where
F
I
=
_

e
N
I
QdΩ −
_
Γ
e
q
N
I
¯ q
n
dΓ (2.50)
The approximate weak form may now be written as
G
h
=
N
el

e=1
W
I
(M
IJ
˙
T
J
+ K
IJ
T
J
− F
I
) = 0 (2.51)
and since W
I
is an arbitrary parameter, the set of equations to be solved is
N
el

e=1
(M
IJ
˙
T
J
+ K
IJ
T
J
− F
I
) = 0 (2.52)
In matrix notation we can write the above as
M
˙
T + KT = F (2.53)
which for the transient problem is a large set of ordinary diﬀerential equations to be
solved for the nodal temperature vector, T. For problems where the rate of tempera-
ture,
˙
T, may be neglected, the steady state problem
KT = F (2.54)
results.
2.6 Implementation of elements into FEAP
The implementation of a ﬁnite element development into the general purpose program
FEAP (Finite Element Analysis Program) is accomplished by writing a subprogram
named ELMTnn (nn = 01 to 50) [26, 27, 20]. The subroutine must input the material
parameters, compute the ﬁnite element arrays, and output any desired quantities. In
addition, the element routine performs basic computations to obtain nodal values for
contour plots of element variables (e.g., the thermal ﬂux for the heat equation, stresses
for mechanics problems, etc.).
The basic arrays to be computed in each element for a steady state heat equation are
K
IJ
=
_

e
N
I,i
k N
J,i
dΩ (2.55)
and
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 13
F
I
=
_

e
N
I
QdΩ −
_
Γ
eq
N
I
¯ q
n
dΓ (2.56)
For a transient problem is is necessary to also compute
M
IJ
=
_

e
N
I
ρ c N
J
dΩ (2.57)
The above integrals are normally computed using numerical quadrature, where for
example
K
IJ
=
L

l=1
N
I,i

l
) k N
J,i

l
)j(ξ
l
)w
l
(2.58)
where j(ξ) is the determinant of J evaluated at the quadrature point ξ
l
and w
l
are
FEAP is a general non-linear ﬁnite element solution system, hence it needs to compute
a residual for the equations (see FEAP User and Programmer Manual for details). For
the linear heat equation the residual may be expressed as
R = F − KT − M
˙
T (2.59)
A solution to a problem is achieved when
R = 0 (2.60)
Each array is computed for a single element as described in the section of the FEAP
Programmer Manual on adding an element. The listing included in Appendix A sum-
marizes an element for the linear heat transfer problem. Both steady state and transient
solutions are permitted. The heat capacity array, M, is included separately to permit
solution of the general linear eigenproblem
KΦ = MΦΛ (2.61)
which can be used to assess the values of basic time parameters in a problem. The
routine uses basic features included in the FEAP system to generate shape functions,
An example of a solution to a problem is the computation of the temperature in a
rectangular region encasing a circular insulator and subjected to a thermal gradient.
The sides of the block are assumed to also be fully insulated. One quadrant of the
region is modeled as shown by the mesh in Figure 2.1.
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 14
Figure 2.1: Mesh for thermal example
The top of the region is exposed to a constant temperature of 10C
o
and the symmetry
axis is assumed to be at zero temperature. The routines indicated in Tables A.1 to A.5
are incorporated into FEAP as a user element and the steady state solution computed.
The contour of temperatures is shown in Figure 2.2.
CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 15
Figure 2.2: Temperature contours for thermal example
Chapter 3
Introduction to Variational
Theorems
3.1 Derivatives of functionals: The variation
The weak form of a diﬀerential equation is also called a variational equation. The
notion of a variation is associated with the concept of a derivative of a functional
(i.e., a function of functions). In order to construct a derivative of a functional, it is
necessary to introduce a scalar parameter which may be used as the limiting parameter
in the derivative [10]. This may be done by introducing a parameter η and deﬁning a
family of functions given by
T
η
(x) = T(x) + η τ(x) (3.1)
The function τ is an arbitrary function and is related to the arbitrary function W
introduced in the construction of the weak form. The function ητ is called the variation
of the function T and often written as δT (τ(x) alone also may be called the variation
of the function) [10].
Introducing the family of functions T
η
into the functional we obtain, using the steady
state heat equation as an example, the result
G
η
= G(W, T
η
) =
_

W
,i
k T
η
,i
dΩ −
_

W QdΩ
+
_
Γ
q
W ¯ q
n
dΓ (3.2)
The derivative of the functional with respect to η now may be constructed using con-
ventional methods of calculus. Thus,
dG

= lim
η→0
G
η
− G
0
η
(3.3)
16
CHAPTER 3. INTRODUCTION TO VARIATIONAL THEOREMS 17
where G
0
is the value of G
η
for η equal to 0. The construction of the derivative of the
functional requires the computation of variations of derivatives of T. Using the above
deﬁnition we obtain
d(T
η
)
,i

=
d

(T
,i
+ ητ
,i
) = τ
,i
(3.4)
With this result in hand, the derivative of the functional with respect to η is given by
dG

=
_

W
,i
k τ
,i
dΩ (3.5)
The limit of the derivative as η goes to zero is called the variation of the functional.
For the linear steady state heat equation the derivative with respect to η is constant,
hence the derivative is a variation of G. We shall deﬁne the derivative of the functional
representing the weak form of a diﬀerential equation as
dG

= A(W, τ) (3.6)
This is a notation commonly used to deﬁne inner products.
3.2 Symmetry of inner products
Symmetry of inner product relations is fundamental to the derivation of variational
theorems. To investigate symmetry of a functional we consider only terms which include
both the dependent variable and the arbitrary function. An inner product is symmetric
if
A(W, τ) = A(τ, W) (3.7)
Symmetry of the inner product resulting from the variation of a weak form is a suﬃcient
condition for the existence of a variational theorem which may also be used to generate
a weak form. Symmetry of the functional A also implies that the tangent matrix
(computed from the second variation of the theorem or the ﬁrst variation of the weak
form) of a Bubnov-Galerkin ﬁnite element method will be symmetric.
A variational theorem, given by a functional Π(T), has a ﬁrst variation which is iden-
tical to the weak form. Thus, given a functional Π(T) we can construct G(W, T) as
lim
η→0
dΠ(T
η
)

= G(τ, T) (3.8)
Note that use of Eq. 3.1 leads to a result where τ replaces W in the weak form. Thus,
for the variational equation to be equivalent to the weak form τ must be an arbitrary
function with the same restrictions as we established in deﬁning W. Variational theo-
rems are quite common for several problem classes; however, often we may only have a
CHAPTER 3. INTRODUCTION TO VARIATIONAL THEOREMS 18
functional G and desire to know if a variational theorem exists. In practice we seldom
need to have the variational theorem, but knowledge that it exists is helpful since it
implies properties of the discrete problem which are beneﬁcial (e.g., symmetry of the
tangent matrices, minimum or stationary value, etc.). Also, existence of a variational
theorem yields a weak form directly by using Eq. 3.8.
The construction of a variational theorem from a weak form is performed as follows
[24]:
1. Check symmetry of the functional A(W, τ). If symmetric then to to 2; otherwise,
stop: no varitational theorem exists.
2. Perform the following substitutions in G(W, T)
W(x) → T(x, t) (3.9)
T(x, t) → ηT(x, t) (3.10)
to deﬁne G(T, ηT)
3. Integrate the functional result from (b) with respect to η over the interval 0 to 1.
The result of the above process gives
Π(T) =
_
1
0
G(T, ηT)dη (3.11)
Performing the variation of Π and setting to zero gives
lim
η→0
dΠ(T
η
)

= G(τ, T) = 0 (3.12)
and a problem commonly referred to as a variational theorem. A variational theorem
is a functional whose ﬁrst variation, when set to zero, yields the governing diﬀerential
equations and boundary conditions associated with some problem.
For the steady state heat equation we have
G(T, ηT) =
_

T
,i
k η T
,i
dΩ −
_

T QdΩ +
_
Γ
q
T ¯ q
n
dΓ (3.13)
The integral is trivial and gives
Π(T) =
1
2
_

T
,i
kT
,i
dΩ −
_

TQdΩ +
_
Γ
q
T ¯ q
n
dΓ (3.14)
CHAPTER 3. INTRODUCTION TO VARIATIONAL THEOREMS 19
Reversing the process, the ﬁrst variation of the variational theorem generates a vari-
ational equation which is the weak form of the partial diﬀerential equation. The ﬁrst
variation is deﬁned by replacing T by
T
η
= T + ητ (3.15)
and performing the derivative deﬁned by Eq. 3.12. The second variation of the theorem
generates the inner product
A(τ, τ) (3.16)
If the second variation is strictly positive (i.e., A is positive for all τ), the variational
theorem is called a minimum principle and the discrete tangent matrix is positive deﬁ-
nite. If the second variation can have either positive or negative values the variational
theorem is a stationary principle and the discrete tangent matrix is indeﬁnite.
The transient heat equation with weak form given by
G =
_

W
_
ρ c
˙
T − Q
_
dΩ +
_

W
,i
k T
,i
dΩ
+
_
Γ
q
W ¯ q
n
dΓ = 0 (3.17)
does not lead to a variational theorm due to the lack of the symmetry condition for
the transient term
A =
_
˙
T, ητ
_
= (η ˙ τ, T) (3.18)
If however, we ﬁrst discretize the transient term using some time integration method,
we can often restore symmetry to the functional and then deduce a variational theorem
for the discrete problem. For example if at each time t
n
we have
T(t
n
) ≈ T
n
(3.19)
then we can approximate the time derivative by the ﬁnite diﬀerence
˙
T(t
n
) ≈
T
n+1
−T
n
t
n+1
−t
n
(3.20)
Letting t
n+1
− t
n
= ∆t and omitting the subscripts for quantities evaluated at t
n+1
,
the rate term which includes both T and τ becomes
A =
_
T
∆t
, ητ
_
=
_
η
τ
∆t
, T
_
(3.21)
CHAPTER 3. INTRODUCTION TO VARIATIONAL THEOREMS 20
since scalars can be moved from either term without aﬀecting the value of the term.
That is,
A = (T, η τ) = (η T, τ) (3.22)
3.3 Variational notation
A formalism for constructing a variation of a functional may be identiﬁed and is similar
to constructing the diﬀerential of a function. The diﬀerential of a function f(x
i
) may
be written as
df =
∂f
∂x
i
dx
i
(3.23)
where x
i
are the set of independent variables. Similarly, we may formally write a ﬁrst
variation as
δΠ =
∂Π
∂u
δu +
∂Π
∂u
,i
δu
,i
+ · · · (3.24)
where u, u
,i
are the dependent variables of the functional, δu is the variation of the
variable (i.e., it is formally the ητ(x)), and δΠ is called the ﬁrst variation of the func-
tional. This construction is a formal process as the indicated partial derivatives have
no direct deﬁnition (indeed the result of the derivative is obtained from Eq. 3.3). How-
ever, applying this construction can be formally performed using usual constructions
for a derivative of a function. For the functional Eq. 3.14, we obtain the result
δΠ =
1
2
_

∂T
,i
(T
,i
k T
,i
) δT
,i
dΩ −
_

∂T
(T Q) δT dΩ
+
_
Γ
q

∂T
(T ¯ q
n
) δT dΓ (3.25)
δΠ =
1
2
_

(k T
,i
+ T
,i
k) δT
,i
dΩ −
_

QδT dΩ +
_
Γ
q
¯ q
n
δT dΓ (3.26)
Collecting terms we have
δΠ =
_

δT
,i
k T
,i
dΩ −
_

QδT dΩ +
_
Γ
q
¯ q
n
δT dΓ (3.27)
which is identical to Eq. 3.2 with δT replacing W, etc.
This formal construction is easy to apply but masks the meaning of a variation. We
may also use the above process to perform linearizations of variational equations in
order to construct solution processes based on Newton’s method. We shall address this
aspect at a later time.
Chapter 4
Small Deformation: Linear
Elasticity
A summary of the governing equations for linear elasticity is given below. The equations
are presented using direct notation. For a presentation using indicial notation see [26,
Chapter 6]. The presentation below assumes small (inﬁnitesimal) deformations and
general three dimensional behavior in a Cartesian coordinate system, x, where the
domain of analysis is Ω with boundary Γ. The dependent variables are given in terms
of the displacement vector, u, the stress tensor, σ, and the strain tensor, . The basic
governing equations are:
1. Balance of linear momentum expressed as
∇· σ + ρ b
m
= ρ ¨ u (4.1)
where ρ is the mass density, b
m
is the body force per unit mass, ∇ is the gradient
operator, and ¨ u is the acceleration.
2. Balance of angular momentum, which leads to symmetry of the stress tensor
σ = σ
T
(4.2)
3. Deformation measures based upon the gradient of the displacement vector, ∇u,
which may be split as follows
∇u = ∇
(s)
u + ∇
(a)
u (4.3)
where the symmetric part is

(s)
u =
1
2
_
∇u + (∇u)
T
¸
(4.4)
21
CHAPTER 4. SMALL DEFORMATION: LINEAR ELASTICITY 22
and the skew symmetric part is

(a)
u =
1
2
_
∇u − (∇u)
T
¸
(4.5)
Based upon this split, the symmetric part deﬁnes the strain
= ∇
(s)
u (4.6)
and the skew symmetric part deﬁnes the spin, or small rotation,
ω = ∇
(a)
u (4.7)
In a three dimensional setting the above tensors have 9 components. However, if
the tensor is symmetric only 6 are independent and if the tensor is skew symmetric
only 3 are independent. The component ordering for each of the tensors is given
by
σ →
_
_
σ
11
σ
12
σ
13
σ
21
σ
22
σ
23
σ
31
σ
32
σ
33
_
_
(4.8)
which from the balance of angular momentum must be symmetric, hence
σ
ij
= σ
ji
(4.9)
The gradient of the displacement has the components ordered as (with no sym-
metries)
∇u →
_
_
u
1,1
u
1,2
u
1,3
u
2,1
u
2,2
u
2,3
u
3,1
u
3,2
u
3,3
_
_
(4.10)
The strain tensor is the symmetric part with components

_
_

11

12

13

21

22

23

31

32

33
_
_
(4.11)
and the symmetry condition

ij
=
ji
(4.12)
The spin tensor is skew symmetric,thus,
ω
ij
= ω
ji
(4.13)
which implies ω
11
= ω
22
= ω
33
= 0. Accordingly,
ω →
_
_
ω
11
ω
12
ω
13
ω
21
ω
22
ω
23
ω
31
ω
32
ω
33
_
_
=
_
_
0 ω
12
ω
13
−ω
12
0 ω
23
−ω
13
−ω
23
0
_
_
(4.14)
CHAPTER 4. SMALL DEFORMATION: LINEAR ELASTICITY 23
The basic equations which are independent of material constitution are completed by
specifying the boundary conditions. For this purpose the boundary, Γ, is split into two
parts:
• Speciﬁed displacements on the part Γ
u
, given as:
u = ¯ u (4.15)
where ¯ u is a speciﬁed quantity; and
• speciﬁed tractions on the part Γ
t
, given as:
t = σ
n
=
¯
t (4.16)
where
¯
t is a speciﬁed quantity.
In the balance of momentum, the body force was speciﬁed per unit of mass. This may
be converted to a body force per unit volume (i.e., unit weight/volume) using
ρ b
m
= b
v
(4.17)
Static or quasi-static problems are considered by omitting the acceleration term from
the momentum equation (Eq. 4.1). Inclusion of intertial forces requires the speciﬁca-
tion of the initial conditions
u(x, 0) = d
0
(x) (4.18)
˙ u(x, 0) = v
0
(x) (4.19)
where d
0
is the initial displacement ﬁeld, and v
0
is the initial velocity ﬁeld.
4.1 Constitutive Equations for Linear Elasticity
The linear theory is completed by specifying the constitutive behavior for the material.
In small deformation analysis the strain is expressed as an additive sum of parts. We
shall consider several alternatives for splits during the course; however, we begin by
considering a linear elastic material with an additional known strain. Accordingly,
=
m
+
0
(4.20)
where
m
is the strain caused by stresses and is called the mechanical part,
0
is a
second part which we assume is a speciﬁed strain. For example,
0
as a thermal strain
is given by

0
=
th
= α(T − T
0
) (4.21)
CHAPTER 4. SMALL DEFORMATION: LINEAR ELASTICITY 24
.LP where T is temperature and T
0
is a stress free temperature. The constitutive
equations relating stress to mechanical strain may be written (in matrix notation,
which is also called Voigt notation) as
σ = D
m
= D( −
0
) (4.22)
where the matrix of stresses is ordered as the vector
σ =
_
σ
11
σ
22
σ
33
σ
12
σ
23
σ
31
¸
T
(4.23)
the matrix of strains is ordered as the vector (note factors of 2 are used to make shearing
components the engineering strains, γ
ij
)
=
_

11

22

33
2
12
2
23
2
31
¸
T
(4.24)
and D is the matrix of elastic constants given by
D =
_
¸
¸
¸
¸
¸
¸
_
D
11
D
12
D
13
D
14
D
15
D
16
D
21
D
22
D
23
D
24
D
25
D
26
D
31
D
32
D
33
D
34
D
35
D
36
D
41
D
42
D
43
D
44
D
45
D
46
D
51
D
52
D
53
D
54
D
55
D
56
D
61
D
62
D
63
D
64
D
65
D
66
_
¸
¸
¸
¸
¸
¸
_
(4.25)
Assuming the existence of a strain energy density, W(
m
), from which stresses are
computed as
σ
ab
=
∂W

m
ab
(4.26)
the elastic modulus matrix is symmetric and satisﬁes
D
ij
= D
ji
(4.27)
Using tensor quantities, the constitutive equation for linear elasticity is written in
indicial notation as:
σ
ab
= C
abcd
(
cd

0
cd
) (4.28)
The transformation from the tensor to the matrix (Voigt) form is accomplished by the
index transformations shown in Table 4.1
Thus, using this table, we have
C
1111
→ D
11
; C
1233
→ D
43
; etc. (4.29)
The above set of equations deﬁnes the governing equations for use in solving linear
elastic boundary value problems in which the inertial forces may be ignored. We next
CHAPTER 4. SMALL DEFORMATION: LINEAR ELASTICITY 25
Tensor Matrix Index
Index 1 2 3 4 5 6
ab 11 22 33 12 23 31
21 32 13
Table 4.1: Transformation of indices from tensor to matrix form
discuss some variational theorems which include the elasticity equations in a form
amenable for ﬁnite element developments.
For the present, we assume that inertial forces may be ignored. The inclusion of inertial
forces precludes the development of variational theorems in a simple form as noted in
the previous chapter. Later, we can add the inertial eﬀects and use time discrete
methods to restore symmetry to the formulation.
Chapter 5
Variational Theorems: Linear
Elasticity
5.1 Hu-Washizu Variational Theorem
Instead of constructing the weak form of the equations and then deducing the existence
of a variational theorem, as done for the thermal problem, a variational theorem which
includes all the equations for the linear theory of elasticity (without inertial forces)
will be stated. The variational theorem is a result of the work of the Chinese scholar,
Hu, and the Japanese scholar, K. Washizu [25], and, thus, is known as the Hu-Washizu
variational theorem. The theorem may be written as
I(u, σ, ) =
1
2
_

T
D dΩ −
_

T
D
0
dΩ
+
_

σ
T
(∇
(s)
u − ) dΩ −
_

u
T
b
v
d

_
Γ
t
u
T
¯
t dΓ −
_
Γ
u
t
T
(u − ¯ u)dΓ = Stationary (5.1)
Note that the integral deﬁning the variational theorem is a scalar; hence, a transpose
may be introduced into each term without changing the meaning. For example,
I =
_

a
T
bdΩ =
_

(a
T
b)
T
dΩ =
_

b
T
a dΩ (5.2)
A variational theorem is stationary when the arguments (e.g., u, σ, ) satisfy the condi-
tions where the ﬁrst variation vanishes. To construct the ﬁrst variation, we proceed as
in the previous chapter. Accordingly, we introduce the variations to the displacement,
U, the stress, S, and the strain, E, as
u
η
= u + η U (5.3)
26
CHAPTER 5. VARIATIONAL THEOREMS: LINEAR ELASTICITY 27
σ
η
= σ + η S (5.4)

η
= + η E (5.5)
and deﬁne the single parameter functional
I
η
= I(u
η
, σ
η
,
η
) (5.6)
The ﬁrst variation is then deﬁned as the derivative of I
η
with respect to η and evaluated
at η = 0. For the Hu-Washizu theorem the ﬁrst variation deﬁning the stationary
condition is given by
dI
η

¸
¸
¸
¸
η=0
=
_

E
T
DdΩ −
_

E
T
D
0
dΩ
+
_

S
T
(∇
(s)
u − )dΩ +
_

σ
T
(∇
(s)
U − E)dΩ

_

U
T
b
v
dΩ −
_
Γ
t
U
T
¯
tdΓ

_
Γ
u
n
T
S(u − ¯ u)dΓ −
_
Γ
u
t
T
UdΓ = 0 (5.7)
The ﬁrst variation may also be constucted using 3.23 for each of the variables. The
result is
δI =
_

δ
T
DdΩ −
_

δ
T
D
0
dΩ
+
_

δσ
T
(∇
(s)
u − )dΩ +
_

σ
T
(∇
(s)
δu − δ)dΩ

_

δu
T
b
v
dΩ −
_
Γ
t
δu
T
¯
tdΓ

_
Γ
u
n
T
δσ(u − ¯ u)dΓ −
_
Γ
u
t
T
δudΓ = 0 (5.8)
and the two forms lead to identical results.
In order to show that the theorem in form 5.7 is equivalent to the equations for linear
elasticity, we need to group all the terms together which multiply each variation func-
tion (e.g., the U, S, E). To accomplish the grouping it is necessary to integrate by
parts the term involving ∇
(s)
U. Accordingly,
_

σ
T

(s)
UdΩ = −
_

U
T
∇· σdΩ +
_
Γ
t
t
T
UdΓ +
_
Γ
u
t
T
UdΓ (5.9)
CHAPTER 5. VARIATIONAL THEOREMS: LINEAR ELASTICITY 28
Grouping all the terms we obtain
dI
η

¸
¸
¸
¸
η=0
=
_

E
T
[D( −
0
) − σ]dΩ
+
_

S
T
(∇
(s)
u − )dΩ −
_

U
T
(∇· σ + b
v
)dΩ
+
_
Γ
t
U
T
(t −
¯
t)dΓ −
_
Γ
u
n
T
S(u − ¯ u)dΓ = 0 (5.10)
The fundamental lemma of the calculus of variations states that each expression mul-
tiplying an arbitrary function in each integral type must vanish at each point in the
domain of the integral. The lemma is easy to prove. Suppose that an expression does
not vanish at a point, then, since the variation is arbitrary, we can assume that it is
equal to the value of the non-vanishing expression. This results in the integral of the
square of a function, which must then be positive, and hence the integral will not be
zero. This leads to a contradiction, and thus the only possibility is that the assumption
of a non-vanishing expression is false.
The expression which multiplies each variation function is called an Euler equation of
the variational theorem. For the Hu-Washizu theorem, the variations multiply the con-
stitutive equation, the strain-displacement equation, the balance of linear momentum,
the traction boundary condition, and the displacement boundary condition. Indeed,
the only equation not contained is the balance of angular momentum.
The Hu-Washizu variational principle will serve as the basis for most of what we need
in the course. There are other variational principles which can be deduced directly
from the principle. Two of these, the Hellinger-Reissner principle and the principle
of minimum potential energy are presented below since they are also often used in
constructing ﬁnite element formulations in linear elasticity.
5.2 Hellinger-Reissner Variational Theorem
The Hellinger-Reissner principle eliminates the strain as a primary dependent variable;
consequently, only the displacement, u, and the stress, σ, remain as arguments in
the functional for which variations are constructed. The strains are eliminated by
developing an expression in terms of the stresses. For linear elasticity this leads to
=
0
+ D
−1
σ (5.11)
The need to develop an expression for strains in terms of stresses limits the application
of the Hellinger-Reissner principle. For example, in ﬁnite deformation elasticity the
development of a relation similar to 5.11 is not possible in general. On the other
CHAPTER 5. VARIATIONAL THEOREMS: LINEAR ELASTICITY 29
hand, the Hellinger-Reissner principle is an important limiting case when considering
problems with constraints (e.g., linear elastic incompressible problems, thin plates as
a limit case of the thick Mindlin-Reissner theory). Thus, we shall on occasion use the
principle in our studies. Introducing 5.11 into the Hu-Washizu principle leads to the
result
I(u, σ) = −
1
2
_

0T
D
0
dΩ −
1
2
_

σ
T
D
−1
σdΩ

_

σ
T

0
dΩ +
_

σ
T

(s)
udΩ −
_

u
T
b
v
dΩ

_
Γ
t
u
T
¯
tdΓ −
_
Γ
u
t
T
(u − ¯ u)dΓ (5.12)
The Euler equations for this principle are

(s)
u =
0
+ D
−1
σ (5.13)
together with 4.1, 4.15 and 4.16. The strain-displacement equations are deduced by
either directly stating 4.6 or comparing 5.11 to 5.13. The ﬁrst term in 5.12 may be
omitted since its ﬁrst variation is zero.
5.3 Minimum Potential Energy Theorem
The principle of minimum potential energy eliminates both the stress, σ, and the strain,
, as arguments of the functional. In addition, the displacement boundary conditions
are assumed to be imposed as a constraint on the principle. The MPE theorem may
be deduced by assuming
= ∇
(s)
u (5.14)
and
u = ¯ u (5.15)
are satisﬁed at each point of Ω and Γ, respectively. Thus, the variational theorem is
given by the integral functional
I(u) =
1
2
_

(∇
(s)
u)
T
D(∇
(s)
u)dΩ −
_

(∇
(s)
u)
T
D
0
dΩ

_

u
T
b
v
dΩ −
_
Γ
t
u
T
¯
tdΓ (5.16)
Since stress does not appear explicitly in the theorem, the constitutive equation must
be given. Accordingly, in addition to 5.14 and 5.15 the relation
σ = D( −
0
) (5.17)
CHAPTER 5. VARIATIONAL THEOREMS: LINEAR ELASTICITY 30
is given.
The principle of minimum potential energy is often used as the basis for developing a
displacement ﬁnite element method.
Chapter 6
Displacement Finite Element
Methods
A variational equation or theorem may be solved using the direct method of the calculus
of variations. In the direct method of the calculus of variations the dependent variables
are expressed as a set of trial functions multiplying parameters. This reduces a steady
state problem to an algebraic process and a transient problem to a set of ordinary
diﬀerential equations. In the ﬁnite element method we divide the region into elements
and perform the approximations on each element. As indicated in Chapter 2 the region
is divided as
Ω ≈ Ω
h
=
M
el

e=1

e
(6.1)
and integrals are deﬁned as
_

( · ) dΩ ≈
_

h
( · ) dΩ =
M
el

e=1
_

e
( · ) dΩ (6.2)
In the above M
el
is the total number of elements in the ﬁnite element mesh. A similar
construction is performed for the boundaries. With this construction the parts of the
variational equation or theorem are evaluated element by element.
The ﬁnite element approximation for displacements in an element is introduced as
u(ξ, t) =
N
el

α=1
N
α
(ξ) u
α
(t) = N
α
(ξ) u
α
(t) (6.3)
where N
α
is the shape function at node α, ξ are natural coordinates for the element,
u
α
are the values of the displacement vector at node α and repeated indices imply
summation over the range of the index. Using the isoparametric concept
31
CHAPTER 6. DISPLACEMENT METHODS 32
x(ξ) = N
α
(ξ) x
α
(6.4)
where x
α
are the cartesian coordinates of nodes, the displacement at each point in an
element may be computed.
In the next sections we consider the computation of the external force (from applied
loads) and the internal force (from stresses) by the ﬁnite element process.
6.1 External Force Computation
In our study we will normally satisfy the displacement boundary conditions u = ¯ u by
setting nodal values of the displacement to the values of ¯ u evaluated at nodes. That
is, we express
¯ u = N
α
(ξ) ¯ u
α
(t) (6.5)
and set
¯ u
α
(t) = ¯ u(x
α
, t) (6.6)
We then will assume the integral over Γ
u
is satisﬁed and may be omitted. This step
is not necessary but is common in most applications. The remaining terms involving
speciﬁed applied loads are due to the body forces, b
v
, and the applied surface tractions,
¯
t. The terms in the variational principal are
Π
f
=
_

e
u
T
b
v
dΩ +
_
Γ
te
u
T
¯
t dΓ (6.7)
Using Eq. 6.3 in Eq. 6.7 yields
Π
f
= (u
α
)
T
__

e
N
α
b
v
dΩ +
_
Γ
te
N
α
¯
t dΓ
_
= (u
α
)
T
F
α
(6.8)
where F
α
denotes the applied nodal force vector at node α and is computed from
F
α
=
_

e
N
α
b
v
dΩ +
_
Γ
te
N
α
¯
t dΓ (6.9)
6.2 Internal Force Computation
The stress divergence term in the Hu-Washizu variational principle is generated from
the variation with respect to the displacements, u, of the term
Π
σ
=
_

e
(∇
(s)
u)
T
σdΩ =

e
_

e
(∇
(s)
u)
T
σdΩ (6.10)
CHAPTER 6. DISPLACEMENT METHODS 33
Using the ﬁnite element approximation for displacement, the symmetric part of the
strains deﬁned by the symmetric part of the deformation gradient in each element is
given by

(s)
u = (u) = B
α
u
α
(6.11)
where B
α
is the strain displacement matrix for the element. If the components of the
strain for 3-dimensional problems are ordered as

T
=
_

11

22

33
2
12
2
23
2
31
¸
(6.12)
and related to the displacement derivatives by

T
=
_
u
1,1
u
2,2
u
3,3
(u
1,2
+ u
2,1
) (u
2,3
+ u
3,2
) (u
3,1
+ u
1,3
)
¸
(6.13)
the strain-displacement matrix is expressed as:
B
α
=
_
¸
¸
¸
¸
¸
¸
_
N
α,1
0 0
0 N
α,2
0
0 0 N
α,3
N
α,2
N
α,1
0
0 N
α,3
N
α,2
N
α,3
0 N
α,1
_
¸
¸
¸
¸
¸
¸
_
(6.14)
where
N
α,i
=
∂N
α
∂x
i
(6.15)
For a 2-dimensional plane strain problem the non-zero strains reduce to

T
=
_

11

22

33
2
12
¸
(6.16)
and are expressed in terms of the displacement derivatives as

T
=
_
u
1,1
u
2,2
0 (u
1,2
+ u
2,1
)
¸
(6.17)
thus, B
α
becomes:
B
α
=
_
¸
¸
_
N
α,1
0
0 N
α,2
0 0
N
α,2
N
α,1
_
¸
¸
_
(6.18)
strains are

T
=
_

11

22

33
2
12
¸
(6.19)
and are expressed in terms of the displacements as

T
=
_
u
1,1
u
2,2
u
1
/x
1
(u
1,2
+ u
2,1
)
¸
(6.20)
CHAPTER 6. DISPLACEMENT METHODS 34
The strain-displacement matrix for axisymmetry, B
α
, becomes:
B
α
=
_
¸
¸
_
N
α,1
0
0 N
α,2
N
α
/x
1
0
N
α,2
N
α,1
_
¸
¸
_
(6.21)
where x
1
, x
2
now denote the axisymmetric coordinates r, z, respectively
1
The stress divergence term for each element may be written as
Π
σ
e
= (u
α
)
T
_

e
(B
α
)
T
σdΩ (6.22)
In the sequel we deﬁne the variation of this term with respect to the nodal displace-
ments, u
α
, the internal stress divergence force. This force is expressed by
P
α
(σ) =
_

e
(B
α
)
T
σdΩ (6.23)
which gives
Π
σ
e
= (u
α
)
T
P
α
(σ) (6.24)
The stress divergence term is a basic ﬁnite element quantity and must produce a
response which is free of spurious modes or locking tendencies. Locking is generally
associated with poor performance at or near the incompressible limit. To study the
locking problem we split the formulation into deviatoric and volumetric terms.
6.3 Split into Deviatoric and Spherical Parts
For problems in mechanics it is common to split the stress and strain tensors into their
deviatoric and spherical parts. For stress the spherical part is the mean stress deﬁned
by
p =
1
3
tr(σ) =
1
3
σ
kk
(6.25)
For inﬁnitesimal strains the spherical part is the volume change deﬁned by
θ = tr() =
kk
(6.26)
The deviatoric part of stress , s, is deﬁned so that its trace is zero. The stress may be
written in terms of the deviatoric and pressure parts (pressure is spherical part) as
σ = s + p 1 (6.27)
1
For axisymmetry it is also necessary to replace the volume element by dΩ → x
1
dx
1
dx
2
and the
surface element by dΓ → x
1
dS where dS is an boundary diﬀerential in the x
1
- x
2
plane.
CHAPTER 6. DISPLACEMENT METHODS 35
where, 1 is the rank two identity tensor, which in matrix notation is given by the vector
m
T
=
_
1 1 1 0 0 0
¸
(6.28)
In matrix form the pressure is given by
p =
1
3
m
T
σ (6.29)
thus, the deviatoric part of stresses now may be computed as
s = σ −
1
3
mm
T
σ =
_
I −
1
3
mm
T
_
σ (6.30)
where, in three dimensions, I is a 6 ×6 identity matrix. We note that the trace of the
stress gives
m
T
σ = 3 p = m
T
s + p m
T
m = m
T
s + 3 p (6.31)
and hence
m
T
s = 0 (6.32)
as required.
For subsequent developments, we deﬁne
I
dev
= I −
1
3
mm
T
(6.33)
as the deviatoric projector. Similarly, the volumetric projector is deﬁned by
I
vol
=
1
3
mm
T
(6.34)
These operators have the following properties
I = I
dev
+ I
vol
(6.35)
I
dev
= I
dev
I
dev
= (I
dev
)
m
(6.36)
I
vol
= I
vol
I
vol
= (I
vol
)
m
(6.37)
and
I
vol
I
dev
= I
dev
I
vol
= 0 (6.38)
In the above m is any positive integer power. We note, however, that inverses to the
projectors do not exist.
Utilizing the above properties, we can operate on the strain to deﬁne its deviatoric and
volumetric parts. Accordingly, the deviatoric and volumetric parts are given by
= e +
1
3
θ 1 (6.39)
CHAPTER 6. DISPLACEMENT METHODS 36
where e is the strain deviator and θ is the change in volume. Using matrix notation
we have
θ = m
T
(6.40)
we obtain
e = I
dev
; m
T
e = 0 (6.41)
The strain-displacement matrix also may now be written as a deviatoric and volumetric
form. Accordingly, we use the strain split
(u) = B
α
u
α
= (B
dev
)
α
u
α
+ (B
vol
)
α
u
α
(6.42)
where
B
dev
= I
dev
B (6.43)
and
B
vol
= I
vol
B =
1
3
mb (6.44)
where
b = m
T
B ; m
T
B
dev
= 0 (6.45)
For 3-dimensional problems
b
α
=
_
N
α,1
N
α,2
N
α,3
¸
(6.46)
is the volumetric strain-displacement matrix for a node α in its basic form. In 2-
dimensional plane problems the volumetric strain-displacement matrix is given by
b
α
=
_
N
α,1
N
α,2
¸
(6.47)
and for 2-dimensional axisymmetric problems
b
α
=
_
N
α,1
+ N
α
/x
1
N
α,2
¸
(6.48)
The deviatoric matrix B
dev
is constructed from Eq. 6.39 and yields for the 3-dimensional
problem
B
dev
=
1
3
_
¸
¸
¸
¸
¸
¸
_
2 N
α,1
−N
α,2
−N
α,3
−N
α,1
2 N
α,2
−N
α,3
−N
α,1
−N
α,2
2 N
α,3
3 N
α,2
3 N
α,1
0
0 3 N
α,3
3 N
α,2
3 N
α,3
0 3 N
α,1
_
¸
¸
¸
¸
¸
¸
_
(6.49)
and for the 2-dimensional plane problem
B
dev
=
1
3
_
¸
¸
_
2 N
α,1
−N
α,2
−N
α,1
2 N
α,2
−N
α,1
−N
α,2
3 N
α,2
3 N
α,1
_
¸
¸
_
(6.50)
CHAPTER 6. DISPLACEMENT METHODS 37
Finally, the deviatoric matrix for the 2-dimensional axisymmetric problem is given by:
B
dev
=
1
3
_
¸
¸
_
(2 N
α,1
−N
α
/x
1
) −N
α,2
−(N
α,1
+ N
α
/x
1
) 2 N
α,2
(2 N
α
/x
1
+ N
α,1
) −N
α,2
3 N
α,2
3 N
α,1
_
¸
¸
_
(6.51)
6.4 Internal Force - Deviatoric and Volumetric Parts
The above split of terms is useful in writing the internal force calculations in terms of
deviatoric and volumetric parts. Accordingly,
P
α
=
_

e
B
T
α
σdΩ =
_

e
B
T
α
(s + p m) dΩ (6.52)
which after rearrangement gives
P
α
=
_

e
B
T
α
s dΩ +
_

e
B
T
α
mp dΩ (6.53)
If we introduce
B = B
dev
+ B
vol
= B
dev
+
1
3
mb (6.54)
and use the properties deﬁned above for products of the deviatoric and volumetric
terms, then
P
α
=
_

e
(B
T
dev
)
α
s dΩ +
_

e
b
T
α
p dΩ (6.55)
Since the volumetric term has no eﬀect on the deviatoric stresses the residual may also
be computed from the simpler form in terms of B
α
alone as
P
α
=
_

e
B
T
α
s dΩ +
_

e
b
T
α
p dΩ (6.56)
Thus, the internal force is composed of the sum of deviatoric and volumetric parts.
6.5 Constitutive Equations for Isotropic Linear Elas-
ticity
The constitutive equation for isotropic linear elasticity may be expressed as
σ = λ1tr() + 2 µ (6.57)
CHAPTER 6. DISPLACEMENT METHODS 38
where λ and µ are the Lam´e parameters which are related to Young’s modulus, E, and
Poisson’s ratio, ν, by
λ =
ν E
(1 + ν)(1 − 2 ν)
; µ =
E
2 (1 + ν)
(6.58)
For diﬀerent values of ν, the Lam´e parameters have the following ranges
0 ≤ ν ≤
1
2
; 0 ≤ λ ≤ ∞ (6.59)
and
0 ≤ ν ≤
1
2
;
E
2
≥ µ ≥
E
3
(6.60)
For an incompressible material ν is
1
2
; and λ is a parameter which causes diﬃculties
since it is inﬁnite. Another parameter which is related to λ and µ is the bulk modulus,
K, which is deﬁned by
K = λ +
2
3
µ =
E
3 (1 − 2 ν)
(6.61)
We note that K also tends to inﬁnity as ν approaches
1
2
.
The constitutive equation for an isotropic material is given in indicial form by
σ
ij
= λδ
ij

kk
+ 2 µ
ij
(6.62)
and for a general linear elastic material by
σ
ij
= c
ijkl

kl
(6.63)
where c
ijkl
are the elastic moduli. For an isotropic material the elastic moduli are then
related by
c
ijkl
= λδ
ij
δ
kl
+ µ(δ
ik
δ
jl
+ δ
il
δ
jk
) (6.64)
We note that the above deﬁnition for the moduli satisﬁes all the necessary symmetry
conditions; that is
c
ijkl
= c
klij
= c
jikl
= c
ijlk
(6.65)
The relations may be transformed to matrix (Voigt) notation following Table 4.1 and
expressed as
σ = D (6.66)
where the elastic moduli are split into
D = λD
λ
+ µD
µ
(6.67)
CHAPTER 6. DISPLACEMENT METHODS 39
with
D
λ
=
_
¸
¸
¸
¸
¸
¸
_
1 1 1 0 0 0
1 1 1 0 0 0
1 1 1 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
_
¸
¸
¸
¸
¸
¸
_
= mm
T
= 3 I
vol
(6.68)
D
µ
=
_
¸
¸
¸
¸
¸
¸
_
2 0 0 0 0 0
0 2 0 0 0 0
0 0 2 0 0 0
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
_
¸
¸
¸
¸
¸
¸
_
(6.69)
used as non-dimensional matrices to split the moduli.
2
If the moduli matrices are premultiplied by I
vol
and I
dev
the following results are ob-
tained
I
vol
D
λ
= D
λ
(6.70)
I
dev
D
λ
= 0 (6.71)
I
vol
D
µ
=
2
3
mm
T
=
2
3
D
λ
(6.72)
and
D
µ
I
dev
= I
dev
D
µ
=
1
3
_
¸
¸
¸
¸
¸
¸
_
4 −2 −2 0 0 0
−2 4 −2 0 0 0
−2 −2 4 0 0 0
0 0 0 3 0 0
0 0 0 0 3 0
0 0 0 0 0 3
_
¸
¸
¸
¸
¸
¸
_
= D
dev
(6.73)
Once D
dev
has been computed it may be noted that
I
dev
D
dev
= D
dev
I
dev
= D
dev
(6.74)
I
vol
D
dev
= D
dev
I
vol
= 0 (6.75)
and, thus, it is a deviatoric quantity.
In the following section, the computation of the element stiﬀness matrix for a displace-
ment approach is given and is based upon the above representations for the moduli.
2
Note that in D
µ
the terms multiplying shears have unit values since engineering shear strains are
used (i.e., γ
ij
= 2
ij
).
CHAPTER 6. DISPLACEMENT METHODS 40
6.6 Stiﬀness for Displacement Formulation
The displacement formulation is accomplished for a linear elastic material by noting
that the constitutive equation is given by (for simplicity
0
is assumed to be zero)
σ = D (6.76)
The strains for a displacement approach are given by
= B
β
u
β
(6.77)
where u
β
are the displacements at node β.
Constructing the deviatoric and volumetric parts may be accomplished by writing
s = I
dev
σ = I
dev
D = I
dev
(λD
λ
+ µD
µ
) (6.78)
and
p m = I
vol
D = I
vol
(λD
λ
+ µD
µ
) (6.79)
If we use the properties of the moduli multiplied by the projectors, the above equations
reduce to
s = µD
dev
= µD
µ
e = µD
µ
(B
dev
)
β
u
β
(6.80)
and
p m = (λ +
2
3
µ) D
λ
= KD
λ
= Km(m
T
) = Kmθ (6.81)
Thus, the pressure constitutive equation is
p = K θ (6.82)
Noting that the volumetric strain may be computed from
θ = b
β
u
β
(6.83)
the pressure for the displacement model may be computed from
p = Kb
β
u
β
(6.84)
We recall from Section 6.2 that
P
α
=
_

e
(B
T
dev
)
α
s dΩ +
_

e
b
T
α
p dΩ (6.85)
Using the above deﬁnitions and identities the internal force vector may be written as
P
α
=
_

e
µ(B
T
dev
)
α
D
µ
(B
dev
)
β
dΩu
β
+
_

e
Kb
α
b
T
β
dΩu
β
(6.86)
CHAPTER 6. DISPLACEMENT METHODS 41
and, thus, for isotropic linear elasticity, the stiﬀness matrix may be deduced as the
sum of the deviatoric and volumetric parts
K
αβ
= (K
dev
)
αβ
+ (K
vol
)
αβ
(6.87)
where
(K
dev
)
αβ
=
_

e
µ(B
T
dev
)
α
D
µ
(B
dev
)
β
dΩ =
_

e
µB
T
α
D
dev
B
β
dΩ (6.88)
and
(K
vol
)
αβ
=
_

e
Kb
α
b
T
β
dΩ =
_

e
KB
T
α
D
λ
B
β
dΩ (6.89)
6.7 Numerical Integration
Generally the computation of integrals for the ﬁnite element arrays is performed us-
ing numerical integration (i.e., quadrature). The use of the same quadrature for each
part of the stress divergence terms given above (in P and K) leads to a conventional
displacement approach for numerically integrated ﬁnite element developments. The
minimum order quadrature which produces a stiﬀness with the correct rank (i.e., num-
ber of element degree-of-freedoms less the number of rigid body modes) will be called
a standard or full quadrature (or integration). The next lowest order of quadrature is
called a reduced quadrature. Alternatively, use of standard quadrature on one term and
A typical integral is evaluated by ﬁrst transforming the integral onto a natural coordi-
nate space
_

e
f(x) dΩ =
_
2
f(x(ξ)) j(ξ) dξ (6.90)
where
_
2
denotes integration over the natural coordinates ξ, dξ denotes dξ
1

2
in
2-dimensions, and j(ξ) is the determinant of the jacobian transformation
J(ξ) =
∂x
∂ξ
(6.91)
Thus
j(ξ) = det J(ξ) (6.92)
The integrals over 2 are approximated using a quadrature formula, thus
_
2
f(x(ξ)) j(ξ) dξ ≈
L

l=1
f(x(ξ
l
)) j(ξ
l
) w
l
(6.93)
CHAPTER 6. DISPLACEMENT METHODS 42
where ξ
l
and w
l
elements in three dimensions and quadrilateral elements in two dimensions, the inte-
gration is generally carried out as a product of one-dimensional Gaussian quadrature.
Thus, for 2-dimensions,
_
2
g(ξ) dξ =
_
1
−1
_
1
−1
g(ξ) dξ
1

2
(6.94)
and for 3-dimensions
_
2
g(ξ) dξ =
_
1
−1
_
1
−1
_
1
−1
g(ξ) dξ
1

2

3
(6.95)
Using quadrature, the stress divergence is given by
P
α
=
L

l=1
B
α

l
)
T
σ(ξ
l
) j(ξ
l
) w
l
(6.96)
and the stiﬀness matrix is computed by quadrature as
K
αβ
=
L

l=1
B
α

l
)
T
D(ξ
l
)B
β

l
) j(ξ
l
) w
l
(6.97)
Similar expressions may be deduced for each of the terms deﬁned by the deviatoric/volu-
metric splits. The use of quadrature reduces the development of ﬁnite element arrays
to an algebraic process involving matrix operations. For example, the basic algorithm
to compute the stress divergence term is given by:
1. Initialize the array P
α
2. Loop over the quadrature points, l
• Compute j(ξ
l
) w
l
= c
• Compute the matrix in the integrand, (e.g., B
α

l
)
T
σ
l
= A
α
).
• Accumulate the array, e.g.,
P
α
← P
α
+ A
α
c (6.98)
3. Repeat step 2 until all quadrature points in element are considered.
Additional steps are involved in computing the entries in each array. For example, the
determination of B
α
requires computation of the derivatives of the shape functions,
N
α,i
, and computation of σ
l
requires an evaluation of the constitutive equation at the
quadrature point. The evaluation of the shape functions is performed using a shape
function subprogram. In FEAP, a shape function routine for 2 dimensions is called
shp2d and is accessed by the call
CHAPTER 6. DISPLACEMENT METHODS 43
call shp2d( xi, xl, shp, xsj, ndm, nel, ix, flag)
where
xi natural coordinate values (ξ
1
, ξ
2
point (input)
xl array of nodal coordinates for element
(xl(ndm,nen)) (input)
shp array of shape functions and derivatives
(shp(3,nen)) (output)
xsj jacobian determinant at quadrature point
(output)
ndm spatial dimension of problems (input)
nel number of nodes on element (between 3 and
9) (input)
ix array of global node numbers on element
(ix(nen)) (input)
flag ﬂag, if false derivatives returned with
respect to x (input); if true
derivatives returned with respect to ξ.
The array of shape functions has the following meanings:
shp(1,A) is N
A,1
shp(2,A) is N
A,2
shp(3,A) is N
A,3
The quadrature points may be obtained by a call to int2d:
call int2d( l, lint, swg )
where
l -number of quadrature points in each direction
(input).
lint -total number of quadrature points (output).
swg -array of natural coordinates and weights (output).
The array of points and weights has the following meanings:
CHAPTER 6. DISPLACEMENT METHODS 44
swg(1,L) is ξ
1,L
swg(2,L) is ξ
2,L
swg(3,L) is w
L
Using the above two utility subprograms a 2-dimensional formulation for displacement
(or mixed) ﬁnite element method can be easily developed for FEAP. An example, is
element elmt01 which is given in Appendix B.
Chapter 7
Mixed Finite Element Methods
7.1 Solutions using the Hu-Washizu Variational The-
orem
A ﬁnite element formulation which is free from locking at the incompressible or nearly
incompressible limit may be developed from a mixed variational approach. In the work
considered here we use the Hu-Washizu variational principle, which we recall may be
written as
Π(u, σ, ) =
1
2
_

T
D dΩ −
_

T
D
0
dΩ
+
_

σ
T
(∇
(s)
u − ) dΩ −
_

u
T
b
v
dΩ

_
Γ
t
u
T
¯
t dΓ −
_
Γ
u
t
T
(u − ¯ u) dΓ = Stationary (7.1)
In the principle, displacements appear up to ﬁrst derivatives, while the stresses and
strains appear without any derivatives. Accordingly, the continuity conditions we may
use in ﬁnite element approximations are C
0
for the displacements and C
−1
for the
stresses and strains (a C
−1
function is one whose ﬁrst integral will be continuous).
Appropriate interpolations for each element are thus
u(ξ) = N
I
(ξ) u
I
(t) (7.2)
σ(ξ) = φ
α
(ξ) σ
α
(t) (7.3)
and
(ξ) = ψ
α
(ξ)
α
(t) (7.4)
45
CHAPTER 7. MIXED FINITE ELEMENT METHODS 46
where φ
α
(ξ) and ψ
α
(ξ) are interpolations which are continuous in each element but
may be discontinuous across element boundaries.
1
The parameters σ
α
and
α
are not
necessarily nodal values and, thus, may have no direct physical meaning.
If, for the present, we ignore the integral for the body force, and the traction and
displacement boundary integrals and consider an isotropic linear elastic material, the
remaining terms may be split into deviatoric and volumetric parts as
Π(u, σ, ) =
1
2
_

µ
T
D
dev
dΩ −
_

µ
T
D
dev

0
dΩ
+
_

s
T
[e(u) − e] dΩ (7.5)
+
1
2
_

K θ
2
dΩ −
_

K θ θ
0
dΩ +
_

p[θ(u) − θ] dΩ
where
e(u) = I
dev

(s)
u (7.6)
and
θ(u) = tr(∇
(s)
u) = ∇· u (7.7)
are the strain-displacement relations for the deviatoric and volumetric parts, respec-
tively.
Constructing the variation for the above split leads to the following Euler equations
which hold in the domain Ω:
1. Balance of Momentum
∇ · (s + 1p) + b
v
= 0 (7.8)
which is also written as
div(s + 1p) + b
v
= 0 (7.9)
2. Strain-Displacement equations
e(u) − e = 0 (7.10)
θ(u) − θ = 0 (7.11)
3. Constitutive equations
µD
dev
− s = 0 (7.12)
K θ − p = 0 (7.13)
1
Strictly, φ
α
and ψ
α
need only be piecewise continuous in each element; however, this makes the
evaluation of integrals over each element more diﬃcult and to date is rarely used.
CHAPTER 7. MIXED FINITE ELEMENT METHODS 47
In addition the boundary conditions for Γ
u
and Γ
t
are obtained.
Using the interpolations described above, the Hu-Washizu variational theorem may be
approximated by summing the integrals over each element. Accordingly,
Π(u, σ, ) ≈ Π
h
(u, σ, ) =

e
Π
e
(u, σ, ) (7.14)
If the deviatoric part is approximated by taking
e = e(u) (7.15)
for each point of Ω, this part of the problem is given as a displacement model. The
variational expression Eq. 7.5 becomes
Π(u, p, θ) =
1
2
_

µ
T
(u) D
dev
(u) dΩ −
_

µ
T
(u) D
dev

0
dΩ
+
1
2
_

K θ
2
dΩ −
_

K θ θ
0
dΩ
+
_

p [θ(u) − θ] dΩ (7.16)
which may be split into integrals over the elements as
Π(u, p, θ) ≈ Π
h
(u, p, θ) =

e
Π
e
(u, p, θ) (7.17)
A mixed approximation may now be used to describe the pressure and the volume
change in each element. Accordingly, we assume
p(ξ) = φ
α
(ξ) p
α
(t) = φ(ξ) p (7.18)
θ(ξ) = φ
α
(ξ) θ
α
(t) = φ(ξ) θ (7.19)
where it is noted that the same approximating functions are used for both p and theta.
If the material is isotropic linear elastic, the use of the same functions will permit an
exact satisfaction of the constitutive equation, Eq. 7.13 at each point of the domain of
an element. For other situations, the constitutive equation may be approximately sat-
isﬁed. Recall that the strain-displacement equations for a ﬁnite element approximation
are given by
(u) = B
I
u
I
(7.20)
Thus, the ﬁnite element approximation for the mixed formulation may be written as
Π
e
(u, p, θ) = (u
I
)
T
_
1
2
_

e
µB
T
I
D
dev
B
J
dΩu
J

_

e
µB
T
I
D
dev

0
dΩ
_
+ θ
T
_
1
2
_

e

T
φdΩθ −
_

e

T
θ
0
dΩ
_
+ p
T
_ _

e
φ
T
b
J
dΩu
J

_

e
φ
T
φdΩθ
_
(7.21)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 48
If we deﬁne the following matrices:
k =
_

e

T
φdΩ (7.22)
π
0
=
_

e

T
θ
0
dΩ (7.23)
h =
_

e
φ
T
φdΩ (7.24)
g
I
=
_

e
φ
T
b
I
dΩ (7.25)
and recall that the deviatoric stiﬀness is deﬁned as
(K
dev
)
IJ
=
_

e
B
T
I
D
dev
B
J
dΩ (7.26)
and denote the eﬀects of initial deviatoric strains as
(P
0
dev
)
I
=
_

e
µB
T
I
D
dev

0
dΩ =
_

e
µB
T
I
D
µ
e
0
dΩ (7.27)
where e
0
are the deviatoric initial strains. The mixed variational terms become
Π
e
(u, p, θ) = (u
I
)
T
_
1
2
(K
dev
)
IJ
u
J
− (P
0
dev
)
I
_
+ θ
T
_
1
2
kθ − π
0
_
+ p
T
_
g
J
u
J
− hθ
¸
(7.28)
If we denote the variations of pressure and volume change as
p
η
= p + η Π (7.29)
θ
η
= θ + η
Θ
(7.30)
the ﬁrst variation of Eq. 7.28 may be written in the matrix form

e

=
_
(U
I
)
T
, Π
T
, Θ
T
¸
_
_
_
_
(K
dev
)
IJ
g
J
0
g
T
I
0 −h
0 −h k
_
_
_
_
u
J
p
θ
_
_

_
_
(P
0
dev
)
I
0
π
0
_
_
_
_
(7.31)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 49
or in variational notation as
δΠ
e
=
_
(δu
I
)
T
, δp
T
, δθ
T
¸
_
_
_
_
(K
dev
)
IJ
g
J
0
g
T
I
0 −h
0 −h k
_
_
_
_
u
J
p
θ
_
_

_
_
(P
0
dev
)
I
0
π
0
_
_
_
_
(7.32)
We note that the parameters p and θ (and their variations Π and Θ) are associated
with a single element, consequently, from the stationarity condition, the last two rows
of the above matrix expression must vanish and may be solved at the element level.
The requirement for a solution to exist is that
2
n
θ
≥ n
p
(7.33)
where n
θ
and n
p
are the number of parameters associated with the volume change and
pressure approximations, respectively. We have satisﬁed this requirement by taking
an equal number for the two approximations. Also, since we used the same functions
for the two approximations, the matrix h is square and positive deﬁnite (provided our
approximating functions are linearly independent), consequently, we may perform the
element solutions by inverting only h. The solution to Eq. 7.32 is
θ = h
−1
g
J
u
J
(7.34)
and the solution to the third row is
p = h
−1
(kθ − π
0
) (7.35)
Substitution of the above results into the ﬁrst equation gives

e

= (U
I
)
T
__
(K
dev
)
IJ
+ g
T
I
h
−1
kh
−1
g
J
¸
u
J
− (P
0
dev
)
I
− g
T
I
h
−1
π
0
_
(7.36)
Finally, by deﬁning a modiﬁed volumetric strain-displacement matrix as
¯
b
I
= h
−1
g
J
(7.37)
the above simpliﬁes to

e

= (U
I
)
T
__
(K
dev
)
IJ
+
¯
b
T
I
k
¯
b
J
_
u
J
− (P
0
dev
)
I

¯
b
T
I
π
0
¸
(7.38)
2
This is a mixed patch test requirement. See [26, Chapter 12].
CHAPTER 7. MIXED FINITE ELEMENT METHODS 50
The volumetric stiﬀness for the mixed formulation is given as
(K
vol
)
IJ
=
¯
b
T
I
k
¯
b
J
(7.39)
and the volumetric initial force by
(P
0
vol
)
I
=
¯
b
T
I
π
0
(7.40)
The stress divergence term for the mixed model formulation is computed from
P
I
=
_

e
B
T
I
(s + p m) dΩ (7.41)
where the deviatoric stress is expressed by the displacement approximation as
s = µD
dev
(B
J
u
J

0
) (7.42)
and the pressure is expressed by the mixed approximation as
p = φ(ξ) h
−1
(kθ − π
0
) (7.43)
7.2 Finite Element Solution for Mixed Formulation
The mixed ﬁnite element solution for the linear elastic problem requires selecting a set
of approximating functions for φ. The number of φ functions will aﬀect the rank of
the volumetric terms. The modiﬁed volumetric stiﬀness has a rank which is given by
rank(K
vol
) = min( rank(
¯
b), rank(k) ) (7.44)
Provided the approximations for φ are linearly independent, and the number is small
compared to the number of degrees-of-freedom on the element, the rank will normally
be that of k. For example, 4-node quadrilateral or 8-node brick elements can use a
single function
φ
1
= 1 (7.45)
for the approximating space. This gives a rank of 1 for the volumetric stiﬀness. The
requirement for the approximation is guided by the principle that: (1) we use the
minimum number of functions which make K have correct rank for a single element,
and (2) the functions produce an element which is invariant with respect to the input
data. For example, if we show that two functions are suﬃcient for a 2-dimensional
element, use of
φ
1
= 1 ; φ
2
= ξ
1
(7.46)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 51
would not be good since the element is not invariant with respect to a permutation in
the deﬁnition of ξ
1
and ξ
2
. Several alternatives are possible, one being
φ
1
= 1 ; φ
2
= ξ
1
+ ξ
2
(7.47)
another is to use 3 functions with
φ
1
= 1 ; φ
2
= ξ
1
; φ
3
= ξ
2
(7.48)
The actual functions selected must be subjected to further evaluations to decide which
best meets the objectives of the problem solution.
An algorithm to implement the above mixed model for linear elasticity where D is
constant in each element may be summarized as:
1. Initialize arrays: g
I
, h, k, π
0
. FEAP will initialize K and the element residual.
2. Loop over quadrature points, l
(a) Compute shape functions: In 2-d problems FEAP uses,
N
I

l
) = shp(3, I, l) (7.49)
N
I
, i(ξ
l
) = shp(i, I, l) (7.50)
(b) Compute the volume element times the quadrature weight
j
l
w
l
= dv(l) (7.51)
3. Loop over quadrature points, l
(a) Compute the volumetric strain matrices, g
I
and h.
4. Invert h and compute
¯
b
I
¯
b
I
= h
−1
g
I
(7.52)
5. Loop over quadrature points, l
(a) Compute strain-displacement matrix, B, and strains,

l
= [I
dev
B
I

l
) +
1
3
mφ(ξ
l
)
¯
b
I
] u
I
(7.53)
(b) Compute quadrature stresses and π
0
σ
l
= D(
l

0
) (7.54)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 52
(c) Compute the residual
R
I
= F
I

lint

l=1
B
T
I

l
) σ
l
j(ξ
l
) W
l
(7.55)
(d) Compute the deviatoric tangent, K
dev
(e) Compute the volumetric local tangent, k
6. Compute the tangent, K
K = K
dev
+
¯
b
T
k
¯
b (7.56)
7.3 Mixed Solutions for Anisotropic Linear Elastic
Materials
A more general form of the Hu-Washizu principle is needed to consider either anisotropic
linear elastic materials or inelastic materials in which there is coupling between volu-
metric and deviatoric eﬀects. In this section we construct the form of the functional
for an anisotropic linear elastic material. Accordingly, we have
σ = D[ −
0
] (7.57)
where D is a symmetric matrix in which there may be coupling between the deviatoric
and volumetric strain eﬀects. It is now assumed that a ﬁnite element solution will be
constructed in which deviatoric strains, e, are computed directly from the displace-
ments but the volumetric strain, θ, is computed from a mixed form. Accordingly,
¯ = I
dev
(u) +
1
3
mθ (7.58)
A stress may be computed from ¯ as
¯ σ = D[I
dev
((u) −
0
) +
1
3
m(θ − θ
0
)] (7.59)
where θ
0
= m
T

0
. The stress may be split into deviatoric and pressure parts as
¯ σ = ¯s + m¯ p (7.60)
where
¯s = I
dev
D[I
dev
((u) −
0
) +
1
3
m(θ − θ
0
)] (7.61)
and
¯ p =
1
3
m
T
D[I
dev
((u) −
0
) +
1
3
m(θ − θ
0
)] (7.62)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 53
If we deﬁne
D
dev
= I
dev
DI
dev
(7.63)
d =
1
3
I
dev
Dm (7.64)
d
vol
=
1
9
m
T
Dm (7.65)
s
0
= − D
dev

0
− dθ
0
(7.66)
and
p
0
= − d
T

0
− d
vol
θ
0
(7.67)
then the stress may be written as
¯ σ = D
dev
(u) + dθ +
1
3
m(d
T
(u) + d
vol
θ) + s
0
+ mp
0
(7.68)
This form of the stress may be multiplied by the virtual ¯ and integrated over the
domain to obtain part of the variational equation associated with the strain energy.
Subsequently, adding the terms associated with the mixed volumetric pressure and
volume change Vainberg’s theorem may be used to obtain a variational theorem. Al-
ternatively, the stress and strain splits may be substituted into 7.1. The result is
Π(u, p, θ) =
1
2
_

_
(u) θ
¸
_
D
dev
d
d
T
d
vol
_ _
(u)
θ
_
dΩ
+
_

((u) s
0
+ θ p
0
) dΩ
+
_

p [θ(u) − θ] dΩ + Π
ext
(7.69)
This form of the variational principle is equivalent to 7.16 which was deduced for
isotropic materials. The added terms in 7.69 are all associated with d which deﬁnes a
coupling between deviatoric and volumetric strains. For isotropy d is zero.
If we introduce ﬁnite element interpolations using standard displacement interpolation
together with the pressure and volume interpolations given by 7.18 and 7.19, the ﬁrst
variation of 7.69 for a single element is
δΠ
e
=
_
δˆ u
T
I
δ
ˆ
θ
T
_
_

e
__
B
T
I
D
dev
B
J
B
T
I

φ
T
d
T
B
J
φ
T
d
vol
φ
_ _
ˆ u
J
ˆ
θ
_
+
_
B
T
I
s
0
φ
T
p
0
_ _
dΩ + δˆ p
T
_

e
φ
T
_
b
J
−φ
¸
dΩ
_
ˆ u
J
ˆ
θ
_
+
_
δˆ u
T
I
δ
ˆ
θ
T
_
_

e
_
b
T
I
φ
−φ
T
φ
_
dΩ ˆ p + δI
ext
(7.70)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 54
The variational equation 7.70 may be expressed in terms of stresses by substituting the
interpolations into 7.61 and 7.62 resulting in
δΠ
e
=
_
δˆ u
T
I
δ
ˆ
θ
T
_
_

e
_
B
T
I
¯s
φ
T
¯ p
_
dΩ
+ δˆ p
T
_

e
φ
T
_
b
J
−φ
¸
dΩ
_
ˆ u
J
ˆ
θ
_
+
_
δˆ u
T
I
δ
ˆ
θ
T
_
_

e
_
b
T
I
φ
−φ
T
φ
_
dΩ ˆ p + δI
ext
(7.71)
Since the interpolations for the pressure and volume change are associated with a single
element it is possible to solve for their parameters at the element level. Accordingly,
the multiple of δˆ p yields
_

e
φ
T
b
J
dΩ ˆ u
J
=
_

e
φ
T
φdΩ
ˆ
θ = h
ˆ
θ (7.72)
which yields
ˆ
θ =
¯
b
I
ˆ u
I
= h
−1
g
I
ˆ u
I
(7.73)
where h and g
I
are as deﬁned in 7.24 and 7.25, respectively. similarly, the equation
multiplying δ
ˆ
θ yields the equation
_

e
φ
T
¯ p dΩ =
_

e
φ
T
φdΩ ˆ p = h ˆ p (7.74)
Using these results, the ﬁrst integral in the variational equation deﬁnes the stress
divergence terms
δΠ
σ
= δˆ u
T
I
__

B
T
I
¯s dΩ +
¯
b
T
I
_

φ
T
¯ p dΩ
_
(7.75)
which upon use of the deﬁnitions for the mixed pressure, p, and the mixed volumetric
strain displacement equation,
¯
b
I
, yields
δΠ
σ
= δˆ u
T
I
_

B
T
I
[¯s + mp] dΩ (7.76)
The stress of the mixed method is deﬁned as
σ = ¯s + mp (7.77)
and, in general, is not equal to ¯ σ. The stress ¯ σ, however, is the stress which is computed
from the constitutive equation for each material. Thus, when we later consider other
material models (e.g., viscoelasticity, plasticity, etc.) the eﬀective material moduli are
the ones computed by linearizing the constitutive equation expressed in terms of the
CHAPTER 7. MIXED FINITE ELEMENT METHODS 55
¯ σ stresses. The residual for a ﬁnite element formulation is most eﬃciently computed
from the mixed stress and we note the result is identical to the form of the standard
displacement model except for the stress expression used.
The tangent matrix may be expressed in terms of the displacements alone by writing
the variational equation 7.70 as
δΠ
e
=
_
δˆ u
T
I
δ
ˆ
θ
T
δˆ p
T
_
_
_
(K
dev
)
IJ
k
I
g
J
k
T
J
k
vol
−h
g
T
I
−h 0
_
_
_
_
ˆ u
I
ˆ
θ
ˆ p
_
_
+ δΠ
0
+ δΠ
ext
(7.78)
Using the solutions to 7.72 and 7.74 the dependence of 7.78 on
ˆ
θ and ˆ p may be
eliminated to give
δΠ
e
= δˆ u
T
I
¯
K
IJ
ˆ u
J
+ δΠ
0
+ δΠ
ext
(7.79)
where
¯
K
IJ
= (K
dev
)
IJ
+ k
I
¯
b
J
+
¯
b
T
I
k
T
J
+
¯
b
T
I
k
vol
¯
b
J
(7.80)
The algorithm for the development of a mixed element based upon the above may be
summarized as:
1. Numerical integration of strain matrices
(a) Compute φ = [1, ξ
1
, ξ
2
, · · · ] (for the 4-node element φ = 1
(b) Compute arrays
h =
_

e
φ
T
φdΩ (7.81)
g
I
=
_

e
φ
T
b
J
dΩ (7.82)
2. Mixed volumetric strain displacement matrix
(a) Compute
¯
b
I
= h
−1
g
I
3. Constitution computation for each quadrature point
(a) Compute
= B
I
ˆ u
I
(7.83)
θ = φ(ξ)
¯
b
I
ˆ u
I
(7.84)
¯ = I
dev
+
1
3
mθ (7.85)
¯ σ = D
[
¯ −
0
] (7.86)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 56
¯ p =
1
3
m
T
¯ σ (7.87)
¯ π =
_

e
φ
T
¯ p dΩ (7.88)
4. Mixed Pressure
(a) Compute p = φ(ξ) h
−1
¯ π
5. Residual and Stiﬀness Integrals
(a) Compute mixed stress σ = I
dev
¯ σ + mp
(b) Compute
R
σ
I
= −
_

e
B
T
I
σdΩ (7.89)
(K
dev
)
IJ
=
_

e
B
T
I
D
dev
B
d
J
Ω (7.90)
k
I
=
_

e
B
T
I
dφdΩ (7.91)
k
vol
=
_

e
φ
T
d
vol
φdΩ (7.92)
6. Stiﬀness assembly
(a) Compute
¯
K
IJ
= (K
dev
)
IJ
+ k
I
¯
b
J
+
¯
b
T
I
k
T
J
+
¯
b
T
I
k
vol
¯
b
J
(7.93)
7.4 Hu-Washizu Variational Theorem: General Prob-
lems
The ﬁnite element approximation for the mixed formulation of a general linear elastic
material (i.e., anistotropic behavior) may be written for a typical element as
Π
e
(u, p, θ) =
1
2
_

e

T
D dΩ −
_

e

T
D
0
dΩ
+ p (∇ · u − θ) dΩ (7.94)
Using the approximations introduced for the isotropic model for the displacement and
mixed volume change gives
= I
dev
B
I
u
I
+
1
3
mφ(ξ) θ (7.95)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 57
which when introduced into the variational theorem gives
Π
e
(u, p, θ) =
1
2
_

e
_
I
dev
B
I
u
I
+
1
3
mφ(ξ) θ
_
T
D
_
I
dev
B
I
u
I
+
1
3
mφ(ξ) θ
_
dΩ −
_

e
_
I
dev
B
I
u
I
+
1
3
mφ(ξ) θ
_
T
D
0
dΩ
+ p
T
_ _

e
φ
T
b
J
dΩu
J

_

e
φ
T
φdΩθ
_
(7.96)
For symmetric D, we can deﬁne the following matrices:
(K
dev
)
IJ
=
_

e
B
T
I
I
dev
DI
dev
B
J
dΩ (7.97)
(K
co
)
J
=
1
3
_

e
φ
T
m
T
DI
dev
B
J
dΩ (7.98)
k =
1
9
_

e
φ
T
m
T
DmφdΩ (7.99)
π
0
=
1
3
_

e
φ
T
mD
0
dΩ (7.100)
h =
_

e
φ
T
φdΩ (7.101)
g
I
=
_

e
φ
T
b
I
dΩ (7.102)
and denote the eﬀects of initial deviatoric strains as
(P
0
dev
)
I
=
_

e
B
T
I
I
dev
D
0
dΩ (7.103)
The mixed variational terms become
Π
e
(u, p, θ) =
1
2
[(u
I
)
T
(K
dev
)
IJ
u
J
+ 2 θ
T
(K
co
)
J
u
J
+ θ
T
kθ]
− (u
I
)
T
(P
0
dev
)
I
− θ
T
π
0
+ p
T
g
I
u
I
− p
T
hθ (7.104)
The ﬁrst variation of Eq. 7.104 may be written in the matrix form

e

=
_
(U
I
)
T
, Π
T
, Θ
T
¸
_
_
_
_
(K
dev
)
IJ
g
I
(K
co
)
I
g
T
J
0 −h
(K
T
co
)
J
−h k
_
_
_
_
u
J
p
θ
_
_

_
_
(P
0
dev
)
I
0
π
0
_
_
_
_
(7.105)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 58
Recall that the terms which multiply the variations in pressure, Π, and the variation in
the volume change, Θ, are associated with individual elements, and, thus, the second
row of Eq. 7.105 may be solved at the element level to give the parameters for the
volume change, θ, as
θ = h
−1
g
J
u
J
(7.106)
and the solution to the third row is
p = h
−1
[(K
co
)
J
u
J
+ kθ − π
0
] (7.107)
Deﬁning a modiﬁed volumetric strain-displacement matrix as
¯
b
I
= h
−1
g
J
(7.108)
Substitution of the above results into the ﬁrst equation gives

e

= (U
I
)
T
__
(K
dev
)
IJ
+
¯
b
T
I
(K
co
)
J
+ (K
T
co
)
I
¯
b
J
+
¯
b
T
I
k
¯
b
J
¸
u
J
− (P
0
dev
)
I

¯
b
T
I
π
0
_
(7.109)
Thus, the stiﬀness matrix for the general anisotropic linear elastic formulation is given
by
K
IJ
= (K
dev
)
IJ
+
¯
b
T
I
(K
co
)
J
+ (K
T
co
)
I
¯
b
J
+
¯
b
T
I
k
¯
b
J
(7.110)
This operation may be performed after all the integrals over the element are evaluated.
The matrices which involve the elastic moduli may be simpliﬁed by deﬁning some
reduced terms. Accordingly, we let
3
d =
1
3
Dm (7.112)
Also, deﬁne
d
vol
=
1
9
m
T
Dm =
1
3
m
T
d (7.113)
Then
DI
dev
= D − dm
T
(7.114)
or
I
dev
D = D − md
T
(7.115)
3
If D is not symmetric, equations Eq.7.112 through Eq.7.117 must be modiﬁed. Essentially, this
requires a computation of two d terms as
d
R
= Dm ; d
L
= D
T
m (7.111)
and using these in the remaining equations instead of d (note, when D is symmetric the d
R
and d
L
terms are equal).
CHAPTER 7. MIXED FINITE ELEMENT METHODS 59
which gives
1
3
I
dev
Dm =
1
3
(D − md
T
) m = d − d
vol
m = d
dev
(7.116)
Finally, the deviatoric part of the modulus is now deﬁned in terms of the above as
D
dev
= I
dev
DI
dev
= D − dm
T
− md
T
+ d
vol
mm
T
(7.117)
For isotropy, the above expressions reduce to:
d
vol
= K (7.118)
d =
_
K K K 0 0 0
¸
T
(7.119)
d
dev
= 0 (7.120)
and
D
dev
=
1
3
µ
_
¸
¸
¸
¸
¸
¸
_
4 −2 −2 0 0 0
−2 4 −2 0 0 0
−2 −2 4 0 0 0
0 0 0 3 0 0
0 0 0 0 3 0
0 0 0 0 0 3
_
¸
¸
¸
¸
¸
¸
_
(7.121)
The matrices for the mixed treatment of the symmetric D anisotropic case are com-
puted as:
(K
dev
)
IJ
=
_

e
B
T
I
D
dev
B
J
dΩ (7.122)
(K
co
)
J
=
_

e
φ
T
d
T
dev
B
J
dΩ (7.123)
and
k =
_

e
d
vol
φ
T
φdΩ (7.124)
The matrix for the initial strains is computed as
π
0
=
_

e
φ
T
d
T

0
dΩ (7.125)
which is a 1 ×6 vector.
CHAPTER 7. MIXED FINITE ELEMENT METHODS 60
7.4.1 Example: Interpolations linear for u and constant φ
As an example, we consider the case where the set of shape functions for the displace-
ments is the tri-linear interpolation
N
I
(ξ) =
1
8
(1 + ξ
I
1
ξ
1
) (1 + ξ
I
2
ξ
2
) (1 + ξ
I
3
ξ
3
) (7.126)
where ξ
I
i
are the values of the natural coordinates at the I-node. The interpolation for
the pressure (and volume change) is constant
φ
1
= 1 (7.127)
This element is often called B1-P0 (order 1 interpolations for the brick element, order 0
for the pressure/volume change). In 2-dimensions the element is Q1-P0, for the order 1
quadrilateral. Higher order elements are also deﬁned, for example, the Q2-P1 element
uses quadratic interpolation for displacements (Lagrange interpolations) and linear for
the pressure with
φ(ξ) =
_
1 ξ
1
ξ
2
¸
(7.128)
Alternatively, it is possible to use the interpolations
φ(ξ) =
_
1 x
1
(ξ) x
2
(ξ)
¸
(7.129)
The matrices for the B1-P0 (or Q1-P0) element reduce to
(K
co
)
J
=
_

e
d
T
dev
B
J
dΩ (7.130)
which is a column vector (of size 1 ×24). For isotropy, this matrix is zero. The volume
stiﬀness becomes:
k =
_

e
d
vol
dΩ (7.131)
which is a 1 ×1 matrix and for constant d
vol
becomes
k = k
vol
= d
vol

e
(7.132)
where Ω
e
is the volume of the element. For isotropy k
vol
is the bulk modulus times the
element volume. The other matrices in the stiﬀness are
h = h =
_

e
dΩ = Ω
e
(7.133)
and
g
I
=
_

e
b
I
dΩ (7.134)
CHAPTER 7. MIXED FINITE ELEMENT METHODS 61
which gives the modiﬁed volumetric strain-displacement equation
¯
b
I
=
1

e
g
I
=
1

e
_

e
b
I
dΩ (7.135)
The initial strain term is computed as
π
0
=
_

e
d
T

0
dΩ (7.136)
For the mixed element the internal force is computed using
P
I
=
_

e
B
T
I
(s(u) + p m) dΩ (7.137)
where the deviatoric part of the stress is computed from the displacement form, and
the pressure is computed from the mixed form
p =
1

e
_
[(K
co
)
J
+ k
vol
¯
b
J
] u
J
− π
0
_
(7.138)
Chapter 8
Enhanced Strain Mixed Method
8.1 Hu-Washizu Variational Theorem for Linear Elas-
ticity
An alternative to the mixed ﬁnite element method discussed previously is given by
the enhanced strain method [16]. The enhanced strain method is related to earlier
works which utilized incompatible displacement modes; however, the method does not
have the deﬁciencies which are present in the earlier works. Enhanced strains provide
great ﬂexibility in designing accurate ﬁnite element models for problems which have
constraints or other similar types of diﬃculties. In the enhanced strain method we
again use the Hu-Washizu variational principle, which we recall may be written for
linear elasticity as
Π(u, σ, =
1
2
_

T
D dΩ −
_

T
D
0
dΩ
+
_

σ
T
(∇
(s)
u − ) dΩ −
_

u
T
b
v
dΩ

_
Γ
t
u
T
¯
t dΓ −
_
Γ
u
t
T
(u − ¯ u) dΓ = Stationary (8.1)
The strain tensor is expressed as an additive sum of the symmetric gradient of the
displacement vector, ∇
(s)
u, and the enhanced strains, ˜, and written as
(u, ˜) = ∇
(s)
u + ˜(ξ) (8.2)
62
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 63
If we again ignore the integral for the body force and the traction and displacement
boundary integrals, upon use of Eq. 8.2 the remaining terms become
Π(u, σ, ˜) =
1
2
_

(∇
(s)
u + ˜)
T
D(∇
(s)
u + ˜) dΩ

_

(∇
(s)
u + ˜)
T
D
0
dΩ −
_

σ
T
˜ dΩ (8.3)
Introducing the variations for each function as
u
η
= u + η U (8.4)
σ
η
= σ + η S (8.5)
and
˜
η
= ˜ + η
˜
E (8.6)
the variation for the above enhanced principle is given by

=
_

(∇
(s)
U)
T
D(∇
(s)
u + ˜ −
0
) dΩ
+
_

˜
E
T
[D(∇
(s)
u + ˜ −
0
) − σ] dΩ −
_

S
T
˜ dΩ (8.7)
After integration by parts of the variation of the displacement gradient term (and also
considering the body force term), the following Euler equations are obtained for the
domain Ω:
1. Balance of momentum
div [D(∇
(s)
u + ˜ −
0
)] + b
v
= 0 (8.8)
2. Strain-displacement equations on the enhanced modes
˜ = 0 (8.9)
3. Constitutive equations
D(∇
(s)
u + ˜ −
0
) − σ = 0 (8.10)
In addition the boundary conditions for Γ
u
and Γ
t
are obtained. We note Eq. 8.9
implies that, at the solution, the enhanced strains must vanish. Substitution of this
result into the remaining equations yields the appropriate displacement equations of
equilibrium and constitutive equation for linear elasticity, from 8.8 and 8.10, respec-
tively. While the enhanced strains vanish pointwise at a solution, in an approximate
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 64
scheme based upon the enhanced strain method this is not the case. The enhanced
strains will only vanish in some integral sense over each element, just as the balance of
momentum and constitutive equations are approximated by ﬁnite element solutions.
In the enhanced strain principle, displacements appear up to ﬁrst derivatives, while
the stresses and enhanced strains appear without any derivatives. Accordingly, the
continuity conditions we may use in ﬁnite element approximations again are C
0
for the
displacements and C
−1
for the stresses and enhanced strains. Appropriate interpola-
tions for displacements and stresses are the same as given previously for each element,
and are thus
u(ξ) = N
I
(ξ) u
I
(t) (8.11)
and
σ(ξ) = φ
α
(ξ) σ
α
(t) (8.12)
respectively. The choice of appropriate approximating functions φ
α
will be aﬀected by
the strain approximation, as will be shown below. From Eq. 8.2, the strain approxi-
mations are now given by
(u, ˜) = ∇
(s)
u + ˜(ξ) (8.13)
where the approximations for the enhanced strains are assumed as
˜(ξ) = ψ
α
(ξ) ˜
α
(t) (8.14)
It should be noted that diﬀerent interpolations are introduced for the stress and the
enhanced strain terms.
Using the interpolations described above, the Hu-Washizu variational theorem may be
approximated by summing the integrals over each element. Accordingly,
Π(u, σ, ) = Π
h
(u, σ, ) ≈

e
Π
e
(u, σ, ) (8.15)
The variational expression in each element becomes
Π
e
(u, σ, ˜) =
1
2
_

e
(∇
(s)
u + ˜)
T
D(∇
(s)
u + ˜) dΩ

_

e
(∇
(s)
u + ˜)
T
D
0
dΩ −
_

e
σ
T
˜ dΩ (8.16)
Substituting the approximations for displacements, stresses, and enhanced strains and
replacing with
u
I
η
= u
I
+ η U
I
(8.17)
σ
α
η
= σ
α
+ η S
α
(8.18)
and
˜
α
η
= ˜
α
+ η
˜
E
α
(8.19)
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 65
gives the ﬁrst variation in each element as

e

=
_
(U
I
)
T
, (
˜
E
α
)
T
, (S
α
)
T
_
_
_
_
_
K
IJ
˜
Γ
βI
0
˜
Γ
T
αJ
˜
H
αβ
Q
αβ
0 Q
T
βα
0
_
_
_
_
u
J
˜
β
σ
β
_
_

_
_
P
0
I
˜ π
0
α
0
_
_
_
_
(8.20)
where
K
IJ
=
_

e
B
T
I
DB
J
dΩ (8.21)
˜
Γ
αJ
=
_

e
ψ
T
α
DB
J
dΩ (8.22)
˜
H
αβ
=
_

e
ψ
T
α

β
dΩ (8.23)
Q
αβ
=
_

e
φ
T
α
ψ
β
dΩ (8.24)
P
0
I
=
_

e
B
T
I
D
0
dΩ (8.25)
˜ π
0
α
=
_

e
ψ
T
α
D
0
dΩ (8.26)
The discrete Euler equation generated by the third equation of Eq. 8.20 is given in
each element by
Q
αβ

β
= 0 (8.27)
There are at least three possible ways this may be used:
1. The
β
are zero, which is not a useful result.
2. The interpolations for φ
α
are orthogonal to the interpolations ψ
β
, which means
that
Q
αβ
= 0 (8.28)
which is the solution to be followed here. This is not perfect since we will not
obtain a method to compute the σ
β
directly from the variational formulation.
3. A combination of options (a) and (b).
For a formulation which satisﬁes Eq. 8.28, the variational equations in each element
reduce to

e

=
_
(U
I
)
T
, (
˜
E
α
)
T
_
__
K
IJ
˜
Γ
βI
˜
Γ
T
αJ
˜
H
αβ
_
_
u
J
˜
β
_

_
P
0
I
˜ π
0
α
_
_
(8.29)
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 66
Since the interpolations for the enhanced strains are assumed for each element inde-
pendently, the second of Eq. 8.29 may be solved at the element level giving
˜
β
= (
˜
H
αβ
)
−1
_
˜ π
0
α

˜
Γ
αJ
u
J
_
(8.30)
which may be substituted into the ﬁrst equation to give

e

= (U
I
)
T
˜
K
IJ
u
J

˜
P
0
I
(8.31)
where
˜
K
IJ
= K
IJ

˜
Γ
T
βI
(
˜
H
αβ
)
−1
˜
Γ
αJ
(8.32)
and
˜
P
0
I
= P
0
I

˜
Γ
T
βI
(
˜
H
αβ
)
−1
˜ π
0
α
(8.33)
8.2 Stresses in the Enhanced Method
Since the stresses based upon the mixed approximation are no longer available an
alternative is needed for computations. Simo and Rifai suggest using a least square
projection technique to obtain the stresses; however, the stresses which are directly
utilized in the variational equation Eq. 8.7 may be deduced as
˜ σ = D(∇
(s)
u + ˜ −
0
) (8.34)
In subsequent development we shall use these stresses for all calculations of arrays, as
well as, for outputs and stress projections to nodes. Thus, the variation in each element
may be written

e

¸
¸
¸
¸
η=0
=
_

e
(∇
(s)
U)
T
˜ σdΩ +
_

e
˜
E
T
˜ σdΩ (8.35)
It is noted that the orthogonality condition
_

e
S
T
˜ dΩ = 0 (8.36)
has been incorporated in the above variation.
With the above description, the residual in each element becomes:
R
I
= F
I

_

e
B
T
I
˜ σdΩ (8.37)
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 67
for the contribution in the element to the global residual. Similarly, the residual for
the enhanced modes is computed from
˜
R
α
= −
_

e
ψ
T
α
˜ σdΩ (8.38)
Note that the residual for the enhanced modes will vanish at a solution since it belongs
to a single element.
8.3 Construction of Enhanced Modes
The construction of the enhanced modes depends crucially on the orthogonality require-
ment being satisﬁed for each element. Based upon the study of the shape functions
using the alternative representation we recall that the gradient of the displacement
involves a constant part and a part which depends only on the determinant of the jaco-
bian matrix, j(ξ), the constant part of the jacobian matrix, J
0
coordinates, ξ. Accordingly, it is useful to express the enhanced strains in a similar
form. Using tensor notation we introduce the representations
˜ =
j
0
j(ξ)
J
−T
0
˜
E(ξ) J
−1
0
(8.39)
which represents a transformation of the local enhanced strains,
˜
E, expressed on the
bi-unit square to the global strains, ˜, using the transformation deﬁned at the element
center. The weighting by the jacobian determinant terms is motivated by the gradient
of the shape functions. Similarly, a transformation of the local stresses, Σ, on the
bi-unit square element to the global stresses, σ, is given by
σ = J
0
Σ(ξ) J
T
0
(8.40)
These transformations have the property that
tr(σ˜) =
j
0
j(ξ)
tr(Σ
˜
E) (8.41)
The transformations may also be written in matrix form as
˜ =
j
0
j(ξ)
F
−1
0
˜
E(ξ) (8.42)
σ =
j
0
j(ξ)
F
T
0
Σ(ξ) (8.43)
where for 2-dimensional problems
˜
E
T
=
_
˜
E
11
˜
E
22
˜
E
33
2
˜
E
12
¸
(8.44)
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 68
Σ
T
=
_
Σ
11
Σ
22
Σ
33
Σ
12
¸
(8.45)
and ˜ and σ have similar ordering. The matrix F
0
is given by
F
0
=
_
¸
¸
_
(J
11
0
)
2
J
21
0
J
12
0
0 2 J
11
0
J
12
0
J
12
0
J
21
0
(J
22
0
)
2
0 2 J
21
0
J
22
0
0 0 1 0
J
11
0
J
21
0
J
12
0
J
22
0
0 J
11
0
J
22
0
+ J
12
0
J
21
0
_
¸
¸
_
(8.46)
In matrix form Eq. 8.41 may be written as
σ
T
˜ =
j
0
j(ξ)
Σ
T
˜
E (8.47)
The integral over the element becomes
_

e
σ
T
˜ dΩ = j
0
_
2
Σ
T
˜
Ed2 = 0 (8.48)
Thus, the satisfaction of the orthogonality condition may be accomplished by con-
structing the interpolations in the natural coordinate system and transforming to the
global frame using Eq. 8.42 and Eq. 8.43. A number of alternatives are discussed in
the paper by Simo and Rifai [16]. Here we consider the simplest form, which indeed is
identical to the modiﬁed incompatible mode formulation [22]. It should be noted how-
ever, that no ad-hoc assumptions are required in the enhanced formulation, contrary
to what is necessary when using incompatible modes.
For the simplest form, the interpolations
Σ =
_
¸
¸
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
¸
¸
_
_
¸
¸
_
Σ
1
Σ
2
Σ
3
Σ
4
_
¸
¸
_
(8.49)
for the stress and
˜
E =
_
¸
¸
_
ξ
1
0 0 0
0 ξ
2
0 0
0 0 0 0
0 0 ξ
1
ξ
2
_
¸
¸
_
_
¸
¸
_
E
1
E
2
E
3
E
4
_
¸
¸
_
(8.50)
for the enhanced strains are used. The integrals of natural coordinates over the bi-linear
(2-dimensional) element obey the following properties
_
2
ξ
p
i
d2 =
_
0 if p odd
4
p+1
if p even
(8.51)
Thus, the interpolations given by Eqs. 8.49 and Eqs. 8.50 contain only linear polyno-
mials in ξ and, thus, satisfy the orthogonality condition Eq. 8.48. These interpolations
have been incorporated into the element routine elmt04 which has been developed for
a linear elastic-viscoelastic material, as well as, for non-linear materials.
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 69
8.4 Non-Linear Elasticity
For a non-linear, hyperelastic material the stresses are computed from a strain energy
density function, W(), through
σ =
∂W

(8.52)
The partial derivative is understood in terms of components, where
σ
ij
=
∂W

ij
(8.53)
We note that for the linear material model discussed previously that
W() =
1
2

T
D −
T
D
0
(8.54)
For the enhanced formulation the computation of stresses is given by
˜ σ =
∂W

¸
¸
¸
¸
= ∇
(s)
u + ˜
(8.55)
In subsequent development we shall use these stresses for all calculations of arrays, as
well as, for outputs and stress projections to nodes. Thus, for the enhanced formulation
the variation in each element may be written as (see Eqs. 8.35 to 8.38)

e

=
_

e
(∇
(s)
U)
T
˜ σdΩ +
_

e
˜
E
T
˜ σdΩ (8.56)
In a manner identical to the linear elastic material, the residual in each element be-
comes:
R
I
= F
I

_

e
B
T
I
˜ σdΩ (8.57)
Similarly, the residual for the enhanced modes is computed from
˜
R
α
= −
_

e
ψ
T
α
˜ σdΩ = 0 (8.58)
We note above that at a solution the residual,
˜
R
α
, should vanish independently in each
element.
8.5 Solution Strategy: Newton’s Method
The solution to a non-linear problem is commonly computed using a sequence of linear
approximations. A popular scheme is Newton’s method, which may be summarized as:
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 70
1. Given the set of equations
f (x) = 0 (8.59)
where x are the dependent variables.
2. Construct the linear part of f about a current point x
(i)
as
f
(i+1)
≈ f
(i)
+
∂f
∂x
¸
¸
¸
¸
x=x
(i)
dx
(i+1)
= 0 (8.60)
where dx
(i+1)
is an increment of x.
3. Solve the linear problem
dx
(i+1)
= − (F
(i)
)
−1
f
(i)
; F
(i)
=
∂f
∂x
¸
¸
¸
¸
x=x
(i)
(8.61)
and update the solution as
x
(i+1)
= x
(i)
+ dx
(i+1)
(8.62)
In the above, F
(i)
, is the Jacobian or tangent matrix for the equations.
4. Repeat steps b.) and c.) until the solution converges to within a tolerance, tol.
Convergence may be assessed from
| dx
(i+1)
| < tol | x
(i+1)
| (8.63)
where | x | is the length of the vector, x.
Using Newton’s method on the set of equations deﬁned by Eq. 8.57 and Eq. 8.58 above
gives the problem
_
R
(i+1)
I
˜
R
(i+1)
α
_

_
R
(i)
I
˜
R
(i)
α
_

_
K
(i)
IJ
˜
Γ
(i)
βI
˜
Γ
(i)
αJ
T
˜
H
(i)
αβ
_
_
du
J(i+1)

β(i+1)
_
=
_
ˆ
R
(i+1)
I
0
_
(8.64)
In the above, the terms in the Jacobian are deﬁned as
K
(i)
IJ
= −
∂R
I
∂u
J
¸
¸
¸
¸
(i)
(8.65)
which expands to
K
(i)
IJ
=
_

e
B
I
∂ ˜ σ

¸
¸
¸
¸
(i)

∂u
J
dΩ =
_

e
B
I
˜
D
(i)
t
B
J
dΩ (8.66)
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 71
where
˜
D
(i)
t
=
∂ ˜ σ

¸
¸
¸
¸
(i)
(8.67)
deﬁne the tangent moduli for the material. For the non-linear elastic material
˜
D
(i)
t
=

2
W
∂ ∂
¸
¸
¸
¸
(i)
(8.68)
Similarly,
˜
Γ
(i)
αJ
=
_

e
ψ
T
α
˜
D
(i)
t
B
J
dΩ (8.69)
and
˜
H
(i)
αβ
=
_

e
ψ
T
α
˜
D
(i)
t
ψ
β
dΩ (8.70)
Since the second equation in Eq. 8.64 is complete at the element level, we may perform
a partial solution by static condensation. Accordingly,

β(i+1)
= (
˜
H
(i)
αβ
)
−1
[
˜
R
(i)
α

˜
Γ
(i)
αJ
du
J(i+1)
] (8.71)
which may be substituted into the ﬁrst equation to give
ˆ
R
(i+1)
I
=
˜
R
(i)
I

˜
K
(i)
IJ
du
J(i+1)
(8.72)
where
˜
R
(i)
I
= R
(i)
I

˜
Γ
(i)
βI
T (
˜
H
(i)
αβ
)
−1
˜
R
(i)
α
(8.73)
and
˜
K
(i)
IJ
= K
(i)
IJ

˜
Γ
(i)T
βI
(
˜
H
(i)
αβ
)
−1
˜
Γ
(i)
αJ
(8.74)
The reduced ﬁrst equations may be assembled into the global equations. Thus after
I
, the assembled equations become

e
˜
K
(i)
IJ
du
J(i+1)
=

e
˜
R
(i)
I
+ F
I
(8.75)
which may be solved for the incremental nodal displacements, du
J(i+1)
. After the solve,
the new nodal displacements are updated
u
J(i+1)
= u
J(i)
+ du
J(i+1)
(8.76)
The incremental displacements also may be substituted back into Eq. 8.71 to compute
the increments to the enhanced modes; these may then be used for the update
˜
β(i+1)
= ˜
β(i)
+ d˜
β(i+1)
(8.77)
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 72
It should be noted that these last steps may not be performed until after the element
arrays are assembled and the resulting global problem is solved for the incremental
nodal displacements. Consequently, for this algorithm, it is necessary to save the
arrays used in Eq. 8.71 for the later update of the enhanced modes. In the enhanced
element for 2-dimensional plane strain applications in FEAP, the arrays are moved into
history arrays using a pmove routine. This requires additional storage for the enhanced
formulation with respect to that needed for a displacement or a mixed B-bar type of
formulation. It is possible to modify the above algorithm such that the additional
storage is reduced to saving only the current values of the enhanced mode parameters,
˜
β(i)
.
The alternate algorithm is given by linearizing the residual,
˜
R
α
, with respect to ˜
β
only.
Accordingly, with u
J
(i) known we enter each element calculation with the enhanced
strain parameters at the values ˜
β(i−1)
and perform the following steps.
1. For k = 0 set
˜
β(i,k)
= ˜
β(i−1)
(8.78)
where a single superscript i denotes the value of ˜
β
computed in the last global
iteration.
2. Compute the linear part of
˜
R
α
as
˜
R
α
(u
J
(i), ˜
β(i,k)
) −
˜
H
(i,k)
αβ

β(i,k+1)
= 0 (8.79)
where now
˜
H
(i,k)
αβ
=
_

e
ψ
T
α
˜
D
(i,k)
t
ψ
β
dΩ (8.80)
with
˜
D
(i,k)
t
=
∂ ˜ σ

¸
¸
¸
¸

(s)
u
(i)
+ ˜
(i,k)
(8.81)
3. Solve for the increment

β(i,k+1)
= (
˜
H
(i,k)
αβ
)
−1
˜
R
(i,k)
α
(8.82)
4. Update the solution
˜
β(i,k+1)
= ˜
β(i,k)
+ d˜
β(i,k+1)
(8.83)
5. Set k ← k + 1 and repeat Steps 2. to 4. until convergence achieved (or a set
number of k-steps is completed).
6. Set
˜
β(i)
= ˜
β(i,k+1)
(8.84)
and save for the next global iteration, as well as use for subsequent steps for the
global i-iterations or to compute stresses.
CHAPTER 8. ENHANCED STRAIN MIXED METHOD 73
Vector Deﬁnition Description
u
J
(i) Current solution value
at each node, J.
∆u
J
(i) = u
J
(i) −u
J
(t
n
) Diﬀerence between current
and previous solution
du
J
(i) = u
J
(i) −u
J
(i −1) Increment from last iteration
Table 8.1: Element Local Arrays
Array Description Problems
ul(ndf,nen,1) local u
J
(i) All
ul(ndf,nen,2) local ∆u
J
(i) All
ul(ndf,nen,3) local du
J
(i) All
ul(ndf,nen,4) local ˙ u
J
(i) Transient
ul(ndf,nen,5) local ˙ u
J
(i) Transient
ul(ndf,nen,6) local ˙ u
J
(i −1) Transient
ul(ndf,nen,7) used for b.c. on u
J
(i) All
1
Table 8.2: Element Local Arrays
The only information to be stored is the ˜
β(i)
. The algorithm requires repeated com-
putation of R
(i,k)
α
and H
(i,k)
αβ
; however, using only 2 or 3 iterations generally suﬃces
(even though convergence may not be achieved for the ﬁrst few values of the i-global
iterations). Once the k-iteration is completed, linearization with respect to both u
J
and ˜
B
is performed, leading to Eq. 8.72 to Eq. 8.75 for the global steps. If the k
iteration is converged, the R
(i)
α
is zero in Eq. 8.72 to Eq. 8.75 thus simplifying slightly
the steps involved.
While the above process has been illustrated for the non-linear elastic material, it may
be directly extended to any material for which we can iteratively compute the stresses,
˜ σ
(i)
, and the tangent moduli,
˜
D
(i)
t
. In subsequent presentations we shall discuss the
construction of these steps for linear viscoelastic materials, elasto-plastic materials,
and a class of viscoplastic materials.
In FEAP, the u
J
(i) nodal displacement vector and the ∆u
J
(i) and du
J
(i) nodal in-
cremental vectors are retained in global arrays. The global arrays are passed to each
element in a local array, ul(ndf,nen,i). The deﬁnitions of the entries in the local
array are given in Table 8.1.
The array ul contains information for the current element according to the deﬁnitions
in Table 8.2.
Chapter 9
Linear Viscoelasticity
9.1 Isotropic Model
The representation of a constitutive equation for linear viscoelasticity may be in the
form of either a diﬀerential equation or an integral equation form. In the discussion
to be presented here we assume the material is linear and isotropic. Accordingly, in
matrix form the stress and strain may be split as
σ = s + mp (9.1)
and
= e +
1
3
mθ (9.2)
where σ is the Cauchy stress, s is the stress deviator, and p is the mean (pressure)
stress deﬁned in matrix form as
p =
1
3
m
T
σ (9.3)
is strain, e is the strain deviator, and θ is the volume change deﬁned in matrix form
as
θ = m
T
(9.4)
In the presentation given here we assume that the pressure-volume parts of the behavior
are governed by a linear elastic model
p = K θ (9.5)
where K is the bulk elastic modulus deﬁned in terms of Young’s modulus and Poisson’s
ratio as
K =
E
3 (1 −2 ν)
(9.6)
74
CHAPTER 9. LINEAR VISCOELASTICITY 75
The deviatoric parts are assumed to satisfy a linear viscoelastic model.
Linear viscoelastic behavior may be stated in the form of diﬀerential equation models
or in the form of integral equations. In the diﬀerential equation model the constitutive
equation may be written as
P(s) = 2 GQ(e) (9.7)
where P and Q are diﬀerential operators expressed as
P = p
m

m
∂t
m
+ p
m−1

m−1
∂t
m−1
+ · · · + p
0
(9.8)
Q = q
m

m
∂t
m
+ q
m−1

m−1
∂t
m−1
+ · · · + q
0
(9.9)
and
G =
E
2 (1 + ν)
(9.10)
is identical to the elastic shear modulus. Alternatively, the operator may be written as
s = 2 G(µ
0
e +
N

i=1
µ
i
q
i
) (9.11)
˙ q
i
+
1
λ
i
q
i
= ˙ e (9.12)
This form of the representation is equivalent to a generalized Maxwell model (a set of
Maxwell models in parallel). The set of ﬁrst order diﬀerential equations may be inte-
grated for speciﬁed strains, e. The integral for each term is given by the homogeneous
diﬀerential equation solution, q
i
h
,
q
i
h
(t) = C exp
−t
λ
i
(9.13)
and variation of parameters on C to give
q
i
(t) =
_
t
−∞
exp −
t −τ
λ
1
˙ e(τ) dτ (9.14)
An advantage to the diﬀerential equation form is that it may be easily extended to
include aging eﬀects by making the parameters time dependent.
An alternative to the linear viscoelastic model in diﬀerential form is to use an integral
equation form. The integral form equivalent to the above is expressed in terms of the
relaxation modulus function. The relaxation modulus function is deﬁned in terms of an
idealized experiment in which, at time labeled zero (t = 0), a specimen is subjected to
a constant strain, e
0
, and the stress response, s(t), is measured. For a linear material
CHAPTER 9. LINEAR VISCOELASTICITY 76
a unique relation is obtained which is independent of the magnitude of the applied
strain. This relation may be written as
s(t) = 2 G(t) e
0
(9.15)
where G(t) is deﬁned as the shear relaxation modulus function. Using linearity and
superposition for an arbitrary state of strain yields an integral equation speciﬁed as
s(t) =
_
t
−∞
G(t −τ) ˙ e(τ) dτ (9.16)
It is noted that the above form is a generalization of the Maxwell material. Indeed the
integral equation form may be deﬁned as a generalized Maxwell model by assuming
the shear relaxation modulus function in the Prony series form
G(t) = G
0
+
N

i=1
G
i
exp
−t
λ
i
(9.17)
or the alternate form
G(t) = G(µ
0
+
N

i=1
µ
i
exp
−t
λ
i
) (9.18)
where
µ
0
+
N

i=1
µ
i
= 1 (9.19)
With this form the integral equation form is identical to the diﬀerential equation model
for the generalized Maxwell material. In the subsequent discussion we will consider the
generalized Maxwell material and let N be 1 (i.e., the standard linear solid). The
addition of more terms may be easily accommodated based upon the one term repre-
sentation. Accordingly,
G(t) = G(µ
0
+ µ
1
exp
−t
λ
1
) (9.20)
where
µ
0
+ µ
1
= 1 (9.21)
In applications involving a linear viscoelastic model, it is usually assumed that the
material is undisturbed until a time identiﬁed as zero. At time zero a strain may be
suddenly applied and then varied over subsequent time. The integral representation
for the the model may be simpliﬁed by dividing the integral into
_
t
−∞
(·) dτ =
_
0

−∞
(·) dτ +
_
0
+
0

(·) dτ +
_
t
0
+
(·) dτ (9.22)
CHAPTER 9. LINEAR VISCOELASTICITY 77
The ﬁrst term is zero, the second term includes a jump term associated with e
0
at
time zero, and the last term covers the subsequent history of strain. The result of this
separation when applied to Eq. 9.16 gives
s(t) = 2 G(t) e
0
+ 2
_
t
0
G(t −τ) ˙ e(τ) dτ (9.23)
where subsequently the 0 limit on the integral is understood as 0
+
.
Substitution of Eq. 9.20 into Eq. 9.23 gives
s(t) = 2 G[µ
0
e(t) + µ
1
exp
−t
λ
1
(e
0
+
_
t
0
exp
t
λ
1
˙ e(τ) dτ)] (9.24)
It remains to evaluate the integral. Accordingly, we divide the integral as
_
t
0
(·) dτ =
_
t
n
0
(·) dτ +
_
t
t
n
(·) dτ (9.25)
If we deﬁne the integral as
i
1
(t) =
_
t
0
exp
τ
λ
1
˙ e(τ) dτ (9.26)
the above separation gives
i
1
(t) = i
1
(t
n
) +
_
t
t
n
exp
τ
λ
1
˙ e(τ) dτ (9.27)
Including the negative exponential multiplier term gives
h
1
= exp
−t
λ
1
i
1
(9.28)
and then
h
1
(t) = exp
−∆t
λ
1
h
1
n
+ ∆h (9.29)
where
∆h = exp
−t
λ
1
_
t
t
n
exp
τ
λ
1
˙ e(τ) dτ (9.30)
The strain rate is now approximated as constant over each time increment t
n
to t, thus
˙ e(τ) ≈
e(t) −e
n
∆t
; t
n
≤ τ ≤ t (9.31)
where e
n
denotes the value of the strain at time t
n
and ∆t denotes the time increment
t − t
n
. A numerical approximation to ∆h may be employed and one proposal uses a
midpoint (one-point) approximation for the integral as [8]
∆h = exp
−∆t

1
(e −e
n
) (9.32)
CHAPTER 9. LINEAR VISCOELASTICITY 78
The recursion then becomes
_
t
t
n
exp
τ
λ
1
dτ ≈ ∆t exp
t
n+
1
2
λ
1
(9.33)
where t
n+
1
2
denotes the time
1
2
(t
n
+ t). Thus, the recurrsion is now given in the form
h
1
(t) = exp
−t
λ
1
i
1
(t) = exp
−∆t
λ
1
[h
1
n
+ exp
∆t
2 λ
1
(e − e
n
)] (9.34)
and simpliﬁes to
h
1
(t) = exp
−∆t
2 λ
1
[exp
−∆t
2 λ
1
h
1
n
+ (e − e
n
)] (9.35)
This form requires only one evaluation of an exponential term. Furthermore, a zero
value of the time step produces a correct answer, as well as, a very large value of the
time step producing a zero value. Thus, the form is doubly asymptotically accurate.
The use of ﬁnite diﬀerence approximations on the diﬀerential equation form directly
does not produce this property.
While the above form is easy to evaluate it has problems when the size of the time
step is changed. Thus, a more stable form is used in FEAP where the integral over the
time step is evaluated in closed form [23]. The result gives
∆h =
λ
1
∆t
_
1 − exp
−∆t
λ
1
_
(e −e
n
) (9.36)
This approximation produces a singular ratio for zero time steps; however, the limit
value is well behaved at a unit value. For very small time steps a series expansion may
be used to yield accurate values. This form gives a recursion which is stable for small
and large time steps and gives smooth transitions under variable time steps. It may
also be extended for use with thermorheologically simple materials.
The constitutive equation now has the simple form
s(t) = 2 G[µ
0
e(t) + µ
1
h
1
(t)] (9.37)
The inclusion of more terms in the series reduces to evaluation of additional h
i
(t)
integral recursions. The required storage is increased by a need to preserve the h
i
for
each quadrature point in the problem and each term in the series.
The implementation of the viscoelastic model into a Newton solution process requires
the computation of the tangent tensor. Accordingly, we need to compute
∂s

=
∂s
∂e
∂e

=
∂s
∂e
I
dev
(9.38)
CHAPTER 9. LINEAR VISCOELASTICITY 79
where I
dev
is the deviatoric operator identiﬁed previously. The partial derivative with
respect to the deviatoric stress is given by
∂s
∂e
= 2 G[µ
0
I + µ
1
∂h
1
∂e
] (9.39)
If we let
∆h
1
= ∆h
1
(∆t) (e − e
n
) (9.40)
the derivative of the last term in Eq. 9.39 becomes
∂h
1
∂e
= ∆h
1
(∆t), I (9.41)
Thus, the tangent tensor is given by
∂s

= 2 G[µ
0
+ µ
1
∆h
1
(∆t)] I
dev
(9.42)
The only modiﬁcation from a linear elastic material is the substitution of the factor
G
visc
= G[µ
0
+ µ
1
∆h
1
(∆t)] (9.43)
for the elastic shear modulus. Again we note that for zero ∆t the full elastic modulus
is recovered, whereas for very large increments the equilibrium modulus µ
G
0
is used.
The above formulation is incorporated into the subroutine viscoe. Note the simplicity
of the additional coding needed to include the linear viscoelastic formulation. Since
the material is linear, use of the consistently derived tangent modulus terms leads to
convergence in one iteration (the second iteration produces a zero residual).
Chapter 10
Plasticity Type Formulations
10.1 Plasticity Constitutive Equations
The constitutive equations for a material which behaves according to a plasticity type
formulation for deformation states which exceed the elastic limit may be expressed by
assuming that the strains are decomposed according to
=
e
+
p
(10.1)
where
e
are the elastic strains and
p
are the inelastic strains. If the material is non-
linear hyper-elastic we may deduce the stress from the expression for the elastic strain
energy as
σ =
∂W

¸
¸
¸
¸

e
(10.2)
where W is the strain energy density and is expressed as a function of the elastic
strains and σ and
e
are stress and strain energy conjugates. For a linear hyper-elastic
material the stress to elastic strain relation is given by
σ = D
e
= D( −
p

0
) (10.3)
In the following discussion we limit our comments to linear elastic materials and also
set
0
zero. The inelastic component of the strain rate is related to the gradient of a
˙
p
= ˙ γ
∂f
∂σ
(10.4)
where f is a loading function and ˙ γ is a scalar rate term called the plastic consistency
parameter. The plastic consistency parameter, ˙ γ, is zero for elastic behavior and pos-
itive for plastic behavior. A back stress is deﬁned as α which is related to the plastic
80
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 81
strain rate through
˙ α =
2
3
H
kin
˙
p
=
2
3
H
kin
˙ γ
∂f
∂σ
(10.5)
where H
kin
is a kinematic hardening modulus. The yield surface is deﬁned in an as-
sociative manner, using the same function as the loading function, and is expressed
as
f(σ, α, ¯ e
p
) = F(Σ) − Y (¯ e
p
) (10.6)
where the stress and back stress appear in the form
Σ = σ − α (10.7)
and
Y (¯ e
p
) = Y
0
+ H
iso
¯ e
p
(10.8)
is a function which measures the size of the current yield surface. Commonly, Y
0
is
related to σ
y
, the yield stress in uniaxial tension. The isotropic hardening behavior of
the material is included in Y through an eﬀective or accumulated plastic strain deﬁned
by
¯ e
p
=
_
t
0
_
2
3
˙
p
· ˙
p
_1
2
dτ (10.9)
In Eq. 10.8 H
iso
is an isotropic hardening modulus. In the present study both the
isotropic and the kinematic hardening moduli will be assumed as constants. Using the
deﬁnition of the plastic strain rate the eﬀective plastic strain may also be written as
¯ e
p
=
_
t
0
˙ γ
_
2
3
∂f
∂σ
·
∂f
∂σ
_1
2
dt (10.10)
thus it is evident that ¯ e
p
is dependent on the integral of ˙ γ and the particular load-
ing/yield function used to describe the material.
Generally, the model is completed by describing a scalar function, g, which describes
the limit behavior of the model. Diﬀerent limit equations may be written for rate
independent plasticity, rate dependent plasticity, and generalized plasticity models.
The simplest relation is for classical, associative, rate independent plasticity where
g = f(σ, α, ¯ e
p
) ≤ 0 (10.11)
is used. Later alternative forms will be introduced to represent other types of material
behavior.
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 82
10.2 Solution Algorithm for the Constitutive Equa-
tions
The solution of the above set of equations may be eﬀected numerically using a variety
of algorithms. A very eﬀective method to integrate plasticity equations is the operator
split method with a return map concept [17, 18, 3, 4] the algorithm may be extended to
include a variety of modeling concepts for the limit behavior; however, for the present
we restrict our attention to plasticity as deﬁned by Eq. 10.11 above. Accordingly, a
discrete solution at time t
n
is deﬁned in terms of the state σ
n
, α
n
,
p
n
, and ¯ e
p
n
. The
solution is then advanced to time t
n+1
by specifying the strain,
n+1
.
To initiate the solution at t
n+1
a trial state is computed assuming the step is entirely
elastic. Recall that a step is elastic when ˙ γ is zero. This implies that there will be no
change to
p
, α, or ¯ e
p
during an elastic increment. The step is initiated by taking the
trial values for plastic quantities

pTR
n+1
=
p
n
(10.12)
α
TR
n+1
= α
n
(10.13)
¯ e
pTR
n+1
= ¯ e
p
n
(10.14)
and
˙ γ
TR
n+1
= 0 (10.15)
Thus for linear elasticity
σ
TR
n+1
= D(
n+1

pTR
n+1
) (10.16)
The trial stress given by Eq. 10.16 is checked in Eq. 10.6 and Eq. 10.11 to determine
if the step is elastic or whether inelastic terms should be active. If the state at t
n+1
is elastic the stresses (as well as other state variables) are set equal to the trial value;
otherwise, a correction is required to include the inelastic terms.
For an inelastic step the stresses must satisfy (1.2) for the time t
n+1
which requires
the rate equations for
p
and α to be integrated over the time increment. Accordingly,
integrating non-linear terms using a backward Euler implicit method between t
n
and
t
n+1
, the plastic strain is given by

p
n+1
=
p
n
+ λ
n+1
∂f
∂σ
¸
¸
¸
¸
n+1
(10.17)
and the back stress by
α
p
n+1
= α
p
n
+
ˆ
H
kin
λ
n+1
∂f
∂σ
¸
¸
¸
¸
n+1
(10.18)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 83
where
ˆ
H
kin
is a constant kinematic hardening parameter and the integral of the con-
sistency parameter is given by
λ
n+1
= γ
n+1
− γ
n
(10.19)
Similarly, evaluating Eq. 10.11 at t
n+1
gives
g
n+1
= f
n+1
= 0 (10.20)
Finally, integration of Eq. 10.4 produces
¯ e
p
n+1
= ¯ e
p
n
+ λ
n+1
(
3
2
∂f
∂σ
·
∂f
∂σ
)
1
2
¸
¸
¸
¸
n+1
(10.21)
The set of equations Eqs. 10.3, 10.18 and 10.20 constitute a set of non-linear equations
in terms of the values of σ
n+1
, α
n+1
and λ
n+1
which must be solved for each stress point
and each time step of interest. A Newton method may be used to solve the equations.
To simplify the notation the subscripts on n + 1 are omitted. The iteration counter is
shown as a superscript (j) and initial iterate values are taken as the trial stress and
zero λ
(
0). The iterative solution is continued until the norm the residuals are within
acceptable tolerance values (e.g., normally, half machine precision relative to the initial
iterate values are used since Newton’s method then guarantees that machine precision
is achieved if the next iteration is checked). Before proceeding with Newton’s method
we note that the following relations hold
∂f
∂σ
=
∂f
∂Σ
∂Σ
∂σ
=
∂f
∂Σ
(10.22)
and
∂f
∂α
=
∂f
∂Σ
∂Σ
∂α
= −
∂f
∂Σ
(10.23)
Thus, treating the equations Eq. 10.18 and Eq. 10.20 as residual equations, in the
form
R
(j)
σ
= −
p
n
− D
−1
σ
(j)
− λ
∂f
∂Σ
(10.24)
R
(j)
α
= α
n
+
ˆ
H
kin
λ
∂f
∂Σ
− α
(j)
(10.25)
and
R
(j)
f
= −f(σ
(j)
, α
(j)
, ¯ e
p
(j)) (10.26)
we may linearize the equations to obtain (note the iteration counter j is omitted in the
coeﬃcient array for simplicity)
_
¸
_
ˆ
C −λ

2
f
∂Σ
2

∂λ

∂f
∂Σ
)

ˆ
H
kin
λ

2
f
∂Σ
2
(I +
ˆ
H
kin
λ

2
f
∂Σ
2
) −
ˆ
H
kin

∂λ

∂f
∂Σ
)
∂f
∂Σ
T

∂f
∂Σ
T
−A
_
¸
_
_
_

(j+1)

(j+1)

(j+1)
_
_
=
_
¸
_
R
(j)
σ
T
(j)
α
R
(j)
f
_
¸
_
(10.27)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 84
where
A =
∂Y
∂¯ e
p
∂¯ e
p
∂λ
(10.28)
and
ˆ
C = D
−1
+ λ

2
f
∂Σ
2
(10.29)
The solutions to Eq. 10.27 are computed and added to obtain the next iterates. Ac-
cordingly,
σ
(j+1)
= σ
(j)
+ dσ
(j+1)
(10.30)
α
(j+1)
= α
(j)
+ dα
(j+1)
(10.31)
and
λ
(j+1)
= λ
(j)
+ dλ
(j+1)
(10.32)
deﬁne the next iterates. The solution is terminated whenever the norms of the residuals
are smaller than a selected small tolerance.
Once convergence is achieved for each stress point evaluation (i.e., to compute the stress
at each Gauss point for a given strain), the stresses may be used in the ﬁnite element
to compute each element residual. In addition it is necessary to compute the tangent
moduli, D
t
, for use in the element stiﬀness matrix (if one is used) for the next iteration
on the momentum balance equation. That is we need to perform a new solution to
see if the strains we used to compute the stresses are correct. This is accomplished, as
before, by solving

e
K
IJ
u
J
= F
I
+

e
R
I
(10.33)
where K
IJ
and R
I
are the element stiﬀness and residual, respectively. The computation
of the tangent moduli may be obtained by noting that the computation of the last stress
increment in the Newton solution of Eq. 10.27 may be written as
_
_

(j+1)

(j+1)

(j+1)
_
_
=
_
¸
_
ˆ
C −λ

2
f
∂Σ
2

∂λ

∂f
∂Σ
)

ˆ
H
kin
λ

2
f
∂Σ
2
(I +
ˆ
H
kin
λ

2
f
∂Σ
2
) −
ˆ
H
kin

∂λ

∂f
∂Σ
)
∂f
∂Σ
T

∂f
∂Σ
T
−A
_
¸
_
−1
_
¸
_
R
(j)
σ
T
(j)
α
R
(j)
f
_
¸
_
(10.34)
At convergence for the given strain, , the residuals will vanish; however, if we now
consider a linearization with respect to strain only R
σ
contributes to the change. The
linearization of the residuals with respect to an increment of strain yields
R
σ
= d ; R
α
= 0 ; R
f
= 0 (10.35)
Denoting the inverse matrix as
_
_
ˆ
D
11
ˆ
D
12
ˆ
D
13
ˆ
D
21
ˆ
D
22
ˆ
D
23
ˆ
D
31
ˆ
D
32
ˆ
D
33
_
_
=
_
¸
_
ˆ
C −λ

2
f
∂Σ
2

∂λ

∂f
∂Σ
)

ˆ
H
kin
λ
∂f
∂Σ
(I +
ˆ
H
kin
λ

2
f
∂Σ
2
) −
ˆ
H
kin

∂λ

∂f
∂Σ
)
∂f
∂Σ
T

∂f
∂Σ
T
−A
_
¸
_
−1
(10.36)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 85
The ﬁnal result for the linearization with respect to strain gives
_
_

(j+1)

(j+1)

(j+1)
_
_
=
_
_
ˆ
D
11
d
ˆ
D
21
d
ˆ
D
31
d
_
_
(10.37)
thus, the tangent moduli for the next momentum iteration is
D
t
=
ˆ
D
11
(10.38)
Except for giving the form of f this completes the speciﬁcation of the general algorithm.
Multiple yield surfaces may be included by modifying Eq. 10.4 to
˙
p
=
K

k=1
˙ γ
k
∂f
k
∂σ
(10.39)
with each part of the yield surface described by a separate function
g
k
= f
k
(σ, α, ¯ e
p
) ≤ 0 (10.40)
An active condition for each surface is denoted when ˙ γ
k
≥ 0. Thus, usually only 1 or
2 of the surfaces are active at any time.
As constitutive equations become complex the speciﬁcation of the parameters is more
diﬃcult. A systematic procedure for determining the parameters from experimental
data is given by Ju, et. al. [13]. The method provides the best estimates for the
parameters and their sensitivities to errors or inconsistencies in the data.
10.3 Isotropic plasticity: J
2
Model
As in previous developments, the strain is again split into deviatoric, e, and volumetric
(spherical), θ, parts and expressed in matrix notation as
= e +
1
3
mθ (10.41)
where
θ = m
T
(10.42)
For our study on inelastic behavior, the decomposition into elastic and plastic parts
may now be expressed as
e = e
e
+ e
p
(10.43)
and
θ = θ
e
+ θ
p
(10.44)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 86
The stress also is split into the deviatoric, s, and pressure (spherical) parts as
σ = s + mp (10.45)
where
p =
1
3
m
T
σ (10.46)
With the above splits the isotropic linear elastic constitutive equations are given by
p = K (θ − θ
p
) (10.47)
and
s = 2 G(e − e
p
) (10.48)
In the developments below we restrict plasticity to the deviatoric part only. Thus θ
p
vanishes and the yield function can depend only on the deviatoric part of the stress.
For isotropic materials the yield and loading function may be expressed in terms of the
invariants of stress and back stress. The invariants of s are denoted as J
1
, J
2
, and J
3
and given by
J
1
= m
T
s = 0 (10.49)
J
2
=
1
2
s
T
s (10.50)
and
J
3
= det (s) (10.51)
The simplest formulation is where the function depends only on J
2
. We write this
model using
_
2 J
2
= (s
T
s)
1
2
= s (10.52)
and including the back stress, the limit equation as
g = f(s, α, ¯ e
p
) = s − α − Y ¯ e
p
≤ 0 (10.53)
where Y is the radius of the yield function which is related to the uniaxial yield stress,
σ
y
, through
Y =
_
2
3

y
+ H
iso
¯ e
p
) (10.54)
and, thus, includes the eﬀects of linear isotropic hardening. The back stress adjusted
value Σ is given by
Σ = s − α (10.55)
A simple calculation shows that
∂f
∂σ
=
∂f
∂s
I
dev
;
∂f
∂s
=
∂f
∂Σ
∂Σ
∂s
=
∂f
∂Σ
;
∂f
∂α
=
∂f
∂Σ
∂Σ
∂α
= −
∂f
∂Σ
(10.56)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 87
(where we recall that I
dev
= I −
1
3
mm
T
), and
∂f
∂Σ
=
Σ
Σ
= n =
∂f
∂σ
(10.57)
Thus, the evolution of the back stress satisﬁes
˙ α =
2
3
H
kin
˙ γ
∂f
∂σ
=
2
3
H
kin
˙ γ n (10.58)
Noting that at the initial state α is zero, we can conclude that the back stress evolves
such that it is a purely deviatoric quantity. thus,
m
T
α = 0 (10.59)
With this fact we then have the following important properties
m
T
Σ = 0 ; m
T
n = 0 ; n
T
n = 1 (10.60)
Based upon the above all aspects of the J
2
plasticity model are restricted to deviatoric
components only. Thus, our model is completed by giving the evolution equations for
plastic strain and eﬀective plastic strain in the form
˙ e
p
= ˙ γ n (10.61)
¯ e
p
=
_
t
0
_
2
3
˙ γ dτ (10.62)
The discrete form of the isotropic J
2
model is given by the equations
p
n+1
= K θ
n+1
(10.63)
s
n+1
= 2 G(e
n+1
− e
p
n+1
) (10.64)
e
p
n+1
= e
p
n
+ λ
n+1
n
n+1
(10.65)
α
n+1
= α
n
+
2
3
H
kin
λ
n+1
n
n+1
(10.66)
¯ e
p
n+1
= ¯ e
p
n
+
_
2
3
λ
n+1
(10.67)
g
n+1
= Σ
n+1
− Y
n+1
≤ 0 (10.68)
Σ
n+1
= s
n+1
− α
n+1
(10.69)
and
Y
n+1
=
_
2
3

y
+ H
iso
¯ e
p
n+1
) = Y
n
+
2
3
H
iso
λ
n+1
(10.70)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 88
The solution of the J
2
model is straight forward and may be accomplished by solving
only a scalar equation in λ
n+1
. The solution is performed using a trial state based
upon the assumption that λ
TR
n+1
is zero. Accordingly,
e
pTR
n+1
= e
p
n
; α
TR
n+1
= α
n
; ¯ e
pTR
n+1
= ¯ e
p
n
(10.71)
This yields the trial deviator stresses
s
TR
n+1
= 2 G(e
n+1
− e
pTR
n+1
) = 2 G(e
n+1
− e
p
n
) (10.72)
which may be used to check the limit equation, g
n+1
. If the limit equation is satisﬁed
then the trial values deﬁne the solution at t
n+1
. If the trial values violate the limit
equation, it is necessary to perform the second part of the algorithm. The second
part of the algorithm solves the discrete rate equations using the trial values as initial
conditions. Accordingly, using Eqs. 10.65, 10.71 and 10.72 in Eq. 10.64 gives
s
n+1
= s
TR
n+1
− 2 Gλ
n+1
n
n+1
(10.73)
Next subtracting Eq. 10.66 from Eq. 10.73 gives
Σ
n+1
= s
n+1
− α
n+1
= s
TR
n+1
− α
n
− 2 (G +
1
3
H
kin
) λ
n+1
n
n+1
(10.74)
Noting that Σ
n+1
= Σ
n+1
n
n+1
Eq. 10.74 gives also that s
TR
n+1
− α
n
is in the
direction n
n+1
, and may be written as the scalar equation
_
Σ
n+1
− Σ
TR
n+1
+ 2 (G +
1
3
H
kin
) λ
n+1
_
n
n+1
= 0 (10.75)
that is, the coeﬃcient must vanish to obtain a solution. In the above
Σ
TR
n+1
= s
TR
n+1
− α
n
(10.76)
Combining Eq. 10.75 with Eq. 10.68 and Eq. 10.70 yields the scalar equation,
Σ
TR
− Y
n
= 2 [G +
1
3
(H
iso
+ H
kin
)] λ
n+1
(10.77)
Once λ
n+1
is known it may be combined with the result
n
n+1
= n
TR
n+1
=
Σ
TR
n+1
Σ
TR
n+1

(10.78)
to give the stress as
s
n+1
= s
TR
n+1
− 2 Gλ
n+1
n
n+1
(10.79)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 89
as well as the plastic strain and the back stress. In practice λ
n+1
is reduced slightly to
that s
n+1
is always slightly outside the limit yield condition. Accordingly, the solution
to Eq. 10.78 is perturbed as
λ
n+1
=
Σ
TR
n+1
− Y
n
(1 + tol)
2 [G +
1
3
(H
iso
+ H
kin
)]
(10.80)
where tol is a small value, say 10
−8
.
The solution of the problem, as shown above, does not require a linearization or an
iteration process. If non-linear isotropic hardening is included or alternative forms for
the limit equation are employed, the equation equivalent to Eq. 10.77 will be nonlinear
in λ
n+1
and a linearization and iteration process are required to obtain a solution.
Similarly, if non-linear kinematic hardening is introduced the reduction to a scalar
equation may also be complicated.
Once the converged value for λ
n+1
is known and the stress state determined, a check
on satisfaction of the momentum equation must be made. If the momentum equation
is not satisﬁed for the current time, t
n+1
, another iteration is necessary to improve
the estimate for the state of strain,
n+1
. If a Newton type solution method is used it
is necessary to compute an appropriate tangent modulus matrix for each stress point
in the analysis. For the solution process developed here, this may be achieved by
selecting as primary dependent variables the stress, s
n+1
, the back stress, α
n+1
, and
the consistency parameter, λ
n+1
. Writing appropriate residual equations as
R
s
= e − e
p
n

1
2 G
s − λn = 0 (10.81)
R
α
= α − α
n

2
3
λH
kin
n = 0 (10.82)
and
R
f
= Y
n
+
2
3
H
iso
λ − Σ = 0 (10.83)
In the above we have deleted speciﬁc reference to the values at t
n+1
complexity in the linearization performed below. It should be understood that λ de-
notes the value of the solution in the t
n+1
step, i.e., previously given as λ
n+1
, and etc.
for all the other variables. We note that for the current strain, e, the above equations
are satisﬁed; however, to proceed to the next iteration of the momentum equation we
consider a linearization of the above equations with respect to a change in the strain
also, which we denote by de. Accordingly, the linearization n becomes
∂n
∂Σ
=
1
Σ
(I − nn
T
) =
1
Σ
N (10.84)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 90
Using this result, the linearization of Eqs. 10.81 to 10.83 yields the set of equations
_
¸
_
1
2 G
I +
λ
Σ
N −
λ
Σ
TR

N n

2 λH
kin
3 Σ
N I +
2 λH
kin
3 Σ
N −
2
3
H
kin
n
n
T
−n
T

2
3
H
iso
_
¸
_
_
_
ds

_
_
=
_
_
de
0
0
_
_
(10.85)
The inverse to the coeﬃcient matrix may be computed by ﬁrst solving the ﬁrst two
equation for ds and dα in terms of dλ and de, and then substituting the result into the
third equation to obtain a ﬁnal expression for dλ in terms of de. This also permits the
substitution of alternative limit equations without changing the solution to the ﬁrst
part. Accordingly, we consider
_
1
2 G
I
+
λ
Σ
N

λ
Σ
N

2 λH
kin
3 Σ
TR

N I +
2 λH
kin
3 Σ
N
_
_
ds

_
=
_
de − ndλ

2
3
H
kin
ndλ
_
(10.86)
The solution to this equation is
1
_
ds

_
=
_
2 G(I − BN) BN
2 GC N I − C N
_ _
de − ndλ

2
3
H
kin
ndλ
_
(10.87)
where B and C are given by
B =
2 Gλ
Σ D
=
2 Gλ
Σ
TR

; C =
2 H
kin
λ
3 Σ D
=
2 H
kin
λ
3 Σ
TR

(10.88)
and where we have noted that
D = 1 + 2 (G +
H
kin
3
)
λ
Σ
=
Σ
TR

Σ
(10.89)
This result may be substituted into the third equation in Eq. 10.85 to obtain
2 Gn
T
de = 2 [G +
1
3
(H
iso
+ H
kin
)] dλ (10.90)
Substituting this result back into the ﬁrst of equation Eq. 10.87 yields the incremental
equation which yields the tangent modulus matrix for the algorithm. Thus, we obtain
ds = 2 G[I − B(I − nn
T
) − Ann
T
] de (10.91)
where
A =
G
G +
1
3
(H
iso
+ H
kin
)
(10.92)
Finally, for the diﬀerential strains, d, the tangent becomes
ds = 2 G
_
I
dev
− B(I
dev
− nn
T
) − Ann
T
¸
d (10.93)
1
See Appendix E for a discussion on the inverse of this type of matrix.
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 91
10.4 Isotropic viscoplasticity: J
2
model
The previous section presented the formulation and solution algorithm for a J
2
classical
plasticity model. In this section we show how such a formulation may be easily extended
to include rate eﬀects in the inelastic behavior. The model selected for exposition is
classical viscoplasticity as introduced by Prager for one-dimension and extended to full
three-dimensional form by Perzyna [15].
For the viscoplastic model considered here, the only modiﬁcation to the formulation
is the replacement of the limit equation for g. Indeed, other models representing the
problems of generalized plasticity and generalized viscoplasticity can also be developed
by such replacement. In viscoplasticity, the relationship for g becomes a constitutive
equation describing the evolution for the consistency parameter, ˙ γ. Accordingly, we
write
g = Φ[f(s, α, ¯ e
p
)] − ζ ˙ γ ≤ 0 (10.94)
where the yield condition, f, still is given by
f(s, α, ¯ e
p
) = s − α − Y ¯ e
p
(10.95)
with Y the radius of the yield function which is related to a uniaxial yield stress, σ
y
,
through
Y =
_
2
3

y
+ H
iso
¯ e
p
) (10.96)
and, thus, includes the eﬀects of linear isotropic hardening. For viscoplasticity, Y and
σ
y
relate to the inelastic behavior in the limit as ˙ γ tends to zero (the rate independent
limit). For loading rates which are ﬁnite, the stress state may lie outside the yield
surface. The function Φ together with the parameter ζ determine the rate dependency
of the model. Perzyna considers alternatives for representing Φ; however, here we
restrict our attention to the simple case given by
Φ(f) = (f)
m
(10.97)
where m is a positive integer power. Other functional forms for Φ may be considered
without conceptual diﬃculty. All the other equations for the model remain as given in
Section 10.3.
For trial stress values for which the yield function deﬁned by Eq. 10.95 exceeds zero,
the behavior is inelastic and the return map solution for the viscoplastic model is given
by Eqs. 10.63 to 10.67, 10.69 and 10.70. The formulation is completed by integration
of the constitutive equation Eq. 10.94 for the time interval t
n
to t
n+1
(i.e., ∆t) using
a backward Euler evaluation of the integrals to obtain
∆t Φ[f (s
n+1
, α
n+1
, ¯ e
p
n+1
)] − ζ λ
n+1
= 0 (10.98)
CHAPTER 10. PLASTICITY TYPE FORMULATIONS 92
The discrete consistency parameter may be obtained by combining the scalar coeﬃ-
cient from Eq. 10.75 with Eq. 10.98 to obtain a single nonlinear equation in λ
n+1
.
Accordingly, we obtain
∆t Φ[ Σ
TR
− Y
n
− 2 (G +
1
3
(H
iso
+ H
kin
)) λ
n+1
] − ζ λ
n+1
= 0 (10.99)
For the simple model used here for Φ, the above becomes
∆t
_
Σ
TR
− Y
n
− 2 [G +
1
3
(H
iso
+ H
kin
)] λ
n+1
_
m
− ζ λ
n+1
= 0 (10.100)
In general, the above equation is nonlinear and must be solved numerically. For the
case where m = 1 the equation is linear and has the solution
λ
n+1
=
∆t ( Σ
TR
n+1
− Y
n
(1 + tol))
ζ + 2 ∆t [G +
1
3
(H
iso
+ H
kin
)]
(10.101)
Comparing Eq. 10.101 to Eq. 10.80 we can observe that the limit solution for ζ zero
is identical to the classical plasticity problem. The stress, back stress, plastic strain,
and eﬀective plastic strain are computed using the same expressions as for the classical
plasticity model. For nonzero ζ, the presence of ∆t in Eq. 10.101 implies a rate
dependency, with results depending on time durations for applying and changing loads
on the body. The extension to higher powers of m may be constructed using a Newton
solution scheme to solve the non-linear scalar equation.
Chapter 11
Augmented Lagrangian
Formulations
11.1 Constraint Equations - Introduction
The solution of many problems requires imposition of constraints as part of the for-
mulation. For example, if it is desired to solve the incompressible equations for linear
elasticity it is necessary to impose the constraint
tr() = m
T
= 0 (11.1)
Another type of constraint is to impose boundary conditions on a node, where we wish
to impose the condition for node I that
u
I
= ¯ u
I
(11.2)
in which ¯ u denotes a speciﬁed value. This type of constraint can be made more general
by letting the degrees-of-freedom be associated with a rotated local coordinate system
(e.g., a spherical coordinate frame) where now
u
I
= T
I
u
I
= ¯ u
I
(11.3)
in which T
I
is an orthogonal rotation matrix which transforms the degrees-of-freedom
from the global Cartesian to the prime system. Many other conditions could be given;
however, the above suﬃce for the present. The inclusion of the constraints into the ﬁnite
element problem may be performed by several diﬀerent approaches. For constraints of
the type Eq. 11.2 it is easy to directly eliminate the variables associated with u
I
, as is
done in FEAP. On the other hand the inclusion of Eq. 11.1 or Eq. 11.3 presents more
diﬃculty to implement. Thus, an alternative method is needed to implement general
types of constraints. A common approach is to use penalty methods; however, these
93
CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS 94
are sensitive to the value of the penalty parameter selected. A better approach, which
is numerically superior, is to use an augmented Lagrangian approach. This method is
an extension to the penalty method and uses values for the penalty parameter which
lead to a better conditioned numerical problem. In the sequel we ﬁrst consider penalty
methods, based upon a mixed formulation. Subsequently, we show how to extend the
mixed penalty treatment to the augmented Lagrangian algorithm which is based on an
iterative update procedure generally attributed to Uzawa [1].
11.2 Mixed Penalty Methods for Constraints
Consider a general constraint equation expressed as
g(u) = 0 (11.4)
which is to be imposed for some part of the domain, Ω
c
. The constraint may be
included as part of the problem formulation by supplementing the variational problem,
Π(u), with the term
Π
c
(u, λ) =
_

c
λ
T
g(u) dΩ (11.5)
Deﬁne the variations as
u
η
= u + η U (11.6)
and
λ
η
= λ + η Λ (11.7)
The variation of the integral gives the added terms

¸
¸
¸
¸
η=0
=
_

c
Λ
T
g(u) dΩ +
_

c
U
T
G
T
λdΩ (11.8)
where
G =
∂g
∂u
(11.9)
The Euler equation for the ﬁrst integral leads to the constraint equation.
g(u) = 0 (11.10)
for each point in Ω
c
, and the second equation leads to a term which is combined with
the variation of the original variational theorem to generate revised Euler equations
for the problem.
In a ﬁnite element matrix setting we can approximate the λ in each element as
λ = N
λ
α
(x) λ
α
(11.11)
CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS 95
and use the usual isoparametric interpolations for u. Thus, Eq. 11.8 generates the
matrix problem
G =

¸
¸
¸
¸
η=0
=
_

α
)
T
(U
I
)
T
¸
_
P
λ
α
P
λ
I
_
(11.12)
where
P
λ
α
=

e
_

ce
N
λ
α
g(u) dΩ (11.13)
and
P
λ
I
=

e
_

ce
N
I
G
T
λdΩ (11.14)
For non-linear constraint equations it is necessary to linearize this expression for com-
bination with the remaining part of the problem. Performing the linearization leads to
the problem
_

α
)
T
(U
I
)
T
¸
_
0 K
λ
αJ
K
λ

K
λ
IJ
_ _

β
du
J
_
(11.15)
where
K
λ

=

e
_

ce
N
I
G
T
N
λ
β
dΩ (11.16)
K
λ
αJ
=

e
_

ce
N
λ
α
GN
J
dΩ = (K
λ

)
T
(11.17)
and
K
λ
IJ
=

e
_

ce
N
I
λ
T

2
g
∂u∂u
N
J
dΩ =

e
_

ce
N
I
λ
T
∂G
∂u
N
J
dΩ (11.18)
The diﬃculty with the above formulation lies in the fact that there are no terms
in Eq. 11.15 which are associated with the diagonals for the λ degrees-of-freedom.
Moreover, if the constraints are linear there are no terms on the diagonals for any of
the degrees-of-freedom. This greatly, complicates a solution process since for a direct
solution the equations must be ordered to eliminate the displacement equations prior
to the Lagrange multiplier equations. Furthermore, iterative methods are even more
diﬃcult to consider. The deﬁciency associated with the diagonals for the Lagrange
multiplier equations may be removed by adding a regularization term to Eq. 11.5. The
modiﬁcation to the variational term considered takes the form
Π
c
(u, λ) =
_

c
λ
T
g(u) dΩ −
_

c
1
k
λ
T
λdΩ (11.19)
where k is a penalty parameter introduced such that in the limit as k goes to inﬁnity
the original problem is recovered.

¸
¸
¸
¸
η=0
=
_

c
Λ
T
(g(u) −
1
k
λ) dΩ +
_

c
U
T
G
T
λdΩ (11.20)
CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS 96
The Euler equation for the ﬁrst integral now gives the constraint equation.
g(u) −
1
k
λ = 0 (11.21)
for each point in Ω
c
. It is evident that the solution converges to satisfy the constraint
only in the limit when k is inﬁnity. The linearization of Eq. 11.20 gives the matrix
problem
_

α
)
T
(U
I
)
T
¸
_
K
λ
αβ
K
λ
αJ
K
λ

K
λ
IJ
_ _

β
du
J
_
(11.22)
where
K
λ
αβ
=

e
_

ce
N
λ
α
1
k
I N
λ
β
dΩ (11.23)
Many cases for constraints permit the elimination of the equations for λ
α
at a local
level. Thus, if a Newton solution scheme is employed the residual equations may be
written as
_
R
λ
α
R
I
(u) + R
λ
I
_
=
_
−P
λ
α
+ K
λ
αβ
λ
β
R
(
I
u
)
− P
λ
I
_

_
−K
λ
αβ
K
λ
αJ
K
λ

K
IJ
+ K
λ
IJ
_ _

β
du
J
_
=
_
0
0
_
(11.24)
This gives the set of equations to solve for the increment as
_
−K
λ
αβ
K
λ
αJ
K
λ

K
IJ
+ K
λ
IJ
_ _

β
du
J
_
=
_
−P
λ
α
+ K
λ
αβ
λ
β
R
I
(u) − P
λ
I
_
(11.25)
Solving the ﬁrst row of Eq. 11.25 gives

β
= (K
λ
αβ
)
−1
_
P
λ
α
+ K
λ
αJ
du
J
¸
− λ
β
(11.26)
Since the residual equation for λ
β
is linear it may be solved to give
λ
β
=
_
K
λ
αβ
_
−1
P
λ
α
(11.27)
and this simpliﬁes Eq. 11.26 to

β
=
_
K
λ
αβ
_
−1
K
λ
αJ
du
J
(11.28)
which when substituted into the second of Eq. 11.25 once again yields a displacement
model for the problem which is expressed as
ˆ
K
IJ
du
J
= R
I
(u) − P
λ
I
(11.29)
where
ˆ
K
IJ
= K
IJ
+ K
λ
IJ
+ K
λ

_
K
λ
αβ
_
−1
K
λ
αJ
(11.30)
CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS 97
The above solution process deﬁnes a perturbed Lagrangian form of the penalty solu-
tion process. In order to yield a solution which provides an adequate satisfaction of
the constraint equation, fairly large values for the penalty parameter should be used
(generally on the order of about half machine precision, e.g., 10
6
or 10
7
). The values
used then yield stiﬀness modiﬁcations for the second term on the right hand side of
Eq. 11.30 which are several orders larger than components appearing in the stiﬀness,
K
IJ
. If the values are too large, ill conditioning for the solution to the linear equations
will result; if too small, the constraint may be violated by an unacceptable amount.
Furthermore, iterative solutions become very diﬃcult for these large penalty values.
Consequently, an alternative approach is needed. In the next section, the augmented
Lagrangian method is introduced as an alternative.
11.3 Augmented Lagrangian Method for Constraints
The augmented Lagrangian strategy presented is a simple modiﬁcation to the perturbed
Lagrangian form which now becomes
Π
c
(u, λ, λ
A
) =
_

c
(λ + λ
A
)
T
g(u) dΩ −
_

c
1
k
λ
T
λdΩ (11.31)
where λ
A
is the augmented term. The variation to Eq. 11.31 gives

¸
¸
¸
¸
η=0
=
_

c
Λ
T
_
g(u) −
1
k
λ
_
dΩ +
_

c
U
T
G
T
(λ + λ
A
) dΩ (11.32)
+
_

c
Λ
T
A
g(u) dΩ (11.33)
The Euler equation for the variation of λ gives the equation
g(u) −
1
k
λ = 0 (11.34)
which may be used to compute λ. The variation for λ
A
gives the constraint equation
g(u) = 0 (11.35)
and, thus, the constraint equation is satisﬁed independently of the value of the penalty
parameter, k, and we also conclude that λ must vanish at the solution. Using, these
facts we also note that the algorithm merely reduces to the original Lagrange multiplier
method, but with λ
A
used as the multiplier. The method may be made computation-
ally viable by making the determination of λ
A
an iterative algorithm. The Uzawa
algorithm is the simplest algorithm which may be considered. In the Uzawa algorithm
we introduce an outer iteration loop for the augmentation. For each step in the analysis
we assume:
CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS 98
1. Let j be the augmentation iteration counter. For each time, t
n+1
, set j to zero
and take the initial value of the augmented multiplier as
λ
β(j)
A
= λ
β
A
(t
n
) (11.36)
where the dependence on the n+1 step on the left side is implied. Let λ
β
A
(0) = 0.
2. Solve the non-linear problem
_
−K
λ
αβ
K
λ
αJ
K
λ

K
IJ
+ K
λ
IJ
_ _

β
du
J
_
=
_
−P
λ
α
+ K
λ
αβ
λ
β
R
(
I
u
)
− (P
A
)
λ
I
_
=
_
0
R
I
(u) − (P
A
)
λ
I
_
(11.37)
where
λ
β
=
_
K
λ
αβ
_
−1
P
λ
α
(11.38)
and
(P
A
)
λ
I
=

e
_

ce
N
I
G
T
N
λ
β

β
+ λ
β(j)
A
) dΩ (11.39)
In the above the iteration aspects for the incremental problem are not shown.
3. After the iteration for the incremental problem converges update the augmented
parameters using
λ
β(j+1)
A
= λ
β(j)
A
+ λ
β
(11.40)
where
λ
β
=
_
K
λ
αβ
_
−1
P
λ
α
(11.41)
is computed using the converged solution from step 2.
4. Check convergence for the augmented step. If the constraint is satisﬁed to within
a speciﬁed tolerance, or the change in the λ
β
is less than some tolerance times
λ
β(j+1)
A
proceed to the next time and repeat steps 1 to 3.
If not converged increase the j counter and repeat steps 2 and 3.
To perform the above algorithm it is necessary for the penalty parameter k to be large
enough for the iteration to converge. All that is required is that the terms in the added
stiﬀness be somewhat larger than the original stiﬀness terms. The convergence rate for
the augmented iteration is generally linear, not quadratic as in a Newton solution. The
larger k is made the more rapid the convergence. Thus, it is desirable for the value to
be at least one or two orders in magnitude larger than the conventional stiﬀness terms
(as compared with the six or seven used in a penalty approach). Use of values with
this range in magnitudes leads to 3-6 augmented steps for most problems. The number
of non-linear iterations will decrease for the later augmented steps since the violation
in the constraint is becoming less and less.
CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS 99
Using the above augmented Lagrangian approach to satisfy the incompressibility con-
straint leads to a particularly simple update. For the constant pressure/volume element
there is only one equation for each element. Thus the equations to be solved are scalar.
For the enhanced element there is one equation at each Gauss point so it is also easy to
modify. For more complex situations, involving multi-point constraints, the situation
is slightly more complex. Augmented approaches have been used to solve a variety
of problems in ﬁnite element methods. In some cases (for example, frictional contact
problems) it is possible to augment in a way which renders a problem which origi-
nally has an un-symmetric tangent matrices to one which is symmetric. In general,
the method is the one of current choice since, as a special case, it also includes an
option of penalty solution through the perturbed Lagrangian approach (merely omit
all augmented steps!).
Chapter 12
Transient Analysis of Non-Linear
Problems
The variational equation for a quasi-static problem solved by the ﬁnite element method
is expressed as

η

¸
¸
¸
¸
η=0
=

e
(U
I
)
T
__

e
B
T
I
˜ σdΩ −
_

e
N
I
b
v
dΩ −
_
Γ
e
N
I
¯
t dΓ
_
= 0 (12.1)
where ˜ σ is computed for a displacement, mixed, or enhanced method as described
in previous chapters. In order to extend the variational equation to accommodate
transient analysis, the body force vector, b
v
, is replaced by
b
v
← b
v
− ρ ¨ u (12.2)
in which ¨ u is the acceleration vector. With this replacement the variational equation
becomes

η

¸
¸
¸
¸
η=0
=

e
(U
I
)
T
__

e
B
T
I
˜ σdΩ +
_

e
N
I
ρ ¨ udΩ −
_

e
N
I
bdΩ

_
Γ
e
N
I
¯
t dΓ
_
= 0 (12.3)
which leads to the residual equation for each node
R
I
= F
I

e
_

e
B
T
I
˜ σdΩ −

e
_

e
N
I
ρ ¨ udΩ (12.4)
100
CHAPTER 12. TRANSIENT ANALYSIS 101
The last term is the inertia contribution to the momentum equation. For continuum
problems the acceleration is computed from the isoparametric interpolations as
¨ u = N
J
(x) ¨ u
J
(t) (12.5)
thus, the inertia term may be written as

e
_

e
N
I
ρ ¨ udΩ = M
IJ
¨ u
J
(12.6)
where M
IJ
is the mass matrix for the problem. If we deﬁne
P
I
( ˜ σ) =

e
_

e
B
I
˜ σdΩ (12.7)
the residual equation becomes
R
I
= F
I
− P
I
( ˜ σ) − M
IJ
¨ u
J
(12.8)
or, by ignoring the nodal indices, in the total matrix form as
R = F − P( ˜ σ) − M¨ u (12.9)
In general, the above equation is a non-linear set of ordinary diﬀerential equations.
The practical solution of the equations is accomplished using a time marching scheme,
as described in the next section.
12.2 Newmark Solution of Momentum Equations
In this section we illustrate the solution of Eq. 12.9 by a time marching process using
the classical Newmark method of solution [14]. The Newmark method is a one-step
method which may be used to advance a solution from time t
n
to time t
n+1
. The
method is self starting, consequently, given the initial conditions,
u(0) =
¯
d
0
; ˙ u(0) = ¯ v
0
(12.10)
where
¯
d
0
and ¯ v
0
are the initial displacement and velocity vectors, the solution at the
ﬁrst increment may be determined. The Newmark method uses approximations to the
displacements, velocities, and accelerations and these are given as:
u
n
≈ u(t
n
) (12.11)
v
n
≈ ˙ u(t
n
) (12.12)
CHAPTER 12. TRANSIENT ANALYSIS 102
and
a
n
≈ ¨ u(t
n
) (12.13)
The initial state is completed by solving the residual equation at time zero. Accordingly,
R
0
= F
0
− P( ˜ σ
0
) − Ma
0
= 0 (12.14)
which yields the solution
a
0
= M
−1
[F
0
− P( ˜ σ
0
)] (12.15)
this is combined with the initial conditions
u
0
=
¯
d
0
; v
0
= ¯ v
0
(12.16)
to give a complete state at time zero.
The Newmark formulas to advance a solution are given by
u
n+1
= u
n
+ ∆t v
n
+ ∆t
2
_
(
1
2
− β) a
n
+ β a
n+1
_
(12.17)
and
v
n+1
= v
n
+ ∆t [(1 − γ) a
n
+ γ a
n+1
] (12.18)
in which β and γ are numerical parameters which control the stability and numeri-
cal dissipation, respectively. For γ =
1
2
there is no numerical dissipation, whereas
for γ >
1
2
numerical dissipation is introduced. The values of β control primarily
the stability but also inﬂuence the form of the matrix problem. A β of zero leads
to a formulation which is called explicit, where for no damping, the solution for the
acceleration, a
n+1
, involves only the mass matrix. For a diagonal mass this solution
step is very eﬃcient; however, in general the method is only conditionally stable and
very small time steps are needed. For β non-zero, the method is implicit and a solu-
tion step normally involves linearization of the momentum equation and an iterative
solution process based on Newton’s method. The advantage of implicit solutions is
improved stability, where quite large time steps may usually be taken. For example,
for β = 0.25, the method for linear problems is unconditionally stable. This method
is commonly called trapezoidal rule or constant average acceleration. Values of β less
than 0.25 should not be used since they are only conditionally stable with allowable
time steps not much larger than the explicit scheme.
The advancement of a solution from one step to the next is completed by combining Eq.
12.17 and Eq. 12.18 with the momentum equation written at time t
n+1
. Accordingly,
R
n+1
= F
n+1
− P( ˜ σ
n+1
) − Ma
n+1
= 0 (12.19)
In order to advance the solution to the next time it is necessary to recast the problem in
an iterative form. This involves selecting appropriate values for the variables to initiate
CHAPTER 12. TRANSIENT ANALYSIS 103
the step, linearization of the equations, solution of the linearized equations, and updat-
ing of the variables. Since the Newmark formulas are linear and have scalar coeﬃcients
they may be directly used in the momentum equation to reduce the primary unknowns
to a single vector. This vector may be the displacements, u
n+1
, the velocities, v
n+1
,
or the accelerations, a
n+1
. For the explicit case the only viable choice is accelerations.
In the sequel we will address the implicit case and use the displacements, u
n+1
, as the
primary unknowns. For an implicit solution it is best to select the initial value for the
iterate as
u
(
n+1
0) = u
n
(12.20)
Any other choice may perturb the displacements in such a way to cause false inelastic
values, especially near boundaries, which impede convergence of the Newton method.
With the choice Eq. 12.20, the values of the initial state which satisfy the Newmark
formulas are given by
v
(
n+1
0) =
_
1 −
γ
β
_
v
n
+ ∆t
_
1 −
γ
2 β
_
a
n
(12.21)
and
a
(
n+1
0) = −
1
β ∆t
v
n
+
_
1 −
1
2 β
_
a
n
(12.22)
Linearizing the Newmark formulas leads to the results
du
(i+1)
n+1
= β ∆t
2
da
(i+1)
n+1
(12.23)
and
dv
(i+1)
n+1
= γ ∆t da
(i+1)
n+1
(12.24)
Thus the appropriate update formulas (which also satisfy the Newmark formulas) are
given by
u
(i+1)
n+1
= u
(i)
n+1
+ du
(i+1)
n+1
(12.25)
v
(i+1)
n+1
= v
(i)
n+1
+
γ
β ∆t
du
(i+1)
n+1
(12.26)
and
a
(i+1)
n+1
= a
(i)
n+1
+
1
β ∆t
2
du
(i+1)
n+1
(12.27)
The linearization of the momentum equation leads to
K
(i)
t
du
(i+1)
n+1
= R
(i)
n+1
(12.28)
where
K
(i)
t
= −
_
∂R
∂u
+
∂R
∂v
∂v
∂u
+
∂R
∂a
∂a
∂u
_
(12.29)
or
K
(i)
t
= K
t
+
γ
β ∆t
C
t
+
1
β ∆t
2
M (12.30)
In Eq. 12.30, K
t
is the tangent stiﬀness matrix as computed for the quasi-static
problem, C
t
is a tangent damping matrix, and M is the mass matrix introduced above.
CHAPTER 12. TRANSIENT ANALYSIS 104
12.3 Hilber-Hughes-Taylor (HHT) Algorithm
The Newmark algorighm given in the previous section can be altered by considering
the residual for the momentum equation to be given by
R
n+α
= F
n+1
− P(u
n+α
, v
n+α
) − Ma
n+1
= 0 (12.31)
where the displacement at the intermeiate point is given by
u
n+α
= (1 −α) u
n
+ αu
n+1
(12.32)
and the velocity by
v
n+α
= (1 −α) v
n
+ αv
n+1
(12.33)
In the above t
n+α
= (1 − α)t
n
+ αt
n+1
. This algorithm is called the Hilber-Hughes-
Taylor α-method or, for short, the HHT-method [9] and has been analysed extensively
for stability and dissipative properties by Hughes [12]. To reduce the properties to a
single parameter, the relations,
β =
α
2
4
(12.34)
and
α + γ =
3
2
(12.35)
are employed.
Linearization of 12.31 gives the tangent matrix
K

=
1
β∆t
2
M +
αγ
β∆t
C + αK (12.36)
for use in the Newton method
K

du
n+1
= R
n+α
(12.37)
Chapter 13
Finite Deformation Formulations:
Elasticity
13.1 Kinematics and Deformation
The basic equations for solid mechanics may be found in standard references on the
subject (e.g., see Chadwick [2] or Gurtin [7]). Solution by ﬁnite element methods
is considered by Crisﬁeld in [3], by Curnier in [5], and by Zienkiewicz and Taylor in
[27].. Here only a summary of the basic equations is presented. A body B has material
points whose positions are given by the vector X in a ﬁxed reference conﬁguration
1
,

0
, in a three dimensional space. In Cartesian coordinates the position vector may be
described in terms of its components as:
XS = X
A
E
A
; A = 1, 3 (13.1)
where E
A
are unit base vectors. After the body is loaded each material point is
described by its position vector, x, in the current conﬁguration, Ω. The position
vector in the current conﬁguration may be given in terms of its components as
x = x
a
e
a
; a = 1, 3 (13.2)
where e
a
are unit base vectors for the current time. In our discussion, common origins
and directions are used for the reference and current coordinates. The position vector
at the current time is related to the reference conﬁguration position vector through the
mapping
x = φ(X, t) (13.3)
1
As much as possible we adopt the notation that upper case letters refer to quantities deﬁned in
the reference conﬁguration and lower case letters to quantities deﬁned in the current conﬁguration.
Exceptions occur when quantities are related to both the reference and current conﬁgurations.
105
CHAPTER 13. FINITE DEFORMATION 106
When common origins and directions for the coordinate frames are used, a displacement
vector, u, may be introduced as the change between the two frames. Accordingly,
x = 1X + u (13.4)
is used. In the above 1 is a rank two shifter tensor between the two coordinate frames,
and is given by
1 = δ
a
Ae
a
E
T
A
; a, A = 1, 3 (13.5)
where δ
aA
is a Kronecker delta quantity such that
δ
aA
=
_
1 if a = A
0 if a = A
(13.6)
In component form we then have
x
a
= δ
aA
X
A
+ u
a
(13.7)
A fundamental measure of deformation is described by the deformation gradient relative
to X given by
F =
∂φ
∂X
(13.8)
subject to the constraint
J = det F > 0 (13.9)
to ensure that material volume elements remain positive. The determinant of the
deformation gradient maps a volume element in the reference conﬁguration into one in
the reference conﬁguration, that is
dv = det FdV = J dV (13.10)
where dV is a volume element in the reference conﬁguration and dv its corresponding
form in the current conﬁguration.
The deformation gradient relates the current conﬁguration to the reference conﬁgura-
tion, consequently it has components deﬁned as
F = F
aA
e
a
E
T
A
(13.11)
The deformation gradient may be expressed in terms of the displacement as
F = 1 +
∂u
∂X
(13.12)
Using F directly complicates the development of constitutive equations and it is com-
mon to introduce deformation measures which are related completely to either the
CHAPTER 13. FINITE DEFORMATION 107
reference or the current conﬁgurations. Accordingly, for the reference conﬁguration,
the right Cauchy-Green deformation tensor, C, is introduced as
C = F
T
F (13.13)
Alternatively the Green strain tensor, E (do not confuse with the base vectors), is
introduced as
E =
1
2
(C − 1
0
) (13.14)
where 1
0
is the rank two identity tensor with respect to the reference conﬁguration and
is given by
1
0
= δ
AB
E
A
E
T
B
(13.15)
and δ
AB
is a Kronecker delta for the reference conﬁguration. The Green strain may be
expressed in terms of the displacements as
E =
1
2
_
1
T
∂u
∂X
+ (
∂u
∂X
)
T
1 + (
∂u
∂X
)
T
∂u
∂X
_
(13.16)
Deﬁning the displacement vector for the reference conﬁguration as
U = 1u (13.17)
with components
U
A
= δ
aA
u
a
(13.18)
the components of the Green strain may be written in the familiar form
E
IJ
=
1
2
_
∂U
I
∂X
J
+
∂U
J
∂X
I
+
∂U
K
∂X
I
∂U
K
∂X
J
_
(13.19)
In the current conﬁguration a common deformation measure is the left Cauchy-Green
deformation tensor, b, expressed as
b = FF
T
(13.20)
The Almansi strain tensor, e, may be expressed in terms of b as
e =
1
2
_
1
t
− b
−1
_
(13.21)
where 1
t
is the rank two identity tensor with respect to the current conﬁguration and
is given by
1
t
= δ
ab
e
a
e
T
b
(13.22)
and δ
ab
is a Kronecker delta for the current conﬁguration.
CHAPTER 13. FINITE DEFORMATION 108
13.2 Stress and Traction Measures
Stress measures the amount of force per unit of area. In ﬁnite deformation problems
care must be taken to describe the conﬁguration to which stress is measured. The
Cauchy stress, σ, and the Kirchhoﬀ stress, τ, are measures deﬁned with respect to the
current conﬁguration. They are related through the determinant of the deformation
τ = τ
ab
e
a
e
T
b
= J σ = J σ
ab
e
a
e
T
b
(13.23)
The second Piola-Kirchhoﬀ stress, S, is a stress measure with respect to the reference
conﬁguration and has components
S = S
AB
E
A
E
T
B
(13.24)
The second Piola-Kirchhoﬀ stress is related to the Kirchhoﬀ stress through
τ = FSF
T
(13.25)
Finally, the ﬁrst Piola-Kirchhoﬀ stress, P, is related to S through
P = FS = P
aA
e
a
E
T
A
(13.26)
which gives
τ = PF
T
(13.27)
For the current conﬁguration traction is given by
t = σ
T
n (13.28)
where n is an unit outward pointing normal to a surface deﬁned in the current conﬁgu-
ration. This form of the traction may be related to reference surface quantity through
force relations deﬁned as
t ds = t
0
dS (13.29)
where ds and dS are surface elements in the current and reference conﬁgurations,
respectively, and t
0
is traction on the reference conﬁguration. Note that the direction
of the traction component is preserved during the transformation and, thus, remains
related to the current conﬁguration forces. The reference conﬁguration traction is
deduced from the ﬁrst Piola-Kirchhoﬀ stress through
t
0
= PN (13.30)
where N is an unit outward pointing normal to the reference surface. Using the deﬁ-
nition for traction and stresses we obtain
F
T
nds = J NdS (13.31)
and
ds = J
_
N ·
_
C
−1
N

1
2
dS (13.32)
to relate changes in the surface area and transformation of the normals.
CHAPTER 13. FINITE DEFORMATION 109
13.3 Balance of Momentum
The balance of momentum for a solid body consists of two parts: balance of linear
momentum; and balance of angular momentum. The balance of linear momentum may
be expressed by integrating the surface and body loads over the body. Accordingly, for
a body force per unit mass, b
m
the resultant force, R, acting on a body is given by
_

ρ b
m
dv +
_
∂Ω
t ds = R (13.33)
where rho is the mass density per unit volume and ∂Ω is the surface area of the body,
both for the current conﬁguration. The mass density in the current conﬁguration is
related to the reference conﬁguration mass density, rho
0
, through
ρ
0
= J ρ (13.34)
The total linear momentum of the body is given by
p =
_

ρ v dv (13.35)
The balance of linear momentum describes the translational equilibrium of a body (or
any part of a body) and is obtained by equating the resultant force, R, to the rate of
change of the body momentum, ˙ p. Accordingly,
_

ρ b
m
dv +
_
∂Ω
t
d
s =
_

ρ ˙ v dv (13.36)
Introducing the relationship between traction and stress and using the divergence prin-
cipal leads to the balance of linear momentum relation
_

[div σ + ρ (b
m
− ˙ v)] dv = 0 (13.37)
where div is the divergence with respect to the current conﬁguration, that is,
div σ =
∂σ
ab
∂x
a
e
b
(13.38)
Since the above result must hold for any part of a body a local form for balance of
linear momentum may be deduced as
div σ + ρ b
m
= ρ ˙ v (13.39)
This relation is also called the local equilibrium equation for a body.
CHAPTER 13. FINITE DEFORMATION 110
Similar relations may be constructed for the balance of angular momentum and lead
to the requirement
σ = σ
T
(13.40)
that is, the Cauchy stress tensor is symmetric and, thus, has only six independent
components.
The balance of momentum may also be written for the reference conﬁguration using
results deduced above. Accordingly, we may write the integrals with respect to the
reference body as
_

0
ρ
0
b
m
dV +
_
∂Ω
0
t
0
dS =
_

0
ρ
0
˙ v dV (13.41)
where the deﬁnitions for rho, σ and nds in terms of reference conﬁguration quantities
have been used. Using the divergence principle on the traction term leads to the result
_

0
[Div P + ρ
0
(b
m
− ˙ v)] dV = 0 (13.42)
which has the local form
Div P + ρ
0
b
m
= ρ
0
˙ v (13.43)
In these relations Div is the divergence with respect to the reference conﬁguration
coordinates
Div P =
∂P
aA
∂X
A
e
a
(13.44)
We also note that the symmetry of the Cauchy stress tensor, σ, leads to the corre-
sponding requirement on P
FP
T
= PF
T
(13.45)
and subsequently to the symmetry of the second Piola-Kirchhoﬀ stress tensor
S = S
T
(13.46)
13.4 Boundary Conditions
The basic boundary conditions for a solid region consist of two types: displacement
boundary conditions and traction boundary conditions.
Boundary conditions are deﬁned on each part of the boundary for which a component
or components of a vector may be speciﬁed without solution of any auxiliary problem.
The conditions are usually given in terms of their components with respect to a local
coordinate system deﬁned by the orthogonal basis, e

a
, a = 1, 2, 3. At each point on
the boundary one (and only one) boundary condition must be speciﬁed for all three
CHAPTER 13. FINITE DEFORMATION 111
directions of the basis e

a
. Generally, these conditions may be a mixture of displacement
and traction boundary conditions.
For displacement boundary conditions, components of the position vector, x, may be
expressed with respect to the basis as
x = x

a
e

a
(13.47)
Boundary conditions may now be given for each component by requiring
2
x

a
= ¯ x

a
(13.48)
for each point on the displacement boundary, ∂Ω
u
. The boundary condition may also
be expressed in terms of components of the displacement vector, u. Accordingly,
u = x − 1X = u

a
e

a
(13.49)
deﬁne components of the displacement with respect to the prime coordinates. Thus,
boundary conditions may now be given for each displacement component by requiring
u

a
= ¯ u

a
(13.50)
In general, the boundary condition is non-linear unless points in the reference conﬁg-
uration can be identiﬁed easily (such as ﬁxed points).
The second type of boundary condition is a traction boundary condition. Using the
orthogonal basis described above, the traction vector t may be written as
t = t

a
e

a
(13.51)
Traction boundary conditions may be given for each component by requiring
t

a
=
¯
t

a
(13.52)
for each point on the boundary, ∂Ω
t
.
13.5 Initial Conditions
Initial conditions describe the state of a body at the start of an analysis. The conditions
describe the initial kinematic state with respect to the reference conﬁguration used
to deﬁne the body and the initial state of stress in this position. In addition, for
constitutive equations the initial values for internal variables which evolve in time
must be given.
2
A speciﬁed quantity is indicated by a superposed bar, (¯·).
CHAPTER 13. FINITE DEFORMATION 112
The initial conditions for the kinematic state consist of specifying the position and
velocity at time zero. Accordingly,
x(0) = φ(X, 0) =
¯
d
0
(X) (13.53)
and
v(0) =
˙
φ(X, 0) = ¯ v
0
(X) (13.54)
are speciﬁed at each point in the body.
The initial conditions for stresses are speciﬁed as
σ(0) = ¯ σ
0
(13.55)
at each point in the body.
13.6 Material Constitution - Finite Elasticity
The equations are completed by specifying a material constitution. As an example, we
consider a ﬁnite deformation form for hyperelasticity. Thus, we postulate the existence
of a strain energy density function, W, from which stresses are computed by taking a
derivative with respect to a deformation measure. For a strain energy density expressed
in terms of the right Cauchy-Green deformation tensor, C, the second Piola-Kirchhoﬀ
stress tensor is computed as
S = 2
∂ W
∂ C
(13.56)
For an isotropic material the strain energy density depends only on the three invariants
of the deformation. Here we consider the three invariants as
I
C
= tr C = C
KK
(13.57)
II
C
=
1
2
(I
2
C
− tr C
2
) =
1
2
(C
KK
C
LL
− C
KL
C
LK
) (13.58)
and
III
C
= det C = J
2
(13.59)
and write the strain energy density as
W(C) = W(I
C
, II
C
, J) (13.60)
We select J instead of III
C
as the measure of the volume change. Thus, the stress is
computed as
S = 2
∂ W
∂ I
C
∂ I
C
∂ C
+ 2
∂ W
∂ II
C
∂ II
C
∂ C
+ 2
∂ W
∂ J
∂ J
∂ C
(13.61)
CHAPTER 13. FINITE DEFORMATION 113
The derivatives of the invariants may be evaluated as
∂ I
C
∂ C
= 1
0
(13.62)
∂ II
C
∂ C
= I
C
1
0
− C (13.63)
∂ J
∂ C
= J C
−1
(13.64)
Thus, the stress is computed to be
S = 2
∂ W
∂ I
C
1
0
+ 2
∂ W
∂ II
C
(I
C
1
0
− C) +
∂ W
∂ J
J C
−1
(13.65)
The second Piola-Kirchhoﬀ stress may be transformed to the Kirchhoﬀ stress using Eq.
13.25, and gives
τ = 2
∂ W
∂ I
C
b + 2
∂ W
∂ II
C
(I
C
b − 1
t
) +
∂ W
∂ J
J 1
t
(13.66)
As an example, we consider the case of a Neo-Hookean material which includes a
compressibility eﬀect. The strain energy density is expressed as
W(I
C
, J) = µ(I
C
− 3 − 2 ln J) +
1
2
λ(J − 1)
2
(13.67)
The material constants λ and µ have been selected to give the same response in small
deformations as a linear elastic material using the Lam´e moduli. Substitution into Eq.
13.65 gives
S = 2 µ(1
0
− C
−1
) + λJ (J − 1) C
−1
(13.68)
which may be transformed to give the Kirchhoﬀ stress
τ = 2 µ(b − 1
t
) + λJ (J − 1) 1
t
(13.69)
The Cauchy stress is then obtained from
σ =
τ
J
(13.70)
Some formulations require computation of the elastic moduli for the ﬁnite elasticity
model. The elastic moduli with repect to the reference conﬁguration are deduced from
CI = 4

2
W
∂C∂C
(13.71)
The spatial elasticities related to the Cauchy stress, σ, are obtained by the push forward
c
ijkl
=
1
J
F
iI
F
jJ
F
kK
F
lL
C
IJKL
(13.72)
CHAPTER 13. FINITE DEFORMATION 114
For the Neo-Hookean model the material modul with respect to the reference conﬁgu-
ration are given as
C
IJKL
= λJ (2 J
−1
) C
−1
IJ
C
−1
KL
− 2 (µ − λJ (J − 1)) C
−1
IK
C
−1
JL
(13.73)
Transforming to spatial quantities gives
c
ijkl
= λ(2 J
−1
) δ
ij
δ
kl
− 2 (
µ
J
− λ(J − 1)) δ
ik
δ
jl
(13.74)
Other forms of constitutive equations may be introduced using appropriate expansions
of the strain energy density function. As an alternative, an elastic formulation may
also be expressed in terms of the principal stretches (which are the sqare root of the
eigenvalues of C); however, the computations are quite delicate (see [19]).
13.7 Variational Description
A variational theorem for ﬁnite elasticity may be written in the reference conﬁguration
as
Π(u) =
_

0
W(C(u)) dV −
_

0
u
T
ρ
0
b
m
dV −
_
∂Ω
0t
u
T
¯
t
0
dS (13.75)
where
¯
t
0
denotes the speciﬁed tractions in the reference conﬁguration and ∂Ω
0
t is the
traction boundary for the reference conﬁguration. In a ﬁnite element formulation, the
basic element arrays evolve from the balance of linear momentum equations written
as a variational equation. Accordingly, in the reference conﬁguration a variational
equation is deﬁned from Eq. 13.75 as
3
δΠ =
_

0
∂ W
∂ C
δCdV −
_

0
δu
T
ρ
0
b
m
dV −
_
∂Ω
0
t
δu
T
¯
t
0
dS = 0 (13.77)
where δu is a variation of the displacement (often called a virtual displacement) which
is arbitrary except at the kinematic boundary condition locations, ∂Ω
u
, where, for con-
venience, it vanishes. Since a virtual displacement is an arbitrary function, satisfaction
of the variational equation implies satisfaction of the balance of linear momentum at
3
Since the notation for ﬁnite deformation includes use of upper and lower case letters, the notation
for a variation to a quantity is written as δ. Thus,
u
eta
→ u + δu (13.76)
Furthermore, matrix notation is used as much as possible to express the variational equation.
CHAPTER 13. FINITE DEFORMATION 115
each point in the body as well as the traction boundary conditions. We note that using
Eq. 13.26, Eq. 13.56 and constructing the variation of C, the ﬁrst term reduces to
∂ W
∂ C
δC =
1
2
SδC = δF
T
P (13.78)
Furthermore, by introducing the inertial forces through the body force as
b
m
→ b
m
− ˙ v = b
m
− ¨ x (13.79)
where v is the velocity vector, the variational equation may be written as
δΠ =
_

0
δu
T
ρ
0
˙ v dV +
_

0
δF
T
PdV (13.80)

_

0
δu
T
ρ
0
b
m
dV −
_
∂Ω
0
δu
T
¯
t
0
dS = 0 (13.81)
This is the variational equation form of the equations which is used for subsequent
development of the ﬁnite element arrays. The ﬁrst term side represents the inertial
terms. For static and quasi-static problems this term may be neglected. The second
term is the stress divergence term which also may be given in terms of the second
Piola-Kirchhoﬀ stress as
δF
T
P = δF
T
FS =
1
2
δCS = δES (13.82)
where symmetry of the second Piola-Kirchhoﬀ stress is noted. The third and fourth
terms of the variational equation represent the eﬀects of body and surface traction
The above variational equation may be transformed to the current conﬁguration as
δΠ =
_

δu
T
ρ ˙ v dv +
_

∇(δu)
T
σ] dv (13.83)

_

δu
T
ρ b
m
dv −
_
∂Ω
t
δu
T
¯
t ds = 0 (13.84)
The last result is identical to the conventional, small deformation formulation found
in earlier chapters and in ﬁnite element texts (e.g., see Hughes [12] or Zienkiewicz
and Taylor [26, 27]) except that integrals are performed over the deformed current
conﬁguration.
Representations with respect to a ﬁxed reference conﬁguration are introduced to sim-
plify the development of the basic relations. Some authors refer to the case where
the reference conﬁguration is the initial description of the body (at time zero) as a
total Lagrangian description and to one which is referred to the previous computed
conﬁguration as an updated Lagrangian description. For the development considered
here it is not important which is selected since ultimately all integrations are carried
out over the current conﬁguration; and, either a total or an updated description can
be transformed to the current state.
CHAPTER 13. FINITE DEFORMATION 116
13.8 Linearized Equations
The stress divergence term may be written in many forms, as shown above. To solve
a boundary value problem the nonlinear equations may be linearized and solved as
a sequence of linear problems. The linearization should be considered in a reference
conﬁguration representation. In this section it is expedient to again use a tensor form
of the equations instead of the matrix form used above. Accordingly, a formulation
based upon the second Piola-Kirchhoﬀ stress and written in tensor form is considered
for the linearization step.
_

0
tr (δF
T
FS) dV =
_

0
tr (δFSF
T
) dV (13.85)
which are equivalent forms. In the above, the trace operation denotes the following
step (reference conﬁguration tensors are used as an example, but other forms also hold)
tr (AB) = A
IJ
B
JI
(13.86)
Note that in the reference conﬁguration the domain, Ω
0
is ﬁxed (i.e, does not change)
which is not true for a formulation considered directly in the current conﬁguration.
Consequently, a linearization may be written as
∆(
_

0
tr (δFSF
T
dV ) =
_

0
tr (δFS∆F
T
) dV +
_

0
tr (δF∆SF
T
) dV (13.87)
We also note that for a continuum problem ∆(δF) vanishes, which is not true for
problems in beams, plates and shells and, thus, additional terms are necessary. The
linearization may be transformed to the current conﬁguration and expressed in terms
of quantities associated with the Cauchy stress. Accordingly, using
δF = ∇(δu) F (13.88)
and a similar expression for ∆F gives
∆(
_

0
tr (δFSF
T
dV ) =
_

tr (∇(δu) σ∇(∆u)
T
) dv
+ +
_

tr (∇(δu) ∆σ) dv (13.89)
where
∆σ =
1
J
F∆SF
T
(13.90)
The ﬁrst term on the right hand side leads to the geometric stiﬀness term in a ﬁnite
element formulation, whereas, the second term depends on the material constitution
and leads to the material part of the stiﬀness.
CHAPTER 13. FINITE DEFORMATION 117
The material part involves ∆S which is computed for each particular constitutive
relation. This will be discussed later for a particular constitutive equation; however,
in general we seek an expression of the form
∆S = CI ∆E (13.91)
where CI are the material moduli for the material constitution expressed in the reference
conﬁguration. When used with the deﬁnition of ∆σ this may be transformed to the
current conﬁguration as
∆σ = c I ∆ (13.92)
where c I are the material moduli expressed in the current conﬁguration. The moduli
are related through
J c I = FFCI F
T
F
T
(13.93)
In the above ∆ is the symmetric part of the gradient of the incremental displacement.
It is expressed as
∆ =
1
2
[∇∆u + (∇∆u)
T
] (13.94)
Substitution of the above into the term for the material part of the stiﬀness yields
_

tr (∇(δu) ∆σ) dv =
_

tr (δ c I ∆)) dv (13.95)
which we note is also identical to the form of the linear problem.
13.9 Element Technology
A ﬁnite element discretization may be constructed by dividing the body into ﬁnite
elements. Accordingly, we have
Ω ≈ Ω
h
=

e

e
(13.96)
where Ω
e
is the domain of an individual element, e, and Ω
h
is the domain covered
by all the elements. We note that in general Ω
h
is an approximation to the domain
of the real body since the elements only have mapped polygonal shapes. With this
approximation the integrals in the variational equation may be approximated as
_

( · ) dv ≈
_

h
( · ) dv =

e
_

e
( · ) dv (13.97)
Using this approximation the variational equation become

e
_ _

e
δu
T
ρ ˙ v dv +
_

e
∇(δu)
T
σdv
_
=
CHAPTER 13. FINITE DEFORMATION 118

e
_ _

e
δu
T
ρ b
m
dv +
_
∂Ω
et
δu
T
¯
t ds
_
(13.99)
An approximate variational solution may be developed by writing trial solutions and
test functions for the motions and virtual displacements, respectively. Adopting an
isoparametric formulation (e.g., see [12, 26, 27]) we may write for a typical element
X = N
I
(ξ) X
I
; I = 1, 2, , nen (13.100)
where nen is the number of nodes deﬁning an element, I are node labels for the ele-
ment, N
I
(ξ) are shape functions for node I which maintain suitable continuity between
contiguous elements and X
I
are the coordinates for node I. Similarly, we may write
approximations for the current conﬁguration as
x = N
I
(ξ) x
I
(t) (13.101)
the displacements as
u = N
I
(ξ) u
I
(t) (13.102)
the incremental displacements as
∆u = N
I
(ξ) ∆u
I
(t) (13.103)
and the virtual displacements as
δu = N
I
(ξ) δu
I
(13.104)
Time dependence is included in the nodal parameters for the current position and
displacements.
13.10 Consistent and Lumped Mass Matrices
Using the above approximations we may discretize the terms in the variational equation
for each element. Accordingly, the ﬁrst term becomes
_

e
δu
T
ρ ˙ v dv = (δu
T
)
I
_

e
N
I
ρ N
J
dv 1
t
¨ x
J
(13.105)
where summation convention is implied for the a and b indices. The integral for the
shape functions deﬁnes the consistent mass matrix for the analysis which may be writ-
ten as
M
IJ
=
_

e
N
I
ρ N
J
dv 1
t
(13.106)
For procedures to construct a lumped mass see either [26, Appendix 8] or [11].
CHAPTER 13. FINITE DEFORMATION 119
13.11 Stress Divergence Matrix
The stress divergence term may be expanded by noting symmetry of σ to give
_

e
tr [∇(δu) σ] dv =
_

e
tr [δ σ] dv (13.107)
where δ is given by
δ =
1
2
_
∇(δu) + (∇(δu))
T
¸
(13.108)
Introducing matrix notation for σ and δ as
σ =
_
σ
11
σ
22
σ
33
σ
12
σ
23
σ
31
¸
T
(13.109)
and
δ =
_
δ
11
δ
22
δ
33
2 δ
12
2 δ
23
2 δ
31
¸
T
(13.110)
the stress divergence term may be written as
_

e
δ
T
σdv (13.111)
Expressing the δ in terms of the virtual displacements gives
δ =
_
∂δu
1
∂x
1
∂δu
2
∂x
2
∂δu
3
∂x
3
∂δu
1
∂x
2
+
∂δu
2
∂x
1
∂δu
2
∂x
3
+
∂δu
3
∂x
2
∂δu
3
∂x
1
+
∂δu
1
∂x
3
¸
T
(13.112)
Using the interpolations for the virtual displacements in each element leads to the
matrix relation
δ =
_
¸
¸
¸
¸
¸
¸
_
N
I,1
0 0
0 N
I,2
0
0 0 N
I,3
N
I,2
N
I,1
0
0 N
I,3
N
I,2
N
I,3
0 N
I,1
_
¸
¸
¸
¸
¸
¸
_
_
_
δu
I
1
δu
I
2
δu
I
3
_
_
= B
I
δu
I
(13.113)
In the above, the notation
N
I,1
=
∂N
I
∂x
1
(13.114)
has been used for conciseness. The B
I
matrix describes the transformation from the
virtual displacements, δu
I
to the δ. The stress divergence term may now be written
as
_

e
δ
T
σdv = (δu
I
)
T
_

e
B
T
I
σdv (13.115)
The above expressions are identical to results obtained for the linear problem except
that all calculations are based upon coordinates in the current (deformed) conﬁgura-
tion.
CHAPTER 13. FINITE DEFORMATION 120
13.12 Geometric stiﬀness
The geometric stiﬀness for a ﬁnite element formulation is obtained by substituting the
interpolations described above into the geometric term for a single element. Accord-
ingly,
_

e
tr (∇(δu) σ∇(∆u)
T
) dv = (δu
I
)
T
_

e
tr (∇N
T
I
σ∇N
J
) dv ∆u
J
(13.116)
Evaluation of the right hand side of the above expression leads to the geometric stiﬀness
matrix which is given by
(K
g
)
IJ
=
_

e
tr ( ∇N
T
I
σ∇N
J
) dv 1
t
(13.117)
In component form the expression for the geometric stiﬀness is given as
(K
g
ij
)
IJ
=
_

e
∂ N
I
∂ x
k
σ
kl
∂ N
J
∂ x
l
dv δ
ij
(13.118)
13.13 Material tangent matrix - standard B matrix
formulation
The material tangent matrix is deduced from the term
_

e
tr (∇(δu) ∆σ) dv =
_

e
tr (δ c I ∆) dv (13.119)
which is evaluated for a typical element. In matrix notation the right hand side becomes
_

e
tr (δ c I ∆) dv =
_

e
δ
T
D∆ dv (13.120)
where D denotes the material moduli in the current conﬁguration given in the matrix
representation introduced for the linear problem. Furthermore, substitution of the
ﬁnite element interpolations into the incremental strain term leads to the result in
matrix form
∆ = B
J
∆u
J
(13.121)
Thus, the material tangent is computed from
_

e
δ
T
D∆ dv = (δu
I
)
T
_

e
B
T
I
DB
J
dv ∆u
J
(13.122)
CHAPTER 13. FINITE DEFORMATION 121
and the material tangent matrix is given by
(K
m
)
IJ
=
_

e
B
T
I
DB
J
dv (13.123)
which again is identical to the linear problem except that all steps are performed for
the current conﬁguration.
The right hand side terms may be discretized by introducing the interpolations for the
virtual displacement. Accordingly, the body force may be given as
_

e
δu
T
ρ b
m
dv = (δu
I
)
T
_

e
N
I
ρ b
m
dv (13.124)
_
∂Ω
e
δu
T
¯
tds = (δu
I
)
T
_
∂Ω
et
N
I
¯
t ds (13.125)
13.15 Basic ﬁnite element formulation
Accumulating all terms together, the variational equation may be written as

e
(δu
I
)
T
_
M
IJ
¨ x
J
+
_

e
B
T
I
σdv − f
I
_
= 0 (13.126)
where f
I
is the sum of the body and surface traction terms.
f
I
=
_

e
N
I
ρ b
m
dv +
_
∂Ω
e
t
N
I
¯
t ds (13.127)
Since δu
I
is arbitrary, the variational equation leads to the discrete balance of linear
momentum equations

e
_
M
IJ
¨ x
J
+
_

e
B
T
I
σdv − f
I
_
= 0 (13.128)
which may be written in the compact matrix form
M¨ x + N(σ) = f (13.129)
CHAPTER 13. FINITE DEFORMATION 122
where N(σ) is the stress divergence vector.
Solution of this set of equations together with satisfying the material constitution and
the displacement boundary conditions, yields the solution to a problem. A common
solution procedure is to use a Newton type solution method and solve a sequence of
linear problems. Accordingly, in a Newton Method we write the momentum equation
as
R = f − M¨ x − N(σ) = 0 (13.130)
A linearization of this set of equations gives the result
M∆¨ u + K
t
∆u = R (13.131)
where
K
t
= K
m
+ K
g
(13.132)
The above description is for a standard displacement type formulation. We refer to the
method as the standard B-matrix formulation.
13.16 Mixed formulation
In the mixed formulation used, a modiﬁed deformation gradient, (as described in [19]),
is used. The mixed formulation is used to permit solution of incompressible and nearly
incompressible materials, as well as, compressible solutions which can be treated by
a standard B matrix formulations. Thus, the modiﬁed deformation gradient is based
upon a separation into volumetric and deviatoric parts as
F = F
vol
F
dev
(13.133)
where F
vol
measures volumetric part and F
dev
the deviatoric part of deformation. Since
det F measures the volumetric part we have
J = det F = det F
vol
det F
dev
(13.134)
det F
vol
= J (13.135)
and
det F
dev
= 1 (13.136)
This may be accomplished by using
F
vol
= J
1
3
1 (13.137)
for the volumetric part which gives
F
dev
= J

1
2
F (13.138)
CHAPTER 13. FINITE DEFORMATION 123
for the deviatoric part. The modiﬁed deformation gradient may then be constructed
by replacing the volumetric part by a mixed treatment. Accordingly, we deﬁne
˜
F =
_
θ
J
_1
3
F (13.139)
as the modiﬁed tensor. In the above expression, θ is a mixed representation for the de-
terminant of the deformation gradient. The modiﬁed right Cauchy-Green deformation
tensor is then computed from
˜
C =
˜
F
T
˜
F (13.140)
with similar deﬁnitions for
˜
E and
˜
b. The virtual modiﬁed deformation gradient is now
given by
δ
˜
F =
_
δθ
3 θ
1
t
+ (∇δu −
1
3
div δu1
t
)
_
˜
F (13.141)
A three ﬁeld variational statement of the problem is completed by adding to the motion,
φ, and mixed determinant of the modiﬁed deformation gradient, θ, the mixed pressure,
p.
Π(u, θ, p) =
_

0
W
_
˜
C(u, θ)
_
dV +
_

0
p (J − θ) dV (13.142)

_

0
u
T
ρ
0
b
m
dV −
_
∂Ω
0t
u
T
¯
t
0
dS (13.143)
A variational equation including the eﬀects of inertia may be constructed following
steps above and written as (see, [19])
_

δu
T
ρ ˙ v dv +
_

∇δu
T
( ˜ σ
dev
+p 1
t
) dv =
_

δu
T
ρ b
m
dv +
_
∂Ω
δu
T
¯
t ds (13.144)
for the linear momentum equation,
_

δθ(
tr ˜ σ

p
J
) dv = 0 (13.145)
for the relationship between the mixed pressure and the trace of the stress, and
_

δp (1 −
θ
J
) dv = 0 (13.146)
for the relation between the mixed and the determinant of the deformation gradient.
In the above, the modiﬁed Cauchy stress, ˜ σ, and the modiﬁed Kirchhoﬀ stress, ˜ τ, are
related to the modiﬁed second Piola-Kirchhoﬀ stress by
J ˜ σ = ˜ τ =
˜
F
˜
S
˜
F
T
(13.147)
CHAPTER 13. FINITE DEFORMATION 124
where
˜
S is computed using
˜
C as the deformation measure. The deviatoric part of the
stress, ˜ σ
dev
, is then computed using
˜ σ
dev
= (I −
1
3
1
t
1
T
t
) ˜ σ = I
dev
˜ σ (13.148)
where I is a rank four identity tensor. The spherical part of the stress is given by the
mixed pressure, p, not tr ˜ σ. The mixed pressure p is computed from tr ˜ σ using the
variational equation given above. Thus, the stress in this approach is computed using
σ = p 1
t
+ ˜ σ
dev
(13.149)
A ﬁnite element implementation for the above may be deduced using the isoparametric
interpolations given above for X, x, u, and δu. In addition interpolations for θ, δθ, p,
and δp must be given. In the low order elements the above functions are all taken as
constant in each element. Discretization of the modiﬁed momentum equation gives
M¨ x + N( ˜ σ
dev
+ p 1
t
) = F (13.150)
where the stress divergence vector for a typical node is given by
N
I
( ˜ σ
dev
+ p 1
t
) =

e
_

e
B
T
I
( ˜ σ
dev
+ p 1
t
) dv (13.151)
The pressure, p, appearing in the above relations may be obtained by ﬁrst computing
the mixed volume, θ, using the third variational equation. Accordingly, for each ele-
ment (with the constant interpolations for θ and p) integration of the third variational
equation yields a solution
θ =

e

e0
(13.152)
for each element, where Ω
e
is the volume of the element in the current conﬁguration
and Ω
e0
is the volume in the undeformed reference conﬁguration. The θ may now be
used to deﬁne the modiﬁed deformation quantity and the modiﬁed stress state, ˜ σ may
be determined in each element. Finally, use of the second variational equation yields
the mixed pressure as
p =
1

e0
_

e
tr ˜ σ
3 θ
dv (13.153)
in each element. This may be combined with the deviatoric part of ˜ σ to deﬁne the
mixed stress, σ, in each element. A tangent matrix may be computed for the mixed
formulation. Details for the construction are included in [19].
Bibliography
[1] K.J. Arrow, L. Hurwicz, and H. Uzawa. Studies in Non-Linear Programming.
Stanford University Press, Stanford, CA, 1958.
[2] P. Chadwick. Continuum Mechanics. John Wiley and Sons, New York, 1976.
[3] M.A. Crisﬁeld. Non-linear Finite Element Analysis of Solids and Structures, vol-
ume 1. John Wiley and Sons, Chichester, 1991.
[4] M.A. Crisﬁeld. Non-linear Finite Element Analysis of Solids and Structures, vol-
ume 2. John Wiley and Sons, Chichester, 1997.
[5] A. Curnier. Computational Methods in Solid Mechanics. Kluwer Academic Pub-
lishers, Dordrecht, 1993.
[6] G.H. Golub and C.F. Van Loan. Matrix Computations. The Johns Hopkins Uni-
versity Press, Baltimore MD, 2 edition, 1989.
[7] Mortin E. Gurtin. An Introduction to Continuum Mechanics. Academic Press,
New York, 1981.
[8] L.R. Herrmann and F.E. Peterson. A numerical procedure for viscoelastic stress
analysis. In Proceedings 7th ICRPG Mechanical Behavior Working Group, Or-
lando, FL, 1968.
[9] H.M. Hilber, T.J.R. Hughes, and R.L. Taylor. Improved numerical dissipation for
time integration algorithms in structural dynamics. Earthquake Engineering and
Structural Dynamics, 5:282–292, 1977.
[10] F.B. Hildebrand. Methods of Applied Mathematics. Prentice-Hall, Englewood
Cliﬀs NJ, 2 edition, 1965.
[11] E. Hinton, T. Rock, and O.C. Zienkiewicz. A note on mass lumping and related
processes in the ﬁnite element method. Earthquake Engineering and Structural
Dynamics, 4:245–249, 1976.
[12] Thomas J.R. Hughes. The Finite Element Method. Prentice-Hall, 1987.
125
BIBLIOGRAPHY 126
[13] J.W. Ju, J.C. Simo, K.S. Pister, and R.L. Taylor. A parameter estimation algo-
rithm for inelastic material behavior. In C.S. Desai, editor, Constitutive Equations
for Engineering Materials. January 1987.
[14] N. Newmark. A method of computation for structural dynamics. Journal of the
Engineering Mechanics Division, 85:67–94, 1959.
[15] P. Perzyna. Fundamental problems in viscoplasticity. Advances in Applied Me-
chanics, 9:243–377, 1966.
[16] J.C. Simo and M.S. Rifai. A class of mixed assumed strain methods and the
method of incompatible modes. International Journal for Numerical Methods in
Engineering, 29:1595–1638, 1990.
[17] J.C. Simo and R.L. Taylor. Consistent tangent operators for rate-independent
elastoplasticity. Computer Methods in Applied Mechanics and Engineering,
48:101–118, 1985.
[18] J.C. Simo and R.L. Taylor. A return mapping algorithm for plane stress elastoplas-
ticity. International Journal for Numerical Methods in Engineering, 22:649–670,
1986.
[19] J.C. Simo and R.L. Taylor. Quasi-incompressible ﬁnite elasticity in principal
stretches. continuum basis and numerical algorithms. Computer Methods in Ap-
plied Mechanics and Engineering, 85:273–310, 1991.
[20] R.L. Taylor. FEAP - A Finite Element Analysis Program, Programmer Manual.
University of California, Berkeley. http://www.ce.berkeley.edu/~rlt.
[21] R.L. Taylor. FEAP - A Finite Element Analysis Program, User Manual. Univer-
sity of California, Berkeley. http://www.ce.berkeley.edu/~rlt.
[22] R.L. Taylor, P.J. Beresford, and E.L. Wilson. A non-conforming element for stress
analysis. International Journal for Numerical Methods in Engineering, 10:1211–
1219, 1976.
[23] R.L. Taylor, K.S. Pister, and G.L. Goudreau. Thermomechanical analysis of vis-
coelastic solids. International Journal for Numerical Methods in Engineering,
2:45–79, 1970.
[24] M.M. Vainberg. Variational Methods for the Study of Nonlinear Operators.
Holden-Day Inc., San Francisco, CA, 1964.
[25] K. Washizu. Variational Methods in Elasticity and Plasticity. Pergamon Press,
New York, 3 edition, 1982.
BIBLIOGRAPHY 127
[26] O.C. Zienkiewicz and R.L. Taylor. The Finite Element Method, volume 1.
McGraw-Hill, London, 4 edition, 1989.
[27] O.C. Zienkiewicz and R.L. Taylor. The Finite Element Method, volume 2.
McGraw-Hill, London, 4 edition, 1991.
Appendix A
Heat Transfer Element
This appendix contains a FEAP user subprogram to formulate the ﬁnite element arrays
needed to solve two-dimensional plane or axisymmetric linear heat transfer problems.
Table A.1 is the interface program to FEAP and Table A.2 is the subprogram to input
material parameters for the type of region, thermal conductivity, heat capacity, and
density for an isotropic Fourier material model. Table A.3 presents the routine used to
compute the element tangent and residual arrays and Tables A.4 and A.8 the routines
to output numerical values and nodal projections for the heat ﬂux. Nodal projections
of the heat ﬂux may then be plotted in graphics mode by FEAP. Table A.6 deﬁnes the
Fourier model and Table A.7 a routine to compute coordinates in elements. The heat
capacity array has been coded separtately in (Table A.8) to permit solution of general
linear eigenproblem.
128
APPENDIX A. HEAT TRANSFER ELEMENT 129
subroutine elmt02(d,ul,xl,ix,tl,s,r,ndf,ndm,nst,isw)
c Two dimensional heat transfer element
implicit none
include ’cdata.h’
include ’eldata.h’
include ’prstrs.h’
include ’comblk.h’
integer ndf,ndm,nst,isw, ix(*)
real*8 d(*),ul(ndf,*),xl(ndm,*),tl(*),s(nst,*),r(*),shp(3,9)
c Input material properties
if(isw.eq.1) then
call inpt02(d)
c Check of mesh if desired (chec)
elseif(isw.eq.2) then
call ckisop(ix,xl,shp,ndm)
c Compute conductivity (stiffness) matrix
elseif(isw.eq.3 .or. isw.eq.6) then
call stif02(d,ul,xl,ix,s,r,ndf,ndm,nst)
c Compute heat flux and print at center of element
elseif(isw.eq.4) then
call strs02(d,ul,xl,ix,ndf,ndm)
c Compute heat capacity (mass) matrix
elseif(isw.eq.5) then
call capa02(d,xl,ix,s,r,ndf,ndm,nst)
c Compute nodal heat flux for print/plots
elseif(isw.eq.8) then
call stcn02(ix,d,xl,ul,shp,hr(nph),hr(nph+numnp),
& ndf,ndm,nel,numnp)
endif
end
Table A.1: Element Routine for Heat Transfer
APPENDIX A. HEAT TRANSFER ELEMENT 130
subroutine inpt02(d)
implicit none
include ’iofile.h’
logical errck, tinput, pcomp, rflag
character name*15, wlab(2)*6
real*8 d(*),td(1)
data wlab/’ Plane’,’Axisym’/
c Input material parameters
d(4) = 1.d0
rflag = .true.
do while(rflag)
errck = tinput(name,1,td,1)
if(pcomp(name,’cond’,4)) then
d(1) = td(1)
elseif(pcomp(name,’spec’,4)) then
d(2) = td(1)
elseif(pcomp(name,’dens’,4)) then
d(3) = td(1)
elseif(pcomp(name,’plan’,4)) then
d(4) = 1.d0
elseif(pcomp(name,’axis’,4)) then
d(4) = 2.d0
elseif(pcomp(name,’ ’,4)) then
rflag = .false.
endif
end do ! while
if(ior.lt.0) write(*,2000) d(1),d(2),d(3),wlab(int(d(4)))
write(iow,2000) d(1),d(2),d(3),wlab(int(d(4)))
d(2) = d(2)*d(3)
d(5) = 2 ! number of quadrature points/direction
2000 format(5x,’Linear Heat Conduction Element’//
& 5x,’Conductivity ’,e12.5/5x,’Specific Heat’,e12.5/
& 5x,’Density ’,e12.5/5x,a6,’ Analysis’)
end
Table A.2: Input Routine for Heat Transfer Element
APPENDIX A. HEAT TRANSFER ELEMENT 131
subroutine stif02(d,ul,xl,ix,s,r,ndf,ndm,nst)
implicit none
include ’eldata.h’
include ’eltran.h’
integer ndf,ndm,nst, i,j, l,lint, ix(*)
real*8 d(*),ul(ndf,*),xl(ndm,*),s(nst,*),r(ndf,*)
c Compute tangent matrix (linear), and residual
l = nint(d(5))
call int2d(l,lint,sg)
do l = 1,lint
call shape(sg(1,l),sg(2,l),xl,shp,xsj,ndm,nel,ix,.false.)
do j = 0,nel-1
a1 = d(1)*shp(1,j+1)*xsj*sg(3,l)
a2 = d(1)*shp(2,j+1)*xsj*sg(3,l)
a3 = d(2)*shp(3,j+1)*xsj*sg(3,l)
do i = 0,nel-1
s(i*ndf+1,j*ndf+1) = s(i*ndf+1,j*ndf+1)
& + (a1*shp(1,i+1) + a2*shp(2,i+1))*ctan(1)
& + a3*shp(3,i+1)*ctan(2)
end do
end do
end do
end
Table A.3: Stiﬀness for Heat Transfer Element
APPENDIX A. HEAT TRANSFER ELEMENT 132
subroutine strs02(d,ul,xl,ix,ndf,ndm)
implicit none
include ’bdata.h’
include ’cdata.h’
include ’eldata.h’
include ’fdata.h’
include ’iofile.h’
integer ndf,ndm, ix(*)
c Compute thermal gradient and heat flux
call shape(0.0d0,0.0d0,xl,shp,xsj,ndm,nel,ix,.false.)
mct = mct - 1
if(mct.le.0) then
mct = 50
endif
2001 format(a1,20a4//5x,’element flux’//’ elmt matl 1-coord 2-coord’
2002 format(2i5,2f9.3,4e12.3)
end
Table A.4: Output Routine for Heat Transfer Element
APPENDIX A. HEAT TRANSFER ELEMENT 133
subroutine stcn02(ix,d,xl,ul,shp,dt,st,ndf,ndm,nel,numnp)
implicit none
integer ndf,ndm,nel,numnp, j,l,ll,lint, ix(*)
real*8 xsj,xg, tdot
real*8 dt(numnp),st(numnp,*),xl(ndm,*),shp(3,4),d(*)
c Lumped projection routine
l = max(2,nint(d(5)))
call int2d(l,lint,sg)
do l = 1,lint
call shape(sg(1,l),sg(2,l),xl,shp,xsj,ndm,nel,ix,.false.)
xsj = xsj*sg(3,l)
do j = 1,nel
ll = iabs(ix(j))
if(ll.gt.0) then
xg = xsj*shp(3,j)
dt(ll) = dt(ll) + xg
st(ll,1) = st(ll,1) + flux(1)*xg
st(ll,2) = st(ll,2) + flux(2)*xg
endif
end do
end do
end
Table A.5: Flux Projection Routine for Heat Transfer Element
APPENDIX A. HEAT TRANSFER ELEMENT 134
implicit none
include ’cdata.h’
integer ndm,nel, i
real*8 tdot
tdot = 0.0d0
do i = 1,nel
tdot = tdot + shp(3,i)*ul(1,i,4)
end do
end
Table A.6: Thermal Gradient and Flux
implicit none
integer i,ndm,nel
c Compute element coordinate value
do i = 1,nel
end do
end
Table A.7: Coordinate in Element
APPENDIX A. HEAT TRANSFER ELEMENT 135
subroutine capa02(d,xl,ix,s,r,ndf,ndm,nst)
implicit none
include ’eldata.h’
integer ndf,ndm,nst, i,j, l,lint, ix(*)
real*8 d(*),xl(ndm,*),s(nst,*),r(ndf,*), shp(3,9),sg(3,9)
c Compute heat capacity matrix
l = nint(d(5))
call int2d(l,lint,sg)
do l = 1,lint
call shape(sg(1,l),sg(2,l),xl,shp,xsj,ndm,nel,ix,.false.)
xsj = xsj*sg(3,l)
do j = 0,nel-1
shj = d(2)*shp(3,j+1)*xsj
r(1,j+1) = r(1,j+1) + shj
do i = 0,nel-1
s(i*ndf+1,j*ndf+1) = s(i*ndf+1,j*ndf+1) + shj*shp(3,i+1)
end do
end do
end do
end
Table A.8: Heat Capacity Routine for Heat Transfer Element
Appendix B
Solid Elements
B.1 Displacement elements
Displacement elements are computed using the virtual work equation written in terms
of assumed element displacments. All elements for continuum (solids) analysis use
isoparametric displacement ﬁelds expressed as
u =

I
N
I
(ξ) u
I
(B.1)
where N
I
(ξ) are shape functions and u
I
are nodal displacements. Computation of the
derivatives appearing in the strain-displacement matrices is performed as described in
Appendix D.
The strain-displacement matrices for each node are given by:
1. Three dimensional problems
=
_

x

y

z
γ
xy
γ
yz
γ
zx
¸
T
(B.2)
B
I
=
_
¸
¸
¸
¸
¸
¸
_
N
I,x
0 0
0 N
I,y
0
0 0 N
I,z
N
I,y
N
I,x
0
0 N
I,z
N
I,y
N
I,z
N
I,x
_
¸
¸
¸
¸
¸
¸
_
(B.3)
2. Two dimensional plane problems
=
_

x

y

z
γ
xy
¸
T
(B.4)
136
APPENDIX B. SOLID ELEMENTS 137
B
I
=
_
¸
¸
_
N
I,x
0
0 N
I,y
0 0
N
I,y
N
I,x
_
¸
¸
_
(B.5)
3. Two dimensional axisymmetric
=
_

r

z

θ
γ
rz
¸
T
(B.6)
B
I
=
_
¸
¸
_
N
I,r
0
0 N
I,z
N
I
r
0
N
I,z
N
I,r
_
¸
¸
_
(B.7)
Appendix C
Structural Elements
C.1 Truss elements
C.2 Frame elements
The current frame elements permit analysis of small displacement, second order dis-
placement, and ﬁnite displacement theories. Each element is a two node element with
linear displacement interpolations in each element.
C.2.1 Small displacement element
The strain-displacement relations for the small-displacement theory for plane bending
in the x
1
−x
2
global coordinate frame are given as
u
e
1
(z
1
, z
2
) = u
1
(z
1
) −z
2
θ(z
1
)
u
e
2
(z
1
, z
2
) = u
2
(z
1
) (C.1)
where z
1
and z
2
are coordinates and u
1
, u
2
and θ are displacement functions along the
z
1
-axis of the frame element.
These displacements give non-zero strains on each cross section expressed by
=
_

1
γ
12
_
=
_
−z
2
κ
γ
_
=
_
u
1,1
−z
2
θ
,1
u
2
, 1 −θ
_
(C.2)
where is the axial strain, κ the change in curvature and γ is the transverse shearing
strain for the cross section.
Two types of material constitution are considered:
138
APPENDIX C. STRUCTURAL ELEMENTS 139
1. Resultant theory where
_
_
_
N
M
V
_
_
_
=
_
_
EA 0 0
0 EI 0
0 0 kGA
_
_
_
_
_

κ
γ
_
_
_
(C.3)
2. Integration on the cross section where
_
N
M
_
=
_
A
_
1
z
2
_
σ
1
( −z
2
κ) dA (C.4)
C.3 Plate elements
C.4 Shell elements
Appendix D
Isoparametric Shape Functions for
Elements
D.1 Conventional Representation
The shape functions for the bilinear quadrilateral isoparametric element are given by
N
I
(ξ) =
1
4
(1 + ξ
I
1
ξ
1
) (1 + ξ
I
2
ξ
2
) (D.1)
Using these shape functions, the derivatives with respect to the natural coordinates
are computed to be
∂N
I
∂ξ
1
=
1
4
ξ
I
1
(1 + ξ
I
2
ξ
2
) (D.2)
and
∂N
I
∂ξ
2
=
1
4
ξ
I
2
(1 + ξ
I
1
ξ
1
) (D.3)
Using the shape functions, the interpolation for the global Cartesian coordinates may
be expressed in each element as
x = N
I
(ξ) x
I
(D.4)
where x
I
are the values of coordinates at the nodes of the element and the repeated
index I implies summation over the 4 nodes describing the quadrilateral element.
The derivatives of the shape functions with respect to the global coordinates, x, are
computed using the chain rule. Accordingly,
∂N
I
∂ξ
α
=
∂x
i
∂ξ
α
∂N
I
∂x
i
(D.5)
140
APPENDIX D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 141
which may be written in direct (matrix) notation as

ξ
N
I
= ∇
x
N
I
J (D.6)
When solved for the derivatives with respect to the global coordinates we obtain

x
N
I
= ∇
ξ
N
I
J
−1
(D.7)
In the above

x
N
I
=
_
∂N
I
∂x
1
∂N
I
∂x
2
_
(D.8)

ξ
N
I
=
_
∂N
I
∂ξ
1
∂N
I
∂ξ
2
_
(D.9)
and
J(ξ) =
_
∂x
1
∂ξ
1
∂x
1
∂ξ
2
∂x
2
∂ξ
1
∂x
2
∂ξ
2
_
(D.10)
Using the shape functions D.1 for the 4-node element, the terms in J(ξ) have the
structure
J

=
∂x
i
∂ξ
α
=
1
4
4

I=1
x
I
i
ξ
I
α
+
1
4
4

I=1
x
I
i
ξ
I
α
ξ
I
β
ξ
β
(D.11)
where
1
β = mod(α, 2) + 1 (D.12)
The constant part of J is evaluated at the point ξ = 0 (commonly named the element
center), and is given by
∂N
I
∂ξ
α
=
1
4
ξ
I
α
(D.13)
thus
J

(0) =
∂x
i
∂ξ
α
¸
¸
¸
¸
ξ=0
=
1
4
4

I=1
x
I
i
ξ
I
α
(D.14)
describe the derivatives of the coordinates at the element center. We denote the jaco-
bian at the center as J
0
, that is
J
0
= J(0) (D.15)
The global derivatives of the shape functions at the element center become

x
N
I
(0) = ∇
x
iN
I
(0) J
−1
0
(D.16)
1
Note that mod(i, j) = i −
i
j
j where
i
j
is evaluated in integer arithmetic. Thus, mod(1, 2) and
mod(3, 2) are both evaluated to be 1, while mod(2, 2) and mod(4, 2) are 0.
APPENDIX D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 142
In subsequent developments we use the notation
b
iI
=
∂N
I
∂x
i
¸
¸
¸
¸
ξ=0
(D.17)
to denote the derivatives of the shape functions at the element center.
In subsequent descriptions we will deﬁne
∆J
iαβ
=
1
4
4

I=1
x
I
i
ξ
I
α
ξ
I
β
=
1
4
4

I=1
x
I
i
ξ
I
1
ξ
I
2
= ∆J
i
(D.18)
which is the coeﬃcient to the spatially varying part of the jacobian transformation.
That is, the jacobian determinant may be expressed as
_
J
11
(ξ) J
12
(ξ)
J
21
(ξ) J
22
(ξ)
_
=
_
(J
0
)
11
(J
0
)
21
(J
0
)
21
(J
0
)
22
_
+
_
∆J
112
ξ
2
∆J
121
ξ
1
∆J
212
ξ
2
∆J
221
ξ
1
_
(D.19)
which in matrix notation may be written as
J(ξ) = J
0
+ ∆JΞ (D.20)
where
Ξ =
_
ξ
2
0
0 ξ
1
_
(D.21)
and
∆J =
_
∆J
1
∆J
1
∆J
2
∆J
2
_
(D.22)
D.2 Alternative Representation in Two Dimensions
An alternative representation for the shape functions has been proposed by Belytschko.
In the development of stabilized elements he introduced the representation
N
I
(ξ) =
1
4
δ
I
+
2

i=1
b
iI
(x
i
− x
0
i
) + Γ
I
h(ξ) (D.23)
where x
i
are the element global cartesian coordinates,
x
0
i
=
4

I=1
x
I
i
N
I
(0) =
1
4
(x
1
i
+ x
2
i
+ x
3
i
+ x
4
i
) (D.24)
are the values of the global coordinates at the element center,
h(ξ) = ξ
1
ξ
2
(D.25)
APPENDIX D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 143
and δ
I
and Γ
I
are constant parameters associated with node I. These parameters may
be evaluated by deﬁning the shape functions at each node and using the fact that
N
I

J
) = δ
IJ
(D.26)
where δ
IJ
is the Kronecker delta function for the nodes. Evaluating the alternative
shape function expression at each node gives
N
I

J
) =
1
4
δ
I
+
2

i=1
b
iI
(x
J
i
− x
0
i
) + Γ
I
h(ξ
J
) (D.27)
Introducing the notation
1
T
=
_
1 1 1 1
¸
(D.28)
h
T
=
_
1 −1 1 −1
¸
(D.29)
x
T
i
=
_
x
1
i
x
2
i
x
3
i
x
4
i
¸
(D.30)
b
T
i
=
_
b
i1
b
i2
b
i3
b
i4
¸
(D.31)
and the parameter vectors
δ
T
=
_
δ
1
δ
2
δ
3
δ
4
¸
(D.32)
Γ
T
=
_
Γ
1
Γ
2
Γ
3
Γ
4
¸
(D.33)
The shape functions at the nodes may be written in the matrix form
I =
1
4
δ 1
T
+
2

i=1
b
i
(x
i
− x
0
i
1)
T
+ Γh
T
(D.34)
Note that the rows in the expression are associated with the I in the N
I
shape functions,
while the columns are associated with the J where the ξ
J
are evaluated. The I is a
4 × 4 identity matrix for the element. Using this form, the parameters δ and Γ may
be easily computed. First by multiplying (from the right) by 1, we obtain
I 1 = 1 = δ (D.35)
In obtaining this result we note that
1
T
1 = 4 (D.36)
and
x
T
i
1 = 4 x
0
i
(D.37)
which gives
(x
i
− x
0
i
1)
T
1 = 0 (D.38)
Finally, we note that
h
T
1 = 0 (D.39)
APPENDIX D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 144
Next by multiplying (again from the right) by h, we get
h
T
h = 4 (D.40)
I h = h =
2

i=1
x
h
i
b
i
+ 4 Γ (D.41)
where
2
x
h
i
= x
T
i
h (D.42)
Thus, the parameters for Γ are computed as
Γ =
1
4
[h −
2

i=1
x
h
i
b
i
] (D.43)
It remains to compute the b
i
.
D.3 Derivatives of Alternative Formulation
Using the alternative expression for the shape functions, the derivatives with respect
to the global coordinates, x
i
, are given by
∂N
I
∂x
i
= b
iI
+ Γ
I
∂h
∂x
i
(D.44)
where the b
iI
are constant over the entire element and are computed by the conventional
expressions at the center of the element. The derivatives of the function h may also be
computed using the chain rule and are given by

x
h = ∇
ξ
hJ
−1
(D.45)
For the speciﬁc functional expression for h, the gradient with respect to the natural
coordinates is given by

ξ
h =
_
ξ
2
ξ
1
_
(D.46)
Furthermore, the inverse for the jacobian matrix is given by
J
−1
=
1
j(ξ)
_
∂x
2
∂ξ
2

∂x
1
∂ξ
2

∂x
2
∂ξ
1
∂x
1
∂ξ
1
_
(D.47)
2
The factor x
h
i
is sometimes called an hour glass shape, and when the coordinate, x, is replaced
by the displacement, u, the factor u
h
i
, deﬁnes the magnitude of the hour glass mode.
APPENDIX D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 145
where j(ξ) is the determinant of the jacobian transformation matrix, J. Recall that the
derivative of a global coordinate with respect to a natural coordinate has a constant
and a linear part. For the speciﬁc form of the h(ξ) function the product of the linear
part vanishes and the relationship for the gradient simpliﬁes to

x
h =
j
0
j(ξ)

ξ
hJ
−1
0
(D.48)
where j
0
is the value of the jacobian determinant evaluated at the element center. The
jacobian determinant at the center of the element is computed to be
j
0
= (J
0
)
11
(J
0
)
22
− (J
0
)
21
(J
0
)
12
(D.49)
We note also that the jacobian determinant at any location in the element may be
expressed as
j(ξ) = j
0
+ j
1
ξ
1
+ j
2
ξ
2
(D.50)
where
j
1
= (J
0
)
11
∆J
22
− (J
0
)
21
∆J
12
(D.51)
j
2
= ∆J
11
(J
0
)
22
− ∆J
21
(J
0
)
12
(D.52)
With the above deﬁnitions and
b
I
=
_
b
1I
b
2I
_
(D.53)
the gradient of the displacement may be written as

x
u = ∇
x
N
I
u
I
=
_
b
I
+
j
0
j(ξ)

ξ
hJ
−1
0
Γ
I
_
u
I
(D.54)
The structure of this representation is useful knowledge when we consider the construc-
tion of the enhanced part of the strains in Chapter 8.
Appendix E
Properties for J
2
plasticity models
The solution of the J
2
functions in the form
∂f
∂Σ
= n (E.1)
where
n =
Σ
Σ
(E.2)
and
Σ = s − α (E.3)
We note that n has the properties
1
T
n = 0 ; n
T
n = 1 (E.4)
In the derivation of the tangent the derivative of n leads to

2
f
∂Σ∂Σ
=
∂n
∂Σ
=
1
Σ
(1 − nn
T
) (E.5)
which appears in several location in the tangent matrices. The inversion of the tangent
matrices may be simpliﬁed using the Sherman-Morrison-Woodbury formula which is
described on page 51 in Reference [6].
(A + UV
T
)
−1
= A
−1
− A
−1
UWV
T
A
−1
(E.6)
where
W = (I + V
T
A
−1
U)
−1
(E.7)
In the above A is an ntimesn matrix, U, V are n×k matrices, where k ≤ n, and W
is a k × k matrix. The inverse may be proved by multiplying the results together to
recover the identity matrix. In the case of the deviatoric model A is diagonal and U
146
APPENDIX E. PROPERTIES FOR J
2
PLASTICITY MODELS 147
and V are proportional to n which is rank 1, thus leading to a scalar W (i.e., a 1 ×1
matrix).
There are some properties which need to be noted:
nn
T
(nn
T
) = nn
T
(E.8)
(I − nn
T
) n = 0 (E.9)
and
(I − nn
T
) (I − nn
T
) = I − nn
T
(E.10)
E.1 Example 1
Consider the matrix
H
1
= AI + Bnn
T
(E.11)
Using the Sherman-Morrison-Woodward formula the inverse is given by noting that U
is equal to B
n
and V is equal to n, thus
H
−1
1
=
1
A
I − (
1
A
I) (Bn)W n
T
(
1
A
I) (E.12)
where
W = (1 +
B
A
)
−1
=
A
A + B
(E.13)
The above simpliﬁes to
H
−1
1
=
1
A
(I −
B
A + B
nn
T
) (E.14)
E.2 Example 2
Consider the matrix
H
2
= C I + D(I − nn
T
) (E.15)
which may be rewritten as
H
2
= (C + D) I − Dnn
T
(E.16)
for which the solution from example 1 gives
H
−1
2
=
1
C + D
(I +
D
C
nn
T
) (E.17)
APPENDIX E. PROPERTIES FOR J
2
PLASTICITY MODELS 148
Recollecting into the original type of matrices gives
H
−1
2
=
1
C
[I −
D
C + D
(I − nn
T
)] (E.18)
A slightly more general form for an inverse results in considering the case with kinematic
hardening. In this case we encounter a matrix of the form
H =
_
AI + B(I −nn
T
) C (I −nn
T
)
D(I −nn
T
) E I + F (I −nn
T
)
_
(E.19)
The inverse may be written as
H
−1
=
_
a I + b (I −nn
T
) c (I −nn
T
)
d (I −nn
T
) e I + f (I −nn
T
)
_
(E.20)
where
a =
1
A
; e =
1
E
(E.21)
and the remaining coeﬃcients obtained by solving the small matrix problem
_
A + B C
D E + F
_ _
b c
d f
_
= −
_
B C
D F
_ _
a 0
0 e
_
(E.22)
The solution to (A.11b) is given by
_
b c
d f
_
= −
1
G
_
E + F −C
−D A + B
_ _
B C
D F
_ _
a 0
0 e
_
(E.23)
where
G = (A + B) (E + F) − C D (E.24)
The inverse may be proved by multiplying the two matrices together and show that
the result is an identity matrix.
Appendix F
Matrix Form for Equations of Solids
F.1 Stress and Strain
Generally the equations of mechanics are expressed using tensor forms. However, it
is traditional for ﬁnite element calculations to be performed using matrix forms. This
appendix summarizes the transformation of quantities from tensor to matrix form. We
begin by writing the forms for stress and strain in a matrix form involving both 9 and 6-
component forms. The advantage of using the 9-component form is not apparent until
considering constitutive equations where direct use of the transformation between the
two forms avoids possibility of errors by factors of two.
First we show the transformation for the stress and strain tensors into their matrix
representations. Here, for example, the components of the stress in tensor form may
be given as
σ
ij
=
_
_
σ
11
σ
12
σ
13
σ
21
σ
22
σ
23
σ
31
σ
32
σ
33
_
_
(F.1)
and reordered into the 9-component vector as
σ =
_
σ
11
σ
22
σ
33
σ
12
σ
21
σ
23
σ
32
σ
31
σ
13
¸
T
(F.2)
Conservation of angular momentum requires the stress to be symmetric, thus satisfying
σ
ij
= σ
ji
(F.3)
This permits the independent components of stress to be written in a 6-component
matrix form as
σ =
_
σ
11
σ
22
σ
33
σ
12
σ
23
σ
31
¸
T
(F.4)
In the sequel we shall use an underscore to indicate a 9-component form and omit the
underscore for the 6-component form.
149
APPENDIX F. MATRIX FORM FOR EQUATIONS OF SOLIDS 150
The 6-component form may be related to the 9-component form using a simple pro-
jector matrix, P, deﬁned by
P =
1
2
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
2 0 0 0 0 0
0 2 0 0 0 0
0 0 2 0 0 0
0 0 0 1 0 0
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 1 0
0 0 0 0 0 1
0 0 0 0 0 1
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(F.5)
giving
σ = P
T
σ (F.6)
In a similar manner we can write the components of the strain tensor as

ij
=
_
_

11

12

13

21

22

23

31

32

33
_
_
(F.7)
and reordered into the 9-component vector as
=
_

11

22

33

12

21

23

32

31

13
¸
T
(F.8)
Strain-displacement relations give symmetry of strain as

ij
=
ji
(F.9)
This permits the independent components of strain to be written in a 6-component
matrix form as
=
_

11

22

33
γ
12
γ
23
γ
31
¸
T
(F.10)
where γ
iij
are the engineering components of the shearing strain given by
γ
ij
= 2
ij
(F.11)
F.2 Split into Deviatoric and Spherical Components
Using the matrix form we can write the split of stress and strain in their deviator and
spherical components as
σ = s +mp (F.12)
APPENDIX F. MATRIX FORM FOR EQUATIONS OF SOLIDS 151
and
= e +
1
3

v
(F.13)
where p and ε
v
are the pressure and volume change, respectively, given by
p =
1
3
m
T
σ (F.14)
and
ε
v
= m
T
(F.15)
The matrix m is a trace projector deﬁned by
m =
_
1 1 1 0 0 0 0 0 0
¸
(F.16)
The splits may also be written in 6-component form as
σ = s +mp (F.17)
and
= e +
1
3

v
(F.18)
where
s =
_
s
11
s
22
s
33
s
12
s
23
s
31
¸
T
(F.19)
and
e =
_
e
11
e
22
e
33
2 e
12
2 e
23
2 e
31
¸
T
(F.20)
These also are related to their 9-component form using the P projector and may be
written as
s = P
T
s and e = Pe (F.21)
The 6-component projector mis likewise related to its 9-component counterpart through
m = P
T
m =
_
1 1 1 0 0 0
¸
T
(F.22)
Using the above matrix forms we can obtain expressions for the deviatoric stress and
strain matrices in terms of the full stress and strain values. Accordingly, for the stress
we have the two relations
σ = s +
1
3
mm
T
σ (F.23)
and
σ = s +
1
3
mm
T
σ (F.24)
which solve to give
s = σ −
1
3
mm
T
σ =
_
I −
1
3
mm
T
_
σ (F.25)
APPENDIX F. MATRIX FORM FOR EQUATIONS OF SOLIDS 152
and
s = σ −
1
3
mm
T
σ =
_
I −
1
3
mm
T
_
σ (F.26)
where I and I are identity matrices of size 9 and 6, respectively. We deﬁne the two
deviatoric projectors as
I
dev
= I −
1
3
mm
T
and I
dev
= I −
1
3
mm
T
(F.27)
Similarly for strains we have the deviatoric relations
e = −
1
3
mm
T
= I
dev
(F.28)
and
e = −
1
3
mm
T
= I
dev
(F.29)
F.3 Linear Elastic Constitutive Equations
Let us now consider the relations for linear elastic constitutive equations. In index
notation these are expressed as
σ
ij
= C
ijkl

kl
(F.30)
where C
ijkl
are the elastic moduli and possess the minor symmetries
C
ijkl
= C
jikl
= C
ijlk
(F.31)
From notions of hyperelasticity where stress is deduced from the stored energy function
W() as
σ
ij
=
∂W

ij
(F.32)
the elastic constants also possess the major symmetries
C
ijkl
= C
klij
(F.33)
We introduce the matrix forms for linear elasticity as
σ = D (F.34)
and
σ = D (F.35)
where D is a 9 × 9 matrix of elastic constants and D is a 6 × 6 matrix of elastic
constants.
APPENDIX F. MATRIX FORM FOR EQUATIONS OF SOLIDS 153
Construction of D follows directly from C
ijkl
using the index maps shown in Table F.1.
Applying the projector rules (which shows why we only need the two forms given above)
we obtain
σ = P
T
σ = P
T
D = P
T
DP = D (F.36)
which gives the relation between the two elastic moduli as
D = P
T
DP (F.37)
Entries in D use the index maps shown in Table F.2.
F.3.1 Example: Isotropic behavior
As an example we consider the isotropic linear elastic relations expressed in terms of
the Lam`e parameters as
σ
ij
= λδ
ij
ε
v
+ 2 µ
ij
(F.38)
where in index form ε
v
=
kk
. Writing the relationship for the constitution as
σ
ij
= C
ijkl

kl
(F.39)
we obtain the tensor form of the elastic moduli as
C
ijkl
= λδ
ij
δ
kl
+ 2 µI
ijkl
(F.40)
where I
ijkl
is the rank-4 tensor identity. This may be directly related to a matrix form
as
σ = D (F.41)
where
D = λmm
T
+ 2 µI (F.42)
Form Index
Matrix 1 2 3 4 5 6 7 8 9
Tensor 11 22 33 12 21 23 32 31 13
Table F.1: Matrix and tensor index maps
Form Index
Matrix 1 2 3 4 5 6
Tensor 11 22 33 12 & 21 23 & 32 31 & 13
Table F.2: Matrix and tensor index maps
APPENDIX F. MATRIX FORM FOR EQUATIONS OF SOLIDS 154
Applying the projector as indicated in Eq. F.37 we obtain the 6 ×6 matrix form as
D = λmm
T
+ 2 µP
T
P (F.43)
where m is given by Eq. F.22 and
P
T
P = I
0
=
1
2
_
¸
¸
¸
¸
¸
¸
_
2 0 0 0 0 0
0 2 0 0 0 0
0 0 2 0 0 0
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
_
¸
¸
¸
¸
¸
¸
_
(F.44)
Thus we can also write Eq. F.43 as
D = λmm
T
+ 2 µI
0
(F.45)
We note that this gives the shear equations with the correct factors to match the use
of the engineering components. While this may be obtained also by merely writing
Eq. F.38 for each of the independent stress components and introducing the deﬁnition
for engineering shearing strain, the above process provides a direct way to construct
the constitutive model for a wide range of material behavior. One of which is classical
elasto-plasticity which we will consider later.

Contents
1 Introduction 2 Introduction to Strong and Weak Forms 2.1 2.2 2.3 2.4 2.5 2.6 Strong form for problems in engineering . . . . . . . . . . . . . . . . . 1 3 3 4 5 7 8 12 16 16 17 20 21 23 26 26 28 29

Construction of a weak form . . . . . . . . . . . . . . . . . . . . . . . . Heat conduction problem: Strong form . . . . . . . . . . . . . . . . . . Heat conduction problem: Weak form . . . . . . . . . . . . . . . . . . . Approximate solutions: The ﬁnite element method . . . . . . . . . . . . Implementation of elements into FEAP . . . . . . . . . . . . . . . . . .

3 Introduction to Variational Theorems 3.1 3.2 3.3 Derivatives of functionals: The variation . . . . . . . . . . . . . . . . . Symmetry of inner products . . . . . . . . . . . . . . . . . . . . . . . . Variational notation . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4 Small Deformation: Linear Elasticity 4.1 Constitutive Equations for Linear Elasticity . . . . . . . . . . . . . . .

5 Variational Theorems: Linear Elasticity 5.1 5.2 5.3 Hu-Washizu Variational Theorem . . . . . . . . . . . . . . . . . . . . . Hellinger-Reissner Variational Theorem . . . . . . . . . . . . . . . . . . Minimum Potential Energy Theorem . . . . . . . . . . . . . . . . . . . i

CONTENTS 6 Displacement Methods 6.1 6.2 6.3 6.4 6.5 6.6 6.7 External Force Computation . . . . . . . . . . . . . . . . . . . . . . . . Internal Force Computation . . . . . . . . . . . . . . . . . . . . . . . . Split into Deviatoric and Spherical Parts . . . . . . . . . . . . . . . . . Internal Force - Deviatoric and Volumetric Parts . . . . . . . . . . . . . Constitutive Equations for Isotropic Linear Elasticity . . . . . . . . . . Stiﬀness for Displacement Formulation . . . . . . . . . . . . . . . . . . Numerical Integration . . . . . . . . . . . . . . . . . . . . . . . . . . .

ii 31 32 32 34 37 37 40 41 45 45 50 52 56 60 62 . . . . . . . . . 62 66 67 69 69 74 74 80 80 82

7 Mixed Finite Element Methods 7.1 7.2 7.3 7.4 Solutions using the Hu-Washizu Variational Theorem . . . . . . . . . . Finite Element Solution for Mixed Formulation . . . . . . . . . . . . . Mixed Solutions for Anisotropic Linear Elastic Materials . . . . . . . . Hu-Washizu Variational Theorem: General Problems . . . . . . . . . . 7.4.1 Example: Interpolations linear for u and constant φ . . . . . . .

8 Enhanced Strain Mixed Method 8.1 8.2 8.3 8.4 8.5 Hu-Washizu Variational Theorem for Linear Elasticity

Stresses in the Enhanced Method . . . . . . . . . . . . . . . . . . . . . Construction of Enhanced Modes . . . . . . . . . . . . . . . . . . . . . Non-Linear Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . Solution Strategy: Newton’s Method . . . . . . . . . . . . . . . . . . .

9 Linear Viscoelasticity 9.1 Isotropic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10 Plasticity Type Formulations 10.1 Plasticity Constitutive Equations . . . . . . . . . . . . . . . . . . . . . 10.2 Solution Algorithm for the Constitutive Equations . . . . . . . . . . . .

. . . . . . . . . . . 11. . . . . . . . . . . . . 116 13. . . .11Stress Divergence Matrix . . . . .8 Linearized Equations . . 111 13. . . . . . . 101 12. . . 110 13. . . . . . . . . . .4 Boundary Conditions . . .1 Adding the transient terms .Introduction . . . . . . 112 13. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Kinematics and Deformation . . . . . . 114 13. . . . . . . . . . . . . . . . . .4 Isotropic viscoplasticity: J2 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 13. . .CONTENTS 10. . . . . . . . . . . . . . . . . . . . . . . . . . 11. . . . . . . .Finite Elasticity . . . .3 Augmented Lagrangian Method for Constraints . . . . .12Geometric stiﬀness . . . .1 Constraint Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Balance of Momentum . . . . . . . . . . . . . . . . . . . .2 Stress and Traction Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 13. . . 117 13. . . . . . . . . . . . . .7 Variational Description . .14Loading terms . . 108 13.3 Isotropic plasticity: J2 Model . . . 12 Transient Analysis iii 85 91 93 93 94 97 100 12. . . 11 Augmented Lagrangian Formulations 11. . . . . . . . . . . . . . .9 Element Technology . . . . . . 121 . . . 120 13.standard B matrix formulation . . . . . . . . . . . . . . . . . . .10Consistent and Lumped Mass Matrices . . . . . . . . . . . . .6 Material Constitution . . . . . . . 105 13. . . . . . . .2 Mixed Penalty Methods for Constraints . . . . . . . . . . . . . . . . . . . . . . . 104 13 Finite Deformation 105 13. . . . . . . . .2 Newmark Solution of Momentum Equations . . . . . . . 120 13. . . . . .3 Hilber-Hughes-Taylor (HHT) Algorithm . . . . . . . . . 119 13. . .5 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13Material tangent matrix . . . . . . 100 12. . . . . . . . . . 10.

. . . . .4 Shell elements . . . . . . . .3 Linear Elastic Constitutive Equations . . 138 C. . . 152 F. . . . . . . . 144 E Properties for J2 plasticity models 146 E. . . . . . 136 C Structural Elements C. . . . . . . . . . .1 Example 1 . . . .2 Example 2 . . . . . . .1 Stress and Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 D Isoparametric Shape Functions for Elements 140 D. . . . . . . . . . . . . . . . . . . . .2 Frame elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 . . . . . . . . 122 A Heat Transfer Element B Solid Elements 128 136 B. . . . . . . . . . .CONTENTS iv 13.16Mixed formulation .3 Plate elements .1 Truss elements 138 . . . . . . . . . . . . . . . . . . . . . . 147 E. . . . . . . . . . . . . 149 F. . . . . . . . . . . . 121 13. . . . . . . . . . . . . . . . . . . . . . . . . . .2 Alternative Representation in Two Dimensions . . . . 150 F. . . . . . . .1 Small displacement element . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15Basic ﬁnite element formulation . . . . . . . . . . . . . . . . . . . . . . . .3 Derivatives of Alternative Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 C. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Example: Isotropic behavior . . . .3. . . . . . . . 147 F Matrix Form for Equations of Solids 149 F. . . . . . .1 Conventional Representation . . . . . . . . . . . . . . . . . 142 D.2 Split into Deviatoric and Spherical Components . 138 C. . . . . . . . . . . . . . . . . 140 D. . . . . . .2. .1 Displacement elements . . . . . . . 138 C. . . . . . . . . . .

Chapters 4 and 5 provides a summary of the linear elasticity problem in its strong and weak forms. This report presents the background necessary to understand the formulations which are employed to develop the two and three dimensional continuum elements which are provided with the FEAP system. Special attention is given to methods which can handle anisotropic elastic models where the elasticity tangent matrix is fully populated. Chapter 6 discusses implentation for displacement (irreducible) based ﬁnite element methods. Chapters 7 and 8 then discuss alternative mixed methods for treating problems which include constraints leading to near incompressibility. whereas. The strong form of a problem is given as a set of partial diﬀerential equations. This is an essential feature required to handle both inelastic and non-linear constitutive models. Chapter 9 presents a generalization of the linear elastic constitutive model to that for linear viscoelasticity.Chapter 1 Introduction The Finite Element Analysis Program FEAP may be used to solve a wide variety of problems in linear and non-linear solid continuum mechanics. The linear heat equation is used as an example problem to describe some of the details concerning use of strong and weak forms. In this report. Chapters 2 and 3 provide an introduction to problem formulation in both a strong and a weak form. A variational statement provides a convenient basis for constructing the ﬁnite element model. General mixed and enhanced strain methods are presented as alternatives to develop low order ﬁnite elements that perform well at the nearly incompressible regime. Vainberg’s theorem is introduced to indicate when a variational theorem exists for a given variational equation. the weak form of a problem is associated with either variational equations or variational theorems. For applications involving an isotropic model and strong 1 . Companion manuals are available which describe the use of the program [21] and information for those who wish to modify the program by adding user developed modules [20].

Full details are provided for the case of isotropic models. . Chapter 10 presents the general algorithm employed in the FEAP system to model plasticity type presentations. The Newmark method and some of its variants (e. Alternative representations for linear viscoelastic behavior are presented in the form of diﬀerential models and integral equations. as well as.CHAPTER 1. A discussion is presented for both rate and rate independent models. Chapter 11 discusses methods used in FEAP to solve constraints included in a ﬁnite element model. INTRODUCTION 2 deviatoric relaxation compared to the spherical problem. and enhanced strain methods. for the problem of intermittant contact between contiguous bodies.g. a situation can arise at large times in which the response is nearly incompressible – thus requiring use of elements that perform well in this regime. A ﬁnal option is the use of Lagrange multipliers to include the constraint. Such constraints are evident in going to the fully incompressible case. The latter provides a basis for constructing an accurate time integration method which is employed in the FEAP system. The simplest approach is use of a penalty approach to embed the constraint without the introduction of additional parameters in the algebraic problem.. Chapter 12 presents a discussion for extension of problems to the fully transient case. as well as. Chapter 13 presents a summary for extending the methods discussed in the ﬁrst twelve chapters to the ﬁnite deformation problem. The formulation used is based on a return map algorithm for which analytic tangent matrices for use in a Newton solution algorithm can be obtained. All of these methods are used as part of the FEAP system. The chapter presents a summary for diﬀerent deformation and stress measures used in solid mechanics together with a discussion on treating hyper-elastic constitutive models. An extension using the Uzawa algorithm for an augmented Lagrangian treatment is then considered and avoids the need for large penalty parameters – which can lead to numerical ill-conditioning of the algebraic problem. an energy-momentum conserving method) are discussed as methods to solve the transient algorithm by a discrete time stepping method. for a generalized plasticity model. Finally. mixed. It is shown that general elements which closely follow the representations used for the small deformation case can be developed using displacement.

1) deﬁned on the region 0 ≤ x ≤ a.one case being the linear wave equation in one space dimension and time. Linear equations are characterized by the appearance of the dependent variable(s) in linear form only. y) ∂x2 ∂y 2 (2. in general it is not possible to treat general boundary conditions or problem shapes using this approach. Some equations admit use of solutions written as series of products of one dimensional functions for which exact solutions may be constructed for each function. 0 ≤ y ≤ b with the boundary condition u = 0 on all edges.1 Strong form for problems in engineering Many problems in engineering are modeled using partial diﬀerential equations (PDE). The diﬀerential equations may be either linear or non-linear. Very few partial diﬀerential equations may be solved in closed form . As an example consider the Poisson equation ∂2u ∂2u + = q(x.2) which when substituted into the equation yields 3 . Again. This diﬀerential equation may be solved by writing u as a product form u= m n sin( nπy mπx ) sin( )umn a b (2.Chapter 2 Introduction to Strong and Weak Forms 2. non-linear equations include nonlinear terms also. The set of partial diﬀerential equations describing such problems is often referred to as the strong form of the problem. whereas.

Integrate the result of 1. use of a ﬁnite set of terms leads to an approxiamte solution with the accuracy depending on the number of terms used. Replace the boundary conditions by an appropriate construction. it is not possible to get an exact solution in closed form. 2. Indeed. over the domain of consideration.e. y) of interest. Multiply the diﬀerential equation by an arbitrary function which contracts the equations to a scalar. More general solutions may be constructed using separable solution. Accordingly. Integrate by parts using Green’s theorem to reduce derivatives to their minimum order. Indeed. Consequently. In the sequel. It is well established that the ﬁnite element method is one of the most powerful methods to solve general problems represented as sets of partial diﬀerential equations. . we will be concerned with the construction of approximate solutions based on the ﬁnite element method. Meshes constructed by subdivision are sometimes referred to as a Ritz sequence due to their similarity with solutions constructed in series form from variational equations. INTRODUCTION TO STRONG AND WEAK FORMS 4 m n mπ a 2 + nπ b 2 sin( mπx nπy ) sin( )umn = q(x. Evaluation of the solution requires the summation of the series for each point (x. This is similar to a series solution in that each mesh used to construct an FE solution represents a particular number of terms. y) a b (2.2 Construction of a weak form A weak form of a set of diﬀerential equations to be solved by the ﬁnite element method is constructed by considering 4 steps: 1. if sequences of meshes are constructed by subdivision the concept of a series is also obtained since by constraining the added nodes to have values deﬁned by a subdivision the results for the previous mesh is recovered .CHAPTER 2. the solutions are obtained only in series form. 2.in essence this is the result for fewer terms in the series.3) The solution may now be completed by expanding the right hand side as a double sine series (i.. 3. The weak form will be the basis for constructing our ﬁnite element solutions. again. Ω. however. 4. Fourier series) and matching terms between the left and right sides. we now direct our attention to rewriting the set of equations in a form we call the weak form of the problem.

i = and in time by ∂T ˙ T = ∂t In addition. Ω. T is temperature. and q n = qi n i = qn ¯ (2. Q is the volumetric heat generation per unit volume per unit time.6) ai b i = i=1 ai b i (2.. ρ is density. the divergence of the ﬂux may be written as d qi. The equations hold for all points xi in the domain of interest.5) (2.i = i=1 ∂qi ∂xi (2.3 Heat conduction problem: Strong form The above steps are made more concrete by considering an example. and t is time. INTRODUCTION TO STRONG AND WEAK FORMS 5 2. c is speciﬁc heat. The following notation is introduced for use throughout this report.8) Boundary conditions are given by ¯ T (xj .4) where: d is the spatial dimension of the problem.CHAPTER 2.7) With this notation. qi is the component of the heat ﬂux in the xi direction. t) = T ¯ where T is a speciﬁed temperature for points xj on the boundary. summation convention is used where d ∂( · ) ∂xi (2. Partial derivatives in space will be denoted by ( · ). ΓT . The governing partial diﬀerential equation set for the transient heat conduction equation is given by d − i=1 ∂qi ∂T + Q = ρc ∂xi ∂t (2.10) (2.9) .

2. Initial conditions are given by ¯ T (xi .CHAPTER 2. For an isotropic material kij = kδij (2. second rank thermal conductivity tensor.i + Q = ρcT (2. it is only necessary to approximate ﬁrst derivatives of functions to obtain a solution. Γq . .i ).12) where kij is a symmetric.13) in which δij is the Kronecker delta function (δij = 1 for i = j.4. Ω. we show that. expressed as a weak form. 0) = T0 (xi ) (2. INTRODUCTION TO STRONG AND WEAK FORMS 6 where q nn is a speciﬁed ﬂux for points xj on the ﬂux boundary. The equations are completed by giving a relationship between the gradient of temperature and the heat ﬂux (called the thermal constitutive equation). In the following. The Fourier law is a linear relationship given as qi = − kij T. at time zero.15) The equation is a second order diﬀerential equation and for isotropic materials with constant k is expanded for two dimensional plane bodies as k ∂2T ∂2T + ∂x2 ∂x2 1 2 + Q = ρc ∂T ∂t (2.11) for points in the domain.16) We note that it is necessary to compute second derivatives of the temperature to compute a solution to the diﬀerential equation.i (2.14 into Eq. Hence for an isotropic material the Fourier law becomes qi = − kT. The result is ˙ (kT. Thus. = 0 for i = j).j (2. 2.14) The diﬀerential equation may be expressed in terms of temperature by substituting Eq. the solution process is simpliﬁed by considering weak (variational) forms. The partial diﬀerential equation together with the boundary and initial conditions is called the strong form of the problem. and ni are direction ¯ cosines of the unit outward pointing normal to the boundary.

i dΩ = Ω Γ φ ni d Γ (2. φ is the product of two functions. qi .4 Heat conduction problem: Weak form In step 1.i V dΩ + Γ (U V )ni dΓ (2.i dΩ = Ω Γ (2. Applying the integration by parts to the heat equation gives G(W. which transforms the set of diﬀerential equations onto a scalar function.24) + Γ W qi ni dΓ = 0 .19) Normally.20) (U V )ni dΓ (2. Green’s theorem is given by φ. INTRODUCTION TO STRONG AND WEAK FORMS 7 2.18) In step 3 we integrate by parts the terms involving the spatial derivatives (i.17) ˙ W (xi ) ρcT − Q + qi. T ) = Ω ˙ W (xi ) ρcT − Q dΩ − Ω W. Thus ˙ g(W. T ) = W (xi ) ρcT − Q + qi..23) which we observe is an integration by parts. qi .21) The left hand side expands to give [U V. Thus. the thermal ﬂux vector in our case). we multiply Eq. T ) = Ω = 0 (2. Thus for φ = VU we have (U V ).i V ] dΩ = Ω Γ (U V )ni dΓ (2.i dΩ = − Ω Ω U. Ω.i dΩ = 0 (2.i In step 2 we integrate over the domain. 2.22) which may be rearranged as U V. G(W. qi .e. The equation is ﬁrst written on one side of an equal sign.i qi d Ω (2.i + U.CHAPTER 2.4 by an arbitrary function W (xi ).

This leads to weaker conditions to deﬁne solutions of the problem and thus the notion of a weak form is established.24 completes step 4 and we obtain the ﬁnal expression G(W.27) + Γq W qn dΓ = 0 ¯ We note that the above form only involves ﬁrst derivatives of quantities instead of the second derivatives in the original diﬀerential equation.10 may be used for the part on Γq and (without any loss in what we need to do) we can set W to zero on Γu (Note that W is arbitrary. Eq. 2.28) . the boundary term may be split into two parts and expressed as W qn dΓ = Γ ΓT 8 W qn dΓ + Γq W qn dΓ (2. 2. there are no additional equations that can be used to give any additional reductions.26) + Γq W qn dΓ = 0 ¯ If in addition to the use of the boundary condition we assume that the Fourier law is satisﬁed at each point in Ω the above integral becomes G = Ω W ˙ ρ c T − Q dΩ + Ω W. qi . we deﬁne each integral as a sum of integrals over each element. hence our equation must be valid even if W is zero for some parts of the domain).i qi dΩ (2. we let Nel Ω ≈ Ωh = e=1 Ωe (2. Accordingly. INTRODUCTION TO STRONG AND WEAK FORMS Introducing qn .5 Approximate solutions: The ﬁnite element method For ﬁnite element approximate solutions. Substituting all the above into Eq.i k T. thus. Furthermore.25) Now the boundary condition Eq. 2. 2. T ) = Ω ˙ W (xi ) ρcT − Q dΩ − Ω W.27 is said to be irreducible [26. Chapter 9].i dΩ (2.CHAPTER 2.

CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 9 where Ωh is the approximation to the domain created by the set of elements.i qi dΩ (2.32) for coordinates and Nel T = I=1 NI (ξ)T I (t) (2. Ωe is the domain of a typical element and Nel is the number of nodes attached to the element. Commonly. Standard element interpolation functions which maintain C 0 continuity are discussed in any standard book on the ﬁnite element method (e. See [26. With this approximation. deﬁnes a class called a C 0 function. but not its ﬁrst derivatives..30) + e=1 Γeq W qn dΓ = 0 ¯ Introducing the Fourier law the above integral becomes Nel Nel G ≈ Gh = e=1 Nel Ωe W ˙ ρcT − Q dΩ + e=1 Ωe W. Chapter 7]). which maintain the C 0 condition.31) + e=1 Γeq W qn dΓ = 0 ¯ In order for the above integrals to be well deﬁned.g. Isoparametric elements.29) Thus our heat equation integral becomes Nel Nel G ≈ Gh = e=1 Nel Ωe W ˙ ρcT − Q dΩ − e=1 Ωe W.i dΩ (2. surface integrals between adjacent elements must vanish. the ﬁrst derivatives of W and T may be discontinuous in Ω. The case where only the function is continuous.33) . satisfy the conditions Nel xi = I=1 NI (ξ)xI i (2. Integrals may now be summed over each element and written as Nel (·) dΩ ≈ Ω Ωh (·) dΩ = e=1 Ωe (·) dΩ (2. This occurs under the condition that both W and T are continuous in Ω.i kT. the ﬁnite element method uses isoparametric elements to construct C 0 functions in Ωh .

CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS

10

for temperature. Similar expressions are used for other quantities also. In the above, I refers to a node number, NI is a speciﬁed spatial function called a shape function for node I, ξ are natural coordinates for the element, xI are values of the coordinates at i node I, T I (t) are time dependent nodal values of temperature, and nel is the number of nodes connected to an element. Standard shape functions, for which all the nodal parameters have the value of approximations to the variable, satisfy the condition
Nel

NI (ξ) = 1
I=1

(2.34)

This ensures the approximations contain the terms (1, xi ) and thus lead to convergent solutions. In summation convention, the above interpolations are written as xi = NI (ξ) xI i and T = NI (ξ) T I (t) The weight function may also be expressed as W = NI (ξ) W I (2.37) (2.36) (2.35)

where W I are arbitrary parameters. This form of approximation is attributed to Galerkin (or Bubnov-Galerkin) and the approximate solution process is often called a Galerkin method. It is also possible to use a diﬀerent approximation for the weighting functions than for the dependent variable, leading to a method called the PetrovGalerkin process. The shape functions for a 4-node quadrilateral element in two-dimensions may be written as 1 I I (1 + ξ1 ξ1 )(1 + ξ2 ξ2 ) (2.38) 4 where ξiI are values of the natural coordinates at node I. Later we also will use an alternative representation for these shape functions; however, the above suﬃces for most developments. Derivatives for isoparametric elements may be constructed using the chain rule. Accordingly, we may write NI (ξ) = ∂NI ∂NI ∂xj ∂NI = = Jji ∂ξi ∂xj ∂ξi ∂xj where the Jacobian transformation between coordinates is deﬁned by (2.39)

CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS

11

Jji =

∂xj ∂ξi

(2.40)

The above constitutes a set of linear equations which may be solved at each natural coordinate value (e.g., quadrature point) to specify the derivatives of the shape functions. Accordingly ∂NI ∂NI −1 = J (2.41) ∂xj ∂ξi ji Using the derivatives of the shape functions we may write the gradient of the temperature in two dimensions as T,x1 T,x2 NI,x1 I T (t) NI,x2

=

(2.42)

Similarly, the gradient of the weighting function is expressed as W,x1 W,x2 NI,x1 WI NI,x2

=

(2.43)

Finally the rate of temperature change in each element is written as ˙ ˙ T = NI (ξ) T I (t) (2.44)

With the above deﬁnitions available, we can write the terms in the weak form for each element as ˙ ˙ W ρcT dΩ = W I MIJ T J (2.45)
Ωe

where MIJ =
Ωe

NI ρ c NJ dΩ

(2.46)

deﬁnes the element heat capacity matrix. Similarly, the term W,i k T,i dΩ = W I KIJ T J
Ωe

(2.47)

where KIJ =
Ωe

NI,i k NJ,i dΩ

(2.48)

deﬁnes the element conductivity matrix. Finally, W Q dΩ −
Ωe Γeq

W qn dΓ = W I FI ¯

(2.49)

CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS where FI =
Ωe

12

NI Q dΩ −
Γe q

NI qn dΓ ¯

(2.50)

The approximate weak form may now be written as
Nel

Gh =
e=1

˙ W I (MIJ T J + KIJ T J − FI ) = 0

(2.51)

and since W I is an arbitrary parameter, the set of equations to be solved is
Nel

˙ (MIJ T J + KIJ T J − FI ) = 0
e=1

(2.52)

In matrix notation we can write the above as ˙ MT + KT = F (2.53)

which for the transient problem is a large set of ordinary diﬀerential equations to be solved for the nodal temperature vector, T. For problems where the rate of tempera˙ ture, T, may be neglected, the steady state problem KT = F results. (2.54)

2.6

Implementation of elements into FEAP

The implementation of a ﬁnite element development into the general purpose program FEAP (Finite Element Analysis Program) is accomplished by writing a subprogram named ELMTnn (nn = 01 to 50) [26, 27, 20]. The subroutine must input the material parameters, compute the ﬁnite element arrays, and output any desired quantities. In addition, the element routine performs basic computations to obtain nodal values for contour plots of element variables (e.g., the thermal ﬂux for the heat equation, stresses for mechanics problems, etc.). The basic arrays to be computed in each element for a steady state heat equation are KIJ =
Ωe

NI,i k NJ,i dΩ

(2.55)

and

where for example L KIJ = l=1 NI.59) Each array is computed for a single element as described in the section of the FEAP Programmer Manual on adding an element.1. INTRODUCTION TO STRONG AND WEAK FORMS 13 FI = Ωe NI Q dΩ − Γeq NI qn dΓ ¯ (2.i (ξ l )j(ξ l )wl (2. is included separately to permit solution of the general linear eigenproblem KΦ = MΦΛ (2. . One quadrant of the region is modeled as shown by the mesh in Figure 2.56) For a transient problem is is necessary to also compute MIJ = Ωe NI ρ c NJ dΩ (2. Both steady state and transient solutions are permitted. perform numerical quadrature. The routine uses basic features included in the FEAP system to generate shape functions. M. The listing included in Appendix A summarizes an element for the linear heat transfer problem. For the linear heat equation the residual may be expressed as ˙ R = F − KT − MT A solution to a problem is achieved when R = 0 (2.CHAPTER 2.58) where j(ξ) is the determinant of J evaluated at the quadrature point ξ l and wl are quadrature weights.61) which can be used to assess the values of basic time parameters in a problem. hence it needs to compute a residual for the equations (see FEAP User and Programmer Manual for details).57) The above integrals are normally computed using numerical quadrature.i (ξ l ) k NJ. The heat capacity array.60) (2. FEAP is a general non-linear ﬁnite element solution system. An example of a solution to a problem is the computation of the temperature in a rectangular region encasing a circular insulator and subjected to a thermal gradient. etc. The sides of the block are assumed to also be fully insulated.

2. The routines indicated in Tables A.1: Mesh for thermal example The top of the region is exposed to a constant temperature of 10C o and the symmetry axis is assumed to be at zero temperature. .CHAPTER 2.1 to A.5 are incorporated into FEAP as a user element and the steady state solution computed. INTRODUCTION TO STRONG AND WEAK FORMS 14 Figure 2. The contour of temperatures is shown in Figure 2.

CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 15 Figure 2.2: Temperature contours for thermal example .

.Chapter 3 Introduction to Variational Theorems 3. The function ητ is called the variation of the function T and often written as δT (τ (x) alone also may be called the variation of the function) [10].1) The function τ is an arbitrary function and is related to the arbitrary function W introduced in the construction of the weak form. Introducing the family of functions T η into the functional we obtain.i dΩ − Ω W Q dΩ (3. The notion of a variation is associated with the concept of a derivative of a functional (i. using the steady state heat equation as an example. the result Gη = G(W. a function of functions). Thus.e. it is necessary to introduce a scalar parameter which may be used as the limiting parameter in the derivative [10]. T η ) = Ω η W. This may be done by introducing a parameter η and deﬁning a family of functions given by T η (x) = T (x) + η τ (x) (3.1 Derivatives of functionals: The variation The weak form of a diﬀerential equation is also called a variational equation.i k T.2) + Γq W qn dΓ ¯ The derivative of the functional with respect to η now may be constructed using conventional methods of calculus. In order to construct a derivative of a functional.3) . dG Gη − G0 = lim η→0 dη η 16 (3.

The construction of the derivative of the functional requires the computation of variations of derivatives of T .4) dη dη With this result in hand. for the variational equation to be equivalent to the weak form τ must be an arbitrary function with the same restrictions as we established in deﬁning W .i dΩ Ω (3. the derivative of the functional with respect to η is given by dG = dη W. A variational theorem. Symmetry of the functional A also implies that the tangent matrix (computed from the second variation of the theorem or the ﬁrst variation of the weak form) of a Bubnov-Galerkin ﬁnite element method will be symmetric.1 leads to a result where τ replaces W in the weak form.7) Symmetry of the inner product resulting from the variation of a weak form is a suﬃcient condition for the existence of a variational theorem which may also be used to generate a weak form.i ) = τ. often we may only have a . hence the derivative is a variation of G. τ ) = A(τ. An inner product is symmetric if A(W. Using the above deﬁnition we obtain d(T η ). Thus. T ) η→0 dη lim (3.i + ητ.5) The limit of the derivative as η goes to zero is called the variation of the functional. For the linear steady state heat equation the derivative with respect to η is constant.i k τ. To investigate symmetry of a functional we consider only terms which include both the dependent variable and the arbitrary function. W ) (3.i d = (T. given by a functional Π(T ).8) Note that use of Eq. τ ) dη This is a notation commonly used to deﬁne inner products. has a ﬁrst variation which is identical to the weak form. Variational theorems are quite common for several problem classes. (3. We shall deﬁne the derivative of the functional representing the weak form of a diﬀerential equation as dG = A(W. T ) as dΠ(T η ) = G(τ. Thus.CHAPTER 3.6) 3.i (3. given a functional Π(T ) we can construct G(W. INTRODUCTION TO VARIATIONAL THEOREMS 17 where G0 is the value of Gη for η equal to 0. 3.2 Symmetry of inner products Symmetry of inner product relations is fundamental to the derivation of variational theorems. however.

11) Performing the variation of Π and setting to zero gives dΠ(T η ) = G(τ. If symmetric then to to 2. ηT )dη (3. 3. A variational theorem is a functional whose ﬁrst variation. The result of the above process gives 1 (3. Perform the following substitutions in G(W.12) and a problem commonly referred to as a variational theorem.).9) (3.i k η T. For the steady state heat equation we have G(T. INTRODUCTION TO VARIATIONAL THEOREMS 18 functional G and desire to know if a variational theorem exists.13) The integral is trivial and gives Π(T ) = 1 2 T.10) Π(T ) = 0 G(T.i kT. t) to deﬁne G(T. The construction of a variational theorem from a weak form is performed as follows [24]: 1. T ) = 0 η→0 dη lim (3.i dΩ − Ω T Q dΩ + Γq T qn dΓ ¯ (3. ηT ) 3. Integrate the functional result from (b) with respect to η over the interval 0 to 1. symmetry of the tangent matrices. when set to zero. yields the governing diﬀerential equations and boundary conditions associated with some problem.14) . Also. t) → ηT (x. T ) W (x) → T (x.g. etc. otherwise.8.CHAPTER 3. existence of a variational theorem yields a weak form directly by using Eq. 2. minimum or stationary value. In practice we seldom need to have the variational theorem.i dΩ − Ω Ω T QdΩ + Γq T qn dΓ ¯ (3. Check symmetry of the functional A(W. stop: no varitational theorem exists. t) T (x. but knowledge that it exists is helpful since it implies properties of the discrete problem which are beneﬁcial (e.. ηT ) = Ω T. τ ).

The ﬁrst variation is deﬁned by replacing T by T η = T + ητ (3. the variational theorem is called a minimum principle and the discrete tangent matrix is positive deﬁnite. If the second variation can have either positive or negative values the variational theorem is a stationary principle and the discrete tangent matrix is indeﬁnite. we can often restore symmetry to the functional and then deduce a variational theorem for the discrete problem. For example if at each time tn we have T (tn ) ≈ Tn then we can approximate the time derivative by the ﬁnite diﬀerence Tn+1 − Tn ˙ T (tn ) ≈ tn+1 − tn (3.18) If however.CHAPTER 3. The transient heat equation with weak form given by G = Ω W ˙ ρ c T − Q dΩ + Ω W.16) If the second variation is strictly positive (i.15) and performing the derivative deﬁned by Eq.17) + Γq W qn dΓ = 0 ¯ does not lead to a variational theorm due to the lack of the symmetry condition for the transient term ˙ A = T .i k T.e. T ) ˙ (3.i dΩ (3.12. τ ) (3.T ∆t A= = η (3. the ﬁrst variation of the variational theorem generates a variational equation which is the weak form of the partial diﬀerential equation..19) Letting tn+1 − tn = ∆t and omitting the subscripts for quantities evaluated at tn+1 . we ﬁrst discretize the transient term using some time integration method. ητ ∆t τ . the rate term which includes both T and τ becomes T . ητ = (η τ . The second variation of the theorem generates the inner product A(τ. A is positive for all τ ).21) . INTRODUCTION TO VARIATIONAL THEOREMS 19 Reversing the process.20) (3. 3.

This construction is a formal process as the indicated partial derivatives have no direct deﬁnition (indeed the result of the derivative is obtained from Eq.25) Performing the derivatives leads to δΠ = 1 2 (k T.24) ∂u ∂u.i k) δT.i dΩ − Ω ∂T. it is formally the ητ (x)). This formal construction is easy to apply but masks the meaning of a variation.i ∂ (T qn ) δT dΓ ¯ Γq ∂T Ω ∂ (T Q) δT dΩ ∂T (3. Similarly. A = (T. δu is the variation of the variable (i. u. τ ) (3.3 Variational notation A formalism for constructing a variation of a functional may be identiﬁed and is similar to constructing the diﬀerential of a function.26) Collecting terms we have δΠ = Ω δT. we may formally write a ﬁrst variation as ∂Π ∂Π δΠ = δu + δu. For the functional Eq. etc.i where u. . That is. 3. 3.i dΩ − Ω Q δT dΩ + Γq qn δT dΓ ¯ (3.i dΩ − Ω Ω Q δT dΩ + Γq qn δT dΓ ¯ (3.3). INTRODUCTION TO VARIATIONAL THEOREMS 20 since scalars can be moved from either term without aﬀecting the value of the term. We shall address this aspect at a later time. However.i k T.i + T.22) 3.i ) δT. applying this construction can be formally performed using usual constructions for a derivative of a function.27) which is identical to Eq.23) ∂xi where xi are the set of independent variables.i k T. 3.14. we obtain the result δΠ = + 1 2 ∂ (T. and δΠ is called the ﬁrst variation of the functional.i are the dependent variables of the functional. η τ ) = (η T.i + · · · (3.CHAPTER 3.e. We may also use the above process to perform linearizations of variational equations in order to construct solution processes based on Newton’s method..2 with δT replacing W . The diﬀerential of a function f (xi ) may be written as ∂f df = dxi (3.

which leads to symmetry of the stress tensor σ = σT 3. where the domain of analysis is Ω with boundary Γ.2) u.1) is the gradient 2. Balance of linear momentum expressed as ¨ · σ + ρ bm = ρ u where ρ is the mass density. the stress tensor. ¨ operator. u.3) u + (a) u u = 1 2 21 u + ( u)T (4. The basic governing equations are: 1.Chapter 4 Small Deformation: Linear Elasticity A summary of the governing equations for linear elasticity is given below. σ. Deformation measures based upon the gradient of the displacement vector. bm is the body force per unit mass. The presentation below assumes small (inﬁnitesimal) deformations and general three dimensional behavior in a Cartesian coordinate system. Chapter 6]. (4. and the strain tensor. For a presentation using indicial notation see [26. The dependent variables are given in terms of the displacement vector.4) . . x. Balance of angular momentum. and u is the acceleration. (4. which may be split as follows u = where the symmetric part is (s) (s) (4. The equations are presented using direct notation.

ω = (a) u (4. ωij = ωji which implies ω11 = ω22 = ω33  ω11 ω12 ω21 ω22 ω → ω31 ω32 = 0. However.11) and the symmetry condition ij = ji (4.CHAPTER 4.2 u2.7) In a three dimensional setting the above tensors have 9 components.13)  ω13 ω23  0 (4.1 (4. The component ordering for each of the tensors is given by   σ11 σ12 σ13 σ → σ21 σ22 σ23  (4.10) The strain tensor is the symmetric part with components   11 12 22 32 13 23 →  21 31 33  (4.   ω13 0 ω12  = −ω12 ω23 0 ω33 −ω13 −ω23 (4.5) Based upon this split.6) and the skew symmetric part deﬁnes the spin.3 u2.8) σ31 σ32 σ33 which from the balance of angular momentum must be symmetric.thus.3  u3.12) The spin tensor is skew symmetric. if the tensor is symmetric only 6 are independent and if the tensor is skew symmetric only 3 are independent. SMALL DEFORMATION: LINEAR ELASTICITY and the skew symmetric part is (a) 22 u = 1 2 u − ( u)T (4.1 u2. the symmetric part deﬁnes the strain = (s) u (4.2 u1.9) components ordered as (with no sym u1.1 u → u3.2 u3. hence σij = σji The gradient of the displacement has the metries)  u1.3 (4.14) . or small rotation. Accordingly.

1 Constitutive Equations for Linear Elasticity The linear theory is completed by specifying the constitutive behavior for the material. = m + 0 (4.1). we begin by considering a linear elastic material with an additional known strain. t In the balance of momentum.19) 4. however.16) (4. For this purpose the boundary. 0) = d0 (x) (4.15) Static or quasi-static problems are considered by omitting the acceleration term from the momentum equation (Eq. and v0 is the initial velocity ﬁeld. the body force was speciﬁed per unit of mass. 0 is a second part which we assume is a speciﬁed strain. Inclusion of intertial forces requires the speciﬁcation of the initial conditions u(x. For example. 0 as a thermal strain is given by 0 = th = α(T − T0 ) (4. is split into two parts: • Speciﬁed displacements on the part Γu . unit weight/volume) using ρ bm = bv (4. 4.e. 0) = v0 (x) where d0 is the initial displacement ﬁeld. We shall consider several alternatives for splits during the course.20) where m is the strain caused by stresses and is called the mechanical part.18) ˙ u(x.21) . Γ.CHAPTER 4. given as: ¯ u = u ¯ where u is a speciﬁed quantity. This may be converted to a body force per unit volume (i. and • speciﬁed tractions on the part Γt . In small deformation analysis the strain is expressed as an additive sum of parts. (4. SMALL DEFORMATION: LINEAR ELASTICITY 23 The basic equations which are independent of material constitution are completed by specifying the boundary conditions. Accordingly. given as: t = σn = ¯ t where ¯ is a speciﬁed quantity..17) (4.

27) Using tensor quantities. from which stresses are (4.26) (4. The constitutive equations relating stress to mechanical strain may be written (in matrix notation.24) and D is the matrix of elastic constants  D11 D12 D21 D22  D D32 D =  31 D41 D42  D51 D52 D61 D62 given by D13 D23 D33 D43 D53 D63 D14 D24 D34 D44 D54 D64 D15 D25 D35 D45 D55 D65  D16 D26   D36   D46   D56  D66 m (4. using this table.29) The above set of equations deﬁnes the governing equations for use in solving linear elastic boundary value problems in which the inertial forces may be ignored.28) cd The transformation from the tensor to the matrix (Voigt) form is accomplished by the index transformations shown in Table 4. which is also called Voigt notation) as σ = D m = D( − 0 ) (4.LP where T is temperature and T0 is a stress free temperature. etc.22) where the matrix of stresses is ordered as the vector σ = σ11 σ22 σ33 σ12 σ23 σ31 T (4. we have C1111 → D11 . C1233 → D43 .23) the matrix of strains is ordered as the vector (note factors of 2 are used to make shearing components the engineering strains. the constitutive equation for linear elasticity is written in indicial notation as: σab = Cabcd ( cd − 0 ) (4. (4.1 Thus. W ( computed as ∂W σab = ∂ m ab the elastic modulus matrix is symmetric and satisﬁes Dij = Dji ). We next .25) Assuming the existence of a strain energy density.CHAPTER 4. SMALL DEFORMATION: LINEAR ELASTICITY 24 . γij ) = 11 22 33 2 12 2 23 2 T 31 (4.

we assume that inertial forces may be ignored.CHAPTER 4. SMALL DEFORMATION: LINEAR ELASTICITY Tensor Matrix Index Index 1 2 3 4 5 6 ab 11 22 33 12 23 31 21 32 13 Table 4.1: Transformation of indices from tensor to matrix form 25 discuss some variational theorems which include the elasticity equations in a form amenable for ﬁnite element developments. The inclusion of inertial forces precludes the development of variational theorems in a simple form as noted in the previous chapter. we can add the inertial eﬀects and use time discrete methods to restore symmetry to the formulation. Later. For the present. .

3) . we introduce the variations to the displacement. and the Japanese scholar. hence. K. To construct the ﬁrst variation. a transpose may be introduced into each term without changing the meaning.1 Hu-Washizu Variational Theorem Instead of constructing the weak form of the equations and then deducing the existence of a variational theorem. as uη = u + η U 26 (5. Accordingly. σ. we proceed as in the previous chapter. u.1) − Γt uT ¯ dΓ − t Γu ¯ tT (u − u)dΓ = Stationary Note that the integral deﬁning the variational theorem is a scalar. σ. ) = + Ω 1 2 T Ω D dΩ − Ω (s) T D 0 dΩ σT ( u − ) dΩ − Ω uT bv d Ω (5. and the strain. as done for the thermal problem. E.Chapter 5 Variational Theorems: Linear Elasticity 5.g. Washizu [25]. The theorem may be written as I(u. For example. a variational theorem which includes all the equations for the linear theory of elasticity (without inertial forces) will be stated. Hu.. and. I = Ω aT b dΩ = Ω (aT b)T dΩ = Ω bT a dΩ (5. U. the stress.2) A variational theorem is stationary when the arguments (e. The variational theorem is a result of the work of the Chinese scholar. thus. S. is known as the Hu-Washizu variational theorem. ) satisfy the conditions where the ﬁrst variation vanishes.

The result is δI = Ω δ T D dΩ − Ω (s) δ T D 0 dΩ σT ( Ω (s) + Ω δσ T ( u − )dΩ + δuT ¯ tdΓ Γt δu − δ )dΩ − Ω δuT bv dΩ − − Γu ¯ nT δσ(u − u)dΓ − Γu tT δudΓ = 0 (5. the U. To accomplish the grouping it is necessary to integrate by parts the term involving (s) U. VARIATIONAL THEOREMS: LINEAR ELASTICITY ση = σ + η S η 27 (5. For the Hu-Washizu theorem the ﬁrst variation deﬁning the stationary condition is given by dI η dη = η=0 Ω ET D dΩ − Ω ET D 0 dΩ )dΩ + Ω + Ω ST ( (s) u − σT ( (s) U − E)dΩ − Ω UT bv dΩ − Γt UT ¯ tdΓ tT UdΓ = 0 Γu − Γu ¯ nT S(u − u)dΓ − (5. E). S.6) The ﬁrst variation is then deﬁned as the derivative of I η with respect to η and evaluated at η = 0. η ) (5.23 for each of the variables.CHAPTER 5.8) and the two forms lead to identical results.9) .4) (5. In order to show that the theorem in form 5.5) = + ηE and deﬁne the single parameter functional I η = I(uη . σT Ω (s) UdΩ = − Ω UT · σdΩ + Γt tT UdΓ + Γu tT UdΓ (5. σ η .7) The ﬁrst variation may also be constucted using 3.7 is equivalent to the equations for linear elasticity. we need to group all the terms together which multiply each variation function (e..g. Accordingly.

the Hellinger-Reissner principle and the principle of minimum potential energy are presented below since they are also often used in constructing ﬁnite element formulations in linear elasticity. This results in the integral of the square of a function. VARIATIONAL THEOREMS: LINEAR ELASTICITY Grouping all the terms we obtain dI η dη = η=0 Ω 28 ET [D( ST ( Ω (s) − u − 0 ) − σ]dΩ )dΩ − Ω + + Γt UT ( · σ + bv )dΩ (5. and the displacement boundary condition.11) The need to develop an expression for strains in terms of stresses limits the application of the Hellinger-Reissner principle. u. since the variation is arbitrary. There are other variational principles which can be deduced directly from the principle. The expression which multiplies each variation function is called an Euler equation of the variational theorem. the strain-displacement equation. and the stress. the balance of linear momentum. the only equation not contained is the balance of angular momentum. The strains are eliminated by developing an expression in terms of the stresses. we can assume that it is equal to the value of the non-vanishing expression. For the Hu-Washizu theorem. Two of these. which must then be positive. consequently. and hence the integral will not be zero. then. 5.CHAPTER 5. Indeed. only the displacement. and thus the only possibility is that the assumption of a non-vanishing expression is false. in ﬁnite deformation elasticity the development of a relation similar to 5.11 is not possible in general. For linear elasticity this leads to = 0 + D−1 σ (5.2 Hellinger-Reissner Variational Theorem The Hellinger-Reissner principle eliminates the strain as a primary dependent variable. For example.10) UT (t − ¯ t)dΓ − Γu ¯ nT S(u − u)dΓ = 0 The fundamental lemma of the calculus of variations states that each expression multiplying an arbitrary function in each integral type must vanish at each point in the domain of the integral. On the other . the variations multiply the constitutive equation. The Hu-Washizu variational principle will serve as the basis for most of what we need in the course. σ. Suppose that an expression does not vanish at a point. the traction boundary condition. remain as arguments in the functional for which variations are constructed. The lemma is easy to prove. This leads to a contradiction.

13) together with 4.11 into the Hu-Washizu principle leads to the result I(u. respectively. σ. and the strain. σ) = − − Ω 1 2 0T Ω 0 D 0 dΩ − σT Ω T 1 2 σ T D−1 σdΩ Ω (s) σT dΩ + udΩ − Ω uT bv dΩ (5.. 4. thin plates as a limit case of the thick Mindlin-Reissner theory). the constitutive equation must be given.13. as arguments of the functional. Introducing 5. The strain-displacement equations are deduced by either directly stating 4. Thus. the Hellinger-Reissner principle is an important limiting case when considering problems with constraints (e.3 Minimum Potential Energy Theorem The principle of minimum potential energy eliminates both the stress.17) . VARIATIONAL THEOREMS: LINEAR ELASTICITY 29 hand. in addition to 5. the displacement boundary conditions are assumed to be imposed as a constraint on the principle.12) − Γt u tdΓ − Γu T¯ ¯ t (u − u)dΓ The Euler equations for this principle are (s) u = 0 + D−1 σ (5.6 or comparing 5.15 the relation σ = D( − 0 ) (5.g.12 may be omitted since its ﬁrst variation is zero. In addition.15 and 4. Thus.15) are satisﬁed at each point of Ω and Γ.1. linear elastic incompressible problems. we shall on occasion use the principle in our studies.14) and ¯ u = u (5. The MPE theorem may be deduced by assuming = (s) u (5.16.16) uT bv dΩ − Γt uT ¯ tdΓ Since stress does not appear explicitly in the theorem. Accordingly. the variational theorem is given by the integral functional I(u) = − Ω 1 2 ( Ω (s) u)T D( (s) u)dΩ − Ω ( (s) u)T D 0 dΩ (5.11 to 5.14 and 5. .CHAPTER 5. 5. The ﬁrst term in 5.

30 The principle of minimum potential energy is often used as the basis for developing a displacement ﬁnite element method.CHAPTER 5. VARIATIONAL THEOREMS: LINEAR ELASTICITY is given. .

2) In the above Mel is the total number of elements in the ﬁnite element mesh.Chapter 6 Displacement Finite Element Methods A variational equation or theorem may be solved using the direct method of the calculus of variations. A similar construction is performed for the boundaries. t) = α=1 Nα (ξ) uα (t) = Nα (ξ) uα (t) (6. The ﬁnite element approximation for displacements in an element is introduced as Nel u(ξ.3) where Nα is the shape function at node α. This reduces a steady state problem to an algebraic process and a transient problem to a set of ordinary diﬀerential equations. ξ are natural coordinates for the element. As indicated in Chapter 2 the region is divided as Mel Ω ≈ Ωh = e=1 Ωe (6. Using the isoparametric concept 31 . In the ﬁnite element method we divide the region into elements and perform the approximations on each element.1) and integrals are deﬁned as Mel ( · ) dΩ ≈ Ω Ωh ( · ) dΩ = e=1 Ωe ( · ) dΩ (6. uα are the values of the displacement vector at node α and repeated indices imply summation over the range of the index. In the direct method of the calculus of variations the dependent variables are expressed as a set of trial functions multiplying parameters. With this construction the parts of the variational equation or theorem are evaluated element by element.

we express ¯ ¯ u = Nα (ξ) uα (t) (6. The remaining terms involving speciﬁed applied loads are due to the body forces.5) and set ¯ ¯ uα (t) = u(xα . This step is not necessary but is common in most applications. of the term Πσ = Ωe ( (s) u)T σ dΩ = e Ωe ( (s) u)T σ dΩ (6. DISPLACEMENT METHODS 32 x(ξ) = Nα (ξ) xα (6. 6. In the next sections we consider the computation of the external force (from applied loads) and the internal force (from stresses) by the ﬁnite element process.7 yields Πf = (uα )T Ωe Nα bv dΩ + Γte Nα ¯ dΓ t = (uα )T Fα (6.4) where xα are the cartesian coordinates of nodes.3 in Eq. u.6) We then will assume the integral over Γu is satisﬁed and may be omitted. and the applied surface tractions.CHAPTER 6. ¯ The terms in the variational principal are t.9) 6. Πf = Ωe uT bv dΩ + Γte uT ¯ dΓ t (6. 6.10) .2 Internal Force Computation The stress divergence term in the Hu-Washizu variational principle is generated from the variation with respect to the displacements.8) where Fα denotes the applied nodal force vector at node α and is computed from Fα = Ωe Nα bv dΩ + Γte Nα ¯ dΓ t (6. bv . 6.7) Using Eq. t) (6.1 External Force Computation ¯ In our study we will normally satisfy the displacement boundary conditions u = u by ¯ setting nodal values of the displacement to the values of u evaluated at nodes. the displacement at each point in an element may be computed. That is.

18) Finally.13) the strain-displacement matrix is expressed  Nα.2 + u2.1 + u1.16) and are expressed in terms of the displacement derivatives as T = u1.2  0 Nα.1  0   0 Bα =  Nα.2 u1 /x1 (u1.11) where Bα is the strain displacement matrix for the element.15) = 11 22 33 2 12 (6.1 ) (u2.20) .2 0 (u1.1 u2. for a 2-dimensional axisymmetric problem (with no torsional loading) the strains are T = (6.1 (6.1 u2. DISPLACEMENT METHODS 33 Using the ﬁnite element approximation for displacement.14) ∂Nα ∂xi For a 2-dimensional plane strain problem the non-zero strains reduce to Nα.1 0   Nα.3 Nα.1 (6.3 where (6.19) 11 22 33 2 12 and are expressed in terms of the displacements as T = u1.1 )   Nα.3 ) as:  0 0 Nα.12) and related to the displacement derivatives by T = u1.3   Nα. If the components of the strain for 3-dimensional problems are ordered as T = 11 22 33 2 12 2 23 2 31 (6.2 + u2.1 u2.CHAPTER 6.1 0  0 Nα.1 ) (6.17) thus.3 + u3.3 (u1.2   0 Nα.2 Nα.2 0   0 Nα.2 + u2. the symmetric part of the strains deﬁned by the symmetric part of the deformation gradient in each element is given by (s) u = (u) = Bα uα (6.2   =   0 0  Nα.2 u3.i = T (6.2 ) (u3. Bα becomes: Bα (6.

Locking is generally associated with poor performance at or near the incompressible limit.2   Bα =  Nα /x1 0  Nα.26) The deviatoric part of stress .23) which gives Πσe = (uα )T Pα (σ) (6. is deﬁned so that its trace is zero.x2 plane. becomes:   Nα.21) (Bα )T σ dΩ (6. To study the locking problem we split the formulation into deviatoric and volumetric terms.1 0  0 Nα. uα . For stress the spherical part is the mean stress deﬁned by 1 1 (6.24) The stress divergence term is a basic ﬁnite element quantity and must produce a response which is free of spurious modes or locking tendencies. x2 now denote the axisymmetric coordinates r.25) p = tr(σ) = σkk 3 3 For inﬁnitesimal strains the spherical part is the volume change deﬁned by θ = tr( ) = kk (6. Bα . the internal stress divergence force.2 Nα. z. 6.22) In the sequel we deﬁne the variation of this term with respect to the nodal displacements. The stress may be written in terms of the deviatoric and pressure parts (pressure is spherical part) as σ = s + p1 (6. DISPLACEMENT METHODS The strain-displacement matrix for axisymmetry. s.27) 1 For axisymmetry it is also necessary to replace the volume element by dΩ → x1 dx1 dx2 and the surface element by dΓ → x1 dS where dS is an boundary diﬀerential in the x1 . respectively1 The stress divergence term for each element may be written as Πσe = (uα )T Ωe 34 (6.1 where x1 .CHAPTER 6. .3 Split into Deviatoric and Spherical Parts For problems in mechanics it is common to split the stress and strain tensors into their deviatoric and spherical parts. This force is expressed by Pα (σ) = Ωe (Bα )T σ dΩ (6.

35) (6. the deviatoric and volumetric parts are given by = e + 1 θ1 3 (6. we deﬁne Idev = I − 1 m mT 3 (6. For subsequent developments.28) In matrix form the pressure is given by p = 1 T m σ 3 (6.31) and hence mT s = 0 as required.36) (6.39) (6. in three dimensions.CHAPTER 6. Accordingly.32) as the deviatoric projector. Utilizing the above properties. I is a 6 × 6 identity matrix. 1 is the rank two identity tensor. which in matrix notation is given by the vector mT = 1 1 1 0 0 0 (6. that inverses to the projectors do not exist. We note. the volumetric projector is deﬁned by Ivol = 1 m mT 3 (6. We note that the trace of the stress gives mT σ = 3 p = mT s + p mT m = mT s + 3 p (6.29) thus.38) In the above m is any positive integer power. however. DISPLACEMENT METHODS 35 where.37) . the deviatoric part of stresses now may be computed as s = σ − 1 m mT σ = 3 I − 1 m mT 3 σ (6.33) (6.34) These operators have the following properties I = Idev + Ivol Idev = Idev Idev = (Idev )m Ivol = Ivol Ivol = (Ivol )m and Ivol Idev = Idev Ivol = 0 (6.30) where. Similarly. we can operate on the strain to deﬁne its deviatoric and volumetric parts.

1    0 3 Nα.1 and for the 2-dimensional plane problem  Bdev 2 Nα.2 −Nα.3 3 Nα.2   −Nα.1 −Nα.2  −Nα. DISPLACEMENT METHODS 36 where e is the strain deviator and θ is the change in volume.1 + Nα /x1 Nα.45) 1 mb 3 (6. mT e = 0 (6.46) mT Bdev = 0 (6.2 Nα.1 3 Nα.3 −Nα.41) The strain-displacement matrix also may now be written as a deviatoric and volumetric form.39 and yields for the 3-dimensional problem   2 Nα.CHAPTER 6.43) (6.2 (6. For 3-dimensional problems bα = Nα.1 (6.42) is the volumetric strain-displacement matrix for a node α in its basic form.1 1 −Nα.2 2 Nα.48) The deviatoric matrix Bdev is constructed from Eq.3    (6.3 0 3 Nα.2 −Nα.2  3 Nα.2 3 Nα.1 Nα. In 2dimensional plane problems the volumetric strain-displacement matrix is given by bα = Nα.3    1 −Nα.2 (6. Using matrix notation we have θ = mT (6.2 2 Nα.2  3 Nα.49) Bdev = 0  3  3 Nα.50) . Accordingly.1 2 Nα. 6.1 −Nα.1 Nα. we use the strain split (u) = Bα uα = (Bdev )α uα + (Bvol )α uα where Bdev = Idev B and Bvol = Ivol B = where b = mT B .40) we obtain e = Idev .44) (6.47) and for 2-dimensional axisymmetric problems bα = Nα.1  = 3 −Nα.3 (6.

53) If we introduce 1 mb (6.5 Constitutive Equations for Isotropic Linear Elasticity The constitutive equation for isotropic linear elasticity may be expressed as σ = λ 1 tr( ) + 2 µ (6.56) Thus.2 1 −(Nα. the internal force is composed of the sum of deviatoric and volumetric parts. Pα = Ωe BT σ dΩ = α Ωe BT (s + p m) dΩ α (6.CHAPTER 6.1 ) −Nα. DISPLACEMENT METHODS 37 Finally.52) which after rearrangement gives Pα = Ωe BT s dΩ + α Ωe BT m p dΩ α (6. the deviatoric matrix for the 2-dimensional axisymmetric problem is given by:   (2 Nα.4 Internal Force .1 + Nα /x1 ) 2 Nα.57) .2  3 Nα. then B = Bdev + Bvol = Bdev + Pα = Ωe (BT )α s dΩ + dev Ωe bT p dΩ α (6.2 3 Nα.55) Since the volumetric term has no eﬀect on the deviatoric stresses the residual may also be computed from the simpler form in terms of Bα alone as Pα = Ωe BT s dΩ + α Ωe bT p dΩ α (6.51) Bdev = 3  (2 Nα /x1 + Nα.2    (6. 6.1 − Nα /x1 ) −Nα. Accordingly.54) 3 and use the properties deﬁned above for products of the deviatoric and volumetric terms.Deviatoric and Volumetric Parts The above split of terms is useful in writing the internal force calculations in terms of deviatoric and volumetric parts.1 6.

58) For diﬀerent values of ν.61) 1 We note that K also tends to inﬁnity as ν approaches 2 . that is cijkl = cklij = cjikl = cijlk (6. µ = E 2 (1 + ν) (6.62) (6. DISPLACEMENT METHODS 38 where λ and µ are the Lam´ parameters which are related to Young’s modulus. ≥ µ ≥ (6.66) where the elastic moduli are split into D = λ Dλ + µ Dµ (6. Another parameter which is related to λ and µ is the bulk modulus. which is deﬁned by 0 ≤ ν ≤ K = λ + E 2 µ = 3 3 (1 − 2 ν) (6.65) The relations may be transformed to matrix (Voigt) notation following Table 4.CHAPTER 6. and λ is a parameter which causes diﬃculties since it is inﬁnite. 0 ≤ λ ≤ ∞ (6. The constitutive equation for an isotropic material is given in indicial form by σij = λ δij and for a general linear elastic material by σij = cijkl kl kk + 2µ ij (6.1 and expressed as σ = D (6. For an isotropic material the elastic moduli are then related by cijkl = λ δij δkl + µ (δik δjl + δil δjk ) (6. the Lam´ parameters have the following ranges e 0 ≤ ν ≤ and 1 2 . and e Poisson’s ratio.64) We note that the above deﬁnition for the moduli satisﬁes all the necessary symmetry conditions.60) 2 2 3 1 For an incompressible material ν is 2 .63) where cijkl are the elastic moduli. K.59) 1 E E . by λ = νE (1 + ν)(1 − 2 ν) .67) . ν. E.

γij = 2 ij ). DISPLACEMENT METHODS with  1 1  1 =  0  0 0  2 0  0 =  0  0 0 1 1 1 0 0 0 0 2 0 0 0 0 1 1 1 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0  0 0  0  = m mT = 3 Ivol 0  0 0  0 0  0  0  0 1 39 Dλ (6.72) Dµ Idev = Idev Dµ (6.e. thus.CHAPTER 6. 2 (6.70) (6.69) used as non-dimensional matrices to split the moduli..73) Once Ddev has been computed it may be noted that Idev Ddev = Ddev Idev = Ddev Ivol Ddev = Ddev Ivol = 0 and. it is a deviatoric quantity.75) . Note that in Dµ the terms multiplying shears have unit values since engineering shear strains are used (i.68) Dµ (6. the computation of the element stiﬀness matrix for a displacement approach is given and is based upon the above representations for the moduli.74) (6.2 If the moduli matrices are premultiplied by Ivol and Idev the following results are obtained Ivol Dλ = Dλ Idev Dλ = 0 2 2 m m T = Dλ Ivol Dµ = 3 3 and  4 −2 −2 0 0 0 −2 4 −2 0 0 0   1 −2 −2 4 0 0 0   = Ddev = 0 0 3 0 0 3 0   0 0 0 0 3 0 0 0 0 0 0 3  (6.71) (6. In the following section.

2 that Pα = Ωe (6.83) (6.80) 2 µ) Dλ = K Dλ 3 Thus.78) (6.76) p = Kθ Noting that the volumetric strain may be computed from θ = bβ uβ the pressure for the displacement model may be computed from p = K bβ uβ We recall from Section 6.82) (6.85) Using the above deﬁnitions and identities the internal force vector may be written as Pα = Ωe µ (BT )α Dµ (Bdev )β dΩ uβ + dev Ωe K bα bT dΩ uβ β (6. the pressure constitutive equation is p m = (λ + and = K m(mT ) = K m θ (6.81) = Idev (λ Dλ + µ Dµ ) (6.86) .84) (BT )α s dΩ + dev Ωe bT p dΩ α (6.CHAPTER 6.77) (6.79) If we use the properties of the moduli multiplied by the projectors.6 Stiﬀness for Displacement Formulation The displacement formulation is accomplished for a linear elastic material by noting that the constitutive equation is given by (for simplicity 0 is assumed to be zero) σ = D The strains for a displacement approach are given by = Bβ uβ where uβ are the displacements at node β. DISPLACEMENT METHODS 40 6. the above equations reduce to s = µ Ddev = µDµ e = µDµ (Bdev )β uβ (6. Constructing the deviatoric and volumetric parts may be accomplished by writing s = Idev σ = Idev D and p m = Ivol D = Ivol (λ Dλ + µ Dµ ) (6.

e. A typical integral is evaluated by ﬁrst transforming the integral onto a natural coordinate space f (x) dΩ = Ωe 2 f (x(ξ)) j(ξ) dξ (6. thus. DISPLACEMENT METHODS 41 and.89) 6. The use of the same quadrature for each part of the stress divergence terms given above (in P and K) leads to a conventional displacement approach for numerically integrated ﬁnite element developments. and j(ξ) is the determinant of the jacobian transformation J(ξ) = Thus j(ξ) = det J(ξ) The integrals over 2 are approximated using a quadrature formula. Alternatively. The next lowest order of quadrature is called a reduced quadrature..7 Numerical Integration Generally the computation of integrals for the ﬁnite element arrays is performed using numerical integration (i. The minimum order quadrature which produces a stiﬀness with the correct rank (i.. quadrature). number of element degree-of-freedoms less the number of rigid body modes) will be called a standard or full quadrature (or integration).92) f (x(ξ)) j(ξ) dξ ≈ l=1 f (x(ξ l )) j(ξ l ) wl (6. for isotropic linear elasticity.CHAPTER 6. thus L 2 ∂x ∂ξ (6.e. the stiﬀness matrix may be deduced as the sum of the deviatoric and volumetric parts Kαβ = (Kdev )αβ + (Kvol )αβ where (Kdev )αβ and (Kvol )αβ = Ωe (6.91) (6. dξ denotes dξ1 dξ2 in 2-dimensions.88) K bα bT dΩ = β Ωe K BT Dλ Bβ dΩ α (6. use of standard quadrature on one term and reduced quadrature on another leads to a method called selective reduced quadrature.90) where 2 denotes integration over the natural coordinates ξ.93) .87) = Ωe µ (BT )α Dµ (Bdev )β dΩ = dev Ωe µ BT Ddev Bβ dΩ α (6.

and computation of σ l requires an evaluation of the constitutive equation at the quadrature point. Additional steps are involved in computing the entries in each array.95) Using quadrature. the determination of Bα requires computation of the derivatives of the shape functions. e. (e. Repeat step 2 until all quadrature points in element are considered. The use of quadrature reduces the development of ﬁnite element arrays to an algebraic process involving matrix operations. For brick elements in three dimensions and quadrilateral elements in two dimensions.. l • Compute j(ξ l ) wl = c • Compute the matrix in the integrand. 1 1 g(ξ) dξ = 2 −1 −1 g(ξ) dξ1 dξ2 (6. For example.i . DISPLACEMENT METHODS 42 where ξ l and wl are quadrature points and quadrature weights.. a shape function routine for 2 dimensions is called shp2d and is accessed by the call (6. Initialize the array Pα 2. Loop over the quadrature points. respectively.97) Similar expressions may be deduced for each of the terms deﬁned by the deviatoric/volumetric splits. In FEAP. • Accumulate the array. for 2-dimensions. the basic algorithm to compute the stress divergence term is given by: 1. For example.96) and the stiﬀness matrix is computed by quadrature as L Kαβ = l=1 Bα (ξ l )T D (ξ l )Bβ (ξ l ) j(ξ l ) wl (6.94) and for 3-dimensions 1 1 −1 1 g(ξ) dξ = 2 −1 −1 g(ξ) dξ1 dξ2 dξ3 (6. Thus.g.CHAPTER 6. the stress divergence is given by L Pα = l=1 Bα (ξ l )T σ(ξ l ) j(ξ l ) wl (6. Nα. the integration is generally carried out as a product of one-dimensional Gaussian quadrature.g. Pα ← Pα + Aα c 3.98) . Bα (ξ l )T σ l = Aα ). The evaluation of the shape functions is performed using a shape function subprogram.

swg -array of natural coordinates and weights (output). swg ) where -number of quadrature points in each direction (input). nel. flag) where xi natural coordinate values (ξ1 .CHAPTER 6. l The array of points and weights has the following meanings: . 43 The array of shape functions has the following meanings: shp(1.1 shp(2.A) is NA.nen)) (input) shp array of shape functions and derivatives (shp(3.nen)) (output) xsj jacobian determinant at quadrature point (output) ndm spatial dimension of problems (input) nel number of nodes on element (between 3 and 9) (input) ix array of global node numbers on element (ix(nen)) (input) flag ﬂag. DISPLACEMENT METHODS call shp2d( xi.A) is NA. ξ2 ) at quadrature point (input) xl array of nodal coordinates for element (xl(ndm.A) is NA. shp. xl. if false derivatives returned with respect to x (input).2 shp(3. if true derivatives returned with respect to ξ.3 The quadrature points may be obtained by a call to int2d: call int2d( l. ndm. lint. lint -total number of quadrature points (output). ix. xsj.

L swg(3. . DISPLACEMENT METHODS swg(1. An example.L) is ξ2.L swg(2.L) is ξ1. is element elmt01 which is given in Appendix B.CHAPTER 6.L) is wL 44 Using the above two utility subprograms a 2-dimensional formulation for displacement (or mixed) ﬁnite element method can be easily developed for FEAP.

2) (7. σ.3) (7.Chapter 7 Mixed Finite Element Methods 7. while the stresses and strains appear without any derivatives. Appropriate interpolations for each element are thus u(ξ) = NI (ξ) uI (t) σ(ξ) = φα (ξ) σ α (t) and (ξ) = ψα (ξ) α (7. which we recall may be written as Π(u.4) (t) 45 . In the work considered here we use the Hu-Washizu variational principle. the continuity conditions we may use in ﬁnite element approximations are C 0 for the displacements and C −1 for the stresses and strains (a C −1 function is one whose ﬁrst integral will be continuous). ) = + Ω 1 2 T Ω D dΩ − Ω (s) T D 0 dΩ uT bv dΩ σT ( T u − ) dΩ − Ω T − Γt u ¯ dΓ − t Γu ¯ t (u − u) dΓ = Stationary (7.1) In the principle. displacements appear up to ﬁrst derivatives. Accordingly.1 Solutions using the Hu-Washizu Variational Theorem A ﬁnite element formulation which is free from locking at the incompressible or nearly incompressible limit may be developed from a mixed variational approach.

Balance of Momentum · (s + 1 p) + bv = 0 which is also written as div(s + 1 p) + bv = 0 2.13) Kθ − p = 0 (7. however. ) = + Ω 1 2 µ Ω T Ddev dΩ − Ω µ T Ddev 0 dΩ (7. φα and ψα need only be piecewise continuous in each element. Constructing the variation for the above split leads to the following Euler equations which hold in the domain Ω: 1.9) (7. may have no direct physical meaning. MIXED FINITE ELEMENT METHODS 46 where φα (ξ) and ψα (ξ) are interpolations which are continuous in each element but may be discontinuous across element boundaries.6) (7. and the traction and displacement boundary integrals and consider an isotropic linear elastic material. the remaining terms may be split into deviatoric and volumetric parts as Π(u. Constitutive equations µ Ddev − s = 0 (7.11) (7. σ. respectively.5) sT [e(u) − e] dΩ 1 2 K θ2 dΩ − Ω Ω + where K θ θ0 dΩ + Ω p[θ(u) − θ] dΩ e(u) = Idev and θ(u) = tr( (s) (s) u ·u (7.CHAPTER 7. thus.7) u) = are the strain-displacement relations for the deviatoric and volumetric parts. .10) (7. Strain-Displacement equations e(u) − e = 0 θ(u) − θ = 0 3.1 The parameters σ α and α are not necessarily nodal values and. we ignore the integral for the body force.12) (7. this makes the evaluation of integrals over each element more diﬃcult and to date is rarely used.8) 1 Strictly. If. for the present.

) (7. the Hu-Washizu variational theorem may be approximated by summing the integrals over each element.20) Thus. ) ≈ Πh (u. 47 Using the interpolations described above. this part of the problem is given as a displacement model. σ. θ) (7. The variational expression Eq. the ﬁnite element approximation for the mixed formulation may be written as Πe (u. θ) = e Πe (u. MIXED FINITE ELEMENT METHODS In addition the boundary conditions for Γu and Γt are obtained. σ. Recall that the strain-displacement equations for a ﬁnite element approximation are given by (u) = BI uI (7.16) p [θ(u) − θ] dΩ which may be split into integrals over the elements as Π(u. Accordingly. p.5 becomes Π(u. 7. σ. θ) ≈ Πh (u. 7. For other situations.17) A mixed approximation may now be used to describe the pressure and the volume change in each element.19) where it is noted that the same approximating functions are used for both p and theta. p.13 at each point of the domain of an element. Accordingly.15) for each point of Ω. θ) = (uI )T + θT + pT Ωe 1 2 1 2 Ωe µ BT Ddev BJ dΩ uJ − I Ωe Ωe µ BT Ddev I 0 dΩ K φT φ dΩ θ − Ωe K φT θ0 dΩ φT φ dΩ θ Ωe φT bJ dΩ uJ − (7.CHAPTER 7. p. p. the constitutive equation may be approximately satisﬁed.21) .14) If the deviatoric part is approximated by taking e = e(u) (7. θ) = 1 2 1 + 2 + Ω µ Ω T (u) Ddev (u) dΩ − Ω µ T (u) Ddev 0 dΩ K θ2 dΩ − Ω Ω K θ θ0 dΩ (7. p.18) (7. Π(u. Eq. we assume p(ξ) = φα (ξ) pα (t) = φ(ξ) p θ(ξ) = φα (ξ) θα (t) = φ(ξ) θ (7. ) = e Πe (u. If the material is isotropic linear elastic. the use of the same functions will permit an exact satisfaction of the constitutive equation.

27) where e0 are the deviatoric initial strains. θ) = (uI )T + θT + pT 1 (Kdev )IJ uJ − (P0 )I dev 2 1 k θ − π0 2 g J uJ − h θ (7. 7. MIXED FINITE ELEMENT METHODS If we deﬁne the following matrices: k = Ωe 48 K φT φ dΩ K φT θ0 dΩ Ωe (7. p.26) and denote the eﬀects of initial deviatoric strains as (P0 )I = dev Ωe µ BT Ddev I 0 dΩ = Ωe µ BT Dµ e0 dΩ I (7.28) If we denote the variations of pressure and volume change as pη = p + η Π θη = θ + η Θ the ﬁrst variation of Eq.28 may be written in the matrix form    J (Kdev )IJ gJ 0 u dΠe T T  I T T  p  gI 0 −h = (U ) .23) (7.31) .24) (7. Θ dη 0 −h k θ  0  (Pdev )I  0  − π0 (7. Π .25) φT φ dΩ φT bI dΩ Ωe gI = and recall that the deviatoric stiﬀness is deﬁned as (Kdev )IJ = Ωe BT Ddev BJ dΩ I (7.CHAPTER 7.29) (7. The mixed variational terms become Πe (u.30) (7.22) π0 = h = Ωe (7.

by deﬁning a modiﬁed volumetric strain-displacement matrix as ¯ bI = h−1 gJ the above simpliﬁes to dΠe = (UI )T dη 2 (7.38) This is a mixed patch test requirement. .36) (7. MIXED FINITE ELEMENT METHODS or in variational notation as δΠe    J (Kdev )IJ gJ 0 u T  I T T T  p gI 0 −h = (δu ) .37) ¯ ¯ (Kdev )IJ + bT k bJ I ¯ uJ − (P0 )I − bT π 0 dev I (7.35) (7.33) where nθ and np are the number of parameters associated with the volume change and pressure approximations.CHAPTER 7. since we used the same functions for the two approximations. the last two rows of the above matrix expression must vanish and may be solved at the element level. δθ 0 −h k θ  0  (Pdev )I  0  − π0 49 (7. we may perform the element solutions by inverting only h. respectively.32 is θ = h−1 gJ uJ and the solution to the third row is p = h−1 (k θ − π 0 ) Substitution of the above results into the ﬁrst equation gives dΠe T = (UI )T (Kdev )IJ + gI h−1 k h−1 gJ dη T − (P0 )I − gI h−1 π 0 dev uJ (7. 7. We have satisﬁed this requirement by taking an equal number for the two approximations.34) Finally. the matrix h is square and positive deﬁnite (provided our approximating functions are linearly independent). Also. δp . consequently. See [26. from the stationarity condition. The requirement for a solution to exist is that2 nθ ≥ np (7.32) We note that the parameters p and θ (and their variations Π and Θ) are associated with a single element. consequently. Chapter 12]. The solution to Eq.

41) where the deviatoric stress is expressed by the displacement approximation as s = µ Ddev (BJ uJ − 0 ) (7. and the number is small compared to the number of degrees-of-freedom on the element. The requirement for the approximation is guided by the principle that: (1) we use the minimum number of functions which make K have correct rank for a single element.2 Finite Element Solution for Mixed Formulation The mixed ﬁnite element solution for the linear elastic problem requires selecting a set of approximating functions for φ.46) . the rank will normally be that of k.43) 7. 4-node quadrilateral or 8-node brick elements can use a single function φ1 = 1 (7.42) and the pressure is expressed by the mixed approximation as p = φ(ξ) h−1 (k θ − π 0 ) (7. use of φ1 = 1 .CHAPTER 7. The modiﬁed volumetric stiﬀness has a rank which is given by ¯ rank(Kvol ) = min ( rank(b). rank(k) ) (7. MIXED FINITE ELEMENT METHODS The volumetric stiﬀness for the mixed formulation is given as ¯ ¯ (Kvol )IJ = bT k bJ I and the volumetric initial force by ¯ (P0 )I = bT π 0 vol I The stress divergence term for the mixed model formulation is computed from PI = Ωe 50 (7.44) Provided the approximations for φ are linearly independent.39) (7. For example. and (2) the functions produce an element which is invariant with respect to the input data.40) BT (s + p m) dΩ I (7. This gives a rank of 1 for the volumetric stiﬀness. For example.45) for the approximating space. if we show that two functions are suﬃcient for a 2-dimensional element. The number of φ functions will aﬀect the rank of the volumetric terms. φ2 = ξ1 (7.

FEAP will initialize K and the element residual. gI and h. Invert h and compute bI 5. l) NI . NI (ξ l ) = shp(3. another is to use 3 functions with φ1 = 1 .49) (7. Loop over quadrature points. I.52) = [Idev BI (ξ l ) + 1 ¯ m φ(ξ l ) bI ] uI 3 (7.47) The actual functions selected must be subjected to further evaluations to decide which best meets the objectives of the problem solution. l (a) Compute strain-displacement matrix. I. l (7. B. one being φ1 = 1 .50) (7. Loop over quadrature points. MIXED FINITE ELEMENT METHODS 51 would not be good since the element is not invariant with respect to a permutation in the deﬁnition of ξ1 and ξ2 . An algorithm to implement the above mixed model for linear elasticity where D is constant in each element may be summarized as: 1. Several alternatives are possible. k.48) φ2 = ξ1 + ξ2 (7. φ3 = ξ2 (7. i(ξ l ) = shp(i. φ2 = ξ1 . Loop over quadrature points. π 0 . ¯ 4.53) (b) Compute quadrature stresses and π 0 σl = D ( l − 0 ) (7. l (a) Compute the volumetric strain matrices.51) ¯ bI = h−1 gI (7. 2. l (a) Compute shape functions: In 2-d problems FEAP uses. and strains.54) .CHAPTER 7. h. l) (b) Compute the volume element times the quadrature weight jl wl = dv(l) 3. Initialize arrays: gI .

55) (d) Compute the deviatoric tangent. In this section we construct the form of the functional for an anisotropic linear elastic material.CHAPTER 7.60) 1 m (θ − θ0 )] 3 1 m (θ − θ0 )] 3 where ¯ = Idev D [Idev ( (u) − s and p = ¯ 1 T m D [Idev ( (u) − 3 0 ) + 0 (7. are computed directly from the displacements but the volumetric strain. Accordingly. K ¯ ¯ K = Kdev + bT k b (7.3 Mixed Solutions for Anisotropic Linear Elastic Materials A more general form of the Hu-Washizu principle is needed to consider either anisotropic linear elastic materials or inelastic materials in which there is coupling between volumetric and deviatoric eﬀects.59) .57) where D is a symmetric matrix in which there may be coupling between the deviatoric and volumetric strain eﬀects. It is now assumed that a ﬁnite element solution will be constructed in which deviatoric strains. ¯ = Idev (u) + A stress may be computed from ¯ as ¯ σ = D [Idev ( (u) − where θ0 = mT 0 0 1 mθ 3 1 m (θ − θ0 )] 3 (7.56) 7. e. k 6. Accordingly. Kdev (e) Compute the volumetric local tangent. we have σ = D[ − 0 ] (7. θ. Compute the tangent.61) (7. MIXED FINITE ELEMENT METHODS (c) Compute the residual lint 52 RI = F I − l=1 BT (ξ l ) σ l j(ξ l ) Wl I (7.58) ) + (7. is computed from a mixed form.62) ) + . The stress may be split into deviatoric and pressure parts as ¯ σ = ¯ + mp s ¯ (7.

adding the terms associated with the mixed volumetric pressure and volume change Vainberg’s theorem may be used to obtain a variational theorem.16 which was deduced for isotropic materials. MIXED FINITE ELEMENT METHODS If we deﬁne Ddev = Idev D Idev 1 d = Idev D m 3 1 dvol = mT D m 9 0 s = − Ddev 0 − d θ0 and p 0 = − dT then the stress may be written as ¯ σ = Ddev (u) + d θ + 1 m (dT (u) + dvol θ) + s0 + m p0 3 0 53 (7.68) This form of the stress may be multiplied by the virtual ¯ and integrated over the domain to obtain part of the variational equation associated with the strain energy. p.66) (7.1. If we introduce ﬁnite element interpolations using standard displacement interpolation together with the pressure and volume interpolations given by 7. For isotropy d is zero.69 for a single element is δΠe = + + ˆ ˆI δ uT δ θ BT s0 I T 0 φ p ˆ ˆI δ uT δ θ T Ωe BT Ddev BJ BT d φ I I φT dT BJ φT dvol φ φT bJ −φ dΩ Ωe ˆ uJ ˆ θ ˆ uJ ˆ θ (7. the ﬁrst variation of 7. Alternatively.19. The added terms in 7.64) (7.63) (7.18 and 7.70) ˆ dΩ + δ pT T Ωe −φ φ bT φ I T ˆ dΩ p + δIext . the stress and strain splits may be substituted into 7.CHAPTER 7.69) This form of the variational principle is equivalent to 7. Subsequently. θ) = + Ω 1 2 (u) θ Ω Ddev d dT dvol (u) dΩ θ ( (u) s0 + θ p0 ) dΩ p [θ(u) − θ] dΩ + Πext Ω + (7.65) (7.69 are all associated with d which deﬁnes a coupling between deviatoric and volumetric strains. The result is Π(u.67) − dvol θ0 (7.

the equation ˆ multiplying δ θ yields the equation φT p dΩ = ¯ Ωe Ωe ˆ ˆ φT φ dΩ p = h p (7.74) Using these results.CHAPTER 7. is the stress which is computed from the constitutive equation for each material. respectively. plasticity. and the mixed volumetric ¯ strain displacement equation. however. the ﬁrst integral in the variational equation deﬁnes the stress divergence terms ˆI δΠσ = δ uT Ω ¯ BT ¯ dΩ + bT I s I Ω φT p dΩ ¯ (7.24 and 7. in general.61 and 7.77) ¯ ¯ and.62 resulting in δΠe = ˆI δ uT ˆT δθ Ωe BT ¯ Is dΩ φT p ¯ ˆ uJ ˆ θ (7.70 may be expressed in terms of stresses by substituting the interpolations into 7. yields ˆI δΠσ = δ uT Ω BT [¯ + m p] dΩ I s (7. etc. Thus. p.) the eﬀective material moduli are the ones computed by linearizing the constitutive equation expressed in terms of the . MIXED FINITE ELEMENT METHODS 54 The variational equation 7. similarly. ˆ the multiple of δ p yields ˆ φT bJ dΩ uJ = Ωe Ωe ˆ ˆ φT φ dΩ θ = h θ (7.72) which yields ˆ ¯ ˆ ˆ θ = bI uI = h−1 gI uI (7. viscoelasticity.75) which upon use of the deﬁnitions for the mixed pressure. is not equal to σ.25.73) where h and gI are as deﬁned in 7.g.. Accordingly. bI . when we later consider other material models (e.76) The stress of the mixed method is deﬁned as σ = ¯ + mp s (7.71) ˆ + δ pT Ωe φT bJ −φ dΩ ˆT δθ Ωe + ˆI δ uT bT φ I ˆ dΩ p + δIext − φT φ Since the interpolations for the pressure and volume change are associated with a single element it is possible to solve for their parameters at the element level. The stress σ.

78) ˆ ˆ Using the solutions to 7.86) ¯ σ = D[ ¯ − ] .81) (7.82) gI = 2.CHAPTER 7. The tangent matrix may be expressed in terms of the displacements alone by writing the variational equation 7. · · · ] (for the 4-node element φ = 1 (b) Compute arrays h = Ωe φT φ dΩ φT bJ dΩ Ωe (7. Constitution computation for each quadrature point (a) Compute ˆ = BI uI ¯ ˆ θ = φ(ξ) bI uI ¯ = Idev + 1 mθ 3 0 (7. The residual for a ﬁnite element formulation is most eﬃciently computed from the mixed stress and we note the result is identical to the form of the standard displacement model except for the stress expression used.74 the dependence of 7.85) (7. Numerical integration of strain matrices (a) Compute φ = [1.80) The algorithm for the development of a mixed element based upon the above may be summarized as: 1.84) (7. Mixed volumetric strain displacement matrix ¯ (a) Compute bI = h−1 gI 3. ξ1 . ξ2 .78 on θ and p may be eliminated to give ˆI ¯ ˆ δΠe = δ uT KIJ uJ + δΠ0 + δΠext (7. MIXED FINITE ELEMENT METHODS 55 ¯ σ stresses.79) where ¯ ¯ ¯ ¯ ¯ KIJ = (Kdev )IJ + kI bJ + bT kT + bT kvol bJ I J I (7.83) (7.70 as    ˆ uI (Kdev )IJ kI gJ ˆ ˆT kvol − h  θ  δΠe = δ uT δ θ δ pT  kT ˆI ˆ J T gI −h 0 ˆ p + δΠ0 + δΠext (7.72 and 7.

92) (Kdev )IJ = kI = Ωe BT d φ dΩ I φT dvol φ dΩ Ωe kvol = 6.87) (7. Residual and Stiﬀness Integrals (7.4 Hu-Washizu Variational Theorem: General Problems The ﬁnite element approximation for the mixed formulation of a general linear elastic material (i. MIXED FINITE ELEMENT METHODS 1 T ¯ m σ 3 φT p dΩ ¯ Ωe 56 p = ¯ ¯ π = 4.89) (7..94) · u − θ) dΩ Using the approximations introduced for the isotropic model for the displacement and mixed volume change gives = Idev BI uI + 1 m φ(ξ) θ 3 (7.93) 7.95) . Mixed Pressure ¯ (a) Compute p = φ(ξ) h−1 π 5. Stiﬀness assembly (a) Compute ¯ ¯ ¯ ¯ ¯ KIJ = (Kdev )IJ + kI bJ + bT kT + bT kvol bJ I J I (7.91) (7.90) (7. anistotropic behavior) may be written for a typical element as Πe (u. θ) = 1 2 + p( T Ωe D dΩ − Ωe T D 0 dΩ (7. p.e.88) ¯ (a) Compute mixed stress σ = Idev σ + m p (b) Compute Rσ = − I Ωe BT σ dΩ I BT Ddev Bd Ω J I Ωe (7.CHAPTER 7.

ΘT   gJ dη (KT )J −h k θ co  0  (Pdev )I −  0  π0 (7.104 may be written in the matrix form    J (Kdev )IJ gI (Kco )I u dΠe T 0 −h   p  = (UI )T .98) (7. 7. θ) = 1 [(uI )T (Kdev )IJ uJ + 2 θ T (Kco )J uJ + θ T k θ] 2 − (uI )T (P0 )I − θ T π 0 + pT gI uI − pT h θ dev (7.101) (7. θ) = + 1 2 Idev BI uI + Ωe 57 1 m φ(ξ) θ 3 T D Idev BI uI T I 1 m φ(ξ) θ dΩ − 3 Ωe Ωe 1 Idev BI u + m φ(ξ) θ 3 φT φ dΩ θ Ωe D 0 dΩ (7.103) The mixed variational terms become Πe (u. we can deﬁne the following matrices: (Kdev )IJ = Ωe BT Idev D Idev BJ dΩ I 1 3 φT mT D Idev BJ dΩ Ωe (7.100) (7.104) The ﬁrst variation of Eq.CHAPTER 7.105) . ΠT .102) (Kco )J = k = 1 9 φT mT D m φ dΩ Ωe π0 = 1 3 φT m D Ωe 0 dΩ h = Ωe φT φ dΩ φT bI dΩ Ωe gI = and denote the eﬀects of initial deviatoric strains as (P0 )I = dev Ωe BT Idev D I 0 dΩ (7. MIXED FINITE ELEMENT METHODS which when introduced into the variational theorem gives Πe (u.97) (7. p. p.96) + pT φT bJ dΩ uJ − For symmetric D.99) (7.

equations Eq.117 must be modiﬁed. and.106) and the solution to the third row is p = h−1 [(Kco )J uJ + k θ − π 0 ] Deﬁning a modiﬁed volumetric strain-displacement matrix as ¯ bI = h−1 gJ Substitution of the above results into the ﬁrst equation gives dΠe ¯ ¯ ¯ ¯ = (UI )T (Kdev )IJ + bT (Kco )J + (KT )I bJ + bT k bJ I co I dη ¯ − (P0 )I − bT π 0 dev I uJ (7. dL = DT m (7.115) 3 If D is not symmetric. deﬁne dvol = Then D Idev = D − d mT or Idev D = D − m dT (7. The matrices which involve the elastic moduli may be simpliﬁed by deﬁning some reduced terms. the second row of Eq. are associated with individual elements. MIXED FINITE ELEMENT METHODS 58 Recall that the terms which multiply the variations in pressure.114) dR = D m .113) (7. Accordingly. Θ.112) 1 1 T m D m = mT d 9 3 (7.112 through Eq. Π.107) Thus. and the variation in the volume change. thus. we let3 d = Also. as θ = h−1 gJ uJ (7. 7.110) I co I This operation may be performed after all the integrals over the element are evaluated.7.7.105 may be solved at the element level to give the parameters for the volume change. when D is symmetric the dR and dL terms are equal). Essentially. the stiﬀness matrix for the general anisotropic linear elastic formulation is given by ¯ ¯ ¯ ¯ KIJ = (Kdev )IJ + bT (Kco )J + (KT )I bJ + bT k bJ (7.111) and using these in the remaining equations instead of d (note.CHAPTER 7.108) (7.109) (7. this requires a computation of two d terms as 1 Dm 3 (7. . θ.

116) Finally. MIXED FINITE ELEMENT METHODS which gives 1 1 Idev D m = (D − m dT ) m = d − dvol m = ddev 3 3 59 (7. the deviatoric part of the modulus is now deﬁned in terms of the above as Ddev = Idev D Idev = D − d mT − m dT + dvol m mT For isotropy.124) The matrix for the initial strains is computed as π0 = Ωe φT dT 0 dΩ (7.CHAPTER 7.117) (7. the above expressions reduce to: dvol = K d = K K K 0 0 0 ddev = 0 and  4 −2 −2 0 0 0 −2 4 −2 0 0 0   1 −2 −2 4 0 0 0   = µ 0 0 3 0 0 3 0  0 0 0 0 3 0 0 0 0 0 0 3  T (7. .119) (7.123) (Kco )J = and k = Ωe dvol φT φ dΩ (7.122) (7.118) (7.121) The matrices for the mixed treatment of the symmetric D anisotropic case are computed as: (Kdev )IJ = Ωe BT Ddev BJ dΩ I φT dT BJ dΩ dev Ωe (7.120) Ddev (7.125) which is a 1 × 6 vector.

127) This element is often called B1-P0 (order 1 interpolations for the brick element. order 0 for the pressure/volume change). this matrix is zero. the Q2-P1 element uses quadratic interpolation for displacements (Lagrange interpolations) and linear for the pressure with φ(ξ) = 1 ξ1 ξ2 (7. for example.CHAPTER 7.131) k = Ωe which is a 1 × 1 matrix and for constant dvol becomes k = kvol = dvol Ωe (7.4. we consider the case where the set of shape functions for the displacements is the tri-linear interpolation NI (ξ) = 1 I I I (1 + ξ1 ξ1 ) (1 + ξ2 ξ2 ) (1 + ξ3 ξ3 ) 8 (7. The interpolation for the pressure (and volume change) is constant φ1 = 1 (7.128) Alternatively. The other matrices in the stiﬀness are h = h = Ωe dΩ = Ωe (7. for the order 1 quadrilateral. Higher order elements are also deﬁned. The volume stiﬀness becomes: dvol dΩ (7. MIXED FINITE ELEMENT METHODS 60 7.134) .132) where Ωe is the volume of the element.130) which is a column vector (of size 1 × 24). it is possible to use the interpolations φ(ξ) = 1 x1 (ξ) x2 (ξ) (7.1 Example: Interpolations linear for u and constant φ As an example.129) The matrices for the B1-P0 (or Q1-P0) element reduce to (Kco )J = Ωe dT BJ dΩ dev (7. For isotropy kvol is the bulk modulus times the element volume. In 2-dimensions the element is Q1-P0.126) where ξiI are the values of the natural coordinates at the I-node. For isotropy.133) and gI = Ωe bI dΩ (7.

CHAPTER 7.138) .136) For the mixed element the internal force is computed using PI = Ωe BT (s(u) + p m) dΩ I (7.135) dT 0 dΩ (7. MIXED FINITE ELEMENT METHODS which gives the modiﬁed volumetric strain-displacement equation 1 1 ¯ bI = gI = Ωe Ωe The initial strain term is computed as π0 = Ωe 61 bI dΩ Ωe (7. and the pressure is computed from the mixed form p = 1 ¯ [(Kco )J + kvol bJ ] uJ − π 0 Ωe (7.137) where the deviatoric part of the stress is computed from the displacement form.

Enhanced strains provide great ﬂexibility in designing accurate ﬁnite element models for problems which have constraints or other similar types of diﬃculties. and written as (u. the method does not have the deﬁciencies which are present in the earlier works. ˜) = (s) u + ˜(ξ) (8. In the enhanced strain method we again use the Hu-Washizu variational principle. σ. and the enhanced strains.2) 62 . however. The enhanced strain method is related to earlier works which utilized incompatible displacement modes. which we recall may be written for linear elasticity as Π(u. ˜.1 Hu-Washizu Variational Theorem for Linear Elasticity An alternative to the mixed ﬁnite element method discussed previously is given by the enhanced strain method [16].1) The strain tensor is expressed as an additive sum of the symmetric gradient of the displacement vector. (s) u.Chapter 8 Enhanced Strain Mixed Method 8. = + Ω 1 2 T Ω D dΩ − Ω (s) T D 0 dΩ uT bv dΩ σT ( u − ) dΩ − Ω − Γt uT ¯ dΓ − t Γu ¯ tT (u − u) dΓ = Stationary (8.

˜) = − Ω 1 2 ( Ω (s) u + ˜)T D ( 0 (s) u + ˜) dΩ σ T ˜ dΩ Ω ( (s) u + ˜)T D dΩ − (8. 8.8 and 8.9 implies that.4) (8. Substitution of this result into the remaining equations yields the appropriate displacement equations of equilibrium and constitutive equation for linear elasticity. 8.5) (8.6) ( Ω (s) U)T D ( (s) (s) u + ˜ − 0 0 ) dΩ ST ˜ dΩ Ω ˜ ET [D ( u + ˜ − ) − σ] dΩ − (8. the enhanced strains must vanish. Balance of momentum div [D ( (s) u + ˜ − 0 )] + bv = 0 (8.8) 2. We note Eq. respectively.10. ENHANCED STRAIN MIXED METHOD 63 If we again ignore the integral for the body force and the traction and displacement boundary integrals. from 8. upon use of Eq. Constitutive equations D( (s) (8.10) In addition the boundary conditions for Γu and Γt are obtained.2 the remaining terms become Π(u.7) After integration by parts of the variation of the displacement gradient term (and also considering the body force term). While the enhanced strains vanish pointwise at a solution.CHAPTER 8. at the solution. Strain-displacement equations on the enhanced modes ˜ = 0 3.3) Introducing the variations for each function as uη = u + η U ση = σ + η S and ˜ ˜η = ˜ + η E the variation for the above enhanced principle is given by dΠ = dη + Ω (8. the following Euler equations are obtained for the domain Ω: 1. σ.9) u + ˜ − 0 ) − σ = 0 (8. in an approximate .

σ. The enhanced strains will only vanish in some integral sense over each element.12) respectively. the strain approximations are now given by (u. Accordingly. displacements appear up to ﬁrst derivatives.14) It should be noted that diﬀerent interpolations are introduced for the stress and the enhanced strain terms. stresses. Appropriate interpolations for displacements and stresses are the same as given previously for each element. as will be shown below. σ. ) ≈ e Πe (u. 8. From Eq.15) The variational expression in each element becomes Πe (u. Using the interpolations described above.11) and σ(ξ) = φα (ξ) σ α (t) (8. ENHANCED STRAIN MIXED METHOD 64 scheme based upon the enhanced strain method this is not the case.13) where the approximations for the enhanced strains are assumed as ˜(ξ) = ψ α (ξ) ˜α (t) (8. ˜) = (s) u + ˜(ξ) (8. σ.19) (8.CHAPTER 8.16) Substituting the approximations for displacements. Π(u.2. the continuity conditions we may use in ﬁnite element approximations again are C 0 for the displacements and C −1 for the stresses and enhanced strains. while the stresses and enhanced strains appear without any derivatives. and enhanced strains and replacing with uI = uI + η UI (8. The choice of appropriate approximating functions φα will be aﬀected by the strain approximation. the Hu-Washizu variational theorem may be approximated by summing the integrals over each element. ) (8.18) . ) = Πh (u.17) η σ α = σ α + η Sα η and ˜ ˜α = ˜α + η Eα η (8. ˜) = − Ωe 1 2 ( Ωe (s) u + ˜)T D ( 0 (s) u + ˜) dΩ σ T ˜ dΩ Ωe ( (s) u + ˜)T D dΩ − (8. just as the balance of momentum and constitutive equations are approximated by ﬁnite element solutions. In the enhanced strain principle. Accordingly. σ. and are thus u(ξ) = NI (ξ) uI (t) (8.

(Eα )T ˜ KIJ ΓβI T ˜ ˜ ΓαJ Hαβ uJ ˜β − P0 I ˜α π0 (8.24) (8.27) There are at least three possible ways this may be used: 1.23) (8. ENHANCED STRAIN MIXED METHOD gives the ﬁrst variation in each element as dΠe ˜ = (UI )T . This is not perfect since we will not obtain a method to compute the σ β directly from the variational formulation. 8. 8. (Eα )T .28.22) (8. which means that Qαβ = 0 (8.20 is given in each element by Qαβ β = 0 (8.26) ˜ ΓαJ = ˜ Hαβ = Ωe ψ T D ψ β dΩ α φT ψ β dΩ α Ωe Qαβ = P0 = I Ωe BT D I ψT D α Ωe 0 dΩ dΩ ˜α π0 = 0 The discrete Euler equation generated by the third equation of Eq. which is not a useful result.28) which is the solution to be followed here. A combination of options (a) and (b).21) (8. For a formulation which satisﬁes Eq. 2. 3.20) (8. the variational equations in each element reduce to dΠe = dη ˜ (UI )T . The interpolations for φα are orthogonal to the interpolations ψ β . The β are zero. (Sα )T dη where KIJ = Ωe 65    J   0  ˜ KIJ ΓβI 0 u PI  ΓT H ˜ ˜ αβ Qαβ   ˜β  − π 0  ˜α αJ β T 0 σ 0 Qβα 0 BT D BJ dΩ I ψ T D BJ dΩ α Ωe (8.25) (8.CHAPTER 8.29) .

32) (8.CHAPTER 8. the residual in each element becomes: RI = FI − Ωe BT σ dΩ I ˜ (8.2 Stresses in the Enhanced Method Since the stresses based upon the mixed approximation are no longer available an alternative is needed for computations. as well as. the second of Eq. With the above description. Simo and Rifai suggest using a least square projection technique to obtain the stresses. for outputs and stress projections to nodes.29 may be solved at the element level giving ˜ ˜β = (Hαβ )−1 ˜ ˜α π 0 − ΓαJ uJ (8. Thus.33) 8. ENHANCED STRAIN MIXED METHOD 66 Since the interpolations for the enhanced strains are assumed for each element independently. the variation in each element may be written dΠe dη = η=0 Ωe ( (s) ˜ U)T σ dΩ + Ωe ˜ ˜ ET σ dΩ (8.35) It is noted that the orthogonality condition ST ˜ dΩ = 0 Ωe (8.34) In subsequent development we shall use these stresses for all calculations of arrays.31) (8. 8. however.7 may be deduced as ˜ σ = D( (s) u + ˜ − 0 ) (8.30) which may be substituted into the ﬁrst equation to give dΠe ˜ ˜ = (UI )T KIJ uJ − P0 I dη where and ˜ ˜T ˜ ˜ KIJ = KIJ − ΓβI (Hαβ )−1 ΓαJ ˜ ˜T ˜ ˜α P0 = P0 − ΓβI (Hαβ )−1 π 0 I I (8.36) has been incorporated in the above variation. the stresses which are directly utilized in the variational equation Eq. 8.37) .

the residual for the enhanced modes is computed from ˜ Rα = − Ωe ψ T σ dΩ α ˜ (8.43) . Similarly. it is useful to express the enhanced strains in a similar form.44) j0 −1 ˜ F E(ξ) j(ξ) 0 j0 T F Σ(ξ) j(ξ) 0 (8. σ. j(ξ). Similarly.3 Construction of Enhanced Modes The construction of the enhanced modes depends crucially on the orthogonality requirement being satisﬁed for each element. and gradients of local coordinates. ENHANCED STRAIN MIXED METHOD 67 for the contribution in the element to the global residual. Based upon the study of the shape functions using the alternative representation we recall that the gradient of the displacement involves a constant part and a part which depends only on the determinant of the jacobian matrix. the constant part of the jacobian matrix. J0 . on the bi-unit square element to the global stresses. expressed on the bi-unit square to the global strains. The weighting by the jacobian determinant terms is motivated by the gradient of the shape functions. ξ. ˜.CHAPTER 8. a transformation of the local stresses.41) (8.42) (8. Σ. is given by σ = J0 Σ(ξ) JT 0 These transformations have the property that tr(σ ˜) = j0 ˜ tr(Σ E) j(ξ) (8.39) ˜ which represents a transformation of the local enhanced strains. using the transformation deﬁned at the element center. 8.40) The transformations may also be written in matrix form as ˜ = σ = where for 2-dimensional problems ˜ ET = ˜ ˜ ˜ ˜ E11 E22 E33 2 E12 (8. E.38) Note that the residual for the enhanced modes will vanish at a solution since it belongs to a single element. Accordingly. Using tensor notation we introduce the representations ˜ = j0 −T ˜ J0 E(ξ) J−1 0 j(ξ) (8.

thus. The matrix F0 is given by   11 2 21 12 11 12 (J0 ) J0 J0 0 2 J0 J0 22 21 22  J 12 J 21 (J0 )2 0 2 J0 J0  F0 =  0 0   0 0 1 0 11 21 12 22 11 22 12 21 J0 J0 J0 J0 0 J0 J0 + J0 J0 In matrix form Eq. contrary to what is necessary when using incompatible modes. It should be noted however.50) for the enhanced strains are used.42 and Eq. for non-linear materials. the interpolations  1 0 0 1 Σ =  0 0 0 0 for the stress and 0 0 1 0  0 0  0 1  1 Σ Σ2   3 Σ  Σ4  1 E E 2   3 E  E4 (8. that no ad-hoc assumptions are required in the enhanced formulation. 8. A number of alternatives are discussed in the paper by Simo and Rifai [16]. The integrals of natural coordinates over the bi-linear (2-dimensional) element obey the following properties ξip d2 = 0 4 p+1 2 if p odd if p even (8. 8.51) Thus. . satisfy the orthogonality condition Eq. These interpolations have been incorporated into the element routine elmt04 which has been developed for a linear elastic-viscoelastic material.48.50 contain only linear polynomials in ξ and. the satisfaction of the orthogonality condition may be accomplished by constructing the interpolations in the natural coordinate system and transforming to the global frame using Eq. 8.CHAPTER 8.41 may be written as σT ˜ = The integral over the element becomes σ T ˜ dΩ = j0 Ωe (8. as well as.49)   ξ1 0 0 0  0 ξ2 0 0  ˜  E =  0 0 0 0 0 0 ξ1 ξ2 (8. the interpolations given by Eqs.43.46) j0 ˜ ΣT E j(ξ) (8. which indeed is identical to the modiﬁed incompatible mode formulation [22]. Here we consider the simplest form. 8.49 and Eqs. ENHANCED STRAIN MIXED METHOD ΣT = Σ11 Σ22 Σ33 Σ12 68 (8. 8. 8. For the simplest form.48) Thus.47) ˜ ΣT E d2 = 0 2 (8.45) and ˜ and σ have similar ordering.

W ( ).38) dΠe = dη ( Ωe (s) ˜ U)T σ dΩ + Ωe ˜ ˜ ET σ dΩ (8. ENHANCED STRAIN MIXED METHOD 69 8.55) = (s) u + ˜ In subsequent development we shall use these stresses for all calculations of arrays. Thus. through ∂W σ = (8. Rα . for outputs and stress projections to nodes.54) For the enhanced formulation the computation of stresses is given by ˜ σ = ∂W ∂ (8. where σij = ∂W ∂ ij (8. for the enhanced formulation the variation in each element may be written as (see Eqs.58) ˜ We note above that at a solution the residual.35 to 8.52) ∂ The partial derivative is understood in terms of components.57) I ˜ Ωe Similarly. should vanish independently in each element.5 Solution Strategy: Newton’s Method The solution to a non-linear problem is commonly computed using a sequence of linear approximations. as well as. the residual in each element becomes: RI = FI − BT σ dΩ (8. A popular scheme is Newton’s method.56) In a manner identical to the linear elastic material.4 Non-Linear Elasticity For a non-linear. 8. 8.53) We note that for the linear material model discussed previously that W( ) = 1 2 T D − T D 0 (8. which may be summarized as: . hyperelastic material the stresses are computed from a strain energy density function. the residual for the enhanced modes is computed from ˜ Rα = − Ωe ψ T σ dΩ = 0 α ˜ (8.CHAPTER 8.

2. tol. 8. x.57 and Eq. Convergence may be assessed from | dx(i+1) | < tol | x(i+1) | where | x | is the length of the vector.65) which expands to (i) KIJ = Ωe ˜ ∂σ BI ∂ (i) ∂ dΩ = ∂uJ ˜ (i) BI Dt BJ dΩ Ωe (8.62) 4. . ENHANCED STRAIN MIXED METHOD 1. Using Newton’s method on the set of equations deﬁned by Eq. F(i) = ∂f ∂x ∂f ∂x dx(i+1) = 0 x=x(i) 70 (8.60) (8.58 above gives the problem RI ˜ (i+1) Rα (i+1) (8. Solve the linear problem dx(i+1) = − (F(i) )−1 f (i) and update the solution as x(i+1) = x(i) + dx(i+1) In the above.66) . Repeat steps b. the terms in the Jacobian are deﬁned as (i) KIJ ∂RI = − ∂uJ (i) (8. 8.64) In the above.59) (8.) and c.61) x=x(i) (8. Given the set of equations f (x) = 0 where x are the dependent variables. is the Jacobian or tangent matrix for the equations. Construct the linear part of f about a current point x(i) as f (i+1) ≈ f (i) + where dx(i+1) is an increment of x.63) ≈ = RI ˜ (i) Rα ˆ (i+1) RI 0 (i) − (i) ˜ (i) KIJ ΓβI ˜ (i) ˜ (i) ΓαJ T Hαβ duJ(i+1) d˜β(i+1) (8.) until the solution converges to within a tolerance. F(i) .CHAPTER 8. 3.

FI . duJ(i+1) . 8. ˜ ˜ (i) ˜ (i) d˜β(i+1) = (Hαβ )−1 [R(i) − ΓαJ duJ(i+1) ] α which may be substituted into the ﬁrst equation to give ˆ (i+1) = R(i) − K(i) duJ(i+1) ˜ ˜ RI I IJ where and (i) ˜ (i) ˜ (i) ˜ (i) ˜ RI = RI − ΓβI T (Hαβ )−1 R(i) α (i) ˜ (i) ˜ (i)T ˜ (i) ˜ (i) KIJ = KIJ − ΓβI (Hαβ )−1 ΓαJ (8.71) (8. Thus after adding any nodal forces.70) Since the second equation in Eq. After the solve.68) ˜ (i) ψ T Dt BJ dΩ α ˜ (i) ψ T Dt ψ β dΩ α Ωe (8. ˜ (i) ΓαJ = Ωe ∂2W ∂ ∂ (i) (8.76) The incremental displacements also may be substituted back into Eq.67) deﬁne the tangent moduli for the material.64 is complete at the element level.71 to compute the increments to the enhanced modes. we may perform a partial solution by static condensation.74) The reduced ﬁrst equations may be assembled into the global equations.69) and ˜ (i) Hαβ = (8. Accordingly. 8. For the non-linear elastic material ˜ (i) Dt = Similarly.75) which may be solved for the incremental nodal displacements.72) (8.CHAPTER 8. ENHANCED STRAIN MIXED METHOD where ˜ ∂σ ˜ (i) Dt = ∂ 71 (i) (8. the assembled equations become ˜ (i) KIJ duJ(i+1) = e e ˜ (i) RI + F I (8. these may then be used for the update ˜β(i+1) = ˜β(i) + d˜β(i+1) (8. the new nodal displacements are updated uJ(i+1) = uJ(i) + duJ(i+1) (8.77) .73) (8.

for this algorithm.k+1) (8. until convergence achieved (or a set number of k-steps is completed). 1. It is possible to modify the above algorithm such that the additional storage is reduced to saving only the current values of the enhanced mode parameters.k) Rα (uJ (i).CHAPTER 8. This requires additional storage for the enhanced formulation with respect to that needed for a displacement or a mixed B-bar type of formulation. Compute the linear part of Rα as ˜ ˜ (i.80) with ˜ ˜ (i.k+1) = (Hαβ )−1 R(i.k) α 4. as well as use for subsequent steps for the global i-iterations or to compute stresses. (8. 8. 6. to 4.83) 5.78) where a single superscript i denotes the value of ˜β computed in the last global iteration. Update the solution ˜β(i.k) ψ T Dt ψ β dΩ α (8.k+1) = ˜β(i.k) 3.k) ˜ d˜β(i. ENHANCED STRAIN MIXED METHOD 72 It should be noted that these last steps may not be performed until after the element arrays are assembled and the resulting global problem is solved for the incremental nodal displacements. J Accordingly. ˜β(i.k) ) − Hαβ d˜β(i. Set ˜β(i) = ˜β(i.82) .k+1) = 0 where now ˜ (i. with respect to ˜β only. In the enhanced element for 2-dimensional plane strain applications in FEAP. For k = 0 set ˜β(i. the arrays are moved into history arrays using a pmove routine.81) (s) u(i) + ˜(i. Solve for the increment ˜ (i. ˜ The alternate algorithm is given by linearizing the residual.k+1) (8.79) ˜ (i.k) + d˜β(i.k) Hαβ = Ωe (8.84) and save for the next global iteration. ˜β(i) . Rα . it is necessary to save the arrays used in Eq. Consequently. Set k ← k + 1 and repeat Steps 2.71 for the later update of the enhanced modes. with u (i) known we enter each element calculation with the enhanced strain parameters at the values ˜β(i−1) and perform the following steps.k) = ˜β(i−1) (8.k) = ∂ σ Dt ∂ (8. ˜ 2.

.k) putation of Rα and Hαβ . While the above process has been illustrated for the non-linear elastic material. on uJ (i) Problems All All All Transient Transient Transient All1 73 Table 8.nen.4) ul(ndf.i).nen. however. 8.5) ul(ndf. linearization with respect to both uJ and ˜B is performed.nen. using only 2 or 3 iterations generally suﬃces (even though convergence may not be achieved for the ﬁrst few values of the i-global iterations).nen. ˜ (i) ˜ σ (i) .nen.2) ul(ndf. ∆uJ (i) = uJ (i) − uJ (tn ) Diﬀerence between current and previous solution duJ (i) = uJ (i) − uJ (i − 1) Increment from last iteration Table 8.CHAPTER 8.3) ul(ndf.nen.2: Element Local Arrays The only information to be stored is the ˜β(i) . The deﬁnitions of the entries in the local array are given in Table 8.nen.72 to Eq.1: Element Local Arrays Array ul(ndf. If the k (i) iteration is converged.7) Description local uJ (i) local ∆uJ (i) local duJ (i) ˙ local uJ (i) ˙ local uJ (i) ˙ local uJ (i − 1) used for b.72 to Eq. The global arrays are passed to each element in a local array.2. In FEAP.75 thus simplifying slightly the steps involved.nen. The algorithm requires repeated com(i. Dt . ul(ndf. and a class of viscoplastic materials.k) (i.1.75 for the global steps.6) ul(ndf. 8.1) ul(ndf. J. ENHANCED STRAIN MIXED METHOD Vector Deﬁnition uJ (i) Description Current solution value at each node.c. elasto-plastic materials. and the tangent moduli. leading to Eq. 8. The array ul contains information for the current element according to the deﬁnitions in Table 8. Once the k-iteration is completed. the uJ (i) nodal displacement vector and the ∆uJ (i) and duJ (i) nodal incremental vectors are retained in global arrays. 8. In subsequent presentations we shall discuss the construction of these steps for linear viscoelastic materials. the Rα is zero in Eq. it may be directly extended to any material for which we can iteratively compute the stresses.

2) 3 where σ is the Cauchy stress. s is the stress deviator.1 Isotropic Model The representation of a constitutive equation for linear viscoelasticity may be in the form of either a diﬀerential equation or an integral equation form.3) 3 is strain.4) = e + In the presentation given here we assume that the pressure-volume parts of the behavior are governed by a linear elastic model p = Kθ (9. and θ is the volume change deﬁned in matrix form as θ = mT (9.1) 1 mθ (9.Chapter 9 Linear Viscoelasticity 9. In the discussion to be presented here we assume the material is linear and isotropic.5) where K is the bulk elastic modulus deﬁned in terms of Young’s modulus and Poisson’s ratio as E (9. Accordingly.6) K = 3 (1 − 2 ν) 74 . e is the strain deviator. in matrix form the stress and strain may be split as σ = s + mp and (9. and p is the mean (pressure) stress deﬁned in matrix form as 1 p = mT σ (9.

is measured. LINEAR VISCOELASTICITY The deviatoric parts are assumed to satisfy a linear viscoelastic model. In the diﬀerential equation model the constitutive equation may be written as P (s) = 2 G Q(e) (9. The set of ﬁrst order diﬀerential equations may be integrated for speciﬁed strains. Alternatively. the operator may be written as N s = 2 G (µ0 e + i=1 µi qi ) (9. 75 Linear viscoelastic behavior may be stated in the form of diﬀerential equation models or in the form of integral equations.14) An advantage to the diﬀerential equation form is that it may be easily extended to include aging eﬀects by making the parameters time dependent.13) qi (t) = −∞ exp − t−τ ˙ e(τ ) dτ λ1 (9. The integral form equivalent to the above is expressed in terms of the relaxation modulus function.8) (9. The relaxation modulus function is deﬁned in terms of an idealized experiment in which.12) This form of the representation is equivalent to a generalized Maxwell model (a set of Maxwell models in parallel). An alternative to the linear viscoelastic model in diﬀerential form is to use an integral equation form.11) ˙ qi + 1 i ˙ q = e λi (9. a specimen is subjected to a constant strain. at time labeled zero (t = 0). s(t).CHAPTER 9. The integral for each term is given by the homogeneous diﬀerential equation solution.7) where P and Q are diﬀerential operators expressed as P = pm Q = qm and G = ∂ m−1 ∂m + pm−1 m−1 + · · · + p0 ∂tm ∂t ∂m ∂ m−1 + qm−1 m−1 + · · · + q0 ∂tm ∂t E 2 (1 + ν) (9. e0 .10) is identical to the elastic shear modulus. h qi (t) = C exp h and variation of parameters on C to give t −t λi (9.9) (9. e. qi . and the stress response. For a linear material .

21) In applications involving a linear viscoelastic model.22) .17) or the alternate form G(t) = G (µ0 + N µi exp i=1 −t ) λi (9. Using linearity and superposition for an arbitrary state of strain yields an integral equation speciﬁed as t s(t) = −∞ ˙ G(t − τ ) e(τ ) dτ (9. −t G(t) = G (µ0 + µ1 exp ) (9.e.19) With this form the integral equation form is identical to the diﬀerential equation model for the generalized Maxwell material.18) where µ0 + N µi = 1 i=1 (9. LINEAR VISCOELASTICITY 76 a unique relation is obtained which is independent of the magnitude of the applied strain. In the subsequent discussion we will consider the generalized Maxwell material and let N be 1 (i.. Indeed the integral equation form may be deﬁned as a generalized Maxwell model by assuming the shear relaxation modulus function in the Prony series form N G(t) = G0 + i=1 Gi exp −t λi (9.20) λ1 where µ0 + µ1 = 1 (9.15) where G(t) is deﬁned as the shear relaxation modulus function.CHAPTER 9. Accordingly. it is usually assumed that the material is undisturbed until a time identiﬁed as zero. The addition of more terms may be easily accommodated based upon the one term representation.16) It is noted that the above form is a generalization of the Maxwell material. the standard linear solid). This relation may be written as s(t) = 2 G(t) e0 (9. The integral representation for the the model may be simpliﬁed by dividing the integral into t 0− 0+ t (·) dτ = −∞ −∞ (·) dτ + 0− (·) dτ + 0+ (·) dτ (9. At time zero a strain may be suddenly applied and then varied over subsequent time.

Accordingly. we divide the integral as t tn t (·) dτ = 0 0 (·) dτ + tn (·) dτ (9.CHAPTER 9.31) ∆t where en denotes the value of the strain at time tn and ∆t denotes the time increment t − tn .16 gives t s(t) = 2 G(t) e0 + 2 0 ˙ G(t − τ ) e(τ ) dτ (9. A numerical approximation to ∆h may be employed and one proposal uses a midpoint (one-point) approximation for the integral as [8] ∆h = exp −∆t (e − en ) 2λ1 (9.23) where subsequently the 0 limit on the integral is understood as 0+ .23 gives s(t) = 2 G [µ0 e(t) + µ1 exp −t (e0 + λ1 t exp 0 t ˙ e(τ ) dτ )] λ1 (9.30) λ1 tn λ1 The strain rate is now approximated as constant over each time increment tn to t. tn ≤ τ ≤ t (9.28) −∆t 1 h + ∆h λ1 n (9.25) If we deﬁne the integral as t i1 (t) = 0 exp τ ˙ e(τ ) dτ λ1 (9. LINEAR VISCOELASTICITY 77 The ﬁrst term is zero. 9.32) . The result of this separation when applied to Eq. 9. 9. Substitution of Eq.27) Including the negative exponential multiplier term gives h1 = exp and then h1 (t) = exp where −t 1 i λ1 (9. the second term includes a jump term associated with e0 at time zero.20 into Eq.26) the above separation gives t i (t) = i (tn ) + tn 1 1 exp τ ˙ e(τ ) dτ λ1 (9. thus ∆h = exp ˙ e(τ ) ≈ e(t) − en . and the last term covers the subsequent history of strain.29) −t t τ ˙ exp e(τ ) dτ (9.24) It remains to evaluate the integral.

Thus. For very small time steps a series expansion may be used to yield accurate values.36) This approximation produces a singular ratio for zero time steps. the form is doubly asymptotically accurate.34) h1 (t) = exp − ∆t − ∆t 1 [exp h + (e − en )] 2 λ1 2 λ1 n (9. While the above form is easy to evaluate it has problems when the size of the time step is changed. The use of ﬁnite diﬀerence approximations on the diﬀerential equation form directly does not produce this property. a very large value of the time step producing a zero value. however.35) This form requires only one evaluation of an exponential term. LINEAR VISCOELASTICITY The recursion then becomes t 78 exp tn tn+ 1 τ 2 dτ ≈ ∆t exp λ1 λ1 (9. The required storage is increased by a need to preserve the hi for each quadrature point in the problem and each term in the series. as well as. Thus.33) where tn+ 1 denotes the time 1 (tn + t). Thus. This form gives a recursion which is stable for small and large time steps and gives smooth transitions under variable time steps. The result gives ∆h = λ1 ∆t 1 − exp −∆t λ1 (e − en ) (9. a zero value of the time step produces a correct answer. we need to compute ∂s ∂ = ∂s ∂e ∂e ∂ = ∂s Idev ∂e (9. a more stable form is used in FEAP where the integral over the time step is evaluated in closed form [23]. Accordingly. The constitutive equation now has the simple form s(t) = 2 G [µ0 e(t) + µ1 h1 (t)] (9. It may also be extended for use with thermorheologically simple materials. the limit value is well behaved at a unit value. The implementation of the viscoelastic model into a Newton solution process requires the computation of the tangent tensor. the recurrsion is now given in the form 2 2 h1 (t) = exp and simpliﬁes to −t 1 −∆t 1 ∆t i (t) = exp [hn + exp (e − en )] λ1 λ1 2 λ1 (9.37) The inclusion of more terms in the series reduces to evaluation of additional hi (t) integral recursions.38) .CHAPTER 9. Furthermore.

Again we note that for zero ∆t the full elastic modulus is recovered. Note the simplicity of the additional coding needed to include the linear viscoelastic formulation. 0 The above formulation is incorporated into the subroutine viscoe.43) for the elastic shear modulus.40) ∂h1 = 2 G [µ0 I + µ1 ] ∂e (9.42) = ∆h1 (∆t).39) The only modiﬁcation from a linear elastic material is the substitution of the factor Gvisc = G [µ0 + µ1 ∆h1 (∆t)] (9. the tangent tensor is given by ∂s = 2 G [µ0 + µ1 ∆h1 (∆t)] Idev ∂ (9. LINEAR VISCOELASTICITY 79 where Idev is the deviatoric operator identiﬁed previously. .CHAPTER 9.39 becomes ∂h1 ∂e Thus. whereas for very large increments the equilibrium modulus µG is used. 9. The partial derivative with respect to the deviatoric stress is given by ∂s ∂e If we let ∆h1 = ∆h1 (∆t) (e − en ) the derivative of the last term in Eq. use of the consistently derived tangent modulus terms leads to convergence in one iteration (the second iteration produces a zero residual). Since the material is linear. I (9.41) (9.

2) ∂ e where W is the strain energy density and is expressed as a function of the elastic strains and σ and e are stress and strain energy conjugates. Accordingly. A back stress is deﬁned as α which is related to the plastic 80 . γ.1) where e are the elastic strains and p are the inelastic strains. For a linear hyper-elastic material the stress to elastic strain relation is given by σ = D e = D( − p − 0 ) (10. ˙p = γ ˙ ∂f ∂σ (10. is zero for elastic behavior and pos˙ itive for plastic behavior. The inelastic component of the strain rate is related to the gradient of a loading function with respect to stress. If the material is nonlinear hyper-elastic we may deduce the stress from the expression for the elastic strain energy as ∂W σ = (10. The plastic consistency parameter.Chapter 10 Plasticity Type Formulations 10.1 Plasticity Constitutive Equations The constitutive equations for a material which behaves according to a plasticity type formulation for deformation states which exceed the elastic limit may be expressed by assuming that the strains are decomposed according to = e + p (10.3) In the following discussion we limit our comments to linear elastic materials and also set 0 zero.4) where f is a loading function and γ is a scalar rate term called the plastic consistency ˙ parameter.

ep ) ≤ 0 ¯ (10. and generalized plasticity models. rate dependent plasticity. In the present study both the isotropic and the kinematic hardening moduli will be assumed as constants. rate independent plasticity where g = f (σ.10) thus it is evident that ep is dependent on the integral of γ and the particular load¯ ˙ ing/yield function used to describe the material. Diﬀerent limit equations may be written for rate independent plasticity.9) 3 0 In Eq. Later alternative forms will be introduced to represent other types of material behavior.8 Hiso is an isotropic hardening modulus. . Generally. the model is completed by describing a scalar function.5) 3 3 ∂σ where Hkin is a kinematic hardening modulus.11) is used. The simplest relation is for classical. Commonly. The yield surface is deﬁned in an associative manner. the yield stress in uniaxial tension. using the same function as the loading function. Y0 is related to σy . α. PLASTICITY TYPE FORMULATIONS strain rate through 81 2 2 ∂f Hkin ˙ p = Hkin γ ˙ (10. 10.6) ˙ α = where the stress and back stress appear in the form Σ = σ − α and Y (¯p ) = Y0 + Hiso ep e ¯ (10. The isotropic hardening behavior of the material is included in Y through an eﬀective or accumulated plastic strain deﬁned by 1 t 2 p p 2 p ˙ · ˙ e = ¯ dτ (10. and is expressed as f (σ. associative.CHAPTER 10. Using the deﬁnition of the plastic strain rate the eﬀective plastic strain may also be written as t (10.8) is a function which measures the size of the current yield surface. g. α. which describes the limit behavior of the model. ep ) = F (Σ) − Y (¯p ) ¯ e (10.7) ep = ¯ 0 γ ˙ 2 ∂f ∂f · 3 ∂σ ∂σ 1 2 dt (10.

To initiate the solution at tn+1 a trial state is computed assuming the step is entirely elastic. 10. 18. α. and ep . The ¯n n solution is then advanced to time tn+1 by specifying the strain. a discrete solution at time tn is deﬁned in terms of the state σ n .CHAPTER 10. for the present we restrict our attention to plasticity as deﬁned by Eq. For an inelastic step the stresses must satisfy (1.11 to determine if the step is elastic or whether inelastic terms should be active. PLASTICITY TYPE FORMULATIONS 82 10. The step is initiated by taking the ¯ trial values for plastic quantities pT R p (10. αn .12) n+1 = n αT R = αn n+1 epT R = ep ¯n+1 ¯n and γn+1 = 0 ˙ TR Thus for linear elasticity σT R = D ( n+1 n+1 (10.16 is checked in Eq. however. n+1 . or ep during an elastic increment.18) n+1 . 10. otherwise. integrating non-linear terms using a backward Euler implicit method between tn and tn+1 .2 Solution Algorithm for the Constitutive Equations The solution of the above set of equations may be eﬀected numerically using a variety of algorithms. 10. p .16) The trial stress given by Eq. This implies that there will be no ˙ change to p .17) n+1 and the back stress by ∂f p ˆ αp n+1 = αn + Hkin λn+1 ∂σ (10. Accordingly. Accordingly. A very eﬀective method to integrate plasticity equations is the operator split method with a return map concept [17.14) (10. a correction is required to include the inelastic terms. 10.15) pT R n+1 ) − (10. 3.13) (10. 4] the algorithm may be extended to include a variety of modeling concepts for the limit behavior.6 and Eq.2) for the time tn+1 which requires the rate equations for p and α to be integrated over the time increment.11 above. If the state at tn+1 is elastic the stresses (as well as other state variables) are set equal to the trial value. Recall that a step is elastic when γ is zero. the plastic strain is given by p n+1 = p n + λn+1 ∂f ∂σ (10.

treating the equations Eq.g. The iteration counter is shown as a superscript (j) and initial iterate values are taken as the trial stress and zero λ( 0). Before proceeding with Newton’s method we note that the following relations hold ∂f ∂f ∂Σ ∂f = = ∂σ ∂Σ ∂σ ∂Σ and (10.4 produces ¯n ep ¯n+1 = ep + λn+1 ( 3 ∂f ∂f 1 · )2 2 ∂σ ∂σ (10.18 and 10.CHAPTER 10.20 constitute a set of non-linear equations in terms of the values of σ n+1 .3.. 10. 10.18 and Eq.20 as residual equations.22) ∂f ∂f ∂Σ ∂f = = − (10. integration of Eq.19) (10. To simplify the notation the subscripts on n + 1 are omitted. half machine precision relative to the initial iterate values are used since Newton’s method then guarantees that machine precision is achieved if the next iteration is checked). 10. 10. ep (j)) ¯ (10.11 at tn+1 gives gn+1 = fn+1 = 0 Finally.24) σ n ∂Σ ∂f ˆ R(j) = αn + Hkin λ − α(j) (10.27) −Hkin λ ∂Σ2 (I + Hkin λ ∂Σ2 ) −Hkin ∂λ (λ ∂Σ ) dα T T (j+1) (j) ∂f ∂f dλ Rf − ∂Σ −A ∂Σ . evaluating Eq. PLASTICITY TYPE FORMULATIONS 83 ˆ where Hkin is a constant kinematic hardening parameter and the integral of the consistency parameter is given by λn+1 = γn+1 − γn Similarly. in the form ∂f R(j) = − p − D−1 σ (j) − λ (10. normally. The iterative solution is continued until the norm the residuals are within acceptable tolerance values (e. 10. 10.20) The set of equations Eqs. αn+1 and λn+1 which must be solved for each stress point and each time step of interest.25) α ∂Σ and (j) Rf = − f (σ (j) .21) n+1 (10.26) we may linearize the equations to obtain (note the iteration counter j is omitted in the coeﬃcient array for simplicity)     (j)  ∂2 ∂f ∂ ˆ (λ ∂Σ ) C −λ ∂Σf2 Rσ dσ (j+1) ∂λ  (j)   ˆ ∂f   ∂2f ∂2f ∂ (j+1)  ˆ ˆ = Tα  (10.23) ∂α ∂Σ ∂α ∂Σ Thus. A Newton method may be used to solve the equations. α(j) .

27 are computed and added to obtain the next iterates. for use in the element stiﬀness matrix (if one is used) for the next iteration on the momentum balance equation. by solving KIJ uJ = FI + RI (10. to compute the stress at each Gauss point for a given strain).36)  ˆ ∂Σ ∂λ ∂Σ  ∂Σ T T ∂f ˆ ˆ ˆ D31 D32 D33 − ∂f −A ∂Σ ∂Σ . respectively. Rα = 0 . PLASTICITY TYPE FORMULATIONS where A = and 84 ∂Y ∂¯p e p ∂λ ∂¯ e (10.CHAPTER 10.32) deﬁne the next iterates. . This is accomplished.33) e e (10. the stresses may be used in the ﬁnite element to compute each element residual.35) Denoting the inverse matrix as  −1 ˆ ∂2 ∂f ∂ ˆ ˆ ˆ  C −λ ∂Σf2 (λ ∂Σ ) D11 D12 D13 ∂λ ˆ ˆ ˆ ˆ ˆ D21 D22 D23  = −Hkin λ ∂f (I + Hkin λ ∂ 2 f2 ) −Hkin ∂ (λ ∂f ) (10. if we now consider a linearization with respect to strain only Rσ contributes to the change.29) The solutions to Eq. In addition it is necessary to compute the tangent moduli. The solution is terminated whenever the norms of the residuals are smaller than a selected small tolerance. Once convergence is achieved for each stress point evaluation (i.28) ∂2f ˆ C = D−1 + λ ∂Σ2 (10. The linearization of the residuals with respect to an increment of strain yields Rσ = d .30) α(j+1) = α(j) + dα(j+1) and λ(j+1) = λ(j) + dλ(j+1) (10. That is we need to perform a new solution to see if the strains we used to compute the stresses are correct. Dt . however. The computation of the tangent moduli may be obtained by noting that the computation of the last stress increment in the Newton solution of Eq.31) where KIJ and RI are the element stiﬀness and residual. the residuals will vanish. Accordingly. as before. σ (j+1) = σ (j) + dσ (j+1) (10.e.27 may be written as  −1  (j)   (j+1)  ∂2 ∂f ∂ ˆ C −λ ∂Σf2 (λ ∂Σ ) Rσ dσ ∂λ ˆ ˆ ˆ dα(j+1)  = − Hkin λ ∂ 2 f2 (I + Hkin λ ∂ 2 f2 ) − Hkin ∂ (λ ∂f ) T(j)    α  ∂λ ∂Σ  ∂Σ ∂Σ (j) ∂f T ∂f T dλ(j+1) Rf − ∂Σ −A ∂Σ (10.34) At convergence for the given strain.. R f = 0 (10. 10. 10.

10. Thus.43) and θ = θe + θp (10. the strain is again split into deviatoric. the decomposition into elastic and plastic parts may now be expressed as e = ee + ep (10. al.CHAPTER 10.38) Except for giving the form of f this completes the speciﬁcation of the general algorithm. The method provides the best estimates for the parameters and their sensitivities to errors or inconsistencies in the data.40) An active condition for each surface is denoted when γk ≥ 0. 10.41) . [13]. the tangent moduli for the next momentum iteration is ˆ Dt = D11 85 (10. Multiple yield surfaces may be included by modifying Eq.3 Isotropic plasticity: J2 Model As in previous developments. PLASTICITY TYPE FORMULATIONS The ﬁnal result for the linearization with respect to strain gives  (j+1)  ˆ  D11 d dσ ˆ dα(j+1)  = D21 d  (j+1) ˆ dλ D31 d thus. ep ) ≤ 0 ¯ (10.37) (10. parts and expressed in matrix notation as = e + where θ = mT (10. As constitutive equations become complex the speciﬁcation of the parameters is more diﬃcult. and volumetric (spherical). e. usually only 1 or ˙ 2 of the surfaces are active at any time.42) For our study on inelastic behavior.4 to K ˙p = k=1 γk ˙ ∂fk ∂σ (10.44) 1 mθ 3 (10. α. et. A systematic procedure for determining the parameters from experimental data is given by Ju.39) with each part of the yield surface described by a separate function gk = fk (σ. θ.

through 2 Y = (σy + Hiso ep ) ¯ (10. ∂f ∂f ∂Σ ∂f = = − (10. includes the eﬀects of linear isotropic hardening.47) where Y is the radius of the yield function which is related to the uniaxial yield stress. The invariants of s are denoted as J1 . s. For isotropic materials the yield and loading function may be expressed in terms of the invariants of stress and back stress.46) 3 With the above splits the isotropic linear elastic constitutive equations are given by p = p = K (θ − θp ) and s = 2 G (e − ep ) (10.55) A simple calculation shows that ∂f ∂f = Idev ∂σ ∂s . and J3 and given by J1 = m T s = 0 (10. the limit equation as g = f (s. ep ) = ¯ s − α − Y ep ≤ 0 ¯ (10. thus.54) 3 and. σy .49) J2 = and J3 = det (s) (10.48) In the developments below we restrict plasticity to the deviatoric part only. We write this model using 1 2 J2 = (sT s) 2 = s (10. and pressure (spherical) parts as σ = s + mp where 86 (10.53) 1 T s s 2 (10.51) The simplest formulation is where the function depends only on J2 . PLASTICITY TYPE FORMULATIONS The stress also is split into the deviatoric. α.45) 1 T m σ (10.56) ∂α ∂Σ ∂α ∂Σ .CHAPTER 10. Thus θp vanishes and the yield function can depend only on the deviatoric part of the stress.50) (10.52) and including the back stress. The back stress adjusted value Σ is given by Σ = s − α (10. J2 . ∂f ∂f ∂Σ ∂f = = ∂s ∂Σ ∂s ∂Σ .

63) (10. mT n = 0 . and Σ Σ = n = ∂f ∂σ (10. the evolution of the back stress satisﬁes ˙ α = 2 ∂f 2 Hkin γ ˙ = Hkin γ n ˙ 3 ∂σ 3 (10.59) Based upon the above all aspects of the J2 plasticity model are restricted to deviatoric components only. we can conclude that the back stress evolves such that it is a purely deviatoric quantity.57) Thus. mT α = 0 With this fact we then have the following important properties mT Σ = 0 .61) (10. our model is completed by giving the evolution equations for plastic strain and eﬀective plastic strain in the form ˙ ep = γ n ˙ t (10.69) αn+1 = αn + 2 Hkin λn+1 nn+1 3 2 λn+1 3 − Yn+1 ≤ 0 ep ¯n+1 = ep + ¯n gn+1 = Σn+1 Σn+1 = sn+1 − αn+1 and Yn+1 = 2 2 (σy + Hiso ep ) = Yn + Hiso λn+1 ¯n+1 3 3 (10.68) (10.65) (10.60) (10.CHAPTER 10. nT n = 1 (10. Thus.58) Noting that at the initial state α is zero.67) (10.62) ep = ¯ 0 2 γ dτ ˙ 3 The discrete form of the isotropic J2 model is given by the equations pn+1 = K θn+1 sn+1 = 2 G (en+1 − ep ) n+1 p ep n+1 = en + λn+1 nn+1 (10.66) (10.70) .64) (10. thus. PLASTICITY TYPE FORMULATIONS (where we recall that Idev = I − ∂f = ∂Σ 1 3 87 m mT ).

10. 10.75 with Eq.79) nT R n+1 = ΣT R n+1 ΣT R n+1 (10.CHAPTER 10.75) that is. the coeﬃcient must vanish to obtain a solution. 10. epT R = ep ¯n+1 ¯n (10. If the limit equation is satisﬁed then the trial values deﬁne the solution at tn+1 .73) Noting that Σn+1 = Σn+1 nn+1 Eq. and may be written as the scalar equation Σn+1 − ΣT R n+1 + 2 (G + 1 Hkin ) λn+1 nn+1 = 0 3 (10. 10. Accordingly. ΣT R − Yn = 2 [G + 1 (Hiso + Hkin )] λn+1 3 (10. Accordingly. 10. gn+1 .77) (10.76) Once λn+1 is known it may be combined with the result nn+1 = to give the stress as sn+1 = sT R − 2 G λn+1 nn+1 n+1 (10.74 gives also that sT R − αn is in the n+1 direction nn+1 .64 gives sn+1 = sT R − 2 G λn+1 nn+1 n+1 Next subtracting Eq.70 yields the scalar equation. 10.71 and 10. using Eqs. The second part of the algorithm solves the discrete rate equations using the trial values as initial conditions.78) .66 from Eq. The solution is performed using a trial state based upon the assumption that λT R is zero. If the trial values violate the limit equation.65. n+1 epT R = ep n+1 n . αT R = αn n+1 . 10. 10.74) (10.72 in Eq.71) This yields the trial deviator stresses sT R = 2 G (en+1 − epT R ) = 2 G (en+1 − ep ) n+1 n n+1 (10.72) which may be used to check the limit equation.73 gives Σn+1 = sn+1 − αn+1 = sT R − αn − 2 (G + n+1 1 Hkin ) λn+1 nn+1 3 (10.68 and Eq. In the above ΣT R = sT R − αn n+1 n+1 Combining Eq. PLASTICITY TYPE FORMULATIONS 88 The solution of the J2 model is straight forward and may be accomplished by solving only a scalar equation in λn+1 . 10. it is necessary to perform the second part of the algorithm.

77 will be nonlinear in λn+1 and a linearization and iteration process are required to obtain a solution. We note that for the current strain. to avoid added complexity in the linearization performed below. In practice λn+1 is reduced slightly to that sn+1 is always slightly outside the limit yield condition. this may be achieved by selecting as primary dependent variables the stress. the equation equivalent to Eq. if non-linear kinematic hardening is introduced the reduction to a scalar equation may also be complicated. Accordingly. Accordingly. previously given as λn+1 . Similarly.. If non-linear isotropic hardening is included or alternative forms for the limit equation are employed. It should be understood that λ denotes the value of the solution in the tn+1 step. which we denote by de. for all the other variables.82) Rα = α − αn − and R f = Yn + 2 Hiso λ − Σ = 0 (10.80) 1 3 where tol is a small value. say 10−8 . a check on satisfaction of the momentum equation must be made. however. and etc. the linearization n becomes ∂n = ∂Σ 1 Σ (I − n nT ) = 1 Σ N (10. Once the converged value for λn+1 is known and the stress state determined. and the consistency parameter.84) .81) (10. the back stress. sn+1 . For the solution process developed here. tn+1 .78 is perturbed as λn+1 = ΣT R n+1 2 [G + − Yn (1 + tol) (Hiso + Hkin )] (10. αn+1 . as shown above. the above equations are satisﬁed. does not require a linearization or an iteration process.83) 3 In the above we have deleted speciﬁc reference to the values at tn+1 . to proceed to the next iteration of the momentum equation we consider a linearization of the above equations with respect to a change in the strain also. The solution of the problem. PLASTICITY TYPE FORMULATIONS 89 as well as the plastic strain and the back stress. If the momentum equation is not satisﬁed for the current time. Writing appropriate residual equations as Rs = e − e p − n 1 s − λn = 0 2G 2 λ Hkin n = 0 3 (10. another iteration is necessary to improve the estimate for the state of strain. 10.CHAPTER 10. λn+1 . i. If a Newton type solution method is used it is necessary to compute an appropriate tangent modulus matrix for each stress point in the analysis. n+1 .e. the solution to Eq. e. 10.

for the diﬀerential strains. . and then substituting the result into the third equation to obtain a ﬁnal expression for dλ in terms of de.91) G (Hiso + Hkin ) (10. PLASTICITY TYPE FORMULATIONS 90 Using this result. we consider 1 I λ N + Σ 2G 2 H − 3 λΣTkin N R λ − Σ N λ I + 2 3 Hkin N Σ ds dα = de − n dλ − 2 Hkin n dλ 3 (10.92) Finally. Accordingly. 10.90) Substituting this result back into the ﬁrst of equation Eq. d . This also permits the substitution of alternative limit equations without changing the solution to the ﬁrst part.93) See Appendix E for a discussion on the inverse of this type of matrix.85 to obtain 2 G nT de = 2 [G + 1 (Hiso + Hkin )] dλ 3 (10. Thus.88) and where we have noted that D = 1 + 2 (G + Hkin ) 3 λ Σ = ΣT R Σ (10.85) I + 3 Σ N − 3 Hkin n dα =  − 3 Σ N 2 T T dλ 0 n −n − 3 Hiso The inverse to the coeﬃcient matrix may be computed by ﬁrst solving the ﬁrst two equation for ds and dα in terms of dλ and de.89) This result may be substituted into the third equation in Eq. 10.CHAPTER 10.81 to 10.87 yields the incremental equation which yields the tangent modulus matrix for the algorithm.87) where B and C are given by B = 2Gλ = Σ D 2Gλ ΣT R .86) The solution to this equation is1 ds dα = 2 G (I − B N) BN 2GC N I − CN de − n dλ − 2 Hkin n dλ 3 (10. C = 2 Hkin λ 2 Hkin λ = 3 Σ D 3 ΣT R (10. the tangent becomes ds = 2 G Idev − B (Idev − n nT ) − A n nT d 1 (10. the linearization of Eqs. we obtain ds = 2 G [I − B (I − n nT ) − A n nT ] de where A = G + 1 3 (10.83 yields the set of equations      1 λ I + Σ N − Σλ R N n T ds de 2G    2 λ Hkin 2 λ Hkin 2 0 (10. 10.

Perzyna considers alternatives for representing Φ. through 2 (σy + Hiso ep ) ¯ (10.98) . Accordingly. ep )] − ζ λn+1 = 0 ¯n+1 (10. ep ) = ¯ s − α − Y ep ¯ (10. 10. α. In this section we show how such a formulation may be easily extended to include rate eﬀects in the inelastic behavior. In viscoplasticity. 10. α. f . however. All the other equations for the model remain as given in Section 10. 10. still is given by f (s. ∆t) using a backward Euler evaluation of the integrals to obtain ∆t Φ [f (sn+1 . αn+1 . here we restrict our attention to the simple case given by Φ (f ) = (f )m (10.4 Isotropic viscoplasticity: J2 model The previous section presented the formulation and solution algorithm for a J2 classical plasticity model. For trial stress values for which the yield function deﬁned by Eq. the behavior is inelastic and the return map solution for the viscoplastic model is given by Eqs.67. other models representing the problems of generalized plasticity and generalized viscoplasticity can also be developed by such replacement.96) Y = 3 and. γ. The model selected for exposition is classical viscoplasticity as introduced by Prager for one-dimension and extended to full three-dimensional form by Perzyna [15].3.63 to 10..70. For viscoplasticity. 10. includes the eﬀects of linear isotropic hardening. The function Φ together with the parameter ζ determine the rate dependency of the model. For the viscoplastic model considered here. we ˙ write g = Φ [f (s. ep )] − ζ γ ≤ 0 ¯ ˙ (10.94) where the yield condition. The formulation is completed by integration of the constitutive equation Eq.CHAPTER 10. Indeed. For loading rates which are ﬁnite.95 exceeds zero. PLASTICITY TYPE FORMULATIONS 91 10. the only modiﬁcation to the formulation is the replacement of the limit equation for g.97) where m is a positive integer power. the relationship for g becomes a constitutive equation describing the evolution for the consistency parameter.95) with Y the radius of the yield function which is related to a uniaxial yield stress. Y and σy relate to the inelastic behavior in the limit as γ tends to zero (the rate independent ˙ limit). σy .69 and 10. Other functional forms for Φ may be considered without conceptual diﬃculty. the stress state may lie outside the yield surface.94 for the time interval tn to tn+1 (i.e. thus.

Accordingly. 10.98 to obtain a single nonlinear equation in λn+1 . 10. back stress. plastic strain. and eﬀective plastic strain are computed using the same expressions as for the classical plasticity model.99) For the simple model used here for Φ. The stress. the above becomes ∆t ΣT R − Yn − 2 [G + 1 (Hiso + Hkin )] λn+1 3 m − ζ λn+1 = 0 (10. the above equation is nonlinear and must be solved numerically. 10.100) In general.CHAPTER 10. with results depending on time durations for applying and changing loads on the body.101 implies a rate dependency.75 with Eq. we obtain ∆t Φ [ ΣT R − Yn − 2 (G + 1 (Hiso + Hkin )) λn+1 ] − ζ λn+1 = 0 3 (10.101 to Eq.101) Comparing Eq.80 we can observe that the limit solution for ζ zero is identical to the classical plasticity problem. PLASTICITY TYPE FORMULATIONS 92 The discrete consistency parameter may be obtained by combining the scalar coeﬃcient from Eq. The extension to higher powers of m may be constructed using a Newton solution scheme to solve the non-linear scalar equation. 10. the presence of ∆t in Eq. For the case where m = 1 the equation is linear and has the solution λn+1 ∆t ( ΣT R − Yn (1 + tol)) n+1 = ζ + 2 ∆t [G + 1 (Hiso + Hkin )] 3 (10. 10. . For nonzero ζ.

A common approach is to use penalty methods. Many other conditions could be given.1 Constraint Equations . however. a spherical coordinate frame) where now ¯ u I = T I uI = u I (11. an alternative method is needed to implement general types of constraints. these 93 . 11. the above suﬃce for the present.g.Introduction The solution of many problems requires imposition of constraints as part of the formulation. This type of constraint can be made more general by letting the degrees-of-freedom be associated with a rotated local coordinate system (e. as is done in FEAP. For example. Thus. 11. however. For constraints of the type Eq. The inclusion of the constraints into the ﬁnite element problem may be performed by several diﬀerent approaches.2) ¯ in which u denotes a speciﬁed value.. 11. where we wish to impose the condition for node I that ¯ uI = uI (11.1 or Eq.Chapter 11 Augmented Lagrangian Formulations 11.2 it is easy to directly eliminate the variables associated with uI . if it is desired to solve the incompressible equations for linear elasticity it is necessary to impose the constraint tr( ) = mT = 0 (11.3 presents more diﬃculty to implement.1) Another type of constraint is to impose boundary conditions on a node.3) in which TI is an orthogonal rotation matrix which transforms the degrees-of-freedom from the global Cartesian to the prime system. On the other hand the inclusion of Eq.

CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS

94

are sensitive to the value of the penalty parameter selected. A better approach, which is numerically superior, is to use an augmented Lagrangian approach. This method is an extension to the penalty method and uses values for the penalty parameter which lead to a better conditioned numerical problem. In the sequel we ﬁrst consider penalty methods, based upon a mixed formulation. Subsequently, we show how to extend the mixed penalty treatment to the augmented Lagrangian algorithm which is based on an iterative update procedure generally attributed to Uzawa [1].

11.2

Mixed Penalty Methods for Constraints

Consider a general constraint equation expressed as g(u) = 0 (11.4)

which is to be imposed for some part of the domain, Ωc . The constraint may be included as part of the problem formulation by supplementing the variational problem, Π(u), with the term Πc (u, λ) =
Ωc

λT g(u) dΩ

(11.5)

Deﬁne the variations as uη = u + η U and λη = λ + η Λ The variation of the integral gives the added terms dΠcη dη where =
η=0 Ωc

(11.6) (11.7)

ΛT g(u) dΩ +
Ωc

UT GT λ dΩ

(11.8)

∂g ∂u The Euler equation for the ﬁrst integral leads to the constraint equation. G = g(u) = 0

(11.9)

(11.10)

for each point in Ωc , and the second equation leads to a term which is combined with the variation of the original variational theorem to generate revised Euler equations for the problem. In a ﬁnite element matrix setting we can approximate the λ in each element as
λ λ = Nα (x) λα

(11.11)

CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS

95

and use the usual isoparametric interpolations for u. Thus, Eq. 11.8 generates the matrix problem dΠcη Pλ α G = = (Λα )T (UI )T (11.12) Pλ dη η=0 I where Pλ = α
e Ωce λ Nα g(u) dΩ

(11.13)

and Pλ = I
e Ωce

NI GT λ dΩ

(11.14)

For non-linear constraint equations it is necessary to linearize this expression for combination with the remaining part of the problem. Performing the linearization leads to the problem 0 Kλ dλβ αJ (Λα )T (UI )T (11.15) Kλ Kλ duJ IJ Iβ where Kλ = Iβ
e Ωce λ Nα G NJ dΩ = (Kλ )T Jα e Ωce λ NI GT Nβ dΩ

(11.16) (11.17)

Kλ = αJ and Kλ = IJ
e Ωce

NI λT

∂2g NJ dΩ = ∂u∂u

NI λT
e Ωce

∂G NJ dΩ ∂u

(11.18)

The diﬃculty with the above formulation lies in the fact that there are no terms in Eq. 11.15 which are associated with the diagonals for the λ degrees-of-freedom. Moreover, if the constraints are linear there are no terms on the diagonals for any of the degrees-of-freedom. This greatly, complicates a solution process since for a direct solution the equations must be ordered to eliminate the displacement equations prior to the Lagrange multiplier equations. Furthermore, iterative methods are even more diﬃcult to consider. The deﬁciency associated with the diagonals for the Lagrange multiplier equations may be removed by adding a regularization term to Eq. 11.5. The modiﬁcation to the variational term considered takes the form Πc (u, λ) =
Ωc

λT g(u) dΩ −
Ωc

1 T λ λ dΩ k

(11.19)

where k is a penalty parameter introduced such that in the limit as k goes to inﬁnity the original problem is recovered. dΠcη dη =
η=0 Ωc

ΛT (g(u) −

1 λ) dΩ + k

UT GT λ dΩ
Ωc

(11.20)

CHAPTER 11. AUGMENTED LAGRANGIAN FORMULATIONS The Euler equation for the ﬁrst integral now gives the constraint equation. g(u) − 1 λ = 0 k

96

(11.21)

for each point in Ωc . It is evident that the solution converges to satisfy the constraint only in the limit when k is inﬁnity. The linearization of Eq. 11.20 gives the matrix problem Kλ Kλ dλβ αJ αβ (Λα )T (UI )T (11.22) Kλ Kλ duJ IJ Iβ where Kλ = αβ
e Ωce λ Nα

1 λ I Nβ dΩ k

(11.23)

Many cases for constraints permit the elimination of the equations for λα at a local level. Thus, if a Newton solution scheme is employed the residual equations may be written as Rλ α RI (u) + Rλ I = − Pλ + Kλ λβ α αβ ( ) RI u − Pλ I − − Kλ Kλ αJ αβ Kλ KIJ + Kλ IJ Iβ dλβ duJ = 0 0 (11.24)

This gives the set of equations to solve for the increment as − Kλ Kλ αJ αβ Kλ KIJ + Kλ IJ Iβ Solving the ﬁrst row of Eq. 11.25 gives dλβ = (Kλ )−1 Pλ + Kλ duJ αJ αβ α − λβ (11.26) dλβ duJ = − Pλ + Kλ λβ α αβ RI (u) − Pλ I (11.25)

Since the residual equation for λβ is linear it may be solved to give λβ = and this simpliﬁes Eq. 11.26 to dλβ = Kλ αβ
−1

Kλ αβ

−1

Pλ α

(11.27)

Kλ duJ αJ

(11.28)

which when substituted into the second of Eq. 11.25 once again yields a displacement model for the problem which is expressed as ˆ KIJ duJ = RI (u) − Pλ I where ˆ KIJ = KIJ + Kλ + Kλ Kλ IJ Iβ αβ
−1

(11.29) Kλ αJ

(11.30)

λ. e.31 gives dΠcη dη = η=0 Ωc ΛT g(u) − + 1 λ k dΩ + Ωc UT GT (λ + λA ) dΩ (11. λA ) = Ωc (λ + λA )T g(u) dΩ − Ωc 1 T λ λ dΩ k (11.g.35) and. k. the constraint may be violated by an unacceptable amount. The Uzawa algorithm is the simplest algorithm which may be considered. 11. the augmented Lagrangian method is introduced as an alternative. ill conditioning for the solution to the linear equations will result.31) where λA is the augmented term.CHAPTER 11.32) (11. KIJ . If the values are too large.33) ΛT g(u) dΩ A Ωc The Euler equation for the variation of λ gives the equation g(u) − 1 λ = 0 k (11. Using. For each step in the analysis we assume: . The variation to Eq. but with λA used as the multiplier. thus. The method may be made computationally viable by making the determination of λA an iterative algorithm. The values used then yield stiﬀness modiﬁcations for the second term on the right hand side of Eq.3 Augmented Lagrangian Method for Constraints The augmented Lagrangian strategy presented is a simple modiﬁcation to the perturbed Lagrangian form which now becomes Πc (u. In order to yield a solution which provides an adequate satisfaction of the constraint equation.. iterative solutions become very diﬃcult for these large penalty values. 11. these facts we also note that the algorithm merely reduces to the original Lagrange multiplier method. fairly large values for the penalty parameter should be used (generally on the order of about half machine precision. Furthermore. if too small.34) which may be used to compute λ.30 which are several orders larger than components appearing in the stiﬀness. an alternative approach is needed. The variation for λA gives the constraint equation g(u) = 0 (11. Consequently. 11. the constraint equation is satisﬁed independently of the value of the penalty parameter. 106 or 107 ). and we also conclude that λ must vanish at the solution. In the next section. In the Uzawa algorithm we introduce an outer iteration loop for the augmentation. AUGMENTED LAGRANGIAN FORMULATIONS 97 The above solution process deﬁnes a perturbed Lagrangian form of the penalty solution process.

The convergence rate for the augmented iteration is generally linear. If not converged increase the j counter and repeat steps 2 and 3. 3. Thus. . or the change in the λβ is less than some tolerance times β(j+1) λA proceed to the next time and repeat steps 1 to 3. If the constraint is satisﬁed to within a speciﬁed tolerance. tn+1 . AUGMENTED LAGRANGIAN FORMULATIONS 98 1. Let j be the augmentation iteration counter. To perform the above algorithm it is necessary for the penalty parameter k to be large enough for the iteration to converge.CHAPTER 11. not quadratic as in a Newton solution.39) Pλ α In the above the iteration aspects for the incremental problem are not shown.38) (11. All that is required is that the terms in the added stiﬀness be somewhat larger than the original stiﬀness terms. 4. Use of values with this range in magnitudes leads to 3-6 augmented steps for most problems. The larger k is made the more rapid the convergence.41) is computed using the converged solution from step 2. set j to zero and take the initial value of the augmented multiplier as λA β(j) = λβ (tn ) A (11. Check convergence for the augmented step.37) (11. Solve the non-linear problem − Kλ Kλ αJ αβ Kλ KIJ + Kλ IJ Iβ where λβ = and (PA )λ = I e Ωce λ NI GT Nβ (λβ + λA ) dΩ β(j) dλβ duJ = − Pλ + Kλ λβ α αβ ( RI u) − (PA )λ I Kλ αβ −1 = 0 RI (u) − (PA )λ I (11. The number of non-linear iterations will decrease for the later augmented steps since the violation in the constraint is becoming less and less. For each time. After the iteration for the incremental problem converges update the augmented parameters using β(j+1) β(j) λA = λA + λβ (11. it is desirable for the value to be at least one or two orders in magnitude larger than the conventional stiﬀness terms (as compared with the six or seven used in a penalty approach). Let λβ (0) = 0.36) where the dependence on the n+1 step on the left side is implied. A 2.40) where λβ = Kλ αβ −1 Pλ α (11.

involving multi-point constraints. In some cases (for example. as a special case. For the constant pressure/volume element there is only one equation for each element. Thus the equations to be solved are scalar.CHAPTER 11. frictional contact problems) it is possible to augment in a way which renders a problem which originally has an un-symmetric tangent matrices to one which is symmetric. For the enhanced element there is one equation at each Gauss point so it is also easy to modify. AUGMENTED LAGRANGIAN FORMULATIONS 99 Using the above augmented Lagrangian approach to satisfy the incompressibility constraint leads to a particularly simple update. the situation is slightly more complex. it also includes an option of penalty solution through the perturbed Lagrangian approach (merely omit all augmented steps!). Augmented approaches have been used to solve a variety of problems in ﬁnite element methods. . For more complex situations. the method is the one of current choice since. In general.

1) ˜ where σ is computed for a displacement. is replaced by ¨ bv ← bv − ρ u (12.1 Adding the transient terms The variational equation for a quasi-static problem solved by the ﬁnite element method is expressed as dΠη dη = η=0 e (UI )T Ωe BT σ dΩ − I ˜ Ωe NI bv dΩ − Γe NI ¯ dΓ = 0 t (12. bv . With this replacement the variational equation becomes dΠη dη = η=0 e (UI )T Ωe BT σ dΩ + I ˜ Ωe ¨ NI ρ u dΩ − Ωe NI b dΩ (12. mixed.4) 100 . or enhanced method as described in previous chapters. In order to extend the variational equation to accommodate transient analysis.2) ¨ in which u is the acceleration vector.3) − Γe NI ¯ dΓ = 0 t which leads to the residual equation for each node RI = FI − e Ωe BT σ dΩ − I ˜ e Ωe ¨ NI ρ u dΩ (12.Chapter 12 Transient Analysis of Non-Linear Problems 12. the body force vector.

7) the residual equation becomes ˜ ¨ RI = FI − PI (σ) − MIJ uJ or. and accelerations and these are given as: un ≈ u(tn ) ˙ vn ≈ u(tn ) (12. The practical solution of the equations is accomplished using a time marching scheme. in the total matrix form as ˜ ¨ R = F − P(σ) − M u (12. by ignoring the nodal indices. 12.5) (12.12) . For continuum problems the acceleration is computed from the isoparametric interpolations as ¨ ¨ u = NJ (x) uJ (t) thus. the above equation is a non-linear set of ordinary diﬀerential equations. If we deﬁne ˜ PI (σ) = e Ωe ˜ BI σ dΩ (12. consequently.8) In general. TRANSIENT ANALYSIS 101 The last term is the inertia contribution to the momentum equation. the solution at the ﬁrst increment may be determined.CHAPTER 12. ˙ ¯ u(0) = v0 (12. ¯ u(0) = d0 . given the initial conditions. as described in the next section. the inertia term may be written as ¨ ¨ NI ρ u dΩ = MIJ uJ e Ωe (12. 12. The Newmark method uses approximations to the displacements.9 by a time marching process using the classical Newmark method of solution [14]. The Newmark method is a one-step method which may be used to advance a solution from time tn to time tn+1 . The method is self starting.6) where MIJ is the mass matrix for the problem.11) (12. velocities.10) ¯ ¯ where d0 and v0 are the initial displacement and velocity vectors.2 Newmark Solution of Momentum Equations In this section we illustrate the solution of Eq.9) (12.

12.19) 1 − β) an + β an+1 2 (12. For γ = 1 there is no numerical dissipation. for β = 0. ¯ v0 = v0 (12. The values of β control primarily 2 the stability but also inﬂuence the form of the matrix problem. the method is implicit and a solution step normally involves linearization of the momentum equation and an iterative solution process based on Newton’s method. For a diagonal mass this solution step is very eﬃcient. The Newmark formulas to advance a solution are given by un+1 = un + ∆t vn + ∆t2 ( and vn+1 = vn + ∆t [(1 − γ) an + γ an+1 ] (12. A β of zero leads to a formulation which is called explicit.16) (12. ˜ R0 = F0 − P(σ 0 ) − M a0 = 0 which yields the solution ˜ a0 = M−1 [F0 − P(σ 0 )] this is combined with the initial conditions ¯ u0 = d0 to give a complete state at time zero. where quite large time steps may usually be taken.17) . For example.17 and Eq. respectively. 12. where for no damping. the solution for the acceleration.18 with the momentum equation written at time tn+1 . ˜ Rn+1 = Fn+1 − P(σ n+1 ) − M an+1 = 0 (12. whereas 2 for γ > 1 numerical dissipation is introduced.25 should not be used since they are only conditionally stable with allowable time steps not much larger than the explicit scheme. TRANSIENT ANALYSIS and ¨ an ≈ u(tn ) 102 (12. Accordingly. The advantage of implicit solutions is improved stability. For β non-zero. involves only the mass matrix.18) in which β and γ are numerical parameters which control the stability and numerical dissipation. however.13) The initial state is completed by solving the residual equation at time zero. This method is commonly called trapezoidal rule or constant average acceleration.CHAPTER 12. Values of β less than 0. Accordingly. This involves selecting appropriate values for the variables to initiate .25. the method for linear problems is unconditionally stable. an+1 .14) In order to advance the solution to the next time it is necessary to recast the problem in an iterative form. in general the method is only conditionally stable and very small time steps are needed. The advancement of a solution from one step to the next is completed by combining Eq.15) (12.

un+1 .30. In the sequel we will address the implicit case and use the displacements. linearization of the equations. solution of the linearized equations. For an implicit solution it is best to select the initial value for the iterate as ( un+1 0) = un (12. With the choice Eq.25) γ (i+1) (i) (i+1) vn+1 = vn+1 + dun+1 (12. TRANSIENT ANALYSIS 103 the step. which impede convergence of the Newton method. and M is the mass matrix introduced above.20.20) Any other choice may perturb the displacements in such a way to cause false inelastic values. This vector may be the displacements. Kt = Kt + .30) β ∆t β ∆t2 In Eq. 12. and updating of the variables.28) (12.26) β ∆t and 1 (i+1) (i) (i+1) an+1 = an+1 + du (12.23) (12. as the primary unknowns. the values of the initial state which satisfy the Newmark formulas are given by vn+1 0) = and ( ( 1 − γ β vn + ∆t 1− γ 2β an an (12. 12. or the accelerations. an+1 . vn+1 .22) = β ∆t2 dan+1 = γ ∆t dan+1 (i+1) (12. Since the Newmark formulas are linear and have scalar coeﬃcients they may be directly used in the momentum equation to reduce the primary unknowns to a single vector.24) (i+1) Thus the appropriate update formulas (which also satisfy the Newmark formulas) are given by (i+1) (i) (i+1) un+1 = un+1 + dun+1 (12. especially near boundaries. un+1 . For the explicit case the only viable choice is accelerations.CHAPTER 12. Kt is the tangent stiﬀness matrix as computed for the quasi-static problem.29) = − (i) ∂R ∂R ∂v ∂R ∂a + + ∂u ∂v ∂u ∂a ∂u γ 1 Ct + M (12. Ct is a tangent damping matrix.21) 1 1 vn + 1 − β ∆t 2β Linearizing the Newmark formulas leads to the results an+1 0) = − dun+1 and dvn+1 (i+1) (i+1) (12. the velocities.27) β ∆t2 n+1 The linearization of the momentum equation leads to Kt where Kt or (i) (i) dun+1 (i+1) = Rn+1 (i) (12.

the relations.33) (12. α2 4 β= and (12. 3 2 (12. vn+α ) − M an+1 = 0 where the displacement at the intermeiate point is given by un+α = (1 − α) un + α un+1 and the velocity by vn+α = (1 − α) vn + α vn+1 (12.3 Hilber-Hughes-Taylor (HHT) Algorithm The Newmark algorighm given in the previous section can be altered by considering the residual for the momentum equation to be given by Rn+α = Fn+1 − P(un+α .34) α+γ = are employed. TRANSIENT ANALYSIS 104 12. for short.CHAPTER 12. To reduce the properties to a single parameter. the HHT-method [9] and has been analysed extensively for stability and dissipative properties by Hughes [12].35) Linearization of 12. This algorithm is called the Hilber-HughesTaylor α-method or.31 gives the tangent matrix K = for use in the Newton method K dun+1 = Rn+α (12.31) In the above tn+α = (1 − α)tn + αtn+1 .37) 1 αγ M+ C + αK 2 β∆t β∆t (12.32) (12.36) .

3) 1 As much as possible we adopt the notation that upper case letters refer to quantities deﬁned in the reference conﬁguration and lower case letters to quantities deﬁned in the current conﬁguration.2) where ea are unit base vectors for the current time. common origins and directions are used for the reference and current coordinates. Ω0 . Ω. Exceptions occur when quantities are related to both the reference and current conﬁgurations. In our discussion..Chapter 13 Finite Deformation Formulations: Elasticity 13.1 Kinematics and Deformation The basic equations for solid mechanics may be found in standard references on the subject (e. The position vector at the current time is related to the reference conﬁguration position vector through the mapping x = φ(X. see Chadwick [2] or Gurtin [7]). in the current conﬁguration. a = 1.. t) (13. in a three dimensional space. In Cartesian coordinates the position vector may be described in terms of its components as: XS = XA EA . The position vector in the current conﬁguration may be given in terms of its components as x = xa ea . by Curnier in [5]. After the body is loaded each material point is described by its position vector. A = 1. Here only a summary of the basic equations is presented.1) where EA are unit base vectors. A body B has material points whose positions are given by the vector X in a ﬁxed reference conﬁguration1 . 3 (13. 3 (13. 105 . Solution by ﬁnite element methods is considered by Crisﬁeld in [3].g. x. and by Zienkiewicz and Taylor in [27].

9) to ensure that material volume elements remain positive. a displacement vector.11) Using F directly complicates the development of constitutive equations and it is common to introduce deformation measures which are related completely to either the . and is given by (13. 3 A where δaA is a Kronecker delta quantity such that δaA = In component form we then have xa = δaA XA + ua (13. FINITE DEFORMATION 106 When common origins and directions for the coordinate frames are used.6) A fundamental measure of deformation is described by the deformation gradient relative to X given by ∂φ (13.8) F = ∂X subject to the constraint J = det F > 0 (13. The deformation gradient relates the current conﬁguration to the reference conﬁguration. consequently it has components deﬁned as F = FaA ea ET A The deformation gradient may be expressed in terms of the displacement as F = 1 + ∂u ∂X (13. A = 1. a. The determinant of the deformation gradient maps a volume element in the reference conﬁguration into one in the reference conﬁguration. u. x = 1X + u (13.CHAPTER 13.5) 1 = δa Aea ET .10) where dV is a volume element in the reference conﬁguration and dv its corresponding form in the current conﬁguration.12) (13.4) is used. In the above 1 is a rank two shifter tensor between the two coordinate frames. that is dv = det F dV = J dV (13. Accordingly.7) 1 if a = A 0 if a = A (13. may be introduced as the change between the two frames.

E (do not confuse with the base vectors).16) Deﬁning the displacement vector for the reference conﬁguration as U = 1u with components UA = δaA ua the components of the Green strain may be written in the familiar form EIJ = 1 2 ∂UI ∂UJ ∂UK ∂UK + + ∂XJ ∂XI ∂XI ∂XJ (13.19) (13. expressed as b = F FT The Almansi strain tensor. may be expressed in terms of b as e = 1 1t − b−1 2 (13.CHAPTER 13. Accordingly. FINITE DEFORMATION 107 reference or the current conﬁgurations.18) (13. b.17) In the current conﬁguration a common deformation measure is the left Cauchy-Green deformation tensor. is introduced as 1 E = (C − 10 ) (13. . e. C.22) b and δab is a Kronecker delta for the current conﬁguration. is introduced as C = FT F (13.14) 2 where 10 is the rank two identity tensor with respect to the reference conﬁguration and is given by 10 = δAB EA ET (13. for the reference conﬁguration. the right Cauchy-Green deformation tensor.21) (13.20) where 1t is the rank two identity tensor with respect to the current conﬁguration and is given by 1t = δab ea eT (13.13) Alternatively the Green strain tensor. The Green strain may be expressed in terms of the displacements as E = 1 2 1T ∂u ∂u T ∂u T ∂u + ( ) 1 + ( ) ∂X ∂X ∂X ∂X (13.15) B and δAB is a Kronecker delta for the reference conﬁguration.

respectively. Note that the direction of the traction component is preserved during the transformation and. In ﬁnite deformation problems care must be taken to describe the conﬁguration to which stress is measured. and t0 is traction on the reference conﬁguration.CHAPTER 13. σ.31) (13. FINITE DEFORMATION 108 13.27) (13.29) where ds and dS are surface elements in the current and reference conﬁgurations. thus. remains related to the current conﬁguration forces. τ . P.32) . the ﬁrst Piola-Kirchhoﬀ stress. They are related through the determinant of the deformation gradient as τ = τab ea eT = J σ = J σab ea eT (13.2 Stress and Traction Measures Stress measures the amount of force per unit of area. are measures deﬁned with respect to the current conﬁguration.26) (13.23) b b The second Piola-Kirchhoﬀ stress. and the Kirchhoﬀ stress.25) (13. This form of the traction may be related to reference surface quantity through force relations deﬁned as t ds = t0 dS (13. ds = J N · C−1 N and (13. Using the deﬁnition for traction and stresses we obtain FT n ds = J N dS 1 dS 2 to relate changes in the surface area and transformation of the normals. S. is a stress measure with respect to the reference conﬁguration and has components S = SAB EA ET B The second Piola-Kirchhoﬀ stress is related to the Kirchhoﬀ stress through τ = F S FT Finally.28) (13.24) where n is an unit outward pointing normal to a surface deﬁned in the current conﬁguration.30) where N is an unit outward pointing normal to the reference surface. The reference conﬁguration traction is deduced from the ﬁrst Piola-Kirchhoﬀ stress through t0 = P N (13. is related to S through P = F S = PaA ea ET A which gives τ = P FT For the current conﬁguration traction is given by t = σT n (13. The Cauchy stress.

35) The balance of linear momentum describes the translational equilibrium of a body (or any part of a body) and is obtained by equating the resultant force. (13. rho0 . R. to the rate of ˙ change of the body momentum. Accordingly. through ρ0 = J ρ The total linear momentum of the body is given by p = Ω (13. and balance of angular momentum. bm the resultant force. Accordingly.34) ρ v dv (13. FINITE DEFORMATION 109 13.37) where div is the divergence with respect to the current conﬁguration.39) . R.38) Since the above result must hold for any part of a body a local form for balance of linear momentum may be deduced as ˙ div σ + ρ bm = ρ v This relation is also called the local equilibrium equation for a body. div σ = ∂σab eb ∂xa (13. that is. for a body force per unit mass. ρ bm dv + Ω ∂Ω td s = Ω ˙ ρ v dv (13.36) Introducing the relationship between traction and stress and using the divergence principal leads to the balance of linear momentum relation ˙ [div σ + ρ (bm − v)] dv = 0 Ω (13.3 Balance of Momentum The balance of momentum for a solid body consists of two parts: balance of linear momentum.CHAPTER 13. acting on a body is given by ρ bm dv + Ω ∂Ω t ds = R (13. The balance of linear momentum may be expressed by integrating the surface and body loads over the body. The mass density in the current conﬁguration is related to the reference conﬁguration mass density.33) where rho is the mass density per unit volume and ∂Ω is the surface area of the body. both for the current conﬁguration. p.

CHAPTER 13. At each point on the boundary one (and only one) boundary condition must be speciﬁed for all three .4 Boundary Conditions The basic boundary conditions for a solid region consist of two types: displacement boundary conditions and traction boundary conditions.45) and subsequently to the symmetry of the second Piola-Kirchhoﬀ stress tensor S = ST (13. the Cauchy stress tensor is symmetric and.42) which has the local form ˙ Div P + ρ0 bm = ρ0 v (13.46) 13. leads to the corresponding requirement on P F PT = P FT (13.44) Div P = ∂XA We also note that the symmetry of the Cauchy stress tensor. thus. has only six independent components. 3.41) where the deﬁnitions for rho. Boundary conditions are deﬁned on each part of the boundary for which a component or components of a vector may be speciﬁed without solution of any auxiliary problem. ea . Accordingly. σ. σ and n ds in terms of reference conﬁguration quantities have been used. The conditions are usually given in terms of their components with respect to a local coordinate system deﬁned by the orthogonal basis. we may write the integrals with respect to the reference body as ρ0 bm dV + Ω0 ∂Ω0 t0 dS = Ω0 ˙ ρ0 v dV (13.40) that is. a = 1. Using the divergence principle on the traction term leads to the result ˙ [Div P + ρ0 (bm − v)] dV = 0 Ω0 (13.43) In these relations Div is the divergence with respect to the reference conﬁguration coordinates ∂PaA ea (13. 2. FINITE DEFORMATION 110 Similar relations may be constructed for the balance of angular momentum and lead to the requirement σ = σT (13. The balance of momentum may also be written for the reference conﬁguration using results deduced above.

Thus.51) 13. Generally. x.52) (13. The second type of boundary condition is a traction boundary condition.49) deﬁne components of the displacement with respect to the prime coordinates. Accordingly. for constitutive equations the initial values for internal variables which evolve in time must be given. the traction vector t may be written as t = t a ea Traction boundary conditions may be given for each component by requiring ¯ ta = ta for each point on the boundary. Using the orthogonal basis described above. the boundary condition is non-linear unless points in the reference conﬁguration can be identiﬁed easily (such as ﬁxed points). The conditions describe the initial kinematic state with respect to the reference conﬁguration used to deﬁne the body and the initial state of stress in this position. components of the position vector. may be expressed with respect to the basis as x = x a ea Boundary conditions may now be given for each component by requiring2 xa = xa ¯ (13.47) for each point on the displacement boundary. 2 A speciﬁed quantity is indicated by a superposed bar. these conditions may be a mixture of displacement and traction boundary conditions. FINITE DEFORMATION 111 directions of the basis ea . ∂Ωt . ∂Ωu . u = x − 1 X = ua ea (13. boundary conditions may now be given for each displacement component by requiring ua = u a ¯ (13. .48) (13. In addition. For displacement boundary conditions.CHAPTER 13. The boundary condition may also be expressed in terms of components of the displacement vector. u.5 Initial Conditions Initial conditions describe the state of a body at the start of an analysis.50) In general. (13. (¯ ·).

Finite Elasticity The equations are completed by specifying a material constitution.60) (13.53) 13. W .61) ∂ IC ∂ C ∂ IIC ∂ C ∂J ∂C . Thus.54) (13. J) (13. For a strain energy density expressed in terms of the right Cauchy-Green deformation tensor. (13. 0) = d0 (X) and ˙ ¯ v(0) = φ(X. FINITE DEFORMATION 112 The initial conditions for the kinematic state consist of specifying the position and velocity at time zero. Accordingly.58) We select J instead of IIIC as the measure of the volume change. Thus. we postulate the existence of a strain energy density function. ¯ x(0) = φ(X. As an example.59) 1 2 1 (IC − tr C2 ) = (CKK CLL − CKL CLK ) 2 2 (13. 0) = v0 (X) are speciﬁed at each point in the body. IIC . Here we consider the three invariants as IC = tr C = CKK IIC = and IIIC = det C = J 2 and write the strain energy density as W (C) = W (IC .56) S = 2 ∂C For an isotropic material the strain energy density depends only on the three invariants of the deformation. from which stresses are computed by taking a derivative with respect to a deformation measure.57) (13.55) (13. we consider a ﬁnite deformation form for hyperelasticity. the stress is computed as ∂ W ∂ IC ∂ W ∂ IIC ∂W ∂J S = 2 + 2 + 2 (13. C.CHAPTER 13. the second Piola-Kirchhoﬀ stress tensor is computed as ∂W (13.6 Material Constitution . The initial conditions for stresses are speciﬁed as ¯ σ(0) = σ 0 at each point in the body.

25.62) (13.70) (13. e 13. 13. the stress is computed to be S = 2 ∂W ∂W ∂W 10 + 2 (IC 10 − C) + J C−1 ∂ IC ∂ IIC ∂J 113 (13.72) .CHAPTER 13. FINITE DEFORMATION The derivatives of the invariants may be evaluated as ∂ IC = 10 ∂C ∂ IIC = IC 10 − C ∂C ∂J = J C−1 ∂C Thus.65 gives S = 2 µ (10 − C−1 ) + λ J (J − 1) C−1 (13.68) which may be transformed to give the Kirchhoﬀ stress τ = 2 µ (b − 1t ) + λ J (J − 1) 1t The Cauchy stress is then obtained from σ = τ J (13. The strain energy density is expressed as W (IC .66) As an example. J) = µ (IC − 3 − 2 ln J) + 1 λ (J − 1)2 2 (13.67) The material constants λ and µ have been selected to give the same response in small deformations as a linear elastic material using the Lam´ moduli.65) The second Piola-Kirchhoﬀ stress may be transformed to the Kirchhoﬀ stress using Eq.64) (13. are obtained by the push forward cijkl = 1 FiI FjJ FkK FlL CIJKL J (13. σ.69) Some formulations require computation of the elastic moduli for the ﬁnite elasticity model.71) The spatial elasticities related to the Cauchy stress. The elastic moduli with repect to the reference conﬁguration are deduced from C= 4 I ∂2W ∂C ∂C (13.63) (13. we consider the case of a Neo-Hookean material which includes a compressibility eﬀect. and gives τ = 2 ∂W ∂W ∂W b + 2 (IC b − 1t ) + J 1t ∂ IC ∂ IIC ∂J (13. Substitution into Eq.

74) Other forms of constitutive equations may be introduced using appropriate expansions of the strain energy density function. Accordingly. the basic element arrays evolve from the balance of linear momentum equations written as a variational equation.77) where δu is a variation of the displacement (often called a virtual displacement) which is arbitrary except at the kinematic boundary condition locations. Thus. satisfaction of the variational equation implies satisfaction of the balance of linear momentum at Since the notation for ﬁnite deformation includes use of upper and lower case letters. matrix notation is used as much as possible to express the variational equation. however. for convenience. In a ﬁnite element formulation. FINITE DEFORMATION 114 For the Neo-Hookean model the material modul with respect to the reference conﬁguration are given as −1 −1 −1 −1 CIJKL = λ J (2 J −1 ) CIJ CKL − 2 (µ − λ J (J − 1)) CIK CJL (13. in the reference conﬁguration a variational equation is deﬁned from Eq. ueta → u + δu Furthermore. 13. Since a virtual displacement is an arbitrary function. 13.75) where ¯0 denotes the speciﬁed tractions in the reference conﬁguration and ∂Ω0 t is the t traction boundary for the reference conﬁguration.7 Variational Description A variational theorem for ﬁnite elasticity may be written in the reference conﬁguration as Π(u) = Ω0 W (C(u)) dV − Ω0 uT ρ0 bm dV − ∂Ω0t uT ¯0 dS t (13. the notation for a variation to a quantity is written as δ. (13.75 as3 δΠ = Ω0 ∂W δC dV − ∂C δuT ρ0 bm dV − Ω0 ∂Ω0 t δuT ¯0 dS = 0 t (13. where. it vanishes. the computations are quite delicate (see [19]). ∂Ωu .CHAPTER 13. As an alternative. an elastic formulation may also be expressed in terms of the principal stretches (which are the sqare root of the eigenvalues of C).76) 3 .73) Transforming to spatial quantities gives cijkl = λ (2 J −1 ) δij δkl − 2 ( µ − λ (J − 1)) δik δjl J (13.

26. Eq.80) (13. Some authors refer to the case where the reference conﬁguration is the initial description of the body (at time zero) as a total Lagrangian description and to one which is referred to the previous computed conﬁguration as an updated Lagrangian description.81) − Ω0 δuT ρ0 bm dV − ∂Ω0 This is the variational equation form of the equations which is used for subsequent development of the ﬁnite element arrays.82) 2 where symmetry of the second Piola-Kirchhoﬀ stress is noted. The third and fourth terms of the variational equation represent the eﬀects of body and surface traction loadings.78) (13. 13. 13..56 and constructing the variation of C. the ﬁrst term reduces to ∂W 1 δC = S δC = δFT P ∂C 2 Furthermore. The second term is the stress divergence term which also may be given in terms of the second Piola-Kirchhoﬀ stress as 1 δFT P = δFT F S = δC S = δE S (13. The above variational equation may be transformed to the current conﬁguration as δΠ = Ω ˙ δuT ρ v dv + Ω (δu)T σ] dv δuT ¯ ds = 0 t ∂Ωt (13.79) ˙ δuT ρ0 v dV + Ω0 δFT P dV δuT ¯0 dS = 0 t (13. and. by introducing the inertial forces through the body force as ˙ ¨ bm → bm − v = bm − x where v is the velocity vector. either a total or an updated description can be transformed to the current state.84) − Ω δuT ρ bm dv − The last result is identical to the conventional. The ﬁrst term side represents the inertial terms. For static and quasi-static problems this term may be neglected. the variational equation may be written as δΠ = Ω0 (13. 27]) except that integrals are performed over the deformed current conﬁguration. We note that using Eq. . FINITE DEFORMATION 115 each point in the body as well as the traction boundary conditions.g. small deformation formulation found in earlier chapters and in ﬁnite element texts (e. Representations with respect to a ﬁxed reference conﬁguration are introduced to simplify the development of the basic relations.83) (13. see Hughes [12] or Zienkiewicz and Taylor [26. For the development considered here it is not important which is selected since ultimately all integrations are carried out over the current conﬁguration.CHAPTER 13.

which is not true for problems in beams. Accordingly. but other forms also hold) tr (A B) = AIJ BJI (13. the trace operation denotes the following step (reference conﬁguration tensors are used as an example. The linearization should be considered in a reference conﬁguration representation.CHAPTER 13. does not change) which is not true for a formulation considered directly in the current conﬁguration. In the above.90) J The ﬁrst term on the right hand side leads to the geometric stiﬀness term in a ﬁnite element formulation. additional terms are necessary.85) which are equivalent forms. FINITE DEFORMATION 116 13. a linearization may be written as ∆( Ω0 tr (δF S FT dV ) = Ω0 tr (δF S ∆FT ) dV + Ω0 tr (δF ∆S FT ) dV (13. a formulation based upon the second Piola-Kirchhoﬀ stress and written in tensor form is considered for the linearization step. the second term depends on the material constitution and leads to the material part of the stiﬀness.8 Linearized Equations The stress divergence term may be written in many forms. thus. using δF = and a similar expression for ∆F gives ∆( Ω0 (δu) F (13.89) + + Ω tr ( (δu) ∆σ) dv where 1 F ∆S FT (13. ∆σ = .87) We also note that for a continuum problem ∆ (δF) vanishes.86) Note that in the reference conﬁguration the domain.88) tr (δF S FT dV ) = Ω tr ( (δu) σ (∆u)T ) dv (13. Accordingly. To solve a boundary value problem the nonlinear equations may be linearized and solved as a sequence of linear problems. In this section it is expedient to again use a tensor form of the equations instead of the matrix form used above. Consequently. tr (δFT F S) dV = Ω0 Ω0 tr (δF S FT ) dV (13.e. The linearization may be transformed to the current conﬁguration and expressed in terms of quantities associated with the Cauchy stress. plates and shells and. whereas. as shown above. Ω0 is ﬁxed (i.

97) Using this approximation the variational equation become ˙ δuT ρ v dv + e Ωe Ωe (δu)T σ dv = . 13.9 Element Technology A ﬁnite element discretization may be constructed by dividing the body into ﬁnite elements.91) where Care the material moduli for the material constitution expressed in the reference I conﬁguration. Accordingly. With this approximation the integrals in the variational equation may be approximated as ( · ) dv ≈ Ω Ωh ( · ) dv = e Ωe ( · ) dv (13.93) In the above ∆ is the symmetric part of the gradient of the incremental displacement. FINITE DEFORMATION 117 The material part involves ∆S which is computed for each particular constitutive relation.CHAPTER 13. It is expressed as 1 ∆ = [ ∆u + ( ∆u)T ] (13. we have Ω ≈ Ωh = e Ωe (13.92) where I are the material moduli expressed in the current conﬁguration.96) where Ωe is the domain of an individual element. and Ωh is the domain covered by all the elements.95) which we note is also identical to the form of the linear problem. We note that in general Ωh is an approximation to the domain of the real body since the elements only have mapped polygonal shapes. The moduli c are related through J I = F F C F T FT c I (13.94) 2 Substitution of the above into the term for the material part of the stiﬀness yields tr ( (δu) ∆σ) dv = Ω Ω tr (δ I ∆ )) dv c (13. however. When used with the deﬁnition of ∆σ this may be transformed to the current conﬁguration as ∆σ = I ∆ c (13. This will be discussed later for a particular constitutive equation. in general we seek an expression of the form ∆S = C∆E I (13. e.

respectively. . NI (ξ) are shape functions for node I which maintain suitable continuity between contiguous elements and XI are the coordinates for node I.100) where nen is the number of nodes deﬁning an element.103) (13. we may write approximations for the current conﬁguration as x = NI (ξ) xI (t) the displacements as u = NI (ξ) uI (t) the incremental displacements as ∆u = NI (ξ) ∆uI (t) and the virtual displacements as δu = NI (ξ) δuI (13.104) (13. 27]) we may write for a typical element X = NI (ξ) XI .106) Ωe For procedures to construct a lumped mass see either [26. I are node labels for the element.101) Time dependence is included in the nodal parameters for the current position and displacements.102) (13. The integral for the shape functions deﬁnes the consistent mass matrix for the analysis which may be written as MIJ = NI ρ NJ dv 1t (13. I = 1. the ﬁrst term becomes ˙ δuT ρ v dv = (δuT )I Ωe Ωe ¨ NI ρ NJ dv 1t xJ (13.105) where summation convention is implied for the a and b indices. Adopting an isoparametric formulation (e. Similarly. 13. Appendix 8] or [11]. FINITE DEFORMATION δuT ρ bm dv + e Ωe ∂Ωet 118 δuT ¯ ds t (13. Accordingly. 2.99) An approximate variational solution may be developed by writing trial solutions and test functions for the motions and virtual displacements. 26.10 Consistent and Lumped Mass Matrices Using the above approximations we may discretize the terms in the variational equation for each element. see [12.. nen (13.g. .CHAPTER 13.

11 Stress Divergence Matrix The stress divergence term may be expanded by noting symmetry of σ to give tr [ Ωe (δu) σ] dv = Ωe tr [δ σ] dv (13.CHAPTER 13. The BI matrix describes the transformation from the virtual displacements.3   1    δuI =  = BI δuI 2 NI.2 0   I    δu  0 0 NI.111) Expressing the δ in terms of the virtual displacements gives δ = ∂δu1 ∂x1 ∂δu2 ∂x2 ∂δu3 ∂x3 ∂δu1 ∂x2 + ∂δu2 ∂x1 ∂δu2 ∂x3 + ∂δu3 ∂x2 ∂δu3 ∂x1 + ∂δu1 ∂x3 T (13.107) where δ is given by 1 (δu) + ( (δu))T 2 Introducing matrix notation for σ and δ as δ = σ = and δ = δ 11 (13.1 = δ (13. δuI to the δ .109) T 31 δ 33 2δ 12 2δ 23 2δ (13.112) Using the interpolations for the virtual displacements in each element leads to the matrix relation   NI.114) ∂x1 has been used for conciseness.1 0 0  0 NI. .113) In the above.1 NI.2  δu3 NI.108) σ11 σ22 σ33 σ12 σ23 σ31 δ 22 T (13.110) the stress divergence term may be written as δ Ωe T σ dv (13.1 0  I    0 NI. The stress divergence term may now be written as δ T σ dv = (δuI )T BT σ dv (13. the notation ∂NI (13.3 NI.115) I Ωe Ωe The above expressions are identical to results obtained for the linear problem except that all calculations are based upon coordinates in the current (deformed) conﬁguration.2 NI. FINITE DEFORMATION 119 13.3 0 NI.

In matrix notation the right hand side becomes tr (δ I ∆ ) dv = c Ωe Ωe δ T D ∆ dv (13.120) where D denotes the material moduli in the current conﬁguration given in the matrix representation introduced for the linear problem.121) Thus.118) 13. Accordingly. the material tangent is computed from δ Ωe T D ∆ dv = (δuI )T Ωe BT D BJ dv ∆uJ I (13.119) which is evaluated for a typical element.12 Geometric stiﬀness The geometric stiﬀness for a ﬁnite element formulation is obtained by substituting the interpolations described above into the geometric term for a single element.CHAPTER 13. substitution of the ﬁnite element interpolations into the incremental strain term leads to the result in matrix form ∆ = BJ ∆uJ (13. FINITE DEFORMATION 120 13. tr ( (δu) σ Ωe (∆u)T ) dv = (δuI )T Ωe T tr ( NI σ NJ ) dv ∆uJ (13.117) In component form the expression for the geometric stiﬀness is given as (Kg )IJ = ij Ωe ∂ NI ∂ NJ σkl dv δij ∂ xk ∂ xl (13.122) .116) Evaluation of the right hand side of the above expression leads to the geometric stiﬀness matrix which is given by (Kg )IJ = Ωe T tr ( NI σ NJ ) dv 1t (13.standard B matrix formulation The material tangent matrix is deduced from the term tr ( (δu) ∆σ) dv = Ωe Ωe tr (δ I ∆ ) dv c (13. Furthermore.13 Material tangent matrix .

the body force may be given as δuT ρ bm dv = (δuI )T Ωe Ωe NI ρ bm dv (13.126) where fI is the sum of the body and surface traction terms.123) which again is identical to the linear problem except that all steps are performed for the current conﬁguration.124) and the boundary loading is δuT ¯ tds = (δuI )T ∂Ωe ∂Ωet NI ¯ ds t (13.15 Basic ﬁnite element formulation Accumulating all terms together. 13.128) which may be written in the compact matrix form ¨ M x + N(σ) = f (13.125) 13.127) Since δuI is arbitrary.CHAPTER 13. the variational equation may be written as (δuI )T e ¨ MIJ xJ + Ωe BT σ dv − fI I = 0 (13. fI = Ωe NI ρ bm dv + ∂Ωe t NI ¯ ds t (13. the variational equation leads to the discrete balance of linear momentum equations ¨ MIJ xJ + e Ωe BT σ dv − fI I = 0 (13.14 Loading terms The right hand side terms may be discretized by introducing the interpolations for the virtual displacement.129) . Accordingly. FINITE DEFORMATION and the material tangent matrix is given by (Km )IJ = Ωe 121 BT D BJ dv I (13.

134) (13. FINITE DEFORMATION where N(σ) is the stress divergence vector. yields the solution to a problem.CHAPTER 13. compressible solutions which can be treated by a standard B matrix formulations. Since det F measures the volumetric part we have J = det F = det Fvol det Fdev which leads to the result det Fvol = J and det Fdev = 1 This may be accomplished by using Fvol = J 3 1 for the volumetric part which gives Fdev = J − 2 F 1 1 (13.133) where Fvol measures volumetric part and Fdev the deviatoric part of deformation.138) . the modiﬁed deformation gradient is based upon a separation into volumetric and deviatoric parts as F = Fvol Fdev (13. a modiﬁed deformation gradient.16 Mixed formulation In the mixed formulation used.137) (13.130) A linearization of this set of equations gives the result M ∆¨ + Kt ∆u = R u where Kt = Km + Kg (13. (13.136) (13.135) (13. 122 Solution of this set of equations together with satisfying the material constitution and the displacement boundary conditions. is used.131) 13. Accordingly. as well as. We refer to the method as the standard B-matrix formulation.132) The above description is for a standard displacement type formulation. (as described in [19]). A common solution procedure is to use a Newton type solution method and solve a sequence of linear problems. The mixed formulation is used to permit solution of incompressible and nearly incompressible materials. in a Newton Method we write the momentum equation as ¨ R = f − M x − N(σ) = 0 (13. Thus.

τ . [19]) ˙ δuT ρ v dv + Ω Ω ˜ δuT (σ dev + p 1t ) dv = Ω δuT ρ bm dv + ∂Ω δuT ¯ ds (13. ˜ Π(u. θ is a mixed representation for the determinant of the deformation gradient. and δp (1 − Ω θ ) dv = 0 J (13. and the modiﬁed Kirchhoﬀ stress. θ) dV + p (J − θ) dV (13. Accordingly. θ.142) Ω0 Ω0 − Ω0 uT ρ0 bm dV − ∂Ω0t t uT ¯0 dS (13.139) as the modiﬁed tensor.144) t for the linear momentum equation. and mixed determinant of the modiﬁed deformation gradient.143) A variational equation including the eﬀects of inertia may be constructed following steps above and written as (see. FINITE DEFORMATION 123 for the deviatoric part. the modiﬁed Cauchy stress.145) for the relationship between the mixed pressure and the trace of the stress. δθ( Ω ˜ trσ p − ) dv = 0 3θ J (13. the mixed pressure. The modiﬁed deformation gradient may then be constructed by replacing the volumetric part by a mixed treatment.141) 3θ 3 A three ﬁeld variational statement of the problem is completed by adding to the motion. φ. The modiﬁed right Cauchy-Green deformation tensor is then computed from ˜ ˜ ˜ C = FT F (13. p) = W C(u. In the above expression. σ. we deﬁne ˜ F = θ J 1 3 F (13.146) for the relation between the mixed and the determinant of the deformation gradient. ˜ ˜ In the above. The virtual modiﬁed deformation gradient is now given by δθ 1 ˜ ˜ δF = 1t + ( δu − div δu 1t ) F (13. p.140) ˜ ˜ with similar deﬁnitions for E and b. are related to the modiﬁed second Piola-Kirchhoﬀ stress by ˜˜˜ ˜ ˜ J σ = τ = F S FT (13. θ.CHAPTER 13.147) .

σ. p. p.148) where I is a rank four identity tensor. using the third variational equation. appearing in the above relations may be obtained by ﬁrst computing the mixed volume. for each element (with the constant interpolations for θ and p) integration of the third variational equation yields a solution Ωe θ= (13.CHAPTER 13. This may be combined with the deviatoric part of σ to deﬁne the mixed stress. the stress in this approach is computed using ˜ σ = p 1t + σ dev (13. σ dev . and δu.149) A ﬁnite element implementation for the above may be deduced using the isoparametric interpolations given above for X.151) The pressure.152) Ωe0 for each element. and δp must be given.150) BT (σ dev + p 1t ) dv I ˜ (13. not trσ. The spherical part of the stress is given by the ˜ ˜ mixed pressure. FINITE DEFORMATION 124 ˜ ˜ where S is computed using C as the deformation measure. is then computed using ˜ σ dev = (I − 1 ˜ 1t 1T ) σ = Idev σ t ˜ 3 (13. σ may be determined in each element. . A tangent matrix may be computed for the mixed formulation. The mixed pressure p is computed from trσ using the variational equation given above. The θ may now be ˜ used to deﬁne the modiﬁed deformation quantity and the modiﬁed stress state. in each element. x. u. p. use of the second variational equation yields the mixed pressure as ˜ 1 trσ p = dv (13. Finally. Discretization of the modiﬁed momentum equation gives ¨ ˜ M x + N(σ dev + p 1t ) = F where the stress divergence vector for a typical node is given by ˜ NI (σ dev + p 1t ) = e Ωe (13. Details for the construction are included in [19]. The deviatoric part of the ˜ stress. In the low order elements the above functions are all taken as constant in each element. Accordingly. δθ.153) Ωe0 Ωe 3 θ ˜ in each element. In addition interpolations for θ. θ. where Ωe is the volume of the element in the current conﬁguration and Ωe0 is the volume in the undeformed reference conﬁguration. Thus.

[10] F. Hilber. Hinton. 2 edition. volume 1. New York. 4:245–249.R. Golub and C. Chichester. Improved numerical dissipation for time integration algorithms in structural dynamics.R. Arrow. [12] Thomas J. Curnier. 1993. Matrix Computations. and R.A. 1991. 2 edition. Studies in Non-Linear Programming.B. Englewood Cliﬀs NJ.M. [8] L. Gurtin. [4] M. Dordrecht. Stanford University Press. Hurwicz. 1977. T. Continuum Mechanics. [11] E. volume 2. Uzawa.C. John Wiley and Sons. Hughes. Non-linear Finite Element Analysis of Solids and Structures.F. Van Loan. A numerical procedure for viscoelastic stress analysis. Kluwer Academic Publishers. 1976. Earthquake Engineering and Structural Dynamics.R. Crisﬁeld. The Johns Hopkins University Press.Bibliography [1] K. 1958. and H. [7] Mortin E. Chichester. Chadwick. [6] G. [2] P. and O. T. Peterson. Baltimore MD. Prentice-Hall. 1989. Earthquake Engineering and Structural Dynamics. John Wiley and Sons. Orlando. [9] H. Crisﬁeld. 1968. The Finite Element Method. 125 . Non-linear Finite Element Analysis of Solids and Structures. John Wiley and Sons. [5] A. FL. Prentice-Hall. Taylor. Rock. Herrmann and F. Hughes. Stanford. 1987. Hildebrand.J.J. Methods of Applied Mathematics. 1981. CA. In Proceedings 7th ICRPG Mechanical Behavior Working Group.H. [3] M. L. Academic Press. New York. 1965. A note on mass lumping and related processes in the ﬁnite element method. 5:282–292. Computational Methods in Solid Mechanics. An Introduction to Continuum Mechanics.A. Zienkiewicz. 1997. 1976.E.L.

L. Simo and R. New York. 1986. . Berkeley.W. Vainberg. 10:1211– 1219.L. Perzyna. International Journal for Numerical Methods in Engineering. 1976. and R. University of California.A Finite Element Analysis Program. Pergamon Press. Berkeley. International Journal for Numerical Methods in Engineering. [15] P. 85:67–94. [25] K. [18] J. Taylor.L.C. Computer Methods in Applied Mechanics and Engineering. Taylor. Taylor. A method of computation for structural dynamics.berkeley. Thermomechanical analysis of viscoelastic solids.L. [20] R. Taylor. K.M. A class of mixed assumed strain methods and the method of incompatible modes. Simo. CA. [22] R. In C. and E. A return mapping algorithm for plane stress elastoplasticity. 85:273–310.edu/~rlt. 1990. 1982. Desai. International Journal for Numerical Methods in Engineering. 1964. Washizu. continuum basis and numerical algorithms. January 1987. Rifai.S. Fundamental problems in viscoplasticity. 3 edition. Variational Methods for the Study of Nonlinear Operators.BIBLIOGRAPHY 126 [13] J. [19] J. 22:649–670. Computer Methods in Applied Mechanics and Engineering. User Manual. A non-conforming element for stress analysis. Consistent tangent operators for rate-independent elastoplasticity. University of California. Journal of the Engineering Mechanics Division.S.ce. Wilson.S.L. 1970. [24] M. FEAP . Pister. [23] R. Variational Methods in Elasticity and Plasticity. [17] J.L. Simo and M. Constitutive Equations for Engineering Materials.C.S. Simo and R.L. International Journal for Numerical Methods in Engineering.C. K. Simo and R. Taylor. [14] N. http://www.A Finite Element Analysis Program. J. Advances in Applied Mechanics. 9:243–377.berkeley.edu/~rlt. Taylor. editor. Pister.L. Taylor.J.C. A parameter estimation algorithm for inelastic material behavior. Newmark. P. Ju. Quasi-incompressible ﬁnite elasticity in principal stretches. 1991. 1985. FEAP . 48:101–118. Goudreau. San Francisco. Taylor. Holden-Day Inc.ce.L. [21] R. 2:45–79.L. http://www. and G. 1966. 29:1595–1638. Programmer Manual.C. Beresford. [16] J. 1959..

C. The Finite Element Method.L. McGraw-Hill. [27] O. Zienkiewicz and R. London. The Finite Element Method. Taylor. Taylor. 4 edition. .C. McGraw-Hill. volume 2.L. 1989. Zienkiewicz and R. 4 edition. 1991. London.BIBLIOGRAPHY 127 [26] O. volume 1.

The heat capacity array has been coded separtately in (Table A. and density for an isotropic Fourier material model.6 deﬁnes the Fourier model and Table A. Table A. Nodal projections of the heat ﬂux may then be plotted in graphics mode by FEAP.1 is the interface program to FEAP and Table A.4 and A. 128 . Table A.3 presents the routine used to compute the element tangent and residual arrays and Tables A.Appendix A Heat Transfer Element This appendix contains a FEAP user subprogram to formulate the ﬁnite element arrays needed to solve two-dimensional plane or axisymmetric linear heat transfer problems.8) to permit solution of general linear eigenproblem. thermal conductivity. Table A.7 a routine to compute coordinates in elements.2 is the subprogram to input material parameters for the type of region. heat capacity.8 the routines to output numerical values and nodal projections for the heat ﬂux.

xl.nst) Compute heat flux and print at center of element elseif(isw.9) Input material properties if(isw.xl.s.ndm.5) then call capa02(d. HEAT TRANSFER ELEMENT subroutine elmt02(d.*).r.ndf.nst.d. ix(*) d(*).shp(3.h’ ’prstrs.h’ 129 ndf.or.h’ ’comblk.shp.8) then call stcn02(ix.xl.xl(ndm.ndm.h’ ’eldata.ndm) Compute conductivity (stiffness) matrix elseif(isw.xl.*).isw) c Two dimensional heat transfer element implicit include include include include integer real*8 c none ’cdata.eq.nst) Compute nodal heat flux for print/plots elseif(isw.s.eq.eq. isw.nst.ndm.1) then call inpt02(d) Check of mesh if desired (chec) elseif(isw.ndf.eq.tl(*).numnp) endif end Table A.ndf.2) then call ckisop(ix.APPENDIX A.r(*).hr(nph).4) then call strs02(d.ndm) Compute heat capacity (mass) matrix elseif(isw.ndm.xl.ix.tl.eq.ul.isw.ul.*).3 .ix.xl.ul.s(nst. & ndf.ndm.1: Element Routine for Heat Transfer c c c c c .ix.shp.nel.r.eq.6) then call stif02(d.ndf.hr(nph+numnp).eq.ul.s.r.ix.ul(ndf.

’ Analysis’) end Table A.4)) then d(4) = 2.4)) then d(4) = 1.4)) then d(3) = td(1) elseif(pcomp(name. pcomp. tinput.’spec’.1.’Density ’.d0 elseif(pcomp(name.5/5x.’ ’.5/ 5x.5/5x.d(3).’Conductivity ’.e12.e12.wlab(int(d(4))) d(2) = d(2)*d(3) d(5) = 2 ! number of quadrature points/direction 2000 & & format(5x.d(3).4)) then d(2) = td(1) elseif(pcomp(name.4)) then rflag = .td.d(2).d0 elseif(pcomp(name.false. endif end do ! while if(ior.1) if(pcomp(name.h’ errck.d0 rflag = .e12.’Axisym’/ 130 logical character real*8 data c Input material parameters d(4) = 1.’Linear Heat Conduction Element’// 5x.2000) d(1).a6.’Specific Heat’. wlab(2)*6 d(*).4)) then d(1) = td(1) elseif(pcomp(name.2000) d(1).wlab(int(d(4))) write(iow. rflag name*15.td(1) wlab/’ Plane’.’plan’. HEAT TRANSFER ELEMENT subroutine inpt02(d) implicit include none ’iofile. do while(rflag) errck = tinput(name.0) write(*.APPENDIX A.’dens’.lt.’axis’.true.’cond’.2: Input Routine for Heat Transfer Element .d(2).

false.nel-1 a1 = d(1)*shp(1.shp.i+1))*ctan(1) & + a3*shp(3.nst) implicit include include integer real*8 real*8 real*8 c none ’eldata.xl.j+1) .s.h’ ndf.s(nst.i+1) + a2*shp(2.j*ndf+1) & + (a1*shp(1.ul.h’ ’eltran. gradt(2).9).a3*tdot do i = 0.xsj.l) a3 = d(2)*shp(3.) call flux02(d. a1.l).xl.l).j+1)*xsj*sg(3.APPENDIX A.ndm.nst.9).sg(2.shp.nel.r(ndf.nel.ndm.nel-1 s(i*ndf+1.eq.r. and residual l = nint(d(5)) call int2d(l.ul.3: Stiﬀness for Heat Transfer Element . gradt. tdot.j+1)*xsj*sg(3.l) a2 = d(1)*shp(2.i+1)*ctan(2) end do end do end do end Table A.sg(3.lint.flux.a3.a2*gradt(2) .lint.j+1) = r(1. HEAT TRANSFER ELEMENT 131 subroutine stif02(d.xl.nel) do j = 0.flux(2) Compute tangent matrix (linear).j.j*ndf+1) = s(i*ndf+1.tdot) if(nint(d(4))..a2.xl(ndm. i.*). ix(*) xsj.ndm.j+1)*xsj*sg(3.l) r(1.ndf.ndf.*) shp(3.ix.2) xsj = xsj*radi02(shp.ndm.ix.lint call shape(sg(1.a1*gradt(1) .sg) do l = 1.*).ul(ndf. l. radi02 d(*).*).

xl.ndf.0 . pfr) write(*.nel.0 .xx.3.2002) n.nel.’ 1-flux 2-flux 1-grad format(2i5.gradt(2).ndm) implicit include include include include include integer real*8 real*8 c none ’bdata. tdot.ul.1).xl.1 if(mct.ix.yy.h’ ndf.shp(3.le.head mct = 50 endif xx = radi02(shp.xl(ndm.shp.2002) n.4e12.*). ix(*) xx.*).ma.0d0.ndm.0.) call flux02(d.yy.tdot) mct = mct .2f9.head if(ior.lt.. xsj.flux.2001) o.xsj.flux(2).ma.h’ ’cdata.false.2001) o.0) then write(iow.xl(1.ndm.3) 2-coord’ 2-grad’) end Table A.’element flux’//’ elmt matl 1-coord .ndm. pfr) write(*.0d0.gradt 2001 & 2002 format(a1.1).flux.gradt if(ior.yy. radi02 d(*).ix.ndm.ndf.and.lt.h’ ’fdata.ul.flux.h’ ’iofile.h’ ’eldata. HEAT TRANSFER ELEMENT 132 subroutine strs02(d.xl(2.nel) yy = radi02(shp.nel) write(iow.9) Compute thermal gradient and heat flux call shape(0.20a4//5x.4: Output Routine for Heat Transfer Element .and. gradt.APPENDIX A.xx.shp.ul(ndf.

gradt.l.ul.ndm.flux(2).xl.*).numnp) implicit none integer real*8 real*8 real*8 c ndf.9) Lumped projection routine l = max(2.4).shp.APPENDIX A.*).shp(3.false. ix(*) xsj.nel..st.ul.ndf.sg(3.l).0) then xg = xsj*shp(3.nel.nel.st(numnp.*).shp.2) + flux(2)*xg endif end do end do end Table A.flux.numnp. HEAT TRANSFER ELEMENT 133 subroutine stcn02(ix.xsj.1) = st(ll.l) do j = 1.nel.2) = st(ll.5: Flux Projection Routine for Heat Transfer Element .ll.nel ll = iabs(ix(j)) if(ll.j) dt(ll) = dt(ll) + xg st(ll.sg(2.dt.lint call shape(sg(1.xl.1) + flux(1)*xg st(ll.xg.gt.shp.lint. j.ndf.ndm.ix.nint(d(5))) call int2d(l. tdot dt(numnp).) call flux02(d.lint.xl(ndm.tdot) xsj = xsj*sg(3.d.ndm.sg) do l = 1.d(*) gradt(2).ul(ndf.l).

ndm.*).ul(ndf. gradt.i)*xl(1.0d0 gradt(2) = 0.1) tdot = tdot + shp(3.i.nel.ul.ndf.4) end do flux(1) = -d(1)*gradt(1) flux(2) = -d(1)*gradt(2) end Table A.i)*ul(1.i.flux(*) 134 gradt(1) = 0.nel radi02 = radi02 + shp(3.i.i) end do end Table A.nel.i)*ul(1. shp(3.tdot) implicit include integer real*8 real*8 none ’cdata.1) gradt(2) = gradt(2) + shp(2.flux.0d0 do i = 1.h’ ndm. HEAT TRANSFER ELEMENT subroutine flux02(d.7: Coordinate in Element .gradt(*).nen. xl(ndm.nel radi02.*).nel gradt(1) = gradt(1) + shp(1.i)*ul(1.nel) implicit none integer real*8 c i.ndm.xl.*).6: Thermal Gradient and Flux function radi02(shp.*) Compute element coordinate value radi02 = 0. i tdot d(*).0d0 tdot = 0.shp(3.d0 do i = 1.shp.APPENDIX A.

l).xl(ndm. i.nel) do j = 0. shp(3.ndm.j+1) = r(1.j.9) Compute heat capacity matrix l = nint(d(5)) call int2d(l.j*ndf+1) = s(i*ndf+1.r.i+1) end do end do end do end Table A.sg) do l = 1.2) xsj = xsj*radi02(shp.*).lint.j*ndf+1) + shj*shp(3.sg(3.false.eq. l. shj.h’ ndf.xl.ndm.APPENDIX A.) xsj = xsj*sg(3.sg(2.xl.l) if(nint(d(4)).s.s(nst.l).*).ix.r(ndf.nel-1 shj = d(2)*shp(3.9).xl.ndm.nst..nel.ndm.*).lint.j+1) + shj do i = 0.8: Heat Capacity Routine for Heat Transfer Element .nst) implicit include integer real*8 real*8 c none ’eldata.nel-1 s(i*ndf+1. ix(*) xsj. radi02 d(*).shp.ndf.xsj. HEAT TRANSFER ELEMENT 135 subroutine capa02(d.ix.lint call shape(sg(1.j+1)*xsj r(1.

All elements for continuum (solids) analysis use isoparametric displacement ﬁelds expressed as u= I NI (ξ) uI (B.y  NI.Appendix B Solid Elements B. Two dimensional plane problems = x y  γxy γyz γzx  0 0 NI.y   0 NI.z   NI.3) z γxy T (B.2) (B. Three dimensional problems = x y z NI.y 0   0 NI.x 0   NI.z NI.x  0   0 BI =   NI.1) where NI (ξ) are shape functions and uI are nodal displacements.4) 136 .1 Displacement elements Displacement elements are computed using the virtual work equation written in terms of assumed element displacments. Computation of the derivatives appearing in the strain-displacement matrices is performed as described in Appendix D.x T (B. The strain-displacement matrices for each node are given by: 1.z 2.

y NI.5) γrz T (B. SOLID ELEMENTS  NI.x 3.y   BI =   0 0  NI.APPENDIX B. Two dimensional axisymmetric = r z θ 137  (B.z   BI =  NI  0  r NI.7) .x 0  0 NI.z NI.r  (B.6)  NI.r 0  0 NI.

1 − θ (C.Appendix C Structural Elements C.1 Small displacement element The strain-displacement relations for the small-displacement theory for plane bending in the x1 − x2 global coordinate frame are given as ue (z1 .1 − z2 θ. u2 and θ are displacement functions along the z1 -axis of the frame element. z2 ) = u2 (z1 ) 2 (C. and ﬁnite displacement theories. z2 ) = u1 (z1 ) − z2 θ(z1 ) 1 ue (z1 .1) where z1 and z2 are coordinates and u1 .2 Truss elements Frame elements The current frame elements permit analysis of small displacement. κ the change in curvature and γ is the transverse shearing strain for the cross section. C.1 C.2. second order displacement. These displacements give non-zero strains on each cross section expressed by = 1 γ12 = − z2 κ γ = u1.2) where is the axial strain. Each element is a two node element with linear displacement interpolations in each element. Two types of material constitution are considered: 138 .1 u2 .

APPENDIX C. STRUCTURAL ELEMENTS 1. Resultant theory where      EA 0 0  N     0 EI  M 0 κ =     V γ 0 0 kGA 2. Integration on the cross section where N M =
A

139

(C.3)

1 z2

σ1 ( − z2 κ) dA

(C.4)

C.3 C.4

Plate elements Shell elements

Appendix D Isoparametric Shape Functions for Elements
D.1 Conventional Representation

The shape functions for the bilinear quadrilateral isoparametric element are given by NI (ξ) = 1 I I (1 + ξ1 ξ1 ) (1 + ξ2 ξ2 ) 4 (D.1)

Using these shape functions, the derivatives with respect to the natural coordinates are computed to be ∂NI 1 I I = ξ1 (1 + ξ2 ξ2 ) (D.2) ∂ξ1 4 and ∂NI 1 I I = ξ2 (1 + ξ1 ξ1 ) ∂ξ2 4 (D.3)

Using the shape functions, the interpolation for the global Cartesian coordinates may be expressed in each element as x = NI (ξ) xI (D.4)

where xI are the values of coordinates at the nodes of the element and the repeated index I implies summation over the 4 nodes describing the quadrilateral element. The derivatives of the shape functions with respect to the global coordinates, x, are computed using the chain rule. Accordingly, ∂NI ∂xi ∂NI = ∂ξα ∂ξα ∂xi 140 (D.5)

APPENDIX D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS which may be written in direct (matrix) notation as
ξ NI

141

=

x NI

J

(D.6)

When solved for the derivatives with respect to the global coordinates we obtain
x NI

=

ξ NI

J−1

(D.7)

In the above
x NI

=

∂NI ∂x1 ∂NI ∂x2 ∂NI ∂ξ1 ∂NI ∂ξ2 ∂x1 ∂ξ1 ∂x2 ∂ξ1 ∂x1 ∂ξ2 ∂x2 ∂ξ2

(D.8)

ξ NI =

(D.9)

and J(ξ) = (D.10)

Using the shape functions D.1 for the 4-node element, the terms in J(ξ) have the structure 4 4 ∂xi 1 1 I I I I Jiα = = xi ξα + xI ξα ξβ ξβ (D.11) i ∂ξα 4 I=1 4 I=1 where1 β = mod(α, 2) + 1 (D.12) The constant part of J is evaluated at the point ξ = 0 (commonly named the element center), and is given by ∂NI 1 I = ξα (D.13) ∂ξα 4 thus Jiα (0) = ∂xi ∂ξα =
ξ=0

1 4

4 I xI ξα i I=1

(D.14)

describe the derivatives of the coordinates at the element center. We denote the jacobian at the center as J0 , that is J0 = J(0) (D.15) The global derivatives of the shape functions at the element center become
x NI (0)
1

=

−1 x iNI (0) J0

(D.16)

i i Note that mod(i, j) = i − j j where j is evaluated in integer arithmetic. Thus, mod(1, 2) and mod(3, 2) are both evaluated to be 1, while mod(2, 2) and mod(4, 2) are 0.

22) ξ2 0 0 ξ1 (D.20) D. the jacobian determinant may be expressed as J11 (ξ) J12 (ξ) J21 (ξ) J22 (ξ) = (J0 )11 (J0 )21 (J0 )21 (J0 )22 + ∆J112 ξ2 ∆J121 ξ1 ∆J212 ξ2 ∆J221 ξ1 (D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS In subsequent developments we use the notation biI = ∂NI ∂xi 142 (D. 4 x0 i = I=1 xI NI (0) = i 1 1 (x + x2 + x3 + x4 ) i i i 4 i (D.23) where xi are the element global cartesian coordinates.18) which is the coeﬃcient to the spatially varying part of the jacobian transformation.24) are the values of the global coordinates at the element center.17) ξ=0 to denote the derivatives of the shape functions at the element center. In the development of stabilized elements he introduced the representation 1 NI (ξ) = δI + 4 2 biI (xi − x0 ) + ΓI h(ξ) i i=1 (D. That is.19) which in matrix notation may be written as J(ξ) = J0 + ∆J Ξ where Ξ = and ∆J = ∆J1 ∆J1 ∆J2 ∆J2 (D.2 Alternative Representation in Two Dimensions An alternative representation for the shape functions has been proposed by Belytschko. In subsequent descriptions we will deﬁne ∆ Jiαβ 1 = 4 4 xI i I=1 I I ξα ξβ 1 = 4 4 I I xI ξ1 ξ2 = ∆ Ji i I=1 (D.25) .APPENDIX D.21) (D. h(ξ) = ξ1 ξ2 (D.

30) (D.31) 1 δI + 4 2 biI (xJ − x0 ) + ΓI h(ξ J ) i i i=1 (D.26) where δIJ is the Kronecker delta function for the nodes.29) (D.37) (D.39) (D.33) 1 1 1 1 1 −1 1 −1 x1 x2 x3 x4 i i i i bi1 bi2 bi3 bi4 (D.27) The shape functions at the nodes may be written in the matrix form 1 I = δ 1T + 4 2 bi (xi − x0 1)T + Γ hT i i=1 (D. Evaluating the alternative shape function expression at each node gives NI (ξ J ) = Introducing the notation 1T = hT = xT = i bT = i and the parameter vectors δT = ΓT = δ1 δ2 δ3 δ4 Γ1 Γ2 Γ3 Γ4 (D. These parameters may be evaluated by deﬁning the shape functions at each node and using the fact that NI (ξ J ) = δIJ (D.APPENDIX D. First by multiplying (from the right) by 1. we obtain I1 = 1 = δ In obtaining this result we note that 1T 1 = 4 and xT 1 = 4 x0 i i which gives (xi − x0 1)T 1 = 0 i Finally.36) (D. while the columns are associated with the J where the ξJ are evaluated.34) Note that the rows in the expression are associated with the I in the NI shape functions. the parameters δ and Γ may be easily computed.35) .38) (D. Using this form. The I is a 4 × 4 identity matrix for the element. we note that hT 1 = 0 (D.32) (D.28) (D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 143 and δI and ΓI are constant parameters associated with node I.

47) The factor xh is sometimes called an hour glass shape. the parameters for Γ are computed as 1 Γ = [h − 4 It remains to compute the bi . The derivatives of the function h may also be computed using the chain rule and are given by x h = ξ h J−1 (D. x.APPENDIX D. and when the coordinate.45) For the speciﬁc functional expression for h. we get hT h = 4 2 144 (D. the factor uh . ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS Next by multiplying (again from the right) by h. are given by ∂h ∂NI = biI + ΓI ∂xi ∂xi (D. deﬁnes the magnitude of the hour glass mode.40) (D.42) x h bi ] i i=1 (D. xi . the inverse for the jacobian matrix is given by J 2 −1 1 = j(ξ) ∂x2 ∂ξ2 2 − ∂x1 ∂ξ 1 − ∂x2 ∂ξ ∂x1 ∂ξ1 (D.44) where the biI are constant over the entire element and are computed by the conventional expressions at the center of the element.43) D. 2 (D. the gradient with respect to the natural coordinates is given by ξ2 (D. the derivatives with respect to the global coordinates. u.3 Derivatives of Alternative Formulation Using the alternative expression for the shape functions.41) Ih = h = i=1 x h bi + 4 Γ i where2 xh = xT h i i Thus. is replaced i by the displacement.46) ξh = ξ1 Furthermore. i .

54) The structure of this representation is useful knowledge when we consider the construction of the enhanced part of the strains in Chapter 8. .APPENDIX D.52) the gradient of the displacement may be written as x u = x N I uI = bI + j0 j(ξ) ξ h J−1 ΓI 0 uI (D.53) (D. J.50) where j1 = (J0 )11 ∆ J22 − (J0 )21 ∆ J12 j2 = ∆ J11 (J0 )22 − ∆ J21 (J0 )12 With the above deﬁnitions and bI = b1I b2I (D. For the speciﬁc form of the h(ξ) function the product of the linear part vanishes and the relationship for the gradient simpliﬁes to x h = j0 j(ξ) ξ h J−1 0 (D. Recall that the derivative of a global coordinate with respect to a natural coordinate has a constant and a linear part. The jacobian determinant at the center of the element is computed to be j0 = (J0 )11 (J0 )22 − (J0 )21 (J0 )12 (D.48) where j0 is the value of the jacobian determinant evaluated at the element center.51) (D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 145 where j(ξ) is the determinant of the jacobian transformation matrix.49) We note also that the jacobian determinant at any location in the element may be expressed as j(ξ) = j0 + j1 ξ1 + j2 ξ2 (D.

The inverse may be proved by multiplying the results together to recover the identity matrix. In the case of the deviatoric model A is diagonal and U 146 (E. V are n × k matrices.4) (E.5) which appears in several location in the tangent matrices. U. (A + U VT )−1 = A−1 − A−1 U W VT A−1 where W = (I + VT A−1 U)−1 (E.2) Σ and Σ = s − α We note that n has the properties 1T n = 0 .Appendix E Properties for J2 plasticity models The solution of the J2 plasticity model leads to derivatives of the yield and loading functions in the form ∂f = n (E. where k ≤ n. nT n = 1 (E.6) .3) In the derivation of the tangent the derivative of n leads to ∂2f ∂n = = ∂Σ ∂Σ ∂Σ 1 Σ (1 − n nT ) (E.7) In the above A is an ntimesn matrix.1) ∂Σ where Σ n = (E. and W is a k × k matrix. The inversion of the tangent matrices may be simpliﬁed using the Sherman-Morrison-Woodbury formula which is described on page 51 in Reference [6].

APPENDIX E. PROPERTIES FOR J2 PLASTICITY MODELS

147

and V are proportional to n which is rank 1, thus leading to a scalar W (i.e., a 1 × 1 matrix). There are some properties which need to be noted: n nT (n nT ) = n nT (I − n nT ) n = 0 and (I − n nT ) (I − n nT ) = I − n nT (E.10) (E.8) (E.9)

E.1

Example 1
H1 = A I + B n nT (E.11)

Consider the matrix Using the Sherman-Morrison-Woodward formula the inverse is given by noting that U is equal to B n and V is equal to n, thus H−1 = 1 where W = (1 + The above simpliﬁes to H−1 = 1 1 B (I − n nT ) A A+B (E.14) 1 1 1 I − ( I) (B n)W nT ( I) A A A B −1 A ) = A A + B (E.12)

(E.13)

E.2

Example 2
H2 = C I + D (I − n nT ) (E.15)

Consider the matrix which may be rewritten as H2 = (C + D) I − D n nT for which the solution from example 1 gives H−1 = 2 D 1 (I + n nT ) C + D C (E.17) (E.16)

APPENDIX E. PROPERTIES FOR J2 PLASTICITY MODELS Recollecting into the original type of matrices gives H−1 = 2 1 D [I − (I − n nT )] C C + D

148

(E.18)

A slightly more general form for an inverse results in considering the case with kinematic hardening. In this case we encounter a matrix of the form H = A I + B (I − n nT ) C (I − n nT ) D (I − n nT ) E I + F (I − n nT ) (E.19)

The inverse may be written as H−1 = where a I + b (I − n nT ) c (I − n nT ) T d (I − n n ) e I + f (I − n nT ) (E.20)

1 1 ; e = A E and the remaining coeﬃcients obtained by solving the small matrix problem a = A + B C D E + F The solution to (A.11b) is given by b c d f where G = (A + B) (E + F ) − C D = − 1 E + F −C −D A + B G B C D F a 0 0 e b c d f = − B C D F a 0 0 e

(E.21)

(E.22)

(E.23)

(E.24)

The inverse may be proved by multiplying the two matrices together and show that the result is an identity matrix.

Appendix F Matrix Form for Equations of Solids
F.1 Stress and Strain

Generally the equations of mechanics are expressed using tensor forms. However, it is traditional for ﬁnite element calculations to be performed using matrix forms. This appendix summarizes the transformation of quantities from tensor to matrix form. We begin by writing the forms for stress and strain in a matrix form involving both 9 and 6component forms. The advantage of using the 9-component form is not apparent until considering constitutive equations where direct use of the transformation between the two forms avoids possibility of errors by factors of two. First we show the transformation for the stress and strain tensors into their matrix representations. Here, for example, the components of the stress in tensor form may be given as   σ11 σ12 σ13 σij =  σ21 σ22 σ23  (F.1) σ31 σ32 σ33 and reordered into the 9-component vector as σ= σ11 σ22 σ33 σ12 σ21 σ23 σ32 σ31 σ13
T

(F.2)

Conservation of angular momentum requires the stress to be symmetric, thus satisfying σij = σji (F.3)

This permits the independent components of stress to be written in a 6-component matrix form as T σ = σ11 σ22 σ33 σ12 σ23 σ31 (F.4) In the sequel we shall use an underscore to indicate a 9-component form and omit the underscore for the 6-component form. 149

2 Split into Deviatoric and Spherical Components Using the matrix form we can write the split of stress and strain in their deviator and spherical components as (F. P.6) =  (F.7) and reordered into the 9-component vector as = T 11 22 33 12 21 23 32 31 13 (F. deﬁned by  2  0   0   0 1  P=  0 2   0   0   0 0 giving σ = PT σ In a similar manner we can write the components of the strain tensor as   11 21 31 12 22 32 13 23 33 ij 150 to the 9-component form using a simple pro0 2 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 1 1               (F. MATRIX FORM FOR EQUATIONS OF SOLIDS The 6-component form may be related jector matrix.APPENDIX F.5) (F.9) This permits the independent components of strain to be written in a 6-component matrix form as T = (F.8) Strain-displacement relations give symmetry of strain as ij = ji (F.10) 11 22 33 γ12 γ23 γ31 where γiij are the engineering components of the shearing strain given by γij = 2 ij (F.12) σ = s + mp .11) F.

17) (F.13) (F.22) Using the above matrix forms we can obtain expressions for the deviatoric stress and strain matrices in terms of the full stress and strain values.18) (F. respectively. Accordingly.14) (F.24) σ = s + m mT σ 3 which solve to give 1 1 s = σ − m mT σ = I − m mT σ (F.20) These also are related to their 9-component form using the P projector and may be written as s = PT s and e = P e (F.19) T (F.APPENDIX F.16) The splits may also be written in 6-component form as σ = s + mp and =e+ where s= and e= e11 e22 e33 2 e12 2 e23 2 e31 s11 s22 s33 s12 s23 s31 1 m εv 3 T (F.25) 3 3 . given by =e+ p= and εv = m T The matrix m is a trace projector deﬁned by m= 1 1 1 0 0 0 0 0 0 1 T m σ 3 (F.23) 3 and 1 (F.21) The 6-component projector m is likewise related to its 9-component counterpart through m = PT m = 1 1 1 0 0 0 T (F. MATRIX FORM FOR EQUATIONS OF SOLIDS and 151 1 m εv 3 where p and εv are the pressure and volume change. for the stress we have the two relations 1 σ = s + m mT σ (F.15) (F.

MATRIX FORM FOR EQUATIONS OF SOLIDS and s=σ− 152 1 1 m mT σ = I − m mT 3 3 σ (F. In index notation these are expressed as σij = Cijkl kl (F.32) ∂ ij the elastic constants also possess the major symmetries Cijkl = Cklij We introduce the matrix forms for linear elasticity as σ=D and σ=D (F.3 Linear Elastic Constitutive Equations Let us now consider the relations for linear elastic constitutive equations.28) (F. respectively. (F.27) Similarly for strains we have the deviatoric relations e= − and e= − 1 m mT = Idev 3 1 m mT = Idev 3 (F.33) .APPENDIX F.31) From notions of hyperelasticity where stress is deduced from the stored energy function W ( ) as ∂W σij = (F.30) where Cijkl are the elastic moduli and possess the minor symmetries Cijkl = Cjikl = Cijlk (F. We deﬁne the two deviatoric projectors as Idev = I − 1 m mT 3 and Idev = I − 1 m mT 3 (F.35) where D is a 9 × 9 matrix of elastic constants and D is a 6 × 6 matrix of elastic constants.34) (F.29) F.26) where I and I are identity matrices of size 9 and 6.

This may be directly related to a matrix form as (F. Applying the projector rules (which shows why we only need the two forms given above) we obtain σ = PT σ = PT D = PT D P = D (F. (F.37) F.1: Matrix and tensor index maps Form Index Matrix 1 2 3 4 5 6 Tensor 11 22 33 12 & 21 23 & 32 31 & 13 Table F. Writing the relationship for the constitution as σij = Cijkl kl (F.2.36) which gives the relation between the two elastic moduli as D = PT D P Entries in D use the index maps shown in Table F.42) .38) where in index form εv = kk .39) we obtain the tensor form of the elastic moduli as Cijkl = λ δij δkl + 2 µ Iijkl (F.41) σ=D where D = λ m mT + 2 µ I Form Index Matrix 1 2 3 4 5 6 7 8 9 Tensor 11 22 33 12 21 23 32 31 13 Table F.1.40) where Iijkl is the rank-4 tensor identity.3. MATRIX FORM FOR EQUATIONS OF SOLIDS 153 Construction of D follows directly from Cijkl using the index maps shown in Table F.1 Example: Isotropic behavior As an example we consider the isotropic linear elastic relations expressed in terms of the Lam` parameters as e σij = λ δij εv + 2 µ ij (F.APPENDIX F.2: Matrix and tensor index maps (F.

MATRIX FORM FOR EQUATIONS OF SOLIDS 154 Applying the projector as indicated in Eq.45) We note that this gives the shear equations with the correct factors to match the use of the engineering components.43)   1  T P P = I0 =  2    Thus we can also write Eq. F. F. One of which is classical elasto-plasticity which we will consider later.37 we obtain the 6 × 6 matrix form as D = λ m mT + 2 µ PT P where m is given by Eq. F.43 as D = λ m mT + 2 µ I0 (F.22 and  2 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1         (F.APPENDIX F.38 for each of the independent stress components and introducing the deﬁnition for engineering shearing strain. . F. While this may be obtained also by merely writing Eq. the above process provides a direct way to construct the constitutive model for a wide range of material behavior.44) (F.