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
quadrature weights.
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,
perform numerical quadrature, etc.
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)
Performing the derivatives leads to
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)
Finally, for a 2-dimensional axisymmetric problem (with no torsional loading) the
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
reduced quadrature on another leads to a method called selective reduced quadrature.
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
are quadrature points and quadrature weights, respectively. For brick
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
) at quadrature
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
, and gradients of local
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
adding any nodal forces, F
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
loading function with respect to stress. Accordingly,
˙
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
, to avoid added
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
12.1 Adding the transient terms
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
gradient as
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
loadings.
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.
13.14 Loading terms
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)
and the boundary loading is
_
∂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
MƬ 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)
which leads to the result
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 xsj, a1,a2,a3, tdot, radi02
real*8 d(*),ul(ndf,*),xl(ndm,*),s(nst,*),r(ndf,*)
real*8 shp(3,9),sg(3,9), gradt(2),flux(2)
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.)
call flux02(d,shp,ul,ndf,nel, gradt,flux,tdot)
if(nint(d(4)).eq.2) xsj = xsj*radi02(shp,xl,ndm,nel)
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)
r(1,j+1) = r(1,j+1) - a1*gradt(1) - a2*gradt(2) - a3*tdot
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(*)
real*8 xx,yy, xsj, tdot, radi02
real*8 d(*),ul(ndf,*),xl(ndm,*),gradt(2),flux(2),shp(3,9)
c Compute thermal gradient and heat flux
call shape(0.0d0,0.0d0,xl,shp,xsj,ndm,nel,ix,.false.)
call flux02(d,shp,ul,ndf,nel, gradt,flux,tdot)
mct = mct - 1
if(mct.le.0) then
write(iow,2001) o,head
if(ior.lt.0 .and. pfr) write(*,2001) o,head
mct = 50
endif
xx = radi02(shp,xl(1,1),ndm,nel)
yy = radi02(shp,xl(2,1),ndm,nel)
write(iow,2002) n,ma,xx,yy,flux,gradt
if(ior.lt.0 .and. pfr) write(*,2002) n,ma,xx,yy,flux,gradt
2001 format(a1,20a4//5x,’element flux’//’ elmt matl 1-coord 2-coord’
& ,’ 1-flux 2-flux 1-grad 2-grad’)
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(*)
real*8 gradt(2),flux(2),ul(ndf,*),sg(3,9)
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.)
call flux02(d,shp,ul,ndf,nel, gradt,flux,tdot)
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
subroutine flux02(d,shp,ul,ndf,nel, gradt,flux,tdot)
implicit none
include ’cdata.h’
integer ndm,nel, i
real*8 tdot
real*8 d(*),shp(3,*),ul(ndf,nen,*),gradt(*),flux(*)
gradt(1) = 0.0d0
gradt(2) = 0.0d0
tdot = 0.0d0
do i = 1,nel
gradt(1) = gradt(1) + shp(1,i)*ul(1,i,1)
gradt(2) = gradt(2) + shp(2,i)*ul(1,i,1)
tdot = tdot + shp(3,i)*ul(1,i,4)
end do
flux(1) = -d(1)*gradt(1)
flux(2) = -d(1)*gradt(2)
end
Table A.6: Thermal Gradient and Flux
function radi02(shp,xl,ndm,nel)
implicit none
integer i,ndm,nel
real*8 radi02, shp(3,*), xl(ndm,*)
c Compute element coordinate value
radi02 = 0.d0
do i = 1,nel
radi02 = radi02 + shp(3,i)*xl(1,i)
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 xsj, shj, radi02
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)
if(nint(d(4)).eq.2) xsj = xsj*radi02(shp,xl,ndm,nel)
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
plasticity model leads to derivatives of the yield and loading
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 . . . . . . . . . . . .

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

. . . . 138 C. . . . . . . . . . . . . . . . . . . 142 D. . 152 F. . . . . . . . .2 Example 2 . . . . 144 E Properties for J2 plasticity models 146 E. 122 A Heat Transfer Element B Solid Elements 128 136 B. . . . . . . . . . 138 C. . . . .1 Small displacement element .1 Example: Isotropic behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 F. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Frame elements . . . . . . . . . . 138 C. . . . . . . . . . . . . . . . . . . . . . . .15Basic finite element formulation . . . . . . . . . . 121 13. . . . . . . .1 Conventional Representation . .1 Stress and Strain . . . . . . . . . . . . . . . . . . . . . .1 Displacement elements . . . . . . . . . . . 147 F Matrix Form for Equations of Solids 149 F. . . . . . 136 C Structural Elements C. . 139 D Isoparametric Shape Functions for Elements 140 D. . . . . . . . . . . . .3 Derivatives of Alternative Formulation . . . . . . . . . . . 153 . . . . .16Mixed formulation . . . . . . . . . . . . . . . . . . . . . . . . . 147 E. .2 Alternative Representation in Two Dimensions .2 Split into Deviatoric and Spherical Components . . . . . . . . . . . . . . . . . . . . . . .4 Shell elements . .2. .1 Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .CONTENTS iv 13. . . . . . . . . . . .3. . . . . . . . . 140 D. . . . . . . . . . . . 150 F. . . . . .1 Truss elements 138 . . . . . . . . . . 139 C. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Plate elements . . . . . . . . . . . . . . . . . .3 Linear Elastic Constitutive Equations . . . . . . .

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. Chapters 7 and 8 then discuss alternative mixed methods for treating problems which include constraints leading to near incompressibility. whereas. Chapters 2 and 3 provide an introduction to problem formulation in both a strong and a weak form. The linear heat equation is used as an example problem to describe some of the details concerning use of strong and weak forms. 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]. Chapters 4 and 5 provides a summary of the linear elasticity problem in its strong and weak forms. General mixed and enhanced strain methods are presented as alternatives to develop low order finite elements that perform well at the nearly incompressible regime. Chapter 6 discusses implentation for displacement (irreducible) based finite element methods.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. For applications involving an isotropic model and strong 1 . Vainberg’s theorem is introduced to indicate when a variational theorem exists for a given variational equation. This is an essential feature required to handle both inelastic and non-linear constitutive models. The strong form of a problem is given as a set of partial differential equations. A variational statement provides a convenient basis for constructing the finite element model. In this report. Special attention is given to methods which can handle anisotropic elastic models where the elasticity tangent matrix is fully populated. the weak form of a problem is associated with either variational equations or variational theorems. Chapter 9 presents a generalization of the linear elastic constitutive model to that for linear viscoelasticity.

The Newmark method and some of its variants (e. Chapter 11 discusses methods used in FEAP to solve constraints included in a finite element model. The latter provides a basis for constructing an accurate time integration method which is employed in the FEAP system. A discussion is presented for both rate and rate independent models. Such constraints are evident in going to the fully incompressible case. The simplest approach is use of a penalty approach to embed the constraint without the introduction of additional parameters in the algebraic problem. It is shown that general elements which closely follow the representations used for the small deformation case can be developed using displacement. Chapter 10 presents the general algorithm employed in the FEAP system to model plasticity type presentations. as well as. Finally. 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. A final option is the use of Lagrange multipliers to include the constraint.. INTRODUCTION 2 deviatoric relaxation compared to the spherical 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. All of these methods are used as part of the FEAP system. 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. mixed.CHAPTER 1. Chapter 13 presents a summary for extending the methods discussed in the first twelve chapters to the finite deformation problem. for the problem of intermittant contact between contiguous bodies. Chapter 12 presents a discussion for extension of problems to the fully transient case. as well as.g. 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. and enhanced strain methods. Alternative representations for linear viscoelastic behavior are presented in the form of differential models and integral equations. an energy-momentum conserving method) are discussed as methods to solve the transient algorithm by a discrete time stepping method. .

y) ∂x2 ∂y 2 (2. This differential equation may be solved by writing u as a product form u= m n sin( nπy mπx ) sin( )umn a b (2. in general it is not possible to treat general boundary conditions or problem shapes using this approach. 0 ≤ y ≤ b with the boundary condition u = 0 on all edges. 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. Very few partial differential equations may be solved in closed form . The set of partial differential equations describing such problems is often referred to as the strong form of the problem.2) which when substituted into the equation yields 3 . The differential equations may be either linear or non-linear.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. whereas.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). As an example consider the Poisson equation ∂2u ∂2u + = q(x.1) defined on the region 0 ≤ x ≤ a. non-linear equations include nonlinear terms also.

however. it is not possible to get an exact solution in closed form.in essence this is the result for fewer terms in the series. Ω. use of a finite set of terms leads to an approxiamte solution with the accuracy depending on the number of terms used. This is similar to a series solution in that each mesh used to construct an FE solution represents a particular number of terms. Integrate the result of 1. 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. y) a b (2. Evaluation of the solution requires the summation of the series for each point (x. Indeed. Indeed. again. More general solutions may be constructed using separable solution. 4. Multiply the differential equation by an arbitrary function which contracts the equations to a scalar. 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. Integrate by parts using Green’s theorem to reduce derivatives to their minimum order.CHAPTER 2.e. Consequently. 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 . The weak form will be the basis for constructing our finite element solutions. In the sequel. 2. 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. Replace the boundary conditions by an appropriate construction. we now direct our attention to rewriting the set of equations in a form we call the weak form of the problem.. over the domain of consideration. 3. INTRODUCTION TO STRONG AND WEAK FORMS 4 m n mπ a 2 + nπ b 2 sin( mπx nπy ) sin( )umn = q(x. the solutions are obtained only in series form. y) of interest. . Fourier series) and matching terms between the left and right sides.3) The solution may now be completed by expanding the right hand side as a double sine series (i. Accordingly.

and q n = qi n i = qn ¯ (2. c is specific heat. Partial derivatives in space will be denoted by ( · ).9) .i = i=1 ∂qi ∂xi (2. ΓT . Q is the volumetric heat generation per unit volume per unit time. The equations hold for all points xi in the domain of interest.. qi is the component of the heat flux in the xi direction. Ω.4) where: d is the spatial dimension of the problem. summation convention is used where d ∂( · ) ∂xi (2. t) = T ¯ where T is a specified temperature for points xj on the boundary. The following notation is introduced for use throughout this report.i = and in time by ∂T ˙ T = ∂t In addition. the divergence of the flux may be written as d qi. The governing partial differential equation set for the transient heat conduction equation is given by d − i=1 ∂qi ∂T + Q = ρc ∂xi ∂t (2.5) (2.10) (2. ρ is density.7) With this notation.CHAPTER 2.6) ai b i = i=1 ai b i (2. and t is time.3 Heat conduction problem: Strong form The above steps are made more concrete by considering an example. T is temperature. INTRODUCTION TO STRONG AND WEAK FORMS 5 2.8) Boundary conditions are given by ¯ T (xj .

16) We note that it is necessary to compute second derivatives of the temperature to compute a solution to the differential equation. 0) = T0 (xi ) (2. at time zero. The partial differential equation together with the boundary and initial conditions is called the strong form of the problem. The Fourier law is a linear relationship given as qi = − kij T. The equations are completed by giving a relationship between the gradient of temperature and the heat flux (called the thermal constitutive equation).CHAPTER 2.i ). Hence for an isotropic material the Fourier law becomes qi = − kT. we show that.14 into Eq.j (2. In the following.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. Γq . Thus.11) for points in the domain. = 0 for i = j). expressed as a weak form. it is only necessary to approximate first derivatives of functions to obtain a solution.12) where kij is a symmetric. For an isotropic material kij = kδij (2. second rank thermal conductivity tensor.13) in which δij is the Kronecker delta function (δij = 1 for i = j. Ω.4. 2. . INTRODUCTION TO STRONG AND WEAK FORMS 6 where q nn is a specified flux for points xj on the flux boundary. the solution process is simplified by considering weak (variational) forms. Initial conditions are given by ¯ T (xi . The result is ˙ (kT. and ni are direction ¯ cosines of the unit outward pointing normal to the boundary.14) The differential equation may be expressed in terms of temperature by substituting Eq.i + Q = ρcT (2. 2.i (2.

23) which we observe is an integration by parts.18) In step 3 we integrate by parts the terms involving the spatial derivatives (i.i dΩ = 0 (2. The equation is first written on one side of an equal sign.21) The left hand side expands to give [U V.24) + Γ W qi ni dΓ = 0 .i dΩ = Ω Γ φ ni d Γ (2.i V dΩ + Γ (U V )ni dΓ (2. Thus.20) (U V )ni dΓ (2. qi . T ) = Ω ˙ W (xi ) ρcT − Q dΩ − Ω W.CHAPTER 2. we multiply Eq. which transforms the set of differential equations onto a scalar function.i In step 2 we integrate over the domain.4 Heat conduction problem: Weak form In step 1. Green’s theorem is given by φ.i + U.4 by an arbitrary function W (xi ). qi .i V ] dΩ = Ω Γ (U V )ni dΓ (2.e.i dΩ = − Ω Ω U. Thus for φ = VU we have (U V ).i dΩ = Ω Γ (2.22) which may be rearranged as U V.17) ˙ W (xi ) ρcT − Q + qi. T ) = W (xi ) ρcT − Q + qi. qi .. the thermal flux vector in our case). T ) = Ω = 0 (2. Ω. INTRODUCTION TO STRONG AND WEAK FORMS 7 2. φ is the product of two functions. 2. G(W.19) Normally. Thus ˙ g(W.i qi d Ω (2. Applying the integration by parts to the heat equation gives G(W.

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.28) . we define each integral as a sum of integrals over each element. Accordingly.i k T. thus. Eq. 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.CHAPTER 2. hence our equation must be valid even if W is zero for some parts of the domain).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.24 completes step 4 and we obtain the final expression G(W.i dΩ (2. we let Nel Ω ≈ Ωh = e=1 Ωe (2.25) Now the boundary condition Eq. there are no additional equations that can be used to give any additional reductions. This leads to weaker conditions to define solutions of the problem and thus the notion of a weak form is established. 2.27 is said to be irreducible [26. Substituting all the above into Eq. 2. T ) = Ω ˙ W (xi ) ρcT − Q dΩ − Ω W. 2.5 Approximate solutions: The finite element method For finite element approximate solutions. Furthermore. 2.i qi dΩ (2. Chapter 9]. qi .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. INTRODUCTION TO STRONG AND WEAK FORMS Introducing qn .

defines a class called a C 0 function. Ωe is the domain of a typical element and Nel is the number of nodes attached to the element. 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. which maintain the C 0 condition. Integrals may now be summed over each element and written as Nel (·) dΩ ≈ Ω Ωh (·) dΩ = e=1 Ωe (·) dΩ (2. Standard element interpolation functions which maintain C 0 continuity are discussed in any standard book on the finite element method (e.29) Thus our heat equation integral becomes Nel Nel G ≈ Gh = e=1 Nel Ωe W ˙ ρcT − Q dΩ − e=1 Ωe W. INTRODUCTION TO STRONG AND WEAK FORMS 9 where Ωh is the approximation to the domain created by the set of elements. The case where only the function is continuous.i qi dΩ (2.31) + e=1 Γeq W qn dΓ = 0 ¯ In order for the above integrals to be well defined.CHAPTER 2.i kT.32) for coordinates and Nel T = I=1 NI (ξ)T I (t) (2. See [26. Chapter 7]). satisfy the conditions Nel xi = I=1 NI (ξ)xI i (2. This occurs under the condition that both W and T are continuous in Ω. Commonly. the finite element method uses isoparametric elements to construct C 0 functions in Ωh .g. the first derivatives of W and T may be discontinuous in Ω. With this approximation. Isoparametric elements.33) . surface integrals between adjacent elements must vanish.i dΩ (2..

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

The routine uses basic features included in the FEAP system to generate shape functions. hence it needs to compute a residual for the equations (see FEAP User and Programmer Manual for details). FEAP is a general non-linear finite element solution system. is included separately to permit solution of the general linear eigenproblem KΦ = MΦΛ (2.i (ξ l )j(ξ l )wl (2. 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. where for example L KIJ = l=1 NI. 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. perform numerical quadrature.i (ξ l ) k NJ.58) where j(ξ) is the determinant of J evaluated at the quadrature point ξ l and wl are quadrature weights.57) The above integrals are normally computed using numerical quadrature. The sides of the block are assumed to also be fully insulated. The listing included in Appendix A summarizes an element for the linear heat transfer problem. Both steady state and transient solutions are permitted. INTRODUCTION TO STRONG AND WEAK FORMS 13 FI = Ωe NI Q dΩ − Γeq NI qn dΓ ¯ (2. M.60) (2. etc.61) which can be used to assess the values of basic time parameters in a problem.59) Each array is computed for a single element as described in the section of the FEAP Programmer Manual on adding an element. . The heat capacity array.CHAPTER 2.1. 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.

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.5 are incorporated into FEAP as a user element and the steady state solution computed.2. The contour of temperatures is shown in Figure 2. . The routines indicated in Tables A.1 to A.

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

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.3) .Chapter 3 Introduction to Variational Theorems 3.i k T. a function of functions). 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].e.. it is necessary to introduce a scalar parameter which may be used as the limiting parameter in the derivative [10]. Introducing the family of functions T η into the functional we obtain. using the steady state heat equation as an example. 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. the result Gη = G(W. dG Gη − G0 = lim η→0 dη η 16 (3. In order to construct a derivative of a functional.2) + Γq W qn dΓ ¯ The derivative of the functional with respect to η now may be constructed using conventional methods of calculus.1) The function τ is an arbitrary function and is related to the arbitrary function W introduced in the construction of the weak form.i dΩ − Ω W Q dΩ (3. Thus. T η ) = Ω η W.

INTRODUCTION TO VARIATIONAL THEOREMS 17 where G0 is the value of Gη for η equal to 0.CHAPTER 3. To investigate symmetry of a functional we consider only terms which include both the dependent variable and the arbitrary function. Variational theorems are quite common for several problem classes.i d = (T.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. often we may only have a .i ) = τ. the derivative of the functional with respect to η is given by dG = dη W. τ ) = A(τ.8) Note that use of Eq. 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 . A variational theorem. (3. has a first variation which is identical to the weak form.6) 3.i dΩ Ω (3.4) dη dη With this result in hand. given by a functional Π(T ).5) The limit of the derivative as η goes to zero is called the variation of the functional. 3. τ ) dη This is a notation commonly used to define inner products. We shall define the derivative of the functional representing the weak form of a differential equation as dG = A(W. For the linear steady state heat equation the derivative with respect to η is constant.1 leads to a result where τ replaces W in the weak form.i + ητ.2 Symmetry of inner products Symmetry of inner product relations is fundamental to the derivation of variational theorems. Thus. given a functional Π(T ) we can construct G(W. T ) η→0 dη lim (3. W ) (3. Thus. The construction of the derivative of the functional requires the computation of variations of derivatives of T . 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. hence the derivative is a variation of G. An inner product is symmetric if A(W. Using the above definition we obtain d(T η ). however.i (3.i k τ. T ) as dΠ(T η ) = G(τ.

For the steady state heat equation we have G(T.g.10) Π(T ) = 0 G(T.14) . etc. when set to zero.12) and a problem commonly referred to as a variational theorem. but knowledge that it exists is helpful since it implies properties of the discrete problem which are beneficial (e. ηT ) 3. A variational theorem is a functional whose first variation. τ ). otherwise.13) The integral is trivial and gives Π(T ) = 1 2 T.i dΩ − Ω Ω T QdΩ + Γq T qn dΓ ¯ (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. symmetry of the tangent matrices. ηT )dη (3. 2.8. Also. 3.i k η T. t) to define G(T. Check symmetry of the functional A(W. Perform the following substitutions in G(W. t) T (x. If symmetric then to to 2.11) Performing the variation of Π and setting to zero gives dΠ(T η ) = G(τ. T ) = 0 η→0 dη lim (3.9) (3. The result of the above process gives 1 (3. Integrate the functional result from (b) with respect to η over the interval 0 to 1.i kT. The construction of a variational theorem from a weak form is performed as follows [24]: 1. T ) W (x) → T (x. minimum or stationary value. yields the governing differential equations and boundary conditions associated with some problem..CHAPTER 3.i dΩ − Ω T Q dΩ + Γq T qn dΓ ¯ (3. t) → ηT (x. existence of a variational theorem yields a weak form directly by using Eq. ηT ) = Ω T. stop: no varitational theorem exists.).

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 rate term which includes both T and τ becomes T .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 . ητ ∆t τ .12.15) and performing the derivative defined by Eq.19) Letting tn+1 − tn = ∆t and omitting the subscripts for quantities evaluated at tn+1 .i dΩ (3. The first variation is defined by replacing T by T η = T + ητ (3.CHAPTER 3.i k T.16) If the second variation is strictly positive (i. 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. the variational theorem is called a minimum principle and the discrete tangent matrix is positive definite. The second variation of the theorem generates the inner product A(τ.e. ητ = (η τ . the first variation of the variational theorem generates a variational equation which is the weak form of the partial differential equation. INTRODUCTION TO VARIATIONAL THEOREMS 19 Reversing the process. T ) ˙ (3.18) If however. The transient heat equation with weak form given by G = Ω W ˙ ρ c T − Q dΩ + Ω W.T ∆t A= = η (3.20) (3. we can often restore symmetry to the functional and then deduce a variational theorem for the discrete problem.21) . 3. we first discretize the transient term using some time integration method.. τ ) (3. A is positive for all τ ).

e. applying this construction can be formally performed using usual constructions for a derivative of a function. INTRODUCTION TO VARIATIONAL THEOREMS 20 since scalars can be moved from either term without affecting the value of the term. u. 3. A = (T. However.i k T. δu is the variation of the variable (i.i dΩ − Ω ∂T.23) ∂xi where xi are the set of independent variables.i are the dependent variables of the functional. we obtain the result δΠ = + 1 2 ∂ (T.CHAPTER 3.. etc. This formal construction is easy to apply but masks the meaning of a variation.i k T.i where u.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. 3.2 with δT replacing W . 3.i ∂ (T qn ) δT dΓ ¯ Γq ∂T Ω ∂ (T Q) δT dΩ ∂T (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.24) ∂u ∂u. Similarly.22) 3. We shall address this aspect at a later time. That is. .i + T. it is formally the ητ (x)). The differential of a function f (xi ) may be written as ∂f df = dxi (3. η τ ) = (η T.i k) δT. we may formally write a first variation as ∂Π ∂Π δΠ = δu + δu. and δΠ is called the first variation of the functional.14. We may also use the above process to perform linearizations of variational equations in order to construct solution processes based on Newton’s method. τ ) (3.i ) δT.25) Performing the derivatives leads to δΠ = 1 2 (k T.i dΩ − Ω Ω Q δT dΩ + Γq qn δT dΓ ¯ (3.i + · · · (3.i dΩ − Ω Q δT dΩ + Γq qn δT dΓ ¯ (3. For the functional Eq.3).27) which is identical to Eq.26) Collecting terms we have δΠ = Ω δT.

(4. bm is the body force per unit mass. The dependent variables are given in terms of the displacement vector. and u is the acceleration.2) u. . The basic governing equations are: 1. which leads to symmetry of the stress tensor σ = σT 3. (4. the stress tensor. x. ¨ operator.3) u + (a) u u = 1 2 21 u + ( u)T (4. u. where the domain of analysis is Ω with boundary Γ. The equations are presented using direct notation. For a presentation using indicial notation see [26. Chapter 6]. and the strain tensor. The presentation below assumes small (infinitesimal) deformations and general three dimensional behavior in a Cartesian coordinate system. Deformation measures based upon the gradient of the displacement vector.Chapter 4 Small Deformation: Linear Elasticity A summary of the governing equations for linear elasticity is given below. σ. Balance of angular momentum. Balance of linear momentum expressed as ¨ · σ + ρ bm = ρ u where ρ is the mass density.1) is the gradient 2.4) . which may be split as follows u = where the symmetric part is (s) (s) (4.

8) σ31 σ32 σ33 which from the balance of angular momentum must be symmetric. However.9) components ordered as (with no sym u1.1 u → u3.1 (4.3 u2.   ω13 0 ω12  = −ω12 ω23 0 ω33 −ω13 −ω23 (4.5) Based upon this split.2 u1.3 (4.7) In a three dimensional setting the above tensors have 9 components.10) The strain tensor is the symmetric part with components   11 12 22 32 13 23 →  21 31 33  (4.14) . Accordingly.2 u3.6) and the skew symmetric part defines the spin. or small rotation.2 u2.thus.12) The spin tensor is skew symmetric.13)  ω13 ω23  0 (4. ω = (a) u (4.3  u3. 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   σ11 σ12 σ13 σ → σ21 σ22 σ23  (4.1 u2. ωij = ωji which implies ω11 = ω22 = ω33  ω11 ω12 ω21 ω22 ω → ω31 ω32 = 0.CHAPTER 4. the symmetric part defines the strain = (s) u (4.11) and the symmetry condition ij = ji (4. SMALL DEFORMATION: LINEAR ELASTICITY and the skew symmetric part is (a) 22 u = 1 2 u − ( u)T (4. hence σij = σji The gradient of the displacement has the metries)  u1.

Γ.15) Static or quasi-static problems are considered by omitting the acceleration term from the momentum equation (Eq.16) (4. is split into two parts: • Specified displacements on the part Γu . the body force was specified per unit of mass.CHAPTER 4. and • specified tractions on the part Γt . given as: ¯ u = u ¯ where u is a specified quantity. We shall consider several alternatives for splits during the course.17) (4. however. For example. = m + 0 (4. we begin by considering a linear elastic material with an additional known strain.1). unit weight/volume) using ρ bm = bv (4. In small deformation analysis the strain is expressed as an additive sum of parts. 0) = v0 (x) where d0 is the initial displacement field. 0) = d0 (x) (4.18) ˙ u(x. and v0 is the initial velocity field. SMALL DEFORMATION: LINEAR ELASTICITY 23 The basic equations which are independent of material constitution are completed by specifying the boundary conditions.e. 0 as a thermal strain is given by 0 = th = α(T − T0 ) (4. Accordingly. Inclusion of intertial forces requires the specification of the initial conditions u(x. t In the balance of momentum. given as: t = σn = ¯ t where ¯ is a specified quantity. 0 is a second part which we assume is a specified strain.19) 4. 4.20) where m is the strain caused by stresses and is called the mechanical part. This may be converted to a body force per unit volume (i..21) .1 Constitutive Equations for Linear Elasticity The linear theory is completed by specifying the constitutive behavior for the material. (4. For this purpose the boundary.

(4. C1233 → D43 .1 Thus. W ( computed as ∂W σab = ∂ m ab the elastic modulus matrix is symmetric and satisfies Dij = Dji ).27) Using tensor quantities. from which stresses are (4.23) the matrix of strains is ordered as the vector (note factors of 2 are used to make shearing components the engineering strains.26) (4.28) 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. The constitutive equations relating stress to mechanical strain may be written (in matrix notation. using this table.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. etc.22) where the matrix of stresses is ordered as the vector σ = σ11 σ22 σ33 σ12 σ23 σ31 T (4. We next . γij ) = 11 22 33 2 12 2 23 2 T 31 (4.CHAPTER 4. we have C1111 → D11 .LP where T is temperature and T0 is a stress free temperature.25) Assuming the existence of a strain energy density. the constitutive equation for linear elasticity is written in indicial notation as: σab = Cabcd ( cd − 0 ) (4. SMALL DEFORMATION: LINEAR ELASTICITY 24 .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.

CHAPTER 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. 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. For the present. . we assume that inertial forces may be ignored. we can add the inertial effects and use time discrete methods to restore symmetry to the formulation. Later.

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

To accomplish the grouping it is necessary to integrate by parts the term involving (s) U.6) The first variation is then defined as the derivative of I η with respect to η and evaluated at η = 0. E).23 for each of the variables.9) . σT Ω (s) UdΩ = − Ω UT · σdΩ + Γt tT UdΓ + Γu tT UdΓ (5. VARIATIONAL THEOREMS: LINEAR ELASTICITY ση = σ + η S η 27 (5. the U. σ η .7 is equivalent to the equations for linear elasticity.CHAPTER 5. Accordingly.4) (5. 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. we need to group all the terms together which multiply each variation function (e.8) and the two forms lead to identical results. In order to show that the theorem in form 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 − (5.5) = + ηE and define the single parameter functional I η = I(uη .. η ) (5.7) The first variation may also be constucted using 3.g. S.

u. which must then be positive. Suppose that an expression does not vanish at a point. The Hu-Washizu variational principle will serve as the basis for most of what we need in the course. σ. 5. The expression which multiplies each variation function is called an Euler equation of the variational theorem.11 is not possible in general. The strains are eliminated by developing an expression in terms of the stresses. For linear elasticity this leads to = 0 + D−1 σ (5. On the other . the variations multiply the constitutive equation.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.2 Hellinger-Reissner Variational Theorem The Hellinger-Reissner principle eliminates the strain as a primary dependent variable. the balance of linear momentum.CHAPTER 5. the traction boundary condition. consequently. then.11) The need to develop an expression for strains in terms of stresses limits the application of the Hellinger-Reissner principle. 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. Indeed. This leads to a contradiction. and the stress. For example. and the displacement boundary condition. There are other variational principles which can be deduced directly from the principle. The lemma is easy to prove. only the displacement. Two of these. and hence the integral will not be zero. remain as arguments in the functional for which variations are constructed. This results in the integral of the square of a function. since the variation is arbitrary. and thus the only possibility is that the assumption of a non-vanishing expression is false. 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. For the Hu-Washizu theorem. we can assume that it is equal to the value of the non-vanishing expression. in finite deformation elasticity the development of a relation similar to 5. the strain-displacement equation. the only equation not contained is the balance of angular momentum.

12 may be omitted since its first variation is zero.15 and 4.g.. we shall on occasion use the principle in our studies.15) are satisfied at each point of Ω and Γ. 5. linear elastic incompressible problems.16. 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 (5.14 and 5.11 to 5. Accordingly.CHAPTER 5. . In addition. in addition to 5.6 or comparing 5. Thus.17) .12) − Γt u tdΓ − Γu T¯ ¯ t (u − u)dΓ The Euler equations for this principle are (s) u = 0 + D−1 σ (5.13) together with 4. The strain-displacement equations are deduced by either directly stating 4. Thus.13. 4. respectively. the displacement boundary conditions are assumed to be imposed as a constraint on the principle.11 into the Hu-Washizu principle leads to the result I(u.1. 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. thin plates as a limit case of the thick Mindlin-Reissner theory). σ. Introducing 5.15 the relation σ = D( − 0 ) (5.3 Minimum Potential Energy Theorem The principle of minimum potential energy eliminates both the stress.14) and ¯ u = u (5. the constitutive equation must be given. The MPE theorem may be deduced by assuming = (s) u (5. as arguments of the functional.16) uT bv dΩ − Γt uT ¯ tdΓ Since stress does not appear explicitly in the theorem. VARIATIONAL THEOREMS: LINEAR ELASTICITY 29 hand. and the strain. The first term in 5. the Hellinger-Reissner principle is an important limiting case when considering problems with constraints (e.

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

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

6. we express ¯ ¯ u = Nα (ξ) uα (t) (6.6) We then will assume the integral over Γu is satisfied and may be omitted. of the term Πσ = Ωe ( (s) u)T σ dΩ = e Ωe ( (s) u)T σ dΩ (6. That is.3 in Eq. bv . This step is not necessary but is common in most applications.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 (6.9) 6. The remaining terms involving specified applied loads are due to the body forces.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.2 Internal Force Computation The stress divergence term in the Hu-Washizu variational principle is generated from the variation with respect to the displacements. Πf = Ωe uT bv dΩ + Γte uT ¯ dΓ t (6.7 yields Πf = (uα )T Ωe Nα bv dΩ + Γte Nα ¯ dΓ t = (uα )T Fα (6. ¯ The terms in the variational principal are t.5) and set ¯ ¯ uα (t) = u(xα . 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.10) . u.7) Using Eq. the displacement at each point in an element may be computed.CHAPTER 6. 6.4) where xα are the cartesian coordinates of nodes. DISPLACEMENT METHODS 32 x(ξ) = Nα (ξ) xα (6. 6. t) (6. and the applied surface tractions.

13) the strain-displacement matrix is expressed  Nα.2 0 (u1.1 ) (u2.2   =   0 0  Nα.1 0  0 Nα.20) .3   Nα.2 u3.1 (6.2 + u2.CHAPTER 6.2 0   0 Nα.15) = 11 22 33 2 12 (6.2 u1 /x1 (u1. for a 2-dimensional axisymmetric problem (with no torsional loading) the strains are T = (6.2 Nα. DISPLACEMENT METHODS 33 Using the finite element approximation for displacement.18) Finally.1 )   Nα.1 u2.3 + u3.19) 11 22 33 2 12 and are expressed in terms of the displacements as T = u1.2 + u2.2  0 Nα.2 ) (u3.12) and related to the displacement derivatives by T = u1.i = T (6.1 + u1.3 where (6.11) where Bα 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 (6.14) ∂Nα ∂xi For a 2-dimensional plane strain problem the non-zero strains reduce to Nα.2 + u2.1  0   0 Bα =  Nα.16) and are expressed in terms of the displacement derivatives as T = u1. Bα becomes: Bα (6. 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 (6.17) thus.2   0 Nα.1 u2.3 Nα.3 ) as:  0 0 Nα.3 (u1.1 ) (6.1 u2.1 (6.1 0   Nα.

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.2   Bα =  Nα /x1 0  Nα. This force is expressed by Pα (σ) = Ωe (Bα )T σ dΩ (6.1 0  0 Nα.26) The deviatoric part of stress . becomes:   Nα. x2 now denote the axisymmetric coordinates r.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 . Locking is generally associated with poor performance at or near the incompressible limit.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. respectively1 The stress divergence term for each element may be written as Πσe = (uα )T Ωe 34 (6. 6.23) which gives Πσe = (uα )T Pα (σ) (6. uα . The stress may be written in terms of the deviatoric and pressure parts (pressure is spherical part) as σ = s + p1 (6.22) In the sequel we define the variation of this term with respect to the nodal displacements.25) p = tr(σ) = σkk 3 3 For infinitesimal strains the spherical part is the volume change defined by θ = tr( ) = kk (6.2 Nα.x2 plane.21) (Bα )T σ dΩ (6. DISPLACEMENT METHODS The strain-displacement matrix for axisymmetry. For stress the spherical part is the mean stress defined by 1 1 (6. is defined so that its trace is zero. To study the locking problem we split the formulation into deviatoric and volumetric terms.CHAPTER 6. the internal stress divergence force.1 where x1 . . Bα . s. z.

37) .38) In the above m is any positive integer power. DISPLACEMENT METHODS 35 where.36) (6. Similarly. however.35) (6. 1 is the rank two identity tensor.39) (6.32) as the deviatoric projector. we can operate on the strain to define its deviatoric and volumetric parts.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. which in matrix notation is given by the vector mT = 1 1 1 0 0 0 (6. For subsequent developments. Utilizing the above properties. Accordingly. I is a 6 × 6 identity matrix. We note.31) and hence mT s = 0 as required.29) thus. We note that the trace of the stress gives mT σ = 3 p = mT s + p mT m = mT s + 3 p (6.33) (6.30) where. we define Idev = I − 1 m mT 3 (6.28) In matrix form the pressure is given by p = 1 T m σ 3 (6. the deviatoric part of stresses now may be computed as s = σ − 1 m mT σ = 3 I − 1 m mT 3 σ (6. the volumetric projector is defined by Ivol = 1 m mT 3 (6. in three dimensions.CHAPTER 6. that inverses to the projectors do not exist. the deviatoric and volumetric parts are given by = e + 1 θ1 3 (6.

In 2dimensional plane problems the volumetric strain-displacement matrix is given by bα = Nα.1    0 3 Nα.1 Nα.40) we obtain e = Idev .2 Nα.1 1 −Nα. mT e = 0 (6.2   −Nα.1 −Nα.1 + Nα /x1 Nα.2 2 Nα.2 −Nα.45) 1 mb 3 (6. Accordingly.3 3 Nα.3    1 −Nα.44) (6.2 2 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 .1  = 3 −Nα.3 0 3 Nα.2 −Nα.3    (6.48) The deviatoric matrix Bdev is constructed from Eq. 6.1 (6. DISPLACEMENT METHODS 36 where e is the strain deviator and θ is the change in volume.42) is the volumetric strain-displacement matrix for a node α in its basic form.1 −Nα.2  −Nα.43) (6.47) and for 2-dimensional axisymmetric problems bα = Nα. Using matrix notation we have θ = mT (6.2 3 Nα.2  3 Nα.3 −Nα.2 (6.1 2 Nα.CHAPTER 6.50) .49) Bdev = 0  3  3 Nα.3 (6.1 Nα.1 3 Nα.2 (6. For 3-dimensional problems bα = Nα.46) mT Bdev = 0 (6.39 and yields for the 3-dimensional problem   2 Nα.41) The strain-displacement matrix also may now be written as a deviatoric and volumetric form.1 and for the 2-dimensional plane problem  Bdev 2 Nα.2  3 Nα.

1 ) −Nα.52) which after rearrangement gives Pα = Ωe BT s dΩ + α Ωe BT m p dΩ α (6.2 1 −(Nα.2    (6.CHAPTER 6. Pα = Ωe BT σ dΩ = α Ωe BT (s + p m) dΩ α (6. 6.51) Bdev = 3  (2 Nα /x1 + Nα.1 6. the internal force is composed of the sum of deviatoric and volumetric parts.2 3 Nα. then B = Bdev + Bvol = Bdev + Pα = Ωe (BT )α s dΩ + dev Ωe bT p dΩ α (6.4 Internal Force .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 (6.54) 3 and use the properties defined above for products of the deviatoric and volumetric terms.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. the deviatoric matrix for the 2-dimensional axisymmetric problem is given by:   (2 Nα.56) Thus.2  3 Nα.1 + Nα /x1 ) 2 Nα.57) . Accordingly.53) If we introduce 1 mb (6. DISPLACEMENT METHODS 37 Finally.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 − Nα /x1 ) −Nα.

ν.58) For different values of ν. 0 ≤ λ ≤ ∞ (6. Another parameter which is related to λ and µ is the bulk modulus. that is cijkl = cklij = cjikl = cijlk (6. K.64) We note that the above definition for the moduli satisfies all the necessary symmetry conditions.62) (6. and e Poisson’s ratio. and λ is a parameter which causes difficulties since it is infinite. which is defined by 0 ≤ ν ≤ K = λ + E 2 µ = 3 3 (1 − 2 ν) (6.61) 1 We note that K also tends to infinity as ν approaches 2 . the Lam´ parameters have the following ranges e 0 ≤ ν ≤ and 1 2 .65) The relations may be transformed to matrix (Voigt) notation following Table 4.63) where cijkl are the elastic moduli. For an isotropic material the elastic moduli are then related by cijkl = λ δij δkl + µ (δik δjl + δil δjk ) (6.1 and expressed as σ = D (6.CHAPTER 6. DISPLACEMENT METHODS 38 where λ and µ are the Lam´ parameters which are related to Young’s modulus. µ = E 2 (1 + ν) (6. E.59) 1 E E . 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. by λ = νE (1 + ν)(1 − 2 ν) . ≥ µ ≥ (6.66) where the elastic moduli are split into D = λ Dλ + µ Dµ (6.67) .60) 2 2 3 1 For an incompressible material ν is 2 .

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.CHAPTER 6.. the computation of the element stiffness matrix for a displacement approach is given and is based upon the above representations for the moduli. γij = 2 ij ).73) Once Ddev has been computed it may be noted that Idev Ddev = Ddev Idev = Ddev Ivol Ddev = Ddev Ivol = 0 and.70) (6.72) Dµ Idev = Idev Dµ (6.75) .e. 2 (6. it is a deviatoric quantity. thus. In the following section.71) (6. Note that in Dµ the terms multiplying shears have unit values since engineering shear strains are used (i.74) (6. 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 (6.68) Dµ (6.69) used as non-dimensional matrices to split the moduli.

79) If we use the properties of the moduli multiplied by the projectors.78) (6.CHAPTER 6.84) (BT )α s dΩ + dev Ωe bT p dΩ α (6.82) (6. the above equations reduce to s = µ Ddev = µDµ e = µDµ (Bdev )β uβ (6. DISPLACEMENT METHODS 40 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 6.81) = Idev (λ Dλ + µ Dµ ) (6.86) .2 that Pα = Ωe (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 ) (6.6 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.83) (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 (6.80) 2 µ) Dλ = K Dλ 3 Thus.77) (6.

92) f (x(ξ)) j(ξ) dξ ≈ l=1 f (x(ξ l )) j(ξ l ) wl (6.87) = Ωe µ (BT )α Dµ (Bdev )β dΩ = dev Ωe µ BT Ddev Bβ dΩ α (6.93) . 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 L 2 ∂x ∂ξ (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.e. number of element degree-of-freedoms less the number of rigid body modes) will be called a standard or full quadrature (or integration). The minimum order quadrature which produces a stiffness with the correct rank (i.90) where 2 denotes integration over the natural coordinates ξ.89) 6. The next lowest order of quadrature is called a reduced quadrature. 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.7 Numerical Integration Generally the computation of integrals for the finite element arrays is performed using numerical integration (i. DISPLACEMENT METHODS 41 and.91) (6. for isotropic linear elasticity. Alternatively. quadrature).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.CHAPTER 6. dξ denotes dξ1 dξ2 in 2-dimensions. thus... use of standard quadrature on one term and reduced quadrature on another leads to a method called selective reduced quadrature.88) K bα bT dΩ = β Ωe K BT Dλ Bβ dΩ α (6.

The evaluation of the shape functions is performed using a shape function subprogram. the basic algorithm to compute the stress divergence term is given by: 1.CHAPTER 6. Pα ← Pα + Aα c 3. Nα.98) . Thus.97) Similar expressions may be deduced for each of the terms defined by the deviatoric/volumetric splits. a shape function routine for 2 dimensions is called shp2d and is accessed by the call (6.94) and for 3-dimensions 1 1 −1 1 g(ξ) dξ = 2 −1 −1 g(ξ) dξ1 dξ2 dξ3 (6. • Accumulate the array. DISPLACEMENT METHODS 42 where ξ l and wl are quadrature points and quadrature weights.g. l • Compute j(ξ l ) wl = c • Compute the matrix in the integrand. e.. Loop over the quadrature points. In FEAP.i . The use of quadrature reduces the development of finite element arrays to an algebraic process involving matrix operations. (e. the determination of Bα requires computation of the derivatives of the shape functions.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. Bα (ξ l )T σ l = Aα ). For example. the integration is generally carried out as a product of one-dimensional Gaussian quadrature.g. respectively. the stress divergence is given by L Pα = l=1 Bα (ξ l )T σ(ξ l ) j(ξ l ) wl (6. Initialize the array Pα 2. For example. For brick elements in three dimensions and quadrilateral elements in two dimensions. for 2-dimensions.. 1 1 g(ξ) dξ = 2 −1 −1 g(ξ) dξ1 dξ2 (6. Additional steps are involved in computing the entries in each array. and computation of σ l requires an evaluation of the constitutive equation at the quadrature point. Repeat step 2 until all quadrature points in element are considered.95) Using quadrature.

ix. swg ) where -number of quadrature points in each direction (input). l The array of points and weights has the following meanings: . if true derivatives returned with respect to ξ. ξ2 ) at quadrature point (input) xl array of nodal coordinates for element (xl(ndm.nen)) (input) shp array of shape functions and derivatives (shp(3.A) is NA.3 The quadrature points may be obtained by a call to int2d: call int2d( l. nel. lint -total number of quadrature points (output). swg -array of natural coordinates and weights (output).1 shp(2. DISPLACEMENT METHODS call shp2d( xi. xl.A) is NA. lint. shp.A) is NA.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. flag) where xi natural coordinate values (ξ1 . xsj. if false derivatives returned with respect to x (input).CHAPTER 6. ndm.2 shp(3. 43 The array of shape functions has the following meanings: shp(1.

L) is ξ1.L swg(2.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. DISPLACEMENT METHODS swg(1. is element elmt01 which is given in Appendix B. An example. .L swg(3.CHAPTER 6.L) is ξ2.

) = + 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. 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. In the work considered here we use the Hu-Washizu variational principle.4) (t) 45 .Chapter 7 Mixed Finite Element Methods 7. displacements appear up to first derivatives.3) (7. Accordingly.2) (7.1) In the principle. σ. 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). while the stresses and strains appear without any derivatives. which we recall may be written as Π(u.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.

Constructing the variation for the above split leads to the following Euler equations which hold in the domain Ω: 1. Strain-Displacement equations e(u) − e = 0 θ(u) − θ = 0 3. thus.CHAPTER 7. ) = + Ω 1 2 µ Ω T Ddev dΩ − Ω µ T Ddev 0 dΩ (7. If.13) Kθ − p = 0 (7.8) 1 Strictly. however.1 The parameters σ α and α are not necessarily nodal values and. Constitutive equations µ Ddev − s = 0 (7. σ.10) (7.12) (7. for the present.7) u) = are the strain-displacement relations for the deviatoric and volumetric parts. the remaining terms may be split into deviatoric and volumetric parts as Π(u. 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. . respectively. this makes the evaluation of integrals over each element more difficult and to date is rarely used. Balance of Momentum · (s + 1 p) + bv = 0 which is also written as div(s + 1 p) + bv = 0 2.11) (7. and the traction and displacement boundary integrals and consider an isotropic linear elastic material.6) (7.9) (7.5) 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. may have no direct physical meaning. we ignore the integral for the body force. φα and ψα need only be piecewise continuous in each element.

the finite element approximation for the mixed formulation may be written as Πe (u. 7. p. p. If the material is isotropic linear elastic. MIXED FINITE ELEMENT METHODS In addition the boundary conditions for Γu and Γt are obtained. 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. Accordingly. Π(u. σ.13 at each point of the domain of an element.16) p [θ(u) − θ] dΩ which may be split into integrals over the elements as Π(u. p.15) for each point of Ω. ) (7. the use of the same functions will permit an exact satisfaction of the constitutive equation. we assume p(ξ) = φα (ξ) pα (t) = φ(ξ) p θ(ξ) = φα (ξ) θα (t) = φ(ξ) θ (7.5 becomes Π(u. 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. σ. 7. θ) = e Πe (u. ) ≈ Πh (u. this part of the problem is given as a displacement model. The variational expression Eq.21) . the constitutive equation may be approximately satisfied. σ. Eq. the Hu-Washizu variational theorem may be approximated by summing the integrals over each element.18) (7. Recall that the strain-displacement equations for a finite element approximation are given by (u) = BI uI (7.20) Thus.19) where it is noted that the same approximating functions are used for both p and theta.17) A mixed approximation may now be used to describe the pressure and the volume change in each element. 47 Using the interpolations described above. For other situations. Accordingly. θ) ≈ Πh (u. ) = e Πe (u. p. θ) (7. p.14) If the deviatoric part is approximated by taking e = e(u) (7.

30) (7. The mixed variational terms become Πe (u. MIXED FINITE ELEMENT METHODS If we define the following matrices: k = Ωe 48 K φT φ dΩ K φT θ0 dΩ Ωe (7.CHAPTER 7.29) (7.24) (7.31) . p. θ) = (uI )T + θT + pT 1 (Kdev )IJ uJ − (P0 )I dev 2 1 k θ − π0 2 g J uJ − h θ (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. 7.22) π0 = h = Ωe (7.23) (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 (7.26) 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. Θ dη 0 −h k θ  0  (Pdev )I  0  − π0 (7.27) where e0 are the deviatoric initial strains.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 ) .

38) This is a mixed patch test requirement. δθ 0 −h k θ  0  (Pdev )I  0  − π0 49 (7. since we used the same functions for the two approximations. The requirement for a solution to exist is that2 nθ ≥ np (7. the matrix h is square and positive definite (provided our approximating functions are linearly independent).CHAPTER 7. See [26.35) (7. we may perform the element solutions by inverting only h.32) We note that the parameters p and θ (and their variations Π and Θ) are associated with a single element. 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 ) . consequently. Chapter 12].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. from the stationarity condition. .37) ¯ ¯ (Kdev )IJ + bT k bJ I ¯ uJ − (P0 )I − bT π 0 dev I (7.33) where nθ and np are the number of parameters associated with the volume change and pressure approximations. the last two rows of the above matrix expression must vanish and may be solved at the element level. consequently. The solution to Eq. Also. 7. We have satisfied this requirement by taking an equal number for the two approximations. by defining a modified volumetric strain-displacement matrix as ¯ bI = h−1 gJ the above simplifies to dΠe = (UI )T dη 2 (7.36) (7.34) Finally. respectively. δp .

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.43) 7.CHAPTER 7. For example. and (2) the functions produce an element which is invariant with respect to the input data. 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. and the number is small compared to the number of degrees-of-freedom on the element.41) where the deviatoric stress is expressed by the displacement approximation as s = µ Ddev (BJ uJ − 0 ) (7. use of φ1 = 1 .40) BT (s + p m) dΩ I (7.44) Provided the approximations for φ are linearly independent. if we show that two functions are sufficient for a 2-dimensional element.46) .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. For example.39) (7. This gives a rank of 1 for the volumetric stiffness. 4-node quadrilateral or 8-node brick elements can use a single function φ1 = 1 (7. φ2 = ξ1 (7.45) for the approximating space. The number of φ functions will affect the rank of the volumetric terms. the rank will normally be that of k. rank(k) ) (7. The modified volumetric stiffness has a rank which is given by ¯ rank(Kvol ) = min ( rank(b).42) and the pressure is expressed by the mixed approximation as p = φ(ξ) h−1 (k θ − π 0 ) (7.

and strains. ¯ 4. B. 2. one being φ1 = 1 . l (a) Compute shape functions: In 2-d problems FEAP uses.52) = [Idev BI (ξ l ) + 1 ¯ m φ(ξ l ) bI ] uI 3 (7. φ2 = ξ1 .53) (b) Compute quadrature stresses and π 0 σl = D ( l − 0 ) (7. h. I.54) . Loop over quadrature points. Several alternatives are possible. Loop over quadrature points. l (7. i(ξ l ) = shp(i. k.51) ¯ bI = h−1 gI (7. Initialize arrays: gI . An algorithm to implement the above mixed model for linear elasticity where D is constant in each element may be summarized as: 1.49) (7. Loop over quadrature points. Invert h and compute bI 5. FEAP will initialize K and the element residual. φ3 = ξ2 (7. l) (b) Compute the volume element times the quadrature weight jl wl = dv(l) 3.50) (7. gI and h.48) φ2 = ξ1 + ξ2 (7. I. l) NI . NI (ξ l ) = shp(3. π 0 .47) The actual functions selected must be subjected to further evaluations to decide which best meets the objectives of the problem solution. another is to use 3 functions with φ1 = 1 . l (a) Compute strain-displacement matrix. l (a) Compute the volumetric strain matrices. 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 .CHAPTER 7.

58) ) + (7. The stress may be split into deviatoric and pressure parts as ¯ σ = ¯ + mp s ¯ (7. ¯ = 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. Kdev (e) Compute the volumetric local tangent. are computed directly from the displacements but the volumetric strain. In this section we construct the form of the functional for an anisotropic linear elastic material.59) .55) (d) Compute the deviatoric tangent. Compute the tangent.57) where D is a symmetric matrix in which there may be coupling between the deviatoric and volumetric strain effects. k 6. e.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. is computed from a mixed form.CHAPTER 7.61) (7. Accordingly.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. we have σ = D[ − 0 ] (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. Accordingly.56) 7. K ¯ ¯ K = Kdev + bT k b (7.62) ) + . It is now assumed that a finite element solution will be constructed in which deviatoric strains. θ.

1. Alternatively. Subsequently. p.CHAPTER 7.16 which was deduced for isotropic materials. For isotropy d is zero.66) (7. 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 (7.19. The added terms in 7.18 and 7. If we introduce finite element interpolations using standard displacement interpolation together with the pressure and volume interpolations given by 7.64) (7.65) (7. adding the terms associated with the mixed volumetric pressure and volume change Vainberg’s theorem may be used to obtain a variational theorem.69 are all associated with d which defines a coupling between deviatoric and volumetric strains.69) This form of the variational principle is equivalent to 7.63) (7.67) − dvol θ0 (7. 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. the first variation of 7.70) ˆ dΩ + δ pT T Ωe −φ φ bT φ I T ˆ dΩ p + δIext . The result is Π(u.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. the stress and strain splits may be substituted into 7.69 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.

plasticity.75) which upon use of the definitions for the mixed pressure.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.77) ¯ ¯ and.62 resulting in δΠe = ˆI δ uT ˆT δθ Ωe BT ¯ Is dΩ φT p ¯ ˆ uJ ˆ θ (7. viscoelasticity.g. in general. ˆ the multiple of δ p yields ˆ φT bJ dΩ uJ = Ωe Ωe ˆ ˆ φT φ dΩ θ = h θ (7. bI . the equation ˆ multiplying δ θ yields the equation φT p dΩ = ¯ Ωe Ωe ˆ ˆ φT φ dΩ p = h p (7.24 and 7. is the stress which is computed from the constitutive equation for each material. p.72) which yields ˆ ¯ ˆ ˆ θ = bI uI = h−1 gI uI (7.CHAPTER 7.70 may be expressed in terms of stresses by substituting the interpolations into 7. and the mixed volumetric ¯ strain displacement equation. MIXED FINITE ELEMENT METHODS 54 The variational equation 7. however.74) Using these results. is not equal to σ.25.73) where h and gI are as defined in 7. similarly.61 and 7.) the effective material moduli are the ones computed by linearizing the constitutive equation expressed in terms of the . The stress σ. Thus. Accordingly. when we later consider other material models (e. yields ˆI δΠσ = δ uT Ω BT [¯ + m p] dΩ I s (7.76) The stress of the mixed method is defined as σ = ¯ + mp s (7. respectively. 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. etc..

Constitution computation for each quadrature point (a) Compute ˆ = BI uI ¯ ˆ θ = φ(ξ) bI uI ¯ = Idev + 1 mθ 3 0 (7.85) (7. MIXED FINITE ELEMENT METHODS 55 ¯ σ stresses.86) ¯ σ = D[ ¯ − ] . ξ1 .74 the dependence of 7. Numerical integration of strain matrices (a) Compute φ = [1.84) (7.83) (7.82) gI = 2. · · · ] (for the 4-node element φ = 1 (b) Compute arrays h = Ωe φT φ dΩ φT bJ dΩ Ωe (7.72 and 7. Mixed volumetric strain displacement matrix ¯ (a) Compute bI = h−1 gI 3.78 on θ and p may be eliminated to give ˆI ¯ ˆ δΠe = δ uT KIJ uJ + δΠ0 + δΠext (7. ξ2 . The tangent matrix may be expressed in terms of the displacements alone by writing the variational equation 7.81) (7.80) The algorithm for the development of a mixed element based upon the above may be summarized as: 1.CHAPTER 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 (7.79) where ¯ ¯ ¯ ¯ ¯ KIJ = (Kdev )IJ + kI bJ + bT kT + bT kvol bJ I J I (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.78) ˆ ˆ Using the solutions to 7.

Residual and Stiffness Integrals (7.e. MIXED FINITE ELEMENT METHODS 1 T ¯ m σ 3 φT p dΩ ¯ Ωe 56 p = ¯ ¯ π = 4.95) .92) (Kdev )IJ = kI = Ωe BT d φ dΩ I φT dvol φ dΩ Ωe kvol = 6. p.91) (7.4 Hu-Washizu Variational Theorem: General Problems The finite element approximation for the mixed formulation of a general linear elastic material (i.93) 7.CHAPTER 7.. θ) = 1 2 + p( T Ωe D dΩ − Ωe T D 0 dΩ (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 (7.90) (7. anistotropic behavior) may be written for a typical element as Πe (u. Stiffness assembly (a) Compute ¯ ¯ ¯ ¯ ¯ KIJ = (Kdev )IJ + kI bJ + bT kT + bT kvol bJ I J I (7.88) ¯ (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. Mixed Pressure ¯ (a) Compute p = φ(ξ) h−1 π 5.89) (7.87) (7.

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. 7.96) + pT φT bJ dΩ uJ − For symmetric D. 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. p. p.97) (7.101) (7.103) The mixed variational terms become Πe (u.CHAPTER 7.104) The first variation of Eq.100) (7. ΠT .105) . MIXED FINITE ELEMENT METHODS which when introduced into the variational theorem gives 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 (7.98) (7.104 may be written in the matrix form    J (Kdev )IJ gI (Kco )I u dΠe T 0 −h   p  = (UI )T .99) (7. ΘT   gJ dη (KT )J −h k θ co  0  (Pdev )I −  0  π0 (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.

θ. MIXED FINITE ELEMENT METHODS 58 Recall that the terms which multiply the variations in pressure. and the variation in the volume change.110) I co I This operation may be performed after all the integrals over the element are evaluated.109) (7.7. 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. as θ = h−1 gJ uJ (7. are associated with individual elements. equations Eq. this requires a computation of two d terms as 1 Dm 3 (7.111) and using these in the remaining equations instead of d (note. . the second row of Eq. dL = DT m (7.108) (7. and. Π. Θ. Essentially.114) dR = D m .107) Thus. when D is symmetric the dR and dL terms are equal). The matrices which involve the elastic moduli may be simplified by defining some reduced terms.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 (7.113) (7.117 must be modified. Accordingly. thus.112) 1 1 T m D m = mT d 9 3 (7. 7.112 through Eq.7.CHAPTER 7. define dvol = Then D Idev = D − d mT or Idev D = D − m dT (7.115) 3 If D is not symmetric. we let3 d = Also.105 may be solved at the element level to give the parameters for the volume change.

119) (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 (7.116) Finally.117) (7.125) which is a 1 × 6 vector.123) (Kco )J = and k = Ωe dvol φT φ dΩ (7.118) (7. MIXED FINITE ELEMENT METHODS which gives 1 1 Idev D m = (D − m dT ) m = d − dvol m = ddev 3 3 59 (7.CHAPTER 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 (7.122) (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.120) Ddev (7.124) The matrix for the initial strains is computed as π0 = Ωe φT dT 0 dΩ (7.

130) which is a column vector (of size 1 × 24).128) Alternatively. For isotropy kvol is the bulk modulus times the element volume. The other matrices in the stiffness are h = h = Ωe dΩ = Ωe (7. In 2-dimensions the element is Q1-P0. for example.133) and gI = Ωe bI dΩ (7. order 0 for the pressure/volume change).1 Example: Interpolations linear for u and constant φ As an example. this matrix is zero. For isotropy.126) where ξiI are the values of the natural coordinates at the I-node. The interpolation for the pressure (and volume change) is constant φ1 = 1 (7.131) k = Ωe which is a 1 × 1 matrix and for constant dvol becomes k = kvol = dvol Ωe (7.127) This element is often called B1-P0 (order 1 interpolations for the brick element. the Q2-P1 element uses quadratic interpolation for displacements (Lagrange interpolations) and linear for the pressure with φ(ξ) = 1 ξ1 ξ2 (7.132) where Ωe is the volume of the element. Higher order elements are also defined. it is possible to use the interpolations φ(ξ) = 1 x1 (ξ) x2 (ξ) (7. MIXED FINITE ELEMENT METHODS 60 7.CHAPTER 7. 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 volume stiffness becomes: dvol dΩ (7.134) .129) The matrices for the B1-P0 (or Q1-P0) element reduce to (Kco )J = Ωe dT BJ dΩ dev (7. for the order 1 quadrilateral.4.

135) dT 0 dΩ (7.137) where the deviatoric part of the stress is computed from the displacement form.136) For the mixed element the internal force is computed using PI = Ωe BT (s(u) + p m) dΩ I (7.CHAPTER 7. and the pressure is computed from the mixed form p = 1 ¯ [(Kco )J + kvol bJ ] uJ − π 0 Ωe (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 (7.138) .

and the enhanced strains. however. the method does not have the deficiencies which are present in the earlier works. σ. Enhanced strains provide great flexibility in designing accurate finite element models for problems which have constraints or other similar types of difficulties. ˜) = (s) u + ˜(ξ) (8.1 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]. (s) u. = + 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. ˜.Chapter 8 Enhanced Strain Mixed Method 8.1) The strain tensor is expressed as an additive sum of the symmetric gradient of the displacement vector. 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. and written as (u. The enhanced strain method is related to earlier works which utilized incompatible displacement modes.2) 62 .

the following Euler equations are obtained for the domain Ω: 1. at the solution. ˜) = − Ω 1 2 ( Ω (s) u + ˜)T D ( 0 (s) u + ˜) dΩ σ T ˜ dΩ Ω ( (s) u + ˜)T D dΩ − (8.10) In addition the boundary conditions for Γu and Γt are obtained.7) After integration by parts of the variation of the displacement gradient term (and also considering the body force term). ENHANCED STRAIN MIXED METHOD 63 If we again ignore the integral for the body force and the traction and displacement boundary integrals. from 8. 8. Balance of momentum div [D ( (s) u + ˜ − 0 )] + bv = 0 (8.2 the remaining terms become Π(u. upon use of Eq. Substitution of this result into the remaining equations yields the appropriate displacement equations of equilibrium and constitutive equation for linear elasticity.10. Constitutive equations D( (s) (8. 8.9) u + ˜ − 0 ) − σ = 0 (8.6) ( Ω (s) U)T D ( (s) (s) u + ˜ − 0 0 ) dΩ ST ˜ dΩ Ω ˜ ET [D ( u + ˜ − ) − σ] dΩ − (8. in an approximate .8 and 8. We note Eq.5) (8.9 implies that. While the enhanced strains vanish pointwise at a solution. σ.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. Strain-displacement equations on the enhanced modes ˜ = 0 3.CHAPTER 8. respectively.4) (8.8) 2. the enhanced strains must vanish.

CHAPTER 8. 8.17) η σ α = σ α + η Sα η and ˜ ˜α = ˜α + η Eα η (8.11) and σ(ξ) = φα (ξ) σ α (t) (8. ˜) = − Ωe 1 2 ( Ωe (s) u + ˜)T D ( 0 (s) u + ˜) dΩ σ T ˜ dΩ Ωe ( (s) u + ˜)T D dΩ − (8.18) . stresses. the strain approximations are now given by (u. σ.19) (8. ) (8. Π(u. σ. as will be shown below.15) The variational expression in each element becomes Πe (u. σ. the Hu-Washizu variational theorem may be approximated by summing the integrals over each element. The choice of appropriate approximating functions φα will be affected by the strain approximation. 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.13) where the approximations for the enhanced strains are assumed as ˜(ξ) = ψ α (ξ) ˜α (t) (8. and are thus u(ξ) = NI (ξ) uI (t) (8. 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. Appropriate interpolations for displacements and stresses are the same as given previously for each element.16) Substituting the approximations for displacements. ˜) = (s) u + ˜(ξ) (8.12) respectively. ) ≈ e Πe (u. In the enhanced strain principle. From Eq. Accordingly. ) = Πh (u.14) It should be noted that different interpolations are introduced for the stress and the enhanced strain terms. displacements appear up to first derivatives. while the stresses and enhanced strains appear without any derivatives. σ. Using the interpolations described above. Accordingly. and enhanced strains and replacing with uI = uI + η UI (8. ENHANCED STRAIN MIXED METHOD 64 scheme based upon the enhanced strain method this is not the case.2.

28. 8. which means that Qαβ = 0 (8.28) which is the solution to be followed here. For a formulation which satisfies Eq. (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. the variational equations in each element reduce to dΠe = dη ˜ (UI )T . (Eα )T ˜ KIJ ΓβI T ˜ ˜ ΓαJ Hαβ uJ ˜β − P0 I ˜α π0 (8. The β are zero.23) (8. A combination of options (a) and (b). 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.27) There are at least three possible ways this may be used: 1. The interpolations for φα are orthogonal to the interpolations ψ β . 2.21) (8.20 is given in each element by Qαβ β = 0 (8. ENHANCED STRAIN MIXED METHOD gives the first variation in each element as dΠe ˜ = (UI )T .22) (8.29) .CHAPTER 8. This is not perfect since we will not obtain a method to compute the σ β directly from the variational formulation. (Eα )T .25) (8.24) (8. 3.20) (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. the variation in each element may be written dΠe dη = η=0 Ωe ( (s) ˜ U)T σ dΩ + Ωe ˜ ˜ ET σ dΩ (8.32) (8. however.31) (8. as well as. the residual in each element becomes: RI = FI − Ωe BT σ dΩ I ˜ (8.33) 8.29 may be solved at the element level giving ˜ ˜β = (Hαβ )−1 ˜ ˜α π 0 − ΓαJ uJ (8. Simo and Rifai suggest using a least square projection technique to obtain the stresses. With the above description. the second of Eq.35) It is noted that the orthogonality condition ST ˜ dΩ = 0 Ωe (8.37) . the stresses which are directly utilized in the variational equation Eq.36) has been incorporated in the above variation. ENHANCED STRAIN MIXED METHOD 66 Since the interpolations for the enhanced strains are assumed for each element independently.34) In subsequent development we shall use these stresses for all calculations of arrays. 8.7 may be deduced as ˜ σ = D( (s) u + ˜ − 0 ) (8.CHAPTER 8. Thus. for outputs and stress projections to nodes.2 Stresses in the Enhanced Method Since the stresses based upon the mixed approximation are no longer available an alternative is needed for computations. 8.

44) j0 −1 ˜ F E(ξ) j(ξ) 0 j0 T F Σ(ξ) j(ξ) 0 (8.3 Construction of Enhanced Modes The construction of the enhanced modes depends crucially on the orthogonality requirement being satisfied for each element. expressed on the bi-unit square to the global strains. it is useful to express the enhanced strains in a similar form. Σ. and gradients of local coordinates. j(ξ). 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. ξ. ˜. E.CHAPTER 8. 8. the constant part of the jacobian matrix. Accordingly.41) (8. ENHANCED STRAIN MIXED METHOD 67 for the contribution in the element to the global residual.42) (8. is given by σ = J0 Σ(ξ) JT 0 These transformations have the property that tr(σ ˜) = j0 ˜ tr(Σ E) j(ξ) (8.38) Note that the residual for the enhanced modes will vanish at a solution since it belongs to a single element. the residual for the enhanced modes is computed from ˜ Rα = − Ωe ψ T σ dΩ α ˜ (8. The weighting by the jacobian determinant terms is motivated by the gradient of the shape functions. on the bi-unit square element to the global stresses. using the transformation defined at the element center. Using tensor notation we introduce the representations ˜ = j0 −T ˜ J0 E(ξ) J−1 0 j(ξ) (8. Similarly. Similarly. J0 .43) .40) The transformations may also be written in matrix form as ˜ = σ = where for 2-dimensional problems ˜ ET = ˜ ˜ ˜ ˜ E11 E22 E33 2 E12 (8.39) ˜ which represents a transformation of the local enhanced strains. a transformation of the local stresses. σ.

48.51) Thus. for non-linear materials. which indeed is identical to the modified incompatible mode formulation [22]. Here we consider the simplest form. A number of alternatives are discussed in the paper by Simo and Rifai [16]. satisfy the orthogonality condition Eq. 8. 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. the interpolations given by Eqs.46) j0 ˜ ΣT E j(ξ) (8. These interpolations have been incorporated into the element routine elmt04 which has been developed for a linear elastic-viscoelastic material. 8. thus.43. 8. For the simplest form. 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.50) for the enhanced strains are used.CHAPTER 8.49)   ξ1 0 0 0  0 ξ2 0 0  ˜  E =  0 0 0 0 0 0 ξ1 ξ2 (8. 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. that no ad-hoc assumptions are required in the enhanced formulation. contrary to what is necessary when using incompatible modes. It should be noted however.48) Thus.42 and Eq.45) and ˜ and σ have similar ordering. as well as.47) ˜ ΣT E d2 = 0 2 (8. ENHANCED STRAIN MIXED METHOD ΣT = Σ11 Σ22 Σ33 Σ12 68 (8. 8. 8.41 may be written as σT ˜ = The integral over the element becomes σ T ˜ dΩ = j0 Ωe (8.50 contain only linear polynomials in ξ and. 8. .49 and Eqs. 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.

53) We note that for the linear material model discussed previously that W( ) = 1 2 T D − T D 0 (8. the residual for the enhanced modes is computed from ˜ Rα = − Ωe ψ T σ dΩ = 0 α ˜ (8.38) dΠe = dη ( Ωe (s) ˜ U)T σ dΩ + Ωe ˜ ˜ ET σ dΩ (8. for outputs and stress projections to nodes. ENHANCED STRAIN MIXED METHOD 69 8. Rα .5 Solution Strategy: Newton’s Method The solution to a non-linear problem is commonly computed using a sequence of linear approximations. 8.56) In a manner identical to the linear elastic material. 8.58) ˜ We note above that at a solution the residual.CHAPTER 8. the residual in each element becomes: RI = FI − BT σ dΩ (8.55) = (s) u + ˜ In subsequent development we shall use these stresses for all calculations of arrays.35 to 8.57) I ˜ Ωe Similarly. A popular scheme is Newton’s method. W ( ).4 Non-Linear Elasticity For a non-linear. should vanish independently in each element. Thus. which may be summarized as: . through ∂W σ = (8. hyperelastic material the stresses are computed from a strain energy density function. for the enhanced formulation the variation in each element may be written as (see Eqs.52) ∂ The partial derivative is understood in terms of components.54) For the enhanced formulation the computation of stresses is given by ˜ σ = ∂W ∂ (8. as well as. where σij = ∂W ∂ ij (8.

Repeat steps b.65) which expands to (i) KIJ = Ωe ˜ ∂σ BI ∂ (i) ∂ dΩ = ∂uJ ˜ (i) BI Dt BJ dΩ Ωe (8.) until the solution converges to within a tolerance.59) (8.CHAPTER 8. ENHANCED STRAIN MIXED METHOD 1. 8. Convergence may be assessed from | dx(i+1) | < tol | x(i+1) | where | x | is the length of the vector. tol.60) (8. x.61) x=x(i) (8.63) ≈ = 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. 2. the terms in the Jacobian are defined as (i) KIJ ∂RI = − ∂uJ (i) (8.) and c.57 and Eq. 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. 8.64) In the above. Given the set of equations f (x) = 0 where x are the dependent variables.66) . 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. . Using Newton’s method on the set of equations defined by Eq. is the Jacobian or tangent matrix for the equations.58 above gives the problem RI ˜ (i+1) Rα (i+1) (8. 3.62) 4. F(i) . F(i) = ∂f ∂x ∂f ∂x dx(i+1) = 0 x=x(i) 70 (8.

After the solve.70) Since the second equation in Eq. these may then be used for the update ˜β(i+1) = ˜β(i) + d˜β(i+1) (8.69) and ˜ (i) Hαβ = (8.74) The reduced first equations may be assembled into the global equations.67) define the tangent moduli for the material.71) (8. ˜ (i) ΓαJ = Ωe ∂2W ∂ ∂ (i) (8.71 to compute the increments to the enhanced modes.72) (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. the assembled equations become ˜ (i) KIJ duJ(i+1) = e e ˜ (i) RI + F I (8. Accordingly. Thus after adding any nodal forces. duJ(i+1) . we may perform a partial solution by static condensation. For the non-linear elastic material ˜ (i) Dt = Similarly.76) The incremental displacements also may be substituted back into Eq. 8.73) (8. the new nodal displacements are updated uJ(i+1) = uJ(i) + duJ(i+1) (8. ENHANCED STRAIN MIXED METHOD where ˜ ∂σ ˜ (i) Dt = ∂ 71 (i) (8.77) .68) ˜ (i) ψ T Dt BJ dΩ α ˜ (i) ψ T Dt ψ β dΩ α Ωe (8.64 is complete at the element level. FI . 8.75) which may be solved for the incremental nodal displacements.CHAPTER 8.

Set k ← k + 1 and repeat Steps 2. 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. This requires additional storage for the enhanced formulation with respect to that needed for a displacement or a mixed B-bar type of formulation.k+1) (8.84) and save for the next global iteration. Update the solution ˜β(i.78) where a single superscript i denotes the value of ˜β computed in the last global iteration.83) 5. In the enhanced element for 2-dimensional plane strain applications in FEAP. 6.k) 3.k) + d˜β(i. until convergence achieved (or a set number of k-steps is completed).79) ˜ (i.k) Rα (uJ (i). ˜ 2. ˜ The alternate algorithm is given by linearizing the residual. Solve for the increment ˜ (i.k) ) − Hαβ d˜β(i.CHAPTER 8.k) α 4. 8.80) with ˜ ˜ (i. Compute the linear part of Rα as ˜ ˜ (i.81) (s) u(i) + ˜(i. Set ˜β(i) = ˜β(i. for this algorithm.71 for the later update of the enhanced modes.k+1) (8. it is necessary to save the arrays used in Eq.k+1) = ˜β(i. to 4. the arrays are moved into history arrays using a pmove routine.k) ψ T Dt ψ β dΩ α (8. 1.k) Hαβ = Ωe (8.k) ˜ d˜β(i. as well as use for subsequent steps for the global i-iterations or to compute stresses. Consequently. For k = 0 set ˜β(i. (8.k) = ∂ σ Dt ∂ (8. Rα . ˜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. 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. with respect to ˜β only.82) . J Accordingly.k+1) = (Hαβ )−1 R(i.k+1) = 0 where now ˜ (i.k) = ˜β(i−1) (8. ˜β(i) .

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). In FEAP.72 to Eq.1) ul(ndf.k) putation of Rα and Hαβ .i). 8.nen. on uJ (i) Problems All All All Transient Transient Transient All1 73 Table 8.nen. ∆uJ (i) = uJ (i) − uJ (tn ) Difference between current and previous solution duJ (i) = uJ (i) − uJ (i − 1) Increment from last iteration Table 8.4) ul(ndf. . The algorithm requires repeated com(i.nen.3) ul(ndf. 8. If the k (i) iteration is converged.nen.75 thus simplifying slightly the steps involved. The definitions of the entries in the local array are given in Table 8. In subsequent presentations we shall discuss the construction of these steps for linear viscoelastic materials.nen. Dt .nen. the uJ (i) nodal displacement vector and the ∆uJ (i) and duJ (i) nodal incremental vectors are retained in global arrays.nen.72 to Eq. ENHANCED STRAIN MIXED METHOD Vector Definition uJ (i) Description Current solution value at each node. J.2.75 for the global steps.c. While the above process has been illustrated for the non-linear elastic material. 8. it may be directly extended to any material for which we can iteratively compute the stresses.2: Element Local Arrays The only information to be stored is the ˜β(i) . ul(ndf. The array ul contains information for the current element according to the definitions in Table 8. ˜ (i) ˜ σ (i) . Once the k-iteration is completed. and a class of viscoplastic materials. the Rα is zero in Eq.k) (i.1: Element Local Arrays Array ul(ndf.1.6) ul(ndf. elasto-plastic materials.7) Description local uJ (i) local ∆uJ (i) local duJ (i) ˙ local uJ (i) ˙ local uJ (i) ˙ local uJ (i − 1) used for b.nen. 8. The global arrays are passed to each element in a local array.5) ul(ndf. however. linearization with respect to both uJ and ˜B is performed. and the tangent moduli. leading to Eq.2) ul(ndf.CHAPTER 8.

in matrix form the stress and strain may be split as σ = s + mp and (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.Chapter 9 Linear Viscoelasticity 9.6) K = 3 (1 − 2 ν) 74 .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. e is the strain deviator. and p is the mean (pressure) stress defined in matrix form as 1 p = mT σ (9. Accordingly. s is the stress deviator.1) 1 mθ (9. and θ is the volume change defined in matrix form as θ = mT (9.3) 3 is strain. In the discussion to be presented here we assume the material is linear and isotropic.2) 3 where σ is the Cauchy stress.5) where K is the bulk elastic modulus defined in terms of Young’s modulus and Poisson’s ratio as E (9.

at time labeled zero (t = 0). a specimen is subjected to a constant strain.13) qi (t) = −∞ exp − t−τ ˙ e(τ ) dτ λ1 (9. is measured. e.9) (9. The integral for each term is given by the homogeneous differential equation solution. The relaxation modulus function is defined in terms of an idealized experiment in which. Alternatively.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. and the stress response.CHAPTER 9. 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.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. qi . LINEAR VISCOELASTICITY The deviatoric parts are assumed to satisfy a linear viscoelastic model.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. s(t). 75 Linear viscoelastic behavior may be stated in the form of differential equation models or in the form of integral equations. The set of first order differential equations may be integrated for specified strains.11) ˙ qi + 1 i ˙ q = e λi (9. the operator may be written as N s = 2 G (µ0 e + i=1 µi qi ) (9.8) (9. e0 .12) This form of the representation is equivalent to a generalized Maxwell model (a set of Maxwell models in parallel). In the differential equation model the constitutive equation may be written as P (s) = 2 G Q(e) (9. For a linear material .

20) λ1 where µ0 + µ1 = 1 (9.15) where G(t) is defined as the shear relaxation modulus function. This relation may be written as s(t) = 2 G(t) e0 (9.19) With this form the integral equation form is identical to the differential equation model for the generalized Maxwell material.CHAPTER 9.22) . At time zero a strain may be suddenly applied and then varied over subsequent time.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 N G(t) = G0 + i=1 Gi exp −t λi (9. LINEAR VISCOELASTICITY 76 a unique relation is obtained which is independent of the magnitude of the applied strain.17) or the alternate form G(t) = G (µ0 + N µi exp i=1 −t ) λi (9.18) where µ0 + N µi = 1 i=1 (9..e. 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. The addition of more terms may be easily accommodated based upon the one term representation. In the subsequent discussion we will consider the generalized Maxwell material and let N be 1 (i. 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 (9.21) In applications involving a linear viscoelastic model. it is usually assumed that the material is undisturbed until a time identified as zero. the standard linear solid). −t G(t) = G (µ0 + µ1 exp ) (9. Accordingly.

30) λ1 tn λ1 The strain rate is now approximated as constant over each time increment tn to t. 9.24) It remains to evaluate the integral.27) Including the negative exponential multiplier term gives h1 = exp and then h1 (t) = exp where −t 1 i λ1 (9.28) −∆t 1 h + ∆h λ1 n (9.29) −t t τ ˙ exp e(τ ) dτ (9. Accordingly.16 gives t s(t) = 2 G(t) e0 + 2 0 ˙ G(t − τ ) e(τ ) dτ (9. thus ∆h = exp ˙ e(τ ) ≈ e(t) − en . tn ≤ τ ≤ t (9.20 into Eq. 9. The result of this separation when applied to Eq.31) ∆t where en denotes the value of the strain at time tn and ∆t denotes the time increment t − tn . and the last term covers the subsequent history of strain. Substitution of Eq. the second term includes a jump term associated with e0 at time zero.26) the above separation gives t i (t) = i (tn ) + tn 1 1 exp τ ˙ e(τ ) dτ λ1 (9.32) . 9.23 gives s(t) = 2 G [µ0 e(t) + µ1 exp −t (e0 + λ1 t exp 0 t ˙ e(τ ) dτ )] λ1 (9. LINEAR VISCOELASTICITY 77 The first term is zero.25) If we define the integral as t i1 (t) = 0 exp τ ˙ e(τ ) dτ λ1 (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.CHAPTER 9.23) where subsequently the 0 limit on the integral is understood as 0+ . we divide the integral as t tn t (·) dτ = 0 0 (·) dτ + tn (·) dτ (9.

35) This form requires only one evaluation of an exponential term. The constitutive equation now has the simple form s(t) = 2 G [µ0 e(t) + µ1 h1 (t)] (9. the form is doubly asymptotically accurate. Thus. a very large value of the time step producing a zero value. Thus.38) . While the above form is easy to evaluate it has problems when the size of the time step is changed. It may also be extended for use with thermorheologically simple materials. 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. we need to compute ∂s ∂ = ∂s ∂e ∂e ∂ = ∂s Idev ∂e (9. Furthermore. 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. 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 (9.36) This approximation produces a singular ratio for zero time steps.CHAPTER 9. The use of finite difference approximations on the differential equation form directly does not produce this property. For very small time steps a series expansion may be used to yield accurate values. Accordingly. the limit value is well behaved at a unit value. as well as.37) The inclusion of more terms in the series reduces to evaluation of additional hi (t) integral recursions. a zero value of the time step produces a correct answer.33) where tn+ 1 denotes the time 1 (tn + t). Thus. however.34) h1 (t) = exp − ∆t − ∆t 1 [exp h + (e − en )] 2 λ1 2 λ1 n (9. The implementation of the viscoelastic model into a Newton solution process requires the computation of the tangent tensor. a more stable form is used in FEAP where the integral over the time step is evaluated in closed form [23].

43) for the elastic shear modulus. Note the simplicity of the additional coding needed to include the linear viscoelastic formulation. whereas for very large increments the equilibrium modulus µG is used.39) The only modification from a linear elastic material is the substitution of the factor Gvisc = G [µ0 + µ1 ∆h1 (∆t)] (9. use of the consistently derived tangent modulus terms leads to convergence in one iteration (the second iteration produces a zero residual).CHAPTER 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. 9.40) ∂h1 = 2 G [µ0 I + µ1 ] ∂e (9. Again we note that for zero ∆t the full elastic modulus is recovered.39 becomes ∂h1 ∂e Thus. I (9. Since the material is linear.42) = ∆h1 (∆t). 0 The above formulation is incorporated into the subroutine viscoe. LINEAR VISCOELASTICITY 79 where Idev is the deviatoric operator identified previously.41) (9. the tangent tensor is given by ∂s = 2 G [µ0 + µ1 ∆h1 (∆t)] Idev ∂ (9. .

If the material is nonlinear hyper-elastic we may deduce the stress from the expression for the elastic strain energy as ∂W σ = (10. For a linear hyper-elastic material the stress to elastic strain relation is given by σ = D e = D( − p − 0 ) (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. γ. Accordingly. ˙p = γ ˙ ∂f ∂σ (10.4) where f is a loading function and γ is a scalar rate term called the plastic consistency ˙ parameter. A back stress is defined as α which is related to the plastic 80 .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 (10.2) ∂ 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.Chapter 10 Plasticity Type Formulations 10. The plastic consistency parameter.3) In the following discussion we limit our comments to linear elastic materials and also set 0 zero.1) where e are the elastic strains and p are the inelastic strains.

10. Different limit equations may be written for rate independent plasticity. the model is completed by describing a scalar function.11) is used. Y0 is related to σy . and is expressed as f (σ. α. ep ) ≤ 0 ¯ (10. Using the definition of the plastic strain rate the effective plastic strain may also be written as t (10. The simplest relation is for classical. the yield stress in uniaxial tension.7) ep = ¯ 0 γ ˙ 2 ∂f ∂f · 3 ∂σ ∂σ 1 2 dt (10. rate dependent plasticity. which describes the limit behavior of the model. .6) ˙ α = where the stress and back stress appear in the form Σ = σ − α and Y (¯p ) = Y0 + Hiso ep e ¯ (10. Generally.8 Hiso is an isotropic hardening modulus.8) is a function which measures the size of the current yield surface. PLASTICITY TYPE FORMULATIONS strain rate through 81 2 2 ∂f Hkin ˙ p = Hkin γ ˙ (10. Commonly. 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. using the same function as the loading function.CHAPTER 10.10) 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. rate independent plasticity where g = f (σ. The yield surface is defined in an associative manner. associative. In the present study both the isotropic and the kinematic hardening moduli will be assumed as constants. ep ) = F (Σ) − Y (¯p ) ¯ e (10.5) 3 3 ∂σ where Hkin is a kinematic hardening modulus. g.9) 3 0 In Eq. α. Later alternative forms will be introduced to represent other types of material behavior. and generalized plasticity models.

the plastic strain is given by p n+1 = p n + λn+1 ∂f ∂σ (10. and ep . a discrete solution at time tn is defined in terms of the state σ n .16 is checked in Eq.13) (10. for the present we restrict our attention to plasticity as defined by Eq. A very effective method to integrate plasticity equations is the operator split method with a return map concept [17. If the state at tn+1 is elastic the stresses (as well as other state variables) are set equal to the trial value. αn . a correction is required to include the inelastic terms. Accordingly.2 Solution Algorithm for the Constitutive Equations The solution of the above set of equations may be effected numerically using a variety of algorithms. Accordingly. The ¯n n solution is then advanced to time tn+1 by specifying the strain. 10. α. PLASTICITY TYPE FORMULATIONS 82 10. The step is initiated by taking the ¯ trial values for plastic quantities pT R p (10. 10. This implies that there will be no ˙ change to p .14) (10.CHAPTER 10.11 to determine if the step is elastic or whether inelastic terms should be active.16) The trial stress given by Eq. otherwise. 3. n+1 . 4] the algorithm may be extended to include a variety of modeling concepts for the limit behavior.6 and Eq. 10. however.2) for the time tn+1 which requires the rate equations for p and α to be integrated over the time increment. To initiate the solution at tn+1 a trial state is computed assuming the step is entirely elastic. or ep during an elastic increment. integrating non-linear terms using a backward Euler implicit method between tn and tn+1 . 18.17) n+1 and the back stress by ∂f p ˆ αp n+1 = αn + Hkin λn+1 ∂σ (10. For an inelastic step the stresses must satisfy (1.12) 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 (10.11 above.18) n+1 . p . 10. Recall that a step is elastic when γ is zero.15) pT R n+1 ) − (10.

10.3.CHAPTER 10. 10.24) σ n ∂Σ ∂f ˆ R(j) = αn + Hkin λ − α(j) (10.25) α ∂Σ and (j) Rf = − f (σ (j) . 10.. The iteration counter is shown as a superscript (j) and initial iterate values are taken as the trial stress and zero λ( 0). normally. ep (j)) ¯ (10. To simplify the notation the subscripts on n + 1 are omitted.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  (10.21) n+1 (10. The iterative solution is continued until the norm the residuals are within acceptable tolerance values (e. 10.22) ∂f ∂f ∂Σ ∂f = = − (10. 10. in the form ∂f R(j) = − p − D−1 σ (j) − λ (10. 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).4 produces ¯n ep ¯n+1 = ep + λn+1 ( 3 ∂f ∂f 1 · )2 2 ∂σ ∂σ (10.18 and Eq. A Newton method may be used to solve the equations.g.20 as residual equations.19) (10. Before proceeding with Newton’s method we note that the following relations hold ∂f ∂f ∂Σ ∂f = = ∂σ ∂Σ ∂σ ∂Σ and (10. integration of Eq. treating the equations Eq.20 constitute a set of non-linear equations in terms of the values of σ n+1 .23) ∂α ∂Σ ∂α ∂Σ Thus. 10.11 at tn+1 gives gn+1 = fn+1 = 0 Finally. 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.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.18 and 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. α(j) .

if we now consider a linearization with respect to strain only Rσ contributes to the change. Dt . respectively.30) α(j+1) = α(j) + dα(j+1) and λ(j+1) = λ(j) + dλ(j+1) (10.27 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. 10.34) At convergence for the given strain. R f = 0 (10.27 are computed and added to obtain the next iterates. the residuals will vanish.31) where KIJ and RI are the element stiffness and 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. the stresses may be used in the finite element to compute each element residual.e. Rα = 0 . The linearization of the residuals with respect to an increment of strain yields Rσ = d . however.CHAPTER 10. 10.29) The solutions to Eq. as before. This is accomplished. Once convergence is achieved for each stress point evaluation (i. The solution is terminated whenever the norms of the residuals are smaller than a selected small tolerance. That is we need to perform a new solution to see if the strains we used to compute the stresses are correct. . by solving KIJ uJ = FI + RI (10.28) ∂2f ˆ C = D−1 + λ ∂Σ2 (10. Accordingly. σ (j+1) = σ (j) + dσ (j+1) (10.32) define the next iterates. In addition it is necessary to compute the tangent moduli. PLASTICITY TYPE FORMULATIONS where A = and 84 ∂Y ∂¯p e p ∂λ ∂¯ e (10. for use in the element stiffness matrix (if one is used) for the next iteration on the momentum balance equation.. 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.36)  ˆ ∂Σ ∂λ ∂Σ  ∂Σ T T ∂f ˆ ˆ ˆ D31 D32 D33 − ∂f −A ∂Σ ∂Σ .33) e e (10. to compute the stress at each Gauss point for a given strain).

[13].39) with each part of the yield surface described by a separate function gk = fk (σ.CHAPTER 10. usually only 1 or ˙ 2 of the surfaces are active at any time. 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. The method provides the best estimates for the parameters and their sensitivities to errors or inconsistencies in the data. Thus. et. α. θ. Multiple yield surfaces may be included by modifying Eq.38) Except for giving the form of f this completes the specification of the general algorithm. 10. the decomposition into elastic and plastic parts may now be expressed as e = ee + ep (10.4 to K ˙p = k=1 γk ˙ ∂fk ∂σ (10.3 Isotropic plasticity: J2 Model As in previous developments.40) An active condition for each surface is denoted when γk ≥ 0.41) . 10.43) and θ = θe + θp (10. the strain is again split into deviatoric. e. the tangent moduli for the next momentum iteration is ˆ Dt = D11 85 (10.44) 1 mθ 3 (10.42) For our study on inelastic behavior. As constitutive equations become complex the specification of the parameters is more difficult. ep ) ≤ 0 ¯ (10.37) (10. al. and volumetric (spherical). A systematic procedure for determining the parameters from experimental data is given by Ju. parts and expressed in matrix notation as = e + where θ = mT (10.

CHAPTER 10. and pressure (spherical) parts as σ = s + mp where 86 (10.49) J2 = and J3 = det (s) (10. J2 . thus. the limit equation as g = f (s. ep ) = ¯ s − α − Y ep ≤ 0 ¯ (10. For isotropic materials the yield and loading function may be expressed in terms of the invariants of stress and back stress. and J3 and given by J1 = m T s = 0 (10.47) where Y is the radius of the yield function which is related to the uniaxial yield stress.52) and including the back stress. through 2 Y = (σy + Hiso ep ) ¯ (10.46) 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 .51) The simplest formulation is where the function depends only on J2 . We write this model using 1 2 J2 = (sT s) 2 = s (10.54) 3 and. The back stress adjusted value Σ is given by Σ = s − α (10.50) (10.53) 1 T s s 2 (10. α. PLASTICITY TYPE FORMULATIONS The stress also is split into the deviatoric. includes the effects of linear isotropic hardening. ∂f ∂f ∂Σ ∂f = = − (10.45) 1 T m σ (10. s. The invariants of s are denoted as J1 .56) ∂α ∂Σ ∂α ∂Σ .48) In the developments below we restrict plasticity to the deviatoric part only. ∂f ∂f ∂Σ ∂f = = ∂s ∂Σ ∂s ∂Σ . Thus θp vanishes and the yield function can depend only on the deviatoric part of the stress. σy .

57) Thus.60) (10. thus.67) (10. PLASTICITY TYPE FORMULATIONS (where we recall that Idev = I − ∂f = ∂Σ 1 3 87 m mT ). Thus.63) (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.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. nT n = 1 (10.65) (10. the evolution of the back stress satisfies ˙ α = 2 ∂f 2 Hkin γ ˙ = Hkin γ n ˙ 3 ∂σ 3 (10.58) Noting that at the initial state α is zero. our model is completed by giving the evolution equations for plastic strain and effective plastic strain in the form ˙ ep = γ n ˙ t (10.CHAPTER 10.64) (10. mT n = 0 .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 (10.70) .61) (10. mT α = 0 With this fact we then have the following important properties mT Σ = 0 .66) (10.68) (10. and Σ Σ = n = ∂f ∂σ (10.

10. 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. Accordingly. n+1 epT R = ep n+1 n . 10. 10.71 and 10. If the trial values violate 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. If the limit equation is satisfied then the trial values define the solution at tn+1 . 10. 10.74 gives also that sT R − αn is in the n+1 direction nn+1 .CHAPTER 10.71) 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. 10. epT R = ep ¯n+1 ¯n (10. ΣT R − Yn = 2 [G + 1 (Hiso + Hkin )] λn+1 3 (10. αT R = αn n+1 .75 with Eq.64 gives sn+1 = sT R − 2 G λn+1 nn+1 n+1 Next subtracting Eq.72) which may be used to check the limit equation.73) Noting that Σn+1 = Σn+1 nn+1 Eq.68 and Eq.72 in Eq.79) nT R n+1 = ΣT R n+1 ΣT R n+1 (10.78) . The second part of the algorithm solves the discrete rate equations using the trial values as initial conditions. using Eqs.75) that is. In the above ΣT R = sT R − αn n+1 n+1 Combining Eq.70 yields the scalar equation. 10. The solution is performed using a trial state based upon the assumption that λT R is zero.66 from Eq.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. Accordingly. 10. 10. the coefficient must vanish to obtain a solution. it is necessary to perform the second part of the algorithm.65. 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 . gn+1 .74) (10.

e. e. If the momentum equation is not satisfied for the current time. n+1 . 10. for all the other variables.78 is perturbed as λn+1 = ΣT R n+1 2 [G + − Yn (1 + tol) (Hiso + Hkin )] (10.84) . and etc. to avoid added complexity in the linearization performed below. Similarly. say 10−8 . It should be understood that λ denotes the value of the solution in the tn+1 step. 10. the equation equivalent to Eq.82) Rα = α − αn − and R f = Yn + 2 Hiso λ − Σ = 0 (10. λn+1 . however.80) 1 3 where tol is a small value. a check on satisfaction of the momentum equation must be made. Accordingly. Writing appropriate residual equations as Rs = e − e p − n 1 s − λn = 0 2G 2 λ Hkin n = 0 3 (10. For the solution process developed here. 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.83) 3 In the above we have deleted specific reference to the values at tn+1 . In practice λn+1 is reduced slightly to that sn+1 is always slightly outside the limit yield condition. which we denote by de. tn+1 . the back stress. if non-linear kinematic hardening is introduced the reduction to a scalar equation may also be complicated. previously given as λn+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.CHAPTER 10. another iteration is necessary to improve the estimate for the state of strain. Accordingly. PLASTICITY TYPE FORMULATIONS 89 as well as the plastic strain and the back stress. Once the converged value for λn+1 is known and the stress state determined. this may be achieved by selecting as primary dependent variables the stress. We note that for the current strain. i. as shown above. sn+1 .77 will be nonlinear in λn+1 and a linearization and iteration process are required to obtain a solution.81) (10. the solution to Eq. 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. and the consistency parameter. the linearization n becomes ∂n = ∂Σ 1 Σ (I − n nT ) = 1 Σ N (10. the above equations are satisfied. αn+1 .

90) Substituting this result back into the first of equation Eq.87 yields the incremental equation which yields the tangent modulus matrix for the algorithm. Thus. . d . Accordingly.92) Finally.89) This result may be substituted into the third equation in Eq. 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.91) G (Hiso + Hkin ) (10.81 to 10.87) where B and C are given by B = 2Gλ = Σ D 2Gλ ΣT R .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 (10.88) and where we have noted that D = 1 + 2 (G + Hkin ) 3 λ Σ = ΣT R Σ (10. This also permits the substitution of alternative limit equations without changing the solution to the first part.CHAPTER 10. the linearization of Eqs. and then substituting the result into the third equation to obtain a final expression for dλ in terms of de. 10. for the differential strains. C = 2 Hkin λ 2 Hkin λ = 3 Σ D 3 ΣT R (10. we obtain ds = 2 G [I − B (I − n nT ) − A n nT ] de where A = G + 1 3 (10. 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. 10.86) 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 (10.85 to obtain 2 G nT de = 2 [G + 1 (Hiso + Hkin )] dλ 3 (10.93) See Appendix E for a discussion on the inverse of this type of matrix. the tangent becomes ds = 2 G Idev − B (Idev − n nT ) − A n nT d 1 (10. PLASTICITY TYPE FORMULATIONS 90 Using this result.

α.3.70. For loading rates which are finite. The function Φ together with the parameter ζ determine the rate dependency of the model. Accordingly. ep )] − ζ γ ≤ 0 ¯ ˙ (10. the relationship for g becomes a constitutive equation describing the evolution for the consistency parameter. Perzyna considers alternatives for representing Φ. Indeed. In this section we show how such a formulation may be easily extended to include rate effects in the inelastic behavior. 10. αn+1 . through 2 (σy + Hiso ep ) ¯ (10. 10.4 Isotropic viscoplasticity: J2 model The previous section presented the formulation and solution algorithm for a J2 classical plasticity model.CHAPTER 10.94 for the time interval tn to tn+1 (i. f . 10.96) Y = 3 and.95) with Y the radius of the yield function which is related to a uniaxial yield stress. For viscoplasticity.e. ∆t) using a backward Euler evaluation of the integrals to obtain ∆t Φ [f (sn+1 . we ˙ write g = Φ [f (s.97) where m is a positive integer power. The formulation is completed by integration of the constitutive equation Eq.. the only modification to the formulation is the replacement of the limit equation for g. For the viscoplastic model considered here. includes the effects of linear isotropic hardening.94) where the yield condition. thus. other models representing the problems of generalized plasticity and generalized viscoplasticity can also be developed by such replacement.98) .63 to 10. Other functional forms for Φ may be considered without conceptual difficulty. however. All the other equations for the model remain as given in Section 10. Y and σy relate to the inelastic behavior in the limit as γ tends to zero (the rate independent ˙ limit).69 and 10. α. still is given by f (s.95 exceeds zero. the behavior is inelastic and the return map solution for the viscoplastic model is given by Eqs. γ.67. 10. In viscoplasticity. here we restrict our attention to the simple case given by Φ (f ) = (f )m (10. the stress state may lie outside the yield surface. 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]. ep ) = ¯ s − α − Y ep ¯ (10. PLASTICITY TYPE FORMULATIONS 91 10. σy . ep )] − ζ λn+1 = 0 ¯n+1 (10. For trial stress values for which the yield function defined by Eq.

For nonzero ζ. The stress. we obtain ∆t Φ [ ΣT R − Yn − 2 (G + 1 (Hiso + Hkin )) λn+1 ] − ζ λn+1 = 0 3 (10. 10. and effective plastic strain are computed using the same expressions as for the classical plasticity model.100) In general. 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. .98 to obtain a single nonlinear equation in λn+1 . PLASTICITY TYPE FORMULATIONS 92 The discrete consistency parameter may be obtained by combining the scalar coefficient from Eq. back stress.101) Comparing Eq. 10. the above equation is nonlinear and must be solved numerically.101 implies a rate dependency. with results depending on time durations for applying and changing loads on the body. the above becomes ∆t ΣT R − Yn − 2 [G + 1 (Hiso + Hkin )] λn+1 3 m − ζ λn+1 = 0 (10. 10. Accordingly. 10.75 with Eq. The extension to higher powers of m may be constructed using a Newton solution scheme to solve the non-linear scalar equation.99) For the simple model used here for Φ. 10. plastic strain.101 to Eq.CHAPTER 10. the presence of ∆t in Eq.80 we can observe that the limit solution for ζ zero is identical to the classical plasticity problem.

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

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)

KIJ .31 gives dΠcη dη = η=0 Ωc ΛT g(u) − + 1 λ k dΩ + Ωc UT GT (λ + λA ) dΩ (11. The variation to Eq. ill conditioning for the solution to the linear equations will result.30 which are several orders larger than components appearing in the stiffness. thus.CHAPTER 11.33) ΛT g(u) dΩ A Ωc The Euler equation for the variation of λ gives the equation g(u) − 1 λ = 0 k (11. For each step in the analysis we assume: . In order to yield a solution which provides an adequate satisfaction of the constraint equation. k. Furthermore. 11. an alternative approach is needed. the augmented Lagrangian method is introduced as an alternative. but with λA used as the multiplier.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. Using. iterative solutions become very difficult for these large penalty values. the constraint equation is satisfied independently of the value of the penalty parameter. The method may be made computationally viable by making the determination of λA an iterative algorithm.32) (11. the constraint may be violated by an unacceptable amount.31) where λA is the augmented term. if too small. In the Uzawa algorithm we introduce an outer iteration loop for the augmentation. these facts we also note that the algorithm merely reduces to the original Lagrange multiplier method. In the next section. 106 or 107 ). fairly large values for the penalty parameter should be used (generally on the order of about half machine precision. λ.. λA ) = Ωc (λ + λA )T g(u) dΩ − Ωc 1 T λ λ dΩ k (11. The values used then yield stiffness modifications for the second term on the right hand side of Eq. The Uzawa algorithm is the simplest algorithm which may be considered.g. and we also conclude that λ must vanish at the solution.35) and. The variation for λA gives the constraint equation g(u) = 0 (11. e.34) which may be used to compute λ. Consequently. 11. 11. AUGMENTED LAGRANGIAN FORMULATIONS 97 The above solution process defines a perturbed Lagrangian form of the penalty solution process. If the values are too large.

4. For each time. Let λβ (0) = 0. . The number of non-linear iterations will decrease for the later augmented steps since the violation in the constraint is becoming less and less. set j to zero and take the initial value of the augmented multiplier as λA β(j) = λβ (tn ) A (11. tn+1 . The larger k is made the more rapid the convergence. To perform the above algorithm it is necessary for the penalty parameter k to be large enough for the iteration to converge. After the iteration for the incremental problem converges update the augmented parameters using β(j+1) β(j) λA = λA + λβ (11. The convergence rate for the augmented iteration is generally linear.CHAPTER 11. 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. Use of values with this range in magnitudes leads to 3-6 augmented steps for most problems. 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. A 2.38) (11. AUGMENTED LAGRANGIAN FORMULATIONS 98 1.40) where λβ = Kλ αβ −1 Pλ α (11.39) Pλ α In the above the iteration aspects for the incremental problem are not shown. If the constraint is satisfied to within a specified tolerance. Let j be the augmentation iteration counter. Check convergence for the augmented step. 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).41) is computed using the converged solution from step 2.37) (11. If not converged increase the j counter and repeat steps 2 and 3. 3. not quadratic as in a Newton solution. All that is required is that the terms in the added stiffness be somewhat larger than the original stiffness terms.36) where the dependence on the n+1 step on the left side is implied.

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

2) ¨ in which u is the acceleration vector.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 (12.3) − 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. or enhanced method as described in previous chapters.1) ˜ where σ is computed for a displacement. bv .4) 100 . In order to extend the variational equation to accommodate transient analysis. is replaced by ¨ bv ← bv − ρ u (12. mixed. 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. the body force vector.

11) (12. in the total matrix form as ˜ ¨ R = F − P(σ) − M u (12.6) where MIJ is the mass matrix for the problem.12) .7) the residual equation becomes ˜ ¨ RI = FI − PI (σ) − MIJ uJ or. as described in the next section. and accelerations and these are given as: un ≈ u(tn ) ˙ vn ≈ u(tn ) (12. ˙ ¯ u(0) = v0 (12.8) In general. consequently. the inertia term may be written as ¨ ¨ NI ρ u dΩ = MIJ uJ e Ωe (12. The Newmark method is a one-step method which may be used to advance a solution from time tn to time tn+1 .10) ¯ ¯ where d0 and v0 are the initial displacement and velocity vectors. 12. the above equation is a non-linear set of ordinary differential equations. by ignoring the nodal indices.5) (12. The Newmark method uses approximations to the displacements.9) (12.9 by a time marching process using the classical Newmark method of solution [14]. The practical solution of the equations is accomplished using a time marching scheme. velocities. The method is self starting. TRANSIENT ANALYSIS 101 The last term is the inertia contribution to the momentum equation. the solution at the first increment may be determined.2 Newmark Solution of Momentum Equations In this section we illustrate the solution of Eq. given the initial conditions. If we define ˜ PI (σ) = e Ωe ˜ BI σ dΩ (12. For continuum problems the acceleration is computed from the isoparametric interpolations as ¨ ¨ u = NJ (x) uJ (t) thus. ¯ u(0) = d0 . 12.CHAPTER 12.

an+1 . 12. 12. however.14) In order to advance the solution to the next time it is necessary to recast the problem in an iterative form. 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 ] (12.16) (12. involves only the mass matrix. where quite large time steps may usually be taken. For γ = 1 there is no numerical dissipation. The advantage of implicit solutions is improved stability. 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 example.CHAPTER 12. Accordingly. whereas 2 for γ > 1 numerical dissipation is introduced.17 and Eq. ¯ v0 = v0 (12. For a diagonal mass this solution step is very efficient.17) . A β of zero leads to a formulation which is called explicit. the method for linear problems is unconditionally stable. ˜ 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.18 with the momentum equation written at time tn+1 . This involves selecting appropriate values for the variables to initiate .25. This method is commonly called trapezoidal rule or constant average acceleration. where for no damping. Accordingly. for β = 0. TRANSIENT ANALYSIS and ¨ an ≈ u(tn ) 102 (12. For β non-zero.19) 1 − β) an + β an+1 2 (12.13) The initial state is completed by solving the residual equation at time zero. the solution for the acceleration. Values of β less than 0. The values of β control primarily 2 the stability but also influence the form of the matrix problem. in general the method is only conditionally stable and very small time steps are needed.15) (12. ˜ Rn+1 = Fn+1 − P(σ n+1 ) − M an+1 = 0 (12.18) in which β and γ are numerical parameters which control the stability and numerical dissipation. respectively. The advancement of a solution from one step to the next is completed by combining Eq.25 should not be used since they are only conditionally stable with allowable time steps not much larger than the explicit scheme.

26) β ∆t and 1 (i+1) (i) (i+1) an+1 = an+1 + du (12. For an implicit solution it is best to select the initial value for the iterate as ( un+1 0) = un (12. as the primary unknowns. an+1 . With the choice Eq.23) (12. the velocities. 12. un+1 . Ct is a tangent damping matrix. linearization of the equations. which impede convergence of the Newton method. In the sequel we will address the implicit case and use the displacements.30) β ∆t β ∆t2 In Eq. This vector may be the displacements. especially near boundaries.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) (12.24) (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 (12.29) = − (i) ∂R ∂R ∂v ∂R ∂a + + ∂u ∂v ∂u ∂a ∂u γ 1 Ct + M (12.28) (12. solution of the linearized equations. 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. un+1 . vn+1 . and M is the mass matrix introduced above. and updating 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. Kt = Kt + .25) γ (i+1) (i) (i+1) vn+1 = vn+1 + dun+1 (12. For the explicit case the only viable choice is accelerations. or the accelerations.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) (12.20) Any other choice may perturb the displacements in such a way to cause false inelastic values.30.22) = β ∆t2 dan+1 = γ ∆t dan+1 (i+1) (12.CHAPTER 12. 12.20. TRANSIENT ANALYSIS 103 the step. Kt is the tangent stiffness matrix as computed for the quasi-static problem.

TRANSIENT ANALYSIS 104 12. α2 4 β= and (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. the HHT-method [9] and has been analysed extensively for stability and dissipative properties by Hughes [12].33) (12. for short.31 gives the tangent matrix K = for use in the Newton method K dun+1 = Rn+α (12. the relations.36) .35) Linearization of 12.31) In the above tn+α = (1 − α)tn + αtn+1 . This algorithm is called the Hilber-HughesTaylor α-method or.34) α+γ = are employed.37) 1 αγ M+ C + αK 2 β∆t β∆t (12.CHAPTER 12. To reduce the properties to a single parameter. 3 2 (12.32) (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 .

a = 1.2) where ea are unit base vectors for the current time. by Curnier in [5]. see Chadwick [2] or Gurtin [7]). t) (13. in the current configuration. 105 . Ω. Solution by finite element methods is considered by Crisfield in [3]. x. 3 (13. and by Zienkiewicz and Taylor in [27].1 Kinematics and Deformation The basic equations for solid mechanics may be found in standard references on the subject (e. common origins and directions are used for the reference and current coordinates. The position vector in the current configuration may be given in terms of its components as x = xa ea . Here only a summary of the basic equations is presented. After the body is loaded each material point is described by its position vector.Chapter 13 Finite Deformation Formulations: Elasticity 13..1) where EA are unit base vectors.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. In Cartesian coordinates the position vector may be described in terms of its components as: XS = XA EA . Exceptions occur when quantities are related to both the reference and current configurations. in a three dimensional space. The position vector at the current time is related to the reference configuration position vector through the mapping x = φ(X.g. Ω0 . A body B has material points whose positions are given by the vector X in a fixed reference configuration1 . 3 (13. In our discussion. A = 1..

In the above 1 is a rank two shifter tensor between the two coordinate frames.8) F = ∂X subject to the constraint J = det F > 0 (13. may be introduced as the change between the two frames. x = 1X + u (13.CHAPTER 13. and is given by (13. that is dv = det F dV = J dV (13. 3 A where δaA is a Kronecker delta quantity such that δaA = In component form we then have xa = δaA XA + ua (13. a.7) 1 if a = A 0 if a = A (13. The determinant of the deformation gradient maps a volume element in the reference configuration into one in the reference configuration.4) is used.5) 1 = δa Aea ET . 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 (13.10) where dV is a volume element in the reference configuration and dv its corresponding form in the current configuration.6) A fundamental measure of deformation is described by the deformation gradient relative to X given by ∂φ (13. a displacement vector. A = 1. FINITE DEFORMATION 106 When common origins and directions for the coordinate frames are used. The deformation gradient relates the current configuration to the reference configuration.12) (13.11) Using F directly complicates the development of constitutive equations and it is common to introduce deformation measures which are related completely to either the .9) to ensure that material volume elements remain positive. u. Accordingly.

the right Cauchy-Green deformation tensor. . for the reference configuration. Accordingly. e. E (do not confuse with the base vectors). may be expressed in terms of b as e = 1 1t − b−1 2 (13. 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.CHAPTER 13.19) (13. b.15) B and δAB is a Kronecker delta for the reference configuration.22) b and δab is a Kronecker delta for the current configuration. is introduced as 1 E = (C − 10 ) (13. C.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. is introduced as C = FT F (13.13) Alternatively the Green strain tensor. expressed as b = F FT The Almansi strain tensor.17) In the current configuration a common deformation measure is the left Cauchy-Green deformation tensor.20) where 1t is the rank two identity tensor with respect to the current configuration and is given by 1t = δab ea eT (13. FINITE DEFORMATION 107 reference or the current configurations.14) 2 where 10 is the rank two identity tensor with respect to the reference configuration and is given by 10 = δAB EA ET (13.21) (13.18) (13.

and t0 is traction on the reference configuration. respectively.24) where n is an unit outward pointing normal to a surface defined in the current configuration.25) (13.23) b b The second Piola-Kirchhoff stress. 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.29) where ds and dS are surface elements in the current and reference configurations.30) where N is an unit outward pointing normal to the reference surface.2 Stress and Traction Measures Stress measures the amount of force per unit of area. This form of the traction may be related to reference surface quantity through force relations defined as t ds = t0 dS (13. S.CHAPTER 13. σ. P. the first Piola-Kirchhoff stress. remains related to the current configuration forces. They are related through the determinant of the deformation gradient as τ = τab ea eT = J σ = J σab ea eT (13. and the Kirchhoff stress.32) .31) (13.28) (13. τ . In finite deformation problems care must be taken to describe the configuration to which stress is measured. 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. are measures defined with respect to the current configuration. thus.27) (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. ds = J N · C−1 N and (13.26) (13. FINITE DEFORMATION 108 13. Note that the direction of the traction component is preserved during the transformation and. The Cauchy stress. The reference configuration traction is deduced from the first Piola-Kirchhoff stress through t0 = P N (13.

to the rate of ˙ change of the body momentum. R. both for the current configuration.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. through ρ0 = J ρ The total linear momentum of the body is given by p = Ω (13.33) where rho is the mass density per unit volume and ∂Ω is the surface area of the body. R.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. (13. Accordingly. ρ bm dv + Ω ∂Ω td s = Ω ˙ 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. rho0 . FINITE DEFORMATION 109 13.CHAPTER 13. for a body force per unit mass. and balance of angular momentum. acting on a body is given by ρ bm dv + Ω ∂Ω t ds = R (13.37) where div is the divergence with respect to the current configuration. The mass density in the current configuration is related to the reference configuration mass density. div σ = ∂σab eb ∂xa (13. p.3 Balance of Momentum The balance of momentum for a solid body consists of two parts: balance of linear momentum. that is.34) ρ v dv (13. Accordingly. bm the resultant force.39) . The balance of linear momentum may be expressed by integrating the surface and body loads over the body.

CHAPTER 13. leads to the corresponding requirement on P F PT = P FT (13.45) and subsequently to the symmetry of the second Piola-Kirchhoff stress tensor S = ST (13. thus. 2.43) In these relations Div is the divergence with respect to the reference configuration coordinates ∂PaA ea (13. the Cauchy stress tensor is symmetric and. 3. At each point on the boundary one (and only one) boundary condition must be specified for all three . Accordingly. FINITE DEFORMATION 110 Similar relations may be constructed for the balance of angular momentum and lead to the requirement σ = σT (13.40) that is. 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 .41) where the definitions for rho. σ. The conditions are usually given in terms of their components with respect to a local coordinate system defined by the orthogonal basis. has only six independent components.44) Div P = ∂XA We also note that the symmetry of the Cauchy stress tensor.42) which has the local form ˙ Div P + ρ0 bm = ρ0 v (13. σ and n ds in terms of reference configuration quantities have been used. The balance of momentum may also be written for the reference configuration using results deduced above. we may write the integrals with respect to the reference body as ρ0 bm dV + Ω0 ∂Ω0 t0 dS = Ω0 ˙ ρ0 v dV (13.4 Boundary Conditions The basic boundary conditions for a solid region consist of two types: displacement boundary conditions and traction boundary conditions.46) 13. a = 1. Using the divergence principle on the traction term leads to the result ˙ [Div P + ρ0 (bm − v)] dV = 0 Ω0 (13.

49) define components of the displacement with respect to the prime coordinates.CHAPTER 13.51) 13. components of the position vector. For displacement boundary conditions. the boundary condition is non-linear unless points in the reference configuration can be identified easily (such as fixed points). boundary conditions may now be given for each displacement component by requiring ua = u a ¯ (13. (¯ ·). 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 ¯ (13.5 Initial Conditions Initial conditions describe the state of a body at the start of an analysis. u = x − 1 X = ua ea (13. ∂Ωt .50) In general. FINITE DEFORMATION 111 directions of the basis ea . The boundary condition may also be expressed in terms of components of the displacement vector. Generally. Using the orthogonal basis described above. these conditions may be a mixture of displacement and traction boundary conditions. u. (13.47) for each point on the displacement boundary.48) (13. Accordingly. 2 A specified quantity is indicated by a superposed bar. . The second type of boundary condition is a traction boundary condition. for constitutive equations the initial values for internal variables which evolve in time must be given. 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.52) (13. ∂Ωu . 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. In addition. Thus. x.

59) 1 2 1 (IC − tr C2 ) = (CKK CLL − CKL CLK ) 2 2 (13. Thus. (13.54) (13.Finite Elasticity The equations are completed by specifying a material constitution.57) (13. the second Piola-Kirchhoff stress tensor is computed as ∂W (13.CHAPTER 13. we postulate the existence of a strain energy density function. 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. 0) = v0 (X) are specified at each point in the body. 0) = d0 (X) and ˙ ¯ v(0) = φ(X.61) ∂ IC ∂ C ∂ IIC ∂ C ∂J ∂C .60) (13. The initial conditions for stresses are specified as ¯ σ(0) = σ 0 at each point in the body. W .58) We select J instead of IIIC as the measure of the volume change. IIC . the stress is computed as ∂ W ∂ IC ∂ W ∂ IIC ∂W ∂J S = 2 + 2 + 2 (13.6 Material Constitution . Thus. J) (13. C. ¯ x(0) = φ(X.55) (13. As an example. FINITE DEFORMATION 112 The initial conditions for the kinematic state consist of specifying the position and velocity at time zero. For a strain energy density expressed in terms of the right Cauchy-Green deformation tensor. we consider a finite deformation form for hyperelasticity. from which stresses are computed by taking a derivative with respect to a deformation measure.53) 13. Accordingly.

The elastic moduli with repect to the reference configuration are deduced from C= 4 I ∂2W ∂C ∂C (13.71) The spatial elasticities related to the Cauchy stress.65) The second Piola-Kirchhoff stress may be transformed to the Kirchhoff stress using Eq. σ.72) .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 gives S = 2 µ (10 − C−1 ) + λ J (J − 1) C−1 (13. and gives τ = 2 ∂W ∂W ∂W b + 2 (IC b − 1t ) + J 1t ∂ IC ∂ IIC ∂J (13. the stress is computed to be S = 2 ∂W ∂W ∂W 10 + 2 (IC 10 − C) + J C−1 ∂ IC ∂ IIC ∂J 113 (13.25. we consider the case of a Neo-Hookean material which includes a compressibility effect. 13. J) = µ (IC − 3 − 2 ln J) + 1 λ (J − 1)2 2 (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.CHAPTER 13.69) Some formulations require computation of the elastic moduli for the finite elasticity model. The strain energy density is expressed as W (IC . are obtained by the push forward cijkl = 1 FiI FjJ FkK FlL CIJKL J (13.63) (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 (13.64) (13. Substitution into Eq.70) (13.62) (13.66) As an example. e 13.

the computations are quite delicate (see [19]). however. Accordingly. (13.75 as3 δΠ = Ω0 ∂W δC dV − ∂C δuT ρ0 bm dV − Ω0 ∂Ω0 t δuT ¯0 dS = 0 t (13. ∂Ωu . Since a virtual displacement is an arbitrary function.CHAPTER 13. in the reference configuration a variational equation is defined from Eq.73) Transforming to spatial quantities gives cijkl = λ (2 J −1 ) δij δkl − 2 ( µ − λ (J − 1)) δik δjl J (13.74) Other forms of constitutive equations may be introduced using appropriate expansions of the strain energy density function. 13. an elastic formulation may also be expressed in terms of the principal stretches (which are the sqare root of the eigenvalues of C). for convenience. the basic element arrays evolve from the balance of linear momentum equations written as a variational equation.76) 3 .77) where δu is a variation of the displacement (often called a virtual displacement) which is arbitrary except at the kinematic boundary condition locations. In a finite element formulation.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. As an alternative. the notation for a variation to a quantity is written as δ. where. 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. 13.75) where ¯0 denotes the specified tractions in the reference configuration and ∂Ω0 t is the t traction boundary for the reference configuration. 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. it vanishes. ueta → u + δu Furthermore. matrix notation is used as much as possible to express the variational equation. Thus.

26. We note that using Eq.83) (13.81) − 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.78) (13. 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. 13.82) 2 where symmetry of the second Piola-Kirchhoff stress is noted. 13. 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. see Hughes [12] or Zienkiewicz and Taylor [26. small deformation formulation found in earlier chapters and in finite element texts (e.g. For the development considered here it is not important which is selected since ultimately all integrations are carried out over the current configuration. 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. and. The third and fourth terms of the variational equation represent the effects of body and surface traction loadings. the first term reduces to ∂W 1 δC = S δC = δFT P ∂C 2 Furthermore. Eq.56 and constructing the variation of C. FINITE DEFORMATION 115 each point in the body as well as the traction boundary conditions. For static and quasi-static problems this term may be neglected. . 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. 27]) except that integrals are performed over the deformed current configuration. Representations with respect to a fixed reference configuration are introduced to simplify the development of the basic relations. by introducing the inertial forces through the body force as ˙ ¨ bm → bm − v = bm − x where v is the velocity vector.80) (13.CHAPTER 13. the variational equation may be written as δΠ = Ω0 (13. The first term side represents the inertial terms.

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

This will be discussed later for a particular constitutive equation.96) where Ωe is the domain of an individual element. When used with the definition of ∆σ this may be transformed to the current configuration as ∆σ = I ∆ c (13. in general we seek an expression of the form ∆S = C∆E I (13. It is expressed as 1 ∆ = [ ∆u + ( ∆u)T ] (13.92) where I are the material moduli expressed in the current configuration. The moduli c are related through J I = F F C F T FT c I (13. 13.91) where Care the material moduli for the material constitution expressed in the reference I configuration. FINITE DEFORMATION 117 The material part involves ∆S which is computed for each particular constitutive relation.97) Using this approximation the variational equation become ˙ δuT ρ v dv + e Ωe Ωe (δu)T σ dv = . With this approximation the integrals in the variational equation may be approximated as ( · ) dv ≈ Ω Ωh ( · ) dv = e Ωe ( · ) dv (13.CHAPTER 13. We note that in general Ωh is an approximation to the domain of the real body since the elements only have mapped polygonal shapes.9 Element Technology A finite element discretization may be constructed by dividing the body into finite elements. and Ωh is the domain covered by all the elements. we have Ω ≈ Ωh = e Ωe (13.93) In the above ∆ is the symmetric part of the gradient of the incremental displacement. however.95) which we note is also identical to the form of the linear problem. e. Accordingly.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.

CHAPTER 13..103) (13. . Appendix 8] or [11].101) Time dependence is included in the nodal parameters for the current position and displacements. I are node labels for the element. see [12. 27]) we may write for a typical element X = NI (ξ) XI . 13.10 Consistent and Lumped Mass Matrices Using the above approximations we may discretize the terms in the variational equation for each element. . nen (13. Similarly. the first term becomes ˙ δuT ρ v dv = (δuT )I Ωe Ωe ¨ NI ρ NJ dv 1t xJ (13.102) (13. Adopting an isoparametric formulation (e.106) Ωe For procedures to construct a lumped mass see either [26. Accordingly. I = 1. NI (ξ) are shape functions for node I which maintain suitable continuity between contiguous elements and XI are the coordinates for node I. respectively.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 written as MIJ = NI ρ NJ dv 1t (13.g. 26. 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 (13.100) where nen is the number of nodes defining an element.104) (13. FINITE DEFORMATION δuT ρ bm dv + e Ωe ∂Ωet 118 δuT ¯ ds t (13.99) An approximate variational solution may be developed by writing trial solutions and test functions for the motions and virtual displacements. 2.

3   1    δuI =  = BI δuI 2 NI.3 0 NI. The BI matrix describes the transformation from the virtual displacements.1 NI.2  δu3 NI.109) T 31 δ 33 2δ 12 2δ 23 2δ (13.108) σ11 σ22 σ33 σ12 σ23 σ31 δ 22 T (13.3 NI.1 0 0  0 NI.2 NI.1 = δ (13.114) ∂x1 has been used for conciseness.1 0  I    0 NI. The stress divergence term may now be written as δ T σ dv = (δuI )T BT σ dv (13.2 0   I    δu  0 0 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.110) the stress divergence term may be written as δ Ωe T σ dv (13.112) Using the interpolations for the virtual displacements in each element leads to the matrix relation   NI.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. δuI to the δ .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. .113) In the above. the notation ∂NI (13.111) 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.CHAPTER 13. FINITE DEFORMATION 119 13.

120) where D denotes the material moduli in the current configuration given in the matrix representation introduced for the linear problem. In matrix notation the right hand side becomes tr (δ I ∆ ) dv = c Ωe Ωe δ T D ∆ dv (13.116) 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 (13.117) In component form the expression for the geometric stiffness is given as (Kg )IJ = ij Ωe ∂ NI ∂ NJ σkl dv δij ∂ xk ∂ xl (13.13 Material tangent matrix . substitution of the finite element interpolations into the incremental strain term leads to the result in matrix form ∆ = BJ ∆uJ (13.119) which is evaluated for a typical element. Furthermore.121) Thus.118) 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.CHAPTER 13.122) . tr ( (δu) σ Ωe (∆u)T ) dv = (δuI )T Ωe T tr ( NI σ NJ ) dv ∆uJ (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. FINITE DEFORMATION 120 13. Accordingly. the material tangent is computed from δ Ωe T D ∆ dv = (δuI )T Ωe BT D BJ dv ∆uJ I (13.

the variational equation may be written as (δuI )T e ¨ MIJ xJ + Ωe BT σ dv − fI I = 0 (13. FINITE DEFORMATION and the material tangent matrix is given by (Km )IJ = Ωe 121 BT D BJ dv I (13. 13.15 Basic finite element formulation Accumulating all terms together.125) 13. the variational equation leads to the discrete balance of linear momentum equations ¨ MIJ xJ + e Ωe BT σ dv − fI I = 0 (13. fI = Ωe NI ρ bm dv + ∂Ωe t NI ¯ ds t (13.123) which again is identical to the linear problem except that all steps are performed for the current configuration. Accordingly.129) .128) which may be written in the compact matrix form ¨ M x + N(σ) = f (13.CHAPTER 13.127) Since δuI is arbitrary.126) where fI is the sum of the body and surface traction terms.14 Loading terms The right hand side terms may be discretized by introducing the interpolations for the virtual displacement. the body force may be given as δuT ρ bm dv = (δuI )T Ωe Ωe NI ρ bm dv (13.124) and the boundary loading is δuT ¯ tds = (δuI )T ∂Ωe ∂Ωet NI ¯ ds t (13.

We refer to the method as the standard B-matrix formulation.138) .135) (13. yields the solution to a problem. a modified deformation gradient. Thus. in a Newton Method we write the momentum equation as ¨ R = f − M x − N(σ) = 0 (13. is used. The mixed formulation is used to permit solution of incompressible and nearly incompressible materials. as well as.136) (13.133) where Fvol measures volumetric part and Fdev the deviatoric part of deformation.130) A linearization of this set of equations gives the result M ∆¨ + Kt ∆u = R u where Kt = Km + Kg (13. (13. Accordingly. 122 Solution of this set of equations together with satisfying the material constitution and the displacement boundary conditions. (as described in [19]).131) 13. compressible solutions which can be treated by a standard B matrix formulations.137) (13. the modified deformation gradient is based upon a separation into volumetric and deviatoric parts as F = Fvol Fdev (13. FINITE DEFORMATION where N(σ) is the stress divergence vector. A common solution procedure is to use a Newton type solution method and solve a sequence of linear problems.134) (13.132) The above description is for a standard displacement type formulation.CHAPTER 13.16 Mixed formulation In the mixed formulation used. 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.

we define ˜ F = θ J 1 3 F (13. the modified Cauchy stress. φ. ˜ Π(u. δθ( Ω ˜ trσ p − ) dv = 0 3θ J (13. ˜ ˜ In the above. θ is a mixed representation for the determinant of the deformation gradient. The modified deformation gradient may then be constructed by replacing the volumetric part by a mixed treatment. p.141) 3θ 3 A three field variational statement of the problem is completed by adding to the motion.140) ˜ ˜ with similar definitions for E and b. In the above expression.146) for the relation between the mixed and the determinant of the deformation gradient. the mixed pressure.144) t for the linear momentum equation. [19]) ˙ δuT ρ v dv + Ω Ω ˜ δuT (σ dev + p 1t ) dv = Ω δuT ρ bm dv + ∂Ω δuT ¯ ds (13. Accordingly. and the modified Kirchhoff stress.142) Ω0 Ω0 − Ω0 uT ρ0 bm dV − ∂Ω0t t uT ¯0 dS (13. θ.139) as the modified tensor. The virtual modified deformation gradient is now given by δθ 1 ˜ ˜ δF = 1t + ( δu − div δu 1t ) F (13. and mixed determinant of the modified deformation gradient. The modified right Cauchy-Green deformation tensor is then computed from ˜ ˜ ˜ C = FT F (13. and δp (1 − Ω θ ) dv = 0 J (13.143) A variational equation including the effects of inertia may be constructed following steps above and written as (see. are related to the modified second Piola-Kirchhoff stress by ˜˜˜ ˜ ˜ J σ = τ = F S FT (13. θ) dV + p (J − θ) dV (13. σ. τ . FINITE DEFORMATION 123 for the deviatoric part.CHAPTER 13.147) . θ. p) = W C(u.145) for the relationship between the mixed pressure and the trace of the stress.

The θ may now be ˜ used to define the modified deformation quantity and the modified stress state. The spherical part of the stress is given by the ˜ ˜ mixed pressure. Finally. and δu. using the third variational equation. In addition interpolations for θ. 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. A tangent matrix may be computed for the mixed formulation. p.151) The pressure. x. appearing in the above relations may be obtained by first computing the mixed volume. .149) A finite element implementation for the above may be deduced using the isoparametric interpolations given above for X. p. In the low order elements the above functions are all taken as constant in each element. use of the second variational equation yields the mixed pressure as ˜ 1 trσ p = dv (13. σ dev . in each element. Thus. σ may be determined in each element.150) BT (σ dev + p 1t ) dv I ˜ (13. The mixed pressure p is computed from trσ using the variational equation given above. FINITE DEFORMATION 124 ˜ ˜ where S is computed using C as the deformation measure. σ. δθ. and δp must be given. This may be combined with the deviatoric part of σ to define the mixed stress. not trσ.CHAPTER 13.148) where I is a rank four identity tensor. The deviatoric part of the ˜ stress. p. Details for the construction are included in [19]. θ. the stress in this approach is computed using ˜ σ = p 1t + σ dev (13.153) Ωe0 Ωe 3 θ ˜ in each element.152) Ωe0 for each element. Accordingly. where Ωe is the volume of the element in the current configuration and Ωe0 is the volume in the undeformed reference configuration. u. for each element (with the constant interpolations for θ and p) integration of the third variational equation yields a solution Ωe θ= (13. is then computed using ˜ σ dev = (I − 1 ˜ 1t 1T ) σ = Idev σ t ˜ 3 (13.

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

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

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

and density for an isotropic Fourier material model. heat capacity.7 a routine to compute coordinates in elements. 128 . The heat capacity array has been coded separtately in (Table A.8) to permit solution of general linear eigenproblem. Nodal projections of the heat flux may then be plotted in graphics mode by FEAP.6 defines the Fourier model and Table A.3 presents the routine used to compute the element tangent and residual arrays and Tables A. Table A. thermal conductivity.8 the routines to output numerical values and nodal projections for the heat flux.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.2 is the subprogram to input material parameters for the type of region.4 and A.1 is the interface program to FEAP and Table A. Table A. Table A.

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

true.wlab(int(d(4))) d(2) = d(2)*d(3) d(5) = 2 ! number of quadrature points/direction 2000 & & format(5x.lt.’Conductivity ’.’Specific Heat’.e12.’Linear Heat Conduction Element’// 5x.a6.4)) then d(2) = td(1) elseif(pcomp(name.wlab(int(d(4))) write(iow.2000) d(1).5/5x.td.4)) then d(4) = 2.’ ’.4)) then d(1) = td(1) elseif(pcomp(name.4)) then rflag = .td(1) wlab/’ Plane’.’ Analysis’) end Table A.false.APPENDIX A.d(2).0) write(*.’plan’.d0 elseif(pcomp(name.h’ errck.4)) then d(4) = 1.5/5x.1) if(pcomp(name.’Density ’.’dens’.’Axisym’/ 130 logical character real*8 data c Input material parameters d(4) = 1.’axis’. wlab(2)*6 d(*).2000) d(1). tinput.4)) then d(3) = td(1) elseif(pcomp(name. do while(rflag) errck = tinput(name.’spec’.2: Input Routine for Heat Transfer Element . endif end do ! while if(ior.’cond’.5/ 5x. HEAT TRANSFER ELEMENT subroutine inpt02(d) implicit include none ’iofile.d0 rflag = .d(3).e12.d0 elseif(pcomp(name.e12.d(3).d(2).1. rflag name*15. pcomp.

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

’ 1-flux 2-flux 1-grad format(2i5.shp. HEAT TRANSFER ELEMENT 132 subroutine strs02(d.2002) n.yy.xl.*).ul.ndm.ndf.1 if(mct. xsj.xl(ndm.nel) write(iow.h’ ’iofile.ndm.le.flux(2).0 .2002) n.xl(2.ma.9) Compute thermal gradient and heat flux call shape(0.and.shp.xl. pfr) write(*.ix.flux.h’ ’eldata..false.*).3) 2-coord’ 2-grad’) end Table A.0 .h’ ’fdata.lt. ix(*) xx.ndm.ndm.ndm) implicit include include include include include integer real*8 real*8 c none ’bdata.xl(1.2f9.’element flux’//’ elmt matl 1-coord .4: Output Routine for Heat Transfer Element .head if(ior.flux.ix.and.tdot) mct = mct . radi02 d(*).nel.gradt 2001 & 2002 format(a1.ndf.h’ ’cdata.0.ul.2001) o.0d0.shp(3.1). gradt.0d0.xsj.ma.3.h’ ndf.yy.APPENDIX A.yy.flux.) call flux02(d.lt.head mct = 50 endif xx = radi02(shp.nel.0) then write(iow.nel) yy = radi02(shp.xx.gradt if(ior. tdot.1).20a4//5x.xx.2001) o.ul(ndf.4e12.gradt(2). pfr) write(*.

j) dt(ll) = dt(ll) + xg st(ll.l) do j = 1.2) + flux(2)*xg endif end do end do end Table A.sg(3.flux.d(*) gradt(2).ndm.nel.4).xsj.1) = st(ll.ndm.5: Flux Projection Routine for Heat Transfer Element .ul(ndf.l).l).shp.APPENDIX A.st(numnp.2) = st(ll.*).shp.ix.xl.lint.9) Lumped projection routine l = max(2. j.gt.ndf.nel.ll.nel ll = iabs(ix(j)) if(ll.xl(ndm.lint..) call flux02(d.numnp.ul.d. gradt.*).false.ul.xg. ix(*) xsj.0) then xg = xsj*shp(3.tdot) xsj = xsj*sg(3.ndf.lint call shape(sg(1.nel.*).sg(2.sg) do l = 1.nel. tdot dt(numnp).shp.numnp) implicit none integer real*8 real*8 real*8 c ndf.l.st.shp(3. HEAT TRANSFER ELEMENT 133 subroutine stcn02(ix.nint(d(5))) call int2d(l.flux(2).ndm.xl.1) + flux(1)*xg st(ll.dt.

i)*ul(1.0d0 do i = 1. HEAT TRANSFER ELEMENT subroutine flux02(d.nel.0d0 tdot = 0.h’ ndm.shp(3.flux.*) Compute element coordinate value radi02 = 0.nel) implicit none integer real*8 c i.ul. gradt.gradt(*).nel.i)*ul(1.flux(*) 134 gradt(1) = 0.ndm.*).i)*xl(1.i.nen.i.ndf.7: Coordinate in Element .0d0 gradt(2) = 0.xl.nel gradt(1) = gradt(1) + shp(1. i tdot d(*).*).i) end do end Table A.ul(ndf.APPENDIX A. shp(3.shp.1) tdot = tdot + shp(3.1) gradt(2) = gradt(2) + shp(2.ndm.tdot) implicit include integer real*8 real*8 none ’cdata.i.nel radi02. xl(ndm.4) end do flux(1) = -d(1)*gradt(1) flux(2) = -d(1)*gradt(2) end Table A.*).6: Thermal Gradient and Flux function radi02(shp.d0 do i = 1.nel radi02 = radi02 + shp(3.i)*ul(1.

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

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

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

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

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.

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.24) are the values of the global coordinates at the element center.21) (D. 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.23) where xi are the element global cartesian coordinates.22) ξ2 0 0 ξ1 (D. 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.APPENDIX D.2 Alternative Representation in Two Dimensions An alternative representation for the shape functions has been proposed by Belytschko. h(ξ) = ξ1 ξ2 (D.25) .20) D.17) ξ=0 to denote the derivatives of the shape functions at the element center.18) which is the coefficient to the spatially varying part of the jacobian transformation.19) which in matrix notation may be written as J(ξ) = J0 + ∆J Ξ where Ξ = and ∆J = ∆J1 ∆J1 ∆J2 ∆J2 (D. That is. 4 x0 i = I=1 xI NI (0) = i 1 1 (x + x2 + x3 + x4 ) i i i 4 i (D.

29) (D. 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.35) . First by multiplying (from the right) by 1. while the columns are associated with the J where the ξJ are evaluated. the parameters δ and Γ may be easily computed. 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.30) (D.38) (D. These parameters may be evaluated by defining the shape functions at each node and using the fact that NI (ξ J ) = δIJ (D.32) (D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 143 and δI and ΓI are constant parameters associated with node I.APPENDIX D.26) where δIJ is the Kronecker delta function for the nodes. we note that hT 1 = 0 (D. The I is a 4 × 4 identity matrix for the element.36) (D.34) Note that the rows in the expression are associated with the I in the NI shape functions.37) (D.28) (D. Using this form.33) 1 1 1 1 1 −1 1 −1 x1 x2 x3 x4 i i i i bi1 bi2 bi3 bi4 (D.31) 1 δI + 4 2 biI (xJ − x0 ) + ΓI h(ξ J ) i i i=1 (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.39) (D.

40) (D.47) The factor xh is sometimes called an hour glass shape. are given by ∂h ∂NI = biI + ΓI ∂xi ∂xi (D. i .42) x h bi ] i i=1 (D. defines the magnitude of the hour glass mode.46) ξh = ξ1 Furthermore. is replaced i by the displacement.3 Derivatives of Alternative Formulation Using the alternative expression for the shape functions. x. the derivatives with respect to the global coordinates.41) Ih = h = i=1 x h bi + 4 Γ i where2 xh = xT h i i Thus. and when the coordinate. 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. xi .43) D. we get hT h = 4 2 144 (D.45) For the specific functional expression for h. the factor uh . ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS Next by multiplying (again from the right) by h. the parameters for Γ are computed as 1 Γ = [h − 4 It remains to compute the bi .APPENDIX D. u. the gradient with respect to the natural coordinates is given by ξ2 (D. 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. 2 (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.48) where j0 is the value of the jacobian determinant evaluated at the element center. J. The jacobian determinant at the center of the element is computed to be j0 = (J0 )11 (J0 )22 − (J0 )21 (J0 )12 (D.51) (D.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. . Recall that the derivative of a global coordinate with respect to a natural coordinate has a constant and a linear part.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. 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.53) (D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 145 where j(ξ) is the determinant of the jacobian transformation matrix.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.

nT n = 1 (E.1) ∂Σ where Σ n = (E.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. and W is a k × k matrix. The inverse may be proved by multiplying the results together to recover the identity matrix.2) Σ and Σ = s − α We note that n has the properties 1T n = 0 . 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 + U VT )−1 = A−1 − A−1 U W VT A−1 where W = (I + VT A−1 U)−1 (E. U. In the case of the deviatoric model A is diagonal and U 146 (E. where k ≤ n.5) which appears in several location in the tangent matrices.4) (E.7) In the above A is an ntimesn matrix. V are n × k matrices.3) In the derivation of the tangent the derivative of n leads to ∂2f ∂n = = ∂Σ ∂Σ ∂Σ 1 Σ (1 − n nT ) (E.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

5) (F.12) σ = s + mp . P.7) and reordered into the 9-component vector as = T 11 22 33 12 21 23 32 31 13 (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 (F.10) 11 22 33 γ12 γ23 γ31 where γiij are the engineering components of the shearing strain given by γij = 2 ij (F.11) F.APPENDIX F.8) Strain-displacement relations give symmetry of strain as ij = ji (F.6) =  (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.9) This permits the independent components of strain to be written in a 6-component matrix form as T = (F. MATRIX FORM FOR EQUATIONS OF SOLIDS The 6-component form may be related jector matrix.

MATRIX FORM FOR EQUATIONS OF SOLIDS and 151 1 m εv 3 where p and εv are the pressure and volume change.14) (F. respectively. 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.15) (F. Accordingly.APPENDIX F.25) 3 3 .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.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.13) (F.23) 3 and 1 (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. for the stress we have the two relations 1 σ = s + m mT σ (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.24) σ = s + m mT σ 3 which solve to give 1 1 s = σ − m mT σ = I − m mT σ (F.17) (F.18) (F.

respectively. (F.3 Linear Elastic Constitutive Equations Let us now consider the relations for linear elastic constitutive equations.34) (F. We define the two deviatoric projectors as Idev = I − 1 m mT 3 and Idev = I − 1 m mT 3 (F. MATRIX FORM FOR EQUATIONS OF SOLIDS and s=σ− 152 1 1 m mT σ = I − m mT 3 3 σ (F.APPENDIX F.35) where D is a 9 × 9 matrix of elastic constants and D is a 6 × 6 matrix of elastic constants.27) Similarly for strains we have the deviatoric relations e= − and e= − 1 m mT = Idev 3 1 m mT = Idev 3 (F. In index notation these are expressed as σij = Cijkl kl (F.33) .31) From notions of hyperelasticity where stress is deduced from the stored energy function W ( ) as ∂W σij = (F.26) where I and I are identity matrices of size 9 and 6.29) F.28) (F.30) where Cijkl are the elastic moduli and possess the minor symmetries Cijkl = Cjikl = Cijlk (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.

This may be directly related to a matrix form as (F.40) where Iijkl is the rank-4 tensor identity.2.1: 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.APPENDIX F.2: Matrix and tensor index maps (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.38) where in index form εv = kk .3.39) we obtain the tensor form of the elastic moduli as Cijkl = λ δij δkl + 2 µ Iijkl (F.42) .37) F. MATRIX FORM FOR EQUATIONS OF SOLIDS 153 Construction of D follows directly from Cijkl using the index maps shown in Table F. Writing the relationship for the constitution as σij = Cijkl kl (F.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. (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 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.1.

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

Sign up to vote on this title
UsefulNot useful